diff --git "a/exp/log/log-train-2023-02-28-10-38-34-1" "b/exp/log/log-train-2023-02-28-10-38-34-1"
new file mode 100644--- /dev/null
+++ "b/exp/log/log-train-2023-02-28-10-38-34-1"
@@ -0,0 +1,95870 @@
+2023-02-28 10:38:34,447 INFO [train.py:1094] (1/2) Training started
+2023-02-28 10:38:34,447 INFO [train.py:1104] (1/2) Device: cuda:1
+2023-02-28 10:38:34,449 INFO [train.py:1113] (1/2) {'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 10:38:34,449 INFO [train.py:1115] (1/2) About to create model
+2023-02-28 10:38:34,599 INFO [zipformer.py:405] (1/2) 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 10:38:34,607 INFO [train.py:536] (1/2) Use giga
+2023-02-28 10:38:34,611 INFO [train.py:1119] (1/2) Number of model parameters: 6061029
+2023-02-28 10:38:36,322 INFO [train.py:1134] (1/2) Using DDP
+2023-02-28 10:38:36,474 INFO [librispeech.py:43] (1/2) 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] (1/2) About to get train-clean-360 cuts from data/fbank/librispeech_cuts_train-clean-360.jsonl.gz
+2023-02-28 10:38:36,478 INFO [librispeech.py:53] (1/2) About to get train-other-500 cuts from data/fbank/librispeech_cuts_train-other-500.jsonl.gz
+2023-02-28 10:38:36,479 INFO [train.py:1189] (1/2) Using the XL subset of GigaSpeech (10k hours)
+2023-02-28 10:38:36,480 INFO [gigaspeech.py:46] (1/2) About to get train-XL cuts
+2023-02-28 10:38:36,669 INFO [gigaspeech.py:60] (1/2) Loading 1998 splits
+2023-02-28 10:38:45,342 INFO [asr_datamodule.py:165] (1/2) Enable MUSAN
+2023-02-28 10:38:45,343 INFO [asr_datamodule.py:175] (1/2) Enable SpecAugment
+2023-02-28 10:38:45,343 INFO [asr_datamodule.py:176] (1/2) Time warp factor: 80
+2023-02-28 10:38:45,343 INFO [asr_datamodule.py:189] (1/2) About to create train dataset
+2023-02-28 10:38:45,343 INFO [asr_datamodule.py:220] (1/2) Using DynamicBucketingSampler.
+2023-02-28 10:38:49,685 INFO [asr_datamodule.py:229] (1/2) About to create train dataloader
+2023-02-28 10:38:49,686 INFO [asr_datamodule.py:165] (1/2) Enable MUSAN
+2023-02-28 10:38:49,686 INFO [asr_datamodule.py:175] (1/2) Enable SpecAugment
+2023-02-28 10:38:49,686 INFO [asr_datamodule.py:176] (1/2) Time warp factor: 80
+2023-02-28 10:38:49,687 INFO [asr_datamodule.py:189] (1/2) About to create train dataset
+2023-02-28 10:38:49,687 INFO [asr_datamodule.py:220] (1/2) Using DynamicBucketingSampler.
+2023-02-28 10:38:53,629 INFO [asr_datamodule.py:229] (1/2) About to create train dataloader
+2023-02-28 10:38:53,630 INFO [librispeech.py:68] (1/2) About to get dev-clean cuts from data/fbank/librispeech_cuts_dev-clean.jsonl.gz
+2023-02-28 10:38:53,631 INFO [librispeech.py:73] (1/2) About to get dev-other cuts from data/fbank/librispeech_cuts_dev-other.jsonl.gz
+2023-02-28 10:38:53,632 INFO [asr_datamodule.py:242] (1/2) About to create dev dataset
+2023-02-28 10:38:53,843 INFO [asr_datamodule.py:261] (1/2) About to create dev dataloader
+2023-02-28 10:39:29,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=6.52 vs. limit=2.0
+2023-02-28 10:39:31,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1860, 4.1700, 4.1814, 4.1809], device='cuda:1'), covar=tensor([0.0024, 0.0024, 0.0029, 0.0026], device='cuda:1'), in_proj_covar=tensor([0.0027, 0.0026, 0.0027, 0.0027], device='cuda:1'), out_proj_covar=tensor([1.8915e-05, 1.8449e-05, 1.7936e-05, 1.7283e-05], device='cuda:1')
+2023-02-28 10:39:58,431 INFO [train.py:968] (1/2) Epoch 1, batch 50, giga_loss[loss=1.527, simple_loss=1.348, pruned_loss=1.594, over 28871.00 frames. ], tot_loss[loss=2.493, simple_loss=2.252, pruned_loss=2.314, over 1266973.89 frames. ], libri_tot_loss[loss=2.61, simple_loss=2.359, pruned_loss=2.434, over 256300.48 frames. ], giga_tot_loss[loss=2.501, simple_loss=2.259, pruned_loss=2.318, over 1057786.75 frames. ], batch size: 112, lr: 2.75e-02, grad_scale: 1.0
+2023-02-28 10:40:33,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:40:43,136 INFO [train.py:968] (1/2) Epoch 1, batch 100, giga_loss[loss=1.161, simple_loss=1.008, pruned_loss=1.224, over 27581.00 frames. ], tot_loss[loss=1.846, simple_loss=1.644, pruned_loss=1.825, over 2239670.66 frames. ], libri_tot_loss[loss=2.191, simple_loss=1.967, pruned_loss=2.101, over 366406.49 frames. ], giga_tot_loss[loss=1.829, simple_loss=1.628, pruned_loss=1.812, over 2001648.86 frames. ], batch size: 472, lr: 3.00e-02, grad_scale: 1.0
+2023-02-28 10:40:43,652 INFO [optim.py:369] (1/2) 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 10:40:55,222 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=6.97 vs. limit=2.0
+2023-02-28 10:41:21,765 INFO [train.py:968] (1/2) Epoch 1, batch 150, giga_loss[loss=1.046, simple_loss=0.9039, pruned_loss=1.049, over 28301.00 frames. ], tot_loss[loss=1.524, simple_loss=1.343, pruned_loss=1.548, over 3012141.81 frames. ], libri_tot_loss[loss=1.83, simple_loss=1.63, pruned_loss=1.789, over 528175.81 frames. ], giga_tot_loss[loss=1.52, simple_loss=1.339, pruned_loss=1.548, over 2735699.96 frames. ], batch size: 369, lr: 3.25e-02, grad_scale: 1.0
+2023-02-28 10:41:59,725 INFO [train.py:968] (1/2) Epoch 1, batch 200, giga_loss[loss=0.9493, simple_loss=0.8111, pruned_loss=0.9412, over 28527.00 frames. ], tot_loss[loss=1.333, simple_loss=1.165, pruned_loss=1.357, over 3607517.31 frames. ], libri_tot_loss[loss=1.666, simple_loss=1.476, pruned_loss=1.637, over 660082.74 frames. ], giga_tot_loss[loss=1.33, simple_loss=1.162, pruned_loss=1.356, over 3330103.60 frames. ], batch size: 307, lr: 3.50e-02, grad_scale: 1.0
+2023-02-28 10:42:00,290 INFO [optim.py:369] (1/2) 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 10:42:14,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=10.48 vs. limit=5.0
+2023-02-28 10:42:22,206 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:42:24,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:42:38,384 INFO [train.py:968] (1/2) Epoch 1, batch 250, giga_loss[loss=0.8816, simple_loss=0.7433, pruned_loss=0.8694, over 28947.00 frames. ], tot_loss[loss=1.208, simple_loss=1.048, pruned_loss=1.223, over 4069226.83 frames. ], libri_tot_loss[loss=1.596, simple_loss=1.409, pruned_loss=1.569, over 738265.41 frames. ], giga_tot_loss[loss=1.201, simple_loss=1.042, pruned_loss=1.218, over 3822837.47 frames. ], batch size: 145, lr: 3.75e-02, grad_scale: 1.0
+2023-02-28 10:42:47,410 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:43:09,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8887, 3.9443, 3.8075, 3.7200], device='cuda:1'), covar=tensor([0.0128, 0.0101, 0.0236, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0026, 0.0025, 0.0027, 0.0025], device='cuda:1'), out_proj_covar=tensor([1.8861e-05, 1.7962e-05, 1.8053e-05, 1.7115e-05], device='cuda:1')
+2023-02-28 10:43:16,652 INFO [train.py:968] (1/2) Epoch 1, batch 300, giga_loss[loss=0.8872, simple_loss=0.7493, pruned_loss=0.8261, over 28790.00 frames. ], tot_loss[loss=1.121, simple_loss=0.9664, pruned_loss=1.122, over 4428885.28 frames. ], libri_tot_loss[loss=1.473, simple_loss=1.291, pruned_loss=1.442, over 914857.76 frames. ], giga_tot_loss[loss=1.118, simple_loss=0.9636, pruned_loss=1.121, over 4179882.28 frames. ], batch size: 262, lr: 4.00e-02, grad_scale: 1.0
+2023-02-28 10:43:17,223 INFO [optim.py:369] (1/2) 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 10:43:33,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-02-28 10:43:49,634 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 350, giga_loss[loss=0.8104, simple_loss=0.6772, pruned_loss=0.7476, over 28934.00 frames. ], tot_loss[loss=1.052, simple_loss=0.9018, pruned_loss=1.035, over 4705531.83 frames. ], libri_tot_loss[loss=1.446, simple_loss=1.265, pruned_loss=1.413, over 964511.09 frames. ], giga_tot_loss[loss=1.047, simple_loss=0.8975, pruned_loss=1.032, over 4498008.10 frames. ], batch size: 106, lr: 4.24e-02, grad_scale: 1.0
+2023-02-28 10:44:14,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=2.10 vs. limit=2.0
+2023-02-28 10:44:17,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=5.22 vs. limit=2.0
+2023-02-28 10:44:36,800 INFO [train.py:968] (1/2) Epoch 1, batch 400, giga_loss[loss=0.8704, simple_loss=0.7257, pruned_loss=0.7761, over 28849.00 frames. ], tot_loss[loss=1.005, simple_loss=0.8564, pruned_loss=0.9704, over 4931223.28 frames. ], libri_tot_loss[loss=1.41, simple_loss=1.23, pruned_loss=1.371, over 1038236.05 frames. ], giga_tot_loss[loss=0.9992, simple_loss=0.8521, pruned_loss=0.9671, over 4752767.29 frames. ], batch size: 186, lr: 4.49e-02, grad_scale: 2.0
+2023-02-28 10:44:37,965 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 450, giga_loss[loss=0.8563, simple_loss=0.7175, pruned_loss=0.7256, over 28290.00 frames. ], tot_loss[loss=0.971, simple_loss=0.824, pruned_loss=0.9169, over 5103846.90 frames. ], libri_tot_loss[loss=1.36, simple_loss=1.182, pruned_loss=1.31, over 1156573.71 frames. ], giga_tot_loss[loss=0.9652, simple_loss=0.8197, pruned_loss=0.9143, over 4943676.11 frames. ], batch size: 368, lr: 4.74e-02, grad_scale: 2.0
+2023-02-28 10:45:33,369 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=476.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:45:53,904 INFO [train.py:968] (1/2) Epoch 1, batch 500, giga_loss[loss=0.8364, simple_loss=0.7002, pruned_loss=0.6885, over 28891.00 frames. ], tot_loss[loss=0.9402, simple_loss=0.7955, pruned_loss=0.8659, over 5231303.63 frames. ], libri_tot_loss[loss=1.328, simple_loss=1.151, pruned_loss=1.268, over 1250439.87 frames. ], giga_tot_loss[loss=0.9336, simple_loss=0.7906, pruned_loss=0.8632, over 5091183.60 frames. ], batch size: 174, lr: 4.99e-02, grad_scale: 2.0
+2023-02-28 10:45:54,431 INFO [optim.py:369] (1/2) 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 10:46:30,327 INFO [train.py:968] (1/2) Epoch 1, batch 550, libri_loss[loss=1.056, simple_loss=0.8793, pruned_loss=0.8553, over 29290.00 frames. ], tot_loss[loss=0.9175, simple_loss=0.774, pruned_loss=0.8243, over 5336868.33 frames. ], libri_tot_loss[loss=1.276, simple_loss=1.1, pruned_loss=1.198, over 1430435.94 frames. ], giga_tot_loss[loss=0.9089, simple_loss=0.7677, pruned_loss=0.8218, over 5203608.53 frames. ], batch size: 94, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:46:57,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.83 vs. limit=5.0
+2023-02-28 10:47:11,998 INFO [train.py:968] (1/2) Epoch 1, batch 600, giga_loss[loss=0.7718, simple_loss=0.6496, pruned_loss=0.5936, over 28991.00 frames. ], tot_loss[loss=0.8877, simple_loss=0.7482, pruned_loss=0.7759, over 5416778.38 frames. ], libri_tot_loss[loss=1.258, simple_loss=1.083, pruned_loss=1.173, over 1497263.07 frames. ], giga_tot_loss[loss=0.8791, simple_loss=0.7419, pruned_loss=0.7732, over 5304128.65 frames. ], batch size: 100, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:47:12,566 INFO [optim.py:369] (1/2) 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 10:47:15,330 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-02-28 10:47:56,302 INFO [train.py:968] (1/2) Epoch 1, batch 650, giga_loss[loss=0.7498, simple_loss=0.6335, pruned_loss=0.5586, over 29004.00 frames. ], tot_loss[loss=0.8566, simple_loss=0.7226, pruned_loss=0.7271, over 5482386.00 frames. ], libri_tot_loss[loss=1.24, simple_loss=1.066, pruned_loss=1.148, over 1563114.80 frames. ], giga_tot_loss[loss=0.8486, simple_loss=0.7166, pruned_loss=0.7248, over 5387376.94 frames. ], batch size: 145, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:48:05,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0927, 0.9898, 1.3638, 1.3345], device='cuda:1'), covar=tensor([1.4598, 1.9198, 1.1616, 1.0240], device='cuda:1'), in_proj_covar=tensor([0.0104, 0.0122, 0.0104, 0.0089], device='cuda:1'), out_proj_covar=tensor([7.6266e-05, 8.2356e-05, 6.9673e-05, 6.3704e-05], device='cuda:1')
+2023-02-28 10:48:26,892 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:48:36,489 INFO [train.py:968] (1/2) Epoch 1, batch 700, giga_loss[loss=0.6976, simple_loss=0.5992, pruned_loss=0.4898, over 28913.00 frames. ], tot_loss[loss=0.831, simple_loss=0.7017, pruned_loss=0.6854, over 5531936.56 frames. ], libri_tot_loss[loss=1.216, simple_loss=1.044, pruned_loss=1.111, over 1670314.35 frames. ], giga_tot_loss[loss=0.8216, simple_loss=0.6945, pruned_loss=0.6827, over 5447626.66 frames. ], batch size: 106, lr: 4.98e-02, grad_scale: 2.0
+2023-02-28 10:48:37,085 INFO [optim.py:369] (1/2) 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 10:48:46,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:49:16,902 INFO [train.py:968] (1/2) Epoch 1, batch 750, giga_loss[loss=0.6363, simple_loss=0.5394, pruned_loss=0.4526, over 23828.00 frames. ], tot_loss[loss=0.805, simple_loss=0.6809, pruned_loss=0.6454, over 5556925.24 frames. ], libri_tot_loss[loss=1.186, simple_loss=1.016, pruned_loss=1.067, over 1790014.48 frames. ], giga_tot_loss[loss=0.7951, simple_loss=0.6732, pruned_loss=0.6433, over 5484799.90 frames. ], batch size: 705, lr: 4.97e-02, grad_scale: 2.0
+2023-02-28 10:49:41,981 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:49:46,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8178, 0.8506, 0.8429, 0.5740], device='cuda:1'), covar=tensor([0.8542, 1.1597, 1.1565, 1.4226], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0116, 0.0111, 0.0111], device='cuda:1'), out_proj_covar=tensor([6.5748e-05, 7.6508e-05, 7.0077e-05, 7.4541e-05], device='cuda:1')
+2023-02-28 10:49:58,783 INFO [train.py:968] (1/2) Epoch 1, batch 800, giga_loss[loss=0.741, simple_loss=0.633, pruned_loss=0.5099, over 29007.00 frames. ], tot_loss[loss=0.7768, simple_loss=0.6587, pruned_loss=0.6056, over 5585719.49 frames. ], libri_tot_loss[loss=1.181, simple_loss=1.013, pruned_loss=1.06, over 1810583.40 frames. ], giga_tot_loss[loss=0.7682, simple_loss=0.6519, pruned_loss=0.6034, over 5527835.32 frames. ], batch size: 164, lr: 4.97e-02, grad_scale: 4.0
+2023-02-28 10:49:59,309 INFO [optim.py:369] (1/2) 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 10:50:25,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8634, 0.8095, 1.0793, 0.6745], device='cuda:1'), covar=tensor([0.7682, 1.2073, 0.4968, 1.5177], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0151, 0.0101, 0.0163], device='cuda:1'), out_proj_covar=tensor([8.0986e-05, 1.0355e-04, 6.2954e-05, 1.1844e-04], device='cuda:1')
+2023-02-28 10:50:44,171 INFO [train.py:968] (1/2) Epoch 1, batch 850, giga_loss[loss=0.9798, simple_loss=0.7949, pruned_loss=0.7332, over 28316.00 frames. ], tot_loss[loss=0.7806, simple_loss=0.6621, pruned_loss=0.5929, over 5605026.44 frames. ], libri_tot_loss[loss=1.171, simple_loss=1.003, pruned_loss=1.046, over 1851513.91 frames. ], giga_tot_loss[loss=0.7732, simple_loss=0.6562, pruned_loss=0.5911, over 5556739.30 frames. ], batch size: 368, lr: 4.96e-02, grad_scale: 4.0
+2023-02-28 10:50:45,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:50:47,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=855.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:50:49,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:51:12,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:51:21,724 INFO [train.py:968] (1/2) Epoch 1, batch 900, giga_loss[loss=0.814, simple_loss=0.6969, pruned_loss=0.5424, over 28929.00 frames. ], tot_loss[loss=0.7936, simple_loss=0.6729, pruned_loss=0.5888, over 5634767.41 frames. ], libri_tot_loss[loss=1.142, simple_loss=0.9773, pruned_loss=1.006, over 1970634.08 frames. ], giga_tot_loss[loss=0.7862, simple_loss=0.6671, pruned_loss=0.5873, over 5589431.05 frames. ], batch size: 112, lr: 4.96e-02, grad_scale: 4.0
+2023-02-28 10:51:22,899 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 950, giga_loss[loss=0.8116, simple_loss=0.693, pruned_loss=0.5368, over 28970.00 frames. ], tot_loss[loss=0.7984, simple_loss=0.6777, pruned_loss=0.5788, over 5638997.42 frames. ], libri_tot_loss[loss=1.133, simple_loss=0.9692, pruned_loss=0.9922, over 2010151.18 frames. ], giga_tot_loss[loss=0.7923, simple_loss=0.673, pruned_loss=0.5778, over 5600785.98 frames. ], batch size: 213, lr: 4.96e-02, grad_scale: 4.0
+2023-02-28 10:52:08,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=955.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:52:08,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8196, 2.8731, 3.4267, 1.9526], device='cuda:1'), covar=tensor([0.3808, 0.2702, 0.2994, 0.4018], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0171, 0.0224, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-02-28 10:52:12,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 10:52:27,050 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:52:37,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=994.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:52:39,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:52:40,494 INFO [train.py:968] (1/2) Epoch 1, batch 1000, giga_loss[loss=0.8174, simple_loss=0.6969, pruned_loss=0.5359, over 27912.00 frames. ], tot_loss[loss=0.7957, simple_loss=0.6773, pruned_loss=0.5637, over 5651065.76 frames. ], libri_tot_loss[loss=1.107, simple_loss=0.9473, pruned_loss=0.9552, over 2125876.06 frames. ], giga_tot_loss[loss=0.7901, simple_loss=0.6728, pruned_loss=0.5637, over 5613125.91 frames. ], batch size: 412, lr: 4.95e-02, grad_scale: 4.0
+2023-02-28 10:52:41,028 INFO [optim.py:369] (1/2) 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 10:52:57,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:53:14,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1337, 2.6055, 3.3696, 1.4922], device='cuda:1'), covar=tensor([0.3354, 0.2883, 0.2040, 0.5349], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0151, 0.0199, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1')
+2023-02-28 10:53:14,858 INFO [train.py:968] (1/2) Epoch 1, batch 1050, libri_loss[loss=0.8375, simple_loss=0.7153, pruned_loss=0.5414, over 29362.00 frames. ], tot_loss[loss=0.7918, simple_loss=0.675, pruned_loss=0.5499, over 5671094.07 frames. ], libri_tot_loss[loss=1.078, simple_loss=0.9216, pruned_loss=0.9135, over 2272522.29 frames. ], giga_tot_loss[loss=0.7862, simple_loss=0.6707, pruned_loss=0.5503, over 5633324.77 frames. ], batch size: 92, lr: 4.95e-02, grad_scale: 4.0
+2023-02-28 10:53:27,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1064.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:53:52,900 INFO [train.py:968] (1/2) Epoch 1, batch 1100, giga_loss[loss=0.6785, simple_loss=0.5905, pruned_loss=0.4207, over 28509.00 frames. ], tot_loss[loss=0.7772, simple_loss=0.6657, pruned_loss=0.5279, over 5669814.91 frames. ], libri_tot_loss[loss=1.044, simple_loss=0.8927, pruned_loss=0.8673, over 2423907.69 frames. ], giga_tot_loss[loss=0.7742, simple_loss=0.6633, pruned_loss=0.531, over 5637064.30 frames. ], batch size: 85, lr: 4.94e-02, grad_scale: 4.0
+2023-02-28 10:53:54,648 INFO [optim.py:369] (1/2) 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 10:54:32,742 INFO [train.py:968] (1/2) Epoch 1, batch 1150, libri_loss[loss=0.7776, simple_loss=0.6734, pruned_loss=0.4822, over 29534.00 frames. ], tot_loss[loss=0.7662, simple_loss=0.6581, pruned_loss=0.5109, over 5672611.65 frames. ], libri_tot_loss[loss=1.028, simple_loss=0.8792, pruned_loss=0.8448, over 2509355.65 frames. ], giga_tot_loss[loss=0.7638, simple_loss=0.6561, pruned_loss=0.5141, over 5641485.48 frames. ], batch size: 80, lr: 4.94e-02, grad_scale: 4.0
+2023-02-28 10:54:36,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:55:11,734 INFO [train.py:968] (1/2) Epoch 1, batch 1200, libri_loss[loss=0.8432, simple_loss=0.711, pruned_loss=0.5399, over 29746.00 frames. ], tot_loss[loss=0.7575, simple_loss=0.6521, pruned_loss=0.4969, over 5681001.18 frames. ], libri_tot_loss[loss=1.003, simple_loss=0.8586, pruned_loss=0.8118, over 2639372.97 frames. ], giga_tot_loss[loss=0.7561, simple_loss=0.6508, pruned_loss=0.5008, over 5650585.52 frames. ], batch size: 87, lr: 4.93e-02, grad_scale: 8.0
+2023-02-28 10:55:12,299 INFO [optim.py:369] (1/2) 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 10:55:16,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:55:19,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1210.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:55:43,680 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:55:44,077 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 10:55:50,442 INFO [train.py:968] (1/2) Epoch 1, batch 1250, giga_loss[loss=0.7044, simple_loss=0.6184, pruned_loss=0.4215, over 28807.00 frames. ], tot_loss[loss=0.745, simple_loss=0.6442, pruned_loss=0.4797, over 5684760.48 frames. ], libri_tot_loss[loss=0.9921, simple_loss=0.8496, pruned_loss=0.7966, over 2700814.75 frames. ], giga_tot_loss[loss=0.7439, simple_loss=0.6432, pruned_loss=0.4834, over 5659576.87 frames. ], batch size: 99, lr: 4.92e-02, grad_scale: 8.0
+2023-02-28 10:56:31,631 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 1300, giga_loss[loss=0.6844, simple_loss=0.6108, pruned_loss=0.3972, over 28886.00 frames. ], tot_loss[loss=0.7427, simple_loss=0.6436, pruned_loss=0.4713, over 5690537.10 frames. ], libri_tot_loss[loss=0.986, simple_loss=0.8442, pruned_loss=0.7871, over 2748649.28 frames. ], giga_tot_loss[loss=0.741, simple_loss=0.6422, pruned_loss=0.4738, over 5667740.78 frames. ], batch size: 186, lr: 4.92e-02, grad_scale: 8.0
+2023-02-28 10:56:33,288 INFO [optim.py:369] (1/2) 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 10:56:33,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:56:54,916 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:56:54,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:56:58,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:57:09,570 INFO [train.py:968] (1/2) Epoch 1, batch 1350, giga_loss[loss=0.6517, simple_loss=0.5916, pruned_loss=0.3671, over 28801.00 frames. ], tot_loss[loss=0.7312, simple_loss=0.6371, pruned_loss=0.456, over 5686804.34 frames. ], libri_tot_loss[loss=0.9813, simple_loss=0.8404, pruned_loss=0.7801, over 2779895.71 frames. ], giga_tot_loss[loss=0.7296, simple_loss=0.6358, pruned_loss=0.458, over 5667212.61 frames. ], batch size: 262, lr: 4.91e-02, grad_scale: 8.0
+2023-02-28 10:57:17,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:57:32,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5175, 2.5717, 3.3483, 2.3326], device='cuda:1'), covar=tensor([0.2012, 0.2006, 0.1076, 0.2822], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0173, 0.0201, 0.0240], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-02-28 10:57:47,743 INFO [train.py:968] (1/2) Epoch 1, batch 1400, giga_loss[loss=0.682, simple_loss=0.5994, pruned_loss=0.4005, over 28487.00 frames. ], tot_loss[loss=0.7222, simple_loss=0.6318, pruned_loss=0.4437, over 5698169.50 frames. ], libri_tot_loss[loss=0.9682, simple_loss=0.8296, pruned_loss=0.7625, over 2856981.55 frames. ], giga_tot_loss[loss=0.7209, simple_loss=0.6309, pruned_loss=0.4455, over 5677858.71 frames. ], batch size: 60, lr: 4.91e-02, grad_scale: 8.0
+2023-02-28 10:57:48,914 INFO [optim.py:369] (1/2) 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 10:58:10,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0195, 0.9402, 0.9667, 0.9817], device='cuda:1'), covar=tensor([0.5942, 0.7633, 0.7141, 0.5713], device='cuda:1'), in_proj_covar=tensor([0.0139, 0.0160, 0.0153, 0.0121], device='cuda:1'), out_proj_covar=tensor([8.8067e-05, 9.7962e-05, 9.5162e-05, 7.7211e-05], device='cuda:1')
+2023-02-28 10:58:24,307 INFO [train.py:968] (1/2) Epoch 1, batch 1450, giga_loss[loss=0.6975, simple_loss=0.6051, pruned_loss=0.4143, over 27566.00 frames. ], tot_loss[loss=0.7142, simple_loss=0.6268, pruned_loss=0.4333, over 5690538.44 frames. ], libri_tot_loss[loss=0.952, simple_loss=0.8159, pruned_loss=0.741, over 2951079.31 frames. ], giga_tot_loss[loss=0.7132, simple_loss=0.6264, pruned_loss=0.4351, over 5677527.64 frames. ], batch size: 472, lr: 4.90e-02, grad_scale: 4.0
+2023-02-28 10:58:40,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1473.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:58:42,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1476.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:58:44,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1479.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:58:46,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7150, 2.0749, 3.1395, 1.6833], device='cuda:1'), covar=tensor([0.2194, 0.2549, 0.1049, 0.3918], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0231, 0.0256, 0.0330], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 10:58:46,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1482.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:58:58,031 INFO [train.py:968] (1/2) Epoch 1, batch 1500, giga_loss[loss=0.6442, simple_loss=0.5886, pruned_loss=0.3566, over 28923.00 frames. ], tot_loss[loss=0.6997, simple_loss=0.6184, pruned_loss=0.4176, over 5695128.16 frames. ], libri_tot_loss[loss=0.9384, simple_loss=0.8053, pruned_loss=0.7228, over 3029525.20 frames. ], giga_tot_loss[loss=0.6989, simple_loss=0.6181, pruned_loss=0.4194, over 5688111.65 frames. ], batch size: 164, lr: 4.89e-02, grad_scale: 4.0
+2023-02-28 10:58:59,015 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:59:02,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1504.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:59:02,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1505.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 10:59:06,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1511.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:59:22,362 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1533.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 10:59:32,621 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 1550, giga_loss[loss=0.6244, simple_loss=0.574, pruned_loss=0.3421, over 28632.00 frames. ], tot_loss[loss=0.689, simple_loss=0.6117, pruned_loss=0.4061, over 5701562.53 frames. ], libri_tot_loss[loss=0.9324, simple_loss=0.8003, pruned_loss=0.7131, over 3083288.74 frames. ], giga_tot_loss[loss=0.6869, simple_loss=0.6104, pruned_loss=0.4066, over 5695557.95 frames. ], batch size: 85, lr: 4.89e-02, grad_scale: 4.0
+2023-02-28 10:59:54,367 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-02-28 11:00:15,270 INFO [train.py:968] (1/2) Epoch 1, batch 1600, giga_loss[loss=0.758, simple_loss=0.6186, pruned_loss=0.4717, over 27636.00 frames. ], tot_loss[loss=0.6859, simple_loss=0.6084, pruned_loss=0.4019, over 5693578.50 frames. ], libri_tot_loss[loss=0.923, simple_loss=0.793, pruned_loss=0.7009, over 3139190.85 frames. ], giga_tot_loss[loss=0.6844, simple_loss=0.6075, pruned_loss=0.4027, over 5685303.55 frames. ], batch size: 472, lr: 4.88e-02, grad_scale: 8.0
+2023-02-28 11:00:16,507 INFO [optim.py:369] (1/2) 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 11:00:25,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1615.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:00:53,083 INFO [train.py:968] (1/2) Epoch 1, batch 1650, giga_loss[loss=0.6777, simple_loss=0.5969, pruned_loss=0.3881, over 28955.00 frames. ], tot_loss[loss=0.686, simple_loss=0.6067, pruned_loss=0.4007, over 5704923.13 frames. ], libri_tot_loss[loss=0.9122, simple_loss=0.7845, pruned_loss=0.6868, over 3207746.66 frames. ], giga_tot_loss[loss=0.6847, simple_loss=0.606, pruned_loss=0.4014, over 5693687.81 frames. ], batch size: 112, lr: 4.87e-02, grad_scale: 4.0
+2023-02-28 11:01:31,800 INFO [train.py:968] (1/2) Epoch 1, batch 1700, giga_loss[loss=0.6531, simple_loss=0.5842, pruned_loss=0.3667, over 28565.00 frames. ], tot_loss[loss=0.6832, simple_loss=0.6039, pruned_loss=0.397, over 5710108.81 frames. ], libri_tot_loss[loss=0.9051, simple_loss=0.7788, pruned_loss=0.6769, over 3259241.29 frames. ], giga_tot_loss[loss=0.6817, simple_loss=0.603, pruned_loss=0.3974, over 5699404.37 frames. ], batch size: 65, lr: 4.87e-02, grad_scale: 4.0
+2023-02-28 11:01:33,832 INFO [optim.py:369] (1/2) 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 11:02:14,737 INFO [train.py:968] (1/2) Epoch 1, batch 1750, giga_loss[loss=0.6063, simple_loss=0.5545, pruned_loss=0.3317, over 28894.00 frames. ], tot_loss[loss=0.6742, simple_loss=0.5971, pruned_loss=0.3891, over 5698885.41 frames. ], libri_tot_loss[loss=0.8977, simple_loss=0.773, pruned_loss=0.667, over 3311783.85 frames. ], giga_tot_loss[loss=0.6726, simple_loss=0.596, pruned_loss=0.3893, over 5686522.69 frames. ], batch size: 174, lr: 4.86e-02, grad_scale: 4.0
+2023-02-28 11:02:20,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1758.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:02:23,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1761.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:02:45,707 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:02:53,507 INFO [train.py:968] (1/2) Epoch 1, batch 1800, giga_loss[loss=0.5983, simple_loss=0.5516, pruned_loss=0.3241, over 28914.00 frames. ], tot_loss[loss=0.661, simple_loss=0.5884, pruned_loss=0.3779, over 5696102.14 frames. ], libri_tot_loss[loss=0.8977, simple_loss=0.773, pruned_loss=0.667, over 3311783.85 frames. ], giga_tot_loss[loss=0.6597, simple_loss=0.5876, pruned_loss=0.378, over 5686480.07 frames. ], batch size: 99, lr: 4.85e-02, grad_scale: 4.0
+2023-02-28 11:02:55,670 INFO [optim.py:369] (1/2) 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 11:03:12,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1827.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:03:31,126 INFO [train.py:968] (1/2) Epoch 1, batch 1850, giga_loss[loss=0.5741, simple_loss=0.5489, pruned_loss=0.2992, over 28965.00 frames. ], tot_loss[loss=0.654, simple_loss=0.5845, pruned_loss=0.3709, over 5688115.50 frames. ], libri_tot_loss[loss=0.8904, simple_loss=0.7673, pruned_loss=0.6574, over 3359608.31 frames. ], giga_tot_loss[loss=0.6526, simple_loss=0.5835, pruned_loss=0.3709, over 5679783.68 frames. ], batch size: 145, lr: 4.84e-02, grad_scale: 4.0
+2023-02-28 11:04:11,942 INFO [train.py:968] (1/2) Epoch 1, batch 1900, giga_loss[loss=0.6033, simple_loss=0.5595, pruned_loss=0.3242, over 28830.00 frames. ], tot_loss[loss=0.6416, simple_loss=0.5779, pruned_loss=0.36, over 5688730.76 frames. ], libri_tot_loss[loss=0.8858, simple_loss=0.7639, pruned_loss=0.6508, over 3394245.72 frames. ], giga_tot_loss[loss=0.6397, simple_loss=0.5765, pruned_loss=0.3596, over 5682134.92 frames. ], batch size: 99, lr: 4.83e-02, grad_scale: 4.0
+2023-02-28 11:04:15,689 INFO [optim.py:369] (1/2) 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 11:04:21,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3940, 1.0918, 0.9607, 1.3016], device='cuda:1'), covar=tensor([0.9180, 1.2737, 1.1453, 1.4837], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0490, 0.0494, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 11:04:28,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3722, 1.3636, 1.3272, 1.1398], device='cuda:1'), covar=tensor([0.4037, 0.3823, 0.3430, 0.3251], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0219, 0.0186, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1')
+2023-02-28 11:04:30,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1921.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:04:51,053 INFO [train.py:968] (1/2) Epoch 1, batch 1950, giga_loss[loss=0.5151, simple_loss=0.4936, pruned_loss=0.2682, over 28739.00 frames. ], tot_loss[loss=0.6249, simple_loss=0.5668, pruned_loss=0.3473, over 5677251.81 frames. ], libri_tot_loss[loss=0.8678, simple_loss=0.7502, pruned_loss=0.6292, over 3496938.13 frames. ], giga_tot_loss[loss=0.6233, simple_loss=0.5656, pruned_loss=0.3471, over 5670629.31 frames. ], batch size: 242, lr: 4.83e-02, grad_scale: 4.0
+2023-02-28 11:05:01,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1959.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:05:27,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2105, 2.4348, 3.2569, 2.3589], device='cuda:1'), covar=tensor([0.1445, 0.1397, 0.0501, 0.1910], device='cuda:1'), in_proj_covar=tensor([0.0323, 0.0296, 0.0349, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1')
+2023-02-28 11:05:34,818 INFO [train.py:968] (1/2) Epoch 1, batch 2000, giga_loss[loss=0.5271, simple_loss=0.5058, pruned_loss=0.2742, over 28943.00 frames. ], tot_loss[loss=0.6058, simple_loss=0.5538, pruned_loss=0.3334, over 5667688.93 frames. ], libri_tot_loss[loss=0.8589, simple_loss=0.7432, pruned_loss=0.6175, over 3556323.07 frames. ], giga_tot_loss[loss=0.6029, simple_loss=0.5518, pruned_loss=0.3323, over 5667418.28 frames. ], batch size: 106, lr: 4.82e-02, grad_scale: 8.0
+2023-02-28 11:05:37,472 INFO [optim.py:369] (1/2) 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 11:05:38,235 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2004.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:06:19,128 INFO [train.py:968] (1/2) Epoch 1, batch 2050, giga_loss[loss=0.5367, simple_loss=0.4794, pruned_loss=0.297, over 23543.00 frames. ], tot_loss[loss=0.5782, simple_loss=0.5364, pruned_loss=0.3135, over 5659353.93 frames. ], libri_tot_loss[loss=0.8565, simple_loss=0.7416, pruned_loss=0.6138, over 3579617.51 frames. ], giga_tot_loss[loss=0.5748, simple_loss=0.534, pruned_loss=0.312, over 5657228.39 frames. ], batch size: 705, lr: 4.81e-02, grad_scale: 8.0
+2023-02-28 11:06:33,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2064.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:06:36,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2067.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:07:00,065 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2096.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:07:02,275 INFO [train.py:968] (1/2) Epoch 1, batch 2100, giga_loss[loss=0.5302, simple_loss=0.5114, pruned_loss=0.2745, over 28987.00 frames. ], tot_loss[loss=0.5689, simple_loss=0.5313, pruned_loss=0.306, over 5652694.23 frames. ], libri_tot_loss[loss=0.8546, simple_loss=0.74, pruned_loss=0.6109, over 3593681.66 frames. ], giga_tot_loss[loss=0.5652, simple_loss=0.5288, pruned_loss=0.3041, over 5656817.64 frames. ], batch size: 136, lr: 4.80e-02, grad_scale: 8.0
+2023-02-28 11:07:04,224 INFO [optim.py:369] (1/2) 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 11:07:09,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4281, 2.2849, 3.5533, 1.5781], device='cuda:1'), covar=tensor([0.0919, 0.1108, 0.0617, 0.1695], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0264, 0.0374, 0.0321], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1')
+2023-02-28 11:07:40,170 INFO [train.py:968] (1/2) Epoch 1, batch 2150, libri_loss[loss=0.6744, simple_loss=0.6151, pruned_loss=0.3669, over 29534.00 frames. ], tot_loss[loss=0.5657, simple_loss=0.531, pruned_loss=0.3023, over 5673773.78 frames. ], libri_tot_loss[loss=0.8421, simple_loss=0.7312, pruned_loss=0.5949, over 3682854.87 frames. ], giga_tot_loss[loss=0.5595, simple_loss=0.5264, pruned_loss=0.299, over 5669933.24 frames. ], batch size: 82, lr: 4.79e-02, grad_scale: 8.0
+2023-02-28 11:08:15,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1842, 1.1308, 1.2496, 0.5263], device='cuda:1'), covar=tensor([0.1662, 0.2175, 0.1044, 0.5209], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0277, 0.0241, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003], device='cuda:1')
+2023-02-28 11:08:18,311 INFO [train.py:968] (1/2) Epoch 1, batch 2200, giga_loss[loss=0.4606, simple_loss=0.4604, pruned_loss=0.2304, over 28602.00 frames. ], tot_loss[loss=0.5586, simple_loss=0.5268, pruned_loss=0.2968, over 5680397.72 frames. ], libri_tot_loss[loss=0.8287, simple_loss=0.7214, pruned_loss=0.5783, over 3777504.29 frames. ], giga_tot_loss[loss=0.5503, simple_loss=0.5208, pruned_loss=0.292, over 5671340.69 frames. ], batch size: 71, lr: 4.78e-02, grad_scale: 4.0
+2023-02-28 11:08:19,912 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2202.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:08:20,753 INFO [optim.py:369] (1/2) 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 11:08:41,031 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2231.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:08:54,927 INFO [train.py:968] (1/2) Epoch 1, batch 2250, giga_loss[loss=0.4413, simple_loss=0.4468, pruned_loss=0.2179, over 28353.00 frames. ], tot_loss[loss=0.5512, simple_loss=0.5224, pruned_loss=0.2912, over 5690639.97 frames. ], libri_tot_loss[loss=0.8194, simple_loss=0.715, pruned_loss=0.5658, over 3855313.92 frames. ], giga_tot_loss[loss=0.5402, simple_loss=0.5144, pruned_loss=0.2847, over 5680595.17 frames. ], batch size: 65, lr: 4.77e-02, grad_scale: 4.0
+2023-02-28 11:08:55,105 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2250.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:09:03,170 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2263.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:09:18,009 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-02-28 11:09:31,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2285, 0.9290, 0.9035, 0.7386], device='cuda:1'), covar=tensor([0.3027, 0.3910, 0.4279, 0.6440], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0441, 0.0416, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 11:09:34,748 INFO [train.py:968] (1/2) Epoch 1, batch 2300, giga_loss[loss=0.4414, simple_loss=0.4431, pruned_loss=0.2198, over 28703.00 frames. ], tot_loss[loss=0.5349, simple_loss=0.5121, pruned_loss=0.2798, over 5696476.73 frames. ], libri_tot_loss[loss=0.8185, simple_loss=0.7143, pruned_loss=0.5645, over 3862257.36 frames. ], giga_tot_loss[loss=0.5259, simple_loss=0.5055, pruned_loss=0.2745, over 5690386.12 frames. ], batch size: 85, lr: 4.77e-02, grad_scale: 4.0
+2023-02-28 11:09:37,486 INFO [optim.py:369] (1/2) 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 11:10:00,050 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2334.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:10:04,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7810, 1.1142, 1.4931, 1.2541], device='cuda:1'), covar=tensor([0.3128, 0.4460, 0.3608, 0.6487], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0423, 0.0467, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005], device='cuda:1')
+2023-02-28 11:10:09,203 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2345.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:10:10,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2348.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:10:12,099 INFO [train.py:968] (1/2) Epoch 1, batch 2350, giga_loss[loss=0.473, simple_loss=0.4672, pruned_loss=0.2394, over 28515.00 frames. ], tot_loss[loss=0.5243, simple_loss=0.5053, pruned_loss=0.2725, over 5702537.85 frames. ], libri_tot_loss[loss=0.8128, simple_loss=0.7104, pruned_loss=0.5569, over 3912980.12 frames. ], giga_tot_loss[loss=0.5139, simple_loss=0.4976, pruned_loss=0.2662, over 5692948.44 frames. ], batch size: 78, lr: 4.76e-02, grad_scale: 4.0
+2023-02-28 11:10:33,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2377.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:10:35,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2379.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:10:50,396 INFO [train.py:968] (1/2) Epoch 1, batch 2400, giga_loss[loss=0.4401, simple_loss=0.4441, pruned_loss=0.2181, over 28874.00 frames. ], tot_loss[loss=0.5143, simple_loss=0.4986, pruned_loss=0.2656, over 5705016.05 frames. ], libri_tot_loss[loss=0.8116, simple_loss=0.7097, pruned_loss=0.5545, over 3932809.42 frames. ], giga_tot_loss[loss=0.5041, simple_loss=0.491, pruned_loss=0.2595, over 5695659.90 frames. ], batch size: 86, lr: 4.75e-02, grad_scale: 8.0
+2023-02-28 11:10:53,453 INFO [optim.py:369] (1/2) 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 11:11:20,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9116, 2.1352, 3.1864, 1.7158], device='cuda:1'), covar=tensor([0.1663, 0.1484, 0.0511, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0313, 0.0380, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 11:11:27,202 INFO [train.py:968] (1/2) Epoch 1, batch 2450, giga_loss[loss=0.4642, simple_loss=0.4558, pruned_loss=0.2363, over 28632.00 frames. ], tot_loss[loss=0.5081, simple_loss=0.4949, pruned_loss=0.2611, over 5712399.55 frames. ], libri_tot_loss[loss=0.8081, simple_loss=0.7076, pruned_loss=0.5485, over 3981534.44 frames. ], giga_tot_loss[loss=0.4951, simple_loss=0.4852, pruned_loss=0.2532, over 5700553.42 frames. ], batch size: 92, lr: 4.74e-02, grad_scale: 8.0
+2023-02-28 11:11:48,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2857, 0.8660, 1.4245, 0.4706], device='cuda:1'), covar=tensor([0.2999, 0.2984, 0.2009, 0.6610], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0317, 0.0362, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1')
+2023-02-28 11:11:48,547 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2477.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:11:50,571 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2480.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:12:05,856 INFO [train.py:968] (1/2) Epoch 1, batch 2500, giga_loss[loss=0.4395, simple_loss=0.4439, pruned_loss=0.2176, over 28425.00 frames. ], tot_loss[loss=0.4987, simple_loss=0.4887, pruned_loss=0.2547, over 5718303.75 frames. ], libri_tot_loss[loss=0.8059, simple_loss=0.7061, pruned_loss=0.545, over 4008626.36 frames. ], giga_tot_loss[loss=0.4858, simple_loss=0.4791, pruned_loss=0.2468, over 5707883.12 frames. ], batch size: 71, lr: 4.73e-02, grad_scale: 8.0
+2023-02-28 11:12:08,668 INFO [optim.py:369] (1/2) 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 11:12:12,211 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2509.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:12:21,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2522.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:12:23,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2525.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:12:42,081 INFO [train.py:968] (1/2) Epoch 1, batch 2550, giga_loss[loss=0.4761, simple_loss=0.4641, pruned_loss=0.244, over 28558.00 frames. ], tot_loss[loss=0.4964, simple_loss=0.4875, pruned_loss=0.2529, over 5724716.85 frames. ], libri_tot_loss[loss=0.8029, simple_loss=0.7042, pruned_loss=0.5392, over 4060567.06 frames. ], giga_tot_loss[loss=0.4794, simple_loss=0.475, pruned_loss=0.2423, over 5715016.38 frames. ], batch size: 60, lr: 4.72e-02, grad_scale: 8.0
+2023-02-28 11:12:45,210 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2554.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:13:20,160 INFO [train.py:968] (1/2) Epoch 1, batch 2600, giga_loss[loss=0.4593, simple_loss=0.4669, pruned_loss=0.2258, over 28758.00 frames. ], tot_loss[loss=0.4934, simple_loss=0.4856, pruned_loss=0.2508, over 5724513.28 frames. ], libri_tot_loss[loss=0.7987, simple_loss=0.7013, pruned_loss=0.5333, over 4104158.85 frames. ], giga_tot_loss[loss=0.4755, simple_loss=0.4725, pruned_loss=0.2396, over 5714478.32 frames. ], batch size: 119, lr: 4.71e-02, grad_scale: 8.0
+2023-02-28 11:13:22,951 INFO [optim.py:369] (1/2) 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 11:13:24,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2606.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:13:39,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2625.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:13:48,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2638.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:13:57,776 INFO [train.py:968] (1/2) Epoch 1, batch 2650, giga_loss[loss=0.4326, simple_loss=0.4487, pruned_loss=0.2082, over 28401.00 frames. ], tot_loss[loss=0.4923, simple_loss=0.486, pruned_loss=0.2495, over 5728177.32 frames. ], libri_tot_loss[loss=0.7926, simple_loss=0.6975, pruned_loss=0.5249, over 4164315.36 frames. ], giga_tot_loss[loss=0.4718, simple_loss=0.4708, pruned_loss=0.2367, over 5715922.26 frames. ], batch size: 65, lr: 4.70e-02, grad_scale: 8.0
+2023-02-28 11:14:35,946 INFO [train.py:968] (1/2) Epoch 1, batch 2700, giga_loss[loss=0.4283, simple_loss=0.4448, pruned_loss=0.2059, over 28451.00 frames. ], tot_loss[loss=0.5006, simple_loss=0.4918, pruned_loss=0.2548, over 5727740.92 frames. ], libri_tot_loss[loss=0.7857, simple_loss=0.6926, pruned_loss=0.5172, over 4215608.49 frames. ], giga_tot_loss[loss=0.4804, simple_loss=0.4771, pruned_loss=0.2421, over 5713367.46 frames. ], batch size: 71, lr: 4.69e-02, grad_scale: 8.0
+2023-02-28 11:14:39,230 INFO [optim.py:369] (1/2) 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 11:14:51,644 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6861, 1.8205, 1.2946, 1.0916], device='cuda:1'), covar=tensor([0.2342, 0.3585, 0.2506, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0373, 0.0296, 0.0279], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 11:15:02,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=2733.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:15:15,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2749.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:15:15,676 INFO [train.py:968] (1/2) Epoch 1, batch 2750, giga_loss[loss=0.5862, simple_loss=0.5562, pruned_loss=0.3081, over 28904.00 frames. ], tot_loss[loss=0.5096, simple_loss=0.4984, pruned_loss=0.2605, over 5723817.75 frames. ], libri_tot_loss[loss=0.7791, simple_loss=0.6879, pruned_loss=0.5102, over 4257142.60 frames. ], giga_tot_loss[loss=0.4915, simple_loss=0.4852, pruned_loss=0.2491, over 5708527.46 frames. ], batch size: 186, lr: 4.68e-02, grad_scale: 8.0
+2023-02-28 11:15:18,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2752.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:29,827 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2768.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:31,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2771.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:15:40,759 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2781.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:15:40,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2781.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:15:43,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2784.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:15:56,865 INFO [train.py:968] (1/2) Epoch 1, batch 2800, giga_loss[loss=0.6948, simple_loss=0.6053, pruned_loss=0.3921, over 26496.00 frames. ], tot_loss[loss=0.5238, simple_loss=0.5087, pruned_loss=0.2696, over 5704470.63 frames. ], libri_tot_loss[loss=0.7709, simple_loss=0.6824, pruned_loss=0.5015, over 4297979.65 frames. ], giga_tot_loss[loss=0.507, simple_loss=0.4964, pruned_loss=0.259, over 5700782.15 frames. ], batch size: 555, lr: 4.67e-02, grad_scale: 8.0
+2023-02-28 11:15:57,382 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2800.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:15:59,597 INFO [optim.py:369] (1/2) 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 11:16:09,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2813.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:16:38,320 INFO [train.py:968] (1/2) Epoch 1, batch 2850, giga_loss[loss=0.531, simple_loss=0.5284, pruned_loss=0.2668, over 28779.00 frames. ], tot_loss[loss=0.5327, simple_loss=0.5157, pruned_loss=0.2749, over 5703812.55 frames. ], libri_tot_loss[loss=0.7613, simple_loss=0.6758, pruned_loss=0.4919, over 4350466.72 frames. ], giga_tot_loss[loss=0.518, simple_loss=0.5048, pruned_loss=0.2657, over 5697118.25 frames. ], batch size: 119, lr: 4.66e-02, grad_scale: 8.0
+2023-02-28 11:17:23,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6358, 1.3844, 1.1012, 1.3533], device='cuda:1'), covar=tensor([0.2635, 0.2518, 0.2890, 0.4375], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0518, 0.0552, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005], device='cuda:1')
+2023-02-28 11:17:25,614 INFO [train.py:968] (1/2) Epoch 1, batch 2900, giga_loss[loss=0.5585, simple_loss=0.5126, pruned_loss=0.3022, over 23633.00 frames. ], tot_loss[loss=0.5384, simple_loss=0.5213, pruned_loss=0.2778, over 5709110.80 frames. ], libri_tot_loss[loss=0.7575, simple_loss=0.673, pruned_loss=0.4872, over 4388601.07 frames. ], giga_tot_loss[loss=0.5242, simple_loss=0.511, pruned_loss=0.2688, over 5699729.52 frames. ], batch size: 705, lr: 4.65e-02, grad_scale: 4.0
+2023-02-28 11:17:29,062 INFO [optim.py:369] (1/2) 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 11:17:44,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6513, 1.7205, 1.7964, 1.3446], device='cuda:1'), covar=tensor([0.3153, 0.3073, 0.2439, 0.4466], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0467, 0.0456, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 11:18:04,341 INFO [train.py:968] (1/2) Epoch 1, batch 2950, giga_loss[loss=0.7123, simple_loss=0.6256, pruned_loss=0.3995, over 26507.00 frames. ], tot_loss[loss=0.5442, simple_loss=0.5263, pruned_loss=0.2811, over 5708966.12 frames. ], libri_tot_loss[loss=0.7522, simple_loss=0.6694, pruned_loss=0.4819, over 4418166.48 frames. ], giga_tot_loss[loss=0.5326, simple_loss=0.5179, pruned_loss=0.2738, over 5698498.63 frames. ], batch size: 555, lr: 4.64e-02, grad_scale: 4.0
+2023-02-28 11:18:51,985 INFO [train.py:968] (1/2) Epoch 1, batch 3000, giga_loss[loss=0.5835, simple_loss=0.5556, pruned_loss=0.3057, over 28880.00 frames. ], tot_loss[loss=0.5491, simple_loss=0.5292, pruned_loss=0.2846, over 5686624.69 frames. ], libri_tot_loss[loss=0.7471, simple_loss=0.6657, pruned_loss=0.477, over 4446109.36 frames. ], giga_tot_loss[loss=0.5397, simple_loss=0.5224, pruned_loss=0.2786, over 5675848.21 frames. ], batch size: 99, lr: 4.63e-02, grad_scale: 4.0
+2023-02-28 11:18:51,985 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 11:18:57,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2135, 1.4039, 1.1373, 1.2954], device='cuda:1'), covar=tensor([0.1984, 0.2054, 0.2112, 0.2783], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0347, 0.0287, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1')
+2023-02-28 11:19:00,219 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 17535MB
+2023-02-28 11:19:04,571 INFO [optim.py:369] (1/2) 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 11:19:15,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 11:19:37,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5279, 1.6726, 1.7958, 1.3512], device='cuda:1'), covar=tensor([0.2711, 0.2097, 0.1567, 0.3727], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0459, 0.0448, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 11:19:38,809 INFO [train.py:968] (1/2) Epoch 1, batch 3050, giga_loss[loss=0.5056, simple_loss=0.509, pruned_loss=0.2511, over 28859.00 frames. ], tot_loss[loss=0.5301, simple_loss=0.5161, pruned_loss=0.272, over 5697975.16 frames. ], libri_tot_loss[loss=0.7378, simple_loss=0.6593, pruned_loss=0.4681, over 4494855.91 frames. ], giga_tot_loss[loss=0.5225, simple_loss=0.5108, pruned_loss=0.2671, over 5683952.28 frames. ], batch size: 112, lr: 4.62e-02, grad_scale: 4.0
+2023-02-28 11:19:48,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4804, 1.5313, 1.6674, 0.5386], device='cuda:1'), covar=tensor([0.0660, 0.0640, 0.0388, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0303, 0.0268, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-02-28 11:20:01,506 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3078.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:20:18,652 INFO [train.py:968] (1/2) Epoch 1, batch 3100, giga_loss[loss=0.4985, simple_loss=0.5034, pruned_loss=0.2468, over 28837.00 frames. ], tot_loss[loss=0.5208, simple_loss=0.5098, pruned_loss=0.2659, over 5705255.56 frames. ], libri_tot_loss[loss=0.7337, simple_loss=0.6564, pruned_loss=0.4639, over 4523132.26 frames. ], giga_tot_loss[loss=0.5137, simple_loss=0.5049, pruned_loss=0.2613, over 5690396.67 frames. ], batch size: 112, lr: 4.61e-02, grad_scale: 4.0
+2023-02-28 11:20:21,602 INFO [optim.py:369] (1/2) 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 11:20:24,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3108.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:20:35,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3887, 0.7953, 1.3466, 0.5109], device='cuda:1'), covar=tensor([0.1129, 0.1604, 0.1170, 0.2847], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0339, 0.0401, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 11:20:51,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-02-28 11:20:59,213 INFO [train.py:968] (1/2) Epoch 1, batch 3150, giga_loss[loss=0.4998, simple_loss=0.5011, pruned_loss=0.2493, over 28730.00 frames. ], tot_loss[loss=0.5172, simple_loss=0.5083, pruned_loss=0.263, over 5707579.47 frames. ], libri_tot_loss[loss=0.729, simple_loss=0.6532, pruned_loss=0.4591, over 4547048.63 frames. ], giga_tot_loss[loss=0.5103, simple_loss=0.5035, pruned_loss=0.2585, over 5701174.40 frames. ], batch size: 242, lr: 4.60e-02, grad_scale: 4.0
+2023-02-28 11:21:40,493 INFO [train.py:968] (1/2) Epoch 1, batch 3200, giga_loss[loss=0.5305, simple_loss=0.5234, pruned_loss=0.2687, over 28831.00 frames. ], tot_loss[loss=0.5205, simple_loss=0.5122, pruned_loss=0.2644, over 5705413.05 frames. ], libri_tot_loss[loss=0.7276, simple_loss=0.6523, pruned_loss=0.4574, over 4557184.47 frames. ], giga_tot_loss[loss=0.5143, simple_loss=0.5079, pruned_loss=0.2604, over 5701728.17 frames. ], batch size: 112, lr: 4.59e-02, grad_scale: 8.0
+2023-02-28 11:21:43,925 INFO [optim.py:369] (1/2) 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 11:21:52,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3217.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:22:14,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7445, 1.4122, 1.0624, 1.2567], device='cuda:1'), covar=tensor([0.1364, 0.1415, 0.1987, 0.2529], device='cuda:1'), in_proj_covar=tensor([0.0523, 0.0546, 0.0561, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006], device='cuda:1')
+2023-02-28 11:22:18,342 INFO [train.py:968] (1/2) Epoch 1, batch 3250, giga_loss[loss=0.527, simple_loss=0.5286, pruned_loss=0.2627, over 29077.00 frames. ], tot_loss[loss=0.5256, simple_loss=0.5165, pruned_loss=0.2674, over 5710280.00 frames. ], libri_tot_loss[loss=0.7235, simple_loss=0.6498, pruned_loss=0.4532, over 4583362.08 frames. ], giga_tot_loss[loss=0.5199, simple_loss=0.5123, pruned_loss=0.2638, over 5704473.49 frames. ], batch size: 155, lr: 4.58e-02, grad_scale: 8.0
+2023-02-28 11:22:20,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3251.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:22:21,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3254.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:22:38,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7510, 1.0511, 1.3985, 0.7659], device='cuda:1'), covar=tensor([0.0982, 0.1413, 0.1175, 0.2956], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0362, 0.0403, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1')
+2023-02-28 11:22:45,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3283.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:22:59,554 INFO [train.py:968] (1/2) Epoch 1, batch 3300, giga_loss[loss=0.4735, simple_loss=0.4837, pruned_loss=0.2316, over 28542.00 frames. ], tot_loss[loss=0.5265, simple_loss=0.5173, pruned_loss=0.2679, over 5702391.74 frames. ], libri_tot_loss[loss=0.7195, simple_loss=0.6472, pruned_loss=0.4487, over 4612819.05 frames. ], giga_tot_loss[loss=0.5203, simple_loss=0.5127, pruned_loss=0.264, over 5696322.65 frames. ], batch size: 71, lr: 4.57e-02, grad_scale: 8.0
+2023-02-28 11:23:02,907 INFO [optim.py:369] (1/2) 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 11:23:23,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3330.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:23:37,895 INFO [train.py:968] (1/2) Epoch 1, batch 3350, giga_loss[loss=0.4695, simple_loss=0.4862, pruned_loss=0.2264, over 28979.00 frames. ], tot_loss[loss=0.5243, simple_loss=0.516, pruned_loss=0.2663, over 5706079.62 frames. ], libri_tot_loss[loss=0.7161, simple_loss=0.645, pruned_loss=0.4451, over 4638003.64 frames. ], giga_tot_loss[loss=0.5184, simple_loss=0.5116, pruned_loss=0.2626, over 5698144.42 frames. ], batch size: 164, lr: 4.56e-02, grad_scale: 8.0
+2023-02-28 11:24:00,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5564, 1.3025, 1.8469, 0.6833], device='cuda:1'), covar=tensor([0.0742, 0.0971, 0.0422, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0233, 0.0206, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-02-28 11:24:03,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-02-28 11:24:19,732 INFO [train.py:968] (1/2) Epoch 1, batch 3400, giga_loss[loss=0.4985, simple_loss=0.5111, pruned_loss=0.243, over 28924.00 frames. ], tot_loss[loss=0.5204, simple_loss=0.5137, pruned_loss=0.2635, over 5717450.38 frames. ], libri_tot_loss[loss=0.7136, simple_loss=0.6434, pruned_loss=0.4424, over 4656681.41 frames. ], giga_tot_loss[loss=0.5148, simple_loss=0.5097, pruned_loss=0.26, over 5708856.88 frames. ], batch size: 174, lr: 4.55e-02, grad_scale: 8.0
+2023-02-28 11:24:23,540 INFO [optim.py:369] (1/2) 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 11:24:28,633 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 3450, giga_loss[loss=0.523, simple_loss=0.5115, pruned_loss=0.2672, over 28731.00 frames. ], tot_loss[loss=0.5219, simple_loss=0.5151, pruned_loss=0.2644, over 5725312.72 frames. ], libri_tot_loss[loss=0.7058, simple_loss=0.6384, pruned_loss=0.4347, over 4703070.89 frames. ], giga_tot_loss[loss=0.5163, simple_loss=0.511, pruned_loss=0.2608, over 5714260.55 frames. ], batch size: 119, lr: 4.54e-02, grad_scale: 4.0
+2023-02-28 11:25:01,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3453.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:25:09,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 11:25:13,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3106, 1.2086, 0.9764, 1.0812], device='cuda:1'), covar=tensor([0.1463, 0.1402, 0.2022, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0564, 0.0572, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0004, 0.0006], device='cuda:1')
+2023-02-28 11:25:25,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-02-28 11:25:34,850 INFO [train.py:968] (1/2) Epoch 1, batch 3500, giga_loss[loss=0.4824, simple_loss=0.4926, pruned_loss=0.2361, over 28852.00 frames. ], tot_loss[loss=0.5203, simple_loss=0.5152, pruned_loss=0.2628, over 5705927.20 frames. ], libri_tot_loss[loss=0.7008, simple_loss=0.6349, pruned_loss=0.4298, over 4718947.95 frames. ], giga_tot_loss[loss=0.515, simple_loss=0.5113, pruned_loss=0.2594, over 5711375.79 frames. ], batch size: 119, lr: 4.53e-02, grad_scale: 4.0
+2023-02-28 11:25:38,720 INFO [optim.py:369] (1/2) 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 11:25:58,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7395, 2.7402, 4.7137, 2.4320], device='cuda:1'), covar=tensor([0.1349, 0.1227, 0.0177, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0359, 0.0410, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 11:26:07,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-02-28 11:26:11,900 INFO [train.py:968] (1/2) Epoch 1, batch 3550, giga_loss[loss=0.5196, simple_loss=0.5308, pruned_loss=0.2543, over 28707.00 frames. ], tot_loss[loss=0.5179, simple_loss=0.515, pruned_loss=0.2603, over 5707597.97 frames. ], libri_tot_loss[loss=0.6968, simple_loss=0.6325, pruned_loss=0.4257, over 4746753.62 frames. ], giga_tot_loss[loss=0.5124, simple_loss=0.5111, pruned_loss=0.2569, over 5708970.01 frames. ], batch size: 262, lr: 4.51e-02, grad_scale: 4.0
+2023-02-28 11:26:46,829 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.95 vs. limit=5.0
+2023-02-28 11:26:47,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3592.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:26:49,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5018, 1.4445, 1.2635, 1.2253], device='cuda:1'), covar=tensor([0.1306, 0.2074, 0.1230, 0.2626], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0477, 0.0366, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 11:26:51,247 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3596.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:26:54,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3599.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:26:54,348 INFO [train.py:968] (1/2) Epoch 1, batch 3600, giga_loss[loss=0.5132, simple_loss=0.5092, pruned_loss=0.2586, over 29141.00 frames. ], tot_loss[loss=0.5135, simple_loss=0.5127, pruned_loss=0.2571, over 5710100.45 frames. ], libri_tot_loss[loss=0.6957, simple_loss=0.6316, pruned_loss=0.4247, over 4752408.86 frames. ], giga_tot_loss[loss=0.5092, simple_loss=0.5097, pruned_loss=0.2544, over 5710460.47 frames. ], batch size: 113, lr: 4.50e-02, grad_scale: 8.0
+2023-02-28 11:26:58,389 INFO [optim.py:369] (1/2) 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 11:27:14,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3628.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:27:33,577 INFO [train.py:968] (1/2) Epoch 1, batch 3650, giga_loss[loss=0.4476, simple_loss=0.4659, pruned_loss=0.2146, over 28716.00 frames. ], tot_loss[loss=0.504, simple_loss=0.5066, pruned_loss=0.2507, over 5716992.30 frames. ], libri_tot_loss[loss=0.6942, simple_loss=0.6307, pruned_loss=0.4231, over 4760532.90 frames. ], giga_tot_loss[loss=0.5002, simple_loss=0.5038, pruned_loss=0.2482, over 5718708.56 frames. ], batch size: 99, lr: 4.49e-02, grad_scale: 8.0
+2023-02-28 11:27:42,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1606, 2.3155, 4.2071, 2.1054], device='cuda:1'), covar=tensor([0.1570, 0.1272, 0.0208, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0391, 0.0447, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 11:27:58,487 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3681.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:28:12,488 INFO [train.py:968] (1/2) Epoch 1, batch 3700, giga_loss[loss=0.5461, simple_loss=0.5376, pruned_loss=0.2773, over 28248.00 frames. ], tot_loss[loss=0.5091, simple_loss=0.5083, pruned_loss=0.2549, over 5715524.53 frames. ], libri_tot_loss[loss=0.6921, simple_loss=0.6291, pruned_loss=0.421, over 4777743.52 frames. ], giga_tot_loss[loss=0.5049, simple_loss=0.5056, pruned_loss=0.2522, over 5714155.65 frames. ], batch size: 368, lr: 4.48e-02, grad_scale: 8.0
+2023-02-28 11:28:16,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3705.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:28:16,524 INFO [optim.py:369] (1/2) 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 11:28:38,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3735.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:28:40,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3738.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:28:43,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5483, 1.1521, 1.6056, 0.6927], device='cuda:1'), covar=tensor([0.1398, 0.1261, 0.1049, 0.2975], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0362, 0.0436, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 11:28:48,314 INFO [train.py:968] (1/2) Epoch 1, batch 3750, giga_loss[loss=0.4367, simple_loss=0.465, pruned_loss=0.2042, over 28598.00 frames. ], tot_loss[loss=0.5056, simple_loss=0.5062, pruned_loss=0.2525, over 5721028.59 frames. ], libri_tot_loss[loss=0.6891, simple_loss=0.6271, pruned_loss=0.4177, over 4801226.43 frames. ], giga_tot_loss[loss=0.5004, simple_loss=0.5028, pruned_loss=0.249, over 5719476.48 frames. ], batch size: 60, lr: 4.47e-02, grad_scale: 8.0
+2023-02-28 11:29:03,456 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3767.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:29:10,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-02-28 11:29:19,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3787.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:29:29,547 INFO [train.py:968] (1/2) Epoch 1, batch 3800, giga_loss[loss=0.4592, simple_loss=0.4745, pruned_loss=0.222, over 28859.00 frames. ], tot_loss[loss=0.504, simple_loss=0.5052, pruned_loss=0.2514, over 5723419.84 frames. ], libri_tot_loss[loss=0.685, simple_loss=0.6243, pruned_loss=0.4138, over 4827062.72 frames. ], giga_tot_loss[loss=0.4987, simple_loss=0.5018, pruned_loss=0.2478, over 5719548.16 frames. ], batch size: 92, lr: 4.46e-02, grad_scale: 8.0
+2023-02-28 11:29:35,361 INFO [optim.py:369] (1/2) 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 11:30:05,491 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 3850, giga_loss[loss=0.467, simple_loss=0.4841, pruned_loss=0.2249, over 28967.00 frames. ], tot_loss[loss=0.5055, simple_loss=0.5067, pruned_loss=0.2521, over 5728059.53 frames. ], libri_tot_loss[loss=0.6828, simple_loss=0.623, pruned_loss=0.4113, over 4845665.53 frames. ], giga_tot_loss[loss=0.4998, simple_loss=0.5029, pruned_loss=0.2483, over 5724239.46 frames. ], batch size: 106, lr: 4.45e-02, grad_scale: 8.0
+2023-02-28 11:30:09,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3851.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:30:31,135 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-02-28 11:30:31,468 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 3900, giga_loss[loss=0.4701, simple_loss=0.4854, pruned_loss=0.2274, over 28744.00 frames. ], tot_loss[loss=0.5, simple_loss=0.5039, pruned_loss=0.248, over 5723352.31 frames. ], libri_tot_loss[loss=0.6788, simple_loss=0.6203, pruned_loss=0.4076, over 4866504.27 frames. ], giga_tot_loss[loss=0.4952, simple_loss=0.5008, pruned_loss=0.2448, over 5717375.58 frames. ], batch size: 99, lr: 4.44e-02, grad_scale: 8.0
+2023-02-28 11:30:52,815 INFO [optim.py:369] (1/2) 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 11:31:10,886 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3930.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:31:11,680 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-02-28 11:31:13,139 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=3939.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:31:26,084 INFO [train.py:968] (1/2) Epoch 1, batch 3950, giga_loss[loss=0.4526, simple_loss=0.4777, pruned_loss=0.2137, over 28561.00 frames. ], tot_loss[loss=0.4994, simple_loss=0.5033, pruned_loss=0.2478, over 5725421.05 frames. ], libri_tot_loss[loss=0.676, simple_loss=0.6185, pruned_loss=0.4048, over 4887071.38 frames. ], giga_tot_loss[loss=0.4946, simple_loss=0.5001, pruned_loss=0.2445, over 5717592.54 frames. ], batch size: 85, lr: 4.43e-02, grad_scale: 4.0
+2023-02-28 11:31:35,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3962.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:32:04,494 INFO [train.py:968] (1/2) Epoch 1, batch 4000, giga_loss[loss=0.3982, simple_loss=0.4289, pruned_loss=0.1837, over 28471.00 frames. ], tot_loss[loss=0.4958, simple_loss=0.5002, pruned_loss=0.2457, over 5726264.01 frames. ], libri_tot_loss[loss=0.6697, simple_loss=0.6141, pruned_loss=0.3989, over 4923679.57 frames. ], giga_tot_loss[loss=0.4903, simple_loss=0.4968, pruned_loss=0.2419, over 5716424.82 frames. ], batch size: 60, lr: 4.42e-02, grad_scale: 8.0
+2023-02-28 11:32:09,444 INFO [optim.py:369] (1/2) 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 11:32:43,295 INFO [train.py:968] (1/2) Epoch 1, batch 4050, giga_loss[loss=0.4766, simple_loss=0.4949, pruned_loss=0.2292, over 28645.00 frames. ], tot_loss[loss=0.4916, simple_loss=0.4965, pruned_loss=0.2433, over 5719796.37 frames. ], libri_tot_loss[loss=0.6678, simple_loss=0.6128, pruned_loss=0.397, over 4937578.36 frames. ], giga_tot_loss[loss=0.4864, simple_loss=0.4933, pruned_loss=0.2397, over 5710289.93 frames. ], batch size: 262, lr: 4.41e-02, grad_scale: 8.0
+2023-02-28 11:32:47,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4056.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:32:57,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4069.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:33:12,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-02-28 11:33:19,599 INFO [train.py:968] (1/2) Epoch 1, batch 4100, libri_loss[loss=0.5583, simple_loss=0.5492, pruned_loss=0.2836, over 28617.00 frames. ], tot_loss[loss=0.4877, simple_loss=0.4931, pruned_loss=0.2411, over 5717472.60 frames. ], libri_tot_loss[loss=0.6635, simple_loss=0.6101, pruned_loss=0.3926, over 4966329.10 frames. ], giga_tot_loss[loss=0.4811, simple_loss=0.4888, pruned_loss=0.2367, over 5707965.08 frames. ], batch size: 106, lr: 4.40e-02, grad_scale: 4.0
+2023-02-28 11:33:25,010 INFO [optim.py:369] (1/2) 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 11:33:58,406 INFO [train.py:968] (1/2) Epoch 1, batch 4150, giga_loss[loss=0.5034, simple_loss=0.4984, pruned_loss=0.2542, over 28875.00 frames. ], tot_loss[loss=0.4899, simple_loss=0.4938, pruned_loss=0.243, over 5715857.33 frames. ], libri_tot_loss[loss=0.6593, simple_loss=0.6074, pruned_loss=0.3887, over 4992225.42 frames. ], giga_tot_loss[loss=0.4831, simple_loss=0.4892, pruned_loss=0.2385, over 5704825.75 frames. ], batch size: 186, lr: 4.39e-02, grad_scale: 4.0
+2023-02-28 11:34:36,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4199.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:34:37,279 INFO [train.py:968] (1/2) Epoch 1, batch 4200, giga_loss[loss=0.4591, simple_loss=0.4689, pruned_loss=0.2246, over 28868.00 frames. ], tot_loss[loss=0.4892, simple_loss=0.4923, pruned_loss=0.2431, over 5713412.00 frames. ], libri_tot_loss[loss=0.6553, simple_loss=0.6047, pruned_loss=0.385, over 5014676.32 frames. ], giga_tot_loss[loss=0.483, simple_loss=0.4881, pruned_loss=0.239, over 5700629.67 frames. ], batch size: 186, lr: 4.38e-02, grad_scale: 4.0
+2023-02-28 11:34:39,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4202.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:34:44,098 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4231.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:35:11,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-02-28 11:35:17,645 INFO [train.py:968] (1/2) Epoch 1, batch 4250, giga_loss[loss=0.5086, simple_loss=0.4991, pruned_loss=0.2591, over 28789.00 frames. ], tot_loss[loss=0.487, simple_loss=0.4896, pruned_loss=0.2422, over 5710661.07 frames. ], libri_tot_loss[loss=0.6542, simple_loss=0.6041, pruned_loss=0.3838, over 5023542.32 frames. ], giga_tot_loss[loss=0.4815, simple_loss=0.4858, pruned_loss=0.2386, over 5698965.43 frames. ], batch size: 284, lr: 4.36e-02, grad_scale: 4.0
+2023-02-28 11:35:44,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3706, 1.6851, 0.8994, 1.5322], device='cuda:1'), covar=tensor([0.1030, 0.2128, 0.2023, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0575, 0.0518, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002], device='cuda:1')
+2023-02-28 11:35:55,565 INFO [train.py:968] (1/2) Epoch 1, batch 4300, libri_loss[loss=0.5854, simple_loss=0.5688, pruned_loss=0.301, over 29379.00 frames. ], tot_loss[loss=0.4831, simple_loss=0.4861, pruned_loss=0.2401, over 5716353.11 frames. ], libri_tot_loss[loss=0.6526, simple_loss=0.6031, pruned_loss=0.382, over 5038803.31 frames. ], giga_tot_loss[loss=0.4768, simple_loss=0.4817, pruned_loss=0.236, over 5705566.34 frames. ], batch size: 92, lr: 4.35e-02, grad_scale: 4.0
+2023-02-28 11:36:01,881 INFO [optim.py:369] (1/2) 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 11:36:06,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4314.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:36:07,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1867, 3.0280, 4.5164, 2.3708], device='cuda:1'), covar=tensor([0.1359, 0.0837, 0.0226, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0365, 0.0442, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 11:36:09,327 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4319.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:36:31,185 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-02-28 11:36:31,996 INFO [train.py:968] (1/2) Epoch 1, batch 4350, giga_loss[loss=0.4241, simple_loss=0.4492, pruned_loss=0.1995, over 28870.00 frames. ], tot_loss[loss=0.4791, simple_loss=0.4823, pruned_loss=0.238, over 5708494.44 frames. ], libri_tot_loss[loss=0.6495, simple_loss=0.601, pruned_loss=0.3791, over 5050552.94 frames. ], giga_tot_loss[loss=0.4731, simple_loss=0.4781, pruned_loss=0.234, over 5703797.14 frames. ], batch size: 186, lr: 4.34e-02, grad_scale: 4.0
+2023-02-28 11:37:10,509 INFO [train.py:968] (1/2) Epoch 1, batch 4400, giga_loss[loss=0.4577, simple_loss=0.4679, pruned_loss=0.2237, over 28982.00 frames. ], tot_loss[loss=0.4759, simple_loss=0.4802, pruned_loss=0.2358, over 5697333.89 frames. ], libri_tot_loss[loss=0.6473, simple_loss=0.5997, pruned_loss=0.3766, over 5056260.80 frames. ], giga_tot_loss[loss=0.4687, simple_loss=0.4748, pruned_loss=0.2312, over 5706236.51 frames. ], batch size: 136, lr: 4.33e-02, grad_scale: 8.0
+2023-02-28 11:37:17,194 INFO [optim.py:369] (1/2) 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 11:37:32,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-02-28 11:37:35,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5234, 1.9896, 3.4336, 1.6460], device='cuda:1'), covar=tensor([0.0787, 0.1252, 0.0662, 0.1528], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0381, 0.0545, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0005, 0.0003], device='cuda:1')
+2023-02-28 11:37:46,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4444.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:37:50,950 INFO [train.py:968] (1/2) Epoch 1, batch 4450, giga_loss[loss=0.5338, simple_loss=0.5189, pruned_loss=0.2743, over 28930.00 frames. ], tot_loss[loss=0.4749, simple_loss=0.4805, pruned_loss=0.2347, over 5703679.90 frames. ], libri_tot_loss[loss=0.6447, simple_loss=0.598, pruned_loss=0.3742, over 5073368.01 frames. ], giga_tot_loss[loss=0.4681, simple_loss=0.4754, pruned_loss=0.2304, over 5706958.89 frames. ], batch size: 186, lr: 4.32e-02, grad_scale: 8.0
+2023-02-28 11:37:56,709 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4457.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:37:58,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4460.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:38:22,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4489.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 11:38:30,108 INFO [train.py:968] (1/2) Epoch 1, batch 4500, giga_loss[loss=0.4352, simple_loss=0.4611, pruned_loss=0.2046, over 28995.00 frames. ], tot_loss[loss=0.481, simple_loss=0.4853, pruned_loss=0.2383, over 5683721.90 frames. ], libri_tot_loss[loss=0.6427, simple_loss=0.5968, pruned_loss=0.3718, over 5082093.95 frames. ], giga_tot_loss[loss=0.4722, simple_loss=0.4789, pruned_loss=0.2328, over 5698165.43 frames. ], batch size: 136, lr: 4.31e-02, grad_scale: 4.0
+2023-02-28 11:38:37,529 INFO [optim.py:369] (1/2) 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 11:38:57,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4534.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:39:07,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-02-28 11:39:08,436 INFO [train.py:968] (1/2) Epoch 1, batch 4550, giga_loss[loss=0.4134, simple_loss=0.4535, pruned_loss=0.1867, over 28818.00 frames. ], tot_loss[loss=0.4818, simple_loss=0.4872, pruned_loss=0.2382, over 5697477.21 frames. ], libri_tot_loss[loss=0.6383, simple_loss=0.594, pruned_loss=0.3676, over 5113721.11 frames. ], giga_tot_loss[loss=0.4722, simple_loss=0.4801, pruned_loss=0.2322, over 5701867.98 frames. ], batch size: 186, lr: 4.30e-02, grad_scale: 4.0
+2023-02-28 11:39:42,468 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4587.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:39:45,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4590.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 11:39:48,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8841, 1.4227, 1.4146, 1.2933], device='cuda:1'), covar=tensor([0.1166, 0.2986, 0.1891, 0.3687], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0608, 0.0462, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0005, 0.0004, 0.0005], device='cuda:1')
+2023-02-28 11:39:52,932 INFO [train.py:968] (1/2) Epoch 1, batch 4600, giga_loss[loss=0.4595, simple_loss=0.4786, pruned_loss=0.2202, over 28686.00 frames. ], tot_loss[loss=0.4775, simple_loss=0.4855, pruned_loss=0.2348, over 5690842.29 frames. ], libri_tot_loss[loss=0.6375, simple_loss=0.5934, pruned_loss=0.3669, over 5117711.50 frames. ], giga_tot_loss[loss=0.47, simple_loss=0.4799, pruned_loss=0.23, over 5693424.71 frames. ], batch size: 262, lr: 4.29e-02, grad_scale: 4.0
+2023-02-28 11:39:59,436 INFO [optim.py:369] (1/2) 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 11:40:06,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6316, 2.0066, 2.2383, 1.3729], device='cuda:1'), covar=tensor([0.1002, 0.1742, 0.0593, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0339, 0.0269, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001], device='cuda:1')
+2023-02-28 11:40:08,606 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 4650, giga_loss[loss=0.4682, simple_loss=0.4759, pruned_loss=0.2302, over 28852.00 frames. ], tot_loss[loss=0.4822, simple_loss=0.4884, pruned_loss=0.238, over 5693355.98 frames. ], libri_tot_loss[loss=0.6353, simple_loss=0.5918, pruned_loss=0.3645, over 5142682.59 frames. ], giga_tot_loss[loss=0.4729, simple_loss=0.4818, pruned_loss=0.232, over 5690714.45 frames. ], batch size: 112, lr: 4.28e-02, grad_scale: 4.0
+2023-02-28 11:40:54,592 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=4681.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:41:04,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4694.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:41:08,876 INFO [train.py:968] (1/2) Epoch 1, batch 4700, giga_loss[loss=0.4347, simple_loss=0.4576, pruned_loss=0.206, over 28829.00 frames. ], tot_loss[loss=0.4815, simple_loss=0.4878, pruned_loss=0.2376, over 5694871.10 frames. ], libri_tot_loss[loss=0.632, simple_loss=0.5897, pruned_loss=0.3612, over 5161021.25 frames. ], giga_tot_loss[loss=0.4712, simple_loss=0.4805, pruned_loss=0.231, over 5696524.99 frames. ], batch size: 186, lr: 4.27e-02, grad_scale: 4.0
+2023-02-28 11:41:15,057 INFO [optim.py:369] (1/2) 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 11:41:47,058 INFO [train.py:968] (1/2) Epoch 1, batch 4750, giga_loss[loss=0.4967, simple_loss=0.4951, pruned_loss=0.2492, over 28807.00 frames. ], tot_loss[loss=0.4842, simple_loss=0.4893, pruned_loss=0.2396, over 5698006.26 frames. ], libri_tot_loss[loss=0.6259, simple_loss=0.5855, pruned_loss=0.3559, over 5193485.16 frames. ], giga_tot_loss[loss=0.4747, simple_loss=0.4826, pruned_loss=0.2334, over 5691327.48 frames. ], batch size: 119, lr: 4.26e-02, grad_scale: 4.0
+2023-02-28 11:42:26,055 INFO [train.py:968] (1/2) Epoch 1, batch 4800, giga_loss[loss=0.4817, simple_loss=0.4878, pruned_loss=0.2378, over 27951.00 frames. ], tot_loss[loss=0.4849, simple_loss=0.4896, pruned_loss=0.2401, over 5696418.56 frames. ], libri_tot_loss[loss=0.6248, simple_loss=0.5847, pruned_loss=0.3548, over 5200650.27 frames. ], giga_tot_loss[loss=0.4765, simple_loss=0.4837, pruned_loss=0.2346, over 5691513.88 frames. ], batch size: 412, lr: 4.25e-02, grad_scale: 8.0
+2023-02-28 11:42:35,671 INFO [optim.py:369] (1/2) 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 11:42:56,959 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4837.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:42:58,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4840.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:43:06,451 INFO [train.py:968] (1/2) Epoch 1, batch 4850, giga_loss[loss=0.4725, simple_loss=0.487, pruned_loss=0.229, over 28875.00 frames. ], tot_loss[loss=0.4886, simple_loss=0.493, pruned_loss=0.242, over 5699276.62 frames. ], libri_tot_loss[loss=0.6238, simple_loss=0.5842, pruned_loss=0.3537, over 5211313.53 frames. ], giga_tot_loss[loss=0.4807, simple_loss=0.4875, pruned_loss=0.237, over 5692270.81 frames. ], batch size: 199, lr: 4.24e-02, grad_scale: 8.0
+2023-02-28 11:43:22,065 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4869.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:43:33,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0009, 1.9713, 2.4007, 1.6597], device='cuda:1'), covar=tensor([0.1825, 0.1523, 0.1173, 0.3023], device='cuda:1'), in_proj_covar=tensor([0.0529, 0.0548, 0.0549, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 11:43:45,539 INFO [train.py:968] (1/2) Epoch 1, batch 4900, giga_loss[loss=0.4586, simple_loss=0.4839, pruned_loss=0.2166, over 28711.00 frames. ], tot_loss[loss=0.492, simple_loss=0.4958, pruned_loss=0.2441, over 5709254.62 frames. ], libri_tot_loss[loss=0.6222, simple_loss=0.5832, pruned_loss=0.3521, over 5223516.64 frames. ], giga_tot_loss[loss=0.4846, simple_loss=0.4905, pruned_loss=0.2393, over 5701552.63 frames. ], batch size: 284, lr: 4.23e-02, grad_scale: 8.0
+2023-02-28 11:43:51,569 INFO [optim.py:369] (1/2) 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 11:43:51,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4909.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:44:07,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 11:44:23,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6341, 1.4056, 1.6749, 0.7395], device='cuda:1'), covar=tensor([0.0752, 0.0588, 0.0599, 0.1433], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0402, 0.0457, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 11:44:26,301 INFO [train.py:968] (1/2) Epoch 1, batch 4950, giga_loss[loss=0.4825, simple_loss=0.4913, pruned_loss=0.2368, over 28631.00 frames. ], tot_loss[loss=0.486, simple_loss=0.4925, pruned_loss=0.2397, over 5712878.60 frames. ], libri_tot_loss[loss=0.6215, simple_loss=0.5827, pruned_loss=0.3515, over 5230311.27 frames. ], giga_tot_loss[loss=0.4795, simple_loss=0.4879, pruned_loss=0.2355, over 5704959.59 frames. ], batch size: 336, lr: 4.22e-02, grad_scale: 8.0
+2023-02-28 11:45:05,727 INFO [train.py:968] (1/2) Epoch 1, batch 5000, giga_loss[loss=0.4678, simple_loss=0.4903, pruned_loss=0.2227, over 28915.00 frames. ], tot_loss[loss=0.4825, simple_loss=0.491, pruned_loss=0.237, over 5721942.33 frames. ], libri_tot_loss[loss=0.6197, simple_loss=0.5817, pruned_loss=0.3496, over 5246857.24 frames. ], giga_tot_loss[loss=0.4758, simple_loss=0.4862, pruned_loss=0.2327, over 5711756.78 frames. ], batch size: 164, lr: 4.20e-02, grad_scale: 8.0
+2023-02-28 11:45:12,617 INFO [optim.py:369] (1/2) 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 11:45:25,076 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-02-28 11:45:43,896 INFO [train.py:968] (1/2) Epoch 1, batch 5050, giga_loss[loss=0.4715, simple_loss=0.4961, pruned_loss=0.2235, over 28632.00 frames. ], tot_loss[loss=0.4812, simple_loss=0.4902, pruned_loss=0.2361, over 5726317.88 frames. ], libri_tot_loss[loss=0.6176, simple_loss=0.5804, pruned_loss=0.3476, over 5261304.68 frames. ], giga_tot_loss[loss=0.4744, simple_loss=0.4853, pruned_loss=0.2318, over 5715930.11 frames. ], batch size: 242, lr: 4.19e-02, grad_scale: 8.0
+2023-02-28 11:45:45,738 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5052.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:45:48,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5055.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:45:48,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=5056.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:46:10,821 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5084.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:46:23,691 INFO [train.py:968] (1/2) Epoch 1, batch 5100, giga_loss[loss=0.3895, simple_loss=0.4255, pruned_loss=0.1768, over 28756.00 frames. ], tot_loss[loss=0.4732, simple_loss=0.4847, pruned_loss=0.2309, over 5723194.88 frames. ], libri_tot_loss[loss=0.6159, simple_loss=0.5792, pruned_loss=0.3461, over 5270070.55 frames. ], giga_tot_loss[loss=0.4674, simple_loss=0.4806, pruned_loss=0.2271, over 5713330.94 frames. ], batch size: 119, lr: 4.18e-02, grad_scale: 8.0
+2023-02-28 11:46:31,064 INFO [optim.py:369] (1/2) 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 11:47:06,168 INFO [train.py:968] (1/2) Epoch 1, batch 5150, giga_loss[loss=0.4124, simple_loss=0.4329, pruned_loss=0.1959, over 28672.00 frames. ], tot_loss[loss=0.4689, simple_loss=0.481, pruned_loss=0.2284, over 5721792.72 frames. ], libri_tot_loss[loss=0.6132, simple_loss=0.5773, pruned_loss=0.3437, over 5278178.83 frames. ], giga_tot_loss[loss=0.4627, simple_loss=0.4768, pruned_loss=0.2243, over 5717952.24 frames. ], batch size: 85, lr: 4.17e-02, grad_scale: 8.0
+2023-02-28 11:47:48,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5199.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:47:48,706 INFO [train.py:968] (1/2) Epoch 1, batch 5200, giga_loss[loss=0.5387, simple_loss=0.5104, pruned_loss=0.2835, over 26544.00 frames. ], tot_loss[loss=0.4613, simple_loss=0.4753, pruned_loss=0.2236, over 5722850.65 frames. ], libri_tot_loss[loss=0.6128, simple_loss=0.5771, pruned_loss=0.3434, over 5281312.94 frames. ], giga_tot_loss[loss=0.4561, simple_loss=0.4718, pruned_loss=0.2202, over 5719022.91 frames. ], batch size: 555, lr: 4.16e-02, grad_scale: 8.0
+2023-02-28 11:47:50,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5202.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:47:55,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5152, 1.0132, 1.4823, 0.9160], device='cuda:1'), covar=tensor([0.0626, 0.0459, 0.0470, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0558, 0.0413, 0.0469, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 11:47:56,879 INFO [optim.py:369] (1/2) 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 11:48:12,201 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=5231.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 11:48:20,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-02-28 11:48:28,786 INFO [train.py:968] (1/2) Epoch 1, batch 5250, giga_loss[loss=0.4958, simple_loss=0.5099, pruned_loss=0.2408, over 28666.00 frames. ], tot_loss[loss=0.4626, simple_loss=0.4768, pruned_loss=0.2242, over 5718853.81 frames. ], libri_tot_loss[loss=0.6103, simple_loss=0.5753, pruned_loss=0.3411, over 5298233.85 frames. ], giga_tot_loss[loss=0.4564, simple_loss=0.4727, pruned_loss=0.2201, over 5712086.91 frames. ], batch size: 307, lr: 4.15e-02, grad_scale: 8.0
+2023-02-28 11:49:07,522 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-02-28 11:49:12,211 INFO [train.py:968] (1/2) Epoch 1, batch 5300, giga_loss[loss=0.502, simple_loss=0.5093, pruned_loss=0.2474, over 28929.00 frames. ], tot_loss[loss=0.4648, simple_loss=0.4803, pruned_loss=0.2247, over 5704566.46 frames. ], libri_tot_loss[loss=0.6088, simple_loss=0.5744, pruned_loss=0.3397, over 5297462.51 frames. ], giga_tot_loss[loss=0.4587, simple_loss=0.4762, pruned_loss=0.2207, over 5706485.53 frames. ], batch size: 227, lr: 4.14e-02, grad_scale: 8.0
+2023-02-28 11:49:19,527 INFO [optim.py:369] (1/2) 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 11:49:27,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0764, 1.9491, 1.3763, 1.4018], device='cuda:1'), covar=tensor([0.1007, 0.0985, 0.1456, 0.1968], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0680, 0.0647, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 11:49:47,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-02-28 11:49:47,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 11:49:50,463 INFO [train.py:968] (1/2) Epoch 1, batch 5350, giga_loss[loss=0.531, simple_loss=0.5249, pruned_loss=0.2686, over 27884.00 frames. ], tot_loss[loss=0.4702, simple_loss=0.4826, pruned_loss=0.2289, over 5699206.45 frames. ], libri_tot_loss[loss=0.6046, simple_loss=0.5712, pruned_loss=0.3364, over 5314301.71 frames. ], giga_tot_loss[loss=0.4636, simple_loss=0.4786, pruned_loss=0.2243, over 5699996.05 frames. ], batch size: 412, lr: 4.13e-02, grad_scale: 4.0
+2023-02-28 11:50:19,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=2.03 vs. limit=2.0
+2023-02-28 11:50:30,170 INFO [train.py:968] (1/2) Epoch 1, batch 5400, giga_loss[loss=0.4686, simple_loss=0.4719, pruned_loss=0.2326, over 28966.00 frames. ], tot_loss[loss=0.4717, simple_loss=0.4824, pruned_loss=0.2305, over 5706122.94 frames. ], libri_tot_loss[loss=0.6037, simple_loss=0.5707, pruned_loss=0.3355, over 5323080.58 frames. ], giga_tot_loss[loss=0.4654, simple_loss=0.4784, pruned_loss=0.2262, over 5704134.41 frames. ], batch size: 136, lr: 4.12e-02, grad_scale: 4.0
+2023-02-28 11:50:38,955 INFO [optim.py:369] (1/2) 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 11:50:40,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-02-28 11:50:50,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7627, 1.9074, 1.9076, 1.3710], device='cuda:1'), covar=tensor([0.0813, 0.1561, 0.0814, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0458, 0.0327, 0.0309], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 11:51:11,242 INFO [train.py:968] (1/2) Epoch 1, batch 5450, giga_loss[loss=0.5064, simple_loss=0.4969, pruned_loss=0.2579, over 28384.00 frames. ], tot_loss[loss=0.4705, simple_loss=0.4799, pruned_loss=0.2306, over 5703453.64 frames. ], libri_tot_loss[loss=0.6001, simple_loss=0.5682, pruned_loss=0.3325, over 5339310.95 frames. ], giga_tot_loss[loss=0.4649, simple_loss=0.4764, pruned_loss=0.2267, over 5697494.05 frames. ], batch size: 368, lr: 4.11e-02, grad_scale: 4.0
+2023-02-28 11:51:19,547 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-02-28 11:51:46,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8562, 2.0228, 3.0103, 1.8776], device='cuda:1'), covar=tensor([0.0609, 0.1153, 0.0804, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0433, 0.0594, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0006, 0.0004], device='cuda:1')
+2023-02-28 11:51:54,009 INFO [train.py:968] (1/2) Epoch 1, batch 5500, giga_loss[loss=0.429, simple_loss=0.4502, pruned_loss=0.2039, over 28994.00 frames. ], tot_loss[loss=0.4662, simple_loss=0.4758, pruned_loss=0.2283, over 5703822.66 frames. ], libri_tot_loss[loss=0.5985, simple_loss=0.5672, pruned_loss=0.3312, over 5347853.37 frames. ], giga_tot_loss[loss=0.4613, simple_loss=0.4727, pruned_loss=0.225, over 5696466.51 frames. ], batch size: 164, lr: 4.10e-02, grad_scale: 4.0
+2023-02-28 11:52:02,433 INFO [optim.py:369] (1/2) 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 11:52:28,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4929, 1.5768, 0.9320, 1.1136], device='cuda:1'), covar=tensor([0.1191, 0.0979, 0.1705, 0.1902], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0677, 0.0648, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 11:52:33,354 INFO [train.py:968] (1/2) Epoch 1, batch 5550, giga_loss[loss=0.5199, simple_loss=0.495, pruned_loss=0.2724, over 24424.00 frames. ], tot_loss[loss=0.4662, simple_loss=0.4751, pruned_loss=0.2287, over 5711395.18 frames. ], libri_tot_loss[loss=0.5953, simple_loss=0.5651, pruned_loss=0.3281, over 5370744.69 frames. ], giga_tot_loss[loss=0.4595, simple_loss=0.4704, pruned_loss=0.2243, over 5699047.89 frames. ], batch size: 705, lr: 4.09e-02, grad_scale: 4.0
+2023-02-28 11:53:16,071 INFO [train.py:968] (1/2) Epoch 1, batch 5600, giga_loss[loss=0.4441, simple_loss=0.4691, pruned_loss=0.2096, over 28873.00 frames. ], tot_loss[loss=0.4622, simple_loss=0.4724, pruned_loss=0.226, over 5721341.66 frames. ], libri_tot_loss[loss=0.5929, simple_loss=0.5634, pruned_loss=0.3262, over 5384076.82 frames. ], giga_tot_loss[loss=0.4558, simple_loss=0.468, pruned_loss=0.2218, over 5707218.01 frames. ], batch size: 174, lr: 4.08e-02, grad_scale: 8.0
+2023-02-28 11:53:23,675 INFO [optim.py:369] (1/2) 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 11:53:56,155 INFO [train.py:968] (1/2) Epoch 1, batch 5650, giga_loss[loss=0.4179, simple_loss=0.4453, pruned_loss=0.1952, over 28891.00 frames. ], tot_loss[loss=0.4547, simple_loss=0.4667, pruned_loss=0.2214, over 5717581.97 frames. ], libri_tot_loss[loss=0.5895, simple_loss=0.5611, pruned_loss=0.3235, over 5389438.69 frames. ], giga_tot_loss[loss=0.4484, simple_loss=0.4623, pruned_loss=0.2173, over 5712407.58 frames. ], batch size: 174, lr: 4.07e-02, grad_scale: 8.0
+2023-02-28 11:54:36,579 INFO [train.py:968] (1/2) Epoch 1, batch 5700, libri_loss[loss=0.5923, simple_loss=0.5774, pruned_loss=0.3036, over 27791.00 frames. ], tot_loss[loss=0.4476, simple_loss=0.4603, pruned_loss=0.2174, over 5707684.39 frames. ], libri_tot_loss[loss=0.5891, simple_loss=0.561, pruned_loss=0.3228, over 5385751.22 frames. ], giga_tot_loss[loss=0.4405, simple_loss=0.4553, pruned_loss=0.2129, over 5712557.37 frames. ], batch size: 115, lr: 4.06e-02, grad_scale: 8.0
+2023-02-28 11:54:45,165 INFO [optim.py:369] (1/2) 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 11:55:16,894 INFO [train.py:968] (1/2) Epoch 1, batch 5750, giga_loss[loss=0.4857, simple_loss=0.4922, pruned_loss=0.2396, over 28603.00 frames. ], tot_loss[loss=0.4478, simple_loss=0.4606, pruned_loss=0.2175, over 5702864.63 frames. ], libri_tot_loss[loss=0.5876, simple_loss=0.56, pruned_loss=0.3214, over 5390826.38 frames. ], giga_tot_loss[loss=0.4397, simple_loss=0.4547, pruned_loss=0.2123, over 5711190.41 frames. ], batch size: 336, lr: 4.05e-02, grad_scale: 8.0
+2023-02-28 11:55:54,286 INFO [train.py:968] (1/2) Epoch 1, batch 5800, giga_loss[loss=0.4253, simple_loss=0.4574, pruned_loss=0.1966, over 29045.00 frames. ], tot_loss[loss=0.4499, simple_loss=0.4631, pruned_loss=0.2184, over 5714787.34 frames. ], libri_tot_loss[loss=0.5858, simple_loss=0.5588, pruned_loss=0.3197, over 5405108.93 frames. ], giga_tot_loss[loss=0.4412, simple_loss=0.4568, pruned_loss=0.2129, over 5717253.38 frames. ], batch size: 155, lr: 4.04e-02, grad_scale: 8.0
+2023-02-28 11:56:01,629 INFO [optim.py:369] (1/2) 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 11:56:25,405 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-02-28 11:56:36,516 INFO [train.py:968] (1/2) Epoch 1, batch 5850, giga_loss[loss=0.5207, simple_loss=0.5192, pruned_loss=0.2611, over 28831.00 frames. ], tot_loss[loss=0.4541, simple_loss=0.4677, pruned_loss=0.2203, over 5720402.59 frames. ], libri_tot_loss[loss=0.5844, simple_loss=0.558, pruned_loss=0.3184, over 5414675.98 frames. ], giga_tot_loss[loss=0.4461, simple_loss=0.4617, pruned_loss=0.2152, over 5719354.75 frames. ], batch size: 119, lr: 4.03e-02, grad_scale: 8.0
+2023-02-28 11:56:58,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1076, 2.3986, 2.2952, 1.7323], device='cuda:1'), covar=tensor([0.0631, 0.1128, 0.0527, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0319, 0.0487, 0.0345, 0.0324], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 11:57:08,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0923, 1.6934, 1.5244, 1.4498], device='cuda:1'), covar=tensor([0.0616, 0.1347, 0.1032, 0.1587], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0732, 0.0565, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0006], device='cuda:1')
+2023-02-28 11:57:18,956 INFO [train.py:968] (1/2) Epoch 1, batch 5900, giga_loss[loss=0.4486, simple_loss=0.4726, pruned_loss=0.2123, over 28902.00 frames. ], tot_loss[loss=0.4575, simple_loss=0.4713, pruned_loss=0.2218, over 5715000.95 frames. ], libri_tot_loss[loss=0.5826, simple_loss=0.5568, pruned_loss=0.317, over 5421748.57 frames. ], giga_tot_loss[loss=0.4508, simple_loss=0.4663, pruned_loss=0.2177, over 5711952.79 frames. ], batch size: 186, lr: 4.02e-02, grad_scale: 8.0
+2023-02-28 11:57:27,290 INFO [optim.py:369] (1/2) 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 11:57:53,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-02-28 11:57:57,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3879, 1.4407, 2.1042, 0.1819], device='cuda:1'), covar=tensor([0.0956, 0.0665, 0.0363, 0.1932], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0399, 0.0374, 0.0598], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1')
+2023-02-28 11:58:01,373 INFO [train.py:968] (1/2) Epoch 1, batch 5950, libri_loss[loss=0.5016, simple_loss=0.5135, pruned_loss=0.2449, over 26010.00 frames. ], tot_loss[loss=0.461, simple_loss=0.4749, pruned_loss=0.2236, over 5718434.66 frames. ], libri_tot_loss[loss=0.5805, simple_loss=0.5557, pruned_loss=0.3149, over 5437031.00 frames. ], giga_tot_loss[loss=0.4529, simple_loss=0.4688, pruned_loss=0.2185, over 5712868.53 frames. ], batch size: 136, lr: 4.01e-02, grad_scale: 8.0
+2023-02-28 11:58:10,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 11:58:43,527 INFO [train.py:968] (1/2) Epoch 1, batch 6000, giga_loss[loss=0.4347, simple_loss=0.4649, pruned_loss=0.2023, over 29056.00 frames. ], tot_loss[loss=0.4671, simple_loss=0.4793, pruned_loss=0.2274, over 5712674.89 frames. ], libri_tot_loss[loss=0.5789, simple_loss=0.5546, pruned_loss=0.3136, over 5441544.71 frames. ], giga_tot_loss[loss=0.4595, simple_loss=0.4736, pruned_loss=0.2226, over 5710246.65 frames. ], batch size: 128, lr: 4.00e-02, grad_scale: 8.0
+2023-02-28 11:58:43,527 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 11:58:51,740 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 18556MB
+2023-02-28 11:58:58,806 INFO [optim.py:369] (1/2) 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 11:59:30,349 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 6050, giga_loss[loss=0.5206, simple_loss=0.5065, pruned_loss=0.2673, over 28887.00 frames. ], tot_loss[loss=0.4821, simple_loss=0.4884, pruned_loss=0.2379, over 5708201.40 frames. ], libri_tot_loss[loss=0.5772, simple_loss=0.5535, pruned_loss=0.3121, over 5450621.00 frames. ], giga_tot_loss[loss=0.4756, simple_loss=0.4835, pruned_loss=0.2338, over 5702919.72 frames. ], batch size: 174, lr: 3.99e-02, grad_scale: 8.0
+2023-02-28 12:00:17,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-02-28 12:00:19,897 INFO [train.py:968] (1/2) Epoch 1, batch 6100, giga_loss[loss=0.5138, simple_loss=0.5169, pruned_loss=0.2554, over 28961.00 frames. ], tot_loss[loss=0.4939, simple_loss=0.4956, pruned_loss=0.2461, over 5706873.11 frames. ], libri_tot_loss[loss=0.5735, simple_loss=0.551, pruned_loss=0.3092, over 5471077.98 frames. ], giga_tot_loss[loss=0.4881, simple_loss=0.4913, pruned_loss=0.2425, over 5694330.99 frames. ], batch size: 174, lr: 3.98e-02, grad_scale: 4.0
+2023-02-28 12:00:31,038 INFO [optim.py:369] (1/2) 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 12:00:33,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-02-28 12:01:07,646 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6148.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:01:08,858 INFO [train.py:968] (1/2) Epoch 1, batch 6150, giga_loss[loss=0.4863, simple_loss=0.4964, pruned_loss=0.2381, over 28915.00 frames. ], tot_loss[loss=0.5055, simple_loss=0.5033, pruned_loss=0.2538, over 5687202.93 frames. ], libri_tot_loss[loss=0.5727, simple_loss=0.5505, pruned_loss=0.3083, over 5476026.94 frames. ], giga_tot_loss[loss=0.5004, simple_loss=0.4994, pruned_loss=0.2507, over 5677021.86 frames. ], batch size: 186, lr: 3.97e-02, grad_scale: 4.0
+2023-02-28 12:01:37,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-02-28 12:01:57,772 INFO [train.py:968] (1/2) Epoch 1, batch 6200, libri_loss[loss=0.524, simple_loss=0.5251, pruned_loss=0.2614, over 29374.00 frames. ], tot_loss[loss=0.5159, simple_loss=0.5101, pruned_loss=0.2609, over 5674322.60 frames. ], libri_tot_loss[loss=0.5712, simple_loss=0.5499, pruned_loss=0.3068, over 5476635.00 frames. ], giga_tot_loss[loss=0.5112, simple_loss=0.5061, pruned_loss=0.2581, over 5671833.64 frames. ], batch size: 92, lr: 3.96e-02, grad_scale: 4.0
+2023-02-28 12:02:06,399 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 6250, giga_loss[loss=0.5223, simple_loss=0.5201, pruned_loss=0.2623, over 28804.00 frames. ], tot_loss[loss=0.5247, simple_loss=0.5157, pruned_loss=0.2668, over 5683326.32 frames. ], libri_tot_loss[loss=0.5691, simple_loss=0.5487, pruned_loss=0.3049, over 5489126.69 frames. ], giga_tot_loss[loss=0.5215, simple_loss=0.5127, pruned_loss=0.2651, over 5675193.89 frames. ], batch size: 112, lr: 3.95e-02, grad_scale: 4.0
+2023-02-28 12:02:43,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6251.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:03:28,565 INFO [train.py:968] (1/2) Epoch 1, batch 6300, giga_loss[loss=0.5709, simple_loss=0.5475, pruned_loss=0.2971, over 28922.00 frames. ], tot_loss[loss=0.5312, simple_loss=0.5201, pruned_loss=0.2711, over 5665853.46 frames. ], libri_tot_loss[loss=0.5674, simple_loss=0.5477, pruned_loss=0.3035, over 5485670.19 frames. ], giga_tot_loss[loss=0.5292, simple_loss=0.5179, pruned_loss=0.2702, over 5667182.40 frames. ], batch size: 213, lr: 3.94e-02, grad_scale: 4.0
+2023-02-28 12:03:39,709 INFO [optim.py:369] (1/2) 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 12:04:21,662 INFO [train.py:968] (1/2) Epoch 1, batch 6350, giga_loss[loss=0.5614, simple_loss=0.5337, pruned_loss=0.2945, over 28565.00 frames. ], tot_loss[loss=0.5363, simple_loss=0.5227, pruned_loss=0.275, over 5647455.60 frames. ], libri_tot_loss[loss=0.5667, simple_loss=0.5474, pruned_loss=0.3029, over 5491925.18 frames. ], giga_tot_loss[loss=0.5349, simple_loss=0.5208, pruned_loss=0.2745, over 5645103.37 frames. ], batch size: 307, lr: 3.93e-02, grad_scale: 4.0
+2023-02-28 12:04:56,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5063, 1.5137, 1.3757, 1.4857], device='cuda:1'), covar=tensor([0.0427, 0.0590, 0.0692, 0.0545], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0768, 0.0571, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0006], device='cuda:1')
+2023-02-28 12:04:57,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6383.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:05:16,417 INFO [train.py:968] (1/2) Epoch 1, batch 6400, giga_loss[loss=0.5792, simple_loss=0.5571, pruned_loss=0.3007, over 28991.00 frames. ], tot_loss[loss=0.5427, simple_loss=0.5256, pruned_loss=0.2799, over 5632747.87 frames. ], libri_tot_loss[loss=0.5653, simple_loss=0.5464, pruned_loss=0.3018, over 5498174.28 frames. ], giga_tot_loss[loss=0.5423, simple_loss=0.5246, pruned_loss=0.28, over 5627693.72 frames. ], batch size: 164, lr: 3.92e-02, grad_scale: 8.0
+2023-02-28 12:05:27,575 INFO [optim.py:369] (1/2) 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 12:05:35,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9088, 2.5711, 2.3952, 1.7850], device='cuda:1'), covar=tensor([0.2230, 0.1168, 0.1060, 0.2967], device='cuda:1'), in_proj_covar=tensor([0.0595, 0.0573, 0.0544, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0009], device='cuda:1')
+2023-02-28 12:05:35,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6418.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:05:55,161 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6436.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:06:09,328 INFO [train.py:968] (1/2) Epoch 1, batch 6450, giga_loss[loss=0.5033, simple_loss=0.4998, pruned_loss=0.2533, over 28069.00 frames. ], tot_loss[loss=0.5505, simple_loss=0.5298, pruned_loss=0.2856, over 5619177.60 frames. ], libri_tot_loss[loss=0.5643, simple_loss=0.5459, pruned_loss=0.3009, over 5504419.73 frames. ], giga_tot_loss[loss=0.5508, simple_loss=0.5292, pruned_loss=0.2862, over 5611352.45 frames. ], batch size: 77, lr: 3.91e-02, grad_scale: 8.0
+2023-02-28 12:07:01,414 INFO [train.py:968] (1/2) Epoch 1, batch 6500, giga_loss[loss=0.6119, simple_loss=0.5511, pruned_loss=0.3363, over 27559.00 frames. ], tot_loss[loss=0.5544, simple_loss=0.5328, pruned_loss=0.288, over 5613954.29 frames. ], libri_tot_loss[loss=0.563, simple_loss=0.5451, pruned_loss=0.2998, over 5506115.80 frames. ], giga_tot_loss[loss=0.5554, simple_loss=0.5327, pruned_loss=0.2891, over 5608556.41 frames. ], batch size: 472, lr: 3.90e-02, grad_scale: 8.0
+2023-02-28 12:07:13,637 INFO [optim.py:369] (1/2) 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 12:07:27,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6523.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:07:32,755 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6528.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:07:53,986 INFO [train.py:968] (1/2) Epoch 1, batch 6550, giga_loss[loss=0.5462, simple_loss=0.5258, pruned_loss=0.2833, over 29105.00 frames. ], tot_loss[loss=0.5561, simple_loss=0.5335, pruned_loss=0.2894, over 5628310.59 frames. ], libri_tot_loss[loss=0.5622, simple_loss=0.5447, pruned_loss=0.299, over 5512223.20 frames. ], giga_tot_loss[loss=0.5576, simple_loss=0.5335, pruned_loss=0.2908, over 5620381.19 frames. ], batch size: 128, lr: 3.89e-02, grad_scale: 4.0
+2023-02-28 12:08:05,538 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6563.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:08:08,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6564.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:08:11,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8858, 1.5449, 1.2497, 1.2245], device='cuda:1'), covar=tensor([0.1174, 0.1334, 0.1772, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.0563, 0.0702, 0.0688, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 12:08:37,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6593.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:08:39,625 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6596.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:08:42,680 INFO [train.py:968] (1/2) Epoch 1, batch 6600, giga_loss[loss=0.5142, simple_loss=0.5125, pruned_loss=0.258, over 28969.00 frames. ], tot_loss[loss=0.5528, simple_loss=0.5309, pruned_loss=0.2873, over 5635982.82 frames. ], libri_tot_loss[loss=0.5609, simple_loss=0.544, pruned_loss=0.2979, over 5519041.44 frames. ], giga_tot_loss[loss=0.5549, simple_loss=0.5314, pruned_loss=0.2892, over 5625534.70 frames. ], batch size: 136, lr: 3.89e-02, grad_scale: 4.0
+2023-02-28 12:08:46,318 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-02-28 12:08:46,886 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6604.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:08:55,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4128, 1.1486, 1.6841, 0.6543], device='cuda:1'), covar=tensor([0.0594, 0.0417, 0.0331, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0627, 0.0453, 0.0502, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 12:08:56,971 INFO [optim.py:369] (1/2) 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 12:09:09,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6626.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:09:20,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9562, 2.4898, 4.6174, 2.7125], device='cuda:1'), covar=tensor([0.1669, 0.1009, 0.0215, 0.0641], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0421, 0.0507, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 12:09:31,590 INFO [train.py:968] (1/2) Epoch 1, batch 6650, giga_loss[loss=0.6421, simple_loss=0.573, pruned_loss=0.3556, over 26545.00 frames. ], tot_loss[loss=0.548, simple_loss=0.5287, pruned_loss=0.2836, over 5639300.01 frames. ], libri_tot_loss[loss=0.5578, simple_loss=0.5421, pruned_loss=0.2953, over 5533583.97 frames. ], giga_tot_loss[loss=0.5522, simple_loss=0.5302, pruned_loss=0.2871, over 5621484.52 frames. ], batch size: 555, lr: 3.88e-02, grad_scale: 4.0
+2023-02-28 12:09:41,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6659.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:09:46,634 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6666.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:09:51,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6669.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:10:20,587 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-02-28 12:10:20,939 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6698.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:10:22,845 INFO [train.py:968] (1/2) Epoch 1, batch 6700, giga_loss[loss=0.5409, simple_loss=0.5301, pruned_loss=0.2758, over 28667.00 frames. ], tot_loss[loss=0.5471, simple_loss=0.5291, pruned_loss=0.2825, over 5649899.23 frames. ], libri_tot_loss[loss=0.5567, simple_loss=0.5414, pruned_loss=0.2945, over 5537204.64 frames. ], giga_tot_loss[loss=0.5512, simple_loss=0.5308, pruned_loss=0.2858, over 5633709.83 frames. ], batch size: 307, lr: 3.87e-02, grad_scale: 4.0
+2023-02-28 12:10:31,256 INFO [optim.py:369] (1/2) 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 12:10:40,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-02-28 12:11:10,054 INFO [train.py:968] (1/2) Epoch 1, batch 6750, giga_loss[loss=0.5435, simple_loss=0.5284, pruned_loss=0.2793, over 28256.00 frames. ], tot_loss[loss=0.5469, simple_loss=0.5289, pruned_loss=0.2824, over 5622141.67 frames. ], libri_tot_loss[loss=0.5539, simple_loss=0.5394, pruned_loss=0.2924, over 5540576.98 frames. ], giga_tot_loss[loss=0.5523, simple_loss=0.5316, pruned_loss=0.2865, over 5610263.18 frames. ], batch size: 368, lr: 3.86e-02, grad_scale: 4.0
+2023-02-28 12:11:18,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6758.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:11:30,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6769.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:11:33,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:11:57,955 INFO [train.py:968] (1/2) Epoch 1, batch 6800, libri_loss[loss=0.5565, simple_loss=0.5367, pruned_loss=0.2881, over 19033.00 frames. ], tot_loss[loss=0.5414, simple_loss=0.5256, pruned_loss=0.2787, over 5616720.37 frames. ], libri_tot_loss[loss=0.5521, simple_loss=0.5382, pruned_loss=0.2909, over 5539139.53 frames. ], giga_tot_loss[loss=0.547, simple_loss=0.5284, pruned_loss=0.2828, over 5611665.00 frames. ], batch size: 187, lr: 3.85e-02, grad_scale: 8.0
+2023-02-28 12:12:00,040 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6801.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:12:12,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6811.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:12:12,458 INFO [optim.py:369] (1/2) 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 12:12:50,803 INFO [train.py:968] (1/2) Epoch 1, batch 6850, giga_loss[loss=0.4591, simple_loss=0.4826, pruned_loss=0.2178, over 28967.00 frames. ], tot_loss[loss=0.5337, simple_loss=0.5213, pruned_loss=0.273, over 5615745.25 frames. ], libri_tot_loss[loss=0.5501, simple_loss=0.5368, pruned_loss=0.2894, over 5544565.22 frames. ], giga_tot_loss[loss=0.5397, simple_loss=0.5245, pruned_loss=0.2775, over 5608385.70 frames. ], batch size: 164, lr: 3.84e-02, grad_scale: 8.0
+2023-02-28 12:12:56,040 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6855.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:12:57,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2258, 1.2318, 0.8131, 1.0087], device='cuda:1'), covar=tensor([0.0989, 0.0915, 0.1799, 0.1424], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0676, 0.0648, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 12:12:59,833 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=6860.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:13:24,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1547, 1.5188, 0.9207, 1.3653], device='cuda:1'), covar=tensor([0.1316, 0.1805, 0.1466, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0630, 0.0782, 0.0737, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 12:13:38,030 INFO [train.py:968] (1/2) Epoch 1, batch 6900, giga_loss[loss=0.5074, simple_loss=0.5045, pruned_loss=0.2552, over 28764.00 frames. ], tot_loss[loss=0.5246, simple_loss=0.5161, pruned_loss=0.2665, over 5638380.51 frames. ], libri_tot_loss[loss=0.5478, simple_loss=0.5351, pruned_loss=0.2878, over 5551532.20 frames. ], giga_tot_loss[loss=0.531, simple_loss=0.5199, pruned_loss=0.2711, over 5627704.28 frames. ], batch size: 119, lr: 3.83e-02, grad_scale: 8.0
+2023-02-28 12:13:39,948 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6901.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:13:41,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6903.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:13:44,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:13:58,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4078, 1.6788, 1.5531, 1.2359], device='cuda:1'), covar=tensor([0.0978, 0.1184, 0.1389, 0.2184], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0683, 0.0656, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 12:14:10,963 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6933.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:15,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6938.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:28,902 INFO [train.py:968] (1/2) Epoch 1, batch 6950, giga_loss[loss=0.5148, simple_loss=0.5188, pruned_loss=0.2554, over 28975.00 frames. ], tot_loss[loss=0.5235, simple_loss=0.5151, pruned_loss=0.2659, over 5644997.95 frames. ], libri_tot_loss[loss=0.5455, simple_loss=0.5335, pruned_loss=0.2861, over 5562275.12 frames. ], giga_tot_loss[loss=0.53, simple_loss=0.519, pruned_loss=0.2705, over 5629274.64 frames. ], batch size: 213, lr: 3.82e-02, grad_scale: 4.0
+2023-02-28 12:14:34,094 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6954.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:36,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6957.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:40,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4248, 2.0413, 1.8912, 1.4668], device='cuda:1'), covar=tensor([0.2358, 0.1227, 0.1236, 0.3423], device='cuda:1'), in_proj_covar=tensor([0.0598, 0.0543, 0.0536, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0010], device='cuda:1')
+2023-02-28 12:14:50,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6971.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:14:56,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6979.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:15:01,468 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:15:17,893 INFO [train.py:968] (1/2) Epoch 1, batch 7000, giga_loss[loss=0.4809, simple_loss=0.4898, pruned_loss=0.236, over 28905.00 frames. ], tot_loss[loss=0.52, simple_loss=0.5128, pruned_loss=0.2635, over 5647867.75 frames. ], libri_tot_loss[loss=0.5447, simple_loss=0.533, pruned_loss=0.2853, over 5561293.43 frames. ], giga_tot_loss[loss=0.5254, simple_loss=0.5159, pruned_loss=0.2674, over 5637906.60 frames. ], batch size: 119, lr: 3.81e-02, grad_scale: 4.0
+2023-02-28 12:15:28,516 INFO [optim.py:369] (1/2) 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 12:15:39,180 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7022.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:15:51,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7034.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:02,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7046.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:04,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7049.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:04,952 INFO [train.py:968] (1/2) Epoch 1, batch 7050, giga_loss[loss=0.5643, simple_loss=0.5451, pruned_loss=0.2918, over 28740.00 frames. ], tot_loss[loss=0.5168, simple_loss=0.5111, pruned_loss=0.2613, over 5658073.57 frames. ], libri_tot_loss[loss=0.5427, simple_loss=0.5317, pruned_loss=0.2839, over 5567791.63 frames. ], giga_tot_loss[loss=0.5223, simple_loss=0.5143, pruned_loss=0.2652, over 5646114.56 frames. ], batch size: 262, lr: 3.80e-02, grad_scale: 4.0
+2023-02-28 12:16:41,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7078.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:44,345 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7081.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:16:46,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7084.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:02,599 INFO [train.py:968] (1/2) Epoch 1, batch 7100, giga_loss[loss=0.4764, simple_loss=0.4921, pruned_loss=0.2303, over 28860.00 frames. ], tot_loss[loss=0.5138, simple_loss=0.5098, pruned_loss=0.2589, over 5661832.55 frames. ], libri_tot_loss[loss=0.5423, simple_loss=0.5314, pruned_loss=0.2835, over 5568438.23 frames. ], giga_tot_loss[loss=0.5182, simple_loss=0.5125, pruned_loss=0.262, over 5652590.67 frames. ], batch size: 186, lr: 3.79e-02, grad_scale: 4.0
+2023-02-28 12:17:19,098 INFO [optim.py:369] (1/2) 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 12:17:19,306 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7113.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:21,306 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7114.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:23,466 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7117.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:23,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.12 vs. limit=2.0
+2023-02-28 12:17:27,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7122.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:31,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7125.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:17:51,312 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 7150, giga_loss[loss=0.6383, simple_loss=0.5596, pruned_loss=0.3585, over 26689.00 frames. ], tot_loss[loss=0.5041, simple_loss=0.5043, pruned_loss=0.2519, over 5673683.89 frames. ], libri_tot_loss[loss=0.5392, simple_loss=0.5293, pruned_loss=0.2812, over 5581962.52 frames. ], giga_tot_loss[loss=0.5092, simple_loss=0.5073, pruned_loss=0.2555, over 5657354.98 frames. ], batch size: 555, lr: 3.79e-02, grad_scale: 4.0
+2023-02-28 12:17:58,516 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7154.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:18:07,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1818, 0.8464, 1.0483, 0.3797], device='cuda:1'), covar=tensor([0.0475, 0.0408, 0.0421, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0614, 0.0449, 0.0508, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 12:18:15,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3257, 1.3955, 1.1232, 1.5595], device='cuda:1'), covar=tensor([0.0954, 0.1196, 0.0967, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0639, 0.0750, 0.0713, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0005, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 12:18:25,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7177.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:18:28,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7180.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:18:50,225 INFO [train.py:968] (1/2) Epoch 1, batch 7200, giga_loss[loss=0.44, simple_loss=0.4796, pruned_loss=0.2002, over 29024.00 frames. ], tot_loss[loss=0.5034, simple_loss=0.5057, pruned_loss=0.2506, over 5667755.04 frames. ], libri_tot_loss[loss=0.538, simple_loss=0.5285, pruned_loss=0.2803, over 5578071.72 frames. ], giga_tot_loss[loss=0.5079, simple_loss=0.5084, pruned_loss=0.2537, over 5659740.85 frames. ], batch size: 136, lr: 3.78e-02, grad_scale: 8.0
+2023-02-28 12:19:03,405 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7209.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:19:05,305 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4056, 1.7376, 2.9381, 1.3669], device='cuda:1'), covar=tensor([0.1487, 0.0970, 0.0395, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0421, 0.0505, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0006, 0.0004], device='cuda:1')
+2023-02-28 12:19:05,719 INFO [optim.py:369] (1/2) 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 12:19:11,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2353, 1.5803, 1.5789, 1.1968], device='cuda:1'), covar=tensor([0.1394, 0.1345, 0.1140, 0.2188], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0593, 0.0491, 0.0618], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:19:19,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7230.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:19:24,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7235.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:19:38,391 INFO [train.py:968] (1/2) Epoch 1, batch 7250, giga_loss[loss=0.5283, simple_loss=0.5256, pruned_loss=0.2655, over 28966.00 frames. ], tot_loss[loss=0.5068, simple_loss=0.5076, pruned_loss=0.253, over 5664903.96 frames. ], libri_tot_loss[loss=0.535, simple_loss=0.5261, pruned_loss=0.2782, over 5588555.75 frames. ], giga_tot_loss[loss=0.5121, simple_loss=0.5113, pruned_loss=0.2565, over 5651507.77 frames. ], batch size: 164, lr: 3.77e-02, grad_scale: 4.0
+2023-02-28 12:20:03,218 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7271.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:20:20,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4716, 1.8287, 1.3021, 0.7492], device='cuda:1'), covar=tensor([0.0750, 0.0454, 0.0618, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0640, 0.0469, 0.0541, 0.0604], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 12:20:30,932 INFO [train.py:968] (1/2) Epoch 1, batch 7300, giga_loss[loss=0.4603, simple_loss=0.4807, pruned_loss=0.22, over 29022.00 frames. ], tot_loss[loss=0.5118, simple_loss=0.5105, pruned_loss=0.2566, over 5677682.12 frames. ], libri_tot_loss[loss=0.535, simple_loss=0.5261, pruned_loss=0.2781, over 5593336.11 frames. ], giga_tot_loss[loss=0.5157, simple_loss=0.5132, pruned_loss=0.2591, over 5663898.99 frames. ], batch size: 128, lr: 3.76e-02, grad_scale: 4.0
+2023-02-28 12:20:32,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7302.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:20:42,975 INFO [optim.py:369] (1/2) 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 12:20:58,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4641, 1.4738, 0.9055, 1.2609], device='cuda:1'), covar=tensor([0.0937, 0.0901, 0.1638, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0699, 0.0678, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 12:21:11,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0216, 1.2292, 1.1926, 1.1008], device='cuda:1'), covar=tensor([0.1516, 0.1150, 0.1297, 0.2465], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0598, 0.0498, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:21:15,320 INFO [train.py:968] (1/2) Epoch 1, batch 7350, giga_loss[loss=0.4307, simple_loss=0.4552, pruned_loss=0.2031, over 28742.00 frames. ], tot_loss[loss=0.509, simple_loss=0.5086, pruned_loss=0.2547, over 5670333.69 frames. ], libri_tot_loss[loss=0.5314, simple_loss=0.5241, pruned_loss=0.2752, over 5593804.73 frames. ], giga_tot_loss[loss=0.5142, simple_loss=0.5119, pruned_loss=0.2583, over 5662653.04 frames. ], batch size: 119, lr: 3.75e-02, grad_scale: 4.0
+2023-02-28 12:21:38,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7373.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:21:41,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7376.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:21:45,903 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7378.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:21:47,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7381.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:00,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:03,891 INFO [train.py:968] (1/2) Epoch 1, batch 7400, giga_loss[loss=0.4632, simple_loss=0.4692, pruned_loss=0.2286, over 28760.00 frames. ], tot_loss[loss=0.5076, simple_loss=0.5064, pruned_loss=0.2544, over 5645456.79 frames. ], libri_tot_loss[loss=0.5304, simple_loss=0.5236, pruned_loss=0.2743, over 5577013.85 frames. ], giga_tot_loss[loss=0.5122, simple_loss=0.5092, pruned_loss=0.2576, over 5656577.87 frames. ], batch size: 99, lr: 3.74e-02, grad_scale: 4.0
+2023-02-28 12:22:09,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7405.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:14,594 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7410.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:22:17,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7190, 2.1263, 1.9913, 1.4126], device='cuda:1'), covar=tensor([0.1807, 0.0966, 0.0982, 0.2624], device='cuda:1'), in_proj_covar=tensor([0.0580, 0.0544, 0.0524, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0010], device='cuda:1')
+2023-02-28 12:22:18,162 INFO [optim.py:369] (1/2) 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 12:22:42,087 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7441.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:22:48,724 INFO [train.py:968] (1/2) Epoch 1, batch 7450, giga_loss[loss=0.5537, simple_loss=0.5412, pruned_loss=0.2832, over 28971.00 frames. ], tot_loss[loss=0.5064, simple_loss=0.5052, pruned_loss=0.2538, over 5659496.37 frames. ], libri_tot_loss[loss=0.5297, simple_loss=0.5232, pruned_loss=0.2736, over 5584128.91 frames. ], giga_tot_loss[loss=0.5101, simple_loss=0.5072, pruned_loss=0.2565, over 5664237.13 frames. ], batch size: 164, lr: 3.73e-02, grad_scale: 4.0
+2023-02-28 12:23:40,182 INFO [train.py:968] (1/2) Epoch 1, batch 7500, giga_loss[loss=0.5034, simple_loss=0.5038, pruned_loss=0.2515, over 28559.00 frames. ], tot_loss[loss=0.5038, simple_loss=0.5042, pruned_loss=0.2517, over 5667046.65 frames. ], libri_tot_loss[loss=0.528, simple_loss=0.5222, pruned_loss=0.2723, over 5584175.79 frames. ], giga_tot_loss[loss=0.5076, simple_loss=0.5062, pruned_loss=0.2545, over 5673421.85 frames. ], batch size: 336, lr: 3.72e-02, grad_scale: 4.0
+2023-02-28 12:23:51,775 INFO [optim.py:369] (1/2) 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 12:24:09,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2003, 1.2560, 0.9996, 1.2180], device='cuda:1'), covar=tensor([0.1030, 0.1641, 0.1132, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0651, 0.0789, 0.0743, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0006, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 12:24:17,644 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7540.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:24:19,584 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7543.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:24:25,989 INFO [train.py:968] (1/2) Epoch 1, batch 7550, giga_loss[loss=0.4435, simple_loss=0.4766, pruned_loss=0.2053, over 28982.00 frames. ], tot_loss[loss=0.4988, simple_loss=0.5021, pruned_loss=0.2478, over 5670923.12 frames. ], libri_tot_loss[loss=0.5267, simple_loss=0.5213, pruned_loss=0.2712, over 5579139.70 frames. ], giga_tot_loss[loss=0.5024, simple_loss=0.504, pruned_loss=0.2504, over 5683766.59 frames. ], batch size: 164, lr: 3.72e-02, grad_scale: 4.0
+2023-02-28 12:24:32,895 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7556.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:24:45,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7572.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:25:15,579 INFO [train.py:968] (1/2) Epoch 1, batch 7600, giga_loss[loss=0.5093, simple_loss=0.4784, pruned_loss=0.2701, over 23424.00 frames. ], tot_loss[loss=0.4995, simple_loss=0.5025, pruned_loss=0.2482, over 5673228.61 frames. ], libri_tot_loss[loss=0.5261, simple_loss=0.5209, pruned_loss=0.2707, over 5581898.93 frames. ], giga_tot_loss[loss=0.5026, simple_loss=0.5041, pruned_loss=0.2505, over 5681700.29 frames. ], batch size: 705, lr: 3.71e-02, grad_scale: 8.0
+2023-02-28 12:25:26,986 INFO [optim.py:369] (1/2) 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 12:25:39,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-02-28 12:25:55,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7646.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:26:00,604 INFO [train.py:968] (1/2) Epoch 1, batch 7650, giga_loss[loss=0.5077, simple_loss=0.5042, pruned_loss=0.2557, over 27947.00 frames. ], tot_loss[loss=0.4999, simple_loss=0.5021, pruned_loss=0.2488, over 5679404.62 frames. ], libri_tot_loss[loss=0.524, simple_loss=0.5196, pruned_loss=0.2691, over 5590009.37 frames. ], giga_tot_loss[loss=0.5034, simple_loss=0.5041, pruned_loss=0.2513, over 5681783.08 frames. ], batch size: 412, lr: 3.70e-02, grad_scale: 4.0
+2023-02-28 12:26:19,696 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7668.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:26:27,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:26:49,261 INFO [train.py:968] (1/2) Epoch 1, batch 7700, giga_loss[loss=0.506, simple_loss=0.501, pruned_loss=0.2555, over 28773.00 frames. ], tot_loss[loss=0.4953, simple_loss=0.4986, pruned_loss=0.2459, over 5686551.98 frames. ], libri_tot_loss[loss=0.523, simple_loss=0.519, pruned_loss=0.2683, over 5594715.45 frames. ], giga_tot_loss[loss=0.4985, simple_loss=0.5004, pruned_loss=0.2483, over 5685513.93 frames. ], batch size: 92, lr: 3.69e-02, grad_scale: 4.0
+2023-02-28 12:27:06,555 INFO [optim.py:369] (1/2) 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 12:27:36,626 INFO [train.py:968] (1/2) Epoch 1, batch 7750, libri_loss[loss=0.4914, simple_loss=0.505, pruned_loss=0.239, over 29551.00 frames. ], tot_loss[loss=0.4986, simple_loss=0.4999, pruned_loss=0.2487, over 5684314.27 frames. ], libri_tot_loss[loss=0.5216, simple_loss=0.518, pruned_loss=0.2672, over 5602764.00 frames. ], giga_tot_loss[loss=0.5016, simple_loss=0.5015, pruned_loss=0.2509, over 5679108.22 frames. ], batch size: 83, lr: 3.68e-02, grad_scale: 4.0
+2023-02-28 12:28:15,178 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7789.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:28:18,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7792.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:28:26,784 INFO [train.py:968] (1/2) Epoch 1, batch 7800, giga_loss[loss=0.4788, simple_loss=0.4875, pruned_loss=0.235, over 28979.00 frames. ], tot_loss[loss=0.4993, simple_loss=0.4998, pruned_loss=0.2494, over 5693957.87 frames. ], libri_tot_loss[loss=0.5212, simple_loss=0.5178, pruned_loss=0.2669, over 5607543.51 frames. ], giga_tot_loss[loss=0.5017, simple_loss=0.501, pruned_loss=0.2512, over 5686728.01 frames. ], batch size: 213, lr: 3.67e-02, grad_scale: 4.0
+2023-02-28 12:28:27,778 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=7800.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:28:42,141 INFO [optim.py:369] (1/2) 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 12:28:44,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7816.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:28:49,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:28:49,285 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-02-28 12:28:49,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7821.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:28:53,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7823.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:29:17,598 INFO [train.py:968] (1/2) Epoch 1, batch 7850, giga_loss[loss=0.4478, simple_loss=0.4609, pruned_loss=0.2173, over 28832.00 frames. ], tot_loss[loss=0.4953, simple_loss=0.4964, pruned_loss=0.247, over 5696973.18 frames. ], libri_tot_loss[loss=0.5202, simple_loss=0.5172, pruned_loss=0.2661, over 5612040.72 frames. ], giga_tot_loss[loss=0.4976, simple_loss=0.4976, pruned_loss=0.2488, over 5688306.59 frames. ], batch size: 119, lr: 3.66e-02, grad_scale: 4.0
+2023-02-28 12:29:19,799 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7852.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:29:59,774 INFO [train.py:968] (1/2) Epoch 1, batch 7900, giga_loss[loss=0.4481, simple_loss=0.4566, pruned_loss=0.2198, over 28581.00 frames. ], tot_loss[loss=0.4923, simple_loss=0.4944, pruned_loss=0.2451, over 5698131.77 frames. ], libri_tot_loss[loss=0.5187, simple_loss=0.5164, pruned_loss=0.2649, over 5610860.99 frames. ], giga_tot_loss[loss=0.4949, simple_loss=0.4956, pruned_loss=0.2472, over 5694359.68 frames. ], batch size: 71, lr: 3.66e-02, grad_scale: 4.0
+2023-02-28 12:30:15,437 INFO [optim.py:369] (1/2) 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 12:30:29,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:30:49,796 INFO [train.py:968] (1/2) Epoch 1, batch 7950, giga_loss[loss=0.5102, simple_loss=0.5101, pruned_loss=0.2551, over 28654.00 frames. ], tot_loss[loss=0.4926, simple_loss=0.4946, pruned_loss=0.2453, over 5687408.35 frames. ], libri_tot_loss[loss=0.5171, simple_loss=0.5153, pruned_loss=0.2637, over 5613439.39 frames. ], giga_tot_loss[loss=0.4956, simple_loss=0.4961, pruned_loss=0.2476, over 5683433.81 frames. ], batch size: 262, lr: 3.65e-02, grad_scale: 4.0
+2023-02-28 12:30:59,686 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7959.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:31:02,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7962.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 12:31:30,261 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7991.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 12:31:39,144 INFO [train.py:968] (1/2) Epoch 1, batch 8000, giga_loss[loss=0.4889, simple_loss=0.5097, pruned_loss=0.234, over 28875.00 frames. ], tot_loss[loss=0.4932, simple_loss=0.4957, pruned_loss=0.2453, over 5680831.67 frames. ], libri_tot_loss[loss=0.5159, simple_loss=0.5145, pruned_loss=0.2629, over 5611615.43 frames. ], giga_tot_loss[loss=0.4961, simple_loss=0.4972, pruned_loss=0.2475, over 5681663.03 frames. ], batch size: 112, lr: 3.64e-02, grad_scale: 8.0
+2023-02-28 12:31:54,698 INFO [optim.py:369] (1/2) 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 12:31:56,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8017.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:32:18,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8043.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:32:22,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7971, 2.0440, 2.6871, 0.4626], device='cuda:1'), covar=tensor([0.0912, 0.0823, 0.0437, 0.2449], device='cuda:1'), in_proj_covar=tensor([0.0655, 0.0527, 0.0426, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 12:32:23,987 INFO [train.py:968] (1/2) Epoch 1, batch 8050, giga_loss[loss=0.5138, simple_loss=0.5064, pruned_loss=0.2606, over 28003.00 frames. ], tot_loss[loss=0.4894, simple_loss=0.494, pruned_loss=0.2424, over 5679884.74 frames. ], libri_tot_loss[loss=0.5145, simple_loss=0.5135, pruned_loss=0.2619, over 5619564.69 frames. ], giga_tot_loss[loss=0.4924, simple_loss=0.4956, pruned_loss=0.2446, over 5674971.29 frames. ], batch size: 412, lr: 3.63e-02, grad_scale: 8.0
+2023-02-28 12:32:49,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8074.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:32:52,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8077.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:33:15,914 INFO [train.py:968] (1/2) Epoch 1, batch 8100, giga_loss[loss=0.4446, simple_loss=0.4578, pruned_loss=0.2157, over 28494.00 frames. ], tot_loss[loss=0.4953, simple_loss=0.4975, pruned_loss=0.2465, over 5675663.35 frames. ], libri_tot_loss[loss=0.5139, simple_loss=0.513, pruned_loss=0.2613, over 5624621.04 frames. ], giga_tot_loss[loss=0.4978, simple_loss=0.4988, pruned_loss=0.2484, over 5668191.34 frames. ], batch size: 78, lr: 3.62e-02, grad_scale: 4.0
+2023-02-28 12:33:21,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1619, 2.6726, 2.3827, 2.0889], device='cuda:1'), covar=tensor([0.0601, 0.0756, 0.0393, 0.0325], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0691, 0.0513, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0005, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 12:33:22,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8106.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:33:32,100 INFO [optim.py:369] (1/2) 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 12:33:49,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8136.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:04,845 INFO [train.py:968] (1/2) Epoch 1, batch 8150, giga_loss[loss=0.5916, simple_loss=0.5372, pruned_loss=0.323, over 23779.00 frames. ], tot_loss[loss=0.4988, simple_loss=0.4999, pruned_loss=0.2488, over 5685583.31 frames. ], libri_tot_loss[loss=0.5131, simple_loss=0.5127, pruned_loss=0.2606, over 5630992.00 frames. ], giga_tot_loss[loss=0.5011, simple_loss=0.5009, pruned_loss=0.2507, over 5674979.21 frames. ], batch size: 705, lr: 3.62e-02, grad_scale: 4.0
+2023-02-28 12:34:32,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8175.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:44,322 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8186.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:48,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8189.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:34:55,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8195.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:35:01,106 INFO [train.py:968] (1/2) Epoch 1, batch 8200, giga_loss[loss=0.4488, simple_loss=0.4694, pruned_loss=0.2141, over 28615.00 frames. ], tot_loss[loss=0.5032, simple_loss=0.502, pruned_loss=0.2522, over 5680227.23 frames. ], libri_tot_loss[loss=0.5132, simple_loss=0.5129, pruned_loss=0.2607, over 5632022.04 frames. ], giga_tot_loss[loss=0.5049, simple_loss=0.5025, pruned_loss=0.2536, over 5671186.25 frames. ], batch size: 78, lr: 3.61e-02, grad_scale: 4.0
+2023-02-28 12:35:17,442 INFO [optim.py:369] (1/2) 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 12:35:22,915 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8218.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:35:53,060 INFO [train.py:968] (1/2) Epoch 1, batch 8250, giga_loss[loss=0.5046, simple_loss=0.5063, pruned_loss=0.2515, over 28646.00 frames. ], tot_loss[loss=0.5077, simple_loss=0.5043, pruned_loss=0.2555, over 5672279.94 frames. ], libri_tot_loss[loss=0.5134, simple_loss=0.5132, pruned_loss=0.2606, over 5623870.75 frames. ], giga_tot_loss[loss=0.5087, simple_loss=0.5043, pruned_loss=0.2566, over 5673060.74 frames. ], batch size: 242, lr: 3.60e-02, grad_scale: 4.0
+2023-02-28 12:36:43,136 INFO [train.py:968] (1/2) Epoch 1, batch 8300, giga_loss[loss=0.4785, simple_loss=0.4956, pruned_loss=0.2307, over 29035.00 frames. ], tot_loss[loss=0.5109, simple_loss=0.5058, pruned_loss=0.2581, over 5659852.50 frames. ], libri_tot_loss[loss=0.5134, simple_loss=0.5133, pruned_loss=0.2604, over 5620363.76 frames. ], giga_tot_loss[loss=0.5118, simple_loss=0.5055, pruned_loss=0.259, over 5664708.04 frames. ], batch size: 155, lr: 3.59e-02, grad_scale: 4.0
+2023-02-28 12:36:59,028 INFO [optim.py:369] (1/2) 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 12:37:01,804 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8318.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:37:06,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8321.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:37:11,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7967, 1.9964, 3.7826, 2.1972], device='cuda:1'), covar=tensor([0.1711, 0.1174, 0.0284, 0.0664], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0452, 0.0528, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0006, 0.0004], device='cuda:1')
+2023-02-28 12:37:31,433 INFO [train.py:968] (1/2) Epoch 1, batch 8350, giga_loss[loss=0.5168, simple_loss=0.5155, pruned_loss=0.259, over 28884.00 frames. ], tot_loss[loss=0.5093, simple_loss=0.5044, pruned_loss=0.2571, over 5654695.30 frames. ], libri_tot_loss[loss=0.5123, simple_loss=0.5127, pruned_loss=0.2595, over 5614504.81 frames. ], giga_tot_loss[loss=0.5109, simple_loss=0.5046, pruned_loss=0.2586, over 5664931.68 frames. ], batch size: 199, lr: 3.58e-02, grad_scale: 4.0
+2023-02-28 12:37:33,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8350.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:37:52,969 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 12:38:09,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8392.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:38:16,663 INFO [train.py:968] (1/2) Epoch 1, batch 8400, giga_loss[loss=0.5983, simple_loss=0.5443, pruned_loss=0.3261, over 26546.00 frames. ], tot_loss[loss=0.5073, simple_loss=0.5034, pruned_loss=0.2556, over 5656773.71 frames. ], libri_tot_loss[loss=0.5114, simple_loss=0.5124, pruned_loss=0.2587, over 5619618.68 frames. ], giga_tot_loss[loss=0.5093, simple_loss=0.5037, pruned_loss=0.2575, over 5661235.11 frames. ], batch size: 555, lr: 3.58e-02, grad_scale: 8.0
+2023-02-28 12:38:30,230 INFO [optim.py:369] (1/2) 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 12:38:31,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4744, 1.6098, 1.6546, 1.2522], device='cuda:1'), covar=tensor([0.2090, 0.1157, 0.1124, 0.2777], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0530, 0.0516, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0008, 0.0011], device='cuda:1')
+2023-02-28 12:38:51,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5153, 1.4296, 1.0437, 1.1253], device='cuda:1'), covar=tensor([0.0823, 0.0780, 0.1206, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0725, 0.0683, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:38:59,231 INFO [train.py:968] (1/2) Epoch 1, batch 8450, giga_loss[loss=0.4625, simple_loss=0.4453, pruned_loss=0.2398, over 26551.00 frames. ], tot_loss[loss=0.4989, simple_loss=0.4985, pruned_loss=0.2496, over 5662022.99 frames. ], libri_tot_loss[loss=0.511, simple_loss=0.512, pruned_loss=0.2584, over 5623038.63 frames. ], giga_tot_loss[loss=0.5008, simple_loss=0.4988, pruned_loss=0.2514, over 5663248.57 frames. ], batch size: 555, lr: 3.57e-02, grad_scale: 8.0
+2023-02-28 12:39:45,078 INFO [train.py:968] (1/2) Epoch 1, batch 8500, giga_loss[loss=0.5851, simple_loss=0.5429, pruned_loss=0.3137, over 28743.00 frames. ], tot_loss[loss=0.4923, simple_loss=0.4938, pruned_loss=0.2454, over 5669683.65 frames. ], libri_tot_loss[loss=0.5107, simple_loss=0.5119, pruned_loss=0.2581, over 5627607.75 frames. ], giga_tot_loss[loss=0.4937, simple_loss=0.4939, pruned_loss=0.2468, over 5666949.74 frames. ], batch size: 284, lr: 3.56e-02, grad_scale: 8.0
+2023-02-28 12:39:55,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8511.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:02,488 INFO [optim.py:369] (1/2) 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 12:40:21,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8535.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:23,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8538.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:34,925 INFO [train.py:968] (1/2) Epoch 1, batch 8550, giga_loss[loss=0.4699, simple_loss=0.4481, pruned_loss=0.2459, over 23807.00 frames. ], tot_loss[loss=0.4909, simple_loss=0.4916, pruned_loss=0.2451, over 5665275.21 frames. ], libri_tot_loss[loss=0.5111, simple_loss=0.5122, pruned_loss=0.2584, over 5621989.33 frames. ], giga_tot_loss[loss=0.4914, simple_loss=0.4913, pruned_loss=0.2458, over 5667747.32 frames. ], batch size: 705, lr: 3.55e-02, grad_scale: 4.0
+2023-02-28 12:40:52,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8567.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:40:55,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8570.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:41:28,333 INFO [train.py:968] (1/2) Epoch 1, batch 8600, giga_loss[loss=0.4674, simple_loss=0.4747, pruned_loss=0.2301, over 28684.00 frames. ], tot_loss[loss=0.4925, simple_loss=0.4921, pruned_loss=0.2464, over 5659618.67 frames. ], libri_tot_loss[loss=0.5106, simple_loss=0.5119, pruned_loss=0.2579, over 5626144.52 frames. ], giga_tot_loss[loss=0.4931, simple_loss=0.4918, pruned_loss=0.2472, over 5658294.85 frames. ], batch size: 307, lr: 3.54e-02, grad_scale: 4.0
+2023-02-28 12:41:42,402 INFO [optim.py:369] (1/2) 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 12:42:15,789 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-02-28 12:42:20,647 INFO [train.py:968] (1/2) Epoch 1, batch 8650, giga_loss[loss=0.5398, simple_loss=0.5228, pruned_loss=0.2783, over 27656.00 frames. ], tot_loss[loss=0.4951, simple_loss=0.4945, pruned_loss=0.2479, over 5652645.78 frames. ], libri_tot_loss[loss=0.5095, simple_loss=0.5112, pruned_loss=0.257, over 5630073.26 frames. ], giga_tot_loss[loss=0.496, simple_loss=0.4942, pruned_loss=0.2489, over 5649278.45 frames. ], batch size: 474, lr: 3.54e-02, grad_scale: 4.0
+2023-02-28 12:42:24,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8654.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:42:24,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7639, 1.4565, 1.3236, 1.3527], device='cuda:1'), covar=tensor([0.0794, 0.1557, 0.1255, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0844, 0.0650, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0008, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 12:42:27,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8657.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:42:27,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=8658.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:42:57,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8686.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:08,288 INFO [train.py:968] (1/2) Epoch 1, batch 8700, giga_loss[loss=0.4334, simple_loss=0.4807, pruned_loss=0.193, over 28886.00 frames. ], tot_loss[loss=0.4965, simple_loss=0.4981, pruned_loss=0.2474, over 5657101.24 frames. ], libri_tot_loss[loss=0.5072, simple_loss=0.5095, pruned_loss=0.2555, over 5627380.68 frames. ], giga_tot_loss[loss=0.499, simple_loss=0.4991, pruned_loss=0.2494, over 5656879.56 frames. ], batch size: 174, lr: 3.53e-02, grad_scale: 4.0
+2023-02-28 12:43:21,946 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8713.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:23,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8716.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:24,105 INFO [optim.py:369] (1/2) 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 12:43:54,458 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8745.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:43:58,913 INFO [train.py:968] (1/2) Epoch 1, batch 8750, giga_loss[loss=0.5278, simple_loss=0.5224, pruned_loss=0.2666, over 28298.00 frames. ], tot_loss[loss=0.4949, simple_loss=0.4997, pruned_loss=0.245, over 5668475.83 frames. ], libri_tot_loss[loss=0.5063, simple_loss=0.509, pruned_loss=0.2548, over 5632830.14 frames. ], giga_tot_loss[loss=0.4974, simple_loss=0.5006, pruned_loss=0.2471, over 5664164.34 frames. ], batch size: 368, lr: 3.52e-02, grad_scale: 4.0
+2023-02-28 12:44:43,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2757, 1.0986, 1.1223, 0.6312], device='cuda:1'), covar=tensor([0.0545, 0.0443, 0.0493, 0.0734], device='cuda:1'), in_proj_covar=tensor([0.0622, 0.0492, 0.0550, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 12:44:45,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0400, 1.2948, 1.3867, 1.0517], device='cuda:1'), covar=tensor([0.1482, 0.1342, 0.1142, 0.2249], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0577, 0.0471, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:44:45,375 INFO [train.py:968] (1/2) Epoch 1, batch 8800, giga_loss[loss=0.5869, simple_loss=0.5572, pruned_loss=0.3083, over 27844.00 frames. ], tot_loss[loss=0.4983, simple_loss=0.5019, pruned_loss=0.2474, over 5669321.65 frames. ], libri_tot_loss[loss=0.5039, simple_loss=0.5074, pruned_loss=0.2531, over 5640858.51 frames. ], giga_tot_loss[loss=0.5022, simple_loss=0.5039, pruned_loss=0.2502, over 5659982.40 frames. ], batch size: 412, lr: 3.51e-02, grad_scale: 8.0
+2023-02-28 12:44:58,571 INFO [optim.py:369] (1/2) 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 12:45:31,727 INFO [train.py:968] (1/2) Epoch 1, batch 8850, giga_loss[loss=0.5506, simple_loss=0.5322, pruned_loss=0.2845, over 27902.00 frames. ], tot_loss[loss=0.5018, simple_loss=0.504, pruned_loss=0.2498, over 5662637.43 frames. ], libri_tot_loss[loss=0.5035, simple_loss=0.5072, pruned_loss=0.2527, over 5646216.38 frames. ], giga_tot_loss[loss=0.5051, simple_loss=0.5056, pruned_loss=0.2523, over 5650751.36 frames. ], batch size: 412, lr: 3.51e-02, grad_scale: 8.0
+2023-02-28 12:46:15,102 INFO [train.py:968] (1/2) Epoch 1, batch 8900, giga_loss[loss=0.5551, simple_loss=0.5343, pruned_loss=0.2879, over 28273.00 frames. ], tot_loss[loss=0.5058, simple_loss=0.5061, pruned_loss=0.2527, over 5668823.70 frames. ], libri_tot_loss[loss=0.503, simple_loss=0.507, pruned_loss=0.2523, over 5652825.19 frames. ], giga_tot_loss[loss=0.5089, simple_loss=0.5076, pruned_loss=0.2551, over 5653671.86 frames. ], batch size: 65, lr: 3.50e-02, grad_scale: 8.0
+2023-02-28 12:46:29,718 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 8950, giga_loss[loss=0.5145, simple_loss=0.4893, pruned_loss=0.2699, over 23487.00 frames. ], tot_loss[loss=0.5036, simple_loss=0.5038, pruned_loss=0.2517, over 5658423.11 frames. ], libri_tot_loss[loss=0.5011, simple_loss=0.5057, pruned_loss=0.2509, over 5661240.60 frames. ], giga_tot_loss[loss=0.5078, simple_loss=0.506, pruned_loss=0.2548, over 5639518.13 frames. ], batch size: 705, lr: 3.49e-02, grad_scale: 4.0
+2023-02-28 12:47:50,017 INFO [train.py:968] (1/2) Epoch 1, batch 9000, giga_loss[loss=0.4998, simple_loss=0.4998, pruned_loss=0.2499, over 28566.00 frames. ], tot_loss[loss=0.499, simple_loss=0.5007, pruned_loss=0.2486, over 5649086.10 frames. ], libri_tot_loss[loss=0.5011, simple_loss=0.506, pruned_loss=0.2507, over 5649496.43 frames. ], giga_tot_loss[loss=0.5023, simple_loss=0.5021, pruned_loss=0.2513, over 5643974.17 frames. ], batch size: 307, lr: 3.48e-02, grad_scale: 4.0
+2023-02-28 12:47:50,017 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 12:47:54,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4691, 1.4684, 1.1296, 1.0653], device='cuda:1'), covar=tensor([0.0968, 0.0915, 0.1445, 0.1671], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0703, 0.0686, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:47:55,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0462, 1.1198, 1.3338, 0.3417], device='cuda:1'), covar=tensor([0.0529, 0.0397, 0.0381, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0499, 0.0554, 0.0608], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 12:47:58,245 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19157MB
+2023-02-28 12:48:16,133 INFO [optim.py:369] (1/2) 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 12:48:31,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9033.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:48:37,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-02-28 12:48:44,636 INFO [train.py:968] (1/2) Epoch 1, batch 9050, giga_loss[loss=0.5399, simple_loss=0.4975, pruned_loss=0.2911, over 23715.00 frames. ], tot_loss[loss=0.4983, simple_loss=0.4994, pruned_loss=0.2486, over 5643446.18 frames. ], libri_tot_loss[loss=0.5006, simple_loss=0.5057, pruned_loss=0.2502, over 5641787.24 frames. ], giga_tot_loss[loss=0.5013, simple_loss=0.5006, pruned_loss=0.2511, over 5646977.60 frames. ], batch size: 705, lr: 3.48e-02, grad_scale: 4.0
+2023-02-28 12:49:36,348 INFO [train.py:968] (1/2) Epoch 1, batch 9100, giga_loss[loss=0.5428, simple_loss=0.5401, pruned_loss=0.2728, over 28875.00 frames. ], tot_loss[loss=0.5017, simple_loss=0.5008, pruned_loss=0.2513, over 5648982.14 frames. ], libri_tot_loss[loss=0.5002, simple_loss=0.5055, pruned_loss=0.2499, over 5648409.39 frames. ], giga_tot_loss[loss=0.5044, simple_loss=0.5018, pruned_loss=0.2535, over 5645927.76 frames. ], batch size: 174, lr: 3.47e-02, grad_scale: 2.0
+2023-02-28 12:49:52,845 INFO [optim.py:369] (1/2) 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 12:50:17,526 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-02-28 12:50:23,457 INFO [train.py:968] (1/2) Epoch 1, batch 9150, giga_loss[loss=0.4853, simple_loss=0.4927, pruned_loss=0.239, over 28672.00 frames. ], tot_loss[loss=0.4994, simple_loss=0.4994, pruned_loss=0.2497, over 5647562.99 frames. ], libri_tot_loss[loss=0.4981, simple_loss=0.5042, pruned_loss=0.2484, over 5658138.65 frames. ], giga_tot_loss[loss=0.5035, simple_loss=0.5011, pruned_loss=0.253, over 5636102.63 frames. ], batch size: 92, lr: 3.46e-02, grad_scale: 2.0
+2023-02-28 12:50:27,173 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9153.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:50:41,650 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-02-28 12:50:49,203 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9176.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:50:51,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9179.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:50:53,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8748, 1.8271, 3.1983, 0.4958], device='cuda:1'), covar=tensor([0.0826, 0.0733, 0.0247, 0.1839], device='cuda:1'), in_proj_covar=tensor([0.0693, 0.0563, 0.0499, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 12:51:00,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 12:51:10,909 INFO [train.py:968] (1/2) Epoch 1, batch 9200, giga_loss[loss=0.4374, simple_loss=0.4545, pruned_loss=0.2101, over 28967.00 frames. ], tot_loss[loss=0.4933, simple_loss=0.4948, pruned_loss=0.2459, over 5663750.74 frames. ], libri_tot_loss[loss=0.4976, simple_loss=0.504, pruned_loss=0.2479, over 5662260.66 frames. ], giga_tot_loss[loss=0.4971, simple_loss=0.4962, pruned_loss=0.2489, over 5650893.15 frames. ], batch size: 136, lr: 3.46e-02, grad_scale: 4.0
+2023-02-28 12:51:19,297 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9208.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:51:28,565 INFO [optim.py:369] (1/2) 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 12:51:51,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6035, 1.6592, 0.8769, 1.1657], device='cuda:1'), covar=tensor([0.1254, 0.1194, 0.2511, 0.1971], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0709, 0.0705, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:51:58,940 INFO [train.py:968] (1/2) Epoch 1, batch 9250, libri_loss[loss=0.4092, simple_loss=0.4347, pruned_loss=0.1918, over 29342.00 frames. ], tot_loss[loss=0.4926, simple_loss=0.4939, pruned_loss=0.2457, over 5660973.65 frames. ], libri_tot_loss[loss=0.4972, simple_loss=0.5038, pruned_loss=0.2475, over 5666613.33 frames. ], giga_tot_loss[loss=0.4959, simple_loss=0.4951, pruned_loss=0.2484, over 5646666.61 frames. ], batch size: 71, lr: 3.45e-02, grad_scale: 4.0
+2023-02-28 12:52:47,494 INFO [train.py:968] (1/2) Epoch 1, batch 9300, giga_loss[loss=0.5268, simple_loss=0.524, pruned_loss=0.2648, over 28369.00 frames. ], tot_loss[loss=0.4906, simple_loss=0.494, pruned_loss=0.2436, over 5659145.40 frames. ], libri_tot_loss[loss=0.4968, simple_loss=0.5035, pruned_loss=0.2472, over 5665066.42 frames. ], giga_tot_loss[loss=0.4935, simple_loss=0.4951, pruned_loss=0.2459, over 5648992.81 frames. ], batch size: 71, lr: 3.44e-02, grad_scale: 4.0
+2023-02-28 12:53:08,965 INFO [optim.py:369] (1/2) 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 12:53:10,146 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 9350, giga_loss[loss=0.4853, simple_loss=0.4963, pruned_loss=0.2372, over 28712.00 frames. ], tot_loss[loss=0.4923, simple_loss=0.4959, pruned_loss=0.2443, over 5666397.92 frames. ], libri_tot_loss[loss=0.4949, simple_loss=0.5023, pruned_loss=0.2459, over 5673104.11 frames. ], giga_tot_loss[loss=0.4962, simple_loss=0.4976, pruned_loss=0.2474, over 5650432.49 frames. ], batch size: 284, lr: 3.43e-02, grad_scale: 4.0
+2023-02-28 12:53:58,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1135, 1.5269, 1.5650, 1.2104], device='cuda:1'), covar=tensor([0.1528, 0.1304, 0.1201, 0.2478], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0590, 0.0470, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0009], device='cuda:1')
+2023-02-28 12:54:03,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9379.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:54:21,771 INFO [train.py:968] (1/2) Epoch 1, batch 9400, giga_loss[loss=0.5108, simple_loss=0.5064, pruned_loss=0.2575, over 28585.00 frames. ], tot_loss[loss=0.4947, simple_loss=0.4965, pruned_loss=0.2464, over 5661568.20 frames. ], libri_tot_loss[loss=0.4947, simple_loss=0.502, pruned_loss=0.2458, over 5675427.49 frames. ], giga_tot_loss[loss=0.498, simple_loss=0.4981, pruned_loss=0.2489, over 5646662.06 frames. ], batch size: 307, lr: 3.43e-02, grad_scale: 4.0
+2023-02-28 12:54:26,805 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:54:39,735 INFO [optim.py:369] (1/2) 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 12:55:11,359 INFO [train.py:968] (1/2) Epoch 1, batch 9450, giga_loss[loss=0.4812, simple_loss=0.5006, pruned_loss=0.2309, over 28661.00 frames. ], tot_loss[loss=0.4968, simple_loss=0.4996, pruned_loss=0.247, over 5661769.95 frames. ], libri_tot_loss[loss=0.4943, simple_loss=0.5016, pruned_loss=0.2455, over 5670104.95 frames. ], giga_tot_loss[loss=0.4998, simple_loss=0.5011, pruned_loss=0.2492, over 5653781.99 frames. ], batch size: 262, lr: 3.42e-02, grad_scale: 4.0
+2023-02-28 12:55:25,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0160, 0.5695, 0.9950, 0.0691], device='cuda:1'), covar=tensor([0.0573, 0.0663, 0.0564, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0669, 0.0558, 0.0523, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 12:55:45,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-02-28 12:55:56,873 INFO [train.py:968] (1/2) Epoch 1, batch 9500, giga_loss[loss=0.4329, simple_loss=0.4703, pruned_loss=0.1978, over 28816.00 frames. ], tot_loss[loss=0.4951, simple_loss=0.5002, pruned_loss=0.245, over 5649719.24 frames. ], libri_tot_loss[loss=0.4933, simple_loss=0.5009, pruned_loss=0.2449, over 5657460.68 frames. ], giga_tot_loss[loss=0.4983, simple_loss=0.502, pruned_loss=0.2474, over 5655009.60 frames. ], batch size: 186, lr: 3.41e-02, grad_scale: 4.0
+2023-02-28 12:56:15,912 INFO [optim.py:369] (1/2) 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 12:56:20,598 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-02-28 12:56:22,442 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 9550, giga_loss[loss=0.485, simple_loss=0.4952, pruned_loss=0.2375, over 28724.00 frames. ], tot_loss[loss=0.4951, simple_loss=0.5016, pruned_loss=0.2443, over 5657739.22 frames. ], libri_tot_loss[loss=0.4924, simple_loss=0.5002, pruned_loss=0.2442, over 5658766.65 frames. ], giga_tot_loss[loss=0.4985, simple_loss=0.5036, pruned_loss=0.2467, over 5660800.28 frames. ], batch size: 242, lr: 3.41e-02, grad_scale: 4.0
+2023-02-28 12:56:48,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1698, 1.1950, 0.8697, 1.0703], device='cuda:1'), covar=tensor([0.0782, 0.0715, 0.1264, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0694, 0.0678, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:57:29,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7493, 1.7707, 1.1101, 1.4179], device='cuda:1'), covar=tensor([0.1007, 0.0962, 0.1541, 0.1587], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0703, 0.0676, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 12:57:30,429 INFO [train.py:968] (1/2) Epoch 1, batch 9600, giga_loss[loss=0.4656, simple_loss=0.4902, pruned_loss=0.2205, over 28985.00 frames. ], tot_loss[loss=0.499, simple_loss=0.5043, pruned_loss=0.2468, over 5661512.05 frames. ], libri_tot_loss[loss=0.4916, simple_loss=0.4998, pruned_loss=0.2436, over 5661277.53 frames. ], giga_tot_loss[loss=0.5026, simple_loss=0.5063, pruned_loss=0.2495, over 5661568.01 frames. ], batch size: 164, lr: 3.40e-02, grad_scale: 8.0
+2023-02-28 12:57:41,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2821, 1.5831, 2.5908, 0.3635], device='cuda:1'), covar=tensor([0.0984, 0.0879, 0.0449, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.0683, 0.0568, 0.0535, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 12:57:48,764 INFO [optim.py:369] (1/2) 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 12:58:14,415 INFO [train.py:968] (1/2) Epoch 1, batch 9650, giga_loss[loss=0.4874, simple_loss=0.4964, pruned_loss=0.2392, over 28870.00 frames. ], tot_loss[loss=0.5034, simple_loss=0.5068, pruned_loss=0.25, over 5675343.78 frames. ], libri_tot_loss[loss=0.4907, simple_loss=0.4994, pruned_loss=0.2428, over 5668425.62 frames. ], giga_tot_loss[loss=0.5073, simple_loss=0.5089, pruned_loss=0.2529, over 5669206.48 frames. ], batch size: 186, lr: 3.39e-02, grad_scale: 4.0
+2023-02-28 12:58:38,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9671.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:58:41,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:59:03,324 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 9700, giga_loss[loss=0.4754, simple_loss=0.4879, pruned_loss=0.2315, over 28872.00 frames. ], tot_loss[loss=0.5039, simple_loss=0.5061, pruned_loss=0.2508, over 5658284.48 frames. ], libri_tot_loss[loss=0.4902, simple_loss=0.4992, pruned_loss=0.2424, over 5670488.99 frames. ], giga_tot_loss[loss=0.5075, simple_loss=0.508, pruned_loss=0.2535, over 5651648.23 frames. ], batch size: 145, lr: 3.38e-02, grad_scale: 4.0
+2023-02-28 12:59:11,022 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9703.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 12:59:24,755 INFO [optim.py:369] (1/2) 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 12:59:51,595 INFO [train.py:968] (1/2) Epoch 1, batch 9750, giga_loss[loss=0.4563, simple_loss=0.4752, pruned_loss=0.2187, over 28741.00 frames. ], tot_loss[loss=0.5033, simple_loss=0.5055, pruned_loss=0.2506, over 5667434.10 frames. ], libri_tot_loss[loss=0.4884, simple_loss=0.4979, pruned_loss=0.2412, over 5676880.31 frames. ], giga_tot_loss[loss=0.5084, simple_loss=0.5085, pruned_loss=0.2542, over 5656392.25 frames. ], batch size: 119, lr: 3.38e-02, grad_scale: 4.0
+2023-02-28 12:59:56,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9754.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:00:19,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9781.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:00:35,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2576, 0.9041, 1.5361, 0.0976], device='cuda:1'), covar=tensor([0.0526, 0.0582, 0.0468, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0693, 0.0576, 0.0551, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0004], device='cuda:1')
+2023-02-28 13:00:35,476 INFO [train.py:968] (1/2) Epoch 1, batch 9800, giga_loss[loss=0.4721, simple_loss=0.4956, pruned_loss=0.2243, over 28987.00 frames. ], tot_loss[loss=0.4982, simple_loss=0.5035, pruned_loss=0.2465, over 5677222.48 frames. ], libri_tot_loss[loss=0.4865, simple_loss=0.4967, pruned_loss=0.2398, over 5682475.16 frames. ], giga_tot_loss[loss=0.5044, simple_loss=0.5073, pruned_loss=0.2508, over 5663015.57 frames. ], batch size: 164, lr: 3.37e-02, grad_scale: 2.0
+2023-02-28 13:00:42,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9807.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:00:53,023 INFO [optim.py:369] (1/2) 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 13:01:00,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1430, 1.4572, 1.4982, 1.1823], device='cuda:1'), covar=tensor([0.1364, 0.1157, 0.1134, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0561, 0.0471, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0009], device='cuda:1')
+2023-02-28 13:01:07,162 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1237, 1.2283, 0.9740, 1.2581], device='cuda:1'), covar=tensor([0.1108, 0.1478, 0.1072, 0.1331], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0782, 0.0785, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005], device='cuda:1')
+2023-02-28 13:01:09,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9838.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:01:14,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9841.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:01:22,050 INFO [train.py:968] (1/2) Epoch 1, batch 9850, giga_loss[loss=0.4923, simple_loss=0.5178, pruned_loss=0.2334, over 28856.00 frames. ], tot_loss[loss=0.4937, simple_loss=0.502, pruned_loss=0.2427, over 5676486.03 frames. ], libri_tot_loss[loss=0.4857, simple_loss=0.4962, pruned_loss=0.2392, over 5686126.77 frames. ], giga_tot_loss[loss=0.4996, simple_loss=0.5055, pruned_loss=0.2469, over 5661806.83 frames. ], batch size: 112, lr: 3.36e-02, grad_scale: 2.0
+2023-02-28 13:01:37,928 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:01:39,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4011, 1.4683, 1.4582, 1.1917], device='cuda:1'), covar=tensor([0.1740, 0.0995, 0.1008, 0.2474], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0498, 0.0491, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0008, 0.0012], device='cuda:1')
+2023-02-28 13:02:07,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9897.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:09,668 INFO [train.py:968] (1/2) Epoch 1, batch 9900, giga_loss[loss=0.5382, simple_loss=0.5256, pruned_loss=0.2754, over 28790.00 frames. ], tot_loss[loss=0.4971, simple_loss=0.5041, pruned_loss=0.2451, over 5681873.93 frames. ], libri_tot_loss[loss=0.4851, simple_loss=0.4957, pruned_loss=0.2388, over 5691372.92 frames. ], giga_tot_loss[loss=0.5026, simple_loss=0.5075, pruned_loss=0.2489, over 5665262.33 frames. ], batch size: 284, lr: 3.36e-02, grad_scale: 2.0
+2023-02-28 13:02:09,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9900.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:28,878 INFO [optim.py:369] (1/2) 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 13:02:34,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9924.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:37,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9927.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:02:38,525 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 9950, giga_loss[loss=0.4578, simple_loss=0.4801, pruned_loss=0.2178, over 29030.00 frames. ], tot_loss[loss=0.4996, simple_loss=0.5048, pruned_loss=0.2472, over 5676036.58 frames. ], libri_tot_loss[loss=0.4849, simple_loss=0.4956, pruned_loss=0.2386, over 5694545.30 frames. ], giga_tot_loss[loss=0.5043, simple_loss=0.5077, pruned_loss=0.2504, over 5659888.04 frames. ], batch size: 128, lr: 3.35e-02, grad_scale: 2.0
+2023-02-28 13:03:08,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9956.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:03:26,244 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=9977.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:03:46,623 INFO [train.py:968] (1/2) Epoch 1, batch 10000, libri_loss[loss=0.442, simple_loss=0.4677, pruned_loss=0.2081, over 29548.00 frames. ], tot_loss[loss=0.496, simple_loss=0.5019, pruned_loss=0.2451, over 5672132.65 frames. ], libri_tot_loss[loss=0.484, simple_loss=0.4954, pruned_loss=0.2378, over 5698760.42 frames. ], giga_tot_loss[loss=0.5011, simple_loss=0.5048, pruned_loss=0.2487, over 5654152.02 frames. ], batch size: 79, lr: 3.34e-02, grad_scale: 4.0
+2023-02-28 13:04:04,356 INFO [optim.py:369] (1/2) 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 13:04:32,250 INFO [train.py:968] (1/2) Epoch 1, batch 10050, giga_loss[loss=0.4661, simple_loss=0.4819, pruned_loss=0.2251, over 28903.00 frames. ], tot_loss[loss=0.494, simple_loss=0.4994, pruned_loss=0.2443, over 5663067.58 frames. ], libri_tot_loss[loss=0.4827, simple_loss=0.4946, pruned_loss=0.2368, over 5695945.98 frames. ], giga_tot_loss[loss=0.4997, simple_loss=0.5027, pruned_loss=0.2483, over 5649921.18 frames. ], batch size: 186, lr: 3.34e-02, grad_scale: 4.0
+2023-02-28 13:04:34,288 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10051.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:05:20,478 INFO [train.py:968] (1/2) Epoch 1, batch 10100, giga_loss[loss=0.5135, simple_loss=0.4843, pruned_loss=0.2714, over 23696.00 frames. ], tot_loss[loss=0.4883, simple_loss=0.4948, pruned_loss=0.2409, over 5647920.63 frames. ], libri_tot_loss[loss=0.4825, simple_loss=0.4946, pruned_loss=0.2367, over 5678784.68 frames. ], giga_tot_loss[loss=0.4932, simple_loss=0.4975, pruned_loss=0.2444, over 5650737.92 frames. ], batch size: 705, lr: 3.33e-02, grad_scale: 4.0
+2023-02-28 13:05:41,541 INFO [optim.py:369] (1/2) 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 13:05:57,406 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10136.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:06:14,563 INFO [train.py:968] (1/2) Epoch 1, batch 10150, giga_loss[loss=0.501, simple_loss=0.5018, pruned_loss=0.2501, over 28295.00 frames. ], tot_loss[loss=0.4891, simple_loss=0.494, pruned_loss=0.2421, over 5637665.16 frames. ], libri_tot_loss[loss=0.4815, simple_loss=0.4939, pruned_loss=0.2359, over 5675794.33 frames. ], giga_tot_loss[loss=0.4941, simple_loss=0.4967, pruned_loss=0.2457, over 5641802.34 frames. ], batch size: 368, lr: 3.32e-02, grad_scale: 4.0
+2023-02-28 13:06:17,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2012, 1.2905, 0.9638, 1.1545], device='cuda:1'), covar=tensor([0.1085, 0.1527, 0.1042, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0785, 0.0788, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005], device='cuda:1')
+2023-02-28 13:06:42,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10182.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:06:59,875 INFO [train.py:968] (1/2) Epoch 1, batch 10200, giga_loss[loss=0.5432, simple_loss=0.526, pruned_loss=0.2802, over 28264.00 frames. ], tot_loss[loss=0.4861, simple_loss=0.492, pruned_loss=0.2401, over 5657461.07 frames. ], libri_tot_loss[loss=0.4796, simple_loss=0.4927, pruned_loss=0.2346, over 5682817.67 frames. ], giga_tot_loss[loss=0.4921, simple_loss=0.4953, pruned_loss=0.2444, over 5653276.10 frames. ], batch size: 368, lr: 3.32e-02, grad_scale: 4.0
+2023-02-28 13:07:10,790 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2920, 0.9648, 1.4781, 0.8689], device='cuda:1'), covar=tensor([0.0451, 0.0392, 0.0280, 0.0484], device='cuda:1'), in_proj_covar=tensor([0.0691, 0.0522, 0.0577, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 13:07:17,584 INFO [optim.py:369] (1/2) 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 13:07:24,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.10 vs. limit=2.0
+2023-02-28 13:07:43,339 INFO [train.py:968] (1/2) Epoch 1, batch 10250, giga_loss[loss=0.4133, simple_loss=0.4534, pruned_loss=0.1866, over 29022.00 frames. ], tot_loss[loss=0.4818, simple_loss=0.4894, pruned_loss=0.2371, over 5658676.69 frames. ], libri_tot_loss[loss=0.4788, simple_loss=0.4923, pruned_loss=0.2339, over 5688999.01 frames. ], giga_tot_loss[loss=0.4875, simple_loss=0.4923, pruned_loss=0.2413, over 5648604.28 frames. ], batch size: 155, lr: 3.31e-02, grad_scale: 4.0
+2023-02-28 13:07:45,265 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-02-28 13:08:09,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8847, 1.4113, 1.4189, 1.3916], device='cuda:1'), covar=tensor([0.0657, 0.1443, 0.0941, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0883, 0.0648, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 13:08:30,195 INFO [train.py:968] (1/2) Epoch 1, batch 10300, giga_loss[loss=0.507, simple_loss=0.4962, pruned_loss=0.2589, over 26559.00 frames. ], tot_loss[loss=0.4754, simple_loss=0.486, pruned_loss=0.2324, over 5656625.51 frames. ], libri_tot_loss[loss=0.4787, simple_loss=0.4922, pruned_loss=0.2338, over 5690999.45 frames. ], giga_tot_loss[loss=0.4799, simple_loss=0.4883, pruned_loss=0.2358, over 5646684.60 frames. ], batch size: 555, lr: 3.30e-02, grad_scale: 4.0
+2023-02-28 13:08:50,882 INFO [optim.py:369] (1/2) 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 13:08:57,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10325.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:09:00,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10328.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:09:20,282 INFO [train.py:968] (1/2) Epoch 1, batch 10350, libri_loss[loss=0.4727, simple_loss=0.5008, pruned_loss=0.2223, over 29642.00 frames. ], tot_loss[loss=0.4709, simple_loss=0.4833, pruned_loss=0.2292, over 5667831.63 frames. ], libri_tot_loss[loss=0.4771, simple_loss=0.4911, pruned_loss=0.2328, over 5696374.08 frames. ], giga_tot_loss[loss=0.4757, simple_loss=0.4859, pruned_loss=0.2327, over 5654371.09 frames. ], batch size: 91, lr: 3.30e-02, grad_scale: 4.0
+2023-02-28 13:09:22,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:09:27,676 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10357.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:10:08,180 INFO [train.py:968] (1/2) Epoch 1, batch 10400, giga_loss[loss=0.4412, simple_loss=0.4613, pruned_loss=0.2105, over 28746.00 frames. ], tot_loss[loss=0.47, simple_loss=0.4825, pruned_loss=0.2288, over 5670432.01 frames. ], libri_tot_loss[loss=0.4766, simple_loss=0.4909, pruned_loss=0.2323, over 5696002.42 frames. ], giga_tot_loss[loss=0.4742, simple_loss=0.4846, pruned_loss=0.2319, over 5659762.97 frames. ], batch size: 242, lr: 3.29e-02, grad_scale: 8.0
+2023-02-28 13:10:34,607 INFO [optim.py:369] (1/2) 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 13:10:39,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10426.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:10:44,488 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 10450, giga_loss[loss=0.4514, simple_loss=0.4679, pruned_loss=0.2175, over 28770.00 frames. ], tot_loss[loss=0.465, simple_loss=0.4779, pruned_loss=0.226, over 5663847.21 frames. ], libri_tot_loss[loss=0.4768, simple_loss=0.4912, pruned_loss=0.2324, over 5691349.88 frames. ], giga_tot_loss[loss=0.4678, simple_loss=0.479, pruned_loss=0.2283, over 5658886.41 frames. ], batch size: 243, lr: 3.28e-02, grad_scale: 4.0
+2023-02-28 13:11:06,766 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10455.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:11:45,204 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10495.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:11:47,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10498.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:11:50,087 INFO [train.py:968] (1/2) Epoch 1, batch 10500, giga_loss[loss=0.5745, simple_loss=0.5391, pruned_loss=0.3049, over 26658.00 frames. ], tot_loss[loss=0.4641, simple_loss=0.4771, pruned_loss=0.2256, over 5661577.71 frames. ], libri_tot_loss[loss=0.4765, simple_loss=0.491, pruned_loss=0.2322, over 5693500.96 frames. ], giga_tot_loss[loss=0.4666, simple_loss=0.478, pruned_loss=0.2275, over 5655453.33 frames. ], batch size: 555, lr: 3.28e-02, grad_scale: 4.0
+2023-02-28 13:12:01,042 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:12:14,904 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10527.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:12:36,499 INFO [train.py:968] (1/2) Epoch 1, batch 10550, giga_loss[loss=0.5614, simple_loss=0.5408, pruned_loss=0.291, over 28987.00 frames. ], tot_loss[loss=0.4684, simple_loss=0.4806, pruned_loss=0.2281, over 5664203.92 frames. ], libri_tot_loss[loss=0.4764, simple_loss=0.4911, pruned_loss=0.232, over 5697284.06 frames. ], giga_tot_loss[loss=0.4702, simple_loss=0.4811, pruned_loss=0.2297, over 5655409.48 frames. ], batch size: 136, lr: 3.27e-02, grad_scale: 4.0
+2023-02-28 13:12:52,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-02-28 13:12:55,202 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:12:58,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10572.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:13:22,041 INFO [train.py:968] (1/2) Epoch 1, batch 10600, libri_loss[loss=0.4679, simple_loss=0.5064, pruned_loss=0.2147, over 29548.00 frames. ], tot_loss[loss=0.467, simple_loss=0.4797, pruned_loss=0.2272, over 5657197.79 frames. ], libri_tot_loss[loss=0.4773, simple_loss=0.4917, pruned_loss=0.2325, over 5698840.92 frames. ], giga_tot_loss[loss=0.4675, simple_loss=0.4792, pruned_loss=0.2279, over 5647741.65 frames. ], batch size: 89, lr: 3.27e-02, grad_scale: 1.0
+2023-02-28 13:13:23,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10601.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:13:47,175 INFO [optim.py:369] (1/2) 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 13:14:12,807 INFO [train.py:968] (1/2) Epoch 1, batch 10650, giga_loss[loss=0.4665, simple_loss=0.4837, pruned_loss=0.2246, over 28547.00 frames. ], tot_loss[loss=0.465, simple_loss=0.4779, pruned_loss=0.2261, over 5659318.73 frames. ], libri_tot_loss[loss=0.4774, simple_loss=0.4918, pruned_loss=0.2326, over 5701832.65 frames. ], giga_tot_loss[loss=0.4651, simple_loss=0.4772, pruned_loss=0.2265, over 5648515.81 frames. ], batch size: 71, lr: 3.26e-02, grad_scale: 1.0
+2023-02-28 13:14:16,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-02-28 13:14:17,210 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10654.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:14:20,625 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10657.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:14:47,065 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10686.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:15:00,129 INFO [train.py:968] (1/2) Epoch 1, batch 10700, giga_loss[loss=0.4561, simple_loss=0.4764, pruned_loss=0.2179, over 28966.00 frames. ], tot_loss[loss=0.4678, simple_loss=0.4795, pruned_loss=0.228, over 5655793.30 frames. ], libri_tot_loss[loss=0.4777, simple_loss=0.4921, pruned_loss=0.2327, over 5699560.95 frames. ], giga_tot_loss[loss=0.4674, simple_loss=0.4784, pruned_loss=0.2282, over 5648675.52 frames. ], batch size: 164, lr: 3.25e-02, grad_scale: 1.0
+2023-02-28 13:15:10,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0319, 2.3697, 4.7697, 2.8938], device='cuda:1'), covar=tensor([0.1473, 0.0978, 0.0171, 0.0425], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0443, 0.0535, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004], device='cuda:1')
+2023-02-28 13:15:25,330 INFO [optim.py:369] (1/2) 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 13:15:52,292 INFO [train.py:968] (1/2) Epoch 1, batch 10750, giga_loss[loss=0.53, simple_loss=0.5211, pruned_loss=0.2695, over 28625.00 frames. ], tot_loss[loss=0.475, simple_loss=0.4846, pruned_loss=0.2327, over 5653955.10 frames. ], libri_tot_loss[loss=0.4779, simple_loss=0.4923, pruned_loss=0.2328, over 5702625.25 frames. ], giga_tot_loss[loss=0.4745, simple_loss=0.4834, pruned_loss=0.2327, over 5645181.60 frames. ], batch size: 336, lr: 3.25e-02, grad_scale: 1.0
+2023-02-28 13:16:04,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-02-28 13:16:19,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 13:16:36,575 INFO [train.py:968] (1/2) Epoch 1, batch 10800, giga_loss[loss=0.4442, simple_loss=0.4615, pruned_loss=0.2134, over 28804.00 frames. ], tot_loss[loss=0.4756, simple_loss=0.4856, pruned_loss=0.2328, over 5669034.02 frames. ], libri_tot_loss[loss=0.4772, simple_loss=0.4918, pruned_loss=0.2323, over 5708143.65 frames. ], giga_tot_loss[loss=0.4758, simple_loss=0.4849, pruned_loss=0.2333, over 5655344.34 frames. ], batch size: 99, lr: 3.24e-02, grad_scale: 2.0
+2023-02-28 13:16:38,760 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=10802.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:16:41,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10807.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:16:56,185 INFO [optim.py:369] (1/2) 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 13:17:00,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10830.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:17:17,520 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 10850, giga_loss[loss=0.4834, simple_loss=0.4912, pruned_loss=0.2378, over 28926.00 frames. ], tot_loss[loss=0.4774, simple_loss=0.4873, pruned_loss=0.2338, over 5680569.65 frames. ], libri_tot_loss[loss=0.4765, simple_loss=0.4914, pruned_loss=0.2318, over 5714868.23 frames. ], giga_tot_loss[loss=0.4782, simple_loss=0.4869, pruned_loss=0.2347, over 5662057.35 frames. ], batch size: 100, lr: 3.23e-02, grad_scale: 2.0
+2023-02-28 13:17:32,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7870, 1.4970, 1.1735, 1.2839], device='cuda:1'), covar=tensor([0.0817, 0.1066, 0.1150, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0711, 0.0673, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 13:18:09,538 INFO [train.py:968] (1/2) Epoch 1, batch 10900, giga_loss[loss=0.4833, simple_loss=0.5003, pruned_loss=0.2332, over 28881.00 frames. ], tot_loss[loss=0.4804, simple_loss=0.4888, pruned_loss=0.2361, over 5683990.32 frames. ], libri_tot_loss[loss=0.4763, simple_loss=0.4913, pruned_loss=0.2316, over 5718382.55 frames. ], giga_tot_loss[loss=0.4813, simple_loss=0.4886, pruned_loss=0.237, over 5665505.43 frames. ], batch size: 186, lr: 3.23e-02, grad_scale: 2.0
+2023-02-28 13:18:31,784 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 10950, giga_loss[loss=0.4612, simple_loss=0.4851, pruned_loss=0.2186, over 28816.00 frames. ], tot_loss[loss=0.4788, simple_loss=0.4889, pruned_loss=0.2344, over 5675673.37 frames. ], libri_tot_loss[loss=0.475, simple_loss=0.4904, pruned_loss=0.2308, over 5722707.56 frames. ], giga_tot_loss[loss=0.4807, simple_loss=0.4894, pruned_loss=0.236, over 5656300.69 frames. ], batch size: 119, lr: 3.22e-02, grad_scale: 2.0
+2023-02-28 13:18:58,933 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10950.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:19:02,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10953.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:19:23,244 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10973.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:19:25,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10976.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:19:31,970 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10982.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:19:49,378 INFO [train.py:968] (1/2) Epoch 1, batch 11000, giga_loss[loss=0.5582, simple_loss=0.5054, pruned_loss=0.3055, over 23356.00 frames. ], tot_loss[loss=0.4775, simple_loss=0.4876, pruned_loss=0.2336, over 5667352.04 frames. ], libri_tot_loss[loss=0.4739, simple_loss=0.4895, pruned_loss=0.23, over 5726607.42 frames. ], giga_tot_loss[loss=0.4801, simple_loss=0.4889, pruned_loss=0.2357, over 5647390.27 frames. ], batch size: 705, lr: 3.22e-02, grad_scale: 2.0
+2023-02-28 13:19:56,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11005.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:20:14,592 INFO [optim.py:369] (1/2) 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 13:20:42,322 INFO [train.py:968] (1/2) Epoch 1, batch 11050, giga_loss[loss=0.4872, simple_loss=0.487, pruned_loss=0.2437, over 28579.00 frames. ], tot_loss[loss=0.4791, simple_loss=0.4877, pruned_loss=0.2353, over 5664268.58 frames. ], libri_tot_loss[loss=0.4735, simple_loss=0.4891, pruned_loss=0.2299, over 5729274.10 frames. ], giga_tot_loss[loss=0.4817, simple_loss=0.489, pruned_loss=0.2371, over 5645286.19 frames. ], batch size: 336, lr: 3.21e-02, grad_scale: 2.0
+2023-02-28 13:21:05,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1954, 1.6091, 1.4906, 1.2553], device='cuda:1'), covar=tensor([0.1079, 0.1086, 0.0950, 0.1686], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0572, 0.0465, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-02-28 13:21:25,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-02-28 13:21:37,736 INFO [train.py:968] (1/2) Epoch 1, batch 11100, libri_loss[loss=0.4775, simple_loss=0.496, pruned_loss=0.2295, over 29531.00 frames. ], tot_loss[loss=0.4778, simple_loss=0.4863, pruned_loss=0.2346, over 5654799.03 frames. ], libri_tot_loss[loss=0.4728, simple_loss=0.4887, pruned_loss=0.2293, over 5731666.45 frames. ], giga_tot_loss[loss=0.4807, simple_loss=0.4877, pruned_loss=0.2368, over 5635180.60 frames. ], batch size: 81, lr: 3.20e-02, grad_scale: 2.0
+2023-02-28 13:22:01,016 INFO [optim.py:369] (1/2) 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 13:22:15,725 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11137.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:22:26,363 INFO [train.py:968] (1/2) Epoch 1, batch 11150, libri_loss[loss=0.3996, simple_loss=0.4297, pruned_loss=0.1847, over 29502.00 frames. ], tot_loss[loss=0.4741, simple_loss=0.4834, pruned_loss=0.2324, over 5655549.07 frames. ], libri_tot_loss[loss=0.4723, simple_loss=0.4884, pruned_loss=0.229, over 5731677.43 frames. ], giga_tot_loss[loss=0.477, simple_loss=0.4848, pruned_loss=0.2346, over 5637735.11 frames. ], batch size: 70, lr: 3.20e-02, grad_scale: 2.0
+2023-02-28 13:22:43,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11167.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:22:54,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11177.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:23:12,560 INFO [train.py:968] (1/2) Epoch 1, batch 11200, giga_loss[loss=0.5437, simple_loss=0.525, pruned_loss=0.2812, over 28700.00 frames. ], tot_loss[loss=0.4751, simple_loss=0.484, pruned_loss=0.2331, over 5665454.30 frames. ], libri_tot_loss[loss=0.4722, simple_loss=0.4885, pruned_loss=0.2288, over 5735922.95 frames. ], giga_tot_loss[loss=0.4776, simple_loss=0.4849, pruned_loss=0.2351, over 5645688.77 frames. ], batch size: 262, lr: 3.19e-02, grad_scale: 4.0
+2023-02-28 13:23:37,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11222.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:23:38,254 INFO [optim.py:369] (1/2) 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 13:24:03,444 INFO [train.py:968] (1/2) Epoch 1, batch 11250, giga_loss[loss=0.4922, simple_loss=0.5016, pruned_loss=0.2414, over 28599.00 frames. ], tot_loss[loss=0.4753, simple_loss=0.4836, pruned_loss=0.2335, over 5663844.72 frames. ], libri_tot_loss[loss=0.4721, simple_loss=0.4884, pruned_loss=0.2287, over 5736796.34 frames. ], giga_tot_loss[loss=0.4774, simple_loss=0.4844, pruned_loss=0.2352, over 5647154.04 frames. ], batch size: 242, lr: 3.19e-02, grad_scale: 4.0
+2023-02-28 13:24:16,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-02-28 13:24:35,114 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11278.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:24:52,412 INFO [train.py:968] (1/2) Epoch 1, batch 11300, giga_loss[loss=0.5396, simple_loss=0.5324, pruned_loss=0.2734, over 28665.00 frames. ], tot_loss[loss=0.4783, simple_loss=0.4854, pruned_loss=0.2356, over 5656211.40 frames. ], libri_tot_loss[loss=0.4723, simple_loss=0.4885, pruned_loss=0.2288, over 5730817.87 frames. ], giga_tot_loss[loss=0.4798, simple_loss=0.4858, pruned_loss=0.2369, over 5646899.95 frames. ], batch size: 307, lr: 3.18e-02, grad_scale: 4.0
+2023-02-28 13:25:03,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-02-28 13:25:12,600 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11320.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:25:14,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11323.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:25:15,151 INFO [optim.py:369] (1/2) 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 13:25:41,587 INFO [train.py:968] (1/2) Epoch 1, batch 11350, giga_loss[loss=0.5695, simple_loss=0.5331, pruned_loss=0.3029, over 26600.00 frames. ], tot_loss[loss=0.4807, simple_loss=0.487, pruned_loss=0.2372, over 5655736.68 frames. ], libri_tot_loss[loss=0.4723, simple_loss=0.4886, pruned_loss=0.2288, over 5733571.92 frames. ], giga_tot_loss[loss=0.482, simple_loss=0.4873, pruned_loss=0.2384, over 5644821.38 frames. ], batch size: 555, lr: 3.17e-02, grad_scale: 4.0
+2023-02-28 13:25:43,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:25:58,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11365.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:26:00,329 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11368.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:26:27,839 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:26:29,572 INFO [train.py:968] (1/2) Epoch 1, batch 11400, libri_loss[loss=0.504, simple_loss=0.5184, pruned_loss=0.2448, over 27527.00 frames. ], tot_loss[loss=0.4818, simple_loss=0.4879, pruned_loss=0.2379, over 5646902.30 frames. ], libri_tot_loss[loss=0.4722, simple_loss=0.4886, pruned_loss=0.2286, over 5731160.40 frames. ], giga_tot_loss[loss=0.4832, simple_loss=0.488, pruned_loss=0.2392, over 5639124.76 frames. ], batch size: 115, lr: 3.17e-02, grad_scale: 4.0
+2023-02-28 13:26:44,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7027, 1.4047, 1.4313, 0.8856], device='cuda:1'), covar=tensor([0.0450, 0.0400, 0.0368, 0.0548], device='cuda:1'), in_proj_covar=tensor([0.0681, 0.0497, 0.0554, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 13:26:52,587 INFO [optim.py:369] (1/2) 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 13:27:17,944 INFO [train.py:968] (1/2) Epoch 1, batch 11450, giga_loss[loss=0.5485, simple_loss=0.53, pruned_loss=0.2835, over 27898.00 frames. ], tot_loss[loss=0.4817, simple_loss=0.4873, pruned_loss=0.238, over 5646835.91 frames. ], libri_tot_loss[loss=0.471, simple_loss=0.4879, pruned_loss=0.2278, over 5733313.76 frames. ], giga_tot_loss[loss=0.4841, simple_loss=0.4881, pruned_loss=0.2401, over 5636348.83 frames. ], batch size: 412, lr: 3.16e-02, grad_scale: 4.0
+2023-02-28 13:28:07,983 INFO [train.py:968] (1/2) Epoch 1, batch 11500, giga_loss[loss=0.5181, simple_loss=0.4908, pruned_loss=0.2727, over 23562.00 frames. ], tot_loss[loss=0.4771, simple_loss=0.4845, pruned_loss=0.2348, over 5654238.52 frames. ], libri_tot_loss[loss=0.4707, simple_loss=0.4877, pruned_loss=0.2276, over 5734174.20 frames. ], giga_tot_loss[loss=0.4793, simple_loss=0.4852, pruned_loss=0.2366, over 5644972.41 frames. ], batch size: 705, lr: 3.16e-02, grad_scale: 4.0
+2023-02-28 13:28:20,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11512.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:28:32,631 INFO [optim.py:369] (1/2) 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 13:28:50,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11542.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:28:59,260 INFO [train.py:968] (1/2) Epoch 1, batch 11550, giga_loss[loss=0.5103, simple_loss=0.5073, pruned_loss=0.2567, over 28608.00 frames. ], tot_loss[loss=0.4781, simple_loss=0.4853, pruned_loss=0.2354, over 5651601.35 frames. ], libri_tot_loss[loss=0.4701, simple_loss=0.4872, pruned_loss=0.2272, over 5735845.57 frames. ], giga_tot_loss[loss=0.4804, simple_loss=0.4863, pruned_loss=0.2373, over 5642060.13 frames. ], batch size: 307, lr: 3.15e-02, grad_scale: 4.0
+2023-02-28 13:29:15,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11567.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:29:43,712 INFO [train.py:968] (1/2) Epoch 1, batch 11600, giga_loss[loss=0.5008, simple_loss=0.4997, pruned_loss=0.2509, over 27934.00 frames. ], tot_loss[loss=0.4756, simple_loss=0.4848, pruned_loss=0.2332, over 5669987.32 frames. ], libri_tot_loss[loss=0.4694, simple_loss=0.4869, pruned_loss=0.2267, over 5739548.79 frames. ], giga_tot_loss[loss=0.4784, simple_loss=0.4859, pruned_loss=0.2354, over 5656534.66 frames. ], batch size: 412, lr: 3.15e-02, grad_scale: 8.0
+2023-02-28 13:29:57,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1030, 2.5142, 3.8976, 1.7472], device='cuda:1'), covar=tensor([0.0487, 0.0903, 0.0856, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0671, 0.0460, 0.0777, 0.0526], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 13:30:09,565 INFO [optim.py:369] (1/2) 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 13:30:35,055 INFO [train.py:968] (1/2) Epoch 1, batch 11650, giga_loss[loss=0.4643, simple_loss=0.4859, pruned_loss=0.2214, over 28986.00 frames. ], tot_loss[loss=0.4812, simple_loss=0.4883, pruned_loss=0.2371, over 5657882.73 frames. ], libri_tot_loss[loss=0.4695, simple_loss=0.4871, pruned_loss=0.2267, over 5740931.71 frames. ], giga_tot_loss[loss=0.4835, simple_loss=0.4889, pruned_loss=0.239, over 5644370.41 frames. ], batch size: 164, lr: 3.14e-02, grad_scale: 2.0
+2023-02-28 13:30:38,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11653.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:30:39,337 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11655.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:30:41,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11658.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:08,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1222, 1.2623, 0.9832, 1.1608], device='cuda:1'), covar=tensor([0.0935, 0.1148, 0.0912, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0796, 0.0808, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005], device='cuda:1')
+2023-02-28 13:31:12,622 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11685.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:16,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11687.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:17,101 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 11700, libri_loss[loss=0.3836, simple_loss=0.4206, pruned_loss=0.1733, over 29384.00 frames. ], tot_loss[loss=0.4856, simple_loss=0.4907, pruned_loss=0.2403, over 5658729.27 frames. ], libri_tot_loss[loss=0.4688, simple_loss=0.4866, pruned_loss=0.2262, over 5742357.94 frames. ], giga_tot_loss[loss=0.4883, simple_loss=0.4918, pruned_loss=0.2424, over 5645519.44 frames. ], batch size: 67, lr: 3.13e-02, grad_scale: 2.0
+2023-02-28 13:31:35,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1154, 1.3066, 1.0893, 0.7441], device='cuda:1'), covar=tensor([0.0580, 0.0331, 0.0361, 0.0550], device='cuda:1'), in_proj_covar=tensor([0.0689, 0.0506, 0.0574, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 13:31:43,677 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11717.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:31:51,331 INFO [optim.py:369] (1/2) 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 13:32:16,497 INFO [train.py:968] (1/2) Epoch 1, batch 11750, giga_loss[loss=0.4783, simple_loss=0.4934, pruned_loss=0.2316, over 28900.00 frames. ], tot_loss[loss=0.484, simple_loss=0.4896, pruned_loss=0.2392, over 5658523.96 frames. ], libri_tot_loss[loss=0.4685, simple_loss=0.4864, pruned_loss=0.226, over 5744131.55 frames. ], giga_tot_loss[loss=0.4865, simple_loss=0.4906, pruned_loss=0.2412, over 5645904.27 frames. ], batch size: 136, lr: 3.13e-02, grad_scale: 2.0
+2023-02-28 13:32:37,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1182, 1.8847, 2.8784, 1.5675], device='cuda:1'), covar=tensor([0.0741, 0.0999, 0.1121, 0.1666], device='cuda:1'), in_proj_covar=tensor([0.0658, 0.0449, 0.0762, 0.0513], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 13:32:37,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3008, 1.3839, 1.1071, 1.2138], device='cuda:1'), covar=tensor([0.0570, 0.0749, 0.0974, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0559, 0.0918, 0.0686, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 13:32:59,357 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11796.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 13:33:03,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11799.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:33:03,750 INFO [train.py:968] (1/2) Epoch 1, batch 11800, giga_loss[loss=0.4911, simple_loss=0.4973, pruned_loss=0.2425, over 28258.00 frames. ], tot_loss[loss=0.4833, simple_loss=0.4902, pruned_loss=0.2382, over 5657920.31 frames. ], libri_tot_loss[loss=0.468, simple_loss=0.486, pruned_loss=0.2257, over 5745002.70 frames. ], giga_tot_loss[loss=0.4861, simple_loss=0.4914, pruned_loss=0.2404, over 5644556.80 frames. ], batch size: 368, lr: 3.12e-02, grad_scale: 2.0
+2023-02-28 13:33:12,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9143, 1.1594, 0.8065, 1.0760], device='cuda:1'), covar=tensor([0.1357, 0.1581, 0.1426, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0814, 0.0814, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0005, 0.0005], device='cuda:1')
+2023-02-28 13:33:27,697 INFO [optim.py:369] (1/2) 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 13:33:29,479 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11828.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 13:33:36,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2993, 1.6424, 1.5211, 1.3396], device='cuda:1'), covar=tensor([0.1348, 0.1031, 0.1107, 0.2045], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0547, 0.0458, 0.0562], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-02-28 13:33:52,778 INFO [train.py:968] (1/2) Epoch 1, batch 11850, giga_loss[loss=0.5631, simple_loss=0.5256, pruned_loss=0.3003, over 26602.00 frames. ], tot_loss[loss=0.4825, simple_loss=0.4903, pruned_loss=0.2373, over 5648898.14 frames. ], libri_tot_loss[loss=0.4685, simple_loss=0.4865, pruned_loss=0.226, over 5738987.31 frames. ], giga_tot_loss[loss=0.4846, simple_loss=0.491, pruned_loss=0.2391, over 5641540.72 frames. ], batch size: 555, lr: 3.12e-02, grad_scale: 2.0
+2023-02-28 13:34:41,192 INFO [train.py:968] (1/2) Epoch 1, batch 11900, giga_loss[loss=0.4879, simple_loss=0.4905, pruned_loss=0.2427, over 28771.00 frames. ], tot_loss[loss=0.4788, simple_loss=0.4879, pruned_loss=0.2349, over 5647152.64 frames. ], libri_tot_loss[loss=0.4679, simple_loss=0.4861, pruned_loss=0.2255, over 5739496.39 frames. ], giga_tot_loss[loss=0.4812, simple_loss=0.4888, pruned_loss=0.2368, over 5639935.02 frames. ], batch size: 119, lr: 3.11e-02, grad_scale: 2.0
+2023-02-28 13:34:41,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7734, 2.2981, 1.7452, 1.6716], device='cuda:1'), covar=tensor([0.0558, 0.0581, 0.0471, 0.0325], device='cuda:1'), in_proj_covar=tensor([0.0654, 0.0797, 0.0610, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0006, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 13:35:04,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=11925.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:35:04,801 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11942.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:35:27,179 INFO [train.py:968] (1/2) Epoch 1, batch 11950, giga_loss[loss=0.4343, simple_loss=0.4674, pruned_loss=0.2006, over 28661.00 frames. ], tot_loss[loss=0.4759, simple_loss=0.4857, pruned_loss=0.2331, over 5659544.77 frames. ], libri_tot_loss[loss=0.4678, simple_loss=0.4859, pruned_loss=0.2255, over 5742876.70 frames. ], giga_tot_loss[loss=0.4781, simple_loss=0.4866, pruned_loss=0.2348, over 5649252.87 frames. ], batch size: 242, lr: 3.11e-02, grad_scale: 2.0
+2023-02-28 13:36:19,054 INFO [train.py:968] (1/2) Epoch 1, batch 12000, giga_loss[loss=0.4816, simple_loss=0.4622, pruned_loss=0.2505, over 23576.00 frames. ], tot_loss[loss=0.4783, simple_loss=0.487, pruned_loss=0.2348, over 5646910.06 frames. ], libri_tot_loss[loss=0.4679, simple_loss=0.486, pruned_loss=0.2255, over 5743256.31 frames. ], giga_tot_loss[loss=0.48, simple_loss=0.4876, pruned_loss=0.2362, over 5638140.20 frames. ], batch size: 705, lr: 3.10e-02, grad_scale: 4.0
+2023-02-28 13:36:19,054 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 13:36:27,811 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19327MB
+2023-02-28 13:36:52,185 INFO [optim.py:369] (1/2) 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 13:36:54,739 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 12050, giga_loss[loss=0.4127, simple_loss=0.4463, pruned_loss=0.1895, over 28569.00 frames. ], tot_loss[loss=0.4779, simple_loss=0.4868, pruned_loss=0.2345, over 5657407.23 frames. ], libri_tot_loss[loss=0.4671, simple_loss=0.4856, pruned_loss=0.2249, over 5746757.93 frames. ], giga_tot_loss[loss=0.4801, simple_loss=0.4877, pruned_loss=0.2363, over 5645775.94 frames. ], batch size: 78, lr: 3.10e-02, grad_scale: 4.0
+2023-02-28 13:37:54,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4841, 1.5461, 1.0141, 1.1803], device='cuda:1'), covar=tensor([0.0888, 0.0758, 0.1326, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0707, 0.0693, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 13:37:54,733 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12085.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:37:57,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12088.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:38:01,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:38:07,661 INFO [train.py:968] (1/2) Epoch 1, batch 12100, giga_loss[loss=0.4047, simple_loss=0.4383, pruned_loss=0.1855, over 28940.00 frames. ], tot_loss[loss=0.4712, simple_loss=0.4819, pruned_loss=0.2303, over 5675540.49 frames. ], libri_tot_loss[loss=0.4668, simple_loss=0.4854, pruned_loss=0.2247, over 5748968.64 frames. ], giga_tot_loss[loss=0.4733, simple_loss=0.4827, pruned_loss=0.2319, over 5663165.24 frames. ], batch size: 145, lr: 3.09e-02, grad_scale: 4.0
+2023-02-28 13:38:24,980 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:38:58,411 INFO [train.py:968] (1/2) Epoch 1, batch 12150, giga_loss[loss=0.471, simple_loss=0.4905, pruned_loss=0.2258, over 29006.00 frames. ], tot_loss[loss=0.4732, simple_loss=0.4829, pruned_loss=0.2318, over 5671738.65 frames. ], libri_tot_loss[loss=0.4669, simple_loss=0.4855, pruned_loss=0.2247, over 5751497.55 frames. ], giga_tot_loss[loss=0.4749, simple_loss=0.4834, pruned_loss=0.2332, over 5658748.81 frames. ], batch size: 145, lr: 3.08e-02, grad_scale: 4.0
+2023-02-28 13:39:06,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5731, 1.8448, 3.8872, 2.5992], device='cuda:1'), covar=tensor([0.1475, 0.0999, 0.0248, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0436, 0.0548, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004], device='cuda:1')
+2023-02-28 13:39:32,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-02-28 13:39:49,164 INFO [train.py:968] (1/2) Epoch 1, batch 12200, giga_loss[loss=0.5132, simple_loss=0.5197, pruned_loss=0.2533, over 28532.00 frames. ], tot_loss[loss=0.4772, simple_loss=0.4859, pruned_loss=0.2343, over 5672762.17 frames. ], libri_tot_loss[loss=0.467, simple_loss=0.4857, pruned_loss=0.2247, over 5753515.88 frames. ], giga_tot_loss[loss=0.4786, simple_loss=0.4862, pruned_loss=0.2355, over 5659744.65 frames. ], batch size: 336, lr: 3.08e-02, grad_scale: 4.0
+2023-02-28 13:39:57,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12209.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:40:08,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2126, 1.1326, 1.2900, 0.7302], device='cuda:1'), covar=tensor([0.0400, 0.0290, 0.0266, 0.0453], device='cuda:1'), in_proj_covar=tensor([0.0682, 0.0482, 0.0553, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 13:40:11,938 INFO [optim.py:369] (1/2) 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 13:40:24,502 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12239.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:40:35,290 INFO [train.py:968] (1/2) Epoch 1, batch 12250, giga_loss[loss=0.4971, simple_loss=0.4984, pruned_loss=0.2479, over 28271.00 frames. ], tot_loss[loss=0.4775, simple_loss=0.4859, pruned_loss=0.2345, over 5672563.28 frames. ], libri_tot_loss[loss=0.4661, simple_loss=0.485, pruned_loss=0.2241, over 5757432.09 frames. ], giga_tot_loss[loss=0.4797, simple_loss=0.4867, pruned_loss=0.2363, over 5656443.85 frames. ], batch size: 368, lr: 3.07e-02, grad_scale: 4.0
+2023-02-28 13:41:03,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6554, 1.8396, 3.4312, 2.0429], device='cuda:1'), covar=tensor([0.1379, 0.0941, 0.0293, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0450, 0.0572, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0004], device='cuda:1')
+2023-02-28 13:41:26,156 INFO [train.py:968] (1/2) Epoch 1, batch 12300, giga_loss[loss=0.4459, simple_loss=0.4797, pruned_loss=0.2061, over 28875.00 frames. ], tot_loss[loss=0.4731, simple_loss=0.4838, pruned_loss=0.2312, over 5682646.02 frames. ], libri_tot_loss[loss=0.4658, simple_loss=0.4849, pruned_loss=0.224, over 5759639.32 frames. ], giga_tot_loss[loss=0.475, simple_loss=0.4845, pruned_loss=0.2328, over 5666997.93 frames. ], batch size: 174, lr: 3.07e-02, grad_scale: 4.0
+2023-02-28 13:41:27,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12300.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:41:38,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12313.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:41:48,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7950, 2.4716, 3.5613, 1.7294], device='cuda:1'), covar=tensor([0.0589, 0.0767, 0.0874, 0.1669], device='cuda:1'), in_proj_covar=tensor([0.0693, 0.0451, 0.0781, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 13:41:51,029 INFO [optim.py:369] (1/2) 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 13:42:05,187 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 12350, giga_loss[loss=0.4799, simple_loss=0.4955, pruned_loss=0.2321, over 28635.00 frames. ], tot_loss[loss=0.4711, simple_loss=0.4829, pruned_loss=0.2297, over 5676529.89 frames. ], libri_tot_loss[loss=0.4645, simple_loss=0.4839, pruned_loss=0.2231, over 5755486.59 frames. ], giga_tot_loss[loss=0.4742, simple_loss=0.4843, pruned_loss=0.2321, over 5663861.14 frames. ], batch size: 307, lr: 3.06e-02, grad_scale: 4.0
+2023-02-28 13:42:36,398 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12377.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:42:44,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3167, 1.2916, 1.0590, 0.9927], device='cuda:1'), covar=tensor([0.1022, 0.1308, 0.0990, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0804, 0.0833, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 13:43:00,442 INFO [train.py:968] (1/2) Epoch 1, batch 12400, giga_loss[loss=0.4254, simple_loss=0.4515, pruned_loss=0.1997, over 28872.00 frames. ], tot_loss[loss=0.4708, simple_loss=0.483, pruned_loss=0.2293, over 5681299.03 frames. ], libri_tot_loss[loss=0.4645, simple_loss=0.484, pruned_loss=0.223, over 5756069.16 frames. ], giga_tot_loss[loss=0.4733, simple_loss=0.484, pruned_loss=0.2313, over 5670257.22 frames. ], batch size: 199, lr: 3.06e-02, grad_scale: 8.0
+2023-02-28 13:43:03,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12403.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:43:28,231 INFO [optim.py:369] (1/2) 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 13:43:45,164 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12443.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:43:47,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 12450, giga_loss[loss=0.5889, simple_loss=0.5514, pruned_loss=0.3132, over 28011.00 frames. ], tot_loss[loss=0.4694, simple_loss=0.4817, pruned_loss=0.2286, over 5664322.20 frames. ], libri_tot_loss[loss=0.4649, simple_loss=0.4843, pruned_loss=0.2233, over 5747882.60 frames. ], giga_tot_loss[loss=0.4711, simple_loss=0.4821, pruned_loss=0.23, over 5662559.54 frames. ], batch size: 412, lr: 3.05e-02, grad_scale: 8.0
+2023-02-28 13:44:06,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12467.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:44:08,206 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12469.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:44:14,801 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12475.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:44:40,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1841, 1.3355, 1.3103, 1.3564], device='cuda:1'), covar=tensor([0.1444, 0.1142, 0.1130, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0560, 0.0463, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0008, 0.0010], device='cuda:1')
+2023-02-28 13:44:41,273 INFO [train.py:968] (1/2) Epoch 1, batch 12500, giga_loss[loss=0.4015, simple_loss=0.4368, pruned_loss=0.1831, over 28895.00 frames. ], tot_loss[loss=0.4684, simple_loss=0.4801, pruned_loss=0.2283, over 5661607.85 frames. ], libri_tot_loss[loss=0.4644, simple_loss=0.4839, pruned_loss=0.223, over 5748838.71 frames. ], giga_tot_loss[loss=0.4701, simple_loss=0.4808, pruned_loss=0.2297, over 5658168.23 frames. ], batch size: 174, lr: 3.05e-02, grad_scale: 8.0
+2023-02-28 13:45:06,667 INFO [optim.py:369] (1/2) 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 13:45:27,971 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12546.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:45:30,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12549.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:45:30,584 INFO [train.py:968] (1/2) Epoch 1, batch 12550, giga_loss[loss=0.4535, simple_loss=0.457, pruned_loss=0.225, over 28776.00 frames. ], tot_loss[loss=0.4667, simple_loss=0.4777, pruned_loss=0.2279, over 5671046.08 frames. ], libri_tot_loss[loss=0.4648, simple_loss=0.4842, pruned_loss=0.2232, over 5749600.51 frames. ], giga_tot_loss[loss=0.4677, simple_loss=0.478, pruned_loss=0.2287, over 5667262.20 frames. ], batch size: 99, lr: 3.04e-02, grad_scale: 4.0
+2023-02-28 13:46:00,075 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12578.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:06,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12584.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:20,520 INFO [train.py:968] (1/2) Epoch 1, batch 12600, giga_loss[loss=0.4221, simple_loss=0.4467, pruned_loss=0.1988, over 28831.00 frames. ], tot_loss[loss=0.4638, simple_loss=0.4749, pruned_loss=0.2263, over 5678165.42 frames. ], libri_tot_loss[loss=0.4644, simple_loss=0.4839, pruned_loss=0.223, over 5751068.60 frames. ], giga_tot_loss[loss=0.4649, simple_loss=0.4753, pruned_loss=0.2273, over 5672971.70 frames. ], batch size: 119, lr: 3.04e-02, grad_scale: 4.0
+2023-02-28 13:46:29,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12610.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:35,137 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12613.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:35,762 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12614.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:46:47,302 INFO [optim.py:369] (1/2) 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 13:47:03,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12642.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:47:08,980 INFO [train.py:968] (1/2) Epoch 1, batch 12650, libri_loss[loss=0.4114, simple_loss=0.4547, pruned_loss=0.184, over 29515.00 frames. ], tot_loss[loss=0.4624, simple_loss=0.4738, pruned_loss=0.2255, over 5692680.81 frames. ], libri_tot_loss[loss=0.464, simple_loss=0.4838, pruned_loss=0.2226, over 5755812.33 frames. ], giga_tot_loss[loss=0.4636, simple_loss=0.4739, pruned_loss=0.2267, over 5682314.25 frames. ], batch size: 81, lr: 3.03e-02, grad_scale: 4.0
+2023-02-28 13:47:45,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12688.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:47:51,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-02-28 13:47:55,340 INFO [train.py:968] (1/2) Epoch 1, batch 12700, giga_loss[loss=0.4799, simple_loss=0.4872, pruned_loss=0.2363, over 28806.00 frames. ], tot_loss[loss=0.4615, simple_loss=0.4735, pruned_loss=0.2247, over 5680970.96 frames. ], libri_tot_loss[loss=0.4628, simple_loss=0.4828, pruned_loss=0.2219, over 5750640.17 frames. ], giga_tot_loss[loss=0.4636, simple_loss=0.4742, pruned_loss=0.2265, over 5674643.82 frames. ], batch size: 119, lr: 3.03e-02, grad_scale: 4.0
+2023-02-28 13:48:10,160 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:48:22,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12727.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:24,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12730.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:44,296 INFO [train.py:968] (1/2) Epoch 1, batch 12750, giga_loss[loss=0.4109, simple_loss=0.4487, pruned_loss=0.1866, over 28891.00 frames. ], tot_loss[loss=0.4556, simple_loss=0.4706, pruned_loss=0.2203, over 5676508.44 frames. ], libri_tot_loss[loss=0.4616, simple_loss=0.4818, pruned_loss=0.2211, over 5744143.49 frames. ], giga_tot_loss[loss=0.4583, simple_loss=0.4718, pruned_loss=0.2224, over 5674494.27 frames. ], batch size: 174, lr: 3.02e-02, grad_scale: 4.0
+2023-02-28 13:48:47,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12752.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:50,742 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12757.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:51,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12759.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:48:53,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12760.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:49:11,923 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-02-28 13:49:21,415 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12789.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:49:32,058 INFO [train.py:968] (1/2) Epoch 1, batch 12800, giga_loss[loss=0.4349, simple_loss=0.4518, pruned_loss=0.209, over 27545.00 frames. ], tot_loss[loss=0.4464, simple_loss=0.4647, pruned_loss=0.2141, over 5676582.17 frames. ], libri_tot_loss[loss=0.4591, simple_loss=0.4796, pruned_loss=0.2197, over 5750641.73 frames. ], giga_tot_loss[loss=0.4506, simple_loss=0.4671, pruned_loss=0.217, over 5666803.31 frames. ], batch size: 472, lr: 3.02e-02, grad_scale: 8.0
+2023-02-28 13:49:58,544 INFO [optim.py:369] (1/2) 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 13:50:01,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12831.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:06,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12834.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:10,185 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=12839.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:14,756 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 12850, libri_loss[loss=0.4026, simple_loss=0.4378, pruned_loss=0.1837, over 29539.00 frames. ], tot_loss[loss=0.4371, simple_loss=0.4585, pruned_loss=0.2078, over 5677333.14 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4785, pruned_loss=0.2189, over 5756218.93 frames. ], giga_tot_loss[loss=0.4411, simple_loss=0.4609, pruned_loss=0.2106, over 5662022.59 frames. ], batch size: 81, lr: 3.01e-02, grad_scale: 4.0
+2023-02-28 13:50:31,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12859.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:35,020 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12863.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:50:55,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5304, 1.4391, 1.8608, 0.3112], device='cuda:1'), covar=tensor([0.0642, 0.0626, 0.0538, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0689, 0.0708, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 13:51:04,309 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12891.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:09,230 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12895.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:11,797 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 12900, giga_loss[loss=0.3593, simple_loss=0.4101, pruned_loss=0.1543, over 28895.00 frames. ], tot_loss[loss=0.4269, simple_loss=0.4517, pruned_loss=0.201, over 5673530.99 frames. ], libri_tot_loss[loss=0.4565, simple_loss=0.4775, pruned_loss=0.2181, over 5759253.94 frames. ], giga_tot_loss[loss=0.4308, simple_loss=0.4541, pruned_loss=0.2037, over 5657031.39 frames. ], batch size: 186, lr: 3.01e-02, grad_scale: 4.0
+2023-02-28 13:51:41,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12927.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:51:41,840 INFO [optim.py:369] (1/2) 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 13:52:06,606 INFO [train.py:968] (1/2) Epoch 1, batch 12950, giga_loss[loss=0.396, simple_loss=0.4471, pruned_loss=0.1725, over 28903.00 frames. ], tot_loss[loss=0.4176, simple_loss=0.4467, pruned_loss=0.1943, over 5671213.12 frames. ], libri_tot_loss[loss=0.4559, simple_loss=0.4771, pruned_loss=0.2177, over 5761539.19 frames. ], giga_tot_loss[loss=0.4207, simple_loss=0.4486, pruned_loss=0.1964, over 5655126.43 frames. ], batch size: 227, lr: 3.00e-02, grad_scale: 4.0
+2023-02-28 13:52:42,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12987.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:52:45,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12990.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:52:54,961 INFO [train.py:968] (1/2) Epoch 1, batch 13000, giga_loss[loss=0.3847, simple_loss=0.4326, pruned_loss=0.1683, over 29059.00 frames. ], tot_loss[loss=0.4126, simple_loss=0.4444, pruned_loss=0.1904, over 5672715.36 frames. ], libri_tot_loss[loss=0.4533, simple_loss=0.475, pruned_loss=0.2162, over 5764685.25 frames. ], giga_tot_loss[loss=0.4162, simple_loss=0.4468, pruned_loss=0.1927, over 5654428.23 frames. ], batch size: 155, lr: 3.00e-02, grad_scale: 2.0
+2023-02-28 13:53:16,625 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13019.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:53:26,082 INFO [optim.py:369] (1/2) 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 13:53:46,866 INFO [train.py:968] (1/2) Epoch 1, batch 13050, giga_loss[loss=0.3791, simple_loss=0.4287, pruned_loss=0.1647, over 28804.00 frames. ], tot_loss[loss=0.4103, simple_loss=0.443, pruned_loss=0.1888, over 5670913.18 frames. ], libri_tot_loss[loss=0.452, simple_loss=0.474, pruned_loss=0.2154, over 5763438.93 frames. ], giga_tot_loss[loss=0.4133, simple_loss=0.4451, pruned_loss=0.1908, over 5655684.88 frames. ], batch size: 284, lr: 2.99e-02, grad_scale: 2.0
+2023-02-28 13:53:52,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3347, 1.3560, 1.3666, 1.1364], device='cuda:1'), covar=tensor([0.0652, 0.0944, 0.1022, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0883, 0.0669, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 13:54:34,883 INFO [train.py:968] (1/2) Epoch 1, batch 13100, giga_loss[loss=0.4079, simple_loss=0.4229, pruned_loss=0.1965, over 26621.00 frames. ], tot_loss[loss=0.4073, simple_loss=0.4403, pruned_loss=0.1872, over 5666541.53 frames. ], libri_tot_loss[loss=0.4501, simple_loss=0.4724, pruned_loss=0.2143, over 5764526.67 frames. ], giga_tot_loss[loss=0.4099, simple_loss=0.4423, pruned_loss=0.1888, over 5649965.31 frames. ], batch size: 555, lr: 2.99e-02, grad_scale: 2.0
+2023-02-28 13:55:03,149 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 13150, giga_loss[loss=0.4456, simple_loss=0.4674, pruned_loss=0.2119, over 28927.00 frames. ], tot_loss[loss=0.4019, simple_loss=0.4364, pruned_loss=0.1837, over 5677345.93 frames. ], libri_tot_loss[loss=0.4484, simple_loss=0.471, pruned_loss=0.2132, over 5767378.20 frames. ], giga_tot_loss[loss=0.4043, simple_loss=0.4384, pruned_loss=0.1851, over 5659133.12 frames. ], batch size: 227, lr: 2.98e-02, grad_scale: 2.0
+2023-02-28 13:55:31,749 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=13156.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:56:09,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9362, 1.7183, 1.4386, 1.3840], device='cuda:1'), covar=tensor([0.0849, 0.0947, 0.1120, 0.1373], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0685, 0.0669, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 13:56:18,271 INFO [train.py:968] (1/2) Epoch 1, batch 13200, giga_loss[loss=0.3674, simple_loss=0.388, pruned_loss=0.1734, over 24193.00 frames. ], tot_loss[loss=0.4018, simple_loss=0.4363, pruned_loss=0.1837, over 5670737.71 frames. ], libri_tot_loss[loss=0.4481, simple_loss=0.4708, pruned_loss=0.2131, over 5767706.30 frames. ], giga_tot_loss[loss=0.4037, simple_loss=0.4378, pruned_loss=0.1848, over 5655840.86 frames. ], batch size: 705, lr: 2.98e-02, grad_scale: 4.0
+2023-02-28 13:56:32,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13214.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:56:46,732 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 13250, giga_loss[loss=0.3653, simple_loss=0.3884, pruned_loss=0.1711, over 24366.00 frames. ], tot_loss[loss=0.4012, simple_loss=0.4357, pruned_loss=0.1833, over 5671552.60 frames. ], libri_tot_loss[loss=0.4467, simple_loss=0.4695, pruned_loss=0.2123, over 5765894.53 frames. ], giga_tot_loss[loss=0.4023, simple_loss=0.4369, pruned_loss=0.1838, over 5657906.91 frames. ], batch size: 705, lr: 2.97e-02, grad_scale: 4.0
+2023-02-28 13:57:53,744 INFO [train.py:968] (1/2) Epoch 1, batch 13300, giga_loss[loss=0.3866, simple_loss=0.4293, pruned_loss=0.1719, over 28564.00 frames. ], tot_loss[loss=0.3957, simple_loss=0.4321, pruned_loss=0.1797, over 5674893.23 frames. ], libri_tot_loss[loss=0.4439, simple_loss=0.4672, pruned_loss=0.2107, over 5768792.39 frames. ], giga_tot_loss[loss=0.3975, simple_loss=0.4339, pruned_loss=0.1805, over 5658114.93 frames. ], batch size: 307, lr: 2.97e-02, grad_scale: 4.0
+2023-02-28 13:58:24,748 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 13350, giga_loss[loss=0.3548, simple_loss=0.4062, pruned_loss=0.1517, over 28666.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4277, pruned_loss=0.1754, over 5678982.44 frames. ], libri_tot_loss[loss=0.4434, simple_loss=0.4667, pruned_loss=0.2103, over 5771249.27 frames. ], giga_tot_loss[loss=0.3902, simple_loss=0.4289, pruned_loss=0.1758, over 5661848.39 frames. ], batch size: 262, lr: 2.96e-02, grad_scale: 4.0
+2023-02-28 13:58:54,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13357.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:58:58,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13360.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:59:26,112 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13389.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 13:59:37,446 INFO [train.py:968] (1/2) Epoch 1, batch 13400, giga_loss[loss=0.3727, simple_loss=0.4182, pruned_loss=0.1636, over 29033.00 frames. ], tot_loss[loss=0.3856, simple_loss=0.4241, pruned_loss=0.1735, over 5671781.59 frames. ], libri_tot_loss[loss=0.4422, simple_loss=0.4658, pruned_loss=0.2096, over 5774732.89 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4246, pruned_loss=0.1733, over 5651726.63 frames. ], batch size: 155, lr: 2.96e-02, grad_scale: 4.0
+2023-02-28 13:59:46,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2738, 1.3749, 1.1272, 0.8931], device='cuda:1'), covar=tensor([0.0452, 0.0345, 0.0369, 0.0483], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0484, 0.0541, 0.0603], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 14:00:08,930 INFO [optim.py:369] (1/2) 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 14:00:21,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2564, 1.3461, 1.0804, 1.1898], device='cuda:1'), covar=tensor([0.1027, 0.1054, 0.0903, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0747, 0.0817, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 14:00:31,340 INFO [train.py:968] (1/2) Epoch 1, batch 13450, giga_loss[loss=0.3473, simple_loss=0.3954, pruned_loss=0.1495, over 28627.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4223, pruned_loss=0.1739, over 5653302.30 frames. ], libri_tot_loss[loss=0.4413, simple_loss=0.465, pruned_loss=0.2091, over 5768513.65 frames. ], giga_tot_loss[loss=0.3847, simple_loss=0.4226, pruned_loss=0.1734, over 5639541.46 frames. ], batch size: 242, lr: 2.95e-02, grad_scale: 4.0
+2023-02-28 14:01:15,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6930, 1.8195, 3.2318, 1.9482], device='cuda:1'), covar=tensor([0.1279, 0.0862, 0.0291, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0431, 0.0529, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0004], device='cuda:1')
+2023-02-28 14:01:27,292 INFO [train.py:968] (1/2) Epoch 1, batch 13500, libri_loss[loss=0.4706, simple_loss=0.4833, pruned_loss=0.2289, over 19387.00 frames. ], tot_loss[loss=0.3865, simple_loss=0.4231, pruned_loss=0.1749, over 5640681.60 frames. ], libri_tot_loss[loss=0.4414, simple_loss=0.465, pruned_loss=0.2091, over 5759058.08 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4231, pruned_loss=0.1744, over 5638778.25 frames. ], batch size: 187, lr: 2.95e-02, grad_scale: 4.0
+2023-02-28 14:02:05,656 INFO [optim.py:369] (1/2) 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 14:02:08,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13531.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:02:28,126 INFO [train.py:968] (1/2) Epoch 1, batch 13550, giga_loss[loss=0.3749, simple_loss=0.4285, pruned_loss=0.1607, over 29054.00 frames. ], tot_loss[loss=0.3876, simple_loss=0.4251, pruned_loss=0.1751, over 5635512.46 frames. ], libri_tot_loss[loss=0.4408, simple_loss=0.4646, pruned_loss=0.2088, over 5760965.33 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4249, pruned_loss=0.1745, over 5630662.37 frames. ], batch size: 213, lr: 2.94e-02, grad_scale: 4.0
+2023-02-28 14:03:31,199 INFO [train.py:968] (1/2) Epoch 1, batch 13600, giga_loss[loss=0.3735, simple_loss=0.4195, pruned_loss=0.1637, over 28761.00 frames. ], tot_loss[loss=0.3907, simple_loss=0.4283, pruned_loss=0.1766, over 5639271.93 frames. ], libri_tot_loss[loss=0.4402, simple_loss=0.4641, pruned_loss=0.2084, over 5761797.72 frames. ], giga_tot_loss[loss=0.39, simple_loss=0.428, pruned_loss=0.176, over 5633217.73 frames. ], batch size: 243, lr: 2.94e-02, grad_scale: 8.0
+2023-02-28 14:04:04,692 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 13650, giga_loss[loss=0.3844, simple_loss=0.4246, pruned_loss=0.1721, over 28650.00 frames. ], tot_loss[loss=0.3937, simple_loss=0.43, pruned_loss=0.1787, over 5637143.46 frames. ], libri_tot_loss[loss=0.4385, simple_loss=0.4626, pruned_loss=0.2074, over 5756882.94 frames. ], giga_tot_loss[loss=0.393, simple_loss=0.4299, pruned_loss=0.1781, over 5632388.25 frames. ], batch size: 92, lr: 2.93e-02, grad_scale: 4.0
+2023-02-28 14:04:26,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6229, 2.2107, 3.3023, 1.5393], device='cuda:1'), covar=tensor([0.0774, 0.1001, 0.1256, 0.2045], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0437, 0.0732, 0.0504], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 14:04:58,582 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:05:00,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:05:23,614 INFO [train.py:968] (1/2) Epoch 1, batch 13700, giga_loss[loss=0.3862, simple_loss=0.4289, pruned_loss=0.1717, over 28390.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4274, pruned_loss=0.1762, over 5651914.58 frames. ], libri_tot_loss[loss=0.4368, simple_loss=0.4615, pruned_loss=0.2063, over 5759702.15 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4273, pruned_loss=0.1757, over 5641973.40 frames. ], batch size: 368, lr: 2.93e-02, grad_scale: 4.0
+2023-02-28 14:05:26,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7876, 1.5401, 1.5040, 1.4508], device='cuda:1'), covar=tensor([0.0802, 0.1476, 0.1059, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0880, 0.0657, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 14:05:31,761 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:06:25,966 INFO [train.py:968] (1/2) Epoch 1, batch 13750, giga_loss[loss=0.3759, simple_loss=0.4306, pruned_loss=0.1606, over 28502.00 frames. ], tot_loss[loss=0.3856, simple_loss=0.4249, pruned_loss=0.1731, over 5645602.68 frames. ], libri_tot_loss[loss=0.4361, simple_loss=0.4608, pruned_loss=0.206, over 5761249.08 frames. ], giga_tot_loss[loss=0.3849, simple_loss=0.4248, pruned_loss=0.1725, over 5634129.29 frames. ], batch size: 336, lr: 2.92e-02, grad_scale: 4.0
+2023-02-28 14:07:24,807 INFO [train.py:968] (1/2) Epoch 1, batch 13800, libri_loss[loss=0.4276, simple_loss=0.4556, pruned_loss=0.1998, over 25815.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4233, pruned_loss=0.1716, over 5634917.61 frames. ], libri_tot_loss[loss=0.4358, simple_loss=0.4604, pruned_loss=0.2058, over 5743723.27 frames. ], giga_tot_loss[loss=0.3813, simple_loss=0.4224, pruned_loss=0.1701, over 5638374.65 frames. ], batch size: 136, lr: 2.92e-02, grad_scale: 4.0
+2023-02-28 14:08:02,376 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 13850, giga_loss[loss=0.3483, simple_loss=0.3709, pruned_loss=0.1628, over 24126.00 frames. ], tot_loss[loss=0.3809, simple_loss=0.4197, pruned_loss=0.171, over 5643051.85 frames. ], libri_tot_loss[loss=0.4351, simple_loss=0.4599, pruned_loss=0.2054, over 5747812.42 frames. ], giga_tot_loss[loss=0.3788, simple_loss=0.4187, pruned_loss=0.1694, over 5639947.94 frames. ], batch size: 705, lr: 2.91e-02, grad_scale: 2.0
+2023-02-28 14:09:21,722 INFO [train.py:968] (1/2) Epoch 1, batch 13900, giga_loss[loss=0.3617, simple_loss=0.4017, pruned_loss=0.1608, over 28947.00 frames. ], tot_loss[loss=0.381, simple_loss=0.4198, pruned_loss=0.1711, over 5648774.68 frames. ], libri_tot_loss[loss=0.4337, simple_loss=0.4588, pruned_loss=0.2045, over 5741324.71 frames. ], giga_tot_loss[loss=0.3787, simple_loss=0.4186, pruned_loss=0.1694, over 5648222.57 frames. ], batch size: 284, lr: 2.91e-02, grad_scale: 2.0
+2023-02-28 14:09:57,078 INFO [optim.py:369] (1/2) 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 14:10:17,120 INFO [train.py:968] (1/2) Epoch 1, batch 13950, giga_loss[loss=0.4234, simple_loss=0.4463, pruned_loss=0.2002, over 27545.00 frames. ], tot_loss[loss=0.3805, simple_loss=0.42, pruned_loss=0.1705, over 5659633.30 frames. ], libri_tot_loss[loss=0.4339, simple_loss=0.4588, pruned_loss=0.2047, over 5742684.72 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4179, pruned_loss=0.168, over 5655659.61 frames. ], batch size: 472, lr: 2.90e-02, grad_scale: 2.0
+2023-02-28 14:10:41,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0047, 1.0271, 0.8095, 1.1588], device='cuda:1'), covar=tensor([0.1608, 0.1806, 0.1606, 0.1778], device='cuda:1'), in_proj_covar=tensor([0.0846, 0.0816, 0.0863, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 14:11:17,350 INFO [train.py:968] (1/2) Epoch 1, batch 14000, giga_loss[loss=0.4129, simple_loss=0.4617, pruned_loss=0.182, over 28798.00 frames. ], tot_loss[loss=0.3815, simple_loss=0.4222, pruned_loss=0.1703, over 5667390.51 frames. ], libri_tot_loss[loss=0.4329, simple_loss=0.458, pruned_loss=0.2041, over 5743378.14 frames. ], giga_tot_loss[loss=0.3781, simple_loss=0.4202, pruned_loss=0.168, over 5661161.04 frames. ], batch size: 174, lr: 2.90e-02, grad_scale: 4.0
+2023-02-28 14:11:56,224 INFO [optim.py:369] (1/2) 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 14:12:22,402 INFO [train.py:968] (1/2) Epoch 1, batch 14050, giga_loss[loss=0.3883, simple_loss=0.4244, pruned_loss=0.1761, over 28673.00 frames. ], tot_loss[loss=0.3791, simple_loss=0.4206, pruned_loss=0.1688, over 5670709.65 frames. ], libri_tot_loss[loss=0.4322, simple_loss=0.4575, pruned_loss=0.2037, over 5744974.28 frames. ], giga_tot_loss[loss=0.3766, simple_loss=0.4192, pruned_loss=0.167, over 5663723.22 frames. ], batch size: 262, lr: 2.90e-02, grad_scale: 4.0
+2023-02-28 14:13:00,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4676, 2.0656, 1.8913, 1.3792], device='cuda:1'), covar=tensor([0.1803, 0.0772, 0.0835, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0453, 0.0429, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0014], device='cuda:1')
+2023-02-28 14:13:29,433 INFO [train.py:968] (1/2) Epoch 1, batch 14100, giga_loss[loss=0.3699, simple_loss=0.4171, pruned_loss=0.1614, over 28886.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.4183, pruned_loss=0.1674, over 5682329.78 frames. ], libri_tot_loss[loss=0.4315, simple_loss=0.457, pruned_loss=0.2033, over 5749249.22 frames. ], giga_tot_loss[loss=0.3738, simple_loss=0.4167, pruned_loss=0.1655, over 5671291.27 frames. ], batch size: 213, lr: 2.89e-02, grad_scale: 4.0
+2023-02-28 14:14:04,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14130.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:14:05,230 INFO [optim.py:369] (1/2) 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 14:14:15,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-02-28 14:14:33,956 INFO [train.py:968] (1/2) Epoch 1, batch 14150, giga_loss[loss=0.4226, simple_loss=0.4477, pruned_loss=0.1988, over 28405.00 frames. ], tot_loss[loss=0.3804, simple_loss=0.4211, pruned_loss=0.1699, over 5659431.09 frames. ], libri_tot_loss[loss=0.4309, simple_loss=0.4565, pruned_loss=0.2028, over 5741269.12 frames. ], giga_tot_loss[loss=0.378, simple_loss=0.4197, pruned_loss=0.1682, over 5656035.13 frames. ], batch size: 368, lr: 2.89e-02, grad_scale: 4.0
+2023-02-28 14:15:37,455 INFO [train.py:968] (1/2) Epoch 1, batch 14200, giga_loss[loss=0.3632, simple_loss=0.4025, pruned_loss=0.162, over 26851.00 frames. ], tot_loss[loss=0.381, simple_loss=0.4245, pruned_loss=0.1688, over 5650105.91 frames. ], libri_tot_loss[loss=0.4306, simple_loss=0.4562, pruned_loss=0.2027, over 5734329.56 frames. ], giga_tot_loss[loss=0.378, simple_loss=0.4227, pruned_loss=0.1667, over 5650990.98 frames. ], batch size: 555, lr: 2.88e-02, grad_scale: 4.0
+2023-02-28 14:16:12,673 INFO [optim.py:369] (1/2) 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 14:16:35,769 INFO [train.py:968] (1/2) Epoch 1, batch 14250, giga_loss[loss=0.4185, simple_loss=0.4641, pruned_loss=0.1864, over 28926.00 frames. ], tot_loss[loss=0.3779, simple_loss=0.4233, pruned_loss=0.1662, over 5649558.63 frames. ], libri_tot_loss[loss=0.4293, simple_loss=0.4554, pruned_loss=0.2018, over 5737235.83 frames. ], giga_tot_loss[loss=0.3749, simple_loss=0.4215, pruned_loss=0.1642, over 5644935.58 frames. ], batch size: 213, lr: 2.88e-02, grad_scale: 4.0
+2023-02-28 14:17:15,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1659, 3.1044, 2.2905, 1.9173], device='cuda:1'), covar=tensor([0.0725, 0.0568, 0.0478, 0.0373], device='cuda:1'), in_proj_covar=tensor([0.0720, 0.0795, 0.0659, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 14:17:22,668 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-02-28 14:17:34,475 INFO [train.py:968] (1/2) Epoch 1, batch 14300, giga_loss[loss=0.3551, simple_loss=0.411, pruned_loss=0.1496, over 28670.00 frames. ], tot_loss[loss=0.3759, simple_loss=0.423, pruned_loss=0.1644, over 5656215.10 frames. ], libri_tot_loss[loss=0.4292, simple_loss=0.4554, pruned_loss=0.2017, over 5740280.23 frames. ], giga_tot_loss[loss=0.3727, simple_loss=0.421, pruned_loss=0.1622, over 5648217.28 frames. ], batch size: 307, lr: 2.87e-02, grad_scale: 4.0
+2023-02-28 14:18:05,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6681, 1.4368, 1.6687, 0.3070], device='cuda:1'), covar=tensor([0.0453, 0.0502, 0.0512, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0764, 0.0727, 0.0760, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 14:18:13,687 INFO [optim.py:369] (1/2) 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 14:18:41,591 INFO [train.py:968] (1/2) Epoch 1, batch 14350, giga_loss[loss=0.4471, simple_loss=0.4612, pruned_loss=0.2165, over 26948.00 frames. ], tot_loss[loss=0.376, simple_loss=0.423, pruned_loss=0.1645, over 5652878.38 frames. ], libri_tot_loss[loss=0.4292, simple_loss=0.4553, pruned_loss=0.2017, over 5732838.56 frames. ], giga_tot_loss[loss=0.373, simple_loss=0.4213, pruned_loss=0.1624, over 5651905.36 frames. ], batch size: 555, lr: 2.87e-02, grad_scale: 4.0
+2023-02-28 14:19:03,419 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-02-28 14:19:36,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-02-28 14:19:44,160 INFO [train.py:968] (1/2) Epoch 1, batch 14400, giga_loss[loss=0.3432, simple_loss=0.3959, pruned_loss=0.1452, over 28909.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.422, pruned_loss=0.1654, over 5656008.63 frames. ], libri_tot_loss[loss=0.4286, simple_loss=0.4549, pruned_loss=0.2014, over 5730356.86 frames. ], giga_tot_loss[loss=0.3737, simple_loss=0.4205, pruned_loss=0.1635, over 5656258.22 frames. ], batch size: 164, lr: 2.86e-02, grad_scale: 8.0
+2023-02-28 14:20:29,206 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 14450, giga_loss[loss=0.3809, simple_loss=0.4017, pruned_loss=0.1801, over 24701.00 frames. ], tot_loss[loss=0.3797, simple_loss=0.4238, pruned_loss=0.1677, over 5655364.25 frames. ], libri_tot_loss[loss=0.4281, simple_loss=0.4544, pruned_loss=0.2011, over 5731357.00 frames. ], giga_tot_loss[loss=0.3773, simple_loss=0.4226, pruned_loss=0.166, over 5653531.55 frames. ], batch size: 705, lr: 2.86e-02, grad_scale: 8.0
+2023-02-28 14:21:19,948 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14465.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:22:12,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14496.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:22:17,095 INFO [train.py:968] (1/2) Epoch 1, batch 14500, giga_loss[loss=0.3304, simple_loss=0.3918, pruned_loss=0.1345, over 28869.00 frames. ], tot_loss[loss=0.3743, simple_loss=0.4198, pruned_loss=0.1644, over 5672420.99 frames. ], libri_tot_loss[loss=0.4272, simple_loss=0.4539, pruned_loss=0.2004, over 5736457.13 frames. ], giga_tot_loss[loss=0.3718, simple_loss=0.4184, pruned_loss=0.1626, over 5664584.50 frames. ], batch size: 227, lr: 2.85e-02, grad_scale: 8.0
+2023-02-28 14:22:27,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14505.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:22:37,197 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14513.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 14:23:03,214 INFO [optim.py:369] (1/2) 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 14:23:20,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8215, 2.3190, 1.8323, 1.6743], device='cuda:1'), covar=tensor([0.0648, 0.0671, 0.0505, 0.0390], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0767, 0.0646, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 14:23:27,450 INFO [train.py:968] (1/2) Epoch 1, batch 14550, libri_loss[loss=0.3935, simple_loss=0.4299, pruned_loss=0.1785, over 29570.00 frames. ], tot_loss[loss=0.373, simple_loss=0.418, pruned_loss=0.164, over 5670985.08 frames. ], libri_tot_loss[loss=0.4265, simple_loss=0.4534, pruned_loss=0.2, over 5740023.49 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4164, pruned_loss=0.1621, over 5659944.83 frames. ], batch size: 78, lr: 2.85e-02, grad_scale: 4.0
+2023-02-28 14:23:34,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9095, 2.0592, 3.9983, 2.5246], device='cuda:1'), covar=tensor([0.1392, 0.0942, 0.0243, 0.0469], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0431, 0.0533, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0004], device='cuda:1')
+2023-02-28 14:24:16,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6685, 3.1124, 4.3762, 2.0424], device='cuda:1'), covar=tensor([0.0509, 0.0746, 0.1028, 0.1657], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0445, 0.0740, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 14:24:23,657 INFO [train.py:968] (1/2) Epoch 1, batch 14600, giga_loss[loss=0.341, simple_loss=0.3954, pruned_loss=0.1433, over 29009.00 frames. ], tot_loss[loss=0.3729, simple_loss=0.4175, pruned_loss=0.1642, over 5666690.44 frames. ], libri_tot_loss[loss=0.4243, simple_loss=0.452, pruned_loss=0.1984, over 5729609.98 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4155, pruned_loss=0.1621, over 5661947.81 frames. ], batch size: 136, lr: 2.85e-02, grad_scale: 4.0
+2023-02-28 14:24:38,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9150, 1.7163, 1.4827, 1.4232], device='cuda:1'), covar=tensor([0.0719, 0.1329, 0.1017, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0548, 0.0886, 0.0661, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 14:25:05,347 INFO [optim.py:369] (1/2) 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 14:25:25,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14648.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:25:26,896 INFO [train.py:968] (1/2) Epoch 1, batch 14650, giga_loss[loss=0.485, simple_loss=0.4783, pruned_loss=0.2458, over 26899.00 frames. ], tot_loss[loss=0.3775, simple_loss=0.4208, pruned_loss=0.1671, over 5668537.67 frames. ], libri_tot_loss[loss=0.4238, simple_loss=0.4517, pruned_loss=0.1981, over 5723543.19 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4185, pruned_loss=0.1649, over 5669049.64 frames. ], batch size: 555, lr: 2.84e-02, grad_scale: 4.0
+2023-02-28 14:25:28,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2191, 1.4578, 1.3667, 1.2206], device='cuda:1'), covar=tensor([0.1321, 0.0927, 0.1130, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0506, 0.0426, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0008, 0.0010], device='cuda:1')
+2023-02-28 14:25:28,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14651.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:26:03,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14680.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:26:14,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6751, 2.2402, 3.4445, 1.7794], device='cuda:1'), covar=tensor([0.0599, 0.0867, 0.0920, 0.1597], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0448, 0.0728, 0.0506], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 14:26:28,345 INFO [train.py:968] (1/2) Epoch 1, batch 14700, giga_loss[loss=0.3877, simple_loss=0.4328, pruned_loss=0.1713, over 28784.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4249, pruned_loss=0.1695, over 5676428.05 frames. ], libri_tot_loss[loss=0.4224, simple_loss=0.4508, pruned_loss=0.1972, over 5730286.72 frames. ], giga_tot_loss[loss=0.3788, simple_loss=0.4229, pruned_loss=0.1674, over 5668790.72 frames. ], batch size: 174, lr: 2.84e-02, grad_scale: 4.0
+2023-02-28 14:27:09,471 INFO [optim.py:369] (1/2) 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 14:27:16,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1793, 1.5923, 1.8489, 1.5489], device='cuda:1'), covar=tensor([0.0673, 0.1764, 0.0816, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0874, 0.0667, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 14:27:30,900 INFO [train.py:968] (1/2) Epoch 1, batch 14750, giga_loss[loss=0.3229, simple_loss=0.3881, pruned_loss=0.1289, over 28939.00 frames. ], tot_loss[loss=0.3799, simple_loss=0.4222, pruned_loss=0.1688, over 5683317.36 frames. ], libri_tot_loss[loss=0.4213, simple_loss=0.4499, pruned_loss=0.1965, over 5733577.24 frames. ], giga_tot_loss[loss=0.3777, simple_loss=0.421, pruned_loss=0.1672, over 5673250.73 frames. ], batch size: 145, lr: 2.83e-02, grad_scale: 4.0
+2023-02-28 14:27:48,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9992, 2.4540, 3.7034, 1.7622], device='cuda:1'), covar=tensor([0.0529, 0.0849, 0.0868, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0441, 0.0721, 0.0510], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 14:28:40,268 INFO [train.py:968] (1/2) Epoch 1, batch 14800, giga_loss[loss=0.4147, simple_loss=0.4426, pruned_loss=0.1934, over 27639.00 frames. ], tot_loss[loss=0.3829, simple_loss=0.4236, pruned_loss=0.1711, over 5673945.38 frames. ], libri_tot_loss[loss=0.4207, simple_loss=0.4494, pruned_loss=0.1961, over 5737084.65 frames. ], giga_tot_loss[loss=0.381, simple_loss=0.4225, pruned_loss=0.1697, over 5662098.61 frames. ], batch size: 474, lr: 2.83e-02, grad_scale: 8.0
+2023-02-28 14:28:47,521 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14806.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:28:47,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1697, 1.2147, 1.0181, 1.3185], device='cuda:1'), covar=tensor([0.1408, 0.1487, 0.1205, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0772, 0.0833, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 14:29:18,453 INFO [optim.py:369] (1/2) 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 14:29:29,678 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 14850, giga_loss[loss=0.3913, simple_loss=0.4326, pruned_loss=0.175, over 28892.00 frames. ], tot_loss[loss=0.3835, simple_loss=0.4239, pruned_loss=0.1716, over 5655405.38 frames. ], libri_tot_loss[loss=0.4206, simple_loss=0.4494, pruned_loss=0.196, over 5719476.25 frames. ], giga_tot_loss[loss=0.3813, simple_loss=0.4226, pruned_loss=0.17, over 5660734.33 frames. ], batch size: 227, lr: 2.82e-02, grad_scale: 8.0
+2023-02-28 14:30:15,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14871.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:30:24,832 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=14880.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:30:25,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8636, 2.1214, 1.9850, 1.7832], device='cuda:1'), covar=tensor([0.0624, 0.1874, 0.1106, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0885, 0.0675, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 14:30:32,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14888.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 14:30:49,210 INFO [train.py:968] (1/2) Epoch 1, batch 14900, giga_loss[loss=0.3567, simple_loss=0.4207, pruned_loss=0.1464, over 28945.00 frames. ], tot_loss[loss=0.3836, simple_loss=0.4254, pruned_loss=0.1709, over 5666167.62 frames. ], libri_tot_loss[loss=0.4193, simple_loss=0.4485, pruned_loss=0.1952, over 5723166.89 frames. ], giga_tot_loss[loss=0.3816, simple_loss=0.4242, pruned_loss=0.1695, over 5665199.02 frames. ], batch size: 213, lr: 2.82e-02, grad_scale: 8.0
+2023-02-28 14:31:33,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1593, 1.3196, 0.9422, 1.1095], device='cuda:1'), covar=tensor([0.0589, 0.0516, 0.0971, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0685, 0.0674, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 14:31:46,058 INFO [optim.py:369] (1/2) 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 14:32:11,961 INFO [train.py:968] (1/2) Epoch 1, batch 14950, giga_loss[loss=0.3421, simple_loss=0.3992, pruned_loss=0.1425, over 28784.00 frames. ], tot_loss[loss=0.3834, simple_loss=0.4255, pruned_loss=0.1707, over 5665117.85 frames. ], libri_tot_loss[loss=0.419, simple_loss=0.4482, pruned_loss=0.195, over 5723985.45 frames. ], giga_tot_loss[loss=0.3819, simple_loss=0.4247, pruned_loss=0.1695, over 5663058.39 frames. ], batch size: 119, lr: 2.82e-02, grad_scale: 4.0
+2023-02-28 14:33:03,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-02-28 14:33:11,173 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:33:18,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:33:36,413 INFO [train.py:968] (1/2) Epoch 1, batch 15000, giga_loss[loss=0.3291, simple_loss=0.3772, pruned_loss=0.1405, over 29078.00 frames. ], tot_loss[loss=0.3772, simple_loss=0.4202, pruned_loss=0.1671, over 5677979.23 frames. ], libri_tot_loss[loss=0.4189, simple_loss=0.4481, pruned_loss=0.195, over 5724715.75 frames. ], giga_tot_loss[loss=0.3755, simple_loss=0.4193, pruned_loss=0.1659, over 5674997.89 frames. ], batch size: 199, lr: 2.81e-02, grad_scale: 4.0
+2023-02-28 14:33:36,414 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 14:33:44,992 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19327MB
+2023-02-28 14:34:02,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15014.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:03,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15015.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:05,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15017.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:26,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15031.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 14:34:27,034 INFO [optim.py:369] (1/2) 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 14:34:30,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15034.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 14:34:40,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6938, 2.0966, 1.9541, 0.6556], device='cuda:1'), covar=tensor([0.0666, 0.0567, 0.0599, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0755, 0.0742, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 14:34:48,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15046.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:34:52,424 INFO [train.py:968] (1/2) Epoch 1, batch 15050, giga_loss[loss=0.3288, simple_loss=0.3866, pruned_loss=0.1355, over 28364.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4128, pruned_loss=0.1632, over 5683435.54 frames. ], libri_tot_loss[loss=0.4185, simple_loss=0.4478, pruned_loss=0.1947, over 5729560.94 frames. ], giga_tot_loss[loss=0.3675, simple_loss=0.4116, pruned_loss=0.1617, over 5675786.48 frames. ], batch size: 368, lr: 2.81e-02, grad_scale: 4.0
+2023-02-28 14:35:10,889 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15063.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 14:35:51,700 INFO [train.py:968] (1/2) Epoch 1, batch 15100, libri_loss[loss=0.4281, simple_loss=0.4546, pruned_loss=0.2008, over 29139.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4107, pruned_loss=0.1624, over 5688161.91 frames. ], libri_tot_loss[loss=0.4181, simple_loss=0.4473, pruned_loss=0.1945, over 5735825.66 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4087, pruned_loss=0.1601, over 5674381.19 frames. ], batch size: 101, lr: 2.80e-02, grad_scale: 4.0
+2023-02-28 14:35:56,055 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-02-28 14:36:22,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-02-28 14:36:32,384 INFO [optim.py:369] (1/2) 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 14:36:32,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=15133.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:36:53,122 INFO [train.py:968] (1/2) Epoch 1, batch 15150, giga_loss[loss=0.3582, simple_loss=0.4118, pruned_loss=0.1523, over 28994.00 frames. ], tot_loss[loss=0.3687, simple_loss=0.4111, pruned_loss=0.1632, over 5687863.50 frames. ], libri_tot_loss[loss=0.4168, simple_loss=0.4464, pruned_loss=0.1937, over 5740196.50 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4096, pruned_loss=0.1613, over 5671716.14 frames. ], batch size: 155, lr: 2.80e-02, grad_scale: 4.0
+2023-02-28 14:37:26,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15181.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:37:51,057 INFO [train.py:968] (1/2) Epoch 1, batch 15200, giga_loss[loss=0.3687, simple_loss=0.4145, pruned_loss=0.1615, over 28770.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4102, pruned_loss=0.1622, over 5680753.50 frames. ], libri_tot_loss[loss=0.416, simple_loss=0.4457, pruned_loss=0.1933, over 5743538.08 frames. ], giga_tot_loss[loss=0.365, simple_loss=0.4089, pruned_loss=0.1605, over 5663940.18 frames. ], batch size: 99, lr: 2.79e-02, grad_scale: 8.0
+2023-02-28 14:38:01,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.21 vs. limit=2.0
+2023-02-28 14:38:35,706 INFO [optim.py:369] (1/2) 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 14:38:42,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0759, 1.3840, 0.9996, 1.3158], device='cuda:1'), covar=tensor([0.1485, 0.1488, 0.1428, 0.1481], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0759, 0.0831, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 14:38:57,271 INFO [train.py:968] (1/2) Epoch 1, batch 15250, giga_loss[loss=0.36, simple_loss=0.3972, pruned_loss=0.1615, over 26818.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4071, pruned_loss=0.1584, over 5677832.17 frames. ], libri_tot_loss[loss=0.4156, simple_loss=0.4454, pruned_loss=0.193, over 5747036.20 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.4055, pruned_loss=0.1566, over 5659626.00 frames. ], batch size: 555, lr: 2.79e-02, grad_scale: 8.0
+2023-02-28 14:39:02,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15255.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:39:10,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4433, 1.6017, 1.6204, 1.5320], device='cuda:1'), covar=tensor([0.0961, 0.0872, 0.0861, 0.1416], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0520, 0.0441, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0010], device='cuda:1')
+2023-02-28 14:39:20,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5420, 1.9276, 1.4318, 1.1191], device='cuda:1'), covar=tensor([0.0521, 0.0281, 0.0323, 0.0480], device='cuda:1'), in_proj_covar=tensor([0.0641, 0.0491, 0.0548, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 14:39:20,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8064, 2.1541, 2.3191, 0.5704], device='cuda:1'), covar=tensor([0.0701, 0.0651, 0.0522, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0757, 0.0772, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 14:40:01,062 INFO [train.py:968] (1/2) Epoch 1, batch 15300, giga_loss[loss=0.3255, simple_loss=0.3808, pruned_loss=0.1351, over 29065.00 frames. ], tot_loss[loss=0.3592, simple_loss=0.4048, pruned_loss=0.1568, over 5672063.29 frames. ], libri_tot_loss[loss=0.4148, simple_loss=0.4448, pruned_loss=0.1925, over 5751526.40 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4029, pruned_loss=0.1549, over 5651885.54 frames. ], batch size: 155, lr: 2.79e-02, grad_scale: 8.0
+2023-02-28 14:40:36,417 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15327.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:40:48,307 INFO [optim.py:369] (1/2) 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 14:41:12,615 INFO [train.py:968] (1/2) Epoch 1, batch 15350, giga_loss[loss=0.3504, simple_loss=0.4073, pruned_loss=0.1467, over 28667.00 frames. ], tot_loss[loss=0.3589, simple_loss=0.4049, pruned_loss=0.1565, over 5676153.09 frames. ], libri_tot_loss[loss=0.4146, simple_loss=0.4447, pruned_loss=0.1924, over 5745445.29 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4028, pruned_loss=0.1544, over 5663611.86 frames. ], batch size: 307, lr: 2.78e-02, grad_scale: 8.0
+2023-02-28 14:41:23,350 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15356.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:42:24,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15398.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:42:26,035 INFO [train.py:968] (1/2) Epoch 1, batch 15400, giga_loss[loss=0.3648, simple_loss=0.4066, pruned_loss=0.1615, over 28850.00 frames. ], tot_loss[loss=0.3589, simple_loss=0.4056, pruned_loss=0.1561, over 5686545.71 frames. ], libri_tot_loss[loss=0.4146, simple_loss=0.4447, pruned_loss=0.1924, over 5745445.29 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.404, pruned_loss=0.1545, over 5676784.72 frames. ], batch size: 164, lr: 2.78e-02, grad_scale: 8.0
+2023-02-28 14:42:27,279 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15401.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:42:37,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3227, 1.2629, 1.3701, 0.6461], device='cuda:1'), covar=tensor([0.0406, 0.0247, 0.0298, 0.0481], device='cuda:1'), in_proj_covar=tensor([0.0622, 0.0470, 0.0559, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003], device='cuda:1')
+2023-02-28 14:43:04,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15430.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:43:08,763 INFO [optim.py:369] (1/2) 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 14:43:31,761 INFO [train.py:968] (1/2) Epoch 1, batch 15450, giga_loss[loss=0.3964, simple_loss=0.427, pruned_loss=0.1829, over 28928.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4073, pruned_loss=0.1584, over 5681921.77 frames. ], libri_tot_loss[loss=0.4139, simple_loss=0.444, pruned_loss=0.192, over 5740988.77 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4058, pruned_loss=0.1566, over 5676983.33 frames. ], batch size: 227, lr: 2.77e-02, grad_scale: 8.0
+2023-02-28 14:44:42,368 INFO [train.py:968] (1/2) Epoch 1, batch 15500, giga_loss[loss=0.3391, simple_loss=0.3971, pruned_loss=0.1405, over 28481.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4067, pruned_loss=0.1582, over 5676076.11 frames. ], libri_tot_loss[loss=0.4138, simple_loss=0.444, pruned_loss=0.1919, over 5734061.94 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4051, pruned_loss=0.1566, over 5676968.72 frames. ], batch size: 370, lr: 2.77e-02, grad_scale: 4.0
+2023-02-28 14:44:54,023 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15508.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:45:21,667 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 15550, giga_loss[loss=0.3699, simple_loss=0.4198, pruned_loss=0.16, over 28392.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.406, pruned_loss=0.1561, over 5663267.59 frames. ], libri_tot_loss[loss=0.4131, simple_loss=0.4436, pruned_loss=0.1914, over 5736471.32 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.4045, pruned_loss=0.1546, over 5660776.03 frames. ], batch size: 369, lr: 2.77e-02, grad_scale: 4.0
+2023-02-28 14:46:25,863 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=15588.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:46:40,716 INFO [train.py:968] (1/2) Epoch 1, batch 15600, giga_loss[loss=0.3765, simple_loss=0.43, pruned_loss=0.1615, over 28772.00 frames. ], tot_loss[loss=0.3628, simple_loss=0.4098, pruned_loss=0.1579, over 5661957.83 frames. ], libri_tot_loss[loss=0.4123, simple_loss=0.4429, pruned_loss=0.191, over 5732203.56 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4079, pruned_loss=0.1558, over 5660380.31 frames. ], batch size: 174, lr: 2.76e-02, grad_scale: 8.0
+2023-02-28 14:47:28,250 INFO [optim.py:369] (1/2) 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 14:47:47,558 INFO [train.py:968] (1/2) Epoch 1, batch 15650, giga_loss[loss=0.3638, simple_loss=0.4146, pruned_loss=0.1565, over 28920.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4111, pruned_loss=0.1582, over 5657775.91 frames. ], libri_tot_loss[loss=0.4121, simple_loss=0.4427, pruned_loss=0.1909, over 5733085.54 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4096, pruned_loss=0.1565, over 5655493.88 frames. ], batch size: 284, lr: 2.76e-02, grad_scale: 4.0
+2023-02-28 14:47:50,016 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15651.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:47:54,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15654.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:48:29,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15683.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:48:50,389 INFO [train.py:968] (1/2) Epoch 1, batch 15700, giga_loss[loss=0.3579, simple_loss=0.4092, pruned_loss=0.1533, over 28462.00 frames. ], tot_loss[loss=0.3647, simple_loss=0.4114, pruned_loss=0.1591, over 5651473.35 frames. ], libri_tot_loss[loss=0.4119, simple_loss=0.4426, pruned_loss=0.1907, over 5735286.29 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4099, pruned_loss=0.1574, over 5646607.26 frames. ], batch size: 369, lr: 2.75e-02, grad_scale: 4.0
+2023-02-28 14:49:35,282 INFO [optim.py:369] (1/2) 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 14:49:52,245 INFO [train.py:968] (1/2) Epoch 1, batch 15750, giga_loss[loss=0.3142, simple_loss=0.3744, pruned_loss=0.127, over 28487.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4101, pruned_loss=0.1583, over 5646658.63 frames. ], libri_tot_loss[loss=0.4116, simple_loss=0.4424, pruned_loss=0.1905, over 5725458.00 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4086, pruned_loss=0.1567, over 5650335.56 frames. ], batch size: 368, lr: 2.75e-02, grad_scale: 4.0
+2023-02-28 14:50:55,928 INFO [train.py:968] (1/2) Epoch 1, batch 15800, giga_loss[loss=0.345, simple_loss=0.3917, pruned_loss=0.1492, over 28654.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4077, pruned_loss=0.1566, over 5652167.03 frames. ], libri_tot_loss[loss=0.4102, simple_loss=0.4413, pruned_loss=0.1896, over 5729011.44 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4065, pruned_loss=0.1553, over 5649908.10 frames. ], batch size: 85, lr: 2.75e-02, grad_scale: 4.0
+2023-02-28 14:51:18,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-02-28 14:51:39,219 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 15850, giga_loss[loss=0.3729, simple_loss=0.4186, pruned_loss=0.1636, over 28732.00 frames. ], tot_loss[loss=0.3592, simple_loss=0.4058, pruned_loss=0.1563, over 5665582.31 frames. ], libri_tot_loss[loss=0.4084, simple_loss=0.4398, pruned_loss=0.1885, over 5733086.24 frames. ], giga_tot_loss[loss=0.3579, simple_loss=0.4053, pruned_loss=0.1553, over 5658043.26 frames. ], batch size: 307, lr: 2.74e-02, grad_scale: 4.0
+2023-02-28 14:52:59,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2449, 1.3991, 1.2868, 0.8176], device='cuda:1'), covar=tensor([0.0378, 0.0241, 0.0282, 0.0460], device='cuda:1'), in_proj_covar=tensor([0.0641, 0.0499, 0.0541, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 14:53:01,296 INFO [train.py:968] (1/2) Epoch 1, batch 15900, giga_loss[loss=0.3839, simple_loss=0.4241, pruned_loss=0.1718, over 28060.00 frames. ], tot_loss[loss=0.3588, simple_loss=0.4061, pruned_loss=0.1558, over 5668641.01 frames. ], libri_tot_loss[loss=0.4081, simple_loss=0.4397, pruned_loss=0.1883, over 5734833.03 frames. ], giga_tot_loss[loss=0.3572, simple_loss=0.4051, pruned_loss=0.1546, over 5659979.15 frames. ], batch size: 412, lr: 2.74e-02, grad_scale: 4.0
+2023-02-28 14:53:09,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8009, 1.9793, 1.2186, 1.6157], device='cuda:1'), covar=tensor([0.0826, 0.0741, 0.1217, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0649, 0.0645, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 14:53:43,001 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 15950, giga_loss[loss=0.3629, simple_loss=0.4087, pruned_loss=0.1585, over 28704.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4073, pruned_loss=0.1561, over 5675747.40 frames. ], libri_tot_loss[loss=0.4065, simple_loss=0.4386, pruned_loss=0.1873, over 5739281.74 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4064, pruned_loss=0.1549, over 5662641.80 frames. ], batch size: 307, lr: 2.73e-02, grad_scale: 4.0
+2023-02-28 14:54:20,127 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:55:06,842 INFO [train.py:968] (1/2) Epoch 1, batch 16000, giga_loss[loss=0.3866, simple_loss=0.428, pruned_loss=0.1726, over 28476.00 frames. ], tot_loss[loss=0.3621, simple_loss=0.4082, pruned_loss=0.158, over 5669734.66 frames. ], libri_tot_loss[loss=0.4054, simple_loss=0.4378, pruned_loss=0.1866, over 5745099.09 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4071, pruned_loss=0.1565, over 5651201.27 frames. ], batch size: 336, lr: 2.73e-02, grad_scale: 8.0
+2023-02-28 14:55:34,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5257, 1.6155, 2.8456, 1.7039], device='cuda:1'), covar=tensor([0.1315, 0.0899, 0.0381, 0.0733], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0431, 0.0552, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 14:55:52,432 INFO [optim.py:369] (1/2) 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 14:56:08,285 INFO [train.py:968] (1/2) Epoch 1, batch 16050, giga_loss[loss=0.3935, simple_loss=0.437, pruned_loss=0.175, over 28425.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4126, pruned_loss=0.1609, over 5659546.70 frames. ], libri_tot_loss[loss=0.4051, simple_loss=0.4376, pruned_loss=0.1863, over 5733434.62 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4113, pruned_loss=0.1594, over 5653190.58 frames. ], batch size: 368, lr: 2.73e-02, grad_scale: 4.0
+2023-02-28 14:56:56,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.97 vs. limit=5.0
+2023-02-28 14:57:09,571 INFO [train.py:968] (1/2) Epoch 1, batch 16100, giga_loss[loss=0.3929, simple_loss=0.4394, pruned_loss=0.1732, over 28911.00 frames. ], tot_loss[loss=0.3717, simple_loss=0.4166, pruned_loss=0.1634, over 5653941.40 frames. ], libri_tot_loss[loss=0.4047, simple_loss=0.4373, pruned_loss=0.1861, over 5737673.43 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4153, pruned_loss=0.1619, over 5643832.29 frames. ], batch size: 155, lr: 2.72e-02, grad_scale: 4.0
+2023-02-28 14:57:16,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16106.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:57:19,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16109.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:57:24,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6054, 1.9647, 1.9157, 0.3674], device='cuda:1'), covar=tensor([0.0752, 0.0632, 0.0585, 0.1471], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0795, 0.0803, 0.0805], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 14:57:40,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-02-28 14:57:48,059 INFO [optim.py:369] (1/2) 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 14:57:50,882 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16138.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 14:58:05,647 INFO [train.py:968] (1/2) Epoch 1, batch 16150, giga_loss[loss=0.3466, simple_loss=0.4075, pruned_loss=0.1428, over 29056.00 frames. ], tot_loss[loss=0.3738, simple_loss=0.4186, pruned_loss=0.1646, over 5651795.46 frames. ], libri_tot_loss[loss=0.4036, simple_loss=0.4365, pruned_loss=0.1854, over 5736444.12 frames. ], giga_tot_loss[loss=0.3718, simple_loss=0.4175, pruned_loss=0.163, over 5641196.25 frames. ], batch size: 136, lr: 2.72e-02, grad_scale: 4.0
+2023-02-28 14:58:13,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-02-28 14:59:08,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6896, 2.4296, 1.6658, 1.7056], device='cuda:1'), covar=tensor([0.0891, 0.0729, 0.0649, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0779, 0.0669, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 14:59:19,917 INFO [train.py:968] (1/2) Epoch 1, batch 16200, giga_loss[loss=0.3899, simple_loss=0.4284, pruned_loss=0.1757, over 28360.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.4163, pruned_loss=0.1628, over 5651317.51 frames. ], libri_tot_loss[loss=0.4033, simple_loss=0.4364, pruned_loss=0.1852, over 5734806.48 frames. ], giga_tot_loss[loss=0.3692, simple_loss=0.4153, pruned_loss=0.1615, over 5643237.42 frames. ], batch size: 368, lr: 2.72e-02, grad_scale: 4.0
+2023-02-28 15:00:04,893 INFO [optim.py:369] (1/2) 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 15:00:18,805 INFO [train.py:968] (1/2) Epoch 1, batch 16250, giga_loss[loss=0.3586, simple_loss=0.4055, pruned_loss=0.1559, over 28927.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4145, pruned_loss=0.162, over 5657083.39 frames. ], libri_tot_loss[loss=0.4038, simple_loss=0.4367, pruned_loss=0.1855, over 5731213.63 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4125, pruned_loss=0.1598, over 5651500.09 frames. ], batch size: 106, lr: 2.71e-02, grad_scale: 4.0
+2023-02-28 15:00:57,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16276.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:01:30,042 INFO [train.py:968] (1/2) Epoch 1, batch 16300, giga_loss[loss=0.3359, simple_loss=0.3938, pruned_loss=0.139, over 28719.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4134, pruned_loss=0.1608, over 5664571.21 frames. ], libri_tot_loss[loss=0.4037, simple_loss=0.4367, pruned_loss=0.1854, over 5729266.49 frames. ], giga_tot_loss[loss=0.3645, simple_loss=0.4115, pruned_loss=0.1587, over 5660364.87 frames. ], batch size: 262, lr: 2.71e-02, grad_scale: 4.0
+2023-02-28 15:02:14,124 INFO [optim.py:369] (1/2) 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 15:02:27,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6873, 3.0556, 4.4426, 1.9155], device='cuda:1'), covar=tensor([0.0348, 0.0645, 0.0742, 0.1607], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0432, 0.0724, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 15:02:32,039 INFO [train.py:968] (1/2) Epoch 1, batch 16350, giga_loss[loss=0.3105, simple_loss=0.3668, pruned_loss=0.1271, over 28850.00 frames. ], tot_loss[loss=0.368, simple_loss=0.413, pruned_loss=0.1615, over 5655410.55 frames. ], libri_tot_loss[loss=0.4033, simple_loss=0.4365, pruned_loss=0.1852, over 5723309.27 frames. ], giga_tot_loss[loss=0.3648, simple_loss=0.411, pruned_loss=0.1593, over 5654855.20 frames. ], batch size: 106, lr: 2.70e-02, grad_scale: 4.0
+2023-02-28 15:03:24,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-02-28 15:03:28,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16394.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:03:32,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4919, 3.0100, 4.2848, 1.6283], device='cuda:1'), covar=tensor([0.0591, 0.0846, 0.1090, 0.2269], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0441, 0.0725, 0.0503], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 15:03:34,352 INFO [train.py:968] (1/2) Epoch 1, batch 16400, libri_loss[loss=0.4234, simple_loss=0.4595, pruned_loss=0.1937, over 27768.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.4104, pruned_loss=0.1605, over 5656908.72 frames. ], libri_tot_loss[loss=0.4034, simple_loss=0.4365, pruned_loss=0.1853, over 5726243.01 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4082, pruned_loss=0.1581, over 5652377.60 frames. ], batch size: 116, lr: 2.70e-02, grad_scale: 8.0
+2023-02-28 15:03:59,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-02-28 15:04:16,493 INFO [optim.py:369] (1/2) 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 15:04:21,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2766, 1.2954, 1.2448, 0.6981], device='cuda:1'), covar=tensor([0.0367, 0.0266, 0.0289, 0.0454], device='cuda:1'), in_proj_covar=tensor([0.0658, 0.0508, 0.0545, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 15:04:26,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-02-28 15:04:34,749 INFO [train.py:968] (1/2) Epoch 1, batch 16450, giga_loss[loss=0.3559, simple_loss=0.4121, pruned_loss=0.1498, over 28894.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4094, pruned_loss=0.159, over 5661180.45 frames. ], libri_tot_loss[loss=0.4031, simple_loss=0.4361, pruned_loss=0.1851, over 5730819.53 frames. ], giga_tot_loss[loss=0.3604, simple_loss=0.4073, pruned_loss=0.1568, over 5651967.46 frames. ], batch size: 213, lr: 2.70e-02, grad_scale: 8.0
+2023-02-28 15:05:10,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2334, 1.2832, 1.0459, 1.2785], device='cuda:1'), covar=tensor([0.1466, 0.1479, 0.1211, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0840, 0.0765, 0.0832, 0.0850], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:05:10,976 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 16500, giga_loss[loss=0.3374, simple_loss=0.4128, pruned_loss=0.131, over 29009.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4076, pruned_loss=0.1561, over 5670460.48 frames. ], libri_tot_loss[loss=0.4026, simple_loss=0.4358, pruned_loss=0.1847, over 5732659.83 frames. ], giga_tot_loss[loss=0.3566, simple_loss=0.4055, pruned_loss=0.1538, over 5659854.58 frames. ], batch size: 120, lr: 2.69e-02, grad_scale: 4.0
+2023-02-28 15:05:47,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3232, 2.8957, 3.3111, 0.9981], device='cuda:1'), covar=tensor([0.0622, 0.0544, 0.0319, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0795, 0.0788, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 15:06:17,612 INFO [optim.py:369] (1/2) 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 15:06:25,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16542.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:06:34,138 INFO [train.py:968] (1/2) Epoch 1, batch 16550, giga_loss[loss=0.3799, simple_loss=0.4362, pruned_loss=0.1619, over 29011.00 frames. ], tot_loss[loss=0.359, simple_loss=0.4097, pruned_loss=0.1542, over 5684585.92 frames. ], libri_tot_loss[loss=0.4026, simple_loss=0.4358, pruned_loss=0.1847, over 5735161.89 frames. ], giga_tot_loss[loss=0.3555, simple_loss=0.4074, pruned_loss=0.1518, over 5672980.60 frames. ], batch size: 285, lr: 2.69e-02, grad_scale: 2.0
+2023-02-28 15:06:48,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-02-28 15:07:33,107 INFO [train.py:968] (1/2) Epoch 1, batch 16600, giga_loss[loss=0.3469, simple_loss=0.4057, pruned_loss=0.144, over 29001.00 frames. ], tot_loss[loss=0.3624, simple_loss=0.4126, pruned_loss=0.156, over 5683307.88 frames. ], libri_tot_loss[loss=0.4021, simple_loss=0.4354, pruned_loss=0.1845, over 5738549.97 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.4109, pruned_loss=0.1538, over 5670119.21 frames. ], batch size: 285, lr: 2.69e-02, grad_scale: 2.0
+2023-02-28 15:07:48,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7055, 1.7912, 4.2138, 2.6686], device='cuda:1'), covar=tensor([0.1571, 0.1091, 0.0212, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0426, 0.0526, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 15:08:17,096 INFO [optim.py:369] (1/2) 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 15:08:32,937 INFO [train.py:968] (1/2) Epoch 1, batch 16650, giga_loss[loss=0.4026, simple_loss=0.4351, pruned_loss=0.1851, over 28629.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4117, pruned_loss=0.156, over 5682178.55 frames. ], libri_tot_loss[loss=0.4006, simple_loss=0.4342, pruned_loss=0.1836, over 5744506.06 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4104, pruned_loss=0.154, over 5663783.96 frames. ], batch size: 262, lr: 2.68e-02, grad_scale: 2.0
+2023-02-28 15:08:35,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16651.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:09:40,124 INFO [train.py:968] (1/2) Epoch 1, batch 16700, giga_loss[loss=0.3305, simple_loss=0.3945, pruned_loss=0.1333, over 28675.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4129, pruned_loss=0.1577, over 5669193.56 frames. ], libri_tot_loss[loss=0.4, simple_loss=0.4339, pruned_loss=0.1832, over 5738256.74 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4115, pruned_loss=0.1556, over 5657736.07 frames. ], batch size: 307, lr: 2.68e-02, grad_scale: 2.0
+2023-02-28 15:10:08,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3446, 1.1845, 0.9932, 1.1188], device='cuda:1'), covar=tensor([0.1454, 0.1560, 0.1222, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0771, 0.0831, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:10:30,731 INFO [optim.py:369] (1/2) 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 15:10:34,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-02-28 15:10:48,175 INFO [train.py:968] (1/2) Epoch 1, batch 16750, giga_loss[loss=0.3266, simple_loss=0.3912, pruned_loss=0.131, over 28554.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4111, pruned_loss=0.1559, over 5663509.14 frames. ], libri_tot_loss[loss=0.3995, simple_loss=0.4335, pruned_loss=0.1828, over 5738654.66 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4097, pruned_loss=0.1539, over 5651837.58 frames. ], batch size: 336, lr: 2.67e-02, grad_scale: 2.0
+2023-02-28 15:11:09,155 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=16764.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:11:17,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16769.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:11:56,529 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16794.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:12:01,659 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16797.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:12:04,546 INFO [train.py:968] (1/2) Epoch 1, batch 16800, giga_loss[loss=0.3645, simple_loss=0.4185, pruned_loss=0.1552, over 28465.00 frames. ], tot_loss[loss=0.3603, simple_loss=0.4107, pruned_loss=0.1549, over 5653321.25 frames. ], libri_tot_loss[loss=0.3993, simple_loss=0.4332, pruned_loss=0.1827, over 5731683.48 frames. ], giga_tot_loss[loss=0.3578, simple_loss=0.4096, pruned_loss=0.153, over 5648368.50 frames. ], batch size: 336, lr: 2.67e-02, grad_scale: 4.0
+2023-02-28 15:12:41,483 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16826.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:12:56,297 INFO [optim.py:369] (1/2) 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 15:13:09,797 INFO [train.py:968] (1/2) Epoch 1, batch 16850, giga_loss[loss=0.3988, simple_loss=0.4469, pruned_loss=0.1753, over 28815.00 frames. ], tot_loss[loss=0.3654, simple_loss=0.4143, pruned_loss=0.1582, over 5663780.54 frames. ], libri_tot_loss[loss=0.3985, simple_loss=0.4324, pruned_loss=0.1823, over 5738426.45 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4132, pruned_loss=0.1559, over 5650251.78 frames. ], batch size: 263, lr: 2.67e-02, grad_scale: 4.0
+2023-02-28 15:13:15,515 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 16900, giga_loss[loss=0.3682, simple_loss=0.423, pruned_loss=0.1567, over 28878.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4165, pruned_loss=0.1595, over 5671492.60 frames. ], libri_tot_loss[loss=0.3974, simple_loss=0.4315, pruned_loss=0.1817, over 5742423.61 frames. ], giga_tot_loss[loss=0.3659, simple_loss=0.4162, pruned_loss=0.1578, over 5655495.17 frames. ], batch size: 155, lr: 2.66e-02, grad_scale: 4.0
+2023-02-28 15:14:34,823 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16912.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:14:39,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16915.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:14:41,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16917.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:14:46,692 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 15:15:12,232 INFO [optim.py:369] (1/2) 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 15:15:21,840 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 16950, giga_loss[loss=0.4043, simple_loss=0.4212, pruned_loss=0.1937, over 26892.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4153, pruned_loss=0.1588, over 5680319.52 frames. ], libri_tot_loss[loss=0.3972, simple_loss=0.4314, pruned_loss=0.1815, over 5745229.35 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4149, pruned_loss=0.1573, over 5664353.87 frames. ], batch size: 555, lr: 2.66e-02, grad_scale: 4.0
+2023-02-28 15:16:21,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3296, 1.5019, 1.4561, 1.2977], device='cuda:1'), covar=tensor([0.1695, 0.0827, 0.0786, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0420, 0.0400, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0016], device='cuda:1')
+2023-02-28 15:16:27,141 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16999.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:16:30,893 INFO [train.py:968] (1/2) Epoch 1, batch 17000, giga_loss[loss=0.356, simple_loss=0.4099, pruned_loss=0.1511, over 28505.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4124, pruned_loss=0.1574, over 5689022.00 frames. ], libri_tot_loss[loss=0.3957, simple_loss=0.4304, pruned_loss=0.1805, over 5753942.55 frames. ], giga_tot_loss[loss=0.362, simple_loss=0.4121, pruned_loss=0.1559, over 5665020.32 frames. ], batch size: 370, lr: 2.66e-02, grad_scale: 4.0
+2023-02-28 15:17:10,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17028.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:17:16,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0520, 1.1687, 1.1225, 1.0748], device='cuda:1'), covar=tensor([0.1292, 0.0893, 0.1123, 0.1910], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0495, 0.0432, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0010], device='cuda:1')
+2023-02-28 15:17:24,938 INFO [optim.py:369] (1/2) 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 15:17:43,304 INFO [train.py:968] (1/2) Epoch 1, batch 17050, giga_loss[loss=0.3122, simple_loss=0.3844, pruned_loss=0.12, over 28724.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4103, pruned_loss=0.1549, over 5693221.79 frames. ], libri_tot_loss[loss=0.3951, simple_loss=0.4301, pruned_loss=0.1801, over 5758567.97 frames. ], giga_tot_loss[loss=0.3582, simple_loss=0.4098, pruned_loss=0.1533, over 5667861.10 frames. ], batch size: 307, lr: 2.65e-02, grad_scale: 4.0
+2023-02-28 15:17:58,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17060.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:18:03,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17063.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:18:39,101 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:18:47,555 INFO [train.py:968] (1/2) Epoch 1, batch 17100, libri_loss[loss=0.3527, simple_loss=0.4014, pruned_loss=0.152, over 29565.00 frames. ], tot_loss[loss=0.3575, simple_loss=0.4086, pruned_loss=0.1532, over 5691972.09 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4299, pruned_loss=0.1797, over 5763009.52 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4077, pruned_loss=0.1514, over 5665369.72 frames. ], batch size: 77, lr: 2.65e-02, grad_scale: 4.0
+2023-02-28 15:19:35,072 INFO [optim.py:369] (1/2) 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 15:19:36,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17139.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:19:48,677 INFO [train.py:968] (1/2) Epoch 1, batch 17150, giga_loss[loss=0.3534, simple_loss=0.4108, pruned_loss=0.148, over 28971.00 frames. ], tot_loss[loss=0.3596, simple_loss=0.4104, pruned_loss=0.1545, over 5691835.69 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4301, pruned_loss=0.1796, over 5763869.33 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4088, pruned_loss=0.1523, over 5667216.34 frames. ], batch size: 186, lr: 2.65e-02, grad_scale: 4.0
+2023-02-28 15:19:51,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9493, 2.4864, 1.8753, 1.8289], device='cuda:1'), covar=tensor([0.0733, 0.0720, 0.0588, 0.0388], device='cuda:1'), in_proj_covar=tensor([0.0746, 0.0791, 0.0676, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 15:20:10,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3094, 1.5121, 1.4878, 0.3740], device='cuda:1'), covar=tensor([0.0676, 0.0632, 0.0652, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0824, 0.0822, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 15:20:44,817 INFO [train.py:968] (1/2) Epoch 1, batch 17200, giga_loss[loss=0.4547, simple_loss=0.4752, pruned_loss=0.217, over 27506.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4157, pruned_loss=0.1589, over 5692281.75 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4303, pruned_loss=0.1795, over 5769527.31 frames. ], giga_tot_loss[loss=0.363, simple_loss=0.4135, pruned_loss=0.1563, over 5663934.30 frames. ], batch size: 472, lr: 2.64e-02, grad_scale: 8.0
+2023-02-28 15:21:28,052 INFO [optim.py:369] (1/2) 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 15:21:38,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3588, 1.3178, 1.2079, 1.1463], device='cuda:1'), covar=tensor([0.0596, 0.0843, 0.1004, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0522, 0.0864, 0.0646, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 15:21:40,831 INFO [train.py:968] (1/2) Epoch 1, batch 17250, libri_loss[loss=0.3432, simple_loss=0.3772, pruned_loss=0.1546, over 29655.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4155, pruned_loss=0.1597, over 5696194.91 frames. ], libri_tot_loss[loss=0.3938, simple_loss=0.4297, pruned_loss=0.179, over 5772000.40 frames. ], giga_tot_loss[loss=0.3645, simple_loss=0.414, pruned_loss=0.1575, over 5668882.09 frames. ], batch size: 69, lr: 2.64e-02, grad_scale: 4.0
+2023-02-28 15:21:53,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17261.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:22:17,068 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17282.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:22:20,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17285.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:22:20,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1533, 2.0772, 2.9659, 1.3709], device='cuda:1'), covar=tensor([0.0859, 0.0929, 0.1195, 0.2021], device='cuda:1'), in_proj_covar=tensor([0.0628, 0.0435, 0.0739, 0.0510], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 15:22:34,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17299.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:22:34,467 INFO [train.py:968] (1/2) Epoch 1, batch 17300, giga_loss[loss=0.3738, simple_loss=0.4193, pruned_loss=0.1642, over 28881.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4112, pruned_loss=0.1582, over 5679231.00 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4289, pruned_loss=0.1783, over 5764227.33 frames. ], giga_tot_loss[loss=0.3615, simple_loss=0.4101, pruned_loss=0.1565, over 5661485.55 frames. ], batch size: 186, lr: 2.63e-02, grad_scale: 4.0
+2023-02-28 15:22:53,303 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17314.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:23:05,988 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-02-28 15:23:22,493 INFO [optim.py:369] (1/2) 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 15:23:35,449 INFO [train.py:968] (1/2) Epoch 1, batch 17350, giga_loss[loss=0.3793, simple_loss=0.4089, pruned_loss=0.1748, over 26946.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4106, pruned_loss=0.1587, over 5667746.13 frames. ], libri_tot_loss[loss=0.3923, simple_loss=0.4287, pruned_loss=0.178, over 5765821.18 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4097, pruned_loss=0.1574, over 5651054.85 frames. ], batch size: 555, lr: 2.63e-02, grad_scale: 4.0
+2023-02-28 15:24:34,836 INFO [train.py:968] (1/2) Epoch 1, batch 17400, libri_loss[loss=0.4485, simple_loss=0.4696, pruned_loss=0.2137, over 29671.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.42, pruned_loss=0.1668, over 5667759.80 frames. ], libri_tot_loss[loss=0.3931, simple_loss=0.4293, pruned_loss=0.1785, over 5767743.94 frames. ], giga_tot_loss[loss=0.3744, simple_loss=0.4186, pruned_loss=0.1651, over 5651247.79 frames. ], batch size: 91, lr: 2.63e-02, grad_scale: 4.0
+2023-02-28 15:25:16,926 INFO [optim.py:369] (1/2) 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 15:25:26,037 INFO [train.py:968] (1/2) Epoch 1, batch 17450, giga_loss[loss=0.5106, simple_loss=0.5241, pruned_loss=0.2485, over 28665.00 frames. ], tot_loss[loss=0.3946, simple_loss=0.4341, pruned_loss=0.1775, over 5675002.72 frames. ], libri_tot_loss[loss=0.3936, simple_loss=0.4298, pruned_loss=0.1788, over 5769674.95 frames. ], giga_tot_loss[loss=0.3921, simple_loss=0.4326, pruned_loss=0.1758, over 5658768.28 frames. ], batch size: 307, lr: 2.62e-02, grad_scale: 4.0
+2023-02-28 15:26:13,104 INFO [train.py:968] (1/2) Epoch 1, batch 17500, giga_loss[loss=0.4805, simple_loss=0.4937, pruned_loss=0.2337, over 28903.00 frames. ], tot_loss[loss=0.4013, simple_loss=0.4393, pruned_loss=0.1817, over 5674826.53 frames. ], libri_tot_loss[loss=0.394, simple_loss=0.4301, pruned_loss=0.179, over 5768824.13 frames. ], giga_tot_loss[loss=0.3991, simple_loss=0.4379, pruned_loss=0.1802, over 5661498.60 frames. ], batch size: 106, lr: 2.62e-02, grad_scale: 4.0
+2023-02-28 15:26:48,840 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 17550, giga_loss[loss=0.3935, simple_loss=0.4185, pruned_loss=0.1843, over 28869.00 frames. ], tot_loss[loss=0.3966, simple_loss=0.4341, pruned_loss=0.1795, over 5680624.44 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.4296, pruned_loss=0.1786, over 5772800.81 frames. ], giga_tot_loss[loss=0.3956, simple_loss=0.4337, pruned_loss=0.1788, over 5663587.71 frames. ], batch size: 99, lr: 2.62e-02, grad_scale: 4.0
+2023-02-28 15:27:27,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17582.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:27:44,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9373, 1.8448, 2.2412, 1.8573], device='cuda:1'), covar=tensor([0.1089, 0.0982, 0.0935, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0504, 0.0426, 0.0528], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0011], device='cuda:1')
+2023-02-28 15:27:44,707 INFO [train.py:968] (1/2) Epoch 1, batch 17600, giga_loss[loss=0.3114, simple_loss=0.3609, pruned_loss=0.1309, over 28825.00 frames. ], tot_loss[loss=0.3872, simple_loss=0.4251, pruned_loss=0.1746, over 5690097.00 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4292, pruned_loss=0.1783, over 5774809.32 frames. ], giga_tot_loss[loss=0.3868, simple_loss=0.4251, pruned_loss=0.1743, over 5673691.85 frames. ], batch size: 119, lr: 2.61e-02, grad_scale: 8.0
+2023-02-28 15:28:17,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17636.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:28:20,340 INFO [optim.py:369] (1/2) 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 15:28:28,312 INFO [train.py:968] (1/2) Epoch 1, batch 17650, giga_loss[loss=0.3392, simple_loss=0.3822, pruned_loss=0.1481, over 28317.00 frames. ], tot_loss[loss=0.3786, simple_loss=0.4172, pruned_loss=0.17, over 5678237.61 frames. ], libri_tot_loss[loss=0.3937, simple_loss=0.43, pruned_loss=0.1788, over 5754936.20 frames. ], giga_tot_loss[loss=0.3773, simple_loss=0.4163, pruned_loss=0.1691, over 5679958.94 frames. ], batch size: 368, lr: 2.61e-02, grad_scale: 4.0
+2023-02-28 15:28:51,876 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:29:16,075 INFO [train.py:968] (1/2) Epoch 1, batch 17700, giga_loss[loss=0.2815, simple_loss=0.3362, pruned_loss=0.1134, over 28499.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4065, pruned_loss=0.1638, over 5682018.38 frames. ], libri_tot_loss[loss=0.394, simple_loss=0.4302, pruned_loss=0.1789, over 5756532.37 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4054, pruned_loss=0.1629, over 5681116.80 frames. ], batch size: 65, lr: 2.61e-02, grad_scale: 4.0
+2023-02-28 15:29:47,799 INFO [optim.py:369] (1/2) 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 15:29:56,183 INFO [train.py:968] (1/2) Epoch 1, batch 17750, giga_loss[loss=0.35, simple_loss=0.3863, pruned_loss=0.1569, over 29041.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4002, pruned_loss=0.1603, over 5689323.02 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4307, pruned_loss=0.1792, over 5762996.11 frames. ], giga_tot_loss[loss=0.3572, simple_loss=0.3975, pruned_loss=0.1584, over 5679398.03 frames. ], batch size: 106, lr: 2.60e-02, grad_scale: 4.0
+2023-02-28 15:29:59,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17755.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:30:16,179 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=17776.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:30:19,312 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17779.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:30:21,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17782.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:30:32,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2474, 2.5043, 5.0538, 3.1166], device='cuda:1'), covar=tensor([0.1346, 0.0905, 0.0170, 0.0410], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0432, 0.0540, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 15:30:36,119 INFO [train.py:968] (1/2) Epoch 1, batch 17800, giga_loss[loss=0.3248, simple_loss=0.3729, pruned_loss=0.1384, over 28992.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.3949, pruned_loss=0.1568, over 5693621.12 frames. ], libri_tot_loss[loss=0.3948, simple_loss=0.4311, pruned_loss=0.1793, over 5764195.95 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.3915, pruned_loss=0.1547, over 5682909.17 frames. ], batch size: 164, lr: 2.60e-02, grad_scale: 4.0
+2023-02-28 15:30:45,758 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17811.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:30:50,765 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17817.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:30:53,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:31:00,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6486, 1.5713, 1.0518, 1.1901], device='cuda:1'), covar=tensor([0.0991, 0.1143, 0.1414, 0.1562], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0676, 0.0644, 0.0620], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 15:31:07,120 INFO [optim.py:369] (1/2) 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 15:31:16,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17849.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:31:17,327 INFO [train.py:968] (1/2) Epoch 1, batch 17850, giga_loss[loss=0.4023, simple_loss=0.4082, pruned_loss=0.1982, over 26663.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.3909, pruned_loss=0.1539, over 5700629.47 frames. ], libri_tot_loss[loss=0.3953, simple_loss=0.4316, pruned_loss=0.1795, over 5765148.55 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.3869, pruned_loss=0.1514, over 5689752.19 frames. ], batch size: 555, lr: 2.60e-02, grad_scale: 4.0
+2023-02-28 15:31:35,706 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-02-28 15:32:02,925 INFO [train.py:968] (1/2) Epoch 1, batch 17900, giga_loss[loss=0.2958, simple_loss=0.3377, pruned_loss=0.1269, over 28578.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3856, pruned_loss=0.1501, over 5695194.25 frames. ], libri_tot_loss[loss=0.3958, simple_loss=0.4321, pruned_loss=0.1798, over 5766083.37 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3814, pruned_loss=0.1476, over 5685092.81 frames. ], batch size: 85, lr: 2.59e-02, grad_scale: 4.0
+2023-02-28 15:32:35,486 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 17950, giga_loss[loss=0.2898, simple_loss=0.3442, pruned_loss=0.1177, over 28459.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3828, pruned_loss=0.1488, over 5697719.23 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.4328, pruned_loss=0.1803, over 5766328.27 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3783, pruned_loss=0.146, over 5688465.53 frames. ], batch size: 60, lr: 2.59e-02, grad_scale: 4.0
+2023-02-28 15:32:53,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17957.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:33:32,254 INFO [train.py:968] (1/2) Epoch 1, batch 18000, giga_loss[loss=0.3197, simple_loss=0.3628, pruned_loss=0.1383, over 28817.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3785, pruned_loss=0.1457, over 5698376.34 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4331, pruned_loss=0.1805, over 5764413.89 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3739, pruned_loss=0.1429, over 5691778.01 frames. ], batch size: 186, lr: 2.59e-02, grad_scale: 8.0
+2023-02-28 15:33:32,255 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 15:33:38,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1257, 1.1419, 1.3756, 0.6917], device='cuda:1'), covar=tensor([0.0336, 0.0243, 0.0235, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0680, 0.0513, 0.0560, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 15:33:41,882 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19327MB
+2023-02-28 15:33:51,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8932, 2.3861, 2.5577, 1.6839], device='cuda:1'), covar=tensor([0.1481, 0.0662, 0.0609, 0.1872], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0401, 0.0382, 0.0617], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0016], device='cuda:1')
+2023-02-28 15:34:16,598 INFO [optim.py:369] (1/2) 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 15:34:27,060 INFO [train.py:968] (1/2) Epoch 1, batch 18050, giga_loss[loss=0.3068, simple_loss=0.3582, pruned_loss=0.1277, over 28855.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3755, pruned_loss=0.1443, over 5688547.90 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.433, pruned_loss=0.1805, over 5763675.36 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3708, pruned_loss=0.1413, over 5683057.46 frames. ], batch size: 174, lr: 2.58e-02, grad_scale: 4.0
+2023-02-28 15:34:47,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2882, 1.1928, 1.1694, 1.0643], device='cuda:1'), covar=tensor([0.0793, 0.1100, 0.1265, 0.1045], device='cuda:1'), in_proj_covar=tensor([0.0559, 0.0902, 0.0678, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 15:35:02,240 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18093.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:07,820 INFO [train.py:968] (1/2) Epoch 1, batch 18100, giga_loss[loss=0.3096, simple_loss=0.3598, pruned_loss=0.1297, over 28631.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3746, pruned_loss=0.1443, over 5694523.23 frames. ], libri_tot_loss[loss=0.398, simple_loss=0.4338, pruned_loss=0.1812, over 5768293.94 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3682, pruned_loss=0.1402, over 5683693.68 frames. ], batch size: 307, lr: 2.58e-02, grad_scale: 4.0
+2023-02-28 15:35:08,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18100.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:11,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18103.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:30,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18123.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:39,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18130.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:43,238 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18132.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:35:49,403 INFO [optim.py:369] (1/2) 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 15:35:56,376 INFO [train.py:968] (1/2) Epoch 1, batch 18150, giga_loss[loss=0.2563, simple_loss=0.3189, pruned_loss=0.09684, over 28677.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3713, pruned_loss=0.1413, over 5707725.13 frames. ], libri_tot_loss[loss=0.3988, simple_loss=0.4347, pruned_loss=0.1815, over 5771386.67 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3636, pruned_loss=0.1365, over 5694350.28 frames. ], batch size: 119, lr: 2.58e-02, grad_scale: 4.0
+2023-02-28 15:35:57,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18151.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:36:16,334 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-02-28 15:36:24,451 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-02-28 15:36:38,778 INFO [train.py:968] (1/2) Epoch 1, batch 18200, libri_loss[loss=0.4037, simple_loss=0.4481, pruned_loss=0.1796, over 29491.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3724, pruned_loss=0.1432, over 5708012.28 frames. ], libri_tot_loss[loss=0.3997, simple_loss=0.4354, pruned_loss=0.182, over 5775005.45 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3642, pruned_loss=0.138, over 5692632.84 frames. ], batch size: 85, lr: 2.57e-02, grad_scale: 4.0
+2023-02-28 15:37:09,293 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-02-28 15:37:24,856 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 18250, giga_loss[loss=0.4636, simple_loss=0.4809, pruned_loss=0.2232, over 28614.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.3872, pruned_loss=0.1539, over 5706077.51 frames. ], libri_tot_loss[loss=0.3996, simple_loss=0.4352, pruned_loss=0.182, over 5776730.56 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.38, pruned_loss=0.1494, over 5691387.78 frames. ], batch size: 336, lr: 2.57e-02, grad_scale: 4.0
+2023-02-28 15:37:55,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18273.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:37:57,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18276.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:38:04,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2534, 1.2885, 1.2175, 1.1888], device='cuda:1'), covar=tensor([0.0804, 0.0861, 0.1079, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0894, 0.0679, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 15:38:08,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-02-28 15:38:12,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7656, 2.1592, 1.8129, 1.6838], device='cuda:1'), covar=tensor([0.0694, 0.0736, 0.0573, 0.0390], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0799, 0.0666, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 15:38:12,589 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18294.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:38:15,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18297.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:38:16,766 INFO [train.py:968] (1/2) Epoch 1, batch 18300, libri_loss[loss=0.4312, simple_loss=0.4821, pruned_loss=0.1901, over 29366.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4058, pruned_loss=0.1664, over 5694749.43 frames. ], libri_tot_loss[loss=0.4008, simple_loss=0.4362, pruned_loss=0.1827, over 5768011.84 frames. ], giga_tot_loss[loss=0.3607, simple_loss=0.3981, pruned_loss=0.1616, over 5688417.87 frames. ], batch size: 92, lr: 2.57e-02, grad_scale: 4.0
+2023-02-28 15:38:20,165 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18305.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:38:37,259 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18326.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:38:49,255 INFO [optim.py:369] (1/2) 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 15:38:50,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9182, 1.2012, 0.7819, 1.0259], device='cuda:1'), covar=tensor([0.1174, 0.0873, 0.1859, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0656, 0.0642, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 15:38:56,782 INFO [train.py:968] (1/2) Epoch 1, batch 18350, giga_loss[loss=0.3962, simple_loss=0.4327, pruned_loss=0.1798, over 28715.00 frames. ], tot_loss[loss=0.3831, simple_loss=0.4183, pruned_loss=0.1739, over 5704484.14 frames. ], libri_tot_loss[loss=0.4011, simple_loss=0.4367, pruned_loss=0.1828, over 5768620.62 frames. ], giga_tot_loss[loss=0.3755, simple_loss=0.4114, pruned_loss=0.1698, over 5697791.00 frames. ], batch size: 92, lr: 2.56e-02, grad_scale: 4.0
+2023-02-28 15:38:59,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18353.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:39:39,055 INFO [train.py:968] (1/2) Epoch 1, batch 18400, giga_loss[loss=0.4134, simple_loss=0.4541, pruned_loss=0.1863, over 28670.00 frames. ], tot_loss[loss=0.39, simple_loss=0.4257, pruned_loss=0.1771, over 5699616.85 frames. ], libri_tot_loss[loss=0.4022, simple_loss=0.4377, pruned_loss=0.1833, over 5771563.85 frames. ], giga_tot_loss[loss=0.3824, simple_loss=0.4187, pruned_loss=0.1731, over 5689191.50 frames. ], batch size: 242, lr: 2.56e-02, grad_scale: 8.0
+2023-02-28 15:40:00,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8447, 2.1387, 3.8649, 2.5830], device='cuda:1'), covar=tensor([0.1940, 0.1178, 0.0404, 0.0598], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0431, 0.0521, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 15:40:10,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7587, 1.7512, 1.1241, 1.2488], device='cuda:1'), covar=tensor([0.0860, 0.0977, 0.1266, 0.1398], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0653, 0.0639, 0.0598], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 15:40:15,832 INFO [optim.py:369] (1/2) 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 15:40:22,548 INFO [train.py:968] (1/2) Epoch 1, batch 18450, giga_loss[loss=0.3741, simple_loss=0.4274, pruned_loss=0.1604, over 28813.00 frames. ], tot_loss[loss=0.3904, simple_loss=0.4281, pruned_loss=0.1764, over 5700368.95 frames. ], libri_tot_loss[loss=0.4026, simple_loss=0.4381, pruned_loss=0.1835, over 5772821.70 frames. ], giga_tot_loss[loss=0.3839, simple_loss=0.4221, pruned_loss=0.1728, over 5689833.75 frames. ], batch size: 186, lr: 2.56e-02, grad_scale: 8.0
+2023-02-28 15:40:42,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18468.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:40:54,334 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 15:41:11,185 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18498.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:41:12,320 INFO [train.py:968] (1/2) Epoch 1, batch 18500, giga_loss[loss=0.3963, simple_loss=0.4421, pruned_loss=0.1753, over 28922.00 frames. ], tot_loss[loss=0.3888, simple_loss=0.4279, pruned_loss=0.1748, over 5694504.08 frames. ], libri_tot_loss[loss=0.4027, simple_loss=0.4382, pruned_loss=0.1836, over 5773291.00 frames. ], giga_tot_loss[loss=0.3834, simple_loss=0.4229, pruned_loss=0.1719, over 5685192.11 frames. ], batch size: 174, lr: 2.55e-02, grad_scale: 8.0
+2023-02-28 15:41:46,952 INFO [optim.py:369] (1/2) 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 15:41:54,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7334, 2.6953, 4.5198, 1.9725], device='cuda:1'), covar=tensor([0.0367, 0.0788, 0.0657, 0.1725], device='cuda:1'), in_proj_covar=tensor([0.0643, 0.0439, 0.0736, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0007, 0.0006], device='cuda:1')
+2023-02-28 15:41:56,727 INFO [train.py:968] (1/2) Epoch 1, batch 18550, giga_loss[loss=0.3867, simple_loss=0.4275, pruned_loss=0.173, over 29014.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.4288, pruned_loss=0.1753, over 5696308.30 frames. ], libri_tot_loss[loss=0.4027, simple_loss=0.4384, pruned_loss=0.1835, over 5775763.50 frames. ], giga_tot_loss[loss=0.3851, simple_loss=0.4245, pruned_loss=0.1729, over 5685489.94 frames. ], batch size: 136, lr: 2.55e-02, grad_scale: 8.0
+2023-02-28 15:42:11,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18567.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:42:14,259 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18568.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:42:14,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1478, 1.2604, 1.0784, 1.3690], device='cuda:1'), covar=tensor([0.1503, 0.1595, 0.1208, 0.1534], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0760, 0.0804, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:42:41,717 INFO [train.py:968] (1/2) Epoch 1, batch 18600, giga_loss[loss=0.3814, simple_loss=0.4246, pruned_loss=0.1692, over 28716.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4328, pruned_loss=0.1788, over 5702819.09 frames. ], libri_tot_loss[loss=0.4036, simple_loss=0.4393, pruned_loss=0.184, over 5777166.08 frames. ], giga_tot_loss[loss=0.3906, simple_loss=0.4284, pruned_loss=0.1763, over 5691484.32 frames. ], batch size: 60, lr: 2.55e-02, grad_scale: 8.0
+2023-02-28 15:42:52,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18611.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:42:55,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18614.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:16,675 INFO [optim.py:369] (1/2) 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 15:43:16,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18641.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:20,350 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18643.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:21,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18644.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:43:25,637 INFO [train.py:968] (1/2) Epoch 1, batch 18650, giga_loss[loss=0.4205, simple_loss=0.4558, pruned_loss=0.1926, over 28859.00 frames. ], tot_loss[loss=0.4, simple_loss=0.4365, pruned_loss=0.1817, over 5703077.41 frames. ], libri_tot_loss[loss=0.4045, simple_loss=0.4399, pruned_loss=0.1846, over 5779706.99 frames. ], giga_tot_loss[loss=0.3953, simple_loss=0.4324, pruned_loss=0.1791, over 5690123.38 frames. ], batch size: 199, lr: 2.54e-02, grad_scale: 8.0
+2023-02-28 15:43:45,769 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 18700, giga_loss[loss=0.4277, simple_loss=0.4614, pruned_loss=0.1969, over 28986.00 frames. ], tot_loss[loss=0.4016, simple_loss=0.4394, pruned_loss=0.182, over 5709865.32 frames. ], libri_tot_loss[loss=0.4051, simple_loss=0.4406, pruned_loss=0.1848, over 5780686.57 frames. ], giga_tot_loss[loss=0.3973, simple_loss=0.4354, pruned_loss=0.1796, over 5697054.55 frames. ], batch size: 106, lr: 2.54e-02, grad_scale: 8.0
+2023-02-28 15:44:30,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18728.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:44:42,472 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 18750, giga_loss[loss=0.4175, simple_loss=0.4534, pruned_loss=0.1908, over 28290.00 frames. ], tot_loss[loss=0.4025, simple_loss=0.4409, pruned_loss=0.1821, over 5699493.77 frames. ], libri_tot_loss[loss=0.406, simple_loss=0.4412, pruned_loss=0.1854, over 5770308.62 frames. ], giga_tot_loss[loss=0.3982, simple_loss=0.4372, pruned_loss=0.1796, over 5696975.01 frames. ], batch size: 368, lr: 2.54e-02, grad_scale: 4.0
+2023-02-28 15:45:30,599 INFO [train.py:968] (1/2) Epoch 1, batch 18800, giga_loss[loss=0.422, simple_loss=0.4626, pruned_loss=0.1907, over 29083.00 frames. ], tot_loss[loss=0.4034, simple_loss=0.4424, pruned_loss=0.1822, over 5699482.32 frames. ], libri_tot_loss[loss=0.4064, simple_loss=0.4417, pruned_loss=0.1856, over 5773724.84 frames. ], giga_tot_loss[loss=0.3995, simple_loss=0.439, pruned_loss=0.18, over 5693100.09 frames. ], batch size: 136, lr: 2.53e-02, grad_scale: 8.0
+2023-02-28 15:46:04,908 INFO [optim.py:369] (1/2) 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 15:46:13,747 INFO [train.py:968] (1/2) Epoch 1, batch 18850, giga_loss[loss=0.4217, simple_loss=0.4528, pruned_loss=0.1953, over 28219.00 frames. ], tot_loss[loss=0.3993, simple_loss=0.4406, pruned_loss=0.179, over 5697978.79 frames. ], libri_tot_loss[loss=0.4067, simple_loss=0.4419, pruned_loss=0.1857, over 5774531.22 frames. ], giga_tot_loss[loss=0.396, simple_loss=0.4377, pruned_loss=0.1772, over 5692065.32 frames. ], batch size: 368, lr: 2.53e-02, grad_scale: 8.0
+2023-02-28 15:46:31,557 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18871.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:46:33,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18874.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:46:40,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18883.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:46:53,420 INFO [train.py:968] (1/2) Epoch 1, batch 18900, giga_loss[loss=0.3508, simple_loss=0.4079, pruned_loss=0.1469, over 28779.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4383, pruned_loss=0.1761, over 5706798.43 frames. ], libri_tot_loss[loss=0.4084, simple_loss=0.4435, pruned_loss=0.1867, over 5774464.01 frames. ], giga_tot_loss[loss=0.3909, simple_loss=0.4346, pruned_loss=0.1736, over 5700791.97 frames. ], batch size: 284, lr: 2.53e-02, grad_scale: 4.0
+2023-02-28 15:46:54,175 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=18901.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:46:56,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18903.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:47:00,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1200, 1.2816, 1.2870, 1.0581], device='cuda:1'), covar=tensor([0.1175, 0.1263, 0.0913, 0.1665], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0507, 0.0417, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0010, 0.0008, 0.0011], device='cuda:1')
+2023-02-28 15:47:29,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18942.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:47:29,884 INFO [optim.py:369] (1/2) 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 15:47:30,244 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 18950, giga_loss[loss=0.4627, simple_loss=0.4766, pruned_loss=0.2244, over 28292.00 frames. ], tot_loss[loss=0.398, simple_loss=0.4398, pruned_loss=0.1782, over 5701043.83 frames. ], libri_tot_loss[loss=0.4091, simple_loss=0.4439, pruned_loss=0.1872, over 5773492.54 frames. ], giga_tot_loss[loss=0.3935, simple_loss=0.4363, pruned_loss=0.1754, over 5695386.12 frames. ], batch size: 369, lr: 2.53e-02, grad_scale: 4.0
+2023-02-28 15:47:49,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9182, 2.5778, 2.4061, 1.5097], device='cuda:1'), covar=tensor([0.1411, 0.0553, 0.0598, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0380, 0.0360, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0016], device='cuda:1')
+2023-02-28 15:48:20,913 INFO [train.py:968] (1/2) Epoch 1, batch 19000, giga_loss[loss=0.4181, simple_loss=0.4388, pruned_loss=0.1987, over 28791.00 frames. ], tot_loss[loss=0.4047, simple_loss=0.4422, pruned_loss=0.1836, over 5688974.24 frames. ], libri_tot_loss[loss=0.409, simple_loss=0.4438, pruned_loss=0.1871, over 5771799.51 frames. ], giga_tot_loss[loss=0.4012, simple_loss=0.4394, pruned_loss=0.1814, over 5684412.39 frames. ], batch size: 99, lr: 2.52e-02, grad_scale: 4.0
+2023-02-28 15:48:27,684 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 15:48:59,128 INFO [optim.py:369] (1/2) 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 15:49:04,846 INFO [train.py:968] (1/2) Epoch 1, batch 19050, giga_loss[loss=0.3948, simple_loss=0.4297, pruned_loss=0.1799, over 28850.00 frames. ], tot_loss[loss=0.4089, simple_loss=0.444, pruned_loss=0.1869, over 5690607.92 frames. ], libri_tot_loss[loss=0.4095, simple_loss=0.4444, pruned_loss=0.1873, over 5773249.56 frames. ], giga_tot_loss[loss=0.4055, simple_loss=0.4413, pruned_loss=0.1849, over 5683778.99 frames. ], batch size: 199, lr: 2.52e-02, grad_scale: 4.0
+2023-02-28 15:49:19,483 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19067.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:32,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19085.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:33,573 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19086.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:35,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19088.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:37,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19089.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:49:46,685 INFO [train.py:968] (1/2) Epoch 1, batch 19100, giga_loss[loss=0.4906, simple_loss=0.485, pruned_loss=0.2481, over 28622.00 frames. ], tot_loss[loss=0.4106, simple_loss=0.4442, pruned_loss=0.1884, over 5689652.48 frames. ], libri_tot_loss[loss=0.4102, simple_loss=0.4449, pruned_loss=0.1877, over 5765944.55 frames. ], giga_tot_loss[loss=0.4072, simple_loss=0.4415, pruned_loss=0.1865, over 5688332.68 frames. ], batch size: 71, lr: 2.52e-02, grad_scale: 4.0
+2023-02-28 15:50:01,000 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19117.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:50:01,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3024, 1.2013, 1.1371, 1.4943], device='cuda:1'), covar=tensor([0.1275, 0.1487, 0.0997, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.0840, 0.0779, 0.0804, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:50:01,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19118.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:50:23,404 INFO [optim.py:369] (1/2) 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 15:50:29,142 INFO [train.py:968] (1/2) Epoch 1, batch 19150, giga_loss[loss=0.3902, simple_loss=0.4275, pruned_loss=0.1765, over 28895.00 frames. ], tot_loss[loss=0.4092, simple_loss=0.4428, pruned_loss=0.1878, over 5699521.96 frames. ], libri_tot_loss[loss=0.4106, simple_loss=0.4454, pruned_loss=0.1879, over 5769798.21 frames. ], giga_tot_loss[loss=0.4061, simple_loss=0.4401, pruned_loss=0.1861, over 5692847.62 frames. ], batch size: 112, lr: 2.51e-02, grad_scale: 4.0
+2023-02-28 15:50:43,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2056, 1.1360, 1.0562, 1.1475], device='cuda:1'), covar=tensor([0.1302, 0.1575, 0.1191, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0845, 0.0786, 0.0811, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:51:15,227 INFO [train.py:968] (1/2) Epoch 1, batch 19200, giga_loss[loss=0.3931, simple_loss=0.431, pruned_loss=0.1776, over 28083.00 frames. ], tot_loss[loss=0.4092, simple_loss=0.4427, pruned_loss=0.1878, over 5683119.58 frames. ], libri_tot_loss[loss=0.4115, simple_loss=0.4461, pruned_loss=0.1884, over 5760361.86 frames. ], giga_tot_loss[loss=0.4059, simple_loss=0.4399, pruned_loss=0.1859, over 5684938.18 frames. ], batch size: 77, lr: 2.51e-02, grad_scale: 8.0
+2023-02-28 15:51:49,883 INFO [optim.py:369] (1/2) 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 15:51:55,176 INFO [train.py:968] (1/2) Epoch 1, batch 19250, giga_loss[loss=0.3786, simple_loss=0.4309, pruned_loss=0.1632, over 28676.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.441, pruned_loss=0.185, over 5680723.75 frames. ], libri_tot_loss[loss=0.4113, simple_loss=0.4462, pruned_loss=0.1882, over 5753016.79 frames. ], giga_tot_loss[loss=0.4028, simple_loss=0.4384, pruned_loss=0.1836, over 5685053.01 frames. ], batch size: 242, lr: 2.51e-02, grad_scale: 8.0
+2023-02-28 15:51:58,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2782, 1.3801, 1.1455, 1.3121], device='cuda:1'), covar=tensor([0.0652, 0.0756, 0.0949, 0.0712], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0912, 0.0676, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 15:51:58,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2703, 1.3628, 1.1356, 1.5861], device='cuda:1'), covar=tensor([0.1455, 0.1618, 0.1174, 0.1656], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0796, 0.0826, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:52:02,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19258.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:52:18,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19276.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:52:39,657 INFO [train.py:968] (1/2) Epoch 1, batch 19300, giga_loss[loss=0.3229, simple_loss=0.3834, pruned_loss=0.1312, over 28998.00 frames. ], tot_loss[loss=0.3999, simple_loss=0.437, pruned_loss=0.1814, over 5679310.23 frames. ], libri_tot_loss[loss=0.4119, simple_loss=0.4467, pruned_loss=0.1886, over 5748484.22 frames. ], giga_tot_loss[loss=0.3969, simple_loss=0.4343, pruned_loss=0.1798, over 5685362.07 frames. ], batch size: 164, lr: 2.50e-02, grad_scale: 8.0
+2023-02-28 15:53:01,654 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19323.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:53:16,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 15:53:17,917 INFO [optim.py:369] (1/2) 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 15:53:20,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2894, 1.2343, 1.1748, 1.4027], device='cuda:1'), covar=tensor([0.1315, 0.1345, 0.0988, 0.1438], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0777, 0.0817, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:53:23,154 INFO [train.py:968] (1/2) Epoch 1, batch 19350, giga_loss[loss=0.357, simple_loss=0.3979, pruned_loss=0.1581, over 28683.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4291, pruned_loss=0.1754, over 5678697.65 frames. ], libri_tot_loss[loss=0.4129, simple_loss=0.4473, pruned_loss=0.1892, over 5750488.87 frames. ], giga_tot_loss[loss=0.3863, simple_loss=0.426, pruned_loss=0.1733, over 5680316.88 frames. ], batch size: 307, lr: 2.50e-02, grad_scale: 4.0
+2023-02-28 15:53:36,508 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-02-28 15:54:07,747 INFO [train.py:968] (1/2) Epoch 1, batch 19400, giga_loss[loss=0.2925, simple_loss=0.3634, pruned_loss=0.1108, over 28888.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4215, pruned_loss=0.17, over 5685117.08 frames. ], libri_tot_loss[loss=0.4127, simple_loss=0.4472, pruned_loss=0.1891, over 5755865.95 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4184, pruned_loss=0.1678, over 5679319.09 frames. ], batch size: 136, lr: 2.50e-02, grad_scale: 4.0
+2023-02-28 15:54:09,809 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19401.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:13,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19404.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:27,946 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19419.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:54:29,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19422.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:54:38,684 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19433.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:49,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19442.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:54:50,935 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 19450, giga_loss[loss=0.3485, simple_loss=0.3907, pruned_loss=0.1532, over 28691.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.414, pruned_loss=0.1646, over 5680543.99 frames. ], libri_tot_loss[loss=0.4131, simple_loss=0.4477, pruned_loss=0.1893, over 5749895.33 frames. ], giga_tot_loss[loss=0.3672, simple_loss=0.4103, pruned_loss=0.1621, over 5679075.33 frames. ], batch size: 242, lr: 2.49e-02, grad_scale: 4.0
+2023-02-28 15:54:57,987 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19451.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 15:55:43,527 INFO [train.py:968] (1/2) Epoch 1, batch 19500, giga_loss[loss=0.3344, simple_loss=0.3898, pruned_loss=0.1395, over 29110.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.414, pruned_loss=0.1646, over 5687617.86 frames. ], libri_tot_loss[loss=0.4128, simple_loss=0.4475, pruned_loss=0.1891, over 5754746.11 frames. ], giga_tot_loss[loss=0.3672, simple_loss=0.4102, pruned_loss=0.1621, over 5680377.27 frames. ], batch size: 164, lr: 2.49e-02, grad_scale: 4.0
+2023-02-28 15:56:00,039 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19522.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 15:56:22,686 INFO [optim.py:369] (1/2) 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 15:56:26,851 INFO [train.py:968] (1/2) Epoch 1, batch 19550, giga_loss[loss=0.3359, simple_loss=0.3906, pruned_loss=0.1406, over 28936.00 frames. ], tot_loss[loss=0.3731, simple_loss=0.4154, pruned_loss=0.1654, over 5687968.58 frames. ], libri_tot_loss[loss=0.4134, simple_loss=0.4479, pruned_loss=0.1894, over 5740413.25 frames. ], giga_tot_loss[loss=0.3681, simple_loss=0.4112, pruned_loss=0.1625, over 5692571.20 frames. ], batch size: 213, lr: 2.49e-02, grad_scale: 4.0
+2023-02-28 15:56:58,855 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19585.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:57:01,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19588.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:57:10,944 INFO [train.py:968] (1/2) Epoch 1, batch 19600, giga_loss[loss=0.3436, simple_loss=0.3876, pruned_loss=0.1498, over 28540.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4126, pruned_loss=0.1636, over 5691906.33 frames. ], libri_tot_loss[loss=0.4139, simple_loss=0.4483, pruned_loss=0.1897, over 5741886.10 frames. ], giga_tot_loss[loss=0.3652, simple_loss=0.4087, pruned_loss=0.1608, over 5693947.07 frames. ], batch size: 85, lr: 2.49e-02, grad_scale: 8.0
+2023-02-28 15:57:26,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19617.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:57:51,793 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 19650, giga_loss[loss=0.3347, simple_loss=0.3802, pruned_loss=0.1446, over 28723.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4079, pruned_loss=0.1605, over 5705646.31 frames. ], libri_tot_loss[loss=0.4144, simple_loss=0.4488, pruned_loss=0.19, over 5744213.10 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4039, pruned_loss=0.1577, over 5704626.36 frames. ], batch size: 92, lr: 2.48e-02, grad_scale: 4.0
+2023-02-28 15:58:34,560 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 19700, giga_loss[loss=0.3042, simple_loss=0.3597, pruned_loss=0.1243, over 28511.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4049, pruned_loss=0.1584, over 5699353.30 frames. ], libri_tot_loss[loss=0.4166, simple_loss=0.4507, pruned_loss=0.1913, over 5729534.04 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.3991, pruned_loss=0.1543, over 5711331.97 frames. ], batch size: 71, lr: 2.48e-02, grad_scale: 4.0
+2023-02-28 15:59:04,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8872, 1.4901, 1.5813, 1.4075], device='cuda:1'), covar=tensor([0.0691, 0.1683, 0.0991, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0906, 0.0661, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 15:59:11,393 INFO [optim.py:369] (1/2) 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 15:59:13,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6591, 1.3466, 1.4728, 1.2428], device='cuda:1'), covar=tensor([0.0631, 0.1272, 0.0815, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0908, 0.0668, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 15:59:14,851 INFO [train.py:968] (1/2) Epoch 1, batch 19750, giga_loss[loss=0.312, simple_loss=0.3679, pruned_loss=0.128, over 29050.00 frames. ], tot_loss[loss=0.3596, simple_loss=0.4036, pruned_loss=0.1578, over 5702844.31 frames. ], libri_tot_loss[loss=0.4178, simple_loss=0.4518, pruned_loss=0.1919, over 5729412.68 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.3965, pruned_loss=0.153, over 5712015.34 frames. ], batch size: 155, lr: 2.48e-02, grad_scale: 4.0
+2023-02-28 15:59:46,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5180, 1.5024, 1.2047, 1.4302], device='cuda:1'), covar=tensor([0.1396, 0.1626, 0.1207, 0.1875], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0779, 0.0805, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 15:59:46,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0611, 2.5320, 3.7421, 1.8265], device='cuda:1'), covar=tensor([0.0503, 0.0713, 0.0791, 0.1779], device='cuda:1'), in_proj_covar=tensor([0.0679, 0.0436, 0.0747, 0.0512], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 15:59:47,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=19787.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 15:59:56,350 INFO [train.py:968] (1/2) Epoch 1, batch 19800, giga_loss[loss=0.2929, simple_loss=0.3485, pruned_loss=0.1186, over 28810.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4008, pruned_loss=0.1562, over 5713059.84 frames. ], libri_tot_loss[loss=0.4192, simple_loss=0.4531, pruned_loss=0.1927, over 5732600.34 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.3931, pruned_loss=0.1509, over 5716939.47 frames. ], batch size: 99, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 16:00:31,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19841.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:00:33,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19844.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:00:34,069 INFO [optim.py:369] (1/2) 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 16:00:38,262 INFO [train.py:968] (1/2) Epoch 1, batch 19850, giga_loss[loss=0.2985, simple_loss=0.36, pruned_loss=0.1185, over 28799.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.3976, pruned_loss=0.154, over 5704369.29 frames. ], libri_tot_loss[loss=0.4198, simple_loss=0.4535, pruned_loss=0.193, over 5725365.64 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3903, pruned_loss=0.1491, over 5714359.27 frames. ], batch size: 199, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 16:00:58,122 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19873.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:01:17,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19897.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 16:01:18,878 INFO [train.py:968] (1/2) Epoch 1, batch 19900, giga_loss[loss=0.3178, simple_loss=0.3753, pruned_loss=0.1301, over 28978.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.3955, pruned_loss=0.153, over 5707482.64 frames. ], libri_tot_loss[loss=0.421, simple_loss=0.4546, pruned_loss=0.1937, over 5727076.57 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3879, pruned_loss=0.148, over 5713594.26 frames. ], batch size: 136, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 16:01:45,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-02-28 16:01:56,894 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 19950, giga_loss[loss=0.3268, simple_loss=0.3718, pruned_loss=0.1409, over 28798.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.3932, pruned_loss=0.1513, over 5719268.91 frames. ], libri_tot_loss[loss=0.4221, simple_loss=0.4555, pruned_loss=0.1943, over 5731604.39 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3848, pruned_loss=0.1457, over 5719623.71 frames. ], batch size: 199, lr: 2.47e-02, grad_scale: 4.0
+2023-02-28 16:02:40,062 INFO [train.py:968] (1/2) Epoch 1, batch 20000, giga_loss[loss=0.3381, simple_loss=0.3847, pruned_loss=0.1457, over 28299.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3908, pruned_loss=0.1495, over 5723898.59 frames. ], libri_tot_loss[loss=0.4226, simple_loss=0.4559, pruned_loss=0.1947, over 5734775.71 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3828, pruned_loss=0.144, over 5721182.14 frames. ], batch size: 71, lr: 2.46e-02, grad_scale: 8.0
+2023-02-28 16:02:48,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20011.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:03:11,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20040.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 16:03:14,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20043.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 16:03:14,918 INFO [optim.py:369] (1/2) 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 16:03:18,504 INFO [train.py:968] (1/2) Epoch 1, batch 20050, giga_loss[loss=0.3385, simple_loss=0.3919, pruned_loss=0.1426, over 28958.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3918, pruned_loss=0.1504, over 5735541.40 frames. ], libri_tot_loss[loss=0.4238, simple_loss=0.4567, pruned_loss=0.1954, over 5739405.82 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3831, pruned_loss=0.1443, over 5729116.85 frames. ], batch size: 174, lr: 2.46e-02, grad_scale: 8.0
+2023-02-28 16:03:22,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4327, 1.3034, 1.2969, 0.8028], device='cuda:1'), covar=tensor([0.0435, 0.0344, 0.0334, 0.0458], device='cuda:1'), in_proj_covar=tensor([0.0691, 0.0546, 0.0563, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:03:36,880 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20072.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 16:03:48,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9626, 2.2760, 4.7918, 3.1818], device='cuda:1'), covar=tensor([0.1435, 0.0929, 0.0179, 0.0452], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0430, 0.0548, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 16:04:03,453 INFO [train.py:968] (1/2) Epoch 1, batch 20100, giga_loss[loss=0.4184, simple_loss=0.4491, pruned_loss=0.1939, over 28726.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.3993, pruned_loss=0.1565, over 5719932.64 frames. ], libri_tot_loss[loss=0.4251, simple_loss=0.458, pruned_loss=0.1962, over 5733210.08 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.39, pruned_loss=0.15, over 5721066.77 frames. ], batch size: 262, lr: 2.46e-02, grad_scale: 8.0
+2023-02-28 16:04:47,919 INFO [optim.py:369] (1/2) 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 16:04:52,599 INFO [train.py:968] (1/2) Epoch 1, batch 20150, giga_loss[loss=0.4129, simple_loss=0.4509, pruned_loss=0.1874, over 29007.00 frames. ], tot_loss[loss=0.3666, simple_loss=0.4077, pruned_loss=0.1628, over 5709863.34 frames. ], libri_tot_loss[loss=0.4248, simple_loss=0.4576, pruned_loss=0.196, over 5728145.77 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.3988, pruned_loss=0.1566, over 5715159.39 frames. ], batch size: 164, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:05:03,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20162.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:05:31,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6278, 1.4561, 1.3758, 1.3837], device='cuda:1'), covar=tensor([0.0658, 0.1097, 0.1007, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0898, 0.0681, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 16:05:37,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8618, 2.4770, 3.5961, 1.6589], device='cuda:1'), covar=tensor([0.0573, 0.0778, 0.0930, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.0674, 0.0439, 0.0760, 0.0519], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0005, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 16:05:43,978 INFO [train.py:968] (1/2) Epoch 1, batch 20200, giga_loss[loss=0.4019, simple_loss=0.4423, pruned_loss=0.1808, over 28809.00 frames. ], tot_loss[loss=0.3822, simple_loss=0.4194, pruned_loss=0.1726, over 5694501.17 frames. ], libri_tot_loss[loss=0.4247, simple_loss=0.4577, pruned_loss=0.1959, over 5732432.82 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4111, pruned_loss=0.1669, over 5694480.65 frames. ], batch size: 199, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:06:24,602 INFO [optim.py:369] (1/2) 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 16:06:29,859 INFO [train.py:968] (1/2) Epoch 1, batch 20250, giga_loss[loss=0.3875, simple_loss=0.436, pruned_loss=0.1695, over 28784.00 frames. ], tot_loss[loss=0.3886, simple_loss=0.4254, pruned_loss=0.1759, over 5694763.24 frames. ], libri_tot_loss[loss=0.4245, simple_loss=0.4576, pruned_loss=0.1958, over 5735835.91 frames. ], giga_tot_loss[loss=0.3804, simple_loss=0.4184, pruned_loss=0.1712, over 5691086.52 frames. ], batch size: 199, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:07:20,211 INFO [train.py:968] (1/2) Epoch 1, batch 20300, giga_loss[loss=0.4091, simple_loss=0.4498, pruned_loss=0.1842, over 28934.00 frames. ], tot_loss[loss=0.3939, simple_loss=0.4307, pruned_loss=0.1786, over 5681599.90 frames. ], libri_tot_loss[loss=0.4246, simple_loss=0.4576, pruned_loss=0.1958, over 5738385.40 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4246, pruned_loss=0.1746, over 5675829.78 frames. ], batch size: 199, lr: 2.45e-02, grad_scale: 8.0
+2023-02-28 16:07:25,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20305.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:07:29,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20308.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:07:47,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5703, 1.4131, 1.3916, 0.9231], device='cuda:1'), covar=tensor([0.0410, 0.0322, 0.0273, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0680, 0.0524, 0.0572, 0.0607], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:07:57,776 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20337.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:08:05,632 INFO [optim.py:369] (1/2) 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 16:08:09,640 INFO [train.py:968] (1/2) Epoch 1, batch 20350, giga_loss[loss=0.415, simple_loss=0.4506, pruned_loss=0.1897, over 28916.00 frames. ], tot_loss[loss=0.4018, simple_loss=0.437, pruned_loss=0.1833, over 5667751.96 frames. ], libri_tot_loss[loss=0.425, simple_loss=0.458, pruned_loss=0.196, over 5730863.14 frames. ], giga_tot_loss[loss=0.3955, simple_loss=0.4315, pruned_loss=0.1797, over 5669895.08 frames. ], batch size: 106, lr: 2.44e-02, grad_scale: 4.0
+2023-02-28 16:08:21,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4556, 1.7823, 1.5508, 1.3581], device='cuda:1'), covar=tensor([0.1341, 0.0633, 0.0683, 0.1873], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0365, 0.0353, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0017], device='cuda:1')
+2023-02-28 16:08:41,912 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20386.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:08:53,524 INFO [train.py:968] (1/2) Epoch 1, batch 20400, giga_loss[loss=0.323, simple_loss=0.3776, pruned_loss=0.1342, over 28965.00 frames. ], tot_loss[loss=0.4042, simple_loss=0.4384, pruned_loss=0.185, over 5662700.90 frames. ], libri_tot_loss[loss=0.4252, simple_loss=0.4579, pruned_loss=0.1962, over 5725701.43 frames. ], giga_tot_loss[loss=0.3986, simple_loss=0.4337, pruned_loss=0.1817, over 5667907.52 frames. ], batch size: 227, lr: 2.44e-02, grad_scale: 4.0
+2023-02-28 16:09:04,338 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20413.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:09:34,366 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 20450, libri_loss[loss=0.3425, simple_loss=0.3875, pruned_loss=0.1487, over 29393.00 frames. ], tot_loss[loss=0.3968, simple_loss=0.4325, pruned_loss=0.1806, over 5677194.32 frames. ], libri_tot_loss[loss=0.4248, simple_loss=0.4573, pruned_loss=0.1962, over 5731451.99 frames. ], giga_tot_loss[loss=0.3915, simple_loss=0.4283, pruned_loss=0.1774, over 5673468.34 frames. ], batch size: 67, lr: 2.44e-02, grad_scale: 2.0
+2023-02-28 16:10:03,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20482.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:10:08,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4660, 1.8470, 1.6246, 1.5065], device='cuda:1'), covar=tensor([0.0733, 0.0851, 0.0664, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0790, 0.0656, 0.0588], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0005, 0.0004], device='cuda:1')
+2023-02-28 16:10:19,498 INFO [train.py:968] (1/2) Epoch 1, batch 20500, giga_loss[loss=0.3868, simple_loss=0.4292, pruned_loss=0.1722, over 28830.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.428, pruned_loss=0.1757, over 5691208.51 frames. ], libri_tot_loss[loss=0.424, simple_loss=0.4568, pruned_loss=0.1957, over 5734078.03 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4246, pruned_loss=0.1732, over 5684973.15 frames. ], batch size: 145, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:10:25,118 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20505.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:10:44,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20529.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:10:46,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20532.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:10:52,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-02-28 16:10:54,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9245, 1.6653, 1.2418, 1.4763], device='cuda:1'), covar=tensor([0.0694, 0.1005, 0.1177, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0646, 0.0628, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 16:11:01,104 INFO [optim.py:369] (1/2) 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 16:11:03,799 INFO [train.py:968] (1/2) Epoch 1, batch 20550, giga_loss[loss=0.4197, simple_loss=0.4554, pruned_loss=0.192, over 27941.00 frames. ], tot_loss[loss=0.3895, simple_loss=0.4282, pruned_loss=0.1754, over 5691584.35 frames. ], libri_tot_loss[loss=0.4237, simple_loss=0.4565, pruned_loss=0.1955, over 5736648.16 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4253, pruned_loss=0.1734, over 5683946.27 frames. ], batch size: 412, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:11:13,948 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20561.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:11:42,699 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20597.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:11:44,745 INFO [train.py:968] (1/2) Epoch 1, batch 20600, giga_loss[loss=0.3921, simple_loss=0.4361, pruned_loss=0.174, over 28913.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4331, pruned_loss=0.1786, over 5696141.92 frames. ], libri_tot_loss[loss=0.4244, simple_loss=0.4572, pruned_loss=0.1958, over 5739077.96 frames. ], giga_tot_loss[loss=0.3908, simple_loss=0.4293, pruned_loss=0.1761, over 5686264.99 frames. ], batch size: 119, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:12:28,293 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 20650, giga_loss[loss=0.4138, simple_loss=0.4481, pruned_loss=0.1898, over 28521.00 frames. ], tot_loss[loss=0.398, simple_loss=0.4356, pruned_loss=0.1802, over 5691310.93 frames. ], libri_tot_loss[loss=0.4251, simple_loss=0.4579, pruned_loss=0.1962, over 5732036.10 frames. ], giga_tot_loss[loss=0.3934, simple_loss=0.4316, pruned_loss=0.1776, over 5688610.45 frames. ], batch size: 85, lr: 2.43e-02, grad_scale: 2.0
+2023-02-28 16:13:12,481 INFO [train.py:968] (1/2) Epoch 1, batch 20700, giga_loss[loss=0.4073, simple_loss=0.4256, pruned_loss=0.1945, over 23838.00 frames. ], tot_loss[loss=0.4001, simple_loss=0.4375, pruned_loss=0.1814, over 5690652.50 frames. ], libri_tot_loss[loss=0.4247, simple_loss=0.4576, pruned_loss=0.1959, over 5725436.40 frames. ], giga_tot_loss[loss=0.396, simple_loss=0.4338, pruned_loss=0.1791, over 5693029.51 frames. ], batch size: 705, lr: 2.42e-02, grad_scale: 2.0
+2023-02-28 16:13:38,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6108, 1.8356, 3.9529, 2.7735], device='cuda:1'), covar=tensor([0.1467, 0.0985, 0.0243, 0.0473], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0425, 0.0543, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 16:13:57,387 INFO [optim.py:369] (1/2) 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 16:13:58,661 INFO [train.py:968] (1/2) Epoch 1, batch 20750, giga_loss[loss=0.392, simple_loss=0.4312, pruned_loss=0.1764, over 28761.00 frames. ], tot_loss[loss=0.4031, simple_loss=0.4392, pruned_loss=0.1835, over 5683697.37 frames. ], libri_tot_loss[loss=0.4253, simple_loss=0.4581, pruned_loss=0.1962, over 5730276.63 frames. ], giga_tot_loss[loss=0.3986, simple_loss=0.4353, pruned_loss=0.181, over 5680190.66 frames. ], batch size: 262, lr: 2.42e-02, grad_scale: 2.0
+2023-02-28 16:14:25,639 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20781.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:14:25,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7720, 2.1066, 1.8127, 1.6616], device='cuda:1'), covar=tensor([0.0756, 0.0775, 0.0578, 0.0393], device='cuda:1'), in_proj_covar=tensor([0.0753, 0.0817, 0.0663, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0004], device='cuda:1')
+2023-02-28 16:14:33,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20788.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:14:41,003 INFO [train.py:968] (1/2) Epoch 1, batch 20800, giga_loss[loss=0.3765, simple_loss=0.4224, pruned_loss=0.1652, over 28702.00 frames. ], tot_loss[loss=0.4042, simple_loss=0.4399, pruned_loss=0.1842, over 5691459.93 frames. ], libri_tot_loss[loss=0.4253, simple_loss=0.4583, pruned_loss=0.1962, over 5735666.33 frames. ], giga_tot_loss[loss=0.3997, simple_loss=0.4359, pruned_loss=0.1818, over 5682189.36 frames. ], batch size: 262, lr: 2.42e-02, grad_scale: 4.0
+2023-02-28 16:14:42,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6238, 1.9062, 1.7824, 0.6538], device='cuda:1'), covar=tensor([0.0772, 0.0669, 0.0754, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0856, 0.0869, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 16:14:51,056 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:15:21,290 INFO [train.py:968] (1/2) Epoch 1, batch 20850, libri_loss[loss=0.441, simple_loss=0.4841, pruned_loss=0.1989, over 26089.00 frames. ], tot_loss[loss=0.403, simple_loss=0.4397, pruned_loss=0.1832, over 5698499.83 frames. ], libri_tot_loss[loss=0.4256, simple_loss=0.4584, pruned_loss=0.1964, over 5735560.46 frames. ], giga_tot_loss[loss=0.3987, simple_loss=0.4359, pruned_loss=0.1807, over 5690439.78 frames. ], batch size: 137, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:15:26,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20857.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:15:45,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20880.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:16:03,848 INFO [train.py:968] (1/2) Epoch 1, batch 20900, giga_loss[loss=0.3663, simple_loss=0.4145, pruned_loss=0.159, over 28661.00 frames. ], tot_loss[loss=0.4007, simple_loss=0.4389, pruned_loss=0.1813, over 5696462.52 frames. ], libri_tot_loss[loss=0.4259, simple_loss=0.4587, pruned_loss=0.1965, over 5734620.61 frames. ], giga_tot_loss[loss=0.3967, simple_loss=0.4354, pruned_loss=0.179, over 5690725.60 frames. ], batch size: 92, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:16:29,951 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20931.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:16:32,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20934.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:16:43,502 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 20950, giga_loss[loss=0.3737, simple_loss=0.4287, pruned_loss=0.1594, over 28932.00 frames. ], tot_loss[loss=0.3994, simple_loss=0.4394, pruned_loss=0.1797, over 5697464.04 frames. ], libri_tot_loss[loss=0.4256, simple_loss=0.4585, pruned_loss=0.1964, over 5731459.65 frames. ], giga_tot_loss[loss=0.3955, simple_loss=0.4361, pruned_loss=0.1775, over 5693858.95 frames. ], batch size: 174, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:16:56,616 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:04,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20972.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:12,946 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=20981.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:27,493 INFO [train.py:968] (1/2) Epoch 1, batch 21000, giga_loss[loss=0.3809, simple_loss=0.4234, pruned_loss=0.1692, over 28482.00 frames. ], tot_loss[loss=0.3961, simple_loss=0.4371, pruned_loss=0.1775, over 5695059.84 frames. ], libri_tot_loss[loss=0.4258, simple_loss=0.4586, pruned_loss=0.1965, over 5731285.11 frames. ], giga_tot_loss[loss=0.3927, simple_loss=0.4343, pruned_loss=0.1755, over 5692150.02 frames. ], batch size: 71, lr: 2.41e-02, grad_scale: 4.0
+2023-02-28 16:17:27,494 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 16:17:37,401 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19327MB
+2023-02-28 16:17:37,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21000.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:39,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21003.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:54,691 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21023.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:17:57,868 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21026.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:18:02,698 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:18:16,940 INFO [train.py:968] (1/2) Epoch 1, batch 21050, giga_loss[loss=0.3889, simple_loss=0.4246, pruned_loss=0.1766, over 28906.00 frames. ], tot_loss[loss=0.3933, simple_loss=0.4345, pruned_loss=0.176, over 5710016.80 frames. ], libri_tot_loss[loss=0.4257, simple_loss=0.4584, pruned_loss=0.1965, over 5735392.42 frames. ], giga_tot_loss[loss=0.3896, simple_loss=0.4316, pruned_loss=0.1738, over 5702906.81 frames. ], batch size: 285, lr: 2.40e-02, grad_scale: 4.0
+2023-02-28 16:18:21,179 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21055.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:18:22,765 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 16:18:55,972 INFO [train.py:968] (1/2) Epoch 1, batch 21100, giga_loss[loss=0.3678, simple_loss=0.4121, pruned_loss=0.1617, over 28435.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.4317, pruned_loss=0.1739, over 5714973.13 frames. ], libri_tot_loss[loss=0.4254, simple_loss=0.4581, pruned_loss=0.1963, over 5739758.33 frames. ], giga_tot_loss[loss=0.3862, simple_loss=0.4291, pruned_loss=0.1717, over 5704895.63 frames. ], batch size: 65, lr: 2.40e-02, grad_scale: 4.0
+2023-02-28 16:19:09,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21115.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:19:12,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:19:32,837 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:19:33,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8963, 1.6078, 1.3309, 1.3289], device='cuda:1'), covar=tensor([0.0666, 0.0828, 0.1114, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0637, 0.0640, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 16:19:34,988 INFO [train.py:968] (1/2) Epoch 1, batch 21150, giga_loss[loss=0.3952, simple_loss=0.4055, pruned_loss=0.1925, over 23705.00 frames. ], tot_loss[loss=0.3909, simple_loss=0.432, pruned_loss=0.1749, over 5719491.41 frames. ], libri_tot_loss[loss=0.4252, simple_loss=0.458, pruned_loss=0.1963, over 5746533.39 frames. ], giga_tot_loss[loss=0.3867, simple_loss=0.4289, pruned_loss=0.1722, over 5704227.35 frames. ], batch size: 705, lr: 2.40e-02, grad_scale: 4.0
+2023-02-28 16:19:40,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21156.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:20:06,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21186.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:20:18,608 INFO [train.py:968] (1/2) Epoch 1, batch 21200, libri_loss[loss=0.4167, simple_loss=0.4585, pruned_loss=0.1874, over 29109.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.433, pruned_loss=0.1761, over 5701307.70 frames. ], libri_tot_loss[loss=0.4252, simple_loss=0.4579, pruned_loss=0.1962, over 5732425.27 frames. ], giga_tot_loss[loss=0.3882, simple_loss=0.4297, pruned_loss=0.1733, over 5701167.18 frames. ], batch size: 101, lr: 2.40e-02, grad_scale: 8.0
+2023-02-28 16:20:59,358 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 21250, giga_loss[loss=0.3587, simple_loss=0.4181, pruned_loss=0.1497, over 28899.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4322, pruned_loss=0.1742, over 5714227.14 frames. ], libri_tot_loss[loss=0.425, simple_loss=0.4578, pruned_loss=0.1961, over 5734504.74 frames. ], giga_tot_loss[loss=0.3866, simple_loss=0.4294, pruned_loss=0.1719, over 5711951.80 frames. ], batch size: 145, lr: 2.39e-02, grad_scale: 8.0
+2023-02-28 16:21:41,601 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 21300, giga_loss[loss=0.364, simple_loss=0.4229, pruned_loss=0.1526, over 28705.00 frames. ], tot_loss[loss=0.3914, simple_loss=0.4331, pruned_loss=0.1748, over 5706279.49 frames. ], libri_tot_loss[loss=0.4258, simple_loss=0.4581, pruned_loss=0.1967, over 5738293.32 frames. ], giga_tot_loss[loss=0.3866, simple_loss=0.4298, pruned_loss=0.1717, over 5700359.36 frames. ], batch size: 60, lr: 2.39e-02, grad_scale: 4.0
+2023-02-28 16:21:43,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21302.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:21:44,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5743, 1.6468, 1.7142, 0.3769], device='cuda:1'), covar=tensor([0.0725, 0.0648, 0.0651, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.0846, 0.0834, 0.0861, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 16:22:06,143 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21329.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:07,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21331.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:08,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21332.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:22,747 INFO [optim.py:369] (1/2) 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 16:22:23,954 INFO [train.py:968] (1/2) Epoch 1, batch 21350, giga_loss[loss=0.322, simple_loss=0.3788, pruned_loss=0.1326, over 28919.00 frames. ], tot_loss[loss=0.3855, simple_loss=0.4296, pruned_loss=0.1707, over 5718155.81 frames. ], libri_tot_loss[loss=0.4257, simple_loss=0.458, pruned_loss=0.1967, over 5740100.26 frames. ], giga_tot_loss[loss=0.3815, simple_loss=0.4269, pruned_loss=0.168, over 5711625.80 frames. ], batch size: 112, lr: 2.39e-02, grad_scale: 4.0
+2023-02-28 16:22:29,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21356.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:33,432 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21361.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:22:40,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6621, 1.6022, 1.4575, 1.8661], device='cuda:1'), covar=tensor([0.1400, 0.1566, 0.1191, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0765, 0.0818, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 16:22:57,076 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-02-28 16:23:02,411 INFO [train.py:968] (1/2) Epoch 1, batch 21400, giga_loss[loss=0.3359, simple_loss=0.3953, pruned_loss=0.1382, over 28571.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4304, pruned_loss=0.1716, over 5720721.36 frames. ], libri_tot_loss[loss=0.4255, simple_loss=0.4577, pruned_loss=0.1967, over 5737716.07 frames. ], giga_tot_loss[loss=0.3818, simple_loss=0.427, pruned_loss=0.1683, over 5716731.22 frames. ], batch size: 78, lr: 2.38e-02, grad_scale: 4.0
+2023-02-28 16:23:15,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2712, 1.8652, 1.7817, 1.1847], device='cuda:1'), covar=tensor([0.0359, 0.0326, 0.0235, 0.0390], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0551, 0.0549, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:23:19,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1583, 1.8397, 1.8172, 1.0247], device='cuda:1'), covar=tensor([0.0420, 0.0311, 0.0233, 0.0434], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0552, 0.0548, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:23:38,778 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 1, batch 21450, giga_loss[loss=0.389, simple_loss=0.4349, pruned_loss=0.1715, over 28783.00 frames. ], tot_loss[loss=0.3856, simple_loss=0.4286, pruned_loss=0.1713, over 5715592.93 frames. ], libri_tot_loss[loss=0.4259, simple_loss=0.4578, pruned_loss=0.197, over 5729489.39 frames. ], giga_tot_loss[loss=0.3806, simple_loss=0.4254, pruned_loss=0.1679, over 5719271.46 frames. ], batch size: 284, lr: 2.38e-02, grad_scale: 4.0
+2023-02-28 16:23:46,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7624, 1.6074, 1.4650, 0.8319], device='cuda:1'), covar=tensor([0.0466, 0.0341, 0.0287, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0714, 0.0546, 0.0545, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:23:59,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8303, 2.0012, 4.0502, 2.9032], device='cuda:1'), covar=tensor([0.1357, 0.0920, 0.0230, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0420, 0.0523, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 16:24:24,195 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21499.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:24:24,542 INFO [train.py:968] (1/2) Epoch 1, batch 21500, giga_loss[loss=0.3409, simple_loss=0.3947, pruned_loss=0.1435, over 28993.00 frames. ], tot_loss[loss=0.3813, simple_loss=0.4248, pruned_loss=0.1689, over 5715228.55 frames. ], libri_tot_loss[loss=0.4265, simple_loss=0.4581, pruned_loss=0.1975, over 5733346.45 frames. ], giga_tot_loss[loss=0.3758, simple_loss=0.4213, pruned_loss=0.1652, over 5714385.64 frames. ], batch size: 145, lr: 2.38e-02, grad_scale: 2.0
+2023-02-28 16:24:26,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21502.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:24:34,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21511.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:24:49,070 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21531.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:25:05,463 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 21550, giga_loss[loss=0.3339, simple_loss=0.3951, pruned_loss=0.1363, over 28238.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4246, pruned_loss=0.1693, over 5711849.15 frames. ], libri_tot_loss[loss=0.427, simple_loss=0.4582, pruned_loss=0.1979, over 5725917.19 frames. ], giga_tot_loss[loss=0.3754, simple_loss=0.4207, pruned_loss=0.1651, over 5717367.23 frames. ], batch size: 77, lr: 2.38e-02, grad_scale: 2.0
+2023-02-28 16:25:06,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9413, 1.6842, 1.6536, 1.6843], device='cuda:1'), covar=tensor([0.0805, 0.1505, 0.1067, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0910, 0.0698, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 16:25:48,523 INFO [train.py:968] (1/2) Epoch 1, batch 21600, giga_loss[loss=0.3503, simple_loss=0.3957, pruned_loss=0.1524, over 28707.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4262, pruned_loss=0.172, over 5711335.83 frames. ], libri_tot_loss[loss=0.4274, simple_loss=0.4584, pruned_loss=0.1982, over 5727709.54 frames. ], giga_tot_loss[loss=0.3794, simple_loss=0.4226, pruned_loss=0.1681, over 5714040.85 frames. ], batch size: 119, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:26:05,110 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-02-28 16:26:30,045 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 21650, giga_loss[loss=0.354, simple_loss=0.3979, pruned_loss=0.155, over 28958.00 frames. ], tot_loss[loss=0.3842, simple_loss=0.4242, pruned_loss=0.1721, over 5716057.61 frames. ], libri_tot_loss[loss=0.4279, simple_loss=0.4586, pruned_loss=0.1986, over 5730405.94 frames. ], giga_tot_loss[loss=0.3787, simple_loss=0.4208, pruned_loss=0.1683, over 5715552.08 frames. ], batch size: 227, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:26:31,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7323, 1.8051, 1.2354, 1.3665], device='cuda:1'), covar=tensor([0.0952, 0.1312, 0.1090, 0.0692], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0846, 0.0695, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0005], device='cuda:1')
+2023-02-28 16:27:10,435 INFO [train.py:968] (1/2) Epoch 1, batch 21700, giga_loss[loss=0.348, simple_loss=0.3953, pruned_loss=0.1504, over 28736.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4227, pruned_loss=0.1719, over 5710324.83 frames. ], libri_tot_loss[loss=0.4284, simple_loss=0.4588, pruned_loss=0.1991, over 5725994.81 frames. ], giga_tot_loss[loss=0.3769, simple_loss=0.4186, pruned_loss=0.1676, over 5713940.48 frames. ], batch size: 284, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:27:14,701 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21705.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:27:28,284 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21721.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:27:52,653 INFO [optim.py:369] (1/2) 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 16:27:52,665 INFO [train.py:968] (1/2) Epoch 1, batch 21750, giga_loss[loss=0.3226, simple_loss=0.3777, pruned_loss=0.1338, over 29059.00 frames. ], tot_loss[loss=0.379, simple_loss=0.4183, pruned_loss=0.1698, over 5710613.03 frames. ], libri_tot_loss[loss=0.428, simple_loss=0.4582, pruned_loss=0.1989, over 5728263.79 frames. ], giga_tot_loss[loss=0.3734, simple_loss=0.4149, pruned_loss=0.166, over 5710936.58 frames. ], batch size: 128, lr: 2.37e-02, grad_scale: 4.0
+2023-02-28 16:28:20,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9931, 1.1144, 0.9463, 1.2183], device='cuda:1'), covar=tensor([0.1836, 0.2098, 0.1769, 0.2106], device='cuda:1'), in_proj_covar=tensor([0.0861, 0.0776, 0.0810, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 16:28:32,443 INFO [train.py:968] (1/2) Epoch 1, batch 21800, giga_loss[loss=0.3711, simple_loss=0.4158, pruned_loss=0.1632, over 27949.00 frames. ], tot_loss[loss=0.377, simple_loss=0.4166, pruned_loss=0.1687, over 5716751.78 frames. ], libri_tot_loss[loss=0.4287, simple_loss=0.4586, pruned_loss=0.1994, over 5734207.06 frames. ], giga_tot_loss[loss=0.3702, simple_loss=0.4121, pruned_loss=0.1642, over 5710960.47 frames. ], batch size: 412, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:28:46,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7372, 1.2080, 1.3726, 0.9274], device='cuda:1'), covar=tensor([0.0391, 0.0356, 0.0253, 0.0389], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0536, 0.0547, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:28:48,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3279, 1.6458, 1.5308, 0.3791], device='cuda:1'), covar=tensor([0.0763, 0.0660, 0.0897, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0879, 0.0902, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 16:28:50,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:29:16,069 INFO [optim.py:369] (1/2) 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 16:29:16,081 INFO [train.py:968] (1/2) Epoch 1, batch 21850, giga_loss[loss=0.358, simple_loss=0.4166, pruned_loss=0.1497, over 28534.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4166, pruned_loss=0.1683, over 5706103.78 frames. ], libri_tot_loss[loss=0.4294, simple_loss=0.4591, pruned_loss=0.1998, over 5727321.99 frames. ], giga_tot_loss[loss=0.3697, simple_loss=0.4118, pruned_loss=0.1638, over 5706584.15 frames. ], batch size: 307, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:29:48,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21886.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:30:01,865 INFO [train.py:968] (1/2) Epoch 1, batch 21900, giga_loss[loss=0.3721, simple_loss=0.4243, pruned_loss=0.16, over 28508.00 frames. ], tot_loss[loss=0.3783, simple_loss=0.4185, pruned_loss=0.1691, over 5703826.20 frames. ], libri_tot_loss[loss=0.4296, simple_loss=0.4591, pruned_loss=0.2, over 5729248.34 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4144, pruned_loss=0.1651, over 5702215.76 frames. ], batch size: 336, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:30:14,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-02-28 16:30:18,292 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-02-28 16:30:46,644 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 21950, giga_loss[loss=0.3471, simple_loss=0.408, pruned_loss=0.1431, over 28868.00 frames. ], tot_loss[loss=0.3802, simple_loss=0.4209, pruned_loss=0.1697, over 5713714.45 frames. ], libri_tot_loss[loss=0.4305, simple_loss=0.4597, pruned_loss=0.2007, over 5733953.78 frames. ], giga_tot_loss[loss=0.3734, simple_loss=0.4163, pruned_loss=0.1653, over 5707808.24 frames. ], batch size: 174, lr: 2.36e-02, grad_scale: 4.0
+2023-02-28 16:31:00,649 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:02,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21966.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:21,425 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=21988.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:26,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21995.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:31,029 INFO [train.py:968] (1/2) Epoch 1, batch 22000, giga_loss[loss=0.335, simple_loss=0.3966, pruned_loss=0.1367, over 28963.00 frames. ], tot_loss[loss=0.3812, simple_loss=0.4224, pruned_loss=0.17, over 5707803.82 frames. ], libri_tot_loss[loss=0.4313, simple_loss=0.4601, pruned_loss=0.2013, over 5736547.01 frames. ], giga_tot_loss[loss=0.3744, simple_loss=0.4178, pruned_loss=0.1655, over 5700455.15 frames. ], batch size: 136, lr: 2.35e-02, grad_scale: 8.0
+2023-02-28 16:31:38,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22006.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:31:51,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0991, 2.0077, 2.8613, 1.3844], device='cuda:1'), covar=tensor([0.0767, 0.0869, 0.0995, 0.1917], device='cuda:1'), in_proj_covar=tensor([0.0688, 0.0444, 0.0756, 0.0513], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 16:31:57,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22029.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:01,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:14,673 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-02-28 16:32:15,504 INFO [train.py:968] (1/2) Epoch 1, batch 22050, giga_loss[loss=0.3869, simple_loss=0.4186, pruned_loss=0.1776, over 28685.00 frames. ], tot_loss[loss=0.38, simple_loss=0.4216, pruned_loss=0.1692, over 5695018.83 frames. ], libri_tot_loss[loss=0.4321, simple_loss=0.4602, pruned_loss=0.202, over 5731833.32 frames. ], giga_tot_loss[loss=0.3724, simple_loss=0.4167, pruned_loss=0.164, over 5692849.25 frames. ], batch size: 99, lr: 2.35e-02, grad_scale: 4.0
+2023-02-28 16:32:16,261 INFO [optim.py:369] (1/2) 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 16:32:27,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22061.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:45,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22080.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:58,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22096.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:32:59,590 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-02-28 16:33:00,462 INFO [train.py:968] (1/2) Epoch 1, batch 22100, giga_loss[loss=0.429, simple_loss=0.4584, pruned_loss=0.1998, over 28780.00 frames. ], tot_loss[loss=0.3805, simple_loss=0.4222, pruned_loss=0.1694, over 5691379.88 frames. ], libri_tot_loss[loss=0.4335, simple_loss=0.4612, pruned_loss=0.2029, over 5725540.15 frames. ], giga_tot_loss[loss=0.3721, simple_loss=0.4166, pruned_loss=0.1638, over 5694987.73 frames. ], batch size: 227, lr: 2.35e-02, grad_scale: 4.0
+2023-02-28 16:33:05,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5513, 2.2460, 1.7594, 1.5441], device='cuda:1'), covar=tensor([0.0914, 0.0981, 0.0797, 0.0523], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0847, 0.0693, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0005], device='cuda:1')
+2023-02-28 16:33:42,518 INFO [train.py:968] (1/2) Epoch 1, batch 22150, giga_loss[loss=0.3878, simple_loss=0.4187, pruned_loss=0.1785, over 28488.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4248, pruned_loss=0.1724, over 5698331.02 frames. ], libri_tot_loss[loss=0.4333, simple_loss=0.4607, pruned_loss=0.203, over 5730648.66 frames. ], giga_tot_loss[loss=0.3768, simple_loss=0.4197, pruned_loss=0.1669, over 5695772.33 frames. ], batch size: 78, lr: 2.35e-02, grad_scale: 4.0
+2023-02-28 16:33:43,147 INFO [optim.py:369] (1/2) 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 16:33:47,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22156.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:34:17,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6767, 3.0488, 4.2899, 1.8449], device='cuda:1'), covar=tensor([0.0404, 0.0667, 0.0740, 0.1818], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0449, 0.0747, 0.0524], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0005, 0.0007, 0.0006], device='cuda:1')
+2023-02-28 16:34:25,941 INFO [train.py:968] (1/2) Epoch 1, batch 22200, giga_loss[loss=0.3748, simple_loss=0.4267, pruned_loss=0.1614, over 28724.00 frames. ], tot_loss[loss=0.387, simple_loss=0.4265, pruned_loss=0.1738, over 5704626.50 frames. ], libri_tot_loss[loss=0.4342, simple_loss=0.4614, pruned_loss=0.2035, over 5733091.87 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.4214, pruned_loss=0.1686, over 5699992.20 frames. ], batch size: 284, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:34:44,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22223.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:34:46,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22226.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:01,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22239.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:03,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22242.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:03,749 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 22250, libri_loss[loss=0.5471, simple_loss=0.5444, pruned_loss=0.275, over 29546.00 frames. ], tot_loss[loss=0.391, simple_loss=0.4301, pruned_loss=0.176, over 5705911.97 frames. ], libri_tot_loss[loss=0.4363, simple_loss=0.4629, pruned_loss=0.2048, over 5736378.77 frames. ], giga_tot_loss[loss=0.3819, simple_loss=0.4239, pruned_loss=0.17, over 5698643.39 frames. ], batch size: 89, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:35:09,206 INFO [optim.py:369] (1/2) 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 16:35:13,088 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22255.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:26,612 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:35:27,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4497, 1.7144, 3.9429, 2.7292], device='cuda:1'), covar=tensor([0.1569, 0.1158, 0.0245, 0.0477], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0430, 0.0552, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 16:35:45,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-02-28 16:35:47,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-02-28 16:35:47,610 INFO [train.py:968] (1/2) Epoch 1, batch 22300, libri_loss[loss=0.5115, simple_loss=0.5108, pruned_loss=0.256, over 29120.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.433, pruned_loss=0.1779, over 5716817.28 frames. ], libri_tot_loss[loss=0.438, simple_loss=0.4639, pruned_loss=0.2061, over 5743633.90 frames. ], giga_tot_loss[loss=0.3834, simple_loss=0.4256, pruned_loss=0.1706, over 5702848.08 frames. ], batch size: 101, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:35:57,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2212, 1.1513, 1.0346, 0.0395], device='cuda:1'), covar=tensor([0.0374, 0.0509, 0.0704, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0885, 0.0908, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 16:36:06,968 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-02-28 16:36:25,071 INFO [train.py:968] (1/2) Epoch 1, batch 22350, giga_loss[loss=0.3752, simple_loss=0.4178, pruned_loss=0.1663, over 28843.00 frames. ], tot_loss[loss=0.3998, simple_loss=0.4368, pruned_loss=0.1814, over 5718773.48 frames. ], libri_tot_loss[loss=0.4409, simple_loss=0.4656, pruned_loss=0.2081, over 5747947.39 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4272, pruned_loss=0.1717, over 5701222.19 frames. ], batch size: 112, lr: 2.34e-02, grad_scale: 4.0
+2023-02-28 16:36:25,783 INFO [optim.py:369] (1/2) 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 16:36:34,312 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22363.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:36:34,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22363.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:36:48,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22381.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:37:03,557 INFO [train.py:968] (1/2) Epoch 1, batch 22400, giga_loss[loss=0.4344, simple_loss=0.4677, pruned_loss=0.2005, over 28701.00 frames. ], tot_loss[loss=0.3996, simple_loss=0.4374, pruned_loss=0.1809, over 5718517.18 frames. ], libri_tot_loss[loss=0.442, simple_loss=0.4664, pruned_loss=0.2088, over 5743869.46 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4281, pruned_loss=0.1715, over 5706863.61 frames. ], batch size: 284, lr: 2.33e-02, grad_scale: 8.0
+2023-02-28 16:37:40,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2547, 2.0925, 1.5149, 0.8749], device='cuda:1'), covar=tensor([0.0431, 0.0302, 0.0277, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0548, 0.0578, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:37:49,152 INFO [train.py:968] (1/2) Epoch 1, batch 22450, giga_loss[loss=0.3361, simple_loss=0.3999, pruned_loss=0.1361, over 28908.00 frames. ], tot_loss[loss=0.3979, simple_loss=0.4364, pruned_loss=0.1797, over 5716738.08 frames. ], libri_tot_loss[loss=0.4419, simple_loss=0.4662, pruned_loss=0.2087, over 5744716.11 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4291, pruned_loss=0.1724, over 5706848.87 frames. ], batch size: 174, lr: 2.33e-02, grad_scale: 8.0
+2023-02-28 16:37:51,105 INFO [optim.py:369] (1/2) 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 16:38:22,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4950, 1.8666, 1.6469, 0.4503], device='cuda:1'), covar=tensor([0.1143, 0.0887, 0.1017, 0.1956], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0888, 0.0917, 0.0854], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 16:38:32,581 INFO [train.py:968] (1/2) Epoch 1, batch 22500, giga_loss[loss=0.3926, simple_loss=0.4424, pruned_loss=0.1714, over 28583.00 frames. ], tot_loss[loss=0.3955, simple_loss=0.4345, pruned_loss=0.1782, over 5713874.18 frames. ], libri_tot_loss[loss=0.4421, simple_loss=0.4664, pruned_loss=0.2089, over 5745598.53 frames. ], giga_tot_loss[loss=0.3864, simple_loss=0.4285, pruned_loss=0.1721, over 5705127.52 frames. ], batch size: 336, lr: 2.33e-02, grad_scale: 4.0
+2023-02-28 16:38:37,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22506.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:41,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22509.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:51,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22523.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:51,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9500, 1.8116, 1.4392, 1.3493], device='cuda:1'), covar=tensor([0.0726, 0.0957, 0.0987, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0631, 0.0626, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 16:38:52,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22524.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:54,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22527.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:38:56,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22531.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:02,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22538.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:15,604 INFO [train.py:968] (1/2) Epoch 1, batch 22550, giga_loss[loss=0.3309, simple_loss=0.3864, pruned_loss=0.1377, over 28954.00 frames. ], tot_loss[loss=0.3924, simple_loss=0.4319, pruned_loss=0.1765, over 5715984.42 frames. ], libri_tot_loss[loss=0.444, simple_loss=0.4674, pruned_loss=0.2103, over 5748722.47 frames. ], giga_tot_loss[loss=0.3823, simple_loss=0.4253, pruned_loss=0.1697, over 5705533.00 frames. ], batch size: 136, lr: 2.33e-02, grad_scale: 4.0
+2023-02-28 16:39:17,023 INFO [optim.py:369] (1/2) 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 16:39:22,263 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22556.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:32,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:39:34,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7321, 1.6971, 1.2274, 1.4163], device='cuda:1'), covar=tensor([0.0772, 0.0839, 0.1208, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0633, 0.0634, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 16:39:59,473 INFO [train.py:968] (1/2) Epoch 1, batch 22600, giga_loss[loss=0.3116, simple_loss=0.3721, pruned_loss=0.1255, over 29008.00 frames. ], tot_loss[loss=0.3856, simple_loss=0.4262, pruned_loss=0.1725, over 5717523.64 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.468, pruned_loss=0.2108, over 5749488.81 frames. ], giga_tot_loss[loss=0.3758, simple_loss=0.4198, pruned_loss=0.1659, over 5707981.72 frames. ], batch size: 164, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:40:00,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9991, 2.2403, 4.7767, 3.1108], device='cuda:1'), covar=tensor([0.2137, 0.1372, 0.0449, 0.0520], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0428, 0.0562, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 16:40:06,605 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22608.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:40:14,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22618.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:40:24,530 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22630.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:40:38,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1821, 1.7510, 1.7338, 1.5059], device='cuda:1'), covar=tensor([0.0624, 0.1501, 0.0924, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0547, 0.0900, 0.0675, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 16:40:39,251 INFO [train.py:968] (1/2) Epoch 1, batch 22650, libri_loss[loss=0.4895, simple_loss=0.4788, pruned_loss=0.2501, over 29601.00 frames. ], tot_loss[loss=0.3856, simple_loss=0.4259, pruned_loss=0.1726, over 5718648.94 frames. ], libri_tot_loss[loss=0.4466, simple_loss=0.469, pruned_loss=0.2121, over 5752827.81 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4184, pruned_loss=0.1648, over 5706870.70 frames. ], batch size: 74, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:40:40,752 INFO [optim.py:369] (1/2) 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 16:40:59,088 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:41:01,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:41:22,409 INFO [train.py:968] (1/2) Epoch 1, batch 22700, libri_loss[loss=0.4712, simple_loss=0.4847, pruned_loss=0.2288, over 26247.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4298, pruned_loss=0.175, over 5709941.83 frames. ], libri_tot_loss[loss=0.4472, simple_loss=0.469, pruned_loss=0.2127, over 5752058.67 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4216, pruned_loss=0.1662, over 5699487.47 frames. ], batch size: 136, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:41:26,163 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22706.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:41:54,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22738.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:04,671 INFO [train.py:968] (1/2) Epoch 1, batch 22750, giga_loss[loss=0.378, simple_loss=0.414, pruned_loss=0.171, over 28751.00 frames. ], tot_loss[loss=0.3895, simple_loss=0.431, pruned_loss=0.174, over 5706184.33 frames. ], libri_tot_loss[loss=0.4478, simple_loss=0.4692, pruned_loss=0.2131, over 5749482.15 frames. ], giga_tot_loss[loss=0.3777, simple_loss=0.4236, pruned_loss=0.1659, over 5699265.13 frames. ], batch size: 99, lr: 2.32e-02, grad_scale: 4.0
+2023-02-28 16:42:05,788 INFO [optim.py:369] (1/2) 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 16:42:14,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22761.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:16,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22764.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:26,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=22778.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:36,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5816, 1.6662, 1.5496, 0.3765], device='cuda:1'), covar=tensor([0.0603, 0.0591, 0.0874, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0888, 0.0932, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 16:42:38,482 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22793.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:42:43,108 INFO [train.py:968] (1/2) Epoch 1, batch 22800, giga_loss[loss=0.3609, simple_loss=0.396, pruned_loss=0.1629, over 28909.00 frames. ], tot_loss[loss=0.3888, simple_loss=0.4295, pruned_loss=0.1741, over 5703883.54 frames. ], libri_tot_loss[loss=0.448, simple_loss=0.4693, pruned_loss=0.2133, over 5750035.09 frames. ], giga_tot_loss[loss=0.3779, simple_loss=0.4227, pruned_loss=0.1666, over 5696748.78 frames. ], batch size: 106, lr: 2.31e-02, grad_scale: 8.0
+2023-02-28 16:43:26,380 INFO [train.py:968] (1/2) Epoch 1, batch 22850, giga_loss[loss=0.3353, simple_loss=0.38, pruned_loss=0.1453, over 28829.00 frames. ], tot_loss[loss=0.388, simple_loss=0.4274, pruned_loss=0.1743, over 5707371.76 frames. ], libri_tot_loss[loss=0.4487, simple_loss=0.4697, pruned_loss=0.2139, over 5753336.37 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.4209, pruned_loss=0.1671, over 5697866.47 frames. ], batch size: 99, lr: 2.31e-02, grad_scale: 8.0
+2023-02-28 16:43:29,184 INFO [optim.py:369] (1/2) 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 16:43:51,624 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22881.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:43:53,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22884.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:44:03,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6101, 1.2719, 1.4470, 0.8174], device='cuda:1'), covar=tensor([0.0430, 0.0332, 0.0258, 0.0417], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0540, 0.0581, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:44:05,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22898.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:44:06,951 INFO [train.py:968] (1/2) Epoch 1, batch 22900, giga_loss[loss=0.3363, simple_loss=0.378, pruned_loss=0.1473, over 28498.00 frames. ], tot_loss[loss=0.387, simple_loss=0.4254, pruned_loss=0.1743, over 5714396.71 frames. ], libri_tot_loss[loss=0.4496, simple_loss=0.4703, pruned_loss=0.2144, over 5756510.28 frames. ], giga_tot_loss[loss=0.3761, simple_loss=0.4185, pruned_loss=0.1669, over 5702763.16 frames. ], batch size: 85, lr: 2.31e-02, grad_scale: 8.0
+2023-02-28 16:44:17,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22913.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:44:42,405 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22944.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:44:46,232 INFO [train.py:968] (1/2) Epoch 1, batch 22950, giga_loss[loss=0.3535, simple_loss=0.394, pruned_loss=0.1565, over 28811.00 frames. ], tot_loss[loss=0.3857, simple_loss=0.4231, pruned_loss=0.1742, over 5717397.16 frames. ], libri_tot_loss[loss=0.4496, simple_loss=0.47, pruned_loss=0.2146, over 5758582.27 frames. ], giga_tot_loss[loss=0.3745, simple_loss=0.416, pruned_loss=0.1665, over 5704757.27 frames. ], batch size: 112, lr: 2.31e-02, grad_scale: 4.0
+2023-02-28 16:44:48,658 INFO [optim.py:369] (1/2) 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 16:44:57,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3075, 1.3359, 1.4205, 0.1623], device='cuda:1'), covar=tensor([0.0630, 0.0713, 0.0968, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0914, 0.0947, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 16:45:13,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:45:26,897 INFO [train.py:968] (1/2) Epoch 1, batch 23000, giga_loss[loss=0.4502, simple_loss=0.4663, pruned_loss=0.2171, over 27608.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4245, pruned_loss=0.1756, over 5713873.86 frames. ], libri_tot_loss[loss=0.4509, simple_loss=0.4709, pruned_loss=0.2155, over 5753034.95 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.417, pruned_loss=0.1678, over 5707245.97 frames. ], batch size: 472, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:45:31,625 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23005.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:45:52,400 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23032.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:45:59,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23041.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:01,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23044.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:06,367 INFO [train.py:968] (1/2) Epoch 1, batch 23050, giga_loss[loss=0.3118, simple_loss=0.3552, pruned_loss=0.1342, over 28497.00 frames. ], tot_loss[loss=0.3842, simple_loss=0.4214, pruned_loss=0.1735, over 5721654.52 frames. ], libri_tot_loss[loss=0.4513, simple_loss=0.471, pruned_loss=0.2158, over 5757210.17 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4137, pruned_loss=0.1656, over 5711510.13 frames. ], batch size: 71, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:46:08,356 INFO [optim.py:369] (1/2) 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 16:46:25,534 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23073.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:35,309 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23087.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:46:38,636 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 23100, giga_loss[loss=0.3562, simple_loss=0.4026, pruned_loss=0.1549, over 28075.00 frames. ], tot_loss[loss=0.3758, simple_loss=0.414, pruned_loss=0.1688, over 5713330.63 frames. ], libri_tot_loss[loss=0.4508, simple_loss=0.4704, pruned_loss=0.2156, over 5757960.54 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4078, pruned_loss=0.1623, over 5704151.86 frames. ], batch size: 412, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:47:01,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23119.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:07,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23126.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:09,474 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23129.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:23,912 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23148.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:24,821 INFO [train.py:968] (1/2) Epoch 1, batch 23150, giga_loss[loss=0.3632, simple_loss=0.4003, pruned_loss=0.1631, over 28764.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.4119, pruned_loss=0.1678, over 5707669.85 frames. ], libri_tot_loss[loss=0.4517, simple_loss=0.4709, pruned_loss=0.2163, over 5749586.09 frames. ], giga_tot_loss[loss=0.3615, simple_loss=0.4037, pruned_loss=0.1596, over 5706544.43 frames. ], batch size: 262, lr: 2.30e-02, grad_scale: 4.0
+2023-02-28 16:47:25,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23151.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:26,754 INFO [optim.py:369] (1/2) 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 16:47:26,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23153.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:29,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6030, 1.9664, 3.8570, 2.6686], device='cuda:1'), covar=tensor([0.1595, 0.1018, 0.0276, 0.0546], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0436, 0.0577, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0008, 0.0005], device='cuda:1')
+2023-02-28 16:47:31,284 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23158.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:50,365 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23180.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:47:55,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23187.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:48:07,716 INFO [train.py:968] (1/2) Epoch 1, batch 23200, giga_loss[loss=0.408, simple_loss=0.4237, pruned_loss=0.1962, over 24000.00 frames. ], tot_loss[loss=0.3741, simple_loss=0.4126, pruned_loss=0.1678, over 5709939.58 frames. ], libri_tot_loss[loss=0.4522, simple_loss=0.4712, pruned_loss=0.2166, over 5751108.25 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4054, pruned_loss=0.1607, over 5707304.41 frames. ], batch size: 705, lr: 2.29e-02, grad_scale: 8.0
+2023-02-28 16:48:24,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23218.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:48:39,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4042, 1.1824, 1.1791, 1.2046], device='cuda:1'), covar=tensor([0.1374, 0.1681, 0.1185, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0765, 0.0811, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 16:48:50,536 INFO [train.py:968] (1/2) Epoch 1, batch 23250, giga_loss[loss=0.3354, simple_loss=0.3949, pruned_loss=0.138, over 28904.00 frames. ], tot_loss[loss=0.3773, simple_loss=0.4163, pruned_loss=0.1692, over 5704176.46 frames. ], libri_tot_loss[loss=0.453, simple_loss=0.4718, pruned_loss=0.2171, over 5743404.53 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4087, pruned_loss=0.1619, over 5707836.73 frames. ], batch size: 145, lr: 2.29e-02, grad_scale: 8.0
+2023-02-28 16:48:53,221 INFO [optim.py:369] (1/2) 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 16:49:08,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3127, 1.4362, 0.9873, 1.2584], device='cuda:1'), covar=tensor([0.0597, 0.0518, 0.0934, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0636, 0.0639, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 16:49:30,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23296.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:49:32,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23299.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:49:32,485 INFO [train.py:968] (1/2) Epoch 1, batch 23300, giga_loss[loss=0.4236, simple_loss=0.4569, pruned_loss=0.1951, over 28658.00 frames. ], tot_loss[loss=0.3841, simple_loss=0.4222, pruned_loss=0.173, over 5700927.85 frames. ], libri_tot_loss[loss=0.4535, simple_loss=0.472, pruned_loss=0.2175, over 5739425.52 frames. ], giga_tot_loss[loss=0.3727, simple_loss=0.4144, pruned_loss=0.1655, over 5706115.37 frames. ], batch size: 262, lr: 2.29e-02, grad_scale: 4.0
+2023-02-28 16:49:56,587 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23328.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:50:13,230 INFO [train.py:968] (1/2) Epoch 1, batch 23350, giga_loss[loss=0.3678, simple_loss=0.4219, pruned_loss=0.1569, over 28808.00 frames. ], tot_loss[loss=0.3902, simple_loss=0.4276, pruned_loss=0.1764, over 5680735.41 frames. ], libri_tot_loss[loss=0.454, simple_loss=0.4723, pruned_loss=0.2178, over 5720241.26 frames. ], giga_tot_loss[loss=0.3794, simple_loss=0.4202, pruned_loss=0.1693, over 5703002.66 frames. ], batch size: 199, lr: 2.29e-02, grad_scale: 4.0
+2023-02-28 16:50:16,640 INFO [optim.py:369] (1/2) 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 16:50:17,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8471, 1.8160, 3.6698, 2.8989], device='cuda:1'), covar=tensor([0.1248, 0.1013, 0.0280, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0435, 0.0577, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0005, 0.0008, 0.0005], device='cuda:1')
+2023-02-28 16:50:54,287 INFO [train.py:968] (1/2) Epoch 1, batch 23400, libri_loss[loss=0.3921, simple_loss=0.4073, pruned_loss=0.1884, over 29357.00 frames. ], tot_loss[loss=0.3938, simple_loss=0.4309, pruned_loss=0.1784, over 5685871.81 frames. ], libri_tot_loss[loss=0.4536, simple_loss=0.4717, pruned_loss=0.2177, over 5725355.51 frames. ], giga_tot_loss[loss=0.3831, simple_loss=0.4239, pruned_loss=0.1712, over 5697739.10 frames. ], batch size: 67, lr: 2.29e-02, grad_scale: 4.0
+2023-02-28 16:51:01,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23407.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:51:40,305 INFO [train.py:968] (1/2) Epoch 1, batch 23450, libri_loss[loss=0.4755, simple_loss=0.4877, pruned_loss=0.2316, over 29642.00 frames. ], tot_loss[loss=0.397, simple_loss=0.4335, pruned_loss=0.1803, over 5689677.42 frames. ], libri_tot_loss[loss=0.4534, simple_loss=0.4716, pruned_loss=0.2176, over 5728187.57 frames. ], giga_tot_loss[loss=0.3878, simple_loss=0.4274, pruned_loss=0.1741, over 5695889.27 frames. ], batch size: 91, lr: 2.28e-02, grad_scale: 4.0
+2023-02-28 16:51:45,298 INFO [optim.py:369] (1/2) 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 16:52:03,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4860, 1.7929, 1.5446, 1.3755], device='cuda:1'), covar=tensor([0.1433, 0.0622, 0.0705, 0.1878], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0354, 0.0344, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0019], device='cuda:1')
+2023-02-28 16:52:32,649 INFO [train.py:968] (1/2) Epoch 1, batch 23500, giga_loss[loss=0.4419, simple_loss=0.4378, pruned_loss=0.2231, over 23736.00 frames. ], tot_loss[loss=0.4099, simple_loss=0.4415, pruned_loss=0.1891, over 5671210.07 frames. ], libri_tot_loss[loss=0.4533, simple_loss=0.4715, pruned_loss=0.2175, over 5720785.36 frames. ], giga_tot_loss[loss=0.4017, simple_loss=0.4361, pruned_loss=0.1837, over 5682162.90 frames. ], batch size: 705, lr: 2.28e-02, grad_scale: 4.0
+2023-02-28 16:52:33,834 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23501.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:53:16,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9425, 1.9440, 1.6373, 0.9883], device='cuda:1'), covar=tensor([0.0391, 0.0285, 0.0248, 0.0366], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0554, 0.0602, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 16:53:25,763 INFO [train.py:968] (1/2) Epoch 1, batch 23550, giga_loss[loss=0.5071, simple_loss=0.5079, pruned_loss=0.2531, over 27596.00 frames. ], tot_loss[loss=0.4187, simple_loss=0.448, pruned_loss=0.1947, over 5679179.84 frames. ], libri_tot_loss[loss=0.4528, simple_loss=0.471, pruned_loss=0.2173, over 5723310.63 frames. ], giga_tot_loss[loss=0.4122, simple_loss=0.4438, pruned_loss=0.1903, over 5685075.12 frames. ], batch size: 472, lr: 2.28e-02, grad_scale: 4.0
+2023-02-28 16:53:26,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23550.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:53:28,659 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:53:30,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6959, 2.0958, 1.6163, 1.5377], device='cuda:1'), covar=tensor([0.0807, 0.0870, 0.0737, 0.0521], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0819, 0.0690, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0006, 0.0005], device='cuda:1')
+2023-02-28 16:53:36,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23562.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:53:56,469 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23582.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:54:06,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23593.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:54:13,292 INFO [train.py:968] (1/2) Epoch 1, batch 23600, giga_loss[loss=0.4763, simple_loss=0.4877, pruned_loss=0.2325, over 28644.00 frames. ], tot_loss[loss=0.4294, simple_loss=0.4552, pruned_loss=0.2018, over 5680421.39 frames. ], libri_tot_loss[loss=0.4531, simple_loss=0.4711, pruned_loss=0.2176, over 5724881.69 frames. ], giga_tot_loss[loss=0.4228, simple_loss=0.451, pruned_loss=0.1973, over 5682062.35 frames. ], batch size: 307, lr: 2.28e-02, grad_scale: 8.0
+2023-02-28 16:54:23,879 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23610.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:54:35,813 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=23622.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:55:04,722 INFO [train.py:968] (1/2) Epoch 1, batch 23650, giga_loss[loss=0.4635, simple_loss=0.4766, pruned_loss=0.2251, over 28658.00 frames. ], tot_loss[loss=0.4417, simple_loss=0.4627, pruned_loss=0.2103, over 5648535.28 frames. ], libri_tot_loss[loss=0.454, simple_loss=0.4717, pruned_loss=0.2181, over 5705228.22 frames. ], giga_tot_loss[loss=0.4353, simple_loss=0.4586, pruned_loss=0.206, over 5667550.33 frames. ], batch size: 85, lr: 2.27e-02, grad_scale: 8.0
+2023-02-28 16:55:07,491 INFO [optim.py:369] (1/2) 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 16:55:55,643 INFO [train.py:968] (1/2) Epoch 1, batch 23700, giga_loss[loss=0.4572, simple_loss=0.4863, pruned_loss=0.2141, over 29059.00 frames. ], tot_loss[loss=0.4514, simple_loss=0.4696, pruned_loss=0.2166, over 5654372.72 frames. ], libri_tot_loss[loss=0.4545, simple_loss=0.4719, pruned_loss=0.2185, over 5711676.19 frames. ], giga_tot_loss[loss=0.4455, simple_loss=0.4659, pruned_loss=0.2126, over 5661965.95 frames. ], batch size: 155, lr: 2.27e-02, grad_scale: 4.0
+2023-02-28 16:56:00,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23705.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:03,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23708.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:32,086 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23736.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:32,753 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23737.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:34,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23739.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:56:42,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9049, 1.6099, 1.4687, 1.4527], device='cuda:1'), covar=tensor([0.0639, 0.1161, 0.0985, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0913, 0.0683, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 16:56:42,832 INFO [train.py:968] (1/2) Epoch 1, batch 23750, giga_loss[loss=0.454, simple_loss=0.471, pruned_loss=0.2185, over 28645.00 frames. ], tot_loss[loss=0.4546, simple_loss=0.472, pruned_loss=0.2186, over 5662773.36 frames. ], libri_tot_loss[loss=0.4549, simple_loss=0.472, pruned_loss=0.2189, over 5712754.76 frames. ], giga_tot_loss[loss=0.4496, simple_loss=0.469, pruned_loss=0.2151, over 5666934.15 frames. ], batch size: 262, lr: 2.27e-02, grad_scale: 4.0
+2023-02-28 16:56:49,483 INFO [optim.py:369] (1/2) 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 16:57:00,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0893, 1.1302, 0.9524, 0.9288], device='cuda:1'), covar=tensor([0.1426, 0.1586, 0.1189, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0787, 0.0836, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 16:57:02,726 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23768.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:57:34,469 INFO [train.py:968] (1/2) Epoch 1, batch 23800, giga_loss[loss=0.3928, simple_loss=0.4252, pruned_loss=0.1802, over 28862.00 frames. ], tot_loss[loss=0.4596, simple_loss=0.4747, pruned_loss=0.2223, over 5656930.08 frames. ], libri_tot_loss[loss=0.4557, simple_loss=0.4726, pruned_loss=0.2194, over 5712685.77 frames. ], giga_tot_loss[loss=0.4548, simple_loss=0.4717, pruned_loss=0.219, over 5659257.27 frames. ], batch size: 119, lr: 2.27e-02, grad_scale: 4.0
+2023-02-28 16:58:29,183 INFO [train.py:968] (1/2) Epoch 1, batch 23850, giga_loss[loss=0.4485, simple_loss=0.4775, pruned_loss=0.2097, over 29051.00 frames. ], tot_loss[loss=0.4649, simple_loss=0.4776, pruned_loss=0.2262, over 5644660.27 frames. ], libri_tot_loss[loss=0.4564, simple_loss=0.4729, pruned_loss=0.2199, over 5716623.51 frames. ], giga_tot_loss[loss=0.4607, simple_loss=0.4751, pruned_loss=0.2232, over 5641137.82 frames. ], batch size: 155, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 16:58:34,453 INFO [optim.py:369] (1/2) 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 16:58:50,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1953, 1.2922, 1.1503, 1.1438], device='cuda:1'), covar=tensor([0.1356, 0.1582, 0.1114, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0839, 0.0789, 0.0840, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 16:58:54,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23876.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 16:59:02,252 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 23900, giga_loss[loss=0.5344, simple_loss=0.5171, pruned_loss=0.2759, over 27613.00 frames. ], tot_loss[loss=0.4721, simple_loss=0.4819, pruned_loss=0.2312, over 5629726.88 frames. ], libri_tot_loss[loss=0.4566, simple_loss=0.4729, pruned_loss=0.2201, over 5707413.00 frames. ], giga_tot_loss[loss=0.4688, simple_loss=0.4799, pruned_loss=0.2288, over 5632940.48 frames. ], batch size: 472, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 17:00:26,469 INFO [train.py:968] (1/2) Epoch 1, batch 23950, giga_loss[loss=0.4233, simple_loss=0.4577, pruned_loss=0.1944, over 28858.00 frames. ], tot_loss[loss=0.4757, simple_loss=0.4836, pruned_loss=0.2338, over 5602991.48 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4735, pruned_loss=0.2208, over 5706003.80 frames. ], giga_tot_loss[loss=0.4723, simple_loss=0.4817, pruned_loss=0.2314, over 5605968.15 frames. ], batch size: 145, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 17:00:32,734 INFO [optim.py:369] (1/2) 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 17:00:46,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5475, 1.2602, 1.5178, 0.9001], device='cuda:1'), covar=tensor([0.0290, 0.0241, 0.0170, 0.0321], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0545, 0.0592, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 17:01:04,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23985.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:01:16,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23997.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:01:18,479 INFO [train.py:968] (1/2) Epoch 1, batch 24000, libri_loss[loss=0.3787, simple_loss=0.4057, pruned_loss=0.1759, over 29361.00 frames. ], tot_loss[loss=0.4728, simple_loss=0.4811, pruned_loss=0.2323, over 5614202.46 frames. ], libri_tot_loss[loss=0.4572, simple_loss=0.4731, pruned_loss=0.2207, over 5706498.26 frames. ], giga_tot_loss[loss=0.4709, simple_loss=0.4802, pruned_loss=0.2308, over 5613317.04 frames. ], batch size: 67, lr: 2.26e-02, grad_scale: 8.0
+2023-02-28 17:01:18,479 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 17:01:23,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1962, 1.1287, 1.2584, 0.6493], device='cuda:1'), covar=tensor([0.0329, 0.0278, 0.0223, 0.0394], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0545, 0.0582, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 17:01:26,963 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19327MB
+2023-02-28 17:01:45,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24019.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:01:47,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24022.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:02:14,853 INFO [train.py:968] (1/2) Epoch 1, batch 24050, giga_loss[loss=0.5653, simple_loss=0.5432, pruned_loss=0.2937, over 28262.00 frames. ], tot_loss[loss=0.4686, simple_loss=0.4783, pruned_loss=0.2295, over 5628767.74 frames. ], libri_tot_loss[loss=0.457, simple_loss=0.4729, pruned_loss=0.2206, over 5707501.38 frames. ], giga_tot_loss[loss=0.4675, simple_loss=0.4779, pruned_loss=0.2285, over 5626341.26 frames. ], batch size: 369, lr: 2.26e-02, grad_scale: 4.0
+2023-02-28 17:02:18,189 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24051.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:02:21,562 INFO [optim.py:369] (1/2) 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 17:02:59,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 17:03:01,134 INFO [train.py:968] (1/2) Epoch 1, batch 24100, giga_loss[loss=0.4448, simple_loss=0.4726, pruned_loss=0.2084, over 28812.00 frames. ], tot_loss[loss=0.4665, simple_loss=0.4774, pruned_loss=0.2278, over 5630299.82 frames. ], libri_tot_loss[loss=0.4565, simple_loss=0.4721, pruned_loss=0.2205, over 5712872.39 frames. ], giga_tot_loss[loss=0.4663, simple_loss=0.4779, pruned_loss=0.2273, over 5620699.16 frames. ], batch size: 112, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:03:32,797 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:03:36,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24131.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:03:46,138 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24140.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:03:48,329 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 24150, libri_loss[loss=0.5071, simple_loss=0.5159, pruned_loss=0.2491, over 29202.00 frames. ], tot_loss[loss=0.4706, simple_loss=0.4806, pruned_loss=0.2303, over 5625530.10 frames. ], libri_tot_loss[loss=0.4574, simple_loss=0.4727, pruned_loss=0.2211, over 5717033.11 frames. ], giga_tot_loss[loss=0.4698, simple_loss=0.4805, pruned_loss=0.2296, over 5612090.72 frames. ], batch size: 97, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:03:59,656 INFO [optim.py:369] (1/2) 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 17:04:04,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24160.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:04:18,706 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24172.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:04:46,912 INFO [train.py:968] (1/2) Epoch 1, batch 24200, libri_loss[loss=0.4942, simple_loss=0.4987, pruned_loss=0.2448, over 29370.00 frames. ], tot_loss[loss=0.469, simple_loss=0.4802, pruned_loss=0.2289, over 5631899.37 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4729, pruned_loss=0.2216, over 5721549.92 frames. ], giga_tot_loss[loss=0.4681, simple_loss=0.4802, pruned_loss=0.228, over 5614472.64 frames. ], batch size: 92, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:05:26,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5413, 1.5669, 1.5250, 1.3689], device='cuda:1'), covar=tensor([0.1198, 0.1098, 0.1089, 0.1846], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0502, 0.0407, 0.0512], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0012], device='cuda:1')
+2023-02-28 17:05:39,494 INFO [train.py:968] (1/2) Epoch 1, batch 24250, giga_loss[loss=0.4416, simple_loss=0.4429, pruned_loss=0.2201, over 23926.00 frames. ], tot_loss[loss=0.462, simple_loss=0.4758, pruned_loss=0.2241, over 5631840.35 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4724, pruned_loss=0.2213, over 5727872.62 frames. ], giga_tot_loss[loss=0.462, simple_loss=0.4763, pruned_loss=0.2238, over 5609194.91 frames. ], batch size: 705, lr: 2.25e-02, grad_scale: 4.0
+2023-02-28 17:05:44,810 INFO [optim.py:369] (1/2) 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 17:05:45,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24256.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:06:08,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1764, 1.3414, 1.1600, 1.1815], device='cuda:1'), covar=tensor([0.1229, 0.0820, 0.1104, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0493, 0.0401, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0012], device='cuda:1')
+2023-02-28 17:06:25,211 INFO [train.py:968] (1/2) Epoch 1, batch 24300, giga_loss[loss=0.4202, simple_loss=0.4577, pruned_loss=0.1914, over 28960.00 frames. ], tot_loss[loss=0.4568, simple_loss=0.4731, pruned_loss=0.2202, over 5649107.05 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4725, pruned_loss=0.2214, over 5733563.48 frames. ], giga_tot_loss[loss=0.4566, simple_loss=0.4735, pruned_loss=0.2199, over 5622908.77 frames. ], batch size: 199, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:06:26,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1108, 1.1233, 1.0571, 1.2449], device='cuda:1'), covar=tensor([0.1580, 0.1722, 0.1298, 0.1685], device='cuda:1'), in_proj_covar=tensor([0.0878, 0.0818, 0.0862, 0.0904], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 17:06:42,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3552, 1.1242, 1.3690, 0.8935], device='cuda:1'), covar=tensor([0.0460, 0.0300, 0.0223, 0.0357], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0530, 0.0564, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 17:07:16,094 INFO [train.py:968] (1/2) Epoch 1, batch 24350, giga_loss[loss=0.4703, simple_loss=0.4809, pruned_loss=0.2299, over 28915.00 frames. ], tot_loss[loss=0.4528, simple_loss=0.4702, pruned_loss=0.2177, over 5645373.25 frames. ], libri_tot_loss[loss=0.4571, simple_loss=0.4719, pruned_loss=0.2211, over 5736004.71 frames. ], giga_tot_loss[loss=0.4532, simple_loss=0.471, pruned_loss=0.2176, over 5621144.75 frames. ], batch size: 186, lr: 2.24e-02, grad_scale: 2.0
+2023-02-28 17:07:23,933 INFO [optim.py:369] (1/2) 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 17:08:06,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24399.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:06,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-02-28 17:08:06,375 INFO [train.py:968] (1/2) Epoch 1, batch 24400, giga_loss[loss=0.4214, simple_loss=0.4511, pruned_loss=0.1959, over 28877.00 frames. ], tot_loss[loss=0.4486, simple_loss=0.4674, pruned_loss=0.2149, over 5646666.45 frames. ], libri_tot_loss[loss=0.4573, simple_loss=0.472, pruned_loss=0.2213, over 5738201.13 frames. ], giga_tot_loss[loss=0.4486, simple_loss=0.4679, pruned_loss=0.2147, over 5623775.09 frames. ], batch size: 199, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:08:09,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24402.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:36,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=24430.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:37,062 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24431.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:08:55,666 INFO [train.py:968] (1/2) Epoch 1, batch 24450, giga_loss[loss=0.424, simple_loss=0.4618, pruned_loss=0.1931, over 28814.00 frames. ], tot_loss[loss=0.4471, simple_loss=0.466, pruned_loss=0.2141, over 5648461.15 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4724, pruned_loss=0.2217, over 5741859.75 frames. ], giga_tot_loss[loss=0.4462, simple_loss=0.4659, pruned_loss=0.2133, over 5623912.64 frames. ], batch size: 284, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:09:02,756 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 24500, giga_loss[loss=0.367, simple_loss=0.4192, pruned_loss=0.1574, over 29010.00 frames. ], tot_loss[loss=0.4457, simple_loss=0.4654, pruned_loss=0.213, over 5645222.48 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4724, pruned_loss=0.2217, over 5739035.38 frames. ], giga_tot_loss[loss=0.4449, simple_loss=0.4652, pruned_loss=0.2123, over 5627050.38 frames. ], batch size: 136, lr: 2.24e-02, grad_scale: 4.0
+2023-02-28 17:10:43,649 INFO [train.py:968] (1/2) Epoch 1, batch 24550, giga_loss[loss=0.3826, simple_loss=0.427, pruned_loss=0.1691, over 28671.00 frames. ], tot_loss[loss=0.4406, simple_loss=0.4627, pruned_loss=0.2092, over 5644830.38 frames. ], libri_tot_loss[loss=0.4577, simple_loss=0.4722, pruned_loss=0.2216, over 5729941.58 frames. ], giga_tot_loss[loss=0.4398, simple_loss=0.4625, pruned_loss=0.2085, over 5637809.80 frames. ], batch size: 85, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:10:51,135 INFO [optim.py:369] (1/2) 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 17:11:05,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7818, 1.3650, 1.4081, 1.4304], device='cuda:1'), covar=tensor([0.0745, 0.1434, 0.1067, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0879, 0.0674, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:11:35,951 INFO [train.py:968] (1/2) Epoch 1, batch 24600, giga_loss[loss=0.4196, simple_loss=0.469, pruned_loss=0.1851, over 28945.00 frames. ], tot_loss[loss=0.4359, simple_loss=0.461, pruned_loss=0.2054, over 5653696.75 frames. ], libri_tot_loss[loss=0.4586, simple_loss=0.4729, pruned_loss=0.2221, over 5727807.79 frames. ], giga_tot_loss[loss=0.4339, simple_loss=0.46, pruned_loss=0.2039, over 5647534.29 frames. ], batch size: 199, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:12:31,093 INFO [train.py:968] (1/2) Epoch 1, batch 24650, giga_loss[loss=0.4213, simple_loss=0.4626, pruned_loss=0.19, over 28566.00 frames. ], tot_loss[loss=0.4359, simple_loss=0.463, pruned_loss=0.2044, over 5666523.36 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4725, pruned_loss=0.2218, over 5730817.21 frames. ], giga_tot_loss[loss=0.4343, simple_loss=0.4623, pruned_loss=0.2032, over 5657678.18 frames. ], batch size: 60, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:12:37,141 INFO [optim.py:369] (1/2) 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 17:12:56,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-02-28 17:13:23,145 INFO [train.py:968] (1/2) Epoch 1, batch 24700, giga_loss[loss=0.4197, simple_loss=0.4634, pruned_loss=0.188, over 28528.00 frames. ], tot_loss[loss=0.4388, simple_loss=0.4645, pruned_loss=0.2065, over 5649074.04 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4724, pruned_loss=0.2217, over 5725439.45 frames. ], giga_tot_loss[loss=0.4372, simple_loss=0.4638, pruned_loss=0.2053, over 5645933.01 frames. ], batch size: 60, lr: 2.23e-02, grad_scale: 4.0
+2023-02-28 17:14:13,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2733, 1.7978, 1.6689, 1.8064], device='cuda:1'), covar=tensor([0.0695, 0.1529, 0.1067, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0915, 0.0674, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:14:14,046 INFO [train.py:968] (1/2) Epoch 1, batch 24750, giga_loss[loss=0.4264, simple_loss=0.4531, pruned_loss=0.1998, over 28890.00 frames. ], tot_loss[loss=0.4383, simple_loss=0.4645, pruned_loss=0.206, over 5666225.28 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4723, pruned_loss=0.2215, over 5722436.44 frames. ], giga_tot_loss[loss=0.4369, simple_loss=0.4638, pruned_loss=0.205, over 5664830.79 frames. ], batch size: 106, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:14:18,577 INFO [optim.py:369] (1/2) 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 17:14:23,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4023, 1.5795, 1.3763, 1.3321], device='cuda:1'), covar=tensor([0.1199, 0.0976, 0.1031, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0500, 0.0398, 0.0512], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0011, 0.0009, 0.0012], device='cuda:1')
+2023-02-28 17:15:03,024 INFO [train.py:968] (1/2) Epoch 1, batch 24800, giga_loss[loss=0.3912, simple_loss=0.4263, pruned_loss=0.1781, over 28891.00 frames. ], tot_loss[loss=0.4348, simple_loss=0.4615, pruned_loss=0.2041, over 5672535.11 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4725, pruned_loss=0.2214, over 5716010.45 frames. ], giga_tot_loss[loss=0.4334, simple_loss=0.4607, pruned_loss=0.2031, over 5676424.72 frames. ], batch size: 112, lr: 2.22e-02, grad_scale: 8.0
+2023-02-28 17:15:04,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5380, 1.7187, 1.6097, 0.6404], device='cuda:1'), covar=tensor([0.0737, 0.0693, 0.0780, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0952, 0.0972, 0.0955, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 17:15:07,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24805.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:15:47,098 INFO [train.py:968] (1/2) Epoch 1, batch 24850, giga_loss[loss=0.4341, simple_loss=0.4561, pruned_loss=0.2061, over 28808.00 frames. ], tot_loss[loss=0.4355, simple_loss=0.4605, pruned_loss=0.2053, over 5675808.67 frames. ], libri_tot_loss[loss=0.4571, simple_loss=0.4721, pruned_loss=0.2211, over 5721578.24 frames. ], giga_tot_loss[loss=0.4341, simple_loss=0.4598, pruned_loss=0.2042, over 5672506.70 frames. ], batch size: 243, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:15:54,406 INFO [optim.py:369] (1/2) 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 17:16:32,095 INFO [train.py:968] (1/2) Epoch 1, batch 24900, giga_loss[loss=0.415, simple_loss=0.4461, pruned_loss=0.192, over 28857.00 frames. ], tot_loss[loss=0.4356, simple_loss=0.46, pruned_loss=0.2056, over 5660998.91 frames. ], libri_tot_loss[loss=0.4573, simple_loss=0.4723, pruned_loss=0.2212, over 5706654.72 frames. ], giga_tot_loss[loss=0.4335, simple_loss=0.4589, pruned_loss=0.2041, over 5668889.51 frames. ], batch size: 186, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:17:06,065 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=24940.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:17:14,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24948.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:17:16,608 INFO [train.py:968] (1/2) Epoch 1, batch 24950, giga_loss[loss=0.4418, simple_loss=0.4815, pruned_loss=0.2011, over 28697.00 frames. ], tot_loss[loss=0.4327, simple_loss=0.4593, pruned_loss=0.2031, over 5658800.51 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4724, pruned_loss=0.2214, over 5698086.04 frames. ], giga_tot_loss[loss=0.4307, simple_loss=0.4581, pruned_loss=0.2016, over 5671303.33 frames. ], batch size: 99, lr: 2.22e-02, grad_scale: 4.0
+2023-02-28 17:17:18,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:17:29,738 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=24962.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:17:49,451 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24980.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:18:07,176 INFO [train.py:968] (1/2) Epoch 1, batch 25000, giga_loss[loss=0.4688, simple_loss=0.4541, pruned_loss=0.2417, over 23789.00 frames. ], tot_loss[loss=0.4306, simple_loss=0.4581, pruned_loss=0.2015, over 5658097.87 frames. ], libri_tot_loss[loss=0.4574, simple_loss=0.4722, pruned_loss=0.2213, over 5694002.16 frames. ], giga_tot_loss[loss=0.4283, simple_loss=0.4571, pruned_loss=0.1998, over 5671150.24 frames. ], batch size: 705, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:18:31,540 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-02-28 17:18:34,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5563, 1.6692, 3.4084, 2.5790], device='cuda:1'), covar=tensor([0.1423, 0.1058, 0.0346, 0.0784], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0451, 0.0605, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0005], device='cuda:1')
+2023-02-28 17:18:52,490 INFO [train.py:968] (1/2) Epoch 1, batch 25050, giga_loss[loss=0.4988, simple_loss=0.4976, pruned_loss=0.25, over 28393.00 frames. ], tot_loss[loss=0.4306, simple_loss=0.4576, pruned_loss=0.2018, over 5662148.24 frames. ], libri_tot_loss[loss=0.4571, simple_loss=0.4717, pruned_loss=0.2212, over 5696760.35 frames. ], giga_tot_loss[loss=0.4283, simple_loss=0.4567, pruned_loss=0.1999, over 5669463.88 frames. ], batch size: 368, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:19:01,699 INFO [optim.py:369] (1/2) 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 17:19:07,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4557, 1.9607, 1.6927, 0.3855], device='cuda:1'), covar=tensor([0.0835, 0.0639, 0.0755, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0926, 0.0946, 0.0941, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 17:19:37,742 INFO [train.py:968] (1/2) Epoch 1, batch 25100, giga_loss[loss=0.4326, simple_loss=0.4587, pruned_loss=0.2033, over 28584.00 frames. ], tot_loss[loss=0.4307, simple_loss=0.4572, pruned_loss=0.2021, over 5675126.71 frames. ], libri_tot_loss[loss=0.4577, simple_loss=0.4722, pruned_loss=0.2216, over 5701805.71 frames. ], giga_tot_loss[loss=0.4271, simple_loss=0.4555, pruned_loss=0.1993, over 5675028.64 frames. ], batch size: 307, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:20:29,420 INFO [train.py:968] (1/2) Epoch 1, batch 25150, giga_loss[loss=0.4306, simple_loss=0.4615, pruned_loss=0.1999, over 28919.00 frames. ], tot_loss[loss=0.43, simple_loss=0.4559, pruned_loss=0.2021, over 5663051.80 frames. ], libri_tot_loss[loss=0.457, simple_loss=0.4717, pruned_loss=0.2211, over 5705495.29 frames. ], giga_tot_loss[loss=0.4272, simple_loss=0.4546, pruned_loss=0.1999, over 5658964.83 frames. ], batch size: 174, lr: 2.21e-02, grad_scale: 4.0
+2023-02-28 17:20:38,147 INFO [optim.py:369] (1/2) 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 17:20:39,051 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25158.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:21:15,064 INFO [train.py:968] (1/2) Epoch 1, batch 25200, giga_loss[loss=0.3749, simple_loss=0.4154, pruned_loss=0.1672, over 29159.00 frames. ], tot_loss[loss=0.4305, simple_loss=0.4554, pruned_loss=0.2028, over 5669350.04 frames. ], libri_tot_loss[loss=0.4564, simple_loss=0.4712, pruned_loss=0.2208, over 5709382.72 frames. ], giga_tot_loss[loss=0.4281, simple_loss=0.4545, pruned_loss=0.2009, over 5661745.22 frames. ], batch size: 128, lr: 2.21e-02, grad_scale: 8.0
+2023-02-28 17:21:23,005 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-02-28 17:21:52,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25241.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:22:02,159 INFO [train.py:968] (1/2) Epoch 1, batch 25250, giga_loss[loss=0.4193, simple_loss=0.4531, pruned_loss=0.1927, over 28880.00 frames. ], tot_loss[loss=0.4331, simple_loss=0.4566, pruned_loss=0.2048, over 5677862.89 frames. ], libri_tot_loss[loss=0.4562, simple_loss=0.471, pruned_loss=0.2207, over 5716554.16 frames. ], giga_tot_loss[loss=0.4305, simple_loss=0.4554, pruned_loss=0.2028, over 5664200.78 frames. ], batch size: 174, lr: 2.20e-02, grad_scale: 4.0
+2023-02-28 17:22:04,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1618, 1.7454, 1.5266, 1.2753], device='cuda:1'), covar=tensor([0.1531, 0.0586, 0.0719, 0.2340], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0346, 0.0332, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0020], device='cuda:1')
+2023-02-28 17:22:09,181 INFO [optim.py:369] (1/2) 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 17:22:47,788 INFO [train.py:968] (1/2) Epoch 1, batch 25300, giga_loss[loss=0.3523, simple_loss=0.408, pruned_loss=0.1483, over 28953.00 frames. ], tot_loss[loss=0.4298, simple_loss=0.4539, pruned_loss=0.2028, over 5674317.13 frames. ], libri_tot_loss[loss=0.4567, simple_loss=0.4712, pruned_loss=0.2211, over 5710538.03 frames. ], giga_tot_loss[loss=0.4266, simple_loss=0.4525, pruned_loss=0.2004, over 5667424.98 frames. ], batch size: 145, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:22:53,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2384, 1.7061, 1.7233, 1.7120], device='cuda:1'), covar=tensor([0.0688, 0.1737, 0.1095, 0.1307], device='cuda:1'), in_proj_covar=tensor([0.0565, 0.0916, 0.0685, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:23:04,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25315.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:23:26,222 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 25350, giga_loss[loss=0.4744, simple_loss=0.4803, pruned_loss=0.2342, over 27965.00 frames. ], tot_loss[loss=0.4313, simple_loss=0.4545, pruned_loss=0.2041, over 5668276.04 frames. ], libri_tot_loss[loss=0.4571, simple_loss=0.4714, pruned_loss=0.2213, over 5714458.50 frames. ], giga_tot_loss[loss=0.4279, simple_loss=0.4528, pruned_loss=0.2016, over 5658580.89 frames. ], batch size: 412, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:23:48,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5536, 1.5300, 1.4744, 1.4521], device='cuda:1'), covar=tensor([0.0599, 0.0893, 0.0730, 0.0722], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0903, 0.0674, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:23:48,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2284, 1.0911, 1.0061, 0.9023], device='cuda:1'), covar=tensor([0.0273, 0.0236, 0.0213, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0547, 0.0585, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 17:23:49,108 INFO [optim.py:369] (1/2) 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 17:24:14,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25388.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:24:27,199 INFO [train.py:968] (1/2) Epoch 1, batch 25400, giga_loss[loss=0.3933, simple_loss=0.4411, pruned_loss=0.1727, over 28950.00 frames. ], tot_loss[loss=0.4325, simple_loss=0.456, pruned_loss=0.2045, over 5668275.57 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.4719, pruned_loss=0.2217, over 5715817.72 frames. ], giga_tot_loss[loss=0.4284, simple_loss=0.4537, pruned_loss=0.2016, over 5657826.03 frames. ], batch size: 213, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:24:42,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8676, 2.0173, 2.0876, 1.7049], device='cuda:1'), covar=tensor([0.0678, 0.1756, 0.0972, 0.1516], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0900, 0.0667, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:25:14,210 INFO [train.py:968] (1/2) Epoch 1, batch 25450, giga_loss[loss=0.3782, simple_loss=0.4253, pruned_loss=0.1655, over 28850.00 frames. ], tot_loss[loss=0.4288, simple_loss=0.4549, pruned_loss=0.2013, over 5671582.99 frames. ], libri_tot_loss[loss=0.458, simple_loss=0.4722, pruned_loss=0.2219, over 5716990.63 frames. ], giga_tot_loss[loss=0.4251, simple_loss=0.4527, pruned_loss=0.1987, over 5662163.87 frames. ], batch size: 112, lr: 2.20e-02, grad_scale: 2.0
+2023-02-28 17:25:24,503 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:25:27,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25461.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:25:47,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25480.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:25:48,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25483.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:25:55,268 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25490.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:26:04,918 INFO [train.py:968] (1/2) Epoch 1, batch 25500, giga_loss[loss=0.4199, simple_loss=0.4525, pruned_loss=0.1936, over 28879.00 frames. ], tot_loss[loss=0.428, simple_loss=0.4548, pruned_loss=0.2006, over 5666701.42 frames. ], libri_tot_loss[loss=0.4585, simple_loss=0.4726, pruned_loss=0.2222, over 5718499.01 frames. ], giga_tot_loss[loss=0.4244, simple_loss=0.4525, pruned_loss=0.1981, over 5657464.73 frames. ], batch size: 227, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:26:19,901 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25512.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:26:37,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5121, 1.3105, 1.3155, 0.9562], device='cuda:1'), covar=tensor([0.0453, 0.0363, 0.0256, 0.0394], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0561, 0.0608, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 17:26:39,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25533.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:26:54,261 INFO [train.py:968] (1/2) Epoch 1, batch 25550, libri_loss[loss=0.4052, simple_loss=0.4233, pruned_loss=0.1936, over 29669.00 frames. ], tot_loss[loss=0.4282, simple_loss=0.4545, pruned_loss=0.2009, over 5673389.72 frames. ], libri_tot_loss[loss=0.4585, simple_loss=0.4725, pruned_loss=0.2223, over 5722497.12 frames. ], giga_tot_loss[loss=0.4247, simple_loss=0.4524, pruned_loss=0.1985, over 5661290.07 frames. ], batch size: 69, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:27:03,030 INFO [optim.py:369] (1/2) 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 17:27:44,062 INFO [train.py:968] (1/2) Epoch 1, batch 25600, libri_loss[loss=0.4451, simple_loss=0.4533, pruned_loss=0.2185, over 29371.00 frames. ], tot_loss[loss=0.436, simple_loss=0.4591, pruned_loss=0.2064, over 5658159.17 frames. ], libri_tot_loss[loss=0.4584, simple_loss=0.4722, pruned_loss=0.2223, over 5722621.52 frames. ], giga_tot_loss[loss=0.4326, simple_loss=0.4572, pruned_loss=0.204, over 5646810.87 frames. ], batch size: 71, lr: 2.19e-02, grad_scale: 4.0
+2023-02-28 17:28:00,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25616.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:28:36,577 INFO [train.py:968] (1/2) Epoch 1, batch 25650, giga_loss[loss=0.3994, simple_loss=0.4341, pruned_loss=0.1824, over 28932.00 frames. ], tot_loss[loss=0.4368, simple_loss=0.4589, pruned_loss=0.2074, over 5660859.00 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.472, pruned_loss=0.2219, over 5723398.48 frames. ], giga_tot_loss[loss=0.4342, simple_loss=0.4575, pruned_loss=0.2055, over 5650281.96 frames. ], batch size: 174, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:28:47,733 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25676.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:29:10,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25679.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:29:15,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1214, 1.5504, 1.6396, 1.4762], device='cuda:1'), covar=tensor([0.0654, 0.1510, 0.1036, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0547, 0.0885, 0.0677, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:29:28,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4849, 2.3633, 2.4357, 2.2721], device='cuda:1'), covar=tensor([0.0470, 0.1420, 0.0860, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0545, 0.0883, 0.0672, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:29:32,660 INFO [train.py:968] (1/2) Epoch 1, batch 25700, giga_loss[loss=0.4053, simple_loss=0.4358, pruned_loss=0.1874, over 28994.00 frames. ], tot_loss[loss=0.4394, simple_loss=0.46, pruned_loss=0.2095, over 5665281.51 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.472, pruned_loss=0.2219, over 5723398.48 frames. ], giga_tot_loss[loss=0.4374, simple_loss=0.4589, pruned_loss=0.208, over 5657049.26 frames. ], batch size: 136, lr: 2.19e-02, grad_scale: 2.0
+2023-02-28 17:29:43,191 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25708.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:29:47,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8265, 1.7606, 3.2581, 2.7465], device='cuda:1'), covar=tensor([0.1156, 0.0937, 0.0301, 0.0460], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0447, 0.0581, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 17:30:21,056 INFO [train.py:968] (1/2) Epoch 1, batch 25750, giga_loss[loss=0.399, simple_loss=0.4349, pruned_loss=0.1815, over 28731.00 frames. ], tot_loss[loss=0.4414, simple_loss=0.4611, pruned_loss=0.2109, over 5658608.47 frames. ], libri_tot_loss[loss=0.4575, simple_loss=0.4716, pruned_loss=0.2217, over 5725053.93 frames. ], giga_tot_loss[loss=0.4401, simple_loss=0.4605, pruned_loss=0.2099, over 5650065.25 frames. ], batch size: 92, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:30:31,452 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25759.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:30:32,607 INFO [optim.py:369] (1/2) 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 17:30:34,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25762.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:30:35,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25763.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:31:02,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25791.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:31:09,268 INFO [train.py:968] (1/2) Epoch 1, batch 25800, giga_loss[loss=0.4448, simple_loss=0.4572, pruned_loss=0.2162, over 27923.00 frames. ], tot_loss[loss=0.4423, simple_loss=0.4615, pruned_loss=0.2115, over 5664572.04 frames. ], libri_tot_loss[loss=0.4582, simple_loss=0.472, pruned_loss=0.2222, over 5729519.23 frames. ], giga_tot_loss[loss=0.4401, simple_loss=0.4603, pruned_loss=0.21, over 5652054.55 frames. ], batch size: 412, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:31:31,934 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=25821.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:31:56,009 INFO [train.py:968] (1/2) Epoch 1, batch 25850, giga_loss[loss=0.4074, simple_loss=0.4528, pruned_loss=0.181, over 28833.00 frames. ], tot_loss[loss=0.4395, simple_loss=0.4609, pruned_loss=0.2091, over 5670818.43 frames. ], libri_tot_loss[loss=0.4584, simple_loss=0.4723, pruned_loss=0.2223, over 5730431.86 frames. ], giga_tot_loss[loss=0.4375, simple_loss=0.4596, pruned_loss=0.2077, over 5659552.35 frames. ], batch size: 186, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:31:58,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5275, 2.1384, 1.6300, 1.5952], device='cuda:1'), covar=tensor([0.1231, 0.1219, 0.1075, 0.0676], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0858, 0.0714, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0006, 0.0005], device='cuda:1')
+2023-02-28 17:32:08,298 INFO [optim.py:369] (1/2) 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 17:32:11,118 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-02-28 17:32:51,238 INFO [train.py:968] (1/2) Epoch 1, batch 25900, giga_loss[loss=0.4171, simple_loss=0.4318, pruned_loss=0.2013, over 27583.00 frames. ], tot_loss[loss=0.4291, simple_loss=0.4535, pruned_loss=0.2024, over 5659839.66 frames. ], libri_tot_loss[loss=0.4584, simple_loss=0.4723, pruned_loss=0.2223, over 5730431.86 frames. ], giga_tot_loss[loss=0.4275, simple_loss=0.4525, pruned_loss=0.2013, over 5651071.12 frames. ], batch size: 472, lr: 2.18e-02, grad_scale: 2.0
+2023-02-28 17:32:56,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25906.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:32:58,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25909.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:33:27,818 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 25950, giga_loss[loss=0.4102, simple_loss=0.4448, pruned_loss=0.1878, over 29054.00 frames. ], tot_loss[loss=0.4258, simple_loss=0.4509, pruned_loss=0.2003, over 5673190.98 frames. ], libri_tot_loss[loss=0.4585, simple_loss=0.4725, pruned_loss=0.2223, over 5735532.60 frames. ], giga_tot_loss[loss=0.4233, simple_loss=0.4493, pruned_loss=0.1987, over 5659061.70 frames. ], batch size: 155, lr: 2.17e-02, grad_scale: 2.0
+2023-02-28 17:33:46,512 INFO [optim.py:369] (1/2) 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 17:34:24,912 INFO [train.py:968] (1/2) Epoch 1, batch 26000, giga_loss[loss=0.4012, simple_loss=0.4356, pruned_loss=0.1834, over 28982.00 frames. ], tot_loss[loss=0.4242, simple_loss=0.4493, pruned_loss=0.1995, over 5670897.48 frames. ], libri_tot_loss[loss=0.4589, simple_loss=0.4727, pruned_loss=0.2226, over 5724911.32 frames. ], giga_tot_loss[loss=0.4209, simple_loss=0.4471, pruned_loss=0.1974, over 5667737.82 frames. ], batch size: 136, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:35:17,769 INFO [train.py:968] (1/2) Epoch 1, batch 26050, giga_loss[loss=0.3753, simple_loss=0.4177, pruned_loss=0.1665, over 28453.00 frames. ], tot_loss[loss=0.4249, simple_loss=0.4499, pruned_loss=0.1999, over 5672342.18 frames. ], libri_tot_loss[loss=0.4591, simple_loss=0.4727, pruned_loss=0.2227, over 5724182.65 frames. ], giga_tot_loss[loss=0.4213, simple_loss=0.4476, pruned_loss=0.1975, over 5669507.04 frames. ], batch size: 65, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:35:28,466 INFO [optim.py:369] (1/2) 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 17:35:44,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-02-28 17:36:03,253 INFO [train.py:968] (1/2) Epoch 1, batch 26100, giga_loss[loss=0.3943, simple_loss=0.4446, pruned_loss=0.172, over 29034.00 frames. ], tot_loss[loss=0.4301, simple_loss=0.4547, pruned_loss=0.2027, over 5664965.80 frames. ], libri_tot_loss[loss=0.4588, simple_loss=0.4724, pruned_loss=0.2226, over 5711355.24 frames. ], giga_tot_loss[loss=0.4263, simple_loss=0.4524, pruned_loss=0.2002, over 5672190.06 frames. ], batch size: 128, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:36:54,433 INFO [train.py:968] (1/2) Epoch 1, batch 26150, giga_loss[loss=0.4972, simple_loss=0.5081, pruned_loss=0.2432, over 28545.00 frames. ], tot_loss[loss=0.4285, simple_loss=0.4567, pruned_loss=0.2002, over 5670310.74 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.4716, pruned_loss=0.2223, over 5715028.19 frames. ], giga_tot_loss[loss=0.4256, simple_loss=0.4553, pruned_loss=0.198, over 5672144.00 frames. ], batch size: 336, lr: 2.17e-02, grad_scale: 4.0
+2023-02-28 17:37:05,108 INFO [optim.py:369] (1/2) 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 17:37:41,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26196.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:37:44,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4722, 3.4338, 2.2586, 2.1383], device='cuda:1'), covar=tensor([0.0910, 0.0738, 0.0779, 0.0468], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0861, 0.0710, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 17:37:45,267 INFO [train.py:968] (1/2) Epoch 1, batch 26200, giga_loss[loss=0.3859, simple_loss=0.4313, pruned_loss=0.1702, over 28442.00 frames. ], tot_loss[loss=0.4311, simple_loss=0.4591, pruned_loss=0.2016, over 5674852.13 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4713, pruned_loss=0.2222, over 5715841.05 frames. ], giga_tot_loss[loss=0.4289, simple_loss=0.4581, pruned_loss=0.1998, over 5675429.38 frames. ], batch size: 65, lr: 2.16e-02, grad_scale: 4.0
+2023-02-28 17:38:36,243 INFO [train.py:968] (1/2) Epoch 1, batch 26250, giga_loss[loss=0.3739, simple_loss=0.4234, pruned_loss=0.1622, over 28160.00 frames. ], tot_loss[loss=0.4359, simple_loss=0.462, pruned_loss=0.2049, over 5676145.60 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.471, pruned_loss=0.2221, over 5718869.76 frames. ], giga_tot_loss[loss=0.4341, simple_loss=0.4614, pruned_loss=0.2034, over 5673337.65 frames. ], batch size: 77, lr: 2.16e-02, grad_scale: 2.0
+2023-02-28 17:38:37,950 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26251.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:38:49,719 INFO [optim.py:369] (1/2) 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 17:39:24,943 INFO [train.py:968] (1/2) Epoch 1, batch 26300, giga_loss[loss=0.4021, simple_loss=0.4441, pruned_loss=0.1801, over 28615.00 frames. ], tot_loss[loss=0.4384, simple_loss=0.4636, pruned_loss=0.2067, over 5668749.62 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4713, pruned_loss=0.2222, over 5710868.57 frames. ], giga_tot_loss[loss=0.4365, simple_loss=0.4627, pruned_loss=0.2052, over 5672929.22 frames. ], batch size: 71, lr: 2.16e-02, grad_scale: 2.0
+2023-02-28 17:39:39,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6568, 1.4766, 1.5224, 1.2732], device='cuda:1'), covar=tensor([0.0784, 0.1278, 0.1198, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0911, 0.0683, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 17:39:44,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3787, 1.6430, 1.3823, 1.4715], device='cuda:1'), covar=tensor([0.1094, 0.0526, 0.0630, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0344, 0.0330, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0021], device='cuda:1')
+2023-02-28 17:40:03,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26339.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:40:06,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26342.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:40:09,951 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1070, 1.0886, 0.9825, 1.1039], device='cuda:1'), covar=tensor([0.1476, 0.1603, 0.1236, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0808, 0.0850, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 17:40:14,639 INFO [train.py:968] (1/2) Epoch 1, batch 26350, giga_loss[loss=0.5188, simple_loss=0.5007, pruned_loss=0.2685, over 26494.00 frames. ], tot_loss[loss=0.4405, simple_loss=0.464, pruned_loss=0.2086, over 5670338.39 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4712, pruned_loss=0.2223, over 5709367.48 frames. ], giga_tot_loss[loss=0.4387, simple_loss=0.4632, pruned_loss=0.2071, over 5674287.76 frames. ], batch size: 555, lr: 2.16e-02, grad_scale: 2.0
+2023-02-28 17:40:19,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3515, 1.9629, 1.6095, 1.3277], device='cuda:1'), covar=tensor([0.1490, 0.0638, 0.0728, 0.1839], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0342, 0.0331, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0021], device='cuda:1')
+2023-02-28 17:40:25,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1976, 1.4310, 1.1664, 0.2436], device='cuda:1'), covar=tensor([0.0490, 0.0425, 0.0533, 0.0739], device='cuda:1'), in_proj_covar=tensor([0.0964, 0.0968, 0.0960, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 17:40:26,031 INFO [optim.py:369] (1/2) 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 17:40:36,112 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26371.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:41:00,302 INFO [train.py:968] (1/2) Epoch 1, batch 26400, giga_loss[loss=0.3655, simple_loss=0.4149, pruned_loss=0.1581, over 28936.00 frames. ], tot_loss[loss=0.4382, simple_loss=0.4615, pruned_loss=0.2074, over 5686699.64 frames. ], libri_tot_loss[loss=0.4572, simple_loss=0.4705, pruned_loss=0.2219, over 5718214.23 frames. ], giga_tot_loss[loss=0.4365, simple_loss=0.4612, pruned_loss=0.2059, over 5680897.99 frames. ], batch size: 145, lr: 2.16e-02, grad_scale: 4.0
+2023-02-28 17:41:33,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9424, 2.4533, 2.1749, 1.8575], device='cuda:1'), covar=tensor([0.0994, 0.0985, 0.0767, 0.0494], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0887, 0.0719, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 17:41:47,648 INFO [train.py:968] (1/2) Epoch 1, batch 26450, giga_loss[loss=0.3997, simple_loss=0.4384, pruned_loss=0.1805, over 28772.00 frames. ], tot_loss[loss=0.4389, simple_loss=0.4612, pruned_loss=0.2083, over 5685114.71 frames. ], libri_tot_loss[loss=0.4576, simple_loss=0.471, pruned_loss=0.2221, over 5717010.15 frames. ], giga_tot_loss[loss=0.4369, simple_loss=0.4603, pruned_loss=0.2067, over 5681041.23 frames. ], batch size: 243, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:42:00,388 INFO [optim.py:369] (1/2) 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 17:42:07,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4818, 1.3504, 1.3007, 1.5224], device='cuda:1'), covar=tensor([0.1682, 0.1824, 0.1377, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.0878, 0.0794, 0.0835, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 17:42:22,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1306, 1.8027, 1.3883, 1.4431], device='cuda:1'), covar=tensor([0.0658, 0.0903, 0.1084, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0618, 0.0628, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 17:42:39,268 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26495.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:42:44,159 INFO [train.py:968] (1/2) Epoch 1, batch 26500, giga_loss[loss=0.4353, simple_loss=0.4628, pruned_loss=0.2039, over 28865.00 frames. ], tot_loss[loss=0.4351, simple_loss=0.4584, pruned_loss=0.2059, over 5681908.06 frames. ], libri_tot_loss[loss=0.4574, simple_loss=0.4708, pruned_loss=0.222, over 5716962.09 frames. ], giga_tot_loss[loss=0.4336, simple_loss=0.4579, pruned_loss=0.2047, over 5678640.45 frames. ], batch size: 186, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:43:30,907 INFO [train.py:968] (1/2) Epoch 1, batch 26550, giga_loss[loss=0.4251, simple_loss=0.4578, pruned_loss=0.1962, over 28334.00 frames. ], tot_loss[loss=0.4351, simple_loss=0.4585, pruned_loss=0.2058, over 5683093.31 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4712, pruned_loss=0.2223, over 5721123.90 frames. ], giga_tot_loss[loss=0.433, simple_loss=0.4574, pruned_loss=0.2043, over 5675945.59 frames. ], batch size: 368, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:43:41,227 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:44:19,166 INFO [train.py:968] (1/2) Epoch 1, batch 26600, giga_loss[loss=0.4682, simple_loss=0.4786, pruned_loss=0.2289, over 27902.00 frames. ], tot_loss[loss=0.4352, simple_loss=0.4574, pruned_loss=0.2065, over 5671976.89 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4712, pruned_loss=0.2223, over 5723697.79 frames. ], giga_tot_loss[loss=0.4332, simple_loss=0.4564, pruned_loss=0.205, over 5663273.41 frames. ], batch size: 412, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:44:42,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26626.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:44:59,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6354, 1.6855, 3.8984, 2.7836], device='cuda:1'), covar=tensor([0.1501, 0.1172, 0.0272, 0.0459], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0457, 0.0612, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 17:45:02,656 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26643.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:45:07,120 INFO [train.py:968] (1/2) Epoch 1, batch 26650, libri_loss[loss=0.5196, simple_loss=0.5159, pruned_loss=0.2617, over 20185.00 frames. ], tot_loss[loss=0.4307, simple_loss=0.4542, pruned_loss=0.2036, over 5659409.98 frames. ], libri_tot_loss[loss=0.4586, simple_loss=0.4718, pruned_loss=0.2227, over 5718061.05 frames. ], giga_tot_loss[loss=0.4279, simple_loss=0.4525, pruned_loss=0.2017, over 5657304.41 frames. ], batch size: 187, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:45:18,194 INFO [optim.py:369] (1/2) 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 17:45:52,809 INFO [train.py:968] (1/2) Epoch 1, batch 26700, giga_loss[loss=0.5004, simple_loss=0.5049, pruned_loss=0.2479, over 27902.00 frames. ], tot_loss[loss=0.4331, simple_loss=0.4565, pruned_loss=0.2049, over 5644094.96 frames. ], libri_tot_loss[loss=0.4595, simple_loss=0.4723, pruned_loss=0.2233, over 5700287.49 frames. ], giga_tot_loss[loss=0.4293, simple_loss=0.4541, pruned_loss=0.2022, over 5656882.99 frames. ], batch size: 412, lr: 2.15e-02, grad_scale: 4.0
+2023-02-28 17:46:05,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-02-28 17:46:28,257 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-02-28 17:46:38,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4281, 1.6819, 1.4066, 0.2428], device='cuda:1'), covar=tensor([0.0572, 0.0486, 0.0754, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0949, 0.0974, 0.0934, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 17:46:42,429 INFO [train.py:968] (1/2) Epoch 1, batch 26750, libri_loss[loss=0.4638, simple_loss=0.4771, pruned_loss=0.2252, over 27907.00 frames. ], tot_loss[loss=0.4371, simple_loss=0.4599, pruned_loss=0.2071, over 5652766.98 frames. ], libri_tot_loss[loss=0.4594, simple_loss=0.4723, pruned_loss=0.2232, over 5702906.98 frames. ], giga_tot_loss[loss=0.4336, simple_loss=0.4579, pruned_loss=0.2047, over 5659555.88 frames. ], batch size: 116, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:46:55,733 INFO [optim.py:369] (1/2) 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 17:47:04,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26769.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 17:47:09,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26772.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:47:33,477 INFO [train.py:968] (1/2) Epoch 1, batch 26800, giga_loss[loss=0.3788, simple_loss=0.4297, pruned_loss=0.1639, over 28980.00 frames. ], tot_loss[loss=0.4377, simple_loss=0.4602, pruned_loss=0.2076, over 5654021.99 frames. ], libri_tot_loss[loss=0.4591, simple_loss=0.4721, pruned_loss=0.223, over 5705205.99 frames. ], giga_tot_loss[loss=0.4349, simple_loss=0.4584, pruned_loss=0.2056, over 5656303.85 frames. ], batch size: 164, lr: 2.14e-02, grad_scale: 8.0
+2023-02-28 17:47:36,103 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26801.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:47:42,291 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=26807.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:47:47,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-02-28 17:47:50,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4128, 1.7395, 1.4603, 1.4381], device='cuda:1'), covar=tensor([0.1470, 0.0676, 0.0752, 0.1825], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0348, 0.0333, 0.0551], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0021], device='cuda:1')
+2023-02-28 17:47:57,482 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-02-28 17:48:18,992 INFO [train.py:968] (1/2) Epoch 1, batch 26850, giga_loss[loss=0.4082, simple_loss=0.4679, pruned_loss=0.1742, over 28822.00 frames. ], tot_loss[loss=0.435, simple_loss=0.4604, pruned_loss=0.2048, over 5673000.00 frames. ], libri_tot_loss[loss=0.4596, simple_loss=0.4725, pruned_loss=0.2233, over 5708388.80 frames. ], giga_tot_loss[loss=0.4318, simple_loss=0.4584, pruned_loss=0.2026, over 5671021.89 frames. ], batch size: 119, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:48:25,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6429, 1.6046, 1.2871, 1.2303], device='cuda:1'), covar=tensor([0.0793, 0.0922, 0.1142, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0627, 0.0618, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 17:48:29,162 INFO [optim.py:369] (1/2) 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 17:48:35,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:49:05,423 INFO [train.py:968] (1/2) Epoch 1, batch 26900, giga_loss[loss=0.3935, simple_loss=0.4515, pruned_loss=0.1678, over 28856.00 frames. ], tot_loss[loss=0.4289, simple_loss=0.4585, pruned_loss=0.1997, over 5678407.07 frames. ], libri_tot_loss[loss=0.4585, simple_loss=0.4719, pruned_loss=0.2226, over 5709653.02 frames. ], giga_tot_loss[loss=0.4267, simple_loss=0.4572, pruned_loss=0.1981, over 5674390.76 frames. ], batch size: 119, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:49:12,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6386, 2.2416, 1.8360, 1.6597], device='cuda:1'), covar=tensor([0.1148, 0.1079, 0.0921, 0.0567], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0863, 0.0708, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 17:49:25,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8449, 2.1152, 1.8461, 1.1677], device='cuda:1'), covar=tensor([0.0500, 0.0395, 0.0527, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0943, 0.0928, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 17:49:39,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26936.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:49:50,980 INFO [train.py:968] (1/2) Epoch 1, batch 26950, giga_loss[loss=0.4576, simple_loss=0.4871, pruned_loss=0.214, over 28776.00 frames. ], tot_loss[loss=0.4327, simple_loss=0.462, pruned_loss=0.2017, over 5675505.70 frames. ], libri_tot_loss[loss=0.4591, simple_loss=0.4722, pruned_loss=0.223, over 5703059.37 frames. ], giga_tot_loss[loss=0.4296, simple_loss=0.4604, pruned_loss=0.1994, over 5676757.85 frames. ], batch size: 284, lr: 2.14e-02, grad_scale: 4.0
+2023-02-28 17:50:01,543 INFO [optim.py:369] (1/2) 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 17:50:36,307 INFO [train.py:968] (1/2) Epoch 1, batch 27000, giga_loss[loss=0.5574, simple_loss=0.515, pruned_loss=0.2999, over 23693.00 frames. ], tot_loss[loss=0.4388, simple_loss=0.4657, pruned_loss=0.2059, over 5669539.19 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4716, pruned_loss=0.2225, over 5698655.45 frames. ], giga_tot_loss[loss=0.4364, simple_loss=0.4647, pruned_loss=0.204, over 5673550.93 frames. ], batch size: 705, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:50:36,307 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 17:50:42,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2253, 1.2296, 1.1258, 1.1635], device='cuda:1'), covar=tensor([0.1985, 0.2074, 0.1665, 0.2000], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0821, 0.0863, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 17:50:44,992 INFO [train.py:1012] (1/2) Epoch 1, validation: loss=0.3094, simple_loss=0.3959, pruned_loss=0.1115, over 944034.00 frames. 
+2023-02-28 17:50:44,992 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19345MB
+2023-02-28 17:50:57,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27013.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:00,393 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27016.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:02,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27018.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:04,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-02-28 17:51:30,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27045.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:51:33,708 INFO [train.py:968] (1/2) Epoch 1, batch 27050, giga_loss[loss=0.4658, simple_loss=0.484, pruned_loss=0.2238, over 28861.00 frames. ], tot_loss[loss=0.4471, simple_loss=0.4702, pruned_loss=0.212, over 5665847.03 frames. ], libri_tot_loss[loss=0.4596, simple_loss=0.4726, pruned_loss=0.2233, over 5700784.17 frames. ], giga_tot_loss[loss=0.4435, simple_loss=0.4684, pruned_loss=0.2093, over 5666047.61 frames. ], batch size: 199, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:51:46,298 INFO [optim.py:369] (1/2) 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 17:51:58,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-02-28 17:52:02,373 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27079.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:52:06,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27082.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:52:24,565 INFO [train.py:968] (1/2) Epoch 1, batch 27100, giga_loss[loss=0.3909, simple_loss=0.4231, pruned_loss=0.1794, over 28910.00 frames. ], tot_loss[loss=0.4506, simple_loss=0.4713, pruned_loss=0.2149, over 5642150.16 frames. ], libri_tot_loss[loss=0.4591, simple_loss=0.4721, pruned_loss=0.223, over 5694429.19 frames. ], giga_tot_loss[loss=0.4479, simple_loss=0.4703, pruned_loss=0.2128, over 5647324.70 frames. ], batch size: 112, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:52:35,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27111.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:52:43,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6376, 1.9117, 1.5619, 1.4670], device='cuda:1'), covar=tensor([0.0749, 0.0955, 0.0735, 0.0497], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0878, 0.0710, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 17:53:04,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-02-28 17:53:10,858 INFO [train.py:968] (1/2) Epoch 1, batch 27150, libri_loss[loss=0.4534, simple_loss=0.4625, pruned_loss=0.2221, over 29552.00 frames. ], tot_loss[loss=0.4467, simple_loss=0.4686, pruned_loss=0.2124, over 5637866.90 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4713, pruned_loss=0.2226, over 5677649.39 frames. ], giga_tot_loss[loss=0.4448, simple_loss=0.4684, pruned_loss=0.2106, over 5655512.80 frames. ], batch size: 75, lr: 2.13e-02, grad_scale: 4.0
+2023-02-28 17:53:22,909 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27161.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:53:24,487 INFO [optim.py:369] (1/2) 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 17:53:27,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27164.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:53:42,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27182.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:53:55,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27193.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:54:01,038 INFO [train.py:968] (1/2) Epoch 1, batch 27200, giga_loss[loss=0.3402, simple_loss=0.4139, pruned_loss=0.1332, over 28791.00 frames. ], tot_loss[loss=0.445, simple_loss=0.4681, pruned_loss=0.2109, over 5628266.78 frames. ], libri_tot_loss[loss=0.4582, simple_loss=0.4712, pruned_loss=0.2226, over 5670020.19 frames. ], giga_tot_loss[loss=0.4435, simple_loss=0.468, pruned_loss=0.2095, over 5647465.34 frames. ], batch size: 99, lr: 2.13e-02, grad_scale: 8.0
+2023-02-28 17:54:12,823 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-02-28 17:54:55,082 INFO [train.py:968] (1/2) Epoch 1, batch 27250, giga_loss[loss=0.4009, simple_loss=0.4453, pruned_loss=0.1782, over 28720.00 frames. ], tot_loss[loss=0.4387, simple_loss=0.4657, pruned_loss=0.2059, over 5642051.30 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.4712, pruned_loss=0.2225, over 5671340.09 frames. ], giga_tot_loss[loss=0.4375, simple_loss=0.4656, pruned_loss=0.2047, over 5655626.47 frames. ], batch size: 66, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:55:09,521 INFO [optim.py:369] (1/2) 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 17:55:41,175 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=27297.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 17:55:42,914 INFO [train.py:968] (1/2) Epoch 1, batch 27300, libri_loss[loss=0.4971, simple_loss=0.5, pruned_loss=0.2471, over 29528.00 frames. ], tot_loss[loss=0.4395, simple_loss=0.4661, pruned_loss=0.2064, over 5650019.91 frames. ], libri_tot_loss[loss=0.4583, simple_loss=0.4713, pruned_loss=0.2226, over 5677242.42 frames. ], giga_tot_loss[loss=0.4379, simple_loss=0.4657, pruned_loss=0.205, over 5654623.59 frames. ], batch size: 82, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:55:50,549 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-02-28 17:56:05,334 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27325.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:07,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27328.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:24,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=27343.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:29,990 INFO [train.py:968] (1/2) Epoch 1, batch 27350, giga_loss[loss=0.4476, simple_loss=0.4646, pruned_loss=0.2153, over 27938.00 frames. ], tot_loss[loss=0.4401, simple_loss=0.466, pruned_loss=0.207, over 5663417.09 frames. ], libri_tot_loss[loss=0.4573, simple_loss=0.4704, pruned_loss=0.2221, over 5686314.00 frames. ], giga_tot_loss[loss=0.4388, simple_loss=0.4663, pruned_loss=0.2056, over 5658401.04 frames. ], batch size: 412, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:56:39,874 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27357.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:56:40,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-02-28 17:56:46,060 INFO [optim.py:369] (1/2) 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 17:57:14,527 INFO [train.py:968] (1/2) Epoch 1, batch 27400, giga_loss[loss=0.4267, simple_loss=0.4546, pruned_loss=0.1994, over 28572.00 frames. ], tot_loss[loss=0.4393, simple_loss=0.4648, pruned_loss=0.2069, over 5674801.70 frames. ], libri_tot_loss[loss=0.4567, simple_loss=0.4699, pruned_loss=0.2217, over 5691322.28 frames. ], giga_tot_loss[loss=0.4382, simple_loss=0.4654, pruned_loss=0.2055, over 5665526.13 frames. ], batch size: 336, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:58:10,696 INFO [train.py:968] (1/2) Epoch 1, batch 27450, libri_loss[loss=0.4034, simple_loss=0.4201, pruned_loss=0.1934, over 29653.00 frames. ], tot_loss[loss=0.4358, simple_loss=0.4607, pruned_loss=0.2055, over 5656990.43 frames. ], libri_tot_loss[loss=0.4557, simple_loss=0.4691, pruned_loss=0.2211, over 5695655.99 frames. ], giga_tot_loss[loss=0.4354, simple_loss=0.4616, pruned_loss=0.2046, over 5645011.60 frames. ], batch size: 69, lr: 2.12e-02, grad_scale: 2.0
+2023-02-28 17:58:23,649 INFO [optim.py:369] (1/2) 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 17:59:02,131 INFO [train.py:968] (1/2) Epoch 1, batch 27500, giga_loss[loss=0.3342, simple_loss=0.3963, pruned_loss=0.136, over 28565.00 frames. ], tot_loss[loss=0.435, simple_loss=0.4593, pruned_loss=0.2053, over 5650602.63 frames. ], libri_tot_loss[loss=0.4557, simple_loss=0.4692, pruned_loss=0.2211, over 5699749.68 frames. ], giga_tot_loss[loss=0.4343, simple_loss=0.4598, pruned_loss=0.2044, over 5636764.16 frames. ], batch size: 60, lr: 2.11e-02, grad_scale: 2.0
+2023-02-28 17:59:45,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=27541.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 17:59:52,525 INFO [train.py:968] (1/2) Epoch 1, batch 27550, giga_loss[loss=0.4453, simple_loss=0.4647, pruned_loss=0.2129, over 28309.00 frames. ], tot_loss[loss=0.4337, simple_loss=0.4578, pruned_loss=0.2048, over 5651760.82 frames. ], libri_tot_loss[loss=0.4563, simple_loss=0.4697, pruned_loss=0.2214, over 5692844.53 frames. ], giga_tot_loss[loss=0.4322, simple_loss=0.4576, pruned_loss=0.2035, over 5646774.38 frames. ], batch size: 368, lr: 2.11e-02, grad_scale: 2.0
+2023-02-28 18:00:05,356 INFO [optim.py:369] (1/2) 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 18:00:18,252 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7019, 1.7754, 1.8579, 1.6896], device='cuda:1'), covar=tensor([0.0668, 0.2083, 0.1221, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0892, 0.0670, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 18:00:39,534 INFO [train.py:968] (1/2) Epoch 1, batch 27600, giga_loss[loss=0.4423, simple_loss=0.4668, pruned_loss=0.2089, over 28277.00 frames. ], tot_loss[loss=0.4361, simple_loss=0.4588, pruned_loss=0.2067, over 5652299.53 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4693, pruned_loss=0.2209, over 5695964.33 frames. ], giga_tot_loss[loss=0.4351, simple_loss=0.4587, pruned_loss=0.2057, over 5644246.07 frames. ], batch size: 368, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:01:01,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8585, 2.2998, 1.6790, 1.5632], device='cuda:1'), covar=tensor([0.1010, 0.1102, 0.0931, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0862, 0.0709, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:01:23,393 INFO [train.py:968] (1/2) Epoch 1, batch 27650, giga_loss[loss=0.3738, simple_loss=0.4248, pruned_loss=0.1614, over 29113.00 frames. ], tot_loss[loss=0.4326, simple_loss=0.4564, pruned_loss=0.2044, over 5658882.99 frames. ], libri_tot_loss[loss=0.4547, simple_loss=0.4688, pruned_loss=0.2203, over 5702931.47 frames. ], giga_tot_loss[loss=0.432, simple_loss=0.4565, pruned_loss=0.2037, over 5644690.45 frames. ], batch size: 155, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:01:39,540 INFO [optim.py:369] (1/2) 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 18:01:46,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27672.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:01:47,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5935, 1.0902, 1.4767, 0.9847], device='cuda:1'), covar=tensor([0.0360, 0.0395, 0.0218, 0.0363], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0576, 0.0597, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 18:02:10,248 INFO [train.py:968] (1/2) Epoch 1, batch 27700, giga_loss[loss=0.3615, simple_loss=0.4117, pruned_loss=0.1556, over 28695.00 frames. ], tot_loss[loss=0.4262, simple_loss=0.4532, pruned_loss=0.1996, over 5664724.11 frames. ], libri_tot_loss[loss=0.455, simple_loss=0.469, pruned_loss=0.2205, over 5697096.19 frames. ], giga_tot_loss[loss=0.4248, simple_loss=0.4527, pruned_loss=0.1984, over 5657882.63 frames. ], batch size: 119, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:02:28,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27718.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:02:49,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6094, 1.9378, 1.7387, 1.5038], device='cuda:1'), covar=tensor([0.1113, 0.0846, 0.0986, 0.1752], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0491, 0.0398, 0.0498], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:1')
+2023-02-28 18:03:00,146 INFO [train.py:968] (1/2) Epoch 1, batch 27750, giga_loss[loss=0.442, simple_loss=0.4642, pruned_loss=0.2099, over 28883.00 frames. ], tot_loss[loss=0.4214, simple_loss=0.4503, pruned_loss=0.1963, over 5667250.92 frames. ], libri_tot_loss[loss=0.4548, simple_loss=0.4687, pruned_loss=0.2204, over 5700655.74 frames. ], giga_tot_loss[loss=0.4199, simple_loss=0.4498, pruned_loss=0.195, over 5657930.59 frames. ], batch size: 227, lr: 2.11e-02, grad_scale: 4.0
+2023-02-28 18:03:18,036 INFO [optim.py:369] (1/2) 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 18:03:49,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7723, 2.0017, 1.6733, 1.6030], device='cuda:1'), covar=tensor([0.0840, 0.0850, 0.0717, 0.0459], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0872, 0.0718, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:03:52,878 INFO [train.py:968] (1/2) Epoch 1, batch 27800, libri_loss[loss=0.463, simple_loss=0.4862, pruned_loss=0.2199, over 27676.00 frames. ], tot_loss[loss=0.4201, simple_loss=0.4488, pruned_loss=0.1957, over 5647218.01 frames. ], libri_tot_loss[loss=0.4548, simple_loss=0.4688, pruned_loss=0.2204, over 5700866.97 frames. ], giga_tot_loss[loss=0.4185, simple_loss=0.4482, pruned_loss=0.1945, over 5639335.65 frames. ], batch size: 116, lr: 2.10e-02, grad_scale: 4.0
+2023-02-28 18:03:56,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6808, 1.2659, 1.4953, 0.9964], device='cuda:1'), covar=tensor([0.0342, 0.0321, 0.0219, 0.0346], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0564, 0.0597, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 18:04:10,214 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27815.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:04:12,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27818.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:04:49,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27847.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:04:51,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5510, 1.9570, 1.7081, 1.5406], device='cuda:1'), covar=tensor([0.1074, 0.1223, 0.0901, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0864, 0.0724, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:04:52,713 INFO [train.py:968] (1/2) Epoch 1, batch 27850, giga_loss[loss=0.3698, simple_loss=0.4157, pruned_loss=0.1619, over 28852.00 frames. ], tot_loss[loss=0.4147, simple_loss=0.4441, pruned_loss=0.1926, over 5669661.72 frames. ], libri_tot_loss[loss=0.4552, simple_loss=0.4691, pruned_loss=0.2206, over 5702938.13 frames. ], giga_tot_loss[loss=0.4128, simple_loss=0.4432, pruned_loss=0.1912, over 5661270.67 frames. ], batch size: 112, lr: 2.10e-02, grad_scale: 2.0
+2023-02-28 18:05:03,706 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27861.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:05:09,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27864.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:05:10,010 INFO [optim.py:369] (1/2) 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 18:05:37,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:05:41,877 INFO [train.py:968] (1/2) Epoch 1, batch 27900, giga_loss[loss=0.4299, simple_loss=0.4694, pruned_loss=0.1951, over 29013.00 frames. ], tot_loss[loss=0.419, simple_loss=0.447, pruned_loss=0.1955, over 5660518.50 frames. ], libri_tot_loss[loss=0.4554, simple_loss=0.4693, pruned_loss=0.2208, over 5696108.10 frames. ], giga_tot_loss[loss=0.4165, simple_loss=0.4456, pruned_loss=0.1937, over 5659345.08 frames. ], batch size: 136, lr: 2.10e-02, grad_scale: 2.0
+2023-02-28 18:05:43,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6216, 2.0147, 1.6813, 1.6046], device='cuda:1'), covar=tensor([0.1037, 0.1141, 0.0912, 0.0563], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0865, 0.0724, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:05:53,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-02-28 18:05:56,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27916.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:06:29,422 INFO [train.py:968] (1/2) Epoch 1, batch 27950, giga_loss[loss=0.3912, simple_loss=0.4425, pruned_loss=0.17, over 28864.00 frames. ], tot_loss[loss=0.4202, simple_loss=0.4484, pruned_loss=0.196, over 5657571.70 frames. ], libri_tot_loss[loss=0.4554, simple_loss=0.4693, pruned_loss=0.2208, over 5702209.32 frames. ], giga_tot_loss[loss=0.4171, simple_loss=0.4466, pruned_loss=0.1938, over 5650353.53 frames. ], batch size: 174, lr: 2.10e-02, grad_scale: 2.0
+2023-02-28 18:06:46,525 INFO [optim.py:369] (1/2) 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 18:07:22,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7176, 1.7228, 1.3444, 1.4090], device='cuda:1'), covar=tensor([0.0763, 0.0893, 0.1137, 0.1136], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0613, 0.0600, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 18:07:23,049 INFO [train.py:968] (1/2) Epoch 1, batch 28000, giga_loss[loss=0.3443, simple_loss=0.394, pruned_loss=0.1473, over 28819.00 frames. ], tot_loss[loss=0.4172, simple_loss=0.4465, pruned_loss=0.194, over 5654975.60 frames. ], libri_tot_loss[loss=0.4556, simple_loss=0.4694, pruned_loss=0.2209, over 5704281.00 frames. ], giga_tot_loss[loss=0.4143, simple_loss=0.4448, pruned_loss=0.1919, over 5647063.12 frames. ], batch size: 99, lr: 2.10e-02, grad_scale: 4.0
+2023-02-28 18:08:06,011 INFO [train.py:968] (1/2) Epoch 1, batch 28050, giga_loss[loss=0.3922, simple_loss=0.4368, pruned_loss=0.1737, over 28856.00 frames. ], tot_loss[loss=0.4202, simple_loss=0.4486, pruned_loss=0.1959, over 5643277.42 frames. ], libri_tot_loss[loss=0.457, simple_loss=0.4706, pruned_loss=0.2217, over 5689574.59 frames. ], giga_tot_loss[loss=0.4153, simple_loss=0.4454, pruned_loss=0.1926, over 5647461.50 frames. ], batch size: 174, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:08:15,927 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28059.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:08:17,876 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28062.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:08:19,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7914, 2.7079, 3.4929, 1.6022], device='cuda:1'), covar=tensor([0.0627, 0.0749, 0.1006, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0497, 0.0846, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0006], device='cuda:1')
+2023-02-28 18:08:20,840 INFO [optim.py:369] (1/2) 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 18:08:31,773 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-02-28 18:08:42,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28091.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:08:50,306 INFO [train.py:968] (1/2) Epoch 1, batch 28100, giga_loss[loss=0.4401, simple_loss=0.4678, pruned_loss=0.2062, over 29006.00 frames. ], tot_loss[loss=0.4213, simple_loss=0.449, pruned_loss=0.1969, over 5636763.01 frames. ], libri_tot_loss[loss=0.4568, simple_loss=0.4703, pruned_loss=0.2216, over 5676756.71 frames. ], giga_tot_loss[loss=0.4168, simple_loss=0.4461, pruned_loss=0.1937, over 5650082.83 frames. ], batch size: 164, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:08:57,193 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28107.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:09:36,774 INFO [train.py:968] (1/2) Epoch 1, batch 28150, giga_loss[loss=0.4518, simple_loss=0.4717, pruned_loss=0.2159, over 29005.00 frames. ], tot_loss[loss=0.4265, simple_loss=0.4529, pruned_loss=0.2001, over 5646044.23 frames. ], libri_tot_loss[loss=0.4579, simple_loss=0.4711, pruned_loss=0.2223, over 5679664.40 frames. ], giga_tot_loss[loss=0.4213, simple_loss=0.4496, pruned_loss=0.1965, over 5653501.89 frames. ], batch size: 136, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:09:51,764 INFO [optim.py:369] (1/2) 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 18:10:22,033 INFO [train.py:968] (1/2) Epoch 1, batch 28200, libri_loss[loss=0.3942, simple_loss=0.4214, pruned_loss=0.1835, over 29436.00 frames. ], tot_loss[loss=0.43, simple_loss=0.4557, pruned_loss=0.2021, over 5663826.60 frames. ], libri_tot_loss[loss=0.4584, simple_loss=0.4714, pruned_loss=0.2227, over 5688015.08 frames. ], giga_tot_loss[loss=0.4238, simple_loss=0.452, pruned_loss=0.1978, over 5660787.17 frames. ], batch size: 70, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:10:40,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7921, 1.9149, 1.6672, 1.4594], device='cuda:1'), covar=tensor([0.0586, 0.0755, 0.0635, 0.0427], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0887, 0.0719, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:11:15,748 INFO [train.py:968] (1/2) Epoch 1, batch 28250, giga_loss[loss=0.45, simple_loss=0.443, pruned_loss=0.2285, over 23356.00 frames. ], tot_loss[loss=0.4341, simple_loss=0.4582, pruned_loss=0.205, over 5655068.28 frames. ], libri_tot_loss[loss=0.4581, simple_loss=0.471, pruned_loss=0.2225, over 5691388.81 frames. ], giga_tot_loss[loss=0.4291, simple_loss=0.4554, pruned_loss=0.2014, over 5649618.62 frames. ], batch size: 705, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:11:31,612 INFO [optim.py:369] (1/2) 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 18:12:04,772 INFO [train.py:968] (1/2) Epoch 1, batch 28300, giga_loss[loss=0.4536, simple_loss=0.4489, pruned_loss=0.2291, over 23564.00 frames. ], tot_loss[loss=0.4339, simple_loss=0.4576, pruned_loss=0.2051, over 5654599.07 frames. ], libri_tot_loss[loss=0.4578, simple_loss=0.4708, pruned_loss=0.2224, over 5695081.22 frames. ], giga_tot_loss[loss=0.4297, simple_loss=0.4553, pruned_loss=0.2021, over 5646108.36 frames. ], batch size: 705, lr: 2.09e-02, grad_scale: 4.0
+2023-02-28 18:12:29,800 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-02-28 18:12:57,615 INFO [train.py:968] (1/2) Epoch 1, batch 28350, giga_loss[loss=0.3712, simple_loss=0.4232, pruned_loss=0.1596, over 28890.00 frames. ], tot_loss[loss=0.4316, simple_loss=0.4576, pruned_loss=0.2028, over 5664313.13 frames. ], libri_tot_loss[loss=0.4568, simple_loss=0.4701, pruned_loss=0.2218, over 5700453.03 frames. ], giga_tot_loss[loss=0.4286, simple_loss=0.4561, pruned_loss=0.2005, over 5651982.04 frames. ], batch size: 213, lr: 2.08e-02, grad_scale: 4.0
+2023-02-28 18:13:12,484 INFO [optim.py:369] (1/2) 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 18:13:48,669 INFO [train.py:968] (1/2) Epoch 1, batch 28400, giga_loss[loss=0.4898, simple_loss=0.4802, pruned_loss=0.2497, over 26636.00 frames. ], tot_loss[loss=0.4321, simple_loss=0.4584, pruned_loss=0.2029, over 5661598.05 frames. ], libri_tot_loss[loss=0.4564, simple_loss=0.4697, pruned_loss=0.2216, over 5692504.53 frames. ], giga_tot_loss[loss=0.4297, simple_loss=0.4573, pruned_loss=0.201, over 5657724.58 frames. ], batch size: 555, lr: 2.08e-02, grad_scale: 8.0
+2023-02-28 18:14:41,379 INFO [train.py:968] (1/2) Epoch 1, batch 28450, giga_loss[loss=0.4191, simple_loss=0.4485, pruned_loss=0.1948, over 28931.00 frames. ], tot_loss[loss=0.4307, simple_loss=0.4564, pruned_loss=0.2025, over 5656787.57 frames. ], libri_tot_loss[loss=0.4568, simple_loss=0.4701, pruned_loss=0.2218, over 5685307.22 frames. ], giga_tot_loss[loss=0.428, simple_loss=0.4551, pruned_loss=0.2005, over 5660156.25 frames. ], batch size: 145, lr: 2.08e-02, grad_scale: 4.0
+2023-02-28 18:15:02,100 INFO [optim.py:369] (1/2) 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 18:15:19,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28482.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:15:40,615 INFO [train.py:968] (1/2) Epoch 1, batch 28500, giga_loss[loss=0.3639, simple_loss=0.4201, pruned_loss=0.1539, over 28913.00 frames. ], tot_loss[loss=0.4281, simple_loss=0.4545, pruned_loss=0.2009, over 5668587.18 frames. ], libri_tot_loss[loss=0.4566, simple_loss=0.4699, pruned_loss=0.2217, over 5689930.68 frames. ], giga_tot_loss[loss=0.4257, simple_loss=0.4533, pruned_loss=0.1991, over 5666621.97 frames. ], batch size: 174, lr: 2.08e-02, grad_scale: 4.0
+2023-02-28 18:16:28,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-02-28 18:16:38,447 INFO [train.py:968] (1/2) Epoch 1, batch 28550, libri_loss[loss=0.4839, simple_loss=0.4975, pruned_loss=0.2351, over 25886.00 frames. ], tot_loss[loss=0.4242, simple_loss=0.4515, pruned_loss=0.1984, over 5669173.17 frames. ], libri_tot_loss[loss=0.4563, simple_loss=0.4698, pruned_loss=0.2214, over 5689480.91 frames. ], giga_tot_loss[loss=0.4219, simple_loss=0.4503, pruned_loss=0.1967, over 5667804.04 frames. ], batch size: 136, lr: 2.08e-02, grad_scale: 2.0
+2023-02-28 18:16:56,436 INFO [optim.py:369] (1/2) 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 18:17:16,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5733, 1.8637, 1.6315, 0.4409], device='cuda:1'), covar=tensor([0.0983, 0.0742, 0.0835, 0.1587], device='cuda:1'), in_proj_covar=tensor([0.1002, 0.0990, 0.0982, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 18:17:27,513 INFO [train.py:968] (1/2) Epoch 1, batch 28600, giga_loss[loss=0.5484, simple_loss=0.5177, pruned_loss=0.2895, over 26596.00 frames. ], tot_loss[loss=0.4263, simple_loss=0.4525, pruned_loss=0.2, over 5670736.12 frames. ], libri_tot_loss[loss=0.456, simple_loss=0.4697, pruned_loss=0.2212, over 5690837.50 frames. ], giga_tot_loss[loss=0.4244, simple_loss=0.4515, pruned_loss=0.1987, over 5668386.40 frames. ], batch size: 555, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:17:27,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7264, 1.6319, 4.0709, 2.9720], device='cuda:1'), covar=tensor([0.1487, 0.1252, 0.0279, 0.0440], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0459, 0.0608, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 18:17:34,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.37 vs. limit=5.0
+2023-02-28 18:17:51,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28622.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:17:55,712 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28625.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:17:57,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28628.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:18:15,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28648.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:18:18,640 INFO [train.py:968] (1/2) Epoch 1, batch 28650, giga_loss[loss=0.4853, simple_loss=0.484, pruned_loss=0.2433, over 27569.00 frames. ], tot_loss[loss=0.4262, simple_loss=0.4518, pruned_loss=0.2002, over 5645159.65 frames. ], libri_tot_loss[loss=0.456, simple_loss=0.4696, pruned_loss=0.2212, over 5683914.92 frames. ], giga_tot_loss[loss=0.4241, simple_loss=0.4507, pruned_loss=0.1988, over 5649655.00 frames. ], batch size: 472, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:18:25,046 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:18:44,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4177, 1.7576, 1.4614, 1.4449], device='cuda:1'), covar=tensor([0.1374, 0.0587, 0.0704, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0336, 0.0324, 0.0543], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0014, 0.0022], device='cuda:1')
+2023-02-28 18:18:45,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28680.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:19:03,476 INFO [train.py:968] (1/2) Epoch 1, batch 28700, giga_loss[loss=0.4106, simple_loss=0.4431, pruned_loss=0.189, over 28926.00 frames. ], tot_loss[loss=0.4277, simple_loss=0.4532, pruned_loss=0.2011, over 5658471.52 frames. ], libri_tot_loss[loss=0.4555, simple_loss=0.4694, pruned_loss=0.2208, over 5692867.18 frames. ], giga_tot_loss[loss=0.4255, simple_loss=0.4519, pruned_loss=0.1995, over 5652945.29 frames. ], batch size: 112, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:19:55,844 INFO [train.py:968] (1/2) Epoch 1, batch 28750, giga_loss[loss=0.5415, simple_loss=0.5058, pruned_loss=0.2886, over 26607.00 frames. ], tot_loss[loss=0.4328, simple_loss=0.4563, pruned_loss=0.2046, over 5654657.00 frames. ], libri_tot_loss[loss=0.4553, simple_loss=0.4694, pruned_loss=0.2206, over 5696829.29 frames. ], giga_tot_loss[loss=0.4307, simple_loss=0.455, pruned_loss=0.2032, over 5645728.30 frames. ], batch size: 555, lr: 2.07e-02, grad_scale: 2.0
+2023-02-28 18:20:09,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4496, 1.6615, 1.4678, 1.5054], device='cuda:1'), covar=tensor([0.1280, 0.0520, 0.0692, 0.1626], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0330, 0.0326, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0014, 0.0022], device='cuda:1')
+2023-02-28 18:20:11,866 INFO [optim.py:369] (1/2) 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 18:20:44,070 INFO [train.py:968] (1/2) Epoch 1, batch 28800, giga_loss[loss=0.4081, simple_loss=0.4417, pruned_loss=0.1872, over 28342.00 frames. ], tot_loss[loss=0.4332, simple_loss=0.4568, pruned_loss=0.2048, over 5654222.48 frames. ], libri_tot_loss[loss=0.4549, simple_loss=0.4692, pruned_loss=0.2203, over 5698770.71 frames. ], giga_tot_loss[loss=0.4317, simple_loss=0.4558, pruned_loss=0.2038, over 5645170.63 frames. ], batch size: 368, lr: 2.07e-02, grad_scale: 4.0
+2023-02-28 18:21:18,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5171, 1.4929, 3.4908, 2.4894], device='cuda:1'), covar=tensor([0.1438, 0.1202, 0.0348, 0.0514], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0462, 0.0608, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 18:21:29,987 INFO [train.py:968] (1/2) Epoch 1, batch 28850, giga_loss[loss=0.4167, simple_loss=0.4408, pruned_loss=0.1963, over 28530.00 frames. ], tot_loss[loss=0.435, simple_loss=0.4572, pruned_loss=0.2064, over 5657904.60 frames. ], libri_tot_loss[loss=0.454, simple_loss=0.4686, pruned_loss=0.2197, over 5706749.20 frames. ], giga_tot_loss[loss=0.4337, simple_loss=0.4564, pruned_loss=0.2055, over 5641216.71 frames. ], batch size: 85, lr: 2.07e-02, grad_scale: 4.0
+2023-02-28 18:21:47,730 INFO [optim.py:369] (1/2) 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 18:21:56,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=28879.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:22:03,440 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 28900, giga_loss[loss=0.3996, simple_loss=0.436, pruned_loss=0.1816, over 28257.00 frames. ], tot_loss[loss=0.4353, simple_loss=0.4571, pruned_loss=0.2067, over 5665149.03 frames. ], libri_tot_loss[loss=0.4533, simple_loss=0.4681, pruned_loss=0.2192, over 5711899.29 frames. ], giga_tot_loss[loss=0.4344, simple_loss=0.4567, pruned_loss=0.2061, over 5645956.56 frames. ], batch size: 77, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:23:03,671 INFO [train.py:968] (1/2) Epoch 1, batch 28950, giga_loss[loss=0.4108, simple_loss=0.4469, pruned_loss=0.1874, over 28716.00 frames. ], tot_loss[loss=0.4365, simple_loss=0.4582, pruned_loss=0.2074, over 5658121.31 frames. ], libri_tot_loss[loss=0.4533, simple_loss=0.4683, pruned_loss=0.2191, over 5715757.74 frames. ], giga_tot_loss[loss=0.4352, simple_loss=0.4572, pruned_loss=0.2066, over 5637553.21 frames. ], batch size: 284, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:23:22,704 INFO [optim.py:369] (1/2) 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 18:23:52,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28997.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:23:55,426 INFO [train.py:968] (1/2) Epoch 1, batch 29000, giga_loss[loss=0.4171, simple_loss=0.4583, pruned_loss=0.188, over 29013.00 frames. ], tot_loss[loss=0.433, simple_loss=0.4568, pruned_loss=0.2046, over 5660224.94 frames. ], libri_tot_loss[loss=0.4529, simple_loss=0.468, pruned_loss=0.2189, over 5717478.53 frames. ], giga_tot_loss[loss=0.4322, simple_loss=0.4562, pruned_loss=0.2041, over 5642227.58 frames. ], batch size: 213, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:24:07,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-02-28 18:24:17,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29023.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:24:31,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0154, 2.6450, 2.0713, 1.9556], device='cuda:1'), covar=tensor([0.1025, 0.0955, 0.0779, 0.0517], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0866, 0.0711, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:24:42,538 INFO [train.py:968] (1/2) Epoch 1, batch 29050, giga_loss[loss=0.391, simple_loss=0.4362, pruned_loss=0.1729, over 28872.00 frames. ], tot_loss[loss=0.4335, simple_loss=0.4578, pruned_loss=0.2046, over 5659653.51 frames. ], libri_tot_loss[loss=0.4528, simple_loss=0.4681, pruned_loss=0.2188, over 5710120.27 frames. ], giga_tot_loss[loss=0.4324, simple_loss=0.457, pruned_loss=0.2039, over 5650344.67 frames. ], batch size: 199, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:24:47,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29055.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:25:00,136 INFO [optim.py:369] (1/2) 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 18:25:23,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2305, 2.9962, 3.9368, 1.6735], device='cuda:1'), covar=tensor([0.0517, 0.0635, 0.0882, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0504, 0.0857, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-02-28 18:25:29,615 INFO [train.py:968] (1/2) Epoch 1, batch 29100, giga_loss[loss=0.4673, simple_loss=0.4799, pruned_loss=0.2273, over 29087.00 frames. ], tot_loss[loss=0.434, simple_loss=0.4579, pruned_loss=0.2051, over 5670933.10 frames. ], libri_tot_loss[loss=0.4518, simple_loss=0.4673, pruned_loss=0.2181, over 5713882.09 frames. ], giga_tot_loss[loss=0.4337, simple_loss=0.4578, pruned_loss=0.2048, over 5659286.67 frames. ], batch size: 128, lr: 2.06e-02, grad_scale: 4.0
+2023-02-28 18:26:05,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29140.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:26:07,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29143.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:26:11,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-02-28 18:26:13,074 INFO [train.py:968] (1/2) Epoch 1, batch 29150, giga_loss[loss=0.3632, simple_loss=0.4119, pruned_loss=0.1572, over 29072.00 frames. ], tot_loss[loss=0.4359, simple_loss=0.4591, pruned_loss=0.2063, over 5675904.91 frames. ], libri_tot_loss[loss=0.4518, simple_loss=0.4673, pruned_loss=0.2182, over 5709819.84 frames. ], giga_tot_loss[loss=0.4352, simple_loss=0.4588, pruned_loss=0.2058, over 5669089.05 frames. ], batch size: 155, lr: 2.06e-02, grad_scale: 2.0
+2023-02-28 18:26:27,624 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:26:31,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29169.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 18:26:33,652 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29172.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:26:59,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29198.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 18:26:59,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29198.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:27:00,325 INFO [train.py:968] (1/2) Epoch 1, batch 29200, giga_loss[loss=0.4184, simple_loss=0.4468, pruned_loss=0.1949, over 28924.00 frames. ], tot_loss[loss=0.4367, simple_loss=0.4596, pruned_loss=0.2069, over 5670707.19 frames. ], libri_tot_loss[loss=0.4514, simple_loss=0.4668, pruned_loss=0.218, over 5713509.52 frames. ], giga_tot_loss[loss=0.4361, simple_loss=0.4596, pruned_loss=0.2063, over 5660853.45 frames. ], batch size: 213, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:27:03,065 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29230.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:27:54,782 INFO [train.py:968] (1/2) Epoch 1, batch 29250, giga_loss[loss=0.394, simple_loss=0.4423, pruned_loss=0.1728, over 28875.00 frames. ], tot_loss[loss=0.434, simple_loss=0.4591, pruned_loss=0.2044, over 5661209.47 frames. ], libri_tot_loss[loss=0.4512, simple_loss=0.4667, pruned_loss=0.2179, over 5705484.68 frames. ], giga_tot_loss[loss=0.4334, simple_loss=0.4591, pruned_loss=0.2039, over 5660294.32 frames. ], batch size: 174, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:27:58,520 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=29254.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:27:58,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29254.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:28:06,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29260.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:28:13,214 INFO [optim.py:369] (1/2) 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 18:28:39,957 INFO [train.py:968] (1/2) Epoch 1, batch 29300, giga_loss[loss=0.356, simple_loss=0.4056, pruned_loss=0.1532, over 28803.00 frames. ], tot_loss[loss=0.431, simple_loss=0.4573, pruned_loss=0.2024, over 5657907.23 frames. ], libri_tot_loss[loss=0.4508, simple_loss=0.4665, pruned_loss=0.2176, over 5699926.17 frames. ], giga_tot_loss[loss=0.4306, simple_loss=0.4573, pruned_loss=0.2019, over 5660566.74 frames. ], batch size: 119, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:29:04,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3159, 1.3064, 1.2567, 1.5129], device='cuda:1'), covar=tensor([0.1558, 0.1686, 0.1259, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0799, 0.0862, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 18:29:25,807 INFO [train.py:968] (1/2) Epoch 1, batch 29350, giga_loss[loss=0.4315, simple_loss=0.4581, pruned_loss=0.2024, over 28608.00 frames. ], tot_loss[loss=0.4283, simple_loss=0.4548, pruned_loss=0.2009, over 5657943.54 frames. ], libri_tot_loss[loss=0.4502, simple_loss=0.466, pruned_loss=0.2171, over 5702157.31 frames. ], giga_tot_loss[loss=0.428, simple_loss=0.4549, pruned_loss=0.2005, over 5656967.84 frames. ], batch size: 307, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:29:45,162 INFO [optim.py:369] (1/2) 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 18:30:08,419 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=29395.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:11,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:13,178 INFO [train.py:968] (1/2) Epoch 1, batch 29400, giga_loss[loss=0.4794, simple_loss=0.4873, pruned_loss=0.2358, over 28869.00 frames. ], tot_loss[loss=0.4318, simple_loss=0.4569, pruned_loss=0.2033, over 5654970.65 frames. ], libri_tot_loss[loss=0.4502, simple_loss=0.466, pruned_loss=0.2172, over 5701686.50 frames. ], giga_tot_loss[loss=0.4312, simple_loss=0.4569, pruned_loss=0.2027, over 5654313.49 frames. ], batch size: 145, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:30:15,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29400.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:18,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29403.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:21,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:44,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29429.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:30:50,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29435.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:31:07,665 INFO [train.py:968] (1/2) Epoch 1, batch 29450, giga_loss[loss=0.4591, simple_loss=0.4787, pruned_loss=0.2198, over 27605.00 frames. ], tot_loss[loss=0.4335, simple_loss=0.4581, pruned_loss=0.2044, over 5654824.01 frames. ], libri_tot_loss[loss=0.45, simple_loss=0.4658, pruned_loss=0.2171, over 5704747.84 frames. ], giga_tot_loss[loss=0.4329, simple_loss=0.4581, pruned_loss=0.2038, over 5650961.34 frames. ], batch size: 472, lr: 2.05e-02, grad_scale: 4.0
+2023-02-28 18:31:23,594 INFO [optim.py:369] (1/2) 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 18:31:39,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-02-28 18:31:53,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9365, 3.5482, 4.6910, 1.8538], device='cuda:1'), covar=tensor([0.0383, 0.0504, 0.0689, 0.1807], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0500, 0.0835, 0.0543], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0006], device='cuda:1')
+2023-02-28 18:31:55,898 INFO [train.py:968] (1/2) Epoch 1, batch 29500, giga_loss[loss=0.3641, simple_loss=0.4152, pruned_loss=0.1565, over 28902.00 frames. ], tot_loss[loss=0.4334, simple_loss=0.4574, pruned_loss=0.2047, over 5665573.85 frames. ], libri_tot_loss[loss=0.4499, simple_loss=0.4658, pruned_loss=0.217, over 5709404.51 frames. ], giga_tot_loss[loss=0.4327, simple_loss=0.4573, pruned_loss=0.204, over 5657400.68 frames. ], batch size: 164, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:32:44,256 INFO [train.py:968] (1/2) Epoch 1, batch 29550, giga_loss[loss=0.4196, simple_loss=0.4508, pruned_loss=0.1941, over 28516.00 frames. ], tot_loss[loss=0.4314, simple_loss=0.4553, pruned_loss=0.2038, over 5648550.03 frames. ], libri_tot_loss[loss=0.4498, simple_loss=0.4657, pruned_loss=0.217, over 5702238.57 frames. ], giga_tot_loss[loss=0.4307, simple_loss=0.4552, pruned_loss=0.2031, over 5647651.12 frames. ], batch size: 307, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:33:01,675 INFO [optim.py:369] (1/2) 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 18:33:08,667 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-02-28 18:33:32,474 INFO [train.py:968] (1/2) Epoch 1, batch 29600, giga_loss[loss=0.4321, simple_loss=0.4628, pruned_loss=0.2007, over 28884.00 frames. ], tot_loss[loss=0.432, simple_loss=0.4559, pruned_loss=0.204, over 5660952.14 frames. ], libri_tot_loss[loss=0.4499, simple_loss=0.4659, pruned_loss=0.217, over 5706720.03 frames. ], giga_tot_loss[loss=0.4309, simple_loss=0.4554, pruned_loss=0.2032, over 5654778.99 frames. ], batch size: 186, lr: 2.04e-02, grad_scale: 8.0
+2023-02-28 18:33:33,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9782, 1.5568, 1.4702, 1.4102], device='cuda:1'), covar=tensor([0.0684, 0.1572, 0.1288, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0526, 0.0878, 0.0665, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 18:34:01,596 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=29629.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:34:01,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29629.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:34:22,463 INFO [train.py:968] (1/2) Epoch 1, batch 29650, libri_loss[loss=0.4077, simple_loss=0.4347, pruned_loss=0.1904, over 29569.00 frames. ], tot_loss[loss=0.432, simple_loss=0.4563, pruned_loss=0.2038, over 5658157.17 frames. ], libri_tot_loss[loss=0.4492, simple_loss=0.4655, pruned_loss=0.2165, over 5709819.60 frames. ], giga_tot_loss[loss=0.4314, simple_loss=0.4561, pruned_loss=0.2034, over 5649338.05 frames. ], batch size: 76, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:34:42,436 INFO [optim.py:369] (1/2) 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 18:35:09,766 INFO [train.py:968] (1/2) Epoch 1, batch 29700, giga_loss[loss=0.3895, simple_loss=0.4288, pruned_loss=0.1751, over 28998.00 frames. ], tot_loss[loss=0.4282, simple_loss=0.4542, pruned_loss=0.2011, over 5656434.64 frames. ], libri_tot_loss[loss=0.4488, simple_loss=0.4653, pruned_loss=0.2162, over 5697568.44 frames. ], giga_tot_loss[loss=0.4277, simple_loss=0.454, pruned_loss=0.2007, over 5659239.54 frames. ], batch size: 136, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:35:48,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8751, 2.4095, 1.9171, 1.5964], device='cuda:1'), covar=tensor([0.0922, 0.0648, 0.0885, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0480, 0.0388, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:1')
+2023-02-28 18:35:59,987 INFO [train.py:968] (1/2) Epoch 1, batch 29750, giga_loss[loss=0.4152, simple_loss=0.4535, pruned_loss=0.1884, over 29000.00 frames. ], tot_loss[loss=0.4264, simple_loss=0.4536, pruned_loss=0.1996, over 5659105.15 frames. ], libri_tot_loss[loss=0.4487, simple_loss=0.4654, pruned_loss=0.216, over 5696903.11 frames. ], giga_tot_loss[loss=0.4258, simple_loss=0.4531, pruned_loss=0.1992, over 5661054.71 frames. ], batch size: 155, lr: 2.04e-02, grad_scale: 4.0
+2023-02-28 18:36:18,807 INFO [optim.py:369] (1/2) 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 18:36:21,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29770.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:36:22,435 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:36:24,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29775.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:36:31,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-02-28 18:36:46,677 INFO [train.py:968] (1/2) Epoch 1, batch 29800, giga_loss[loss=0.387, simple_loss=0.4314, pruned_loss=0.1713, over 28654.00 frames. ], tot_loss[loss=0.4286, simple_loss=0.455, pruned_loss=0.2011, over 5659240.61 frames. ], libri_tot_loss[loss=0.4482, simple_loss=0.4649, pruned_loss=0.2157, over 5703683.48 frames. ], giga_tot_loss[loss=0.4277, simple_loss=0.4546, pruned_loss=0.2004, over 5653041.44 frames. ], batch size: 60, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:36:51,459 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29804.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:37:04,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4435, 1.8650, 1.4517, 1.2459], device='cuda:1'), covar=tensor([0.1027, 0.0835, 0.0923, 0.1652], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0478, 0.0387, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:1')
+2023-02-28 18:37:34,392 INFO [train.py:968] (1/2) Epoch 1, batch 29850, giga_loss[loss=0.3946, simple_loss=0.4344, pruned_loss=0.1774, over 28779.00 frames. ], tot_loss[loss=0.4288, simple_loss=0.455, pruned_loss=0.2013, over 5663228.37 frames. ], libri_tot_loss[loss=0.4486, simple_loss=0.4651, pruned_loss=0.216, over 5707446.96 frames. ], giga_tot_loss[loss=0.4272, simple_loss=0.4543, pruned_loss=0.2001, over 5653539.49 frames. ], batch size: 99, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:37:53,237 INFO [optim.py:369] (1/2) 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 18:38:21,844 INFO [train.py:968] (1/2) Epoch 1, batch 29900, giga_loss[loss=0.4434, simple_loss=0.4638, pruned_loss=0.2115, over 28657.00 frames. ], tot_loss[loss=0.4285, simple_loss=0.4544, pruned_loss=0.2013, over 5670323.72 frames. ], libri_tot_loss[loss=0.449, simple_loss=0.4655, pruned_loss=0.2163, over 5708297.43 frames. ], giga_tot_loss[loss=0.4266, simple_loss=0.4534, pruned_loss=0.1999, over 5661423.21 frames. ], batch size: 242, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:38:37,051 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29913.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:38:39,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29916.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:39:03,861 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29945.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:39:08,397 INFO [train.py:968] (1/2) Epoch 1, batch 29950, giga_loss[loss=0.3916, simple_loss=0.4314, pruned_loss=0.1759, over 28634.00 frames. ], tot_loss[loss=0.4275, simple_loss=0.4531, pruned_loss=0.201, over 5665682.28 frames. ], libri_tot_loss[loss=0.4491, simple_loss=0.4656, pruned_loss=0.2163, over 5712194.18 frames. ], giga_tot_loss[loss=0.4253, simple_loss=0.4518, pruned_loss=0.1994, over 5653696.26 frames. ], batch size: 262, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:39:22,035 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8239, 1.6130, 1.7106, 1.5492], device='cuda:1'), covar=tensor([0.0715, 0.1350, 0.0825, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0891, 0.0669, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 18:39:27,431 INFO [optim.py:369] (1/2) 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 18:39:59,534 INFO [train.py:968] (1/2) Epoch 1, batch 30000, giga_loss[loss=0.4486, simple_loss=0.4625, pruned_loss=0.2173, over 28953.00 frames. ], tot_loss[loss=0.4229, simple_loss=0.4492, pruned_loss=0.1984, over 5674981.10 frames. ], libri_tot_loss[loss=0.4496, simple_loss=0.4661, pruned_loss=0.2165, over 5713324.16 frames. ], giga_tot_loss[loss=0.4202, simple_loss=0.4474, pruned_loss=0.1965, over 5663359.42 frames. ], batch size: 164, lr: 2.03e-02, grad_scale: 8.0
+2023-02-28 18:39:59,535 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 18:40:08,175 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-02-28 18:40:12,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30004.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:40:46,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5994, 1.3426, 1.4132, 0.6214], device='cuda:1'), covar=tensor([0.0321, 0.0280, 0.0214, 0.0339], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0592, 0.0614, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 18:40:53,662 INFO [train.py:968] (1/2) Epoch 1, batch 30050, giga_loss[loss=0.3955, simple_loss=0.4255, pruned_loss=0.1827, over 28901.00 frames. ], tot_loss[loss=0.42, simple_loss=0.4463, pruned_loss=0.1969, over 5684582.94 frames. ], libri_tot_loss[loss=0.4495, simple_loss=0.466, pruned_loss=0.2165, over 5707390.84 frames. ], giga_tot_loss[loss=0.4175, simple_loss=0.4446, pruned_loss=0.1952, over 5679366.34 frames. ], batch size: 106, lr: 2.03e-02, grad_scale: 4.0
+2023-02-28 18:40:59,367 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:41:42,109 INFO [train.py:968] (1/2) Epoch 1, batch 30100, giga_loss[loss=0.4003, simple_loss=0.4385, pruned_loss=0.1811, over 28238.00 frames. ], tot_loss[loss=0.4211, simple_loss=0.4466, pruned_loss=0.1978, over 5678372.49 frames. ], libri_tot_loss[loss=0.4492, simple_loss=0.4658, pruned_loss=0.2163, over 5692745.95 frames. ], giga_tot_loss[loss=0.4186, simple_loss=0.445, pruned_loss=0.1961, over 5686362.17 frames. ], batch size: 368, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:42:20,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-02-28 18:42:32,681 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30147.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:42:34,348 INFO [train.py:968] (1/2) Epoch 1, batch 30150, libri_loss[loss=0.4636, simple_loss=0.4675, pruned_loss=0.2298, over 19063.00 frames. ], tot_loss[loss=0.4178, simple_loss=0.4449, pruned_loss=0.1954, over 5660087.25 frames. ], libri_tot_loss[loss=0.4494, simple_loss=0.466, pruned_loss=0.2165, over 5675649.98 frames. ], giga_tot_loss[loss=0.4154, simple_loss=0.4433, pruned_loss=0.1937, over 5681918.91 frames. ], batch size: 187, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:42:35,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30150.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:42:55,664 INFO [optim.py:369] (1/2) 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 18:43:03,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3922, 3.3928, 4.1350, 1.9536], device='cuda:1'), covar=tensor([0.0433, 0.0471, 0.0809, 0.1577], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0503, 0.0842, 0.0550], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-02-28 18:43:05,681 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30179.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:43:21,503 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30195.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:43:27,708 INFO [train.py:968] (1/2) Epoch 1, batch 30200, giga_loss[loss=0.3622, simple_loss=0.4193, pruned_loss=0.1525, over 28894.00 frames. ], tot_loss[loss=0.4106, simple_loss=0.4407, pruned_loss=0.1902, over 5654954.67 frames. ], libri_tot_loss[loss=0.4483, simple_loss=0.4649, pruned_loss=0.2158, over 5669058.52 frames. ], giga_tot_loss[loss=0.4088, simple_loss=0.4398, pruned_loss=0.1889, over 5678016.55 frames. ], batch size: 199, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:43:58,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6284, 3.5357, 4.3939, 2.0229], device='cuda:1'), covar=tensor([0.0386, 0.0484, 0.0771, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.0753, 0.0493, 0.0825, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0006], device='cuda:1')
+2023-02-28 18:44:23,852 INFO [train.py:968] (1/2) Epoch 1, batch 30250, giga_loss[loss=0.3385, simple_loss=0.3948, pruned_loss=0.1412, over 28884.00 frames. ], tot_loss[loss=0.4022, simple_loss=0.4353, pruned_loss=0.1846, over 5633302.17 frames. ], libri_tot_loss[loss=0.4481, simple_loss=0.4647, pruned_loss=0.2157, over 5658515.06 frames. ], giga_tot_loss[loss=0.4005, simple_loss=0.4344, pruned_loss=0.1833, over 5660952.11 frames. ], batch size: 186, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:44:26,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 18:44:30,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7573, 2.7984, 1.9720, 1.7568], device='cuda:1'), covar=tensor([0.1581, 0.1175, 0.1059, 0.0689], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0839, 0.0712, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:44:44,882 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 30300, giga_loss[loss=0.3803, simple_loss=0.4085, pruned_loss=0.1761, over 26763.00 frames. ], tot_loss[loss=0.3941, simple_loss=0.4298, pruned_loss=0.1792, over 5633091.94 frames. ], libri_tot_loss[loss=0.4477, simple_loss=0.4643, pruned_loss=0.2156, over 5662732.42 frames. ], giga_tot_loss[loss=0.3922, simple_loss=0.429, pruned_loss=0.1777, over 5650783.94 frames. ], batch size: 555, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:45:35,415 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30319.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:46:01,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9225, 2.8984, 3.6340, 1.6716], device='cuda:1'), covar=tensor([0.0631, 0.0621, 0.0951, 0.2001], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0494, 0.0822, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0006], device='cuda:1')
+2023-02-28 18:46:06,785 INFO [train.py:968] (1/2) Epoch 1, batch 30350, giga_loss[loss=0.3451, simple_loss=0.4022, pruned_loss=0.144, over 28542.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.424, pruned_loss=0.1729, over 5642385.46 frames. ], libri_tot_loss[loss=0.4469, simple_loss=0.4636, pruned_loss=0.2151, over 5666971.64 frames. ], giga_tot_loss[loss=0.3831, simple_loss=0.4232, pruned_loss=0.1714, over 5652035.61 frames. ], batch size: 336, lr: 2.02e-02, grad_scale: 4.0
+2023-02-28 18:46:24,736 INFO [optim.py:369] (1/2) 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 18:46:57,034 INFO [train.py:968] (1/2) Epoch 1, batch 30400, giga_loss[loss=0.308, simple_loss=0.3791, pruned_loss=0.1185, over 28562.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.4195, pruned_loss=0.1667, over 5641075.72 frames. ], libri_tot_loss[loss=0.4458, simple_loss=0.4625, pruned_loss=0.2145, over 5671155.36 frames. ], giga_tot_loss[loss=0.3748, simple_loss=0.4191, pruned_loss=0.1653, over 5644778.61 frames. ], batch size: 85, lr: 2.01e-02, grad_scale: 8.0
+2023-02-28 18:47:28,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30430.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:47:37,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7850, 2.1830, 1.6232, 1.5679], device='cuda:1'), covar=tensor([0.1122, 0.1071, 0.0901, 0.0636], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0845, 0.0721, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 18:47:49,249 INFO [train.py:968] (1/2) Epoch 1, batch 30450, giga_loss[loss=0.3613, simple_loss=0.4143, pruned_loss=0.1541, over 28636.00 frames. ], tot_loss[loss=0.3773, simple_loss=0.42, pruned_loss=0.1672, over 5643272.43 frames. ], libri_tot_loss[loss=0.4447, simple_loss=0.4616, pruned_loss=0.2139, over 5678250.26 frames. ], giga_tot_loss[loss=0.3746, simple_loss=0.4192, pruned_loss=0.165, over 5638813.36 frames. ], batch size: 85, lr: 2.01e-02, grad_scale: 4.0
+2023-02-28 18:47:55,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6826, 1.9312, 1.6752, 1.4778], device='cuda:1'), covar=tensor([0.1000, 0.0753, 0.0880, 0.1560], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0469, 0.0389, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:1')
+2023-02-28 18:48:02,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30462.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:48:06,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30465.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:48:12,255 INFO [optim.py:369] (1/2) 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 18:48:34,200 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 30500, giga_loss[loss=0.3925, simple_loss=0.4133, pruned_loss=0.1859, over 26669.00 frames. ], tot_loss[loss=0.3769, simple_loss=0.4193, pruned_loss=0.1673, over 5632775.72 frames. ], libri_tot_loss[loss=0.4445, simple_loss=0.4611, pruned_loss=0.2139, over 5672105.38 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.4179, pruned_loss=0.1642, over 5633958.41 frames. ], batch size: 555, lr: 2.01e-02, grad_scale: 4.0
+2023-02-28 18:48:41,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5101, 1.3144, 1.2455, 1.3140], device='cuda:1'), covar=tensor([0.1699, 0.1941, 0.1549, 0.2524], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0758, 0.0860, 0.0912], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 18:49:00,531 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 30550, giga_loss[loss=0.3648, simple_loss=0.4141, pruned_loss=0.1577, over 28975.00 frames. ], tot_loss[loss=0.3698, simple_loss=0.4143, pruned_loss=0.1626, over 5637352.64 frames. ], libri_tot_loss[loss=0.4446, simple_loss=0.4611, pruned_loss=0.214, over 5674626.69 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4128, pruned_loss=0.1597, over 5635462.16 frames. ], batch size: 213, lr: 2.01e-02, grad_scale: 2.0
+2023-02-28 18:49:50,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30570.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:49:53,926 INFO [optim.py:369] (1/2) 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 18:49:54,751 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30573.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:49:57,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30576.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:50:20,759 INFO [train.py:968] (1/2) Epoch 1, batch 30600, giga_loss[loss=0.3189, simple_loss=0.3836, pruned_loss=0.1271, over 28961.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.4107, pruned_loss=0.1603, over 5640203.06 frames. ], libri_tot_loss[loss=0.4426, simple_loss=0.4592, pruned_loss=0.213, over 5679201.91 frames. ], giga_tot_loss[loss=0.3615, simple_loss=0.4093, pruned_loss=0.1568, over 5633187.70 frames. ], batch size: 106, lr: 2.01e-02, grad_scale: 2.0
+2023-02-28 18:50:22,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30602.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:50:24,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-02-28 18:50:27,136 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30605.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:50:40,496 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-02-28 18:50:43,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2879, 1.4448, 1.1846, 1.3206], device='cuda:1'), covar=tensor([0.1221, 0.0686, 0.0699, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0318, 0.0320, 0.0532], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0016, 0.0014, 0.0023], device='cuda:1')
+2023-02-28 18:50:55,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4995, 2.0029, 1.6383, 0.5861], device='cuda:1'), covar=tensor([0.0894, 0.0776, 0.1005, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0992, 0.1045, 0.1006, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-02-28 18:51:01,205 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-02-28 18:51:11,698 INFO [train.py:968] (1/2) Epoch 1, batch 30650, giga_loss[loss=0.3332, simple_loss=0.3902, pruned_loss=0.1381, over 28626.00 frames. ], tot_loss[loss=0.3669, simple_loss=0.412, pruned_loss=0.1609, over 5648190.38 frames. ], libri_tot_loss[loss=0.4422, simple_loss=0.4588, pruned_loss=0.2128, over 5683566.93 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4104, pruned_loss=0.1573, over 5637939.87 frames. ], batch size: 336, lr: 2.01e-02, grad_scale: 2.0
+2023-02-28 18:51:31,558 INFO [optim.py:369] (1/2) 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 18:51:33,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3228, 3.3148, 4.1064, 1.7742], device='cuda:1'), covar=tensor([0.0463, 0.0516, 0.0719, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0491, 0.0813, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 18:51:59,921 INFO [train.py:968] (1/2) Epoch 1, batch 30700, giga_loss[loss=0.3136, simple_loss=0.378, pruned_loss=0.1246, over 28959.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4102, pruned_loss=0.1591, over 5654860.01 frames. ], libri_tot_loss[loss=0.4414, simple_loss=0.4581, pruned_loss=0.2123, over 5691138.72 frames. ], giga_tot_loss[loss=0.359, simple_loss=0.4081, pruned_loss=0.155, over 5638736.41 frames. ], batch size: 145, lr: 2.00e-02, grad_scale: 2.0
+2023-02-28 18:52:10,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30713.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:52:13,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30716.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:52:44,731 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30745.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:52:48,736 INFO [train.py:968] (1/2) Epoch 1, batch 30750, giga_loss[loss=0.2952, simple_loss=0.3779, pruned_loss=0.1062, over 28911.00 frames. ], tot_loss[loss=0.3589, simple_loss=0.4066, pruned_loss=0.1556, over 5664774.47 frames. ], libri_tot_loss[loss=0.4399, simple_loss=0.4567, pruned_loss=0.2116, over 5696981.33 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4047, pruned_loss=0.1515, over 5646156.64 frames. ], batch size: 174, lr: 2.00e-02, grad_scale: 2.0
+2023-02-28 18:52:52,599 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=30753.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:53:11,722 INFO [optim.py:369] (1/2) 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 18:53:39,518 INFO [train.py:968] (1/2) Epoch 1, batch 30800, giga_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.1181, over 28424.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4022, pruned_loss=0.1522, over 5645672.25 frames. ], libri_tot_loss[loss=0.4397, simple_loss=0.4566, pruned_loss=0.2114, over 5688824.93 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3998, pruned_loss=0.1479, over 5636946.94 frames. ], batch size: 65, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:54:31,570 INFO [train.py:968] (1/2) Epoch 1, batch 30850, giga_loss[loss=0.3499, simple_loss=0.395, pruned_loss=0.1524, over 27981.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4002, pruned_loss=0.1515, over 5641793.55 frames. ], libri_tot_loss[loss=0.439, simple_loss=0.4561, pruned_loss=0.211, over 5682950.42 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3975, pruned_loss=0.1472, over 5638599.53 frames. ], batch size: 412, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:54:53,590 INFO [optim.py:369] (1/2) 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 18:55:17,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:55:25,890 INFO [train.py:968] (1/2) Epoch 1, batch 30900, giga_loss[loss=0.3059, simple_loss=0.3439, pruned_loss=0.134, over 24083.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3976, pruned_loss=0.1506, over 5633997.79 frames. ], libri_tot_loss[loss=0.4381, simple_loss=0.4552, pruned_loss=0.2104, over 5685073.13 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3956, pruned_loss=0.1469, over 5629237.12 frames. ], batch size: 705, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:56:22,315 INFO [train.py:968] (1/2) Epoch 1, batch 30950, giga_loss[loss=0.3973, simple_loss=0.4215, pruned_loss=0.1866, over 26512.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.3984, pruned_loss=0.1511, over 5622995.50 frames. ], libri_tot_loss[loss=0.4379, simple_loss=0.455, pruned_loss=0.2105, over 5684126.07 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3965, pruned_loss=0.1477, over 5619601.72 frames. ], batch size: 555, lr: 2.00e-02, grad_scale: 4.0
+2023-02-28 18:56:33,956 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-02-28 18:56:49,874 INFO [optim.py:369] (1/2) 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 18:56:55,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30977.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:57:22,077 INFO [train.py:968] (1/2) Epoch 1, batch 31000, giga_loss[loss=0.3629, simple_loss=0.4211, pruned_loss=0.1524, over 28534.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4028, pruned_loss=0.1524, over 5636212.93 frames. ], libri_tot_loss[loss=0.437, simple_loss=0.4542, pruned_loss=0.2099, over 5687419.38 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4013, pruned_loss=0.1493, over 5630005.15 frames. ], batch size: 336, lr: 2.00e-02, grad_scale: 2.0
+2023-02-28 18:58:04,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31036.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:06,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31039.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:16,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31048.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:17,713 INFO [train.py:968] (1/2) Epoch 1, batch 31050, giga_loss[loss=0.3076, simple_loss=0.3779, pruned_loss=0.1187, over 28939.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4054, pruned_loss=0.154, over 5653117.12 frames. ], libri_tot_loss[loss=0.4366, simple_loss=0.4537, pruned_loss=0.2098, over 5691369.93 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4027, pruned_loss=0.1495, over 5643070.13 frames. ], batch size: 174, lr: 1.99e-02, grad_scale: 2.0
+2023-02-28 18:58:45,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31068.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:58:52,181 INFO [optim.py:369] (1/2) 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 18:58:58,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7127, 2.0719, 1.7379, 0.3558], device='cuda:1'), covar=tensor([0.0894, 0.0753, 0.0901, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0989, 0.1052, 0.0996, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 18:59:04,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1958, 3.1212, 3.9227, 1.9768], device='cuda:1'), covar=tensor([0.0600, 0.0606, 0.1068, 0.1780], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0489, 0.0797, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 18:59:20,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7655, 1.5974, 1.6043, 1.5087], device='cuda:1'), covar=tensor([0.0764, 0.1421, 0.0859, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0862, 0.0656, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 18:59:27,822 INFO [train.py:968] (1/2) Epoch 1, batch 31100, giga_loss[loss=0.2935, simple_loss=0.3363, pruned_loss=0.1254, over 24663.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.403, pruned_loss=0.1516, over 5664369.86 frames. ], libri_tot_loss[loss=0.4356, simple_loss=0.4528, pruned_loss=0.2092, over 5695429.15 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.4008, pruned_loss=0.1476, over 5652268.36 frames. ], batch size: 705, lr: 1.99e-02, grad_scale: 2.0
+2023-02-28 18:59:55,230 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31120.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 18:59:58,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31123.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:00:05,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:00:33,752 INFO [train.py:968] (1/2) Epoch 1, batch 31150, giga_loss[loss=0.369, simple_loss=0.411, pruned_loss=0.1635, over 26709.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.399, pruned_loss=0.1476, over 5659449.97 frames. ], libri_tot_loss[loss=0.4352, simple_loss=0.4525, pruned_loss=0.209, over 5696509.00 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3973, pruned_loss=0.1444, over 5648913.95 frames. ], batch size: 555, lr: 1.99e-02, grad_scale: 2.0
+2023-02-28 19:00:37,829 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31152.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:01:07,885 INFO [optim.py:369] (1/2) 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 19:01:41,057 INFO [train.py:968] (1/2) Epoch 1, batch 31200, giga_loss[loss=0.3351, simple_loss=0.3946, pruned_loss=0.1378, over 28817.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3961, pruned_loss=0.144, over 5664528.38 frames. ], libri_tot_loss[loss=0.4343, simple_loss=0.4517, pruned_loss=0.2085, over 5697805.90 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3947, pruned_loss=0.141, over 5654738.30 frames. ], batch size: 227, lr: 1.99e-02, grad_scale: 4.0
+2023-02-28 19:01:47,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6152, 2.1429, 1.6430, 1.6328], device='cuda:1'), covar=tensor([0.1200, 0.1076, 0.0996, 0.0593], device='cuda:1'), in_proj_covar=tensor([0.0830, 0.0823, 0.0728, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 19:01:51,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5333, 2.0431, 1.4512, 1.4161], device='cuda:1'), covar=tensor([0.0963, 0.0911, 0.0964, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0818, 0.0724, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 19:02:36,881 INFO [train.py:968] (1/2) Epoch 1, batch 31250, giga_loss[loss=0.3197, simple_loss=0.3769, pruned_loss=0.1313, over 28980.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3944, pruned_loss=0.1448, over 5661504.00 frames. ], libri_tot_loss[loss=0.4332, simple_loss=0.4506, pruned_loss=0.2078, over 5692613.45 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3917, pruned_loss=0.1401, over 5657106.37 frames. ], batch size: 186, lr: 1.99e-02, grad_scale: 4.0
+2023-02-28 19:03:02,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:05,486 INFO [optim.py:369] (1/2) 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 19:03:06,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31274.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:33,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31296.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:37,945 INFO [train.py:968] (1/2) Epoch 1, batch 31300, giga_loss[loss=0.3353, simple_loss=0.3895, pruned_loss=0.1406, over 28947.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3933, pruned_loss=0.1449, over 5664887.74 frames. ], libri_tot_loss[loss=0.4323, simple_loss=0.4499, pruned_loss=0.2073, over 5697908.96 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3902, pruned_loss=0.1398, over 5655969.75 frames. ], batch size: 145, lr: 1.99e-02, grad_scale: 4.0
+2023-02-28 19:03:40,667 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31303.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:03:57,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31318.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:04:06,723 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31327.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:04:29,917 INFO [train.py:968] (1/2) Epoch 1, batch 31350, giga_loss[loss=0.3435, simple_loss=0.4038, pruned_loss=0.1416, over 28957.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3949, pruned_loss=0.1484, over 5671964.72 frames. ], libri_tot_loss[loss=0.4329, simple_loss=0.4499, pruned_loss=0.208, over 5702932.65 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3901, pruned_loss=0.141, over 5658768.72 frames. ], batch size: 155, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:04:34,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7150, 2.1484, 1.7164, 1.6444], device='cuda:1'), covar=tensor([0.1013, 0.0935, 0.0822, 0.0520], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0795, 0.0698, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 19:04:47,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1980, 1.5696, 1.4525, 1.2212], device='cuda:1'), covar=tensor([0.1505, 0.0613, 0.0717, 0.1878], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0313, 0.0309, 0.0524], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0016, 0.0015, 0.0023], device='cuda:1')
+2023-02-28 19:05:00,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 19:05:00,429 INFO [optim.py:369] (1/2) 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 19:05:27,606 INFO [train.py:968] (1/2) Epoch 1, batch 31400, giga_loss[loss=0.3381, simple_loss=0.3995, pruned_loss=0.1384, over 28362.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.397, pruned_loss=0.1486, over 5661611.21 frames. ], libri_tot_loss[loss=0.4323, simple_loss=0.4493, pruned_loss=0.2076, over 5696329.43 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3924, pruned_loss=0.1418, over 5656541.70 frames. ], batch size: 368, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:05:32,833 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31404.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:05:34,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5140, 1.4045, 1.3168, 1.0285], device='cuda:1'), covar=tensor([0.0332, 0.0307, 0.0191, 0.0329], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0547, 0.0570, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 19:05:57,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31423.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:06:31,639 INFO [train.py:968] (1/2) Epoch 1, batch 31450, giga_loss[loss=0.3211, simple_loss=0.3864, pruned_loss=0.1279, over 28435.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.3983, pruned_loss=0.1482, over 5666606.79 frames. ], libri_tot_loss[loss=0.4313, simple_loss=0.4486, pruned_loss=0.207, over 5697766.40 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3939, pruned_loss=0.1416, over 5660629.38 frames. ], batch size: 336, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:06:34,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3925, 1.3038, 1.1734, 1.3888], device='cuda:1'), covar=tensor([0.1313, 0.1439, 0.1162, 0.1577], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0745, 0.0842, 0.0901], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 19:06:45,448 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-02-28 19:06:50,291 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-02-28 19:06:58,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8201, 2.9517, 3.6027, 1.8077], device='cuda:1'), covar=tensor([0.0558, 0.0494, 0.0819, 0.1683], device='cuda:1'), in_proj_covar=tensor([0.0716, 0.0482, 0.0781, 0.0521], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 19:07:01,497 INFO [optim.py:369] (1/2) 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 19:07:15,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-02-28 19:07:35,068 INFO [train.py:968] (1/2) Epoch 1, batch 31500, giga_loss[loss=0.3714, simple_loss=0.4167, pruned_loss=0.1631, over 28928.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3932, pruned_loss=0.1446, over 5667801.34 frames. ], libri_tot_loss[loss=0.4308, simple_loss=0.4482, pruned_loss=0.2067, over 5703746.48 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3886, pruned_loss=0.1379, over 5656975.69 frames. ], batch size: 284, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:08:41,009 INFO [train.py:968] (1/2) Epoch 1, batch 31550, giga_loss[loss=0.3495, simple_loss=0.3818, pruned_loss=0.1586, over 24827.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3935, pruned_loss=0.145, over 5670319.87 frames. ], libri_tot_loss[loss=0.4288, simple_loss=0.4465, pruned_loss=0.2056, over 5701314.84 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3897, pruned_loss=0.1388, over 5662133.42 frames. ], batch size: 705, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:09:03,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2855, 1.2638, 1.1116, 1.3660], device='cuda:1'), covar=tensor([0.1387, 0.0659, 0.0743, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0310, 0.0304, 0.0514], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0016, 0.0014, 0.0023], device='cuda:1')
+2023-02-28 19:09:04,825 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31566.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:09:08,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:09:15,073 INFO [optim.py:369] (1/2) 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 19:09:47,306 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31598.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:09:49,079 INFO [train.py:968] (1/2) Epoch 1, batch 31600, giga_loss[loss=0.3276, simple_loss=0.4011, pruned_loss=0.1271, over 28070.00 frames. ], tot_loss[loss=0.3444, simple_loss=0.3973, pruned_loss=0.1458, over 5662217.55 frames. ], libri_tot_loss[loss=0.4287, simple_loss=0.4464, pruned_loss=0.2056, over 5702366.27 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3941, pruned_loss=0.1407, over 5654782.76 frames. ], batch size: 412, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:10:47,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-02-28 19:10:56,089 INFO [train.py:968] (1/2) Epoch 1, batch 31650, giga_loss[loss=0.3293, simple_loss=0.4017, pruned_loss=0.1285, over 28626.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3992, pruned_loss=0.1439, over 5667427.08 frames. ], libri_tot_loss[loss=0.4284, simple_loss=0.4462, pruned_loss=0.2054, over 5704474.05 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3961, pruned_loss=0.139, over 5659042.71 frames. ], batch size: 262, lr: 1.98e-02, grad_scale: 2.0
+2023-02-28 19:11:00,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4452, 1.8415, 1.4880, 0.5607], device='cuda:1'), covar=tensor([0.0713, 0.0640, 0.0902, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0982, 0.1037, 0.0981, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 19:11:26,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31671.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:11:31,025 INFO [optim.py:369] (1/2) 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 19:11:52,341 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31693.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:11:52,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2377, 1.8632, 1.8084, 1.6113], device='cuda:1'), covar=tensor([0.0676, 0.1455, 0.0960, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0867, 0.0651, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 19:12:00,533 INFO [train.py:968] (1/2) Epoch 1, batch 31700, giga_loss[loss=0.3195, simple_loss=0.3913, pruned_loss=0.1239, over 29028.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3987, pruned_loss=0.1423, over 5658744.21 frames. ], libri_tot_loss[loss=0.429, simple_loss=0.4466, pruned_loss=0.2057, over 5706554.32 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.395, pruned_loss=0.137, over 5649664.30 frames. ], batch size: 285, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:12:02,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31702.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:13:01,581 INFO [train.py:968] (1/2) Epoch 1, batch 31750, giga_loss[loss=0.2979, simple_loss=0.3672, pruned_loss=0.1143, over 27609.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3965, pruned_loss=0.1397, over 5659952.65 frames. ], libri_tot_loss[loss=0.4285, simple_loss=0.4462, pruned_loss=0.2054, over 5709524.25 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3933, pruned_loss=0.1349, over 5649601.26 frames. ], batch size: 472, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:13:36,894 INFO [optim.py:369] (1/2) 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 19:13:40,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31779.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:14:03,335 INFO [train.py:968] (1/2) Epoch 1, batch 31800, giga_loss[loss=0.3793, simple_loss=0.4173, pruned_loss=0.1706, over 27641.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3981, pruned_loss=0.1424, over 5657534.61 frames. ], libri_tot_loss[loss=0.4279, simple_loss=0.4458, pruned_loss=0.205, over 5705511.99 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3944, pruned_loss=0.1369, over 5651383.55 frames. ], batch size: 472, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:14:07,791 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31802.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:14:23,421 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31817.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:14:54,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31836.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:00,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31839.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:10,363 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31845.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:11,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31846.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:14,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31848.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:15:16,415 INFO [train.py:968] (1/2) Epoch 1, batch 31850, giga_loss[loss=0.3698, simple_loss=0.4185, pruned_loss=0.1606, over 28393.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.399, pruned_loss=0.1443, over 5663717.12 frames. ], libri_tot_loss[loss=0.4274, simple_loss=0.4454, pruned_loss=0.2048, over 5707383.57 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3956, pruned_loss=0.1391, over 5656306.69 frames. ], batch size: 368, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:15:46,354 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:15:59,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31877.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:16:16,238 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=31884.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:16:38,955 INFO [train.py:968] (1/2) Epoch 1, batch 31900, giga_loss[loss=0.3289, simple_loss=0.3814, pruned_loss=0.1382, over 28390.00 frames. ], tot_loss[loss=0.347, simple_loss=0.4011, pruned_loss=0.1465, over 5665263.60 frames. ], libri_tot_loss[loss=0.4272, simple_loss=0.4451, pruned_loss=0.2046, over 5700053.81 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3979, pruned_loss=0.1415, over 5664500.18 frames. ], batch size: 369, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:17:10,190 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31922.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:17:17,249 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 31950, giga_loss[loss=0.3161, simple_loss=0.3806, pruned_loss=0.1258, over 28838.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3959, pruned_loss=0.143, over 5665412.26 frames. ], libri_tot_loss[loss=0.4265, simple_loss=0.4446, pruned_loss=0.2042, over 5698146.06 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3925, pruned_loss=0.1378, over 5665708.31 frames. ], batch size: 145, lr: 1.97e-02, grad_scale: 2.0
+2023-02-28 19:17:55,196 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31954.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:18:21,741 INFO [optim.py:369] (1/2) 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 19:18:55,723 INFO [train.py:968] (1/2) Epoch 1, batch 32000, giga_loss[loss=0.2972, simple_loss=0.3663, pruned_loss=0.114, over 28605.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3945, pruned_loss=0.1422, over 5668052.22 frames. ], libri_tot_loss[loss=0.4257, simple_loss=0.444, pruned_loss=0.2037, over 5702199.75 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3914, pruned_loss=0.1374, over 5664287.91 frames. ], batch size: 307, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:19:35,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3817, 1.8755, 1.4542, 1.2379], device='cuda:1'), covar=tensor([0.1111, 0.0862, 0.1055, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0459, 0.0382, 0.0478], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:1')
+2023-02-28 19:19:51,981 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.28 vs. limit=5.0
+2023-02-28 19:19:59,198 INFO [train.py:968] (1/2) Epoch 1, batch 32050, giga_loss[loss=0.3868, simple_loss=0.4264, pruned_loss=0.1736, over 28455.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3954, pruned_loss=0.1446, over 5669123.62 frames. ], libri_tot_loss[loss=0.4247, simple_loss=0.4432, pruned_loss=0.2031, over 5706112.75 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.392, pruned_loss=0.1392, over 5661373.15 frames. ], batch size: 370, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:20:34,638 INFO [optim.py:369] (1/2) 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 19:21:03,736 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32097.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:21:05,554 INFO [train.py:968] (1/2) Epoch 1, batch 32100, libri_loss[loss=0.4134, simple_loss=0.4403, pruned_loss=0.1932, over 29531.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3994, pruned_loss=0.1463, over 5673242.83 frames. ], libri_tot_loss[loss=0.4238, simple_loss=0.4425, pruned_loss=0.2026, over 5707393.93 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3963, pruned_loss=0.1412, over 5665174.11 frames. ], batch size: 89, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:21:32,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2842, 1.6646, 1.3347, 1.4154], device='cuda:1'), covar=tensor([0.1401, 0.0601, 0.0699, 0.1747], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0306, 0.0304, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0016, 0.0015, 0.0024], device='cuda:1')
+2023-02-28 19:21:55,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-02-28 19:22:05,569 INFO [train.py:968] (1/2) Epoch 1, batch 32150, giga_loss[loss=0.298, simple_loss=0.3405, pruned_loss=0.1277, over 24489.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3994, pruned_loss=0.147, over 5666009.76 frames. ], libri_tot_loss[loss=0.4232, simple_loss=0.442, pruned_loss=0.2022, over 5703440.96 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3962, pruned_loss=0.1418, over 5662353.64 frames. ], batch size: 705, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:22:23,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-02-28 19:22:39,196 INFO [optim.py:369] (1/2) 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 19:22:39,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8310, 2.1693, 1.8266, 1.7665], device='cuda:1'), covar=tensor([0.0901, 0.0738, 0.0830, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0449, 0.0378, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0012], device='cuda:1')
+2023-02-28 19:22:40,479 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32177.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:22:43,758 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32181.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:22:53,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7483, 1.5244, 1.3634, 1.4594], device='cuda:1'), covar=tensor([0.0589, 0.1178, 0.1064, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0860, 0.0635, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 19:23:07,302 INFO [train.py:968] (1/2) Epoch 1, batch 32200, giga_loss[loss=0.3253, simple_loss=0.3873, pruned_loss=0.1316, over 28375.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3982, pruned_loss=0.1473, over 5656251.84 frames. ], libri_tot_loss[loss=0.422, simple_loss=0.441, pruned_loss=0.2015, over 5692657.80 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3951, pruned_loss=0.1418, over 5661212.53 frames. ], batch size: 368, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:23:18,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6339, 2.1803, 1.6518, 1.5569], device='cuda:1'), covar=tensor([0.1159, 0.1129, 0.0964, 0.0661], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0811, 0.0701, 0.0622], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 19:23:26,179 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32216.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:24:06,994 INFO [train.py:968] (1/2) Epoch 1, batch 32250, giga_loss[loss=0.3617, simple_loss=0.4076, pruned_loss=0.1579, over 28629.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.398, pruned_loss=0.1477, over 5653736.38 frames. ], libri_tot_loss[loss=0.421, simple_loss=0.4402, pruned_loss=0.2009, over 5690051.65 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.395, pruned_loss=0.1423, over 5659753.16 frames. ], batch size: 307, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:24:16,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-02-28 19:24:20,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32259.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 19:24:39,562 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:24:44,465 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 32300, libri_loss[loss=0.3919, simple_loss=0.4193, pruned_loss=0.1822, over 25805.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3994, pruned_loss=0.1474, over 5658632.42 frames. ], libri_tot_loss[loss=0.4208, simple_loss=0.4401, pruned_loss=0.2008, over 5690384.36 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3963, pruned_loss=0.1422, over 5662807.73 frames. ], batch size: 136, lr: 1.96e-02, grad_scale: 4.0
+2023-02-28 19:25:29,314 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32308.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:25:31,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3026, 1.5225, 1.2719, 1.2264], device='cuda:1'), covar=tensor([0.1029, 0.0852, 0.0967, 0.1526], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0468, 0.0386, 0.0486], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0012, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 19:25:48,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32320.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:25:54,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32323.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:26:28,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=32343.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:26:37,410 INFO [train.py:968] (1/2) Epoch 1, batch 32350, libri_loss[loss=0.3693, simple_loss=0.3918, pruned_loss=0.1734, over 29364.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.4003, pruned_loss=0.1471, over 5661877.79 frames. ], libri_tot_loss[loss=0.4205, simple_loss=0.4395, pruned_loss=0.2007, over 5686298.52 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3976, pruned_loss=0.1419, over 5667423.81 frames. ], batch size: 71, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:26:39,855 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:27:17,917 INFO [optim.py:369] (1/2) 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 19:27:34,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3878, 1.3075, 1.4169, 0.7009], device='cuda:1'), covar=tensor([0.0367, 0.0319, 0.0199, 0.0421], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0552, 0.0603, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 19:27:53,110 INFO [train.py:968] (1/2) Epoch 1, batch 32400, giga_loss[loss=0.357, simple_loss=0.4053, pruned_loss=0.1544, over 28100.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.398, pruned_loss=0.1453, over 5656237.55 frames. ], libri_tot_loss[loss=0.4197, simple_loss=0.4388, pruned_loss=0.2003, over 5681170.30 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3956, pruned_loss=0.1403, over 5664651.62 frames. ], batch size: 412, lr: 1.95e-02, grad_scale: 8.0
+2023-02-28 19:27:56,245 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32402.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:27:58,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32405.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:28:38,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32434.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 19:28:57,865 INFO [train.py:968] (1/2) Epoch 1, batch 32450, giga_loss[loss=0.2649, simple_loss=0.3296, pruned_loss=0.1001, over 28085.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3922, pruned_loss=0.143, over 5654038.66 frames. ], libri_tot_loss[loss=0.4186, simple_loss=0.438, pruned_loss=0.1996, over 5677362.12 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3898, pruned_loss=0.1381, over 5664364.35 frames. ], batch size: 412, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:29:28,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32472.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:29:33,186 INFO [optim.py:369] (1/2) 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 19:30:06,128 INFO [train.py:968] (1/2) Epoch 1, batch 32500, giga_loss[loss=0.266, simple_loss=0.3209, pruned_loss=0.1056, over 24171.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3847, pruned_loss=0.1387, over 5655307.60 frames. ], libri_tot_loss[loss=0.4183, simple_loss=0.4378, pruned_loss=0.1994, over 5682013.49 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3822, pruned_loss=0.1339, over 5659230.72 frames. ], batch size: 705, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:31:11,963 INFO [train.py:968] (1/2) Epoch 1, batch 32550, giga_loss[loss=0.336, simple_loss=0.3902, pruned_loss=0.1409, over 28082.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3856, pruned_loss=0.1398, over 5654432.00 frames. ], libri_tot_loss[loss=0.4178, simple_loss=0.4374, pruned_loss=0.1991, over 5684225.74 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3835, pruned_loss=0.1357, over 5655200.53 frames. ], batch size: 412, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:31:20,625 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:31:59,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32591.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:32:09,941 INFO [train.py:968] (1/2) Epoch 1, batch 32600, libri_loss[loss=0.3885, simple_loss=0.4102, pruned_loss=0.1834, over 29543.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3888, pruned_loss=0.1428, over 5655939.61 frames. ], libri_tot_loss[loss=0.417, simple_loss=0.4367, pruned_loss=0.1987, over 5690965.47 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.386, pruned_loss=0.138, over 5649516.19 frames. ], batch size: 77, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:32:27,609 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32615.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:32:31,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32618.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:32:40,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2306, 1.2903, 1.1841, 1.1913], device='cuda:1'), covar=tensor([0.0646, 0.0703, 0.0962, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0857, 0.0659, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 19:33:09,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32646.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:33:11,981 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32647.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:33:14,353 INFO [train.py:968] (1/2) Epoch 1, batch 32650, giga_loss[loss=0.3154, simple_loss=0.3837, pruned_loss=0.1236, over 28945.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3854, pruned_loss=0.139, over 5656220.66 frames. ], libri_tot_loss[loss=0.417, simple_loss=0.4367, pruned_loss=0.1987, over 5690965.47 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3833, pruned_loss=0.1354, over 5651221.23 frames. ], batch size: 199, lr: 1.95e-02, grad_scale: 4.0
+2023-02-28 19:33:19,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 19:33:48,694 INFO [optim.py:369] (1/2) 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 19:33:55,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32683.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:34:18,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 32700, giga_loss[loss=0.3053, simple_loss=0.3718, pruned_loss=0.1194, over 28603.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3843, pruned_loss=0.1374, over 5665661.11 frames. ], libri_tot_loss[loss=0.4167, simple_loss=0.4366, pruned_loss=0.1984, over 5694140.26 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3813, pruned_loss=0.1331, over 5657971.23 frames. ], batch size: 307, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:34:21,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32702.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:34:43,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32718.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:00,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32731.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:05,163 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32734.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:08,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32737.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:10,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0820, 1.1813, 1.0775, 1.0365], device='cuda:1'), covar=tensor([0.1524, 0.1485, 0.1231, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0739, 0.0829, 0.0898], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 19:35:23,859 INFO [train.py:968] (1/2) Epoch 1, batch 32750, giga_loss[loss=0.3089, simple_loss=0.3674, pruned_loss=0.1252, over 29033.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3843, pruned_loss=0.1388, over 5665441.80 frames. ], libri_tot_loss[loss=0.4163, simple_loss=0.4362, pruned_loss=0.1982, over 5702225.73 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3802, pruned_loss=0.1332, over 5650460.21 frames. ], batch size: 165, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:35:44,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32766.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:35:57,924 INFO [optim.py:369] (1/2) 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 19:36:15,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32789.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:36:17,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32792.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:36:27,009 INFO [train.py:968] (1/2) Epoch 1, batch 32800, giga_loss[loss=0.2935, simple_loss=0.3684, pruned_loss=0.1093, over 28432.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3863, pruned_loss=0.1399, over 5651108.50 frames. ], libri_tot_loss[loss=0.4172, simple_loss=0.4367, pruned_loss=0.1988, over 5687180.44 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3813, pruned_loss=0.1333, over 5652158.25 frames. ], batch size: 60, lr: 1.94e-02, grad_scale: 8.0
+2023-02-28 19:36:28,882 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-02-28 19:36:52,160 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32821.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:36:59,065 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32826.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:03,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32829.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:31,541 INFO [train.py:968] (1/2) Epoch 1, batch 32850, giga_loss[loss=0.317, simple_loss=0.3786, pruned_loss=0.1276, over 28332.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3865, pruned_loss=0.1401, over 5651745.04 frames. ], libri_tot_loss[loss=0.4175, simple_loss=0.437, pruned_loss=0.1991, over 5687605.54 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3807, pruned_loss=0.1329, over 5651274.69 frames. ], batch size: 368, lr: 1.94e-02, grad_scale: 8.0
+2023-02-28 19:37:42,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32858.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:46,859 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32861.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:37:50,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32864.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:38:08,959 INFO [optim.py:369] (1/2) 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 19:38:30,501 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:38:36,661 INFO [train.py:968] (1/2) Epoch 1, batch 32900, giga_loss[loss=0.3232, simple_loss=0.3827, pruned_loss=0.1318, over 28926.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3881, pruned_loss=0.1418, over 5651163.44 frames. ], libri_tot_loss[loss=0.4182, simple_loss=0.4373, pruned_loss=0.1995, over 5680616.49 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3827, pruned_loss=0.135, over 5655691.95 frames. ], batch size: 284, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:39:38,860 INFO [train.py:968] (1/2) Epoch 1, batch 32950, giga_loss[loss=0.3421, simple_loss=0.4175, pruned_loss=0.1333, over 28956.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3858, pruned_loss=0.1396, over 5644136.88 frames. ], libri_tot_loss[loss=0.4181, simple_loss=0.4373, pruned_loss=0.1995, over 5675829.20 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3805, pruned_loss=0.1329, over 5651225.21 frames. ], batch size: 186, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:40:10,558 INFO [optim.py:369] (1/2) 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 19:40:16,770 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-02-28 19:40:28,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3511, 1.7524, 2.1724, 1.1057], device='cuda:1'), covar=tensor([0.0611, 0.0521, 0.0936, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0475, 0.0757, 0.0522], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 19:40:34,562 INFO [train.py:968] (1/2) Epoch 1, batch 33000, giga_loss[loss=0.3536, simple_loss=0.4183, pruned_loss=0.1444, over 28893.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.389, pruned_loss=0.1404, over 5638913.44 frames. ], libri_tot_loss[loss=0.4181, simple_loss=0.4371, pruned_loss=0.1995, over 5664011.75 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.383, pruned_loss=0.1328, over 5654422.12 frames. ], batch size: 145, lr: 1.94e-02, grad_scale: 4.0
+2023-02-28 19:40:34,563 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 19:40:42,966 INFO [train.py:1012] (1/2) Epoch 1, validation: loss=0.2718, simple_loss=0.3588, pruned_loss=0.0924, over 944034.00 frames. 
+2023-02-28 19:40:42,966 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-02-28 19:41:14,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.44 vs. limit=5.0
+2023-02-28 19:41:20,552 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7532, 1.6297, 1.2104, 1.2737], device='cuda:1'), covar=tensor([0.0765, 0.0733, 0.0979, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0554, 0.0577, 0.0531], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0008, 0.0008], device='cuda:1')
+2023-02-28 19:41:44,286 INFO [train.py:968] (1/2) Epoch 1, batch 33050, giga_loss[loss=0.321, simple_loss=0.3956, pruned_loss=0.1232, over 28710.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3916, pruned_loss=0.141, over 5644744.01 frames. ], libri_tot_loss[loss=0.4176, simple_loss=0.4368, pruned_loss=0.1992, over 5668043.26 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3863, pruned_loss=0.1342, over 5653022.50 frames. ], batch size: 243, lr: 1.93e-02, grad_scale: 4.0
+2023-02-28 19:42:20,258 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 33100, giga_loss[loss=0.3281, simple_loss=0.3899, pruned_loss=0.1331, over 28525.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3914, pruned_loss=0.1414, over 5640469.17 frames. ], libri_tot_loss[loss=0.4166, simple_loss=0.4359, pruned_loss=0.1987, over 5671436.93 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3872, pruned_loss=0.1353, over 5643642.09 frames. ], batch size: 336, lr: 1.93e-02, grad_scale: 4.0
+2023-02-28 19:43:30,783 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-02-28 19:43:44,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0193, 1.2261, 0.8682, 0.9442], device='cuda:1'), covar=tensor([0.0714, 0.0539, 0.1103, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0563, 0.0593, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 19:43:52,862 INFO [train.py:968] (1/2) Epoch 1, batch 33150, giga_loss[loss=0.387, simple_loss=0.4336, pruned_loss=0.1701, over 28819.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3921, pruned_loss=0.1418, over 5649813.26 frames. ], libri_tot_loss[loss=0.4163, simple_loss=0.4358, pruned_loss=0.1984, over 5675050.99 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3878, pruned_loss=0.1358, over 5648405.76 frames. ], batch size: 243, lr: 1.93e-02, grad_scale: 4.0
+2023-02-28 19:44:25,591 INFO [optim.py:369] (1/2) 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 19:44:53,632 INFO [train.py:968] (1/2) Epoch 1, batch 33200, giga_loss[loss=0.3265, simple_loss=0.3834, pruned_loss=0.1348, over 29032.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.389, pruned_loss=0.1389, over 5658040.33 frames. ], libri_tot_loss[loss=0.4164, simple_loss=0.4358, pruned_loss=0.1985, over 5676581.35 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3851, pruned_loss=0.1336, over 5655482.66 frames. ], batch size: 155, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:45:50,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-02-28 19:45:58,708 INFO [train.py:968] (1/2) Epoch 1, batch 33250, giga_loss[loss=0.2672, simple_loss=0.3348, pruned_loss=0.09984, over 29122.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3871, pruned_loss=0.1382, over 5660836.85 frames. ], libri_tot_loss[loss=0.4161, simple_loss=0.4353, pruned_loss=0.1985, over 5677219.12 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3837, pruned_loss=0.133, over 5657949.66 frames. ], batch size: 120, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:46:32,407 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 33300, giga_loss[loss=0.3153, simple_loss=0.3823, pruned_loss=0.1241, over 28701.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3858, pruned_loss=0.1377, over 5657068.59 frames. ], libri_tot_loss[loss=0.4162, simple_loss=0.4354, pruned_loss=0.1985, over 5671268.70 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3818, pruned_loss=0.1322, over 5660442.54 frames. ], batch size: 262, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:47:35,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6436, 1.5648, 1.5217, 1.0757], device='cuda:1'), covar=tensor([0.0431, 0.0294, 0.0216, 0.0388], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0532, 0.0595, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 19:48:00,017 INFO [train.py:968] (1/2) Epoch 1, batch 33350, giga_loss[loss=0.34, simple_loss=0.4079, pruned_loss=0.1361, over 28644.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3887, pruned_loss=0.1395, over 5663350.27 frames. ], libri_tot_loss[loss=0.4153, simple_loss=0.4347, pruned_loss=0.1979, over 5675491.22 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3849, pruned_loss=0.1339, over 5662084.73 frames. ], batch size: 307, lr: 1.93e-02, grad_scale: 8.0
+2023-02-28 19:48:09,897 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33359.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:48:11,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1113, 1.8228, 1.6454, 1.6649], device='cuda:1'), covar=tensor([0.0786, 0.1590, 0.1137, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0520, 0.0866, 0.0644, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 19:48:39,083 INFO [optim.py:369] (1/2) 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 19:49:03,231 INFO [train.py:968] (1/2) Epoch 1, batch 33400, giga_loss[loss=0.3604, simple_loss=0.4176, pruned_loss=0.1516, over 28448.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3905, pruned_loss=0.1411, over 5667631.86 frames. ], libri_tot_loss[loss=0.4143, simple_loss=0.4341, pruned_loss=0.1973, over 5678337.30 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3867, pruned_loss=0.1357, over 5664034.98 frames. ], batch size: 336, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:49:38,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3330, 2.3814, 1.7528, 0.9971], device='cuda:1'), covar=tensor([0.0394, 0.0205, 0.0231, 0.0380], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0536, 0.0608, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 19:50:09,142 INFO [train.py:968] (1/2) Epoch 1, batch 33450, giga_loss[loss=0.3881, simple_loss=0.4275, pruned_loss=0.1743, over 28616.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3915, pruned_loss=0.1421, over 5663559.08 frames. ], libri_tot_loss[loss=0.4132, simple_loss=0.4332, pruned_loss=0.1966, over 5674789.73 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3882, pruned_loss=0.1369, over 5662750.05 frames. ], batch size: 307, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:50:29,344 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33462.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:50:52,322 INFO [optim.py:369] (1/2) 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 19:51:18,845 INFO [train.py:968] (1/2) Epoch 1, batch 33500, giga_loss[loss=0.3784, simple_loss=0.4344, pruned_loss=0.1612, over 28626.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3957, pruned_loss=0.1443, over 5662972.88 frames. ], libri_tot_loss[loss=0.4134, simple_loss=0.4333, pruned_loss=0.1967, over 5678368.36 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3924, pruned_loss=0.1393, over 5659024.33 frames. ], batch size: 242, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:52:10,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33548.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:52:12,029 INFO [train.py:968] (1/2) Epoch 1, batch 33550, giga_loss[loss=0.3162, simple_loss=0.391, pruned_loss=0.1207, over 28922.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3978, pruned_loss=0.1451, over 5668192.49 frames. ], libri_tot_loss[loss=0.4126, simple_loss=0.4328, pruned_loss=0.1962, over 5687308.46 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3943, pruned_loss=0.1396, over 5656388.42 frames. ], batch size: 164, lr: 1.92e-02, grad_scale: 4.0
+2023-02-28 19:52:52,480 INFO [optim.py:369] (1/2) 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 19:53:22,760 INFO [train.py:968] (1/2) Epoch 1, batch 33600, giga_loss[loss=0.2931, simple_loss=0.3313, pruned_loss=0.1275, over 24671.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3968, pruned_loss=0.1444, over 5669495.19 frames. ], libri_tot_loss[loss=0.4125, simple_loss=0.4327, pruned_loss=0.1962, over 5691735.65 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3935, pruned_loss=0.1392, over 5656107.99 frames. ], batch size: 705, lr: 1.92e-02, grad_scale: 8.0
+2023-02-28 19:54:09,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-02-28 19:54:38,629 INFO [train.py:968] (1/2) Epoch 1, batch 33650, giga_loss[loss=0.3565, simple_loss=0.3915, pruned_loss=0.1607, over 26903.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.394, pruned_loss=0.1428, over 5670741.72 frames. ], libri_tot_loss[loss=0.4129, simple_loss=0.4329, pruned_loss=0.1965, over 5692392.97 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.391, pruned_loss=0.1383, over 5659623.84 frames. ], batch size: 555, lr: 1.92e-02, grad_scale: 8.0
+2023-02-28 19:55:17,463 INFO [optim.py:369] (1/2) 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 19:55:18,032 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.44 vs. limit=5.0
+2023-02-28 19:55:43,869 INFO [train.py:968] (1/2) Epoch 1, batch 33700, libri_loss[loss=0.4037, simple_loss=0.4318, pruned_loss=0.1878, over 29083.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3944, pruned_loss=0.1436, over 5670246.15 frames. ], libri_tot_loss[loss=0.4128, simple_loss=0.4328, pruned_loss=0.1964, over 5698514.89 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3911, pruned_loss=0.1385, over 5655017.13 frames. ], batch size: 101, lr: 1.92e-02, grad_scale: 8.0
+2023-02-28 19:56:23,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0528, 2.3120, 4.9666, 3.3346], device='cuda:1'), covar=tensor([0.1405, 0.1069, 0.0204, 0.0337], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0459, 0.0575, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 19:56:27,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33734.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:56:43,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4035, 1.7434, 1.3344, 1.3169], device='cuda:1'), covar=tensor([0.1046, 0.0779, 0.1048, 0.1508], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0453, 0.0382, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 19:56:50,257 INFO [train.py:968] (1/2) Epoch 1, batch 33750, giga_loss[loss=0.4055, simple_loss=0.4309, pruned_loss=0.1901, over 28090.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3939, pruned_loss=0.144, over 5659389.38 frames. ], libri_tot_loss[loss=0.4123, simple_loss=0.4324, pruned_loss=0.1961, over 5699083.76 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3907, pruned_loss=0.1391, over 5645978.37 frames. ], batch size: 412, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 19:57:29,585 INFO [optim.py:369] (1/2) 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 19:57:55,025 INFO [train.py:968] (1/2) Epoch 1, batch 33800, libri_loss[loss=0.3998, simple_loss=0.4283, pruned_loss=0.1856, over 28708.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3904, pruned_loss=0.1422, over 5659991.76 frames. ], libri_tot_loss[loss=0.4113, simple_loss=0.4317, pruned_loss=0.1954, over 5699146.55 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3874, pruned_loss=0.1374, over 5647714.80 frames. ], batch size: 106, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 19:58:19,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 19:58:41,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33837.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:58:45,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7707, 2.4916, 1.9134, 1.8203], device='cuda:1'), covar=tensor([0.1472, 0.1243, 0.1012, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0825, 0.0711, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 19:58:57,513 INFO [train.py:968] (1/2) Epoch 1, batch 33850, giga_loss[loss=0.2792, simple_loss=0.3597, pruned_loss=0.09936, over 28492.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3895, pruned_loss=0.1412, over 5657578.65 frames. ], libri_tot_loss[loss=0.4107, simple_loss=0.4313, pruned_loss=0.1951, over 5703054.93 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3867, pruned_loss=0.1367, over 5643567.29 frames. ], batch size: 336, lr: 1.91e-02, grad_scale: 2.0
+2023-02-28 19:59:16,096 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=33865.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:59:31,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33877.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:59:36,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33880.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 19:59:36,357 INFO [optim.py:369] (1/2) 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 19:59:59,562 INFO [train.py:968] (1/2) Epoch 1, batch 33900, giga_loss[loss=0.3206, simple_loss=0.387, pruned_loss=0.1271, over 28609.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3876, pruned_loss=0.1384, over 5671740.08 frames. ], libri_tot_loss[loss=0.4102, simple_loss=0.4309, pruned_loss=0.1947, over 5706534.99 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3847, pruned_loss=0.1338, over 5656337.85 frames. ], batch size: 307, lr: 1.91e-02, grad_scale: 2.0
+2023-02-28 20:00:12,252 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33909.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:00:21,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0126, 1.0899, 0.9904, 0.8284], device='cuda:1'), covar=tensor([0.1670, 0.1748, 0.1519, 0.1737], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0761, 0.0847, 0.0904], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 20:00:28,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33923.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:00:30,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3539, 1.2972, 1.2944, 0.9351], device='cuda:1'), covar=tensor([0.0460, 0.0343, 0.0247, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0549, 0.0603, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 20:00:56,149 INFO [train.py:968] (1/2) Epoch 1, batch 33950, giga_loss[loss=0.3492, simple_loss=0.4124, pruned_loss=0.143, over 28595.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3912, pruned_loss=0.1385, over 5683279.68 frames. ], libri_tot_loss[loss=0.4101, simple_loss=0.4309, pruned_loss=0.1946, over 5710782.71 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3879, pruned_loss=0.1336, over 5666458.44 frames. ], batch size: 336, lr: 1.91e-02, grad_scale: 2.0
+2023-02-28 20:01:37,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33980.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:01:37,594 INFO [optim.py:369] (1/2) 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 20:01:40,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:02:01,237 INFO [train.py:968] (1/2) Epoch 1, batch 34000, giga_loss[loss=0.4242, simple_loss=0.466, pruned_loss=0.1912, over 28756.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3924, pruned_loss=0.1377, over 5676440.18 frames. ], libri_tot_loss[loss=0.41, simple_loss=0.4308, pruned_loss=0.1946, over 5712715.76 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3896, pruned_loss=0.1333, over 5661281.95 frames. ], batch size: 262, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 20:02:14,327 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34012.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:05,141 INFO [train.py:968] (1/2) Epoch 1, batch 34050, giga_loss[loss=0.328, simple_loss=0.3977, pruned_loss=0.1291, over 28878.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3925, pruned_loss=0.1375, over 5669763.57 frames. ], libri_tot_loss[loss=0.4091, simple_loss=0.4301, pruned_loss=0.194, over 5706752.50 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.39, pruned_loss=0.1334, over 5661376.27 frames. ], batch size: 174, lr: 1.91e-02, grad_scale: 4.0
+2023-02-28 20:03:28,397 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34066.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:35,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34069.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:35,889 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34070.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:03:49,642 INFO [optim.py:369] (1/2) 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 20:04:16,254 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 34100, giga_loss[loss=0.2927, simple_loss=0.3707, pruned_loss=0.1073, over 28980.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3928, pruned_loss=0.1379, over 5664758.80 frames. ], libri_tot_loss[loss=0.4092, simple_loss=0.4301, pruned_loss=0.1942, over 5698675.31 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3904, pruned_loss=0.1338, over 5663847.24 frames. ], batch size: 186, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:04:46,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2956, 1.5609, 1.2747, 1.3142], device='cuda:1'), covar=tensor([0.1278, 0.0611, 0.0684, 0.1551], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0287, 0.0295, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0024], device='cuda:1')
+2023-02-28 20:05:07,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8298, 1.4999, 1.4238, 1.4033], device='cuda:1'), covar=tensor([0.0665, 0.1515, 0.1087, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0855, 0.0636, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 20:05:25,352 INFO [train.py:968] (1/2) Epoch 1, batch 34150, giga_loss[loss=0.3162, simple_loss=0.3873, pruned_loss=0.1225, over 29035.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3926, pruned_loss=0.1381, over 5659379.51 frames. ], libri_tot_loss[loss=0.4092, simple_loss=0.4302, pruned_loss=0.1941, over 5695673.24 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3894, pruned_loss=0.1331, over 5660497.71 frames. ], batch size: 200, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:05:37,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.04 vs. limit=2.0
+2023-02-28 20:06:07,567 INFO [optim.py:369] (1/2) 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 20:06:12,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2929, 1.2669, 1.2665, 0.9613], device='cuda:1'), covar=tensor([0.0351, 0.0275, 0.0187, 0.0307], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0543, 0.0605, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 20:06:33,487 INFO [train.py:968] (1/2) Epoch 1, batch 34200, giga_loss[loss=0.293, simple_loss=0.3752, pruned_loss=0.1053, over 28912.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3932, pruned_loss=0.1386, over 5655860.22 frames. ], libri_tot_loss[loss=0.4088, simple_loss=0.4298, pruned_loss=0.1939, over 5693871.00 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3896, pruned_loss=0.1327, over 5657610.32 frames. ], batch size: 145, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:07:33,267 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 34250, giga_loss[loss=0.3129, simple_loss=0.3875, pruned_loss=0.1191, over 28364.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3944, pruned_loss=0.1392, over 5634807.58 frames. ], libri_tot_loss[loss=0.4089, simple_loss=0.4297, pruned_loss=0.194, over 5676432.03 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3912, pruned_loss=0.1339, over 5650931.05 frames. ], batch size: 65, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:08:25,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4610, 2.0573, 1.4646, 0.5029], device='cuda:1'), covar=tensor([0.1324, 0.1138, 0.0979, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0989, 0.1072, 0.1007, 0.0921], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 20:08:25,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6288, 2.0779, 1.6394, 1.5757], device='cuda:1'), covar=tensor([0.1154, 0.1075, 0.0984, 0.0639], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0817, 0.0701, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0005], device='cuda:1')
+2023-02-28 20:08:25,609 INFO [optim.py:369] (1/2) 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 20:08:33,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2016, 1.4287, 1.1208, 0.4576], device='cuda:1'), covar=tensor([0.0591, 0.0549, 0.0660, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0997, 0.1080, 0.1011, 0.0927], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 20:08:52,320 INFO [train.py:968] (1/2) Epoch 1, batch 34300, giga_loss[loss=0.4469, simple_loss=0.4696, pruned_loss=0.2122, over 29007.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3988, pruned_loss=0.1417, over 5655163.01 frames. ], libri_tot_loss[loss=0.409, simple_loss=0.4299, pruned_loss=0.1941, over 5680594.08 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3954, pruned_loss=0.1362, over 5663772.47 frames. ], batch size: 186, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:09:40,306 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34332.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:09:55,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5589, 1.8814, 2.0072, 1.5253], device='cuda:1'), covar=tensor([0.0725, 0.2086, 0.1033, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0851, 0.0637, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 20:09:59,886 INFO [train.py:968] (1/2) Epoch 1, batch 34350, giga_loss[loss=0.3666, simple_loss=0.4154, pruned_loss=0.1589, over 28860.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3989, pruned_loss=0.1424, over 5660409.55 frames. ], libri_tot_loss[loss=0.4093, simple_loss=0.4302, pruned_loss=0.1942, over 5674662.04 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3949, pruned_loss=0.1363, over 5670961.74 frames. ], batch size: 227, lr: 1.90e-02, grad_scale: 4.0
+2023-02-28 20:10:46,195 INFO [optim.py:369] (1/2) 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 20:10:51,246 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34383.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:10:54,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34386.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:11:10,572 INFO [train.py:968] (1/2) Epoch 1, batch 34400, giga_loss[loss=0.3336, simple_loss=0.3941, pruned_loss=0.1366, over 29024.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3955, pruned_loss=0.1402, over 5674789.17 frames. ], libri_tot_loss[loss=0.4091, simple_loss=0.4302, pruned_loss=0.194, over 5679406.79 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3915, pruned_loss=0.1343, over 5678732.97 frames. ], batch size: 285, lr: 1.90e-02, grad_scale: 8.0
+2023-02-28 20:11:33,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34415.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:12:08,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34437.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:12:21,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34445.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:12:30,243 INFO [train.py:968] (1/2) Epoch 1, batch 34450, giga_loss[loss=0.2615, simple_loss=0.3463, pruned_loss=0.08839, over 28890.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3918, pruned_loss=0.1368, over 5682663.84 frames. ], libri_tot_loss[loss=0.4088, simple_loss=0.43, pruned_loss=0.1938, over 5684333.78 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.388, pruned_loss=0.1309, over 5681269.10 frames. ], batch size: 164, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:12:34,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 20:13:07,518 INFO [optim.py:369] (1/2) 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 20:13:33,417 INFO [train.py:968] (1/2) Epoch 1, batch 34500, giga_loss[loss=0.3068, simple_loss=0.3502, pruned_loss=0.1317, over 24678.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3901, pruned_loss=0.1354, over 5680505.55 frames. ], libri_tot_loss[loss=0.4093, simple_loss=0.4304, pruned_loss=0.1941, over 5674401.97 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3858, pruned_loss=0.1292, over 5687757.92 frames. ], batch size: 705, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:14:23,882 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 34550, giga_loss[loss=0.3606, simple_loss=0.4167, pruned_loss=0.1522, over 28878.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3908, pruned_loss=0.1357, over 5680822.88 frames. ], libri_tot_loss[loss=0.409, simple_loss=0.4302, pruned_loss=0.1939, over 5677816.49 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.387, pruned_loss=0.1302, over 5683538.56 frames. ], batch size: 112, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:15:17,126 INFO [optim.py:369] (1/2) 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 20:15:21,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5160, 1.7113, 1.3763, 1.2858], device='cuda:1'), covar=tensor([0.1315, 0.0513, 0.0657, 0.1709], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0290, 0.0296, 0.0501], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0025], device='cuda:1')
+2023-02-28 20:15:26,094 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34588.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:15:29,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34591.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:15:41,649 INFO [train.py:968] (1/2) Epoch 1, batch 34600, giga_loss[loss=0.3304, simple_loss=0.3944, pruned_loss=0.1332, over 28931.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3949, pruned_loss=0.1394, over 5673789.90 frames. ], libri_tot_loss[loss=0.4084, simple_loss=0.4298, pruned_loss=0.1935, over 5681829.50 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3914, pruned_loss=0.134, over 5672413.23 frames. ], batch size: 164, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:16:06,425 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34620.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:16:43,480 INFO [train.py:968] (1/2) Epoch 1, batch 34650, giga_loss[loss=0.3161, simple_loss=0.377, pruned_loss=0.1276, over 28965.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3942, pruned_loss=0.1403, over 5670560.19 frames. ], libri_tot_loss[loss=0.4083, simple_loss=0.4297, pruned_loss=0.1934, over 5684234.44 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3911, pruned_loss=0.1354, over 5667183.83 frames. ], batch size: 186, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:16:45,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4539, 1.4961, 1.2579, 1.2749], device='cuda:1'), covar=tensor([0.0755, 0.0638, 0.1091, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0560, 0.0594, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 20:17:22,777 INFO [optim.py:369] (1/2) 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 20:17:43,515 INFO [train.py:968] (1/2) Epoch 1, batch 34700, giga_loss[loss=0.3798, simple_loss=0.4196, pruned_loss=0.17, over 28359.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3925, pruned_loss=0.1408, over 5673443.22 frames. ], libri_tot_loss[loss=0.4085, simple_loss=0.4298, pruned_loss=0.1936, over 5687992.56 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.389, pruned_loss=0.1355, over 5667132.44 frames. ], batch size: 368, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:17:54,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34707.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:18:42,097 INFO [train.py:968] (1/2) Epoch 1, batch 34750, giga_loss[loss=0.3357, simple_loss=0.3841, pruned_loss=0.1437, over 28112.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3939, pruned_loss=0.1427, over 5662410.00 frames. ], libri_tot_loss[loss=0.4081, simple_loss=0.4295, pruned_loss=0.1934, over 5680050.46 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3906, pruned_loss=0.1376, over 5664268.00 frames. ], batch size: 412, lr: 1.89e-02, grad_scale: 4.0
+2023-02-28 20:19:17,243 INFO [optim.py:369] (1/2) 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 20:19:34,316 INFO [train.py:968] (1/2) Epoch 1, batch 34800, giga_loss[loss=0.372, simple_loss=0.4202, pruned_loss=0.1619, over 28024.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4031, pruned_loss=0.1504, over 5662057.87 frames. ], libri_tot_loss[loss=0.4072, simple_loss=0.4289, pruned_loss=0.1927, over 5683067.01 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.4, pruned_loss=0.1453, over 5660132.75 frames. ], batch size: 412, lr: 1.89e-02, grad_scale: 8.0
+2023-02-28 20:19:44,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34812.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:20:06,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-02-28 20:20:09,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5220, 1.5261, 1.2206, 1.2278], device='cuda:1'), covar=tensor([0.0637, 0.0623, 0.0968, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0571, 0.0588, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 20:20:24,155 INFO [train.py:968] (1/2) Epoch 1, batch 34850, giga_loss[loss=0.4042, simple_loss=0.4485, pruned_loss=0.1799, over 28726.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4138, pruned_loss=0.1573, over 5675835.72 frames. ], libri_tot_loss[loss=0.4068, simple_loss=0.4287, pruned_loss=0.1925, over 5686138.32 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4113, pruned_loss=0.153, over 5671540.61 frames. ], batch size: 85, lr: 1.88e-02, grad_scale: 8.0
+2023-02-28 20:20:24,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34850.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:20:26,410 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34853.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:20:54,198 INFO [optim.py:369] (1/2) 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 20:20:54,467 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34882.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:21:09,145 INFO [train.py:968] (1/2) Epoch 1, batch 34900, giga_loss[loss=0.3759, simple_loss=0.4289, pruned_loss=0.1615, over 28934.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4175, pruned_loss=0.1607, over 5677944.05 frames. ], libri_tot_loss[loss=0.4069, simple_loss=0.4288, pruned_loss=0.1925, over 5688650.00 frames. ], giga_tot_loss[loss=0.364, simple_loss=0.4151, pruned_loss=0.1564, over 5671824.78 frames. ], batch size: 227, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:21:19,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34913.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:21:51,777 INFO [train.py:968] (1/2) Epoch 1, batch 34950, giga_loss[loss=0.3587, simple_loss=0.3661, pruned_loss=0.1756, over 24135.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4138, pruned_loss=0.1598, over 5682910.80 frames. ], libri_tot_loss[loss=0.408, simple_loss=0.4298, pruned_loss=0.1931, over 5693841.70 frames. ], giga_tot_loss[loss=0.3604, simple_loss=0.4106, pruned_loss=0.1551, over 5673097.77 frames. ], batch size: 705, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:21:55,578 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34958.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:17,183 INFO [optim.py:369] (1/2) 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 20:22:19,421 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=34986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:20,253 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34987.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:32,520 INFO [train.py:968] (1/2) Epoch 1, batch 35000, giga_loss[loss=0.3221, simple_loss=0.3793, pruned_loss=0.1324, over 29016.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.4072, pruned_loss=0.1571, over 5691041.99 frames. ], libri_tot_loss[loss=0.4079, simple_loss=0.4298, pruned_loss=0.193, over 5701812.18 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4037, pruned_loss=0.1519, over 5675518.60 frames. ], batch size: 155, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:22:39,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5651, 1.5556, 3.3717, 2.6197], device='cuda:1'), covar=tensor([0.1537, 0.1386, 0.0372, 0.0502], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0468, 0.0587, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 20:22:45,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35015.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:22:50,264 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35019.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:00,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4271, 1.4597, 1.3053, 1.3396], device='cuda:1'), covar=tensor([0.0536, 0.0672, 0.0940, 0.0647], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0858, 0.0636, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 20:23:14,922 INFO [train.py:968] (1/2) Epoch 1, batch 35050, giga_loss[loss=0.293, simple_loss=0.3521, pruned_loss=0.117, over 29001.00 frames. ], tot_loss[loss=0.3518, simple_loss=0.3987, pruned_loss=0.1525, over 5690706.44 frames. ], libri_tot_loss[loss=0.4083, simple_loss=0.4302, pruned_loss=0.1931, over 5698138.23 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3949, pruned_loss=0.1472, over 5681146.00 frames. ], batch size: 164, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:23:20,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35056.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:22,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35059.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:42,961 INFO [optim.py:369] (1/2) 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 20:23:47,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35088.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:23:57,794 INFO [train.py:968] (1/2) Epoch 1, batch 35100, giga_loss[loss=0.3245, simple_loss=0.3677, pruned_loss=0.1406, over 28182.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3908, pruned_loss=0.1488, over 5686993.88 frames. ], libri_tot_loss[loss=0.4089, simple_loss=0.4307, pruned_loss=0.1936, over 5700419.91 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3864, pruned_loss=0.143, over 5676871.88 frames. ], batch size: 368, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:24:00,730 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35103.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:24:41,825 INFO [train.py:968] (1/2) Epoch 1, batch 35150, giga_loss[loss=0.3336, simple_loss=0.351, pruned_loss=0.1582, over 23951.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3832, pruned_loss=0.1443, over 5688834.91 frames. ], libri_tot_loss[loss=0.4093, simple_loss=0.4311, pruned_loss=0.1937, over 5704745.42 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3785, pruned_loss=0.1387, over 5676905.59 frames. ], batch size: 705, lr: 1.88e-02, grad_scale: 4.0
+2023-02-28 20:25:10,780 INFO [optim.py:369] (1/2) 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 20:25:26,395 INFO [train.py:968] (1/2) Epoch 1, batch 35200, giga_loss[loss=0.3031, simple_loss=0.3618, pruned_loss=0.1222, over 28787.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3788, pruned_loss=0.1419, over 5693248.81 frames. ], libri_tot_loss[loss=0.4097, simple_loss=0.4315, pruned_loss=0.194, over 5698201.64 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3739, pruned_loss=0.1364, over 5688761.11 frames. ], batch size: 186, lr: 1.88e-02, grad_scale: 8.0
+2023-02-28 20:26:11,582 INFO [train.py:968] (1/2) Epoch 1, batch 35250, giga_loss[loss=0.3453, simple_loss=0.3849, pruned_loss=0.1529, over 29042.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3744, pruned_loss=0.1393, over 5692734.87 frames. ], libri_tot_loss[loss=0.4104, simple_loss=0.4321, pruned_loss=0.1944, over 5701119.50 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3691, pruned_loss=0.1338, over 5686363.06 frames. ], batch size: 164, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:26:23,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1942, 1.6765, 1.4607, 1.4163], device='cuda:1'), covar=tensor([0.1678, 0.0587, 0.0679, 0.2177], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0289, 0.0288, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0024], device='cuda:1')
+2023-02-28 20:26:42,625 INFO [optim.py:369] (1/2) 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 20:26:54,889 INFO [train.py:968] (1/2) Epoch 1, batch 35300, giga_loss[loss=0.2961, simple_loss=0.3588, pruned_loss=0.1167, over 29005.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3697, pruned_loss=0.1361, over 5681233.52 frames. ], libri_tot_loss[loss=0.411, simple_loss=0.4326, pruned_loss=0.1947, over 5694117.39 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.364, pruned_loss=0.1305, over 5682142.34 frames. ], batch size: 155, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:27:43,439 INFO [train.py:968] (1/2) Epoch 1, batch 35350, giga_loss[loss=0.2806, simple_loss=0.3335, pruned_loss=0.1138, over 28956.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.365, pruned_loss=0.1331, over 5677050.15 frames. ], libri_tot_loss[loss=0.4117, simple_loss=0.4332, pruned_loss=0.1951, over 5696855.08 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.359, pruned_loss=0.1274, over 5675076.31 frames. ], batch size: 106, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:27:52,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35361.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:28:12,316 INFO [optim.py:369] (1/2) 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 20:28:16,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35390.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:28:20,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35394.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:28:24,962 INFO [train.py:968] (1/2) Epoch 1, batch 35400, giga_loss[loss=0.2991, simple_loss=0.3533, pruned_loss=0.1225, over 28320.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3638, pruned_loss=0.1323, over 5681400.68 frames. ], libri_tot_loss[loss=0.4126, simple_loss=0.4341, pruned_loss=0.1956, over 5695994.87 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3556, pruned_loss=0.125, over 5679547.12 frames. ], batch size: 368, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:28:51,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2234, 1.3303, 1.1356, 1.2461], device='cuda:1'), covar=tensor([0.1475, 0.1526, 0.1240, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0774, 0.0844, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 20:29:05,155 INFO [train.py:968] (1/2) Epoch 1, batch 35450, giga_loss[loss=0.3059, simple_loss=0.3495, pruned_loss=0.1311, over 28970.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3651, pruned_loss=0.134, over 5691449.86 frames. ], libri_tot_loss[loss=0.4141, simple_loss=0.4355, pruned_loss=0.1964, over 5703231.09 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3545, pruned_loss=0.125, over 5682903.55 frames. ], batch size: 106, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:29:27,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35478.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:29:31,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-02-28 20:29:32,662 INFO [optim.py:369] (1/2) 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 20:29:34,544 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35485.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:29:46,925 INFO [train.py:968] (1/2) Epoch 1, batch 35500, giga_loss[loss=0.3017, simple_loss=0.3461, pruned_loss=0.1287, over 28821.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3618, pruned_loss=0.1322, over 5693615.23 frames. ], libri_tot_loss[loss=0.4145, simple_loss=0.4359, pruned_loss=0.1965, over 5703446.04 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3514, pruned_loss=0.1234, over 5686212.46 frames. ], batch size: 119, lr: 1.87e-02, grad_scale: 4.0
+2023-02-28 20:29:52,209 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35504.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:29:55,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35507.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:17,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35533.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:18,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35536.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:18,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35536.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:20,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4193, 1.6688, 1.2344, 1.4964], device='cuda:1'), covar=tensor([0.1236, 0.0596, 0.0687, 0.1554], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0288, 0.0287, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0024], device='cuda:1')
+2023-02-28 20:30:20,158 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35537.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:22,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35540.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:31,738 INFO [train.py:968] (1/2) Epoch 1, batch 35550, libri_loss[loss=0.5436, simple_loss=0.5146, pruned_loss=0.2863, over 29571.00 frames. ], tot_loss[loss=0.313, simple_loss=0.361, pruned_loss=0.1325, over 5678062.51 frames. ], libri_tot_loss[loss=0.416, simple_loss=0.4371, pruned_loss=0.1974, over 5697050.08 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3483, pruned_loss=0.1218, over 5677083.67 frames. ], batch size: 76, lr: 1.87e-02, grad_scale: 1.0
+2023-02-28 20:30:42,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-02-28 20:30:44,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35565.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:30:47,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35569.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:31:03,194 INFO [optim.py:369] (1/2) 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 20:31:16,094 INFO [train.py:968] (1/2) Epoch 1, batch 35600, giga_loss[loss=0.2998, simple_loss=0.3452, pruned_loss=0.1272, over 29064.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3594, pruned_loss=0.1321, over 5680856.99 frames. ], libri_tot_loss[loss=0.4164, simple_loss=0.4373, pruned_loss=0.1977, over 5699957.64 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3482, pruned_loss=0.1226, over 5677352.30 frames. ], batch size: 136, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:31:26,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3986, 1.8860, 1.5529, 0.5006], device='cuda:1'), covar=tensor([0.1011, 0.0809, 0.0951, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0979, 0.0995, 0.0981, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 20:31:36,898 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35621.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:31:39,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35624.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:32:04,901 INFO [train.py:968] (1/2) Epoch 1, batch 35650, giga_loss[loss=0.3688, simple_loss=0.4173, pruned_loss=0.1602, over 28775.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3722, pruned_loss=0.1398, over 5685500.74 frames. ], libri_tot_loss[loss=0.4166, simple_loss=0.4376, pruned_loss=0.1978, over 5702237.96 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3624, pruned_loss=0.1316, over 5680541.20 frames. ], batch size: 199, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:32:07,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35653.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:32:41,191 INFO [optim.py:369] (1/2) 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 20:32:42,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1903, 1.2824, 0.8748, 1.0382], device='cuda:1'), covar=tensor([0.0670, 0.0566, 0.1168, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0585, 0.0600, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 20:32:54,028 INFO [train.py:968] (1/2) Epoch 1, batch 35700, giga_loss[loss=0.405, simple_loss=0.4535, pruned_loss=0.1783, over 28896.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.39, pruned_loss=0.1518, over 5683653.88 frames. ], libri_tot_loss[loss=0.4168, simple_loss=0.4377, pruned_loss=0.198, over 5703312.77 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3821, pruned_loss=0.1451, over 5678761.23 frames. ], batch size: 174, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:33:28,109 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35738.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:33:40,447 INFO [train.py:968] (1/2) Epoch 1, batch 35750, giga_loss[loss=0.3295, simple_loss=0.3902, pruned_loss=0.1344, over 28777.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4018, pruned_loss=0.1589, over 5686123.39 frames. ], libri_tot_loss[loss=0.4173, simple_loss=0.4379, pruned_loss=0.1983, over 5705239.28 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.395, pruned_loss=0.153, over 5680491.94 frames. ], batch size: 92, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:33:56,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35767.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:33:59,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3816, 1.3852, 3.5654, 2.7957], device='cuda:1'), covar=tensor([0.1393, 0.1306, 0.0266, 0.0494], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0458, 0.0571, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:1')
+2023-02-28 20:34:01,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5491, 1.6411, 1.4117, 1.7502], device='cuda:1'), covar=tensor([0.1122, 0.0452, 0.0584, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0283, 0.0278, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0017, 0.0015, 0.0024], device='cuda:1')
+2023-02-28 20:34:12,335 INFO [optim.py:369] (1/2) 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 20:34:24,974 INFO [train.py:968] (1/2) Epoch 1, batch 35800, giga_loss[loss=0.3634, simple_loss=0.4154, pruned_loss=0.1557, over 28918.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.4085, pruned_loss=0.1613, over 5684417.06 frames. ], libri_tot_loss[loss=0.4174, simple_loss=0.4381, pruned_loss=0.1984, over 5706847.40 frames. ], giga_tot_loss[loss=0.3574, simple_loss=0.4026, pruned_loss=0.1561, over 5678401.63 frames. ], batch size: 186, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:35:10,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2342, 2.7457, 2.2232, 1.8832], device='cuda:1'), covar=tensor([0.0919, 0.0633, 0.0862, 0.1592], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0460, 0.0376, 0.0479], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0012, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 20:35:12,524 INFO [train.py:968] (1/2) Epoch 1, batch 35850, giga_loss[loss=0.3362, simple_loss=0.4007, pruned_loss=0.1358, over 28824.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.4085, pruned_loss=0.1586, over 5676399.30 frames. ], libri_tot_loss[loss=0.4181, simple_loss=0.4386, pruned_loss=0.1988, over 5705470.96 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.403, pruned_loss=0.1537, over 5672705.19 frames. ], batch size: 174, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:35:24,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35860.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:35:49,638 INFO [optim.py:369] (1/2) 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 20:36:01,393 INFO [train.py:968] (1/2) Epoch 1, batch 35900, giga_loss[loss=0.3882, simple_loss=0.4096, pruned_loss=0.1835, over 23657.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.409, pruned_loss=0.158, over 5661861.26 frames. ], libri_tot_loss[loss=0.4184, simple_loss=0.4388, pruned_loss=0.1991, over 5698064.20 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4043, pruned_loss=0.1536, over 5665072.88 frames. ], batch size: 705, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:36:20,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=35923.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:36:46,679 INFO [train.py:968] (1/2) Epoch 1, batch 35950, giga_loss[loss=0.3963, simple_loss=0.426, pruned_loss=0.1833, over 28756.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4105, pruned_loss=0.1592, over 5679475.73 frames. ], libri_tot_loss[loss=0.4185, simple_loss=0.4388, pruned_loss=0.1991, over 5700463.58 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.406, pruned_loss=0.1545, over 5679122.11 frames. ], batch size: 99, lr: 1.86e-02, grad_scale: 2.0
+2023-02-28 20:37:16,585 INFO [optim.py:369] (1/2) 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 20:37:27,418 INFO [train.py:968] (1/2) Epoch 1, batch 36000, giga_loss[loss=0.3473, simple_loss=0.4046, pruned_loss=0.145, over 28899.00 frames. ], tot_loss[loss=0.3698, simple_loss=0.4147, pruned_loss=0.1625, over 5683408.70 frames. ], libri_tot_loss[loss=0.4199, simple_loss=0.44, pruned_loss=0.1999, over 5705656.55 frames. ], giga_tot_loss[loss=0.3619, simple_loss=0.4094, pruned_loss=0.1572, over 5678112.15 frames. ], batch size: 99, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:37:27,418 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 20:37:35,987 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-02-28 20:37:39,632 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36003.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:37:41,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36006.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:38:05,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36035.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:38:17,296 INFO [train.py:968] (1/2) Epoch 1, batch 36050, giga_loss[loss=0.398, simple_loss=0.4409, pruned_loss=0.1776, over 28665.00 frames. ], tot_loss[loss=0.3748, simple_loss=0.4194, pruned_loss=0.1651, over 5692731.70 frames. ], libri_tot_loss[loss=0.4212, simple_loss=0.4413, pruned_loss=0.2006, over 5708788.87 frames. ], giga_tot_loss[loss=0.3659, simple_loss=0.4134, pruned_loss=0.1592, over 5684984.34 frames. ], batch size: 307, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:38:25,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5657, 1.1543, 1.3486, 0.9520], device='cuda:1'), covar=tensor([0.0394, 0.0301, 0.0255, 0.0385], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0552, 0.0638, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 20:38:46,018 INFO [optim.py:369] (1/2) 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 20:38:57,252 INFO [train.py:968] (1/2) Epoch 1, batch 36100, giga_loss[loss=0.38, simple_loss=0.4358, pruned_loss=0.1621, over 28373.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.422, pruned_loss=0.1655, over 5708737.23 frames. ], libri_tot_loss[loss=0.4212, simple_loss=0.4412, pruned_loss=0.2006, over 5714165.66 frames. ], giga_tot_loss[loss=0.3683, simple_loss=0.4167, pruned_loss=0.1599, over 5697369.77 frames. ], batch size: 65, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:39:08,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36113.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:39:34,933 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 36150, libri_loss[loss=0.5482, simple_loss=0.532, pruned_loss=0.2822, over 29509.00 frames. ], tot_loss[loss=0.3794, simple_loss=0.4246, pruned_loss=0.1671, over 5696984.31 frames. ], libri_tot_loss[loss=0.423, simple_loss=0.4425, pruned_loss=0.2017, over 5714439.43 frames. ], giga_tot_loss[loss=0.3705, simple_loss=0.419, pruned_loss=0.161, over 5687184.79 frames. ], batch size: 84, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:39:52,811 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=36164.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:40:08,555 INFO [optim.py:369] (1/2) 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 20:40:21,698 INFO [train.py:968] (1/2) Epoch 1, batch 36200, giga_loss[loss=0.3776, simple_loss=0.4322, pruned_loss=0.1615, over 28990.00 frames. ], tot_loss[loss=0.3784, simple_loss=0.4248, pruned_loss=0.166, over 5696999.11 frames. ], libri_tot_loss[loss=0.4233, simple_loss=0.4427, pruned_loss=0.2019, over 5707835.92 frames. ], giga_tot_loss[loss=0.3701, simple_loss=0.4197, pruned_loss=0.1602, over 5693904.33 frames. ], batch size: 106, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:40:26,145 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 36250, libri_loss[loss=0.4447, simple_loss=0.4721, pruned_loss=0.2087, over 29545.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4242, pruned_loss=0.1645, over 5702387.45 frames. ], libri_tot_loss[loss=0.4241, simple_loss=0.4434, pruned_loss=0.2024, over 5711544.98 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4191, pruned_loss=0.1587, over 5696102.66 frames. ], batch size: 82, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:41:07,901 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36256.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:41:10,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36259.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:41:15,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-02-28 20:41:31,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36285.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:41:32,214 INFO [optim.py:369] (1/2) 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 20:41:34,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36288.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:41:34,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36288.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:41:42,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36298.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:41:43,996 INFO [train.py:968] (1/2) Epoch 1, batch 36300, giga_loss[loss=0.358, simple_loss=0.4197, pruned_loss=0.1482, over 28910.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.4231, pruned_loss=0.1632, over 5701439.59 frames. ], libri_tot_loss[loss=0.4251, simple_loss=0.4441, pruned_loss=0.2031, over 5713760.45 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4181, pruned_loss=0.1572, over 5694434.20 frames. ], batch size: 174, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:41:53,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5773, 1.6589, 3.5157, 2.8336], device='cuda:1'), covar=tensor([0.1447, 0.1222, 0.0279, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0454, 0.0553, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:1')
+2023-02-28 20:41:58,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36317.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:42:27,028 INFO [train.py:968] (1/2) Epoch 1, batch 36350, giga_loss[loss=0.4305, simple_loss=0.4477, pruned_loss=0.2067, over 28563.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.422, pruned_loss=0.1627, over 5693635.84 frames. ], libri_tot_loss[loss=0.4248, simple_loss=0.444, pruned_loss=0.2029, over 5714550.65 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.418, pruned_loss=0.158, over 5687399.12 frames. ], batch size: 336, lr: 1.85e-02, grad_scale: 4.0
+2023-02-28 20:43:03,135 INFO [optim.py:369] (1/2) 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 20:43:15,380 INFO [train.py:968] (1/2) Epoch 1, batch 36400, giga_loss[loss=0.4422, simple_loss=0.4609, pruned_loss=0.2117, over 28683.00 frames. ], tot_loss[loss=0.3798, simple_loss=0.4245, pruned_loss=0.1675, over 5691082.92 frames. ], libri_tot_loss[loss=0.4252, simple_loss=0.4444, pruned_loss=0.203, over 5715764.51 frames. ], giga_tot_loss[loss=0.3736, simple_loss=0.4208, pruned_loss=0.1632, over 5684681.57 frames. ], batch size: 92, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:43:29,353 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-02-28 20:43:40,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4516, 1.7823, 1.5631, 1.4710], device='cuda:1'), covar=tensor([0.1057, 0.1256, 0.0936, 0.0683], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0831, 0.0704, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0010, 0.0007, 0.0006], device='cuda:1')
+2023-02-28 20:43:51,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36441.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:43:53,474 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 36450, giga_loss[loss=0.3645, simple_loss=0.4147, pruned_loss=0.1572, over 29012.00 frames. ], tot_loss[loss=0.386, simple_loss=0.4274, pruned_loss=0.1723, over 5694815.60 frames. ], libri_tot_loss[loss=0.426, simple_loss=0.445, pruned_loss=0.2035, over 5721396.59 frames. ], giga_tot_loss[loss=0.3791, simple_loss=0.4233, pruned_loss=0.1674, over 5683896.41 frames. ], batch size: 128, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:44:08,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-02-28 20:44:18,359 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-02-28 20:44:18,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36473.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:44:31,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6513, 2.2849, 1.8722, 0.4644], device='cuda:1'), covar=tensor([0.1528, 0.0995, 0.1042, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1034, 0.1036, 0.1034, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 20:44:33,444 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 36500, giga_loss[loss=0.4685, simple_loss=0.463, pruned_loss=0.237, over 28621.00 frames. ], tot_loss[loss=0.3854, simple_loss=0.4258, pruned_loss=0.1725, over 5695541.77 frames. ], libri_tot_loss[loss=0.4259, simple_loss=0.4449, pruned_loss=0.2035, over 5723069.03 frames. ], giga_tot_loss[loss=0.3796, simple_loss=0.4225, pruned_loss=0.1684, over 5685083.23 frames. ], batch size: 85, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:45:21,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36539.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:45:29,986 INFO [train.py:968] (1/2) Epoch 1, batch 36550, giga_loss[loss=0.3941, simple_loss=0.4339, pruned_loss=0.1772, over 28621.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4232, pruned_loss=0.1703, over 5705063.29 frames. ], libri_tot_loss[loss=0.4265, simple_loss=0.4455, pruned_loss=0.2038, over 5724529.89 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.4198, pruned_loss=0.1665, over 5695295.58 frames. ], batch size: 336, lr: 1.84e-02, grad_scale: 8.0
+2023-02-28 20:45:53,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=36574.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:45:58,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36579.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:46:04,156 INFO [optim.py:369] (1/2) 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 20:46:13,626 INFO [train.py:968] (1/2) Epoch 1, batch 36600, giga_loss[loss=0.424, simple_loss=0.4519, pruned_loss=0.198, over 27964.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4231, pruned_loss=0.1703, over 5697757.22 frames. ], libri_tot_loss[loss=0.4277, simple_loss=0.4465, pruned_loss=0.2045, over 5718342.82 frames. ], giga_tot_loss[loss=0.3754, simple_loss=0.419, pruned_loss=0.1659, over 5695075.57 frames. ], batch size: 412, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:47:03,380 INFO [train.py:968] (1/2) Epoch 1, batch 36650, giga_loss[loss=0.3484, simple_loss=0.4071, pruned_loss=0.1448, over 28784.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4191, pruned_loss=0.1655, over 5697979.34 frames. ], libri_tot_loss[loss=0.4275, simple_loss=0.4465, pruned_loss=0.2043, over 5720285.52 frames. ], giga_tot_loss[loss=0.3696, simple_loss=0.4156, pruned_loss=0.1618, over 5693891.16 frames. ], batch size: 284, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:47:11,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7937, 1.8830, 4.6361, 3.2316], device='cuda:1'), covar=tensor([0.1527, 0.1182, 0.0195, 0.0458], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0462, 0.0562, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:1')
+2023-02-28 20:47:32,914 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36682.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:47:37,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36685.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:47:39,782 INFO [optim.py:369] (1/2) 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 20:47:51,816 INFO [train.py:968] (1/2) Epoch 1, batch 36700, giga_loss[loss=0.3036, simple_loss=0.367, pruned_loss=0.1201, over 28846.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.4132, pruned_loss=0.161, over 5694138.44 frames. ], libri_tot_loss[loss=0.4278, simple_loss=0.4466, pruned_loss=0.2044, over 5723292.12 frames. ], giga_tot_loss[loss=0.3623, simple_loss=0.4099, pruned_loss=0.1574, over 5687999.43 frames. ], batch size: 227, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:48:03,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6887, 1.5703, 1.2276, 1.2710], device='cuda:1'), covar=tensor([0.0808, 0.0862, 0.1134, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0568, 0.0582, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 20:48:05,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36714.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:48:12,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36722.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:48:15,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36725.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:48:38,449 INFO [train.py:968] (1/2) Epoch 1, batch 36750, giga_loss[loss=0.3044, simple_loss=0.3604, pruned_loss=0.1242, over 28924.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4085, pruned_loss=0.1583, over 5703488.40 frames. ], libri_tot_loss[loss=0.4295, simple_loss=0.448, pruned_loss=0.2055, over 5724981.13 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4036, pruned_loss=0.1532, over 5696416.03 frames. ], batch size: 213, lr: 1.84e-02, grad_scale: 4.0
+2023-02-28 20:48:42,715 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36754.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:49:18,008 INFO [optim.py:369] (1/2) 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 20:49:32,465 INFO [train.py:968] (1/2) Epoch 1, batch 36800, giga_loss[loss=0.3083, simple_loss=0.3666, pruned_loss=0.125, over 28720.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4012, pruned_loss=0.1535, over 5679591.00 frames. ], libri_tot_loss[loss=0.4303, simple_loss=0.4487, pruned_loss=0.2059, over 5719812.17 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3956, pruned_loss=0.1478, over 5677731.71 frames. ], batch size: 99, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:49:51,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8253, 1.7282, 1.3273, 1.5267], device='cuda:1'), covar=tensor([0.0646, 0.0803, 0.1071, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0562, 0.0584, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 20:50:10,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3344, 1.1169, 1.2236, 1.6662], device='cuda:1'), covar=tensor([0.1703, 0.1799, 0.1337, 0.1752], device='cuda:1'), in_proj_covar=tensor([0.0896, 0.0785, 0.0861, 0.0913], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 20:50:23,402 INFO [train.py:968] (1/2) Epoch 1, batch 36850, giga_loss[loss=0.358, simple_loss=0.4086, pruned_loss=0.1537, over 28867.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3955, pruned_loss=0.1484, over 5681670.84 frames. ], libri_tot_loss[loss=0.4305, simple_loss=0.4489, pruned_loss=0.2061, over 5721947.61 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3903, pruned_loss=0.1433, over 5677780.72 frames. ], batch size: 227, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:50:51,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4872, 3.2233, 4.2066, 1.9413], device='cuda:1'), covar=tensor([0.0407, 0.0524, 0.0617, 0.1769], device='cuda:1'), in_proj_covar=tensor([0.0686, 0.0470, 0.0744, 0.0523], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 20:50:56,520 INFO [optim.py:369] (1/2) 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 20:50:56,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1850, 1.1599, 1.1023, 1.3048], device='cuda:1'), covar=tensor([0.1650, 0.1754, 0.1350, 0.1742], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0786, 0.0853, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 20:51:08,469 INFO [train.py:968] (1/2) Epoch 1, batch 36900, giga_loss[loss=0.3151, simple_loss=0.3741, pruned_loss=0.1281, over 28936.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3966, pruned_loss=0.1493, over 5673264.48 frames. ], libri_tot_loss[loss=0.4312, simple_loss=0.4494, pruned_loss=0.2064, over 5713110.70 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3915, pruned_loss=0.1442, over 5677622.43 frames. ], batch size: 112, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:51:50,732 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 36950, giga_loss[loss=0.3016, simple_loss=0.3636, pruned_loss=0.1198, over 29059.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3947, pruned_loss=0.1476, over 5685502.69 frames. ], libri_tot_loss[loss=0.4318, simple_loss=0.4501, pruned_loss=0.2067, over 5711432.84 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3896, pruned_loss=0.1428, over 5690352.39 frames. ], batch size: 155, lr: 1.83e-02, grad_scale: 8.0
+2023-02-28 20:51:54,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3903, 1.7342, 1.8294, 1.6149], device='cuda:1'), covar=tensor([0.0802, 0.1918, 0.1124, 0.1471], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0870, 0.0648, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 20:52:24,197 INFO [optim.py:369] (1/2) 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 20:52:32,801 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=36998.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:52:33,913 INFO [train.py:968] (1/2) Epoch 1, batch 37000, giga_loss[loss=0.3074, simple_loss=0.364, pruned_loss=0.1254, over 28819.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.3966, pruned_loss=0.1503, over 5682378.09 frames. ], libri_tot_loss[loss=0.4343, simple_loss=0.4523, pruned_loss=0.2082, over 5713518.45 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3887, pruned_loss=0.1434, over 5683599.69 frames. ], batch size: 199, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:52:35,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37001.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:52:38,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37005.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:52:59,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9720, 2.9143, 3.7312, 1.7656], device='cuda:1'), covar=tensor([0.0541, 0.0601, 0.0724, 0.1865], device='cuda:1'), in_proj_covar=tensor([0.0717, 0.0479, 0.0765, 0.0524], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0005, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 20:53:13,901 INFO [train.py:968] (1/2) Epoch 1, batch 37050, giga_loss[loss=0.3302, simple_loss=0.3776, pruned_loss=0.1414, over 28808.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3937, pruned_loss=0.1485, over 5694950.33 frames. ], libri_tot_loss[loss=0.4351, simple_loss=0.4531, pruned_loss=0.2085, over 5715322.08 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3857, pruned_loss=0.1417, over 5693917.13 frames. ], batch size: 119, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:53:44,467 INFO [optim.py:369] (1/2) 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 20:53:48,212 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:53:48,801 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37093.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:53:50,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37095.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:53:54,924 INFO [train.py:968] (1/2) Epoch 1, batch 37100, giga_loss[loss=0.3587, simple_loss=0.399, pruned_loss=0.1592, over 28028.00 frames. ], tot_loss[loss=0.3425, simple_loss=0.391, pruned_loss=0.147, over 5707651.81 frames. ], libri_tot_loss[loss=0.4362, simple_loss=0.4541, pruned_loss=0.2092, over 5720004.20 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3823, pruned_loss=0.1396, over 5702372.55 frames. ], batch size: 412, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:54:12,757 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 37150, giga_loss[loss=0.4394, simple_loss=0.4437, pruned_loss=0.2176, over 26749.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3889, pruned_loss=0.1458, over 5715912.47 frames. ], libri_tot_loss[loss=0.4369, simple_loss=0.4549, pruned_loss=0.2094, over 5724500.70 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3795, pruned_loss=0.1379, over 5707374.58 frames. ], batch size: 555, lr: 1.83e-02, grad_scale: 4.0
+2023-02-28 20:55:06,114 INFO [optim.py:369] (1/2) 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 20:55:15,245 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37198.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 20:55:16,383 INFO [train.py:968] (1/2) Epoch 1, batch 37200, giga_loss[loss=0.2753, simple_loss=0.3326, pruned_loss=0.109, over 28892.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3853, pruned_loss=0.1437, over 5701759.65 frames. ], libri_tot_loss[loss=0.4378, simple_loss=0.4556, pruned_loss=0.21, over 5713884.17 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3764, pruned_loss=0.1361, over 5704419.32 frames. ], batch size: 119, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:55:43,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8258, 2.5337, 2.0094, 1.9880], device='cuda:1'), covar=tensor([0.1382, 0.1108, 0.0967, 0.0581], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0857, 0.0714, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 20:55:57,205 INFO [train.py:968] (1/2) Epoch 1, batch 37250, giga_loss[loss=0.2587, simple_loss=0.3284, pruned_loss=0.09448, over 28879.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3828, pruned_loss=0.1419, over 5708006.66 frames. ], libri_tot_loss[loss=0.4383, simple_loss=0.4562, pruned_loss=0.2103, over 5716116.22 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3735, pruned_loss=0.1341, over 5707902.91 frames. ], batch size: 174, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:56:17,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5671, 1.3931, 1.3982, 1.7949], device='cuda:1'), covar=tensor([0.1738, 0.2023, 0.1435, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.0904, 0.0786, 0.0852, 0.0925], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 20:56:27,583 INFO [optim.py:369] (1/2) 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 20:56:37,846 INFO [train.py:968] (1/2) Epoch 1, batch 37300, libri_loss[loss=0.5498, simple_loss=0.5368, pruned_loss=0.2815, over 29592.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3805, pruned_loss=0.1406, over 5718348.80 frames. ], libri_tot_loss[loss=0.4396, simple_loss=0.4572, pruned_loss=0.211, over 5720789.48 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3703, pruned_loss=0.132, over 5713838.33 frames. ], batch size: 74, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:56:40,756 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37304.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:19,702 INFO [train.py:968] (1/2) Epoch 1, batch 37350, giga_loss[loss=0.2922, simple_loss=0.3519, pruned_loss=0.1162, over 29076.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3776, pruned_loss=0.1384, over 5718264.20 frames. ], libri_tot_loss[loss=0.4404, simple_loss=0.4578, pruned_loss=0.2115, over 5713427.65 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3683, pruned_loss=0.1304, over 5720590.10 frames. ], batch size: 155, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 20:57:20,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37351.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:39,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37373.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:41,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37376.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:57:44,760 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:58:00,976 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37397.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:58:02,607 INFO [train.py:968] (1/2) Epoch 1, batch 37400, libri_loss[loss=0.5228, simple_loss=0.5243, pruned_loss=0.2607, over 29741.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3787, pruned_loss=0.1392, over 5717430.58 frames. ], libri_tot_loss[loss=0.442, simple_loss=0.4592, pruned_loss=0.2124, over 5715538.96 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3675, pruned_loss=0.1298, over 5717510.53 frames. ], batch size: 87, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 20:58:30,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3663, 1.1663, 1.2277, 1.0821], device='cuda:1'), covar=tensor([0.0350, 0.0287, 0.0220, 0.0346], device='cuda:1'), in_proj_covar=tensor([0.0828, 0.0574, 0.0659, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 20:58:43,622 INFO [train.py:968] (1/2) Epoch 1, batch 37450, giga_loss[loss=0.3513, simple_loss=0.4028, pruned_loss=0.1499, over 28685.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3833, pruned_loss=0.1423, over 5718196.42 frames. ], libri_tot_loss[loss=0.4426, simple_loss=0.4599, pruned_loss=0.2127, over 5718897.54 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3723, pruned_loss=0.1332, over 5715297.80 frames. ], batch size: 242, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 20:59:00,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37468.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 20:59:18,439 INFO [optim.py:369] (1/2) 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 20:59:22,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-02-28 20:59:26,382 INFO [train.py:968] (1/2) Epoch 1, batch 37500, giga_loss[loss=0.4205, simple_loss=0.4545, pruned_loss=0.1932, over 28280.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3921, pruned_loss=0.1488, over 5705592.19 frames. ], libri_tot_loss[loss=0.4443, simple_loss=0.4613, pruned_loss=0.2136, over 5710824.93 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3801, pruned_loss=0.1389, over 5709447.97 frames. ], batch size: 368, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 20:59:30,650 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8438, 1.6821, 4.0058, 3.0822], device='cuda:1'), covar=tensor([0.1542, 0.1392, 0.0280, 0.0437], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0465, 0.0584, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 20:59:39,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37516.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:43,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37519.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:43,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37519.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:47,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37522.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:47,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37523.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 20:59:50,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37526.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:00:14,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-02-28 21:00:15,060 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 37550, giga_loss[loss=0.3844, simple_loss=0.4232, pruned_loss=0.1728, over 28479.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4025, pruned_loss=0.1571, over 5699195.97 frames. ], libri_tot_loss[loss=0.4446, simple_loss=0.4616, pruned_loss=0.2138, over 5716240.01 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3908, pruned_loss=0.1474, over 5697361.26 frames. ], batch size: 85, lr: 1.82e-02, grad_scale: 4.0
+2023-02-28 21:00:17,270 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37551.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:00:21,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37555.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:00:36,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5418, 1.4153, 1.1557, 1.2054], device='cuda:1'), covar=tensor([0.0523, 0.0481, 0.0772, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0565, 0.0573, 0.0528], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 21:00:37,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37573.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:00:52,192 INFO [optim.py:369] (1/2) 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 21:01:02,086 INFO [train.py:968] (1/2) Epoch 1, batch 37600, giga_loss[loss=0.369, simple_loss=0.4277, pruned_loss=0.1551, over 28868.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4107, pruned_loss=0.1637, over 5686437.36 frames. ], libri_tot_loss[loss=0.4437, simple_loss=0.4608, pruned_loss=0.2133, over 5720592.77 frames. ], giga_tot_loss[loss=0.3555, simple_loss=0.4008, pruned_loss=0.1551, over 5680438.14 frames. ], batch size: 227, lr: 1.82e-02, grad_scale: 8.0
+2023-02-28 21:01:17,632 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37611.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:01:20,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37614.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:01:28,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6395, 1.7179, 3.3842, 2.7336], device='cuda:1'), covar=tensor([0.1540, 0.1287, 0.0335, 0.0519], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0464, 0.0581, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:1')
+2023-02-28 21:01:28,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3074, 1.2850, 1.2240, 1.2250], device='cuda:1'), covar=tensor([0.0958, 0.1466, 0.1385, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0876, 0.0664, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 21:01:40,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-02-28 21:01:45,266 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37643.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:01:51,007 INFO [train.py:968] (1/2) Epoch 1, batch 37650, giga_loss[loss=0.3583, simple_loss=0.4151, pruned_loss=0.1507, over 28944.00 frames. ], tot_loss[loss=0.3753, simple_loss=0.4166, pruned_loss=0.167, over 5689684.51 frames. ], libri_tot_loss[loss=0.4446, simple_loss=0.4613, pruned_loss=0.2139, over 5726432.94 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4067, pruned_loss=0.158, over 5678484.60 frames. ], batch size: 227, lr: 1.81e-02, grad_scale: 8.0
+2023-02-28 21:01:55,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8122, 1.4021, 1.5752, 1.0132], device='cuda:1'), covar=tensor([0.0333, 0.0342, 0.0230, 0.0395], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0571, 0.0649, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 21:02:00,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37661.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:02:18,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37679.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:02:28,739 INFO [optim.py:369] (1/2) 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 21:02:37,465 INFO [train.py:968] (1/2) Epoch 1, batch 37700, libri_loss[loss=0.4496, simple_loss=0.47, pruned_loss=0.2146, over 27876.00 frames. ], tot_loss[loss=0.3806, simple_loss=0.4216, pruned_loss=0.1698, over 5682180.87 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4614, pruned_loss=0.2141, over 5727968.80 frames. ], giga_tot_loss[loss=0.3677, simple_loss=0.4127, pruned_loss=0.1614, over 5670995.65 frames. ], batch size: 115, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:02:39,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4404, 1.7201, 1.4045, 1.2316], device='cuda:1'), covar=tensor([0.0991, 0.0861, 0.0897, 0.1506], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0452, 0.0359, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0012, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 21:02:55,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37716.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:02:57,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37719.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:03:02,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37726.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:03:18,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37748.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:03:19,726 INFO [train.py:968] (1/2) Epoch 1, batch 37750, libri_loss[loss=0.4681, simple_loss=0.4771, pruned_loss=0.2296, over 29524.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4289, pruned_loss=0.1759, over 5691009.23 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4614, pruned_loss=0.2141, over 5732331.41 frames. ], giga_tot_loss[loss=0.3773, simple_loss=0.4201, pruned_loss=0.1673, over 5675906.44 frames. ], batch size: 84, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:03:26,841 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=37757.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:03:39,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:03:55,222 INFO [optim.py:369] (1/2) 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 21:04:04,437 INFO [train.py:968] (1/2) Epoch 1, batch 37800, libri_loss[loss=0.4569, simple_loss=0.4716, pruned_loss=0.2211, over 29655.00 frames. ], tot_loss[loss=0.3816, simple_loss=0.4216, pruned_loss=0.1708, over 5689054.68 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4612, pruned_loss=0.2142, over 5735189.02 frames. ], giga_tot_loss[loss=0.3698, simple_loss=0.4139, pruned_loss=0.1628, over 5673329.36 frames. ], batch size: 88, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:04:23,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37822.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:04:25,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37825.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:04:45,912 INFO [train.py:968] (1/2) Epoch 1, batch 37850, giga_loss[loss=0.3099, simple_loss=0.3794, pruned_loss=0.1202, over 28835.00 frames. ], tot_loss[loss=0.372, simple_loss=0.4153, pruned_loss=0.1643, over 5696599.16 frames. ], libri_tot_loss[loss=0.4447, simple_loss=0.4612, pruned_loss=0.2141, over 5738803.33 frames. ], giga_tot_loss[loss=0.3612, simple_loss=0.4083, pruned_loss=0.1571, over 5679936.18 frames. ], batch size: 186, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:04:49,103 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37854.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:04,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37869.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:07,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37872.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:22,059 INFO [optim.py:369] (1/2) 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 21:05:31,211 INFO [train.py:968] (1/2) Epoch 1, batch 37900, giga_loss[loss=0.4563, simple_loss=0.4571, pruned_loss=0.2278, over 26777.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4138, pruned_loss=0.1625, over 5687208.34 frames. ], libri_tot_loss[loss=0.4445, simple_loss=0.461, pruned_loss=0.214, over 5728616.38 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4074, pruned_loss=0.1558, over 5681316.25 frames. ], batch size: 555, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:05:32,827 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37901.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:42,816 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37915.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:05:45,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37918.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:06:10,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37947.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:06:13,503 INFO [train.py:968] (1/2) Epoch 1, batch 37950, giga_loss[loss=0.3964, simple_loss=0.4398, pruned_loss=0.1765, over 28708.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4148, pruned_loss=0.1621, over 5690824.70 frames. ], libri_tot_loss[loss=0.4454, simple_loss=0.4617, pruned_loss=0.2145, over 5728721.53 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4082, pruned_loss=0.1554, over 5685472.92 frames. ], batch size: 71, lr: 1.81e-02, grad_scale: 4.0
+2023-02-28 21:06:38,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6640, 1.4528, 1.4497, 0.8805], device='cuda:1'), covar=tensor([0.0303, 0.0245, 0.0227, 0.0345], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0569, 0.0657, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 21:06:48,058 INFO [optim.py:369] (1/2) 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 21:06:57,713 INFO [train.py:968] (1/2) Epoch 1, batch 38000, giga_loss[loss=0.4211, simple_loss=0.4366, pruned_loss=0.2029, over 23791.00 frames. ], tot_loss[loss=0.3727, simple_loss=0.4176, pruned_loss=0.1638, over 5676502.30 frames. ], libri_tot_loss[loss=0.4456, simple_loss=0.4619, pruned_loss=0.2147, over 5717986.06 frames. ], giga_tot_loss[loss=0.3633, simple_loss=0.4115, pruned_loss=0.1575, over 5681257.33 frames. ], batch size: 705, lr: 1.81e-02, grad_scale: 8.0
+2023-02-28 21:07:00,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7105, 1.8218, 3.8973, 2.9187], device='cuda:1'), covar=tensor([0.1470, 0.1172, 0.0260, 0.0575], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0459, 0.0590, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:07:29,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38036.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:07:30,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4033, 1.7615, 1.5549, 1.5425], device='cuda:1'), covar=tensor([0.1136, 0.1284, 0.0980, 0.0581], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0837, 0.0708, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:07:42,111 INFO [train.py:968] (1/2) Epoch 1, batch 38050, giga_loss[loss=0.3963, simple_loss=0.4327, pruned_loss=0.1799, over 28546.00 frames. ], tot_loss[loss=0.3764, simple_loss=0.4206, pruned_loss=0.1661, over 5678935.98 frames. ], libri_tot_loss[loss=0.4458, simple_loss=0.4623, pruned_loss=0.2146, over 5719852.93 frames. ], giga_tot_loss[loss=0.3665, simple_loss=0.414, pruned_loss=0.1595, over 5679941.57 frames. ], batch size: 307, lr: 1.80e-02, grad_scale: 8.0
+2023-02-28 21:08:20,587 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 38100, giga_loss[loss=0.3609, simple_loss=0.4136, pruned_loss=0.1541, over 29072.00 frames. ], tot_loss[loss=0.3792, simple_loss=0.4225, pruned_loss=0.1679, over 5691349.11 frames. ], libri_tot_loss[loss=0.4467, simple_loss=0.4628, pruned_loss=0.2152, over 5721699.43 frames. ], giga_tot_loss[loss=0.3696, simple_loss=0.4161, pruned_loss=0.1615, over 5690080.50 frames. ], batch size: 128, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:08:56,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38132.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:09:12,471 INFO [train.py:968] (1/2) Epoch 1, batch 38150, libri_loss[loss=0.5226, simple_loss=0.5239, pruned_loss=0.2607, over 19603.00 frames. ], tot_loss[loss=0.3806, simple_loss=0.4235, pruned_loss=0.1688, over 5681157.20 frames. ], libri_tot_loss[loss=0.4473, simple_loss=0.4635, pruned_loss=0.2155, over 5713452.71 frames. ], giga_tot_loss[loss=0.3706, simple_loss=0.4168, pruned_loss=0.1622, over 5687672.58 frames. ], batch size: 187, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:09:35,175 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38179.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:09:37,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38182.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:09:46,218 INFO [optim.py:369] (1/2) 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 21:09:51,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-02-28 21:09:53,994 INFO [train.py:968] (1/2) Epoch 1, batch 38200, giga_loss[loss=0.4185, simple_loss=0.4491, pruned_loss=0.1939, over 28771.00 frames. ], tot_loss[loss=0.382, simple_loss=0.4246, pruned_loss=0.1697, over 5695298.36 frames. ], libri_tot_loss[loss=0.4481, simple_loss=0.4642, pruned_loss=0.216, over 5718324.65 frames. ], giga_tot_loss[loss=0.3718, simple_loss=0.4177, pruned_loss=0.1629, over 5695487.10 frames. ], batch size: 119, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:10:05,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38211.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:10:38,678 INFO [train.py:968] (1/2) Epoch 1, batch 38250, giga_loss[loss=0.3891, simple_loss=0.4496, pruned_loss=0.1643, over 28871.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.4238, pruned_loss=0.1668, over 5700179.23 frames. ], libri_tot_loss[loss=0.4481, simple_loss=0.4643, pruned_loss=0.2159, over 5719021.03 frames. ], giga_tot_loss[loss=0.3704, simple_loss=0.4181, pruned_loss=0.1613, over 5699632.11 frames. ], batch size: 199, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:10:59,498 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38275.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:11:01,577 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:11:19,014 INFO [train.py:968] (1/2) Epoch 1, batch 38300, giga_loss[loss=0.3409, simple_loss=0.4051, pruned_loss=0.1383, over 28674.00 frames. ], tot_loss[loss=0.3791, simple_loss=0.4247, pruned_loss=0.1667, over 5689394.09 frames. ], libri_tot_loss[loss=0.449, simple_loss=0.4647, pruned_loss=0.2166, over 5702152.23 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.4179, pruned_loss=0.1594, over 5703234.34 frames. ], batch size: 78, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:11:26,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38307.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:11:57,580 INFO [train.py:968] (1/2) Epoch 1, batch 38350, giga_loss[loss=0.4035, simple_loss=0.4458, pruned_loss=0.1806, over 28863.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.4249, pruned_loss=0.1662, over 5692998.69 frames. ], libri_tot_loss[loss=0.4487, simple_loss=0.4645, pruned_loss=0.2164, over 5704537.27 frames. ], giga_tot_loss[loss=0.3678, simple_loss=0.4182, pruned_loss=0.1587, over 5701663.71 frames. ], batch size: 99, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:12:16,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=38372.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:12:16,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9496, 2.9522, 2.2711, 1.9416], device='cuda:1'), covar=tensor([0.1287, 0.1109, 0.0984, 0.0648], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0863, 0.0729, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:12:21,881 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0258, 2.9172, 3.7362, 1.5832], device='cuda:1'), covar=tensor([0.0500, 0.0600, 0.0806, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0485, 0.0783, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0006], device='cuda:1')
+2023-02-28 21:12:32,321 INFO [optim.py:369] (1/2) 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 21:12:38,129 INFO [train.py:968] (1/2) Epoch 1, batch 38400, giga_loss[loss=0.3706, simple_loss=0.4097, pruned_loss=0.1658, over 28928.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4215, pruned_loss=0.1637, over 5697767.31 frames. ], libri_tot_loss[loss=0.4482, simple_loss=0.4642, pruned_loss=0.2162, over 5707801.59 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4158, pruned_loss=0.1571, over 5701762.66 frames. ], batch size: 112, lr: 1.80e-02, grad_scale: 8.0
+2023-02-28 21:12:46,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0551, 1.1764, 1.0814, 0.8148], device='cuda:1'), covar=tensor([0.1628, 0.1739, 0.1415, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.0886, 0.0781, 0.0857, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 21:13:21,418 INFO [train.py:968] (1/2) Epoch 1, batch 38450, giga_loss[loss=0.3081, simple_loss=0.3725, pruned_loss=0.1219, over 28809.00 frames. ], tot_loss[loss=0.3705, simple_loss=0.4183, pruned_loss=0.1613, over 5706466.79 frames. ], libri_tot_loss[loss=0.4481, simple_loss=0.4641, pruned_loss=0.216, over 5710673.81 frames. ], giga_tot_loss[loss=0.362, simple_loss=0.4132, pruned_loss=0.1554, over 5706955.56 frames. ], batch size: 119, lr: 1.80e-02, grad_scale: 4.0
+2023-02-28 21:13:26,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4116, 1.4820, 3.4162, 2.7290], device='cuda:1'), covar=tensor([0.1478, 0.1269, 0.0309, 0.0464], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0458, 0.0593, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:13:30,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0645, 2.9215, 3.7949, 1.6646], device='cuda:1'), covar=tensor([0.0517, 0.0658, 0.0792, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0482, 0.0772, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:13:46,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3272, 1.4133, 1.2249, 1.2966], device='cuda:1'), covar=tensor([0.1303, 0.0616, 0.0625, 0.1618], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0270, 0.0266, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0017, 0.0015, 0.0025], device='cuda:1')
+2023-02-28 21:13:47,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9043, 1.6412, 1.2515, 1.3562], device='cuda:1'), covar=tensor([0.0686, 0.0794, 0.1190, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0556, 0.0593, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 21:13:52,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6979, 2.1382, 1.8498, 1.6729], device='cuda:1'), covar=tensor([0.1296, 0.1422, 0.1025, 0.0654], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0851, 0.0717, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:13:58,314 INFO [optim.py:369] (1/2) 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 21:14:03,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-02-28 21:14:03,579 INFO [train.py:968] (1/2) Epoch 1, batch 38500, giga_loss[loss=0.3479, simple_loss=0.4051, pruned_loss=0.1453, over 28961.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4152, pruned_loss=0.1588, over 5717090.77 frames. ], libri_tot_loss[loss=0.4479, simple_loss=0.4639, pruned_loss=0.2159, over 5713847.30 frames. ], giga_tot_loss[loss=0.3589, simple_loss=0.4107, pruned_loss=0.1535, over 5714735.53 frames. ], batch size: 164, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:14:49,049 INFO [train.py:968] (1/2) Epoch 1, batch 38550, giga_loss[loss=0.3594, simple_loss=0.4163, pruned_loss=0.1513, over 29023.00 frames. ], tot_loss[loss=0.3658, simple_loss=0.4145, pruned_loss=0.1586, over 5714223.20 frames. ], libri_tot_loss[loss=0.4484, simple_loss=0.4643, pruned_loss=0.2163, over 5714942.07 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4105, pruned_loss=0.1539, over 5711461.51 frames. ], batch size: 128, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:15:21,543 INFO [optim.py:369] (1/2) 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 21:15:28,096 INFO [train.py:968] (1/2) Epoch 1, batch 38600, libri_loss[loss=0.5113, simple_loss=0.5092, pruned_loss=0.2567, over 19162.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4162, pruned_loss=0.1603, over 5705269.76 frames. ], libri_tot_loss[loss=0.4477, simple_loss=0.4638, pruned_loss=0.2158, over 5710374.26 frames. ], giga_tot_loss[loss=0.3606, simple_loss=0.4116, pruned_loss=0.1548, over 5707846.88 frames. ], batch size: 186, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:16:07,509 INFO [train.py:968] (1/2) Epoch 1, batch 38650, giga_loss[loss=0.3519, simple_loss=0.4052, pruned_loss=0.1493, over 28745.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.416, pruned_loss=0.1591, over 5707624.12 frames. ], libri_tot_loss[loss=0.4482, simple_loss=0.4641, pruned_loss=0.2162, over 5711476.48 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4115, pruned_loss=0.1537, over 5708647.33 frames. ], batch size: 60, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:16:42,008 INFO [optim.py:369] (1/2) 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 21:16:47,515 INFO [train.py:968] (1/2) Epoch 1, batch 38700, giga_loss[loss=0.394, simple_loss=0.4354, pruned_loss=0.1763, over 28661.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4125, pruned_loss=0.1551, over 5709988.66 frames. ], libri_tot_loss[loss=0.4479, simple_loss=0.4638, pruned_loss=0.216, over 5712290.10 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.409, pruned_loss=0.1508, over 5710074.79 frames. ], batch size: 262, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:17:07,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0472, 0.9781, 0.9676, 0.4754], device='cuda:1'), covar=tensor([0.0273, 0.0289, 0.0250, 0.0327], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0573, 0.0624, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 21:17:09,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4933, 1.8467, 1.5532, 0.4680], device='cuda:1'), covar=tensor([0.0566, 0.0465, 0.0694, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.1029, 0.1014, 0.1051, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 21:17:26,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38747.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:17:28,647 INFO [train.py:968] (1/2) Epoch 1, batch 38750, giga_loss[loss=0.3836, simple_loss=0.4326, pruned_loss=0.1673, over 29080.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4134, pruned_loss=0.1569, over 5712087.58 frames. ], libri_tot_loss[loss=0.4481, simple_loss=0.4638, pruned_loss=0.2162, over 5711875.97 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4082, pruned_loss=0.1505, over 5712127.06 frames. ], batch size: 155, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:17:31,843 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=38753.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:18:02,442 INFO [optim.py:369] (1/2) 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 21:18:09,109 INFO [train.py:968] (1/2) Epoch 1, batch 38800, giga_loss[loss=0.3124, simple_loss=0.3718, pruned_loss=0.1264, over 28892.00 frames. ], tot_loss[loss=0.3622, simple_loss=0.4115, pruned_loss=0.1564, over 5717628.11 frames. ], libri_tot_loss[loss=0.4455, simple_loss=0.4618, pruned_loss=0.2146, over 5720478.84 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.407, pruned_loss=0.1503, over 5710053.71 frames. ], batch size: 119, lr: 1.79e-02, grad_scale: 8.0
+2023-02-28 21:18:22,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9918, 2.9280, 2.2126, 0.8562], device='cuda:1'), covar=tensor([0.1441, 0.0703, 0.0990, 0.2070], device='cuda:1'), in_proj_covar=tensor([0.1047, 0.1027, 0.1051, 0.0934], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 21:18:49,237 INFO [train.py:968] (1/2) Epoch 1, batch 38850, giga_loss[loss=0.3255, simple_loss=0.3832, pruned_loss=0.1339, over 28570.00 frames. ], tot_loss[loss=0.3596, simple_loss=0.4089, pruned_loss=0.1552, over 5711104.86 frames. ], libri_tot_loss[loss=0.4457, simple_loss=0.4621, pruned_loss=0.2147, over 5723910.23 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4041, pruned_loss=0.1491, over 5701930.84 frames. ], batch size: 85, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:19:20,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38890.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:19:21,695 INFO [optim.py:369] (1/2) 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 21:19:21,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38893.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:19:26,629 INFO [train.py:968] (1/2) Epoch 1, batch 38900, giga_loss[loss=0.3732, simple_loss=0.4206, pruned_loss=0.1628, over 28825.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.4076, pruned_loss=0.1548, over 5706310.58 frames. ], libri_tot_loss[loss=0.4458, simple_loss=0.4623, pruned_loss=0.2147, over 5713351.56 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4021, pruned_loss=0.1484, over 5708369.07 frames. ], batch size: 186, lr: 1.79e-02, grad_scale: 4.0
+2023-02-28 21:19:45,222 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 38950, giga_loss[loss=0.3419, simple_loss=0.3881, pruned_loss=0.1479, over 28691.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4105, pruned_loss=0.1581, over 5701250.44 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4616, pruned_loss=0.214, over 5710149.77 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.4045, pruned_loss=0.1511, over 5705248.96 frames. ], batch size: 99, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:20:43,278 INFO [optim.py:369] (1/2) 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 21:20:50,321 INFO [train.py:968] (1/2) Epoch 1, batch 39000, libri_loss[loss=0.4705, simple_loss=0.496, pruned_loss=0.2224, over 29556.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4084, pruned_loss=0.1573, over 5697057.59 frames. ], libri_tot_loss[loss=0.4444, simple_loss=0.4613, pruned_loss=0.2137, over 5712937.48 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4031, pruned_loss=0.1512, over 5697597.09 frames. ], batch size: 89, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:20:50,321 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 21:20:59,056 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-02-28 21:21:21,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7952, 1.9540, 3.9610, 3.0582], device='cuda:1'), covar=tensor([0.1447, 0.1222, 0.0281, 0.0417], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0462, 0.0594, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:21:33,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7331, 1.7818, 1.2177, 1.3989], device='cuda:1'), covar=tensor([0.0590, 0.0606, 0.1081, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0552, 0.0578, 0.0519], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 21:21:39,871 INFO [train.py:968] (1/2) Epoch 1, batch 39050, giga_loss[loss=0.3503, simple_loss=0.3999, pruned_loss=0.1504, over 28951.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.4064, pruned_loss=0.1563, over 5698805.52 frames. ], libri_tot_loss[loss=0.4433, simple_loss=0.4607, pruned_loss=0.2129, over 5706806.44 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4009, pruned_loss=0.1501, over 5704083.50 frames. ], batch size: 145, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:22:13,724 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 39100, giga_loss[loss=0.3463, simple_loss=0.3973, pruned_loss=0.1476, over 28953.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4036, pruned_loss=0.1547, over 5709399.58 frames. ], libri_tot_loss[loss=0.4426, simple_loss=0.4603, pruned_loss=0.2125, over 5710497.44 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.3981, pruned_loss=0.1488, over 5710052.96 frames. ], batch size: 186, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:22:41,821 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:22:43,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5921, 1.5378, 1.3886, 1.6499], device='cuda:1'), covar=tensor([0.1415, 0.1505, 0.1183, 0.1657], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0774, 0.0850, 0.0902], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-02-28 21:22:59,850 INFO [train.py:968] (1/2) Epoch 1, batch 39150, giga_loss[loss=0.305, simple_loss=0.366, pruned_loss=0.122, over 28882.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.402, pruned_loss=0.1544, over 5705132.24 frames. ], libri_tot_loss[loss=0.4432, simple_loss=0.4609, pruned_loss=0.2127, over 5714463.79 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3959, pruned_loss=0.1481, over 5701943.13 frames. ], batch size: 199, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:23:15,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-02-28 21:23:35,816 INFO [optim.py:369] (1/2) 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 21:23:40,337 INFO [train.py:968] (1/2) Epoch 1, batch 39200, giga_loss[loss=0.353, simple_loss=0.3953, pruned_loss=0.1554, over 28494.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4018, pruned_loss=0.154, over 5712552.04 frames. ], libri_tot_loss[loss=0.4438, simple_loss=0.4615, pruned_loss=0.2131, over 5718552.71 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3941, pruned_loss=0.1464, over 5706035.57 frames. ], batch size: 71, lr: 1.78e-02, grad_scale: 8.0
+2023-02-28 21:23:55,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39216.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:24:01,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5343, 1.5874, 3.7533, 2.8552], device='cuda:1'), covar=tensor([0.1683, 0.1448, 0.0313, 0.0601], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0463, 0.0588, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:24:15,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-02-28 21:24:25,880 INFO [train.py:968] (1/2) Epoch 1, batch 39250, giga_loss[loss=0.4246, simple_loss=0.4453, pruned_loss=0.202, over 28759.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4052, pruned_loss=0.1557, over 5700262.12 frames. ], libri_tot_loss[loss=0.4431, simple_loss=0.4609, pruned_loss=0.2126, over 5712274.31 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.398, pruned_loss=0.1485, over 5701265.39 frames. ], batch size: 92, lr: 1.78e-02, grad_scale: 8.0
+2023-02-28 21:24:43,214 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39271.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:24:45,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39274.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:25:02,812 INFO [optim.py:369] (1/2) 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 21:25:08,030 INFO [train.py:968] (1/2) Epoch 1, batch 39300, giga_loss[loss=0.3916, simple_loss=0.4324, pruned_loss=0.1754, over 28063.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.4066, pruned_loss=0.1561, over 5674752.01 frames. ], libri_tot_loss[loss=0.4425, simple_loss=0.4605, pruned_loss=0.2123, over 5696686.57 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4001, pruned_loss=0.1494, over 5689768.57 frames. ], batch size: 412, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:25:11,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39303.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:25:51,134 INFO [train.py:968] (1/2) Epoch 1, batch 39350, giga_loss[loss=0.3084, simple_loss=0.3785, pruned_loss=0.1191, over 28972.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4085, pruned_loss=0.1563, over 5671308.11 frames. ], libri_tot_loss[loss=0.4436, simple_loss=0.4614, pruned_loss=0.213, over 5683729.49 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4011, pruned_loss=0.1489, over 5694528.75 frames. ], batch size: 164, lr: 1.78e-02, grad_scale: 4.0
+2023-02-28 21:26:26,901 INFO [optim.py:369] (1/2) 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 21:26:31,767 INFO [train.py:968] (1/2) Epoch 1, batch 39400, giga_loss[loss=0.3355, simple_loss=0.4016, pruned_loss=0.1347, over 28738.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4104, pruned_loss=0.1573, over 5678603.94 frames. ], libri_tot_loss[loss=0.4437, simple_loss=0.4615, pruned_loss=0.213, over 5691554.18 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4021, pruned_loss=0.1489, over 5689936.39 frames. ], batch size: 262, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:27:16,993 INFO [train.py:968] (1/2) Epoch 1, batch 39450, giga_loss[loss=0.3159, simple_loss=0.3717, pruned_loss=0.13, over 28765.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4072, pruned_loss=0.1544, over 5689365.59 frames. ], libri_tot_loss[loss=0.4434, simple_loss=0.4612, pruned_loss=0.2128, over 5693697.36 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.4005, pruned_loss=0.1475, over 5696371.18 frames. ], batch size: 99, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:27:54,511 INFO [optim.py:369] (1/2) 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 21:27:57,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-02-28 21:28:00,545 INFO [train.py:968] (1/2) Epoch 1, batch 39500, giga_loss[loss=0.3478, simple_loss=0.3989, pruned_loss=0.1484, over 28889.00 frames. ], tot_loss[loss=0.359, simple_loss=0.4078, pruned_loss=0.1551, over 5696608.12 frames. ], libri_tot_loss[loss=0.4437, simple_loss=0.4616, pruned_loss=0.2129, over 5697612.77 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4011, pruned_loss=0.1484, over 5698719.71 frames. ], batch size: 106, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:28:13,977 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-02-28 21:28:41,299 INFO [train.py:968] (1/2) Epoch 1, batch 39550, giga_loss[loss=0.3947, simple_loss=0.4198, pruned_loss=0.1848, over 28698.00 frames. ], tot_loss[loss=0.3627, simple_loss=0.4103, pruned_loss=0.1575, over 5701147.32 frames. ], libri_tot_loss[loss=0.4446, simple_loss=0.4623, pruned_loss=0.2134, over 5690577.13 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4031, pruned_loss=0.1505, over 5709628.50 frames. ], batch size: 92, lr: 1.77e-02, grad_scale: 4.0
+2023-02-28 21:29:08,855 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39578.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:29:11,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.10 vs. limit=2.0
+2023-02-28 21:29:21,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39591.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:29:23,434 INFO [optim.py:369] (1/2) 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 21:29:29,320 INFO [train.py:968] (1/2) Epoch 1, batch 39600, giga_loss[loss=0.3661, simple_loss=0.4031, pruned_loss=0.1646, over 23647.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4123, pruned_loss=0.1588, over 5701665.62 frames. ], libri_tot_loss[loss=0.4446, simple_loss=0.4623, pruned_loss=0.2135, over 5693656.65 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4057, pruned_loss=0.1522, over 5705867.64 frames. ], batch size: 705, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:29:57,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2457, 2.4429, 2.9184, 1.7934], device='cuda:1'), covar=tensor([0.0581, 0.0612, 0.0840, 0.1477], device='cuda:1'), in_proj_covar=tensor([0.0713, 0.0481, 0.0764, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:30:10,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2519, 1.4097, 1.2835, 1.4445], device='cuda:1'), covar=tensor([0.1733, 0.1754, 0.1376, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0789, 0.0852, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 21:30:10,804 INFO [train.py:968] (1/2) Epoch 1, batch 39650, libri_loss[loss=0.3812, simple_loss=0.4119, pruned_loss=0.1753, over 29349.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4163, pruned_loss=0.1615, over 5694599.62 frames. ], libri_tot_loss[loss=0.4441, simple_loss=0.462, pruned_loss=0.2131, over 5687107.06 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4103, pruned_loss=0.1554, over 5703753.98 frames. ], batch size: 67, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:30:21,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3116, 2.0573, 1.6230, 1.3673], device='cuda:1'), covar=tensor([0.1168, 0.0461, 0.0569, 0.1427], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0266, 0.0269, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0018, 0.0016, 0.0026], device='cuda:1')
+2023-02-28 21:30:34,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:30:47,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1685, 1.5100, 1.3315, 0.2024], device='cuda:1'), covar=tensor([0.1017, 0.0871, 0.1356, 0.1644], device='cuda:1'), in_proj_covar=tensor([0.1080, 0.1061, 0.1114, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 21:30:48,556 INFO [optim.py:369] (1/2) 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 21:30:52,727 INFO [train.py:968] (1/2) Epoch 1, batch 39700, giga_loss[loss=0.3593, simple_loss=0.4222, pruned_loss=0.1482, over 28916.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4169, pruned_loss=0.161, over 5704972.74 frames. ], libri_tot_loss[loss=0.4442, simple_loss=0.4622, pruned_loss=0.2131, over 5689466.16 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4112, pruned_loss=0.1554, over 5710106.40 frames. ], batch size: 164, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:30:55,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1285, 1.1289, 0.9736, 0.9140], device='cuda:1'), covar=tensor([0.0598, 0.0614, 0.1016, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0566, 0.0576, 0.0522], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 21:30:58,725 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 21:31:22,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39734.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:31:24,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39737.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:31:35,542 INFO [train.py:968] (1/2) Epoch 1, batch 39750, giga_loss[loss=0.3751, simple_loss=0.4271, pruned_loss=0.1615, over 28732.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.417, pruned_loss=0.1606, over 5704205.14 frames. ], libri_tot_loss[loss=0.4446, simple_loss=0.4627, pruned_loss=0.2133, over 5692499.67 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4115, pruned_loss=0.1552, over 5705642.97 frames. ], batch size: 242, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:31:49,056 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39766.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:32:13,230 INFO [optim.py:369] (1/2) 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 21:32:17,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5478, 1.5300, 1.1427, 1.2822], device='cuda:1'), covar=tensor([0.0741, 0.0841, 0.1193, 0.1124], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0574, 0.0586, 0.0529], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 21:32:17,972 INFO [train.py:968] (1/2) Epoch 1, batch 39800, giga_loss[loss=0.3886, simple_loss=0.4387, pruned_loss=0.1692, over 28953.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4188, pruned_loss=0.162, over 5702294.60 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4628, pruned_loss=0.2134, over 5691182.98 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4129, pruned_loss=0.1561, over 5704910.36 frames. ], batch size: 213, lr: 1.77e-02, grad_scale: 8.0
+2023-02-28 21:32:53,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5509, 1.9368, 1.6161, 0.8343], device='cuda:1'), covar=tensor([0.1330, 0.0921, 0.1353, 0.1733], device='cuda:1'), in_proj_covar=tensor([0.1078, 0.1068, 0.1123, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 21:32:59,404 INFO [train.py:968] (1/2) Epoch 1, batch 39850, giga_loss[loss=0.3458, simple_loss=0.4083, pruned_loss=0.1416, over 28993.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4171, pruned_loss=0.1603, over 5706206.24 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4629, pruned_loss=0.2134, over 5691483.19 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4119, pruned_loss=0.1551, over 5708176.79 frames. ], batch size: 227, lr: 1.76e-02, grad_scale: 8.0
+2023-02-28 21:33:00,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4052, 1.2781, 1.3779, 0.7560], device='cuda:1'), covar=tensor([0.0305, 0.0250, 0.0206, 0.0292], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0605, 0.0674, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 21:33:04,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=39856.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:33:23,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7037, 2.0323, 1.7136, 1.6610], device='cuda:1'), covar=tensor([0.1301, 0.1283, 0.1059, 0.0658], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0825, 0.0708, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:33:34,305 INFO [optim.py:369] (1/2) 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 21:33:38,566 INFO [train.py:968] (1/2) Epoch 1, batch 39900, giga_loss[loss=0.3676, simple_loss=0.4169, pruned_loss=0.1591, over 28743.00 frames. ], tot_loss[loss=0.3669, simple_loss=0.4148, pruned_loss=0.1595, over 5703749.78 frames. ], libri_tot_loss[loss=0.4449, simple_loss=0.4629, pruned_loss=0.2135, over 5686356.65 frames. ], giga_tot_loss[loss=0.3589, simple_loss=0.4096, pruned_loss=0.1541, over 5710449.65 frames. ], batch size: 284, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:34:21,096 INFO [train.py:968] (1/2) Epoch 1, batch 39950, giga_loss[loss=0.4015, simple_loss=0.4297, pruned_loss=0.1866, over 26578.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.409, pruned_loss=0.1554, over 5706188.15 frames. ], libri_tot_loss[loss=0.445, simple_loss=0.463, pruned_loss=0.2134, over 5687901.05 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4041, pruned_loss=0.1505, over 5710346.22 frames. ], batch size: 555, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:34:23,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39953.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:34:58,991 INFO [optim.py:369] (1/2) 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 21:35:02,626 INFO [train.py:968] (1/2) Epoch 1, batch 40000, giga_loss[loss=0.3456, simple_loss=0.4114, pruned_loss=0.1399, over 28371.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4064, pruned_loss=0.1532, over 5709298.47 frames. ], libri_tot_loss[loss=0.4449, simple_loss=0.463, pruned_loss=0.2134, over 5689294.12 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4011, pruned_loss=0.1477, over 5711867.45 frames. ], batch size: 369, lr: 1.76e-02, grad_scale: 8.0
+2023-02-28 21:35:17,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-02-28 21:35:29,538 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40032.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:35:37,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5543, 2.3289, 1.7021, 1.5915], device='cuda:1'), covar=tensor([0.1157, 0.1202, 0.1104, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0818, 0.0708, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:35:43,932 INFO [train.py:968] (1/2) Epoch 1, batch 40050, giga_loss[loss=0.3297, simple_loss=0.403, pruned_loss=0.1282, over 29065.00 frames. ], tot_loss[loss=0.3575, simple_loss=0.4089, pruned_loss=0.153, over 5717317.08 frames. ], libri_tot_loss[loss=0.4442, simple_loss=0.4625, pruned_loss=0.213, over 5694792.79 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4038, pruned_loss=0.1477, over 5714965.71 frames. ], batch size: 128, lr: 1.76e-02, grad_scale: 8.0
+2023-02-28 21:35:45,970 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40052.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:35:53,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 21:36:06,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-02-28 21:36:25,173 INFO [optim.py:369] (1/2) 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 21:36:25,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40096.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:36:27,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40099.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:36:28,223 INFO [train.py:968] (1/2) Epoch 1, batch 40100, giga_loss[loss=0.3432, simple_loss=0.4087, pruned_loss=0.1388, over 27932.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4107, pruned_loss=0.154, over 5702208.95 frames. ], libri_tot_loss[loss=0.4444, simple_loss=0.4626, pruned_loss=0.2131, over 5687935.18 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4058, pruned_loss=0.1487, over 5706893.54 frames. ], batch size: 412, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:36:39,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5474, 1.4213, 1.3623, 1.2468], device='cuda:1'), covar=tensor([0.0726, 0.1200, 0.1176, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0846, 0.0655, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 21:36:52,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40128.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:36:56,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9091, 2.7251, 1.8869, 1.8672], device='cuda:1'), covar=tensor([0.0877, 0.0695, 0.0886, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0450, 0.0360, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 21:37:04,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4117, 1.8362, 1.5942, 1.5855], device='cuda:1'), covar=tensor([0.1037, 0.1309, 0.0985, 0.0614], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0817, 0.0709, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:37:09,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-02-28 21:37:10,992 INFO [train.py:968] (1/2) Epoch 1, batch 40150, giga_loss[loss=0.3164, simple_loss=0.3719, pruned_loss=0.1304, over 28819.00 frames. ], tot_loss[loss=0.3611, simple_loss=0.4111, pruned_loss=0.1556, over 5707140.04 frames. ], libri_tot_loss[loss=0.4448, simple_loss=0.4629, pruned_loss=0.2133, over 5691257.37 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4062, pruned_loss=0.1503, over 5707949.14 frames. ], batch size: 112, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:37:18,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40160.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:37:45,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40195.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:37:46,128 INFO [optim.py:369] (1/2) 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 21:37:47,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40198.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:37:48,469 INFO [train.py:968] (1/2) Epoch 1, batch 40200, giga_loss[loss=0.3155, simple_loss=0.3763, pruned_loss=0.1274, over 29033.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.413, pruned_loss=0.1592, over 5700961.54 frames. ], libri_tot_loss[loss=0.4441, simple_loss=0.4626, pruned_loss=0.2128, over 5686175.22 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.407, pruned_loss=0.1529, over 5706739.43 frames. ], batch size: 164, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:37:53,964 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40208.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:38:07,930 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40231.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:38:28,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0618, 2.1120, 2.6334, 1.7225], device='cuda:1'), covar=tensor([0.0586, 0.1759, 0.0840, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0861, 0.0659, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 21:38:28,921 INFO [train.py:968] (1/2) Epoch 1, batch 40250, giga_loss[loss=0.3318, simple_loss=0.383, pruned_loss=0.1403, over 28504.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4122, pruned_loss=0.1606, over 5705923.74 frames. ], libri_tot_loss[loss=0.4428, simple_loss=0.4615, pruned_loss=0.2121, over 5696103.60 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.406, pruned_loss=0.1537, over 5702441.85 frames. ], batch size: 85, lr: 1.76e-02, grad_scale: 4.0
+2023-02-28 21:39:02,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4459, 2.1417, 1.6227, 1.0205], device='cuda:1'), covar=tensor([0.0369, 0.0325, 0.0244, 0.0404], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0629, 0.0695, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 21:39:07,136 INFO [optim.py:369] (1/2) 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 21:39:10,683 INFO [train.py:968] (1/2) Epoch 1, batch 40300, giga_loss[loss=0.2914, simple_loss=0.3536, pruned_loss=0.1146, over 29041.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.4106, pruned_loss=0.1604, over 5716512.44 frames. ], libri_tot_loss[loss=0.4424, simple_loss=0.4612, pruned_loss=0.2117, over 5701056.17 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4048, pruned_loss=0.1541, over 5709587.83 frames. ], batch size: 155, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:39:37,141 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-02-28 21:39:50,277 INFO [train.py:968] (1/2) Epoch 1, batch 40350, giga_loss[loss=0.3215, simple_loss=0.3692, pruned_loss=0.1369, over 28756.00 frames. ], tot_loss[loss=0.3647, simple_loss=0.4094, pruned_loss=0.16, over 5717255.51 frames. ], libri_tot_loss[loss=0.4426, simple_loss=0.4614, pruned_loss=0.2119, over 5698287.88 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4029, pruned_loss=0.1532, over 5714124.03 frames. ], batch size: 99, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:40:05,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4282, 1.8469, 1.4530, 1.2807], device='cuda:1'), covar=tensor([0.1231, 0.0921, 0.1231, 0.1932], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0442, 0.0358, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 21:40:10,448 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40374.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:40:13,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40377.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:40:28,546 INFO [optim.py:369] (1/2) 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 21:40:31,281 INFO [train.py:968] (1/2) Epoch 1, batch 40400, giga_loss[loss=0.3149, simple_loss=0.3656, pruned_loss=0.1321, over 29020.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.406, pruned_loss=0.1574, over 5724971.99 frames. ], libri_tot_loss[loss=0.4421, simple_loss=0.4611, pruned_loss=0.2116, over 5704174.80 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.3997, pruned_loss=0.1507, over 5717786.83 frames. ], batch size: 136, lr: 1.75e-02, grad_scale: 8.0
+2023-02-28 21:40:36,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:40:37,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40407.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:41:12,991 INFO [train.py:968] (1/2) Epoch 1, batch 40450, giga_loss[loss=0.2854, simple_loss=0.3474, pruned_loss=0.1117, over 28870.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4007, pruned_loss=0.1542, over 5716063.95 frames. ], libri_tot_loss[loss=0.4424, simple_loss=0.4613, pruned_loss=0.2118, over 5699572.27 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3942, pruned_loss=0.1475, over 5715127.66 frames. ], batch size: 199, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:41:49,826 INFO [optim.py:369] (1/2) 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 21:41:52,525 INFO [train.py:968] (1/2) Epoch 1, batch 40500, giga_loss[loss=0.3162, simple_loss=0.3773, pruned_loss=0.1275, over 28735.00 frames. ], tot_loss[loss=0.35, simple_loss=0.3964, pruned_loss=0.1518, over 5720933.72 frames. ], libri_tot_loss[loss=0.4409, simple_loss=0.46, pruned_loss=0.2109, over 5706228.24 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3894, pruned_loss=0.1445, over 5714949.91 frames. ], batch size: 262, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:42:19,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40535.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:42:29,477 INFO [train.py:968] (1/2) Epoch 1, batch 40550, giga_loss[loss=0.3545, simple_loss=0.3997, pruned_loss=0.1546, over 28731.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.3956, pruned_loss=0.1508, over 5698765.30 frames. ], libri_tot_loss[loss=0.4413, simple_loss=0.4603, pruned_loss=0.2112, over 5684648.49 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3882, pruned_loss=0.1432, over 5714081.51 frames. ], batch size: 92, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:42:29,791 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40550.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 21:42:32,531 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40553.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:42:54,821 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40582.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 21:42:56,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40583.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:42:59,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:43:09,790 INFO [train.py:968] (1/2) Epoch 1, batch 40600, giga_loss[loss=0.3132, simple_loss=0.3616, pruned_loss=0.1324, over 28430.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4006, pruned_loss=0.1541, over 5706248.34 frames. ], libri_tot_loss[loss=0.4406, simple_loss=0.4598, pruned_loss=0.2107, over 5693016.49 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3927, pruned_loss=0.1462, over 5711654.16 frames. ], batch size: 78, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:43:18,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40611.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:43:51,038 INFO [train.py:968] (1/2) Epoch 1, batch 40650, libri_loss[loss=0.417, simple_loss=0.4265, pruned_loss=0.2037, over 29367.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4043, pruned_loss=0.1556, over 5713061.63 frames. ], libri_tot_loss[loss=0.4399, simple_loss=0.4593, pruned_loss=0.2102, over 5698687.15 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.3966, pruned_loss=0.1478, over 5712640.79 frames. ], batch size: 67, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:44:02,765 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40665.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:12,246 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40678.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:14,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40681.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:20,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4978, 2.1152, 1.5153, 1.2840], device='cuda:1'), covar=tensor([0.0951, 0.0717, 0.0989, 0.1509], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0444, 0.0361, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 21:44:23,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-02-28 21:44:27,274 INFO [optim.py:369] (1/2) 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 21:44:29,981 INFO [train.py:968] (1/2) Epoch 1, batch 40700, giga_loss[loss=0.3495, simple_loss=0.4043, pruned_loss=0.1474, over 28732.00 frames. ], tot_loss[loss=0.3608, simple_loss=0.4078, pruned_loss=0.1569, over 5715734.71 frames. ], libri_tot_loss[loss=0.4394, simple_loss=0.4589, pruned_loss=0.21, over 5701880.58 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4007, pruned_loss=0.1497, over 5712803.80 frames. ], batch size: 99, lr: 1.75e-02, grad_scale: 4.0
+2023-02-28 21:44:37,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40710.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:49,324 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40723.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:52,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40726.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:44:54,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40729.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:45:07,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-02-28 21:45:11,804 INFO [train.py:968] (1/2) Epoch 1, batch 40750, giga_loss[loss=0.4145, simple_loss=0.438, pruned_loss=0.1955, over 23892.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4106, pruned_loss=0.1583, over 5712581.91 frames. ], libri_tot_loss[loss=0.4381, simple_loss=0.458, pruned_loss=0.2091, over 5698711.22 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4041, pruned_loss=0.1516, over 5714295.77 frames. ], batch size: 705, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:45:19,740 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40758.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:45:51,465 INFO [optim.py:369] (1/2) 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 21:45:54,029 INFO [train.py:968] (1/2) Epoch 1, batch 40800, giga_loss[loss=0.3704, simple_loss=0.4172, pruned_loss=0.1618, over 28561.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4133, pruned_loss=0.1601, over 5704167.88 frames. ], libri_tot_loss[loss=0.4383, simple_loss=0.4582, pruned_loss=0.2092, over 5693583.16 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4069, pruned_loss=0.1535, over 5709836.69 frames. ], batch size: 60, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:46:45,120 INFO [train.py:968] (1/2) Epoch 1, batch 40850, giga_loss[loss=0.4882, simple_loss=0.5008, pruned_loss=0.2378, over 28448.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.4194, pruned_loss=0.1668, over 5687350.40 frames. ], libri_tot_loss[loss=0.4379, simple_loss=0.4578, pruned_loss=0.209, over 5687293.14 frames. ], giga_tot_loss[loss=0.3676, simple_loss=0.4137, pruned_loss=0.1608, over 5697311.58 frames. ], batch size: 336, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:46:59,924 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-02-28 21:47:20,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=40885.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:47:33,119 INFO [optim.py:369] (1/2) 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 21:47:35,464 INFO [train.py:968] (1/2) Epoch 1, batch 40900, giga_loss[loss=0.3708, simple_loss=0.4176, pruned_loss=0.162, over 28877.00 frames. ], tot_loss[loss=0.3891, simple_loss=0.4281, pruned_loss=0.175, over 5668720.95 frames. ], libri_tot_loss[loss=0.4383, simple_loss=0.4581, pruned_loss=0.2093, over 5680646.54 frames. ], giga_tot_loss[loss=0.381, simple_loss=0.4229, pruned_loss=0.1695, over 5682228.66 frames. ], batch size: 112, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:47:54,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 21:48:21,450 INFO [train.py:968] (1/2) Epoch 1, batch 40950, giga_loss[loss=0.4455, simple_loss=0.4721, pruned_loss=0.2095, over 28992.00 frames. ], tot_loss[loss=0.3961, simple_loss=0.434, pruned_loss=0.1791, over 5679366.70 frames. ], libri_tot_loss[loss=0.438, simple_loss=0.4581, pruned_loss=0.2089, over 5687494.06 frames. ], giga_tot_loss[loss=0.3885, simple_loss=0.429, pruned_loss=0.174, over 5684030.08 frames. ], batch size: 213, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:48:34,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40963.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:48:57,336 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:49:04,955 INFO [optim.py:369] (1/2) 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 21:49:06,311 INFO [train.py:968] (1/2) Epoch 1, batch 41000, giga_loss[loss=0.4139, simple_loss=0.4516, pruned_loss=0.1881, over 28975.00 frames. ], tot_loss[loss=0.4052, simple_loss=0.4398, pruned_loss=0.1854, over 5659956.63 frames. ], libri_tot_loss[loss=0.438, simple_loss=0.4581, pruned_loss=0.2089, over 5680302.90 frames. ], giga_tot_loss[loss=0.3981, simple_loss=0.4351, pruned_loss=0.1806, over 5669619.60 frames. ], batch size: 145, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:49:45,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41040.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:49:49,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41044.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:49:53,497 INFO [train.py:968] (1/2) Epoch 1, batch 41050, giga_loss[loss=0.498, simple_loss=0.4944, pruned_loss=0.2508, over 27867.00 frames. ], tot_loss[loss=0.4155, simple_loss=0.4468, pruned_loss=0.1921, over 5671676.49 frames. ], libri_tot_loss[loss=0.4372, simple_loss=0.4576, pruned_loss=0.2084, over 5685036.74 frames. ], giga_tot_loss[loss=0.4097, simple_loss=0.4431, pruned_loss=0.1881, over 5674589.94 frames. ], batch size: 412, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:50:39,529 INFO [optim.py:369] (1/2) 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 21:50:40,145 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 41100, giga_loss[loss=0.5176, simple_loss=0.505, pruned_loss=0.2651, over 27449.00 frames. ], tot_loss[loss=0.4226, simple_loss=0.4515, pruned_loss=0.1968, over 5651195.59 frames. ], libri_tot_loss[loss=0.4361, simple_loss=0.457, pruned_loss=0.2076, over 5680255.06 frames. ], giga_tot_loss[loss=0.4182, simple_loss=0.4488, pruned_loss=0.1939, over 5658168.21 frames. ], batch size: 472, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:50:48,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:50:51,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41109.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:15,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41129.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:17,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41132.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:24,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41138.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:51:35,874 INFO [train.py:968] (1/2) Epoch 1, batch 41150, giga_loss[loss=0.5233, simple_loss=0.5094, pruned_loss=0.2686, over 28282.00 frames. ], tot_loss[loss=0.424, simple_loss=0.4523, pruned_loss=0.1978, over 5652451.83 frames. ], libri_tot_loss[loss=0.4361, simple_loss=0.4571, pruned_loss=0.2076, over 5674638.73 frames. ], giga_tot_loss[loss=0.4203, simple_loss=0.45, pruned_loss=0.1953, over 5663230.32 frames. ], batch size: 368, lr: 1.74e-02, grad_scale: 4.0
+2023-02-28 21:51:49,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41161.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:52:08,572 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 21:52:15,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41183.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:52:19,541 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:52:33,128 INFO [train.py:968] (1/2) Epoch 1, batch 41200, giga_loss[loss=0.4442, simple_loss=0.4747, pruned_loss=0.2068, over 28776.00 frames. ], tot_loss[loss=0.4349, simple_loss=0.4585, pruned_loss=0.2057, over 5626178.96 frames. ], libri_tot_loss[loss=0.4371, simple_loss=0.4577, pruned_loss=0.2083, over 5678724.93 frames. ], giga_tot_loss[loss=0.4308, simple_loss=0.4559, pruned_loss=0.2029, over 5630217.91 frames. ], batch size: 174, lr: 1.74e-02, grad_scale: 8.0
+2023-02-28 21:52:43,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4905, 1.5895, 1.3643, 1.4381], device='cuda:1'), covar=tensor([0.1133, 0.0442, 0.0558, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0261, 0.0263, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0018, 0.0016, 0.0027], device='cuda:1')
+2023-02-28 21:52:46,224 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41215.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:13,219 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41241.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:15,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 41250, giga_loss[loss=0.5074, simple_loss=0.5057, pruned_loss=0.2546, over 28740.00 frames. ], tot_loss[loss=0.44, simple_loss=0.4614, pruned_loss=0.2093, over 5626122.75 frames. ], libri_tot_loss[loss=0.4368, simple_loss=0.4576, pruned_loss=0.208, over 5685271.51 frames. ], giga_tot_loss[loss=0.4369, simple_loss=0.4595, pruned_loss=0.2072, over 5621013.23 frames. ], batch size: 284, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:53:29,249 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41258.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:30,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41260.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:53:43,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5518, 2.7181, 3.2828, 1.5668], device='cuda:1'), covar=tensor([0.0670, 0.0705, 0.1055, 0.1858], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0509, 0.0827, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-02-28 21:53:45,604 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41273.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:54:08,837 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41293.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:54:13,072 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 41300, giga_loss[loss=0.4318, simple_loss=0.4705, pruned_loss=0.1965, over 28219.00 frames. ], tot_loss[loss=0.4434, simple_loss=0.4642, pruned_loss=0.2113, over 5632907.19 frames. ], libri_tot_loss[loss=0.4351, simple_loss=0.4564, pruned_loss=0.2069, over 5687204.18 frames. ], giga_tot_loss[loss=0.4428, simple_loss=0.464, pruned_loss=0.2108, over 5624771.07 frames. ], batch size: 77, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:54:44,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6852, 2.3466, 1.6326, 1.3677], device='cuda:1'), covar=tensor([0.0922, 0.0624, 0.0834, 0.1541], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0453, 0.0362, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0013], device='cuda:1')
+2023-02-28 21:55:05,720 INFO [train.py:968] (1/2) Epoch 1, batch 41350, giga_loss[loss=0.4848, simple_loss=0.4902, pruned_loss=0.2397, over 28712.00 frames. ], tot_loss[loss=0.4458, simple_loss=0.4656, pruned_loss=0.213, over 5618080.25 frames. ], libri_tot_loss[loss=0.4352, simple_loss=0.4565, pruned_loss=0.2069, over 5667206.42 frames. ], giga_tot_loss[loss=0.4456, simple_loss=0.4657, pruned_loss=0.2128, over 5627561.28 frames. ], batch size: 262, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:55:26,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3753, 1.3912, 2.9312, 2.4867], device='cuda:1'), covar=tensor([0.1384, 0.1210, 0.0408, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0473, 0.0623, 0.0475], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:55:44,450 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:55:53,474 INFO [train.py:968] (1/2) Epoch 1, batch 41400, libri_loss[loss=0.4962, simple_loss=0.4874, pruned_loss=0.2525, over 25990.00 frames. ], tot_loss[loss=0.4433, simple_loss=0.4631, pruned_loss=0.2117, over 5618881.58 frames. ], libri_tot_loss[loss=0.4347, simple_loss=0.4563, pruned_loss=0.2066, over 5664987.73 frames. ], giga_tot_loss[loss=0.4437, simple_loss=0.4636, pruned_loss=0.212, over 5627638.02 frames. ], batch size: 136, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:55:53,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6777, 2.2012, 1.8390, 1.7843], device='cuda:1'), covar=tensor([0.1270, 0.1310, 0.1045, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0846, 0.0706, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 21:55:56,644 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41403.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:00,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41406.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:11,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41419.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:32,011 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41438.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:56:47,334 INFO [train.py:968] (1/2) Epoch 1, batch 41450, giga_loss[loss=0.5992, simple_loss=0.5484, pruned_loss=0.325, over 26730.00 frames. ], tot_loss[loss=0.4412, simple_loss=0.4615, pruned_loss=0.2104, over 5619853.56 frames. ], libri_tot_loss[loss=0.4346, simple_loss=0.4561, pruned_loss=0.2065, over 5668431.38 frames. ], giga_tot_loss[loss=0.4418, simple_loss=0.4621, pruned_loss=0.2107, over 5623123.91 frames. ], batch size: 555, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:56:52,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1061, 1.1856, 1.0969, 1.2217], device='cuda:1'), covar=tensor([0.1809, 0.1770, 0.1501, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0793, 0.0872, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 21:57:36,328 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=41495.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:57:38,800 INFO [optim.py:369] (1/2) 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 21:57:40,728 INFO [train.py:968] (1/2) Epoch 1, batch 41500, giga_loss[loss=0.4898, simple_loss=0.4947, pruned_loss=0.2425, over 28282.00 frames. ], tot_loss[loss=0.4386, simple_loss=0.4605, pruned_loss=0.2083, over 5615488.45 frames. ], libri_tot_loss[loss=0.4346, simple_loss=0.456, pruned_loss=0.2066, over 5670929.65 frames. ], giga_tot_loss[loss=0.4391, simple_loss=0.4612, pruned_loss=0.2085, over 5615154.34 frames. ], batch size: 368, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:58:32,972 INFO [train.py:968] (1/2) Epoch 1, batch 41550, libri_loss[loss=0.4727, simple_loss=0.4881, pruned_loss=0.2286, over 29291.00 frames. ], tot_loss[loss=0.4414, simple_loss=0.4624, pruned_loss=0.2102, over 5611361.49 frames. ], libri_tot_loss[loss=0.4341, simple_loss=0.4556, pruned_loss=0.2063, over 5675171.27 frames. ], giga_tot_loss[loss=0.4424, simple_loss=0.4634, pruned_loss=0.2107, over 5605554.07 frames. ], batch size: 94, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 21:58:47,250 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41562.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:58:52,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41565.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:58:57,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-02-28 21:59:22,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41594.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 21:59:28,377 INFO [optim.py:369] (1/2) 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 21:59:29,046 INFO [train.py:968] (1/2) Epoch 1, batch 41600, giga_loss[loss=0.3986, simple_loss=0.4438, pruned_loss=0.1767, over 28863.00 frames. ], tot_loss[loss=0.4395, simple_loss=0.461, pruned_loss=0.2089, over 5608247.02 frames. ], libri_tot_loss[loss=0.4341, simple_loss=0.4556, pruned_loss=0.2063, over 5679980.08 frames. ], giga_tot_loss[loss=0.4404, simple_loss=0.462, pruned_loss=0.2094, over 5597934.14 frames. ], batch size: 174, lr: 1.73e-02, grad_scale: 8.0
+2023-02-28 22:00:03,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41633.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:00:18,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5024, 1.5304, 2.9384, 2.6807], device='cuda:1'), covar=tensor([0.1380, 0.1175, 0.0434, 0.0410], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0470, 0.0614, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 22:00:18,655 INFO [train.py:968] (1/2) Epoch 1, batch 41650, giga_loss[loss=0.4202, simple_loss=0.4582, pruned_loss=0.1911, over 29043.00 frames. ], tot_loss[loss=0.4332, simple_loss=0.4579, pruned_loss=0.2042, over 5623320.12 frames. ], libri_tot_loss[loss=0.4334, simple_loss=0.4551, pruned_loss=0.2059, over 5683129.69 frames. ], giga_tot_loss[loss=0.4347, simple_loss=0.4593, pruned_loss=0.2051, over 5610322.89 frames. ], batch size: 128, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 22:00:36,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41668.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:00:58,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 22:01:09,899 INFO [optim.py:369] (1/2) 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 22:01:09,912 INFO [train.py:968] (1/2) Epoch 1, batch 41700, giga_loss[loss=0.3787, simple_loss=0.4233, pruned_loss=0.167, over 29034.00 frames. ], tot_loss[loss=0.4278, simple_loss=0.4548, pruned_loss=0.2004, over 5637099.49 frames. ], libri_tot_loss[loss=0.4331, simple_loss=0.4548, pruned_loss=0.2057, over 5686266.12 frames. ], giga_tot_loss[loss=0.4292, simple_loss=0.4561, pruned_loss=0.2011, over 5623135.99 frames. ], batch size: 128, lr: 1.73e-02, grad_scale: 4.0
+2023-02-28 22:01:22,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-02-28 22:02:01,557 INFO [train.py:968] (1/2) Epoch 1, batch 41750, giga_loss[loss=0.3792, simple_loss=0.4259, pruned_loss=0.1663, over 29016.00 frames. ], tot_loss[loss=0.4224, simple_loss=0.4509, pruned_loss=0.1969, over 5637879.61 frames. ], libri_tot_loss[loss=0.4321, simple_loss=0.4541, pruned_loss=0.2051, over 5690957.41 frames. ], giga_tot_loss[loss=0.4242, simple_loss=0.4526, pruned_loss=0.1978, over 5621775.25 frames. ], batch size: 128, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:02:19,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41765.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:02:30,395 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41776.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:02:32,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41779.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:02:35,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-02-28 22:02:44,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1089, 2.3248, 2.8982, 1.3842], device='cuda:1'), covar=tensor([0.0860, 0.0693, 0.1103, 0.1845], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0508, 0.0834, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-02-28 22:02:53,801 INFO [optim.py:369] (1/2) 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 22:02:53,814 INFO [train.py:968] (1/2) Epoch 1, batch 41800, giga_loss[loss=0.4193, simple_loss=0.4524, pruned_loss=0.1931, over 28607.00 frames. ], tot_loss[loss=0.4189, simple_loss=0.4485, pruned_loss=0.1947, over 5640127.52 frames. ], libri_tot_loss[loss=0.4321, simple_loss=0.4539, pruned_loss=0.2052, over 5694296.53 frames. ], giga_tot_loss[loss=0.4201, simple_loss=0.4499, pruned_loss=0.1952, over 5623355.96 frames. ], batch size: 336, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:03:04,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41808.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:07,983 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41811.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:09,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41813.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:11,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41814.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:40,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41843.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:03:42,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3695, 1.7639, 1.4463, 1.4287], device='cuda:1'), covar=tensor([0.1213, 0.0500, 0.0576, 0.1531], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0267, 0.0264, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0019, 0.0016, 0.0027], device='cuda:1')
+2023-02-28 22:03:47,544 INFO [train.py:968] (1/2) Epoch 1, batch 41850, giga_loss[loss=0.3952, simple_loss=0.4361, pruned_loss=0.1771, over 28884.00 frames. ], tot_loss[loss=0.4176, simple_loss=0.4476, pruned_loss=0.1937, over 5634158.89 frames. ], libri_tot_loss[loss=0.4323, simple_loss=0.4541, pruned_loss=0.2053, over 5683758.59 frames. ], giga_tot_loss[loss=0.4181, simple_loss=0.4484, pruned_loss=0.1939, over 5629058.60 frames. ], batch size: 186, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:04:06,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:04:38,147 INFO [optim.py:369] (1/2) 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 22:04:38,159 INFO [train.py:968] (1/2) Epoch 1, batch 41900, giga_loss[loss=0.4262, simple_loss=0.4232, pruned_loss=0.2146, over 23437.00 frames. ], tot_loss[loss=0.4144, simple_loss=0.4457, pruned_loss=0.1915, over 5640947.81 frames. ], libri_tot_loss[loss=0.4326, simple_loss=0.4543, pruned_loss=0.2054, over 5686140.30 frames. ], giga_tot_loss[loss=0.4145, simple_loss=0.4462, pruned_loss=0.1914, over 5634572.48 frames. ], batch size: 705, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:04:49,473 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41908.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:04:51,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41911.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:05:23,562 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41940.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:05:34,808 INFO [train.py:968] (1/2) Epoch 1, batch 41950, giga_loss[loss=0.4321, simple_loss=0.4627, pruned_loss=0.2008, over 28842.00 frames. ], tot_loss[loss=0.4103, simple_loss=0.443, pruned_loss=0.1887, over 5640171.18 frames. ], libri_tot_loss[loss=0.4323, simple_loss=0.4542, pruned_loss=0.2052, over 5690563.30 frames. ], giga_tot_loss[loss=0.41, simple_loss=0.4432, pruned_loss=0.1884, over 5629693.89 frames. ], batch size: 199, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:05:39,548 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41956.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:05:41,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41959.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:06:11,490 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:06:21,357 INFO [train.py:968] (1/2) Epoch 1, batch 42000, giga_loss[loss=0.485, simple_loss=0.4979, pruned_loss=0.2361, over 27978.00 frames. ], tot_loss[loss=0.4087, simple_loss=0.4437, pruned_loss=0.1868, over 5652743.46 frames. ], libri_tot_loss[loss=0.4309, simple_loss=0.4532, pruned_loss=0.2043, over 5696067.43 frames. ], giga_tot_loss[loss=0.4087, simple_loss=0.4443, pruned_loss=0.1865, over 5637098.85 frames. ], batch size: 412, lr: 1.72e-02, grad_scale: 8.0
+2023-02-28 22:06:21,357 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 22:06:28,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7622, 1.6863, 3.4206, 2.9483], device='cuda:1'), covar=tensor([0.1767, 0.1455, 0.0418, 0.0445], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0464, 0.0604, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 22:06:30,178 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-02-28 22:06:40,916 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42013.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:06:45,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42016.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:06:46,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9879, 1.5876, 1.5453, 1.4301], device='cuda:1'), covar=tensor([0.0616, 0.1288, 0.1099, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0833, 0.0639, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 22:07:15,657 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42045.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:07:16,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42047.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:07:19,473 INFO [train.py:968] (1/2) Epoch 1, batch 42050, giga_loss[loss=0.4206, simple_loss=0.454, pruned_loss=0.1936, over 28248.00 frames. ], tot_loss[loss=0.4098, simple_loss=0.4456, pruned_loss=0.1869, over 5645885.27 frames. ], libri_tot_loss[loss=0.4304, simple_loss=0.4526, pruned_loss=0.204, over 5680183.37 frames. ], giga_tot_loss[loss=0.4096, simple_loss=0.4463, pruned_loss=0.1865, over 5646806.78 frames. ], batch size: 368, lr: 1.72e-02, grad_scale: 8.0
+2023-02-28 22:08:08,439 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 42100, giga_loss[loss=0.3987, simple_loss=0.4416, pruned_loss=0.1779, over 28533.00 frames. ], tot_loss[loss=0.4117, simple_loss=0.4472, pruned_loss=0.1881, over 5657793.40 frames. ], libri_tot_loss[loss=0.4299, simple_loss=0.4525, pruned_loss=0.2037, over 5682988.29 frames. ], giga_tot_loss[loss=0.4116, simple_loss=0.4477, pruned_loss=0.1877, over 5655374.17 frames. ], batch size: 85, lr: 1.72e-02, grad_scale: 8.0
+2023-02-28 22:08:54,209 INFO [train.py:968] (1/2) Epoch 1, batch 42150, giga_loss[loss=0.3957, simple_loss=0.4343, pruned_loss=0.1785, over 28823.00 frames. ], tot_loss[loss=0.4119, simple_loss=0.4468, pruned_loss=0.1884, over 5657758.39 frames. ], libri_tot_loss[loss=0.4297, simple_loss=0.4524, pruned_loss=0.2035, over 5686074.76 frames. ], giga_tot_loss[loss=0.4114, simple_loss=0.4472, pruned_loss=0.1879, over 5652678.37 frames. ], batch size: 99, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:09:42,986 INFO [train.py:968] (1/2) Epoch 1, batch 42200, giga_loss[loss=0.378, simple_loss=0.4223, pruned_loss=0.1669, over 28910.00 frames. ], tot_loss[loss=0.4107, simple_loss=0.4451, pruned_loss=0.1881, over 5671870.48 frames. ], libri_tot_loss[loss=0.4291, simple_loss=0.4521, pruned_loss=0.203, over 5690058.79 frames. ], giga_tot_loss[loss=0.4105, simple_loss=0.4455, pruned_loss=0.1878, over 5663671.22 frames. ], batch size: 199, lr: 1.72e-02, grad_scale: 4.0
+2023-02-28 22:09:43,657 INFO [optim.py:369] (1/2) 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 22:10:31,742 INFO [train.py:968] (1/2) Epoch 1, batch 42250, giga_loss[loss=0.3616, simple_loss=0.4111, pruned_loss=0.156, over 28831.00 frames. ], tot_loss[loss=0.4114, simple_loss=0.444, pruned_loss=0.1894, over 5670715.97 frames. ], libri_tot_loss[loss=0.4294, simple_loss=0.4523, pruned_loss=0.2033, over 5696801.81 frames. ], giga_tot_loss[loss=0.4103, simple_loss=0.4438, pruned_loss=0.1884, over 5657007.57 frames. ], batch size: 174, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:11:23,499 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 42300, giga_loss[loss=0.3628, simple_loss=0.4159, pruned_loss=0.1549, over 29079.00 frames. ], tot_loss[loss=0.4088, simple_loss=0.4423, pruned_loss=0.1876, over 5670028.95 frames. ], libri_tot_loss[loss=0.4294, simple_loss=0.4522, pruned_loss=0.2033, over 5698595.72 frames. ], giga_tot_loss[loss=0.4077, simple_loss=0.4422, pruned_loss=0.1866, over 5657236.11 frames. ], batch size: 128, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:11:27,008 INFO [optim.py:369] (1/2) 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 22:12:04,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42341.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 22:12:11,149 INFO [train.py:968] (1/2) Epoch 1, batch 42350, giga_loss[loss=0.3526, simple_loss=0.4229, pruned_loss=0.1412, over 28863.00 frames. ], tot_loss[loss=0.4048, simple_loss=0.4408, pruned_loss=0.1844, over 5662166.44 frames. ], libri_tot_loss[loss=0.4292, simple_loss=0.452, pruned_loss=0.2032, over 5682167.94 frames. ], giga_tot_loss[loss=0.4037, simple_loss=0.4406, pruned_loss=0.1834, over 5667309.12 frames. ], batch size: 112, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:13:03,523 INFO [train.py:968] (1/2) Epoch 1, batch 42400, giga_loss[loss=0.3524, simple_loss=0.4061, pruned_loss=0.1493, over 28583.00 frames. ], tot_loss[loss=0.4041, simple_loss=0.4403, pruned_loss=0.1839, over 5664535.38 frames. ], libri_tot_loss[loss=0.4289, simple_loss=0.4518, pruned_loss=0.2029, over 5684893.59 frames. ], giga_tot_loss[loss=0.403, simple_loss=0.4402, pruned_loss=0.183, over 5666006.55 frames. ], batch size: 85, lr: 1.71e-02, grad_scale: 8.0
+2023-02-28 22:13:04,166 INFO [optim.py:369] (1/2) 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 22:13:10,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7421, 2.0696, 1.7155, 0.9074], device='cuda:1'), covar=tensor([0.0773, 0.0592, 0.0717, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.1075, 0.1076, 0.1085, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 22:13:25,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42422.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:13:52,119 INFO [train.py:968] (1/2) Epoch 1, batch 42450, libri_loss[loss=0.4666, simple_loss=0.4911, pruned_loss=0.2211, over 29663.00 frames. ], tot_loss[loss=0.4036, simple_loss=0.4397, pruned_loss=0.1838, over 5661650.07 frames. ], libri_tot_loss[loss=0.4287, simple_loss=0.4518, pruned_loss=0.2028, over 5680909.24 frames. ], giga_tot_loss[loss=0.4023, simple_loss=0.4394, pruned_loss=0.1826, over 5665279.11 frames. ], batch size: 88, lr: 1.71e-02, grad_scale: 8.0
+2023-02-28 22:14:24,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3892, 1.3564, 1.3119, 0.8883], device='cuda:1'), covar=tensor([0.0520, 0.0356, 0.0295, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0912, 0.0630, 0.0715, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 22:14:38,788 INFO [train.py:968] (1/2) Epoch 1, batch 42500, giga_loss[loss=0.3448, simple_loss=0.3991, pruned_loss=0.1453, over 28739.00 frames. ], tot_loss[loss=0.4004, simple_loss=0.437, pruned_loss=0.1819, over 5665655.15 frames. ], libri_tot_loss[loss=0.4288, simple_loss=0.4518, pruned_loss=0.2029, over 5674151.93 frames. ], giga_tot_loss[loss=0.399, simple_loss=0.4365, pruned_loss=0.1807, over 5673933.70 frames. ], batch size: 119, lr: 1.71e-02, grad_scale: 8.0
+2023-02-28 22:14:39,488 INFO [optim.py:369] (1/2) 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 22:15:29,243 INFO [train.py:968] (1/2) Epoch 1, batch 42550, libri_loss[loss=0.4204, simple_loss=0.4541, pruned_loss=0.1934, over 29536.00 frames. ], tot_loss[loss=0.4023, simple_loss=0.4375, pruned_loss=0.1836, over 5666046.21 frames. ], libri_tot_loss[loss=0.4283, simple_loss=0.4515, pruned_loss=0.2025, over 5679822.02 frames. ], giga_tot_loss[loss=0.401, simple_loss=0.437, pruned_loss=0.1825, over 5667089.18 frames. ], batch size: 81, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:15:41,554 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42568.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:15:59,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3698, 1.4619, 1.1473, 1.2069], device='cuda:1'), covar=tensor([0.1152, 0.0481, 0.0603, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0258, 0.0256, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0019, 0.0016, 0.0027], device='cuda:1')
+2023-02-28 22:16:12,195 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42597.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:16:15,514 INFO [train.py:968] (1/2) Epoch 1, batch 42600, giga_loss[loss=0.4387, simple_loss=0.4481, pruned_loss=0.2146, over 28575.00 frames. ], tot_loss[loss=0.4038, simple_loss=0.4381, pruned_loss=0.1848, over 5670018.38 frames. ], libri_tot_loss[loss=0.429, simple_loss=0.4522, pruned_loss=0.2029, over 5673597.85 frames. ], giga_tot_loss[loss=0.4015, simple_loss=0.4367, pruned_loss=0.1831, over 5676628.23 frames. ], batch size: 85, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:16:18,259 INFO [optim.py:369] (1/2) 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 22:17:04,760 INFO [train.py:968] (1/2) Epoch 1, batch 42650, libri_loss[loss=0.4397, simple_loss=0.4717, pruned_loss=0.2039, over 29299.00 frames. ], tot_loss[loss=0.401, simple_loss=0.4353, pruned_loss=0.1833, over 5672664.44 frames. ], libri_tot_loss[loss=0.4278, simple_loss=0.4513, pruned_loss=0.2021, over 5678331.54 frames. ], giga_tot_loss[loss=0.3994, simple_loss=0.4346, pruned_loss=0.1821, over 5673437.81 frames. ], batch size: 94, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:17:29,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42674.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:17:57,315 INFO [train.py:968] (1/2) Epoch 1, batch 42700, libri_loss[loss=0.415, simple_loss=0.4501, pruned_loss=0.19, over 29650.00 frames. ], tot_loss[loss=0.401, simple_loss=0.4348, pruned_loss=0.1835, over 5660797.79 frames. ], libri_tot_loss[loss=0.4279, simple_loss=0.4515, pruned_loss=0.2021, over 5682915.24 frames. ], giga_tot_loss[loss=0.3991, simple_loss=0.4338, pruned_loss=0.1823, over 5657026.18 frames. ], batch size: 88, lr: 1.71e-02, grad_scale: 4.0
+2023-02-28 22:17:58,642 INFO [optim.py:369] (1/2) 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 22:18:13,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42716.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 22:18:46,841 INFO [train.py:968] (1/2) Epoch 1, batch 42750, giga_loss[loss=0.3511, simple_loss=0.4091, pruned_loss=0.1466, over 28994.00 frames. ], tot_loss[loss=0.4021, simple_loss=0.4356, pruned_loss=0.1843, over 5654024.14 frames. ], libri_tot_loss[loss=0.428, simple_loss=0.4517, pruned_loss=0.2022, over 5680404.35 frames. ], giga_tot_loss[loss=0.3999, simple_loss=0.4342, pruned_loss=0.1828, over 5652922.59 frames. ], batch size: 136, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:18:48,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8049, 1.5716, 1.5542, 0.8795], device='cuda:1'), covar=tensor([0.0464, 0.0340, 0.0250, 0.0485], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0650, 0.0731, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 22:18:57,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4866, 1.4806, 1.1384, 1.1563], device='cuda:1'), covar=tensor([0.0688, 0.0725, 0.0946, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0567, 0.0574, 0.0516], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 22:19:07,092 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:19:24,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4460, 1.5721, 1.3397, 1.4417], device='cuda:1'), covar=tensor([0.1106, 0.0441, 0.0589, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0256, 0.0260, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0018, 0.0017, 0.0027], device='cuda:1')
+2023-02-28 22:19:32,797 INFO [train.py:968] (1/2) Epoch 1, batch 42800, libri_loss[loss=0.3989, simple_loss=0.4356, pruned_loss=0.1811, over 29525.00 frames. ], tot_loss[loss=0.4016, simple_loss=0.4359, pruned_loss=0.1836, over 5668037.81 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4512, pruned_loss=0.2016, over 5687324.59 frames. ], giga_tot_loss[loss=0.3998, simple_loss=0.4348, pruned_loss=0.1824, over 5660004.55 frames. ], batch size: 81, lr: 1.70e-02, grad_scale: 8.0
+2023-02-28 22:19:35,159 INFO [optim.py:369] (1/2) 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 22:19:49,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42817.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:19:53,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:20:20,565 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 42850, giga_loss[loss=0.3981, simple_loss=0.4438, pruned_loss=0.1761, over 29089.00 frames. ], tot_loss[loss=0.4006, simple_loss=0.4361, pruned_loss=0.1826, over 5671700.10 frames. ], libri_tot_loss[loss=0.427, simple_loss=0.451, pruned_loss=0.2015, over 5687031.57 frames. ], giga_tot_loss[loss=0.3991, simple_loss=0.4352, pruned_loss=0.1815, over 5665149.82 frames. ], batch size: 128, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:20:32,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-02-28 22:20:33,080 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42859.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 22:20:35,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42862.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 22:20:41,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-02-28 22:20:54,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-02-28 22:20:55,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4423, 1.4534, 1.0203, 1.0842], device='cuda:1'), covar=tensor([0.0711, 0.0696, 0.1168, 0.0975], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0553, 0.0564, 0.0499], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 22:21:00,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42891.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 22:21:09,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=42899.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:21:10,246 INFO [train.py:968] (1/2) Epoch 1, batch 42900, libri_loss[loss=0.4009, simple_loss=0.4275, pruned_loss=0.1872, over 20166.00 frames. ], tot_loss[loss=0.4002, simple_loss=0.4364, pruned_loss=0.182, over 5667911.02 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4514, pruned_loss=0.2017, over 5681810.59 frames. ], giga_tot_loss[loss=0.398, simple_loss=0.4349, pruned_loss=0.1805, over 5667754.06 frames. ], batch size: 187, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:21:12,584 INFO [optim.py:369] (1/2) 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 22:21:52,108 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-02-28 22:22:04,289 INFO [train.py:968] (1/2) Epoch 1, batch 42950, libri_loss[loss=0.3568, simple_loss=0.3905, pruned_loss=0.1615, over 29665.00 frames. ], tot_loss[loss=0.4019, simple_loss=0.4373, pruned_loss=0.1832, over 5672952.82 frames. ], libri_tot_loss[loss=0.427, simple_loss=0.4511, pruned_loss=0.2015, over 5677470.81 frames. ], giga_tot_loss[loss=0.3998, simple_loss=0.4361, pruned_loss=0.1818, over 5675639.69 frames. ], batch size: 69, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:22:44,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4119, 1.3515, 1.0250, 1.0705], device='cuda:1'), covar=tensor([0.0632, 0.0678, 0.1000, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0567, 0.0575, 0.0508], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 22:22:51,408 INFO [train.py:968] (1/2) Epoch 1, batch 43000, libri_loss[loss=0.4278, simple_loss=0.4604, pruned_loss=0.1975, over 29523.00 frames. ], tot_loss[loss=0.4054, simple_loss=0.4393, pruned_loss=0.1858, over 5680791.23 frames. ], libri_tot_loss[loss=0.4257, simple_loss=0.4501, pruned_loss=0.2006, over 5684716.59 frames. ], giga_tot_loss[loss=0.4042, simple_loss=0.4388, pruned_loss=0.1849, over 5676290.71 frames. ], batch size: 83, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:22:55,397 INFO [optim.py:369] (1/2) 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 22:23:09,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3008, 1.3397, 1.1591, 1.4141], device='cuda:1'), covar=tensor([0.1815, 0.1757, 0.1478, 0.1861], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0783, 0.0853, 0.0934], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 22:23:22,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8373, 2.9640, 3.5482, 1.6385], device='cuda:1'), covar=tensor([0.0627, 0.0581, 0.1057, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0530, 0.0872, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:1')
+2023-02-28 22:23:37,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43042.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:23:46,448 INFO [train.py:968] (1/2) Epoch 1, batch 43050, giga_loss[loss=0.4423, simple_loss=0.4556, pruned_loss=0.2145, over 27981.00 frames. ], tot_loss[loss=0.4094, simple_loss=0.441, pruned_loss=0.1889, over 5678651.67 frames. ], libri_tot_loss[loss=0.4257, simple_loss=0.4502, pruned_loss=0.2006, over 5688546.12 frames. ], giga_tot_loss[loss=0.408, simple_loss=0.4403, pruned_loss=0.1879, over 5671592.83 frames. ], batch size: 412, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:24:02,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7849, 2.0871, 1.3220, 1.1501], device='cuda:1'), covar=tensor([0.0495, 0.0331, 0.0376, 0.0469], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0610, 0.0702, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 22:24:05,947 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-02-28 22:24:24,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6211, 1.7719, 1.4088, 1.4586], device='cuda:1'), covar=tensor([0.1500, 0.2093, 0.1512, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0839, 0.0699, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 22:24:42,042 INFO [train.py:968] (1/2) Epoch 1, batch 43100, giga_loss[loss=0.4114, simple_loss=0.4385, pruned_loss=0.1922, over 28226.00 frames. ], tot_loss[loss=0.4132, simple_loss=0.4431, pruned_loss=0.1917, over 5674129.28 frames. ], libri_tot_loss[loss=0.4257, simple_loss=0.4502, pruned_loss=0.2006, over 5688627.54 frames. ], giga_tot_loss[loss=0.4119, simple_loss=0.4425, pruned_loss=0.1907, over 5668335.17 frames. ], batch size: 77, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:24:46,160 INFO [optim.py:369] (1/2) 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 22:25:26,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43147.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:25:28,683 INFO [train.py:968] (1/2) Epoch 1, batch 43150, giga_loss[loss=0.3619, simple_loss=0.4178, pruned_loss=0.153, over 28838.00 frames. ], tot_loss[loss=0.4147, simple_loss=0.444, pruned_loss=0.1927, over 5662412.17 frames. ], libri_tot_loss[loss=0.4258, simple_loss=0.4502, pruned_loss=0.2007, over 5684549.08 frames. ], giga_tot_loss[loss=0.4131, simple_loss=0.4432, pruned_loss=0.1915, over 5660946.00 frames. ], batch size: 174, lr: 1.70e-02, grad_scale: 4.0
+2023-02-28 22:25:59,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3785, 1.7257, 1.4016, 0.4558], device='cuda:1'), covar=tensor([0.1185, 0.0885, 0.0829, 0.1446], device='cuda:1'), in_proj_covar=tensor([0.1095, 0.1075, 0.1082, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 22:26:18,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6913, 1.7408, 1.5630, 0.9978], device='cuda:1'), covar=tensor([0.0530, 0.0358, 0.0316, 0.0492], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0616, 0.0703, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 22:26:18,486 INFO [train.py:968] (1/2) Epoch 1, batch 43200, giga_loss[loss=0.3463, simple_loss=0.406, pruned_loss=0.1433, over 28508.00 frames. ], tot_loss[loss=0.4132, simple_loss=0.4429, pruned_loss=0.1917, over 5663715.44 frames. ], libri_tot_loss[loss=0.426, simple_loss=0.4504, pruned_loss=0.2008, over 5687677.28 frames. ], giga_tot_loss[loss=0.4116, simple_loss=0.442, pruned_loss=0.1906, over 5659167.63 frames. ], batch size: 71, lr: 1.70e-02, grad_scale: 8.0
+2023-02-28 22:26:21,582 INFO [optim.py:369] (1/2) 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 22:26:29,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2499, 1.2558, 1.2095, 1.1748], device='cuda:1'), covar=tensor([0.1476, 0.1567, 0.1249, 0.1519], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0804, 0.0867, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 22:26:51,061 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-02-28 22:27:06,256 INFO [train.py:968] (1/2) Epoch 1, batch 43250, giga_loss[loss=0.4607, simple_loss=0.4671, pruned_loss=0.2271, over 24042.00 frames. ], tot_loss[loss=0.4119, simple_loss=0.4429, pruned_loss=0.1904, over 5655499.20 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4513, pruned_loss=0.2016, over 5682290.79 frames. ], giga_tot_loss[loss=0.409, simple_loss=0.4411, pruned_loss=0.1885, over 5656703.85 frames. ], batch size: 705, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:27:07,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43252.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:26,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43272.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:27,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43274.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:44,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43290.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:47,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43293.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:27:53,279 INFO [train.py:968] (1/2) Epoch 1, batch 43300, libri_loss[loss=0.413, simple_loss=0.4393, pruned_loss=0.1934, over 29572.00 frames. ], tot_loss[loss=0.4054, simple_loss=0.4386, pruned_loss=0.1861, over 5656372.59 frames. ], libri_tot_loss[loss=0.4279, simple_loss=0.4519, pruned_loss=0.202, over 5682967.08 frames. ], giga_tot_loss[loss=0.4022, simple_loss=0.4364, pruned_loss=0.184, over 5656097.70 frames. ], batch size: 75, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:27:58,261 INFO [optim.py:369] (1/2) 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 22:28:12,071 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43322.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:28:37,828 INFO [train.py:968] (1/2) Epoch 1, batch 43350, libri_loss[loss=0.3488, simple_loss=0.3899, pruned_loss=0.1538, over 29652.00 frames. ], tot_loss[loss=0.4045, simple_loss=0.4375, pruned_loss=0.1858, over 5664179.45 frames. ], libri_tot_loss[loss=0.4274, simple_loss=0.4516, pruned_loss=0.2016, over 5683738.32 frames. ], giga_tot_loss[loss=0.4012, simple_loss=0.4354, pruned_loss=0.1835, over 5662078.01 frames. ], batch size: 73, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:29:14,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7860, 1.6719, 3.4830, 3.0005], device='cuda:1'), covar=tensor([0.1302, 0.1191, 0.0332, 0.0344], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0472, 0.0619, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 22:29:21,824 INFO [train.py:968] (1/2) Epoch 1, batch 43400, giga_loss[loss=0.3643, simple_loss=0.4054, pruned_loss=0.1616, over 29044.00 frames. ], tot_loss[loss=0.4042, simple_loss=0.4369, pruned_loss=0.1857, over 5675020.18 frames. ], libri_tot_loss[loss=0.4276, simple_loss=0.4519, pruned_loss=0.2016, over 5690790.61 frames. ], giga_tot_loss[loss=0.4005, simple_loss=0.4343, pruned_loss=0.1833, over 5666257.75 frames. ], batch size: 128, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:29:26,263 INFO [optim.py:369] (1/2) 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 22:29:29,977 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0853, 1.7830, 1.6456, 1.6629], device='cuda:1'), covar=tensor([0.0743, 0.1464, 0.1190, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0510, 0.0840, 0.0644, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 22:29:40,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43417.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:29:40,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43417.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:29:42,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43420.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:30:09,041 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43449.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:30:09,370 INFO [train.py:968] (1/2) Epoch 1, batch 43450, giga_loss[loss=0.4551, simple_loss=0.4692, pruned_loss=0.2205, over 27658.00 frames. ], tot_loss[loss=0.4043, simple_loss=0.4366, pruned_loss=0.186, over 5678255.67 frames. ], libri_tot_loss[loss=0.4263, simple_loss=0.451, pruned_loss=0.2008, over 5694509.32 frames. ], giga_tot_loss[loss=0.4017, simple_loss=0.4347, pruned_loss=0.1843, over 5667168.34 frames. ], batch size: 472, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:30:59,439 INFO [train.py:968] (1/2) Epoch 1, batch 43500, giga_loss[loss=0.4104, simple_loss=0.4543, pruned_loss=0.1832, over 29000.00 frames. ], tot_loss[loss=0.409, simple_loss=0.4407, pruned_loss=0.1887, over 5669079.13 frames. ], libri_tot_loss[loss=0.4258, simple_loss=0.4506, pruned_loss=0.2006, over 5693450.42 frames. ], giga_tot_loss[loss=0.4071, simple_loss=0.4394, pruned_loss=0.1873, over 5661041.24 frames. ], batch size: 213, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:31:02,188 INFO [optim.py:369] (1/2) 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 22:31:16,564 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-02-28 22:31:48,131 INFO [train.py:968] (1/2) Epoch 1, batch 43550, giga_loss[loss=0.4056, simple_loss=0.4653, pruned_loss=0.1729, over 29001.00 frames. ], tot_loss[loss=0.4082, simple_loss=0.4431, pruned_loss=0.1866, over 5664800.87 frames. ], libri_tot_loss[loss=0.4266, simple_loss=0.4511, pruned_loss=0.2011, over 5687072.66 frames. ], giga_tot_loss[loss=0.4054, simple_loss=0.4414, pruned_loss=0.1846, over 5664247.18 frames. ], batch size: 128, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:31:57,278 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43560.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:32:01,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:32:31,773 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43592.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:32:38,158 INFO [train.py:968] (1/2) Epoch 1, batch 43600, giga_loss[loss=0.4676, simple_loss=0.4595, pruned_loss=0.2379, over 23730.00 frames. ], tot_loss[loss=0.4075, simple_loss=0.4435, pruned_loss=0.1857, over 5654949.52 frames. ], libri_tot_loss[loss=0.4263, simple_loss=0.4508, pruned_loss=0.2009, over 5679168.98 frames. ], giga_tot_loss[loss=0.4049, simple_loss=0.4422, pruned_loss=0.1838, over 5660625.53 frames. ], batch size: 705, lr: 1.69e-02, grad_scale: 8.0
+2023-02-28 22:32:43,462 INFO [optim.py:369] (1/2) 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 22:32:45,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3832, 1.6727, 1.3726, 0.4172], device='cuda:1'), covar=tensor([0.0808, 0.0750, 0.0838, 0.1320], device='cuda:1'), in_proj_covar=tensor([0.1078, 0.1085, 0.1089, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 22:33:05,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43627.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:33:24,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43647.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:33:27,397 INFO [train.py:968] (1/2) Epoch 1, batch 43650, libri_loss[loss=0.3674, simple_loss=0.3955, pruned_loss=0.1697, over 29398.00 frames. ], tot_loss[loss=0.41, simple_loss=0.4453, pruned_loss=0.1873, over 5661523.13 frames. ], libri_tot_loss[loss=0.4256, simple_loss=0.4501, pruned_loss=0.2005, over 5684401.27 frames. ], giga_tot_loss[loss=0.4082, simple_loss=0.4447, pruned_loss=0.1858, over 5660808.29 frames. ], batch size: 67, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:33:49,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2784, 1.3798, 1.2157, 1.7035], device='cuda:1'), covar=tensor([0.1879, 0.1842, 0.1454, 0.2063], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0805, 0.0866, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 22:33:49,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9310, 2.6776, 1.9632, 1.7906], device='cuda:1'), covar=tensor([0.1392, 0.1335, 0.1087, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0836, 0.0701, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 22:34:16,775 INFO [train.py:968] (1/2) Epoch 1, batch 43700, giga_loss[loss=0.4281, simple_loss=0.4589, pruned_loss=0.1987, over 28615.00 frames. ], tot_loss[loss=0.4108, simple_loss=0.4457, pruned_loss=0.1879, over 5668231.12 frames. ], libri_tot_loss[loss=0.4253, simple_loss=0.4501, pruned_loss=0.2003, over 5689751.95 frames. ], giga_tot_loss[loss=0.4092, simple_loss=0.4452, pruned_loss=0.1866, over 5662245.56 frames. ], batch size: 262, lr: 1.69e-02, grad_scale: 4.0
+2023-02-28 22:34:21,815 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 43750, giga_loss[loss=0.3436, simple_loss=0.3953, pruned_loss=0.1459, over 28905.00 frames. ], tot_loss[loss=0.4124, simple_loss=0.4458, pruned_loss=0.1895, over 5674726.26 frames. ], libri_tot_loss[loss=0.4243, simple_loss=0.4494, pruned_loss=0.1996, over 5694916.14 frames. ], giga_tot_loss[loss=0.4115, simple_loss=0.4458, pruned_loss=0.1886, over 5664487.84 frames. ], batch size: 112, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:35:21,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43770.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:26,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43773.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:43,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43790.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:45,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43793.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:35:52,827 INFO [train.py:968] (1/2) Epoch 1, batch 43800, giga_loss[loss=0.3406, simple_loss=0.3874, pruned_loss=0.1469, over 28949.00 frames. ], tot_loss[loss=0.4101, simple_loss=0.4435, pruned_loss=0.1884, over 5663143.89 frames. ], libri_tot_loss[loss=0.4241, simple_loss=0.4492, pruned_loss=0.1995, over 5689411.86 frames. ], giga_tot_loss[loss=0.4095, simple_loss=0.4436, pruned_loss=0.1877, over 5659378.88 frames. ], batch size: 106, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:35:55,370 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:36:15,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2820, 1.8356, 1.4407, 0.5327], device='cuda:1'), covar=tensor([0.1266, 0.0689, 0.0895, 0.1501], device='cuda:1'), in_proj_covar=tensor([0.1098, 0.1085, 0.1101, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 22:36:16,324 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43822.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:36:42,959 INFO [train.py:968] (1/2) Epoch 1, batch 43850, giga_loss[loss=0.4753, simple_loss=0.4841, pruned_loss=0.2332, over 28979.00 frames. ], tot_loss[loss=0.4087, simple_loss=0.4419, pruned_loss=0.1878, over 5665411.36 frames. ], libri_tot_loss[loss=0.4238, simple_loss=0.4492, pruned_loss=0.1992, over 5687824.46 frames. ], giga_tot_loss[loss=0.4081, simple_loss=0.4419, pruned_loss=0.1872, over 5662949.29 frames. ], batch size: 136, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:37:31,567 INFO [train.py:968] (1/2) Epoch 1, batch 43900, giga_loss[loss=0.3769, simple_loss=0.4219, pruned_loss=0.1659, over 28741.00 frames. ], tot_loss[loss=0.4065, simple_loss=0.4397, pruned_loss=0.1866, over 5668422.00 frames. ], libri_tot_loss[loss=0.4228, simple_loss=0.4485, pruned_loss=0.1985, over 5691665.58 frames. ], giga_tot_loss[loss=0.4066, simple_loss=0.4401, pruned_loss=0.1866, over 5662389.24 frames. ], batch size: 60, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:37:37,597 INFO [optim.py:369] (1/2) 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 22:37:43,354 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=43907.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:38:24,118 INFO [train.py:968] (1/2) Epoch 1, batch 43950, giga_loss[loss=0.4184, simple_loss=0.452, pruned_loss=0.1924, over 28601.00 frames. ], tot_loss[loss=0.409, simple_loss=0.4414, pruned_loss=0.1883, over 5681703.70 frames. ], libri_tot_loss[loss=0.4226, simple_loss=0.4484, pruned_loss=0.1983, over 5697595.27 frames. ], giga_tot_loss[loss=0.4088, simple_loss=0.4415, pruned_loss=0.1881, over 5670968.05 frames. ], batch size: 336, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:39:17,517 INFO [train.py:968] (1/2) Epoch 1, batch 44000, giga_loss[loss=0.391, simple_loss=0.4315, pruned_loss=0.1753, over 28985.00 frames. ], tot_loss[loss=0.4118, simple_loss=0.4427, pruned_loss=0.1905, over 5672510.17 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4486, pruned_loss=0.1984, over 5697899.07 frames. ], giga_tot_loss[loss=0.4114, simple_loss=0.4426, pruned_loss=0.1901, over 5663488.78 frames. ], batch size: 213, lr: 1.68e-02, grad_scale: 8.0
+2023-02-28 22:39:22,157 INFO [optim.py:369] (1/2) 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 22:40:06,148 INFO [train.py:968] (1/2) Epoch 1, batch 44050, giga_loss[loss=0.4092, simple_loss=0.4404, pruned_loss=0.189, over 28873.00 frames. ], tot_loss[loss=0.4075, simple_loss=0.4391, pruned_loss=0.1879, over 5677742.52 frames. ], libri_tot_loss[loss=0.4228, simple_loss=0.4487, pruned_loss=0.1985, over 5700884.66 frames. ], giga_tot_loss[loss=0.4069, simple_loss=0.4389, pruned_loss=0.1875, over 5667633.32 frames. ], batch size: 227, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:40:24,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9792, 2.8913, 1.7189, 1.8193], device='cuda:1'), covar=tensor([0.0834, 0.0554, 0.0910, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0480, 0.0364, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:1')
+2023-02-28 22:40:53,722 INFO [train.py:968] (1/2) Epoch 1, batch 44100, giga_loss[loss=0.4009, simple_loss=0.439, pruned_loss=0.1814, over 28676.00 frames. ], tot_loss[loss=0.4056, simple_loss=0.4382, pruned_loss=0.1865, over 5661092.22 frames. ], libri_tot_loss[loss=0.4233, simple_loss=0.4491, pruned_loss=0.1987, over 5683049.03 frames. ], giga_tot_loss[loss=0.4044, simple_loss=0.4374, pruned_loss=0.1857, over 5667750.27 frames. ], batch size: 262, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:41:00,001 INFO [optim.py:369] (1/2) 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 22:41:50,176 INFO [train.py:968] (1/2) Epoch 1, batch 44150, giga_loss[loss=0.4, simple_loss=0.4423, pruned_loss=0.1789, over 28717.00 frames. ], tot_loss[loss=0.4064, simple_loss=0.4395, pruned_loss=0.1866, over 5660967.02 frames. ], libri_tot_loss[loss=0.4228, simple_loss=0.4488, pruned_loss=0.1984, over 5682694.78 frames. ], giga_tot_loss[loss=0.4058, simple_loss=0.4392, pruned_loss=0.1862, over 5666378.98 frames. ], batch size: 119, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:42:18,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44180.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:42:40,118 INFO [train.py:968] (1/2) Epoch 1, batch 44200, libri_loss[loss=0.431, simple_loss=0.452, pruned_loss=0.205, over 27826.00 frames. ], tot_loss[loss=0.4081, simple_loss=0.4406, pruned_loss=0.1878, over 5669572.57 frames. ], libri_tot_loss[loss=0.4226, simple_loss=0.4486, pruned_loss=0.1983, over 5685502.62 frames. ], giga_tot_loss[loss=0.4076, simple_loss=0.4404, pruned_loss=0.1874, over 5671227.07 frames. ], batch size: 115, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:42:45,164 INFO [optim.py:369] (1/2) 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 22:42:49,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-02-28 22:43:06,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6230, 1.9981, 1.7511, 1.6565], device='cuda:1'), covar=tensor([0.1121, 0.1392, 0.0990, 0.0641], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0846, 0.0710, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 22:43:30,460 INFO [train.py:968] (1/2) Epoch 1, batch 44250, giga_loss[loss=0.4431, simple_loss=0.4874, pruned_loss=0.1994, over 29039.00 frames. ], tot_loss[loss=0.4097, simple_loss=0.4419, pruned_loss=0.1888, over 5667833.76 frames. ], libri_tot_loss[loss=0.4224, simple_loss=0.4484, pruned_loss=0.1982, over 5690097.27 frames. ], giga_tot_loss[loss=0.4092, simple_loss=0.4417, pruned_loss=0.1883, over 5664368.99 frames. ], batch size: 106, lr: 1.68e-02, grad_scale: 4.0
+2023-02-28 22:44:00,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44282.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:44:16,175 INFO [train.py:968] (1/2) Epoch 1, batch 44300, giga_loss[loss=0.3166, simple_loss=0.3943, pruned_loss=0.1194, over 28519.00 frames. ], tot_loss[loss=0.4086, simple_loss=0.4432, pruned_loss=0.187, over 5665779.97 frames. ], libri_tot_loss[loss=0.4233, simple_loss=0.4489, pruned_loss=0.1988, over 5684881.46 frames. ], giga_tot_loss[loss=0.407, simple_loss=0.4424, pruned_loss=0.1858, over 5666237.39 frames. ], batch size: 85, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:44:18,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44302.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:44:23,351 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 1, batch 44350, giga_loss[loss=0.4927, simple_loss=0.4739, pruned_loss=0.2557, over 23624.00 frames. ], tot_loss[loss=0.4064, simple_loss=0.4434, pruned_loss=0.1847, over 5675803.63 frames. ], libri_tot_loss[loss=0.4227, simple_loss=0.4484, pruned_loss=0.1985, over 5689319.69 frames. ], giga_tot_loss[loss=0.4053, simple_loss=0.4432, pruned_loss=0.1837, over 5671812.61 frames. ], batch size: 705, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:45:05,517 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44352.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:45:55,155 INFO [train.py:968] (1/2) Epoch 1, batch 44400, giga_loss[loss=0.3882, simple_loss=0.4338, pruned_loss=0.1713, over 29033.00 frames. ], tot_loss[loss=0.4087, simple_loss=0.4458, pruned_loss=0.1858, over 5687195.86 frames. ], libri_tot_loss[loss=0.4216, simple_loss=0.4476, pruned_loss=0.1978, over 5689206.14 frames. ], giga_tot_loss[loss=0.4084, simple_loss=0.4462, pruned_loss=0.1853, over 5683713.45 frames. ], batch size: 164, lr: 1.67e-02, grad_scale: 8.0
+2023-02-28 22:46:00,124 INFO [optim.py:369] (1/2) 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 22:46:11,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-02-28 22:46:18,422 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44425.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:46:20,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-02-28 22:46:21,979 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44428.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:46:41,347 INFO [train.py:968] (1/2) Epoch 1, batch 44450, giga_loss[loss=0.5649, simple_loss=0.5136, pruned_loss=0.3081, over 23340.00 frames. ], tot_loss[loss=0.4169, simple_loss=0.4502, pruned_loss=0.1918, over 5666450.83 frames. ], libri_tot_loss[loss=0.422, simple_loss=0.4478, pruned_loss=0.1981, over 5681388.32 frames. ], giga_tot_loss[loss=0.4159, simple_loss=0.4503, pruned_loss=0.1908, over 5670397.44 frames. ], batch size: 705, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:46:51,019 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44457.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:47:31,074 INFO [train.py:968] (1/2) Epoch 1, batch 44500, giga_loss[loss=0.4035, simple_loss=0.4466, pruned_loss=0.1802, over 28884.00 frames. ], tot_loss[loss=0.421, simple_loss=0.4521, pruned_loss=0.195, over 5644721.40 frames. ], libri_tot_loss[loss=0.4222, simple_loss=0.4479, pruned_loss=0.1983, over 5675537.26 frames. ], giga_tot_loss[loss=0.4198, simple_loss=0.4522, pruned_loss=0.1937, over 5653384.48 frames. ], batch size: 174, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:47:37,436 INFO [optim.py:369] (1/2) 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 22:48:05,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0712, 1.2103, 1.0067, 0.1080], device='cuda:1'), covar=tensor([0.0736, 0.0762, 0.1044, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.1068, 0.1095, 0.0933], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 22:48:16,775 INFO [train.py:968] (1/2) Epoch 1, batch 44550, giga_loss[loss=0.4049, simple_loss=0.4401, pruned_loss=0.1848, over 28819.00 frames. ], tot_loss[loss=0.4209, simple_loss=0.4516, pruned_loss=0.1951, over 5654820.19 frames. ], libri_tot_loss[loss=0.4208, simple_loss=0.4468, pruned_loss=0.1974, over 5680125.69 frames. ], giga_tot_loss[loss=0.4212, simple_loss=0.4528, pruned_loss=0.1948, over 5656756.81 frames. ], batch size: 119, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:48:22,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44555.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:49:02,861 INFO [train.py:968] (1/2) Epoch 1, batch 44600, giga_loss[loss=0.416, simple_loss=0.4558, pruned_loss=0.1881, over 28930.00 frames. ], tot_loss[loss=0.4186, simple_loss=0.4502, pruned_loss=0.1935, over 5660401.65 frames. ], libri_tot_loss[loss=0.4213, simple_loss=0.4471, pruned_loss=0.1977, over 5683303.04 frames. ], giga_tot_loss[loss=0.4184, simple_loss=0.451, pruned_loss=0.1929, over 5658526.94 frames. ], batch size: 213, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:49:09,340 INFO [optim.py:369] (1/2) 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 22:49:49,583 INFO [train.py:968] (1/2) Epoch 1, batch 44650, giga_loss[loss=0.4033, simple_loss=0.4467, pruned_loss=0.1799, over 28731.00 frames. ], tot_loss[loss=0.4136, simple_loss=0.4487, pruned_loss=0.1893, over 5668295.55 frames. ], libri_tot_loss[loss=0.4208, simple_loss=0.4468, pruned_loss=0.1974, over 5687110.13 frames. ], giga_tot_loss[loss=0.4138, simple_loss=0.4497, pruned_loss=0.189, over 5662669.48 frames. ], batch size: 242, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:50:03,214 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-02-28 22:50:18,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44677.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:50:36,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44698.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:50:37,826 INFO [train.py:968] (1/2) Epoch 1, batch 44700, giga_loss[loss=0.4102, simple_loss=0.4433, pruned_loss=0.1886, over 28624.00 frames. ], tot_loss[loss=0.4147, simple_loss=0.4502, pruned_loss=0.1896, over 5667470.23 frames. ], libri_tot_loss[loss=0.4203, simple_loss=0.4465, pruned_loss=0.1971, over 5690237.42 frames. ], giga_tot_loss[loss=0.4151, simple_loss=0.4513, pruned_loss=0.1895, over 5659939.19 frames. ], batch size: 92, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:50:39,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44701.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:50:40,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-02-28 22:50:44,371 INFO [optim.py:369] (1/2) 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 22:50:51,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-02-28 22:51:09,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44727.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:51:11,958 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 44750, giga_loss[loss=0.4619, simple_loss=0.4775, pruned_loss=0.2232, over 27820.00 frames. ], tot_loss[loss=0.4165, simple_loss=0.4512, pruned_loss=0.1909, over 5669912.93 frames. ], libri_tot_loss[loss=0.4206, simple_loss=0.4467, pruned_loss=0.1973, over 5690999.24 frames. ], giga_tot_loss[loss=0.4166, simple_loss=0.4519, pruned_loss=0.1907, over 5663330.15 frames. ], batch size: 412, lr: 1.67e-02, grad_scale: 4.0
+2023-02-28 22:52:17,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8003, 1.7572, 4.4164, 3.2934], device='cuda:1'), covar=tensor([0.1443, 0.1255, 0.0255, 0.0476], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0475, 0.0632, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 22:52:22,778 INFO [train.py:968] (1/2) Epoch 1, batch 44800, giga_loss[loss=0.511, simple_loss=0.4974, pruned_loss=0.2623, over 26584.00 frames. ], tot_loss[loss=0.4165, simple_loss=0.4506, pruned_loss=0.1912, over 5678662.50 frames. ], libri_tot_loss[loss=0.4214, simple_loss=0.4475, pruned_loss=0.1977, over 5691817.75 frames. ], giga_tot_loss[loss=0.4158, simple_loss=0.4506, pruned_loss=0.1905, over 5672569.74 frames. ], batch size: 555, lr: 1.67e-02, grad_scale: 8.0
+2023-02-28 22:52:28,490 INFO [optim.py:369] (1/2) 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 22:52:45,590 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44820.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:52:47,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44823.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:53:12,677 INFO [train.py:968] (1/2) Epoch 1, batch 44850, giga_loss[loss=0.3692, simple_loss=0.4083, pruned_loss=0.165, over 28933.00 frames. ], tot_loss[loss=0.4157, simple_loss=0.4485, pruned_loss=0.1914, over 5661099.98 frames. ], libri_tot_loss[loss=0.4216, simple_loss=0.4475, pruned_loss=0.1978, over 5695166.55 frames. ], giga_tot_loss[loss=0.4149, simple_loss=0.4485, pruned_loss=0.1907, over 5652560.41 frames. ], batch size: 136, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:53:14,415 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44852.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:53:36,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44870.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:53:38,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44873.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:54:01,558 INFO [train.py:968] (1/2) Epoch 1, batch 44900, giga_loss[loss=0.4049, simple_loss=0.4362, pruned_loss=0.1868, over 28700.00 frames. ], tot_loss[loss=0.4122, simple_loss=0.4453, pruned_loss=0.1896, over 5663526.27 frames. ], libri_tot_loss[loss=0.4211, simple_loss=0.4472, pruned_loss=0.1975, over 5696321.48 frames. ], giga_tot_loss[loss=0.4118, simple_loss=0.4455, pruned_loss=0.189, over 5654882.95 frames. ], batch size: 284, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:54:04,684 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44902.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:54:09,442 INFO [optim.py:369] (1/2) 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 22:54:21,309 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=44920.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 22:54:39,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-02-28 22:54:44,260 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 44950, giga_loss[loss=0.4259, simple_loss=0.4548, pruned_loss=0.1985, over 28312.00 frames. ], tot_loss[loss=0.4085, simple_loss=0.442, pruned_loss=0.1876, over 5665074.30 frames. ], libri_tot_loss[loss=0.4208, simple_loss=0.447, pruned_loss=0.1973, over 5701399.51 frames. ], giga_tot_loss[loss=0.4082, simple_loss=0.4422, pruned_loss=0.1871, over 5652762.84 frames. ], batch size: 368, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:55:36,569 INFO [train.py:968] (1/2) Epoch 1, batch 45000, giga_loss[loss=0.4778, simple_loss=0.4868, pruned_loss=0.2344, over 28696.00 frames. ], tot_loss[loss=0.4077, simple_loss=0.4407, pruned_loss=0.1873, over 5660995.70 frames. ], libri_tot_loss[loss=0.4215, simple_loss=0.4476, pruned_loss=0.1977, over 5695850.75 frames. ], giga_tot_loss[loss=0.4065, simple_loss=0.4403, pruned_loss=0.1863, over 5655806.28 frames. ], batch size: 242, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:55:36,569 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 22:55:44,895 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-02-28 22:55:51,059 INFO [optim.py:369] (1/2) 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 22:55:51,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4447, 1.9182, 1.5020, 0.5542], device='cuda:1'), covar=tensor([0.1108, 0.0713, 0.1108, 0.1429], device='cuda:1'), in_proj_covar=tensor([0.1110, 0.1083, 0.1111, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 22:56:08,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3682, 1.2118, 1.2063, 1.5240], device='cuda:1'), covar=tensor([0.1747, 0.1715, 0.1408, 0.2152], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0785, 0.0861, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 22:56:27,532 INFO [train.py:968] (1/2) Epoch 1, batch 45050, giga_loss[loss=0.418, simple_loss=0.4513, pruned_loss=0.1923, over 28753.00 frames. ], tot_loss[loss=0.4049, simple_loss=0.439, pruned_loss=0.1854, over 5669828.40 frames. ], libri_tot_loss[loss=0.4212, simple_loss=0.4475, pruned_loss=0.1975, over 5702030.85 frames. ], giga_tot_loss[loss=0.4036, simple_loss=0.4384, pruned_loss=0.1844, over 5658805.75 frames. ], batch size: 284, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:57:13,904 INFO [train.py:968] (1/2) Epoch 1, batch 45100, giga_loss[loss=0.3881, simple_loss=0.4289, pruned_loss=0.1736, over 28862.00 frames. ], tot_loss[loss=0.3994, simple_loss=0.4357, pruned_loss=0.1816, over 5667150.90 frames. ], libri_tot_loss[loss=0.4206, simple_loss=0.447, pruned_loss=0.1971, over 5706977.07 frames. ], giga_tot_loss[loss=0.3985, simple_loss=0.4354, pruned_loss=0.1808, over 5652897.51 frames. ], batch size: 199, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:57:20,539 INFO [optim.py:369] (1/2) 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 22:57:58,354 INFO [train.py:968] (1/2) Epoch 1, batch 45150, giga_loss[loss=0.4566, simple_loss=0.4701, pruned_loss=0.2215, over 26672.00 frames. ], tot_loss[loss=0.3964, simple_loss=0.4339, pruned_loss=0.1795, over 5678826.95 frames. ], libri_tot_loss[loss=0.4205, simple_loss=0.4468, pruned_loss=0.1971, over 5710627.73 frames. ], giga_tot_loss[loss=0.3953, simple_loss=0.4336, pruned_loss=0.1785, over 5663572.30 frames. ], batch size: 555, lr: 1.66e-02, grad_scale: 2.0
+2023-02-28 22:58:50,718 INFO [train.py:968] (1/2) Epoch 1, batch 45200, giga_loss[loss=0.3988, simple_loss=0.4299, pruned_loss=0.1839, over 28842.00 frames. ], tot_loss[loss=0.3957, simple_loss=0.4332, pruned_loss=0.1791, over 5669024.36 frames. ], libri_tot_loss[loss=0.4202, simple_loss=0.4466, pruned_loss=0.1969, over 5714319.00 frames. ], giga_tot_loss[loss=0.3946, simple_loss=0.4329, pruned_loss=0.1782, over 5652531.96 frames. ], batch size: 186, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 22:59:00,258 INFO [optim.py:369] (1/2) 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 22:59:39,411 INFO [train.py:968] (1/2) Epoch 1, batch 45250, giga_loss[loss=0.3976, simple_loss=0.4294, pruned_loss=0.1829, over 28340.00 frames. ], tot_loss[loss=0.3949, simple_loss=0.4312, pruned_loss=0.1793, over 5658731.03 frames. ], libri_tot_loss[loss=0.4206, simple_loss=0.447, pruned_loss=0.1971, over 5718950.03 frames. ], giga_tot_loss[loss=0.3931, simple_loss=0.4302, pruned_loss=0.1779, over 5639986.48 frames. ], batch size: 368, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 23:00:21,910 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 45300, giga_loss[loss=0.3726, simple_loss=0.4112, pruned_loss=0.167, over 28765.00 frames. ], tot_loss[loss=0.393, simple_loss=0.4296, pruned_loss=0.1782, over 5663041.87 frames. ], libri_tot_loss[loss=0.4207, simple_loss=0.4472, pruned_loss=0.1971, over 5723474.04 frames. ], giga_tot_loss[loss=0.3908, simple_loss=0.4283, pruned_loss=0.1767, over 5642920.79 frames. ], batch size: 92, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 23:00:35,055 INFO [optim.py:369] (1/2) 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 23:00:43,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45318.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:01:01,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2780, 1.5421, 1.2613, 1.3937], device='cuda:1'), covar=tensor([0.1016, 0.0619, 0.0545, 0.1231], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0253, 0.0252, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0019, 0.0017, 0.0028], device='cuda:1')
+2023-02-28 23:01:11,184 INFO [train.py:968] (1/2) Epoch 1, batch 45350, giga_loss[loss=0.3931, simple_loss=0.4085, pruned_loss=0.1888, over 23541.00 frames. ], tot_loss[loss=0.3948, simple_loss=0.4317, pruned_loss=0.179, over 5652948.18 frames. ], libri_tot_loss[loss=0.4208, simple_loss=0.4474, pruned_loss=0.1971, over 5720060.07 frames. ], giga_tot_loss[loss=0.3925, simple_loss=0.4301, pruned_loss=0.1774, over 5638563.86 frames. ], batch size: 705, lr: 1.66e-02, grad_scale: 4.0
+2023-02-28 23:02:00,847 INFO [train.py:968] (1/2) Epoch 1, batch 45400, giga_loss[loss=0.3802, simple_loss=0.4064, pruned_loss=0.177, over 23200.00 frames. ], tot_loss[loss=0.3933, simple_loss=0.4311, pruned_loss=0.1778, over 5648733.84 frames. ], libri_tot_loss[loss=0.4213, simple_loss=0.4478, pruned_loss=0.1974, over 5721955.58 frames. ], giga_tot_loss[loss=0.3908, simple_loss=0.4294, pruned_loss=0.1761, over 5635103.48 frames. ], batch size: 705, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:02:09,383 INFO [optim.py:369] (1/2) 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 23:02:34,738 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45438.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:02:38,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45441.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:02:44,728 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 1, batch 45450, libri_loss[loss=0.4759, simple_loss=0.4911, pruned_loss=0.2304, over 19632.00 frames. ], tot_loss[loss=0.3942, simple_loss=0.4311, pruned_loss=0.1787, over 5609941.65 frames. ], libri_tot_loss[loss=0.4229, simple_loss=0.4488, pruned_loss=0.1985, over 5689677.63 frames. ], giga_tot_loss[loss=0.3896, simple_loss=0.428, pruned_loss=0.1756, over 5627319.54 frames. ], batch size: 187, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:02:56,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45461.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:02:58,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45464.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:03,962 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45470.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:07,833 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=45474.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:24,547 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45493.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:03:30,667 INFO [train.py:968] (1/2) Epoch 1, batch 45500, giga_loss[loss=0.3278, simple_loss=0.3859, pruned_loss=0.1349, over 28157.00 frames. ], tot_loss[loss=0.3947, simple_loss=0.431, pruned_loss=0.1792, over 5592644.96 frames. ], libri_tot_loss[loss=0.4239, simple_loss=0.4495, pruned_loss=0.1992, over 5663040.40 frames. ], giga_tot_loss[loss=0.3898, simple_loss=0.4278, pruned_loss=0.1759, over 5629680.74 frames. ], batch size: 77, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:03:40,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9618, 1.0374, 0.9802, 0.4671], device='cuda:1'), covar=tensor([0.0341, 0.0317, 0.0296, 0.0366], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0621, 0.0747, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 23:03:41,087 INFO [optim.py:369] (1/2) 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 23:04:20,452 INFO [train.py:968] (1/2) Epoch 1, batch 45550, giga_loss[loss=0.4487, simple_loss=0.4745, pruned_loss=0.2115, over 28874.00 frames. ], tot_loss[loss=0.3996, simple_loss=0.4348, pruned_loss=0.1822, over 5603063.01 frames. ], libri_tot_loss[loss=0.4242, simple_loss=0.4497, pruned_loss=0.1994, over 5653921.20 frames. ], giga_tot_loss[loss=0.3954, simple_loss=0.4321, pruned_loss=0.1794, over 5640182.92 frames. ], batch size: 174, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:05:09,209 INFO [train.py:968] (1/2) Epoch 1, batch 45600, giga_loss[loss=0.4454, simple_loss=0.4762, pruned_loss=0.2072, over 28945.00 frames. ], tot_loss[loss=0.4039, simple_loss=0.4383, pruned_loss=0.1847, over 5573367.34 frames. ], libri_tot_loss[loss=0.4258, simple_loss=0.4507, pruned_loss=0.2005, over 5609298.60 frames. ], giga_tot_loss[loss=0.3986, simple_loss=0.4349, pruned_loss=0.1812, over 5643510.21 frames. ], batch size: 164, lr: 1.65e-02, grad_scale: 8.0
+2023-02-28 23:05:16,863 INFO [optim.py:369] (1/2) 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 23:05:51,786 INFO [train.py:968] (1/2) Epoch 1, batch 45650, giga_loss[loss=0.4423, simple_loss=0.4416, pruned_loss=0.2215, over 23568.00 frames. ], tot_loss[loss=0.4086, simple_loss=0.4415, pruned_loss=0.1879, over 5522184.49 frames. ], libri_tot_loss[loss=0.4273, simple_loss=0.4517, pruned_loss=0.2014, over 5533593.01 frames. ], giga_tot_loss[loss=0.4024, simple_loss=0.4375, pruned_loss=0.1837, over 5647290.03 frames. ], batch size: 705, lr: 1.65e-02, grad_scale: 4.0
+2023-02-28 23:06:15,163 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-02-28 23:07:27,469 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 50, giga_loss[loss=0.3819, simple_loss=0.4404, pruned_loss=0.1617, over 28933.00 frames. ], tot_loss[loss=0.3941, simple_loss=0.4432, pruned_loss=0.1725, over 1267613.41 frames. ], libri_tot_loss[loss=0.4048, simple_loss=0.4453, pruned_loss=0.1822, over 229226.31 frames. ], giga_tot_loss[loss=0.3924, simple_loss=0.443, pruned_loss=0.1709, over 1081430.82 frames. ], batch size: 213, lr: 1.62e-02, grad_scale: 4.0
+2023-02-28 23:07:55,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=45741.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:08:24,239 INFO [train.py:968] (1/2) Epoch 2, batch 100, libri_loss[loss=0.3638, simple_loss=0.4206, pruned_loss=0.1536, over 29661.00 frames. ], tot_loss[loss=0.3808, simple_loss=0.4312, pruned_loss=0.1652, over 2245598.03 frames. ], libri_tot_loss[loss=0.3784, simple_loss=0.4253, pruned_loss=0.1658, over 425124.84 frames. ], giga_tot_loss[loss=0.3822, simple_loss=0.433, pruned_loss=0.1657, over 1965170.65 frames. ], batch size: 91, lr: 1.62e-02, grad_scale: 4.0
+2023-02-28 23:08:44,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2898, 1.3640, 1.2712, 1.2500], device='cuda:1'), covar=tensor([0.0753, 0.0860, 0.1210, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0825, 0.0641, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 23:09:00,594 INFO [optim.py:369] (1/2) 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 23:09:10,274 INFO [train.py:968] (1/2) Epoch 2, batch 150, giga_loss[loss=0.309, simple_loss=0.346, pruned_loss=0.136, over 23911.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4105, pruned_loss=0.1526, over 3009668.33 frames. ], libri_tot_loss[loss=0.3685, simple_loss=0.4168, pruned_loss=0.16, over 505493.27 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.411, pruned_loss=0.1525, over 2745932.46 frames. ], batch size: 705, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:09:15,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45824.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:09:37,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45849.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:09:39,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-02-28 23:09:52,619 INFO [train.py:968] (1/2) Epoch 2, batch 200, giga_loss[loss=0.3253, simple_loss=0.372, pruned_loss=0.1394, over 28527.00 frames. ], tot_loss[loss=0.3425, simple_loss=0.3958, pruned_loss=0.1447, over 3595109.83 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.4148, pruned_loss=0.1581, over 646628.08 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3947, pruned_loss=0.1439, over 3330103.60 frames. ], batch size: 307, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:10:18,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3741, 1.8626, 1.4295, 1.1232], device='cuda:1'), covar=tensor([0.0910, 0.0872, 0.0906, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0471, 0.0361, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:1')
+2023-02-28 23:10:27,786 INFO [optim.py:369] (1/2) 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 23:10:36,030 INFO [train.py:968] (1/2) Epoch 2, batch 250, giga_loss[loss=0.2553, simple_loss=0.3178, pruned_loss=0.09642, over 28590.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3809, pruned_loss=0.1359, over 4057442.20 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.4148, pruned_loss=0.1581, over 646628.08 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3794, pruned_loss=0.1349, over 3851184.53 frames. ], batch size: 71, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:10:58,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3025, 1.3309, 1.2667, 1.2922], device='cuda:1'), covar=tensor([0.1918, 0.1916, 0.1601, 0.2032], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0775, 0.0864, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 23:11:14,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2458, 1.7272, 1.7778, 1.5748], device='cuda:1'), covar=tensor([0.0742, 0.1809, 0.1248, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0852, 0.0647, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 23:11:16,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45967.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:17,664 INFO [train.py:968] (1/2) Epoch 2, batch 300, giga_loss[loss=0.3035, simple_loss=0.3531, pruned_loss=0.1269, over 28874.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.373, pruned_loss=0.1316, over 4414536.86 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4165, pruned_loss=0.159, over 818898.35 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3696, pruned_loss=0.1295, over 4203629.96 frames. ], batch size: 213, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:11:19,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45970.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:39,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45992.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:41,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45995.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:48,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45999.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:11:59,249 INFO [optim.py:369] (1/2) 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 23:12:06,624 INFO [train.py:968] (1/2) Epoch 2, batch 350, giga_loss[loss=0.2823, simple_loss=0.3486, pruned_loss=0.108, over 28773.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3629, pruned_loss=0.1256, over 4694812.08 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4167, pruned_loss=0.1582, over 869654.16 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3595, pruned_loss=0.1237, over 4517143.78 frames. ], batch size: 174, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:12:11,901 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46024.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:12:21,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8553, 1.8254, 4.4212, 3.3765], device='cuda:1'), covar=tensor([0.1496, 0.1384, 0.0227, 0.0442], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0483, 0.0620, 0.0488], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 23:12:22,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3464, 1.4422, 1.3078, 1.2530], device='cuda:1'), covar=tensor([0.1849, 0.1872, 0.1654, 0.1861], device='cuda:1'), in_proj_covar=tensor([0.0949, 0.0793, 0.0880, 0.0942], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 23:12:46,719 INFO [train.py:968] (1/2) Epoch 2, batch 400, libri_loss[loss=0.3507, simple_loss=0.4175, pruned_loss=0.1419, over 29509.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3598, pruned_loss=0.1232, over 4925013.57 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4178, pruned_loss=0.1572, over 1090174.69 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.354, pruned_loss=0.1201, over 4737374.18 frames. ], batch size: 84, lr: 1.61e-02, grad_scale: 8.0
+2023-02-28 23:12:46,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3037, 4.1226, 5.0458, 2.0683], device='cuda:1'), covar=tensor([0.0344, 0.0377, 0.0708, 0.1862], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0494, 0.0793, 0.0538], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-02-28 23:13:10,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46095.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:13:20,300 INFO [optim.py:369] (1/2) 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 23:13:25,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46116.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:13:26,888 INFO [train.py:968] (1/2) Epoch 2, batch 450, giga_loss[loss=0.2451, simple_loss=0.3045, pruned_loss=0.09281, over 28573.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3578, pruned_loss=0.1226, over 5102107.54 frames. ], libri_tot_loss[loss=0.3639, simple_loss=0.4156, pruned_loss=0.1561, over 1232931.40 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3519, pruned_loss=0.1193, over 4927738.42 frames. ], batch size: 85, lr: 1.61e-02, grad_scale: 8.0
+2023-02-28 23:13:42,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4647, 2.3494, 1.4784, 1.2447], device='cuda:1'), covar=tensor([0.0901, 0.0850, 0.0870, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0469, 0.0361, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:1')
+2023-02-28 23:14:11,205 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 500, giga_loss[loss=0.2376, simple_loss=0.3034, pruned_loss=0.08589, over 28429.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3541, pruned_loss=0.1204, over 5228297.80 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4151, pruned_loss=0.1553, over 1345042.65 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.348, pruned_loss=0.1172, over 5075474.44 frames. ], batch size: 71, lr: 1.61e-02, grad_scale: 8.0
+2023-02-28 23:14:45,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-02-28 23:14:48,992 INFO [optim.py:369] (1/2) 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 23:14:56,631 INFO [train.py:968] (1/2) Epoch 2, batch 550, giga_loss[loss=0.263, simple_loss=0.3293, pruned_loss=0.09831, over 29038.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3524, pruned_loss=0.1193, over 5335664.63 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.4139, pruned_loss=0.1536, over 1502304.56 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3457, pruned_loss=0.1158, over 5192876.46 frames. ], batch size: 128, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:15:08,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7689, 1.4586, 1.2674, 1.2481], device='cuda:1'), covar=tensor([0.0620, 0.0805, 0.0995, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0535, 0.0566, 0.0502], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 23:15:17,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-02-28 23:15:25,659 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46252.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:15:32,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46259.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:15:34,332 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 600, giga_loss[loss=0.2802, simple_loss=0.3421, pruned_loss=0.1091, over 28856.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3501, pruned_loss=0.118, over 5403920.42 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.4128, pruned_loss=0.1528, over 1640310.48 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3429, pruned_loss=0.1144, over 5284723.66 frames. ], batch size: 174, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:15:45,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 23:16:03,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46291.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:16:22,223 INFO [optim.py:369] (1/2) 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 23:16:24,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7263, 1.4564, 1.2831, 1.1860], device='cuda:1'), covar=tensor([0.0611, 0.0741, 0.1002, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0538, 0.0570, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-02-28 23:16:30,298 INFO [train.py:968] (1/2) Epoch 2, batch 650, giga_loss[loss=0.2626, simple_loss=0.3296, pruned_loss=0.09775, over 28883.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3469, pruned_loss=0.116, over 5472319.43 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4138, pruned_loss=0.1534, over 1724659.43 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3397, pruned_loss=0.1122, over 5369676.17 frames. ], batch size: 174, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:17:20,665 INFO [train.py:968] (1/2) Epoch 2, batch 700, giga_loss[loss=0.2534, simple_loss=0.3162, pruned_loss=0.09534, over 28885.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3419, pruned_loss=0.1129, over 5514283.61 frames. ], libri_tot_loss[loss=0.3609, simple_loss=0.4142, pruned_loss=0.1538, over 1735578.13 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3359, pruned_loss=0.1098, over 5440749.32 frames. ], batch size: 112, lr: 1.61e-02, grad_scale: 4.0
+2023-02-28 23:17:38,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-02-28 23:17:58,329 INFO [optim.py:369] (1/2) 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 23:18:05,575 INFO [train.py:968] (1/2) Epoch 2, batch 750, giga_loss[loss=0.2948, simple_loss=0.3419, pruned_loss=0.1238, over 28761.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.34, pruned_loss=0.1122, over 5549624.98 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4155, pruned_loss=0.155, over 1816709.39 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3335, pruned_loss=0.1085, over 5485973.87 frames. ], batch size: 262, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:18:28,362 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-02-28 23:18:49,614 INFO [train.py:968] (1/2) Epoch 2, batch 800, giga_loss[loss=0.315, simple_loss=0.3441, pruned_loss=0.143, over 24044.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.339, pruned_loss=0.1126, over 5579026.49 frames. ], libri_tot_loss[loss=0.3639, simple_loss=0.4163, pruned_loss=0.1558, over 1897317.80 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3324, pruned_loss=0.1087, over 5523117.06 frames. ], batch size: 705, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:18:51,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46470.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:19:26,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6242, 1.4333, 1.2245, 1.3000], device='cuda:1'), covar=tensor([0.0675, 0.0781, 0.1053, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0534, 0.0561, 0.0498], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 23:19:35,059 INFO [optim.py:369] (1/2) 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 23:19:41,752 INFO [train.py:968] (1/2) Epoch 2, batch 850, giga_loss[loss=0.4852, simple_loss=0.5013, pruned_loss=0.2345, over 28297.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3503, pruned_loss=0.1202, over 5606460.92 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4163, pruned_loss=0.1565, over 1957168.70 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3441, pruned_loss=0.1164, over 5558041.13 frames. ], batch size: 65, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:19:54,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2875, 1.2504, 2.8920, 2.5700], device='cuda:1'), covar=tensor([0.1470, 0.1437, 0.0416, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0469, 0.0604, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 23:20:05,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46542.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:20:16,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-02-28 23:20:31,320 INFO [train.py:968] (1/2) Epoch 2, batch 900, giga_loss[loss=0.3744, simple_loss=0.4241, pruned_loss=0.1623, over 28888.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3671, pruned_loss=0.1302, over 5625701.57 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.4159, pruned_loss=0.1557, over 2069873.15 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3608, pruned_loss=0.1266, over 5582981.56 frames. ], batch size: 186, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:21:05,061 INFO [optim.py:369] (1/2) 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 23:21:06,965 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46613.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:10,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46616.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:12,642 INFO [train.py:968] (1/2) Epoch 2, batch 950, giga_loss[loss=0.3783, simple_loss=0.4284, pruned_loss=0.1641, over 28557.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3813, pruned_loss=0.1394, over 5623284.54 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4164, pruned_loss=0.1563, over 2201100.67 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3746, pruned_loss=0.1356, over 5596969.16 frames. ], batch size: 336, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:21:19,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46627.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:36,432 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46645.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:21:54,917 INFO [train.py:968] (1/2) Epoch 2, batch 1000, libri_loss[loss=0.3616, simple_loss=0.3979, pruned_loss=0.1627, over 29347.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3904, pruned_loss=0.1442, over 5637929.17 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.416, pruned_loss=0.1565, over 2273941.11 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3848, pruned_loss=0.1409, over 5612569.48 frames. ], batch size: 67, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:21:55,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46669.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:22:07,962 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46685.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:22:09,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46688.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:22:29,085 INFO [optim.py:369] (1/2) 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 23:22:32,695 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 1050, libri_loss[loss=0.4235, simple_loss=0.4504, pruned_loss=0.1982, over 19376.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3945, pruned_loss=0.1444, over 5643887.77 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.4163, pruned_loss=0.1567, over 2346271.50 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3894, pruned_loss=0.1415, over 5634425.85 frames. ], batch size: 187, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:23:20,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-02-28 23:23:27,801 INFO [train.py:968] (1/2) Epoch 2, batch 1100, libri_loss[loss=0.362, simple_loss=0.4264, pruned_loss=0.1488, over 29666.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3947, pruned_loss=0.1431, over 5638197.29 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4164, pruned_loss=0.1567, over 2389551.33 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3903, pruned_loss=0.1405, over 5636128.64 frames. ], batch size: 91, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:23:28,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46770.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:23:32,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46773.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:23:58,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46802.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:24:07,203 INFO [optim.py:369] (1/2) 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 23:24:13,333 INFO [train.py:968] (1/2) Epoch 2, batch 1150, giga_loss[loss=0.3532, simple_loss=0.4008, pruned_loss=0.1528, over 28848.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.398, pruned_loss=0.1459, over 5646163.62 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4176, pruned_loss=0.1577, over 2441570.15 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3939, pruned_loss=0.1434, over 5641485.48 frames. ], batch size: 119, lr: 1.60e-02, grad_scale: 4.0
+2023-02-28 23:24:57,108 INFO [train.py:968] (1/2) Epoch 2, batch 1200, giga_loss[loss=0.3657, simple_loss=0.4247, pruned_loss=0.1534, over 28770.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.401, pruned_loss=0.1478, over 5668685.56 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.417, pruned_loss=0.1568, over 2691126.26 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3965, pruned_loss=0.1454, over 5649011.30 frames. ], batch size: 99, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:25:30,667 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-02-28 23:25:35,489 INFO [optim.py:369] (1/2) 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 23:25:43,305 INFO [train.py:968] (1/2) Epoch 2, batch 1250, libri_loss[loss=0.4177, simple_loss=0.4691, pruned_loss=0.1832, over 29237.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4038, pruned_loss=0.1498, over 5674663.66 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4167, pruned_loss=0.1562, over 2783300.40 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3999, pruned_loss=0.1478, over 5655427.30 frames. ], batch size: 94, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:25:53,829 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-02-28 23:26:26,378 INFO [train.py:968] (1/2) Epoch 2, batch 1300, giga_loss[loss=0.3605, simple_loss=0.4176, pruned_loss=0.1517, over 28875.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.407, pruned_loss=0.1506, over 5692015.94 frames. ], libri_tot_loss[loss=0.3635, simple_loss=0.4159, pruned_loss=0.1555, over 2890825.28 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.4039, pruned_loss=0.1491, over 5669509.32 frames. ], batch size: 227, lr: 1.60e-02, grad_scale: 8.0
+2023-02-28 23:26:39,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-02-28 23:26:43,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46988.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:26:45,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7895, 2.0176, 1.6620, 1.6420], device='cuda:1'), covar=tensor([0.1088, 0.1049, 0.0878, 0.0552], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0825, 0.0699, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 23:26:50,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=46995.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:27:04,975 INFO [optim.py:369] (1/2) 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 23:27:09,829 INFO [train.py:968] (1/2) Epoch 2, batch 1350, giga_loss[loss=0.3963, simple_loss=0.4491, pruned_loss=0.1717, over 29015.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4096, pruned_loss=0.1521, over 5687998.65 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.4159, pruned_loss=0.1556, over 2920682.69 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4071, pruned_loss=0.1509, over 5668701.17 frames. ], batch size: 145, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:27:12,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-02-28 23:27:33,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47044.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:27:56,384 INFO [train.py:968] (1/2) Epoch 2, batch 1400, giga_loss[loss=0.3664, simple_loss=0.4187, pruned_loss=0.1571, over 28675.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.41, pruned_loss=0.1512, over 5691104.98 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4157, pruned_loss=0.1554, over 2935661.27 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4081, pruned_loss=0.1503, over 5674904.32 frames. ], batch size: 262, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:28:32,887 INFO [optim.py:369] (1/2) 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 23:28:37,725 INFO [train.py:968] (1/2) Epoch 2, batch 1450, giga_loss[loss=0.3274, simple_loss=0.4106, pruned_loss=0.1222, over 28701.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4091, pruned_loss=0.1487, over 5697775.67 frames. ], libri_tot_loss[loss=0.3623, simple_loss=0.4152, pruned_loss=0.1547, over 3073106.77 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4075, pruned_loss=0.1481, over 5682189.68 frames. ], batch size: 60, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:28:51,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2875, 1.4311, 2.9231, 2.5781], device='cuda:1'), covar=tensor([0.1448, 0.1232, 0.0384, 0.0518], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0471, 0.0596, 0.0483], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 23:28:53,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-02-28 23:29:21,297 INFO [train.py:968] (1/2) Epoch 2, batch 1500, giga_loss[loss=0.3309, simple_loss=0.3932, pruned_loss=0.1343, over 28889.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.4074, pruned_loss=0.1465, over 5701935.27 frames. ], libri_tot_loss[loss=0.3622, simple_loss=0.4152, pruned_loss=0.1546, over 3115072.61 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4061, pruned_loss=0.146, over 5687084.32 frames. ], batch size: 186, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:29:22,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9749, 1.8617, 4.9968, 3.5099], device='cuda:1'), covar=tensor([0.1509, 0.1375, 0.0179, 0.0362], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0475, 0.0600, 0.0485], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 23:29:38,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47187.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:29:40,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47190.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:29:50,119 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=47203.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:29:56,933 INFO [optim.py:369] (1/2) 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 23:30:03,140 INFO [train.py:968] (1/2) Epoch 2, batch 1550, giga_loss[loss=0.3251, simple_loss=0.3927, pruned_loss=0.1287, over 28920.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.4047, pruned_loss=0.1435, over 5712401.07 frames. ], libri_tot_loss[loss=0.3619, simple_loss=0.4149, pruned_loss=0.1544, over 3156300.36 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.4036, pruned_loss=0.1431, over 5698316.14 frames. ], batch size: 136, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:30:03,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47219.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:30:39,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2704, 3.0818, 3.9690, 1.7775], device='cuda:1'), covar=tensor([0.0416, 0.0561, 0.0830, 0.1892], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0496, 0.0772, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-02-28 23:30:52,162 INFO [train.py:968] (1/2) Epoch 2, batch 1600, libri_loss[loss=0.4061, simple_loss=0.4497, pruned_loss=0.1813, over 29484.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.4064, pruned_loss=0.1467, over 5698148.20 frames. ], libri_tot_loss[loss=0.362, simple_loss=0.415, pruned_loss=0.1545, over 3197395.97 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4054, pruned_loss=0.1461, over 5684099.97 frames. ], batch size: 85, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:31:32,907 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 1650, giga_loss[loss=0.3762, simple_loss=0.423, pruned_loss=0.1648, over 28797.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4093, pruned_loss=0.1515, over 5706012.92 frames. ], libri_tot_loss[loss=0.3622, simple_loss=0.4153, pruned_loss=0.1546, over 3250988.58 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4082, pruned_loss=0.1509, over 5691408.04 frames. ], batch size: 119, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:32:15,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47363.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:32:21,375 INFO [train.py:968] (1/2) Epoch 2, batch 1700, giga_loss[loss=0.3539, simple_loss=0.401, pruned_loss=0.1534, over 28975.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4122, pruned_loss=0.1553, over 5708271.07 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4162, pruned_loss=0.1551, over 3344947.88 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4107, pruned_loss=0.1545, over 5698674.86 frames. ], batch size: 136, lr: 1.59e-02, grad_scale: 8.0
+2023-02-28 23:32:22,495 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47370.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:32:59,281 INFO [optim.py:369] (1/2) 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 23:33:05,218 INFO [train.py:968] (1/2) Epoch 2, batch 1750, giga_loss[loss=0.3417, simple_loss=0.3967, pruned_loss=0.1434, over 28566.00 frames. ], tot_loss[loss=0.361, simple_loss=0.4109, pruned_loss=0.1555, over 5702707.03 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4163, pruned_loss=0.155, over 3408209.83 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4095, pruned_loss=0.155, over 5690648.08 frames. ], batch size: 71, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:33:08,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-02-28 23:33:48,391 INFO [train.py:968] (1/2) Epoch 2, batch 1800, giga_loss[loss=0.328, simple_loss=0.3813, pruned_loss=0.1373, over 29031.00 frames. ], tot_loss[loss=0.359, simple_loss=0.4083, pruned_loss=0.1548, over 5695546.23 frames. ], libri_tot_loss[loss=0.3618, simple_loss=0.4149, pruned_loss=0.1543, over 3503065.90 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4077, pruned_loss=0.1548, over 5681314.17 frames. ], batch size: 136, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:33:54,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7004, 1.5807, 1.3486, 1.3260], device='cuda:1'), covar=tensor([0.0717, 0.0776, 0.0923, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0527, 0.0565, 0.0500], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 23:34:20,746 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47506.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:25,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47509.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:30,432 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47513.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:30,799 INFO [optim.py:369] (1/2) 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 23:34:32,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47516.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:34:34,395 INFO [train.py:968] (1/2) Epoch 2, batch 1850, giga_loss[loss=0.3478, simple_loss=0.4058, pruned_loss=0.1449, over 28854.00 frames. ], tot_loss[loss=0.3588, simple_loss=0.4084, pruned_loss=0.1546, over 5698393.42 frames. ], libri_tot_loss[loss=0.3625, simple_loss=0.4155, pruned_loss=0.1547, over 3550554.93 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4074, pruned_loss=0.1543, over 5683587.59 frames. ], batch size: 186, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:34:54,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47538.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:35:01,782 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47545.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:35:27,844 INFO [train.py:968] (1/2) Epoch 2, batch 1900, giga_loss[loss=0.3196, simple_loss=0.384, pruned_loss=0.1276, over 28677.00 frames. ], tot_loss[loss=0.355, simple_loss=0.4063, pruned_loss=0.1519, over 5698253.08 frames. ], libri_tot_loss[loss=0.3625, simple_loss=0.4159, pruned_loss=0.1546, over 3608043.43 frames. ], giga_tot_loss[loss=0.3543, simple_loss=0.4051, pruned_loss=0.1517, over 5682508.19 frames. ], batch size: 242, lr: 1.59e-02, grad_scale: 4.0
+2023-02-28 23:35:34,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8339, 1.6275, 1.3967, 1.5083], device='cuda:1'), covar=tensor([0.0649, 0.0663, 0.0848, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0528, 0.0556, 0.0497], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 23:35:40,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47578.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:35:54,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0501, 1.7485, 1.4484, 1.6985], device='cuda:1'), covar=tensor([0.0557, 0.0728, 0.0872, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0531, 0.0557, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 23:36:23,212 INFO [optim.py:369] (1/2) 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 23:36:28,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3753, 1.7527, 1.5992, 1.5663], device='cuda:1'), covar=tensor([0.1193, 0.1407, 0.0992, 0.0631], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0824, 0.0704, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 23:36:28,933 INFO [train.py:968] (1/2) Epoch 2, batch 1950, giga_loss[loss=0.3346, simple_loss=0.3869, pruned_loss=0.1412, over 28823.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.4007, pruned_loss=0.1478, over 5693741.88 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.4154, pruned_loss=0.1544, over 3642237.18 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.3999, pruned_loss=0.1477, over 5678715.79 frames. ], batch size: 284, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:36:43,742 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 2000, giga_loss[loss=0.2802, simple_loss=0.3415, pruned_loss=0.1094, over 28550.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3933, pruned_loss=0.1426, over 5687465.61 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.4158, pruned_loss=0.1547, over 3718721.29 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3918, pruned_loss=0.1422, over 5670297.70 frames. ], batch size: 71, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:37:58,891 INFO [optim.py:369] (1/2) 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 23:38:02,275 INFO [train.py:968] (1/2) Epoch 2, batch 2050, giga_loss[loss=0.277, simple_loss=0.3424, pruned_loss=0.1058, over 29011.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3883, pruned_loss=0.1393, over 5686755.89 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4163, pruned_loss=0.1551, over 3801965.92 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3859, pruned_loss=0.1382, over 5667511.70 frames. ], batch size: 128, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:38:04,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47721.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:38:07,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47724.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:38:13,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2416, 1.2197, 1.1528, 1.0127], device='cuda:1'), covar=tensor([0.1365, 0.1244, 0.1081, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0774, 0.0861, 0.0924], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 23:38:36,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 2100, giga_loss[loss=0.3306, simple_loss=0.3916, pruned_loss=0.1348, over 28846.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3838, pruned_loss=0.1374, over 5664884.02 frames. ], libri_tot_loss[loss=0.3631, simple_loss=0.4162, pruned_loss=0.155, over 3821422.70 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3817, pruned_loss=0.1363, over 5649038.89 frames. ], batch size: 99, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:38:57,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=47772.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:39:16,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-02-28 23:39:31,943 INFO [optim.py:369] (1/2) 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 23:39:34,687 INFO [train.py:968] (1/2) Epoch 2, batch 2150, giga_loss[loss=0.3331, simple_loss=0.3851, pruned_loss=0.1406, over 28735.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3858, pruned_loss=0.138, over 5667951.92 frames. ], libri_tot_loss[loss=0.3631, simple_loss=0.4164, pruned_loss=0.1549, over 3867848.51 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3831, pruned_loss=0.1368, over 5662651.18 frames. ], batch size: 99, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:40:14,066 INFO [train.py:968] (1/2) Epoch 2, batch 2200, libri_loss[loss=0.4689, simple_loss=0.4943, pruned_loss=0.2217, over 28674.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3861, pruned_loss=0.1378, over 5685617.64 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4177, pruned_loss=0.1557, over 3936262.44 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3822, pruned_loss=0.1358, over 5675992.83 frames. ], batch size: 106, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:40:52,676 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 2250, giga_loss[loss=0.2994, simple_loss=0.3514, pruned_loss=0.1237, over 28763.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3834, pruned_loss=0.1361, over 5692803.20 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4189, pruned_loss=0.1563, over 3984847.41 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3787, pruned_loss=0.1336, over 5680521.38 frames. ], batch size: 99, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:41:15,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4729, 2.0779, 1.5602, 0.6470], device='cuda:1'), covar=tensor([0.1978, 0.1329, 0.1422, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1086, 0.1048, 0.1077, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 23:41:39,413 INFO [train.py:968] (1/2) Epoch 2, batch 2300, giga_loss[loss=0.3484, simple_loss=0.3799, pruned_loss=0.1585, over 28817.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3802, pruned_loss=0.1344, over 5688535.68 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4192, pruned_loss=0.1564, over 3999845.89 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3758, pruned_loss=0.1321, over 5689091.07 frames. ], batch size: 86, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:42:03,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5266, 2.1573, 1.4048, 1.1796], device='cuda:1'), covar=tensor([0.0969, 0.0676, 0.0920, 0.1652], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0448, 0.0340, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0014], device='cuda:1')
+2023-02-28 23:42:15,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48009.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:42:20,204 INFO [optim.py:369] (1/2) 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 23:42:22,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-02-28 23:42:23,020 INFO [train.py:968] (1/2) Epoch 2, batch 2350, giga_loss[loss=0.2634, simple_loss=0.3329, pruned_loss=0.09696, over 28744.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3761, pruned_loss=0.1314, over 5693651.05 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4199, pruned_loss=0.1566, over 4028454.92 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3717, pruned_loss=0.1291, over 5691223.59 frames. ], batch size: 119, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:43:06,489 INFO [train.py:968] (1/2) Epoch 2, batch 2400, giga_loss[loss=0.3335, simple_loss=0.3679, pruned_loss=0.1496, over 24438.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.375, pruned_loss=0.1311, over 5695027.22 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4218, pruned_loss=0.1578, over 4073884.62 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3692, pruned_loss=0.128, over 5689402.90 frames. ], batch size: 710, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:43:35,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9097, 1.0598, 0.9408, 0.4707], device='cuda:1'), covar=tensor([0.0373, 0.0293, 0.0328, 0.0413], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0645, 0.0734, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-02-28 23:43:42,994 INFO [optim.py:369] (1/2) 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 23:43:43,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-02-28 23:43:46,337 INFO [train.py:968] (1/2) Epoch 2, batch 2450, giga_loss[loss=0.287, simple_loss=0.3512, pruned_loss=0.1114, over 28869.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3713, pruned_loss=0.129, over 5703820.67 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4213, pruned_loss=0.1574, over 4108497.22 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3662, pruned_loss=0.1263, over 5697167.47 frames. ], batch size: 145, lr: 1.58e-02, grad_scale: 8.0
+2023-02-28 23:43:55,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48131.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:06,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48147.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:10,748 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48152.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:13,755 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 2500, giga_loss[loss=0.2957, simple_loss=0.3505, pruned_loss=0.1204, over 28919.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3681, pruned_loss=0.1272, over 5712353.07 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4218, pruned_loss=0.1575, over 4135601.81 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.363, pruned_loss=0.1245, over 5704309.23 frames. ], batch size: 186, lr: 1.58e-02, grad_scale: 4.0
+2023-02-28 23:44:28,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3277, 1.2819, 1.2937, 1.1857], device='cuda:1'), covar=tensor([0.1732, 0.1883, 0.1451, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0784, 0.0848, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 23:44:35,206 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48181.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:37,428 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48184.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:44:51,597 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48202.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:45:03,896 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 2550, giga_loss[loss=0.2931, simple_loss=0.3455, pruned_loss=0.1203, over 28518.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3678, pruned_loss=0.1267, over 5721367.66 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4235, pruned_loss=0.1582, over 4193903.23 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3609, pruned_loss=0.1231, over 5711431.45 frames. ], batch size: 78, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:45:33,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3521, 1.2256, 1.0233, 1.3934], device='cuda:1'), covar=tensor([0.1217, 0.0566, 0.0602, 0.1409], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0239, 0.0239, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0020, 0.0017, 0.0029], device='cuda:1')
+2023-02-28 23:45:42,604 INFO [train.py:968] (1/2) Epoch 2, batch 2600, libri_loss[loss=0.3825, simple_loss=0.4278, pruned_loss=0.1686, over 29364.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3677, pruned_loss=0.1264, over 5724505.28 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4246, pruned_loss=0.1591, over 4256624.96 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3594, pruned_loss=0.1219, over 5714031.49 frames. ], batch size: 71, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:46:00,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48290.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:46:02,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48293.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:46:17,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5089, 1.3822, 1.2773, 1.5628], device='cuda:1'), covar=tensor([0.1627, 0.1695, 0.1383, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0771, 0.0841, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0007], device='cuda:1')
+2023-02-28 23:46:18,980 INFO [optim.py:369] (1/2) 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 23:46:22,438 INFO [train.py:968] (1/2) Epoch 2, batch 2650, giga_loss[loss=0.267, simple_loss=0.3356, pruned_loss=0.09921, over 28643.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3678, pruned_loss=0.1265, over 5718750.32 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4257, pruned_loss=0.1596, over 4306558.23 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3584, pruned_loss=0.1213, over 5716740.34 frames. ], batch size: 242, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:46:25,684 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48322.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:46:31,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2633, 1.6736, 1.1857, 1.1556], device='cuda:1'), covar=tensor([0.0869, 0.0821, 0.0856, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0446, 0.0342, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0010, 0.0014], device='cuda:1')
+2023-02-28 23:47:07,071 INFO [train.py:968] (1/2) Epoch 2, batch 2700, giga_loss[loss=0.374, simple_loss=0.4202, pruned_loss=0.1639, over 28725.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3707, pruned_loss=0.1289, over 5723474.63 frames. ], libri_tot_loss[loss=0.3723, simple_loss=0.4256, pruned_loss=0.1595, over 4337949.36 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3622, pruned_loss=0.1242, over 5718814.29 frames. ], batch size: 262, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:47:50,580 INFO [optim.py:369] (1/2) 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 23:47:51,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-02-28 23:47:54,477 INFO [train.py:968] (1/2) Epoch 2, batch 2750, giga_loss[loss=0.3921, simple_loss=0.4346, pruned_loss=0.1748, over 28916.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3776, pruned_loss=0.1342, over 5718649.14 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4254, pruned_loss=0.1593, over 4375682.42 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3698, pruned_loss=0.13, over 5711612.78 frames. ], batch size: 145, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:48:39,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48465.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:48:41,708 INFO [train.py:968] (1/2) Epoch 2, batch 2800, giga_loss[loss=0.4608, simple_loss=0.4779, pruned_loss=0.2219, over 28234.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3848, pruned_loss=0.139, over 5719594.44 frames. ], libri_tot_loss[loss=0.3717, simple_loss=0.4251, pruned_loss=0.1592, over 4411419.61 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3778, pruned_loss=0.1352, over 5711645.10 frames. ], batch size: 368, lr: 1.57e-02, grad_scale: 8.0
+2023-02-28 23:49:21,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48506.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:49:31,006 INFO [optim.py:369] (1/2) 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 23:49:32,409 INFO [train.py:968] (1/2) Epoch 2, batch 2850, giga_loss[loss=0.4244, simple_loss=0.4531, pruned_loss=0.1979, over 28745.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3946, pruned_loss=0.1469, over 5701746.14 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4253, pruned_loss=0.1594, over 4418592.51 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3887, pruned_loss=0.1438, over 5694831.19 frames. ], batch size: 284, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:50:12,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48556.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:50:25,351 INFO [train.py:968] (1/2) Epoch 2, batch 2900, giga_loss[loss=0.3354, simple_loss=0.4033, pruned_loss=0.1337, over 28973.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.3989, pruned_loss=0.1479, over 5709774.20 frames. ], libri_tot_loss[loss=0.3723, simple_loss=0.4256, pruned_loss=0.1595, over 4425759.55 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3941, pruned_loss=0.1454, over 5703750.89 frames. ], batch size: 145, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:50:35,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48577.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:50:59,870 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48605.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:51:08,888 INFO [optim.py:369] (1/2) 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 23:51:10,210 INFO [train.py:968] (1/2) Epoch 2, batch 2950, giga_loss[loss=0.3512, simple_loss=0.4151, pruned_loss=0.1437, over 29035.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4052, pruned_loss=0.1521, over 5710387.42 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4251, pruned_loss=0.1596, over 4477081.01 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4008, pruned_loss=0.1496, over 5698770.98 frames. ], batch size: 155, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:51:43,965 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48649.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:51:46,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48652.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:51:51,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5055, 1.3324, 1.3830, 1.3178], device='cuda:1'), covar=tensor([0.0678, 0.1072, 0.1068, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0827, 0.0635, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 23:51:57,801 INFO [train.py:968] (1/2) Epoch 2, batch 3000, libri_loss[loss=0.3669, simple_loss=0.4257, pruned_loss=0.1541, over 29529.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4123, pruned_loss=0.1582, over 5694343.98 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4249, pruned_loss=0.1595, over 4521578.35 frames. ], giga_tot_loss[loss=0.3604, simple_loss=0.4085, pruned_loss=0.1561, over 5682192.59 frames. ], batch size: 82, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:51:57,801 INFO [train.py:1003] (1/2) Computing validation loss
+2023-02-28 23:52:07,256 INFO [train.py:1012] (1/2) Epoch 2, validation: loss=0.294, simple_loss=0.3813, pruned_loss=0.1034, over 944034.00 frames. 
+2023-02-28 23:52:07,257 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-02-28 23:52:17,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48681.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:52:35,518 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48699.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:52:38,854 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:52:52,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48717.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:52:54,062 INFO [train.py:968] (1/2) Epoch 2, batch 3050, libri_loss[loss=0.3668, simple_loss=0.4227, pruned_loss=0.1554, over 26011.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.4081, pruned_loss=0.1547, over 5687786.86 frames. ], libri_tot_loss[loss=0.3717, simple_loss=0.4246, pruned_loss=0.1594, over 4563065.73 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4047, pruned_loss=0.153, over 5682787.99 frames. ], batch size: 136, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:52:55,202 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48720.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:52:57,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48723.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:53:06,620 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48731.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:53:28,383 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48752.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:53:45,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9044, 1.5872, 1.4895, 1.5466], device='cuda:1'), covar=tensor([0.0665, 0.1276, 0.1232, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0822, 0.0632, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-02-28 23:53:46,203 INFO [train.py:968] (1/2) Epoch 2, batch 3100, giga_loss[loss=0.34, simple_loss=0.401, pruned_loss=0.1394, over 28895.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.402, pruned_loss=0.1493, over 5694588.25 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4248, pruned_loss=0.1595, over 4582354.31 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3989, pruned_loss=0.1478, over 5688455.74 frames. ], batch size: 145, lr: 1.57e-02, grad_scale: 4.0
+2023-02-28 23:53:46,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4446, 3.2577, 2.6579, 2.3960], device='cuda:1'), covar=tensor([0.1280, 0.0974, 0.0843, 0.0550], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0816, 0.0704, 0.0603], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-02-28 23:54:06,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48790.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:54:37,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=48816.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:54:38,211 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 3150, giga_loss[loss=0.3425, simple_loss=0.3971, pruned_loss=0.144, over 29126.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3996, pruned_loss=0.1465, over 5704198.76 frames. ], libri_tot_loss[loss=0.3718, simple_loss=0.4247, pruned_loss=0.1594, over 4602268.81 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3969, pruned_loss=0.1452, over 5696413.00 frames. ], batch size: 128, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:54:44,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-02-28 23:54:59,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48840.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:55:23,484 INFO [train.py:968] (1/2) Epoch 2, batch 3200, giga_loss[loss=0.3786, simple_loss=0.4295, pruned_loss=0.1638, over 28512.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.4009, pruned_loss=0.1467, over 5709005.06 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4239, pruned_loss=0.1587, over 4645462.20 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3986, pruned_loss=0.1457, over 5697932.51 frames. ], batch size: 336, lr: 1.56e-02, grad_scale: 8.0
+2023-02-28 23:56:05,188 INFO [optim.py:369] (1/2) 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 23:56:06,409 INFO [train.py:968] (1/2) Epoch 2, batch 3250, giga_loss[loss=0.3257, simple_loss=0.3931, pruned_loss=0.1291, over 29069.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4036, pruned_loss=0.1476, over 5713513.04 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4234, pruned_loss=0.1583, over 4664399.57 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.4018, pruned_loss=0.147, over 5701993.21 frames. ], batch size: 128, lr: 1.56e-02, grad_scale: 8.0
+2023-02-28 23:56:51,398 INFO [train.py:968] (1/2) Epoch 2, batch 3300, giga_loss[loss=0.3609, simple_loss=0.4116, pruned_loss=0.1551, over 28772.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4054, pruned_loss=0.1489, over 5714259.77 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4233, pruned_loss=0.1583, over 4701563.84 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4034, pruned_loss=0.1481, over 5699519.77 frames. ], batch size: 284, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:57:00,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:57:02,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48983.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:57:06,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48986.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:57:08,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-02-28 23:57:31,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49015.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:57:33,804 INFO [optim.py:369] (1/2) 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 23:57:34,522 INFO [train.py:968] (1/2) Epoch 2, batch 3350, giga_loss[loss=0.3606, simple_loss=0.4135, pruned_loss=0.1539, over 28905.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4066, pruned_loss=0.1501, over 5715298.35 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.423, pruned_loss=0.158, over 4741762.07 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4048, pruned_loss=0.1495, over 5698469.88 frames. ], batch size: 106, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:58:18,136 INFO [train.py:968] (1/2) Epoch 2, batch 3400, giga_loss[loss=0.4041, simple_loss=0.4451, pruned_loss=0.1815, over 29087.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4079, pruned_loss=0.1516, over 5718256.86 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4229, pruned_loss=0.158, over 4761817.93 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4063, pruned_loss=0.151, over 5704235.80 frames. ], batch size: 155, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:58:39,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49092.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:58:46,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2970, 1.2608, 1.2651, 1.1699], device='cuda:1'), covar=tensor([0.1785, 0.1871, 0.1448, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0778, 0.0852, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-02-28 23:59:01,611 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 3450, giga_loss[loss=0.3678, simple_loss=0.4165, pruned_loss=0.1596, over 28898.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4086, pruned_loss=0.152, over 5726706.64 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4231, pruned_loss=0.1581, over 4789744.30 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4068, pruned_loss=0.1513, over 5711787.89 frames. ], batch size: 174, lr: 1.56e-02, grad_scale: 4.0
+2023-02-28 23:59:06,411 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49123.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:59:08,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49126.0, num_to_drop=1, layers_to_drop={0}
+2023-02-28 23:59:09,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2739, 1.4839, 1.1392, 1.3047], device='cuda:1'), covar=tensor([0.1123, 0.0503, 0.0557, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0229, 0.0233, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0019, 0.0017, 0.0029], device='cuda:1')
+2023-02-28 23:59:10,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2962, 1.8235, 1.4135, 0.3596], device='cuda:1'), covar=tensor([0.1226, 0.0731, 0.1122, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.1087, 0.1052, 0.1093, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-02-28 23:59:12,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-02-28 23:59:15,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9245, 1.1070, 0.7896, 0.8439], device='cuda:1'), covar=tensor([0.0782, 0.0641, 0.1553, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0527, 0.0556, 0.0486], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-02-28 23:59:33,322 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49155.0, num_to_drop=1, layers_to_drop={1}
+2023-02-28 23:59:41,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49165.0, num_to_drop=0, layers_to_drop=set()
+2023-02-28 23:59:43,583 INFO [train.py:968] (1/2) Epoch 2, batch 3500, giga_loss[loss=0.3325, simple_loss=0.4015, pruned_loss=0.1317, over 28938.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4103, pruned_loss=0.1528, over 5727796.80 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4227, pruned_loss=0.1578, over 4820887.88 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4088, pruned_loss=0.1523, over 5712913.21 frames. ], batch size: 136, lr: 1.56e-02, grad_scale: 4.0
+2023-03-01 00:00:03,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:22,791 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:26,072 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 3550, giga_loss[loss=0.3806, simple_loss=0.4354, pruned_loss=0.1629, over 29025.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4095, pruned_loss=0.1508, over 5722947.94 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4232, pruned_loss=0.1582, over 4836378.08 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4078, pruned_loss=0.15, over 5709503.56 frames. ], batch size: 155, lr: 1.56e-02, grad_scale: 4.0
+2023-03-01 00:00:35,888 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:38,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:00:41,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:08,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:12,433 INFO [train.py:968] (1/2) Epoch 2, batch 3600, giga_loss[loss=0.3467, simple_loss=0.4028, pruned_loss=0.1453, over 28896.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4092, pruned_loss=0.1494, over 5726763.95 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4231, pruned_loss=0.1583, over 4860216.13 frames. ], giga_tot_loss[loss=0.3524, simple_loss=0.4076, pruned_loss=0.1486, over 5714442.62 frames. ], batch size: 145, lr: 1.56e-02, grad_scale: 8.0
+2023-03-01 00:01:49,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:51,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:01:57,684 INFO [optim.py:369] (1/2) 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 00:01:58,467 INFO [train.py:968] (1/2) Epoch 2, batch 3650, giga_loss[loss=0.3248, simple_loss=0.3864, pruned_loss=0.1316, over 29008.00 frames. ], tot_loss[loss=0.351, simple_loss=0.407, pruned_loss=0.1475, over 5728452.61 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4228, pruned_loss=0.158, over 4885970.41 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4056, pruned_loss=0.1469, over 5714667.88 frames. ], batch size: 136, lr: 1.56e-02, grad_scale: 8.0
+2023-03-01 00:02:08,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3453, 1.5556, 1.0100, 1.1838], device='cuda:1'), covar=tensor([0.0782, 0.0610, 0.1244, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0528, 0.0553, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 00:02:15,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:17,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:19,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:45,288 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:02:47,196 INFO [train.py:968] (1/2) Epoch 2, batch 3700, giga_loss[loss=0.3207, simple_loss=0.3893, pruned_loss=0.126, over 28951.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.4048, pruned_loss=0.1467, over 5724654.93 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4224, pruned_loss=0.1577, over 4911246.86 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.4035, pruned_loss=0.146, over 5717809.73 frames. ], batch size: 145, lr: 1.56e-02, grad_scale: 4.0
+2023-03-01 00:03:30,751 INFO [optim.py:369] (1/2) 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 00:03:30,764 INFO [train.py:968] (1/2) Epoch 2, batch 3750, giga_loss[loss=0.3222, simple_loss=0.38, pruned_loss=0.1322, over 28643.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.4011, pruned_loss=0.1443, over 5722620.41 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4221, pruned_loss=0.1576, over 4921314.25 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.4001, pruned_loss=0.1438, over 5715400.09 frames. ], batch size: 60, lr: 1.56e-02, grad_scale: 4.0
+2023-03-01 00:04:13,829 INFO [train.py:968] (1/2) Epoch 2, batch 3800, giga_loss[loss=0.3725, simple_loss=0.4214, pruned_loss=0.1618, over 28739.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.4026, pruned_loss=0.1453, over 5727106.56 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4226, pruned_loss=0.158, over 4945458.51 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.4008, pruned_loss=0.1442, over 5723896.03 frames. ], batch size: 262, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:04:53,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8992, 1.7860, 1.3128, 1.5194], device='cuda:1'), covar=tensor([0.0695, 0.0771, 0.1068, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0522, 0.0548, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 00:04:55,275 INFO [optim.py:369] (1/2) 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 00:04:55,288 INFO [train.py:968] (1/2) Epoch 2, batch 3850, giga_loss[loss=0.358, simple_loss=0.4062, pruned_loss=0.1549, over 28021.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.4024, pruned_loss=0.1453, over 5728184.64 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4221, pruned_loss=0.1577, over 4964848.66 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.4009, pruned_loss=0.1444, over 5722143.47 frames. ], batch size: 412, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:05:34,281 INFO [train.py:968] (1/2) Epoch 2, batch 3900, giga_loss[loss=0.3839, simple_loss=0.4286, pruned_loss=0.1696, over 28898.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.4027, pruned_loss=0.1449, over 5725422.31 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.422, pruned_loss=0.1576, over 4985660.81 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.4012, pruned_loss=0.1439, over 5718945.36 frames. ], batch size: 106, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:05:52,127 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:06:07,754 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 2, batch 3950, giga_loss[loss=0.327, simple_loss=0.3876, pruned_loss=0.1332, over 28756.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.4018, pruned_loss=0.1434, over 5715742.10 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.422, pruned_loss=0.1577, over 5005240.81 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.4001, pruned_loss=0.1421, over 5714862.69 frames. ], batch size: 99, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:06:26,096 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:06:57,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7961, 1.5635, 1.5678, 1.1618], device='cuda:1'), covar=tensor([0.0362, 0.0343, 0.0256, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0886, 0.0637, 0.0727, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 00:06:58,542 INFO [train.py:968] (1/2) Epoch 2, batch 4000, giga_loss[loss=0.339, simple_loss=0.4005, pruned_loss=0.1387, over 29034.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.4015, pruned_loss=0.1434, over 5719377.37 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4219, pruned_loss=0.1576, over 5009729.60 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.4, pruned_loss=0.1424, over 5717879.79 frames. ], batch size: 164, lr: 1.55e-02, grad_scale: 8.0
+2023-03-01 00:07:39,265 INFO [train.py:968] (1/2) Epoch 2, batch 4050, libri_loss[loss=0.3353, simple_loss=0.4047, pruned_loss=0.133, over 29500.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3999, pruned_loss=0.1436, over 5711758.32 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.421, pruned_loss=0.157, over 5032851.12 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3989, pruned_loss=0.1429, over 5708562.84 frames. ], batch size: 85, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:07:39,961 INFO [optim.py:369] (1/2) 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 00:07:45,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4331, 1.3303, 3.3925, 2.8015], device='cuda:1'), covar=tensor([0.1476, 0.1420, 0.0304, 0.0507], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0470, 0.0593, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:1')
+2023-03-01 00:07:49,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:07:51,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:03,447 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:05,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:16,417 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 4100, giga_loss[loss=0.2982, simple_loss=0.3637, pruned_loss=0.1163, over 28840.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.397, pruned_loss=0.142, over 5715527.53 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4208, pruned_loss=0.157, over 5046093.54 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.396, pruned_loss=0.1412, over 5710125.47 frames. ], batch size: 119, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:08:29,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:08:58,623 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 4150, giga_loss[loss=0.3746, simple_loss=0.423, pruned_loss=0.1631, over 28787.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3961, pruned_loss=0.1425, over 5699483.39 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4213, pruned_loss=0.1574, over 5060092.36 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3944, pruned_loss=0.1411, over 5700649.99 frames. ], batch size: 284, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:09:01,834 INFO [optim.py:369] (1/2) 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 00:09:05,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=49825.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:09:40,680 INFO [train.py:968] (1/2) Epoch 2, batch 4200, giga_loss[loss=0.3323, simple_loss=0.3789, pruned_loss=0.1429, over 28886.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3975, pruned_loss=0.1446, over 5702312.70 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4212, pruned_loss=0.1574, over 5079893.84 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3955, pruned_loss=0.1432, over 5699682.78 frames. ], batch size: 112, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:10:21,176 INFO [train.py:968] (1/2) Epoch 2, batch 4250, giga_loss[loss=0.3022, simple_loss=0.3645, pruned_loss=0.1199, over 28695.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.396, pruned_loss=0.1446, over 5713157.58 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.42, pruned_loss=0.157, over 5112578.95 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3946, pruned_loss=0.1434, over 5703671.34 frames. ], batch size: 92, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:10:21,829 INFO [optim.py:369] (1/2) 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 00:11:06,015 INFO [train.py:968] (1/2) Epoch 2, batch 4300, giga_loss[loss=0.3078, simple_loss=0.3633, pruned_loss=0.1261, over 28603.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3928, pruned_loss=0.1429, over 5715345.22 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4192, pruned_loss=0.1564, over 5135274.62 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3916, pruned_loss=0.142, over 5703020.24 frames. ], batch size: 85, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:11:34,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:11:48,358 INFO [train.py:968] (1/2) Epoch 2, batch 4350, giga_loss[loss=0.3302, simple_loss=0.3835, pruned_loss=0.1384, over 28866.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3903, pruned_loss=0.1424, over 5706664.76 frames. ], libri_tot_loss[loss=0.3663, simple_loss=0.4193, pruned_loss=0.1566, over 5133301.39 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.389, pruned_loss=0.1414, over 5703133.60 frames. ], batch size: 227, lr: 1.55e-02, grad_scale: 4.0
+2023-03-01 00:11:48,977 INFO [optim.py:369] (1/2) 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 00:12:06,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2833, 1.2735, 1.2213, 1.2000], device='cuda:1'), covar=tensor([0.0706, 0.0924, 0.1211, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0831, 0.0634, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 00:12:39,179 INFO [train.py:968] (1/2) Epoch 2, batch 4400, giga_loss[loss=0.3595, simple_loss=0.4066, pruned_loss=0.1562, over 29009.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.387, pruned_loss=0.14, over 5713277.14 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4197, pruned_loss=0.1568, over 5144568.80 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3853, pruned_loss=0.1389, over 5708080.00 frames. ], batch size: 164, lr: 1.55e-02, grad_scale: 8.0
+2023-03-01 00:13:25,805 INFO [train.py:968] (1/2) Epoch 2, batch 4450, giga_loss[loss=0.3734, simple_loss=0.4195, pruned_loss=0.1636, over 28494.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.386, pruned_loss=0.1389, over 5711838.34 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4194, pruned_loss=0.1569, over 5156778.66 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.384, pruned_loss=0.1376, over 5711176.70 frames. ], batch size: 336, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:13:27,243 INFO [optim.py:369] (1/2) 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 00:13:39,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1139, 1.1336, 1.1120, 0.9736], device='cuda:1'), covar=tensor([0.1730, 0.1826, 0.1588, 0.1709], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0766, 0.0850, 0.0901], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 00:13:53,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:13:55,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:03,716 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=50159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:14,520 INFO [train.py:968] (1/2) Epoch 2, batch 4500, libri_loss[loss=0.3546, simple_loss=0.4161, pruned_loss=0.1466, over 26010.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3903, pruned_loss=0.1413, over 5703866.75 frames. ], libri_tot_loss[loss=0.3672, simple_loss=0.4199, pruned_loss=0.1573, over 5171779.23 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3877, pruned_loss=0.1395, over 5702512.71 frames. ], batch size: 136, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:14:22,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:31,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:14:39,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50200.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:14:48,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-01 00:14:55,088 INFO [train.py:968] (1/2) Epoch 2, batch 4550, giga_loss[loss=0.3893, simple_loss=0.4387, pruned_loss=0.1699, over 28701.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3939, pruned_loss=0.1431, over 5703862.62 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4198, pruned_loss=0.1574, over 5186248.83 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3914, pruned_loss=0.1414, over 5699435.40 frames. ], batch size: 242, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:14:57,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 00:14:57,199 INFO [optim.py:369] (1/2) 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 00:15:22,951 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6133, 1.4660, 1.1003, 1.2656], device='cuda:1'), covar=tensor([0.0627, 0.0691, 0.0993, 0.1048], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0534, 0.0568, 0.0501], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 00:15:38,002 INFO [train.py:968] (1/2) Epoch 2, batch 4600, libri_loss[loss=0.3823, simple_loss=0.4372, pruned_loss=0.1637, over 29649.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3971, pruned_loss=0.1445, over 5696130.55 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4203, pruned_loss=0.1578, over 5200297.81 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3938, pruned_loss=0.1423, over 5695328.32 frames. ], batch size: 88, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:15:44,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.21 vs. limit=2.0
+2023-03-01 00:16:23,458 INFO [train.py:968] (1/2) Epoch 2, batch 4650, giga_loss[loss=0.3977, simple_loss=0.4225, pruned_loss=0.1864, over 26632.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3975, pruned_loss=0.1441, over 5691770.54 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4215, pruned_loss=0.1589, over 5217010.78 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3934, pruned_loss=0.1412, over 5687263.70 frames. ], batch size: 555, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:16:25,054 INFO [optim.py:369] (1/2) 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 00:16:36,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:16:39,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:16:44,235 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50343.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:16:46,415 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50346.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:17:04,553 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:17:07,614 INFO [train.py:968] (1/2) Epoch 2, batch 4700, giga_loss[loss=0.3474, simple_loss=0.3996, pruned_loss=0.1476, over 28948.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3962, pruned_loss=0.143, over 5702179.04 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4213, pruned_loss=0.1587, over 5223976.06 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.393, pruned_loss=0.1406, over 5696717.70 frames. ], batch size: 136, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:17:12,321 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50375.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:17:49,866 INFO [train.py:968] (1/2) Epoch 2, batch 4750, giga_loss[loss=0.4491, simple_loss=0.4776, pruned_loss=0.2104, over 27588.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3973, pruned_loss=0.1442, over 5700701.37 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4214, pruned_loss=0.1589, over 5233777.86 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3943, pruned_loss=0.142, over 5694080.18 frames. ], batch size: 472, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:17:51,241 INFO [optim.py:369] (1/2) 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 00:18:27,854 INFO [train.py:968] (1/2) Epoch 2, batch 4800, giga_loss[loss=0.3187, simple_loss=0.3874, pruned_loss=0.125, over 28965.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4018, pruned_loss=0.1485, over 5713933.33 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4224, pruned_loss=0.16, over 5279471.66 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.397, pruned_loss=0.1449, over 5695381.26 frames. ], batch size: 164, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:19:12,393 INFO [train.py:968] (1/2) Epoch 2, batch 4850, giga_loss[loss=0.3366, simple_loss=0.3925, pruned_loss=0.1404, over 28725.00 frames. ], tot_loss[loss=0.3518, simple_loss=0.404, pruned_loss=0.1498, over 5706905.28 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4224, pruned_loss=0.16, over 5294047.63 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3997, pruned_loss=0.1467, over 5688763.34 frames. ], batch size: 99, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:19:15,007 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:19:19,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1048, 1.2744, 1.1696, 1.0433], device='cuda:1'), covar=tensor([0.0687, 0.0804, 0.1076, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0810, 0.0633, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 00:19:24,112 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:19:53,597 INFO [train.py:968] (1/2) Epoch 2, batch 4900, giga_loss[loss=0.3791, simple_loss=0.4228, pruned_loss=0.1677, over 28549.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.407, pruned_loss=0.1509, over 5715124.77 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.423, pruned_loss=0.16, over 5318728.95 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4023, pruned_loss=0.148, over 5694495.51 frames. ], batch size: 78, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:20:13,071 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=50590.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:20:35,358 INFO [train.py:968] (1/2) Epoch 2, batch 4950, giga_loss[loss=0.3841, simple_loss=0.428, pruned_loss=0.17, over 28600.00 frames. ], tot_loss[loss=0.355, simple_loss=0.408, pruned_loss=0.151, over 5713851.52 frames. ], libri_tot_loss[loss=0.3716, simple_loss=0.423, pruned_loss=0.1601, over 5317054.57 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.404, pruned_loss=0.1484, over 5706379.22 frames. ], batch size: 60, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:20:37,869 INFO [optim.py:369] (1/2) 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 00:21:08,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6435, 2.9586, 2.1906, 0.8179], device='cuda:1'), covar=tensor([0.2166, 0.0919, 0.1173, 0.2302], device='cuda:1'), in_proj_covar=tensor([0.1107, 0.1089, 0.1116, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 00:21:15,441 INFO [train.py:968] (1/2) Epoch 2, batch 5000, giga_loss[loss=0.353, simple_loss=0.3956, pruned_loss=0.1552, over 23852.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4079, pruned_loss=0.1507, over 5714231.90 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4233, pruned_loss=0.1602, over 5336443.12 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.404, pruned_loss=0.1482, over 5702992.43 frames. ], batch size: 705, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:21:23,314 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:21:25,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:21:51,681 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:21:58,966 INFO [train.py:968] (1/2) Epoch 2, batch 5050, giga_loss[loss=0.3137, simple_loss=0.3802, pruned_loss=0.1236, over 28934.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.4051, pruned_loss=0.148, over 5721548.54 frames. ], libri_tot_loss[loss=0.3717, simple_loss=0.4232, pruned_loss=0.1601, over 5339297.91 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.402, pruned_loss=0.1461, over 5711864.35 frames. ], batch size: 227, lr: 1.54e-02, grad_scale: 4.0
+2023-03-01 00:22:00,876 INFO [optim.py:369] (1/2) 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 00:22:38,259 INFO [train.py:968] (1/2) Epoch 2, batch 5100, giga_loss[loss=0.2947, simple_loss=0.3615, pruned_loss=0.1139, over 28334.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.4053, pruned_loss=0.148, over 5731019.42 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4236, pruned_loss=0.1602, over 5361707.17 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.4018, pruned_loss=0.1459, over 5716420.21 frames. ], batch size: 65, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:22:42,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8139, 1.5115, 1.1329, 1.2452], device='cuda:1'), covar=tensor([0.0692, 0.0898, 0.1148, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0540, 0.0571, 0.0503], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 00:23:19,568 INFO [train.py:968] (1/2) Epoch 2, batch 5150, giga_loss[loss=0.3769, simple_loss=0.4199, pruned_loss=0.1669, over 29021.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4047, pruned_loss=0.1479, over 5717946.35 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.424, pruned_loss=0.1604, over 5367141.17 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.4009, pruned_loss=0.1457, over 5711778.76 frames. ], batch size: 136, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:23:21,415 INFO [optim.py:369] (1/2) 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 00:24:01,856 INFO [train.py:968] (1/2) Epoch 2, batch 5200, giga_loss[loss=0.2875, simple_loss=0.3534, pruned_loss=0.1108, over 28664.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3992, pruned_loss=0.1444, over 5718424.20 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4238, pruned_loss=0.1603, over 5368055.23 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3961, pruned_loss=0.1426, over 5718124.99 frames. ], batch size: 242, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:24:11,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3452, 1.9598, 1.6470, 0.4327], device='cuda:1'), covar=tensor([0.1311, 0.0785, 0.1330, 0.1715], device='cuda:1'), in_proj_covar=tensor([0.1133, 0.1108, 0.1130, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 00:24:24,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:24:43,822 INFO [train.py:968] (1/2) Epoch 2, batch 5250, giga_loss[loss=0.4163, simple_loss=0.4459, pruned_loss=0.1933, over 27623.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3958, pruned_loss=0.1421, over 5719888.68 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.424, pruned_loss=0.1604, over 5374225.86 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3929, pruned_loss=0.1404, over 5718832.84 frames. ], batch size: 472, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:24:48,464 INFO [optim.py:369] (1/2) 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 00:25:00,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=50939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:25:22,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:25:25,784 INFO [train.py:968] (1/2) Epoch 2, batch 5300, giga_loss[loss=0.3289, simple_loss=0.4042, pruned_loss=0.1268, over 28916.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3975, pruned_loss=0.1424, over 5701491.84 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4244, pruned_loss=0.1609, over 5371635.42 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3944, pruned_loss=0.1403, over 5708563.59 frames. ], batch size: 227, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:25:42,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 00:25:44,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6565, 1.4545, 3.8296, 2.9858], device='cuda:1'), covar=tensor([0.1485, 0.1475, 0.0323, 0.0591], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0474, 0.0614, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:26:06,426 INFO [train.py:968] (1/2) Epoch 2, batch 5350, giga_loss[loss=0.3384, simple_loss=0.3858, pruned_loss=0.1455, over 28497.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.4015, pruned_loss=0.1444, over 5706625.79 frames. ], libri_tot_loss[loss=0.3739, simple_loss=0.425, pruned_loss=0.1614, over 5391736.13 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3976, pruned_loss=0.1417, over 5705885.31 frames. ], batch size: 85, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:26:10,043 INFO [optim.py:369] (1/2) 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 00:26:22,720 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:26:25,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:26:37,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2387, 1.3739, 1.1018, 1.3729], device='cuda:1'), covar=tensor([0.1788, 0.1756, 0.1551, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0771, 0.0851, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 00:26:48,265 INFO [train.py:968] (1/2) Epoch 2, batch 5400, giga_loss[loss=0.328, simple_loss=0.3875, pruned_loss=0.1343, over 28930.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.4032, pruned_loss=0.1467, over 5704685.43 frames. ], libri_tot_loss[loss=0.3748, simple_loss=0.4256, pruned_loss=0.162, over 5402379.75 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3992, pruned_loss=0.1437, over 5700046.62 frames. ], batch size: 174, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:26:49,930 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:27:18,355 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51108.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:27:21,036 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 5450, giga_loss[loss=0.3861, simple_loss=0.4112, pruned_loss=0.1805, over 28621.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.4034, pruned_loss=0.1485, over 5711633.59 frames. ], libri_tot_loss[loss=0.3745, simple_loss=0.4254, pruned_loss=0.1618, over 5421247.29 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3993, pruned_loss=0.1456, over 5702682.26 frames. ], batch size: 78, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:27:28,924 INFO [optim.py:369] (1/2) 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 00:27:33,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7149, 2.9899, 1.6750, 1.5310], device='cuda:1'), covar=tensor([0.0879, 0.0529, 0.0907, 0.1408], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0440, 0.0342, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0011, 0.0014], device='cuda:1')
+2023-03-01 00:27:45,062 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51140.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:28:02,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8092, 2.2643, 1.9997, 1.8844], device='cuda:1'), covar=tensor([0.1262, 0.1339, 0.0999, 0.0665], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0811, 0.0710, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:28:10,858 INFO [train.py:968] (1/2) Epoch 2, batch 5500, giga_loss[loss=0.3068, simple_loss=0.3642, pruned_loss=0.1247, over 29107.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4012, pruned_loss=0.1493, over 5708944.42 frames. ], libri_tot_loss[loss=0.3743, simple_loss=0.4251, pruned_loss=0.1618, over 5428706.33 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3978, pruned_loss=0.1468, over 5698931.31 frames. ], batch size: 128, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:28:17,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:28:39,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51201.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:28:39,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2685, 1.2592, 1.3322, 0.9988], device='cuda:1'), covar=tensor([0.0575, 0.0411, 0.0263, 0.0396], device='cuda:1'), in_proj_covar=tensor([0.0982, 0.0686, 0.0780, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 00:28:54,511 INFO [train.py:968] (1/2) Epoch 2, batch 5550, giga_loss[loss=0.2866, simple_loss=0.3484, pruned_loss=0.1124, over 28557.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.3989, pruned_loss=0.1492, over 5707680.80 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4253, pruned_loss=0.1621, over 5434955.40 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3957, pruned_loss=0.1468, over 5698010.48 frames. ], batch size: 60, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:28:57,194 INFO [optim.py:369] (1/2) 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 00:29:39,255 INFO [train.py:968] (1/2) Epoch 2, batch 5600, giga_loss[loss=0.3223, simple_loss=0.3673, pruned_loss=0.1386, over 28678.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3954, pruned_loss=0.1468, over 5707698.83 frames. ], libri_tot_loss[loss=0.3752, simple_loss=0.4257, pruned_loss=0.1623, over 5436141.76 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3923, pruned_loss=0.1446, over 5701487.76 frames. ], batch size: 92, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:29:49,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-01 00:29:58,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1161, 1.2118, 1.0520, 0.8509], device='cuda:1'), covar=tensor([0.0411, 0.0310, 0.0269, 0.0352], device='cuda:1'), in_proj_covar=tensor([0.0974, 0.0679, 0.0783, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 00:30:11,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4706, 2.1751, 1.7075, 1.7727], device='cuda:1'), covar=tensor([0.0485, 0.0642, 0.0784, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0531, 0.0556, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 00:30:17,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:30:22,903 INFO [train.py:968] (1/2) Epoch 2, batch 5650, giga_loss[loss=0.33, simple_loss=0.3828, pruned_loss=0.1386, over 28685.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3938, pruned_loss=0.1455, over 5712240.81 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4257, pruned_loss=0.1622, over 5442753.13 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3906, pruned_loss=0.1435, over 5707162.93 frames. ], batch size: 262, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:30:25,375 INFO [optim.py:369] (1/2) 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 00:30:38,258 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:31:00,013 INFO [train.py:968] (1/2) Epoch 2, batch 5700, giga_loss[loss=0.3051, simple_loss=0.3484, pruned_loss=0.1309, over 28844.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3882, pruned_loss=0.1415, over 5710949.65 frames. ], libri_tot_loss[loss=0.3749, simple_loss=0.4258, pruned_loss=0.1621, over 5444804.17 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3845, pruned_loss=0.1394, over 5713932.72 frames. ], batch size: 99, lr: 1.53e-02, grad_scale: 8.0
+2023-03-01 00:31:34,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-01 00:31:43,418 INFO [train.py:968] (1/2) Epoch 2, batch 5750, giga_loss[loss=0.2778, simple_loss=0.3488, pruned_loss=0.1034, over 28242.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3828, pruned_loss=0.1381, over 5713519.78 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4261, pruned_loss=0.1623, over 5447634.35 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3792, pruned_loss=0.136, over 5715643.48 frames. ], batch size: 77, lr: 1.53e-02, grad_scale: 4.0
+2023-03-01 00:31:44,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9823, 1.9150, 1.5868, 1.7327], device='cuda:1'), covar=tensor([0.1611, 0.2141, 0.1591, 0.0939], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0810, 0.0711, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:31:47,180 INFO [optim.py:369] (1/2) 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 00:32:12,285 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:32:13,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:32:15,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:32:19,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 00:32:21,483 INFO [train.py:968] (1/2) Epoch 2, batch 5800, giga_loss[loss=0.3747, simple_loss=0.4162, pruned_loss=0.1666, over 27854.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3851, pruned_loss=0.1397, over 5715017.81 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.4263, pruned_loss=0.1626, over 5457946.01 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3809, pruned_loss=0.1371, over 5714616.79 frames. ], batch size: 412, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:32:38,562 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:33:03,673 INFO [train.py:968] (1/2) Epoch 2, batch 5850, giga_loss[loss=0.3409, simple_loss=0.3958, pruned_loss=0.143, over 28979.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3889, pruned_loss=0.1416, over 5716295.23 frames. ], libri_tot_loss[loss=0.3753, simple_loss=0.4258, pruned_loss=0.1624, over 5461266.30 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3856, pruned_loss=0.1396, over 5715279.92 frames. ], batch size: 136, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:33:07,670 INFO [optim.py:369] (1/2) 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 00:33:21,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4816, 1.2835, 1.3038, 1.5865], device='cuda:1'), covar=tensor([0.1687, 0.1820, 0.1451, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0780, 0.0845, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 00:33:28,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:33:39,015 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51565.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:33:41,427 INFO [train.py:968] (1/2) Epoch 2, batch 5900, giga_loss[loss=0.3703, simple_loss=0.4197, pruned_loss=0.1605, over 28636.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3928, pruned_loss=0.1428, over 5725321.16 frames. ], libri_tot_loss[loss=0.3748, simple_loss=0.4257, pruned_loss=0.1619, over 5480350.26 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3888, pruned_loss=0.1407, over 5717458.82 frames. ], batch size: 262, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:33:47,247 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51576.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:33:50,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4845, 3.5663, 4.2164, 1.9559], device='cuda:1'), covar=tensor([0.0351, 0.0354, 0.0634, 0.1479], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0500, 0.0774, 0.0529], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 00:34:23,614 INFO [train.py:968] (1/2) Epoch 2, batch 5950, giga_loss[loss=0.4659, simple_loss=0.4878, pruned_loss=0.222, over 27925.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3973, pruned_loss=0.1456, over 5715140.73 frames. ], libri_tot_loss[loss=0.3746, simple_loss=0.4254, pruned_loss=0.1619, over 5481546.73 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3937, pruned_loss=0.1436, over 5713246.63 frames. ], batch size: 412, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:34:28,407 INFO [optim.py:369] (1/2) 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 00:35:01,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6141, 1.5730, 1.1925, 1.3270], device='cuda:1'), covar=tensor([0.0781, 0.0831, 0.1204, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0541, 0.0564, 0.0504], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 00:35:05,342 INFO [train.py:968] (1/2) Epoch 2, batch 6000, libri_loss[loss=0.4257, simple_loss=0.4629, pruned_loss=0.1942, over 29469.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4013, pruned_loss=0.1479, over 5715605.04 frames. ], libri_tot_loss[loss=0.3755, simple_loss=0.4262, pruned_loss=0.1624, over 5494085.99 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3968, pruned_loss=0.1453, over 5710020.30 frames. ], batch size: 85, lr: 1.52e-02, grad_scale: 8.0
+2023-03-01 00:35:05,342 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 00:35:09,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7981, 3.1431, 3.4954, 1.5540], device='cuda:1'), covar=tensor([0.0605, 0.0552, 0.0778, 0.2166], device='cuda:1'), in_proj_covar=tensor([0.0744, 0.0499, 0.0778, 0.0530], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 00:35:14,524 INFO [train.py:1012] (1/2) Epoch 2, validation: loss=0.288, simple_loss=0.3799, pruned_loss=0.09807, over 944034.00 frames. 
+2023-03-01 00:35:14,524 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 00:35:23,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8898, 1.0655, 0.9722, 0.6233], device='cuda:1'), covar=tensor([0.0379, 0.0348, 0.0281, 0.0379], device='cuda:1'), in_proj_covar=tensor([0.0992, 0.0697, 0.0802, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 00:35:35,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:35:39,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:35:47,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3341, 1.7815, 1.3470, 1.4835], device='cuda:1'), covar=tensor([0.0895, 0.0371, 0.0468, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0226, 0.0232, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0020, 0.0018, 0.0029], device='cuda:1')
+2023-03-01 00:35:55,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:35:59,660 INFO [train.py:968] (1/2) Epoch 2, batch 6050, giga_loss[loss=0.3224, simple_loss=0.3885, pruned_loss=0.1282, over 28215.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4047, pruned_loss=0.151, over 5712290.08 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4264, pruned_loss=0.1627, over 5500551.62 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.4007, pruned_loss=0.1485, over 5705109.92 frames. ], batch size: 368, lr: 1.52e-02, grad_scale: 8.0
+2023-03-01 00:35:59,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51719.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:36:04,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51722.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:36:05,798 INFO [optim.py:369] (1/2) 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 00:36:06,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:36:29,686 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51751.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:36:29,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8303, 2.2871, 1.8369, 1.8172], device='cuda:1'), covar=tensor([0.1130, 0.1235, 0.0950, 0.0657], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0798, 0.0692, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:36:47,723 INFO [train.py:968] (1/2) Epoch 2, batch 6100, giga_loss[loss=0.3884, simple_loss=0.4295, pruned_loss=0.1736, over 29081.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4135, pruned_loss=0.1597, over 5709439.73 frames. ], libri_tot_loss[loss=0.3757, simple_loss=0.4263, pruned_loss=0.1625, over 5510314.88 frames. ], giga_tot_loss[loss=0.3627, simple_loss=0.41, pruned_loss=0.1577, over 5699715.67 frames. ], batch size: 128, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:36:52,740 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:37:03,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:37:36,941 INFO [train.py:968] (1/2) Epoch 2, batch 6150, libri_loss[loss=0.3455, simple_loss=0.4045, pruned_loss=0.1433, over 29536.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.4192, pruned_loss=0.1651, over 5699533.42 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4257, pruned_loss=0.1621, over 5520259.06 frames. ], giga_tot_loss[loss=0.3722, simple_loss=0.4167, pruned_loss=0.1639, over 5686927.26 frames. ], batch size: 79, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:37:41,767 INFO [optim.py:369] (1/2) 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 00:37:46,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5258, 1.6661, 1.1827, 1.1263], device='cuda:1'), covar=tensor([0.0442, 0.0388, 0.0430, 0.0427], device='cuda:1'), in_proj_covar=tensor([0.0978, 0.0685, 0.0780, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 00:37:48,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:37:58,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2008, 1.2315, 1.1971, 1.1831], device='cuda:1'), covar=tensor([0.0660, 0.0777, 0.1044, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0823, 0.0640, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 00:38:13,386 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:38:15,651 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 6200, giga_loss[loss=0.4985, simple_loss=0.5032, pruned_loss=0.2469, over 28621.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4267, pruned_loss=0.1717, over 5682339.02 frames. ], libri_tot_loss[loss=0.3752, simple_loss=0.4259, pruned_loss=0.1623, over 5519365.95 frames. ], giga_tot_loss[loss=0.383, simple_loss=0.4246, pruned_loss=0.1707, over 5675942.45 frames. ], batch size: 336, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:38:44,599 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=51888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:38:45,168 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 6250, giga_loss[loss=0.3982, simple_loss=0.4354, pruned_loss=0.1805, over 28888.00 frames. ], tot_loss[loss=0.3903, simple_loss=0.4301, pruned_loss=0.1753, over 5682895.57 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4256, pruned_loss=0.1622, over 5537585.23 frames. ], giga_tot_loss[loss=0.3895, simple_loss=0.4288, pruned_loss=0.1752, over 5668106.67 frames. ], batch size: 186, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:39:14,777 INFO [optim.py:369] (1/2) 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 00:39:30,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51940.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 00:39:57,886 INFO [train.py:968] (1/2) Epoch 2, batch 6300, giga_loss[loss=0.3839, simple_loss=0.4284, pruned_loss=0.1697, over 28779.00 frames. ], tot_loss[loss=0.4, simple_loss=0.4369, pruned_loss=0.1816, over 5686984.66 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4255, pruned_loss=0.162, over 5542311.95 frames. ], giga_tot_loss[loss=0.4001, simple_loss=0.4361, pruned_loss=0.182, over 5674280.87 frames. ], batch size: 119, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:40:01,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:40:04,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:40:34,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:40:48,377 INFO [train.py:968] (1/2) Epoch 2, batch 6350, giga_loss[loss=0.4245, simple_loss=0.4543, pruned_loss=0.1974, over 28826.00 frames. ], tot_loss[loss=0.4074, simple_loss=0.4417, pruned_loss=0.1866, over 5663988.92 frames. ], libri_tot_loss[loss=0.3749, simple_loss=0.4256, pruned_loss=0.1621, over 5551015.05 frames. ], giga_tot_loss[loss=0.4082, simple_loss=0.4414, pruned_loss=0.1875, over 5649809.15 frames. ], batch size: 199, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:40:53,257 INFO [optim.py:369] (1/2) 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 00:41:39,556 INFO [train.py:968] (1/2) Epoch 2, batch 6400, giga_loss[loss=0.3508, simple_loss=0.4097, pruned_loss=0.1459, over 28680.00 frames. ], tot_loss[loss=0.4114, simple_loss=0.4439, pruned_loss=0.1894, over 5663972.71 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4255, pruned_loss=0.1619, over 5557922.91 frames. ], giga_tot_loss[loss=0.4129, simple_loss=0.4441, pruned_loss=0.1908, over 5648253.95 frames. ], batch size: 71, lr: 1.52e-02, grad_scale: 8.0
+2023-03-01 00:41:54,157 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:41:57,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52086.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:42:28,470 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52115.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:42:31,879 INFO [train.py:968] (1/2) Epoch 2, batch 6450, giga_loss[loss=0.4572, simple_loss=0.4699, pruned_loss=0.2223, over 28301.00 frames. ], tot_loss[loss=0.4175, simple_loss=0.4471, pruned_loss=0.194, over 5645473.05 frames. ], libri_tot_loss[loss=0.3739, simple_loss=0.425, pruned_loss=0.1614, over 5568111.26 frames. ], giga_tot_loss[loss=0.4206, simple_loss=0.4483, pruned_loss=0.1964, over 5625921.68 frames. ], batch size: 368, lr: 1.52e-02, grad_scale: 4.0
+2023-03-01 00:42:37,455 INFO [optim.py:369] (1/2) 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 00:42:53,309 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:43:04,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:43:10,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4642, 1.2797, 1.1996, 1.6981], device='cuda:1'), covar=tensor([0.1674, 0.1849, 0.1465, 0.1971], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0800, 0.0869, 0.0931], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 00:43:14,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:43:26,179 INFO [train.py:968] (1/2) Epoch 2, batch 6500, giga_loss[loss=0.4179, simple_loss=0.4607, pruned_loss=0.1875, over 28745.00 frames. ], tot_loss[loss=0.4241, simple_loss=0.4514, pruned_loss=0.1984, over 5622984.84 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4248, pruned_loss=0.161, over 5564995.63 frames. ], giga_tot_loss[loss=0.4279, simple_loss=0.4531, pruned_loss=0.2014, over 5612090.59 frames. ], batch size: 99, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:44:17,371 INFO [train.py:968] (1/2) Epoch 2, batch 6550, giga_loss[loss=0.418, simple_loss=0.4491, pruned_loss=0.1935, over 28891.00 frames. ], tot_loss[loss=0.4273, simple_loss=0.4539, pruned_loss=0.2003, over 5622135.53 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4247, pruned_loss=0.1609, over 5571811.97 frames. ], giga_tot_loss[loss=0.4315, simple_loss=0.4558, pruned_loss=0.2036, over 5608337.52 frames. ], batch size: 145, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:44:23,864 INFO [optim.py:369] (1/2) 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 00:45:00,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:06,600 INFO [train.py:968] (1/2) Epoch 2, batch 6600, giga_loss[loss=0.4033, simple_loss=0.4391, pruned_loss=0.1838, over 29103.00 frames. ], tot_loss[loss=0.4237, simple_loss=0.4511, pruned_loss=0.1982, over 5639905.22 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4238, pruned_loss=0.1603, over 5577995.54 frames. ], giga_tot_loss[loss=0.4295, simple_loss=0.4542, pruned_loss=0.2023, over 5624782.83 frames. ], batch size: 155, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:45:12,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8248, 2.1649, 1.8761, 1.7788], device='cuda:1'), covar=tensor([0.1161, 0.1458, 0.1008, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0823, 0.0702, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:45:30,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:34,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:39,856 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:41,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:45:55,461 INFO [train.py:968] (1/2) Epoch 2, batch 6650, giga_loss[loss=0.5569, simple_loss=0.5326, pruned_loss=0.2906, over 28307.00 frames. ], tot_loss[loss=0.4226, simple_loss=0.4498, pruned_loss=0.1977, over 5639933.55 frames. ], libri_tot_loss[loss=0.3721, simple_loss=0.4237, pruned_loss=0.1602, over 5585977.77 frames. ], giga_tot_loss[loss=0.4284, simple_loss=0.4529, pruned_loss=0.2019, over 5622306.48 frames. ], batch size: 368, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:46:00,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7660, 1.7721, 4.0629, 3.0947], device='cuda:1'), covar=tensor([0.1571, 0.1353, 0.0302, 0.0500], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0486, 0.0638, 0.0505], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:46:02,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:46:03,220 INFO [optim.py:369] (1/2) 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 00:46:14,136 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:46:14,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6170, 1.4562, 1.0739, 1.2637], device='cuda:1'), covar=tensor([0.0608, 0.0657, 0.0986, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0551, 0.0569, 0.0516], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 00:46:47,417 INFO [train.py:968] (1/2) Epoch 2, batch 6700, giga_loss[loss=0.3746, simple_loss=0.4235, pruned_loss=0.1629, over 28883.00 frames. ], tot_loss[loss=0.4212, simple_loss=0.4496, pruned_loss=0.1963, over 5644355.39 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4236, pruned_loss=0.1601, over 5590830.56 frames. ], giga_tot_loss[loss=0.4266, simple_loss=0.4526, pruned_loss=0.2003, over 5626769.40 frames. ], batch size: 227, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:46:48,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 00:47:20,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:47:24,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:47:33,164 INFO [train.py:968] (1/2) Epoch 2, batch 6750, giga_loss[loss=0.4731, simple_loss=0.4601, pruned_loss=0.243, over 23661.00 frames. ], tot_loss[loss=0.4192, simple_loss=0.449, pruned_loss=0.1947, over 5643754.04 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4229, pruned_loss=0.1598, over 5592430.78 frames. ], giga_tot_loss[loss=0.4265, simple_loss=0.4532, pruned_loss=0.1999, over 5630859.52 frames. ], batch size: 705, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:47:40,516 INFO [optim.py:369] (1/2) 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 00:47:45,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=2.19 vs. limit=2.0
+2023-03-01 00:47:47,610 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:47:54,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:48:24,677 INFO [train.py:968] (1/2) Epoch 2, batch 6800, giga_loss[loss=0.3978, simple_loss=0.4392, pruned_loss=0.1782, over 28776.00 frames. ], tot_loss[loss=0.4189, simple_loss=0.4491, pruned_loss=0.1943, over 5627015.37 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4226, pruned_loss=0.1594, over 5595683.09 frames. ], giga_tot_loss[loss=0.426, simple_loss=0.4533, pruned_loss=0.1994, over 5614594.43 frames. ], batch size: 119, lr: 1.51e-02, grad_scale: 8.0
+2023-03-01 00:49:04,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4734, 1.9112, 1.4650, 0.5961], device='cuda:1'), covar=tensor([0.0984, 0.0866, 0.1140, 0.1439], device='cuda:1'), in_proj_covar=tensor([0.1124, 0.1140, 0.1126, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 00:49:10,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:49:14,584 INFO [train.py:968] (1/2) Epoch 2, batch 6850, giga_loss[loss=0.5204, simple_loss=0.5015, pruned_loss=0.2696, over 26452.00 frames. ], tot_loss[loss=0.4154, simple_loss=0.4466, pruned_loss=0.1921, over 5617759.64 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4226, pruned_loss=0.1594, over 5594540.24 frames. ], giga_tot_loss[loss=0.4222, simple_loss=0.4505, pruned_loss=0.1969, over 5609002.69 frames. ], batch size: 555, lr: 1.51e-02, grad_scale: 8.0
+2023-03-01 00:49:26,865 INFO [optim.py:369] (1/2) 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 00:50:08,608 INFO [train.py:968] (1/2) Epoch 2, batch 6900, giga_loss[loss=0.3476, simple_loss=0.408, pruned_loss=0.1436, over 28615.00 frames. ], tot_loss[loss=0.4103, simple_loss=0.4441, pruned_loss=0.1883, over 5624967.89 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4225, pruned_loss=0.1594, over 5600080.84 frames. ], giga_tot_loss[loss=0.4164, simple_loss=0.4476, pruned_loss=0.1926, over 5613520.67 frames. ], batch size: 85, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:50:27,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:50:48,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6169, 2.2725, 1.5846, 1.2741], device='cuda:1'), covar=tensor([0.0887, 0.0645, 0.0875, 0.1487], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0449, 0.0346, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0011, 0.0014], device='cuda:1')
+2023-03-01 00:51:01,310 INFO [train.py:968] (1/2) Epoch 2, batch 6950, giga_loss[loss=0.49, simple_loss=0.4834, pruned_loss=0.2483, over 26537.00 frames. ], tot_loss[loss=0.4055, simple_loss=0.441, pruned_loss=0.185, over 5640565.81 frames. ], libri_tot_loss[loss=0.3704, simple_loss=0.4223, pruned_loss=0.1592, over 5605940.00 frames. ], giga_tot_loss[loss=0.4113, simple_loss=0.4443, pruned_loss=0.1891, over 5626770.94 frames. ], batch size: 555, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:51:08,212 INFO [optim.py:369] (1/2) 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 00:51:39,529 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:51:44,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:51:51,937 INFO [train.py:968] (1/2) Epoch 2, batch 7000, libri_loss[loss=0.3221, simple_loss=0.3899, pruned_loss=0.1272, over 29664.00 frames. ], tot_loss[loss=0.4009, simple_loss=0.4377, pruned_loss=0.1821, over 5645915.58 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4216, pruned_loss=0.1587, over 5609963.25 frames. ], giga_tot_loss[loss=0.4068, simple_loss=0.4413, pruned_loss=0.1862, over 5631869.01 frames. ], batch size: 88, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:52:11,273 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:52:40,745 INFO [train.py:968] (1/2) Epoch 2, batch 7050, giga_loss[loss=0.3925, simple_loss=0.4262, pruned_loss=0.1793, over 28617.00 frames. ], tot_loss[loss=0.3972, simple_loss=0.4348, pruned_loss=0.1798, over 5652958.59 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4215, pruned_loss=0.1586, over 5613634.46 frames. ], giga_tot_loss[loss=0.4025, simple_loss=0.438, pruned_loss=0.1835, over 5639325.63 frames. ], batch size: 92, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:52:47,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 00:52:48,932 INFO [optim.py:369] (1/2) 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 00:53:30,126 INFO [train.py:968] (1/2) Epoch 2, batch 7100, giga_loss[loss=0.3789, simple_loss=0.4232, pruned_loss=0.1673, over 28625.00 frames. ], tot_loss[loss=0.3958, simple_loss=0.4342, pruned_loss=0.1787, over 5658836.81 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4214, pruned_loss=0.1585, over 5616814.89 frames. ], giga_tot_loss[loss=0.4007, simple_loss=0.4371, pruned_loss=0.1821, over 5645883.01 frames. ], batch size: 85, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:53:33,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8862, 1.5498, 1.5616, 1.4132], device='cuda:1'), covar=tensor([0.0721, 0.1569, 0.1030, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0832, 0.0641, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 00:53:50,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6961, 2.0815, 1.4885, 1.6430], device='cuda:1'), covar=tensor([0.1004, 0.0371, 0.0513, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0230, 0.0230, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0021, 0.0018, 0.0030], device='cuda:1')
+2023-03-01 00:54:04,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9586, 1.0776, 0.9852, 0.4114], device='cuda:1'), covar=tensor([0.0391, 0.0353, 0.0314, 0.0476], device='cuda:1'), in_proj_covar=tensor([0.0951, 0.0688, 0.0773, 0.0784], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 00:54:09,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:54:22,980 INFO [train.py:968] (1/2) Epoch 2, batch 7150, giga_loss[loss=0.4186, simple_loss=0.432, pruned_loss=0.2026, over 26562.00 frames. ], tot_loss[loss=0.3936, simple_loss=0.4332, pruned_loss=0.177, over 5653681.76 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4219, pruned_loss=0.159, over 5612476.21 frames. ], giga_tot_loss[loss=0.398, simple_loss=0.4356, pruned_loss=0.1802, over 5648493.30 frames. ], batch size: 555, lr: 1.51e-02, grad_scale: 4.0
+2023-03-01 00:54:31,460 INFO [optim.py:369] (1/2) 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 00:55:00,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4516, 1.3242, 1.2728, 1.4661], device='cuda:1'), covar=tensor([0.1784, 0.1984, 0.1722, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.0941, 0.0805, 0.0879, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 00:55:15,676 INFO [train.py:968] (1/2) Epoch 2, batch 7200, giga_loss[loss=0.4046, simple_loss=0.4591, pruned_loss=0.1751, over 28873.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4312, pruned_loss=0.1737, over 5668104.02 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4221, pruned_loss=0.1591, over 5616365.28 frames. ], giga_tot_loss[loss=0.3928, simple_loss=0.4331, pruned_loss=0.1763, over 5661220.76 frames. ], batch size: 174, lr: 1.50e-02, grad_scale: 8.0
+2023-03-01 00:55:44,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4394, 1.8595, 1.6373, 1.6380], device='cuda:1'), covar=tensor([0.1443, 0.1716, 0.1240, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0824, 0.0717, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:55:49,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-01 00:56:08,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8961, 2.4860, 2.3393, 1.8112], device='cuda:1'), covar=tensor([0.1038, 0.0358, 0.0386, 0.1209], device='cuda:1'), in_proj_covar=tensor([0.0331, 0.0230, 0.0229, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0021, 0.0018, 0.0030], device='cuda:1')
+2023-03-01 00:56:12,210 INFO [train.py:968] (1/2) Epoch 2, batch 7250, libri_loss[loss=0.3432, simple_loss=0.4045, pruned_loss=0.1409, over 27903.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4318, pruned_loss=0.1724, over 5661641.14 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4211, pruned_loss=0.1584, over 5621519.32 frames. ], giga_tot_loss[loss=0.3927, simple_loss=0.4344, pruned_loss=0.1755, over 5653276.02 frames. ], batch size: 116, lr: 1.50e-02, grad_scale: 8.0
+2023-03-01 00:56:18,049 INFO [optim.py:369] (1/2) 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 00:56:37,940 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=52948.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 00:56:40,509 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:56:43,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:56:51,926 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 7300, giga_loss[loss=0.486, simple_loss=0.4808, pruned_loss=0.2456, over 26689.00 frames. ], tot_loss[loss=0.3908, simple_loss=0.4339, pruned_loss=0.1739, over 5661604.17 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4208, pruned_loss=0.1582, over 5629779.95 frames. ], giga_tot_loss[loss=0.3955, simple_loss=0.4367, pruned_loss=0.1772, over 5648831.03 frames. ], batch size: 555, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:57:18,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:57:40,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6276, 1.3815, 1.0882, 1.1464], device='cuda:1'), covar=tensor([0.0544, 0.0614, 0.0964, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0539, 0.0551, 0.0494], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 00:57:53,243 INFO [train.py:968] (1/2) Epoch 2, batch 7350, libri_loss[loss=0.3622, simple_loss=0.4278, pruned_loss=0.1483, over 29762.00 frames. ], tot_loss[loss=0.3898, simple_loss=0.433, pruned_loss=0.1733, over 5681733.85 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4204, pruned_loss=0.1578, over 5636821.85 frames. ], giga_tot_loss[loss=0.3945, simple_loss=0.4359, pruned_loss=0.1766, over 5666160.77 frames. ], batch size: 87, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:58:03,843 INFO [optim.py:369] (1/2) 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 00:58:41,696 INFO [train.py:968] (1/2) Epoch 2, batch 7400, giga_loss[loss=0.3642, simple_loss=0.41, pruned_loss=0.1592, over 28948.00 frames. ], tot_loss[loss=0.388, simple_loss=0.431, pruned_loss=0.1725, over 5680312.14 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.42, pruned_loss=0.1574, over 5641197.54 frames. ], giga_tot_loss[loss=0.3926, simple_loss=0.4339, pruned_loss=0.1757, over 5664975.09 frames. ], batch size: 112, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:58:56,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:59:21,804 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:59:34,547 INFO [train.py:968] (1/2) Epoch 2, batch 7450, giga_loss[loss=0.3728, simple_loss=0.4074, pruned_loss=0.169, over 27958.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4308, pruned_loss=0.1745, over 5669940.41 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4202, pruned_loss=0.1575, over 5645021.73 frames. ], giga_tot_loss[loss=0.394, simple_loss=0.4331, pruned_loss=0.1775, over 5654597.41 frames. ], batch size: 412, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 00:59:43,637 INFO [optim.py:369] (1/2) 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 00:59:46,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2906, 1.7753, 5.2800, 3.6294], device='cuda:1'), covar=tensor([0.1499, 0.1519, 0.0255, 0.0419], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0484, 0.0626, 0.0508], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 00:59:53,162 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 00:59:57,290 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.70 vs. limit=5.0
+2023-03-01 01:00:00,247 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 7500, giga_loss[loss=0.346, simple_loss=0.3979, pruned_loss=0.147, over 28704.00 frames. ], tot_loss[loss=0.3877, simple_loss=0.4291, pruned_loss=0.1732, over 5686058.45 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4201, pruned_loss=0.1574, over 5652867.65 frames. ], giga_tot_loss[loss=0.3918, simple_loss=0.4313, pruned_loss=0.1761, over 5667429.33 frames. ], batch size: 71, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:00:48,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8676, 1.6560, 1.3107, 1.4130], device='cuda:1'), covar=tensor([0.0626, 0.0736, 0.1021, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0543, 0.0564, 0.0501], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 01:01:18,990 INFO [train.py:968] (1/2) Epoch 2, batch 7550, giga_loss[loss=0.4459, simple_loss=0.4631, pruned_loss=0.2143, over 28266.00 frames. ], tot_loss[loss=0.3862, simple_loss=0.4292, pruned_loss=0.1716, over 5697851.30 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4204, pruned_loss=0.1575, over 5656844.74 frames. ], giga_tot_loss[loss=0.3895, simple_loss=0.4308, pruned_loss=0.174, over 5680075.49 frames. ], batch size: 368, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:01:22,238 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:01:24,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:01:26,650 INFO [optim.py:369] (1/2) 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 01:01:33,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3489, 1.7241, 1.3785, 0.3306], device='cuda:1'), covar=tensor([0.1069, 0.0885, 0.1345, 0.1615], device='cuda:1'), in_proj_covar=tensor([0.1136, 0.1138, 0.1135, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 01:01:55,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:02:06,866 INFO [train.py:968] (1/2) Epoch 2, batch 7600, giga_loss[loss=0.4046, simple_loss=0.4372, pruned_loss=0.186, over 28371.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4278, pruned_loss=0.1693, over 5704930.30 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4199, pruned_loss=0.1573, over 5661796.20 frames. ], giga_tot_loss[loss=0.3864, simple_loss=0.4296, pruned_loss=0.1716, over 5687137.25 frames. ], batch size: 368, lr: 1.50e-02, grad_scale: 8.0
+2023-03-01 01:02:53,286 INFO [train.py:968] (1/2) Epoch 2, batch 7650, giga_loss[loss=0.451, simple_loss=0.4684, pruned_loss=0.2168, over 28754.00 frames. ], tot_loss[loss=0.3845, simple_loss=0.4282, pruned_loss=0.1704, over 5700232.36 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4196, pruned_loss=0.1572, over 5665531.43 frames. ], giga_tot_loss[loss=0.388, simple_loss=0.4303, pruned_loss=0.1728, over 5683698.56 frames. ], batch size: 242, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:02:58,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53323.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:03:02,890 INFO [optim.py:369] (1/2) 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 01:03:39,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-01 01:03:45,272 INFO [train.py:968] (1/2) Epoch 2, batch 7700, giga_loss[loss=0.3985, simple_loss=0.4354, pruned_loss=0.1808, over 28561.00 frames. ], tot_loss[loss=0.3843, simple_loss=0.4273, pruned_loss=0.1707, over 5700287.94 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4195, pruned_loss=0.1571, over 5666758.78 frames. ], giga_tot_loss[loss=0.3873, simple_loss=0.4291, pruned_loss=0.1727, over 5686347.02 frames. ], batch size: 307, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:04:28,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8410, 1.4894, 1.2972, 1.3784], device='cuda:1'), covar=tensor([0.0608, 0.0827, 0.0931, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0545, 0.0568, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 01:04:38,076 INFO [train.py:968] (1/2) Epoch 2, batch 7750, giga_loss[loss=0.4299, simple_loss=0.4524, pruned_loss=0.2037, over 28311.00 frames. ], tot_loss[loss=0.3852, simple_loss=0.4272, pruned_loss=0.1716, over 5687990.28 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4196, pruned_loss=0.1569, over 5669397.20 frames. ], giga_tot_loss[loss=0.3881, simple_loss=0.4287, pruned_loss=0.1737, over 5674868.55 frames. ], batch size: 368, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:04:47,605 INFO [optim.py:369] (1/2) 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 01:05:05,652 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:05:09,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7202, 2.1030, 1.8221, 1.7567], device='cuda:1'), covar=tensor([0.1255, 0.1400, 0.1016, 0.0664], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0830, 0.0709, 0.0595], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 01:05:24,351 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53466.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:05:25,873 INFO [train.py:968] (1/2) Epoch 2, batch 7800, giga_loss[loss=0.3652, simple_loss=0.4145, pruned_loss=0.1579, over 28850.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4266, pruned_loss=0.1715, over 5686445.92 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4196, pruned_loss=0.1569, over 5664225.13 frames. ], giga_tot_loss[loss=0.3877, simple_loss=0.428, pruned_loss=0.1736, over 5681913.99 frames. ], batch size: 285, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:05:26,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53469.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:05:54,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53498.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 01:06:14,861 INFO [train.py:968] (1/2) Epoch 2, batch 7850, giga_loss[loss=0.4263, simple_loss=0.4329, pruned_loss=0.2099, over 23795.00 frames. ], tot_loss[loss=0.3835, simple_loss=0.4253, pruned_loss=0.1708, over 5695497.55 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4196, pruned_loss=0.1567, over 5669255.99 frames. ], giga_tot_loss[loss=0.3864, simple_loss=0.4266, pruned_loss=0.1731, over 5687980.38 frames. ], batch size: 705, lr: 1.50e-02, grad_scale: 4.0
+2023-03-01 01:06:15,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:06:25,906 INFO [optim.py:369] (1/2) 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 01:07:00,254 INFO [train.py:968] (1/2) Epoch 2, batch 7900, giga_loss[loss=0.4784, simple_loss=0.4795, pruned_loss=0.2386, over 26554.00 frames. ], tot_loss[loss=0.3812, simple_loss=0.4233, pruned_loss=0.1695, over 5695164.27 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4195, pruned_loss=0.1564, over 5670439.57 frames. ], giga_tot_loss[loss=0.3843, simple_loss=0.4246, pruned_loss=0.172, over 5689076.57 frames. ], batch size: 555, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:07:48,894 INFO [train.py:968] (1/2) Epoch 2, batch 7950, giga_loss[loss=0.3566, simple_loss=0.4045, pruned_loss=0.1544, over 28293.00 frames. ], tot_loss[loss=0.3808, simple_loss=0.4226, pruned_loss=0.1695, over 5701542.72 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4197, pruned_loss=0.1567, over 5672987.13 frames. ], giga_tot_loss[loss=0.3832, simple_loss=0.4234, pruned_loss=0.1714, over 5694787.15 frames. ], batch size: 368, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:08:00,585 INFO [optim.py:369] (1/2) 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 01:08:05,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:08:30,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7095, 3.7811, 4.4115, 2.0843], device='cuda:1'), covar=tensor([0.0368, 0.0412, 0.0683, 0.1595], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0531, 0.0822, 0.0546], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 01:08:37,951 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:08:40,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:08:43,799 INFO [train.py:968] (1/2) Epoch 2, batch 8000, giga_loss[loss=0.444, simple_loss=0.4435, pruned_loss=0.2223, over 23476.00 frames. ], tot_loss[loss=0.3842, simple_loss=0.4249, pruned_loss=0.1718, over 5685627.77 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4197, pruned_loss=0.1567, over 5672987.13 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4255, pruned_loss=0.1733, over 5680369.82 frames. ], batch size: 705, lr: 1.49e-02, grad_scale: 8.0
+2023-03-01 01:09:01,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6100, 2.2061, 1.4227, 1.2642], device='cuda:1'), covar=tensor([0.0944, 0.0674, 0.0927, 0.1560], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0471, 0.0350, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 01:09:04,966 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:09:07,173 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:09:17,641 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:09:27,623 INFO [train.py:968] (1/2) Epoch 2, batch 8050, giga_loss[loss=0.365, simple_loss=0.4168, pruned_loss=0.1566, over 28952.00 frames. ], tot_loss[loss=0.3834, simple_loss=0.4252, pruned_loss=0.1708, over 5673953.80 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.42, pruned_loss=0.1568, over 5658233.98 frames. ], giga_tot_loss[loss=0.3852, simple_loss=0.4257, pruned_loss=0.1724, over 5683098.49 frames. ], batch size: 112, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:09:37,304 INFO [optim.py:369] (1/2) 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 01:09:45,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4785, 1.9693, 1.4504, 0.4480], device='cuda:1'), covar=tensor([0.1174, 0.0699, 0.1079, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.1118, 0.1116, 0.1111, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 01:10:03,753 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:10:11,727 INFO [train.py:968] (1/2) Epoch 2, batch 8100, giga_loss[loss=0.5201, simple_loss=0.5002, pruned_loss=0.27, over 26534.00 frames. ], tot_loss[loss=0.3839, simple_loss=0.426, pruned_loss=0.1709, over 5670168.90 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4199, pruned_loss=0.1568, over 5666223.84 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4267, pruned_loss=0.1727, over 5670507.91 frames. ], batch size: 555, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:10:46,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7518, 1.5862, 1.1933, 1.2972], device='cuda:1'), covar=tensor([0.0654, 0.0759, 0.1034, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0542, 0.0561, 0.0497], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 01:11:00,431 INFO [train.py:968] (1/2) Epoch 2, batch 8150, giga_loss[loss=0.3615, simple_loss=0.4096, pruned_loss=0.1567, over 28659.00 frames. ], tot_loss[loss=0.384, simple_loss=0.426, pruned_loss=0.171, over 5675490.17 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4192, pruned_loss=0.1565, over 5668344.68 frames. ], giga_tot_loss[loss=0.3867, simple_loss=0.4272, pruned_loss=0.1731, over 5674001.99 frames. ], batch size: 242, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:11:09,698 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:11:58,816 INFO [train.py:968] (1/2) Epoch 2, batch 8200, giga_loss[loss=0.3912, simple_loss=0.44, pruned_loss=0.1712, over 28982.00 frames. ], tot_loss[loss=0.3875, simple_loss=0.4285, pruned_loss=0.1733, over 5680708.59 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4195, pruned_loss=0.1566, over 5671971.56 frames. ], giga_tot_loss[loss=0.3896, simple_loss=0.4293, pruned_loss=0.175, over 5676350.33 frames. ], batch size: 145, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:12:37,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4983, 2.8082, 3.2833, 1.7542], device='cuda:1'), covar=tensor([0.0615, 0.0554, 0.0910, 0.1475], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0535, 0.0838, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:1')
+2023-03-01 01:12:56,911 INFO [train.py:968] (1/2) Epoch 2, batch 8250, giga_loss[loss=0.3634, simple_loss=0.4175, pruned_loss=0.1547, over 28980.00 frames. ], tot_loss[loss=0.3905, simple_loss=0.4298, pruned_loss=0.1756, over 5675242.92 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4192, pruned_loss=0.1564, over 5666746.33 frames. ], giga_tot_loss[loss=0.3929, simple_loss=0.4309, pruned_loss=0.1774, over 5676402.32 frames. ], batch size: 164, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:13:08,145 INFO [optim.py:369] (1/2) 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 01:13:49,037 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 8300, giga_loss[loss=0.543, simple_loss=0.5128, pruned_loss=0.2866, over 26358.00 frames. ], tot_loss[loss=0.3931, simple_loss=0.4306, pruned_loss=0.1778, over 5669402.52 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4187, pruned_loss=0.1562, over 5673866.04 frames. ], giga_tot_loss[loss=0.396, simple_loss=0.4322, pruned_loss=0.1799, over 5664299.30 frames. ], batch size: 555, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:13:52,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:14:21,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:14:32,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:14:43,626 INFO [train.py:968] (1/2) Epoch 2, batch 8350, giga_loss[loss=0.3446, simple_loss=0.3979, pruned_loss=0.1456, over 28999.00 frames. ], tot_loss[loss=0.3972, simple_loss=0.4329, pruned_loss=0.1807, over 5661156.56 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4192, pruned_loss=0.1566, over 5666594.13 frames. ], giga_tot_loss[loss=0.3995, simple_loss=0.4339, pruned_loss=0.1825, over 5662963.22 frames. ], batch size: 128, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:14:53,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 01:14:53,990 INFO [optim.py:369] (1/2) 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 01:15:32,134 INFO [train.py:968] (1/2) Epoch 2, batch 8400, giga_loss[loss=0.3833, simple_loss=0.4187, pruned_loss=0.1739, over 28563.00 frames. ], tot_loss[loss=0.3947, simple_loss=0.431, pruned_loss=0.1792, over 5661217.17 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4196, pruned_loss=0.1567, over 5669912.34 frames. ], giga_tot_loss[loss=0.3967, simple_loss=0.4317, pruned_loss=0.1808, over 5659659.26 frames. ], batch size: 336, lr: 1.49e-02, grad_scale: 8.0
+2023-03-01 01:15:36,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:15:48,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:15:57,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5229, 1.8416, 1.5716, 1.5535], device='cuda:1'), covar=tensor([0.1412, 0.1722, 0.1268, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0827, 0.0714, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 01:16:20,683 INFO [train.py:968] (1/2) Epoch 2, batch 8450, giga_loss[loss=0.3524, simple_loss=0.4014, pruned_loss=0.1517, over 28690.00 frames. ], tot_loss[loss=0.3923, simple_loss=0.4302, pruned_loss=0.1772, over 5667847.95 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4197, pruned_loss=0.1566, over 5673696.85 frames. ], giga_tot_loss[loss=0.3943, simple_loss=0.4308, pruned_loss=0.1789, over 5663040.88 frames. ], batch size: 262, lr: 1.49e-02, grad_scale: 8.0
+2023-03-01 01:16:30,474 INFO [optim.py:369] (1/2) 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 01:16:37,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:16:52,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:16:55,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8240, 2.2722, 1.7905, 1.8311], device='cuda:1'), covar=tensor([0.0574, 0.0698, 0.0873, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0544, 0.0563, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 01:16:56,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:02,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0452, 1.7817, 1.7185, 1.6161], device='cuda:1'), covar=tensor([0.0807, 0.1522, 0.1194, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0814, 0.0630, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 01:17:08,976 INFO [train.py:968] (1/2) Epoch 2, batch 8500, libri_loss[loss=0.3076, simple_loss=0.3649, pruned_loss=0.1252, over 29399.00 frames. ], tot_loss[loss=0.3881, simple_loss=0.4274, pruned_loss=0.1744, over 5643359.30 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4203, pruned_loss=0.1572, over 5653137.17 frames. ], giga_tot_loss[loss=0.3896, simple_loss=0.4276, pruned_loss=0.1758, over 5657960.98 frames. ], batch size: 71, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:17:22,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:48,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:51,560 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:17:51,927 INFO [train.py:968] (1/2) Epoch 2, batch 8550, giga_loss[loss=0.3977, simple_loss=0.434, pruned_loss=0.1808, over 28775.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4259, pruned_loss=0.1735, over 5658537.89 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4208, pruned_loss=0.1576, over 5658561.84 frames. ], giga_tot_loss[loss=0.3876, simple_loss=0.4257, pruned_loss=0.1747, over 5665313.97 frames. ], batch size: 66, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:18:00,304 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:04,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:05,893 INFO [optim.py:369] (1/2) 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 01:18:21,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:32,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:42,846 INFO [train.py:968] (1/2) Epoch 2, batch 8600, giga_loss[loss=0.5389, simple_loss=0.5052, pruned_loss=0.2863, over 26586.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4241, pruned_loss=0.1728, over 5654179.46 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4213, pruned_loss=0.1577, over 5648704.79 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4236, pruned_loss=0.1738, over 5667703.86 frames. ], batch size: 555, lr: 1.49e-02, grad_scale: 4.0
+2023-03-01 01:18:43,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5985, 2.1410, 1.6096, 0.7070], device='cuda:1'), covar=tensor([0.1221, 0.0729, 0.1185, 0.1519], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.1123, 0.1115, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 01:18:52,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:18:54,585 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:19:26,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=54310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:19:35,310 INFO [train.py:968] (1/2) Epoch 2, batch 8650, giga_loss[loss=0.3534, simple_loss=0.4076, pruned_loss=0.1496, over 28935.00 frames. ], tot_loss[loss=0.3845, simple_loss=0.4233, pruned_loss=0.1729, over 5649679.03 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4208, pruned_loss=0.1573, over 5652703.10 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4235, pruned_loss=0.1743, over 5656840.35 frames. ], batch size: 145, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:19:40,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7870, 2.1452, 1.7764, 1.7514], device='cuda:1'), covar=tensor([0.1119, 0.0999, 0.0895, 0.0541], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0831, 0.0726, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0011, 0.0009, 0.0006], device='cuda:1')
+2023-03-01 01:19:48,361 INFO [optim.py:369] (1/2) 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 01:20:25,239 INFO [train.py:968] (1/2) Epoch 2, batch 8700, giga_loss[loss=0.4285, simple_loss=0.4603, pruned_loss=0.1984, over 28901.00 frames. ], tot_loss[loss=0.3877, simple_loss=0.4263, pruned_loss=0.1746, over 5654414.08 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4204, pruned_loss=0.1571, over 5660134.58 frames. ], giga_tot_loss[loss=0.3899, simple_loss=0.4268, pruned_loss=0.1765, over 5653311.49 frames. ], batch size: 66, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:20:49,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9598, 1.6528, 1.6578, 1.4738], device='cuda:1'), covar=tensor([0.0767, 0.1545, 0.1052, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0834, 0.0637, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 01:21:16,800 INFO [train.py:968] (1/2) Epoch 2, batch 8750, libri_loss[loss=0.3292, simple_loss=0.3828, pruned_loss=0.1378, over 29549.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4289, pruned_loss=0.1734, over 5661898.28 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4202, pruned_loss=0.1569, over 5666288.69 frames. ], giga_tot_loss[loss=0.3908, simple_loss=0.4298, pruned_loss=0.1759, over 5654992.02 frames. ], batch size: 77, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:21:28,060 INFO [optim.py:369] (1/2) 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 01:21:39,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-01 01:22:09,613 INFO [train.py:968] (1/2) Epoch 2, batch 8800, libri_loss[loss=0.3021, simple_loss=0.361, pruned_loss=0.1216, over 29652.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.43, pruned_loss=0.1721, over 5673386.44 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4197, pruned_loss=0.1566, over 5667465.14 frames. ], giga_tot_loss[loss=0.3903, simple_loss=0.4315, pruned_loss=0.1746, over 5666908.29 frames. ], batch size: 69, lr: 1.48e-02, grad_scale: 8.0
+2023-03-01 01:22:56,574 INFO [train.py:968] (1/2) Epoch 2, batch 8850, giga_loss[loss=0.3642, simple_loss=0.4199, pruned_loss=0.1542, over 28818.00 frames. ], tot_loss[loss=0.3909, simple_loss=0.4325, pruned_loss=0.1746, over 5673671.94 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4186, pruned_loss=0.156, over 5672209.49 frames. ], giga_tot_loss[loss=0.395, simple_loss=0.4349, pruned_loss=0.1775, over 5664253.39 frames. ], batch size: 60, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:23:01,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6637, 1.3862, 1.0777, 1.2395], device='cuda:1'), covar=tensor([0.0604, 0.0722, 0.0903, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0535, 0.0555, 0.0501], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 01:23:06,099 INFO [optim.py:369] (1/2) 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 01:23:42,202 INFO [train.py:968] (1/2) Epoch 2, batch 8900, giga_loss[loss=0.3738, simple_loss=0.4257, pruned_loss=0.161, over 28854.00 frames. ], tot_loss[loss=0.3936, simple_loss=0.4343, pruned_loss=0.1765, over 5669461.19 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.4183, pruned_loss=0.1557, over 5682597.29 frames. ], giga_tot_loss[loss=0.3986, simple_loss=0.4372, pruned_loss=0.18, over 5652197.62 frames. ], batch size: 199, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:24:29,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3154, 1.5733, 1.2250, 1.3566], device='cuda:1'), covar=tensor([0.0998, 0.0461, 0.0510, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0225, 0.0229, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0021, 0.0019, 0.0031], device='cuda:1')
+2023-03-01 01:24:32,846 INFO [train.py:968] (1/2) Epoch 2, batch 8950, giga_loss[loss=0.4215, simple_loss=0.432, pruned_loss=0.2056, over 23743.00 frames. ], tot_loss[loss=0.3932, simple_loss=0.4334, pruned_loss=0.1765, over 5667523.16 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4182, pruned_loss=0.1554, over 5683885.58 frames. ], giga_tot_loss[loss=0.3977, simple_loss=0.436, pruned_loss=0.1797, over 5652566.61 frames. ], batch size: 705, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:24:36,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4036, 1.5744, 1.2753, 0.7693], device='cuda:1'), covar=tensor([0.0566, 0.0344, 0.0321, 0.0509], device='cuda:1'), in_proj_covar=tensor([0.0964, 0.0702, 0.0787, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 01:24:39,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 01:24:46,253 INFO [optim.py:369] (1/2) 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 01:24:47,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7609, 1.6083, 4.0519, 3.2221], device='cuda:1'), covar=tensor([0.1606, 0.1444, 0.0318, 0.0443], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0489, 0.0638, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 01:24:51,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 01:25:27,106 INFO [train.py:968] (1/2) Epoch 2, batch 9000, libri_loss[loss=0.4306, simple_loss=0.46, pruned_loss=0.2006, over 20341.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.4328, pruned_loss=0.178, over 5639781.44 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4181, pruned_loss=0.1554, over 5677542.86 frames. ], giga_tot_loss[loss=0.3983, simple_loss=0.435, pruned_loss=0.1808, over 5633898.74 frames. ], batch size: 188, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:25:27,107 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 01:25:36,185 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 01:26:22,807 INFO [train.py:968] (1/2) Epoch 2, batch 9050, giga_loss[loss=0.3957, simple_loss=0.4187, pruned_loss=0.1863, over 28808.00 frames. ], tot_loss[loss=0.3917, simple_loss=0.4306, pruned_loss=0.1764, over 5647042.29 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4181, pruned_loss=0.1555, over 5673386.52 frames. ], giga_tot_loss[loss=0.3955, simple_loss=0.4327, pruned_loss=0.1792, over 5644732.25 frames. ], batch size: 99, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:26:29,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3115, 1.5736, 1.3290, 1.3305], device='cuda:1'), covar=tensor([0.1164, 0.0461, 0.0521, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0226, 0.0228, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0021, 0.0019, 0.0031], device='cuda:1')
+2023-03-01 01:26:33,610 INFO [optim.py:369] (1/2) 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 01:27:13,570 INFO [train.py:968] (1/2) Epoch 2, batch 9100, giga_loss[loss=0.4562, simple_loss=0.4697, pruned_loss=0.2214, over 27565.00 frames. ], tot_loss[loss=0.394, simple_loss=0.4314, pruned_loss=0.1782, over 5655980.03 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4182, pruned_loss=0.1556, over 5680430.30 frames. ], giga_tot_loss[loss=0.3976, simple_loss=0.4334, pruned_loss=0.1809, over 5647631.86 frames. ], batch size: 472, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:28:04,569 INFO [train.py:968] (1/2) Epoch 2, batch 9150, giga_loss[loss=0.3987, simple_loss=0.4333, pruned_loss=0.1821, over 28499.00 frames. ], tot_loss[loss=0.394, simple_loss=0.4313, pruned_loss=0.1784, over 5652707.38 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4185, pruned_loss=0.1556, over 5683943.80 frames. ], giga_tot_loss[loss=0.3975, simple_loss=0.433, pruned_loss=0.181, over 5642257.18 frames. ], batch size: 60, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:28:18,758 INFO [optim.py:369] (1/2) 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 01:28:57,788 INFO [train.py:968] (1/2) Epoch 2, batch 9200, giga_loss[loss=0.3765, simple_loss=0.4183, pruned_loss=0.1674, over 28867.00 frames. ], tot_loss[loss=0.3931, simple_loss=0.4303, pruned_loss=0.178, over 5653476.35 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4182, pruned_loss=0.1555, over 5684867.87 frames. ], giga_tot_loss[loss=0.3967, simple_loss=0.4321, pruned_loss=0.1806, over 5644140.23 frames. ], batch size: 199, lr: 1.48e-02, grad_scale: 8.0
+2023-03-01 01:29:46,786 INFO [train.py:968] (1/2) Epoch 2, batch 9250, giga_loss[loss=0.3547, simple_loss=0.4025, pruned_loss=0.1535, over 28692.00 frames. ], tot_loss[loss=0.3902, simple_loss=0.4278, pruned_loss=0.1763, over 5657584.99 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4187, pruned_loss=0.1557, over 5685209.47 frames. ], giga_tot_loss[loss=0.3934, simple_loss=0.4291, pruned_loss=0.1788, over 5649422.50 frames. ], batch size: 92, lr: 1.48e-02, grad_scale: 8.0
+2023-03-01 01:30:01,888 INFO [optim.py:369] (1/2) 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 01:30:03,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-01 01:30:24,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-01 01:30:37,558 INFO [train.py:968] (1/2) Epoch 2, batch 9300, giga_loss[loss=0.4682, simple_loss=0.4589, pruned_loss=0.2387, over 23619.00 frames. ], tot_loss[loss=0.3907, simple_loss=0.4281, pruned_loss=0.1766, over 5644380.50 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.419, pruned_loss=0.1558, over 5682331.74 frames. ], giga_tot_loss[loss=0.3932, simple_loss=0.429, pruned_loss=0.1788, over 5640284.10 frames. ], batch size: 705, lr: 1.48e-02, grad_scale: 4.0
+2023-03-01 01:31:30,786 INFO [train.py:968] (1/2) Epoch 2, batch 9350, giga_loss[loss=0.385, simple_loss=0.4266, pruned_loss=0.1717, over 28869.00 frames. ], tot_loss[loss=0.3915, simple_loss=0.4298, pruned_loss=0.1767, over 5651785.63 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4188, pruned_loss=0.1558, over 5677216.33 frames. ], giga_tot_loss[loss=0.3943, simple_loss=0.4308, pruned_loss=0.1788, over 5652056.36 frames. ], batch size: 199, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:31:44,780 INFO [optim.py:369] (1/2) 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 01:32:09,057 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=55057.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 01:32:16,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2289, 2.4897, 1.6091, 0.9856], device='cuda:1'), covar=tensor([0.0452, 0.0326, 0.0352, 0.0579], device='cuda:1'), in_proj_covar=tensor([0.0969, 0.0710, 0.0801, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 01:32:20,010 INFO [train.py:968] (1/2) Epoch 2, batch 9400, giga_loss[loss=0.4474, simple_loss=0.4478, pruned_loss=0.2235, over 23672.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.4317, pruned_loss=0.1785, over 5649298.69 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4186, pruned_loss=0.1557, over 5680840.26 frames. ], giga_tot_loss[loss=0.3971, simple_loss=0.4329, pruned_loss=0.1806, over 5645905.26 frames. ], batch size: 705, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:33:08,888 INFO [train.py:968] (1/2) Epoch 2, batch 9450, libri_loss[loss=0.3583, simple_loss=0.4207, pruned_loss=0.148, over 29518.00 frames. ], tot_loss[loss=0.3947, simple_loss=0.4319, pruned_loss=0.1787, over 5654599.36 frames. ], libri_tot_loss[loss=0.3649, simple_loss=0.4186, pruned_loss=0.1556, over 5686170.02 frames. ], giga_tot_loss[loss=0.3976, simple_loss=0.4331, pruned_loss=0.1811, over 5646292.91 frames. ], batch size: 82, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:33:22,660 INFO [optim.py:369] (1/2) 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 01:33:56,027 INFO [train.py:968] (1/2) Epoch 2, batch 9500, giga_loss[loss=0.3536, simple_loss=0.4213, pruned_loss=0.1429, over 29018.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.4333, pruned_loss=0.1759, over 5665795.76 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4192, pruned_loss=0.1557, over 5691492.40 frames. ], giga_tot_loss[loss=0.3955, simple_loss=0.4343, pruned_loss=0.1783, over 5653502.01 frames. ], batch size: 136, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:34:28,290 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 01:34:44,921 INFO [train.py:968] (1/2) Epoch 2, batch 9550, libri_loss[loss=0.4029, simple_loss=0.4498, pruned_loss=0.178, over 29245.00 frames. ], tot_loss[loss=0.3941, simple_loss=0.4358, pruned_loss=0.1762, over 5664228.09 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.4192, pruned_loss=0.1559, over 5685733.49 frames. ], giga_tot_loss[loss=0.3966, simple_loss=0.4367, pruned_loss=0.1782, over 5659429.97 frames. ], batch size: 94, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:34:55,725 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-01 01:34:56,506 INFO [optim.py:369] (1/2) 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 01:35:20,859 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=55255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:35:21,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7789, 2.8645, 1.7191, 1.4462], device='cuda:1'), covar=tensor([0.0874, 0.0430, 0.0877, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0454, 0.0344, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0014], device='cuda:1')
+2023-03-01 01:35:31,788 INFO [train.py:968] (1/2) Epoch 2, batch 9600, giga_loss[loss=0.3893, simple_loss=0.4381, pruned_loss=0.1702, over 28864.00 frames. ], tot_loss[loss=0.3961, simple_loss=0.4381, pruned_loss=0.1771, over 5675613.74 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4187, pruned_loss=0.1556, over 5690633.45 frames. ], giga_tot_loss[loss=0.3995, simple_loss=0.4398, pruned_loss=0.1796, over 5666745.51 frames. ], batch size: 112, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:35:50,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 01:35:57,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4476, 1.2901, 1.1367, 1.7248], device='cuda:1'), covar=tensor([0.1738, 0.1867, 0.1557, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0798, 0.0875, 0.0923], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 01:36:03,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.89 vs. limit=5.0
+2023-03-01 01:36:22,041 INFO [train.py:968] (1/2) Epoch 2, batch 9650, giga_loss[loss=0.5038, simple_loss=0.5079, pruned_loss=0.2499, over 28596.00 frames. ], tot_loss[loss=0.4025, simple_loss=0.4418, pruned_loss=0.1816, over 5671634.30 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.419, pruned_loss=0.156, over 5689739.06 frames. ], giga_tot_loss[loss=0.4056, simple_loss=0.4435, pruned_loss=0.1839, over 5664949.33 frames. ], batch size: 307, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:36:34,020 INFO [optim.py:369] (1/2) 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 01:37:11,486 INFO [train.py:968] (1/2) Epoch 2, batch 9700, libri_loss[loss=0.3254, simple_loss=0.3899, pruned_loss=0.1304, over 29586.00 frames. ], tot_loss[loss=0.4042, simple_loss=0.4423, pruned_loss=0.1831, over 5667428.54 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.4185, pruned_loss=0.1555, over 5693785.16 frames. ], giga_tot_loss[loss=0.4081, simple_loss=0.4446, pruned_loss=0.1858, over 5658074.41 frames. ], batch size: 74, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:38:04,143 INFO [train.py:968] (1/2) Epoch 2, batch 9750, giga_loss[loss=0.3916, simple_loss=0.4402, pruned_loss=0.1714, over 28732.00 frames. ], tot_loss[loss=0.4041, simple_loss=0.442, pruned_loss=0.1831, over 5666370.13 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4184, pruned_loss=0.1554, over 5695437.13 frames. ], giga_tot_loss[loss=0.4079, simple_loss=0.4442, pruned_loss=0.1858, over 5656922.02 frames. ], batch size: 262, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:38:15,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1564, 1.2779, 1.1952, 1.3315], device='cuda:1'), covar=tensor([0.1034, 0.0500, 0.0515, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0220, 0.0224, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0021, 0.0019, 0.0031], device='cuda:1')
+2023-03-01 01:38:15,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55432.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:38:16,311 INFO [optim.py:369] (1/2) 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 01:38:49,151 INFO [train.py:968] (1/2) Epoch 2, batch 9800, giga_loss[loss=0.391, simple_loss=0.4375, pruned_loss=0.1723, over 28648.00 frames. ], tot_loss[loss=0.401, simple_loss=0.4401, pruned_loss=0.1809, over 5675528.74 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4184, pruned_loss=0.1555, over 5701384.30 frames. ], giga_tot_loss[loss=0.405, simple_loss=0.4425, pruned_loss=0.1837, over 5662019.95 frames. ], batch size: 262, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:39:37,706 INFO [train.py:968] (1/2) Epoch 2, batch 9850, giga_loss[loss=0.4017, simple_loss=0.4588, pruned_loss=0.1723, over 29025.00 frames. ], tot_loss[loss=0.3983, simple_loss=0.4395, pruned_loss=0.1786, over 5672000.24 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4183, pruned_loss=0.1555, over 5695463.33 frames. ], giga_tot_loss[loss=0.4023, simple_loss=0.442, pruned_loss=0.1813, over 5665807.02 frames. ], batch size: 136, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:39:53,009 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 2, batch 9900, giga_loss[loss=0.3998, simple_loss=0.4394, pruned_loss=0.1801, over 28867.00 frames. ], tot_loss[loss=0.3974, simple_loss=0.4396, pruned_loss=0.1775, over 5672593.47 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4186, pruned_loss=0.1557, over 5695104.04 frames. ], giga_tot_loss[loss=0.4007, simple_loss=0.4416, pruned_loss=0.1799, over 5667796.46 frames. ], batch size: 145, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:40:31,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55575.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:40:33,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55578.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:41:02,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=55604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:41:02,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2494, 1.2206, 1.1064, 1.3240], device='cuda:1'), covar=tensor([0.1709, 0.1855, 0.1508, 0.1960], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0812, 0.0878, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 01:41:06,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55607.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 01:41:09,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7064, 1.5049, 1.2114, 1.2215], device='cuda:1'), covar=tensor([0.0575, 0.0639, 0.0964, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0538, 0.0559, 0.0501], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 01:41:16,561 INFO [train.py:968] (1/2) Epoch 2, batch 9950, giga_loss[loss=0.4693, simple_loss=0.4829, pruned_loss=0.2278, over 27937.00 frames. ], tot_loss[loss=0.3988, simple_loss=0.4402, pruned_loss=0.1787, over 5671649.71 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4182, pruned_loss=0.1554, over 5699993.60 frames. ], giga_tot_loss[loss=0.4029, simple_loss=0.4428, pruned_loss=0.1815, over 5662496.04 frames. ], batch size: 412, lr: 1.47e-02, grad_scale: 2.0
+2023-03-01 01:41:29,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:41:34,184 INFO [optim.py:369] (1/2) 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 01:42:07,094 INFO [train.py:968] (1/2) Epoch 2, batch 10000, libri_loss[loss=0.3718, simple_loss=0.4416, pruned_loss=0.1511, over 29198.00 frames. ], tot_loss[loss=0.3981, simple_loss=0.4393, pruned_loss=0.1784, over 5667892.15 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.419, pruned_loss=0.1557, over 5703787.38 frames. ], giga_tot_loss[loss=0.4015, simple_loss=0.4411, pruned_loss=0.181, over 5656057.83 frames. ], batch size: 97, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:42:55,249 INFO [train.py:968] (1/2) Epoch 2, batch 10050, giga_loss[loss=0.3816, simple_loss=0.3966, pruned_loss=0.1833, over 23343.00 frames. ], tot_loss[loss=0.3964, simple_loss=0.4367, pruned_loss=0.1781, over 5653962.15 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4188, pruned_loss=0.1556, over 5704813.14 frames. ], giga_tot_loss[loss=0.4, simple_loss=0.4387, pruned_loss=0.1807, over 5642981.62 frames. ], batch size: 705, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:43:12,532 INFO [optim.py:369] (1/2) 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 01:43:46,768 INFO [train.py:968] (1/2) Epoch 2, batch 10100, giga_loss[loss=0.4423, simple_loss=0.4526, pruned_loss=0.216, over 26658.00 frames. ], tot_loss[loss=0.3956, simple_loss=0.4351, pruned_loss=0.1781, over 5666735.19 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4189, pruned_loss=0.1556, over 5708871.84 frames. ], giga_tot_loss[loss=0.399, simple_loss=0.437, pruned_loss=0.1806, over 5653739.68 frames. ], batch size: 555, lr: 1.47e-02, grad_scale: 4.0
+2023-03-01 01:43:48,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5739, 2.2792, 1.6620, 0.6643], device='cuda:1'), covar=tensor([0.1481, 0.0774, 0.1359, 0.1873], device='cuda:1'), in_proj_covar=tensor([0.1127, 0.1115, 0.1135, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 01:43:50,569 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:43:55,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:44:24,200 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 10150, giga_loss[loss=0.4254, simple_loss=0.4313, pruned_loss=0.2098, over 23226.00 frames. ], tot_loss[loss=0.3943, simple_loss=0.4331, pruned_loss=0.1777, over 5649821.10 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4194, pruned_loss=0.1559, over 5702198.97 frames. ], giga_tot_loss[loss=0.397, simple_loss=0.4343, pruned_loss=0.1798, over 5644909.32 frames. ], batch size: 705, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:44:56,653 INFO [optim.py:369] (1/2) 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 01:45:28,882 INFO [train.py:968] (1/2) Epoch 2, batch 10200, giga_loss[loss=0.3982, simple_loss=0.4357, pruned_loss=0.1803, over 28775.00 frames. ], tot_loss[loss=0.3949, simple_loss=0.4329, pruned_loss=0.1785, over 5652846.50 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.4195, pruned_loss=0.1558, over 5702949.15 frames. ], giga_tot_loss[loss=0.3976, simple_loss=0.4341, pruned_loss=0.1806, over 5647194.39 frames. ], batch size: 66, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:46:14,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:46:16,404 INFO [train.py:968] (1/2) Epoch 2, batch 10250, giga_loss[loss=0.392, simple_loss=0.425, pruned_loss=0.1795, over 28946.00 frames. ], tot_loss[loss=0.3919, simple_loss=0.4307, pruned_loss=0.1765, over 5656105.21 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4194, pruned_loss=0.1556, over 5705113.60 frames. ], giga_tot_loss[loss=0.3949, simple_loss=0.432, pruned_loss=0.1789, over 5648529.47 frames. ], batch size: 106, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:46:31,949 INFO [optim.py:369] (1/2) 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 01:47:04,094 INFO [train.py:968] (1/2) Epoch 2, batch 10300, giga_loss[loss=0.3348, simple_loss=0.3907, pruned_loss=0.1394, over 28594.00 frames. ], tot_loss[loss=0.3842, simple_loss=0.4266, pruned_loss=0.1709, over 5673133.93 frames. ], libri_tot_loss[loss=0.3653, simple_loss=0.4193, pruned_loss=0.1556, over 5712173.91 frames. ], giga_tot_loss[loss=0.3873, simple_loss=0.4279, pruned_loss=0.1734, over 5659333.66 frames. ], batch size: 85, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:47:12,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:47:30,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1072, 3.2750, 3.8374, 1.5307], device='cuda:1'), covar=tensor([0.0481, 0.0488, 0.0769, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0542, 0.0845, 0.0546], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0006, 0.0010, 0.0007], device='cuda:1')
+2023-03-01 01:47:52,705 INFO [train.py:968] (1/2) Epoch 2, batch 10350, giga_loss[loss=0.3707, simple_loss=0.4225, pruned_loss=0.1594, over 28834.00 frames. ], tot_loss[loss=0.3781, simple_loss=0.4225, pruned_loss=0.1669, over 5667784.23 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4188, pruned_loss=0.1552, over 5715537.81 frames. ], giga_tot_loss[loss=0.3817, simple_loss=0.4242, pruned_loss=0.1696, over 5652479.77 frames. ], batch size: 186, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:48:09,247 INFO [optim.py:369] (1/2) 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 01:48:30,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:48:34,754 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 10400, giga_loss[loss=0.3623, simple_loss=0.4121, pruned_loss=0.1562, over 28684.00 frames. ], tot_loss[loss=0.378, simple_loss=0.4226, pruned_loss=0.1667, over 5677195.35 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4187, pruned_loss=0.1552, over 5720519.26 frames. ], giga_tot_loss[loss=0.3811, simple_loss=0.4241, pruned_loss=0.1691, over 5659495.50 frames. ], batch size: 307, lr: 1.46e-02, grad_scale: 8.0
+2023-03-01 01:48:56,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1466, 2.2639, 1.4554, 1.1794], device='cuda:1'), covar=tensor([0.0415, 0.0339, 0.0380, 0.0515], device='cuda:1'), in_proj_covar=tensor([0.0966, 0.0710, 0.0775, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 01:49:05,704 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:49:37,194 INFO [train.py:968] (1/2) Epoch 2, batch 10450, giga_loss[loss=0.3399, simple_loss=0.3914, pruned_loss=0.1442, over 28690.00 frames. ], tot_loss[loss=0.3784, simple_loss=0.421, pruned_loss=0.1678, over 5661301.35 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4192, pruned_loss=0.1556, over 5713081.17 frames. ], giga_tot_loss[loss=0.3803, simple_loss=0.4218, pruned_loss=0.1694, over 5654040.92 frames. ], batch size: 66, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:49:39,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:49:41,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:49:52,686 INFO [optim.py:369] (1/2) 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 01:49:59,028 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:50:09,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0802, 1.2107, 1.0959, 0.8582], device='cuda:1'), covar=tensor([0.1774, 0.1878, 0.1510, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0959, 0.0821, 0.0887, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 01:50:13,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:50:30,672 INFO [train.py:968] (1/2) Epoch 2, batch 10500, giga_loss[loss=0.344, simple_loss=0.392, pruned_loss=0.148, over 28909.00 frames. ], tot_loss[loss=0.3763, simple_loss=0.4188, pruned_loss=0.1669, over 5662669.72 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4192, pruned_loss=0.1556, over 5714808.75 frames. ], giga_tot_loss[loss=0.378, simple_loss=0.4194, pruned_loss=0.1683, over 5654970.13 frames. ], batch size: 227, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:51:20,197 INFO [train.py:968] (1/2) Epoch 2, batch 10550, giga_loss[loss=0.3685, simple_loss=0.4282, pruned_loss=0.1544, over 28866.00 frames. ], tot_loss[loss=0.3778, simple_loss=0.4205, pruned_loss=0.1675, over 5667081.07 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4188, pruned_loss=0.1553, over 5716972.88 frames. ], giga_tot_loss[loss=0.3799, simple_loss=0.4214, pruned_loss=0.1691, over 5658095.55 frames. ], batch size: 145, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:51:37,676 INFO [optim.py:369] (1/2) 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 01:51:42,121 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:51:50,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2430, 1.2151, 0.9901, 1.3460], device='cuda:1'), covar=tensor([0.1109, 0.0520, 0.0536, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0327, 0.0219, 0.0225, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0021, 0.0019, 0.0032], device='cuda:1')
+2023-03-01 01:52:08,405 INFO [train.py:968] (1/2) Epoch 2, batch 10600, giga_loss[loss=0.3915, simple_loss=0.4248, pruned_loss=0.1791, over 27619.00 frames. ], tot_loss[loss=0.3784, simple_loss=0.4215, pruned_loss=0.1677, over 5652677.18 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4193, pruned_loss=0.1554, over 5710730.81 frames. ], giga_tot_loss[loss=0.3801, simple_loss=0.4218, pruned_loss=0.1692, over 5649317.19 frames. ], batch size: 472, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:52:57,689 INFO [train.py:968] (1/2) Epoch 2, batch 10650, libri_loss[loss=0.3762, simple_loss=0.4352, pruned_loss=0.1586, over 29535.00 frames. ], tot_loss[loss=0.3751, simple_loss=0.4193, pruned_loss=0.1655, over 5654617.88 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4194, pruned_loss=0.1553, over 5715330.66 frames. ], giga_tot_loss[loss=0.3769, simple_loss=0.4195, pruned_loss=0.1672, over 5645607.78 frames. ], batch size: 84, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:53:15,140 INFO [optim.py:369] (1/2) 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 01:53:25,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1553, 1.2532, 1.1564, 0.6942], device='cuda:1'), covar=tensor([0.0395, 0.0336, 0.0306, 0.0415], device='cuda:1'), in_proj_covar=tensor([0.0959, 0.0704, 0.0760, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 01:53:48,693 INFO [train.py:968] (1/2) Epoch 2, batch 10700, giga_loss[loss=0.3788, simple_loss=0.4198, pruned_loss=0.1689, over 28573.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.4198, pruned_loss=0.1665, over 5655065.32 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4203, pruned_loss=0.1559, over 5717625.66 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4191, pruned_loss=0.1675, over 5644735.48 frames. ], batch size: 336, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:54:01,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-01 01:54:16,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8541, 1.7880, 3.9009, 3.2448], device='cuda:1'), covar=tensor([0.1446, 0.1301, 0.0301, 0.0439], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0493, 0.0635, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 01:54:37,381 INFO [train.py:968] (1/2) Epoch 2, batch 10750, giga_loss[loss=0.3948, simple_loss=0.4315, pruned_loss=0.179, over 28916.00 frames. ], tot_loss[loss=0.3814, simple_loss=0.4236, pruned_loss=0.1696, over 5668315.60 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4204, pruned_loss=0.156, over 5722445.21 frames. ], giga_tot_loss[loss=0.3821, simple_loss=0.4229, pruned_loss=0.1707, over 5654061.00 frames. ], batch size: 112, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:54:56,919 INFO [optim.py:369] (1/2) 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 01:55:12,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5732, 1.4571, 3.5291, 2.8498], device='cuda:1'), covar=tensor([0.1514, 0.1415, 0.0345, 0.0486], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0491, 0.0628, 0.0519], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 01:55:27,468 INFO [train.py:968] (1/2) Epoch 2, batch 10800, giga_loss[loss=0.4525, simple_loss=0.4714, pruned_loss=0.2167, over 28896.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4269, pruned_loss=0.1715, over 5670633.29 frames. ], libri_tot_loss[loss=0.3665, simple_loss=0.4208, pruned_loss=0.1561, over 5726721.30 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.4263, pruned_loss=0.1728, over 5653348.47 frames. ], batch size: 199, lr: 1.46e-02, grad_scale: 8.0
+2023-03-01 01:55:32,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4151, 2.1043, 1.4359, 1.3682], device='cuda:1'), covar=tensor([0.0794, 0.0543, 0.0733, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0459, 0.0346, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 01:56:10,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:56:12,190 INFO [train.py:968] (1/2) Epoch 2, batch 10850, giga_loss[loss=0.4043, simple_loss=0.4421, pruned_loss=0.1832, over 28786.00 frames. ], tot_loss[loss=0.3867, simple_loss=0.4282, pruned_loss=0.1726, over 5671422.43 frames. ], libri_tot_loss[loss=0.3659, simple_loss=0.4203, pruned_loss=0.1558, over 5729134.52 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4284, pruned_loss=0.1747, over 5652831.01 frames. ], batch size: 119, lr: 1.46e-02, grad_scale: 4.0
+2023-03-01 01:56:23,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-01 01:56:29,374 INFO [optim.py:369] (1/2) 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 01:56:31,030 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:56:46,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8596, 2.1257, 1.3643, 1.0411], device='cuda:1'), covar=tensor([0.0553, 0.0426, 0.0421, 0.0561], device='cuda:1'), in_proj_covar=tensor([0.0985, 0.0728, 0.0784, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 01:56:59,920 INFO [train.py:968] (1/2) Epoch 2, batch 10900, libri_loss[loss=0.4022, simple_loss=0.4441, pruned_loss=0.1801, over 29407.00 frames. ], tot_loss[loss=0.3889, simple_loss=0.4296, pruned_loss=0.1741, over 5676984.27 frames. ], libri_tot_loss[loss=0.3658, simple_loss=0.4202, pruned_loss=0.1557, over 5730022.12 frames. ], giga_tot_loss[loss=0.3914, simple_loss=0.4302, pruned_loss=0.1763, over 5659486.50 frames. ], batch size: 92, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 01:57:05,778 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-01 01:57:51,987 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:57:53,643 INFO [train.py:968] (1/2) Epoch 2, batch 10950, giga_loss[loss=0.411, simple_loss=0.4506, pruned_loss=0.1857, over 28081.00 frames. ], tot_loss[loss=0.3911, simple_loss=0.4318, pruned_loss=0.1752, over 5680512.72 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.42, pruned_loss=0.1555, over 5729662.43 frames. ], giga_tot_loss[loss=0.3938, simple_loss=0.4326, pruned_loss=0.1775, over 5666224.49 frames. ], batch size: 412, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 01:58:13,647 INFO [optim.py:369] (1/2) 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 01:58:38,348 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:58:41,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:58:47,702 INFO [train.py:968] (1/2) Epoch 2, batch 11000, giga_loss[loss=0.3645, simple_loss=0.407, pruned_loss=0.161, over 28657.00 frames. ], tot_loss[loss=0.3906, simple_loss=0.432, pruned_loss=0.1747, over 5667105.54 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4196, pruned_loss=0.1552, over 5724572.00 frames. ], giga_tot_loss[loss=0.3939, simple_loss=0.4333, pruned_loss=0.1772, over 5658936.62 frames. ], batch size: 92, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 01:59:11,713 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 01:59:44,535 INFO [train.py:968] (1/2) Epoch 2, batch 11050, giga_loss[loss=0.4166, simple_loss=0.4519, pruned_loss=0.1906, over 28533.00 frames. ], tot_loss[loss=0.3954, simple_loss=0.4342, pruned_loss=0.1783, over 5646346.42 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4196, pruned_loss=0.1554, over 5716146.49 frames. ], giga_tot_loss[loss=0.398, simple_loss=0.4353, pruned_loss=0.1803, over 5646528.70 frames. ], batch size: 336, lr: 1.45e-02, grad_scale: 2.0
+2023-03-01 02:00:03,044 INFO [optim.py:369] (1/2) 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 02:00:29,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:00:32,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:00:41,988 INFO [train.py:968] (1/2) Epoch 2, batch 11100, giga_loss[loss=0.3994, simple_loss=0.4327, pruned_loss=0.1831, over 28030.00 frames. ], tot_loss[loss=0.3949, simple_loss=0.4328, pruned_loss=0.1784, over 5634413.60 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4194, pruned_loss=0.1553, over 5717885.10 frames. ], giga_tot_loss[loss=0.3973, simple_loss=0.434, pruned_loss=0.1803, over 5632019.74 frames. ], batch size: 77, lr: 1.45e-02, grad_scale: 2.0
+2023-03-01 02:01:09,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:01:35,515 INFO [train.py:968] (1/2) Epoch 2, batch 11150, giga_loss[loss=0.3559, simple_loss=0.4022, pruned_loss=0.1549, over 28757.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.4296, pruned_loss=0.1753, over 5648477.75 frames. ], libri_tot_loss[loss=0.3643, simple_loss=0.419, pruned_loss=0.1548, over 5721866.76 frames. ], giga_tot_loss[loss=0.3938, simple_loss=0.4314, pruned_loss=0.1781, over 5639774.72 frames. ], batch size: 92, lr: 1.45e-02, grad_scale: 2.0
+2023-03-01 02:01:54,869 INFO [optim.py:369] (1/2) 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 02:02:04,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6384, 2.5991, 1.4409, 1.3443], device='cuda:1'), covar=tensor([0.0850, 0.0518, 0.0913, 0.1454], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0461, 0.0345, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 02:02:27,423 INFO [train.py:968] (1/2) Epoch 2, batch 11200, giga_loss[loss=0.3717, simple_loss=0.4227, pruned_loss=0.1604, over 28680.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4285, pruned_loss=0.1756, over 5647816.04 frames. ], libri_tot_loss[loss=0.3643, simple_loss=0.419, pruned_loss=0.1548, over 5724658.62 frames. ], giga_tot_loss[loss=0.3933, simple_loss=0.4301, pruned_loss=0.1782, over 5637246.45 frames. ], batch size: 85, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:03:06,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:03:08,657 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=56916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 02:03:10,476 INFO [train.py:968] (1/2) Epoch 2, batch 11250, giga_loss[loss=0.3879, simple_loss=0.4245, pruned_loss=0.1757, over 28466.00 frames. ], tot_loss[loss=0.3887, simple_loss=0.4277, pruned_loss=0.1749, over 5647742.73 frames. ], libri_tot_loss[loss=0.3644, simple_loss=0.4189, pruned_loss=0.155, over 5711134.16 frames. ], giga_tot_loss[loss=0.3922, simple_loss=0.4294, pruned_loss=0.1775, over 5649253.82 frames. ], batch size: 65, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:03:31,182 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 11300, giga_loss[loss=0.3624, simple_loss=0.4199, pruned_loss=0.1525, over 28698.00 frames. ], tot_loss[loss=0.3887, simple_loss=0.4273, pruned_loss=0.175, over 5648924.47 frames. ], libri_tot_loss[loss=0.3643, simple_loss=0.4188, pruned_loss=0.1549, over 5712602.84 frames. ], giga_tot_loss[loss=0.3919, simple_loss=0.4288, pruned_loss=0.1775, over 5647854.93 frames. ], batch size: 60, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:04:51,245 INFO [train.py:968] (1/2) Epoch 2, batch 11350, libri_loss[loss=0.3716, simple_loss=0.4281, pruned_loss=0.1575, over 29519.00 frames. ], tot_loss[loss=0.39, simple_loss=0.4281, pruned_loss=0.176, over 5658482.01 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.4182, pruned_loss=0.1545, over 5721020.48 frames. ], giga_tot_loss[loss=0.3943, simple_loss=0.4302, pruned_loss=0.1792, over 5647403.15 frames. ], batch size: 84, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:04:52,741 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-01 02:05:15,119 INFO [optim.py:369] (1/2) 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 02:05:31,856 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:05:33,938 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 11400, giga_loss[loss=0.365, simple_loss=0.4197, pruned_loss=0.1551, over 28911.00 frames. ], tot_loss[loss=0.3947, simple_loss=0.4312, pruned_loss=0.1791, over 5661456.57 frames. ], libri_tot_loss[loss=0.3637, simple_loss=0.4182, pruned_loss=0.1545, over 5723095.12 frames. ], giga_tot_loss[loss=0.3983, simple_loss=0.433, pruned_loss=0.1818, over 5650218.41 frames. ], batch size: 145, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:06:00,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:06:30,746 INFO [train.py:968] (1/2) Epoch 2, batch 11450, giga_loss[loss=0.4153, simple_loss=0.4474, pruned_loss=0.1916, over 28739.00 frames. ], tot_loss[loss=0.3949, simple_loss=0.4312, pruned_loss=0.1793, over 5647191.84 frames. ], libri_tot_loss[loss=0.3635, simple_loss=0.4181, pruned_loss=0.1544, over 5718786.66 frames. ], giga_tot_loss[loss=0.3989, simple_loss=0.4332, pruned_loss=0.1823, over 5639339.89 frames. ], batch size: 284, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:06:34,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-01 02:06:51,177 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 11500, giga_loss[loss=0.3895, simple_loss=0.4321, pruned_loss=0.1734, over 28631.00 frames. ], tot_loss[loss=0.3951, simple_loss=0.4313, pruned_loss=0.1794, over 5639141.38 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.4183, pruned_loss=0.1544, over 5706886.12 frames. ], giga_tot_loss[loss=0.399, simple_loss=0.433, pruned_loss=0.1825, over 5642070.74 frames. ], batch size: 60, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:07:50,031 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:08:11,605 INFO [train.py:968] (1/2) Epoch 2, batch 11550, giga_loss[loss=0.4857, simple_loss=0.4836, pruned_loss=0.2439, over 26612.00 frames. ], tot_loss[loss=0.3948, simple_loss=0.4316, pruned_loss=0.179, over 5646191.13 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.4185, pruned_loss=0.1544, over 5706867.09 frames. ], giga_tot_loss[loss=0.3985, simple_loss=0.433, pruned_loss=0.182, over 5647400.85 frames. ], batch size: 555, lr: 1.45e-02, grad_scale: 4.0
+2023-03-01 02:08:15,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-01 02:08:29,841 INFO [optim.py:369] (1/2) 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 02:09:00,078 INFO [train.py:968] (1/2) Epoch 2, batch 11600, giga_loss[loss=0.4094, simple_loss=0.45, pruned_loss=0.1843, over 28938.00 frames. ], tot_loss[loss=0.3969, simple_loss=0.4332, pruned_loss=0.1803, over 5640014.16 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4182, pruned_loss=0.1542, over 5702400.83 frames. ], giga_tot_loss[loss=0.4012, simple_loss=0.435, pruned_loss=0.1837, over 5643556.66 frames. ], batch size: 213, lr: 1.45e-02, grad_scale: 8.0
+2023-03-01 02:09:22,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 02:09:32,419 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:09:37,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5048, 1.4946, 1.1097, 1.1425], device='cuda:1'), covar=tensor([0.0626, 0.0658, 0.1042, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0535, 0.0555, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 02:09:47,521 INFO [train.py:968] (1/2) Epoch 2, batch 11650, giga_loss[loss=0.3408, simple_loss=0.4018, pruned_loss=0.1399, over 28840.00 frames. ], tot_loss[loss=0.3932, simple_loss=0.4315, pruned_loss=0.1774, over 5661952.11 frames. ], libri_tot_loss[loss=0.3623, simple_loss=0.4175, pruned_loss=0.1536, over 5707714.01 frames. ], giga_tot_loss[loss=0.3983, simple_loss=0.434, pruned_loss=0.1813, over 5658458.90 frames. ], batch size: 99, lr: 1.45e-02, grad_scale: 8.0
+2023-03-01 02:09:51,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2902, 1.6620, 1.4741, 1.4988], device='cuda:1'), covar=tensor([0.1294, 0.1681, 0.1181, 0.0736], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0834, 0.0726, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0006], device='cuda:1')
+2023-03-01 02:10:10,042 INFO [optim.py:369] (1/2) 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 02:10:43,003 INFO [train.py:968] (1/2) Epoch 2, batch 11700, giga_loss[loss=0.3708, simple_loss=0.4209, pruned_loss=0.1603, over 28968.00 frames. ], tot_loss[loss=0.3965, simple_loss=0.4334, pruned_loss=0.1798, over 5651481.74 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.4174, pruned_loss=0.1534, over 5710233.76 frames. ], giga_tot_loss[loss=0.4014, simple_loss=0.4358, pruned_loss=0.1835, over 5645449.97 frames. ], batch size: 213, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:11:01,415 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:11:03,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:11:36,391 INFO [train.py:968] (1/2) Epoch 2, batch 11750, libri_loss[loss=0.3632, simple_loss=0.4252, pruned_loss=0.1506, over 29175.00 frames. ], tot_loss[loss=0.3981, simple_loss=0.4348, pruned_loss=0.1807, over 5657450.17 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.417, pruned_loss=0.153, over 5715502.81 frames. ], giga_tot_loss[loss=0.4036, simple_loss=0.4375, pruned_loss=0.1848, over 5646230.85 frames. ], batch size: 101, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:11:48,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 02:11:53,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57437.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 02:11:56,365 INFO [optim.py:369] (1/2) 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 02:12:21,521 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57466.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 02:12:24,434 INFO [train.py:968] (1/2) Epoch 2, batch 11800, giga_loss[loss=0.3676, simple_loss=0.4205, pruned_loss=0.1573, over 28643.00 frames. ], tot_loss[loss=0.3975, simple_loss=0.4339, pruned_loss=0.1805, over 5649617.69 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4165, pruned_loss=0.1527, over 5709497.24 frames. ], giga_tot_loss[loss=0.403, simple_loss=0.4368, pruned_loss=0.1846, over 5645230.42 frames. ], batch size: 85, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:12:39,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4684, 1.5444, 0.8975, 1.3629], device='cuda:1'), covar=tensor([0.0939, 0.0842, 0.1997, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0540, 0.0556, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 02:13:08,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4339, 1.7108, 1.2824, 1.4885], device='cuda:1'), covar=tensor([0.1132, 0.0429, 0.0530, 0.1314], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0218, 0.0222, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0022, 0.0020, 0.0032], device='cuda:1')
+2023-03-01 02:13:14,332 INFO [train.py:968] (1/2) Epoch 2, batch 11850, giga_loss[loss=0.3839, simple_loss=0.4328, pruned_loss=0.1675, over 28639.00 frames. ], tot_loss[loss=0.3955, simple_loss=0.4335, pruned_loss=0.1787, over 5646697.74 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.4161, pruned_loss=0.1525, over 5706302.55 frames. ], giga_tot_loss[loss=0.4012, simple_loss=0.4367, pruned_loss=0.1829, over 5644521.41 frames. ], batch size: 92, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:13:35,443 INFO [optim.py:369] (1/2) 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 02:14:06,460 INFO [train.py:968] (1/2) Epoch 2, batch 11900, giga_loss[loss=0.4094, simple_loss=0.4464, pruned_loss=0.1863, over 28596.00 frames. ], tot_loss[loss=0.3938, simple_loss=0.4329, pruned_loss=0.1774, over 5649838.91 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4159, pruned_loss=0.1524, over 5710476.75 frames. ], giga_tot_loss[loss=0.3993, simple_loss=0.4359, pruned_loss=0.1813, over 5643154.50 frames. ], batch size: 307, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:14:09,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:14:15,086 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-01 02:14:15,915 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:14:26,945 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:14:54,817 INFO [train.py:968] (1/2) Epoch 2, batch 11950, giga_loss[loss=0.3491, simple_loss=0.4046, pruned_loss=0.1468, over 28838.00 frames. ], tot_loss[loss=0.3921, simple_loss=0.4316, pruned_loss=0.1763, over 5647073.65 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4158, pruned_loss=0.1522, over 5712437.69 frames. ], giga_tot_loss[loss=0.3972, simple_loss=0.4344, pruned_loss=0.18, over 5639004.85 frames. ], batch size: 174, lr: 1.44e-02, grad_scale: 2.0
+2023-03-01 02:15:02,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-01 02:15:10,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1175, 3.4299, 3.7885, 1.6346], device='cuda:1'), covar=tensor([0.0550, 0.0488, 0.0918, 0.1861], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0549, 0.0850, 0.0556], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0007], device='cuda:1')
+2023-03-01 02:15:11,024 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:15:12,685 INFO [optim.py:369] (1/2) 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 02:15:39,720 INFO [train.py:968] (1/2) Epoch 2, batch 12000, giga_loss[loss=0.4213, simple_loss=0.4533, pruned_loss=0.1947, over 28912.00 frames. ], tot_loss[loss=0.389, simple_loss=0.4293, pruned_loss=0.1744, over 5660881.81 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4159, pruned_loss=0.1522, over 5718248.26 frames. ], giga_tot_loss[loss=0.3939, simple_loss=0.4318, pruned_loss=0.178, over 5647620.17 frames. ], batch size: 227, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:15:39,720 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 02:15:48,565 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 02:15:55,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:16:31,034 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 2, batch 12050, libri_loss[loss=0.396, simple_loss=0.4439, pruned_loss=0.1741, over 26106.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.4319, pruned_loss=0.1767, over 5648073.55 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4164, pruned_loss=0.1526, over 5707793.20 frames. ], giga_tot_loss[loss=0.3971, simple_loss=0.434, pruned_loss=0.1801, over 5645543.14 frames. ], batch size: 136, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:16:54,771 INFO [optim.py:369] (1/2) 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 02:17:00,775 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:17:14,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:17:17,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:17:22,076 INFO [train.py:968] (1/2) Epoch 2, batch 12100, giga_loss[loss=0.3345, simple_loss=0.3962, pruned_loss=0.1364, over 28969.00 frames. ], tot_loss[loss=0.3924, simple_loss=0.4317, pruned_loss=0.1765, over 5635462.81 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4164, pruned_loss=0.1525, over 5695685.86 frames. ], giga_tot_loss[loss=0.3966, simple_loss=0.4337, pruned_loss=0.1798, over 5642366.63 frames. ], batch size: 106, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:17:28,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5007, 2.3039, 1.4709, 1.3247], device='cuda:1'), covar=tensor([0.0882, 0.0728, 0.0912, 0.1458], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0462, 0.0344, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 02:17:44,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:13,925 INFO [train.py:968] (1/2) Epoch 2, batch 12150, libri_loss[loss=0.3673, simple_loss=0.4258, pruned_loss=0.1543, over 29238.00 frames. ], tot_loss[loss=0.3907, simple_loss=0.4301, pruned_loss=0.1757, over 5643376.01 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.417, pruned_loss=0.1528, over 5684212.77 frames. ], giga_tot_loss[loss=0.3947, simple_loss=0.4316, pruned_loss=0.1789, over 5656080.84 frames. ], batch size: 97, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:18:14,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:17,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:33,400 INFO [optim.py:369] (1/2) 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 02:18:44,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:18:44,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1981, 1.2323, 1.0917, 1.2113], device='cuda:1'), covar=tensor([0.2216, 0.2189, 0.1893, 0.2095], device='cuda:1'), in_proj_covar=tensor([0.0954, 0.0798, 0.0883, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 02:18:47,161 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:00,858 INFO [train.py:968] (1/2) Epoch 2, batch 12200, libri_loss[loss=0.3617, simple_loss=0.4204, pruned_loss=0.1516, over 27679.00 frames. ], tot_loss[loss=0.3906, simple_loss=0.4298, pruned_loss=0.1757, over 5652656.38 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.4172, pruned_loss=0.1531, over 5688340.74 frames. ], giga_tot_loss[loss=0.3947, simple_loss=0.4314, pruned_loss=0.1791, over 5657833.65 frames. ], batch size: 115, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:19:19,940 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=57888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:26,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5660, 1.9468, 1.6544, 1.6882], device='cuda:1'), covar=tensor([0.1360, 0.1799, 0.1256, 0.0753], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0825, 0.0720, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0006], device='cuda:1')
+2023-03-01 02:19:31,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-01 02:19:33,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:36,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:37,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:39,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:19:48,307 INFO [train.py:968] (1/2) Epoch 2, batch 12250, giga_loss[loss=0.5177, simple_loss=0.5033, pruned_loss=0.2661, over 26686.00 frames. ], tot_loss[loss=0.3924, simple_loss=0.4311, pruned_loss=0.1768, over 5656149.80 frames. ], libri_tot_loss[loss=0.3618, simple_loss=0.4175, pruned_loss=0.1531, over 5689498.02 frames. ], giga_tot_loss[loss=0.3963, simple_loss=0.4325, pruned_loss=0.18, over 5658552.47 frames. ], batch size: 555, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:19:54,159 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8064, 1.4914, 1.4832, 1.5179], device='cuda:1'), covar=tensor([0.0798, 0.1657, 0.1210, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0829, 0.0631, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 02:20:09,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:20:11,906 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:20:37,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:20:39,831 INFO [train.py:968] (1/2) Epoch 2, batch 12300, libri_loss[loss=0.4109, simple_loss=0.4563, pruned_loss=0.1828, over 25787.00 frames. ], tot_loss[loss=0.3936, simple_loss=0.432, pruned_loss=0.1776, over 5652404.30 frames. ], libri_tot_loss[loss=0.3622, simple_loss=0.4178, pruned_loss=0.1533, over 5687840.46 frames. ], giga_tot_loss[loss=0.3968, simple_loss=0.433, pruned_loss=0.1803, over 5655262.19 frames. ], batch size: 136, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:21:21,177 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 12350, giga_loss[loss=0.3939, simple_loss=0.4345, pruned_loss=0.1767, over 28746.00 frames. ], tot_loss[loss=0.3912, simple_loss=0.4308, pruned_loss=0.1757, over 5665357.70 frames. ], libri_tot_loss[loss=0.3619, simple_loss=0.4176, pruned_loss=0.1531, over 5689239.69 frames. ], giga_tot_loss[loss=0.3949, simple_loss=0.4322, pruned_loss=0.1789, over 5665417.97 frames. ], batch size: 243, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:21:49,509 INFO [optim.py:369] (1/2) 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 02:21:52,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9140, 1.7394, 4.7974, 3.7462], device='cuda:1'), covar=tensor([0.1498, 0.1394, 0.0250, 0.0331], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0493, 0.0648, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 02:22:15,612 INFO [train.py:968] (1/2) Epoch 2, batch 12400, giga_loss[loss=0.363, simple_loss=0.4136, pruned_loss=0.1562, over 28877.00 frames. ], tot_loss[loss=0.3887, simple_loss=0.4292, pruned_loss=0.1741, over 5658865.11 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.4173, pruned_loss=0.1529, over 5696533.36 frames. ], giga_tot_loss[loss=0.3931, simple_loss=0.4311, pruned_loss=0.1776, over 5651458.27 frames. ], batch size: 112, lr: 1.44e-02, grad_scale: 8.0
+2023-03-01 02:22:19,433 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-01 02:22:45,911 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:22:48,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:22:53,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:22:56,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:02,684 INFO [train.py:968] (1/2) Epoch 2, batch 12450, libri_loss[loss=0.2639, simple_loss=0.3315, pruned_loss=0.09812, over 29652.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4285, pruned_loss=0.1726, over 5675802.21 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4166, pruned_loss=0.1524, over 5702501.89 frames. ], giga_tot_loss[loss=0.392, simple_loss=0.431, pruned_loss=0.1765, over 5663641.36 frames. ], batch size: 73, lr: 1.44e-02, grad_scale: 4.0
+2023-03-01 02:23:09,035 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=58126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:13,800 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:25,570 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:25,880 INFO [optim.py:369] (1/2) 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 02:23:39,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:42,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:51,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:23:53,117 INFO [train.py:968] (1/2) Epoch 2, batch 12500, giga_loss[loss=0.4127, simple_loss=0.4492, pruned_loss=0.1881, over 28699.00 frames. ], tot_loss[loss=0.3854, simple_loss=0.4276, pruned_loss=0.1716, over 5688464.32 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4162, pruned_loss=0.1521, over 5706782.88 frames. ], giga_tot_loss[loss=0.3906, simple_loss=0.4302, pruned_loss=0.1754, over 5674219.69 frames. ], batch size: 284, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:24:01,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 02:24:11,854 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:24:40,914 INFO [train.py:968] (1/2) Epoch 2, batch 12550, giga_loss[loss=0.3489, simple_loss=0.3975, pruned_loss=0.1502, over 28104.00 frames. ], tot_loss[loss=0.387, simple_loss=0.428, pruned_loss=0.173, over 5680338.04 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4164, pruned_loss=0.1523, over 5711207.92 frames. ], giga_tot_loss[loss=0.3916, simple_loss=0.4303, pruned_loss=0.1765, over 5664056.74 frames. ], batch size: 77, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:24:51,313 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6158, 1.6267, 3.7639, 2.9453], device='cuda:1'), covar=tensor([0.1381, 0.1278, 0.0280, 0.0410], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0483, 0.0637, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 02:24:51,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:25:06,359 INFO [optim.py:369] (1/2) 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 02:25:28,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:25:33,550 INFO [train.py:968] (1/2) Epoch 2, batch 12600, giga_loss[loss=0.4484, simple_loss=0.4513, pruned_loss=0.2228, over 26531.00 frames. ], tot_loss[loss=0.3863, simple_loss=0.4265, pruned_loss=0.173, over 5676221.98 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4165, pruned_loss=0.1523, over 5712184.88 frames. ], giga_tot_loss[loss=0.39, simple_loss=0.4283, pruned_loss=0.1758, over 5662597.67 frames. ], batch size: 555, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:26:14,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:26:16,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:26:23,216 INFO [train.py:968] (1/2) Epoch 2, batch 12650, giga_loss[loss=0.33, simple_loss=0.3838, pruned_loss=0.1381, over 28957.00 frames. ], tot_loss[loss=0.3825, simple_loss=0.4227, pruned_loss=0.1711, over 5677631.14 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4165, pruned_loss=0.1522, over 5705772.30 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.4243, pruned_loss=0.1737, over 5672016.37 frames. ], batch size: 164, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:26:44,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3477, 1.6346, 1.1378, 1.3467], device='cuda:1'), covar=tensor([0.1073, 0.0444, 0.0554, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0217, 0.0221, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0032], device='cuda:1')
+2023-03-01 02:26:47,724 INFO [optim.py:369] (1/2) 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 02:26:49,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:11,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4970, 2.6076, 1.4259, 1.3466], device='cuda:1'), covar=tensor([0.0842, 0.0591, 0.0841, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0461, 0.0346, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 02:27:15,578 INFO [train.py:968] (1/2) Epoch 2, batch 12700, giga_loss[loss=0.3777, simple_loss=0.4219, pruned_loss=0.1668, over 28755.00 frames. ], tot_loss[loss=0.3819, simple_loss=0.4217, pruned_loss=0.171, over 5682480.79 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4169, pruned_loss=0.1525, over 5707136.50 frames. ], giga_tot_loss[loss=0.3843, simple_loss=0.4226, pruned_loss=0.173, over 5676652.11 frames. ], batch size: 284, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:27:23,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:25,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:54,934 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:27:57,859 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:28:01,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:28:09,300 INFO [train.py:968] (1/2) Epoch 2, batch 12750, giga_loss[loss=0.4488, simple_loss=0.4568, pruned_loss=0.2205, over 26616.00 frames. ], tot_loss[loss=0.3817, simple_loss=0.4211, pruned_loss=0.1712, over 5689477.06 frames. ], libri_tot_loss[loss=0.3604, simple_loss=0.4165, pruned_loss=0.1521, over 5709787.97 frames. ], giga_tot_loss[loss=0.3845, simple_loss=0.4222, pruned_loss=0.1734, over 5682056.42 frames. ], batch size: 555, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:28:15,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 02:28:32,116 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:28:33,840 INFO [optim.py:369] (1/2) 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 02:29:00,036 INFO [train.py:968] (1/2) Epoch 2, batch 12800, giga_loss[loss=0.3919, simple_loss=0.4305, pruned_loss=0.1767, over 27592.00 frames. ], tot_loss[loss=0.3793, simple_loss=0.4203, pruned_loss=0.1691, over 5684292.79 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4163, pruned_loss=0.152, over 5712386.78 frames. ], giga_tot_loss[loss=0.382, simple_loss=0.4215, pruned_loss=0.1712, over 5675698.87 frames. ], batch size: 472, lr: 1.43e-02, grad_scale: 8.0
+2023-03-01 02:29:37,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:29:55,477 INFO [train.py:968] (1/2) Epoch 2, batch 12850, giga_loss[loss=0.4206, simple_loss=0.449, pruned_loss=0.1961, over 28580.00 frames. ], tot_loss[loss=0.3722, simple_loss=0.4166, pruned_loss=0.1639, over 5686150.13 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.4155, pruned_loss=0.1515, over 5717792.30 frames. ], giga_tot_loss[loss=0.3757, simple_loss=0.4183, pruned_loss=0.1665, over 5673825.87 frames. ], batch size: 307, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:30:13,189 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=58534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:30:16,938 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=58537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:30:20,071 INFO [optim.py:369] (1/2) 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 02:30:43,649 INFO [train.py:968] (1/2) Epoch 2, batch 12900, giga_loss[loss=0.3096, simple_loss=0.3795, pruned_loss=0.1198, over 28657.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4133, pruned_loss=0.1606, over 5677495.79 frames. ], libri_tot_loss[loss=0.359, simple_loss=0.4151, pruned_loss=0.1515, over 5720032.44 frames. ], giga_tot_loss[loss=0.3708, simple_loss=0.4151, pruned_loss=0.1633, over 5663338.21 frames. ], batch size: 242, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:31:38,099 INFO [train.py:968] (1/2) Epoch 2, batch 12950, giga_loss[loss=0.3051, simple_loss=0.3469, pruned_loss=0.1317, over 24117.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.407, pruned_loss=0.1547, over 5673764.38 frames. ], libri_tot_loss[loss=0.3583, simple_loss=0.4144, pruned_loss=0.1511, over 5724099.48 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.4089, pruned_loss=0.1572, over 5658205.47 frames. ], batch size: 705, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:32:07,978 INFO [optim.py:369] (1/2) 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 02:32:12,260 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:32:14,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:32:35,017 INFO [train.py:968] (1/2) Epoch 2, batch 13000, giga_loss[loss=0.2937, simple_loss=0.3695, pruned_loss=0.1089, over 28628.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4022, pruned_loss=0.15, over 5679937.10 frames. ], libri_tot_loss[loss=0.3582, simple_loss=0.414, pruned_loss=0.1512, over 5728778.03 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4038, pruned_loss=0.1519, over 5661472.09 frames. ], batch size: 242, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:32:41,526 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:33:22,773 INFO [train.py:968] (1/2) Epoch 2, batch 13050, giga_loss[loss=0.3772, simple_loss=0.4275, pruned_loss=0.1634, over 28776.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.4008, pruned_loss=0.1468, over 5672947.90 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4134, pruned_loss=0.1512, over 5725290.12 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4022, pruned_loss=0.1483, over 5659660.76 frames. ], batch size: 284, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:33:48,373 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 13100, giga_loss[loss=0.3227, simple_loss=0.3946, pruned_loss=0.1254, over 28950.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3993, pruned_loss=0.145, over 5668352.40 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.4126, pruned_loss=0.1508, over 5730115.72 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.4007, pruned_loss=0.1464, over 5651865.41 frames. ], batch size: 136, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:35:12,637 INFO [train.py:968] (1/2) Epoch 2, batch 13150, giga_loss[loss=0.2896, simple_loss=0.3628, pruned_loss=0.1082, over 28905.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3971, pruned_loss=0.1427, over 5675446.06 frames. ], libri_tot_loss[loss=0.3567, simple_loss=0.4122, pruned_loss=0.1506, over 5733405.45 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3984, pruned_loss=0.1438, over 5658140.17 frames. ], batch size: 174, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:35:36,032 INFO [optim.py:369] (1/2) 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 02:35:53,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-01 02:36:07,858 INFO [train.py:968] (1/2) Epoch 2, batch 13200, giga_loss[loss=0.3255, simple_loss=0.3856, pruned_loss=0.1327, over 28970.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3933, pruned_loss=0.1399, over 5671671.41 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4118, pruned_loss=0.1503, over 5735873.73 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3944, pruned_loss=0.1409, over 5654921.58 frames. ], batch size: 213, lr: 1.43e-02, grad_scale: 8.0
+2023-03-01 02:36:50,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:36:53,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:37:00,104 INFO [train.py:968] (1/2) Epoch 2, batch 13250, libri_loss[loss=0.3288, simple_loss=0.3902, pruned_loss=0.1337, over 29241.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3912, pruned_loss=0.1385, over 5665712.52 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4116, pruned_loss=0.1503, over 5730454.76 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3918, pruned_loss=0.139, over 5656132.03 frames. ], batch size: 94, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:37:11,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3004, 1.4262, 1.0401, 1.3647], device='cuda:1'), covar=tensor([0.1077, 0.0435, 0.0545, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0214, 0.0220, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:1')
+2023-03-01 02:37:23,351 INFO [optim.py:369] (1/2) 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 02:37:47,030 INFO [train.py:968] (1/2) Epoch 2, batch 13300, libri_loss[loss=0.3111, simple_loss=0.3645, pruned_loss=0.1288, over 29341.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3908, pruned_loss=0.1382, over 5674998.79 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4094, pruned_loss=0.1492, over 5736551.58 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3925, pruned_loss=0.1392, over 5658711.11 frames. ], batch size: 67, lr: 1.43e-02, grad_scale: 4.0
+2023-03-01 02:38:37,214 INFO [train.py:968] (1/2) Epoch 2, batch 13350, giga_loss[loss=0.3724, simple_loss=0.4141, pruned_loss=0.1653, over 28051.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3883, pruned_loss=0.1362, over 5670356.44 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.4087, pruned_loss=0.1489, over 5737282.75 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3898, pruned_loss=0.1369, over 5655272.60 frames. ], batch size: 412, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:38:47,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7288, 1.5650, 1.5401, 1.5154], device='cuda:1'), covar=tensor([0.0606, 0.1085, 0.1145, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0799, 0.0613, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 02:38:53,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 02:38:59,826 INFO [optim.py:369] (1/2) 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 02:39:08,074 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:09,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-01 02:39:10,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:10,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:15,410 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:26,724 INFO [train.py:968] (1/2) Epoch 2, batch 13400, giga_loss[loss=0.3621, simple_loss=0.3894, pruned_loss=0.1674, over 24165.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3856, pruned_loss=0.1337, over 5677980.74 frames. ], libri_tot_loss[loss=0.3518, simple_loss=0.4072, pruned_loss=0.1482, over 5743039.53 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3872, pruned_loss=0.1343, over 5657564.19 frames. ], batch size: 705, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:39:32,256 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:42,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:39:45,703 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 13450, giga_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1162, over 28542.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3817, pruned_loss=0.1307, over 5675910.52 frames. ], libri_tot_loss[loss=0.3519, simple_loss=0.4071, pruned_loss=0.1483, over 5745971.73 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3826, pruned_loss=0.1308, over 5655854.57 frames. ], batch size: 336, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:40:28,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4963, 1.4763, 1.1314, 1.1253], device='cuda:1'), covar=tensor([0.0667, 0.0590, 0.0942, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0523, 0.0558, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 02:40:47,632 INFO [optim.py:369] (1/2) 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 02:41:13,588 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 13500, giga_loss[loss=0.3483, simple_loss=0.3963, pruned_loss=0.1502, over 28321.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5665142.47 frames. ], libri_tot_loss[loss=0.3507, simple_loss=0.406, pruned_loss=0.1476, over 5748185.75 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.379, pruned_loss=0.1291, over 5644669.67 frames. ], batch size: 368, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:41:29,706 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.64 vs. limit=5.0
+2023-03-01 02:42:10,076 INFO [train.py:968] (1/2) Epoch 2, batch 13550, giga_loss[loss=0.2909, simple_loss=0.3567, pruned_loss=0.1125, over 28234.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3774, pruned_loss=0.1297, over 5658481.11 frames. ], libri_tot_loss[loss=0.3502, simple_loss=0.4056, pruned_loss=0.1474, over 5749840.21 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3782, pruned_loss=0.1299, over 5640048.45 frames. ], batch size: 77, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:42:28,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 02:42:31,518 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 02:42:38,714 INFO [optim.py:369] (1/2) 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 02:42:42,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-01 02:43:13,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5763, 1.6794, 1.5133, 0.9132], device='cuda:1'), covar=tensor([0.0485, 0.0323, 0.0295, 0.0496], device='cuda:1'), in_proj_covar=tensor([0.1030, 0.0710, 0.0788, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 02:43:13,732 INFO [train.py:968] (1/2) Epoch 2, batch 13600, giga_loss[loss=0.342, simple_loss=0.4031, pruned_loss=0.1404, over 28947.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3778, pruned_loss=0.1302, over 5640987.53 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.405, pruned_loss=0.147, over 5751539.38 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3788, pruned_loss=0.1305, over 5624084.65 frames. ], batch size: 186, lr: 1.42e-02, grad_scale: 8.0
+2023-03-01 02:43:31,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4393, 1.6245, 1.3041, 1.4127], device='cuda:1'), covar=tensor([0.1114, 0.0420, 0.0526, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0212, 0.0218, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:1')
+2023-03-01 02:43:43,781 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:43:57,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-01 02:44:09,490 INFO [train.py:968] (1/2) Epoch 2, batch 13650, giga_loss[loss=0.3113, simple_loss=0.3846, pruned_loss=0.1189, over 28955.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3814, pruned_loss=0.1309, over 5655582.92 frames. ], libri_tot_loss[loss=0.3489, simple_loss=0.4045, pruned_loss=0.1467, over 5755621.91 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3818, pruned_loss=0.1309, over 5635318.86 frames. ], batch size: 164, lr: 1.42e-02, grad_scale: 8.0
+2023-03-01 02:44:42,619 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 13700, libri_loss[loss=0.2648, simple_loss=0.3278, pruned_loss=0.1009, over 29505.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3818, pruned_loss=0.1307, over 5654372.06 frames. ], libri_tot_loss[loss=0.3482, simple_loss=0.4038, pruned_loss=0.1463, over 5758953.41 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3822, pruned_loss=0.1307, over 5632626.66 frames. ], batch size: 70, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:46:12,030 INFO [train.py:968] (1/2) Epoch 2, batch 13750, giga_loss[loss=0.2848, simple_loss=0.3516, pruned_loss=0.109, over 28747.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3823, pruned_loss=0.1311, over 5652345.37 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.4032, pruned_loss=0.1457, over 5758849.40 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3826, pruned_loss=0.1312, over 5631211.29 frames. ], batch size: 119, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:46:48,891 INFO [optim.py:369] (1/2) 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 02:46:54,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:47:22,866 INFO [train.py:968] (1/2) Epoch 2, batch 13800, giga_loss[loss=0.3778, simple_loss=0.4113, pruned_loss=0.1721, over 26791.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3804, pruned_loss=0.1295, over 5656123.66 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.4032, pruned_loss=0.1457, over 5758849.40 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3806, pruned_loss=0.1295, over 5639674.73 frames. ], batch size: 555, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:47:43,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-01 02:48:23,437 INFO [train.py:968] (1/2) Epoch 2, batch 13850, giga_loss[loss=0.272, simple_loss=0.3515, pruned_loss=0.09625, over 28649.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3781, pruned_loss=0.1265, over 5648403.30 frames. ], libri_tot_loss[loss=0.347, simple_loss=0.4028, pruned_loss=0.1456, over 5752901.30 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3782, pruned_loss=0.1262, over 5637530.16 frames. ], batch size: 262, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:48:49,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2695, 1.4123, 1.4641, 1.3451], device='cuda:1'), covar=tensor([0.0777, 0.0885, 0.0924, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0809, 0.0630, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 02:48:53,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:48:53,862 INFO [optim.py:369] (1/2) 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 02:49:14,869 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:49:29,820 INFO [train.py:968] (1/2) Epoch 2, batch 13900, giga_loss[loss=0.3309, simple_loss=0.3573, pruned_loss=0.1522, over 24355.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3758, pruned_loss=0.1258, over 5653347.41 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.4024, pruned_loss=0.1455, over 5756359.88 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3754, pruned_loss=0.1252, over 5639023.08 frames. ], batch size: 705, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:49:46,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0079, 3.3051, 3.7568, 1.8095], device='cuda:1'), covar=tensor([0.0456, 0.0461, 0.0738, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.0736, 0.0528, 0.0766, 0.0523], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 02:50:00,283 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:50:04,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:50:34,719 INFO [train.py:968] (1/2) Epoch 2, batch 13950, giga_loss[loss=0.3619, simple_loss=0.4089, pruned_loss=0.1575, over 28529.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3743, pruned_loss=0.1259, over 5660334.67 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.4018, pruned_loss=0.1452, over 5758600.99 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3741, pruned_loss=0.1254, over 5645657.99 frames. ], batch size: 336, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:50:42,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:51:31,350 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 02:51:32,232 INFO [train.py:968] (1/2) Epoch 2, batch 14000, giga_loss[loss=0.3169, simple_loss=0.3796, pruned_loss=0.1271, over 28960.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3728, pruned_loss=0.1253, over 5669518.83 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.4006, pruned_loss=0.1446, over 5762948.37 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3728, pruned_loss=0.1247, over 5650731.37 frames. ], batch size: 186, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:51:40,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:51:52,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:51:55,570 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:52:13,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5081, 1.4873, 2.8251, 2.4501], device='cuda:1'), covar=tensor([0.1275, 0.1238, 0.0431, 0.1245], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0488, 0.0606, 0.0489], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 02:52:30,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:52:31,366 INFO [train.py:968] (1/2) Epoch 2, batch 14050, giga_loss[loss=0.313, simple_loss=0.3786, pruned_loss=0.1237, over 28067.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3749, pruned_loss=0.1254, over 5681252.46 frames. ], libri_tot_loss[loss=0.3444, simple_loss=0.4002, pruned_loss=0.1443, over 5767348.14 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3744, pruned_loss=0.1246, over 5659186.83 frames. ], batch size: 412, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:53:05,681 INFO [optim.py:369] (1/2) 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 02:53:36,566 INFO [train.py:968] (1/2) Epoch 2, batch 14100, giga_loss[loss=0.2916, simple_loss=0.3677, pruned_loss=0.1078, over 29036.00 frames. ], tot_loss[loss=0.316, simple_loss=0.378, pruned_loss=0.127, over 5688431.26 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.3997, pruned_loss=0.1443, over 5769631.22 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3775, pruned_loss=0.1259, over 5666949.61 frames. ], batch size: 155, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:54:41,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:54:42,908 INFO [train.py:968] (1/2) Epoch 2, batch 14150, giga_loss[loss=0.3, simple_loss=0.3635, pruned_loss=0.1183, over 29121.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1242, over 5679249.86 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3993, pruned_loss=0.1441, over 5761746.46 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.373, pruned_loss=0.123, over 5666333.94 frames. ], batch size: 200, lr: 1.42e-02, grad_scale: 4.0
+2023-03-01 02:54:44,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:55:20,985 INFO [optim.py:369] (1/2) 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 02:55:26,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:55:50,628 INFO [train.py:968] (1/2) Epoch 2, batch 14200, giga_loss[loss=0.3273, simple_loss=0.3879, pruned_loss=0.1334, over 28936.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3747, pruned_loss=0.1254, over 5684044.33 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.3988, pruned_loss=0.1438, over 5762285.27 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3739, pruned_loss=0.1241, over 5671180.55 frames. ], batch size: 186, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 02:56:26,437 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=59895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 02:57:01,564 INFO [train.py:968] (1/2) Epoch 2, batch 14250, giga_loss[loss=0.3688, simple_loss=0.3949, pruned_loss=0.1714, over 24701.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3774, pruned_loss=0.1267, over 5663715.90 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.399, pruned_loss=0.1439, over 5765154.81 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.376, pruned_loss=0.1252, over 5648828.34 frames. ], batch size: 705, lr: 1.41e-02, grad_scale: 2.0
+2023-03-01 02:57:17,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:58:06,357 INFO [train.py:968] (1/2) Epoch 2, batch 14300, giga_loss[loss=0.3272, simple_loss=0.3781, pruned_loss=0.1382, over 26800.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3798, pruned_loss=0.1251, over 5667606.24 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3981, pruned_loss=0.1433, over 5768857.58 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3791, pruned_loss=0.124, over 5650079.54 frames. ], batch size: 555, lr: 1.41e-02, grad_scale: 2.0
+2023-03-01 02:58:51,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5573, 1.5721, 1.0947, 1.3833], device='cuda:1'), covar=tensor([0.0764, 0.0697, 0.1034, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0526, 0.0556, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 02:59:08,820 INFO [train.py:968] (1/2) Epoch 2, batch 14350, giga_loss[loss=0.2909, simple_loss=0.3739, pruned_loss=0.1039, over 28548.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3792, pruned_loss=0.1235, over 5644461.74 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3982, pruned_loss=0.1434, over 5757867.84 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3781, pruned_loss=0.1222, over 5638675.08 frames. ], batch size: 336, lr: 1.41e-02, grad_scale: 2.0
+2023-03-01 02:59:42,219 INFO [optim.py:369] (1/2) 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 03:00:12,256 INFO [train.py:968] (1/2) Epoch 2, batch 14400, giga_loss[loss=0.3177, simple_loss=0.3887, pruned_loss=0.1234, over 28692.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3792, pruned_loss=0.1223, over 5657207.32 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3984, pruned_loss=0.1435, over 5758521.58 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.378, pruned_loss=0.1211, over 5651161.52 frames. ], batch size: 262, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:00:21,175 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:00:25,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:00:49,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4685, 1.9470, 1.8539, 1.9074], device='cuda:1'), covar=tensor([0.0669, 0.1686, 0.1118, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0796, 0.0625, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 03:01:02,769 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:01:17,501 INFO [train.py:968] (1/2) Epoch 2, batch 14450, giga_loss[loss=0.2995, simple_loss=0.3651, pruned_loss=0.1169, over 29087.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3814, pruned_loss=0.1249, over 5666106.63 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3982, pruned_loss=0.1432, over 5761489.03 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3803, pruned_loss=0.1238, over 5656971.94 frames. ], batch size: 200, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:01:53,551 INFO [optim.py:369] (1/2) 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 03:02:25,675 INFO [train.py:968] (1/2) Epoch 2, batch 14500, libri_loss[loss=0.3696, simple_loss=0.4203, pruned_loss=0.1594, over 29544.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3801, pruned_loss=0.1253, over 5666110.10 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3983, pruned_loss=0.1432, over 5763894.19 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3789, pruned_loss=0.1241, over 5654988.39 frames. ], batch size: 83, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:03:16,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0942, 1.1571, 1.1565, 1.1644], device='cuda:1'), covar=tensor([0.0564, 0.0649, 0.0862, 0.0633], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0793, 0.0619, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 03:03:16,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9570, 1.0191, 0.8484, 0.1766], device='cuda:1'), covar=tensor([0.0749, 0.0817, 0.0893, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.1099, 0.1097, 0.1135, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 03:03:36,479 INFO [train.py:968] (1/2) Epoch 2, batch 14550, libri_loss[loss=0.2749, simple_loss=0.3378, pruned_loss=0.106, over 28197.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3808, pruned_loss=0.126, over 5674093.17 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3972, pruned_loss=0.1423, over 5766873.06 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3801, pruned_loss=0.1252, over 5658931.89 frames. ], batch size: 62, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:04:16,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:05:02,459 INFO [train.py:968] (1/2) Epoch 2, batch 14600, giga_loss[loss=0.3155, simple_loss=0.3765, pruned_loss=0.1272, over 29142.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3754, pruned_loss=0.1223, over 5676092.88 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3971, pruned_loss=0.1422, over 5764344.80 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3747, pruned_loss=0.1215, over 5664696.52 frames. ], batch size: 200, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:05:06,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:05:23,484 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:06:09,438 INFO [train.py:968] (1/2) Epoch 2, batch 14650, giga_loss[loss=0.3242, simple_loss=0.3895, pruned_loss=0.1294, over 28510.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3737, pruned_loss=0.1216, over 5664439.94 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3964, pruned_loss=0.1418, over 5758355.03 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3732, pruned_loss=0.1209, over 5658518.58 frames. ], batch size: 370, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:06:51,952 INFO [optim.py:369] (1/2) 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 03:06:58,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7316, 3.1048, 1.9284, 1.5105], device='cuda:1'), covar=tensor([0.0893, 0.0526, 0.0794, 0.1425], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0442, 0.0344, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 03:07:14,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3791, 1.9677, 1.5027, 1.4963], device='cuda:1'), covar=tensor([0.0973, 0.0478, 0.0465, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0315, 0.0209, 0.0218, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:1')
+2023-03-01 03:07:21,452 INFO [train.py:968] (1/2) Epoch 2, batch 14700, giga_loss[loss=0.3023, simple_loss=0.3636, pruned_loss=0.1206, over 28929.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3728, pruned_loss=0.1218, over 5675051.37 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.3963, pruned_loss=0.1417, over 5760781.34 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3721, pruned_loss=0.1209, over 5666476.32 frames. ], batch size: 112, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:08:15,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:08:20,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:08:22,188 INFO [train.py:968] (1/2) Epoch 2, batch 14750, libri_loss[loss=0.3179, simple_loss=0.3786, pruned_loss=0.1286, over 29500.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3789, pruned_loss=0.1255, over 5687345.81 frames. ], libri_tot_loss[loss=0.3396, simple_loss=0.3961, pruned_loss=0.1416, over 5763876.23 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3778, pruned_loss=0.1242, over 5674604.17 frames. ], batch size: 81, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:08:44,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:02,209 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:02,231 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:02,548 INFO [optim.py:369] (1/2) 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 03:09:18,430 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:09:33,325 INFO [train.py:968] (1/2) Epoch 2, batch 14800, giga_loss[loss=0.3755, simple_loss=0.407, pruned_loss=0.172, over 28049.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3793, pruned_loss=0.1262, over 5682833.06 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3956, pruned_loss=0.1412, over 5765888.86 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3785, pruned_loss=0.1251, over 5669127.25 frames. ], batch size: 412, lr: 1.41e-02, grad_scale: 8.0
+2023-03-01 03:09:38,262 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:10:36,064 INFO [train.py:968] (1/2) Epoch 2, batch 14850, giga_loss[loss=0.3358, simple_loss=0.398, pruned_loss=0.1368, over 28544.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3787, pruned_loss=0.1275, over 5674922.98 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.3948, pruned_loss=0.1408, over 5759890.88 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3784, pruned_loss=0.1268, over 5667609.16 frames. ], batch size: 336, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:11:02,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-01 03:11:14,332 INFO [optim.py:369] (1/2) 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 03:11:38,688 INFO [train.py:968] (1/2) Epoch 2, batch 14900, giga_loss[loss=0.3557, simple_loss=0.4026, pruned_loss=0.1545, over 28366.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3799, pruned_loss=0.1291, over 5669354.62 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.3945, pruned_loss=0.1408, over 5754919.25 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3795, pruned_loss=0.1282, over 5666135.48 frames. ], batch size: 369, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:12:12,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=60591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:12:49,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:12:52,100 INFO [train.py:968] (1/2) Epoch 2, batch 14950, giga_loss[loss=0.3293, simple_loss=0.3892, pruned_loss=0.1347, over 28743.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3812, pruned_loss=0.129, over 5670027.47 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3945, pruned_loss=0.1408, over 5757467.80 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3807, pruned_loss=0.1281, over 5663861.68 frames. ], batch size: 243, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:13:28,434 INFO [optim.py:369] (1/2) 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 03:13:45,429 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 15000, giga_loss[loss=0.3004, simple_loss=0.373, pruned_loss=0.114, over 28844.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3825, pruned_loss=0.1289, over 5663941.24 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3944, pruned_loss=0.1407, over 5748260.48 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.382, pruned_loss=0.1279, over 5664741.07 frames. ], batch size: 227, lr: 1.41e-02, grad_scale: 4.0
+2023-03-01 03:14:08,120 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 03:14:17,215 INFO [train.py:1012] (1/2) Epoch 2, validation: loss=0.2578, simple_loss=0.3473, pruned_loss=0.08416, over 944034.00 frames. 
+2023-03-01 03:14:17,216 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 03:15:38,802 INFO [train.py:968] (1/2) Epoch 2, batch 15050, giga_loss[loss=0.2717, simple_loss=0.3404, pruned_loss=0.1015, over 28858.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3792, pruned_loss=0.1264, over 5667166.67 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3939, pruned_loss=0.1405, over 5751528.46 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.379, pruned_loss=0.1257, over 5663541.57 frames. ], batch size: 145, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:16:20,768 INFO [optim.py:369] (1/2) 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 03:16:37,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:16:40,905 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 15100, giga_loss[loss=0.2739, simple_loss=0.3385, pruned_loss=0.1046, over 28807.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3742, pruned_loss=0.1244, over 5690887.40 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3932, pruned_loss=0.1402, over 5756635.41 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1237, over 5680974.47 frames. ], batch size: 174, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:16:52,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6568, 1.5486, 1.4858, 2.0597], device='cuda:1'), covar=tensor([0.1734, 0.1665, 0.1465, 0.1714], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0769, 0.0871, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 03:17:19,655 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:17:31,910 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:17:33,010 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-01 03:17:34,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:17:43,433 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 15150, giga_loss[loss=0.2669, simple_loss=0.3364, pruned_loss=0.09873, over 28746.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3684, pruned_loss=0.1217, over 5685314.35 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3929, pruned_loss=0.14, over 5757611.93 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3679, pruned_loss=0.1206, over 5674531.09 frames. ], batch size: 92, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:17:55,570 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:18:06,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1772, 1.3475, 1.0524, 1.4010], device='cuda:1'), covar=tensor([0.1105, 0.0443, 0.0525, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0208, 0.0215, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:1')
+2023-03-01 03:18:10,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:18:13,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:18:27,717 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:18:53,597 INFO [train.py:968] (1/2) Epoch 2, batch 15200, giga_loss[loss=0.342, simple_loss=0.3919, pruned_loss=0.1461, over 28744.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3701, pruned_loss=0.1235, over 5687740.99 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3924, pruned_loss=0.1398, over 5763004.89 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3691, pruned_loss=0.1221, over 5671029.45 frames. ], batch size: 262, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:19:14,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4115, 1.2434, 1.1457, 1.6160], device='cuda:1'), covar=tensor([0.1944, 0.1871, 0.1600, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0773, 0.0862, 0.0920], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 03:19:51,929 INFO [train.py:968] (1/2) Epoch 2, batch 15250, giga_loss[loss=0.3667, simple_loss=0.3894, pruned_loss=0.172, over 24442.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3722, pruned_loss=0.1254, over 5684307.87 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3927, pruned_loss=0.1399, over 5765497.26 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3707, pruned_loss=0.1238, over 5667198.95 frames. ], batch size: 705, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:20:31,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2217, 1.3995, 1.1397, 1.0576], device='cuda:1'), covar=tensor([0.0388, 0.0301, 0.0283, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.1019, 0.0706, 0.0787, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 03:20:31,690 INFO [optim.py:369] (1/2) 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 03:20:38,388 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:41,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:52,403 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:55,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:55,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:20:57,580 INFO [train.py:968] (1/2) Epoch 2, batch 15300, giga_loss[loss=0.3171, simple_loss=0.3783, pruned_loss=0.128, over 28932.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3692, pruned_loss=0.123, over 5671041.10 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3927, pruned_loss=0.1399, over 5765795.78 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3678, pruned_loss=0.1217, over 5656912.75 frames. ], batch size: 284, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:21:09,260 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:13,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:19,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:26,919 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:30,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:34,579 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:21:52,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:22:06,012 INFO [train.py:968] (1/2) Epoch 2, batch 15350, giga_loss[loss=0.3003, simple_loss=0.365, pruned_loss=0.1179, over 28861.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3669, pruned_loss=0.1201, over 5672312.29 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3922, pruned_loss=0.1397, over 5766976.58 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3661, pruned_loss=0.1191, over 5659460.71 frames. ], batch size: 174, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:22:09,125 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:22:38,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3471, 1.9001, 1.9525, 1.7913], device='cuda:1'), covar=tensor([0.0843, 0.1787, 0.1149, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0788, 0.0615, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 03:22:42,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6411, 1.4790, 1.2245, 1.3459], device='cuda:1'), covar=tensor([0.0632, 0.0586, 0.0913, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0518, 0.0566, 0.0487], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 03:22:45,165 INFO [optim.py:369] (1/2) 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 03:23:18,322 INFO [train.py:968] (1/2) Epoch 2, batch 15400, giga_loss[loss=0.2741, simple_loss=0.3496, pruned_loss=0.09924, over 28952.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3659, pruned_loss=0.1203, over 5674472.28 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3919, pruned_loss=0.1396, over 5770143.35 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3648, pruned_loss=0.1191, over 5658832.96 frames. ], batch size: 155, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:23:55,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5897, 2.4360, 1.4861, 1.3334], device='cuda:1'), covar=tensor([0.0868, 0.0588, 0.0920, 0.1449], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0436, 0.0339, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 03:24:14,742 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:24:16,742 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=61111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:24:17,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:24:27,301 INFO [train.py:968] (1/2) Epoch 2, batch 15450, giga_loss[loss=0.3387, simple_loss=0.3967, pruned_loss=0.1403, over 28807.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3665, pruned_loss=0.1194, over 5680820.83 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3918, pruned_loss=0.1395, over 5761751.94 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3652, pruned_loss=0.1181, over 5674349.34 frames. ], batch size: 119, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:24:59,824 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:25:21,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6650, 3.0585, 3.3849, 1.4839], device='cuda:1'), covar=tensor([0.0559, 0.0453, 0.0834, 0.1963], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0523, 0.0748, 0.0535], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 03:25:36,217 INFO [train.py:968] (1/2) Epoch 2, batch 15500, giga_loss[loss=0.3759, simple_loss=0.4122, pruned_loss=0.1699, over 28194.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3673, pruned_loss=0.1204, over 5682060.12 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3912, pruned_loss=0.1393, over 5757552.26 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3662, pruned_loss=0.1191, over 5679171.01 frames. ], batch size: 412, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:26:44,802 INFO [train.py:968] (1/2) Epoch 2, batch 15550, giga_loss[loss=0.2793, simple_loss=0.3463, pruned_loss=0.1061, over 28867.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3693, pruned_loss=0.1225, over 5688936.84 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3912, pruned_loss=0.1394, over 5761418.73 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5680872.34 frames. ], batch size: 284, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:27:14,238 INFO [optim.py:369] (1/2) 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 03:27:40,816 INFO [train.py:968] (1/2) Epoch 2, batch 15600, libri_loss[loss=0.2718, simple_loss=0.3289, pruned_loss=0.1073, over 29323.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3664, pruned_loss=0.1205, over 5677008.80 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.39, pruned_loss=0.1387, over 5758347.53 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3653, pruned_loss=0.119, over 5669689.95 frames. ], batch size: 67, lr: 1.40e-02, grad_scale: 8.0
+2023-03-01 03:28:42,164 INFO [train.py:968] (1/2) Epoch 2, batch 15650, giga_loss[loss=0.3286, simple_loss=0.398, pruned_loss=0.1296, over 28521.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5667426.20 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3897, pruned_loss=0.1385, over 5759015.60 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3665, pruned_loss=0.1179, over 5659389.51 frames. ], batch size: 336, lr: 1.40e-02, grad_scale: 8.0
+2023-03-01 03:29:19,007 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 15700, libri_loss[loss=0.3463, simple_loss=0.4098, pruned_loss=0.1414, over 28660.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5666639.04 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3898, pruned_loss=0.1385, over 5761186.89 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3702, pruned_loss=0.1195, over 5656057.73 frames. ], batch size: 106, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:30:39,612 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-01 03:30:47,482 INFO [train.py:968] (1/2) Epoch 2, batch 15750, giga_loss[loss=0.3105, simple_loss=0.3767, pruned_loss=0.1222, over 28950.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.374, pruned_loss=0.1226, over 5670909.22 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3898, pruned_loss=0.1386, over 5764245.08 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3725, pruned_loss=0.1209, over 5657282.05 frames. ], batch size: 213, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:31:05,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6191, 4.1750, 2.4929, 2.2548], device='cuda:1'), covar=tensor([0.0690, 0.0241, 0.0700, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0432, 0.0336, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 03:31:07,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2570, 1.3184, 1.1477, 1.5878], device='cuda:1'), covar=tensor([0.2316, 0.2021, 0.1835, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0768, 0.0858, 0.0907], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 03:31:22,495 INFO [optim.py:369] (1/2) 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 03:31:24,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4836, 1.4007, 1.1314, 1.1160], device='cuda:1'), covar=tensor([0.0699, 0.0636, 0.1074, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0515, 0.0561, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 03:31:28,493 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=61454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:31:46,484 INFO [train.py:968] (1/2) Epoch 2, batch 15800, giga_loss[loss=0.3765, simple_loss=0.4092, pruned_loss=0.1719, over 26804.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5661492.73 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3892, pruned_loss=0.1382, over 5766331.26 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3731, pruned_loss=0.1221, over 5645861.36 frames. ], batch size: 555, lr: 1.40e-02, grad_scale: 4.0
+2023-03-01 03:32:04,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8257, 1.6988, 1.5557, 1.6025], device='cuda:1'), covar=tensor([0.0659, 0.1016, 0.1017, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0800, 0.0620, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 03:32:09,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:32:13,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 03:32:48,751 INFO [train.py:968] (1/2) Epoch 2, batch 15850, giga_loss[loss=0.2758, simple_loss=0.3504, pruned_loss=0.1006, over 28981.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1223, over 5659967.56 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3891, pruned_loss=0.1381, over 5765526.84 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.371, pruned_loss=0.1209, over 5646478.63 frames. ], batch size: 136, lr: 1.40e-02, grad_scale: 2.0
+2023-03-01 03:33:05,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3468, 1.8795, 1.4405, 1.5776], device='cuda:1'), covar=tensor([0.1105, 0.0429, 0.0503, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0310, 0.0206, 0.0214, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0022, 0.0020, 0.0033], device='cuda:1')
+2023-03-01 03:33:29,057 INFO [optim.py:369] (1/2) 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 03:33:54,009 INFO [train.py:968] (1/2) Epoch 2, batch 15900, giga_loss[loss=0.3134, simple_loss=0.3782, pruned_loss=0.1243, over 28497.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3709, pruned_loss=0.1211, over 5662178.06 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3893, pruned_loss=0.1382, over 5766242.91 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3694, pruned_loss=0.1195, over 5648156.67 frames. ], batch size: 336, lr: 1.39e-02, grad_scale: 2.0
+2023-03-01 03:34:49,099 INFO [train.py:968] (1/2) Epoch 2, batch 15950, giga_loss[loss=0.2816, simple_loss=0.3525, pruned_loss=0.1053, over 28892.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1207, over 5678631.78 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3881, pruned_loss=0.1375, over 5771847.11 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.368, pruned_loss=0.1193, over 5658633.90 frames. ], batch size: 186, lr: 1.39e-02, grad_scale: 2.0
+2023-03-01 03:35:02,159 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:35:04,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:35:26,924 INFO [optim.py:369] (1/2) 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 03:35:43,761 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 16000, giga_loss[loss=0.3626, simple_loss=0.4099, pruned_loss=0.1577, over 28619.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3705, pruned_loss=0.1214, over 5680730.19 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3881, pruned_loss=0.1373, over 5773522.72 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3694, pruned_loss=0.1201, over 5661244.86 frames. ], batch size: 307, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:37:02,557 INFO [train.py:968] (1/2) Epoch 2, batch 16050, giga_loss[loss=0.3032, simple_loss=0.3679, pruned_loss=0.1192, over 28646.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3726, pruned_loss=0.1226, over 5677415.35 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3884, pruned_loss=0.1375, over 5773938.45 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3713, pruned_loss=0.1213, over 5660892.02 frames. ], batch size: 242, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:37:08,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3202, 1.3248, 1.1852, 1.4064], device='cuda:1'), covar=tensor([0.1909, 0.1874, 0.1617, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0762, 0.0857, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 03:37:37,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7106, 1.4864, 3.4165, 2.7282], device='cuda:1'), covar=tensor([0.1510, 0.1554, 0.0369, 0.0513], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0492, 0.0612, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 03:37:44,498 INFO [optim.py:369] (1/2) 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 03:38:09,294 INFO [train.py:968] (1/2) Epoch 2, batch 16100, giga_loss[loss=0.2794, simple_loss=0.3449, pruned_loss=0.1069, over 28517.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3735, pruned_loss=0.1242, over 5659999.57 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3882, pruned_loss=0.1374, over 5764588.26 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5653390.12 frames. ], batch size: 85, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:39:07,442 INFO [train.py:968] (1/2) Epoch 2, batch 16150, giga_loss[loss=0.3678, simple_loss=0.4232, pruned_loss=0.1562, over 28589.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3773, pruned_loss=0.1264, over 5644002.37 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3881, pruned_loss=0.1372, over 5748457.10 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1251, over 5651213.01 frames. ], batch size: 307, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:39:20,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:39:22,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1696, 1.3912, 1.1263, 0.8355], device='cuda:1'), covar=tensor([0.0488, 0.0328, 0.0306, 0.0457], device='cuda:1'), in_proj_covar=tensor([0.1003, 0.0710, 0.0765, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 03:39:30,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3882, 1.9666, 1.3324, 1.1527], device='cuda:1'), covar=tensor([0.0923, 0.0604, 0.0877, 0.1537], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0435, 0.0335, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 03:39:41,118 INFO [optim.py:369] (1/2) 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 03:39:46,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5738, 2.0382, 1.6146, 1.6479], device='cuda:1'), covar=tensor([0.1286, 0.1258, 0.1051, 0.0704], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0768, 0.0684, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 03:40:02,134 INFO [train.py:968] (1/2) Epoch 2, batch 16200, giga_loss[loss=0.2871, simple_loss=0.3704, pruned_loss=0.1019, over 28867.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3789, pruned_loss=0.1266, over 5649721.44 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3877, pruned_loss=0.1369, over 5752871.29 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3779, pruned_loss=0.1255, over 5647715.42 frames. ], batch size: 174, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:41:06,889 INFO [train.py:968] (1/2) Epoch 2, batch 16250, giga_loss[loss=0.3469, simple_loss=0.4023, pruned_loss=0.1458, over 28547.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3806, pruned_loss=0.1278, over 5644638.74 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3877, pruned_loss=0.1369, over 5748625.43 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3797, pruned_loss=0.1268, over 5644880.52 frames. ], batch size: 370, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:41:25,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 03:41:35,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7192, 2.7377, 1.5465, 1.4163], device='cuda:1'), covar=tensor([0.0763, 0.0501, 0.0836, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0433, 0.0337, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 03:41:56,948 INFO [optim.py:369] (1/2) 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 03:42:24,054 INFO [train.py:968] (1/2) Epoch 2, batch 16300, giga_loss[loss=0.2784, simple_loss=0.3481, pruned_loss=0.1044, over 29082.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3786, pruned_loss=0.1268, over 5651348.16 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3875, pruned_loss=0.1368, over 5750227.86 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3779, pruned_loss=0.126, over 5649240.02 frames. ], batch size: 100, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:42:28,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:42:33,407 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:43:14,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:43:34,055 INFO [train.py:968] (1/2) Epoch 2, batch 16350, giga_loss[loss=0.2855, simple_loss=0.3623, pruned_loss=0.1044, over 28825.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3759, pruned_loss=0.1247, over 5659832.89 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3875, pruned_loss=0.1368, over 5751765.11 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3753, pruned_loss=0.124, over 5655964.14 frames. ], batch size: 174, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:43:48,657 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 03:44:14,911 INFO [optim.py:369] (1/2) 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 03:44:38,106 INFO [train.py:968] (1/2) Epoch 2, batch 16400, giga_loss[loss=0.3501, simple_loss=0.4029, pruned_loss=0.1486, over 28648.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3752, pruned_loss=0.1249, over 5664416.83 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3871, pruned_loss=0.1366, over 5755471.14 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3748, pruned_loss=0.1241, over 5655653.73 frames. ], batch size: 262, lr: 1.39e-02, grad_scale: 8.0
+2023-03-01 03:44:48,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 03:45:41,092 INFO [train.py:968] (1/2) Epoch 2, batch 16450, giga_loss[loss=0.3122, simple_loss=0.3689, pruned_loss=0.1277, over 28721.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1242, over 5657458.66 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3866, pruned_loss=0.1363, over 5755497.97 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1235, over 5648254.70 frames. ], batch size: 243, lr: 1.39e-02, grad_scale: 8.0
+2023-03-01 03:45:47,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:46:02,040 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:46:18,165 INFO [optim.py:369] (1/2) 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 03:46:23,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-01 03:46:36,709 INFO [train.py:968] (1/2) Epoch 2, batch 16500, giga_loss[loss=0.3331, simple_loss=0.3939, pruned_loss=0.1361, over 28102.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3722, pruned_loss=0.1242, over 5659793.94 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3867, pruned_loss=0.1364, over 5755509.87 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3714, pruned_loss=0.1233, over 5648995.81 frames. ], batch size: 412, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:46:37,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-01 03:46:55,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-01 03:47:35,715 INFO [train.py:968] (1/2) Epoch 2, batch 16550, giga_loss[loss=0.3425, simple_loss=0.3996, pruned_loss=0.1427, over 28816.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1236, over 5663359.24 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3864, pruned_loss=0.1361, over 5749683.10 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1227, over 5657351.27 frames. ], batch size: 145, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:48:16,185 INFO [optim.py:369] (1/2) 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 03:48:45,401 INFO [train.py:968] (1/2) Epoch 2, batch 16600, giga_loss[loss=0.2973, simple_loss=0.3807, pruned_loss=0.1069, over 28956.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3718, pruned_loss=0.1212, over 5674009.64 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3861, pruned_loss=0.1361, over 5751679.23 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3712, pruned_loss=0.1203, over 5666086.55 frames. ], batch size: 284, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:49:33,473 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:49:48,488 INFO [train.py:968] (1/2) Epoch 2, batch 16650, libri_loss[loss=0.4201, simple_loss=0.4414, pruned_loss=0.1994, over 20048.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.373, pruned_loss=0.1193, over 5670260.36 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3862, pruned_loss=0.1363, over 5741232.59 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3723, pruned_loss=0.1181, over 5672937.37 frames. ], batch size: 187, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:50:27,230 INFO [optim.py:369] (1/2) 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 03:50:28,258 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-01 03:50:52,368 INFO [train.py:968] (1/2) Epoch 2, batch 16700, giga_loss[loss=0.2884, simple_loss=0.366, pruned_loss=0.1054, over 28908.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3739, pruned_loss=0.1198, over 5663030.24 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3862, pruned_loss=0.1363, over 5741232.59 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3734, pruned_loss=0.1189, over 5665113.79 frames. ], batch size: 284, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:52:03,108 INFO [train.py:968] (1/2) Epoch 2, batch 16750, giga_loss[loss=0.3356, simple_loss=0.402, pruned_loss=0.1346, over 28297.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3736, pruned_loss=0.12, over 5659938.75 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.386, pruned_loss=0.1361, over 5743925.93 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3732, pruned_loss=0.1193, over 5658227.39 frames. ], batch size: 368, lr: 1.39e-02, grad_scale: 4.0
+2023-03-01 03:52:42,850 INFO [optim.py:369] (1/2) 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 03:53:08,521 INFO [train.py:968] (1/2) Epoch 2, batch 16800, giga_loss[loss=0.3215, simple_loss=0.3828, pruned_loss=0.1301, over 27603.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3736, pruned_loss=0.1208, over 5658990.45 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3854, pruned_loss=0.1358, over 5750579.51 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3733, pruned_loss=0.1199, over 5648410.09 frames. ], batch size: 472, lr: 1.38e-02, grad_scale: 8.0
+2023-03-01 03:53:55,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:54:07,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:54:11,915 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:54:14,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 03:54:14,963 INFO [train.py:968] (1/2) Epoch 2, batch 16850, libri_loss[loss=0.3189, simple_loss=0.3843, pruned_loss=0.1268, over 28496.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.373, pruned_loss=0.1194, over 5658581.04 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.385, pruned_loss=0.1354, over 5743354.26 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3727, pruned_loss=0.1184, over 5652583.64 frames. ], batch size: 106, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:55:02,656 INFO [optim.py:369] (1/2) 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 03:55:27,722 INFO [train.py:968] (1/2) Epoch 2, batch 16900, giga_loss[loss=0.3412, simple_loss=0.3988, pruned_loss=0.1418, over 28975.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3725, pruned_loss=0.1189, over 5661197.92 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3842, pruned_loss=0.135, over 5747483.24 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3726, pruned_loss=0.118, over 5650781.68 frames. ], batch size: 106, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:55:38,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0785, 3.4185, 3.7952, 1.7964], device='cuda:1'), covar=tensor([0.0433, 0.0478, 0.0800, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0523, 0.0749, 0.0531], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 03:56:41,703 INFO [train.py:968] (1/2) Epoch 2, batch 16950, giga_loss[loss=0.3209, simple_loss=0.3907, pruned_loss=0.1255, over 28115.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3763, pruned_loss=0.1207, over 5672372.49 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3839, pruned_loss=0.1347, over 5752051.36 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3765, pruned_loss=0.1199, over 5657184.67 frames. ], batch size: 412, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:57:15,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:20,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:26,462 INFO [optim.py:369] (1/2) 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 03:57:29,552 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:32,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:57:47,489 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 17000, giga_loss[loss=0.289, simple_loss=0.3609, pruned_loss=0.1086, over 28939.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3771, pruned_loss=0.1212, over 5670265.70 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3842, pruned_loss=0.1349, over 5745202.19 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3768, pruned_loss=0.12, over 5662017.62 frames. ], batch size: 112, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 03:57:54,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6390, 2.1178, 1.7106, 1.7259], device='cuda:1'), covar=tensor([0.1513, 0.1441, 0.1165, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0771, 0.0687, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0011, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 03:58:02,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:58:09,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8356, 1.5703, 1.4140, 1.5016], device='cuda:1'), covar=tensor([0.0670, 0.1485, 0.1175, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0800, 0.0607, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 03:58:12,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:58:17,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 03:59:07,077 INFO [train.py:968] (1/2) Epoch 2, batch 17050, libri_loss[loss=0.3352, simple_loss=0.3956, pruned_loss=0.1374, over 29663.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3759, pruned_loss=0.1214, over 5668262.13 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3845, pruned_loss=0.1351, over 5744077.21 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3752, pruned_loss=0.12, over 5661622.60 frames. ], batch size: 91, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:00:01,022 INFO [optim.py:369] (1/2) 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 04:00:01,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-01 04:00:24,368 INFO [train.py:968] (1/2) Epoch 2, batch 17100, giga_loss[loss=0.2763, simple_loss=0.3567, pruned_loss=0.09797, over 28696.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.374, pruned_loss=0.1199, over 5673051.62 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3849, pruned_loss=0.1352, over 5741355.57 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.373, pruned_loss=0.1186, over 5669524.92 frames. ], batch size: 307, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:00:47,613 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:00:47,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0112, 1.9819, 1.4026, 1.6477], device='cuda:1'), covar=tensor([0.0565, 0.0514, 0.0938, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0508, 0.0553, 0.0477], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 04:01:34,476 INFO [train.py:968] (1/2) Epoch 2, batch 17150, giga_loss[loss=0.332, simple_loss=0.3895, pruned_loss=0.1373, over 28947.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3727, pruned_loss=0.1189, over 5672040.17 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3849, pruned_loss=0.1352, over 5744202.53 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3716, pruned_loss=0.1174, over 5664450.10 frames. ], batch size: 284, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:01:43,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:01:46,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:02:18,736 INFO [optim.py:369] (1/2) 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 04:02:27,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:02:43,308 INFO [train.py:968] (1/2) Epoch 2, batch 17200, giga_loss[loss=0.3189, simple_loss=0.3823, pruned_loss=0.1277, over 27536.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3721, pruned_loss=0.1183, over 5671788.60 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3847, pruned_loss=0.135, over 5744392.55 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3713, pruned_loss=0.117, over 5665056.61 frames. ], batch size: 472, lr: 1.38e-02, grad_scale: 8.0
+2023-03-01 04:02:45,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8646, 1.8777, 1.2326, 1.5883], device='cuda:1'), covar=tensor([0.0580, 0.0540, 0.0987, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0502, 0.0548, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 04:03:08,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:03:40,550 INFO [train.py:968] (1/2) Epoch 2, batch 17250, giga_loss[loss=0.3659, simple_loss=0.4185, pruned_loss=0.1566, over 28139.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3753, pruned_loss=0.1204, over 5674014.31 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3842, pruned_loss=0.1346, over 5749668.04 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3747, pruned_loss=0.1194, over 5662088.42 frames. ], batch size: 412, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:04:03,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:04:17,235 INFO [optim.py:369] (1/2) 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 04:04:35,508 INFO [train.py:968] (1/2) Epoch 2, batch 17300, giga_loss[loss=0.2913, simple_loss=0.3708, pruned_loss=0.1059, over 28901.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3758, pruned_loss=0.1214, over 5679387.07 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3837, pruned_loss=0.1342, over 5751936.81 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3755, pruned_loss=0.1205, over 5665339.30 frames. ], batch size: 112, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:04:44,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2608, 3.6961, 4.0057, 1.5795], device='cuda:1'), covar=tensor([0.0456, 0.0430, 0.0698, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0535, 0.0756, 0.0534], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 04:04:49,399 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=62978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:05:37,844 INFO [train.py:968] (1/2) Epoch 2, batch 17350, giga_loss[loss=0.2899, simple_loss=0.3645, pruned_loss=0.1076, over 28711.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3729, pruned_loss=0.1211, over 5671340.90 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3843, pruned_loss=0.1345, over 5751376.59 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3719, pruned_loss=0.1197, over 5659303.04 frames. ], batch size: 242, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:05:58,418 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:02,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:06,949 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:20,679 INFO [optim.py:369] (1/2) 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 04:06:28,653 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:33,730 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:06:36,357 INFO [train.py:968] (1/2) Epoch 2, batch 17400, giga_loss[loss=0.2852, simple_loss=0.3538, pruned_loss=0.1083, over 28623.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3723, pruned_loss=0.1216, over 5666846.62 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3844, pruned_loss=0.1347, over 5753874.59 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3711, pruned_loss=0.1201, over 5652531.19 frames. ], batch size: 242, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:06:48,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4944, 2.5525, 1.5290, 1.2646], device='cuda:1'), covar=tensor([0.0900, 0.0476, 0.0891, 0.1478], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0433, 0.0331, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0011, 0.0015], device='cuda:1')
+2023-03-01 04:07:10,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5812, 2.0683, 1.4688, 0.7113], device='cuda:1'), covar=tensor([0.1182, 0.0788, 0.1292, 0.1502], device='cuda:1'), in_proj_covar=tensor([0.1162, 0.1186, 0.1185, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 04:07:30,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1399, 1.1658, 1.0033, 0.9403], device='cuda:1'), covar=tensor([0.0581, 0.0513, 0.0944, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0506, 0.0546, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 04:07:32,296 INFO [train.py:968] (1/2) Epoch 2, batch 17450, giga_loss[loss=0.5018, simple_loss=0.5193, pruned_loss=0.2422, over 28639.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3772, pruned_loss=0.1263, over 5668694.00 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3834, pruned_loss=0.1341, over 5758719.13 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3769, pruned_loss=0.1253, over 5650447.48 frames. ], batch size: 307, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:08:09,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3662, 1.8109, 1.3606, 0.3494], device='cuda:1'), covar=tensor([0.0973, 0.0653, 0.1142, 0.1371], device='cuda:1'), in_proj_covar=tensor([0.1143, 0.1161, 0.1164, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 04:08:15,304 INFO [optim.py:369] (1/2) 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 04:08:20,499 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 2, batch 17500, giga_loss[loss=0.3741, simple_loss=0.3985, pruned_loss=0.1748, over 23814.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3898, pruned_loss=0.1351, over 5672836.30 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3834, pruned_loss=0.134, over 5761183.28 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3895, pruned_loss=0.1343, over 5655073.74 frames. ], batch size: 705, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:08:45,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:08:47,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:09:07,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6845, 1.5047, 1.2522, 1.2410], device='cuda:1'), covar=tensor([0.0618, 0.0650, 0.0945, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0510, 0.0550, 0.0471], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 04:09:14,937 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:09:16,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-01 04:09:17,120 INFO [train.py:968] (1/2) Epoch 2, batch 17550, giga_loss[loss=0.2948, simple_loss=0.3591, pruned_loss=0.1153, over 28838.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3968, pruned_loss=0.1401, over 5681068.51 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.383, pruned_loss=0.1336, over 5763224.87 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3971, pruned_loss=0.1399, over 5663678.90 frames. ], batch size: 119, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:09:32,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6409, 1.3591, 1.4189, 1.3375], device='cuda:1'), covar=tensor([0.0700, 0.1179, 0.1230, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0802, 0.0616, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 04:09:47,320 INFO [optim.py:369] (1/2) 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 04:10:03,753 INFO [train.py:968] (1/2) Epoch 2, batch 17600, giga_loss[loss=0.3088, simple_loss=0.3733, pruned_loss=0.1221, over 28716.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.396, pruned_loss=0.1411, over 5670401.16 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3832, pruned_loss=0.1337, over 5752872.11 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3965, pruned_loss=0.1411, over 5662761.58 frames. ], batch size: 284, lr: 1.38e-02, grad_scale: 8.0
+2023-03-01 04:10:33,662 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:10:35,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:10:43,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:10:48,545 INFO [train.py:968] (1/2) Epoch 2, batch 17650, giga_loss[loss=0.3486, simple_loss=0.3817, pruned_loss=0.1578, over 26583.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3894, pruned_loss=0.1381, over 5673899.51 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3836, pruned_loss=0.1341, over 5747296.67 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3897, pruned_loss=0.1379, over 5670583.13 frames. ], batch size: 555, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:11:02,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:11:25,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:11:25,651 INFO [optim.py:369] (1/2) 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 04:11:38,515 INFO [train.py:968] (1/2) Epoch 2, batch 17700, giga_loss[loss=0.2926, simple_loss=0.3491, pruned_loss=0.1181, over 28938.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3809, pruned_loss=0.1338, over 5673740.78 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3837, pruned_loss=0.1342, over 5739617.06 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3811, pruned_loss=0.1335, over 5676066.06 frames. ], batch size: 106, lr: 1.38e-02, grad_scale: 4.0
+2023-03-01 04:11:58,516 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.17 vs. limit=2.0
+2023-03-01 04:12:21,568 INFO [train.py:968] (1/2) Epoch 2, batch 17750, giga_loss[loss=0.2926, simple_loss=0.3398, pruned_loss=0.1227, over 27692.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3719, pruned_loss=0.1289, over 5681467.03 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3843, pruned_loss=0.1345, over 5741838.42 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3713, pruned_loss=0.1285, over 5680139.88 frames. ], batch size: 472, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:12:36,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:12:54,246 INFO [optim.py:369] (1/2) 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 04:12:55,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:12:58,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:13:08,062 INFO [train.py:968] (1/2) Epoch 2, batch 17800, giga_loss[loss=0.2736, simple_loss=0.3325, pruned_loss=0.1074, over 28577.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3632, pruned_loss=0.1241, over 5689258.98 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3846, pruned_loss=0.1346, over 5745296.59 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3621, pruned_loss=0.1235, over 5684150.19 frames. ], batch size: 307, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:13:23,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:13:30,935 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:13:32,850 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 17850, giga_loss[loss=0.3084, simple_loss=0.3479, pruned_loss=0.1345, over 26581.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3583, pruned_loss=0.1218, over 5692466.32 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.385, pruned_loss=0.1348, over 5748574.81 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3566, pruned_loss=0.1209, over 5684663.09 frames. ], batch size: 555, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:13:57,276 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:14:06,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:14:16,760 INFO [optim.py:369] (1/2) 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 04:14:28,561 INFO [train.py:968] (1/2) Epoch 2, batch 17900, libri_loss[loss=0.3328, simple_loss=0.3953, pruned_loss=0.1351, over 29531.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3563, pruned_loss=0.1204, over 5706659.25 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3855, pruned_loss=0.1351, over 5756020.51 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3532, pruned_loss=0.1188, over 5690925.66 frames. ], batch size: 80, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:14:35,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:14:37,169 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:15:04,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:15:09,230 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 17950, giga_loss[loss=0.3121, simple_loss=0.3593, pruned_loss=0.1324, over 28632.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.353, pruned_loss=0.1186, over 5700376.56 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3858, pruned_loss=0.1352, over 5755649.11 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3494, pruned_loss=0.1167, over 5686800.14 frames. ], batch size: 307, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:15:46,071 INFO [optim.py:369] (1/2) 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 04:15:57,780 INFO [train.py:968] (1/2) Epoch 2, batch 18000, libri_loss[loss=0.4684, simple_loss=0.4974, pruned_loss=0.2197, over 29544.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3489, pruned_loss=0.1162, over 5696343.91 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3863, pruned_loss=0.1356, over 5757790.36 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3451, pruned_loss=0.1141, over 5682845.98 frames. ], batch size: 89, lr: 1.37e-02, grad_scale: 8.0
+2023-03-01 04:15:57,780 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 04:16:06,576 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 04:16:21,015 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:16:23,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:16:51,509 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:16:53,435 INFO [train.py:968] (1/2) Epoch 2, batch 18050, giga_loss[loss=0.243, simple_loss=0.3136, pruned_loss=0.08622, over 28875.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3458, pruned_loss=0.1141, over 5706307.67 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3869, pruned_loss=0.1358, over 5761389.24 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3413, pruned_loss=0.1117, over 5691164.41 frames. ], batch size: 145, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:17:23,543 INFO [optim.py:369] (1/2) 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 04:17:36,333 INFO [train.py:968] (1/2) Epoch 2, batch 18100, libri_loss[loss=0.3022, simple_loss=0.3629, pruned_loss=0.1208, over 29657.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3447, pruned_loss=0.1137, over 5697274.90 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3877, pruned_loss=0.1361, over 5758359.07 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3387, pruned_loss=0.1106, over 5685581.37 frames. ], batch size: 69, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:18:16,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:18:17,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:18:20,211 INFO [train.py:968] (1/2) Epoch 2, batch 18150, giga_loss[loss=0.2232, simple_loss=0.294, pruned_loss=0.0762, over 28969.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3418, pruned_loss=0.1121, over 5697974.38 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3879, pruned_loss=0.136, over 5761815.45 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3358, pruned_loss=0.1092, over 5684134.03 frames. ], batch size: 136, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:18:52,898 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 18200, giga_loss[loss=0.2524, simple_loss=0.3079, pruned_loss=0.09844, over 28587.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3377, pruned_loss=0.1091, over 5706315.60 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3882, pruned_loss=0.1361, over 5763065.85 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.332, pruned_loss=0.1064, over 5693381.51 frames. ], batch size: 85, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:19:54,912 INFO [train.py:968] (1/2) Epoch 2, batch 18250, giga_loss[loss=0.2771, simple_loss=0.335, pruned_loss=0.1096, over 28984.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3346, pruned_loss=0.1077, over 5708252.39 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3886, pruned_loss=0.1363, over 5767492.85 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3281, pruned_loss=0.1045, over 5692021.41 frames. ], batch size: 227, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:20:27,291 INFO [optim.py:369] (1/2) 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 04:20:37,296 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 18300, giga_loss[loss=0.3027, simple_loss=0.3726, pruned_loss=0.1164, over 29050.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3421, pruned_loss=0.1123, over 5708957.78 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3888, pruned_loss=0.1364, over 5769387.37 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3352, pruned_loss=0.1089, over 5692400.23 frames. ], batch size: 128, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:20:54,012 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=63979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:21:04,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:21:30,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7053, 1.4622, 1.2682, 1.2672], device='cuda:1'), covar=tensor([0.0626, 0.0734, 0.0987, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0507, 0.0538, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 04:21:32,421 INFO [train.py:968] (1/2) Epoch 2, batch 18350, giga_loss[loss=0.3824, simple_loss=0.4305, pruned_loss=0.1672, over 28633.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3592, pruned_loss=0.1234, over 5708079.36 frames. ], libri_tot_loss[loss=0.3309, simple_loss=0.3888, pruned_loss=0.1365, over 5770658.97 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3533, pruned_loss=0.1204, over 5692907.29 frames. ], batch size: 242, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:22:04,359 INFO [optim.py:369] (1/2) 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 04:22:13,678 INFO [train.py:968] (1/2) Epoch 2, batch 18400, giga_loss[loss=0.3703, simple_loss=0.4227, pruned_loss=0.159, over 28734.00 frames. ], tot_loss[loss=0.32, simple_loss=0.374, pruned_loss=0.133, over 5710151.56 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3902, pruned_loss=0.1376, over 5773545.11 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3672, pruned_loss=0.1293, over 5692559.71 frames. ], batch size: 262, lr: 1.37e-02, grad_scale: 8.0
+2023-03-01 04:22:41,449 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 04:22:56,529 INFO [train.py:968] (1/2) Epoch 2, batch 18450, libri_loss[loss=0.3891, simple_loss=0.4328, pruned_loss=0.1727, over 29499.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3847, pruned_loss=0.1388, over 5712639.40 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3912, pruned_loss=0.1383, over 5776485.75 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3781, pruned_loss=0.1351, over 5694560.61 frames. ], batch size: 85, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:23:08,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:23:11,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:23:30,210 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:23:43,140 INFO [train.py:968] (1/2) Epoch 2, batch 18500, giga_loss[loss=0.3326, simple_loss=0.3905, pruned_loss=0.1373, over 28731.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3887, pruned_loss=0.1396, over 5708782.99 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3913, pruned_loss=0.1384, over 5778613.37 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3834, pruned_loss=0.1367, over 5691417.63 frames. ], batch size: 92, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:23:58,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:24:00,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:24:11,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:24:26,019 INFO [train.py:968] (1/2) Epoch 2, batch 18550, giga_loss[loss=0.3208, simple_loss=0.3923, pruned_loss=0.1246, over 28766.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3894, pruned_loss=0.1381, over 5703081.78 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.392, pruned_loss=0.1387, over 5781136.39 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3844, pruned_loss=0.1355, over 5685375.98 frames. ], batch size: 186, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:24:33,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3823, 1.5011, 1.1669, 1.3249], device='cuda:1'), covar=tensor([0.1045, 0.0453, 0.0481, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0201, 0.0205, 0.0362], device='cuda:1'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:1')
+2023-03-01 04:24:42,442 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:25:00,745 INFO [optim.py:369] (1/2) 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 04:25:12,035 INFO [train.py:968] (1/2) Epoch 2, batch 18600, giga_loss[loss=0.3331, simple_loss=0.3853, pruned_loss=0.1405, over 28505.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3915, pruned_loss=0.139, over 5698606.71 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3925, pruned_loss=0.1389, over 5781879.73 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3869, pruned_loss=0.1367, over 5682016.11 frames. ], batch size: 71, lr: 1.37e-02, grad_scale: 4.0
+2023-03-01 04:25:31,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8407, 2.3250, 2.0405, 1.9380], device='cuda:1'), covar=tensor([0.1404, 0.1383, 0.1036, 0.0652], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0796, 0.0710, 0.0595], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 04:25:46,755 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:25:47,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4200, 1.8795, 1.2782, 0.7420], device='cuda:1'), covar=tensor([0.1989, 0.1038, 0.1186, 0.1811], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.1113, 0.1156, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 04:25:54,275 INFO [train.py:968] (1/2) Epoch 2, batch 18650, giga_loss[loss=0.3319, simple_loss=0.393, pruned_loss=0.1354, over 28941.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3943, pruned_loss=0.1412, over 5701632.31 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3933, pruned_loss=0.1391, over 5784967.09 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3899, pruned_loss=0.1392, over 5682783.65 frames. ], batch size: 213, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:26:04,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:06,212 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:06,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:09,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:10,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:26,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:27,721 INFO [optim.py:369] (1/2) 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 04:26:33,149 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:34,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:26:36,912 INFO [train.py:968] (1/2) Epoch 2, batch 18700, giga_loss[loss=0.3228, simple_loss=0.3945, pruned_loss=0.1256, over 28838.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3975, pruned_loss=0.1435, over 5697593.55 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3949, pruned_loss=0.1401, over 5773778.04 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3925, pruned_loss=0.141, over 5689687.48 frames. ], batch size: 174, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:27:18,424 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:27:20,072 INFO [train.py:968] (1/2) Epoch 2, batch 18750, giga_loss[loss=0.3521, simple_loss=0.409, pruned_loss=0.1477, over 28920.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.401, pruned_loss=0.1456, over 5700473.77 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3948, pruned_loss=0.14, over 5775789.60 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3973, pruned_loss=0.1439, over 5691157.74 frames. ], batch size: 145, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:27:32,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-01 04:27:50,956 INFO [optim.py:369] (1/2) 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 04:28:00,873 INFO [train.py:968] (1/2) Epoch 2, batch 18800, giga_loss[loss=0.3512, simple_loss=0.4088, pruned_loss=0.1468, over 28268.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.4023, pruned_loss=0.145, over 5712761.25 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3949, pruned_loss=0.14, over 5778952.53 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3995, pruned_loss=0.1438, over 5700879.92 frames. ], batch size: 368, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:28:11,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:15,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:31,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:32,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:41,333 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 18850, giga_loss[loss=0.3384, simple_loss=0.4083, pruned_loss=0.1343, over 28854.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.4038, pruned_loss=0.1452, over 5704239.16 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3952, pruned_loss=0.1402, over 5772769.70 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.4013, pruned_loss=0.1441, over 5699295.92 frames. ], batch size: 145, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:28:56,414 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:28:59,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3257, 1.8565, 1.6296, 1.6598], device='cuda:1'), covar=tensor([0.0864, 0.1960, 0.1404, 0.1540], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0817, 0.0636, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 04:29:00,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-01 04:29:20,199 INFO [optim.py:369] (1/2) 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 04:29:31,533 INFO [train.py:968] (1/2) Epoch 2, batch 18900, giga_loss[loss=0.327, simple_loss=0.3939, pruned_loss=0.1301, over 28573.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.4043, pruned_loss=0.1446, over 5703365.15 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3955, pruned_loss=0.1402, over 5775571.02 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.4023, pruned_loss=0.1438, over 5695568.74 frames. ], batch size: 336, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:29:39,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:30:07,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:30:12,131 INFO [train.py:968] (1/2) Epoch 2, batch 18950, giga_loss[loss=0.3468, simple_loss=0.4086, pruned_loss=0.1425, over 28673.00 frames. ], tot_loss[loss=0.344, simple_loss=0.4029, pruned_loss=0.1426, over 5703577.12 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.396, pruned_loss=0.1404, over 5778600.10 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.4011, pruned_loss=0.1418, over 5692931.60 frames. ], batch size: 242, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:30:17,574 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 2, batch 19000, giga_loss[loss=0.3816, simple_loss=0.4096, pruned_loss=0.1768, over 23689.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.4012, pruned_loss=0.1407, over 5710563.37 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3967, pruned_loss=0.1409, over 5781246.35 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3992, pruned_loss=0.1396, over 5698422.28 frames. ], batch size: 705, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:30:59,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2149, 1.8231, 5.3028, 3.7323], device='cuda:1'), covar=tensor([0.1590, 0.1578, 0.0198, 0.0358], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0482, 0.0603, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 04:31:06,900 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 19050, giga_loss[loss=0.4263, simple_loss=0.4392, pruned_loss=0.2067, over 27915.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.4032, pruned_loss=0.1433, over 5700779.03 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3972, pruned_loss=0.1413, over 5773603.88 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.4012, pruned_loss=0.1421, over 5695464.41 frames. ], batch size: 412, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:31:37,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7411, 1.5452, 1.5705, 1.8989], device='cuda:1'), covar=tensor([0.1635, 0.1671, 0.1311, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0769, 0.0839, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0006, 0.0006], device='cuda:1')
+2023-03-01 04:31:38,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:31:40,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:31:55,263 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:08,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:10,814 INFO [optim.py:369] (1/2) 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 04:32:11,216 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:14,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:24,053 INFO [train.py:968] (1/2) Epoch 2, batch 19100, giga_loss[loss=0.4385, simple_loss=0.4364, pruned_loss=0.2203, over 23889.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4063, pruned_loss=0.149, over 5689064.90 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3979, pruned_loss=0.1417, over 5775502.15 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4042, pruned_loss=0.1478, over 5681484.78 frames. ], batch size: 705, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:32:41,886 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:32:44,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-01 04:32:45,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:09,654 INFO [train.py:968] (1/2) Epoch 2, batch 19150, giga_loss[loss=0.3493, simple_loss=0.4042, pruned_loss=0.1472, over 28968.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4068, pruned_loss=0.1509, over 5684869.06 frames. ], libri_tot_loss[loss=0.341, simple_loss=0.3982, pruned_loss=0.1419, over 5769434.50 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4051, pruned_loss=0.1499, over 5682414.69 frames. ], batch size: 227, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:33:15,164 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:18,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:33,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-01 04:33:43,043 INFO [optim.py:369] (1/2) 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 04:33:43,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:45,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:33:51,556 INFO [train.py:968] (1/2) Epoch 2, batch 19200, giga_loss[loss=0.3581, simple_loss=0.4021, pruned_loss=0.1571, over 28730.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4066, pruned_loss=0.152, over 5695050.73 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3983, pruned_loss=0.142, over 5772955.92 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4053, pruned_loss=0.1514, over 5688079.62 frames. ], batch size: 60, lr: 1.36e-02, grad_scale: 8.0
+2023-03-01 04:34:35,859 INFO [train.py:968] (1/2) Epoch 2, batch 19250, giga_loss[loss=0.3741, simple_loss=0.4138, pruned_loss=0.1672, over 28945.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.405, pruned_loss=0.1512, over 5705470.02 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3985, pruned_loss=0.142, over 5776512.04 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.404, pruned_loss=0.151, over 5694190.98 frames. ], batch size: 213, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:34:46,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:34:48,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:35:11,481 INFO [optim.py:369] (1/2) 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 04:35:19,253 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:35:20,957 INFO [train.py:968] (1/2) Epoch 2, batch 19300, giga_loss[loss=0.3599, simple_loss=0.4171, pruned_loss=0.1514, over 28685.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4057, pruned_loss=0.1518, over 5693697.00 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3988, pruned_loss=0.1423, over 5776017.85 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4048, pruned_loss=0.1516, over 5682796.74 frames. ], batch size: 119, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:35:33,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=64982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:35:47,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:36:03,998 INFO [train.py:968] (1/2) Epoch 2, batch 19350, giga_loss[loss=0.292, simple_loss=0.3576, pruned_loss=0.1132, over 28922.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.403, pruned_loss=0.1483, over 5699953.28 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3992, pruned_loss=0.1424, over 5777487.97 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.402, pruned_loss=0.1482, over 5688911.12 frames. ], batch size: 199, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:36:27,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:36:41,278 INFO [optim.py:369] (1/2) 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 04:36:50,518 INFO [train.py:968] (1/2) Epoch 2, batch 19400, giga_loss[loss=0.292, simple_loss=0.358, pruned_loss=0.113, over 28644.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3975, pruned_loss=0.1436, over 5698768.62 frames. ], libri_tot_loss[loss=0.3421, simple_loss=0.3994, pruned_loss=0.1424, over 5779057.46 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3966, pruned_loss=0.1436, over 5685822.42 frames. ], batch size: 307, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:37:16,023 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:37:26,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7636, 1.6763, 4.2612, 3.0859], device='cuda:1'), covar=tensor([0.1493, 0.1469, 0.0259, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0476, 0.0607, 0.0488], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 04:37:32,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:37:37,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2737, 2.0205, 1.2866, 1.3450], device='cuda:1'), covar=tensor([0.1211, 0.0374, 0.0480, 0.1653], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0202, 0.0204, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:1')
+2023-03-01 04:37:39,689 INFO [train.py:968] (1/2) Epoch 2, batch 19450, giga_loss[loss=0.3217, simple_loss=0.3861, pruned_loss=0.1287, over 28766.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3916, pruned_loss=0.1398, over 5690399.33 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3995, pruned_loss=0.1424, over 5774055.88 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3907, pruned_loss=0.1398, over 5681555.23 frames. ], batch size: 284, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:37:58,618 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:02,809 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:04,008 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:04,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:13,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9951, 3.3787, 3.6978, 1.7228], device='cuda:1'), covar=tensor([0.0392, 0.0378, 0.0623, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0516, 0.0758, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0006, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 04:38:15,144 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 19500, giga_loss[loss=0.3354, simple_loss=0.3766, pruned_loss=0.1471, over 28004.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3851, pruned_loss=0.1356, over 5682795.07 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3996, pruned_loss=0.1425, over 5765305.28 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3838, pruned_loss=0.1354, over 5681368.01 frames. ], batch size: 412, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:38:31,495 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:38:32,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7644, 1.5340, 4.1404, 3.2195], device='cuda:1'), covar=tensor([0.1395, 0.1404, 0.0262, 0.0403], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0478, 0.0606, 0.0490], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 04:38:36,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4831, 1.8905, 1.5891, 1.5630], device='cuda:1'), covar=tensor([0.1366, 0.1747, 0.1298, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0798, 0.0707, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 04:39:04,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3596, 1.4088, 1.2845, 0.8760], device='cuda:1'), covar=tensor([0.0492, 0.0385, 0.0295, 0.0494], device='cuda:1'), in_proj_covar=tensor([0.0964, 0.0707, 0.0774, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 04:39:14,812 INFO [train.py:968] (1/2) Epoch 2, batch 19550, giga_loss[loss=0.2917, simple_loss=0.3576, pruned_loss=0.1129, over 28666.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3793, pruned_loss=0.1319, over 5683170.56 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3996, pruned_loss=0.1425, over 5760494.06 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3778, pruned_loss=0.1316, over 5683280.75 frames. ], batch size: 60, lr: 1.36e-02, grad_scale: 2.0
+2023-03-01 04:39:29,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:39:33,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3197, 1.9264, 1.4509, 0.3857], device='cuda:1'), covar=tensor([0.1812, 0.0884, 0.1296, 0.2088], device='cuda:1'), in_proj_covar=tensor([0.1145, 0.1107, 0.1139, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 04:39:49,699 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:39:52,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:40:02,344 INFO [train.py:968] (1/2) Epoch 2, batch 19600, giga_loss[loss=0.4048, simple_loss=0.4279, pruned_loss=0.1908, over 26508.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3808, pruned_loss=0.1327, over 5683143.72 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.4002, pruned_loss=0.1427, over 5759942.99 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3788, pruned_loss=0.1321, over 5682682.22 frames. ], batch size: 555, lr: 1.36e-02, grad_scale: 4.0
+2023-03-01 04:40:21,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5957, 1.4459, 3.6238, 2.8657], device='cuda:1'), covar=tensor([0.1566, 0.1568, 0.0287, 0.0720], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0482, 0.0602, 0.0489], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0007, 0.0006], device='cuda:1')
+2023-03-01 04:40:21,214 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:40:48,077 INFO [train.py:968] (1/2) Epoch 2, batch 19650, giga_loss[loss=0.2625, simple_loss=0.333, pruned_loss=0.09594, over 28637.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3796, pruned_loss=0.131, over 5677809.63 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.401, pruned_loss=0.1431, over 5741544.94 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3767, pruned_loss=0.1298, over 5692607.61 frames. ], batch size: 60, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:41:17,677 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:41:29,397 INFO [train.py:968] (1/2) Epoch 2, batch 19700, giga_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.117, over 28619.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3775, pruned_loss=0.1296, over 5690425.09 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.4007, pruned_loss=0.1429, over 5744440.99 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3747, pruned_loss=0.1284, over 5697554.10 frames. ], batch size: 307, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:41:29,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:41:32,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7593, 2.1906, 1.7981, 1.8331], device='cuda:1'), covar=tensor([0.1003, 0.0360, 0.0419, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0295, 0.0196, 0.0199, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0030, 0.0022, 0.0020, 0.0034], device='cuda:1')
+2023-03-01 04:41:35,436 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:41:37,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:42:02,948 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:42:09,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:42:12,415 INFO [train.py:968] (1/2) Epoch 2, batch 19750, giga_loss[loss=0.3152, simple_loss=0.3737, pruned_loss=0.1284, over 28842.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3736, pruned_loss=0.1277, over 5702140.64 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.4008, pruned_loss=0.1429, over 5744826.79 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3712, pruned_loss=0.1267, over 5707191.47 frames. ], batch size: 186, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:42:35,411 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:42:51,745 INFO [train.py:968] (1/2) Epoch 2, batch 19800, giga_loss[loss=0.2609, simple_loss=0.3374, pruned_loss=0.09214, over 28922.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3722, pruned_loss=0.1273, over 5699216.45 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.402, pruned_loss=0.1436, over 5736378.96 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3684, pruned_loss=0.1254, over 5708828.38 frames. ], batch size: 145, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:42:53,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:18,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:21,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:32,231 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 19850, giga_loss[loss=0.3157, simple_loss=0.3719, pruned_loss=0.1297, over 28252.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3686, pruned_loss=0.1255, over 5703956.17 frames. ], libri_tot_loss[loss=0.3444, simple_loss=0.4019, pruned_loss=0.1435, over 5737194.06 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3655, pruned_loss=0.124, over 5710528.49 frames. ], batch size: 368, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:43:38,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:43:48,561 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:08,640 INFO [optim.py:369] (1/2) 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 04:44:08,955 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:11,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:16,624 INFO [train.py:968] (1/2) Epoch 2, batch 19900, giga_loss[loss=0.2987, simple_loss=0.3614, pruned_loss=0.118, over 28613.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3678, pruned_loss=0.1249, over 5715341.57 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.4032, pruned_loss=0.144, over 5740093.34 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3628, pruned_loss=0.1226, over 5717245.56 frames. ], batch size: 336, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:44:32,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:33,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-01 04:44:52,258 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:54,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:44:55,249 INFO [train.py:968] (1/2) Epoch 2, batch 19950, giga_loss[loss=0.3273, simple_loss=0.3908, pruned_loss=0.1319, over 28295.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3666, pruned_loss=0.1242, over 5718665.72 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.404, pruned_loss=0.1443, over 5744005.51 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3608, pruned_loss=0.1215, over 5715967.14 frames. ], batch size: 368, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:45:12,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6184, 1.4369, 3.9889, 3.0394], device='cuda:1'), covar=tensor([0.1469, 0.1442, 0.0282, 0.0407], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0478, 0.0609, 0.0495], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 04:45:17,705 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:25,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:27,590 INFO [optim.py:369] (1/2) 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 04:45:27,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:30,822 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:33,044 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 20000, giga_loss[loss=0.3281, simple_loss=0.3777, pruned_loss=0.1392, over 27653.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3661, pruned_loss=0.1242, over 5708179.29 frames. ], libri_tot_loss[loss=0.3478, simple_loss=0.4054, pruned_loss=0.1451, over 5735884.59 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3589, pruned_loss=0.1207, over 5712810.18 frames. ], batch size: 472, lr: 1.35e-02, grad_scale: 8.0
+2023-03-01 04:45:44,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:51,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:45:56,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:46:17,408 INFO [train.py:968] (1/2) Epoch 2, batch 20050, giga_loss[loss=0.2563, simple_loss=0.3207, pruned_loss=0.09592, over 29100.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3626, pruned_loss=0.1218, over 5720023.12 frames. ], libri_tot_loss[loss=0.3484, simple_loss=0.406, pruned_loss=0.1454, over 5738561.69 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3558, pruned_loss=0.1184, over 5721029.37 frames. ], batch size: 128, lr: 1.35e-02, grad_scale: 8.0
+2023-03-01 04:46:36,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:46:50,910 INFO [optim.py:369] (1/2) 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 04:46:56,603 INFO [train.py:968] (1/2) Epoch 2, batch 20100, giga_loss[loss=0.2773, simple_loss=0.3424, pruned_loss=0.1062, over 28982.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3622, pruned_loss=0.1215, over 5719300.03 frames. ], libri_tot_loss[loss=0.3497, simple_loss=0.4074, pruned_loss=0.146, over 5735862.16 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3544, pruned_loss=0.1176, over 5722077.42 frames. ], batch size: 145, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:47:06,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-01 04:47:22,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4119, 1.3558, 1.3193, 1.3663], device='cuda:1'), covar=tensor([0.1824, 0.1954, 0.1577, 0.2063], device='cuda:1'), in_proj_covar=tensor([0.0955, 0.0794, 0.0867, 0.0918], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 04:47:36,138 INFO [train.py:968] (1/2) Epoch 2, batch 20150, giga_loss[loss=0.2712, simple_loss=0.3369, pruned_loss=0.1028, over 28931.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3612, pruned_loss=0.1208, over 5729837.78 frames. ], libri_tot_loss[loss=0.3508, simple_loss=0.4082, pruned_loss=0.1467, over 5738033.07 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3532, pruned_loss=0.1166, over 5729756.96 frames. ], batch size: 136, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:47:40,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:48:05,216 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-01 04:48:05,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2096, 1.5760, 1.1143, 1.4287], device='cuda:1'), covar=tensor([0.0890, 0.0364, 0.0456, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0197, 0.0201, 0.0352], device='cuda:1'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:1')
+2023-03-01 04:48:15,388 INFO [optim.py:369] (1/2) 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 04:48:21,930 INFO [train.py:968] (1/2) Epoch 2, batch 20200, giga_loss[loss=0.3266, simple_loss=0.3835, pruned_loss=0.1348, over 28841.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3675, pruned_loss=0.126, over 5720145.17 frames. ], libri_tot_loss[loss=0.3519, simple_loss=0.409, pruned_loss=0.1474, over 5741425.04 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3596, pruned_loss=0.1215, over 5716733.58 frames. ], batch size: 112, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:48:39,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:48:42,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:49:12,202 INFO [train.py:968] (1/2) Epoch 2, batch 20250, giga_loss[loss=0.3823, simple_loss=0.4233, pruned_loss=0.1706, over 28247.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3759, pruned_loss=0.132, over 5715531.55 frames. ], libri_tot_loss[loss=0.3528, simple_loss=0.4097, pruned_loss=0.148, over 5739365.91 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3683, pruned_loss=0.1276, over 5714342.74 frames. ], batch size: 368, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:49:12,472 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:49:17,465 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=65925.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:49:56,323 INFO [optim.py:369] (1/2) 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 04:50:00,849 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 20300, giga_loss[loss=0.3405, simple_loss=0.399, pruned_loss=0.141, over 28957.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3877, pruned_loss=0.1416, over 5700810.26 frames. ], libri_tot_loss[loss=0.354, simple_loss=0.4107, pruned_loss=0.1487, over 5742127.76 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3802, pruned_loss=0.1372, over 5696453.00 frames. ], batch size: 128, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:50:03,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:50:32,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:50:42,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5355, 1.4167, 1.3527, 1.7726], device='cuda:1'), covar=tensor([0.1947, 0.2013, 0.1640, 0.2094], device='cuda:1'), in_proj_covar=tensor([0.0960, 0.0798, 0.0868, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 04:50:49,106 INFO [train.py:968] (1/2) Epoch 2, batch 20350, libri_loss[loss=0.3275, simple_loss=0.3965, pruned_loss=0.1293, over 29541.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3921, pruned_loss=0.143, over 5701770.03 frames. ], libri_tot_loss[loss=0.3547, simple_loss=0.4112, pruned_loss=0.1491, over 5746910.58 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.385, pruned_loss=0.1389, over 5692333.40 frames. ], batch size: 82, lr: 1.35e-02, grad_scale: 2.0
+2023-03-01 04:50:50,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-01 04:51:26,178 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:51:39,022 INFO [train.py:968] (1/2) Epoch 2, batch 20400, giga_loss[loss=0.3349, simple_loss=0.3998, pruned_loss=0.135, over 29057.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3976, pruned_loss=0.1455, over 5686671.60 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4114, pruned_loss=0.1492, over 5747611.41 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3917, pruned_loss=0.1422, over 5678104.87 frames. ], batch size: 155, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:52:28,467 INFO [train.py:968] (1/2) Epoch 2, batch 20450, giga_loss[loss=0.3484, simple_loss=0.406, pruned_loss=0.1454, over 28556.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4049, pruned_loss=0.1512, over 5684223.06 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4122, pruned_loss=0.15, over 5752642.44 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3989, pruned_loss=0.1478, over 5670101.60 frames. ], batch size: 71, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:52:46,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 04:53:04,811 INFO [optim.py:369] (1/2) 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 04:53:10,702 INFO [train.py:968] (1/2) Epoch 2, batch 20500, giga_loss[loss=0.3029, simple_loss=0.3702, pruned_loss=0.1178, over 28495.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.4072, pruned_loss=0.1533, over 5680716.27 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4124, pruned_loss=0.1501, over 5752407.23 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4022, pruned_loss=0.1505, over 5668062.98 frames. ], batch size: 71, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:53:34,394 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:53:36,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:53:54,733 INFO [train.py:968] (1/2) Epoch 2, batch 20550, giga_loss[loss=0.3313, simple_loss=0.3666, pruned_loss=0.148, over 23911.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.399, pruned_loss=0.1466, over 5677650.44 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4128, pruned_loss=0.1505, over 5746405.24 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3943, pruned_loss=0.1439, over 5670583.88 frames. ], batch size: 705, lr: 1.35e-02, grad_scale: 4.0
+2023-03-01 04:54:02,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:54:05,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 04:54:10,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-01 04:54:13,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-01 04:54:33,162 INFO [optim.py:369] (1/2) 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 04:54:39,745 INFO [train.py:968] (1/2) Epoch 2, batch 20600, giga_loss[loss=0.3276, simple_loss=0.3934, pruned_loss=0.1309, over 28760.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3971, pruned_loss=0.1445, over 5696140.64 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4133, pruned_loss=0.1513, over 5748984.45 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3926, pruned_loss=0.1416, over 5687287.39 frames. ], batch size: 262, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:55:09,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66300.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:55:25,634 INFO [train.py:968] (1/2) Epoch 2, batch 20650, giga_loss[loss=0.3055, simple_loss=0.3739, pruned_loss=0.1186, over 28808.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3978, pruned_loss=0.1447, over 5696702.33 frames. ], libri_tot_loss[loss=0.3586, simple_loss=0.4138, pruned_loss=0.1517, over 5752002.48 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3935, pruned_loss=0.1419, over 5685674.18 frames. ], batch size: 119, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:56:00,128 INFO [optim.py:369] (1/2) 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 04:56:04,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6746, 1.6662, 1.6102, 1.8053], device='cuda:1'), covar=tensor([0.1841, 0.1818, 0.1465, 0.2104], device='cuda:1'), in_proj_covar=tensor([0.0964, 0.0808, 0.0875, 0.0924], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 04:56:07,023 INFO [train.py:968] (1/2) Epoch 2, batch 20700, giga_loss[loss=0.3335, simple_loss=0.3921, pruned_loss=0.1375, over 28856.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.4006, pruned_loss=0.146, over 5701646.29 frames. ], libri_tot_loss[loss=0.3589, simple_loss=0.414, pruned_loss=0.152, over 5755628.18 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3967, pruned_loss=0.1433, over 5688079.38 frames. ], batch size: 112, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:56:11,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2390, 1.6154, 1.1916, 0.4777], device='cuda:1'), covar=tensor([0.1130, 0.0671, 0.0823, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.1133, 0.1128, 0.1161, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 04:56:57,232 INFO [train.py:968] (1/2) Epoch 2, batch 20750, giga_loss[loss=0.3267, simple_loss=0.383, pruned_loss=0.1353, over 28816.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.4033, pruned_loss=0.1483, over 5687643.99 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4142, pruned_loss=0.1522, over 5738057.37 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3999, pruned_loss=0.1459, over 5692010.81 frames. ], batch size: 66, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:57:18,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66443.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:57:21,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66446.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 04:57:32,451 INFO [optim.py:369] (1/2) 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 04:57:43,208 INFO [train.py:968] (1/2) Epoch 2, batch 20800, giga_loss[loss=0.324, simple_loss=0.384, pruned_loss=0.132, over 28623.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4059, pruned_loss=0.1509, over 5682654.96 frames. ], libri_tot_loss[loss=0.3601, simple_loss=0.4147, pruned_loss=0.1527, over 5740269.00 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4025, pruned_loss=0.1485, over 5683311.29 frames. ], batch size: 78, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:57:47,593 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66475.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 04:58:29,831 INFO [train.py:968] (1/2) Epoch 2, batch 20850, giga_loss[loss=0.3758, simple_loss=0.4208, pruned_loss=0.1654, over 28795.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.408, pruned_loss=0.1528, over 5675327.47 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4157, pruned_loss=0.1534, over 5734003.76 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4044, pruned_loss=0.1503, over 5679789.49 frames. ], batch size: 99, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:58:49,215 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 04:59:00,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66553.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 04:59:07,452 INFO [optim.py:369] (1/2) 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 04:59:14,485 INFO [train.py:968] (1/2) Epoch 2, batch 20900, giga_loss[loss=0.3326, simple_loss=0.3917, pruned_loss=0.1367, over 28308.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4079, pruned_loss=0.1527, over 5685521.63 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4169, pruned_loss=0.1543, over 5737510.37 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4037, pruned_loss=0.1498, over 5684413.59 frames. ], batch size: 77, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 04:59:14,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1136, 0.9431, 0.8193, 1.2644], device='cuda:1'), covar=tensor([0.1069, 0.0434, 0.0517, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0194, 0.0197, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:1')
+2023-03-01 04:59:44,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66606.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 04:59:53,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4128, 1.4106, 1.0556, 1.1974], device='cuda:1'), covar=tensor([0.0568, 0.0537, 0.0882, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0506, 0.0547, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 04:59:53,860 INFO [train.py:968] (1/2) Epoch 2, batch 20950, giga_loss[loss=0.4127, simple_loss=0.4374, pruned_loss=0.194, over 26718.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4073, pruned_loss=0.1511, over 5685376.37 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4169, pruned_loss=0.1543, over 5726610.63 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4039, pruned_loss=0.1488, over 5693188.94 frames. ], batch size: 555, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:00:19,019 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:00:20,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 05:00:27,973 INFO [optim.py:369] (1/2) 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 05:00:32,855 INFO [train.py:968] (1/2) Epoch 2, batch 21000, giga_loss[loss=0.3986, simple_loss=0.442, pruned_loss=0.1777, over 28600.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4059, pruned_loss=0.1486, over 5685102.50 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4171, pruned_loss=0.1547, over 5727697.06 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.4028, pruned_loss=0.1463, over 5689496.14 frames. ], batch size: 307, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:00:32,855 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 05:00:38,143 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2469, 1.6669, 1.3292, 0.3007], device='cuda:1'), covar=tensor([0.1011, 0.0786, 0.1281, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.1134, 0.1114, 0.1154, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 05:00:41,509 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 05:01:05,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8361, 1.7426, 3.7921, 3.0552], device='cuda:1'), covar=tensor([0.1454, 0.1410, 0.0294, 0.0545], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0482, 0.0616, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 05:01:22,402 INFO [train.py:968] (1/2) Epoch 2, batch 21050, giga_loss[loss=0.3013, simple_loss=0.3658, pruned_loss=0.1184, over 28797.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4061, pruned_loss=0.1474, over 5694577.81 frames. ], libri_tot_loss[loss=0.3644, simple_loss=0.4179, pruned_loss=0.1554, over 5731766.37 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.4025, pruned_loss=0.1447, over 5693436.03 frames. ], batch size: 99, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:01:46,748 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66749.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:01:49,974 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66752.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:01:56,636 INFO [optim.py:369] (1/2) 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 05:02:02,927 INFO [train.py:968] (1/2) Epoch 2, batch 21100, giga_loss[loss=0.3552, simple_loss=0.4043, pruned_loss=0.153, over 28882.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.4038, pruned_loss=0.1459, over 5695281.45 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4181, pruned_loss=0.1556, over 5728920.54 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.4008, pruned_loss=0.1436, over 5696492.67 frames. ], batch size: 112, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:02:13,315 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66781.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:02:39,360 INFO [train.py:968] (1/2) Epoch 2, batch 21150, giga_loss[loss=0.3155, simple_loss=0.3767, pruned_loss=0.1271, over 28839.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.4031, pruned_loss=0.1462, over 5712627.26 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4201, pruned_loss=0.1578, over 5735684.14 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3983, pruned_loss=0.1421, over 5706618.95 frames. ], batch size: 119, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:03:15,635 INFO [optim.py:369] (1/2) 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 05:03:20,234 INFO [train.py:968] (1/2) Epoch 2, batch 21200, giga_loss[loss=0.3227, simple_loss=0.3836, pruned_loss=0.1309, over 28896.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.4, pruned_loss=0.1441, over 5714537.92 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4199, pruned_loss=0.1579, over 5740040.87 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3959, pruned_loss=0.1403, over 5705360.35 frames. ], batch size: 227, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:03:57,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:03:58,539 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=66914.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:04:04,214 INFO [train.py:968] (1/2) Epoch 2, batch 21250, giga_loss[loss=0.3641, simple_loss=0.3903, pruned_loss=0.1689, over 23365.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3987, pruned_loss=0.1439, over 5707744.92 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4197, pruned_loss=0.158, over 5740087.64 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3952, pruned_loss=0.1405, over 5700074.99 frames. ], batch size: 705, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:04:12,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66928.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:04:40,221 INFO [optim.py:369] (1/2) 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 05:04:45,515 INFO [train.py:968] (1/2) Epoch 2, batch 21300, giga_loss[loss=0.2999, simple_loss=0.3724, pruned_loss=0.1137, over 28512.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.4009, pruned_loss=0.1457, over 5704306.73 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4202, pruned_loss=0.1587, over 5731214.86 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3971, pruned_loss=0.142, over 5706008.68 frames. ], batch size: 71, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:05:25,576 INFO [train.py:968] (1/2) Epoch 2, batch 21350, libri_loss[loss=0.3597, simple_loss=0.4073, pruned_loss=0.156, over 29562.00 frames. ], tot_loss[loss=0.347, simple_loss=0.402, pruned_loss=0.1461, over 5698135.97 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4205, pruned_loss=0.1594, over 5724211.00 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.398, pruned_loss=0.142, over 5705959.36 frames. ], batch size: 79, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:05:29,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9902, 1.2346, 1.0536, 0.1214], device='cuda:1'), covar=tensor([0.0993, 0.0883, 0.1486, 0.1933], device='cuda:1'), in_proj_covar=tensor([0.1136, 0.1105, 0.1141, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 05:05:31,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:05:55,875 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:05:57,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:06:01,294 INFO [optim.py:369] (1/2) 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 05:06:06,588 INFO [train.py:968] (1/2) Epoch 2, batch 21400, libri_loss[loss=0.4089, simple_loss=0.4392, pruned_loss=0.1893, over 29550.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3992, pruned_loss=0.143, over 5702281.17 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4208, pruned_loss=0.1598, over 5726779.52 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3956, pruned_loss=0.1393, over 5705730.16 frames. ], batch size: 77, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:06:08,899 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67071.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:06:10,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67074.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:06:15,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-01 05:06:21,795 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:06:35,106 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67103.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:06:42,000 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-01 05:06:47,355 INFO [train.py:968] (1/2) Epoch 2, batch 21450, giga_loss[loss=0.3141, simple_loss=0.3853, pruned_loss=0.1214, over 28474.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3985, pruned_loss=0.1419, over 5710911.16 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.421, pruned_loss=0.1601, over 5726007.76 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3951, pruned_loss=0.1384, over 5713792.92 frames. ], batch size: 65, lr: 1.34e-02, grad_scale: 4.0
+2023-03-01 05:06:52,259 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67124.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:07:22,627 INFO [optim.py:369] (1/2) 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 05:07:26,184 INFO [train.py:968] (1/2) Epoch 2, batch 21500, giga_loss[loss=0.3323, simple_loss=0.3882, pruned_loss=0.1382, over 28872.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3973, pruned_loss=0.1417, over 5720792.21 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.421, pruned_loss=0.1604, over 5728744.25 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3939, pruned_loss=0.1381, over 5720335.95 frames. ], batch size: 119, lr: 1.34e-02, grad_scale: 2.0
+2023-03-01 05:07:26,505 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:07:29,456 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 2, batch 21550, giga_loss[loss=0.2925, simple_loss=0.3561, pruned_loss=0.1145, over 28685.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3959, pruned_loss=0.1419, over 5726466.63 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4213, pruned_loss=0.1614, over 5739543.26 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3915, pruned_loss=0.137, over 5715964.86 frames. ], batch size: 92, lr: 1.34e-02, grad_scale: 2.0
+2023-03-01 05:08:46,728 INFO [optim.py:369] (1/2) 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 05:08:51,476 INFO [train.py:968] (1/2) Epoch 2, batch 21600, giga_loss[loss=0.3608, simple_loss=0.4145, pruned_loss=0.1536, over 28759.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3926, pruned_loss=0.1393, over 5716111.28 frames. ], libri_tot_loss[loss=0.3725, simple_loss=0.4217, pruned_loss=0.1617, over 5732245.64 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3886, pruned_loss=0.135, over 5714619.44 frames. ], batch size: 284, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:08:55,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3942, 1.3472, 1.2973, 0.7775], device='cuda:1'), covar=tensor([0.0589, 0.0439, 0.0337, 0.0579], device='cuda:1'), in_proj_covar=tensor([0.1039, 0.0715, 0.0812, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 05:09:08,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67289.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:09:28,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2833, 1.7383, 1.4018, 0.3411], device='cuda:1'), covar=tensor([0.1029, 0.0760, 0.1263, 0.1590], device='cuda:1'), in_proj_covar=tensor([0.1142, 0.1110, 0.1145, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 05:09:30,487 INFO [train.py:968] (1/2) Epoch 2, batch 21650, libri_loss[loss=0.3396, simple_loss=0.3892, pruned_loss=0.145, over 29590.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3929, pruned_loss=0.1401, over 5716446.11 frames. ], libri_tot_loss[loss=0.3725, simple_loss=0.4215, pruned_loss=0.1617, over 5727879.14 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3891, pruned_loss=0.136, over 5719054.64 frames. ], batch size: 74, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:09:57,367 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:10:08,284 INFO [optim.py:369] (1/2) 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 05:10:11,732 INFO [train.py:968] (1/2) Epoch 2, batch 21700, giga_loss[loss=0.3202, simple_loss=0.3744, pruned_loss=0.1331, over 28962.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3935, pruned_loss=0.1417, over 5711080.64 frames. ], libri_tot_loss[loss=0.3741, simple_loss=0.4227, pruned_loss=0.1628, over 5723209.84 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3887, pruned_loss=0.1369, over 5716001.81 frames. ], batch size: 106, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:10:47,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2369, 1.3772, 0.9003, 1.3075], device='cuda:1'), covar=tensor([0.0989, 0.0412, 0.0529, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0195, 0.0197, 0.0351], device='cuda:1'), out_proj_covar=tensor([0.0030, 0.0023, 0.0020, 0.0034], device='cuda:1')
+2023-03-01 05:10:52,423 INFO [train.py:968] (1/2) Epoch 2, batch 21750, giga_loss[loss=0.2725, simple_loss=0.3347, pruned_loss=0.1052, over 28729.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3925, pruned_loss=0.1424, over 5710526.71 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4226, pruned_loss=0.1634, over 5727994.27 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3874, pruned_loss=0.137, over 5709924.93 frames. ], batch size: 99, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:11:03,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67432.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:11:06,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:11:27,798 INFO [optim.py:369] (1/2) 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 05:11:28,031 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67464.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:11:32,070 INFO [train.py:968] (1/2) Epoch 2, batch 21800, giga_loss[loss=0.2887, simple_loss=0.346, pruned_loss=0.1157, over 28974.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3903, pruned_loss=0.1417, over 5717925.24 frames. ], libri_tot_loss[loss=0.3755, simple_loss=0.423, pruned_loss=0.164, over 5729254.03 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3854, pruned_loss=0.1367, over 5716020.43 frames. ], batch size: 106, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:11:37,784 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:11:57,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67499.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:12:14,574 INFO [train.py:968] (1/2) Epoch 2, batch 21850, giga_loss[loss=0.2959, simple_loss=0.3549, pruned_loss=0.1185, over 28883.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3876, pruned_loss=0.1409, over 5712644.64 frames. ], libri_tot_loss[loss=0.3757, simple_loss=0.4229, pruned_loss=0.1642, over 5734007.17 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3829, pruned_loss=0.1361, over 5706403.02 frames. ], batch size: 186, lr: 1.33e-02, grad_scale: 2.0
+2023-03-01 05:12:50,507 INFO [optim.py:369] (1/2) 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 05:12:55,953 INFO [train.py:968] (1/2) Epoch 2, batch 21900, libri_loss[loss=0.3792, simple_loss=0.4217, pruned_loss=0.1684, over 29560.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.387, pruned_loss=0.1405, over 5714272.65 frames. ], libri_tot_loss[loss=0.3767, simple_loss=0.4236, pruned_loss=0.1649, over 5736936.20 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3819, pruned_loss=0.1355, over 5706123.56 frames. ], batch size: 78, lr: 1.33e-02, grad_scale: 2.0
+2023-03-01 05:13:06,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-01 05:13:32,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7444, 2.1308, 1.9131, 1.9091], device='cuda:1'), covar=tensor([0.1212, 0.1454, 0.1038, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0795, 0.0703, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 05:13:39,962 INFO [train.py:968] (1/2) Epoch 2, batch 21950, giga_loss[loss=0.358, simple_loss=0.4101, pruned_loss=0.1529, over 27934.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3864, pruned_loss=0.1396, over 5712827.63 frames. ], libri_tot_loss[loss=0.3772, simple_loss=0.4239, pruned_loss=0.1652, over 5739530.74 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3814, pruned_loss=0.135, over 5703623.93 frames. ], batch size: 412, lr: 1.33e-02, grad_scale: 2.0
+2023-03-01 05:13:43,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-01 05:14:00,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67642.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:14:03,192 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67645.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:14:20,106 INFO [optim.py:369] (1/2) 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 05:14:23,033 INFO [train.py:968] (1/2) Epoch 2, batch 22000, libri_loss[loss=0.3829, simple_loss=0.4221, pruned_loss=0.1719, over 25988.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3898, pruned_loss=0.1412, over 5711448.87 frames. ], libri_tot_loss[loss=0.3784, simple_loss=0.4245, pruned_loss=0.1661, over 5739723.30 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.384, pruned_loss=0.1357, over 5703109.88 frames. ], batch size: 136, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:14:29,544 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67674.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:14:53,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-01 05:15:06,204 INFO [train.py:968] (1/2) Epoch 2, batch 22050, giga_loss[loss=0.3318, simple_loss=0.3956, pruned_loss=0.1341, over 28675.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3919, pruned_loss=0.1417, over 5716838.38 frames. ], libri_tot_loss[loss=0.3795, simple_loss=0.4252, pruned_loss=0.1669, over 5744685.85 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3857, pruned_loss=0.1359, over 5704802.95 frames. ], batch size: 262, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:15:12,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:15:22,643 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 05:15:29,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-01 05:15:48,355 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 22100, giga_loss[loss=0.3099, simple_loss=0.3812, pruned_loss=0.1193, over 28991.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3925, pruned_loss=0.1412, over 5709053.13 frames. ], libri_tot_loss[loss=0.3802, simple_loss=0.4255, pruned_loss=0.1675, over 5746429.38 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.387, pruned_loss=0.1358, over 5697673.13 frames. ], batch size: 164, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:16:09,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 05:16:37,737 INFO [train.py:968] (1/2) Epoch 2, batch 22150, giga_loss[loss=0.3869, simple_loss=0.4309, pruned_loss=0.1715, over 27944.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3921, pruned_loss=0.1402, over 5704245.95 frames. ], libri_tot_loss[loss=0.3811, simple_loss=0.4261, pruned_loss=0.1681, over 5745042.48 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3866, pruned_loss=0.135, over 5695658.85 frames. ], batch size: 412, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:17:04,050 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:17:17,747 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:17:18,011 INFO [train.py:968] (1/2) Epoch 2, batch 22200, giga_loss[loss=0.3076, simple_loss=0.3713, pruned_loss=0.122, over 28823.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.392, pruned_loss=0.1405, over 5701383.89 frames. ], libri_tot_loss[loss=0.3812, simple_loss=0.4258, pruned_loss=0.1683, over 5738398.98 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3867, pruned_loss=0.1351, over 5698450.20 frames. ], batch size: 186, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:17:19,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:17:44,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:17:51,346 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=67906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:18:02,001 INFO [train.py:968] (1/2) Epoch 2, batch 22250, giga_loss[loss=0.32, simple_loss=0.3782, pruned_loss=0.1309, over 28767.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3934, pruned_loss=0.1417, over 5702716.31 frames. ], libri_tot_loss[loss=0.382, simple_loss=0.4264, pruned_loss=0.1688, over 5739349.69 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3884, pruned_loss=0.1368, over 5699283.50 frames. ], batch size: 119, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:18:40,814 INFO [optim.py:369] (1/2) 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 05:18:46,457 INFO [train.py:968] (1/2) Epoch 2, batch 22300, giga_loss[loss=0.3493, simple_loss=0.4089, pruned_loss=0.1449, over 28740.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.396, pruned_loss=0.1436, over 5705250.62 frames. ], libri_tot_loss[loss=0.383, simple_loss=0.427, pruned_loss=0.1695, over 5738963.39 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3908, pruned_loss=0.1385, over 5702045.70 frames. ], batch size: 242, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:19:07,356 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:19:09,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:19:27,315 INFO [train.py:968] (1/2) Epoch 2, batch 22350, giga_loss[loss=0.3196, simple_loss=0.3853, pruned_loss=0.1269, over 28685.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3996, pruned_loss=0.1466, over 5703220.69 frames. ], libri_tot_loss[loss=0.3839, simple_loss=0.4275, pruned_loss=0.1702, over 5744075.75 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3941, pruned_loss=0.1411, over 5694939.34 frames. ], batch size: 262, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:19:32,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:19:59,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2629, 1.9540, 1.2734, 1.1548], device='cuda:1'), covar=tensor([0.0852, 0.0626, 0.0879, 0.1349], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0436, 0.0336, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0012, 0.0015], device='cuda:1')
+2023-03-01 05:20:05,747 INFO [optim.py:369] (1/2) 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 05:20:09,008 INFO [train.py:968] (1/2) Epoch 2, batch 22400, giga_loss[loss=0.338, simple_loss=0.4029, pruned_loss=0.1365, over 28877.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4025, pruned_loss=0.1485, over 5710484.89 frames. ], libri_tot_loss[loss=0.3855, simple_loss=0.4284, pruned_loss=0.1713, over 5746495.01 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3965, pruned_loss=0.1425, over 5700796.16 frames. ], batch size: 199, lr: 1.33e-02, grad_scale: 8.0
+2023-03-01 05:20:11,772 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:20:21,775 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:20:39,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 05:20:46,535 INFO [train.py:968] (1/2) Epoch 2, batch 22450, giga_loss[loss=0.3604, simple_loss=0.4071, pruned_loss=0.1569, over 28436.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.4037, pruned_loss=0.1489, over 5712292.80 frames. ], libri_tot_loss[loss=0.3865, simple_loss=0.4291, pruned_loss=0.1719, over 5744125.90 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3971, pruned_loss=0.1424, over 5704584.74 frames. ], batch size: 71, lr: 1.33e-02, grad_scale: 8.0
+2023-03-01 05:21:24,547 INFO [optim.py:369] (1/2) 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 05:21:26,680 INFO [train.py:968] (1/2) Epoch 2, batch 22500, giga_loss[loss=0.3175, simple_loss=0.3861, pruned_loss=0.1245, over 28836.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4039, pruned_loss=0.1485, over 5715709.73 frames. ], libri_tot_loss[loss=0.3875, simple_loss=0.4299, pruned_loss=0.1726, over 5746789.71 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3972, pruned_loss=0.1421, over 5706471.05 frames. ], batch size: 227, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:22:10,795 INFO [train.py:968] (1/2) Epoch 2, batch 22550, giga_loss[loss=0.3006, simple_loss=0.3728, pruned_loss=0.1142, over 28831.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4035, pruned_loss=0.1481, over 5719965.48 frames. ], libri_tot_loss[loss=0.3874, simple_loss=0.4296, pruned_loss=0.1726, over 5750529.22 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3976, pruned_loss=0.1421, over 5708422.22 frames. ], batch size: 119, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:22:50,589 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 22600, giga_loss[loss=0.3487, simple_loss=0.4031, pruned_loss=0.1472, over 28784.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.4028, pruned_loss=0.1474, over 5716191.91 frames. ], libri_tot_loss[loss=0.3878, simple_loss=0.4298, pruned_loss=0.1729, over 5751426.85 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3971, pruned_loss=0.1418, over 5705273.50 frames. ], batch size: 243, lr: 1.33e-02, grad_scale: 4.0
+2023-03-01 05:23:02,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4790, 1.6756, 1.3587, 1.0014], device='cuda:1'), covar=tensor([0.0731, 0.0359, 0.0353, 0.0534], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0735, 0.0858, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 05:23:03,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 22650, giga_loss[loss=0.2914, simple_loss=0.3571, pruned_loss=0.1128, over 28928.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3988, pruned_loss=0.1447, over 5719933.30 frames. ], libri_tot_loss[loss=0.3885, simple_loss=0.4302, pruned_loss=0.1734, over 5753777.18 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3934, pruned_loss=0.1392, over 5708662.51 frames. ], batch size: 112, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:23:39,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2947, 2.2673, 1.4025, 1.1771], device='cuda:1'), covar=tensor([0.0866, 0.0613, 0.0871, 0.1424], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0444, 0.0339, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0016, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 05:24:15,109 INFO [optim.py:369] (1/2) 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 05:24:16,863 INFO [train.py:968] (1/2) Epoch 2, batch 22700, giga_loss[loss=0.3473, simple_loss=0.3964, pruned_loss=0.1491, over 28311.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3944, pruned_loss=0.142, over 5719559.78 frames. ], libri_tot_loss[loss=0.3888, simple_loss=0.4303, pruned_loss=0.1737, over 5754173.09 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3888, pruned_loss=0.1363, over 5708556.02 frames. ], batch size: 368, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:24:20,933 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:24:46,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4325, 1.9921, 1.6695, 1.5154], device='cuda:1'), covar=tensor([0.0485, 0.0639, 0.0810, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0507, 0.0547, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 05:24:55,636 INFO [train.py:968] (1/2) Epoch 2, batch 22750, giga_loss[loss=0.3271, simple_loss=0.397, pruned_loss=0.1286, over 28990.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3956, pruned_loss=0.1433, over 5706067.00 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4311, pruned_loss=0.1747, over 5740818.16 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3886, pruned_loss=0.1361, over 5706453.14 frames. ], batch size: 213, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:25:00,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:03,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:24,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:30,973 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:33,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:25:38,127 INFO [optim.py:369] (1/2) 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 05:25:40,359 INFO [train.py:968] (1/2) Epoch 2, batch 22800, giga_loss[loss=0.3085, simple_loss=0.3689, pruned_loss=0.124, over 28653.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3966, pruned_loss=0.1422, over 5703574.65 frames. ], libri_tot_loss[loss=0.3906, simple_loss=0.4312, pruned_loss=0.175, over 5741726.98 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3902, pruned_loss=0.1357, over 5702382.01 frames. ], batch size: 78, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:25:51,825 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68480.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:26:13,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4533, 1.9608, 1.5417, 0.4522], device='cuda:1'), covar=tensor([0.1267, 0.0799, 0.1192, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.1105, 0.1160, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 05:26:22,122 INFO [train.py:968] (1/2) Epoch 2, batch 22850, giga_loss[loss=0.3185, simple_loss=0.3795, pruned_loss=0.1287, over 28804.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3983, pruned_loss=0.1424, over 5699764.44 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4307, pruned_loss=0.1748, over 5736771.88 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3925, pruned_loss=0.1363, over 5702084.83 frames. ], batch size: 174, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:27:02,765 INFO [optim.py:369] (1/2) 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 05:27:04,903 INFO [train.py:968] (1/2) Epoch 2, batch 22900, giga_loss[loss=0.3423, simple_loss=0.3937, pruned_loss=0.1454, over 29044.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3955, pruned_loss=0.1416, over 5699915.10 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4305, pruned_loss=0.1748, over 5739332.30 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3907, pruned_loss=0.1364, over 5699084.73 frames. ], batch size: 128, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:27:07,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2680, 1.5927, 1.0941, 1.4309], device='cuda:1'), covar=tensor([0.1034, 0.0409, 0.0496, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0289, 0.0198, 0.0198, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0031, 0.0024, 0.0021, 0.0035], device='cuda:1')
+2023-03-01 05:27:25,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:29,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:36,350 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:38,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:27:49,653 INFO [train.py:968] (1/2) Epoch 2, batch 22950, libri_loss[loss=0.4334, simple_loss=0.4617, pruned_loss=0.2025, over 19835.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3941, pruned_loss=0.1428, over 5689238.69 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4306, pruned_loss=0.1749, over 5731632.14 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3893, pruned_loss=0.1376, over 5695333.97 frames. ], batch size: 186, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:27:52,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:28:01,602 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:28:29,972 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 2, batch 23000, giga_loss[loss=0.3903, simple_loss=0.4224, pruned_loss=0.1791, over 28906.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3917, pruned_loss=0.1422, over 5693432.80 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.431, pruned_loss=0.1753, over 5723720.03 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.387, pruned_loss=0.1373, over 5705526.08 frames. ], batch size: 174, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:29:02,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7621, 3.1933, 3.4384, 1.5902], device='cuda:1'), covar=tensor([0.0682, 0.0521, 0.1102, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0545, 0.0801, 0.0566], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 05:29:14,355 INFO [train.py:968] (1/2) Epoch 2, batch 23050, giga_loss[loss=0.3133, simple_loss=0.3767, pruned_loss=0.1249, over 28864.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3924, pruned_loss=0.1446, over 5692801.16 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4324, pruned_loss=0.1766, over 5725168.62 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3864, pruned_loss=0.1386, over 5700421.71 frames. ], batch size: 186, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:29:40,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:29:54,332 INFO [optim.py:369] (1/2) 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 05:29:55,578 INFO [train.py:968] (1/2) Epoch 2, batch 23100, giga_loss[loss=0.3509, simple_loss=0.4011, pruned_loss=0.1504, over 28331.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3906, pruned_loss=0.1426, over 5708136.51 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.4327, pruned_loss=0.1769, over 5727875.60 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.385, pruned_loss=0.1371, over 5711479.72 frames. ], batch size: 368, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:30:09,260 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9518, 1.0512, 0.8043, 0.2391], device='cuda:1'), covar=tensor([0.0757, 0.0699, 0.0899, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.1133, 0.1111, 0.1166, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 05:30:15,002 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:30:36,303 INFO [train.py:968] (1/2) Epoch 2, batch 23150, giga_loss[loss=0.3428, simple_loss=0.3819, pruned_loss=0.1518, over 28893.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3862, pruned_loss=0.1402, over 5707577.66 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4331, pruned_loss=0.1774, over 5729427.21 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3805, pruned_loss=0.1349, over 5708376.15 frames. ], batch size: 112, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:31:09,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68855.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:31:18,770 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 23200, giga_loss[loss=0.3319, simple_loss=0.3816, pruned_loss=0.1412, over 27714.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3795, pruned_loss=0.1358, over 5702272.08 frames. ], libri_tot_loss[loss=0.394, simple_loss=0.4332, pruned_loss=0.1774, over 5726520.07 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3747, pruned_loss=0.1314, over 5705599.33 frames. ], batch size: 472, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:31:34,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-01 05:31:39,478 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:31:41,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:31:59,949 INFO [train.py:968] (1/2) Epoch 2, batch 23250, giga_loss[loss=0.3441, simple_loss=0.3908, pruned_loss=0.1487, over 28643.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3782, pruned_loss=0.1348, over 5709820.64 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4336, pruned_loss=0.1777, over 5730881.00 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3728, pruned_loss=0.13, over 5708047.69 frames. ], batch size: 85, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:32:05,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:32:44,451 INFO [optim.py:369] (1/2) 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 05:32:45,730 INFO [train.py:968] (1/2) Epoch 2, batch 23300, giga_loss[loss=0.3406, simple_loss=0.4017, pruned_loss=0.1398, over 29144.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3801, pruned_loss=0.1354, over 5711822.49 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4329, pruned_loss=0.1774, over 5733339.01 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3756, pruned_loss=0.1312, over 5707987.43 frames. ], batch size: 155, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:32:47,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=68971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:33:11,142 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68998.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:33:13,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69001.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:33:28,449 INFO [train.py:968] (1/2) Epoch 2, batch 23350, giga_loss[loss=0.3337, simple_loss=0.3969, pruned_loss=0.1353, over 28712.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3856, pruned_loss=0.1386, over 5706840.41 frames. ], libri_tot_loss[loss=0.3941, simple_loss=0.4332, pruned_loss=0.1775, over 5727642.65 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3807, pruned_loss=0.1343, over 5709046.80 frames. ], batch size: 262, lr: 1.32e-02, grad_scale: 8.0
+2023-03-01 05:33:38,393 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69030.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:33:56,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3848, 2.3703, 1.4347, 1.2369], device='cuda:1'), covar=tensor([0.0836, 0.0530, 0.0846, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0444, 0.0331, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0016, 0.0012, 0.0015], device='cuda:1')
+2023-03-01 05:34:10,039 INFO [optim.py:369] (1/2) 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 05:34:10,706 INFO [train.py:968] (1/2) Epoch 2, batch 23400, libri_loss[loss=0.4443, simple_loss=0.4669, pruned_loss=0.2108, over 20286.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3901, pruned_loss=0.1412, over 5700598.55 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4321, pruned_loss=0.1769, over 5723267.62 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3857, pruned_loss=0.1372, over 5706527.45 frames. ], batch size: 186, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:34:18,813 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:34:53,268 INFO [train.py:968] (1/2) Epoch 2, batch 23450, libri_loss[loss=0.4255, simple_loss=0.455, pruned_loss=0.198, over 25898.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3937, pruned_loss=0.1431, over 5697978.78 frames. ], libri_tot_loss[loss=0.3925, simple_loss=0.4316, pruned_loss=0.1767, over 5722568.16 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3897, pruned_loss=0.1392, over 5703111.26 frames. ], batch size: 136, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:35:41,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:35:41,973 INFO [optim.py:369] (1/2) 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 05:35:42,742 INFO [train.py:968] (1/2) Epoch 2, batch 23500, giga_loss[loss=0.3407, simple_loss=0.3968, pruned_loss=0.1423, over 28873.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3961, pruned_loss=0.1443, over 5681752.50 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4318, pruned_loss=0.1769, over 5714226.04 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3925, pruned_loss=0.1408, over 5693687.14 frames. ], batch size: 213, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:36:28,078 INFO [train.py:968] (1/2) Epoch 2, batch 23550, giga_loss[loss=0.3728, simple_loss=0.4141, pruned_loss=0.1657, over 28744.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4043, pruned_loss=0.1523, over 5685014.58 frames. ], libri_tot_loss[loss=0.3937, simple_loss=0.4325, pruned_loss=0.1775, over 5718551.28 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.3999, pruned_loss=0.1481, over 5689741.57 frames. ], batch size: 92, lr: 1.32e-02, grad_scale: 2.0
+2023-03-01 05:36:54,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3081, 1.7787, 1.5017, 1.3470], device='cuda:1'), covar=tensor([0.1099, 0.0400, 0.0462, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0195, 0.0197, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0031, 0.0023, 0.0021, 0.0035], device='cuda:1')
+2023-03-01 05:37:18,332 INFO [optim.py:369] (1/2) 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 05:37:18,344 INFO [train.py:968] (1/2) Epoch 2, batch 23600, giga_loss[loss=0.3487, simple_loss=0.4097, pruned_loss=0.1439, over 29009.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.41, pruned_loss=0.1576, over 5674071.01 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4325, pruned_loss=0.1777, over 5711319.11 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.4061, pruned_loss=0.1538, over 5683212.93 frames. ], batch size: 155, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:38:00,270 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:03,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:10,050 INFO [train.py:968] (1/2) Epoch 2, batch 23650, giga_loss[loss=0.3951, simple_loss=0.4448, pruned_loss=0.1727, over 28857.00 frames. ], tot_loss[loss=0.3727, simple_loss=0.4176, pruned_loss=0.1639, over 5674048.20 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4323, pruned_loss=0.1778, over 5712588.90 frames. ], giga_tot_loss[loss=0.3675, simple_loss=0.4143, pruned_loss=0.1604, over 5679289.44 frames. ], batch size: 145, lr: 1.32e-02, grad_scale: 4.0
+2023-03-01 05:38:33,005 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:36,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:38:56,788 INFO [optim.py:369] (1/2) 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 05:38:56,803 INFO [train.py:968] (1/2) Epoch 2, batch 23700, giga_loss[loss=0.5285, simple_loss=0.5141, pruned_loss=0.2714, over 27678.00 frames. ], tot_loss[loss=0.383, simple_loss=0.4245, pruned_loss=0.1708, over 5666663.00 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4332, pruned_loss=0.1787, over 5704278.29 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4206, pruned_loss=0.1667, over 5676493.96 frames. ], batch size: 472, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:39:02,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2970, 1.6032, 1.1886, 1.2897], device='cuda:1'), covar=tensor([0.1013, 0.0384, 0.0453, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0192, 0.0194, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0031, 0.0023, 0.0021, 0.0034], device='cuda:1')
+2023-03-01 05:39:21,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2757, 2.3013, 1.3826, 1.6128], device='cuda:1'), covar=tensor([0.0625, 0.0718, 0.1059, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0528, 0.0554, 0.0484], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 05:39:48,973 INFO [train.py:968] (1/2) Epoch 2, batch 23750, libri_loss[loss=0.4337, simple_loss=0.4672, pruned_loss=0.2001, over 29678.00 frames. ], tot_loss[loss=0.3959, simple_loss=0.433, pruned_loss=0.1794, over 5650830.02 frames. ], libri_tot_loss[loss=0.3957, simple_loss=0.4336, pruned_loss=0.179, over 5698476.98 frames. ], giga_tot_loss[loss=0.3905, simple_loss=0.4294, pruned_loss=0.1758, over 5662112.80 frames. ], batch size: 91, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:40:05,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:40:23,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:40:28,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-01 05:40:39,467 INFO [optim.py:369] (1/2) 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 05:40:39,479 INFO [train.py:968] (1/2) Epoch 2, batch 23800, giga_loss[loss=0.6116, simple_loss=0.5607, pruned_loss=0.3312, over 26541.00 frames. ], tot_loss[loss=0.4015, simple_loss=0.4374, pruned_loss=0.1828, over 5654792.39 frames. ], libri_tot_loss[loss=0.3958, simple_loss=0.4336, pruned_loss=0.179, over 5702109.22 frames. ], giga_tot_loss[loss=0.3972, simple_loss=0.4345, pruned_loss=0.1799, over 5659241.74 frames. ], batch size: 555, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:40:55,502 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:40:59,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:41:26,038 INFO [train.py:968] (1/2) Epoch 2, batch 23850, giga_loss[loss=0.387, simple_loss=0.4292, pruned_loss=0.1724, over 28897.00 frames. ], tot_loss[loss=0.406, simple_loss=0.4402, pruned_loss=0.1859, over 5643046.05 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.434, pruned_loss=0.1792, over 5692235.66 frames. ], giga_tot_loss[loss=0.4024, simple_loss=0.4377, pruned_loss=0.1835, over 5654929.57 frames. ], batch size: 227, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:41:27,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:02,323 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:08,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:17,669 INFO [optim.py:369] (1/2) 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 05:42:17,682 INFO [train.py:968] (1/2) Epoch 2, batch 23900, libri_loss[loss=0.4565, simple_loss=0.4802, pruned_loss=0.2163, over 29513.00 frames. ], tot_loss[loss=0.4097, simple_loss=0.4422, pruned_loss=0.1886, over 5637225.47 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4345, pruned_loss=0.1795, over 5695012.63 frames. ], giga_tot_loss[loss=0.4066, simple_loss=0.4399, pruned_loss=0.1866, over 5642745.43 frames. ], batch size: 81, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:42:44,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:42:49,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:43:08,265 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:43:08,659 INFO [train.py:968] (1/2) Epoch 2, batch 23950, giga_loss[loss=0.3959, simple_loss=0.4342, pruned_loss=0.1788, over 28847.00 frames. ], tot_loss[loss=0.4139, simple_loss=0.4446, pruned_loss=0.1916, over 5637921.40 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4342, pruned_loss=0.1796, over 5696402.09 frames. ], giga_tot_loss[loss=0.4118, simple_loss=0.4432, pruned_loss=0.1902, over 5640048.87 frames. ], batch size: 174, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:43:21,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:43:29,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69638.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:44:03,186 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:44:06,497 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 24000, giga_loss[loss=0.4961, simple_loss=0.4705, pruned_loss=0.2608, over 23477.00 frames. ], tot_loss[loss=0.4207, simple_loss=0.4492, pruned_loss=0.1961, over 5625550.06 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4343, pruned_loss=0.1799, over 5696762.48 frames. ], giga_tot_loss[loss=0.4192, simple_loss=0.4483, pruned_loss=0.195, over 5625258.71 frames. ], batch size: 705, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:44:06,509 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 05:44:15,957 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 05:45:06,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:45:11,154 INFO [train.py:968] (1/2) Epoch 2, batch 24050, giga_loss[loss=0.3744, simple_loss=0.4213, pruned_loss=0.1637, over 29032.00 frames. ], tot_loss[loss=0.42, simple_loss=0.4478, pruned_loss=0.1961, over 5611883.09 frames. ], libri_tot_loss[loss=0.3973, simple_loss=0.4345, pruned_loss=0.18, over 5700293.66 frames. ], giga_tot_loss[loss=0.419, simple_loss=0.4472, pruned_loss=0.1954, over 5606813.91 frames. ], batch size: 155, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:45:18,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-01 05:45:20,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7393, 1.5699, 1.2840, 1.4404], device='cuda:1'), covar=tensor([0.0667, 0.0797, 0.1023, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0525, 0.0555, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 05:45:52,201 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:46:00,275 INFO [train.py:968] (1/2) Epoch 2, batch 24100, giga_loss[loss=0.3989, simple_loss=0.4356, pruned_loss=0.1811, over 28905.00 frames. ], tot_loss[loss=0.4192, simple_loss=0.4466, pruned_loss=0.1959, over 5629314.38 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4343, pruned_loss=0.1799, over 5706062.59 frames. ], giga_tot_loss[loss=0.4193, simple_loss=0.4467, pruned_loss=0.196, over 5617520.85 frames. ], batch size: 285, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:46:01,658 INFO [optim.py:369] (1/2) 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 05:46:21,278 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=69787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:46:45,908 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 24150, giga_loss[loss=0.3944, simple_loss=0.4402, pruned_loss=0.1743, over 28838.00 frames. ], tot_loss[loss=0.4175, simple_loss=0.4456, pruned_loss=0.1947, over 5625556.34 frames. ], libri_tot_loss[loss=0.3972, simple_loss=0.4345, pruned_loss=0.1799, over 5707803.52 frames. ], giga_tot_loss[loss=0.4178, simple_loss=0.4456, pruned_loss=0.195, over 5613874.91 frames. ], batch size: 243, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:47:51,185 INFO [train.py:968] (1/2) Epoch 2, batch 24200, giga_loss[loss=0.5125, simple_loss=0.4953, pruned_loss=0.2649, over 26487.00 frames. ], tot_loss[loss=0.4143, simple_loss=0.4445, pruned_loss=0.1921, over 5622872.81 frames. ], libri_tot_loss[loss=0.3973, simple_loss=0.4346, pruned_loss=0.18, over 5708835.33 frames. ], giga_tot_loss[loss=0.4146, simple_loss=0.4445, pruned_loss=0.1923, over 5612039.78 frames. ], batch size: 555, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:47:52,188 INFO [optim.py:369] (1/2) 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 05:48:43,297 INFO [train.py:968] (1/2) Epoch 2, batch 24250, giga_loss[loss=0.5229, simple_loss=0.4974, pruned_loss=0.2742, over 26503.00 frames. ], tot_loss[loss=0.416, simple_loss=0.446, pruned_loss=0.193, over 5623723.12 frames. ], libri_tot_loss[loss=0.3981, simple_loss=0.4349, pruned_loss=0.1806, over 5712409.53 frames. ], giga_tot_loss[loss=0.416, simple_loss=0.446, pruned_loss=0.193, over 5609850.23 frames. ], batch size: 555, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:48:49,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5822, 1.4914, 1.2599, 1.3566], device='cuda:1'), covar=tensor([0.0515, 0.0516, 0.0759, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0521, 0.0550, 0.0481], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 05:48:56,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:02,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:21,092 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:24,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:49:35,979 INFO [train.py:968] (1/2) Epoch 2, batch 24300, giga_loss[loss=0.3255, simple_loss=0.3916, pruned_loss=0.1297, over 28898.00 frames. ], tot_loss[loss=0.4127, simple_loss=0.4439, pruned_loss=0.1907, over 5635307.11 frames. ], libri_tot_loss[loss=0.3981, simple_loss=0.4348, pruned_loss=0.1807, over 5716108.27 frames. ], giga_tot_loss[loss=0.4128, simple_loss=0.4441, pruned_loss=0.1908, over 5619574.57 frames. ], batch size: 145, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:49:37,466 INFO [optim.py:369] (1/2) 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 05:49:58,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6784, 2.1045, 1.7802, 1.7643], device='cuda:1'), covar=tensor([0.1435, 0.1599, 0.1180, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0814, 0.0704, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 05:50:00,461 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:50:06,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:50:27,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70013.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:50:31,007 INFO [train.py:968] (1/2) Epoch 2, batch 24350, giga_loss[loss=0.4191, simple_loss=0.4532, pruned_loss=0.1925, over 28609.00 frames. ], tot_loss[loss=0.4074, simple_loss=0.4409, pruned_loss=0.187, over 5640393.12 frames. ], libri_tot_loss[loss=0.3975, simple_loss=0.4343, pruned_loss=0.1804, over 5720532.28 frames. ], giga_tot_loss[loss=0.4084, simple_loss=0.4418, pruned_loss=0.1875, over 5621796.10 frames. ], batch size: 262, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:50:55,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:20,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=70067.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:51:21,648 INFO [train.py:968] (1/2) Epoch 2, batch 24400, giga_loss[loss=0.3949, simple_loss=0.4375, pruned_loss=0.1762, over 27982.00 frames. ], tot_loss[loss=0.4015, simple_loss=0.437, pruned_loss=0.183, over 5635320.47 frames. ], libri_tot_loss[loss=0.3973, simple_loss=0.434, pruned_loss=0.1803, over 5719467.29 frames. ], giga_tot_loss[loss=0.4027, simple_loss=0.4381, pruned_loss=0.1837, over 5619027.64 frames. ], batch size: 412, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:51:22,363 INFO [optim.py:369] (1/2) 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 05:51:27,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:30,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:33,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:34,970 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:39,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:51:56,962 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:03,467 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 24450, giga_loss[loss=0.3471, simple_loss=0.4029, pruned_loss=0.1456, over 29048.00 frames. ], tot_loss[loss=0.3952, simple_loss=0.4325, pruned_loss=0.179, over 5643095.62 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4335, pruned_loss=0.1802, over 5724083.59 frames. ], giga_tot_loss[loss=0.3965, simple_loss=0.4338, pruned_loss=0.1796, over 5623643.49 frames. ], batch size: 155, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:52:28,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:28,112 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:31,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:47,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70156.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:52:50,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70159.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 05:52:51,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0852, 1.2941, 1.0008, 0.2200], device='cuda:1'), covar=tensor([0.0752, 0.0706, 0.1207, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.1162, 0.1137, 0.1180, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 05:52:52,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:52:59,883 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:00,261 INFO [train.py:968] (1/2) Epoch 2, batch 24500, giga_loss[loss=0.3601, simple_loss=0.4097, pruned_loss=0.1552, over 28834.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.4307, pruned_loss=0.1773, over 5645621.13 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4336, pruned_loss=0.1801, over 5725190.61 frames. ], giga_tot_loss[loss=0.3936, simple_loss=0.4316, pruned_loss=0.1778, over 5627124.01 frames. ], batch size: 199, lr: 1.31e-02, grad_scale: 8.0
+2023-03-01 05:53:00,845 INFO [optim.py:369] (1/2) 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 05:53:16,262 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4828, 2.5423, 1.5043, 1.2616], device='cuda:1'), covar=tensor([0.0780, 0.0674, 0.0787, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0451, 0.0328, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0016, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 05:53:16,302 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:16,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7967, 1.9839, 1.7870, 1.7526], device='cuda:1'), covar=tensor([0.1188, 0.1338, 0.0954, 0.0653], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0815, 0.0704, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 05:53:18,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:20,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70188.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:53:20,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7045, 1.5493, 1.1365, 1.2795], device='cuda:1'), covar=tensor([0.0719, 0.0841, 0.1076, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0515, 0.0534, 0.0476], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 05:53:46,972 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:53:50,215 INFO [train.py:968] (1/2) Epoch 2, batch 24550, giga_loss[loss=0.3423, simple_loss=0.4005, pruned_loss=0.142, over 29018.00 frames. ], tot_loss[loss=0.3926, simple_loss=0.4305, pruned_loss=0.1774, over 5632432.93 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.4333, pruned_loss=0.1799, over 5718257.63 frames. ], giga_tot_loss[loss=0.3935, simple_loss=0.4314, pruned_loss=0.1779, over 5623023.72 frames. ], batch size: 128, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:54:02,196 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:05,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:39,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:46,149 INFO [train.py:968] (1/2) Epoch 2, batch 24600, giga_loss[loss=0.3223, simple_loss=0.388, pruned_loss=0.1283, over 28815.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.4292, pruned_loss=0.1755, over 5635125.93 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.433, pruned_loss=0.1797, over 5712250.79 frames. ], giga_tot_loss[loss=0.3911, simple_loss=0.4302, pruned_loss=0.1761, over 5630626.69 frames. ], batch size: 119, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:54:48,303 INFO [optim.py:369] (1/2) 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 05:54:56,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:54:59,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:23,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:27,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:29,199 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:55:39,462 INFO [train.py:968] (1/2) Epoch 2, batch 24650, giga_loss[loss=0.3147, simple_loss=0.387, pruned_loss=0.1212, over 28941.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4275, pruned_loss=0.1732, over 5642667.45 frames. ], libri_tot_loss[loss=0.3973, simple_loss=0.4336, pruned_loss=0.1805, over 5707253.48 frames. ], giga_tot_loss[loss=0.3865, simple_loss=0.4275, pruned_loss=0.1727, over 5642252.72 frames. ], batch size: 136, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:55:42,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2922, 1.7526, 1.2481, 0.5473], device='cuda:1'), covar=tensor([0.1347, 0.0757, 0.1145, 0.1604], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.1153, 0.1195, 0.1027], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 05:55:43,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6460, 3.2263, 1.7596, 1.4560], device='cuda:1'), covar=tensor([0.0914, 0.0461, 0.0897, 0.1496], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0456, 0.0332, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0016, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 05:55:58,380 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:56:29,874 INFO [train.py:968] (1/2) Epoch 2, batch 24700, giga_loss[loss=0.3542, simple_loss=0.4201, pruned_loss=0.1441, over 28994.00 frames. ], tot_loss[loss=0.3845, simple_loss=0.428, pruned_loss=0.1705, over 5651599.75 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4333, pruned_loss=0.1803, over 5706705.64 frames. ], giga_tot_loss[loss=0.3843, simple_loss=0.4282, pruned_loss=0.1701, over 5651007.59 frames. ], batch size: 136, lr: 1.31e-02, grad_scale: 4.0
+2023-03-01 05:56:33,432 INFO [optim.py:369] (1/2) 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 05:56:47,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8803, 1.6791, 1.3253, 1.4405], device='cuda:1'), covar=tensor([0.0525, 0.0621, 0.0829, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0520, 0.0539, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 05:57:27,214 INFO [train.py:968] (1/2) Epoch 2, batch 24750, giga_loss[loss=0.4095, simple_loss=0.4164, pruned_loss=0.2014, over 23474.00 frames. ], tot_loss[loss=0.3866, simple_loss=0.43, pruned_loss=0.1717, over 5639459.34 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.433, pruned_loss=0.1802, over 5695937.15 frames. ], giga_tot_loss[loss=0.3864, simple_loss=0.4303, pruned_loss=0.1713, over 5647299.04 frames. ], batch size: 705, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 05:57:41,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0170, 3.1661, 2.0666, 0.7186], device='cuda:1'), covar=tensor([0.1518, 0.0599, 0.0996, 0.1589], device='cuda:1'), in_proj_covar=tensor([0.1186, 0.1146, 0.1179, 0.1025], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 05:57:50,474 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 05:58:03,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4046, 2.2817, 1.4877, 1.2772], device='cuda:1'), covar=tensor([0.0791, 0.0588, 0.0798, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0451, 0.0329, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0016, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 05:58:18,410 INFO [train.py:968] (1/2) Epoch 2, batch 24800, giga_loss[loss=0.3772, simple_loss=0.4226, pruned_loss=0.1659, over 28893.00 frames. ], tot_loss[loss=0.3901, simple_loss=0.432, pruned_loss=0.1741, over 5654125.20 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4331, pruned_loss=0.1804, over 5699934.12 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4321, pruned_loss=0.1734, over 5655638.57 frames. ], batch size: 112, lr: 1.30e-02, grad_scale: 8.0
+2023-03-01 05:58:19,837 INFO [optim.py:369] (1/2) 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 05:58:51,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=70503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 05:59:05,395 INFO [train.py:968] (1/2) Epoch 2, batch 24850, giga_loss[loss=0.4036, simple_loss=0.4331, pruned_loss=0.1871, over 28533.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4311, pruned_loss=0.1738, over 5664269.73 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4327, pruned_loss=0.1801, over 5692760.65 frames. ], giga_tot_loss[loss=0.3891, simple_loss=0.4316, pruned_loss=0.1733, over 5669595.09 frames. ], batch size: 85, lr: 1.30e-02, grad_scale: 8.0
+2023-03-01 05:59:52,699 INFO [train.py:968] (1/2) Epoch 2, batch 24900, giga_loss[loss=0.3508, simple_loss=0.4023, pruned_loss=0.1497, over 28692.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4291, pruned_loss=0.1733, over 5667615.42 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4323, pruned_loss=0.1798, over 5689738.48 frames. ], giga_tot_loss[loss=0.3878, simple_loss=0.4297, pruned_loss=0.1729, over 5673524.40 frames. ], batch size: 262, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 05:59:55,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-01 05:59:57,012 INFO [optim.py:369] (1/2) 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 06:00:07,473 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70585.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:00:10,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70588.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 06:00:36,857 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 24950, giga_loss[loss=0.3958, simple_loss=0.4274, pruned_loss=0.1821, over 28338.00 frames. ], tot_loss[loss=0.3863, simple_loss=0.4273, pruned_loss=0.1727, over 5672717.21 frames. ], libri_tot_loss[loss=0.3956, simple_loss=0.4321, pruned_loss=0.1796, over 5694918.25 frames. ], giga_tot_loss[loss=0.3864, simple_loss=0.4278, pruned_loss=0.1725, over 5672439.22 frames. ], batch size: 368, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:00:41,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2198, 1.5536, 1.2523, 1.3569], device='cuda:1'), covar=tensor([0.0922, 0.0538, 0.0464, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0196, 0.0197, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0032, 0.0024, 0.0021, 0.0036], device='cuda:1')
+2023-03-01 06:01:11,896 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=70653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:01:19,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5978, 1.9779, 1.6629, 1.6621], device='cuda:1'), covar=tensor([0.1395, 0.1745, 0.1261, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0816, 0.0706, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 06:01:26,041 INFO [train.py:968] (1/2) Epoch 2, batch 25000, giga_loss[loss=0.3357, simple_loss=0.4024, pruned_loss=0.1345, over 28993.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4267, pruned_loss=0.1716, over 5670141.29 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4326, pruned_loss=0.18, over 5691547.20 frames. ], giga_tot_loss[loss=0.3842, simple_loss=0.4266, pruned_loss=0.1709, over 5671318.04 frames. ], batch size: 106, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:01:27,969 INFO [optim.py:369] (1/2) 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 06:02:14,566 INFO [train.py:968] (1/2) Epoch 2, batch 25050, giga_loss[loss=0.4125, simple_loss=0.4451, pruned_loss=0.1899, over 27918.00 frames. ], tot_loss[loss=0.3813, simple_loss=0.4252, pruned_loss=0.1687, over 5681532.39 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4326, pruned_loss=0.1802, over 5694738.39 frames. ], giga_tot_loss[loss=0.3804, simple_loss=0.425, pruned_loss=0.1679, over 5679566.27 frames. ], batch size: 412, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:03:03,358 INFO [train.py:968] (1/2) Epoch 2, batch 25100, giga_loss[loss=0.3648, simple_loss=0.4164, pruned_loss=0.1566, over 28930.00 frames. ], tot_loss[loss=0.3834, simple_loss=0.4265, pruned_loss=0.1701, over 5680245.81 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.4326, pruned_loss=0.1803, over 5700358.96 frames. ], giga_tot_loss[loss=0.3821, simple_loss=0.4262, pruned_loss=0.169, over 5673352.27 frames. ], batch size: 106, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:03:05,693 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.5807, 1.5509, 1.4460, 1.4867], device='cuda:1'), covar=tensor([0.0876, 0.1333, 0.1162, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0819, 0.0641, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 06:03:55,259 INFO [train.py:968] (1/2) Epoch 2, batch 25150, giga_loss[loss=0.3483, simple_loss=0.3994, pruned_loss=0.1486, over 28973.00 frames. ], tot_loss[loss=0.3805, simple_loss=0.4243, pruned_loss=0.1683, over 5676693.70 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4327, pruned_loss=0.1803, over 5694350.54 frames. ], giga_tot_loss[loss=0.3792, simple_loss=0.4238, pruned_loss=0.1672, over 5675356.66 frames. ], batch size: 136, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:03:58,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0873, 1.2709, 1.0394, 0.1415], device='cuda:1'), covar=tensor([0.0885, 0.0867, 0.1387, 0.1666], device='cuda:1'), in_proj_covar=tensor([0.1205, 0.1166, 0.1194, 0.1036], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 06:04:46,269 INFO [train.py:968] (1/2) Epoch 2, batch 25200, giga_loss[loss=0.3955, simple_loss=0.4379, pruned_loss=0.1765, over 28645.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4238, pruned_loss=0.1687, over 5673599.25 frames. ], libri_tot_loss[loss=0.3975, simple_loss=0.4334, pruned_loss=0.1808, over 5698139.19 frames. ], giga_tot_loss[loss=0.3785, simple_loss=0.4226, pruned_loss=0.1671, over 5668720.86 frames. ], batch size: 307, lr: 1.30e-02, grad_scale: 8.0
+2023-03-01 06:04:48,463 INFO [optim.py:369] (1/2) 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 06:04:54,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:05:33,980 INFO [train.py:968] (1/2) Epoch 2, batch 25250, giga_loss[loss=0.403, simple_loss=0.4328, pruned_loss=0.1866, over 28571.00 frames. ], tot_loss[loss=0.3814, simple_loss=0.4236, pruned_loss=0.1696, over 5669550.19 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4332, pruned_loss=0.1805, over 5702323.16 frames. ], giga_tot_loss[loss=0.3794, simple_loss=0.4225, pruned_loss=0.1682, over 5661041.03 frames. ], batch size: 336, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:05:56,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4196, 1.7492, 1.0648, 0.9771], device='cuda:1'), covar=tensor([0.0655, 0.0460, 0.0523, 0.0631], device='cuda:1'), in_proj_covar=tensor([0.1041, 0.0773, 0.0859, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 06:06:25,313 INFO [train.py:968] (1/2) Epoch 2, batch 25300, giga_loss[loss=0.4064, simple_loss=0.4224, pruned_loss=0.1951, over 23614.00 frames. ], tot_loss[loss=0.3839, simple_loss=0.4247, pruned_loss=0.1716, over 5669548.72 frames. ], libri_tot_loss[loss=0.3975, simple_loss=0.4335, pruned_loss=0.1807, over 5703176.54 frames. ], giga_tot_loss[loss=0.382, simple_loss=0.4235, pruned_loss=0.1702, over 5662001.47 frames. ], batch size: 705, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:06:28,621 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 25350, giga_loss[loss=0.4353, simple_loss=0.461, pruned_loss=0.2048, over 27986.00 frames. ], tot_loss[loss=0.3813, simple_loss=0.4226, pruned_loss=0.17, over 5672227.10 frames. ], libri_tot_loss[loss=0.3976, simple_loss=0.4337, pruned_loss=0.1807, over 5703467.84 frames. ], giga_tot_loss[loss=0.3795, simple_loss=0.4214, pruned_loss=0.1688, over 5665603.64 frames. ], batch size: 412, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:07:19,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:07:22,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:07:25,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:07:48,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:08:01,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3366, 2.0221, 1.4794, 0.5236], device='cuda:1'), covar=tensor([0.1527, 0.0910, 0.1475, 0.1893], device='cuda:1'), in_proj_covar=tensor([0.1200, 0.1163, 0.1203, 0.1034], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 06:08:05,388 INFO [train.py:968] (1/2) Epoch 2, batch 25400, giga_loss[loss=0.4173, simple_loss=0.4396, pruned_loss=0.1974, over 28790.00 frames. ], tot_loss[loss=0.3808, simple_loss=0.4217, pruned_loss=0.17, over 5669499.27 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4332, pruned_loss=0.1804, over 5706333.54 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.4207, pruned_loss=0.1689, over 5660508.41 frames. ], batch size: 99, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:08:09,895 INFO [optim.py:369] (1/2) 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 06:08:21,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4228, 1.8978, 1.4654, 0.6715], device='cuda:1'), covar=tensor([0.1336, 0.0778, 0.1033, 0.1621], device='cuda:1'), in_proj_covar=tensor([0.1185, 0.1160, 0.1196, 0.1026], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 06:08:25,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5029, 1.3564, 1.2937, 1.7105], device='cuda:1'), covar=tensor([0.1890, 0.2083, 0.1653, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.0963, 0.0823, 0.0877, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 06:08:52,578 INFO [train.py:968] (1/2) Epoch 2, batch 25450, giga_loss[loss=0.3843, simple_loss=0.4269, pruned_loss=0.1709, over 28898.00 frames. ], tot_loss[loss=0.3821, simple_loss=0.4225, pruned_loss=0.1708, over 5666452.47 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4323, pruned_loss=0.1798, over 5704105.64 frames. ], giga_tot_loss[loss=0.381, simple_loss=0.422, pruned_loss=0.17, over 5659428.51 frames. ], batch size: 227, lr: 1.30e-02, grad_scale: 2.0
+2023-03-01 06:09:42,995 INFO [train.py:968] (1/2) Epoch 2, batch 25500, giga_loss[loss=0.3521, simple_loss=0.4163, pruned_loss=0.1439, over 28973.00 frames. ], tot_loss[loss=0.3804, simple_loss=0.4226, pruned_loss=0.1691, over 5668522.22 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4326, pruned_loss=0.1799, over 5705008.69 frames. ], giga_tot_loss[loss=0.3792, simple_loss=0.4217, pruned_loss=0.1683, over 5661342.00 frames. ], batch size: 164, lr: 1.30e-02, grad_scale: 2.0
+2023-03-01 06:09:44,598 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:09:47,017 INFO [optim.py:369] (1/2) 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 06:09:47,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:10:11,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:10:25,005 INFO [train.py:968] (1/2) Epoch 2, batch 25550, giga_loss[loss=0.378, simple_loss=0.429, pruned_loss=0.1635, over 28878.00 frames. ], tot_loss[loss=0.3796, simple_loss=0.4221, pruned_loss=0.1685, over 5661342.87 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4325, pruned_loss=0.1801, over 5695212.67 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.421, pruned_loss=0.167, over 5662139.91 frames. ], batch size: 186, lr: 1.30e-02, grad_scale: 2.0
+2023-03-01 06:10:29,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4017, 1.2280, 1.2291, 1.6051], device='cuda:1'), covar=tensor([0.1925, 0.2041, 0.1693, 0.2058], device='cuda:1'), in_proj_covar=tensor([0.0977, 0.0820, 0.0889, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 06:11:17,075 INFO [train.py:968] (1/2) Epoch 2, batch 25600, giga_loss[loss=0.3447, simple_loss=0.3982, pruned_loss=0.1456, over 28938.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4227, pruned_loss=0.1688, over 5654558.53 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4322, pruned_loss=0.18, over 5696135.61 frames. ], giga_tot_loss[loss=0.3787, simple_loss=0.422, pruned_loss=0.1677, over 5654232.55 frames. ], batch size: 145, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:11:22,454 INFO [optim.py:369] (1/2) 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 06:12:06,903 INFO [train.py:968] (1/2) Epoch 2, batch 25650, giga_loss[loss=0.4261, simple_loss=0.4486, pruned_loss=0.2018, over 27945.00 frames. ], tot_loss[loss=0.3832, simple_loss=0.4244, pruned_loss=0.171, over 5648124.59 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4325, pruned_loss=0.1802, over 5686367.31 frames. ], giga_tot_loss[loss=0.3817, simple_loss=0.4236, pruned_loss=0.1699, over 5655439.82 frames. ], batch size: 412, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:12:39,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3568, 2.1053, 1.5143, 0.5504], device='cuda:1'), covar=tensor([0.1300, 0.0712, 0.1236, 0.1510], device='cuda:1'), in_proj_covar=tensor([0.1185, 0.1165, 0.1207, 0.1024], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 06:12:56,101 INFO [train.py:968] (1/2) Epoch 2, batch 25700, giga_loss[loss=0.3376, simple_loss=0.3938, pruned_loss=0.1407, over 28128.00 frames. ], tot_loss[loss=0.3875, simple_loss=0.4268, pruned_loss=0.1741, over 5640931.89 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4322, pruned_loss=0.18, over 5683180.89 frames. ], giga_tot_loss[loss=0.3863, simple_loss=0.4262, pruned_loss=0.1732, over 5648760.48 frames. ], batch size: 77, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:13:01,799 INFO [optim.py:369] (1/2) 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 06:13:15,018 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=71385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:13:33,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2154, 1.2952, 1.2041, 1.1970], device='cuda:1'), covar=tensor([0.1893, 0.1982, 0.1701, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.0977, 0.0823, 0.0891, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 06:13:33,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4951, 1.9703, 1.8003, 1.7354], device='cuda:1'), covar=tensor([0.1196, 0.1556, 0.1042, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0819, 0.0709, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 06:13:50,614 INFO [train.py:968] (1/2) Epoch 2, batch 25750, giga_loss[loss=0.4471, simple_loss=0.4615, pruned_loss=0.2164, over 28627.00 frames. ], tot_loss[loss=0.389, simple_loss=0.427, pruned_loss=0.1754, over 5646222.60 frames. ], libri_tot_loss[loss=0.3957, simple_loss=0.4318, pruned_loss=0.1798, over 5678507.46 frames. ], giga_tot_loss[loss=0.3881, simple_loss=0.4268, pruned_loss=0.1748, over 5655715.07 frames. ], batch size: 85, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:13:53,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=71423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:14:44,812 INFO [train.py:968] (1/2) Epoch 2, batch 25800, giga_loss[loss=0.5095, simple_loss=0.5043, pruned_loss=0.2574, over 24103.00 frames. ], tot_loss[loss=0.3935, simple_loss=0.4295, pruned_loss=0.1787, over 5624785.29 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4322, pruned_loss=0.18, over 5669621.05 frames. ], giga_tot_loss[loss=0.3924, simple_loss=0.4289, pruned_loss=0.1779, over 5639623.25 frames. ], batch size: 705, lr: 1.30e-02, grad_scale: 4.0
+2023-03-01 06:14:49,219 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 25850, libri_loss[loss=0.3442, simple_loss=0.3814, pruned_loss=0.1535, over 29638.00 frames. ], tot_loss[loss=0.3928, simple_loss=0.4292, pruned_loss=0.1782, over 5630882.64 frames. ], libri_tot_loss[loss=0.3958, simple_loss=0.4319, pruned_loss=0.1799, over 5663937.90 frames. ], giga_tot_loss[loss=0.3921, simple_loss=0.4289, pruned_loss=0.1777, over 5646423.48 frames. ], batch size: 69, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:16:09,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7593, 2.0575, 1.8235, 1.7308], device='cuda:1'), covar=tensor([0.1363, 0.1507, 0.1044, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0830, 0.0715, 0.0604], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 06:16:19,299 INFO [train.py:968] (1/2) Epoch 2, batch 25900, giga_loss[loss=0.3319, simple_loss=0.3749, pruned_loss=0.1444, over 28805.00 frames. ], tot_loss[loss=0.3923, simple_loss=0.4286, pruned_loss=0.178, over 5637257.73 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4322, pruned_loss=0.1801, over 5666051.27 frames. ], giga_tot_loss[loss=0.3913, simple_loss=0.428, pruned_loss=0.1773, over 5647417.50 frames. ], batch size: 92, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:16:23,927 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 25950, giga_loss[loss=0.3456, simple_loss=0.4076, pruned_loss=0.1418, over 28872.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.4267, pruned_loss=0.174, over 5659076.89 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4325, pruned_loss=0.1805, over 5671631.64 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4258, pruned_loss=0.173, over 5661800.89 frames. ], batch size: 145, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:17:52,800 INFO [train.py:968] (1/2) Epoch 2, batch 26000, giga_loss[loss=0.4172, simple_loss=0.4323, pruned_loss=0.2011, over 27583.00 frames. ], tot_loss[loss=0.3822, simple_loss=0.4224, pruned_loss=0.171, over 5651309.12 frames. ], libri_tot_loss[loss=0.3972, simple_loss=0.4328, pruned_loss=0.1808, over 5674154.37 frames. ], giga_tot_loss[loss=0.3804, simple_loss=0.4213, pruned_loss=0.1698, over 5651071.12 frames. ], batch size: 472, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:17:57,295 INFO [optim.py:369] (1/2) 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 06:18:15,882 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-01 06:18:41,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3164, 1.8215, 1.7348, 1.7159], device='cuda:1'), covar=tensor([0.0754, 0.1760, 0.1259, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0791, 0.0623, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 06:18:41,635 INFO [train.py:968] (1/2) Epoch 2, batch 26050, giga_loss[loss=0.3429, simple_loss=0.3992, pruned_loss=0.1433, over 28813.00 frames. ], tot_loss[loss=0.3782, simple_loss=0.4194, pruned_loss=0.1685, over 5666081.15 frames. ], libri_tot_loss[loss=0.3969, simple_loss=0.4326, pruned_loss=0.1806, over 5679990.52 frames. ], giga_tot_loss[loss=0.3767, simple_loss=0.4185, pruned_loss=0.1674, over 5660339.27 frames. ], batch size: 243, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:19:21,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:19:29,898 INFO [train.py:968] (1/2) Epoch 2, batch 26100, giga_loss[loss=0.3624, simple_loss=0.4038, pruned_loss=0.1605, over 28666.00 frames. ], tot_loss[loss=0.3772, simple_loss=0.4181, pruned_loss=0.1682, over 5675964.28 frames. ], libri_tot_loss[loss=0.398, simple_loss=0.4333, pruned_loss=0.1814, over 5681728.01 frames. ], giga_tot_loss[loss=0.3748, simple_loss=0.4165, pruned_loss=0.1665, over 5669765.41 frames. ], batch size: 307, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:19:36,631 INFO [optim.py:369] (1/2) 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 06:19:59,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 06:20:02,463 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 26150, giga_loss[loss=0.4446, simple_loss=0.4477, pruned_loss=0.2207, over 23646.00 frames. ], tot_loss[loss=0.3788, simple_loss=0.4195, pruned_loss=0.169, over 5666742.19 frames. ], libri_tot_loss[loss=0.3987, simple_loss=0.4338, pruned_loss=0.1818, over 5674372.21 frames. ], giga_tot_loss[loss=0.3761, simple_loss=0.4177, pruned_loss=0.1673, over 5667629.85 frames. ], batch size: 705, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:21:10,384 INFO [train.py:968] (1/2) Epoch 2, batch 26200, giga_loss[loss=0.4438, simple_loss=0.4871, pruned_loss=0.2003, over 28848.00 frames. ], tot_loss[loss=0.3833, simple_loss=0.4245, pruned_loss=0.1711, over 5679063.83 frames. ], libri_tot_loss[loss=0.3985, simple_loss=0.4338, pruned_loss=0.1817, over 5679545.10 frames. ], giga_tot_loss[loss=0.3809, simple_loss=0.4228, pruned_loss=0.1695, over 5675121.00 frames. ], batch size: 174, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:21:17,349 INFO [optim.py:369] (1/2) 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 06:21:44,580 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:21:49,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:01,125 INFO [train.py:968] (1/2) Epoch 2, batch 26250, giga_loss[loss=0.3824, simple_loss=0.4418, pruned_loss=0.1615, over 29018.00 frames. ], tot_loss[loss=0.3833, simple_loss=0.4269, pruned_loss=0.1698, over 5678340.13 frames. ], libri_tot_loss[loss=0.3985, simple_loss=0.4338, pruned_loss=0.1817, over 5684132.61 frames. ], giga_tot_loss[loss=0.3811, simple_loss=0.4255, pruned_loss=0.1683, over 5670993.60 frames. ], batch size: 128, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:22:14,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:21,035 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:23,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:22:49,229 INFO [train.py:968] (1/2) Epoch 2, batch 26300, giga_loss[loss=0.4809, simple_loss=0.4952, pruned_loss=0.2333, over 28847.00 frames. ], tot_loss[loss=0.3859, simple_loss=0.4293, pruned_loss=0.1713, over 5686741.13 frames. ], libri_tot_loss[loss=0.3987, simple_loss=0.4337, pruned_loss=0.1818, over 5687757.01 frames. ], giga_tot_loss[loss=0.3837, simple_loss=0.428, pruned_loss=0.1697, over 5677768.30 frames. ], batch size: 199, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:22:54,253 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:23:27,406 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:23:36,331 INFO [train.py:968] (1/2) Epoch 2, batch 26350, giga_loss[loss=0.3518, simple_loss=0.4008, pruned_loss=0.1514, over 28206.00 frames. ], tot_loss[loss=0.3905, simple_loss=0.4322, pruned_loss=0.1744, over 5677454.43 frames. ], libri_tot_loss[loss=0.3996, simple_loss=0.4344, pruned_loss=0.1824, over 5684635.18 frames. ], giga_tot_loss[loss=0.3876, simple_loss=0.4305, pruned_loss=0.1724, over 5673305.62 frames. ], batch size: 77, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:23:54,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2764, 1.8073, 1.3441, 0.3973], device='cuda:1'), covar=tensor([0.1103, 0.0747, 0.1323, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.1179, 0.1174, 0.1184, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 06:24:21,741 INFO [train.py:968] (1/2) Epoch 2, batch 26400, giga_loss[loss=0.3821, simple_loss=0.4197, pruned_loss=0.1722, over 28553.00 frames. ], tot_loss[loss=0.3921, simple_loss=0.4329, pruned_loss=0.1757, over 5677319.63 frames. ], libri_tot_loss[loss=0.3992, simple_loss=0.4342, pruned_loss=0.1821, over 5686295.13 frames. ], giga_tot_loss[loss=0.3898, simple_loss=0.4317, pruned_loss=0.174, over 5672677.61 frames. ], batch size: 336, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:24:27,884 INFO [optim.py:369] (1/2) 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 06:25:11,174 INFO [train.py:968] (1/2) Epoch 2, batch 26450, giga_loss[loss=0.4358, simple_loss=0.45, pruned_loss=0.2108, over 26494.00 frames. ], tot_loss[loss=0.392, simple_loss=0.4321, pruned_loss=0.176, over 5668053.43 frames. ], libri_tot_loss[loss=0.3992, simple_loss=0.4342, pruned_loss=0.1821, over 5674358.56 frames. ], giga_tot_loss[loss=0.3902, simple_loss=0.4311, pruned_loss=0.1746, over 5674287.76 frames. ], batch size: 555, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:25:43,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 06:26:00,603 INFO [train.py:968] (1/2) Epoch 2, batch 26500, giga_loss[loss=0.3477, simple_loss=0.4021, pruned_loss=0.1466, over 28842.00 frames. ], tot_loss[loss=0.3892, simple_loss=0.4295, pruned_loss=0.1745, over 5681243.76 frames. ], libri_tot_loss[loss=0.3984, simple_loss=0.4336, pruned_loss=0.1816, over 5681425.54 frames. ], giga_tot_loss[loss=0.3883, simple_loss=0.4292, pruned_loss=0.1737, over 5679881.77 frames. ], batch size: 174, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:26:07,346 INFO [optim.py:369] (1/2) 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 06:26:13,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4894, 2.5836, 1.4174, 1.2984], device='cuda:1'), covar=tensor([0.0861, 0.0523, 0.0855, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0459, 0.0334, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 06:26:49,575 INFO [train.py:968] (1/2) Epoch 2, batch 26550, giga_loss[loss=0.4173, simple_loss=0.445, pruned_loss=0.1948, over 27585.00 frames. ], tot_loss[loss=0.3877, simple_loss=0.4275, pruned_loss=0.1739, over 5675813.34 frames. ], libri_tot_loss[loss=0.3984, simple_loss=0.4336, pruned_loss=0.1815, over 5675155.22 frames. ], giga_tot_loss[loss=0.3866, simple_loss=0.4271, pruned_loss=0.1731, over 5680672.59 frames. ], batch size: 472, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:27:41,770 INFO [train.py:968] (1/2) Epoch 2, batch 26600, giga_loss[loss=0.3914, simple_loss=0.431, pruned_loss=0.1759, over 28928.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4252, pruned_loss=0.1724, over 5675054.06 frames. ], libri_tot_loss[loss=0.398, simple_loss=0.4333, pruned_loss=0.1813, over 5677866.51 frames. ], giga_tot_loss[loss=0.3842, simple_loss=0.425, pruned_loss=0.1718, over 5676435.56 frames. ], batch size: 119, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:27:49,001 INFO [optim.py:369] (1/2) 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 06:28:23,666 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:28:31,986 INFO [train.py:968] (1/2) Epoch 2, batch 26650, giga_loss[loss=0.3536, simple_loss=0.4014, pruned_loss=0.1529, over 28877.00 frames. ], tot_loss[loss=0.3868, simple_loss=0.4264, pruned_loss=0.1736, over 5668148.09 frames. ], libri_tot_loss[loss=0.3983, simple_loss=0.4335, pruned_loss=0.1815, over 5670964.20 frames. ], giga_tot_loss[loss=0.3858, simple_loss=0.4259, pruned_loss=0.1728, over 5675099.82 frames. ], batch size: 186, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:28:51,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3301, 1.3088, 1.1906, 1.6436], device='cuda:1'), covar=tensor([0.1699, 0.1563, 0.1369, 0.1845], device='cuda:1'), in_proj_covar=tensor([0.0987, 0.0827, 0.0900, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 06:29:18,663 INFO [train.py:968] (1/2) Epoch 2, batch 26700, giga_loss[loss=0.3084, simple_loss=0.3634, pruned_loss=0.1267, over 28251.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4254, pruned_loss=0.1737, over 5642224.97 frames. ], libri_tot_loss[loss=0.3994, simple_loss=0.4342, pruned_loss=0.1822, over 5645532.71 frames. ], giga_tot_loss[loss=0.3844, simple_loss=0.4242, pruned_loss=0.1723, over 5671807.14 frames. ], batch size: 77, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:29:24,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 06:29:26,892 INFO [optim.py:369] (1/2) 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 06:29:32,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:30:06,398 INFO [train.py:968] (1/2) Epoch 2, batch 26750, giga_loss[loss=0.3356, simple_loss=0.3948, pruned_loss=0.1382, over 28916.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4234, pruned_loss=0.1731, over 5633640.08 frames. ], libri_tot_loss[loss=0.3988, simple_loss=0.4337, pruned_loss=0.1819, over 5651423.27 frames. ], giga_tot_loss[loss=0.3832, simple_loss=0.4226, pruned_loss=0.1719, over 5652201.94 frames. ], batch size: 145, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:30:54,583 INFO [train.py:968] (1/2) Epoch 2, batch 26800, giga_loss[loss=0.3988, simple_loss=0.4382, pruned_loss=0.1797, over 28543.00 frames. ], tot_loss[loss=0.386, simple_loss=0.425, pruned_loss=0.1735, over 5640869.36 frames. ], libri_tot_loss[loss=0.3989, simple_loss=0.4337, pruned_loss=0.182, over 5648872.00 frames. ], giga_tot_loss[loss=0.3843, simple_loss=0.424, pruned_loss=0.1722, over 5658137.28 frames. ], batch size: 336, lr: 1.29e-02, grad_scale: 8.0
+2023-03-01 06:30:59,425 INFO [optim.py:369] (1/2) 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 06:31:05,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0376, 4.2369, 4.7306, 2.1764], device='cuda:1'), covar=tensor([0.0341, 0.0324, 0.0618, 0.1587], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0575, 0.0834, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 06:31:46,412 INFO [train.py:968] (1/2) Epoch 2, batch 26850, giga_loss[loss=0.464, simple_loss=0.477, pruned_loss=0.2254, over 27510.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4262, pruned_loss=0.1734, over 5649837.49 frames. ], libri_tot_loss[loss=0.3982, simple_loss=0.4331, pruned_loss=0.1816, over 5655285.58 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4257, pruned_loss=0.1725, over 5658138.64 frames. ], batch size: 472, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:31:48,494 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:31:52,969 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:31:55,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:32:27,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6737, 2.6179, 1.4768, 1.4622], device='cuda:1'), covar=tensor([0.0877, 0.0568, 0.0941, 0.1488], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0459, 0.0333, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 06:32:28,149 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:32:36,982 INFO [train.py:968] (1/2) Epoch 2, batch 26900, giga_loss[loss=0.4014, simple_loss=0.4349, pruned_loss=0.184, over 28267.00 frames. ], tot_loss[loss=0.3882, simple_loss=0.4273, pruned_loss=0.1746, over 5649209.01 frames. ], libri_tot_loss[loss=0.3976, simple_loss=0.4328, pruned_loss=0.1813, over 5659138.25 frames. ], giga_tot_loss[loss=0.3876, simple_loss=0.4271, pruned_loss=0.1741, over 5652059.55 frames. ], batch size: 368, lr: 1.29e-02, grad_scale: 4.0
+2023-03-01 06:32:44,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-01 06:32:45,355 INFO [optim.py:369] (1/2) 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 06:33:02,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 06:33:25,514 INFO [train.py:968] (1/2) Epoch 2, batch 26950, giga_loss[loss=0.3461, simple_loss=0.4178, pruned_loss=0.1372, over 28925.00 frames. ], tot_loss[loss=0.3877, simple_loss=0.4277, pruned_loss=0.1739, over 5656061.28 frames. ], libri_tot_loss[loss=0.3975, simple_loss=0.4325, pruned_loss=0.1812, over 5649760.16 frames. ], giga_tot_loss[loss=0.3872, simple_loss=0.4277, pruned_loss=0.1733, over 5667202.05 frames. ], batch size: 213, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:34:12,133 INFO [train.py:968] (1/2) Epoch 2, batch 27000, giga_loss[loss=0.3713, simple_loss=0.4279, pruned_loss=0.1573, over 28919.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4279, pruned_loss=0.1708, over 5667906.66 frames. ], libri_tot_loss[loss=0.3974, simple_loss=0.4323, pruned_loss=0.1813, over 5657326.67 frames. ], giga_tot_loss[loss=0.3841, simple_loss=0.4279, pruned_loss=0.1701, over 5669876.92 frames. ], batch size: 186, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:34:12,133 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 06:34:17,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7441, 3.3381, 3.4403, 1.6530], device='cuda:1'), covar=tensor([0.0722, 0.0530, 0.0969, 0.2283], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0568, 0.0828, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 06:34:21,353 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 06:34:26,750 INFO [optim.py:369] (1/2) 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 06:34:38,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:34:50,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6176, 2.0943, 1.3021, 1.0828], device='cuda:1'), covar=tensor([0.0621, 0.0441, 0.0459, 0.0561], device='cuda:1'), in_proj_covar=tensor([0.1056, 0.0793, 0.0895, 0.0904], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 06:35:07,031 INFO [train.py:968] (1/2) Epoch 2, batch 27050, giga_loss[loss=0.3796, simple_loss=0.4412, pruned_loss=0.159, over 29073.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.4308, pruned_loss=0.1715, over 5679282.01 frames. ], libri_tot_loss[loss=0.3977, simple_loss=0.4325, pruned_loss=0.1815, over 5663990.08 frames. ], giga_tot_loss[loss=0.3856, simple_loss=0.4305, pruned_loss=0.1703, over 5675362.32 frames. ], batch size: 155, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:35:07,233 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=72719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:35:50,299 INFO [train.py:968] (1/2) Epoch 2, batch 27100, giga_loss[loss=0.3548, simple_loss=0.4056, pruned_loss=0.152, over 28995.00 frames. ], tot_loss[loss=0.3885, simple_loss=0.4319, pruned_loss=0.1725, over 5685225.64 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4313, pruned_loss=0.1806, over 5672049.27 frames. ], giga_tot_loss[loss=0.3884, simple_loss=0.4328, pruned_loss=0.1721, over 5675440.97 frames. ], batch size: 128, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:35:57,389 INFO [optim.py:369] (1/2) 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 06:36:45,858 INFO [train.py:968] (1/2) Epoch 2, batch 27150, giga_loss[loss=0.4062, simple_loss=0.4462, pruned_loss=0.183, over 28751.00 frames. ], tot_loss[loss=0.3935, simple_loss=0.4347, pruned_loss=0.1762, over 5679174.92 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4313, pruned_loss=0.1806, over 5673821.83 frames. ], giga_tot_loss[loss=0.3935, simple_loss=0.4354, pruned_loss=0.1758, over 5670136.98 frames. ], batch size: 284, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:36:49,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1247, 1.6774, 1.2717, 1.2685], device='cuda:1'), covar=tensor([0.1053, 0.0464, 0.0444, 0.1307], device='cuda:1'), in_proj_covar=tensor([0.0291, 0.0196, 0.0196, 0.0348], device='cuda:1'), out_proj_covar=tensor([0.0033, 0.0025, 0.0022, 0.0037], device='cuda:1')
+2023-03-01 06:36:52,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 06:36:55,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:36:59,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:37:29,606 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:37:36,508 INFO [train.py:968] (1/2) Epoch 2, batch 27200, giga_loss[loss=0.3973, simple_loss=0.4307, pruned_loss=0.182, over 28934.00 frames. ], tot_loss[loss=0.3969, simple_loss=0.4361, pruned_loss=0.1789, over 5662629.05 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4307, pruned_loss=0.1799, over 5677770.18 frames. ], giga_tot_loss[loss=0.3978, simple_loss=0.4373, pruned_loss=0.1791, over 5651562.23 frames. ], batch size: 106, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:37:44,627 INFO [optim.py:369] (1/2) 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 06:38:03,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:38:14,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-01 06:38:20,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1860, 1.2585, 1.0197, 1.5380], device='cuda:1'), covar=tensor([0.2014, 0.1979, 0.1897, 0.2141], device='cuda:1'), in_proj_covar=tensor([0.0964, 0.0810, 0.0884, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 06:38:24,989 INFO [train.py:968] (1/2) Epoch 2, batch 27250, giga_loss[loss=0.3749, simple_loss=0.426, pruned_loss=0.1619, over 28999.00 frames. ], tot_loss[loss=0.394, simple_loss=0.4339, pruned_loss=0.1771, over 5669559.53 frames. ], libri_tot_loss[loss=0.395, simple_loss=0.4305, pruned_loss=0.1798, over 5683765.10 frames. ], giga_tot_loss[loss=0.3949, simple_loss=0.4352, pruned_loss=0.1773, over 5654972.65 frames. ], batch size: 128, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:38:43,739 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-01 06:39:06,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6622, 1.4190, 1.4637, 1.3735], device='cuda:1'), covar=tensor([0.0684, 0.1327, 0.1172, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0809, 0.0624, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 06:39:15,393 INFO [train.py:968] (1/2) Epoch 2, batch 27300, giga_loss[loss=0.4298, simple_loss=0.4703, pruned_loss=0.1946, over 28877.00 frames. ], tot_loss[loss=0.3919, simple_loss=0.4325, pruned_loss=0.1757, over 5657404.46 frames. ], libri_tot_loss[loss=0.3941, simple_loss=0.4298, pruned_loss=0.1793, over 5685652.19 frames. ], giga_tot_loss[loss=0.3933, simple_loss=0.4342, pruned_loss=0.1762, over 5643911.43 frames. ], batch size: 186, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:39:21,703 INFO [optim.py:369] (1/2) 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 06:40:02,645 INFO [train.py:968] (1/2) Epoch 2, batch 27350, giga_loss[loss=0.3694, simple_loss=0.4278, pruned_loss=0.1555, over 28835.00 frames. ], tot_loss[loss=0.3889, simple_loss=0.4318, pruned_loss=0.173, over 5666767.44 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4299, pruned_loss=0.1794, over 5686760.00 frames. ], giga_tot_loss[loss=0.3897, simple_loss=0.4331, pruned_loss=0.1731, over 5654653.52 frames. ], batch size: 119, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:40:23,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:40:25,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:40:50,330 INFO [train.py:968] (1/2) Epoch 2, batch 27400, giga_loss[loss=0.4383, simple_loss=0.4366, pruned_loss=0.2201, over 23483.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4315, pruned_loss=0.1721, over 5670211.46 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4296, pruned_loss=0.1791, over 5694561.26 frames. ], giga_tot_loss[loss=0.3887, simple_loss=0.4329, pruned_loss=0.1722, over 5652979.39 frames. ], batch size: 705, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:40:53,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:40:57,797 INFO [optim.py:369] (1/2) 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 06:41:18,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:41:42,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7095, 1.9976, 1.8376, 1.7763], device='cuda:1'), covar=tensor([0.1202, 0.1432, 0.0964, 0.0670], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0838, 0.0719, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 06:41:43,431 INFO [train.py:968] (1/2) Epoch 2, batch 27450, giga_loss[loss=0.3689, simple_loss=0.4164, pruned_loss=0.1607, over 28580.00 frames. ], tot_loss[loss=0.3897, simple_loss=0.4327, pruned_loss=0.1734, over 5672343.21 frames. ], libri_tot_loss[loss=0.3934, simple_loss=0.4291, pruned_loss=0.1788, over 5692326.50 frames. ], giga_tot_loss[loss=0.3907, simple_loss=0.4342, pruned_loss=0.1736, over 5659725.61 frames. ], batch size: 71, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:42:24,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6265, 4.0985, 4.3949, 1.7665], device='cuda:1'), covar=tensor([0.0459, 0.0357, 0.0742, 0.1839], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0572, 0.0826, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0007], device='cuda:1')
+2023-03-01 06:42:29,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-01 06:42:30,668 INFO [train.py:968] (1/2) Epoch 2, batch 27500, giga_loss[loss=0.3581, simple_loss=0.4084, pruned_loss=0.1539, over 28556.00 frames. ], tot_loss[loss=0.3918, simple_loss=0.4337, pruned_loss=0.1749, over 5677704.31 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4298, pruned_loss=0.179, over 5697589.58 frames. ], giga_tot_loss[loss=0.392, simple_loss=0.4345, pruned_loss=0.1747, over 5662190.39 frames. ], batch size: 336, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:42:40,128 INFO [optim.py:369] (1/2) 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 06:43:25,315 INFO [train.py:968] (1/2) Epoch 2, batch 27550, giga_loss[loss=0.3336, simple_loss=0.391, pruned_loss=0.1381, over 28857.00 frames. ], tot_loss[loss=0.3904, simple_loss=0.4314, pruned_loss=0.1748, over 5669306.47 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4301, pruned_loss=0.1792, over 5700358.59 frames. ], giga_tot_loss[loss=0.3903, simple_loss=0.4317, pruned_loss=0.1745, over 5654259.46 frames. ], batch size: 174, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:43:45,258 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:43:47,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:44:15,761 INFO [train.py:968] (1/2) Epoch 2, batch 27600, giga_loss[loss=0.4656, simple_loss=0.4738, pruned_loss=0.2286, over 27965.00 frames. ], tot_loss[loss=0.3906, simple_loss=0.4303, pruned_loss=0.1754, over 5654797.70 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4304, pruned_loss=0.1794, over 5701691.08 frames. ], giga_tot_loss[loss=0.3901, simple_loss=0.4304, pruned_loss=0.1749, over 5640415.39 frames. ], batch size: 412, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:44:16,377 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:44:27,714 INFO [optim.py:369] (1/2) 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 06:44:37,266 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 27650, giga_loss[loss=0.4772, simple_loss=0.477, pruned_loss=0.2387, over 27648.00 frames. ], tot_loss[loss=0.3889, simple_loss=0.4287, pruned_loss=0.1746, over 5653591.59 frames. ], libri_tot_loss[loss=0.3955, simple_loss=0.4311, pruned_loss=0.1799, over 5695542.99 frames. ], giga_tot_loss[loss=0.3875, simple_loss=0.428, pruned_loss=0.1736, over 5646947.61 frames. ], batch size: 474, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:45:48,773 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=73357.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:45:56,958 INFO [train.py:968] (1/2) Epoch 2, batch 27700, giga_loss[loss=0.4007, simple_loss=0.4364, pruned_loss=0.1826, over 28682.00 frames. ], tot_loss[loss=0.3881, simple_loss=0.4272, pruned_loss=0.1746, over 5637544.67 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4309, pruned_loss=0.1798, over 5685254.34 frames. ], giga_tot_loss[loss=0.387, simple_loss=0.4266, pruned_loss=0.1737, over 5640376.53 frames. ], batch size: 92, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:46:05,249 INFO [optim.py:369] (1/2) 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 06:46:44,075 INFO [train.py:968] (1/2) Epoch 2, batch 27750, giga_loss[loss=0.38, simple_loss=0.4275, pruned_loss=0.1662, over 28989.00 frames. ], tot_loss[loss=0.3858, simple_loss=0.4254, pruned_loss=0.1731, over 5637333.97 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4309, pruned_loss=0.1798, over 5677522.59 frames. ], giga_tot_loss[loss=0.3847, simple_loss=0.4249, pruned_loss=0.1722, over 5645973.30 frames. ], batch size: 145, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:46:53,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-01 06:47:05,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4879, 2.5600, 1.4711, 1.2433], device='cuda:1'), covar=tensor([0.0921, 0.0527, 0.0957, 0.1507], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0462, 0.0334, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 06:47:23,162 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 06:47:31,651 INFO [train.py:968] (1/2) Epoch 2, batch 27800, giga_loss[loss=0.3368, simple_loss=0.3988, pruned_loss=0.1375, over 28854.00 frames. ], tot_loss[loss=0.3771, simple_loss=0.4202, pruned_loss=0.1669, over 5651871.65 frames. ], libri_tot_loss[loss=0.395, simple_loss=0.4307, pruned_loss=0.1796, over 5682485.94 frames. ], giga_tot_loss[loss=0.3761, simple_loss=0.4198, pruned_loss=0.1662, over 5653700.62 frames. ], batch size: 199, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:47:41,419 INFO [optim.py:369] (1/2) 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 06:47:42,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-01 06:47:59,420 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-01 06:48:23,354 INFO [train.py:968] (1/2) Epoch 2, batch 27850, libri_loss[loss=0.4027, simple_loss=0.4314, pruned_loss=0.187, over 29574.00 frames. ], tot_loss[loss=0.3731, simple_loss=0.418, pruned_loss=0.1641, over 5660704.10 frames. ], libri_tot_loss[loss=0.3949, simple_loss=0.4306, pruned_loss=0.1796, over 5685346.54 frames. ], giga_tot_loss[loss=0.372, simple_loss=0.4175, pruned_loss=0.1633, over 5659079.37 frames. ], batch size: 78, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:49:16,142 INFO [train.py:968] (1/2) Epoch 2, batch 27900, giga_loss[loss=0.4515, simple_loss=0.4592, pruned_loss=0.2219, over 27604.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4182, pruned_loss=0.1653, over 5649938.78 frames. ], libri_tot_loss[loss=0.3952, simple_loss=0.4306, pruned_loss=0.1799, over 5687103.59 frames. ], giga_tot_loss[loss=0.3729, simple_loss=0.4176, pruned_loss=0.1641, over 5646633.31 frames. ], batch size: 472, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:49:25,755 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 27950, giga_loss[loss=0.3583, simple_loss=0.4034, pruned_loss=0.1566, over 28855.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4135, pruned_loss=0.1626, over 5663252.40 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4302, pruned_loss=0.1796, over 5690324.83 frames. ], giga_tot_loss[loss=0.3683, simple_loss=0.4132, pruned_loss=0.1617, over 5657385.88 frames. ], batch size: 199, lr: 1.28e-02, grad_scale: 4.0
+2023-03-01 06:51:03,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:51:08,714 INFO [train.py:968] (1/2) Epoch 2, batch 28000, giga_loss[loss=0.3824, simple_loss=0.4271, pruned_loss=0.1688, over 28912.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.4143, pruned_loss=0.1637, over 5660316.00 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.43, pruned_loss=0.1794, over 5690160.44 frames. ], giga_tot_loss[loss=0.3698, simple_loss=0.4139, pruned_loss=0.1628, over 5655235.90 frames. ], batch size: 145, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:51:18,448 INFO [optim.py:369] (1/2) 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 06:51:23,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=73686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:51:59,042 INFO [train.py:968] (1/2) Epoch 2, batch 28050, libri_loss[loss=0.4366, simple_loss=0.4518, pruned_loss=0.2107, over 29615.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.4171, pruned_loss=0.1651, over 5660541.58 frames. ], libri_tot_loss[loss=0.395, simple_loss=0.4306, pruned_loss=0.1797, over 5697443.18 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4157, pruned_loss=0.1636, over 5649138.82 frames. ], batch size: 74, lr: 1.28e-02, grad_scale: 8.0
+2023-03-01 06:52:11,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73732.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 06:52:49,893 INFO [train.py:968] (1/2) Epoch 2, batch 28100, giga_loss[loss=0.3874, simple_loss=0.4232, pruned_loss=0.1758, over 28609.00 frames. ], tot_loss[loss=0.376, simple_loss=0.4189, pruned_loss=0.1666, over 5657041.88 frames. ], libri_tot_loss[loss=0.3956, simple_loss=0.4309, pruned_loss=0.1801, over 5697953.32 frames. ], giga_tot_loss[loss=0.3734, simple_loss=0.4173, pruned_loss=0.1648, over 5646776.75 frames. ], batch size: 336, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:53:01,300 INFO [optim.py:369] (1/2) 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 06:53:26,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:53:29,166 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:53:39,963 INFO [train.py:968] (1/2) Epoch 2, batch 28150, giga_loss[loss=0.3708, simple_loss=0.4172, pruned_loss=0.1623, over 28304.00 frames. ], tot_loss[loss=0.3752, simple_loss=0.4185, pruned_loss=0.166, over 5659716.83 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4314, pruned_loss=0.1803, over 5699908.88 frames. ], giga_tot_loss[loss=0.3726, simple_loss=0.4167, pruned_loss=0.1643, over 5649650.33 frames. ], batch size: 368, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:53:54,989 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:54:25,993 INFO [train.py:968] (1/2) Epoch 2, batch 28200, giga_loss[loss=0.3398, simple_loss=0.3985, pruned_loss=0.1406, over 29097.00 frames. ], tot_loss[loss=0.377, simple_loss=0.4196, pruned_loss=0.1672, over 5663470.17 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4317, pruned_loss=0.1804, over 5708187.28 frames. ], giga_tot_loss[loss=0.3738, simple_loss=0.4173, pruned_loss=0.1652, over 5646084.67 frames. ], batch size: 155, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:54:30,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73875.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 06:54:32,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-01 06:54:32,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73878.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:54:34,251 INFO [optim.py:369] (1/2) 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 06:55:01,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73907.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 06:55:02,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-01 06:55:12,044 INFO [train.py:968] (1/2) Epoch 2, batch 28250, giga_loss[loss=0.3628, simple_loss=0.4147, pruned_loss=0.1555, over 28977.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4226, pruned_loss=0.1694, over 5674956.93 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4318, pruned_loss=0.1805, over 5712902.16 frames. ], giga_tot_loss[loss=0.3777, simple_loss=0.4205, pruned_loss=0.1674, over 5656105.87 frames. ], batch size: 106, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:55:59,193 INFO [train.py:968] (1/2) Epoch 2, batch 28300, giga_loss[loss=0.3659, simple_loss=0.4249, pruned_loss=0.1534, over 28889.00 frames. ], tot_loss[loss=0.3827, simple_loss=0.4247, pruned_loss=0.1704, over 5675095.55 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4317, pruned_loss=0.1804, over 5715064.01 frames. ], giga_tot_loss[loss=0.3801, simple_loss=0.4229, pruned_loss=0.1687, over 5656986.26 frames. ], batch size: 174, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:56:09,538 INFO [optim.py:369] (1/2) 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 06:56:13,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=73983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:56:53,078 INFO [train.py:968] (1/2) Epoch 2, batch 28350, giga_loss[loss=0.3671, simple_loss=0.4156, pruned_loss=0.1593, over 28635.00 frames. ], tot_loss[loss=0.3838, simple_loss=0.4256, pruned_loss=0.171, over 5657683.60 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4319, pruned_loss=0.1806, over 5704281.39 frames. ], giga_tot_loss[loss=0.3814, simple_loss=0.4239, pruned_loss=0.1694, over 5653241.06 frames. ], batch size: 307, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 06:57:30,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:57:38,296 INFO [train.py:968] (1/2) Epoch 2, batch 28400, giga_loss[loss=0.3978, simple_loss=0.4305, pruned_loss=0.1825, over 28589.00 frames. ], tot_loss[loss=0.3844, simple_loss=0.4258, pruned_loss=0.1715, over 5652625.01 frames. ], libri_tot_loss[loss=0.3978, simple_loss=0.4329, pruned_loss=0.1813, over 5700334.10 frames. ], giga_tot_loss[loss=0.3808, simple_loss=0.4233, pruned_loss=0.1692, over 5650353.73 frames. ], batch size: 307, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 06:57:50,360 INFO [optim.py:369] (1/2) 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 06:58:29,652 INFO [train.py:968] (1/2) Epoch 2, batch 28450, giga_loss[loss=0.4114, simple_loss=0.4468, pruned_loss=0.188, over 29065.00 frames. ], tot_loss[loss=0.388, simple_loss=0.4288, pruned_loss=0.1736, over 5648212.74 frames. ], libri_tot_loss[loss=0.3979, simple_loss=0.4327, pruned_loss=0.1815, over 5695525.59 frames. ], giga_tot_loss[loss=0.3848, simple_loss=0.4268, pruned_loss=0.1713, over 5649327.81 frames. ], batch size: 113, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 06:58:39,764 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:59:22,268 INFO [train.py:968] (1/2) Epoch 2, batch 28500, giga_loss[loss=0.4152, simple_loss=0.4464, pruned_loss=0.192, over 29066.00 frames. ], tot_loss[loss=0.389, simple_loss=0.4301, pruned_loss=0.1739, over 5640228.72 frames. ], libri_tot_loss[loss=0.3986, simple_loss=0.4331, pruned_loss=0.182, over 5679713.35 frames. ], giga_tot_loss[loss=0.3854, simple_loss=0.4281, pruned_loss=0.1714, over 5654660.29 frames. ], batch size: 113, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 06:59:31,196 INFO [optim.py:369] (1/2) 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 06:59:57,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 06:59:59,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:00:11,324 INFO [train.py:968] (1/2) Epoch 2, batch 28550, giga_loss[loss=0.3707, simple_loss=0.4157, pruned_loss=0.1628, over 28785.00 frames. ], tot_loss[loss=0.3863, simple_loss=0.4273, pruned_loss=0.1726, over 5655617.09 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4317, pruned_loss=0.1812, over 5685780.21 frames. ], giga_tot_loss[loss=0.3846, simple_loss=0.4269, pruned_loss=0.1711, over 5660700.40 frames. ], batch size: 284, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:00:30,504 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:00:34,725 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-01 07:01:05,385 INFO [train.py:968] (1/2) Epoch 2, batch 28600, giga_loss[loss=0.4449, simple_loss=0.466, pruned_loss=0.2119, over 28273.00 frames. ], tot_loss[loss=0.3853, simple_loss=0.4262, pruned_loss=0.1722, over 5662344.17 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4315, pruned_loss=0.1808, over 5689499.76 frames. ], giga_tot_loss[loss=0.3841, simple_loss=0.4259, pruned_loss=0.1711, over 5662595.43 frames. ], batch size: 368, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:01:23,814 INFO [optim.py:369] (1/2) 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 07:01:44,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 07:01:47,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0837, 1.1565, 1.0880, 0.8837], device='cuda:1'), covar=tensor([0.1750, 0.1843, 0.1550, 0.1618], device='cuda:1'), in_proj_covar=tensor([0.0983, 0.0823, 0.0892, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 07:01:50,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2103, 1.3482, 1.1829, 0.6802], device='cuda:1'), covar=tensor([0.0566, 0.0472, 0.0371, 0.0539], device='cuda:1'), in_proj_covar=tensor([0.1045, 0.0775, 0.0880, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 07:02:05,301 INFO [train.py:968] (1/2) Epoch 2, batch 28650, giga_loss[loss=0.3377, simple_loss=0.3865, pruned_loss=0.1445, over 28293.00 frames. ], tot_loss[loss=0.3845, simple_loss=0.4253, pruned_loss=0.1718, over 5661352.49 frames. ], libri_tot_loss[loss=0.397, simple_loss=0.4319, pruned_loss=0.1811, over 5682894.99 frames. ], giga_tot_loss[loss=0.3828, simple_loss=0.4245, pruned_loss=0.1705, over 5667156.19 frames. ], batch size: 71, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:02:09,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4176, 2.1547, 1.4134, 1.2756], device='cuda:1'), covar=tensor([0.0933, 0.0578, 0.0859, 0.1497], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0465, 0.0333, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 07:02:41,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3624, 1.7346, 1.5331, 1.6188], device='cuda:1'), covar=tensor([0.1228, 0.1726, 0.1126, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0835, 0.0727, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 07:02:48,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:02:57,486 INFO [train.py:968] (1/2) Epoch 2, batch 28700, giga_loss[loss=0.3875, simple_loss=0.4364, pruned_loss=0.1694, over 28860.00 frames. ], tot_loss[loss=0.3812, simple_loss=0.4229, pruned_loss=0.1698, over 5668580.13 frames. ], libri_tot_loss[loss=0.3968, simple_loss=0.4319, pruned_loss=0.1809, over 5686877.96 frames. ], giga_tot_loss[loss=0.3798, simple_loss=0.4222, pruned_loss=0.1687, over 5669409.54 frames. ], batch size: 174, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:03:07,182 INFO [optim.py:369] (1/2) 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 07:03:48,448 INFO [train.py:968] (1/2) Epoch 2, batch 28750, giga_loss[loss=0.3482, simple_loss=0.4027, pruned_loss=0.1468, over 28890.00 frames. ], tot_loss[loss=0.3809, simple_loss=0.4222, pruned_loss=0.1698, over 5657311.11 frames. ], libri_tot_loss[loss=0.3971, simple_loss=0.4322, pruned_loss=0.1809, over 5690096.03 frames. ], giga_tot_loss[loss=0.3794, simple_loss=0.4213, pruned_loss=0.1688, over 5655123.67 frames. ], batch size: 112, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:04:27,072 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:04:39,181 INFO [train.py:968] (1/2) Epoch 2, batch 28800, giga_loss[loss=0.4314, simple_loss=0.4551, pruned_loss=0.2038, over 28325.00 frames. ], tot_loss[loss=0.3804, simple_loss=0.4218, pruned_loss=0.1695, over 5665145.31 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4318, pruned_loss=0.1806, over 5694057.65 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.4213, pruned_loss=0.1687, over 5659226.34 frames. ], batch size: 368, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 07:04:50,681 INFO [optim.py:369] (1/2) 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 07:05:11,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:05:11,860 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:05:14,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:05:29,644 INFO [train.py:968] (1/2) Epoch 2, batch 28850, giga_loss[loss=0.4348, simple_loss=0.4552, pruned_loss=0.2072, over 27551.00 frames. ], tot_loss[loss=0.3836, simple_loss=0.4236, pruned_loss=0.1718, over 5657450.79 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4316, pruned_loss=0.1806, over 5693610.29 frames. ], giga_tot_loss[loss=0.3825, simple_loss=0.4231, pruned_loss=0.1709, over 5652416.59 frames. ], batch size: 472, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 07:05:41,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:05:46,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1717, 1.2520, 0.9099, 0.9433], device='cuda:1'), covar=tensor([0.0430, 0.0406, 0.0389, 0.0448], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0791, 0.0886, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 07:06:17,974 INFO [train.py:968] (1/2) Epoch 2, batch 28900, giga_loss[loss=0.3851, simple_loss=0.4332, pruned_loss=0.1685, over 28898.00 frames. ], tot_loss[loss=0.387, simple_loss=0.4258, pruned_loss=0.1741, over 5664281.17 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4312, pruned_loss=0.1804, over 5701010.82 frames. ], giga_tot_loss[loss=0.3861, simple_loss=0.4255, pruned_loss=0.1734, over 5652288.89 frames. ], batch size: 199, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:06:28,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7055, 1.4804, 3.7306, 2.9850], device='cuda:1'), covar=tensor([0.1562, 0.1614, 0.0346, 0.0532], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0504, 0.0667, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 07:06:35,207 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 28950, giga_loss[loss=0.3923, simple_loss=0.4273, pruned_loss=0.1787, over 28655.00 frames. ], tot_loss[loss=0.3872, simple_loss=0.4255, pruned_loss=0.1745, over 5648716.00 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4314, pruned_loss=0.1806, over 5700524.09 frames. ], giga_tot_loss[loss=0.386, simple_loss=0.4249, pruned_loss=0.1736, over 5639061.67 frames. ], batch size: 262, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:07:18,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8453, 1.2491, 4.7513, 3.5378], device='cuda:1'), covar=tensor([0.1535, 0.1864, 0.0271, 0.0420], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0503, 0.0664, 0.0534], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 07:07:29,128 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:07:31,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3741, 1.7836, 1.4828, 0.5825], device='cuda:1'), covar=tensor([0.0983, 0.0740, 0.1155, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.1193, 0.1171, 0.1188, 0.1036], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 07:07:37,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:07:40,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:07:42,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8126, 2.4286, 2.0331, 2.1101], device='cuda:1'), covar=tensor([0.0519, 0.0617, 0.0756, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0517, 0.0543, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 07:07:49,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2007, 1.1742, 0.9876, 0.9684], device='cuda:1'), covar=tensor([0.0474, 0.0347, 0.0701, 0.0711], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0519, 0.0546, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 07:08:02,488 INFO [train.py:968] (1/2) Epoch 2, batch 29000, giga_loss[loss=0.3589, simple_loss=0.4114, pruned_loss=0.1532, over 28969.00 frames. ], tot_loss[loss=0.389, simple_loss=0.4262, pruned_loss=0.1758, over 5652993.78 frames. ], libri_tot_loss[loss=0.3957, simple_loss=0.4308, pruned_loss=0.1803, over 5701494.62 frames. ], giga_tot_loss[loss=0.3884, simple_loss=0.4262, pruned_loss=0.1753, over 5643446.73 frames. ], batch size: 145, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:08:06,531 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:08:12,307 INFO [optim.py:369] (1/2) 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 07:08:35,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 07:08:47,267 INFO [train.py:968] (1/2) Epoch 2, batch 29050, giga_loss[loss=0.4151, simple_loss=0.4428, pruned_loss=0.1937, over 28570.00 frames. ], tot_loss[loss=0.3902, simple_loss=0.427, pruned_loss=0.1767, over 5640263.80 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4314, pruned_loss=0.1807, over 5692676.29 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4263, pruned_loss=0.1757, over 5639390.09 frames. ], batch size: 307, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:08:55,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6567, 4.0599, 4.4257, 1.7422], device='cuda:1'), covar=tensor([0.0454, 0.0376, 0.0788, 0.1915], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0579, 0.0838, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 07:09:38,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-01 07:09:40,864 INFO [train.py:968] (1/2) Epoch 2, batch 29100, giga_loss[loss=0.3626, simple_loss=0.4279, pruned_loss=0.1487, over 28907.00 frames. ], tot_loss[loss=0.3887, simple_loss=0.4268, pruned_loss=0.1753, over 5640428.65 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4313, pruned_loss=0.1806, over 5693872.90 frames. ], giga_tot_loss[loss=0.3878, simple_loss=0.4263, pruned_loss=0.1747, over 5638499.61 frames. ], batch size: 145, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:09:53,946 INFO [optim.py:369] (1/2) 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 07:10:03,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3501, 1.4194, 1.1156, 1.3711], device='cuda:1'), covar=tensor([0.1025, 0.0426, 0.0477, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0193, 0.0194, 0.0342], device='cuda:1'), out_proj_covar=tensor([0.0033, 0.0025, 0.0022, 0.0037], device='cuda:1')
+2023-03-01 07:10:28,272 INFO [train.py:968] (1/2) Epoch 2, batch 29150, giga_loss[loss=0.4034, simple_loss=0.4393, pruned_loss=0.1838, over 29158.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4271, pruned_loss=0.1747, over 5654698.92 frames. ], libri_tot_loss[loss=0.3957, simple_loss=0.4309, pruned_loss=0.1802, over 5698992.18 frames. ], giga_tot_loss[loss=0.3879, simple_loss=0.4269, pruned_loss=0.1744, over 5647573.82 frames. ], batch size: 113, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:10:41,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:11:05,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3100, 2.9306, 3.0962, 1.9179], device='cuda:1'), covar=tensor([0.0646, 0.0604, 0.1056, 0.1412], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0582, 0.0851, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 07:11:15,241 INFO [train.py:968] (1/2) Epoch 2, batch 29200, giga_loss[loss=0.3679, simple_loss=0.4183, pruned_loss=0.1587, over 28481.00 frames. ], tot_loss[loss=0.3888, simple_loss=0.4276, pruned_loss=0.1751, over 5652051.16 frames. ], libri_tot_loss[loss=0.3956, simple_loss=0.4309, pruned_loss=0.1801, over 5688802.69 frames. ], giga_tot_loss[loss=0.3883, simple_loss=0.4273, pruned_loss=0.1746, over 5654104.01 frames. ], batch size: 60, lr: 1.27e-02, grad_scale: 8.0
+2023-03-01 07:11:25,102 INFO [optim.py:369] (1/2) 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 07:11:25,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6251, 1.4126, 1.4119, 1.4934], device='cuda:1'), covar=tensor([0.0761, 0.1260, 0.1142, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0805, 0.0634, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 07:11:26,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.21 vs. limit=5.0
+2023-03-01 07:11:27,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4379, 1.8656, 5.5052, 4.0700], device='cuda:1'), covar=tensor([0.1415, 0.1510, 0.0248, 0.0307], device='cuda:1'), in_proj_covar=tensor([0.0520, 0.0504, 0.0669, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 07:11:59,682 INFO [train.py:968] (1/2) Epoch 2, batch 29250, giga_loss[loss=0.4251, simple_loss=0.4498, pruned_loss=0.2002, over 28617.00 frames. ], tot_loss[loss=0.3917, simple_loss=0.4295, pruned_loss=0.177, over 5647867.62 frames. ], libri_tot_loss[loss=0.3953, simple_loss=0.4306, pruned_loss=0.1799, over 5673618.67 frames. ], giga_tot_loss[loss=0.3915, simple_loss=0.4295, pruned_loss=0.1767, over 5662027.82 frames. ], batch size: 336, lr: 1.27e-02, grad_scale: 4.0
+2023-03-01 07:12:23,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1883, 1.2448, 1.2771, 1.2391], device='cuda:1'), covar=tensor([0.0763, 0.0956, 0.1183, 0.0901], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0814, 0.0640, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 07:12:48,300 INFO [train.py:968] (1/2) Epoch 2, batch 29300, giga_loss[loss=0.3435, simple_loss=0.4024, pruned_loss=0.1423, over 28853.00 frames. ], tot_loss[loss=0.3899, simple_loss=0.4284, pruned_loss=0.1757, over 5656426.96 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4302, pruned_loss=0.1796, over 5677198.37 frames. ], giga_tot_loss[loss=0.3901, simple_loss=0.4287, pruned_loss=0.1757, over 5664103.74 frames. ], batch size: 186, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:12:53,844 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:12:57,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:13:01,766 INFO [optim.py:369] (1/2) 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 07:13:08,085 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=74988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:13:17,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5346, 1.4750, 1.1528, 1.3140], device='cuda:1'), covar=tensor([0.0751, 0.0735, 0.1083, 0.0921], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0524, 0.0554, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 07:13:27,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:13:28,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:13:38,795 INFO [train.py:968] (1/2) Epoch 2, batch 29350, libri_loss[loss=0.356, simple_loss=0.3949, pruned_loss=0.1585, over 29663.00 frames. ], tot_loss[loss=0.3904, simple_loss=0.4296, pruned_loss=0.1756, over 5653155.37 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.4302, pruned_loss=0.1795, over 5674057.80 frames. ], giga_tot_loss[loss=0.3904, simple_loss=0.4299, pruned_loss=0.1755, over 5661329.23 frames. ], batch size: 73, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:13:41,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5949, 2.4812, 1.7468, 1.6600], device='cuda:1'), covar=tensor([0.0955, 0.0283, 0.0408, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0287, 0.0194, 0.0197, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0033, 0.0025, 0.0023, 0.0038], device='cuda:1')
+2023-03-01 07:13:44,330 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-01 07:14:28,845 INFO [train.py:968] (1/2) Epoch 2, batch 29400, giga_loss[loss=0.3497, simple_loss=0.4098, pruned_loss=0.1448, over 28663.00 frames. ], tot_loss[loss=0.3884, simple_loss=0.4287, pruned_loss=0.174, over 5653990.08 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4295, pruned_loss=0.1792, over 5681387.02 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4296, pruned_loss=0.1742, over 5653078.12 frames. ], batch size: 60, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:14:40,178 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 29450, giga_loss[loss=0.4063, simple_loss=0.4146, pruned_loss=0.199, over 23885.00 frames. ], tot_loss[loss=0.3864, simple_loss=0.4272, pruned_loss=0.1728, over 5658794.75 frames. ], libri_tot_loss[loss=0.3936, simple_loss=0.4292, pruned_loss=0.179, over 5677270.44 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4281, pruned_loss=0.1729, over 5661305.44 frames. ], batch size: 705, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:15:46,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:15:48,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:16:01,781 INFO [train.py:968] (1/2) Epoch 2, batch 29500, giga_loss[loss=0.365, simple_loss=0.4197, pruned_loss=0.1551, over 28878.00 frames. ], tot_loss[loss=0.3865, simple_loss=0.4267, pruned_loss=0.1732, over 5655061.83 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4286, pruned_loss=0.1786, over 5683065.77 frames. ], giga_tot_loss[loss=0.3875, simple_loss=0.4279, pruned_loss=0.1735, over 5651140.95 frames. ], batch size: 186, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:16:14,476 INFO [optim.py:369] (1/2) 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 07:16:14,916 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:16:14,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3418, 1.8241, 1.2932, 1.3576], device='cuda:1'), covar=tensor([0.0942, 0.0385, 0.0459, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0192, 0.0193, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0033, 0.0025, 0.0022, 0.0037], device='cuda:1')
+2023-03-01 07:16:49,548 INFO [train.py:968] (1/2) Epoch 2, batch 29550, giga_loss[loss=0.4394, simple_loss=0.4546, pruned_loss=0.2121, over 27580.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4281, pruned_loss=0.1738, over 5669223.62 frames. ], libri_tot_loss[loss=0.3924, simple_loss=0.4284, pruned_loss=0.1782, over 5688640.68 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4292, pruned_loss=0.1743, over 5660284.45 frames. ], batch size: 472, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:16:50,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9644, 2.3432, 1.9575, 1.9619], device='cuda:1'), covar=tensor([0.0935, 0.1247, 0.0929, 0.0627], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0828, 0.0719, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 07:17:28,688 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-01 07:17:42,659 INFO [train.py:968] (1/2) Epoch 2, batch 29600, giga_loss[loss=0.425, simple_loss=0.4475, pruned_loss=0.2013, over 27613.00 frames. ], tot_loss[loss=0.3893, simple_loss=0.4287, pruned_loss=0.175, over 5659475.53 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4281, pruned_loss=0.1781, over 5688057.89 frames. ], giga_tot_loss[loss=0.3904, simple_loss=0.4299, pruned_loss=0.1755, over 5652412.58 frames. ], batch size: 472, lr: 1.26e-02, grad_scale: 8.0
+2023-03-01 07:17:58,419 INFO [optim.py:369] (1/2) 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 07:18:17,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:18:26,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2915, 1.7171, 1.2904, 1.4428], device='cuda:1'), covar=tensor([0.0883, 0.0486, 0.0461, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0194, 0.0194, 0.0342], device='cuda:1'), out_proj_covar=tensor([0.0033, 0.0025, 0.0023, 0.0038], device='cuda:1')
+2023-03-01 07:18:34,694 INFO [train.py:968] (1/2) Epoch 2, batch 29650, giga_loss[loss=0.3312, simple_loss=0.3928, pruned_loss=0.1348, over 28960.00 frames. ], tot_loss[loss=0.3883, simple_loss=0.4273, pruned_loss=0.1747, over 5664051.70 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4282, pruned_loss=0.178, over 5690370.13 frames. ], giga_tot_loss[loss=0.3891, simple_loss=0.4281, pruned_loss=0.175, over 5656196.01 frames. ], batch size: 174, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:18:45,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0823, 3.5427, 3.8062, 1.7814], device='cuda:1'), covar=tensor([0.0445, 0.0422, 0.0727, 0.1764], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0580, 0.0840, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 07:19:07,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:19:13,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:19:18,388 INFO [train.py:968] (1/2) Epoch 2, batch 29700, giga_loss[loss=0.51, simple_loss=0.5019, pruned_loss=0.2591, over 28268.00 frames. ], tot_loss[loss=0.3896, simple_loss=0.4278, pruned_loss=0.1757, over 5637433.20 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4287, pruned_loss=0.1784, over 5674070.51 frames. ], giga_tot_loss[loss=0.3895, simple_loss=0.428, pruned_loss=0.1755, over 5644938.37 frames. ], batch size: 368, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:19:34,098 INFO [optim.py:369] (1/2) 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 07:19:59,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4119, 1.8339, 1.3868, 0.4388], device='cuda:1'), covar=tensor([0.1019, 0.0861, 0.1399, 0.1411], device='cuda:1'), in_proj_covar=tensor([0.1204, 0.1180, 0.1202, 0.1044], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 07:20:09,072 INFO [train.py:968] (1/2) Epoch 2, batch 29750, libri_loss[loss=0.3553, simple_loss=0.3902, pruned_loss=0.1602, over 29631.00 frames. ], tot_loss[loss=0.3908, simple_loss=0.4288, pruned_loss=0.1764, over 5649360.28 frames. ], libri_tot_loss[loss=0.3926, simple_loss=0.4286, pruned_loss=0.1783, over 5673664.69 frames. ], giga_tot_loss[loss=0.3908, simple_loss=0.4291, pruned_loss=0.1763, over 5655134.44 frames. ], batch size: 69, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:20:09,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-01 07:20:34,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-01 07:20:47,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4718, 1.3292, 1.2605, 1.4917], device='cuda:1'), covar=tensor([0.1593, 0.1514, 0.1260, 0.1785], device='cuda:1'), in_proj_covar=tensor([0.0984, 0.0816, 0.0895, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 07:20:59,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-01 07:21:00,430 INFO [train.py:968] (1/2) Epoch 2, batch 29800, giga_loss[loss=0.3545, simple_loss=0.4095, pruned_loss=0.1498, over 28925.00 frames. ], tot_loss[loss=0.3913, simple_loss=0.4292, pruned_loss=0.1767, over 5643243.19 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.429, pruned_loss=0.1784, over 5679199.20 frames. ], giga_tot_loss[loss=0.391, simple_loss=0.4291, pruned_loss=0.1764, over 5642430.42 frames. ], batch size: 227, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:21:13,958 INFO [optim.py:369] (1/2) 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 07:21:35,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:21:39,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:21:47,152 INFO [train.py:968] (1/2) Epoch 2, batch 29850, giga_loss[loss=0.3541, simple_loss=0.4018, pruned_loss=0.1532, over 28799.00 frames. ], tot_loss[loss=0.3871, simple_loss=0.4269, pruned_loss=0.1736, over 5667852.51 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.4293, pruned_loss=0.1787, over 5683412.67 frames. ], giga_tot_loss[loss=0.3864, simple_loss=0.4265, pruned_loss=0.1731, over 5662925.86 frames. ], batch size: 99, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:22:07,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:22:34,321 INFO [train.py:968] (1/2) Epoch 2, batch 29900, giga_loss[loss=0.3481, simple_loss=0.4027, pruned_loss=0.1467, over 28921.00 frames. ], tot_loss[loss=0.3879, simple_loss=0.4279, pruned_loss=0.174, over 5666128.82 frames. ], libri_tot_loss[loss=0.3936, simple_loss=0.4297, pruned_loss=0.1788, over 5686182.98 frames. ], giga_tot_loss[loss=0.3869, simple_loss=0.4271, pruned_loss=0.1733, over 5659264.37 frames. ], batch size: 164, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:22:51,000 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 29950, giga_loss[loss=0.3893, simple_loss=0.4357, pruned_loss=0.1715, over 28915.00 frames. ], tot_loss[loss=0.3865, simple_loss=0.4269, pruned_loss=0.1731, over 5673072.24 frames. ], libri_tot_loss[loss=0.3934, simple_loss=0.4292, pruned_loss=0.1787, over 5692958.82 frames. ], giga_tot_loss[loss=0.3857, simple_loss=0.4266, pruned_loss=0.1724, over 5660631.49 frames. ], batch size: 136, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:23:24,414 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:23:56,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3693, 1.8686, 1.3650, 0.6275], device='cuda:1'), covar=tensor([0.1452, 0.0854, 0.1147, 0.1743], device='cuda:1'), in_proj_covar=tensor([0.1212, 0.1173, 0.1211, 0.1033], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 07:24:07,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=75663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:24:13,756 INFO [train.py:968] (1/2) Epoch 2, batch 30000, giga_loss[loss=0.3394, simple_loss=0.3934, pruned_loss=0.1426, over 29085.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4254, pruned_loss=0.1722, over 5660910.43 frames. ], libri_tot_loss[loss=0.3939, simple_loss=0.4296, pruned_loss=0.179, over 5683513.79 frames. ], giga_tot_loss[loss=0.3836, simple_loss=0.4247, pruned_loss=0.1713, over 5659691.05 frames. ], batch size: 155, lr: 1.26e-02, grad_scale: 8.0
+2023-03-01 07:24:13,757 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 07:24:21,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8599, 1.4407, 3.1969, 2.9847], device='cuda:1'), covar=tensor([0.1523, 0.1723, 0.0494, 0.0521], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0501, 0.0652, 0.0533], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 07:24:22,530 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 07:24:31,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:24:36,516 INFO [optim.py:369] (1/2) 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 07:25:07,937 INFO [train.py:968] (1/2) Epoch 2, batch 30050, giga_loss[loss=0.3353, simple_loss=0.3869, pruned_loss=0.1418, over 28950.00 frames. ], tot_loss[loss=0.3851, simple_loss=0.425, pruned_loss=0.1726, over 5661639.73 frames. ], libri_tot_loss[loss=0.3931, simple_loss=0.4291, pruned_loss=0.1785, over 5689390.85 frames. ], giga_tot_loss[loss=0.3845, simple_loss=0.4248, pruned_loss=0.1721, over 5654552.09 frames. ], batch size: 112, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:25:12,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3652, 3.8333, 3.9684, 1.9385], device='cuda:1'), covar=tensor([0.0521, 0.0517, 0.1189, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0586, 0.0848, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 07:25:17,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:25:39,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3082, 1.2522, 1.1228, 1.2976], device='cuda:1'), covar=tensor([0.1848, 0.1972, 0.1689, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.0983, 0.0826, 0.0900, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 07:25:51,369 INFO [train.py:968] (1/2) Epoch 2, batch 30100, giga_loss[loss=0.3357, simple_loss=0.3821, pruned_loss=0.1447, over 28928.00 frames. ], tot_loss[loss=0.383, simple_loss=0.4232, pruned_loss=0.1715, over 5670167.06 frames. ], libri_tot_loss[loss=0.3922, simple_loss=0.4285, pruned_loss=0.178, over 5694823.53 frames. ], giga_tot_loss[loss=0.383, simple_loss=0.4233, pruned_loss=0.1713, over 5658435.08 frames. ], batch size: 66, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:26:09,046 INFO [optim.py:369] (1/2) 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 07:26:42,842 INFO [train.py:968] (1/2) Epoch 2, batch 30150, giga_loss[loss=0.3892, simple_loss=0.422, pruned_loss=0.1781, over 27878.00 frames. ], tot_loss[loss=0.3794, simple_loss=0.4197, pruned_loss=0.1696, over 5660768.91 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4288, pruned_loss=0.1783, over 5678762.50 frames. ], giga_tot_loss[loss=0.3787, simple_loss=0.4194, pruned_loss=0.169, over 5666195.69 frames. ], batch size: 412, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:26:43,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:26:47,365 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:14,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:33,098 INFO [train.py:968] (1/2) Epoch 2, batch 30200, giga_loss[loss=0.3975, simple_loss=0.4235, pruned_loss=0.1858, over 28812.00 frames. ], tot_loss[loss=0.3773, simple_loss=0.4176, pruned_loss=0.1685, over 5678480.61 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4289, pruned_loss=0.1784, over 5681668.81 frames. ], giga_tot_loss[loss=0.3765, simple_loss=0.4172, pruned_loss=0.1679, over 5680186.14 frames. ], batch size: 112, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:27:38,390 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:41,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:27:47,670 INFO [optim.py:369] (1/2) 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 07:28:03,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9965, 1.4973, 4.9161, 3.4684], device='cuda:1'), covar=tensor([0.2284, 0.2132, 0.0477, 0.0575], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0504, 0.0673, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 07:28:12,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-01 07:28:12,936 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:28:25,119 INFO [train.py:968] (1/2) Epoch 2, batch 30250, giga_loss[loss=0.4245, simple_loss=0.4532, pruned_loss=0.1979, over 27598.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.4172, pruned_loss=0.1681, over 5681334.55 frames. ], libri_tot_loss[loss=0.3927, simple_loss=0.4289, pruned_loss=0.1783, over 5681550.83 frames. ], giga_tot_loss[loss=0.376, simple_loss=0.4168, pruned_loss=0.1676, over 5682575.67 frames. ], batch size: 472, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:29:10,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-01 07:29:15,374 INFO [train.py:968] (1/2) Epoch 2, batch 30300, giga_loss[loss=0.3746, simple_loss=0.4199, pruned_loss=0.1647, over 28481.00 frames. ], tot_loss[loss=0.3747, simple_loss=0.4167, pruned_loss=0.1664, over 5682534.34 frames. ], libri_tot_loss[loss=0.3932, simple_loss=0.4291, pruned_loss=0.1786, over 5685656.92 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.416, pruned_loss=0.1655, over 5679722.64 frames. ], batch size: 78, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:29:30,867 INFO [optim.py:369] (1/2) 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 07:29:45,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:30:12,145 INFO [train.py:968] (1/2) Epoch 2, batch 30350, giga_loss[loss=0.3009, simple_loss=0.343, pruned_loss=0.1293, over 24350.00 frames. ], tot_loss[loss=0.367, simple_loss=0.4121, pruned_loss=0.161, over 5662644.78 frames. ], libri_tot_loss[loss=0.3933, simple_loss=0.4291, pruned_loss=0.1787, over 5680860.52 frames. ], giga_tot_loss[loss=0.3655, simple_loss=0.4113, pruned_loss=0.1599, over 5665422.36 frames. ], batch size: 705, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:30:32,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:30:36,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5029, 1.5160, 1.1008, 1.3617], device='cuda:1'), covar=tensor([0.0640, 0.0571, 0.0995, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0507, 0.0540, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 07:31:03,442 INFO [train.py:968] (1/2) Epoch 2, batch 30400, giga_loss[loss=0.302, simple_loss=0.3713, pruned_loss=0.1164, over 28508.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.4057, pruned_loss=0.1544, over 5659153.50 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4286, pruned_loss=0.1784, over 5681645.12 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4053, pruned_loss=0.1536, over 5660560.06 frames. ], batch size: 71, lr: 1.26e-02, grad_scale: 8.0
+2023-03-01 07:31:18,040 INFO [optim.py:369] (1/2) 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 07:31:51,885 INFO [train.py:968] (1/2) Epoch 2, batch 30450, libri_loss[loss=0.3659, simple_loss=0.4105, pruned_loss=0.1607, over 29528.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4016, pruned_loss=0.1507, over 5648974.59 frames. ], libri_tot_loss[loss=0.3917, simple_loss=0.4275, pruned_loss=0.1779, over 5673649.37 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4012, pruned_loss=0.1495, over 5654970.71 frames. ], batch size: 80, lr: 1.26e-02, grad_scale: 4.0
+2023-03-01 07:32:11,813 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:14,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:28,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7511, 2.4474, 1.7944, 0.8513], device='cuda:1'), covar=tensor([0.1281, 0.0750, 0.1276, 0.1438], device='cuda:1'), in_proj_covar=tensor([0.1191, 0.1160, 0.1198, 0.1026], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 07:32:42,409 INFO [train.py:968] (1/2) Epoch 2, batch 30500, giga_loss[loss=0.3074, simple_loss=0.3803, pruned_loss=0.1172, over 28928.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3961, pruned_loss=0.1454, over 5655358.77 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4264, pruned_loss=0.1774, over 5678391.43 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3963, pruned_loss=0.1443, over 5655589.54 frames. ], batch size: 145, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:32:43,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4670, 1.9326, 1.3320, 0.6623], device='cuda:1'), covar=tensor([0.1167, 0.0785, 0.1324, 0.1559], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.1157, 0.1197, 0.1023], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 07:32:44,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:46,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9113, 1.5083, 4.1222, 3.0622], device='cuda:1'), covar=tensor([0.2027, 0.2158, 0.0564, 0.0619], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0497, 0.0665, 0.0523], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 07:32:54,402 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:57,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:32:58,494 INFO [optim.py:369] (1/2) 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 07:33:30,280 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:33:36,238 INFO [train.py:968] (1/2) Epoch 2, batch 30550, giga_loss[loss=0.3065, simple_loss=0.3765, pruned_loss=0.1183, over 28860.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3932, pruned_loss=0.1411, over 5643926.43 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4259, pruned_loss=0.1771, over 5679264.68 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3935, pruned_loss=0.1402, over 5643140.59 frames. ], batch size: 99, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:34:13,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-01 07:34:32,362 INFO [train.py:968] (1/2) Epoch 2, batch 30600, giga_loss[loss=0.3424, simple_loss=0.3923, pruned_loss=0.1462, over 27963.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3941, pruned_loss=0.1419, over 5633231.51 frames. ], libri_tot_loss[loss=0.3903, simple_loss=0.426, pruned_loss=0.1773, over 5675366.05 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3936, pruned_loss=0.1403, over 5635257.38 frames. ], batch size: 412, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:34:50,944 INFO [optim.py:369] (1/2) 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 07:34:52,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3379, 1.7897, 1.3348, 0.6763], device='cuda:1'), covar=tensor([0.1189, 0.0750, 0.1121, 0.1410], device='cuda:1'), in_proj_covar=tensor([0.1206, 0.1176, 0.1201, 0.1034], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 07:34:57,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6584, 4.0474, 4.2667, 1.8605], device='cuda:1'), covar=tensor([0.0486, 0.0498, 0.1116, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0567, 0.0816, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 07:35:21,962 INFO [train.py:968] (1/2) Epoch 2, batch 30650, giga_loss[loss=0.3246, simple_loss=0.3841, pruned_loss=0.1326, over 28739.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3908, pruned_loss=0.1395, over 5637942.73 frames. ], libri_tot_loss[loss=0.3886, simple_loss=0.4245, pruned_loss=0.1763, over 5681738.83 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3908, pruned_loss=0.138, over 5632845.15 frames. ], batch size: 92, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:36:10,454 INFO [train.py:968] (1/2) Epoch 2, batch 30700, giga_loss[loss=0.355, simple_loss=0.3967, pruned_loss=0.1566, over 27523.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3868, pruned_loss=0.1366, over 5646355.24 frames. ], libri_tot_loss[loss=0.3872, simple_loss=0.4232, pruned_loss=0.1756, over 5689046.73 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3868, pruned_loss=0.1348, over 5634308.94 frames. ], batch size: 472, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:36:27,822 INFO [optim.py:369] (1/2) 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 07:37:00,724 INFO [train.py:968] (1/2) Epoch 2, batch 30750, giga_loss[loss=0.3198, simple_loss=0.3897, pruned_loss=0.1249, over 28978.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3854, pruned_loss=0.1358, over 5649855.85 frames. ], libri_tot_loss[loss=0.3861, simple_loss=0.4223, pruned_loss=0.1749, over 5694424.37 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3853, pruned_loss=0.1339, over 5634343.27 frames. ], batch size: 128, lr: 1.25e-02, grad_scale: 2.0
+2023-03-01 07:37:49,156 INFO [train.py:968] (1/2) Epoch 2, batch 30800, giga_loss[loss=0.339, simple_loss=0.3908, pruned_loss=0.1436, over 27912.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3856, pruned_loss=0.1356, over 5653476.01 frames. ], libri_tot_loss[loss=0.3851, simple_loss=0.4214, pruned_loss=0.1744, over 5698389.90 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3856, pruned_loss=0.1338, over 5636797.06 frames. ], batch size: 412, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:38:09,754 INFO [optim.py:369] (1/2) 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 07:38:28,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5534, 1.6668, 1.5476, 0.9030], device='cuda:1'), covar=tensor([0.0525, 0.0399, 0.0283, 0.0531], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.0777, 0.0885, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 07:38:30,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-01 07:38:42,102 INFO [train.py:968] (1/2) Epoch 2, batch 30850, giga_loss[loss=0.2838, simple_loss=0.3557, pruned_loss=0.1059, over 28894.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3832, pruned_loss=0.1329, over 5651472.29 frames. ], libri_tot_loss[loss=0.3853, simple_loss=0.4215, pruned_loss=0.1745, over 5690884.70 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3825, pruned_loss=0.1307, over 5644039.78 frames. ], batch size: 227, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:39:14,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9913, 1.1727, 1.1017, 0.6401], device='cuda:1'), covar=tensor([0.0410, 0.0396, 0.0330, 0.0421], device='cuda:1'), in_proj_covar=tensor([0.1072, 0.0766, 0.0873, 0.0921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 07:39:32,266 INFO [train.py:968] (1/2) Epoch 2, batch 30900, giga_loss[loss=0.3532, simple_loss=0.4041, pruned_loss=0.1511, over 27932.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3795, pruned_loss=0.1298, over 5643618.94 frames. ], libri_tot_loss[loss=0.3842, simple_loss=0.4204, pruned_loss=0.174, over 5685183.69 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3788, pruned_loss=0.1274, over 5641900.76 frames. ], batch size: 412, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:39:50,164 INFO [optim.py:369] (1/2) 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 07:40:23,777 INFO [train.py:968] (1/2) Epoch 2, batch 30950, giga_loss[loss=0.2923, simple_loss=0.3618, pruned_loss=0.1114, over 28707.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3739, pruned_loss=0.1262, over 5637717.68 frames. ], libri_tot_loss[loss=0.3831, simple_loss=0.4195, pruned_loss=0.1734, over 5685503.47 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3733, pruned_loss=0.1238, over 5635380.64 frames. ], batch size: 284, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:41:04,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=76662.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:41:09,843 INFO [train.py:968] (1/2) Epoch 2, batch 31000, giga_loss[loss=0.4032, simple_loss=0.4189, pruned_loss=0.1938, over 26698.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.374, pruned_loss=0.1275, over 5653289.91 frames. ], libri_tot_loss[loss=0.3813, simple_loss=0.4178, pruned_loss=0.1724, over 5694880.95 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1245, over 5640915.30 frames. ], batch size: 555, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:41:29,109 INFO [optim.py:369] (1/2) 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 07:41:29,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=76687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:41:47,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-01 07:42:03,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 31050, giga_loss[loss=0.3076, simple_loss=0.3752, pruned_loss=0.12, over 28523.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3723, pruned_loss=0.1268, over 5637469.96 frames. ], libri_tot_loss[loss=0.381, simple_loss=0.4177, pruned_loss=0.1722, over 5697115.15 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3713, pruned_loss=0.1242, over 5625578.23 frames. ], batch size: 336, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:42:59,998 INFO [train.py:968] (1/2) Epoch 2, batch 31100, giga_loss[loss=0.3361, simple_loss=0.3996, pruned_loss=0.1363, over 28745.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3747, pruned_loss=0.1285, over 5635244.56 frames. ], libri_tot_loss[loss=0.3805, simple_loss=0.417, pruned_loss=0.172, over 5700759.96 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3736, pruned_loss=0.1257, over 5621285.34 frames. ], batch size: 284, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:43:21,812 INFO [optim.py:369] (1/2) 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 07:43:58,931 INFO [train.py:968] (1/2) Epoch 2, batch 31150, giga_loss[loss=0.2956, simple_loss=0.373, pruned_loss=0.1091, over 28943.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5648361.31 frames. ], libri_tot_loss[loss=0.3797, simple_loss=0.4163, pruned_loss=0.1715, over 5705571.86 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.377, pruned_loss=0.1263, over 5631780.25 frames. ], batch size: 227, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:45:03,954 INFO [train.py:968] (1/2) Epoch 2, batch 31200, giga_loss[loss=0.3292, simple_loss=0.3915, pruned_loss=0.1334, over 28971.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.379, pruned_loss=0.1287, over 5655799.64 frames. ], libri_tot_loss[loss=0.3791, simple_loss=0.4157, pruned_loss=0.1712, over 5698373.91 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3781, pruned_loss=0.1263, over 5648383.08 frames. ], batch size: 213, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:45:33,383 INFO [optim.py:369] (1/2) 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 07:46:19,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6071, 2.5822, 1.4342, 1.4126], device='cuda:1'), covar=tensor([0.0822, 0.0458, 0.0916, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0454, 0.0334, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 07:46:19,548 INFO [train.py:968] (1/2) Epoch 2, batch 31250, giga_loss[loss=0.2842, simple_loss=0.3532, pruned_loss=0.1076, over 28026.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.376, pruned_loss=0.1263, over 5665596.76 frames. ], libri_tot_loss[loss=0.3783, simple_loss=0.4151, pruned_loss=0.1708, over 5702202.03 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3752, pruned_loss=0.1239, over 5655679.73 frames. ], batch size: 412, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:47:25,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6917, 1.6615, 5.5753, 3.9270], device='cuda:1'), covar=tensor([0.1200, 0.1565, 0.0215, 0.0368], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0493, 0.0640, 0.0506], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 07:47:26,153 INFO [train.py:968] (1/2) Epoch 2, batch 31300, libri_loss[loss=0.3033, simple_loss=0.3567, pruned_loss=0.1249, over 29576.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1237, over 5662334.51 frames. ], libri_tot_loss[loss=0.3779, simple_loss=0.4148, pruned_loss=0.1705, over 5702625.23 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3734, pruned_loss=0.1216, over 5653666.92 frames. ], batch size: 76, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:47:33,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3593, 1.7759, 1.3100, 1.4076], device='cuda:1'), covar=tensor([0.1024, 0.0368, 0.0445, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0186, 0.0194, 0.0343], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0025, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 07:47:36,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4819, 1.9891, 1.6167, 1.7166], device='cuda:1'), covar=tensor([0.1432, 0.1568, 0.1165, 0.0771], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0805, 0.0712, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 07:47:52,862 INFO [optim.py:369] (1/2) 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 07:48:30,784 INFO [train.py:968] (1/2) Epoch 2, batch 31350, giga_loss[loss=0.3462, simple_loss=0.3959, pruned_loss=0.1483, over 28535.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3725, pruned_loss=0.1221, over 5658684.02 frames. ], libri_tot_loss[loss=0.3783, simple_loss=0.415, pruned_loss=0.1708, over 5697296.23 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3714, pruned_loss=0.1198, over 5656491.74 frames. ], batch size: 78, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:48:50,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77037.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:49:18,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:49:26,816 INFO [train.py:968] (1/2) Epoch 2, batch 31400, giga_loss[loss=0.2905, simple_loss=0.353, pruned_loss=0.114, over 28491.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.37, pruned_loss=0.1225, over 5663367.56 frames. ], libri_tot_loss[loss=0.3772, simple_loss=0.4141, pruned_loss=0.1701, over 5695905.93 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3681, pruned_loss=0.1192, over 5660524.97 frames. ], batch size: 369, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:49:51,059 INFO [optim.py:369] (1/2) 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 07:49:54,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77091.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:50:29,856 INFO [train.py:968] (1/2) Epoch 2, batch 31450, giga_loss[loss=0.271, simple_loss=0.3409, pruned_loss=0.1005, over 28747.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3688, pruned_loss=0.1221, over 5665238.52 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.4128, pruned_loss=0.1693, over 5701927.78 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3671, pruned_loss=0.1188, over 5656700.73 frames. ], batch size: 243, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:51:32,570 INFO [train.py:968] (1/2) Epoch 2, batch 31500, giga_loss[loss=0.3158, simple_loss=0.3832, pruned_loss=0.1243, over 28598.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.369, pruned_loss=0.1219, over 5667561.59 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4129, pruned_loss=0.1694, over 5702612.14 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.119, over 5660051.22 frames. ], batch size: 307, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:51:47,149 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77180.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:51:49,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77183.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:51:55,334 INFO [optim.py:369] (1/2) 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 07:52:08,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7870, 1.3021, 3.8211, 3.0173], device='cuda:1'), covar=tensor([0.1541, 0.1780, 0.0319, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0496, 0.0640, 0.0511], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 07:52:17,308 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:21,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:25,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6859, 1.8178, 1.4852, 1.5877], device='cuda:1'), covar=tensor([0.1945, 0.2320, 0.1669, 0.1174], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0803, 0.0714, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 07:52:27,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:52:33,732 INFO [train.py:968] (1/2) Epoch 2, batch 31550, giga_loss[loss=0.3388, simple_loss=0.4046, pruned_loss=0.1365, over 28649.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1235, over 5664183.59 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.4127, pruned_loss=0.1694, over 5699664.47 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3698, pruned_loss=0.1197, over 5658937.84 frames. ], batch size: 242, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:52:39,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5895, 2.0456, 1.7308, 1.7798], device='cuda:1'), covar=tensor([0.1314, 0.1447, 0.1076, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0806, 0.0718, 0.0617], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 07:52:50,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:52,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 07:52:56,088 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 07:52:56,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77237.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:53:23,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.43 vs. limit=2.0
+2023-03-01 07:53:38,002 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77266.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 07:53:42,175 INFO [train.py:968] (1/2) Epoch 2, batch 31600, giga_loss[loss=0.2341, simple_loss=0.2941, pruned_loss=0.08703, over 24678.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1232, over 5663335.61 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.4127, pruned_loss=0.1697, over 5701746.63 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3691, pruned_loss=0.119, over 5656434.25 frames. ], batch size: 705, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:54:04,209 INFO [optim.py:369] (1/2) 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 07:54:48,121 INFO [train.py:968] (1/2) Epoch 2, batch 31650, giga_loss[loss=0.2784, simple_loss=0.3446, pruned_loss=0.1061, over 28217.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3697, pruned_loss=0.1225, over 5672176.92 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4122, pruned_loss=0.1694, over 5706635.07 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3666, pruned_loss=0.1178, over 5661015.60 frames. ], batch size: 412, lr: 1.25e-02, grad_scale: 8.0
+2023-03-01 07:55:56,963 INFO [train.py:968] (1/2) Epoch 2, batch 31700, giga_loss[loss=0.3044, simple_loss=0.3772, pruned_loss=0.1158, over 28716.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3737, pruned_loss=0.1253, over 5676384.04 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4117, pruned_loss=0.1691, over 5706874.35 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3704, pruned_loss=0.1205, over 5666928.14 frames. ], batch size: 262, lr: 1.25e-02, grad_scale: 4.0
+2023-03-01 07:56:21,435 INFO [optim.py:369] (1/2) 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 07:56:46,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0777, 1.1236, 1.1336, 1.0411], device='cuda:1'), covar=tensor([0.0919, 0.1085, 0.1380, 0.1032], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0782, 0.0612, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 07:57:02,120 INFO [train.py:968] (1/2) Epoch 2, batch 31750, giga_loss[loss=0.2962, simple_loss=0.3814, pruned_loss=0.1055, over 28588.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3778, pruned_loss=0.1261, over 5652447.49 frames. ], libri_tot_loss[loss=0.3748, simple_loss=0.4114, pruned_loss=0.1691, over 5701371.00 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3744, pruned_loss=0.1212, over 5649590.82 frames. ], batch size: 307, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 07:57:17,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3558, 1.2147, 1.1777, 1.6126], device='cuda:1'), covar=tensor([0.2133, 0.2070, 0.1751, 0.2341], device='cuda:1'), in_proj_covar=tensor([0.0985, 0.0796, 0.0894, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 07:57:19,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 07:58:08,435 INFO [train.py:968] (1/2) Epoch 2, batch 31800, giga_loss[loss=0.3016, simple_loss=0.3825, pruned_loss=0.1103, over 28921.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3782, pruned_loss=0.124, over 5660082.27 frames. ], libri_tot_loss[loss=0.3744, simple_loss=0.4111, pruned_loss=0.1688, over 5706152.58 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3751, pruned_loss=0.1193, over 5652719.23 frames. ], batch size: 284, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 07:58:33,818 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:59:12,684 INFO [train.py:968] (1/2) Epoch 2, batch 31850, giga_loss[loss=0.2657, simple_loss=0.3483, pruned_loss=0.09152, over 28948.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3778, pruned_loss=0.1218, over 5654465.98 frames. ], libri_tot_loss[loss=0.3744, simple_loss=0.4112, pruned_loss=0.1688, over 5698220.21 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.375, pruned_loss=0.1177, over 5656017.51 frames. ], batch size: 213, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 07:59:20,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 08:00:14,273 INFO [train.py:968] (1/2) Epoch 2, batch 31900, giga_loss[loss=0.3733, simple_loss=0.425, pruned_loss=0.1607, over 27570.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3772, pruned_loss=0.1216, over 5651012.65 frames. ], libri_tot_loss[loss=0.3748, simple_loss=0.4115, pruned_loss=0.169, over 5697422.65 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3741, pruned_loss=0.1173, over 5652006.24 frames. ], batch size: 472, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:00:38,313 INFO [optim.py:369] (1/2) 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 08:00:59,264 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:01:03,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:01:18,227 INFO [train.py:968] (1/2) Epoch 2, batch 31950, giga_loss[loss=0.3163, simple_loss=0.3807, pruned_loss=0.1259, over 28729.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.378, pruned_loss=0.1238, over 5651541.27 frames. ], libri_tot_loss[loss=0.3744, simple_loss=0.4113, pruned_loss=0.1688, over 5702804.07 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3746, pruned_loss=0.119, over 5646134.87 frames. ], batch size: 262, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:02:28,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=77666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:02:31,173 INFO [train.py:968] (1/2) Epoch 2, batch 32000, libri_loss[loss=0.3763, simple_loss=0.4106, pruned_loss=0.171, over 28665.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3787, pruned_loss=0.1251, over 5655189.22 frames. ], libri_tot_loss[loss=0.3743, simple_loss=0.4111, pruned_loss=0.1687, over 5693641.03 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3749, pruned_loss=0.1198, over 5657040.70 frames. ], batch size: 106, lr: 1.24e-02, grad_scale: 8.0
+2023-03-01 08:02:33,250 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:03:29,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0817, 1.2178, 1.1018, 0.8282], device='cuda:1'), covar=tensor([0.1845, 0.1859, 0.1600, 0.1872], device='cuda:1'), in_proj_covar=tensor([0.0967, 0.0794, 0.0878, 0.0933], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 08:03:46,075 INFO [train.py:968] (1/2) Epoch 2, batch 32050, giga_loss[loss=0.2375, simple_loss=0.3227, pruned_loss=0.0761, over 29136.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3777, pruned_loss=0.1247, over 5670660.33 frames. ], libri_tot_loss[loss=0.3735, simple_loss=0.4105, pruned_loss=0.1682, over 5697018.87 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3745, pruned_loss=0.12, over 5668386.08 frames. ], batch size: 146, lr: 1.24e-02, grad_scale: 8.0
+2023-03-01 08:04:33,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3361, 1.4864, 1.3535, 0.8741], device='cuda:1'), covar=tensor([0.0525, 0.0391, 0.0301, 0.0467], device='cuda:1'), in_proj_covar=tensor([0.1087, 0.0758, 0.0859, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 08:04:33,909 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:04:34,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-01 08:04:37,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:04:59,041 INFO [train.py:968] (1/2) Epoch 2, batch 32100, giga_loss[loss=0.2712, simple_loss=0.352, pruned_loss=0.09522, over 28872.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.374, pruned_loss=0.1223, over 5675976.75 frames. ], libri_tot_loss[loss=0.3735, simple_loss=0.4105, pruned_loss=0.1683, over 5702345.06 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3706, pruned_loss=0.1174, over 5668639.35 frames. ], batch size: 164, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:05:14,615 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:05:25,956 INFO [optim.py:369] (1/2) 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 08:05:28,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 08:06:08,155 INFO [train.py:968] (1/2) Epoch 2, batch 32150, giga_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.09186, over 29081.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.372, pruned_loss=0.1218, over 5657159.67 frames. ], libri_tot_loss[loss=0.3736, simple_loss=0.4104, pruned_loss=0.1684, over 5694522.46 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3687, pruned_loss=0.1169, over 5657985.28 frames. ], batch size: 175, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:06:46,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3674, 2.0212, 1.3551, 1.5348], device='cuda:1'), covar=tensor([0.0948, 0.0305, 0.0437, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0189, 0.0192, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 08:07:09,647 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 32200, giga_loss[loss=0.2988, simple_loss=0.3759, pruned_loss=0.1108, over 28759.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3733, pruned_loss=0.1226, over 5655340.66 frames. ], libri_tot_loss[loss=0.3733, simple_loss=0.4102, pruned_loss=0.1682, over 5687338.09 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3706, pruned_loss=0.1186, over 5661775.06 frames. ], batch size: 119, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:07:45,380 INFO [optim.py:369] (1/2) 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 08:08:19,694 INFO [train.py:968] (1/2) Epoch 2, batch 32250, giga_loss[loss=0.3751, simple_loss=0.4138, pruned_loss=0.1683, over 26845.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.377, pruned_loss=0.1243, over 5651663.00 frames. ], libri_tot_loss[loss=0.3733, simple_loss=0.4101, pruned_loss=0.1683, over 5679708.22 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3743, pruned_loss=0.1203, over 5662940.76 frames. ], batch size: 555, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:09:25,330 INFO [train.py:968] (1/2) Epoch 2, batch 32300, giga_loss[loss=0.2742, simple_loss=0.3406, pruned_loss=0.1039, over 28682.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3762, pruned_loss=0.1252, over 5654052.19 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4102, pruned_loss=0.1683, over 5682795.82 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3731, pruned_loss=0.1207, over 5660033.19 frames. ], batch size: 60, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:09:37,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:09:41,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7973, 2.4882, 1.7723, 0.6966], device='cuda:1'), covar=tensor([0.2212, 0.1331, 0.1547, 0.2275], device='cuda:1'), in_proj_covar=tensor([0.1202, 0.1188, 0.1231, 0.1026], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 08:09:53,657 INFO [optim.py:369] (1/2) 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 08:10:13,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:10:17,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:10:17,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 32350, libri_loss[loss=0.2872, simple_loss=0.3382, pruned_loss=0.1181, over 28497.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.376, pruned_loss=0.1262, over 5661508.44 frames. ], libri_tot_loss[loss=0.3723, simple_loss=0.4092, pruned_loss=0.1677, over 5686822.17 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3736, pruned_loss=0.1222, over 5661875.90 frames. ], batch size: 63, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:10:52,753 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:10:56,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:11:01,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:11:33,293 INFO [train.py:968] (1/2) Epoch 2, batch 32400, giga_loss[loss=0.3147, simple_loss=0.3831, pruned_loss=0.1232, over 28904.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5666243.85 frames. ], libri_tot_loss[loss=0.3729, simple_loss=0.4094, pruned_loss=0.1681, over 5690300.52 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3744, pruned_loss=0.1226, over 5662937.25 frames. ], batch size: 227, lr: 1.24e-02, grad_scale: 8.0
+2023-03-01 08:11:33,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5942, 1.3661, 1.3704, 1.3939], device='cuda:1'), covar=tensor([0.0794, 0.1613, 0.1436, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0802, 0.0623, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 08:11:35,227 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:11:51,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9632, 2.8084, 2.3376, 2.1563], device='cuda:1'), covar=tensor([0.1521, 0.1330, 0.0989, 0.0737], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0790, 0.0701, 0.0604], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:12:08,791 INFO [optim.py:369] (1/2) 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 08:12:17,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-01 08:12:47,653 INFO [train.py:968] (1/2) Epoch 2, batch 32450, giga_loss[loss=0.2991, simple_loss=0.3735, pruned_loss=0.1124, over 28335.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3783, pruned_loss=0.1262, over 5668272.08 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.4093, pruned_loss=0.168, over 5694976.97 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3756, pruned_loss=0.1219, over 5660677.41 frames. ], batch size: 368, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:12:51,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7267, 1.6181, 1.5536, 1.5727], device='cuda:1'), covar=tensor([0.0762, 0.1606, 0.1278, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0802, 0.0625, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 08:12:51,825 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:12:56,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:13:04,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6522, 2.3686, 1.9736, 1.9297], device='cuda:1'), covar=tensor([0.1686, 0.1525, 0.1128, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0791, 0.0699, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:13:05,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4382, 2.6360, 1.4568, 1.2217], device='cuda:1'), covar=tensor([0.0902, 0.0469, 0.0913, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0455, 0.0335, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 08:13:27,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3462, 1.2636, 1.0902, 1.1910], device='cuda:1'), covar=tensor([0.0558, 0.0390, 0.0776, 0.0716], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0494, 0.0543, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 08:13:44,220 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:13:54,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4476, 2.5743, 1.4443, 1.2844], device='cuda:1'), covar=tensor([0.0903, 0.0583, 0.1011, 0.1489], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0457, 0.0336, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 08:14:09,298 INFO [train.py:968] (1/2) Epoch 2, batch 32500, giga_loss[loss=0.3138, simple_loss=0.3844, pruned_loss=0.1216, over 28731.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3777, pruned_loss=0.125, over 5669373.52 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.4093, pruned_loss=0.168, over 5696371.39 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3752, pruned_loss=0.1211, over 5661949.41 frames. ], batch size: 262, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:14:33,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:37,149 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:39,177 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:14:44,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:14:57,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9584, 2.6801, 1.8978, 1.0356], device='cuda:1'), covar=tensor([0.1854, 0.1108, 0.1393, 0.1980], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.1183, 0.1217, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 08:15:16,163 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 32550, giga_loss[loss=0.2998, simple_loss=0.361, pruned_loss=0.1193, over 28468.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.124, over 5669790.69 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.409, pruned_loss=0.1678, over 5691911.10 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 5667454.10 frames. ], batch size: 336, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:15:19,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:16:13,415 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.05 vs. limit=2.0
+2023-03-01 08:16:26,964 INFO [train.py:968] (1/2) Epoch 2, batch 32600, giga_loss[loss=0.2941, simple_loss=0.3584, pruned_loss=0.1149, over 27678.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3671, pruned_loss=0.1205, over 5668970.01 frames. ], libri_tot_loss[loss=0.3726, simple_loss=0.4092, pruned_loss=0.168, over 5693695.90 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3647, pruned_loss=0.1168, over 5665360.54 frames. ], batch size: 472, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:16:31,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2732, 1.7153, 1.1720, 1.3068], device='cuda:1'), covar=tensor([0.1006, 0.0337, 0.0469, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0184, 0.0190, 0.0337], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0025, 0.0023, 0.0038], device='cuda:1')
+2023-03-01 08:16:55,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4337, 1.9152, 1.6103, 1.6628], device='cuda:1'), covar=tensor([0.1133, 0.1286, 0.0948, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0785, 0.0701, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:17:00,316 INFO [optim.py:369] (1/2) 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 08:17:34,594 INFO [train.py:968] (1/2) Epoch 2, batch 32650, giga_loss[loss=0.2971, simple_loss=0.3644, pruned_loss=0.1149, over 29042.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3651, pruned_loss=0.1195, over 5666636.57 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.4092, pruned_loss=0.1681, over 5696783.56 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3625, pruned_loss=0.1159, over 5660521.32 frames. ], batch size: 285, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:18:33,257 INFO [train.py:968] (1/2) Epoch 2, batch 32700, giga_loss[loss=0.2881, simple_loss=0.3641, pruned_loss=0.1061, over 28901.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3677, pruned_loss=0.122, over 5662308.99 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4096, pruned_loss=0.1685, over 5698404.65 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3643, pruned_loss=0.1177, over 5655347.79 frames. ], batch size: 145, lr: 1.24e-02, grad_scale: 2.0
+2023-03-01 08:18:34,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2005, 1.0915, 0.9726, 0.9862], device='cuda:1'), covar=tensor([0.0567, 0.0474, 0.0887, 0.0749], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0500, 0.0542, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 08:18:53,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:18:58,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3577, 1.8913, 1.3149, 0.6518], device='cuda:1'), covar=tensor([0.1752, 0.0945, 0.1489, 0.1812], device='cuda:1'), in_proj_covar=tensor([0.1210, 0.1191, 0.1230, 0.1025], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 08:19:00,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8019, 1.2303, 3.4779, 2.9165], device='cuda:1'), covar=tensor([0.1447, 0.1797, 0.0424, 0.0491], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0499, 0.0652, 0.0503], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 08:19:01,292 INFO [optim.py:369] (1/2) 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 08:19:23,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0832, 2.3115, 1.4784, 1.2785], device='cuda:1'), covar=tensor([0.0601, 0.0431, 0.0457, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.1095, 0.0742, 0.0855, 0.0910], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 08:19:26,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-03-01 08:19:31,177 INFO [train.py:968] (1/2) Epoch 2, batch 32750, giga_loss[loss=0.2604, simple_loss=0.3413, pruned_loss=0.08976, over 28888.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3665, pruned_loss=0.1213, over 5661243.31 frames. ], libri_tot_loss[loss=0.3729, simple_loss=0.4093, pruned_loss=0.1683, over 5701563.63 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3627, pruned_loss=0.1165, over 5651460.12 frames. ], batch size: 164, lr: 1.24e-02, grad_scale: 2.0
+2023-03-01 08:20:06,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:20:34,757 INFO [train.py:968] (1/2) Epoch 2, batch 32800, giga_loss[loss=0.3119, simple_loss=0.3757, pruned_loss=0.124, over 28589.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3653, pruned_loss=0.1192, over 5665300.64 frames. ], libri_tot_loss[loss=0.3733, simple_loss=0.4096, pruned_loss=0.1685, over 5704379.05 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.114, over 5653796.39 frames. ], batch size: 242, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:20:47,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5763, 1.5621, 1.2817, 1.1366], device='cuda:1'), covar=tensor([0.0565, 0.0489, 0.0428, 0.0537], device='cuda:1'), in_proj_covar=tensor([0.1081, 0.0736, 0.0844, 0.0902], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 08:20:58,723 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 08:21:03,085 INFO [optim.py:369] (1/2) 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 08:21:20,901 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-01 08:21:40,509 INFO [train.py:968] (1/2) Epoch 2, batch 32850, giga_loss[loss=0.3118, simple_loss=0.3723, pruned_loss=0.1257, over 28438.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.363, pruned_loss=0.1177, over 5668120.19 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.4093, pruned_loss=0.1683, over 5706311.22 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3594, pruned_loss=0.1134, over 5657230.46 frames. ], batch size: 368, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:21:54,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:21:57,365 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:22:13,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3358, 4.7865, 5.0196, 2.1115], device='cuda:1'), covar=tensor([0.0278, 0.0231, 0.0545, 0.1698], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0547, 0.0774, 0.0536], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:22:38,734 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 32900, giga_loss[loss=0.2844, simple_loss=0.3319, pruned_loss=0.1184, over 24506.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3624, pruned_loss=0.1174, over 5662024.50 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.409, pruned_loss=0.1683, over 5708464.74 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3587, pruned_loss=0.1127, over 5650440.88 frames. ], batch size: 705, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:23:12,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 08:23:13,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:23:19,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:23:19,338 INFO [optim.py:369] (1/2) 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 08:23:23,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6899, 4.1131, 4.3810, 1.7606], device='cuda:1'), covar=tensor([0.0344, 0.0294, 0.0617, 0.1814], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0553, 0.0779, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:23:39,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5877, 1.3720, 1.2846, 1.2272], device='cuda:1'), covar=tensor([0.0672, 0.0618, 0.0978, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0505, 0.0546, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 08:23:57,094 INFO [train.py:968] (1/2) Epoch 2, batch 32950, giga_loss[loss=0.2582, simple_loss=0.3373, pruned_loss=0.0895, over 29029.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3627, pruned_loss=0.1174, over 5646687.30 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4085, pruned_loss=0.168, over 5692948.43 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3593, pruned_loss=0.113, over 5648783.51 frames. ], batch size: 175, lr: 1.24e-02, grad_scale: 4.0
+2023-03-01 08:23:58,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:23:59,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0635, 2.4176, 1.7158, 1.2282], device='cuda:1'), covar=tensor([0.0656, 0.0306, 0.0337, 0.0582], device='cuda:1'), in_proj_covar=tensor([0.1098, 0.0733, 0.0839, 0.0912], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 08:24:02,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-01 08:24:05,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:24:15,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-01 08:25:05,328 INFO [train.py:968] (1/2) Epoch 2, batch 33000, giga_loss[loss=0.35, simple_loss=0.4082, pruned_loss=0.1459, over 28764.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3647, pruned_loss=0.1198, over 5634867.56 frames. ], libri_tot_loss[loss=0.3731, simple_loss=0.4089, pruned_loss=0.1686, over 5676904.47 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3606, pruned_loss=0.1146, over 5651067.09 frames. ], batch size: 263, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:25:05,328 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 08:25:12,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4885, 1.3992, 1.4057, 1.4355], device='cuda:1'), covar=tensor([0.0882, 0.1446, 0.1453, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0792, 0.0615, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 08:25:14,380 INFO [train.py:1012] (1/2) Epoch 2, validation: loss=0.2412, simple_loss=0.3357, pruned_loss=0.07337, over 944034.00 frames. 
+2023-03-01 08:25:14,380 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 08:25:19,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1495, 1.2775, 1.1411, 0.9923], device='cuda:1'), covar=tensor([0.1676, 0.1583, 0.1367, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0963, 0.0785, 0.0881, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 08:25:34,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3753, 1.9147, 1.3735, 0.6745], device='cuda:1'), covar=tensor([0.1203, 0.0813, 0.1340, 0.1466], device='cuda:1'), in_proj_covar=tensor([0.1178, 0.1160, 0.1201, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 08:25:37,704 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-01 08:25:41,508 INFO [optim.py:369] (1/2) 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 08:26:18,594 INFO [train.py:968] (1/2) Epoch 2, batch 33050, giga_loss[loss=0.2961, simple_loss=0.3519, pruned_loss=0.1201, over 27605.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3628, pruned_loss=0.1181, over 5644022.68 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4084, pruned_loss=0.1682, over 5680450.36 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3595, pruned_loss=0.1138, over 5653196.93 frames. ], batch size: 472, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:27:02,118 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-01 08:27:14,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4705, 2.5747, 1.3154, 1.2007], device='cuda:1'), covar=tensor([0.0892, 0.0477, 0.0918, 0.1441], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0444, 0.0336, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 08:27:20,444 INFO [train.py:968] (1/2) Epoch 2, batch 33100, giga_loss[loss=0.3073, simple_loss=0.3788, pruned_loss=0.1179, over 28943.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1183, over 5652049.16 frames. ], libri_tot_loss[loss=0.3723, simple_loss=0.4082, pruned_loss=0.1682, over 5682037.81 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3625, pruned_loss=0.1141, over 5657346.61 frames. ], batch size: 106, lr: 1.23e-02, grad_scale: 2.0
+2023-03-01 08:27:50,169 INFO [optim.py:369] (1/2) 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 08:27:53,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4460, 1.9050, 1.5603, 1.6046], device='cuda:1'), covar=tensor([0.1177, 0.1322, 0.1040, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0778, 0.0706, 0.0604], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:28:23,249 INFO [train.py:968] (1/2) Epoch 2, batch 33150, giga_loss[loss=0.3105, simple_loss=0.3863, pruned_loss=0.1173, over 28652.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3683, pruned_loss=0.1191, over 5656499.34 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4078, pruned_loss=0.168, over 5686616.22 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3654, pruned_loss=0.1151, over 5656099.30 frames. ], batch size: 92, lr: 1.23e-02, grad_scale: 2.0
+2023-03-01 08:29:06,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=78853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:29:23,609 INFO [train.py:968] (1/2) Epoch 2, batch 33200, giga_loss[loss=0.3761, simple_loss=0.4039, pruned_loss=0.1741, over 26810.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3699, pruned_loss=0.1207, over 5653290.51 frames. ], libri_tot_loss[loss=0.3723, simple_loss=0.408, pruned_loss=0.1683, over 5692217.46 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3661, pruned_loss=0.1155, over 5646473.94 frames. ], batch size: 555, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:30:00,684 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 33250, giga_loss[loss=0.3509, simple_loss=0.3902, pruned_loss=0.1558, over 26812.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3693, pruned_loss=0.1201, over 5658378.08 frames. ], libri_tot_loss[loss=0.372, simple_loss=0.4077, pruned_loss=0.1681, over 5694386.61 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3662, pruned_loss=0.1157, over 5650692.49 frames. ], batch size: 555, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:30:53,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1806, 0.9394, 0.8087, 1.3177], device='cuda:1'), covar=tensor([0.1048, 0.0419, 0.0517, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0280, 0.0186, 0.0189, 0.0336], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 08:30:56,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6081, 2.4895, 1.6565, 0.7388], device='cuda:1'), covar=tensor([0.2519, 0.1251, 0.1701, 0.2437], device='cuda:1'), in_proj_covar=tensor([0.1180, 0.1164, 0.1215, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 08:31:10,167 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 33300, giga_loss[loss=0.2963, simple_loss=0.3665, pruned_loss=0.1131, over 27917.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3678, pruned_loss=0.1192, over 5655064.64 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4069, pruned_loss=0.1676, over 5689238.07 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3651, pruned_loss=0.1151, over 5653341.82 frames. ], batch size: 412, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:31:36,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0768, 1.4200, 4.1360, 3.1356], device='cuda:1'), covar=tensor([0.1341, 0.1663, 0.0328, 0.0463], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0493, 0.0629, 0.0491], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 08:32:01,559 INFO [optim.py:369] (1/2) 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 08:32:10,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:32:32,921 INFO [train.py:968] (1/2) Epoch 2, batch 33350, libri_loss[loss=0.3268, simple_loss=0.3634, pruned_loss=0.1451, over 29483.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3672, pruned_loss=0.1194, over 5654149.22 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4067, pruned_loss=0.1673, over 5688925.76 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3639, pruned_loss=0.1146, over 5651596.33 frames. ], batch size: 70, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:32:54,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4783, 2.2187, 1.5888, 0.4049], device='cuda:1'), covar=tensor([0.1635, 0.1035, 0.1665, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.1181, 0.1172, 0.1209, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 08:33:32,797 INFO [train.py:968] (1/2) Epoch 2, batch 33400, giga_loss[loss=0.3374, simple_loss=0.3911, pruned_loss=0.1419, over 28640.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.365, pruned_loss=0.119, over 5654830.89 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4066, pruned_loss=0.1674, over 5682728.15 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1144, over 5657195.30 frames. ], batch size: 262, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:33:34,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-01 08:34:05,946 INFO [optim.py:369] (1/2) 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 08:34:38,798 INFO [train.py:968] (1/2) Epoch 2, batch 33450, giga_loss[loss=0.3169, simple_loss=0.3823, pruned_loss=0.1257, over 28509.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3661, pruned_loss=0.119, over 5666592.61 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4056, pruned_loss=0.1667, over 5687487.21 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3636, pruned_loss=0.115, over 5663698.61 frames. ], batch size: 336, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:35:15,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:35:18,419 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:35:47,871 INFO [train.py:968] (1/2) Epoch 2, batch 33500, giga_loss[loss=0.2562, simple_loss=0.3403, pruned_loss=0.08603, over 28999.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3672, pruned_loss=0.1191, over 5665788.13 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4051, pruned_loss=0.1662, over 5689680.91 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3652, pruned_loss=0.1157, over 5661010.25 frames. ], batch size: 155, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:35:54,214 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 2, batch 33550, giga_loss[loss=0.2984, simple_loss=0.3593, pruned_loss=0.1187, over 28479.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3691, pruned_loss=0.121, over 5669116.86 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4051, pruned_loss=0.1661, over 5695488.79 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3665, pruned_loss=0.1172, over 5659704.76 frames. ], batch size: 78, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:37:09,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:37:10,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8601, 1.5739, 1.2906, 1.2945], device='cuda:1'), covar=tensor([0.0689, 0.0697, 0.0955, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0497, 0.0539, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 08:37:27,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5022, 1.2328, 2.9087, 2.6539], device='cuda:1'), covar=tensor([0.1366, 0.1617, 0.0437, 0.0508], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0489, 0.0639, 0.0492], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 08:38:05,436 INFO [train.py:968] (1/2) Epoch 2, batch 33600, giga_loss[loss=0.3207, simple_loss=0.3848, pruned_loss=0.1283, over 28943.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3739, pruned_loss=0.1236, over 5674725.01 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4053, pruned_loss=0.1663, over 5698058.42 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3712, pruned_loss=0.1198, over 5664633.91 frames. ], batch size: 106, lr: 1.23e-02, grad_scale: 8.0
+2023-03-01 08:38:32,744 INFO [optim.py:369] (1/2) 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 08:39:04,041 INFO [train.py:968] (1/2) Epoch 2, batch 33650, giga_loss[loss=0.3088, simple_loss=0.3751, pruned_loss=0.1213, over 29036.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3761, pruned_loss=0.124, over 5661210.78 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4051, pruned_loss=0.1661, over 5692163.07 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3737, pruned_loss=0.1205, over 5658064.37 frames. ], batch size: 128, lr: 1.23e-02, grad_scale: 8.0
+2023-03-01 08:39:10,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:39:57,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:40:09,334 INFO [train.py:968] (1/2) Epoch 2, batch 33700, giga_loss[loss=0.304, simple_loss=0.3707, pruned_loss=0.1186, over 28954.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3747, pruned_loss=0.1234, over 5664096.63 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4046, pruned_loss=0.1658, over 5696640.27 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3722, pruned_loss=0.1196, over 5656563.92 frames. ], batch size: 213, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:40:11,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:40:17,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:40:46,235 INFO [optim.py:369] (1/2) 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 08:40:52,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1803, 1.5464, 1.2007, 1.3442], device='cuda:1'), covar=tensor([0.0918, 0.0457, 0.0446, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0184, 0.0188, 0.0335], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 08:40:58,531 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:41:23,035 INFO [train.py:968] (1/2) Epoch 2, batch 33750, giga_loss[loss=0.288, simple_loss=0.3598, pruned_loss=0.1081, over 28834.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3721, pruned_loss=0.122, over 5670074.07 frames. ], libri_tot_loss[loss=0.3672, simple_loss=0.404, pruned_loss=0.1653, over 5702715.02 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3699, pruned_loss=0.1182, over 5657682.27 frames. ], batch size: 164, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:41:25,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.02 vs. limit=2.0
+2023-03-01 08:42:02,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3615, 1.2802, 1.1796, 1.8881], device='cuda:1'), covar=tensor([0.1963, 0.1956, 0.1675, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.0973, 0.0796, 0.0884, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 08:42:23,927 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:42:25,270 INFO [train.py:968] (1/2) Epoch 2, batch 33800, giga_loss[loss=0.3343, simple_loss=0.3975, pruned_loss=0.1355, over 29037.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3727, pruned_loss=0.1235, over 5669708.33 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4037, pruned_loss=0.1652, over 5703895.65 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3701, pruned_loss=0.119, over 5657627.55 frames. ], batch size: 285, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:42:27,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:42:45,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3599, 2.0063, 1.3230, 1.5749], device='cuda:1'), covar=tensor([0.0889, 0.0287, 0.0410, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0184, 0.0188, 0.0333], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 08:43:01,131 INFO [optim.py:369] (1/2) 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 08:43:06,832 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:43:31,570 INFO [train.py:968] (1/2) Epoch 2, batch 33850, giga_loss[loss=0.352, simple_loss=0.4051, pruned_loss=0.1495, over 28962.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1235, over 5662054.14 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4041, pruned_loss=0.1656, over 5700089.65 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3689, pruned_loss=0.1183, over 5654272.53 frames. ], batch size: 199, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:44:38,678 INFO [train.py:968] (1/2) Epoch 2, batch 33900, giga_loss[loss=0.2671, simple_loss=0.3398, pruned_loss=0.09718, over 28946.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3718, pruned_loss=0.124, over 5669853.17 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.404, pruned_loss=0.1657, over 5702867.32 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3682, pruned_loss=0.1186, over 5660037.34 frames. ], batch size: 227, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:45:13,650 INFO [optim.py:369] (1/2) 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 08:45:47,380 INFO [train.py:968] (1/2) Epoch 2, batch 33950, giga_loss[loss=0.2782, simple_loss=0.3505, pruned_loss=0.1029, over 28788.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3698, pruned_loss=0.1234, over 5652584.74 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4033, pruned_loss=0.1652, over 5705608.30 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3671, pruned_loss=0.1189, over 5641580.30 frames. ], batch size: 119, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:46:38,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 08:46:40,025 INFO [train.py:968] (1/2) Epoch 2, batch 34000, giga_loss[loss=0.2857, simple_loss=0.3708, pruned_loss=0.1003, over 28992.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.369, pruned_loss=0.1216, over 5669180.48 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4024, pruned_loss=0.1644, over 5714587.87 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.366, pruned_loss=0.1167, over 5649325.15 frames. ], batch size: 285, lr: 1.23e-02, grad_scale: 8.0
+2023-03-01 08:46:56,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79678.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 08:47:17,195 INFO [optim.py:369] (1/2) 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 08:47:42,329 INFO [train.py:968] (1/2) Epoch 2, batch 34050, giga_loss[loss=0.2854, simple_loss=0.3738, pruned_loss=0.09848, over 28548.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3685, pruned_loss=0.1195, over 5672139.35 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4021, pruned_loss=0.1642, over 5709435.94 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3656, pruned_loss=0.1147, over 5660337.27 frames. ], batch size: 71, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:47:52,214 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:47:57,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4688, 1.4440, 1.1002, 1.1968], device='cuda:1'), covar=tensor([0.0715, 0.0586, 0.1039, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0491, 0.0537, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 08:48:02,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:48:22,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-01 08:48:31,287 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:48:42,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79768.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 08:48:42,579 INFO [train.py:968] (1/2) Epoch 2, batch 34100, giga_loss[loss=0.3711, simple_loss=0.4242, pruned_loss=0.159, over 28348.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3711, pruned_loss=0.1191, over 5665617.16 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.402, pruned_loss=0.1642, over 5702235.38 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3683, pruned_loss=0.1146, over 5662582.42 frames. ], batch size: 368, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:49:17,226 INFO [optim.py:369] (1/2) 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 08:49:44,293 INFO [train.py:968] (1/2) Epoch 2, batch 34150, giga_loss[loss=0.3622, simple_loss=0.4141, pruned_loss=0.1551, over 28675.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3722, pruned_loss=0.1194, over 5664401.27 frames. ], libri_tot_loss[loss=0.3651, simple_loss=0.4019, pruned_loss=0.1641, over 5704534.13 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3696, pruned_loss=0.1152, over 5659327.94 frames. ], batch size: 242, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:50:27,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3491, 1.8106, 1.3465, 0.7042], device='cuda:1'), covar=tensor([0.1415, 0.0828, 0.1158, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.1197, 0.1177, 0.1201, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 08:50:46,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4904, 1.4537, 1.1413, 1.1820], device='cuda:1'), covar=tensor([0.0666, 0.0532, 0.0967, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0496, 0.0549, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 08:50:51,256 INFO [train.py:968] (1/2) Epoch 2, batch 34200, giga_loss[loss=0.3072, simple_loss=0.3503, pruned_loss=0.132, over 24607.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3727, pruned_loss=0.1207, over 5654851.50 frames. ], libri_tot_loss[loss=0.3637, simple_loss=0.4008, pruned_loss=0.1633, over 5693739.61 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.37, pruned_loss=0.1156, over 5658294.81 frames. ], batch size: 705, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:51:01,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:51:04,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:51:27,937 INFO [optim.py:369] (1/2) 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 08:51:49,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:52:02,666 INFO [train.py:968] (1/2) Epoch 2, batch 34250, libri_loss[loss=0.329, simple_loss=0.3747, pruned_loss=0.1416, over 29579.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.372, pruned_loss=0.1193, over 5656889.83 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4005, pruned_loss=0.1631, over 5686366.13 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3698, pruned_loss=0.115, over 5666083.86 frames. ], batch size: 78, lr: 1.23e-02, grad_scale: 4.0
+2023-03-01 08:52:15,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9168, 3.4722, 3.6657, 1.6448], device='cuda:1'), covar=tensor([0.0524, 0.0473, 0.0774, 0.1904], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0550, 0.0780, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:53:14,397 INFO [train.py:968] (1/2) Epoch 2, batch 34300, giga_loss[loss=0.2323, simple_loss=0.2948, pruned_loss=0.08484, over 24954.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3708, pruned_loss=0.1184, over 5655793.44 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4001, pruned_loss=0.1628, over 5690471.19 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3689, pruned_loss=0.1143, over 5658744.06 frames. ], batch size: 705, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:53:27,849 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=79978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:53:48,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7339, 1.2678, 3.8512, 2.8961], device='cuda:1'), covar=tensor([0.1540, 0.1885, 0.0292, 0.0604], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0497, 0.0631, 0.0499], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 08:53:53,958 INFO [optim.py:369] (1/2) 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 08:54:06,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4893, 2.9791, 1.4534, 1.2317], device='cuda:1'), covar=tensor([0.0992, 0.0496, 0.0974, 0.1574], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0443, 0.0330, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 08:54:25,012 INFO [train.py:968] (1/2) Epoch 2, batch 34350, giga_loss[loss=0.3359, simple_loss=0.4012, pruned_loss=0.1353, over 28480.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3718, pruned_loss=0.119, over 5658147.55 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4002, pruned_loss=0.1628, over 5692994.53 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3695, pruned_loss=0.1145, over 5657182.68 frames. ], batch size: 336, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:54:54,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4912, 2.2630, 1.4914, 0.7215], device='cuda:1'), covar=tensor([0.2095, 0.0947, 0.1627, 0.2011], device='cuda:1'), in_proj_covar=tensor([0.1205, 0.1178, 0.1197, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 08:55:07,400 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:55:11,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80053.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 08:55:18,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6047, 1.9654, 1.6627, 1.7832], device='cuda:1'), covar=tensor([0.1236, 0.1315, 0.1065, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0765, 0.0695, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 08:55:30,502 INFO [train.py:968] (1/2) Epoch 2, batch 34400, giga_loss[loss=0.2874, simple_loss=0.372, pruned_loss=0.1014, over 28959.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3755, pruned_loss=0.1206, over 5654317.16 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4004, pruned_loss=0.1632, over 5686697.74 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3729, pruned_loss=0.116, over 5659431.80 frames. ], batch size: 145, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 08:55:39,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5895, 1.5672, 1.5010, 2.0956], device='cuda:1'), covar=tensor([0.1928, 0.1858, 0.1494, 0.1833], device='cuda:1'), in_proj_covar=tensor([0.0975, 0.0801, 0.0890, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 08:56:00,846 INFO [optim.py:369] (1/2) 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 08:56:09,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:56:29,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0382, 3.8474, 1.7802, 1.6478], device='cuda:1'), covar=tensor([0.0808, 0.0333, 0.0817, 0.1359], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0448, 0.0333, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 08:56:35,036 INFO [train.py:968] (1/2) Epoch 2, batch 34450, giga_loss[loss=0.3177, simple_loss=0.3839, pruned_loss=0.1257, over 28441.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3763, pruned_loss=0.1211, over 5673435.40 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.3999, pruned_loss=0.1628, over 5694560.48 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3738, pruned_loss=0.1162, over 5669753.87 frames. ], batch size: 336, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 08:56:55,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:57:11,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80143.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 08:57:23,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2023, 1.5883, 1.2704, 1.4977], device='cuda:1'), covar=tensor([0.0959, 0.0375, 0.0440, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0184, 0.0187, 0.0332], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 08:57:29,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:57:45,134 INFO [train.py:968] (1/2) Epoch 2, batch 34500, giga_loss[loss=0.2928, simple_loss=0.3644, pruned_loss=0.1106, over 28682.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3745, pruned_loss=0.1208, over 5681671.89 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.3998, pruned_loss=0.1627, over 5698578.95 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3722, pruned_loss=0.1162, over 5674681.54 frames. ], batch size: 262, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:58:24,626 INFO [optim.py:369] (1/2) 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 08:58:25,135 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80196.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 08:58:28,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80199.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 08:58:55,052 INFO [train.py:968] (1/2) Epoch 2, batch 34550, giga_loss[loss=0.3098, simple_loss=0.3767, pruned_loss=0.1215, over 28882.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3718, pruned_loss=0.1192, over 5682638.98 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4, pruned_loss=0.1627, over 5697220.73 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.369, pruned_loss=0.1143, over 5677831.89 frames. ], batch size: 227, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 08:59:07,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80228.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 08:59:33,765 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 08:59:40,046 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 34600, giga_loss[loss=0.272, simple_loss=0.3508, pruned_loss=0.09659, over 28885.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3685, pruned_loss=0.1158, over 5684534.89 frames. ], libri_tot_loss[loss=0.3625, simple_loss=0.3998, pruned_loss=0.1626, over 5692085.36 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3659, pruned_loss=0.1111, over 5684876.87 frames. ], batch size: 174, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:00:15,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:00:15,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 09:00:16,500 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:00:19,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:00:28,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:00:32,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80289.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:00:42,739 INFO [optim.py:369] (1/2) 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 09:00:59,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:01:11,249 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80318.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:01:11,625 INFO [train.py:968] (1/2) Epoch 2, batch 34650, giga_loss[loss=0.2794, simple_loss=0.3494, pruned_loss=0.1047, over 27610.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3696, pruned_loss=0.1169, over 5688878.56 frames. ], libri_tot_loss[loss=0.363, simple_loss=0.4002, pruned_loss=0.1629, over 5696387.66 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3666, pruned_loss=0.1121, over 5685350.22 frames. ], batch size: 472, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:01:28,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5424, 1.5586, 1.5921, 1.5467], device='cuda:1'), covar=tensor([0.0912, 0.1413, 0.1014, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0784, 0.0613, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 09:01:30,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4679, 1.3947, 1.1165, 1.2419], device='cuda:1'), covar=tensor([0.0688, 0.0602, 0.0961, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0505, 0.0555, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 09:01:55,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:02:13,840 INFO [train.py:968] (1/2) Epoch 2, batch 34700, giga_loss[loss=0.3241, simple_loss=0.3907, pruned_loss=0.1287, over 28919.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3715, pruned_loss=0.118, over 5678342.34 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4003, pruned_loss=0.163, over 5690055.00 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3685, pruned_loss=0.1133, over 5680852.72 frames. ], batch size: 227, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:02:46,164 INFO [optim.py:369] (1/2) 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 09:03:07,970 INFO [train.py:968] (1/2) Epoch 2, batch 34750, giga_loss[loss=0.3271, simple_loss=0.3916, pruned_loss=0.1313, over 28324.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3741, pruned_loss=0.1207, over 5674202.03 frames. ], libri_tot_loss[loss=0.363, simple_loss=0.4001, pruned_loss=0.163, over 5695029.26 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3708, pruned_loss=0.1153, over 5670782.00 frames. ], batch size: 368, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:03:18,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:04:10,477 INFO [train.py:968] (1/2) Epoch 2, batch 34800, libri_loss[loss=0.3085, simple_loss=0.3537, pruned_loss=0.1317, over 29651.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3703, pruned_loss=0.1194, over 5665538.87 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4, pruned_loss=0.1629, over 5689153.19 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3674, pruned_loss=0.1144, over 5667247.97 frames. ], batch size: 73, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 09:04:16,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4847, 1.3847, 1.4048, 1.3106], device='cuda:1'), covar=tensor([0.0818, 0.1264, 0.1293, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0781, 0.0613, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 09:04:44,572 INFO [optim.py:369] (1/2) 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 09:04:45,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:04:49,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:04:57,722 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-01 09:05:14,443 INFO [train.py:968] (1/2) Epoch 2, batch 34850, giga_loss[loss=0.2823, simple_loss=0.3537, pruned_loss=0.1055, over 28953.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.37, pruned_loss=0.1198, over 5660750.62 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.3999, pruned_loss=0.1629, over 5681936.29 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3673, pruned_loss=0.1152, over 5668581.52 frames. ], batch size: 199, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:05:22,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:05:25,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:07,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:07,433 INFO [train.py:968] (1/2) Epoch 2, batch 34900, giga_loss[loss=0.4307, simple_loss=0.4696, pruned_loss=0.1959, over 28505.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3741, pruned_loss=0.1244, over 5658377.27 frames. ], libri_tot_loss[loss=0.362, simple_loss=0.3992, pruned_loss=0.1624, over 5688402.78 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3713, pruned_loss=0.1194, over 5658041.98 frames. ], batch size: 85, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:06:09,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:36,294 INFO [optim.py:369] (1/2) 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 09:06:40,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:06:44,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3134, 1.2789, 1.1203, 1.7782], device='cuda:1'), covar=tensor([0.1918, 0.1804, 0.1638, 0.1978], device='cuda:1'), in_proj_covar=tensor([0.0958, 0.0777, 0.0879, 0.0931], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 09:06:56,329 INFO [train.py:968] (1/2) Epoch 2, batch 34950, giga_loss[loss=0.3658, simple_loss=0.423, pruned_loss=0.1543, over 28688.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3856, pruned_loss=0.1322, over 5670482.39 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.3993, pruned_loss=0.1625, over 5690721.24 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3832, pruned_loss=0.1279, over 5668115.85 frames. ], batch size: 262, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:07:43,280 INFO [train.py:968] (1/2) Epoch 2, batch 35000, giga_loss[loss=0.3962, simple_loss=0.4449, pruned_loss=0.1737, over 28907.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3943, pruned_loss=0.1384, over 5670939.69 frames. ], libri_tot_loss[loss=0.3614, simple_loss=0.3988, pruned_loss=0.162, over 5688808.26 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3924, pruned_loss=0.1345, over 5670577.98 frames. ], batch size: 186, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:07:46,135 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:07:48,324 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:07:48,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:08:07,903 INFO [optim.py:369] (1/2) 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 09:08:14,893 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:08:23,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2539, 1.4398, 1.2929, 0.7066], device='cuda:1'), covar=tensor([0.0618, 0.0448, 0.0318, 0.0580], device='cuda:1'), in_proj_covar=tensor([0.1089, 0.0767, 0.0849, 0.0924], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 09:08:26,306 INFO [train.py:968] (1/2) Epoch 2, batch 35050, giga_loss[loss=0.3035, simple_loss=0.3637, pruned_loss=0.1217, over 28707.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3945, pruned_loss=0.1395, over 5670144.34 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.3994, pruned_loss=0.1624, over 5683787.56 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3925, pruned_loss=0.1356, over 5673335.58 frames. ], batch size: 242, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:09:10,204 INFO [train.py:968] (1/2) Epoch 2, batch 35100, giga_loss[loss=0.3833, simple_loss=0.4045, pruned_loss=0.1811, over 27555.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3888, pruned_loss=0.1374, over 5681314.00 frames. ], libri_tot_loss[loss=0.3619, simple_loss=0.3994, pruned_loss=0.1622, over 5691720.16 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3868, pruned_loss=0.1332, over 5676114.42 frames. ], batch size: 472, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:09:19,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6602, 4.9297, 5.3834, 2.2529], device='cuda:1'), covar=tensor([0.0347, 0.0354, 0.0808, 0.1747], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0564, 0.0785, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 09:09:23,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-01 09:09:25,194 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:09:34,956 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 35150, giga_loss[loss=0.2873, simple_loss=0.3498, pruned_loss=0.1124, over 28906.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3798, pruned_loss=0.1323, over 5681215.27 frames. ], libri_tot_loss[loss=0.3622, simple_loss=0.3998, pruned_loss=0.1624, over 5693746.87 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3778, pruned_loss=0.1287, over 5675408.57 frames. ], batch size: 186, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:10:30,426 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 35200, giga_loss[loss=0.2687, simple_loss=0.3258, pruned_loss=0.1058, over 28606.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3718, pruned_loss=0.1285, over 5685761.37 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4003, pruned_loss=0.1626, over 5695074.44 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3691, pruned_loss=0.1246, over 5679649.80 frames. ], batch size: 92, lr: 1.22e-02, grad_scale: 8.0
+2023-03-01 09:10:56,155 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=80887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:11:03,549 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 35250, giga_loss[loss=0.2413, simple_loss=0.3064, pruned_loss=0.08812, over 28396.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3637, pruned_loss=0.1243, over 5684543.16 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4005, pruned_loss=0.1625, over 5697520.95 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3605, pruned_loss=0.1203, over 5677063.76 frames. ], batch size: 60, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:11:36,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0175, 1.6048, 3.2646, 2.9959], device='cuda:1'), covar=tensor([0.1232, 0.1543, 0.0357, 0.0548], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0502, 0.0654, 0.0515], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 09:12:04,729 INFO [train.py:968] (1/2) Epoch 2, batch 35300, giga_loss[loss=0.2619, simple_loss=0.332, pruned_loss=0.09592, over 28917.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3573, pruned_loss=0.1207, over 5692196.67 frames. ], libri_tot_loss[loss=0.3631, simple_loss=0.4008, pruned_loss=0.1627, over 5702133.02 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3533, pruned_loss=0.1162, over 5681658.82 frames. ], batch size: 174, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:12:29,918 INFO [optim.py:369] (1/2) 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 09:12:48,740 INFO [train.py:968] (1/2) Epoch 2, batch 35350, giga_loss[loss=0.3024, simple_loss=0.3534, pruned_loss=0.1257, over 28579.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3546, pruned_loss=0.1196, over 5683295.92 frames. ], libri_tot_loss[loss=0.3634, simple_loss=0.4012, pruned_loss=0.1628, over 5686549.50 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3504, pruned_loss=0.1152, over 5689660.96 frames. ], batch size: 336, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:12:53,693 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 09:13:14,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2177, 1.4564, 1.1688, 1.4228], device='cuda:1'), covar=tensor([0.0916, 0.0480, 0.0449, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0180, 0.0187, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:13:19,546 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 35400, giga_loss[loss=0.3806, simple_loss=0.3946, pruned_loss=0.1833, over 26638.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3494, pruned_loss=0.1162, over 5677990.77 frames. ], libri_tot_loss[loss=0.3637, simple_loss=0.4015, pruned_loss=0.163, over 5683720.75 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3457, pruned_loss=0.1125, over 5685615.44 frames. ], batch size: 555, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:13:59,461 INFO [optim.py:369] (1/2) 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 09:14:19,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4851, 2.1900, 1.5386, 1.3048], device='cuda:1'), covar=tensor([0.0848, 0.0578, 0.0828, 0.1436], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0448, 0.0326, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 09:14:19,853 INFO [train.py:968] (1/2) Epoch 2, batch 35450, giga_loss[loss=0.3055, simple_loss=0.36, pruned_loss=0.1255, over 28865.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3463, pruned_loss=0.1149, over 5673481.72 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4013, pruned_loss=0.1627, over 5687409.85 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3428, pruned_loss=0.1116, over 5676354.89 frames. ], batch size: 112, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:14:30,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:14:59,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:06,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:07,678 INFO [train.py:968] (1/2) Epoch 2, batch 35500, giga_loss[loss=0.2891, simple_loss=0.3364, pruned_loss=0.1209, over 26603.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3428, pruned_loss=0.113, over 5673929.40 frames. ], libri_tot_loss[loss=0.3638, simple_loss=0.4017, pruned_loss=0.1629, over 5687632.35 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3393, pruned_loss=0.1097, over 5675761.00 frames. ], batch size: 555, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:15:27,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:30,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:15:30,467 INFO [optim.py:369] (1/2) 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 09:15:48,971 INFO [train.py:968] (1/2) Epoch 2, batch 35550, giga_loss[loss=0.3021, simple_loss=0.3433, pruned_loss=0.1304, over 26647.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3415, pruned_loss=0.1124, over 5684833.26 frames. ], libri_tot_loss[loss=0.3645, simple_loss=0.4024, pruned_loss=0.1633, over 5693939.24 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3365, pruned_loss=0.1082, over 5680582.17 frames. ], batch size: 555, lr: 1.22e-02, grad_scale: 4.0
+2023-03-01 09:15:55,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:16:00,588 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:16:25,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:16:31,390 INFO [train.py:968] (1/2) Epoch 2, batch 35600, giga_loss[loss=0.2691, simple_loss=0.3264, pruned_loss=0.106, over 28897.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3384, pruned_loss=0.1109, over 5684687.25 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4028, pruned_loss=0.1636, over 5688072.11 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3325, pruned_loss=0.1062, over 5685940.83 frames. ], batch size: 213, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:16:57,371 INFO [optim.py:369] (1/2) 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 09:17:02,775 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:17:04,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:17:15,699 INFO [train.py:968] (1/2) Epoch 2, batch 35650, giga_loss[loss=0.2364, simple_loss=0.3061, pruned_loss=0.08337, over 28821.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3365, pruned_loss=0.1098, over 5676690.30 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4036, pruned_loss=0.1639, over 5684739.79 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3293, pruned_loss=0.1042, over 5680917.90 frames. ], batch size: 199, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:17:22,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7743, 2.2766, 1.9747, 1.8926], device='cuda:1'), covar=tensor([0.1566, 0.1771, 0.1273, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0815, 0.0731, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 09:17:31,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2351, 1.3498, 0.8989, 0.8418], device='cuda:1'), covar=tensor([0.0554, 0.0359, 0.0394, 0.0520], device='cuda:1'), in_proj_covar=tensor([0.1089, 0.0776, 0.0858, 0.0931], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 09:17:31,972 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:17:57,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3690, 1.7367, 1.3049, 1.4402], device='cuda:1'), covar=tensor([0.0976, 0.0457, 0.0441, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0181, 0.0185, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:18:04,548 INFO [train.py:968] (1/2) Epoch 2, batch 35700, giga_loss[loss=0.1961, simple_loss=0.2634, pruned_loss=0.06446, over 28429.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.333, pruned_loss=0.1081, over 5670965.30 frames. ], libri_tot_loss[loss=0.3657, simple_loss=0.4037, pruned_loss=0.1639, over 5685383.20 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3263, pruned_loss=0.1029, over 5673620.59 frames. ], batch size: 60, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:18:08,912 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:11,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:30,750 INFO [optim.py:369] (1/2) 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 09:18:38,607 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:40,450 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:42,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:18:52,122 INFO [train.py:968] (1/2) Epoch 2, batch 35750, giga_loss[loss=0.2697, simple_loss=0.3349, pruned_loss=0.1022, over 28208.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3407, pruned_loss=0.1129, over 5674733.08 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4039, pruned_loss=0.1641, over 5687584.09 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3347, pruned_loss=0.1082, over 5674783.77 frames. ], batch size: 77, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:19:08,702 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:19:25,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6859, 1.4568, 1.1491, 1.2366], device='cuda:1'), covar=tensor([0.0605, 0.0618, 0.0937, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0499, 0.0543, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 09:19:39,022 INFO [train.py:968] (1/2) Epoch 2, batch 35800, giga_loss[loss=0.3349, simple_loss=0.4002, pruned_loss=0.1348, over 28841.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3573, pruned_loss=0.1232, over 5669564.71 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4041, pruned_loss=0.1644, over 5681506.09 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3516, pruned_loss=0.1186, over 5675473.94 frames. ], batch size: 145, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:20:07,518 INFO [optim.py:369] (1/2) 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 09:20:14,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:20:18,666 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 35850, giga_loss[loss=0.3628, simple_loss=0.418, pruned_loss=0.1538, over 28999.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3715, pruned_loss=0.1314, over 5671741.08 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.405, pruned_loss=0.1649, over 5681362.08 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3653, pruned_loss=0.1265, over 5676424.99 frames. ], batch size: 164, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:20:40,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8186, 1.4720, 4.1181, 3.1653], device='cuda:1'), covar=tensor([0.1549, 0.1760, 0.0267, 0.0454], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0488, 0.0641, 0.0508], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 09:20:48,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:21:13,149 INFO [train.py:968] (1/2) Epoch 2, batch 35900, libri_loss[loss=0.4301, simple_loss=0.4591, pruned_loss=0.2005, over 27789.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3814, pruned_loss=0.1365, over 5678518.90 frames. ], libri_tot_loss[loss=0.3682, simple_loss=0.4058, pruned_loss=0.1653, over 5684905.50 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3751, pruned_loss=0.1314, over 5678989.12 frames. ], batch size: 116, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:21:36,286 INFO [optim.py:369] (1/2) 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 09:21:40,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0654, 3.2127, 2.2219, 1.0816], device='cuda:1'), covar=tensor([0.1916, 0.0710, 0.1390, 0.2018], device='cuda:1'), in_proj_covar=tensor([0.1183, 0.1136, 0.1203, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 09:21:53,102 INFO [train.py:968] (1/2) Epoch 2, batch 35950, giga_loss[loss=0.328, simple_loss=0.3972, pruned_loss=0.1294, over 28694.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.385, pruned_loss=0.137, over 5690323.15 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4065, pruned_loss=0.1657, over 5693632.20 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3785, pruned_loss=0.1316, over 5682632.32 frames. ], batch size: 262, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:22:19,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:22:21,914 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:22:32,882 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-01 09:22:39,056 INFO [train.py:968] (1/2) Epoch 2, batch 36000, giga_loss[loss=0.2796, simple_loss=0.3514, pruned_loss=0.1039, over 28456.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3855, pruned_loss=0.1358, over 5673506.68 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4062, pruned_loss=0.1655, over 5692759.06 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3799, pruned_loss=0.1308, over 5667595.95 frames. ], batch size: 65, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:22:39,057 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 09:22:43,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4285, 1.4966, 1.4206, 1.3869], device='cuda:1'), covar=tensor([0.0931, 0.1358, 0.1420, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0815, 0.0641, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 09:22:47,849 INFO [train.py:1012] (1/2) Epoch 2, validation: loss=0.2604, simple_loss=0.3566, pruned_loss=0.08209, over 944034.00 frames. 
+2023-03-01 09:22:47,850 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 09:22:57,749 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:05,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:08,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:12,778 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=81692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:23:14,063 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:23:31,229 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 36050, giga_loss[loss=0.3471, simple_loss=0.407, pruned_loss=0.1436, over 28398.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3876, pruned_loss=0.1372, over 5680276.93 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4073, pruned_loss=0.1662, over 5698610.73 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3815, pruned_loss=0.1316, over 5669429.18 frames. ], batch size: 368, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:24:13,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7867, 1.4334, 3.9285, 3.0398], device='cuda:1'), covar=tensor([0.1471, 0.1749, 0.0309, 0.0477], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0484, 0.0638, 0.0507], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 09:24:16,331 INFO [train.py:968] (1/2) Epoch 2, batch 36100, giga_loss[loss=0.3417, simple_loss=0.3875, pruned_loss=0.1479, over 28853.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3904, pruned_loss=0.1393, over 5694818.66 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4075, pruned_loss=0.1664, over 5703305.56 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3849, pruned_loss=0.134, over 5681724.19 frames. ], batch size: 112, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:24:41,960 INFO [optim.py:369] (1/2) 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 09:24:50,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3021, 1.6971, 1.1783, 1.4205], device='cuda:1'), covar=tensor([0.0910, 0.0460, 0.0424, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0271, 0.0179, 0.0181, 0.0326], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:24:59,704 INFO [train.py:968] (1/2) Epoch 2, batch 36150, giga_loss[loss=0.3396, simple_loss=0.3931, pruned_loss=0.143, over 28830.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3932, pruned_loss=0.1418, over 5688565.68 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4072, pruned_loss=0.1661, over 5706868.70 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3888, pruned_loss=0.1374, over 5675068.12 frames. ], batch size: 99, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:25:38,523 INFO [train.py:968] (1/2) Epoch 2, batch 36200, giga_loss[loss=0.3663, simple_loss=0.4188, pruned_loss=0.1569, over 28998.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3974, pruned_loss=0.1435, over 5703462.55 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4085, pruned_loss=0.1669, over 5709092.63 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3924, pruned_loss=0.1386, over 5690442.94 frames. ], batch size: 136, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:25:51,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:26:05,384 INFO [optim.py:369] (1/2) 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 09:26:23,145 INFO [train.py:968] (1/2) Epoch 2, batch 36250, giga_loss[loss=0.332, simple_loss=0.4033, pruned_loss=0.1303, over 29029.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.4, pruned_loss=0.1442, over 5697173.21 frames. ], libri_tot_loss[loss=0.3716, simple_loss=0.4089, pruned_loss=0.1671, over 5705886.21 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3955, pruned_loss=0.1397, over 5689348.23 frames. ], batch size: 164, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:26:43,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3798, 1.2664, 1.2040, 1.4069], device='cuda:1'), covar=tensor([0.1926, 0.1883, 0.1584, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.0991, 0.0809, 0.0898, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 09:27:07,420 INFO [train.py:968] (1/2) Epoch 2, batch 36300, giga_loss[loss=0.3278, simple_loss=0.4009, pruned_loss=0.1274, over 29044.00 frames. ], tot_loss[loss=0.3444, simple_loss=0.4013, pruned_loss=0.1437, over 5700983.99 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4093, pruned_loss=0.1673, over 5707071.72 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3974, pruned_loss=0.1397, over 5693716.34 frames. ], batch size: 155, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:27:32,881 INFO [optim.py:369] (1/2) 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 09:27:38,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7722, 1.9653, 1.7059, 1.7099], device='cuda:1'), covar=tensor([0.1304, 0.1448, 0.1104, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0789, 0.0714, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 09:27:46,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2591, 4.0771, 5.0174, 2.2701], device='cuda:1'), covar=tensor([0.0298, 0.0351, 0.0569, 0.1708], device='cuda:1'), in_proj_covar=tensor([0.0728, 0.0529, 0.0757, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 09:27:50,745 INFO [train.py:968] (1/2) Epoch 2, batch 36350, giga_loss[loss=0.3187, simple_loss=0.3908, pruned_loss=0.1233, over 28403.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.4002, pruned_loss=0.1416, over 5695102.33 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4096, pruned_loss=0.1675, over 5704708.50 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3967, pruned_loss=0.138, over 5691087.92 frames. ], batch size: 65, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:27:58,213 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:00,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:25,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:29,304 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:32,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:28:33,412 INFO [train.py:968] (1/2) Epoch 2, batch 36400, giga_loss[loss=0.288, simple_loss=0.3678, pruned_loss=0.1041, over 28766.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3969, pruned_loss=0.1377, over 5690647.55 frames. ], libri_tot_loss[loss=0.3728, simple_loss=0.41, pruned_loss=0.1678, over 5695574.21 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3937, pruned_loss=0.1344, over 5696094.33 frames. ], batch size: 242, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:28:33,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:29:01,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5571, 1.4062, 1.3078, 1.2866], device='cuda:1'), covar=tensor([0.0564, 0.0509, 0.0926, 0.0661], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0498, 0.0540, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 09:29:01,648 INFO [optim.py:369] (1/2) 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 09:29:16,383 INFO [train.py:968] (1/2) Epoch 2, batch 36450, giga_loss[loss=0.3832, simple_loss=0.4276, pruned_loss=0.1694, over 28801.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3951, pruned_loss=0.1365, over 5694738.67 frames. ], libri_tot_loss[loss=0.3735, simple_loss=0.4106, pruned_loss=0.1682, over 5699780.64 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3915, pruned_loss=0.1325, over 5695117.88 frames. ], batch size: 284, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:30:01,691 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8992, 1.4654, 3.0728, 2.8834], device='cuda:1'), covar=tensor([0.1610, 0.1890, 0.0739, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0490, 0.0637, 0.0509], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 09:30:03,816 INFO [train.py:968] (1/2) Epoch 2, batch 36500, giga_loss[loss=0.3076, simple_loss=0.3853, pruned_loss=0.1149, over 28519.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.398, pruned_loss=0.1403, over 5680320.65 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4119, pruned_loss=0.169, over 5694462.44 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3937, pruned_loss=0.1358, over 5684861.98 frames. ], batch size: 71, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:30:31,936 INFO [optim.py:369] (1/2) 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 09:30:39,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:42,345 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:44,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:45,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:30:49,051 INFO [train.py:968] (1/2) Epoch 2, batch 36550, giga_loss[loss=0.423, simple_loss=0.4463, pruned_loss=0.1999, over 28967.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.4015, pruned_loss=0.1461, over 5691259.90 frames. ], libri_tot_loss[loss=0.3753, simple_loss=0.4122, pruned_loss=0.1691, over 5702406.62 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3974, pruned_loss=0.1415, over 5687354.77 frames. ], batch size: 213, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:30:51,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8067, 1.3969, 3.9833, 3.2187], device='cuda:1'), covar=tensor([0.1546, 0.1852, 0.0332, 0.0483], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0493, 0.0643, 0.0514], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 09:31:07,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:31:09,296 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 36600, giga_loss[loss=0.3018, simple_loss=0.36, pruned_loss=0.1218, over 28482.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4033, pruned_loss=0.1492, over 5690761.58 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.413, pruned_loss=0.1695, over 5701759.24 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3991, pruned_loss=0.1448, over 5687833.45 frames. ], batch size: 71, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:32:00,411 INFO [optim.py:369] (1/2) 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 09:32:16,274 INFO [train.py:968] (1/2) Epoch 2, batch 36650, giga_loss[loss=0.3143, simple_loss=0.3739, pruned_loss=0.1273, over 28815.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4017, pruned_loss=0.1491, over 5690203.67 frames. ], libri_tot_loss[loss=0.3763, simple_loss=0.4133, pruned_loss=0.1697, over 5704648.27 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.398, pruned_loss=0.1451, over 5685153.83 frames. ], batch size: 119, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:32:57,737 INFO [train.py:968] (1/2) Epoch 2, batch 36700, giga_loss[loss=0.2952, simple_loss=0.3572, pruned_loss=0.1166, over 28707.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4004, pruned_loss=0.1488, over 5698811.19 frames. ], libri_tot_loss[loss=0.3764, simple_loss=0.4133, pruned_loss=0.1697, over 5706615.25 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3972, pruned_loss=0.1452, over 5692928.30 frames. ], batch size: 92, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:33:04,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9953, 2.3893, 2.6122, 2.1340], device='cuda:1'), covar=tensor([0.0762, 0.1712, 0.1031, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0805, 0.0632, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 09:33:08,925 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:33:17,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-01 09:33:28,586 INFO [optim.py:369] (1/2) 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 09:33:43,387 INFO [train.py:968] (1/2) Epoch 2, batch 36750, giga_loss[loss=0.3009, simple_loss=0.3721, pruned_loss=0.1148, over 29034.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3984, pruned_loss=0.1466, over 5701356.58 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4133, pruned_loss=0.1696, over 5707266.94 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3956, pruned_loss=0.1435, over 5695930.97 frames. ], batch size: 155, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:34:02,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:34:11,852 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:34:15,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-01 09:34:29,569 INFO [train.py:968] (1/2) Epoch 2, batch 36800, giga_loss[loss=0.3375, simple_loss=0.3938, pruned_loss=0.1406, over 28746.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3959, pruned_loss=0.1435, over 5698111.38 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4134, pruned_loss=0.1695, over 5706531.86 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3931, pruned_loss=0.1404, over 5694301.49 frames. ], batch size: 262, lr: 1.21e-02, grad_scale: 8.0
+2023-03-01 09:34:31,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1392, 1.1931, 1.0893, 1.1642], device='cuda:1'), covar=tensor([0.2048, 0.2013, 0.1773, 0.1892], device='cuda:1'), in_proj_covar=tensor([0.0985, 0.0807, 0.0876, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 09:34:34,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8723, 2.0705, 2.0306, 1.9934], device='cuda:1'), covar=tensor([0.1376, 0.1504, 0.0962, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0805, 0.0711, 0.0611], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 09:34:42,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-01 09:35:01,533 INFO [optim.py:369] (1/2) 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 09:35:16,181 INFO [train.py:968] (1/2) Epoch 2, batch 36850, giga_loss[loss=0.3064, simple_loss=0.3714, pruned_loss=0.1207, over 28004.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3912, pruned_loss=0.1401, over 5694245.27 frames. ], libri_tot_loss[loss=0.3773, simple_loss=0.4143, pruned_loss=0.1702, over 5709086.03 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3878, pruned_loss=0.1367, over 5688713.84 frames. ], batch size: 412, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:36:02,843 INFO [train.py:968] (1/2) Epoch 2, batch 36900, giga_loss[loss=0.2804, simple_loss=0.3452, pruned_loss=0.1078, over 28966.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3836, pruned_loss=0.1349, over 5695508.08 frames. ], libri_tot_loss[loss=0.3777, simple_loss=0.4147, pruned_loss=0.1704, over 5701954.35 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3801, pruned_loss=0.1314, over 5698184.57 frames. ], batch size: 128, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:36:16,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:36:20,092 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:36:50,476 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 36950, giga_loss[loss=0.2518, simple_loss=0.3015, pruned_loss=0.101, over 23326.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3766, pruned_loss=0.131, over 5675195.27 frames. ], libri_tot_loss[loss=0.3782, simple_loss=0.4151, pruned_loss=0.1707, over 5704349.95 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3728, pruned_loss=0.1273, over 5674838.18 frames. ], batch size: 705, lr: 1.21e-02, grad_scale: 4.0
+2023-03-01 09:37:26,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-01 09:37:35,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 09:37:43,806 INFO [train.py:968] (1/2) Epoch 2, batch 37000, giga_loss[loss=0.3071, simple_loss=0.3628, pruned_loss=0.1257, over 28699.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3743, pruned_loss=0.1288, over 5679779.59 frames. ], libri_tot_loss[loss=0.3785, simple_loss=0.4155, pruned_loss=0.1708, over 5706311.62 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3707, pruned_loss=0.1255, over 5677680.77 frames. ], batch size: 60, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:37:44,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0705, 2.0908, 2.4119, 2.0531], device='cuda:1'), covar=tensor([0.0641, 0.1507, 0.0931, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0811, 0.0639, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 09:37:48,614 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-01 09:38:13,362 INFO [optim.py:369] (1/2) 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 09:38:27,864 INFO [train.py:968] (1/2) Epoch 2, batch 37050, giga_loss[loss=0.2991, simple_loss=0.3543, pruned_loss=0.1219, over 28688.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3765, pruned_loss=0.1308, over 5675719.26 frames. ], libri_tot_loss[loss=0.3797, simple_loss=0.4164, pruned_loss=0.1715, over 5701198.48 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3718, pruned_loss=0.1264, over 5678737.87 frames. ], batch size: 92, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:38:44,349 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:38:53,877 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 09:38:57,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:39:10,526 INFO [train.py:968] (1/2) Epoch 2, batch 37100, giga_loss[loss=0.4195, simple_loss=0.4323, pruned_loss=0.2034, over 24202.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3763, pruned_loss=0.131, over 5685231.93 frames. ], libri_tot_loss[loss=0.3805, simple_loss=0.4169, pruned_loss=0.172, over 5703409.61 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3715, pruned_loss=0.1266, over 5685463.48 frames. ], batch size: 705, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:39:36,583 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 37150, giga_loss[loss=0.3777, simple_loss=0.4164, pruned_loss=0.1695, over 27997.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3745, pruned_loss=0.1303, over 5681690.35 frames. ], libri_tot_loss[loss=0.3815, simple_loss=0.4178, pruned_loss=0.1726, over 5698986.25 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3691, pruned_loss=0.1254, over 5684986.04 frames. ], batch size: 412, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:39:55,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:40:28,959 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 09:40:30,371 INFO [train.py:968] (1/2) Epoch 2, batch 37200, giga_loss[loss=0.3255, simple_loss=0.3816, pruned_loss=0.1347, over 28006.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.372, pruned_loss=0.1287, over 5695878.38 frames. ], libri_tot_loss[loss=0.3818, simple_loss=0.4183, pruned_loss=0.1726, over 5702565.18 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3661, pruned_loss=0.1237, over 5694993.83 frames. ], batch size: 412, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:40:36,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2090, 1.2357, 0.9780, 1.2894], device='cuda:1'), covar=tensor([0.1025, 0.0414, 0.0458, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0178, 0.0180, 0.0323], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:40:42,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2322, 3.7511, 3.8941, 1.5889], device='cuda:1'), covar=tensor([0.0478, 0.0340, 0.0668, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.0746, 0.0541, 0.0770, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 09:40:45,366 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82886.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 09:40:54,238 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:40:57,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:40:57,520 INFO [optim.py:369] (1/2) 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 09:41:10,430 INFO [train.py:968] (1/2) Epoch 2, batch 37250, giga_loss[loss=0.2759, simple_loss=0.3381, pruned_loss=0.1068, over 29014.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3693, pruned_loss=0.1271, over 5705285.73 frames. ], libri_tot_loss[loss=0.3824, simple_loss=0.4189, pruned_loss=0.1729, over 5704403.77 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3634, pruned_loss=0.1223, over 5702950.63 frames. ], batch size: 213, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:41:18,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:41:41,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0943, 1.6839, 1.5004, 1.5693], device='cuda:1'), covar=tensor([0.0500, 0.0592, 0.0830, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0491, 0.0532, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 09:41:43,282 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=82962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:41:43,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7262, 3.2604, 3.4425, 1.6125], device='cuda:1'), covar=tensor([0.0640, 0.0462, 0.0862, 0.2026], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0545, 0.0777, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 09:41:46,792 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:41:48,591 INFO [train.py:968] (1/2) Epoch 2, batch 37300, giga_loss[loss=0.2878, simple_loss=0.35, pruned_loss=0.1127, over 28975.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3683, pruned_loss=0.1263, over 5715848.63 frames. ], libri_tot_loss[loss=0.3832, simple_loss=0.4199, pruned_loss=0.1732, over 5711269.14 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3607, pruned_loss=0.1203, over 5707988.06 frames. ], batch size: 164, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:41:48,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:42:12,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:42:30,194 INFO [train.py:968] (1/2) Epoch 2, batch 37350, libri_loss[loss=0.41, simple_loss=0.4539, pruned_loss=0.183, over 27841.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3663, pruned_loss=0.1253, over 5712661.74 frames. ], libri_tot_loss[loss=0.3837, simple_loss=0.4206, pruned_loss=0.1734, over 5713279.78 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3584, pruned_loss=0.1192, over 5704748.20 frames. ], batch size: 116, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:43:07,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3783, 1.7592, 1.5834, 1.5859], device='cuda:1'), covar=tensor([0.1342, 0.1739, 0.1124, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0807, 0.0727, 0.0618], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 09:43:10,868 INFO [train.py:968] (1/2) Epoch 2, batch 37400, giga_loss[loss=0.2266, simple_loss=0.2964, pruned_loss=0.07845, over 28551.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3617, pruned_loss=0.1223, over 5716959.37 frames. ], libri_tot_loss[loss=0.3842, simple_loss=0.4209, pruned_loss=0.1738, over 5715466.40 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3548, pruned_loss=0.1168, over 5708858.20 frames. ], batch size: 71, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:43:20,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5932, 1.3303, 2.9409, 2.5840], device='cuda:1'), covar=tensor([0.1421, 0.1557, 0.0429, 0.0541], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0493, 0.0645, 0.0520], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 09:43:39,056 INFO [optim.py:369] (1/2) 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 09:43:47,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:43:47,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8587, 1.5486, 1.1996, 1.3236], device='cuda:1'), covar=tensor([0.0724, 0.0814, 0.1003, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0490, 0.0533, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 09:43:50,260 INFO [train.py:968] (1/2) Epoch 2, batch 37450, giga_loss[loss=0.2546, simple_loss=0.3229, pruned_loss=0.09314, over 28949.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3584, pruned_loss=0.12, over 5723854.28 frames. ], libri_tot_loss[loss=0.3849, simple_loss=0.4216, pruned_loss=0.1741, over 5717700.88 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3517, pruned_loss=0.115, over 5715582.30 frames. ], batch size: 186, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:44:14,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1313, 1.1875, 1.2342, 1.0863], device='cuda:1'), covar=tensor([0.0906, 0.1069, 0.1382, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0797, 0.0636, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 09:44:37,078 INFO [train.py:968] (1/2) Epoch 2, batch 37500, libri_loss[loss=0.3776, simple_loss=0.4204, pruned_loss=0.1674, over 29540.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3587, pruned_loss=0.1205, over 5733533.61 frames. ], libri_tot_loss[loss=0.3857, simple_loss=0.4225, pruned_loss=0.1744, over 5724312.28 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3505, pruned_loss=0.1143, over 5721032.96 frames. ], batch size: 78, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:44:40,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1074, 3.5884, 1.9388, 1.6369], device='cuda:1'), covar=tensor([0.0853, 0.0374, 0.0861, 0.1435], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0438, 0.0322, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 09:44:59,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2146, 4.4125, 4.8673, 2.0801], device='cuda:1'), covar=tensor([0.0351, 0.0308, 0.0641, 0.1938], device='cuda:1'), in_proj_covar=tensor([0.0759, 0.0544, 0.0776, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 09:45:03,317 INFO [optim.py:369] (1/2) 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 09:45:17,844 INFO [train.py:968] (1/2) Epoch 2, batch 37550, giga_loss[loss=0.3108, simple_loss=0.3671, pruned_loss=0.1272, over 28996.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3607, pruned_loss=0.1222, over 5721854.34 frames. ], libri_tot_loss[loss=0.3872, simple_loss=0.4239, pruned_loss=0.1752, over 5716642.23 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3509, pruned_loss=0.1149, over 5718487.66 frames. ], batch size: 128, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:45:25,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4054, 1.6458, 1.1625, 1.4184], device='cuda:1'), covar=tensor([0.1005, 0.0388, 0.0476, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0264, 0.0176, 0.0180, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0034, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:45:38,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2002, 1.2218, 1.2064, 1.1734], device='cuda:1'), covar=tensor([0.0818, 0.0938, 0.1285, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0791, 0.0628, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 09:45:52,658 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:45:55,560 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:45:56,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83261.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 09:46:02,967 INFO [train.py:968] (1/2) Epoch 2, batch 37600, giga_loss[loss=0.3683, simple_loss=0.4029, pruned_loss=0.1668, over 23769.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3642, pruned_loss=0.1245, over 5714934.21 frames. ], libri_tot_loss[loss=0.3872, simple_loss=0.4239, pruned_loss=0.1752, over 5718413.87 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3558, pruned_loss=0.1182, over 5710644.51 frames. ], batch size: 705, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:46:20,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:46:33,596 INFO [optim.py:369] (1/2) 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 09:46:48,140 INFO [train.py:968] (1/2) Epoch 2, batch 37650, libri_loss[loss=0.3703, simple_loss=0.4188, pruned_loss=0.1609, over 29563.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3703, pruned_loss=0.1286, over 5710720.34 frames. ], libri_tot_loss[loss=0.3871, simple_loss=0.4241, pruned_loss=0.1751, over 5715212.70 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3617, pruned_loss=0.1221, over 5710448.58 frames. ], batch size: 76, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:47:06,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:47:26,754 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 37700, giga_loss[loss=0.4205, simple_loss=0.4464, pruned_loss=0.1974, over 27933.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3801, pruned_loss=0.1359, over 5696344.44 frames. ], libri_tot_loss[loss=0.3879, simple_loss=0.4247, pruned_loss=0.1755, over 5712250.73 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3717, pruned_loss=0.1296, over 5698730.71 frames. ], batch size: 412, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:48:09,468 INFO [optim.py:369] (1/2) 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 09:48:09,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83404.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 09:48:13,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83407.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 09:48:27,332 INFO [train.py:968] (1/2) Epoch 2, batch 37750, giga_loss[loss=0.3454, simple_loss=0.4143, pruned_loss=0.1383, over 28589.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3884, pruned_loss=0.1418, over 5684872.18 frames. ], libri_tot_loss[loss=0.3884, simple_loss=0.4253, pruned_loss=0.1758, over 5717175.26 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3799, pruned_loss=0.1353, over 5681779.55 frames. ], batch size: 307, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:48:46,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 37800, giga_loss[loss=0.3177, simple_loss=0.3927, pruned_loss=0.1214, over 28547.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3922, pruned_loss=0.1427, over 5685162.70 frames. ], libri_tot_loss[loss=0.3886, simple_loss=0.4254, pruned_loss=0.1759, over 5718248.38 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3853, pruned_loss=0.1374, over 5681615.36 frames. ], batch size: 65, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:49:27,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:49:31,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:49:50,917 INFO [optim.py:369] (1/2) 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 09:49:58,236 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 37850, giga_loss[loss=0.331, simple_loss=0.3944, pruned_loss=0.1338, over 28632.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3979, pruned_loss=0.1459, over 5678687.93 frames. ], libri_tot_loss[loss=0.3887, simple_loss=0.4256, pruned_loss=0.1759, over 5720947.18 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3918, pruned_loss=0.1412, over 5672978.70 frames. ], batch size: 85, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:50:50,446 INFO [train.py:968] (1/2) Epoch 2, batch 37900, libri_loss[loss=0.3724, simple_loss=0.414, pruned_loss=0.1654, over 29638.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4027, pruned_loss=0.1499, over 5675455.92 frames. ], libri_tot_loss[loss=0.3889, simple_loss=0.4257, pruned_loss=0.1761, over 5723563.22 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3969, pruned_loss=0.1451, over 5666817.70 frames. ], batch size: 88, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:50:52,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3378, 1.9136, 1.2664, 1.4102], device='cuda:1'), covar=tensor([0.0995, 0.0349, 0.0421, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0266, 0.0174, 0.0178, 0.0322], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:51:08,187 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=83589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:51:19,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2869, 4.6247, 4.9427, 2.1679], device='cuda:1'), covar=tensor([0.0357, 0.0280, 0.0638, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0544, 0.0765, 0.0550], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 09:51:21,010 INFO [optim.py:369] (1/2) 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 09:51:22,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1493, 1.3604, 0.9854, 1.3372], device='cuda:1'), covar=tensor([0.1065, 0.0412, 0.0485, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0174, 0.0178, 0.0321], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:51:32,085 INFO [train.py:968] (1/2) Epoch 2, batch 37950, giga_loss[loss=0.3134, simple_loss=0.3801, pruned_loss=0.1233, over 28206.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3935, pruned_loss=0.1427, over 5680769.42 frames. ], libri_tot_loss[loss=0.3889, simple_loss=0.4255, pruned_loss=0.1761, over 5716890.97 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 5678585.44 frames. ], batch size: 77, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:51:50,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8814, 3.1056, 2.0814, 0.5846], device='cuda:1'), covar=tensor([0.2196, 0.0798, 0.1343, 0.2405], device='cuda:1'), in_proj_covar=tensor([0.1192, 0.1131, 0.1226, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 09:51:59,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-01 09:52:16,883 INFO [train.py:968] (1/2) Epoch 2, batch 38000, giga_loss[loss=0.303, simple_loss=0.3875, pruned_loss=0.1093, over 28895.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3922, pruned_loss=0.1406, over 5685178.87 frames. ], libri_tot_loss[loss=0.3889, simple_loss=0.4256, pruned_loss=0.1761, over 5720419.60 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3874, pruned_loss=0.1364, over 5679616.41 frames. ], batch size: 174, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:52:23,740 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=83677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:52:48,734 INFO [optim.py:369] (1/2) 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 09:53:01,262 INFO [train.py:968] (1/2) Epoch 2, batch 38050, giga_loss[loss=0.3356, simple_loss=0.3785, pruned_loss=0.1463, over 23462.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3913, pruned_loss=0.1396, over 5683472.18 frames. ], libri_tot_loss[loss=0.3894, simple_loss=0.4261, pruned_loss=0.1764, over 5723897.98 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3864, pruned_loss=0.1352, over 5675324.70 frames. ], batch size: 705, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:53:12,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:53:45,694 INFO [train.py:968] (1/2) Epoch 2, batch 38100, giga_loss[loss=0.3348, simple_loss=0.3945, pruned_loss=0.1375, over 28761.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3943, pruned_loss=0.1415, over 5696575.06 frames. ], libri_tot_loss[loss=0.3898, simple_loss=0.4261, pruned_loss=0.1767, over 5726785.69 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3899, pruned_loss=0.1373, over 5687183.03 frames. ], batch size: 60, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:54:18,076 INFO [optim.py:369] (1/2) 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 09:54:30,972 INFO [train.py:968] (1/2) Epoch 2, batch 38150, giga_loss[loss=0.3277, simple_loss=0.3948, pruned_loss=0.1303, over 28948.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3978, pruned_loss=0.1442, over 5684294.29 frames. ], libri_tot_loss[loss=0.3903, simple_loss=0.4265, pruned_loss=0.1771, over 5718019.81 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3932, pruned_loss=0.1396, over 5683799.20 frames. ], batch size: 145, lr: 1.20e-02, grad_scale: 8.0
+2023-03-01 09:55:17,367 INFO [train.py:968] (1/2) Epoch 2, batch 38200, giga_loss[loss=0.3558, simple_loss=0.4021, pruned_loss=0.1547, over 28789.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3991, pruned_loss=0.1455, over 5693467.68 frames. ], libri_tot_loss[loss=0.3898, simple_loss=0.4261, pruned_loss=0.1768, over 5725613.96 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3946, pruned_loss=0.1408, over 5684649.27 frames. ], batch size: 99, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:55:19,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-01 09:55:21,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:55:23,880 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:55:52,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:56:03,951 INFO [train.py:968] (1/2) Epoch 2, batch 38250, giga_loss[loss=0.2908, simple_loss=0.3673, pruned_loss=0.1072, over 29071.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3999, pruned_loss=0.146, over 5696980.47 frames. ], libri_tot_loss[loss=0.3897, simple_loss=0.426, pruned_loss=0.1766, over 5724497.95 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.396, pruned_loss=0.1419, over 5690295.67 frames. ], batch size: 128, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:56:43,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:56:48,742 INFO [train.py:968] (1/2) Epoch 2, batch 38300, giga_loss[loss=0.3884, simple_loss=0.4272, pruned_loss=0.1748, over 27593.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4007, pruned_loss=0.1472, over 5695144.94 frames. ], libri_tot_loss[loss=0.3894, simple_loss=0.4257, pruned_loss=0.1765, over 5727309.90 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3973, pruned_loss=0.1434, over 5686622.62 frames. ], batch size: 472, lr: 1.20e-02, grad_scale: 4.0
+2023-03-01 09:57:12,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3013, 1.6342, 1.2688, 1.4038], device='cuda:1'), covar=tensor([0.0959, 0.0359, 0.0413, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0172, 0.0178, 0.0317], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0023, 0.0039], device='cuda:1')
+2023-03-01 09:57:19,856 INFO [optim.py:369] (1/2) 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 09:57:33,820 INFO [train.py:968] (1/2) Epoch 2, batch 38350, giga_loss[loss=0.3133, simple_loss=0.3821, pruned_loss=0.1222, over 28963.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.402, pruned_loss=0.1476, over 5701711.38 frames. ], libri_tot_loss[loss=0.3903, simple_loss=0.4267, pruned_loss=0.177, over 5727725.32 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3979, pruned_loss=0.1435, over 5693689.30 frames. ], batch size: 164, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 09:58:03,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:58:04,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-01 09:58:14,698 INFO [train.py:968] (1/2) Epoch 2, batch 38400, giga_loss[loss=0.3383, simple_loss=0.4003, pruned_loss=0.1382, over 28539.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.4016, pruned_loss=0.1456, over 5708228.98 frames. ], libri_tot_loss[loss=0.391, simple_loss=0.4272, pruned_loss=0.1774, over 5727221.93 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3975, pruned_loss=0.1415, over 5702062.94 frames. ], batch size: 85, lr: 1.19e-02, grad_scale: 8.0
+2023-03-01 09:58:47,646 INFO [optim.py:369] (1/2) 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 09:58:47,913 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:58:49,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:58:56,818 INFO [train.py:968] (1/2) Epoch 2, batch 38450, giga_loss[loss=0.3152, simple_loss=0.3871, pruned_loss=0.1216, over 28991.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.4025, pruned_loss=0.1451, over 5711680.90 frames. ], libri_tot_loss[loss=0.3914, simple_loss=0.4276, pruned_loss=0.1776, over 5732291.60 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3978, pruned_loss=0.1404, over 5701101.18 frames. ], batch size: 155, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 09:59:11,830 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:59:15,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7687, 1.5156, 4.1481, 3.2508], device='cuda:1'), covar=tensor([0.1721, 0.1795, 0.0284, 0.0499], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0487, 0.0659, 0.0526], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 09:59:21,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:59:30,989 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-01 09:59:35,721 INFO [train.py:968] (1/2) Epoch 2, batch 38500, giga_loss[loss=0.3494, simple_loss=0.4042, pruned_loss=0.1472, over 28448.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.4024, pruned_loss=0.1447, over 5715083.77 frames. ], libri_tot_loss[loss=0.3911, simple_loss=0.4275, pruned_loss=0.1774, over 5735715.58 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3978, pruned_loss=0.1399, over 5702929.00 frames. ], batch size: 71, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 09:59:59,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 09:59:59,849 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 10:00:01,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:00:08,670 INFO [optim.py:369] (1/2) 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 10:00:16,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3672, 2.2672, 1.3909, 1.2348], device='cuda:1'), covar=tensor([0.0830, 0.0584, 0.0838, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0438, 0.0320, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 10:00:18,512 INFO [train.py:968] (1/2) Epoch 2, batch 38550, giga_loss[loss=0.3257, simple_loss=0.3872, pruned_loss=0.1321, over 28621.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3996, pruned_loss=0.1433, over 5699477.73 frames. ], libri_tot_loss[loss=0.3919, simple_loss=0.428, pruned_loss=0.1779, over 5721920.91 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3949, pruned_loss=0.1383, over 5701889.90 frames. ], batch size: 78, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:00:26,363 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:00:59,551 INFO [train.py:968] (1/2) Epoch 2, batch 38600, giga_loss[loss=0.3302, simple_loss=0.3932, pruned_loss=0.1336, over 29089.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3975, pruned_loss=0.1419, over 5702919.02 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4279, pruned_loss=0.1778, over 5717778.53 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.393, pruned_loss=0.137, over 5707834.75 frames. ], batch size: 128, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:01:08,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6561, 3.2000, 1.7564, 1.4589], device='cuda:1'), covar=tensor([0.0855, 0.0407, 0.0842, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0440, 0.0320, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 10:01:08,304 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 10:01:29,915 INFO [optim.py:369] (1/2) 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 10:01:38,847 INFO [train.py:968] (1/2) Epoch 2, batch 38650, giga_loss[loss=0.325, simple_loss=0.3827, pruned_loss=0.1336, over 28670.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3959, pruned_loss=0.1407, over 5711897.69 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4273, pruned_loss=0.1771, over 5717762.73 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3916, pruned_loss=0.136, over 5715492.61 frames. ], batch size: 60, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:01:59,845 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:02:24,363 INFO [train.py:968] (1/2) Epoch 2, batch 38700, giga_loss[loss=0.3491, simple_loss=0.4086, pruned_loss=0.1448, over 29023.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3968, pruned_loss=0.1419, over 5711439.77 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4274, pruned_loss=0.1771, over 5720572.34 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3925, pruned_loss=0.1374, over 5711461.51 frames. ], batch size: 128, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:02:27,324 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 10:02:31,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:02:58,876 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 38750, giga_loss[loss=0.3165, simple_loss=0.3734, pruned_loss=0.1298, over 28649.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3967, pruned_loss=0.1416, over 5709300.64 frames. ], libri_tot_loss[loss=0.3906, simple_loss=0.4273, pruned_loss=0.177, over 5721549.48 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3933, pruned_loss=0.138, over 5708511.29 frames. ], batch size: 78, lr: 1.19e-02, grad_scale: 2.0
+2023-03-01 10:03:33,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5574, 1.2207, 3.2917, 2.9760], device='cuda:1'), covar=tensor([0.1596, 0.1785, 0.0342, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0478, 0.0639, 0.0514], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 10:03:48,824 INFO [train.py:968] (1/2) Epoch 2, batch 38800, giga_loss[loss=0.3261, simple_loss=0.3937, pruned_loss=0.1293, over 29059.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3961, pruned_loss=0.1402, over 5696987.06 frames. ], libri_tot_loss[loss=0.3909, simple_loss=0.4274, pruned_loss=0.1771, over 5706126.72 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3927, pruned_loss=0.1365, over 5709603.93 frames. ], batch size: 128, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:04:05,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:04:19,665 INFO [optim.py:369] (1/2) 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 10:04:30,274 INFO [train.py:968] (1/2) Epoch 2, batch 38850, giga_loss[loss=0.3124, simple_loss=0.3738, pruned_loss=0.1256, over 28882.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3946, pruned_loss=0.1385, over 5702983.62 frames. ], libri_tot_loss[loss=0.3909, simple_loss=0.4275, pruned_loss=0.1772, over 5708930.32 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3913, pruned_loss=0.135, over 5710185.19 frames. ], batch size: 199, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:04:37,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:05:08,310 INFO [train.py:968] (1/2) Epoch 2, batch 38900, giga_loss[loss=0.3267, simple_loss=0.3897, pruned_loss=0.1318, over 28638.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3951, pruned_loss=0.1395, over 5696984.03 frames. ], libri_tot_loss[loss=0.3913, simple_loss=0.4278, pruned_loss=0.1774, over 5702669.22 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3911, pruned_loss=0.1351, over 5707736.35 frames. ], batch size: 336, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:05:30,107 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:05:40,179 INFO [optim.py:369] (1/2) 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 10:05:48,955 INFO [train.py:968] (1/2) Epoch 2, batch 38950, giga_loss[loss=0.3171, simple_loss=0.3827, pruned_loss=0.1257, over 28391.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3942, pruned_loss=0.1397, over 5695731.02 frames. ], libri_tot_loss[loss=0.3919, simple_loss=0.4282, pruned_loss=0.1778, over 5696870.46 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3894, pruned_loss=0.1344, over 5709996.59 frames. ], batch size: 368, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:05:58,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:06:32,407 INFO [train.py:968] (1/2) Epoch 2, batch 39000, giga_loss[loss=0.27, simple_loss=0.3417, pruned_loss=0.09914, over 29064.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3901, pruned_loss=0.1372, over 5696976.06 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.428, pruned_loss=0.1776, over 5701949.13 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3856, pruned_loss=0.1322, over 5703559.94 frames. ], batch size: 128, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:06:32,407 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 10:06:41,495 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 10:06:43,080 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:06:44,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:07:08,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:07:11,043 INFO [optim.py:369] (1/2) 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 10:07:18,586 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 39050, giga_loss[loss=0.3231, simple_loss=0.3797, pruned_loss=0.1332, over 28430.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3877, pruned_loss=0.1361, over 5688603.82 frames. ], libri_tot_loss[loss=0.3914, simple_loss=0.4278, pruned_loss=0.1775, over 5692702.23 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3832, pruned_loss=0.1312, over 5702877.64 frames. ], batch size: 65, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:07:22,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 10:07:28,299 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:07:48,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:07:54,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-01 10:07:59,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1020, 1.6789, 4.4804, 3.3575], device='cuda:1'), covar=tensor([0.1486, 0.1671, 0.0261, 0.0593], device='cuda:1'), in_proj_covar=tensor([0.0510, 0.0488, 0.0653, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 10:08:01,999 INFO [train.py:968] (1/2) Epoch 2, batch 39100, giga_loss[loss=0.3351, simple_loss=0.3941, pruned_loss=0.1381, over 28987.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3905, pruned_loss=0.1392, over 5684139.10 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4289, pruned_loss=0.1784, over 5685317.00 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.385, pruned_loss=0.1334, over 5702762.49 frames. ], batch size: 106, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:08:05,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3602, 1.4198, 1.2884, 1.7024], device='cuda:1'), covar=tensor([0.2113, 0.2007, 0.1836, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.0966, 0.0807, 0.0881, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 10:08:31,920 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 39150, giga_loss[loss=0.2733, simple_loss=0.3388, pruned_loss=0.1039, over 28961.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3873, pruned_loss=0.1374, over 5680537.06 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4289, pruned_loss=0.1784, over 5681180.92 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.382, pruned_loss=0.1319, over 5699178.99 frames. ], batch size: 106, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:08:47,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1690, 2.7327, 2.9068, 1.4202], device='cuda:1'), covar=tensor([0.0678, 0.0503, 0.0895, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0558, 0.0779, 0.0556], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 10:09:15,909 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:19,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:20,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:23,320 INFO [train.py:968] (1/2) Epoch 2, batch 39200, libri_loss[loss=0.4465, simple_loss=0.4738, pruned_loss=0.2096, over 29114.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3855, pruned_loss=0.1368, over 5694290.41 frames. ], libri_tot_loss[loss=0.392, simple_loss=0.4282, pruned_loss=0.1779, over 5687411.21 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.38, pruned_loss=0.1311, over 5704097.50 frames. ], batch size: 101, lr: 1.19e-02, grad_scale: 8.0
+2023-03-01 10:09:40,647 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:42,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 10:09:43,533 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:45,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:09:53,184 INFO [optim.py:369] (1/2) 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 10:10:00,960 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=84918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:10:01,418 INFO [train.py:968] (1/2) Epoch 2, batch 39250, giga_loss[loss=0.2625, simple_loss=0.3301, pruned_loss=0.09744, over 28946.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3826, pruned_loss=0.1353, over 5697144.87 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4279, pruned_loss=0.1776, over 5680638.06 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3774, pruned_loss=0.1299, over 5711413.31 frames. ], batch size: 136, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:10:09,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:10:42,525 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 39300, libri_loss[loss=0.3646, simple_loss=0.4037, pruned_loss=0.1627, over 29340.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3816, pruned_loss=0.135, over 5685089.50 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.4279, pruned_loss=0.1777, over 5673585.30 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.376, pruned_loss=0.1294, over 5702611.02 frames. ], batch size: 71, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:11:14,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:17,433 INFO [optim.py:369] (1/2) 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 10:11:17,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:19,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:26,645 INFO [train.py:968] (1/2) Epoch 2, batch 39350, giga_loss[loss=0.3393, simple_loss=0.4074, pruned_loss=0.1356, over 29025.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3807, pruned_loss=0.1341, over 5682975.47 frames. ], libri_tot_loss[loss=0.3919, simple_loss=0.4282, pruned_loss=0.1778, over 5668497.38 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.375, pruned_loss=0.1285, over 5701289.32 frames. ], batch size: 164, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:11:32,633 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:11:45,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:12:13,334 INFO [train.py:968] (1/2) Epoch 2, batch 39400, giga_loss[loss=0.3277, simple_loss=0.3855, pruned_loss=0.135, over 28797.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.384, pruned_loss=0.1361, over 5686843.42 frames. ], libri_tot_loss[loss=0.3916, simple_loss=0.428, pruned_loss=0.1776, over 5673499.32 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3787, pruned_loss=0.1309, over 5697380.28 frames. ], batch size: 119, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:12:42,976 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:12:47,910 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:12:50,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:12:56,753 INFO [train.py:968] (1/2) Epoch 2, batch 39450, giga_loss[loss=0.3177, simple_loss=0.3787, pruned_loss=0.1283, over 29099.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3878, pruned_loss=0.1383, over 5687247.85 frames. ], libri_tot_loss[loss=0.3917, simple_loss=0.4281, pruned_loss=0.1777, over 5679185.48 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3818, pruned_loss=0.1325, over 5690999.02 frames. ], batch size: 128, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:13:17,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:13:23,310 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:13:25,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:13:29,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5570, 3.9624, 4.2553, 1.5049], device='cuda:1'), covar=tensor([0.0561, 0.0601, 0.1144, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0553, 0.0792, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 10:13:41,714 INFO [train.py:968] (1/2) Epoch 2, batch 39500, giga_loss[loss=0.4514, simple_loss=0.4597, pruned_loss=0.2216, over 26544.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3879, pruned_loss=0.1369, over 5693850.06 frames. ], libri_tot_loss[loss=0.3914, simple_loss=0.4279, pruned_loss=0.1774, over 5683608.55 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3826, pruned_loss=0.1319, over 5693115.93 frames. ], batch size: 555, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:13:52,901 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:14:03,807 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:14:15,741 INFO [optim.py:369] (1/2) 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 10:14:22,989 INFO [train.py:968] (1/2) Epoch 2, batch 39550, giga_loss[loss=0.326, simple_loss=0.3899, pruned_loss=0.131, over 28629.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3893, pruned_loss=0.1373, over 5680818.38 frames. ], libri_tot_loss[loss=0.3917, simple_loss=0.4282, pruned_loss=0.1776, over 5672614.38 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.383, pruned_loss=0.1312, over 5691255.98 frames. ], batch size: 336, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:14:31,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2088, 1.7199, 1.2406, 0.4750], device='cuda:1'), covar=tensor([0.1242, 0.0736, 0.1334, 0.1577], device='cuda:1'), in_proj_covar=tensor([0.1250, 0.1167, 0.1227, 0.1064], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 10:14:48,873 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:14:51,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:15:06,371 INFO [train.py:968] (1/2) Epoch 2, batch 39600, giga_loss[loss=0.3439, simple_loss=0.403, pruned_loss=0.1424, over 28491.00 frames. ], tot_loss[loss=0.329, simple_loss=0.387, pruned_loss=0.1354, over 5689049.80 frames. ], libri_tot_loss[loss=0.3913, simple_loss=0.4279, pruned_loss=0.1774, over 5673666.07 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3815, pruned_loss=0.13, over 5696485.94 frames. ], batch size: 336, lr: 1.19e-02, grad_scale: 8.0
+2023-03-01 10:15:14,004 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:15:23,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 10:15:27,325 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:15:51,952 INFO [train.py:968] (1/2) Epoch 2, batch 39650, giga_loss[loss=0.3084, simple_loss=0.3776, pruned_loss=0.1196, over 28695.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3867, pruned_loss=0.1356, over 5685551.09 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4282, pruned_loss=0.1777, over 5664982.86 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3813, pruned_loss=0.1303, over 5699048.87 frames. ], batch size: 262, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:16:10,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0871, 1.7657, 1.0867, 1.0559], device='cuda:1'), covar=tensor([0.1110, 0.0754, 0.1082, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0444, 0.0323, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 10:16:12,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:16:31,556 INFO [train.py:968] (1/2) Epoch 2, batch 39700, giga_loss[loss=0.3042, simple_loss=0.3674, pruned_loss=0.1205, over 28837.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3878, pruned_loss=0.1365, over 5704146.47 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4286, pruned_loss=0.1778, over 5670777.15 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3822, pruned_loss=0.1313, over 5710261.48 frames. ], batch size: 112, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:16:45,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5399, 2.7414, 1.6097, 1.3922], device='cuda:1'), covar=tensor([0.0861, 0.0485, 0.0931, 0.1422], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0443, 0.0324, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 10:17:00,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2839, 4.1874, 2.1501, 1.9644], device='cuda:1'), covar=tensor([0.0756, 0.0384, 0.0841, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0445, 0.0325, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0017, 0.0012, 0.0016], device='cuda:1')
+2023-03-01 10:17:01,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:17:07,753 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 39750, giga_loss[loss=0.3674, simple_loss=0.4105, pruned_loss=0.1621, over 23647.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3909, pruned_loss=0.1385, over 5706574.82 frames. ], libri_tot_loss[loss=0.3921, simple_loss=0.4288, pruned_loss=0.1777, over 5677894.49 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3851, pruned_loss=0.1332, over 5705867.64 frames. ], batch size: 705, lr: 1.19e-02, grad_scale: 4.0
+2023-03-01 10:17:17,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9033, 4.1171, 4.5640, 1.8022], device='cuda:1'), covar=tensor([0.0349, 0.0337, 0.0586, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0547, 0.0779, 0.0550], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 10:17:21,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4864, 1.2273, 1.1644, 1.6201], device='cuda:1'), covar=tensor([0.2016, 0.2168, 0.1874, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.0978, 0.0803, 0.0883, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 10:17:24,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6510, 1.9522, 1.7718, 1.7795], device='cuda:1'), covar=tensor([0.1450, 0.1631, 0.1131, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0792, 0.0702, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0011, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 10:17:31,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:17:33,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:17:59,894 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 39800, giga_loss[loss=0.3073, simple_loss=0.3801, pruned_loss=0.1172, over 29032.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3937, pruned_loss=0.1399, over 5706388.61 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4293, pruned_loss=0.1782, over 5679766.74 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3884, pruned_loss=0.135, over 5704582.70 frames. ], batch size: 155, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:18:20,497 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5898, 2.1860, 1.9402, 1.7482], device='cuda:1'), covar=tensor([0.0844, 0.0282, 0.0329, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0174, 0.0178, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0026, 0.0024, 0.0040], device='cuda:1')
+2023-03-01 10:18:25,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:18:31,410 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:18:31,913 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 39850, giga_loss[loss=0.3329, simple_loss=0.3871, pruned_loss=0.1393, over 28564.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3953, pruned_loss=0.1409, over 5708528.88 frames. ], libri_tot_loss[loss=0.3941, simple_loss=0.4304, pruned_loss=0.1789, over 5678301.27 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.389, pruned_loss=0.1352, over 5709739.10 frames. ], batch size: 78, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:18:59,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:02,002 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 10:19:02,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:19,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:19,782 INFO [train.py:968] (1/2) Epoch 2, batch 39900, giga_loss[loss=0.3414, simple_loss=0.4046, pruned_loss=0.1391, over 28789.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3973, pruned_loss=0.1421, over 5705903.17 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4311, pruned_loss=0.1791, over 5682625.12 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3906, pruned_loss=0.1361, over 5703879.91 frames. ], batch size: 262, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:19:26,687 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:19:53,369 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 39950, giga_loss[loss=0.2835, simple_loss=0.3514, pruned_loss=0.1078, over 28720.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3983, pruned_loss=0.1428, over 5708336.73 frames. ], libri_tot_loss[loss=0.3946, simple_loss=0.431, pruned_loss=0.179, over 5686144.94 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3925, pruned_loss=0.1375, over 5704093.23 frames. ], batch size: 99, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:20:40,516 INFO [train.py:968] (1/2) Epoch 2, batch 40000, giga_loss[loss=0.3116, simple_loss=0.3723, pruned_loss=0.1254, over 28272.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3981, pruned_loss=0.1427, over 5713413.32 frames. ], libri_tot_loss[loss=0.3943, simple_loss=0.4308, pruned_loss=0.1789, over 5690110.04 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3929, pruned_loss=0.1378, over 5706975.01 frames. ], batch size: 77, lr: 1.18e-02, grad_scale: 8.0
+2023-03-01 10:21:12,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 10:21:14,724 INFO [optim.py:369] (1/2) 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 10:21:15,683 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:21:17,567 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 40050, giga_loss[loss=0.3368, simple_loss=0.3936, pruned_loss=0.14, over 28887.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3958, pruned_loss=0.1413, over 5716004.76 frames. ], libri_tot_loss[loss=0.3955, simple_loss=0.4317, pruned_loss=0.1796, over 5690864.05 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3901, pruned_loss=0.136, over 5710707.35 frames. ], batch size: 227, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:21:20,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:21:41,756 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 40100, giga_loss[loss=0.2729, simple_loss=0.3406, pruned_loss=0.1026, over 28573.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.392, pruned_loss=0.1389, over 5699523.89 frames. ], libri_tot_loss[loss=0.3959, simple_loss=0.4321, pruned_loss=0.1798, over 5675355.72 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3861, pruned_loss=0.1335, over 5710489.13 frames. ], batch size: 71, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:22:10,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-01 10:22:39,147 INFO [optim.py:369] (1/2) 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 10:22:44,799 INFO [train.py:968] (1/2) Epoch 2, batch 40150, libri_loss[loss=0.4197, simple_loss=0.4534, pruned_loss=0.193, over 29651.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3896, pruned_loss=0.1374, over 5706331.00 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4326, pruned_loss=0.1801, over 5679945.68 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3835, pruned_loss=0.1319, over 5711482.38 frames. ], batch size: 91, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:22:46,309 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:19,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:22,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:24,421 INFO [train.py:968] (1/2) Epoch 2, batch 40200, giga_loss[loss=0.3813, simple_loss=0.4492, pruned_loss=0.1567, over 28659.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3907, pruned_loss=0.1364, over 5703617.12 frames. ], libri_tot_loss[loss=0.3959, simple_loss=0.4322, pruned_loss=0.1798, over 5671927.78 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3852, pruned_loss=0.1313, over 5715083.31 frames. ], batch size: 336, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:23:30,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:31,557 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:37,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:23:49,446 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:24:05,939 INFO [optim.py:369] (1/2) 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 10:24:12,423 INFO [train.py:968] (1/2) Epoch 2, batch 40250, giga_loss[loss=0.2785, simple_loss=0.3424, pruned_loss=0.1073, over 28602.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3926, pruned_loss=0.1369, over 5700649.05 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4322, pruned_loss=0.1799, over 5676664.58 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3876, pruned_loss=0.132, over 5706178.62 frames. ], batch size: 92, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:24:15,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=85923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:24:52,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5083, 1.4507, 1.2967, 1.8851], device='cuda:1'), covar=tensor([0.1785, 0.1843, 0.1589, 0.1993], device='cuda:1'), in_proj_covar=tensor([0.0974, 0.0787, 0.0871, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 10:24:55,685 INFO [train.py:968] (1/2) Epoch 2, batch 40300, giga_loss[loss=0.3133, simple_loss=0.3782, pruned_loss=0.1242, over 29032.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3919, pruned_loss=0.1369, over 5707686.03 frames. ], libri_tot_loss[loss=0.3962, simple_loss=0.4326, pruned_loss=0.1799, over 5680963.84 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3868, pruned_loss=0.1321, over 5708583.95 frames. ], batch size: 136, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:25:31,030 INFO [optim.py:369] (1/2) 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 10:25:36,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:37,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:37,442 INFO [train.py:968] (1/2) Epoch 2, batch 40350, giga_loss[loss=0.3028, simple_loss=0.3598, pruned_loss=0.1229, over 28893.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3912, pruned_loss=0.1384, over 5710059.31 frames. ], libri_tot_loss[loss=0.396, simple_loss=0.4323, pruned_loss=0.1798, over 5686668.95 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3864, pruned_loss=0.1337, over 5706139.31 frames. ], batch size: 112, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:25:38,842 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:43,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:25:46,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:26:03,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:26:12,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:26:13,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5355, 2.0778, 1.5607, 0.7020], device='cuda:1'), covar=tensor([0.1484, 0.0873, 0.1390, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.1215, 0.1134, 0.1216, 0.1041], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 10:26:21,704 INFO [train.py:968] (1/2) Epoch 2, batch 40400, giga_loss[loss=0.3764, simple_loss=0.4064, pruned_loss=0.1732, over 26782.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3892, pruned_loss=0.1386, over 5698268.52 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4328, pruned_loss=0.1803, over 5678479.78 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3847, pruned_loss=0.1342, over 5702087.46 frames. ], batch size: 555, lr: 1.18e-02, grad_scale: 8.0
+2023-03-01 10:26:54,590 INFO [optim.py:369] (1/2) 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 10:27:00,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7178, 3.2459, 3.3469, 1.6509], device='cuda:1'), covar=tensor([0.0630, 0.0521, 0.0878, 0.2129], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0564, 0.0787, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 10:27:00,873 INFO [train.py:968] (1/2) Epoch 2, batch 40450, giga_loss[loss=0.3366, simple_loss=0.3891, pruned_loss=0.1421, over 28389.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3874, pruned_loss=0.1387, over 5714371.66 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.4326, pruned_loss=0.1803, over 5687745.70 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3824, pruned_loss=0.1336, over 5710119.72 frames. ], batch size: 368, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:27:17,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6457, 2.1109, 1.3293, 1.1479], device='cuda:1'), covar=tensor([0.0852, 0.0546, 0.0606, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.1134, 0.0803, 0.0888, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 10:27:48,451 INFO [train.py:968] (1/2) Epoch 2, batch 40500, giga_loss[loss=0.3283, simple_loss=0.3836, pruned_loss=0.1366, over 28931.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3856, pruned_loss=0.1373, over 5720375.97 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4325, pruned_loss=0.1802, over 5688848.97 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3815, pruned_loss=0.1332, over 5716153.19 frames. ], batch size: 213, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:27:49,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2893, 1.4789, 1.1904, 0.9700], device='cuda:1'), covar=tensor([0.0775, 0.0515, 0.0428, 0.0614], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.0803, 0.0884, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 10:27:59,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1828, 1.6692, 1.3347, 1.4281], device='cuda:1'), covar=tensor([0.0905, 0.0386, 0.0430, 0.1057], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0173, 0.0178, 0.0314], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0027, 0.0024, 0.0040], device='cuda:1')
+2023-03-01 10:28:11,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:28:11,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6281, 1.4640, 1.3890, 1.5209], device='cuda:1'), covar=tensor([0.0790, 0.1506, 0.1339, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0788, 0.0629, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 10:28:22,387 INFO [optim.py:369] (1/2) 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 10:28:28,768 INFO [train.py:968] (1/2) Epoch 2, batch 40550, giga_loss[loss=0.2998, simple_loss=0.3515, pruned_loss=0.124, over 28928.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3823, pruned_loss=0.1355, over 5705200.51 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.4326, pruned_loss=0.1803, over 5672970.79 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.378, pruned_loss=0.1313, over 5715778.36 frames. ], batch size: 106, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:28:38,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2129, 2.8006, 2.9458, 1.4032], device='cuda:1'), covar=tensor([0.0689, 0.0485, 0.0973, 0.2008], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0558, 0.0792, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 10:28:44,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7553, 1.4824, 1.6738, 1.4695], device='cuda:1'), covar=tensor([0.0858, 0.1704, 0.1237, 0.1447], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0791, 0.0629, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 10:28:57,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:29:12,223 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:29:12,628 INFO [train.py:968] (1/2) Epoch 2, batch 40600, giga_loss[loss=0.2392, simple_loss=0.316, pruned_loss=0.08117, over 29114.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3767, pruned_loss=0.132, over 5711225.55 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4325, pruned_loss=0.1802, over 5677499.31 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3725, pruned_loss=0.1279, over 5716285.17 frames. ], batch size: 155, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:29:17,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1494, 1.2446, 0.9668, 1.0024], device='cuda:1'), covar=tensor([0.0674, 0.0523, 0.1117, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0499, 0.0533, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 10:29:41,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:29:49,139 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 40650, giga_loss[loss=0.243, simple_loss=0.3188, pruned_loss=0.08359, over 28883.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3725, pruned_loss=0.1293, over 5709506.12 frames. ], libri_tot_loss[loss=0.3967, simple_loss=0.4327, pruned_loss=0.1803, over 5677718.10 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3682, pruned_loss=0.1253, over 5713698.74 frames. ], batch size: 174, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:29:59,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86323.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:30:11,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:13,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:20,083 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 40700, giga_loss[loss=0.3378, simple_loss=0.385, pruned_loss=0.1453, over 28860.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3762, pruned_loss=0.1318, over 5707308.87 frames. ], libri_tot_loss[loss=0.3966, simple_loss=0.4327, pruned_loss=0.1803, over 5682878.36 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3708, pruned_loss=0.1271, over 5707234.54 frames. ], batch size: 112, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:30:38,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:45,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2751, 4.2543, 2.0619, 2.0663], device='cuda:1'), covar=tensor([0.0807, 0.0388, 0.0878, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0452, 0.0330, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0018, 0.0013, 0.0016], device='cuda:1')
+2023-03-01 10:30:54,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1755, 1.2125, 1.0050, 1.0583], device='cuda:1'), covar=tensor([0.0577, 0.0490, 0.0986, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0497, 0.0531, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 10:30:58,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:30:58,758 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:00,806 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:04,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:14,202 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 40750, giga_loss[loss=0.3013, simple_loss=0.3627, pruned_loss=0.12, over 28630.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3782, pruned_loss=0.132, over 5714063.10 frames. ], libri_tot_loss[loss=0.3958, simple_loss=0.4319, pruned_loss=0.1798, over 5687453.82 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3736, pruned_loss=0.1277, over 5710458.20 frames. ], batch size: 78, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:31:27,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:41,002 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:44,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:46,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:31:59,538 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:32:04,734 INFO [train.py:968] (1/2) Epoch 2, batch 40800, giga_loss[loss=0.3422, simple_loss=0.3976, pruned_loss=0.1434, over 28869.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1358, over 5712707.75 frames. ], libri_tot_loss[loss=0.3963, simple_loss=0.4322, pruned_loss=0.1802, over 5690097.18 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3789, pruned_loss=0.1308, over 5707892.29 frames. ], batch size: 145, lr: 1.18e-02, grad_scale: 8.0
+2023-03-01 10:32:08,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:32:38,485 INFO [optim.py:369] (1/2) 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 10:32:42,058 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-01 10:32:45,015 INFO [train.py:968] (1/2) Epoch 2, batch 40850, giga_loss[loss=0.3307, simple_loss=0.3891, pruned_loss=0.1362, over 28767.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3881, pruned_loss=0.1374, over 5716490.13 frames. ], libri_tot_loss[loss=0.3968, simple_loss=0.4327, pruned_loss=0.1804, over 5689810.55 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3819, pruned_loss=0.1318, over 5713857.77 frames. ], batch size: 119, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:33:01,683 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:33:03,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:33:26,973 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:33:27,387 INFO [train.py:968] (1/2) Epoch 2, batch 40900, giga_loss[loss=0.2827, simple_loss=0.3575, pruned_loss=0.1039, over 28997.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3893, pruned_loss=0.1376, over 5712465.70 frames. ], libri_tot_loss[loss=0.3965, simple_loss=0.4325, pruned_loss=0.1802, over 5685625.11 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3833, pruned_loss=0.1321, over 5713974.79 frames. ], batch size: 136, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:33:38,616 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-01 10:34:06,450 INFO [optim.py:369] (1/2) 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 10:34:14,634 INFO [train.py:968] (1/2) Epoch 2, batch 40950, libri_loss[loss=0.3987, simple_loss=0.4378, pruned_loss=0.1798, over 29672.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3921, pruned_loss=0.1397, over 5705941.35 frames. ], libri_tot_loss[loss=0.3964, simple_loss=0.4325, pruned_loss=0.1801, over 5685211.29 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3865, pruned_loss=0.1346, over 5708057.48 frames. ], batch size: 88, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:34:37,658 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:34:44,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:35:09,431 INFO [train.py:968] (1/2) Epoch 2, batch 41000, giga_loss[loss=0.3601, simple_loss=0.412, pruned_loss=0.1541, over 28592.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.4007, pruned_loss=0.1484, over 5689545.95 frames. ], libri_tot_loss[loss=0.3961, simple_loss=0.4322, pruned_loss=0.1799, over 5688337.34 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.396, pruned_loss=0.1441, over 5688461.05 frames. ], batch size: 336, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:35:40,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86698.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:35:54,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9917, 3.5445, 3.6940, 1.5733], device='cuda:1'), covar=tensor([0.0546, 0.0494, 0.0945, 0.2098], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0569, 0.0794, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 10:35:54,883 INFO [optim.py:369] (1/2) 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 10:35:59,906 INFO [train.py:968] (1/2) Epoch 2, batch 41050, giga_loss[loss=0.3415, simple_loss=0.3985, pruned_loss=0.1423, over 28849.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4066, pruned_loss=0.1533, over 5687729.44 frames. ], libri_tot_loss[loss=0.3951, simple_loss=0.4315, pruned_loss=0.1794, over 5690230.48 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4032, pruned_loss=0.15, over 5685032.90 frames. ], batch size: 186, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:36:01,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1761, 1.9714, 1.1641, 1.2883], device='cuda:1'), covar=tensor([0.0935, 0.0587, 0.0855, 0.1320], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0453, 0.0325, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0018, 0.0013, 0.0016], device='cuda:1')
+2023-03-01 10:36:08,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:36:18,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3040, 1.2919, 1.2297, 1.3854], device='cuda:1'), covar=tensor([0.1867, 0.1831, 0.1599, 0.1724], device='cuda:1'), in_proj_covar=tensor([0.0970, 0.0783, 0.0868, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 10:36:32,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5237, 1.6130, 1.3470, 1.0409], device='cuda:1'), covar=tensor([0.0546, 0.0369, 0.0332, 0.0486], device='cuda:1'), in_proj_covar=tensor([0.1155, 0.0824, 0.0902, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 10:36:42,950 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2622, 1.3163, 1.2332, 1.2465], device='cuda:1'), covar=tensor([0.1432, 0.1356, 0.1185, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.0974, 0.0789, 0.0875, 0.0931], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 10:36:47,580 INFO [train.py:968] (1/2) Epoch 2, batch 41100, giga_loss[loss=0.3808, simple_loss=0.4242, pruned_loss=0.1687, over 28626.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4142, pruned_loss=0.1608, over 5668777.31 frames. ], libri_tot_loss[loss=0.3951, simple_loss=0.4314, pruned_loss=0.1794, over 5686698.37 frames. ], giga_tot_loss[loss=0.3628, simple_loss=0.4108, pruned_loss=0.1574, over 5669975.79 frames. ], batch size: 78, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:36:48,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:03,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:06,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:29,314 INFO [optim.py:369] (1/2) 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 10:37:35,158 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:35,503 INFO [train.py:968] (1/2) Epoch 2, batch 41150, giga_loss[loss=0.4454, simple_loss=0.4671, pruned_loss=0.2119, over 28231.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.4217, pruned_loss=0.1678, over 5672503.82 frames. ], libri_tot_loss[loss=0.3953, simple_loss=0.4316, pruned_loss=0.1795, over 5689736.05 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4187, pruned_loss=0.1648, over 5670487.68 frames. ], batch size: 368, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:37:37,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:47,058 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:49,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:50,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:37:52,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86841.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:37:57,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86844.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:38:19,901 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:38:20,194 INFO [train.py:968] (1/2) Epoch 2, batch 41200, giga_loss[loss=0.475, simple_loss=0.4675, pruned_loss=0.2413, over 23447.00 frames. ], tot_loss[loss=0.3888, simple_loss=0.429, pruned_loss=0.1743, over 5677869.19 frames. ], libri_tot_loss[loss=0.3959, simple_loss=0.4323, pruned_loss=0.1798, over 5700178.35 frames. ], giga_tot_loss[loss=0.3838, simple_loss=0.4255, pruned_loss=0.171, over 5666047.14 frames. ], batch size: 705, lr: 1.18e-02, grad_scale: 4.0
+2023-03-01 10:38:22,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:38:24,370 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:38:57,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:39:07,679 INFO [optim.py:369] (1/2) 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 10:39:09,278 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:39:11,768 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 41250, giga_loss[loss=0.3645, simple_loss=0.4117, pruned_loss=0.1587, over 28977.00 frames. ], tot_loss[loss=0.3933, simple_loss=0.432, pruned_loss=0.1773, over 5659034.55 frames. ], libri_tot_loss[loss=0.3957, simple_loss=0.4321, pruned_loss=0.1796, over 5692257.90 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4293, pruned_loss=0.1748, over 5655780.92 frames. ], batch size: 213, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:39:28,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8173, 1.5512, 1.5447, 1.5506], device='cuda:1'), covar=tensor([0.0708, 0.1347, 0.1086, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0808, 0.0645, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 10:39:47,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:05,154 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:06,716 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-01 10:40:07,143 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:10,593 INFO [train.py:968] (1/2) Epoch 2, batch 41300, giga_loss[loss=0.4246, simple_loss=0.4514, pruned_loss=0.1989, over 28671.00 frames. ], tot_loss[loss=0.3971, simple_loss=0.4337, pruned_loss=0.1803, over 5629915.64 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4312, pruned_loss=0.179, over 5677024.29 frames. ], giga_tot_loss[loss=0.3951, simple_loss=0.4325, pruned_loss=0.1789, over 5640706.50 frames. ], batch size: 284, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:40:15,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=86973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:24,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:28,311 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:43,031 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:40:43,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3550, 2.0751, 1.4563, 1.5727], device='cuda:1'), covar=tensor([0.0780, 0.0273, 0.0339, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0171, 0.0174, 0.0311], device='cuda:1'), out_proj_covar=tensor([0.0035, 0.0027, 0.0024, 0.0040], device='cuda:1')
+2023-03-01 10:41:00,188 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:41:00,694 INFO [optim.py:369] (1/2) 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 10:41:02,856 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 41350, giga_loss[loss=0.4922, simple_loss=0.4922, pruned_loss=0.2461, over 27876.00 frames. ], tot_loss[loss=0.4014, simple_loss=0.4362, pruned_loss=0.1833, over 5616828.84 frames. ], libri_tot_loss[loss=0.3942, simple_loss=0.4309, pruned_loss=0.1788, over 5679969.09 frames. ], giga_tot_loss[loss=0.4002, simple_loss=0.4356, pruned_loss=0.1824, over 5621700.70 frames. ], batch size: 412, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:42:00,967 INFO [train.py:968] (1/2) Epoch 2, batch 41400, giga_loss[loss=0.4682, simple_loss=0.4722, pruned_loss=0.2321, over 27559.00 frames. ], tot_loss[loss=0.4096, simple_loss=0.4416, pruned_loss=0.1888, over 5615473.79 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4312, pruned_loss=0.1788, over 5679295.52 frames. ], giga_tot_loss[loss=0.4088, simple_loss=0.441, pruned_loss=0.1883, over 5618173.06 frames. ], batch size: 472, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:42:49,235 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 41450, giga_loss[loss=0.553, simple_loss=0.5208, pruned_loss=0.2926, over 26645.00 frames. ], tot_loss[loss=0.4121, simple_loss=0.4436, pruned_loss=0.1903, over 5628562.87 frames. ], libri_tot_loss[loss=0.3947, simple_loss=0.4314, pruned_loss=0.1791, over 5683050.45 frames. ], giga_tot_loss[loss=0.4116, simple_loss=0.4433, pruned_loss=0.1899, over 5626051.62 frames. ], batch size: 555, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:43:36,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6486, 1.4774, 1.1738, 1.3586], device='cuda:1'), covar=tensor([0.0491, 0.0489, 0.0727, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0497, 0.0529, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 10:43:39,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:43:41,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:43:47,606 INFO [train.py:968] (1/2) Epoch 2, batch 41500, giga_loss[loss=0.4136, simple_loss=0.4452, pruned_loss=0.191, over 28235.00 frames. ], tot_loss[loss=0.4101, simple_loss=0.4417, pruned_loss=0.1893, over 5641540.51 frames. ], libri_tot_loss[loss=0.3945, simple_loss=0.4311, pruned_loss=0.1789, over 5691866.17 frames. ], giga_tot_loss[loss=0.4107, simple_loss=0.4422, pruned_loss=0.1896, over 5629465.98 frames. ], batch size: 368, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:44:06,334 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:44:25,346 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:44:26,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:44:30,032 INFO [optim.py:369] (1/2) 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 10:44:30,341 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 41550, libri_loss[loss=0.3805, simple_loss=0.4336, pruned_loss=0.1637, over 29662.00 frames. ], tot_loss[loss=0.4066, simple_loss=0.4388, pruned_loss=0.1872, over 5640379.11 frames. ], libri_tot_loss[loss=0.3934, simple_loss=0.4304, pruned_loss=0.1782, over 5700242.87 frames. ], giga_tot_loss[loss=0.4087, simple_loss=0.4403, pruned_loss=0.1886, over 5620420.35 frames. ], batch size: 91, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:44:46,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0163, 3.5981, 3.7556, 1.5866], device='cuda:1'), covar=tensor([0.0564, 0.0492, 0.0914, 0.2162], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0575, 0.0819, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 10:45:24,197 INFO [train.py:968] (1/2) Epoch 2, batch 41600, giga_loss[loss=0.4293, simple_loss=0.4653, pruned_loss=0.1966, over 27997.00 frames. ], tot_loss[loss=0.4043, simple_loss=0.4379, pruned_loss=0.1853, over 5637925.85 frames. ], libri_tot_loss[loss=0.3929, simple_loss=0.4301, pruned_loss=0.1779, over 5695202.21 frames. ], giga_tot_loss[loss=0.4067, simple_loss=0.4396, pruned_loss=0.1869, over 5624599.56 frames. ], batch size: 412, lr: 1.17e-02, grad_scale: 8.0
+2023-03-01 10:45:28,075 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:46:14,157 INFO [optim.py:369] (1/2) 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 10:46:18,910 INFO [train.py:968] (1/2) Epoch 2, batch 41650, giga_loss[loss=0.5065, simple_loss=0.4993, pruned_loss=0.2569, over 27985.00 frames. ], tot_loss[loss=0.407, simple_loss=0.4397, pruned_loss=0.1871, over 5614039.56 frames. ], libri_tot_loss[loss=0.3928, simple_loss=0.4298, pruned_loss=0.1779, over 5689629.27 frames. ], giga_tot_loss[loss=0.4092, simple_loss=0.4414, pruned_loss=0.1885, over 5607001.98 frames. ], batch size: 412, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:46:50,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:46:54,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-01 10:46:54,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:46:56,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-01 10:46:57,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:01,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:04,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:13,855 INFO [train.py:968] (1/2) Epoch 2, batch 41700, giga_loss[loss=0.3898, simple_loss=0.4049, pruned_loss=0.1874, over 23846.00 frames. ], tot_loss[loss=0.4066, simple_loss=0.4392, pruned_loss=0.187, over 5600776.72 frames. ], libri_tot_loss[loss=0.3919, simple_loss=0.4292, pruned_loss=0.1774, over 5694361.45 frames. ], giga_tot_loss[loss=0.4094, simple_loss=0.4414, pruned_loss=0.1887, over 5589470.05 frames. ], batch size: 705, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:47:30,955 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:34,077 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:47:43,339 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87395.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:47:52,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 10:48:00,063 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 41750, giga_loss[loss=0.341, simple_loss=0.4101, pruned_loss=0.1359, over 29066.00 frames. ], tot_loss[loss=0.3991, simple_loss=0.4347, pruned_loss=0.1817, over 5616399.50 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4291, pruned_loss=0.1773, over 5696298.07 frames. ], giga_tot_loss[loss=0.4018, simple_loss=0.4368, pruned_loss=0.1834, over 5603397.34 frames. ], batch size: 128, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:48:57,105 INFO [train.py:968] (1/2) Epoch 2, batch 41800, giga_loss[loss=0.3645, simple_loss=0.4276, pruned_loss=0.1507, over 28945.00 frames. ], tot_loss[loss=0.3948, simple_loss=0.4331, pruned_loss=0.1783, over 5631856.00 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.4291, pruned_loss=0.1773, over 5698472.30 frames. ], giga_tot_loss[loss=0.397, simple_loss=0.4348, pruned_loss=0.1797, over 5618771.44 frames. ], batch size: 227, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:49:22,415 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:49:25,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:49:46,902 INFO [optim.py:369] (1/2) 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 10:49:50,082 INFO [train.py:968] (1/2) Epoch 2, batch 41850, giga_loss[loss=0.414, simple_loss=0.4464, pruned_loss=0.1908, over 28316.00 frames. ], tot_loss[loss=0.3896, simple_loss=0.4291, pruned_loss=0.175, over 5632053.29 frames. ], libri_tot_loss[loss=0.3911, simple_loss=0.4285, pruned_loss=0.1769, over 5702266.99 frames. ], giga_tot_loss[loss=0.3919, simple_loss=0.4309, pruned_loss=0.1765, over 5616903.73 frames. ], batch size: 368, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:49:56,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:50:16,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 10:50:40,788 INFO [train.py:968] (1/2) Epoch 2, batch 41900, giga_loss[loss=0.3641, simple_loss=0.4108, pruned_loss=0.1587, over 28383.00 frames. ], tot_loss[loss=0.3861, simple_loss=0.4267, pruned_loss=0.1728, over 5623483.40 frames. ], libri_tot_loss[loss=0.3906, simple_loss=0.4281, pruned_loss=0.1765, over 5697559.98 frames. ], giga_tot_loss[loss=0.3884, simple_loss=0.4286, pruned_loss=0.1742, over 5612536.30 frames. ], batch size: 71, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:50:53,768 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:51:30,561 INFO [train.py:968] (1/2) Epoch 2, batch 41950, giga_loss[loss=0.3536, simple_loss=0.4121, pruned_loss=0.1476, over 28982.00 frames. ], tot_loss[loss=0.3839, simple_loss=0.4254, pruned_loss=0.1712, over 5645618.62 frames. ], libri_tot_loss[loss=0.3913, simple_loss=0.4287, pruned_loss=0.1769, over 5699570.16 frames. ], giga_tot_loss[loss=0.385, simple_loss=0.4262, pruned_loss=0.1718, over 5633056.03 frames. ], batch size: 164, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:51:34,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:51:56,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1236, 1.2681, 1.1513, 1.0120], device='cuda:1'), covar=tensor([0.1652, 0.1643, 0.1471, 0.1483], device='cuda:1'), in_proj_covar=tensor([0.0991, 0.0814, 0.0901, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 10:51:58,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:52:19,505 INFO [train.py:968] (1/2) Epoch 2, batch 42000, giga_loss[loss=0.3847, simple_loss=0.4287, pruned_loss=0.1703, over 29082.00 frames. ], tot_loss[loss=0.3829, simple_loss=0.4245, pruned_loss=0.1707, over 5652163.91 frames. ], libri_tot_loss[loss=0.3912, simple_loss=0.4286, pruned_loss=0.1769, over 5702506.84 frames. ], giga_tot_loss[loss=0.3837, simple_loss=0.4253, pruned_loss=0.171, over 5638870.22 frames. ], batch size: 128, lr: 1.17e-02, grad_scale: 8.0
+2023-03-01 10:52:19,506 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 10:52:28,084 INFO [train.py:1012] (1/2) Epoch 2, validation: loss=0.2594, simple_loss=0.3583, pruned_loss=0.08022, over 944034.00 frames. 
+2023-03-01 10:52:28,085 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 10:53:11,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3466, 1.3701, 1.2213, 1.3397], device='cuda:1'), covar=tensor([0.1739, 0.1746, 0.1515, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0987, 0.0812, 0.0896, 0.0934], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 10:53:14,286 INFO [optim.py:369] (1/2) 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 10:53:18,129 INFO [train.py:968] (1/2) Epoch 2, batch 42050, libri_loss[loss=0.3951, simple_loss=0.4399, pruned_loss=0.1752, over 29662.00 frames. ], tot_loss[loss=0.381, simple_loss=0.4232, pruned_loss=0.1694, over 5653079.32 frames. ], libri_tot_loss[loss=0.3908, simple_loss=0.4283, pruned_loss=0.1767, over 5708480.03 frames. ], giga_tot_loss[loss=0.3817, simple_loss=0.4239, pruned_loss=0.1697, over 5635282.05 frames. ], batch size: 91, lr: 1.17e-02, grad_scale: 8.0
+2023-03-01 10:53:23,307 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:53:27,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:53:57,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:53:58,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:54:09,536 INFO [train.py:968] (1/2) Epoch 2, batch 42100, giga_loss[loss=0.4026, simple_loss=0.4504, pruned_loss=0.1774, over 27984.00 frames. ], tot_loss[loss=0.376, simple_loss=0.4204, pruned_loss=0.1658, over 5652025.67 frames. ], libri_tot_loss[loss=0.3915, simple_loss=0.4287, pruned_loss=0.1771, over 5710975.58 frames. ], giga_tot_loss[loss=0.3755, simple_loss=0.4204, pruned_loss=0.1653, over 5634325.62 frames. ], batch size: 412, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:54:12,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87770.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:54:35,664 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:54:38,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:55:07,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4935, 2.7663, 1.6081, 1.3244], device='cuda:1'), covar=tensor([0.0833, 0.0477, 0.0836, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0455, 0.0323, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 10:55:08,135 INFO [optim.py:369] (1/2) 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 10:55:10,460 INFO [train.py:968] (1/2) Epoch 2, batch 42150, giga_loss[loss=0.3682, simple_loss=0.4201, pruned_loss=0.1581, over 28580.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.4212, pruned_loss=0.1632, over 5658650.57 frames. ], libri_tot_loss[loss=0.3912, simple_loss=0.4284, pruned_loss=0.177, over 5712068.70 frames. ], giga_tot_loss[loss=0.3736, simple_loss=0.4214, pruned_loss=0.1629, over 5643488.50 frames. ], batch size: 307, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:55:14,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:55:46,179 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:55:57,811 INFO [train.py:968] (1/2) Epoch 2, batch 42200, giga_loss[loss=0.392, simple_loss=0.4329, pruned_loss=0.1755, over 28763.00 frames. ], tot_loss[loss=0.3768, simple_loss=0.4235, pruned_loss=0.165, over 5669606.61 frames. ], libri_tot_loss[loss=0.3905, simple_loss=0.4278, pruned_loss=0.1766, over 5715174.51 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4241, pruned_loss=0.1649, over 5653832.55 frames. ], batch size: 243, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:56:24,103 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:56:39,944 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87913.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:56:42,218 INFO [optim.py:369] (1/2) 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 10:56:43,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 10:56:44,869 INFO [train.py:968] (1/2) Epoch 2, batch 42250, giga_loss[loss=0.3369, simple_loss=0.3934, pruned_loss=0.1402, over 29058.00 frames. ], tot_loss[loss=0.3793, simple_loss=0.4243, pruned_loss=0.1671, over 5652241.17 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4276, pruned_loss=0.1769, over 5698070.62 frames. ], giga_tot_loss[loss=0.3789, simple_loss=0.4248, pruned_loss=0.1665, over 5652464.51 frames. ], batch size: 128, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:57:09,502 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87945.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 10:57:31,786 INFO [train.py:968] (1/2) Epoch 2, batch 42300, giga_loss[loss=0.3462, simple_loss=0.3929, pruned_loss=0.1497, over 28466.00 frames. ], tot_loss[loss=0.3776, simple_loss=0.4227, pruned_loss=0.1663, over 5660564.53 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4273, pruned_loss=0.1765, over 5699256.36 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.4233, pruned_loss=0.1659, over 5658533.36 frames. ], batch size: 85, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:57:39,514 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:57:59,426 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=87998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:57:59,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 10:58:19,367 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 42350, giga_loss[loss=0.3385, simple_loss=0.3944, pruned_loss=0.1413, over 28880.00 frames. ], tot_loss[loss=0.3797, simple_loss=0.4225, pruned_loss=0.1685, over 5654080.37 frames. ], libri_tot_loss[loss=0.3906, simple_loss=0.4276, pruned_loss=0.1768, over 5698521.08 frames. ], giga_tot_loss[loss=0.3793, simple_loss=0.4228, pruned_loss=0.1679, over 5652951.54 frames. ], batch size: 199, lr: 1.17e-02, grad_scale: 2.0
+2023-03-01 10:58:50,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2078, 1.3386, 1.0859, 1.3528], device='cuda:1'), covar=tensor([0.0943, 0.0432, 0.0438, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0262, 0.0174, 0.0176, 0.0313], device='cuda:1'), out_proj_covar=tensor([0.0036, 0.0027, 0.0024, 0.0040], device='cuda:1')
+2023-03-01 10:58:51,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-01 10:59:06,459 INFO [train.py:968] (1/2) Epoch 2, batch 42400, giga_loss[loss=0.4143, simple_loss=0.4449, pruned_loss=0.1918, over 27920.00 frames. ], tot_loss[loss=0.3772, simple_loss=0.4206, pruned_loss=0.1669, over 5658076.73 frames. ], libri_tot_loss[loss=0.3898, simple_loss=0.4271, pruned_loss=0.1762, over 5696108.93 frames. ], giga_tot_loss[loss=0.3771, simple_loss=0.421, pruned_loss=0.1667, over 5657773.97 frames. ], batch size: 412, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 10:59:57,297 INFO [optim.py:369] (1/2) 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 11:00:00,939 INFO [train.py:968] (1/2) Epoch 2, batch 42450, giga_loss[loss=0.3424, simple_loss=0.4053, pruned_loss=0.1397, over 29081.00 frames. ], tot_loss[loss=0.3743, simple_loss=0.4199, pruned_loss=0.1643, over 5665223.56 frames. ], libri_tot_loss[loss=0.3897, simple_loss=0.427, pruned_loss=0.1761, over 5697180.38 frames. ], giga_tot_loss[loss=0.3742, simple_loss=0.4202, pruned_loss=0.1641, over 5663824.61 frames. ], batch size: 155, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:00:10,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:00:21,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:00:26,736 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 42500, libri_loss[loss=0.34, simple_loss=0.3875, pruned_loss=0.1463, over 29344.00 frames. ], tot_loss[loss=0.3739, simple_loss=0.4202, pruned_loss=0.1639, over 5677753.87 frames. ], libri_tot_loss[loss=0.3898, simple_loss=0.4273, pruned_loss=0.1762, over 5700918.77 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.42, pruned_loss=0.1632, over 5672256.99 frames. ], batch size: 71, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:00:50,031 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:01:12,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1311, 3.6693, 3.8592, 1.8172], device='cuda:1'), covar=tensor([0.0502, 0.0449, 0.0780, 0.1760], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0589, 0.0816, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:01:28,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7192, 1.5944, 1.3419, 1.5091], device='cuda:1'), covar=tensor([0.1147, 0.1977, 0.1290, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0810, 0.0722, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 11:01:32,258 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 42550, giga_loss[loss=0.341, simple_loss=0.3988, pruned_loss=0.1416, over 28990.00 frames. ], tot_loss[loss=0.3761, simple_loss=0.4212, pruned_loss=0.1655, over 5671033.52 frames. ], libri_tot_loss[loss=0.3898, simple_loss=0.4271, pruned_loss=0.1763, over 5706044.47 frames. ], giga_tot_loss[loss=0.3751, simple_loss=0.4211, pruned_loss=0.1646, over 5661233.66 frames. ], batch size: 136, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:01:45,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:02:24,006 INFO [train.py:968] (1/2) Epoch 2, batch 42600, giga_loss[loss=0.3082, simple_loss=0.373, pruned_loss=0.1217, over 28959.00 frames. ], tot_loss[loss=0.3748, simple_loss=0.4198, pruned_loss=0.1649, over 5677815.47 frames. ], libri_tot_loss[loss=0.3901, simple_loss=0.4273, pruned_loss=0.1764, over 5706775.25 frames. ], giga_tot_loss[loss=0.3738, simple_loss=0.4195, pruned_loss=0.1641, over 5669401.31 frames. ], batch size: 164, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:02:27,967 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:02:28,057 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:02:32,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:03:05,619 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:03:05,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8866, 1.7189, 1.6028, 1.5402], device='cuda:1'), covar=tensor([0.0843, 0.1619, 0.1327, 0.1359], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0801, 0.0634, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 11:03:12,732 INFO [optim.py:369] (1/2) 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 11:03:16,723 INFO [train.py:968] (1/2) Epoch 2, batch 42650, giga_loss[loss=0.3491, simple_loss=0.4109, pruned_loss=0.1437, over 28909.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4189, pruned_loss=0.1649, over 5668939.83 frames. ], libri_tot_loss[loss=0.3901, simple_loss=0.4274, pruned_loss=0.1764, over 5702460.49 frames. ], giga_tot_loss[loss=0.3731, simple_loss=0.4183, pruned_loss=0.1639, over 5665085.12 frames. ], batch size: 145, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:03:46,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-01 11:03:50,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:03:56,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:07,546 INFO [train.py:968] (1/2) Epoch 2, batch 42700, giga_loss[loss=0.3885, simple_loss=0.4255, pruned_loss=0.1757, over 28970.00 frames. ], tot_loss[loss=0.3728, simple_loss=0.4174, pruned_loss=0.1641, over 5677714.64 frames. ], libri_tot_loss[loss=0.3904, simple_loss=0.4278, pruned_loss=0.1765, over 5706254.22 frames. ], giga_tot_loss[loss=0.3711, simple_loss=0.4163, pruned_loss=0.163, over 5670588.15 frames. ], batch size: 145, lr: 1.17e-02, grad_scale: 4.0
+2023-03-01 11:04:12,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:13,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:14,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:18,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6505, 1.7001, 0.9136, 1.4843], device='cuda:1'), covar=tensor([0.0740, 0.0720, 0.1719, 0.1045], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0499, 0.0530, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 11:04:44,139 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:04:54,890 INFO [optim.py:369] (1/2) 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 11:04:55,201 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 2, batch 42750, giga_loss[loss=0.343, simple_loss=0.393, pruned_loss=0.1465, over 28776.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4165, pruned_loss=0.1643, over 5685649.62 frames. ], libri_tot_loss[loss=0.3895, simple_loss=0.4272, pruned_loss=0.176, over 5711338.36 frames. ], giga_tot_loss[loss=0.3716, simple_loss=0.416, pruned_loss=0.1636, over 5674700.16 frames. ], batch size: 119, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:04:59,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:05:16,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3379, 1.9953, 1.4314, 0.6459], device='cuda:1'), covar=tensor([0.1299, 0.0781, 0.1302, 0.1511], device='cuda:1'), in_proj_covar=tensor([0.1230, 0.1169, 0.1229, 0.1050], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 11:05:28,544 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:05:47,980 INFO [train.py:968] (1/2) Epoch 2, batch 42800, libri_loss[loss=0.4193, simple_loss=0.462, pruned_loss=0.1883, over 29081.00 frames. ], tot_loss[loss=0.3732, simple_loss=0.4164, pruned_loss=0.165, over 5670782.71 frames. ], libri_tot_loss[loss=0.3903, simple_loss=0.4278, pruned_loss=0.1764, over 5703630.91 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4152, pruned_loss=0.1639, over 5667928.42 frames. ], batch size: 101, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:05:57,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4739, 4.0102, 4.2228, 1.9143], device='cuda:1'), covar=tensor([0.0474, 0.0392, 0.0752, 0.1820], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0596, 0.0843, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:06:15,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=88493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:17,833 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:19,976 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:22,072 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:24,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:32,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6339, 1.8247, 1.6796, 1.7343], device='cuda:1'), covar=tensor([0.0862, 0.1135, 0.0824, 0.0570], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0817, 0.0720, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 11:06:35,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=88516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:36,324 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 42850, giga_loss[loss=0.3511, simple_loss=0.4073, pruned_loss=0.1474, over 28690.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4167, pruned_loss=0.166, over 5650052.02 frames. ], libri_tot_loss[loss=0.3904, simple_loss=0.428, pruned_loss=0.1764, over 5698601.65 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4154, pruned_loss=0.1648, over 5651326.82 frames. ], batch size: 262, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:06:40,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=88519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:49,327 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:06:52,594 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:07:06,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=88548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:07:25,135 INFO [train.py:968] (1/2) Epoch 2, batch 42900, giga_loss[loss=0.4225, simple_loss=0.4478, pruned_loss=0.1986, over 27902.00 frames. ], tot_loss[loss=0.3728, simple_loss=0.4164, pruned_loss=0.1646, over 5660335.83 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4278, pruned_loss=0.1763, over 5699763.28 frames. ], giga_tot_loss[loss=0.3711, simple_loss=0.4152, pruned_loss=0.1635, over 5659095.41 frames. ], batch size: 412, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:07:43,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3315, 1.9211, 1.4763, 0.4827], device='cuda:1'), covar=tensor([0.1276, 0.0805, 0.1128, 0.1649], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.1170, 0.1212, 0.1042], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 11:08:14,042 INFO [optim.py:369] (1/2) 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 11:08:15,419 INFO [train.py:968] (1/2) Epoch 2, batch 42950, giga_loss[loss=0.4005, simple_loss=0.4306, pruned_loss=0.1852, over 27629.00 frames. ], tot_loss[loss=0.372, simple_loss=0.4166, pruned_loss=0.1638, over 5665271.78 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4281, pruned_loss=0.1767, over 5698091.49 frames. ], giga_tot_loss[loss=0.37, simple_loss=0.4152, pruned_loss=0.1624, over 5665519.62 frames. ], batch size: 472, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:08:55,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8145, 3.4423, 3.5766, 1.7951], device='cuda:1'), covar=tensor([0.0559, 0.0459, 0.0880, 0.1895], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0596, 0.0842, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:09:04,753 INFO [train.py:968] (1/2) Epoch 2, batch 43000, giga_loss[loss=0.4094, simple_loss=0.447, pruned_loss=0.1859, over 28580.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4167, pruned_loss=0.1631, over 5673860.13 frames. ], libri_tot_loss[loss=0.3903, simple_loss=0.4278, pruned_loss=0.1764, over 5701923.20 frames. ], giga_tot_loss[loss=0.3698, simple_loss=0.4157, pruned_loss=0.162, over 5669865.24 frames. ], batch size: 307, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:09:22,808 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=88684.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:09:57,987 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 43050, giga_loss[loss=0.4883, simple_loss=0.4868, pruned_loss=0.2449, over 27637.00 frames. ], tot_loss[loss=0.3732, simple_loss=0.4178, pruned_loss=0.1643, over 5662516.48 frames. ], libri_tot_loss[loss=0.3905, simple_loss=0.428, pruned_loss=0.1765, over 5684528.94 frames. ], giga_tot_loss[loss=0.3714, simple_loss=0.4166, pruned_loss=0.1631, over 5673526.85 frames. ], batch size: 472, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:10:48,387 INFO [train.py:968] (1/2) Epoch 2, batch 43100, giga_loss[loss=0.4676, simple_loss=0.4693, pruned_loss=0.2329, over 28643.00 frames. ], tot_loss[loss=0.3783, simple_loss=0.4213, pruned_loss=0.1676, over 5665748.50 frames. ], libri_tot_loss[loss=0.3902, simple_loss=0.4278, pruned_loss=0.1763, over 5679679.00 frames. ], giga_tot_loss[loss=0.3766, simple_loss=0.4203, pruned_loss=0.1665, over 5678495.44 frames. ], batch size: 307, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:11:12,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6522, 1.4596, 1.2610, 1.2846], device='cuda:1'), covar=tensor([0.0577, 0.0605, 0.0899, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0504, 0.0538, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 11:11:40,441 INFO [optim.py:369] (1/2) 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 11:11:42,038 INFO [train.py:968] (1/2) Epoch 2, batch 43150, giga_loss[loss=0.381, simple_loss=0.4216, pruned_loss=0.1702, over 28943.00 frames. ], tot_loss[loss=0.3799, simple_loss=0.4215, pruned_loss=0.1691, over 5675171.61 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4276, pruned_loss=0.1762, over 5684159.81 frames. ], giga_tot_loss[loss=0.3785, simple_loss=0.4207, pruned_loss=0.1682, over 5681071.41 frames. ], batch size: 106, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:12:33,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=88868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:12:34,061 INFO [train.py:968] (1/2) Epoch 2, batch 43200, giga_loss[loss=0.348, simple_loss=0.3948, pruned_loss=0.1506, over 28788.00 frames. ], tot_loss[loss=0.3812, simple_loss=0.4218, pruned_loss=0.1703, over 5660516.48 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4276, pruned_loss=0.1762, over 5670975.48 frames. ], giga_tot_loss[loss=0.38, simple_loss=0.4211, pruned_loss=0.1694, over 5676701.46 frames. ], batch size: 99, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:12:57,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4649, 2.3399, 1.4986, 1.4261], device='cuda:1'), covar=tensor([0.0745, 0.0510, 0.0698, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0455, 0.0322, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 11:13:02,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-01 11:13:27,321 INFO [optim.py:369] (1/2) 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 11:13:29,772 INFO [train.py:968] (1/2) Epoch 2, batch 43250, giga_loss[loss=0.4826, simple_loss=0.4798, pruned_loss=0.2427, over 26572.00 frames. ], tot_loss[loss=0.3848, simple_loss=0.4241, pruned_loss=0.1728, over 5645716.59 frames. ], libri_tot_loss[loss=0.3901, simple_loss=0.4277, pruned_loss=0.1762, over 5672276.61 frames. ], giga_tot_loss[loss=0.3837, simple_loss=0.4234, pruned_loss=0.172, over 5657318.19 frames. ], batch size: 555, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:14:13,984 INFO [train.py:968] (1/2) Epoch 2, batch 43300, giga_loss[loss=0.3348, simple_loss=0.3899, pruned_loss=0.1398, over 29011.00 frames. ], tot_loss[loss=0.3842, simple_loss=0.4235, pruned_loss=0.1725, over 5659229.73 frames. ], libri_tot_loss[loss=0.3901, simple_loss=0.4279, pruned_loss=0.1762, over 5679225.20 frames. ], giga_tot_loss[loss=0.3832, simple_loss=0.4227, pruned_loss=0.1718, over 5661556.77 frames. ], batch size: 128, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:14:30,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3777, 1.8065, 1.7610, 1.6289], device='cuda:1'), covar=tensor([0.0576, 0.0827, 0.0872, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0509, 0.0544, 0.0472], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 11:14:35,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4407, 1.3956, 1.0970, 1.2310], device='cuda:1'), covar=tensor([0.0527, 0.0515, 0.0923, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0509, 0.0544, 0.0473], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 11:14:55,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:14:58,527 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:15:02,404 INFO [train.py:968] (1/2) Epoch 2, batch 43350, giga_loss[loss=0.397, simple_loss=0.4402, pruned_loss=0.1769, over 28939.00 frames. ], tot_loss[loss=0.3811, simple_loss=0.4224, pruned_loss=0.1699, over 5662059.37 frames. ], libri_tot_loss[loss=0.3904, simple_loss=0.4281, pruned_loss=0.1763, over 5677481.98 frames. ], giga_tot_loss[loss=0.3798, simple_loss=0.4214, pruned_loss=0.1691, over 5664773.53 frames. ], batch size: 186, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:15:24,660 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:15:38,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2543, 4.8437, 4.9665, 2.0436], device='cuda:1'), covar=tensor([0.0341, 0.0298, 0.0635, 0.1823], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0598, 0.0840, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:15:38,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89059.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:15:48,473 INFO [train.py:968] (1/2) Epoch 2, batch 43400, giga_loss[loss=0.3209, simple_loss=0.3869, pruned_loss=0.1274, over 28876.00 frames. ], tot_loss[loss=0.3758, simple_loss=0.4192, pruned_loss=0.1662, over 5655674.90 frames. ], libri_tot_loss[loss=0.3911, simple_loss=0.4287, pruned_loss=0.1767, over 5674243.38 frames. ], giga_tot_loss[loss=0.3738, simple_loss=0.4176, pruned_loss=0.165, over 5660529.03 frames. ], batch size: 174, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:16:20,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2782, 1.5586, 1.2966, 1.5693], device='cuda:1'), covar=tensor([0.0953, 0.0367, 0.0397, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0172, 0.0175, 0.0312], device='cuda:1'), out_proj_covar=tensor([0.0036, 0.0027, 0.0024, 0.0041], device='cuda:1')
+2023-03-01 11:16:28,984 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 11:16:36,464 INFO [optim.py:369] (1/2) 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 11:16:38,179 INFO [train.py:968] (1/2) Epoch 2, batch 43450, giga_loss[loss=0.3793, simple_loss=0.4216, pruned_loss=0.1685, over 28900.00 frames. ], tot_loss[loss=0.3737, simple_loss=0.4168, pruned_loss=0.1653, over 5660611.04 frames. ], libri_tot_loss[loss=0.3906, simple_loss=0.4283, pruned_loss=0.1764, over 5678099.36 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4158, pruned_loss=0.1644, over 5660681.83 frames. ], batch size: 145, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:17:25,750 INFO [train.py:968] (1/2) Epoch 2, batch 43500, giga_loss[loss=0.3652, simple_loss=0.4059, pruned_loss=0.1623, over 29044.00 frames. ], tot_loss[loss=0.3723, simple_loss=0.4153, pruned_loss=0.1646, over 5670079.22 frames. ], libri_tot_loss[loss=0.3897, simple_loss=0.4277, pruned_loss=0.1758, over 5682476.65 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4147, pruned_loss=0.1642, over 5666257.75 frames. ], batch size: 128, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:17:45,915 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 11:17:55,332 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89202.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:17:57,862 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:18:09,644 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 43550, giga_loss[loss=0.3858, simple_loss=0.4302, pruned_loss=0.1708, over 28870.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4146, pruned_loss=0.1643, over 5679351.56 frames. ], libri_tot_loss[loss=0.3893, simple_loss=0.4274, pruned_loss=0.1756, over 5691674.62 frames. ], giga_tot_loss[loss=0.3707, simple_loss=0.414, pruned_loss=0.1637, over 5667308.13 frames. ], batch size: 186, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:18:15,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-01 11:18:23,902 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89234.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:18:50,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 11:18:59,031 INFO [train.py:968] (1/2) Epoch 2, batch 43600, giga_loss[loss=0.4507, simple_loss=0.4726, pruned_loss=0.2144, over 27925.00 frames. ], tot_loss[loss=0.3763, simple_loss=0.4188, pruned_loss=0.1669, over 5653378.42 frames. ], libri_tot_loss[loss=0.3894, simple_loss=0.4275, pruned_loss=0.1756, over 5674791.63 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.418, pruned_loss=0.1662, over 5659906.49 frames. ], batch size: 412, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:19:45,494 INFO [optim.py:369] (1/2) 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 11:19:45,506 INFO [train.py:968] (1/2) Epoch 2, batch 43650, giga_loss[loss=0.3645, simple_loss=0.4275, pruned_loss=0.1508, over 28994.00 frames. ], tot_loss[loss=0.376, simple_loss=0.4214, pruned_loss=0.1653, over 5665123.70 frames. ], libri_tot_loss[loss=0.3893, simple_loss=0.4275, pruned_loss=0.1756, over 5682735.58 frames. ], giga_tot_loss[loss=0.3747, simple_loss=0.4204, pruned_loss=0.1644, over 5662325.77 frames. ], batch size: 155, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:20:42,171 INFO [train.py:968] (1/2) Epoch 2, batch 43700, giga_loss[loss=0.4204, simple_loss=0.4669, pruned_loss=0.187, over 28621.00 frames. ], tot_loss[loss=0.3756, simple_loss=0.4221, pruned_loss=0.1646, over 5665071.36 frames. ], libri_tot_loss[loss=0.389, simple_loss=0.4272, pruned_loss=0.1754, over 5684283.07 frames. ], giga_tot_loss[loss=0.3748, simple_loss=0.4216, pruned_loss=0.164, over 5661419.88 frames. ], batch size: 336, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:21:27,285 INFO [optim.py:369] (1/2) 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 11:21:27,299 INFO [train.py:968] (1/2) Epoch 2, batch 43750, giga_loss[loss=0.3383, simple_loss=0.4043, pruned_loss=0.1362, over 29000.00 frames. ], tot_loss[loss=0.381, simple_loss=0.4256, pruned_loss=0.1682, over 5670731.89 frames. ], libri_tot_loss[loss=0.3886, simple_loss=0.4266, pruned_loss=0.1753, over 5691435.50 frames. ], giga_tot_loss[loss=0.3801, simple_loss=0.4256, pruned_loss=0.1673, over 5660573.20 frames. ], batch size: 128, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:21:29,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:22:16,184 INFO [train.py:968] (1/2) Epoch 2, batch 43800, giga_loss[loss=0.3438, simple_loss=0.4011, pruned_loss=0.1432, over 28826.00 frames. ], tot_loss[loss=0.3823, simple_loss=0.4261, pruned_loss=0.1692, over 5678778.49 frames. ], libri_tot_loss[loss=0.3876, simple_loss=0.4258, pruned_loss=0.1747, over 5698269.53 frames. ], giga_tot_loss[loss=0.3822, simple_loss=0.4268, pruned_loss=0.1688, over 5663679.58 frames. ], batch size: 112, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:23:02,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7763, 2.5151, 1.7831, 0.8982], device='cuda:1'), covar=tensor([0.1576, 0.0845, 0.1230, 0.1649], device='cuda:1'), in_proj_covar=tensor([0.1220, 0.1178, 0.1222, 0.1056], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 11:23:02,409 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 43850, libri_loss[loss=0.3236, simple_loss=0.3696, pruned_loss=0.1387, over 28481.00 frames. ], tot_loss[loss=0.3829, simple_loss=0.4259, pruned_loss=0.17, over 5679711.01 frames. ], libri_tot_loss[loss=0.3874, simple_loss=0.4255, pruned_loss=0.1746, over 5701981.92 frames. ], giga_tot_loss[loss=0.383, simple_loss=0.4267, pruned_loss=0.1697, over 5663675.73 frames. ], batch size: 63, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:23:47,496 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:23:52,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-01 11:23:54,150 INFO [train.py:968] (1/2) Epoch 2, batch 43900, libri_loss[loss=0.4112, simple_loss=0.4441, pruned_loss=0.1891, over 29531.00 frames. ], tot_loss[loss=0.3827, simple_loss=0.4249, pruned_loss=0.1702, over 5675009.95 frames. ], libri_tot_loss[loss=0.3875, simple_loss=0.4257, pruned_loss=0.1747, over 5705801.54 frames. ], giga_tot_loss[loss=0.3826, simple_loss=0.4254, pruned_loss=0.1699, over 5658131.16 frames. ], batch size: 89, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:23:58,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4317, 2.3990, 1.4363, 1.3153], device='cuda:1'), covar=tensor([0.0761, 0.0455, 0.0733, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0457, 0.0328, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 11:24:39,086 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 2, batch 43950, giga_loss[loss=0.3524, simple_loss=0.4078, pruned_loss=0.1484, over 28983.00 frames. ], tot_loss[loss=0.3797, simple_loss=0.4223, pruned_loss=0.1686, over 5666416.73 frames. ], libri_tot_loss[loss=0.3883, simple_loss=0.4264, pruned_loss=0.1751, over 5696016.72 frames. ], giga_tot_loss[loss=0.3786, simple_loss=0.4219, pruned_loss=0.1677, over 5659618.31 frames. ], batch size: 164, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:25:05,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0413, 1.5939, 1.3326, 1.4606], device='cuda:1'), covar=tensor([0.0588, 0.0759, 0.0916, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0491, 0.0527, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 11:25:26,631 INFO [train.py:968] (1/2) Epoch 2, batch 44000, giga_loss[loss=0.375, simple_loss=0.4219, pruned_loss=0.164, over 28937.00 frames. ], tot_loss[loss=0.38, simple_loss=0.4219, pruned_loss=0.1691, over 5675048.09 frames. ], libri_tot_loss[loss=0.3884, simple_loss=0.4266, pruned_loss=0.1751, over 5700544.07 frames. ], giga_tot_loss[loss=0.3788, simple_loss=0.4213, pruned_loss=0.1682, over 5665172.27 frames. ], batch size: 112, lr: 1.16e-02, grad_scale: 8.0
+2023-03-01 11:26:04,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7952, 1.5462, 1.5882, 1.4870], device='cuda:1'), covar=tensor([0.0832, 0.1552, 0.1325, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0810, 0.0641, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 11:26:20,721 INFO [train.py:968] (1/2) Epoch 2, batch 44050, giga_loss[loss=0.3982, simple_loss=0.4392, pruned_loss=0.1786, over 28978.00 frames. ], tot_loss[loss=0.3801, simple_loss=0.4218, pruned_loss=0.1692, over 5682083.04 frames. ], libri_tot_loss[loss=0.3883, simple_loss=0.4265, pruned_loss=0.1751, over 5704725.82 frames. ], giga_tot_loss[loss=0.3791, simple_loss=0.4213, pruned_loss=0.1684, over 5670044.86 frames. ], batch size: 213, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:26:21,312 INFO [optim.py:369] (1/2) 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 11:26:37,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4091, 1.3290, 1.2007, 1.5577], device='cuda:1'), covar=tensor([0.1828, 0.1868, 0.1659, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1005, 0.0816, 0.0906, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 11:27:09,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:27:12,822 INFO [train.py:968] (1/2) Epoch 2, batch 44100, giga_loss[loss=0.4366, simple_loss=0.4503, pruned_loss=0.2114, over 26654.00 frames. ], tot_loss[loss=0.3827, simple_loss=0.4229, pruned_loss=0.1712, over 5674852.30 frames. ], libri_tot_loss[loss=0.3878, simple_loss=0.4262, pruned_loss=0.1747, over 5708463.60 frames. ], giga_tot_loss[loss=0.3822, simple_loss=0.4228, pruned_loss=0.1708, over 5661360.80 frames. ], batch size: 555, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:27:14,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3714, 1.4523, 1.2099, 0.9785], device='cuda:1'), covar=tensor([0.0627, 0.0466, 0.0358, 0.0526], device='cuda:1'), in_proj_covar=tensor([0.1120, 0.0833, 0.0906, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 11:27:35,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:28:00,116 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:28:00,570 INFO [train.py:968] (1/2) Epoch 2, batch 44150, giga_loss[loss=0.3464, simple_loss=0.3945, pruned_loss=0.1491, over 28788.00 frames. ], tot_loss[loss=0.381, simple_loss=0.4212, pruned_loss=0.1704, over 5662712.55 frames. ], libri_tot_loss[loss=0.3882, simple_loss=0.4265, pruned_loss=0.175, over 5693330.18 frames. ], giga_tot_loss[loss=0.3801, simple_loss=0.4207, pruned_loss=0.1698, over 5663822.72 frames. ], batch size: 99, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:28:01,125 INFO [optim.py:369] (1/2) 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 11:28:06,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3816, 1.3150, 1.3246, 1.6362], device='cuda:1'), covar=tensor([0.1825, 0.1823, 0.1547, 0.1882], device='cuda:1'), in_proj_covar=tensor([0.1005, 0.0817, 0.0909, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 11:28:13,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89832.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:28:40,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-03-01 11:28:42,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8164, 1.5000, 4.1515, 3.2039], device='cuda:1'), covar=tensor([0.1701, 0.1798, 0.0341, 0.0510], device='cuda:1'), in_proj_covar=tensor([0.0536, 0.0507, 0.0689, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 11:28:45,305 INFO [train.py:968] (1/2) Epoch 2, batch 44200, giga_loss[loss=0.3918, simple_loss=0.4408, pruned_loss=0.1714, over 28005.00 frames. ], tot_loss[loss=0.379, simple_loss=0.4202, pruned_loss=0.1689, over 5664946.78 frames. ], libri_tot_loss[loss=0.3885, simple_loss=0.4267, pruned_loss=0.1751, over 5687341.99 frames. ], giga_tot_loss[loss=0.3778, simple_loss=0.4194, pruned_loss=0.1681, over 5670718.40 frames. ], batch size: 412, lr: 1.16e-02, grad_scale: 4.0
+2023-03-01 11:29:16,516 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=89902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:29:31,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2217, 3.7845, 3.9250, 2.0647], device='cuda:1'), covar=tensor([0.0420, 0.0386, 0.0690, 0.1848], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0599, 0.0835, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:29:36,260 INFO [train.py:968] (1/2) Epoch 2, batch 44250, giga_loss[loss=0.3633, simple_loss=0.4178, pruned_loss=0.1544, over 28708.00 frames. ], tot_loss[loss=0.3803, simple_loss=0.4217, pruned_loss=0.1695, over 5658483.64 frames. ], libri_tot_loss[loss=0.3884, simple_loss=0.4267, pruned_loss=0.1751, over 5690201.86 frames. ], giga_tot_loss[loss=0.3791, simple_loss=0.4209, pruned_loss=0.1687, over 5659977.71 frames. ], batch size: 242, lr: 1.16e-02, grad_scale: 2.0
+2023-03-01 11:29:37,540 INFO [optim.py:369] (1/2) 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 11:29:48,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3085, 1.4223, 1.1969, 1.4268], device='cuda:1'), covar=tensor([0.1892, 0.1962, 0.1712, 0.2083], device='cuda:1'), in_proj_covar=tensor([0.1004, 0.0829, 0.0914, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 11:29:53,415 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=89937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:29:55,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=89938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:29:59,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=89941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:30:22,169 INFO [train.py:968] (1/2) Epoch 2, batch 44300, giga_loss[loss=0.392, simple_loss=0.4315, pruned_loss=0.1762, over 28974.00 frames. ], tot_loss[loss=0.3835, simple_loss=0.4244, pruned_loss=0.1713, over 5661361.43 frames. ], libri_tot_loss[loss=0.3891, simple_loss=0.4273, pruned_loss=0.1755, over 5684542.47 frames. ], giga_tot_loss[loss=0.3817, simple_loss=0.4232, pruned_loss=0.1701, over 5667652.22 frames. ], batch size: 213, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:30:25,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=89970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:30:33,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4949, 4.9366, 5.1846, 2.5048], device='cuda:1'), covar=tensor([0.0364, 0.0333, 0.0585, 0.1538], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0597, 0.0829, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:31:16,016 INFO [train.py:968] (1/2) Epoch 2, batch 44350, giga_loss[loss=0.3412, simple_loss=0.4148, pruned_loss=0.1338, over 28902.00 frames. ], tot_loss[loss=0.3823, simple_loss=0.4233, pruned_loss=0.1707, over 5658072.03 frames. ], libri_tot_loss[loss=0.3892, simple_loss=0.4273, pruned_loss=0.1755, over 5685529.57 frames. ], giga_tot_loss[loss=0.3808, simple_loss=0.4222, pruned_loss=0.1697, over 5661902.07 frames. ], batch size: 136, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:31:17,975 INFO [optim.py:369] (1/2) 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 11:32:03,421 INFO [train.py:968] (1/2) Epoch 2, batch 44400, giga_loss[loss=0.4, simple_loss=0.444, pruned_loss=0.178, over 28294.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4241, pruned_loss=0.1687, over 5662090.28 frames. ], libri_tot_loss[loss=0.3887, simple_loss=0.4269, pruned_loss=0.1752, over 5689760.17 frames. ], giga_tot_loss[loss=0.3798, simple_loss=0.4235, pruned_loss=0.1681, over 5660904.55 frames. ], batch size: 368, lr: 1.15e-02, grad_scale: 4.0
+2023-03-01 11:32:09,067 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:14,065 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:15,919 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:34,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4401, 1.9214, 1.6731, 1.7199], device='cuda:1'), covar=tensor([0.1337, 0.1599, 0.1143, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0817, 0.0732, 0.0623], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:32:41,771 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:32:48,217 INFO [train.py:968] (1/2) Epoch 2, batch 44450, giga_loss[loss=0.432, simple_loss=0.4708, pruned_loss=0.1966, over 28736.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.425, pruned_loss=0.1662, over 5682775.75 frames. ], libri_tot_loss[loss=0.3897, simple_loss=0.4276, pruned_loss=0.1759, over 5692533.28 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4239, pruned_loss=0.165, over 5679121.41 frames. ], batch size: 284, lr: 1.15e-02, grad_scale: 4.0
+2023-03-01 11:32:50,099 INFO [optim.py:369] (1/2) 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 11:33:11,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:33:37,780 INFO [train.py:968] (1/2) Epoch 2, batch 44500, giga_loss[loss=0.3895, simple_loss=0.4432, pruned_loss=0.1679, over 28942.00 frames. ], tot_loss[loss=0.3826, simple_loss=0.4286, pruned_loss=0.1683, over 5666810.15 frames. ], libri_tot_loss[loss=0.39, simple_loss=0.4278, pruned_loss=0.1761, over 5674994.74 frames. ], giga_tot_loss[loss=0.3809, simple_loss=0.4276, pruned_loss=0.1672, over 5679308.30 frames. ], batch size: 164, lr: 1.15e-02, grad_scale: 4.0
+2023-03-01 11:34:04,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:34:10,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1690, 1.7863, 1.2997, 0.3903], device='cuda:1'), covar=tensor([0.1194, 0.0795, 0.1218, 0.1534], device='cuda:1'), in_proj_covar=tensor([0.1228, 0.1181, 0.1215, 0.1067], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 11:34:15,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90207.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:34:24,667 INFO [train.py:968] (1/2) Epoch 2, batch 44550, giga_loss[loss=0.4807, simple_loss=0.4759, pruned_loss=0.2427, over 26611.00 frames. ], tot_loss[loss=0.3881, simple_loss=0.4317, pruned_loss=0.1722, over 5615558.88 frames. ], libri_tot_loss[loss=0.3907, simple_loss=0.4284, pruned_loss=0.1766, over 5622129.33 frames. ], giga_tot_loss[loss=0.3859, simple_loss=0.4304, pruned_loss=0.1707, over 5674440.90 frames. ], batch size: 555, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:34:25,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3514, 4.3982, 2.4093, 2.2887], device='cuda:1'), covar=tensor([0.0727, 0.0350, 0.0698, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0458, 0.0325, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 11:34:26,618 INFO [optim.py:369] (1/2) 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 11:35:03,985 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:12,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9239, 2.9191, 2.0652, 1.8570], device='cuda:1'), covar=tensor([0.0615, 0.0469, 0.0626, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0459, 0.0324, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 11:35:15,566 INFO [train.py:968] (1/2) Epoch 2, batch 44600, libri_loss[loss=0.4329, simple_loss=0.4586, pruned_loss=0.2036, over 20717.00 frames. ], tot_loss[loss=0.3922, simple_loss=0.4335, pruned_loss=0.1754, over 5554766.36 frames. ], libri_tot_loss[loss=0.3918, simple_loss=0.429, pruned_loss=0.1773, over 5564532.56 frames. ], giga_tot_loss[loss=0.3894, simple_loss=0.4319, pruned_loss=0.1734, over 5653004.78 frames. ], batch size: 187, lr: 1.15e-02, grad_scale: 2.0
+2023-03-01 11:35:21,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:29,755 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:31,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:35:35,597 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-01 11:36:14,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2952, 1.7199, 1.5368, 1.5669], device='cuda:1'), covar=tensor([0.1378, 0.1842, 0.1194, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0805, 0.0720, 0.0617], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 11:36:29,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-01 11:36:32,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:36:37,882 INFO [optim.py:369] (1/2) 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 11:36:52,205 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:36:54,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:36:55,101 INFO [train.py:968] (1/2) Epoch 3, batch 50, giga_loss[loss=0.3804, simple_loss=0.4375, pruned_loss=0.1617, over 28689.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4167, pruned_loss=0.1492, over 1267250.44 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3885, pruned_loss=0.1347, over 173169.73 frames. ], giga_tot_loss[loss=0.3618, simple_loss=0.4208, pruned_loss=0.1514, over 1127881.09 frames. ], batch size: 262, lr: 1.10e-02, grad_scale: 2.0
+2023-03-01 11:37:06,198 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90350.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 11:37:09,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90353.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:37:24,516 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:37:38,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90382.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 11:37:45,765 INFO [train.py:968] (1/2) Epoch 3, batch 100, giga_loss[loss=0.2921, simple_loss=0.3587, pruned_loss=0.1127, over 28479.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4074, pruned_loss=0.1438, over 2237677.22 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.394, pruned_loss=0.1359, over 255023.70 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4089, pruned_loss=0.1447, over 2075987.45 frames. ], batch size: 71, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:38:13,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:14,515 INFO [optim.py:369] (1/2) 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 11:38:15,943 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:34,140 INFO [train.py:968] (1/2) Epoch 3, batch 150, giga_loss[loss=0.254, simple_loss=0.3241, pruned_loss=0.09198, over 28432.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3912, pruned_loss=0.135, over 3006119.57 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3925, pruned_loss=0.1348, over 367234.30 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3915, pruned_loss=0.1352, over 2819496.67 frames. ], batch size: 71, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:38:38,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-01 11:38:38,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:42,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:38:59,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6482, 3.0065, 1.5906, 1.2460], device='cuda:1'), covar=tensor([0.0871, 0.0350, 0.0847, 0.1442], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0451, 0.0322, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 11:39:13,491 INFO [train.py:968] (1/2) Epoch 3, batch 200, giga_loss[loss=0.3031, simple_loss=0.351, pruned_loss=0.1276, over 28481.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1262, over 3607247.47 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3896, pruned_loss=0.1323, over 476278.73 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3746, pruned_loss=0.1262, over 3413894.40 frames. ], batch size: 78, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:39:41,479 INFO [optim.py:369] (1/2) 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 11:39:42,381 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:39:59,805 INFO [train.py:968] (1/2) Epoch 3, batch 250, giga_loss[loss=0.2756, simple_loss=0.3136, pruned_loss=0.1188, over 23869.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3619, pruned_loss=0.1192, over 4059578.18 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.388, pruned_loss=0.1311, over 530518.98 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3609, pruned_loss=0.1189, over 3889195.90 frames. ], batch size: 705, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:40:38,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1392, 1.7375, 1.4271, 0.3971], device='cuda:1'), covar=tensor([0.1460, 0.1015, 0.1659, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.1220, 0.1171, 0.1207, 0.1056], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 11:40:43,992 INFO [train.py:968] (1/2) Epoch 3, batch 300, giga_loss[loss=0.2188, simple_loss=0.2913, pruned_loss=0.07317, over 28948.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3514, pruned_loss=0.1136, over 4425912.36 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3893, pruned_loss=0.1317, over 609537.22 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3497, pruned_loss=0.1128, over 4270481.32 frames. ], batch size: 136, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:40:45,373 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:40:47,496 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:41:17,463 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=90623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:41:18,829 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:41:25,221 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 350, giga_loss[loss=0.2369, simple_loss=0.3047, pruned_loss=0.08451, over 28478.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3458, pruned_loss=0.1107, over 4706005.14 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3943, pruned_loss=0.1342, over 739577.72 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3424, pruned_loss=0.1091, over 4553005.86 frames. ], batch size: 78, lr: 1.09e-02, grad_scale: 2.0
+2023-03-01 11:41:44,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 11:42:02,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6377, 4.0265, 4.3152, 1.9398], device='cuda:1'), covar=tensor([0.0373, 0.0353, 0.0649, 0.1794], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0578, 0.0799, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:42:16,454 INFO [train.py:968] (1/2) Epoch 3, batch 400, giga_loss[loss=0.29, simple_loss=0.351, pruned_loss=0.1144, over 28751.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3414, pruned_loss=0.1079, over 4932409.61 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.391, pruned_loss=0.1325, over 917352.63 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3373, pruned_loss=0.1059, over 4775490.76 frames. ], batch size: 284, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:42:43,337 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 450, giga_loss[loss=0.3286, simple_loss=0.3611, pruned_loss=0.1481, over 26623.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3387, pruned_loss=0.1069, over 5092780.20 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3919, pruned_loss=0.133, over 991655.32 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3345, pruned_loss=0.1049, over 4956943.29 frames. ], batch size: 555, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:43:05,299 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:43:24,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 11:43:34,250 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=90776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:43:36,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=90779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:43:48,258 INFO [train.py:968] (1/2) Epoch 3, batch 500, giga_loss[loss=0.268, simple_loss=0.3284, pruned_loss=0.1038, over 28198.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3351, pruned_loss=0.1047, over 5227010.30 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3902, pruned_loss=0.1321, over 1040551.59 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3315, pruned_loss=0.103, over 5113685.93 frames. ], batch size: 368, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:44:04,546 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 3, batch 550, giga_loss[loss=0.244, simple_loss=0.3203, pruned_loss=0.08381, over 28990.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3327, pruned_loss=0.1035, over 5327670.84 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3897, pruned_loss=0.1316, over 1137203.94 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3289, pruned_loss=0.1017, over 5226541.37 frames. ], batch size: 136, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:44:36,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 11:44:46,001 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:45:22,561 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=90889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:45:23,972 INFO [train.py:968] (1/2) Epoch 3, batch 600, giga_loss[loss=0.2409, simple_loss=0.3064, pruned_loss=0.08767, over 28858.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3302, pruned_loss=0.102, over 5417782.53 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3893, pruned_loss=0.1318, over 1232071.26 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3261, pruned_loss=0.09998, over 5327758.60 frames. ], batch size: 199, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:45:32,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=90898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:45:32,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4507, 1.3684, 1.3300, 1.5137], device='cuda:1'), covar=tensor([0.2067, 0.2083, 0.1780, 0.2166], device='cuda:1'), in_proj_covar=tensor([0.1031, 0.0834, 0.0920, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 11:45:56,766 INFO [optim.py:369] (1/2) 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 11:46:11,856 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 11:46:15,512 INFO [train.py:968] (1/2) Epoch 3, batch 650, giga_loss[loss=0.2923, simple_loss=0.3493, pruned_loss=0.1177, over 28325.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3261, pruned_loss=0.09932, over 5479182.59 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3892, pruned_loss=0.1315, over 1278771.42 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3222, pruned_loss=0.09743, over 5404121.05 frames. ], batch size: 368, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:46:26,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1935, 1.6273, 1.2401, 0.8086], device='cuda:1'), covar=tensor([0.1492, 0.0763, 0.1073, 0.1583], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1165, 0.1227, 0.1057], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 11:46:59,445 INFO [train.py:968] (1/2) Epoch 3, batch 700, giga_loss[loss=0.2686, simple_loss=0.3266, pruned_loss=0.1053, over 28945.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3249, pruned_loss=0.09901, over 5528497.65 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3904, pruned_loss=0.1316, over 1370658.79 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3203, pruned_loss=0.09685, over 5461632.16 frames. ], batch size: 227, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:47:07,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:47:29,987 INFO [optim.py:369] (1/2) 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 11:47:45,963 INFO [train.py:968] (1/2) Epoch 3, batch 750, libri_loss[loss=0.3394, simple_loss=0.4014, pruned_loss=0.1387, over 29538.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3233, pruned_loss=0.09814, over 5563175.49 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3903, pruned_loss=0.1317, over 1527097.43 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3174, pruned_loss=0.09526, over 5497732.82 frames. ], batch size: 89, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:47:46,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:47:48,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:48:00,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7031, 2.0584, 1.8899, 1.8485], device='cuda:1'), covar=tensor([0.1572, 0.1907, 0.1286, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0809, 0.0741, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:48:13,449 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:48:31,773 INFO [train.py:968] (1/2) Epoch 3, batch 800, giga_loss[loss=0.3583, simple_loss=0.404, pruned_loss=0.1563, over 28191.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3249, pruned_loss=0.09959, over 5590255.45 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3907, pruned_loss=0.1321, over 1592259.40 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3192, pruned_loss=0.09676, over 5534349.85 frames. ], batch size: 368, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:48:58,516 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:49:19,966 INFO [train.py:968] (1/2) Epoch 3, batch 850, giga_loss[loss=0.3838, simple_loss=0.4258, pruned_loss=0.1709, over 28723.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3394, pruned_loss=0.1084, over 5611659.52 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3905, pruned_loss=0.1316, over 1694868.64 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3337, pruned_loss=0.1057, over 5564011.34 frames. ], batch size: 92, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:49:22,344 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:49:26,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:49:52,719 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 900, libri_loss[loss=0.3819, simple_loss=0.4373, pruned_loss=0.1632, over 29074.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.354, pruned_loss=0.1168, over 5628092.23 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3889, pruned_loss=0.1307, over 1816051.23 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3487, pruned_loss=0.1143, over 5585483.92 frames. ], batch size: 101, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:50:36,922 INFO [optim.py:369] (1/2) 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 11:50:40,142 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 950, giga_loss[loss=0.3926, simple_loss=0.4352, pruned_loss=0.175, over 29021.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3657, pruned_loss=0.1233, over 5637782.02 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3896, pruned_loss=0.131, over 1889515.28 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3608, pruned_loss=0.1211, over 5605997.75 frames. ], batch size: 136, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:50:56,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=91242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:11,141 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:14,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:15,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:51:17,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2364, 3.0797, 1.3341, 1.2026], device='cuda:1'), covar=tensor([0.0802, 0.0358, 0.0558, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.1136, 0.0838, 0.0911, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 11:51:25,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3314, 2.9812, 3.0249, 1.8642], device='cuda:1'), covar=tensor([0.0551, 0.0455, 0.0831, 0.1584], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0566, 0.0800, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 11:51:37,167 INFO [train.py:968] (1/2) Epoch 3, batch 1000, giga_loss[loss=0.2806, simple_loss=0.364, pruned_loss=0.09858, over 28947.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3713, pruned_loss=0.1248, over 5646581.11 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3916, pruned_loss=0.1325, over 1966941.97 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3665, pruned_loss=0.1225, over 5618900.42 frames. ], batch size: 164, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:51:40,664 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:52:24,926 INFO [train.py:968] (1/2) Epoch 3, batch 1050, giga_loss[loss=0.3249, simple_loss=0.3963, pruned_loss=0.1268, over 28717.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3735, pruned_loss=0.1241, over 5648976.71 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3909, pruned_loss=0.1319, over 2019109.32 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3697, pruned_loss=0.1225, over 5629497.24 frames. ], batch size: 262, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:52:50,432 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:52:53,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:07,986 INFO [train.py:968] (1/2) Epoch 3, batch 1100, giga_loss[loss=0.3356, simple_loss=0.3988, pruned_loss=0.1362, over 27574.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3765, pruned_loss=0.1255, over 5657978.94 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3914, pruned_loss=0.132, over 2133505.16 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3729, pruned_loss=0.1239, over 5636142.39 frames. ], batch size: 472, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:53:15,721 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:22,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:25,341 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=91410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:36,314 INFO [optim.py:369] (1/2) 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 11:53:51,186 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:53:52,254 INFO [train.py:968] (1/2) Epoch 3, batch 1150, giga_loss[loss=0.2956, simple_loss=0.3626, pruned_loss=0.1143, over 28849.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3788, pruned_loss=0.1277, over 5663470.85 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3924, pruned_loss=0.1326, over 2203414.27 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3754, pruned_loss=0.1262, over 5646011.41 frames. ], batch size: 119, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:54:38,028 INFO [train.py:968] (1/2) Epoch 3, batch 1200, libri_loss[loss=0.3269, simple_loss=0.3899, pruned_loss=0.132, over 29549.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3825, pruned_loss=0.1305, over 5672673.58 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3923, pruned_loss=0.1324, over 2313368.83 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3794, pruned_loss=0.1292, over 5651827.81 frames. ], batch size: 79, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:55:04,139 INFO [optim.py:369] (1/2) 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 11:55:17,376 INFO [train.py:968] (1/2) Epoch 3, batch 1250, giga_loss[loss=0.3174, simple_loss=0.386, pruned_loss=0.1244, over 29037.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3852, pruned_loss=0.1318, over 5682116.80 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3913, pruned_loss=0.132, over 2449625.58 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3829, pruned_loss=0.1309, over 5661444.19 frames. ], batch size: 155, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:55:39,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0574, 1.3183, 1.1702, 1.1788], device='cuda:1'), covar=tensor([0.1122, 0.0424, 0.0431, 0.1464], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0168, 0.0171, 0.0309], device='cuda:1'), out_proj_covar=tensor([0.0037, 0.0027, 0.0024, 0.0042], device='cuda:1')
+2023-03-01 11:56:03,286 INFO [train.py:968] (1/2) Epoch 3, batch 1300, giga_loss[loss=0.3468, simple_loss=0.4031, pruned_loss=0.1452, over 28876.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3884, pruned_loss=0.1331, over 5677212.37 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3902, pruned_loss=0.1313, over 2524573.13 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3868, pruned_loss=0.1327, over 5665008.13 frames. ], batch size: 199, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:56:23,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=91617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:56:28,155 INFO [optim.py:369] (1/2) 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 11:56:45,085 INFO [train.py:968] (1/2) Epoch 3, batch 1350, giga_loss[loss=0.3297, simple_loss=0.3937, pruned_loss=0.1328, over 29123.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.391, pruned_loss=0.1341, over 5683081.58 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3909, pruned_loss=0.1317, over 2607921.52 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3895, pruned_loss=0.1337, over 5668667.17 frames. ], batch size: 128, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:57:15,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=91678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:57:26,416 INFO [train.py:968] (1/2) Epoch 3, batch 1400, giga_loss[loss=0.341, simple_loss=0.4073, pruned_loss=0.1373, over 29025.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3916, pruned_loss=0.1333, over 5690992.71 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3915, pruned_loss=0.1322, over 2682521.56 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3901, pruned_loss=0.1328, over 5681375.69 frames. ], batch size: 155, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:57:56,587 INFO [optim.py:369] (1/2) 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 11:58:10,068 INFO [train.py:968] (1/2) Epoch 3, batch 1450, libri_loss[loss=0.3655, simple_loss=0.4279, pruned_loss=0.1516, over 29583.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3914, pruned_loss=0.1322, over 5692100.83 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.393, pruned_loss=0.1333, over 2746086.28 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3896, pruned_loss=0.1313, over 5680807.56 frames. ], batch size: 78, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 11:58:25,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=91760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:58:28,271 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 1500, giga_loss[loss=0.3051, simple_loss=0.3793, pruned_loss=0.1155, over 29066.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3897, pruned_loss=0.1299, over 5705128.95 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3939, pruned_loss=0.1338, over 2807168.80 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3878, pruned_loss=0.1289, over 5693721.26 frames. ], batch size: 155, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 11:58:51,377 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=91792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 11:59:17,964 INFO [optim.py:369] (1/2) 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 11:59:34,917 INFO [train.py:968] (1/2) Epoch 3, batch 1550, giga_loss[loss=0.3151, simple_loss=0.3872, pruned_loss=0.1215, over 29008.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.388, pruned_loss=0.1288, over 5700330.04 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.393, pruned_loss=0.1333, over 2853329.54 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3868, pruned_loss=0.1282, over 5688856.27 frames. ], batch size: 155, lr: 1.09e-02, grad_scale: 4.0
+2023-03-01 12:00:08,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3869, 2.4605, 1.4376, 1.2160], device='cuda:1'), covar=tensor([0.1011, 0.0485, 0.0827, 0.1534], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0447, 0.0317, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 12:00:19,721 INFO [train.py:968] (1/2) Epoch 3, batch 1600, giga_loss[loss=0.3325, simple_loss=0.3833, pruned_loss=0.1408, over 28977.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3905, pruned_loss=0.1329, over 5701448.54 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3933, pruned_loss=0.1338, over 2883599.85 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3894, pruned_loss=0.1321, over 5690809.62 frames. ], batch size: 227, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:00:54,141 INFO [optim.py:369] (1/2) 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 12:01:03,752 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=91935.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:01:08,424 INFO [train.py:968] (1/2) Epoch 3, batch 1650, giga_loss[loss=0.3459, simple_loss=0.403, pruned_loss=0.1444, over 28234.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.395, pruned_loss=0.1389, over 5696745.21 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3945, pruned_loss=0.1344, over 2945617.17 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3935, pruned_loss=0.1381, over 5695704.80 frames. ], batch size: 368, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:01:16,112 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 12:01:44,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3222, 1.6241, 1.2678, 1.4340], device='cuda:1'), covar=tensor([0.0947, 0.0356, 0.0413, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0257, 0.0168, 0.0169, 0.0304], device='cuda:1'), out_proj_covar=tensor([0.0037, 0.0028, 0.0024, 0.0041], device='cuda:1')
+2023-03-01 12:01:49,311 INFO [train.py:968] (1/2) Epoch 3, batch 1700, giga_loss[loss=0.3755, simple_loss=0.3983, pruned_loss=0.1763, over 23682.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3955, pruned_loss=0.1401, over 5706456.73 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3951, pruned_loss=0.1342, over 3089346.92 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3941, pruned_loss=0.1398, over 5697504.70 frames. ], batch size: 705, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:02:17,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6073, 4.9560, 5.2294, 2.7967], device='cuda:1'), covar=tensor([0.0344, 0.0313, 0.0723, 0.1467], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0564, 0.0788, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 12:02:21,919 INFO [optim.py:369] (1/2) 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 12:02:36,738 INFO [train.py:968] (1/2) Epoch 3, batch 1750, libri_loss[loss=0.4119, simple_loss=0.4305, pruned_loss=0.1967, over 29318.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3933, pruned_loss=0.1397, over 5694570.89 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3949, pruned_loss=0.1344, over 3144820.94 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3922, pruned_loss=0.1394, over 5684139.94 frames. ], batch size: 71, lr: 1.09e-02, grad_scale: 8.0
+2023-03-01 12:02:46,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:03:18,079 INFO [train.py:968] (1/2) Epoch 3, batch 1800, giga_loss[loss=0.3249, simple_loss=0.3845, pruned_loss=0.1327, over 28862.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3916, pruned_loss=0.1388, over 5695667.55 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3948, pruned_loss=0.1342, over 3249680.85 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3907, pruned_loss=0.1388, over 5682999.83 frames. ], batch size: 145, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:03:41,535 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.90 vs. limit=5.0
+2023-03-01 12:03:44,935 INFO [optim.py:369] (1/2) 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 12:03:55,716 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 12:04:01,576 INFO [train.py:968] (1/2) Epoch 3, batch 1850, giga_loss[loss=0.3319, simple_loss=0.3792, pruned_loss=0.1423, over 23643.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3899, pruned_loss=0.1369, over 5681302.27 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3948, pruned_loss=0.134, over 3292487.71 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3891, pruned_loss=0.1371, over 5675805.95 frames. ], batch size: 705, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:04:37,696 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 3, batch 1900, giga_loss[loss=0.2947, simple_loss=0.3608, pruned_loss=0.1143, over 29015.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3864, pruned_loss=0.1334, over 5681354.03 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3946, pruned_loss=0.1338, over 3322496.17 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3858, pruned_loss=0.1337, over 5681783.97 frames. ], batch size: 106, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:04:57,512 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:05:00,675 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:05:22,743 INFO [optim.py:369] (1/2) 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 12:05:27,008 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:05:38,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9462, 4.4430, 4.6645, 2.2801], device='cuda:1'), covar=tensor([0.0300, 0.0267, 0.0539, 0.1597], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0560, 0.0779, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 12:05:38,665 INFO [train.py:968] (1/2) Epoch 3, batch 1950, giga_loss[loss=0.2553, simple_loss=0.3276, pruned_loss=0.09151, over 28975.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3802, pruned_loss=0.1293, over 5667771.67 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3947, pruned_loss=0.1339, over 3375210.88 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3795, pruned_loss=0.1294, over 5673879.97 frames. ], batch size: 213, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:05:51,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8933, 2.8809, 2.2353, 2.0899], device='cuda:1'), covar=tensor([0.1564, 0.1368, 0.1121, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0802, 0.0734, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 12:05:54,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0999, 1.2099, 1.2258, 1.2375], device='cuda:1'), covar=tensor([0.1024, 0.1034, 0.1446, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0778, 0.0626, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 12:06:20,103 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 2000, giga_loss[loss=0.2419, simple_loss=0.2945, pruned_loss=0.09461, over 23223.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3735, pruned_loss=0.1253, over 5660977.58 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.395, pruned_loss=0.1339, over 3425177.43 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3724, pruned_loss=0.1253, over 5662052.65 frames. ], batch size: 705, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:06:43,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92310.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:06:58,837 INFO [optim.py:369] (1/2) 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 12:07:13,119 INFO [train.py:968] (1/2) Epoch 3, batch 2050, giga_loss[loss=0.3443, simple_loss=0.3977, pruned_loss=0.1455, over 28943.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3686, pruned_loss=0.1226, over 5659254.54 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3952, pruned_loss=0.1339, over 3485162.50 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3671, pruned_loss=0.1223, over 5655771.87 frames. ], batch size: 136, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:07:20,097 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=92348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:07:43,344 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 2100, giga_loss[loss=0.3732, simple_loss=0.4083, pruned_loss=0.169, over 26636.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3687, pruned_loss=0.1221, over 5663442.60 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3952, pruned_loss=0.1336, over 3532136.21 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.367, pruned_loss=0.1219, over 5657141.99 frames. ], batch size: 555, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:08:24,072 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 2150, giga_loss[loss=0.287, simple_loss=0.3555, pruned_loss=0.1092, over 28998.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3708, pruned_loss=0.1233, over 5677694.29 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3965, pruned_loss=0.1342, over 3598592.10 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3679, pruned_loss=0.1224, over 5669946.18 frames. ], batch size: 164, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:08:38,479 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-01 12:08:44,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8546, 3.9753, 1.6500, 1.7337], device='cuda:1'), covar=tensor([0.0974, 0.0384, 0.1023, 0.1552], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0444, 0.0321, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 12:08:45,083 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92453.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:08:47,336 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92456.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:09:13,013 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92485.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:09:16,612 INFO [train.py:968] (1/2) Epoch 3, batch 2200, giga_loss[loss=0.2849, simple_loss=0.3513, pruned_loss=0.1093, over 28803.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.123, over 5685634.08 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3969, pruned_loss=0.1342, over 3707322.76 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3667, pruned_loss=0.1218, over 5673262.61 frames. ], batch size: 66, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:09:47,697 INFO [optim.py:369] (1/2) 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 12:09:59,514 INFO [train.py:968] (1/2) Epoch 3, batch 2250, giga_loss[loss=0.2893, simple_loss=0.3551, pruned_loss=0.1118, over 28717.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3664, pruned_loss=0.1209, over 5696603.60 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3973, pruned_loss=0.1346, over 3740095.15 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3627, pruned_loss=0.1195, over 5683917.64 frames. ], batch size: 284, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:10:07,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6988, 1.5531, 1.2440, 1.3487], device='cuda:1'), covar=tensor([0.0655, 0.0622, 0.0970, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0480, 0.0526, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 12:10:15,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4471, 2.2490, 1.6622, 0.6570], device='cuda:1'), covar=tensor([0.1728, 0.0745, 0.1269, 0.1660], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1150, 0.1222, 0.1046], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 12:10:41,522 INFO [train.py:968] (1/2) Epoch 3, batch 2300, giga_loss[loss=0.2624, simple_loss=0.3324, pruned_loss=0.09626, over 28885.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3622, pruned_loss=0.1178, over 5706473.78 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3975, pruned_loss=0.1346, over 3793694.98 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3584, pruned_loss=0.1164, over 5691964.24 frames. ], batch size: 174, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:11:12,226 INFO [optim.py:369] (1/2) 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 12:11:23,355 INFO [train.py:968] (1/2) Epoch 3, batch 2350, giga_loss[loss=0.2657, simple_loss=0.3398, pruned_loss=0.09582, over 28734.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3604, pruned_loss=0.1168, over 5710373.26 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.398, pruned_loss=0.1348, over 3864761.77 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3557, pruned_loss=0.1149, over 5693919.68 frames. ], batch size: 284, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:11:39,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:12:05,020 INFO [train.py:968] (1/2) Epoch 3, batch 2400, giga_loss[loss=0.2583, simple_loss=0.3312, pruned_loss=0.09275, over 29002.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3577, pruned_loss=0.1157, over 5709871.93 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3987, pruned_loss=0.1352, over 3893350.02 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3532, pruned_loss=0.1137, over 5695968.11 frames. ], batch size: 164, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:12:28,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:12:30,411 INFO [optim.py:369] (1/2) 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 12:12:30,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-01 12:12:41,260 INFO [train.py:968] (1/2) Epoch 3, batch 2450, giga_loss[loss=0.2864, simple_loss=0.3398, pruned_loss=0.1165, over 28222.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3562, pruned_loss=0.1146, over 5715003.73 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.4001, pruned_loss=0.1356, over 3968528.81 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3501, pruned_loss=0.1121, over 5700272.65 frames. ], batch size: 77, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:12:47,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=92749.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:13:19,587 INFO [train.py:968] (1/2) Epoch 3, batch 2500, libri_loss[loss=0.3556, simple_loss=0.4327, pruned_loss=0.1392, over 28538.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3527, pruned_loss=0.1124, over 5723196.66 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.4, pruned_loss=0.1353, over 4015724.64 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3468, pruned_loss=0.11, over 5707998.11 frames. ], batch size: 106, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:13:28,968 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:13:32,717 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:13:56,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:14:02,891 INFO [train.py:968] (1/2) Epoch 3, batch 2550, giga_loss[loss=0.294, simple_loss=0.3512, pruned_loss=0.1184, over 27988.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.351, pruned_loss=0.1114, over 5728853.39 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.401, pruned_loss=0.1358, over 4043923.81 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.345, pruned_loss=0.1088, over 5714397.37 frames. ], batch size: 412, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:14:22,755 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:14:24,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:14:25,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7853, 5.0603, 5.4419, 2.6163], device='cuda:1'), covar=tensor([0.0317, 0.0289, 0.0629, 0.1619], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0554, 0.0777, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 12:14:41,786 INFO [train.py:968] (1/2) Epoch 3, batch 2600, giga_loss[loss=0.2739, simple_loss=0.3385, pruned_loss=0.1047, over 28832.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3508, pruned_loss=0.1113, over 5731046.82 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.4023, pruned_loss=0.1364, over 4099031.06 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3437, pruned_loss=0.1082, over 5714737.93 frames. ], batch size: 174, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:14:42,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=92892.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:14:44,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=92895.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:14:48,003 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=92898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:15:08,739 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:15:21,597 INFO [train.py:968] (1/2) Epoch 3, batch 2650, giga_loss[loss=0.2986, simple_loss=0.3547, pruned_loss=0.1213, over 28233.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3519, pruned_loss=0.1126, over 5724455.14 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.4032, pruned_loss=0.137, over 4134247.50 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3445, pruned_loss=0.1091, over 5716271.94 frames. ], batch size: 77, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:16:09,105 INFO [train.py:968] (1/2) Epoch 3, batch 2700, giga_loss[loss=0.288, simple_loss=0.3471, pruned_loss=0.1144, over 28858.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3569, pruned_loss=0.1165, over 5722456.65 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.4033, pruned_loss=0.137, over 4151600.35 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3506, pruned_loss=0.1136, over 5714745.12 frames. ], batch size: 112, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:16:32,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:16:38,985 INFO [optim.py:369] (1/2) 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 12:16:41,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7001, 2.4072, 1.7807, 1.9705], device='cuda:1'), covar=tensor([0.0519, 0.0626, 0.0911, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0482, 0.0522, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 12:16:53,896 INFO [train.py:968] (1/2) Epoch 3, batch 2750, giga_loss[loss=0.4057, simple_loss=0.4468, pruned_loss=0.1823, over 28678.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3643, pruned_loss=0.1215, over 5719090.09 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.4033, pruned_loss=0.137, over 4177697.01 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3587, pruned_loss=0.1189, over 5710684.03 frames. ], batch size: 262, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:17:45,502 INFO [train.py:968] (1/2) Epoch 3, batch 2800, giga_loss[loss=0.3843, simple_loss=0.412, pruned_loss=0.1783, over 23597.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3727, pruned_loss=0.1278, over 5701693.88 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.403, pruned_loss=0.1368, over 4203573.95 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3681, pruned_loss=0.1258, over 5692339.12 frames. ], batch size: 705, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:18:05,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:18:16,317 INFO [optim.py:369] (1/2) 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 12:18:31,077 INFO [train.py:968] (1/2) Epoch 3, batch 2850, giga_loss[loss=0.334, simple_loss=0.3896, pruned_loss=0.1392, over 27854.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.377, pruned_loss=0.1291, over 5707900.52 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.4029, pruned_loss=0.1367, over 4227462.87 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.373, pruned_loss=0.1274, over 5699230.86 frames. ], batch size: 412, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:19:20,921 INFO [train.py:968] (1/2) Epoch 3, batch 2900, giga_loss[loss=0.4368, simple_loss=0.4534, pruned_loss=0.2101, over 26499.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3842, pruned_loss=0.1332, over 5709301.29 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.4029, pruned_loss=0.1368, over 4252412.42 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3807, pruned_loss=0.1318, over 5699793.33 frames. ], batch size: 555, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:19:24,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4944, 1.5232, 1.0393, 0.9071], device='cuda:1'), covar=tensor([0.0467, 0.0428, 0.0413, 0.0568], device='cuda:1'), in_proj_covar=tensor([0.1094, 0.0811, 0.0880, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 12:19:50,197 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:19:52,059 INFO [optim.py:369] (1/2) 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 12:20:07,496 INFO [train.py:968] (1/2) Epoch 3, batch 2950, giga_loss[loss=0.4295, simple_loss=0.4588, pruned_loss=0.2002, over 27969.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.391, pruned_loss=0.1383, over 5688716.32 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.4031, pruned_loss=0.1371, over 4283946.24 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3878, pruned_loss=0.1369, over 5684917.00 frames. ], batch size: 412, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:20:51,034 INFO [train.py:968] (1/2) Epoch 3, batch 3000, giga_loss[loss=0.2934, simple_loss=0.3546, pruned_loss=0.1161, over 28755.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3908, pruned_loss=0.138, over 5685272.99 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.4031, pruned_loss=0.1373, over 4307141.88 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3881, pruned_loss=0.1368, over 5680109.26 frames. ], batch size: 284, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:20:51,034 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 12:20:56,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2419, 1.3637, 0.9597, 1.3842], device='cuda:1'), covar=tensor([0.1012, 0.0379, 0.0489, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0164, 0.0167, 0.0302], device='cuda:1'), out_proj_covar=tensor([0.0037, 0.0027, 0.0024, 0.0042], device='cuda:1')
+2023-03-01 12:20:59,911 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 12:21:00,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:21:07,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-01 12:21:30,789 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:21:34,464 INFO [optim.py:369] (1/2) 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 12:21:51,221 INFO [train.py:968] (1/2) Epoch 3, batch 3050, giga_loss[loss=0.2844, simple_loss=0.3591, pruned_loss=0.1048, over 29017.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3844, pruned_loss=0.1331, over 5690777.17 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.4028, pruned_loss=0.1371, over 4322788.39 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3822, pruned_loss=0.1322, over 5685143.54 frames. ], batch size: 136, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:22:12,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:22:32,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7621, 1.2594, 3.9587, 3.2708], device='cuda:1'), covar=tensor([0.1669, 0.1904, 0.0309, 0.0480], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0502, 0.0667, 0.0546], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 12:22:32,762 INFO [train.py:968] (1/2) Epoch 3, batch 3100, giga_loss[loss=0.2862, simple_loss=0.3584, pruned_loss=0.107, over 29068.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3826, pruned_loss=0.1309, over 5705287.59 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.4026, pruned_loss=0.1374, over 4383066.68 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3804, pruned_loss=0.1299, over 5695098.17 frames. ], batch size: 136, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:22:36,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:22:46,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:23:06,245 INFO [optim.py:369] (1/2) 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 12:23:08,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:23:21,480 INFO [train.py:968] (1/2) Epoch 3, batch 3150, giga_loss[loss=0.3274, simple_loss=0.3899, pruned_loss=0.1324, over 29038.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3838, pruned_loss=0.1314, over 5708091.84 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.4031, pruned_loss=0.1379, over 4398409.86 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3815, pruned_loss=0.1301, over 5698094.78 frames. ], batch size: 213, lr: 1.08e-02, grad_scale: 4.0
+2023-03-01 12:23:43,002 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 12:24:00,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4834, 2.1327, 1.4593, 0.7526], device='cuda:1'), covar=tensor([0.2274, 0.1196, 0.1366, 0.2156], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1138, 0.1232, 0.1045], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 12:24:03,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:05,008 INFO [train.py:968] (1/2) Epoch 3, batch 3200, libri_loss[loss=0.3958, simple_loss=0.447, pruned_loss=0.1723, over 27924.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3878, pruned_loss=0.1335, over 5708680.02 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.4035, pruned_loss=0.1381, over 4426510.00 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3853, pruned_loss=0.1321, over 5698869.67 frames. ], batch size: 115, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:24:19,598 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=93508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:22,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3071, 1.2146, 1.1496, 1.3582], device='cuda:1'), covar=tensor([0.1991, 0.1961, 0.1686, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.1021, 0.0814, 0.0902, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 12:24:36,257 INFO [optim.py:369] (1/2) 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 12:24:44,142 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:46,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:24:48,534 INFO [train.py:968] (1/2) Epoch 3, batch 3250, giga_loss[loss=0.3176, simple_loss=0.3871, pruned_loss=0.124, over 29063.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3889, pruned_loss=0.1338, over 5707273.37 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.4032, pruned_loss=0.1381, over 4473442.54 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3866, pruned_loss=0.1326, over 5701800.30 frames. ], batch size: 128, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:25:13,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:25:34,998 INFO [train.py:968] (1/2) Epoch 3, batch 3300, giga_loss[loss=0.3796, simple_loss=0.4215, pruned_loss=0.1689, over 27958.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3905, pruned_loss=0.1356, over 5694951.43 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.403, pruned_loss=0.1378, over 4484724.25 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3886, pruned_loss=0.1348, over 5696212.37 frames. ], batch size: 412, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:25:41,279 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:03,697 INFO [optim.py:369] (1/2) 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 12:26:06,709 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:09,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:16,019 INFO [train.py:968] (1/2) Epoch 3, batch 3350, giga_loss[loss=0.3068, simple_loss=0.3726, pruned_loss=0.1205, over 28756.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3904, pruned_loss=0.1357, over 5704025.85 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.4023, pruned_loss=0.1375, over 4523449.48 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.389, pruned_loss=0.1352, over 5702679.69 frames. ], batch size: 99, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:26:36,206 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:39,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:26:50,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6377, 2.1342, 1.8854, 1.8340], device='cuda:1'), covar=tensor([0.1530, 0.1670, 0.1141, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0794, 0.0730, 0.0620], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 12:27:01,859 INFO [train.py:968] (1/2) Epoch 3, batch 3400, giga_loss[loss=0.4237, simple_loss=0.4521, pruned_loss=0.1977, over 26578.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3904, pruned_loss=0.1354, over 5705873.84 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.4023, pruned_loss=0.1373, over 4541945.48 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3891, pruned_loss=0.1352, over 5709996.10 frames. ], batch size: 555, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:27:08,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:32,405 INFO [optim.py:369] (1/2) 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 12:27:42,591 INFO [train.py:968] (1/2) Epoch 3, batch 3450, giga_loss[loss=0.3408, simple_loss=0.4028, pruned_loss=0.1394, over 28744.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3903, pruned_loss=0.1344, over 5703354.82 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.4025, pruned_loss=0.1373, over 4549718.73 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.389, pruned_loss=0.1341, over 5715236.59 frames. ], batch size: 284, lr: 1.08e-02, grad_scale: 8.0
+2023-03-01 12:27:43,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:44,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:45,474 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:27:49,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3315, 1.2737, 1.2021, 1.4221], device='cuda:1'), covar=tensor([0.2036, 0.1900, 0.1726, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.1015, 0.0815, 0.0902, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 12:28:09,587 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:12,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:23,133 INFO [train.py:968] (1/2) Epoch 3, batch 3500, giga_loss[loss=0.2829, simple_loss=0.3534, pruned_loss=0.1062, over 28656.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3898, pruned_loss=0.1331, over 5708800.23 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.4014, pruned_loss=0.1367, over 4596248.22 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3892, pruned_loss=0.1332, over 5712467.66 frames. ], batch size: 78, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:28:30,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1303, 1.1502, 1.0479, 1.5071], device='cuda:1'), covar=tensor([0.2311, 0.2086, 0.2016, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.1012, 0.0811, 0.0897, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 12:28:34,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:40,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:42,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:28:49,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-01 12:28:57,330 INFO [optim.py:369] (1/2) 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 12:28:57,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7183, 1.4542, 1.4153, 1.9179], device='cuda:1'), covar=tensor([0.1824, 0.1867, 0.1604, 0.1903], device='cuda:1'), in_proj_covar=tensor([0.1016, 0.0817, 0.0898, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 12:29:08,026 INFO [train.py:968] (1/2) Epoch 3, batch 3550, giga_loss[loss=0.3051, simple_loss=0.3768, pruned_loss=0.1167, over 29035.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3915, pruned_loss=0.1333, over 5713767.23 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.4012, pruned_loss=0.1367, over 4632848.08 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3909, pruned_loss=0.1333, over 5713255.27 frames. ], batch size: 106, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:29:08,820 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:08,856 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:11,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:37,043 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:43,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=93883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:44,953 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:47,974 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:29:48,915 INFO [train.py:968] (1/2) Epoch 3, batch 3600, giga_loss[loss=0.3342, simple_loss=0.3867, pruned_loss=0.1409, over 28303.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3895, pruned_loss=0.1316, over 5715801.19 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.4009, pruned_loss=0.1365, over 4651625.84 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.389, pruned_loss=0.1317, over 5712871.49 frames. ], batch size: 368, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:30:12,560 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:14,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:16,889 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:30:32,300 INFO [train.py:968] (1/2) Epoch 3, batch 3650, giga_loss[loss=0.2923, simple_loss=0.3607, pruned_loss=0.112, over 28872.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3873, pruned_loss=0.1305, over 5722953.40 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.4002, pruned_loss=0.136, over 4676537.60 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3872, pruned_loss=0.1309, over 5717437.50 frames. ], batch size: 112, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:30:33,474 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 12:30:37,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=93947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:39,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=93950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:41,070 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=93953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:30:46,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3373, 1.4028, 1.2235, 1.5970], device='cuda:1'), covar=tensor([0.1929, 0.1590, 0.1392, 0.1794], device='cuda:1'), in_proj_covar=tensor([0.1027, 0.0832, 0.0908, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 12:31:03,771 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 3700, giga_loss[loss=0.3099, simple_loss=0.3729, pruned_loss=0.1234, over 28860.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3855, pruned_loss=0.1293, over 5718695.46 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.4005, pruned_loss=0.1361, over 4702499.95 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3849, pruned_loss=0.1293, over 5714523.75 frames. ], batch size: 99, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:31:14,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7160, 1.2165, 3.6483, 2.9493], device='cuda:1'), covar=tensor([0.1684, 0.1963, 0.0359, 0.0583], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0496, 0.0666, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 12:31:41,937 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:31:42,987 INFO [optim.py:369] (1/2) 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 12:31:45,659 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:31:57,759 INFO [train.py:968] (1/2) Epoch 3, batch 3750, libri_loss[loss=0.3219, simple_loss=0.3998, pruned_loss=0.1221, over 27710.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3844, pruned_loss=0.1285, over 5726648.14 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.4003, pruned_loss=0.1358, over 4723431.29 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3838, pruned_loss=0.1287, over 5722750.74 frames. ], batch size: 116, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:32:11,590 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 3800, giga_loss[loss=0.3484, simple_loss=0.409, pruned_loss=0.1439, over 28930.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3866, pruned_loss=0.1305, over 5727528.33 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.4006, pruned_loss=0.1361, over 4740794.71 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3857, pruned_loss=0.1303, over 5722130.17 frames. ], batch size: 145, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:33:07,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-01 12:33:09,488 INFO [optim.py:369] (1/2) 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 12:33:20,185 INFO [train.py:968] (1/2) Epoch 3, batch 3850, giga_loss[loss=0.2822, simple_loss=0.3581, pruned_loss=0.1031, over 28889.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3867, pruned_loss=0.1305, over 5726698.41 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3998, pruned_loss=0.1358, over 4764099.88 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3862, pruned_loss=0.1304, over 5719172.15 frames. ], batch size: 112, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:33:39,602 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:34:03,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1116, 0.9923, 0.7209, 1.3688], device='cuda:1'), covar=tensor([0.0843, 0.0327, 0.0435, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0163, 0.0166, 0.0299], device='cuda:1'), out_proj_covar=tensor([0.0037, 0.0027, 0.0024, 0.0042], device='cuda:1')
+2023-03-01 12:34:04,099 INFO [train.py:968] (1/2) Epoch 3, batch 3900, giga_loss[loss=0.289, simple_loss=0.3598, pruned_loss=0.1091, over 28756.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3848, pruned_loss=0.1282, over 5723645.06 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.4, pruned_loss=0.136, over 4786061.63 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3841, pruned_loss=0.1279, over 5714862.69 frames. ], batch size: 99, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:34:06,726 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-01 12:34:36,087 INFO [optim.py:369] (1/2) 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 12:34:44,452 INFO [train.py:968] (1/2) Epoch 3, batch 3950, giga_loss[loss=0.3159, simple_loss=0.3808, pruned_loss=0.1255, over 29088.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3834, pruned_loss=0.1272, over 5727388.22 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.4001, pruned_loss=0.136, over 4807617.23 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3825, pruned_loss=0.1268, over 5717789.01 frames. ], batch size: 155, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:35:24,324 INFO [train.py:968] (1/2) Epoch 3, batch 4000, libri_loss[loss=0.2698, simple_loss=0.3383, pruned_loss=0.1006, over 29666.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3808, pruned_loss=0.1263, over 5718893.10 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3986, pruned_loss=0.1352, over 4842006.43 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3806, pruned_loss=0.1263, over 5708631.39 frames. ], batch size: 69, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:35:42,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5356, 2.0778, 2.0607, 1.9439], device='cuda:1'), covar=tensor([0.0905, 0.1526, 0.1262, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0777, 0.0632, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 12:35:53,215 INFO [optim.py:369] (1/2) 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 12:36:01,877 INFO [train.py:968] (1/2) Epoch 3, batch 4050, giga_loss[loss=0.2807, simple_loss=0.3529, pruned_loss=0.1042, over 28876.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.378, pruned_loss=0.1251, over 5720316.38 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3981, pruned_loss=0.1349, over 4858199.10 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.378, pruned_loss=0.1252, over 5709539.73 frames. ], batch size: 186, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:36:28,591 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:36:41,616 INFO [train.py:968] (1/2) Epoch 3, batch 4100, libri_loss[loss=0.3363, simple_loss=0.4, pruned_loss=0.1363, over 29534.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1255, over 5708293.69 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3983, pruned_loss=0.135, over 4894930.49 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3766, pruned_loss=0.1251, over 5701128.77 frames. ], batch size: 82, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:37:14,060 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 4150, giga_loss[loss=0.2873, simple_loss=0.3525, pruned_loss=0.1111, over 28548.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3769, pruned_loss=0.1257, over 5705205.42 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3976, pruned_loss=0.1346, over 4916134.92 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1254, over 5702188.45 frames. ], batch size: 71, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:38:01,990 INFO [train.py:968] (1/2) Epoch 3, batch 4200, giga_loss[loss=0.301, simple_loss=0.3648, pruned_loss=0.1186, over 28976.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5711449.90 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3975, pruned_loss=0.1345, over 4945007.98 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3758, pruned_loss=0.1265, over 5703925.04 frames. ], batch size: 227, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:38:14,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7161, 1.3018, 4.0463, 3.3135], device='cuda:1'), covar=tensor([0.1542, 0.1853, 0.0284, 0.0468], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0494, 0.0659, 0.0538], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 12:38:34,772 INFO [optim.py:369] (1/2) 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 12:38:37,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6137, 1.5837, 1.1436, 0.9830], device='cuda:1'), covar=tensor([0.0510, 0.0484, 0.0438, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.1119, 0.0827, 0.0894, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 12:38:40,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0903, 1.7125, 4.9372, 3.4481], device='cuda:1'), covar=tensor([0.1512, 0.1644, 0.0235, 0.0438], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0494, 0.0657, 0.0537], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 12:38:42,349 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 4250, giga_loss[loss=0.3079, simple_loss=0.3618, pruned_loss=0.1271, over 28659.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3747, pruned_loss=0.1259, over 5710762.33 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3975, pruned_loss=0.1346, over 4966791.63 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3732, pruned_loss=0.1253, over 5702145.05 frames. ], batch size: 85, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:39:25,416 INFO [train.py:968] (1/2) Epoch 3, batch 4300, giga_loss[loss=0.3118, simple_loss=0.3742, pruned_loss=0.1247, over 28587.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3719, pruned_loss=0.125, over 5707818.14 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3976, pruned_loss=0.1347, over 4977381.00 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3704, pruned_loss=0.1243, over 5701578.80 frames. ], batch size: 307, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:39:35,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4779, 2.1227, 1.4503, 0.4835], device='cuda:1'), covar=tensor([0.1294, 0.0895, 0.1469, 0.1718], device='cuda:1'), in_proj_covar=tensor([0.1223, 0.1155, 0.1242, 0.1045], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 12:39:45,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94616.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:39:46,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8948, 1.4661, 4.3776, 3.4147], device='cuda:1'), covar=tensor([0.1610, 0.1899, 0.0273, 0.0546], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0497, 0.0667, 0.0544], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 12:39:55,849 INFO [optim.py:369] (1/2) 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 12:40:06,320 INFO [train.py:968] (1/2) Epoch 3, batch 4350, giga_loss[loss=0.2633, simple_loss=0.3341, pruned_loss=0.09623, over 28491.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.369, pruned_loss=0.1232, over 5711036.62 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3976, pruned_loss=0.1348, over 4998807.70 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.367, pruned_loss=0.1223, over 5706141.16 frames. ], batch size: 78, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:40:37,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:40:40,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:40:44,716 INFO [train.py:968] (1/2) Epoch 3, batch 4400, giga_loss[loss=0.3237, simple_loss=0.3877, pruned_loss=0.1299, over 28989.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3681, pruned_loss=0.1226, over 5715531.30 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3975, pruned_loss=0.1348, over 5015538.30 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3661, pruned_loss=0.1217, over 5709339.29 frames. ], batch size: 164, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:41:07,049 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:41:20,935 INFO [optim.py:369] (1/2) 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 12:41:33,924 INFO [train.py:968] (1/2) Epoch 3, batch 4450, giga_loss[loss=0.3543, simple_loss=0.3994, pruned_loss=0.1546, over 26786.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3726, pruned_loss=0.1252, over 5704792.35 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3975, pruned_loss=0.1348, over 5025094.08 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3706, pruned_loss=0.1243, over 5700802.72 frames. ], batch size: 555, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:41:34,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=94741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:41:40,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:42:15,369 INFO [train.py:968] (1/2) Epoch 3, batch 4500, giga_loss[loss=0.4338, simple_loss=0.4476, pruned_loss=0.21, over 26670.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3747, pruned_loss=0.1261, over 5705360.42 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.397, pruned_loss=0.1344, over 5047056.35 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.373, pruned_loss=0.1255, over 5697285.71 frames. ], batch size: 555, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:42:50,246 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 4550, giga_loss[loss=0.4293, simple_loss=0.4505, pruned_loss=0.204, over 23850.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3767, pruned_loss=0.1265, over 5703654.45 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3967, pruned_loss=0.1345, over 5059892.80 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3753, pruned_loss=0.1258, over 5694758.94 frames. ], batch size: 705, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:43:26,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-01 12:43:46,599 INFO [train.py:968] (1/2) Epoch 3, batch 4600, libri_loss[loss=0.2856, simple_loss=0.3526, pruned_loss=0.1093, over 29570.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3776, pruned_loss=0.1263, over 5699661.32 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3971, pruned_loss=0.1348, over 5079603.45 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3757, pruned_loss=0.1254, over 5688977.98 frames. ], batch size: 74, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:43:47,565 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=94892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:43:49,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=94895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:44:14,354 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=94924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:44:19,746 INFO [optim.py:369] (1/2) 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 12:44:28,468 INFO [train.py:968] (1/2) Epoch 3, batch 4650, giga_loss[loss=0.2801, simple_loss=0.3583, pruned_loss=0.101, over 28912.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3774, pruned_loss=0.126, over 5699833.12 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3972, pruned_loss=0.135, over 5091932.96 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3754, pruned_loss=0.1249, over 5697031.33 frames. ], batch size: 145, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:44:59,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6493, 2.4245, 1.6634, 0.7919], device='cuda:1'), covar=tensor([0.2350, 0.1233, 0.1549, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.1211, 0.1130, 0.1244, 0.1038], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 12:45:07,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-01 12:45:09,624 INFO [train.py:968] (1/2) Epoch 3, batch 4700, giga_loss[loss=0.2952, simple_loss=0.3586, pruned_loss=0.1159, over 28869.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3784, pruned_loss=0.1269, over 5703929.70 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.3977, pruned_loss=0.135, over 5110956.37 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3759, pruned_loss=0.1258, over 5698082.22 frames. ], batch size: 112, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:45:09,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=94991.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:45:43,289 INFO [optim.py:369] (1/2) 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 12:45:51,896 INFO [train.py:968] (1/2) Epoch 3, batch 4750, giga_loss[loss=0.3807, simple_loss=0.4183, pruned_loss=0.1715, over 26676.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3801, pruned_loss=0.1284, over 5700939.76 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.398, pruned_loss=0.1351, over 5128704.25 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3774, pruned_loss=0.1273, over 5693580.36 frames. ], batch size: 555, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:46:33,227 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-01 12:46:34,811 INFO [train.py:968] (1/2) Epoch 3, batch 4800, giga_loss[loss=0.3016, simple_loss=0.3699, pruned_loss=0.1167, over 28874.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3829, pruned_loss=0.1302, over 5690435.63 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.398, pruned_loss=0.1351, over 5145050.07 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3801, pruned_loss=0.1292, over 5689249.99 frames. ], batch size: 119, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:46:54,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:47:04,697 INFO [optim.py:369] (1/2) 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 12:47:10,065 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95134.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:47:12,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95137.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:47:14,808 INFO [train.py:968] (1/2) Epoch 3, batch 4850, libri_loss[loss=0.2882, simple_loss=0.358, pruned_loss=0.1093, over 29482.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3863, pruned_loss=0.1321, over 5699776.55 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3986, pruned_loss=0.1356, over 5168716.24 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3832, pruned_loss=0.1307, over 5694495.51 frames. ], batch size: 70, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:47:34,594 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95166.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:47:42,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-01 12:47:55,665 INFO [train.py:968] (1/2) Epoch 3, batch 4900, giga_loss[loss=0.3292, simple_loss=0.3941, pruned_loss=0.1321, over 29137.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3874, pruned_loss=0.1323, over 5714742.75 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3978, pruned_loss=0.1351, over 5186685.68 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3853, pruned_loss=0.1315, over 5706310.78 frames. ], batch size: 113, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:48:28,547 INFO [optim.py:369] (1/2) 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 12:48:36,484 INFO [train.py:968] (1/2) Epoch 3, batch 4950, giga_loss[loss=0.2977, simple_loss=0.3627, pruned_loss=0.1163, over 28765.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3894, pruned_loss=0.1339, over 5712727.38 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3986, pruned_loss=0.1357, over 5199828.97 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.387, pruned_loss=0.1328, over 5703710.16 frames. ], batch size: 119, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:48:52,269 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=95259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:48:54,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=95262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:49:20,315 INFO [train.py:968] (1/2) Epoch 3, batch 5000, giga_loss[loss=0.3165, simple_loss=0.3735, pruned_loss=0.1297, over 28639.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3896, pruned_loss=0.1336, over 5722135.17 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3997, pruned_loss=0.1363, over 5219406.38 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3865, pruned_loss=0.1322, over 5711186.21 frames. ], batch size: 85, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:49:20,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=95291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:49:54,498 INFO [optim.py:369] (1/2) 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 12:50:00,659 INFO [train.py:968] (1/2) Epoch 3, batch 5050, giga_loss[loss=0.3884, simple_loss=0.4333, pruned_loss=0.1718, over 27883.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3891, pruned_loss=0.133, over 5727590.20 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.4002, pruned_loss=0.1364, over 5232959.38 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.386, pruned_loss=0.1317, over 5715721.11 frames. ], batch size: 412, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:50:41,762 INFO [train.py:968] (1/2) Epoch 3, batch 5100, giga_loss[loss=0.2898, simple_loss=0.3506, pruned_loss=0.1145, over 28648.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3876, pruned_loss=0.1324, over 5725906.21 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.4003, pruned_loss=0.1365, over 5254982.58 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3847, pruned_loss=0.1312, over 5711867.87 frames. ], batch size: 85, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:50:42,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2067, 1.7865, 1.8131, 1.6967], device='cuda:1'), covar=tensor([0.0988, 0.2013, 0.1442, 0.1566], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0774, 0.0621, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 12:50:57,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 12:51:18,057 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 5150, giga_loss[loss=0.2748, simple_loss=0.3421, pruned_loss=0.1037, over 28511.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3837, pruned_loss=0.13, over 5718099.89 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.4006, pruned_loss=0.1366, over 5257094.19 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3807, pruned_loss=0.1288, over 5718051.24 frames. ], batch size: 85, lr: 1.07e-02, grad_scale: 4.0
+2023-03-01 12:51:48,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-01 12:52:03,986 INFO [train.py:968] (1/2) Epoch 3, batch 5200, giga_loss[loss=0.2883, simple_loss=0.3566, pruned_loss=0.11, over 28918.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3812, pruned_loss=0.1284, over 5723212.65 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.4008, pruned_loss=0.1367, over 5279234.01 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.1271, over 5719632.24 frames. ], batch size: 145, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:52:11,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:52:35,938 INFO [optim.py:369] (1/2) 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 12:52:44,187 INFO [train.py:968] (1/2) Epoch 3, batch 5250, giga_loss[loss=0.2878, simple_loss=0.3663, pruned_loss=0.1047, over 28959.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3815, pruned_loss=0.1279, over 5717424.22 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.401, pruned_loss=0.1369, over 5291376.54 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3783, pruned_loss=0.1265, over 5711479.62 frames. ], batch size: 145, lr: 1.07e-02, grad_scale: 8.0
+2023-03-01 12:53:30,429 INFO [train.py:968] (1/2) Epoch 3, batch 5300, giga_loss[loss=0.2808, simple_loss=0.3475, pruned_loss=0.1071, over 28497.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3826, pruned_loss=0.1274, over 5710712.36 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.4012, pruned_loss=0.1372, over 5295731.34 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3799, pruned_loss=0.126, over 5705885.31 frames. ], batch size: 85, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 12:53:47,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 12:54:03,793 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 5350, giga_loss[loss=0.2894, simple_loss=0.3641, pruned_loss=0.1073, over 28930.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3835, pruned_loss=0.1289, over 5706476.78 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.4008, pruned_loss=0.1371, over 5306704.74 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3813, pruned_loss=0.1277, over 5700046.62 frames. ], batch size: 174, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 12:54:21,429 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:54:27,746 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:54:45,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4568, 1.9161, 1.3445, 0.8950], device='cuda:1'), covar=tensor([0.2108, 0.1234, 0.1040, 0.2004], device='cuda:1'), in_proj_covar=tensor([0.1213, 0.1127, 0.1240, 0.1035], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 12:54:52,324 INFO [train.py:968] (1/2) Epoch 3, batch 5400, giga_loss[loss=0.3207, simple_loss=0.3713, pruned_loss=0.135, over 28782.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.382, pruned_loss=0.1294, over 5700113.01 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.401, pruned_loss=0.1374, over 5308693.51 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3798, pruned_loss=0.128, over 5699656.80 frames. ], batch size: 119, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:55:29,820 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=95732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:55:30,292 INFO [optim.py:369] (1/2) 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 12:55:36,837 INFO [train.py:968] (1/2) Epoch 3, batch 5450, giga_loss[loss=0.3461, simple_loss=0.3939, pruned_loss=0.1492, over 29005.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3812, pruned_loss=0.1305, over 5702327.42 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.4006, pruned_loss=0.1372, over 5320495.28 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3793, pruned_loss=0.1295, over 5698203.13 frames. ], batch size: 128, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:56:02,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9187, 1.6057, 1.4151, 1.3965], device='cuda:1'), covar=tensor([0.0586, 0.0726, 0.0867, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0488, 0.0525, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 12:56:18,698 INFO [train.py:968] (1/2) Epoch 3, batch 5500, giga_loss[loss=0.3149, simple_loss=0.3687, pruned_loss=0.1306, over 28973.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3791, pruned_loss=0.1303, over 5705055.93 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.4013, pruned_loss=0.1378, over 5328591.78 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3767, pruned_loss=0.1289, over 5699572.81 frames. ], batch size: 213, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:56:58,889 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 5550, giga_loss[loss=0.2968, simple_loss=0.3537, pruned_loss=0.12, over 28851.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3772, pruned_loss=0.1298, over 5708960.03 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.4016, pruned_loss=0.138, over 5334299.69 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3748, pruned_loss=0.1284, over 5702815.70 frames. ], batch size: 112, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:57:29,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=95873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:57:45,513 INFO [train.py:968] (1/2) Epoch 3, batch 5600, giga_loss[loss=0.307, simple_loss=0.3642, pruned_loss=0.125, over 28685.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3767, pruned_loss=0.1295, over 5719495.53 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.4018, pruned_loss=0.1385, over 5359227.02 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3736, pruned_loss=0.1276, over 5707162.93 frames. ], batch size: 262, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 12:57:52,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5159, 1.6135, 1.2389, 0.9861], device='cuda:1'), covar=tensor([0.0809, 0.0567, 0.0486, 0.0725], device='cuda:1'), in_proj_covar=tensor([0.1175, 0.0862, 0.0926, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 12:58:18,780 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 5650, giga_loss[loss=0.2598, simple_loss=0.3301, pruned_loss=0.09469, over 29018.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3718, pruned_loss=0.1267, over 5728459.31 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.4018, pruned_loss=0.1387, over 5374670.39 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3685, pruned_loss=0.1248, over 5714381.05 frames. ], batch size: 164, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:59:04,490 INFO [train.py:968] (1/2) Epoch 3, batch 5700, giga_loss[loss=0.2763, simple_loss=0.3476, pruned_loss=0.1024, over 28727.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3672, pruned_loss=0.1239, over 5722541.07 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.4022, pruned_loss=0.1389, over 5377689.23 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3636, pruned_loss=0.1219, over 5716009.43 frames. ], batch size: 284, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:59:21,762 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:59:25,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:59:28,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 12:59:34,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 12:59:38,980 INFO [optim.py:369] (1/2) 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 12:59:40,415 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 5750, giga_loss[loss=0.2943, simple_loss=0.3636, pruned_loss=0.1125, over 28839.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3688, pruned_loss=0.1247, over 5714944.44 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.4021, pruned_loss=0.1387, over 5381589.81 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3651, pruned_loss=0.1229, over 5715362.30 frames. ], batch size: 186, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 12:59:49,243 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 12:59:55,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7479, 2.0555, 1.8754, 1.7933], device='cuda:1'), covar=tensor([0.1445, 0.1603, 0.1146, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0793, 0.0738, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 13:00:15,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9420, 1.7216, 1.2689, 1.5497], device='cuda:1'), covar=tensor([0.0541, 0.0596, 0.0901, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0482, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 13:00:23,385 INFO [train.py:968] (1/2) Epoch 3, batch 5800, giga_loss[loss=0.3117, simple_loss=0.3778, pruned_loss=0.1228, over 28416.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3729, pruned_loss=0.1267, over 5721453.08 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.4022, pruned_loss=0.1388, over 5398866.46 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3687, pruned_loss=0.1246, over 5718020.08 frames. ], batch size: 71, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:00:39,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:00:39,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-01 13:00:43,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1533, 1.9995, 1.2281, 1.1798], device='cuda:1'), covar=tensor([0.0834, 0.0550, 0.0799, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0449, 0.0317, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 13:01:00,273 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 5850, giga_loss[loss=0.3308, simple_loss=0.3929, pruned_loss=0.1344, over 28257.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3758, pruned_loss=0.1276, over 5726897.24 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.4014, pruned_loss=0.1384, over 5417069.52 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3719, pruned_loss=0.1257, over 5720376.75 frames. ], batch size: 368, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:01:31,229 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:33,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:37,908 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:41,544 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:01:49,034 INFO [train.py:968] (1/2) Epoch 3, batch 5900, giga_loss[loss=0.3038, simple_loss=0.3686, pruned_loss=0.1195, over 28513.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3784, pruned_loss=0.1283, over 5719756.54 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.4013, pruned_loss=0.1384, over 5423383.72 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3751, pruned_loss=0.1268, over 5713093.72 frames. ], batch size: 60, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:02:00,587 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:01,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:02:03,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8346, 1.1811, 4.2418, 3.3677], device='cuda:1'), covar=tensor([0.1599, 0.2008, 0.0333, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0491, 0.0670, 0.0535], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 13:02:07,222 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:28,112 INFO [optim.py:369] (1/2) 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 13:02:32,415 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:32,751 INFO [train.py:968] (1/2) Epoch 3, batch 5950, giga_loss[loss=0.4, simple_loss=0.4257, pruned_loss=0.1872, over 23900.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3812, pruned_loss=0.1299, over 5706807.47 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.4012, pruned_loss=0.1384, over 5426557.15 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5708647.66 frames. ], batch size: 705, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:02:40,617 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:42,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:02:57,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-01 13:03:07,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:03:17,413 INFO [train.py:968] (1/2) Epoch 3, batch 6000, giga_loss[loss=0.4125, simple_loss=0.4412, pruned_loss=0.192, over 27627.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3824, pruned_loss=0.1305, over 5707582.35 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.4006, pruned_loss=0.138, over 5437470.51 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3799, pruned_loss=0.1294, over 5705684.40 frames. ], batch size: 472, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 13:03:17,413 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 13:03:26,065 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 13:04:06,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-01 13:04:09,536 INFO [optim.py:369] (1/2) 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 13:04:15,890 INFO [train.py:968] (1/2) Epoch 3, batch 6050, libri_loss[loss=0.3053, simple_loss=0.3712, pruned_loss=0.1197, over 29516.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3907, pruned_loss=0.1383, over 5707098.63 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.4005, pruned_loss=0.1379, over 5439799.16 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3888, pruned_loss=0.1376, over 5704628.57 frames. ], batch size: 70, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:04:47,245 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:04:59,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:05:03,603 INFO [train.py:968] (1/2) Epoch 3, batch 6100, giga_loss[loss=0.3832, simple_loss=0.4431, pruned_loss=0.1617, over 28999.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3986, pruned_loss=0.1454, over 5685609.20 frames. ], libri_tot_loss[loss=0.3381, simple_loss=0.4004, pruned_loss=0.1379, over 5439314.83 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3971, pruned_loss=0.1449, over 5691287.01 frames. ], batch size: 155, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:05:05,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4610, 2.9810, 1.5235, 1.2979], device='cuda:1'), covar=tensor([0.0846, 0.0412, 0.0834, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0453, 0.0319, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0018, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 13:05:31,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-01 13:05:46,023 INFO [optim.py:369] (1/2) 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 13:05:53,002 INFO [train.py:968] (1/2) Epoch 3, batch 6150, giga_loss[loss=0.4024, simple_loss=0.442, pruned_loss=0.1814, over 28958.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.407, pruned_loss=0.1521, over 5677946.58 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.4004, pruned_loss=0.1378, over 5446768.71 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.4057, pruned_loss=0.152, over 5681058.25 frames. ], batch size: 136, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:06:28,171 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 6200, giga_loss[loss=0.3713, simple_loss=0.4159, pruned_loss=0.1633, over 28888.00 frames. ], tot_loss[loss=0.3655, simple_loss=0.4132, pruned_loss=0.1589, over 5669286.26 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.4004, pruned_loss=0.138, over 5455571.36 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4123, pruned_loss=0.159, over 5668106.67 frames. ], batch size: 186, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:07:10,306 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 13:07:17,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:07:21,042 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:07:29,772 INFO [train.py:968] (1/2) Epoch 3, batch 6250, giga_loss[loss=0.3499, simple_loss=0.4121, pruned_loss=0.1439, over 28948.00 frames. ], tot_loss[loss=0.374, simple_loss=0.4195, pruned_loss=0.1642, over 5676151.53 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.4008, pruned_loss=0.1383, over 5464001.04 frames. ], giga_tot_loss[loss=0.3743, simple_loss=0.419, pruned_loss=0.1648, over 5674338.92 frames. ], batch size: 145, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:07:39,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5386, 1.4810, 1.3057, 1.7525], device='cuda:1'), covar=tensor([0.2358, 0.2252, 0.1973, 0.2295], device='cuda:1'), in_proj_covar=tensor([0.0996, 0.0814, 0.0889, 0.0924], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:07:45,317 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:07:46,677 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:08:05,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96580.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:08:17,441 INFO [train.py:968] (1/2) Epoch 3, batch 6300, libri_loss[loss=0.3758, simple_loss=0.4307, pruned_loss=0.1604, over 25902.00 frames. ], tot_loss[loss=0.38, simple_loss=0.4237, pruned_loss=0.1681, over 5663774.28 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.4007, pruned_loss=0.1383, over 5476505.94 frames. ], giga_tot_loss[loss=0.3816, simple_loss=0.4241, pruned_loss=0.1695, over 5658691.72 frames. ], batch size: 136, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:08:40,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3390, 1.6775, 1.2621, 1.4156], device='cuda:1'), covar=tensor([0.0883, 0.0338, 0.0387, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0162, 0.0163, 0.0296], device='cuda:1'), out_proj_covar=tensor([0.0037, 0.0028, 0.0025, 0.0042], device='cuda:1')
+2023-03-01 13:08:44,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:09:06,322 INFO [optim.py:369] (1/2) 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 13:09:11,895 INFO [train.py:968] (1/2) Epoch 3, batch 6350, giga_loss[loss=0.4423, simple_loss=0.4549, pruned_loss=0.2148, over 27885.00 frames. ], tot_loss[loss=0.385, simple_loss=0.4265, pruned_loss=0.1717, over 5644816.91 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.4011, pruned_loss=0.1385, over 5470761.30 frames. ], giga_tot_loss[loss=0.3862, simple_loss=0.4267, pruned_loss=0.1728, over 5647726.65 frames. ], batch size: 412, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:09:32,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8713, 2.6212, 1.9334, 2.1274], device='cuda:1'), covar=tensor([0.0763, 0.0223, 0.0330, 0.0851], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0162, 0.0163, 0.0295], device='cuda:1'), out_proj_covar=tensor([0.0037, 0.0028, 0.0025, 0.0042], device='cuda:1')
+2023-03-01 13:10:05,819 INFO [train.py:968] (1/2) Epoch 3, batch 6400, giga_loss[loss=0.3439, simple_loss=0.3978, pruned_loss=0.1449, over 28845.00 frames. ], tot_loss[loss=0.3891, simple_loss=0.4281, pruned_loss=0.1751, over 5628221.98 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3997, pruned_loss=0.1377, over 5480371.02 frames. ], giga_tot_loss[loss=0.3924, simple_loss=0.43, pruned_loss=0.1774, over 5625749.43 frames. ], batch size: 145, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 13:10:30,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6957, 2.0473, 1.8360, 1.7945], device='cuda:1'), covar=tensor([0.1107, 0.1224, 0.0973, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0791, 0.0732, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 13:10:34,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:10:41,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96723.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:10:45,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96726.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:10:55,841 INFO [optim.py:369] (1/2) 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 13:11:02,785 INFO [train.py:968] (1/2) Epoch 3, batch 6450, libri_loss[loss=0.4078, simple_loss=0.4564, pruned_loss=0.1796, over 20425.00 frames. ], tot_loss[loss=0.3979, simple_loss=0.4337, pruned_loss=0.181, over 5610413.59 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3998, pruned_loss=0.1377, over 5481637.92 frames. ], giga_tot_loss[loss=0.4017, simple_loss=0.4358, pruned_loss=0.1838, over 5611402.60 frames. ], batch size: 187, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:11:12,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:20,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96755.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:11:23,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:27,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:50,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=96783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:57,792 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:11:58,159 INFO [train.py:968] (1/2) Epoch 3, batch 6500, giga_loss[loss=0.3868, simple_loss=0.4295, pruned_loss=0.1721, over 28891.00 frames. ], tot_loss[loss=0.402, simple_loss=0.4368, pruned_loss=0.1836, over 5608049.49 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3997, pruned_loss=0.1375, over 5484453.20 frames. ], giga_tot_loss[loss=0.406, simple_loss=0.439, pruned_loss=0.1865, over 5608337.52 frames. ], batch size: 145, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:11:58,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 13:12:31,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2518, 1.3503, 1.2078, 1.2906], device='cuda:1'), covar=tensor([0.1847, 0.1615, 0.1426, 0.1602], device='cuda:1'), in_proj_covar=tensor([0.1005, 0.0814, 0.0899, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:12:43,314 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 6550, giga_loss[loss=0.367, simple_loss=0.4092, pruned_loss=0.1624, over 28963.00 frames. ], tot_loss[loss=0.4, simple_loss=0.4352, pruned_loss=0.1825, over 5632099.40 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3998, pruned_loss=0.1376, over 5495106.79 frames. ], giga_tot_loss[loss=0.4047, simple_loss=0.4377, pruned_loss=0.1858, over 5625621.92 frames. ], batch size: 227, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:12:54,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:12:57,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3925, 3.9937, 4.1345, 1.5882], device='cuda:1'), covar=tensor([0.0493, 0.0408, 0.0855, 0.2025], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0599, 0.0822, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 13:13:39,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5236, 2.3018, 1.6629, 0.5631], device='cuda:1'), covar=tensor([0.1508, 0.0887, 0.1485, 0.1955], device='cuda:1'), in_proj_covar=tensor([0.1269, 0.1204, 0.1266, 0.1087], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 13:13:39,548 INFO [train.py:968] (1/2) Epoch 3, batch 6600, libri_loss[loss=0.3642, simple_loss=0.429, pruned_loss=0.1497, over 29658.00 frames. ], tot_loss[loss=0.3982, simple_loss=0.4335, pruned_loss=0.1815, over 5632301.47 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3998, pruned_loss=0.1376, over 5502989.99 frames. ], giga_tot_loss[loss=0.4033, simple_loss=0.4362, pruned_loss=0.1852, over 5623134.94 frames. ], batch size: 91, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:13:39,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:13:42,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:14:12,245 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=96923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:14:22,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=96933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:14:23,809 INFO [optim.py:369] (1/2) 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 13:14:26,952 INFO [train.py:968] (1/2) Epoch 3, batch 6650, giga_loss[loss=0.3762, simple_loss=0.4221, pruned_loss=0.1651, over 28614.00 frames. ], tot_loss[loss=0.3949, simple_loss=0.4322, pruned_loss=0.1787, over 5640377.53 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.4001, pruned_loss=0.1377, over 5514752.48 frames. ], giga_tot_loss[loss=0.401, simple_loss=0.4354, pruned_loss=0.1832, over 5626439.58 frames. ], batch size: 307, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:15:12,228 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 13:15:19,511 INFO [train.py:968] (1/2) Epoch 3, batch 6700, giga_loss[loss=0.4738, simple_loss=0.4859, pruned_loss=0.2308, over 27493.00 frames. ], tot_loss[loss=0.3937, simple_loss=0.4321, pruned_loss=0.1776, over 5645583.44 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3999, pruned_loss=0.1376, over 5520709.11 frames. ], giga_tot_loss[loss=0.3995, simple_loss=0.4353, pruned_loss=0.1819, over 5630744.15 frames. ], batch size: 472, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:15:20,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=96991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:15:22,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=96994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:15:29,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5424, 1.5770, 1.6569, 1.6290], device='cuda:1'), covar=tensor([0.0810, 0.1201, 0.0971, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0792, 0.0634, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 13:15:54,530 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:16:07,110 INFO [optim.py:369] (1/2) 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 13:16:13,119 INFO [train.py:968] (1/2) Epoch 3, batch 6750, giga_loss[loss=0.4501, simple_loss=0.4619, pruned_loss=0.2191, over 27454.00 frames. ], tot_loss[loss=0.3945, simple_loss=0.4329, pruned_loss=0.1781, over 5630388.22 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3998, pruned_loss=0.1377, over 5527675.43 frames. ], giga_tot_loss[loss=0.4002, simple_loss=0.436, pruned_loss=0.1821, over 5614580.22 frames. ], batch size: 472, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:16:48,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:16:53,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:17:08,840 INFO [train.py:968] (1/2) Epoch 3, batch 6800, giga_loss[loss=0.3585, simple_loss=0.4133, pruned_loss=0.1518, over 28665.00 frames. ], tot_loss[loss=0.39, simple_loss=0.4296, pruned_loss=0.1751, over 5615251.41 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3998, pruned_loss=0.1377, over 5523347.63 frames. ], giga_tot_loss[loss=0.395, simple_loss=0.4325, pruned_loss=0.1788, over 5607150.71 frames. ], batch size: 262, lr: 1.06e-02, grad_scale: 8.0
+2023-03-01 13:17:11,602 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:17:51,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2638, 1.3572, 0.8935, 0.8646], device='cuda:1'), covar=tensor([0.0552, 0.0435, 0.0440, 0.0545], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.0864, 0.0916, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 13:17:56,503 INFO [optim.py:369] (1/2) 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 13:18:00,163 INFO [train.py:968] (1/2) Epoch 3, batch 6850, giga_loss[loss=0.3149, simple_loss=0.3807, pruned_loss=0.1245, over 28913.00 frames. ], tot_loss[loss=0.3833, simple_loss=0.4261, pruned_loss=0.1702, over 5627538.99 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3997, pruned_loss=0.1377, over 5528997.00 frames. ], giga_tot_loss[loss=0.388, simple_loss=0.4288, pruned_loss=0.1736, over 5617429.10 frames. ], batch size: 213, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:18:01,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8091, 3.4187, 3.4898, 1.5778], device='cuda:1'), covar=tensor([0.0734, 0.0689, 0.1335, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0602, 0.0818, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 13:18:18,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:18:26,916 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 6900, giga_loss[loss=0.3379, simple_loss=0.3979, pruned_loss=0.139, over 29007.00 frames. ], tot_loss[loss=0.3761, simple_loss=0.4213, pruned_loss=0.1654, over 5638997.14 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3993, pruned_loss=0.1376, over 5537323.51 frames. ], giga_tot_loss[loss=0.3823, simple_loss=0.4252, pruned_loss=0.1697, over 5628375.82 frames. ], batch size: 106, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:19:30,211 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:19:33,170 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 6950, giga_loss[loss=0.3642, simple_loss=0.4135, pruned_loss=0.1575, over 28627.00 frames. ], tot_loss[loss=0.3734, simple_loss=0.4195, pruned_loss=0.1637, over 5643165.87 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3994, pruned_loss=0.1377, over 5542868.08 frames. ], giga_tot_loss[loss=0.3788, simple_loss=0.4228, pruned_loss=0.1674, over 5631151.76 frames. ], batch size: 262, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:19:58,172 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:20:23,415 INFO [train.py:968] (1/2) Epoch 3, batch 7000, giga_loss[loss=0.3474, simple_loss=0.406, pruned_loss=0.1444, over 28960.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4178, pruned_loss=0.1627, over 5653066.65 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3994, pruned_loss=0.1377, over 5552949.02 frames. ], giga_tot_loss[loss=0.3769, simple_loss=0.421, pruned_loss=0.1664, over 5637431.87 frames. ], batch size: 164, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:20:36,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:20:37,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:21:06,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:21:09,434 INFO [optim.py:369] (1/2) 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 13:21:14,985 INFO [train.py:968] (1/2) Epoch 3, batch 7050, giga_loss[loss=0.3365, simple_loss=0.4005, pruned_loss=0.1362, over 29005.00 frames. ], tot_loss[loss=0.3703, simple_loss=0.4171, pruned_loss=0.1617, over 5667148.72 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3985, pruned_loss=0.1371, over 5563294.47 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.4209, pruned_loss=0.1659, over 5647941.13 frames. ], batch size: 155, lr: 1.06e-02, grad_scale: 4.0
+2023-03-01 13:22:04,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4877, 1.4070, 1.3248, 1.7236], device='cuda:1'), covar=tensor([0.2137, 0.2099, 0.1804, 0.2182], device='cuda:1'), in_proj_covar=tensor([0.1014, 0.0830, 0.0911, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:22:13,472 INFO [train.py:968] (1/2) Epoch 3, batch 7100, giga_loss[loss=0.379, simple_loss=0.4233, pruned_loss=0.1673, over 28610.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4156, pruned_loss=0.1599, over 5666009.26 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3987, pruned_loss=0.1372, over 5566830.34 frames. ], giga_tot_loss[loss=0.3728, simple_loss=0.4187, pruned_loss=0.1634, over 5648860.83 frames. ], batch size: 336, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:22:15,214 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:22:32,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 13:22:41,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 13:22:52,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 13:23:00,851 INFO [optim.py:369] (1/2) 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 13:23:04,650 INFO [train.py:968] (1/2) Epoch 3, batch 7150, giga_loss[loss=0.3678, simple_loss=0.4302, pruned_loss=0.1527, over 28884.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4125, pruned_loss=0.1558, over 5679359.46 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3983, pruned_loss=0.1369, over 5575023.77 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4157, pruned_loss=0.1593, over 5660651.02 frames. ], batch size: 213, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:24:07,404 INFO [train.py:968] (1/2) Epoch 3, batch 7200, giga_loss[loss=0.3541, simple_loss=0.4196, pruned_loss=0.1443, over 29035.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4144, pruned_loss=0.1553, over 5671936.38 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3985, pruned_loss=0.1371, over 5579434.67 frames. ], giga_tot_loss[loss=0.3666, simple_loss=0.4169, pruned_loss=0.1582, over 5654385.72 frames. ], batch size: 155, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:24:55,565 INFO [optim.py:369] (1/2) 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 13:24:57,851 INFO [train.py:968] (1/2) Epoch 3, batch 7250, giga_loss[loss=0.3457, simple_loss=0.3964, pruned_loss=0.1474, over 28773.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4157, pruned_loss=0.1565, over 5669406.16 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3982, pruned_loss=0.1369, over 5582960.52 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4182, pruned_loss=0.1591, over 5653388.86 frames. ], batch size: 119, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:25:05,452 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:25:33,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1390, 1.2037, 1.1189, 1.3089], device='cuda:1'), covar=tensor([0.2066, 0.1984, 0.1766, 0.1879], device='cuda:1'), in_proj_covar=tensor([0.0991, 0.0811, 0.0891, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:25:51,192 INFO [train.py:968] (1/2) Epoch 3, batch 7300, giga_loss[loss=0.3328, simple_loss=0.3908, pruned_loss=0.1374, over 28624.00 frames. ], tot_loss[loss=0.3647, simple_loss=0.4156, pruned_loss=0.1569, over 5678938.83 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3981, pruned_loss=0.1367, over 5587439.26 frames. ], giga_tot_loss[loss=0.3686, simple_loss=0.418, pruned_loss=0.1596, over 5664325.87 frames. ], batch size: 92, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:25:55,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1862, 1.2473, 1.1492, 1.1906], device='cuda:1'), covar=tensor([0.1894, 0.1865, 0.1646, 0.1703], device='cuda:1'), in_proj_covar=tensor([0.0998, 0.0817, 0.0899, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:26:35,543 INFO [optim.py:369] (1/2) 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 13:26:39,019 INFO [train.py:968] (1/2) Epoch 3, batch 7350, giga_loss[loss=0.3454, simple_loss=0.3957, pruned_loss=0.1476, over 28929.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4141, pruned_loss=0.1568, over 5676750.00 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3981, pruned_loss=0.1368, over 5591407.28 frames. ], giga_tot_loss[loss=0.3672, simple_loss=0.4162, pruned_loss=0.1591, over 5662785.27 frames. ], batch size: 213, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:26:52,844 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:27:25,007 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:27:28,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:27:28,433 INFO [train.py:968] (1/2) Epoch 3, batch 7400, libri_loss[loss=0.3076, simple_loss=0.3752, pruned_loss=0.12, over 29506.00 frames. ], tot_loss[loss=0.3628, simple_loss=0.4119, pruned_loss=0.1568, over 5675829.31 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3978, pruned_loss=0.1365, over 5598999.03 frames. ], giga_tot_loss[loss=0.3667, simple_loss=0.4143, pruned_loss=0.1595, over 5659532.69 frames. ], batch size: 70, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:27:53,706 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:10,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:12,651 INFO [optim.py:369] (1/2) 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 13:28:16,712 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:17,053 INFO [train.py:968] (1/2) Epoch 3, batch 7450, giga_loss[loss=0.4294, simple_loss=0.45, pruned_loss=0.2044, over 27496.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4112, pruned_loss=0.1564, over 5682096.83 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3979, pruned_loss=0.1365, over 5606213.26 frames. ], giga_tot_loss[loss=0.3657, simple_loss=0.4133, pruned_loss=0.159, over 5664575.63 frames. ], batch size: 472, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:28:46,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:47,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=97768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:28:53,635 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 13:29:05,672 INFO [train.py:968] (1/2) Epoch 3, batch 7500, giga_loss[loss=0.3345, simple_loss=0.399, pruned_loss=0.135, over 28906.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4112, pruned_loss=0.1544, over 5685574.86 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3983, pruned_loss=0.1368, over 5597598.06 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4131, pruned_loss=0.1569, over 5681867.74 frames. ], batch size: 174, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:29:46,719 INFO [optim.py:369] (1/2) 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 13:29:49,456 INFO [train.py:968] (1/2) Epoch 3, batch 7550, giga_loss[loss=0.372, simple_loss=0.4232, pruned_loss=0.1604, over 28739.00 frames. ], tot_loss[loss=0.359, simple_loss=0.4112, pruned_loss=0.1534, over 5694310.05 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3982, pruned_loss=0.1368, over 5604865.62 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4131, pruned_loss=0.1559, over 5687446.64 frames. ], batch size: 262, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:30:27,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3148, 1.7123, 1.4995, 1.5843], device='cuda:1'), covar=tensor([0.1326, 0.1733, 0.1145, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0798, 0.0734, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 13:30:37,247 INFO [train.py:968] (1/2) Epoch 3, batch 7600, giga_loss[loss=0.3233, simple_loss=0.3862, pruned_loss=0.1303, over 28820.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4114, pruned_loss=0.1542, over 5693064.82 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.398, pruned_loss=0.1367, over 5611865.19 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4134, pruned_loss=0.1567, over 5683333.12 frames. ], batch size: 186, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:30:53,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=97911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:30:57,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=97914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:31:24,444 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 7650, giga_loss[loss=0.3326, simple_loss=0.3891, pruned_loss=0.1381, over 28951.00 frames. ], tot_loss[loss=0.3596, simple_loss=0.4103, pruned_loss=0.1545, over 5698948.54 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3974, pruned_loss=0.1362, over 5619160.55 frames. ], giga_tot_loss[loss=0.3636, simple_loss=0.4127, pruned_loss=0.1572, over 5686700.38 frames. ], batch size: 213, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:31:29,305 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=97943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:31:33,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=97948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:32:12,887 INFO [train.py:968] (1/2) Epoch 3, batch 7700, libri_loss[loss=0.3555, simple_loss=0.4153, pruned_loss=0.1478, over 29536.00 frames. ], tot_loss[loss=0.3601, simple_loss=0.4101, pruned_loss=0.1551, over 5682803.40 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3976, pruned_loss=0.1364, over 5619910.37 frames. ], giga_tot_loss[loss=0.3641, simple_loss=0.4124, pruned_loss=0.1579, over 5674337.91 frames. ], batch size: 89, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:32:48,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:33:02,679 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 7750, giga_loss[loss=0.3577, simple_loss=0.4058, pruned_loss=0.1549, over 28563.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.4095, pruned_loss=0.155, over 5692789.36 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3977, pruned_loss=0.1365, over 5625702.83 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4113, pruned_loss=0.1574, over 5682067.42 frames. ], batch size: 307, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:33:12,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1418, 1.3906, 1.2440, 0.8900], device='cuda:1'), covar=tensor([0.2014, 0.1845, 0.1692, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.1016, 0.0820, 0.0900, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:33:29,438 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98067.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:33:53,130 INFO [train.py:968] (1/2) Epoch 3, batch 7800, giga_loss[loss=0.3952, simple_loss=0.4254, pruned_loss=0.1825, over 29044.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.4082, pruned_loss=0.1546, over 5687813.31 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3978, pruned_loss=0.1366, over 5617517.75 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4097, pruned_loss=0.1567, over 5688135.83 frames. ], batch size: 155, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:34:12,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:34:20,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:34:29,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5572, 2.2938, 1.5529, 0.8546], device='cuda:1'), covar=tensor([0.1866, 0.0912, 0.1579, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1174, 0.1241, 0.1071], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 13:34:39,508 INFO [optim.py:369] (1/2) 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 13:34:40,799 INFO [train.py:968] (1/2) Epoch 3, batch 7850, libri_loss[loss=0.3437, simple_loss=0.4106, pruned_loss=0.1384, over 29518.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4076, pruned_loss=0.1548, over 5691680.69 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.398, pruned_loss=0.1366, over 5624000.21 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.409, pruned_loss=0.1571, over 5688069.95 frames. ], batch size: 89, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:34:42,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:35:06,677 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:35:09,670 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 7900, giga_loss[loss=0.4129, simple_loss=0.4239, pruned_loss=0.201, over 23813.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.4067, pruned_loss=0.1544, over 5687259.56 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3981, pruned_loss=0.1366, over 5619682.69 frames. ], giga_tot_loss[loss=0.3606, simple_loss=0.4079, pruned_loss=0.1567, over 5690126.21 frames. ], batch size: 705, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:35:39,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:14,439 INFO [optim.py:369] (1/2) 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 13:36:16,100 INFO [train.py:968] (1/2) Epoch 3, batch 7950, giga_loss[loss=0.3256, simple_loss=0.3875, pruned_loss=0.1318, over 28953.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4072, pruned_loss=0.1544, over 5683513.18 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3982, pruned_loss=0.1366, over 5621767.10 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4082, pruned_loss=0.1564, over 5684832.58 frames. ], batch size: 227, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:36:27,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:29,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:30,893 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:33,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:53,944 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:58,136 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:36:58,212 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:00,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:02,294 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:02,715 INFO [train.py:968] (1/2) Epoch 3, batch 8000, giga_loss[loss=0.3368, simple_loss=0.3974, pruned_loss=0.1381, over 28952.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.408, pruned_loss=0.1542, over 5681581.95 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.399, pruned_loss=0.1371, over 5622941.53 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4084, pruned_loss=0.1558, over 5683098.49 frames. ], batch size: 112, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:37:05,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-01 13:37:26,373 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:30,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:37:32,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-01 13:37:48,574 INFO [optim.py:369] (1/2) 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 13:37:52,038 INFO [train.py:968] (1/2) Epoch 3, batch 8050, giga_loss[loss=0.4071, simple_loss=0.4348, pruned_loss=0.1897, over 27626.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.407, pruned_loss=0.153, over 5667834.55 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3991, pruned_loss=0.1373, over 5627320.96 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4073, pruned_loss=0.1543, over 5666072.73 frames. ], batch size: 472, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:38:41,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7037, 3.3864, 3.3984, 1.6812], device='cuda:1'), covar=tensor([0.0643, 0.0469, 0.0924, 0.1920], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0602, 0.0835, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 13:38:42,504 INFO [train.py:968] (1/2) Epoch 3, batch 8100, giga_loss[loss=0.3587, simple_loss=0.4118, pruned_loss=0.1528, over 28702.00 frames. ], tot_loss[loss=0.3571, simple_loss=0.4077, pruned_loss=0.1532, over 5682652.53 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3993, pruned_loss=0.1374, over 5635488.53 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4081, pruned_loss=0.1547, over 5675317.39 frames. ], batch size: 262, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:38:42,988 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 13:38:59,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6765, 1.2710, 3.4478, 2.9111], device='cuda:1'), covar=tensor([0.1535, 0.1735, 0.0381, 0.0541], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0499, 0.0676, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 13:39:28,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1944, 1.3121, 1.1870, 1.2880], device='cuda:1'), covar=tensor([0.2031, 0.1832, 0.1729, 0.1966], device='cuda:1'), in_proj_covar=tensor([0.1016, 0.0809, 0.0906, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:39:33,623 INFO [optim.py:369] (1/2) 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 13:39:35,895 INFO [train.py:968] (1/2) Epoch 3, batch 8150, giga_loss[loss=0.6151, simple_loss=0.5607, pruned_loss=0.3347, over 26570.00 frames. ], tot_loss[loss=0.3635, simple_loss=0.4119, pruned_loss=0.1575, over 5675293.10 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3993, pruned_loss=0.1375, over 5631343.60 frames. ], giga_tot_loss[loss=0.3653, simple_loss=0.4124, pruned_loss=0.1591, over 5674538.58 frames. ], batch size: 555, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:39:37,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:40:01,164 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:40:04,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:40:12,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8473, 1.7454, 1.2806, 1.4337], device='cuda:1'), covar=tensor([0.0560, 0.0570, 0.0850, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0493, 0.0534, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 13:40:29,036 INFO [train.py:968] (1/2) Epoch 3, batch 8200, giga_loss[loss=0.3805, simple_loss=0.4239, pruned_loss=0.1685, over 28817.00 frames. ], tot_loss[loss=0.3646, simple_loss=0.4119, pruned_loss=0.1586, over 5681942.45 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3985, pruned_loss=0.137, over 5640088.91 frames. ], giga_tot_loss[loss=0.3676, simple_loss=0.4134, pruned_loss=0.1609, over 5674716.03 frames. ], batch size: 284, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:40:36,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:40:37,061 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:41:21,572 INFO [optim.py:369] (1/2) 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 13:41:22,826 INFO [train.py:968] (1/2) Epoch 3, batch 8250, giga_loss[loss=0.5594, simple_loss=0.5243, pruned_loss=0.2972, over 26358.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4143, pruned_loss=0.162, over 5662242.87 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3988, pruned_loss=0.1373, over 5631848.58 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4153, pruned_loss=0.1638, over 5664299.30 frames. ], batch size: 555, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:41:39,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9318, 1.3751, 3.6896, 3.2256], device='cuda:1'), covar=tensor([0.1474, 0.1779, 0.0359, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0527, 0.0505, 0.0680, 0.0545], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 13:42:08,733 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98585.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:42:11,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98588.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 13:42:13,024 INFO [train.py:968] (1/2) Epoch 3, batch 8300, giga_loss[loss=0.3633, simple_loss=0.4126, pruned_loss=0.157, over 28486.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4162, pruned_loss=0.1643, over 5665380.38 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3995, pruned_loss=0.1378, over 5637670.18 frames. ], giga_tot_loss[loss=0.3744, simple_loss=0.4169, pruned_loss=0.166, over 5662275.60 frames. ], batch size: 71, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:42:37,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98617.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 13:42:57,336 INFO [optim.py:369] (1/2) 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 13:42:58,076 INFO [train.py:968] (1/2) Epoch 3, batch 8350, giga_loss[loss=0.3428, simple_loss=0.3873, pruned_loss=0.1492, over 28743.00 frames. ], tot_loss[loss=0.371, simple_loss=0.4149, pruned_loss=0.1635, over 5663347.81 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3996, pruned_loss=0.1377, over 5639798.67 frames. ], giga_tot_loss[loss=0.373, simple_loss=0.4155, pruned_loss=0.1652, over 5659393.23 frames. ], batch size: 92, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:43:13,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 8400, giga_loss[loss=0.3285, simple_loss=0.3831, pruned_loss=0.137, over 28690.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.414, pruned_loss=0.1623, over 5669326.71 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3991, pruned_loss=0.1374, over 5643671.23 frames. ], giga_tot_loss[loss=0.3719, simple_loss=0.4152, pruned_loss=0.1643, over 5663040.88 frames. ], batch size: 262, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:43:50,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6485, 1.4674, 1.3159, 2.0531], device='cuda:1'), covar=tensor([0.1889, 0.1943, 0.1748, 0.2000], device='cuda:1'), in_proj_covar=tensor([0.1018, 0.0816, 0.0908, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 13:43:58,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3276, 2.6742, 1.3488, 1.2553], device='cuda:1'), covar=tensor([0.1003, 0.0487, 0.0976, 0.1505], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0461, 0.0321, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 13:44:24,327 INFO [optim.py:369] (1/2) 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 13:44:24,980 INFO [train.py:968] (1/2) Epoch 3, batch 8450, giga_loss[loss=0.3167, simple_loss=0.3713, pruned_loss=0.1311, over 28564.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4105, pruned_loss=0.1583, over 5666732.34 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3993, pruned_loss=0.1376, over 5649999.28 frames. ], giga_tot_loss[loss=0.3666, simple_loss=0.4118, pruned_loss=0.1607, over 5657126.37 frames. ], batch size: 307, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:44:50,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6163, 4.2090, 4.3549, 1.9850], device='cuda:1'), covar=tensor([0.0477, 0.0381, 0.0788, 0.1876], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0600, 0.0829, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 13:44:55,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5487, 2.7022, 1.4754, 1.2945], device='cuda:1'), covar=tensor([0.0840, 0.0418, 0.0794, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0457, 0.0315, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 13:45:08,461 INFO [train.py:968] (1/2) Epoch 3, batch 8500, giga_loss[loss=0.3149, simple_loss=0.3748, pruned_loss=0.1275, over 28949.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4078, pruned_loss=0.1565, over 5674794.61 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3992, pruned_loss=0.1376, over 5651334.29 frames. ], giga_tot_loss[loss=0.3628, simple_loss=0.4089, pruned_loss=0.1584, over 5666181.60 frames. ], batch size: 112, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:45:18,582 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=98799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:22,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=98802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:23,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:47,948 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=98831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:45:56,061 INFO [optim.py:369] (1/2) 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 13:45:56,073 INFO [train.py:968] (1/2) Epoch 3, batch 8550, giga_loss[loss=0.3053, simple_loss=0.3736, pruned_loss=0.1185, over 28759.00 frames. ], tot_loss[loss=0.3586, simple_loss=0.4057, pruned_loss=0.1558, over 5677446.95 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3993, pruned_loss=0.1376, over 5654320.00 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4067, pruned_loss=0.1577, over 5668239.25 frames. ], batch size: 99, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:46:30,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=98874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:46:47,922 INFO [train.py:968] (1/2) Epoch 3, batch 8600, libri_loss[loss=0.3045, simple_loss=0.3657, pruned_loss=0.1217, over 29607.00 frames. ], tot_loss[loss=0.3606, simple_loss=0.4064, pruned_loss=0.1574, over 5667548.54 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3989, pruned_loss=0.1375, over 5656779.15 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.4076, pruned_loss=0.1592, over 5658174.95 frames. ], batch size: 76, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:47:33,147 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=98936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:47:36,242 INFO [optim.py:369] (1/2) 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 13:47:36,255 INFO [train.py:968] (1/2) Epoch 3, batch 8650, giga_loss[loss=0.3949, simple_loss=0.4358, pruned_loss=0.177, over 28952.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4095, pruned_loss=0.1591, over 5667609.77 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.399, pruned_loss=0.1376, over 5661550.62 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4107, pruned_loss=0.1609, over 5655788.68 frames. ], batch size: 227, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:48:25,181 INFO [train.py:968] (1/2) Epoch 3, batch 8700, giga_loss[loss=0.3448, simple_loss=0.4159, pruned_loss=0.1369, over 28583.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4128, pruned_loss=0.1581, over 5657254.69 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3994, pruned_loss=0.138, over 5651416.36 frames. ], giga_tot_loss[loss=0.3664, simple_loss=0.4135, pruned_loss=0.1596, over 5657894.64 frames. ], batch size: 85, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:48:49,199 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:48:51,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:48:53,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4057, 2.8624, 1.4615, 1.1944], device='cuda:1'), covar=tensor([0.0888, 0.0442, 0.0841, 0.1472], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0458, 0.0318, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 13:49:07,839 INFO [optim.py:369] (1/2) 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 13:49:07,851 INFO [train.py:968] (1/2) Epoch 3, batch 8750, giga_loss[loss=0.4249, simple_loss=0.4565, pruned_loss=0.1966, over 27932.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4143, pruned_loss=0.1573, over 5673456.38 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3987, pruned_loss=0.1372, over 5658762.36 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4162, pruned_loss=0.1601, over 5667621.43 frames. ], batch size: 412, lr: 1.05e-02, grad_scale: 4.0
+2023-03-01 13:49:15,665 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:49:53,278 INFO [train.py:968] (1/2) Epoch 3, batch 8800, giga_loss[loss=0.433, simple_loss=0.464, pruned_loss=0.201, over 28293.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4167, pruned_loss=0.1591, over 5664931.21 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.399, pruned_loss=0.1375, over 5652099.04 frames. ], giga_tot_loss[loss=0.3704, simple_loss=0.4182, pruned_loss=0.1614, over 5666729.46 frames. ], batch size: 369, lr: 1.05e-02, grad_scale: 8.0
+2023-03-01 13:50:40,622 INFO [train.py:968] (1/2) Epoch 3, batch 8850, giga_loss[loss=0.3547, simple_loss=0.4077, pruned_loss=0.1508, over 28853.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4193, pruned_loss=0.162, over 5651690.48 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3992, pruned_loss=0.1377, over 5650878.67 frames. ], giga_tot_loss[loss=0.374, simple_loss=0.4204, pruned_loss=0.1638, over 5653945.71 frames. ], batch size: 199, lr: 1.05e-02, grad_scale: 2.0
+2023-03-01 13:50:41,961 INFO [optim.py:369] (1/2) 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 13:51:13,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:51:27,166 INFO [train.py:968] (1/2) Epoch 3, batch 8900, libri_loss[loss=0.3262, simple_loss=0.3991, pruned_loss=0.1267, over 29474.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4194, pruned_loss=0.1629, over 5654738.47 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3997, pruned_loss=0.1379, over 5657049.72 frames. ], giga_tot_loss[loss=0.3749, simple_loss=0.4203, pruned_loss=0.1648, over 5650928.03 frames. ], batch size: 85, lr: 1.05e-02, grad_scale: 2.0
+2023-03-01 13:52:16,048 INFO [train.py:968] (1/2) Epoch 3, batch 8950, giga_loss[loss=0.4115, simple_loss=0.4411, pruned_loss=0.191, over 28883.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4183, pruned_loss=0.1633, over 5643858.97 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.3997, pruned_loss=0.1378, over 5662174.34 frames. ], giga_tot_loss[loss=0.3751, simple_loss=0.4194, pruned_loss=0.1654, over 5636259.55 frames. ], batch size: 199, lr: 1.05e-02, grad_scale: 2.0
+2023-03-01 13:52:17,567 INFO [optim.py:369] (1/2) 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 13:52:49,282 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:52:55,908 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:52:57,901 INFO [train.py:968] (1/2) Epoch 3, batch 9000, giga_loss[loss=0.3211, simple_loss=0.3709, pruned_loss=0.1356, over 28883.00 frames. ], tot_loss[loss=0.369, simple_loss=0.4154, pruned_loss=0.1613, over 5650229.78 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3994, pruned_loss=0.1375, over 5660842.21 frames. ], giga_tot_loss[loss=0.3729, simple_loss=0.4173, pruned_loss=0.1642, over 5645319.42 frames. ], batch size: 106, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:52:57,901 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 13:53:07,098 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 13:53:26,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:53:34,855 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:53:36,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:53:48,944 INFO [train.py:968] (1/2) Epoch 3, batch 9050, giga_loss[loss=0.4277, simple_loss=0.455, pruned_loss=0.2002, over 28785.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4147, pruned_loss=0.1611, over 5658462.56 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3992, pruned_loss=0.1371, over 5668858.09 frames. ], giga_tot_loss[loss=0.3733, simple_loss=0.417, pruned_loss=0.1648, over 5646567.64 frames. ], batch size: 284, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:53:51,169 INFO [optim.py:369] (1/2) 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 13:54:01,062 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:54:12,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:54:37,362 INFO [train.py:968] (1/2) Epoch 3, batch 9100, giga_loss[loss=0.3688, simple_loss=0.4188, pruned_loss=0.1594, over 28866.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4146, pruned_loss=0.1613, over 5664311.84 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3987, pruned_loss=0.1366, over 5676974.22 frames. ], giga_tot_loss[loss=0.3747, simple_loss=0.4177, pruned_loss=0.1658, over 5646702.53 frames. ], batch size: 119, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:55:23,551 INFO [train.py:968] (1/2) Epoch 3, batch 9150, giga_loss[loss=0.3565, simple_loss=0.4013, pruned_loss=0.1559, over 28748.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4148, pruned_loss=0.1619, over 5659043.37 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3993, pruned_loss=0.1372, over 5679160.91 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4171, pruned_loss=0.1656, over 5642244.22 frames. ], batch size: 262, lr: 1.04e-02, grad_scale: 2.0
+2023-03-01 13:55:26,845 INFO [optim.py:369] (1/2) 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 13:55:28,038 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:55:36,946 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:55:39,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5122, 1.5078, 1.2034, 1.3049], device='cuda:1'), covar=tensor([0.0599, 0.0531, 0.0933, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0489, 0.0534, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 13:55:39,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:55:40,706 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-01 13:55:47,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-01 13:56:06,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:56:11,231 INFO [train.py:968] (1/2) Epoch 3, batch 9200, libri_loss[loss=0.3532, simple_loss=0.4199, pruned_loss=0.1432, over 29475.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.4116, pruned_loss=0.1598, over 5666855.35 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3997, pruned_loss=0.1373, over 5683895.49 frames. ], giga_tot_loss[loss=0.3697, simple_loss=0.4133, pruned_loss=0.1631, over 5648975.37 frames. ], batch size: 85, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:56:56,933 INFO [train.py:968] (1/2) Epoch 3, batch 9250, giga_loss[loss=0.399, simple_loss=0.4129, pruned_loss=0.1926, over 23532.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4107, pruned_loss=0.1591, over 5653061.81 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3997, pruned_loss=0.1374, over 5680814.31 frames. ], giga_tot_loss[loss=0.3689, simple_loss=0.4125, pruned_loss=0.1626, over 5640382.06 frames. ], batch size: 705, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:56:58,186 INFO [optim.py:369] (1/2) 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 13:57:25,996 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-01 13:57:48,292 INFO [train.py:968] (1/2) Epoch 3, batch 9300, giga_loss[loss=0.4159, simple_loss=0.4559, pruned_loss=0.1879, over 27915.00 frames. ], tot_loss[loss=0.3657, simple_loss=0.4124, pruned_loss=0.1595, over 5662643.54 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3997, pruned_loss=0.1374, over 5682782.46 frames. ], giga_tot_loss[loss=0.3694, simple_loss=0.414, pruned_loss=0.1624, over 5650697.89 frames. ], batch size: 412, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:57:48,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0011, 1.4756, 1.3765, 1.2264], device='cuda:1'), covar=tensor([0.0961, 0.0329, 0.0339, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0164, 0.0164, 0.0297], device='cuda:1'), out_proj_covar=tensor([0.0039, 0.0029, 0.0025, 0.0044], device='cuda:1')
+2023-03-01 13:58:18,596 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:58:34,925 INFO [train.py:968] (1/2) Epoch 3, batch 9350, giga_loss[loss=0.3582, simple_loss=0.4072, pruned_loss=0.1546, over 28733.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4149, pruned_loss=0.1613, over 5664667.74 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3996, pruned_loss=0.1372, over 5682669.89 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4165, pruned_loss=0.1642, over 5654584.37 frames. ], batch size: 262, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:58:36,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2528, 1.6346, 1.2383, 1.4313], device='cuda:1'), covar=tensor([0.0878, 0.0414, 0.0405, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0163, 0.0163, 0.0297], device='cuda:1'), out_proj_covar=tensor([0.0039, 0.0029, 0.0025, 0.0044], device='cuda:1')
+2023-03-01 13:58:36,970 INFO [optim.py:369] (1/2) 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 13:58:53,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:58:55,477 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:58:57,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 13:59:23,568 INFO [train.py:968] (1/2) Epoch 3, batch 9400, giga_loss[loss=0.3492, simple_loss=0.4107, pruned_loss=0.1439, over 28652.00 frames. ], tot_loss[loss=0.3702, simple_loss=0.4151, pruned_loss=0.1626, over 5649420.81 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3994, pruned_loss=0.1372, over 5673215.82 frames. ], giga_tot_loss[loss=0.3737, simple_loss=0.4168, pruned_loss=0.1653, over 5649165.14 frames. ], batch size: 85, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 13:59:28,813 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-01 13:59:46,258 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:00:07,890 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 9450, giga_loss[loss=0.4653, simple_loss=0.4722, pruned_loss=0.2292, over 27565.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4158, pruned_loss=0.1599, over 5655709.13 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.3996, pruned_loss=0.1374, over 5677848.46 frames. ], giga_tot_loss[loss=0.3712, simple_loss=0.4173, pruned_loss=0.1626, over 5650731.25 frames. ], batch size: 472, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:00:10,107 INFO [optim.py:369] (1/2) 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 14:00:51,334 INFO [train.py:968] (1/2) Epoch 3, batch 9500, giga_loss[loss=0.4261, simple_loss=0.4681, pruned_loss=0.1921, over 28630.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.417, pruned_loss=0.1586, over 5655704.11 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3997, pruned_loss=0.1375, over 5671198.02 frames. ], giga_tot_loss[loss=0.3702, simple_loss=0.4184, pruned_loss=0.161, over 5657826.67 frames. ], batch size: 71, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:01:01,809 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:06,103 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:11,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:18,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:32,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:01:34,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3300, 1.4491, 0.8833, 1.1738], device='cuda:1'), covar=tensor([0.0745, 0.0624, 0.1456, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0481, 0.0521, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 14:01:37,759 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 9550, giga_loss[loss=0.3428, simple_loss=0.4043, pruned_loss=0.1406, over 28940.00 frames. ], tot_loss[loss=0.369, simple_loss=0.4194, pruned_loss=0.1593, over 5667956.69 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3991, pruned_loss=0.1371, over 5674846.60 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4213, pruned_loss=0.1619, over 5666212.58 frames. ], batch size: 227, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:01:41,620 INFO [optim.py:369] (1/2) 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 14:01:51,056 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-01 14:01:54,175 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-01 14:01:58,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99861.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:02:21,379 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:02:25,460 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 9600, giga_loss[loss=0.3621, simple_loss=0.4162, pruned_loss=0.154, over 28938.00 frames. ], tot_loss[loss=0.3738, simple_loss=0.4221, pruned_loss=0.1627, over 5670094.69 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3989, pruned_loss=0.1369, over 5678513.00 frames. ], giga_tot_loss[loss=0.3774, simple_loss=0.4242, pruned_loss=0.1653, over 5665562.59 frames. ], batch size: 136, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:02:50,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:16,301 INFO [train.py:968] (1/2) Epoch 3, batch 9650, giga_loss[loss=0.3838, simple_loss=0.4297, pruned_loss=0.169, over 28677.00 frames. ], tot_loss[loss=0.3791, simple_loss=0.425, pruned_loss=0.1666, over 5638759.25 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.399, pruned_loss=0.137, over 5653522.86 frames. ], giga_tot_loss[loss=0.3825, simple_loss=0.427, pruned_loss=0.1691, over 5658074.41 frames. ], batch size: 262, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:03:18,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-01 14:03:18,726 INFO [optim.py:369] (1/2) 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 14:03:34,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:35,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=99962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:38,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=99965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:03:56,416 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=99982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:02,039 INFO [train.py:968] (1/2) Epoch 3, batch 9700, giga_loss[loss=0.3681, simple_loss=0.4163, pruned_loss=0.1599, over 28923.00 frames. ], tot_loss[loss=0.3802, simple_loss=0.4254, pruned_loss=0.1675, over 5638011.82 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3991, pruned_loss=0.137, over 5649787.63 frames. ], giga_tot_loss[loss=0.3838, simple_loss=0.4275, pruned_loss=0.1701, over 5656019.48 frames. ], batch size: 213, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:04:05,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=99994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:09,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=99997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:44,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:04:47,670 INFO [train.py:968] (1/2) Epoch 3, batch 9750, libri_loss[loss=0.3652, simple_loss=0.4217, pruned_loss=0.1543, over 29261.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4232, pruned_loss=0.165, over 5643501.53 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3991, pruned_loss=0.137, over 5645996.26 frames. ], giga_tot_loss[loss=0.3803, simple_loss=0.4254, pruned_loss=0.1676, over 5661680.35 frames. ], batch size: 94, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:04:49,631 INFO [optim.py:369] (1/2) 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 14:05:17,191 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:05:31,959 INFO [train.py:968] (1/2) Epoch 3, batch 9800, giga_loss[loss=0.3406, simple_loss=0.4052, pruned_loss=0.138, over 28626.00 frames. ], tot_loss[loss=0.3734, simple_loss=0.4223, pruned_loss=0.1623, over 5654319.22 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3989, pruned_loss=0.1369, over 5650335.35 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.4248, pruned_loss=0.1651, over 5665107.55 frames. ], batch size: 92, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:05:32,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:12,106 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-01 14:06:14,644 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:14,916 INFO [train.py:968] (1/2) Epoch 3, batch 9850, giga_loss[loss=0.5129, simple_loss=0.4994, pruned_loss=0.2632, over 26605.00 frames. ], tot_loss[loss=0.3724, simple_loss=0.4222, pruned_loss=0.1613, over 5658877.58 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3983, pruned_loss=0.1365, over 5652936.03 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.4254, pruned_loss=0.1648, over 5665322.73 frames. ], batch size: 555, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:06:17,090 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:06:47,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:54,210 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:06:57,052 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:01,127 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:02,694 INFO [train.py:968] (1/2) Epoch 3, batch 9900, giga_loss[loss=0.3686, simple_loss=0.4214, pruned_loss=0.1578, over 28603.00 frames. ], tot_loss[loss=0.3732, simple_loss=0.4223, pruned_loss=0.162, over 5663365.21 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3972, pruned_loss=0.1358, over 5662654.19 frames. ], giga_tot_loss[loss=0.3797, simple_loss=0.4267, pruned_loss=0.1664, over 5660053.96 frames. ], batch size: 92, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:07:16,968 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-01 14:07:23,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:47,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:07:50,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100236.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:07:51,390 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 9950, giga_loss[loss=0.3418, simple_loss=0.3998, pruned_loss=0.1419, over 28954.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4209, pruned_loss=0.1615, over 5653515.48 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3972, pruned_loss=0.1357, over 5659000.37 frames. ], giga_tot_loss[loss=0.378, simple_loss=0.425, pruned_loss=0.1655, over 5654329.86 frames. ], batch size: 213, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:07:56,178 INFO [optim.py:369] (1/2) 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 14:08:17,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:08:42,701 INFO [train.py:968] (1/2) Epoch 3, batch 10000, giga_loss[loss=0.4115, simple_loss=0.446, pruned_loss=0.1885, over 28819.00 frames. ], tot_loss[loss=0.3715, simple_loss=0.4191, pruned_loss=0.162, over 5647098.99 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3974, pruned_loss=0.1359, over 5660464.37 frames. ], giga_tot_loss[loss=0.3763, simple_loss=0.4222, pruned_loss=0.1651, over 5646538.53 frames. ], batch size: 262, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:09:19,363 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:09:21,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:09:24,058 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 10050, giga_loss[loss=0.3919, simple_loss=0.4326, pruned_loss=0.1756, over 28685.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4164, pruned_loss=0.1607, over 5659292.33 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3973, pruned_loss=0.1359, over 5665464.99 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4195, pruned_loss=0.1638, over 5654304.11 frames. ], batch size: 262, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:09:33,262 INFO [optim.py:369] (1/2) 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 14:09:45,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:09:52,028 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:10:08,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100379.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:10:10,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100382.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:10:19,821 INFO [train.py:968] (1/2) Epoch 3, batch 10100, giga_loss[loss=0.3885, simple_loss=0.4271, pruned_loss=0.175, over 27898.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4148, pruned_loss=0.1605, over 5659160.83 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3981, pruned_loss=0.1365, over 5669574.77 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.417, pruned_loss=0.163, over 5650986.01 frames. ], batch size: 412, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:10:29,619 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100411.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:10:57,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8114, 1.9950, 1.8152, 1.7956], device='cuda:1'), covar=tensor([0.1309, 0.1541, 0.1075, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0816, 0.0752, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 14:11:06,386 INFO [train.py:968] (1/2) Epoch 3, batch 10150, giga_loss[loss=0.3704, simple_loss=0.4173, pruned_loss=0.1618, over 28934.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4146, pruned_loss=0.1616, over 5660614.34 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3978, pruned_loss=0.1362, over 5675795.33 frames. ], giga_tot_loss[loss=0.3731, simple_loss=0.4172, pruned_loss=0.1646, over 5648126.19 frames. ], batch size: 213, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:11:09,351 INFO [optim.py:369] (1/2) 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 14:11:12,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=100448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:11:42,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:11:45,715 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 10200, giga_loss[loss=0.3434, simple_loss=0.3785, pruned_loss=0.1541, over 23912.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4133, pruned_loss=0.1605, over 5648772.62 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3985, pruned_loss=0.1367, over 5667737.63 frames. ], giga_tot_loss[loss=0.3704, simple_loss=0.415, pruned_loss=0.1629, over 5646516.83 frames. ], batch size: 705, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:12:01,767 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:04,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:13,860 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:30,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1331, 3.7339, 3.8155, 1.5873], device='cuda:1'), covar=tensor([0.0582, 0.0636, 0.1168, 0.2130], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0615, 0.0831, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 14:12:32,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:12:39,125 INFO [train.py:968] (1/2) Epoch 3, batch 10250, giga_loss[loss=0.3244, simple_loss=0.3884, pruned_loss=0.1302, over 28599.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.4099, pruned_loss=0.1558, over 5668410.86 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3985, pruned_loss=0.1367, over 5672147.24 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4114, pruned_loss=0.1581, over 5662251.24 frames. ], batch size: 307, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:12:42,936 INFO [optim.py:369] (1/2) 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 14:13:24,296 INFO [train.py:968] (1/2) Epoch 3, batch 10300, giga_loss[loss=0.338, simple_loss=0.4004, pruned_loss=0.1378, over 28870.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4057, pruned_loss=0.1516, over 5653945.38 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3988, pruned_loss=0.1369, over 5670305.14 frames. ], giga_tot_loss[loss=0.3573, simple_loss=0.407, pruned_loss=0.1538, over 5650873.30 frames. ], batch size: 174, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:13:24,594 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:13:28,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:13:51,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2605, 1.2447, 1.0729, 1.5021], device='cuda:1'), covar=tensor([0.2174, 0.2112, 0.2043, 0.2089], device='cuda:1'), in_proj_covar=tensor([0.1017, 0.0819, 0.0905, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 14:13:52,795 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:13:59,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:14:08,503 INFO [train.py:968] (1/2) Epoch 3, batch 10350, giga_loss[loss=0.3716, simple_loss=0.4176, pruned_loss=0.1628, over 28592.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4048, pruned_loss=0.1502, over 5655981.47 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3985, pruned_loss=0.1366, over 5664259.59 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4063, pruned_loss=0.1527, over 5657942.72 frames. ], batch size: 307, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:14:11,765 INFO [optim.py:369] (1/2) 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 14:14:56,644 INFO [train.py:968] (1/2) Epoch 3, batch 10400, giga_loss[loss=0.3304, simple_loss=0.3821, pruned_loss=0.1393, over 28492.00 frames. ], tot_loss[loss=0.351, simple_loss=0.4027, pruned_loss=0.1497, over 5657745.02 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3983, pruned_loss=0.1365, over 5666303.03 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4042, pruned_loss=0.1522, over 5656840.35 frames. ], batch size: 336, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:15:43,130 INFO [train.py:968] (1/2) Epoch 3, batch 10450, giga_loss[loss=0.3314, simple_loss=0.3823, pruned_loss=0.1403, over 28716.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.3996, pruned_loss=0.1485, over 5652362.87 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3982, pruned_loss=0.1363, over 5660542.13 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4009, pruned_loss=0.1508, over 5655950.40 frames. ], batch size: 92, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:15:48,699 INFO [optim.py:369] (1/2) 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 14:16:11,096 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 10500, giga_loss[loss=0.3398, simple_loss=0.4032, pruned_loss=0.1382, over 28783.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4015, pruned_loss=0.1493, over 5662945.93 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3984, pruned_loss=0.1364, over 5668989.57 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4025, pruned_loss=0.1514, over 5658261.98 frames. ], batch size: 199, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:16:28,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=100795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:16:28,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6974, 2.0971, 1.9062, 1.8453], device='cuda:1'), covar=tensor([0.1650, 0.1937, 0.1218, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0810, 0.0738, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 14:17:10,574 INFO [train.py:968] (1/2) Epoch 3, batch 10550, giga_loss[loss=0.4665, simple_loss=0.4811, pruned_loss=0.2259, over 27628.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.4037, pruned_loss=0.1506, over 5651306.72 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3986, pruned_loss=0.1364, over 5663052.21 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4044, pruned_loss=0.1525, over 5652566.50 frames. ], batch size: 472, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:17:15,957 INFO [optim.py:369] (1/2) 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 14:17:31,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9806, 1.3074, 3.7853, 3.2619], device='cuda:1'), covar=tensor([0.1450, 0.1853, 0.0297, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0515, 0.0677, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 14:17:32,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8660, 1.9635, 1.9192, 1.8734], device='cuda:1'), covar=tensor([0.1230, 0.1232, 0.0919, 0.0693], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0812, 0.0739, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 14:18:00,480 INFO [train.py:968] (1/2) Epoch 3, batch 10600, giga_loss[loss=0.3031, simple_loss=0.3655, pruned_loss=0.1203, over 28802.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4018, pruned_loss=0.1496, over 5638818.58 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3983, pruned_loss=0.1363, over 5658083.43 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4026, pruned_loss=0.1515, over 5644806.55 frames. ], batch size: 99, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:18:25,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=100918.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:18:28,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=100921.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:18:45,863 INFO [train.py:968] (1/2) Epoch 3, batch 10650, giga_loss[loss=0.3559, simple_loss=0.4064, pruned_loss=0.1527, over 28896.00 frames. ], tot_loss[loss=0.351, simple_loss=0.402, pruned_loss=0.15, over 5646788.03 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3985, pruned_loss=0.1365, over 5665321.57 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4026, pruned_loss=0.1517, over 5644674.67 frames. ], batch size: 213, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:18:49,123 INFO [optim.py:369] (1/2) 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 14:18:52,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=100950.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:19:36,327 INFO [train.py:968] (1/2) Epoch 3, batch 10700, giga_loss[loss=0.4068, simple_loss=0.4382, pruned_loss=0.1877, over 27496.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4049, pruned_loss=0.152, over 5656985.12 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.399, pruned_loss=0.1367, over 5669735.91 frames. ], giga_tot_loss[loss=0.3562, simple_loss=0.405, pruned_loss=0.1536, over 5650796.06 frames. ], batch size: 472, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:19:52,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:20:06,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0886, 1.4949, 4.8823, 3.6043], device='cuda:1'), covar=tensor([0.1638, 0.1887, 0.0252, 0.0435], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0519, 0.0687, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 14:20:25,298 INFO [train.py:968] (1/2) Epoch 3, batch 10750, giga_loss[loss=0.3679, simple_loss=0.4157, pruned_loss=0.1601, over 28577.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.4089, pruned_loss=0.155, over 5658867.79 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.399, pruned_loss=0.1366, over 5674772.90 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4092, pruned_loss=0.1567, over 5649175.87 frames. ], batch size: 307, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:20:29,754 INFO [optim.py:369] (1/2) 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 14:20:34,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9765, 2.4663, 2.6104, 2.4765], device='cuda:1'), covar=tensor([0.0740, 0.1578, 0.1024, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0794, 0.0637, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 14:21:12,201 INFO [train.py:968] (1/2) Epoch 3, batch 10800, giga_loss[loss=0.3612, simple_loss=0.4104, pruned_loss=0.156, over 28789.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4093, pruned_loss=0.1554, over 5671542.57 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3979, pruned_loss=0.1358, over 5681798.51 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4109, pruned_loss=0.1581, over 5656670.46 frames. ], batch size: 284, lr: 1.04e-02, grad_scale: 8.0
+2023-03-01 14:21:42,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=101123.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:22:00,139 INFO [train.py:968] (1/2) Epoch 3, batch 10850, giga_loss[loss=0.3885, simple_loss=0.4271, pruned_loss=0.1749, over 28870.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4117, pruned_loss=0.1578, over 5674600.29 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3977, pruned_loss=0.1357, over 5686353.10 frames. ], giga_tot_loss[loss=0.3673, simple_loss=0.4135, pruned_loss=0.1605, over 5658547.41 frames. ], batch size: 227, lr: 1.04e-02, grad_scale: 4.0
+2023-03-01 14:22:04,043 INFO [optim.py:369] (1/2) 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 14:22:05,861 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:22:07,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:22:26,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:22:35,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 10900, giga_loss[loss=0.3394, simple_loss=0.4024, pruned_loss=0.1382, over 28938.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4138, pruned_loss=0.159, over 5682326.32 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.398, pruned_loss=0.1359, over 5691486.10 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4155, pruned_loss=0.1618, over 5664374.89 frames. ], batch size: 128, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:23:36,980 INFO [train.py:968] (1/2) Epoch 3, batch 10950, giga_loss[loss=0.352, simple_loss=0.4019, pruned_loss=0.151, over 28645.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4139, pruned_loss=0.1582, over 5673789.75 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3977, pruned_loss=0.1357, over 5694661.30 frames. ], giga_tot_loss[loss=0.3687, simple_loss=0.4157, pruned_loss=0.1608, over 5656809.18 frames. ], batch size: 85, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:23:43,529 INFO [optim.py:369] (1/2) 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 14:24:09,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-01 14:24:25,763 INFO [train.py:968] (1/2) Epoch 3, batch 11000, giga_loss[loss=0.4351, simple_loss=0.4697, pruned_loss=0.2003, over 28608.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4133, pruned_loss=0.1583, over 5669383.64 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3972, pruned_loss=0.1355, over 5699766.92 frames. ], giga_tot_loss[loss=0.3691, simple_loss=0.4156, pruned_loss=0.1613, over 5650398.09 frames. ], batch size: 307, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:24:39,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-01 14:24:47,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:24:50,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=101316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:25:17,793 INFO [train.py:968] (1/2) Epoch 3, batch 11050, giga_loss[loss=0.414, simple_loss=0.4535, pruned_loss=0.1873, over 28846.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4134, pruned_loss=0.1595, over 5644937.32 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3974, pruned_loss=0.1356, over 5695596.83 frames. ], giga_tot_loss[loss=0.3696, simple_loss=0.4152, pruned_loss=0.162, over 5633027.66 frames. ], batch size: 112, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:25:21,284 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:25:22,433 INFO [optim.py:369] (1/2) 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 14:26:09,826 INFO [train.py:968] (1/2) Epoch 3, batch 11100, libri_loss[loss=0.2846, simple_loss=0.346, pruned_loss=0.1116, over 29652.00 frames. ], tot_loss[loss=0.363, simple_loss=0.4107, pruned_loss=0.1577, over 5658985.86 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3963, pruned_loss=0.135, over 5705824.76 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.414, pruned_loss=0.1614, over 5637844.38 frames. ], batch size: 69, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:26:51,138 INFO [train.py:968] (1/2) Epoch 3, batch 11150, giga_loss[loss=0.3998, simple_loss=0.4377, pruned_loss=0.1809, over 28960.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4088, pruned_loss=0.1564, over 5654537.81 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3953, pruned_loss=0.1343, over 5702402.85 frames. ], giga_tot_loss[loss=0.3675, simple_loss=0.4129, pruned_loss=0.161, over 5638516.70 frames. ], batch size: 145, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:26:56,362 INFO [optim.py:369] (1/2) 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 14:27:34,863 INFO [train.py:968] (1/2) Epoch 3, batch 11200, giga_loss[loss=0.3803, simple_loss=0.3969, pruned_loss=0.1819, over 23272.00 frames. ], tot_loss[loss=0.3613, simple_loss=0.4088, pruned_loss=0.1569, over 5660968.30 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3956, pruned_loss=0.1346, over 5707308.43 frames. ], giga_tot_loss[loss=0.3669, simple_loss=0.4121, pruned_loss=0.1609, over 5642455.87 frames. ], batch size: 705, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:27:40,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:27:42,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=101500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:28:21,429 INFO [train.py:968] (1/2) Epoch 3, batch 11250, giga_loss[loss=0.3346, simple_loss=0.3931, pruned_loss=0.138, over 29058.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4083, pruned_loss=0.1568, over 5659081.96 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3958, pruned_loss=0.1348, over 5703257.20 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4112, pruned_loss=0.1604, over 5646788.98 frames. ], batch size: 155, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:28:28,400 INFO [optim.py:369] (1/2) 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 14:29:01,119 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 14:29:12,751 INFO [train.py:968] (1/2) Epoch 3, batch 11300, libri_loss[loss=0.3061, simple_loss=0.3654, pruned_loss=0.1234, over 29348.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4091, pruned_loss=0.1578, over 5659354.27 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3954, pruned_loss=0.1345, over 5705142.34 frames. ], giga_tot_loss[loss=0.367, simple_loss=0.4119, pruned_loss=0.1611, over 5647522.45 frames. ], batch size: 71, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:29:59,744 INFO [train.py:968] (1/2) Epoch 3, batch 11350, giga_loss[loss=0.3849, simple_loss=0.4256, pruned_loss=0.1721, over 28577.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4125, pruned_loss=0.1615, over 5658098.99 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3951, pruned_loss=0.1343, over 5708280.40 frames. ], giga_tot_loss[loss=0.3722, simple_loss=0.4151, pruned_loss=0.1646, over 5645174.27 frames. ], batch size: 336, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:30:00,150 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=101641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:30:02,016 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:30:31,306 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=101673.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:30:48,579 INFO [train.py:968] (1/2) Epoch 3, batch 11400, giga_loss[loss=0.3313, simple_loss=0.385, pruned_loss=0.1388, over 28793.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4132, pruned_loss=0.1617, over 5653189.81 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3954, pruned_loss=0.1344, over 5710629.00 frames. ], giga_tot_loss[loss=0.3724, simple_loss=0.4154, pruned_loss=0.1647, over 5639722.14 frames. ], batch size: 99, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:31:08,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9413, 2.5258, 2.2114, 2.1048], device='cuda:1'), covar=tensor([0.1501, 0.1616, 0.1111, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0816, 0.0739, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 14:31:34,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-01 14:31:42,035 INFO [train.py:968] (1/2) Epoch 3, batch 11450, giga_loss[loss=0.33, simple_loss=0.3925, pruned_loss=0.1337, over 28982.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4148, pruned_loss=0.1642, over 5648726.14 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3952, pruned_loss=0.1343, over 5707807.48 frames. ], giga_tot_loss[loss=0.3754, simple_loss=0.417, pruned_loss=0.1669, over 5639882.77 frames. ], batch size: 164, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:31:48,107 INFO [optim.py:369] (1/2) 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 14:31:54,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0406, 1.4879, 3.6643, 3.0996], device='cuda:1'), covar=tensor([0.1404, 0.1747, 0.0358, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0541, 0.0516, 0.0692, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 14:32:26,991 INFO [train.py:968] (1/2) Epoch 3, batch 11500, libri_loss[loss=0.3093, simple_loss=0.3754, pruned_loss=0.1216, over 29567.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4138, pruned_loss=0.1627, over 5655409.47 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3953, pruned_loss=0.1344, over 5709586.17 frames. ], giga_tot_loss[loss=0.3737, simple_loss=0.416, pruned_loss=0.1657, over 5644972.41 frames. ], batch size: 76, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:32:38,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1398, 1.3458, 1.1643, 0.9628], device='cuda:1'), covar=tensor([0.1863, 0.1875, 0.1681, 0.1666], device='cuda:1'), in_proj_covar=tensor([0.1028, 0.0832, 0.0917, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 14:33:03,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3323, 2.9705, 1.4984, 1.1994], device='cuda:1'), covar=tensor([0.0693, 0.0353, 0.0507, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.0920, 0.0959, 0.1023], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 14:33:16,917 INFO [train.py:968] (1/2) Epoch 3, batch 11550, giga_loss[loss=0.3648, simple_loss=0.4135, pruned_loss=0.1581, over 28866.00 frames. ], tot_loss[loss=0.3733, simple_loss=0.4165, pruned_loss=0.1651, over 5653144.07 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3951, pruned_loss=0.1342, over 5712582.51 frames. ], giga_tot_loss[loss=0.3772, simple_loss=0.4186, pruned_loss=0.1679, over 5641660.43 frames. ], batch size: 186, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:33:23,156 INFO [optim.py:369] (1/2) 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 14:33:49,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=101875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:34:03,664 INFO [train.py:968] (1/2) Epoch 3, batch 11600, giga_loss[loss=0.4067, simple_loss=0.4381, pruned_loss=0.1876, over 27934.00 frames. ], tot_loss[loss=0.3709, simple_loss=0.416, pruned_loss=0.1629, over 5672791.64 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3956, pruned_loss=0.1344, over 5718829.78 frames. ], giga_tot_loss[loss=0.3748, simple_loss=0.4178, pruned_loss=0.1659, over 5656534.66 frames. ], batch size: 412, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:34:09,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 14:34:57,218 INFO [train.py:968] (1/2) Epoch 3, batch 11650, giga_loss[loss=0.3958, simple_loss=0.4322, pruned_loss=0.1797, over 28908.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4165, pruned_loss=0.1637, over 5659635.15 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3954, pruned_loss=0.1342, over 5720841.03 frames. ], giga_tot_loss[loss=0.3756, simple_loss=0.4183, pruned_loss=0.1664, over 5644705.50 frames. ], batch size: 227, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:35:03,241 INFO [optim.py:369] (1/2) 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 14:35:42,800 INFO [train.py:968] (1/2) Epoch 3, batch 11700, giga_loss[loss=0.4052, simple_loss=0.4493, pruned_loss=0.1805, over 28964.00 frames. ], tot_loss[loss=0.3728, simple_loss=0.4174, pruned_loss=0.1641, over 5647284.65 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3951, pruned_loss=0.134, over 5707579.28 frames. ], giga_tot_loss[loss=0.3775, simple_loss=0.4198, pruned_loss=0.1676, over 5644206.92 frames. ], batch size: 136, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:36:06,669 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:36:08,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:36:26,120 INFO [train.py:968] (1/2) Epoch 3, batch 11750, giga_loss[loss=0.3573, simple_loss=0.4104, pruned_loss=0.1522, over 28971.00 frames. ], tot_loss[loss=0.3731, simple_loss=0.4171, pruned_loss=0.1645, over 5642578.69 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.395, pruned_loss=0.1338, over 5704688.03 frames. ], giga_tot_loss[loss=0.3783, simple_loss=0.4199, pruned_loss=0.1684, over 5641431.52 frames. ], batch size: 145, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:36:32,532 INFO [optim.py:369] (1/2) 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 14:36:33,287 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:36:48,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3102, 1.2922, 1.1723, 1.3949], device='cuda:1'), covar=tensor([0.2107, 0.2145, 0.1918, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.1031, 0.0837, 0.0928, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 14:36:56,077 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:37:10,050 INFO [train.py:968] (1/2) Epoch 3, batch 11800, giga_loss[loss=0.3492, simple_loss=0.405, pruned_loss=0.1467, over 28290.00 frames. ], tot_loss[loss=0.3729, simple_loss=0.418, pruned_loss=0.1639, over 5652639.04 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3955, pruned_loss=0.1343, over 5710283.45 frames. ], giga_tot_loss[loss=0.3782, simple_loss=0.4207, pruned_loss=0.1678, over 5644510.71 frames. ], batch size: 368, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:37:55,846 INFO [train.py:968] (1/2) Epoch 3, batch 11850, giga_loss[loss=0.3445, simple_loss=0.4074, pruned_loss=0.1408, over 28879.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4172, pruned_loss=0.1621, over 5656551.51 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3953, pruned_loss=0.1341, over 5715488.70 frames. ], giga_tot_loss[loss=0.3761, simple_loss=0.4201, pruned_loss=0.1661, over 5644247.35 frames. ], batch size: 174, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:38:05,523 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 11900, giga_loss[loss=0.3476, simple_loss=0.4049, pruned_loss=0.1452, over 28730.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.417, pruned_loss=0.1622, over 5658788.69 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3952, pruned_loss=0.1341, over 5718383.84 frames. ], giga_tot_loss[loss=0.3754, simple_loss=0.4196, pruned_loss=0.1656, over 5645846.44 frames. ], batch size: 262, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:39:11,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2397, 1.8272, 1.3779, 0.4255], device='cuda:1'), covar=tensor([0.1266, 0.0888, 0.1468, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.1239, 0.1197, 0.1251, 0.1067], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 14:39:30,013 INFO [train.py:968] (1/2) Epoch 3, batch 11950, giga_loss[loss=0.3437, simple_loss=0.3965, pruned_loss=0.1455, over 28749.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4147, pruned_loss=0.1605, over 5661605.08 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3954, pruned_loss=0.1342, over 5723927.23 frames. ], giga_tot_loss[loss=0.3728, simple_loss=0.4174, pruned_loss=0.1642, over 5644469.12 frames. ], batch size: 284, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:39:35,680 INFO [optim.py:369] (1/2) 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 14:40:12,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5522, 2.4396, 1.6972, 0.6955], device='cuda:1'), covar=tensor([0.1819, 0.0938, 0.1580, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1233, 0.1197, 0.1247, 0.1061], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 14:40:17,294 INFO [train.py:968] (1/2) Epoch 3, batch 12000, giga_loss[loss=0.3917, simple_loss=0.4349, pruned_loss=0.1743, over 28696.00 frames. ], tot_loss[loss=0.3686, simple_loss=0.4153, pruned_loss=0.1609, over 5669758.52 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3948, pruned_loss=0.1339, over 5729590.43 frames. ], giga_tot_loss[loss=0.3741, simple_loss=0.4185, pruned_loss=0.1648, over 5649377.30 frames. ], batch size: 307, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:40:17,294 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 14:40:25,813 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 14:40:44,667 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=102313.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:41:10,809 INFO [train.py:968] (1/2) Epoch 3, batch 12050, giga_loss[loss=0.354, simple_loss=0.4069, pruned_loss=0.1506, over 28949.00 frames. ], tot_loss[loss=0.3698, simple_loss=0.4161, pruned_loss=0.1617, over 5653157.55 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3952, pruned_loss=0.1341, over 5720177.48 frames. ], giga_tot_loss[loss=0.375, simple_loss=0.4189, pruned_loss=0.1655, over 5644059.93 frames. ], batch size: 186, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:41:18,940 INFO [optim.py:369] (1/2) 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 14:41:23,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8077, 4.2018, 1.9040, 1.7401], device='cuda:1'), covar=tensor([0.0856, 0.0314, 0.0810, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0469, 0.0324, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0020, 0.0013, 0.0018], device='cuda:1')
+2023-03-01 14:41:48,291 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 12100, giga_loss[loss=0.3358, simple_loss=0.3888, pruned_loss=0.1414, over 28940.00 frames. ], tot_loss[loss=0.3669, simple_loss=0.4135, pruned_loss=0.1602, over 5670767.45 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3949, pruned_loss=0.1337, over 5724049.24 frames. ], giga_tot_loss[loss=0.3722, simple_loss=0.4165, pruned_loss=0.164, over 5659106.83 frames. ], batch size: 199, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:42:48,257 INFO [train.py:968] (1/2) Epoch 3, batch 12150, giga_loss[loss=0.3677, simple_loss=0.4164, pruned_loss=0.1595, over 28992.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4135, pruned_loss=0.1609, over 5661628.08 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3949, pruned_loss=0.1337, over 5717559.27 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4161, pruned_loss=0.1644, over 5657833.65 frames. ], batch size: 136, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:42:55,248 INFO [optim.py:369] (1/2) 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 14:42:59,551 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 12200, giga_loss[loss=0.3493, simple_loss=0.3988, pruned_loss=0.1499, over 28676.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4148, pruned_loss=0.1622, over 5656536.64 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.395, pruned_loss=0.1337, over 5709734.51 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4169, pruned_loss=0.1651, over 5659977.84 frames. ], batch size: 85, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:43:40,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9423, 3.6177, 3.6488, 1.7623], device='cuda:1'), covar=tensor([0.0461, 0.0428, 0.0761, 0.1910], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0617, 0.0822, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 14:44:25,330 INFO [train.py:968] (1/2) Epoch 3, batch 12250, giga_loss[loss=0.3518, simple_loss=0.4026, pruned_loss=0.1505, over 28911.00 frames. ], tot_loss[loss=0.3713, simple_loss=0.4159, pruned_loss=0.1633, over 5649037.42 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3951, pruned_loss=0.1339, over 5703165.21 frames. ], giga_tot_loss[loss=0.3749, simple_loss=0.4178, pruned_loss=0.166, over 5656005.76 frames. ], batch size: 213, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:44:34,751 INFO [optim.py:369] (1/2) 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 14:45:11,469 INFO [train.py:968] (1/2) Epoch 3, batch 12300, giga_loss[loss=0.3905, simple_loss=0.4453, pruned_loss=0.1678, over 29024.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.4149, pruned_loss=0.1617, over 5663772.51 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.395, pruned_loss=0.1338, over 5704375.35 frames. ], giga_tot_loss[loss=0.3735, simple_loss=0.4172, pruned_loss=0.1649, over 5666591.40 frames. ], batch size: 155, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:45:15,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:45:19,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:45:45,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:45:57,879 INFO [train.py:968] (1/2) Epoch 3, batch 12350, giga_loss[loss=0.3497, simple_loss=0.4065, pruned_loss=0.1465, over 28656.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4141, pruned_loss=0.1601, over 5653759.22 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3954, pruned_loss=0.1339, over 5706928.22 frames. ], giga_tot_loss[loss=0.3715, simple_loss=0.4162, pruned_loss=0.1635, over 5652382.19 frames. ], batch size: 242, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:46:03,600 INFO [optim.py:369] (1/2) 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 14:46:03,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3825, 2.7377, 1.3336, 1.2731], device='cuda:1'), covar=tensor([0.0927, 0.0508, 0.0911, 0.1427], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0464, 0.0320, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 14:46:37,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102688.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:46:40,351 INFO [train.py:968] (1/2) Epoch 3, batch 12400, giga_loss[loss=0.4086, simple_loss=0.4354, pruned_loss=0.191, over 27558.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4139, pruned_loss=0.159, over 5670441.23 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3954, pruned_loss=0.1337, over 5713626.73 frames. ], giga_tot_loss[loss=0.3707, simple_loss=0.4162, pruned_loss=0.1626, over 5662208.32 frames. ], batch size: 472, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:47:28,552 INFO [train.py:968] (1/2) Epoch 3, batch 12450, giga_loss[loss=0.354, simple_loss=0.41, pruned_loss=0.1489, over 28750.00 frames. ], tot_loss[loss=0.3655, simple_loss=0.4136, pruned_loss=0.1587, over 5685760.27 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3955, pruned_loss=0.1337, over 5717377.53 frames. ], giga_tot_loss[loss=0.3698, simple_loss=0.4156, pruned_loss=0.162, over 5675295.32 frames. ], batch size: 119, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:47:40,987 INFO [optim.py:369] (1/2) 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 14:47:46,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=102755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:48:15,533 INFO [train.py:968] (1/2) Epoch 3, batch 12500, giga_loss[loss=0.4043, simple_loss=0.4352, pruned_loss=0.1867, over 28757.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4137, pruned_loss=0.1599, over 5671297.65 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3957, pruned_loss=0.1339, over 5711767.49 frames. ], giga_tot_loss[loss=0.3705, simple_loss=0.4154, pruned_loss=0.1628, over 5666624.87 frames. ], batch size: 284, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:48:53,101 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102831.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 14:48:55,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102834.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:49:00,841 INFO [train.py:968] (1/2) Epoch 3, batch 12550, giga_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1281, over 28908.00 frames. ], tot_loss[loss=0.3641, simple_loss=0.4113, pruned_loss=0.1585, over 5668347.45 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3956, pruned_loss=0.1339, over 5712079.27 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.4131, pruned_loss=0.1614, over 5663192.68 frames. ], batch size: 145, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:49:09,448 INFO [optim.py:369] (1/2) 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 14:49:21,826 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=102863.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 14:49:47,317 INFO [train.py:968] (1/2) Epoch 3, batch 12600, giga_loss[loss=0.3248, simple_loss=0.3766, pruned_loss=0.1365, over 28865.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4078, pruned_loss=0.1566, over 5681230.88 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3958, pruned_loss=0.1341, over 5717233.57 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.4095, pruned_loss=0.1595, over 5671416.45 frames. ], batch size: 174, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:49:55,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=102898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:49:57,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=102901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:50:24,546 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 12650, giga_loss[loss=0.3914, simple_loss=0.4314, pruned_loss=0.1757, over 28820.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4056, pruned_loss=0.1555, over 5688442.11 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3959, pruned_loss=0.134, over 5720744.84 frames. ], giga_tot_loss[loss=0.3617, simple_loss=0.4071, pruned_loss=0.1581, over 5676966.93 frames. ], batch size: 145, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:50:46,239 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 12700, giga_loss[loss=0.3481, simple_loss=0.4007, pruned_loss=0.1478, over 28873.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4036, pruned_loss=0.1543, over 5696695.50 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3958, pruned_loss=0.1341, over 5724565.37 frames. ], giga_tot_loss[loss=0.3593, simple_loss=0.405, pruned_loss=0.1568, over 5683577.35 frames. ], batch size: 186, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:52:11,044 INFO [train.py:968] (1/2) Epoch 3, batch 12750, giga_loss[loss=0.3743, simple_loss=0.4263, pruned_loss=0.1612, over 29022.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.4039, pruned_loss=0.1534, over 5685172.94 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3959, pruned_loss=0.1342, over 5718545.50 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4051, pruned_loss=0.1557, over 5679548.08 frames. ], batch size: 155, lr: 1.03e-02, grad_scale: 4.0
+2023-03-01 14:52:20,779 INFO [optim.py:369] (1/2) 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 14:52:57,368 INFO [train.py:968] (1/2) Epoch 3, batch 12800, giga_loss[loss=0.2927, simple_loss=0.3686, pruned_loss=0.1084, over 28951.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.4015, pruned_loss=0.1494, over 5684843.51 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3953, pruned_loss=0.1337, over 5724431.73 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4032, pruned_loss=0.1523, over 5674086.40 frames. ], batch size: 145, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:53:48,553 INFO [train.py:968] (1/2) Epoch 3, batch 12850, giga_loss[loss=0.3232, simple_loss=0.3768, pruned_loss=0.1348, over 28676.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3966, pruned_loss=0.1447, over 5676007.03 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3945, pruned_loss=0.1332, over 5727657.29 frames. ], giga_tot_loss[loss=0.347, simple_loss=0.3988, pruned_loss=0.1476, over 5663697.51 frames. ], batch size: 85, lr: 1.03e-02, grad_scale: 8.0
+2023-03-01 14:53:57,166 INFO [optim.py:369] (1/2) 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 14:54:39,011 INFO [train.py:968] (1/2) Epoch 3, batch 12900, giga_loss[loss=0.3212, simple_loss=0.3782, pruned_loss=0.1321, over 27631.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3917, pruned_loss=0.1401, over 5671408.84 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.394, pruned_loss=0.133, over 5731293.12 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3939, pruned_loss=0.1428, over 5657545.44 frames. ], batch size: 472, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:55:32,581 INFO [train.py:968] (1/2) Epoch 3, batch 12950, libri_loss[loss=0.2977, simple_loss=0.3576, pruned_loss=0.1189, over 27692.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3869, pruned_loss=0.1352, over 5674881.58 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3935, pruned_loss=0.1328, over 5731674.91 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3889, pruned_loss=0.1375, over 5662924.93 frames. ], batch size: 61, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:55:35,505 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:55:41,890 INFO [optim.py:369] (1/2) 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 14:55:50,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.79 vs. limit=5.0
+2023-03-01 14:56:10,385 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 14:56:19,240 INFO [train.py:968] (1/2) Epoch 3, batch 13000, libri_loss[loss=0.2736, simple_loss=0.337, pruned_loss=0.1051, over 29560.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3852, pruned_loss=0.1319, over 5664848.31 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3929, pruned_loss=0.1325, over 5725766.13 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3873, pruned_loss=0.1341, over 5659613.66 frames. ], batch size: 77, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:56:31,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 14:57:10,452 INFO [train.py:968] (1/2) Epoch 3, batch 13050, giga_loss[loss=0.2987, simple_loss=0.3711, pruned_loss=0.1132, over 28988.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.385, pruned_loss=0.1317, over 5652247.91 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3922, pruned_loss=0.1321, over 5721027.72 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3872, pruned_loss=0.1338, over 5650664.17 frames. ], batch size: 136, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:57:11,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5791, 1.1642, 2.9077, 2.6635], device='cuda:1'), covar=tensor([0.1526, 0.1784, 0.0479, 0.0613], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0504, 0.0680, 0.0542], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 14:57:19,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3912, 1.7399, 1.1491, 1.5213], device='cuda:1'), covar=tensor([0.0878, 0.0350, 0.0430, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0159, 0.0165, 0.0295], device='cuda:1'), out_proj_covar=tensor([0.0040, 0.0029, 0.0026, 0.0045], device='cuda:1')
+2023-03-01 14:57:21,321 INFO [optim.py:369] (1/2) 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 14:57:21,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4447, 4.1347, 4.1330, 1.8887], device='cuda:1'), covar=tensor([0.0380, 0.0325, 0.0773, 0.1917], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0601, 0.0800, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 14:58:02,121 INFO [train.py:968] (1/2) Epoch 3, batch 13100, libri_loss[loss=0.3087, simple_loss=0.3786, pruned_loss=0.1194, over 29148.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3822, pruned_loss=0.1291, over 5657675.17 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3915, pruned_loss=0.1316, over 5723109.45 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3843, pruned_loss=0.1311, over 5653032.46 frames. ], batch size: 101, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:58:13,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2081, 1.1210, 1.0135, 0.9928], device='cuda:1'), covar=tensor([0.0623, 0.0537, 0.1004, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0476, 0.0528, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 14:58:51,127 INFO [train.py:968] (1/2) Epoch 3, batch 13150, giga_loss[loss=0.2994, simple_loss=0.3674, pruned_loss=0.1157, over 28907.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3786, pruned_loss=0.1266, over 5658228.81 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3902, pruned_loss=0.131, over 5719069.75 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3812, pruned_loss=0.1286, over 5656504.54 frames. ], batch size: 227, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 14:58:58,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0041, 1.2315, 1.0570, 0.9148], device='cuda:1'), covar=tensor([0.2242, 0.2012, 0.1968, 0.1868], device='cuda:1'), in_proj_covar=tensor([0.1014, 0.0802, 0.0910, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 14:58:59,779 INFO [optim.py:369] (1/2) 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 14:59:22,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-01 14:59:39,434 INFO [train.py:968] (1/2) Epoch 3, batch 13200, giga_loss[loss=0.3082, simple_loss=0.3723, pruned_loss=0.1221, over 28659.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3772, pruned_loss=0.1258, over 5661468.55 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3901, pruned_loss=0.131, over 5721644.80 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3791, pruned_loss=0.1274, over 5656952.64 frames. ], batch size: 92, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 14:59:40,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:00:15,798 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:00:27,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0387, 1.8669, 1.4015, 1.5982], device='cuda:1'), covar=tensor([0.0595, 0.0576, 0.0899, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0471, 0.0522, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 15:00:30,120 INFO [train.py:968] (1/2) Epoch 3, batch 13250, giga_loss[loss=0.2877, simple_loss=0.3608, pruned_loss=0.1073, over 28936.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3779, pruned_loss=0.1257, over 5669435.46 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3896, pruned_loss=0.1308, over 5725568.02 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3796, pruned_loss=0.1271, over 5660893.30 frames. ], batch size: 199, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:00:42,500 INFO [optim.py:369] (1/2) 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 15:01:21,510 INFO [train.py:968] (1/2) Epoch 3, batch 13300, giga_loss[loss=0.3193, simple_loss=0.3849, pruned_loss=0.1269, over 28462.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3761, pruned_loss=0.1241, over 5666784.94 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3896, pruned_loss=0.1307, over 5726457.18 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3775, pruned_loss=0.1252, over 5659100.20 frames. ], batch size: 336, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:01:53,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:02:11,932 INFO [train.py:968] (1/2) Epoch 3, batch 13350, giga_loss[loss=0.2585, simple_loss=0.343, pruned_loss=0.08697, over 28943.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3723, pruned_loss=0.1209, over 5672423.98 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3884, pruned_loss=0.1301, over 5731845.72 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3739, pruned_loss=0.1221, over 5659183.17 frames. ], batch size: 164, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:02:23,406 INFO [optim.py:369] (1/2) 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 15:02:27,369 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:03:01,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2798, 1.8109, 1.1294, 0.9187], device='cuda:1'), covar=tensor([0.0736, 0.0411, 0.0493, 0.0657], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.0868, 0.0893, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 15:03:01,537 INFO [train.py:968] (1/2) Epoch 3, batch 13400, giga_loss[loss=0.362, simple_loss=0.3914, pruned_loss=0.1663, over 26764.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3675, pruned_loss=0.1178, over 5672929.07 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3874, pruned_loss=0.1296, over 5736418.19 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.369, pruned_loss=0.1188, over 5656179.43 frames. ], batch size: 555, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:03:51,910 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6430, 3.3680, 3.3970, 1.4973], device='cuda:1'), covar=tensor([0.0621, 0.0523, 0.0995, 0.2162], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0595, 0.0777, 0.0554], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 15:03:55,908 INFO [train.py:968] (1/2) Epoch 3, batch 13450, giga_loss[loss=0.2765, simple_loss=0.3482, pruned_loss=0.1024, over 28877.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.1179, over 5656208.84 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3873, pruned_loss=0.1297, over 5740254.94 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3665, pruned_loss=0.1184, over 5637914.33 frames. ], batch size: 227, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:04:06,213 INFO [optim.py:369] (1/2) 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 15:04:12,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=103759.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 15:04:15,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:18,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:47,423 INFO [train.py:968] (1/2) Epoch 3, batch 13500, giga_loss[loss=0.3795, simple_loss=0.4017, pruned_loss=0.1786, over 26676.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3653, pruned_loss=0.1189, over 5657944.03 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3869, pruned_loss=0.1295, over 5742590.32 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.366, pruned_loss=0.1193, over 5640572.72 frames. ], batch size: 555, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:04:50,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:55,959 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=103798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:04:59,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=103801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:05:28,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6743, 1.5280, 1.2199, 1.3507], device='cuda:1'), covar=tensor([0.0716, 0.0737, 0.0984, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0473, 0.0530, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 15:05:34,680 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=103830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:05:48,182 INFO [train.py:968] (1/2) Epoch 3, batch 13550, giga_loss[loss=0.2767, simple_loss=0.3459, pruned_loss=0.1037, over 29156.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3677, pruned_loss=0.1204, over 5646339.22 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3866, pruned_loss=0.1294, over 5743257.37 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5631764.28 frames. ], batch size: 128, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:06:00,161 INFO [optim.py:369] (1/2) 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 15:06:14,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:06:15,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 15:06:41,765 INFO [train.py:968] (1/2) Epoch 3, batch 13600, libri_loss[loss=0.3092, simple_loss=0.3799, pruned_loss=0.1193, over 28759.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3698, pruned_loss=0.1202, over 5654070.66 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3859, pruned_loss=0.1291, over 5748138.80 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3705, pruned_loss=0.1204, over 5634913.23 frames. ], batch size: 106, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:06:54,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=103900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:07:41,716 INFO [train.py:968] (1/2) Epoch 3, batch 13650, giga_loss[loss=0.318, simple_loss=0.3835, pruned_loss=0.1262, over 29003.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3701, pruned_loss=0.1204, over 5647942.56 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.385, pruned_loss=0.1287, over 5747295.04 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3711, pruned_loss=0.1208, over 5631883.15 frames. ], batch size: 186, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:07:54,797 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 13700, giga_loss[loss=0.2852, simple_loss=0.3586, pruned_loss=0.1059, over 28686.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3689, pruned_loss=0.1198, over 5656086.61 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3843, pruned_loss=0.1284, over 5750425.26 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.37, pruned_loss=0.1201, over 5637147.78 frames. ], batch size: 262, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:08:57,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 15:08:59,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:02,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:40,142 INFO [train.py:968] (1/2) Epoch 3, batch 13750, giga_loss[loss=0.265, simple_loss=0.344, pruned_loss=0.09302, over 28687.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1171, over 5659344.28 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3835, pruned_loss=0.128, over 5752967.29 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3677, pruned_loss=0.1176, over 5640995.59 frames. ], batch size: 85, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:09:40,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:42,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:45,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:09:51,983 INFO [optim.py:369] (1/2) 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 15:10:18,779 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:10:29,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6265, 1.6705, 1.5124, 1.8447], device='cuda:1'), covar=tensor([0.2012, 0.1739, 0.1638, 0.1797], device='cuda:1'), in_proj_covar=tensor([0.1012, 0.0810, 0.0920, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 15:10:36,209 INFO [train.py:968] (1/2) Epoch 3, batch 13800, giga_loss[loss=0.2801, simple_loss=0.3572, pruned_loss=0.1015, over 28979.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3649, pruned_loss=0.1148, over 5657428.66 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.383, pruned_loss=0.1276, over 5752861.05 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3659, pruned_loss=0.1152, over 5640026.40 frames. ], batch size: 145, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:11:28,366 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:11:29,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104134.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 15:11:37,981 INFO [train.py:968] (1/2) Epoch 3, batch 13850, giga_loss[loss=0.2696, simple_loss=0.3316, pruned_loss=0.1038, over 29093.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3625, pruned_loss=0.1143, over 5653506.25 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3824, pruned_loss=0.1273, over 5746890.95 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3632, pruned_loss=0.1144, over 5641227.64 frames. ], batch size: 136, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:11:52,766 INFO [optim.py:369] (1/2) 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 15:11:58,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:12:36,763 INFO [train.py:968] (1/2) Epoch 3, batch 13900, giga_loss[loss=0.2747, simple_loss=0.3489, pruned_loss=0.1003, over 28940.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3625, pruned_loss=0.115, over 5658842.85 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.383, pruned_loss=0.1277, over 5749211.61 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3621, pruned_loss=0.1144, over 5645403.32 frames. ], batch size: 155, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:13:34,763 INFO [train.py:968] (1/2) Epoch 3, batch 13950, giga_loss[loss=0.3595, simple_loss=0.4164, pruned_loss=0.1512, over 28945.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1148, over 5666835.06 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3826, pruned_loss=0.1277, over 5749431.94 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3618, pruned_loss=0.1142, over 5654106.85 frames. ], batch size: 186, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:13:47,671 INFO [optim.py:369] (1/2) 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 15:14:05,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:14:14,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104277.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 15:14:17,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104280.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 15:14:25,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5862, 1.9101, 1.3296, 1.5758], device='cuda:1'), covar=tensor([0.0816, 0.0277, 0.0402, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0158, 0.0165, 0.0294], device='cuda:1'), out_proj_covar=tensor([0.0040, 0.0029, 0.0027, 0.0045], device='cuda:1')
+2023-03-01 15:14:29,417 INFO [train.py:968] (1/2) Epoch 3, batch 14000, giga_loss[loss=0.2918, simple_loss=0.3677, pruned_loss=0.108, over 28754.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.364, pruned_loss=0.1153, over 5671413.56 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3824, pruned_loss=0.1279, over 5747997.44 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3631, pruned_loss=0.1141, over 5659588.39 frames. ], batch size: 243, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:14:34,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5753, 1.4876, 1.2386, 1.2773], device='cuda:1'), covar=tensor([0.0659, 0.0583, 0.1013, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0473, 0.0533, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 15:14:49,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7298, 3.4260, 3.4850, 1.6650], device='cuda:1'), covar=tensor([0.0514, 0.0384, 0.0810, 0.2007], device='cuda:1'), in_proj_covar=tensor([0.0735, 0.0588, 0.0754, 0.0547], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 15:14:51,849 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 14050, giga_loss[loss=0.3131, simple_loss=0.3641, pruned_loss=0.131, over 26769.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.367, pruned_loss=0.1164, over 5675005.17 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3821, pruned_loss=0.1278, over 5743757.64 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.366, pruned_loss=0.115, over 5667227.50 frames. ], batch size: 555, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:15:46,297 INFO [optim.py:369] (1/2) 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 15:16:23,503 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:16:32,848 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:16:33,201 INFO [train.py:968] (1/2) Epoch 3, batch 14100, giga_loss[loss=0.2713, simple_loss=0.3315, pruned_loss=0.1056, over 26846.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3619, pruned_loss=0.1133, over 5677139.28 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3813, pruned_loss=0.1274, over 5747167.05 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3613, pruned_loss=0.1122, over 5666281.90 frames. ], batch size: 555, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:17:40,525 INFO [train.py:968] (1/2) Epoch 3, batch 14150, giga_loss[loss=0.2916, simple_loss=0.3582, pruned_loss=0.1125, over 28128.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3637, pruned_loss=0.1148, over 5667570.74 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.381, pruned_loss=0.1272, over 5740192.29 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3633, pruned_loss=0.1139, over 5665296.34 frames. ], batch size: 412, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:17:41,611 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=104442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:17:56,499 INFO [optim.py:369] (1/2) 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 15:17:57,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5625, 2.3973, 2.1767, 2.2044], device='cuda:1'), covar=tensor([0.0803, 0.1582, 0.1190, 0.1207], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0761, 0.0616, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 15:18:45,836 INFO [train.py:968] (1/2) Epoch 3, batch 14200, giga_loss[loss=0.3548, simple_loss=0.4168, pruned_loss=0.1464, over 28838.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3668, pruned_loss=0.1156, over 5658888.99 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3802, pruned_loss=0.1268, over 5742777.56 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3668, pruned_loss=0.115, over 5652612.04 frames. ], batch size: 213, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:18:59,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2800, 1.4950, 1.4209, 1.4810], device='cuda:1'), covar=tensor([0.1022, 0.1426, 0.1208, 0.1143], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0763, 0.0617, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 15:19:06,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:19:37,375 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 14250, giga_loss[loss=0.3463, simple_loss=0.4105, pruned_loss=0.1411, over 28950.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3697, pruned_loss=0.1153, over 5663107.49 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.38, pruned_loss=0.1267, over 5746770.35 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3694, pruned_loss=0.1146, over 5652175.36 frames. ], batch size: 145, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:20:00,925 INFO [optim.py:369] (1/2) 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 15:20:06,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3335, 1.8227, 1.2902, 1.4820], device='cuda:1'), covar=tensor([0.0824, 0.0279, 0.0366, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0158, 0.0163, 0.0295], device='cuda:1'), out_proj_covar=tensor([0.0041, 0.0030, 0.0026, 0.0045], device='cuda:1')
+2023-03-01 15:20:42,700 INFO [train.py:968] (1/2) Epoch 3, batch 14300, giga_loss[loss=0.2698, simple_loss=0.362, pruned_loss=0.08881, over 28920.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3688, pruned_loss=0.1136, over 5654905.69 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3801, pruned_loss=0.1266, over 5750628.46 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3682, pruned_loss=0.1127, over 5639659.64 frames. ], batch size: 145, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:21:40,136 INFO [train.py:968] (1/2) Epoch 3, batch 14350, libri_loss[loss=0.3076, simple_loss=0.3701, pruned_loss=0.1225, over 29560.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3692, pruned_loss=0.1131, over 5669277.07 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3803, pruned_loss=0.1267, over 5754047.68 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3682, pruned_loss=0.1118, over 5651161.52 frames. ], batch size: 77, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:21:44,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:21:53,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:21:56,093 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:21:58,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-01 15:22:11,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-01 15:22:13,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-01 15:22:23,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:22:25,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6241, 2.1471, 1.7872, 1.8581], device='cuda:1'), covar=tensor([0.1524, 0.1540, 0.1200, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0777, 0.0731, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 15:22:27,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:22:32,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:22:39,728 INFO [train.py:968] (1/2) Epoch 3, batch 14400, giga_loss[loss=0.3115, simple_loss=0.3714, pruned_loss=0.1259, over 28124.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3702, pruned_loss=0.1147, over 5675025.53 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3793, pruned_loss=0.1262, over 5757254.21 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.37, pruned_loss=0.1138, over 5655464.09 frames. ], batch size: 412, lr: 1.02e-02, grad_scale: 8.0
+2023-03-01 15:22:59,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:23:33,309 INFO [train.py:968] (1/2) Epoch 3, batch 14450, libri_loss[loss=0.347, simple_loss=0.396, pruned_loss=0.1491, over 19990.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3678, pruned_loss=0.1145, over 5673266.07 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3783, pruned_loss=0.1256, over 5754331.56 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.368, pruned_loss=0.1137, over 5656011.27 frames. ], batch size: 187, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:23:54,643 INFO [optim.py:369] (1/2) 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 15:23:58,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:08,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:44,273 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:48,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:24:48,462 INFO [train.py:968] (1/2) Epoch 3, batch 14500, libri_loss[loss=0.3069, simple_loss=0.3686, pruned_loss=0.1226, over 29687.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3697, pruned_loss=0.1165, over 5674667.02 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3779, pruned_loss=0.1254, over 5757822.99 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.37, pruned_loss=0.1158, over 5655980.78 frames. ], batch size: 88, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:25:29,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=104817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:25:36,347 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:26:05,894 INFO [train.py:968] (1/2) Epoch 3, batch 14550, giga_loss[loss=0.2581, simple_loss=0.3385, pruned_loss=0.0889, over 29025.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3643, pruned_loss=0.1129, over 5684013.30 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3774, pruned_loss=0.1251, over 5759830.46 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3647, pruned_loss=0.1123, over 5665422.15 frames. ], batch size: 213, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:26:27,443 INFO [optim.py:369] (1/2) 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 15:26:34,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8259, 1.5125, 1.1863, 1.4071], device='cuda:1'), covar=tensor([0.0558, 0.0687, 0.0894, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0466, 0.0528, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 15:27:09,585 INFO [train.py:968] (1/2) Epoch 3, batch 14600, giga_loss[loss=0.2634, simple_loss=0.3433, pruned_loss=0.09176, over 28372.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3623, pruned_loss=0.1117, over 5677598.01 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3772, pruned_loss=0.125, over 5759592.76 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3623, pruned_loss=0.1109, over 5659832.52 frames. ], batch size: 368, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:27:19,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:27:23,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:27:29,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:27:32,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=104911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:28:02,843 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:28:10,071 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 14650, giga_loss[loss=0.2532, simple_loss=0.3239, pruned_loss=0.09124, over 28908.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3596, pruned_loss=0.1107, over 5682183.67 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3769, pruned_loss=0.1249, over 5763243.34 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3594, pruned_loss=0.1097, over 5662862.10 frames. ], batch size: 186, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:28:31,155 INFO [optim.py:369] (1/2) 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 15:28:38,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=104960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:28:41,360 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 14700, giga_loss[loss=0.352, simple_loss=0.4205, pruned_loss=0.1418, over 28913.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3642, pruned_loss=0.1135, over 5683130.12 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3763, pruned_loss=0.1247, over 5756531.65 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3644, pruned_loss=0.1127, over 5672336.41 frames. ], batch size: 174, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:29:10,880 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=104992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:29:16,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4138, 1.6599, 1.0970, 0.8489], device='cuda:1'), covar=tensor([0.0778, 0.0471, 0.0471, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.1204, 0.0882, 0.0909, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 15:30:05,655 INFO [train.py:968] (1/2) Epoch 3, batch 14750, giga_loss[loss=0.3369, simple_loss=0.399, pruned_loss=0.1373, over 28352.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3667, pruned_loss=0.1155, over 5687000.52 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3751, pruned_loss=0.1242, over 5763300.33 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3674, pruned_loss=0.1149, over 5668729.79 frames. ], batch size: 368, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:30:26,536 INFO [optim.py:369] (1/2) 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 15:31:06,009 INFO [train.py:968] (1/2) Epoch 3, batch 14800, giga_loss[loss=0.3031, simple_loss=0.3678, pruned_loss=0.1192, over 28564.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3651, pruned_loss=0.1156, over 5689080.31 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.375, pruned_loss=0.1241, over 5764610.34 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3655, pruned_loss=0.1151, over 5671832.63 frames. ], batch size: 78, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:32:07,100 INFO [train.py:968] (1/2) Epoch 3, batch 14850, libri_loss[loss=0.2512, simple_loss=0.3226, pruned_loss=0.08989, over 29554.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5678247.95 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3746, pruned_loss=0.1239, over 5759234.14 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3678, pruned_loss=0.1175, over 5666859.31 frames. ], batch size: 76, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:32:26,328 INFO [optim.py:369] (1/2) 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 15:33:09,464 INFO [train.py:968] (1/2) Epoch 3, batch 14900, giga_loss[loss=0.3177, simple_loss=0.3831, pruned_loss=0.1262, over 28883.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3669, pruned_loss=0.1172, over 5668234.23 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3746, pruned_loss=0.1239, over 5751952.32 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3672, pruned_loss=0.1167, over 5664331.63 frames. ], batch size: 284, lr: 1.02e-02, grad_scale: 4.0
+2023-03-01 15:34:14,132 INFO [train.py:968] (1/2) Epoch 3, batch 14950, libri_loss[loss=0.2615, simple_loss=0.3308, pruned_loss=0.09614, over 29563.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3684, pruned_loss=0.117, over 5665874.87 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1236, over 5748072.32 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3689, pruned_loss=0.1167, over 5663112.28 frames. ], batch size: 74, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:34:32,573 INFO [optim.py:369] (1/2) 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 15:35:20,248 INFO [train.py:968] (1/2) Epoch 3, batch 15000, giga_loss[loss=0.3303, simple_loss=0.386, pruned_loss=0.1373, over 28473.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3674, pruned_loss=0.1161, over 5670505.31 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3736, pruned_loss=0.1235, over 5753835.41 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.368, pruned_loss=0.1156, over 5660185.39 frames. ], batch size: 336, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:35:20,249 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 15:35:29,188 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 15:36:36,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-01 15:36:43,678 INFO [train.py:968] (1/2) Epoch 3, batch 15050, giga_loss[loss=0.2959, simple_loss=0.3616, pruned_loss=0.115, over 28737.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3639, pruned_loss=0.1144, over 5684528.04 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1236, over 5751926.23 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3641, pruned_loss=0.1138, over 5677654.37 frames. ], batch size: 243, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:36:58,706 INFO [optim.py:369] (1/2) 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 15:37:52,565 INFO [train.py:968] (1/2) Epoch 3, batch 15100, giga_loss[loss=0.2486, simple_loss=0.3218, pruned_loss=0.08775, over 28705.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.357, pruned_loss=0.1112, over 5688198.23 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3736, pruned_loss=0.1236, over 5755881.87 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1105, over 5677898.92 frames. ], batch size: 262, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:38:10,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9780, 1.1785, 4.0312, 3.1045], device='cuda:1'), covar=tensor([0.1513, 0.1894, 0.0324, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0498, 0.0649, 0.0511], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 15:38:39,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0312, 1.2232, 0.9060, 0.3294], device='cuda:1'), covar=tensor([0.0867, 0.0857, 0.1340, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.1254, 0.1211, 0.1288, 0.1076], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 15:38:46,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9028, 3.6187, 3.6380, 1.6274], device='cuda:1'), covar=tensor([0.0505, 0.0439, 0.0823, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0594, 0.0754, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 15:38:49,592 INFO [train.py:968] (1/2) Epoch 3, batch 15150, libri_loss[loss=0.3853, simple_loss=0.425, pruned_loss=0.1729, over 29663.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3573, pruned_loss=0.112, over 5688789.55 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3737, pruned_loss=0.1238, over 5759036.35 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3567, pruned_loss=0.1109, over 5675252.67 frames. ], batch size: 88, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:39:04,642 INFO [optim.py:369] (1/2) 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 15:39:24,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9439, 1.2315, 4.3130, 3.3829], device='cuda:1'), covar=tensor([0.1581, 0.2057, 0.0272, 0.0619], device='cuda:1'), in_proj_covar=tensor([0.0522, 0.0506, 0.0658, 0.0517], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 15:39:38,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2179, 1.9048, 1.7092, 1.7779], device='cuda:1'), covar=tensor([0.0630, 0.0698, 0.0852, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0479, 0.0538, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 15:39:42,747 INFO [train.py:968] (1/2) Epoch 3, batch 15200, giga_loss[loss=0.2914, simple_loss=0.3623, pruned_loss=0.1103, over 28819.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3584, pruned_loss=0.1136, over 5684902.11 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1232, over 5762263.43 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3581, pruned_loss=0.1128, over 5668542.19 frames. ], batch size: 227, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:39:59,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.09 vs. limit=2.0
+2023-03-01 15:40:34,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-01 15:40:37,988 INFO [train.py:968] (1/2) Epoch 3, batch 15250, giga_loss[loss=0.2665, simple_loss=0.3446, pruned_loss=0.09416, over 28384.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3578, pruned_loss=0.1131, over 5669683.12 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5756721.51 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3578, pruned_loss=0.1125, over 5660086.46 frames. ], batch size: 369, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:40:58,848 INFO [optim.py:369] (1/2) 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 15:41:40,682 INFO [train.py:968] (1/2) Epoch 3, batch 15300, giga_loss[loss=0.2555, simple_loss=0.3339, pruned_loss=0.0885, over 28557.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3556, pruned_loss=0.1102, over 5671193.33 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3723, pruned_loss=0.123, over 5757629.12 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3554, pruned_loss=0.1096, over 5661998.03 frames. ], batch size: 307, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:41:41,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4211, 4.5873, 2.1973, 2.2347], device='cuda:1'), covar=tensor([0.0658, 0.0169, 0.0702, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0445, 0.0323, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 15:42:24,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4175, 1.3347, 1.4620, 1.3342], device='cuda:1'), covar=tensor([0.0709, 0.1079, 0.1068, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0753, 0.0605, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 15:42:43,041 INFO [train.py:968] (1/2) Epoch 3, batch 15350, libri_loss[loss=0.3575, simple_loss=0.4067, pruned_loss=0.1542, over 19455.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3539, pruned_loss=0.1095, over 5655930.90 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1232, over 5751285.27 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3531, pruned_loss=0.1085, over 5653220.48 frames. ], batch size: 187, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:43:06,029 INFO [optim.py:369] (1/2) 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 15:43:51,963 INFO [train.py:968] (1/2) Epoch 3, batch 15400, libri_loss[loss=0.3083, simple_loss=0.3677, pruned_loss=0.1244, over 29531.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.354, pruned_loss=0.1094, over 5673714.01 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3716, pruned_loss=0.1226, over 5755502.44 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3536, pruned_loss=0.1088, over 5665237.28 frames. ], batch size: 81, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:44:55,774 INFO [train.py:968] (1/2) Epoch 3, batch 15450, giga_loss[loss=0.3489, simple_loss=0.3895, pruned_loss=0.1542, over 26834.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3544, pruned_loss=0.1089, over 5684177.36 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3716, pruned_loss=0.1227, over 5757024.17 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3539, pruned_loss=0.1081, over 5675234.79 frames. ], batch size: 555, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:45:13,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7259, 1.8843, 1.3959, 1.1714], device='cuda:1'), covar=tensor([0.0679, 0.0498, 0.0436, 0.0558], device='cuda:1'), in_proj_covar=tensor([0.1188, 0.0881, 0.0895, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 15:45:14,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2395, 1.2988, 1.2202, 1.5023], device='cuda:1'), covar=tensor([0.2144, 0.1904, 0.1753, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.1013, 0.0808, 0.0922, 0.0934], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 15:45:19,063 INFO [optim.py:369] (1/2) 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 15:45:33,667 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-01 15:45:49,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=105781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:45:58,581 INFO [train.py:968] (1/2) Epoch 3, batch 15500, giga_loss[loss=0.3075, simple_loss=0.3699, pruned_loss=0.1225, over 28892.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3564, pruned_loss=0.1109, over 5688467.56 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1227, over 5760032.47 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3558, pruned_loss=0.11, over 5677234.78 frames. ], batch size: 213, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:46:18,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8725, 1.2511, 3.7160, 2.9925], device='cuda:1'), covar=tensor([0.1546, 0.1911, 0.0316, 0.0696], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0486, 0.0641, 0.0504], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0006], device='cuda:1')
+2023-03-01 15:46:54,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2980, 2.1485, 1.3112, 1.3033], device='cuda:1'), covar=tensor([0.0858, 0.0502, 0.0818, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0438, 0.0316, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 15:47:05,521 INFO [train.py:968] (1/2) Epoch 3, batch 15550, giga_loss[loss=0.2835, simple_loss=0.3261, pruned_loss=0.1205, over 24381.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.355, pruned_loss=0.1101, over 5675940.88 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1228, over 5752139.55 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3543, pruned_loss=0.1092, over 5673670.69 frames. ], batch size: 705, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:47:25,288 INFO [optim.py:369] (1/2) 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 15:47:35,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-01 15:47:46,968 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 15600, giga_loss[loss=0.2681, simple_loss=0.3213, pruned_loss=0.1075, over 24221.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3552, pruned_loss=0.1086, over 5661565.45 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3718, pruned_loss=0.1229, over 5748067.15 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3538, pruned_loss=0.1073, over 5660687.47 frames. ], batch size: 705, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:48:04,728 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=105895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:48:45,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-01 15:48:57,453 INFO [train.py:968] (1/2) Epoch 3, batch 15650, giga_loss[loss=0.2609, simple_loss=0.3506, pruned_loss=0.08559, over 28772.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.357, pruned_loss=0.1084, over 5665262.13 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3712, pruned_loss=0.1225, over 5749831.29 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3559, pruned_loss=0.1073, over 5660380.31 frames. ], batch size: 174, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:49:18,204 INFO [optim.py:369] (1/2) 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 15:49:58,093 INFO [train.py:968] (1/2) Epoch 3, batch 15700, giga_loss[loss=0.3567, simple_loss=0.417, pruned_loss=0.1482, over 28900.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.36, pruned_loss=0.1106, over 5658171.21 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3712, pruned_loss=0.1224, over 5749228.19 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.359, pruned_loss=0.1095, over 5653554.85 frames. ], batch size: 145, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:50:25,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-01 15:50:58,306 INFO [train.py:968] (1/2) Epoch 3, batch 15750, giga_loss[loss=0.2718, simple_loss=0.3462, pruned_loss=0.09869, over 28955.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3612, pruned_loss=0.1116, over 5655814.45 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3711, pruned_loss=0.1223, over 5750759.05 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3603, pruned_loss=0.1107, over 5649208.98 frames. ], batch size: 164, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:51:05,799 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.22 vs. limit=5.0
+2023-03-01 15:51:18,088 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 15800, giga_loss[loss=0.3088, simple_loss=0.3711, pruned_loss=0.1232, over 27634.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3596, pruned_loss=0.111, over 5660243.12 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5753323.34 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3591, pruned_loss=0.1103, over 5648587.90 frames. ], batch size: 472, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:52:57,919 INFO [train.py:968] (1/2) Epoch 3, batch 15850, libri_loss[loss=0.2745, simple_loss=0.3455, pruned_loss=0.1017, over 29547.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3564, pruned_loss=0.109, over 5659559.58 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1214, over 5756093.98 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3565, pruned_loss=0.1085, over 5645601.92 frames. ], batch size: 79, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:53:16,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:53:18,679 INFO [optim.py:369] (1/2) 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 15:53:48,475 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:53:55,102 INFO [train.py:968] (1/2) Epoch 3, batch 15900, giga_loss[loss=0.2625, simple_loss=0.3308, pruned_loss=0.09711, over 29000.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.356, pruned_loss=0.1093, over 5670950.86 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3697, pruned_loss=0.1215, over 5757516.54 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3554, pruned_loss=0.1084, over 5655657.74 frames. ], batch size: 186, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:53:56,712 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:54:58,720 INFO [train.py:968] (1/2) Epoch 3, batch 15950, giga_loss[loss=0.2882, simple_loss=0.3618, pruned_loss=0.1072, over 28098.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3562, pruned_loss=0.1094, over 5675472.25 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1213, over 5760446.44 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3556, pruned_loss=0.1087, over 5659109.04 frames. ], batch size: 412, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:55:16,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7886, 4.5510, 4.4876, 1.8075], device='cuda:1'), covar=tensor([0.0409, 0.0319, 0.0621, 0.2057], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0596, 0.0758, 0.0551], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 15:55:17,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:55:22,448 INFO [optim.py:369] (1/2) 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 15:55:27,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9455, 1.2309, 0.9520, 0.2026], device='cuda:1'), covar=tensor([0.1094, 0.1048, 0.1498, 0.1868], device='cuda:1'), in_proj_covar=tensor([0.1245, 0.1211, 0.1269, 0.1073], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 15:55:39,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:55:55,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7304, 1.6125, 1.1941, 1.3439], device='cuda:1'), covar=tensor([0.0566, 0.0549, 0.0870, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0475, 0.0532, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 15:55:58,627 INFO [train.py:968] (1/2) Epoch 3, batch 16000, giga_loss[loss=0.2973, simple_loss=0.3663, pruned_loss=0.1142, over 28325.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3591, pruned_loss=0.1111, over 5681210.13 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3694, pruned_loss=0.1211, over 5761463.78 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3585, pruned_loss=0.1104, over 5664582.82 frames. ], batch size: 368, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:56:11,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:56:14,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:56:14,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6001, 1.6606, 1.5063, 1.9980], device='cuda:1'), covar=tensor([0.2118, 0.1888, 0.1660, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.0997, 0.0785, 0.0905, 0.0925], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 15:56:50,124 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:56:54,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4278, 1.6863, 1.2885, 1.4203], device='cuda:1'), covar=tensor([0.0890, 0.0347, 0.0387, 0.1005], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0158, 0.0163, 0.0292], device='cuda:1'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:1')
+2023-03-01 15:57:02,598 INFO [train.py:968] (1/2) Epoch 3, batch 16050, giga_loss[loss=0.3233, simple_loss=0.3872, pruned_loss=0.1297, over 28526.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.1121, over 5670265.11 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5762881.39 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3591, pruned_loss=0.1116, over 5652390.34 frames. ], batch size: 336, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 15:57:12,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6349, 1.4662, 1.3932, 1.2868], device='cuda:1'), covar=tensor([0.0605, 0.0481, 0.0690, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0480, 0.0529, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 15:57:21,406 INFO [optim.py:369] (1/2) 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 15:58:01,190 INFO [train.py:968] (1/2) Epoch 3, batch 16100, giga_loss[loss=0.3471, simple_loss=0.4107, pruned_loss=0.1417, over 27976.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3615, pruned_loss=0.1132, over 5676190.43 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5767386.32 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3607, pruned_loss=0.1125, over 5654703.97 frames. ], batch size: 412, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:58:08,020 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:10,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:25,133 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:31,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:47,215 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:58:59,824 INFO [train.py:968] (1/2) Epoch 3, batch 16150, giga_loss[loss=0.2997, simple_loss=0.3747, pruned_loss=0.1124, over 28923.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3641, pruned_loss=0.1145, over 5660629.54 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3684, pruned_loss=0.1203, over 5767951.72 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3639, pruned_loss=0.1141, over 5642092.09 frames. ], batch size: 213, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 15:59:06,258 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 15:59:06,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6010, 2.1422, 1.7638, 1.7517], device='cuda:1'), covar=tensor([0.1581, 0.1596, 0.1238, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0753, 0.0714, 0.0623], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 15:59:20,023 INFO [optim.py:369] (1/2) 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 15:59:32,734 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 16200, giga_loss[loss=0.3221, simple_loss=0.3854, pruned_loss=0.1294, over 28634.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3644, pruned_loss=0.1135, over 5663383.50 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 5771226.48 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3649, pruned_loss=0.1136, over 5643001.86 frames. ], batch size: 262, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:00:31,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4491, 2.9833, 1.4196, 1.3038], device='cuda:1'), covar=tensor([0.1113, 0.0509, 0.1129, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0445, 0.0319, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:1')
+2023-03-01 16:00:36,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4597, 1.4771, 1.3055, 2.0863], device='cuda:1'), covar=tensor([0.2197, 0.2039, 0.1874, 0.2013], device='cuda:1'), in_proj_covar=tensor([0.0995, 0.0794, 0.0906, 0.0921], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 16:01:03,432 INFO [train.py:968] (1/2) Epoch 3, batch 16250, giga_loss[loss=0.2774, simple_loss=0.3494, pruned_loss=0.1028, over 29168.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3647, pruned_loss=0.1146, over 5663200.91 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3672, pruned_loss=0.1195, over 5775006.15 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3653, pruned_loss=0.1146, over 5638898.89 frames. ], batch size: 113, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:01:27,108 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:01:37,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:01:59,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2758, 1.6089, 1.2588, 1.3933], device='cuda:1'), covar=tensor([0.0817, 0.0402, 0.0396, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0156, 0.0163, 0.0288], device='cuda:1'), out_proj_covar=tensor([0.0040, 0.0030, 0.0027, 0.0045], device='cuda:1')
+2023-03-01 16:02:07,019 INFO [train.py:968] (1/2) Epoch 3, batch 16300, giga_loss[loss=0.3043, simple_loss=0.3715, pruned_loss=0.1185, over 28744.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.362, pruned_loss=0.1132, over 5666045.92 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3669, pruned_loss=0.1193, over 5766956.31 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3627, pruned_loss=0.1132, over 5651400.36 frames. ], batch size: 243, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:03:12,713 INFO [train.py:968] (1/2) Epoch 3, batch 16350, giga_loss[loss=0.2641, simple_loss=0.3385, pruned_loss=0.09483, over 28986.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3606, pruned_loss=0.1119, over 5663670.10 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3663, pruned_loss=0.1191, over 5758132.68 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3615, pruned_loss=0.112, over 5658733.88 frames. ], batch size: 213, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:03:36,823 INFO [optim.py:369] (1/2) 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 16:03:39,627 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:19,494 INFO [train.py:968] (1/2) Epoch 3, batch 16400, giga_loss[loss=0.2861, simple_loss=0.3544, pruned_loss=0.1089, over 28466.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3603, pruned_loss=0.1126, over 5659244.58 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3661, pruned_loss=0.119, over 5759739.07 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3612, pruned_loss=0.1128, over 5653292.86 frames. ], batch size: 336, lr: 1.01e-02, grad_scale: 8.0
+2023-03-01 16:04:34,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:38,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:44,452 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=106710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:47,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:04:57,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 16:05:05,514 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:05:15,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:05:21,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-01 16:05:21,775 INFO [train.py:968] (1/2) Epoch 3, batch 16450, giga_loss[loss=0.2617, simple_loss=0.3408, pruned_loss=0.09124, over 28798.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3578, pruned_loss=0.1116, over 5658274.05 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1188, over 5760686.37 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5650905.49 frames. ], batch size: 243, lr: 1.01e-02, grad_scale: 4.0
+2023-03-01 16:05:23,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=106742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:05:46,316 INFO [optim.py:369] (1/2) 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 16:06:19,187 INFO [train.py:968] (1/2) Epoch 3, batch 16500, giga_loss[loss=0.2702, simple_loss=0.3514, pruned_loss=0.09453, over 28944.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3578, pruned_loss=0.111, over 5659637.82 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3655, pruned_loss=0.1185, over 5761593.52 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3585, pruned_loss=0.1112, over 5650359.42 frames. ], batch size: 199, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:06:32,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3326, 1.4348, 1.3769, 1.3715], device='cuda:1'), covar=tensor([0.0923, 0.1359, 0.1239, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0764, 0.0620, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 16:07:21,332 INFO [train.py:968] (1/2) Epoch 3, batch 16550, giga_loss[loss=0.2784, simple_loss=0.3448, pruned_loss=0.106, over 27604.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3571, pruned_loss=0.109, over 5668317.79 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3654, pruned_loss=0.1183, over 5762879.65 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3577, pruned_loss=0.1092, over 5659159.31 frames. ], batch size: 472, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:07:27,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=106845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:07:29,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=106847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:07:44,786 INFO [optim.py:369] (1/2) 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 16:08:17,569 INFO [train.py:968] (1/2) Epoch 3, batch 16600, giga_loss[loss=0.3042, simple_loss=0.3804, pruned_loss=0.114, over 29011.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3591, pruned_loss=0.1073, over 5685240.88 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1184, over 5766341.14 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3592, pruned_loss=0.1072, over 5672980.60 frames. ], batch size: 285, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:09:12,023 INFO [train.py:968] (1/2) Epoch 3, batch 16650, giga_loss[loss=0.3724, simple_loss=0.413, pruned_loss=0.1659, over 27598.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3623, pruned_loss=0.1097, over 5682816.96 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.366, pruned_loss=0.1187, over 5768496.75 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3619, pruned_loss=0.109, over 5668359.56 frames. ], batch size: 472, lr: 1.01e-02, grad_scale: 1.0
+2023-03-01 16:09:33,619 INFO [optim.py:369] (1/2) 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 16:10:05,121 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 16700, giga_loss[loss=0.2927, simple_loss=0.3636, pruned_loss=0.1109, over 29000.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3616, pruned_loss=0.1095, over 5675537.83 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1183, over 5764363.85 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3616, pruned_loss=0.1089, over 5664217.06 frames. ], batch size: 199, lr: 1.01e-02, grad_scale: 1.0
+2023-03-01 16:10:08,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=106991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:10:53,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:11:14,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:11:20,637 INFO [train.py:968] (1/2) Epoch 3, batch 16750, giga_loss[loss=0.2559, simple_loss=0.3141, pruned_loss=0.09887, over 24895.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.361, pruned_loss=0.1093, over 5664521.27 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3654, pruned_loss=0.1182, over 5765106.03 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.361, pruned_loss=0.1089, over 5654530.76 frames. ], batch size: 705, lr: 1.01e-02, grad_scale: 1.0
+2023-03-01 16:11:30,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1732, 1.2791, 0.8893, 0.8836], device='cuda:1'), covar=tensor([0.0582, 0.0466, 0.0396, 0.0483], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.0883, 0.0897, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 16:11:46,979 INFO [optim.py:369] (1/2) 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 16:12:26,305 INFO [train.py:968] (1/2) Epoch 3, batch 16800, giga_loss[loss=0.2697, simple_loss=0.3509, pruned_loss=0.09425, over 28357.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3606, pruned_loss=0.1087, over 5661883.43 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1182, over 5763934.89 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3604, pruned_loss=0.1081, over 5652294.40 frames. ], batch size: 368, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:12:43,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:13:35,264 INFO [train.py:968] (1/2) Epoch 3, batch 16850, giga_loss[loss=0.2798, simple_loss=0.3647, pruned_loss=0.09743, over 28773.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3611, pruned_loss=0.1085, over 5660743.51 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1183, over 5766086.90 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3607, pruned_loss=0.1078, over 5649455.02 frames. ], batch size: 307, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:14:05,130 INFO [optim.py:369] (1/2) 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 16:14:18,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4921, 2.7769, 1.4613, 1.5089], device='cuda:1'), covar=tensor([0.0708, 0.0426, 0.0751, 0.1090], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0438, 0.0313, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:1')
+2023-03-01 16:14:25,343 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:14:28,219 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 16900, giga_loss[loss=0.2919, simple_loss=0.3737, pruned_loss=0.105, over 29001.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3634, pruned_loss=0.11, over 5668657.83 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.365, pruned_loss=0.1181, over 5770154.39 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3636, pruned_loss=0.1094, over 5651897.79 frames. ], batch size: 145, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:15:01,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:15:04,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8058, 3.5023, 3.5046, 1.6592], device='cuda:1'), covar=tensor([0.0536, 0.0416, 0.0939, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0588, 0.0752, 0.0547], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 16:15:09,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5262, 1.3733, 1.2099, 1.0956], device='cuda:1'), covar=tensor([0.0519, 0.0470, 0.0784, 0.0778], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0457, 0.0517, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 16:15:16,709 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 16950, giga_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.1231, over 27611.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.366, pruned_loss=0.1114, over 5673932.48 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1184, over 5773104.22 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3657, pruned_loss=0.1103, over 5654828.69 frames. ], batch size: 472, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:15:49,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:15:52,621 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:16:34,495 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:16:34,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4094, 1.4568, 1.2183, 1.3848], device='cuda:1'), covar=tensor([0.0962, 0.1554, 0.1477, 0.1377], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0757, 0.0603, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 16:16:42,770 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-01 16:16:49,630 INFO [train.py:968] (1/2) Epoch 3, batch 17000, libri_loss[loss=0.2509, simple_loss=0.3221, pruned_loss=0.08988, over 29433.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3646, pruned_loss=0.1107, over 5686679.70 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.118, over 5777000.56 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3647, pruned_loss=0.1101, over 5665381.73 frames. ], batch size: 67, lr: 1.01e-02, grad_scale: 2.0
+2023-03-01 16:17:07,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-01 16:17:09,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5878, 1.1529, 2.9063, 2.7436], device='cuda:1'), covar=tensor([0.1513, 0.1885, 0.0510, 0.0564], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0506, 0.0674, 0.0527], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 16:17:45,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9228, 1.2493, 4.2664, 3.2249], device='cuda:1'), covar=tensor([0.1616, 0.2101, 0.0355, 0.0508], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0509, 0.0681, 0.0533], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 16:17:57,731 INFO [train.py:968] (1/2) Epoch 3, batch 17050, giga_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 28997.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3629, pruned_loss=0.1108, over 5689190.88 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3651, pruned_loss=0.118, over 5779951.78 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.363, pruned_loss=0.1099, over 5666272.21 frames. ], batch size: 155, lr: 1.00e-02, grad_scale: 2.0
+2023-03-01 16:18:27,542 INFO [optim.py:369] (1/2) 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 16:18:30,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=107365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:18:33,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=107368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:18:48,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-01 16:19:09,323 INFO [train.py:968] (1/2) Epoch 3, batch 17100, giga_loss[loss=0.2792, simple_loss=0.3634, pruned_loss=0.09755, over 28871.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3604, pruned_loss=0.1085, over 5685962.48 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.365, pruned_loss=0.118, over 5781893.09 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3604, pruned_loss=0.1077, over 5664974.12 frames. ], batch size: 174, lr: 1.00e-02, grad_scale: 2.0
+2023-03-01 16:19:20,113 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=107397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:20:14,760 INFO [train.py:968] (1/2) Epoch 3, batch 17150, libri_loss[loss=0.2997, simple_loss=0.3635, pruned_loss=0.118, over 29538.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3595, pruned_loss=0.1077, over 5684747.50 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3652, pruned_loss=0.1181, over 5782206.64 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3592, pruned_loss=0.1068, over 5665464.26 frames. ], batch size: 77, lr: 1.00e-02, grad_scale: 2.0
+2023-03-01 16:20:19,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-01 16:20:39,165 INFO [optim.py:369] (1/2) 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 16:21:12,539 INFO [train.py:968] (1/2) Epoch 3, batch 17200, giga_loss[loss=0.346, simple_loss=0.4103, pruned_loss=0.1409, over 28645.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3627, pruned_loss=0.1099, over 5681073.44 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3655, pruned_loss=0.1182, over 5783292.70 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3622, pruned_loss=0.1089, over 5663196.28 frames. ], batch size: 307, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:22:02,926 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107533.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:22:11,431 INFO [train.py:968] (1/2) Epoch 3, batch 17250, giga_loss[loss=0.3011, simple_loss=0.3709, pruned_loss=0.1157, over 28381.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3651, pruned_loss=0.1113, over 5678040.49 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3653, pruned_loss=0.1181, over 5782463.98 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3648, pruned_loss=0.1105, over 5663851.87 frames. ], batch size: 368, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:22:17,492 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:22:20,900 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-01 16:22:29,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 16:22:35,989 INFO [optim.py:369] (1/2) 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 16:22:40,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 16:23:03,510 INFO [train.py:968] (1/2) Epoch 3, batch 17300, giga_loss[loss=0.2651, simple_loss=0.3445, pruned_loss=0.09284, over 28486.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3625, pruned_loss=0.1108, over 5680052.70 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.118, over 5783409.15 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3623, pruned_loss=0.1101, over 5664876.19 frames. ], batch size: 71, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:23:34,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-01 16:23:40,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-01 16:24:02,950 INFO [train.py:968] (1/2) Epoch 3, batch 17350, giga_loss[loss=0.2649, simple_loss=0.3449, pruned_loss=0.09251, over 28738.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3612, pruned_loss=0.1111, over 5668805.45 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.1181, over 5782868.08 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3607, pruned_loss=0.1103, over 5655293.63 frames. ], batch size: 174, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:24:27,890 INFO [optim.py:369] (1/2) 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 16:24:57,761 INFO [train.py:968] (1/2) Epoch 3, batch 17400, giga_loss[loss=0.2973, simple_loss=0.3634, pruned_loss=0.1156, over 28100.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3611, pruned_loss=0.1121, over 5667680.06 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3649, pruned_loss=0.1178, over 5787757.79 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3611, pruned_loss=0.1115, over 5648100.84 frames. ], batch size: 412, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:25:24,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3560, 1.9727, 1.3035, 1.5209], device='cuda:1'), covar=tensor([0.0910, 0.0337, 0.0402, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0245, 0.0157, 0.0164, 0.0291], device='cuda:1'), out_proj_covar=tensor([0.0041, 0.0030, 0.0027, 0.0046], device='cuda:1')
+2023-03-01 16:25:49,723 INFO [train.py:968] (1/2) Epoch 3, batch 17450, giga_loss[loss=0.3447, simple_loss=0.4077, pruned_loss=0.1408, over 28574.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3712, pruned_loss=0.1192, over 5674882.05 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3653, pruned_loss=0.118, over 5788504.93 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3709, pruned_loss=0.1185, over 5655043.40 frames. ], batch size: 336, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:26:07,662 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 17500, giga_loss[loss=0.355, simple_loss=0.4156, pruned_loss=0.1472, over 28896.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3818, pruned_loss=0.126, over 5680683.08 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3653, pruned_loss=0.1179, over 5790366.08 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3817, pruned_loss=0.1256, over 5661514.92 frames. ], batch size: 213, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:26:36,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-01 16:27:18,696 INFO [train.py:968] (1/2) Epoch 3, batch 17550, giga_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1269, over 28682.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3848, pruned_loss=0.1295, over 5677321.24 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3653, pruned_loss=0.1178, over 5791632.94 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3851, pruned_loss=0.1295, over 5658751.81 frames. ], batch size: 262, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:27:37,865 INFO [optim.py:369] (1/2) 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 16:28:02,691 INFO [train.py:968] (1/2) Epoch 3, batch 17600, giga_loss[loss=0.2659, simple_loss=0.3291, pruned_loss=0.1013, over 28596.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3799, pruned_loss=0.1274, over 5681726.96 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1175, over 5791566.12 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3808, pruned_loss=0.1279, over 5664011.95 frames. ], batch size: 85, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:28:11,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=107903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:28:15,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107908.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:28:24,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=107920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:28:36,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 16:28:46,080 INFO [train.py:968] (1/2) Epoch 3, batch 17650, giga_loss[loss=0.267, simple_loss=0.338, pruned_loss=0.09799, over 28903.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3721, pruned_loss=0.1235, over 5694506.15 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3654, pruned_loss=0.1177, over 5794576.67 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.1239, over 5675275.75 frames. ], batch size: 186, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:29:05,343 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 17700, giga_loss[loss=0.2457, simple_loss=0.3161, pruned_loss=0.08765, over 29005.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.364, pruned_loss=0.1193, over 5701310.62 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.366, pruned_loss=0.118, over 5796522.37 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.364, pruned_loss=0.1194, over 5682514.07 frames. ], batch size: 164, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:30:12,520 INFO [train.py:968] (1/2) Epoch 3, batch 17750, giga_loss[loss=0.2192, simple_loss=0.2865, pruned_loss=0.07597, over 28768.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3566, pruned_loss=0.1162, over 5701019.25 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3667, pruned_loss=0.1184, over 5797987.33 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3559, pruned_loss=0.1159, over 5682534.17 frames. ], batch size: 99, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:30:20,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108051.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:30:20,525 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:30:21,947 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108054.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:30:30,087 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:30:30,464 INFO [optim.py:369] (1/2) 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 16:30:32,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:30:47,855 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 17800, libri_loss[loss=0.4276, simple_loss=0.4479, pruned_loss=0.2037, over 19023.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3509, pruned_loss=0.1134, over 5694088.47 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3676, pruned_loss=0.1191, over 5791494.68 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3489, pruned_loss=0.1124, over 5683004.59 frames. ], batch size: 187, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:30:57,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:31:04,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.23 vs. limit=2.0
+2023-03-01 16:31:27,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:31:34,703 INFO [train.py:968] (1/2) Epoch 3, batch 17850, giga_loss[loss=0.2661, simple_loss=0.3265, pruned_loss=0.1029, over 27948.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3469, pruned_loss=0.1116, over 5687835.05 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1194, over 5783942.05 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3444, pruned_loss=0.1104, over 5683995.20 frames. ], batch size: 412, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:31:52,988 INFO [optim.py:369] (1/2) 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 16:32:02,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6568, 2.3333, 1.6356, 0.8856], device='cuda:1'), covar=tensor([0.1683, 0.0932, 0.1540, 0.1818], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1183, 0.1247, 0.1061], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 16:32:15,338 INFO [train.py:968] (1/2) Epoch 3, batch 17900, giga_loss[loss=0.2546, simple_loss=0.3269, pruned_loss=0.09119, over 29006.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3445, pruned_loss=0.1099, over 5701593.85 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 5788280.96 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3409, pruned_loss=0.1084, over 5691256.61 frames. ], batch size: 155, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:32:55,873 INFO [train.py:968] (1/2) Epoch 3, batch 17950, giga_loss[loss=0.2775, simple_loss=0.3318, pruned_loss=0.1116, over 28537.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3408, pruned_loss=0.1083, over 5694318.56 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3693, pruned_loss=0.1199, over 5786385.07 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3371, pruned_loss=0.1066, over 5686288.37 frames. ], batch size: 336, lr: 1.00e-02, grad_scale: 4.0
+2023-03-01 16:33:16,876 INFO [optim.py:369] (1/2) 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 16:33:28,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:33:37,494 INFO [train.py:968] (1/2) Epoch 3, batch 18000, giga_loss[loss=0.2486, simple_loss=0.3105, pruned_loss=0.09335, over 27676.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3371, pruned_loss=0.1058, over 5692816.22 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 5781312.42 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.333, pruned_loss=0.104, over 5688402.44 frames. ], batch size: 472, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:33:37,494 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 16:33:45,890 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 16:34:26,710 INFO [train.py:968] (1/2) Epoch 3, batch 18050, giga_loss[loss=0.2392, simple_loss=0.304, pruned_loss=0.08717, over 28817.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.334, pruned_loss=0.1039, over 5701451.87 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3698, pruned_loss=0.1201, over 5785327.94 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3295, pruned_loss=0.1018, over 5691778.01 frames. ], batch size: 186, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:34:45,737 INFO [optim.py:369] (1/2) 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 16:35:03,688 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 16:35:06,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6740, 1.1619, 3.2943, 2.8412], device='cuda:1'), covar=tensor([0.1500, 0.1961, 0.0394, 0.0728], device='cuda:1'), in_proj_covar=tensor([0.0524, 0.0500, 0.0670, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 16:35:07,615 INFO [train.py:968] (1/2) Epoch 3, batch 18100, giga_loss[loss=0.3088, simple_loss=0.358, pruned_loss=0.1298, over 27886.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3332, pruned_loss=0.1034, over 5697523.37 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3702, pruned_loss=0.1199, over 5788391.74 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3272, pruned_loss=0.1011, over 5682657.94 frames. ], batch size: 412, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:35:34,295 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:35:37,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:35:39,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:35:50,730 INFO [train.py:968] (1/2) Epoch 3, batch 18150, giga_loss[loss=0.195, simple_loss=0.2706, pruned_loss=0.05967, over 28314.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3288, pruned_loss=0.1007, over 5696442.61 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3706, pruned_loss=0.12, over 5788901.07 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3235, pruned_loss=0.09862, over 5683480.08 frames. ], batch size: 65, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:36:03,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:36:12,523 INFO [optim.py:369] (1/2) 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 16:36:39,411 INFO [train.py:968] (1/2) Epoch 3, batch 18200, giga_loss[loss=0.2454, simple_loss=0.3074, pruned_loss=0.09171, over 28653.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3249, pruned_loss=0.09855, over 5707747.15 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3709, pruned_loss=0.1202, over 5790549.40 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3198, pruned_loss=0.09647, over 5694905.90 frames. ], batch size: 262, lr: 1.00e-02, grad_scale: 8.0
+2023-03-01 16:36:50,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=108505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:37:22,363 INFO [train.py:968] (1/2) Epoch 3, batch 18250, giga_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 28698.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3253, pruned_loss=0.0993, over 5704462.68 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3709, pruned_loss=0.1201, over 5790501.80 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3206, pruned_loss=0.0974, over 5693272.15 frames. ], batch size: 262, lr: 9.99e-03, grad_scale: 8.0
+2023-03-01 16:37:48,916 INFO [optim.py:369] (1/2) 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 16:37:53,766 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:37:55,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:38:12,180 INFO [train.py:968] (1/2) Epoch 3, batch 18300, giga_loss[loss=0.3584, simple_loss=0.4119, pruned_loss=0.1525, over 28775.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3392, pruned_loss=0.1077, over 5703293.43 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3714, pruned_loss=0.1205, over 5791053.93 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3346, pruned_loss=0.1057, over 5692840.52 frames. ], batch size: 112, lr: 9.99e-03, grad_scale: 4.0
+2023-03-01 16:38:23,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:38:49,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 16:38:56,971 INFO [train.py:968] (1/2) Epoch 3, batch 18350, giga_loss[loss=0.3329, simple_loss=0.387, pruned_loss=0.1394, over 28615.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3557, pruned_loss=0.118, over 5702601.77 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3715, pruned_loss=0.1205, over 5793493.98 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3515, pruned_loss=0.1163, over 5690514.56 frames. ], batch size: 85, lr: 9.99e-03, grad_scale: 4.0
+2023-03-01 16:39:03,838 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=108648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:39:05,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=108651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:39:17,356 INFO [optim.py:369] (1/2) 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 16:39:29,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1470, 1.7114, 1.7355, 1.7275], device='cuda:1'), covar=tensor([0.0833, 0.1674, 0.1324, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0780, 0.0624, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 16:39:29,674 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=108680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:39:39,891 INFO [train.py:968] (1/2) Epoch 3, batch 18400, giga_loss[loss=0.3208, simple_loss=0.3878, pruned_loss=0.1269, over 28860.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3671, pruned_loss=0.124, over 5697789.33 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3723, pruned_loss=0.1209, over 5785035.59 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3628, pruned_loss=0.1223, over 5693496.88 frames. ], batch size: 199, lr: 9.99e-03, grad_scale: 8.0
+2023-03-01 16:40:02,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:40:22,884 INFO [train.py:968] (1/2) Epoch 3, batch 18450, giga_loss[loss=0.2844, simple_loss=0.3622, pruned_loss=0.1033, over 28697.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3735, pruned_loss=0.1258, over 5700505.40 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3735, pruned_loss=0.1214, over 5787205.25 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3689, pruned_loss=0.1242, over 5691946.42 frames. ], batch size: 92, lr: 9.99e-03, grad_scale: 8.0
+2023-03-01 16:40:38,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6219, 1.5347, 1.2600, 1.2736], device='cuda:1'), covar=tensor([0.0636, 0.0641, 0.0890, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0476, 0.0522, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 16:40:43,153 INFO [optim.py:369] (1/2) 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 16:40:51,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.01 vs. limit=5.0
+2023-03-01 16:41:03,867 INFO [train.py:968] (1/2) Epoch 3, batch 18500, giga_loss[loss=0.3483, simple_loss=0.4063, pruned_loss=0.1452, over 28817.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3763, pruned_loss=0.1261, over 5701270.94 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3731, pruned_loss=0.1211, over 5789526.77 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.373, pruned_loss=0.1251, over 5690859.87 frames. ], batch size: 199, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:41:24,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1603, 1.2088, 1.2465, 1.1961], device='cuda:1'), covar=tensor([0.0849, 0.1004, 0.1315, 0.0929], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0775, 0.0628, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 16:41:36,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:41:50,814 INFO [train.py:968] (1/2) Epoch 3, batch 18550, giga_loss[loss=0.3417, simple_loss=0.3956, pruned_loss=0.1439, over 28527.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.377, pruned_loss=0.1258, over 5696008.68 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3734, pruned_loss=0.1211, over 5789167.69 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3742, pruned_loss=0.1252, over 5685293.15 frames. ], batch size: 336, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:41:58,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2659, 1.5003, 4.8014, 3.4289], device='cuda:1'), covar=tensor([0.1509, 0.1896, 0.0228, 0.0521], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0492, 0.0650, 0.0522], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 16:42:13,108 INFO [optim.py:369] (1/2) 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 16:42:31,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-01 16:42:34,356 INFO [train.py:968] (1/2) Epoch 3, batch 18600, giga_loss[loss=0.3064, simple_loss=0.3733, pruned_loss=0.1197, over 29014.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.38, pruned_loss=0.1285, over 5698042.45 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.374, pruned_loss=0.1216, over 5791053.61 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3773, pruned_loss=0.1276, over 5685489.94 frames. ], batch size: 136, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:42:38,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5869, 1.2270, 3.6716, 2.8508], device='cuda:1'), covar=tensor([0.1619, 0.1892, 0.0309, 0.0618], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0494, 0.0655, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 16:43:10,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3891, 2.4807, 1.3665, 1.3146], device='cuda:1'), covar=tensor([0.0838, 0.0413, 0.0800, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0436, 0.0310, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:1')
+2023-03-01 16:43:17,876 INFO [train.py:968] (1/2) Epoch 3, batch 18650, giga_loss[loss=0.3042, simple_loss=0.3696, pruned_loss=0.1194, over 28822.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3823, pruned_loss=0.1301, over 5703903.71 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.374, pruned_loss=0.1216, over 5791885.79 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3802, pruned_loss=0.1296, over 5691224.45 frames. ], batch size: 99, lr: 9.98e-03, grad_scale: 4.0
+2023-03-01 16:43:23,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 16:43:39,292 INFO [optim.py:369] (1/2) 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 16:43:44,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=108970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:44:01,853 INFO [train.py:968] (1/2) Epoch 3, batch 18700, giga_loss[loss=0.3267, simple_loss=0.3944, pruned_loss=0.1295, over 28859.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3862, pruned_loss=0.1323, over 5702829.78 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.375, pruned_loss=0.122, over 5793157.00 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3839, pruned_loss=0.1318, over 5690123.38 frames. ], batch size: 199, lr: 9.97e-03, grad_scale: 4.0
+2023-03-01 16:44:40,810 INFO [train.py:968] (1/2) Epoch 3, batch 18750, giga_loss[loss=0.3071, simple_loss=0.3738, pruned_loss=0.1202, over 28562.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3888, pruned_loss=0.1332, over 5711366.08 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3753, pruned_loss=0.1223, over 5795739.44 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.387, pruned_loss=0.1327, over 5696551.31 frames. ], batch size: 92, lr: 9.97e-03, grad_scale: 4.0
+2023-03-01 16:45:03,327 INFO [optim.py:369] (1/2) 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 16:45:23,535 INFO [train.py:968] (1/2) Epoch 3, batch 18800, giga_loss[loss=0.308, simple_loss=0.3811, pruned_loss=0.1174, over 28723.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3897, pruned_loss=0.1328, over 5708831.27 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3757, pruned_loss=0.1225, over 5795254.38 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3881, pruned_loss=0.1324, over 5696953.57 frames. ], batch size: 284, lr: 9.97e-03, grad_scale: 8.0
+2023-03-01 16:45:25,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:46:05,545 INFO [train.py:968] (1/2) Epoch 3, batch 18850, giga_loss[loss=0.3287, simple_loss=0.396, pruned_loss=0.1307, over 28615.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3904, pruned_loss=0.1322, over 5706318.96 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.376, pruned_loss=0.1226, over 5796872.21 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.389, pruned_loss=0.1318, over 5694171.24 frames. ], batch size: 336, lr: 9.97e-03, grad_scale: 8.0
+2023-03-01 16:46:24,350 INFO [optim.py:369] (1/2) 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 16:46:26,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6767, 1.7401, 1.0786, 1.4443], device='cuda:1'), covar=tensor([0.0795, 0.0818, 0.1629, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0473, 0.0521, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 16:46:40,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 18900, giga_loss[loss=0.3147, simple_loss=0.3871, pruned_loss=0.1212, over 28858.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3894, pruned_loss=0.1307, over 5701459.45 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3761, pruned_loss=0.1227, over 5797449.85 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3883, pruned_loss=0.1304, over 5691139.81 frames. ], batch size: 174, lr: 9.96e-03, grad_scale: 8.0
+2023-03-01 16:46:54,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:47:19,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0977, 1.8902, 1.7607, 2.0020], device='cuda:1'), covar=tensor([0.0880, 0.1731, 0.1392, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0776, 0.0625, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 16:47:22,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:47:24,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:47:25,576 INFO [train.py:968] (1/2) Epoch 3, batch 18950, giga_loss[loss=0.2997, simple_loss=0.3662, pruned_loss=0.1166, over 29041.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3865, pruned_loss=0.1276, over 5713963.54 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3766, pruned_loss=0.1228, over 5798814.59 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3855, pruned_loss=0.1274, over 5701989.83 frames. ], batch size: 128, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:47:46,506 INFO [optim.py:369] (1/2) 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 16:47:47,387 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:48:04,171 INFO [train.py:968] (1/2) Epoch 3, batch 19000, giga_loss[loss=0.3649, simple_loss=0.421, pruned_loss=0.1544, over 29013.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3871, pruned_loss=0.1282, over 5711007.50 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3769, pruned_loss=0.123, over 5801118.52 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3863, pruned_loss=0.1281, over 5695922.19 frames. ], batch size: 128, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:48:48,512 INFO [train.py:968] (1/2) Epoch 3, batch 19050, giga_loss[loss=0.353, simple_loss=0.4037, pruned_loss=0.1512, over 28678.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.39, pruned_loss=0.1335, over 5689868.35 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3775, pruned_loss=0.1234, over 5792187.40 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3891, pruned_loss=0.1331, over 5684668.33 frames. ], batch size: 262, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:48:50,677 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:48:51,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:48:52,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:49:11,822 INFO [optim.py:369] (1/2) 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 16:49:17,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2882, 1.8855, 1.4636, 0.6407], device='cuda:1'), covar=tensor([0.1469, 0.0803, 0.1124, 0.1727], device='cuda:1'), in_proj_covar=tensor([0.1259, 0.1190, 0.1273, 0.1075], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 16:49:19,003 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:49:31,556 INFO [train.py:968] (1/2) Epoch 3, batch 19100, giga_loss[loss=0.3899, simple_loss=0.4195, pruned_loss=0.1801, over 28434.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3925, pruned_loss=0.1375, over 5690302.66 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3776, pruned_loss=0.1235, over 5793269.31 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3918, pruned_loss=0.1372, over 5684416.93 frames. ], batch size: 85, lr: 9.96e-03, grad_scale: 4.0
+2023-03-01 16:50:02,218 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109430.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:50:10,185 INFO [train.py:968] (1/2) Epoch 3, batch 19150, giga_loss[loss=0.3803, simple_loss=0.4148, pruned_loss=0.1729, over 28438.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.393, pruned_loss=0.1392, over 5693035.37 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3784, pruned_loss=0.1243, over 5788606.67 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3923, pruned_loss=0.1388, over 5688329.01 frames. ], batch size: 60, lr: 9.95e-03, grad_scale: 4.0
+2023-03-01 16:50:30,577 INFO [optim.py:369] (1/2) 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 16:50:47,958 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:50:50,864 INFO [train.py:968] (1/2) Epoch 3, batch 19200, libri_loss[loss=0.3337, simple_loss=0.3943, pruned_loss=0.1366, over 28681.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3907, pruned_loss=0.1381, over 5693208.60 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3791, pruned_loss=0.125, over 5781376.19 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.39, pruned_loss=0.1377, over 5692414.69 frames. ], batch size: 106, lr: 9.95e-03, grad_scale: 8.0
+2023-03-01 16:50:51,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:51:12,972 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:51:31,531 INFO [train.py:968] (1/2) Epoch 3, batch 19250, libri_loss[loss=0.341, simple_loss=0.4073, pruned_loss=0.1373, over 26076.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.39, pruned_loss=0.1371, over 5689935.47 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3801, pruned_loss=0.1256, over 5781838.98 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3888, pruned_loss=0.1366, over 5686213.57 frames. ], batch size: 136, lr: 9.95e-03, grad_scale: 4.0
+2023-03-01 16:51:47,772 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:52:10,235 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:52:12,817 INFO [train.py:968] (1/2) Epoch 3, batch 19300, libri_loss[loss=0.3796, simple_loss=0.4418, pruned_loss=0.1588, over 29164.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3884, pruned_loss=0.135, over 5690559.51 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3806, pruned_loss=0.1258, over 5782840.84 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.387, pruned_loss=0.1345, over 5685823.23 frames. ], batch size: 97, lr: 9.95e-03, grad_scale: 4.0
+2023-03-01 16:52:57,593 INFO [train.py:968] (1/2) Epoch 3, batch 19350, libri_loss[loss=0.2822, simple_loss=0.3456, pruned_loss=0.1094, over 29365.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3849, pruned_loss=0.1322, over 5686575.04 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3805, pruned_loss=0.1257, over 5781334.92 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.384, pruned_loss=0.1319, over 5682535.81 frames. ], batch size: 71, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:53:20,400 INFO [optim.py:369] (1/2) 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 16:53:37,036 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:53:39,629 INFO [train.py:968] (1/2) Epoch 3, batch 19400, libri_loss[loss=0.3063, simple_loss=0.3845, pruned_loss=0.1141, over 29231.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3786, pruned_loss=0.1278, over 5679462.63 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3811, pruned_loss=0.1259, over 5773754.19 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1275, over 5679985.56 frames. ], batch size: 94, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:53:52,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:53:54,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:54:02,207 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:54:05,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5279, 2.8303, 1.4837, 1.3598], device='cuda:1'), covar=tensor([0.0846, 0.0336, 0.0849, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0431, 0.0305, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 16:54:13,880 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-01 16:54:21,776 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:54:27,398 INFO [train.py:968] (1/2) Epoch 3, batch 19450, giga_loss[loss=0.2413, simple_loss=0.3181, pruned_loss=0.08227, over 28908.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.123, over 5682477.08 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3817, pruned_loss=0.1263, over 5774896.51 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3694, pruned_loss=0.1224, over 5680604.86 frames. ], batch size: 227, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:54:52,378 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 19500, giga_loss[loss=0.2653, simple_loss=0.3364, pruned_loss=0.09712, over 28649.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3664, pruned_loss=0.12, over 5682693.03 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3822, pruned_loss=0.1265, over 5770730.91 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3639, pruned_loss=0.119, over 5681261.82 frames. ], batch size: 242, lr: 9.94e-03, grad_scale: 4.0
+2023-03-01 16:55:30,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109805.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:56:01,583 INFO [train.py:968] (1/2) Epoch 3, batch 19550, giga_loss[loss=0.2998, simple_loss=0.3693, pruned_loss=0.1152, over 28291.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3674, pruned_loss=0.1208, over 5675032.56 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3826, pruned_loss=0.1269, over 5762267.43 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3648, pruned_loss=0.1196, over 5680755.99 frames. ], batch size: 368, lr: 9.94e-03, grad_scale: 2.0
+2023-03-01 16:56:26,467 INFO [optim.py:369] (1/2) 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 16:56:33,276 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9057, 3.6500, 1.8365, 1.7090], device='cuda:1'), covar=tensor([0.0788, 0.0293, 0.0806, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0436, 0.0309, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 16:56:45,383 INFO [train.py:968] (1/2) Epoch 3, batch 19600, giga_loss[loss=0.2957, simple_loss=0.3559, pruned_loss=0.1178, over 28299.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.367, pruned_loss=0.12, over 5691155.16 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3829, pruned_loss=0.1269, over 5760847.67 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3644, pruned_loss=0.1189, over 5695246.90 frames. ], batch size: 77, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:56:47,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3631, 1.3432, 1.1964, 1.4496], device='cuda:1'), covar=tensor([0.2011, 0.1991, 0.1802, 0.2085], device='cuda:1'), in_proj_covar=tensor([0.1014, 0.0819, 0.0909, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 16:57:29,518 INFO [train.py:968] (1/2) Epoch 3, batch 19650, giga_loss[loss=0.2461, simple_loss=0.3168, pruned_loss=0.08767, over 28613.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3652, pruned_loss=0.1193, over 5696025.84 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3833, pruned_loss=0.1271, over 5762130.57 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3626, pruned_loss=0.1182, over 5697422.21 frames. ], batch size: 60, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:57:35,396 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=109948.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:57:37,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=109951.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 16:57:47,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=109962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:57:52,909 INFO [optim.py:369] (1/2) 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 16:57:55,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=109972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:58:01,045 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=109980.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 16:58:09,733 INFO [train.py:968] (1/2) Epoch 3, batch 19700, giga_loss[loss=0.3101, simple_loss=0.3637, pruned_loss=0.1283, over 28786.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3615, pruned_loss=0.1171, over 5708237.09 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3841, pruned_loss=0.1275, over 5763534.78 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3586, pruned_loss=0.1158, over 5707229.01 frames. ], batch size: 99, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:58:47,601 INFO [train.py:968] (1/2) Epoch 3, batch 19750, giga_loss[loss=0.2665, simple_loss=0.3333, pruned_loss=0.09985, over 29004.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3604, pruned_loss=0.1167, over 5716854.31 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3855, pruned_loss=0.1284, over 5765141.48 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.356, pruned_loss=0.1145, over 5712905.54 frames. ], batch size: 106, lr: 9.93e-03, grad_scale: 4.0
+2023-03-01 16:58:54,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8711, 3.5743, 3.5886, 1.7190], device='cuda:1'), covar=tensor([0.0533, 0.0403, 0.0706, 0.2191], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0581, 0.0750, 0.0554], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 16:58:57,420 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:05,394 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:59:20,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6634, 1.9762, 1.8485, 1.7503], device='cuda:1'), covar=tensor([0.1354, 0.1711, 0.1073, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0782, 0.0735, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 16:59:25,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:28,523 INFO [train.py:968] (1/2) Epoch 3, batch 19800, giga_loss[loss=0.2421, simple_loss=0.3151, pruned_loss=0.0846, over 28896.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3577, pruned_loss=0.1158, over 5709545.04 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.386, pruned_loss=0.1288, over 5757367.56 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3533, pruned_loss=0.1135, over 5711496.11 frames. ], batch size: 174, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 16:59:39,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 16:59:40,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 16:59:44,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:00:06,675 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:00:10,036 INFO [train.py:968] (1/2) Epoch 3, batch 19850, giga_loss[loss=0.3601, simple_loss=0.402, pruned_loss=0.1591, over 27421.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3559, pruned_loss=0.115, over 5719155.90 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3865, pruned_loss=0.129, over 5759984.17 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3514, pruned_loss=0.1127, over 5717495.12 frames. ], batch size: 472, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 17:00:32,624 INFO [optim.py:369] (1/2) 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 17:00:48,843 INFO [train.py:968] (1/2) Epoch 3, batch 19900, libri_loss[loss=0.3511, simple_loss=0.4224, pruned_loss=0.1399, over 29386.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3551, pruned_loss=0.1146, over 5719968.35 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3877, pruned_loss=0.1296, over 5760626.04 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3493, pruned_loss=0.1118, over 5716198.46 frames. ], batch size: 92, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 17:00:59,784 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:02,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:18,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:20,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:24,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:27,145 INFO [train.py:968] (1/2) Epoch 3, batch 19950, giga_loss[loss=0.2376, simple_loss=0.31, pruned_loss=0.08266, over 28797.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3535, pruned_loss=0.1137, over 5709662.22 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3887, pruned_loss=0.1302, over 5751542.43 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.347, pruned_loss=0.1104, over 5713725.81 frames. ], batch size: 66, lr: 9.92e-03, grad_scale: 2.0
+2023-03-01 17:01:32,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-01 17:01:44,681 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:01:50,399 INFO [optim.py:369] (1/2) 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 17:02:06,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6523, 1.1510, 3.3806, 2.7758], device='cuda:1'), covar=tensor([0.1625, 0.2033, 0.0382, 0.0602], device='cuda:1'), in_proj_covar=tensor([0.0522, 0.0499, 0.0664, 0.0531], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 17:02:08,860 INFO [train.py:968] (1/2) Epoch 3, batch 20000, giga_loss[loss=0.2347, simple_loss=0.3151, pruned_loss=0.07717, over 28932.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3532, pruned_loss=0.1126, over 5723494.02 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3899, pruned_loss=0.1305, over 5758277.28 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3453, pruned_loss=0.109, over 5719143.55 frames. ], batch size: 174, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:02:28,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1133, 1.2764, 1.1234, 0.6797], device='cuda:1'), covar=tensor([0.0816, 0.0526, 0.0487, 0.0660], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.0907, 0.0945, 0.1027], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 17:02:30,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-01 17:02:47,220 INFO [train.py:968] (1/2) Epoch 3, batch 20050, giga_loss[loss=0.2881, simple_loss=0.3482, pruned_loss=0.114, over 28152.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3514, pruned_loss=0.1117, over 5727348.89 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3905, pruned_loss=0.1306, over 5760262.03 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3436, pruned_loss=0.1082, over 5721260.99 frames. ], batch size: 77, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:02:51,615 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:03:07,522 INFO [optim.py:369] (1/2) 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 17:03:22,511 INFO [train.py:968] (1/2) Epoch 3, batch 20100, giga_loss[loss=0.2739, simple_loss=0.3435, pruned_loss=0.1022, over 28951.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3518, pruned_loss=0.1115, over 5729810.88 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3907, pruned_loss=0.1305, over 5755066.44 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3433, pruned_loss=0.1078, over 5728032.69 frames. ], batch size: 227, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:03:42,125 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:03:53,949 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:04:01,468 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:04:03,917 INFO [train.py:968] (1/2) Epoch 3, batch 20150, libri_loss[loss=0.3128, simple_loss=0.3831, pruned_loss=0.1212, over 29555.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3555, pruned_loss=0.114, over 5720420.36 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3915, pruned_loss=0.1309, over 5748668.74 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3468, pruned_loss=0.1101, over 5724201.92 frames. ], batch size: 79, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:04:29,318 INFO [optim.py:369] (1/2) 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 17:04:51,972 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 20200, giga_loss[loss=0.3115, simple_loss=0.375, pruned_loss=0.124, over 29016.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3615, pruned_loss=0.1185, over 5714802.06 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3918, pruned_loss=0.131, over 5750289.98 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.354, pruned_loss=0.1151, over 5715850.17 frames. ], batch size: 128, lr: 9.91e-03, grad_scale: 4.0
+2023-03-01 17:04:53,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:03,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:07,579 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:21,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:05:38,064 INFO [train.py:968] (1/2) Epoch 3, batch 20250, giga_loss[loss=0.3464, simple_loss=0.3996, pruned_loss=0.1466, over 28851.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.372, pruned_loss=0.1264, over 5702126.67 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3919, pruned_loss=0.131, over 5755073.20 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3648, pruned_loss=0.1234, over 5696914.43 frames. ], batch size: 227, lr: 9.90e-03, grad_scale: 4.0
+2023-03-01 17:06:06,450 INFO [optim.py:369] (1/2) 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 17:06:06,813 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:06:08,642 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 20300, giga_loss[loss=0.4327, simple_loss=0.4596, pruned_loss=0.2029, over 27871.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3785, pruned_loss=0.1305, over 5696550.73 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3919, pruned_loss=0.1309, over 5758544.83 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3724, pruned_loss=0.1282, over 5687691.70 frames. ], batch size: 412, lr: 9.90e-03, grad_scale: 4.0
+2023-03-01 17:06:33,321 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 20350, giga_loss[loss=0.3161, simple_loss=0.3853, pruned_loss=0.1235, over 28909.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3838, pruned_loss=0.1329, over 5686563.92 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3927, pruned_loss=0.1314, over 5758104.88 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3781, pruned_loss=0.1305, over 5678925.29 frames. ], batch size: 106, lr: 9.90e-03, grad_scale: 4.0
+2023-03-01 17:07:36,839 INFO [optim.py:369] (1/2) 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 17:07:44,216 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:07:51,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2668, 1.9815, 1.5059, 0.5405], device='cuda:1'), covar=tensor([0.1643, 0.0911, 0.1539, 0.1986], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.1171, 0.1268, 0.1077], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 17:07:57,194 INFO [train.py:968] (1/2) Epoch 3, batch 20400, giga_loss[loss=0.3542, simple_loss=0.408, pruned_loss=0.1502, over 28936.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3893, pruned_loss=0.1364, over 5686568.88 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3925, pruned_loss=0.1313, over 5760383.48 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3847, pruned_loss=0.1347, over 5677033.39 frames. ], batch size: 186, lr: 9.90e-03, grad_scale: 8.0
+2023-03-01 17:08:02,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:08:15,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=110711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:08:40,140 INFO [train.py:968] (1/2) Epoch 3, batch 20450, giga_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.1242, over 28612.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3943, pruned_loss=0.1403, over 5680195.90 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3926, pruned_loss=0.1315, over 5760821.97 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3906, pruned_loss=0.1389, over 5671235.27 frames. ], batch size: 85, lr: 9.89e-03, grad_scale: 8.0
+2023-03-01 17:09:05,939 INFO [optim.py:369] (1/2) 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 17:09:20,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:09:24,992 INFO [train.py:968] (1/2) Epoch 3, batch 20500, giga_loss[loss=0.2843, simple_loss=0.3549, pruned_loss=0.1068, over 28709.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3868, pruned_loss=0.1348, over 5676644.99 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3927, pruned_loss=0.1317, over 5755972.93 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3838, pruned_loss=0.1336, over 5672730.98 frames. ], batch size: 262, lr: 9.89e-03, grad_scale: 8.0
+2023-03-01 17:09:42,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:10:05,175 INFO [train.py:968] (1/2) Epoch 3, batch 20550, giga_loss[loss=0.3287, simple_loss=0.3837, pruned_loss=0.1368, over 28759.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3823, pruned_loss=0.1304, over 5690324.82 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3931, pruned_loss=0.1321, over 5757896.22 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3792, pruned_loss=0.1291, over 5683814.80 frames. ], batch size: 92, lr: 9.89e-03, grad_scale: 8.0
+2023-03-01 17:10:31,117 INFO [optim.py:369] (1/2) 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 17:10:38,137 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:10:42,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=110883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:10:48,828 INFO [train.py:968] (1/2) Epoch 3, batch 20600, giga_loss[loss=0.3135, simple_loss=0.3845, pruned_loss=0.1213, over 29014.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3824, pruned_loss=0.1301, over 5692072.92 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3938, pruned_loss=0.1327, over 5756973.13 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3791, pruned_loss=0.1284, over 5685412.62 frames. ], batch size: 155, lr: 9.89e-03, grad_scale: 2.0
+2023-03-01 17:11:19,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:21,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:28,667 INFO [train.py:968] (1/2) Epoch 3, batch 20650, giga_loss[loss=0.3547, simple_loss=0.4131, pruned_loss=0.1482, over 28945.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3848, pruned_loss=0.1312, over 5687152.98 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3946, pruned_loss=0.1335, over 5752805.60 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3809, pruned_loss=0.129, over 5683020.27 frames. ], batch size: 186, lr: 9.89e-03, grad_scale: 2.0
+2023-03-01 17:11:39,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=110955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:42,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=110958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:43,346 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=110960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:11:55,268 INFO [optim.py:369] (1/2) 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 17:12:09,305 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 20700, giga_loss[loss=0.3172, simple_loss=0.3811, pruned_loss=0.1267, over 28562.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3873, pruned_loss=0.1331, over 5692517.73 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.395, pruned_loss=0.1337, over 5754400.61 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3837, pruned_loss=0.1312, over 5686994.59 frames. ], batch size: 60, lr: 9.88e-03, grad_scale: 2.0
+2023-03-01 17:12:17,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 17:12:22,808 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:38,649 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:41,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:43,326 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:12:45,080 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 20750, libri_loss[loss=0.4015, simple_loss=0.4389, pruned_loss=0.182, over 29536.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3892, pruned_loss=0.1346, over 5694696.59 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3958, pruned_loss=0.1343, over 5745894.81 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3854, pruned_loss=0.1325, over 5696599.87 frames. ], batch size: 79, lr: 9.88e-03, grad_scale: 2.0
+2023-03-01 17:12:59,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0205, 1.2788, 4.6806, 3.4746], device='cuda:1'), covar=tensor([0.1557, 0.1902, 0.0230, 0.0485], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0493, 0.0669, 0.0539], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 17:13:02,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:03,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:09,085 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:13,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7303, 2.2649, 1.2439, 0.9693], device='cuda:1'), covar=tensor([0.0868, 0.0554, 0.0692, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.1178, 0.0901, 0.0946, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 17:13:15,067 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4682, 2.1699, 1.6467, 0.7147], device='cuda:1'), covar=tensor([0.1784, 0.0898, 0.1496, 0.2083], device='cuda:1'), in_proj_covar=tensor([0.1239, 0.1173, 0.1265, 0.1065], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 17:13:19,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:20,252 INFO [optim.py:369] (1/2) 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 17:13:33,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:13:37,747 INFO [train.py:968] (1/2) Epoch 3, batch 20800, giga_loss[loss=0.3955, simple_loss=0.4214, pruned_loss=0.1848, over 23854.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3911, pruned_loss=0.1367, over 5681651.52 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3965, pruned_loss=0.1348, over 5749349.85 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3872, pruned_loss=0.1345, over 5678223.60 frames. ], batch size: 705, lr: 9.88e-03, grad_scale: 4.0
+2023-03-01 17:13:42,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-01 17:14:09,732 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:14:19,306 INFO [train.py:968] (1/2) Epoch 3, batch 20850, giga_loss[loss=0.3387, simple_loss=0.3906, pruned_loss=0.1434, over 28925.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.392, pruned_loss=0.1377, over 5680425.12 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3968, pruned_loss=0.1352, over 5744699.56 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3884, pruned_loss=0.1357, over 5680240.46 frames. ], batch size: 112, lr: 9.88e-03, grad_scale: 4.0
+2023-03-01 17:14:43,832 INFO [optim.py:369] (1/2) 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 17:14:59,013 INFO [train.py:968] (1/2) Epoch 3, batch 20900, giga_loss[loss=0.286, simple_loss=0.3595, pruned_loss=0.1063, over 28465.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3912, pruned_loss=0.1367, over 5682656.55 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.397, pruned_loss=0.1353, over 5736536.36 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3882, pruned_loss=0.135, over 5689413.20 frames. ], batch size: 65, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:15:01,331 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:03,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:16,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:18,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:18,867 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 17:15:26,408 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:28,174 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:30,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:38,105 INFO [train.py:968] (1/2) Epoch 3, batch 20950, giga_loss[loss=0.3178, simple_loss=0.3799, pruned_loss=0.1278, over 27634.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3918, pruned_loss=0.1363, over 5686952.64 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3975, pruned_loss=0.1358, over 5738217.36 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3888, pruned_loss=0.1345, over 5689916.96 frames. ], batch size: 472, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:15:42,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:56,361 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:15:56,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:16:04,083 INFO [optim.py:369] (1/2) 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 17:16:18,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1515, 2.8369, 2.8569, 1.3866], device='cuda:1'), covar=tensor([0.0739, 0.0553, 0.0976, 0.2203], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0598, 0.0765, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 17:16:18,818 INFO [train.py:968] (1/2) Epoch 3, batch 21000, giga_loss[loss=0.3295, simple_loss=0.3981, pruned_loss=0.1304, over 28932.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3914, pruned_loss=0.1342, over 5694235.09 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3978, pruned_loss=0.1361, over 5740482.05 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3887, pruned_loss=0.1325, over 5693858.95 frames. ], batch size: 174, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:16:18,819 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 17:16:27,500 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 17:16:41,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6393, 2.8540, 1.4188, 1.3154], device='cuda:1'), covar=tensor([0.0814, 0.0336, 0.0872, 0.1359], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0438, 0.0304, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 17:17:06,265 INFO [train.py:968] (1/2) Epoch 3, batch 21050, giga_loss[loss=0.3696, simple_loss=0.4197, pruned_loss=0.1597, over 28768.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3897, pruned_loss=0.1332, over 5687163.90 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.398, pruned_loss=0.1365, over 5733085.55 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3873, pruned_loss=0.1315, over 5692966.49 frames. ], batch size: 242, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:17:13,957 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:17:36,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:17:44,656 INFO [train.py:968] (1/2) Epoch 3, batch 21100, giga_loss[loss=0.2908, simple_loss=0.3556, pruned_loss=0.113, over 28905.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3866, pruned_loss=0.131, over 5691910.63 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3985, pruned_loss=0.137, over 5724723.84 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.384, pruned_loss=0.1291, over 5702083.79 frames. ], batch size: 186, lr: 9.87e-03, grad_scale: 4.0
+2023-03-01 17:17:59,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9237, 3.9479, 1.8999, 1.6729], device='cuda:1'), covar=tensor([0.0807, 0.0236, 0.0787, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0439, 0.0304, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 17:18:23,443 INFO [train.py:968] (1/2) Epoch 3, batch 21150, giga_loss[loss=0.2846, simple_loss=0.353, pruned_loss=0.1081, over 28435.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.385, pruned_loss=0.13, over 5698086.86 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3993, pruned_loss=0.1377, over 5725693.99 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3819, pruned_loss=0.1277, over 5704895.63 frames. ], batch size: 65, lr: 9.86e-03, grad_scale: 4.0
+2023-03-01 17:18:48,614 INFO [optim.py:369] (1/2) 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 17:19:04,087 INFO [train.py:968] (1/2) Epoch 3, batch 21200, giga_loss[loss=0.3104, simple_loss=0.3761, pruned_loss=0.1223, over 28865.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3824, pruned_loss=0.1286, over 5702582.69 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3994, pruned_loss=0.1378, over 5728614.88 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3797, pruned_loss=0.1266, over 5705034.79 frames. ], batch size: 199, lr: 9.86e-03, grad_scale: 8.0
+2023-03-01 17:19:17,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:19:32,609 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:19:35,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:19:43,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5085, 1.1129, 3.3940, 2.7705], device='cuda:1'), covar=tensor([0.1697, 0.1981, 0.0365, 0.0640], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0488, 0.0668, 0.0542], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 17:19:46,308 INFO [train.py:968] (1/2) Epoch 3, batch 21250, giga_loss[loss=0.2728, simple_loss=0.3516, pruned_loss=0.09704, over 28492.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3836, pruned_loss=0.1299, over 5702609.03 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3992, pruned_loss=0.1378, over 5728254.44 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3814, pruned_loss=0.1282, over 5704205.30 frames. ], batch size: 71, lr: 9.86e-03, grad_scale: 8.0
+2023-03-01 17:19:53,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5387, 4.8348, 5.2139, 2.1367], device='cuda:1'), covar=tensor([0.0370, 0.0298, 0.0692, 0.1956], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0592, 0.0763, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 17:19:57,638 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:20:05,784 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 17:20:10,575 INFO [optim.py:369] (1/2) 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 17:20:14,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6463, 1.9464, 1.7846, 1.6735], device='cuda:1'), covar=tensor([0.1458, 0.1784, 0.1198, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0794, 0.0739, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 17:20:26,105 INFO [train.py:968] (1/2) Epoch 3, batch 21300, libri_loss[loss=0.3656, simple_loss=0.4149, pruned_loss=0.1582, over 29199.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3849, pruned_loss=0.1304, over 5711026.31 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3991, pruned_loss=0.138, over 5733929.29 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3826, pruned_loss=0.1285, over 5706508.94 frames. ], batch size: 97, lr: 9.86e-03, grad_scale: 8.0
+2023-03-01 17:20:45,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 17:20:47,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:02,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:05,843 INFO [train.py:968] (1/2) Epoch 3, batch 21350, giga_loss[loss=0.3157, simple_loss=0.3832, pruned_loss=0.1241, over 28838.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3844, pruned_loss=0.1297, over 5712831.06 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.3996, pruned_loss=0.1389, over 5740793.60 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3815, pruned_loss=0.127, over 5701704.90 frames. ], batch size: 186, lr: 9.85e-03, grad_scale: 8.0
+2023-03-01 17:21:09,947 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:12,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:30,466 INFO [optim.py:369] (1/2) 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 17:21:36,971 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:21:46,752 INFO [train.py:968] (1/2) Epoch 3, batch 21400, giga_loss[loss=0.3342, simple_loss=0.3905, pruned_loss=0.139, over 27887.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3832, pruned_loss=0.1283, over 5722534.51 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.4, pruned_loss=0.1393, over 5742875.23 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3803, pruned_loss=0.1256, over 5711282.89 frames. ], batch size: 412, lr: 9.85e-03, grad_scale: 8.0
+2023-03-01 17:21:56,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 17:22:11,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:22:22,279 INFO [train.py:968] (1/2) Epoch 3, batch 21450, giga_loss[loss=0.3171, simple_loss=0.3825, pruned_loss=0.1258, over 28749.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3837, pruned_loss=0.1287, over 5721598.41 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.4011, pruned_loss=0.1403, over 5736686.91 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.38, pruned_loss=0.1254, over 5717416.45 frames. ], batch size: 92, lr: 9.85e-03, grad_scale: 4.0
+2023-03-01 17:22:48,838 INFO [optim.py:369] (1/2) 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 17:22:54,180 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:22:55,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=111782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:23:01,600 INFO [train.py:968] (1/2) Epoch 3, batch 21500, giga_loss[loss=0.292, simple_loss=0.3616, pruned_loss=0.1112, over 28838.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3822, pruned_loss=0.1285, over 5729402.77 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.4015, pruned_loss=0.141, over 5743282.65 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3781, pruned_loss=0.1247, over 5719302.53 frames. ], batch size: 174, lr: 9.85e-03, grad_scale: 4.0
+2023-03-01 17:23:18,721 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:23:35,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-01 17:23:40,844 INFO [train.py:968] (1/2) Epoch 3, batch 21550, giga_loss[loss=0.2739, simple_loss=0.343, pruned_loss=0.1024, over 28461.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3783, pruned_loss=0.1262, over 5726406.71 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.4013, pruned_loss=0.1408, over 5747071.94 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3748, pruned_loss=0.1229, over 5714544.84 frames. ], batch size: 65, lr: 9.85e-03, grad_scale: 4.0
+2023-03-01 17:23:56,647 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=111861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:24:02,992 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=111869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:24:05,603 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:24:21,118 INFO [train.py:968] (1/2) Epoch 3, batch 21600, giga_loss[loss=0.3832, simple_loss=0.4189, pruned_loss=0.1737, over 28591.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3783, pruned_loss=0.1267, over 5730886.42 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.4016, pruned_loss=0.1414, over 5750157.85 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3748, pruned_loss=0.1234, over 5718382.04 frames. ], batch size: 71, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:24:29,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=111901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:24:57,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6854, 3.2423, 1.7573, 1.5675], device='cuda:1'), covar=tensor([0.0734, 0.0346, 0.0770, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0434, 0.0302, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 17:25:00,411 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-01 17:25:00,669 INFO [train.py:968] (1/2) Epoch 3, batch 21650, giga_loss[loss=0.244, simple_loss=0.3201, pruned_loss=0.08398, over 28506.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3779, pruned_loss=0.1276, over 5724676.62 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.4011, pruned_loss=0.1412, over 5749575.92 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3751, pruned_loss=0.1247, over 5714386.42 frames. ], batch size: 60, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:25:25,770 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 21700, giga_loss[loss=0.278, simple_loss=0.3463, pruned_loss=0.1048, over 28476.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1286, over 5728018.43 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.4016, pruned_loss=0.1421, over 5754182.15 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.374, pruned_loss=0.125, over 5714546.86 frames. ], batch size: 60, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:25:39,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=111993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:25:48,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6904, 3.0168, 1.6285, 1.4829], device='cuda:1'), covar=tensor([0.0752, 0.0392, 0.0798, 0.1207], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0432, 0.0300, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0017], device='cuda:1')
+2023-03-01 17:26:16,214 INFO [train.py:968] (1/2) Epoch 3, batch 21750, giga_loss[loss=0.276, simple_loss=0.3502, pruned_loss=0.101, over 28896.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3765, pruned_loss=0.1283, over 5731636.26 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.4024, pruned_loss=0.143, over 5759508.70 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.372, pruned_loss=0.1242, over 5715184.70 frames. ], batch size: 199, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:26:23,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0407, 1.2696, 0.8955, 0.3055], device='cuda:1'), covar=tensor([0.0750, 0.0634, 0.1034, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.1181, 0.1277, 0.1072], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 17:26:41,271 INFO [optim.py:369] (1/2) 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 17:26:56,059 INFO [train.py:968] (1/2) Epoch 3, batch 21800, giga_loss[loss=0.3111, simple_loss=0.3666, pruned_loss=0.1278, over 29020.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3734, pruned_loss=0.1269, over 5725262.82 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.4027, pruned_loss=0.1432, over 5761256.69 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3692, pruned_loss=0.1233, over 5710328.37 frames. ], batch size: 164, lr: 9.84e-03, grad_scale: 8.0
+2023-03-01 17:27:13,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7366, 3.4070, 3.4522, 1.6916], device='cuda:1'), covar=tensor([0.0559, 0.0430, 0.0844, 0.2132], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0596, 0.0769, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 17:27:31,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:27:33,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:27:35,267 INFO [train.py:968] (1/2) Epoch 3, batch 21850, giga_loss[loss=0.2482, simple_loss=0.3247, pruned_loss=0.08587, over 28931.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3695, pruned_loss=0.1246, over 5727942.34 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.4021, pruned_loss=0.1432, over 5764678.69 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3658, pruned_loss=0.1211, over 5711569.32 frames. ], batch size: 136, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:27:45,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-01 17:27:58,659 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:28:11,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112184.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 17:28:16,368 INFO [train.py:968] (1/2) Epoch 3, batch 21900, giga_loss[loss=0.3438, simple_loss=0.4017, pruned_loss=0.143, over 28534.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3703, pruned_loss=0.125, over 5721385.97 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.4027, pruned_loss=0.1438, over 5765926.66 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3663, pruned_loss=0.1214, over 5706584.15 frames. ], batch size: 307, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:28:45,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-01 17:28:48,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7296, 2.1396, 1.9963, 1.8451], device='cuda:1'), covar=tensor([0.1744, 0.1796, 0.1272, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0789, 0.0732, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 17:28:56,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:29:01,910 INFO [train.py:968] (1/2) Epoch 3, batch 21950, giga_loss[loss=0.3094, simple_loss=0.3822, pruned_loss=0.1183, over 28508.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3734, pruned_loss=0.1266, over 5715358.47 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.4031, pruned_loss=0.1441, over 5767107.52 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3696, pruned_loss=0.1233, over 5702215.76 frames. ], batch size: 336, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:29:28,727 INFO [optim.py:369] (1/2) 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 17:29:42,319 INFO [train.py:968] (1/2) Epoch 3, batch 22000, giga_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09973, over 28831.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3761, pruned_loss=0.1278, over 5714682.58 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.4035, pruned_loss=0.1447, over 5760860.21 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3717, pruned_loss=0.124, over 5707242.63 frames. ], batch size: 199, lr: 9.83e-03, grad_scale: 8.0
+2023-03-01 17:29:57,464 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 17:30:00,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4213, 1.5030, 1.1478, 0.9165], device='cuda:1'), covar=tensor([0.0828, 0.0629, 0.0455, 0.0619], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.0916, 0.0955, 0.1026], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 17:30:22,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4661, 1.8178, 1.6149, 1.6207], device='cuda:1'), covar=tensor([0.1401, 0.1669, 0.1199, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0775, 0.0722, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 17:30:23,664 INFO [train.py:968] (1/2) Epoch 3, batch 22050, giga_loss[loss=0.2972, simple_loss=0.3722, pruned_loss=0.1111, over 28731.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3777, pruned_loss=0.1277, over 5704867.79 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.4038, pruned_loss=0.145, over 5760268.37 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3735, pruned_loss=0.1241, over 5698912.69 frames. ], batch size: 284, lr: 9.82e-03, grad_scale: 8.0
+2023-03-01 17:30:50,641 INFO [optim.py:369] (1/2) 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 17:30:54,230 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:30:57,139 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 22100, libri_loss[loss=0.3182, simple_loss=0.3642, pruned_loss=0.1361, over 29479.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3785, pruned_loss=0.128, over 5680499.94 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.4045, pruned_loss=0.1458, over 5740843.20 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3736, pruned_loss=0.1238, over 5691266.53 frames. ], batch size: 70, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:31:20,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:31:43,290 INFO [train.py:968] (1/2) Epoch 3, batch 22150, giga_loss[loss=0.3988, simple_loss=0.4374, pruned_loss=0.1802, over 28690.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3783, pruned_loss=0.1279, over 5689924.81 frames. ], libri_tot_loss[loss=0.3482, simple_loss=0.4044, pruned_loss=0.1461, over 5745501.12 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3737, pruned_loss=0.1238, over 5692919.88 frames. ], batch size: 242, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:31:57,126 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:32:09,235 INFO [optim.py:369] (1/2) 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 17:32:23,630 INFO [train.py:968] (1/2) Epoch 3, batch 22200, giga_loss[loss=0.3116, simple_loss=0.3798, pruned_loss=0.1217, over 28760.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3792, pruned_loss=0.1287, over 5696789.90 frames. ], libri_tot_loss[loss=0.3486, simple_loss=0.4045, pruned_loss=0.1463, over 5746409.17 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.375, pruned_loss=0.1249, over 5697517.27 frames. ], batch size: 284, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:33:04,362 INFO [train.py:968] (1/2) Epoch 3, batch 22250, giga_loss[loss=0.3536, simple_loss=0.4011, pruned_loss=0.1531, over 28905.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3808, pruned_loss=0.1302, over 5699121.10 frames. ], libri_tot_loss[loss=0.3492, simple_loss=0.405, pruned_loss=0.1468, over 5746807.14 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3768, pruned_loss=0.1265, over 5698697.79 frames. ], batch size: 174, lr: 9.82e-03, grad_scale: 4.0
+2023-03-01 17:33:19,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112559.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 17:33:30,299 INFO [optim.py:369] (1/2) 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 17:33:44,766 INFO [train.py:968] (1/2) Epoch 3, batch 22300, giga_loss[loss=0.3136, simple_loss=0.3828, pruned_loss=0.1222, over 29086.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3829, pruned_loss=0.1312, over 5699635.45 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.4047, pruned_loss=0.1466, over 5748496.81 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3796, pruned_loss=0.1283, over 5697460.25 frames. ], batch size: 136, lr: 9.81e-03, grad_scale: 4.0
+2023-03-01 17:33:52,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8072, 1.6772, 1.7589, 1.6475], device='cuda:1'), covar=tensor([0.0870, 0.1622, 0.1304, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0760, 0.0622, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 17:34:24,217 INFO [train.py:968] (1/2) Epoch 3, batch 22350, giga_loss[loss=0.3433, simple_loss=0.401, pruned_loss=0.1429, over 28264.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3854, pruned_loss=0.1323, over 5707538.95 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.4058, pruned_loss=0.1475, over 5750443.70 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3815, pruned_loss=0.1289, over 5703397.95 frames. ], batch size: 368, lr: 9.81e-03, grad_scale: 4.0
+2023-03-01 17:34:52,182 INFO [optim.py:369] (1/2) 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 17:35:07,892 INFO [train.py:968] (1/2) Epoch 3, batch 22400, giga_loss[loss=0.3417, simple_loss=0.3979, pruned_loss=0.1428, over 27597.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3855, pruned_loss=0.1318, over 5708784.51 frames. ], libri_tot_loss[loss=0.3509, simple_loss=0.4062, pruned_loss=0.1478, over 5751227.48 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3821, pruned_loss=0.1289, over 5704672.10 frames. ], batch size: 472, lr: 9.81e-03, grad_scale: 8.0
+2023-03-01 17:35:18,860 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112702.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 17:35:20,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112705.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 17:35:43,972 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=112734.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 17:35:49,593 INFO [train.py:968] (1/2) Epoch 3, batch 22450, giga_loss[loss=0.3334, simple_loss=0.401, pruned_loss=0.1329, over 28862.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3856, pruned_loss=0.1314, over 5712691.85 frames. ], libri_tot_loss[loss=0.3523, simple_loss=0.4072, pruned_loss=0.1487, over 5753553.31 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3818, pruned_loss=0.1281, over 5706967.50 frames. ], batch size: 174, lr: 9.81e-03, grad_scale: 4.0
+2023-03-01 17:36:18,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2148, 3.8261, 3.8979, 1.6437], device='cuda:1'), covar=tensor([0.0374, 0.0358, 0.0700, 0.2201], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0598, 0.0778, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 17:36:21,814 INFO [optim.py:369] (1/2) 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 17:36:26,229 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 22500, libri_loss[loss=0.3414, simple_loss=0.3865, pruned_loss=0.1481, over 29658.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3865, pruned_loss=0.132, over 5717249.60 frames. ], libri_tot_loss[loss=0.3528, simple_loss=0.4076, pruned_loss=0.149, over 5755721.11 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3828, pruned_loss=0.1289, over 5710050.62 frames. ], batch size: 73, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:37:08,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=112834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:37:14,575 INFO [train.py:968] (1/2) Epoch 3, batch 22550, giga_loss[loss=0.2716, simple_loss=0.3497, pruned_loss=0.09676, over 29015.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3856, pruned_loss=0.1319, over 5716619.35 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4079, pruned_loss=0.1496, over 5758977.47 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3819, pruned_loss=0.1284, over 5706877.51 frames. ], batch size: 155, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:37:42,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5959, 2.9307, 1.5385, 1.4323], device='cuda:1'), covar=tensor([0.0780, 0.0373, 0.0831, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0445, 0.0310, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:1')
+2023-03-01 17:37:43,485 INFO [optim.py:369] (1/2) 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 17:37:55,283 INFO [train.py:968] (1/2) Epoch 3, batch 22600, giga_loss[loss=0.314, simple_loss=0.3735, pruned_loss=0.1273, over 28976.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3828, pruned_loss=0.1304, over 5711191.33 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4081, pruned_loss=0.1499, over 5757336.15 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3791, pruned_loss=0.1271, over 5704443.71 frames. ], batch size: 213, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:37:57,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2470, 1.2781, 1.1435, 1.5548], device='cuda:1'), covar=tensor([0.2006, 0.1916, 0.1851, 0.2041], device='cuda:1'), in_proj_covar=tensor([0.1016, 0.0809, 0.0915, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 17:38:13,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3783, 1.3759, 1.2737, 1.2881], device='cuda:1'), covar=tensor([0.1202, 0.1781, 0.1573, 0.1587], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0773, 0.0624, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 17:38:34,337 INFO [train.py:968] (1/2) Epoch 3, batch 22650, giga_loss[loss=0.3461, simple_loss=0.4023, pruned_loss=0.1449, over 27934.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3797, pruned_loss=0.1289, over 5707452.98 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.4084, pruned_loss=0.1503, over 5756280.10 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1255, over 5702506.33 frames. ], batch size: 412, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:38:44,941 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:38:55,133 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=112970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:38:59,690 INFO [optim.py:369] (1/2) 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 17:39:00,766 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=112977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:39:03,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=112980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:39:12,067 INFO [train.py:968] (1/2) Epoch 3, batch 22700, giga_loss[loss=0.3155, simple_loss=0.3891, pruned_loss=0.121, over 28846.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3783, pruned_loss=0.1268, over 5708798.47 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.4087, pruned_loss=0.1507, over 5755123.42 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3747, pruned_loss=0.1234, over 5705196.71 frames. ], batch size: 186, lr: 9.80e-03, grad_scale: 4.0
+2023-03-01 17:39:16,202 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 17:39:29,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:39:31,152 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 22750, giga_loss[loss=0.2997, simple_loss=0.3726, pruned_loss=0.1134, over 28944.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.38, pruned_loss=0.1261, over 5706483.61 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4091, pruned_loss=0.1512, over 5757349.91 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3764, pruned_loss=0.1227, over 5701035.64 frames. ], batch size: 213, lr: 9.79e-03, grad_scale: 4.0
+2023-03-01 17:40:18,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 17:40:24,840 INFO [optim.py:369] (1/2) 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 17:40:28,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1924, 1.4532, 1.4006, 1.2641], device='cuda:1'), covar=tensor([0.0711, 0.0673, 0.1117, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0774, 0.0628, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 17:40:34,523 INFO [train.py:968] (1/2) Epoch 3, batch 22800, giga_loss[loss=0.3248, simple_loss=0.3725, pruned_loss=0.1385, over 28870.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3829, pruned_loss=0.1291, over 5707725.83 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4097, pruned_loss=0.152, over 5761669.00 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3787, pruned_loss=0.1249, over 5697926.48 frames. ], batch size: 99, lr: 9.79e-03, grad_scale: 8.0
+2023-03-01 17:41:16,442 INFO [train.py:968] (1/2) Epoch 3, batch 22850, giga_loss[loss=0.3075, simple_loss=0.3714, pruned_loss=0.1218, over 28807.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3809, pruned_loss=0.1296, over 5707903.75 frames. ], libri_tot_loss[loss=0.3578, simple_loss=0.4103, pruned_loss=0.1526, over 5764528.27 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3766, pruned_loss=0.1253, over 5696570.26 frames. ], batch size: 242, lr: 9.79e-03, grad_scale: 8.0
+2023-03-01 17:41:21,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-01 17:41:27,336 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:41:41,625 INFO [optim.py:369] (1/2) 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 17:41:43,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 17:41:51,346 INFO [train.py:968] (1/2) Epoch 3, batch 22900, giga_loss[loss=0.2935, simple_loss=0.3503, pruned_loss=0.1183, over 28555.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3799, pruned_loss=0.1304, over 5709940.28 frames. ], libri_tot_loss[loss=0.3577, simple_loss=0.4099, pruned_loss=0.1527, over 5761356.52 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3751, pruned_loss=0.1257, over 5700553.43 frames. ], batch size: 78, lr: 9.79e-03, grad_scale: 8.0
+2023-03-01 17:42:33,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-01 17:42:34,634 INFO [train.py:968] (1/2) Epoch 3, batch 22950, giga_loss[loss=0.3316, simple_loss=0.3727, pruned_loss=0.1453, over 24209.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3771, pruned_loss=0.1294, over 5713762.51 frames. ], libri_tot_loss[loss=0.358, simple_loss=0.41, pruned_loss=0.153, over 5762298.20 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3728, pruned_loss=0.1252, over 5704786.69 frames. ], batch size: 705, lr: 9.79e-03, grad_scale: 4.0
+2023-03-01 17:43:02,700 INFO [optim.py:369] (1/2) 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 17:43:12,962 INFO [train.py:968] (1/2) Epoch 3, batch 23000, giga_loss[loss=0.3268, simple_loss=0.3847, pruned_loss=0.1345, over 28939.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3767, pruned_loss=0.13, over 5715259.47 frames. ], libri_tot_loss[loss=0.3588, simple_loss=0.4106, pruned_loss=0.1535, over 5765297.48 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3721, pruned_loss=0.1257, over 5704566.59 frames. ], batch size: 174, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:43:21,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:24,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:32,339 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:33,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3180, 1.3304, 1.2099, 1.3565], device='cuda:1'), covar=tensor([0.2124, 0.2153, 0.1974, 0.2248], device='cuda:1'), in_proj_covar=tensor([0.1015, 0.0812, 0.0909, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 17:43:43,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:43,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-01 17:43:46,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:43:53,300 INFO [train.py:968] (1/2) Epoch 3, batch 23050, giga_loss[loss=0.2676, simple_loss=0.3374, pruned_loss=0.09888, over 28921.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3745, pruned_loss=0.1285, over 5724864.64 frames. ], libri_tot_loss[loss=0.3585, simple_loss=0.4103, pruned_loss=0.1534, over 5768765.97 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3703, pruned_loss=0.1246, over 5712207.98 frames. ], batch size: 213, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:43:56,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:44:11,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0144, 3.5795, 2.0470, 2.0558], device='cuda:1'), covar=tensor([0.0747, 0.0441, 0.0763, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0448, 0.0310, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0019, 0.0013, 0.0018], device='cuda:1')
+2023-03-01 17:44:20,925 INFO [optim.py:369] (1/2) 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 17:44:26,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=113385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:44:30,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7921, 3.5008, 3.4578, 1.6142], device='cuda:1'), covar=tensor([0.0492, 0.0416, 0.0757, 0.2195], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0594, 0.0773, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 17:44:30,940 INFO [train.py:968] (1/2) Epoch 3, batch 23100, giga_loss[loss=0.3217, simple_loss=0.3807, pruned_loss=0.1314, over 28008.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3706, pruned_loss=0.1265, over 5713098.17 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.411, pruned_loss=0.154, over 5761460.11 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.366, pruned_loss=0.1224, over 5708652.89 frames. ], batch size: 412, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:44:51,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5324, 1.1782, 2.9120, 2.7164], device='cuda:1'), covar=tensor([0.1456, 0.1778, 0.0398, 0.0577], device='cuda:1'), in_proj_covar=tensor([0.0520, 0.0495, 0.0675, 0.0541], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 17:45:12,565 INFO [train.py:968] (1/2) Epoch 3, batch 23150, giga_loss[loss=0.2675, simple_loss=0.3412, pruned_loss=0.09688, over 28832.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.366, pruned_loss=0.1242, over 5702521.29 frames. ], libri_tot_loss[loss=0.359, simple_loss=0.4104, pruned_loss=0.1538, over 5754853.34 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3621, pruned_loss=0.1206, over 5703080.70 frames. ], batch size: 174, lr: 9.78e-03, grad_scale: 4.0
+2023-03-01 17:45:38,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:45:40,319 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:45:41,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1511, 2.4690, 2.3436, 2.1843], device='cuda:1'), covar=tensor([0.1352, 0.1531, 0.1004, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0768, 0.0726, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 17:45:48,505 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:45:50,368 INFO [train.py:968] (1/2) Epoch 3, batch 23200, giga_loss[loss=0.2883, simple_loss=0.3556, pruned_loss=0.1105, over 28887.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3659, pruned_loss=0.1238, over 5700279.77 frames. ], libri_tot_loss[loss=0.36, simple_loss=0.4111, pruned_loss=0.1545, over 5744234.47 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3615, pruned_loss=0.1198, over 5709622.24 frames. ], batch size: 227, lr: 9.77e-03, grad_scale: 8.0
+2023-03-01 17:45:50,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:02,731 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:15,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:20,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:22,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=113531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:46:30,791 INFO [train.py:968] (1/2) Epoch 3, batch 23250, giga_loss[loss=0.3138, simple_loss=0.3802, pruned_loss=0.1237, over 29010.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3692, pruned_loss=0.1254, over 5696606.16 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.411, pruned_loss=0.1548, over 5737581.86 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3636, pruned_loss=0.1205, over 5708406.97 frames. ], batch size: 164, lr: 9.77e-03, grad_scale: 8.0
+2023-03-01 17:46:45,988 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:47:12,986 INFO [train.py:968] (1/2) Epoch 3, batch 23300, giga_loss[loss=0.3444, simple_loss=0.4074, pruned_loss=0.1407, over 29041.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3731, pruned_loss=0.1269, over 5705445.27 frames. ], libri_tot_loss[loss=0.3602, simple_loss=0.4109, pruned_loss=0.1547, over 5741327.84 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3675, pruned_loss=0.1222, over 5710541.81 frames. ], batch size: 136, lr: 9.77e-03, grad_scale: 4.0
+2023-03-01 17:47:42,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=113627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:47:44,981 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 17:47:54,509 INFO [train.py:968] (1/2) Epoch 3, batch 23350, giga_loss[loss=0.3227, simple_loss=0.3871, pruned_loss=0.1291, over 28803.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3783, pruned_loss=0.1302, over 5699710.09 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4115, pruned_loss=0.1556, over 5735818.79 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3723, pruned_loss=0.1248, over 5707099.85 frames. ], batch size: 99, lr: 9.77e-03, grad_scale: 4.0
+2023-03-01 17:48:02,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9538, 1.3029, 1.1036, 0.1878], device='cuda:1'), covar=tensor([0.1272, 0.1080, 0.1748, 0.2254], device='cuda:1'), in_proj_covar=tensor([0.1255, 0.1170, 0.1279, 0.1081], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 17:48:16,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2490, 1.9886, 1.5196, 0.5554], device='cuda:1'), covar=tensor([0.1707, 0.0852, 0.1600, 0.1996], device='cuda:1'), in_proj_covar=tensor([0.1259, 0.1173, 0.1284, 0.1083], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 17:48:25,309 INFO [optim.py:369] (1/2) 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 17:48:35,209 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 23400, giga_loss[loss=0.3843, simple_loss=0.431, pruned_loss=0.1687, over 28261.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3824, pruned_loss=0.1322, over 5699250.18 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.412, pruned_loss=0.1561, over 5736861.02 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3763, pruned_loss=0.1269, over 5703362.88 frames. ], batch size: 368, lr: 9.77e-03, grad_scale: 4.0
+2023-03-01 17:49:18,946 INFO [train.py:968] (1/2) Epoch 3, batch 23450, giga_loss[loss=0.2946, simple_loss=0.3655, pruned_loss=0.1119, over 28873.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3851, pruned_loss=0.1338, over 5692810.79 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.4123, pruned_loss=0.1564, over 5738160.71 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3794, pruned_loss=0.1287, over 5693687.14 frames. ], batch size: 213, lr: 9.76e-03, grad_scale: 4.0
+2023-03-01 17:49:24,219 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 17:49:51,479 INFO [optim.py:369] (1/2) 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 17:50:02,874 INFO [train.py:968] (1/2) Epoch 3, batch 23500, giga_loss[loss=0.3581, simple_loss=0.4048, pruned_loss=0.1557, over 28760.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.391, pruned_loss=0.1397, over 5689431.25 frames. ], libri_tot_loss[loss=0.3625, simple_loss=0.412, pruned_loss=0.1565, over 5737456.02 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3857, pruned_loss=0.1348, over 5688970.64 frames. ], batch size: 99, lr: 9.76e-03, grad_scale: 4.0
+2023-03-01 17:50:47,765 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=113832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:50:50,341 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 23550, giga_loss[loss=0.3838, simple_loss=0.4317, pruned_loss=0.168, over 29048.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3978, pruned_loss=0.1458, over 5683663.79 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4122, pruned_loss=0.1566, over 5736561.74 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3933, pruned_loss=0.1417, over 5683844.86 frames. ], batch size: 128, lr: 9.76e-03, grad_scale: 4.0
+2023-03-01 17:51:08,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5002, 1.1784, 1.1753, 1.5528], device='cuda:1'), covar=tensor([0.2058, 0.2106, 0.1817, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1018, 0.0813, 0.0907, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 17:51:25,374 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=113864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:51:39,489 INFO [optim.py:369] (1/2) 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 17:51:55,447 INFO [train.py:968] (1/2) Epoch 3, batch 23600, giga_loss[loss=0.3875, simple_loss=0.424, pruned_loss=0.1755, over 28825.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4051, pruned_loss=0.1518, over 5680906.52 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4125, pruned_loss=0.157, over 5736777.85 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4011, pruned_loss=0.148, over 5679718.00 frames. ], batch size: 99, lr: 9.76e-03, grad_scale: 8.0
+2023-03-01 17:52:00,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7050, 2.5918, 1.7481, 0.9154], device='cuda:1'), covar=tensor([0.2643, 0.1042, 0.1576, 0.2280], device='cuda:1'), in_proj_covar=tensor([0.1273, 0.1200, 0.1305, 0.1102], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 17:52:47,680 INFO [train.py:968] (1/2) Epoch 3, batch 23650, giga_loss[loss=0.422, simple_loss=0.4441, pruned_loss=0.2, over 27628.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.411, pruned_loss=0.1575, over 5679228.60 frames. ], libri_tot_loss[loss=0.3637, simple_loss=0.4127, pruned_loss=0.1574, over 5738979.13 frames. ], giga_tot_loss[loss=0.3579, simple_loss=0.4075, pruned_loss=0.1541, over 5675795.55 frames. ], batch size: 472, lr: 9.75e-03, grad_scale: 8.0
+2023-03-01 17:53:26,136 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-01 17:53:27,075 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 23700, giga_loss[loss=0.3633, simple_loss=0.4178, pruned_loss=0.1544, over 28686.00 frames. ], tot_loss[loss=0.3746, simple_loss=0.4193, pruned_loss=0.165, over 5669004.67 frames. ], libri_tot_loss[loss=0.3655, simple_loss=0.414, pruned_loss=0.1584, over 5740178.28 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4153, pruned_loss=0.1614, over 5663105.57 frames. ], batch size: 242, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:53:50,472 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:54:30,325 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 23750, giga_loss[loss=0.3548, simple_loss=0.4089, pruned_loss=0.1504, over 29125.00 frames. ], tot_loss[loss=0.3809, simple_loss=0.4239, pruned_loss=0.169, over 5669878.55 frames. ], libri_tot_loss[loss=0.3661, simple_loss=0.4145, pruned_loss=0.1588, over 5742001.52 frames. ], giga_tot_loss[loss=0.3762, simple_loss=0.4205, pruned_loss=0.1659, over 5662183.22 frames. ], batch size: 155, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:55:07,985 INFO [optim.py:369] (1/2) 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 17:55:17,629 INFO [train.py:968] (1/2) Epoch 3, batch 23800, giga_loss[loss=0.3627, simple_loss=0.4172, pruned_loss=0.1541, over 28897.00 frames. ], tot_loss[loss=0.3847, simple_loss=0.4258, pruned_loss=0.1718, over 5659679.10 frames. ], libri_tot_loss[loss=0.3662, simple_loss=0.4143, pruned_loss=0.159, over 5736945.88 frames. ], giga_tot_loss[loss=0.3814, simple_loss=0.4236, pruned_loss=0.1696, over 5654929.57 frames. ], batch size: 227, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:56:03,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-01 17:56:10,256 INFO [train.py:968] (1/2) Epoch 3, batch 23850, libri_loss[loss=0.3289, simple_loss=0.374, pruned_loss=0.1419, over 29475.00 frames. ], tot_loss[loss=0.387, simple_loss=0.4265, pruned_loss=0.1738, over 5652489.66 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4136, pruned_loss=0.1588, over 5740228.69 frames. ], giga_tot_loss[loss=0.3853, simple_loss=0.4256, pruned_loss=0.1725, over 5644226.25 frames. ], batch size: 70, lr: 9.75e-03, grad_scale: 4.0
+2023-03-01 17:56:15,589 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:56:18,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=114148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:56:50,575 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 17:56:51,593 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 23900, giga_loss[loss=0.5061, simple_loss=0.4982, pruned_loss=0.257, over 27600.00 frames. ], tot_loss[loss=0.3914, simple_loss=0.4293, pruned_loss=0.1768, over 5647302.64 frames. ], libri_tot_loss[loss=0.3656, simple_loss=0.4137, pruned_loss=0.1588, over 5740759.67 frames. ], giga_tot_loss[loss=0.3905, simple_loss=0.4288, pruned_loss=0.1761, over 5638960.87 frames. ], batch size: 472, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 17:57:23,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-01 17:58:02,891 INFO [train.py:968] (1/2) Epoch 3, batch 23950, giga_loss[loss=0.4851, simple_loss=0.472, pruned_loss=0.2491, over 23531.00 frames. ], tot_loss[loss=0.3972, simple_loss=0.4332, pruned_loss=0.1806, over 5615552.29 frames. ], libri_tot_loss[loss=0.3658, simple_loss=0.4136, pruned_loss=0.1589, over 5730860.71 frames. ], giga_tot_loss[loss=0.3968, simple_loss=0.4331, pruned_loss=0.1803, over 5616458.89 frames. ], batch size: 705, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 17:58:44,574 INFO [optim.py:369] (1/2) 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 17:58:57,239 INFO [train.py:968] (1/2) Epoch 3, batch 24000, giga_loss[loss=0.4401, simple_loss=0.4564, pruned_loss=0.2119, over 28328.00 frames. ], tot_loss[loss=0.3991, simple_loss=0.4337, pruned_loss=0.1822, over 5612391.32 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4141, pruned_loss=0.1595, over 5732624.53 frames. ], giga_tot_loss[loss=0.3992, simple_loss=0.434, pruned_loss=0.1823, over 5607107.84 frames. ], batch size: 368, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 17:58:57,239 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 17:59:05,879 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 17:59:52,708 INFO [train.py:968] (1/2) Epoch 3, batch 24050, giga_loss[loss=0.3593, simple_loss=0.4136, pruned_loss=0.1525, over 28985.00 frames. ], tot_loss[loss=0.3968, simple_loss=0.4317, pruned_loss=0.1809, over 5629891.21 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4143, pruned_loss=0.1597, over 5736407.99 frames. ], giga_tot_loss[loss=0.3976, simple_loss=0.4323, pruned_loss=0.1814, over 5619248.47 frames. ], batch size: 164, lr: 9.74e-03, grad_scale: 4.0
+2023-03-01 18:00:34,328 INFO [optim.py:369] (1/2) 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 18:00:38,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:00:44,211 INFO [train.py:968] (1/2) Epoch 3, batch 24100, giga_loss[loss=0.3412, simple_loss=0.3997, pruned_loss=0.1413, over 28853.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.4301, pruned_loss=0.1793, over 5624320.03 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4143, pruned_loss=0.1598, over 5734371.96 frames. ], giga_tot_loss[loss=0.3952, simple_loss=0.4308, pruned_loss=0.1798, over 5616424.14 frames. ], batch size: 136, lr: 9.74e-03, grad_scale: 2.0
+2023-03-01 18:01:05,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:01:36,272 INFO [train.py:968] (1/2) Epoch 3, batch 24150, giga_loss[loss=0.417, simple_loss=0.4289, pruned_loss=0.2025, over 23257.00 frames. ], tot_loss[loss=0.3928, simple_loss=0.4301, pruned_loss=0.1777, over 5615396.73 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4147, pruned_loss=0.1602, over 5734434.72 frames. ], giga_tot_loss[loss=0.3933, simple_loss=0.4305, pruned_loss=0.1781, over 5607236.58 frames. ], batch size: 705, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:01:41,501 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-01 18:02:18,358 INFO [optim.py:369] (1/2) 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 18:02:28,381 INFO [train.py:968] (1/2) Epoch 3, batch 24200, giga_loss[loss=0.3985, simple_loss=0.4468, pruned_loss=0.1751, over 28615.00 frames. ], tot_loss[loss=0.3928, simple_loss=0.4306, pruned_loss=0.1775, over 5622442.14 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4141, pruned_loss=0.1599, over 5737433.31 frames. ], giga_tot_loss[loss=0.3944, simple_loss=0.4318, pruned_loss=0.1784, over 5610415.98 frames. ], batch size: 262, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:03:01,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=114520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:03,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7499, 2.0611, 1.8025, 1.7572], device='cuda:1'), covar=tensor([0.1435, 0.1876, 0.1225, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0782, 0.0723, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 18:03:07,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-01 18:03:08,169 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:10,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 24250, giga_loss[loss=0.3661, simple_loss=0.4167, pruned_loss=0.1578, over 28573.00 frames. ], tot_loss[loss=0.3896, simple_loss=0.4285, pruned_loss=0.1753, over 5626446.49 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4143, pruned_loss=0.1601, over 5739120.71 frames. ], giga_tot_loss[loss=0.3908, simple_loss=0.4295, pruned_loss=0.176, over 5614472.64 frames. ], batch size: 336, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:03:32,878 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-01 18:03:40,350 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=114557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:42,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=114560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:03:43,043 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:04:10,773 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 24300, giga_loss[loss=0.368, simple_loss=0.4192, pruned_loss=0.1583, over 28543.00 frames. ], tot_loss[loss=0.3854, simple_loss=0.4262, pruned_loss=0.1723, over 5631673.76 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4141, pruned_loss=0.1602, over 5740498.60 frames. ], giga_tot_loss[loss=0.3867, simple_loss=0.4274, pruned_loss=0.173, over 5619260.16 frames. ], batch size: 85, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:04:35,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8957, 1.8463, 1.7257, 1.6397], device='cuda:1'), covar=tensor([0.0911, 0.1580, 0.1320, 0.1374], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0777, 0.0634, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 18:05:01,589 INFO [train.py:968] (1/2) Epoch 3, batch 24350, giga_loss[loss=0.4121, simple_loss=0.4464, pruned_loss=0.1889, over 27481.00 frames. ], tot_loss[loss=0.3826, simple_loss=0.4245, pruned_loss=0.1703, over 5636854.26 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4143, pruned_loss=0.1605, over 5745175.74 frames. ], giga_tot_loss[loss=0.3837, simple_loss=0.4255, pruned_loss=0.171, over 5619916.96 frames. ], batch size: 472, lr: 9.73e-03, grad_scale: 2.0
+2023-03-01 18:05:36,585 INFO [optim.py:369] (1/2) 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 18:05:47,595 INFO [train.py:968] (1/2) Epoch 3, batch 24400, giga_loss[loss=0.323, simple_loss=0.3863, pruned_loss=0.1299, over 28753.00 frames. ], tot_loss[loss=0.3755, simple_loss=0.4195, pruned_loss=0.1658, over 5640417.70 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4138, pruned_loss=0.1602, over 5745634.55 frames. ], giga_tot_loss[loss=0.3772, simple_loss=0.4209, pruned_loss=0.1667, over 5623917.45 frames. ], batch size: 262, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:06:08,162 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-01 18:06:33,918 INFO [train.py:968] (1/2) Epoch 3, batch 24450, giga_loss[loss=0.3482, simple_loss=0.399, pruned_loss=0.1487, over 28997.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4164, pruned_loss=0.1634, over 5648923.85 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4134, pruned_loss=0.1599, over 5748678.33 frames. ], giga_tot_loss[loss=0.3736, simple_loss=0.4181, pruned_loss=0.1645, over 5630203.89 frames. ], batch size: 164, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:07:03,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5141, 2.0925, 1.4605, 0.7341], device='cuda:1'), covar=tensor([0.2389, 0.1257, 0.1283, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1194, 0.1274, 0.1087], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 18:07:14,064 INFO [optim.py:369] (1/2) 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 18:07:24,238 INFO [train.py:968] (1/2) Epoch 3, batch 24500, giga_loss[loss=0.3629, simple_loss=0.4165, pruned_loss=0.1546, over 28996.00 frames. ], tot_loss[loss=0.3726, simple_loss=0.417, pruned_loss=0.1641, over 5646044.94 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4134, pruned_loss=0.16, over 5752118.62 frames. ], giga_tot_loss[loss=0.3742, simple_loss=0.4184, pruned_loss=0.165, over 5626249.90 frames. ], batch size: 106, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:08:11,968 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 18:08:21,263 INFO [train.py:968] (1/2) Epoch 3, batch 24550, giga_loss[loss=0.4008, simple_loss=0.4354, pruned_loss=0.1831, over 27620.00 frames. ], tot_loss[loss=0.3718, simple_loss=0.4169, pruned_loss=0.1634, over 5642108.04 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4136, pruned_loss=0.1601, over 5744035.07 frames. ], giga_tot_loss[loss=0.3729, simple_loss=0.4178, pruned_loss=0.164, over 5632342.25 frames. ], batch size: 472, lr: 9.72e-03, grad_scale: 4.0
+2023-03-01 18:08:23,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9531, 1.1683, 3.7877, 3.0656], device='cuda:1'), covar=tensor([0.1480, 0.1909, 0.0353, 0.0578], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0508, 0.0689, 0.0549], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 18:09:05,177 INFO [optim.py:369] (1/2) 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 18:09:13,814 INFO [train.py:968] (1/2) Epoch 3, batch 24600, giga_loss[loss=0.3896, simple_loss=0.4424, pruned_loss=0.1684, over 28126.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4133, pruned_loss=0.1587, over 5656511.45 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4136, pruned_loss=0.1602, over 5745900.49 frames. ], giga_tot_loss[loss=0.3662, simple_loss=0.414, pruned_loss=0.1592, over 5646532.27 frames. ], batch size: 412, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:09:20,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=114895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:10:07,028 INFO [train.py:968] (1/2) Epoch 3, batch 24650, giga_loss[loss=0.3694, simple_loss=0.4334, pruned_loss=0.1527, over 29039.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4147, pruned_loss=0.1569, over 5656793.23 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4141, pruned_loss=0.1607, over 5737959.69 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.4148, pruned_loss=0.1568, over 5654803.68 frames. ], batch size: 155, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:10:51,358 INFO [optim.py:369] (1/2) 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 18:10:59,152 INFO [train.py:968] (1/2) Epoch 3, batch 24700, giga_loss[loss=0.3472, simple_loss=0.3992, pruned_loss=0.1476, over 28576.00 frames. ], tot_loss[loss=0.3648, simple_loss=0.4151, pruned_loss=0.1572, over 5649963.45 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4139, pruned_loss=0.1605, over 5739256.28 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4154, pruned_loss=0.1572, over 5645918.21 frames. ], batch size: 71, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:11:49,262 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 24750, giga_loss[loss=0.351, simple_loss=0.4007, pruned_loss=0.1507, over 28936.00 frames. ], tot_loss[loss=0.366, simple_loss=0.4159, pruned_loss=0.158, over 5669295.88 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.414, pruned_loss=0.1606, over 5742370.73 frames. ], giga_tot_loss[loss=0.3659, simple_loss=0.4161, pruned_loss=0.1579, over 5661793.49 frames. ], batch size: 112, lr: 9.71e-03, grad_scale: 4.0
+2023-03-01 18:11:52,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:12:16,853 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:12:29,215 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 24800, libri_loss[loss=0.2963, simple_loss=0.3487, pruned_loss=0.1219, over 29464.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4132, pruned_loss=0.1568, over 5675106.19 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4133, pruned_loss=0.1604, over 5734233.81 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.414, pruned_loss=0.1568, over 5673384.94 frames. ], batch size: 70, lr: 9.71e-03, grad_scale: 8.0
+2023-03-01 18:12:56,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:13:25,270 INFO [train.py:968] (1/2) Epoch 3, batch 24850, giga_loss[loss=0.3612, simple_loss=0.4147, pruned_loss=0.1538, over 28953.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4133, pruned_loss=0.1584, over 5676619.91 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4139, pruned_loss=0.1609, over 5737167.04 frames. ], giga_tot_loss[loss=0.3646, simple_loss=0.4135, pruned_loss=0.1579, over 5670717.72 frames. ], batch size: 213, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:13:45,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-01 18:14:02,176 INFO [optim.py:369] (1/2) 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 18:14:11,186 INFO [train.py:968] (1/2) Epoch 3, batch 24900, giga_loss[loss=0.3434, simple_loss=0.3931, pruned_loss=0.1468, over 28930.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4126, pruned_loss=0.159, over 5675952.95 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4137, pruned_loss=0.161, over 5740692.09 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4129, pruned_loss=0.1585, over 5667186.61 frames. ], batch size: 106, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:14:57,905 INFO [train.py:968] (1/2) Epoch 3, batch 24950, giga_loss[loss=0.3176, simple_loss=0.3891, pruned_loss=0.1231, over 28587.00 frames. ], tot_loss[loss=0.3612, simple_loss=0.4104, pruned_loss=0.156, over 5684030.78 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4127, pruned_loss=0.1604, over 5743117.44 frames. ], giga_tot_loss[loss=0.3618, simple_loss=0.4115, pruned_loss=0.156, over 5672713.01 frames. ], batch size: 85, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:15:19,343 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:15:50,397 INFO [train.py:968] (1/2) Epoch 3, batch 25000, giga_loss[loss=0.346, simple_loss=0.3957, pruned_loss=0.1482, over 28867.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4094, pruned_loss=0.1537, over 5682493.95 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4127, pruned_loss=0.1604, over 5735268.03 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4102, pruned_loss=0.1536, over 5679099.08 frames. ], batch size: 112, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:16:19,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8207, 2.3568, 1.1726, 1.0322], device='cuda:1'), covar=tensor([0.0953, 0.0552, 0.0658, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.1216, 0.0941, 0.0975, 0.1033], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 18:16:38,437 INFO [train.py:968] (1/2) Epoch 3, batch 25050, giga_loss[loss=0.3442, simple_loss=0.4088, pruned_loss=0.1398, over 28854.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4102, pruned_loss=0.1547, over 5674846.84 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4132, pruned_loss=0.1607, over 5737552.48 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4103, pruned_loss=0.1543, over 5669130.50 frames. ], batch size: 145, lr: 9.70e-03, grad_scale: 4.0
+2023-03-01 18:17:12,765 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-01 18:17:21,098 INFO [optim.py:369] (1/2) 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 18:17:30,377 INFO [train.py:968] (1/2) Epoch 3, batch 25100, giga_loss[loss=0.327, simple_loss=0.3854, pruned_loss=0.1343, over 28843.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4083, pruned_loss=0.1538, over 5680055.20 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.413, pruned_loss=0.1605, over 5738574.85 frames. ], giga_tot_loss[loss=0.3578, simple_loss=0.4085, pruned_loss=0.1535, over 5673797.88 frames. ], batch size: 199, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:17:47,467 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6610, 2.8929, 1.5408, 1.5499], device='cuda:1'), covar=tensor([0.0814, 0.0389, 0.0795, 0.1279], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0461, 0.0315, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 18:18:19,681 INFO [train.py:968] (1/2) Epoch 3, batch 25150, giga_loss[loss=0.3546, simple_loss=0.4111, pruned_loss=0.149, over 28864.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4085, pruned_loss=0.1552, over 5659345.02 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4133, pruned_loss=0.1608, over 5730001.87 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4083, pruned_loss=0.1545, over 5659534.00 frames. ], batch size: 186, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:18:35,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4627, 1.5558, 1.1013, 0.8278], device='cuda:1'), covar=tensor([0.0777, 0.0610, 0.0517, 0.0725], device='cuda:1'), in_proj_covar=tensor([0.1232, 0.0958, 0.0985, 0.1042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 18:18:39,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6476, 3.2149, 1.6800, 1.4620], device='cuda:1'), covar=tensor([0.0847, 0.0361, 0.0800, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0465, 0.0318, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 18:18:41,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3236, 1.3086, 1.1963, 1.3978], device='cuda:1'), covar=tensor([0.1982, 0.2131, 0.1883, 0.2103], device='cuda:1'), in_proj_covar=tensor([0.1039, 0.0850, 0.0934, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 18:18:58,390 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:19:03,501 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 25200, giga_loss[loss=0.3473, simple_loss=0.4046, pruned_loss=0.145, over 29014.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.4085, pruned_loss=0.1561, over 5661927.62 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4135, pruned_loss=0.161, over 5730883.86 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4081, pruned_loss=0.1554, over 5660969.55 frames. ], batch size: 164, lr: 9.69e-03, grad_scale: 8.0
+2023-03-01 18:19:24,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6660, 1.7256, 1.3128, 1.0136], device='cuda:1'), covar=tensor([0.0703, 0.0471, 0.0439, 0.0597], device='cuda:1'), in_proj_covar=tensor([0.1211, 0.0948, 0.0976, 0.1031], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 18:19:24,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6113, 1.9774, 1.7980, 1.6916], device='cuda:1'), covar=tensor([0.1284, 0.2021, 0.1200, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0792, 0.0733, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 18:19:52,539 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-01 18:19:59,892 INFO [train.py:968] (1/2) Epoch 3, batch 25250, libri_loss[loss=0.3915, simple_loss=0.4316, pruned_loss=0.1757, over 26083.00 frames. ], tot_loss[loss=0.3596, simple_loss=0.4075, pruned_loss=0.1558, over 5659002.23 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.414, pruned_loss=0.1613, over 5726212.17 frames. ], giga_tot_loss[loss=0.3582, simple_loss=0.4066, pruned_loss=0.1549, over 5661648.25 frames. ], batch size: 136, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:20:31,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8815, 1.3731, 3.9548, 3.1345], device='cuda:1'), covar=tensor([0.1582, 0.1911, 0.0336, 0.0498], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0508, 0.0699, 0.0551], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 18:20:36,886 INFO [optim.py:369] (1/2) 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 18:20:46,308 INFO [train.py:968] (1/2) Epoch 3, batch 25300, giga_loss[loss=0.3964, simple_loss=0.4299, pruned_loss=0.1815, over 27516.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.4049, pruned_loss=0.1538, over 5666167.69 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4134, pruned_loss=0.1609, over 5728449.70 frames. ], giga_tot_loss[loss=0.3555, simple_loss=0.4044, pruned_loss=0.1533, over 5664765.34 frames. ], batch size: 472, lr: 9.69e-03, grad_scale: 4.0
+2023-03-01 18:21:21,850 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:21:24,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:21:31,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:21:32,693 INFO [train.py:968] (1/2) Epoch 3, batch 25350, giga_loss[loss=0.335, simple_loss=0.395, pruned_loss=0.1376, over 28898.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.4049, pruned_loss=0.1547, over 5669225.62 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4126, pruned_loss=0.1604, over 5735304.57 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.4049, pruned_loss=0.1543, over 5659086.90 frames. ], batch size: 199, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:21:50,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:22:10,820 INFO [optim.py:369] (1/2) 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 18:22:14,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2384, 1.3972, 1.2158, 1.4745], device='cuda:1'), covar=tensor([0.2064, 0.1972, 0.1859, 0.1988], device='cuda:1'), in_proj_covar=tensor([0.1041, 0.0846, 0.0934, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 18:22:19,876 INFO [train.py:968] (1/2) Epoch 3, batch 25400, giga_loss[loss=0.3913, simple_loss=0.4247, pruned_loss=0.179, over 28939.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4064, pruned_loss=0.1553, over 5674119.59 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4128, pruned_loss=0.1607, over 5740178.39 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.406, pruned_loss=0.1545, over 5659714.04 frames. ], batch size: 227, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:22:28,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-01 18:23:02,388 INFO [train.py:968] (1/2) Epoch 3, batch 25450, giga_loss[loss=0.3345, simple_loss=0.3942, pruned_loss=0.1374, over 28782.00 frames. ], tot_loss[loss=0.3588, simple_loss=0.4075, pruned_loss=0.155, over 5680096.43 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4129, pruned_loss=0.1607, over 5745706.79 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4069, pruned_loss=0.1542, over 5661139.46 frames. ], batch size: 99, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:23:24,015 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:24,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=115764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:43,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:43,524 INFO [optim.py:369] (1/2) 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 18:23:46,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=115785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:23:50,528 INFO [train.py:968] (1/2) Epoch 3, batch 25500, giga_loss[loss=0.3485, simple_loss=0.4046, pruned_loss=0.1462, over 28724.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.407, pruned_loss=0.1541, over 5667886.00 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4126, pruned_loss=0.1605, over 5736798.12 frames. ], giga_tot_loss[loss=0.3569, simple_loss=0.4067, pruned_loss=0.1536, over 5660474.36 frames. ], batch size: 99, lr: 9.68e-03, grad_scale: 4.0
+2023-03-01 18:24:15,984 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=115814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:24:39,759 INFO [train.py:968] (1/2) Epoch 3, batch 25550, giga_loss[loss=0.3339, simple_loss=0.3962, pruned_loss=0.1358, over 28941.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.4082, pruned_loss=0.1556, over 5670096.16 frames. ], libri_tot_loss[loss=0.3669, simple_loss=0.4128, pruned_loss=0.1606, over 5740731.72 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4077, pruned_loss=0.155, over 5658534.26 frames. ], batch size: 136, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:24:50,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=115854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:25:16,907 INFO [optim.py:369] (1/2) 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 18:25:25,450 INFO [train.py:968] (1/2) Epoch 3, batch 25600, giga_loss[loss=0.311, simple_loss=0.3704, pruned_loss=0.1257, over 28852.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4098, pruned_loss=0.1574, over 5662646.67 frames. ], libri_tot_loss[loss=0.3668, simple_loss=0.4127, pruned_loss=0.1604, over 5734590.71 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4093, pruned_loss=0.1569, over 5655692.42 frames. ], batch size: 112, lr: 9.67e-03, grad_scale: 8.0
+2023-03-01 18:26:13,632 INFO [train.py:968] (1/2) Epoch 3, batch 25650, giga_loss[loss=0.3842, simple_loss=0.4207, pruned_loss=0.1739, over 28537.00 frames. ], tot_loss[loss=0.3674, simple_loss=0.4126, pruned_loss=0.1611, over 5659875.95 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4136, pruned_loss=0.1611, over 5740150.48 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4113, pruned_loss=0.16, over 5646559.87 frames. ], batch size: 307, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:27:00,094 INFO [optim.py:369] (1/2) 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 18:27:09,355 INFO [train.py:968] (1/2) Epoch 3, batch 25700, giga_loss[loss=0.3786, simple_loss=0.4187, pruned_loss=0.1693, over 28185.00 frames. ], tot_loss[loss=0.3708, simple_loss=0.4144, pruned_loss=0.1636, over 5674298.26 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4138, pruned_loss=0.1614, over 5742413.03 frames. ], giga_tot_loss[loss=0.3691, simple_loss=0.4133, pruned_loss=0.1625, over 5660971.06 frames. ], batch size: 368, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:27:16,080 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=115997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:27:19,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:27:47,541 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:27:58,079 INFO [train.py:968] (1/2) Epoch 3, batch 25750, giga_loss[loss=0.3093, simple_loss=0.3692, pruned_loss=0.1247, over 28769.00 frames. ], tot_loss[loss=0.3728, simple_loss=0.4156, pruned_loss=0.165, over 5662872.08 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4142, pruned_loss=0.1616, over 5746686.72 frames. ], giga_tot_loss[loss=0.3711, simple_loss=0.4143, pruned_loss=0.1639, over 5645894.35 frames. ], batch size: 119, lr: 9.67e-03, grad_scale: 4.0
+2023-03-01 18:28:10,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9790, 1.1002, 0.8433, 0.6395], device='cuda:1'), covar=tensor([0.0621, 0.0581, 0.0474, 0.0571], device='cuda:1'), in_proj_covar=tensor([0.1183, 0.0938, 0.0968, 0.1026], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 18:28:35,717 INFO [optim.py:369] (1/2) 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 18:28:43,445 INFO [train.py:968] (1/2) Epoch 3, batch 25800, giga_loss[loss=0.3492, simple_loss=0.3949, pruned_loss=0.1517, over 28638.00 frames. ], tot_loss[loss=0.37, simple_loss=0.4136, pruned_loss=0.1632, over 5668612.19 frames. ], libri_tot_loss[loss=0.3685, simple_loss=0.4139, pruned_loss=0.1615, over 5746754.89 frames. ], giga_tot_loss[loss=0.3689, simple_loss=0.4128, pruned_loss=0.1625, over 5653451.26 frames. ], batch size: 307, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:29:00,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-01 18:29:28,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:29:30,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:29:31,530 INFO [train.py:968] (1/2) Epoch 3, batch 25850, giga_loss[loss=0.3116, simple_loss=0.386, pruned_loss=0.1186, over 28990.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4135, pruned_loss=0.1629, over 5655875.90 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4136, pruned_loss=0.1613, over 5734307.98 frames. ], giga_tot_loss[loss=0.3693, simple_loss=0.4132, pruned_loss=0.1627, over 5650586.13 frames. ], batch size: 128, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:30:12,538 INFO [optim.py:369] (1/2) 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 18:30:18,003 INFO [train.py:968] (1/2) Epoch 3, batch 25900, libri_loss[loss=0.4042, simple_loss=0.45, pruned_loss=0.1792, over 29245.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4102, pruned_loss=0.1586, over 5667745.89 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.414, pruned_loss=0.1617, over 5736672.65 frames. ], giga_tot_loss[loss=0.3628, simple_loss=0.4095, pruned_loss=0.158, over 5660444.59 frames. ], batch size: 94, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:30:29,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4467, 2.9367, 1.4412, 1.4551], device='cuda:1'), covar=tensor([0.0890, 0.0392, 0.0884, 0.1304], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0460, 0.0316, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 18:31:06,447 INFO [train.py:968] (1/2) Epoch 3, batch 25950, giga_loss[loss=0.3586, simple_loss=0.4094, pruned_loss=0.1539, over 28372.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4076, pruned_loss=0.1567, over 5661359.62 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4145, pruned_loss=0.1621, over 5736232.77 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4065, pruned_loss=0.1559, over 5655051.54 frames. ], batch size: 65, lr: 9.66e-03, grad_scale: 4.0
+2023-03-01 18:31:44,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:45,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:46,380 INFO [optim.py:369] (1/2) 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 18:31:46,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:48,448 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116285.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 18:31:48,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:31:53,791 INFO [train.py:968] (1/2) Epoch 3, batch 26000, giga_loss[loss=0.4294, simple_loss=0.4381, pruned_loss=0.2104, over 27556.00 frames. ], tot_loss[loss=0.3606, simple_loss=0.4066, pruned_loss=0.1573, over 5668168.30 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4147, pruned_loss=0.1624, over 5741223.78 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4052, pruned_loss=0.1561, over 5656591.36 frames. ], batch size: 472, lr: 9.66e-03, grad_scale: 8.0
+2023-03-01 18:31:55,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9375, 4.6820, 2.0203, 1.8316], device='cuda:1'), covar=tensor([0.0829, 0.0246, 0.0792, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0459, 0.0317, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 18:32:08,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4211, 4.0650, 4.1362, 1.9079], device='cuda:1'), covar=tensor([0.0514, 0.0478, 0.0973, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0631, 0.0815, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 18:32:14,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:32:14,909 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:32:26,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-01 18:32:44,013 INFO [train.py:968] (1/2) Epoch 3, batch 26050, giga_loss[loss=0.3576, simple_loss=0.4058, pruned_loss=0.1547, over 29086.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.4058, pruned_loss=0.1564, over 5690398.46 frames. ], libri_tot_loss[loss=0.3704, simple_loss=0.4151, pruned_loss=0.1628, over 5745847.64 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.404, pruned_loss=0.1548, over 5674998.56 frames. ], batch size: 136, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:33:26,750 INFO [optim.py:369] (1/2) 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 18:33:31,536 INFO [train.py:968] (1/2) Epoch 3, batch 26100, giga_loss[loss=0.3667, simple_loss=0.4144, pruned_loss=0.1595, over 28549.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4097, pruned_loss=0.1595, over 5688459.11 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4154, pruned_loss=0.1634, over 5748862.05 frames. ], giga_tot_loss[loss=0.3615, simple_loss=0.4078, pruned_loss=0.1576, over 5671762.20 frames. ], batch size: 336, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:33:45,301 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 26150, giga_loss[loss=0.3464, simple_loss=0.4192, pruned_loss=0.1367, over 29027.00 frames. ], tot_loss[loss=0.3662, simple_loss=0.4132, pruned_loss=0.1595, over 5696426.46 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4153, pruned_loss=0.1634, over 5751186.94 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4118, pruned_loss=0.158, over 5679732.19 frames. ], batch size: 128, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:35:01,636 INFO [optim.py:369] (1/2) 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 18:35:07,718 INFO [train.py:968] (1/2) Epoch 3, batch 26200, giga_loss[loss=0.3349, simple_loss=0.4026, pruned_loss=0.1336, over 28917.00 frames. ], tot_loss[loss=0.3647, simple_loss=0.4142, pruned_loss=0.1576, over 5687968.69 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4154, pruned_loss=0.1636, over 5749698.06 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.413, pruned_loss=0.1561, over 5675347.02 frames. ], batch size: 199, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:35:15,514 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:35:27,565 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:35:29,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-01 18:35:55,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6942, 3.3844, 3.4465, 1.4664], device='cuda:1'), covar=tensor([0.0615, 0.0518, 0.0990, 0.2266], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0633, 0.0811, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 18:35:59,936 INFO [train.py:968] (1/2) Epoch 3, batch 26250, giga_loss[loss=0.3722, simple_loss=0.4197, pruned_loss=0.1623, over 28592.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4162, pruned_loss=0.1591, over 5690001.81 frames. ], libri_tot_loss[loss=0.3718, simple_loss=0.4158, pruned_loss=0.1639, over 5751779.72 frames. ], giga_tot_loss[loss=0.365, simple_loss=0.4148, pruned_loss=0.1576, over 5677388.71 frames. ], batch size: 78, lr: 9.65e-03, grad_scale: 4.0
+2023-03-01 18:36:40,776 INFO [optim.py:369] (1/2) 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 18:36:45,545 INFO [train.py:968] (1/2) Epoch 3, batch 26300, giga_loss[loss=0.3832, simple_loss=0.4343, pruned_loss=0.1661, over 28896.00 frames. ], tot_loss[loss=0.3694, simple_loss=0.4176, pruned_loss=0.1606, over 5679953.24 frames. ], libri_tot_loss[loss=0.3719, simple_loss=0.4158, pruned_loss=0.164, over 5741748.11 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.4165, pruned_loss=0.1592, over 5677737.75 frames. ], batch size: 199, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:37:36,933 INFO [train.py:968] (1/2) Epoch 3, batch 26350, giga_loss[loss=0.3359, simple_loss=0.3987, pruned_loss=0.1365, over 29037.00 frames. ], tot_loss[loss=0.37, simple_loss=0.4172, pruned_loss=0.1614, over 5674887.54 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4154, pruned_loss=0.1637, over 5744286.10 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4168, pruned_loss=0.1606, over 5669853.56 frames. ], batch size: 136, lr: 9.64e-03, grad_scale: 2.0
+2023-03-01 18:37:56,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 18:38:25,016 INFO [optim.py:369] (1/2) 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 18:38:29,614 INFO [train.py:968] (1/2) Epoch 3, batch 26400, giga_loss[loss=0.3507, simple_loss=0.4028, pruned_loss=0.1492, over 28891.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4167, pruned_loss=0.1616, over 5684840.36 frames. ], libri_tot_loss[loss=0.3714, simple_loss=0.4155, pruned_loss=0.1637, over 5744772.67 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4163, pruned_loss=0.1609, over 5680159.49 frames. ], batch size: 213, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:38:53,597 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:38:55,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5514, 1.8603, 1.7163, 1.7075], device='cuda:1'), covar=tensor([0.1328, 0.1774, 0.1109, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0798, 0.0737, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 18:39:18,028 INFO [train.py:968] (1/2) Epoch 3, batch 26450, libri_loss[loss=0.3923, simple_loss=0.4323, pruned_loss=0.1761, over 27679.00 frames. ], tot_loss[loss=0.3693, simple_loss=0.4152, pruned_loss=0.1617, over 5686300.44 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.4154, pruned_loss=0.1638, over 5746929.12 frames. ], giga_tot_loss[loss=0.3685, simple_loss=0.4149, pruned_loss=0.161, over 5679244.80 frames. ], batch size: 116, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:40:02,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:40:06,442 INFO [optim.py:369] (1/2) 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 18:40:10,186 INFO [train.py:968] (1/2) Epoch 3, batch 26500, libri_loss[loss=0.3674, simple_loss=0.4235, pruned_loss=0.1557, over 29652.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4142, pruned_loss=0.1618, over 5694046.78 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4149, pruned_loss=0.1634, over 5751245.27 frames. ], giga_tot_loss[loss=0.3687, simple_loss=0.4144, pruned_loss=0.1615, over 5683202.93 frames. ], batch size: 91, lr: 9.64e-03, grad_scale: 4.0
+2023-03-01 18:40:14,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6578, 2.9554, 1.5692, 1.5796], device='cuda:1'), covar=tensor([0.0781, 0.0366, 0.0822, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0463, 0.0316, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 18:40:21,356 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116803.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 18:40:23,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 18:40:52,344 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=116835.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 18:40:56,471 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:40:58,494 INFO [train.py:968] (1/2) Epoch 3, batch 26550, giga_loss[loss=0.3659, simple_loss=0.4092, pruned_loss=0.1613, over 28896.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4145, pruned_loss=0.1623, over 5669461.49 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4151, pruned_loss=0.1636, over 5735107.80 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4144, pruned_loss=0.1618, over 5673318.36 frames. ], batch size: 174, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:41:19,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3422, 2.0359, 1.4439, 0.6257], device='cuda:1'), covar=tensor([0.1624, 0.0872, 0.1376, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.1204, 0.1275, 0.1094], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 18:41:28,115 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:41:39,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=116884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:41:40,435 INFO [optim.py:369] (1/2) 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 18:41:41,708 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-01 18:41:44,597 INFO [train.py:968] (1/2) Epoch 3, batch 26600, giga_loss[loss=0.449, simple_loss=0.4602, pruned_loss=0.2189, over 26652.00 frames. ], tot_loss[loss=0.3707, simple_loss=0.4151, pruned_loss=0.1631, over 5676407.27 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4151, pruned_loss=0.1635, over 5737895.22 frames. ], giga_tot_loss[loss=0.3704, simple_loss=0.415, pruned_loss=0.1629, over 5676039.05 frames. ], batch size: 555, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:41:51,740 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=116898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:42:18,052 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=116925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:42:20,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=116928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:42:32,447 INFO [train.py:968] (1/2) Epoch 3, batch 26650, giga_loss[loss=0.3991, simple_loss=0.4124, pruned_loss=0.193, over 23697.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4128, pruned_loss=0.162, over 5659352.37 frames. ], libri_tot_loss[loss=0.3711, simple_loss=0.4154, pruned_loss=0.1634, over 5740851.45 frames. ], giga_tot_loss[loss=0.3681, simple_loss=0.4125, pruned_loss=0.1618, over 5655213.72 frames. ], batch size: 705, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:42:49,578 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:43:22,055 INFO [train.py:968] (1/2) Epoch 3, batch 26700, giga_loss[loss=0.3188, simple_loss=0.3812, pruned_loss=0.1282, over 28735.00 frames. ], tot_loss[loss=0.3661, simple_loss=0.411, pruned_loss=0.1606, over 5654328.18 frames. ], libri_tot_loss[loss=0.371, simple_loss=0.4152, pruned_loss=0.1634, over 5735185.30 frames. ], giga_tot_loss[loss=0.3659, simple_loss=0.4108, pruned_loss=0.1605, over 5654961.60 frames. ], batch size: 92, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:43:40,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:44,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:46,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:56,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:43:58,637 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 26750, giga_loss[loss=0.4293, simple_loss=0.4459, pruned_loss=0.2063, over 26575.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4109, pruned_loss=0.1588, over 5657561.51 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4152, pruned_loss=0.1633, over 5729193.32 frames. ], giga_tot_loss[loss=0.3641, simple_loss=0.4108, pruned_loss=0.1587, over 5661457.80 frames. ], batch size: 555, lr: 9.63e-03, grad_scale: 4.0
+2023-03-01 18:44:14,963 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:44:25,504 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:44:50,313 INFO [optim.py:369] (1/2) 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 18:44:57,414 INFO [train.py:968] (1/2) Epoch 3, batch 26800, giga_loss[loss=0.3906, simple_loss=0.4258, pruned_loss=0.1777, over 28011.00 frames. ], tot_loss[loss=0.3674, simple_loss=0.4133, pruned_loss=0.1607, over 5651735.07 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4147, pruned_loss=0.1629, over 5731464.88 frames. ], giga_tot_loss[loss=0.3677, simple_loss=0.4136, pruned_loss=0.1609, over 5651148.68 frames. ], batch size: 412, lr: 9.62e-03, grad_scale: 8.0
+2023-03-01 18:44:57,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:45:05,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:45:06,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4295, 2.0156, 1.4367, 0.6217], device='cuda:1'), covar=tensor([0.1785, 0.0920, 0.1271, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.1269, 0.1214, 0.1278, 0.1100], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 18:45:15,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6595, 3.3354, 3.3751, 1.5944], device='cuda:1'), covar=tensor([0.0697, 0.0614, 0.1095, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0641, 0.0828, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 18:45:23,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-01 18:45:44,164 INFO [train.py:968] (1/2) Epoch 3, batch 26850, giga_loss[loss=0.3248, simple_loss=0.3925, pruned_loss=0.1286, over 28607.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4117, pruned_loss=0.1595, over 5665648.85 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4143, pruned_loss=0.1626, over 5735194.34 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4122, pruned_loss=0.1599, over 5659451.24 frames. ], batch size: 60, lr: 9.62e-03, grad_scale: 8.0
+2023-03-01 18:45:53,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3058, 1.3484, 1.2458, 1.5001], device='cuda:1'), covar=tensor([0.2017, 0.2003, 0.1786, 0.1916], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0850, 0.0941, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 18:46:27,819 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 26900, giga_loss[loss=0.3519, simple_loss=0.4133, pruned_loss=0.1452, over 28737.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.4129, pruned_loss=0.1572, over 5659468.51 frames. ], libri_tot_loss[loss=0.3702, simple_loss=0.4147, pruned_loss=0.1629, over 5723805.13 frames. ], giga_tot_loss[loss=0.3637, simple_loss=0.4129, pruned_loss=0.1572, over 5664728.91 frames. ], batch size: 119, lr: 9.62e-03, grad_scale: 8.0
+2023-03-01 18:46:37,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:46:55,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:47:13,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:47:15,991 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 26950, giga_loss[loss=0.3779, simple_loss=0.4334, pruned_loss=0.1612, over 28843.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.4159, pruned_loss=0.1576, over 5661284.43 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4152, pruned_loss=0.1636, over 5708568.94 frames. ], giga_tot_loss[loss=0.3643, simple_loss=0.4153, pruned_loss=0.1567, over 5677442.36 frames. ], batch size: 199, lr: 9.62e-03, grad_scale: 4.0
+2023-03-01 18:47:43,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:47:49,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:48:00,420 INFO [optim.py:369] (1/2) 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 18:48:03,430 INFO [train.py:968] (1/2) Epoch 3, batch 27000, giga_loss[loss=0.4308, simple_loss=0.4589, pruned_loss=0.2014, over 28352.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4186, pruned_loss=0.1588, over 5675044.11 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.4155, pruned_loss=0.1637, over 5712515.62 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.418, pruned_loss=0.1578, over 5683549.25 frames. ], batch size: 368, lr: 9.61e-03, grad_scale: 4.0
+2023-03-01 18:48:03,430 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 18:48:09,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3582, 2.7455, 1.3093, 1.2940], device='cuda:1'), covar=tensor([0.1003, 0.0478, 0.1104, 0.1656], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0458, 0.0316, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 18:48:12,287 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 18:48:32,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 18:48:48,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7504, 1.2482, 3.4290, 2.9792], device='cuda:1'), covar=tensor([0.1659, 0.1915, 0.0441, 0.0533], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0507, 0.0712, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 18:48:58,240 INFO [train.py:968] (1/2) Epoch 3, batch 27050, giga_loss[loss=0.3763, simple_loss=0.4237, pruned_loss=0.1644, over 28809.00 frames. ], tot_loss[loss=0.3726, simple_loss=0.4211, pruned_loss=0.1621, over 5665741.00 frames. ], libri_tot_loss[loss=0.3713, simple_loss=0.4152, pruned_loss=0.1637, over 5706708.81 frames. ], giga_tot_loss[loss=0.3717, simple_loss=0.4209, pruned_loss=0.1612, over 5676818.41 frames. ], batch size: 186, lr: 9.61e-03, grad_scale: 4.0
+2023-03-01 18:49:13,799 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:49:17,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:49:42,619 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:49:44,898 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7138, 1.5003, 1.2208, 1.3049], device='cuda:1'), covar=tensor([0.0430, 0.0424, 0.0672, 0.0727], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0477, 0.0516, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 18:49:45,941 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 27100, libri_loss[loss=0.3491, simple_loss=0.3989, pruned_loss=0.1496, over 29534.00 frames. ], tot_loss[loss=0.3769, simple_loss=0.4228, pruned_loss=0.1655, over 5656153.64 frames. ], libri_tot_loss[loss=0.3707, simple_loss=0.4146, pruned_loss=0.1634, over 5709893.34 frames. ], giga_tot_loss[loss=0.3768, simple_loss=0.4234, pruned_loss=0.1651, over 5661085.56 frames. ], batch size: 80, lr: 9.61e-03, grad_scale: 2.0
+2023-03-01 18:50:14,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:50:16,183 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 27150, giga_loss[loss=0.3604, simple_loss=0.4157, pruned_loss=0.1526, over 29060.00 frames. ], tot_loss[loss=0.3755, simple_loss=0.421, pruned_loss=0.165, over 5656465.87 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.414, pruned_loss=0.163, over 5713647.23 frames. ], giga_tot_loss[loss=0.3762, simple_loss=0.4222, pruned_loss=0.1651, over 5655767.67 frames. ], batch size: 136, lr: 9.61e-03, grad_scale: 2.0
+2023-03-01 18:50:46,524 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:51:08,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:51:23,377 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 27200, giga_loss[loss=0.3487, simple_loss=0.4122, pruned_loss=0.1426, over 28859.00 frames. ], tot_loss[loss=0.3755, simple_loss=0.4208, pruned_loss=0.1651, over 5652697.86 frames. ], libri_tot_loss[loss=0.3704, simple_loss=0.4142, pruned_loss=0.1633, over 5719339.57 frames. ], giga_tot_loss[loss=0.3758, simple_loss=0.4217, pruned_loss=0.1649, over 5645297.05 frames. ], batch size: 199, lr: 9.61e-03, grad_scale: 4.0
+2023-03-01 18:52:03,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:07,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:15,437 INFO [train.py:968] (1/2) Epoch 3, batch 27250, giga_loss[loss=0.2964, simple_loss=0.3769, pruned_loss=0.108, over 28814.00 frames. ], tot_loss[loss=0.3705, simple_loss=0.4189, pruned_loss=0.1611, over 5641902.49 frames. ], libri_tot_loss[loss=0.3704, simple_loss=0.4142, pruned_loss=0.1634, over 5700912.42 frames. ], giga_tot_loss[loss=0.3707, simple_loss=0.4197, pruned_loss=0.1609, over 5651985.29 frames. ], batch size: 66, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:52:33,474 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:44,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=117570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:52:57,304 INFO [optim.py:369] (1/2) 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 18:53:00,466 INFO [train.py:968] (1/2) Epoch 3, batch 27300, giga_loss[loss=0.3596, simple_loss=0.4155, pruned_loss=0.1519, over 28874.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4169, pruned_loss=0.158, over 5661949.61 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4131, pruned_loss=0.1627, over 5704099.31 frames. ], giga_tot_loss[loss=0.3677, simple_loss=0.4187, pruned_loss=0.1583, over 5665286.55 frames. ], batch size: 186, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:53:05,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8474, 3.5936, 3.5944, 1.8565], device='cuda:1'), covar=tensor([0.0498, 0.0447, 0.0840, 0.1957], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0655, 0.0840, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 18:53:23,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9192, 1.7957, 1.6045, 1.7482], device='cuda:1'), covar=tensor([0.0841, 0.1449, 0.1343, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0774, 0.0631, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 18:53:27,996 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:53:30,192 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:53:34,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:53:59,744 INFO [train.py:968] (1/2) Epoch 3, batch 27350, libri_loss[loss=0.39, simple_loss=0.4334, pruned_loss=0.1732, over 29531.00 frames. ], tot_loss[loss=0.37, simple_loss=0.4192, pruned_loss=0.1604, over 5651665.31 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4137, pruned_loss=0.1629, over 5699622.34 frames. ], giga_tot_loss[loss=0.3704, simple_loss=0.4201, pruned_loss=0.1604, over 5657453.91 frames. ], batch size: 82, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:54:06,930 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:54:45,856 INFO [optim.py:369] (1/2) 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 18:54:47,806 INFO [train.py:968] (1/2) Epoch 3, batch 27400, giga_loss[loss=0.3242, simple_loss=0.3875, pruned_loss=0.1305, over 28554.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4185, pruned_loss=0.1604, over 5661471.95 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4136, pruned_loss=0.1629, over 5701338.94 frames. ], giga_tot_loss[loss=0.3701, simple_loss=0.4195, pruned_loss=0.1603, over 5663750.08 frames. ], batch size: 307, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:55:08,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=117713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:55:12,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=117716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:55:13,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.18 vs. limit=2.0
+2023-03-01 18:55:17,595 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=117722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:55:36,333 INFO [train.py:968] (1/2) Epoch 3, batch 27450, giga_loss[loss=0.3209, simple_loss=0.3742, pruned_loss=0.1339, over 28698.00 frames. ], tot_loss[loss=0.3673, simple_loss=0.4158, pruned_loss=0.1594, over 5643946.45 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4132, pruned_loss=0.1627, over 5685912.61 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.417, pruned_loss=0.1595, over 5659085.80 frames. ], batch size: 99, lr: 9.60e-03, grad_scale: 4.0
+2023-03-01 18:55:40,407 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=117745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 18:56:20,035 INFO [optim.py:369] (1/2) 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 18:56:23,158 INFO [train.py:968] (1/2) Epoch 3, batch 27500, giga_loss[loss=0.3407, simple_loss=0.395, pruned_loss=0.1432, over 28902.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.4153, pruned_loss=0.1608, over 5633230.16 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4135, pruned_loss=0.1628, over 5687878.25 frames. ], giga_tot_loss[loss=0.3687, simple_loss=0.4161, pruned_loss=0.1607, over 5642036.76 frames. ], batch size: 186, lr: 9.59e-03, grad_scale: 4.0
+2023-03-01 18:57:17,436 INFO [train.py:968] (1/2) Epoch 3, batch 27550, giga_loss[loss=0.3309, simple_loss=0.3848, pruned_loss=0.1385, over 28990.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4135, pruned_loss=0.1604, over 5633051.02 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4137, pruned_loss=0.1633, over 5682388.58 frames. ], giga_tot_loss[loss=0.3667, simple_loss=0.414, pruned_loss=0.1598, over 5644823.03 frames. ], batch size: 213, lr: 9.59e-03, grad_scale: 4.0
+2023-03-01 18:57:54,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-01 18:58:00,787 INFO [optim.py:369] (1/2) 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 18:58:03,741 INFO [train.py:968] (1/2) Epoch 3, batch 27600, giga_loss[loss=0.3582, simple_loss=0.4045, pruned_loss=0.1559, over 28618.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4122, pruned_loss=0.1607, over 5638924.28 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4133, pruned_loss=0.163, over 5689196.70 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4129, pruned_loss=0.1604, over 5640839.80 frames. ], batch size: 307, lr: 9.59e-03, grad_scale: 8.0
+2023-03-01 18:58:23,691 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7892, 2.3709, 2.3389, 2.1786], device='cuda:1'), covar=tensor([0.0839, 0.1624, 0.1267, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0775, 0.0633, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 18:58:48,612 INFO [train.py:968] (1/2) Epoch 3, batch 27650, giga_loss[loss=0.353, simple_loss=0.4063, pruned_loss=0.1498, over 28485.00 frames. ], tot_loss[loss=0.3663, simple_loss=0.4116, pruned_loss=0.1604, over 5641242.94 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4127, pruned_loss=0.1625, over 5686802.10 frames. ], giga_tot_loss[loss=0.3669, simple_loss=0.4127, pruned_loss=0.1606, over 5643191.59 frames. ], batch size: 85, lr: 9.59e-03, grad_scale: 8.0
+2023-03-01 18:59:34,581 INFO [optim.py:369] (1/2) 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 18:59:34,883 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 27700, giga_loss[loss=0.3279, simple_loss=0.3879, pruned_loss=0.1339, over 28856.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4088, pruned_loss=0.157, over 5658564.32 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4127, pruned_loss=0.1623, over 5692629.92 frames. ], giga_tot_loss[loss=0.3619, simple_loss=0.4096, pruned_loss=0.1572, over 5653929.21 frames. ], batch size: 119, lr: 9.59e-03, grad_scale: 4.0
+2023-03-01 19:00:20,325 INFO [train.py:968] (1/2) Epoch 3, batch 27750, giga_loss[loss=0.3269, simple_loss=0.3937, pruned_loss=0.1301, over 29076.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.4057, pruned_loss=0.1535, over 5663239.16 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4129, pruned_loss=0.1627, over 5698885.81 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.406, pruned_loss=0.153, over 5652518.43 frames. ], batch size: 155, lr: 9.58e-03, grad_scale: 4.0
+2023-03-01 19:01:10,070 INFO [optim.py:369] (1/2) 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 19:01:11,420 INFO [train.py:968] (1/2) Epoch 3, batch 27800, giga_loss[loss=0.3302, simple_loss=0.3896, pruned_loss=0.1353, over 28793.00 frames. ], tot_loss[loss=0.3536, simple_loss=0.4041, pruned_loss=0.1515, over 5654639.17 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4131, pruned_loss=0.1628, over 5687736.30 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4039, pruned_loss=0.1508, over 5654790.53 frames. ], batch size: 262, lr: 9.58e-03, grad_scale: 4.0
+2023-03-01 19:01:20,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:01:56,794 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:01:59,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:02:05,222 INFO [train.py:968] (1/2) Epoch 3, batch 27850, libri_loss[loss=0.3866, simple_loss=0.4354, pruned_loss=0.1689, over 29515.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4018, pruned_loss=0.1502, over 5648675.96 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4134, pruned_loss=0.1629, over 5690405.60 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4012, pruned_loss=0.1493, over 5645515.89 frames. ], batch size: 81, lr: 9.58e-03, grad_scale: 2.0
+2023-03-01 19:02:28,642 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:02:56,096 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 27900, giga_loss[loss=0.4097, simple_loss=0.4412, pruned_loss=0.1891, over 28335.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.3994, pruned_loss=0.1489, over 5660215.51 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4135, pruned_loss=0.1628, over 5688342.48 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.3984, pruned_loss=0.1479, over 5658556.28 frames. ], batch size: 368, lr: 9.58e-03, grad_scale: 2.0
+2023-03-01 19:03:19,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1893, 1.6631, 1.2662, 0.3993], device='cuda:1'), covar=tensor([0.1229, 0.0705, 0.1199, 0.1633], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1211, 0.1295, 0.1111], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 19:03:48,877 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:03:49,374 INFO [train.py:968] (1/2) Epoch 3, batch 27950, giga_loss[loss=0.3012, simple_loss=0.3758, pruned_loss=0.1133, over 29018.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.402, pruned_loss=0.1507, over 5667080.57 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4138, pruned_loss=0.1629, over 5691583.26 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4007, pruned_loss=0.1495, over 5662292.66 frames. ], batch size: 128, lr: 9.58e-03, grad_scale: 2.0
+2023-03-01 19:03:51,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:04:21,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:04:39,930 INFO [optim.py:369] (1/2) 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 19:04:40,612 INFO [train.py:968] (1/2) Epoch 3, batch 28000, giga_loss[loss=0.3618, simple_loss=0.418, pruned_loss=0.1528, over 28892.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4032, pruned_loss=0.1514, over 5659063.35 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4136, pruned_loss=0.1627, over 5694901.34 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4023, pruned_loss=0.1506, over 5651708.22 frames. ], batch size: 186, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:05:29,463 INFO [train.py:968] (1/2) Epoch 3, batch 28050, giga_loss[loss=0.3303, simple_loss=0.3902, pruned_loss=0.1352, over 28854.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4025, pruned_loss=0.1503, over 5660775.11 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4139, pruned_loss=0.1629, over 5697859.57 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4013, pruned_loss=0.1493, over 5651860.28 frames. ], batch size: 199, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:05:36,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4105, 1.4438, 1.0338, 0.9373], device='cuda:1'), covar=tensor([0.0721, 0.0617, 0.0553, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.1256, 0.0984, 0.1023, 0.1097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-01 19:05:39,353 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 19:06:12,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9807, 1.3957, 4.1650, 3.3197], device='cuda:1'), covar=tensor([0.1540, 0.1945, 0.0280, 0.0749], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0504, 0.0691, 0.0558], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 19:06:19,571 INFO [optim.py:369] (1/2) 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 19:06:20,326 INFO [train.py:968] (1/2) Epoch 3, batch 28100, giga_loss[loss=0.3122, simple_loss=0.367, pruned_loss=0.1287, over 28620.00 frames. ], tot_loss[loss=0.3536, simple_loss=0.4034, pruned_loss=0.1519, over 5656867.79 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4132, pruned_loss=0.1623, over 5702013.37 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4029, pruned_loss=0.1514, over 5645390.49 frames. ], batch size: 92, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:06:58,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 28150, giga_loss[loss=0.4342, simple_loss=0.4606, pruned_loss=0.204, over 28404.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4058, pruned_loss=0.1538, over 5668893.24 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4133, pruned_loss=0.1624, over 5704169.14 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4052, pruned_loss=0.1532, over 5656934.62 frames. ], batch size: 369, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:07:50,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6564, 1.7346, 1.1705, 0.8857], device='cuda:1'), covar=tensor([0.0691, 0.0555, 0.0520, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.1242, 0.0974, 0.1006, 0.1093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 19:07:52,585 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 28200, giga_loss[loss=0.3598, simple_loss=0.4163, pruned_loss=0.1516, over 28689.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4088, pruned_loss=0.1556, over 5663661.43 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4136, pruned_loss=0.1625, over 5698988.14 frames. ], giga_tot_loss[loss=0.3589, simple_loss=0.408, pruned_loss=0.1548, over 5658804.32 frames. ], batch size: 284, lr: 9.57e-03, grad_scale: 4.0
+2023-03-01 19:08:15,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 19:08:42,968 INFO [train.py:968] (1/2) Epoch 3, batch 28250, giga_loss[loss=0.3731, simple_loss=0.4177, pruned_loss=0.1642, over 28957.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4106, pruned_loss=0.157, over 5665539.78 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4137, pruned_loss=0.1628, over 5702499.26 frames. ], giga_tot_loss[loss=0.3609, simple_loss=0.4098, pruned_loss=0.156, over 5657806.41 frames. ], batch size: 213, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:09:02,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9950, 1.1244, 4.1724, 3.4500], device='cuda:1'), covar=tensor([0.2134, 0.2503, 0.0586, 0.0582], device='cuda:1'), in_proj_covar=tensor([0.0541, 0.0505, 0.0697, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 19:09:04,944 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:09:24,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4268, 4.1349, 4.1468, 1.7287], device='cuda:1'), covar=tensor([0.0512, 0.0488, 0.0939, 0.2104], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0646, 0.0825, 0.0590], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 19:09:28,842 INFO [optim.py:369] (1/2) 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 19:09:30,002 INFO [train.py:968] (1/2) Epoch 3, batch 28300, giga_loss[loss=0.3947, simple_loss=0.4151, pruned_loss=0.1871, over 23594.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4114, pruned_loss=0.1585, over 5650547.68 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4132, pruned_loss=0.1627, over 5700975.37 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4111, pruned_loss=0.1576, over 5643376.17 frames. ], batch size: 705, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:09:32,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-01 19:10:11,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2657, 1.9168, 1.3353, 0.5203], device='cuda:1'), covar=tensor([0.1493, 0.0798, 0.1358, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.1297, 0.1233, 0.1284, 0.1118], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 19:10:19,483 INFO [train.py:968] (1/2) Epoch 3, batch 28350, giga_loss[loss=0.3992, simple_loss=0.4426, pruned_loss=0.1779, over 27932.00 frames. ], tot_loss[loss=0.3663, simple_loss=0.4131, pruned_loss=0.1598, over 5653472.94 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4135, pruned_loss=0.1629, over 5702718.84 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4125, pruned_loss=0.1587, over 5645222.84 frames. ], batch size: 412, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:10:43,073 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=118662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:11:07,834 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:11:16,751 INFO [train.py:968] (1/2) Epoch 3, batch 28400, giga_loss[loss=0.3204, simple_loss=0.3906, pruned_loss=0.125, over 28629.00 frames. ], tot_loss[loss=0.3632, simple_loss=0.4123, pruned_loss=0.1571, over 5658594.69 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4134, pruned_loss=0.1629, over 5701544.37 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.412, pruned_loss=0.1562, over 5652421.39 frames. ], batch size: 307, lr: 9.56e-03, grad_scale: 8.0
+2023-03-01 19:11:49,265 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 28450, giga_loss[loss=0.3332, simple_loss=0.3876, pruned_loss=0.1395, over 28950.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4127, pruned_loss=0.1579, over 5666823.57 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4127, pruned_loss=0.1624, over 5707316.82 frames. ], giga_tot_loss[loss=0.3641, simple_loss=0.4131, pruned_loss=0.1576, over 5655342.84 frames. ], batch size: 136, lr: 9.56e-03, grad_scale: 4.0
+2023-03-01 19:12:58,170 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 28500, giga_loss[loss=0.3158, simple_loss=0.376, pruned_loss=0.1277, over 28909.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4118, pruned_loss=0.1579, over 5672865.91 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.413, pruned_loss=0.1625, over 5712450.30 frames. ], giga_tot_loss[loss=0.3633, simple_loss=0.4118, pruned_loss=0.1574, over 5658133.45 frames. ], batch size: 112, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:13:17,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=118808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:13:55,685 INFO [train.py:968] (1/2) Epoch 3, batch 28550, giga_loss[loss=0.3349, simple_loss=0.3935, pruned_loss=0.1382, over 29074.00 frames. ], tot_loss[loss=0.3639, simple_loss=0.4114, pruned_loss=0.1582, over 5675745.17 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4128, pruned_loss=0.1623, over 5708784.94 frames. ], giga_tot_loss[loss=0.3637, simple_loss=0.4116, pruned_loss=0.1579, over 5666532.90 frames. ], batch size: 155, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:14:50,960 INFO [optim.py:369] (1/2) 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 19:14:50,973 INFO [train.py:968] (1/2) Epoch 3, batch 28600, giga_loss[loss=0.3571, simple_loss=0.3923, pruned_loss=0.161, over 28880.00 frames. ], tot_loss[loss=0.3608, simple_loss=0.4087, pruned_loss=0.1565, over 5670921.81 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4131, pruned_loss=0.1625, over 5700795.97 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4085, pruned_loss=0.156, over 5669889.32 frames. ], batch size: 99, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:15:34,006 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 28650, giga_loss[loss=0.3222, simple_loss=0.3847, pruned_loss=0.1299, over 28800.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4095, pruned_loss=0.1577, over 5676316.81 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4129, pruned_loss=0.1623, over 5705012.36 frames. ], giga_tot_loss[loss=0.3621, simple_loss=0.4095, pruned_loss=0.1573, over 5671312.41 frames. ], batch size: 119, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:15:50,953 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=118951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:15:54,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=118954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:16:23,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=118983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:16:28,353 INFO [optim.py:369] (1/2) 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 19:16:28,365 INFO [train.py:968] (1/2) Epoch 3, batch 28700, libri_loss[loss=0.3918, simple_loss=0.4384, pruned_loss=0.1726, over 26159.00 frames. ], tot_loss[loss=0.3629, simple_loss=0.4096, pruned_loss=0.1581, over 5648962.03 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4132, pruned_loss=0.1624, over 5696461.00 frames. ], giga_tot_loss[loss=0.3623, simple_loss=0.4091, pruned_loss=0.1577, over 5652392.17 frames. ], batch size: 136, lr: 9.55e-03, grad_scale: 4.0
+2023-03-01 19:16:45,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5657, 2.8543, 1.5771, 1.3920], device='cuda:1'), covar=tensor([0.0782, 0.0364, 0.0758, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0465, 0.0316, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 19:17:01,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:11,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7182, 1.1838, 3.2634, 2.9267], device='cuda:1'), covar=tensor([0.1550, 0.1900, 0.0450, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0553, 0.0516, 0.0714, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 19:17:11,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:15,011 INFO [train.py:968] (1/2) Epoch 3, batch 28750, giga_loss[loss=0.376, simple_loss=0.4204, pruned_loss=0.1658, over 28943.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.4091, pruned_loss=0.1581, over 5649135.24 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4127, pruned_loss=0.1621, over 5699540.65 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4092, pruned_loss=0.1579, over 5647907.05 frames. ], batch size: 213, lr: 9.54e-03, grad_scale: 2.0
+2023-03-01 19:17:30,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:53,203 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:17:55,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:18:02,421 INFO [train.py:968] (1/2) Epoch 3, batch 28800, giga_loss[loss=0.3508, simple_loss=0.4048, pruned_loss=0.1483, over 28999.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4108, pruned_loss=0.1595, over 5655449.08 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4131, pruned_loss=0.1623, over 5702546.69 frames. ], giga_tot_loss[loss=0.3643, simple_loss=0.4104, pruned_loss=0.1591, over 5651053.00 frames. ], batch size: 136, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:18:03,348 INFO [optim.py:369] (1/2) 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 19:18:08,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:18:15,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-01 19:18:22,521 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:18:53,828 INFO [train.py:968] (1/2) Epoch 3, batch 28850, giga_loss[loss=0.3696, simple_loss=0.4026, pruned_loss=0.1683, over 28667.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4114, pruned_loss=0.1607, over 5650837.05 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4129, pruned_loss=0.1623, over 5706731.55 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4112, pruned_loss=0.1604, over 5642696.51 frames. ], batch size: 92, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:19:05,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5672, 3.3135, 3.3054, 1.9804], device='cuda:1'), covar=tensor([0.0544, 0.0558, 0.1030, 0.1676], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0656, 0.0833, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 19:19:13,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4221, 1.5161, 1.2277, 1.8286], device='cuda:1'), covar=tensor([0.2032, 0.1953, 0.1879, 0.2213], device='cuda:1'), in_proj_covar=tensor([0.1038, 0.0834, 0.0937, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 19:19:25,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6218, 3.4258, 1.6897, 1.5797], device='cuda:1'), covar=tensor([0.0797, 0.0394, 0.0754, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0466, 0.0314, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 19:19:33,646 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:19:37,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:19:43,414 INFO [train.py:968] (1/2) Epoch 3, batch 28900, giga_loss[loss=0.3161, simple_loss=0.3788, pruned_loss=0.1267, over 29023.00 frames. ], tot_loss[loss=0.3682, simple_loss=0.4122, pruned_loss=0.1621, over 5648313.74 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4131, pruned_loss=0.1625, over 5704249.10 frames. ], giga_tot_loss[loss=0.3676, simple_loss=0.4119, pruned_loss=0.1617, over 5643399.51 frames. ], batch size: 155, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:19:44,289 INFO [optim.py:369] (1/2) 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 19:19:52,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:19:56,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:06,505 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:26,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:32,713 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:33,036 INFO [train.py:968] (1/2) Epoch 3, batch 28950, giga_loss[loss=0.3329, simple_loss=0.3966, pruned_loss=0.1346, over 28928.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4124, pruned_loss=0.1629, over 5646102.67 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4129, pruned_loss=0.1625, over 5698251.68 frames. ], giga_tot_loss[loss=0.3687, simple_loss=0.4123, pruned_loss=0.1626, over 5645635.69 frames. ], batch size: 174, lr: 9.54e-03, grad_scale: 4.0
+2023-03-01 19:20:35,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:20:58,368 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-01 19:21:04,934 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:21:27,262 INFO [train.py:968] (1/2) Epoch 3, batch 29000, giga_loss[loss=0.3689, simple_loss=0.4214, pruned_loss=0.1582, over 28825.00 frames. ], tot_loss[loss=0.3691, simple_loss=0.4128, pruned_loss=0.1627, over 5640675.47 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4124, pruned_loss=0.1622, over 5701150.09 frames. ], giga_tot_loss[loss=0.3693, simple_loss=0.4132, pruned_loss=0.1627, over 5637121.95 frames. ], batch size: 174, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:21:27,855 INFO [optim.py:369] (1/2) 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 19:22:12,337 INFO [train.py:968] (1/2) Epoch 3, batch 29050, giga_loss[loss=0.4205, simple_loss=0.4482, pruned_loss=0.1965, over 28268.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.4137, pruned_loss=0.162, over 5638068.89 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4129, pruned_loss=0.1624, over 5687419.48 frames. ], giga_tot_loss[loss=0.3686, simple_loss=0.4136, pruned_loss=0.1618, over 5646687.22 frames. ], batch size: 368, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:23:01,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0311, 2.3020, 2.0829, 2.0111], device='cuda:1'), covar=tensor([0.1546, 0.1741, 0.1144, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0816, 0.0747, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 19:23:01,593 INFO [train.py:968] (1/2) Epoch 3, batch 29100, giga_loss[loss=0.3344, simple_loss=0.3885, pruned_loss=0.1401, over 28981.00 frames. ], tot_loss[loss=0.3721, simple_loss=0.4158, pruned_loss=0.1642, over 5647792.49 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4129, pruned_loss=0.1626, over 5689821.15 frames. ], giga_tot_loss[loss=0.3717, simple_loss=0.4157, pruned_loss=0.1639, over 5651851.42 frames. ], batch size: 106, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:23:02,300 INFO [optim.py:369] (1/2) 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 19:23:12,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:23:31,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2023, 1.2680, 1.0222, 1.1925], device='cuda:1'), covar=tensor([0.0610, 0.0507, 0.0980, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0493, 0.0530, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 19:23:46,971 INFO [train.py:968] (1/2) Epoch 3, batch 29150, giga_loss[loss=0.4307, simple_loss=0.4539, pruned_loss=0.2038, over 27927.00 frames. ], tot_loss[loss=0.3744, simple_loss=0.4176, pruned_loss=0.1656, over 5641157.86 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4133, pruned_loss=0.1629, over 5667268.18 frames. ], giga_tot_loss[loss=0.3738, simple_loss=0.4174, pruned_loss=0.1651, over 5663704.36 frames. ], batch size: 412, lr: 9.53e-03, grad_scale: 4.0
+2023-03-01 19:24:12,141 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:24:30,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:24:33,309 INFO [train.py:968] (1/2) Epoch 3, batch 29200, libri_loss[loss=0.428, simple_loss=0.4679, pruned_loss=0.194, over 25632.00 frames. ], tot_loss[loss=0.3756, simple_loss=0.4185, pruned_loss=0.1664, over 5647461.92 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4135, pruned_loss=0.1629, over 5666593.80 frames. ], giga_tot_loss[loss=0.3752, simple_loss=0.4181, pruned_loss=0.1661, over 5665976.98 frames. ], batch size: 136, lr: 9.53e-03, grad_scale: 8.0
+2023-03-01 19:24:34,008 INFO [optim.py:369] (1/2) 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 19:25:18,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4747, 1.3721, 1.3395, 1.3435], device='cuda:1'), covar=tensor([0.0811, 0.1216, 0.1312, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0780, 0.0635, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 19:25:25,050 INFO [train.py:968] (1/2) Epoch 3, batch 29250, giga_loss[loss=0.4108, simple_loss=0.4503, pruned_loss=0.1856, over 27916.00 frames. ], tot_loss[loss=0.3748, simple_loss=0.4186, pruned_loss=0.1656, over 5649417.45 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.413, pruned_loss=0.1625, over 5671193.24 frames. ], giga_tot_loss[loss=0.3751, simple_loss=0.4188, pruned_loss=0.1657, over 5659795.77 frames. ], batch size: 412, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:25:26,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:25:30,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:25:34,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5790, 1.8767, 1.6360, 1.6561], device='cuda:1'), covar=tensor([0.1364, 0.1797, 0.1186, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0808, 0.0744, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 19:25:57,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=119574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:26:14,914 INFO [train.py:968] (1/2) Epoch 3, batch 29300, libri_loss[loss=0.3059, simple_loss=0.3563, pruned_loss=0.1277, over 29631.00 frames. ], tot_loss[loss=0.3718, simple_loss=0.4169, pruned_loss=0.1633, over 5649033.91 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4124, pruned_loss=0.1621, over 5677475.11 frames. ], giga_tot_loss[loss=0.3729, simple_loss=0.4179, pruned_loss=0.1639, over 5650877.57 frames. ], batch size: 69, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:26:16,137 INFO [optim.py:369] (1/2) 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 19:26:51,075 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3177, 1.3323, 1.1779, 1.4891], device='cuda:1'), covar=tensor([0.1904, 0.1784, 0.1694, 0.1978], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0835, 0.0951, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 19:26:56,188 INFO [train.py:968] (1/2) Epoch 3, batch 29350, giga_loss[loss=0.3372, simple_loss=0.3894, pruned_loss=0.1425, over 28529.00 frames. ], tot_loss[loss=0.3669, simple_loss=0.4139, pruned_loss=0.1599, over 5668424.23 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4125, pruned_loss=0.162, over 5682844.56 frames. ], giga_tot_loss[loss=0.3678, simple_loss=0.4146, pruned_loss=0.1605, over 5664498.41 frames. ], batch size: 307, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:26:59,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8961, 2.1177, 2.4750, 2.0604], device='cuda:1'), covar=tensor([0.0856, 0.1656, 0.1035, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0788, 0.0639, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 19:27:16,407 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-01 19:27:35,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3078, 1.3476, 0.8016, 1.0358], device='cuda:1'), covar=tensor([0.0490, 0.0505, 0.0481, 0.0527], device='cuda:1'), in_proj_covar=tensor([0.1245, 0.0975, 0.0986, 0.1064], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 19:27:41,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1522, 1.5987, 1.2110, 1.3768], device='cuda:1'), covar=tensor([0.0937, 0.0348, 0.0419, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0151, 0.0155, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0043, 0.0032, 0.0029, 0.0048], device='cuda:1')
+2023-03-01 19:27:43,786 INFO [train.py:968] (1/2) Epoch 3, batch 29400, giga_loss[loss=0.3698, simple_loss=0.4224, pruned_loss=0.1586, over 29053.00 frames. ], tot_loss[loss=0.3674, simple_loss=0.4136, pruned_loss=0.1606, over 5656396.12 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4127, pruned_loss=0.1621, over 5685344.46 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.414, pruned_loss=0.1609, over 5650779.43 frames. ], batch size: 155, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:27:45,222 INFO [optim.py:369] (1/2) 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 19:28:28,602 INFO [train.py:968] (1/2) Epoch 3, batch 29450, giga_loss[loss=0.3432, simple_loss=0.3995, pruned_loss=0.1434, over 28856.00 frames. ], tot_loss[loss=0.368, simple_loss=0.4147, pruned_loss=0.1607, over 5667226.71 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4133, pruned_loss=0.1626, over 5687579.79 frames. ], giga_tot_loss[loss=0.3677, simple_loss=0.4146, pruned_loss=0.1604, over 5660381.40 frames. ], batch size: 199, lr: 9.52e-03, grad_scale: 4.0
+2023-03-01 19:29:21,275 INFO [train.py:968] (1/2) Epoch 3, batch 29500, libri_loss[loss=0.3675, simple_loss=0.4205, pruned_loss=0.1572, over 28633.00 frames. ], tot_loss[loss=0.3704, simple_loss=0.4161, pruned_loss=0.1623, over 5659752.41 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4135, pruned_loss=0.1629, over 5689611.77 frames. ], giga_tot_loss[loss=0.3699, simple_loss=0.4159, pruned_loss=0.1619, over 5651904.32 frames. ], batch size: 106, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:29:22,522 INFO [optim.py:369] (1/2) 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 19:30:11,595 INFO [train.py:968] (1/2) Epoch 3, batch 29550, giga_loss[loss=0.4047, simple_loss=0.4395, pruned_loss=0.1849, over 28703.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4146, pruned_loss=0.1622, over 5663681.88 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4134, pruned_loss=0.1627, over 5693029.76 frames. ], giga_tot_loss[loss=0.3692, simple_loss=0.4145, pruned_loss=0.162, over 5653928.00 frames. ], batch size: 262, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:30:14,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:30:29,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=119861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:30:58,444 INFO [train.py:968] (1/2) Epoch 3, batch 29600, giga_loss[loss=0.4019, simple_loss=0.4372, pruned_loss=0.1833, over 28707.00 frames. ], tot_loss[loss=0.3695, simple_loss=0.4141, pruned_loss=0.1624, over 5656776.74 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4133, pruned_loss=0.1627, over 5696285.22 frames. ], giga_tot_loss[loss=0.3694, simple_loss=0.4141, pruned_loss=0.1623, over 5645745.35 frames. ], batch size: 242, lr: 9.51e-03, grad_scale: 8.0
+2023-03-01 19:31:00,208 INFO [optim.py:369] (1/2) 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 19:31:03,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-01 19:31:14,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=119907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:31:33,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7404, 5.4116, 5.4060, 2.4830], device='cuda:1'), covar=tensor([0.0364, 0.0346, 0.0830, 0.1696], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0662, 0.0834, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 19:31:46,146 INFO [train.py:968] (1/2) Epoch 3, batch 29650, giga_loss[loss=0.3472, simple_loss=0.4044, pruned_loss=0.145, over 28962.00 frames. ], tot_loss[loss=0.3696, simple_loss=0.4147, pruned_loss=0.1622, over 5659078.97 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4135, pruned_loss=0.1628, over 5688523.56 frames. ], giga_tot_loss[loss=0.3692, simple_loss=0.4145, pruned_loss=0.162, over 5656832.06 frames. ], batch size: 227, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:32:08,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0919, 1.1898, 4.6137, 3.4832], device='cuda:1'), covar=tensor([0.1595, 0.2010, 0.0302, 0.0512], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0509, 0.0705, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 19:32:35,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=119986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:32:37,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=119989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:32:38,730 INFO [train.py:968] (1/2) Epoch 3, batch 29700, giga_loss[loss=0.3967, simple_loss=0.4317, pruned_loss=0.1808, over 28415.00 frames. ], tot_loss[loss=0.3718, simple_loss=0.4162, pruned_loss=0.1638, over 5652083.81 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4139, pruned_loss=0.163, over 5695397.71 frames. ], giga_tot_loss[loss=0.3714, simple_loss=0.4158, pruned_loss=0.1635, over 5642633.27 frames. ], batch size: 369, lr: 9.51e-03, grad_scale: 4.0
+2023-03-01 19:32:40,725 INFO [optim.py:369] (1/2) 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 19:32:50,017 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:32:52,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:33:02,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:33:18,102 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:33:21,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2391, 1.3106, 1.1357, 0.7237], device='cuda:1'), covar=tensor([0.0640, 0.0620, 0.0387, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.1249, 0.1002, 0.0988, 0.1065], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 19:33:22,104 INFO [train.py:968] (1/2) Epoch 3, batch 29750, giga_loss[loss=0.3532, simple_loss=0.411, pruned_loss=0.1477, over 28668.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4141, pruned_loss=0.161, over 5675906.00 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.4138, pruned_loss=0.1628, over 5701449.43 frames. ], giga_tot_loss[loss=0.3679, simple_loss=0.414, pruned_loss=0.161, over 5661894.53 frames. ], batch size: 262, lr: 9.50e-03, grad_scale: 4.0
+2023-03-01 19:33:43,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=120058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:34:13,602 INFO [train.py:968] (1/2) Epoch 3, batch 29800, giga_loss[loss=0.3559, simple_loss=0.4084, pruned_loss=0.1517, over 28856.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4133, pruned_loss=0.1598, over 5672964.62 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4138, pruned_loss=0.1627, over 5703510.91 frames. ], giga_tot_loss[loss=0.3663, simple_loss=0.4132, pruned_loss=0.1598, over 5659824.05 frames. ], batch size: 199, lr: 9.50e-03, grad_scale: 4.0
+2023-03-01 19:34:16,047 INFO [optim.py:369] (1/2) 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 19:34:27,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 19:35:03,660 INFO [train.py:968] (1/2) Epoch 3, batch 29850, giga_loss[loss=0.3392, simple_loss=0.3959, pruned_loss=0.1412, over 28697.00 frames. ], tot_loss[loss=0.3645, simple_loss=0.4122, pruned_loss=0.1584, over 5656386.71 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4141, pruned_loss=0.1629, over 5686029.73 frames. ], giga_tot_loss[loss=0.3639, simple_loss=0.4118, pruned_loss=0.158, over 5661658.99 frames. ], batch size: 242, lr: 9.50e-03, grad_scale: 2.0
+2023-03-01 19:35:50,038 INFO [train.py:968] (1/2) Epoch 3, batch 29900, giga_loss[loss=0.3207, simple_loss=0.3829, pruned_loss=0.1292, over 28128.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4099, pruned_loss=0.1571, over 5656955.42 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4134, pruned_loss=0.1626, over 5686430.07 frames. ], giga_tot_loss[loss=0.3621, simple_loss=0.4101, pruned_loss=0.157, over 5660192.76 frames. ], batch size: 77, lr: 9.50e-03, grad_scale: 2.0
+2023-03-01 19:35:54,305 INFO [optim.py:369] (1/2) 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 19:36:38,569 INFO [train.py:968] (1/2) Epoch 3, batch 29950, giga_loss[loss=0.301, simple_loss=0.3696, pruned_loss=0.1162, over 28404.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4091, pruned_loss=0.1572, over 5658082.01 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4134, pruned_loss=0.1624, over 5690174.55 frames. ], giga_tot_loss[loss=0.3618, simple_loss=0.4092, pruned_loss=0.1572, over 5656533.54 frames. ], batch size: 60, lr: 9.50e-03, grad_scale: 2.0
+2023-03-01 19:36:41,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3546, 1.4430, 1.1832, 1.4024], device='cuda:1'), covar=tensor([0.0867, 0.0340, 0.0388, 0.0940], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0152, 0.0155, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0043, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 19:36:52,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7590, 2.6926, 1.8561, 0.7230], device='cuda:1'), covar=tensor([0.2592, 0.1231, 0.1557, 0.2723], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.1242, 0.1291, 0.1118], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 19:37:14,035 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4716, 2.5768, 1.5104, 1.3527], device='cuda:1'), covar=tensor([0.0854, 0.0435, 0.0762, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0465, 0.0311, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 19:37:14,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:37:20,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.21 vs. limit=5.0
+2023-03-01 19:37:25,411 INFO [train.py:968] (1/2) Epoch 3, batch 30000, giga_loss[loss=0.3921, simple_loss=0.4248, pruned_loss=0.1796, over 27630.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4071, pruned_loss=0.1563, over 5660336.52 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4133, pruned_loss=0.1623, over 5694531.76 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4072, pruned_loss=0.1563, over 5654871.32 frames. ], batch size: 472, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:37:25,411 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 19:37:34,312 INFO [train.py:1012] (1/2) Epoch 3, validation: loss=0.2555, simple_loss=0.3575, pruned_loss=0.07679, over 944034.00 frames. 
+2023-03-01 19:37:34,312 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 19:37:37,434 INFO [optim.py:369] (1/2) 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 19:37:56,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-01 19:38:24,384 INFO [train.py:968] (1/2) Epoch 3, batch 30050, giga_loss[loss=0.3204, simple_loss=0.3796, pruned_loss=0.1306, over 28658.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4041, pruned_loss=0.1544, over 5675779.07 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4141, pruned_loss=0.1628, over 5695637.10 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.403, pruned_loss=0.1536, over 5669602.39 frames. ], batch size: 262, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:38:46,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1544, 0.8446, 0.7413, 1.3180], device='cuda:1'), covar=tensor([0.0864, 0.0381, 0.0415, 0.0917], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0152, 0.0154, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0043, 0.0032, 0.0029, 0.0048], device='cuda:1')
+2023-03-01 19:39:08,620 INFO [train.py:968] (1/2) Epoch 3, batch 30100, giga_loss[loss=0.3592, simple_loss=0.4054, pruned_loss=0.1565, over 28300.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4019, pruned_loss=0.1534, over 5689434.51 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4136, pruned_loss=0.1623, over 5699264.27 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4013, pruned_loss=0.1531, over 5680950.03 frames. ], batch size: 368, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:39:12,169 INFO [optim.py:369] (1/2) 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 19:39:41,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0139, 1.2571, 4.2553, 3.2924], device='cuda:1'), covar=tensor([0.1592, 0.1965, 0.0354, 0.0508], device='cuda:1'), in_proj_covar=tensor([0.0556, 0.0522, 0.0732, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 19:39:41,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:39:44,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:39:50,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=120433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:39:57,887 INFO [train.py:968] (1/2) Epoch 3, batch 30150, giga_loss[loss=0.3319, simple_loss=0.3829, pruned_loss=0.1404, over 28531.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4009, pruned_loss=0.1523, over 5692506.48 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4137, pruned_loss=0.1622, over 5701486.68 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.3999, pruned_loss=0.1518, over 5683271.01 frames. ], batch size: 60, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:40:12,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8883, 1.7638, 1.6257, 1.6003], device='cuda:1'), covar=tensor([0.0967, 0.1771, 0.1447, 0.1464], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0770, 0.0622, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 19:40:15,896 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:40:46,615 INFO [train.py:968] (1/2) Epoch 3, batch 30200, libri_loss[loss=0.3747, simple_loss=0.4107, pruned_loss=0.1693, over 29404.00 frames. ], tot_loss[loss=0.35, simple_loss=0.3996, pruned_loss=0.1502, over 5692055.07 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4126, pruned_loss=0.1616, over 5706202.38 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.3994, pruned_loss=0.1501, over 5679722.64 frames. ], batch size: 92, lr: 9.49e-03, grad_scale: 4.0
+2023-03-01 19:40:52,030 INFO [optim.py:369] (1/2) 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 19:41:40,860 INFO [train.py:968] (1/2) Epoch 3, batch 30250, giga_loss[loss=0.3454, simple_loss=0.3921, pruned_loss=0.1493, over 26626.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3966, pruned_loss=0.1462, over 5671998.67 frames. ], libri_tot_loss[loss=0.3682, simple_loss=0.4127, pruned_loss=0.1618, over 5696338.57 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.396, pruned_loss=0.1455, over 5670995.44 frames. ], batch size: 555, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:42:18,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=120576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:42:20,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=120579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:42:30,433 INFO [train.py:968] (1/2) Epoch 3, batch 30300, giga_loss[loss=0.3001, simple_loss=0.3615, pruned_loss=0.1194, over 28872.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3924, pruned_loss=0.1424, over 5671010.51 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.412, pruned_loss=0.1614, over 5699900.27 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.392, pruned_loss=0.1417, over 5665781.09 frames. ], batch size: 112, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:42:34,856 INFO [optim.py:369] (1/2) 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 19:42:48,954 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=120608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:43:11,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1181, 1.2788, 0.9741, 0.9706], device='cuda:1'), covar=tensor([0.0733, 0.0518, 0.0420, 0.0542], device='cuda:1'), in_proj_covar=tensor([0.1239, 0.0976, 0.0959, 0.1043], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 19:43:18,032 INFO [train.py:968] (1/2) Epoch 3, batch 30350, giga_loss[loss=0.2922, simple_loss=0.364, pruned_loss=0.1102, over 28854.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3882, pruned_loss=0.139, over 5660106.73 frames. ], libri_tot_loss[loss=0.3663, simple_loss=0.4107, pruned_loss=0.1609, over 5700176.32 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3879, pruned_loss=0.1377, over 5653110.01 frames. ], batch size: 186, lr: 9.48e-03, grad_scale: 2.0
+2023-03-01 19:43:23,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7186, 1.7207, 1.1872, 1.3969], device='cuda:1'), covar=tensor([0.0678, 0.0578, 0.0999, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0472, 0.0520, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 19:44:06,859 INFO [train.py:968] (1/2) Epoch 3, batch 30400, giga_loss[loss=0.3075, simple_loss=0.3673, pruned_loss=0.1238, over 27611.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3839, pruned_loss=0.1347, over 5661976.25 frames. ], libri_tot_loss[loss=0.366, simple_loss=0.4103, pruned_loss=0.1608, over 5703666.64 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3835, pruned_loss=0.1333, over 5652443.56 frames. ], batch size: 472, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:44:11,367 INFO [optim.py:369] (1/2) 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 19:44:29,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-01 19:44:38,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3331, 1.2530, 0.9540, 1.4108], device='cuda:1'), covar=tensor([0.0903, 0.0367, 0.0445, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0150, 0.0156, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 19:44:46,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.15 vs. limit=2.0
+2023-03-01 19:44:58,386 INFO [train.py:968] (1/2) Epoch 3, batch 30450, giga_loss[loss=0.2962, simple_loss=0.3744, pruned_loss=0.109, over 28895.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3829, pruned_loss=0.1324, over 5654104.11 frames. ], libri_tot_loss[loss=0.3659, simple_loss=0.41, pruned_loss=0.1609, over 5707525.11 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3823, pruned_loss=0.1306, over 5641882.15 frames. ], batch size: 227, lr: 9.48e-03, grad_scale: 4.0
+2023-03-01 19:45:49,219 INFO [train.py:968] (1/2) Epoch 3, batch 30500, giga_loss[loss=0.3508, simple_loss=0.401, pruned_loss=0.1503, over 27958.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.383, pruned_loss=0.1323, over 5648464.31 frames. ], libri_tot_loss[loss=0.3652, simple_loss=0.4092, pruned_loss=0.1605, over 5704371.44 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3825, pruned_loss=0.1304, over 5640317.26 frames. ], batch size: 412, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:45:54,689 INFO [optim.py:369] (1/2) 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 19:46:38,517 INFO [train.py:968] (1/2) Epoch 3, batch 30550, giga_loss[loss=0.2638, simple_loss=0.3471, pruned_loss=0.09024, over 28998.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3817, pruned_loss=0.1316, over 5649181.38 frames. ], libri_tot_loss[loss=0.3646, simple_loss=0.4086, pruned_loss=0.1603, over 5709903.29 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3808, pruned_loss=0.1292, over 5635509.19 frames. ], batch size: 155, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:47:34,899 INFO [train.py:968] (1/2) Epoch 3, batch 30600, giga_loss[loss=0.2669, simple_loss=0.3425, pruned_loss=0.09563, over 28707.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.378, pruned_loss=0.1286, over 5643971.42 frames. ], libri_tot_loss[loss=0.3647, simple_loss=0.4087, pruned_loss=0.1604, over 5710509.77 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3772, pruned_loss=0.1265, over 5632410.28 frames. ], batch size: 284, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:47:36,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2994, 1.3565, 0.9793, 1.3837], device='cuda:1'), covar=tensor([0.0876, 0.0367, 0.0439, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0151, 0.0156, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 19:47:39,839 INFO [optim.py:369] (1/2) 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 19:47:43,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8190, 1.6760, 1.3029, 1.4376], device='cuda:1'), covar=tensor([0.0604, 0.0589, 0.0916, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0457, 0.0506, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 19:48:28,037 INFO [train.py:968] (1/2) Epoch 3, batch 30650, giga_loss[loss=0.2921, simple_loss=0.3668, pruned_loss=0.1087, over 28642.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3751, pruned_loss=0.1263, over 5638757.75 frames. ], libri_tot_loss[loss=0.3642, simple_loss=0.4083, pruned_loss=0.1601, over 5703263.49 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3744, pruned_loss=0.1245, over 5635651.09 frames. ], batch size: 242, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:48:52,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2342, 1.5099, 1.1225, 0.6133], device='cuda:1'), covar=tensor([0.0785, 0.0538, 0.0404, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.1209, 0.0928, 0.0935, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 19:49:15,383 INFO [train.py:968] (1/2) Epoch 3, batch 30700, libri_loss[loss=0.3456, simple_loss=0.3848, pruned_loss=0.1532, over 29503.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3753, pruned_loss=0.126, over 5645151.24 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4074, pruned_loss=0.1595, over 5708333.62 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3748, pruned_loss=0.1243, over 5636976.06 frames. ], batch size: 82, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:49:20,938 INFO [optim.py:369] (1/2) 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 19:50:04,208 INFO [train.py:968] (1/2) Epoch 3, batch 30750, giga_loss[loss=0.2657, simple_loss=0.3458, pruned_loss=0.09278, over 28603.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3741, pruned_loss=0.1247, over 5636890.69 frames. ], libri_tot_loss[loss=0.3638, simple_loss=0.4076, pruned_loss=0.1599, over 5692988.82 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3726, pruned_loss=0.122, over 5642793.71 frames. ], batch size: 242, lr: 9.47e-03, grad_scale: 4.0
+2023-03-01 19:50:16,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-01 19:50:34,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:50:55,647 INFO [train.py:968] (1/2) Epoch 3, batch 30800, giga_loss[loss=0.281, simple_loss=0.3605, pruned_loss=0.1007, over 28912.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3718, pruned_loss=0.1224, over 5634148.18 frames. ], libri_tot_loss[loss=0.3637, simple_loss=0.4075, pruned_loss=0.16, over 5684833.85 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3699, pruned_loss=0.1194, over 5645347.36 frames. ], batch size: 136, lr: 9.46e-03, grad_scale: 8.0
+2023-03-01 19:51:00,280 INFO [optim.py:369] (1/2) 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 19:51:19,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7274, 1.9028, 1.7408, 1.7973], device='cuda:1'), covar=tensor([0.1068, 0.1635, 0.1182, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0752, 0.0614, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 19:51:47,105 INFO [train.py:968] (1/2) Epoch 3, batch 30850, giga_loss[loss=0.2833, simple_loss=0.3531, pruned_loss=0.1068, over 28765.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3667, pruned_loss=0.1191, over 5631694.78 frames. ], libri_tot_loss[loss=0.3635, simple_loss=0.4072, pruned_loss=0.1599, over 5688756.56 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3647, pruned_loss=0.1161, over 5636048.73 frames. ], batch size: 284, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:52:35,868 INFO [train.py:968] (1/2) Epoch 3, batch 30900, giga_loss[loss=0.3129, simple_loss=0.3696, pruned_loss=0.1281, over 29058.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3652, pruned_loss=0.1188, over 5643492.58 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4066, pruned_loss=0.1596, over 5692969.86 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.1159, over 5642429.45 frames. ], batch size: 128, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:52:44,181 INFO [optim.py:369] (1/2) 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 19:53:27,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5632, 2.0052, 1.7489, 1.8263], device='cuda:1'), covar=tensor([0.1303, 0.1446, 0.1147, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0783, 0.0740, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 19:53:28,610 INFO [train.py:968] (1/2) Epoch 3, batch 30950, giga_loss[loss=0.3066, simple_loss=0.3778, pruned_loss=0.1177, over 28473.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3635, pruned_loss=0.1183, over 5634763.87 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4062, pruned_loss=0.1596, over 5697173.07 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3616, pruned_loss=0.1153, over 5629218.48 frames. ], batch size: 336, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:54:18,966 INFO [train.py:968] (1/2) Epoch 3, batch 31000, giga_loss[loss=0.3093, simple_loss=0.3717, pruned_loss=0.1234, over 28914.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3666, pruned_loss=0.1209, over 5624960.98 frames. ], libri_tot_loss[loss=0.3623, simple_loss=0.4058, pruned_loss=0.1594, over 5692932.43 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3638, pruned_loss=0.117, over 5621828.26 frames. ], batch size: 112, lr: 9.46e-03, grad_scale: 4.0
+2023-03-01 19:54:26,672 INFO [optim.py:369] (1/2) 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 19:54:31,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3408, 1.7369, 1.5205, 1.4963], device='cuda:1'), covar=tensor([0.1514, 0.1915, 0.1223, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0781, 0.0739, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 19:55:20,580 INFO [train.py:968] (1/2) Epoch 3, batch 31050, giga_loss[loss=0.3821, simple_loss=0.4347, pruned_loss=0.1648, over 28782.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3695, pruned_loss=0.1207, over 5626631.36 frames. ], libri_tot_loss[loss=0.3621, simple_loss=0.4056, pruned_loss=0.1593, over 5684857.21 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3671, pruned_loss=0.1174, over 5630049.54 frames. ], batch size: 243, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:55:31,956 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-01 19:56:20,828 INFO [train.py:968] (1/2) Epoch 3, batch 31100, giga_loss[loss=0.3014, simple_loss=0.3691, pruned_loss=0.1168, over 28885.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3708, pruned_loss=0.1213, over 5646208.56 frames. ], libri_tot_loss[loss=0.3617, simple_loss=0.4052, pruned_loss=0.1591, over 5686787.79 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3682, pruned_loss=0.1176, over 5645549.71 frames. ], batch size: 199, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:56:27,683 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:56:28,105 INFO [optim.py:369] (1/2) 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 19:57:23,434 INFO [train.py:968] (1/2) Epoch 3, batch 31150, giga_loss[loss=0.2772, simple_loss=0.3507, pruned_loss=0.1019, over 29077.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.369, pruned_loss=0.1203, over 5657173.82 frames. ], libri_tot_loss[loss=0.3598, simple_loss=0.4033, pruned_loss=0.1581, over 5689427.64 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3673, pruned_loss=0.1168, over 5653084.02 frames. ], batch size: 128, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:57:27,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 19:57:33,343 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-01 19:57:46,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1947, 3.3026, 2.2650, 0.8827], device='cuda:1'), covar=tensor([0.2251, 0.0882, 0.1439, 0.2690], device='cuda:1'), in_proj_covar=tensor([0.1257, 0.1223, 0.1271, 0.1093], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 19:58:26,188 INFO [train.py:968] (1/2) Epoch 3, batch 31200, giga_loss[loss=0.3003, simple_loss=0.3688, pruned_loss=0.1159, over 28958.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3663, pruned_loss=0.1176, over 5659040.01 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.4031, pruned_loss=0.1579, over 5693373.70 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3645, pruned_loss=0.1143, over 5651837.42 frames. ], batch size: 186, lr: 9.45e-03, grad_scale: 8.0
+2023-03-01 19:58:29,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7933, 4.5688, 4.4676, 2.2039], device='cuda:1'), covar=tensor([0.0332, 0.0276, 0.0658, 0.1829], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0622, 0.0766, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 19:58:32,514 INFO [optim.py:369] (1/2) 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 19:59:29,814 INFO [train.py:968] (1/2) Epoch 3, batch 31250, giga_loss[loss=0.2584, simple_loss=0.3394, pruned_loss=0.08873, over 28443.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3653, pruned_loss=0.1163, over 5657848.65 frames. ], libri_tot_loss[loss=0.359, simple_loss=0.4025, pruned_loss=0.1578, over 5690097.05 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3634, pruned_loss=0.1127, over 5654281.51 frames. ], batch size: 336, lr: 9.45e-03, grad_scale: 4.0
+2023-03-01 19:59:50,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:00:15,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0916, 1.3339, 0.9932, 0.4126], device='cuda:1'), covar=tensor([0.1184, 0.1112, 0.1914, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.1208, 0.1258, 0.1083], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 20:00:26,418 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:00:31,272 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 31300, giga_loss[loss=0.2354, simple_loss=0.2927, pruned_loss=0.089, over 24442.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3643, pruned_loss=0.1168, over 5658590.66 frames. ], libri_tot_loss[loss=0.3585, simple_loss=0.402, pruned_loss=0.1575, over 5689459.59 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3619, pruned_loss=0.1127, over 5655217.30 frames. ], batch size: 705, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:00:38,934 INFO [optim.py:369] (1/2) 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 20:01:08,949 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=121619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:01:32,522 INFO [train.py:968] (1/2) Epoch 3, batch 31350, giga_loss[loss=0.2693, simple_loss=0.3446, pruned_loss=0.09702, over 28704.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3625, pruned_loss=0.1161, over 5652711.14 frames. ], libri_tot_loss[loss=0.3576, simple_loss=0.4012, pruned_loss=0.157, over 5683139.95 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3605, pruned_loss=0.1125, over 5655117.83 frames. ], batch size: 262, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:01:36,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2358, 2.5829, 1.1236, 1.2603], device='cuda:1'), covar=tensor([0.1091, 0.0471, 0.1086, 0.1631], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0454, 0.0317, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 20:02:27,416 INFO [train.py:968] (1/2) Epoch 3, batch 31400, libri_loss[loss=0.326, simple_loss=0.3718, pruned_loss=0.1402, over 29543.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3616, pruned_loss=0.1161, over 5655481.59 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4013, pruned_loss=0.1573, over 5679325.03 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1117, over 5659479.87 frames. ], batch size: 77, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:02:33,847 INFO [optim.py:369] (1/2) 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 20:02:42,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 20:03:24,150 INFO [train.py:968] (1/2) Epoch 3, batch 31450, giga_loss[loss=0.2927, simple_loss=0.3657, pruned_loss=0.1099, over 28371.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3632, pruned_loss=0.1161, over 5655093.99 frames. ], libri_tot_loss[loss=0.3577, simple_loss=0.4011, pruned_loss=0.1571, over 5680785.02 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3603, pruned_loss=0.1119, over 5656629.99 frames. ], batch size: 368, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:04:02,118 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 31500, giga_loss[loss=0.3142, simple_loss=0.3748, pruned_loss=0.1268, over 28704.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3638, pruned_loss=0.1154, over 5654858.07 frames. ], libri_tot_loss[loss=0.3574, simple_loss=0.4008, pruned_loss=0.1571, over 5674994.88 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3612, pruned_loss=0.1114, over 5660899.93 frames. ], batch size: 307, lr: 9.44e-03, grad_scale: 4.0
+2023-03-01 20:04:41,416 INFO [optim.py:369] (1/2) 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 20:04:44,515 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=121800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:05:04,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 20:05:37,339 INFO [train.py:968] (1/2) Epoch 3, batch 31550, giga_loss[loss=0.2503, simple_loss=0.3303, pruned_loss=0.08511, over 28939.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5660079.75 frames. ], libri_tot_loss[loss=0.3573, simple_loss=0.4006, pruned_loss=0.157, over 5679888.26 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3568, pruned_loss=0.1087, over 5660031.44 frames. ], batch size: 164, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:06:21,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2273, 1.3434, 1.0982, 1.2778], device='cuda:1'), covar=tensor([0.2130, 0.1944, 0.1932, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.1034, 0.0823, 0.0940, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 20:06:31,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3498, 1.5724, 1.1848, 1.5794], device='cuda:1'), covar=tensor([0.0750, 0.0296, 0.0358, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0149, 0.0155, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 20:06:40,320 INFO [train.py:968] (1/2) Epoch 3, batch 31600, giga_loss[loss=0.3571, simple_loss=0.4144, pruned_loss=0.1499, over 28950.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3637, pruned_loss=0.1162, over 5668768.64 frames. ], libri_tot_loss[loss=0.358, simple_loss=0.401, pruned_loss=0.1575, over 5684346.46 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3601, pruned_loss=0.1115, over 5664362.64 frames. ], batch size: 186, lr: 9.43e-03, grad_scale: 8.0
+2023-03-01 20:06:48,808 INFO [optim.py:369] (1/2) 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 20:06:55,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3638, 1.3079, 1.1759, 1.5714], device='cuda:1'), covar=tensor([0.2108, 0.1988, 0.1914, 0.2076], device='cuda:1'), in_proj_covar=tensor([0.1032, 0.0819, 0.0937, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 20:07:11,648 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=121914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:07:13,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=121917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:07:16,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4363, 2.7013, 1.8061, 2.2212], device='cuda:1'), covar=tensor([0.0664, 0.0229, 0.0325, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0149, 0.0154, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 20:07:31,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=121932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:07:41,257 INFO [train.py:968] (1/2) Epoch 3, batch 31650, giga_loss[loss=0.2712, simple_loss=0.3616, pruned_loss=0.09042, over 29049.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3667, pruned_loss=0.1168, over 5664135.60 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4008, pruned_loss=0.1575, over 5688921.06 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3633, pruned_loss=0.1121, over 5655982.80 frames. ], batch size: 128, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:07:45,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-01 20:07:47,821 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 31700, giga_loss[loss=0.2961, simple_loss=0.3736, pruned_loss=0.1094, over 28437.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3688, pruned_loss=0.1156, over 5656196.78 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.4001, pruned_loss=0.157, over 5676615.05 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3656, pruned_loss=0.1109, over 5658710.26 frames. ], batch size: 336, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:08:53,197 INFO [optim.py:369] (1/2) 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 20:09:03,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7910, 1.6476, 5.3600, 3.5837], device='cuda:1'), covar=tensor([0.1338, 0.1829, 0.0265, 0.0559], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0508, 0.0684, 0.0540], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 20:09:39,820 INFO [train.py:968] (1/2) Epoch 3, batch 31750, giga_loss[loss=0.2892, simple_loss=0.3682, pruned_loss=0.1051, over 28426.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3684, pruned_loss=0.1147, over 5657197.56 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.3996, pruned_loss=0.1568, over 5685689.24 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3649, pruned_loss=0.1094, over 5650004.83 frames. ], batch size: 368, lr: 9.43e-03, grad_scale: 4.0
+2023-03-01 20:10:10,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1592, 2.3412, 1.1472, 1.2695], device='cuda:1'), covar=tensor([0.0976, 0.0360, 0.0899, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0454, 0.0319, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0019], device='cuda:1')
+2023-03-01 20:10:17,576 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8136, 1.6759, 1.2982, 1.4230], device='cuda:1'), covar=tensor([0.0663, 0.0668, 0.1006, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0474, 0.0529, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 20:10:21,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:10:25,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:10:27,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 20:10:40,782 INFO [train.py:968] (1/2) Epoch 3, batch 31800, giga_loss[loss=0.2936, simple_loss=0.3699, pruned_loss=0.1086, over 28989.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3669, pruned_loss=0.1128, over 5651281.28 frames. ], libri_tot_loss[loss=0.3562, simple_loss=0.3994, pruned_loss=0.1565, over 5680403.53 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3637, pruned_loss=0.1078, over 5649296.75 frames. ], batch size: 213, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:10:42,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8977, 3.6559, 3.6212, 1.6824], device='cuda:1'), covar=tensor([0.0523, 0.0420, 0.0826, 0.2191], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0615, 0.0770, 0.0561], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 20:10:50,624 INFO [optim.py:369] (1/2) 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 20:11:01,348 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:11:50,621 INFO [train.py:968] (1/2) Epoch 3, batch 31850, giga_loss[loss=0.3027, simple_loss=0.3687, pruned_loss=0.1183, over 29067.00 frames. ], tot_loss[loss=0.298, simple_loss=0.368, pruned_loss=0.114, over 5651972.96 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.3993, pruned_loss=0.1564, over 5680378.97 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3651, pruned_loss=0.1096, over 5649908.71 frames. ], batch size: 285, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:12:41,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:12:56,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4356, 2.7188, 1.5070, 1.3800], device='cuda:1'), covar=tensor([0.0834, 0.0413, 0.0836, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0450, 0.0316, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 20:13:06,843 INFO [train.py:968] (1/2) Epoch 3, batch 31900, giga_loss[loss=0.2642, simple_loss=0.3412, pruned_loss=0.09357, over 28924.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3675, pruned_loss=0.1147, over 5658992.65 frames. ], libri_tot_loss[loss=0.3559, simple_loss=0.3991, pruned_loss=0.1563, over 5681668.14 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3651, pruned_loss=0.111, over 5655950.41 frames. ], batch size: 106, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:13:20,175 INFO [optim.py:369] (1/2) 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 20:13:56,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122222.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 20:14:00,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122225.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 20:14:25,596 INFO [train.py:968] (1/2) Epoch 3, batch 31950, giga_loss[loss=0.34, simple_loss=0.3836, pruned_loss=0.1482, over 27482.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3689, pruned_loss=0.1166, over 5671016.85 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.3985, pruned_loss=0.1558, over 5687776.02 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3662, pruned_loss=0.1123, over 5662308.70 frames. ], batch size: 472, lr: 9.42e-03, grad_scale: 4.0
+2023-03-01 20:14:44,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 20:15:39,000 INFO [train.py:968] (1/2) Epoch 3, batch 32000, giga_loss[loss=0.2828, simple_loss=0.3534, pruned_loss=0.1061, over 28940.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3625, pruned_loss=0.1119, over 5673047.14 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.3984, pruned_loss=0.1557, over 5688806.14 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3603, pruned_loss=0.1084, over 5665356.07 frames. ], batch size: 285, lr: 9.42e-03, grad_scale: 8.0
+2023-03-01 20:15:50,988 INFO [optim.py:369] (1/2) 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 20:16:17,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:16:22,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:16:45,416 INFO [train.py:968] (1/2) Epoch 3, batch 32050, giga_loss[loss=0.2889, simple_loss=0.3604, pruned_loss=0.1087, over 28955.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3614, pruned_loss=0.1115, over 5662435.84 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.3984, pruned_loss=0.1558, over 5680157.01 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.359, pruned_loss=0.108, over 5663692.39 frames. ], batch size: 213, lr: 9.41e-03, grad_scale: 8.0
+2023-03-01 20:16:48,304 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:16:59,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:17:51,952 INFO [train.py:968] (1/2) Epoch 3, batch 32100, giga_loss[loss=0.3197, simple_loss=0.391, pruned_loss=0.1242, over 28683.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3609, pruned_loss=0.1123, over 5661528.57 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.3978, pruned_loss=0.1554, over 5680429.94 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3589, pruned_loss=0.109, over 5661749.28 frames. ], batch size: 307, lr: 9.41e-03, grad_scale: 8.0
+2023-03-01 20:18:02,860 INFO [optim.py:369] (1/2) 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 20:18:52,461 INFO [train.py:968] (1/2) Epoch 3, batch 32150, giga_loss[loss=0.2878, simple_loss=0.3721, pruned_loss=0.1018, over 29042.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3653, pruned_loss=0.1144, over 5667417.27 frames. ], libri_tot_loss[loss=0.3541, simple_loss=0.3976, pruned_loss=0.1553, over 5682307.47 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3631, pruned_loss=0.111, over 5665890.24 frames. ], batch size: 155, lr: 9.41e-03, grad_scale: 8.0
+2023-03-01 20:19:21,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-01 20:19:21,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-01 20:19:50,743 INFO [train.py:968] (1/2) Epoch 3, batch 32200, giga_loss[loss=0.3108, simple_loss=0.3872, pruned_loss=0.1172, over 28568.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3654, pruned_loss=0.1152, over 5669519.41 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.3973, pruned_loss=0.155, over 5688227.81 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3631, pruned_loss=0.1116, over 5662609.87 frames. ], batch size: 92, lr: 9.41e-03, grad_scale: 4.0
+2023-03-01 20:20:03,674 INFO [optim.py:369] (1/2) 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 20:20:17,301 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:20:20,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2088, 1.0214, 0.9044, 1.3304], device='cuda:1'), covar=tensor([0.0884, 0.0352, 0.0436, 0.1051], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0150, 0.0155, 0.0277], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0050], device='cuda:1')
+2023-03-01 20:20:53,642 INFO [train.py:968] (1/2) Epoch 3, batch 32250, giga_loss[loss=0.2742, simple_loss=0.3486, pruned_loss=0.09988, over 28917.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3643, pruned_loss=0.1157, over 5670972.88 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.3969, pruned_loss=0.1549, over 5691931.64 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3621, pruned_loss=0.112, over 5661813.62 frames. ], batch size: 213, lr: 9.41e-03, grad_scale: 4.0
+2023-03-01 20:21:22,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4383, 2.0033, 1.4801, 1.5970], device='cuda:1'), covar=tensor([0.0800, 0.0270, 0.0345, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0149, 0.0154, 0.0276], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 20:21:48,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:21:51,528 INFO [train.py:968] (1/2) Epoch 3, batch 32300, giga_loss[loss=0.2732, simple_loss=0.3528, pruned_loss=0.0968, over 29060.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3641, pruned_loss=0.1159, over 5663473.20 frames. ], libri_tot_loss[loss=0.3523, simple_loss=0.3961, pruned_loss=0.1542, over 5687310.63 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.362, pruned_loss=0.1123, over 5660575.50 frames. ], batch size: 128, lr: 9.41e-03, grad_scale: 4.0
+2023-03-01 20:22:02,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3053, 1.3537, 1.1780, 1.3667], device='cuda:1'), covar=tensor([0.2193, 0.1909, 0.1856, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.1035, 0.0815, 0.0930, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 20:22:02,687 INFO [optim.py:369] (1/2) 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 20:22:25,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4359, 1.2626, 1.1090, 1.0799], device='cuda:1'), covar=tensor([0.0642, 0.0596, 0.0966, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0475, 0.0519, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 20:22:41,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1319, 2.9418, 2.8695, 1.2842], device='cuda:1'), covar=tensor([0.0744, 0.0530, 0.1029, 0.2137], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0618, 0.0771, 0.0566], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 20:23:01,277 INFO [train.py:968] (1/2) Epoch 3, batch 32350, giga_loss[loss=0.2764, simple_loss=0.3581, pruned_loss=0.09732, over 28664.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3668, pruned_loss=0.1171, over 5656518.32 frames. ], libri_tot_loss[loss=0.3521, simple_loss=0.3958, pruned_loss=0.1542, over 5679970.11 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3651, pruned_loss=0.1139, over 5659957.18 frames. ], batch size: 307, lr: 9.40e-03, grad_scale: 2.0
+2023-03-01 20:24:12,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4408, 2.9559, 1.4457, 1.4043], device='cuda:1'), covar=tensor([0.0802, 0.0370, 0.0878, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0452, 0.0312, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 20:24:13,608 INFO [train.py:968] (1/2) Epoch 3, batch 32400, giga_loss[loss=0.3043, simple_loss=0.3817, pruned_loss=0.1134, over 28970.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3672, pruned_loss=0.116, over 5670719.97 frames. ], libri_tot_loss[loss=0.3512, simple_loss=0.395, pruned_loss=0.1537, over 5683357.91 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.366, pruned_loss=0.113, over 5669855.67 frames. ], batch size: 128, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:24:29,380 INFO [optim.py:369] (1/2) 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 20:24:45,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:24:53,173 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 32450, giga_loss[loss=0.2491, simple_loss=0.3051, pruned_loss=0.09656, over 24344.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3654, pruned_loss=0.1152, over 5667397.76 frames. ], libri_tot_loss[loss=0.3505, simple_loss=0.3945, pruned_loss=0.1532, over 5689323.85 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.364, pruned_loss=0.112, over 5660810.41 frames. ], batch size: 705, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:26:26,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 20:26:26,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8135, 1.5875, 1.6362, 1.5170], device='cuda:1'), covar=tensor([0.0631, 0.1234, 0.1170, 0.1057], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0766, 0.0614, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 20:26:29,067 INFO [train.py:968] (1/2) Epoch 3, batch 32500, giga_loss[loss=0.2581, simple_loss=0.3293, pruned_loss=0.09348, over 29055.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.361, pruned_loss=0.1138, over 5676304.64 frames. ], libri_tot_loss[loss=0.3498, simple_loss=0.3939, pruned_loss=0.1528, over 5694943.74 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3592, pruned_loss=0.1104, over 5665097.53 frames. ], batch size: 285, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:26:38,429 INFO [optim.py:369] (1/2) 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 20:27:14,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7096, 2.1403, 1.9137, 1.9015], device='cuda:1'), covar=tensor([0.1661, 0.1649, 0.1199, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0761, 0.0732, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 20:27:32,383 INFO [train.py:968] (1/2) Epoch 3, batch 32550, giga_loss[loss=0.2586, simple_loss=0.3323, pruned_loss=0.09242, over 28444.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3542, pruned_loss=0.1106, over 5669571.42 frames. ], libri_tot_loss[loss=0.3493, simple_loss=0.3935, pruned_loss=0.1526, over 5698034.58 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3527, pruned_loss=0.1074, over 5657870.98 frames. ], batch size: 336, lr: 9.40e-03, grad_scale: 4.0
+2023-03-01 20:27:58,512 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=122861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:27:58,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=122861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:03,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=122864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:26,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:33,626 INFO [train.py:968] (1/2) Epoch 3, batch 32600, giga_loss[loss=0.2841, simple_loss=0.3352, pruned_loss=0.1165, over 24652.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3546, pruned_loss=0.1111, over 5657685.50 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.3932, pruned_loss=0.1524, over 5688489.52 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3529, pruned_loss=0.1079, over 5655395.51 frames. ], batch size: 705, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:28:36,259 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=122893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:28:45,856 INFO [optim.py:369] (1/2) 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 20:29:33,046 INFO [train.py:968] (1/2) Epoch 3, batch 32650, giga_loss[loss=0.239, simple_loss=0.3262, pruned_loss=0.07588, over 28940.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3562, pruned_loss=0.1122, over 5655666.71 frames. ], libri_tot_loss[loss=0.3487, simple_loss=0.3929, pruned_loss=0.1522, over 5691334.78 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3546, pruned_loss=0.1093, over 5650863.84 frames. ], batch size: 155, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:29:40,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1223, 3.8873, 3.8350, 1.7256], device='cuda:1'), covar=tensor([0.0449, 0.0458, 0.0835, 0.2171], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0615, 0.0755, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 20:29:59,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=122963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:30:34,827 INFO [train.py:968] (1/2) Epoch 3, batch 32700, giga_loss[loss=0.2815, simple_loss=0.3571, pruned_loss=0.103, over 28674.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3544, pruned_loss=0.1104, over 5658757.89 frames. ], libri_tot_loss[loss=0.3489, simple_loss=0.393, pruned_loss=0.1524, over 5695622.05 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3523, pruned_loss=0.1072, over 5650528.88 frames. ], batch size: 243, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:30:47,378 INFO [optim.py:369] (1/2) 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 20:31:21,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:31:24,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:31:36,846 INFO [train.py:968] (1/2) Epoch 3, batch 32750, libri_loss[loss=0.3187, simple_loss=0.3604, pruned_loss=0.1385, over 29606.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3542, pruned_loss=0.11, over 5671302.09 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.393, pruned_loss=0.1524, over 5702256.95 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3514, pruned_loss=0.1062, over 5657656.51 frames. ], batch size: 74, lr: 9.39e-03, grad_scale: 4.0
+2023-03-01 20:32:01,732 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:32:36,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:32:41,670 INFO [train.py:968] (1/2) Epoch 3, batch 32800, giga_loss[loss=0.2674, simple_loss=0.3408, pruned_loss=0.09698, over 29033.00 frames. ], tot_loss[loss=0.288, simple_loss=0.354, pruned_loss=0.111, over 5661140.07 frames. ], libri_tot_loss[loss=0.3488, simple_loss=0.3929, pruned_loss=0.1524, over 5700720.87 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3506, pruned_loss=0.1066, over 5650460.21 frames. ], batch size: 165, lr: 9.39e-03, grad_scale: 8.0
+2023-03-01 20:32:52,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6429, 2.4059, 1.5322, 0.6481], device='cuda:1'), covar=tensor([0.2752, 0.1569, 0.1865, 0.3018], device='cuda:1'), in_proj_covar=tensor([0.1268, 0.1234, 0.1294, 0.1102], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 20:32:55,301 INFO [optim.py:369] (1/2) 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 20:33:01,445 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:33:07,106 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:33:44,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:33:50,251 INFO [train.py:968] (1/2) Epoch 3, batch 32850, giga_loss[loss=0.2524, simple_loss=0.3112, pruned_loss=0.09685, over 24609.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3525, pruned_loss=0.1087, over 5661284.81 frames. ], libri_tot_loss[loss=0.3487, simple_loss=0.3927, pruned_loss=0.1524, over 5703481.93 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3493, pruned_loss=0.1046, over 5649816.13 frames. ], batch size: 705, lr: 9.38e-03, grad_scale: 8.0
+2023-03-01 20:34:53,174 INFO [train.py:968] (1/2) Epoch 3, batch 32900, giga_loss[loss=0.3249, simple_loss=0.3851, pruned_loss=0.1324, over 28801.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3522, pruned_loss=0.1085, over 5668765.03 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.3923, pruned_loss=0.1519, over 5708399.11 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3492, pruned_loss=0.1047, over 5654358.90 frames. ], batch size: 243, lr: 9.38e-03, grad_scale: 8.0
+2023-03-01 20:35:06,750 INFO [optim.py:369] (1/2) 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 20:35:45,187 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:35:47,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:35:53,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:35:59,434 INFO [train.py:968] (1/2) Epoch 3, batch 32950, giga_loss[loss=0.2453, simple_loss=0.3225, pruned_loss=0.08408, over 28929.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3532, pruned_loss=0.1099, over 5672339.33 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3916, pruned_loss=0.1515, over 5712315.27 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3506, pruned_loss=0.1064, over 5656636.14 frames. ], batch size: 155, lr: 9.38e-03, grad_scale: 4.0
+2023-03-01 20:36:27,465 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:36:29,629 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:37:02,898 INFO [train.py:968] (1/2) Epoch 3, batch 33000, giga_loss[loss=0.2454, simple_loss=0.3284, pruned_loss=0.08122, over 28422.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3505, pruned_loss=0.1066, over 5666132.78 frames. ], libri_tot_loss[loss=0.3471, simple_loss=0.3914, pruned_loss=0.1514, over 5711833.69 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3485, pruned_loss=0.1038, over 5654077.85 frames. ], batch size: 369, lr: 9.38e-03, grad_scale: 4.0
+2023-03-01 20:37:02,898 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 20:37:10,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3958, 1.5898, 1.5700, 1.4869], device='cuda:1'), covar=tensor([0.0839, 0.1100, 0.1315, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0753, 0.0605, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 20:37:11,508 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 20:37:21,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4132, 1.5454, 1.5401, 1.4935], device='cuda:1'), covar=tensor([0.0933, 0.1502, 0.1287, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0754, 0.0605, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 20:37:24,084 INFO [optim.py:369] (1/2) 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 20:38:10,627 INFO [train.py:968] (1/2) Epoch 3, batch 33050, giga_loss[loss=0.2878, simple_loss=0.3588, pruned_loss=0.1084, over 28617.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3543, pruned_loss=0.1073, over 5666706.01 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3917, pruned_loss=0.1516, over 5714122.80 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3522, pruned_loss=0.1045, over 5654754.25 frames. ], batch size: 78, lr: 9.38e-03, grad_scale: 4.0
+2023-03-01 20:38:54,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:38:57,160 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 33100, giga_loss[loss=0.2839, simple_loss=0.3628, pruned_loss=0.1024, over 28866.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3574, pruned_loss=0.1097, over 5664934.69 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3912, pruned_loss=0.1512, over 5719700.80 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.355, pruned_loss=0.1064, over 5648425.05 frames. ], batch size: 145, lr: 9.37e-03, grad_scale: 4.0
+2023-03-01 20:39:21,993 INFO [optim.py:369] (1/2) 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 20:39:36,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:40:09,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6219, 1.6397, 1.2335, 1.4538], device='cuda:1'), covar=tensor([0.0714, 0.0615, 0.0995, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0476, 0.0524, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 20:40:14,984 INFO [train.py:968] (1/2) Epoch 3, batch 33150, giga_loss[loss=0.2577, simple_loss=0.3415, pruned_loss=0.0869, over 28939.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3576, pruned_loss=0.1095, over 5663644.39 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.3911, pruned_loss=0.1511, over 5719849.53 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3555, pruned_loss=0.1066, over 5649944.01 frames. ], batch size: 164, lr: 9.37e-03, grad_scale: 4.0
+2023-03-01 20:41:12,614 INFO [train.py:968] (1/2) Epoch 3, batch 33200, giga_loss[loss=0.2869, simple_loss=0.3547, pruned_loss=0.1096, over 28524.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3578, pruned_loss=0.11, over 5669429.09 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3907, pruned_loss=0.1508, over 5724288.22 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3555, pruned_loss=0.1068, over 5652864.72 frames. ], batch size: 78, lr: 9.37e-03, grad_scale: 8.0
+2023-03-01 20:41:17,078 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-01 20:41:27,731 INFO [optim.py:369] (1/2) 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 20:41:57,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5335, 2.1862, 1.5018, 0.6965], device='cuda:1'), covar=tensor([0.2088, 0.1118, 0.1932, 0.2449], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.1246, 0.1280, 0.1102], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 20:42:14,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6181, 2.4441, 1.5515, 0.6132], device='cuda:1'), covar=tensor([0.2962, 0.1494, 0.1932, 0.2841], device='cuda:1'), in_proj_covar=tensor([0.1264, 0.1241, 0.1275, 0.1100], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 20:42:20,106 INFO [train.py:968] (1/2) Epoch 3, batch 33250, giga_loss[loss=0.2665, simple_loss=0.3227, pruned_loss=0.1051, over 24727.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3551, pruned_loss=0.1076, over 5666571.11 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3907, pruned_loss=0.1508, over 5724288.22 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3533, pruned_loss=0.105, over 5653678.85 frames. ], batch size: 705, lr: 9.37e-03, grad_scale: 8.0
+2023-03-01 20:43:22,500 INFO [train.py:968] (1/2) Epoch 3, batch 33300, giga_loss[loss=0.2735, simple_loss=0.3419, pruned_loss=0.1025, over 28921.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.354, pruned_loss=0.1079, over 5666582.16 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3907, pruned_loss=0.1509, over 5725216.78 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3524, pruned_loss=0.1056, over 5655351.13 frames. ], batch size: 213, lr: 9.37e-03, grad_scale: 8.0
+2023-03-01 20:43:36,874 INFO [optim.py:369] (1/2) 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 20:44:24,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=123640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:44:24,624 INFO [train.py:968] (1/2) Epoch 3, batch 33350, giga_loss[loss=0.257, simple_loss=0.3275, pruned_loss=0.09325, over 28650.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3545, pruned_loss=0.1081, over 5671778.40 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.3908, pruned_loss=0.151, over 5726083.11 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3528, pruned_loss=0.1059, over 5661772.89 frames. ], batch size: 85, lr: 9.37e-03, grad_scale: 4.0
+2023-03-01 20:45:27,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5110, 4.1869, 4.1872, 1.6282], device='cuda:1'), covar=tensor([0.0414, 0.0418, 0.0892, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0610, 0.0744, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 20:45:32,780 INFO [train.py:968] (1/2) Epoch 3, batch 33400, giga_loss[loss=0.3184, simple_loss=0.3796, pruned_loss=0.1286, over 27655.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3582, pruned_loss=0.1105, over 5670220.76 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3907, pruned_loss=0.1509, over 5727731.52 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3565, pruned_loss=0.1083, over 5659882.51 frames. ], batch size: 472, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:45:47,200 INFO [optim.py:369] (1/2) 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 20:45:51,803 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:46:05,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-01 20:46:32,153 INFO [train.py:968] (1/2) Epoch 3, batch 33450, giga_loss[loss=0.3555, simple_loss=0.4007, pruned_loss=0.1552, over 27533.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3586, pruned_loss=0.1113, over 5655806.54 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3907, pruned_loss=0.1509, over 5718692.67 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3564, pruned_loss=0.1085, over 5653750.32 frames. ], batch size: 472, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:47:31,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=123783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:47:36,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=123786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:47:41,675 INFO [train.py:968] (1/2) Epoch 3, batch 33500, giga_loss[loss=0.3022, simple_loss=0.3516, pruned_loss=0.1264, over 24249.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3627, pruned_loss=0.1139, over 5658969.86 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3905, pruned_loss=0.1507, over 5711863.18 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3606, pruned_loss=0.1112, over 5662952.34 frames. ], batch size: 705, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:47:52,378 INFO [optim.py:369] (1/2) 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 20:48:07,817 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=123815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:48:09,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3447, 1.8898, 1.4353, 1.5564], device='cuda:1'), covar=tensor([0.0939, 0.0333, 0.0379, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0148, 0.0153, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0045, 0.0032, 0.0029, 0.0050], device='cuda:1')
+2023-03-01 20:48:12,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=123821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:48:33,028 INFO [train.py:968] (1/2) Epoch 3, batch 33550, giga_loss[loss=0.2908, simple_loss=0.3649, pruned_loss=0.1084, over 27710.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3649, pruned_loss=0.1146, over 5662142.33 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.3902, pruned_loss=0.1505, over 5715786.13 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3625, pruned_loss=0.1113, over 5660095.37 frames. ], batch size: 474, lr: 9.36e-03, grad_scale: 4.0
+2023-03-01 20:49:37,557 INFO [train.py:968] (1/2) Epoch 3, batch 33600, giga_loss[loss=0.2432, simple_loss=0.3016, pruned_loss=0.09243, over 24221.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3639, pruned_loss=0.1134, over 5663692.60 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3893, pruned_loss=0.1499, over 5718629.83 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3622, pruned_loss=0.1103, over 5657936.44 frames. ], batch size: 705, lr: 9.36e-03, grad_scale: 8.0
+2023-03-01 20:49:56,047 INFO [optim.py:369] (1/2) 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 20:50:01,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9707, 2.4754, 2.5634, 2.1701], device='cuda:1'), covar=tensor([0.0587, 0.1372, 0.0924, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0761, 0.0613, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 20:50:07,206 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-01 20:50:31,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2225, 1.4022, 1.1710, 1.5493], device='cuda:1'), covar=tensor([0.2194, 0.1915, 0.1928, 0.2097], device='cuda:1'), in_proj_covar=tensor([0.1022, 0.0805, 0.0929, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 20:50:44,192 INFO [train.py:968] (1/2) Epoch 3, batch 33650, giga_loss[loss=0.2701, simple_loss=0.3415, pruned_loss=0.09931, over 28894.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.364, pruned_loss=0.1145, over 5665461.24 frames. ], libri_tot_loss[loss=0.3444, simple_loss=0.3892, pruned_loss=0.1498, over 5721681.08 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3619, pruned_loss=0.1108, over 5656184.44 frames. ], batch size: 186, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:50:53,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3722, 1.4525, 1.0375, 1.1338], device='cuda:1'), covar=tensor([0.0736, 0.0560, 0.0482, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.1227, 0.0938, 0.0946, 0.1035], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 20:51:46,134 INFO [train.py:968] (1/2) Epoch 3, batch 33700, giga_loss[loss=0.3526, simple_loss=0.3963, pruned_loss=0.1544, over 26834.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3643, pruned_loss=0.1163, over 5650669.78 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3896, pruned_loss=0.1504, over 5709411.93 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3611, pruned_loss=0.1115, over 5653405.11 frames. ], batch size: 555, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:52:04,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9946, 1.1356, 1.0059, 0.6490], device='cuda:1'), covar=tensor([0.0566, 0.0587, 0.0411, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.1232, 0.0935, 0.0946, 0.1041], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 20:52:04,655 INFO [optim.py:369] (1/2) 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 20:52:46,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9738, 1.5994, 0.9560, 0.9554], device='cuda:1'), covar=tensor([0.0577, 0.0421, 0.0562, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0452, 0.0310, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 20:52:47,672 INFO [train.py:968] (1/2) Epoch 3, batch 33750, giga_loss[loss=0.2966, simple_loss=0.3628, pruned_loss=0.1152, over 28039.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5644713.19 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3892, pruned_loss=0.1501, over 5704781.50 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3605, pruned_loss=0.1109, over 5649984.42 frames. ], batch size: 412, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:53:44,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:53:53,603 INFO [train.py:968] (1/2) Epoch 3, batch 33800, giga_loss[loss=0.2783, simple_loss=0.3511, pruned_loss=0.1028, over 28587.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3632, pruned_loss=0.1161, over 5652005.02 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3891, pruned_loss=0.1501, over 5708873.58 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3605, pruned_loss=0.1116, over 5651399.50 frames. ], batch size: 307, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:54:09,990 INFO [optim.py:369] (1/2) 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 20:54:26,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7491, 1.0920, 3.3910, 2.8042], device='cuda:1'), covar=tensor([0.1494, 0.1879, 0.0379, 0.0623], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0505, 0.0680, 0.0529], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 20:54:54,034 INFO [train.py:968] (1/2) Epoch 3, batch 33850, giga_loss[loss=0.2401, simple_loss=0.3096, pruned_loss=0.08533, over 28571.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3612, pruned_loss=0.116, over 5643860.67 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.389, pruned_loss=0.1501, over 5704111.92 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3583, pruned_loss=0.1114, over 5645020.59 frames. ], batch size: 85, lr: 9.35e-03, grad_scale: 4.0
+2023-03-01 20:55:54,510 INFO [train.py:968] (1/2) Epoch 3, batch 33900, giga_loss[loss=0.2474, simple_loss=0.3228, pruned_loss=0.086, over 28543.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3606, pruned_loss=0.1148, over 5640288.00 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.3887, pruned_loss=0.1499, over 5701958.47 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3579, pruned_loss=0.1105, over 5641869.36 frames. ], batch size: 85, lr: 9.34e-03, grad_scale: 4.0
+2023-03-01 20:55:56,445 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-01 20:55:59,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:56:06,288 INFO [optim.py:369] (1/2) 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 20:56:32,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:56:36,405 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:56:44,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5020, 1.9244, 1.2099, 1.1152], device='cuda:1'), covar=tensor([0.0953, 0.0685, 0.0645, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.1233, 0.0932, 0.0952, 0.1052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 20:56:51,502 INFO [train.py:968] (1/2) Epoch 3, batch 33950, libri_loss[loss=0.3223, simple_loss=0.3784, pruned_loss=0.1331, over 29488.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.359, pruned_loss=0.1129, over 5667938.99 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3875, pruned_loss=0.1489, over 5711961.05 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3566, pruned_loss=0.1087, over 5657877.74 frames. ], batch size: 85, lr: 9.34e-03, grad_scale: 4.0
+2023-03-01 20:57:10,391 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:57:19,320 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:57:36,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.61 vs. limit=5.0
+2023-03-01 20:57:45,904 INFO [train.py:968] (1/2) Epoch 3, batch 34000, giga_loss[loss=0.2949, simple_loss=0.3754, pruned_loss=0.1072, over 28941.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3607, pruned_loss=0.1114, over 5665830.53 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3876, pruned_loss=0.1491, over 5701154.67 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3579, pruned_loss=0.1068, over 5667053.04 frames. ], batch size: 213, lr: 9.34e-03, grad_scale: 8.0
+2023-03-01 20:57:59,313 INFO [optim.py:369] (1/2) 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 20:58:39,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:58:41,626 INFO [train.py:968] (1/2) Epoch 3, batch 34050, giga_loss[loss=0.2369, simple_loss=0.3253, pruned_loss=0.07419, over 28457.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3625, pruned_loss=0.1116, over 5667134.68 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3873, pruned_loss=0.1488, over 5707718.88 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3597, pruned_loss=0.1069, over 5661158.90 frames. ], batch size: 71, lr: 9.34e-03, grad_scale: 8.0
+2023-03-01 20:58:42,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 20:59:16,002 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 34100, giga_loss[loss=0.2551, simple_loss=0.3448, pruned_loss=0.08275, over 28878.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3618, pruned_loss=0.1105, over 5669381.79 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3868, pruned_loss=0.1485, over 5710877.92 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3597, pruned_loss=0.1065, over 5661376.27 frames. ], batch size: 174, lr: 9.34e-03, grad_scale: 8.0
+2023-03-01 21:00:02,702 INFO [optim.py:369] (1/2) 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 21:00:16,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 21:00:37,273 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:00:53,155 INFO [train.py:968] (1/2) Epoch 3, batch 34150, giga_loss[loss=0.3193, simple_loss=0.3733, pruned_loss=0.1327, over 26733.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3617, pruned_loss=0.1102, over 5665013.60 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3867, pruned_loss=0.1484, over 5705188.34 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3597, pruned_loss=0.1065, over 5663183.16 frames. ], batch size: 555, lr: 9.34e-03, grad_scale: 2.0
+2023-03-01 21:01:09,606 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:01:09,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0592, 1.2122, 1.1965, 1.1434], device='cuda:1'), covar=tensor([0.0796, 0.0976, 0.1394, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0758, 0.0610, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 21:02:03,311 INFO [train.py:968] (1/2) Epoch 3, batch 34200, giga_loss[loss=0.3186, simple_loss=0.392, pruned_loss=0.1225, over 29009.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3624, pruned_loss=0.1106, over 5659487.19 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3868, pruned_loss=0.1486, over 5699418.25 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3604, pruned_loss=0.1069, over 5661870.39 frames. ], batch size: 213, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:02:22,340 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 34250, giga_loss[loss=0.2594, simple_loss=0.345, pruned_loss=0.08694, over 29011.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3625, pruned_loss=0.1113, over 5657749.93 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3863, pruned_loss=0.1484, over 5700342.38 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3601, pruned_loss=0.1066, over 5656983.24 frames. ], batch size: 186, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:04:14,296 INFO [train.py:968] (1/2) Epoch 3, batch 34300, giga_loss[loss=0.253, simple_loss=0.3445, pruned_loss=0.08079, over 29010.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3639, pruned_loss=0.1121, over 5644960.20 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3864, pruned_loss=0.1486, over 5694068.56 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3614, pruned_loss=0.1073, over 5649322.93 frames. ], batch size: 175, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:04:15,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4313, 1.5411, 1.2371, 0.8436], device='cuda:1'), covar=tensor([0.0705, 0.0514, 0.0414, 0.0653], device='cuda:1'), in_proj_covar=tensor([0.1223, 0.0929, 0.0951, 0.1038], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 21:04:32,706 INFO [optim.py:369] (1/2) 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 21:05:17,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:05:17,539 INFO [train.py:968] (1/2) Epoch 3, batch 34350, giga_loss[loss=0.2843, simple_loss=0.3636, pruned_loss=0.1025, over 28895.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3662, pruned_loss=0.1122, over 5655947.41 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3861, pruned_loss=0.1484, over 5689283.67 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3639, pruned_loss=0.1078, over 5662210.65 frames. ], batch size: 227, lr: 9.33e-03, grad_scale: 2.0
+2023-03-01 21:05:52,798 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:06:29,839 INFO [train.py:968] (1/2) Epoch 3, batch 34400, giga_loss[loss=0.2562, simple_loss=0.3345, pruned_loss=0.08896, over 29196.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3656, pruned_loss=0.1119, over 5669490.78 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3864, pruned_loss=0.1487, over 5691587.87 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3634, pruned_loss=0.1078, over 5672066.79 frames. ], batch size: 200, lr: 9.33e-03, grad_scale: 4.0
+2023-03-01 21:06:55,640 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 34450, libri_loss[loss=0.3597, simple_loss=0.3951, pruned_loss=0.1621, over 29565.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3627, pruned_loss=0.1103, over 5678477.83 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3864, pruned_loss=0.1485, over 5695927.63 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3605, pruned_loss=0.1064, over 5676174.72 frames. ], batch size: 76, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:07:42,393 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124741.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 21:08:37,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:08:40,309 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 34500, giga_loss[loss=0.2931, simple_loss=0.3651, pruned_loss=0.1105, over 28569.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3609, pruned_loss=0.1089, over 5686968.55 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3862, pruned_loss=0.1484, over 5699394.50 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3586, pruned_loss=0.1048, over 5681431.76 frames. ], batch size: 370, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:09:06,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:07,654 INFO [optim.py:369] (1/2) 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 21:09:16,125 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=124815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:29,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=124825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:36,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=124830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:09:53,658 INFO [train.py:968] (1/2) Epoch 3, batch 34550, giga_loss[loss=0.2587, simple_loss=0.344, pruned_loss=0.08668, over 28375.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3592, pruned_loss=0.1076, over 5693014.44 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3862, pruned_loss=0.1485, over 5699471.59 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3569, pruned_loss=0.1035, over 5688163.98 frames. ], batch size: 368, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:10:49,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2563, 1.6604, 1.4933, 1.4750], device='cuda:1'), covar=tensor([0.1352, 0.1776, 0.1105, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0747, 0.0730, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 21:10:52,519 INFO [train.py:968] (1/2) Epoch 3, batch 34600, giga_loss[loss=0.2807, simple_loss=0.3565, pruned_loss=0.1024, over 28878.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.361, pruned_loss=0.1091, over 5683016.90 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3864, pruned_loss=0.1487, over 5693696.84 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3585, pruned_loss=0.1049, over 5683538.56 frames. ], batch size: 112, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:11:05,988 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-01 21:11:10,948 INFO [optim.py:369] (1/2) 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 21:11:45,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9800, 1.2379, 3.5802, 2.9504], device='cuda:1'), covar=tensor([0.1444, 0.1900, 0.0375, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0525, 0.0494, 0.0671, 0.0525], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 21:11:55,760 INFO [train.py:968] (1/2) Epoch 3, batch 34650, giga_loss[loss=0.2883, simple_loss=0.3614, pruned_loss=0.1076, over 28247.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3638, pruned_loss=0.1109, over 5669930.51 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3864, pruned_loss=0.1485, over 5695529.48 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3616, pruned_loss=0.1074, over 5668707.44 frames. ], batch size: 412, lr: 9.32e-03, grad_scale: 4.0
+2023-03-01 21:12:01,197 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:04,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:31,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=124973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:34,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=124976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:12:37,234 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 34700, giga_loss[loss=0.2558, simple_loss=0.3315, pruned_loss=0.09004, over 28965.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3622, pruned_loss=0.1108, over 5672627.69 frames. ], libri_tot_loss[loss=0.3421, simple_loss=0.3867, pruned_loss=0.1487, over 5699934.35 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3596, pruned_loss=0.1067, over 5667183.83 frames. ], batch size: 186, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:13:10,266 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:13:10,576 INFO [optim.py:369] (1/2) 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 21:13:15,776 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6341, 1.6582, 1.1623, 1.4900], device='cuda:1'), covar=tensor([0.0714, 0.0612, 0.1084, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0460, 0.0515, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 21:13:37,257 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:13:44,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2009, 1.3815, 1.2142, 1.2440], device='cuda:1'), covar=tensor([0.1797, 0.1464, 0.1400, 0.1502], device='cuda:1'), in_proj_covar=tensor([0.1020, 0.0798, 0.0918, 0.0920], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 21:13:48,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:13:48,385 INFO [train.py:968] (1/2) Epoch 3, batch 34750, giga_loss[loss=0.3015, simple_loss=0.3741, pruned_loss=0.1145, over 28607.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3597, pruned_loss=0.1101, over 5672761.45 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3863, pruned_loss=0.1484, over 5700596.64 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3574, pruned_loss=0.1065, over 5667403.78 frames. ], batch size: 307, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:14:08,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:14:18,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5009, 2.0672, 1.6577, 1.6441], device='cuda:1'), covar=tensor([0.0870, 0.0285, 0.0356, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0148, 0.0153, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0045, 0.0032, 0.0029, 0.0050], device='cuda:1')
+2023-03-01 21:14:35,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8514, 1.1234, 3.9517, 3.0720], device='cuda:1'), covar=tensor([0.1669, 0.2161, 0.0366, 0.0653], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0504, 0.0686, 0.0538], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 21:14:42,487 INFO [train.py:968] (1/2) Epoch 3, batch 34800, giga_loss[loss=0.352, simple_loss=0.4024, pruned_loss=0.1508, over 28883.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3613, pruned_loss=0.1124, over 5666976.18 frames. ], libri_tot_loss[loss=0.3412, simple_loss=0.3861, pruned_loss=0.1482, over 5697421.26 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3589, pruned_loss=0.1085, over 5664478.39 frames. ], batch size: 112, lr: 9.31e-03, grad_scale: 8.0
+2023-03-01 21:15:00,682 INFO [optim.py:369] (1/2) 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 21:15:10,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125116.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 21:15:33,407 INFO [train.py:968] (1/2) Epoch 3, batch 34850, giga_loss[loss=0.3198, simple_loss=0.3934, pruned_loss=0.1231, over 29041.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3705, pruned_loss=0.1188, over 5665292.05 frames. ], libri_tot_loss[loss=0.3411, simple_loss=0.3859, pruned_loss=0.1481, over 5698451.16 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3687, pruned_loss=0.1157, over 5662262.98 frames. ], batch size: 128, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:16:08,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.82 vs. limit=5.0
+2023-03-01 21:16:11,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:14,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:20,366 INFO [train.py:968] (1/2) Epoch 3, batch 34900, giga_loss[loss=0.3682, simple_loss=0.4227, pruned_loss=0.1569, over 29057.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3801, pruned_loss=0.125, over 5660634.81 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3864, pruned_loss=0.1485, over 5682527.15 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.378, pruned_loss=0.1215, over 5672400.40 frames. ], batch size: 136, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:16:28,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:34,447 INFO [optim.py:369] (1/2) 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 21:16:41,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:16:43,876 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:17:04,462 INFO [train.py:968] (1/2) Epoch 3, batch 34950, giga_loss[loss=0.2823, simple_loss=0.3526, pruned_loss=0.106, over 29050.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3824, pruned_loss=0.1271, over 5667514.68 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.3862, pruned_loss=0.1485, over 5684764.10 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3808, pruned_loss=0.1242, over 5674717.19 frames. ], batch size: 136, lr: 9.31e-03, grad_scale: 4.0
+2023-03-01 21:17:21,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 21:17:24,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125262.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 21:17:48,516 INFO [train.py:968] (1/2) Epoch 3, batch 35000, giga_loss[loss=0.2797, simple_loss=0.3398, pruned_loss=0.1098, over 28865.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3777, pruned_loss=0.1256, over 5674870.18 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.387, pruned_loss=0.149, over 5690576.94 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3756, pruned_loss=0.1222, over 5674842.27 frames. ], batch size: 99, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:17:48,816 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 21:18:01,039 INFO [optim.py:369] (1/2) 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 21:18:28,448 INFO [train.py:968] (1/2) Epoch 3, batch 35050, libri_loss[loss=0.429, simple_loss=0.4426, pruned_loss=0.2077, over 19302.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3727, pruned_loss=0.1243, over 5659365.36 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.3878, pruned_loss=0.1493, over 5677028.83 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3696, pruned_loss=0.1201, over 5672224.78 frames. ], batch size: 187, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:18:30,249 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:18:33,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:18:56,555 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:19:09,006 INFO [train.py:968] (1/2) Epoch 3, batch 35100, giga_loss[loss=0.28, simple_loss=0.344, pruned_loss=0.108, over 28866.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3644, pruned_loss=0.12, over 5663627.71 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3882, pruned_loss=0.1495, over 5668383.99 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3612, pruned_loss=0.116, over 5682014.01 frames. ], batch size: 186, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:19:21,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:19:21,712 INFO [optim.py:369] (1/2) 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 21:19:42,428 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 35150, giga_loss[loss=0.2567, simple_loss=0.3184, pruned_loss=0.09748, over 28198.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3567, pruned_loss=0.1166, over 5661697.60 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3883, pruned_loss=0.1495, over 5666381.18 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3534, pruned_loss=0.1127, over 5678163.52 frames. ], batch size: 368, lr: 9.30e-03, grad_scale: 4.0
+2023-03-01 21:20:35,745 INFO [train.py:968] (1/2) Epoch 3, batch 35200, giga_loss[loss=0.2196, simple_loss=0.2881, pruned_loss=0.0756, over 28626.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.35, pruned_loss=0.1134, over 5650620.13 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3885, pruned_loss=0.1497, over 5650968.36 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3466, pruned_loss=0.1096, over 5678860.61 frames. ], batch size: 85, lr: 9.30e-03, grad_scale: 8.0
+2023-03-01 21:20:48,547 INFO [optim.py:369] (1/2) 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 21:21:02,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0016, 1.8362, 1.8045, 1.7358], device='cuda:1'), covar=tensor([0.0992, 0.1605, 0.1244, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0769, 0.0616, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 21:21:02,731 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 35250, giga_loss[loss=0.2692, simple_loss=0.3401, pruned_loss=0.09914, over 28831.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3458, pruned_loss=0.1111, over 5662933.93 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.3888, pruned_loss=0.1498, over 5646987.31 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3423, pruned_loss=0.1075, over 5689399.53 frames. ], batch size: 174, lr: 9.29e-03, grad_scale: 8.0
+2023-03-01 21:21:25,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:28,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:46,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:48,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:52,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:58,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:21:59,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0071, 1.8551, 1.7492, 1.7850], device='cuda:1'), covar=tensor([0.0928, 0.1686, 0.1462, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0764, 0.0617, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 21:22:00,426 INFO [train.py:968] (1/2) Epoch 3, batch 35300, giga_loss[loss=0.2421, simple_loss=0.3115, pruned_loss=0.08638, over 28554.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3444, pruned_loss=0.1112, over 5663019.20 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3903, pruned_loss=0.1507, over 5645105.13 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3386, pruned_loss=0.106, over 5686890.40 frames. ], batch size: 336, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:22:03,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0849, 1.3259, 1.0457, 0.4378], device='cuda:1'), covar=tensor([0.1219, 0.1138, 0.1828, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.1280, 0.1207, 0.1272, 0.1092], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 21:22:08,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-01 21:22:12,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:22:13,946 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 35350, giga_loss[loss=0.2645, simple_loss=0.3251, pruned_loss=0.1019, over 28866.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3401, pruned_loss=0.1087, over 5665212.58 frames. ], libri_tot_loss[loss=0.3456, simple_loss=0.39, pruned_loss=0.1506, over 5646704.33 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3348, pruned_loss=0.104, over 5683135.21 frames. ], batch size: 186, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:23:26,966 INFO [train.py:968] (1/2) Epoch 3, batch 35400, libri_loss[loss=0.3153, simple_loss=0.3614, pruned_loss=0.1346, over 28503.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3372, pruned_loss=0.1073, over 5663856.37 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3905, pruned_loss=0.1507, over 5650998.77 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3311, pruned_loss=0.1022, over 5674492.77 frames. ], batch size: 63, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:23:41,430 INFO [optim.py:369] (1/2) 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 21:24:00,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=125731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:03,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=125734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:07,831 INFO [train.py:968] (1/2) Epoch 3, batch 35450, giga_loss[loss=0.2825, simple_loss=0.3315, pruned_loss=0.1168, over 28533.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3357, pruned_loss=0.1066, over 5679500.78 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.3912, pruned_loss=0.1511, over 5660776.02 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3279, pruned_loss=0.1003, over 5679625.24 frames. ], batch size: 71, lr: 9.29e-03, grad_scale: 4.0
+2023-03-01 21:24:12,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 21:24:13,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:26,474 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=125763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:30,950 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:24:45,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-01 21:24:47,866 INFO [train.py:968] (1/2) Epoch 3, batch 35500, giga_loss[loss=0.2867, simple_loss=0.3465, pruned_loss=0.1134, over 28279.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3343, pruned_loss=0.1057, over 5686822.21 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.391, pruned_loss=0.1506, over 5665464.09 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3258, pruned_loss=0.09911, over 5683129.93 frames. ], batch size: 368, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:25:01,921 INFO [optim.py:369] (1/2) 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 21:25:06,826 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:25:30,488 INFO [train.py:968] (1/2) Epoch 3, batch 35550, giga_loss[loss=0.2069, simple_loss=0.275, pruned_loss=0.06942, over 28335.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3285, pruned_loss=0.102, over 5691083.11 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3909, pruned_loss=0.1505, over 5667960.26 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3213, pruned_loss=0.09651, over 5686116.39 frames. ], batch size: 78, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:25:37,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1656, 1.3132, 1.0056, 0.7130], device='cuda:1'), covar=tensor([0.0731, 0.0554, 0.0475, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.1206, 0.0903, 0.0935, 0.1034], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 21:25:59,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 21:26:16,384 INFO [train.py:968] (1/2) Epoch 3, batch 35600, giga_loss[loss=0.2314, simple_loss=0.2988, pruned_loss=0.08203, over 28996.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3241, pruned_loss=0.09975, over 5679677.11 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3908, pruned_loss=0.1505, over 5668973.46 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3183, pruned_loss=0.09527, over 5675089.91 frames. ], batch size: 227, lr: 9.28e-03, grad_scale: 8.0
+2023-03-01 21:26:22,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=125896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:26:32,714 INFO [optim.py:369] (1/2) 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 21:26:45,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-01 21:27:00,736 INFO [train.py:968] (1/2) Epoch 3, batch 35650, giga_loss[loss=0.272, simple_loss=0.3356, pruned_loss=0.1041, over 27902.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3262, pruned_loss=0.1022, over 5675711.83 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.3915, pruned_loss=0.1509, over 5669778.31 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3197, pruned_loss=0.09726, over 5671725.11 frames. ], batch size: 412, lr: 9.28e-03, grad_scale: 8.0
+2023-03-01 21:27:44,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-01 21:27:44,573 INFO [train.py:968] (1/2) Epoch 3, batch 35700, libri_loss[loss=0.3897, simple_loss=0.4273, pruned_loss=0.1761, over 25787.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3409, pruned_loss=0.1103, over 5687095.48 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.3927, pruned_loss=0.1516, over 5672770.69 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3324, pruned_loss=0.1041, over 5681715.01 frames. ], batch size: 136, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:27:49,465 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=125996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:28:02,326 INFO [optim.py:369] (1/2) 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 21:28:18,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7013, 2.9865, 1.5337, 1.3421], device='cuda:1'), covar=tensor([0.0847, 0.0354, 0.0874, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0451, 0.0307, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 21:28:30,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:28:31,684 INFO [train.py:968] (1/2) Epoch 3, batch 35750, libri_loss[loss=0.4312, simple_loss=0.4626, pruned_loss=0.1999, over 25987.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3567, pruned_loss=0.1203, over 5684012.20 frames. ], libri_tot_loss[loss=0.3487, simple_loss=0.3933, pruned_loss=0.1521, over 5675932.57 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3481, pruned_loss=0.1138, over 5677475.63 frames. ], batch size: 136, lr: 9.28e-03, grad_scale: 4.0
+2023-03-01 21:28:32,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:28:56,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:29:11,504 INFO [train.py:968] (1/2) Epoch 3, batch 35800, giga_loss[loss=0.316, simple_loss=0.3904, pruned_loss=0.1209, over 28915.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3675, pruned_loss=0.1259, over 5687817.17 frames. ], libri_tot_loss[loss=0.3491, simple_loss=0.3938, pruned_loss=0.1522, over 5680534.57 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3593, pruned_loss=0.1198, over 5678614.76 frames. ], batch size: 285, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:29:28,136 INFO [optim.py:369] (1/2) 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 21:29:35,613 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:29:40,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:29:54,645 INFO [train.py:968] (1/2) Epoch 3, batch 35850, giga_loss[loss=0.3128, simple_loss=0.3729, pruned_loss=0.1264, over 28659.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3754, pruned_loss=0.1295, over 5678849.99 frames. ], libri_tot_loss[loss=0.3503, simple_loss=0.3948, pruned_loss=0.1529, over 5675272.96 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3676, pruned_loss=0.1236, over 5676131.80 frames. ], batch size: 92, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:29:56,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:30:06,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-01 21:30:15,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2285, 1.3499, 1.1407, 1.5173], device='cuda:1'), covar=tensor([0.2239, 0.2039, 0.1938, 0.2194], device='cuda:1'), in_proj_covar=tensor([0.1047, 0.0829, 0.0929, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 21:30:40,762 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:30:41,546 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-01 21:30:41,726 INFO [train.py:968] (1/2) Epoch 3, batch 35900, giga_loss[loss=0.295, simple_loss=0.368, pruned_loss=0.111, over 29041.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3764, pruned_loss=0.1281, over 5673276.81 frames. ], libri_tot_loss[loss=0.3503, simple_loss=0.3948, pruned_loss=0.1529, over 5677512.79 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.37, pruned_loss=0.1232, over 5669211.88 frames. ], batch size: 136, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:30:58,634 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 35950, giga_loss[loss=0.2841, simple_loss=0.3602, pruned_loss=0.104, over 28891.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3768, pruned_loss=0.1271, over 5671219.43 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.3948, pruned_loss=0.1529, over 5678402.02 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.123, over 5667143.58 frames. ], batch size: 174, lr: 9.27e-03, grad_scale: 4.0
+2023-03-01 21:31:49,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:31:52,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:07,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:11,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9709, 4.4426, 4.6727, 2.0664], device='cuda:1'), covar=tensor([0.0327, 0.0322, 0.0615, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0609, 0.0761, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 21:32:11,846 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 36000, giga_loss[loss=0.3042, simple_loss=0.3779, pruned_loss=0.1153, over 28815.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3801, pruned_loss=0.1299, over 5681966.78 frames. ], libri_tot_loss[loss=0.3517, simple_loss=0.3959, pruned_loss=0.1537, over 5680217.60 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3748, pruned_loss=0.1254, over 5677199.18 frames. ], batch size: 199, lr: 9.27e-03, grad_scale: 8.0
+2023-03-01 21:32:12,165 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 21:32:21,066 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 21:32:26,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4989, 2.1281, 1.5697, 0.7489], device='cuda:1'), covar=tensor([0.2016, 0.1027, 0.1729, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.1314, 0.1237, 0.1315, 0.1115], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 21:32:27,793 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:32:43,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:55,753 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:32:57,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:02,483 INFO [train.py:968] (1/2) Epoch 3, batch 36050, giga_loss[loss=0.3754, simple_loss=0.4186, pruned_loss=0.1661, over 28960.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3833, pruned_loss=0.1324, over 5679165.66 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.3965, pruned_loss=0.154, over 5679204.52 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3782, pruned_loss=0.1282, over 5676199.47 frames. ], batch size: 106, lr: 9.26e-03, grad_scale: 8.0
+2023-03-01 21:33:23,709 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:29,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:48,481 INFO [train.py:968] (1/2) Epoch 3, batch 36100, giga_loss[loss=0.3378, simple_loss=0.4081, pruned_loss=0.1337, over 28804.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3853, pruned_loss=0.1326, over 5684996.35 frames. ], libri_tot_loss[loss=0.3524, simple_loss=0.3967, pruned_loss=0.1541, over 5676886.50 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3811, pruned_loss=0.1292, over 5684743.05 frames. ], batch size: 199, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:33:48,609 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:33:56,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4374, 1.4363, 1.4385, 1.4771], device='cuda:1'), covar=tensor([0.1018, 0.1420, 0.1568, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0772, 0.0622, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 21:33:58,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6819, 1.6366, 1.2175, 1.3971], device='cuda:1'), covar=tensor([0.0672, 0.0582, 0.1004, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0471, 0.0517, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 21:34:01,249 INFO [optim.py:369] (1/2) 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 21:34:08,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8068, 1.1886, 3.5405, 3.0615], device='cuda:1'), covar=tensor([0.1582, 0.1997, 0.0333, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0493, 0.0671, 0.0538], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 21:34:26,186 INFO [train.py:968] (1/2) Epoch 3, batch 36150, giga_loss[loss=0.3339, simple_loss=0.4019, pruned_loss=0.133, over 28373.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3883, pruned_loss=0.1334, over 5693105.16 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.3973, pruned_loss=0.1546, over 5673214.90 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3842, pruned_loss=0.1297, over 5697369.77 frames. ], batch size: 65, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:34:26,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-01 21:35:11,941 INFO [train.py:968] (1/2) Epoch 3, batch 36200, giga_loss[loss=0.3161, simple_loss=0.3868, pruned_loss=0.1227, over 28940.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3892, pruned_loss=0.1333, over 5686477.47 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.3972, pruned_loss=0.1545, over 5674389.82 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3861, pruned_loss=0.1304, over 5688708.08 frames. ], batch size: 145, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:35:14,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:35:25,152 INFO [optim.py:369] (1/2) 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 21:35:28,131 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:35:30,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:35:49,591 INFO [train.py:968] (1/2) Epoch 3, batch 36250, giga_loss[loss=0.2964, simple_loss=0.3725, pruned_loss=0.1101, over 28717.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3891, pruned_loss=0.1322, over 5692206.33 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.3975, pruned_loss=0.1546, over 5679925.31 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3861, pruned_loss=0.1292, over 5689330.07 frames. ], batch size: 262, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:35:54,217 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:36:22,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 21:36:29,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3266, 1.4648, 1.2417, 1.3734], device='cuda:1'), covar=tensor([0.0906, 0.0357, 0.0379, 0.1008], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0143, 0.0147, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 21:36:30,058 INFO [train.py:968] (1/2) Epoch 3, batch 36300, giga_loss[loss=0.3395, simple_loss=0.4019, pruned_loss=0.1386, over 28276.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3878, pruned_loss=0.1299, over 5693015.17 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.3979, pruned_loss=0.1549, over 5673258.48 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.385, pruned_loss=0.1271, over 5696517.14 frames. ], batch size: 368, lr: 9.26e-03, grad_scale: 4.0
+2023-03-01 21:36:45,997 INFO [optim.py:369] (1/2) 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 21:37:06,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:37:08,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:37:09,088 INFO [train.py:968] (1/2) Epoch 3, batch 36350, giga_loss[loss=0.3315, simple_loss=0.3944, pruned_loss=0.1343, over 28618.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3863, pruned_loss=0.1285, over 5688994.50 frames. ], libri_tot_loss[loss=0.3547, simple_loss=0.3989, pruned_loss=0.1552, over 5670766.57 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3829, pruned_loss=0.1252, over 5694714.18 frames. ], batch size: 307, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:37:14,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6742, 1.4436, 1.5387, 1.4209], device='cuda:1'), covar=tensor([0.0974, 0.1767, 0.1503, 0.1431], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0766, 0.0626, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 21:37:24,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7190, 4.5313, 1.8327, 1.6815], device='cuda:1'), covar=tensor([0.0706, 0.0274, 0.0697, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0449, 0.0305, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 21:37:31,039 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:37:51,105 INFO [train.py:968] (1/2) Epoch 3, batch 36400, giga_loss[loss=0.4656, simple_loss=0.4678, pruned_loss=0.2317, over 26558.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3883, pruned_loss=0.131, over 5678090.07 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4002, pruned_loss=0.1563, over 5667205.98 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3838, pruned_loss=0.1264, over 5686025.98 frames. ], batch size: 555, lr: 9.25e-03, grad_scale: 8.0
+2023-03-01 21:38:08,425 INFO [optim.py:369] (1/2) 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 21:38:09,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-01 21:38:35,898 INFO [train.py:968] (1/2) Epoch 3, batch 36450, giga_loss[loss=0.3566, simple_loss=0.4126, pruned_loss=0.1503, over 28963.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3908, pruned_loss=0.1354, over 5672628.15 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4007, pruned_loss=0.1566, over 5661268.64 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3865, pruned_loss=0.131, over 5683846.87 frames. ], batch size: 227, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:38:58,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=126766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:39:05,233 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:39:20,310 INFO [train.py:968] (1/2) Epoch 3, batch 36500, giga_loss[loss=0.4156, simple_loss=0.4335, pruned_loss=0.1988, over 26549.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3926, pruned_loss=0.139, over 5678044.98 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.4008, pruned_loss=0.1567, over 5667126.60 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3889, pruned_loss=0.1349, over 5681860.45 frames. ], batch size: 555, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:39:35,080 INFO [optim.py:369] (1/2) 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 21:40:05,136 INFO [train.py:968] (1/2) Epoch 3, batch 36550, giga_loss[loss=0.3193, simple_loss=0.3766, pruned_loss=0.131, over 28746.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3919, pruned_loss=0.1396, over 5681780.06 frames. ], libri_tot_loss[loss=0.3573, simple_loss=0.401, pruned_loss=0.1568, over 5668746.48 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3888, pruned_loss=0.1361, over 5683575.31 frames. ], batch size: 284, lr: 9.25e-03, grad_scale: 4.0
+2023-03-01 21:40:07,030 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=126843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:40:47,566 INFO [train.py:968] (1/2) Epoch 3, batch 36600, giga_loss[loss=0.3023, simple_loss=0.3626, pruned_loss=0.121, over 28568.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3908, pruned_loss=0.1393, over 5697400.61 frames. ], libri_tot_loss[loss=0.3578, simple_loss=0.4016, pruned_loss=0.157, over 5674512.58 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3875, pruned_loss=0.136, over 5694252.51 frames. ], batch size: 85, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:41:04,636 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=126909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:41:06,285 INFO [optim.py:369] (1/2) 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 21:41:07,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=126912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:41:29,455 INFO [train.py:968] (1/2) Epoch 3, batch 36650, giga_loss[loss=0.3264, simple_loss=0.3867, pruned_loss=0.1331, over 28837.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3898, pruned_loss=0.1389, over 5698065.92 frames. ], libri_tot_loss[loss=0.3589, simple_loss=0.4025, pruned_loss=0.1576, over 5676266.40 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.386, pruned_loss=0.1351, over 5694345.75 frames. ], batch size: 145, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:41:29,707 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=126941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:42:11,699 INFO [train.py:968] (1/2) Epoch 3, batch 36700, giga_loss[loss=0.3323, simple_loss=0.3852, pruned_loss=0.1397, over 27654.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.388, pruned_loss=0.1365, over 5695692.37 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.403, pruned_loss=0.1579, over 5679269.19 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3844, pruned_loss=0.133, over 5690534.96 frames. ], batch size: 472, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:42:30,527 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 36750, giga_loss[loss=0.28, simple_loss=0.3343, pruned_loss=0.1129, over 23398.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3851, pruned_loss=0.134, over 5684000.93 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4032, pruned_loss=0.1578, over 5670274.62 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3816, pruned_loss=0.1306, over 5688607.92 frames. ], batch size: 705, lr: 9.24e-03, grad_scale: 4.0
+2023-03-01 21:43:39,155 INFO [train.py:968] (1/2) Epoch 3, batch 36800, giga_loss[loss=0.2494, simple_loss=0.3277, pruned_loss=0.08555, over 28995.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3795, pruned_loss=0.1303, over 5679883.76 frames. ], libri_tot_loss[loss=0.3605, simple_loss=0.4041, pruned_loss=0.1584, over 5659470.32 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3752, pruned_loss=0.1263, over 5692395.33 frames. ], batch size: 164, lr: 9.24e-03, grad_scale: 8.0
+2023-03-01 21:43:58,142 INFO [optim.py:369] (1/2) 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 21:44:13,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6148, 1.6605, 1.1722, 1.0636], device='cuda:1'), covar=tensor([0.0702, 0.0599, 0.0629, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.0906, 0.0967, 0.1025], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 21:44:31,238 INFO [train.py:968] (1/2) Epoch 3, batch 36850, giga_loss[loss=0.2531, simple_loss=0.3282, pruned_loss=0.08904, over 28559.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3704, pruned_loss=0.1243, over 5673670.69 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4043, pruned_loss=0.1586, over 5661942.35 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3665, pruned_loss=0.1206, over 5681795.11 frames. ], batch size: 60, lr: 9.24e-03, grad_scale: 8.0
+2023-03-01 21:44:39,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 36900, giga_loss[loss=0.3198, simple_loss=0.3865, pruned_loss=0.1266, over 28703.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3653, pruned_loss=0.1211, over 5670353.43 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4048, pruned_loss=0.1589, over 5666895.63 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3611, pruned_loss=0.1172, over 5672463.04 frames. ], batch size: 242, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:45:38,723 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:45:40,832 INFO [optim.py:369] (1/2) 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 21:45:46,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:46:06,305 INFO [train.py:968] (1/2) Epoch 3, batch 36950, giga_loss[loss=0.2812, simple_loss=0.3482, pruned_loss=0.1071, over 28574.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.367, pruned_loss=0.1224, over 5666269.16 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.4058, pruned_loss=0.1598, over 5661558.62 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3614, pruned_loss=0.1171, over 5673833.03 frames. ], batch size: 71, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:46:16,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4412, 1.4146, 1.4608, 1.4003], device='cuda:1'), covar=tensor([0.1051, 0.1429, 0.1576, 0.1267], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0767, 0.0633, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 21:46:34,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5331, 1.4398, 1.2075, 1.7820], device='cuda:1'), covar=tensor([0.2042, 0.1948, 0.1870, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.1038, 0.0826, 0.0921, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 21:46:43,897 INFO [train.py:968] (1/2) Epoch 3, batch 37000, giga_loss[loss=0.2839, simple_loss=0.354, pruned_loss=0.1069, over 28893.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3675, pruned_loss=0.1221, over 5685466.32 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4062, pruned_loss=0.1597, over 5668650.98 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3614, pruned_loss=0.1169, over 5685429.18 frames. ], batch size: 145, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:46:45,517 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:46:47,440 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:47:09,962 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 37050, giga_loss[loss=0.4014, simple_loss=0.4279, pruned_loss=0.1874, over 26795.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3686, pruned_loss=0.1235, over 5682308.55 frames. ], libri_tot_loss[loss=0.3636, simple_loss=0.407, pruned_loss=0.1601, over 5665892.83 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3615, pruned_loss=0.1176, over 5685797.87 frames. ], batch size: 555, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:47:40,609 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:47:43,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:48:05,036 INFO [train.py:968] (1/2) Epoch 3, batch 37100, giga_loss[loss=0.2719, simple_loss=0.3365, pruned_loss=0.1037, over 28890.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3653, pruned_loss=0.1218, over 5687354.26 frames. ], libri_tot_loss[loss=0.3641, simple_loss=0.4076, pruned_loss=0.1603, over 5666881.78 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3586, pruned_loss=0.1163, over 5689446.97 frames. ], batch size: 145, lr: 9.23e-03, grad_scale: 4.0
+2023-03-01 21:48:07,025 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:48:19,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7497, 1.6655, 1.1303, 1.3494], device='cuda:1'), covar=tensor([0.0654, 0.0623, 0.0960, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0463, 0.0506, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 21:48:22,666 INFO [optim.py:369] (1/2) 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 21:48:46,483 INFO [train.py:968] (1/2) Epoch 3, batch 37150, giga_loss[loss=0.2716, simple_loss=0.3352, pruned_loss=0.1041, over 28358.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3626, pruned_loss=0.1201, over 5700957.26 frames. ], libri_tot_loss[loss=0.3648, simple_loss=0.4083, pruned_loss=0.1607, over 5671688.46 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3558, pruned_loss=0.1146, over 5698743.83 frames. ], batch size: 77, lr: 9.22e-03, grad_scale: 4.0
+2023-03-01 21:49:03,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3787, 1.4304, 1.2333, 1.4873], device='cuda:1'), covar=tensor([0.2114, 0.2050, 0.1951, 0.2132], device='cuda:1'), in_proj_covar=tensor([0.1034, 0.0821, 0.0916, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 21:49:28,140 INFO [train.py:968] (1/2) Epoch 3, batch 37200, giga_loss[loss=0.2839, simple_loss=0.3469, pruned_loss=0.1105, over 28961.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3591, pruned_loss=0.1178, over 5701087.89 frames. ], libri_tot_loss[loss=0.365, simple_loss=0.4086, pruned_loss=0.1608, over 5665893.65 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3529, pruned_loss=0.1128, over 5705759.25 frames. ], batch size: 145, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:49:31,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7254, 1.9334, 1.8138, 1.7573], device='cuda:1'), covar=tensor([0.1531, 0.1596, 0.1178, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0781, 0.0749, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 21:49:44,469 INFO [optim.py:369] (1/2) 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 21:49:52,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3614, 1.4498, 1.1185, 1.4218], device='cuda:1'), covar=tensor([0.0861, 0.0360, 0.0387, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0143, 0.0147, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 21:50:09,425 INFO [train.py:968] (1/2) Epoch 3, batch 37250, giga_loss[loss=0.2416, simple_loss=0.3142, pruned_loss=0.08446, over 28345.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3567, pruned_loss=0.1166, over 5700181.92 frames. ], libri_tot_loss[loss=0.3654, simple_loss=0.4089, pruned_loss=0.1609, over 5667768.87 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3507, pruned_loss=0.112, over 5702739.43 frames. ], batch size: 60, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:50:11,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5398, 1.4781, 1.5196, 1.5196], device='cuda:1'), covar=tensor([0.1013, 0.1626, 0.1489, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0763, 0.0627, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 21:50:43,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=127584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:50:48,529 INFO [train.py:968] (1/2) Epoch 3, batch 37300, giga_loss[loss=0.2272, simple_loss=0.3066, pruned_loss=0.07388, over 29044.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3555, pruned_loss=0.1165, over 5706466.24 frames. ], libri_tot_loss[loss=0.3664, simple_loss=0.4098, pruned_loss=0.1615, over 5674855.72 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3485, pruned_loss=0.1111, over 5703210.33 frames. ], batch size: 155, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:51:02,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5695, 1.5653, 1.4899, 1.6157], device='cuda:1'), covar=tensor([0.0982, 0.1502, 0.1567, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0770, 0.0630, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 21:51:04,110 INFO [optim.py:369] (1/2) 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 21:51:27,269 INFO [train.py:968] (1/2) Epoch 3, batch 37350, libri_loss[loss=0.3595, simple_loss=0.4185, pruned_loss=0.1503, over 29201.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3538, pruned_loss=0.1153, over 5720145.20 frames. ], libri_tot_loss[loss=0.3666, simple_loss=0.4102, pruned_loss=0.1615, over 5682966.78 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3458, pruned_loss=0.1092, over 5711448.25 frames. ], batch size: 97, lr: 9.22e-03, grad_scale: 8.0
+2023-03-01 21:51:28,235 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:52:06,011 INFO [train.py:968] (1/2) Epoch 3, batch 37400, giga_loss[loss=0.2714, simple_loss=0.3314, pruned_loss=0.1057, over 28859.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3527, pruned_loss=0.1148, over 5727735.34 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4119, pruned_loss=0.1627, over 5688825.05 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3433, pruned_loss=0.1077, over 5716674.99 frames. ], batch size: 199, lr: 9.22e-03, grad_scale: 2.0
+2023-03-01 21:52:26,478 INFO [optim.py:369] (1/2) 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 21:52:36,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=127727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:52:39,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=127730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:52:47,323 INFO [train.py:968] (1/2) Epoch 3, batch 37450, giga_loss[loss=0.251, simple_loss=0.3155, pruned_loss=0.09323, over 28519.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3497, pruned_loss=0.1127, over 5733206.25 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4118, pruned_loss=0.1625, over 5692099.23 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3412, pruned_loss=0.1063, over 5722188.52 frames. ], batch size: 71, lr: 9.21e-03, grad_scale: 2.0
+2023-03-01 21:53:03,259 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=127759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:53:29,969 INFO [train.py:968] (1/2) Epoch 3, batch 37500, giga_loss[loss=0.4057, simple_loss=0.4339, pruned_loss=0.1888, over 26495.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3508, pruned_loss=0.1134, over 5722404.61 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4124, pruned_loss=0.1626, over 5687476.21 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3422, pruned_loss=0.1072, over 5718724.74 frames. ], batch size: 555, lr: 9.21e-03, grad_scale: 2.0
+2023-03-01 21:53:46,783 INFO [optim.py:369] (1/2) 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 21:53:52,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127820.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 21:54:11,688 INFO [train.py:968] (1/2) Epoch 3, batch 37550, libri_loss[loss=0.4169, simple_loss=0.4609, pruned_loss=0.1865, over 28957.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.356, pruned_loss=0.1167, over 5715750.69 frames. ], libri_tot_loss[loss=0.369, simple_loss=0.4129, pruned_loss=0.1626, over 5688362.86 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3467, pruned_loss=0.1102, over 5713053.97 frames. ], batch size: 107, lr: 9.21e-03, grad_scale: 2.0
+2023-03-01 21:54:36,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7201, 3.4772, 3.4247, 1.6049], device='cuda:1'), covar=tensor([0.0530, 0.0426, 0.0776, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0614, 0.0774, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-01 21:54:53,755 INFO [train.py:968] (1/2) Epoch 3, batch 37600, giga_loss[loss=0.298, simple_loss=0.3642, pruned_loss=0.1159, over 28618.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.363, pruned_loss=0.1214, over 5710202.65 frames. ], libri_tot_loss[loss=0.3701, simple_loss=0.4139, pruned_loss=0.1631, over 5692454.81 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3533, pruned_loss=0.1145, over 5705121.95 frames. ], batch size: 71, lr: 9.21e-03, grad_scale: 4.0
+2023-03-01 21:55:03,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2050, 1.1754, 1.0249, 1.1182], device='cuda:1'), covar=tensor([0.0564, 0.0446, 0.0884, 0.0675], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0479, 0.0519, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 21:55:15,720 INFO [optim.py:369] (1/2) 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 21:55:26,049 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=127922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:55:41,697 INFO [train.py:968] (1/2) Epoch 3, batch 37650, giga_loss[loss=0.3921, simple_loss=0.4311, pruned_loss=0.1765, over 28648.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3731, pruned_loss=0.1287, over 5697396.27 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4142, pruned_loss=0.1633, over 5687543.07 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3641, pruned_loss=0.1223, over 5698863.36 frames. ], batch size: 307, lr: 9.21e-03, grad_scale: 4.0
+2023-03-01 21:56:34,586 INFO [train.py:968] (1/2) Epoch 3, batch 37700, giga_loss[loss=0.3314, simple_loss=0.3962, pruned_loss=0.1333, over 28663.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3787, pruned_loss=0.1322, over 5676743.97 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4141, pruned_loss=0.1632, over 5688669.36 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3716, pruned_loss=0.127, over 5676764.75 frames. ], batch size: 242, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:56:56,047 INFO [optim.py:369] (1/2) 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 21:56:59,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:57:00,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2654, 1.9841, 1.3796, 1.4158], device='cuda:1'), covar=tensor([0.0883, 0.0336, 0.0371, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0144, 0.0148, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 21:57:21,654 INFO [train.py:968] (1/2) Epoch 3, batch 37750, giga_loss[loss=0.2992, simple_loss=0.3721, pruned_loss=0.1131, over 28124.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3828, pruned_loss=0.1334, over 5677764.78 frames. ], libri_tot_loss[loss=0.3703, simple_loss=0.4141, pruned_loss=0.1633, over 5689808.01 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3771, pruned_loss=0.1292, over 5676643.41 frames. ], batch size: 77, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:57:35,410 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:57:42,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7615, 1.5670, 1.2431, 1.3191], device='cuda:1'), covar=tensor([0.0610, 0.0601, 0.0973, 0.0959], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0476, 0.0516, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 21:58:08,484 INFO [train.py:968] (1/2) Epoch 3, batch 37800, giga_loss[loss=0.3999, simple_loss=0.4371, pruned_loss=0.1813, over 28882.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3895, pruned_loss=0.138, over 5675789.81 frames. ], libri_tot_loss[loss=0.3706, simple_loss=0.4144, pruned_loss=0.1634, over 5692750.87 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3843, pruned_loss=0.1341, over 5672236.37 frames. ], batch size: 186, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:58:10,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4191, 3.0560, 1.3956, 1.2760], device='cuda:1'), covar=tensor([0.0876, 0.0385, 0.0896, 0.1519], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0450, 0.0301, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0020, 0.0014, 0.0018], device='cuda:1')
+2023-03-01 21:58:15,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2402, 1.6705, 1.2959, 1.4289], device='cuda:1'), covar=tensor([0.0881, 0.0352, 0.0370, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0144, 0.0147, 0.0265], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0050], device='cuda:1')
+2023-03-01 21:58:28,265 INFO [optim.py:369] (1/2) 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 21:58:51,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2585, 1.7247, 1.1193, 1.5164], device='cuda:1'), covar=tensor([0.0880, 0.0356, 0.0398, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0143, 0.0147, 0.0264], device='cuda:1'), out_proj_covar=tensor([0.0044, 0.0032, 0.0029, 0.0049], device='cuda:1')
+2023-03-01 21:58:51,515 INFO [train.py:968] (1/2) Epoch 3, batch 37850, giga_loss[loss=0.25, simple_loss=0.3247, pruned_loss=0.08761, over 28243.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3887, pruned_loss=0.1375, over 5674668.58 frames. ], libri_tot_loss[loss=0.3709, simple_loss=0.4145, pruned_loss=0.1637, over 5695994.41 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3842, pruned_loss=0.1339, over 5668831.00 frames. ], batch size: 368, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:59:07,677 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:10,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:17,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:33,183 INFO [train.py:968] (1/2) Epoch 3, batch 37900, giga_loss[loss=0.2936, simple_loss=0.361, pruned_loss=0.1131, over 28950.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3824, pruned_loss=0.1322, over 5683471.73 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4146, pruned_loss=0.1639, over 5695425.05 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3784, pruned_loss=0.1288, over 5679280.20 frames. ], batch size: 106, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 21:59:34,645 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 21:59:37,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 21:59:50,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4832, 1.4069, 1.4549, 1.3167], device='cuda:1'), covar=tensor([0.0956, 0.1343, 0.1447, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0759, 0.0617, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0007, 0.0008], device='cuda:1')
+2023-03-01 21:59:55,743 INFO [optim.py:369] (1/2) 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 22:00:16,041 INFO [train.py:968] (1/2) Epoch 3, batch 37950, giga_loss[loss=0.2918, simple_loss=0.3625, pruned_loss=0.1105, over 28463.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3811, pruned_loss=0.1304, over 5676590.95 frames. ], libri_tot_loss[loss=0.3721, simple_loss=0.4152, pruned_loss=0.1645, over 5688793.39 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3765, pruned_loss=0.1263, over 5679317.91 frames. ], batch size: 71, lr: 9.20e-03, grad_scale: 4.0
+2023-03-01 22:00:19,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:00:20,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4833, 1.4337, 1.1468, 1.2236], device='cuda:1'), covar=tensor([0.0680, 0.0578, 0.1032, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0471, 0.0519, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 22:00:37,245 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:00:40,968 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-01 22:00:58,693 INFO [train.py:968] (1/2) Epoch 3, batch 38000, giga_loss[loss=0.3075, simple_loss=0.3692, pruned_loss=0.1229, over 28984.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3802, pruned_loss=0.129, over 5683487.11 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4154, pruned_loss=0.1645, over 5692391.05 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3757, pruned_loss=0.1251, over 5682307.08 frames. ], batch size: 66, lr: 9.19e-03, grad_scale: 8.0
+2023-03-01 22:01:03,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:01:19,185 INFO [optim.py:369] (1/2) 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 22:01:41,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-01 22:01:41,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128338.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:01:43,113 INFO [train.py:968] (1/2) Epoch 3, batch 38050, libri_loss[loss=0.389, simple_loss=0.4351, pruned_loss=0.1714, over 29536.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3845, pruned_loss=0.132, over 5688522.05 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4154, pruned_loss=0.1645, over 5697962.47 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3802, pruned_loss=0.1281, over 5682516.65 frames. ], batch size: 84, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:01:43,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128341.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:02:07,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128370.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:02:23,637 INFO [train.py:968] (1/2) Epoch 3, batch 38100, giga_loss[loss=0.3419, simple_loss=0.3988, pruned_loss=0.1425, over 29002.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3875, pruned_loss=0.1344, over 5694357.02 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4156, pruned_loss=0.1647, over 5702185.05 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3831, pruned_loss=0.1302, over 5685582.62 frames. ], batch size: 136, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:02:44,622 INFO [optim.py:369] (1/2) 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 22:02:56,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:06,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:07,288 INFO [train.py:968] (1/2) Epoch 3, batch 38150, giga_loss[loss=0.3564, simple_loss=0.4037, pruned_loss=0.1546, over 27592.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3884, pruned_loss=0.1349, over 5689027.12 frames. ], libri_tot_loss[loss=0.3732, simple_loss=0.4162, pruned_loss=0.1651, over 5695470.53 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3837, pruned_loss=0.1306, over 5687932.01 frames. ], batch size: 472, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:03:10,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7317, 1.5574, 1.2046, 1.3452], device='cuda:1'), covar=tensor([0.0620, 0.0696, 0.1004, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0469, 0.0517, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 22:03:10,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:35,014 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:03:51,988 INFO [train.py:968] (1/2) Epoch 3, batch 38200, giga_loss[loss=0.3904, simple_loss=0.4237, pruned_loss=0.1786, over 28766.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3895, pruned_loss=0.1364, over 5688521.79 frames. ], libri_tot_loss[loss=0.374, simple_loss=0.4167, pruned_loss=0.1657, over 5698726.97 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3849, pruned_loss=0.132, over 5684574.88 frames. ], batch size: 242, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:04:13,631 INFO [optim.py:369] (1/2) 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 22:04:25,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6184, 1.4884, 5.1333, 3.6514], device='cuda:1'), covar=tensor([0.1445, 0.1923, 0.0250, 0.0463], device='cuda:1'), in_proj_covar=tensor([0.0533, 0.0497, 0.0686, 0.0550], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 22:04:34,838 INFO [train.py:968] (1/2) Epoch 3, batch 38250, giga_loss[loss=0.2976, simple_loss=0.3641, pruned_loss=0.1156, over 28786.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3906, pruned_loss=0.1372, over 5689618.92 frames. ], libri_tot_loss[loss=0.3742, simple_loss=0.417, pruned_loss=0.1658, over 5691166.51 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3864, pruned_loss=0.1332, over 5693211.03 frames. ], batch size: 119, lr: 9.19e-03, grad_scale: 4.0
+2023-03-01 22:04:40,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:05:06,241 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:05:08,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8190, 1.1517, 3.9937, 3.0482], device='cuda:1'), covar=tensor([0.1731, 0.2073, 0.0321, 0.0619], device='cuda:1'), in_proj_covar=tensor([0.0532, 0.0496, 0.0685, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 22:05:08,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:05:19,160 INFO [train.py:968] (1/2) Epoch 3, batch 38300, giga_loss[loss=0.3277, simple_loss=0.3932, pruned_loss=0.1311, over 29027.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3901, pruned_loss=0.1355, over 5695770.42 frames. ], libri_tot_loss[loss=0.3745, simple_loss=0.417, pruned_loss=0.1659, over 5692080.68 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3866, pruned_loss=0.1322, over 5697771.45 frames. ], batch size: 128, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:05:20,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4708, 2.0963, 1.6412, 1.8841], device='cuda:1'), covar=tensor([0.0529, 0.0662, 0.0889, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0461, 0.0514, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 22:05:32,542 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:05:40,127 INFO [optim.py:369] (1/2) 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 22:05:42,332 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:06:01,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=128640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:06:02,218 INFO [train.py:968] (1/2) Epoch 3, batch 38350, giga_loss[loss=0.3522, simple_loss=0.4097, pruned_loss=0.1473, over 29033.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.389, pruned_loss=0.1329, over 5704395.22 frames. ], libri_tot_loss[loss=0.3751, simple_loss=0.4174, pruned_loss=0.1663, over 5695535.44 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3855, pruned_loss=0.1295, over 5703075.72 frames. ], batch size: 106, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:06:42,948 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:06:43,386 INFO [train.py:968] (1/2) Epoch 3, batch 38400, libri_loss[loss=0.363, simple_loss=0.414, pruned_loss=0.156, over 25804.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3892, pruned_loss=0.1325, over 5703248.03 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4174, pruned_loss=0.1663, over 5696853.62 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3858, pruned_loss=0.129, over 5701333.92 frames. ], batch size: 136, lr: 9.18e-03, grad_scale: 8.0
+2023-03-01 22:06:45,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:01,636 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:04,515 INFO [optim.py:369] (1/2) 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 22:07:10,292 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:10,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6685, 3.0410, 1.9291, 1.7514], device='cuda:1'), covar=tensor([0.0678, 0.0375, 0.0458, 0.0591], device='cuda:1'), in_proj_covar=tensor([0.1213, 0.0917, 0.0971, 0.1050], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 22:07:24,226 INFO [train.py:968] (1/2) Epoch 3, batch 38450, giga_loss[loss=0.2757, simple_loss=0.3511, pruned_loss=0.1002, over 28466.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3876, pruned_loss=0.1319, over 5705728.41 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.418, pruned_loss=0.1668, over 5700694.43 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3833, pruned_loss=0.1274, over 5701150.12 frames. ], batch size: 71, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:07:40,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:07:43,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:00,872 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=128783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:03,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=128786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:06,236 INFO [train.py:968] (1/2) Epoch 3, batch 38500, giga_loss[loss=0.2952, simple_loss=0.3614, pruned_loss=0.1145, over 28703.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3856, pruned_loss=0.1307, over 5703372.97 frames. ], libri_tot_loss[loss=0.3755, simple_loss=0.4177, pruned_loss=0.1666, over 5695198.07 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3813, pruned_loss=0.126, over 5705803.26 frames. ], batch size: 78, lr: 9.18e-03, grad_scale: 4.0
+2023-03-01 22:08:07,756 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:18,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1784, 4.5673, 4.8188, 2.2942], device='cuda:1'), covar=tensor([0.0338, 0.0334, 0.0644, 0.1821], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0632, 0.0783, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 22:08:18,804 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=128806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:25,403 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=128815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:08:25,873 INFO [optim.py:369] (1/2) 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 22:08:44,797 INFO [train.py:968] (1/2) Epoch 3, batch 38550, giga_loss[loss=0.3035, simple_loss=0.3761, pruned_loss=0.1155, over 28932.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3831, pruned_loss=0.1291, over 5716013.13 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4171, pruned_loss=0.1662, over 5701907.71 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3791, pruned_loss=0.1246, over 5712189.25 frames. ], batch size: 145, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:09:08,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1905, 2.9679, 2.9169, 1.4263], device='cuda:1'), covar=tensor([0.0711, 0.0597, 0.0974, 0.2129], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0626, 0.0776, 0.0566], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 22:09:26,663 INFO [train.py:968] (1/2) Epoch 3, batch 38600, giga_loss[loss=0.2976, simple_loss=0.3688, pruned_loss=0.1132, over 28918.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3831, pruned_loss=0.1297, over 5711032.04 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4174, pruned_loss=0.1663, over 5702435.36 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3792, pruned_loss=0.1254, over 5707679.82 frames. ], batch size: 136, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:09:46,367 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 38650, giga_loss[loss=0.3399, simple_loss=0.3959, pruned_loss=0.142, over 28831.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3837, pruned_loss=0.1302, over 5709277.44 frames. ], libri_tot_loss[loss=0.3753, simple_loss=0.4176, pruned_loss=0.1665, over 5696257.21 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3796, pruned_loss=0.1258, over 5712956.21 frames. ], batch size: 119, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:10:18,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-01 22:10:46,029 INFO [train.py:968] (1/2) Epoch 3, batch 38700, giga_loss[loss=0.2884, simple_loss=0.3629, pruned_loss=0.107, over 28906.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3823, pruned_loss=0.1281, over 5714037.27 frames. ], libri_tot_loss[loss=0.375, simple_loss=0.4174, pruned_loss=0.1663, over 5699166.98 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3788, pruned_loss=0.1244, over 5714553.17 frames. ], batch size: 227, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:10:46,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9396, 2.3610, 2.0489, 1.7984], device='cuda:1'), covar=tensor([0.0461, 0.0619, 0.0803, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0467, 0.0516, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 22:11:07,129 INFO [optim.py:369] (1/2) 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 22:11:25,737 INFO [train.py:968] (1/2) Epoch 3, batch 38750, giga_loss[loss=0.2965, simple_loss=0.3636, pruned_loss=0.1147, over 28899.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3812, pruned_loss=0.1266, over 5709114.90 frames. ], libri_tot_loss[loss=0.3749, simple_loss=0.4174, pruned_loss=0.1662, over 5701275.96 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3781, pruned_loss=0.1233, over 5707634.21 frames. ], batch size: 186, lr: 9.17e-03, grad_scale: 4.0
+2023-03-01 22:12:00,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5969, 4.1630, 4.3315, 1.7847], device='cuda:1'), covar=tensor([0.0430, 0.0378, 0.0796, 0.2154], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0630, 0.0788, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 22:12:03,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1055, 1.2919, 3.4356, 3.2148], device='cuda:1'), covar=tensor([0.1266, 0.1852, 0.0317, 0.0632], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0494, 0.0669, 0.0548], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 22:12:04,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:12:06,874 INFO [train.py:968] (1/2) Epoch 3, batch 38800, giga_loss[loss=0.3125, simple_loss=0.3741, pruned_loss=0.1255, over 28847.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3806, pruned_loss=0.1263, over 5715091.87 frames. ], libri_tot_loss[loss=0.3753, simple_loss=0.4177, pruned_loss=0.1665, over 5704471.45 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3773, pruned_loss=0.1229, over 5711117.40 frames. ], batch size: 112, lr: 9.17e-03, grad_scale: 8.0
+2023-03-01 22:12:26,042 INFO [optim.py:369] (1/2) 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 22:12:48,562 INFO [train.py:968] (1/2) Epoch 3, batch 38850, giga_loss[loss=0.2651, simple_loss=0.3421, pruned_loss=0.09405, over 28675.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3781, pruned_loss=0.1253, over 5714073.26 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4178, pruned_loss=0.1665, over 5705303.09 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3753, pruned_loss=0.1224, over 5710214.45 frames. ], batch size: 60, lr: 9.16e-03, grad_scale: 8.0
+2023-03-01 22:13:16,849 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:13:21,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:13:28,206 INFO [train.py:968] (1/2) Epoch 3, batch 38900, giga_loss[loss=0.2858, simple_loss=0.3517, pruned_loss=0.1099, over 28444.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3766, pruned_loss=0.1253, over 5707886.41 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4173, pruned_loss=0.1661, over 5708909.44 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3729, pruned_loss=0.1213, over 5701704.71 frames. ], batch size: 65, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:13:48,812 INFO [optim.py:369] (1/2) 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 22:13:59,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:14:01,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:14:06,971 INFO [train.py:968] (1/2) Epoch 3, batch 38950, giga_loss[loss=0.3219, simple_loss=0.3838, pruned_loss=0.13, over 28722.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3758, pruned_loss=0.1254, over 5701574.70 frames. ], libri_tot_loss[loss=0.3747, simple_loss=0.4174, pruned_loss=0.166, over 5697977.26 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3712, pruned_loss=0.1207, over 5706877.50 frames. ], batch size: 242, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:14:24,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:14:25,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 22:14:39,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129283.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:14:45,757 INFO [train.py:968] (1/2) Epoch 3, batch 39000, giga_loss[loss=0.3116, simple_loss=0.3726, pruned_loss=0.1253, over 28967.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1268, over 5708956.37 frames. ], libri_tot_loss[loss=0.374, simple_loss=0.4168, pruned_loss=0.1656, over 5705731.65 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3722, pruned_loss=0.1218, over 5706474.07 frames. ], batch size: 227, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:14:45,757 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 22:14:54,201 INFO [train.py:1012] (1/2) Epoch 3, validation: loss=0.2507, simple_loss=0.3503, pruned_loss=0.07558, over 944034.00 frames. 
+2023-03-01 22:14:54,202 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 22:15:16,482 INFO [optim.py:369] (1/2) 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 22:15:21,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:15:23,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:15:34,438 INFO [train.py:968] (1/2) Epoch 3, batch 39050, giga_loss[loss=0.2882, simple_loss=0.357, pruned_loss=0.1097, over 28916.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3749, pruned_loss=0.126, over 5703265.39 frames. ], libri_tot_loss[loss=0.374, simple_loss=0.4168, pruned_loss=0.1656, over 5704935.69 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3705, pruned_loss=0.1215, over 5702108.35 frames. ], batch size: 213, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:15:47,775 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:16:16,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-01 22:16:16,891 INFO [train.py:968] (1/2) Epoch 3, batch 39100, giga_loss[loss=0.278, simple_loss=0.347, pruned_loss=0.1045, over 28951.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3731, pruned_loss=0.1254, over 5705393.11 frames. ], libri_tot_loss[loss=0.3745, simple_loss=0.4172, pruned_loss=0.1659, over 5705420.18 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3684, pruned_loss=0.1208, over 5704083.50 frames. ], batch size: 145, lr: 9.16e-03, grad_scale: 4.0
+2023-03-01 22:16:35,424 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 39150, giga_loss[loss=0.3235, simple_loss=0.3817, pruned_loss=0.1326, over 28673.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3708, pruned_loss=0.1246, over 5706718.34 frames. ], libri_tot_loss[loss=0.3745, simple_loss=0.4171, pruned_loss=0.166, over 5700796.00 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3657, pruned_loss=0.1196, over 5709780.10 frames. ], batch size: 262, lr: 9.15e-03, grad_scale: 4.0
+2023-03-01 22:17:35,865 INFO [train.py:968] (1/2) Epoch 3, batch 39200, libri_loss[loss=0.4074, simple_loss=0.4457, pruned_loss=0.1845, over 29661.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3696, pruned_loss=0.1244, over 5703580.06 frames. ], libri_tot_loss[loss=0.3745, simple_loss=0.4171, pruned_loss=0.1659, over 5706021.89 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3644, pruned_loss=0.1194, over 5701257.26 frames. ], batch size: 91, lr: 9.15e-03, grad_scale: 8.0
+2023-03-01 22:17:57,942 INFO [optim.py:369] (1/2) 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 22:18:18,055 INFO [train.py:968] (1/2) Epoch 3, batch 39250, giga_loss[loss=0.2959, simple_loss=0.3768, pruned_loss=0.1075, over 29025.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3672, pruned_loss=0.1223, over 5703575.42 frames. ], libri_tot_loss[loss=0.3749, simple_loss=0.4174, pruned_loss=0.1662, over 5706576.78 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3626, pruned_loss=0.118, over 5701289.32 frames. ], batch size: 164, lr: 9.15e-03, grad_scale: 8.0
+2023-03-01 22:18:25,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:18:59,517 INFO [train.py:968] (1/2) Epoch 3, batch 39300, giga_loss[loss=0.3066, simple_loss=0.3791, pruned_loss=0.1171, over 28735.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3706, pruned_loss=0.1243, over 5690585.92 frames. ], libri_tot_loss[loss=0.3756, simple_loss=0.4179, pruned_loss=0.1667, over 5698161.22 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.365, pruned_loss=0.1191, over 5696575.50 frames. ], batch size: 284, lr: 9.15e-03, grad_scale: 4.0
+2023-03-01 22:19:19,630 INFO [optim.py:369] (1/2) 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 22:19:35,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6555, 1.7459, 1.4409, 1.0732], device='cuda:1'), covar=tensor([0.0908, 0.0651, 0.0488, 0.0722], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.0928, 0.0970, 0.1067], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 22:19:40,138 INFO [train.py:968] (1/2) Epoch 3, batch 39350, giga_loss[loss=0.336, simple_loss=0.4022, pruned_loss=0.1349, over 28063.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3758, pruned_loss=0.1278, over 5671356.24 frames. ], libri_tot_loss[loss=0.3765, simple_loss=0.4186, pruned_loss=0.1672, over 5683044.29 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3689, pruned_loss=0.1215, over 5689768.57 frames. ], batch size: 412, lr: 9.15e-03, grad_scale: 2.0
+2023-03-01 22:19:40,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1591, 1.3163, 1.1660, 0.7094], device='cuda:1'), covar=tensor([0.0788, 0.0598, 0.0395, 0.0631], device='cuda:1'), in_proj_covar=tensor([0.1227, 0.0924, 0.0965, 0.1061], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 22:19:55,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=129658.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:20:23,857 INFO [train.py:968] (1/2) Epoch 3, batch 39400, giga_loss[loss=0.2891, simple_loss=0.366, pruned_loss=0.106, over 28685.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3761, pruned_loss=0.1266, over 5681939.91 frames. ], libri_tot_loss[loss=0.3761, simple_loss=0.4183, pruned_loss=0.167, over 5684670.44 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3703, pruned_loss=0.1212, over 5694911.72 frames. ], batch size: 262, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:20:26,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:20:28,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:20:49,733 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:21:07,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-01 22:21:09,813 INFO [train.py:968] (1/2) Epoch 3, batch 39450, giga_loss[loss=0.3013, simple_loss=0.3784, pruned_loss=0.1121, over 28698.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3742, pruned_loss=0.1242, over 5684360.69 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4182, pruned_loss=0.1668, over 5685809.09 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.1199, over 5693563.15 frames. ], batch size: 262, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:21:34,833 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-01 22:21:50,885 INFO [train.py:968] (1/2) Epoch 3, batch 39500, giga_loss[loss=0.3215, simple_loss=0.3759, pruned_loss=0.1335, over 28827.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3736, pruned_loss=0.124, over 5686582.16 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.4181, pruned_loss=0.1668, over 5681851.11 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3686, pruned_loss=0.1193, over 5698259.94 frames. ], batch size: 112, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:21:55,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4085, 1.3674, 1.2353, 1.5218], device='cuda:1'), covar=tensor([0.1910, 0.1909, 0.1874, 0.2151], device='cuda:1'), in_proj_covar=tensor([0.1039, 0.0822, 0.0918, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 22:21:58,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=129801.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:22:01,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=129804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:22:02,279 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=129806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:22:10,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7187, 3.1873, 1.7293, 1.3508], device='cuda:1'), covar=tensor([0.0827, 0.0342, 0.0593, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.0920, 0.0961, 0.1053], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 22:22:12,138 INFO [optim.py:369] (1/2) 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 22:22:25,758 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=129833.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:22:33,344 INFO [train.py:968] (1/2) Epoch 3, batch 39550, giga_loss[loss=0.2671, simple_loss=0.3394, pruned_loss=0.09735, over 28832.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3749, pruned_loss=0.1254, over 5695519.24 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4183, pruned_loss=0.167, over 5688119.21 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5699616.22 frames. ], batch size: 119, lr: 9.14e-03, grad_scale: 2.0
+2023-03-01 22:23:13,795 INFO [train.py:968] (1/2) Epoch 3, batch 39600, giga_loss[loss=0.3022, simple_loss=0.3693, pruned_loss=0.1176, over 28824.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3744, pruned_loss=0.1248, over 5711473.03 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4178, pruned_loss=0.1665, over 5692709.32 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3698, pruned_loss=0.1204, over 5710872.56 frames. ], batch size: 227, lr: 9.14e-03, grad_scale: 4.0
+2023-03-01 22:23:38,856 INFO [optim.py:369] (1/2) 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 22:23:58,130 INFO [train.py:968] (1/2) Epoch 3, batch 39650, libri_loss[loss=0.4763, simple_loss=0.4892, pruned_loss=0.2317, over 19067.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3781, pruned_loss=0.1269, over 5700307.50 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.4181, pruned_loss=0.1668, over 5686527.76 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3734, pruned_loss=0.1225, over 5706618.29 frames. ], batch size: 188, lr: 9.14e-03, grad_scale: 4.0
+2023-03-01 22:24:38,120 INFO [train.py:968] (1/2) Epoch 3, batch 39700, giga_loss[loss=0.3452, simple_loss=0.3961, pruned_loss=0.1471, over 28560.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3815, pruned_loss=0.1291, over 5692902.61 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.4181, pruned_loss=0.1667, over 5679552.32 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3763, pruned_loss=0.1242, over 5704071.05 frames. ], batch size: 71, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:24:42,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5155, 1.9871, 1.9571, 1.8768], device='cuda:1'), covar=tensor([0.0996, 0.1546, 0.1188, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0764, 0.0637, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 22:24:48,718 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:25:00,849 INFO [optim.py:369] (1/2) 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 22:25:15,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3565, 1.2455, 5.1417, 3.7310], device='cuda:1'), covar=tensor([0.1495, 0.2077, 0.0256, 0.0440], device='cuda:1'), in_proj_covar=tensor([0.0530, 0.0497, 0.0685, 0.0552], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 22:25:18,587 INFO [train.py:968] (1/2) Epoch 3, batch 39750, giga_loss[loss=0.3764, simple_loss=0.4048, pruned_loss=0.174, over 23956.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3828, pruned_loss=0.1296, over 5698897.08 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4179, pruned_loss=0.1664, over 5682314.69 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3782, pruned_loss=0.1253, over 5705511.87 frames. ], batch size: 705, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:26:01,284 INFO [train.py:968] (1/2) Epoch 3, batch 39800, libri_loss[loss=0.3819, simple_loss=0.4179, pruned_loss=0.1729, over 29557.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3841, pruned_loss=0.1304, over 5704408.88 frames. ], libri_tot_loss[loss=0.3755, simple_loss=0.4181, pruned_loss=0.1664, over 5687479.30 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3795, pruned_loss=0.1261, over 5705438.16 frames. ], batch size: 77, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:26:23,458 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-01 22:26:23,642 INFO [optim.py:369] (1/2) 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 22:26:41,509 INFO [train.py:968] (1/2) Epoch 3, batch 39850, giga_loss[loss=0.4372, simple_loss=0.4699, pruned_loss=0.2023, over 28953.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3852, pruned_loss=0.1311, over 5709619.86 frames. ], libri_tot_loss[loss=0.3754, simple_loss=0.4179, pruned_loss=0.1664, over 5694318.55 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3808, pruned_loss=0.1267, over 5704910.36 frames. ], batch size: 213, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:27:15,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130181.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:27:17,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:27:22,025 INFO [train.py:968] (1/2) Epoch 3, batch 39900, giga_loss[loss=0.3504, simple_loss=0.3979, pruned_loss=0.1514, over 28716.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3846, pruned_loss=0.1307, over 5713791.41 frames. ], libri_tot_loss[loss=0.3751, simple_loss=0.4178, pruned_loss=0.1662, over 5697327.70 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3808, pruned_loss=0.1269, over 5707722.40 frames. ], batch size: 92, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:27:43,757 INFO [optim.py:369] (1/2) 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 22:27:52,398 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6452, 0.9683, 3.3975, 2.8988], device='cuda:1'), covar=tensor([0.1642, 0.2042, 0.0396, 0.0552], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0502, 0.0697, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 22:28:01,223 INFO [train.py:968] (1/2) Epoch 3, batch 39950, giga_loss[loss=0.243, simple_loss=0.3201, pruned_loss=0.08294, over 28557.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3833, pruned_loss=0.1303, over 5708502.03 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.4186, pruned_loss=0.1667, over 5691247.16 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3787, pruned_loss=0.126, over 5710257.94 frames. ], batch size: 85, lr: 9.13e-03, grad_scale: 4.0
+2023-03-01 22:28:35,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-01 22:28:40,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-01 22:28:42,901 INFO [train.py:968] (1/2) Epoch 3, batch 40000, giga_loss[loss=0.2789, simple_loss=0.3382, pruned_loss=0.1098, over 28598.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3788, pruned_loss=0.1276, over 5713536.03 frames. ], libri_tot_loss[loss=0.3764, simple_loss=0.4189, pruned_loss=0.1669, over 5694392.26 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3745, pruned_loss=0.1236, over 5712329.87 frames. ], batch size: 78, lr: 9.12e-03, grad_scale: 8.0
+2023-03-01 22:29:07,040 INFO [optim.py:369] (1/2) 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 22:29:10,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130324.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:29:10,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1772, 2.8668, 2.9255, 1.2701], device='cuda:1'), covar=tensor([0.0735, 0.0661, 0.1142, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0615, 0.0783, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 22:29:12,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130327.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 22:29:22,918 INFO [train.py:968] (1/2) Epoch 3, batch 40050, giga_loss[loss=0.3434, simple_loss=0.4135, pruned_loss=0.1366, over 28371.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3768, pruned_loss=0.1261, over 5706716.26 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4188, pruned_loss=0.1668, over 5687486.18 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3726, pruned_loss=0.1221, over 5711867.45 frames. ], batch size: 369, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:29:36,906 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130356.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 22:29:45,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6270, 1.4678, 1.2358, 1.2981], device='cuda:1'), covar=tensor([0.0529, 0.0543, 0.0826, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0471, 0.0507, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 22:29:56,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:30:04,218 INFO [train.py:968] (1/2) Epoch 3, batch 40100, giga_loss[loss=0.2678, simple_loss=0.354, pruned_loss=0.09077, over 29099.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3789, pruned_loss=0.1258, over 5710628.89 frames. ], libri_tot_loss[loss=0.3761, simple_loss=0.4187, pruned_loss=0.1667, over 5686886.96 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.375, pruned_loss=0.1221, over 5715489.88 frames. ], batch size: 155, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:30:06,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:30:06,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6493, 2.4113, 1.5185, 0.4713], device='cuda:1'), covar=tensor([0.2635, 0.1274, 0.1755, 0.2902], device='cuda:1'), in_proj_covar=tensor([0.1272, 0.1206, 0.1291, 0.1071], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 22:30:07,730 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:30:30,535 INFO [optim.py:369] (1/2) 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 22:30:47,961 INFO [train.py:968] (1/2) Epoch 3, batch 40150, giga_loss[loss=0.3498, simple_loss=0.4064, pruned_loss=0.1465, over 28595.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3792, pruned_loss=0.1249, over 5705888.97 frames. ], libri_tot_loss[loss=0.3758, simple_loss=0.4184, pruned_loss=0.1665, over 5689265.28 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.376, pruned_loss=0.1217, over 5707978.54 frames. ], batch size: 307, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:31:28,099 INFO [train.py:968] (1/2) Epoch 3, batch 40200, giga_loss[loss=0.3366, simple_loss=0.378, pruned_loss=0.1476, over 28556.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3779, pruned_loss=0.1252, over 5705699.63 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4187, pruned_loss=0.1668, over 5688382.91 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3745, pruned_loss=0.1219, over 5708023.93 frames. ], batch size: 78, lr: 9.12e-03, grad_scale: 4.0
+2023-03-01 22:31:51,838 INFO [optim.py:369] (1/2) 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 22:31:54,737 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:31:57,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:32:07,649 INFO [train.py:968] (1/2) Epoch 3, batch 40250, giga_loss[loss=0.2626, simple_loss=0.3236, pruned_loss=0.1008, over 28629.00 frames. ], tot_loss[loss=0.316, simple_loss=0.378, pruned_loss=0.127, over 5709107.56 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4189, pruned_loss=0.1668, over 5691480.50 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3747, pruned_loss=0.1239, over 5708358.43 frames. ], batch size: 85, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:32:09,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3095, 1.3112, 1.1491, 1.3474], device='cuda:1'), covar=tensor([0.2195, 0.2209, 0.2098, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1036, 0.0824, 0.0923, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0007, 0.0007], device='cuda:1')
+2023-03-01 22:32:18,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:32:22,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:32:50,435 INFO [train.py:968] (1/2) Epoch 3, batch 40300, giga_loss[loss=0.3485, simple_loss=0.3963, pruned_loss=0.1504, over 28877.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3772, pruned_loss=0.1283, over 5706183.59 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4186, pruned_loss=0.1666, over 5695594.17 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.374, pruned_loss=0.1252, over 5702121.16 frames. ], batch size: 213, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:33:14,861 INFO [optim.py:369] (1/2) 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 22:33:32,451 INFO [train.py:968] (1/2) Epoch 3, batch 40350, giga_loss[loss=0.3133, simple_loss=0.3654, pruned_loss=0.1306, over 28862.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3746, pruned_loss=0.1273, over 5708968.04 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4186, pruned_loss=0.1666, over 5688342.08 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3717, pruned_loss=0.1246, over 5712066.32 frames. ], batch size: 227, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:34:13,997 INFO [train.py:968] (1/2) Epoch 3, batch 40400, giga_loss[loss=0.3081, simple_loss=0.3667, pruned_loss=0.1247, over 28789.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3737, pruned_loss=0.1271, over 5711618.61 frames. ], libri_tot_loss[loss=0.3767, simple_loss=0.4192, pruned_loss=0.1671, over 5689239.09 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3701, pruned_loss=0.1238, over 5713932.08 frames. ], batch size: 284, lr: 9.11e-03, grad_scale: 8.0
+2023-03-01 22:34:24,130 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:34:26,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:34:34,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0415, 3.7366, 3.7190, 1.7049], device='cuda:1'), covar=tensor([0.0443, 0.0391, 0.0774, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0622, 0.0787, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 22:34:37,550 INFO [optim.py:369] (1/2) 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 22:34:47,702 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 40450, giga_loss[loss=0.2894, simple_loss=0.352, pruned_loss=0.1134, over 28970.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3713, pruned_loss=0.1261, over 5711629.70 frames. ], libri_tot_loss[loss=0.3765, simple_loss=0.419, pruned_loss=0.167, over 5687671.58 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3667, pruned_loss=0.1219, over 5715483.80 frames. ], batch size: 213, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:35:14,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:35:16,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=130770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:35:20,534 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:35:31,784 INFO [train.py:968] (1/2) Epoch 3, batch 40500, giga_loss[loss=0.2905, simple_loss=0.352, pruned_loss=0.1145, over 28605.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3653, pruned_loss=0.1223, over 5710369.27 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4187, pruned_loss=0.1669, over 5682221.34 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3611, pruned_loss=0.1184, over 5718602.42 frames. ], batch size: 336, lr: 9.11e-03, grad_scale: 4.0
+2023-03-01 22:35:55,208 INFO [optim.py:369] (1/2) 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 22:36:00,081 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:36:05,366 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:36:11,891 INFO [train.py:968] (1/2) Epoch 3, batch 40550, giga_loss[loss=0.2736, simple_loss=0.3454, pruned_loss=0.1009, over 28760.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3625, pruned_loss=0.1205, over 5709836.47 frames. ], libri_tot_loss[loss=0.3764, simple_loss=0.4189, pruned_loss=0.167, over 5684071.09 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3576, pruned_loss=0.1162, over 5715498.28 frames. ], batch size: 99, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:36:55,184 INFO [train.py:968] (1/2) Epoch 3, batch 40600, giga_loss[loss=0.3157, simple_loss=0.3799, pruned_loss=0.1258, over 28946.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3652, pruned_loss=0.1218, over 5704727.06 frames. ], libri_tot_loss[loss=0.3763, simple_loss=0.4188, pruned_loss=0.1669, over 5685770.37 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3608, pruned_loss=0.1178, over 5708071.33 frames. ], batch size: 174, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:37:12,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:13,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=130913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:14,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:15,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=130916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:19,659 INFO [optim.py:369] (1/2) 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 22:37:34,822 INFO [train.py:968] (1/2) Epoch 3, batch 40650, giga_loss[loss=0.3011, simple_loss=0.3749, pruned_loss=0.1137, over 29009.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3703, pruned_loss=0.1244, over 5691085.01 frames. ], libri_tot_loss[loss=0.3768, simple_loss=0.4192, pruned_loss=0.1672, over 5667536.84 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3649, pruned_loss=0.1197, over 5709826.52 frames. ], batch size: 174, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:37:35,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:37,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:38,962 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=130945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:37:49,797 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=130958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:38:14,860 INFO [train.py:968] (1/2) Epoch 3, batch 40700, giga_loss[loss=0.3277, simple_loss=0.3909, pruned_loss=0.1323, over 28797.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1266, over 5691368.25 frames. ], libri_tot_loss[loss=0.3774, simple_loss=0.4196, pruned_loss=0.1676, over 5664846.46 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5709055.04 frames. ], batch size: 284, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:38:40,602 INFO [optim.py:369] (1/2) 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 22:38:57,636 INFO [train.py:968] (1/2) Epoch 3, batch 40750, giga_loss[loss=0.2938, simple_loss=0.3654, pruned_loss=0.1111, over 28821.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3771, pruned_loss=0.1272, over 5705929.51 frames. ], libri_tot_loss[loss=0.3773, simple_loss=0.4195, pruned_loss=0.1676, over 5668786.38 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3719, pruned_loss=0.1226, over 5716717.79 frames. ], batch size: 174, lr: 9.10e-03, grad_scale: 4.0
+2023-03-01 22:39:40,337 INFO [train.py:968] (1/2) Epoch 3, batch 40800, giga_loss[loss=0.3072, simple_loss=0.3802, pruned_loss=0.1171, over 28891.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3803, pruned_loss=0.1294, over 5707666.13 frames. ], libri_tot_loss[loss=0.3775, simple_loss=0.4196, pruned_loss=0.1677, over 5673337.72 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3753, pruned_loss=0.1249, over 5712771.07 frames. ], batch size: 199, lr: 9.10e-03, grad_scale: 8.0
+2023-03-01 22:40:05,541 INFO [optim.py:369] (1/2) 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 22:40:26,027 INFO [train.py:968] (1/2) Epoch 3, batch 40850, giga_loss[loss=0.3381, simple_loss=0.3824, pruned_loss=0.1469, over 28867.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3833, pruned_loss=0.1324, over 5705103.32 frames. ], libri_tot_loss[loss=0.3784, simple_loss=0.4203, pruned_loss=0.1682, over 5676639.48 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3783, pruned_loss=0.1278, over 5706664.22 frames. ], batch size: 112, lr: 9.09e-03, grad_scale: 8.0
+2023-03-01 22:40:34,950 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6174, 1.7904, 1.6600, 1.6423], device='cuda:1'), covar=tensor([0.0785, 0.1068, 0.0733, 0.0637], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0768, 0.0749, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 22:40:40,061 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 40900, giga_loss[loss=0.3531, simple_loss=0.412, pruned_loss=0.1471, over 28692.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3941, pruned_loss=0.1426, over 5680635.13 frames. ], libri_tot_loss[loss=0.378, simple_loss=0.42, pruned_loss=0.168, over 5679078.30 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.39, pruned_loss=0.1389, over 5679788.63 frames. ], batch size: 242, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:41:27,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:41:33,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:41:47,462 INFO [optim.py:369] (1/2) 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 22:42:05,252 INFO [train.py:968] (1/2) Epoch 3, batch 40950, giga_loss[loss=0.3409, simple_loss=0.396, pruned_loss=0.143, over 28583.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4011, pruned_loss=0.1479, over 5685098.79 frames. ], libri_tot_loss[loss=0.3783, simple_loss=0.4203, pruned_loss=0.1681, over 5681190.96 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3973, pruned_loss=0.1446, over 5682442.79 frames. ], batch size: 78, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:42:18,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8641, 0.9158, 0.5888, 0.5805], device='cuda:1'), covar=tensor([0.0418, 0.0422, 0.0416, 0.0528], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.0966, 0.1024, 0.1105], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 22:42:44,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-01 22:42:50,650 INFO [train.py:968] (1/2) Epoch 3, batch 41000, giga_loss[loss=0.3538, simple_loss=0.4099, pruned_loss=0.1489, over 28975.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.407, pruned_loss=0.1537, over 5674893.94 frames. ], libri_tot_loss[loss=0.3777, simple_loss=0.4198, pruned_loss=0.1678, over 5684689.87 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.4042, pruned_loss=0.1511, over 5669619.60 frames. ], batch size: 145, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:42:53,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:42:56,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:12,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:16,344 INFO [optim.py:369] (1/2) 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 22:43:20,525 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:24,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-01 22:43:24,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:32,603 INFO [train.py:968] (1/2) Epoch 3, batch 41050, giga_loss[loss=0.4119, simple_loss=0.443, pruned_loss=0.1904, over 27910.00 frames. ], tot_loss[loss=0.3665, simple_loss=0.4135, pruned_loss=0.1598, over 5686070.11 frames. ], libri_tot_loss[loss=0.3774, simple_loss=0.4196, pruned_loss=0.1676, over 5692616.06 frames. ], giga_tot_loss[loss=0.363, simple_loss=0.4111, pruned_loss=0.1574, over 5674501.94 frames. ], batch size: 412, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:43:34,634 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:38,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:40,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:43:43,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:44:03,572 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:44:08,701 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:44:18,198 INFO [train.py:968] (1/2) Epoch 3, batch 41100, giga_loss[loss=0.4523, simple_loss=0.4649, pruned_loss=0.2198, over 27449.00 frames. ], tot_loss[loss=0.3745, simple_loss=0.4192, pruned_loss=0.165, over 5662384.15 frames. ], libri_tot_loss[loss=0.378, simple_loss=0.4201, pruned_loss=0.1679, over 5686151.69 frames. ], giga_tot_loss[loss=0.3711, simple_loss=0.4168, pruned_loss=0.1627, over 5658168.21 frames. ], batch size: 472, lr: 9.09e-03, grad_scale: 4.0
+2023-03-01 22:44:52,374 INFO [optim.py:369] (1/2) 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 22:45:09,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1727, 2.9422, 2.9333, 1.4148], device='cuda:1'), covar=tensor([0.0782, 0.0623, 0.1115, 0.2166], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0627, 0.0797, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 22:45:09,571 INFO [train.py:968] (1/2) Epoch 3, batch 41150, giga_loss[loss=0.36, simple_loss=0.4082, pruned_loss=0.1559, over 28946.00 frames. ], tot_loss[loss=0.3757, simple_loss=0.4199, pruned_loss=0.1657, over 5672227.55 frames. ], libri_tot_loss[loss=0.3779, simple_loss=0.4201, pruned_loss=0.1678, over 5691538.86 frames. ], giga_tot_loss[loss=0.3729, simple_loss=0.4178, pruned_loss=0.164, over 5663264.65 frames. ], batch size: 164, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:45:27,851 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:30,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:38,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-01 22:45:43,875 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=131476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:47,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=131479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:45:58,094 INFO [train.py:968] (1/2) Epoch 3, batch 41200, giga_loss[loss=0.4933, simple_loss=0.4746, pruned_loss=0.256, over 23510.00 frames. ], tot_loss[loss=0.3813, simple_loss=0.4225, pruned_loss=0.1701, over 5643533.23 frames. ], libri_tot_loss[loss=0.3767, simple_loss=0.4191, pruned_loss=0.1672, over 5697712.06 frames. ], giga_tot_loss[loss=0.3803, simple_loss=0.4219, pruned_loss=0.1693, over 5628977.56 frames. ], batch size: 705, lr: 9.08e-03, grad_scale: 8.0
+2023-03-01 22:45:59,150 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=131491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:46:17,169 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:46:39,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8182, 1.7905, 1.4586, 1.5872], device='cuda:1'), covar=tensor([0.0465, 0.0393, 0.0711, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0477, 0.0515, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 22:46:46,578 INFO [train.py:968] (1/2) Epoch 3, batch 41250, giga_loss[loss=0.3516, simple_loss=0.4026, pruned_loss=0.1503, over 29013.00 frames. ], tot_loss[loss=0.3847, simple_loss=0.4244, pruned_loss=0.1725, over 5622661.99 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.4185, pruned_loss=0.1668, over 5684502.15 frames. ], giga_tot_loss[loss=0.3848, simple_loss=0.4248, pruned_loss=0.1724, over 5620718.67 frames. ], batch size: 213, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:47:11,870 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:47:27,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-01 22:47:39,921 INFO [train.py:968] (1/2) Epoch 3, batch 41300, giga_loss[loss=0.3319, simple_loss=0.3896, pruned_loss=0.1371, over 28689.00 frames. ], tot_loss[loss=0.3913, simple_loss=0.4295, pruned_loss=0.1766, over 5631969.83 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4185, pruned_loss=0.1667, over 5689104.26 frames. ], giga_tot_loss[loss=0.3917, simple_loss=0.4299, pruned_loss=0.1768, over 5625478.37 frames. ], batch size: 262, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:48:11,160 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 41350, giga_loss[loss=0.367, simple_loss=0.4179, pruned_loss=0.1581, over 28970.00 frames. ], tot_loss[loss=0.3929, simple_loss=0.4307, pruned_loss=0.1776, over 5637082.59 frames. ], libri_tot_loss[loss=0.3761, simple_loss=0.4187, pruned_loss=0.1667, over 5691653.51 frames. ], giga_tot_loss[loss=0.3933, simple_loss=0.431, pruned_loss=0.1778, over 5629059.32 frames. ], batch size: 164, lr: 9.08e-03, grad_scale: 4.0
+2023-03-01 22:49:24,622 INFO [train.py:968] (1/2) Epoch 3, batch 41400, giga_loss[loss=0.35, simple_loss=0.3966, pruned_loss=0.1517, over 28841.00 frames. ], tot_loss[loss=0.3909, simple_loss=0.4284, pruned_loss=0.1767, over 5636703.48 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4186, pruned_loss=0.1666, over 5693855.48 frames. ], giga_tot_loss[loss=0.3914, simple_loss=0.4287, pruned_loss=0.177, over 5627885.84 frames. ], batch size: 99, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:49:55,551 INFO [optim.py:369] (1/2) 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 22:50:13,491 INFO [train.py:968] (1/2) Epoch 3, batch 41450, giga_loss[loss=0.3596, simple_loss=0.4147, pruned_loss=0.1522, over 28882.00 frames. ], tot_loss[loss=0.3924, simple_loss=0.4291, pruned_loss=0.1779, over 5635720.28 frames. ], libri_tot_loss[loss=0.376, simple_loss=0.4185, pruned_loss=0.1668, over 5700905.92 frames. ], giga_tot_loss[loss=0.3933, simple_loss=0.4298, pruned_loss=0.1784, over 5620134.87 frames. ], batch size: 145, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:50:48,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9379, 1.2392, 3.7569, 3.0733], device='cuda:1'), covar=tensor([0.1629, 0.2061, 0.0379, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0514, 0.0719, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 22:51:02,973 INFO [train.py:968] (1/2) Epoch 3, batch 41500, giga_loss[loss=0.3961, simple_loss=0.439, pruned_loss=0.1767, over 28599.00 frames. ], tot_loss[loss=0.3881, simple_loss=0.4271, pruned_loss=0.1745, over 5621627.53 frames. ], libri_tot_loss[loss=0.3761, simple_loss=0.4186, pruned_loss=0.1668, over 5693633.63 frames. ], giga_tot_loss[loss=0.3889, simple_loss=0.4277, pruned_loss=0.175, over 5614635.97 frames. ], batch size: 336, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:51:18,975 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:51:26,341 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:51:30,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4375, 1.6953, 1.6412, 1.5918], device='cuda:1'), covar=tensor([0.1263, 0.1811, 0.1064, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0785, 0.0746, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 22:51:38,433 INFO [optim.py:369] (1/2) 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 22:52:01,372 INFO [train.py:968] (1/2) Epoch 3, batch 41550, giga_loss[loss=0.4069, simple_loss=0.4139, pruned_loss=0.2, over 23595.00 frames. ], tot_loss[loss=0.3921, simple_loss=0.4298, pruned_loss=0.1772, over 5594082.19 frames. ], libri_tot_loss[loss=0.3761, simple_loss=0.4186, pruned_loss=0.1668, over 5693633.63 frames. ], giga_tot_loss[loss=0.3927, simple_loss=0.4303, pruned_loss=0.1776, over 5588640.56 frames. ], batch size: 705, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:52:03,530 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 22:52:49,291 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=131885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:52:49,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-01 22:52:53,290 INFO [train.py:968] (1/2) Epoch 3, batch 41600, giga_loss[loss=0.3635, simple_loss=0.4094, pruned_loss=0.1588, over 28769.00 frames. ], tot_loss[loss=0.389, simple_loss=0.4278, pruned_loss=0.1751, over 5608449.36 frames. ], libri_tot_loss[loss=0.3762, simple_loss=0.4188, pruned_loss=0.1668, over 5697611.95 frames. ], giga_tot_loss[loss=0.3897, simple_loss=0.4283, pruned_loss=0.1756, over 5598996.92 frames. ], batch size: 92, lr: 9.07e-03, grad_scale: 8.0
+2023-03-01 22:53:24,979 INFO [optim.py:369] (1/2) 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 22:53:42,483 INFO [train.py:968] (1/2) Epoch 3, batch 41650, giga_loss[loss=0.3328, simple_loss=0.3932, pruned_loss=0.1361, over 28553.00 frames. ], tot_loss[loss=0.3825, simple_loss=0.4245, pruned_loss=0.1703, over 5624111.50 frames. ], libri_tot_loss[loss=0.3763, simple_loss=0.4188, pruned_loss=0.1669, over 5698221.75 frames. ], giga_tot_loss[loss=0.3831, simple_loss=0.4249, pruned_loss=0.1707, over 5614611.26 frames. ], batch size: 71, lr: 9.07e-03, grad_scale: 4.0
+2023-03-01 22:53:43,501 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=131942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:53:52,780 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 22:54:30,892 INFO [train.py:968] (1/2) Epoch 3, batch 41700, giga_loss[loss=0.3696, simple_loss=0.4159, pruned_loss=0.1617, over 28608.00 frames. ], tot_loss[loss=0.3786, simple_loss=0.4218, pruned_loss=0.1677, over 5628533.67 frames. ], libri_tot_loss[loss=0.3759, simple_loss=0.4186, pruned_loss=0.1666, over 5697794.08 frames. ], giga_tot_loss[loss=0.3795, simple_loss=0.4225, pruned_loss=0.1683, over 5619791.39 frames. ], batch size: 336, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:55:04,204 INFO [optim.py:369] (1/2) 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 22:55:21,911 INFO [train.py:968] (1/2) Epoch 3, batch 41750, giga_loss[loss=0.4339, simple_loss=0.4342, pruned_loss=0.2168, over 23715.00 frames. ], tot_loss[loss=0.3741, simple_loss=0.4184, pruned_loss=0.1649, over 5622675.23 frames. ], libri_tot_loss[loss=0.3749, simple_loss=0.4178, pruned_loss=0.166, over 5701998.63 frames. ], giga_tot_loss[loss=0.3758, simple_loss=0.4196, pruned_loss=0.166, over 5610402.98 frames. ], batch size: 705, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:55:44,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5037, 1.5471, 5.2142, 3.6317], device='cuda:1'), covar=tensor([0.1471, 0.1770, 0.0260, 0.0462], device='cuda:1'), in_proj_covar=tensor([0.0542, 0.0511, 0.0707, 0.0562], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 22:55:59,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:56:02,266 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 3, batch 41800, giga_loss[loss=0.3175, simple_loss=0.3795, pruned_loss=0.1278, over 28517.00 frames. ], tot_loss[loss=0.3699, simple_loss=0.4151, pruned_loss=0.1624, over 5635203.85 frames. ], libri_tot_loss[loss=0.3737, simple_loss=0.4165, pruned_loss=0.1654, over 5699458.18 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4174, pruned_loss=0.1637, over 5623341.00 frames. ], batch size: 60, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:56:32,401 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:56:38,524 INFO [optim.py:369] (1/2) 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 22:56:53,812 INFO [train.py:968] (1/2) Epoch 3, batch 41850, giga_loss[loss=0.3565, simple_loss=0.4088, pruned_loss=0.1521, over 28941.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.4141, pruned_loss=0.1614, over 5640330.40 frames. ], libri_tot_loss[loss=0.3734, simple_loss=0.4162, pruned_loss=0.1653, over 5698987.87 frames. ], giga_tot_loss[loss=0.3706, simple_loss=0.4162, pruned_loss=0.1625, over 5629854.31 frames. ], batch size: 227, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:57:26,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:57:34,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:57:40,279 INFO [train.py:968] (1/2) Epoch 3, batch 41900, giga_loss[loss=0.4378, simple_loss=0.4318, pruned_loss=0.2219, over 23437.00 frames. ], tot_loss[loss=0.3672, simple_loss=0.4134, pruned_loss=0.1605, over 5642593.41 frames. ], libri_tot_loss[loss=0.3737, simple_loss=0.4164, pruned_loss=0.1655, over 5698104.68 frames. ], giga_tot_loss[loss=0.3685, simple_loss=0.4148, pruned_loss=0.1611, over 5634572.48 frames. ], batch size: 705, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:57:56,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7577, 2.7492, 1.8078, 0.9355], device='cuda:1'), covar=tensor([0.2784, 0.1234, 0.1730, 0.2770], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.1242, 0.1294, 0.1116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 22:58:10,465 INFO [optim.py:369] (1/2) 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 22:58:30,767 INFO [train.py:968] (1/2) Epoch 3, batch 41950, giga_loss[loss=0.3059, simple_loss=0.375, pruned_loss=0.1184, over 28636.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4115, pruned_loss=0.1585, over 5650653.05 frames. ], libri_tot_loss[loss=0.3737, simple_loss=0.4165, pruned_loss=0.1655, over 5707658.95 frames. ], giga_tot_loss[loss=0.365, simple_loss=0.4124, pruned_loss=0.1587, over 5632785.43 frames. ], batch size: 242, lr: 9.06e-03, grad_scale: 4.0
+2023-03-01 22:58:38,130 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:58:48,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 22:59:20,300 INFO [train.py:968] (1/2) Epoch 3, batch 42000, giga_loss[loss=0.3475, simple_loss=0.4084, pruned_loss=0.1433, over 27978.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4098, pruned_loss=0.1545, over 5658489.34 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4152, pruned_loss=0.1648, over 5713866.57 frames. ], giga_tot_loss[loss=0.3609, simple_loss=0.4115, pruned_loss=0.1551, over 5637098.85 frames. ], batch size: 412, lr: 9.05e-03, grad_scale: 8.0
+2023-03-01 22:59:20,300 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 22:59:25,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2675, 1.7538, 1.2684, 0.4881], device='cuda:1'), covar=tensor([0.1475, 0.1095, 0.2000, 0.1867], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.1235, 0.1292, 0.1113], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 22:59:30,612 INFO [train.py:1012] (1/2) Epoch 3, validation: loss=0.2463, simple_loss=0.3444, pruned_loss=0.07409, over 944034.00 frames. 
+2023-03-01 22:59:30,613 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 23:00:02,192 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:04,442 INFO [optim.py:369] (1/2) 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 23:00:06,683 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:08,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:19,554 INFO [train.py:968] (1/2) Epoch 3, batch 42050, giga_loss[loss=0.3867, simple_loss=0.4315, pruned_loss=0.1709, over 28880.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4116, pruned_loss=0.1541, over 5668677.09 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4151, pruned_loss=0.1647, over 5716494.24 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.413, pruned_loss=0.1545, over 5648538.10 frames. ], batch size: 174, lr: 9.05e-03, grad_scale: 8.0
+2023-03-01 23:00:31,899 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:37,341 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:00:42,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-01 23:00:55,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2321, 3.3294, 2.1800, 2.1356], device='cuda:1'), covar=tensor([0.0520, 0.0368, 0.0561, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0456, 0.0308, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-01 23:01:06,845 INFO [train.py:968] (1/2) Epoch 3, batch 42100, libri_loss[loss=0.3415, simple_loss=0.3834, pruned_loss=0.1499, over 29376.00 frames. ], tot_loss[loss=0.3621, simple_loss=0.4129, pruned_loss=0.1556, over 5667638.02 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.4153, pruned_loss=0.1651, over 5712321.55 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4138, pruned_loss=0.1553, over 5653880.30 frames. ], batch size: 67, lr: 9.05e-03, grad_scale: 8.0
+2023-03-01 23:01:19,235 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:22,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:38,665 INFO [optim.py:369] (1/2) 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 23:01:40,357 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:51,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:01:56,442 INFO [train.py:968] (1/2) Epoch 3, batch 42150, giga_loss[loss=0.4425, simple_loss=0.4634, pruned_loss=0.2107, over 28480.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4148, pruned_loss=0.1579, over 5661955.58 frames. ], libri_tot_loss[loss=0.3731, simple_loss=0.4156, pruned_loss=0.1653, over 5713322.25 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4151, pruned_loss=0.1574, over 5649289.64 frames. ], batch size: 336, lr: 9.05e-03, grad_scale: 4.0
+2023-03-01 23:02:01,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=132447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:02:19,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7803, 1.6210, 1.5723, 1.6113], device='cuda:1'), covar=tensor([0.0841, 0.1538, 0.1350, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0763, 0.0626, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 23:02:40,464 INFO [train.py:968] (1/2) Epoch 3, batch 42200, giga_loss[loss=0.3857, simple_loss=0.4282, pruned_loss=0.1716, over 28968.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4119, pruned_loss=0.1566, over 5676898.51 frames. ], libri_tot_loss[loss=0.3725, simple_loss=0.4152, pruned_loss=0.1649, over 5713735.10 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4126, pruned_loss=0.1563, over 5665542.62 frames. ], batch size: 155, lr: 9.05e-03, grad_scale: 4.0
+2023-03-01 23:03:12,356 INFO [optim.py:369] (1/2) 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 23:03:27,070 INFO [train.py:968] (1/2) Epoch 3, batch 42250, giga_loss[loss=0.3959, simple_loss=0.4269, pruned_loss=0.1824, over 27565.00 frames. ], tot_loss[loss=0.3642, simple_loss=0.4119, pruned_loss=0.1583, over 5671520.08 frames. ], libri_tot_loss[loss=0.3721, simple_loss=0.415, pruned_loss=0.1646, over 5716759.20 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4125, pruned_loss=0.1582, over 5658869.82 frames. ], batch size: 472, lr: 9.05e-03, grad_scale: 2.0
+2023-03-01 23:04:16,093 INFO [train.py:968] (1/2) Epoch 3, batch 42300, giga_loss[loss=0.3321, simple_loss=0.3942, pruned_loss=0.135, over 28793.00 frames. ], tot_loss[loss=0.3624, simple_loss=0.4112, pruned_loss=0.1568, over 5666627.07 frames. ], libri_tot_loss[loss=0.373, simple_loss=0.4158, pruned_loss=0.1651, over 5712605.80 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.411, pruned_loss=0.1561, over 5658857.21 frames. ], batch size: 99, lr: 9.04e-03, grad_scale: 2.0
+2023-03-01 23:04:20,118 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-01 23:04:48,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:04:48,864 INFO [optim.py:369] (1/2) 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 23:04:59,862 INFO [train.py:968] (1/2) Epoch 3, batch 42350, giga_loss[loss=0.3359, simple_loss=0.4016, pruned_loss=0.1351, over 28771.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4107, pruned_loss=0.1551, over 5668327.76 frames. ], libri_tot_loss[loss=0.3729, simple_loss=0.4156, pruned_loss=0.1651, over 5706396.07 frames. ], giga_tot_loss[loss=0.3594, simple_loss=0.4104, pruned_loss=0.1542, over 5666780.02 frames. ], batch size: 119, lr: 9.04e-03, grad_scale: 2.0
+2023-03-01 23:05:13,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6883, 3.0523, 1.6056, 1.6578], device='cuda:1'), covar=tensor([0.0714, 0.0390, 0.0758, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0458, 0.0309, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-01 23:05:47,090 INFO [train.py:968] (1/2) Epoch 3, batch 42400, giga_loss[loss=0.3565, simple_loss=0.4118, pruned_loss=0.1507, over 28684.00 frames. ], tot_loss[loss=0.3607, simple_loss=0.4108, pruned_loss=0.1553, over 5667016.32 frames. ], libri_tot_loss[loss=0.3727, simple_loss=0.4156, pruned_loss=0.1649, over 5706198.24 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4106, pruned_loss=0.1545, over 5665063.63 frames. ], batch size: 307, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:05:48,613 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-01 23:06:18,796 INFO [optim.py:369] (1/2) 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 23:06:26,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7308, 1.4989, 1.3391, 1.2368], device='cuda:1'), covar=tensor([0.0504, 0.0475, 0.0856, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0483, 0.0511, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-01 23:06:32,541 INFO [train.py:968] (1/2) Epoch 3, batch 42450, giga_loss[loss=0.3472, simple_loss=0.3965, pruned_loss=0.1489, over 28990.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.4092, pruned_loss=0.1541, over 5668762.06 frames. ], libri_tot_loss[loss=0.3717, simple_loss=0.4149, pruned_loss=0.1643, over 5708057.60 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4093, pruned_loss=0.1536, over 5664388.39 frames. ], batch size: 106, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:06:43,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1359, 1.2658, 1.0943, 0.9353], device='cuda:1'), covar=tensor([0.1883, 0.1813, 0.1759, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.1041, 0.0843, 0.0933, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 23:06:53,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2433, 2.9473, 3.0018, 1.3094], device='cuda:1'), covar=tensor([0.0715, 0.0608, 0.1060, 0.2179], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0653, 0.0824, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:06:53,399 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:06:58,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:16,319 INFO [train.py:968] (1/2) Epoch 3, batch 42500, giga_loss[loss=0.3143, simple_loss=0.3791, pruned_loss=0.1247, over 29104.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4081, pruned_loss=0.1541, over 5682315.88 frames. ], libri_tot_loss[loss=0.3722, simple_loss=0.4152, pruned_loss=0.1646, over 5713647.57 frames. ], giga_tot_loss[loss=0.3571, simple_loss=0.4079, pruned_loss=0.1532, over 5672918.48 frames. ], batch size: 128, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:07:17,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2336, 1.5460, 1.1896, 1.3902], device='cuda:1'), covar=tensor([0.0785, 0.0421, 0.0379, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0142, 0.0146, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0045, 0.0033, 0.0030, 0.0050], device='cuda:1')
+2023-03-01 23:07:17,435 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-01 23:07:23,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:24,811 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:44,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=132822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:07:47,047 INFO [optim.py:369] (1/2) 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 23:08:01,351 INFO [train.py:968] (1/2) Epoch 3, batch 42550, giga_loss[loss=0.3511, simple_loss=0.3989, pruned_loss=0.1517, over 28716.00 frames. ], tot_loss[loss=0.3573, simple_loss=0.4069, pruned_loss=0.1539, over 5665035.48 frames. ], libri_tot_loss[loss=0.3721, simple_loss=0.4152, pruned_loss=0.1645, over 5704724.95 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4066, pruned_loss=0.1531, over 5664895.74 frames. ], batch size: 242, lr: 9.04e-03, grad_scale: 4.0
+2023-03-01 23:08:14,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5873, 1.4558, 1.1600, 1.2353], device='cuda:1'), covar=tensor([0.0517, 0.0485, 0.0885, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0486, 0.0517, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 23:08:49,812 INFO [train.py:968] (1/2) Epoch 3, batch 42600, giga_loss[loss=0.3475, simple_loss=0.4081, pruned_loss=0.1434, over 28829.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4059, pruned_loss=0.1535, over 5678559.15 frames. ], libri_tot_loss[loss=0.3721, simple_loss=0.4153, pruned_loss=0.1645, over 5707231.13 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4054, pruned_loss=0.1527, over 5675439.35 frames. ], batch size: 174, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:09:03,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3813, 1.9591, 1.4827, 1.5564], device='cuda:1'), covar=tensor([0.0805, 0.0302, 0.0346, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0143, 0.0147, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0045, 0.0034, 0.0030, 0.0050], device='cuda:1')
+2023-03-01 23:09:23,390 INFO [optim.py:369] (1/2) 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 23:09:39,351 INFO [train.py:968] (1/2) Epoch 3, batch 42650, giga_loss[loss=0.3875, simple_loss=0.4283, pruned_loss=0.1734, over 28539.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4047, pruned_loss=0.1532, over 5662894.01 frames. ], libri_tot_loss[loss=0.3724, simple_loss=0.4157, pruned_loss=0.1646, over 5692025.21 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4038, pruned_loss=0.1522, over 5673412.45 frames. ], batch size: 336, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:09:40,525 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:09:44,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:03,537 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=132965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:06,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=132968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:10,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:26,380 INFO [train.py:968] (1/2) Epoch 3, batch 42700, giga_loss[loss=0.3328, simple_loss=0.3935, pruned_loss=0.1361, over 29026.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4036, pruned_loss=0.1531, over 5658499.31 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.415, pruned_loss=0.164, over 5698519.24 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.403, pruned_loss=0.1524, over 5660004.26 frames. ], batch size: 155, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:10:34,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=132997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:10:51,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6317, 1.1743, 2.7961, 2.6553], device='cuda:1'), covar=tensor([0.1522, 0.1833, 0.0559, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0512, 0.0717, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 23:11:00,226 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 42750, giga_loss[loss=0.3604, simple_loss=0.4109, pruned_loss=0.155, over 28685.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4018, pruned_loss=0.1517, over 5659507.73 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4138, pruned_loss=0.163, over 5704760.41 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.402, pruned_loss=0.1517, over 5653534.42 frames. ], batch size: 262, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:11:26,644 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:12:04,175 INFO [train.py:968] (1/2) Epoch 3, batch 42800, giga_loss[loss=0.3625, simple_loss=0.4184, pruned_loss=0.1533, over 28919.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4029, pruned_loss=0.1519, over 5656973.98 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4138, pruned_loss=0.1631, over 5696414.61 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.403, pruned_loss=0.1518, over 5660361.61 frames. ], batch size: 199, lr: 9.03e-03, grad_scale: 8.0
+2023-03-01 23:12:31,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3456, 1.3468, 1.1880, 1.4879], device='cuda:1'), covar=tensor([0.2102, 0.2095, 0.1978, 0.2182], device='cuda:1'), in_proj_covar=tensor([0.1043, 0.0836, 0.0932, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 23:12:36,777 INFO [optim.py:369] (1/2) 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 23:12:47,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:12:48,948 INFO [train.py:968] (1/2) Epoch 3, batch 42850, giga_loss[loss=0.4074, simple_loss=0.4364, pruned_loss=0.1892, over 28544.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4029, pruned_loss=0.1508, over 5655818.57 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4135, pruned_loss=0.1629, over 5690504.15 frames. ], giga_tot_loss[loss=0.352, simple_loss=0.4029, pruned_loss=0.1506, over 5663880.88 frames. ], batch size: 336, lr: 9.03e-03, grad_scale: 4.0
+2023-03-01 23:13:32,578 INFO [train.py:968] (1/2) Epoch 3, batch 42900, giga_loss[loss=0.3432, simple_loss=0.3992, pruned_loss=0.1436, over 28827.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4034, pruned_loss=0.1504, over 5672951.97 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4138, pruned_loss=0.163, over 5697802.48 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4028, pruned_loss=0.1498, over 5672136.84 frames. ], batch size: 199, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:14:11,004 INFO [optim.py:369] (1/2) 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 23:14:15,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5076, 1.6763, 1.2343, 0.7817], device='cuda:1'), covar=tensor([0.0939, 0.0565, 0.0516, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.1269, 0.1001, 0.1015, 0.1085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 23:14:26,430 INFO [train.py:968] (1/2) Epoch 3, batch 42950, giga_loss[loss=0.3825, simple_loss=0.4299, pruned_loss=0.1676, over 28902.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4048, pruned_loss=0.1519, over 5678456.89 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4135, pruned_loss=0.1626, over 5701165.51 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4045, pruned_loss=0.1515, over 5674514.05 frames. ], batch size: 145, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:14:30,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0910, 1.6984, 1.3575, 1.4253], device='cuda:1'), covar=tensor([0.0603, 0.0766, 0.0928, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0485, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 23:14:30,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1121, 1.7602, 1.7414, 1.6432], device='cuda:1'), covar=tensor([0.0961, 0.2002, 0.1373, 0.1492], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0773, 0.0631, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 23:15:15,513 INFO [train.py:968] (1/2) Epoch 3, batch 43000, giga_loss[loss=0.3811, simple_loss=0.4314, pruned_loss=0.1654, over 28862.00 frames. ], tot_loss[loss=0.3592, simple_loss=0.408, pruned_loss=0.1552, over 5686254.08 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4132, pruned_loss=0.1622, over 5703064.14 frames. ], giga_tot_loss[loss=0.359, simple_loss=0.4079, pruned_loss=0.1551, over 5681078.10 frames. ], batch size: 174, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:15:54,881 INFO [optim.py:369] (1/2) 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 23:16:08,922 INFO [train.py:968] (1/2) Epoch 3, batch 43050, giga_loss[loss=0.39, simple_loss=0.4241, pruned_loss=0.178, over 27981.00 frames. ], tot_loss[loss=0.3641, simple_loss=0.4102, pruned_loss=0.159, over 5675177.34 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4132, pruned_loss=0.1622, over 5702400.61 frames. ], giga_tot_loss[loss=0.3639, simple_loss=0.4101, pruned_loss=0.1588, over 5671592.83 frames. ], batch size: 412, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:16:59,846 INFO [train.py:968] (1/2) Epoch 3, batch 43100, libri_loss[loss=0.4519, simple_loss=0.4663, pruned_loss=0.2188, over 19902.00 frames. ], tot_loss[loss=0.3684, simple_loss=0.4126, pruned_loss=0.1621, over 5665101.88 frames. ], libri_tot_loss[loss=0.3687, simple_loss=0.4129, pruned_loss=0.1622, over 5696060.44 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.4126, pruned_loss=0.1619, over 5668335.17 frames. ], batch size: 187, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:17:31,756 INFO [optim.py:369] (1/2) 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 23:17:33,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:17:39,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6903, 3.4452, 3.3966, 1.6873], device='cuda:1'), covar=tensor([0.0708, 0.0574, 0.1048, 0.2285], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0652, 0.0829, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:17:43,926 INFO [train.py:968] (1/2) Epoch 3, batch 43150, giga_loss[loss=0.3215, simple_loss=0.3744, pruned_loss=0.1343, over 28850.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4111, pruned_loss=0.1609, over 5655473.46 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4124, pruned_loss=0.1618, over 5690187.28 frames. ], giga_tot_loss[loss=0.3669, simple_loss=0.4117, pruned_loss=0.1611, over 5661940.28 frames. ], batch size: 112, lr: 9.02e-03, grad_scale: 4.0
+2023-03-01 23:17:51,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7766, 2.7261, 1.8999, 0.7659], device='cuda:1'), covar=tensor([0.2835, 0.1262, 0.1600, 0.2982], device='cuda:1'), in_proj_covar=tensor([0.1304, 0.1273, 0.1305, 0.1128], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 23:18:29,029 INFO [train.py:968] (1/2) Epoch 3, batch 43200, giga_loss[loss=0.314, simple_loss=0.3802, pruned_loss=0.1239, over 28508.00 frames. ], tot_loss[loss=0.3655, simple_loss=0.4102, pruned_loss=0.1604, over 5655934.88 frames. ], libri_tot_loss[loss=0.3676, simple_loss=0.4121, pruned_loss=0.1616, over 5691625.07 frames. ], giga_tot_loss[loss=0.3662, simple_loss=0.4109, pruned_loss=0.1608, over 5659167.63 frames. ], batch size: 71, lr: 9.01e-03, grad_scale: 8.0
+2023-03-01 23:18:53,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=133514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:19:01,287 INFO [optim.py:369] (1/2) 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 23:19:17,991 INFO [train.py:968] (1/2) Epoch 3, batch 43250, libri_loss[loss=0.4507, simple_loss=0.4639, pruned_loss=0.2187, over 19573.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4099, pruned_loss=0.1587, over 5646496.71 frames. ], libri_tot_loss[loss=0.3679, simple_loss=0.4122, pruned_loss=0.1618, over 5681699.05 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4102, pruned_loss=0.1587, over 5658545.10 frames. ], batch size: 188, lr: 9.01e-03, grad_scale: 8.0
+2023-03-01 23:19:42,938 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:19:45,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:20:00,980 INFO [train.py:968] (1/2) Epoch 3, batch 43300, libri_loss[loss=0.4105, simple_loss=0.4458, pruned_loss=0.1876, over 28576.00 frames. ], tot_loss[loss=0.3595, simple_loss=0.4071, pruned_loss=0.1559, over 5653144.59 frames. ], libri_tot_loss[loss=0.3683, simple_loss=0.4126, pruned_loss=0.1621, over 5687292.29 frames. ], giga_tot_loss[loss=0.359, simple_loss=0.4069, pruned_loss=0.1556, over 5656677.21 frames. ], batch size: 106, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:20:12,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:20:34,971 INFO [optim.py:369] (1/2) 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 23:20:45,133 INFO [train.py:968] (1/2) Epoch 3, batch 43350, giga_loss[loss=0.3593, simple_loss=0.404, pruned_loss=0.1574, over 29035.00 frames. ], tot_loss[loss=0.357, simple_loss=0.405, pruned_loss=0.1545, over 5662440.73 frames. ], libri_tot_loss[loss=0.3682, simple_loss=0.4125, pruned_loss=0.1619, over 5686286.36 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4047, pruned_loss=0.1541, over 5665143.30 frames. ], batch size: 164, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:20:59,828 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=133657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:21:01,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=133660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:21:15,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3254, 1.4609, 1.2062, 1.6514], device='cuda:1'), covar=tensor([0.2070, 0.2075, 0.1936, 0.1988], device='cuda:1'), in_proj_covar=tensor([0.1048, 0.0850, 0.0943, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 23:21:30,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=133689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:21:31,513 INFO [train.py:968] (1/2) Epoch 3, batch 43400, giga_loss[loss=0.3487, simple_loss=0.3956, pruned_loss=0.151, over 29057.00 frames. ], tot_loss[loss=0.3571, simple_loss=0.4041, pruned_loss=0.155, over 5661555.85 frames. ], libri_tot_loss[loss=0.3682, simple_loss=0.4124, pruned_loss=0.1619, over 5690990.19 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4037, pruned_loss=0.1545, over 5659184.67 frames. ], batch size: 136, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:22:05,916 INFO [optim.py:369] (1/2) 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 23:22:16,393 INFO [train.py:968] (1/2) Epoch 3, batch 43450, giga_loss[loss=0.3705, simple_loss=0.4187, pruned_loss=0.1612, over 27926.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.4056, pruned_loss=0.1562, over 5662688.08 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4129, pruned_loss=0.1625, over 5686330.65 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.4046, pruned_loss=0.1551, over 5663907.94 frames. ], batch size: 412, lr: 9.01e-03, grad_scale: 4.0
+2023-03-01 23:22:40,550 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133767.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 23:22:44,514 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:23:03,290 INFO [train.py:968] (1/2) Epoch 3, batch 43500, giga_loss[loss=0.4055, simple_loss=0.4214, pruned_loss=0.1948, over 23674.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4094, pruned_loss=0.1578, over 5660497.74 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4126, pruned_loss=0.1621, over 5687843.01 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4088, pruned_loss=0.1572, over 5659838.77 frames. ], batch size: 705, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:23:40,260 INFO [optim.py:369] (1/2) 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 23:23:50,420 INFO [train.py:968] (1/2) Epoch 3, batch 43550, giga_loss[loss=0.352, simple_loss=0.4148, pruned_loss=0.1446, over 28918.00 frames. ], tot_loss[loss=0.3616, simple_loss=0.412, pruned_loss=0.1556, over 5663663.28 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4129, pruned_loss=0.1623, over 5681055.13 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4113, pruned_loss=0.1549, over 5669019.57 frames. ], batch size: 227, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:23:55,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4739, 4.1984, 4.1369, 2.0418], device='cuda:1'), covar=tensor([0.0426, 0.0420, 0.0759, 0.1806], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0655, 0.0824, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:24:20,948 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=133867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:24:43,270 INFO [train.py:968] (1/2) Epoch 3, batch 43600, giga_loss[loss=0.3883, simple_loss=0.4385, pruned_loss=0.1691, over 28004.00 frames. ], tot_loss[loss=0.3633, simple_loss=0.4136, pruned_loss=0.1565, over 5656734.25 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4133, pruned_loss=0.1626, over 5680884.47 frames. ], giga_tot_loss[loss=0.362, simple_loss=0.4127, pruned_loss=0.1556, over 5660840.55 frames. ], batch size: 412, lr: 9.00e-03, grad_scale: 8.0
+2023-03-01 23:25:20,832 INFO [optim.py:369] (1/2) 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 23:25:33,107 INFO [train.py:968] (1/2) Epoch 3, batch 43650, giga_loss[loss=0.3764, simple_loss=0.4215, pruned_loss=0.1656, over 28765.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4162, pruned_loss=0.159, over 5653672.59 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4137, pruned_loss=0.163, over 5679231.95 frames. ], giga_tot_loss[loss=0.3655, simple_loss=0.4151, pruned_loss=0.1579, over 5658143.46 frames. ], batch size: 119, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:25:59,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6574, 1.8862, 1.7024, 1.7437], device='cuda:1'), covar=tensor([0.1133, 0.1416, 0.0950, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0788, 0.0752, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 23:26:22,759 INFO [train.py:968] (1/2) Epoch 3, batch 43700, giga_loss[loss=0.3965, simple_loss=0.4409, pruned_loss=0.176, over 28903.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4163, pruned_loss=0.1598, over 5657793.59 frames. ], libri_tot_loss[loss=0.3696, simple_loss=0.4137, pruned_loss=0.1628, over 5683638.06 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4156, pruned_loss=0.159, over 5657152.90 frames. ], batch size: 145, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:26:54,333 INFO [optim.py:369] (1/2) 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 23:27:05,491 INFO [train.py:968] (1/2) Epoch 3, batch 43750, giga_loss[loss=0.3806, simple_loss=0.4267, pruned_loss=0.1673, over 28620.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4157, pruned_loss=0.1602, over 5670483.43 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4141, pruned_loss=0.1629, over 5687245.68 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4148, pruned_loss=0.1594, over 5665655.28 frames. ], batch size: 336, lr: 9.00e-03, grad_scale: 4.0
+2023-03-01 23:27:12,267 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:27:54,234 INFO [train.py:968] (1/2) Epoch 3, batch 43800, giga_loss[loss=0.3215, simple_loss=0.3753, pruned_loss=0.1338, over 28746.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4127, pruned_loss=0.1588, over 5665048.53 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4135, pruned_loss=0.1625, over 5692224.78 frames. ], giga_tot_loss[loss=0.3647, simple_loss=0.4124, pruned_loss=0.1584, over 5656284.10 frames. ], batch size: 66, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:28:23,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5636, 4.2727, 4.3375, 1.7859], device='cuda:1'), covar=tensor([0.0451, 0.0385, 0.0794, 0.2093], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0657, 0.0831, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:28:32,233 INFO [optim.py:369] (1/2) 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 23:28:37,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2010, 1.4935, 1.2181, 0.5237], device='cuda:1'), covar=tensor([0.0697, 0.0582, 0.0713, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.1283, 0.1239, 0.1309, 0.1110], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 23:28:44,223 INFO [train.py:968] (1/2) Epoch 3, batch 43850, giga_loss[loss=0.3469, simple_loss=0.3945, pruned_loss=0.1497, over 28621.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4115, pruned_loss=0.1586, over 5674067.43 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4136, pruned_loss=0.1624, over 5693520.24 frames. ], giga_tot_loss[loss=0.3639, simple_loss=0.4112, pruned_loss=0.1583, over 5665890.34 frames. ], batch size: 85, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:28:46,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134142.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 23:28:47,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:29:39,254 INFO [train.py:968] (1/2) Epoch 3, batch 43900, giga_loss[loss=0.3653, simple_loss=0.4131, pruned_loss=0.1587, over 28861.00 frames. ], tot_loss[loss=0.3641, simple_loss=0.4109, pruned_loss=0.1587, over 5674174.23 frames. ], libri_tot_loss[loss=0.3693, simple_loss=0.4137, pruned_loss=0.1625, over 5694411.64 frames. ], giga_tot_loss[loss=0.3637, simple_loss=0.4106, pruned_loss=0.1584, over 5666877.65 frames. ], batch size: 186, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:29:44,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6205, 1.0747, 3.3717, 2.8842], device='cuda:1'), covar=tensor([0.1836, 0.2205, 0.0451, 0.0625], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0517, 0.0716, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 23:29:56,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4324, 1.6932, 1.3631, 1.4934], device='cuda:1'), covar=tensor([0.0780, 0.0325, 0.0338, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0143, 0.0145, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0045, 0.0034, 0.0030, 0.0051], device='cuda:1')
+2023-03-01 23:30:20,931 INFO [optim.py:369] (1/2) 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 23:30:30,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9246, 1.1860, 4.2486, 3.3907], device='cuda:1'), covar=tensor([0.1714, 0.2179, 0.0332, 0.0649], device='cuda:1'), in_proj_covar=tensor([0.0546, 0.0517, 0.0720, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 23:30:31,204 INFO [train.py:968] (1/2) Epoch 3, batch 43950, giga_loss[loss=0.3537, simple_loss=0.3934, pruned_loss=0.157, over 28710.00 frames. ], tot_loss[loss=0.3664, simple_loss=0.4122, pruned_loss=0.1603, over 5670040.29 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4137, pruned_loss=0.1625, over 5695794.18 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4119, pruned_loss=0.16, over 5662551.32 frames. ], batch size: 119, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:30:32,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:30:52,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-01 23:31:11,594 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 23:31:13,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:31:14,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-01 23:31:17,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:31:17,321 INFO [train.py:968] (1/2) Epoch 3, batch 44000, giga_loss[loss=0.338, simple_loss=0.3961, pruned_loss=0.1399, over 28212.00 frames. ], tot_loss[loss=0.3636, simple_loss=0.4096, pruned_loss=0.1588, over 5675881.57 frames. ], libri_tot_loss[loss=0.3697, simple_loss=0.414, pruned_loss=0.1628, over 5700193.89 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.4091, pruned_loss=0.1583, over 5665847.85 frames. ], batch size: 368, lr: 8.99e-03, grad_scale: 8.0
+2023-03-01 23:31:43,376 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134317.0, num_to_drop=1, layers_to_drop={1}
+2023-03-01 23:31:44,772 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:31:45,684 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-01 23:31:54,010 INFO [optim.py:369] (1/2) 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 23:32:06,343 INFO [train.py:968] (1/2) Epoch 3, batch 44050, giga_loss[loss=0.3269, simple_loss=0.3924, pruned_loss=0.1307, over 28942.00 frames. ], tot_loss[loss=0.3605, simple_loss=0.4072, pruned_loss=0.1569, over 5677097.51 frames. ], libri_tot_loss[loss=0.3695, simple_loss=0.4138, pruned_loss=0.1626, over 5701355.04 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.407, pruned_loss=0.1566, over 5668032.06 frames. ], batch size: 186, lr: 8.99e-03, grad_scale: 4.0
+2023-03-01 23:32:24,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1795, 4.8538, 4.8901, 2.0705], device='cuda:1'), covar=tensor([0.0316, 0.0320, 0.0646, 0.1823], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0661, 0.0838, 0.0588], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:32:48,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:32:55,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:32:57,290 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:32:57,680 INFO [train.py:968] (1/2) Epoch 3, batch 44100, giga_loss[loss=0.3771, simple_loss=0.4299, pruned_loss=0.1621, over 28893.00 frames. ], tot_loss[loss=0.3611, simple_loss=0.4084, pruned_loss=0.1569, over 5665513.62 frames. ], libri_tot_loss[loss=0.3694, simple_loss=0.4137, pruned_loss=0.1626, over 5692484.96 frames. ], giga_tot_loss[loss=0.3607, simple_loss=0.4082, pruned_loss=0.1566, over 5665403.24 frames. ], batch size: 174, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:33:17,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-01 23:33:24,595 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:33:34,360 INFO [optim.py:369] (1/2) 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 23:33:43,374 INFO [train.py:968] (1/2) Epoch 3, batch 44150, libri_loss[loss=0.3368, simple_loss=0.3969, pruned_loss=0.1384, over 29561.00 frames. ], tot_loss[loss=0.3632, simple_loss=0.4103, pruned_loss=0.1581, over 5671249.92 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.4137, pruned_loss=0.1624, over 5696406.70 frames. ], giga_tot_loss[loss=0.3628, simple_loss=0.41, pruned_loss=0.1578, over 5666484.35 frames. ], batch size: 78, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:34:32,451 INFO [train.py:968] (1/2) Epoch 3, batch 44200, giga_loss[loss=0.4264, simple_loss=0.4476, pruned_loss=0.2026, over 28787.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4093, pruned_loss=0.1578, over 5672620.66 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4131, pruned_loss=0.162, over 5700190.83 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4094, pruned_loss=0.1579, over 5664619.94 frames. ], batch size: 284, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:35:09,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1307, 2.9329, 2.8082, 1.4705], device='cuda:1'), covar=tensor([0.0813, 0.0684, 0.1214, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0660, 0.0838, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:35:10,489 INFO [optim.py:369] (1/2) 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 23:35:14,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2556, 2.3963, 1.1914, 1.2625], device='cuda:1'), covar=tensor([0.0918, 0.0444, 0.0953, 0.1487], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0468, 0.0310, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0022, 0.0014, 0.0019], device='cuda:1')
+2023-03-01 23:35:22,272 INFO [train.py:968] (1/2) Epoch 3, batch 44250, giga_loss[loss=0.4342, simple_loss=0.4451, pruned_loss=0.2116, over 26746.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.411, pruned_loss=0.1571, over 5666898.36 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.4131, pruned_loss=0.1621, over 5702074.82 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4111, pruned_loss=0.157, over 5658742.37 frames. ], batch size: 555, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:35:22,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7999, 1.1872, 3.4313, 2.9029], device='cuda:1'), covar=tensor([0.1612, 0.2102, 0.0437, 0.0581], device='cuda:1'), in_proj_covar=tensor([0.0545, 0.0515, 0.0722, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-01 23:35:24,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-01 23:35:35,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6993, 3.5782, 1.5639, 1.4999], device='cuda:1'), covar=tensor([0.0851, 0.0333, 0.0943, 0.1390], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0466, 0.0311, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-01 23:35:42,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:35:46,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:36:01,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-01 23:36:05,456 INFO [train.py:968] (1/2) Epoch 3, batch 44300, giga_loss[loss=0.3988, simple_loss=0.4418, pruned_loss=0.1779, over 28499.00 frames. ], tot_loss[loss=0.3597, simple_loss=0.4113, pruned_loss=0.154, over 5687084.62 frames. ], libri_tot_loss[loss=0.3689, simple_loss=0.4135, pruned_loss=0.1622, over 5705827.88 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.411, pruned_loss=0.1537, over 5676479.05 frames. ], batch size: 336, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:36:08,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:36:36,187 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-01 23:36:41,003 INFO [optim.py:369] (1/2) 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 23:36:54,820 INFO [train.py:968] (1/2) Epoch 3, batch 44350, giga_loss[loss=0.4033, simple_loss=0.4347, pruned_loss=0.1859, over 28544.00 frames. ], tot_loss[loss=0.3627, simple_loss=0.4143, pruned_loss=0.1555, over 5689150.41 frames. ], libri_tot_loss[loss=0.3691, simple_loss=0.4135, pruned_loss=0.1623, over 5707631.25 frames. ], giga_tot_loss[loss=0.3618, simple_loss=0.414, pruned_loss=0.1549, over 5678165.33 frames. ], batch size: 336, lr: 8.98e-03, grad_scale: 4.0
+2023-03-01 23:37:42,254 INFO [train.py:968] (1/2) Epoch 3, batch 44400, giga_loss[loss=0.3722, simple_loss=0.4212, pruned_loss=0.1616, over 28487.00 frames. ], tot_loss[loss=0.3676, simple_loss=0.4173, pruned_loss=0.1589, over 5686418.89 frames. ], libri_tot_loss[loss=0.3688, simple_loss=0.4133, pruned_loss=0.1622, over 5710008.47 frames. ], giga_tot_loss[loss=0.3671, simple_loss=0.4173, pruned_loss=0.1585, over 5674781.23 frames. ], batch size: 65, lr: 8.97e-03, grad_scale: 8.0
+2023-03-01 23:37:47,613 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:37:51,927 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=134702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:38:18,934 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 44450, giga_loss[loss=0.4281, simple_loss=0.4382, pruned_loss=0.209, over 23402.00 frames. ], tot_loss[loss=0.3714, simple_loss=0.4189, pruned_loss=0.1619, over 5658953.26 frames. ], libri_tot_loss[loss=0.3686, simple_loss=0.413, pruned_loss=0.1621, over 5696848.63 frames. ], giga_tot_loss[loss=0.3712, simple_loss=0.4193, pruned_loss=0.1616, over 5661172.65 frames. ], batch size: 705, lr: 8.97e-03, grad_scale: 8.0
+2023-03-01 23:38:54,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=134765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:39:19,489 INFO [train.py:968] (1/2) Epoch 3, batch 44500, giga_loss[loss=0.4231, simple_loss=0.4594, pruned_loss=0.1934, over 28641.00 frames. ], tot_loss[loss=0.3739, simple_loss=0.4202, pruned_loss=0.1638, over 5651510.21 frames. ], libri_tot_loss[loss=0.3684, simple_loss=0.4129, pruned_loss=0.1619, over 5697095.84 frames. ], giga_tot_loss[loss=0.374, simple_loss=0.4206, pruned_loss=0.1637, over 5652537.32 frames. ], batch size: 336, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:39:27,166 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-01 23:39:54,862 INFO [optim.py:369] (1/2) 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 23:39:56,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2243, 2.9767, 2.9714, 1.5764], device='cuda:1'), covar=tensor([0.0716, 0.0587, 0.1051, 0.2052], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0661, 0.0827, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:40:02,641 INFO [train.py:968] (1/2) Epoch 3, batch 44550, giga_loss[loss=0.3607, simple_loss=0.4076, pruned_loss=0.157, over 27527.00 frames. ], tot_loss[loss=0.3725, simple_loss=0.4191, pruned_loss=0.163, over 5653723.26 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4123, pruned_loss=0.1616, over 5695516.91 frames. ], giga_tot_loss[loss=0.3733, simple_loss=0.4202, pruned_loss=0.1632, over 5654395.25 frames. ], batch size: 472, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:40:47,531 INFO [train.py:968] (1/2) Epoch 3, batch 44600, giga_loss[loss=0.4422, simple_loss=0.456, pruned_loss=0.2142, over 26643.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4169, pruned_loss=0.1594, over 5650299.18 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4123, pruned_loss=0.1616, over 5687787.83 frames. ], giga_tot_loss[loss=0.3685, simple_loss=0.418, pruned_loss=0.1596, over 5657233.19 frames. ], batch size: 555, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:41:04,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=134908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:41:08,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=134911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:41:25,910 INFO [optim.py:369] (1/2) 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 23:41:34,115 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=134940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:41:34,514 INFO [train.py:968] (1/2) Epoch 3, batch 44650, giga_loss[loss=0.3555, simple_loss=0.4129, pruned_loss=0.1491, over 28968.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4184, pruned_loss=0.1589, over 5647958.51 frames. ], libri_tot_loss[loss=0.3681, simple_loss=0.4126, pruned_loss=0.1618, over 5679877.58 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.419, pruned_loss=0.1589, over 5660162.81 frames. ], batch size: 213, lr: 8.97e-03, grad_scale: 4.0
+2023-03-01 23:42:19,768 INFO [train.py:968] (1/2) Epoch 3, batch 44700, giga_loss[loss=0.3401, simple_loss=0.398, pruned_loss=0.1411, over 28687.00 frames. ], tot_loss[loss=0.368, simple_loss=0.4184, pruned_loss=0.1588, over 5667521.72 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4125, pruned_loss=0.1616, over 5687829.84 frames. ], giga_tot_loss[loss=0.3685, simple_loss=0.4192, pruned_loss=0.1589, over 5669299.01 frames. ], batch size: 92, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:42:23,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0020, 1.1019, 0.8267, 0.6222], device='cuda:1'), covar=tensor([0.0540, 0.0591, 0.0503, 0.0617], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.0992, 0.1021, 0.1081], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 23:43:01,407 INFO [optim.py:369] (1/2) 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 23:43:09,738 INFO [train.py:968] (1/2) Epoch 3, batch 44750, giga_loss[loss=0.3601, simple_loss=0.4095, pruned_loss=0.1553, over 28711.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4176, pruned_loss=0.1589, over 5671469.64 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4124, pruned_loss=0.1615, over 5690851.14 frames. ], giga_tot_loss[loss=0.3683, simple_loss=0.4184, pruned_loss=0.1591, over 5669790.56 frames. ], batch size: 262, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:43:40,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:43:45,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:43:48,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-01 23:43:57,861 INFO [train.py:968] (1/2) Epoch 3, batch 44800, giga_loss[loss=0.3331, simple_loss=0.3896, pruned_loss=0.1383, over 28603.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.4161, pruned_loss=0.1595, over 5662333.70 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4122, pruned_loss=0.1613, over 5694247.96 frames. ], giga_tot_loss[loss=0.3682, simple_loss=0.417, pruned_loss=0.1597, over 5657772.98 frames. ], batch size: 307, lr: 8.96e-03, grad_scale: 8.0
+2023-03-01 23:44:21,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-01 23:44:39,214 INFO [optim.py:369] (1/2) 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 23:44:47,406 INFO [train.py:968] (1/2) Epoch 3, batch 44850, giga_loss[loss=0.3549, simple_loss=0.4045, pruned_loss=0.1526, over 28541.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.415, pruned_loss=0.16, over 5661076.12 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4125, pruned_loss=0.1616, over 5698233.53 frames. ], giga_tot_loss[loss=0.3677, simple_loss=0.4156, pruned_loss=0.1599, over 5653307.23 frames. ], batch size: 307, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:45:33,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5688, 1.7211, 1.2596, 0.8927], device='cuda:1'), covar=tensor([0.0674, 0.0506, 0.0402, 0.0621], device='cuda:1'), in_proj_covar=tensor([0.1261, 0.0997, 0.1027, 0.1091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 23:45:34,096 INFO [train.py:968] (1/2) Epoch 3, batch 44900, libri_loss[loss=0.3559, simple_loss=0.4101, pruned_loss=0.1509, over 29754.00 frames. ], tot_loss[loss=0.364, simple_loss=0.4115, pruned_loss=0.1582, over 5652888.87 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4121, pruned_loss=0.1614, over 5693105.89 frames. ], giga_tot_loss[loss=0.3644, simple_loss=0.4123, pruned_loss=0.1582, over 5650297.99 frames. ], batch size: 87, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:45:52,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0819, 1.3263, 0.9786, 0.1603], device='cuda:1'), covar=tensor([0.0731, 0.0679, 0.1107, 0.1374], device='cuda:1'), in_proj_covar=tensor([0.1269, 0.1223, 0.1286, 0.1107], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-01 23:45:55,691 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:45:59,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:01,557 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:04,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:12,180 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 44950, libri_loss[loss=0.3461, simple_loss=0.4009, pruned_loss=0.1456, over 29533.00 frames. ], tot_loss[loss=0.3622, simple_loss=0.4097, pruned_loss=0.1574, over 5663606.83 frames. ], libri_tot_loss[loss=0.3677, simple_loss=0.4123, pruned_loss=0.1615, over 5699779.51 frames. ], giga_tot_loss[loss=0.3621, simple_loss=0.41, pruned_loss=0.1571, over 5654260.19 frames. ], batch size: 79, lr: 8.96e-03, grad_scale: 4.0
+2023-03-01 23:46:24,196 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:29,558 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=135252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:46:31,437 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:47:03,557 INFO [train.py:968] (1/2) Epoch 3, batch 45000, libri_loss[loss=0.3284, simple_loss=0.3888, pruned_loss=0.134, over 29663.00 frames. ], tot_loss[loss=0.3617, simple_loss=0.4091, pruned_loss=0.1571, over 5670998.01 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4115, pruned_loss=0.1609, over 5697483.06 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4101, pruned_loss=0.1573, over 5663754.56 frames. ], batch size: 91, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:47:03,557 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-01 23:47:11,676 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-01 23:47:47,227 INFO [optim.py:369] (1/2) 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 23:47:56,013 INFO [train.py:968] (1/2) Epoch 3, batch 45050, giga_loss[loss=0.3489, simple_loss=0.413, pruned_loss=0.1424, over 28939.00 frames. ], tot_loss[loss=0.3588, simple_loss=0.4076, pruned_loss=0.155, over 5668658.67 frames. ], libri_tot_loss[loss=0.3667, simple_loss=0.4116, pruned_loss=0.1609, over 5697652.78 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4082, pruned_loss=0.1551, over 5662124.02 frames. ], batch size: 227, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:48:12,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2563, 1.4570, 1.2635, 1.5071], device='cuda:1'), covar=tensor([0.2000, 0.1816, 0.1731, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.1050, 0.0849, 0.0943, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 23:48:41,841 INFO [train.py:968] (1/2) Epoch 3, batch 45100, giga_loss[loss=0.3709, simple_loss=0.416, pruned_loss=0.163, over 27868.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4047, pruned_loss=0.1517, over 5658076.66 frames. ], libri_tot_loss[loss=0.3675, simple_loss=0.4122, pruned_loss=0.1614, over 5688375.17 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4046, pruned_loss=0.1512, over 5660206.76 frames. ], batch size: 412, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:49:21,000 INFO [optim.py:369] (1/2) 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 23:49:29,181 INFO [train.py:968] (1/2) Epoch 3, batch 45150, giga_loss[loss=0.3186, simple_loss=0.3811, pruned_loss=0.1281, over 28761.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4046, pruned_loss=0.151, over 5665689.72 frames. ], libri_tot_loss[loss=0.3674, simple_loss=0.4123, pruned_loss=0.1612, over 5693918.03 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4041, pruned_loss=0.1504, over 5661957.55 frames. ], batch size: 66, lr: 8.95e-03, grad_scale: 4.0
+2023-03-01 23:50:12,214 INFO [train.py:968] (1/2) Epoch 3, batch 45200, giga_loss[loss=0.34, simple_loss=0.3927, pruned_loss=0.1437, over 28629.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4019, pruned_loss=0.15, over 5656275.52 frames. ], libri_tot_loss[loss=0.3671, simple_loss=0.4119, pruned_loss=0.1612, over 5694797.86 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4015, pruned_loss=0.1492, over 5650655.82 frames. ], batch size: 336, lr: 8.95e-03, grad_scale: 8.0
+2023-03-01 23:50:21,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3891, 4.1029, 4.1063, 1.8549], device='cuda:1'), covar=tensor([0.0435, 0.0383, 0.0766, 0.1855], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0661, 0.0821, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:50:58,597 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 45250, giga_loss[loss=0.3831, simple_loss=0.4168, pruned_loss=0.1747, over 27639.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.3984, pruned_loss=0.1485, over 5646766.68 frames. ], libri_tot_loss[loss=0.367, simple_loss=0.4119, pruned_loss=0.161, over 5698134.30 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.398, pruned_loss=0.1479, over 5638716.64 frames. ], batch size: 472, lr: 8.95e-03, grad_scale: 2.0
+2023-03-01 23:51:25,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=135564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:51:48,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2349, 1.7833, 1.4574, 1.3879], device='cuda:1'), covar=tensor([0.0897, 0.0328, 0.0359, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0142, 0.0145, 0.0261], device='cuda:1'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:1')
+2023-03-01 23:51:49,343 INFO [train.py:968] (1/2) Epoch 3, batch 45300, giga_loss[loss=0.3488, simple_loss=0.4125, pruned_loss=0.1425, over 28751.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4009, pruned_loss=0.1503, over 5648931.41 frames. ], libri_tot_loss[loss=0.3673, simple_loss=0.4124, pruned_loss=0.1612, over 5701245.32 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.3999, pruned_loss=0.1493, over 5638625.93 frames. ], batch size: 284, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:52:27,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:52:31,720 INFO [optim.py:369] (1/2) 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 23:52:37,788 INFO [train.py:968] (1/2) Epoch 3, batch 45350, giga_loss[loss=0.331, simple_loss=0.3927, pruned_loss=0.1347, over 28888.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.403, pruned_loss=0.1508, over 5650435.79 frames. ], libri_tot_loss[loss=0.3678, simple_loss=0.4128, pruned_loss=0.1613, over 5701048.93 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4016, pruned_loss=0.1498, over 5641850.83 frames. ], batch size: 213, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:52:58,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5557, 1.8141, 1.6829, 1.6347], device='cuda:1'), covar=tensor([0.1458, 0.1733, 0.1112, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0794, 0.0750, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:1')
+2023-03-01 23:53:13,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4289, 1.5834, 1.1760, 0.8625], device='cuda:1'), covar=tensor([0.0781, 0.0744, 0.0607, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1001, 0.1048, 0.1094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-01 23:53:21,192 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-01 23:53:24,547 INFO [train.py:968] (1/2) Epoch 3, batch 45400, giga_loss[loss=0.3192, simple_loss=0.3858, pruned_loss=0.1263, over 28943.00 frames. ], tot_loss[loss=0.3518, simple_loss=0.4024, pruned_loss=0.1506, over 5641627.83 frames. ], libri_tot_loss[loss=0.368, simple_loss=0.4131, pruned_loss=0.1615, over 5706194.38 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4007, pruned_loss=0.1493, over 5628223.22 frames. ], batch size: 174, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:54:00,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-01 23:54:02,497 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9389, 1.2077, 4.0642, 3.1120], device='cuda:1'), covar=tensor([0.1640, 0.2088, 0.0315, 0.0617], device='cuda:1'), in_proj_covar=tensor([0.0543, 0.0512, 0.0708, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-01 23:54:03,437 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 3, batch 45450, giga_loss[loss=0.3281, simple_loss=0.3886, pruned_loss=0.1338, over 28672.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.4014, pruned_loss=0.1499, over 5632897.56 frames. ], libri_tot_loss[loss=0.3685, simple_loss=0.4134, pruned_loss=0.1618, over 5699641.48 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.3996, pruned_loss=0.1484, over 5626174.37 frames. ], batch size: 66, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:54:23,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7674, 1.6400, 1.5928, 1.5669], device='cuda:1'), covar=tensor([0.0842, 0.1205, 0.1296, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0777, 0.0644, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-01 23:54:33,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8688, 3.2309, 2.2426, 0.9182], device='cuda:1'), covar=tensor([0.2956, 0.1003, 0.1621, 0.2831], device='cuda:1'), in_proj_covar=tensor([0.1285, 0.1225, 0.1304, 0.1115], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 23:54:39,452 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=135773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:54:43,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=135776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:54:54,838 INFO [train.py:968] (1/2) Epoch 3, batch 45500, giga_loss[loss=0.3515, simple_loss=0.4049, pruned_loss=0.1491, over 28609.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4038, pruned_loss=0.1523, over 5626251.39 frames. ], libri_tot_loss[loss=0.3692, simple_loss=0.414, pruned_loss=0.1622, over 5685845.63 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4016, pruned_loss=0.1504, over 5632349.18 frames. ], batch size: 71, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:55:09,225 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:55:42,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-01 23:55:46,134 INFO [train.py:968] (1/2) Epoch 3, batch 45550, giga_loss[loss=0.3159, simple_loss=0.3871, pruned_loss=0.1223, over 29042.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.4077, pruned_loss=0.1555, over 5638219.61 frames. ], libri_tot_loss[loss=0.3699, simple_loss=0.4146, pruned_loss=0.1627, over 5683645.96 frames. ], giga_tot_loss[loss=0.3562, simple_loss=0.4053, pruned_loss=0.1535, over 5644063.87 frames. ], batch size: 136, lr: 8.94e-03, grad_scale: 2.0
+2023-03-01 23:56:33,146 INFO [train.py:968] (1/2) Epoch 3, batch 45600, giga_loss[loss=0.3528, simple_loss=0.401, pruned_loss=0.1523, over 29001.00 frames. ], tot_loss[loss=0.3613, simple_loss=0.4094, pruned_loss=0.1565, over 5649365.88 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4146, pruned_loss=0.1628, over 5686826.66 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4074, pruned_loss=0.1548, over 5650480.43 frames. ], batch size: 213, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:57:12,560 INFO [optim.py:369] (1/2) 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 23:57:17,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=135939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:57:19,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-01 23:57:19,605 INFO [train.py:968] (1/2) Epoch 3, batch 45650, giga_loss[loss=0.3384, simple_loss=0.3941, pruned_loss=0.1414, over 28794.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4114, pruned_loss=0.1585, over 5638404.10 frames. ], libri_tot_loss[loss=0.37, simple_loss=0.4146, pruned_loss=0.1627, over 5673765.00 frames. ], giga_tot_loss[loss=0.3618, simple_loss=0.4096, pruned_loss=0.157, over 5650373.37 frames. ], batch size: 284, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:58:10,717 INFO [train.py:968] (1/2) Epoch 3, batch 45700, giga_loss[loss=0.3207, simple_loss=0.394, pruned_loss=0.1236, over 28797.00 frames. ], tot_loss[loss=0.3667, simple_loss=0.4128, pruned_loss=0.1603, over 5637700.29 frames. ], libri_tot_loss[loss=0.3698, simple_loss=0.4144, pruned_loss=0.1626, over 5676025.12 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4115, pruned_loss=0.1591, over 5644650.60 frames. ], batch size: 243, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:58:22,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0901, 3.8239, 3.7433, 1.8690], device='cuda:1'), covar=tensor([0.0516, 0.0489, 0.0921, 0.1949], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0664, 0.0822, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-01 23:58:25,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6409, 2.6339, 1.8866, 0.7637], device='cuda:1'), covar=tensor([0.2924, 0.1257, 0.1483, 0.2678], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1231, 0.1308, 0.1116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-01 23:58:44,308 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:58:54,385 INFO [optim.py:369] (1/2) 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 23:59:02,050 INFO [train.py:968] (1/2) Epoch 3, batch 45750, giga_loss[loss=0.3337, simple_loss=0.3999, pruned_loss=0.1338, over 28958.00 frames. ], tot_loss[loss=0.3654, simple_loss=0.4133, pruned_loss=0.1587, over 5596639.44 frames. ], libri_tot_loss[loss=0.3712, simple_loss=0.4153, pruned_loss=0.1635, over 5623184.19 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4115, pruned_loss=0.1568, over 5650058.11 frames. ], batch size: 164, lr: 8.93e-03, grad_scale: 4.0
+2023-03-01 23:59:43,436 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:59:46,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-01 23:59:51,623 INFO [train.py:968] (1/2) Epoch 3, batch 45800, giga_loss[loss=0.3968, simple_loss=0.4358, pruned_loss=0.1789, over 28641.00 frames. ], tot_loss[loss=0.3651, simple_loss=0.4128, pruned_loss=0.1587, over 5563006.09 frames. ], libri_tot_loss[loss=0.3715, simple_loss=0.4154, pruned_loss=0.1638, over 5579991.49 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4112, pruned_loss=0.1569, over 5642687.14 frames. ], batch size: 307, lr: 8.93e-03, grad_scale: 4.0
+2023-03-02 00:00:11,996 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:00:29,726 INFO [optim.py:369] (1/2) 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 00:00:38,116 INFO [train.py:968] (1/2) Epoch 3, batch 45850, giga_loss[loss=0.4083, simple_loss=0.4409, pruned_loss=0.1879, over 28598.00 frames. ], tot_loss[loss=0.3644, simple_loss=0.4116, pruned_loss=0.1586, over 5541494.25 frames. ], libri_tot_loss[loss=0.3725, simple_loss=0.4161, pruned_loss=0.1644, over 5530468.82 frames. ], giga_tot_loss[loss=0.3613, simple_loss=0.4097, pruned_loss=0.1565, over 5647735.02 frames. ], batch size: 336, lr: 8.93e-03, grad_scale: 4.0
+2023-03-02 00:00:41,928 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-02 00:01:45,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6273, 1.9420, 1.8755, 1.7302], device='cuda:1'), covar=tensor([0.1539, 0.1873, 0.1207, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0788, 0.0745, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 00:02:04,856 INFO [train.py:968] (1/2) Epoch 4, batch 50, giga_loss[loss=0.3704, simple_loss=0.4296, pruned_loss=0.1556, over 28426.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.4119, pruned_loss=0.1422, over 1264622.18 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3725, pruned_loss=0.118, over 86227.14 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4147, pruned_loss=0.1438, over 1196466.66 frames. ], batch size: 368, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:02:44,068 INFO [optim.py:369] (1/2) 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 00:02:51,328 INFO [train.py:968] (1/2) Epoch 4, batch 100, giga_loss[loss=0.3245, simple_loss=0.3919, pruned_loss=0.1285, over 28990.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3986, pruned_loss=0.135, over 2241704.32 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3758, pruned_loss=0.1209, over 285874.65 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.4015, pruned_loss=0.1369, over 2057797.43 frames. ], batch size: 136, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:02:54,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 00:03:35,834 INFO [train.py:968] (1/2) Epoch 4, batch 150, libri_loss[loss=0.305, simple_loss=0.3637, pruned_loss=0.1232, over 28661.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3805, pruned_loss=0.1255, over 3010318.61 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3752, pruned_loss=0.1218, over 478400.38 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3819, pruned_loss=0.1263, over 2760598.80 frames. ], batch size: 63, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:04:08,602 INFO [optim.py:369] (1/2) 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 00:04:16,437 INFO [train.py:968] (1/2) Epoch 4, batch 200, giga_loss[loss=0.2591, simple_loss=0.3315, pruned_loss=0.09339, over 28591.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3671, pruned_loss=0.1183, over 3602188.41 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3796, pruned_loss=0.1248, over 576472.57 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3664, pruned_loss=0.1179, over 3366548.99 frames. ], batch size: 336, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:04:43,937 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 4, batch 250, giga_loss[loss=0.2484, simple_loss=0.3175, pruned_loss=0.0896, over 28224.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3559, pruned_loss=0.1128, over 4058133.55 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3791, pruned_loss=0.1243, over 698994.84 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3544, pruned_loss=0.112, over 3831947.28 frames. ], batch size: 77, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:05:04,819 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:05:42,127 INFO [train.py:968] (1/2) Epoch 4, batch 300, giga_loss[loss=0.2178, simple_loss=0.3031, pruned_loss=0.06624, over 28870.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.345, pruned_loss=0.1073, over 4417721.29 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3784, pruned_loss=0.1244, over 751103.15 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3431, pruned_loss=0.1063, over 4224835.04 frames. ], batch size: 145, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:06:26,565 INFO [train.py:968] (1/2) Epoch 4, batch 350, giga_loss[loss=0.2653, simple_loss=0.3298, pruned_loss=0.1003, over 28887.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3377, pruned_loss=0.1036, over 4700065.86 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3802, pruned_loss=0.1254, over 853140.36 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3349, pruned_loss=0.1021, over 4523445.06 frames. ], batch size: 186, lr: 8.34e-03, grad_scale: 4.0
+2023-03-02 00:06:56,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0098, 1.2635, 3.8679, 3.1537], device='cuda:1'), covar=tensor([0.1536, 0.2104, 0.0347, 0.0738], device='cuda:1'), in_proj_covar=tensor([0.0537, 0.0507, 0.0707, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:06:59,640 INFO [optim.py:369] (1/2) 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 00:07:01,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:07:02,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8654, 2.3252, 2.1115, 1.9507], device='cuda:1'), covar=tensor([0.1705, 0.1700, 0.1245, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0792, 0.0762, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 00:07:06,488 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 400, giga_loss[loss=0.2535, simple_loss=0.3179, pruned_loss=0.09456, over 28849.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3333, pruned_loss=0.1009, over 4927735.33 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3785, pruned_loss=0.1234, over 1025539.45 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3297, pruned_loss=0.09921, over 4752767.29 frames. ], batch size: 186, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:07:10,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:07:32,785 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:07:39,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1840, 2.9737, 2.9132, 1.3659], device='cuda:1'), covar=tensor([0.0796, 0.0596, 0.1024, 0.2289], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0635, 0.0786, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 00:07:46,633 INFO [train.py:968] (1/2) Epoch 4, batch 450, giga_loss[loss=0.2365, simple_loss=0.3112, pruned_loss=0.08083, over 28833.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3333, pruned_loss=0.1008, over 5106834.15 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3795, pruned_loss=0.1237, over 1263960.60 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3277, pruned_loss=0.09809, over 4923784.35 frames. ], batch size: 199, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:08:25,733 INFO [optim.py:369] (1/2) 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 00:08:32,021 INFO [train.py:968] (1/2) Epoch 4, batch 500, giga_loss[loss=0.222, simple_loss=0.2905, pruned_loss=0.07676, over 28429.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3298, pruned_loss=0.09891, over 5233403.96 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3788, pruned_loss=0.1225, over 1355251.53 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3245, pruned_loss=0.0966, over 5075474.44 frames. ], batch size: 71, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:08:35,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-02 00:09:05,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4464, 3.0893, 1.5963, 1.3627], device='cuda:1'), covar=tensor([0.0854, 0.0430, 0.0620, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.1263, 0.1005, 0.1019, 0.1085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 00:09:16,709 INFO [train.py:968] (1/2) Epoch 4, batch 550, giga_loss[loss=0.2095, simple_loss=0.2865, pruned_loss=0.06621, over 29009.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.327, pruned_loss=0.09722, over 5337048.06 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3789, pruned_loss=0.1222, over 1466402.95 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3215, pruned_loss=0.09482, over 5198444.01 frames. ], batch size: 136, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:09:47,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7089, 1.6057, 1.5088, 1.9945], device='cuda:1'), covar=tensor([0.2039, 0.2022, 0.1880, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1066, 0.0852, 0.0950, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:09:50,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=136728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:09:54,356 INFO [optim.py:369] (1/2) 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 00:10:03,016 INFO [train.py:968] (1/2) Epoch 4, batch 600, giga_loss[loss=0.257, simple_loss=0.321, pruned_loss=0.09649, over 28935.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3245, pruned_loss=0.09582, over 5414646.65 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3803, pruned_loss=0.123, over 1552682.53 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3188, pruned_loss=0.09326, over 5296728.92 frames. ], batch size: 227, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:10:11,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:10:20,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7468, 3.4764, 3.4571, 1.7026], device='cuda:1'), covar=tensor([0.0554, 0.0517, 0.0821, 0.2263], device='cuda:1'), in_proj_covar=tensor([0.0778, 0.0635, 0.0784, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 00:10:24,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:10:52,050 INFO [train.py:968] (1/2) Epoch 4, batch 650, giga_loss[loss=0.2293, simple_loss=0.3037, pruned_loss=0.07745, over 28911.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3221, pruned_loss=0.09458, over 5481187.74 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3796, pruned_loss=0.1224, over 1659472.79 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3164, pruned_loss=0.09205, over 5377806.11 frames. ], batch size: 174, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:11:29,455 INFO [optim.py:369] (1/2) 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 00:11:29,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 00:11:37,729 INFO [train.py:968] (1/2) Epoch 4, batch 700, giga_loss[loss=0.2271, simple_loss=0.2881, pruned_loss=0.08304, over 28515.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3186, pruned_loss=0.09277, over 5531107.09 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3794, pruned_loss=0.122, over 1722927.10 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3132, pruned_loss=0.0904, over 5443784.17 frames. ], batch size: 78, lr: 8.33e-03, grad_scale: 8.0
+2023-03-02 00:12:27,504 INFO [train.py:968] (1/2) Epoch 4, batch 750, giga_loss[loss=0.2448, simple_loss=0.3152, pruned_loss=0.08718, over 28584.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3152, pruned_loss=0.09088, over 5554006.31 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3798, pruned_loss=0.1221, over 1740269.46 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3107, pruned_loss=0.08887, over 5487192.71 frames. ], batch size: 336, lr: 8.32e-03, grad_scale: 8.0
+2023-03-02 00:12:27,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:31,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:39,536 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=136905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:41,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=136908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:43,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=136910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:56,599 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:12:59,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-02 00:13:02,271 INFO [optim.py:369] (1/2) 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 00:13:05,305 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=136937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:13:11,461 INFO [train.py:968] (1/2) Epoch 4, batch 800, giga_loss[loss=0.3453, simple_loss=0.3781, pruned_loss=0.1562, over 26484.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3155, pruned_loss=0.09166, over 5576101.66 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.382, pruned_loss=0.1234, over 1811897.65 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3101, pruned_loss=0.08915, over 5526460.62 frames. ], batch size: 555, lr: 8.32e-03, grad_scale: 8.0
+2023-03-02 00:14:00,994 INFO [train.py:968] (1/2) Epoch 4, batch 850, giga_loss[loss=0.3126, simple_loss=0.3862, pruned_loss=0.1195, over 28974.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3267, pruned_loss=0.09833, over 5599647.85 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3822, pruned_loss=0.1238, over 1873227.03 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3216, pruned_loss=0.09594, over 5556066.65 frames. ], batch size: 145, lr: 8.32e-03, grad_scale: 8.0
+2023-03-02 00:14:08,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4408, 2.9866, 1.4631, 1.4883], device='cuda:1'), covar=tensor([0.0947, 0.0366, 0.0871, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0451, 0.0308, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-02 00:14:42,234 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=137036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:14:49,465 INFO [train.py:968] (1/2) Epoch 4, batch 900, giga_loss[loss=0.3742, simple_loss=0.4334, pruned_loss=0.1575, over 29041.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.343, pruned_loss=0.1077, over 5627698.47 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3824, pruned_loss=0.1241, over 1932146.85 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3384, pruned_loss=0.1055, over 5590524.90 frames. ], batch size: 155, lr: 8.32e-03, grad_scale: 4.0
+2023-03-02 00:15:01,288 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:03,423 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:26,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:32,065 INFO [train.py:968] (1/2) Epoch 4, batch 950, giga_loss[loss=0.343, simple_loss=0.4097, pruned_loss=0.1382, over 28898.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3556, pruned_loss=0.115, over 5634337.26 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.383, pruned_loss=0.1243, over 2011009.12 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3513, pruned_loss=0.113, over 5599815.06 frames. ], batch size: 227, lr: 8.32e-03, grad_scale: 4.0
+2023-03-02 00:15:41,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:15:42,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-02 00:15:54,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3131, 2.7628, 1.3480, 1.2801], device='cuda:1'), covar=tensor([0.0930, 0.0338, 0.0903, 0.1410], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0455, 0.0308, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-02 00:16:08,802 INFO [optim.py:369] (1/2) 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 00:16:15,512 INFO [train.py:968] (1/2) Epoch 4, batch 1000, giga_loss[loss=0.291, simple_loss=0.3687, pruned_loss=0.1066, over 28947.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3623, pruned_loss=0.1174, over 5646112.43 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3822, pruned_loss=0.1236, over 2088407.74 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3586, pruned_loss=0.1159, over 5614007.28 frames. ], batch size: 174, lr: 8.32e-03, grad_scale: 4.0
+2023-03-02 00:16:20,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8239, 1.7009, 1.7641, 1.7699], device='cuda:1'), covar=tensor([0.0914, 0.1353, 0.1185, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0767, 0.0631, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:16:39,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2192, 1.2756, 1.1151, 1.1935], device='cuda:1'), covar=tensor([0.0625, 0.0453, 0.0910, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0469, 0.0512, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-02 00:16:56,519 INFO [train.py:968] (1/2) Epoch 4, batch 1050, giga_loss[loss=0.2874, simple_loss=0.3491, pruned_loss=0.1129, over 23640.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3646, pruned_loss=0.1166, over 5661246.24 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3812, pruned_loss=0.1226, over 2182878.63 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3616, pruned_loss=0.1156, over 5630291.37 frames. ], batch size: 705, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:17:03,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=137199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:17:35,454 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 1100, giga_loss[loss=0.2879, simple_loss=0.3664, pruned_loss=0.1047, over 28985.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3664, pruned_loss=0.1171, over 5656092.01 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3823, pruned_loss=0.1233, over 2246638.09 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3634, pruned_loss=0.116, over 5635619.47 frames. ], batch size: 145, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:17:42,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:17:45,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:18:01,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5141, 1.4246, 1.1245, 1.3154], device='cuda:1'), covar=tensor([0.0553, 0.0461, 0.0821, 0.0753], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0468, 0.0508, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-02 00:18:07,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4225, 4.0449, 4.1674, 1.8986], device='cuda:1'), covar=tensor([0.0362, 0.0379, 0.0681, 0.1809], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0618, 0.0772, 0.0563], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 00:18:07,674 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 1150, giga_loss[loss=0.447, simple_loss=0.4638, pruned_loss=0.2152, over 26550.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3696, pruned_loss=0.1198, over 5667405.37 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.383, pruned_loss=0.1244, over 2390683.71 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3664, pruned_loss=0.1183, over 5642070.06 frames. ], batch size: 555, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:18:58,375 INFO [optim.py:369] (1/2) 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 00:19:00,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1324, 1.1821, 1.2997, 1.1215], device='cuda:1'), covar=tensor([0.1028, 0.1092, 0.1478, 0.1163], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0780, 0.0643, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:19:03,682 INFO [train.py:968] (1/2) Epoch 4, batch 1200, giga_loss[loss=0.2759, simple_loss=0.3448, pruned_loss=0.1034, over 29046.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.373, pruned_loss=0.1222, over 5659112.00 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3843, pruned_loss=0.1254, over 2454927.11 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3698, pruned_loss=0.1206, over 5652331.93 frames. ], batch size: 106, lr: 8.31e-03, grad_scale: 8.0
+2023-03-02 00:19:48,854 INFO [train.py:968] (1/2) Epoch 4, batch 1250, giga_loss[loss=0.3704, simple_loss=0.4225, pruned_loss=0.1591, over 28695.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3772, pruned_loss=0.125, over 5666927.72 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3847, pruned_loss=0.1257, over 2506732.20 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3743, pruned_loss=0.1236, over 5658020.53 frames. ], batch size: 85, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:20:06,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:20:26,587 INFO [optim.py:369] (1/2) 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 00:20:32,085 INFO [train.py:968] (1/2) Epoch 4, batch 1300, giga_loss[loss=0.3028, simple_loss=0.378, pruned_loss=0.1138, over 28713.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3801, pruned_loss=0.1256, over 5683750.71 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3847, pruned_loss=0.1257, over 2639034.47 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3777, pruned_loss=0.1244, over 5669383.78 frames. ], batch size: 284, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:21:10,094 INFO [train.py:968] (1/2) Epoch 4, batch 1350, giga_loss[loss=0.4117, simple_loss=0.4411, pruned_loss=0.1912, over 26512.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3822, pruned_loss=0.1266, over 5680877.77 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3852, pruned_loss=0.126, over 2719144.93 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.38, pruned_loss=0.1256, over 5665303.91 frames. ], batch size: 555, lr: 8.31e-03, grad_scale: 4.0
+2023-03-02 00:21:44,316 INFO [optim.py:369] (1/2) 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 00:21:48,984 INFO [train.py:968] (1/2) Epoch 4, batch 1400, giga_loss[loss=0.3018, simple_loss=0.3799, pruned_loss=0.1118, over 28940.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3836, pruned_loss=0.1263, over 5697082.90 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3847, pruned_loss=0.1257, over 2843208.51 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3819, pruned_loss=0.1256, over 5677760.51 frames. ], batch size: 174, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:21:56,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:21:56,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4381, 1.3460, 1.1742, 1.6554], device='cuda:1'), covar=tensor([0.2032, 0.1900, 0.1861, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.1057, 0.0844, 0.0940, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:22:00,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:22:17,878 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=137574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:22:27,414 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=137586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:22:34,934 INFO [train.py:968] (1/2) Epoch 4, batch 1450, giga_loss[loss=0.3605, simple_loss=0.4038, pruned_loss=0.1586, over 27566.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3834, pruned_loss=0.1254, over 5694039.81 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3847, pruned_loss=0.1258, over 2928926.28 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.382, pruned_loss=0.1248, over 5677527.64 frames. ], batch size: 472, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:23:03,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1452, 2.3139, 1.1839, 1.1816], device='cuda:1'), covar=tensor([0.0961, 0.0316, 0.0834, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0452, 0.0306, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-02 00:23:08,566 INFO [optim.py:369] (1/2) 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 00:23:12,548 INFO [train.py:968] (1/2) Epoch 4, batch 1500, giga_loss[loss=0.3248, simple_loss=0.4013, pruned_loss=0.1242, over 28923.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3813, pruned_loss=0.1226, over 5704921.09 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3841, pruned_loss=0.1251, over 3015141.02 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3804, pruned_loss=0.1224, over 5688111.65 frames. ], batch size: 164, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:23:21,427 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 00:23:53,327 INFO [train.py:968] (1/2) Epoch 4, batch 1550, giga_loss[loss=0.358, simple_loss=0.4155, pruned_loss=0.1503, over 28808.00 frames. ], tot_loss[loss=0.311, simple_loss=0.38, pruned_loss=0.121, over 5713184.03 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3851, pruned_loss=0.1255, over 3086808.76 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3788, pruned_loss=0.1206, over 5695402.96 frames. ], batch size: 112, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:24:13,289 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=137717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:24:16,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=137720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:24:33,372 INFO [optim.py:369] (1/2) 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 00:24:37,484 INFO [train.py:968] (1/2) Epoch 4, batch 1600, giga_loss[loss=0.3781, simple_loss=0.4167, pruned_loss=0.1697, over 28550.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3815, pruned_loss=0.1236, over 5704184.51 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3849, pruned_loss=0.1253, over 3156342.43 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3804, pruned_loss=0.1233, over 5685686.47 frames. ], batch size: 336, lr: 8.30e-03, grad_scale: 8.0
+2023-03-02 00:24:43,501 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 1650, giga_loss[loss=0.3424, simple_loss=0.3904, pruned_loss=0.1472, over 28926.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3833, pruned_loss=0.1275, over 5712252.62 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3836, pruned_loss=0.1246, over 3224301.05 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3831, pruned_loss=0.1276, over 5693058.10 frames. ], batch size: 199, lr: 8.30e-03, grad_scale: 4.0
+2023-03-02 00:25:44,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8426, 1.0082, 3.8543, 3.1155], device='cuda:1'), covar=tensor([0.1953, 0.2422, 0.0536, 0.0639], device='cuda:1'), in_proj_covar=tensor([0.0531, 0.0505, 0.0697, 0.0570], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:25:59,772 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 1700, giga_loss[loss=0.3475, simple_loss=0.3977, pruned_loss=0.1486, over 28975.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3849, pruned_loss=0.1298, over 5713669.10 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3839, pruned_loss=0.1248, over 3368874.75 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3846, pruned_loss=0.13, over 5698674.86 frames. ], batch size: 136, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:26:48,611 INFO [train.py:968] (1/2) Epoch 4, batch 1750, libri_loss[loss=0.3687, simple_loss=0.4237, pruned_loss=0.1568, over 26102.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3845, pruned_loss=0.1308, over 5695964.28 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3842, pruned_loss=0.1251, over 3420072.53 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3841, pruned_loss=0.1309, over 5690648.08 frames. ], batch size: 136, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:27:12,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1022, 1.3690, 1.1585, 0.8807], device='cuda:1'), covar=tensor([0.2122, 0.2001, 0.1885, 0.1826], device='cuda:1'), in_proj_covar=tensor([0.1059, 0.0848, 0.0939, 0.0942], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:27:16,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3917, 2.9793, 1.4706, 1.3751], device='cuda:1'), covar=tensor([0.0891, 0.0348, 0.0804, 0.1388], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0453, 0.0307, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-02 00:27:23,199 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3021, 1.4399, 1.1780, 1.5187], device='cuda:1'), covar=tensor([0.0853, 0.0365, 0.0376, 0.0901], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0140, 0.0143, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:1')
+2023-03-02 00:27:26,266 INFO [optim.py:369] (1/2) 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 00:27:29,804 INFO [train.py:968] (1/2) Epoch 4, batch 1800, giga_loss[loss=0.3072, simple_loss=0.3635, pruned_loss=0.1255, over 28534.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3825, pruned_loss=0.1304, over 5686847.78 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3845, pruned_loss=0.1252, over 3476706.48 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.382, pruned_loss=0.1307, over 5682311.30 frames. ], batch size: 71, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:28:13,794 INFO [train.py:968] (1/2) Epoch 4, batch 1850, giga_loss[loss=0.3173, simple_loss=0.3636, pruned_loss=0.1355, over 28665.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.381, pruned_loss=0.1289, over 5685213.91 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3849, pruned_loss=0.1254, over 3500679.14 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3805, pruned_loss=0.129, over 5679964.87 frames. ], batch size: 92, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:28:46,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5633, 1.4485, 1.1399, 1.2808], device='cuda:1'), covar=tensor([0.0526, 0.0475, 0.0841, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0475, 0.0511, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-02 00:28:50,892 INFO [optim.py:369] (1/2) 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 00:28:56,494 INFO [train.py:968] (1/2) Epoch 4, batch 1900, giga_loss[loss=0.3109, simple_loss=0.3788, pruned_loss=0.1214, over 28677.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3795, pruned_loss=0.1269, over 5691484.88 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3851, pruned_loss=0.1254, over 3612791.12 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3787, pruned_loss=0.127, over 5682508.19 frames. ], batch size: 242, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:29:00,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4600, 1.5086, 1.3140, 1.5232], device='cuda:1'), covar=tensor([0.2002, 0.1991, 0.1861, 0.2041], device='cuda:1'), in_proj_covar=tensor([0.1058, 0.0845, 0.0936, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:29:44,243 INFO [train.py:968] (1/2) Epoch 4, batch 1950, libri_loss[loss=0.2955, simple_loss=0.381, pruned_loss=0.105, over 29212.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3744, pruned_loss=0.1229, over 5687065.79 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3848, pruned_loss=0.125, over 3664936.19 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3737, pruned_loss=0.1233, over 5679072.57 frames. ], batch size: 97, lr: 8.29e-03, grad_scale: 4.0
+2023-03-02 00:30:26,314 INFO [optim.py:369] (1/2) 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 00:30:30,456 INFO [train.py:968] (1/2) Epoch 4, batch 2000, giga_loss[loss=0.2765, simple_loss=0.3381, pruned_loss=0.1075, over 28550.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3678, pruned_loss=0.1189, over 5679277.96 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3845, pruned_loss=0.1247, over 3717820.92 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3672, pruned_loss=0.1194, over 5670297.70 frames. ], batch size: 71, lr: 8.29e-03, grad_scale: 8.0
+2023-03-02 00:31:12,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7949, 1.6535, 1.7777, 1.6798], device='cuda:1'), covar=tensor([0.1115, 0.1910, 0.1414, 0.1456], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0769, 0.0643, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:31:12,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1717, 1.3283, 4.9877, 3.5420], device='cuda:1'), covar=tensor([0.1530, 0.2041, 0.0229, 0.0517], device='cuda:1'), in_proj_covar=tensor([0.0535, 0.0506, 0.0694, 0.0569], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:31:17,912 INFO [train.py:968] (1/2) Epoch 4, batch 2050, giga_loss[loss=0.2981, simple_loss=0.3558, pruned_loss=0.1202, over 28904.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3617, pruned_loss=0.1152, over 5678510.10 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3846, pruned_loss=0.1246, over 3749271.26 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3609, pruned_loss=0.1155, over 5669455.02 frames. ], batch size: 106, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:31:47,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4022, 1.3999, 1.2933, 1.5008], device='cuda:1'), covar=tensor([0.2128, 0.2113, 0.1878, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1071, 0.0855, 0.0952, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:31:51,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7168, 1.5833, 1.7040, 1.5402], device='cuda:1'), covar=tensor([0.1163, 0.1854, 0.1437, 0.1514], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0769, 0.0643, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:32:03,838 INFO [optim.py:369] (1/2) 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 00:32:08,001 INFO [train.py:968] (1/2) Epoch 4, batch 2100, giga_loss[loss=0.2998, simple_loss=0.3643, pruned_loss=0.1177, over 28567.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3585, pruned_loss=0.1138, over 5662338.66 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3851, pruned_loss=0.1247, over 3792494.28 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3571, pruned_loss=0.1138, over 5650574.25 frames. ], batch size: 307, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:32:37,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2002, 1.9184, 1.4252, 0.4093], device='cuda:1'), covar=tensor([0.2093, 0.1048, 0.1754, 0.2402], device='cuda:1'), in_proj_covar=tensor([0.1283, 0.1212, 0.1292, 0.1100], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 00:32:40,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=138285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:32:43,156 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 2150, giga_loss[loss=0.3139, simple_loss=0.3756, pruned_loss=0.1261, over 27659.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3605, pruned_loss=0.1147, over 5674825.85 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3853, pruned_loss=0.1249, over 3833915.35 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3587, pruned_loss=0.1143, over 5661812.99 frames. ], batch size: 472, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:33:23,836 INFO [optim.py:369] (1/2) 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 00:33:28,970 INFO [train.py:968] (1/2) Epoch 4, batch 2200, giga_loss[loss=0.271, simple_loss=0.3283, pruned_loss=0.1068, over 28435.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3591, pruned_loss=0.1134, over 5690048.91 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3858, pruned_loss=0.1249, over 3883255.31 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3568, pruned_loss=0.1129, over 5676491.42 frames. ], batch size: 71, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:33:38,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 00:34:09,638 INFO [train.py:968] (1/2) Epoch 4, batch 2250, libri_loss[loss=0.3493, simple_loss=0.4078, pruned_loss=0.1454, over 19336.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3574, pruned_loss=0.1126, over 5688091.32 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3859, pruned_loss=0.1249, over 3959654.78 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3545, pruned_loss=0.1116, over 5680905.40 frames. ], batch size: 188, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:34:18,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2277, 1.2304, 0.9972, 1.4077], device='cuda:1'), covar=tensor([0.0805, 0.0368, 0.0381, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0138, 0.0142, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:1')
+2023-03-02 00:34:48,875 INFO [optim.py:369] (1/2) 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 00:34:52,331 INFO [train.py:968] (1/2) Epoch 4, batch 2300, giga_loss[loss=0.2506, simple_loss=0.3246, pruned_loss=0.08835, over 29034.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3558, pruned_loss=0.1121, over 5698303.13 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3867, pruned_loss=0.1256, over 3988604.63 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3525, pruned_loss=0.1108, over 5689850.88 frames. ], batch size: 164, lr: 8.28e-03, grad_scale: 8.0
+2023-03-02 00:35:30,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0158, 2.2008, 1.0395, 1.0892], device='cuda:1'), covar=tensor([0.1218, 0.0503, 0.1088, 0.1715], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0449, 0.0304, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-02 00:35:32,492 INFO [train.py:968] (1/2) Epoch 4, batch 2350, giga_loss[loss=0.301, simple_loss=0.3698, pruned_loss=0.1161, over 28268.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3536, pruned_loss=0.1109, over 5694408.25 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3874, pruned_loss=0.126, over 4043836.81 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3494, pruned_loss=0.109, over 5690934.26 frames. ], batch size: 368, lr: 8.28e-03, grad_scale: 4.0
+2023-03-02 00:36:08,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8085, 1.1026, 3.6156, 2.9339], device='cuda:1'), covar=tensor([0.1566, 0.2098, 0.0312, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0529, 0.0502, 0.0687, 0.0565], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:36:11,335 INFO [optim.py:369] (1/2) 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 00:36:15,919 INFO [train.py:968] (1/2) Epoch 4, batch 2400, giga_loss[loss=0.2952, simple_loss=0.35, pruned_loss=0.1202, over 28827.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.35, pruned_loss=0.1093, over 5694992.37 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3871, pruned_loss=0.1258, over 4071302.20 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3465, pruned_loss=0.1078, over 5689782.89 frames. ], batch size: 99, lr: 8.27e-03, grad_scale: 8.0
+2023-03-02 00:36:54,669 INFO [train.py:968] (1/2) Epoch 4, batch 2450, giga_loss[loss=0.2605, simple_loss=0.3314, pruned_loss=0.09475, over 28869.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3472, pruned_loss=0.1076, over 5695849.79 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3874, pruned_loss=0.1259, over 4106315.29 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3432, pruned_loss=0.1059, over 5697167.47 frames. ], batch size: 145, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:36:54,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1661, 1.7923, 1.3902, 0.4357], device='cuda:1'), covar=tensor([0.1782, 0.1030, 0.2030, 0.2188], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1195, 0.1301, 0.1094], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 00:37:09,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0039, 1.9910, 1.8535, 1.8888], device='cuda:1'), covar=tensor([0.0944, 0.1416, 0.1341, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0766, 0.0635, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:37:29,919 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 2500, giga_loss[loss=0.2285, simple_loss=0.3007, pruned_loss=0.07814, over 28442.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3445, pruned_loss=0.1061, over 5701569.39 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3877, pruned_loss=0.1257, over 4136558.50 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3404, pruned_loss=0.1046, over 5703909.77 frames. ], batch size: 85, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:37:47,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:37:49,057 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=138663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:37:57,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6754, 1.5877, 1.4320, 2.0673], device='cuda:1'), covar=tensor([0.2077, 0.2008, 0.1845, 0.2092], device='cuda:1'), in_proj_covar=tensor([0.1069, 0.0848, 0.0951, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:38:05,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0998, 1.1847, 1.2988, 1.2102], device='cuda:1'), covar=tensor([0.1069, 0.1105, 0.1598, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0767, 0.0635, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:38:13,718 INFO [train.py:968] (1/2) Epoch 4, batch 2550, giga_loss[loss=0.2424, simple_loss=0.3168, pruned_loss=0.08399, over 28970.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3432, pruned_loss=0.1054, over 5704189.42 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3885, pruned_loss=0.126, over 4169301.30 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3386, pruned_loss=0.1035, over 5711844.30 frames. ], batch size: 213, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:38:33,333 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 4, batch 2600, giga_loss[loss=0.2497, simple_loss=0.3152, pruned_loss=0.09215, over 28750.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3401, pruned_loss=0.1035, over 5706706.51 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3883, pruned_loss=0.1259, over 4177927.80 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3363, pruned_loss=0.1021, over 5711877.41 frames. ], batch size: 99, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:39:00,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6948, 3.4956, 3.4329, 1.6004], device='cuda:1'), covar=tensor([0.0583, 0.0442, 0.0789, 0.2147], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0635, 0.0778, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 00:39:34,000 INFO [train.py:968] (1/2) Epoch 4, batch 2650, giga_loss[loss=0.2756, simple_loss=0.3406, pruned_loss=0.1053, over 28744.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3394, pruned_loss=0.1028, over 5717229.17 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3883, pruned_loss=0.1256, over 4235212.60 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3349, pruned_loss=0.1011, over 5717497.88 frames. ], batch size: 284, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:39:43,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:39:45,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:39:45,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=138806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:39:47,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=138809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:40:11,994 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=138838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:40:15,146 INFO [optim.py:369] (1/2) 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 00:40:18,082 INFO [train.py:968] (1/2) Epoch 4, batch 2700, giga_loss[loss=0.305, simple_loss=0.3715, pruned_loss=0.1193, over 28837.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3439, pruned_loss=0.1061, over 5719778.59 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3877, pruned_loss=0.1251, over 4260077.53 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3401, pruned_loss=0.1047, over 5717499.53 frames. ], batch size: 186, lr: 8.27e-03, grad_scale: 4.0
+2023-03-02 00:40:37,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0827, 1.3622, 1.1828, 0.9109], device='cuda:1'), covar=tensor([0.2288, 0.2066, 0.1989, 0.1780], device='cuda:1'), in_proj_covar=tensor([0.1063, 0.0851, 0.0947, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:41:01,487 INFO [train.py:968] (1/2) Epoch 4, batch 2750, giga_loss[loss=0.4265, simple_loss=0.4513, pruned_loss=0.2009, over 26592.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3508, pruned_loss=0.1107, over 5719096.68 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3877, pruned_loss=0.1252, over 4324204.81 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3462, pruned_loss=0.109, over 5711016.00 frames. ], batch size: 555, lr: 8.26e-03, grad_scale: 4.0
+2023-03-02 00:41:07,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-02 00:41:44,420 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 2800, giga_loss[loss=0.3271, simple_loss=0.3885, pruned_loss=0.1328, over 28973.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3596, pruned_loss=0.1164, over 5720866.72 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3878, pruned_loss=0.1252, over 4369632.52 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3549, pruned_loss=0.1147, over 5710637.03 frames. ], batch size: 145, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:42:32,735 INFO [train.py:968] (1/2) Epoch 4, batch 2850, giga_loss[loss=0.32, simple_loss=0.3763, pruned_loss=0.1319, over 28759.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.367, pruned_loss=0.1219, over 5705817.18 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3868, pruned_loss=0.1246, over 4392700.83 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3637, pruned_loss=0.1208, over 5695245.06 frames. ], batch size: 92, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:43:18,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:43:19,406 INFO [optim.py:369] (1/2) 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 00:43:24,981 INFO [train.py:968] (1/2) Epoch 4, batch 2900, giga_loss[loss=0.331, simple_loss=0.3871, pruned_loss=0.1375, over 27608.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3721, pruned_loss=0.1234, over 5712242.85 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3868, pruned_loss=0.1247, over 4399889.32 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3693, pruned_loss=0.1226, over 5703394.64 frames. ], batch size: 472, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:43:51,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:43:55,023 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:44:03,846 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:44:03,880 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:44:05,543 INFO [train.py:968] (1/2) Epoch 4, batch 2950, giga_loss[loss=0.3414, simple_loss=0.4043, pruned_loss=0.1393, over 28596.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3775, pruned_loss=0.1262, over 5710621.97 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.387, pruned_loss=0.1248, over 4429627.81 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.375, pruned_loss=0.1254, over 5700195.58 frames. ], batch size: 307, lr: 8.26e-03, grad_scale: 8.0
+2023-03-02 00:44:51,101 INFO [optim.py:369] (1/2) 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 00:44:53,493 INFO [train.py:968] (1/2) Epoch 4, batch 3000, giga_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1257, over 28559.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3846, pruned_loss=0.132, over 5686557.03 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3868, pruned_loss=0.1246, over 4465829.38 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3827, pruned_loss=0.1317, over 5673599.08 frames. ], batch size: 336, lr: 8.26e-03, grad_scale: 4.0
+2023-03-02 00:44:53,493 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 00:45:01,935 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 00:45:05,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7955, 1.0748, 3.7762, 3.1069], device='cuda:1'), covar=tensor([0.1709, 0.2239, 0.0330, 0.0600], device='cuda:1'), in_proj_covar=tensor([0.0538, 0.0502, 0.0697, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:45:32,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:45:34,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:45:41,625 INFO [train.py:968] (1/2) Epoch 4, batch 3050, libri_loss[loss=0.3353, simple_loss=0.402, pruned_loss=0.1343, over 29230.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3795, pruned_loss=0.1282, over 5700307.89 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.386, pruned_loss=0.1244, over 4515394.96 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3784, pruned_loss=0.1283, over 5683441.59 frames. ], batch size: 94, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:45:55,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6071, 1.6271, 1.1928, 1.3498], device='cuda:1'), covar=tensor([0.0662, 0.0482, 0.0852, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0473, 0.0521, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 00:46:00,078 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:46:15,736 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:46:20,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:46:23,888 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 3100, giga_loss[loss=0.3147, simple_loss=0.3781, pruned_loss=0.1256, over 28371.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3766, pruned_loss=0.1252, over 5701352.52 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3863, pruned_loss=0.1245, over 4553639.09 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3751, pruned_loss=0.1252, over 5689884.54 frames. ], batch size: 78, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:46:40,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8229, 1.0684, 3.9107, 3.2330], device='cuda:1'), covar=tensor([0.1709, 0.2228, 0.0333, 0.0631], device='cuda:1'), in_proj_covar=tensor([0.0528, 0.0494, 0.0687, 0.0555], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0006, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 00:46:44,548 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 3150, giga_loss[loss=0.3142, simple_loss=0.3762, pruned_loss=0.1261, over 29001.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3746, pruned_loss=0.1232, over 5709768.14 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3859, pruned_loss=0.1243, over 4566746.81 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3737, pruned_loss=0.1234, over 5699308.34 frames. ], batch size: 106, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:47:17,878 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:47:53,297 INFO [train.py:968] (1/2) Epoch 4, batch 3200, giga_loss[loss=0.3164, simple_loss=0.3801, pruned_loss=0.1263, over 28623.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3759, pruned_loss=0.1235, over 5711046.02 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3856, pruned_loss=0.1241, over 4586520.86 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3752, pruned_loss=0.1237, over 5700283.40 frames. ], batch size: 85, lr: 8.25e-03, grad_scale: 8.0
+2023-03-02 00:47:55,044 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 4, batch 3250, giga_loss[loss=0.2798, simple_loss=0.3588, pruned_loss=0.1004, over 29051.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3781, pruned_loss=0.1243, over 5715474.52 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3852, pruned_loss=0.1238, over 4618872.96 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3777, pruned_loss=0.1248, over 5702908.52 frames. ], batch size: 155, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:49:18,349 INFO [optim.py:369] (1/2) 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 00:49:19,053 INFO [train.py:968] (1/2) Epoch 4, batch 3300, giga_loss[loss=0.3095, simple_loss=0.3737, pruned_loss=0.1227, over 28634.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3796, pruned_loss=0.1256, over 5709056.98 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.385, pruned_loss=0.1236, over 4638096.47 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3793, pruned_loss=0.1261, over 5696761.68 frames. ], batch size: 78, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:49:25,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:49:29,115 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:50:03,684 INFO [train.py:968] (1/2) Epoch 4, batch 3350, giga_loss[loss=0.3234, simple_loss=0.3879, pruned_loss=0.1295, over 28936.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3812, pruned_loss=0.1272, over 5708422.73 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3854, pruned_loss=0.1237, over 4656396.00 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3807, pruned_loss=0.1276, over 5696815.59 frames. ], batch size: 227, lr: 8.25e-03, grad_scale: 4.0
+2023-03-02 00:50:27,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8098, 1.7371, 1.5906, 1.7362], device='cuda:1'), covar=tensor([0.0996, 0.1511, 0.1301, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0757, 0.0627, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:50:45,953 INFO [optim.py:369] (1/2) 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 00:50:46,699 INFO [train.py:968] (1/2) Epoch 4, batch 3400, giga_loss[loss=0.3134, simple_loss=0.3789, pruned_loss=0.1239, over 28691.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3809, pruned_loss=0.1269, over 5716372.06 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3855, pruned_loss=0.1236, over 4671423.26 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3803, pruned_loss=0.1273, over 5707840.29 frames. ], batch size: 242, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:51:24,919 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 3450, giga_loss[loss=0.3184, simple_loss=0.3859, pruned_loss=0.1255, over 28897.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3819, pruned_loss=0.1273, over 5720917.52 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3858, pruned_loss=0.1238, over 4722533.05 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.381, pruned_loss=0.1276, over 5713705.87 frames. ], batch size: 199, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:51:27,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:29,136 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:32,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:39,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:42,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:51,449 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:55,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:51:57,493 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:04,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:05,828 INFO [optim.py:369] (1/2) 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 00:52:06,458 INFO [train.py:968] (1/2) Epoch 4, batch 3500, giga_loss[loss=0.283, simple_loss=0.3594, pruned_loss=0.1033, over 28955.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3829, pruned_loss=0.1275, over 5711636.61 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.386, pruned_loss=0.1241, over 4746273.31 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3819, pruned_loss=0.1276, over 5710446.64 frames. ], batch size: 112, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:52:16,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-02 00:52:21,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3049, 1.2116, 1.0177, 1.4691], device='cuda:1'), covar=tensor([0.0883, 0.0384, 0.0387, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0140, 0.0142, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0046, 0.0035, 0.0030, 0.0052], device='cuda:1')
+2023-03-02 00:52:29,365 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:36,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=139677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:52:39,731 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 3550, giga_loss[loss=0.2856, simple_loss=0.3623, pruned_loss=0.1045, over 28860.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3827, pruned_loss=0.126, over 5715725.93 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3863, pruned_loss=0.1242, over 4768510.59 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3816, pruned_loss=0.126, over 5711893.02 frames. ], batch size: 99, lr: 8.24e-03, grad_scale: 4.0
+2023-03-02 00:53:30,833 INFO [optim.py:369] (1/2) 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 00:53:31,584 INFO [train.py:968] (1/2) Epoch 4, batch 3600, giga_loss[loss=0.2967, simple_loss=0.3732, pruned_loss=0.1101, over 28683.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3834, pruned_loss=0.126, over 5721148.81 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3859, pruned_loss=0.1241, over 4799253.20 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3828, pruned_loss=0.1262, over 5715667.37 frames. ], batch size: 85, lr: 8.24e-03, grad_scale: 8.0
+2023-03-02 00:54:09,023 INFO [train.py:968] (1/2) Epoch 4, batch 3650, giga_loss[loss=0.3073, simple_loss=0.3799, pruned_loss=0.1173, over 28859.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3807, pruned_loss=0.1242, over 5725957.34 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3854, pruned_loss=0.1237, over 4832370.37 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3805, pruned_loss=0.1247, over 5716214.69 frames. ], batch size: 145, lr: 8.24e-03, grad_scale: 8.0
+2023-03-02 00:54:33,609 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=139820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:54:37,336 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=139823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:54:52,701 INFO [optim.py:369] (1/2) 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 00:54:54,356 INFO [train.py:968] (1/2) Epoch 4, batch 3700, giga_loss[loss=0.3024, simple_loss=0.3668, pruned_loss=0.119, over 27960.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3777, pruned_loss=0.1228, over 5722433.43 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3849, pruned_loss=0.1234, over 4853189.60 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3778, pruned_loss=0.1234, over 5711992.46 frames. ], batch size: 412, lr: 8.24e-03, grad_scale: 8.0
+2023-03-02 00:55:02,544 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=139852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:55:32,322 INFO [train.py:968] (1/2) Epoch 4, batch 3750, libri_loss[loss=0.2906, simple_loss=0.3642, pruned_loss=0.1085, over 29575.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3745, pruned_loss=0.1206, over 5729510.43 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3844, pruned_loss=0.1232, over 4879220.77 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3747, pruned_loss=0.1211, over 5717006.98 frames. ], batch size: 77, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:56:15,181 INFO [optim.py:369] (1/2) 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 00:56:15,807 INFO [train.py:968] (1/2) Epoch 4, batch 3800, giga_loss[loss=0.3199, simple_loss=0.3851, pruned_loss=0.1274, over 29039.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.375, pruned_loss=0.1207, over 5737138.53 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3844, pruned_loss=0.1232, over 4889325.12 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3751, pruned_loss=0.1211, over 5725876.78 frames. ], batch size: 128, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:56:22,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5969, 2.2336, 1.6244, 0.7477], device='cuda:1'), covar=tensor([0.1787, 0.0899, 0.1626, 0.2218], device='cuda:1'), in_proj_covar=tensor([0.1261, 0.1184, 0.1276, 0.1094], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 00:56:25,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8160, 2.3646, 2.2364, 2.3252], device='cuda:1'), covar=tensor([0.0894, 0.1540, 0.1271, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0761, 0.0637, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 00:56:34,179 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:56:41,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=139975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:56:54,354 INFO [train.py:968] (1/2) Epoch 4, batch 3850, giga_loss[loss=0.2842, simple_loss=0.3558, pruned_loss=0.1063, over 28864.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3772, pruned_loss=0.1227, over 5738437.00 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3841, pruned_loss=0.123, over 4933846.75 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3772, pruned_loss=0.1231, over 5722712.37 frames. ], batch size: 99, lr: 8.23e-03, grad_scale: 4.0
+2023-03-02 00:57:03,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 00:57:05,365 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:57:33,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4197, 1.4289, 1.2027, 1.5405], device='cuda:1'), covar=tensor([0.0847, 0.0338, 0.0356, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0137, 0.0141, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:1')
+2023-03-02 00:57:34,687 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 3900, giga_loss[loss=0.3045, simple_loss=0.3796, pruned_loss=0.1147, over 28524.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3782, pruned_loss=0.1227, over 5732264.75 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.384, pruned_loss=0.123, over 4957953.58 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3781, pruned_loss=0.1231, over 5719509.58 frames. ], batch size: 60, lr: 8.23e-03, grad_scale: 4.0
+2023-03-02 00:57:35,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:57:45,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:58:16,591 INFO [train.py:968] (1/2) Epoch 4, batch 3950, giga_loss[loss=0.2771, simple_loss=0.3563, pruned_loss=0.09899, over 28480.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.377, pruned_loss=0.1212, over 5729160.38 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3835, pruned_loss=0.1226, over 4972230.49 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3772, pruned_loss=0.1218, over 5716769.26 frames. ], batch size: 78, lr: 8.23e-03, grad_scale: 4.0
+2023-03-02 00:58:20,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-02 00:58:55,408 INFO [optim.py:369] (1/2) 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 00:58:55,420 INFO [train.py:968] (1/2) Epoch 4, batch 4000, giga_loss[loss=0.3141, simple_loss=0.3795, pruned_loss=0.1244, over 28823.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3764, pruned_loss=0.1209, over 5736365.99 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3837, pruned_loss=0.1226, over 5013587.90 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3761, pruned_loss=0.1213, over 5718812.76 frames. ], batch size: 199, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:59:02,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:04,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:13,107 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:28,073 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:28,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:33,694 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:35,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:37,462 INFO [train.py:968] (1/2) Epoch 4, batch 4050, giga_loss[loss=0.2802, simple_loss=0.3454, pruned_loss=0.1076, over 28854.00 frames. ], tot_loss[loss=0.307, simple_loss=0.374, pruned_loss=0.12, over 5724817.48 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3833, pruned_loss=0.1223, over 5025293.03 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3739, pruned_loss=0.1205, over 5710122.31 frames. ], batch size: 112, lr: 8.23e-03, grad_scale: 8.0
+2023-03-02 00:59:37,852 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-02 00:59:40,668 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:43,050 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:43,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:45,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 00:59:58,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:00:08,081 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:00:18,176 INFO [train.py:968] (1/2) Epoch 4, batch 4100, giga_loss[loss=0.2665, simple_loss=0.3396, pruned_loss=0.09668, over 28661.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3721, pruned_loss=0.1191, over 5721004.04 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3839, pruned_loss=0.1226, over 5049805.42 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3713, pruned_loss=0.1192, over 5705801.05 frames. ], batch size: 307, lr: 8.22e-03, grad_scale: 8.0
+2023-03-02 01:00:18,360 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:00:20,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 01:00:32,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6225, 3.0386, 1.5852, 1.5970], device='cuda:1'), covar=tensor([0.0752, 0.0314, 0.0752, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0441, 0.0303, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0021, 0.0014, 0.0019], device='cuda:1')
+2023-03-02 01:00:56,787 INFO [train.py:968] (1/2) Epoch 4, batch 4150, giga_loss[loss=0.306, simple_loss=0.3631, pruned_loss=0.1245, over 28622.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.371, pruned_loss=0.1192, over 5717886.18 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3839, pruned_loss=0.1227, over 5074584.01 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.37, pruned_loss=0.1191, over 5701276.35 frames. ], batch size: 85, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:01:04,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 01:01:35,341 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 4200, giga_loss[loss=0.2683, simple_loss=0.3451, pruned_loss=0.09571, over 28700.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3707, pruned_loss=0.1202, over 5715894.02 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3838, pruned_loss=0.1226, over 5082793.19 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3699, pruned_loss=0.1201, over 5701352.30 frames. ], batch size: 60, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:01:38,624 INFO [optim.py:369] (1/2) 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 01:01:43,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:02:17,963 INFO [train.py:968] (1/2) Epoch 4, batch 4250, giga_loss[loss=0.3101, simple_loss=0.3685, pruned_loss=0.1259, over 28675.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3694, pruned_loss=0.12, over 5717139.47 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3837, pruned_loss=0.1227, over 5113721.75 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3684, pruned_loss=0.1199, over 5699819.76 frames. ], batch size: 92, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:02:32,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-02 01:02:59,425 INFO [train.py:968] (1/2) Epoch 4, batch 4300, giga_loss[loss=0.2882, simple_loss=0.3539, pruned_loss=0.1112, over 28885.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.367, pruned_loss=0.1192, over 5707408.83 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3833, pruned_loss=0.1227, over 5122399.72 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.119, over 5705097.13 frames. ], batch size: 213, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:03:00,140 INFO [optim.py:369] (1/2) 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 01:03:20,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3384, 2.1262, 1.6132, 0.6675], device='cuda:1'), covar=tensor([0.2128, 0.1144, 0.1911, 0.2432], device='cuda:1'), in_proj_covar=tensor([0.1285, 0.1219, 0.1319, 0.1106], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 01:03:30,269 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:32,247 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:33,011 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-02 01:03:38,793 INFO [train.py:968] (1/2) Epoch 4, batch 4350, giga_loss[loss=0.3218, simple_loss=0.3766, pruned_loss=0.1335, over 28784.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3653, pruned_loss=0.1187, over 5712672.35 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3825, pruned_loss=0.1222, over 5159879.67 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3643, pruned_loss=0.1187, over 5702865.83 frames. ], batch size: 284, lr: 8.22e-03, grad_scale: 4.0
+2023-03-02 01:03:39,143 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:40,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:41,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:46,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:03:54,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:03,362 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:12,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:13,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-02 01:04:16,247 INFO [train.py:968] (1/2) Epoch 4, batch 4400, giga_loss[loss=0.2935, simple_loss=0.3638, pruned_loss=0.1116, over 29056.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3623, pruned_loss=0.1168, over 5713734.13 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3831, pruned_loss=0.1225, over 5171365.44 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3607, pruned_loss=0.1165, over 5707894.93 frames. ], batch size: 136, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:04:17,768 INFO [optim.py:369] (1/2) 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 01:04:19,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9126, 1.9060, 1.2848, 1.5609], device='cuda:1'), covar=tensor([0.0551, 0.0490, 0.0908, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0470, 0.0514, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 01:04:29,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:36,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2723, 1.7165, 1.2216, 1.5295], device='cuda:1'), covar=tensor([0.0685, 0.0276, 0.0345, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0137, 0.0141, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0052], device='cuda:1')
+2023-03-02 01:04:40,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:42,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:04:58,531 INFO [train.py:968] (1/2) Epoch 4, batch 4450, libri_loss[loss=0.2984, simple_loss=0.3566, pruned_loss=0.1201, over 29640.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3614, pruned_loss=0.1161, over 5718801.22 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3828, pruned_loss=0.1223, over 5186026.51 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3599, pruned_loss=0.1159, over 5710689.34 frames. ], batch size: 69, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:05:11,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9909, 1.8035, 1.3839, 1.6514], device='cuda:1'), covar=tensor([0.0617, 0.0692, 0.0979, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0473, 0.0518, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 01:05:22,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:05:44,901 INFO [train.py:968] (1/2) Epoch 4, batch 4500, giga_loss[loss=0.2696, simple_loss=0.3372, pruned_loss=0.101, over 28490.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3645, pruned_loss=0.1177, over 5710443.57 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3828, pruned_loss=0.1223, over 5192964.70 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3632, pruned_loss=0.1174, over 5702490.15 frames. ], batch size: 85, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:05:46,573 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:16,654 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:18,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:20,143 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 01:06:26,144 INFO [train.py:968] (1/2) Epoch 4, batch 4550, giga_loss[loss=0.2998, simple_loss=0.3759, pruned_loss=0.1118, over 28701.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3677, pruned_loss=0.1193, over 5707866.44 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3828, pruned_loss=0.1224, over 5207879.80 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3663, pruned_loss=0.1189, over 5699435.40 frames. ], batch size: 242, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:06:32,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:35,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:41,851 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:46,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:47,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:48,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:50,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:06:59,914 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:10,302 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 01:07:10,550 INFO [train.py:968] (1/2) Epoch 4, batch 4600, giga_loss[loss=0.2933, simple_loss=0.3681, pruned_loss=0.1093, over 28622.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3701, pruned_loss=0.1202, over 5704905.01 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3831, pruned_loss=0.1226, over 5214815.35 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3686, pruned_loss=0.1197, over 5696557.63 frames. ], batch size: 262, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:07:11,155 INFO [optim.py:369] (1/2) 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 01:07:13,754 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:15,340 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:25,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=140761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:27,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=140764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:44,398 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:54,220 INFO [train.py:968] (1/2) Epoch 4, batch 4650, libri_loss[loss=0.3459, simple_loss=0.4006, pruned_loss=0.1456, over 29512.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3718, pruned_loss=0.1209, over 5698182.83 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3836, pruned_loss=0.1231, over 5231132.85 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3699, pruned_loss=0.1201, over 5687677.63 frames. ], batch size: 80, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:07:54,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=140793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:07:57,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5063, 1.5765, 1.4359, 1.6292], device='cuda:1'), covar=tensor([0.1867, 0.1861, 0.1785, 0.2000], device='cuda:1'), in_proj_covar=tensor([0.1050, 0.0843, 0.0939, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 01:08:09,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1859, 0.9393, 0.8847, 1.2906], device='cuda:1'), covar=tensor([0.0781, 0.0340, 0.0393, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0137, 0.0139, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0046, 0.0034, 0.0030, 0.0051], device='cuda:1')
+2023-03-02 01:08:22,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6834, 4.3152, 4.3903, 2.0312], device='cuda:1'), covar=tensor([0.0440, 0.0397, 0.0799, 0.1929], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0622, 0.0771, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 01:08:29,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 01:08:34,479 INFO [train.py:968] (1/2) Epoch 4, batch 4700, giga_loss[loss=0.2828, simple_loss=0.3413, pruned_loss=0.1122, over 28647.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3706, pruned_loss=0.1205, over 5709491.15 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3834, pruned_loss=0.123, over 5244442.10 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.369, pruned_loss=0.1198, over 5697612.22 frames. ], batch size: 85, lr: 8.21e-03, grad_scale: 8.0
+2023-03-02 01:08:35,197 INFO [optim.py:369] (1/2) 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 01:08:57,050 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:02,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=140878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:06,114 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:14,474 INFO [train.py:968] (1/2) Epoch 4, batch 4750, giga_loss[loss=0.2811, simple_loss=0.3558, pruned_loss=0.1032, over 29015.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3718, pruned_loss=0.1212, over 5708520.19 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3837, pruned_loss=0.1231, over 5270002.26 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3697, pruned_loss=0.1205, over 5694304.97 frames. ], batch size: 155, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:09:21,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=140902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:09:55,945 INFO [train.py:968] (1/2) Epoch 4, batch 4800, giga_loss[loss=0.3361, simple_loss=0.3826, pruned_loss=0.1448, over 28432.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3736, pruned_loss=0.123, over 5710694.44 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3842, pruned_loss=0.1236, over 5288363.09 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3712, pruned_loss=0.122, over 5694448.44 frames. ], batch size: 71, lr: 8.20e-03, grad_scale: 8.0
+2023-03-02 01:09:57,860 INFO [optim.py:369] (1/2) 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 01:10:16,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-02 01:10:39,072 INFO [train.py:968] (1/2) Epoch 4, batch 4850, giga_loss[loss=0.3096, simple_loss=0.3798, pruned_loss=0.1197, over 28872.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3765, pruned_loss=0.1245, over 5706740.40 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3847, pruned_loss=0.1238, over 5299899.51 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.374, pruned_loss=0.1236, over 5690954.20 frames. ], batch size: 145, lr: 8.20e-03, grad_scale: 8.0
+2023-03-02 01:10:55,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:10:58,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:02,296 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:06,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:09,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:21,653 INFO [train.py:968] (1/2) Epoch 4, batch 4900, giga_loss[loss=0.3404, simple_loss=0.4018, pruned_loss=0.1395, over 28558.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.378, pruned_loss=0.1245, over 5704361.17 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3852, pruned_loss=0.1241, over 5298685.84 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3755, pruned_loss=0.1235, over 5698940.62 frames. ], batch size: 336, lr: 8.20e-03, grad_scale: 8.0
+2023-03-02 01:11:23,689 INFO [optim.py:369] (1/2) 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 01:11:24,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:29,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:11:39,668 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-02 01:12:01,384 INFO [train.py:968] (1/2) Epoch 4, batch 4950, giga_loss[loss=0.313, simple_loss=0.3746, pruned_loss=0.1257, over 28148.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3799, pruned_loss=0.1255, over 5698265.80 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3863, pruned_loss=0.1247, over 5303013.68 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3769, pruned_loss=0.1242, over 5700702.15 frames. ], batch size: 77, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:12:21,466 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:12:40,914 INFO [train.py:968] (1/2) Epoch 4, batch 5000, giga_loss[loss=0.2909, simple_loss=0.3684, pruned_loss=0.1067, over 28859.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3802, pruned_loss=0.125, over 5697391.46 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.387, pruned_loss=0.125, over 5311926.38 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3769, pruned_loss=0.1237, over 5705536.74 frames. ], batch size: 174, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:12:42,910 INFO [optim.py:369] (1/2) 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 01:12:49,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:02,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:04,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:18,767 INFO [train.py:968] (1/2) Epoch 4, batch 5050, giga_loss[loss=0.3321, simple_loss=0.386, pruned_loss=0.1391, over 28421.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3796, pruned_loss=0.1246, over 5712548.36 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3871, pruned_loss=0.1253, over 5332488.75 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3767, pruned_loss=0.1233, over 5712641.48 frames. ], batch size: 71, lr: 8.20e-03, grad_scale: 4.0
+2023-03-02 01:13:25,892 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:13:25,906 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:01,607 INFO [train.py:968] (1/2) Epoch 4, batch 5100, giga_loss[loss=0.294, simple_loss=0.3624, pruned_loss=0.1128, over 28884.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3793, pruned_loss=0.1248, over 5717852.92 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3874, pruned_loss=0.1254, over 5338251.41 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3766, pruned_loss=0.1237, over 5716355.48 frames. ], batch size: 186, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:14:04,249 INFO [optim.py:369] (1/2) 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 01:14:12,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:29,149 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:39,967 INFO [train.py:968] (1/2) Epoch 4, batch 5150, giga_loss[loss=0.2676, simple_loss=0.3418, pruned_loss=0.09674, over 28895.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3769, pruned_loss=0.1233, over 5699112.69 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3879, pruned_loss=0.1258, over 5336646.57 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3741, pruned_loss=0.122, over 5713523.53 frames. ], batch size: 174, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:14:44,862 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:49,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:14:56,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:15:01,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7300, 1.8831, 1.3949, 1.0210], device='cuda:1'), covar=tensor([0.0881, 0.0573, 0.0539, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.1267, 0.1002, 0.1042, 0.1111], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 01:15:13,558 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:15:20,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:15:23,514 INFO [train.py:968] (1/2) Epoch 4, batch 5200, giga_loss[loss=0.2423, simple_loss=0.3175, pruned_loss=0.0835, over 28406.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3739, pruned_loss=0.1218, over 5701726.10 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3882, pruned_loss=0.126, over 5339724.04 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3712, pruned_loss=0.1206, over 5715509.86 frames. ], batch size: 71, lr: 8.19e-03, grad_scale: 8.0
+2023-03-02 01:15:25,517 INFO [optim.py:369] (1/2) 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 01:16:04,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9196, 1.3647, 3.4016, 2.9071], device='cuda:1'), covar=tensor([0.1461, 0.1883, 0.0400, 0.0553], device='cuda:1'), in_proj_covar=tensor([0.0534, 0.0504, 0.0695, 0.0568], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 01:16:05,243 INFO [train.py:968] (1/2) Epoch 4, batch 5250, giga_loss[loss=0.3225, simple_loss=0.3779, pruned_loss=0.1336, over 28985.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3721, pruned_loss=0.1208, over 5703706.16 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3882, pruned_loss=0.126, over 5343793.89 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3698, pruned_loss=0.1198, over 5714098.79 frames. ], batch size: 136, lr: 8.19e-03, grad_scale: 8.0
+2023-03-02 01:16:12,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:14,405 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:27,307 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:31,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:16:38,569 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 5300, giga_loss[loss=0.2939, simple_loss=0.3794, pruned_loss=0.1042, over 29020.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3734, pruned_loss=0.1203, over 5707864.39 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3882, pruned_loss=0.126, over 5360822.58 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3712, pruned_loss=0.1194, over 5710502.11 frames. ], batch size: 155, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:16:49,704 INFO [optim.py:369] (1/2) 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 01:16:55,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:17:29,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:17:30,165 INFO [train.py:968] (1/2) Epoch 4, batch 5350, giga_loss[loss=0.3097, simple_loss=0.3745, pruned_loss=0.1225, over 29084.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3754, pruned_loss=0.1211, over 5709655.24 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3886, pruned_loss=0.1264, over 5375778.58 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3728, pruned_loss=0.1198, over 5707343.87 frames. ], batch size: 128, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:18:06,959 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:18:11,493 INFO [train.py:968] (1/2) Epoch 4, batch 5400, giga_loss[loss=0.3623, simple_loss=0.4165, pruned_loss=0.154, over 28060.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.375, pruned_loss=0.1219, over 5704996.92 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3885, pruned_loss=0.1263, over 5380015.58 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.373, pruned_loss=0.121, over 5702356.83 frames. ], batch size: 412, lr: 8.19e-03, grad_scale: 4.0
+2023-03-02 01:18:15,430 INFO [optim.py:369] (1/2) 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 01:18:38,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 01:18:42,695 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 5450, giga_loss[loss=0.3634, simple_loss=0.4098, pruned_loss=0.1585, over 28683.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3737, pruned_loss=0.1226, over 5707654.94 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3888, pruned_loss=0.1265, over 5390122.57 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3716, pruned_loss=0.1215, over 5702199.59 frames. ], batch size: 262, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:19:21,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3925, 1.4551, 1.2467, 1.6139], device='cuda:1'), covar=tensor([0.1990, 0.1957, 0.1918, 0.2156], device='cuda:1'), in_proj_covar=tensor([0.1046, 0.0832, 0.0933, 0.0934], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 01:19:27,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:19:29,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:19:33,530 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-02 01:19:33,672 INFO [train.py:968] (1/2) Epoch 4, batch 5500, giga_loss[loss=0.2586, simple_loss=0.3304, pruned_loss=0.09342, over 29011.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3719, pruned_loss=0.1229, over 5707210.67 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3888, pruned_loss=0.1265, over 5400283.57 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3699, pruned_loss=0.1221, over 5701025.66 frames. ], batch size: 145, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:19:37,211 INFO [optim.py:369] (1/2) 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 01:19:54,637 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:08,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:16,488 INFO [train.py:968] (1/2) Epoch 4, batch 5550, giga_loss[loss=0.2761, simple_loss=0.3468, pruned_loss=0.1027, over 28862.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3689, pruned_loss=0.1218, over 5709456.38 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3885, pruned_loss=0.1263, over 5407395.82 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3673, pruned_loss=0.1212, over 5702441.66 frames. ], batch size: 112, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:20:36,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:41,096 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:44,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:20:44,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:21:01,699 INFO [train.py:968] (1/2) Epoch 4, batch 5600, giga_loss[loss=0.2742, simple_loss=0.3457, pruned_loss=0.1014, over 29006.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3672, pruned_loss=0.1209, over 5704200.34 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.389, pruned_loss=0.1268, over 5401906.23 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3653, pruned_loss=0.12, over 5705507.35 frames. ], batch size: 145, lr: 8.18e-03, grad_scale: 8.0
+2023-03-02 01:21:05,255 INFO [optim.py:369] (1/2) 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 01:21:11,089 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:21:36,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 01:21:43,093 INFO [train.py:968] (1/2) Epoch 4, batch 5650, giga_loss[loss=0.2368, simple_loss=0.3037, pruned_loss=0.08497, over 28470.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3659, pruned_loss=0.1206, over 5708794.26 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.389, pruned_loss=0.1269, over 5411484.80 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3639, pruned_loss=0.1197, over 5709264.73 frames. ], batch size: 71, lr: 8.18e-03, grad_scale: 8.0
+2023-03-02 01:21:56,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141810.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 01:22:11,074 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:13,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:15,503 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=141834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:23,667 INFO [train.py:968] (1/2) Epoch 4, batch 5700, giga_loss[loss=0.2461, simple_loss=0.3063, pruned_loss=0.09293, over 28282.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3603, pruned_loss=0.1171, over 5719401.61 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3895, pruned_loss=0.1271, over 5423604.26 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3577, pruned_loss=0.116, over 5715614.05 frames. ], batch size: 77, lr: 8.18e-03, grad_scale: 8.0
+2023-03-02 01:22:27,327 INFO [optim.py:369] (1/2) 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 01:22:36,328 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=141858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:37,520 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:22:38,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=141861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:23:01,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=141890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:23:02,752 INFO [train.py:968] (1/2) Epoch 4, batch 5750, giga_loss[loss=0.2942, simple_loss=0.3623, pruned_loss=0.113, over 28731.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3576, pruned_loss=0.1153, over 5722792.86 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3893, pruned_loss=0.127, over 5436500.06 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3548, pruned_loss=0.1142, over 5716026.30 frames. ], batch size: 307, lr: 8.18e-03, grad_scale: 4.0
+2023-03-02 01:23:03,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 01:23:20,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=141911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:23:34,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5983, 3.3646, 3.3277, 1.8322], device='cuda:1'), covar=tensor([0.0601, 0.0482, 0.0798, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0640, 0.0804, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 01:23:38,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9719, 3.7549, 3.6348, 1.8170], device='cuda:1'), covar=tensor([0.0474, 0.0410, 0.0743, 0.1957], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0637, 0.0799, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 01:23:43,965 INFO [train.py:968] (1/2) Epoch 4, batch 5800, giga_loss[loss=0.3044, simple_loss=0.3635, pruned_loss=0.1227, over 29057.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3598, pruned_loss=0.1163, over 5723933.81 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3898, pruned_loss=0.1274, over 5440745.38 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3569, pruned_loss=0.1151, over 5717326.42 frames. ], batch size: 128, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:23:48,686 INFO [optim.py:369] (1/2) 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 01:24:24,722 INFO [train.py:968] (1/2) Epoch 4, batch 5850, giga_loss[loss=0.2686, simple_loss=0.3449, pruned_loss=0.09622, over 28979.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.364, pruned_loss=0.1182, over 5726362.41 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3897, pruned_loss=0.1273, over 5454747.23 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.361, pruned_loss=0.1169, over 5716080.26 frames. ], batch size: 164, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:25:03,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8077, 1.2410, 3.9618, 3.2037], device='cuda:1'), covar=tensor([0.1680, 0.2024, 0.0349, 0.0551], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0508, 0.0707, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 01:25:08,837 INFO [train.py:968] (1/2) Epoch 4, batch 5900, giga_loss[loss=0.3381, simple_loss=0.396, pruned_loss=0.1402, over 28876.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1196, over 5723773.48 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3895, pruned_loss=0.1272, over 5457060.49 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3651, pruned_loss=0.1187, over 5714914.64 frames. ], batch size: 112, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:25:13,467 INFO [optim.py:369] (1/2) 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 01:25:16,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1244, 1.4631, 1.1721, 0.3937], device='cuda:1'), covar=tensor([0.1100, 0.0716, 0.1032, 0.1743], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.1243, 0.1333, 0.1131], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 01:25:20,135 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:25:22,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:25:49,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:25:57,249 INFO [train.py:968] (1/2) Epoch 4, batch 5950, giga_loss[loss=0.2959, simple_loss=0.3601, pruned_loss=0.1159, over 28225.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5719187.91 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3895, pruned_loss=0.1272, over 5457060.49 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5712292.96 frames. ], batch size: 77, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:26:00,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:26:27,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8932, 3.1986, 2.2662, 1.0067], device='cuda:1'), covar=tensor([0.2886, 0.0935, 0.1495, 0.2567], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.1238, 0.1331, 0.1120], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 01:26:41,859 INFO [train.py:968] (1/2) Epoch 4, batch 6000, giga_loss[loss=0.3723, simple_loss=0.4228, pruned_loss=0.1609, over 28929.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3742, pruned_loss=0.1234, over 5714909.97 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3902, pruned_loss=0.1277, over 5465119.84 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3716, pruned_loss=0.1223, over 5708881.86 frames. ], batch size: 199, lr: 8.17e-03, grad_scale: 8.0
+2023-03-02 01:26:41,859 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 01:26:50,247 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 01:26:50,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6725, 2.0042, 1.8434, 1.7224], device='cuda:1'), covar=tensor([0.1452, 0.1663, 0.1145, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0765, 0.0754, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 01:26:53,718 INFO [optim.py:369] (1/2) 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 01:26:55,747 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 01:27:26,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142185.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 01:27:34,114 INFO [train.py:968] (1/2) Epoch 4, batch 6050, giga_loss[loss=0.3651, simple_loss=0.4148, pruned_loss=0.1577, over 28888.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3812, pruned_loss=0.1302, over 5708700.84 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.39, pruned_loss=0.1275, over 5471300.25 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3792, pruned_loss=0.1294, over 5701811.98 frames. ], batch size: 174, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:27:50,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:28:20,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:28:22,534 INFO [train.py:968] (1/2) Epoch 4, batch 6100, giga_loss[loss=0.3964, simple_loss=0.4407, pruned_loss=0.176, over 28961.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.387, pruned_loss=0.1353, over 5702958.41 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3895, pruned_loss=0.1272, over 5480844.73 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3856, pruned_loss=0.135, over 5694330.99 frames. ], batch size: 174, lr: 8.17e-03, grad_scale: 4.0
+2023-03-02 01:28:22,808 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:28:50,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:29:09,206 INFO [train.py:968] (1/2) Epoch 4, batch 6150, giga_loss[loss=0.3275, simple_loss=0.3962, pruned_loss=0.1293, over 28833.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.394, pruned_loss=0.1407, over 5682568.21 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3894, pruned_loss=0.127, over 5487805.06 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.393, pruned_loss=0.141, over 5676676.43 frames. ], batch size: 145, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:29:24,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:29:39,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:29:46,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142328.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 01:29:48,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142331.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 01:30:01,379 INFO [train.py:968] (1/2) Epoch 4, batch 6200, giga_loss[loss=0.3574, simple_loss=0.4059, pruned_loss=0.1545, over 28302.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4015, pruned_loss=0.1478, over 5677829.97 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3892, pruned_loss=0.127, over 5492162.67 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4009, pruned_loss=0.1483, over 5671833.64 frames. ], batch size: 368, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:30:05,871 INFO [optim.py:369] (1/2) 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 01:30:08,019 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:30:09,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:30:14,775 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142360.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 01:30:37,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 01:30:38,178 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 6250, giga_loss[loss=0.35, simple_loss=0.4091, pruned_loss=0.1455, over 28804.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.4067, pruned_loss=0.1524, over 5685520.82 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3889, pruned_loss=0.1267, over 5504242.08 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4068, pruned_loss=0.1536, over 5675193.89 frames. ], batch size: 112, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:31:33,372 INFO [train.py:968] (1/2) Epoch 4, batch 6300, libri_loss[loss=0.2958, simple_loss=0.3609, pruned_loss=0.1154, over 29584.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4125, pruned_loss=0.1572, over 5679185.77 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3895, pruned_loss=0.127, over 5513577.99 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4127, pruned_loss=0.1588, over 5666593.36 frames. ], batch size: 74, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:31:39,042 INFO [optim.py:369] (1/2) 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 01:32:14,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 01:32:26,062 INFO [train.py:968] (1/2) Epoch 4, batch 6350, giga_loss[loss=0.4254, simple_loss=0.4501, pruned_loss=0.2003, over 28565.00 frames. ], tot_loss[loss=0.3678, simple_loss=0.4148, pruned_loss=0.1604, over 5656996.91 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3891, pruned_loss=0.1268, over 5517602.84 frames. ], giga_tot_loss[loss=0.3698, simple_loss=0.4155, pruned_loss=0.1621, over 5645103.37 frames. ], batch size: 307, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:33:20,195 INFO [train.py:968] (1/2) Epoch 4, batch 6400, giga_loss[loss=0.4323, simple_loss=0.4585, pruned_loss=0.203, over 28247.00 frames. ], tot_loss[loss=0.3731, simple_loss=0.4174, pruned_loss=0.1643, over 5641984.98 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3894, pruned_loss=0.1272, over 5527500.12 frames. ], giga_tot_loss[loss=0.3754, simple_loss=0.4184, pruned_loss=0.1662, over 5626165.73 frames. ], batch size: 368, lr: 8.16e-03, grad_scale: 8.0
+2023-03-02 01:33:27,551 INFO [optim.py:369] (1/2) 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 01:33:50,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5826, 1.3973, 1.1846, 1.2407], device='cuda:1'), covar=tensor([0.0487, 0.0397, 0.0746, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0478, 0.0519, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 01:34:14,580 INFO [train.py:968] (1/2) Epoch 4, batch 6450, giga_loss[loss=0.4577, simple_loss=0.4691, pruned_loss=0.2231, over 27519.00 frames. ], tot_loss[loss=0.3844, simple_loss=0.4246, pruned_loss=0.1721, over 5617556.65 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3901, pruned_loss=0.1278, over 5522419.39 frames. ], giga_tot_loss[loss=0.3861, simple_loss=0.4251, pruned_loss=0.1735, over 5610576.04 frames. ], batch size: 472, lr: 8.16e-03, grad_scale: 4.0
+2023-03-02 01:35:05,614 INFO [train.py:968] (1/2) Epoch 4, batch 6500, giga_loss[loss=0.4217, simple_loss=0.4492, pruned_loss=0.1971, over 28892.00 frames. ], tot_loss[loss=0.3865, simple_loss=0.4269, pruned_loss=0.173, over 5617089.25 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3904, pruned_loss=0.128, over 5530438.83 frames. ], giga_tot_loss[loss=0.3892, simple_loss=0.428, pruned_loss=0.1752, over 5606548.94 frames. ], batch size: 285, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:35:11,556 INFO [optim.py:369] (1/2) 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 01:35:24,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3580, 1.3618, 1.4544, 1.4321], device='cuda:1'), covar=tensor([0.1020, 0.1324, 0.1485, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0758, 0.0631, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 01:35:26,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-02 01:35:44,955 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 6550, giga_loss[loss=0.3926, simple_loss=0.4301, pruned_loss=0.1776, over 28665.00 frames. ], tot_loss[loss=0.3869, simple_loss=0.427, pruned_loss=0.1734, over 5630574.97 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3905, pruned_loss=0.1281, over 5535553.92 frames. ], giga_tot_loss[loss=0.3901, simple_loss=0.4285, pruned_loss=0.1759, over 5619312.44 frames. ], batch size: 262, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:35:59,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=142696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:36:15,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 01:36:31,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 01:36:43,657 INFO [train.py:968] (1/2) Epoch 4, batch 6600, giga_loss[loss=0.3711, simple_loss=0.4157, pruned_loss=0.1633, over 29014.00 frames. ], tot_loss[loss=0.3839, simple_loss=0.4242, pruned_loss=0.1718, over 5640138.59 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3907, pruned_loss=0.1283, over 5546507.99 frames. ], giga_tot_loss[loss=0.3881, simple_loss=0.4263, pruned_loss=0.1749, over 5623679.02 frames. ], batch size: 155, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:36:49,840 INFO [optim.py:369] (1/2) 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 01:37:35,102 INFO [train.py:968] (1/2) Epoch 4, batch 6650, giga_loss[loss=0.423, simple_loss=0.4542, pruned_loss=0.1959, over 27922.00 frames. ], tot_loss[loss=0.3838, simple_loss=0.4243, pruned_loss=0.1717, over 5640822.32 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3907, pruned_loss=0.1282, over 5555285.14 frames. ], giga_tot_loss[loss=0.3886, simple_loss=0.4267, pruned_loss=0.1752, over 5621904.22 frames. ], batch size: 412, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:38:07,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:09,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=142830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:17,368 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=142839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:20,446 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 6700, giga_loss[loss=0.3704, simple_loss=0.4165, pruned_loss=0.1622, over 28323.00 frames. ], tot_loss[loss=0.3793, simple_loss=0.4223, pruned_loss=0.1682, over 5635917.03 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3903, pruned_loss=0.128, over 5550497.62 frames. ], giga_tot_loss[loss=0.3858, simple_loss=0.4259, pruned_loss=0.1728, over 5629076.54 frames. ], batch size: 368, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:38:27,718 INFO [optim.py:369] (1/2) 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 01:38:37,096 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:38:48,642 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=142871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:39:11,629 INFO [train.py:968] (1/2) Epoch 4, batch 6750, giga_loss[loss=0.3473, simple_loss=0.4016, pruned_loss=0.1465, over 28918.00 frames. ], tot_loss[loss=0.3814, simple_loss=0.4238, pruned_loss=0.1695, over 5611750.84 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3904, pruned_loss=0.1281, over 5550193.13 frames. ], giga_tot_loss[loss=0.3868, simple_loss=0.4268, pruned_loss=0.1735, over 5607097.42 frames. ], batch size: 227, lr: 8.15e-03, grad_scale: 4.0
+2023-03-02 01:40:02,327 INFO [train.py:968] (1/2) Epoch 4, batch 6800, giga_loss[loss=0.4059, simple_loss=0.4461, pruned_loss=0.1828, over 28799.00 frames. ], tot_loss[loss=0.3776, simple_loss=0.4214, pruned_loss=0.1669, over 5615390.64 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3905, pruned_loss=0.1282, over 5552177.64 frames. ], giga_tot_loss[loss=0.3826, simple_loss=0.4241, pruned_loss=0.1706, over 5610924.85 frames. ], batch size: 199, lr: 8.15e-03, grad_scale: 8.0
+2023-03-02 01:40:09,246 INFO [optim.py:369] (1/2) 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 01:40:28,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=142966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:40:53,481 INFO [train.py:968] (1/2) Epoch 4, batch 6850, giga_loss[loss=0.3308, simple_loss=0.3912, pruned_loss=0.1352, over 28800.00 frames. ], tot_loss[loss=0.3713, simple_loss=0.4177, pruned_loss=0.1624, over 5609276.35 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3909, pruned_loss=0.1285, over 5549183.82 frames. ], giga_tot_loss[loss=0.3761, simple_loss=0.4203, pruned_loss=0.166, over 5609364.42 frames. ], batch size: 119, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:41:41,425 INFO [train.py:968] (1/2) Epoch 4, batch 6900, libri_loss[loss=0.3244, simple_loss=0.3882, pruned_loss=0.1303, over 29554.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4133, pruned_loss=0.1577, over 5632249.64 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3907, pruned_loss=0.1285, over 5558430.78 frames. ], giga_tot_loss[loss=0.3695, simple_loss=0.4162, pruned_loss=0.1614, over 5625720.05 frames. ], batch size: 79, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:41:48,037 INFO [optim.py:369] (1/2) 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 01:42:03,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6368, 1.1380, 3.4247, 2.9231], device='cuda:1'), covar=tensor([0.1669, 0.1986, 0.0421, 0.0580], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0515, 0.0708, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 01:42:10,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6515, 1.5627, 1.5474, 1.5660], device='cuda:1'), covar=tensor([0.0922, 0.1495, 0.1360, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0752, 0.0634, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 01:42:31,280 INFO [train.py:968] (1/2) Epoch 4, batch 6950, giga_loss[loss=0.4123, simple_loss=0.4472, pruned_loss=0.1887, over 27936.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4114, pruned_loss=0.1563, over 5636143.53 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3905, pruned_loss=0.1284, over 5562124.16 frames. ], giga_tot_loss[loss=0.3667, simple_loss=0.4141, pruned_loss=0.1596, over 5628441.85 frames. ], batch size: 412, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:42:38,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-02 01:42:57,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:43:22,908 INFO [train.py:968] (1/2) Epoch 4, batch 7000, giga_loss[loss=0.3591, simple_loss=0.4065, pruned_loss=0.1559, over 28587.00 frames. ], tot_loss[loss=0.359, simple_loss=0.409, pruned_loss=0.1545, over 5646748.43 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3904, pruned_loss=0.1283, over 5564781.79 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4114, pruned_loss=0.1574, over 5639143.32 frames. ], batch size: 307, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:43:30,970 INFO [optim.py:369] (1/2) 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 01:44:10,854 INFO [train.py:968] (1/2) Epoch 4, batch 7050, libri_loss[loss=0.3609, simple_loss=0.4166, pruned_loss=0.1526, over 20202.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4086, pruned_loss=0.154, over 5649196.57 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3902, pruned_loss=0.1282, over 5563842.24 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4111, pruned_loss=0.157, over 5646670.98 frames. ], batch size: 187, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:44:33,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 01:45:06,732 INFO [train.py:968] (1/2) Epoch 4, batch 7100, giga_loss[loss=0.3023, simple_loss=0.3763, pruned_loss=0.1142, over 28889.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4065, pruned_loss=0.1518, over 5660071.98 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3895, pruned_loss=0.128, over 5572526.64 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4095, pruned_loss=0.155, over 5652240.83 frames. ], batch size: 145, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:45:16,872 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 7150, giga_loss[loss=0.3642, simple_loss=0.4258, pruned_loss=0.1513, over 27990.00 frames. ], tot_loss[loss=0.351, simple_loss=0.4042, pruned_loss=0.1489, over 5668006.20 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3893, pruned_loss=0.1279, over 5579057.81 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.407, pruned_loss=0.1519, over 5657486.27 frames. ], batch size: 412, lr: 8.14e-03, grad_scale: 4.0
+2023-03-02 01:46:31,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5009, 2.0850, 1.8130, 1.9020], device='cuda:1'), covar=tensor([0.0513, 0.0623, 0.0795, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0474, 0.0515, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 01:46:47,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:46:48,930 INFO [train.py:968] (1/2) Epoch 4, batch 7200, libri_loss[loss=0.2852, simple_loss=0.3459, pruned_loss=0.1122, over 29604.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4041, pruned_loss=0.147, over 5673601.05 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3878, pruned_loss=0.127, over 5591683.51 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4085, pruned_loss=0.1513, over 5657884.51 frames. ], batch size: 74, lr: 8.13e-03, grad_scale: 8.0
+2023-03-02 01:46:55,112 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 7250, giga_loss[loss=0.3725, simple_loss=0.3983, pruned_loss=0.1733, over 23662.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4044, pruned_loss=0.1464, over 5667588.22 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3868, pruned_loss=0.1267, over 5593878.47 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4094, pruned_loss=0.1509, over 5655321.56 frames. ], batch size: 705, lr: 8.13e-03, grad_scale: 4.0
+2023-03-02 01:48:01,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 01:48:28,281 INFO [train.py:968] (1/2) Epoch 4, batch 7300, giga_loss[loss=0.3296, simple_loss=0.3904, pruned_loss=0.1344, over 28502.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4056, pruned_loss=0.1481, over 5673440.51 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3869, pruned_loss=0.1267, over 5597007.83 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.4097, pruned_loss=0.1519, over 5662105.06 frames. ], batch size: 336, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:48:37,286 INFO [optim.py:369] (1/2) 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 01:48:48,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2373, 1.4636, 1.2313, 1.4237], device='cuda:1'), covar=tensor([0.0839, 0.0375, 0.0362, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0140, 0.0142, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0053], device='cuda:1')
+2023-03-02 01:48:56,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4278, 1.3548, 5.1214, 3.4358], device='cuda:1'), covar=tensor([0.1495, 0.2001, 0.0273, 0.0597], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0519, 0.0717, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 01:49:07,000 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:10,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:14,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=143492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:14,592 INFO [train.py:968] (1/2) Epoch 4, batch 7350, giga_loss[loss=0.4081, simple_loss=0.4452, pruned_loss=0.1855, over 28673.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.4049, pruned_loss=0.1482, over 5674070.81 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3867, pruned_loss=0.1266, over 5602192.07 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.409, pruned_loss=0.1518, over 5662222.16 frames. ], batch size: 242, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:49:35,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6169, 3.3767, 1.4373, 1.6144], device='cuda:1'), covar=tensor([0.0815, 0.0299, 0.0844, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0468, 0.0313, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 01:49:41,300 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:49:50,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1805, 3.9164, 3.8689, 1.4937], device='cuda:1'), covar=tensor([0.0507, 0.0485, 0.0900, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0672, 0.0822, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 01:50:07,035 INFO [train.py:968] (1/2) Epoch 4, batch 7400, libri_loss[loss=0.4233, simple_loss=0.4645, pruned_loss=0.1911, over 28042.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4036, pruned_loss=0.1489, over 5667944.34 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.387, pruned_loss=0.1269, over 5603770.38 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.4067, pruned_loss=0.1518, over 5657715.27 frames. ], batch size: 116, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:50:15,695 INFO [optim.py:369] (1/2) 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 01:50:18,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4160, 1.4915, 1.1380, 1.5179], device='cuda:1'), covar=tensor([0.0788, 0.0341, 0.0354, 0.0851], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0139, 0.0141, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0052], device='cuda:1')
+2023-03-02 01:50:20,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0861, 3.5068, 2.5644, 0.8110], device='cuda:1'), covar=tensor([0.2871, 0.0913, 0.1386, 0.3192], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1230, 0.1312, 0.1107], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 01:50:53,235 INFO [train.py:968] (1/2) Epoch 4, batch 7450, giga_loss[loss=0.3014, simple_loss=0.3711, pruned_loss=0.1158, over 29050.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4018, pruned_loss=0.148, over 5678123.75 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3868, pruned_loss=0.1267, over 5608532.91 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4046, pruned_loss=0.1508, over 5666735.40 frames. ], batch size: 155, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:51:39,880 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=143635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:51:43,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=143638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:51:43,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4528, 1.5900, 1.1164, 0.9689], device='cuda:1'), covar=tensor([0.0865, 0.0671, 0.0640, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.1311, 0.1033, 0.1084, 0.1134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 01:51:47,626 INFO [train.py:968] (1/2) Epoch 4, batch 7500, giga_loss[loss=0.3418, simple_loss=0.4013, pruned_loss=0.1412, over 28740.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4021, pruned_loss=0.1466, over 5689079.33 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3868, pruned_loss=0.1265, over 5612736.97 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4046, pruned_loss=0.1493, over 5677530.57 frames. ], batch size: 307, lr: 8.13e-03, grad_scale: 2.0
+2023-03-02 01:51:58,742 INFO [optim.py:369] (1/2) 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 01:52:12,915 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=143667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 01:52:20,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-02 01:52:37,102 INFO [train.py:968] (1/2) Epoch 4, batch 7550, giga_loss[loss=0.337, simple_loss=0.4011, pruned_loss=0.1364, over 28974.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.4019, pruned_loss=0.1454, over 5696980.66 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3867, pruned_loss=0.1264, over 5617557.92 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4042, pruned_loss=0.1479, over 5684942.23 frames. ], batch size: 128, lr: 8.12e-03, grad_scale: 2.0
+2023-03-02 01:53:25,150 INFO [train.py:968] (1/2) Epoch 4, batch 7600, giga_loss[loss=0.3558, simple_loss=0.4142, pruned_loss=0.1487, over 29045.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.4016, pruned_loss=0.1453, over 5685538.16 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3866, pruned_loss=0.1264, over 5610639.23 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4037, pruned_loss=0.1476, over 5682798.11 frames. ], batch size: 136, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:53:34,076 INFO [optim.py:369] (1/2) 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 01:54:12,462 INFO [train.py:968] (1/2) Epoch 4, batch 7650, giga_loss[loss=0.3328, simple_loss=0.3868, pruned_loss=0.1393, over 28953.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.4009, pruned_loss=0.1455, over 5692846.52 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3871, pruned_loss=0.1268, over 5618438.53 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4024, pruned_loss=0.1473, over 5685255.52 frames. ], batch size: 136, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:54:13,039 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143793.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 01:55:04,291 INFO [train.py:968] (1/2) Epoch 4, batch 7700, giga_loss[loss=0.393, simple_loss=0.4246, pruned_loss=0.1807, over 27964.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.3997, pruned_loss=0.1457, over 5689706.59 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3867, pruned_loss=0.1266, over 5623802.45 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4015, pruned_loss=0.1477, over 5680179.86 frames. ], batch size: 412, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:55:14,474 INFO [optim.py:369] (1/2) 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 01:55:17,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7127, 2.0809, 1.9091, 1.7883], device='cuda:1'), covar=tensor([0.1585, 0.1766, 0.1167, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0780, 0.0754, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 01:55:28,267 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=143866.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 01:55:54,572 INFO [train.py:968] (1/2) Epoch 4, batch 7750, giga_loss[loss=0.3589, simple_loss=0.4052, pruned_loss=0.1563, over 28945.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3998, pruned_loss=0.1465, over 5690199.23 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3863, pruned_loss=0.1263, over 5626382.64 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4017, pruned_loss=0.1485, over 5680953.12 frames. ], batch size: 227, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:56:44,039 INFO [train.py:968] (1/2) Epoch 4, batch 7800, giga_loss[loss=0.3315, simple_loss=0.3912, pruned_loss=0.1359, over 29104.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3993, pruned_loss=0.1462, over 5699029.76 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3872, pruned_loss=0.1269, over 5629217.07 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4004, pruned_loss=0.1479, over 5691243.77 frames. ], batch size: 136, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:56:45,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2534, 1.9056, 1.6266, 1.4847], device='cuda:1'), covar=tensor([0.0837, 0.0299, 0.0312, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0141, 0.0142, 0.0257], device='cuda:1'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0053], device='cuda:1')
+2023-03-02 01:56:54,276 INFO [optim.py:369] (1/2) 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 01:57:35,582 INFO [train.py:968] (1/2) Epoch 4, batch 7850, giga_loss[loss=0.4696, simple_loss=0.4718, pruned_loss=0.2337, over 26659.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3981, pruned_loss=0.1462, over 5695935.00 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3876, pruned_loss=0.127, over 5632870.05 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.3988, pruned_loss=0.1477, over 5687290.19 frames. ], batch size: 555, lr: 8.12e-03, grad_scale: 4.0
+2023-03-02 01:58:19,919 INFO [train.py:968] (1/2) Epoch 4, batch 7900, giga_loss[loss=0.3482, simple_loss=0.3972, pruned_loss=0.1497, over 28656.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.397, pruned_loss=0.1454, over 5697415.34 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3883, pruned_loss=0.1276, over 5629701.74 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3972, pruned_loss=0.1466, over 5694532.66 frames. ], batch size: 242, lr: 8.11e-03, grad_scale: 2.0
+2023-03-02 01:58:30,173 INFO [optim.py:369] (1/2) 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 01:58:41,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4963, 3.3521, 1.5514, 1.3890], device='cuda:1'), covar=tensor([0.0855, 0.0371, 0.0840, 0.1354], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0476, 0.0315, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 01:59:07,673 INFO [train.py:968] (1/2) Epoch 4, batch 7950, giga_loss[loss=0.4215, simple_loss=0.4407, pruned_loss=0.2011, over 27685.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3981, pruned_loss=0.1463, over 5690657.34 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3878, pruned_loss=0.1272, over 5634649.09 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.3989, pruned_loss=0.148, over 5685286.93 frames. ], batch size: 472, lr: 8.11e-03, grad_scale: 2.0
+2023-03-02 01:59:53,166 INFO [train.py:968] (1/2) Epoch 4, batch 8000, giga_loss[loss=0.3178, simple_loss=0.384, pruned_loss=0.1258, over 28771.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3973, pruned_loss=0.1451, over 5690073.58 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3867, pruned_loss=0.1265, over 5641732.01 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.3993, pruned_loss=0.1476, over 5680674.09 frames. ], batch size: 99, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:00:03,611 INFO [optim.py:369] (1/2) 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 02:00:08,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 02:00:17,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144168.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:00:38,701 INFO [train.py:968] (1/2) Epoch 4, batch 8050, libri_loss[loss=0.292, simple_loss=0.3575, pruned_loss=0.1133, over 29564.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3969, pruned_loss=0.1439, over 5684166.90 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3866, pruned_loss=0.1263, over 5648071.41 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3988, pruned_loss=0.1465, over 5671682.00 frames. ], batch size: 78, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:01:21,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=144241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 02:01:21,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4003, 2.0728, 1.3910, 0.5331], device='cuda:1'), covar=tensor([0.1857, 0.0983, 0.1663, 0.2277], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.1250, 0.1326, 0.1120], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 02:01:22,839 INFO [train.py:968] (1/2) Epoch 4, batch 8100, libri_loss[loss=0.3347, simple_loss=0.4031, pruned_loss=0.1332, over 29393.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.398, pruned_loss=0.1444, over 5687408.60 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.387, pruned_loss=0.1266, over 5655226.38 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.3996, pruned_loss=0.147, over 5672143.24 frames. ], batch size: 92, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:01:32,334 INFO [optim.py:369] (1/2) 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 02:01:32,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 02:02:11,760 INFO [train.py:968] (1/2) Epoch 4, batch 8150, giga_loss[loss=0.3362, simple_loss=0.394, pruned_loss=0.1392, over 28891.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3981, pruned_loss=0.1447, over 5693810.09 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3859, pruned_loss=0.1259, over 5661658.85 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4007, pruned_loss=0.148, over 5676925.91 frames. ], batch size: 199, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:02:28,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144311.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:02:32,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144314.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 02:03:02,592 INFO [train.py:968] (1/2) Epoch 4, batch 8200, giga_loss[loss=0.3515, simple_loss=0.401, pruned_loss=0.1509, over 28615.00 frames. ], tot_loss[loss=0.348, simple_loss=0.4005, pruned_loss=0.1478, over 5689177.07 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3854, pruned_loss=0.1254, over 5667418.54 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4034, pruned_loss=0.1514, over 5671186.25 frames. ], batch size: 78, lr: 8.11e-03, grad_scale: 4.0
+2023-03-02 02:03:02,795 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=144343.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:03:13,663 INFO [optim.py:369] (1/2) 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 02:03:45,169 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144384.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:03:47,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=144387.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:03:52,620 INFO [train.py:968] (1/2) Epoch 4, batch 8250, giga_loss[loss=0.3633, simple_loss=0.4155, pruned_loss=0.1555, over 28646.00 frames. ], tot_loss[loss=0.3513, simple_loss=0.4022, pruned_loss=0.1502, over 5691354.36 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3853, pruned_loss=0.1253, over 5672392.44 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.4049, pruned_loss=0.1536, over 5673060.74 frames. ], batch size: 242, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:04:06,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4162, 1.9549, 1.4694, 1.5398], device='cuda:1'), covar=tensor([0.0775, 0.0329, 0.0318, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0139, 0.0142, 0.0255], device='cuda:1'), out_proj_covar=tensor([0.0047, 0.0035, 0.0031, 0.0053], device='cuda:1')
+2023-03-02 02:04:15,481 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 8300, giga_loss[loss=0.3738, simple_loss=0.408, pruned_loss=0.1698, over 27930.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4044, pruned_loss=0.1533, over 5682327.79 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3856, pruned_loss=0.1256, over 5676705.43 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4066, pruned_loss=0.1562, over 5664314.49 frames. ], batch size: 412, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:04:50,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6394, 1.9561, 1.8649, 1.7744], device='cuda:1'), covar=tensor([0.1438, 0.1581, 0.1104, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0773, 0.0756, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 02:04:54,420 INFO [optim.py:369] (1/2) 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 02:05:27,640 INFO [train.py:968] (1/2) Epoch 4, batch 8350, giga_loss[loss=0.3454, simple_loss=0.4033, pruned_loss=0.1438, over 29060.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4021, pruned_loss=0.1514, over 5687698.04 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3843, pruned_loss=0.1246, over 5686777.73 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4059, pruned_loss=0.1559, over 5663887.96 frames. ], batch size: 155, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:06:12,367 INFO [train.py:968] (1/2) Epoch 4, batch 8400, giga_loss[loss=0.2867, simple_loss=0.3649, pruned_loss=0.1043, over 28300.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4025, pruned_loss=0.1516, over 5681072.58 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3842, pruned_loss=0.1244, over 5685853.84 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4059, pruned_loss=0.1558, over 5663004.13 frames. ], batch size: 77, lr: 8.10e-03, grad_scale: 8.0
+2023-03-02 02:06:21,646 INFO [optim.py:369] (1/2) 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 02:06:58,528 INFO [train.py:968] (1/2) Epoch 4, batch 8450, giga_loss[loss=0.3561, simple_loss=0.38, pruned_loss=0.1661, over 23870.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4011, pruned_loss=0.1496, over 5674976.31 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3844, pruned_loss=0.1245, over 5687802.81 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4037, pruned_loss=0.1531, over 5658802.33 frames. ], batch size: 705, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:07:41,882 INFO [train.py:968] (1/2) Epoch 4, batch 8500, giga_loss[loss=0.4021, simple_loss=0.4391, pruned_loss=0.1825, over 28743.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3983, pruned_loss=0.1472, over 5682517.18 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3848, pruned_loss=0.1249, over 5690749.65 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4003, pruned_loss=0.1501, over 5666949.74 frames. ], batch size: 284, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:07:52,523 INFO [optim.py:369] (1/2) 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 02:08:27,855 INFO [train.py:968] (1/2) Epoch 4, batch 8550, giga_loss[loss=0.343, simple_loss=0.391, pruned_loss=0.1476, over 28676.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.396, pruned_loss=0.1461, over 5690972.09 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3846, pruned_loss=0.1247, over 5697229.78 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.3983, pruned_loss=0.1492, over 5672192.48 frames. ], batch size: 307, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:08:30,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2617, 4.0186, 3.9589, 1.8520], device='cuda:1'), covar=tensor([0.0403, 0.0416, 0.0751, 0.2074], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0678, 0.0826, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 02:08:38,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4234, 1.7130, 1.5503, 1.5287], device='cuda:1'), covar=tensor([0.1053, 0.1507, 0.0971, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0768, 0.0751, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 02:09:01,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-02 02:09:17,529 INFO [train.py:968] (1/2) Epoch 4, batch 8600, giga_loss[loss=0.3103, simple_loss=0.3698, pruned_loss=0.1254, over 28421.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3961, pruned_loss=0.1471, over 5672414.02 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3848, pruned_loss=0.1249, over 5691431.98 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.3977, pruned_loss=0.1496, over 5663417.55 frames. ], batch size: 65, lr: 8.10e-03, grad_scale: 4.0
+2023-03-02 02:09:30,286 INFO [optim.py:369] (1/2) 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 02:09:44,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 02:10:09,515 INFO [train.py:968] (1/2) Epoch 4, batch 8650, libri_loss[loss=0.3786, simple_loss=0.4359, pruned_loss=0.1606, over 29742.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.3984, pruned_loss=0.1489, over 5650416.77 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3855, pruned_loss=0.1253, over 5683630.55 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.3993, pruned_loss=0.1511, over 5649161.62 frames. ], batch size: 87, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:10:18,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=144804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:10:29,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5641, 1.9018, 1.8021, 1.7336], device='cuda:1'), covar=tensor([0.1445, 0.1832, 0.1119, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0767, 0.0746, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 02:10:57,929 INFO [train.py:968] (1/2) Epoch 4, batch 8700, giga_loss[loss=0.303, simple_loss=0.3904, pruned_loss=0.1078, over 28886.00 frames. ], tot_loss[loss=0.3513, simple_loss=0.403, pruned_loss=0.1498, over 5653204.02 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3856, pruned_loss=0.1254, over 5676963.09 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4039, pruned_loss=0.1517, over 5656879.56 frames. ], batch size: 174, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:11:08,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7617, 1.6837, 1.5648, 1.6277], device='cuda:1'), covar=tensor([0.1781, 0.2597, 0.1519, 0.1314], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0763, 0.0746, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 02:11:10,930 INFO [optim.py:369] (1/2) 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 02:11:44,877 INFO [train.py:968] (1/2) Epoch 4, batch 8750, giga_loss[loss=0.4115, simple_loss=0.4562, pruned_loss=0.1833, over 28840.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.405, pruned_loss=0.1486, over 5667093.43 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3855, pruned_loss=0.1253, over 5683594.98 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4064, pruned_loss=0.1509, over 5663447.52 frames. ], batch size: 186, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:12:19,802 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-02 02:12:34,760 INFO [train.py:968] (1/2) Epoch 4, batch 8800, giga_loss[loss=0.3659, simple_loss=0.4192, pruned_loss=0.1563, over 28489.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4069, pruned_loss=0.15, over 5667453.56 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3855, pruned_loss=0.1253, over 5687794.50 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4083, pruned_loss=0.1523, over 5660614.35 frames. ], batch size: 336, lr: 8.09e-03, grad_scale: 8.0
+2023-03-02 02:12:42,888 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 8850, giga_loss[loss=0.3289, simple_loss=0.3913, pruned_loss=0.1332, over 28673.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.409, pruned_loss=0.1523, over 5652164.86 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3859, pruned_loss=0.1254, over 5681934.58 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4103, pruned_loss=0.1546, over 5651170.60 frames. ], batch size: 92, lr: 8.09e-03, grad_scale: 8.0
+2023-03-02 02:13:20,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6765, 1.9892, 1.8815, 1.7701], device='cuda:1'), covar=tensor([0.1158, 0.1390, 0.0951, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0764, 0.0747, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 02:13:22,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3568, 2.8027, 1.3181, 1.3061], device='cuda:1'), covar=tensor([0.0836, 0.0302, 0.0892, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0470, 0.0311, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 02:13:26,151 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-02 02:14:05,217 INFO [train.py:968] (1/2) Epoch 4, batch 8900, giga_loss[loss=0.3537, simple_loss=0.4053, pruned_loss=0.151, over 28725.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4087, pruned_loss=0.1526, over 5659054.27 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3861, pruned_loss=0.1256, over 5685033.36 frames. ], giga_tot_loss[loss=0.3594, simple_loss=0.4098, pruned_loss=0.1545, over 5655203.50 frames. ], batch size: 284, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:14:17,821 INFO [optim.py:369] (1/2) 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 02:14:53,559 INFO [train.py:968] (1/2) Epoch 4, batch 8950, giga_loss[loss=0.3994, simple_loss=0.4194, pruned_loss=0.1897, over 23405.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.407, pruned_loss=0.1528, over 5645130.90 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3857, pruned_loss=0.1254, over 5689350.88 frames. ], giga_tot_loss[loss=0.3596, simple_loss=0.4088, pruned_loss=0.1552, over 5637511.74 frames. ], batch size: 705, lr: 8.09e-03, grad_scale: 4.0
+2023-03-02 02:15:39,825 INFO [train.py:968] (1/2) Epoch 4, batch 9000, giga_loss[loss=0.3663, simple_loss=0.4114, pruned_loss=0.1606, over 28647.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4055, pruned_loss=0.1518, over 5656942.76 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.386, pruned_loss=0.1255, over 5692227.71 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4072, pruned_loss=0.1544, over 5646787.58 frames. ], batch size: 262, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:15:39,826 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 02:15:48,058 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 02:16:00,361 INFO [optim.py:369] (1/2) 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 02:16:22,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:16:32,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8790, 1.6705, 1.2676, 1.5329], device='cuda:1'), covar=tensor([0.0647, 0.0695, 0.1021, 0.0929], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0479, 0.0522, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 02:16:35,313 INFO [train.py:968] (1/2) Epoch 4, batch 9050, giga_loss[loss=0.3498, simple_loss=0.4004, pruned_loss=0.1497, over 28534.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4038, pruned_loss=0.1512, over 5657581.00 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3863, pruned_loss=0.1257, over 5694037.86 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4054, pruned_loss=0.1537, over 5647337.20 frames. ], batch size: 336, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:17:23,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2182, 1.8132, 1.4788, 0.5573], device='cuda:1'), covar=tensor([0.1578, 0.1066, 0.1741, 0.1823], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.1258, 0.1324, 0.1122], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 02:17:25,495 INFO [train.py:968] (1/2) Epoch 4, batch 9100, libri_loss[loss=0.3407, simple_loss=0.4071, pruned_loss=0.1372, over 29629.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4043, pruned_loss=0.152, over 5658187.62 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3863, pruned_loss=0.1257, over 5697243.96 frames. ], giga_tot_loss[loss=0.3576, simple_loss=0.4058, pruned_loss=0.1547, over 5646377.98 frames. ], batch size: 91, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:17:38,492 INFO [optim.py:369] (1/2) 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 02:18:15,869 INFO [train.py:968] (1/2) Epoch 4, batch 9150, giga_loss[loss=0.3093, simple_loss=0.3623, pruned_loss=0.1282, over 29031.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.4033, pruned_loss=0.1515, over 5653468.57 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3864, pruned_loss=0.1257, over 5699594.10 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4046, pruned_loss=0.1538, over 5641704.75 frames. ], batch size: 128, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:18:23,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-02 02:18:25,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7058, 1.6543, 1.4566, 1.9501], device='cuda:1'), covar=tensor([0.2136, 0.1967, 0.1901, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.1072, 0.0860, 0.0958, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 02:18:42,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:18:46,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:19:02,736 INFO [train.py:968] (1/2) Epoch 4, batch 9200, libri_loss[loss=0.3213, simple_loss=0.3954, pruned_loss=0.1236, over 29505.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4002, pruned_loss=0.1496, over 5665429.30 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.386, pruned_loss=0.1253, over 5706672.48 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4021, pruned_loss=0.1526, over 5648413.64 frames. ], batch size: 82, lr: 8.08e-03, grad_scale: 8.0
+2023-03-02 02:19:12,142 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:19:45,822 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:19:49,764 INFO [train.py:968] (1/2) Epoch 4, batch 9250, giga_loss[loss=0.3593, simple_loss=0.4128, pruned_loss=0.1529, over 29020.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.3993, pruned_loss=0.1488, over 5665583.13 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3857, pruned_loss=0.1251, over 5713255.94 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4016, pruned_loss=0.1522, over 5644410.49 frames. ], batch size: 164, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:20:04,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4137, 1.6416, 1.3040, 0.9016], device='cuda:1'), covar=tensor([0.0871, 0.0645, 0.0522, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.1027, 0.1084, 0.1131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 02:20:11,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:20:13,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4233, 5.1008, 5.0543, 2.1172], device='cuda:1'), covar=tensor([0.0336, 0.0310, 0.0750, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0690, 0.0835, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 02:20:37,573 INFO [train.py:968] (1/2) Epoch 4, batch 9300, giga_loss[loss=0.3796, simple_loss=0.4243, pruned_loss=0.1675, over 28514.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.4002, pruned_loss=0.1481, over 5667923.63 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3862, pruned_loss=0.1255, over 5714128.10 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4017, pruned_loss=0.1508, over 5649809.00 frames. ], batch size: 336, lr: 8.08e-03, grad_scale: 4.0
+2023-03-02 02:20:45,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2501, 1.3405, 1.1306, 1.5656], device='cuda:1'), covar=tensor([0.2037, 0.1861, 0.1774, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1069, 0.0863, 0.0962, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 02:20:52,616 INFO [optim.py:369] (1/2) 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 02:20:54,147 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:20:54,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8100, 1.1443, 3.4046, 2.9179], device='cuda:1'), covar=tensor([0.1614, 0.2111, 0.0438, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0559, 0.0520, 0.0733, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 02:21:10,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5824, 2.4307, 1.5775, 0.7470], device='cuda:1'), covar=tensor([0.2602, 0.1216, 0.2165, 0.2796], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.1254, 0.1324, 0.1127], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 02:21:24,350 INFO [train.py:968] (1/2) Epoch 4, batch 9350, libri_loss[loss=0.3032, simple_loss=0.3679, pruned_loss=0.1193, over 29552.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4029, pruned_loss=0.1502, over 5669608.41 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3862, pruned_loss=0.1255, over 5715494.06 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4046, pruned_loss=0.1529, over 5652844.67 frames. ], batch size: 77, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:21:51,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-02 02:22:13,529 INFO [train.py:968] (1/2) Epoch 4, batch 9400, giga_loss[loss=0.3645, simple_loss=0.4131, pruned_loss=0.1579, over 28649.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4044, pruned_loss=0.1527, over 5660444.76 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3862, pruned_loss=0.1255, over 5714029.31 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4059, pruned_loss=0.1551, over 5648125.55 frames. ], batch size: 307, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:22:28,262 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 9450, giga_loss[loss=0.2755, simple_loss=0.3624, pruned_loss=0.09434, over 28803.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4047, pruned_loss=0.1499, over 5672684.83 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3857, pruned_loss=0.1253, over 5719379.90 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4069, pruned_loss=0.1528, over 5656222.48 frames. ], batch size: 112, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:23:46,188 INFO [train.py:968] (1/2) Epoch 4, batch 9500, giga_loss[loss=0.3873, simple_loss=0.4312, pruned_loss=0.1717, over 28306.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4063, pruned_loss=0.149, over 5676024.58 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3861, pruned_loss=0.1255, over 5722734.06 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4082, pruned_loss=0.1516, over 5659132.76 frames. ], batch size: 368, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:24:01,150 INFO [optim.py:369] (1/2) 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 02:24:38,085 INFO [train.py:968] (1/2) Epoch 4, batch 9550, giga_loss[loss=0.3328, simple_loss=0.3984, pruned_loss=0.1336, over 29032.00 frames. ], tot_loss[loss=0.3554, simple_loss=0.4095, pruned_loss=0.1506, over 5681165.68 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.386, pruned_loss=0.1255, over 5725404.83 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4114, pruned_loss=0.153, over 5664867.92 frames. ], batch size: 155, lr: 8.07e-03, grad_scale: 4.0
+2023-03-02 02:25:06,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 02:25:27,320 INFO [train.py:968] (1/2) Epoch 4, batch 9600, giga_loss[loss=0.3346, simple_loss=0.3981, pruned_loss=0.1355, over 28938.00 frames. ], tot_loss[loss=0.3606, simple_loss=0.4126, pruned_loss=0.1544, over 5679605.08 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3855, pruned_loss=0.1253, over 5726485.57 frames. ], giga_tot_loss[loss=0.3639, simple_loss=0.4146, pruned_loss=0.1566, over 5665562.59 frames. ], batch size: 136, lr: 8.07e-03, grad_scale: 8.0
+2023-03-02 02:25:39,573 INFO [optim.py:369] (1/2) 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 02:25:46,349 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:25:46,378 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 9650, giga_loss[loss=0.352, simple_loss=0.4061, pruned_loss=0.1489, over 28856.00 frames. ], tot_loss[loss=0.3633, simple_loss=0.4138, pruned_loss=0.1564, over 5671663.97 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3848, pruned_loss=0.1248, over 5728092.82 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.4169, pruned_loss=0.1596, over 5657685.84 frames. ], batch size: 186, lr: 8.07e-03, grad_scale: 8.0
+2023-03-02 02:26:17,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:26:55,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=145834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:26:57,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-02 02:27:02,536 INFO [train.py:968] (1/2) Epoch 4, batch 9700, giga_loss[loss=0.3161, simple_loss=0.3832, pruned_loss=0.1245, over 28923.00 frames. ], tot_loss[loss=0.3627, simple_loss=0.413, pruned_loss=0.1562, over 5671692.24 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3849, pruned_loss=0.1248, over 5731806.24 frames. ], giga_tot_loss[loss=0.3672, simple_loss=0.4159, pruned_loss=0.1592, over 5656019.48 frames. ], batch size: 213, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:27:15,699 INFO [optim.py:369] (1/2) 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 02:27:27,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5826, 3.2138, 1.6111, 1.4319], device='cuda:1'), covar=tensor([0.0784, 0.0326, 0.0798, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0466, 0.0309, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0022, 0.0014, 0.0019], device='cuda:1')
+2023-03-02 02:27:47,765 INFO [train.py:968] (1/2) Epoch 4, batch 9750, giga_loss[loss=0.3202, simple_loss=0.3904, pruned_loss=0.1251, over 28617.00 frames. ], tot_loss[loss=0.3592, simple_loss=0.4109, pruned_loss=0.1538, over 5677831.48 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3848, pruned_loss=0.1248, over 5734533.57 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4137, pruned_loss=0.1566, over 5662326.85 frames. ], batch size: 242, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:27:50,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9998, 1.1908, 3.7660, 3.1215], device='cuda:1'), covar=tensor([0.1419, 0.1983, 0.0358, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0549, 0.0515, 0.0719, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 02:27:59,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:02,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:08,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6044, 2.1637, 1.3796, 1.1667], device='cuda:1'), covar=tensor([0.1101, 0.0620, 0.0711, 0.0898], device='cuda:1'), in_proj_covar=tensor([0.1304, 0.1034, 0.1073, 0.1122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 02:28:29,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:30,458 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:28:31,777 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 9800, libri_loss[loss=0.3211, simple_loss=0.3942, pruned_loss=0.124, over 29670.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.4098, pruned_loss=0.1508, over 5683293.77 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3845, pruned_loss=0.1247, over 5738464.94 frames. ], giga_tot_loss[loss=0.3604, simple_loss=0.4129, pruned_loss=0.154, over 5665807.02 frames. ], batch size: 88, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:28:44,537 INFO [optim.py:369] (1/2) 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 02:28:56,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=145970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:01,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=145977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:04,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=145980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:10,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=145987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:14,703 INFO [train.py:968] (1/2) Epoch 4, batch 9850, giga_loss[loss=0.3583, simple_loss=0.4189, pruned_loss=0.1489, over 29002.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4105, pruned_loss=0.1507, over 5681933.78 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3845, pruned_loss=0.1245, over 5738636.81 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4139, pruned_loss=0.1542, over 5665998.12 frames. ], batch size: 136, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:29:14,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9035, 1.8566, 1.5942, 1.7559], device='cuda:1'), covar=tensor([0.0886, 0.1610, 0.1323, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0740, 0.0624, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 02:29:17,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4320, 1.5136, 4.8643, 3.3922], device='cuda:1'), covar=tensor([0.1538, 0.2020, 0.0303, 0.0609], device='cuda:1'), in_proj_covar=tensor([0.0548, 0.0515, 0.0718, 0.0577], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 02:29:17,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0606, 2.5931, 1.3845, 1.2967], device='cuda:1'), covar=tensor([0.1046, 0.0529, 0.0733, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.1308, 0.1033, 0.1075, 0.1118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 02:29:32,006 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:29:58,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:30:03,456 INFO [train.py:968] (1/2) Epoch 4, batch 9900, giga_loss[loss=0.3829, simple_loss=0.4326, pruned_loss=0.1666, over 28427.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4116, pruned_loss=0.1522, over 5678428.88 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.384, pruned_loss=0.1244, over 5741070.94 frames. ], giga_tot_loss[loss=0.3637, simple_loss=0.4155, pruned_loss=0.1559, over 5661558.59 frames. ], batch size: 71, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:30:17,857 INFO [optim.py:369] (1/2) 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 02:30:48,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1234, 3.8952, 3.8227, 1.6467], device='cuda:1'), covar=tensor([0.0518, 0.0441, 0.0880, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0693, 0.0833, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 02:30:53,549 INFO [train.py:968] (1/2) Epoch 4, batch 9950, giga_loss[loss=0.39, simple_loss=0.4249, pruned_loss=0.1775, over 27614.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4111, pruned_loss=0.1528, over 5674190.71 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3836, pruned_loss=0.124, over 5746012.94 frames. ], giga_tot_loss[loss=0.3646, simple_loss=0.4154, pruned_loss=0.1569, over 5654448.66 frames. ], batch size: 472, lr: 8.06e-03, grad_scale: 4.0
+2023-03-02 02:31:37,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:31:39,786 INFO [train.py:968] (1/2) Epoch 4, batch 10000, libri_loss[loss=0.3033, simple_loss=0.3725, pruned_loss=0.117, over 29566.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4097, pruned_loss=0.153, over 5671566.21 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.384, pruned_loss=0.1242, over 5750275.62 frames. ], giga_tot_loss[loss=0.3637, simple_loss=0.4136, pruned_loss=0.1569, over 5649919.77 frames. ], batch size: 76, lr: 8.06e-03, grad_scale: 8.0
+2023-03-02 02:31:56,748 INFO [optim.py:369] (1/2) 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 02:32:12,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6948, 2.7695, 1.8965, 0.7533], device='cuda:1'), covar=tensor([0.3062, 0.1330, 0.1695, 0.2878], device='cuda:1'), in_proj_covar=tensor([0.1311, 0.1249, 0.1319, 0.1136], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 02:32:31,268 INFO [train.py:968] (1/2) Epoch 4, batch 10050, giga_loss[loss=0.3863, simple_loss=0.4032, pruned_loss=0.1847, over 23245.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4072, pruned_loss=0.1522, over 5674098.85 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3837, pruned_loss=0.1241, over 5752756.43 frames. ], giga_tot_loss[loss=0.3612, simple_loss=0.4109, pruned_loss=0.1558, over 5653888.55 frames. ], batch size: 705, lr: 8.05e-03, grad_scale: 8.0
+2023-03-02 02:32:54,123 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146214.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:33:10,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3734, 1.4738, 1.1876, 0.8726], device='cuda:1'), covar=tensor([0.0833, 0.0705, 0.0542, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.1037, 0.1093, 0.1134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 02:33:25,870 INFO [train.py:968] (1/2) Epoch 4, batch 10100, giga_loss[loss=0.41, simple_loss=0.4345, pruned_loss=0.1927, over 27546.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4041, pruned_loss=0.1509, over 5666512.14 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3835, pruned_loss=0.124, over 5753249.31 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4073, pruned_loss=0.154, over 5649509.01 frames. ], batch size: 472, lr: 8.05e-03, grad_scale: 8.0
+2023-03-02 02:33:43,401 INFO [optim.py:369] (1/2) 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 02:34:03,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 02:34:07,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:34:10,460 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 10150, giga_loss[loss=0.3928, simple_loss=0.4091, pruned_loss=0.1882, over 23493.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4027, pruned_loss=0.1507, over 5660994.75 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3831, pruned_loss=0.1237, over 5754043.10 frames. ], giga_tot_loss[loss=0.3569, simple_loss=0.4059, pruned_loss=0.1539, over 5645304.69 frames. ], batch size: 705, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:34:34,954 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:34:47,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-02 02:34:58,908 INFO [train.py:968] (1/2) Epoch 4, batch 10200, giga_loss[loss=0.3331, simple_loss=0.3906, pruned_loss=0.1378, over 28976.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4019, pruned_loss=0.1495, over 5673028.04 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.384, pruned_loss=0.1242, over 5760459.46 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4046, pruned_loss=0.1529, over 5650859.12 frames. ], batch size: 136, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:35:10,698 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-02 02:35:14,432 INFO [optim.py:369] (1/2) 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 02:35:17,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:35:25,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6535, 2.1118, 1.2465, 1.1597], device='cuda:1'), covar=tensor([0.0980, 0.0625, 0.0742, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.1032, 0.1081, 0.1123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 02:35:47,482 INFO [train.py:968] (1/2) Epoch 4, batch 10250, giga_loss[loss=0.2964, simple_loss=0.37, pruned_loss=0.1114, over 28687.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3988, pruned_loss=0.1455, over 5683830.32 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3841, pruned_loss=0.1243, over 5763506.43 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4012, pruned_loss=0.1485, over 5661962.05 frames. ], batch size: 262, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:36:03,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:36:38,638 INFO [train.py:968] (1/2) Epoch 4, batch 10300, giga_loss[loss=0.3343, simple_loss=0.391, pruned_loss=0.1389, over 28743.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3949, pruned_loss=0.1418, over 5673104.07 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3841, pruned_loss=0.1243, over 5766126.75 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.397, pruned_loss=0.1446, over 5652113.79 frames. ], batch size: 119, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:36:54,246 INFO [optim.py:369] (1/2) 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 02:37:26,032 INFO [train.py:968] (1/2) Epoch 4, batch 10350, giga_loss[loss=0.3182, simple_loss=0.3791, pruned_loss=0.1286, over 28703.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3954, pruned_loss=0.1415, over 5677575.26 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3841, pruned_loss=0.1241, over 5764744.85 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3974, pruned_loss=0.1445, over 5659052.79 frames. ], batch size: 92, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:37:37,699 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:37:39,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:12,878 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:18,361 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 10400, giga_loss[loss=0.3452, simple_loss=0.3856, pruned_loss=0.1524, over 27626.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3928, pruned_loss=0.1411, over 5676842.21 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3836, pruned_loss=0.1238, over 5767387.48 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3952, pruned_loss=0.1441, over 5658134.31 frames. ], batch size: 472, lr: 8.05e-03, grad_scale: 4.0
+2023-03-02 02:38:28,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:30,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:38:33,370 INFO [optim.py:369] (1/2) 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 02:39:00,838 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:39:02,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146589.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 02:39:06,156 INFO [train.py:968] (1/2) Epoch 4, batch 10450, giga_loss[loss=0.3136, simple_loss=0.3669, pruned_loss=0.1302, over 28285.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3911, pruned_loss=0.1408, over 5669758.55 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3839, pruned_loss=0.1241, over 5765386.56 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3928, pruned_loss=0.1433, over 5654789.72 frames. ], batch size: 77, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:39:11,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3171, 1.6243, 1.2252, 0.7080], device='cuda:1'), covar=tensor([0.1284, 0.0804, 0.0987, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.1299, 0.1235, 0.1309, 0.1120], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 02:39:53,711 INFO [train.py:968] (1/2) Epoch 4, batch 10500, giga_loss[loss=0.3303, simple_loss=0.3871, pruned_loss=0.1367, over 28830.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.394, pruned_loss=0.1427, over 5661701.89 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3843, pruned_loss=0.1243, over 5757596.51 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3951, pruned_loss=0.1447, over 5655807.96 frames. ], batch size: 119, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:40:10,256 INFO [optim.py:369] (1/2) 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 02:40:42,446 INFO [train.py:968] (1/2) Epoch 4, batch 10550, giga_loss[loss=0.3511, simple_loss=0.4057, pruned_loss=0.1482, over 28562.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.396, pruned_loss=0.1438, over 5655441.62 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3845, pruned_loss=0.1244, over 5756627.32 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.397, pruned_loss=0.1457, over 5649870.17 frames. ], batch size: 336, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:40:52,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8475, 1.7766, 1.5840, 2.1747], device='cuda:1'), covar=tensor([0.1973, 0.1952, 0.1780, 0.2069], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.0865, 0.0969, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 02:41:01,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4477, 1.5430, 1.0538, 1.0109], device='cuda:1'), covar=tensor([0.0655, 0.0652, 0.0622, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.1299, 0.1033, 0.1069, 0.1123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 02:41:19,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=146732.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:41:22,057 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=146735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 02:41:28,493 INFO [train.py:968] (1/2) Epoch 4, batch 10600, giga_loss[loss=0.2966, simple_loss=0.3599, pruned_loss=0.1166, over 28802.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3955, pruned_loss=0.1433, over 5655537.27 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3851, pruned_loss=0.1247, over 5758034.63 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3963, pruned_loss=0.1455, over 5644806.55 frames. ], batch size: 99, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:41:44,887 INFO [optim.py:369] (1/2) 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 02:41:50,580 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=146764.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 02:42:17,747 INFO [train.py:968] (1/2) Epoch 4, batch 10650, libri_loss[loss=0.3481, simple_loss=0.4141, pruned_loss=0.141, over 19708.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3945, pruned_loss=0.143, over 5647924.85 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3851, pruned_loss=0.1247, over 5750755.75 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3952, pruned_loss=0.145, over 5645419.07 frames. ], batch size: 186, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:42:28,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3118, 1.4780, 1.2444, 1.4082], device='cuda:1'), covar=tensor([0.0847, 0.0328, 0.0356, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0138, 0.0140, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0048, 0.0035, 0.0031, 0.0053], device='cuda:1')
+2023-03-02 02:43:07,677 INFO [train.py:968] (1/2) Epoch 4, batch 10700, giga_loss[loss=0.3653, simple_loss=0.4241, pruned_loss=0.1533, over 28693.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3974, pruned_loss=0.1453, over 5657903.95 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3854, pruned_loss=0.1249, over 5752046.31 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3978, pruned_loss=0.1469, over 5653411.66 frames. ], batch size: 242, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:43:24,856 INFO [optim.py:369] (1/2) 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 02:43:58,861 INFO [train.py:968] (1/2) Epoch 4, batch 10750, giga_loss[loss=0.313, simple_loss=0.3772, pruned_loss=0.1244, over 28953.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3998, pruned_loss=0.1467, over 5656799.30 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3853, pruned_loss=0.1247, over 5750791.07 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4005, pruned_loss=0.1484, over 5652716.05 frames. ], batch size: 106, lr: 8.04e-03, grad_scale: 4.0
+2023-03-02 02:44:22,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=146917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:44:27,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 02:44:42,657 INFO [train.py:968] (1/2) Epoch 4, batch 10800, giga_loss[loss=0.3462, simple_loss=0.4081, pruned_loss=0.1422, over 29000.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.4009, pruned_loss=0.1475, over 5655322.84 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3851, pruned_loss=0.1245, over 5745050.06 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4023, pruned_loss=0.1502, over 5652910.10 frames. ], batch size: 164, lr: 8.03e-03, grad_scale: 8.0
+2023-03-02 02:44:58,083 INFO [optim.py:369] (1/2) 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 02:45:21,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=146984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:45:29,671 INFO [train.py:968] (1/2) Epoch 4, batch 10850, giga_loss[loss=0.3479, simple_loss=0.3989, pruned_loss=0.1484, over 28851.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4027, pruned_loss=0.1491, over 5667219.69 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3849, pruned_loss=0.1244, over 5749947.68 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4047, pruned_loss=0.1524, over 5657967.24 frames. ], batch size: 119, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:45:57,184 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:46:05,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4682, 2.1725, 1.5823, 0.6519], device='cuda:1'), covar=tensor([0.2209, 0.1146, 0.1770, 0.2341], device='cuda:1'), in_proj_covar=tensor([0.1318, 0.1253, 0.1314, 0.1133], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 02:46:17,742 INFO [train.py:968] (1/2) Epoch 4, batch 10900, libri_loss[loss=0.2937, simple_loss=0.3554, pruned_loss=0.116, over 29498.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4023, pruned_loss=0.1485, over 5679148.39 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3841, pruned_loss=0.1239, over 5755208.97 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4053, pruned_loss=0.1524, over 5664374.89 frames. ], batch size: 70, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:46:33,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4841, 1.5322, 1.3432, 1.7412], device='cuda:1'), covar=tensor([0.1832, 0.1543, 0.1408, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.1079, 0.0860, 0.0965, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 02:46:34,684 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:46:39,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:46:49,826 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-02 02:47:09,442 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 10950, giga_loss[loss=0.3278, simple_loss=0.3898, pruned_loss=0.1329, over 28892.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.4024, pruned_loss=0.1476, over 5672717.54 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3835, pruned_loss=0.1237, over 5759343.00 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4059, pruned_loss=0.1516, over 5654387.01 frames. ], batch size: 155, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:48:02,721 INFO [train.py:968] (1/2) Epoch 4, batch 11000, giga_loss[loss=0.4183, simple_loss=0.4575, pruned_loss=0.1896, over 28608.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4035, pruned_loss=0.1493, over 5666968.03 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3835, pruned_loss=0.1237, over 5760576.78 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.4065, pruned_loss=0.1528, over 5650398.09 frames. ], batch size: 307, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:48:19,436 INFO [optim.py:369] (1/2) 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 02:48:54,147 INFO [train.py:968] (1/2) Epoch 4, batch 11050, giga_loss[loss=0.3511, simple_loss=0.3973, pruned_loss=0.1525, over 27511.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4036, pruned_loss=0.1505, over 5632613.85 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3837, pruned_loss=0.124, over 5744951.80 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4062, pruned_loss=0.1536, over 5631568.21 frames. ], batch size: 472, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:49:43,780 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 11100, giga_loss[loss=0.3262, simple_loss=0.3853, pruned_loss=0.1335, over 28953.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4032, pruned_loss=0.1505, over 5644045.71 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.384, pruned_loss=0.1243, over 5746250.75 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4056, pruned_loss=0.1536, over 5638608.15 frames. ], batch size: 164, lr: 8.03e-03, grad_scale: 4.0
+2023-03-02 02:50:07,201 INFO [optim.py:369] (1/2) 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 02:50:39,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1003, 2.8822, 2.8496, 1.5708], device='cuda:1'), covar=tensor([0.0791, 0.0628, 0.1000, 0.1908], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0691, 0.0835, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 02:50:39,407 INFO [train.py:968] (1/2) Epoch 4, batch 11150, giga_loss[loss=0.4735, simple_loss=0.4754, pruned_loss=0.2358, over 27618.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4023, pruned_loss=0.151, over 5637217.10 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.384, pruned_loss=0.1243, over 5746250.75 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4042, pruned_loss=0.1533, over 5632984.98 frames. ], batch size: 472, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:51:16,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4789, 2.1763, 1.6404, 0.6091], device='cuda:1'), covar=tensor([0.2323, 0.1153, 0.1737, 0.2674], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.1244, 0.1311, 0.1136], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 02:51:29,942 INFO [train.py:968] (1/2) Epoch 4, batch 11200, giga_loss[loss=0.2848, simple_loss=0.3523, pruned_loss=0.1086, over 28466.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.4016, pruned_loss=0.1506, over 5653605.66 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.384, pruned_loss=0.1243, over 5747001.50 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4031, pruned_loss=0.1525, over 5649253.82 frames. ], batch size: 65, lr: 8.02e-03, grad_scale: 8.0
+2023-03-02 02:51:46,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:51:48,442 INFO [optim.py:369] (1/2) 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 02:52:18,219 INFO [train.py:968] (1/2) Epoch 4, batch 11250, giga_loss[loss=0.331, simple_loss=0.3917, pruned_loss=0.1352, over 28912.00 frames. ], tot_loss[loss=0.3518, simple_loss=0.4015, pruned_loss=0.1511, over 5651998.50 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3843, pruned_loss=0.1245, over 5746586.12 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4027, pruned_loss=0.1528, over 5647393.90 frames. ], batch size: 186, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:52:20,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:53:07,329 INFO [train.py:968] (1/2) Epoch 4, batch 11300, giga_loss[loss=0.307, simple_loss=0.3738, pruned_loss=0.1201, over 28891.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.402, pruned_loss=0.1517, over 5650750.92 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3847, pruned_loss=0.1247, over 5748265.24 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.4031, pruned_loss=0.1534, over 5643867.53 frames. ], batch size: 145, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:53:26,612 INFO [optim.py:369] (1/2) 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 02:53:55,381 INFO [train.py:968] (1/2) Epoch 4, batch 11350, giga_loss[loss=0.472, simple_loss=0.4669, pruned_loss=0.2385, over 26656.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.4051, pruned_loss=0.1543, over 5658530.21 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3843, pruned_loss=0.1245, over 5750831.54 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4067, pruned_loss=0.1565, over 5648623.32 frames. ], batch size: 555, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:54:00,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3888, 2.8661, 1.5189, 1.3147], device='cuda:1'), covar=tensor([0.0799, 0.0334, 0.0762, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0474, 0.0313, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 02:54:03,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:05,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=147505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:21,883 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 02:54:28,359 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 02:54:30,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0897, 1.2184, 1.2562, 1.1936], device='cuda:1'), covar=tensor([0.0846, 0.0876, 0.1234, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0764, 0.0637, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 02:54:32,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:35,924 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:54:38,734 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 11400, libri_loss[loss=0.2762, simple_loss=0.3425, pruned_loss=0.1049, over 29648.00 frames. ], tot_loss[loss=0.3575, simple_loss=0.4053, pruned_loss=0.1549, over 5648917.78 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3841, pruned_loss=0.1244, over 5754815.05 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4072, pruned_loss=0.1575, over 5635362.52 frames. ], batch size: 69, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:54:59,485 INFO [optim.py:369] (1/2) 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 02:55:05,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=147569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:55:25,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1553, 1.2792, 1.2350, 1.2974], device='cuda:1'), covar=tensor([0.1244, 0.1517, 0.1684, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0763, 0.0642, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 02:55:27,595 INFO [train.py:968] (1/2) Epoch 4, batch 11450, giga_loss[loss=0.3728, simple_loss=0.419, pruned_loss=0.1633, over 28631.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.4052, pruned_loss=0.1546, over 5652703.44 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3842, pruned_loss=0.1243, over 5750224.82 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4074, pruned_loss=0.1579, over 5642070.74 frames. ], batch size: 60, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:55:32,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5438, 2.3442, 1.5250, 0.5959], device='cuda:1'), covar=tensor([0.3348, 0.1518, 0.1883, 0.3191], device='cuda:1'), in_proj_covar=tensor([0.1324, 0.1264, 0.1347, 0.1154], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 02:55:37,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6742, 1.4839, 1.1739, 1.2638], device='cuda:1'), covar=tensor([0.0571, 0.0557, 0.0943, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0472, 0.0513, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-02 02:55:45,423 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=147612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:56:05,588 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:56:13,280 INFO [train.py:968] (1/2) Epoch 4, batch 11500, giga_loss[loss=0.3508, simple_loss=0.3926, pruned_loss=0.1545, over 28958.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4037, pruned_loss=0.1532, over 5661142.50 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3836, pruned_loss=0.1238, over 5753445.86 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4064, pruned_loss=0.1568, over 5648121.84 frames. ], batch size: 186, lr: 8.02e-03, grad_scale: 4.0
+2023-03-02 02:56:32,226 INFO [optim.py:369] (1/2) 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 02:57:04,260 INFO [train.py:968] (1/2) Epoch 4, batch 11550, giga_loss[loss=0.3675, simple_loss=0.4111, pruned_loss=0.162, over 28001.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4049, pruned_loss=0.1537, over 5651949.97 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3839, pruned_loss=0.124, over 5748545.25 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4072, pruned_loss=0.1569, over 5644203.01 frames. ], batch size: 412, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 02:57:11,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:57:47,102 INFO [train.py:968] (1/2) Epoch 4, batch 11600, giga_loss[loss=0.3225, simple_loss=0.3917, pruned_loss=0.1267, over 29022.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4042, pruned_loss=0.1519, over 5673748.42 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3834, pruned_loss=0.1236, over 5753852.27 frames. ], giga_tot_loss[loss=0.3596, simple_loss=0.4072, pruned_loss=0.1559, over 5659612.18 frames. ], batch size: 136, lr: 8.01e-03, grad_scale: 8.0
+2023-03-02 02:57:48,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2568, 1.4114, 1.1780, 1.5623], device='cuda:1'), covar=tensor([0.2302, 0.2115, 0.2134, 0.2035], device='cuda:1'), in_proj_covar=tensor([0.1074, 0.0864, 0.0964, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 02:57:58,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=147755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:58:01,897 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:58:31,716 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 11650, giga_loss[loss=0.3963, simple_loss=0.4314, pruned_loss=0.1806, over 28685.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4065, pruned_loss=0.1546, over 5659497.40 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3833, pruned_loss=0.1237, over 5754255.40 frames. ], giga_tot_loss[loss=0.3632, simple_loss=0.4095, pruned_loss=0.1584, over 5645907.72 frames. ], batch size: 99, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 02:58:43,196 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:59:12,907 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=147827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 02:59:26,854 INFO [train.py:968] (1/2) Epoch 4, batch 11700, libri_loss[loss=0.3068, simple_loss=0.3817, pruned_loss=0.1159, over 29526.00 frames. ], tot_loss[loss=0.3602, simple_loss=0.4082, pruned_loss=0.1561, over 5651815.48 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3836, pruned_loss=0.1238, over 5746818.65 frames. ], giga_tot_loss[loss=0.365, simple_loss=0.4108, pruned_loss=0.1596, over 5646830.86 frames. ], batch size: 83, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 02:59:45,420 INFO [optim.py:369] (1/2) 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 03:00:12,683 INFO [train.py:968] (1/2) Epoch 4, batch 11750, giga_loss[loss=0.353, simple_loss=0.4045, pruned_loss=0.1508, over 28900.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4077, pruned_loss=0.1559, over 5655979.39 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3835, pruned_loss=0.1238, over 5750942.92 frames. ], giga_tot_loss[loss=0.3648, simple_loss=0.4105, pruned_loss=0.1596, over 5645904.27 frames. ], batch size: 136, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 03:00:39,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3770, 2.6319, 1.4720, 1.3130], device='cuda:1'), covar=tensor([0.0800, 0.0420, 0.0791, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0473, 0.0312, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 03:00:57,810 INFO [train.py:968] (1/2) Epoch 4, batch 11800, giga_loss[loss=0.3674, simple_loss=0.4188, pruned_loss=0.158, over 28258.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4078, pruned_loss=0.1544, over 5650006.45 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3835, pruned_loss=0.1238, over 5744644.52 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4107, pruned_loss=0.1582, over 5644556.80 frames. ], batch size: 368, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 03:01:17,433 INFO [optim.py:369] (1/2) 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 03:01:48,332 INFO [train.py:968] (1/2) Epoch 4, batch 11850, giga_loss[loss=0.3965, simple_loss=0.4226, pruned_loss=0.1852, over 26628.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4071, pruned_loss=0.153, over 5647445.69 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3837, pruned_loss=0.1238, over 5746413.08 frames. ], giga_tot_loss[loss=0.3612, simple_loss=0.4095, pruned_loss=0.1564, over 5639961.02 frames. ], batch size: 555, lr: 8.01e-03, grad_scale: 4.0
+2023-03-02 03:02:01,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:02:17,352 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6092, 2.1554, 1.5058, 0.6927], device='cuda:1'), covar=tensor([0.2555, 0.1444, 0.1524, 0.2538], device='cuda:1'), in_proj_covar=tensor([0.1314, 0.1266, 0.1326, 0.1137], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 03:02:33,952 INFO [train.py:968] (1/2) Epoch 4, batch 11900, giga_loss[loss=0.345, simple_loss=0.3999, pruned_loss=0.1451, over 28551.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4054, pruned_loss=0.1515, over 5650744.44 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.383, pruned_loss=0.1233, over 5750435.70 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.4084, pruned_loss=0.1553, over 5639360.83 frames. ], batch size: 78, lr: 8.00e-03, grad_scale: 4.0
+2023-03-02 03:02:44,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:02:51,948 INFO [optim.py:369] (1/2) 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 03:03:05,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:03:18,932 INFO [train.py:968] (1/2) Epoch 4, batch 11950, giga_loss[loss=0.3161, simple_loss=0.3806, pruned_loss=0.1258, over 28495.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4031, pruned_loss=0.1499, over 5654709.62 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.383, pruned_loss=0.1234, over 5744503.61 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.406, pruned_loss=0.1535, over 5648229.20 frames. ], batch size: 71, lr: 8.00e-03, grad_scale: 4.0
+2023-03-02 03:04:05,542 INFO [train.py:968] (1/2) Epoch 4, batch 12000, giga_loss[loss=0.4188, simple_loss=0.44, pruned_loss=0.1988, over 27585.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4049, pruned_loss=0.151, over 5657621.91 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3831, pruned_loss=0.1232, over 5747435.10 frames. ], giga_tot_loss[loss=0.3584, simple_loss=0.4076, pruned_loss=0.1546, over 5647631.60 frames. ], batch size: 472, lr: 8.00e-03, grad_scale: 8.0
+2023-03-02 03:04:05,542 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 03:04:14,258 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 03:04:25,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:04:27,060 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:04:44,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:04:54,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:04:56,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1419, 1.3997, 1.2363, 1.0034], device='cuda:1'), covar=tensor([0.1925, 0.1925, 0.1822, 0.1718], device='cuda:1'), in_proj_covar=tensor([0.1073, 0.0856, 0.0968, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:05:01,869 INFO [train.py:968] (1/2) Epoch 4, batch 12050, giga_loss[loss=0.3253, simple_loss=0.3948, pruned_loss=0.1279, over 28923.00 frames. ], tot_loss[loss=0.3536, simple_loss=0.4048, pruned_loss=0.1512, over 5649940.18 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.383, pruned_loss=0.1233, over 5741569.44 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4076, pruned_loss=0.1547, over 5644419.55 frames. ], batch size: 145, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:05:10,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:14,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:27,988 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:31,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:48,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 03:05:49,234 INFO [train.py:968] (1/2) Epoch 4, batch 12100, giga_loss[loss=0.362, simple_loss=0.4043, pruned_loss=0.1598, over 28584.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4032, pruned_loss=0.1503, over 5665297.57 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3833, pruned_loss=0.1233, over 5742854.46 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4059, pruned_loss=0.1541, over 5657154.95 frames. ], batch size: 307, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:05:57,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:05:59,821 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:06:11,975 INFO [optim.py:369] (1/2) 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 03:06:14,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5804, 1.7641, 1.1713, 0.9534], device='cuda:1'), covar=tensor([0.0953, 0.0712, 0.0776, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.1311, 0.1058, 0.1083, 0.1118], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 03:06:35,832 INFO [train.py:968] (1/2) Epoch 4, batch 12150, giga_loss[loss=0.2933, simple_loss=0.3699, pruned_loss=0.1084, over 28494.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.4033, pruned_loss=0.1508, over 5668692.69 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3833, pruned_loss=0.1234, over 5742798.50 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.406, pruned_loss=0.1545, over 5659896.41 frames. ], batch size: 60, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:06:47,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:00,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:04,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:07,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2044, 1.4217, 1.1873, 1.2365], device='cuda:1'), covar=tensor([0.2167, 0.2020, 0.2000, 0.1922], device='cuda:1'), in_proj_covar=tensor([0.1071, 0.0855, 0.0963, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:07:24,322 INFO [train.py:968] (1/2) Epoch 4, batch 12200, giga_loss[loss=0.3612, simple_loss=0.4198, pruned_loss=0.1513, over 28721.00 frames. ], tot_loss[loss=0.3573, simple_loss=0.4066, pruned_loss=0.154, over 5673189.60 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3834, pruned_loss=0.1235, over 5746370.45 frames. ], giga_tot_loss[loss=0.3622, simple_loss=0.4093, pruned_loss=0.1575, over 5661392.30 frames. ], batch size: 262, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:07:26,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:30,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:32,529 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:07:46,518 INFO [optim.py:369] (1/2) 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 03:07:56,545 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:08:10,062 INFO [train.py:968] (1/2) Epoch 4, batch 12250, giga_loss[loss=0.3469, simple_loss=0.3982, pruned_loss=0.1478, over 28937.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.4074, pruned_loss=0.1546, over 5667373.99 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3838, pruned_loss=0.1237, over 5749532.55 frames. ], giga_tot_loss[loss=0.3629, simple_loss=0.4097, pruned_loss=0.1581, over 5653188.97 frames. ], batch size: 199, lr: 8.00e-03, grad_scale: 2.0
+2023-03-02 03:08:23,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2909, 1.9784, 1.8479, 1.8494], device='cuda:1'), covar=tensor([0.1095, 0.2105, 0.1531, 0.1518], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0765, 0.0641, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 03:08:42,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:08:57,312 INFO [train.py:968] (1/2) Epoch 4, batch 12300, giga_loss[loss=0.3233, simple_loss=0.3874, pruned_loss=0.1296, over 28838.00 frames. ], tot_loss[loss=0.3548, simple_loss=0.4056, pruned_loss=0.152, over 5680663.12 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3838, pruned_loss=0.1235, over 5748814.27 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4081, pruned_loss=0.1558, over 5667654.15 frames. ], batch size: 186, lr: 7.99e-03, grad_scale: 2.0
+2023-03-02 03:09:18,141 INFO [optim.py:369] (1/2) 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 03:09:44,325 INFO [train.py:968] (1/2) Epoch 4, batch 12350, giga_loss[loss=0.2955, simple_loss=0.3565, pruned_loss=0.1172, over 28766.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4045, pruned_loss=0.1512, over 5671695.60 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3834, pruned_loss=0.1232, over 5752324.12 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4073, pruned_loss=0.1551, over 5656523.03 frames. ], batch size: 99, lr: 7.99e-03, grad_scale: 2.0
+2023-03-02 03:10:29,751 INFO [train.py:968] (1/2) Epoch 4, batch 12400, giga_loss[loss=0.3171, simple_loss=0.3744, pruned_loss=0.1299, over 28638.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4032, pruned_loss=0.1493, over 5683435.79 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.383, pruned_loss=0.1229, over 5756391.73 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4062, pruned_loss=0.1533, over 5666010.86 frames. ], batch size: 71, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:10:48,307 INFO [optim.py:369] (1/2) 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 03:10:55,637 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:10:58,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:11:06,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:11:11,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 03:11:17,590 INFO [train.py:968] (1/2) Epoch 4, batch 12450, giga_loss[loss=0.3467, simple_loss=0.3751, pruned_loss=0.1592, over 23647.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4035, pruned_loss=0.15, over 5668289.70 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.383, pruned_loss=0.1228, over 5746136.01 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4061, pruned_loss=0.1535, over 5662269.55 frames. ], batch size: 705, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:11:25,552 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0319, 1.1372, 0.8313, 0.4519], device='cuda:1'), covar=tensor([0.0758, 0.0667, 0.0584, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.1075, 0.1089, 0.1139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 03:11:28,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:11:48,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=148625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:12:04,884 INFO [train.py:968] (1/2) Epoch 4, batch 12500, giga_loss[loss=0.3988, simple_loss=0.4241, pruned_loss=0.1867, over 26550.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4034, pruned_loss=0.1502, over 5675766.55 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3833, pruned_loss=0.123, over 5748610.46 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4056, pruned_loss=0.1533, over 5667378.62 frames. ], batch size: 555, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:12:25,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4939, 1.6320, 1.3273, 1.5630], device='cuda:1'), covar=tensor([0.0796, 0.0321, 0.0344, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0138, 0.0140, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0048, 0.0036, 0.0032, 0.0054], device='cuda:1')
+2023-03-02 03:12:27,605 INFO [optim.py:369] (1/2) 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 03:12:39,139 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:12:52,361 INFO [train.py:968] (1/2) Epoch 4, batch 12550, giga_loss[loss=0.3218, simple_loss=0.3757, pruned_loss=0.134, over 28868.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4014, pruned_loss=0.1496, over 5662925.44 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3835, pruned_loss=0.1232, over 5738442.58 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4033, pruned_loss=0.1524, over 5663361.41 frames. ], batch size: 174, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:13:09,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-02 03:13:23,518 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:13:26,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:13:38,079 INFO [train.py:968] (1/2) Epoch 4, batch 12600, giga_loss[loss=0.3319, simple_loss=0.3726, pruned_loss=0.1456, over 28655.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3973, pruned_loss=0.1472, over 5678450.32 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3833, pruned_loss=0.123, over 5742047.34 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.3996, pruned_loss=0.1504, over 5673600.65 frames. ], batch size: 92, lr: 7.99e-03, grad_scale: 4.0
+2023-03-02 03:13:50,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3226, 3.2518, 1.3595, 1.3099], device='cuda:1'), covar=tensor([0.0966, 0.0495, 0.0905, 0.1446], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0474, 0.0310, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 03:13:53,579 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:13:58,821 INFO [optim.py:369] (1/2) 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 03:14:01,256 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:04,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:25,758 INFO [train.py:968] (1/2) Epoch 4, batch 12650, giga_loss[loss=0.398, simple_loss=0.4191, pruned_loss=0.1885, over 28403.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3956, pruned_loss=0.1462, over 5673663.39 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3841, pruned_loss=0.1234, over 5731093.61 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.397, pruned_loss=0.1489, over 5678349.39 frames. ], batch size: 78, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:14:30,578 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:30,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0118, 1.1618, 0.9732, 0.6314], device='cuda:1'), covar=tensor([0.0719, 0.0665, 0.0456, 0.0606], device='cuda:1'), in_proj_covar=tensor([0.1351, 0.1085, 0.1102, 0.1143], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1')
+2023-03-02 03:14:37,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6538, 4.3604, 4.3719, 1.8577], device='cuda:1'), covar=tensor([0.0365, 0.0366, 0.0630, 0.2092], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0708, 0.0846, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 03:14:50,801 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=148822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:50,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=148822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:14:53,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4502, 4.7336, 2.2873, 2.1493], device='cuda:1'), covar=tensor([0.0657, 0.0284, 0.0689, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0475, 0.0308, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 03:14:53,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=148825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:15:10,086 INFO [train.py:968] (1/2) Epoch 4, batch 12700, giga_loss[loss=0.4051, simple_loss=0.4308, pruned_loss=0.1898, over 26616.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3956, pruned_loss=0.1461, over 5685904.46 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3845, pruned_loss=0.1239, over 5737555.48 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.3968, pruned_loss=0.1487, over 5682056.42 frames. ], batch size: 555, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:15:18,385 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 03:15:23,352 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=148854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:15:32,472 INFO [optim.py:369] (1/2) 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 03:15:58,361 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 12750, giga_loss[loss=0.3222, simple_loss=0.3833, pruned_loss=0.1306, over 27964.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3968, pruned_loss=0.1461, over 5674748.20 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3844, pruned_loss=0.1239, over 5730615.43 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.398, pruned_loss=0.1486, over 5676489.31 frames. ], batch size: 412, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:16:37,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2866, 1.8836, 1.5871, 1.5626], device='cuda:1'), covar=tensor([0.1590, 0.1906, 0.1276, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0777, 0.0754, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 03:16:50,476 INFO [train.py:968] (1/2) Epoch 4, batch 12800, giga_loss[loss=0.2812, simple_loss=0.3401, pruned_loss=0.1112, over 24211.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3932, pruned_loss=0.1418, over 5669976.44 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3841, pruned_loss=0.1239, over 5733049.10 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3946, pruned_loss=0.1441, over 5668294.26 frames. ], batch size: 705, lr: 7.98e-03, grad_scale: 8.0
+2023-03-02 03:17:14,352 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 12850, giga_loss[loss=0.271, simple_loss=0.3482, pruned_loss=0.09686, over 28976.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3885, pruned_loss=0.1371, over 5671558.27 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3831, pruned_loss=0.1235, over 5737601.95 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3908, pruned_loss=0.1398, over 5663914.13 frames. ], batch size: 155, lr: 7.98e-03, grad_scale: 8.0
+2023-03-02 03:18:28,147 INFO [train.py:968] (1/2) Epoch 4, batch 12900, giga_loss[loss=0.3132, simple_loss=0.3769, pruned_loss=0.1248, over 28466.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3852, pruned_loss=0.1338, over 5671278.24 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3827, pruned_loss=0.1232, over 5743278.58 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3875, pruned_loss=0.1366, over 5657723.71 frames. ], batch size: 336, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:18:30,565 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:18:43,097 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:19:20,824 INFO [train.py:968] (1/2) Epoch 4, batch 12950, giga_loss[loss=0.3151, simple_loss=0.3779, pruned_loss=0.1262, over 27965.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3797, pruned_loss=0.1287, over 5677963.78 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3816, pruned_loss=0.1228, over 5747257.12 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3826, pruned_loss=0.1317, over 5661583.09 frames. ], batch size: 412, lr: 7.98e-03, grad_scale: 4.0
+2023-03-02 03:19:35,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:20:04,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5415, 2.1305, 1.1959, 1.0472], device='cuda:1'), covar=tensor([0.0987, 0.0600, 0.0712, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1039, 0.1040, 0.1090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 03:20:08,864 INFO [train.py:968] (1/2) Epoch 4, batch 13000, giga_loss[loss=0.2791, simple_loss=0.369, pruned_loss=0.09459, over 28596.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3773, pruned_loss=0.1249, over 5679697.30 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3801, pruned_loss=0.1221, over 5750900.47 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3808, pruned_loss=0.1279, over 5661813.65 frames. ], batch size: 336, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:20:34,187 INFO [optim.py:369] (1/2) 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 03:21:04,930 INFO [train.py:968] (1/2) Epoch 4, batch 13050, giga_loss[loss=0.318, simple_loss=0.3848, pruned_loss=0.1256, over 28256.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3764, pruned_loss=0.124, over 5666566.28 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.38, pruned_loss=0.1221, over 5752863.74 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3792, pruned_loss=0.1265, over 5649306.33 frames. ], batch size: 368, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:21:08,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:21:09,540 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:21:13,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:21:42,632 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 13100, giga_loss[loss=0.3308, simple_loss=0.3731, pruned_loss=0.1442, over 26710.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3745, pruned_loss=0.1225, over 5669774.57 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3799, pruned_loss=0.1222, over 5755196.32 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3767, pruned_loss=0.1244, over 5653032.46 frames. ], batch size: 555, lr: 7.97e-03, grad_scale: 2.0
+2023-03-02 03:22:20,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:22:20,641 INFO [optim.py:369] (1/2) 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 03:22:47,833 INFO [train.py:968] (1/2) Epoch 4, batch 13150, giga_loss[loss=0.3122, simple_loss=0.3792, pruned_loss=0.1226, over 28729.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5672459.98 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3795, pruned_loss=0.122, over 5757153.46 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3731, pruned_loss=0.1216, over 5656211.24 frames. ], batch size: 284, lr: 7.97e-03, grad_scale: 2.0
+2023-03-02 03:22:56,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8869, 1.1308, 4.0055, 3.2595], device='cuda:1'), covar=tensor([0.1614, 0.2161, 0.0329, 0.0526], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0500, 0.0705, 0.0562], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:23:35,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:23:37,259 INFO [train.py:968] (1/2) Epoch 4, batch 13200, giga_loss[loss=0.3496, simple_loss=0.4046, pruned_loss=0.1473, over 28844.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3695, pruned_loss=0.1191, over 5664183.01 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3792, pruned_loss=0.122, over 5749758.68 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3713, pruned_loss=0.1204, over 5657165.83 frames. ], batch size: 199, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:23:38,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:01,593 INFO [optim.py:369] (1/2) 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 03:24:07,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:25,573 INFO [train.py:968] (1/2) Epoch 4, batch 13250, giga_loss[loss=0.2669, simple_loss=0.3398, pruned_loss=0.09706, over 28473.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3687, pruned_loss=0.1179, over 5671231.31 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3784, pruned_loss=0.1215, over 5751280.56 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3705, pruned_loss=0.1192, over 5661554.89 frames. ], batch size: 78, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:24:40,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:42,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:24:48,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=149419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:25:09,802 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:25:10,854 INFO [train.py:968] (1/2) Epoch 4, batch 13300, giga_loss[loss=0.3161, simple_loss=0.3679, pruned_loss=0.1322, over 26743.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3672, pruned_loss=0.1165, over 5668783.19 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.378, pruned_loss=0.1214, over 5752420.40 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3687, pruned_loss=0.1176, over 5657591.34 frames. ], batch size: 555, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:25:34,990 INFO [optim.py:369] (1/2) 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 03:25:53,569 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 13350, giga_loss[loss=0.2896, simple_loss=0.3551, pruned_loss=0.112, over 28293.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3641, pruned_loss=0.1139, over 5673867.86 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3773, pruned_loss=0.121, over 5757015.71 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3656, pruned_loss=0.1149, over 5658755.62 frames. ], batch size: 368, lr: 7.97e-03, grad_scale: 4.0
+2023-03-02 03:26:25,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-02 03:26:48,763 INFO [train.py:968] (1/2) Epoch 4, batch 13400, giga_loss[loss=0.3091, simple_loss=0.3547, pruned_loss=0.1317, over 26556.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3604, pruned_loss=0.1119, over 5670281.34 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3763, pruned_loss=0.1206, over 5753084.31 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3621, pruned_loss=0.1128, over 5657651.47 frames. ], batch size: 555, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:27:13,174 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:27:15,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:27:17,258 INFO [optim.py:369] (1/2) 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 03:27:42,986 INFO [train.py:968] (1/2) Epoch 4, batch 13450, giga_loss[loss=0.2919, simple_loss=0.362, pruned_loss=0.1109, over 28866.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3583, pruned_loss=0.1116, over 5656592.03 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3761, pruned_loss=0.1208, over 5751116.85 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1118, over 5645092.41 frames. ], batch size: 174, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:27:44,497 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=149594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:28:19,868 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=149627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:28:21,967 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=149630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:28:36,507 INFO [train.py:968] (1/2) Epoch 4, batch 13500, giga_loss[loss=0.3016, simple_loss=0.366, pruned_loss=0.1186, over 27973.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3578, pruned_loss=0.1119, over 5644752.28 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3761, pruned_loss=0.1208, over 5743526.93 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3583, pruned_loss=0.112, over 5642021.96 frames. ], batch size: 412, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:28:53,514 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:29:33,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-02 03:29:33,535 INFO [train.py:968] (1/2) Epoch 4, batch 13550, giga_loss[loss=0.2663, simple_loss=0.3395, pruned_loss=0.09655, over 28672.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1127, over 5634270.60 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3759, pruned_loss=0.1208, over 5746620.46 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3592, pruned_loss=0.1126, over 5627008.26 frames. ], batch size: 92, lr: 7.96e-03, grad_scale: 2.0
+2023-03-02 03:30:25,524 INFO [train.py:968] (1/2) Epoch 4, batch 13600, giga_loss[loss=0.2987, simple_loss=0.3729, pruned_loss=0.1123, over 28897.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3614, pruned_loss=0.1128, over 5653485.02 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3747, pruned_loss=0.1201, over 5752772.51 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3622, pruned_loss=0.1131, over 5638037.11 frames. ], batch size: 112, lr: 7.96e-03, grad_scale: 4.0
+2023-03-02 03:30:51,848 INFO [optim.py:369] (1/2) 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 03:31:20,001 INFO [train.py:968] (1/2) Epoch 4, batch 13650, giga_loss[loss=0.3133, simple_loss=0.3823, pruned_loss=0.1222, over 28449.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3615, pruned_loss=0.1128, over 5644215.65 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3734, pruned_loss=0.1194, over 5747603.10 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3629, pruned_loss=0.1134, over 5632912.52 frames. ], batch size: 336, lr: 7.96e-03, grad_scale: 4.0
+2023-03-02 03:32:25,115 INFO [train.py:968] (1/2) Epoch 4, batch 13700, giga_loss[loss=0.2638, simple_loss=0.3427, pruned_loss=0.09242, over 29101.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3618, pruned_loss=0.1134, over 5648036.76 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.373, pruned_loss=0.1194, over 5750630.34 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3631, pruned_loss=0.1138, over 5634241.90 frames. ], batch size: 128, lr: 7.96e-03, grad_scale: 4.0
+2023-03-02 03:32:27,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 03:32:52,464 INFO [optim.py:369] (1/2) 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 03:33:20,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2529, 1.3585, 1.1134, 1.5188], device='cuda:1'), covar=tensor([0.2308, 0.2018, 0.2006, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.1075, 0.0844, 0.0964, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:33:21,476 INFO [train.py:968] (1/2) Epoch 4, batch 13750, libri_loss[loss=0.2489, simple_loss=0.3121, pruned_loss=0.09288, over 29343.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3601, pruned_loss=0.1122, over 5657493.10 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3723, pruned_loss=0.1191, over 5753873.20 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3614, pruned_loss=0.1126, over 5640950.93 frames. ], batch size: 67, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:33:26,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-02 03:33:38,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:34:14,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4281, 2.6835, 1.4217, 1.3813], device='cuda:1'), covar=tensor([0.0759, 0.0377, 0.0845, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0463, 0.0314, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 03:34:18,932 INFO [train.py:968] (1/2) Epoch 4, batch 13800, giga_loss[loss=0.2764, simple_loss=0.356, pruned_loss=0.09842, over 28809.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3585, pruned_loss=0.1098, over 5656353.04 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3715, pruned_loss=0.1187, over 5758748.21 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3599, pruned_loss=0.1103, over 5636146.57 frames. ], batch size: 243, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:34:37,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=149958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:34:49,952 INFO [optim.py:369] (1/2) 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 03:35:17,194 INFO [train.py:968] (1/2) Epoch 4, batch 13850, giga_loss[loss=0.267, simple_loss=0.3327, pruned_loss=0.1006, over 28984.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3562, pruned_loss=0.1083, over 5666499.07 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3711, pruned_loss=0.1185, over 5764344.99 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1084, over 5640862.67 frames. ], batch size: 213, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:36:12,909 INFO [train.py:968] (1/2) Epoch 4, batch 13900, giga_loss[loss=0.2294, simple_loss=0.3131, pruned_loss=0.07287, over 28994.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3535, pruned_loss=0.1075, over 5666607.36 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3702, pruned_loss=0.118, over 5762427.78 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3543, pruned_loss=0.1076, over 5642284.74 frames. ], batch size: 155, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:36:16,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3122, 1.5417, 4.5300, 3.3611], device='cuda:1'), covar=tensor([0.1526, 0.1928, 0.0300, 0.0646], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0509, 0.0710, 0.0569], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:36:42,848 INFO [optim.py:369] (1/2) 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 03:36:44,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5666, 1.4149, 1.2090, 1.2609], device='cuda:1'), covar=tensor([0.0598, 0.0514, 0.0922, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0458, 0.0515, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 03:37:04,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7844, 4.5097, 4.4752, 2.0499], device='cuda:1'), covar=tensor([0.0325, 0.0334, 0.0707, 0.1863], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0676, 0.0789, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 03:37:13,907 INFO [train.py:968] (1/2) Epoch 4, batch 13950, giga_loss[loss=0.2622, simple_loss=0.3335, pruned_loss=0.09543, over 28083.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3541, pruned_loss=0.1085, over 5667290.25 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3702, pruned_loss=0.118, over 5763647.80 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3547, pruned_loss=0.1085, over 5646167.42 frames. ], batch size: 412, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:37:29,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4959, 3.3389, 1.4765, 1.4669], device='cuda:1'), covar=tensor([0.0807, 0.0306, 0.0891, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0464, 0.0318, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 03:38:03,528 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150139.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 03:38:08,446 INFO [train.py:968] (1/2) Epoch 4, batch 14000, giga_loss[loss=0.2837, simple_loss=0.3584, pruned_loss=0.1045, over 28861.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.356, pruned_loss=0.1089, over 5677498.99 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.37, pruned_loss=0.1179, over 5763765.76 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.356, pruned_loss=0.1086, over 5656923.29 frames. ], batch size: 186, lr: 7.95e-03, grad_scale: 8.0
+2023-03-02 03:38:39,521 INFO [optim.py:369] (1/2) 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 03:39:12,851 INFO [train.py:968] (1/2) Epoch 4, batch 14050, giga_loss[loss=0.2965, simple_loss=0.3768, pruned_loss=0.1081, over 28440.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3577, pruned_loss=0.1083, over 5682863.74 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3699, pruned_loss=0.1178, over 5765182.53 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3577, pruned_loss=0.1081, over 5664819.43 frames. ], batch size: 336, lr: 7.95e-03, grad_scale: 8.0
+2023-03-02 03:39:24,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-02 03:40:15,419 INFO [train.py:968] (1/2) Epoch 4, batch 14100, libri_loss[loss=0.2914, simple_loss=0.362, pruned_loss=0.1104, over 26026.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3545, pruned_loss=0.1061, over 5680391.19 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3695, pruned_loss=0.1176, over 5763934.80 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3543, pruned_loss=0.1057, over 5663616.88 frames. ], batch size: 136, lr: 7.95e-03, grad_scale: 4.0
+2023-03-02 03:40:43,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4058, 2.2632, 1.6066, 0.6404], device='cuda:1'), covar=tensor([0.2083, 0.1027, 0.2045, 0.2322], device='cuda:1'), in_proj_covar=tensor([0.1295, 0.1246, 0.1317, 0.1111], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 03:40:53,906 INFO [optim.py:369] (1/2) 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 03:41:10,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:41:23,218 INFO [train.py:968] (1/2) Epoch 4, batch 14150, giga_loss[loss=0.2957, simple_loss=0.3636, pruned_loss=0.1139, over 28922.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3548, pruned_loss=0.1069, over 5688516.05 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3693, pruned_loss=0.1175, over 5764840.67 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3547, pruned_loss=0.1066, over 5673757.11 frames. ], batch size: 199, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:42:17,804 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 14200, giga_loss[loss=0.2775, simple_loss=0.3695, pruned_loss=0.09277, over 28963.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.357, pruned_loss=0.108, over 5667530.06 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3689, pruned_loss=0.1171, over 5766247.01 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.357, pruned_loss=0.1078, over 5652666.92 frames. ], batch size: 155, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:43:08,487 INFO [optim.py:369] (1/2) 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 03:43:26,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5291, 1.0203, 2.8839, 2.6572], device='cuda:1'), covar=tensor([0.1710, 0.2163, 0.0450, 0.0611], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0513, 0.0702, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:43:37,513 INFO [train.py:968] (1/2) Epoch 4, batch 14250, giga_loss[loss=0.3104, simple_loss=0.3926, pruned_loss=0.1141, over 28955.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3598, pruned_loss=0.1068, over 5666233.36 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3687, pruned_loss=0.1171, over 5768984.22 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3597, pruned_loss=0.1066, over 5650094.02 frames. ], batch size: 199, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:44:19,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:44:22,144 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 14300, giga_loss[loss=0.2944, simple_loss=0.3741, pruned_loss=0.1074, over 29127.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3591, pruned_loss=0.1054, over 5652374.54 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3691, pruned_loss=0.1173, over 5769050.54 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3585, pruned_loss=0.1047, over 5638089.11 frames. ], batch size: 200, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:44:57,055 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:09,967 INFO [optim.py:369] (1/2) 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 03:45:13,111 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:17,620 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:22,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:45:37,297 INFO [train.py:968] (1/2) Epoch 4, batch 14350, giga_loss[loss=0.2928, simple_loss=0.3711, pruned_loss=0.1073, over 28914.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3593, pruned_loss=0.1046, over 5665628.87 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3692, pruned_loss=0.1174, over 5771668.42 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3586, pruned_loss=0.1037, over 5649293.97 frames. ], batch size: 213, lr: 7.94e-03, grad_scale: 2.0
+2023-03-02 03:45:58,899 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:46:05,786 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 14400, giga_loss[loss=0.2637, simple_loss=0.3412, pruned_loss=0.0931, over 28992.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3603, pruned_loss=0.1062, over 5674108.69 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3686, pruned_loss=0.1172, over 5774160.98 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3599, pruned_loss=0.1053, over 5655864.58 frames. ], batch size: 199, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:46:55,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3343, 2.6259, 1.3052, 1.3562], device='cuda:1'), covar=tensor([0.0808, 0.0346, 0.0831, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0462, 0.0312, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 03:47:05,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2949, 1.4246, 1.1778, 1.4594], device='cuda:1'), covar=tensor([0.2165, 0.1946, 0.1999, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.1065, 0.0830, 0.0953, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:47:13,377 INFO [optim.py:369] (1/2) 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 03:47:44,200 INFO [train.py:968] (1/2) Epoch 4, batch 14450, giga_loss[loss=0.2846, simple_loss=0.36, pruned_loss=0.1046, over 29069.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3587, pruned_loss=0.1065, over 5667844.28 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3687, pruned_loss=0.1173, over 5767607.60 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3581, pruned_loss=0.1055, over 5656876.77 frames. ], batch size: 155, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:48:58,575 INFO [train.py:968] (1/2) Epoch 4, batch 14500, giga_loss[loss=0.2647, simple_loss=0.3459, pruned_loss=0.09176, over 28890.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3594, pruned_loss=0.1073, over 5661109.44 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3682, pruned_loss=0.117, over 5759728.79 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3592, pruned_loss=0.1066, over 5657178.02 frames. ], batch size: 145, lr: 7.94e-03, grad_scale: 4.0
+2023-03-02 03:49:19,222 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150657.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 03:49:25,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 03:49:34,907 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:50:12,003 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=150689.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 03:50:18,279 INFO [train.py:968] (1/2) Epoch 4, batch 14550, giga_loss[loss=0.2564, simple_loss=0.3341, pruned_loss=0.0893, over 28728.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3545, pruned_loss=0.1042, over 5668828.30 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3677, pruned_loss=0.1166, over 5761997.62 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3546, pruned_loss=0.1037, over 5661645.92 frames. ], batch size: 262, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:51:22,772 INFO [train.py:968] (1/2) Epoch 4, batch 14600, giga_loss[loss=0.2761, simple_loss=0.3515, pruned_loss=0.1003, over 28309.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3531, pruned_loss=0.1037, over 5663937.05 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3673, pruned_loss=0.1164, over 5756539.24 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3531, pruned_loss=0.103, over 5659810.76 frames. ], batch size: 368, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:51:54,945 INFO [optim.py:369] (1/2) 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 03:52:12,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3281, 1.5689, 1.3526, 1.5160], device='cuda:1'), covar=tensor([0.2241, 0.1946, 0.1909, 0.2040], device='cuda:1'), in_proj_covar=tensor([0.1064, 0.0834, 0.0956, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:52:22,621 INFO [train.py:968] (1/2) Epoch 4, batch 14650, giga_loss[loss=0.2716, simple_loss=0.3385, pruned_loss=0.1024, over 28908.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3512, pruned_loss=0.1033, over 5671249.40 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3666, pruned_loss=0.1161, over 5758804.05 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1026, over 5662862.10 frames. ], batch size: 186, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:52:43,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 03:53:24,376 INFO [train.py:968] (1/2) Epoch 4, batch 14700, giga_loss[loss=0.2971, simple_loss=0.375, pruned_loss=0.1096, over 28084.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3572, pruned_loss=0.107, over 5679692.96 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3668, pruned_loss=0.1161, over 5757948.99 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3569, pruned_loss=0.1062, over 5672058.73 frames. ], batch size: 412, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:53:33,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=150847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:53:33,961 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 03:54:00,782 INFO [optim.py:369] (1/2) 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 03:54:30,328 INFO [train.py:968] (1/2) Epoch 4, batch 14750, giga_loss[loss=0.2993, simple_loss=0.3719, pruned_loss=0.1133, over 28642.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3586, pruned_loss=0.1082, over 5677284.45 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3662, pruned_loss=0.1158, over 5760040.55 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3587, pruned_loss=0.1077, over 5667969.53 frames. ], batch size: 92, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:54:52,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4027, 1.5478, 1.4849, 1.5015], device='cuda:1'), covar=tensor([0.0870, 0.1289, 0.1279, 0.1116], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0746, 0.0618, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 03:55:17,858 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=150930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:55:32,254 INFO [train.py:968] (1/2) Epoch 4, batch 14800, giga_loss[loss=0.3061, simple_loss=0.3688, pruned_loss=0.1217, over 28910.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3582, pruned_loss=0.1091, over 5684176.73 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3661, pruned_loss=0.1158, over 5762589.08 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3583, pruned_loss=0.1086, over 5673118.02 frames. ], batch size: 213, lr: 7.93e-03, grad_scale: 8.0
+2023-03-02 03:56:07,623 INFO [optim.py:369] (1/2) 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 03:56:22,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1655, 1.4293, 1.1346, 1.6344], device='cuda:1'), covar=tensor([0.2458, 0.2099, 0.2226, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1072, 0.0843, 0.0962, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 03:56:25,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-02 03:56:30,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=150990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:56:32,272 INFO [train.py:968] (1/2) Epoch 4, batch 14850, libri_loss[loss=0.2479, simple_loss=0.3155, pruned_loss=0.09016, over 29639.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3601, pruned_loss=0.1113, over 5683792.25 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3654, pruned_loss=0.1155, over 5765981.25 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3607, pruned_loss=0.1111, over 5668990.12 frames. ], batch size: 73, lr: 7.93e-03, grad_scale: 4.0
+2023-03-02 03:56:32,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=150993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:57:10,463 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:57:36,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 03:57:36,780 INFO [train.py:968] (1/2) Epoch 4, batch 14900, giga_loss[loss=0.3023, simple_loss=0.3741, pruned_loss=0.1152, over 28614.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3607, pruned_loss=0.1113, over 5678172.91 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3651, pruned_loss=0.1153, over 5767169.47 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3613, pruned_loss=0.1112, over 5663248.07 frames. ], batch size: 307, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 03:58:16,754 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 14950, giga_loss[loss=0.2642, simple_loss=0.345, pruned_loss=0.09171, over 28456.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3625, pruned_loss=0.1111, over 5681191.74 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.365, pruned_loss=0.1152, over 5769728.98 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.363, pruned_loss=0.111, over 5665382.72 frames. ], batch size: 336, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 03:59:41,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:00:07,653 INFO [train.py:968] (1/2) Epoch 4, batch 15000, giga_loss[loss=0.2828, simple_loss=0.3507, pruned_loss=0.1075, over 28951.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3611, pruned_loss=0.1101, over 5681559.10 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3647, pruned_loss=0.1151, over 5771746.44 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3617, pruned_loss=0.11, over 5665512.79 frames. ], batch size: 213, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:00:07,654 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 04:00:12,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2878, 1.5643, 1.5005, 1.4988], device='cuda:1'), covar=tensor([0.0955, 0.1215, 0.1411, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0757, 0.0622, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 04:00:16,001 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 04:00:16,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2249, 1.4803, 1.1911, 1.4204], device='cuda:1'), covar=tensor([0.2266, 0.2037, 0.2033, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.1075, 0.0845, 0.0964, 0.0942], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:00:28,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3782, 1.3837, 1.2824, 1.7763], device='cuda:1'), covar=tensor([0.2359, 0.2164, 0.2041, 0.2102], device='cuda:1'), in_proj_covar=tensor([0.1073, 0.0844, 0.0963, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:00:56,316 INFO [optim.py:369] (1/2) 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 04:01:13,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:01:19,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:01:25,072 INFO [train.py:968] (1/2) Epoch 4, batch 15050, giga_loss[loss=0.2725, simple_loss=0.3334, pruned_loss=0.1058, over 28779.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3562, pruned_loss=0.1081, over 5694889.18 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3645, pruned_loss=0.115, over 5773088.68 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3567, pruned_loss=0.1081, over 5680238.00 frames. ], batch size: 99, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:02:01,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:02:21,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2052, 3.0024, 2.9684, 1.3004], device='cuda:1'), covar=tensor([0.0733, 0.0582, 0.0976, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0665, 0.0771, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 04:02:29,525 INFO [train.py:968] (1/2) Epoch 4, batch 15100, giga_loss[loss=0.2792, simple_loss=0.3451, pruned_loss=0.1067, over 27559.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3497, pruned_loss=0.1049, over 5689168.22 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3644, pruned_loss=0.1148, over 5774421.32 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3499, pruned_loss=0.1048, over 5674155.87 frames. ], batch size: 472, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:03:03,823 INFO [optim.py:369] (1/2) 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 04:03:26,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5087, 1.9217, 1.8090, 1.7407], device='cuda:1'), covar=tensor([0.1398, 0.1691, 0.1123, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0778, 0.0737, 0.0742, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 04:03:29,267 INFO [train.py:968] (1/2) Epoch 4, batch 15150, giga_loss[loss=0.2816, simple_loss=0.3527, pruned_loss=0.1053, over 29119.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3508, pruned_loss=0.1062, over 5686518.94 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3643, pruned_loss=0.1149, over 5776415.89 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3508, pruned_loss=0.1059, over 5671149.00 frames. ], batch size: 113, lr: 7.92e-03, grad_scale: 4.0
+2023-03-02 04:03:29,758 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:03:47,093 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:04:28,713 INFO [train.py:968] (1/2) Epoch 4, batch 15200, giga_loss[loss=0.2697, simple_loss=0.3463, pruned_loss=0.09656, over 28931.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3527, pruned_loss=0.1078, over 5681775.57 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3643, pruned_loss=0.115, over 5777649.84 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3527, pruned_loss=0.1075, over 5667800.75 frames. ], batch size: 284, lr: 7.92e-03, grad_scale: 8.0
+2023-03-02 04:05:00,533 INFO [optim.py:369] (1/2) 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 04:05:26,695 INFO [train.py:968] (1/2) Epoch 4, batch 15250, giga_loss[loss=0.2519, simple_loss=0.3292, pruned_loss=0.08736, over 28613.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3498, pruned_loss=0.1055, over 5663855.76 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3643, pruned_loss=0.1152, over 5768457.64 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3493, pruned_loss=0.1048, over 5657241.19 frames. ], batch size: 307, lr: 7.92e-03, grad_scale: 8.0
+2023-03-02 04:06:32,259 INFO [train.py:968] (1/2) Epoch 4, batch 15300, giga_loss[loss=0.2646, simple_loss=0.3362, pruned_loss=0.0965, over 28052.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3478, pruned_loss=0.1031, over 5666295.45 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3643, pruned_loss=0.1152, over 5769165.35 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3473, pruned_loss=0.1025, over 5659981.26 frames. ], batch size: 412, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:06:39,171 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:06:46,010 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:07:24,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 15350, libri_loss[loss=0.3127, simple_loss=0.3774, pruned_loss=0.124, over 28795.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3472, pruned_loss=0.1034, over 5652904.25 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3645, pruned_loss=0.1155, over 5751653.33 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3461, pruned_loss=0.1023, over 5659511.79 frames. ], batch size: 107, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:07:51,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:08:01,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6297, 2.1134, 1.9560, 1.8154], device='cuda:1'), covar=tensor([0.1363, 0.1453, 0.1023, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0738, 0.0748, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 04:08:19,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5252, 1.9354, 1.7956, 1.7326], device='cuda:1'), covar=tensor([0.1341, 0.1474, 0.1039, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0739, 0.0748, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 04:08:49,311 INFO [train.py:968] (1/2) Epoch 4, batch 15400, giga_loss[loss=0.2871, simple_loss=0.3592, pruned_loss=0.1074, over 28732.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3483, pruned_loss=0.1032, over 5672112.63 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3644, pruned_loss=0.1154, over 5754168.02 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3473, pruned_loss=0.1022, over 5673950.36 frames. ], batch size: 262, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:09:22,460 INFO [optim.py:369] (1/2) 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 04:09:28,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:09:52,398 INFO [train.py:968] (1/2) Epoch 4, batch 15450, giga_loss[loss=0.2971, simple_loss=0.363, pruned_loss=0.1156, over 29012.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3494, pruned_loss=0.1039, over 5677882.84 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1155, over 5750854.35 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3479, pruned_loss=0.1027, over 5680741.57 frames. ], batch size: 155, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:10:32,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4312, 4.2118, 4.1548, 1.7757], device='cuda:1'), covar=tensor([0.0378, 0.0358, 0.0651, 0.2004], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0650, 0.0753, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 04:10:41,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1932, 3.9884, 3.8963, 1.7377], device='cuda:1'), covar=tensor([0.0413, 0.0401, 0.0693, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.0764, 0.0652, 0.0755, 0.0574], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 04:10:59,341 INFO [train.py:968] (1/2) Epoch 4, batch 15500, giga_loss[loss=0.2784, simple_loss=0.3493, pruned_loss=0.1038, over 28385.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3507, pruned_loss=0.1053, over 5674008.74 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.365, pruned_loss=0.1157, over 5743427.22 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3492, pruned_loss=0.1041, over 5681493.75 frames. ], batch size: 368, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:11:00,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:02,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:28,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:32,563 INFO [optim.py:369] (1/2) 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 04:11:39,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:39,547 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:11:57,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=151691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:12:00,019 INFO [train.py:968] (1/2) Epoch 4, batch 15550, giga_loss[loss=0.2871, simple_loss=0.3654, pruned_loss=0.1044, over 28877.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3494, pruned_loss=0.1042, over 5653498.52 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.365, pruned_loss=0.1158, over 5725702.71 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.348, pruned_loss=0.103, over 5672719.81 frames. ], batch size: 284, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:12:56,469 INFO [train.py:968] (1/2) Epoch 4, batch 15600, giga_loss[loss=0.2992, simple_loss=0.3805, pruned_loss=0.109, over 28903.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3526, pruned_loss=0.105, over 5646655.92 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3647, pruned_loss=0.1154, over 5728558.34 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3514, pruned_loss=0.1041, over 5657571.99 frames. ], batch size: 155, lr: 7.91e-03, grad_scale: 8.0
+2023-03-02 04:13:34,666 INFO [optim.py:369] (1/2) 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 04:13:52,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3285, 1.7535, 1.2087, 0.6839], device='cuda:1'), covar=tensor([0.2123, 0.1292, 0.1815, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.1286, 0.1249, 0.1324, 0.1108], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 04:13:58,764 INFO [train.py:968] (1/2) Epoch 4, batch 15650, giga_loss[loss=0.3598, simple_loss=0.4153, pruned_loss=0.1521, over 28934.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3555, pruned_loss=0.1064, over 5645176.60 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3647, pruned_loss=0.1155, over 5721799.75 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3544, pruned_loss=0.1054, over 5658765.95 frames. ], batch size: 227, lr: 7.91e-03, grad_scale: 4.0
+2023-03-02 04:14:21,083 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=151811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:14:23,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=151814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:14:48,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6409, 2.9000, 1.6317, 1.6579], device='cuda:1'), covar=tensor([0.0655, 0.0353, 0.0672, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0458, 0.0313, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 04:14:53,617 INFO [train.py:968] (1/2) Epoch 4, batch 15700, libri_loss[loss=0.2672, simple_loss=0.3373, pruned_loss=0.09854, over 29556.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3555, pruned_loss=0.1063, over 5653358.57 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.364, pruned_loss=0.1149, over 5728576.14 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3548, pruned_loss=0.1056, over 5654545.81 frames. ], batch size: 79, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:14:55,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=151843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:15:28,875 INFO [optim.py:369] (1/2) 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 04:15:49,791 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9648, 1.1169, 3.5040, 2.9802], device='cuda:1'), covar=tensor([0.1536, 0.2138, 0.0404, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0555, 0.0515, 0.0699, 0.0560], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:15:53,857 INFO [train.py:968] (1/2) Epoch 4, batch 15750, libri_loss[loss=0.2742, simple_loss=0.351, pruned_loss=0.09868, over 29679.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3561, pruned_loss=0.1072, over 5646992.23 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3642, pruned_loss=0.115, over 5731379.14 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3553, pruned_loss=0.1065, over 5644010.17 frames. ], batch size: 88, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:16:54,809 INFO [train.py:968] (1/2) Epoch 4, batch 15800, giga_loss[loss=0.2811, simple_loss=0.3591, pruned_loss=0.1016, over 28834.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3525, pruned_loss=0.1048, over 5649899.88 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3641, pruned_loss=0.115, over 5734189.90 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3517, pruned_loss=0.1042, over 5644199.17 frames. ], batch size: 243, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:16:59,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 04:17:04,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=151952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:17:32,968 INFO [optim.py:369] (1/2) 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 04:17:52,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6552, 2.4222, 1.8378, 0.8120], device='cuda:1'), covar=tensor([0.2133, 0.1186, 0.1880, 0.2285], device='cuda:1'), in_proj_covar=tensor([0.1311, 0.1277, 0.1337, 0.1128], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 04:17:53,359 INFO [train.py:968] (1/2) Epoch 4, batch 15850, giga_loss[loss=0.3102, simple_loss=0.3762, pruned_loss=0.1221, over 28654.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3512, pruned_loss=0.1044, over 5662779.74 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3633, pruned_loss=0.1145, over 5738676.69 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3511, pruned_loss=0.104, over 5652433.49 frames. ], batch size: 307, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:18:50,822 INFO [train.py:968] (1/2) Epoch 4, batch 15900, giga_loss[loss=0.2898, simple_loss=0.3557, pruned_loss=0.112, over 28663.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3502, pruned_loss=0.1043, over 5674821.12 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3629, pruned_loss=0.1142, over 5743849.57 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.35, pruned_loss=0.104, over 5659404.84 frames. ], batch size: 307, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:19:00,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:19:22,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3327, 4.2118, 4.0675, 1.7502], device='cuda:1'), covar=tensor([0.0474, 0.0415, 0.0740, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0672, 0.0773, 0.0575], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 04:19:22,958 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:19:41,347 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:19:56,729 INFO [train.py:968] (1/2) Epoch 4, batch 15950, giga_loss[loss=0.3562, simple_loss=0.3958, pruned_loss=0.1583, over 26900.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3517, pruned_loss=0.1047, over 5679353.79 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3628, pruned_loss=0.1143, over 5746296.11 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3514, pruned_loss=0.1041, over 5663644.52 frames. ], batch size: 555, lr: 7.90e-03, grad_scale: 4.0
+2023-03-02 04:19:59,420 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:20:04,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:20:39,395 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:21:00,158 INFO [train.py:968] (1/2) Epoch 4, batch 16000, giga_loss[loss=0.2673, simple_loss=0.3435, pruned_loss=0.09549, over 28923.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3527, pruned_loss=0.1056, over 5672560.94 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3627, pruned_loss=0.1142, over 5748901.55 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3525, pruned_loss=0.1051, over 5656781.54 frames. ], batch size: 227, lr: 7.90e-03, grad_scale: 8.0
+2023-03-02 04:21:41,638 INFO [optim.py:369] (1/2) 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 04:22:00,159 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3180, 2.7154, 1.3915, 1.3107], device='cuda:1'), covar=tensor([0.0883, 0.0357, 0.0872, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0455, 0.0311, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 04:22:05,420 INFO [train.py:968] (1/2) Epoch 4, batch 16050, giga_loss[loss=0.2996, simple_loss=0.3685, pruned_loss=0.1154, over 27760.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3539, pruned_loss=0.1067, over 5673139.46 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3621, pruned_loss=0.1139, over 5751792.16 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.354, pruned_loss=0.1064, over 5656644.56 frames. ], batch size: 474, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:22:05,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:06,391 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:08,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:21,598 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:40,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:22:58,263 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 16100, giga_loss[loss=0.2586, simple_loss=0.3434, pruned_loss=0.08689, over 29126.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3566, pruned_loss=0.1081, over 5670506.05 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.361, pruned_loss=0.1132, over 5757694.05 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3576, pruned_loss=0.1082, over 5648902.76 frames. ], batch size: 113, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:23:17,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0812, 3.9006, 3.7980, 1.7464], device='cuda:1'), covar=tensor([0.0486, 0.0415, 0.0841, 0.1948], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0665, 0.0762, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 04:23:31,251 INFO [optim.py:369] (1/2) 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 04:23:37,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6981, 3.5231, 3.4090, 1.6339], device='cuda:1'), covar=tensor([0.0594, 0.0516, 0.0917, 0.2221], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0661, 0.0760, 0.0568], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 04:23:52,747 INFO [train.py:968] (1/2) Epoch 4, batch 16150, giga_loss[loss=0.3269, simple_loss=0.3917, pruned_loss=0.1311, over 28395.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3586, pruned_loss=0.1087, over 5657363.98 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.361, pruned_loss=0.1134, over 5746905.84 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3592, pruned_loss=0.1085, over 5646335.03 frames. ], batch size: 368, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:25:01,454 INFO [train.py:968] (1/2) Epoch 4, batch 16200, giga_loss[loss=0.298, simple_loss=0.3616, pruned_loss=0.1172, over 27791.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3583, pruned_loss=0.1085, over 5652095.55 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3607, pruned_loss=0.1132, over 5748850.98 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.359, pruned_loss=0.1085, over 5640403.37 frames. ], batch size: 476, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:25:08,200 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-02 04:25:10,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3954, 1.2878, 1.1784, 1.6385], device='cuda:1'), covar=tensor([0.2204, 0.2079, 0.2074, 0.2151], device='cuda:1'), in_proj_covar=tensor([0.1047, 0.0830, 0.0945, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:25:38,344 INFO [optim.py:369] (1/2) 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 04:25:58,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.79 vs. limit=5.0
+2023-03-02 04:26:05,528 INFO [train.py:968] (1/2) Epoch 4, batch 16250, giga_loss[loss=0.2812, simple_loss=0.3491, pruned_loss=0.1066, over 28119.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3559, pruned_loss=0.1072, over 5662449.99 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3607, pruned_loss=0.1132, over 5752031.37 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3563, pruned_loss=0.107, over 5647760.73 frames. ], batch size: 412, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:26:50,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6485, 1.1632, 2.8639, 2.6269], device='cuda:1'), covar=tensor([0.1531, 0.1876, 0.0511, 0.0734], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0510, 0.0699, 0.0557], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:27:08,242 INFO [train.py:968] (1/2) Epoch 4, batch 16300, libri_loss[loss=0.2659, simple_loss=0.3416, pruned_loss=0.09506, over 29527.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3543, pruned_loss=0.106, over 5664459.69 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1129, over 5746528.24 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3552, pruned_loss=0.1061, over 5654907.84 frames. ], batch size: 84, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:27:13,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7326, 1.6406, 1.2609, 1.4663], device='cuda:1'), covar=tensor([0.0657, 0.0545, 0.0962, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0451, 0.0516, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 04:27:20,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:27:40,276 INFO [optim.py:369] (1/2) 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 04:28:06,031 INFO [train.py:968] (1/2) Epoch 4, batch 16350, giga_loss[loss=0.2535, simple_loss=0.3352, pruned_loss=0.08586, over 28941.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3547, pruned_loss=0.1069, over 5669965.90 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3604, pruned_loss=0.113, over 5748773.74 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3549, pruned_loss=0.1066, over 5657620.49 frames. ], batch size: 164, lr: 7.89e-03, grad_scale: 4.0
+2023-03-02 04:28:45,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5877, 1.8148, 1.3346, 0.9516], device='cuda:1'), covar=tensor([0.0910, 0.0674, 0.0557, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.0975, 0.1003, 0.1090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 04:29:09,219 INFO [train.py:968] (1/2) Epoch 4, batch 16400, giga_loss[loss=0.2489, simple_loss=0.3068, pruned_loss=0.09551, over 24745.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3531, pruned_loss=0.1072, over 5652359.44 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1129, over 5742382.00 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3535, pruned_loss=0.107, over 5646553.77 frames. ], batch size: 705, lr: 7.89e-03, grad_scale: 8.0
+2023-03-02 04:29:41,384 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:30:08,243 INFO [train.py:968] (1/2) Epoch 4, batch 16450, giga_loss[loss=0.3285, simple_loss=0.3733, pruned_loss=0.1418, over 26957.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3513, pruned_loss=0.1057, over 5660423.72 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1126, over 5746531.10 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3516, pruned_loss=0.1055, over 5650283.48 frames. ], batch size: 555, lr: 7.88e-03, grad_scale: 8.0
+2023-03-02 04:30:17,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:30:20,375 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:30:55,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:31:04,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9055, 1.8491, 1.6831, 2.4217], device='cuda:1'), covar=tensor([0.1957, 0.1814, 0.1602, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.1059, 0.0845, 0.0957, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:31:08,209 INFO [train.py:968] (1/2) Epoch 4, batch 16500, giga_loss[loss=0.2589, simple_loss=0.3327, pruned_loss=0.09258, over 28857.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3508, pruned_loss=0.1045, over 5664338.20 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.36, pruned_loss=0.1128, over 5738205.44 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3507, pruned_loss=0.104, over 5660928.62 frames. ], batch size: 112, lr: 7.88e-03, grad_scale: 4.0
+2023-03-02 04:31:34,938 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:31:46,953 INFO [optim.py:369] (1/2) 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 04:31:52,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6634, 1.6254, 1.1890, 1.4026], device='cuda:1'), covar=tensor([0.0664, 0.0595, 0.1020, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0447, 0.0510, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 04:31:53,935 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-02 04:32:06,423 INFO [train.py:968] (1/2) Epoch 4, batch 16550, giga_loss[loss=0.2856, simple_loss=0.363, pruned_loss=0.1041, over 28216.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3516, pruned_loss=0.1032, over 5672403.91 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.113, over 5742479.93 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3513, pruned_loss=0.1025, over 5664315.47 frames. ], batch size: 412, lr: 7.88e-03, grad_scale: 4.0
+2023-03-02 04:32:07,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5094, 1.8693, 1.7027, 1.6767], device='cuda:1'), covar=tensor([0.1468, 0.1588, 0.1139, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0729, 0.0732, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 04:32:30,414 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=152712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:32:32,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=152715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:33:03,293 INFO [train.py:968] (1/2) Epoch 4, batch 16600, giga_loss[loss=0.2742, simple_loss=0.3559, pruned_loss=0.09623, over 28941.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1026, over 5687039.85 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1131, over 5747347.31 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3531, pruned_loss=0.1016, over 5674447.25 frames. ], batch size: 199, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:33:05,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=152744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:33:38,197 INFO [optim.py:369] (1/2) 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 04:33:57,411 INFO [train.py:968] (1/2) Epoch 4, batch 16650, giga_loss[loss=0.2532, simple_loss=0.3408, pruned_loss=0.08278, over 28944.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3533, pruned_loss=0.1023, over 5676576.67 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3595, pruned_loss=0.1128, over 5743068.38 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3532, pruned_loss=0.1015, over 5667795.49 frames. ], batch size: 145, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:34:53,128 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 16700, giga_loss[loss=0.2831, simple_loss=0.3464, pruned_loss=0.1099, over 26880.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3526, pruned_loss=0.1024, over 5665370.60 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5742626.31 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3527, pruned_loss=0.1017, over 5656754.42 frames. ], batch size: 555, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:35:48,028 INFO [optim.py:369] (1/2) 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 04:36:11,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3748, 3.0744, 1.2905, 1.3669], device='cuda:1'), covar=tensor([0.1076, 0.0443, 0.1048, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0455, 0.0311, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 04:36:11,920 INFO [train.py:968] (1/2) Epoch 4, batch 16750, giga_loss[loss=0.2987, simple_loss=0.3694, pruned_loss=0.114, over 29058.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3525, pruned_loss=0.1025, over 5657517.45 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5743159.18 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3526, pruned_loss=0.102, over 5649898.61 frames. ], batch size: 214, lr: 7.88e-03, grad_scale: 2.0
+2023-03-02 04:36:28,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1598, 2.6215, 2.3569, 2.3063], device='cuda:1'), covar=tensor([0.0485, 0.0594, 0.0725, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0447, 0.0511, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 04:37:21,342 INFO [train.py:968] (1/2) Epoch 4, batch 16800, giga_loss[loss=0.2747, simple_loss=0.3523, pruned_loss=0.09856, over 27605.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3515, pruned_loss=0.1007, over 5655646.22 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5736319.54 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3515, pruned_loss=0.1, over 5654360.35 frames. ], batch size: 472, lr: 7.88e-03, grad_scale: 4.0
+2023-03-02 04:38:05,002 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:38:05,338 INFO [optim.py:369] (1/2) 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 04:38:13,292 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=152982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:38:29,006 INFO [train.py:968] (1/2) Epoch 4, batch 16850, giga_loss[loss=0.2877, simple_loss=0.3772, pruned_loss=0.09905, over 28737.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3528, pruned_loss=0.1018, over 5654312.59 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5739016.73 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3526, pruned_loss=0.1009, over 5648558.02 frames. ], batch size: 242, lr: 7.87e-03, grad_scale: 4.0
+2023-03-02 04:38:29,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=152993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:39:32,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:39:34,371 INFO [train.py:968] (1/2) Epoch 4, batch 16900, giga_loss[loss=0.2767, simple_loss=0.3635, pruned_loss=0.09495, over 28509.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3581, pruned_loss=0.1048, over 5661550.07 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3597, pruned_loss=0.1132, over 5740563.46 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3572, pruned_loss=0.1032, over 5653767.06 frames. ], batch size: 336, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:40:20,682 INFO [optim.py:369] (1/2) 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 04:40:43,345 INFO [train.py:968] (1/2) Epoch 4, batch 16950, giga_loss[loss=0.3276, simple_loss=0.386, pruned_loss=0.1346, over 27646.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3581, pruned_loss=0.1048, over 5674803.17 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.36, pruned_loss=0.1134, over 5744326.97 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.357, pruned_loss=0.1032, over 5663845.44 frames. ], batch size: 472, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:40:53,786 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-02 04:40:56,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6056, 4.4198, 4.3143, 1.8325], device='cuda:1'), covar=tensor([0.0424, 0.0349, 0.0796, 0.2007], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0662, 0.0756, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 04:41:19,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:41:34,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4729, 4.3003, 4.1530, 1.8778], device='cuda:1'), covar=tensor([0.0380, 0.0368, 0.0717, 0.1905], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0660, 0.0756, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 04:41:38,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:41:41,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:41:45,028 INFO [train.py:968] (1/2) Epoch 4, batch 17000, giga_loss[loss=0.2648, simple_loss=0.3448, pruned_loss=0.09237, over 28899.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3568, pruned_loss=0.1048, over 5662166.70 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3607, pruned_loss=0.1137, over 5729962.10 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3553, pruned_loss=0.1029, over 5662747.66 frames. ], batch size: 213, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:42:24,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:42:26,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5216, 3.0916, 1.5435, 1.5415], device='cuda:1'), covar=tensor([0.0767, 0.0425, 0.0783, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0451, 0.0308, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 04:42:34,884 INFO [optim.py:369] (1/2) 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 04:42:45,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:42:49,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:42:59,242 INFO [train.py:968] (1/2) Epoch 4, batch 17050, giga_loss[loss=0.2327, simple_loss=0.327, pruned_loss=0.06922, over 28671.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3549, pruned_loss=0.1032, over 5670942.20 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.361, pruned_loss=0.1138, over 5732073.72 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3533, pruned_loss=0.1015, over 5668664.27 frames. ], batch size: 307, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:43:25,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:43:29,759 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:43:34,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3545, 1.8208, 1.3581, 1.4619], device='cuda:1'), covar=tensor([0.0770, 0.0262, 0.0322, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0135, 0.0141, 0.0253], device='cuda:1'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0055], device='cuda:1')
+2023-03-02 04:44:02,966 INFO [train.py:968] (1/2) Epoch 4, batch 17100, giga_loss[loss=0.308, simple_loss=0.377, pruned_loss=0.1195, over 28764.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3537, pruned_loss=0.1022, over 5674456.59 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.114, over 5738084.22 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3517, pruned_loss=0.1003, over 5665287.16 frames. ], batch size: 263, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:44:42,154 INFO [optim.py:369] (1/2) 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 04:45:00,670 INFO [train.py:968] (1/2) Epoch 4, batch 17150, giga_loss[loss=0.2735, simple_loss=0.3518, pruned_loss=0.09761, over 28680.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3535, pruned_loss=0.1023, over 5679261.57 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.361, pruned_loss=0.1139, over 5742552.12 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3521, pruned_loss=0.1005, over 5665545.27 frames. ], batch size: 262, lr: 7.87e-03, grad_scale: 2.0
+2023-03-02 04:45:58,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4094, 2.9374, 1.3871, 1.3395], device='cuda:1'), covar=tensor([0.0852, 0.0310, 0.0904, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0451, 0.0311, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 04:45:58,601 INFO [train.py:968] (1/2) Epoch 4, batch 17200, giga_loss[loss=0.3182, simple_loss=0.3898, pruned_loss=0.1233, over 28139.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3566, pruned_loss=0.1042, over 5675655.27 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1135, over 5744905.37 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3559, pruned_loss=0.1029, over 5662088.42 frames. ], batch size: 412, lr: 7.87e-03, grad_scale: 4.0
+2023-03-02 04:46:09,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:15,007 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:15,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:17,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:29,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 04:46:36,882 INFO [optim.py:369] (1/2) 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 04:46:47,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:46:51,597 INFO [train.py:968] (1/2) Epoch 4, batch 17250, giga_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08742, over 29129.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3577, pruned_loss=0.1057, over 5670024.82 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5733816.48 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3573, pruned_loss=0.1046, over 5666141.60 frames. ], batch size: 128, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:47:16,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1958, 1.2048, 4.9901, 3.6146], device='cuda:1'), covar=tensor([0.1559, 0.2090, 0.0301, 0.0478], device='cuda:1'), in_proj_covar=tensor([0.0543, 0.0507, 0.0694, 0.0559], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:47:49,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1099, 1.3920, 1.1346, 0.3106], device='cuda:1'), covar=tensor([0.0975, 0.0950, 0.1548, 0.1708], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.1260, 0.1325, 0.1101], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 04:47:51,934 INFO [train.py:968] (1/2) Epoch 4, batch 17300, giga_loss[loss=0.3012, simple_loss=0.3682, pruned_loss=0.1171, over 28491.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.354, pruned_loss=0.1049, over 5656996.40 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1134, over 5725323.90 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3536, pruned_loss=0.1039, over 5660419.19 frames. ], batch size: 336, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:48:32,516 INFO [optim.py:369] (1/2) 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 04:48:43,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5202, 2.0886, 1.8331, 1.7273], device='cuda:1'), covar=tensor([0.1574, 0.1584, 0.1149, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0727, 0.0735, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 04:48:49,894 INFO [train.py:968] (1/2) Epoch 4, batch 17350, libri_loss[loss=0.273, simple_loss=0.346, pruned_loss=0.09999, over 29517.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3539, pruned_loss=0.1058, over 5654376.15 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3598, pruned_loss=0.113, over 5730398.38 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3539, pruned_loss=0.1051, over 5650218.97 frames. ], batch size: 81, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:48:50,235 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:48:50,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:48:52,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:48:58,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:00,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:10,051 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:28,088 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:34,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:49:45,584 INFO [train.py:968] (1/2) Epoch 4, batch 17400, giga_loss[loss=0.3603, simple_loss=0.4199, pruned_loss=0.1503, over 28552.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3595, pruned_loss=0.1099, over 5652899.52 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1127, over 5730858.37 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3598, pruned_loss=0.1096, over 5648668.75 frames. ], batch size: 307, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:50:22,458 INFO [optim.py:369] (1/2) 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 04:50:38,316 INFO [train.py:968] (1/2) Epoch 4, batch 17450, giga_loss[loss=0.3243, simple_loss=0.3923, pruned_loss=0.1281, over 28625.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3723, pruned_loss=0.1184, over 5662039.33 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3596, pruned_loss=0.1129, over 5731445.08 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3724, pruned_loss=0.118, over 5657803.27 frames. ], batch size: 92, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:51:14,084 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=153637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:16,192 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=153640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:19,365 INFO [train.py:968] (1/2) Epoch 4, batch 17500, giga_loss[loss=0.3028, simple_loss=0.3703, pruned_loss=0.1177, over 28262.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3779, pruned_loss=0.1221, over 5673686.17 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5734175.45 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3779, pruned_loss=0.1219, over 5665941.74 frames. ], batch size: 77, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:51:25,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8976, 1.0255, 3.9225, 2.9414], device='cuda:1'), covar=tensor([0.1588, 0.2257, 0.0347, 0.0635], device='cuda:1'), in_proj_covar=tensor([0.0539, 0.0501, 0.0688, 0.0553], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 04:51:38,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3970, 2.9558, 1.5005, 1.3592], device='cuda:1'), covar=tensor([0.0871, 0.0326, 0.0855, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0447, 0.0306, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 04:51:41,756 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=153669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:42,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:51:49,990 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 17550, giga_loss[loss=0.2698, simple_loss=0.3384, pruned_loss=0.1006, over 29018.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3743, pruned_loss=0.1212, over 5671572.98 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.1131, over 5737012.67 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3745, pruned_loss=0.1211, over 5662295.36 frames. ], batch size: 164, lr: 7.86e-03, grad_scale: 4.0
+2023-03-02 04:52:34,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0028, 1.3006, 1.0895, 0.3107], device='cuda:1'), covar=tensor([0.1366, 0.1254, 0.2062, 0.2204], device='cuda:1'), in_proj_covar=tensor([0.1284, 0.1240, 0.1299, 0.1085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 04:52:49,337 INFO [train.py:968] (1/2) Epoch 4, batch 17600, giga_loss[loss=0.2533, simple_loss=0.3126, pruned_loss=0.09696, over 28713.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3678, pruned_loss=0.1184, over 5674759.40 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5730726.97 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3677, pruned_loss=0.1181, over 5672502.71 frames. ], batch size: 92, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:53:18,791 INFO [optim.py:369] (1/2) 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 04:53:32,597 INFO [train.py:968] (1/2) Epoch 4, batch 17650, giga_loss[loss=0.2472, simple_loss=0.3113, pruned_loss=0.09152, over 28918.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3604, pruned_loss=0.1151, over 5681808.16 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3611, pruned_loss=0.1135, over 5729983.12 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3601, pruned_loss=0.1148, over 5679540.86 frames. ], batch size: 213, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:53:33,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-02 04:53:52,846 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=153814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:54:19,122 INFO [train.py:968] (1/2) Epoch 4, batch 17700, libri_loss[loss=0.2443, simple_loss=0.3245, pruned_loss=0.08201, over 29667.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3527, pruned_loss=0.1113, over 5676695.51 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3615, pruned_loss=0.1135, over 5722984.39 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.352, pruned_loss=0.1112, over 5679411.55 frames. ], batch size: 73, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:54:45,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4633, 3.0460, 1.5681, 1.4077], device='cuda:1'), covar=tensor([0.0848, 0.0302, 0.0820, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0451, 0.0305, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 04:54:46,205 INFO [optim.py:369] (1/2) 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 04:54:53,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=153885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:55:01,708 INFO [train.py:968] (1/2) Epoch 4, batch 17750, giga_loss[loss=0.2343, simple_loss=0.3046, pruned_loss=0.08197, over 29041.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3441, pruned_loss=0.1069, over 5679066.69 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3618, pruned_loss=0.1136, over 5724372.86 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.343, pruned_loss=0.1066, over 5679398.03 frames. ], batch size: 106, lr: 7.85e-03, grad_scale: 8.0
+2023-03-02 04:55:43,014 INFO [train.py:968] (1/2) Epoch 4, batch 17800, giga_loss[loss=0.2714, simple_loss=0.3281, pruned_loss=0.1073, over 28949.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3394, pruned_loss=0.1046, over 5677135.51 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3622, pruned_loss=0.1139, over 5717068.01 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3377, pruned_loss=0.104, over 5683443.62 frames. ], batch size: 213, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:56:11,944 INFO [optim.py:369] (1/2) 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 04:56:25,866 INFO [train.py:968] (1/2) Epoch 4, batch 17850, libri_loss[loss=0.2551, simple_loss=0.3293, pruned_loss=0.09044, over 29497.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3368, pruned_loss=0.1032, over 5689706.96 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3628, pruned_loss=0.1141, over 5721483.03 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3345, pruned_loss=0.1022, over 5689962.43 frames. ], batch size: 70, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:56:56,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:56:58,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:57:07,697 INFO [train.py:968] (1/2) Epoch 4, batch 17900, libri_loss[loss=0.3413, simple_loss=0.4066, pruned_loss=0.138, over 29380.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.334, pruned_loss=0.1016, over 5688092.25 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3637, pruned_loss=0.1146, over 5724921.49 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3305, pruned_loss=0.1001, over 5684119.36 frames. ], batch size: 92, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:57:10,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:57:21,073 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:57:47,426 INFO [train.py:968] (1/2) Epoch 4, batch 17950, giga_loss[loss=0.2326, simple_loss=0.2978, pruned_loss=0.08364, over 28835.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3313, pruned_loss=0.09991, over 5694202.52 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3641, pruned_loss=0.1147, over 5729070.83 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.327, pruned_loss=0.09815, over 5686185.37 frames. ], batch size: 112, lr: 7.85e-03, grad_scale: 4.0
+2023-03-02 04:58:04,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 04:58:30,656 INFO [train.py:968] (1/2) Epoch 4, batch 18000, giga_loss[loss=0.241, simple_loss=0.3073, pruned_loss=0.08737, over 28718.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3286, pruned_loss=0.09819, over 5700622.41 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3645, pruned_loss=0.1146, over 5734180.05 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3239, pruned_loss=0.09638, over 5688967.05 frames. ], batch size: 92, lr: 7.84e-03, grad_scale: 8.0
+2023-03-02 04:58:30,656 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 04:58:38,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3068, 2.0052, 1.7022, 1.6148], device='cuda:1'), covar=tensor([0.1649, 0.1931, 0.1307, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0754, 0.0755, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 04:58:39,397 INFO [train.py:1012] (1/2) Epoch 4, validation: loss=0.2448, simple_loss=0.3452, pruned_loss=0.0722, over 944034.00 frames. 
+2023-03-02 04:58:39,397 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 04:59:08,330 INFO [optim.py:369] (1/2) 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 04:59:18,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:59:18,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:59:20,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 04:59:20,984 INFO [train.py:968] (1/2) Epoch 4, batch 18050, giga_loss[loss=0.2689, simple_loss=0.3201, pruned_loss=0.1088, over 26520.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.325, pruned_loss=0.09672, over 5684474.09 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3651, pruned_loss=0.115, over 5724816.60 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3203, pruned_loss=0.09474, over 5682265.17 frames. ], batch size: 555, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 04:59:46,996 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 18100, giga_loss[loss=0.2249, simple_loss=0.2936, pruned_loss=0.07807, over 28619.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3214, pruned_loss=0.09438, over 5694454.68 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3655, pruned_loss=0.1152, over 5729599.87 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3162, pruned_loss=0.09217, over 5687485.51 frames. ], batch size: 85, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:00:37,314 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 18150, giga_loss[loss=0.2091, simple_loss=0.2819, pruned_loss=0.0681, over 28545.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3179, pruned_loss=0.09269, over 5703135.88 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3653, pruned_loss=0.115, over 5733006.89 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3132, pruned_loss=0.09073, over 5694047.24 frames. ], batch size: 71, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:01:15,206 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:01:30,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:01:32,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:01:37,924 INFO [train.py:968] (1/2) Epoch 4, batch 18200, giga_loss[loss=0.296, simple_loss=0.3583, pruned_loss=0.1168, over 28534.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3175, pruned_loss=0.09333, over 5701324.38 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3658, pruned_loss=0.1153, over 5733792.09 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3125, pruned_loss=0.09117, over 5692949.54 frames. ], batch size: 336, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:02:00,732 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:02:13,637 INFO [optim.py:369] (1/2) 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 05:02:27,735 INFO [train.py:968] (1/2) Epoch 4, batch 18250, giga_loss[loss=0.2961, simple_loss=0.3667, pruned_loss=0.1127, over 28727.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.329, pruned_loss=0.1, over 5700369.35 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3657, pruned_loss=0.1152, over 5738149.69 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3237, pruned_loss=0.09773, over 5688539.71 frames. ], batch size: 99, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:03:14,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154442.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:03:15,223 INFO [train.py:968] (1/2) Epoch 4, batch 18300, giga_loss[loss=0.3642, simple_loss=0.4168, pruned_loss=0.1558, over 28911.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3447, pruned_loss=0.1093, over 5689564.93 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3659, pruned_loss=0.1153, over 5730959.84 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.34, pruned_loss=0.1072, over 5686162.72 frames. ], batch size: 186, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:03:35,983 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:03:43,069 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:03:44,319 INFO [optim.py:369] (1/2) 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 05:03:54,784 INFO [train.py:968] (1/2) Epoch 4, batch 18350, giga_loss[loss=0.3007, simple_loss=0.365, pruned_loss=0.1182, over 28800.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3566, pruned_loss=0.1154, over 5701710.06 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3662, pruned_loss=0.1155, over 5733270.12 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3524, pruned_loss=0.1136, over 5696421.70 frames. ], batch size: 99, lr: 7.84e-03, grad_scale: 4.0
+2023-03-02 05:04:23,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4142, 3.1991, 3.1071, 1.8481], device='cuda:1'), covar=tensor([0.0530, 0.0511, 0.0873, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0670, 0.0771, 0.0576], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 05:04:40,348 INFO [train.py:968] (1/2) Epoch 4, batch 18400, giga_loss[loss=0.2688, simple_loss=0.3487, pruned_loss=0.0945, over 28938.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3637, pruned_loss=0.1184, over 5686869.50 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3668, pruned_loss=0.1158, over 5726003.46 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3597, pruned_loss=0.1168, over 5688966.33 frames. ], batch size: 145, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:05:08,210 INFO [optim.py:369] (1/2) 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 05:05:19,923 INFO [train.py:968] (1/2) Epoch 4, batch 18450, giga_loss[loss=0.351, simple_loss=0.4217, pruned_loss=0.1401, over 28627.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.367, pruned_loss=0.1188, over 5689511.12 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.367, pruned_loss=0.1158, over 5728422.72 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3636, pruned_loss=0.1175, over 5687991.57 frames. ], batch size: 307, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:05:41,586 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:06:07,950 INFO [train.py:968] (1/2) Epoch 4, batch 18500, giga_loss[loss=0.3103, simple_loss=0.3666, pruned_loss=0.127, over 28804.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3677, pruned_loss=0.1182, over 5686807.99 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3673, pruned_loss=0.116, over 5731127.51 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3646, pruned_loss=0.1171, over 5682497.51 frames. ], batch size: 99, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:06:09,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-02 05:06:12,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-02 05:06:39,112 INFO [optim.py:369] (1/2) 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 05:06:39,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:06:42,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:06:47,622 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 18550, giga_loss[loss=0.3252, simple_loss=0.3798, pruned_loss=0.1353, over 28584.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3698, pruned_loss=0.1198, over 5697529.65 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3671, pruned_loss=0.1158, over 5737384.56 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3676, pruned_loss=0.1191, over 5686669.00 frames. ], batch size: 85, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:07:12,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1992, 1.2521, 1.0837, 1.3559], device='cuda:1'), covar=tensor([0.2371, 0.2153, 0.2054, 0.2043], device='cuda:1'), in_proj_covar=tensor([0.1061, 0.0844, 0.0946, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 05:07:23,562 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154730.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:07:33,558 INFO [train.py:968] (1/2) Epoch 4, batch 18600, giga_loss[loss=0.3022, simple_loss=0.3723, pruned_loss=0.116, over 28927.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3744, pruned_loss=0.1235, over 5693527.32 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3678, pruned_loss=0.1163, over 5732067.03 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3721, pruned_loss=0.1227, over 5687932.38 frames. ], batch size: 227, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:08:04,755 INFO [optim.py:369] (1/2) 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 05:08:18,249 INFO [train.py:968] (1/2) Epoch 4, batch 18650, giga_loss[loss=0.3031, simple_loss=0.3717, pruned_loss=0.1173, over 29077.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.377, pruned_loss=0.1253, over 5700727.57 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3681, pruned_loss=0.1162, over 5736428.18 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3752, pruned_loss=0.1249, over 5691601.09 frames. ], batch size: 136, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:08:38,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154817.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:08:54,718 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:08:57,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:08:58,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0506, 1.3628, 1.0870, 0.2246], device='cuda:1'), covar=tensor([0.1437, 0.1254, 0.2174, 0.2271], device='cuda:1'), in_proj_covar=tensor([0.1333, 0.1254, 0.1335, 0.1109], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 05:09:00,829 INFO [train.py:968] (1/2) Epoch 4, batch 18700, giga_loss[loss=0.3189, simple_loss=0.3827, pruned_loss=0.1275, over 28937.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3801, pruned_loss=0.1259, over 5704923.85 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3682, pruned_loss=0.1163, over 5737290.04 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3785, pruned_loss=0.1256, over 5696886.50 frames. ], batch size: 186, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:09:01,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:08,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:21,824 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:22,331 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=154867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:09:24,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9753, 1.2097, 4.3460, 3.4114], device='cuda:1'), covar=tensor([0.1728, 0.2141, 0.0268, 0.0658], device='cuda:1'), in_proj_covar=tensor([0.0540, 0.0501, 0.0686, 0.0562], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 05:09:32,153 INFO [optim.py:369] (1/2) 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 05:09:35,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0561, 1.6621, 1.3775, 1.4781], device='cuda:1'), covar=tensor([0.0631, 0.0735, 0.0969, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0463, 0.0524, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 05:09:44,927 INFO [train.py:968] (1/2) Epoch 4, batch 18750, libri_loss[loss=0.3218, simple_loss=0.391, pruned_loss=0.1263, over 29396.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3817, pruned_loss=0.126, over 5711898.55 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3687, pruned_loss=0.1166, over 5741258.65 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3803, pruned_loss=0.1257, over 5701186.76 frames. ], batch size: 92, lr: 7.83e-03, grad_scale: 8.0
+2023-03-02 05:09:56,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 05:10:24,822 INFO [train.py:968] (1/2) Epoch 4, batch 18800, giga_loss[loss=0.3463, simple_loss=0.3988, pruned_loss=0.1469, over 28664.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3821, pruned_loss=0.1255, over 5708094.25 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3687, pruned_loss=0.1166, over 5744437.70 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3813, pruned_loss=0.1255, over 5696159.90 frames. ], batch size: 85, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:10:38,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154960.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:10:40,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154963.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:10:52,235 INFO [optim.py:369] (1/2) 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 05:10:57,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:10:57,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 05:10:58,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=154989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:10:59,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:00,967 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=154992.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:11:01,420 INFO [train.py:968] (1/2) Epoch 4, batch 18850, giga_loss[loss=0.2877, simple_loss=0.3654, pruned_loss=0.105, over 28787.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3808, pruned_loss=0.1236, over 5711638.40 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3688, pruned_loss=0.1165, over 5750011.19 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3806, pruned_loss=0.1239, over 5695527.27 frames. ], batch size: 262, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:11:03,445 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=154995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:05,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=154998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:24,536 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:25,875 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:30,011 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:43,140 INFO [train.py:968] (1/2) Epoch 4, batch 18900, giga_loss[loss=0.3001, simple_loss=0.3721, pruned_loss=0.1141, over 28429.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3788, pruned_loss=0.121, over 5712976.43 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3691, pruned_loss=0.1166, over 5751603.89 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3786, pruned_loss=0.1214, over 5697777.58 frames. ], batch size: 71, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:11:47,828 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 05:11:51,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:11:55,090 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:12:22,403 INFO [train.py:968] (1/2) Epoch 4, batch 18950, giga_loss[loss=0.272, simple_loss=0.3488, pruned_loss=0.09755, over 28875.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3786, pruned_loss=0.1208, over 5714544.09 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3695, pruned_loss=0.1168, over 5753312.86 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3782, pruned_loss=0.1209, over 5700674.86 frames. ], batch size: 119, lr: 7.82e-03, grad_scale: 8.0
+2023-03-02 05:12:31,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-02 05:12:31,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155105.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:12:38,474 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:12:55,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:12:57,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:02,520 INFO [train.py:968] (1/2) Epoch 4, batch 19000, giga_loss[loss=0.3225, simple_loss=0.3764, pruned_loss=0.1343, over 28946.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3811, pruned_loss=0.1242, over 5705384.52 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3705, pruned_loss=0.1171, over 5755395.72 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3804, pruned_loss=0.1243, over 5690151.92 frames. ], batch size: 227, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:13:21,053 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:13:44,255 INFO [train.py:968] (1/2) Epoch 4, batch 19050, giga_loss[loss=0.3503, simple_loss=0.4167, pruned_loss=0.142, over 28524.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3841, pruned_loss=0.1289, over 5676055.66 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3705, pruned_loss=0.1172, over 5738084.30 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3839, pruned_loss=0.1293, over 5677794.17 frames. ], batch size: 60, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:13:47,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:50,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:51,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:13:54,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:14,047 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:17,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:23,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:14:23,987 INFO [train.py:968] (1/2) Epoch 4, batch 19100, giga_loss[loss=0.3116, simple_loss=0.3761, pruned_loss=0.1236, over 28748.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3855, pruned_loss=0.1313, over 5687765.44 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3708, pruned_loss=0.1173, over 5741611.42 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3855, pruned_loss=0.1319, over 5684587.85 frames. ], batch size: 262, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:14:27,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155248.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:14:29,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155251.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:14:55,769 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155280.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:14:56,229 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 19150, giga_loss[loss=0.3296, simple_loss=0.3696, pruned_loss=0.1448, over 23788.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3842, pruned_loss=0.1316, over 5693053.22 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3712, pruned_loss=0.1176, over 5744309.03 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3841, pruned_loss=0.132, over 5687187.92 frames. ], batch size: 705, lr: 7.82e-03, grad_scale: 4.0
+2023-03-02 05:15:49,366 INFO [train.py:968] (1/2) Epoch 4, batch 19200, giga_loss[loss=0.3069, simple_loss=0.3668, pruned_loss=0.1235, over 28917.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3824, pruned_loss=0.1303, over 5696493.13 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3716, pruned_loss=0.1178, over 5744449.47 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3823, pruned_loss=0.1308, over 5690403.15 frames. ], batch size: 213, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:16:09,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-02 05:16:12,852 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:16:21,642 INFO [optim.py:369] (1/2) 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 05:16:26,468 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:16:30,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 19250, giga_loss[loss=0.3293, simple_loss=0.3872, pruned_loss=0.1358, over 28558.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.381, pruned_loss=0.1289, over 5690498.98 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3709, pruned_loss=0.1174, over 5748145.18 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3817, pruned_loss=0.1299, over 5681310.68 frames. ], batch size: 78, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:16:35,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:16:52,343 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 19300, giga_loss[loss=0.278, simple_loss=0.3508, pruned_loss=0.1027, over 28887.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3777, pruned_loss=0.1256, over 5696352.76 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3712, pruned_loss=0.1175, over 5749337.04 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3782, pruned_loss=0.1265, over 5687013.61 frames. ], batch size: 145, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:17:34,506 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:17:50,797 INFO [optim.py:369] (1/2) 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 05:17:57,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:18:02,324 INFO [train.py:968] (1/2) Epoch 4, batch 19350, giga_loss[loss=0.3252, simple_loss=0.3708, pruned_loss=0.1398, over 26499.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3728, pruned_loss=0.1222, over 5687702.81 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3714, pruned_loss=0.1175, over 5751641.36 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3731, pruned_loss=0.123, over 5677403.54 frames. ], batch size: 555, lr: 7.81e-03, grad_scale: 8.0
+2023-03-02 05:18:45,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:18:47,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:18:48,053 INFO [train.py:968] (1/2) Epoch 4, batch 19400, giga_loss[loss=0.2616, simple_loss=0.3296, pruned_loss=0.09677, over 28618.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3663, pruned_loss=0.1182, over 5690262.20 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3715, pruned_loss=0.1175, over 5755315.48 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3664, pruned_loss=0.1188, over 5677231.64 frames. ], batch size: 307, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:19:12,171 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:19:22,644 INFO [optim.py:369] (1/2) 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 05:19:28,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7502, 1.6874, 1.2508, 1.1476], device='cuda:1'), covar=tensor([0.0781, 0.0727, 0.0643, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.1012, 0.1031, 0.1125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 05:19:33,535 INFO [train.py:968] (1/2) Epoch 4, batch 19450, giga_loss[loss=0.2917, simple_loss=0.3496, pruned_loss=0.1169, over 28529.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3602, pruned_loss=0.1148, over 5689423.00 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3716, pruned_loss=0.1177, over 5758064.93 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.36, pruned_loss=0.1152, over 5675571.38 frames. ], batch size: 78, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:20:11,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:20:14,347 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 19500, giga_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1271, over 28762.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3583, pruned_loss=0.1132, over 5693959.20 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3716, pruned_loss=0.1176, over 5759550.56 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.358, pruned_loss=0.1136, over 5681240.23 frames. ], batch size: 99, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:20:27,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9723, 1.7453, 1.2257, 1.4837], device='cuda:1'), covar=tensor([0.0538, 0.0575, 0.0967, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0458, 0.0518, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 05:20:33,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4325, 1.9669, 1.7059, 1.5602], device='cuda:1'), covar=tensor([0.1629, 0.1873, 0.1262, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0749, 0.0751, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 05:20:39,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:20:51,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4069, 1.5695, 0.9204, 1.2842], device='cuda:1'), covar=tensor([0.0761, 0.0648, 0.1511, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0460, 0.0521, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 05:20:54,799 INFO [optim.py:369] (1/2) 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 05:21:05,884 INFO [train.py:968] (1/2) Epoch 4, batch 19550, giga_loss[loss=0.2613, simple_loss=0.3395, pruned_loss=0.09151, over 29066.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.36, pruned_loss=0.1139, over 5705616.64 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3726, pruned_loss=0.1181, over 5763708.22 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3586, pruned_loss=0.1136, over 5690088.04 frames. ], batch size: 136, lr: 7.81e-03, grad_scale: 4.0
+2023-03-02 05:21:47,282 INFO [train.py:968] (1/2) Epoch 4, batch 19600, giga_loss[loss=0.2978, simple_loss=0.3556, pruned_loss=0.12, over 28921.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3593, pruned_loss=0.1133, over 5702474.02 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3734, pruned_loss=0.1183, over 5757701.45 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.357, pruned_loss=0.1127, over 5693350.82 frames. ], batch size: 106, lr: 7.80e-03, grad_scale: 8.0
+2023-03-02 05:21:49,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:22:21,604 INFO [optim.py:369] (1/2) 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 05:22:29,544 INFO [train.py:968] (1/2) Epoch 4, batch 19650, giga_loss[loss=0.2688, simple_loss=0.3346, pruned_loss=0.1015, over 28824.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.357, pruned_loss=0.1123, over 5712619.71 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3738, pruned_loss=0.1184, over 5759645.56 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3547, pruned_loss=0.1118, over 5703001.74 frames. ], batch size: 145, lr: 7.80e-03, grad_scale: 8.0
+2023-03-02 05:23:03,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=155837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:08,544 INFO [train.py:968] (1/2) Epoch 4, batch 19700, giga_loss[loss=0.239, simple_loss=0.3123, pruned_loss=0.08289, over 28958.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3528, pruned_loss=0.1097, over 5722221.67 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3739, pruned_loss=0.1183, over 5762657.41 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3504, pruned_loss=0.1092, over 5710818.16 frames. ], batch size: 213, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:23:15,470 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:41,484 INFO [optim.py:369] (1/2) 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 05:23:45,805 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:47,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:23:49,810 INFO [train.py:968] (1/2) Epoch 4, batch 19750, giga_loss[loss=0.3248, simple_loss=0.3811, pruned_loss=0.1343, over 27975.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3519, pruned_loss=0.1097, over 5722292.39 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.375, pruned_loss=0.1188, over 5763262.04 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3482, pruned_loss=0.1085, over 5711408.97 frames. ], batch size: 412, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:23:56,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3174, 1.3714, 1.0564, 0.7523], device='cuda:1'), covar=tensor([0.0774, 0.0679, 0.0586, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.0986, 0.1021, 0.1112], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 05:24:00,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5425, 2.2413, 1.5926, 0.6179], device='cuda:1'), covar=tensor([0.2297, 0.1196, 0.1988, 0.2713], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1220, 0.1313, 0.1108], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 05:24:11,300 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=155919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:24:31,567 INFO [train.py:968] (1/2) Epoch 4, batch 19800, libri_loss[loss=0.3075, simple_loss=0.3796, pruned_loss=0.1177, over 29567.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3503, pruned_loss=0.1088, over 5728102.83 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3758, pruned_loss=0.119, over 5767374.37 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.346, pruned_loss=0.1073, over 5714440.18 frames. ], batch size: 78, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:24:59,185 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=155980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:00,971 INFO [optim.py:369] (1/2) 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 05:25:01,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=155983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:08,024 INFO [train.py:968] (1/2) Epoch 4, batch 19850, giga_loss[loss=0.2881, simple_loss=0.3511, pruned_loss=0.1126, over 28639.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3479, pruned_loss=0.1075, over 5731228.63 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3758, pruned_loss=0.1189, over 5770773.38 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3437, pruned_loss=0.1062, over 5716392.24 frames. ], batch size: 307, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:25:13,509 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=155998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:25,447 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:25:43,664 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-02 05:25:50,392 INFO [train.py:968] (1/2) Epoch 4, batch 19900, giga_loss[loss=0.3011, simple_loss=0.3598, pruned_loss=0.1211, over 28605.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3456, pruned_loss=0.1069, over 5725775.21 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3761, pruned_loss=0.119, over 5772160.31 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3416, pruned_loss=0.1056, over 5712236.17 frames. ], batch size: 336, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:25:56,553 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:26:21,852 INFO [optim.py:369] (1/2) 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 05:26:29,447 INFO [train.py:968] (1/2) Epoch 4, batch 19950, giga_loss[loss=0.2398, simple_loss=0.3059, pruned_loss=0.08687, over 28792.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3444, pruned_loss=0.106, over 5729022.40 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.377, pruned_loss=0.1193, over 5771528.05 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3394, pruned_loss=0.1043, over 5717017.50 frames. ], batch size: 92, lr: 7.80e-03, grad_scale: 4.0
+2023-03-02 05:26:43,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4588, 1.5670, 1.5127, 1.3872], device='cuda:1'), covar=tensor([0.0874, 0.0955, 0.1307, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0733, 0.0618, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 05:27:07,371 INFO [train.py:968] (1/2) Epoch 4, batch 20000, giga_loss[loss=0.2649, simple_loss=0.3274, pruned_loss=0.1012, over 29016.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3438, pruned_loss=0.1052, over 5728057.83 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3783, pruned_loss=0.1199, over 5765246.66 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3374, pruned_loss=0.1027, over 5722568.70 frames. ], batch size: 136, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:27:10,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3940, 1.4169, 1.1956, 1.6776], device='cuda:1'), covar=tensor([0.2091, 0.2064, 0.2026, 0.2302], device='cuda:1'), in_proj_covar=tensor([0.1073, 0.0846, 0.0960, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 05:27:34,952 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:27:35,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-02 05:27:39,392 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 20050, libri_loss[loss=0.381, simple_loss=0.4509, pruned_loss=0.1556, over 29219.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3428, pruned_loss=0.1048, over 5731331.07 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3787, pruned_loss=0.12, over 5766627.86 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3364, pruned_loss=0.1023, over 5724749.16 frames. ], batch size: 97, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:27:58,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5763, 1.5451, 1.2305, 1.4332], device='cuda:1'), covar=tensor([0.0575, 0.0439, 0.0908, 0.0651], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0460, 0.0519, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 05:28:05,168 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.93 vs. limit=5.0
+2023-03-02 05:28:13,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:28:28,644 INFO [train.py:968] (1/2) Epoch 4, batch 20100, giga_loss[loss=0.2868, simple_loss=0.3562, pruned_loss=0.1088, over 29012.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3456, pruned_loss=0.1074, over 5725895.17 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3793, pruned_loss=0.1204, over 5759025.91 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3394, pruned_loss=0.1049, over 5726254.16 frames. ], batch size: 136, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:29:05,172 INFO [optim.py:369] (1/2) 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 05:29:13,257 INFO [train.py:968] (1/2) Epoch 4, batch 20150, giga_loss[loss=0.3846, simple_loss=0.4213, pruned_loss=0.174, over 26619.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3523, pruned_loss=0.1121, over 5714736.59 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3794, pruned_loss=0.1204, over 5760584.97 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3465, pruned_loss=0.1098, over 5712854.89 frames. ], batch size: 555, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:29:48,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4808, 1.5382, 0.9300, 1.2572], device='cuda:1'), covar=tensor([0.0802, 0.0691, 0.1579, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0462, 0.0522, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 05:30:07,337 INFO [train.py:968] (1/2) Epoch 4, batch 20200, giga_loss[loss=0.3965, simple_loss=0.4384, pruned_loss=0.1773, over 29008.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3623, pruned_loss=0.1194, over 5702163.76 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3795, pruned_loss=0.1205, over 5763069.32 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3573, pruned_loss=0.1175, over 5697827.91 frames. ], batch size: 136, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:30:11,798 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 05:30:32,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:30:36,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:30:39,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:30:48,393 INFO [optim.py:369] (1/2) 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 05:30:56,688 INFO [train.py:968] (1/2) Epoch 4, batch 20250, giga_loss[loss=0.2755, simple_loss=0.348, pruned_loss=0.1016, over 28424.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3694, pruned_loss=0.1239, over 5689302.63 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3796, pruned_loss=0.1205, over 5754741.00 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3653, pruned_loss=0.1224, over 5692474.63 frames. ], batch size: 65, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:31:04,887 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:31:29,477 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 20300, giga_loss[loss=0.3297, simple_loss=0.3914, pruned_loss=0.134, over 28874.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3752, pruned_loss=0.1264, over 5670892.58 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3802, pruned_loss=0.1209, over 5748568.27 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3711, pruned_loss=0.1249, over 5676698.09 frames. ], batch size: 119, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:32:04,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:32:27,126 INFO [optim.py:369] (1/2) 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 05:32:34,409 INFO [train.py:968] (1/2) Epoch 4, batch 20350, giga_loss[loss=0.3265, simple_loss=0.3843, pruned_loss=0.1343, over 28693.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3789, pruned_loss=0.1279, over 5672499.70 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3803, pruned_loss=0.1209, over 5747681.58 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3756, pruned_loss=0.1268, over 5676683.40 frames. ], batch size: 78, lr: 7.79e-03, grad_scale: 4.0
+2023-03-02 05:32:37,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6538, 1.7125, 1.3705, 1.0015], device='cuda:1'), covar=tensor([0.0788, 0.0644, 0.0573, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.1318, 0.1024, 0.1061, 0.1146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 05:32:56,601 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:32:58,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:10,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7296, 2.4378, 1.6926, 0.8613], device='cuda:1'), covar=tensor([0.2216, 0.1193, 0.2055, 0.2284], device='cuda:1'), in_proj_covar=tensor([0.1307, 0.1232, 0.1335, 0.1106], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 05:33:16,175 INFO [train.py:968] (1/2) Epoch 4, batch 20400, libri_loss[loss=0.3352, simple_loss=0.3981, pruned_loss=0.1362, over 20049.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3843, pruned_loss=0.1316, over 5663651.00 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3802, pruned_loss=0.1209, over 5742844.23 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3817, pruned_loss=0.131, over 5669579.59 frames. ], batch size: 186, lr: 7.78e-03, grad_scale: 8.0
+2023-03-02 05:33:20,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:24,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:36,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:38,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:33:52,814 INFO [optim.py:369] (1/2) 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 05:34:00,206 INFO [train.py:968] (1/2) Epoch 4, batch 20450, giga_loss[loss=0.2918, simple_loss=0.3588, pruned_loss=0.1124, over 28861.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3797, pruned_loss=0.1281, over 5658395.70 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3808, pruned_loss=0.1213, over 5726569.35 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3772, pruned_loss=0.1274, over 5675499.13 frames. ], batch size: 227, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:34:06,512 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:34:14,024 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=156609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:34:30,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 05:34:42,375 INFO [train.py:968] (1/2) Epoch 4, batch 20500, giga_loss[loss=0.3053, simple_loss=0.3731, pruned_loss=0.1188, over 28883.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3747, pruned_loss=0.1238, over 5669112.29 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3808, pruned_loss=0.1214, over 5728449.27 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3727, pruned_loss=0.1234, over 5680212.38 frames. ], batch size: 174, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:34:50,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9381, 1.6982, 1.2884, 1.4530], device='cuda:1'), covar=tensor([0.0574, 0.0620, 0.0919, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0461, 0.0522, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 05:35:08,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 05:35:17,433 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:35:27,654 INFO [train.py:968] (1/2) Epoch 4, batch 20550, giga_loss[loss=0.3736, simple_loss=0.4216, pruned_loss=0.1628, over 27671.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3739, pruned_loss=0.1225, over 5676747.02 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3809, pruned_loss=0.1213, over 5728158.97 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1222, over 5685226.91 frames. ], batch size: 472, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:35:29,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:35:31,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:35:57,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=156727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:36:10,906 INFO [train.py:968] (1/2) Epoch 4, batch 20600, giga_loss[loss=0.3199, simple_loss=0.3842, pruned_loss=0.1278, over 28684.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3754, pruned_loss=0.1231, over 5677852.96 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3809, pruned_loss=0.1213, over 5730009.73 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3739, pruned_loss=0.1228, over 5682487.71 frames. ], batch size: 242, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:36:48,582 INFO [optim.py:369] (1/2) 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 05:36:54,594 INFO [train.py:968] (1/2) Epoch 4, batch 20650, giga_loss[loss=0.3171, simple_loss=0.3842, pruned_loss=0.125, over 28883.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.379, pruned_loss=0.1259, over 5684007.98 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.381, pruned_loss=0.1213, over 5729922.64 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3778, pruned_loss=0.1257, over 5686727.57 frames. ], batch size: 112, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:37:00,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-02 05:37:34,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:37:38,656 INFO [train.py:968] (1/2) Epoch 4, batch 20700, giga_loss[loss=0.3537, simple_loss=0.4045, pruned_loss=0.1514, over 28656.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3804, pruned_loss=0.1267, over 5690530.60 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3807, pruned_loss=0.1212, over 5724636.43 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3795, pruned_loss=0.1269, over 5696259.17 frames. ], batch size: 92, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:37:57,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1020, 1.0637, 0.9283, 1.2337], device='cuda:1'), covar=tensor([0.0790, 0.0456, 0.0329, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0133, 0.0137, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0055], device='cuda:1')
+2023-03-02 05:38:19,663 INFO [optim.py:369] (1/2) 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 05:38:27,473 INFO [train.py:968] (1/2) Epoch 4, batch 20750, giga_loss[loss=0.2997, simple_loss=0.3698, pruned_loss=0.1148, over 28947.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3817, pruned_loss=0.1286, over 5673940.65 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3808, pruned_loss=0.1212, over 5723686.25 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.381, pruned_loss=0.1287, over 5679044.25 frames. ], batch size: 227, lr: 7.78e-03, grad_scale: 4.0
+2023-03-02 05:38:50,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5078, 3.3260, 1.5620, 1.4608], device='cuda:1'), covar=tensor([0.0873, 0.0296, 0.0821, 0.1304], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0459, 0.0307, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 05:39:12,353 INFO [train.py:968] (1/2) Epoch 4, batch 20800, giga_loss[loss=0.3015, simple_loss=0.3631, pruned_loss=0.1199, over 28470.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3833, pruned_loss=0.1302, over 5681970.60 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3812, pruned_loss=0.1213, over 5726873.20 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3824, pruned_loss=0.1304, over 5682248.41 frames. ], batch size: 71, lr: 7.77e-03, grad_scale: 8.0
+2023-03-02 05:39:42,454 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=156981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:39:44,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=156984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:39:44,417 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=156984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:39:44,753 INFO [optim.py:369] (1/2) 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 05:39:50,455 INFO [train.py:968] (1/2) Epoch 4, batch 20850, giga_loss[loss=0.3102, simple_loss=0.3722, pruned_loss=0.1241, over 28765.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3821, pruned_loss=0.1287, over 5685207.74 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3813, pruned_loss=0.1213, over 5719808.23 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3814, pruned_loss=0.129, over 5690752.58 frames. ], batch size: 119, lr: 7.77e-03, grad_scale: 8.0
+2023-03-02 05:40:01,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3833, 1.7323, 1.6332, 1.5684], device='cuda:1'), covar=tensor([0.1469, 0.1947, 0.1170, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0755, 0.0758, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 05:40:06,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:40:24,409 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-02 05:40:29,620 INFO [train.py:968] (1/2) Epoch 4, batch 20900, giga_loss[loss=0.3259, simple_loss=0.3842, pruned_loss=0.1338, over 27634.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3826, pruned_loss=0.1278, over 5683732.11 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.382, pruned_loss=0.1217, over 5716208.06 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3813, pruned_loss=0.1279, over 5689916.96 frames. ], batch size: 472, lr: 7.77e-03, grad_scale: 8.0
+2023-03-02 05:40:42,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:40:43,063 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 05:41:02,096 INFO [optim.py:369] (1/2) 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 05:41:09,396 INFO [train.py:968] (1/2) Epoch 4, batch 20950, giga_loss[loss=0.2937, simple_loss=0.3746, pruned_loss=0.1064, over 28877.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3819, pruned_loss=0.1257, over 5694248.30 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.382, pruned_loss=0.1219, over 5722280.21 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3809, pruned_loss=0.1258, over 5692970.61 frames. ], batch size: 112, lr: 7.77e-03, grad_scale: 4.0
+2023-03-02 05:41:38,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:41:41,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:41:49,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:41:49,953 INFO [train.py:968] (1/2) Epoch 4, batch 21000, giga_loss[loss=0.2981, simple_loss=0.3721, pruned_loss=0.112, over 29088.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3808, pruned_loss=0.1248, over 5697634.74 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3822, pruned_loss=0.1222, over 5725964.09 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3797, pruned_loss=0.1246, over 5692742.94 frames. ], batch size: 155, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:41:49,954 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 05:41:59,967 INFO [train.py:1012] (1/2) Epoch 4, validation: loss=0.2497, simple_loss=0.3505, pruned_loss=0.07442, over 944034.00 frames. 
+2023-03-02 05:41:59,968 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 05:42:09,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6194, 1.5512, 1.3374, 1.0722], device='cuda:1'), covar=tensor([0.0854, 0.0861, 0.0580, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.1301, 0.1023, 0.1052, 0.1140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 05:42:12,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:42:34,281 INFO [optim.py:369] (1/2) 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 05:42:38,527 INFO [train.py:968] (1/2) Epoch 4, batch 21050, giga_loss[loss=0.3296, simple_loss=0.3871, pruned_loss=0.136, over 28988.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3781, pruned_loss=0.123, over 5709386.10 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3824, pruned_loss=0.1224, over 5730253.71 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.377, pruned_loss=0.1227, over 5700900.05 frames. ], batch size: 136, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:42:45,827 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:42:47,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:43:10,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:43:16,876 INFO [train.py:968] (1/2) Epoch 4, batch 21100, libri_loss[loss=0.35, simple_loss=0.414, pruned_loss=0.143, over 29384.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3761, pruned_loss=0.1217, over 5713753.35 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3828, pruned_loss=0.1226, over 5733137.22 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3747, pruned_loss=0.1213, over 5704068.46 frames. ], batch size: 92, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:43:45,322 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:43:51,407 INFO [optim.py:369] (1/2) 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 05:43:54,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6121, 2.9472, 1.6209, 1.4072], device='cuda:1'), covar=tensor([0.0806, 0.0489, 0.0676, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.1027, 0.1054, 0.1141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 05:43:55,828 INFO [train.py:968] (1/2) Epoch 4, batch 21150, libri_loss[loss=0.2885, simple_loss=0.3661, pruned_loss=0.1054, over 29521.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3747, pruned_loss=0.1209, over 5719457.67 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3831, pruned_loss=0.1229, over 5735536.13 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3732, pruned_loss=0.1203, over 5708987.63 frames. ], batch size: 80, lr: 7.77e-03, grad_scale: 2.0
+2023-03-02 05:44:26,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1441, 1.8098, 1.4162, 0.4447], device='cuda:1'), covar=tensor([0.2006, 0.1035, 0.2226, 0.2162], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.1217, 0.1333, 0.1106], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 05:44:36,999 INFO [train.py:968] (1/2) Epoch 4, batch 21200, giga_loss[loss=0.3859, simple_loss=0.4229, pruned_loss=0.1744, over 27629.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3761, pruned_loss=0.123, over 5710939.00 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3833, pruned_loss=0.1231, over 5736380.02 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3744, pruned_loss=0.1222, over 5701167.18 frames. ], batch size: 472, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:45:13,284 INFO [optim.py:369] (1/2) 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 05:45:17,434 INFO [train.py:968] (1/2) Epoch 4, batch 21250, giga_loss[loss=0.2955, simple_loss=0.3671, pruned_loss=0.1119, over 28764.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3762, pruned_loss=0.1226, over 5720575.34 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3834, pruned_loss=0.1232, over 5737325.10 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3748, pruned_loss=0.1219, over 5711610.85 frames. ], batch size: 119, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:45:26,777 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:45:58,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 05:45:59,293 INFO [train.py:968] (1/2) Epoch 4, batch 21300, giga_loss[loss=0.2661, simple_loss=0.3518, pruned_loss=0.09018, over 28705.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3772, pruned_loss=0.1228, over 5711916.36 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3841, pruned_loss=0.1239, over 5741325.84 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3753, pruned_loss=0.1217, over 5700359.36 frames. ], batch size: 60, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:45:59,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-02 05:46:11,350 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-02 05:46:37,225 INFO [optim.py:369] (1/2) 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 05:46:41,261 INFO [train.py:968] (1/2) Epoch 4, batch 21350, giga_loss[loss=0.3366, simple_loss=0.4034, pruned_loss=0.1349, over 28755.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3758, pruned_loss=0.1212, over 5723281.70 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3845, pruned_loss=0.1244, over 5744146.07 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3737, pruned_loss=0.1197, over 5711218.21 frames. ], batch size: 242, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:47:01,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:47:15,224 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 21400, giga_loss[loss=0.2992, simple_loss=0.37, pruned_loss=0.1142, over 28538.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.376, pruned_loss=0.1214, over 5730557.63 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.385, pruned_loss=0.1248, over 5748474.44 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3736, pruned_loss=0.1198, over 5716618.10 frames. ], batch size: 85, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:47:28,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 05:47:32,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-02 05:47:41,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4712, 1.6381, 1.5799, 1.6441], device='cuda:1'), covar=tensor([0.1064, 0.1387, 0.1105, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0747, 0.0620, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 05:47:57,018 INFO [optim.py:369] (1/2) 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 05:48:02,326 INFO [train.py:968] (1/2) Epoch 4, batch 21450, giga_loss[loss=0.2775, simple_loss=0.3506, pruned_loss=0.1022, over 28605.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.373, pruned_loss=0.12, over 5732501.73 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3853, pruned_loss=0.125, over 5750179.80 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3708, pruned_loss=0.1185, over 5719724.43 frames. ], batch size: 85, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:48:19,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4145, 1.8260, 1.6825, 1.6040], device='cuda:1'), covar=tensor([0.1398, 0.1866, 0.1134, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0753, 0.0757, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 05:48:22,014 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-02 05:48:41,343 INFO [train.py:968] (1/2) Epoch 4, batch 21500, giga_loss[loss=0.2869, simple_loss=0.3629, pruned_loss=0.1055, over 28260.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.37, pruned_loss=0.1185, over 5711746.64 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3858, pruned_loss=0.1257, over 5734992.81 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3674, pruned_loss=0.1165, over 5714232.11 frames. ], batch size: 368, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:48:49,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:48:55,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:48:58,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:49:03,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 05:49:19,588 INFO [optim.py:369] (1/2) 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 05:49:23,119 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:49:23,508 INFO [train.py:968] (1/2) Epoch 4, batch 21550, giga_loss[loss=0.2908, simple_loss=0.3586, pruned_loss=0.1115, over 29086.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3702, pruned_loss=0.1193, over 5717867.74 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3862, pruned_loss=0.1261, over 5736060.85 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3677, pruned_loss=0.1174, over 5718882.70 frames. ], batch size: 128, lr: 7.76e-03, grad_scale: 4.0
+2023-03-02 05:50:07,151 INFO [train.py:968] (1/2) Epoch 4, batch 21600, giga_loss[loss=0.2753, simple_loss=0.3447, pruned_loss=0.1029, over 28943.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3693, pruned_loss=0.1194, over 5709718.86 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3865, pruned_loss=0.1263, over 5728910.36 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1176, over 5716075.29 frames. ], batch size: 164, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:50:11,475 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:21,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5534, 1.5007, 1.5296, 1.4370], device='cuda:1'), covar=tensor([0.0956, 0.1470, 0.1444, 0.1388], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0739, 0.0622, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 05:50:34,879 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:37,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:39,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2268, 1.4531, 1.0619, 0.6105], device='cuda:1'), covar=tensor([0.0971, 0.0658, 0.0577, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.1036, 0.1063, 0.1145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 05:50:42,902 INFO [optim.py:369] (1/2) 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 05:50:48,750 INFO [train.py:968] (1/2) Epoch 4, batch 21650, giga_loss[loss=0.299, simple_loss=0.3594, pruned_loss=0.1193, over 28702.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3675, pruned_loss=0.1193, over 5702361.49 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3864, pruned_loss=0.1263, over 5720543.03 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5714710.20 frames. ], batch size: 242, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:50:51,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:50:51,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157797.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:50:52,185 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-02 05:50:54,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=157799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:51:15,451 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:51:25,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1918, 1.5542, 1.1746, 1.0909], device='cuda:1'), covar=tensor([0.2155, 0.1906, 0.2049, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.1065, 0.0842, 0.0953, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 05:51:28,027 INFO [train.py:968] (1/2) Epoch 4, batch 21700, giga_loss[loss=0.2766, simple_loss=0.3443, pruned_loss=0.1045, over 29084.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3658, pruned_loss=0.1194, over 5699201.86 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3867, pruned_loss=0.1266, over 5712153.60 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3638, pruned_loss=0.118, over 5715624.55 frames. ], batch size: 155, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:52:02,497 INFO [optim.py:369] (1/2) 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 05:52:07,714 INFO [train.py:968] (1/2) Epoch 4, batch 21750, libri_loss[loss=0.3754, simple_loss=0.4369, pruned_loss=0.157, over 25879.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3642, pruned_loss=0.1193, over 5691091.24 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3872, pruned_loss=0.127, over 5709788.96 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3616, pruned_loss=0.1176, over 5706050.27 frames. ], batch size: 136, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:52:12,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:52:20,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=157910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:52:29,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=157921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:52:31,500 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 21800, giga_loss[loss=0.2902, simple_loss=0.3522, pruned_loss=0.1141, over 28859.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3627, pruned_loss=0.1183, over 5696594.35 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3873, pruned_loss=0.1272, over 5711793.10 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3603, pruned_loss=0.1167, over 5706451.94 frames. ], batch size: 112, lr: 7.75e-03, grad_scale: 8.0
+2023-03-02 05:52:58,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=157953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:53:04,637 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=157961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:53:24,826 INFO [optim.py:369] (1/2) 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 05:53:28,958 INFO [train.py:968] (1/2) Epoch 4, batch 21850, libri_loss[loss=0.3559, simple_loss=0.4213, pruned_loss=0.1453, over 29097.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3638, pruned_loss=0.1187, over 5700649.59 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3886, pruned_loss=0.1282, over 5715783.37 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.36, pruned_loss=0.1162, over 5704210.99 frames. ], batch size: 101, lr: 7.75e-03, grad_scale: 4.0
+2023-03-02 05:53:38,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7564, 2.4362, 1.8169, 0.9158], device='cuda:1'), covar=tensor([0.1638, 0.0887, 0.1502, 0.1805], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.1213, 0.1334, 0.1103], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 05:53:51,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-02 05:53:59,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 05:54:11,275 INFO [train.py:968] (1/2) Epoch 4, batch 21900, giga_loss[loss=0.305, simple_loss=0.3826, pruned_loss=0.1137, over 29018.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3668, pruned_loss=0.1198, over 5703389.97 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3884, pruned_loss=0.1284, over 5718943.06 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3633, pruned_loss=0.1174, over 5703109.88 frames. ], batch size: 164, lr: 7.75e-03, grad_scale: 4.0
+2023-03-02 05:54:18,975 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 05:54:20,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:54:22,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:54:46,306 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 4, batch 21950, giga_loss[loss=0.2993, simple_loss=0.3634, pruned_loss=0.1175, over 28928.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.1201, over 5704929.38 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3886, pruned_loss=0.1291, over 5714737.09 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3641, pruned_loss=0.117, over 5708714.18 frames. ], batch size: 106, lr: 7.75e-03, grad_scale: 4.0
+2023-03-02 05:55:16,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:55:31,830 INFO [train.py:968] (1/2) Epoch 4, batch 22000, giga_loss[loss=0.3175, simple_loss=0.3812, pruned_loss=0.1269, over 28945.00 frames. ], tot_loss[loss=0.307, simple_loss=0.371, pruned_loss=0.1215, over 5699455.78 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.3894, pruned_loss=0.1303, over 5715602.34 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3662, pruned_loss=0.1176, over 5700897.87 frames. ], batch size: 186, lr: 7.74e-03, grad_scale: 8.0
+2023-03-02 05:55:37,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:55:55,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158172.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:56:10,624 INFO [optim.py:369] (1/2) 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 05:56:13,413 INFO [train.py:968] (1/2) Epoch 4, batch 22050, giga_loss[loss=0.2784, simple_loss=0.3579, pruned_loss=0.09948, over 28656.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3702, pruned_loss=0.1206, over 5679802.58 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3895, pruned_loss=0.1307, over 5700369.26 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.366, pruned_loss=0.117, over 5693379.06 frames. ], batch size: 262, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:56:56,205 INFO [train.py:968] (1/2) Epoch 4, batch 22100, giga_loss[loss=0.2995, simple_loss=0.3708, pruned_loss=0.1141, over 28780.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3697, pruned_loss=0.1203, over 5684943.84 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3894, pruned_loss=0.131, over 5700634.96 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.366, pruned_loss=0.1169, over 5694987.73 frames. ], batch size: 227, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:56:59,430 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:15,598 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:17,521 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158274.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:57:33,275 INFO [optim.py:369] (1/2) 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 05:57:36,159 INFO [train.py:968] (1/2) Epoch 4, batch 22150, libri_loss[loss=0.3756, simple_loss=0.4323, pruned_loss=0.1595, over 29523.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.372, pruned_loss=0.1222, over 5692903.36 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3906, pruned_loss=0.1319, over 5704698.92 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3676, pruned_loss=0.1185, over 5696934.70 frames. ], batch size: 82, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:57:36,478 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:38,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:41,030 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:57:54,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158315.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:57:56,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:57:57,908 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-02 05:58:01,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:58:09,917 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 05:58:14,090 INFO [train.py:968] (1/2) Epoch 4, batch 22200, giga_loss[loss=0.3324, simple_loss=0.395, pruned_loss=0.1349, over 28718.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3737, pruned_loss=0.1235, over 5676176.41 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.391, pruned_loss=0.1323, over 5683743.47 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3691, pruned_loss=0.1198, over 5699181.01 frames. ], batch size: 284, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:58:18,942 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158347.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:58:53,662 INFO [optim.py:369] (1/2) 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 05:58:54,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1272, 1.1837, 1.0059, 1.0302], device='cuda:1'), covar=tensor([0.0525, 0.0426, 0.0899, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0467, 0.0525, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 05:58:56,191 INFO [train.py:968] (1/2) Epoch 4, batch 22250, libri_loss[loss=0.4124, simple_loss=0.4616, pruned_loss=0.1816, over 29666.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3771, pruned_loss=0.1256, over 5680668.88 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3925, pruned_loss=0.1335, over 5685107.28 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3717, pruned_loss=0.1213, over 5697460.25 frames. ], batch size: 88, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:59:16,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158417.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 05:59:18,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158420.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 05:59:35,823 INFO [train.py:968] (1/2) Epoch 4, batch 22300, libri_loss[loss=0.3799, simple_loss=0.435, pruned_loss=0.1623, over 29766.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3792, pruned_loss=0.1265, over 5692179.15 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.393, pruned_loss=0.134, over 5687079.30 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3742, pruned_loss=0.1225, over 5703652.21 frames. ], batch size: 87, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 05:59:41,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158449.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:00:03,307 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:05,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=158482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:10,670 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 22350, giga_loss[loss=0.3133, simple_loss=0.3811, pruned_loss=0.1227, over 28805.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3804, pruned_loss=0.1271, over 5686646.73 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3934, pruned_loss=0.1344, over 5679172.72 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3756, pruned_loss=0.1233, over 5703551.45 frames. ], batch size: 199, lr: 7.74e-03, grad_scale: 4.0
+2023-03-02 06:00:27,545 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:30,092 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:00:53,579 INFO [train.py:968] (1/2) Epoch 4, batch 22400, giga_loss[loss=0.2863, simple_loss=0.3497, pruned_loss=0.1114, over 28534.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3816, pruned_loss=0.1277, over 5693386.59 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3946, pruned_loss=0.1354, over 5682050.74 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3764, pruned_loss=0.1236, over 5704530.30 frames. ], batch size: 71, lr: 7.74e-03, grad_scale: 8.0
+2023-03-02 06:01:32,976 INFO [optim.py:369] (1/2) 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 06:01:35,925 INFO [train.py:968] (1/2) Epoch 4, batch 22450, giga_loss[loss=0.3201, simple_loss=0.3853, pruned_loss=0.1275, over 28648.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3821, pruned_loss=0.1279, over 5702287.19 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.395, pruned_loss=0.136, over 5687479.92 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3773, pruned_loss=0.1239, over 5706823.45 frames. ], batch size: 336, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:01:58,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:02:08,260 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0104, 1.8171, 1.4211, 1.5884], device='cuda:1'), covar=tensor([0.1105, 0.1929, 0.1730, 0.1750], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0758, 0.0636, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 06:02:16,722 INFO [train.py:968] (1/2) Epoch 4, batch 22500, giga_loss[loss=0.2515, simple_loss=0.3373, pruned_loss=0.0829, over 29003.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3804, pruned_loss=0.1268, over 5695528.70 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.3951, pruned_loss=0.1362, over 5681607.01 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3762, pruned_loss=0.1233, over 5703967.73 frames. ], batch size: 164, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:02:20,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5662, 4.2389, 4.2632, 1.8224], device='cuda:1'), covar=tensor([0.0392, 0.0367, 0.0675, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0674, 0.0791, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:02:58,019 INFO [optim.py:369] (1/2) 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 06:03:00,039 INFO [train.py:968] (1/2) Epoch 4, batch 22550, giga_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 28602.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3783, pruned_loss=0.1254, over 5705330.31 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3957, pruned_loss=0.1367, over 5688431.00 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.374, pruned_loss=0.1219, over 5706611.03 frames. ], batch size: 336, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:03:43,754 INFO [train.py:968] (1/2) Epoch 4, batch 22600, giga_loss[loss=0.2816, simple_loss=0.3565, pruned_loss=0.1033, over 28743.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3727, pruned_loss=0.1218, over 5708228.95 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3958, pruned_loss=0.1368, over 5689090.00 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.369, pruned_loss=0.1189, over 5708788.97 frames. ], batch size: 284, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:04:02,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=158764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:04:04,114 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=158777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:04:20,956 INFO [optim.py:369] (1/2) 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 06:04:21,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-02 06:04:22,861 INFO [train.py:968] (1/2) Epoch 4, batch 22650, giga_loss[loss=0.2495, simple_loss=0.3243, pruned_loss=0.08732, over 28878.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3715, pruned_loss=0.1208, over 5701987.59 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3964, pruned_loss=0.1374, over 5688677.51 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3675, pruned_loss=0.1174, over 5703181.34 frames. ], batch size: 145, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:04:26,304 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=158796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:05:04,769 INFO [train.py:968] (1/2) Epoch 4, batch 22700, giga_loss[loss=0.3169, simple_loss=0.3962, pruned_loss=0.1188, over 28650.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3733, pruned_loss=0.1204, over 5697425.53 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3966, pruned_loss=0.1377, over 5687903.77 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3695, pruned_loss=0.1172, over 5699114.55 frames. ], batch size: 242, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:05:26,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4955, 1.7962, 1.1653, 0.7216], device='cuda:1'), covar=tensor([0.2437, 0.1406, 0.1479, 0.2548], device='cuda:1'), in_proj_covar=tensor([0.1322, 0.1238, 0.1341, 0.1120], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 06:05:45,248 INFO [optim.py:369] (1/2) 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 06:05:45,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=158890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:05:46,974 INFO [train.py:968] (1/2) Epoch 4, batch 22750, giga_loss[loss=0.3276, simple_loss=0.3884, pruned_loss=0.1334, over 28844.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3747, pruned_loss=0.1208, over 5696407.20 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3968, pruned_loss=0.138, over 5689004.25 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3714, pruned_loss=0.118, over 5696756.30 frames. ], batch size: 243, lr: 7.73e-03, grad_scale: 4.0
+2023-03-02 06:06:26,752 INFO [train.py:968] (1/2) Epoch 4, batch 22800, giga_loss[loss=0.2686, simple_loss=0.3294, pruned_loss=0.1039, over 28664.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3727, pruned_loss=0.1214, over 5688748.86 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3974, pruned_loss=0.1387, over 5683902.70 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3688, pruned_loss=0.1179, over 5694415.92 frames. ], batch size: 78, lr: 7.73e-03, grad_scale: 8.0
+2023-03-02 06:06:29,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-02 06:06:39,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9538, 1.0349, 0.8191, 0.7103], device='cuda:1'), covar=tensor([0.0698, 0.0677, 0.0528, 0.0652], device='cuda:1'), in_proj_covar=tensor([0.1339, 0.1071, 0.1092, 0.1162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 06:07:06,250 INFO [optim.py:369] (1/2) 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 06:07:08,158 INFO [train.py:968] (1/2) Epoch 4, batch 22850, giga_loss[loss=0.3104, simple_loss=0.3649, pruned_loss=0.1279, over 28996.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5698404.50 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.398, pruned_loss=0.1392, over 5688881.83 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.368, pruned_loss=0.1192, over 5698596.25 frames. ], batch size: 136, lr: 7.72e-03, grad_scale: 8.0
+2023-03-02 06:07:39,701 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:07:41,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:07:48,033 INFO [train.py:968] (1/2) Epoch 4, batch 22900, giga_loss[loss=0.2935, simple_loss=0.3569, pruned_loss=0.115, over 29078.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.123, over 5708734.77 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3987, pruned_loss=0.1401, over 5690950.47 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3658, pruned_loss=0.1188, over 5707232.11 frames. ], batch size: 155, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:08:06,124 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:08:27,089 INFO [optim.py:369] (1/2) 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 06:08:28,346 INFO [train.py:968] (1/2) Epoch 4, batch 22950, giga_loss[loss=0.3283, simple_loss=0.3811, pruned_loss=0.1378, over 28879.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3702, pruned_loss=0.1244, over 5698929.03 frames. ], libri_tot_loss[loss=0.3395, simple_loss=0.3985, pruned_loss=0.1402, over 5687753.85 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3657, pruned_loss=0.1204, over 5701859.87 frames. ], batch size: 112, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:08:41,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9309, 1.6994, 1.7328, 1.6622], device='cuda:1'), covar=tensor([0.0916, 0.1748, 0.1529, 0.1422], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0759, 0.0633, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 06:09:08,225 INFO [train.py:968] (1/2) Epoch 4, batch 23000, giga_loss[loss=0.3066, simple_loss=0.3569, pruned_loss=0.1281, over 28785.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3685, pruned_loss=0.1227, over 5709906.92 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3988, pruned_loss=0.1406, over 5688367.60 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3642, pruned_loss=0.1189, over 5711879.05 frames. ], batch size: 99, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:09:11,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2692, 1.2980, 4.4827, 3.2802], device='cuda:1'), covar=tensor([0.1589, 0.2224, 0.0303, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0544, 0.0501, 0.0699, 0.0567], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 06:09:15,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:09:29,305 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:09:34,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:09:46,701 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 23050, giga_loss[loss=0.247, simple_loss=0.3128, pruned_loss=0.09055, over 28490.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.364, pruned_loss=0.1205, over 5710653.26 frames. ], libri_tot_loss[loss=0.3401, simple_loss=0.3987, pruned_loss=0.1408, over 5693753.56 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3598, pruned_loss=0.1168, over 5708111.46 frames. ], batch size: 71, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:09:52,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-02 06:10:14,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159228.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:10:27,871 INFO [train.py:968] (1/2) Epoch 4, batch 23100, giga_loss[loss=0.2939, simple_loss=0.3581, pruned_loss=0.1148, over 28218.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3599, pruned_loss=0.1181, over 5713434.20 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3986, pruned_loss=0.141, over 5700825.89 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3555, pruned_loss=0.1143, over 5705756.44 frames. ], batch size: 368, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:10:40,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1628, 1.7505, 1.3728, 0.4212], device='cuda:1'), covar=tensor([0.1819, 0.1133, 0.2020, 0.2212], device='cuda:1'), in_proj_covar=tensor([0.1323, 0.1241, 0.1347, 0.1116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 06:11:04,656 INFO [optim.py:369] (1/2) 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 06:11:06,055 INFO [train.py:968] (1/2) Epoch 4, batch 23150, giga_loss[loss=0.2749, simple_loss=0.3419, pruned_loss=0.104, over 28947.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3568, pruned_loss=0.1157, over 5713283.83 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.3985, pruned_loss=0.141, over 5700460.29 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3527, pruned_loss=0.1124, over 5707543.71 frames. ], batch size: 106, lr: 7.72e-03, grad_scale: 4.0
+2023-03-02 06:11:07,874 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:11:11,028 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:11:43,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1902, 1.6026, 1.4625, 1.4413], device='cuda:1'), covar=tensor([0.1248, 0.1752, 0.1065, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0755, 0.0760, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 06:11:50,951 INFO [train.py:968] (1/2) Epoch 4, batch 23200, giga_loss[loss=0.2924, simple_loss=0.3577, pruned_loss=0.1136, over 28773.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3588, pruned_loss=0.1163, over 5708777.22 frames. ], libri_tot_loss[loss=0.3404, simple_loss=0.3986, pruned_loss=0.1411, over 5697249.63 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3551, pruned_loss=0.1133, over 5707302.85 frames. ], batch size: 199, lr: 7.72e-03, grad_scale: 8.0
+2023-03-02 06:12:26,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6461, 1.4344, 1.1212, 1.3505], device='cuda:1'), covar=tensor([0.0664, 0.0770, 0.1011, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0461, 0.0510, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:12:31,037 INFO [optim.py:369] (1/2) 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 06:12:32,385 INFO [train.py:968] (1/2) Epoch 4, batch 23250, giga_loss[loss=0.3238, simple_loss=0.3919, pruned_loss=0.1278, over 28934.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3639, pruned_loss=0.119, over 5713714.87 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3993, pruned_loss=0.1419, over 5701512.26 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3595, pruned_loss=0.1154, over 5709191.16 frames. ], batch size: 227, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:13:10,973 INFO [train.py:968] (1/2) Epoch 4, batch 23300, giga_loss[loss=0.3006, simple_loss=0.3695, pruned_loss=0.1158, over 29034.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1227, over 5706643.23 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.4, pruned_loss=0.1425, over 5697036.94 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3649, pruned_loss=0.1184, over 5707793.83 frames. ], batch size: 128, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:13:20,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9766, 1.2305, 0.9585, 0.2742], device='cuda:1'), covar=tensor([0.1126, 0.0970, 0.1709, 0.2011], device='cuda:1'), in_proj_covar=tensor([0.1320, 0.1230, 0.1330, 0.1120], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 06:13:25,589 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:13:49,094 INFO [optim.py:369] (1/2) 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 06:13:50,541 INFO [train.py:968] (1/2) Epoch 4, batch 23350, giga_loss[loss=0.3064, simple_loss=0.3764, pruned_loss=0.1182, over 28945.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3744, pruned_loss=0.125, over 5697801.69 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3999, pruned_loss=0.1427, over 5694596.20 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3693, pruned_loss=0.1209, over 5700559.23 frames. ], batch size: 186, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:14:33,937 INFO [train.py:968] (1/2) Epoch 4, batch 23400, giga_loss[loss=0.3332, simple_loss=0.3922, pruned_loss=0.137, over 27951.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3775, pruned_loss=0.1266, over 5695915.28 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.4002, pruned_loss=0.1431, over 5696898.37 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1226, over 5696075.13 frames. ], batch size: 412, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:14:36,339 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:14:40,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:14:41,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:15:02,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4875, 3.2769, 3.1755, 1.6329], device='cuda:1'), covar=tensor([0.0663, 0.0687, 0.0981, 0.2290], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0678, 0.0790, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 06:15:20,026 INFO [optim.py:369] (1/2) 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 06:15:21,491 INFO [train.py:968] (1/2) Epoch 4, batch 23450, giga_loss[loss=0.4483, simple_loss=0.4692, pruned_loss=0.2137, over 28614.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3838, pruned_loss=0.1331, over 5685333.66 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.4006, pruned_loss=0.1436, over 5692913.99 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3794, pruned_loss=0.1293, over 5688976.80 frames. ], batch size: 307, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:15:33,417 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159603.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:16:08,395 INFO [train.py:968] (1/2) Epoch 4, batch 23500, giga_loss[loss=0.372, simple_loss=0.4171, pruned_loss=0.1635, over 27527.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3895, pruned_loss=0.1383, over 5682790.91 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.4006, pruned_loss=0.1436, over 5697162.79 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3857, pruned_loss=0.1351, over 5681767.45 frames. ], batch size: 472, lr: 7.71e-03, grad_scale: 4.0
+2023-03-02 06:16:48,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=159683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:16:52,846 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:16:56,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:16:56,928 INFO [optim.py:369] (1/2) 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 06:16:57,536 INFO [train.py:968] (1/2) Epoch 4, batch 23550, libri_loss[loss=0.4328, simple_loss=0.4526, pruned_loss=0.2065, over 29768.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3969, pruned_loss=0.1438, over 5689142.15 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.4009, pruned_loss=0.1442, over 5701721.00 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3934, pruned_loss=0.1406, over 5683976.58 frames. ], batch size: 87, lr: 7.71e-03, grad_scale: 4.0
+2023-03-02 06:16:59,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:02,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:24,617 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:31,138 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:17:43,587 INFO [train.py:968] (1/2) Epoch 4, batch 23600, giga_loss[loss=0.3358, simple_loss=0.3972, pruned_loss=0.1372, over 28818.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4013, pruned_loss=0.1483, over 5685795.34 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.4003, pruned_loss=0.1439, over 5707485.06 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.399, pruned_loss=0.1462, over 5675666.26 frames. ], batch size: 243, lr: 7.71e-03, grad_scale: 8.0
+2023-03-02 06:17:46,531 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=159746.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:17:49,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159749.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:18:17,044 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=159778.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:18:31,777 INFO [optim.py:369] (1/2) 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 06:18:31,790 INFO [train.py:968] (1/2) Epoch 4, batch 23650, libri_loss[loss=0.3198, simple_loss=0.3699, pruned_loss=0.1348, over 29328.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.408, pruned_loss=0.1553, over 5677238.04 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.4008, pruned_loss=0.1447, over 5714959.58 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.4059, pruned_loss=0.153, over 5661094.29 frames. ], batch size: 71, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:18:49,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-02 06:19:03,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2080, 1.5591, 1.2501, 1.4604], device='cuda:1'), covar=tensor([0.0833, 0.0317, 0.0344, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0132, 0.0135, 0.0243], device='cuda:1'), out_proj_covar=tensor([0.0049, 0.0036, 0.0033, 0.0055], device='cuda:1')
+2023-03-02 06:19:15,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:19:21,366 INFO [train.py:968] (1/2) Epoch 4, batch 23700, giga_loss[loss=0.3561, simple_loss=0.4075, pruned_loss=0.1523, over 28920.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.4131, pruned_loss=0.159, over 5664375.78 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.4009, pruned_loss=0.1448, over 5707790.72 frames. ], giga_tot_loss[loss=0.3633, simple_loss=0.4116, pruned_loss=0.1575, over 5657894.56 frames. ], batch size: 106, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:20:10,406 INFO [optim.py:369] (1/2) 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 06:20:10,419 INFO [train.py:968] (1/2) Epoch 4, batch 23750, giga_loss[loss=0.3733, simple_loss=0.4189, pruned_loss=0.1638, over 28627.00 frames. ], tot_loss[loss=0.3688, simple_loss=0.415, pruned_loss=0.1613, over 5661326.66 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.4012, pruned_loss=0.1451, over 5709893.88 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4136, pruned_loss=0.16, over 5654088.04 frames. ], batch size: 307, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:20:29,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2436, 1.2062, 1.1393, 1.0447], device='cuda:1'), covar=tensor([0.0514, 0.0423, 0.0792, 0.0711], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0471, 0.0520, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:20:35,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5854, 3.3896, 3.2943, 1.6602], device='cuda:1'), covar=tensor([0.0612, 0.0522, 0.0900, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0688, 0.0806, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:20:42,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=159927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:20:58,346 INFO [train.py:968] (1/2) Epoch 4, batch 23800, giga_loss[loss=0.4478, simple_loss=0.4475, pruned_loss=0.224, over 23689.00 frames. ], tot_loss[loss=0.372, simple_loss=0.4168, pruned_loss=0.1636, over 5663702.70 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.4019, pruned_loss=0.1457, over 5714826.62 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4156, pruned_loss=0.1625, over 5651970.15 frames. ], batch size: 705, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:21:39,812 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=159983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:21:50,871 INFO [optim.py:369] (1/2) 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 06:21:50,884 INFO [train.py:968] (1/2) Epoch 4, batch 23850, libri_loss[loss=0.3679, simple_loss=0.402, pruned_loss=0.1669, over 29555.00 frames. ], tot_loss[loss=0.3785, simple_loss=0.4207, pruned_loss=0.1682, over 5649178.46 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.4019, pruned_loss=0.1458, over 5715807.48 frames. ], giga_tot_loss[loss=0.3773, simple_loss=0.4199, pruned_loss=0.1673, over 5638685.29 frames. ], batch size: 76, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:22:16,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:22:31,669 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:22:50,937 INFO [train.py:968] (1/2) Epoch 4, batch 23900, giga_loss[loss=0.4328, simple_loss=0.4561, pruned_loss=0.2048, over 28784.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.4252, pruned_loss=0.1723, over 5637980.49 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.4028, pruned_loss=0.1467, over 5718583.61 frames. ], giga_tot_loss[loss=0.3831, simple_loss=0.4239, pruned_loss=0.1712, over 5625916.41 frames. ], batch size: 307, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:23:08,880 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:23:23,076 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:23:25,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:23:47,062 INFO [optim.py:369] (1/2) 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 06:23:47,074 INFO [train.py:968] (1/2) Epoch 4, batch 23950, giga_loss[loss=0.3921, simple_loss=0.4261, pruned_loss=0.179, over 28328.00 frames. ], tot_loss[loss=0.3843, simple_loss=0.4237, pruned_loss=0.1724, over 5620719.78 frames. ], libri_tot_loss[loss=0.3482, simple_loss=0.4027, pruned_loss=0.1468, over 5721448.00 frames. ], giga_tot_loss[loss=0.3835, simple_loss=0.4233, pruned_loss=0.1718, over 5607107.84 frames. ], batch size: 368, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:23:55,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:24:14,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4104, 2.7362, 1.4506, 1.3260], device='cuda:1'), covar=tensor([0.0828, 0.0344, 0.0775, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0468, 0.0306, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 06:24:35,138 INFO [train.py:968] (1/2) Epoch 4, batch 24000, giga_loss[loss=0.3192, simple_loss=0.3727, pruned_loss=0.1328, over 28922.00 frames. ], tot_loss[loss=0.3807, simple_loss=0.4208, pruned_loss=0.1703, over 5632638.62 frames. ], libri_tot_loss[loss=0.3476, simple_loss=0.4021, pruned_loss=0.1466, over 5720774.51 frames. ], giga_tot_loss[loss=0.3812, simple_loss=0.4212, pruned_loss=0.1706, over 5620699.84 frames. ], batch size: 106, lr: 7.70e-03, grad_scale: 8.0
+2023-03-02 06:24:35,139 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 06:24:43,740 INFO [train.py:1012] (1/2) Epoch 4, validation: loss=0.2484, simple_loss=0.3515, pruned_loss=0.07269, over 944034.00 frames. 
+2023-03-02 06:24:43,740 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 06:25:29,076 INFO [train.py:968] (1/2) Epoch 4, batch 24050, giga_loss[loss=0.3664, simple_loss=0.4212, pruned_loss=0.1557, over 28838.00 frames. ], tot_loss[loss=0.3796, simple_loss=0.42, pruned_loss=0.1696, over 5631014.52 frames. ], libri_tot_loss[loss=0.3484, simple_loss=0.4026, pruned_loss=0.1471, over 5726332.07 frames. ], giga_tot_loss[loss=0.3802, simple_loss=0.4205, pruned_loss=0.1699, over 5613874.91 frames. ], batch size: 243, lr: 7.70e-03, grad_scale: 4.0
+2023-03-02 06:25:29,670 INFO [optim.py:369] (1/2) 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 06:25:34,891 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:25:37,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:26:13,126 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:26:21,721 INFO [train.py:968] (1/2) Epoch 4, batch 24100, giga_loss[loss=0.4609, simple_loss=0.4652, pruned_loss=0.2283, over 26487.00 frames. ], tot_loss[loss=0.378, simple_loss=0.4201, pruned_loss=0.168, over 5625131.84 frames. ], libri_tot_loss[loss=0.3485, simple_loss=0.4027, pruned_loss=0.1472, over 5724601.94 frames. ], giga_tot_loss[loss=0.3787, simple_loss=0.4206, pruned_loss=0.1684, over 5612039.78 frames. ], batch size: 555, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:26:27,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 24150, giga_loss[loss=0.3531, simple_loss=0.4134, pruned_loss=0.1464, over 29019.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.4209, pruned_loss=0.1682, over 5628001.41 frames. ], libri_tot_loss[loss=0.3491, simple_loss=0.4028, pruned_loss=0.1477, over 5728690.69 frames. ], giga_tot_loss[loss=0.3796, simple_loss=0.4217, pruned_loss=0.1688, over 5611404.25 frames. ], batch size: 136, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:27:12,586 INFO [optim.py:369] (1/2) 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 06:27:23,814 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:28:00,726 INFO [train.py:968] (1/2) Epoch 4, batch 24200, giga_loss[loss=0.2842, simple_loss=0.3611, pruned_loss=0.1037, over 28898.00 frames. ], tot_loss[loss=0.3766, simple_loss=0.4198, pruned_loss=0.1667, over 5635142.75 frames. ], libri_tot_loss[loss=0.3503, simple_loss=0.4036, pruned_loss=0.1484, over 5729909.73 frames. ], giga_tot_loss[loss=0.3766, simple_loss=0.4198, pruned_loss=0.1667, over 5619574.57 frames. ], batch size: 145, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:28:10,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5302, 2.2094, 1.5724, 0.6059], device='cuda:1'), covar=tensor([0.2306, 0.1191, 0.1802, 0.2707], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.1270, 0.1338, 0.1133], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 06:28:17,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-02 06:28:53,596 INFO [train.py:968] (1/2) Epoch 4, batch 24250, giga_loss[loss=0.3725, simple_loss=0.4216, pruned_loss=0.1617, over 28231.00 frames. ], tot_loss[loss=0.3739, simple_loss=0.4182, pruned_loss=0.1647, over 5625887.18 frames. ], libri_tot_loss[loss=0.3506, simple_loss=0.4037, pruned_loss=0.1487, over 5722686.46 frames. ], giga_tot_loss[loss=0.3739, simple_loss=0.4184, pruned_loss=0.1647, over 5617961.03 frames. ], batch size: 77, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:28:54,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4375, 1.8620, 1.7315, 1.6566], device='cuda:1'), covar=tensor([0.1561, 0.1800, 0.1171, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0755, 0.0752, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 06:28:54,475 INFO [optim.py:369] (1/2) 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 06:29:00,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:29:06,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0799, 4.8151, 4.7305, 2.1755], device='cuda:1'), covar=tensor([0.0361, 0.0394, 0.0691, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0698, 0.0806, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:29:06,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 06:29:42,203 INFO [train.py:968] (1/2) Epoch 4, batch 24300, giga_loss[loss=0.3264, simple_loss=0.3828, pruned_loss=0.135, over 28296.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4145, pruned_loss=0.1611, over 5633332.95 frames. ], libri_tot_loss[loss=0.3501, simple_loss=0.4031, pruned_loss=0.1486, over 5727948.02 frames. ], giga_tot_loss[loss=0.3693, simple_loss=0.4155, pruned_loss=0.1616, over 5620083.07 frames. ], batch size: 368, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:29:49,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5625, 1.8244, 1.7835, 1.7214], device='cuda:1'), covar=tensor([0.1311, 0.1643, 0.1041, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0758, 0.0755, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 06:30:28,507 INFO [train.py:968] (1/2) Epoch 4, batch 24350, giga_loss[loss=0.3665, simple_loss=0.407, pruned_loss=0.163, over 28598.00 frames. ], tot_loss[loss=0.3631, simple_loss=0.4108, pruned_loss=0.1577, over 5641584.07 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4028, pruned_loss=0.1485, over 5732105.37 frames. ], giga_tot_loss[loss=0.3645, simple_loss=0.4122, pruned_loss=0.1584, over 5624880.25 frames. ], batch size: 307, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:30:29,056 INFO [optim.py:369] (1/2) 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 06:30:34,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2864, 1.5577, 1.2803, 1.6199], device='cuda:1'), covar=tensor([0.2327, 0.2154, 0.2096, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.1083, 0.0867, 0.0970, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 06:31:14,041 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:31:14,334 INFO [train.py:968] (1/2) Epoch 4, batch 24400, giga_loss[loss=0.4043, simple_loss=0.4379, pruned_loss=0.1854, over 28870.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4088, pruned_loss=0.1565, over 5647406.94 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4027, pruned_loss=0.1486, over 5736300.39 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4103, pruned_loss=0.1573, over 5627858.39 frames. ], batch size: 284, lr: 7.69e-03, grad_scale: 8.0
+2023-03-02 06:31:17,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:31:32,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8713, 1.7957, 1.2480, 1.5623], device='cuda:1'), covar=tensor([0.0591, 0.0566, 0.0975, 0.0895], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0466, 0.0519, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:31:44,728 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:31:49,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6669, 1.5528, 1.6027, 1.4350], device='cuda:1'), covar=tensor([0.1010, 0.1604, 0.1580, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0758, 0.0643, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 06:31:49,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3807, 1.4804, 1.2512, 1.3653], device='cuda:1'), covar=tensor([0.2042, 0.1993, 0.1930, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.1073, 0.0857, 0.0959, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 06:32:02,622 INFO [train.py:968] (1/2) Epoch 4, batch 24450, giga_loss[loss=0.4669, simple_loss=0.4481, pruned_loss=0.2429, over 23241.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4094, pruned_loss=0.1573, over 5643543.89 frames. ], libri_tot_loss[loss=0.3506, simple_loss=0.4032, pruned_loss=0.149, over 5739774.33 frames. ], giga_tot_loss[loss=0.363, simple_loss=0.4103, pruned_loss=0.1578, over 5622116.30 frames. ], batch size: 705, lr: 7.69e-03, grad_scale: 4.0
+2023-03-02 06:32:03,966 INFO [optim.py:369] (1/2) 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 06:32:24,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-02 06:32:35,862 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:32:55,177 INFO [train.py:968] (1/2) Epoch 4, batch 24500, libri_loss[loss=0.2851, simple_loss=0.3436, pruned_loss=0.1133, over 29333.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.408, pruned_loss=0.1554, over 5644111.57 frames. ], libri_tot_loss[loss=0.3505, simple_loss=0.403, pruned_loss=0.149, over 5733755.13 frames. ], giga_tot_loss[loss=0.3606, simple_loss=0.409, pruned_loss=0.1561, over 5629117.59 frames. ], batch size: 71, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:33:28,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=160678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:33:41,979 INFO [train.py:968] (1/2) Epoch 4, batch 24550, libri_loss[loss=0.289, simple_loss=0.3463, pruned_loss=0.1159, over 29685.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.404, pruned_loss=0.1517, over 5644749.21 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4023, pruned_loss=0.1489, over 5717581.93 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4057, pruned_loss=0.1525, over 5643803.18 frames. ], batch size: 73, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:33:45,084 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=160700.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:34:13,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5225, 4.3237, 4.1305, 1.9965], device='cuda:1'), covar=tensor([0.0514, 0.0466, 0.0964, 0.2102], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0715, 0.0830, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:34:31,259 INFO [train.py:968] (1/2) Epoch 4, batch 24600, giga_loss[loss=0.3317, simple_loss=0.4082, pruned_loss=0.1276, over 29015.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4033, pruned_loss=0.1485, over 5654091.61 frames. ], libri_tot_loss[loss=0.3491, simple_loss=0.4015, pruned_loss=0.1484, over 5723472.95 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4053, pruned_loss=0.1496, over 5646425.62 frames. ], batch size: 136, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:34:53,042 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:34:54,764 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 24650, giga_loss[loss=0.3279, simple_loss=0.3843, pruned_loss=0.1358, over 28382.00 frames. ], tot_loss[loss=0.3513, simple_loss=0.4052, pruned_loss=0.1487, over 5667177.84 frames. ], libri_tot_loss[loss=0.3485, simple_loss=0.4007, pruned_loss=0.1481, over 5728212.19 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4078, pruned_loss=0.1499, over 5653689.89 frames. ], batch size: 65, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:35:20,829 INFO [optim.py:369] (1/2) 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 06:35:23,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:35:45,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=160821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:35:47,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=160824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:36:06,195 INFO [train.py:968] (1/2) Epoch 4, batch 24700, giga_loss[loss=0.3313, simple_loss=0.3913, pruned_loss=0.1357, over 28952.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4075, pruned_loss=0.1516, over 5659199.05 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.4014, pruned_loss=0.1488, over 5723117.42 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4091, pruned_loss=0.1519, over 5650704.93 frames. ], batch size: 164, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:36:17,944 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=160853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:36:32,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 06:36:51,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4408, 1.5156, 1.1048, 1.0612], device='cuda:1'), covar=tensor([0.0863, 0.0790, 0.0667, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.1333, 0.1077, 0.1092, 0.1172], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 06:36:52,896 INFO [train.py:968] (1/2) Epoch 4, batch 24750, giga_loss[loss=0.292, simple_loss=0.3668, pruned_loss=0.1086, over 28436.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.407, pruned_loss=0.1515, over 5668021.27 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4014, pruned_loss=0.1493, over 5716408.42 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4084, pruned_loss=0.1514, over 5665306.18 frames. ], batch size: 60, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:36:54,801 INFO [optim.py:369] (1/2) 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 06:37:06,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-02 06:37:35,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 06:37:44,047 INFO [train.py:968] (1/2) Epoch 4, batch 24800, libri_loss[loss=0.3761, simple_loss=0.4271, pruned_loss=0.1625, over 29684.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4052, pruned_loss=0.1507, over 5678938.09 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4013, pruned_loss=0.1492, over 5719391.90 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4065, pruned_loss=0.1508, over 5673326.68 frames. ], batch size: 91, lr: 7.68e-03, grad_scale: 8.0
+2023-03-02 06:37:58,250 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-02 06:38:23,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4609, 4.2300, 4.1426, 1.7463], device='cuda:1'), covar=tensor([0.0412, 0.0420, 0.0749, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0702, 0.0810, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:38:27,599 INFO [train.py:968] (1/2) Epoch 4, batch 24850, giga_loss[loss=0.3239, simple_loss=0.3841, pruned_loss=0.1318, over 28940.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4035, pruned_loss=0.1505, over 5675672.14 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4015, pruned_loss=0.1492, over 5717494.81 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4044, pruned_loss=0.1505, over 5672474.10 frames. ], batch size: 145, lr: 7.68e-03, grad_scale: 4.0
+2023-03-02 06:38:30,156 INFO [optim.py:369] (1/2) 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 06:39:10,361 INFO [train.py:968] (1/2) Epoch 4, batch 24900, giga_loss[loss=0.3642, simple_loss=0.3884, pruned_loss=0.17, over 23595.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4023, pruned_loss=0.1494, over 5669706.14 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.4011, pruned_loss=0.1489, over 5715221.57 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4035, pruned_loss=0.1497, over 5667709.16 frames. ], batch size: 705, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:39:20,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1660, 1.2095, 1.0318, 1.1188], device='cuda:1'), covar=tensor([0.0637, 0.0496, 0.0953, 0.0709], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0465, 0.0514, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:39:37,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 24950, giga_loss[loss=0.288, simple_loss=0.3617, pruned_loss=0.1071, over 28349.00 frames. ], tot_loss[loss=0.347, simple_loss=0.4011, pruned_loss=0.1464, over 5676136.82 frames. ], libri_tot_loss[loss=0.3498, simple_loss=0.4013, pruned_loss=0.1491, over 5711879.25 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.4019, pruned_loss=0.1465, over 5676905.37 frames. ], batch size: 71, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:40:00,065 INFO [optim.py:369] (1/2) 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 06:40:47,224 INFO [train.py:968] (1/2) Epoch 4, batch 25000, giga_loss[loss=0.3847, simple_loss=0.4368, pruned_loss=0.1663, over 28945.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.4024, pruned_loss=0.1479, over 5654230.10 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4013, pruned_loss=0.1493, over 5694095.90 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.403, pruned_loss=0.1478, over 5671027.11 frames. ], batch size: 174, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:41:39,067 INFO [train.py:968] (1/2) Epoch 4, batch 25050, giga_loss[loss=0.3532, simple_loss=0.4101, pruned_loss=0.1481, over 28874.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4011, pruned_loss=0.147, over 5662828.34 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4014, pruned_loss=0.1493, over 5695136.42 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.4015, pruned_loss=0.1469, over 5674848.98 frames. ], batch size: 186, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:41:41,025 INFO [optim.py:369] (1/2) 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 06:42:03,429 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161218.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:42:07,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161221.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 06:42:29,804 INFO [train.py:968] (1/2) Epoch 4, batch 25100, libri_loss[loss=0.3466, simple_loss=0.3907, pruned_loss=0.1512, over 29683.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.4006, pruned_loss=0.1475, over 5661484.46 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4013, pruned_loss=0.1493, over 5697211.69 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.4009, pruned_loss=0.1474, over 5668720.86 frames. ], batch size: 73, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:42:33,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6980, 1.6035, 1.5245, 1.5823], device='cuda:1'), covar=tensor([0.1052, 0.1901, 0.1715, 0.1486], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0757, 0.0643, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 06:42:36,707 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161250.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 06:43:16,532 INFO [train.py:968] (1/2) Epoch 4, batch 25150, giga_loss[loss=0.3438, simple_loss=0.4018, pruned_loss=0.1429, over 28977.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4006, pruned_loss=0.1483, over 5659258.20 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4013, pruned_loss=0.1493, over 5699934.23 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4008, pruned_loss=0.1481, over 5661712.82 frames. ], batch size: 145, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:43:17,663 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:43:18,900 INFO [optim.py:369] (1/2) 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 06:44:00,978 INFO [train.py:968] (1/2) Epoch 4, batch 25200, giga_loss[loss=0.3358, simple_loss=0.3954, pruned_loss=0.1382, over 28807.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.401, pruned_loss=0.1496, over 5668983.19 frames. ], libri_tot_loss[loss=0.3498, simple_loss=0.4012, pruned_loss=0.1492, over 5703720.90 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4013, pruned_loss=0.1496, over 5666714.65 frames. ], batch size: 199, lr: 7.67e-03, grad_scale: 8.0
+2023-03-02 06:44:14,252 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:44:48,104 INFO [train.py:968] (1/2) Epoch 4, batch 25250, giga_loss[loss=0.336, simple_loss=0.3799, pruned_loss=0.1461, over 28582.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3994, pruned_loss=0.1485, over 5668496.63 frames. ], libri_tot_loss[loss=0.3502, simple_loss=0.4016, pruned_loss=0.1494, over 5705791.80 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.3993, pruned_loss=0.1484, over 5664260.96 frames. ], batch size: 85, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:44:51,592 INFO [optim.py:369] (1/2) 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 06:44:57,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2274, 1.4133, 1.0924, 0.7061], device='cuda:1'), covar=tensor([0.0962, 0.0738, 0.0566, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.1341, 0.1084, 0.1089, 0.1178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 06:45:30,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2726, 1.3217, 0.8981, 1.0318], device='cuda:1'), covar=tensor([0.0865, 0.0682, 0.0616, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.1357, 0.1093, 0.1102, 0.1190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 06:45:35,091 INFO [train.py:968] (1/2) Epoch 4, batch 25300, giga_loss[loss=0.3555, simple_loss=0.4018, pruned_loss=0.1546, over 28233.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.3994, pruned_loss=0.1489, over 5668357.63 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.4019, pruned_loss=0.1494, over 5710539.59 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.399, pruned_loss=0.1487, over 5660018.50 frames. ], batch size: 368, lr: 7.67e-03, grad_scale: 4.0
+2023-03-02 06:46:21,376 INFO [train.py:968] (1/2) Epoch 4, batch 25350, giga_loss[loss=0.31, simple_loss=0.3832, pruned_loss=0.1184, over 28958.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4001, pruned_loss=0.1494, over 5666112.99 frames. ], libri_tot_loss[loss=0.351, simple_loss=0.4023, pruned_loss=0.1499, over 5707366.35 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.3993, pruned_loss=0.1489, over 5661061.11 frames. ], batch size: 164, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:46:24,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6422, 2.9087, 1.6385, 1.7390], device='cuda:1'), covar=tensor([0.0757, 0.0495, 0.0764, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0468, 0.0307, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 06:46:24,997 INFO [optim.py:369] (1/2) 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 06:47:06,406 INFO [train.py:968] (1/2) Epoch 4, batch 25400, giga_loss[loss=0.3267, simple_loss=0.396, pruned_loss=0.1287, over 28973.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4008, pruned_loss=0.149, over 5664440.69 frames. ], libri_tot_loss[loss=0.3512, simple_loss=0.4023, pruned_loss=0.15, over 5703942.00 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4002, pruned_loss=0.1484, over 5661342.00 frames. ], batch size: 164, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:47:19,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7539, 3.5820, 3.4353, 1.5928], device='cuda:1'), covar=tensor([0.0571, 0.0522, 0.0867, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0711, 0.0824, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:47:50,863 INFO [train.py:968] (1/2) Epoch 4, batch 25450, giga_loss[loss=0.3101, simple_loss=0.3839, pruned_loss=0.1182, over 28683.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.4005, pruned_loss=0.1479, over 5661648.51 frames. ], libri_tot_loss[loss=0.3513, simple_loss=0.4024, pruned_loss=0.1501, over 5696348.27 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.3999, pruned_loss=0.1473, over 5665238.28 frames. ], batch size: 262, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:47:57,314 INFO [optim.py:369] (1/2) 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 06:48:06,811 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:48:20,688 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 06:48:21,528 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:48:39,666 INFO [train.py:968] (1/2) Epoch 4, batch 25500, giga_loss[loss=0.3897, simple_loss=0.4366, pruned_loss=0.1714, over 28600.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4009, pruned_loss=0.1481, over 5655453.36 frames. ], libri_tot_loss[loss=0.3518, simple_loss=0.4028, pruned_loss=0.1504, over 5698461.51 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.4, pruned_loss=0.1474, over 5655929.76 frames. ], batch size: 336, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:49:03,093 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:49:24,549 INFO [train.py:968] (1/2) Epoch 4, batch 25550, giga_loss[loss=0.3479, simple_loss=0.405, pruned_loss=0.1454, over 29028.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4033, pruned_loss=0.1508, over 5654711.69 frames. ], libri_tot_loss[loss=0.3517, simple_loss=0.4026, pruned_loss=0.1504, over 5700623.39 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4027, pruned_loss=0.1502, over 5652523.20 frames. ], batch size: 128, lr: 7.66e-03, grad_scale: 2.0
+2023-03-02 06:49:31,285 INFO [optim.py:369] (1/2) 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 06:49:44,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-02 06:49:48,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4211, 1.9937, 1.2148, 1.1322], device='cuda:1'), covar=tensor([0.1146, 0.0843, 0.0777, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.1344, 0.1089, 0.1088, 0.1175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 06:49:48,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-02 06:50:04,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:50:05,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3306, 1.3615, 1.0174, 1.0061], device='cuda:1'), covar=tensor([0.0593, 0.0556, 0.0452, 0.0579], device='cuda:1'), in_proj_covar=tensor([0.1347, 0.1092, 0.1090, 0.1173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 06:50:14,148 INFO [train.py:968] (1/2) Epoch 4, batch 25600, giga_loss[loss=0.3313, simple_loss=0.3867, pruned_loss=0.138, over 28938.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4058, pruned_loss=0.154, over 5647621.37 frames. ], libri_tot_loss[loss=0.352, simple_loss=0.4028, pruned_loss=0.1506, over 5691796.01 frames. ], giga_tot_loss[loss=0.356, simple_loss=0.4052, pruned_loss=0.1534, over 5653805.98 frames. ], batch size: 136, lr: 7.66e-03, grad_scale: 4.0
+2023-03-02 06:51:05,429 INFO [train.py:968] (1/2) Epoch 4, batch 25650, giga_loss[loss=0.4298, simple_loss=0.4475, pruned_loss=0.2061, over 27978.00 frames. ], tot_loss[loss=0.3588, simple_loss=0.4065, pruned_loss=0.1556, over 5636487.75 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.403, pruned_loss=0.1507, over 5672741.49 frames. ], giga_tot_loss[loss=0.3579, simple_loss=0.4058, pruned_loss=0.155, over 5657831.00 frames. ], batch size: 412, lr: 7.66e-03, grad_scale: 4.0
+2023-03-02 06:51:08,083 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:51:24,321 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:51:26,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:51:28,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-02 06:51:56,084 INFO [train.py:968] (1/2) Epoch 4, batch 25700, libri_loss[loss=0.4272, simple_loss=0.4641, pruned_loss=0.1951, over 25802.00 frames. ], tot_loss[loss=0.3628, simple_loss=0.4087, pruned_loss=0.1584, over 5628789.58 frames. ], libri_tot_loss[loss=0.3526, simple_loss=0.4033, pruned_loss=0.1509, over 5676049.03 frames. ], giga_tot_loss[loss=0.3618, simple_loss=0.408, pruned_loss=0.1578, over 5642432.35 frames. ], batch size: 136, lr: 7.66e-03, grad_scale: 4.0
+2023-03-02 06:51:56,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=161844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:04,959 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=161853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:25,393 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=161876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:27,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=161879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:52:40,888 INFO [train.py:968] (1/2) Epoch 4, batch 25750, giga_loss[loss=0.3485, simple_loss=0.4056, pruned_loss=0.1458, over 28893.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.4077, pruned_loss=0.1576, over 5644119.49 frames. ], libri_tot_loss[loss=0.3526, simple_loss=0.4034, pruned_loss=0.1509, over 5680894.43 frames. ], giga_tot_loss[loss=0.3609, simple_loss=0.4072, pruned_loss=0.1573, over 5649765.82 frames. ], batch size: 174, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:52:45,666 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 4, batch 25800, giga_loss[loss=0.3201, simple_loss=0.3876, pruned_loss=0.1263, over 28581.00 frames. ], tot_loss[loss=0.3583, simple_loss=0.406, pruned_loss=0.1554, over 5653960.16 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4038, pruned_loss=0.1512, over 5688402.16 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4053, pruned_loss=0.1551, over 5650639.79 frames. ], batch size: 71, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:53:54,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-02 06:54:02,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:54:09,596 INFO [train.py:968] (1/2) Epoch 4, batch 25850, giga_loss[loss=0.2997, simple_loss=0.3751, pruned_loss=0.1122, over 28970.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.4035, pruned_loss=0.1515, over 5666751.30 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4038, pruned_loss=0.1512, over 5689207.57 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4029, pruned_loss=0.1513, over 5663230.82 frames. ], batch size: 136, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:54:13,379 INFO [optim.py:369] (1/2) 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 06:54:15,937 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=161999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:54:57,789 INFO [train.py:968] (1/2) Epoch 4, batch 25900, giga_loss[loss=0.3293, simple_loss=0.3879, pruned_loss=0.1354, over 28682.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4002, pruned_loss=0.1496, over 5658674.61 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.404, pruned_loss=0.1514, over 5692215.48 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.3995, pruned_loss=0.1492, over 5652589.79 frames. ], batch size: 92, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:55:43,812 INFO [train.py:968] (1/2) Epoch 4, batch 25950, giga_loss[loss=0.3647, simple_loss=0.4036, pruned_loss=0.1629, over 28305.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3977, pruned_loss=0.1484, over 5673505.09 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.4038, pruned_loss=0.1514, over 5697745.34 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3972, pruned_loss=0.148, over 5663050.75 frames. ], batch size: 368, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:55:50,145 INFO [optim.py:369] (1/2) 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 06:55:51,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3071, 1.9667, 1.4522, 0.5779], device='cuda:1'), covar=tensor([0.1963, 0.1008, 0.1540, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1330, 0.1265, 0.1328, 0.1118], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 06:56:18,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:22,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:36,661 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 26000, giga_loss[loss=0.4079, simple_loss=0.4227, pruned_loss=0.1966, over 23537.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3968, pruned_loss=0.1481, over 5678703.19 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.404, pruned_loss=0.1514, over 5697799.48 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3962, pruned_loss=0.1477, over 5670160.57 frames. ], batch size: 705, lr: 7.65e-03, grad_scale: 8.0
+2023-03-02 06:56:38,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:56:50,746 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:57:00,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:57:04,176 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 26050, giga_loss[loss=0.3703, simple_loss=0.4225, pruned_loss=0.159, over 28622.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.3996, pruned_loss=0.1496, over 5682471.61 frames. ], libri_tot_loss[loss=0.3541, simple_loss=0.4044, pruned_loss=0.1519, over 5701551.72 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.3984, pruned_loss=0.1487, over 5671571.05 frames. ], batch size: 336, lr: 7.65e-03, grad_scale: 8.0
+2023-03-02 06:57:24,605 INFO [optim.py:369] (1/2) 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 06:57:32,745 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:57:43,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2349, 1.3167, 0.8408, 0.9129], device='cuda:1'), covar=tensor([0.0708, 0.0594, 0.0595, 0.0658], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1079, 0.1086, 0.1153], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 06:57:50,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:58:02,531 INFO [train.py:968] (1/2) Epoch 4, batch 26100, giga_loss[loss=0.3731, simple_loss=0.4307, pruned_loss=0.1578, over 28829.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4017, pruned_loss=0.1489, over 5691956.07 frames. ], libri_tot_loss[loss=0.3527, simple_loss=0.4032, pruned_loss=0.1511, over 5706229.96 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4018, pruned_loss=0.1488, over 5678443.64 frames. ], batch size: 119, lr: 7.65e-03, grad_scale: 8.0
+2023-03-02 06:58:20,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4420, 3.2125, 3.1475, 1.9306], device='cuda:1'), covar=tensor([0.0590, 0.0732, 0.1076, 0.1601], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0715, 0.0823, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 06:58:26,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5652, 3.2324, 1.4527, 1.4352], device='cuda:1'), covar=tensor([0.0808, 0.0372, 0.0966, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0468, 0.0310, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 06:58:42,644 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-02 06:58:47,663 INFO [train.py:968] (1/2) Epoch 4, batch 26150, giga_loss[loss=0.3322, simple_loss=0.3914, pruned_loss=0.1365, over 28823.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4034, pruned_loss=0.1481, over 5678949.74 frames. ], libri_tot_loss[loss=0.353, simple_loss=0.4031, pruned_loss=0.1515, over 5700472.31 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4035, pruned_loss=0.1477, over 5671920.95 frames. ], batch size: 99, lr: 7.65e-03, grad_scale: 4.0
+2023-03-02 06:58:52,357 INFO [optim.py:369] (1/2) 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 06:59:05,913 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:59:07,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:59:21,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4729, 1.7877, 1.7302, 1.6114], device='cuda:1'), covar=tensor([0.1361, 0.1731, 0.1036, 0.1056], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0769, 0.0765, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 06:59:29,177 INFO [train.py:968] (1/2) Epoch 4, batch 26200, giga_loss[loss=0.3583, simple_loss=0.4087, pruned_loss=0.1539, over 28844.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.405, pruned_loss=0.149, over 5690800.77 frames. ], libri_tot_loss[loss=0.3525, simple_loss=0.4026, pruned_loss=0.1512, over 5708857.43 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4057, pruned_loss=0.1487, over 5676652.10 frames. ], batch size: 86, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 06:59:30,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-02 06:59:31,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 06:59:47,395 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 06:59:58,450 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:00:00,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:00:18,972 INFO [train.py:968] (1/2) Epoch 4, batch 26250, giga_loss[loss=0.3563, simple_loss=0.4092, pruned_loss=0.1517, over 28713.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4077, pruned_loss=0.1517, over 5675731.33 frames. ], libri_tot_loss[loss=0.3526, simple_loss=0.4026, pruned_loss=0.1513, over 5697876.44 frames. ], giga_tot_loss[loss=0.3555, simple_loss=0.4082, pruned_loss=0.1514, over 5673132.18 frames. ], batch size: 242, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:00:24,338 INFO [optim.py:369] (1/2) 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 07:00:28,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:01:05,268 INFO [train.py:968] (1/2) Epoch 4, batch 26300, giga_loss[loss=0.4019, simple_loss=0.434, pruned_loss=0.1849, over 28682.00 frames. ], tot_loss[loss=0.357, simple_loss=0.4082, pruned_loss=0.1529, over 5678324.01 frames. ], libri_tot_loss[loss=0.3525, simple_loss=0.4025, pruned_loss=0.1513, over 5701120.80 frames. ], giga_tot_loss[loss=0.3571, simple_loss=0.4089, pruned_loss=0.1527, over 5673087.39 frames. ], batch size: 262, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:01:24,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7952, 1.9019, 1.3280, 1.1206], device='cuda:1'), covar=tensor([0.0890, 0.0775, 0.0702, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.1327, 0.1085, 0.1086, 0.1168], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 07:01:32,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:01:56,509 INFO [train.py:968] (1/2) Epoch 4, batch 26350, giga_loss[loss=0.4278, simple_loss=0.4534, pruned_loss=0.2012, over 28707.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4083, pruned_loss=0.154, over 5681619.65 frames. ], libri_tot_loss[loss=0.3524, simple_loss=0.4024, pruned_loss=0.1512, over 5701416.16 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4089, pruned_loss=0.1538, over 5676985.96 frames. ], batch size: 284, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:02:04,202 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 26400, giga_loss[loss=0.4102, simple_loss=0.4379, pruned_loss=0.1913, over 28258.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4057, pruned_loss=0.1526, over 5685924.05 frames. ], libri_tot_loss[loss=0.3525, simple_loss=0.4025, pruned_loss=0.1513, over 5701621.81 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4062, pruned_loss=0.1526, over 5681933.68 frames. ], batch size: 368, lr: 7.64e-03, grad_scale: 8.0
+2023-03-02 07:02:58,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5146, 2.0084, 1.6931, 1.8336], device='cuda:1'), covar=tensor([0.0519, 0.0661, 0.0817, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0461, 0.0509, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:03:07,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1597, 3.9481, 3.8522, 1.6749], device='cuda:1'), covar=tensor([0.0442, 0.0448, 0.0751, 0.2130], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0723, 0.0826, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:03:25,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:03:38,192 INFO [train.py:968] (1/2) Epoch 4, batch 26450, giga_loss[loss=0.3006, simple_loss=0.3662, pruned_loss=0.1175, over 28757.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4031, pruned_loss=0.1512, over 5690865.42 frames. ], libri_tot_loss[loss=0.3527, simple_loss=0.4027, pruned_loss=0.1513, over 5703337.12 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4033, pruned_loss=0.1511, over 5686111.93 frames. ], batch size: 119, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:03:43,991 INFO [optim.py:369] (1/2) 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 07:03:46,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-02 07:04:06,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:04:24,504 INFO [train.py:968] (1/2) Epoch 4, batch 26500, giga_loss[loss=0.3659, simple_loss=0.4164, pruned_loss=0.1577, over 28935.00 frames. ], tot_loss[loss=0.3536, simple_loss=0.4035, pruned_loss=0.1519, over 5684334.95 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.4024, pruned_loss=0.151, over 5707682.84 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4039, pruned_loss=0.1521, over 5676475.56 frames. ], batch size: 227, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:04:35,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9379, 1.7414, 1.6414, 1.7028], device='cuda:1'), covar=tensor([0.0951, 0.1871, 0.1471, 0.1411], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0761, 0.0642, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 07:05:06,632 INFO [train.py:968] (1/2) Epoch 4, batch 26550, giga_loss[loss=0.3496, simple_loss=0.398, pruned_loss=0.1506, over 28913.00 frames. ], tot_loss[loss=0.3557, simple_loss=0.4045, pruned_loss=0.1534, over 5678312.38 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.403, pruned_loss=0.152, over 5700696.81 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.4045, pruned_loss=0.1527, over 5677500.44 frames. ], batch size: 227, lr: 7.64e-03, grad_scale: 4.0
+2023-03-02 07:05:06,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4939, 1.8693, 1.7447, 1.6837], device='cuda:1'), covar=tensor([0.1343, 0.1700, 0.1047, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0773, 0.0765, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 07:05:14,024 INFO [optim.py:369] (1/2) 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 07:05:36,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:05:40,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:05:52,017 INFO [train.py:968] (1/2) Epoch 4, batch 26600, giga_loss[loss=0.3, simple_loss=0.3701, pruned_loss=0.115, over 28895.00 frames. ], tot_loss[loss=0.355, simple_loss=0.403, pruned_loss=0.1534, over 5670174.29 frames. ], libri_tot_loss[loss=0.353, simple_loss=0.4027, pruned_loss=0.1517, over 5704384.80 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4033, pruned_loss=0.1532, over 5665901.05 frames. ], batch size: 174, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:06:06,573 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=162758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:06:09,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9392, 1.7427, 1.2272, 1.6182], device='cuda:1'), covar=tensor([0.0531, 0.0585, 0.0980, 0.0784], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0461, 0.0513, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:06:38,435 INFO [train.py:968] (1/2) Epoch 4, batch 26650, giga_loss[loss=0.5371, simple_loss=0.5109, pruned_loss=0.2817, over 26658.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4015, pruned_loss=0.1527, over 5654910.57 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4029, pruned_loss=0.152, over 5702176.30 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4015, pruned_loss=0.1523, over 5651954.72 frames. ], batch size: 555, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:06:44,122 INFO [optim.py:369] (1/2) 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 07:07:22,064 INFO [train.py:968] (1/2) Epoch 4, batch 26700, giga_loss[loss=0.3194, simple_loss=0.3901, pruned_loss=0.1244, over 28899.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4018, pruned_loss=0.1516, over 5665050.72 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4028, pruned_loss=0.152, over 5706140.42 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4019, pruned_loss=0.1512, over 5658158.00 frames. ], batch size: 174, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:07:27,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:07:43,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2319, 1.1464, 1.0721, 0.9824], device='cuda:1'), covar=tensor([0.0620, 0.0531, 0.0982, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0461, 0.0513, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:08:07,396 INFO [train.py:968] (1/2) Epoch 4, batch 26750, giga_loss[loss=0.3064, simple_loss=0.3751, pruned_loss=0.1188, over 28851.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.404, pruned_loss=0.1521, over 5669604.19 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.403, pruned_loss=0.152, over 5707985.12 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4038, pruned_loss=0.1518, over 5661176.98 frames. ], batch size: 174, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:08:13,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 07:08:13,278 INFO [optim.py:369] (1/2) 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 07:08:19,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=162904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:08:34,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5097, 1.6757, 1.2481, 0.8462], device='cuda:1'), covar=tensor([0.0916, 0.0724, 0.0610, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.1322, 0.1082, 0.1074, 0.1169], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 07:08:55,782 INFO [train.py:968] (1/2) Epoch 4, batch 26800, giga_loss[loss=0.3146, simple_loss=0.3787, pruned_loss=0.1253, over 28914.00 frames. ], tot_loss[loss=0.3559, simple_loss=0.4049, pruned_loss=0.1535, over 5669847.86 frames. ], libri_tot_loss[loss=0.3539, simple_loss=0.4034, pruned_loss=0.1522, over 5712865.25 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4044, pruned_loss=0.1531, over 5657648.23 frames. ], batch size: 174, lr: 7.63e-03, grad_scale: 8.0
+2023-03-02 07:09:34,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=162988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:09:39,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=162991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:09:40,289 INFO [train.py:968] (1/2) Epoch 4, batch 26850, giga_loss[loss=0.3513, simple_loss=0.4196, pruned_loss=0.1415, over 28887.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4042, pruned_loss=0.1505, over 5678428.99 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.4028, pruned_loss=0.1519, over 5708312.96 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4044, pruned_loss=0.1505, over 5671553.78 frames. ], batch size: 174, lr: 7.63e-03, grad_scale: 8.0
+2023-03-02 07:09:45,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=162999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:09:46,365 INFO [optim.py:369] (1/2) 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 07:09:48,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-02 07:09:50,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-02 07:10:04,717 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:10:25,306 INFO [train.py:968] (1/2) Epoch 4, batch 26900, giga_loss[loss=0.3307, simple_loss=0.4051, pruned_loss=0.1281, over 29064.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.4041, pruned_loss=0.1478, over 5674922.74 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4029, pruned_loss=0.1522, over 5700634.22 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4042, pruned_loss=0.1475, over 5675082.81 frames. ], batch size: 155, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:10:45,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6019, 1.9114, 1.5454, 1.0833], device='cuda:1'), covar=tensor([0.1422, 0.0803, 0.0597, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.1079, 0.1067, 0.1173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 07:11:08,591 INFO [train.py:968] (1/2) Epoch 4, batch 26950, giga_loss[loss=0.387, simple_loss=0.4332, pruned_loss=0.1703, over 28610.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4066, pruned_loss=0.1489, over 5674205.80 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4027, pruned_loss=0.1522, over 5696778.42 frames. ], giga_tot_loss[loss=0.352, simple_loss=0.407, pruned_loss=0.1485, over 5677412.12 frames. ], batch size: 307, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:11:15,917 INFO [optim.py:369] (1/2) 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 07:11:54,012 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:11:54,287 INFO [train.py:968] (1/2) Epoch 4, batch 27000, giga_loss[loss=0.3673, simple_loss=0.4117, pruned_loss=0.1614, over 29051.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4095, pruned_loss=0.1516, over 5677716.31 frames. ], libri_tot_loss[loss=0.3541, simple_loss=0.4031, pruned_loss=0.1526, over 5701071.18 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4096, pruned_loss=0.1509, over 5675619.30 frames. ], batch size: 128, lr: 7.63e-03, grad_scale: 4.0
+2023-03-02 07:11:54,287 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 07:11:58,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2280, 1.4292, 1.2218, 1.2475], device='cuda:1'), covar=tensor([0.2677, 0.2292, 0.2409, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.1093, 0.0855, 0.0979, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 07:12:03,198 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 07:12:05,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:12:31,941 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 27050, giga_loss[loss=0.3562, simple_loss=0.4172, pruned_loss=0.1476, over 28786.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4125, pruned_loss=0.1552, over 5667562.11 frames. ], libri_tot_loss[loss=0.3541, simple_loss=0.4032, pruned_loss=0.1525, over 5702659.17 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.4126, pruned_loss=0.1548, over 5664057.75 frames. ], batch size: 119, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:12:59,839 INFO [optim.py:369] (1/2) 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 07:13:21,097 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:13:25,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9946, 1.7008, 1.3762, 1.4103], device='cuda:1'), covar=tensor([0.0547, 0.0635, 0.0922, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0466, 0.0516, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:13:40,893 INFO [train.py:968] (1/2) Epoch 4, batch 27100, giga_loss[loss=0.3647, simple_loss=0.4083, pruned_loss=0.1605, over 28958.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4128, pruned_loss=0.1574, over 5644324.66 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4032, pruned_loss=0.1527, over 5691552.41 frames. ], giga_tot_loss[loss=0.3635, simple_loss=0.4131, pruned_loss=0.1569, over 5650713.35 frames. ], batch size: 227, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:13:49,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-02 07:14:19,419 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 27150, giga_loss[loss=0.4594, simple_loss=0.4651, pruned_loss=0.2269, over 26632.00 frames. ], tot_loss[loss=0.3628, simple_loss=0.412, pruned_loss=0.1568, over 5647345.17 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4027, pruned_loss=0.1523, over 5696935.98 frames. ], giga_tot_loss[loss=0.3634, simple_loss=0.4129, pruned_loss=0.1569, over 5646556.58 frames. ], batch size: 555, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:14:41,483 INFO [optim.py:369] (1/2) 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 07:14:41,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8139, 2.0406, 1.5765, 1.1910], device='cuda:1'), covar=tensor([0.0922, 0.0611, 0.0490, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.1328, 0.1083, 0.1068, 0.1170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 07:15:04,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-02 07:15:15,783 INFO [train.py:968] (1/2) Epoch 4, batch 27200, giga_loss[loss=0.4712, simple_loss=0.4878, pruned_loss=0.2272, over 28661.00 frames. ], tot_loss[loss=0.3606, simple_loss=0.4111, pruned_loss=0.155, over 5655144.92 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.402, pruned_loss=0.1519, over 5699366.65 frames. ], giga_tot_loss[loss=0.3621, simple_loss=0.4129, pruned_loss=0.1556, over 5650775.36 frames. ], batch size: 262, lr: 7.62e-03, grad_scale: 8.0
+2023-03-02 07:16:02,373 INFO [train.py:968] (1/2) Epoch 4, batch 27250, giga_loss[loss=0.3107, simple_loss=0.3801, pruned_loss=0.1207, over 28720.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4093, pruned_loss=0.1519, over 5664779.77 frames. ], libri_tot_loss[loss=0.3525, simple_loss=0.4016, pruned_loss=0.1517, over 5704215.72 frames. ], giga_tot_loss[loss=0.3582, simple_loss=0.4112, pruned_loss=0.1526, over 5655626.47 frames. ], batch size: 66, lr: 7.62e-03, grad_scale: 8.0
+2023-03-02 07:16:07,794 INFO [optim.py:369] (1/2) 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 07:16:27,440 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:16:29,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:16:30,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2279, 3.4753, 1.2592, 1.4244], device='cuda:1'), covar=tensor([0.1108, 0.0463, 0.1056, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0472, 0.0307, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 07:16:37,983 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-02 07:16:47,936 INFO [train.py:968] (1/2) Epoch 4, batch 27300, giga_loss[loss=0.36, simple_loss=0.4209, pruned_loss=0.1495, over 28704.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4094, pruned_loss=0.1518, over 5667960.03 frames. ], libri_tot_loss[loss=0.353, simple_loss=0.4019, pruned_loss=0.152, over 5707951.20 frames. ], giga_tot_loss[loss=0.3576, simple_loss=0.4109, pruned_loss=0.1521, over 5656394.57 frames. ], batch size: 242, lr: 7.62e-03, grad_scale: 8.0
+2023-03-02 07:16:58,420 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:17:27,466 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:17:37,731 INFO [train.py:968] (1/2) Epoch 4, batch 27350, giga_loss[loss=0.3198, simple_loss=0.3911, pruned_loss=0.1243, over 28907.00 frames. ], tot_loss[loss=0.3574, simple_loss=0.4101, pruned_loss=0.1524, over 5673888.99 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4021, pruned_loss=0.1521, over 5711128.04 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4113, pruned_loss=0.1525, over 5661521.43 frames. ], batch size: 174, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:17:46,530 INFO [optim.py:369] (1/2) 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 07:18:16,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7023, 1.6017, 1.3365, 1.4108], device='cuda:1'), covar=tensor([0.0676, 0.0658, 0.0985, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0460, 0.0512, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:18:25,461 INFO [train.py:968] (1/2) Epoch 4, batch 27400, giga_loss[loss=0.3383, simple_loss=0.3881, pruned_loss=0.1443, over 28932.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.409, pruned_loss=0.1522, over 5681853.78 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4022, pruned_loss=0.152, over 5714473.14 frames. ], giga_tot_loss[loss=0.3574, simple_loss=0.41, pruned_loss=0.1524, over 5668413.92 frames. ], batch size: 145, lr: 7.62e-03, grad_scale: 4.0
+2023-03-02 07:19:13,321 INFO [train.py:968] (1/2) Epoch 4, batch 27450, giga_loss[loss=0.3141, simple_loss=0.376, pruned_loss=0.1261, over 28757.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4069, pruned_loss=0.1527, over 5663593.92 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4021, pruned_loss=0.1519, over 5716224.98 frames. ], giga_tot_loss[loss=0.3569, simple_loss=0.4079, pruned_loss=0.153, over 5650299.42 frames. ], batch size: 119, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:19:15,122 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:19:54,379 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=163633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:20:04,005 INFO [train.py:968] (1/2) Epoch 4, batch 27500, giga_loss[loss=0.364, simple_loss=0.4086, pruned_loss=0.1597, over 28240.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.4052, pruned_loss=0.1526, over 5644571.37 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.4021, pruned_loss=0.1522, over 5707352.27 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4061, pruned_loss=0.1526, over 5640029.29 frames. ], batch size: 368, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:20:20,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-03-02 07:20:50,547 INFO [train.py:968] (1/2) Epoch 4, batch 27550, libri_loss[loss=0.2822, simple_loss=0.3439, pruned_loss=0.1103, over 27274.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4038, pruned_loss=0.1524, over 5645177.93 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4025, pruned_loss=0.1524, over 5699090.83 frames. ], giga_tot_loss[loss=0.3543, simple_loss=0.4042, pruned_loss=0.1522, over 5648019.86 frames. ], batch size: 60, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:20:58,201 INFO [optim.py:369] (1/2) 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 07:21:31,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=163738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:21:33,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=163741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:21:35,672 INFO [train.py:968] (1/2) Epoch 4, batch 27600, giga_loss[loss=0.3589, simple_loss=0.3885, pruned_loss=0.1646, over 23619.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.403, pruned_loss=0.1522, over 5635268.66 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.4026, pruned_loss=0.1525, over 5692620.58 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4033, pruned_loss=0.152, over 5641948.51 frames. ], batch size: 705, lr: 7.61e-03, grad_scale: 8.0
+2023-03-02 07:22:00,041 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=163770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:22:19,566 INFO [train.py:968] (1/2) Epoch 4, batch 27650, giga_loss[loss=0.2952, simple_loss=0.3682, pruned_loss=0.1111, over 29021.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.4007, pruned_loss=0.1498, over 5641666.04 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.403, pruned_loss=0.1528, over 5687734.14 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4006, pruned_loss=0.1494, over 5649402.30 frames. ], batch size: 128, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:22:27,915 INFO [optim.py:369] (1/2) 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 07:23:04,081 INFO [train.py:968] (1/2) Epoch 4, batch 27700, giga_loss[loss=0.3292, simple_loss=0.3897, pruned_loss=0.1344, over 28246.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3962, pruned_loss=0.1442, over 5656563.19 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4026, pruned_loss=0.1524, over 5694232.77 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3964, pruned_loss=0.144, over 5655961.25 frames. ], batch size: 368, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:23:20,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=163858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:23:58,194 INFO [train.py:968] (1/2) Epoch 4, batch 27750, giga_loss[loss=0.3642, simple_loss=0.4099, pruned_loss=0.1592, over 29048.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.395, pruned_loss=0.1427, over 5655907.93 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4024, pruned_loss=0.1523, over 5695340.61 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3952, pruned_loss=0.1426, over 5654268.87 frames. ], batch size: 155, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:24:08,379 INFO [optim.py:369] (1/2) 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 07:24:45,824 INFO [train.py:968] (1/2) Epoch 4, batch 27800, giga_loss[loss=0.3285, simple_loss=0.3903, pruned_loss=0.1333, over 28902.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3945, pruned_loss=0.1433, over 5650286.39 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4023, pruned_loss=0.1522, over 5701070.57 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3945, pruned_loss=0.143, over 5642367.71 frames. ], batch size: 186, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:25:38,598 INFO [train.py:968] (1/2) Epoch 4, batch 27850, giga_loss[loss=0.3278, simple_loss=0.3837, pruned_loss=0.136, over 29069.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3923, pruned_loss=0.1426, over 5674740.21 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4035, pruned_loss=0.1531, over 5706642.85 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.391, pruned_loss=0.1412, over 5662113.56 frames. ], batch size: 136, lr: 7.61e-03, grad_scale: 4.0
+2023-03-02 07:25:47,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:25:48,409 INFO [optim.py:369] (1/2) 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 07:25:51,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:25:56,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:26:04,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6127, 3.3971, 3.3375, 1.4604], device='cuda:1'), covar=tensor([0.0627, 0.0570, 0.0908, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0720, 0.0826, 0.0584], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:26:10,598 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 07:26:18,028 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 27900, giga_loss[loss=0.3924, simple_loss=0.422, pruned_loss=0.1814, over 28942.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3943, pruned_loss=0.1441, over 5672493.79 frames. ], libri_tot_loss[loss=0.3556, simple_loss=0.4041, pruned_loss=0.1535, over 5707706.17 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3925, pruned_loss=0.1424, over 5660458.27 frames. ], batch size: 106, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:27:09,896 INFO [train.py:968] (1/2) Epoch 4, batch 27950, libri_loss[loss=0.3173, simple_loss=0.3758, pruned_loss=0.1294, over 29530.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3963, pruned_loss=0.1449, over 5664709.94 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4041, pruned_loss=0.1533, over 5711675.25 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3944, pruned_loss=0.1434, over 5649738.22 frames. ], batch size: 80, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:27:20,723 INFO [optim.py:369] (1/2) 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 07:27:59,168 INFO [train.py:968] (1/2) Epoch 4, batch 28000, giga_loss[loss=0.3157, simple_loss=0.3744, pruned_loss=0.1285, over 28811.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3976, pruned_loss=0.146, over 5660547.73 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.404, pruned_loss=0.1533, over 5714385.55 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.396, pruned_loss=0.1447, over 5645807.54 frames. ], batch size: 227, lr: 7.60e-03, grad_scale: 8.0
+2023-03-02 07:28:04,484 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=164151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:28:07,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=164154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:28:34,880 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=164183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:28:42,044 INFO [train.py:968] (1/2) Epoch 4, batch 28050, giga_loss[loss=0.3028, simple_loss=0.3751, pruned_loss=0.1152, over 28856.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3986, pruned_loss=0.1468, over 5652241.49 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4051, pruned_loss=0.1538, over 5705519.52 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3962, pruned_loss=0.145, over 5647461.50 frames. ], batch size: 174, lr: 7.60e-03, grad_scale: 8.0
+2023-03-02 07:28:51,711 INFO [optim.py:369] (1/2) 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 07:29:03,901 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-02 07:29:11,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5009, 2.9235, 1.5061, 1.3280], device='cuda:1'), covar=tensor([0.0761, 0.0350, 0.0786, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0473, 0.0311, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 07:29:27,870 INFO [train.py:968] (1/2) Epoch 4, batch 28100, giga_loss[loss=0.3222, simple_loss=0.3833, pruned_loss=0.1306, over 28502.00 frames. ], tot_loss[loss=0.347, simple_loss=0.399, pruned_loss=0.1475, over 5658827.52 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4053, pruned_loss=0.154, over 5708691.11 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3968, pruned_loss=0.1459, over 5651489.35 frames. ], batch size: 71, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:30:12,033 INFO [train.py:968] (1/2) Epoch 4, batch 28150, giga_loss[loss=0.3196, simple_loss=0.3813, pruned_loss=0.129, over 28798.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4023, pruned_loss=0.1501, over 5655717.73 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4055, pruned_loss=0.1542, over 5702916.22 frames. ], giga_tot_loss[loss=0.3486, simple_loss=0.4003, pruned_loss=0.1484, over 5654032.38 frames. ], batch size: 99, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:30:19,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6565, 2.0744, 1.6436, 1.8922], device='cuda:1'), covar=tensor([0.0536, 0.0717, 0.0893, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0464, 0.0520, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:30:22,284 INFO [optim.py:369] (1/2) 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 07:30:51,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0159, 5.7415, 5.5098, 2.7684], device='cuda:1'), covar=tensor([0.0386, 0.0517, 0.0976, 0.1685], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0734, 0.0843, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:30:58,875 INFO [train.py:968] (1/2) Epoch 4, batch 28200, giga_loss[loss=0.3379, simple_loss=0.4007, pruned_loss=0.1375, over 28864.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4022, pruned_loss=0.149, over 5662276.59 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4055, pruned_loss=0.1542, over 5704649.97 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.4006, pruned_loss=0.1477, over 5658990.99 frames. ], batch size: 186, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:31:46,565 INFO [train.py:968] (1/2) Epoch 4, batch 28250, giga_loss[loss=0.33, simple_loss=0.3905, pruned_loss=0.1347, over 28802.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4035, pruned_loss=0.1504, over 5649932.70 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.4058, pruned_loss=0.1542, over 5703182.73 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4019, pruned_loss=0.1493, over 5647325.90 frames. ], batch size: 284, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:31:54,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5019, 3.0676, 1.4395, 1.3936], device='cuda:1'), covar=tensor([0.0876, 0.0326, 0.0861, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0474, 0.0311, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 07:31:56,095 INFO [optim.py:369] (1/2) 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 07:32:26,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2814, 1.4608, 1.1004, 0.8774], device='cuda:1'), covar=tensor([0.1045, 0.0688, 0.0584, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.1349, 0.1109, 0.1089, 0.1183], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 07:32:27,924 INFO [train.py:968] (1/2) Epoch 4, batch 28300, giga_loss[loss=0.37, simple_loss=0.4221, pruned_loss=0.159, over 28872.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4035, pruned_loss=0.1511, over 5647104.51 frames. ], libri_tot_loss[loss=0.3562, simple_loss=0.4051, pruned_loss=0.1537, over 5693554.45 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4028, pruned_loss=0.1505, over 5650798.35 frames. ], batch size: 174, lr: 7.60e-03, grad_scale: 4.0
+2023-03-02 07:32:48,285 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-02 07:33:05,103 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:33:16,914 INFO [train.py:968] (1/2) Epoch 4, batch 28350, giga_loss[loss=0.3186, simple_loss=0.3809, pruned_loss=0.1282, over 27588.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4041, pruned_loss=0.15, over 5635229.24 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4045, pruned_loss=0.1534, over 5677557.74 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.404, pruned_loss=0.1497, over 5651348.78 frames. ], batch size: 472, lr: 7.59e-03, grad_scale: 4.0
+2023-03-02 07:33:27,683 INFO [optim.py:369] (1/2) 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 07:33:30,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3583, 1.5974, 3.6276, 3.1472], device='cuda:1'), covar=tensor([0.1238, 0.1705, 0.0389, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0515, 0.0736, 0.0595], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 07:34:08,155 INFO [train.py:968] (1/2) Epoch 4, batch 28400, giga_loss[loss=0.437, simple_loss=0.4498, pruned_loss=0.2121, over 26636.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4045, pruned_loss=0.1498, over 5638929.77 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4042, pruned_loss=0.1532, over 5669833.26 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4047, pruned_loss=0.1497, over 5657724.58 frames. ], batch size: 555, lr: 7.59e-03, grad_scale: 8.0
+2023-03-02 07:34:14,743 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:34:57,773 INFO [train.py:968] (1/2) Epoch 4, batch 28450, libri_loss[loss=0.3165, simple_loss=0.3704, pruned_loss=0.1313, over 29596.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4032, pruned_loss=0.1503, over 5648202.52 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4043, pruned_loss=0.1533, over 5672536.78 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4033, pruned_loss=0.1502, over 5660156.25 frames. ], batch size: 74, lr: 7.59e-03, grad_scale: 4.0
+2023-03-02 07:34:58,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9712, 1.7314, 1.3490, 1.4449], device='cuda:1'), covar=tensor([0.0609, 0.0694, 0.0981, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0468, 0.0514, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 07:35:10,658 INFO [optim.py:369] (1/2) 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 07:35:54,153 INFO [train.py:968] (1/2) Epoch 4, batch 28500, giga_loss[loss=0.3217, simple_loss=0.3802, pruned_loss=0.1316, over 28398.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4024, pruned_loss=0.1499, over 5647685.26 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4045, pruned_loss=0.1534, over 5661774.75 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4023, pruned_loss=0.1495, over 5666686.86 frames. ], batch size: 78, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:36:49,780 INFO [train.py:968] (1/2) Epoch 4, batch 28550, giga_loss[loss=0.4006, simple_loss=0.4221, pruned_loss=0.1895, over 26546.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4003, pruned_loss=0.1487, over 5659296.54 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.404, pruned_loss=0.1531, over 5668890.59 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4005, pruned_loss=0.1486, over 5667804.04 frames. ], batch size: 555, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:36:58,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3780, 1.4723, 1.2339, 1.4875], device='cuda:1'), covar=tensor([0.1972, 0.2046, 0.1974, 0.1920], device='cuda:1'), in_proj_covar=tensor([0.1091, 0.0868, 0.0975, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 07:37:01,156 INFO [optim.py:369] (1/2) 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 07:37:32,971 INFO [train.py:968] (1/2) Epoch 4, batch 28600, giga_loss[loss=0.3446, simple_loss=0.3915, pruned_loss=0.1488, over 28708.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4019, pruned_loss=0.1505, over 5667311.25 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4044, pruned_loss=0.1535, over 5673214.06 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4016, pruned_loss=0.15, over 5669827.21 frames. ], batch size: 262, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:37:43,554 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:38:21,789 INFO [train.py:968] (1/2) Epoch 4, batch 28650, giga_loss[loss=0.3484, simple_loss=0.4016, pruned_loss=0.1476, over 28683.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4034, pruned_loss=0.1524, over 5658237.16 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.4047, pruned_loss=0.1537, over 5677021.43 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.4029, pruned_loss=0.1517, over 5656667.13 frames. ], batch size: 262, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:38:35,562 INFO [optim.py:369] (1/2) 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 07:39:08,599 INFO [train.py:968] (1/2) Epoch 4, batch 28700, giga_loss[loss=0.4293, simple_loss=0.4507, pruned_loss=0.204, over 28643.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4026, pruned_loss=0.1518, over 5663008.41 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.4035, pruned_loss=0.1528, over 5681638.08 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4033, pruned_loss=0.1521, over 5656680.15 frames. ], batch size: 336, lr: 7.59e-03, grad_scale: 2.0
+2023-03-02 07:39:25,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:39:54,943 INFO [train.py:968] (1/2) Epoch 4, batch 28750, giga_loss[loss=0.4386, simple_loss=0.4545, pruned_loss=0.2113, over 28000.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4041, pruned_loss=0.1536, over 5659269.29 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4036, pruned_loss=0.1526, over 5684427.48 frames. ], giga_tot_loss[loss=0.3562, simple_loss=0.4045, pruned_loss=0.154, over 5651095.62 frames. ], batch size: 412, lr: 7.58e-03, grad_scale: 2.0
+2023-03-02 07:40:07,831 INFO [optim.py:369] (1/2) 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 07:40:24,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=164923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:40:32,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-02 07:40:44,536 INFO [train.py:968] (1/2) Epoch 4, batch 28800, giga_loss[loss=0.4353, simple_loss=0.4388, pruned_loss=0.2159, over 23528.00 frames. ], tot_loss[loss=0.357, simple_loss=0.4052, pruned_loss=0.1544, over 5652760.75 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4037, pruned_loss=0.1526, over 5688959.85 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.4055, pruned_loss=0.1548, over 5641567.74 frames. ], batch size: 705, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:41:03,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=164961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:41:14,340 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 28850, giga_loss[loss=0.326, simple_loss=0.3847, pruned_loss=0.1336, over 28955.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4053, pruned_loss=0.1552, over 5642070.57 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.4041, pruned_loss=0.1529, over 5681959.65 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4052, pruned_loss=0.1552, over 5639513.28 frames. ], batch size: 136, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:41:39,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:41:43,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:41:44,356 INFO [optim.py:369] (1/2) 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 07:42:09,358 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 28900, giga_loss[loss=0.3614, simple_loss=0.4104, pruned_loss=0.1562, over 28257.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.4056, pruned_loss=0.1559, over 5650997.16 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4047, pruned_loss=0.1533, over 5684710.79 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.405, pruned_loss=0.1555, over 5645956.56 frames. ], batch size: 77, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:42:25,822 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:42:40,863 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:42:42,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:05,938 INFO [train.py:968] (1/2) Epoch 4, batch 28950, giga_loss[loss=0.3324, simple_loss=0.3855, pruned_loss=0.1397, over 28814.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4053, pruned_loss=0.1553, over 5646470.46 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4042, pruned_loss=0.1529, over 5688147.34 frames. ], giga_tot_loss[loss=0.3579, simple_loss=0.4052, pruned_loss=0.1553, over 5638991.01 frames. ], batch size: 186, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:43:11,202 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:17,572 INFO [optim.py:369] (1/2) 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 07:43:40,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:50,262 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:43:55,281 INFO [train.py:968] (1/2) Epoch 4, batch 29000, giga_loss[loss=0.351, simple_loss=0.4001, pruned_loss=0.151, over 28667.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.4051, pruned_loss=0.1542, over 5643551.65 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4046, pruned_loss=0.1532, over 5682331.88 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4047, pruned_loss=0.154, over 5642683.44 frames. ], batch size: 242, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:44:41,079 INFO [train.py:968] (1/2) Epoch 4, batch 29050, giga_loss[loss=0.3366, simple_loss=0.3917, pruned_loss=0.1408, over 28907.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4062, pruned_loss=0.1549, over 5643477.53 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4042, pruned_loss=0.153, over 5674271.38 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4062, pruned_loss=0.1549, over 5649452.99 frames. ], batch size: 145, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:44:54,433 INFO [optim.py:369] (1/2) 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 07:44:59,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5836, 1.4443, 1.4743, 1.4255], device='cuda:1'), covar=tensor([0.0889, 0.1395, 0.1365, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0741, 0.0642, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 07:45:29,503 INFO [train.py:968] (1/2) Epoch 4, batch 29100, giga_loss[loss=0.3295, simple_loss=0.4017, pruned_loss=0.1286, over 28993.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.4062, pruned_loss=0.1546, over 5659457.66 frames. ], libri_tot_loss[loss=0.3556, simple_loss=0.4046, pruned_loss=0.1533, over 5677481.93 frames. ], giga_tot_loss[loss=0.3573, simple_loss=0.4059, pruned_loss=0.1544, over 5660937.37 frames. ], batch size: 145, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:45:54,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:45:58,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:46:13,179 INFO [train.py:968] (1/2) Epoch 4, batch 29150, giga_loss[loss=0.3433, simple_loss=0.4015, pruned_loss=0.1425, over 28621.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4077, pruned_loss=0.1559, over 5669217.84 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4044, pruned_loss=0.1531, over 5683178.39 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4077, pruned_loss=0.156, over 5664804.22 frames. ], batch size: 307, lr: 7.58e-03, grad_scale: 4.0
+2023-03-02 07:46:22,034 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:46:23,600 INFO [optim.py:369] (1/2) 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 07:46:55,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:47:01,882 INFO [train.py:968] (1/2) Epoch 4, batch 29200, giga_loss[loss=0.3235, simple_loss=0.3968, pruned_loss=0.1252, over 28913.00 frames. ], tot_loss[loss=0.3599, simple_loss=0.4082, pruned_loss=0.1558, over 5669447.84 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4041, pruned_loss=0.153, over 5687572.82 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4085, pruned_loss=0.156, over 5662120.00 frames. ], batch size: 145, lr: 7.57e-03, grad_scale: 8.0
+2023-03-02 07:47:04,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:47:50,430 INFO [train.py:968] (1/2) Epoch 4, batch 29250, giga_loss[loss=0.3774, simple_loss=0.4295, pruned_loss=0.1626, over 28971.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.409, pruned_loss=0.1549, over 5671719.35 frames. ], libri_tot_loss[loss=0.3559, simple_loss=0.4048, pruned_loss=0.1535, over 5690512.07 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4087, pruned_loss=0.1547, over 5662061.87 frames. ], batch size: 213, lr: 7.57e-03, grad_scale: 4.0
+2023-03-02 07:48:04,923 INFO [optim.py:369] (1/2) 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 07:48:12,653 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 07:48:18,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:48:33,274 INFO [train.py:968] (1/2) Epoch 4, batch 29300, giga_loss[loss=0.3436, simple_loss=0.4019, pruned_loss=0.1426, over 28686.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.407, pruned_loss=0.153, over 5673685.57 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4044, pruned_loss=0.1533, over 5695301.13 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4073, pruned_loss=0.153, over 5660949.90 frames. ], batch size: 336, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:49:10,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:13,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:18,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:21,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:21,965 INFO [train.py:968] (1/2) Epoch 4, batch 29350, giga_loss[loss=0.2835, simple_loss=0.3601, pruned_loss=0.1034, over 29040.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4039, pruned_loss=0.151, over 5671550.13 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4045, pruned_loss=0.1533, over 5697157.63 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.404, pruned_loss=0.151, over 5659457.51 frames. ], batch size: 155, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:49:34,966 INFO [optim.py:369] (1/2) 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 07:49:38,568 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:40,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=165514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:43,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:49:46,828 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:50:03,584 INFO [train.py:968] (1/2) Epoch 4, batch 29400, libri_loss[loss=0.3075, simple_loss=0.3612, pruned_loss=0.1269, over 29373.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.4059, pruned_loss=0.1531, over 5671283.73 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4043, pruned_loss=0.1531, over 5704029.92 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4063, pruned_loss=0.1532, over 5653365.88 frames. ], batch size: 71, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:50:30,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:50:33,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:50:57,173 INFO [train.py:968] (1/2) Epoch 4, batch 29450, giga_loss[loss=0.3059, simple_loss=0.3744, pruned_loss=0.1187, over 28545.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4067, pruned_loss=0.1535, over 5669601.84 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4046, pruned_loss=0.1534, over 5708057.26 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.4068, pruned_loss=0.1534, over 5650739.63 frames. ], batch size: 71, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:51:06,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:51:11,399 INFO [optim.py:369] (1/2) 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 07:51:45,974 INFO [train.py:968] (1/2) Epoch 4, batch 29500, libri_loss[loss=0.4094, simple_loss=0.4543, pruned_loss=0.1823, over 29259.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4054, pruned_loss=0.1531, over 5674495.74 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4048, pruned_loss=0.1536, over 5710852.07 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4053, pruned_loss=0.1528, over 5656323.21 frames. ], batch size: 94, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:51:59,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=165657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:52:03,173 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=165660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:52:28,263 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=165689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:52:32,898 INFO [train.py:968] (1/2) Epoch 4, batch 29550, giga_loss[loss=0.4575, simple_loss=0.465, pruned_loss=0.225, over 26640.00 frames. ], tot_loss[loss=0.3561, simple_loss=0.405, pruned_loss=0.1536, over 5663524.71 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4048, pruned_loss=0.1536, over 5712395.70 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4049, pruned_loss=0.1534, over 5647371.98 frames. ], batch size: 555, lr: 7.57e-03, grad_scale: 2.0
+2023-03-02 07:52:47,634 INFO [optim.py:369] (1/2) 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 07:53:06,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8063, 2.5242, 1.3694, 1.1435], device='cuda:1'), covar=tensor([0.1204, 0.0685, 0.0863, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.1319, 0.1106, 0.1088, 0.1177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 07:53:16,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7502, 1.5168, 1.5514, 1.6549], device='cuda:1'), covar=tensor([0.1066, 0.1834, 0.1676, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0763, 0.0654, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 07:53:21,337 INFO [train.py:968] (1/2) Epoch 4, batch 29600, giga_loss[loss=0.3598, simple_loss=0.4106, pruned_loss=0.1545, over 28639.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4052, pruned_loss=0.1537, over 5667622.90 frames. ], libri_tot_loss[loss=0.3557, simple_loss=0.4046, pruned_loss=0.1534, over 5711067.34 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4053, pruned_loss=0.1536, over 5655144.09 frames. ], batch size: 307, lr: 7.57e-03, grad_scale: 4.0
+2023-03-02 07:53:40,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5914, 3.5256, 1.5260, 1.5359], device='cuda:1'), covar=tensor([0.0836, 0.0330, 0.0816, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0476, 0.0310, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 07:54:09,417 INFO [train.py:968] (1/2) Epoch 4, batch 29650, giga_loss[loss=0.3104, simple_loss=0.3686, pruned_loss=0.1261, over 28729.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4049, pruned_loss=0.1529, over 5663652.44 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4046, pruned_loss=0.1533, over 5714159.00 frames. ], giga_tot_loss[loss=0.3555, simple_loss=0.4051, pruned_loss=0.153, over 5650465.43 frames. ], batch size: 99, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:54:27,182 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 29700, giga_loss[loss=0.3015, simple_loss=0.3645, pruned_loss=0.1193, over 28741.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4044, pruned_loss=0.152, over 5674230.99 frames. ], libri_tot_loss[loss=0.3554, simple_loss=0.4045, pruned_loss=0.1532, over 5717170.52 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4045, pruned_loss=0.1522, over 5660469.39 frames. ], batch size: 92, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:55:37,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-02 07:55:46,189 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 29750, giga_loss[loss=0.3769, simple_loss=0.4169, pruned_loss=0.1685, over 27569.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4043, pruned_loss=0.1512, over 5674187.06 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4051, pruned_loss=0.1534, over 5717173.59 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4039, pruned_loss=0.1511, over 5662282.95 frames. ], batch size: 472, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:56:00,850 INFO [optim.py:369] (1/2) 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 07:56:35,760 INFO [train.py:968] (1/2) Epoch 4, batch 29800, giga_loss[loss=0.3084, simple_loss=0.3761, pruned_loss=0.1204, over 28992.00 frames. ], tot_loss[loss=0.3545, simple_loss=0.4054, pruned_loss=0.1518, over 5661171.03 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4054, pruned_loss=0.1536, over 5710763.92 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4048, pruned_loss=0.1515, over 5656509.45 frames. ], batch size: 136, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:56:38,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=165944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:57:24,664 INFO [train.py:968] (1/2) Epoch 4, batch 29850, giga_loss[loss=0.3211, simple_loss=0.3807, pruned_loss=0.1308, over 28857.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4026, pruned_loss=0.1503, over 5665543.64 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4041, pruned_loss=0.1528, over 5715481.82 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4032, pruned_loss=0.1507, over 5656131.07 frames. ], batch size: 186, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:57:41,075 INFO [optim.py:369] (1/2) 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 07:57:50,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 07:58:04,415 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:58:07,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:58:08,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1810, 1.7421, 1.2331, 0.4299], device='cuda:1'), covar=tensor([0.1224, 0.0865, 0.1479, 0.1895], device='cuda:1'), in_proj_covar=tensor([0.1362, 0.1283, 0.1353, 0.1136], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 07:58:12,505 INFO [train.py:968] (1/2) Epoch 4, batch 29900, giga_loss[loss=0.455, simple_loss=0.4533, pruned_loss=0.2284, over 23497.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4026, pruned_loss=0.1509, over 5670339.43 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4045, pruned_loss=0.153, over 5717930.13 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4027, pruned_loss=0.1509, over 5660016.46 frames. ], batch size: 705, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:58:30,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 07:58:54,074 INFO [train.py:968] (1/2) Epoch 4, batch 29950, giga_loss[loss=0.4559, simple_loss=0.4578, pruned_loss=0.227, over 27567.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4025, pruned_loss=0.1517, over 5642392.20 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.4051, pruned_loss=0.1534, over 5692592.15 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.402, pruned_loss=0.1513, over 5653641.05 frames. ], batch size: 472, lr: 7.56e-03, grad_scale: 2.0
+2023-03-02 07:59:08,432 INFO [optim.py:369] (1/2) 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 07:59:30,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9224, 1.3020, 3.2629, 2.8159], device='cuda:1'), covar=tensor([0.1392, 0.1817, 0.0429, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0521, 0.0745, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 07:59:43,081 INFO [train.py:968] (1/2) Epoch 4, batch 30000, giga_loss[loss=0.3267, simple_loss=0.3806, pruned_loss=0.1364, over 28894.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3965, pruned_loss=0.1475, over 5659786.39 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4043, pruned_loss=0.1529, over 5696685.90 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3968, pruned_loss=0.1476, over 5664489.35 frames. ], batch size: 213, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 07:59:43,082 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 07:59:51,385 INFO [train.py:1012] (1/2) Epoch 4, validation: loss=0.245, simple_loss=0.3484, pruned_loss=0.07078, over 944034.00 frames. 
+2023-03-02 07:59:51,386 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 07:59:51,628 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:00:33,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9114, 1.6987, 0.8868, 0.9924], device='cuda:1'), covar=tensor([0.0614, 0.0411, 0.0633, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0479, 0.0311, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 08:00:34,694 INFO [train.py:968] (1/2) Epoch 4, batch 30050, giga_loss[loss=0.3436, simple_loss=0.3868, pruned_loss=0.1502, over 28901.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.3954, pruned_loss=0.1475, over 5662300.60 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4052, pruned_loss=0.1537, over 5681916.07 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3945, pruned_loss=0.1466, over 5679366.34 frames. ], batch size: 106, lr: 7.56e-03, grad_scale: 4.0
+2023-03-02 08:00:36,829 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:00:50,322 INFO [optim.py:369] (1/2) 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 08:01:24,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6899, 2.4422, 1.7783, 0.7869], device='cuda:1'), covar=tensor([0.2172, 0.1213, 0.1680, 0.2402], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.1287, 0.1355, 0.1138], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 08:01:24,462 INFO [train.py:968] (1/2) Epoch 4, batch 30100, giga_loss[loss=0.3324, simple_loss=0.3835, pruned_loss=0.1407, over 28639.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3948, pruned_loss=0.1471, over 5675740.70 frames. ], libri_tot_loss[loss=0.3559, simple_loss=0.405, pruned_loss=0.1534, over 5683317.02 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3942, pruned_loss=0.1466, over 5687672.55 frames. ], batch size: 78, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:01:37,617 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-02 08:02:11,198 INFO [train.py:968] (1/2) Epoch 4, batch 30150, giga_loss[loss=0.3391, simple_loss=0.3975, pruned_loss=0.1403, over 28273.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3937, pruned_loss=0.1454, over 5677820.04 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4045, pruned_loss=0.1532, over 5689079.70 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3933, pruned_loss=0.1451, over 5681782.32 frames. ], batch size: 368, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:02:29,420 INFO [optim.py:369] (1/2) 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 08:02:35,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 08:02:38,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:02:49,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3349, 2.7499, 1.2951, 1.3338], device='cuda:1'), covar=tensor([0.0863, 0.0355, 0.0903, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0477, 0.0309, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 08:03:01,491 INFO [train.py:968] (1/2) Epoch 4, batch 30200, giga_loss[loss=0.3229, simple_loss=0.3903, pruned_loss=0.1277, over 28740.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3912, pruned_loss=0.1415, over 5677948.27 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4039, pruned_loss=0.1529, over 5691598.41 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3911, pruned_loss=0.1413, over 5678366.46 frames. ], batch size: 262, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:03:27,961 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-02 08:03:51,522 INFO [train.py:968] (1/2) Epoch 4, batch 30250, giga_loss[loss=0.3026, simple_loss=0.3727, pruned_loss=0.1163, over 28724.00 frames. ], tot_loss[loss=0.334, simple_loss=0.389, pruned_loss=0.1396, over 5666656.08 frames. ], libri_tot_loss[loss=0.3551, simple_loss=0.4039, pruned_loss=0.1531, over 5698197.55 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3884, pruned_loss=0.1387, over 5660429.79 frames. ], batch size: 284, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:04:04,999 INFO [optim.py:369] (1/2) 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 08:04:44,318 INFO [train.py:968] (1/2) Epoch 4, batch 30300, giga_loss[loss=0.2896, simple_loss=0.3645, pruned_loss=0.1073, over 28657.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3841, pruned_loss=0.1344, over 5661488.29 frames. ], libri_tot_loss[loss=0.3548, simple_loss=0.4037, pruned_loss=0.153, over 5697783.05 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3837, pruned_loss=0.1336, over 5656488.80 frames. ], batch size: 242, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:04:47,241 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166447.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 08:05:00,446 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:05:02,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:05:31,826 INFO [train.py:968] (1/2) Epoch 4, batch 30350, giga_loss[loss=0.2488, simple_loss=0.316, pruned_loss=0.09085, over 24200.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1305, over 5651811.10 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4037, pruned_loss=0.1531, over 5691909.42 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3795, pruned_loss=0.1294, over 5652028.65 frames. ], batch size: 705, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:05:32,906 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:05:47,286 INFO [optim.py:369] (1/2) 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 08:05:55,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:06:18,080 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:06:21,312 INFO [train.py:968] (1/2) Epoch 4, batch 30400, giga_loss[loss=0.3206, simple_loss=0.3897, pruned_loss=0.1258, over 28759.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3773, pruned_loss=0.1267, over 5645714.59 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4029, pruned_loss=0.1527, over 5694916.20 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3769, pruned_loss=0.1256, over 5642312.19 frames. ], batch size: 284, lr: 7.55e-03, grad_scale: 8.0
+2023-03-02 08:06:50,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:07:13,093 INFO [train.py:968] (1/2) Epoch 4, batch 30450, giga_loss[loss=0.3615, simple_loss=0.412, pruned_loss=0.1556, over 27825.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3778, pruned_loss=0.1265, over 5638563.53 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4023, pruned_loss=0.1525, over 5690354.11 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3775, pruned_loss=0.1254, over 5638288.65 frames. ], batch size: 412, lr: 7.55e-03, grad_scale: 4.0
+2023-03-02 08:07:14,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6217, 3.4009, 1.8925, 1.7139], device='cuda:1'), covar=tensor([0.0787, 0.0311, 0.0574, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.1085, 0.1056, 0.1144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 08:07:30,980 INFO [optim.py:369] (1/2) 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 08:08:02,222 INFO [train.py:968] (1/2) Epoch 4, batch 30500, giga_loss[loss=0.2854, simple_loss=0.356, pruned_loss=0.1073, over 28824.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3764, pruned_loss=0.1257, over 5632507.53 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4018, pruned_loss=0.1522, over 5684922.91 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3761, pruned_loss=0.1245, over 5635466.74 frames. ], batch size: 99, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:08:21,141 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:08:24,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:08:40,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1366, 1.3679, 1.2004, 1.3761], device='cuda:1'), covar=tensor([0.0825, 0.0335, 0.0357, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0132, 0.0136, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:1')
+2023-03-02 08:08:53,807 INFO [train.py:968] (1/2) Epoch 4, batch 30550, libri_loss[loss=0.275, simple_loss=0.3304, pruned_loss=0.1098, over 29642.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1226, over 5629075.63 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4015, pruned_loss=0.1521, over 5681297.52 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3718, pruned_loss=0.1211, over 5633604.51 frames. ], batch size: 73, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:08:54,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:08:59,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:09:10,069 INFO [optim.py:369] (1/2) 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 08:09:12,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-02 08:09:13,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:09:14,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:09:41,184 INFO [train.py:968] (1/2) Epoch 4, batch 30600, giga_loss[loss=0.3274, simple_loss=0.3861, pruned_loss=0.1343, over 28204.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3706, pruned_loss=0.1218, over 5634690.02 frames. ], libri_tot_loss[loss=0.3526, simple_loss=0.4012, pruned_loss=0.152, over 5686759.60 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1197, over 5631912.21 frames. ], batch size: 368, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:09:44,025 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:10:26,455 INFO [train.py:968] (1/2) Epoch 4, batch 30650, giga_loss[loss=0.3498, simple_loss=0.4093, pruned_loss=0.1452, over 27830.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.1219, over 5631214.95 frames. ], libri_tot_loss[loss=0.3525, simple_loss=0.4008, pruned_loss=0.1521, over 5677071.93 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3699, pruned_loss=0.1194, over 5637106.18 frames. ], batch size: 412, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:10:42,317 INFO [optim.py:369] (1/2) 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 08:10:54,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166822.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 08:10:55,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-02 08:11:12,283 INFO [train.py:968] (1/2) Epoch 4, batch 30700, giga_loss[loss=0.2815, simple_loss=0.3648, pruned_loss=0.09909, over 28872.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1209, over 5645023.63 frames. ], libri_tot_loss[loss=0.3516, simple_loss=0.3999, pruned_loss=0.1516, over 5684275.52 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3688, pruned_loss=0.1182, over 5641806.14 frames. ], batch size: 174, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:11:43,714 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:11:43,947 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-02 08:11:54,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0908, 1.3486, 1.1386, 0.9741], device='cuda:1'), covar=tensor([0.2228, 0.1884, 0.1922, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.0851, 0.0986, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:12:05,104 INFO [train.py:968] (1/2) Epoch 4, batch 30750, giga_loss[loss=0.2959, simple_loss=0.3688, pruned_loss=0.1116, over 29002.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3666, pruned_loss=0.1175, over 5646884.51 frames. ], libri_tot_loss[loss=0.3517, simple_loss=0.3999, pruned_loss=0.1517, over 5683389.06 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3651, pruned_loss=0.1148, over 5644802.01 frames. ], batch size: 164, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:12:14,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:12:20,868 INFO [optim.py:369] (1/2) 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 08:12:24,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=166915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:12:33,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=166924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:12:51,218 INFO [train.py:968] (1/2) Epoch 4, batch 30800, giga_loss[loss=0.2488, simple_loss=0.331, pruned_loss=0.08334, over 28963.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3625, pruned_loss=0.1151, over 5637372.58 frames. ], libri_tot_loss[loss=0.3512, simple_loss=0.3992, pruned_loss=0.1516, over 5680702.73 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3605, pruned_loss=0.1116, over 5636653.23 frames. ], batch size: 164, lr: 7.54e-03, grad_scale: 8.0
+2023-03-02 08:13:16,096 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=166965.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 08:13:19,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=166968.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 08:13:29,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 08:13:32,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2618, 1.2747, 1.0570, 1.1616], device='cuda:1'), covar=tensor([0.0575, 0.0424, 0.0916, 0.0710], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0453, 0.0506, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 08:13:43,729 INFO [train.py:968] (1/2) Epoch 4, batch 30850, libri_loss[loss=0.2852, simple_loss=0.3334, pruned_loss=0.1185, over 29489.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3595, pruned_loss=0.1137, over 5643228.42 frames. ], libri_tot_loss[loss=0.3502, simple_loss=0.3983, pruned_loss=0.1511, over 5684127.47 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3579, pruned_loss=0.1107, over 5638583.34 frames. ], batch size: 70, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:13:46,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=166997.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 08:14:01,072 INFO [optim.py:369] (1/2) 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 08:14:19,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4881, 2.3062, 1.6631, 0.7713], device='cuda:1'), covar=tensor([0.2050, 0.1244, 0.1991, 0.2280], device='cuda:1'), in_proj_covar=tensor([0.1350, 0.1275, 0.1348, 0.1116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 08:14:31,866 INFO [train.py:968] (1/2) Epoch 4, batch 30900, giga_loss[loss=0.2887, simple_loss=0.362, pruned_loss=0.1078, over 28692.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3583, pruned_loss=0.1137, over 5643513.11 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.3982, pruned_loss=0.1513, over 5683914.15 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3563, pruned_loss=0.1103, over 5639166.05 frames. ], batch size: 262, lr: 7.54e-03, grad_scale: 4.0
+2023-03-02 08:14:47,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:14:50,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:15:04,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-02 08:15:07,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:15:22,160 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:15:26,966 INFO [train.py:968] (1/2) Epoch 4, batch 30950, giga_loss[loss=0.3306, simple_loss=0.3765, pruned_loss=0.1424, over 26625.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3579, pruned_loss=0.1134, over 5631185.06 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.3973, pruned_loss=0.1509, over 5688579.28 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3563, pruned_loss=0.1103, over 5622970.53 frames. ], batch size: 555, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:15:43,996 INFO [optim.py:369] (1/2) 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 08:16:06,109 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 31000, giga_loss[loss=0.2686, simple_loss=0.348, pruned_loss=0.09457, over 29068.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1146, over 5612913.01 frames. ], libri_tot_loss[loss=0.3489, simple_loss=0.3966, pruned_loss=0.1506, over 5666079.23 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3597, pruned_loss=0.1114, over 5625349.44 frames. ], batch size: 128, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:16:35,182 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:16:44,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 08:16:56,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-02 08:17:13,395 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 08:17:20,618 INFO [train.py:968] (1/2) Epoch 4, batch 31050, libri_loss[loss=0.3199, simple_loss=0.3532, pruned_loss=0.1433, over 29506.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3627, pruned_loss=0.1142, over 5633962.11 frames. ], libri_tot_loss[loss=0.3485, simple_loss=0.3961, pruned_loss=0.1504, over 5670135.41 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3611, pruned_loss=0.1109, over 5639276.68 frames. ], batch size: 70, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:17:40,147 INFO [optim.py:369] (1/2) 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 08:17:44,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3873, 1.4219, 1.2359, 1.8120], device='cuda:1'), covar=tensor([0.2231, 0.2075, 0.2023, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1084, 0.0843, 0.0980, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:17:49,908 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:17:57,638 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 31100, giga_loss[loss=0.2791, simple_loss=0.3477, pruned_loss=0.1052, over 28940.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3649, pruned_loss=0.1159, over 5649121.81 frames. ], libri_tot_loss[loss=0.3488, simple_loss=0.3961, pruned_loss=0.1508, over 5666422.02 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3622, pruned_loss=0.1115, over 5655163.68 frames. ], batch size: 199, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:18:25,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:18:26,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:19:01,368 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 31150, giga_loss[loss=0.3429, simple_loss=0.4051, pruned_loss=0.1403, over 27656.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3613, pruned_loss=0.1137, over 5628187.41 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.3952, pruned_loss=0.1504, over 5653195.62 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3592, pruned_loss=0.1097, over 5645248.84 frames. ], batch size: 472, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:19:25,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:19:39,080 INFO [optim.py:369] (1/2) 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 08:19:54,078 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:20:00,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-02 08:20:21,261 INFO [train.py:968] (1/2) Epoch 4, batch 31200, giga_loss[loss=0.2429, simple_loss=0.3282, pruned_loss=0.07877, over 28696.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3604, pruned_loss=0.112, over 5633578.57 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.395, pruned_loss=0.1504, over 5648868.65 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3583, pruned_loss=0.1082, over 5649962.75 frames. ], batch size: 99, lr: 7.53e-03, grad_scale: 8.0
+2023-03-02 08:21:18,706 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:19,023 INFO [train.py:968] (1/2) Epoch 4, batch 31250, giga_loss[loss=0.2554, simple_loss=0.326, pruned_loss=0.09244, over 29024.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1114, over 5651391.97 frames. ], libri_tot_loss[loss=0.3478, simple_loss=0.3947, pruned_loss=0.1505, over 5656243.50 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3565, pruned_loss=0.107, over 5657778.53 frames. ], batch size: 285, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:21:22,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:42,644 INFO [optim.py:369] (1/2) 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 08:21:52,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:55,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:21:57,684 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:24,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:25,023 INFO [train.py:968] (1/2) Epoch 4, batch 31300, giga_loss[loss=0.2609, simple_loss=0.3398, pruned_loss=0.09097, over 28862.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3553, pruned_loss=0.11, over 5649618.44 frames. ], libri_tot_loss[loss=0.3477, simple_loss=0.3946, pruned_loss=0.1504, over 5658759.82 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3531, pruned_loss=0.1062, over 5652493.06 frames. ], batch size: 174, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:22:26,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:33,825 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:22:58,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:23:20,751 INFO [train.py:968] (1/2) Epoch 4, batch 31350, giga_loss[loss=0.2502, simple_loss=0.3266, pruned_loss=0.08687, over 29011.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3544, pruned_loss=0.1099, over 5654390.49 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.3941, pruned_loss=0.1503, over 5657334.66 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3519, pruned_loss=0.1058, over 5657352.28 frames. ], batch size: 199, lr: 7.53e-03, grad_scale: 4.0
+2023-03-02 08:23:21,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:23:25,636 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:23:32,966 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:24:06,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:24:19,579 INFO [train.py:968] (1/2) Epoch 4, batch 31400, giga_loss[loss=0.2604, simple_loss=0.3421, pruned_loss=0.08931, over 28129.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3555, pruned_loss=0.1098, over 5658687.20 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3935, pruned_loss=0.1501, over 5659637.17 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3535, pruned_loss=0.1063, over 5659112.37 frames. ], batch size: 412, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:25:23,781 INFO [train.py:968] (1/2) Epoch 4, batch 31450, giga_loss[loss=0.2948, simple_loss=0.3686, pruned_loss=0.1105, over 28445.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3573, pruned_loss=0.1097, over 5657109.71 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3936, pruned_loss=0.1502, over 5658564.65 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3553, pruned_loss=0.1064, over 5658548.25 frames. ], batch size: 368, lr: 7.52e-03, grad_scale: 2.0
+2023-03-02 08:25:50,155 INFO [optim.py:369] (1/2) 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 08:26:28,571 INFO [train.py:968] (1/2) Epoch 4, batch 31500, libri_loss[loss=0.3571, simple_loss=0.3968, pruned_loss=0.1587, over 29544.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3526, pruned_loss=0.1067, over 5661924.71 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3931, pruned_loss=0.1499, over 5662184.85 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3509, pruned_loss=0.1037, over 5659717.92 frames. ], batch size: 83, lr: 7.52e-03, grad_scale: 2.0
+2023-03-02 08:26:34,363 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:26:38,113 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:17,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:17,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:37,516 INFO [train.py:968] (1/2) Epoch 4, batch 31550, libri_loss[loss=0.2589, simple_loss=0.3165, pruned_loss=0.1006, over 28517.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3542, pruned_loss=0.1082, over 5667518.83 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.3927, pruned_loss=0.1497, over 5666313.26 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3523, pruned_loss=0.105, over 5662133.42 frames. ], batch size: 63, lr: 7.52e-03, grad_scale: 2.0
+2023-03-02 08:27:41,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7128, 2.3780, 1.6249, 0.7756], device='cuda:1'), covar=tensor([0.3136, 0.1875, 0.1896, 0.2826], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.1273, 0.1351, 0.1126], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 08:27:43,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:55,870 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:27:59,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5720, 1.5504, 1.4956, 1.4957], device='cuda:1'), covar=tensor([0.0985, 0.1728, 0.1549, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0730, 0.0622, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 08:28:01,648 INFO [optim.py:369] (1/2) 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 08:28:34,655 INFO [train.py:968] (1/2) Epoch 4, batch 31600, giga_loss[loss=0.2692, simple_loss=0.3639, pruned_loss=0.08721, over 28994.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3573, pruned_loss=0.1088, over 5664929.27 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.392, pruned_loss=0.1492, over 5662761.00 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3553, pruned_loss=0.1054, over 5663388.24 frames. ], batch size: 284, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:28:36,981 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 08:29:38,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2211, 1.3574, 1.1573, 1.5817], device='cuda:1'), covar=tensor([0.2209, 0.1946, 0.1921, 0.2071], device='cuda:1'), in_proj_covar=tensor([0.1085, 0.0845, 0.0980, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:29:41,443 INFO [train.py:968] (1/2) Epoch 4, batch 31650, giga_loss[loss=0.2754, simple_loss=0.3576, pruned_loss=0.09662, over 27709.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3596, pruned_loss=0.1076, over 5654697.50 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3917, pruned_loss=0.1492, over 5658478.52 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3576, pruned_loss=0.1041, over 5657109.01 frames. ], batch size: 472, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:30:07,574 INFO [optim.py:369] (1/2) 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 08:30:22,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=167825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:30:41,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=167840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:30:44,918 INFO [train.py:968] (1/2) Epoch 4, batch 31700, giga_loss[loss=0.2612, simple_loss=0.3469, pruned_loss=0.08782, over 28426.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3592, pruned_loss=0.106, over 5651604.24 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3914, pruned_loss=0.1489, over 5662334.27 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3574, pruned_loss=0.1029, over 5650004.83 frames. ], batch size: 368, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:30:45,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=167843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:30:46,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1895, 4.0067, 3.8865, 1.6573], device='cuda:1'), covar=tensor([0.0470, 0.0455, 0.0818, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0681, 0.0779, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 08:31:14,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:31:19,162 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=167872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:31:19,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=167872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:31:46,176 INFO [train.py:968] (1/2) Epoch 4, batch 31750, giga_loss[loss=0.2605, simple_loss=0.3408, pruned_loss=0.09004, over 28202.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3584, pruned_loss=0.105, over 5653135.07 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.391, pruned_loss=0.1486, over 5664525.64 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.357, pruned_loss=0.1023, over 5649925.40 frames. ], batch size: 412, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:32:10,346 INFO [optim.py:369] (1/2) 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 08:32:45,713 INFO [train.py:968] (1/2) Epoch 4, batch 31800, giga_loss[loss=0.2863, simple_loss=0.3564, pruned_loss=0.1081, over 29067.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3606, pruned_loss=0.1074, over 5648450.61 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.391, pruned_loss=0.1486, over 5659949.26 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3587, pruned_loss=0.1043, over 5649908.71 frames. ], batch size: 285, lr: 7.52e-03, grad_scale: 4.0
+2023-03-02 08:33:35,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2168, 1.3777, 1.3547, 1.3557], device='cuda:1'), covar=tensor([0.0716, 0.0762, 0.1158, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0736, 0.0619, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 08:33:55,689 INFO [train.py:968] (1/2) Epoch 4, batch 31850, giga_loss[loss=0.3083, simple_loss=0.364, pruned_loss=0.1263, over 26849.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3605, pruned_loss=0.1086, over 5656593.34 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3908, pruned_loss=0.1485, over 5662845.69 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3586, pruned_loss=0.1055, over 5654830.08 frames. ], batch size: 555, lr: 7.51e-03, grad_scale: 4.0
+2023-03-02 08:34:28,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:34:29,937 INFO [optim.py:369] (1/2) 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 08:34:31,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:34:34,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:34:39,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:35:15,589 INFO [train.py:968] (1/2) Epoch 4, batch 31900, giga_loss[loss=0.2817, simple_loss=0.3563, pruned_loss=0.1035, over 28971.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3622, pruned_loss=0.1104, over 5665415.36 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3909, pruned_loss=0.1485, over 5662963.08 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3602, pruned_loss=0.1073, over 5663804.91 frames. ], batch size: 106, lr: 7.51e-03, grad_scale: 4.0
+2023-03-02 08:35:16,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:35:21,914 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:36:00,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2589, 1.5222, 1.2465, 1.5271], device='cuda:1'), covar=tensor([0.0835, 0.0374, 0.0360, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0132, 0.0137, 0.0247], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:1')
+2023-03-02 08:36:29,309 INFO [train.py:968] (1/2) Epoch 4, batch 31950, giga_loss[loss=0.2653, simple_loss=0.3435, pruned_loss=0.09356, over 28785.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3574, pruned_loss=0.1073, over 5669231.59 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.391, pruned_loss=0.1487, over 5665218.99 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3555, pruned_loss=0.1045, over 5666164.77 frames. ], batch size: 262, lr: 7.51e-03, grad_scale: 4.0
+2023-03-02 08:36:29,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5672, 1.5392, 1.1551, 1.3168], device='cuda:1'), covar=tensor([0.0610, 0.0491, 0.0942, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0330, 0.0447, 0.0503, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-02 08:36:52,223 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:37:33,654 INFO [train.py:968] (1/2) Epoch 4, batch 32000, giga_loss[loss=0.307, simple_loss=0.3726, pruned_loss=0.1207, over 28629.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3568, pruned_loss=0.1075, over 5672679.93 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.3903, pruned_loss=0.1481, over 5672437.16 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.355, pruned_loss=0.1046, over 5664002.93 frames. ], batch size: 307, lr: 7.51e-03, grad_scale: 8.0
+2023-03-02 08:38:00,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:38:40,200 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 08:38:43,779 INFO [train.py:968] (1/2) Epoch 4, batch 32050, giga_loss[loss=0.2868, simple_loss=0.3582, pruned_loss=0.1077, over 29010.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3559, pruned_loss=0.1075, over 5657924.36 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.3904, pruned_loss=0.1482, over 5660583.21 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3541, pruned_loss=0.1047, over 5662456.88 frames. ], batch size: 186, lr: 7.51e-03, grad_scale: 8.0
+2023-03-02 08:38:55,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:39:08,757 INFO [optim.py:369] (1/2) 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 08:39:36,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3604, 1.6324, 1.1869, 0.8536], device='cuda:1'), covar=tensor([0.0898, 0.0620, 0.0547, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.1051, 0.1030, 0.1125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1')
+2023-03-02 08:39:43,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-02 08:39:44,822 INFO [train.py:968] (1/2) Epoch 4, batch 32100, giga_loss[loss=0.3019, simple_loss=0.3745, pruned_loss=0.1147, over 28525.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3608, pruned_loss=0.1103, over 5658692.07 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3898, pruned_loss=0.1478, over 5655265.56 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3592, pruned_loss=0.1077, over 5667084.07 frames. ], batch size: 336, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:40:42,205 INFO [train.py:968] (1/2) Epoch 4, batch 32150, giga_loss[loss=0.2649, simple_loss=0.3363, pruned_loss=0.09672, over 28970.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3603, pruned_loss=0.1111, over 5659851.36 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.389, pruned_loss=0.1474, over 5658599.37 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3588, pruned_loss=0.1082, over 5663957.30 frames. ], batch size: 186, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:40:45,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7367, 1.7280, 1.6290, 1.6717], device='cuda:1'), covar=tensor([0.1001, 0.1874, 0.1619, 0.1474], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0734, 0.0621, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 08:40:55,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2619, 1.4051, 1.1723, 1.8653], device='cuda:1'), covar=tensor([0.2286, 0.2201, 0.2203, 0.2180], device='cuda:1'), in_proj_covar=tensor([0.1070, 0.0842, 0.0969, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:41:12,932 INFO [optim.py:369] (1/2) 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 08:41:38,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1267, 1.3542, 0.9056, 0.9358], device='cuda:1'), covar=tensor([0.1013, 0.0691, 0.0569, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.1331, 0.1072, 0.1047, 0.1128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 08:41:45,915 INFO [train.py:968] (1/2) Epoch 4, batch 32200, giga_loss[loss=0.2811, simple_loss=0.3501, pruned_loss=0.1061, over 28953.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3592, pruned_loss=0.1111, over 5664386.84 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.3887, pruned_loss=0.1471, over 5663552.67 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3577, pruned_loss=0.1083, over 5663312.03 frames. ], batch size: 199, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:41:46,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:41:51,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:42:26,148 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:42:48,113 INFO [train.py:968] (1/2) Epoch 4, batch 32250, giga_loss[loss=0.3056, simple_loss=0.372, pruned_loss=0.1196, over 28921.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.359, pruned_loss=0.1109, over 5655249.07 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3881, pruned_loss=0.1468, over 5654451.22 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3579, pruned_loss=0.1085, over 5662723.47 frames. ], batch size: 213, lr: 7.51e-03, grad_scale: 2.0
+2023-03-02 08:43:20,345 INFO [optim.py:369] (1/2) 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 08:43:22,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2365, 1.4230, 1.1956, 1.5934], device='cuda:1'), covar=tensor([0.2294, 0.2041, 0.2063, 0.1979], device='cuda:1'), in_proj_covar=tensor([0.1073, 0.0840, 0.0970, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:43:38,854 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 32300, giga_loss[loss=0.2706, simple_loss=0.3517, pruned_loss=0.0948, over 27764.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.361, pruned_loss=0.1113, over 5659888.27 frames. ], libri_tot_loss[loss=0.3406, simple_loss=0.3879, pruned_loss=0.1466, over 5659428.78 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3598, pruned_loss=0.1088, over 5661371.60 frames. ], batch size: 474, lr: 7.50e-03, grad_scale: 2.0
+2023-03-02 08:44:57,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2765, 1.5304, 1.1644, 1.6413], device='cuda:1'), covar=tensor([0.2115, 0.1950, 0.2074, 0.1845], device='cuda:1'), in_proj_covar=tensor([0.1084, 0.0848, 0.0979, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:44:59,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:45:07,580 INFO [train.py:968] (1/2) Epoch 4, batch 32350, giga_loss[loss=0.2663, simple_loss=0.3433, pruned_loss=0.09466, over 28984.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3613, pruned_loss=0.1104, over 5669609.38 frames. ], libri_tot_loss[loss=0.3402, simple_loss=0.3876, pruned_loss=0.1464, over 5658713.32 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3598, pruned_loss=0.1075, over 5671597.88 frames. ], batch size: 155, lr: 7.50e-03, grad_scale: 2.0
+2023-03-02 08:45:43,332 INFO [optim.py:369] (1/2) 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 08:45:47,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2352, 1.6785, 1.5454, 1.4825], device='cuda:1'), covar=tensor([0.1333, 0.1823, 0.1112, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0778, 0.0746, 0.0762, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 08:46:11,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:46:18,391 INFO [train.py:968] (1/2) Epoch 4, batch 32400, giga_loss[loss=0.2716, simple_loss=0.3335, pruned_loss=0.1048, over 27575.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3601, pruned_loss=0.1103, over 5661679.50 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3879, pruned_loss=0.1467, over 5659473.79 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.358, pruned_loss=0.1068, over 5662633.36 frames. ], batch size: 472, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:46:57,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0161, 5.7930, 5.5248, 2.7523], device='cuda:1'), covar=tensor([0.0376, 0.0476, 0.0925, 0.1608], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0694, 0.0777, 0.0573], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 08:47:19,810 INFO [train.py:968] (1/2) Epoch 4, batch 32450, giga_loss[loss=0.2132, simple_loss=0.2932, pruned_loss=0.06656, over 28709.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3561, pruned_loss=0.1097, over 5663663.28 frames. ], libri_tot_loss[loss=0.3405, simple_loss=0.3877, pruned_loss=0.1467, over 5657640.76 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3538, pruned_loss=0.1059, over 5666347.85 frames. ], batch size: 262, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:47:30,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 08:47:48,856 INFO [optim.py:369] (1/2) 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 08:48:03,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=168626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:48:07,583 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:48:24,816 INFO [train.py:968] (1/2) Epoch 4, batch 32500, giga_loss[loss=0.2722, simple_loss=0.3391, pruned_loss=0.1027, over 28927.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3508, pruned_loss=0.1076, over 5662794.87 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3871, pruned_loss=0.1463, over 5665266.27 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3485, pruned_loss=0.1038, over 5658372.96 frames. ], batch size: 155, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:48:49,574 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:49:04,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:49:07,899 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 32550, giga_loss[loss=0.2576, simple_loss=0.3368, pruned_loss=0.08915, over 28968.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3516, pruned_loss=0.1084, over 5662306.87 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3865, pruned_loss=0.1461, over 5664407.52 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.349, pruned_loss=0.1041, over 5659038.70 frames. ], batch size: 227, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:49:39,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:49:44,386 INFO [optim.py:369] (1/2) 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 08:50:19,271 INFO [train.py:968] (1/2) Epoch 4, batch 32600, giga_loss[loss=0.2603, simple_loss=0.3366, pruned_loss=0.09199, over 28604.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3525, pruned_loss=0.1091, over 5657976.28 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3857, pruned_loss=0.1456, over 5671562.64 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3502, pruned_loss=0.1052, over 5649103.71 frames. ], batch size: 307, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:51:06,801 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-02 08:51:21,094 INFO [train.py:968] (1/2) Epoch 4, batch 32650, giga_loss[loss=0.2321, simple_loss=0.3215, pruned_loss=0.07133, over 28824.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.35, pruned_loss=0.1068, over 5649563.02 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.3857, pruned_loss=0.1457, over 5662700.55 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3479, pruned_loss=0.1033, over 5649686.84 frames. ], batch size: 174, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:51:32,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=168803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:51:38,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7111, 1.6740, 1.3187, 1.4184], device='cuda:1'), covar=tensor([0.0681, 0.0523, 0.0898, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0455, 0.0512, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 08:51:50,673 INFO [optim.py:369] (1/2) 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 08:52:23,105 INFO [train.py:968] (1/2) Epoch 4, batch 32700, giga_loss[loss=0.2622, simple_loss=0.3253, pruned_loss=0.09952, over 26812.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3492, pruned_loss=0.1057, over 5648873.46 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.3858, pruned_loss=0.1458, over 5655154.94 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3465, pruned_loss=0.1017, over 5656297.81 frames. ], batch size: 555, lr: 7.50e-03, grad_scale: 4.0
+2023-03-02 08:52:56,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 08:53:35,147 INFO [train.py:968] (1/2) Epoch 4, batch 32750, giga_loss[loss=0.2529, simple_loss=0.3279, pruned_loss=0.08896, over 29013.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3473, pruned_loss=0.1047, over 5648380.44 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.3857, pruned_loss=0.1458, over 5656460.17 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.345, pruned_loss=0.1014, over 5652856.58 frames. ], batch size: 155, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:54:01,606 INFO [optim.py:369] (1/2) 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 08:54:06,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 08:54:38,122 INFO [train.py:968] (1/2) Epoch 4, batch 32800, giga_loss[loss=0.2933, simple_loss=0.3676, pruned_loss=0.1095, over 28748.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3469, pruned_loss=0.1039, over 5659208.53 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3846, pruned_loss=0.1451, over 5666850.01 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3447, pruned_loss=0.1004, over 5653474.51 frames. ], batch size: 262, lr: 7.49e-03, grad_scale: 8.0
+2023-03-02 08:54:44,766 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=168946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:54:48,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=168949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:55:26,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=168978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:55:40,235 INFO [train.py:968] (1/2) Epoch 4, batch 32850, giga_loss[loss=0.2509, simple_loss=0.3192, pruned_loss=0.09131, over 28947.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3464, pruned_loss=0.104, over 5665138.43 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.384, pruned_loss=0.1448, over 5675878.32 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3438, pruned_loss=0.09989, over 5651965.32 frames. ], batch size: 106, lr: 7.49e-03, grad_scale: 8.0
+2023-03-02 08:55:48,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:56:06,002 INFO [optim.py:369] (1/2) 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 08:56:38,019 INFO [train.py:968] (1/2) Epoch 4, batch 32900, giga_loss[loss=0.2684, simple_loss=0.3408, pruned_loss=0.09797, over 28131.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.348, pruned_loss=0.1059, over 5671959.61 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3836, pruned_loss=0.1447, over 5683624.17 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3451, pruned_loss=0.1014, over 5654403.06 frames. ], batch size: 412, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:57:40,016 INFO [train.py:968] (1/2) Epoch 4, batch 32950, giga_loss[loss=0.2828, simple_loss=0.3576, pruned_loss=0.104, over 28455.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.345, pruned_loss=0.103, over 5667350.41 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3831, pruned_loss=0.1444, over 5686929.36 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3426, pruned_loss=0.09916, over 5650521.82 frames. ], batch size: 369, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:57:42,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3572, 1.3432, 1.1247, 1.5352], device='cuda:1'), covar=tensor([0.2286, 0.2148, 0.2199, 0.2263], device='cuda:1'), in_proj_covar=tensor([0.1072, 0.0835, 0.0973, 0.0942], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:58:07,295 INFO [optim.py:369] (1/2) 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 08:58:35,838 INFO [train.py:968] (1/2) Epoch 4, batch 33000, giga_loss[loss=0.2867, simple_loss=0.3743, pruned_loss=0.09958, over 28893.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3482, pruned_loss=0.1036, over 5674710.64 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3824, pruned_loss=0.144, over 5693803.43 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3457, pruned_loss=0.09949, over 5654422.12 frames. ], batch size: 145, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 08:58:35,838 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 08:58:44,544 INFO [train.py:1012] (1/2) Epoch 4, validation: loss=0.2225, simple_loss=0.3201, pruned_loss=0.06249, over 944034.00 frames. 
+2023-03-02 08:58:44,544 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 08:58:45,894 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:58:50,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:59:07,276 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2476, 2.4247, 1.1441, 1.2994], device='cuda:1'), covar=tensor([0.0874, 0.0344, 0.0883, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0465, 0.0312, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 08:59:07,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1485, 1.3509, 1.1336, 0.9517], device='cuda:1'), covar=tensor([0.2055, 0.1939, 0.2005, 0.1870], device='cuda:1'), in_proj_covar=tensor([0.1071, 0.0836, 0.0972, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 08:59:22,463 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 08:59:42,441 INFO [train.py:968] (1/2) Epoch 4, batch 33050, giga_loss[loss=0.3216, simple_loss=0.3706, pruned_loss=0.1363, over 26888.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3516, pruned_loss=0.1052, over 5672164.24 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3822, pruned_loss=0.1439, over 5696937.18 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3489, pruned_loss=0.1009, over 5652556.16 frames. ], batch size: 555, lr: 7.49e-03, grad_scale: 4.0
+2023-03-02 09:00:12,196 INFO [optim.py:369] (1/2) 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 09:00:46,174 INFO [train.py:968] (1/2) Epoch 4, batch 33100, giga_loss[loss=0.2885, simple_loss=0.3614, pruned_loss=0.1078, over 28373.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3533, pruned_loss=0.1066, over 5652381.23 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3821, pruned_loss=0.1439, over 5690783.18 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3504, pruned_loss=0.1022, over 5641017.50 frames. ], batch size: 368, lr: 7.49e-03, grad_scale: 2.0
+2023-03-02 09:01:25,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2667, 1.7421, 1.5884, 1.5306], device='cuda:1'), covar=tensor([0.1399, 0.1733, 0.1076, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0737, 0.0759, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 09:01:49,020 INFO [train.py:968] (1/2) Epoch 4, batch 33150, giga_loss[loss=0.2872, simple_loss=0.3628, pruned_loss=0.1058, over 28663.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3528, pruned_loss=0.1063, over 5663206.82 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3814, pruned_loss=0.1434, over 5693707.80 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3507, pruned_loss=0.1025, over 5650952.05 frames. ], batch size: 307, lr: 7.49e-03, grad_scale: 2.0
+2023-03-02 09:01:59,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3075, 1.4377, 1.5342, 1.4361], device='cuda:1'), covar=tensor([0.0897, 0.1296, 0.1410, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0735, 0.0623, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 09:02:17,011 INFO [optim.py:369] (1/2) 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 09:02:47,790 INFO [train.py:968] (1/2) Epoch 4, batch 33200, giga_loss[loss=0.2654, simple_loss=0.3432, pruned_loss=0.09381, over 29045.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3507, pruned_loss=0.1045, over 5660798.07 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3816, pruned_loss=0.1436, over 5691438.78 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3484, pruned_loss=0.1009, over 5652931.86 frames. ], batch size: 227, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:03:08,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:03:11,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169362.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 09:03:17,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 09:03:49,165 INFO [train.py:968] (1/2) Epoch 4, batch 33250, giga_loss[loss=0.2436, simple_loss=0.3152, pruned_loss=0.08603, over 28960.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3495, pruned_loss=0.1037, over 5666959.15 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3816, pruned_loss=0.1435, over 5694876.98 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3473, pruned_loss=0.1004, over 5657521.03 frames. ], batch size: 186, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:04:20,536 INFO [optim.py:369] (1/2) 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 09:04:50,414 INFO [train.py:968] (1/2) Epoch 4, batch 33300, giga_loss[loss=0.3405, simple_loss=0.4024, pruned_loss=0.1393, over 28484.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3479, pruned_loss=0.1034, over 5670998.16 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3813, pruned_loss=0.1434, over 5698303.03 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.346, pruned_loss=0.1004, over 5660041.75 frames. ], batch size: 370, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:05:27,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:05:52,310 INFO [train.py:968] (1/2) Epoch 4, batch 33350, giga_loss[loss=0.3003, simple_loss=0.3749, pruned_loss=0.1129, over 28644.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3506, pruned_loss=0.1046, over 5665827.53 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3808, pruned_loss=0.143, over 5692238.25 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3488, pruned_loss=0.1016, over 5662084.73 frames. ], batch size: 307, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:06:15,985 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:06:22,708 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 09:06:24,757 INFO [optim.py:369] (1/2) 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 09:06:29,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2964, 1.8029, 1.5607, 1.5312], device='cuda:1'), covar=tensor([0.1460, 0.1790, 0.1171, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0734, 0.0754, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 09:06:49,522 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-02 09:06:50,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8027, 4.1231, 1.7596, 1.6795], device='cuda:1'), covar=tensor([0.0797, 0.0341, 0.0823, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0463, 0.0312, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 09:06:57,331 INFO [train.py:968] (1/2) Epoch 4, batch 33400, giga_loss[loss=0.2837, simple_loss=0.3611, pruned_loss=0.1032, over 28509.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.352, pruned_loss=0.1055, over 5659394.02 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3808, pruned_loss=0.1431, over 5681986.39 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3502, pruned_loss=0.1026, over 5665404.89 frames. ], batch size: 336, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:07:00,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2161, 1.2126, 1.0683, 1.0843], device='cuda:1'), covar=tensor([0.0647, 0.0448, 0.0956, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0443, 0.0501, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-02 09:07:07,855 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=169552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:08:00,195 INFO [train.py:968] (1/2) Epoch 4, batch 33450, giga_loss[loss=0.3189, simple_loss=0.3836, pruned_loss=0.1272, over 28996.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.354, pruned_loss=0.1072, over 5660568.86 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3812, pruned_loss=0.1434, over 5683488.87 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3515, pruned_loss=0.1038, over 5663779.73 frames. ], batch size: 285, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:08:16,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2300, 1.3921, 1.1595, 1.3347], device='cuda:1'), covar=tensor([0.2104, 0.1992, 0.2073, 0.1772], device='cuda:1'), in_proj_covar=tensor([0.1066, 0.0836, 0.0968, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 09:08:34,290 INFO [optim.py:369] (1/2) 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 09:09:01,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2670, 1.5303, 1.1209, 1.3884], device='cuda:1'), covar=tensor([0.0853, 0.0333, 0.0374, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0132, 0.0137, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:1')
+2023-03-02 09:09:02,455 INFO [train.py:968] (1/2) Epoch 4, batch 33500, giga_loss[loss=0.2573, simple_loss=0.3375, pruned_loss=0.08854, over 28786.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3579, pruned_loss=0.1091, over 5655599.83 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3811, pruned_loss=0.1433, over 5680801.99 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3556, pruned_loss=0.1059, over 5660378.15 frames. ], batch size: 99, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:09:07,574 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-02 09:09:49,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5332, 1.7118, 1.5504, 1.6596], device='cuda:1'), covar=tensor([0.1146, 0.1725, 0.1689, 0.1351], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0731, 0.0621, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 09:09:55,846 INFO [train.py:968] (1/2) Epoch 4, batch 33550, giga_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09786, over 28769.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3581, pruned_loss=0.1085, over 5661077.64 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3804, pruned_loss=0.1428, over 5688924.33 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3561, pruned_loss=0.105, over 5656980.75 frames. ], batch size: 99, lr: 7.48e-03, grad_scale: 4.0
+2023-03-02 09:10:28,733 INFO [optim.py:369] (1/2) 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 09:10:49,052 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:10:50,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169737.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 09:10:54,717 INFO [train.py:968] (1/2) Epoch 4, batch 33600, giga_loss[loss=0.2782, simple_loss=0.3524, pruned_loss=0.102, over 28750.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3589, pruned_loss=0.11, over 5669530.50 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3797, pruned_loss=0.1424, over 5695014.86 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3567, pruned_loss=0.1057, over 5658896.22 frames. ], batch size: 243, lr: 7.48e-03, grad_scale: 8.0
+2023-03-02 09:12:05,453 INFO [train.py:968] (1/2) Epoch 4, batch 33650, giga_loss[loss=0.2893, simple_loss=0.3638, pruned_loss=0.1074, over 29011.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3568, pruned_loss=0.1091, over 5663114.90 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3795, pruned_loss=0.1423, over 5689557.28 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3547, pruned_loss=0.105, over 5658236.60 frames. ], batch size: 136, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:12:06,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.00 vs. limit=2.0
+2023-03-02 09:12:32,610 INFO [optim.py:369] (1/2) 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 09:12:43,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1871, 1.6386, 1.4507, 1.4085], device='cuda:1'), covar=tensor([0.1354, 0.1829, 0.1088, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0735, 0.0753, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 09:13:01,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4407, 2.1805, 1.5357, 0.5826], device='cuda:1'), covar=tensor([0.2235, 0.1231, 0.2261, 0.2724], device='cuda:1'), in_proj_covar=tensor([0.1333, 0.1284, 0.1348, 0.1126], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 09:13:01,518 INFO [train.py:968] (1/2) Epoch 4, batch 33700, giga_loss[loss=0.2641, simple_loss=0.3504, pruned_loss=0.0889, over 28872.00 frames. ], tot_loss[loss=0.287, simple_loss=0.356, pruned_loss=0.109, over 5661785.76 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3785, pruned_loss=0.1416, over 5686890.53 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3543, pruned_loss=0.1051, over 5659030.89 frames. ], batch size: 174, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:13:10,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:13:21,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4293, 4.1889, 4.0642, 1.6653], device='cuda:1'), covar=tensor([0.0577, 0.0582, 0.1023, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0688, 0.0779, 0.0565], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 09:13:48,661 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:13:52,157 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169880.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 09:13:52,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:13:54,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169883.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 09:13:54,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:14:04,647 INFO [train.py:968] (1/2) Epoch 4, batch 33750, libri_loss[loss=0.2698, simple_loss=0.3227, pruned_loss=0.1085, over 29460.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.355, pruned_loss=0.1086, over 5656820.71 frames. ], libri_tot_loss[loss=0.3303, simple_loss=0.3782, pruned_loss=0.1412, over 5689004.69 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3533, pruned_loss=0.1049, over 5652103.41 frames. ], batch size: 70, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:14:29,792 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:14:31,885 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=169912.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 09:14:38,348 INFO [optim.py:369] (1/2) 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 09:14:53,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=169927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:15:13,376 INFO [train.py:968] (1/2) Epoch 4, batch 33800, giga_loss[loss=0.2352, simple_loss=0.2942, pruned_loss=0.08816, over 24475.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3528, pruned_loss=0.1079, over 5650338.13 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3779, pruned_loss=0.1411, over 5691236.06 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3515, pruned_loss=0.1047, over 5644215.97 frames. ], batch size: 705, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:16:10,230 INFO [train.py:968] (1/2) Epoch 4, batch 33850, giga_loss[loss=0.2738, simple_loss=0.3475, pruned_loss=0.1001, over 28910.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3525, pruned_loss=0.108, over 5647927.75 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3778, pruned_loss=0.1408, over 5693309.11 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3507, pruned_loss=0.1047, over 5639959.65 frames. ], batch size: 213, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:16:11,055 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=169993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:13,423 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=169996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:18,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1610, 2.9819, 2.9036, 1.3750], device='cuda:1'), covar=tensor([0.0800, 0.0684, 0.1088, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0677, 0.0761, 0.0556], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 09:16:34,017 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:16:46,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:48,967 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:16:54,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:17:10,036 INFO [train.py:968] (1/2) Epoch 4, batch 33900, giga_loss[loss=0.2543, simple_loss=0.3399, pruned_loss=0.0843, over 28746.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.351, pruned_loss=0.1059, over 5661379.72 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3771, pruned_loss=0.1405, over 5696210.42 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3496, pruned_loss=0.1028, over 5651491.83 frames. ], batch size: 243, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:17:27,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-02 09:17:31,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:17:44,360 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:17:47,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:18:05,069 INFO [train.py:968] (1/2) Epoch 4, batch 33950, giga_loss[loss=0.2539, simple_loss=0.3468, pruned_loss=0.08055, over 28926.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3517, pruned_loss=0.1044, over 5667643.81 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3775, pruned_loss=0.1408, over 5691492.97 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3494, pruned_loss=0.1005, over 5663579.66 frames. ], batch size: 186, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:18:14,862 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:18:27,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4639, 1.7230, 1.2777, 0.9974], device='cuda:1'), covar=tensor([0.1261, 0.0734, 0.0674, 0.0917], device='cuda:1'), in_proj_covar=tensor([0.1339, 0.1073, 0.1071, 0.1141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 09:18:30,830 INFO [optim.py:369] (1/2) 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 09:18:36,765 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:19:00,234 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 34000, giga_loss[loss=0.2639, simple_loss=0.3516, pruned_loss=0.08812, over 28605.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3529, pruned_loss=0.1033, over 5669689.66 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3767, pruned_loss=0.1403, over 5694901.56 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3513, pruned_loss=0.09988, over 5663201.86 frames. ], batch size: 85, lr: 7.47e-03, grad_scale: 8.0
+2023-03-02 09:19:41,932 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:19:57,722 INFO [train.py:968] (1/2) Epoch 4, batch 34050, giga_loss[loss=0.2754, simple_loss=0.3524, pruned_loss=0.09922, over 28915.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3539, pruned_loss=0.1038, over 5672892.33 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3769, pruned_loss=0.1405, over 5701882.91 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3515, pruned_loss=0.09956, over 5660326.75 frames. ], batch size: 164, lr: 7.47e-03, grad_scale: 4.0
+2023-03-02 09:19:59,651 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-02 09:20:32,472 INFO [optim.py:369] (1/2) 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 09:21:06,835 INFO [train.py:968] (1/2) Epoch 4, batch 34100, giga_loss[loss=0.2706, simple_loss=0.3472, pruned_loss=0.09695, over 28915.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3536, pruned_loss=0.1039, over 5669867.60 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3763, pruned_loss=0.1401, over 5699642.40 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3515, pruned_loss=0.09965, over 5660265.04 frames. ], batch size: 186, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:21:29,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3864, 1.6851, 1.1932, 1.1244], device='cuda:1'), covar=tensor([0.1092, 0.0752, 0.0693, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.1334, 0.1057, 0.1054, 0.1134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 09:21:50,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2168, 1.2754, 1.1307, 1.4157], device='cuda:1'), covar=tensor([0.0817, 0.0331, 0.0355, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0132, 0.0136, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:1')
+2023-03-02 09:22:02,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 09:22:09,562 INFO [train.py:968] (1/2) Epoch 4, batch 34150, giga_loss[loss=0.2711, simple_loss=0.3514, pruned_loss=0.09545, over 28386.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3554, pruned_loss=0.1053, over 5661021.50 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3764, pruned_loss=0.1402, over 5685122.06 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3532, pruned_loss=0.1009, over 5665224.32 frames. ], batch size: 336, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:22:47,281 INFO [optim.py:369] (1/2) 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 09:23:19,975 INFO [train.py:968] (1/2) Epoch 4, batch 34200, giga_loss[loss=0.2707, simple_loss=0.3514, pruned_loss=0.09505, over 28397.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3545, pruned_loss=0.1045, over 5651161.86 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3763, pruned_loss=0.1401, over 5678206.98 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3522, pruned_loss=0.1002, over 5659262.53 frames. ], batch size: 368, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:24:23,798 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 34250, giga_loss[loss=0.294, simple_loss=0.3754, pruned_loss=0.1064, over 28879.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3545, pruned_loss=0.1038, over 5644731.62 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3766, pruned_loss=0.1402, over 5669664.78 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3519, pruned_loss=0.0994, over 5658401.71 frames. ], batch size: 213, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:24:44,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3682, 1.9340, 1.6940, 1.5266], device='cuda:1'), covar=tensor([0.1244, 0.1613, 0.1008, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0733, 0.0753, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 09:24:59,488 INFO [optim.py:369] (1/2) 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 09:25:25,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:25:30,282 INFO [train.py:968] (1/2) Epoch 4, batch 34300, giga_loss[loss=0.3138, simple_loss=0.3889, pruned_loss=0.1194, over 28854.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3588, pruned_loss=0.1065, over 5643812.75 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3768, pruned_loss=0.1403, over 5663540.58 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3557, pruned_loss=0.1015, over 5658887.51 frames. ], batch size: 174, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:26:36,204 INFO [train.py:968] (1/2) Epoch 4, batch 34350, giga_loss[loss=0.3186, simple_loss=0.3866, pruned_loss=0.1253, over 28397.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3588, pruned_loss=0.1057, over 5662401.44 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3766, pruned_loss=0.1401, over 5668568.59 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3562, pruned_loss=0.1013, over 5669802.10 frames. ], batch size: 368, lr: 7.46e-03, grad_scale: 2.0
+2023-03-02 09:26:43,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:26:49,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5943, 1.6258, 1.5625, 1.4996], device='cuda:1'), covar=tensor([0.0888, 0.1666, 0.1516, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0734, 0.0616, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 09:26:51,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3047, 1.6494, 1.2490, 1.3444], device='cuda:1'), covar=tensor([0.0753, 0.0324, 0.0342, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0205, 0.0131, 0.0134, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:1')
+2023-03-02 09:27:08,023 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:27:15,285 INFO [optim.py:369] (1/2) 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 09:27:32,517 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:27:36,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:27:45,664 INFO [train.py:968] (1/2) Epoch 4, batch 34400, giga_loss[loss=0.2503, simple_loss=0.3321, pruned_loss=0.08425, over 28632.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3578, pruned_loss=0.1058, over 5664132.97 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.377, pruned_loss=0.1403, over 5663313.84 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3551, pruned_loss=0.1015, over 5674940.13 frames. ], batch size: 60, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:27:59,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=170554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:28:04,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3254, 1.4503, 1.2224, 1.7089], device='cuda:1'), covar=tensor([0.2255, 0.2102, 0.2194, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.1070, 0.0836, 0.0962, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 09:28:10,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:28:35,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 09:28:46,802 INFO [train.py:968] (1/2) Epoch 4, batch 34450, giga_loss[loss=0.2474, simple_loss=0.3282, pruned_loss=0.0833, over 28563.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3563, pruned_loss=0.1058, over 5674624.28 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3771, pruned_loss=0.1403, over 5670831.32 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3531, pruned_loss=0.1008, over 5676714.14 frames. ], batch size: 242, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:29:23,688 INFO [optim.py:369] (1/2) 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 09:29:46,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:29:49,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5722, 1.4454, 1.1417, 1.2542], device='cuda:1'), covar=tensor([0.0597, 0.0485, 0.0867, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0437, 0.0501, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-02 09:29:50,409 INFO [train.py:968] (1/2) Epoch 4, batch 34500, giga_loss[loss=0.2473, simple_loss=0.3353, pruned_loss=0.07965, over 28609.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3538, pruned_loss=0.1033, over 5670845.01 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3767, pruned_loss=0.14, over 5660811.66 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3505, pruned_loss=0.09795, over 5681418.76 frames. ], batch size: 307, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:29:50,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:08,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:10,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:26,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:49,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:30:52,438 INFO [train.py:968] (1/2) Epoch 4, batch 34550, giga_loss[loss=0.3013, simple_loss=0.3735, pruned_loss=0.1145, over 28915.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3526, pruned_loss=0.103, over 5685593.01 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3762, pruned_loss=0.1396, over 5668658.28 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3497, pruned_loss=0.09763, over 5687615.77 frames. ], batch size: 112, lr: 7.46e-03, grad_scale: 4.0
+2023-03-02 09:30:56,677 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:31:01,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:31:20,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8895, 2.5713, 1.4799, 1.4202], device='cuda:1'), covar=tensor([0.1109, 0.0591, 0.0766, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.1052, 0.1061, 0.1137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 09:31:25,632 INFO [optim.py:369] (1/2) 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 09:31:35,617 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 34600, giga_loss[loss=0.2825, simple_loss=0.3659, pruned_loss=0.0996, over 28970.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3532, pruned_loss=0.1035, over 5677275.82 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3757, pruned_loss=0.1394, over 5669578.48 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3508, pruned_loss=0.09856, over 5678410.90 frames. ], batch size: 155, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:32:53,958 INFO [train.py:968] (1/2) Epoch 4, batch 34650, giga_loss[loss=0.3228, simple_loss=0.3927, pruned_loss=0.1264, over 28682.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3566, pruned_loss=0.1057, over 5669135.47 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3754, pruned_loss=0.1392, over 5670806.59 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3547, pruned_loss=0.1018, over 5668988.26 frames. ], batch size: 307, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:32:56,950 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5377, 2.0616, 1.8501, 1.7158], device='cuda:1'), covar=tensor([0.1474, 0.1671, 0.1102, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0730, 0.0755, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 09:33:21,357 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:33:54,487 INFO [train.py:968] (1/2) Epoch 4, batch 34700, giga_loss[loss=0.263, simple_loss=0.3389, pruned_loss=0.09355, over 28908.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.354, pruned_loss=0.1049, over 5673411.31 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3753, pruned_loss=0.1391, over 5671997.34 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3523, pruned_loss=0.1013, over 5672008.71 frames. ], batch size: 136, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:33:59,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.91 vs. limit=2.0
+2023-03-02 09:34:08,093 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:34:52,269 INFO [train.py:968] (1/2) Epoch 4, batch 34750, libri_loss[loss=0.3093, simple_loss=0.3709, pruned_loss=0.1239, over 29675.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3522, pruned_loss=0.1042, over 5670845.51 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.375, pruned_loss=0.1389, over 5675544.90 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3508, pruned_loss=0.1011, over 5666530.25 frames. ], batch size: 91, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:35:22,589 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 34800, giga_loss[loss=0.3861, simple_loss=0.4387, pruned_loss=0.1668, over 28505.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3558, pruned_loss=0.1079, over 5663452.61 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3753, pruned_loss=0.1393, over 5677867.11 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3539, pruned_loss=0.1043, over 5658041.98 frames. ], batch size: 85, lr: 7.45e-03, grad_scale: 8.0
+2023-03-02 09:35:57,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1549, 1.6471, 1.1936, 0.5191], device='cuda:1'), covar=tensor([0.1694, 0.0906, 0.1335, 0.2038], device='cuda:1'), in_proj_covar=tensor([0.1344, 0.1264, 0.1355, 0.1114], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 09:36:01,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=170956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:04,856 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=170959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:07,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=170962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:29,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=170988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:36:34,034 INFO [train.py:968] (1/2) Epoch 4, batch 34850, giga_loss[loss=0.3359, simple_loss=0.4043, pruned_loss=0.1338, over 28935.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3666, pruned_loss=0.1151, over 5667114.16 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3754, pruned_loss=0.1395, over 5672417.17 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3648, pruned_loss=0.1116, over 5667258.98 frames. ], batch size: 164, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:36:43,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-02 09:37:00,981 INFO [optim.py:369] (1/2) 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 09:37:20,346 INFO [train.py:968] (1/2) Epoch 4, batch 34900, giga_loss[loss=0.3053, simple_loss=0.3821, pruned_loss=0.1143, over 28903.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3744, pruned_loss=0.1203, over 5677317.91 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3751, pruned_loss=0.1392, over 5680383.09 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.373, pruned_loss=0.1171, over 5670096.57 frames. ], batch size: 227, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:38:03,275 INFO [train.py:968] (1/2) Epoch 4, batch 34950, giga_loss[loss=0.281, simple_loss=0.3511, pruned_loss=0.1054, over 28716.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3751, pruned_loss=0.1217, over 5682119.93 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3749, pruned_loss=0.139, over 5684368.62 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3742, pruned_loss=0.119, over 5672775.42 frames. ], batch size: 284, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:38:25,146 INFO [optim.py:369] (1/2) 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 09:38:43,437 INFO [train.py:968] (1/2) Epoch 4, batch 35000, giga_loss[loss=0.2741, simple_loss=0.3368, pruned_loss=0.1057, over 28580.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3689, pruned_loss=0.1189, over 5688023.13 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3751, pruned_loss=0.1389, over 5689327.88 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.368, pruned_loss=0.1162, over 5676313.85 frames. ], batch size: 78, lr: 7.45e-03, grad_scale: 4.0
+2023-03-02 09:39:29,474 INFO [train.py:968] (1/2) Epoch 4, batch 35050, giga_loss[loss=0.2438, simple_loss=0.3155, pruned_loss=0.086, over 28542.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3609, pruned_loss=0.115, over 5686180.97 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3752, pruned_loss=0.1388, over 5691971.69 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1124, over 5674378.85 frames. ], batch size: 307, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:39:52,544 INFO [optim.py:369] (1/2) 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 09:40:00,892 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 35100, giga_loss[loss=0.2078, simple_loss=0.2785, pruned_loss=0.06855, over 28614.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3515, pruned_loss=0.1102, over 5690499.58 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3751, pruned_loss=0.1387, over 5693069.83 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3507, pruned_loss=0.1082, over 5680358.82 frames. ], batch size: 60, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:40:26,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 09:40:54,630 INFO [train.py:968] (1/2) Epoch 4, batch 35150, giga_loss[loss=0.254, simple_loss=0.3247, pruned_loss=0.09165, over 28376.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3451, pruned_loss=0.1075, over 5689336.42 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3755, pruned_loss=0.1389, over 5696319.49 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3436, pruned_loss=0.1052, over 5678480.31 frames. ], batch size: 78, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:41:20,268 INFO [optim.py:369] (1/2) 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 09:41:20,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:41:31,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:41:35,741 INFO [train.py:968] (1/2) Epoch 4, batch 35200, giga_loss[loss=0.2502, simple_loss=0.3167, pruned_loss=0.0919, over 28565.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3416, pruned_loss=0.1064, over 5688275.19 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3765, pruned_loss=0.1394, over 5693878.87 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3381, pruned_loss=0.1027, over 5681815.53 frames. ], batch size: 336, lr: 7.44e-03, grad_scale: 8.0
+2023-03-02 09:41:59,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2149, 1.9567, 1.4297, 1.3110], device='cuda:1'), covar=tensor([0.0928, 0.0275, 0.0337, 0.1163], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0128, 0.0132, 0.0241], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0037, 0.0034, 0.0057], device='cuda:1')
+2023-03-02 09:41:59,924 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:42:01,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:42:15,991 INFO [train.py:968] (1/2) Epoch 4, batch 35250, giga_loss[loss=0.2487, simple_loss=0.3123, pruned_loss=0.09255, over 28695.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3404, pruned_loss=0.1067, over 5698100.64 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3765, pruned_loss=0.1394, over 5696830.86 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3363, pruned_loss=0.1026, over 5690182.39 frames. ], batch size: 119, lr: 7.44e-03, grad_scale: 8.0
+2023-03-02 09:42:21,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1730, 3.8967, 3.8743, 1.5988], device='cuda:1'), covar=tensor([0.0551, 0.0535, 0.0810, 0.2306], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0696, 0.0769, 0.0571], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 09:42:27,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:42:42,221 INFO [optim.py:369] (1/2) 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 09:42:47,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-02 09:42:58,753 INFO [train.py:968] (1/2) Epoch 4, batch 35300, libri_loss[loss=0.2879, simple_loss=0.3548, pruned_loss=0.1105, over 29566.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3365, pruned_loss=0.1043, over 5702241.09 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3766, pruned_loss=0.1393, over 5703447.11 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.332, pruned_loss=0.1001, over 5689735.04 frames. ], batch size: 75, lr: 7.44e-03, grad_scale: 8.0
+2023-03-02 09:43:30,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:43:32,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=171483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:43:33,152 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:43:37,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-02 09:43:39,246 INFO [train.py:968] (1/2) Epoch 4, batch 35350, giga_loss[loss=0.2998, simple_loss=0.3307, pruned_loss=0.1345, over 23842.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3352, pruned_loss=0.1045, over 5675607.66 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3779, pruned_loss=0.1403, over 5688657.24 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3293, pruned_loss=0.09937, over 5678446.19 frames. ], batch size: 705, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:43:57,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:44:05,570 INFO [optim.py:369] (1/2) 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 09:44:18,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4292, 1.4951, 1.3347, 1.5145], device='cuda:1'), covar=tensor([0.2147, 0.2073, 0.2099, 0.2195], device='cuda:1'), in_proj_covar=tensor([0.1078, 0.0841, 0.0966, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 09:44:23,215 INFO [train.py:968] (1/2) Epoch 4, batch 35400, giga_loss[loss=0.2485, simple_loss=0.3208, pruned_loss=0.08806, over 28959.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3306, pruned_loss=0.1018, over 5679057.38 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3781, pruned_loss=0.1403, over 5692773.76 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.325, pruned_loss=0.09707, over 5677417.01 frames. ], batch size: 174, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:45:02,674 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 35450, giga_loss[loss=0.2613, simple_loss=0.3283, pruned_loss=0.09712, over 27924.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3276, pruned_loss=0.09983, over 5668192.17 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3787, pruned_loss=0.1407, over 5675984.28 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3219, pruned_loss=0.09509, over 5682435.51 frames. ], batch size: 412, lr: 7.44e-03, grad_scale: 4.0
+2023-03-02 09:45:32,013 INFO [optim.py:369] (1/2) 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 09:45:48,267 INFO [train.py:968] (1/2) Epoch 4, batch 35500, giga_loss[loss=0.2432, simple_loss=0.3027, pruned_loss=0.09182, over 28672.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3258, pruned_loss=0.09935, over 5670017.09 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3792, pruned_loss=0.141, over 5676456.76 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3191, pruned_loss=0.09394, over 5680708.98 frames. ], batch size: 119, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:46:08,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8008, 1.6142, 1.3461, 1.3469], device='cuda:1'), covar=tensor([0.0646, 0.0618, 0.1013, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0450, 0.0513, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 09:46:31,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4385, 3.0554, 1.5117, 1.4331], device='cuda:1'), covar=tensor([0.0875, 0.0416, 0.0845, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0462, 0.0307, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 09:46:32,287 INFO [train.py:968] (1/2) Epoch 4, batch 35550, giga_loss[loss=0.2527, simple_loss=0.3062, pruned_loss=0.09954, over 28829.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3219, pruned_loss=0.0969, over 5676064.99 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.3793, pruned_loss=0.1411, over 5678324.31 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3155, pruned_loss=0.09175, over 5682681.24 frames. ], batch size: 119, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:46:34,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:46:53,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3447, 1.4974, 1.0590, 1.0156], device='cuda:1'), covar=tensor([0.0993, 0.0784, 0.0755, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.1321, 0.1059, 0.1055, 0.1147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 09:46:55,803 INFO [optim.py:369] (1/2) 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 09:47:08,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:47:14,401 INFO [train.py:968] (1/2) Epoch 4, batch 35600, giga_loss[loss=0.1965, simple_loss=0.2712, pruned_loss=0.06087, over 29076.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3202, pruned_loss=0.09644, over 5666881.52 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3798, pruned_loss=0.1413, over 5677666.65 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3124, pruned_loss=0.09045, over 5672323.78 frames. ], batch size: 136, lr: 7.43e-03, grad_scale: 8.0
+2023-03-02 09:47:56,318 INFO [train.py:968] (1/2) Epoch 4, batch 35650, giga_loss[loss=0.3153, simple_loss=0.3717, pruned_loss=0.1295, over 28970.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3232, pruned_loss=0.09861, over 5674219.84 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3807, pruned_loss=0.1417, over 5684479.81 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3145, pruned_loss=0.09225, over 5672343.98 frames. ], batch size: 136, lr: 7.43e-03, grad_scale: 8.0
+2023-03-02 09:48:24,101 INFO [optim.py:369] (1/2) 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 09:48:40,135 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=171839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:48:42,944 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 35700, libri_loss[loss=0.3379, simple_loss=0.3937, pruned_loss=0.141, over 29542.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3361, pruned_loss=0.1051, over 5683629.16 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.381, pruned_loss=0.1418, over 5684462.08 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3283, pruned_loss=0.09945, over 5681980.30 frames. ], batch size: 79, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:48:59,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:49:09,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=171871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:49:24,730 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=171891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:49:25,928 INFO [train.py:968] (1/2) Epoch 4, batch 35750, giga_loss[loss=0.336, simple_loss=0.4015, pruned_loss=0.1353, over 28670.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3506, pruned_loss=0.1137, over 5677040.03 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3811, pruned_loss=0.1416, over 5678942.48 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3429, pruned_loss=0.1082, over 5680212.27 frames. ], batch size: 78, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:49:49,925 INFO [optim.py:369] (1/2) 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 09:50:06,761 INFO [train.py:968] (1/2) Epoch 4, batch 35800, giga_loss[loss=0.32, simple_loss=0.3946, pruned_loss=0.1227, over 28874.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3608, pruned_loss=0.1194, over 5673290.74 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3817, pruned_loss=0.142, over 5675391.60 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3531, pruned_loss=0.1137, over 5679799.51 frames. ], batch size: 186, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:50:24,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=171964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:50:49,055 INFO [train.py:968] (1/2) Epoch 4, batch 35850, giga_loss[loss=0.2928, simple_loss=0.3683, pruned_loss=0.1086, over 28597.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3671, pruned_loss=0.1216, over 5660535.50 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3826, pruned_loss=0.1426, over 5660476.10 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3598, pruned_loss=0.116, over 5677601.67 frames. ], batch size: 78, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:50:59,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:51:02,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:51:15,794 INFO [optim.py:369] (1/2) 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 09:51:30,779 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 35900, giga_loss[loss=0.3015, simple_loss=0.3729, pruned_loss=0.115, over 28691.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5651639.57 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3829, pruned_loss=0.1428, over 5652767.43 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3622, pruned_loss=0.116, over 5671147.98 frames. ], batch size: 262, lr: 7.43e-03, grad_scale: 4.0
+2023-03-02 09:51:42,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4194, 1.5815, 1.3906, 1.5131], device='cuda:1'), covar=tensor([0.0804, 0.0317, 0.0319, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0128, 0.0132, 0.0242], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:1')
+2023-03-02 09:52:20,689 INFO [train.py:968] (1/2) Epoch 4, batch 35950, giga_loss[loss=0.3453, simple_loss=0.4097, pruned_loss=0.1404, over 28602.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1217, over 5656897.03 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3834, pruned_loss=0.1432, over 5660239.46 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3645, pruned_loss=0.1167, over 5665889.95 frames. ], batch size: 242, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:52:32,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:52:35,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:52:37,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:52:45,534 INFO [optim.py:369] (1/2) 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 09:53:01,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:53:04,184 INFO [train.py:968] (1/2) Epoch 4, batch 36000, libri_loss[loss=0.3691, simple_loss=0.4126, pruned_loss=0.1627, over 29520.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1231, over 5674744.28 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3837, pruned_loss=0.1433, over 5666269.04 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 5676505.55 frames. ], batch size: 84, lr: 7.42e-03, grad_scale: 8.0
+2023-03-02 09:53:04,185 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 09:53:12,630 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 09:53:53,546 INFO [train.py:968] (1/2) Epoch 4, batch 36050, giga_loss[loss=0.2966, simple_loss=0.3668, pruned_loss=0.1132, over 29077.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5678397.91 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3838, pruned_loss=0.1431, over 5671975.55 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1209, over 5674775.94 frames. ], batch size: 155, lr: 7.42e-03, grad_scale: 8.0
+2023-03-02 09:53:59,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-02 09:54:19,612 INFO [optim.py:369] (1/2) 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 09:54:37,933 INFO [train.py:968] (1/2) Epoch 4, batch 36100, giga_loss[loss=0.3058, simple_loss=0.3784, pruned_loss=0.1166, over 28971.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3777, pruned_loss=0.1259, over 5687754.18 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.384, pruned_loss=0.1432, over 5674517.03 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3738, pruned_loss=0.1226, over 5682836.07 frames. ], batch size: 213, lr: 7.42e-03, grad_scale: 8.0
+2023-03-02 09:54:49,449 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:54:51,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:54:56,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:55:15,505 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 36150, giga_loss[loss=0.348, simple_loss=0.4081, pruned_loss=0.144, over 28680.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3797, pruned_loss=0.1255, over 5700963.30 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3847, pruned_loss=0.1436, over 5674621.06 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3759, pruned_loss=0.122, over 5697184.93 frames. ], batch size: 262, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:55:40,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-02 09:55:46,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2361, 1.3482, 1.4198, 1.2987], device='cuda:1'), covar=tensor([0.1211, 0.1173, 0.1279, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0761, 0.0636, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 09:55:47,437 INFO [optim.py:369] (1/2) 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 09:55:52,622 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 36200, giga_loss[loss=0.3311, simple_loss=0.4004, pruned_loss=0.1309, over 28897.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3825, pruned_loss=0.1271, over 5699740.51 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3855, pruned_loss=0.144, over 5680860.45 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3785, pruned_loss=0.1235, over 5691309.66 frames. ], batch size: 174, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:56:43,364 INFO [train.py:968] (1/2) Epoch 4, batch 36250, giga_loss[loss=0.2844, simple_loss=0.3651, pruned_loss=0.1019, over 28576.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3826, pruned_loss=0.1261, over 5700427.36 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3856, pruned_loss=0.1439, over 5682603.35 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3792, pruned_loss=0.1228, over 5692866.32 frames. ], batch size: 85, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:56:57,850 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:57:00,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:57:10,645 INFO [optim.py:369] (1/2) 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 09:57:23,510 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 36300, giga_loss[loss=0.3104, simple_loss=0.3829, pruned_loss=0.119, over 28656.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3817, pruned_loss=0.1242, over 5699902.77 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3864, pruned_loss=0.1444, over 5681440.69 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3782, pruned_loss=0.1209, over 5695229.36 frames. ], batch size: 336, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:57:32,262 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8355, 1.1240, 3.5206, 2.9228], device='cuda:1'), covar=tensor([0.1613, 0.2176, 0.0364, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0550, 0.0510, 0.0713, 0.0572], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 09:58:07,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8278, 1.7213, 1.6003, 1.6187], device='cuda:1'), covar=tensor([0.1157, 0.2055, 0.1620, 0.1651], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0754, 0.0633, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 09:58:08,465 INFO [train.py:968] (1/2) Epoch 4, batch 36350, giga_loss[loss=0.2942, simple_loss=0.3681, pruned_loss=0.1102, over 28764.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3781, pruned_loss=0.1207, over 5695734.59 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3866, pruned_loss=0.1445, over 5679092.49 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3752, pruned_loss=0.1179, over 5694197.15 frames. ], batch size: 99, lr: 7.42e-03, grad_scale: 4.0
+2023-03-02 09:58:23,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 09:58:34,359 INFO [optim.py:369] (1/2) 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 09:58:49,000 INFO [train.py:968] (1/2) Epoch 4, batch 36400, giga_loss[loss=0.3465, simple_loss=0.4029, pruned_loss=0.145, over 28536.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3794, pruned_loss=0.1226, over 5695283.89 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3874, pruned_loss=0.1452, over 5686279.31 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3761, pruned_loss=0.1191, over 5687774.39 frames. ], batch size: 336, lr: 7.41e-03, grad_scale: 8.0
+2023-03-02 09:58:51,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2486, 1.6660, 1.1670, 0.5548], device='cuda:1'), covar=tensor([0.1475, 0.0809, 0.1298, 0.2038], device='cuda:1'), in_proj_covar=tensor([0.1354, 0.1257, 0.1355, 0.1111], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 09:59:26,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-02 09:59:29,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5528, 1.4636, 1.1513, 1.2225], device='cuda:1'), covar=tensor([0.0662, 0.0554, 0.1036, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0450, 0.0508, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 09:59:33,551 INFO [train.py:968] (1/2) Epoch 4, batch 36450, giga_loss[loss=0.3199, simple_loss=0.3742, pruned_loss=0.1328, over 28833.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3821, pruned_loss=0.127, over 5693262.38 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3878, pruned_loss=0.1454, over 5691159.69 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3789, pruned_loss=0.1236, over 5683050.63 frames. ], batch size: 112, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:00:00,180 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 36500, libri_loss[loss=0.3611, simple_loss=0.4153, pruned_loss=0.1535, over 29550.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3844, pruned_loss=0.1308, over 5691971.50 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3887, pruned_loss=0.146, over 5685620.82 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.381, pruned_loss=0.1271, over 5688216.64 frames. ], batch size: 79, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:00:39,623 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 10:00:59,222 INFO [train.py:968] (1/2) Epoch 4, batch 36550, giga_loss[loss=0.3527, simple_loss=0.4015, pruned_loss=0.152, over 28574.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3844, pruned_loss=0.1323, over 5687871.08 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.3892, pruned_loss=0.1462, over 5687313.82 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3811, pruned_loss=0.1288, over 5683485.67 frames. ], batch size: 336, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:01:08,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:01:21,772 INFO [optim.py:369] (1/2) 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 10:01:37,285 INFO [train.py:968] (1/2) Epoch 4, batch 36600, giga_loss[loss=0.3192, simple_loss=0.3837, pruned_loss=0.1274, over 28832.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.384, pruned_loss=0.1329, over 5699733.04 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3902, pruned_loss=0.1469, over 5693689.26 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3801, pruned_loss=0.1288, over 5690582.15 frames. ], batch size: 174, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:01:53,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0522, 0.9246, 0.7691, 1.3538], device='cuda:1'), covar=tensor([0.0812, 0.0347, 0.0366, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0128, 0.0132, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0051, 0.0038, 0.0034, 0.0057], device='cuda:1')
+2023-03-02 10:02:18,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2421, 3.0565, 2.9785, 1.2576], device='cuda:1'), covar=tensor([0.0732, 0.0612, 0.0967, 0.2249], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0702, 0.0783, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 10:02:23,163 INFO [train.py:968] (1/2) Epoch 4, batch 36650, giga_loss[loss=0.3046, simple_loss=0.3713, pruned_loss=0.1189, over 28758.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3823, pruned_loss=0.1318, over 5699615.50 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3907, pruned_loss=0.1472, over 5693619.27 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3787, pruned_loss=0.1281, over 5692641.23 frames. ], batch size: 242, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:02:51,264 INFO [optim.py:369] (1/2) 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 10:03:08,715 INFO [train.py:968] (1/2) Epoch 4, batch 36700, giga_loss[loss=0.3349, simple_loss=0.3965, pruned_loss=0.1366, over 28856.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3798, pruned_loss=0.129, over 5696388.77 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3908, pruned_loss=0.1473, over 5694658.17 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3769, pruned_loss=0.126, over 5690097.17 frames. ], batch size: 227, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:03:13,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=172848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:16,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=172851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:37,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5654, 1.7459, 1.5169, 1.8982], device='cuda:1'), covar=tensor([0.2000, 0.1772, 0.1743, 0.1656], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.0846, 0.0974, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 10:03:43,060 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=172880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:48,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=172886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:03:55,174 INFO [train.py:968] (1/2) Epoch 4, batch 36750, giga_loss[loss=0.2832, simple_loss=0.3335, pruned_loss=0.1164, over 23434.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3765, pruned_loss=0.126, over 5695529.27 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.3914, pruned_loss=0.1477, over 5691404.27 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3735, pruned_loss=0.123, over 5693551.07 frames. ], batch size: 705, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:04:26,361 INFO [optim.py:369] (1/2) 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 10:04:43,048 INFO [train.py:968] (1/2) Epoch 4, batch 36800, giga_loss[loss=0.271, simple_loss=0.3407, pruned_loss=0.1007, over 28858.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.371, pruned_loss=0.1228, over 5697002.04 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3913, pruned_loss=0.1476, over 5694313.30 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.12, over 5692791.36 frames. ], batch size: 199, lr: 7.41e-03, grad_scale: 8.0
+2023-03-02 10:05:23,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5999, 1.9611, 1.9264, 1.7566], device='cuda:1'), covar=tensor([0.1668, 0.1792, 0.1153, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0756, 0.0768, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 10:05:31,452 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=172991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:05:32,399 INFO [train.py:968] (1/2) Epoch 4, batch 36850, giga_loss[loss=0.2633, simple_loss=0.3133, pruned_loss=0.1067, over 23342.00 frames. ], tot_loss[loss=0.302, simple_loss=0.365, pruned_loss=0.1195, over 5681028.03 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3926, pruned_loss=0.1486, over 5689596.61 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3612, pruned_loss=0.1157, over 5681644.34 frames. ], batch size: 705, lr: 7.41e-03, grad_scale: 4.0
+2023-03-02 10:06:00,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:06:09,078 INFO [optim.py:369] (1/2) 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 10:06:12,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:06:15,895 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 36900, giga_loss[loss=0.2428, simple_loss=0.3188, pruned_loss=0.0834, over 28499.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3587, pruned_loss=0.1154, over 5673946.60 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.393, pruned_loss=0.1488, over 5692007.24 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3546, pruned_loss=0.1117, over 5672120.64 frames. ], batch size: 71, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:06:40,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:06:40,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-02 10:07:08,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9779, 3.7705, 3.6776, 1.6613], device='cuda:1'), covar=tensor([0.0418, 0.0408, 0.0645, 0.2268], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0702, 0.0771, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 10:07:10,327 INFO [train.py:968] (1/2) Epoch 4, batch 36950, giga_loss[loss=0.2981, simple_loss=0.3623, pruned_loss=0.117, over 28890.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3593, pruned_loss=0.1155, over 5679219.27 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3929, pruned_loss=0.1487, over 5696083.07 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3554, pruned_loss=0.1119, over 5673945.75 frames. ], batch size: 106, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:07:16,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5992, 2.1145, 1.5233, 1.8562], device='cuda:1'), covar=tensor([0.0784, 0.0272, 0.0316, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0129, 0.0133, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:1')
+2023-03-02 10:07:38,045 INFO [optim.py:369] (1/2) 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 10:07:38,912 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:07:50,982 INFO [train.py:968] (1/2) Epoch 4, batch 37000, giga_loss[loss=0.2598, simple_loss=0.3293, pruned_loss=0.09519, over 28999.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3593, pruned_loss=0.1152, over 5691612.04 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3932, pruned_loss=0.1487, over 5696410.50 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3554, pruned_loss=0.1117, over 5687084.51 frames. ], batch size: 213, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:08:19,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3968, 1.3680, 1.4600, 1.3341], device='cuda:1'), covar=tensor([0.0959, 0.1420, 0.1447, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0750, 0.0640, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 10:08:29,227 INFO [train.py:968] (1/2) Epoch 4, batch 37050, giga_loss[loss=0.3477, simple_loss=0.3932, pruned_loss=0.1511, over 26795.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3614, pruned_loss=0.1172, over 5695094.60 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.394, pruned_loss=0.1491, over 5702625.45 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3555, pruned_loss=0.1122, over 5685797.87 frames. ], batch size: 555, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:08:55,465 INFO [optim.py:369] (1/2) 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 10:09:04,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:09:06,108 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-02 10:09:07,743 INFO [train.py:968] (1/2) Epoch 4, batch 37100, giga_loss[loss=0.2351, simple_loss=0.3095, pruned_loss=0.08034, over 28827.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3591, pruned_loss=0.116, over 5683222.15 frames. ], libri_tot_loss[loss=0.3472, simple_loss=0.3951, pruned_loss=0.1497, over 5689296.15 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3524, pruned_loss=0.1104, over 5688546.63 frames. ], batch size: 66, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:09:49,807 INFO [train.py:968] (1/2) Epoch 4, batch 37150, giga_loss[loss=0.2377, simple_loss=0.3119, pruned_loss=0.08175, over 28915.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3561, pruned_loss=0.1142, over 5696802.90 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.3958, pruned_loss=0.1501, over 5691691.67 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.35, pruned_loss=0.1092, over 5698880.23 frames. ], batch size: 145, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:10:14,076 INFO [optim.py:369] (1/2) 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 10:10:27,416 INFO [train.py:968] (1/2) Epoch 4, batch 37200, giga_loss[loss=0.2442, simple_loss=0.3179, pruned_loss=0.08521, over 28711.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3548, pruned_loss=0.1135, over 5700109.89 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.3963, pruned_loss=0.15, over 5689553.11 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.348, pruned_loss=0.1083, over 5704274.45 frames. ], batch size: 92, lr: 7.40e-03, grad_scale: 8.0
+2023-03-02 10:10:44,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:10:57,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2096, 1.6715, 1.1928, 0.5781], device='cuda:1'), covar=tensor([0.2207, 0.1121, 0.1597, 0.2622], device='cuda:1'), in_proj_covar=tensor([0.1340, 0.1243, 0.1341, 0.1098], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 10:11:06,941 INFO [train.py:968] (1/2) Epoch 4, batch 37250, giga_loss[loss=0.3793, simple_loss=0.4174, pruned_loss=0.1706, over 28904.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3527, pruned_loss=0.1127, over 5701464.23 frames. ], libri_tot_loss[loss=0.3485, simple_loss=0.3966, pruned_loss=0.1501, over 5693403.77 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3462, pruned_loss=0.1077, over 5701527.00 frames. ], batch size: 145, lr: 7.40e-03, grad_scale: 8.0
+2023-03-02 10:11:07,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:11:33,890 INFO [optim.py:369] (1/2) 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 10:11:48,447 INFO [train.py:968] (1/2) Epoch 4, batch 37300, giga_loss[loss=0.2603, simple_loss=0.3278, pruned_loss=0.09636, over 28726.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3501, pruned_loss=0.1114, over 5692018.06 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.3972, pruned_loss=0.1504, over 5683486.44 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3438, pruned_loss=0.1067, over 5701946.63 frames. ], batch size: 284, lr: 7.40e-03, grad_scale: 4.0
+2023-03-02 10:12:06,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9445, 1.0760, 3.8570, 3.1228], device='cuda:1'), covar=tensor([0.1677, 0.2376, 0.0335, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0509, 0.0704, 0.0564], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 10:12:27,961 INFO [train.py:968] (1/2) Epoch 4, batch 37350, giga_loss[loss=0.2809, simple_loss=0.3413, pruned_loss=0.1102, over 28629.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3465, pruned_loss=0.1091, over 5706503.86 frames. ], libri_tot_loss[loss=0.3494, simple_loss=0.3976, pruned_loss=0.1506, over 5686914.13 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3405, pruned_loss=0.1046, over 5711520.01 frames. ], batch size: 307, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:12:34,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:12:39,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:12:42,908 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:12:54,920 INFO [optim.py:369] (1/2) 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 10:13:01,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:03,923 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:06,358 INFO [train.py:968] (1/2) Epoch 4, batch 37400, giga_loss[loss=0.2597, simple_loss=0.3239, pruned_loss=0.09774, over 28542.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3454, pruned_loss=0.1082, over 5716096.80 frames. ], libri_tot_loss[loss=0.3509, simple_loss=0.3992, pruned_loss=0.1512, over 5691576.60 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3378, pruned_loss=0.1029, over 5716633.62 frames. ], batch size: 71, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:13:06,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:09,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5466, 3.2650, 1.6909, 1.5753], device='cuda:1'), covar=tensor([0.0887, 0.0338, 0.0790, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0453, 0.0303, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0022, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 10:13:27,162 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:13:39,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-02 10:13:46,307 INFO [train.py:968] (1/2) Epoch 4, batch 37450, giga_loss[loss=0.2813, simple_loss=0.3418, pruned_loss=0.1104, over 28572.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3445, pruned_loss=0.1076, over 5717429.73 frames. ], libri_tot_loss[loss=0.3513, simple_loss=0.3997, pruned_loss=0.1515, over 5687587.47 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3365, pruned_loss=0.1021, over 5722280.93 frames. ], batch size: 85, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:14:05,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:14:18,747 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 37500, giga_loss[loss=0.2817, simple_loss=0.3451, pruned_loss=0.1092, over 28891.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3437, pruned_loss=0.1071, over 5715478.52 frames. ], libri_tot_loss[loss=0.3512, simple_loss=0.3997, pruned_loss=0.1514, over 5688697.53 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3371, pruned_loss=0.1026, over 5718456.90 frames. ], batch size: 174, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:14:33,361 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:14:35,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:15:01,891 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:15:14,247 INFO [train.py:968] (1/2) Epoch 4, batch 37550, giga_loss[loss=0.3254, simple_loss=0.3847, pruned_loss=0.133, over 29006.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3504, pruned_loss=0.1119, over 5711269.36 frames. ], libri_tot_loss[loss=0.3519, simple_loss=0.4004, pruned_loss=0.1517, over 5690478.24 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3431, pruned_loss=0.1068, over 5712395.50 frames. ], batch size: 128, lr: 7.39e-03, grad_scale: 4.0
+2023-03-02 10:15:22,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-03-02 10:15:22,874 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:15:29,253 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-02 10:15:41,110 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 37600, giga_loss[loss=0.3274, simple_loss=0.3875, pruned_loss=0.1337, over 28871.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3569, pruned_loss=0.1164, over 5701833.95 frames. ], libri_tot_loss[loss=0.3518, simple_loss=0.4004, pruned_loss=0.1516, over 5688013.18 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3496, pruned_loss=0.1112, over 5705854.93 frames. ], batch size: 227, lr: 7.39e-03, grad_scale: 8.0
+2023-03-02 10:16:11,682 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=173756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:16:13,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=173759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:16:44,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=173788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:16:47,990 INFO [train.py:968] (1/2) Epoch 4, batch 37650, giga_loss[loss=0.2967, simple_loss=0.3632, pruned_loss=0.1151, over 28749.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3658, pruned_loss=0.1222, over 5696232.48 frames. ], libri_tot_loss[loss=0.3518, simple_loss=0.4005, pruned_loss=0.1515, over 5690338.73 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3593, pruned_loss=0.1178, over 5697597.29 frames. ], batch size: 99, lr: 7.39e-03, grad_scale: 8.0
+2023-03-02 10:17:21,367 INFO [optim.py:369] (1/2) 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 10:17:35,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4712, 2.4806, 1.5299, 1.4434], device='cuda:1'), covar=tensor([0.0669, 0.0301, 0.0654, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0456, 0.0302, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 10:17:37,386 INFO [train.py:968] (1/2) Epoch 4, batch 37700, giga_loss[loss=0.2822, simple_loss=0.3541, pruned_loss=0.1051, over 28439.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3721, pruned_loss=0.1264, over 5685094.32 frames. ], libri_tot_loss[loss=0.353, simple_loss=0.4014, pruned_loss=0.1523, over 5698697.07 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3648, pruned_loss=0.121, over 5678242.39 frames. ], batch size: 78, lr: 7.39e-03, grad_scale: 2.0
+2023-03-02 10:18:13,222 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=173884.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 10:18:20,283 INFO [train.py:968] (1/2) Epoch 4, batch 37750, giga_loss[loss=0.3947, simple_loss=0.4362, pruned_loss=0.1766, over 27849.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3772, pruned_loss=0.1286, over 5690712.68 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4013, pruned_loss=0.1522, over 5703939.04 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3709, pruned_loss=0.1238, over 5680211.04 frames. ], batch size: 412, lr: 7.39e-03, grad_scale: 2.0
+2023-03-02 10:18:40,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1781, 2.9741, 2.8886, 1.3304], device='cuda:1'), covar=tensor([0.0782, 0.0707, 0.0989, 0.2443], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0703, 0.0784, 0.0582], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 10:18:43,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=173918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:18:52,150 INFO [optim.py:369] (1/2) 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 10:19:06,418 INFO [train.py:968] (1/2) Epoch 4, batch 37800, giga_loss[loss=0.5075, simple_loss=0.4903, pruned_loss=0.2623, over 23466.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3815, pruned_loss=0.1309, over 5681079.15 frames. ], libri_tot_loss[loss=0.3522, simple_loss=0.4009, pruned_loss=0.1517, over 5706800.29 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3759, pruned_loss=0.1266, over 5669366.05 frames. ], batch size: 705, lr: 7.39e-03, grad_scale: 2.0
+2023-03-02 10:19:48,885 INFO [train.py:968] (1/2) Epoch 4, batch 37850, giga_loss[loss=0.3094, simple_loss=0.3732, pruned_loss=0.1228, over 28602.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3849, pruned_loss=0.1334, over 5670140.12 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4014, pruned_loss=0.1522, over 5697677.29 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.38, pruned_loss=0.1294, over 5669589.34 frames. ], batch size: 85, lr: 7.38e-03, grad_scale: 2.0
+2023-03-02 10:19:56,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1361, 3.9955, 2.2700, 2.2956], device='cuda:1'), covar=tensor([0.0536, 0.0257, 0.0537, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0454, 0.0301, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0019], device='cuda:1')
+2023-03-02 10:20:00,564 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:20:17,784 INFO [optim.py:369] (1/2) 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 10:20:29,970 INFO [train.py:968] (1/2) Epoch 4, batch 37900, giga_loss[loss=0.2855, simple_loss=0.36, pruned_loss=0.1055, over 28876.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3773, pruned_loss=0.1277, over 5670755.69 frames. ], libri_tot_loss[loss=0.3527, simple_loss=0.4011, pruned_loss=0.1522, over 5690784.36 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3731, pruned_loss=0.1241, over 5675245.37 frames. ], batch size: 112, lr: 7.38e-03, grad_scale: 2.0
+2023-03-02 10:20:44,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:20:46,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5486, 2.0682, 1.7787, 1.7806], device='cuda:1'), covar=tensor([0.0606, 0.0701, 0.0830, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0447, 0.0507, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 10:20:46,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:21:01,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:21:13,430 INFO [train.py:968] (1/2) Epoch 4, batch 37950, giga_loss[loss=0.3971, simple_loss=0.4382, pruned_loss=0.178, over 28852.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3753, pruned_loss=0.1252, over 5681251.26 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.4015, pruned_loss=0.1525, over 5692909.93 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3711, pruned_loss=0.1214, over 5682724.88 frames. ], batch size: 145, lr: 7.38e-03, grad_scale: 2.0
+2023-03-02 10:21:14,324 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:21:44,289 INFO [optim.py:369] (1/2) 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 10:21:55,433 INFO [train.py:968] (1/2) Epoch 4, batch 38000, giga_loss[loss=0.2939, simple_loss=0.3715, pruned_loss=0.1081, over 28841.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3753, pruned_loss=0.1247, over 5669554.46 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4018, pruned_loss=0.1527, over 5684664.34 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3709, pruned_loss=0.1209, over 5678426.71 frames. ], batch size: 174, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:22:10,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1626, 1.4016, 1.2357, 1.3618], device='cuda:1'), covar=tensor([0.0827, 0.0347, 0.0326, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0129, 0.0132, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0058], device='cuda:1')
+2023-03-02 10:22:40,355 INFO [train.py:968] (1/2) Epoch 4, batch 38050, giga_loss[loss=0.309, simple_loss=0.3717, pruned_loss=0.1231, over 28956.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3773, pruned_loss=0.1256, over 5674575.03 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4019, pruned_loss=0.1527, over 5685767.01 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3737, pruned_loss=0.1224, over 5680558.12 frames. ], batch size: 213, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:23:01,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:23:04,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:23:08,860 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 10:23:09,626 INFO [optim.py:369] (1/2) 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 10:23:21,058 INFO [train.py:968] (1/2) Epoch 4, batch 38100, giga_loss[loss=0.3156, simple_loss=0.3753, pruned_loss=0.128, over 28825.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3822, pruned_loss=0.1293, over 5683009.44 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4028, pruned_loss=0.153, over 5691177.12 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3778, pruned_loss=0.1256, over 5682720.37 frames. ], batch size: 119, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:23:30,093 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:23:35,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 10:24:07,141 INFO [train.py:968] (1/2) Epoch 4, batch 38150, giga_loss[loss=0.3219, simple_loss=0.3779, pruned_loss=0.1329, over 28465.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3831, pruned_loss=0.1301, over 5690236.88 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.403, pruned_loss=0.1531, over 5693772.91 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3793, pruned_loss=0.1269, over 5687492.93 frames. ], batch size: 71, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:24:40,630 INFO [optim.py:369] (1/2) 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 10:24:52,958 INFO [train.py:968] (1/2) Epoch 4, batch 38200, giga_loss[loss=0.4114, simple_loss=0.4344, pruned_loss=0.1942, over 26763.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3839, pruned_loss=0.1308, over 5694724.53 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.4035, pruned_loss=0.1533, over 5697439.12 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.38, pruned_loss=0.1275, over 5689227.13 frames. ], batch size: 555, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:25:21,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3413, 4.0874, 4.0368, 1.9162], device='cuda:1'), covar=tensor([0.0418, 0.0372, 0.0641, 0.1872], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0706, 0.0782, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 10:25:28,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9143, 2.3925, 2.3455, 2.0628], device='cuda:1'), covar=tensor([0.1491, 0.1488, 0.0954, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0757, 0.0771, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 10:25:28,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:25:35,900 INFO [train.py:968] (1/2) Epoch 4, batch 38250, giga_loss[loss=0.2723, simple_loss=0.3559, pruned_loss=0.09438, over 29020.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3848, pruned_loss=0.1317, over 5688217.85 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4042, pruned_loss=0.1538, over 5689953.60 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3806, pruned_loss=0.1284, over 5690199.52 frames. ], batch size: 164, lr: 7.38e-03, grad_scale: 4.0
+2023-03-02 10:25:44,679 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 10:25:47,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4072, 4.9993, 5.0191, 1.9070], device='cuda:1'), covar=tensor([0.0336, 0.0393, 0.0625, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0715, 0.0790, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 10:25:47,501 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174405.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 10:25:49,301 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:25:51,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:26:08,160 INFO [optim.py:369] (1/2) 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 10:26:11,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 10:26:19,920 INFO [train.py:968] (1/2) Epoch 4, batch 38300, giga_loss[loss=0.283, simple_loss=0.3615, pruned_loss=0.1023, over 28574.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3839, pruned_loss=0.1302, over 5684681.94 frames. ], libri_tot_loss[loss=0.3559, simple_loss=0.4043, pruned_loss=0.1538, over 5681614.96 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3804, pruned_loss=0.1274, over 5693202.18 frames. ], batch size: 60, lr: 7.37e-03, grad_scale: 4.0
+2023-03-02 10:26:40,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4706, 1.9112, 1.6981, 1.6179], device='cuda:1'), covar=tensor([0.1447, 0.1834, 0.1154, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0757, 0.0776, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 10:26:59,157 INFO [train.py:968] (1/2) Epoch 4, batch 38350, giga_loss[loss=0.3353, simple_loss=0.408, pruned_loss=0.1313, over 27903.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3828, pruned_loss=0.1277, over 5687747.35 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4047, pruned_loss=0.154, over 5676426.10 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3794, pruned_loss=0.1248, over 5700136.38 frames. ], batch size: 412, lr: 7.37e-03, grad_scale: 4.0
+2023-03-02 10:27:12,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 10:27:27,637 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:27:29,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:27:29,993 INFO [optim.py:369] (1/2) 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 10:27:32,420 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:27:39,667 INFO [train.py:968] (1/2) Epoch 4, batch 38400, giga_loss[loss=0.2859, simple_loss=0.356, pruned_loss=0.1079, over 28852.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3836, pruned_loss=0.1275, over 5689939.05 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4048, pruned_loss=0.1542, over 5677855.00 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.38, pruned_loss=0.1242, over 5698883.32 frames. ], batch size: 112, lr: 7.37e-03, grad_scale: 8.0
+2023-03-02 10:27:50,446 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:28:19,592 INFO [train.py:968] (1/2) Epoch 4, batch 38450, giga_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 28647.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3831, pruned_loss=0.1274, over 5694479.38 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4057, pruned_loss=0.155, over 5675561.13 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3785, pruned_loss=0.123, over 5705074.39 frames. ], batch size: 307, lr: 7.37e-03, grad_scale: 4.0
+2023-03-02 10:28:53,150 INFO [optim.py:369] (1/2) 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 10:29:03,498 INFO [train.py:968] (1/2) Epoch 4, batch 38500, giga_loss[loss=0.291, simple_loss=0.3669, pruned_loss=0.1076, over 28998.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3798, pruned_loss=0.1256, over 5695512.22 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4056, pruned_loss=0.1551, over 5675393.93 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3761, pruned_loss=0.1218, over 5704145.50 frames. ], batch size: 145, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:29:27,207 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1736, 1.3523, 1.1101, 1.4736], device='cuda:1'), covar=tensor([0.2597, 0.2269, 0.2291, 0.2146], device='cuda:1'), in_proj_covar=tensor([0.1078, 0.0846, 0.0970, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 10:29:30,603 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:29:38,270 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:29:44,330 INFO [train.py:968] (1/2) Epoch 4, batch 38550, giga_loss[loss=0.295, simple_loss=0.3668, pruned_loss=0.1116, over 28872.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3773, pruned_loss=0.124, over 5700732.36 frames. ], libri_tot_loss[loss=0.3577, simple_loss=0.4055, pruned_loss=0.155, over 5673822.28 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3736, pruned_loss=0.1202, over 5710679.29 frames. ], batch size: 186, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:30:14,859 INFO [optim.py:369] (1/2) 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 10:30:24,427 INFO [train.py:968] (1/2) Epoch 4, batch 38600, giga_loss[loss=0.3582, simple_loss=0.4141, pruned_loss=0.1511, over 28644.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3772, pruned_loss=0.1245, over 5704116.74 frames. ], libri_tot_loss[loss=0.3577, simple_loss=0.4055, pruned_loss=0.155, over 5680081.94 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3733, pruned_loss=0.1204, over 5707387.37 frames. ], batch size: 336, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:30:50,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0909, 1.7504, 1.8212, 1.7071], device='cuda:1'), covar=tensor([0.1196, 0.2161, 0.1524, 0.1516], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0751, 0.0635, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 10:30:56,878 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=174781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:30:59,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:31:00,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:31:06,997 INFO [train.py:968] (1/2) Epoch 4, batch 38650, giga_loss[loss=0.2946, simple_loss=0.3639, pruned_loss=0.1127, over 28959.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3791, pruned_loss=0.1262, over 5711368.25 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4057, pruned_loss=0.1551, over 5683621.01 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3755, pruned_loss=0.1224, over 5711068.72 frames. ], batch size: 106, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:31:16,684 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 10:31:37,725 INFO [optim.py:369] (1/2) 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 10:31:45,773 INFO [train.py:968] (1/2) Epoch 4, batch 38700, giga_loss[loss=0.2931, simple_loss=0.3657, pruned_loss=0.1102, over 28544.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3791, pruned_loss=0.1258, over 5696280.44 frames. ], libri_tot_loss[loss=0.3584, simple_loss=0.406, pruned_loss=0.1554, over 5667407.31 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3752, pruned_loss=0.1217, over 5712383.91 frames. ], batch size: 60, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:32:23,786 INFO [train.py:968] (1/2) Epoch 4, batch 38750, giga_loss[loss=0.2952, simple_loss=0.3673, pruned_loss=0.1115, over 28859.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3781, pruned_loss=0.1243, over 5697248.42 frames. ], libri_tot_loss[loss=0.3583, simple_loss=0.4058, pruned_loss=0.1553, over 5673582.04 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3741, pruned_loss=0.12, over 5705922.26 frames. ], batch size: 106, lr: 7.37e-03, grad_scale: 2.0
+2023-03-02 10:32:34,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=174907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:46,843 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-02 10:32:48,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:50,678 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=174928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:51,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:32:53,874 INFO [optim.py:369] (1/2) 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 10:32:54,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=174931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:33:02,854 INFO [train.py:968] (1/2) Epoch 4, batch 38800, giga_loss[loss=0.2893, simple_loss=0.3659, pruned_loss=0.1064, over 28664.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3773, pruned_loss=0.1238, over 5701535.83 frames. ], libri_tot_loss[loss=0.3585, simple_loss=0.4061, pruned_loss=0.1555, over 5671236.69 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3731, pruned_loss=0.1193, over 5711368.07 frames. ], batch size: 85, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:33:13,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7976, 1.7223, 1.6811, 1.6566], device='cuda:1'), covar=tensor([0.1122, 0.1621, 0.1533, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0737, 0.0625, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 10:33:15,634 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:33:16,983 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=174960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:33:29,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2867, 1.2987, 1.1288, 1.6541], device='cuda:1'), covar=tensor([0.2429, 0.2177, 0.2202, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1082, 0.0843, 0.0967, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 10:33:44,221 INFO [train.py:968] (1/2) Epoch 4, batch 38850, giga_loss[loss=0.2945, simple_loss=0.369, pruned_loss=0.11, over 28503.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3753, pruned_loss=0.1224, over 5706336.67 frames. ], libri_tot_loss[loss=0.358, simple_loss=0.4057, pruned_loss=0.1551, over 5678063.18 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3714, pruned_loss=0.1183, over 5708966.59 frames. ], batch size: 65, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:34:13,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1598, 1.3091, 1.1463, 1.0933], device='cuda:1'), covar=tensor([0.1904, 0.1601, 0.1447, 0.1463], device='cuda:1'), in_proj_covar=tensor([0.1088, 0.0847, 0.0972, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 10:34:15,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0131, 2.1037, 1.6381, 1.7387], device='cuda:1'), covar=tensor([0.0704, 0.0281, 0.0326, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0128, 0.0131, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:1')
+2023-03-02 10:34:15,939 INFO [optim.py:369] (1/2) 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 10:34:25,745 INFO [train.py:968] (1/2) Epoch 4, batch 38900, giga_loss[loss=0.289, simple_loss=0.3607, pruned_loss=0.1087, over 28324.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3722, pruned_loss=0.1207, over 5704662.01 frames. ], libri_tot_loss[loss=0.3581, simple_loss=0.4058, pruned_loss=0.1552, over 5680583.29 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3684, pruned_loss=0.1168, over 5705065.19 frames. ], batch size: 368, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:34:31,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:32,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:33,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:40,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:34:58,456 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:35:05,634 INFO [train.py:968] (1/2) Epoch 4, batch 38950, giga_loss[loss=0.2445, simple_loss=0.3264, pruned_loss=0.08134, over 28313.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3683, pruned_loss=0.1188, over 5708988.52 frames. ], libri_tot_loss[loss=0.3579, simple_loss=0.4057, pruned_loss=0.155, over 5685082.20 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3649, pruned_loss=0.1151, over 5705770.63 frames. ], batch size: 71, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:35:09,948 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:35:35,251 INFO [optim.py:369] (1/2) 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 10:35:45,261 INFO [train.py:968] (1/2) Epoch 4, batch 39000, giga_loss[loss=0.3529, simple_loss=0.419, pruned_loss=0.1434, over 27928.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3696, pruned_loss=0.1202, over 5709393.54 frames. ], libri_tot_loss[loss=0.3592, simple_loss=0.4067, pruned_loss=0.1559, over 5687276.98 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5705342.63 frames. ], batch size: 412, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:35:45,261 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 10:35:53,791 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 10:36:03,566 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 39050, libri_loss[loss=0.3733, simple_loss=0.4211, pruned_loss=0.1628, over 28600.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.1221, over 5704561.69 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.407, pruned_loss=0.156, over 5690395.28 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3663, pruned_loss=0.1177, over 5698888.06 frames. ], batch size: 106, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:36:35,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:36:37,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:36:41,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7090, 1.5510, 1.3405, 1.3135], device='cuda:1'), covar=tensor([0.0575, 0.0588, 0.0872, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0447, 0.0506, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 10:36:43,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:36:45,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:37:01,256 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:37:04,947 INFO [optim.py:369] (1/2) 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 10:37:05,490 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-02 10:37:07,465 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:37:14,675 INFO [train.py:968] (1/2) Epoch 4, batch 39100, giga_loss[loss=0.2593, simple_loss=0.3319, pruned_loss=0.09332, over 28775.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1198, over 5710770.75 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.4071, pruned_loss=0.1559, over 5695460.23 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3625, pruned_loss=0.1155, over 5701883.48 frames. ], batch size: 119, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:37:31,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1539, 1.2570, 0.9481, 0.8814], device='cuda:1'), covar=tensor([0.0886, 0.0687, 0.0497, 0.0725], device='cuda:1'), in_proj_covar=tensor([0.1357, 0.1110, 0.1126, 0.1209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 10:37:47,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1881, 1.7451, 1.2559, 1.4277], device='cuda:1'), covar=tensor([0.0732, 0.0427, 0.0362, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0127, 0.0131, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0034, 0.0057], device='cuda:1')
+2023-03-02 10:37:53,930 INFO [train.py:968] (1/2) Epoch 4, batch 39150, giga_loss[loss=0.3159, simple_loss=0.3739, pruned_loss=0.1289, over 28613.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.364, pruned_loss=0.1184, over 5698753.45 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4072, pruned_loss=0.156, over 5677653.31 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.36, pruned_loss=0.1147, over 5707218.73 frames. ], batch size: 336, lr: 7.36e-03, grad_scale: 4.0
+2023-03-02 10:37:54,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4458, 2.7375, 1.5211, 1.5125], device='cuda:1'), covar=tensor([0.0625, 0.0428, 0.0650, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0458, 0.0305, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 10:37:59,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:38:02,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:38:06,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0390, 3.8438, 3.7379, 1.6580], device='cuda:1'), covar=tensor([0.0499, 0.0438, 0.0735, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0707, 0.0780, 0.0578], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-02 10:38:16,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1697, 1.5136, 1.2031, 1.2572], device='cuda:1'), covar=tensor([0.2086, 0.1899, 0.1973, 0.1664], device='cuda:1'), in_proj_covar=tensor([0.1086, 0.0852, 0.0971, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 10:38:24,237 INFO [optim.py:369] (1/2) 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 10:38:24,452 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:38:35,046 INFO [train.py:968] (1/2) Epoch 4, batch 39200, giga_loss[loss=0.4349, simple_loss=0.4464, pruned_loss=0.2117, over 26721.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3625, pruned_loss=0.1181, over 5691918.18 frames. ], libri_tot_loss[loss=0.36, simple_loss=0.4074, pruned_loss=0.1562, over 5672479.06 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3583, pruned_loss=0.1143, over 5702789.90 frames. ], batch size: 555, lr: 7.36e-03, grad_scale: 8.0
+2023-03-02 10:38:37,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-02 10:39:18,135 INFO [train.py:968] (1/2) Epoch 4, batch 39250, giga_loss[loss=0.2925, simple_loss=0.3579, pruned_loss=0.1135, over 28840.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.36, pruned_loss=0.1162, over 5700252.26 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4071, pruned_loss=0.1559, over 5677911.35 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3559, pruned_loss=0.1126, over 5704578.45 frames. ], batch size: 199, lr: 7.35e-03, grad_scale: 8.0
+2023-03-02 10:39:50,192 INFO [optim.py:369] (1/2) 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 10:40:02,984 INFO [train.py:968] (1/2) Epoch 4, batch 39300, libri_loss[loss=0.345, simple_loss=0.3838, pruned_loss=0.1531, over 27823.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.363, pruned_loss=0.1179, over 5699552.56 frames. ], libri_tot_loss[loss=0.3589, simple_loss=0.4063, pruned_loss=0.1557, over 5681738.29 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3593, pruned_loss=0.1143, over 5700186.09 frames. ], batch size: 61, lr: 7.35e-03, grad_scale: 8.0
+2023-03-02 10:40:08,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5520, 1.7619, 1.2349, 1.0361], device='cuda:1'), covar=tensor([0.1213, 0.0832, 0.0741, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.1354, 0.1103, 0.1119, 0.1190], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 10:40:10,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9391, 3.2442, 2.1668, 0.8015], device='cuda:1'), covar=tensor([0.2931, 0.0985, 0.1666, 0.3013], device='cuda:1'), in_proj_covar=tensor([0.1353, 0.1250, 0.1353, 0.1127], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 10:40:27,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:40:46,777 INFO [train.py:968] (1/2) Epoch 4, batch 39350, giga_loss[loss=0.2921, simple_loss=0.3649, pruned_loss=0.1096, over 28867.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3684, pruned_loss=0.1217, over 5687624.07 frames. ], libri_tot_loss[loss=0.3598, simple_loss=0.4069, pruned_loss=0.1563, over 5677563.67 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3639, pruned_loss=0.1173, over 5692085.95 frames. ], batch size: 186, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:41:24,399 INFO [optim.py:369] (1/2) 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 10:41:24,635 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:41:30,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4448, 1.3917, 1.1192, 1.2552], device='cuda:1'), covar=tensor([0.0611, 0.0530, 0.0960, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0449, 0.0508, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 10:41:34,130 INFO [train.py:968] (1/2) Epoch 4, batch 39400, giga_loss[loss=0.2819, simple_loss=0.3514, pruned_loss=0.1062, over 28697.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3688, pruned_loss=0.1205, over 5689679.19 frames. ], libri_tot_loss[loss=0.3597, simple_loss=0.4069, pruned_loss=0.1563, over 5678751.85 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3652, pruned_loss=0.1169, over 5692139.00 frames. ], batch size: 119, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:42:19,033 INFO [train.py:968] (1/2) Epoch 4, batch 39450, giga_loss[loss=0.3062, simple_loss=0.3735, pruned_loss=0.1194, over 27570.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 5692477.05 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.407, pruned_loss=0.1564, over 5684231.89 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5689646.62 frames. ], batch size: 472, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:42:24,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:42:37,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=175616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:42:39,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=175619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:42:50,860 INFO [optim.py:369] (1/2) 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 10:43:01,162 INFO [train.py:968] (1/2) Epoch 4, batch 39500, giga_loss[loss=0.2798, simple_loss=0.3469, pruned_loss=0.1064, over 28604.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3693, pruned_loss=0.1196, over 5703217.46 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4072, pruned_loss=0.1566, over 5689922.63 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3651, pruned_loss=0.1155, over 5696228.52 frames. ], batch size: 71, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:43:04,579 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=175648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:43:07,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8156, 2.5266, 1.3410, 1.2014], device='cuda:1'), covar=tensor([0.1379, 0.0579, 0.0825, 0.1051], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.1095, 0.1105, 0.1175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 10:43:42,580 INFO [train.py:968] (1/2) Epoch 4, batch 39550, giga_loss[loss=0.3044, simple_loss=0.374, pruned_loss=0.1174, over 28746.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3682, pruned_loss=0.1191, over 5706145.39 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4073, pruned_loss=0.1567, over 5693126.21 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3643, pruned_loss=0.1153, over 5697873.96 frames. ], batch size: 284, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:44:15,604 INFO [optim.py:369] (1/2) 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 10:44:23,830 INFO [train.py:968] (1/2) Epoch 4, batch 39600, giga_loss[loss=0.2816, simple_loss=0.3454, pruned_loss=0.1089, over 28698.00 frames. ], tot_loss[loss=0.303, simple_loss=0.368, pruned_loss=0.119, over 5717789.60 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4073, pruned_loss=0.1566, over 5695284.51 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3646, pruned_loss=0.1157, over 5709628.50 frames. ], batch size: 92, lr: 7.35e-03, grad_scale: 8.0
+2023-03-02 10:44:29,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2307, 1.3924, 1.0488, 0.8526], device='cuda:1'), covar=tensor([0.0987, 0.0782, 0.0530, 0.0736], device='cuda:1'), in_proj_covar=tensor([0.1334, 0.1089, 0.1098, 0.1170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 10:44:43,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:45:03,020 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:45:07,472 INFO [train.py:968] (1/2) Epoch 4, batch 39650, giga_loss[loss=0.3101, simple_loss=0.3816, pruned_loss=0.1193, over 29017.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3721, pruned_loss=0.1219, over 5710530.98 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4076, pruned_loss=0.1569, over 5688484.96 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3683, pruned_loss=0.1182, over 5710774.52 frames. ], batch size: 155, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:45:41,624 INFO [optim.py:369] (1/2) 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 10:45:49,918 INFO [train.py:968] (1/2) Epoch 4, batch 39700, libri_loss[loss=0.3563, simple_loss=0.4066, pruned_loss=0.153, over 29545.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3757, pruned_loss=0.1239, over 5703955.80 frames. ], libri_tot_loss[loss=0.3613, simple_loss=0.4081, pruned_loss=0.1572, over 5689392.91 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3713, pruned_loss=0.1199, over 5703511.53 frames. ], batch size: 80, lr: 7.35e-03, grad_scale: 4.0
+2023-03-02 10:45:58,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5530, 1.5264, 1.4128, 1.4738], device='cuda:1'), covar=tensor([0.1024, 0.1693, 0.1554, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0743, 0.0638, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 10:46:04,368 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 10:46:15,505 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=175874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:46:30,782 INFO [train.py:968] (1/2) Epoch 4, batch 39750, giga_loss[loss=0.3028, simple_loss=0.366, pruned_loss=0.1198, over 28564.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3759, pruned_loss=0.1236, over 5712337.71 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4081, pruned_loss=0.1572, over 5692004.61 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.372, pruned_loss=0.12, over 5709739.10 frames. ], batch size: 78, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:46:31,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2272, 1.3601, 0.8214, 1.0420], device='cuda:1'), covar=tensor([0.0936, 0.0700, 0.0608, 0.0645], device='cuda:1'), in_proj_covar=tensor([0.1361, 0.1119, 0.1128, 0.1197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 10:46:41,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:47:03,384 INFO [optim.py:369] (1/2) 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 10:47:13,322 INFO [train.py:968] (1/2) Epoch 4, batch 39800, giga_loss[loss=0.317, simple_loss=0.3865, pruned_loss=0.1237, over 28732.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3765, pruned_loss=0.1237, over 5709354.58 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.4083, pruned_loss=0.1574, over 5694204.94 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.373, pruned_loss=0.1204, over 5705642.97 frames. ], batch size: 242, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:47:41,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=175975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:47:55,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-02 10:47:55,583 INFO [train.py:968] (1/2) Epoch 4, batch 39850, giga_loss[loss=0.3839, simple_loss=0.4221, pruned_loss=0.1728, over 26611.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3781, pruned_loss=0.125, over 5709897.72 frames. ], libri_tot_loss[loss=0.3615, simple_loss=0.4083, pruned_loss=0.1574, over 5698190.83 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3748, pruned_loss=0.1217, over 5703651.14 frames. ], batch size: 555, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:48:27,084 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 39900, giga_loss[loss=0.3136, simple_loss=0.3735, pruned_loss=0.1268, over 28716.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3795, pruned_loss=0.1264, over 5709203.90 frames. ], libri_tot_loss[loss=0.3627, simple_loss=0.4091, pruned_loss=0.1581, over 5694449.85 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3753, pruned_loss=0.1223, over 5707722.40 frames. ], batch size: 92, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:48:39,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:48:41,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:48:43,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9953, 1.2510, 1.2236, 1.1495], device='cuda:1'), covar=tensor([0.1081, 0.0984, 0.1511, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0746, 0.0638, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 10:48:58,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4140, 2.9917, 1.4808, 1.3684], device='cuda:1'), covar=tensor([0.0752, 0.0303, 0.0795, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0470, 0.0309, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0015, 0.0020], device='cuda:1')
+2023-03-02 10:49:03,747 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:11,655 INFO [train.py:968] (1/2) Epoch 4, batch 39950, giga_loss[loss=0.2804, simple_loss=0.3512, pruned_loss=0.1048, over 29078.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3782, pruned_loss=0.1262, over 5700117.27 frames. ], libri_tot_loss[loss=0.3632, simple_loss=0.4096, pruned_loss=0.1584, over 5682968.75 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3736, pruned_loss=0.1218, over 5710252.20 frames. ], batch size: 128, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:49:20,731 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:33,281 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-02 10:49:33,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:35,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:45,079 INFO [optim.py:369] (1/2) 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 10:49:48,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:49:48,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-02 10:49:51,547 INFO [train.py:968] (1/2) Epoch 4, batch 40000, giga_loss[loss=0.2667, simple_loss=0.3332, pruned_loss=0.1001, over 28846.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.374, pruned_loss=0.1235, over 5708244.52 frames. ], libri_tot_loss[loss=0.363, simple_loss=0.4095, pruned_loss=0.1583, over 5688596.46 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3696, pruned_loss=0.1194, over 5711601.96 frames. ], batch size: 112, lr: 7.34e-03, grad_scale: 8.0
+2023-03-02 10:49:57,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:50:06,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:50:12,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3658, 1.8488, 1.4335, 1.4958], device='cuda:1'), covar=tensor([0.0752, 0.0282, 0.0333, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0128, 0.0131, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0051, 0.0038, 0.0034, 0.0057], device='cuda:1')
+2023-03-02 10:50:18,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3578, 2.0032, 1.4987, 0.7641], device='cuda:1'), covar=tensor([0.2425, 0.1371, 0.2116, 0.2547], device='cuda:1'), in_proj_covar=tensor([0.1347, 0.1240, 0.1358, 0.1123], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 10:50:30,953 INFO [train.py:968] (1/2) Epoch 4, batch 40050, giga_loss[loss=0.291, simple_loss=0.3545, pruned_loss=0.1137, over 28720.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3716, pruned_loss=0.1221, over 5713265.05 frames. ], libri_tot_loss[loss=0.3634, simple_loss=0.4099, pruned_loss=0.1585, over 5693409.87 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1176, over 5712157.98 frames. ], batch size: 99, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:51:03,581 INFO [optim.py:369] (1/2) 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 10:51:10,401 INFO [train.py:968] (1/2) Epoch 4, batch 40100, giga_loss[loss=0.3097, simple_loss=0.3864, pruned_loss=0.1165, over 28903.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.372, pruned_loss=0.1205, over 5720031.88 frames. ], libri_tot_loss[loss=0.3624, simple_loss=0.409, pruned_loss=0.1579, over 5697711.29 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3681, pruned_loss=0.1166, over 5715654.26 frames. ], batch size: 186, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:51:15,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:40,193 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:47,203 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:49,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:51:55,623 INFO [train.py:968] (1/2) Epoch 4, batch 40150, giga_loss[loss=0.2868, simple_loss=0.3614, pruned_loss=0.1061, over 28922.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3731, pruned_loss=0.1199, over 5712370.20 frames. ], libri_tot_loss[loss=0.3623, simple_loss=0.409, pruned_loss=0.1578, over 5699906.92 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3696, pruned_loss=0.1166, over 5707158.83 frames. ], batch size: 213, lr: 7.34e-03, grad_scale: 4.0
+2023-03-02 10:52:04,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 10:52:07,791 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:09,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:15,351 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:29,835 INFO [optim.py:369] (1/2) 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 10:52:32,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:33,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:52:36,597 INFO [train.py:968] (1/2) Epoch 4, batch 40200, giga_loss[loss=0.3013, simple_loss=0.37, pruned_loss=0.1163, over 28301.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3724, pruned_loss=0.1202, over 5693250.98 frames. ], libri_tot_loss[loss=0.3629, simple_loss=0.4095, pruned_loss=0.1582, over 5678951.44 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3687, pruned_loss=0.1167, over 5708734.50 frames. ], batch size: 368, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:53:16,812 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:53:17,181 INFO [train.py:968] (1/2) Epoch 4, batch 40250, giga_loss[loss=0.3066, simple_loss=0.3635, pruned_loss=0.1249, over 28810.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3726, pruned_loss=0.122, over 5697288.79 frames. ], libri_tot_loss[loss=0.3628, simple_loss=0.4096, pruned_loss=0.158, over 5682103.14 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3687, pruned_loss=0.1184, over 5707068.77 frames. ], batch size: 119, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:53:18,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:53:41,202 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:53:50,160 INFO [optim.py:369] (1/2) 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 10:53:50,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0320, 2.4217, 2.2704, 2.1726], device='cuda:1'), covar=tensor([0.1454, 0.1557, 0.1030, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0750, 0.0766, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 10:53:58,840 INFO [train.py:968] (1/2) Epoch 4, batch 40300, giga_loss[loss=0.3157, simple_loss=0.3759, pruned_loss=0.1277, over 27979.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3721, pruned_loss=0.1237, over 5679905.28 frames. ], libri_tot_loss[loss=0.363, simple_loss=0.4096, pruned_loss=0.1582, over 5666981.54 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1201, over 5701658.07 frames. ], batch size: 412, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:54:32,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:54:43,319 INFO [train.py:968] (1/2) Epoch 4, batch 40350, giga_loss[loss=0.2467, simple_loss=0.3163, pruned_loss=0.08858, over 28420.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3694, pruned_loss=0.1227, over 5690738.40 frames. ], libri_tot_loss[loss=0.3634, simple_loss=0.41, pruned_loss=0.1584, over 5664960.70 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3655, pruned_loss=0.1192, over 5710207.19 frames. ], batch size: 71, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:55:17,979 INFO [optim.py:369] (1/2) 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 10:55:27,445 INFO [train.py:968] (1/2) Epoch 4, batch 40400, giga_loss[loss=0.2613, simple_loss=0.3362, pruned_loss=0.0932, over 28953.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3676, pruned_loss=0.1214, over 5703479.30 frames. ], libri_tot_loss[loss=0.3631, simple_loss=0.4098, pruned_loss=0.1582, over 5667078.03 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3644, pruned_loss=0.1185, over 5717259.77 frames. ], batch size: 174, lr: 7.33e-03, grad_scale: 8.0
+2023-03-02 10:56:08,825 INFO [train.py:968] (1/2) Epoch 4, batch 40450, libri_loss[loss=0.2858, simple_loss=0.3492, pruned_loss=0.1112, over 28082.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.365, pruned_loss=0.1203, over 5703864.53 frames. ], libri_tot_loss[loss=0.3633, simple_loss=0.4099, pruned_loss=0.1584, over 5666335.50 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3615, pruned_loss=0.1172, over 5716058.54 frames. ], batch size: 62, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:56:09,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4939, 1.4806, 1.5904, 1.4728], device='cuda:1'), covar=tensor([0.1001, 0.1504, 0.1392, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0761, 0.0650, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 10:56:11,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4562, 1.8805, 1.4927, 1.5969], device='cuda:1'), covar=tensor([0.0769, 0.0290, 0.0335, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0129, 0.0133, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0038, 0.0035, 0.0058], device='cuda:1')
+2023-03-02 10:56:28,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5333, 1.7995, 1.7599, 1.6420], device='cuda:1'), covar=tensor([0.1479, 0.1794, 0.1189, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0750, 0.0769, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 10:56:32,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:35,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:42,530 INFO [optim.py:369] (1/2) 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 10:56:42,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:45,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-02 10:56:50,172 INFO [train.py:968] (1/2) Epoch 4, batch 40500, giga_loss[loss=0.2613, simple_loss=0.3329, pruned_loss=0.0949, over 28605.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3587, pruned_loss=0.1165, over 5707372.44 frames. ], libri_tot_loss[loss=0.3635, simple_loss=0.41, pruned_loss=0.1585, over 5664965.95 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3555, pruned_loss=0.1135, over 5718602.42 frames. ], batch size: 336, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:56:54,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:55,368 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:56:59,113 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:57:09,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1660, 0.9574, 0.8462, 1.3981], device='cuda:1'), covar=tensor([0.0798, 0.0360, 0.0402, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0129, 0.0133, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0039, 0.0035, 0.0058], device='cuda:1')
+2023-03-02 10:57:15,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5754, 3.3008, 1.6006, 1.5516], device='cuda:1'), covar=tensor([0.0794, 0.0302, 0.0825, 0.1231], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0476, 0.0310, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 10:57:29,829 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-02 10:57:31,864 INFO [train.py:968] (1/2) Epoch 4, batch 40550, giga_loss[loss=0.3494, simple_loss=0.3922, pruned_loss=0.1533, over 28616.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3552, pruned_loss=0.1144, over 5709931.00 frames. ], libri_tot_loss[loss=0.3638, simple_loss=0.4101, pruned_loss=0.1587, over 5668808.15 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3518, pruned_loss=0.1113, over 5715890.52 frames. ], batch size: 92, lr: 7.33e-03, grad_scale: 4.0
+2023-03-02 10:57:36,576 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3506, 1.5248, 1.2644, 1.6417], device='cuda:1'), covar=tensor([0.2172, 0.2051, 0.2085, 0.2074], device='cuda:1'), in_proj_covar=tensor([0.1099, 0.0850, 0.0972, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 10:57:48,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=176713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:58:06,996 INFO [optim.py:369] (1/2) 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 10:58:14,182 INFO [train.py:968] (1/2) Epoch 4, batch 40600, giga_loss[loss=0.33, simple_loss=0.3883, pruned_loss=0.1358, over 27600.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3576, pruned_loss=0.1156, over 5699339.32 frames. ], libri_tot_loss[loss=0.3626, simple_loss=0.4091, pruned_loss=0.1581, over 5665443.61 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3542, pruned_loss=0.1124, over 5707732.31 frames. ], batch size: 472, lr: 7.33e-03, grad_scale: 2.0
+2023-03-02 10:58:25,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9865, 1.1126, 0.9359, 0.6281], device='cuda:1'), covar=tensor([0.0781, 0.0760, 0.0534, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.1109, 0.1124, 0.1185], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 10:58:54,039 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:58:54,362 INFO [train.py:968] (1/2) Epoch 4, batch 40650, giga_loss[loss=0.3073, simple_loss=0.3682, pruned_loss=0.1232, over 29020.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3613, pruned_loss=0.1171, over 5697225.84 frames. ], libri_tot_loss[loss=0.3622, simple_loss=0.4086, pruned_loss=0.1578, over 5659948.37 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3577, pruned_loss=0.1137, over 5709364.34 frames. ], batch size: 128, lr: 7.33e-03, grad_scale: 2.0
+2023-03-02 10:58:57,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:19,976 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:28,616 INFO [optim.py:369] (1/2) 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 10:59:33,736 INFO [train.py:968] (1/2) Epoch 4, batch 40700, giga_loss[loss=0.316, simple_loss=0.3758, pruned_loss=0.1281, over 28616.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3648, pruned_loss=0.1183, over 5702873.10 frames. ], libri_tot_loss[loss=0.361, simple_loss=0.4079, pruned_loss=0.1571, over 5666031.31 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3611, pruned_loss=0.1149, over 5708529.70 frames. ], batch size: 307, lr: 7.32e-03, grad_scale: 2.0
+2023-03-02 10:59:45,617 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=176856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:47,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=176859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 10:59:48,052 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=176859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:00:13,521 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=176888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:00:17,539 INFO [train.py:968] (1/2) Epoch 4, batch 40750, giga_loss[loss=0.3095, simple_loss=0.3779, pruned_loss=0.1205, over 28897.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3678, pruned_loss=0.1194, over 5711643.44 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4074, pruned_loss=0.1566, over 5667056.08 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3646, pruned_loss=0.1164, over 5716093.76 frames. ], batch size: 112, lr: 7.32e-03, grad_scale: 2.0
+2023-03-02 11:00:56,053 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 40800, giga_loss[loss=0.2996, simple_loss=0.3718, pruned_loss=0.1137, over 28672.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3714, pruned_loss=0.1216, over 5711764.60 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4077, pruned_loss=0.1569, over 5670845.15 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.368, pruned_loss=0.1185, over 5712278.67 frames. ], batch size: 262, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:01:48,721 INFO [train.py:968] (1/2) Epoch 4, batch 40850, giga_loss[loss=0.3379, simple_loss=0.3958, pruned_loss=0.14, over 28680.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3744, pruned_loss=0.1242, over 5708626.68 frames. ], libri_tot_loss[loss=0.3608, simple_loss=0.4077, pruned_loss=0.1569, over 5674209.48 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3712, pruned_loss=0.1212, over 5706571.72 frames. ], batch size: 242, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:02:08,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:02:16,947 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:02:22,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:02:27,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8010, 1.0581, 3.3335, 2.7803], device='cuda:1'), covar=tensor([0.1524, 0.2084, 0.0402, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0552, 0.0513, 0.0726, 0.0582], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 11:02:33,251 INFO [optim.py:369] (1/2) 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 11:02:40,017 INFO [train.py:968] (1/2) Epoch 4, batch 40900, giga_loss[loss=0.3528, simple_loss=0.4118, pruned_loss=0.1469, over 28534.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3837, pruned_loss=0.1331, over 5680743.83 frames. ], libri_tot_loss[loss=0.3604, simple_loss=0.4073, pruned_loss=0.1567, over 5668738.85 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.381, pruned_loss=0.1305, over 5683622.06 frames. ], batch size: 71, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:02:56,068 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:03:27,495 INFO [train.py:968] (1/2) Epoch 4, batch 40950, giga_loss[loss=0.4022, simple_loss=0.4439, pruned_loss=0.1802, over 28602.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3906, pruned_loss=0.1383, over 5674252.83 frames. ], libri_tot_loss[loss=0.36, simple_loss=0.4071, pruned_loss=0.1564, over 5664682.34 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3882, pruned_loss=0.136, over 5680339.76 frames. ], batch size: 242, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:04:07,572 INFO [optim.py:369] (1/2) 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 11:04:12,791 INFO [train.py:968] (1/2) Epoch 4, batch 41000, giga_loss[loss=0.3916, simple_loss=0.436, pruned_loss=0.1736, over 28602.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3977, pruned_loss=0.1447, over 5647132.80 frames. ], libri_tot_loss[loss=0.3607, simple_loss=0.4077, pruned_loss=0.1568, over 5647721.10 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.395, pruned_loss=0.1422, over 5666281.31 frames. ], batch size: 307, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:04:20,233 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:22,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:35,362 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:37,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:48,106 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:04:55,203 INFO [train.py:968] (1/2) Epoch 4, batch 41050, giga_loss[loss=0.4224, simple_loss=0.4494, pruned_loss=0.1977, over 28820.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4046, pruned_loss=0.151, over 5663567.50 frames. ], libri_tot_loss[loss=0.3606, simple_loss=0.4077, pruned_loss=0.1567, over 5655489.35 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4022, pruned_loss=0.1489, over 5672179.00 frames. ], batch size: 99, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:05:03,540 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:05:23,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5379, 2.0044, 1.2776, 0.7599], device='cuda:1'), covar=tensor([0.2437, 0.1429, 0.1420, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.1386, 0.1283, 0.1379, 0.1144], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 11:05:36,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:05:39,022 INFO [optim.py:369] (1/2) 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 11:05:44,887 INFO [train.py:968] (1/2) Epoch 4, batch 41100, giga_loss[loss=0.3961, simple_loss=0.4307, pruned_loss=0.1807, over 28744.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.411, pruned_loss=0.1568, over 5659249.43 frames. ], libri_tot_loss[loss=0.3612, simple_loss=0.4082, pruned_loss=0.1572, over 5659276.23 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4087, pruned_loss=0.1547, over 5662807.22 frames. ], batch size: 119, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:05:50,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:05:58,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0342, 1.6847, 1.3448, 1.5334], device='cuda:1'), covar=tensor([0.0575, 0.0664, 0.0952, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0458, 0.0506, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 11:06:02,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8671, 0.8982, 0.6724, 0.6918], device='cuda:1'), covar=tensor([0.0513, 0.0560, 0.0418, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.1376, 0.1125, 0.1137, 0.1199], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 11:06:34,889 INFO [train.py:968] (1/2) Epoch 4, batch 41150, giga_loss[loss=0.3788, simple_loss=0.4213, pruned_loss=0.1682, over 28805.00 frames. ], tot_loss[loss=0.3648, simple_loss=0.4122, pruned_loss=0.1587, over 5663084.62 frames. ], libri_tot_loss[loss=0.3603, simple_loss=0.4073, pruned_loss=0.1566, over 5664910.37 frames. ], giga_tot_loss[loss=0.3632, simple_loss=0.4113, pruned_loss=0.1575, over 5660463.00 frames. ], batch size: 243, lr: 7.32e-03, grad_scale: 4.0
+2023-03-02 11:07:23,910 INFO [optim.py:369] (1/2) 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 11:07:32,481 INFO [train.py:968] (1/2) Epoch 4, batch 41200, giga_loss[loss=0.4121, simple_loss=0.4478, pruned_loss=0.1882, over 28896.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4143, pruned_loss=0.1617, over 5634612.10 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4068, pruned_loss=0.1562, over 5667529.82 frames. ], giga_tot_loss[loss=0.3684, simple_loss=0.4142, pruned_loss=0.1613, over 5629728.67 frames. ], batch size: 227, lr: 7.31e-03, grad_scale: 8.0
+2023-03-02 11:08:06,616 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:08:08,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:08:21,773 INFO [train.py:968] (1/2) Epoch 4, batch 41250, giga_loss[loss=0.4407, simple_loss=0.456, pruned_loss=0.2127, over 27919.00 frames. ], tot_loss[loss=0.3739, simple_loss=0.4172, pruned_loss=0.1653, over 5628252.73 frames. ], libri_tot_loss[loss=0.3591, simple_loss=0.4065, pruned_loss=0.1558, over 5671671.49 frames. ], giga_tot_loss[loss=0.3743, simple_loss=0.4177, pruned_loss=0.1654, over 5619677.50 frames. ], batch size: 412, lr: 7.31e-03, grad_scale: 8.0
+2023-03-02 11:08:22,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:08:27,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-02 11:08:37,039 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:09:02,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:09:08,213 INFO [optim.py:369] (1/2) 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 11:09:14,474 INFO [train.py:968] (1/2) Epoch 4, batch 41300, giga_loss[loss=0.4893, simple_loss=0.4863, pruned_loss=0.2461, over 26669.00 frames. ], tot_loss[loss=0.381, simple_loss=0.4225, pruned_loss=0.1697, over 5634993.66 frames. ], libri_tot_loss[loss=0.359, simple_loss=0.4064, pruned_loss=0.1558, over 5676402.45 frames. ], giga_tot_loss[loss=0.3817, simple_loss=0.4232, pruned_loss=0.1701, over 5623316.22 frames. ], batch size: 555, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:10:10,328 INFO [train.py:968] (1/2) Epoch 4, batch 41350, giga_loss[loss=0.5048, simple_loss=0.4916, pruned_loss=0.259, over 26595.00 frames. ], tot_loss[loss=0.3827, simple_loss=0.4235, pruned_loss=0.1709, over 5640209.94 frames. ], libri_tot_loss[loss=0.3597, simple_loss=0.4069, pruned_loss=0.1563, over 5677864.99 frames. ], giga_tot_loss[loss=0.3829, simple_loss=0.4239, pruned_loss=0.171, over 5628921.30 frames. ], batch size: 555, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:10:21,064 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:10:22,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:10:53,973 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:10:54,281 INFO [optim.py:369] (1/2) 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 11:10:57,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:02,008 INFO [train.py:968] (1/2) Epoch 4, batch 41400, giga_loss[loss=0.455, simple_loss=0.4683, pruned_loss=0.2209, over 28551.00 frames. ], tot_loss[loss=0.3804, simple_loss=0.421, pruned_loss=0.1699, over 5637700.04 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4068, pruned_loss=0.1562, over 5677592.27 frames. ], giga_tot_loss[loss=0.381, simple_loss=0.4216, pruned_loss=0.1702, over 5628739.64 frames. ], batch size: 336, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:11:25,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:34,032 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:36,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:11:52,527 INFO [train.py:968] (1/2) Epoch 4, batch 41450, giga_loss[loss=0.3578, simple_loss=0.4144, pruned_loss=0.1506, over 28896.00 frames. ], tot_loss[loss=0.3789, simple_loss=0.4203, pruned_loss=0.1688, over 5634987.21 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4068, pruned_loss=0.1561, over 5680659.56 frames. ], giga_tot_loss[loss=0.3797, simple_loss=0.4208, pruned_loss=0.1693, over 5624654.58 frames. ], batch size: 199, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:12:04,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5133, 3.3247, 3.1900, 1.5693], device='cuda:1'), covar=tensor([0.0656, 0.0588, 0.0877, 0.2239], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0727, 0.0804, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 11:12:08,941 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:12:23,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3351, 1.6648, 1.5092, 1.4343], device='cuda:1'), covar=tensor([0.1381, 0.1869, 0.1103, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0755, 0.0753, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 11:12:36,434 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 41500, giga_loss[loss=0.4698, simple_loss=0.4607, pruned_loss=0.2395, over 23410.00 frames. ], tot_loss[loss=0.377, simple_loss=0.4193, pruned_loss=0.1673, over 5624256.18 frames. ], libri_tot_loss[loss=0.3599, simple_loss=0.4071, pruned_loss=0.1563, over 5685700.50 frames. ], giga_tot_loss[loss=0.3778, simple_loss=0.42, pruned_loss=0.1679, over 5609934.25 frames. ], batch size: 705, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:13:31,303 INFO [train.py:968] (1/2) Epoch 4, batch 41550, giga_loss[loss=0.3769, simple_loss=0.4195, pruned_loss=0.1672, over 27914.00 frames. ], tot_loss[loss=0.3809, simple_loss=0.4222, pruned_loss=0.1698, over 5611224.71 frames. ], libri_tot_loss[loss=0.3597, simple_loss=0.407, pruned_loss=0.1562, over 5691686.47 frames. ], giga_tot_loss[loss=0.3823, simple_loss=0.4231, pruned_loss=0.1707, over 5593010.62 frames. ], batch size: 412, lr: 7.31e-03, grad_scale: 4.0
+2023-03-02 11:14:00,044 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 11:14:15,815 INFO [optim.py:369] (1/2) 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 11:14:21,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0601, 1.2380, 1.0735, 1.1776], device='cuda:1'), covar=tensor([0.0914, 0.0320, 0.0330, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0130, 0.0132, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 11:14:21,761 INFO [train.py:968] (1/2) Epoch 4, batch 41600, giga_loss[loss=0.3696, simple_loss=0.4218, pruned_loss=0.1587, over 28941.00 frames. ], tot_loss[loss=0.3783, simple_loss=0.4203, pruned_loss=0.1681, over 5607635.51 frames. ], libri_tot_loss[loss=0.36, simple_loss=0.4071, pruned_loss=0.1564, over 5684161.99 frames. ], giga_tot_loss[loss=0.3797, simple_loss=0.4214, pruned_loss=0.169, over 5596429.20 frames. ], batch size: 106, lr: 7.31e-03, grad_scale: 8.0
+2023-03-02 11:14:24,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.37 vs. limit=5.0
+2023-03-02 11:14:40,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=177764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:14:42,729 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=177767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:15:07,410 INFO [train.py:968] (1/2) Epoch 4, batch 41650, libri_loss[loss=0.4008, simple_loss=0.428, pruned_loss=0.1868, over 29536.00 frames. ], tot_loss[loss=0.3723, simple_loss=0.4171, pruned_loss=0.1638, over 5628850.80 frames. ], libri_tot_loss[loss=0.3595, simple_loss=0.4066, pruned_loss=0.1562, over 5692569.58 frames. ], giga_tot_loss[loss=0.3744, simple_loss=0.4188, pruned_loss=0.165, over 5609829.96 frames. ], batch size: 81, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:15:11,776 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=177796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:15:55,671 INFO [optim.py:369] (1/2) 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 11:16:01,825 INFO [train.py:968] (1/2) Epoch 4, batch 41700, giga_loss[loss=0.3565, simple_loss=0.3845, pruned_loss=0.1642, over 24232.00 frames. ], tot_loss[loss=0.3685, simple_loss=0.4148, pruned_loss=0.1611, over 5635119.59 frames. ], libri_tot_loss[loss=0.3596, simple_loss=0.4067, pruned_loss=0.1562, over 5692767.73 frames. ], giga_tot_loss[loss=0.3702, simple_loss=0.4162, pruned_loss=0.1621, over 5619828.15 frames. ], batch size: 705, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:16:37,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:16:39,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=177879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:16:53,995 INFO [train.py:968] (1/2) Epoch 4, batch 41750, giga_loss[loss=0.3058, simple_loss=0.3771, pruned_loss=0.1173, over 28745.00 frames. ], tot_loss[loss=0.3631, simple_loss=0.4108, pruned_loss=0.1577, over 5639420.98 frames. ], libri_tot_loss[loss=0.3591, simple_loss=0.4063, pruned_loss=0.1559, over 5696569.18 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4123, pruned_loss=0.1588, over 5623062.46 frames. ], batch size: 284, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:17:27,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1557, 1.3971, 1.1288, 1.4750], device='cuda:1'), covar=tensor([0.2143, 0.2010, 0.2159, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.1105, 0.0864, 0.0985, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 11:17:40,462 INFO [optim.py:369] (1/2) 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 11:17:44,754 INFO [train.py:968] (1/2) Epoch 4, batch 41800, libri_loss[loss=0.33, simple_loss=0.3942, pruned_loss=0.1329, over 29536.00 frames. ], tot_loss[loss=0.3581, simple_loss=0.4073, pruned_loss=0.1545, over 5633020.38 frames. ], libri_tot_loss[loss=0.3587, simple_loss=0.406, pruned_loss=0.1557, over 5691029.80 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4088, pruned_loss=0.1556, over 5623446.79 frames. ], batch size: 83, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:17:48,820 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=177946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:18:32,748 INFO [train.py:968] (1/2) Epoch 4, batch 41850, giga_loss[loss=0.3057, simple_loss=0.3689, pruned_loss=0.1213, over 28515.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4064, pruned_loss=0.1537, over 5636981.28 frames. ], libri_tot_loss[loss=0.3583, simple_loss=0.4057, pruned_loss=0.1555, over 5686918.81 frames. ], giga_tot_loss[loss=0.3587, simple_loss=0.4079, pruned_loss=0.1547, over 5631076.46 frames. ], batch size: 71, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:18:42,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7506, 1.9735, 1.7087, 1.9798], device='cuda:1'), covar=tensor([0.0748, 0.0296, 0.0310, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0129, 0.0132, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0052, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 11:18:56,717 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:18:58,236 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:18:59,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:02,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:13,000 INFO [optim.py:369] (1/2) 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 11:19:18,462 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-02 11:19:18,544 INFO [train.py:968] (1/2) Epoch 4, batch 41900, giga_loss[loss=0.3738, simple_loss=0.3883, pruned_loss=0.1796, over 23437.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4063, pruned_loss=0.1535, over 5642897.78 frames. ], libri_tot_loss[loss=0.3582, simple_loss=0.4057, pruned_loss=0.1554, over 5689736.97 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4075, pruned_loss=0.1543, over 5634572.48 frames. ], batch size: 705, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:19:28,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2810, 1.3848, 1.2101, 1.3705], device='cuda:1'), covar=tensor([0.2422, 0.2342, 0.2288, 0.2114], device='cuda:1'), in_proj_covar=tensor([0.1092, 0.0860, 0.0974, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 11:19:31,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:33,176 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:19:35,331 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:20:14,412 INFO [train.py:968] (1/2) Epoch 4, batch 41950, giga_loss[loss=0.3858, simple_loss=0.4343, pruned_loss=0.1687, over 28842.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4029, pruned_loss=0.1506, over 5640414.63 frames. ], libri_tot_loss[loss=0.3571, simple_loss=0.4047, pruned_loss=0.1548, over 5692985.04 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4048, pruned_loss=0.1518, over 5629693.89 frames. ], batch size: 199, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:20:17,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8713, 1.7173, 1.2562, 1.5401], device='cuda:1'), covar=tensor([0.0534, 0.0496, 0.0912, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0461, 0.0511, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 11:20:58,288 INFO [optim.py:369] (1/2) 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 11:21:04,640 INFO [train.py:968] (1/2) Epoch 4, batch 42000, giga_loss[loss=0.3731, simple_loss=0.4247, pruned_loss=0.1607, over 28008.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4029, pruned_loss=0.1474, over 5650350.85 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4046, pruned_loss=0.1546, over 5696487.78 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4045, pruned_loss=0.1483, over 5637272.72 frames. ], batch size: 412, lr: 7.30e-03, grad_scale: 8.0
+2023-03-02 11:21:04,640 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 11:21:13,475 INFO [train.py:1012] (1/2) Epoch 4, validation: loss=0.2346, simple_loss=0.3352, pruned_loss=0.06698, over 944034.00 frames. 
+2023-03-02 11:21:13,475 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 11:21:45,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:21:56,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7378, 1.5969, 1.2188, 1.3979], device='cuda:1'), covar=tensor([0.0652, 0.0653, 0.1001, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0462, 0.0509, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 11:22:01,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:22:03,605 INFO [train.py:968] (1/2) Epoch 4, batch 42050, giga_loss[loss=0.4543, simple_loss=0.4766, pruned_loss=0.216, over 28612.00 frames. ], tot_loss[loss=0.3506, simple_loss=0.4054, pruned_loss=0.1479, over 5664425.69 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4046, pruned_loss=0.1547, over 5699487.41 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4066, pruned_loss=0.1485, over 5650914.76 frames. ], batch size: 307, lr: 7.30e-03, grad_scale: 8.0
+2023-03-02 11:22:47,019 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 42100, giga_loss[loss=0.3306, simple_loss=0.3917, pruned_loss=0.1348, over 28337.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4059, pruned_loss=0.1489, over 5662149.33 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4046, pruned_loss=0.1547, over 5693670.66 frames. ], giga_tot_loss[loss=0.3526, simple_loss=0.4069, pruned_loss=0.1492, over 5655448.16 frames. ], batch size: 71, lr: 7.30e-03, grad_scale: 4.0
+2023-03-02 11:23:02,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-02 11:23:34,124 INFO [train.py:968] (1/2) Epoch 4, batch 42150, giga_loss[loss=0.3525, simple_loss=0.4037, pruned_loss=0.1507, over 28480.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4059, pruned_loss=0.1497, over 5653815.01 frames. ], libri_tot_loss[loss=0.357, simple_loss=0.4046, pruned_loss=0.1547, over 5690802.27 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4067, pruned_loss=0.1497, over 5649289.64 frames. ], batch size: 336, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:23:58,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:24:14,167 INFO [optim.py:369] (1/2) 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 11:24:18,017 INFO [train.py:968] (1/2) Epoch 4, batch 42200, giga_loss[loss=0.3382, simple_loss=0.3864, pruned_loss=0.1451, over 28784.00 frames. ], tot_loss[loss=0.3509, simple_loss=0.4038, pruned_loss=0.149, over 5665101.73 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4045, pruned_loss=0.1546, over 5685929.57 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4046, pruned_loss=0.149, over 5664899.12 frames. ], batch size: 284, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:24:29,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1674, 1.1596, 1.0388, 0.9193], device='cuda:1'), covar=tensor([0.0478, 0.0358, 0.0700, 0.0683], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0458, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 11:25:04,624 INFO [train.py:968] (1/2) Epoch 4, batch 42250, giga_loss[loss=0.398, simple_loss=0.4277, pruned_loss=0.1841, over 27565.00 frames. ], tot_loss[loss=0.3515, simple_loss=0.4031, pruned_loss=0.15, over 5663875.78 frames. ], libri_tot_loss[loss=0.3564, simple_loss=0.4043, pruned_loss=0.1542, over 5690871.23 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.404, pruned_loss=0.1502, over 5658869.82 frames. ], batch size: 472, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:25:39,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1889, 2.4250, 2.6655, 2.4995], device='cuda:1'), covar=tensor([0.0744, 0.1718, 0.1107, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0758, 0.0641, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 11:25:40,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:25:47,394 INFO [optim.py:369] (1/2) 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 11:25:53,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3782, 1.9873, 1.5138, 0.5753], device='cuda:1'), covar=tensor([0.2520, 0.1196, 0.1753, 0.2676], device='cuda:1'), in_proj_covar=tensor([0.1353, 0.1244, 0.1346, 0.1120], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 11:25:53,436 INFO [train.py:968] (1/2) Epoch 4, batch 42300, giga_loss[loss=0.3186, simple_loss=0.3883, pruned_loss=0.1244, over 28639.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4019, pruned_loss=0.1486, over 5669659.12 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4038, pruned_loss=0.1538, over 5698909.66 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4029, pruned_loss=0.1489, over 5657588.93 frames. ], batch size: 307, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:26:13,348 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:26:16,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:26:39,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.94 vs. limit=5.0
+2023-03-02 11:26:39,517 INFO [train.py:968] (1/2) Epoch 4, batch 42350, giga_loss[loss=0.3247, simple_loss=0.3886, pruned_loss=0.1304, over 28863.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.4006, pruned_loss=0.1456, over 5672010.65 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4034, pruned_loss=0.1536, over 5691360.42 frames. ], giga_tot_loss[loss=0.347, simple_loss=0.4018, pruned_loss=0.1461, over 5667835.58 frames. ], batch size: 199, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:26:42,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:27:26,535 INFO [optim.py:369] (1/2) 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 11:27:30,391 INFO [train.py:968] (1/2) Epoch 4, batch 42400, libri_loss[loss=0.3997, simple_loss=0.4372, pruned_loss=0.1811, over 29252.00 frames. ], tot_loss[loss=0.348, simple_loss=0.4022, pruned_loss=0.1469, over 5671352.77 frames. ], libri_tot_loss[loss=0.356, simple_loss=0.404, pruned_loss=0.154, over 5692941.14 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.4025, pruned_loss=0.1466, over 5666006.55 frames. ], batch size: 97, lr: 7.29e-03, grad_scale: 8.0
+2023-03-02 11:27:35,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:27:50,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=178565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:28:01,000 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:28:03,812 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 42450, giga_loss[loss=0.3595, simple_loss=0.4128, pruned_loss=0.1531, over 28700.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4025, pruned_loss=0.1476, over 5670357.19 frames. ], libri_tot_loss[loss=0.3561, simple_loss=0.404, pruned_loss=0.1541, over 5695235.82 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4028, pruned_loss=0.1473, over 5664010.54 frames. ], batch size: 242, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:28:27,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7651, 1.5904, 1.2453, 1.3515], device='cuda:1'), covar=tensor([0.0603, 0.0601, 0.0968, 0.0895], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0454, 0.0505, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 11:28:27,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-02 11:28:33,790 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 4, batch 42500, giga_loss[loss=0.4096, simple_loss=0.4413, pruned_loss=0.1889, over 27550.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.4004, pruned_loss=0.1463, over 5672624.87 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4045, pruned_loss=0.1543, over 5688559.11 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.4001, pruned_loss=0.1457, over 5672877.10 frames. ], batch size: 472, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:29:27,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3322, 1.6116, 1.2406, 1.3880], device='cuda:1'), covar=tensor([0.0806, 0.0315, 0.0350, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0129, 0.0132, 0.0238], device='cuda:1'), out_proj_covar=tensor([0.0053, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 11:29:53,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 11:29:57,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:29:58,304 INFO [train.py:968] (1/2) Epoch 4, batch 42550, giga_loss[loss=0.3191, simple_loss=0.3811, pruned_loss=0.1286, over 28654.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3998, pruned_loss=0.147, over 5665089.91 frames. ], libri_tot_loss[loss=0.3568, simple_loss=0.4048, pruned_loss=0.1544, over 5692042.24 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3992, pruned_loss=0.1464, over 5662006.11 frames. ], batch size: 262, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:29:59,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3850, 1.3967, 1.2474, 1.7898], device='cuda:1'), covar=tensor([0.2299, 0.2250, 0.2030, 0.2201], device='cuda:1'), in_proj_covar=tensor([0.1097, 0.0865, 0.0974, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 11:29:59,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:12,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=178708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:14,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=178711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:30,958 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:30:47,117 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=178740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:30:50,015 INFO [train.py:968] (1/2) Epoch 4, batch 42600, giga_loss[loss=0.3645, simple_loss=0.4118, pruned_loss=0.1587, over 28994.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.3983, pruned_loss=0.1465, over 5680006.48 frames. ], libri_tot_loss[loss=0.3563, simple_loss=0.4044, pruned_loss=0.1542, over 5696543.36 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3982, pruned_loss=0.146, over 5673316.17 frames. ], batch size: 213, lr: 7.29e-03, grad_scale: 4.0
+2023-03-02 11:31:37,457 INFO [train.py:968] (1/2) Epoch 4, batch 42650, giga_loss[loss=0.2763, simple_loss=0.3516, pruned_loss=0.1005, over 28152.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3966, pruned_loss=0.1455, over 5684796.60 frames. ], libri_tot_loss[loss=0.3562, simple_loss=0.4044, pruned_loss=0.154, over 5701758.71 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3962, pruned_loss=0.1451, over 5674488.94 frames. ], batch size: 77, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:32:24,752 INFO [optim.py:369] (1/2) 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 11:32:27,820 INFO [train.py:968] (1/2) Epoch 4, batch 42700, giga_loss[loss=0.335, simple_loss=0.3849, pruned_loss=0.1426, over 28826.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3967, pruned_loss=0.1466, over 5649571.27 frames. ], libri_tot_loss[loss=0.3569, simple_loss=0.4049, pruned_loss=0.1544, over 5683342.34 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3958, pruned_loss=0.1458, over 5656517.24 frames. ], batch size: 92, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:32:30,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=178845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:32:45,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-02 11:33:16,145 INFO [train.py:968] (1/2) Epoch 4, batch 42750, giga_loss[loss=0.3226, simple_loss=0.3934, pruned_loss=0.1259, over 28912.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3968, pruned_loss=0.1469, over 5653224.23 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4046, pruned_loss=0.1543, over 5687279.75 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3962, pruned_loss=0.1462, over 5654794.13 frames. ], batch size: 112, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:34:01,871 INFO [optim.py:369] (1/2) 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 11:34:05,568 INFO [train.py:968] (1/2) Epoch 4, batch 42800, giga_loss[loss=0.2964, simple_loss=0.3692, pruned_loss=0.1118, over 28694.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3969, pruned_loss=0.1458, over 5669074.45 frames. ], libri_tot_loss[loss=0.3559, simple_loss=0.4042, pruned_loss=0.1538, over 5693119.13 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3965, pruned_loss=0.1454, over 5664244.70 frames. ], batch size: 78, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:34:49,460 INFO [train.py:968] (1/2) Epoch 4, batch 42850, giga_loss[loss=0.334, simple_loss=0.3925, pruned_loss=0.1377, over 28193.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3959, pruned_loss=0.1441, over 5673492.26 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.4035, pruned_loss=0.1532, over 5695336.41 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.396, pruned_loss=0.1441, over 5666828.13 frames. ], batch size: 368, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:35:06,417 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-02 11:35:07,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:35:09,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4730, 3.0129, 1.3861, 1.4795], device='cuda:1'), covar=tensor([0.0801, 0.0378, 0.0847, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0480, 0.0309, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 11:35:17,323 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-02 11:35:32,734 INFO [optim.py:369] (1/2) 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 11:35:34,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-02 11:35:34,694 INFO [train.py:968] (1/2) Epoch 4, batch 42900, giga_loss[loss=0.3297, simple_loss=0.3923, pruned_loss=0.1336, over 28949.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3962, pruned_loss=0.1441, over 5678706.22 frames. ], libri_tot_loss[loss=0.3552, simple_loss=0.4036, pruned_loss=0.1533, over 5701327.63 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3959, pruned_loss=0.1436, over 5667246.64 frames. ], batch size: 145, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:35:42,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2133, 1.3620, 1.1710, 1.3412], device='cuda:1'), covar=tensor([0.2456, 0.2228, 0.2196, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.1089, 0.0857, 0.0965, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 11:36:20,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:36:23,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179090.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:36:25,584 INFO [train.py:968] (1/2) Epoch 4, batch 42950, giga_loss[loss=0.3682, simple_loss=0.4152, pruned_loss=0.1606, over 29006.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3992, pruned_loss=0.1468, over 5689235.09 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4035, pruned_loss=0.1531, over 5705080.29 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3989, pruned_loss=0.1464, over 5676551.17 frames. ], batch size: 155, lr: 7.28e-03, grad_scale: 8.0
+2023-03-02 11:36:42,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-02 11:37:09,211 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 43000, giga_loss[loss=0.306, simple_loss=0.3814, pruned_loss=0.1153, over 28862.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3996, pruned_loss=0.1478, over 5691842.21 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.4021, pruned_loss=0.1522, over 5705511.68 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4005, pruned_loss=0.1481, over 5681078.10 frames. ], batch size: 174, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:37:32,192 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 43050, giga_loss[loss=0.358, simple_loss=0.4027, pruned_loss=0.1567, over 27981.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4016, pruned_loss=0.1509, over 5679244.40 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4022, pruned_loss=0.1523, over 5704316.30 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4022, pruned_loss=0.151, over 5671592.83 frames. ], batch size: 412, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:38:14,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8661, 1.1852, 3.3557, 2.9812], device='cuda:1'), covar=tensor([0.1593, 0.2110, 0.0454, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0565, 0.0523, 0.0741, 0.0595], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 11:38:32,469 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:38:56,667 INFO [train.py:968] (1/2) Epoch 4, batch 43100, giga_loss[loss=0.4635, simple_loss=0.474, pruned_loss=0.2265, over 24199.00 frames. ], tot_loss[loss=0.356, simple_loss=0.404, pruned_loss=0.154, over 5672335.53 frames. ], libri_tot_loss[loss=0.3528, simple_loss=0.4018, pruned_loss=0.152, over 5702129.97 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4048, pruned_loss=0.1543, over 5667776.31 frames. ], batch size: 705, lr: 7.28e-03, grad_scale: 4.0
+2023-03-02 11:39:35,694 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179287.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:39:40,156 INFO [train.py:968] (1/2) Epoch 4, batch 43150, giga_loss[loss=0.3173, simple_loss=0.3728, pruned_loss=0.1309, over 28925.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4052, pruned_loss=0.1553, over 5657756.27 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.402, pruned_loss=0.1522, over 5695159.20 frames. ], giga_tot_loss[loss=0.3584, simple_loss=0.4058, pruned_loss=0.1555, over 5659514.51 frames. ], batch size: 199, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:40:27,152 INFO [optim.py:369] (1/2) 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 11:40:28,458 INFO [train.py:968] (1/2) Epoch 4, batch 43200, libri_loss[loss=0.3213, simple_loss=0.3632, pruned_loss=0.1397, over 29377.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4033, pruned_loss=0.1539, over 5659566.37 frames. ], libri_tot_loss[loss=0.3527, simple_loss=0.4016, pruned_loss=0.1519, over 5696881.88 frames. ], giga_tot_loss[loss=0.3564, simple_loss=0.4042, pruned_loss=0.1543, over 5658792.38 frames. ], batch size: 67, lr: 7.27e-03, grad_scale: 8.0
+2023-03-02 11:40:35,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-02 11:40:44,787 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:40:46,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:40:55,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-02 11:41:04,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:41:08,744 INFO [train.py:968] (1/2) Epoch 4, batch 43250, giga_loss[loss=0.3068, simple_loss=0.3709, pruned_loss=0.1214, over 28978.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.4025, pruned_loss=0.1515, over 5664793.94 frames. ], libri_tot_loss[loss=0.3526, simple_loss=0.4015, pruned_loss=0.1519, over 5693924.19 frames. ], giga_tot_loss[loss=0.3537, simple_loss=0.4034, pruned_loss=0.152, over 5665302.35 frames. ], batch size: 213, lr: 7.27e-03, grad_scale: 8.0
+2023-03-02 11:41:12,286 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:41:57,986 INFO [optim.py:369] (1/2) 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 11:42:00,261 INFO [train.py:968] (1/2) Epoch 4, batch 43300, giga_loss[loss=0.3773, simple_loss=0.405, pruned_loss=0.1748, over 26601.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.3996, pruned_loss=0.1488, over 5656343.30 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4017, pruned_loss=0.1521, over 5695057.02 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4001, pruned_loss=0.1489, over 5655153.65 frames. ], batch size: 555, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:42:16,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:42:18,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179465.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:42:44,436 INFO [train.py:968] (1/2) Epoch 4, batch 43350, giga_loss[loss=0.2728, simple_loss=0.351, pruned_loss=0.09728, over 28642.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.397, pruned_loss=0.147, over 5670485.05 frames. ], libri_tot_loss[loss=0.3526, simple_loss=0.4014, pruned_loss=0.1519, over 5700994.73 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3976, pruned_loss=0.1471, over 5663137.57 frames. ], batch size: 60, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:43:20,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:43:23,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:43:27,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:43:30,823 INFO [optim.py:369] (1/2) 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 11:43:31,750 INFO [train.py:968] (1/2) Epoch 4, batch 43400, giga_loss[loss=0.3003, simple_loss=0.36, pruned_loss=0.1203, over 28125.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3946, pruned_loss=0.1458, over 5657782.36 frames. ], libri_tot_loss[loss=0.3524, simple_loss=0.4012, pruned_loss=0.1518, over 5693856.19 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3952, pruned_loss=0.146, over 5658419.77 frames. ], batch size: 77, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:43:49,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0222, 1.8603, 1.7687, 1.6890], device='cuda:1'), covar=tensor([0.1011, 0.1736, 0.1508, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0760, 0.0646, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 11:43:52,452 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:44:19,607 INFO [train.py:968] (1/2) Epoch 4, batch 43450, giga_loss[loss=0.3102, simple_loss=0.38, pruned_loss=0.1202, over 28924.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3965, pruned_loss=0.1472, over 5669307.21 frames. ], libri_tot_loss[loss=0.353, simple_loss=0.4017, pruned_loss=0.1521, over 5698321.71 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3964, pruned_loss=0.1469, over 5665124.78 frames. ], batch size: 136, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:44:28,846 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:44:30,696 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:44:30,710 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179608.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:44:30,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-02 11:44:33,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179611.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:44:58,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:45:00,776 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:45:02,206 INFO [optim.py:369] (1/2) 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 11:45:03,036 INFO [train.py:968] (1/2) Epoch 4, batch 43500, giga_loss[loss=0.3548, simple_loss=0.4134, pruned_loss=0.1481, over 28622.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4019, pruned_loss=0.1503, over 5671290.04 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4024, pruned_loss=0.1525, over 5701778.75 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4011, pruned_loss=0.1496, over 5664057.50 frames. ], batch size: 242, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:45:23,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=179662.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:45:43,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:45:43,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 11:45:46,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 43550, libri_loss[loss=0.334, simple_loss=0.3935, pruned_loss=0.1372, over 29526.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4034, pruned_loss=0.1483, over 5668867.00 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4017, pruned_loss=0.1522, over 5697454.34 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4035, pruned_loss=0.148, over 5666449.16 frames. ], batch size: 83, lr: 7.27e-03, grad_scale: 4.0
+2023-03-02 11:46:10,825 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:46:41,146 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 43600, giga_loss[loss=0.3767, simple_loss=0.4334, pruned_loss=0.16, over 28890.00 frames. ], tot_loss[loss=0.3511, simple_loss=0.4047, pruned_loss=0.1487, over 5670698.36 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4021, pruned_loss=0.1527, over 5701649.87 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4044, pruned_loss=0.1479, over 5664297.14 frames. ], batch size: 199, lr: 7.27e-03, grad_scale: 8.0
+2023-03-02 11:46:53,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:47:33,076 INFO [train.py:968] (1/2) Epoch 4, batch 43650, giga_loss[loss=0.3188, simple_loss=0.3811, pruned_loss=0.1283, over 28823.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.4093, pruned_loss=0.1532, over 5656999.34 frames. ], libri_tot_loss[loss=0.3546, simple_loss=0.4026, pruned_loss=0.1533, over 5693517.71 frames. ], giga_tot_loss[loss=0.3563, simple_loss=0.4087, pruned_loss=0.152, over 5659694.89 frames. ], batch size: 60, lr: 7.26e-03, grad_scale: 8.0
+2023-03-02 11:47:43,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=179805.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:47:46,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=179808.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:48:14,164 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=179837.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:48:14,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=179837.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:48:20,178 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 43700, giga_loss[loss=0.2935, simple_loss=0.3657, pruned_loss=0.1106, over 28915.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4096, pruned_loss=0.1537, over 5655932.06 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4036, pruned_loss=0.154, over 5694402.23 frames. ], giga_tot_loss[loss=0.3562, simple_loss=0.4084, pruned_loss=0.1521, over 5656200.13 frames. ], batch size: 174, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:49:02,618 INFO [train.py:968] (1/2) Epoch 4, batch 43750, giga_loss[loss=0.3599, simple_loss=0.42, pruned_loss=0.1499, over 29009.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4077, pruned_loss=0.153, over 5659679.83 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.403, pruned_loss=0.1535, over 5688208.36 frames. ], giga_tot_loss[loss=0.3559, simple_loss=0.4074, pruned_loss=0.1522, over 5664411.03 frames. ], batch size: 164, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:49:54,073 INFO [optim.py:369] (1/2) 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 11:49:54,086 INFO [train.py:968] (1/2) Epoch 4, batch 43800, giga_loss[loss=0.3378, simple_loss=0.3925, pruned_loss=0.1415, over 28909.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.4061, pruned_loss=0.1532, over 5659035.38 frames. ], libri_tot_loss[loss=0.3547, simple_loss=0.4027, pruned_loss=0.1533, over 5690767.28 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4062, pruned_loss=0.1527, over 5659857.64 frames. ], batch size: 99, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:49:54,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3783, 2.1296, 1.5473, 0.6479], device='cuda:1'), covar=tensor([0.2341, 0.1143, 0.1874, 0.2495], device='cuda:1'), in_proj_covar=tensor([0.1360, 0.1278, 0.1344, 0.1134], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 11:50:23,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0755, 1.2794, 4.7995, 3.2930], device='cuda:1'), covar=tensor([0.1727, 0.2209, 0.0312, 0.0664], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0529, 0.0746, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 11:50:36,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6782, 2.5301, 1.7786, 1.0353], device='cuda:1'), covar=tensor([0.2239, 0.1155, 0.1951, 0.2206], device='cuda:1'), in_proj_covar=tensor([0.1352, 0.1269, 0.1334, 0.1125], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 11:50:43,051 INFO [train.py:968] (1/2) Epoch 4, batch 43850, giga_loss[loss=0.3396, simple_loss=0.389, pruned_loss=0.1451, over 28776.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.403, pruned_loss=0.1512, over 5670550.23 frames. ], libri_tot_loss[loss=0.3543, simple_loss=0.4024, pruned_loss=0.1531, over 5694013.12 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4034, pruned_loss=0.151, over 5667669.07 frames. ], batch size: 99, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:51:16,816 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:51:33,423 INFO [optim.py:369] (1/2) 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 11:51:33,435 INFO [train.py:968] (1/2) Epoch 4, batch 43900, giga_loss[loss=0.368, simple_loss=0.4162, pruned_loss=0.1599, over 28905.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.4022, pruned_loss=0.1513, over 5671243.38 frames. ], libri_tot_loss[loss=0.3544, simple_loss=0.4025, pruned_loss=0.1532, over 5696701.65 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4024, pruned_loss=0.151, over 5665960.97 frames. ], batch size: 227, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:52:19,183 INFO [train.py:968] (1/2) Epoch 4, batch 43950, giga_loss[loss=0.3639, simple_loss=0.4112, pruned_loss=0.1583, over 28995.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.403, pruned_loss=0.1523, over 5668232.71 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4017, pruned_loss=0.1527, over 5694787.59 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4039, pruned_loss=0.1524, over 5665177.91 frames. ], batch size: 128, lr: 7.26e-03, grad_scale: 4.0
+2023-03-02 11:52:57,340 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:53:09,615 INFO [train.py:968] (1/2) Epoch 4, batch 44000, giga_loss[loss=0.3084, simple_loss=0.3711, pruned_loss=0.1228, over 29027.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4027, pruned_loss=0.1524, over 5670023.48 frames. ], libri_tot_loss[loss=0.353, simple_loss=0.4012, pruned_loss=0.1524, over 5694631.66 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4038, pruned_loss=0.1528, over 5667466.77 frames. ], batch size: 136, lr: 7.26e-03, grad_scale: 8.0
+2023-03-02 11:53:31,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6001, 4.3959, 4.2822, 2.0027], device='cuda:1'), covar=tensor([0.0357, 0.0370, 0.0640, 0.1907], device='cuda:1'), in_proj_covar=tensor([0.0839, 0.0739, 0.0820, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 11:53:50,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:53:53,212 INFO [train.py:968] (1/2) Epoch 4, batch 44050, giga_loss[loss=0.3213, simple_loss=0.3815, pruned_loss=0.1306, over 28915.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4001, pruned_loss=0.1506, over 5668168.53 frames. ], libri_tot_loss[loss=0.3531, simple_loss=0.4014, pruned_loss=0.1525, over 5689406.93 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4009, pruned_loss=0.1508, over 5669315.82 frames. ], batch size: 186, lr: 7.26e-03, grad_scale: 8.0
+2023-03-02 11:54:13,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:54:40,180 INFO [train.py:968] (1/2) Epoch 4, batch 44100, giga_loss[loss=0.3439, simple_loss=0.402, pruned_loss=0.1429, over 28702.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4005, pruned_loss=0.1505, over 5673700.25 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4016, pruned_loss=0.1526, over 5694961.39 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.4009, pruned_loss=0.1505, over 5668782.70 frames. ], batch size: 242, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:54:40,881 INFO [optim.py:369] (1/2) 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 11:54:53,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7114, 1.5730, 1.5635, 1.5161], device='cuda:1'), covar=tensor([0.0955, 0.1637, 0.1515, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0755, 0.0639, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 11:55:08,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:55:26,534 INFO [train.py:968] (1/2) Epoch 4, batch 44150, giga_loss[loss=0.326, simple_loss=0.3876, pruned_loss=0.1322, over 28935.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4024, pruned_loss=0.151, over 5675822.62 frames. ], libri_tot_loss[loss=0.3529, simple_loss=0.4014, pruned_loss=0.1522, over 5701093.03 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4029, pruned_loss=0.1513, over 5665310.12 frames. ], batch size: 227, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:55:34,441 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180299.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:55:38,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:55:49,440 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:56:15,746 INFO [train.py:968] (1/2) Epoch 4, batch 44200, giga_loss[loss=0.4238, simple_loss=0.4513, pruned_loss=0.1981, over 28254.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4049, pruned_loss=0.153, over 5682755.74 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4019, pruned_loss=0.1528, over 5704147.41 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4048, pruned_loss=0.1528, over 5671387.86 frames. ], batch size: 368, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:56:16,501 INFO [optim.py:369] (1/2) 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 11:56:29,908 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180355.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:56:33,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180358.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 11:56:58,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180387.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 11:57:03,217 INFO [train.py:968] (1/2) Epoch 4, batch 44250, giga_loss[loss=0.3728, simple_loss=0.4235, pruned_loss=0.1611, over 28606.00 frames. ], tot_loss[loss=0.3544, simple_loss=0.4042, pruned_loss=0.1523, over 5671975.93 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4019, pruned_loss=0.1526, over 5703619.41 frames. ], giga_tot_loss[loss=0.3543, simple_loss=0.4042, pruned_loss=0.1522, over 5662690.82 frames. ], batch size: 336, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:57:06,207 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 11:57:14,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:57:29,712 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:57:49,459 INFO [train.py:968] (1/2) Epoch 4, batch 44300, giga_loss[loss=0.3492, simple_loss=0.4067, pruned_loss=0.1459, over 28947.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4056, pruned_loss=0.1507, over 5675966.44 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4016, pruned_loss=0.1526, over 5704771.37 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.406, pruned_loss=0.1506, over 5666853.21 frames. ], batch size: 213, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:57:50,087 INFO [optim.py:369] (1/2) 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 11:58:32,512 INFO [train.py:968] (1/2) Epoch 4, batch 44350, giga_loss[loss=0.4594, simple_loss=0.4526, pruned_loss=0.2332, over 23624.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4074, pruned_loss=0.1497, over 5679707.58 frames. ], libri_tot_loss[loss=0.3538, simple_loss=0.402, pruned_loss=0.1528, over 5704715.61 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4075, pruned_loss=0.1494, over 5671812.61 frames. ], batch size: 705, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 11:58:36,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3646, 1.8529, 1.2046, 1.1706], device='cuda:1'), covar=tensor([0.1117, 0.0790, 0.0797, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.1385, 0.1149, 0.1124, 0.1231], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 11:59:20,069 INFO [train.py:968] (1/2) Epoch 4, batch 44400, giga_loss[loss=0.3245, simple_loss=0.3951, pruned_loss=0.1269, over 28980.00 frames. ], tot_loss[loss=0.3556, simple_loss=0.4096, pruned_loss=0.1508, over 5695142.26 frames. ], libri_tot_loss[loss=0.3536, simple_loss=0.4018, pruned_loss=0.1527, over 5710423.97 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.41, pruned_loss=0.1506, over 5683095.93 frames. ], batch size: 155, lr: 7.25e-03, grad_scale: 8.0
+2023-03-02 11:59:20,745 INFO [optim.py:369] (1/2) 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 11:59:22,736 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 11:59:23,381 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:59:25,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 11:59:40,825 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:00:06,389 INFO [train.py:968] (1/2) Epoch 4, batch 44450, giga_loss[loss=0.4356, simple_loss=0.4598, pruned_loss=0.2057, over 27529.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4123, pruned_loss=0.1548, over 5685218.30 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4018, pruned_loss=0.1526, over 5716071.62 frames. ], giga_tot_loss[loss=0.3611, simple_loss=0.413, pruned_loss=0.1546, over 5669574.45 frames. ], batch size: 472, lr: 7.25e-03, grad_scale: 8.0
+2023-03-02 12:00:12,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:00:15,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5957, 3.3673, 3.2348, 1.6686], device='cuda:1'), covar=tensor([0.0658, 0.0741, 0.1108, 0.2070], device='cuda:1'), in_proj_covar=tensor([0.0845, 0.0746, 0.0832, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 12:00:16,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 12:00:54,146 INFO [train.py:968] (1/2) Epoch 4, batch 44500, giga_loss[loss=0.3178, simple_loss=0.3885, pruned_loss=0.1236, over 28899.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.4124, pruned_loss=0.1561, over 5657897.70 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4016, pruned_loss=0.1526, over 5705320.52 frames. ], giga_tot_loss[loss=0.3628, simple_loss=0.4133, pruned_loss=0.1561, over 5654216.77 frames. ], batch size: 174, lr: 7.25e-03, grad_scale: 8.0
+2023-03-02 12:00:55,892 INFO [optim.py:369] (1/2) 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 12:01:10,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5842, 3.3760, 3.2706, 1.9242], device='cuda:1'), covar=tensor([0.0534, 0.0583, 0.0866, 0.1710], device='cuda:1'), in_proj_covar=tensor([0.0835, 0.0739, 0.0821, 0.0590], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 12:01:23,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 12:01:38,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:01:39,026 INFO [train.py:968] (1/2) Epoch 4, batch 44550, giga_loss[loss=0.2934, simple_loss=0.3675, pruned_loss=0.1096, over 28937.00 frames. ], tot_loss[loss=0.3619, simple_loss=0.4117, pruned_loss=0.156, over 5656441.04 frames. ], libri_tot_loss[loss=0.3528, simple_loss=0.4012, pruned_loss=0.1522, over 5698991.05 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4132, pruned_loss=0.1565, over 5658373.17 frames. ], batch size: 174, lr: 7.25e-03, grad_scale: 4.0
+2023-03-02 12:01:52,071 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:01:54,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:02:13,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:02:21,684 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:02:24,257 INFO [train.py:968] (1/2) Epoch 4, batch 44600, giga_loss[loss=0.3377, simple_loss=0.3963, pruned_loss=0.1395, over 28994.00 frames. ], tot_loss[loss=0.3594, simple_loss=0.4103, pruned_loss=0.1543, over 5652021.16 frames. ], libri_tot_loss[loss=0.3532, simple_loss=0.4015, pruned_loss=0.1525, over 5691716.52 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4112, pruned_loss=0.1544, over 5659692.78 frames. ], batch size: 136, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:02:25,571 INFO [optim.py:369] (1/2) 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 12:03:07,385 INFO [train.py:968] (1/2) Epoch 4, batch 44650, giga_loss[loss=0.3456, simple_loss=0.4052, pruned_loss=0.143, over 28277.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4107, pruned_loss=0.1523, over 5660698.54 frames. ], libri_tot_loss[loss=0.3533, simple_loss=0.4016, pruned_loss=0.1525, over 5694499.40 frames. ], giga_tot_loss[loss=0.3583, simple_loss=0.4116, pruned_loss=0.1525, over 5663342.65 frames. ], batch size: 368, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:03:08,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:03:30,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180817.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 12:03:32,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180820.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 12:03:45,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:03:47,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:03:50,696 INFO [train.py:968] (1/2) Epoch 4, batch 44700, giga_loss[loss=0.3077, simple_loss=0.3721, pruned_loss=0.1217, over 28776.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4102, pruned_loss=0.1514, over 5655593.33 frames. ], libri_tot_loss[loss=0.353, simple_loss=0.4014, pruned_loss=0.1523, over 5689823.33 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.4115, pruned_loss=0.1517, over 5660496.32 frames. ], batch size: 92, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:03:52,131 INFO [optim.py:369] (1/2) 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 12:03:57,689 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180849.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 12:04:01,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 12:04:18,247 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:04:36,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 12:04:42,260 INFO [train.py:968] (1/2) Epoch 4, batch 44750, giga_loss[loss=0.3424, simple_loss=0.394, pruned_loss=0.1455, over 28944.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.4105, pruned_loss=0.1525, over 5663666.50 frames. ], libri_tot_loss[loss=0.3527, simple_loss=0.4011, pruned_loss=0.1522, over 5693442.07 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4119, pruned_loss=0.1528, over 5663957.49 frames. ], batch size: 106, lr: 7.24e-03, grad_scale: 2.0
+2023-03-02 12:05:03,056 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-02 12:05:04,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 12:05:07,290 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-02 12:05:22,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=180937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:25,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=180940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:27,432 INFO [train.py:968] (1/2) Epoch 4, batch 44800, giga_loss[loss=0.4031, simple_loss=0.4475, pruned_loss=0.1793, over 28572.00 frames. ], tot_loss[loss=0.357, simple_loss=0.4099, pruned_loss=0.1521, over 5679735.29 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4018, pruned_loss=0.1525, over 5699804.33 frames. ], giga_tot_loss[loss=0.3575, simple_loss=0.4106, pruned_loss=0.1522, over 5673462.80 frames. ], batch size: 307, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:05:30,358 INFO [optim.py:369] (1/2) 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 12:05:45,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=180963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:51,433 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=180969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:05:55,169 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=180973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:06:15,090 INFO [train.py:968] (1/2) Epoch 4, batch 44850, giga_loss[loss=0.4555, simple_loss=0.4754, pruned_loss=0.2178, over 28891.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.409, pruned_loss=0.1533, over 5657647.70 frames. ], libri_tot_loss[loss=0.3535, simple_loss=0.4019, pruned_loss=0.1525, over 5700976.81 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4096, pruned_loss=0.1533, over 5651243.06 frames. ], batch size: 145, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:07:00,642 INFO [train.py:968] (1/2) Epoch 4, batch 44900, giga_loss[loss=0.3555, simple_loss=0.4109, pruned_loss=0.15, over 28765.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.4062, pruned_loss=0.1521, over 5654380.09 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4019, pruned_loss=0.1524, over 5695713.56 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4069, pruned_loss=0.1522, over 5652488.72 frames. ], batch size: 284, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:07:02,856 INFO [optim.py:369] (1/2) 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 12:07:22,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6630, 2.0239, 1.3721, 0.9985], device='cuda:1'), covar=tensor([0.1298, 0.0720, 0.0777, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.1367, 0.1141, 0.1123, 0.1216], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 12:07:50,971 INFO [train.py:968] (1/2) Epoch 4, batch 44950, giga_loss[loss=0.4455, simple_loss=0.4698, pruned_loss=0.2106, over 28312.00 frames. ], tot_loss[loss=0.3519, simple_loss=0.4031, pruned_loss=0.1503, over 5655247.33 frames. ], libri_tot_loss[loss=0.3534, simple_loss=0.4019, pruned_loss=0.1524, over 5696777.99 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4037, pruned_loss=0.1504, over 5652762.84 frames. ], batch size: 368, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:07:51,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8143, 4.5079, 1.7746, 1.6480], device='cuda:1'), covar=tensor([0.0808, 0.0206, 0.0774, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0481, 0.0314, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 12:07:58,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:11,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:14,159 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:33,465 INFO [train.py:968] (1/2) Epoch 4, batch 45000, giga_loss[loss=0.3332, simple_loss=0.3916, pruned_loss=0.1374, over 28940.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4015, pruned_loss=0.1497, over 5663731.77 frames. ], libri_tot_loss[loss=0.3537, simple_loss=0.4024, pruned_loss=0.1526, over 5702633.08 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.4015, pruned_loss=0.1496, over 5655398.21 frames. ], batch size: 106, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:08:33,465 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 12:08:41,615 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 12:08:44,620 INFO [optim.py:369] (1/2) 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 12:08:45,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5144, 1.7820, 1.2171, 0.9882], device='cuda:1'), covar=tensor([0.0848, 0.0614, 0.0564, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.1378, 0.1143, 0.1129, 0.1219], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 12:08:46,156 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:08:50,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2039, 1.4501, 1.2033, 1.3759], device='cuda:1'), covar=tensor([0.2126, 0.1946, 0.1886, 0.1958], device='cuda:1'), in_proj_covar=tensor([0.1108, 0.0873, 0.0983, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 12:09:25,338 INFO [train.py:968] (1/2) Epoch 4, batch 45050, giga_loss[loss=0.4575, simple_loss=0.4669, pruned_loss=0.224, over 26466.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4013, pruned_loss=0.1493, over 5670646.36 frames. ], libri_tot_loss[loss=0.3545, simple_loss=0.4031, pruned_loss=0.153, over 5707605.54 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4006, pruned_loss=0.1487, over 5658344.47 frames. ], batch size: 555, lr: 7.24e-03, grad_scale: 4.0
+2023-03-02 12:09:47,565 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:10:00,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1668, 5.9311, 5.7220, 2.3828], device='cuda:1'), covar=tensor([0.0318, 0.0450, 0.0820, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0834, 0.0741, 0.0810, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 12:10:11,105 INFO [train.py:968] (1/2) Epoch 4, batch 45100, giga_loss[loss=0.2912, simple_loss=0.3639, pruned_loss=0.1093, over 28862.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3981, pruned_loss=0.1458, over 5664747.45 frames. ], libri_tot_loss[loss=0.3542, simple_loss=0.4029, pruned_loss=0.1528, over 5708675.94 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3977, pruned_loss=0.1455, over 5652897.51 frames. ], batch size: 199, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:10:12,958 INFO [optim.py:369] (1/2) 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 12:10:14,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:10:16,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:10:40,668 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:10:53,371 INFO [train.py:968] (1/2) Epoch 4, batch 45150, giga_loss[loss=0.3469, simple_loss=0.4054, pruned_loss=0.1442, over 28944.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3968, pruned_loss=0.1439, over 5660742.46 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.4036, pruned_loss=0.1532, over 5691400.97 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3958, pruned_loss=0.143, over 5664209.05 frames. ], batch size: 227, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:11:43,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:11:47,388 INFO [train.py:968] (1/2) Epoch 4, batch 45200, giga_loss[loss=0.325, simple_loss=0.3861, pruned_loss=0.132, over 29003.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3954, pruned_loss=0.143, over 5650574.21 frames. ], libri_tot_loss[loss=0.355, simple_loss=0.4036, pruned_loss=0.1532, over 5691400.97 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3946, pruned_loss=0.1423, over 5653272.30 frames. ], batch size: 155, lr: 7.23e-03, grad_scale: 8.0
+2023-03-02 12:11:49,436 INFO [optim.py:369] (1/2) 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 12:12:34,508 INFO [train.py:968] (1/2) Epoch 4, batch 45250, giga_loss[loss=0.3409, simple_loss=0.3902, pruned_loss=0.1459, over 28340.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3928, pruned_loss=0.1424, over 5645969.07 frames. ], libri_tot_loss[loss=0.3546, simple_loss=0.4034, pruned_loss=0.153, over 5698682.59 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.392, pruned_loss=0.1418, over 5639986.48 frames. ], batch size: 368, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:13:24,341 INFO [train.py:968] (1/2) Epoch 4, batch 45300, giga_loss[loss=0.3505, simple_loss=0.4009, pruned_loss=0.1501, over 27884.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3928, pruned_loss=0.1432, over 5647570.53 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4035, pruned_loss=0.1531, over 5698005.68 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3918, pruned_loss=0.1424, over 5642590.19 frames. ], batch size: 412, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:13:28,404 INFO [optim.py:369] (1/2) 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 12:13:58,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:01,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:08,301 INFO [train.py:968] (1/2) Epoch 4, batch 45350, giga_loss[loss=0.3563, simple_loss=0.3847, pruned_loss=0.1639, over 23541.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3948, pruned_loss=0.144, over 5647949.00 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4037, pruned_loss=0.1531, over 5702563.17 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3937, pruned_loss=0.1431, over 5638563.86 frames. ], batch size: 705, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:14:15,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:27,468 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:14:30,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7117, 1.9695, 1.9236, 1.7290], device='cuda:1'), covar=tensor([0.1626, 0.1766, 0.1137, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0772, 0.0772, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 12:14:53,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9414, 5.1081, 1.9949, 2.1731], device='cuda:1'), covar=tensor([0.0721, 0.0275, 0.0769, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0482, 0.0312, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 12:14:55,416 INFO [train.py:968] (1/2) Epoch 4, batch 45400, giga_loss[loss=0.3478, simple_loss=0.4037, pruned_loss=0.1459, over 28919.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3951, pruned_loss=0.1435, over 5650387.14 frames. ], libri_tot_loss[loss=0.3549, simple_loss=0.4038, pruned_loss=0.153, over 5704852.78 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3941, pruned_loss=0.1427, over 5640104.00 frames. ], batch size: 227, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:15:01,691 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 4, batch 45450, giga_loss[loss=0.4002, simple_loss=0.4356, pruned_loss=0.1824, over 27501.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3947, pruned_loss=0.1439, over 5628971.24 frames. ], libri_tot_loss[loss=0.3555, simple_loss=0.4043, pruned_loss=0.1533, over 5706652.51 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3933, pruned_loss=0.1429, over 5618415.73 frames. ], batch size: 472, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:15:45,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=181593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:15:55,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2189, 1.8540, 1.4625, 0.4283], device='cuda:1'), covar=tensor([0.1471, 0.0920, 0.1515, 0.1690], device='cuda:1'), in_proj_covar=tensor([0.1365, 0.1260, 0.1330, 0.1135], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 12:16:33,244 INFO [train.py:968] (1/2) Epoch 4, batch 45500, libri_loss[loss=0.3506, simple_loss=0.4082, pruned_loss=0.1465, over 29538.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3943, pruned_loss=0.1438, over 5641733.22 frames. ], libri_tot_loss[loss=0.3553, simple_loss=0.4041, pruned_loss=0.1532, over 5708469.90 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3932, pruned_loss=0.143, over 5631151.51 frames. ], batch size: 83, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:16:33,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:16:37,621 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:16:43,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5780, 1.9744, 1.9018, 1.7157], device='cuda:1'), covar=tensor([0.1480, 0.1807, 0.1083, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0773, 0.0770, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 12:16:53,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3469, 4.1042, 4.0526, 1.9399], device='cuda:1'), covar=tensor([0.0371, 0.0419, 0.0624, 0.1895], device='cuda:1'), in_proj_covar=tensor([0.0850, 0.0751, 0.0823, 0.0598], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 12:17:02,793 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:17:16,009 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 4, batch 45550, giga_loss[loss=0.3591, simple_loss=0.4182, pruned_loss=0.15, over 28834.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3982, pruned_loss=0.1468, over 5637055.88 frames. ], libri_tot_loss[loss=0.3558, simple_loss=0.4046, pruned_loss=0.1535, over 5692425.95 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3969, pruned_loss=0.1458, over 5642226.48 frames. ], batch size: 284, lr: 7.23e-03, grad_scale: 4.0
+2023-03-02 12:17:58,434 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=181736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:18:02,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=181739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:18:06,617 INFO [train.py:968] (1/2) Epoch 4, batch 45600, giga_loss[loss=0.3248, simple_loss=0.3929, pruned_loss=0.1284, over 28898.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3997, pruned_loss=0.1473, over 5610783.80 frames. ], libri_tot_loss[loss=0.3566, simple_loss=0.4051, pruned_loss=0.154, over 5656796.72 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3981, pruned_loss=0.146, over 5644783.93 frames. ], batch size: 112, lr: 7.23e-03, grad_scale: 8.0
+2023-03-02 12:18:09,268 INFO [optim.py:369] (1/2) 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 12:18:19,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6960, 2.4569, 1.7826, 0.7474], device='cuda:1'), covar=tensor([0.2409, 0.1175, 0.1908, 0.2762], device='cuda:1'), in_proj_covar=tensor([0.1380, 0.1275, 0.1343, 0.1142], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 12:18:29,632 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=181768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:18:51,095 INFO [train.py:968] (1/2) Epoch 4, batch 45650, giga_loss[loss=0.4102, simple_loss=0.423, pruned_loss=0.1987, over 23535.00 frames. ], tot_loss[loss=0.352, simple_loss=0.4029, pruned_loss=0.1506, over 5568260.02 frames. ], libri_tot_loss[loss=0.358, simple_loss=0.406, pruned_loss=0.155, over 5602625.34 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.4007, pruned_loss=0.1485, over 5645321.66 frames. ], batch size: 705, lr: 7.22e-03, grad_scale: 8.0
+2023-03-02 12:19:39,962 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 12:19:41,134 INFO [train.py:968] (1/2) Epoch 4, batch 45700, giga_loss[loss=0.3452, simple_loss=0.3979, pruned_loss=0.1463, over 28920.00 frames. ], tot_loss[loss=0.3553, simple_loss=0.4049, pruned_loss=0.1528, over 5560035.96 frames. ], libri_tot_loss[loss=0.359, simple_loss=0.4067, pruned_loss=0.1557, over 5570367.04 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4024, pruned_loss=0.1505, over 5650016.40 frames. ], batch size: 213, lr: 7.22e-03, grad_scale: 2.0
+2023-03-02 12:19:46,488 INFO [optim.py:369] (1/2) 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 12:19:53,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4107, 1.7141, 1.6946, 1.5967], device='cuda:1'), covar=tensor([0.1353, 0.1923, 0.1126, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0765, 0.0763, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 12:20:14,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1949, 1.3029, 0.9869, 0.8464], device='cuda:1'), covar=tensor([0.0869, 0.0741, 0.0622, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.1391, 0.1152, 0.1146, 0.1245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 12:20:18,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 12:20:34,111 INFO [train.py:968] (1/2) Epoch 4, batch 45750, giga_loss[loss=0.3479, simple_loss=0.3957, pruned_loss=0.1501, over 28495.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4058, pruned_loss=0.1514, over 5567054.46 frames. ], libri_tot_loss[loss=0.3594, simple_loss=0.4069, pruned_loss=0.1559, over 5554678.54 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4036, pruned_loss=0.1493, over 5651959.67 frames. ], batch size: 60, lr: 7.22e-03, grad_scale: 2.0
+2023-03-02 12:20:43,483 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-02 12:21:37,984 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=181923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:21:53,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-02 12:22:02,379 INFO [optim.py:369] (1/2) 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 12:22:03,019 INFO [train.py:968] (1/2) Epoch 5, batch 50, giga_loss[loss=0.3397, simple_loss=0.4003, pruned_loss=0.1396, over 28969.00 frames. ], tot_loss[loss=0.337, simple_loss=0.4037, pruned_loss=0.1351, over 1265981.38 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3723, pruned_loss=0.1164, over 115562.37 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.4067, pruned_loss=0.1369, over 1173910.21 frames. ], batch size: 106, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:22:49,361 INFO [train.py:968] (1/2) Epoch 5, batch 100, giga_loss[loss=0.3309, simple_loss=0.3966, pruned_loss=0.1326, over 28593.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3936, pruned_loss=0.1303, over 2235143.58 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3737, pruned_loss=0.1166, over 333013.65 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3968, pruned_loss=0.1324, over 2020233.61 frames. ], batch size: 307, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:23:33,334 INFO [optim.py:369] (1/2) 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 12:23:34,091 INFO [train.py:968] (1/2) Epoch 5, batch 150, giga_loss[loss=0.2912, simple_loss=0.3469, pruned_loss=0.1178, over 28396.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3765, pruned_loss=0.1213, over 2994990.28 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3783, pruned_loss=0.1202, over 459974.41 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3767, pruned_loss=0.1216, over 2760598.80 frames. ], batch size: 65, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:23:47,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:24:15,559 INFO [train.py:968] (1/2) Epoch 5, batch 200, giga_loss[loss=0.2649, simple_loss=0.342, pruned_loss=0.09393, over 28958.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3617, pruned_loss=0.1136, over 3599920.23 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3775, pruned_loss=0.1193, over 594973.54 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3607, pruned_loss=0.1134, over 3354731.65 frames. ], batch size: 164, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:24:20,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-02 12:24:56,493 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 250, giga_loss[loss=0.2529, simple_loss=0.322, pruned_loss=0.09193, over 29016.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3484, pruned_loss=0.1062, over 4061418.62 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3764, pruned_loss=0.1185, over 672386.54 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3467, pruned_loss=0.1056, over 3841803.54 frames. ], batch size: 128, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:25:38,096 INFO [train.py:968] (1/2) Epoch 5, batch 300, giga_loss[loss=0.2288, simple_loss=0.2998, pruned_loss=0.07885, over 28874.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3398, pruned_loss=0.102, over 4421440.19 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3747, pruned_loss=0.1169, over 824421.98 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3372, pruned_loss=0.1011, over 4203629.96 frames. ], batch size: 213, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:25:45,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:25:47,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:26:20,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:26:27,801 INFO [train.py:968] (1/2) Epoch 5, batch 350, giga_loss[loss=0.2178, simple_loss=0.2946, pruned_loss=0.07053, over 28918.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3302, pruned_loss=0.097, over 4691796.53 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3725, pruned_loss=0.1157, over 890651.39 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3277, pruned_loss=0.09607, over 4510925.41 frames. ], batch size: 186, lr: 6.72e-03, grad_scale: 2.0
+2023-03-02 12:26:42,600 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-02 12:27:07,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:27:08,434 INFO [train.py:968] (1/2) Epoch 5, batch 400, giga_loss[loss=0.2426, simple_loss=0.3103, pruned_loss=0.08739, over 28849.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3273, pruned_loss=0.09565, over 4921287.90 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.377, pruned_loss=0.1195, over 1014296.58 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3231, pruned_loss=0.09355, over 4752767.29 frames. ], batch size: 186, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:27:49,513 INFO [optim.py:369] (1/2) 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 12:27:50,469 INFO [train.py:968] (1/2) Epoch 5, batch 450, giga_loss[loss=0.3327, simple_loss=0.3696, pruned_loss=0.1479, over 26649.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.325, pruned_loss=0.09459, over 5094974.69 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3756, pruned_loss=0.1181, over 1111076.01 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3209, pruned_loss=0.09273, over 4945606.73 frames. ], batch size: 555, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:28:09,457 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:28:33,390 INFO [train.py:968] (1/2) Epoch 5, batch 500, giga_loss[loss=0.2407, simple_loss=0.3121, pruned_loss=0.08468, over 28203.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3233, pruned_loss=0.09364, over 5227258.24 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3766, pruned_loss=0.1189, over 1275294.95 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3181, pruned_loss=0.09116, over 5084423.37 frames. ], batch size: 368, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:28:44,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4737, 2.2717, 1.6821, 0.5193], device='cuda:1'), covar=tensor([0.2336, 0.1143, 0.1935, 0.2485], device='cuda:1'), in_proj_covar=tensor([0.1347, 0.1250, 0.1317, 0.1111], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 12:29:10,851 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:29:12,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=182444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:29:15,560 INFO [optim.py:369] (1/2) 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 12:29:16,323 INFO [train.py:968] (1/2) Epoch 5, batch 550, giga_loss[loss=0.211, simple_loss=0.2825, pruned_loss=0.06976, over 28418.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3218, pruned_loss=0.0928, over 5332697.44 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3777, pruned_loss=0.119, over 1388172.38 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.316, pruned_loss=0.09014, over 5206008.48 frames. ], batch size: 71, lr: 6.72e-03, grad_scale: 4.0
+2023-03-02 12:29:34,316 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:29:47,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7846, 1.7884, 1.7109, 1.6588], device='cuda:1'), covar=tensor([0.1297, 0.1746, 0.1659, 0.1536], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0751, 0.0637, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 12:29:52,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-02 12:29:58,209 INFO [train.py:968] (1/2) Epoch 5, batch 600, giga_loss[loss=0.2285, simple_loss=0.304, pruned_loss=0.07653, over 28935.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3193, pruned_loss=0.09125, over 5409946.16 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3764, pruned_loss=0.1181, over 1537775.16 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3132, pruned_loss=0.08849, over 5296728.92 frames. ], batch size: 227, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:30:26,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4642, 1.6500, 1.6118, 1.5817], device='cuda:1'), covar=tensor([0.1164, 0.1545, 0.1206, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0758, 0.0643, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 12:30:40,553 INFO [optim.py:369] (1/2) 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 12:30:42,842 INFO [train.py:968] (1/2) Epoch 5, batch 650, giga_loss[loss=0.251, simple_loss=0.3101, pruned_loss=0.09597, over 28842.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3178, pruned_loss=0.09049, over 5474925.03 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3762, pruned_loss=0.1182, over 1683466.05 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.311, pruned_loss=0.08733, over 5373872.34 frames. ], batch size: 92, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:31:28,147 INFO [train.py:968] (1/2) Epoch 5, batch 700, giga_loss[loss=0.1993, simple_loss=0.2765, pruned_loss=0.06106, over 28682.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3153, pruned_loss=0.08935, over 5518440.23 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3769, pruned_loss=0.1188, over 1758361.42 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3086, pruned_loss=0.08618, over 5439059.62 frames. ], batch size: 262, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:32:10,107 INFO [optim.py:369] (1/2) 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 12:32:10,808 INFO [train.py:968] (1/2) Epoch 5, batch 750, giga_loss[loss=0.1887, simple_loss=0.2611, pruned_loss=0.05816, over 28569.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3129, pruned_loss=0.08783, over 5554152.56 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3765, pruned_loss=0.1186, over 1899174.80 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3056, pruned_loss=0.08433, over 5480592.68 frames. ], batch size: 85, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:32:32,669 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:32:53,371 INFO [train.py:968] (1/2) Epoch 5, batch 800, giga_loss[loss=0.213, simple_loss=0.2873, pruned_loss=0.06936, over 29102.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3102, pruned_loss=0.08676, over 5587092.03 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.377, pruned_loss=0.1187, over 1978394.30 frames. ], giga_tot_loss[loss=0.2349, simple_loss=0.303, pruned_loss=0.08336, over 5523031.88 frames. ], batch size: 128, lr: 6.71e-03, grad_scale: 8.0
+2023-03-02 12:33:21,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3402, 1.9705, 1.7956, 1.6407], device='cuda:1'), covar=tensor([0.1440, 0.1734, 0.1110, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0768, 0.0782, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 12:33:39,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=182746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:33:41,704 INFO [optim.py:369] (1/2) 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 12:33:42,316 INFO [train.py:968] (1/2) Epoch 5, batch 850, giga_loss[loss=0.3065, simple_loss=0.377, pruned_loss=0.118, over 28944.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3184, pruned_loss=0.09176, over 5608129.21 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.377, pruned_loss=0.1188, over 1998008.32 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3126, pruned_loss=0.08901, over 5556639.99 frames. ], batch size: 145, lr: 6.71e-03, grad_scale: 8.0
+2023-03-02 12:34:28,563 INFO [train.py:968] (1/2) Epoch 5, batch 900, giga_loss[loss=0.3219, simple_loss=0.3798, pruned_loss=0.132, over 28717.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3336, pruned_loss=0.1001, over 5625131.65 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.377, pruned_loss=0.1185, over 2144101.53 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3271, pruned_loss=0.09714, over 5580318.85 frames. ], batch size: 99, lr: 6.71e-03, grad_scale: 8.0
+2023-03-02 12:35:11,949 INFO [optim.py:369] (1/2) 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 12:35:11,961 INFO [train.py:968] (1/2) Epoch 5, batch 950, giga_loss[loss=0.327, simple_loss=0.3942, pruned_loss=0.1299, over 28557.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3467, pruned_loss=0.1079, over 5637129.18 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3749, pruned_loss=0.1173, over 2218585.14 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3417, pruned_loss=0.1058, over 5596969.16 frames. ], batch size: 336, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:35:46,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=182889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:35:48,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 1000, giga_loss[loss=0.3272, simple_loss=0.395, pruned_loss=0.1297, over 28286.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3545, pruned_loss=0.1113, over 5647033.64 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3746, pruned_loss=0.117, over 2255359.88 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3505, pruned_loss=0.1096, over 5613215.65 frames. ], batch size: 368, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:36:01,947 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=182907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:36:13,199 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=182921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:36:34,155 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 1050, giga_loss[loss=0.2596, simple_loss=0.3467, pruned_loss=0.08629, over 28810.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3587, pruned_loss=0.1114, over 5666845.61 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3759, pruned_loss=0.1175, over 2345438.01 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3547, pruned_loss=0.1098, over 5634544.20 frames. ], batch size: 186, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:37:21,814 INFO [train.py:968] (1/2) Epoch 5, batch 1100, giga_loss[loss=0.2978, simple_loss=0.3706, pruned_loss=0.1125, over 28791.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3596, pruned_loss=0.1109, over 5667051.96 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3756, pruned_loss=0.1173, over 2416200.21 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3562, pruned_loss=0.1096, over 5636739.00 frames. ], batch size: 186, lr: 6.71e-03, grad_scale: 4.0
+2023-03-02 12:37:22,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-02 12:38:05,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:38:05,703 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 1150, giga_loss[loss=0.3358, simple_loss=0.405, pruned_loss=0.1333, over 28942.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3622, pruned_loss=0.1128, over 5669033.17 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3752, pruned_loss=0.1169, over 2500808.78 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1118, over 5640841.69 frames. ], batch size: 227, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:38:49,608 INFO [train.py:968] (1/2) Epoch 5, batch 1200, giga_loss[loss=0.3241, simple_loss=0.3897, pruned_loss=0.1292, over 28836.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3648, pruned_loss=0.1147, over 5678028.86 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3746, pruned_loss=0.1164, over 2585270.26 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3625, pruned_loss=0.1141, over 5650427.39 frames. ], batch size: 186, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:39:13,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-02 12:39:33,814 INFO [optim.py:369] (1/2) 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 12:39:33,826 INFO [train.py:968] (1/2) Epoch 5, batch 1250, giga_loss[loss=0.2929, simple_loss=0.3761, pruned_loss=0.1049, over 28981.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3675, pruned_loss=0.1166, over 5686926.01 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3739, pruned_loss=0.116, over 2666882.37 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3658, pruned_loss=0.1162, over 5660259.98 frames. ], batch size: 164, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:40:09,647 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:40:13,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:40:16,355 INFO [train.py:968] (1/2) Epoch 5, batch 1300, libri_loss[loss=0.2819, simple_loss=0.3576, pruned_loss=0.1031, over 29545.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3704, pruned_loss=0.1169, over 5695620.17 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3735, pruned_loss=0.1151, over 2777250.08 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3691, pruned_loss=0.1171, over 5668466.88 frames. ], batch size: 76, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:40:35,612 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:40:56,114 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 1350, giga_loss[loss=0.284, simple_loss=0.3632, pruned_loss=0.1023, over 28591.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3725, pruned_loss=0.1176, over 5697018.24 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3731, pruned_loss=0.1147, over 2883565.03 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3716, pruned_loss=0.118, over 5669864.34 frames. ], batch size: 60, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:41:00,936 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:41:24,281 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 1400, giga_loss[loss=0.3238, simple_loss=0.3947, pruned_loss=0.1265, over 29057.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3743, pruned_loss=0.1179, over 5700471.32 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3736, pruned_loss=0.115, over 2928634.51 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3733, pruned_loss=0.1182, over 5676159.85 frames. ], batch size: 136, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:42:20,664 INFO [train.py:968] (1/2) Epoch 5, batch 1450, giga_loss[loss=0.3421, simple_loss=0.3841, pruned_loss=0.1501, over 26569.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3732, pruned_loss=0.116, over 5706370.14 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3734, pruned_loss=0.1147, over 3029076.24 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3725, pruned_loss=0.1164, over 5682121.35 frames. ], batch size: 555, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:42:21,383 INFO [optim.py:369] (1/2) 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 12:43:01,173 INFO [train.py:968] (1/2) Epoch 5, batch 1500, giga_loss[loss=0.3194, simple_loss=0.3862, pruned_loss=0.1263, over 28991.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3733, pruned_loss=0.1155, over 5702486.39 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3737, pruned_loss=0.1148, over 3076484.75 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3726, pruned_loss=0.1157, over 5687627.08 frames. ], batch size: 106, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:43:21,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:43:23,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:43:38,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7686, 1.6842, 1.6254, 1.5554], device='cuda:1'), covar=tensor([0.1172, 0.1787, 0.1548, 0.1579], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0746, 0.0634, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 12:43:42,984 INFO [train.py:968] (1/2) Epoch 5, batch 1550, giga_loss[loss=0.3891, simple_loss=0.4238, pruned_loss=0.1772, over 26596.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3712, pruned_loss=0.1135, over 5710444.60 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3736, pruned_loss=0.1146, over 3132615.34 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3706, pruned_loss=0.1138, over 5695213.33 frames. ], batch size: 555, lr: 6.70e-03, grad_scale: 4.0
+2023-03-02 12:43:43,514 INFO [optim.py:369] (1/2) 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 12:43:48,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:44:05,091 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8469, 1.2165, 3.9615, 3.0816], device='cuda:1'), covar=tensor([0.1598, 0.2130, 0.0362, 0.0628], device='cuda:1'), in_proj_covar=tensor([0.0560, 0.0519, 0.0729, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 12:44:27,755 INFO [train.py:968] (1/2) Epoch 5, batch 1600, libri_loss[loss=0.3246, simple_loss=0.3971, pruned_loss=0.126, over 27955.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3725, pruned_loss=0.1158, over 5700030.73 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3731, pruned_loss=0.1143, over 3238490.52 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3723, pruned_loss=0.1162, over 5683438.16 frames. ], batch size: 116, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:44:38,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3137, 1.9368, 1.4939, 0.4879], device='cuda:1'), covar=tensor([0.2046, 0.1274, 0.1851, 0.2508], device='cuda:1'), in_proj_covar=tensor([0.1363, 0.1263, 0.1353, 0.1129], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 12:45:12,586 INFO [train.py:968] (1/2) Epoch 5, batch 1650, giga_loss[loss=0.3571, simple_loss=0.4028, pruned_loss=0.1558, over 29075.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3757, pruned_loss=0.1206, over 5705297.17 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3734, pruned_loss=0.1145, over 3250957.07 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3753, pruned_loss=0.1209, over 5692201.08 frames. ], batch size: 128, lr: 6.70e-03, grad_scale: 8.0
+2023-03-02 12:45:13,193 INFO [optim.py:369] (1/2) 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 12:45:44,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7526, 1.6163, 1.6373, 1.6197], device='cuda:1'), covar=tensor([0.1076, 0.1748, 0.1604, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0750, 0.0636, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 12:45:58,414 INFO [train.py:968] (1/2) Epoch 5, batch 1700, giga_loss[loss=0.3063, simple_loss=0.3749, pruned_loss=0.1189, over 28581.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3775, pruned_loss=0.1236, over 5712961.10 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3731, pruned_loss=0.1141, over 3315815.14 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3774, pruned_loss=0.1242, over 5698946.15 frames. ], batch size: 307, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:46:05,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0473, 1.1324, 0.9273, 0.6464], device='cuda:1'), covar=tensor([0.0718, 0.0760, 0.0542, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.1367, 0.1141, 0.1147, 0.1218], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 12:46:16,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:46:21,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=183630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:46:40,427 INFO [train.py:968] (1/2) Epoch 5, batch 1750, giga_loss[loss=0.2723, simple_loss=0.3438, pruned_loss=0.1004, over 28227.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3758, pruned_loss=0.1234, over 5704349.64 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3734, pruned_loss=0.1142, over 3402885.42 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3757, pruned_loss=0.1241, over 5689009.68 frames. ], batch size: 77, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:46:41,745 INFO [optim.py:369] (1/2) 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 12:47:09,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2597, 1.3793, 1.3616, 1.3382], device='cuda:1'), covar=tensor([0.0835, 0.1036, 0.1329, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0749, 0.0637, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 12:47:22,690 INFO [train.py:968] (1/2) Epoch 5, batch 1800, giga_loss[loss=0.3367, simple_loss=0.3919, pruned_loss=0.1407, over 28311.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3738, pruned_loss=0.123, over 5697850.91 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3729, pruned_loss=0.114, over 3450807.95 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3741, pruned_loss=0.1238, over 5683525.69 frames. ], batch size: 368, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:48:05,660 INFO [train.py:968] (1/2) Epoch 5, batch 1850, giga_loss[loss=0.405, simple_loss=0.448, pruned_loss=0.181, over 27937.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3732, pruned_loss=0.1223, over 5692892.67 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.373, pruned_loss=0.114, over 3475310.47 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.123, over 5679693.85 frames. ], batch size: 412, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:48:08,069 INFO [optim.py:369] (1/2) 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 12:48:21,163 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:23,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:26,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=183773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:28,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=183776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:30,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3967, 1.4420, 1.1901, 1.7261], device='cuda:1'), covar=tensor([0.2190, 0.2016, 0.2066, 0.2313], device='cuda:1'), in_proj_covar=tensor([0.1112, 0.0862, 0.0980, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 12:48:51,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:52,243 INFO [train.py:968] (1/2) Epoch 5, batch 1900, giga_loss[loss=0.3877, simple_loss=0.4174, pruned_loss=0.179, over 26589.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3717, pruned_loss=0.1206, over 5698268.92 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.373, pruned_loss=0.114, over 3546730.55 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3718, pruned_loss=0.1214, over 5682656.52 frames. ], batch size: 555, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:48:53,201 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=183801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:48:55,877 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=183805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:49:35,284 INFO [train.py:968] (1/2) Epoch 5, batch 1950, giga_loss[loss=0.2534, simple_loss=0.3258, pruned_loss=0.09047, over 28187.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.367, pruned_loss=0.1172, over 5691479.14 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3727, pruned_loss=0.1138, over 3630751.86 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3672, pruned_loss=0.1181, over 5679204.46 frames. ], batch size: 77, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:49:36,647 INFO [optim.py:369] (1/2) 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 12:50:19,171 INFO [train.py:968] (1/2) Epoch 5, batch 2000, giga_loss[loss=0.2873, simple_loss=0.3534, pruned_loss=0.1106, over 28975.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3605, pruned_loss=0.1129, over 5687224.28 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3726, pruned_loss=0.1135, over 3707509.20 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1138, over 5671826.26 frames. ], batch size: 128, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:50:44,816 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-02 12:51:05,760 INFO [train.py:968] (1/2) Epoch 5, batch 2050, giga_loss[loss=0.2752, simple_loss=0.3392, pruned_loss=0.1056, over 28601.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3552, pruned_loss=0.1101, over 5672546.15 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3729, pruned_loss=0.1136, over 3715842.63 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3548, pruned_loss=0.1108, over 5670443.05 frames. ], batch size: 336, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:51:09,857 INFO [optim.py:369] (1/2) 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 12:51:57,041 INFO [train.py:968] (1/2) Epoch 5, batch 2100, giga_loss[loss=0.2671, simple_loss=0.3201, pruned_loss=0.107, over 23568.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3532, pruned_loss=0.1097, over 5653290.56 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3732, pruned_loss=0.1138, over 3737914.28 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3525, pruned_loss=0.1101, over 5649253.39 frames. ], batch size: 705, lr: 6.69e-03, grad_scale: 8.0
+2023-03-02 12:52:03,197 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:52:33,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-02 12:52:36,012 INFO [train.py:968] (1/2) Epoch 5, batch 2150, giga_loss[loss=0.2688, simple_loss=0.339, pruned_loss=0.09933, over 28881.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3548, pruned_loss=0.1098, over 5674750.65 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3731, pruned_loss=0.1138, over 3823433.01 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3536, pruned_loss=0.11, over 5662762.96 frames. ], batch size: 112, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:52:37,914 INFO [optim.py:369] (1/2) 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 12:52:44,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3789, 2.8615, 1.4656, 1.4034], device='cuda:1'), covar=tensor([0.0892, 0.0311, 0.0815, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0463, 0.0307, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 12:52:54,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-02 12:52:59,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1227, 4.6801, 2.2117, 2.1161], device='cuda:1'), covar=tensor([0.0801, 0.0280, 0.0715, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0464, 0.0307, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 12:53:17,245 INFO [train.py:968] (1/2) Epoch 5, batch 2200, giga_loss[loss=0.2703, simple_loss=0.3432, pruned_loss=0.09873, over 29077.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3538, pruned_loss=0.1088, over 5690001.32 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3741, pruned_loss=0.114, over 3872234.29 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3518, pruned_loss=0.1087, over 5677816.34 frames. ], batch size: 155, lr: 6.69e-03, grad_scale: 4.0
+2023-03-02 12:54:02,509 INFO [train.py:968] (1/2) Epoch 5, batch 2250, giga_loss[loss=0.3502, simple_loss=0.3916, pruned_loss=0.1544, over 26768.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3514, pruned_loss=0.1076, over 5693108.81 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3749, pruned_loss=0.1143, over 3902822.15 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.349, pruned_loss=0.1073, over 5680351.25 frames. ], batch size: 555, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:54:03,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-02 12:54:04,640 INFO [optim.py:369] (1/2) 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 12:54:23,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:54:43,285 INFO [train.py:968] (1/2) Epoch 5, batch 2300, giga_loss[loss=0.2422, simple_loss=0.322, pruned_loss=0.08116, over 29095.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3492, pruned_loss=0.1063, over 5707307.29 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3756, pruned_loss=0.1146, over 3961231.92 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3461, pruned_loss=0.1056, over 5692298.91 frames. ], batch size: 155, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:54:58,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7853, 1.1543, 1.0442, 1.0459], device='cuda:1'), covar=tensor([0.1333, 0.1168, 0.1788, 0.1377], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0747, 0.0638, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 12:55:04,302 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-02 12:55:12,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:55:25,094 INFO [train.py:968] (1/2) Epoch 5, batch 2350, giga_loss[loss=0.249, simple_loss=0.3235, pruned_loss=0.08722, over 28809.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3459, pruned_loss=0.1044, over 5707450.18 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3751, pruned_loss=0.1142, over 3999025.70 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3432, pruned_loss=0.104, over 5692689.89 frames. ], batch size: 199, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:55:27,911 INFO [optim.py:369] (1/2) 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 12:55:57,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5497, 1.9889, 1.7096, 1.6609], device='cuda:1'), covar=tensor([0.0785, 0.0291, 0.0302, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0126, 0.0130, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0053, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 12:56:09,562 INFO [train.py:968] (1/2) Epoch 5, batch 2400, giga_loss[loss=0.2295, simple_loss=0.3004, pruned_loss=0.07932, over 28718.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3432, pruned_loss=0.1034, over 5705343.42 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3758, pruned_loss=0.1145, over 4026723.88 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3403, pruned_loss=0.1027, over 5691837.36 frames. ], batch size: 71, lr: 6.68e-03, grad_scale: 8.0
+2023-03-02 12:56:24,742 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:56:26,354 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-02 12:56:26,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:56:37,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2688, 3.3776, 1.3047, 1.3786], device='cuda:1'), covar=tensor([0.1025, 0.0372, 0.0936, 0.1429], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0469, 0.0309, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 12:56:49,814 INFO [train.py:968] (1/2) Epoch 5, batch 2450, giga_loss[loss=0.2413, simple_loss=0.3135, pruned_loss=0.0845, over 28903.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3409, pruned_loss=0.1018, over 5715272.29 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3765, pruned_loss=0.1149, over 4081277.05 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3371, pruned_loss=0.1007, over 5700074.22 frames. ], batch size: 145, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:56:50,793 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:56:52,691 INFO [optim.py:369] (1/2) 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 12:56:57,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 12:57:13,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:57:14,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-02 12:57:27,968 INFO [train.py:968] (1/2) Epoch 5, batch 2500, libri_loss[loss=0.3043, simple_loss=0.3798, pruned_loss=0.1144, over 29535.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3389, pruned_loss=0.1008, over 5723230.10 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3759, pruned_loss=0.1142, over 4144270.23 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3351, pruned_loss=0.09982, over 5705130.11 frames. ], batch size: 84, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:58:03,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:58:08,804 INFO [train.py:968] (1/2) Epoch 5, batch 2550, giga_loss[loss=0.2557, simple_loss=0.3294, pruned_loss=0.09099, over 28976.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3375, pruned_loss=0.09978, over 5732105.21 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3767, pruned_loss=0.1146, over 4196806.02 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3329, pruned_loss=0.09841, over 5712706.98 frames. ], batch size: 174, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:58:11,368 INFO [optim.py:369] (1/2) 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 12:58:52,071 INFO [train.py:968] (1/2) Epoch 5, batch 2600, giga_loss[loss=0.2507, simple_loss=0.329, pruned_loss=0.08623, over 28285.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3366, pruned_loss=0.09896, over 5722837.79 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3776, pruned_loss=0.1151, over 4229317.64 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3316, pruned_loss=0.09729, over 5711685.84 frames. ], batch size: 368, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:59:11,488 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:59:13,974 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 12:59:31,971 INFO [train.py:968] (1/2) Epoch 5, batch 2650, giga_loss[loss=0.3482, simple_loss=0.3931, pruned_loss=0.1517, over 26629.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3356, pruned_loss=0.09861, over 5719854.22 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.378, pruned_loss=0.1151, over 4261205.29 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3305, pruned_loss=0.0969, over 5715539.39 frames. ], batch size: 555, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 12:59:35,059 INFO [optim.py:369] (1/2) 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 12:59:40,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:00:15,960 INFO [train.py:968] (1/2) Epoch 5, batch 2700, giga_loss[loss=0.2768, simple_loss=0.3495, pruned_loss=0.1021, over 28837.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3403, pruned_loss=0.1017, over 5724522.84 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3791, pruned_loss=0.1155, over 4294195.78 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3347, pruned_loss=0.09979, over 5717499.53 frames. ], batch size: 186, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 13:00:19,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7806, 2.8023, 1.9152, 0.7095], device='cuda:1'), covar=tensor([0.3102, 0.1510, 0.1736, 0.3120], device='cuda:1'), in_proj_covar=tensor([0.1360, 0.1256, 0.1342, 0.1118], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 13:00:26,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:00:59,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2968, 1.1603, 1.0427, 1.4350], device='cuda:1'), covar=tensor([0.0771, 0.0372, 0.0350, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0125, 0.0129, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0053, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 13:01:02,721 INFO [train.py:968] (1/2) Epoch 5, batch 2750, libri_loss[loss=0.31, simple_loss=0.3911, pruned_loss=0.1145, over 26192.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3461, pruned_loss=0.1055, over 5717228.68 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3795, pruned_loss=0.1155, over 4330127.28 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3406, pruned_loss=0.1037, over 5711323.15 frames. ], batch size: 136, lr: 6.68e-03, grad_scale: 4.0
+2023-03-02 13:01:06,010 INFO [optim.py:369] (1/2) 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 13:01:34,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3737, 1.5257, 1.1713, 1.5311], device='cuda:1'), covar=tensor([0.0800, 0.0339, 0.0341, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0126, 0.0130, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 13:01:50,306 INFO [train.py:968] (1/2) Epoch 5, batch 2800, giga_loss[loss=0.2644, simple_loss=0.3392, pruned_loss=0.09482, over 28450.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3533, pruned_loss=0.1105, over 5710891.87 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3787, pruned_loss=0.1151, over 4353638.64 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3491, pruned_loss=0.1093, over 5703731.77 frames. ], batch size: 71, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:02:34,492 INFO [train.py:968] (1/2) Epoch 5, batch 2850, giga_loss[loss=0.3074, simple_loss=0.3749, pruned_loss=0.1199, over 28873.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3624, pruned_loss=0.117, over 5699412.12 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3792, pruned_loss=0.1154, over 4397096.70 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.358, pruned_loss=0.1158, over 5696168.96 frames. ], batch size: 99, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:02:36,049 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:02:37,027 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:02:41,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:02:56,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7389, 1.9326, 1.6375, 1.5939], device='cuda:1'), covar=tensor([0.1358, 0.1996, 0.1706, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0739, 0.0634, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 13:03:08,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:03:20,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4125, 2.4938, 2.3147, 2.2551], device='cuda:1'), covar=tensor([0.1115, 0.1570, 0.1245, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0737, 0.0634, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 13:03:23,495 INFO [train.py:968] (1/2) Epoch 5, batch 2900, libri_loss[loss=0.332, simple_loss=0.3947, pruned_loss=0.1346, over 29255.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1183, over 5707724.90 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3795, pruned_loss=0.1158, over 4430415.29 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3631, pruned_loss=0.1171, over 5704023.30 frames. ], batch size: 94, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:03:41,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=184818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:04:06,562 INFO [train.py:968] (1/2) Epoch 5, batch 2950, giga_loss[loss=0.2883, simple_loss=0.3647, pruned_loss=0.1059, over 28707.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3727, pruned_loss=0.1215, over 5696277.10 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3795, pruned_loss=0.116, over 4455179.73 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3692, pruned_loss=0.1204, over 5699834.68 frames. ], batch size: 262, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:04:10,312 INFO [optim.py:369] (1/2) 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 13:04:37,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=184879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:04:53,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-02 13:04:54,682 INFO [train.py:968] (1/2) Epoch 5, batch 3000, giga_loss[loss=0.2921, simple_loss=0.3692, pruned_loss=0.1074, over 28944.00 frames. ], tot_loss[loss=0.316, simple_loss=0.379, pruned_loss=0.1265, over 5679694.62 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3794, pruned_loss=0.116, over 4490802.61 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3762, pruned_loss=0.1259, over 5677746.19 frames. ], batch size: 164, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:04:54,682 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 13:05:02,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6222, 1.5691, 1.2693, 1.3901], device='cuda:1'), covar=tensor([0.0602, 0.0528, 0.0986, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0454, 0.0514, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:05:03,544 INFO [train.py:1012] (1/2) Epoch 5, validation: loss=0.2496, simple_loss=0.3486, pruned_loss=0.0753, over 944034.00 frames. 
+2023-03-02 13:05:03,544 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 13:05:45,976 INFO [train.py:968] (1/2) Epoch 5, batch 3050, giga_loss[loss=0.2604, simple_loss=0.3332, pruned_loss=0.09382, over 28910.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.374, pruned_loss=0.1225, over 5691785.42 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3792, pruned_loss=0.1161, over 4525991.20 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3718, pruned_loss=0.1221, over 5685045.82 frames. ], batch size: 186, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:05:50,371 INFO [optim.py:369] (1/2) 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 13:05:57,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=184961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:05:59,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=184964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:06:17,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2419, 1.3786, 1.1713, 1.4315], device='cuda:1'), covar=tensor([0.0817, 0.0346, 0.0356, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0125, 0.0130, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 13:06:17,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1419, 1.3870, 1.0738, 0.8851], device='cuda:1'), covar=tensor([0.1000, 0.0821, 0.0612, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.1353, 0.1127, 0.1147, 0.1209], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 13:06:23,290 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=184993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:06:28,236 INFO [train.py:968] (1/2) Epoch 5, batch 3100, giga_loss[loss=0.2942, simple_loss=0.3661, pruned_loss=0.1112, over 28371.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3699, pruned_loss=0.1189, over 5691863.02 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3792, pruned_loss=0.1163, over 4557001.92 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.368, pruned_loss=0.1186, over 5689884.54 frames. ], batch size: 78, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:06:52,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5019, 2.1573, 1.6093, 0.6555], device='cuda:1'), covar=tensor([0.2283, 0.1164, 0.2021, 0.2577], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.1260, 0.1356, 0.1122], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 13:06:59,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4839, 1.8076, 1.7590, 1.6669], device='cuda:1'), covar=tensor([0.1322, 0.1604, 0.1033, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0742, 0.0766, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 13:07:11,466 INFO [train.py:968] (1/2) Epoch 5, batch 3150, giga_loss[loss=0.2691, simple_loss=0.3412, pruned_loss=0.09844, over 28899.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3677, pruned_loss=0.1168, over 5684719.77 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3793, pruned_loss=0.1166, over 4581569.39 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3658, pruned_loss=0.1163, over 5697271.65 frames. ], batch size: 112, lr: 6.67e-03, grad_scale: 4.0
+2023-03-02 13:07:14,973 INFO [optim.py:369] (1/2) 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 13:07:49,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9834, 1.8218, 1.3835, 1.4820], device='cuda:1'), covar=tensor([0.0625, 0.0689, 0.0998, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0451, 0.0509, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:07:55,401 INFO [train.py:968] (1/2) Epoch 5, batch 3200, giga_loss[loss=0.2989, simple_loss=0.3658, pruned_loss=0.116, over 28678.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3698, pruned_loss=0.1181, over 5693678.30 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.379, pruned_loss=0.1167, over 4613844.73 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3681, pruned_loss=0.1176, over 5699207.23 frames. ], batch size: 284, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:08:04,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-02 13:08:16,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:08:22,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2531, 2.7030, 1.4655, 1.1283], device='cuda:1'), covar=tensor([0.0871, 0.0354, 0.0757, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0466, 0.0308, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:08:32,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-02 13:08:36,101 INFO [train.py:968] (1/2) Epoch 5, batch 3250, giga_loss[loss=0.2983, simple_loss=0.3691, pruned_loss=0.1137, over 28851.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3721, pruned_loss=0.1189, over 5703196.86 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3789, pruned_loss=0.1167, over 4651882.07 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3707, pruned_loss=0.1186, over 5702334.25 frames. ], batch size: 199, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:08:40,761 INFO [optim.py:369] (1/2) 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 13:09:10,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6921, 3.1604, 1.6735, 1.4590], device='cuda:1'), covar=tensor([0.0786, 0.0313, 0.0749, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0466, 0.0308, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:09:10,829 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-02 13:09:19,707 INFO [train.py:968] (1/2) Epoch 5, batch 3300, giga_loss[loss=0.3007, simple_loss=0.371, pruned_loss=0.1152, over 28772.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3732, pruned_loss=0.1197, over 5705301.20 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3784, pruned_loss=0.1166, over 4694021.31 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3722, pruned_loss=0.1197, over 5699519.77 frames. ], batch size: 284, lr: 6.67e-03, grad_scale: 8.0
+2023-03-02 13:09:20,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2085, 1.6247, 1.2504, 1.4094], device='cuda:1'), covar=tensor([0.0770, 0.0365, 0.0328, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0126, 0.0131, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0039, 0.0035, 0.0060], device='cuda:1')
+2023-03-02 13:09:46,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-02 13:10:03,730 INFO [train.py:968] (1/2) Epoch 5, batch 3350, giga_loss[loss=0.2915, simple_loss=0.3575, pruned_loss=0.1127, over 28882.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3749, pruned_loss=0.1216, over 5703044.38 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3787, pruned_loss=0.1168, over 4723463.01 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3738, pruned_loss=0.1215, over 5698859.53 frames. ], batch size: 186, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:10:06,797 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:10:12,211 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:10:20,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:10:22,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:10:43,686 INFO [train.py:968] (1/2) Epoch 5, batch 3400, giga_loss[loss=0.2909, simple_loss=0.3576, pruned_loss=0.1121, over 28515.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3759, pruned_loss=0.1226, over 5709347.95 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3788, pruned_loss=0.1171, over 4755012.71 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3749, pruned_loss=0.1225, over 5703667.14 frames. ], batch size: 78, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:10:45,391 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:11:13,451 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:11:29,914 INFO [train.py:968] (1/2) Epoch 5, batch 3450, giga_loss[loss=0.3475, simple_loss=0.4046, pruned_loss=0.1452, over 28621.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3752, pruned_loss=0.1217, over 5718262.68 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3789, pruned_loss=0.117, over 4766398.38 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3743, pruned_loss=0.1216, over 5712210.20 frames. ], batch size: 307, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:11:35,166 INFO [optim.py:369] (1/2) 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 13:11:44,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3844, 1.4462, 1.5455, 1.4281], device='cuda:1'), covar=tensor([0.1088, 0.1288, 0.1438, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0742, 0.0642, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 13:12:11,418 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:13,153 INFO [train.py:968] (1/2) Epoch 5, batch 3500, giga_loss[loss=0.2866, simple_loss=0.3678, pruned_loss=0.1027, over 28987.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3753, pruned_loss=0.1211, over 5707571.00 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3788, pruned_loss=0.117, over 4768346.68 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3747, pruned_loss=0.1211, over 5709017.19 frames. ], batch size: 136, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:12:13,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:34,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0154, 3.3738, 2.1601, 2.0200], device='cuda:1'), covar=tensor([0.0610, 0.0240, 0.0588, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0466, 0.0306, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0023, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:12:36,966 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:39,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6192, 4.3224, 4.2824, 1.7998], device='cuda:1'), covar=tensor([0.0371, 0.0391, 0.0636, 0.2105], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0733, 0.0790, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:12:55,146 INFO [train.py:968] (1/2) Epoch 5, batch 3550, giga_loss[loss=0.2938, simple_loss=0.3676, pruned_loss=0.11, over 28456.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3745, pruned_loss=0.1192, over 5703487.44 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3792, pruned_loss=0.1172, over 4776074.65 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3737, pruned_loss=0.1191, over 5710392.55 frames. ], batch size: 78, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:12:57,416 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:12:59,217 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 3600, giga_loss[loss=0.3338, simple_loss=0.398, pruned_loss=0.1348, over 27893.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3755, pruned_loss=0.119, over 5711328.25 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3791, pruned_loss=0.1171, over 4806155.49 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3748, pruned_loss=0.119, over 5714554.07 frames. ], batch size: 412, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:13:56,854 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:14:18,719 INFO [train.py:968] (1/2) Epoch 5, batch 3650, giga_loss[loss=0.2826, simple_loss=0.35, pruned_loss=0.1076, over 28672.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.373, pruned_loss=0.1174, over 5713800.79 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3791, pruned_loss=0.1171, over 4825007.27 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3723, pruned_loss=0.1175, over 5715588.75 frames. ], batch size: 85, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:14:22,714 INFO [optim.py:369] (1/2) 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 13:15:00,795 INFO [train.py:968] (1/2) Epoch 5, batch 3700, giga_loss[loss=0.2419, simple_loss=0.3236, pruned_loss=0.08008, over 28557.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3711, pruned_loss=0.1169, over 5715282.15 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3789, pruned_loss=0.117, over 4855039.73 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3705, pruned_loss=0.117, over 5713469.12 frames. ], batch size: 60, lr: 6.66e-03, grad_scale: 8.0
+2023-03-02 13:15:28,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:15:40,751 INFO [train.py:968] (1/2) Epoch 5, batch 3750, giga_loss[loss=0.2749, simple_loss=0.346, pruned_loss=0.1019, over 28643.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.37, pruned_loss=0.1163, over 5724426.21 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3785, pruned_loss=0.1168, over 4901343.28 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3696, pruned_loss=0.1166, over 5715400.09 frames. ], batch size: 60, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:15:47,923 INFO [optim.py:369] (1/2) 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 13:16:26,011 INFO [train.py:968] (1/2) Epoch 5, batch 3800, giga_loss[loss=0.2823, simple_loss=0.3527, pruned_loss=0.106, over 28238.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3702, pruned_loss=0.1164, over 5732855.55 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3783, pruned_loss=0.1167, over 4924552.02 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3698, pruned_loss=0.1167, over 5723144.92 frames. ], batch size: 77, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:16:30,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8734, 3.1696, 2.1006, 0.8702], device='cuda:1'), covar=tensor([0.3097, 0.0932, 0.1765, 0.2934], device='cuda:1'), in_proj_covar=tensor([0.1340, 0.1241, 0.1328, 0.1103], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 13:16:31,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:17:06,193 INFO [train.py:968] (1/2) Epoch 5, batch 3850, giga_loss[loss=0.2887, simple_loss=0.3533, pruned_loss=0.1121, over 28674.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3707, pruned_loss=0.117, over 5734798.33 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3778, pruned_loss=0.1164, over 4957165.12 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3704, pruned_loss=0.1174, over 5722736.15 frames. ], batch size: 92, lr: 6.66e-03, grad_scale: 4.0
+2023-03-02 13:17:11,336 INFO [optim.py:369] (1/2) 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 13:17:12,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-02 13:17:27,901 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:17:29,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:17:46,098 INFO [train.py:968] (1/2) Epoch 5, batch 3900, giga_loss[loss=0.3709, simple_loss=0.4182, pruned_loss=0.1618, over 27568.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.372, pruned_loss=0.1172, over 5729551.80 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.378, pruned_loss=0.1164, over 4982363.02 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3715, pruned_loss=0.1176, over 5718144.28 frames. ], batch size: 472, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:17:53,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6261, 2.0343, 1.9050, 1.7850], device='cuda:1'), covar=tensor([0.1698, 0.1924, 0.1241, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0749, 0.0776, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 13:17:55,561 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:12,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:31,888 INFO [train.py:968] (1/2) Epoch 5, batch 3950, giga_loss[loss=0.2747, simple_loss=0.3498, pruned_loss=0.09978, over 28950.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3703, pruned_loss=0.1154, over 5727106.59 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.378, pruned_loss=0.1167, over 5000416.67 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3698, pruned_loss=0.1156, over 5715238.37 frames. ], batch size: 106, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:18:32,108 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:34,212 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:18:41,187 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=185861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:18:58,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1381, 1.5010, 1.1981, 0.3724], device='cuda:1'), covar=tensor([0.1273, 0.0831, 0.1121, 0.1848], device='cuda:1'), in_proj_covar=tensor([0.1357, 0.1257, 0.1339, 0.1109], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 13:19:00,960 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=185882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:19:10,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=185897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:19:12,960 INFO [train.py:968] (1/2) Epoch 5, batch 4000, giga_loss[loss=0.3082, simple_loss=0.3697, pruned_loss=0.1234, over 28400.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3693, pruned_loss=0.1154, over 5728061.79 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3777, pruned_loss=0.1166, over 5012563.21 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3691, pruned_loss=0.1156, over 5717432.96 frames. ], batch size: 65, lr: 6.65e-03, grad_scale: 8.0
+2023-03-02 13:19:54,187 INFO [train.py:968] (1/2) Epoch 5, batch 4050, giga_loss[loss=0.3081, simple_loss=0.3778, pruned_loss=0.1192, over 29009.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3673, pruned_loss=0.1147, over 5716631.95 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3777, pruned_loss=0.1165, over 5025836.97 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3669, pruned_loss=0.1148, over 5709029.02 frames. ], batch size: 136, lr: 6.65e-03, grad_scale: 8.0
+2023-03-02 13:19:59,545 INFO [optim.py:369] (1/2) 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 13:20:11,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=185971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:20:14,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=185974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:20:33,284 INFO [train.py:968] (1/2) Epoch 5, batch 4100, giga_loss[loss=0.279, simple_loss=0.3513, pruned_loss=0.1033, over 28876.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3657, pruned_loss=0.1141, over 5718260.18 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3782, pruned_loss=0.1171, over 5055993.75 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3646, pruned_loss=0.1137, over 5709539.73 frames. ], batch size: 186, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:20:37,205 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:20:37,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0856, 1.2144, 1.2737, 1.1549], device='cuda:1'), covar=tensor([0.1117, 0.1071, 0.1637, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0734, 0.0635, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 13:21:00,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8621, 3.3283, 1.8503, 1.6547], device='cuda:1'), covar=tensor([0.0788, 0.0271, 0.0741, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0464, 0.0307, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0023, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:21:07,536 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:21:10,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:21:15,672 INFO [train.py:968] (1/2) Epoch 5, batch 4150, giga_loss[loss=0.2815, simple_loss=0.3539, pruned_loss=0.1046, over 28787.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3633, pruned_loss=0.1134, over 5709136.96 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3778, pruned_loss=0.117, over 5064406.01 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3626, pruned_loss=0.1131, over 5700649.99 frames. ], batch size: 284, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:21:17,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7661, 1.6450, 1.2745, 1.4094], device='cuda:1'), covar=tensor([0.0545, 0.0536, 0.0944, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0446, 0.0507, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:21:18,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4592, 1.6619, 1.1518, 1.3850], device='cuda:1'), covar=tensor([0.0742, 0.0301, 0.0350, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0125, 0.0129, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 13:21:20,294 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-02 13:21:21,263 INFO [optim.py:369] (1/2) 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 13:21:34,692 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:21:57,769 INFO [train.py:968] (1/2) Epoch 5, batch 4200, giga_loss[loss=0.3048, simple_loss=0.3687, pruned_loss=0.1205, over 28652.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3633, pruned_loss=0.114, over 5710269.47 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3776, pruned_loss=0.1168, over 5088325.93 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3625, pruned_loss=0.1138, over 5699293.24 frames. ], batch size: 262, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:22:40,253 INFO [train.py:968] (1/2) Epoch 5, batch 4250, giga_loss[loss=0.2702, simple_loss=0.3403, pruned_loss=0.1, over 28851.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3619, pruned_loss=0.1138, over 5712991.26 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.378, pruned_loss=0.1171, over 5103595.51 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3606, pruned_loss=0.1133, over 5704339.94 frames. ], batch size: 112, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:22:41,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:22:48,505 INFO [optim.py:369] (1/2) 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 13:23:05,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4527, 1.8065, 1.7739, 1.6713], device='cuda:1'), covar=tensor([0.1350, 0.1630, 0.1068, 0.1055], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0747, 0.0771, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 13:23:20,286 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186192.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 13:23:25,262 INFO [train.py:968] (1/2) Epoch 5, batch 4300, giga_loss[loss=0.3097, simple_loss=0.3663, pruned_loss=0.1266, over 28840.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3597, pruned_loss=0.1131, over 5714431.60 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3784, pruned_loss=0.1173, over 5115669.22 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.358, pruned_loss=0.1125, over 5704735.81 frames. ], batch size: 199, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:23:33,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4376, 5.0404, 5.0773, 2.1666], device='cuda:1'), covar=tensor([0.0301, 0.0322, 0.0542, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0736, 0.0797, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:23:54,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:24:06,599 INFO [train.py:968] (1/2) Epoch 5, batch 4350, giga_loss[loss=0.292, simple_loss=0.3586, pruned_loss=0.1127, over 28653.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3577, pruned_loss=0.1125, over 5715032.39 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3784, pruned_loss=0.1173, over 5134752.68 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3559, pruned_loss=0.1119, over 5703549.86 frames. ], batch size: 307, lr: 6.65e-03, grad_scale: 4.0
+2023-03-02 13:24:12,967 INFO [optim.py:369] (1/2) 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 13:24:38,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-02 13:24:48,337 INFO [train.py:968] (1/2) Epoch 5, batch 4400, giga_loss[loss=0.2938, simple_loss=0.358, pruned_loss=0.1148, over 28568.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3542, pruned_loss=0.1106, over 5717575.35 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3786, pruned_loss=0.1175, over 5138323.91 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3525, pruned_loss=0.1099, over 5708011.45 frames. ], batch size: 336, lr: 6.65e-03, grad_scale: 8.0
+2023-03-02 13:25:26,598 INFO [train.py:968] (1/2) Epoch 5, batch 4450, giga_loss[loss=0.3017, simple_loss=0.3765, pruned_loss=0.1135, over 28650.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3545, pruned_loss=0.1102, over 5712524.05 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3788, pruned_loss=0.1176, over 5151050.50 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3523, pruned_loss=0.1094, over 5711064.70 frames. ], batch size: 307, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:25:34,297 INFO [optim.py:369] (1/2) 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 13:25:49,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3180, 1.8440, 1.5952, 1.4920], device='cuda:1'), covar=tensor([0.1428, 0.1635, 0.1165, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0748, 0.0769, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 13:25:55,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:25:57,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:26:02,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5309, 1.3931, 1.2268, 1.2032], device='cuda:1'), covar=tensor([0.0465, 0.0417, 0.0724, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0444, 0.0506, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:26:10,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6039, 3.6883, 1.6445, 1.4923], device='cuda:1'), covar=tensor([0.0776, 0.0288, 0.0792, 0.1245], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0475, 0.0311, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:26:12,936 INFO [train.py:968] (1/2) Epoch 5, batch 4500, giga_loss[loss=0.2918, simple_loss=0.3536, pruned_loss=0.115, over 28799.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3589, pruned_loss=0.1128, over 5707014.20 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.379, pruned_loss=0.1179, over 5162138.68 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3566, pruned_loss=0.1118, over 5703353.49 frames. ], batch size: 119, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:26:24,976 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:26:38,852 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-02 13:26:59,518 INFO [train.py:968] (1/2) Epoch 5, batch 4550, giga_loss[loss=0.3021, simple_loss=0.367, pruned_loss=0.1185, over 29088.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.1141, over 5699442.26 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3788, pruned_loss=0.1181, over 5167699.87 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3596, pruned_loss=0.1131, over 5700481.09 frames. ], batch size: 128, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:27:05,066 INFO [optim.py:369] (1/2) 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 13:27:23,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6474, 3.4397, 3.3371, 1.6299], device='cuda:1'), covar=tensor([0.0577, 0.0506, 0.0771, 0.2296], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0727, 0.0788, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:27:37,530 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:27:44,581 INFO [train.py:968] (1/2) Epoch 5, batch 4600, giga_loss[loss=0.3707, simple_loss=0.4165, pruned_loss=0.1625, over 27588.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.364, pruned_loss=0.1152, over 5693990.62 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3788, pruned_loss=0.1181, over 5182585.89 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3622, pruned_loss=0.1143, over 5691066.85 frames. ], batch size: 472, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:28:10,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:28:31,124 INFO [train.py:968] (1/2) Epoch 5, batch 4650, giga_loss[loss=0.3108, simple_loss=0.3772, pruned_loss=0.1222, over 28637.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3634, pruned_loss=0.1142, over 5690697.62 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3789, pruned_loss=0.1182, over 5182597.96 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3618, pruned_loss=0.1135, over 5690492.08 frames. ], batch size: 336, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:28:31,934 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:28:44,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186567.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 13:29:10,055 INFO [train.py:968] (1/2) Epoch 5, batch 4700, giga_loss[loss=0.2921, simple_loss=0.357, pruned_loss=0.1136, over 28661.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3621, pruned_loss=0.1136, over 5702305.70 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3777, pruned_loss=0.1177, over 5207060.88 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3612, pruned_loss=0.1133, over 5698118.66 frames. ], batch size: 85, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:29:44,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3291, 1.4663, 1.2611, 1.4963], device='cuda:1'), covar=tensor([0.1964, 0.1854, 0.1834, 0.1986], device='cuda:1'), in_proj_covar=tensor([0.1102, 0.0864, 0.0975, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 13:29:55,021 INFO [train.py:968] (1/2) Epoch 5, batch 4750, giga_loss[loss=0.2964, simple_loss=0.3698, pruned_loss=0.1115, over 28740.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3627, pruned_loss=0.1143, over 5701557.64 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3779, pruned_loss=0.1178, over 5214219.11 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5696123.92 frames. ], batch size: 284, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:29:59,275 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:30:03,229 INFO [optim.py:369] (1/2) 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 13:30:14,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:30:15,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7694, 1.6635, 1.6582, 1.6852], device='cuda:1'), covar=tensor([0.0925, 0.1738, 0.1536, 0.1408], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0736, 0.0634, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 13:30:16,933 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 4800, giga_loss[loss=0.3381, simple_loss=0.3892, pruned_loss=0.1435, over 27931.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3649, pruned_loss=0.116, over 5703526.80 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3782, pruned_loss=0.1181, over 5230997.23 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3636, pruned_loss=0.1154, over 5694685.37 frames. ], batch size: 412, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:30:38,813 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:30:47,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=186710.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 13:30:49,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=186713.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 13:31:18,138 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=186742.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 13:31:23,437 INFO [train.py:968] (1/2) Epoch 5, batch 4850, libri_loss[loss=0.2616, simple_loss=0.3337, pruned_loss=0.09472, over 29506.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.368, pruned_loss=0.1178, over 5706653.32 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3775, pruned_loss=0.1178, over 5254094.37 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3671, pruned_loss=0.1175, over 5693107.15 frames. ], batch size: 70, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:31:28,766 INFO [optim.py:369] (1/2) 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 13:32:05,838 INFO [train.py:968] (1/2) Epoch 5, batch 4900, giga_loss[loss=0.2727, simple_loss=0.3525, pruned_loss=0.09648, over 29041.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3695, pruned_loss=0.1178, over 5715514.51 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3778, pruned_loss=0.118, over 5263876.88 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3685, pruned_loss=0.1174, over 5702117.31 frames. ], batch size: 155, lr: 6.64e-03, grad_scale: 8.0
+2023-03-02 13:32:12,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6505, 1.7024, 1.5106, 1.1565], device='cuda:1'), covar=tensor([0.0783, 0.0710, 0.0462, 0.0713], device='cuda:1'), in_proj_covar=tensor([0.1388, 0.1155, 0.1171, 0.1230], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 13:32:18,344 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:32:41,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 13:32:50,082 INFO [train.py:968] (1/2) Epoch 5, batch 4950, giga_loss[loss=0.3065, simple_loss=0.3718, pruned_loss=0.1206, over 27994.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3714, pruned_loss=0.1188, over 5714212.07 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3782, pruned_loss=0.1182, over 5271860.04 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3703, pruned_loss=0.1184, over 5702669.96 frames. ], batch size: 412, lr: 6.64e-03, grad_scale: 4.0
+2023-03-02 13:32:56,226 INFO [optim.py:369] (1/2) 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 13:33:02,639 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:02,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:25,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5942, 1.6614, 1.4532, 1.0368], device='cuda:1'), covar=tensor([0.1321, 0.0878, 0.0753, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.1386, 0.1165, 0.1176, 0.1235], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 13:33:26,749 INFO [train.py:968] (1/2) Epoch 5, batch 5000, giga_loss[loss=0.2785, simple_loss=0.3544, pruned_loss=0.1013, over 28889.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.371, pruned_loss=0.1178, over 5720383.28 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3787, pruned_loss=0.1186, over 5292276.91 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3694, pruned_loss=0.1172, over 5708401.74 frames. ], batch size: 119, lr: 6.64e-03, grad_scale: 4.0
+2023-03-02 13:33:27,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9909, 1.1114, 1.1961, 1.0494], device='cuda:1'), covar=tensor([0.0997, 0.1112, 0.1491, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0742, 0.0639, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 13:33:33,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3911, 2.7921, 1.9476, 1.7473], device='cuda:1'), covar=tensor([0.0979, 0.0525, 0.0640, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.1384, 0.1162, 0.1173, 0.1231], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 13:33:34,468 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:49,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=186926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:33:53,649 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:34:08,810 INFO [train.py:968] (1/2) Epoch 5, batch 5050, giga_loss[loss=0.2927, simple_loss=0.3662, pruned_loss=0.1096, over 28922.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3712, pruned_loss=0.118, over 5726897.64 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3788, pruned_loss=0.1187, over 5302928.23 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3697, pruned_loss=0.1175, over 5715540.68 frames. ], batch size: 213, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:34:16,877 INFO [optim.py:369] (1/2) 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 13:34:17,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:34:30,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=186979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:34:35,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7987, 3.5475, 3.4392, 1.5388], device='cuda:1'), covar=tensor([0.0550, 0.0466, 0.0841, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0726, 0.0790, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 13:34:40,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8974, 2.4001, 2.2396, 1.9890], device='cuda:1'), covar=tensor([0.1525, 0.1569, 0.1094, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0738, 0.0769, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 13:34:48,611 INFO [train.py:968] (1/2) Epoch 5, batch 5100, giga_loss[loss=0.2575, simple_loss=0.336, pruned_loss=0.08951, over 28413.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3708, pruned_loss=0.1177, over 5725953.79 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3795, pruned_loss=0.1189, over 5315568.32 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3689, pruned_loss=0.1171, over 5716773.27 frames. ], batch size: 60, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:34:57,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3899, 3.1704, 1.4875, 1.3055], device='cuda:1'), covar=tensor([0.0869, 0.0334, 0.0888, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0482, 0.0311, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:34:58,575 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:01,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:12,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:24,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:32,395 INFO [train.py:968] (1/2) Epoch 5, batch 5150, giga_loss[loss=0.2331, simple_loss=0.3072, pruned_loss=0.07943, over 28511.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3678, pruned_loss=0.116, over 5723605.93 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3796, pruned_loss=0.1189, over 5325731.50 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.366, pruned_loss=0.1154, over 5714485.12 frames. ], batch size: 71, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:35:39,726 INFO [optim.py:369] (1/2) 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 13:35:49,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:35:51,236 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 5200, giga_loss[loss=0.3191, simple_loss=0.3783, pruned_loss=0.1299, over 28537.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3644, pruned_loss=0.1141, over 5724862.45 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3797, pruned_loss=0.119, over 5326874.84 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3629, pruned_loss=0.1136, over 5718534.08 frames. ], batch size: 336, lr: 6.63e-03, grad_scale: 8.0
+2023-03-02 13:36:16,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:36:16,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:36:50,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1850, 1.3714, 1.0781, 0.7569], device='cuda:1'), covar=tensor([0.1090, 0.0843, 0.0657, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.1368, 0.1155, 0.1164, 0.1217], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 13:36:57,076 INFO [train.py:968] (1/2) Epoch 5, batch 5250, giga_loss[loss=0.2652, simple_loss=0.3481, pruned_loss=0.0911, over 29068.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.364, pruned_loss=0.1136, over 5724429.17 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3803, pruned_loss=0.1194, over 5337909.00 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3619, pruned_loss=0.1127, over 5716508.30 frames. ], batch size: 155, lr: 6.63e-03, grad_scale: 8.0
+2023-03-02 13:37:03,817 INFO [optim.py:369] (1/2) 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 13:37:11,669 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:37:13,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:37:27,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:37:37,167 INFO [train.py:968] (1/2) Epoch 5, batch 5300, giga_loss[loss=0.3082, simple_loss=0.3855, pruned_loss=0.1155, over 28731.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3658, pruned_loss=0.1132, over 5710374.51 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3803, pruned_loss=0.1194, over 5351572.60 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3636, pruned_loss=0.1123, over 5706395.18 frames. ], batch size: 242, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:37:39,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:38:04,939 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:38:13,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:38:17,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4001, 1.7174, 1.2733, 0.7143], device='cuda:1'), covar=tensor([0.2999, 0.1885, 0.1415, 0.2909], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.1251, 0.1348, 0.1120], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 13:38:18,729 INFO [train.py:968] (1/2) Epoch 5, batch 5350, giga_loss[loss=0.3424, simple_loss=0.3994, pruned_loss=0.1428, over 28596.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3675, pruned_loss=0.1143, over 5700363.30 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3808, pruned_loss=0.1197, over 5354661.78 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.365, pruned_loss=0.1132, over 5700292.51 frames. ], batch size: 307, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:38:28,273 INFO [optim.py:369] (1/2) 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 13:38:49,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:01,265 INFO [train.py:968] (1/2) Epoch 5, batch 5400, libri_loss[loss=0.3117, simple_loss=0.3893, pruned_loss=0.117, over 29498.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.367, pruned_loss=0.1151, over 5704006.78 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3808, pruned_loss=0.1197, over 5364589.93 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3647, pruned_loss=0.1142, over 5701007.01 frames. ], batch size: 85, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:39:05,238 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:07,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:14,587 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 13:39:26,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1944, 1.4184, 1.2132, 1.4932], device='cuda:1'), covar=tensor([0.0757, 0.0371, 0.0355, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0126, 0.0130, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0039, 0.0036, 0.0059], device='cuda:1')
+2023-03-02 13:39:27,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:29,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:29,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:42,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4866, 1.9608, 1.8177, 1.7106], device='cuda:1'), covar=tensor([0.1446, 0.1672, 0.1159, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0732, 0.0768, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 13:39:42,405 INFO [train.py:968] (1/2) Epoch 5, batch 5450, giga_loss[loss=0.2863, simple_loss=0.3446, pruned_loss=0.1141, over 28445.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3659, pruned_loss=0.116, over 5703231.19 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3809, pruned_loss=0.1198, over 5374649.34 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3637, pruned_loss=0.1151, over 5697786.82 frames. ], batch size: 78, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:39:45,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:39:51,230 INFO [optim.py:369] (1/2) 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 13:39:55,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:12,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:15,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:18,399 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-02 13:40:23,842 INFO [train.py:968] (1/2) Epoch 5, batch 5500, libri_loss[loss=0.3206, simple_loss=0.3911, pruned_loss=0.125, over 29510.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3641, pruned_loss=0.1161, over 5711480.80 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3809, pruned_loss=0.1198, over 5385126.48 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3621, pruned_loss=0.1153, over 5703579.09 frames. ], batch size: 81, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:40:35,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-02 13:40:38,977 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:45,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:48,496 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:40:48,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:05,323 INFO [train.py:968] (1/2) Epoch 5, batch 5550, giga_loss[loss=0.2702, simple_loss=0.3399, pruned_loss=0.1002, over 28898.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3617, pruned_loss=0.1152, over 5711876.29 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3811, pruned_loss=0.1199, over 5398231.32 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3595, pruned_loss=0.1144, over 5701074.97 frames. ], batch size: 145, lr: 6.63e-03, grad_scale: 4.0
+2023-03-02 13:41:05,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:07,721 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:41:13,689 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:28,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:29,529 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:29,749 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 13:41:32,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:33,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:45,594 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:47,309 INFO [train.py:968] (1/2) Epoch 5, batch 5600, libri_loss[loss=0.323, simple_loss=0.4014, pruned_loss=0.1223, over 29300.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3601, pruned_loss=0.1143, over 5714628.05 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3807, pruned_loss=0.1197, over 5408642.89 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3579, pruned_loss=0.1136, over 5707040.94 frames. ], batch size: 94, lr: 6.62e-03, grad_scale: 8.0
+2023-03-02 13:41:47,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:55,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8070, 1.1598, 3.4744, 3.0579], device='cuda:1'), covar=tensor([0.1589, 0.2120, 0.0380, 0.0584], device='cuda:1'), in_proj_covar=tensor([0.0559, 0.0513, 0.0723, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 13:41:55,940 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:41:59,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2821, 1.3818, 1.4167, 1.2562], device='cuda:1'), covar=tensor([0.1023, 0.1358, 0.1692, 0.1421], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0745, 0.0640, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 13:42:11,604 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:42:29,041 INFO [train.py:968] (1/2) Epoch 5, batch 5650, giga_loss[loss=0.2543, simple_loss=0.3292, pruned_loss=0.08966, over 28895.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3576, pruned_loss=0.1129, over 5720929.62 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3809, pruned_loss=0.1199, over 5421094.75 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3552, pruned_loss=0.1121, over 5710548.05 frames. ], batch size: 213, lr: 6.62e-03, grad_scale: 8.0
+2023-03-02 13:42:36,542 INFO [optim.py:369] (1/2) 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 13:43:10,235 INFO [train.py:968] (1/2) Epoch 5, batch 5700, giga_loss[loss=0.2353, simple_loss=0.3066, pruned_loss=0.08195, over 28654.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3522, pruned_loss=0.1101, over 5723215.16 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3812, pruned_loss=0.1202, over 5428405.32 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3496, pruned_loss=0.1091, over 5712536.58 frames. ], batch size: 242, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:43:17,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:43:27,130 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:43:28,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2885, 2.3830, 1.3233, 1.3696], device='cuda:1'), covar=tensor([0.0829, 0.0393, 0.0812, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0484, 0.0308, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:43:29,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:43:38,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-02 13:43:50,892 INFO [train.py:968] (1/2) Epoch 5, batch 5750, giga_loss[loss=0.2703, simple_loss=0.3436, pruned_loss=0.09855, over 28873.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3503, pruned_loss=0.1092, over 5720478.78 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3811, pruned_loss=0.1202, over 5436435.22 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3477, pruned_loss=0.1081, over 5711479.05 frames. ], batch size: 112, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:43:52,399 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:44:00,506 INFO [optim.py:369] (1/2) 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 13:44:14,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:44:32,119 INFO [train.py:968] (1/2) Epoch 5, batch 5800, giga_loss[loss=0.3455, simple_loss=0.4158, pruned_loss=0.1376, over 28831.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3526, pruned_loss=0.11, over 5728573.03 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3805, pruned_loss=0.1198, over 5447977.44 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3503, pruned_loss=0.1092, over 5717536.71 frames. ], batch size: 243, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:45:13,976 INFO [train.py:968] (1/2) Epoch 5, batch 5850, libri_loss[loss=0.3787, simple_loss=0.4333, pruned_loss=0.162, over 29529.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3574, pruned_loss=0.1125, over 5728977.94 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3809, pruned_loss=0.1201, over 5454818.41 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3549, pruned_loss=0.1115, over 5717753.84 frames. ], batch size: 81, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:45:15,004 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:23,633 INFO [optim.py:369] (1/2) 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 13:45:27,912 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:42,548 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:45:55,283 INFO [train.py:968] (1/2) Epoch 5, batch 5900, libri_loss[loss=0.2738, simple_loss=0.3352, pruned_loss=0.1062, over 29652.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3613, pruned_loss=0.1143, over 5719304.66 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3809, pruned_loss=0.1201, over 5458653.04 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3587, pruned_loss=0.1133, over 5715300.12 frames. ], batch size: 69, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:45:59,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3535, 3.1911, 1.5110, 1.3809], device='cuda:1'), covar=tensor([0.0881, 0.0354, 0.0826, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0483, 0.0311, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:46:02,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=187806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:08,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4710, 1.6106, 1.3246, 1.0692], device='cuda:1'), covar=tensor([0.1024, 0.0797, 0.0651, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.1380, 0.1152, 0.1158, 0.1206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 13:46:16,256 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:20,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:31,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0987, 1.0942, 0.8845, 1.2988], device='cuda:1'), covar=tensor([0.0766, 0.0377, 0.0367, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0127, 0.0131, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0040, 0.0036, 0.0060], device='cuda:1')
+2023-03-02 13:46:42,895 INFO [train.py:968] (1/2) Epoch 5, batch 5950, giga_loss[loss=0.2943, simple_loss=0.3629, pruned_loss=0.1128, over 28225.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3643, pruned_loss=0.1157, over 5717317.83 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.381, pruned_loss=0.1201, over 5463155.13 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5712292.96 frames. ], batch size: 77, lr: 6.62e-03, grad_scale: 2.0
+2023-03-02 13:46:46,210 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:46:52,213 INFO [optim.py:369] (1/2) 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 13:47:26,056 INFO [train.py:968] (1/2) Epoch 5, batch 6000, libri_loss[loss=0.2726, simple_loss=0.353, pruned_loss=0.09613, over 29535.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3664, pruned_loss=0.1166, over 5707152.02 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3806, pruned_loss=0.1199, over 5462946.95 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3645, pruned_loss=0.116, over 5709304.46 frames. ], batch size: 81, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:47:26,056 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 13:47:35,025 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 13:47:39,984 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:47:42,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2954, 1.6064, 1.3687, 1.5314], device='cuda:1'), covar=tensor([0.0642, 0.0264, 0.0292, 0.0683], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0126, 0.0130, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0039, 0.0035, 0.0059], device='cuda:1')
+2023-03-02 13:47:46,805 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:47:49,427 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=187916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:08,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:12,169 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:21,267 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=187949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:21,569 INFO [train.py:968] (1/2) Epoch 5, batch 6050, giga_loss[loss=0.3157, simple_loss=0.3815, pruned_loss=0.1249, over 28888.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3739, pruned_loss=0.1237, over 5702264.35 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3805, pruned_loss=0.1198, over 5470346.56 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3723, pruned_loss=0.1234, over 5701811.98 frames. ], batch size: 174, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:48:24,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=187952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:32,263 INFO [optim.py:369] (1/2) 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 13:48:37,804 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=187981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:48:51,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3726, 1.9850, 1.4402, 0.5869], device='cuda:1'), covar=tensor([0.2422, 0.1301, 0.1825, 0.2540], device='cuda:1'), in_proj_covar=tensor([0.1370, 0.1259, 0.1353, 0.1135], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 13:49:08,193 INFO [train.py:968] (1/2) Epoch 5, batch 6100, giga_loss[loss=0.4044, simple_loss=0.4409, pruned_loss=0.1839, over 28652.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3792, pruned_loss=0.1281, over 5690344.94 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3809, pruned_loss=0.12, over 5473028.98 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3775, pruned_loss=0.1277, over 5693839.18 frames. ], batch size: 262, lr: 6.62e-03, grad_scale: 4.0
+2023-03-02 13:49:57,051 INFO [train.py:968] (1/2) Epoch 5, batch 6150, giga_loss[loss=0.3359, simple_loss=0.398, pruned_loss=0.1369, over 28507.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3862, pruned_loss=0.1337, over 5680173.40 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3803, pruned_loss=0.1197, over 5488680.26 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3853, pruned_loss=0.1341, over 5676224.56 frames. ], batch size: 336, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:50:07,708 INFO [optim.py:369] (1/2) 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 13:50:47,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 13:50:48,046 INFO [train.py:968] (1/2) Epoch 5, batch 6200, giga_loss[loss=0.4174, simple_loss=0.4431, pruned_loss=0.1959, over 27498.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3936, pruned_loss=0.1401, over 5678618.03 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3802, pruned_loss=0.1196, over 5496768.32 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3932, pruned_loss=0.1409, over 5671502.61 frames. ], batch size: 472, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:51:07,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-02 13:51:24,757 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 6250, giga_loss[loss=0.4246, simple_loss=0.4631, pruned_loss=0.1931, over 28953.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3996, pruned_loss=0.1455, over 5682224.63 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3802, pruned_loss=0.1195, over 5504658.59 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.3998, pruned_loss=0.1469, over 5674309.25 frames. ], batch size: 227, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:51:44,489 INFO [optim.py:369] (1/2) 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 13:52:22,574 INFO [train.py:968] (1/2) Epoch 5, batch 6300, giga_loss[loss=0.3664, simple_loss=0.4166, pruned_loss=0.1581, over 28943.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4051, pruned_loss=0.1505, over 5676396.48 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3797, pruned_loss=0.1194, over 5512343.84 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.4059, pruned_loss=0.1522, over 5666593.36 frames. ], batch size: 164, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:53:16,130 INFO [train.py:968] (1/2) Epoch 5, batch 6350, giga_loss[loss=0.3199, simple_loss=0.3855, pruned_loss=0.1272, over 28967.00 frames. ], tot_loss[loss=0.3573, simple_loss=0.4077, pruned_loss=0.1535, over 5652420.73 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3802, pruned_loss=0.1198, over 5512894.56 frames. ], giga_tot_loss[loss=0.3596, simple_loss=0.4086, pruned_loss=0.1553, over 5648379.11 frames. ], batch size: 106, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:53:27,399 INFO [optim.py:369] (1/2) 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 13:53:45,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:49,254 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:52,582 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:53,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:53:57,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=188291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:54:05,921 INFO [train.py:968] (1/2) Epoch 5, batch 6400, libri_loss[loss=0.2736, simple_loss=0.3462, pruned_loss=0.1005, over 29550.00 frames. ], tot_loss[loss=0.3611, simple_loss=0.4095, pruned_loss=0.1564, over 5641336.44 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3802, pruned_loss=0.1198, over 5527411.91 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4114, pruned_loss=0.1592, over 5629568.62 frames. ], batch size: 79, lr: 6.61e-03, grad_scale: 8.0
+2023-03-02 13:54:20,835 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 6450, giga_loss[loss=0.3699, simple_loss=0.4169, pruned_loss=0.1614, over 28978.00 frames. ], tot_loss[loss=0.3682, simple_loss=0.4139, pruned_loss=0.1613, over 5625399.61 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3799, pruned_loss=0.1197, over 5532761.29 frames. ], giga_tot_loss[loss=0.3726, simple_loss=0.4163, pruned_loss=0.1644, over 5613851.32 frames. ], batch size: 213, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:55:14,546 INFO [optim.py:369] (1/2) 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 13:55:55,388 INFO [train.py:968] (1/2) Epoch 5, batch 6500, giga_loss[loss=0.4372, simple_loss=0.4565, pruned_loss=0.209, over 27588.00 frames. ], tot_loss[loss=0.3729, simple_loss=0.4175, pruned_loss=0.1642, over 5610385.73 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.38, pruned_loss=0.1198, over 5526837.19 frames. ], giga_tot_loss[loss=0.3768, simple_loss=0.4196, pruned_loss=0.167, over 5608025.15 frames. ], batch size: 472, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:56:20,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:20,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6034, 3.2990, 1.6343, 1.4356], device='cuda:1'), covar=tensor([0.0830, 0.0279, 0.0767, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0480, 0.0309, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 13:56:22,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:26,331 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:28,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:30,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=188434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:33,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=188437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:46,096 INFO [train.py:968] (1/2) Epoch 5, batch 6550, giga_loss[loss=0.407, simple_loss=0.4427, pruned_loss=0.1856, over 28224.00 frames. ], tot_loss[loss=0.373, simple_loss=0.4174, pruned_loss=0.1643, over 5621178.12 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3801, pruned_loss=0.1198, over 5530859.60 frames. ], giga_tot_loss[loss=0.377, simple_loss=0.4196, pruned_loss=0.1672, over 5617361.00 frames. ], batch size: 368, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:56:51,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:56:59,340 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:57:02,274 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=188466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 13:57:12,633 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 13:57:36,281 INFO [train.py:968] (1/2) Epoch 5, batch 6600, libri_loss[loss=0.3383, simple_loss=0.4074, pruned_loss=0.1346, over 27753.00 frames. ], tot_loss[loss=0.3716, simple_loss=0.4158, pruned_loss=0.1637, over 5626441.78 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3807, pruned_loss=0.1203, over 5536157.79 frames. ], giga_tot_loss[loss=0.3759, simple_loss=0.418, pruned_loss=0.1669, over 5621567.58 frames. ], batch size: 116, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:58:26,003 INFO [train.py:968] (1/2) Epoch 5, batch 6650, giga_loss[loss=0.3366, simple_loss=0.397, pruned_loss=0.1381, over 28970.00 frames. ], tot_loss[loss=0.3679, simple_loss=0.4134, pruned_loss=0.1612, over 5632221.03 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3806, pruned_loss=0.1202, over 5546442.16 frames. ], giga_tot_loss[loss=0.373, simple_loss=0.4161, pruned_loss=0.1649, over 5621694.20 frames. ], batch size: 186, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:58:40,076 INFO [optim.py:369] (1/2) 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 13:59:19,527 INFO [train.py:968] (1/2) Epoch 5, batch 6700, giga_loss[loss=0.4043, simple_loss=0.4421, pruned_loss=0.1832, over 28684.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4143, pruned_loss=0.161, over 5639248.57 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3807, pruned_loss=0.1203, over 5547919.95 frames. ], giga_tot_loss[loss=0.3723, simple_loss=0.4166, pruned_loss=0.164, over 5630071.48 frames. ], batch size: 242, lr: 6.61e-03, grad_scale: 4.0
+2023-03-02 13:59:24,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 14:00:11,634 INFO [train.py:968] (1/2) Epoch 5, batch 6750, giga_loss[loss=0.3599, simple_loss=0.416, pruned_loss=0.1519, over 28734.00 frames. ], tot_loss[loss=0.3697, simple_loss=0.4154, pruned_loss=0.162, over 5618231.26 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3812, pruned_loss=0.1208, over 5552972.33 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.4173, pruned_loss=0.1646, over 5607795.93 frames. ], batch size: 262, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:00:22,589 INFO [optim.py:369] (1/2) 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 14:00:36,153 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:00:54,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-02 14:01:00,386 INFO [train.py:968] (1/2) Epoch 5, batch 6800, libri_loss[loss=0.2568, simple_loss=0.3354, pruned_loss=0.08911, over 29373.00 frames. ], tot_loss[loss=0.3658, simple_loss=0.4128, pruned_loss=0.1594, over 5626084.44 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3813, pruned_loss=0.121, over 5566488.41 frames. ], giga_tot_loss[loss=0.3706, simple_loss=0.4155, pruned_loss=0.1629, over 5607731.57 frames. ], batch size: 71, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:01:45,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1421, 2.0505, 1.4223, 1.4285], device='cuda:1'), covar=tensor([0.0894, 0.0285, 0.0332, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0127, 0.0131, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0060], device='cuda:1')
+2023-03-02 14:01:48,506 INFO [train.py:968] (1/2) Epoch 5, batch 6850, giga_loss[loss=0.3384, simple_loss=0.3993, pruned_loss=0.1388, over 28860.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4084, pruned_loss=0.1542, over 5631755.26 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3807, pruned_loss=0.1206, over 5573416.15 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.4119, pruned_loss=0.1583, over 5612479.79 frames. ], batch size: 174, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:02:02,876 INFO [optim.py:369] (1/2) 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 14:02:37,708 INFO [train.py:968] (1/2) Epoch 5, batch 6900, giga_loss[loss=0.2773, simple_loss=0.3499, pruned_loss=0.1023, over 28688.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4048, pruned_loss=0.1502, over 5644484.06 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3805, pruned_loss=0.1205, over 5578546.72 frames. ], giga_tot_loss[loss=0.3579, simple_loss=0.408, pruned_loss=0.1539, over 5625538.74 frames. ], batch size: 78, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:02:39,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9683, 1.0679, 0.9071, 0.5821], device='cuda:1'), covar=tensor([0.0729, 0.0831, 0.0500, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1188, 0.1169, 0.1242], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 14:02:41,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2328, 3.0286, 1.6018, 1.5084], device='cuda:1'), covar=tensor([0.1229, 0.0622, 0.0822, 0.1032], device='cuda:1'), in_proj_covar=tensor([0.1411, 0.1188, 0.1170, 0.1243], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 14:03:00,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9162, 3.1572, 2.2144, 0.8091], device='cuda:1'), covar=tensor([0.3595, 0.1214, 0.1812, 0.3606], device='cuda:1'), in_proj_covar=tensor([0.1387, 0.1265, 0.1358, 0.1134], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 14:03:29,143 INFO [train.py:968] (1/2) Epoch 5, batch 6950, giga_loss[loss=0.377, simple_loss=0.4223, pruned_loss=0.1659, over 27936.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4025, pruned_loss=0.1482, over 5644409.69 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3805, pruned_loss=0.1205, over 5580189.99 frames. ], giga_tot_loss[loss=0.3539, simple_loss=0.4053, pruned_loss=0.1513, over 5628441.85 frames. ], batch size: 412, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:03:41,089 INFO [optim.py:369] (1/2) 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 14:03:54,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=188874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:04:06,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-02 14:04:18,886 INFO [train.py:968] (1/2) Epoch 5, batch 7000, giga_loss[loss=0.2996, simple_loss=0.3698, pruned_loss=0.1147, over 29033.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.4, pruned_loss=0.1464, over 5651788.60 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3809, pruned_loss=0.1209, over 5581699.13 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4021, pruned_loss=0.1488, over 5638750.07 frames. ], batch size: 128, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:04:53,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3108, 1.3441, 1.3199, 1.4659], device='cuda:1'), covar=tensor([0.0753, 0.0319, 0.0306, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0127, 0.0131, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0060], device='cuda:1')
+2023-03-02 14:04:59,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3178, 1.4451, 1.2272, 1.6699], device='cuda:1'), covar=tensor([0.2142, 0.2088, 0.2014, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.1121, 0.0880, 0.0990, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 14:05:07,259 INFO [train.py:968] (1/2) Epoch 5, batch 7050, giga_loss[loss=0.3563, simple_loss=0.4119, pruned_loss=0.1503, over 28969.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3997, pruned_loss=0.1461, over 5659045.92 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3803, pruned_loss=0.1207, over 5587330.10 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4024, pruned_loss=0.149, over 5645602.57 frames. ], batch size: 227, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:05:22,274 INFO [optim.py:369] (1/2) 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 14:06:03,935 INFO [train.py:968] (1/2) Epoch 5, batch 7100, giga_loss[loss=0.3816, simple_loss=0.4176, pruned_loss=0.1728, over 28011.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3993, pruned_loss=0.1453, over 5664827.93 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.38, pruned_loss=0.1205, over 5590604.47 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4018, pruned_loss=0.1479, over 5651990.62 frames. ], batch size: 412, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:06:21,576 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4717, 3.4499, 1.5652, 1.4189], device='cuda:1'), covar=tensor([0.0843, 0.0264, 0.0814, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0477, 0.0308, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 14:06:59,017 INFO [train.py:968] (1/2) Epoch 5, batch 7150, libri_loss[loss=0.3416, simple_loss=0.404, pruned_loss=0.1397, over 29525.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3965, pruned_loss=0.142, over 5672194.21 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3801, pruned_loss=0.1206, over 5595534.61 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3987, pruned_loss=0.1443, over 5658409.65 frames. ], batch size: 84, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:07:00,195 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:07:38,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1405, 1.9308, 1.9110, 1.6638], device='cuda:1'), covar=tensor([0.0970, 0.1673, 0.1288, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0743, 0.0641, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 14:07:53,257 INFO [train.py:968] (1/2) Epoch 5, batch 7200, giga_loss[loss=0.3109, simple_loss=0.3863, pruned_loss=0.1177, over 29024.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3968, pruned_loss=0.1402, over 5670576.47 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3793, pruned_loss=0.1203, over 5597418.01 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3999, pruned_loss=0.143, over 5659740.85 frames. ], batch size: 136, lr: 6.60e-03, grad_scale: 8.0
+2023-03-02 14:08:37,238 INFO [train.py:968] (1/2) Epoch 5, batch 7250, giga_loss[loss=0.362, simple_loss=0.4199, pruned_loss=0.1521, over 29071.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3989, pruned_loss=0.1412, over 5670621.61 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3793, pruned_loss=0.1206, over 5608302.88 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.4023, pruned_loss=0.1442, over 5655252.41 frames. ], batch size: 128, lr: 6.60e-03, grad_scale: 4.0
+2023-03-02 14:08:54,485 INFO [optim.py:369] (1/2) 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 14:08:56,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4563, 1.4824, 1.2596, 1.3719], device='cuda:1'), covar=tensor([0.1206, 0.1669, 0.1664, 0.1537], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0742, 0.0637, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 14:09:08,075 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7028, 2.2691, 1.9561, 2.0839], device='cuda:1'), covar=tensor([0.0742, 0.0246, 0.0282, 0.0738], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0127, 0.0131, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0060], device='cuda:1')
+2023-03-02 14:09:24,686 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:09:26,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:09:28,474 INFO [train.py:968] (1/2) Epoch 5, batch 7300, giga_loss[loss=0.3971, simple_loss=0.43, pruned_loss=0.1821, over 27976.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.4, pruned_loss=0.1427, over 5678211.80 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3794, pruned_loss=0.1208, over 5614414.58 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4031, pruned_loss=0.1454, over 5661770.53 frames. ], batch size: 412, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:09:45,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-02 14:09:50,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1401, 1.1703, 3.6501, 3.1573], device='cuda:1'), covar=tensor([0.1531, 0.2134, 0.0429, 0.0671], device='cuda:1'), in_proj_covar=tensor([0.0569, 0.0526, 0.0745, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 14:09:56,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:10:00,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5264, 1.4918, 1.5328, 1.4463], device='cuda:1'), covar=tensor([0.1129, 0.1679, 0.1610, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0745, 0.0635, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 14:10:02,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3482, 3.1815, 1.4073, 1.3821], device='cuda:1'), covar=tensor([0.0890, 0.0329, 0.0835, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0479, 0.0310, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 14:10:15,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:10:15,644 INFO [train.py:968] (1/2) Epoch 5, batch 7350, giga_loss[loss=0.2969, simple_loss=0.3639, pruned_loss=0.1149, over 28742.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3975, pruned_loss=0.1407, over 5680795.93 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3793, pruned_loss=0.1206, over 5621503.99 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.4006, pruned_loss=0.1436, over 5662653.04 frames. ], batch size: 119, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:10:27,577 INFO [optim.py:369] (1/2) 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 14:11:03,662 INFO [train.py:968] (1/2) Epoch 5, batch 7400, giga_loss[loss=0.2798, simple_loss=0.3443, pruned_loss=0.1076, over 28760.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3956, pruned_loss=0.1411, over 5675864.64 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.379, pruned_loss=0.1206, over 5628448.20 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3988, pruned_loss=0.144, over 5656577.87 frames. ], batch size: 99, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:11:24,755 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 5, batch 7450, giga_loss[loss=0.3346, simple_loss=0.4001, pruned_loss=0.1345, over 28971.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3948, pruned_loss=0.1408, over 5683512.47 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3793, pruned_loss=0.1207, over 5634167.47 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3976, pruned_loss=0.1437, over 5664237.13 frames. ], batch size: 164, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:12:02,032 INFO [optim.py:369] (1/2) 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 14:12:28,219 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:12:30,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:12:34,862 INFO [train.py:968] (1/2) Epoch 5, batch 7500, giga_loss[loss=0.3262, simple_loss=0.395, pruned_loss=0.1287, over 28945.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3937, pruned_loss=0.1393, over 5695090.62 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.379, pruned_loss=0.1204, over 5643740.87 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3969, pruned_loss=0.1427, over 5672647.23 frames. ], batch size: 213, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:12:58,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:12:59,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:13:22,294 INFO [train.py:968] (1/2) Epoch 5, batch 7550, giga_loss[loss=0.3055, simple_loss=0.376, pruned_loss=0.1175, over 28812.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3934, pruned_loss=0.1376, over 5705106.78 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3788, pruned_loss=0.1204, over 5649422.32 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3964, pruned_loss=0.1408, over 5683200.60 frames. ], batch size: 186, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:13:28,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:13:33,859 INFO [optim.py:369] (1/2) 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 14:13:39,057 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:14:05,255 INFO [train.py:968] (1/2) Epoch 5, batch 7600, giga_loss[loss=0.2736, simple_loss=0.3563, pruned_loss=0.0954, over 29111.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3933, pruned_loss=0.1371, over 5709790.49 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3783, pruned_loss=0.1199, over 5659329.42 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3969, pruned_loss=0.1409, over 5685991.97 frames. ], batch size: 155, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:14:29,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7158, 1.5296, 1.2000, 1.3558], device='cuda:1'), covar=tensor([0.0623, 0.0638, 0.0834, 0.1048], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0458, 0.0508, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 14:14:46,466 INFO [train.py:968] (1/2) Epoch 5, batch 7650, giga_loss[loss=0.3844, simple_loss=0.4304, pruned_loss=0.1692, over 28702.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3925, pruned_loss=0.1371, over 5710940.91 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.378, pruned_loss=0.1198, over 5664713.67 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3959, pruned_loss=0.1407, over 5688043.15 frames. ], batch size: 262, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:15:03,136 INFO [optim.py:369] (1/2) 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 14:15:34,555 INFO [train.py:968] (1/2) Epoch 5, batch 7700, giga_loss[loss=0.3394, simple_loss=0.3966, pruned_loss=0.1411, over 28312.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3909, pruned_loss=0.1371, over 5702659.84 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3777, pruned_loss=0.1194, over 5666520.24 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3943, pruned_loss=0.1406, over 5683753.14 frames. ], batch size: 368, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:16:15,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:16:22,326 INFO [train.py:968] (1/2) Epoch 5, batch 7750, giga_loss[loss=0.4222, simple_loss=0.4373, pruned_loss=0.2035, over 26727.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3917, pruned_loss=0.1383, over 5693018.68 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3776, pruned_loss=0.1192, over 5666707.55 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.395, pruned_loss=0.142, over 5678626.24 frames. ], batch size: 555, lr: 6.59e-03, grad_scale: 8.0
+2023-03-02 14:16:34,910 INFO [optim.py:369] (1/2) 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 14:17:03,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:17:05,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:17:06,160 INFO [train.py:968] (1/2) Epoch 5, batch 7800, giga_loss[loss=0.3574, simple_loss=0.4096, pruned_loss=0.1526, over 28965.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3911, pruned_loss=0.1381, over 5697603.39 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3773, pruned_loss=0.1189, over 5667680.57 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3946, pruned_loss=0.1422, over 5686215.62 frames. ], batch size: 213, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:17:48,593 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=189740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:17:56,259 INFO [train.py:968] (1/2) Epoch 5, batch 7850, giga_loss[loss=0.3424, simple_loss=0.3918, pruned_loss=0.1465, over 28599.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.389, pruned_loss=0.137, over 5701501.89 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3774, pruned_loss=0.1189, over 5671939.73 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3919, pruned_loss=0.1406, over 5689159.20 frames. ], batch size: 307, lr: 6.59e-03, grad_scale: 4.0
+2023-03-02 14:18:14,188 INFO [optim.py:369] (1/2) 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 14:18:44,353 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:18:45,526 INFO [train.py:968] (1/2) Epoch 5, batch 7900, giga_loss[loss=0.3518, simple_loss=0.3963, pruned_loss=0.1536, over 27540.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3887, pruned_loss=0.1378, over 5702814.78 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3775, pruned_loss=0.1189, over 5673346.03 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.391, pruned_loss=0.1407, over 5692076.31 frames. ], batch size: 472, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:19:14,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:22,560 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:24,086 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:24,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:27,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:28,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=189847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:30,317 INFO [train.py:968] (1/2) Epoch 5, batch 7950, giga_loss[loss=0.371, simple_loss=0.3951, pruned_loss=0.1734, over 24068.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3892, pruned_loss=0.1386, over 5687749.65 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3775, pruned_loss=0.1188, over 5666462.66 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3913, pruned_loss=0.1414, over 5686867.26 frames. ], batch size: 705, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:19:47,845 INFO [optim.py:369] (1/2) 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 14:19:54,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:19:56,051 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:20:14,165 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 8000, giga_loss[loss=0.3523, simple_loss=0.412, pruned_loss=0.1463, over 28904.00 frames. ], tot_loss[loss=0.335, simple_loss=0.391, pruned_loss=0.1394, over 5679200.43 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3779, pruned_loss=0.1192, over 5662940.55 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3926, pruned_loss=0.1419, over 5682104.94 frames. ], batch size: 99, lr: 6.58e-03, grad_scale: 8.0
+2023-03-02 14:20:53,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4924, 1.5464, 1.3264, 1.6658], device='cuda:1'), covar=tensor([0.1733, 0.1530, 0.1390, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.0869, 0.0987, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 14:20:57,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:20:59,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:20:59,903 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-02 14:21:06,506 INFO [train.py:968] (1/2) Epoch 5, batch 8050, giga_loss[loss=0.3299, simple_loss=0.3901, pruned_loss=0.1348, over 28744.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3912, pruned_loss=0.139, over 5673059.45 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3777, pruned_loss=0.119, over 5666583.75 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3927, pruned_loss=0.1413, over 5672366.32 frames. ], batch size: 284, lr: 6.58e-03, grad_scale: 8.0
+2023-03-02 14:21:18,707 INFO [optim.py:369] (1/2) 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 14:21:27,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=189973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:28,811 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:30,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:30,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5062, 1.6810, 1.3283, 1.0022], device='cuda:1'), covar=tensor([0.1303, 0.0961, 0.0762, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1171, 0.1174, 0.1239], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 14:21:42,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=189990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:44,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=189993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:21:51,592 INFO [train.py:968] (1/2) Epoch 5, batch 8100, giga_loss[loss=0.3386, simple_loss=0.3712, pruned_loss=0.153, over 23695.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3919, pruned_loss=0.1391, over 5662978.84 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3779, pruned_loss=0.1192, over 5665297.82 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3934, pruned_loss=0.1415, over 5663613.37 frames. ], batch size: 705, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:21:57,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:07,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:09,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 14:22:14,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:25,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:22:26,115 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-02 14:22:42,097 INFO [train.py:968] (1/2) Epoch 5, batch 8150, giga_loss[loss=0.4439, simple_loss=0.4693, pruned_loss=0.2092, over 27905.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3924, pruned_loss=0.1395, over 5676943.02 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.378, pruned_loss=0.1192, over 5669235.14 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3939, pruned_loss=0.1419, over 5674146.07 frames. ], batch size: 412, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:22:52,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3396, 1.7921, 1.4658, 1.3384], device='cuda:1'), covar=tensor([0.0767, 0.0291, 0.0311, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0127, 0.0130, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0054, 0.0040, 0.0036, 0.0060], device='cuda:1')
+2023-03-02 14:23:01,723 INFO [optim.py:369] (1/2) 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 14:23:30,645 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:23:32,423 INFO [train.py:968] (1/2) Epoch 5, batch 8200, libri_loss[loss=0.2766, simple_loss=0.3555, pruned_loss=0.0988, over 29669.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.394, pruned_loss=0.1415, over 5677579.41 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3773, pruned_loss=0.1188, over 5675092.78 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3962, pruned_loss=0.1443, over 5670154.95 frames. ], batch size: 73, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:23:49,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:23:55,126 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:24:22,324 INFO [train.py:968] (1/2) Epoch 5, batch 8250, giga_loss[loss=0.356, simple_loss=0.4063, pruned_loss=0.1528, over 28848.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3942, pruned_loss=0.1425, over 5689230.89 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3767, pruned_loss=0.1186, over 5681352.13 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3969, pruned_loss=0.1456, over 5677744.64 frames. ], batch size: 199, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:24:32,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:24:34,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:24:40,789 INFO [optim.py:369] (1/2) 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 14:25:06,675 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:25:15,525 INFO [train.py:968] (1/2) Epoch 5, batch 8300, giga_loss[loss=0.3799, simple_loss=0.4233, pruned_loss=0.1683, over 28235.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.396, pruned_loss=0.1453, over 5677820.95 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3765, pruned_loss=0.1184, over 5684904.01 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.3986, pruned_loss=0.1482, over 5665750.42 frames. ], batch size: 368, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:26:08,034 INFO [train.py:968] (1/2) Epoch 5, batch 8350, giga_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1293, over 28765.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.398, pruned_loss=0.1475, over 5669691.07 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.377, pruned_loss=0.1186, over 5688309.67 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.3999, pruned_loss=0.15, over 5657119.15 frames. ], batch size: 284, lr: 6.58e-03, grad_scale: 4.0
+2023-03-02 14:26:11,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3195, 1.7703, 1.3132, 0.5241], device='cuda:1'), covar=tensor([0.1495, 0.1033, 0.1690, 0.2143], device='cuda:1'), in_proj_covar=tensor([0.1378, 0.1290, 0.1387, 0.1142], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 14:26:16,816 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:18,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:22,435 INFO [optim.py:369] (1/2) 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 14:26:26,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:45,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 14:26:47,231 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:26:55,025 INFO [train.py:968] (1/2) Epoch 5, batch 8400, giga_loss[loss=0.3153, simple_loss=0.3756, pruned_loss=0.1274, over 28571.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.397, pruned_loss=0.1471, over 5671961.10 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.377, pruned_loss=0.1186, over 5690500.16 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.3988, pruned_loss=0.1494, over 5659994.91 frames. ], batch size: 71, lr: 6.58e-03, grad_scale: 8.0
+2023-03-02 14:27:40,000 INFO [train.py:968] (1/2) Epoch 5, batch 8450, giga_loss[loss=0.2871, simple_loss=0.3575, pruned_loss=0.1083, over 28629.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3959, pruned_loss=0.1444, over 5679330.46 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3767, pruned_loss=0.1182, over 5694038.22 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.398, pruned_loss=0.1472, over 5666403.35 frames. ], batch size: 92, lr: 6.57e-03, grad_scale: 8.0
+2023-03-02 14:27:55,255 INFO [optim.py:369] (1/2) 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 14:28:23,984 INFO [train.py:968] (1/2) Epoch 5, batch 8500, giga_loss[loss=0.3098, simple_loss=0.374, pruned_loss=0.1227, over 28980.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3926, pruned_loss=0.1415, over 5677965.33 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3771, pruned_loss=0.1187, over 5700129.93 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3944, pruned_loss=0.144, over 5661000.53 frames. ], batch size: 164, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:28:30,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:28:34,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:28:37,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:07,749 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:12,856 INFO [train.py:968] (1/2) Epoch 5, batch 8550, giga_loss[loss=0.308, simple_loss=0.3689, pruned_loss=0.1236, over 28765.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3901, pruned_loss=0.1399, over 5685262.35 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3772, pruned_loss=0.1188, over 5704280.23 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3917, pruned_loss=0.1422, over 5667667.63 frames. ], batch size: 242, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:29:28,424 INFO [optim.py:369] (1/2) 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 14:29:33,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:47,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2604, 1.4631, 1.1983, 1.3356], device='cuda:1'), covar=tensor([0.2086, 0.2020, 0.2011, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1120, 0.0875, 0.0990, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 14:29:54,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:29:55,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6120, 4.4193, 4.2662, 1.6586], device='cuda:1'), covar=tensor([0.0425, 0.0440, 0.0818, 0.2160], device='cuda:1'), in_proj_covar=tensor([0.0858, 0.0766, 0.0827, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 14:29:57,636 INFO [train.py:968] (1/2) Epoch 5, batch 8600, giga_loss[loss=0.349, simple_loss=0.4096, pruned_loss=0.1442, over 28921.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3898, pruned_loss=0.1404, over 5679097.38 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3782, pruned_loss=0.1196, over 5698557.48 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3903, pruned_loss=0.1419, over 5669062.64 frames. ], batch size: 164, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:30:31,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190532.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:30:44,709 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:30:51,583 INFO [train.py:968] (1/2) Epoch 5, batch 8650, giga_loss[loss=0.2917, simple_loss=0.3626, pruned_loss=0.1104, over 29069.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3904, pruned_loss=0.1416, over 5655805.01 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3779, pruned_loss=0.1193, over 5699207.48 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3914, pruned_loss=0.1436, over 5646324.81 frames. ], batch size: 136, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:30:52,786 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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 14:31:24,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:31:42,340 INFO [train.py:968] (1/2) Epoch 5, batch 8700, giga_loss[loss=0.3877, simple_loss=0.4357, pruned_loss=0.1699, over 27945.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3941, pruned_loss=0.1424, over 5664123.68 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3778, pruned_loss=0.1192, over 5700515.17 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3952, pruned_loss=0.1444, over 5655156.83 frames. ], batch size: 412, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:31:57,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:02,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:20,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=190639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:22,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=190642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:27,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:32:31,103 INFO [train.py:968] (1/2) Epoch 5, batch 8750, giga_loss[loss=0.3265, simple_loss=0.3999, pruned_loss=0.1266, over 28445.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3965, pruned_loss=0.1417, over 5668498.46 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3778, pruned_loss=0.1194, over 5705648.78 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3979, pruned_loss=0.1436, over 5655663.48 frames. ], batch size: 78, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:32:48,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5737, 1.6017, 1.3353, 1.9133], device='cuda:1'), covar=tensor([0.2013, 0.2015, 0.2028, 0.2029], device='cuda:1'), in_proj_covar=tensor([0.1123, 0.0876, 0.0995, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 14:32:48,653 INFO [optim.py:369] (1/2) 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 14:32:51,390 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=190671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:33:14,599 INFO [train.py:968] (1/2) Epoch 5, batch 8800, giga_loss[loss=0.3364, simple_loss=0.3952, pruned_loss=0.1388, over 28945.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.397, pruned_loss=0.1412, over 5683270.86 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3773, pruned_loss=0.1193, over 5711818.04 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3994, pruned_loss=0.144, over 5665789.59 frames. ], batch size: 136, lr: 6.57e-03, grad_scale: 8.0
+2023-03-02 14:33:24,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:33:50,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-02 14:33:58,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5415, 1.5755, 1.0809, 1.2916], device='cuda:1'), covar=tensor([0.0553, 0.0464, 0.0896, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0450, 0.0501, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 14:33:58,527 INFO [train.py:968] (1/2) Epoch 5, batch 8850, giga_loss[loss=0.4073, simple_loss=0.4396, pruned_loss=0.1875, over 27567.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3993, pruned_loss=0.143, over 5681558.99 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3771, pruned_loss=0.1191, over 5714900.94 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.402, pruned_loss=0.146, over 5663884.48 frames. ], batch size: 472, lr: 6.57e-03, grad_scale: 8.0
+2023-03-02 14:34:14,592 INFO [optim.py:369] (1/2) 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 14:34:36,306 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:34:37,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 14:34:42,645 INFO [train.py:968] (1/2) Epoch 5, batch 8900, giga_loss[loss=0.3007, simple_loss=0.3733, pruned_loss=0.1141, over 28984.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.4004, pruned_loss=0.1444, over 5668317.15 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.377, pruned_loss=0.1189, over 5710102.03 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4035, pruned_loss=0.1478, over 5656387.47 frames. ], batch size: 213, lr: 6.57e-03, grad_scale: 4.0
+2023-03-02 14:35:06,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-02 14:35:25,728 INFO [train.py:968] (1/2) Epoch 5, batch 8950, giga_loss[loss=0.2894, simple_loss=0.355, pruned_loss=0.1119, over 29018.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3988, pruned_loss=0.1442, over 5668876.28 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3768, pruned_loss=0.119, over 5716752.79 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4023, pruned_loss=0.1478, over 5651670.85 frames. ], batch size: 136, lr: 6.57e-03, grad_scale: 1.0
+2023-03-02 14:35:47,557 INFO [optim.py:369] (1/2) 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 14:36:17,695 INFO [train.py:968] (1/2) Epoch 5, batch 9000, giga_loss[loss=0.2785, simple_loss=0.3476, pruned_loss=0.1046, over 28917.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3977, pruned_loss=0.1446, over 5653115.12 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3766, pruned_loss=0.1188, over 5717667.31 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4009, pruned_loss=0.148, over 5638148.67 frames. ], batch size: 112, lr: 6.57e-03, grad_scale: 1.0
+2023-03-02 14:36:17,696 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 14:36:27,677 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 14:36:33,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:36:36,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2110, 2.4401, 1.2527, 1.2535], device='cuda:1'), covar=tensor([0.0883, 0.0355, 0.0829, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0487, 0.0313, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 14:36:41,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=190919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:37:12,374 INFO [train.py:968] (1/2) Epoch 5, batch 9050, giga_loss[loss=0.357, simple_loss=0.4062, pruned_loss=0.1539, over 28715.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3958, pruned_loss=0.1441, over 5657147.98 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3767, pruned_loss=0.1189, over 5716591.88 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3986, pruned_loss=0.1471, over 5644977.52 frames. ], batch size: 262, lr: 6.56e-03, grad_scale: 1.0
+2023-03-02 14:37:31,311 INFO [optim.py:369] (1/2) 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 14:37:46,760 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=190984.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:37:59,908 INFO [train.py:968] (1/2) Epoch 5, batch 9100, giga_loss[loss=0.3112, simple_loss=0.3759, pruned_loss=0.1233, over 29010.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3954, pruned_loss=0.1445, over 5665495.34 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3762, pruned_loss=0.1186, over 5720473.14 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.3984, pruned_loss=0.1477, over 5651374.29 frames. ], batch size: 136, lr: 6.56e-03, grad_scale: 1.0
+2023-03-02 14:38:03,007 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 9150, giga_loss[loss=0.343, simple_loss=0.4006, pruned_loss=0.1427, over 28786.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3964, pruned_loss=0.1457, over 5655348.45 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3759, pruned_loss=0.1183, over 5725481.52 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.3995, pruned_loss=0.1491, over 5638458.14 frames. ], batch size: 243, lr: 6.56e-03, grad_scale: 1.0
+2023-03-02 14:38:52,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191050.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:38:54,733 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191053.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:39:03,715 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:06,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:08,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:12,139 INFO [optim.py:369] (1/2) 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 14:39:22,489 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191082.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:39:24,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:32,256 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:39:37,198 INFO [train.py:968] (1/2) Epoch 5, batch 9200, giga_loss[loss=0.3156, simple_loss=0.3782, pruned_loss=0.1266, over 28924.00 frames. ], tot_loss[loss=0.3425, simple_loss=0.395, pruned_loss=0.145, over 5662900.48 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3764, pruned_loss=0.1186, over 5725709.31 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.3974, pruned_loss=0.148, over 5647506.76 frames. ], batch size: 145, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:39:58,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2493, 1.9086, 1.9264, 1.6975], device='cuda:1'), covar=tensor([0.1089, 0.2263, 0.1657, 0.1718], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0740, 0.0635, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 14:40:24,271 INFO [train.py:968] (1/2) Epoch 5, batch 9250, giga_loss[loss=0.3815, simple_loss=0.4221, pruned_loss=0.1704, over 28668.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3918, pruned_loss=0.1429, over 5666198.71 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3766, pruned_loss=0.1186, over 5728991.70 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.394, pruned_loss=0.1459, over 5649329.60 frames. ], batch size: 262, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:40:41,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:40:43,305 INFO [optim.py:369] (1/2) 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 14:41:07,294 INFO [train.py:968] (1/2) Epoch 5, batch 9300, giga_loss[loss=0.3121, simple_loss=0.3689, pruned_loss=0.1276, over 28433.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3909, pruned_loss=0.141, over 5667304.62 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3761, pruned_loss=0.1182, over 5735921.33 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3937, pruned_loss=0.1447, over 5644770.50 frames. ], batch size: 65, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:41:09,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3895, 4.2276, 4.0630, 1.8345], device='cuda:1'), covar=tensor([0.0433, 0.0431, 0.0716, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.0858, 0.0770, 0.0827, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 14:41:09,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5710, 1.7930, 1.1644, 1.0227], device='cuda:1'), covar=tensor([0.1123, 0.0811, 0.0837, 0.1010], device='cuda:1'), in_proj_covar=tensor([0.1389, 0.1167, 0.1158, 0.1225], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 14:41:34,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:41:36,173 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:41:47,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2958, 1.4307, 1.2918, 1.3014], device='cuda:1'), covar=tensor([0.1937, 0.1884, 0.1865, 0.1747], device='cuda:1'), in_proj_covar=tensor([0.1124, 0.0876, 0.0997, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 14:41:51,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4956, 1.5968, 1.5492, 1.5844], device='cuda:1'), covar=tensor([0.0877, 0.1250, 0.1080, 0.1055], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0743, 0.0636, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 14:41:54,933 INFO [train.py:968] (1/2) Epoch 5, batch 9350, giga_loss[loss=0.3483, simple_loss=0.3978, pruned_loss=0.1494, over 28808.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3936, pruned_loss=0.142, over 5675701.22 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3761, pruned_loss=0.1181, over 5741092.06 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3964, pruned_loss=0.1458, over 5650888.64 frames. ], batch size: 119, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:41:58,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3373, 1.5531, 1.0362, 0.8387], device='cuda:1'), covar=tensor([0.0897, 0.0735, 0.0670, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.1374, 0.1156, 0.1148, 0.1215], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 14:42:02,489 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:42:06,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2358, 4.0398, 3.9155, 1.8658], device='cuda:1'), covar=tensor([0.0415, 0.0446, 0.0732, 0.1885], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0769, 0.0832, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 14:42:12,809 INFO [optim.py:369] (1/2) 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 14:42:42,550 INFO [train.py:968] (1/2) Epoch 5, batch 9400, libri_loss[loss=0.2571, simple_loss=0.327, pruned_loss=0.09359, over 29401.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3949, pruned_loss=0.1435, over 5666695.54 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3758, pruned_loss=0.1179, over 5745921.79 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.398, pruned_loss=0.1475, over 5640523.08 frames. ], batch size: 67, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:42:50,884 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:42:54,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:43:12,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-02 14:43:21,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:43:28,185 INFO [train.py:968] (1/2) Epoch 5, batch 9450, giga_loss[loss=0.3422, simple_loss=0.4128, pruned_loss=0.1358, over 28584.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3958, pruned_loss=0.1438, over 5673479.82 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3759, pruned_loss=0.118, over 5748371.94 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3988, pruned_loss=0.1478, over 5647870.61 frames. ], batch size: 85, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:43:35,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191359.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:43:47,461 INFO [optim.py:369] (1/2) 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 14:43:54,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:43:55,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5740, 3.6069, 1.6491, 1.5062], device='cuda:1'), covar=tensor([0.0763, 0.0245, 0.0828, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0483, 0.0311, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 14:44:15,915 INFO [train.py:968] (1/2) Epoch 5, batch 9500, giga_loss[loss=0.4413, simple_loss=0.4449, pruned_loss=0.2189, over 23725.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3967, pruned_loss=0.1421, over 5675490.57 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3754, pruned_loss=0.1177, over 5752563.40 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.4001, pruned_loss=0.1461, over 5649767.83 frames. ], batch size: 705, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:44:54,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:45:00,228 INFO [train.py:968] (1/2) Epoch 5, batch 9550, giga_loss[loss=0.3575, simple_loss=0.4139, pruned_loss=0.1505, over 28966.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3975, pruned_loss=0.1407, over 5685330.73 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3748, pruned_loss=0.1174, over 5755283.07 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.4013, pruned_loss=0.1449, over 5660098.82 frames. ], batch size: 227, lr: 6.56e-03, grad_scale: 2.0
+2023-03-02 14:45:10,023 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:45:18,812 INFO [optim.py:369] (1/2) 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 14:45:19,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7651, 1.0421, 3.4129, 2.9456], device='cuda:1'), covar=tensor([0.1744, 0.2332, 0.0478, 0.0732], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0525, 0.0747, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 14:45:47,325 INFO [train.py:968] (1/2) Epoch 5, batch 9600, giga_loss[loss=0.3109, simple_loss=0.3813, pruned_loss=0.1203, over 28791.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3993, pruned_loss=0.1408, over 5688522.69 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3751, pruned_loss=0.1176, over 5754593.91 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.4027, pruned_loss=0.1444, over 5667495.59 frames. ], batch size: 199, lr: 6.56e-03, grad_scale: 4.0
+2023-03-02 14:45:50,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191502.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:45:54,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191505.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 14:46:08,424 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:10,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:12,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:22,138 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 14:46:35,850 INFO [train.py:968] (1/2) Epoch 5, batch 9650, giga_loss[loss=0.3098, simple_loss=0.3745, pruned_loss=0.1226, over 29059.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.4034, pruned_loss=0.1456, over 5679688.09 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3749, pruned_loss=0.1175, over 5748795.46 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4069, pruned_loss=0.1492, over 5666705.89 frames. ], batch size: 128, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:46:38,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:46:54,456 INFO [optim.py:369] (1/2) 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 14:47:01,074 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:47:10,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:47:14,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:47:22,661 INFO [train.py:968] (1/2) Epoch 5, batch 9700, giga_loss[loss=0.4324, simple_loss=0.4415, pruned_loss=0.2116, over 26528.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4051, pruned_loss=0.1482, over 5671897.20 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3754, pruned_loss=0.1179, over 5752130.74 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4083, pruned_loss=0.1515, over 5656312.04 frames. ], batch size: 555, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:47:27,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-02 14:47:39,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=191618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:48:09,830 INFO [train.py:968] (1/2) Epoch 5, batch 9750, giga_loss[loss=0.3448, simple_loss=0.4066, pruned_loss=0.1415, over 28732.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.4048, pruned_loss=0.1484, over 5673759.79 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3753, pruned_loss=0.1179, over 5755470.38 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.408, pruned_loss=0.1517, over 5656922.02 frames. ], batch size: 262, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:48:29,948 INFO [optim.py:369] (1/2) 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 14:48:51,473 INFO [train.py:968] (1/2) Epoch 5, batch 9800, giga_loss[loss=0.3182, simple_loss=0.3849, pruned_loss=0.1257, over 28648.00 frames. ], tot_loss[loss=0.347, simple_loss=0.4023, pruned_loss=0.1458, over 5680538.04 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3746, pruned_loss=0.1175, over 5758704.36 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4062, pruned_loss=0.1496, over 5662019.95 frames. ], batch size: 262, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:49:39,397 INFO [train.py:968] (1/2) Epoch 5, batch 9850, giga_loss[loss=0.3529, simple_loss=0.4141, pruned_loss=0.1458, over 28871.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.4028, pruned_loss=0.1445, over 5672167.11 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3747, pruned_loss=0.1176, over 5749688.84 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4059, pruned_loss=0.1474, over 5666262.09 frames. ], batch size: 186, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:49:46,627 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=191758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:50:01,846 INFO [optim.py:369] (1/2) 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 14:50:09,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3317, 3.0134, 1.3054, 1.2956], device='cuda:1'), covar=tensor([0.0857, 0.0339, 0.0870, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0482, 0.0313, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 14:50:24,600 INFO [train.py:968] (1/2) Epoch 5, batch 9900, giga_loss[loss=0.4805, simple_loss=0.4775, pruned_loss=0.2418, over 23785.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.4036, pruned_loss=0.1446, over 5664932.80 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.375, pruned_loss=0.1178, over 5742925.20 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4061, pruned_loss=0.1471, over 5664636.92 frames. ], batch size: 705, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:51:00,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:51:13,121 INFO [train.py:968] (1/2) Epoch 5, batch 9950, giga_loss[loss=0.367, simple_loss=0.4195, pruned_loss=0.1573, over 28620.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.4043, pruned_loss=0.1457, over 5667393.53 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3745, pruned_loss=0.1174, over 5746953.55 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4076, pruned_loss=0.1489, over 5661178.53 frames. ], batch size: 262, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:51:24,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-02 14:51:33,503 INFO [optim.py:369] (1/2) 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 14:52:02,040 INFO [train.py:968] (1/2) Epoch 5, batch 10000, giga_loss[loss=0.3391, simple_loss=0.3944, pruned_loss=0.1419, over 28669.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4043, pruned_loss=0.1467, over 5661650.43 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.375, pruned_loss=0.1176, over 5750511.76 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4073, pruned_loss=0.1498, over 5651780.31 frames. ], batch size: 284, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:52:02,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:52:50,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 14:52:50,603 INFO [train.py:968] (1/2) Epoch 5, batch 10050, libri_loss[loss=0.3679, simple_loss=0.4183, pruned_loss=0.1588, over 19064.00 frames. ], tot_loss[loss=0.3495, simple_loss=0.4034, pruned_loss=0.1479, over 5641153.10 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3753, pruned_loss=0.118, over 5739375.55 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4059, pruned_loss=0.1505, over 5642726.95 frames. ], batch size: 188, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:52:52,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=191953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:53:00,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4311, 1.9848, 1.4569, 0.7430], device='cuda:1'), covar=tensor([0.2201, 0.1073, 0.1569, 0.2289], device='cuda:1'), in_proj_covar=tensor([0.1359, 0.1256, 0.1345, 0.1129], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 14:53:10,402 INFO [optim.py:369] (1/2) 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 14:53:17,087 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=191979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:53:19,057 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=191982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:53:35,515 INFO [train.py:968] (1/2) Epoch 5, batch 10100, giga_loss[loss=0.3619, simple_loss=0.4091, pruned_loss=0.1573, over 27856.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.4008, pruned_loss=0.1463, over 5659038.30 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3758, pruned_loss=0.1182, over 5743817.51 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4034, pruned_loss=0.1491, over 5653388.98 frames. ], batch size: 412, lr: 6.55e-03, grad_scale: 4.0
+2023-03-02 14:53:45,440 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:54:16,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:54:19,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:54:21,964 INFO [train.py:968] (1/2) Epoch 5, batch 10150, giga_loss[loss=0.3735, simple_loss=0.3944, pruned_loss=0.1763, over 23226.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3973, pruned_loss=0.1444, over 5654492.52 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3754, pruned_loss=0.1181, over 5747189.85 frames. ], giga_tot_loss[loss=0.3477, simple_loss=0.4002, pruned_loss=0.1476, over 5644909.32 frames. ], batch size: 705, lr: 6.55e-03, grad_scale: 2.0
+2023-03-02 14:54:29,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0706, 1.1735, 1.1853, 1.1206], device='cuda:1'), covar=tensor([0.0848, 0.0988, 0.1479, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0740, 0.0635, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 14:54:45,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 14:54:46,558 INFO [optim.py:369] (1/2) 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 14:54:48,447 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:07,190 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:10,012 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:10,353 INFO [train.py:968] (1/2) Epoch 5, batch 10200, giga_loss[loss=0.3538, simple_loss=0.4044, pruned_loss=0.1516, over 28978.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3983, pruned_loss=0.1463, over 5656161.20 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3757, pruned_loss=0.1181, over 5747192.06 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4009, pruned_loss=0.1494, over 5646652.65 frames. ], batch size: 136, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:55:16,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-02 14:55:27,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7170, 2.2534, 1.2735, 1.1508], device='cuda:1'), covar=tensor([0.1083, 0.0745, 0.1043, 0.1143], device='cuda:1'), in_proj_covar=tensor([0.1404, 0.1183, 0.1171, 0.1246], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 14:55:38,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=192128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:43,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:55:57,450 INFO [train.py:968] (1/2) Epoch 5, batch 10250, giga_loss[loss=0.3087, simple_loss=0.3773, pruned_loss=0.1201, over 28974.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3957, pruned_loss=0.1442, over 5659932.94 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3757, pruned_loss=0.118, over 5749245.05 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.3981, pruned_loss=0.1471, over 5649260.82 frames. ], batch size: 227, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:56:18,252 INFO [optim.py:369] (1/2) 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 14:56:42,546 INFO [train.py:968] (1/2) Epoch 5, batch 10300, libri_loss[loss=0.2764, simple_loss=0.3534, pruned_loss=0.0997, over 29549.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3913, pruned_loss=0.1388, over 5668122.13 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3756, pruned_loss=0.1179, over 5752426.35 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3937, pruned_loss=0.1417, over 5654809.77 frames. ], batch size: 75, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:57:05,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 14:57:28,425 INFO [train.py:968] (1/2) Epoch 5, batch 10350, giga_loss[loss=0.3137, simple_loss=0.3822, pruned_loss=0.1226, over 29085.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3883, pruned_loss=0.1355, over 5669914.63 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3759, pruned_loss=0.1181, over 5755957.27 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3903, pruned_loss=0.1383, over 5653969.17 frames. ], batch size: 128, lr: 6.54e-03, grad_scale: 2.0
+2023-03-02 14:57:49,153 INFO [optim.py:369] (1/2) 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 14:57:51,751 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=192276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:57:54,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=192279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 14:58:14,127 INFO [train.py:968] (1/2) Epoch 5, batch 10400, giga_loss[loss=0.2956, simple_loss=0.3643, pruned_loss=0.1134, over 28509.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3877, pruned_loss=0.1346, over 5669738.36 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3758, pruned_loss=0.1179, over 5749057.82 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3898, pruned_loss=0.1375, over 5659707.02 frames. ], batch size: 78, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 14:58:22,135 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 10450, libri_loss[loss=0.3095, simple_loss=0.3892, pruned_loss=0.1149, over 27777.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3851, pruned_loss=0.134, over 5670407.27 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3754, pruned_loss=0.1175, over 5751331.61 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3875, pruned_loss=0.1373, over 5657687.59 frames. ], batch size: 116, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 14:59:24,002 INFO [optim.py:369] (1/2) 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 14:59:26,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2581, 1.3165, 1.0407, 1.4417], device='cuda:1'), covar=tensor([0.0772, 0.0338, 0.0352, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0127, 0.0131, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0055, 0.0041, 0.0036, 0.0061], device='cuda:1')
+2023-03-02 14:59:51,627 INFO [train.py:968] (1/2) Epoch 5, batch 10500, giga_loss[loss=0.2995, simple_loss=0.357, pruned_loss=0.121, over 28830.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3831, pruned_loss=0.1338, over 5664084.85 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3755, pruned_loss=0.1175, over 5752932.01 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.385, pruned_loss=0.1366, over 5651386.54 frames. ], batch size: 284, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:00:02,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3471, 2.0044, 1.7432, 1.5903], device='cuda:1'), covar=tensor([0.1617, 0.1978, 0.1311, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0764, 0.0783, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:00:24,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-02 15:00:30,384 INFO [train.py:968] (1/2) Epoch 5, batch 10550, giga_loss[loss=0.3447, simple_loss=0.4008, pruned_loss=0.1443, over 28830.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.387, pruned_loss=0.136, over 5665191.57 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3761, pruned_loss=0.1178, over 5748629.51 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3884, pruned_loss=0.1386, over 5655807.96 frames. ], batch size: 119, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:00:49,901 INFO [optim.py:369] (1/2) 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 15:01:04,395 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 15:01:10,711 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 10600, giga_loss[loss=0.3708, simple_loss=0.4198, pruned_loss=0.161, over 28562.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3895, pruned_loss=0.1377, over 5652338.36 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3758, pruned_loss=0.1176, over 5741862.35 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3911, pruned_loss=0.1403, over 5649870.17 frames. ], batch size: 336, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:01:56,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:01:58,330 INFO [train.py:968] (1/2) Epoch 5, batch 10650, giga_loss[loss=0.3252, simple_loss=0.3836, pruned_loss=0.1334, over 28907.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.388, pruned_loss=0.1369, over 5650084.33 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3758, pruned_loss=0.1178, over 5745954.43 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3897, pruned_loss=0.1395, over 5641830.82 frames. ], batch size: 227, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:02:09,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-02 15:02:18,923 INFO [optim.py:369] (1/2) 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 15:02:39,564 INFO [train.py:968] (1/2) Epoch 5, batch 10700, libri_loss[loss=0.3072, simple_loss=0.3856, pruned_loss=0.1144, over 29635.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3883, pruned_loss=0.1369, over 5659251.56 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3767, pruned_loss=0.118, over 5751251.70 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3894, pruned_loss=0.1396, over 5644003.30 frames. ], batch size: 91, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:02:50,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:02:51,722 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:03:21,750 INFO [train.py:968] (1/2) Epoch 5, batch 10750, giga_loss[loss=0.5236, simple_loss=0.5093, pruned_loss=0.2689, over 27496.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3909, pruned_loss=0.1388, over 5666600.87 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.377, pruned_loss=0.1182, over 5751853.22 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3918, pruned_loss=0.1415, over 5650796.06 frames. ], batch size: 472, lr: 6.54e-03, grad_scale: 4.0
+2023-03-02 15:03:42,666 INFO [optim.py:369] (1/2) 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 15:04:04,112 INFO [train.py:968] (1/2) Epoch 5, batch 10800, giga_loss[loss=0.3541, simple_loss=0.4152, pruned_loss=0.1465, over 28296.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.393, pruned_loss=0.1399, over 5661923.76 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3772, pruned_loss=0.1181, over 5750932.90 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3943, pruned_loss=0.1431, over 5646168.03 frames. ], batch size: 71, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:04:17,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3990, 1.7771, 1.3156, 1.6572], device='cuda:1'), covar=tensor([0.0783, 0.0286, 0.0329, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0126, 0.0130, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0061], device='cuda:1')
+2023-03-02 15:04:30,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3734, 1.4496, 1.2389, 1.4487], device='cuda:1'), covar=tensor([0.1852, 0.1747, 0.1630, 0.1819], device='cuda:1'), in_proj_covar=tensor([0.1120, 0.0876, 0.0992, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 15:04:36,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4827, 1.7311, 1.7582, 1.6792], device='cuda:1'), covar=tensor([0.1322, 0.1661, 0.1021, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0761, 0.0782, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:04:45,160 INFO [train.py:968] (1/2) Epoch 5, batch 10850, libri_loss[loss=0.3124, simple_loss=0.3836, pruned_loss=0.1205, over 29533.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.394, pruned_loss=0.1405, over 5659031.93 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3774, pruned_loss=0.1182, over 5741330.46 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3955, pruned_loss=0.1438, over 5651788.76 frames. ], batch size: 84, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:05:05,148 INFO [optim.py:369] (1/2) 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 15:05:08,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1644, 3.9566, 3.8125, 1.7340], device='cuda:1'), covar=tensor([0.0506, 0.0501, 0.0857, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0783, 0.0838, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 15:05:13,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:05:21,543 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:05:23,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3537, 1.7760, 1.7232, 1.5707], device='cuda:1'), covar=tensor([0.1396, 0.1789, 0.1038, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0760, 0.0782, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:05:29,011 INFO [train.py:968] (1/2) Epoch 5, batch 10900, giga_loss[loss=0.3624, simple_loss=0.4075, pruned_loss=0.1587, over 28808.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3941, pruned_loss=0.1404, over 5677004.43 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3768, pruned_loss=0.1178, over 5746614.33 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3964, pruned_loss=0.1441, over 5664303.09 frames. ], batch size: 199, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:05:50,662 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=192826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:06:11,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4737, 1.5007, 1.5159, 1.4482], device='cuda:1'), covar=tensor([0.1644, 0.2673, 0.1431, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0756, 0.0777, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:06:11,974 INFO [train.py:968] (1/2) Epoch 5, batch 10950, giga_loss[loss=0.3378, simple_loss=0.4037, pruned_loss=0.136, over 28924.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3947, pruned_loss=0.1411, over 5682336.12 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3765, pruned_loss=0.1175, over 5750511.85 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3973, pruned_loss=0.145, over 5666566.05 frames. ], batch size: 227, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:06:31,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:06:34,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9291, 1.3353, 1.0529, 0.1743], device='cuda:1'), covar=tensor([0.1315, 0.1113, 0.1815, 0.1973], device='cuda:1'), in_proj_covar=tensor([0.1386, 0.1281, 0.1375, 0.1147], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 15:06:36,087 INFO [optim.py:369] (1/2) 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 15:06:59,029 INFO [train.py:968] (1/2) Epoch 5, batch 11000, giga_loss[loss=0.3717, simple_loss=0.4185, pruned_loss=0.1625, over 27993.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3958, pruned_loss=0.141, over 5673819.67 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3763, pruned_loss=0.1175, over 5753638.17 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3984, pruned_loss=0.1446, over 5657001.98 frames. ], batch size: 412, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:07:23,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:07:50,649 INFO [train.py:968] (1/2) Epoch 5, batch 11050, giga_loss[loss=0.3341, simple_loss=0.3813, pruned_loss=0.1435, over 28822.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3955, pruned_loss=0.1411, over 5667387.49 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3763, pruned_loss=0.1175, over 5757255.32 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3982, pruned_loss=0.1446, over 5648575.35 frames. ], batch size: 99, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:08:13,315 INFO [optim.py:369] (1/2) 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 15:08:24,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:27,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=192989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:39,808 INFO [train.py:968] (1/2) Epoch 5, batch 11100, giga_loss[loss=0.4121, simple_loss=0.436, pruned_loss=0.1941, over 28579.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3964, pruned_loss=0.1428, over 5662155.48 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3765, pruned_loss=0.1176, over 5757264.86 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3988, pruned_loss=0.1461, over 5645286.19 frames. ], batch size: 336, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:08:51,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:54,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:08:57,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:32,400 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:37,164 INFO [train.py:968] (1/2) Epoch 5, batch 11150, giga_loss[loss=0.3475, simple_loss=0.4001, pruned_loss=0.1475, over 28887.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3958, pruned_loss=0.1431, over 5654787.45 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3762, pruned_loss=0.1174, over 5760560.94 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3984, pruned_loss=0.1465, over 5635891.70 frames. ], batch size: 112, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:09:43,791 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2629, 1.4649, 1.1527, 1.3823], device='cuda:1'), covar=tensor([0.0774, 0.0329, 0.0346, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0126, 0.0130, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0055, 0.0040, 0.0036, 0.0061], device='cuda:1')
+2023-03-02 15:09:45,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2633, 1.3875, 4.7436, 3.3595], device='cuda:1'), covar=tensor([0.1648, 0.2110, 0.0313, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0530, 0.0746, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 15:09:51,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:51,929 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:54,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:09:57,635 INFO [optim.py:369] (1/2) 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 15:10:19,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:20,668 INFO [train.py:968] (1/2) Epoch 5, batch 11200, giga_loss[loss=0.3351, simple_loss=0.3996, pruned_loss=0.1352, over 28867.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3924, pruned_loss=0.141, over 5664552.25 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3754, pruned_loss=0.117, over 5764797.05 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.3959, pruned_loss=0.145, over 5642392.08 frames. ], batch size: 174, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:10:21,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4952, 1.7653, 1.8685, 1.7536], device='cuda:1'), covar=tensor([0.1215, 0.1466, 0.0936, 0.0984], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0760, 0.0784, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:10:32,278 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 15:10:47,228 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:48,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:50,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:50,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:10:52,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:06,051 INFO [train.py:968] (1/2) Epoch 5, batch 11250, giga_loss[loss=0.3171, simple_loss=0.3679, pruned_loss=0.1331, over 29000.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3924, pruned_loss=0.1416, over 5666224.93 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3754, pruned_loss=0.1167, over 5766028.11 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3956, pruned_loss=0.1455, over 5645214.84 frames. ], batch size: 106, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:11:15,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:16,741 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:19,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:23,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:11:28,315 INFO [optim.py:369] (1/2) 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 15:11:53,448 INFO [train.py:968] (1/2) Epoch 5, batch 11300, giga_loss[loss=0.3777, simple_loss=0.3978, pruned_loss=0.1788, over 23643.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.392, pruned_loss=0.1414, over 5663417.81 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3755, pruned_loss=0.1167, over 5766221.31 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3948, pruned_loss=0.1453, over 5643800.26 frames. ], batch size: 705, lr: 6.53e-03, grad_scale: 8.0
+2023-03-02 15:11:54,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:12:04,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8431, 1.0979, 3.6454, 2.9800], device='cuda:1'), covar=tensor([0.1726, 0.2229, 0.0399, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0529, 0.0747, 0.0607], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 15:12:11,712 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-02 15:12:15,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:12:24,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2502, 1.3444, 1.4426, 1.3888], device='cuda:1'), covar=tensor([0.0798, 0.0828, 0.1078, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0752, 0.0646, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 15:12:42,251 INFO [train.py:968] (1/2) Epoch 5, batch 11350, giga_loss[loss=0.3574, simple_loss=0.4077, pruned_loss=0.1536, over 28721.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3938, pruned_loss=0.143, over 5665817.63 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3761, pruned_loss=0.1171, over 5768620.52 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3961, pruned_loss=0.1466, over 5644729.64 frames. ], batch size: 284, lr: 6.53e-03, grad_scale: 4.0
+2023-03-02 15:13:03,145 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 11400, giga_loss[loss=0.381, simple_loss=0.4255, pruned_loss=0.1682, over 28211.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3956, pruned_loss=0.145, over 5664682.42 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.376, pruned_loss=0.1169, over 5771715.23 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.3979, pruned_loss=0.1486, over 5642885.70 frames. ], batch size: 368, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:13:34,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:13:35,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:13:40,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:13:40,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6125, 2.1149, 1.3011, 1.2350], device='cuda:1'), covar=tensor([0.1044, 0.0698, 0.0874, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.1406, 0.1197, 0.1195, 0.1254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 15:13:41,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:13:42,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5316, 1.8274, 1.8058, 1.6327], device='cuda:1'), covar=tensor([0.1431, 0.1834, 0.1081, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0754, 0.0780, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:14:00,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:06,503 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:08,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:12,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:15,777 INFO [train.py:968] (1/2) Epoch 5, batch 11450, giga_loss[loss=0.3888, simple_loss=0.4102, pruned_loss=0.1837, over 23532.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3964, pruned_loss=0.1457, over 5657049.80 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3756, pruned_loss=0.1167, over 5771935.54 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.3991, pruned_loss=0.1494, over 5637300.72 frames. ], batch size: 705, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:14:41,422 INFO [optim.py:369] (1/2) 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 15:14:44,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:14:51,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1678, 1.3060, 1.1497, 0.9354], device='cuda:1'), covar=tensor([0.2118, 0.2063, 0.2077, 0.1821], device='cuda:1'), in_proj_covar=tensor([0.1120, 0.0871, 0.0992, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 15:14:52,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:15:07,423 INFO [train.py:968] (1/2) Epoch 5, batch 11500, giga_loss[loss=0.3107, simple_loss=0.3691, pruned_loss=0.1261, over 28683.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3971, pruned_loss=0.1472, over 5651846.72 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3756, pruned_loss=0.1167, over 5771935.54 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.3992, pruned_loss=0.1502, over 5636475.77 frames. ], batch size: 92, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:15:22,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7998, 2.6726, 1.9192, 0.6844], device='cuda:1'), covar=tensor([0.2926, 0.1416, 0.1631, 0.3176], device='cuda:1'), in_proj_covar=tensor([0.1376, 0.1287, 0.1377, 0.1158], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 15:15:47,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:15:54,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:15:57,168 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-02 15:16:00,898 INFO [train.py:968] (1/2) Epoch 5, batch 11550, giga_loss[loss=0.3207, simple_loss=0.3887, pruned_loss=0.1264, over 28953.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.3971, pruned_loss=0.1471, over 5661331.78 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3753, pruned_loss=0.1165, over 5773419.59 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.3992, pruned_loss=0.1498, over 5646863.46 frames. ], batch size: 136, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:16:25,890 INFO [optim.py:369] (1/2) 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 15:16:41,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7507, 1.7012, 1.2844, 1.4506], device='cuda:1'), covar=tensor([0.0644, 0.0533, 0.0973, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0459, 0.0506, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 15:16:50,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4040, 1.3852, 1.0563, 1.1175], device='cuda:1'), covar=tensor([0.0597, 0.0489, 0.0943, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0460, 0.0506, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 15:16:52,475 INFO [train.py:968] (1/2) Epoch 5, batch 11600, giga_loss[loss=0.3526, simple_loss=0.4068, pruned_loss=0.1492, over 28713.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3985, pruned_loss=0.148, over 5656767.83 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3754, pruned_loss=0.1165, over 5775651.26 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4006, pruned_loss=0.1509, over 5640490.89 frames. ], batch size: 262, lr: 6.52e-03, grad_scale: 8.0
+2023-03-02 15:16:55,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:17:17,100 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:17:21,062 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 11650, giga_loss[loss=0.4406, simple_loss=0.4678, pruned_loss=0.2068, over 28669.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3984, pruned_loss=0.147, over 5673842.55 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3756, pruned_loss=0.1166, over 5778713.18 frames. ], giga_tot_loss[loss=0.3503, simple_loss=0.4004, pruned_loss=0.1501, over 5655528.86 frames. ], batch size: 284, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:17:46,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:18:01,745 INFO [optim.py:369] (1/2) 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 15:18:09,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:18:11,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:18:22,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193599.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:18:22,909 INFO [train.py:968] (1/2) Epoch 5, batch 11700, giga_loss[loss=0.4786, simple_loss=0.4767, pruned_loss=0.2402, over 26462.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3988, pruned_loss=0.1469, over 5675760.44 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3757, pruned_loss=0.1167, over 5782940.08 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4015, pruned_loss=0.1506, over 5652599.99 frames. ], batch size: 555, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:18:30,956 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:18:42,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:11,169 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:13,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:14,317 INFO [train.py:968] (1/2) Epoch 5, batch 11750, giga_loss[loss=0.3697, simple_loss=0.418, pruned_loss=0.1607, over 28746.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4015, pruned_loss=0.1496, over 5667935.92 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3757, pruned_loss=0.1167, over 5784269.96 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4042, pruned_loss=0.1533, over 5645694.10 frames. ], batch size: 99, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:19:39,058 INFO [optim.py:369] (1/2) 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 15:19:39,932 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:19:40,605 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 11800, giga_loss[loss=0.3844, simple_loss=0.4178, pruned_loss=0.1755, over 28555.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.4004, pruned_loss=0.1491, over 5668609.84 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.375, pruned_loss=0.1162, over 5787187.56 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.4037, pruned_loss=0.1533, over 5645556.34 frames. ], batch size: 336, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:20:40,808 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193742.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:20:42,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193745.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:20:45,660 INFO [train.py:968] (1/2) Epoch 5, batch 11850, giga_loss[loss=0.347, simple_loss=0.401, pruned_loss=0.1465, over 28481.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4018, pruned_loss=0.1495, over 5672129.28 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3755, pruned_loss=0.1165, over 5790804.60 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4048, pruned_loss=0.1535, over 5647042.43 frames. ], batch size: 336, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:21:09,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193774.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:21:12,457 INFO [optim.py:369] (1/2) 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 15:21:14,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4627, 1.6349, 1.3728, 0.8884], device='cuda:1'), covar=tensor([0.1238, 0.0930, 0.0661, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1194, 0.1182, 0.1255], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 15:21:31,229 INFO [train.py:968] (1/2) Epoch 5, batch 11900, giga_loss[loss=0.3155, simple_loss=0.3808, pruned_loss=0.1251, over 28990.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4025, pruned_loss=0.1488, over 5666509.48 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3758, pruned_loss=0.1167, over 5790352.79 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4055, pruned_loss=0.1529, over 5642901.19 frames. ], batch size: 213, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:21:41,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:21:47,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:22:09,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-02 15:22:17,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=193847.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:22:20,047 INFO [train.py:968] (1/2) Epoch 5, batch 11950, giga_loss[loss=0.3177, simple_loss=0.3865, pruned_loss=0.1245, over 28969.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.4024, pruned_loss=0.1485, over 5657000.66 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3762, pruned_loss=0.1169, over 5779830.53 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.4049, pruned_loss=0.152, over 5644883.59 frames. ], batch size: 136, lr: 6.52e-03, grad_scale: 4.0
+2023-03-02 15:22:44,848 INFO [optim.py:369] (1/2) 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 15:23:03,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2637, 1.6815, 1.2833, 0.4699], device='cuda:1'), covar=tensor([0.1262, 0.0851, 0.1246, 0.1899], device='cuda:1'), in_proj_covar=tensor([0.1379, 0.1288, 0.1379, 0.1146], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 15:23:06,781 INFO [train.py:968] (1/2) Epoch 5, batch 12000, giga_loss[loss=0.2939, simple_loss=0.3552, pruned_loss=0.1164, over 28697.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.4005, pruned_loss=0.1473, over 5652725.53 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3762, pruned_loss=0.1168, over 5780136.26 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.403, pruned_loss=0.1508, over 5640487.32 frames. ], batch size: 99, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:23:06,782 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 15:23:16,110 INFO [train.py:1012] (1/2) Epoch 5, validation: loss=0.2404, simple_loss=0.3437, pruned_loss=0.06857, over 944034.00 frames. 
+2023-03-02 15:23:16,110 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 15:24:04,117 INFO [train.py:968] (1/2) Epoch 5, batch 12050, giga_loss[loss=0.3233, simple_loss=0.3827, pruned_loss=0.1319, over 28817.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.4003, pruned_loss=0.1471, over 5658029.35 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.376, pruned_loss=0.1168, over 5780005.52 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4026, pruned_loss=0.1502, over 5647480.88 frames. ], batch size: 112, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:24:06,737 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=193952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:10,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=193955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:34,753 INFO [optim.py:369] (1/2) 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 15:24:36,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=193979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:39,904 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=193984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:24:54,331 INFO [train.py:968] (1/2) Epoch 5, batch 12100, giga_loss[loss=0.4239, simple_loss=0.4579, pruned_loss=0.195, over 28930.00 frames. ], tot_loss[loss=0.3484, simple_loss=0.4012, pruned_loss=0.1478, over 5656161.45 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3761, pruned_loss=0.1167, over 5782287.34 frames. ], giga_tot_loss[loss=0.3526, simple_loss=0.4033, pruned_loss=0.1509, over 5643327.56 frames. ], batch size: 227, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:25:04,365 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:25:44,431 INFO [train.py:968] (1/2) Epoch 5, batch 12150, giga_loss[loss=0.3265, simple_loss=0.3821, pruned_loss=0.1355, over 28957.00 frames. ], tot_loss[loss=0.3466, simple_loss=0.3994, pruned_loss=0.147, over 5671107.67 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3761, pruned_loss=0.1166, over 5782905.90 frames. ], giga_tot_loss[loss=0.3502, simple_loss=0.4012, pruned_loss=0.1496, over 5659986.67 frames. ], batch size: 164, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:25:46,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:25:51,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7686, 1.8229, 1.5975, 1.6906], device='cuda:1'), covar=tensor([0.1135, 0.1755, 0.1510, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0754, 0.0645, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 15:26:07,737 INFO [optim.py:369] (1/2) 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 15:26:30,738 INFO [train.py:968] (1/2) Epoch 5, batch 12200, giga_loss[loss=0.3307, simple_loss=0.3911, pruned_loss=0.1351, over 28954.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.4005, pruned_loss=0.1485, over 5661856.68 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3766, pruned_loss=0.1168, over 5774017.22 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4023, pruned_loss=0.1513, over 5657127.28 frames. ], batch size: 136, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:26:52,429 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:26:54,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:26:56,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5544, 2.2332, 1.6086, 0.7219], device='cuda:1'), covar=tensor([0.2354, 0.1184, 0.1787, 0.2635], device='cuda:1'), in_proj_covar=tensor([0.1384, 0.1298, 0.1378, 0.1149], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 15:27:15,841 INFO [train.py:968] (1/2) Epoch 5, batch 12250, giga_loss[loss=0.3725, simple_loss=0.4206, pruned_loss=0.1622, over 28992.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4017, pruned_loss=0.1491, over 5669846.58 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3766, pruned_loss=0.1168, over 5777675.60 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4037, pruned_loss=0.1523, over 5659757.10 frames. ], batch size: 128, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:27:20,928 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:27:32,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7786, 2.2719, 2.1022, 1.8461], device='cuda:1'), covar=tensor([0.1505, 0.1689, 0.1081, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0759, 0.0780, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:27:42,084 INFO [optim.py:369] (1/2) 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 15:27:47,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-02 15:27:54,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:00,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:02,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:03,532 INFO [train.py:968] (1/2) Epoch 5, batch 12300, giga_loss[loss=0.3502, simple_loss=0.4049, pruned_loss=0.1477, over 28637.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4018, pruned_loss=0.1494, over 5667052.37 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3765, pruned_loss=0.1168, over 5779414.84 frames. ], giga_tot_loss[loss=0.3545, simple_loss=0.404, pruned_loss=0.1525, over 5655563.55 frames. ], batch size: 307, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:28:25,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194222.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:28:30,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:28:48,898 INFO [train.py:968] (1/2) Epoch 5, batch 12350, giga_loss[loss=0.3683, simple_loss=0.4258, pruned_loss=0.1554, over 29070.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4014, pruned_loss=0.1483, over 5669930.73 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3771, pruned_loss=0.1172, over 5770044.37 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4036, pruned_loss=0.1516, over 5664570.08 frames. ], batch size: 128, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:28:50,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3139, 5.0845, 2.1157, 2.3704], device='cuda:1'), covar=tensor([0.0751, 0.0245, 0.0800, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0488, 0.0315, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 15:29:16,132 INFO [optim.py:369] (1/2) 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 15:29:17,120 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 12400, giga_loss[loss=0.4256, simple_loss=0.4355, pruned_loss=0.2079, over 23819.00 frames. ], tot_loss[loss=0.348, simple_loss=0.4007, pruned_loss=0.1477, over 5657198.83 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.377, pruned_loss=0.1171, over 5769558.20 frames. ], giga_tot_loss[loss=0.3532, simple_loss=0.4033, pruned_loss=0.1515, over 5650145.05 frames. ], batch size: 705, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:29:47,464 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 15:30:06,872 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:30:08,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:30:14,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5017, 3.5425, 1.5253, 1.5413], device='cuda:1'), covar=tensor([0.0878, 0.0373, 0.0846, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0487, 0.0315, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 15:30:22,775 INFO [train.py:968] (1/2) Epoch 5, batch 12450, giga_loss[loss=0.4153, simple_loss=0.4438, pruned_loss=0.1934, over 27642.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3996, pruned_loss=0.1461, over 5662482.91 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.377, pruned_loss=0.1172, over 5761430.96 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4018, pruned_loss=0.1492, over 5662965.15 frames. ], batch size: 472, lr: 6.51e-03, grad_scale: 8.0
+2023-03-02 15:30:36,040 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:30:36,664 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194365.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:30:38,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194368.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:30:47,942 INFO [optim.py:369] (1/2) 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 15:30:52,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:31:09,215 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194397.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:31:10,886 INFO [train.py:968] (1/2) Epoch 5, batch 12500, giga_loss[loss=0.3323, simple_loss=0.3986, pruned_loss=0.133, over 28871.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.399, pruned_loss=0.1451, over 5677105.74 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3773, pruned_loss=0.1173, over 5762510.79 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4007, pruned_loss=0.1477, over 5675789.84 frames. ], batch size: 174, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:32:00,399 INFO [train.py:968] (1/2) Epoch 5, batch 12550, libri_loss[loss=0.3331, simple_loss=0.4008, pruned_loss=0.1327, over 29762.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3973, pruned_loss=0.1445, over 5671519.34 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3777, pruned_loss=0.1175, over 5764121.36 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3985, pruned_loss=0.1466, over 5668087.18 frames. ], batch size: 87, lr: 6.51e-03, grad_scale: 4.0
+2023-03-02 15:32:18,326 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:32:28,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3614, 1.5296, 1.3217, 1.5493], device='cuda:1'), covar=tensor([0.2115, 0.2021, 0.1987, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.1134, 0.0879, 0.0998, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 15:32:29,941 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 12600, giga_loss[loss=0.3173, simple_loss=0.3803, pruned_loss=0.1272, over 29003.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3956, pruned_loss=0.1444, over 5665470.79 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3778, pruned_loss=0.1175, over 5761964.64 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3966, pruned_loss=0.1463, over 5664086.05 frames. ], batch size: 164, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:33:17,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:33:19,298 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 12650, libri_loss[loss=0.3109, simple_loss=0.3872, pruned_loss=0.1173, over 29533.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3919, pruned_loss=0.1423, over 5678712.22 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3778, pruned_loss=0.1175, over 5765452.09 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3931, pruned_loss=0.1444, over 5672131.24 frames. ], batch size: 83, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:33:45,375 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:34:04,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-02 15:34:04,874 INFO [optim.py:369] (1/2) 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 15:34:07,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8323, 1.6685, 1.1928, 1.4291], device='cuda:1'), covar=tensor([0.0588, 0.0611, 0.1031, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0455, 0.0504, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 15:34:28,173 INFO [train.py:968] (1/2) Epoch 5, batch 12700, giga_loss[loss=0.3027, simple_loss=0.3574, pruned_loss=0.124, over 28403.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3888, pruned_loss=0.1406, over 5686299.64 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3772, pruned_loss=0.1171, over 5767534.26 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3904, pruned_loss=0.1429, over 5678349.39 frames. ], batch size: 78, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:35:14,794 INFO [train.py:968] (1/2) Epoch 5, batch 12750, giga_loss[loss=0.355, simple_loss=0.3849, pruned_loss=0.1626, over 23987.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3893, pruned_loss=0.1411, over 5685857.37 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3779, pruned_loss=0.1176, over 5767553.14 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3903, pruned_loss=0.1432, over 5677291.57 frames. ], batch size: 705, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:35:17,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:35:39,237 INFO [optim.py:369] (1/2) 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 15:36:00,490 INFO [train.py:968] (1/2) Epoch 5, batch 12800, giga_loss[loss=0.3604, simple_loss=0.4133, pruned_loss=0.1537, over 28581.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.389, pruned_loss=0.1394, over 5677668.83 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3782, pruned_loss=0.118, over 5758795.62 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3898, pruned_loss=0.1413, over 5676909.86 frames. ], batch size: 307, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:36:12,395 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:36:46,245 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:36:48,935 INFO [train.py:968] (1/2) Epoch 5, batch 12850, giga_loss[loss=0.3901, simple_loss=0.4144, pruned_loss=0.1829, over 26766.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3864, pruned_loss=0.1355, over 5676032.98 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3779, pruned_loss=0.1178, over 5760748.02 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3874, pruned_loss=0.1375, over 5672445.49 frames. ], batch size: 555, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:36:57,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=194755.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:37:12,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-02 15:37:20,292 INFO [optim.py:369] (1/2) 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 15:37:34,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4337, 2.3063, 1.6648, 2.2653], device='cuda:1'), covar=tensor([0.0504, 0.0496, 0.0814, 0.0739], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0446, 0.0498, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 15:37:35,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:37:37,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:37:38,016 INFO [train.py:968] (1/2) Epoch 5, batch 12900, giga_loss[loss=0.3495, simple_loss=0.401, pruned_loss=0.149, over 28589.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3832, pruned_loss=0.1323, over 5653752.76 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3778, pruned_loss=0.1179, over 5743840.74 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3843, pruned_loss=0.1343, over 5663254.40 frames. ], batch size: 336, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:37:39,865 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-02 15:37:41,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1873, 1.2697, 4.2762, 3.2870], device='cuda:1'), covar=tensor([0.1632, 0.2219, 0.0360, 0.0625], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0527, 0.0747, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 15:38:06,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=194827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:38:17,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=194840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:38:25,967 INFO [train.py:968] (1/2) Epoch 5, batch 12950, giga_loss[loss=0.2801, simple_loss=0.3521, pruned_loss=0.1041, over 27631.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3778, pruned_loss=0.1274, over 5657326.09 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3766, pruned_loss=0.1173, over 5748984.82 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.38, pruned_loss=0.13, over 5657545.44 frames. ], batch size: 472, lr: 6.50e-03, grad_scale: 8.0
+2023-03-02 15:38:54,950 INFO [optim.py:369] (1/2) 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 15:38:59,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-02 15:39:12,904 INFO [train.py:968] (1/2) Epoch 5, batch 13000, libri_loss[loss=0.3041, simple_loss=0.3792, pruned_loss=0.1146, over 28102.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.1231, over 5664453.81 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3749, pruned_loss=0.1164, over 5749684.66 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3767, pruned_loss=0.1265, over 5659984.66 frames. ], batch size: 116, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:39:50,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 15:39:58,107 INFO [train.py:968] (1/2) Epoch 5, batch 13050, giga_loss[loss=0.2998, simple_loss=0.3715, pruned_loss=0.1141, over 28309.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3722, pruned_loss=0.1205, over 5665164.71 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3745, pruned_loss=0.1164, over 5750042.87 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3752, pruned_loss=0.1234, over 5659180.66 frames. ], batch size: 368, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:40:29,022 INFO [optim.py:369] (1/2) 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 15:40:30,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3274, 1.4718, 1.2219, 1.4610], device='cuda:1'), covar=tensor([0.2376, 0.2126, 0.2232, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.1131, 0.0869, 0.0995, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 15:40:33,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=194983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:40:35,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=194986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:40:49,649 INFO [train.py:968] (1/2) Epoch 5, batch 13100, libri_loss[loss=0.3238, simple_loss=0.3815, pruned_loss=0.1331, over 29596.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3716, pruned_loss=0.1197, over 5653290.99 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3745, pruned_loss=0.1166, over 5744264.70 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.374, pruned_loss=0.1218, over 5651865.41 frames. ], batch size: 74, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:40:53,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3280, 1.4495, 1.0342, 1.1776], device='cuda:1'), covar=tensor([0.0743, 0.0563, 0.0597, 0.0597], device='cuda:1'), in_proj_covar=tensor([0.1386, 0.1158, 0.1170, 0.1222], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 15:41:04,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:41:15,700 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:41:38,443 INFO [train.py:968] (1/2) Epoch 5, batch 13150, giga_loss[loss=0.2944, simple_loss=0.3613, pruned_loss=0.1138, over 27991.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3707, pruned_loss=0.1186, over 5653683.18 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3741, pruned_loss=0.1164, over 5736505.81 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3728, pruned_loss=0.1205, over 5657840.47 frames. ], batch size: 412, lr: 6.50e-03, grad_scale: 4.0
+2023-03-02 15:42:07,578 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 5, batch 13200, giga_loss[loss=0.2784, simple_loss=0.351, pruned_loss=0.1029, over 28545.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3679, pruned_loss=0.1168, over 5646535.73 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3742, pruned_loss=0.1166, over 5727818.83 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3694, pruned_loss=0.1182, over 5655876.93 frames. ], batch size: 336, lr: 6.49e-03, grad_scale: 8.0
+2023-03-02 15:42:52,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:43:02,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195130.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:43:21,220 INFO [train.py:968] (1/2) Epoch 5, batch 13250, giga_loss[loss=0.258, simple_loss=0.341, pruned_loss=0.0875, over 28593.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3664, pruned_loss=0.116, over 5653516.98 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3744, pruned_loss=0.117, over 5731380.58 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3673, pruned_loss=0.1168, over 5656576.52 frames. ], batch size: 262, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:43:50,063 INFO [optim.py:369] (1/2) 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 15:44:10,728 INFO [train.py:968] (1/2) Epoch 5, batch 13300, giga_loss[loss=0.2958, simple_loss=0.3651, pruned_loss=0.1133, over 28800.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3657, pruned_loss=0.115, over 5659673.59 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3738, pruned_loss=0.1167, over 5731074.24 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3669, pruned_loss=0.1158, over 5661388.84 frames. ], batch size: 174, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:44:17,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9983, 1.2909, 1.2737, 1.2120], device='cuda:1'), covar=tensor([0.1120, 0.0994, 0.1569, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0727, 0.0621, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 15:44:28,490 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 15:44:37,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:44:39,976 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:44:47,830 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:44:56,773 INFO [train.py:968] (1/2) Epoch 5, batch 13350, giga_loss[loss=0.2553, simple_loss=0.3379, pruned_loss=0.08631, over 28932.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3642, pruned_loss=0.1137, over 5664398.61 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3728, pruned_loss=0.1161, over 5735985.30 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3657, pruned_loss=0.1148, over 5658612.83 frames. ], batch size: 145, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:45:08,815 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:12,391 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:15,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:22,399 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195273.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:45:25,865 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:45:44,530 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:45:49,062 INFO [train.py:968] (1/2) Epoch 5, batch 13400, giga_loss[loss=0.267, simple_loss=0.3399, pruned_loss=0.09706, over 28293.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3606, pruned_loss=0.1107, over 5666605.38 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3725, pruned_loss=0.116, over 5738561.38 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.362, pruned_loss=0.1117, over 5658755.62 frames. ], batch size: 368, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:45:54,123 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195305.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:45:54,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3539, 1.4862, 1.3154, 1.6532], device='cuda:1'), covar=tensor([0.2548, 0.2273, 0.2254, 0.2106], device='cuda:1'), in_proj_covar=tensor([0.1113, 0.0855, 0.0976, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 15:46:14,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3594, 2.5695, 1.4085, 1.3330], device='cuda:1'), covar=tensor([0.0779, 0.0352, 0.0806, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0483, 0.0317, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 15:46:40,765 INFO [train.py:968] (1/2) Epoch 5, batch 13450, libri_loss[loss=0.3398, simple_loss=0.3846, pruned_loss=0.1475, over 29545.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3566, pruned_loss=0.1082, over 5667532.01 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.372, pruned_loss=0.1159, over 5742497.97 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3577, pruned_loss=0.1089, over 5656179.43 frames. ], batch size: 78, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:47:09,648 INFO [optim.py:369] (1/2) 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 15:47:33,421 INFO [train.py:968] (1/2) Epoch 5, batch 13500, giga_loss[loss=0.2762, simple_loss=0.3472, pruned_loss=0.1026, over 28877.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3544, pruned_loss=0.108, over 5651132.84 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3713, pruned_loss=0.1154, over 5745433.84 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3556, pruned_loss=0.1087, over 5637914.33 frames. ], batch size: 227, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:47:36,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:47:38,518 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:48:14,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:48:26,840 INFO [train.py:968] (1/2) Epoch 5, batch 13550, giga_loss[loss=0.2745, simple_loss=0.3491, pruned_loss=0.09999, over 28980.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3552, pruned_loss=0.1095, over 5645897.46 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3713, pruned_loss=0.1156, over 5736209.38 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3559, pruned_loss=0.1099, over 5642107.26 frames. ], batch size: 136, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:48:43,282 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:48:59,958 INFO [optim.py:369] (1/2) 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 15:49:27,829 INFO [train.py:968] (1/2) Epoch 5, batch 13600, giga_loss[loss=0.3018, simple_loss=0.3688, pruned_loss=0.1174, over 28815.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3579, pruned_loss=0.1114, over 5636559.57 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3712, pruned_loss=0.1156, over 5738716.57 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3583, pruned_loss=0.1116, over 5630066.67 frames. ], batch size: 199, lr: 6.49e-03, grad_scale: 8.0
+2023-03-02 15:49:39,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9986, 3.8481, 3.6800, 1.8271], device='cuda:1'), covar=tensor([0.0507, 0.0476, 0.0751, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0752, 0.0790, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 15:50:13,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4730, 2.5300, 1.5274, 1.5768], device='cuda:1'), covar=tensor([0.0649, 0.0274, 0.0645, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0477, 0.0316, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 15:50:16,851 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:50:21,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:50:22,106 INFO [train.py:968] (1/2) Epoch 5, batch 13650, giga_loss[loss=0.3427, simple_loss=0.4119, pruned_loss=0.1368, over 28645.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3596, pruned_loss=0.1106, over 5645070.42 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3708, pruned_loss=0.1154, over 5742105.64 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.36, pruned_loss=0.1109, over 5634913.23 frames. ], batch size: 307, lr: 6.49e-03, grad_scale: 8.0
+2023-03-02 15:50:52,895 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:50:55,913 INFO [optim.py:369] (1/2) 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 15:51:08,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8774, 1.8264, 1.2589, 1.5409], device='cuda:1'), covar=tensor([0.0609, 0.0521, 0.0917, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0439, 0.0497, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 15:51:16,456 INFO [train.py:968] (1/2) Epoch 5, batch 13700, giga_loss[loss=0.3511, simple_loss=0.418, pruned_loss=0.1421, over 28963.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3593, pruned_loss=0.1103, over 5645112.05 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3694, pruned_loss=0.1146, over 5739830.82 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3606, pruned_loss=0.111, over 5634543.44 frames. ], batch size: 213, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:51:33,901 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 13750, giga_loss[loss=0.2961, simple_loss=0.3542, pruned_loss=0.1191, over 26703.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3589, pruned_loss=0.1104, over 5649062.31 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3689, pruned_loss=0.1144, over 5743297.60 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3601, pruned_loss=0.1111, over 5635457.88 frames. ], batch size: 555, lr: 6.49e-03, grad_scale: 4.0
+2023-03-02 15:52:33,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0334, 1.1354, 3.8013, 3.0021], device='cuda:1'), covar=tensor([0.1563, 0.2267, 0.0317, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0528, 0.0738, 0.0595], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 15:52:41,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7183, 1.0927, 2.9056, 2.7638], device='cuda:1'), covar=tensor([0.1456, 0.2048, 0.0449, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0526, 0.0736, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 15:52:50,661 INFO [optim.py:369] (1/2) 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 15:53:15,752 INFO [train.py:968] (1/2) Epoch 5, batch 13800, giga_loss[loss=0.2861, simple_loss=0.362, pruned_loss=0.1051, over 28715.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3561, pruned_loss=0.1081, over 5654103.24 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3681, pruned_loss=0.114, over 5746956.60 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3575, pruned_loss=0.1089, over 5637735.03 frames. ], batch size: 262, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:53:54,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-02 15:54:12,277 INFO [train.py:968] (1/2) Epoch 5, batch 13850, giga_loss[loss=0.2618, simple_loss=0.3344, pruned_loss=0.09457, over 28894.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3552, pruned_loss=0.1064, over 5652283.19 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3681, pruned_loss=0.1142, over 5743319.20 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3558, pruned_loss=0.1065, over 5639504.84 frames. ], batch size: 199, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:54:22,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:23,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4003, 1.8349, 1.7052, 1.6516], device='cuda:1'), covar=tensor([0.1581, 0.1749, 0.1236, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0734, 0.0769, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:54:24,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:46,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:47,696 INFO [optim.py:369] (1/2) 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 15:54:54,396 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=195785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:54:59,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:55:09,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5144, 2.1187, 1.8616, 1.7565], device='cuda:1'), covar=tensor([0.1637, 0.1757, 0.1234, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0735, 0.0769, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 15:55:13,628 INFO [train.py:968] (1/2) Epoch 5, batch 13900, giga_loss[loss=0.2915, simple_loss=0.3453, pruned_loss=0.1189, over 26835.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3522, pruned_loss=0.1049, over 5643716.20 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3682, pruned_loss=0.1143, over 5734133.16 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3523, pruned_loss=0.1048, over 5639349.78 frames. ], batch size: 555, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:55:25,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195812.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:55:59,093 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=195839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:56:10,888 INFO [train.py:968] (1/2) Epoch 5, batch 13950, giga_loss[loss=0.2574, simple_loss=0.34, pruned_loss=0.08739, over 28702.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3513, pruned_loss=0.1049, over 5654589.83 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3682, pruned_loss=0.1142, over 5736564.17 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3512, pruned_loss=0.1048, over 5648055.97 frames. ], batch size: 262, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:56:48,880 INFO [optim.py:369] (1/2) 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 15:57:09,176 INFO [train.py:968] (1/2) Epoch 5, batch 14000, giga_loss[loss=0.3388, simple_loss=0.4015, pruned_loss=0.1381, over 27771.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.351, pruned_loss=0.1048, over 5663391.41 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3679, pruned_loss=0.114, over 5740919.56 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3508, pruned_loss=0.1047, over 5652259.72 frames. ], batch size: 474, lr: 6.48e-03, grad_scale: 8.0
+2023-03-02 15:57:32,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:57:36,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:07,363 INFO [train.py:968] (1/2) Epoch 5, batch 14050, giga_loss[loss=0.2719, simple_loss=0.3593, pruned_loss=0.09223, over 28558.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3521, pruned_loss=0.1044, over 5672120.94 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.368, pruned_loss=0.1142, over 5743536.30 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3514, pruned_loss=0.1039, over 5659848.45 frames. ], batch size: 307, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:58:13,888 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:15,768 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195955.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 15:58:20,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195958.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:58:43,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8808, 1.9942, 1.8262, 1.8773], device='cuda:1'), covar=tensor([0.0998, 0.1658, 0.1503, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0736, 0.0633, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 15:58:46,986 INFO [optim.py:369] (1/2) 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 15:58:47,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=195982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:51,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=195985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:58:54,311 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=195987.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 15:59:09,428 INFO [train.py:968] (1/2) Epoch 5, batch 14100, giga_loss[loss=0.2404, simple_loss=0.3267, pruned_loss=0.07712, over 28937.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3557, pruned_loss=0.1061, over 5680046.75 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3681, pruned_loss=0.1143, over 5745239.85 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3549, pruned_loss=0.1054, over 5667538.89 frames. ], batch size: 145, lr: 6.48e-03, grad_scale: 4.0
+2023-03-02 15:59:25,747 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 15:59:41,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0086, 1.1017, 3.7134, 3.0322], device='cuda:1'), covar=tensor([0.1541, 0.2202, 0.0340, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0527, 0.0733, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 16:00:11,899 INFO [train.py:968] (1/2) Epoch 5, batch 14150, giga_loss[loss=0.2789, simple_loss=0.3505, pruned_loss=0.1037, over 28473.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3515, pruned_loss=0.1039, over 5674920.17 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3677, pruned_loss=0.1142, over 5739938.33 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3508, pruned_loss=0.1032, over 5667902.62 frames. ], batch size: 336, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:00:55,472 INFO [optim.py:369] (1/2) 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 16:01:17,633 INFO [train.py:968] (1/2) Epoch 5, batch 14200, giga_loss[loss=0.2602, simple_loss=0.3416, pruned_loss=0.08943, over 28888.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3524, pruned_loss=0.1049, over 5666887.61 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3674, pruned_loss=0.1142, over 5733692.79 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3519, pruned_loss=0.1042, over 5665665.35 frames. ], batch size: 227, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:02:25,659 INFO [train.py:968] (1/2) Epoch 5, batch 14250, giga_loss[loss=0.3204, simple_loss=0.3902, pruned_loss=0.1253, over 28068.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3555, pruned_loss=0.1054, over 5659996.87 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3665, pruned_loss=0.1137, over 5738478.84 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3556, pruned_loss=0.1051, over 5652579.11 frames. ], batch size: 412, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:02:37,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=196160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:03:04,449 INFO [optim.py:369] (1/2) 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 16:03:23,838 INFO [train.py:968] (1/2) Epoch 5, batch 14300, giga_loss[loss=0.2502, simple_loss=0.3459, pruned_loss=0.07721, over 28914.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3579, pruned_loss=0.1045, over 5665152.16 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3664, pruned_loss=0.1136, over 5743149.61 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3577, pruned_loss=0.104, over 5653038.56 frames. ], batch size: 164, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:03:30,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-02 16:03:48,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3011, 1.8997, 1.4104, 1.6385], device='cuda:1'), covar=tensor([0.0705, 0.0240, 0.0294, 0.0713], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0126, 0.0131, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 16:04:21,443 INFO [train.py:968] (1/2) Epoch 5, batch 14350, giga_loss[loss=0.2666, simple_loss=0.3492, pruned_loss=0.09206, over 28694.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3566, pruned_loss=0.1027, over 5651026.24 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3665, pruned_loss=0.1138, over 5744905.73 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3562, pruned_loss=0.102, over 5638623.78 frames. ], batch size: 262, lr: 6.48e-03, grad_scale: 2.0
+2023-03-02 16:05:03,017 INFO [optim.py:369] (1/2) 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 16:05:23,477 INFO [train.py:968] (1/2) Epoch 5, batch 14400, giga_loss[loss=0.276, simple_loss=0.3492, pruned_loss=0.1014, over 29043.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3564, pruned_loss=0.1016, over 5665276.62 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3665, pruned_loss=0.1138, over 5746380.53 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.356, pruned_loss=0.101, over 5653424.81 frames. ], batch size: 128, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:05:27,706 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=196303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:05:30,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=196306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:06:08,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=196335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:06:28,130 INFO [train.py:968] (1/2) Epoch 5, batch 14450, giga_loss[loss=0.2891, simple_loss=0.3596, pruned_loss=0.1093, over 28792.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3569, pruned_loss=0.1031, over 5669614.95 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3662, pruned_loss=0.1135, over 5748505.95 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3567, pruned_loss=0.1027, over 5657510.38 frames. ], batch size: 263, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:07:10,543 INFO [optim.py:369] (1/2) 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 16:07:33,119 INFO [train.py:968] (1/2) Epoch 5, batch 14500, giga_loss[loss=0.2977, simple_loss=0.3715, pruned_loss=0.1119, over 28457.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.356, pruned_loss=0.1038, over 5668004.10 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3658, pruned_loss=0.1134, over 5751354.95 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.356, pruned_loss=0.1034, over 5654077.33 frames. ], batch size: 336, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:08:17,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2987, 1.4793, 1.2009, 1.4404], device='cuda:1'), covar=tensor([0.0772, 0.0341, 0.0350, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0127, 0.0131, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 16:08:50,033 INFO [train.py:968] (1/2) Epoch 5, batch 14550, libri_loss[loss=0.2283, simple_loss=0.3011, pruned_loss=0.07777, over 29345.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3552, pruned_loss=0.1036, over 5678385.69 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3655, pruned_loss=0.1132, over 5752678.71 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3554, pruned_loss=0.1034, over 5665367.42 frames. ], batch size: 67, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:09:34,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4930, 4.2515, 1.5890, 1.6237], device='cuda:1'), covar=tensor([0.1108, 0.0301, 0.1033, 0.1550], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0482, 0.0322, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0017, 0.0021], device='cuda:1')
+2023-03-02 16:09:40,362 INFO [optim.py:369] (1/2) 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 16:10:07,129 INFO [train.py:968] (1/2) Epoch 5, batch 14600, giga_loss[loss=0.2858, simple_loss=0.3555, pruned_loss=0.1081, over 27531.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3502, pruned_loss=0.1007, over 5669295.16 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3652, pruned_loss=0.1131, over 5754350.94 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3506, pruned_loss=0.1005, over 5656922.49 frames. ], batch size: 472, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:11:07,200 INFO [train.py:968] (1/2) Epoch 5, batch 14650, giga_loss[loss=0.3052, simple_loss=0.3615, pruned_loss=0.1245, over 28001.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3484, pruned_loss=0.09997, over 5678219.64 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3635, pruned_loss=0.1121, over 5761096.33 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3496, pruned_loss=0.1002, over 5658610.94 frames. ], batch size: 412, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:11:43,474 INFO [optim.py:369] (1/2) 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 16:12:03,376 INFO [train.py:968] (1/2) Epoch 5, batch 14700, giga_loss[loss=0.2604, simple_loss=0.3346, pruned_loss=0.09312, over 28742.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3481, pruned_loss=0.1003, over 5682246.83 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3625, pruned_loss=0.1114, over 5756884.08 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 5666977.47 frames. ], batch size: 99, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:12:59,765 INFO [train.py:968] (1/2) Epoch 5, batch 14750, giga_loss[loss=0.2995, simple_loss=0.3807, pruned_loss=0.1092, over 28577.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3551, pruned_loss=0.1044, over 5685126.83 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3628, pruned_loss=0.1116, over 5755374.11 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3556, pruned_loss=0.1043, over 5671805.99 frames. ], batch size: 307, lr: 6.47e-03, grad_scale: 4.0
+2023-03-02 16:13:15,493 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9974, 1.3463, 1.0736, 0.2281], device='cuda:1'), covar=tensor([0.1501, 0.1272, 0.2249, 0.2200], device='cuda:1'), in_proj_covar=tensor([0.1356, 0.1275, 0.1368, 0.1138], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 16:13:38,729 INFO [optim.py:369] (1/2) 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 16:13:56,581 INFO [train.py:968] (1/2) Epoch 5, batch 14800, giga_loss[loss=0.3093, simple_loss=0.3726, pruned_loss=0.123, over 28796.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3543, pruned_loss=0.1052, over 5687270.23 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3618, pruned_loss=0.1112, over 5759667.31 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3554, pruned_loss=0.1052, over 5670678.64 frames. ], batch size: 99, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:14:13,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=196714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:14:55,705 INFO [train.py:968] (1/2) Epoch 5, batch 14850, giga_loss[loss=0.272, simple_loss=0.3514, pruned_loss=0.09629, over 28756.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.355, pruned_loss=0.1066, over 5674798.25 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3621, pruned_loss=0.1115, over 5754114.21 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3554, pruned_loss=0.1062, over 5664141.74 frames. ], batch size: 262, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:15:01,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=196753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:15:14,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 16:15:15,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3208, 1.9075, 1.3910, 1.5806], device='cuda:1'), covar=tensor([0.0763, 0.0346, 0.0337, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0125, 0.0132, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 16:15:35,130 INFO [optim.py:369] (1/2) 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 16:15:54,384 INFO [train.py:968] (1/2) Epoch 5, batch 14900, giga_loss[loss=0.2814, simple_loss=0.3544, pruned_loss=0.1042, over 28938.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3543, pruned_loss=0.1065, over 5675974.96 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3613, pruned_loss=0.1111, over 5756049.29 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3552, pruned_loss=0.1064, over 5663989.88 frames. ], batch size: 213, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:17:02,991 INFO [train.py:968] (1/2) Epoch 5, batch 14950, giga_loss[loss=0.2815, simple_loss=0.3513, pruned_loss=0.1059, over 27531.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.356, pruned_loss=0.1064, over 5676747.91 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3616, pruned_loss=0.1112, over 5757307.30 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3564, pruned_loss=0.1062, over 5665489.82 frames. ], batch size: 472, lr: 6.47e-03, grad_scale: 8.0
+2023-03-02 16:17:55,883 INFO [optim.py:369] (1/2) 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 16:18:17,907 INFO [train.py:968] (1/2) Epoch 5, batch 15000, giga_loss[loss=0.3232, simple_loss=0.3703, pruned_loss=0.138, over 26933.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3564, pruned_loss=0.1059, over 5667901.25 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3614, pruned_loss=0.111, over 5750848.88 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3569, pruned_loss=0.1059, over 5663561.65 frames. ], batch size: 555, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:18:17,907 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 16:18:26,278 INFO [train.py:1012] (1/2) Epoch 5, validation: loss=0.2221, simple_loss=0.3187, pruned_loss=0.06272, over 944034.00 frames. 
+2023-03-02 16:18:26,279 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 16:19:37,505 INFO [train.py:968] (1/2) Epoch 5, batch 15050, giga_loss[loss=0.2488, simple_loss=0.3193, pruned_loss=0.08909, over 28935.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.353, pruned_loss=0.1038, over 5677033.05 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3612, pruned_loss=0.1109, over 5753930.92 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3533, pruned_loss=0.1037, over 5668752.66 frames. ], batch size: 106, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:20:09,662 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=196971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:20:17,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6513, 1.6687, 1.4836, 1.8053], device='cuda:1'), covar=tensor([0.2120, 0.2002, 0.1881, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1115, 0.0856, 0.0987, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 16:20:24,149 INFO [optim.py:369] (1/2) 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 16:20:47,659 INFO [train.py:968] (1/2) Epoch 5, batch 15100, giga_loss[loss=0.2661, simple_loss=0.3196, pruned_loss=0.1063, over 26957.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.347, pruned_loss=0.1014, over 5685558.30 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3613, pruned_loss=0.111, over 5754691.27 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3471, pruned_loss=0.1012, over 5678169.91 frames. ], batch size: 555, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:21:17,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9079, 1.0978, 3.7555, 3.0863], device='cuda:1'), covar=tensor([0.1625, 0.2202, 0.0379, 0.0581], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0526, 0.0730, 0.0582], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 16:21:28,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5394, 1.6455, 1.5031, 1.5779], device='cuda:1'), covar=tensor([0.0989, 0.1737, 0.1552, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0734, 0.0627, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 16:21:40,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6339, 2.1991, 1.3711, 1.3064], device='cuda:1'), covar=tensor([0.1231, 0.0747, 0.0882, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.1409, 0.1126, 0.1152, 0.1224], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 16:21:52,063 INFO [train.py:968] (1/2) Epoch 5, batch 15150, giga_loss[loss=0.2977, simple_loss=0.3678, pruned_loss=0.1138, over 28422.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3435, pruned_loss=0.09986, over 5685130.78 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3612, pruned_loss=0.1109, over 5757601.00 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3433, pruned_loss=0.09955, over 5675284.31 frames. ], batch size: 369, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:22:34,952 INFO [optim.py:369] (1/2) 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 16:22:40,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:22:53,435 INFO [train.py:968] (1/2) Epoch 5, batch 15200, giga_loss[loss=0.3184, simple_loss=0.3865, pruned_loss=0.1251, over 28994.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.346, pruned_loss=0.1021, over 5679379.77 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3612, pruned_loss=0.1109, over 5757601.00 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3459, pruned_loss=0.1019, over 5671716.14 frames. ], batch size: 155, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:23:06,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-02 16:23:21,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:23:34,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1281, 1.9702, 1.5002, 1.6081], device='cuda:1'), covar=tensor([0.0555, 0.0515, 0.0857, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0439, 0.0498, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 16:23:44,594 INFO [train.py:968] (1/2) Epoch 5, batch 15250, giga_loss[loss=0.2933, simple_loss=0.3616, pruned_loss=0.1125, over 28667.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3464, pruned_loss=0.1021, over 5680675.46 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3607, pruned_loss=0.1105, over 5764782.02 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.346, pruned_loss=0.1019, over 5664388.34 frames. ], batch size: 92, lr: 6.46e-03, grad_scale: 8.0
+2023-03-02 16:24:03,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1344, 2.1187, 1.2237, 1.2907], device='cuda:1'), covar=tensor([0.0951, 0.0507, 0.0817, 0.1331], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0472, 0.0312, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 16:24:16,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:24:28,608 INFO [optim.py:369] (1/2) 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 16:24:47,799 INFO [train.py:968] (1/2) Epoch 5, batch 15300, giga_loss[loss=0.2423, simple_loss=0.3308, pruned_loss=0.07692, over 28681.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3434, pruned_loss=0.09951, over 5676975.29 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3604, pruned_loss=0.1103, over 5767264.19 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.343, pruned_loss=0.09924, over 5659214.96 frames. ], batch size: 242, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:25:25,741 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:25:28,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:25:43,220 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:25:48,125 INFO [train.py:968] (1/2) Epoch 5, batch 15350, giga_loss[loss=0.2546, simple_loss=0.3308, pruned_loss=0.08917, over 29091.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3419, pruned_loss=0.0981, over 5673778.74 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3603, pruned_loss=0.1103, over 5769642.76 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3413, pruned_loss=0.09759, over 5655292.16 frames. ], batch size: 113, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:25:54,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6650, 1.6902, 1.6378, 1.6070], device='cuda:1'), covar=tensor([0.0954, 0.1652, 0.1444, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0722, 0.0625, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 16:26:05,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:14,351 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:20,310 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:20,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:26:36,019 INFO [optim.py:369] (1/2) 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 16:26:53,972 INFO [train.py:968] (1/2) Epoch 5, batch 15400, libri_loss[loss=0.3129, simple_loss=0.3764, pruned_loss=0.1247, over 29294.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3414, pruned_loss=0.09808, over 5681139.09 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3598, pruned_loss=0.1102, over 5772043.81 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3407, pruned_loss=0.09746, over 5661530.88 frames. ], batch size: 94, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:26:58,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:27:50,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7355, 1.6385, 1.2852, 1.3028], device='cuda:1'), covar=tensor([0.0633, 0.0556, 0.0941, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0440, 0.0496, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 16:27:54,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:27:59,761 INFO [train.py:968] (1/2) Epoch 5, batch 15450, giga_loss[loss=0.2671, simple_loss=0.3459, pruned_loss=0.09412, over 28651.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3429, pruned_loss=0.09834, over 5687787.94 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.36, pruned_loss=0.1105, over 5764907.63 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.342, pruned_loss=0.09742, over 5678360.40 frames. ], batch size: 307, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:28:44,704 INFO [optim.py:369] (1/2) 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 16:29:02,049 INFO [train.py:968] (1/2) Epoch 5, batch 15500, giga_loss[loss=0.2831, simple_loss=0.352, pruned_loss=0.1071, over 28792.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3434, pruned_loss=0.09916, over 5692182.15 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3595, pruned_loss=0.1103, over 5765777.97 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3428, pruned_loss=0.09836, over 5682498.44 frames. ], batch size: 243, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:29:12,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3309, 1.8452, 1.3413, 0.5697], device='cuda:1'), covar=tensor([0.2377, 0.1259, 0.1801, 0.2451], device='cuda:1'), in_proj_covar=tensor([0.1369, 0.1298, 0.1372, 0.1140], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 16:29:29,413 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 16:30:09,552 INFO [train.py:968] (1/2) Epoch 5, batch 15550, giga_loss[loss=0.2754, simple_loss=0.3589, pruned_loss=0.09594, over 28469.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3428, pruned_loss=0.09917, over 5690719.06 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3592, pruned_loss=0.1102, over 5767747.13 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3424, pruned_loss=0.09849, over 5680179.30 frames. ], batch size: 336, lr: 6.46e-03, grad_scale: 4.0
+2023-03-02 16:30:55,702 INFO [optim.py:369] (1/2) 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 16:31:02,743 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:31:06,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:31:14,735 INFO [train.py:968] (1/2) Epoch 5, batch 15600, libri_loss[loss=0.319, simple_loss=0.3816, pruned_loss=0.1282, over 29122.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3422, pruned_loss=0.09769, over 5677406.78 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3594, pruned_loss=0.1103, over 5767827.68 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3415, pruned_loss=0.09689, over 5668177.42 frames. ], batch size: 101, lr: 6.45e-03, grad_scale: 8.0
+2023-03-02 16:31:43,160 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:31:58,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9047, 2.5140, 1.3765, 1.4902], device='cuda:1'), covar=tensor([0.1161, 0.0653, 0.0848, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.1421, 0.1128, 0.1158, 0.1234], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 16:31:59,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2146, 1.5333, 4.0447, 3.2990], device='cuda:1'), covar=tensor([0.1509, 0.2056, 0.0361, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0565, 0.0527, 0.0732, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 16:32:03,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-02 16:32:10,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-02 16:32:15,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:32:20,947 INFO [train.py:968] (1/2) Epoch 5, batch 15650, giga_loss[loss=0.2774, simple_loss=0.348, pruned_loss=0.1034, over 26967.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09853, over 5665892.36 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3594, pruned_loss=0.1103, over 5766162.80 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3447, pruned_loss=0.09776, over 5658923.46 frames. ], batch size: 555, lr: 6.45e-03, grad_scale: 8.0
+2023-03-02 16:33:07,352 INFO [optim.py:369] (1/2) 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 16:33:07,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9045, 1.1094, 3.7252, 2.9575], device='cuda:1'), covar=tensor([0.1654, 0.2311, 0.0387, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0565, 0.0526, 0.0730, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 16:33:23,891 INFO [train.py:968] (1/2) Epoch 5, batch 15700, giga_loss[loss=0.2531, simple_loss=0.3348, pruned_loss=0.08565, over 28693.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3483, pruned_loss=0.1003, over 5664323.17 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3593, pruned_loss=0.1103, over 5766887.78 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3477, pruned_loss=0.09957, over 5656659.74 frames. ], batch size: 307, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:33:54,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:34:24,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=197649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:34:24,820 INFO [train.py:968] (1/2) Epoch 5, batch 15750, giga_loss[loss=0.2652, simple_loss=0.3441, pruned_loss=0.09314, over 28927.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3497, pruned_loss=0.1013, over 5660309.28 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3593, pruned_loss=0.1102, over 5766671.12 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3489, pruned_loss=0.1005, over 5651885.18 frames. ], batch size: 213, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:35:10,736 INFO [optim.py:369] (1/2) 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 16:35:16,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:35:20,205 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 15800, libri_loss[loss=0.2668, simple_loss=0.3416, pruned_loss=0.09603, over 29377.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3483, pruned_loss=0.1007, over 5649505.20 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3591, pruned_loss=0.1102, over 5759487.97 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3476, pruned_loss=0.09996, over 5647743.90 frames. ], batch size: 92, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:35:57,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:36:30,092 INFO [train.py:968] (1/2) Epoch 5, batch 15850, giga_loss[loss=0.2505, simple_loss=0.3358, pruned_loss=0.08254, over 28959.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3458, pruned_loss=0.09947, over 5645676.69 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3591, pruned_loss=0.1104, over 5754248.81 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3448, pruned_loss=0.09831, over 5645601.92 frames. ], batch size: 136, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:36:48,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:36:52,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:08,531 INFO [optim.py:369] (1/2) 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 16:37:18,941 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=197792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:21,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=197795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:23,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:37:26,497 INFO [train.py:968] (1/2) Epoch 5, batch 15900, giga_loss[loss=0.26, simple_loss=0.3321, pruned_loss=0.09392, over 27679.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.345, pruned_loss=0.09961, over 5659709.27 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3587, pruned_loss=0.1101, over 5755719.62 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3441, pruned_loss=0.09859, over 5654932.41 frames. ], batch size: 472, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:37:38,344 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197808.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 16:37:57,032 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=197824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:38:12,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1807, 2.0725, 1.5982, 1.8869], device='cuda:1'), covar=tensor([0.0625, 0.0563, 0.0815, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0441, 0.0499, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 16:38:31,580 INFO [train.py:968] (1/2) Epoch 5, batch 15950, giga_loss[loss=0.3108, simple_loss=0.3832, pruned_loss=0.1192, over 28755.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3456, pruned_loss=0.1001, over 5665701.29 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3586, pruned_loss=0.11, over 5757507.96 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3449, pruned_loss=0.09928, over 5659568.50 frames. ], batch size: 243, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:39:04,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4862, 2.0966, 1.8331, 1.7051], device='cuda:1'), covar=tensor([0.1538, 0.1772, 0.1188, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0717, 0.0764, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 16:39:13,666 INFO [optim.py:369] (1/2) 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 16:39:14,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9001, 1.1833, 3.3733, 2.9008], device='cuda:1'), covar=tensor([0.1636, 0.2197, 0.0458, 0.0685], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0527, 0.0734, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 16:39:29,467 INFO [train.py:968] (1/2) Epoch 5, batch 16000, giga_loss[loss=0.3103, simple_loss=0.3779, pruned_loss=0.1213, over 28656.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.348, pruned_loss=0.1011, over 5671863.43 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3586, pruned_loss=0.1099, over 5757090.03 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3471, pruned_loss=0.1003, over 5664915.90 frames. ], batch size: 307, lr: 6.45e-03, grad_scale: 8.0
+2023-03-02 16:39:45,167 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:39:52,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6424, 1.7523, 1.6377, 1.6501], device='cuda:1'), covar=tensor([0.1137, 0.1650, 0.1504, 0.1470], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0728, 0.0626, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 16:39:56,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 16:40:05,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0705, 1.1988, 0.9706, 0.7707], device='cuda:1'), covar=tensor([0.0772, 0.0732, 0.0552, 0.0739], device='cuda:1'), in_proj_covar=tensor([0.1432, 0.1135, 0.1159, 0.1237], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 16:40:23,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=197938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:40:37,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3125, 1.6698, 1.1880, 1.0765], device='cuda:1'), covar=tensor([0.1302, 0.0785, 0.0749, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1134, 0.1161, 0.1236], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 16:40:39,617 INFO [train.py:968] (1/2) Epoch 5, batch 16050, giga_loss[loss=0.2461, simple_loss=0.3286, pruned_loss=0.08182, over 28873.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3483, pruned_loss=0.1021, over 5657012.91 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3578, pruned_loss=0.1096, over 5751555.47 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.348, pruned_loss=0.1015, over 5653954.67 frames. ], batch size: 174, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:40:52,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-02 16:41:21,914 INFO [optim.py:369] (1/2) 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 16:41:36,937 INFO [train.py:968] (1/2) Epoch 5, batch 16100, giga_loss[loss=0.3414, simple_loss=0.4066, pruned_loss=0.1381, over 28674.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3511, pruned_loss=0.1038, over 5662683.83 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3577, pruned_loss=0.1095, over 5756192.98 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3507, pruned_loss=0.1032, over 5653631.76 frames. ], batch size: 307, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:41:48,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4829, 1.6737, 1.2562, 1.0858], device='cuda:1'), covar=tensor([0.1006, 0.0739, 0.0625, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.1424, 0.1123, 0.1156, 0.1224], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 16:42:39,972 INFO [train.py:968] (1/2) Epoch 5, batch 16150, giga_loss[loss=0.3119, simple_loss=0.388, pruned_loss=0.1179, over 28034.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3546, pruned_loss=0.1058, over 5651681.57 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3575, pruned_loss=0.1094, over 5756885.87 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3545, pruned_loss=0.1053, over 5642634.14 frames. ], batch size: 412, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:43:20,303 INFO [optim.py:369] (1/2) 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 16:43:36,822 INFO [train.py:968] (1/2) Epoch 5, batch 16200, giga_loss[loss=0.2561, simple_loss=0.3409, pruned_loss=0.08562, over 29056.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3547, pruned_loss=0.1052, over 5654799.72 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3567, pruned_loss=0.1089, over 5760245.35 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3553, pruned_loss=0.1051, over 5641196.25 frames. ], batch size: 136, lr: 6.45e-03, grad_scale: 4.0
+2023-03-02 16:44:04,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2155, 1.6844, 1.5330, 1.4634], device='cuda:1'), covar=tensor([0.1076, 0.1519, 0.0935, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0721, 0.0765, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 16:44:46,488 INFO [train.py:968] (1/2) Epoch 5, batch 16250, giga_loss[loss=0.2615, simple_loss=0.3409, pruned_loss=0.09102, over 28865.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3535, pruned_loss=0.1045, over 5655883.79 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3564, pruned_loss=0.1086, over 5760158.01 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3542, pruned_loss=0.1047, over 5642905.42 frames. ], batch size: 263, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:44:50,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6196, 1.6274, 1.5899, 1.5448], device='cuda:1'), covar=tensor([0.1034, 0.1890, 0.1614, 0.1581], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0732, 0.0627, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 16:45:23,650 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:45:42,207 INFO [train.py:968] (1/2) Epoch 5, batch 16300, giga_loss[loss=0.2763, simple_loss=0.3502, pruned_loss=0.1012, over 28948.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3506, pruned_loss=0.1031, over 5665154.20 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3555, pruned_loss=0.1082, over 5757190.64 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3518, pruned_loss=0.1034, over 5653330.91 frames. ], batch size: 186, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:46:18,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2845, 1.2391, 1.1159, 1.4867], device='cuda:1'), covar=tensor([0.0757, 0.0385, 0.0359, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0126, 0.0132, 0.0235], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 16:46:48,515 INFO [train.py:968] (1/2) Epoch 5, batch 16350, libri_loss[loss=0.2751, simple_loss=0.35, pruned_loss=0.1001, over 29544.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3498, pruned_loss=0.1024, over 5671888.46 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.355, pruned_loss=0.1079, over 5760062.29 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3511, pruned_loss=0.1028, over 5658371.20 frames. ], batch size: 82, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:47:13,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 16:47:30,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:47:30,843 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:47:33,963 INFO [optim.py:369] (1/2) 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 16:47:50,972 INFO [train.py:968] (1/2) Epoch 5, batch 16400, giga_loss[loss=0.2565, simple_loss=0.3266, pruned_loss=0.09319, over 28982.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3498, pruned_loss=0.1033, over 5668074.71 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3549, pruned_loss=0.1078, over 5761820.01 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3508, pruned_loss=0.1036, over 5654276.59 frames. ], batch size: 120, lr: 6.44e-03, grad_scale: 8.0
+2023-03-02 16:48:08,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:48:23,827 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 16:48:26,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198329.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 16:48:52,242 INFO [train.py:968] (1/2) Epoch 5, batch 16450, libri_loss[loss=0.282, simple_loss=0.361, pruned_loss=0.1015, over 29668.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3473, pruned_loss=0.1022, over 5660038.23 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3548, pruned_loss=0.1077, over 5757144.59 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3479, pruned_loss=0.1024, over 5650905.49 frames. ], batch size: 88, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:49:01,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198358.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 16:49:37,458 INFO [optim.py:369] (1/2) 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 16:49:53,131 INFO [train.py:968] (1/2) Epoch 5, batch 16500, giga_loss[loss=0.3007, simple_loss=0.3718, pruned_loss=0.1148, over 28794.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3467, pruned_loss=0.1013, over 5653922.38 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3544, pruned_loss=0.1075, over 5750173.73 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3474, pruned_loss=0.1015, over 5650901.62 frames. ], batch size: 243, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:49:56,504 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-02 16:50:25,254 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:50:28,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:50:53,155 INFO [train.py:968] (1/2) Epoch 5, batch 16550, giga_loss[loss=0.2603, simple_loss=0.332, pruned_loss=0.09433, over 27604.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.345, pruned_loss=0.09875, over 5661575.27 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3547, pruned_loss=0.1077, over 5748739.08 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3453, pruned_loss=0.09867, over 5659159.31 frames. ], batch size: 472, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:51:01,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:04,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:05,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:11,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9247, 1.7561, 1.4234, 1.5564], device='cuda:1'), covar=tensor([0.0673, 0.0603, 0.0936, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0440, 0.0503, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 16:51:20,115 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:34,458 INFO [optim.py:369] (1/2) 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 16:51:35,123 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:51:41,293 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 16:51:49,340 INFO [train.py:968] (1/2) Epoch 5, batch 16600, giga_loss[loss=0.3073, simple_loss=0.3902, pruned_loss=0.1122, over 29011.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.346, pruned_loss=0.09681, over 5679573.54 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3541, pruned_loss=0.1074, over 5752028.06 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3465, pruned_loss=0.09676, over 5672980.60 frames. ], batch size: 285, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:51:58,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1074, 1.9072, 1.5755, 1.8219], device='cuda:1'), covar=tensor([0.0640, 0.0619, 0.0841, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0436, 0.0498, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 16:52:46,157 INFO [train.py:968] (1/2) Epoch 5, batch 16650, giga_loss[loss=0.2751, simple_loss=0.3575, pruned_loss=0.09635, over 28964.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3479, pruned_loss=0.09722, over 5675905.17 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3543, pruned_loss=0.1074, over 5751373.33 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3479, pruned_loss=0.09692, over 5669465.53 frames. ], batch size: 285, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:53:05,891 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-02 16:53:32,450 INFO [optim.py:369] (1/2) 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 16:53:48,484 INFO [train.py:968] (1/2) Epoch 5, batch 16700, giga_loss[loss=0.2551, simple_loss=0.3297, pruned_loss=0.09028, over 27586.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.348, pruned_loss=0.09761, over 5660492.27 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3545, pruned_loss=0.1076, over 5743920.01 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3477, pruned_loss=0.09711, over 5660168.65 frames. ], batch size: 472, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:54:05,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5658, 1.6233, 1.6006, 1.5717], device='cuda:1'), covar=tensor([0.1510, 0.1904, 0.1278, 0.1198], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0715, 0.0767, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 16:54:56,085 INFO [train.py:968] (1/2) Epoch 5, batch 16750, giga_loss[loss=0.264, simple_loss=0.3465, pruned_loss=0.09074, over 28365.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3481, pruned_loss=0.09809, over 5659165.13 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3544, pruned_loss=0.1076, over 5745371.91 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3479, pruned_loss=0.09744, over 5655829.11 frames. ], batch size: 368, lr: 6.44e-03, grad_scale: 4.0
+2023-03-02 16:55:11,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:55:54,324 INFO [optim.py:369] (1/2) 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 16:56:09,047 INFO [train.py:968] (1/2) Epoch 5, batch 16800, giga_loss[loss=0.2628, simple_loss=0.3386, pruned_loss=0.09347, over 28939.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09741, over 5657456.42 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3543, pruned_loss=0.1076, over 5746720.27 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3473, pruned_loss=0.09678, over 5652163.91 frames. ], batch size: 186, lr: 6.44e-03, grad_scale: 8.0
+2023-03-02 16:56:19,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4289, 1.4402, 1.1194, 1.1772], device='cuda:1'), covar=tensor([0.0608, 0.0433, 0.0866, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0439, 0.0503, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 16:57:09,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 16:57:19,045 INFO [train.py:968] (1/2) Epoch 5, batch 16850, giga_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1132, over 28473.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3472, pruned_loss=0.09656, over 5659754.78 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3541, pruned_loss=0.1075, over 5749150.43 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3472, pruned_loss=0.09598, over 5651903.22 frames. ], batch size: 336, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 16:57:22,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4062, 1.4304, 1.2379, 1.7532], device='cuda:1'), covar=tensor([0.1956, 0.1868, 0.1919, 0.1858], device='cuda:1'), in_proj_covar=tensor([0.1106, 0.0851, 0.0979, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 16:58:11,921 INFO [optim.py:369] (1/2) 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 16:58:28,139 INFO [train.py:968] (1/2) Epoch 5, batch 16900, giga_loss[loss=0.287, simple_loss=0.3818, pruned_loss=0.09611, over 28719.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3519, pruned_loss=0.0996, over 5659232.76 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3545, pruned_loss=0.1079, over 5745478.73 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3513, pruned_loss=0.09842, over 5652982.52 frames. ], batch size: 262, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 16:58:31,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:58:33,559 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:59:16,613 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=198835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:59:32,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=198849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 16:59:33,003 INFO [train.py:968] (1/2) Epoch 5, batch 16950, giga_loss[loss=0.272, simple_loss=0.3385, pruned_loss=0.1028, over 28602.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3524, pruned_loss=0.09956, over 5657601.67 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3539, pruned_loss=0.1076, over 5736692.39 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3524, pruned_loss=0.09868, over 5657578.28 frames. ], batch size: 85, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 16:59:44,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2937, 1.7932, 1.3275, 0.5612], device='cuda:1'), covar=tensor([0.1507, 0.1134, 0.2035, 0.2168], device='cuda:1'), in_proj_covar=tensor([0.1367, 0.1289, 0.1357, 0.1133], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 16:59:58,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8593, 1.1530, 3.6896, 3.0828], device='cuda:1'), covar=tensor([0.2236, 0.2636, 0.0695, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0522, 0.0721, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 17:00:24,750 INFO [optim.py:369] (1/2) 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 17:00:29,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=198892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:00:43,456 INFO [train.py:968] (1/2) Epoch 5, batch 17000, giga_loss[loss=0.253, simple_loss=0.3184, pruned_loss=0.09386, over 24794.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3509, pruned_loss=0.09944, over 5658561.03 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3534, pruned_loss=0.1072, over 5731203.12 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3513, pruned_loss=0.09887, over 5661016.01 frames. ], batch size: 705, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:01:49,428 INFO [train.py:968] (1/2) Epoch 5, batch 17050, giga_loss[loss=0.2771, simple_loss=0.3549, pruned_loss=0.09962, over 28782.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.09912, over 5673123.80 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3534, pruned_loss=0.1071, over 5736883.33 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3495, pruned_loss=0.0985, over 5667741.98 frames. ], batch size: 243, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:02:10,162 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3883, 1.9970, 1.3779, 0.7827], device='cuda:1'), covar=tensor([0.2539, 0.1399, 0.2210, 0.2561], device='cuda:1'), in_proj_covar=tensor([0.1364, 0.1281, 0.1358, 0.1128], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 17:02:43,726 INFO [optim.py:369] (1/2) 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 17:02:50,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=198992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:02:54,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=198995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:03:01,508 INFO [train.py:968] (1/2) Epoch 5, batch 17100, giga_loss[loss=0.2819, simple_loss=0.3398, pruned_loss=0.112, over 24551.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3469, pruned_loss=0.09679, over 5671492.54 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.353, pruned_loss=0.1068, over 5740552.50 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3473, pruned_loss=0.09636, over 5662166.18 frames. ], batch size: 705, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:03:11,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:03:34,653 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:04:04,022 INFO [train.py:968] (1/2) Epoch 5, batch 17150, giga_loss[loss=0.3165, simple_loss=0.373, pruned_loss=0.13, over 26856.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3466, pruned_loss=0.09676, over 5668621.07 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3528, pruned_loss=0.1067, over 5736509.15 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.347, pruned_loss=0.09633, over 5663877.73 frames. ], batch size: 555, lr: 6.43e-03, grad_scale: 4.0
+2023-03-02 17:04:12,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0666, 1.1174, 4.2738, 3.1573], device='cuda:1'), covar=tensor([0.1683, 0.2305, 0.0308, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0523, 0.0734, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 17:04:51,759 INFO [optim.py:369] (1/2) 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 17:05:05,388 INFO [train.py:968] (1/2) Epoch 5, batch 17200, giga_loss[loss=0.3049, simple_loss=0.3762, pruned_loss=0.1169, over 28881.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.349, pruned_loss=0.09844, over 5668933.28 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3527, pruned_loss=0.1067, over 5739017.54 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3493, pruned_loss=0.09799, over 5662098.66 frames. ], batch size: 284, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 17:06:07,188 INFO [train.py:968] (1/2) Epoch 5, batch 17250, giga_loss[loss=0.2657, simple_loss=0.3255, pruned_loss=0.103, over 24596.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.352, pruned_loss=0.1003, over 5670726.50 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3529, pruned_loss=0.1067, over 5740970.67 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.352, pruned_loss=0.09988, over 5662394.22 frames. ], batch size: 705, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 17:06:30,664 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 17:06:48,782 INFO [optim.py:369] (1/2) 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 17:07:05,760 INFO [train.py:968] (1/2) Epoch 5, batch 17300, giga_loss[loss=0.2301, simple_loss=0.289, pruned_loss=0.08563, over 24386.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.35, pruned_loss=0.1008, over 5659226.52 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.353, pruned_loss=0.1067, over 5734525.45 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1002, over 5657171.26 frames. ], batch size: 705, lr: 6.43e-03, grad_scale: 8.0
+2023-03-02 17:08:02,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:08:06,644 INFO [train.py:968] (1/2) Epoch 5, batch 17350, giga_loss[loss=0.2568, simple_loss=0.3331, pruned_loss=0.09024, over 28924.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3488, pruned_loss=0.1008, over 5660001.91 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.353, pruned_loss=0.1067, over 5734949.82 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3487, pruned_loss=0.1004, over 5657596.37 frames. ], batch size: 227, lr: 6.43e-03, grad_scale: 2.0
+2023-03-02 17:08:26,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:08:38,058 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:08:40,462 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.03 vs. limit=2.0
+2023-03-02 17:08:55,690 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 17400, giga_loss[loss=0.3211, simple_loss=0.3875, pruned_loss=0.1273, over 28968.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3522, pruned_loss=0.1041, over 5654034.51 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3529, pruned_loss=0.1067, over 5737485.76 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3522, pruned_loss=0.1037, over 5648394.06 frames. ], batch size: 186, lr: 6.43e-03, grad_scale: 2.0
+2023-03-02 17:09:32,072 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:09:40,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 17:09:51,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1710, 1.6098, 1.4370, 1.3661], device='cuda:1'), covar=tensor([0.1091, 0.1407, 0.0895, 0.0975], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0712, 0.0761, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 17:09:59,665 INFO [train.py:968] (1/2) Epoch 5, batch 17450, libri_loss[loss=0.2417, simple_loss=0.3186, pruned_loss=0.08244, over 29576.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3629, pruned_loss=0.1107, over 5664421.32 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3528, pruned_loss=0.1066, over 5738910.21 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.363, pruned_loss=0.1105, over 5657887.58 frames. ], batch size: 74, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:10:30,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:10:36,563 INFO [optim.py:369] (1/2) 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 17:10:45,080 INFO [train.py:968] (1/2) Epoch 5, batch 17500, giga_loss[loss=0.3485, simple_loss=0.4036, pruned_loss=0.1466, over 28955.00 frames. ], tot_loss[loss=0.3, simple_loss=0.37, pruned_loss=0.115, over 5661232.51 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.353, pruned_loss=0.1068, over 5730751.88 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.37, pruned_loss=0.1148, over 5662211.49 frames. ], batch size: 106, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:10:45,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6759, 1.5894, 1.1755, 1.3823], device='cuda:1'), covar=tensor([0.0663, 0.0570, 0.0969, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0441, 0.0502, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 17:10:53,963 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:10:59,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:11:11,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-02 17:11:13,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2170, 1.7438, 1.3892, 0.4908], device='cuda:1'), covar=tensor([0.1829, 0.1254, 0.1962, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1371, 0.1310, 0.1362, 0.1137], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 17:11:21,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:11:30,730 INFO [train.py:968] (1/2) Epoch 5, batch 17550, giga_loss[loss=0.2723, simple_loss=0.3377, pruned_loss=0.1034, over 28869.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3703, pruned_loss=0.1169, over 5666796.16 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3527, pruned_loss=0.1065, over 5735033.75 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3712, pruned_loss=0.1172, over 5661858.60 frames. ], batch size: 199, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:12:02,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 17:12:06,561 INFO [optim.py:369] (1/2) 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 17:12:12,723 INFO [train.py:968] (1/2) Epoch 5, batch 17600, giga_loss[loss=0.2903, simple_loss=0.3504, pruned_loss=0.1151, over 28627.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3633, pruned_loss=0.1135, over 5677920.51 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3533, pruned_loss=0.1068, over 5735925.79 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3636, pruned_loss=0.1136, over 5672361.94 frames. ], batch size: 307, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:12:37,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:12:40,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:12:59,906 INFO [train.py:968] (1/2) Epoch 5, batch 17650, giga_loss[loss=0.2342, simple_loss=0.3118, pruned_loss=0.07832, over 29010.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3557, pruned_loss=0.1101, over 5683064.83 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3534, pruned_loss=0.1069, over 5738622.39 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3559, pruned_loss=0.1102, over 5675505.59 frames. ], batch size: 155, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:13:05,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:13:13,443 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:13:34,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0920, 4.5632, 1.9849, 2.2053], device='cuda:1'), covar=tensor([0.0770, 0.0195, 0.0767, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0472, 0.0312, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 17:13:35,358 INFO [optim.py:369] (1/2) 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 17:13:43,652 INFO [train.py:968] (1/2) Epoch 5, batch 17700, giga_loss[loss=0.2502, simple_loss=0.3198, pruned_loss=0.09034, over 28828.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3477, pruned_loss=0.1064, over 5684084.95 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3538, pruned_loss=0.1072, over 5734331.36 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3474, pruned_loss=0.1061, over 5680486.57 frames. ], batch size: 199, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:14:00,972 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 17750, giga_loss[loss=0.2695, simple_loss=0.3287, pruned_loss=0.1052, over 28937.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3408, pruned_loss=0.1028, over 5686356.54 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3547, pruned_loss=0.1076, over 5728862.81 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3394, pruned_loss=0.1022, over 5686505.88 frames. ], batch size: 227, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:14:28,466 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:14:38,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4597, 1.4494, 1.5221, 1.4264], device='cuda:1'), covar=tensor([0.1055, 0.1597, 0.1334, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0744, 0.0630, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 17:14:46,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0716, 1.3100, 3.9385, 3.1265], device='cuda:1'), covar=tensor([0.1585, 0.2116, 0.0320, 0.0646], device='cuda:1'), in_proj_covar=tensor([0.0559, 0.0523, 0.0730, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 17:15:01,955 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:15:07,701 INFO [train.py:968] (1/2) Epoch 5, batch 17800, giga_loss[loss=0.2504, simple_loss=0.3148, pruned_loss=0.09299, over 28982.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3362, pruned_loss=0.1004, over 5681042.32 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3554, pruned_loss=0.1078, over 5723297.27 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3339, pruned_loss=0.09948, over 5684280.64 frames. ], batch size: 106, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:15:49,955 INFO [train.py:968] (1/2) Epoch 5, batch 17850, giga_loss[loss=0.2246, simple_loss=0.2963, pruned_loss=0.07646, over 28910.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.333, pruned_loss=0.09897, over 5673502.76 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3559, pruned_loss=0.1081, over 5708312.47 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3302, pruned_loss=0.09782, over 5689907.51 frames. ], batch size: 199, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:16:02,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:06,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:13,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-02 17:16:29,809 INFO [optim.py:369] (1/2) 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 17:16:30,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:33,816 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:34,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:35,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:16:37,675 INFO [train.py:968] (1/2) Epoch 5, batch 17900, giga_loss[loss=0.254, simple_loss=0.3219, pruned_loss=0.09307, over 28632.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3286, pruned_loss=0.09677, over 5675707.33 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3557, pruned_loss=0.1078, over 5709725.69 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3263, pruned_loss=0.09593, over 5686800.14 frames. ], batch size: 307, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:16:58,351 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:09,133 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=199838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:11,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=199841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:17,336 INFO [train.py:968] (1/2) Epoch 5, batch 17950, libri_loss[loss=0.2355, simple_loss=0.315, pruned_loss=0.078, over 29491.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3248, pruned_loss=0.09472, over 5678893.79 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3555, pruned_loss=0.1077, over 5715259.94 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3223, pruned_loss=0.09378, over 5681783.38 frames. ], batch size: 70, lr: 6.42e-03, grad_scale: 2.0
+2023-03-02 17:17:18,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:22,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5915, 1.9738, 1.8921, 1.7293], device='cuda:1'), covar=tensor([0.1566, 0.1767, 0.1192, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0734, 0.0780, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 17:17:34,709 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=199870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:17:45,869 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199887.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:17:50,924 INFO [optim.py:369] (1/2) 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 17:17:59,109 INFO [train.py:968] (1/2) Epoch 5, batch 18000, giga_loss[loss=0.2203, simple_loss=0.2974, pruned_loss=0.0716, over 28251.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3227, pruned_loss=0.09342, over 5686705.82 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3562, pruned_loss=0.1082, over 5711086.04 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.319, pruned_loss=0.09183, over 5692048.50 frames. ], batch size: 65, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:17:59,109 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 17:18:07,927 INFO [train.py:1012] (1/2) Epoch 5, validation: loss=0.234, simple_loss=0.3355, pruned_loss=0.06627, over 944034.00 frames. 
+2023-03-02 17:18:07,928 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 17:18:24,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4268, 1.6124, 1.2425, 1.8518], device='cuda:1'), covar=tensor([0.2202, 0.2103, 0.2130, 0.2296], device='cuda:1'), in_proj_covar=tensor([0.1113, 0.0859, 0.0988, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 17:18:33,246 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=199928.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:18:48,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=199944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:18:52,662 INFO [train.py:968] (1/2) Epoch 5, batch 18050, giga_loss[loss=0.2442, simple_loss=0.2968, pruned_loss=0.09586, over 23631.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.32, pruned_loss=0.0924, over 5676698.61 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3565, pruned_loss=0.1082, over 5710612.18 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3162, pruned_loss=0.09087, over 5680784.46 frames. ], batch size: 705, lr: 6.42e-03, grad_scale: 4.0
+2023-03-02 17:19:22,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4919, 3.1018, 1.4863, 1.5250], device='cuda:1'), covar=tensor([0.0835, 0.0309, 0.0838, 0.1267], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0472, 0.0311, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 17:19:30,968 INFO [optim.py:369] (1/2) 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 17:19:36,468 INFO [train.py:968] (1/2) Epoch 5, batch 18100, giga_loss[loss=0.2254, simple_loss=0.2936, pruned_loss=0.07861, over 28553.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3164, pruned_loss=0.0904, over 5684709.69 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3569, pruned_loss=0.1083, over 5713887.35 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3122, pruned_loss=0.08868, over 5684266.36 frames. ], batch size: 78, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:20:23,395 INFO [train.py:968] (1/2) Epoch 5, batch 18150, giga_loss[loss=0.2117, simple_loss=0.2829, pruned_loss=0.07025, over 29108.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3138, pruned_loss=0.08878, over 5693431.41 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1085, over 5712796.95 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3093, pruned_loss=0.08689, over 5693260.82 frames. ], batch size: 128, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:20:58,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:21:00,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:21:02,681 INFO [optim.py:369] (1/2) 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 17:21:11,639 INFO [train.py:968] (1/2) Epoch 5, batch 18200, giga_loss[loss=0.1999, simple_loss=0.2725, pruned_loss=0.06369, over 28959.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3106, pruned_loss=0.08718, over 5695881.95 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3573, pruned_loss=0.1084, over 5716966.29 frames. ], giga_tot_loss[loss=0.238, simple_loss=0.3056, pruned_loss=0.08517, over 5691494.88 frames. ], batch size: 145, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:21:28,433 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:21:34,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3784, 1.5575, 1.1123, 1.1295], device='cuda:1'), covar=tensor([0.1164, 0.0876, 0.0809, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.1400, 0.1126, 0.1165, 0.1245], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 17:21:54,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2543, 1.2080, 1.0986, 0.9695], device='cuda:1'), covar=tensor([0.0648, 0.0496, 0.1008, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0439, 0.0499, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 17:21:59,861 INFO [train.py:968] (1/2) Epoch 5, batch 18250, giga_loss[loss=0.2379, simple_loss=0.3094, pruned_loss=0.08319, over 29035.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3174, pruned_loss=0.09135, over 5698135.48 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3572, pruned_loss=0.1082, over 5718345.11 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3129, pruned_loss=0.08964, over 5693100.59 frames. ], batch size: 164, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:22:16,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:22:35,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5368, 1.4549, 1.1755, 1.1829], device='cuda:1'), covar=tensor([0.0558, 0.0460, 0.0871, 0.0787], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0437, 0.0497, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 17:22:39,946 INFO [optim.py:369] (1/2) 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 17:22:46,305 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-02 17:22:46,428 INFO [train.py:968] (1/2) Epoch 5, batch 18300, giga_loss[loss=0.3443, simple_loss=0.405, pruned_loss=0.1418, over 28633.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3321, pruned_loss=0.09954, over 5691603.33 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3572, pruned_loss=0.1082, over 5712686.73 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3277, pruned_loss=0.09784, over 5692907.29 frames. ], batch size: 242, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:23:07,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-02 17:23:09,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:23:28,144 INFO [train.py:968] (1/2) Epoch 5, batch 18350, giga_loss[loss=0.2977, simple_loss=0.3719, pruned_loss=0.1117, over 29026.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3454, pruned_loss=0.1073, over 5689485.86 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3578, pruned_loss=0.1086, over 5708597.38 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3409, pruned_loss=0.1055, over 5693122.91 frames. ], batch size: 155, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:23:39,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200262.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:24:05,387 INFO [optim.py:369] (1/2) 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 17:24:10,765 INFO [train.py:968] (1/2) Epoch 5, batch 18400, giga_loss[loss=0.286, simple_loss=0.3627, pruned_loss=0.1047, over 28764.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3554, pruned_loss=0.1122, over 5695023.42 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3583, pruned_loss=0.1088, over 5711459.91 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3511, pruned_loss=0.1106, over 5694851.81 frames. ], batch size: 119, lr: 6.41e-03, grad_scale: 8.0
+2023-03-02 17:24:13,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:14,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200303.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:24:21,434 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:23,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:47,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:24:53,985 INFO [train.py:968] (1/2) Epoch 5, batch 18450, giga_loss[loss=0.2965, simple_loss=0.3735, pruned_loss=0.1098, over 28536.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3595, pruned_loss=0.1129, over 5687915.04 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3588, pruned_loss=0.109, over 5707231.38 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3556, pruned_loss=0.1115, over 5690763.16 frames. ], batch size: 336, lr: 6.41e-03, grad_scale: 8.0
+2023-03-02 17:25:07,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-02 17:25:11,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:25:13,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:25:36,072 INFO [optim.py:369] (1/2) 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 17:25:42,485 INFO [train.py:968] (1/2) Epoch 5, batch 18500, giga_loss[loss=0.3233, simple_loss=0.3843, pruned_loss=0.1311, over 28566.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3607, pruned_loss=0.112, over 5684017.39 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3588, pruned_loss=0.109, over 5707231.38 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.111, over 5686234.12 frames. ], batch size: 71, lr: 6.41e-03, grad_scale: 8.0
+2023-03-02 17:25:44,104 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:25:48,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200405.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:25:51,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200408.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:26:09,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6649, 1.6558, 1.6303, 1.6707], device='cuda:1'), covar=tensor([0.0901, 0.1326, 0.1389, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0737, 0.0626, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 17:26:15,954 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:26:24,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200446.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:26:26,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200449.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:26:27,001 INFO [train.py:968] (1/2) Epoch 5, batch 18550, libri_loss[loss=0.3557, simple_loss=0.3887, pruned_loss=0.1613, over 28715.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3644, pruned_loss=0.1145, over 5679865.62 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.36, pruned_loss=0.1096, over 5704943.86 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3611, pruned_loss=0.1133, over 5682045.69 frames. ], batch size: 63, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:26:46,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-02 17:26:51,101 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200478.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:27:04,740 INFO [optim.py:369] (1/2) 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 17:27:11,695 INFO [train.py:968] (1/2) Epoch 5, batch 18600, giga_loss[loss=0.3884, simple_loss=0.4179, pruned_loss=0.1794, over 26496.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3675, pruned_loss=0.1172, over 5682222.04 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3602, pruned_loss=0.1098, over 5708147.54 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3648, pruned_loss=0.1162, over 5680865.74 frames. ], batch size: 555, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:27:31,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3734, 1.4745, 1.0195, 1.1037], device='cuda:1'), covar=tensor([0.0945, 0.0864, 0.0776, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.1396, 0.1137, 0.1165, 0.1239], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 17:27:56,244 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-02 17:27:57,574 INFO [train.py:968] (1/2) Epoch 5, batch 18650, giga_loss[loss=0.3643, simple_loss=0.421, pruned_loss=0.1538, over 28209.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3698, pruned_loss=0.1186, over 5695535.62 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3603, pruned_loss=0.1099, over 5712133.78 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3676, pruned_loss=0.1179, over 5690368.38 frames. ], batch size: 368, lr: 6.41e-03, grad_scale: 4.0
+2023-03-02 17:28:14,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200570.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:28:29,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4329, 1.9447, 1.2912, 1.1334], device='cuda:1'), covar=tensor([0.1319, 0.0820, 0.0910, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.1377, 0.1125, 0.1151, 0.1222], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 17:28:33,724 INFO [optim.py:369] (1/2) 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 17:28:40,413 INFO [train.py:968] (1/2) Epoch 5, batch 18700, giga_loss[loss=0.3347, simple_loss=0.397, pruned_loss=0.1362, over 28969.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3738, pruned_loss=0.121, over 5690621.56 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3606, pruned_loss=0.1099, over 5705644.13 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.372, pruned_loss=0.1205, over 5692547.53 frames. ], batch size: 164, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:29:19,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2834, 1.7694, 1.2984, 1.4143], device='cuda:1'), covar=tensor([0.0722, 0.0269, 0.0318, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0125, 0.0128, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 17:29:20,763 INFO [train.py:968] (1/2) Epoch 5, batch 18750, libri_loss[loss=0.27, simple_loss=0.3429, pruned_loss=0.09856, over 29587.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3749, pruned_loss=0.1204, over 5695669.52 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3608, pruned_loss=0.11, over 5702124.03 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3739, pruned_loss=0.1203, over 5700879.92 frames. ], batch size: 75, lr: 6.40e-03, grad_scale: 2.0
+2023-03-02 17:29:38,002 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=200672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:29:42,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:29:55,808 INFO [optim.py:369] (1/2) 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 17:30:00,518 INFO [train.py:968] (1/2) Epoch 5, batch 18800, giga_loss[loss=0.3549, simple_loss=0.399, pruned_loss=0.1554, over 26545.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3764, pruned_loss=0.1206, over 5691583.98 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3611, pruned_loss=0.1101, over 5698917.99 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3756, pruned_loss=0.1207, over 5698936.61 frames. ], batch size: 555, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:30:10,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3505, 1.3673, 1.1916, 1.5231], device='cuda:1'), covar=tensor([0.2157, 0.2081, 0.2098, 0.2006], device='cuda:1'), in_proj_covar=tensor([0.1107, 0.0859, 0.0983, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 17:30:31,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3268, 1.7341, 1.3372, 1.5195], device='cuda:1'), covar=tensor([0.0824, 0.0305, 0.0328, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0124, 0.0129, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 17:30:43,670 INFO [train.py:968] (1/2) Epoch 5, batch 18850, giga_loss[loss=0.2942, simple_loss=0.3696, pruned_loss=0.1094, over 28573.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3771, pruned_loss=0.1201, over 5691275.33 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3618, pruned_loss=0.1103, over 5700680.12 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3761, pruned_loss=0.1201, over 5695568.74 frames. ], batch size: 336, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:31:06,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3559, 3.4487, 1.4612, 1.3656], device='cuda:1'), covar=tensor([0.0869, 0.0231, 0.0816, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0468, 0.0306, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 17:31:09,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2169, 1.6387, 1.1895, 0.6544], device='cuda:1'), covar=tensor([0.2451, 0.1404, 0.1626, 0.2413], device='cuda:1'), in_proj_covar=tensor([0.1348, 0.1269, 0.1341, 0.1102], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 17:31:18,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7266, 1.5454, 1.1562, 1.2590], device='cuda:1'), covar=tensor([0.0660, 0.0635, 0.1037, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0438, 0.0504, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 17:31:18,841 INFO [optim.py:369] (1/2) 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 17:31:23,072 INFO [train.py:968] (1/2) Epoch 5, batch 18900, giga_loss[loss=0.3427, simple_loss=0.3974, pruned_loss=0.144, over 28298.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3755, pruned_loss=0.118, over 5695543.06 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3623, pruned_loss=0.1104, over 5707157.31 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3747, pruned_loss=0.1182, over 5692722.21 frames. ], batch size: 368, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:31:40,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=200820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:31:42,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=200823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:32:05,090 INFO [train.py:968] (1/2) Epoch 5, batch 18950, giga_loss[loss=0.2831, simple_loss=0.3591, pruned_loss=0.1036, over 28707.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3729, pruned_loss=0.1153, over 5710464.45 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3622, pruned_loss=0.1102, over 5713038.68 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3726, pruned_loss=0.1159, over 5703028.76 frames. ], batch size: 119, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:32:07,502 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=200852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:32:16,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4107, 2.9193, 1.5290, 1.3611], device='cuda:1'), covar=tensor([0.0782, 0.0297, 0.0703, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0466, 0.0304, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 17:32:42,109 INFO [optim.py:369] (1/2) 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 17:32:47,024 INFO [train.py:968] (1/2) Epoch 5, batch 19000, giga_loss[loss=0.2683, simple_loss=0.3495, pruned_loss=0.09349, over 28394.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3733, pruned_loss=0.1161, over 5705916.12 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3623, pruned_loss=0.1101, over 5717601.36 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3733, pruned_loss=0.1167, over 5695730.99 frames. ], batch size: 60, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:33:28,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=200945.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:33:31,366 INFO [train.py:968] (1/2) Epoch 5, batch 19050, giga_loss[loss=0.3254, simple_loss=0.3832, pruned_loss=0.1337, over 28758.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3767, pruned_loss=0.1214, over 5689830.60 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3629, pruned_loss=0.1104, over 5713016.58 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3765, pruned_loss=0.1219, over 5685502.52 frames. ], batch size: 284, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:34:10,372 INFO [optim.py:369] (1/2) 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 17:34:14,640 INFO [train.py:968] (1/2) Epoch 5, batch 19100, giga_loss[loss=0.3062, simple_loss=0.3821, pruned_loss=0.1151, over 28725.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3791, pruned_loss=0.1248, over 5693179.95 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3634, pruned_loss=0.1107, over 5716569.22 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3789, pruned_loss=0.1254, over 5685477.20 frames. ], batch size: 78, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:34:24,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-02 17:34:36,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7425, 2.8328, 1.9289, 0.8738], device='cuda:1'), covar=tensor([0.3068, 0.1290, 0.1828, 0.2886], device='cuda:1'), in_proj_covar=tensor([0.1372, 0.1290, 0.1358, 0.1130], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 17:34:46,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:34:55,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:34:57,061 INFO [train.py:968] (1/2) Epoch 5, batch 19150, libri_loss[loss=0.308, simple_loss=0.3866, pruned_loss=0.1147, over 29213.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3784, pruned_loss=0.1254, over 5696025.85 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3638, pruned_loss=0.1108, over 5719035.94 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3781, pruned_loss=0.1259, over 5687236.53 frames. ], batch size: 94, lr: 6.40e-03, grad_scale: 4.0
+2023-03-02 17:35:31,252 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201088.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 17:35:35,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201091.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:35:37,127 INFO [optim.py:369] (1/2) 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 17:35:41,588 INFO [train.py:968] (1/2) Epoch 5, batch 19200, giga_loss[loss=0.3269, simple_loss=0.3807, pruned_loss=0.1366, over 28945.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3761, pruned_loss=0.1243, over 5704906.45 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3638, pruned_loss=0.1106, over 5722679.42 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3762, pruned_loss=0.1252, over 5694190.98 frames. ], batch size: 213, lr: 6.40e-03, grad_scale: 8.0
+2023-03-02 17:35:57,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201120.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 17:36:05,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 17:36:21,503 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 17:36:26,235 INFO [train.py:968] (1/2) Epoch 5, batch 19250, giga_loss[loss=0.3111, simple_loss=0.377, pruned_loss=0.1226, over 28685.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3759, pruned_loss=0.1237, over 5679633.17 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3647, pruned_loss=0.111, over 5709004.14 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3755, pruned_loss=0.1244, over 5682524.36 frames. ], batch size: 307, lr: 6.40e-03, grad_scale: 8.0
+2023-03-02 17:36:58,755 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:37:00,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:37:01,015 INFO [optim.py:369] (1/2) 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 17:37:06,119 INFO [train.py:968] (1/2) Epoch 5, batch 19300, giga_loss[loss=0.2722, simple_loss=0.3454, pruned_loss=0.09948, over 28760.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3733, pruned_loss=0.1208, over 5689594.63 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.365, pruned_loss=0.1111, over 5712884.40 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.373, pruned_loss=0.1217, over 5687620.46 frames. ], batch size: 119, lr: 6.40e-03, grad_scale: 8.0
+2023-03-02 17:37:08,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1411, 1.6716, 1.2462, 0.3909], device='cuda:1'), covar=tensor([0.1971, 0.1190, 0.2134, 0.2587], device='cuda:1'), in_proj_covar=tensor([0.1375, 0.1298, 0.1367, 0.1139], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 17:37:28,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:37:54,053 INFO [train.py:968] (1/2) Epoch 5, batch 19350, giga_loss[loss=0.2523, simple_loss=0.3312, pruned_loss=0.08672, over 28881.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3692, pruned_loss=0.1175, over 5682348.91 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3653, pruned_loss=0.1111, over 5707149.11 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3689, pruned_loss=0.1184, over 5685299.52 frames. ], batch size: 227, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:38:31,415 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 19400, giga_loss[loss=0.2519, simple_loss=0.331, pruned_loss=0.08638, over 28955.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3633, pruned_loss=0.1141, over 5681518.70 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3654, pruned_loss=0.111, over 5710793.79 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.363, pruned_loss=0.115, over 5679523.25 frames. ], batch size: 227, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:39:25,009 INFO [train.py:968] (1/2) Epoch 5, batch 19450, giga_loss[loss=0.247, simple_loss=0.3188, pruned_loss=0.08756, over 28506.00 frames. ], tot_loss[loss=0.289, simple_loss=0.357, pruned_loss=0.1105, over 5684788.95 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3659, pruned_loss=0.1113, over 5712060.21 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3562, pruned_loss=0.1109, over 5681772.87 frames. ], batch size: 85, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:39:25,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-02 17:40:07,907 INFO [optim.py:369] (1/2) 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 17:40:13,858 INFO [train.py:968] (1/2) Epoch 5, batch 19500, giga_loss[loss=0.264, simple_loss=0.3365, pruned_loss=0.09575, over 28713.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3524, pruned_loss=0.1078, over 5681919.82 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3665, pruned_loss=0.1116, over 5708515.07 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3508, pruned_loss=0.1079, over 5682596.89 frames. ], batch size: 284, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:40:26,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:41:01,750 INFO [train.py:968] (1/2) Epoch 5, batch 19550, giga_loss[loss=0.3026, simple_loss=0.3647, pruned_loss=0.1203, over 28935.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3529, pruned_loss=0.1076, over 5679374.78 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3672, pruned_loss=0.112, over 5702794.45 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3509, pruned_loss=0.1072, over 5683982.12 frames. ], batch size: 136, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:41:13,180 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:41:41,870 INFO [optim.py:369] (1/2) 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 17:41:46,051 INFO [train.py:968] (1/2) Epoch 5, batch 19600, giga_loss[loss=0.268, simple_loss=0.3336, pruned_loss=0.1012, over 28637.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3533, pruned_loss=0.1078, over 5683896.60 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3675, pruned_loss=0.1121, over 5697339.55 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.351, pruned_loss=0.1073, over 5692607.61 frames. ], batch size: 60, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:41:49,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3638, 1.4367, 1.2000, 1.9617], device='cuda:1'), covar=tensor([0.2305, 0.2247, 0.2266, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.1114, 0.0863, 0.0982, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 17:41:55,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1651, 1.2757, 1.0836, 1.3591], device='cuda:1'), covar=tensor([0.0760, 0.0415, 0.0350, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0125, 0.0129, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 17:42:11,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2881, 1.1782, 1.1385, 1.0345], device='cuda:1'), covar=tensor([0.0646, 0.0543, 0.0974, 0.0771], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0445, 0.0507, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 17:42:30,499 INFO [train.py:968] (1/2) Epoch 5, batch 19650, giga_loss[loss=0.2461, simple_loss=0.3109, pruned_loss=0.09064, over 28671.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3516, pruned_loss=0.1075, over 5692365.45 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3676, pruned_loss=0.1121, over 5698901.76 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3496, pruned_loss=0.1071, over 5697664.15 frames. ], batch size: 92, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:42:36,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:42:40,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=201559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:43:01,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=201588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:43:05,053 INFO [optim.py:369] (1/2) 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 17:43:10,859 INFO [train.py:968] (1/2) Epoch 5, batch 19700, giga_loss[loss=0.3086, simple_loss=0.3686, pruned_loss=0.1243, over 28026.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3492, pruned_loss=0.1057, over 5707927.26 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3683, pruned_loss=0.1122, over 5704863.83 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3464, pruned_loss=0.1051, over 5706883.89 frames. ], batch size: 412, lr: 6.39e-03, grad_scale: 8.0
+2023-03-02 17:43:49,564 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:43:52,247 INFO [train.py:968] (1/2) Epoch 5, batch 19750, giga_loss[loss=0.275, simple_loss=0.334, pruned_loss=0.108, over 28735.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3462, pruned_loss=0.1042, over 5714038.40 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3684, pruned_loss=0.1121, over 5709706.41 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3434, pruned_loss=0.1036, over 5709019.24 frames. ], batch size: 99, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:44:05,648 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=201667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:44:25,947 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 17:44:28,764 INFO [optim.py:369] (1/2) 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 17:44:33,364 INFO [train.py:968] (1/2) Epoch 5, batch 19800, giga_loss[loss=0.2455, simple_loss=0.3196, pruned_loss=0.0857, over 28689.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3444, pruned_loss=0.1037, over 5709972.55 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3687, pruned_loss=0.1123, over 5702256.08 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3413, pruned_loss=0.1028, over 5712018.18 frames. ], batch size: 262, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:44:58,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7944, 1.0864, 3.4614, 2.8921], device='cuda:1'), covar=tensor([0.1761, 0.2393, 0.0407, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0525, 0.0734, 0.0603], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 17:45:11,732 INFO [train.py:968] (1/2) Epoch 5, batch 19850, giga_loss[loss=0.2663, simple_loss=0.3399, pruned_loss=0.09632, over 28755.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3439, pruned_loss=0.1034, over 5702152.93 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3705, pruned_loss=0.1133, over 5687620.47 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3388, pruned_loss=0.1015, over 5718048.56 frames. ], batch size: 284, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:45:17,547 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-02 17:45:46,735 INFO [optim.py:369] (1/2) 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 17:45:50,972 INFO [train.py:968] (1/2) Epoch 5, batch 19900, giga_loss[loss=0.2375, simple_loss=0.3145, pruned_loss=0.08025, over 28896.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3413, pruned_loss=0.102, over 5708068.79 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3706, pruned_loss=0.1133, over 5693579.36 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3366, pruned_loss=0.1002, over 5715904.99 frames. ], batch size: 145, lr: 6.39e-03, grad_scale: 4.0
+2023-03-02 17:46:20,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5093, 2.0748, 2.1994, 2.0359], device='cuda:1'), covar=tensor([0.1089, 0.2171, 0.1524, 0.1638], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0743, 0.0634, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 17:46:24,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=201839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:46:31,723 INFO [train.py:968] (1/2) Epoch 5, batch 19950, giga_loss[loss=0.2416, simple_loss=0.3129, pruned_loss=0.08518, over 28797.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3394, pruned_loss=0.1012, over 5710987.11 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3707, pruned_loss=0.1131, over 5697884.26 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3349, pruned_loss=0.09968, over 5713725.81 frames. ], batch size: 66, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:47:11,678 INFO [optim.py:369] (1/2) 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 17:47:12,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6091, 1.6952, 1.7005, 1.5737], device='cuda:1'), covar=tensor([0.0884, 0.1147, 0.1215, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0742, 0.0633, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 17:47:14,303 INFO [train.py:968] (1/2) Epoch 5, batch 20000, giga_loss[loss=0.2271, simple_loss=0.3023, pruned_loss=0.07593, over 28798.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3384, pruned_loss=0.1004, over 5718279.75 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3713, pruned_loss=0.1132, over 5699821.98 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3332, pruned_loss=0.09865, over 5719623.71 frames. ], batch size: 199, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:47:33,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4408, 1.5604, 1.3227, 1.7487], device='cuda:1'), covar=tensor([0.2054, 0.2000, 0.1970, 0.1994], device='cuda:1'), in_proj_covar=tensor([0.1115, 0.0859, 0.0981, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 17:47:52,885 INFO [train.py:968] (1/2) Epoch 5, batch 20050, giga_loss[loss=0.2503, simple_loss=0.3148, pruned_loss=0.09296, over 28878.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3371, pruned_loss=0.09986, over 5724282.33 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3724, pruned_loss=0.1138, over 5704879.29 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3313, pruned_loss=0.09768, over 5721490.59 frames. ], batch size: 112, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:47:56,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0669, 1.3741, 1.0954, 0.9611], device='cuda:1'), covar=tensor([0.2216, 0.2100, 0.2207, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1115, 0.0858, 0.0982, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 17:48:16,380 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=201982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:48:18,362 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:48:29,186 INFO [train.py:968] (1/2) Epoch 5, batch 20100, giga_loss[loss=0.2845, simple_loss=0.3472, pruned_loss=0.1109, over 28639.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3366, pruned_loss=0.09929, over 5733740.47 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3726, pruned_loss=0.1138, over 5708776.13 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3305, pruned_loss=0.09709, over 5728554.68 frames. ], batch size: 78, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:48:42,285 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:48:48,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:49:02,332 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:49:07,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:49:09,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0320, 1.2341, 1.3265, 1.0915], device='cuda:1'), covar=tensor([0.1014, 0.0950, 0.1337, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0745, 0.0637, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 17:49:14,099 INFO [train.py:968] (1/2) Epoch 5, batch 20150, giga_loss[loss=0.3024, simple_loss=0.3639, pruned_loss=0.1205, over 28619.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3418, pruned_loss=0.1031, over 5728710.74 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3731, pruned_loss=0.1139, over 5713623.64 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3357, pruned_loss=0.1009, over 5720488.58 frames. ], batch size: 284, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:49:17,451 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:50:00,337 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=202099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:50:03,238 INFO [train.py:968] (1/2) Epoch 5, batch 20200, giga_loss[loss=0.3396, simple_loss=0.396, pruned_loss=0.1416, over 29007.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3485, pruned_loss=0.1077, over 5724942.91 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3728, pruned_loss=0.1137, over 5717497.30 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3434, pruned_loss=0.106, over 5715159.39 frames. ], batch size: 164, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:50:37,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5783, 1.4716, 1.1894, 1.2996], device='cuda:1'), covar=tensor([0.0550, 0.0487, 0.0844, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0447, 0.0506, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 17:50:52,406 INFO [train.py:968] (1/2) Epoch 5, batch 20250, giga_loss[loss=0.2971, simple_loss=0.3708, pruned_loss=0.1116, over 28989.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3578, pruned_loss=0.1144, over 5704896.42 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3728, pruned_loss=0.1137, over 5720429.09 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3532, pruned_loss=0.113, over 5694142.36 frames. ], batch size: 136, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:51:05,863 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:08,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:17,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-02 17:51:21,855 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:25,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:32,640 INFO [optim.py:369] (1/2) 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 17:51:32,940 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:51:36,529 INFO [train.py:968] (1/2) Epoch 5, batch 20300, giga_loss[loss=0.291, simple_loss=0.3596, pruned_loss=0.1112, over 28978.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3628, pruned_loss=0.1167, over 5701646.26 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3728, pruned_loss=0.1136, over 5723191.23 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3588, pruned_loss=0.1157, over 5690331.96 frames. ], batch size: 136, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:51:42,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5865, 2.3544, 2.0112, 2.0750], device='cuda:1'), covar=tensor([0.1006, 0.1736, 0.1469, 0.1455], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0743, 0.0632, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 17:51:55,283 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:52:25,460 INFO [train.py:968] (1/2) Epoch 5, batch 20350, giga_loss[loss=0.3504, simple_loss=0.4111, pruned_loss=0.1448, over 28609.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3684, pruned_loss=0.1198, over 5686059.53 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3727, pruned_loss=0.1136, over 5725046.56 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3653, pruned_loss=0.1191, over 5675272.14 frames. ], batch size: 307, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:52:48,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-02 17:52:53,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5096, 1.0090, 2.8492, 2.7053], device='cuda:1'), covar=tensor([0.1711, 0.2136, 0.0510, 0.0727], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0514, 0.0723, 0.0589], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 17:53:13,354 INFO [optim.py:369] (1/2) 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 17:53:16,259 INFO [train.py:968] (1/2) Epoch 5, batch 20400, giga_loss[loss=0.304, simple_loss=0.3736, pruned_loss=0.1172, over 28556.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3744, pruned_loss=0.1235, over 5679775.06 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3727, pruned_loss=0.1136, over 5726058.32 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5670101.60 frames. ], batch size: 71, lr: 6.38e-03, grad_scale: 8.0
+2023-03-02 17:53:56,627 INFO [train.py:968] (1/2) Epoch 5, batch 20450, giga_loss[loss=0.2841, simple_loss=0.3531, pruned_loss=0.1076, over 28965.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.378, pruned_loss=0.1262, over 5677500.74 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3736, pruned_loss=0.1143, over 5726410.90 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3754, pruned_loss=0.1255, over 5667907.52 frames. ], batch size: 227, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:54:27,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4116, 3.2212, 1.3849, 1.5199], device='cuda:1'), covar=tensor([0.0778, 0.0221, 0.0752, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0463, 0.0301, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 17:54:28,523 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:54:41,781 INFO [train.py:968] (1/2) Epoch 5, batch 20500, giga_loss[loss=0.3211, simple_loss=0.3886, pruned_loss=0.1268, over 28961.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5684416.25 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3737, pruned_loss=0.1144, over 5731121.71 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3694, pruned_loss=0.1205, over 5671191.96 frames. ], batch size: 145, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:54:52,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:55:06,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:55:24,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-02 17:55:24,640 INFO [train.py:968] (1/2) Epoch 5, batch 20550, giga_loss[loss=0.2713, simple_loss=0.3503, pruned_loss=0.09608, over 28760.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3696, pruned_loss=0.1187, over 5701552.00 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.374, pruned_loss=0.1146, over 5734483.79 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3675, pruned_loss=0.1182, over 5687287.39 frames. ], batch size: 262, lr: 6.38e-03, grad_scale: 4.0
+2023-03-02 17:55:26,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 17:55:45,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:56:04,764 INFO [optim.py:369] (1/2) 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 17:56:06,812 INFO [train.py:968] (1/2) Epoch 5, batch 20600, giga_loss[loss=0.291, simple_loss=0.3687, pruned_loss=0.1067, over 28776.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3692, pruned_loss=0.118, over 5701313.66 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3739, pruned_loss=0.1146, over 5738699.13 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3675, pruned_loss=0.1177, over 5685292.65 frames. ], batch size: 119, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:56:22,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-02 17:56:50,759 INFO [train.py:968] (1/2) Epoch 5, batch 20650, giga_loss[loss=0.4534, simple_loss=0.4621, pruned_loss=0.2223, over 26704.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3721, pruned_loss=0.1196, over 5701535.41 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3741, pruned_loss=0.1148, over 5740940.96 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3705, pruned_loss=0.1193, over 5686342.98 frames. ], batch size: 555, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:56:51,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2985, 1.2329, 4.9611, 3.4777], device='cuda:1'), covar=tensor([0.1603, 0.2375, 0.0276, 0.0639], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0524, 0.0740, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 17:56:56,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:56:58,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:10,233 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:13,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:22,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:29,974 INFO [optim.py:369] (1/2) 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 17:57:33,378 INFO [train.py:968] (1/2) Epoch 5, batch 20700, libri_loss[loss=0.2789, simple_loss=0.3499, pruned_loss=0.104, over 29579.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3748, pruned_loss=0.1216, over 5709073.23 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3755, pruned_loss=0.1159, over 5746716.33 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3722, pruned_loss=0.1205, over 5689925.91 frames. ], batch size: 74, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:57:38,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:50,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:57:53,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:58:00,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7129, 1.8152, 1.3427, 1.0722], device='cuda:1'), covar=tensor([0.0999, 0.0716, 0.0724, 0.0962], device='cuda:1'), in_proj_covar=tensor([0.1403, 0.1163, 0.1198, 0.1270], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 17:58:18,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 17:58:19,095 INFO [train.py:968] (1/2) Epoch 5, batch 20750, giga_loss[loss=0.2889, simple_loss=0.358, pruned_loss=0.1099, over 28608.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3762, pruned_loss=0.1229, over 5705432.51 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3761, pruned_loss=0.1163, over 5747281.60 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3736, pruned_loss=0.1217, over 5689327.35 frames. ], batch size: 85, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:59:03,856 INFO [optim.py:369] (1/2) 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 17:59:06,699 INFO [train.py:968] (1/2) Epoch 5, batch 20800, giga_loss[loss=0.4063, simple_loss=0.4427, pruned_loss=0.185, over 28948.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3773, pruned_loss=0.1243, over 5696919.78 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3766, pruned_loss=0.1167, over 5751704.13 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3748, pruned_loss=0.1231, over 5678786.81 frames. ], batch size: 186, lr: 6.37e-03, grad_scale: 8.0
+2023-03-02 17:59:07,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1272, 2.9950, 2.8232, 1.5036], device='cuda:1'), covar=tensor([0.0816, 0.0643, 0.0944, 0.2174], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0755, 0.0788, 0.0590], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 17:59:40,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3007, 3.2108, 1.4125, 1.3859], device='cuda:1'), covar=tensor([0.1238, 0.0415, 0.0965, 0.1616], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0467, 0.0301, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0020], device='cuda:1')
+2023-03-02 17:59:49,205 INFO [train.py:968] (1/2) Epoch 5, batch 20850, giga_loss[loss=0.3433, simple_loss=0.396, pruned_loss=0.1453, over 28858.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3776, pruned_loss=0.1249, over 5693712.75 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3768, pruned_loss=0.1168, over 5745222.09 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3755, pruned_loss=0.1241, over 5683543.87 frames. ], batch size: 199, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 17:59:56,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=202758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:00:28,261 INFO [optim.py:369] (1/2) 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 18:00:29,644 INFO [train.py:968] (1/2) Epoch 5, batch 20900, giga_loss[loss=0.3565, simple_loss=0.4136, pruned_loss=0.1497, over 28713.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3768, pruned_loss=0.1235, over 5702932.31 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3764, pruned_loss=0.1166, over 5746079.53 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3754, pruned_loss=0.1233, over 5692662.47 frames. ], batch size: 262, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:01:08,893 INFO [train.py:968] (1/2) Epoch 5, batch 20950, giga_loss[loss=0.2971, simple_loss=0.3698, pruned_loss=0.1122, over 28330.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3762, pruned_loss=0.1216, over 5702167.11 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.377, pruned_loss=0.117, over 5746730.39 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3746, pruned_loss=0.1213, over 5692001.49 frames. ], batch size: 368, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:01:50,844 INFO [optim.py:369] (1/2) 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 18:01:51,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8335, 1.0722, 3.8411, 3.0190], device='cuda:1'), covar=tensor([0.1789, 0.2419, 0.0358, 0.0693], device='cuda:1'), in_proj_covar=tensor([0.0566, 0.0522, 0.0738, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 18:01:52,189 INFO [train.py:968] (1/2) Epoch 5, batch 21000, giga_loss[loss=0.299, simple_loss=0.3737, pruned_loss=0.1121, over 28714.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3763, pruned_loss=0.1203, over 5698705.49 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3769, pruned_loss=0.1169, over 5739007.42 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3751, pruned_loss=0.1201, over 5695984.86 frames. ], batch size: 119, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:01:52,190 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 18:02:00,844 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 18:02:01,746 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=202901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:02:03,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=202904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:02:26,449 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=202933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:02:34,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0834, 1.8357, 1.5588, 1.5985], device='cuda:1'), covar=tensor([0.0621, 0.0689, 0.0802, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0443, 0.0500, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:02:40,115 INFO [train.py:968] (1/2) Epoch 5, batch 21050, giga_loss[loss=0.368, simple_loss=0.4137, pruned_loss=0.1611, over 28983.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.376, pruned_loss=0.1206, over 5695389.94 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3777, pruned_loss=0.1179, over 5736993.14 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3744, pruned_loss=0.1197, over 5693754.94 frames. ], batch size: 213, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:02:48,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0241, 1.8956, 1.3935, 1.6034], device='cuda:1'), covar=tensor([0.0607, 0.0628, 0.0897, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0445, 0.0501, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:03:16,744 INFO [optim.py:369] (1/2) 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 18:03:18,069 INFO [train.py:968] (1/2) Epoch 5, batch 21100, libri_loss[loss=0.2657, simple_loss=0.3427, pruned_loss=0.09439, over 29562.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3744, pruned_loss=0.1201, over 5709255.10 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3783, pruned_loss=0.1187, over 5739945.41 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3725, pruned_loss=0.1187, over 5704215.95 frames. ], batch size: 80, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:03:56,899 INFO [train.py:968] (1/2) Epoch 5, batch 21150, giga_loss[loss=0.2873, simple_loss=0.3632, pruned_loss=0.1057, over 28615.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3729, pruned_loss=0.1192, over 5698834.85 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3787, pruned_loss=0.1194, over 5727506.77 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3707, pruned_loss=0.1176, over 5704985.56 frames. ], batch size: 242, lr: 6.37e-03, grad_scale: 4.0
+2023-03-02 18:04:32,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7405, 1.1617, 3.6482, 3.0606], device='cuda:1'), covar=tensor([0.1788, 0.2272, 0.0377, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0557, 0.0515, 0.0731, 0.0595], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 18:04:37,150 INFO [optim.py:369] (1/2) 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 18:04:38,600 INFO [train.py:968] (1/2) Epoch 5, batch 21200, giga_loss[loss=0.3008, simple_loss=0.3691, pruned_loss=0.1163, over 28316.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3713, pruned_loss=0.1186, over 5696331.87 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3793, pruned_loss=0.12, over 5723414.18 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3688, pruned_loss=0.1166, over 5704288.74 frames. ], batch size: 77, lr: 6.37e-03, grad_scale: 8.0
+2023-03-02 18:05:22,122 INFO [train.py:968] (1/2) Epoch 5, batch 21250, giga_loss[loss=0.2999, simple_loss=0.3682, pruned_loss=0.1158, over 28930.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3728, pruned_loss=0.12, over 5700257.56 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3795, pruned_loss=0.1202, over 5727102.56 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3704, pruned_loss=0.1183, over 5702655.09 frames. ], batch size: 227, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:05:34,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 18:06:02,137 INFO [optim.py:369] (1/2) 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 18:06:02,810 INFO [train.py:968] (1/2) Epoch 5, batch 21300, giga_loss[loss=0.2985, simple_loss=0.3724, pruned_loss=0.1123, over 28539.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3724, pruned_loss=0.1194, over 5698338.30 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3798, pruned_loss=0.1207, over 5719663.41 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3701, pruned_loss=0.1176, over 5707169.21 frames. ], batch size: 307, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:06:17,590 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-02 18:06:20,925 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203223.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:06:43,220 INFO [train.py:968] (1/2) Epoch 5, batch 21350, giga_loss[loss=0.3128, simple_loss=0.3734, pruned_loss=0.1261, over 28670.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3713, pruned_loss=0.1186, over 5699041.51 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3795, pruned_loss=0.121, over 5724439.42 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3695, pruned_loss=0.1167, over 5701373.76 frames. ], batch size: 92, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:06:53,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1395, 1.5041, 1.4550, 1.2976], device='cuda:1'), covar=tensor([0.1521, 0.2063, 0.1217, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0741, 0.0781, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 18:07:23,237 INFO [optim.py:369] (1/2) 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 18:07:24,267 INFO [train.py:968] (1/2) Epoch 5, batch 21400, giga_loss[loss=0.2624, simple_loss=0.3341, pruned_loss=0.09537, over 28919.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3696, pruned_loss=0.1163, over 5705545.63 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3795, pruned_loss=0.1211, over 5719812.29 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.368, pruned_loss=0.1148, over 5711625.80 frames. ], batch size: 112, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:08:05,446 INFO [train.py:968] (1/2) Epoch 5, batch 21450, giga_loss[loss=0.2784, simple_loss=0.347, pruned_loss=0.1049, over 28980.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3692, pruned_loss=0.116, over 5707238.86 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3801, pruned_loss=0.1217, over 5714675.88 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3671, pruned_loss=0.1141, over 5717253.72 frames. ], batch size: 136, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:08:47,385 INFO [optim.py:369] (1/2) 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 18:08:48,031 INFO [train.py:968] (1/2) Epoch 5, batch 21500, giga_loss[loss=0.2885, simple_loss=0.3619, pruned_loss=0.1076, over 28987.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3655, pruned_loss=0.1142, over 5713723.71 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.38, pruned_loss=0.1218, over 5715490.50 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3639, pruned_loss=0.1126, over 5720753.03 frames. ], batch size: 164, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:09:14,245 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 21550, giga_loss[loss=0.2403, simple_loss=0.3214, pruned_loss=0.07958, over 28998.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3629, pruned_loss=0.1133, over 5711441.86 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.38, pruned_loss=0.1222, over 5718546.73 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3613, pruned_loss=0.1116, over 5714239.69 frames. ], batch size: 145, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:09:54,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5430, 2.2720, 1.6703, 0.8229], device='cuda:1'), covar=tensor([0.2609, 0.1226, 0.1917, 0.2664], device='cuda:1'), in_proj_covar=tensor([0.1370, 0.1273, 0.1360, 0.1140], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 18:09:55,285 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:10:01,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3200, 4.0995, 3.9867, 1.9431], device='cuda:1'), covar=tensor([0.0450, 0.0468, 0.0684, 0.1886], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0744, 0.0772, 0.0586], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:10:10,653 INFO [optim.py:369] (1/2) 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 18:10:11,220 INFO [train.py:968] (1/2) Epoch 5, batch 21600, giga_loss[loss=0.3255, simple_loss=0.3847, pruned_loss=0.1331, over 28708.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3634, pruned_loss=0.1141, over 5708874.53 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3803, pruned_loss=0.1226, over 5710970.13 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3618, pruned_loss=0.1122, over 5718669.87 frames. ], batch size: 284, lr: 6.36e-03, grad_scale: 8.0
+2023-03-02 18:10:38,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5943, 5.2203, 5.1471, 2.3895], device='cuda:1'), covar=tensor([0.0273, 0.0346, 0.0530, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0745, 0.0774, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:10:56,067 INFO [train.py:968] (1/2) Epoch 5, batch 21650, giga_loss[loss=0.2647, simple_loss=0.3353, pruned_loss=0.09702, over 28962.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.363, pruned_loss=0.1146, over 5708940.50 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3804, pruned_loss=0.1226, over 5711686.28 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3615, pruned_loss=0.1131, over 5716001.81 frames. ], batch size: 106, lr: 6.36e-03, grad_scale: 8.0
+2023-03-02 18:11:36,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203598.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:11:37,673 INFO [optim.py:369] (1/2) 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 18:11:37,685 INFO [train.py:968] (1/2) Epoch 5, batch 21700, giga_loss[loss=0.2862, simple_loss=0.3535, pruned_loss=0.1094, over 28909.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5705460.90 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.381, pruned_loss=0.1231, over 5712029.10 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5710750.29 frames. ], batch size: 174, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:12:16,180 INFO [train.py:968] (1/2) Epoch 5, batch 21750, giga_loss[loss=0.247, simple_loss=0.3169, pruned_loss=0.08852, over 28974.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3611, pruned_loss=0.1151, over 5713807.91 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3818, pruned_loss=0.1238, over 5714411.52 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3581, pruned_loss=0.1127, over 5716020.43 frames. ], batch size: 106, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:12:58,290 INFO [optim.py:369] (1/2) 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 18:12:58,304 INFO [train.py:968] (1/2) Epoch 5, batch 21800, giga_loss[loss=0.2558, simple_loss=0.3257, pruned_loss=0.09297, over 28883.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.358, pruned_loss=0.1135, over 5705928.53 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3826, pruned_loss=0.1244, over 5715824.69 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3544, pruned_loss=0.1109, over 5706403.02 frames. ], batch size: 186, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:13:21,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4341, 1.7211, 1.3497, 1.4901], device='cuda:1'), covar=tensor([0.0745, 0.0283, 0.0329, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0124, 0.0127, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 18:13:31,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203741.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:13:33,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203744.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:13:37,336 INFO [train.py:968] (1/2) Epoch 5, batch 21850, giga_loss[loss=0.3332, simple_loss=0.3917, pruned_loss=0.1373, over 28861.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3579, pruned_loss=0.1138, over 5712524.01 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3834, pruned_loss=0.1252, over 5720626.26 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3534, pruned_loss=0.1106, over 5708258.20 frames. ], batch size: 186, lr: 6.36e-03, grad_scale: 4.0
+2023-03-02 18:13:37,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3423, 1.8304, 1.5134, 1.5586], device='cuda:1'), covar=tensor([0.0764, 0.0278, 0.0309, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0124, 0.0127, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 18:13:59,323 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203773.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:14:14,388 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-02 18:14:21,940 INFO [optim.py:369] (1/2) 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 18:14:21,957 INFO [train.py:968] (1/2) Epoch 5, batch 21900, giga_loss[loss=0.2817, simple_loss=0.344, pruned_loss=0.1097, over 28432.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3587, pruned_loss=0.1147, over 5706647.19 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3832, pruned_loss=0.1252, over 5719192.27 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3551, pruned_loss=0.112, over 5704210.99 frames. ], batch size: 71, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:14:27,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:14:35,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4387, 1.6207, 1.4191, 0.9974], device='cuda:1'), covar=tensor([0.1526, 0.0920, 0.0727, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.1417, 0.1188, 0.1205, 0.1274], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 18:14:41,947 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-02 18:14:55,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5310, 1.5505, 1.5314, 1.4615], device='cuda:1'), covar=tensor([0.1142, 0.1816, 0.1567, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0722, 0.0624, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 18:15:05,842 INFO [train.py:968] (1/2) Epoch 5, batch 21950, giga_loss[loss=0.3217, simple_loss=0.3801, pruned_loss=0.1317, over 28972.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.362, pruned_loss=0.1161, over 5707721.66 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.383, pruned_loss=0.1253, over 5721128.90 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3588, pruned_loss=0.1136, over 5703843.39 frames. ], batch size: 136, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:15:08,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=203853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:15:14,797 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=203860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:15:25,593 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-02 18:15:49,413 INFO [optim.py:369] (1/2) 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 18:15:49,426 INFO [train.py:968] (1/2) Epoch 5, batch 22000, giga_loss[loss=0.2737, simple_loss=0.3388, pruned_loss=0.1043, over 28594.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.364, pruned_loss=0.1166, over 5710967.66 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3837, pruned_loss=0.126, over 5724736.40 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3604, pruned_loss=0.1138, over 5704405.17 frames. ], batch size: 85, lr: 6.35e-03, grad_scale: 8.0
+2023-03-02 18:16:13,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1556, 1.3203, 1.3726, 1.3359], device='cuda:1'), covar=tensor([0.0947, 0.0973, 0.1182, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0728, 0.0627, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 18:16:22,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0817, 1.1579, 4.0731, 3.0092], device='cuda:1'), covar=tensor([0.1550, 0.2261, 0.0298, 0.0725], device='cuda:1'), in_proj_covar=tensor([0.0554, 0.0512, 0.0725, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 18:16:22,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 18:16:28,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7579, 2.6153, 1.6841, 0.8395], device='cuda:1'), covar=tensor([0.3637, 0.1548, 0.2348, 0.3738], device='cuda:1'), in_proj_covar=tensor([0.1353, 0.1252, 0.1352, 0.1128], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 18:16:32,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:16:32,972 INFO [train.py:968] (1/2) Epoch 5, batch 22050, giga_loss[loss=0.2674, simple_loss=0.3469, pruned_loss=0.09396, over 28649.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3654, pruned_loss=0.1166, over 5706646.12 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3839, pruned_loss=0.1264, over 5728214.64 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.362, pruned_loss=0.1138, over 5697833.76 frames. ], batch size: 242, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:16:35,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:00,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=203981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:07,108 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-02 18:17:14,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=203996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:16,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=203999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:17,100 INFO [train.py:968] (1/2) Epoch 5, batch 22100, giga_loss[loss=0.2437, simple_loss=0.3212, pruned_loss=0.0831, over 28991.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3643, pruned_loss=0.1154, over 5706963.91 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3838, pruned_loss=0.1266, over 5732027.07 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3613, pruned_loss=0.1128, over 5696171.55 frames. ], batch size: 136, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:17:18,150 INFO [optim.py:369] (1/2) 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 18:17:24,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-02 18:17:37,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 18:17:42,543 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:17:59,063 INFO [train.py:968] (1/2) Epoch 5, batch 22150, giga_loss[loss=0.3419, simple_loss=0.3975, pruned_loss=0.1431, over 26713.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3646, pruned_loss=0.1159, over 5709153.03 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3845, pruned_loss=0.1273, over 5735028.96 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3613, pruned_loss=0.1131, over 5697505.05 frames. ], batch size: 555, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:18:35,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 18:18:41,805 INFO [train.py:968] (1/2) Epoch 5, batch 22200, giga_loss[loss=0.3094, simple_loss=0.3575, pruned_loss=0.1306, over 23723.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3658, pruned_loss=0.117, over 5703835.03 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3848, pruned_loss=0.1276, over 5734184.51 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3627, pruned_loss=0.1144, over 5695005.59 frames. ], batch size: 705, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:18:43,319 INFO [optim.py:369] (1/2) 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 18:19:15,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:19:23,025 INFO [train.py:968] (1/2) Epoch 5, batch 22250, giga_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08837, over 28388.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3683, pruned_loss=0.1187, over 5705789.16 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3853, pruned_loss=0.1283, over 5729667.74 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3647, pruned_loss=0.1155, over 5701923.48 frames. ], batch size: 65, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:19:31,054 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:20:04,142 INFO [train.py:968] (1/2) Epoch 5, batch 22300, giga_loss[loss=0.3162, simple_loss=0.3799, pruned_loss=0.1263, over 28703.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3706, pruned_loss=0.1199, over 5701819.37 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3856, pruned_loss=0.1285, over 5732876.46 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3672, pruned_loss=0.117, over 5695587.57 frames. ], batch size: 60, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:20:04,740 INFO [optim.py:369] (1/2) 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 18:20:16,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2406, 1.7771, 1.3166, 0.4259], device='cuda:1'), covar=tensor([0.1849, 0.1244, 0.2202, 0.2564], device='cuda:1'), in_proj_covar=tensor([0.1365, 0.1263, 0.1367, 0.1137], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 18:20:35,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:20:38,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:20:45,186 INFO [train.py:968] (1/2) Epoch 5, batch 22350, giga_loss[loss=0.3126, simple_loss=0.3739, pruned_loss=0.1256, over 29039.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3726, pruned_loss=0.1207, over 5710697.96 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3862, pruned_loss=0.129, over 5735804.62 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3691, pruned_loss=0.1178, over 5702370.88 frames. ], batch size: 128, lr: 6.35e-03, grad_scale: 4.0
+2023-03-02 18:21:19,592 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204293.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:21:25,250 INFO [train.py:968] (1/2) Epoch 5, batch 22400, giga_loss[loss=0.4007, simple_loss=0.4369, pruned_loss=0.1823, over 28924.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3747, pruned_loss=0.1219, over 5712398.34 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3872, pruned_loss=0.1298, over 5735707.90 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3706, pruned_loss=0.1187, over 5705338.20 frames. ], batch size: 186, lr: 6.35e-03, grad_scale: 8.0
+2023-03-02 18:21:25,863 INFO [optim.py:369] (1/2) 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 18:21:50,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-02 18:22:05,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3849, 2.0035, 1.5102, 1.5366], device='cuda:1'), covar=tensor([0.0717, 0.0251, 0.0316, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0124, 0.0127, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0062], device='cuda:1')
+2023-03-02 18:22:05,733 INFO [train.py:968] (1/2) Epoch 5, batch 22450, giga_loss[loss=0.2941, simple_loss=0.3619, pruned_loss=0.1131, over 29090.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3754, pruned_loss=0.1222, over 5718377.90 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3878, pruned_loss=0.1304, over 5739313.91 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3713, pruned_loss=0.119, over 5708905.84 frames. ], batch size: 128, lr: 6.35e-03, grad_scale: 8.0
+2023-03-02 18:22:30,955 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:22:33,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:22:48,024 INFO [train.py:968] (1/2) Epoch 5, batch 22500, giga_loss[loss=0.277, simple_loss=0.345, pruned_loss=0.1045, over 28874.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3757, pruned_loss=0.1225, over 5720090.97 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3881, pruned_loss=0.1308, over 5740724.98 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3717, pruned_loss=0.1191, over 5710629.10 frames. ], batch size: 112, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:22:48,700 INFO [optim.py:369] (1/2) 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 18:22:55,967 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:23:08,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2448, 1.5097, 1.2553, 1.1433], device='cuda:1'), covar=tensor([0.2154, 0.2037, 0.2174, 0.1989], device='cuda:1'), in_proj_covar=tensor([0.1100, 0.0860, 0.0974, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 18:23:28,381 INFO [train.py:968] (1/2) Epoch 5, batch 22550, giga_loss[loss=0.2773, simple_loss=0.3485, pruned_loss=0.1031, over 28879.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3745, pruned_loss=0.1218, over 5710408.94 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3886, pruned_loss=0.1312, over 5734537.68 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3707, pruned_loss=0.1187, over 5707581.54 frames. ], batch size: 213, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:23:45,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:23:52,327 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 18:24:10,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 18:24:11,804 INFO [train.py:968] (1/2) Epoch 5, batch 22600, giga_loss[loss=0.2722, simple_loss=0.341, pruned_loss=0.1017, over 28883.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3716, pruned_loss=0.1202, over 5699894.55 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3893, pruned_loss=0.132, over 5726070.95 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3676, pruned_loss=0.1168, over 5705386.35 frames. ], batch size: 186, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:24:13,083 INFO [optim.py:369] (1/2) 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 18:24:24,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:24:28,451 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-02 18:24:39,387 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 22650, giga_loss[loss=0.2586, simple_loss=0.3321, pruned_loss=0.09251, over 28801.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.367, pruned_loss=0.1176, over 5698878.42 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3898, pruned_loss=0.1326, over 5725907.26 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3629, pruned_loss=0.114, over 5702846.01 frames. ], batch size: 119, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:24:56,017 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=204556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:25:31,223 INFO [train.py:968] (1/2) Epoch 5, batch 22700, giga_loss[loss=0.2716, simple_loss=0.3555, pruned_loss=0.09381, over 28846.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3662, pruned_loss=0.1161, over 5705209.56 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3898, pruned_loss=0.1329, over 5728759.21 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3624, pruned_loss=0.1126, over 5705196.71 frames. ], batch size: 186, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:25:32,381 INFO [optim.py:369] (1/2) 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 18:25:43,284 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 22750, libri_loss[loss=0.3189, simple_loss=0.3811, pruned_loss=0.1283, over 29516.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3688, pruned_loss=0.1162, over 5696277.71 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3902, pruned_loss=0.1333, over 5724072.25 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3648, pruned_loss=0.1126, over 5700073.99 frames. ], batch size: 81, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:26:20,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:22,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 18:26:23,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:30,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204668.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 18:26:36,105 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:41,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:48,274 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:26:55,682 INFO [train.py:968] (1/2) Epoch 5, batch 22800, giga_loss[loss=0.2521, simple_loss=0.329, pruned_loss=0.0876, over 28915.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3685, pruned_loss=0.1159, over 5696281.75 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3904, pruned_loss=0.1334, over 5724291.87 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.365, pruned_loss=0.1129, over 5698695.49 frames. ], batch size: 174, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:26:56,938 INFO [optim.py:369] (1/2) 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 18:27:01,990 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 22850, giga_loss[loss=0.3558, simple_loss=0.3943, pruned_loss=0.1586, over 26727.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.366, pruned_loss=0.116, over 5684834.95 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3901, pruned_loss=0.1333, over 5716151.34 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3632, pruned_loss=0.1135, over 5694628.79 frames. ], batch size: 555, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:27:45,304 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:27:47,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:27:47,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4762, 1.8472, 1.7807, 1.6307], device='cuda:1'), covar=tensor([0.1483, 0.1687, 0.1180, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0728, 0.0770, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 18:27:57,404 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.04 vs. limit=2.0
+2023-03-02 18:28:07,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 18:28:08,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:28:17,575 INFO [train.py:968] (1/2) Epoch 5, batch 22900, giga_loss[loss=0.2548, simple_loss=0.3218, pruned_loss=0.0939, over 28657.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3645, pruned_loss=0.1169, over 5695937.40 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3899, pruned_loss=0.1334, over 5719144.93 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.362, pruned_loss=0.1145, over 5700563.46 frames. ], batch size: 119, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:28:18,884 INFO [optim.py:369] (1/2) 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 18:28:27,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204811.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:28:30,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:28:45,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8198, 3.6799, 3.4709, 1.7227], device='cuda:1'), covar=tensor([0.0574, 0.0560, 0.0894, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.0841, 0.0760, 0.0794, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:28:53,991 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=204843.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 18:28:55,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:28:58,686 INFO [train.py:968] (1/2) Epoch 5, batch 22950, giga_loss[loss=0.3136, simple_loss=0.373, pruned_loss=0.1271, over 28888.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3631, pruned_loss=0.117, over 5703540.87 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3903, pruned_loss=0.1337, over 5718591.06 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3603, pruned_loss=0.1145, over 5707411.77 frames. ], batch size: 112, lr: 6.34e-03, grad_scale: 8.0
+2023-03-02 18:29:19,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6747, 1.4444, 1.2881, 1.2783], device='cuda:1'), covar=tensor([0.0502, 0.0440, 0.0811, 0.0663], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0447, 0.0501, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:29:39,664 INFO [train.py:968] (1/2) Epoch 5, batch 23000, giga_loss[loss=0.3264, simple_loss=0.3867, pruned_loss=0.133, over 28574.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3633, pruned_loss=0.1182, over 5705780.72 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.3904, pruned_loss=0.1339, over 5721470.03 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3605, pruned_loss=0.1158, over 5706010.96 frames. ], batch size: 336, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:29:40,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3398, 1.9601, 1.4150, 0.6575], device='cuda:1'), covar=tensor([0.2559, 0.1241, 0.2173, 0.2960], device='cuda:1'), in_proj_covar=tensor([0.1364, 0.1260, 0.1361, 0.1140], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 18:29:41,591 INFO [optim.py:369] (1/2) 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 18:30:05,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=204931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:30:17,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8558, 1.2510, 3.3615, 2.8945], device='cuda:1'), covar=tensor([0.1577, 0.2089, 0.0407, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0519, 0.0741, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 18:30:20,354 INFO [train.py:968] (1/2) Epoch 5, batch 23050, libri_loss[loss=0.374, simple_loss=0.4185, pruned_loss=0.1647, over 29539.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3604, pruned_loss=0.1162, over 5714409.24 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3908, pruned_loss=0.1343, over 5725339.09 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3573, pruned_loss=0.1136, over 5710753.94 frames. ], batch size: 83, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:30:49,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=204988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:30:51,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=204991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:30:57,974 INFO [train.py:968] (1/2) Epoch 5, batch 23100, giga_loss[loss=0.2502, simple_loss=0.3246, pruned_loss=0.08783, over 28934.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3566, pruned_loss=0.1146, over 5715377.92 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3914, pruned_loss=0.1348, over 5728849.02 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.353, pruned_loss=0.1117, over 5709039.22 frames. ], batch size: 174, lr: 6.34e-03, grad_scale: 4.0
+2023-03-02 18:31:00,238 INFO [optim.py:369] (1/2) 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 18:31:15,592 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:31:37,646 INFO [train.py:968] (1/2) Epoch 5, batch 23150, giga_loss[loss=0.2655, simple_loss=0.3275, pruned_loss=0.1017, over 29041.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3526, pruned_loss=0.1121, over 5716018.42 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3912, pruned_loss=0.1349, over 5733519.68 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3488, pruned_loss=0.1091, over 5706311.99 frames. ], batch size: 128, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:31:57,944 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:31:59,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:32:18,628 INFO [train.py:968] (1/2) Epoch 5, batch 23200, giga_loss[loss=0.2429, simple_loss=0.31, pruned_loss=0.08785, over 28495.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3528, pruned_loss=0.1116, over 5720896.60 frames. ], libri_tot_loss[loss=0.3305, simple_loss=0.3911, pruned_loss=0.1349, over 5735438.01 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3492, pruned_loss=0.1088, over 5711183.72 frames. ], batch size: 60, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:32:22,490 INFO [optim.py:369] (1/2) 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 18:32:24,779 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:32:39,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3452, 4.2236, 4.0072, 1.7910], device='cuda:1'), covar=tensor([0.0357, 0.0347, 0.0613, 0.1865], device='cuda:1'), in_proj_covar=tensor([0.0830, 0.0760, 0.0789, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:32:59,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2538, 1.5437, 1.0443, 1.1571], device='cuda:1'), covar=tensor([0.1200, 0.0908, 0.0908, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1184, 0.1193, 0.1256], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 18:33:01,874 INFO [train.py:968] (1/2) Epoch 5, batch 23250, giga_loss[loss=0.2828, simple_loss=0.358, pruned_loss=0.1038, over 29018.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3568, pruned_loss=0.1138, over 5711232.12 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3912, pruned_loss=0.135, over 5729862.73 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3529, pruned_loss=0.1108, over 5707313.44 frames. ], batch size: 136, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:33:43,035 INFO [train.py:968] (1/2) Epoch 5, batch 23300, libri_loss[loss=0.3528, simple_loss=0.4041, pruned_loss=0.1508, over 25764.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3615, pruned_loss=0.1164, over 5707542.22 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3913, pruned_loss=0.1354, over 5728763.49 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3574, pruned_loss=0.1132, over 5705083.71 frames. ], batch size: 136, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:33:45,207 INFO [optim.py:369] (1/2) 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 18:33:54,683 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:34:22,457 INFO [train.py:968] (1/2) Epoch 5, batch 23350, giga_loss[loss=0.2826, simple_loss=0.3513, pruned_loss=0.107, over 28754.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5711512.32 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3913, pruned_loss=0.1355, over 5734220.39 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3613, pruned_loss=0.1148, over 5703840.46 frames. ], batch size: 119, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:34:47,432 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-02 18:35:04,324 INFO [train.py:968] (1/2) Epoch 5, batch 23400, giga_loss[loss=0.3414, simple_loss=0.4048, pruned_loss=0.139, over 28887.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3682, pruned_loss=0.1193, over 5709672.23 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3911, pruned_loss=0.1355, over 5738627.29 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3644, pruned_loss=0.1162, over 5699141.48 frames. ], batch size: 227, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:35:07,643 INFO [optim.py:369] (1/2) 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 18:35:33,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-02 18:35:47,713 INFO [train.py:968] (1/2) Epoch 5, batch 23450, giga_loss[loss=0.3118, simple_loss=0.38, pruned_loss=0.1217, over 28972.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.371, pruned_loss=0.1211, over 5704518.17 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3916, pruned_loss=0.1361, over 5737626.85 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3671, pruned_loss=0.1178, over 5696478.72 frames. ], batch size: 174, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:36:36,524 INFO [train.py:968] (1/2) Epoch 5, batch 23500, giga_loss[loss=0.3096, simple_loss=0.3726, pruned_loss=0.1233, over 29002.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3779, pruned_loss=0.1281, over 5701740.63 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.391, pruned_loss=0.1361, over 5742144.99 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3749, pruned_loss=0.1252, over 5690056.32 frames. ], batch size: 128, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:36:40,204 INFO [optim.py:369] (1/2) 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 18:37:23,260 INFO [train.py:968] (1/2) Epoch 5, batch 23550, libri_loss[loss=0.2944, simple_loss=0.3424, pruned_loss=0.1233, over 29373.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.382, pruned_loss=0.1312, over 5698491.05 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3911, pruned_loss=0.1364, over 5741468.95 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3791, pruned_loss=0.1284, over 5687887.37 frames. ], batch size: 67, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:38:09,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3151, 1.4149, 1.3334, 1.3468], device='cuda:1'), covar=tensor([0.1044, 0.1360, 0.1728, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0742, 0.0635, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 18:38:14,875 INFO [train.py:968] (1/2) Epoch 5, batch 23600, giga_loss[loss=0.3145, simple_loss=0.384, pruned_loss=0.1225, over 29026.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3911, pruned_loss=0.1387, over 5693977.63 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3917, pruned_loss=0.1371, over 5745150.13 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3881, pruned_loss=0.1357, over 5681023.67 frames. ], batch size: 155, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:38:17,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:38:18,260 INFO [optim.py:369] (1/2) 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 18:39:03,372 INFO [train.py:968] (1/2) Epoch 5, batch 23650, giga_loss[loss=0.334, simple_loss=0.3917, pruned_loss=0.1381, over 28999.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3977, pruned_loss=0.145, over 5681618.37 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3924, pruned_loss=0.1378, over 5738765.18 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3947, pruned_loss=0.1421, over 5675249.78 frames. ], batch size: 174, lr: 6.33e-03, grad_scale: 8.0
+2023-03-02 18:39:44,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:39:45,716 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=205591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:39:54,537 INFO [train.py:968] (1/2) Epoch 5, batch 23700, giga_loss[loss=0.3691, simple_loss=0.4164, pruned_loss=0.161, over 28967.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.4052, pruned_loss=0.152, over 5668466.54 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3924, pruned_loss=0.1379, over 5740324.81 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.403, pruned_loss=0.1498, over 5660987.06 frames. ], batch size: 136, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:40:01,262 INFO [optim.py:369] (1/2) 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 18:40:35,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7794, 1.6818, 1.1635, 1.4222], device='cuda:1'), covar=tensor([0.0604, 0.0551, 0.0964, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0445, 0.0496, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 18:40:42,323 INFO [train.py:968] (1/2) Epoch 5, batch 23750, giga_loss[loss=0.4033, simple_loss=0.4428, pruned_loss=0.1819, over 28658.00 frames. ], tot_loss[loss=0.3572, simple_loss=0.4074, pruned_loss=0.1535, over 5674458.86 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3923, pruned_loss=0.138, over 5742810.19 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4063, pruned_loss=0.1522, over 5663790.64 frames. ], batch size: 262, lr: 6.33e-03, grad_scale: 4.0
+2023-03-02 18:41:33,271 INFO [train.py:968] (1/2) Epoch 5, batch 23800, giga_loss[loss=0.3468, simple_loss=0.4064, pruned_loss=0.1436, over 28814.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.4095, pruned_loss=0.1562, over 5668983.87 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3922, pruned_loss=0.138, over 5744312.11 frames. ], giga_tot_loss[loss=0.3598, simple_loss=0.4089, pruned_loss=0.1553, over 5658144.48 frames. ], batch size: 186, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:41:40,646 INFO [optim.py:369] (1/2) 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 18:42:11,353 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=205732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:42:15,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=205735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:42:29,927 INFO [train.py:968] (1/2) Epoch 5, batch 23850, giga_loss[loss=0.3274, simple_loss=0.3896, pruned_loss=0.1326, over 28661.00 frames. ], tot_loss[loss=0.3649, simple_loss=0.4117, pruned_loss=0.1591, over 5648491.37 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3922, pruned_loss=0.138, over 5744964.55 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.4113, pruned_loss=0.1586, over 5639030.76 frames. ], batch size: 262, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:42:45,523 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=205764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:43:25,324 INFO [train.py:968] (1/2) Epoch 5, batch 23900, giga_loss[loss=0.4441, simple_loss=0.4639, pruned_loss=0.2122, over 27613.00 frames. ], tot_loss[loss=0.3715, simple_loss=0.416, pruned_loss=0.1635, over 5639488.38 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3921, pruned_loss=0.138, over 5742975.21 frames. ], giga_tot_loss[loss=0.3713, simple_loss=0.416, pruned_loss=0.1633, over 5632940.48 frames. ], batch size: 472, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:43:35,195 INFO [optim.py:369] (1/2) 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 18:44:21,337 INFO [train.py:968] (1/2) Epoch 5, batch 23950, giga_loss[loss=0.4232, simple_loss=0.4282, pruned_loss=0.2091, over 23646.00 frames. ], tot_loss[loss=0.3756, simple_loss=0.4181, pruned_loss=0.1666, over 5618511.30 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3921, pruned_loss=0.1382, over 5747628.20 frames. ], giga_tot_loss[loss=0.3764, simple_loss=0.4188, pruned_loss=0.167, over 5606042.58 frames. ], batch size: 705, lr: 6.32e-03, grad_scale: 2.0
+2023-03-02 18:44:25,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4527, 1.3664, 1.1604, 1.4203], device='cuda:1'), covar=tensor([0.0730, 0.0330, 0.0322, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0125, 0.0129, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0056, 0.0041, 0.0037, 0.0063], device='cuda:1')
+2023-03-02 18:44:51,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:45:07,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-02 18:45:14,061 INFO [train.py:968] (1/2) Epoch 5, batch 24000, giga_loss[loss=0.3777, simple_loss=0.4154, pruned_loss=0.17, over 28855.00 frames. ], tot_loss[loss=0.3734, simple_loss=0.416, pruned_loss=0.1654, over 5625427.08 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3923, pruned_loss=0.1383, over 5748207.15 frames. ], giga_tot_loss[loss=0.3744, simple_loss=0.4168, pruned_loss=0.166, over 5613317.04 frames. ], batch size: 174, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:45:14,061 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 18:45:22,825 INFO [train.py:1012] (1/2) Epoch 5, validation: loss=0.2388, simple_loss=0.3426, pruned_loss=0.06746, over 944034.00 frames. 
+2023-03-02 18:45:22,826 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19455MB
+2023-03-02 18:45:28,666 INFO [optim.py:369] (1/2) 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 18:46:07,860 INFO [train.py:968] (1/2) Epoch 5, batch 24050, giga_loss[loss=0.4146, simple_loss=0.4251, pruned_loss=0.2021, over 23671.00 frames. ], tot_loss[loss=0.37, simple_loss=0.4133, pruned_loss=0.1633, over 5632624.96 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3924, pruned_loss=0.1386, over 5742549.92 frames. ], giga_tot_loss[loss=0.3714, simple_loss=0.4144, pruned_loss=0.1642, over 5624690.94 frames. ], batch size: 705, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:46:25,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=205966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:46:53,580 INFO [train.py:968] (1/2) Epoch 5, batch 24100, giga_loss[loss=0.3676, simple_loss=0.4223, pruned_loss=0.1565, over 28703.00 frames. ], tot_loss[loss=0.3666, simple_loss=0.4117, pruned_loss=0.1607, over 5635025.27 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.392, pruned_loss=0.1384, over 5747014.53 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4135, pruned_loss=0.1622, over 5621475.47 frames. ], batch size: 262, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:47:00,752 INFO [optim.py:369] (1/2) 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 18:47:17,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:47:20,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:47:44,620 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 24150, giga_loss[loss=0.4032, simple_loss=0.4374, pruned_loss=0.1845, over 27600.00 frames. ], tot_loss[loss=0.3681, simple_loss=0.4133, pruned_loss=0.1614, over 5623166.29 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3926, pruned_loss=0.139, over 5745607.46 frames. ], giga_tot_loss[loss=0.37, simple_loss=0.4146, pruned_loss=0.1627, over 5610522.96 frames. ], batch size: 472, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:47:48,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:48:41,740 INFO [train.py:968] (1/2) Epoch 5, batch 24200, giga_loss[loss=0.3239, simple_loss=0.3874, pruned_loss=0.1302, over 28933.00 frames. ], tot_loss[loss=0.3677, simple_loss=0.4137, pruned_loss=0.1609, over 5624785.73 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3931, pruned_loss=0.1394, over 5743566.74 frames. ], giga_tot_loss[loss=0.3691, simple_loss=0.4146, pruned_loss=0.1618, over 5615333.28 frames. ], batch size: 213, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:48:48,923 INFO [optim.py:369] (1/2) 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 18:48:51,924 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:48:54,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:49:18,767 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:49:22,969 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 24250, giga_loss[loss=0.2916, simple_loss=0.3637, pruned_loss=0.1097, over 28648.00 frames. ], tot_loss[loss=0.3634, simple_loss=0.4105, pruned_loss=0.1581, over 5623682.97 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.393, pruned_loss=0.1396, over 5747134.21 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4118, pruned_loss=0.1592, over 5609995.96 frames. ], batch size: 71, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:49:52,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4091, 1.7190, 1.3266, 1.0327], device='cuda:1'), covar=tensor([0.1474, 0.0933, 0.0756, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.1405, 0.1190, 0.1194, 0.1267], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 18:49:54,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2351, 1.4943, 1.2264, 0.8516], device='cuda:1'), covar=tensor([0.1017, 0.0807, 0.0516, 0.0917], device='cuda:1'), in_proj_covar=tensor([0.1405, 0.1190, 0.1193, 0.1266], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 18:50:17,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-02 18:50:22,031 INFO [train.py:968] (1/2) Epoch 5, batch 24300, giga_loss[loss=0.3469, simple_loss=0.4083, pruned_loss=0.1427, over 28862.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4076, pruned_loss=0.1542, over 5641250.71 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3925, pruned_loss=0.1393, over 5749321.08 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4094, pruned_loss=0.1556, over 5626335.77 frames. ], batch size: 174, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:50:29,981 INFO [optim.py:369] (1/2) 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 18:51:11,812 INFO [train.py:968] (1/2) Epoch 5, batch 24350, giga_loss[loss=0.3454, simple_loss=0.3956, pruned_loss=0.1477, over 28279.00 frames. ], tot_loss[loss=0.353, simple_loss=0.404, pruned_loss=0.151, over 5632572.77 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3925, pruned_loss=0.1395, over 5743731.69 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4058, pruned_loss=0.1523, over 5622042.66 frames. ], batch size: 368, lr: 6.32e-03, grad_scale: 4.0
+2023-03-02 18:51:44,009 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:51:56,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5504, 1.5026, 1.4508, 1.4137], device='cuda:1'), covar=tensor([0.0873, 0.1332, 0.1444, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0743, 0.0635, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 18:51:59,989 INFO [train.py:968] (1/2) Epoch 5, batch 24400, giga_loss[loss=0.3281, simple_loss=0.3856, pruned_loss=0.1353, over 28877.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4006, pruned_loss=0.1483, over 5636346.86 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3926, pruned_loss=0.1397, over 5744885.25 frames. ], giga_tot_loss[loss=0.3505, simple_loss=0.4021, pruned_loss=0.1494, over 5623775.09 frames. ], batch size: 199, lr: 6.32e-03, grad_scale: 8.0
+2023-03-02 18:52:05,923 INFO [optim.py:369] (1/2) 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 18:52:48,397 INFO [train.py:968] (1/2) Epoch 5, batch 24450, giga_loss[loss=0.3538, simple_loss=0.4084, pruned_loss=0.1496, over 28811.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.4002, pruned_loss=0.1484, over 5637443.51 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3927, pruned_loss=0.1397, over 5745673.70 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4014, pruned_loss=0.1493, over 5626005.31 frames. ], batch size: 119, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:53:34,355 INFO [train.py:968] (1/2) Epoch 5, batch 24500, giga_loss[loss=0.3267, simple_loss=0.3895, pruned_loss=0.132, over 28930.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3993, pruned_loss=0.148, over 5643155.49 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3924, pruned_loss=0.1398, over 5748967.68 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.401, pruned_loss=0.1491, over 5626171.23 frames. ], batch size: 227, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:53:40,048 INFO [optim.py:369] (1/2) 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 18:53:59,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:54:24,312 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:54:24,753 INFO [train.py:968] (1/2) Epoch 5, batch 24550, giga_loss[loss=0.3372, simple_loss=0.4005, pruned_loss=0.137, over 28149.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3982, pruned_loss=0.146, over 5653393.87 frames. ], libri_tot_loss[loss=0.3359, simple_loss=0.3924, pruned_loss=0.1397, over 5749688.66 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.3996, pruned_loss=0.147, over 5637325.43 frames. ], batch size: 77, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:55:17,165 INFO [train.py:968] (1/2) Epoch 5, batch 24600, giga_loss[loss=0.4203, simple_loss=0.4476, pruned_loss=0.1965, over 26539.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3971, pruned_loss=0.1433, over 5650996.02 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3928, pruned_loss=0.1401, over 5739530.56 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3979, pruned_loss=0.1437, over 5646624.98 frames. ], batch size: 555, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:55:23,885 INFO [optim.py:369] (1/2) 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 18:55:31,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:55:40,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4180, 1.5250, 1.4099, 1.4053], device='cuda:1'), covar=tensor([0.1027, 0.1389, 0.1500, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0737, 0.0631, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 18:56:07,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2602, 1.6979, 1.3950, 1.4869], device='cuda:1'), covar=tensor([0.0765, 0.0328, 0.0319, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0125, 0.0129, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:1')
+2023-03-02 18:56:12,129 INFO [train.py:968] (1/2) Epoch 5, batch 24650, giga_loss[loss=0.421, simple_loss=0.4412, pruned_loss=0.2004, over 26751.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3985, pruned_loss=0.1421, over 5656192.92 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3926, pruned_loss=0.1401, over 5740501.91 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3992, pruned_loss=0.1425, over 5651665.18 frames. ], batch size: 555, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:56:30,120 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:56:32,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:56:51,842 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=206590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:56:59,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:57:01,520 INFO [train.py:968] (1/2) Epoch 5, batch 24700, giga_loss[loss=0.3355, simple_loss=0.3987, pruned_loss=0.1362, over 29049.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3995, pruned_loss=0.1439, over 5648253.40 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3926, pruned_loss=0.1402, over 5733013.92 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.4004, pruned_loss=0.1442, over 5648834.11 frames. ], batch size: 155, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:57:08,512 INFO [optim.py:369] (1/2) 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 18:57:52,781 INFO [train.py:968] (1/2) Epoch 5, batch 24750, giga_loss[loss=0.3017, simple_loss=0.3708, pruned_loss=0.1163, over 28920.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3998, pruned_loss=0.144, over 5658484.30 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3927, pruned_loss=0.1403, over 5723595.85 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.4004, pruned_loss=0.1441, over 5666230.13 frames. ], batch size: 145, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 18:57:57,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:58:00,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:58:00,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 18:58:33,157 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 24800, giga_loss[loss=0.3735, simple_loss=0.4124, pruned_loss=0.1673, over 27600.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3996, pruned_loss=0.1452, over 5661383.52 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3928, pruned_loss=0.1407, over 5714806.33 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.4001, pruned_loss=0.1451, over 5673671.06 frames. ], batch size: 472, lr: 6.31e-03, grad_scale: 8.0
+2023-03-02 18:58:51,460 INFO [optim.py:369] (1/2) 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 18:59:17,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4328, 1.5352, 1.2740, 1.6759], device='cuda:1'), covar=tensor([0.1967, 0.1829, 0.1817, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.1113, 0.0865, 0.0988, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 18:59:31,764 INFO [train.py:968] (1/2) Epoch 5, batch 24850, giga_loss[loss=0.2986, simple_loss=0.3654, pruned_loss=0.1159, over 28836.00 frames. ], tot_loss[loss=0.3444, simple_loss=0.398, pruned_loss=0.1454, over 5664064.08 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3929, pruned_loss=0.1406, over 5717614.87 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3984, pruned_loss=0.1454, over 5670602.28 frames. ], batch size: 99, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:00:16,749 INFO [train.py:968] (1/2) Epoch 5, batch 24900, giga_loss[loss=0.3337, simple_loss=0.4096, pruned_loss=0.129, over 29049.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.396, pruned_loss=0.1432, over 5667093.64 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3926, pruned_loss=0.1405, over 5720350.91 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3966, pruned_loss=0.1434, over 5669192.44 frames. ], batch size: 164, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:00:17,882 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=206801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:00:20,303 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:00:35,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:00:43,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:00:48,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-02 19:01:01,364 INFO [train.py:968] (1/2) Epoch 5, batch 24950, giga_loss[loss=0.3554, simple_loss=0.4183, pruned_loss=0.1463, over 28974.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3957, pruned_loss=0.1411, over 5679337.70 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.393, pruned_loss=0.1408, over 5722657.20 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3959, pruned_loss=0.1411, over 5677967.94 frames. ], batch size: 213, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:01:36,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8241, 1.5928, 1.2232, 1.4214], device='cuda:1'), covar=tensor([0.0671, 0.0672, 0.0983, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0448, 0.0494, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:01:50,669 INFO [train.py:968] (1/2) Epoch 5, batch 25000, giga_loss[loss=0.28, simple_loss=0.355, pruned_loss=0.1025, over 28819.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3962, pruned_loss=0.1417, over 5673534.95 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3931, pruned_loss=0.141, over 5722533.19 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3964, pruned_loss=0.1415, over 5671886.85 frames. ], batch size: 119, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:01:57,808 INFO [optim.py:369] (1/2) 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 19:02:38,781 INFO [train.py:968] (1/2) Epoch 5, batch 25050, giga_loss[loss=0.325, simple_loss=0.3914, pruned_loss=0.1293, over 28906.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3946, pruned_loss=0.1406, over 5680201.03 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3929, pruned_loss=0.1408, over 5724711.12 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.395, pruned_loss=0.1406, over 5675554.61 frames. ], batch size: 186, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:02:52,817 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=206965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:02:54,318 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=206970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:03:00,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-02 19:03:27,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=206999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:03:28,187 INFO [train.py:968] (1/2) Epoch 5, batch 25100, giga_loss[loss=0.341, simple_loss=0.3754, pruned_loss=0.1533, over 23461.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3946, pruned_loss=0.1416, over 5670898.28 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3929, pruned_loss=0.141, over 5719240.81 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3949, pruned_loss=0.1415, over 5671929.71 frames. ], batch size: 705, lr: 6.31e-03, grad_scale: 4.0
+2023-03-02 19:03:35,264 INFO [optim.py:369] (1/2) 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 19:03:41,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6449, 1.8508, 1.3341, 1.1325], device='cuda:1'), covar=tensor([0.1409, 0.0972, 0.0896, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.1421, 0.1218, 0.1210, 0.1284], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 19:03:54,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7581, 1.5725, 1.2601, 1.3038], device='cuda:1'), covar=tensor([0.0580, 0.0599, 0.0885, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0448, 0.0495, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:04:13,843 INFO [train.py:968] (1/2) Epoch 5, batch 25150, giga_loss[loss=0.3135, simple_loss=0.3743, pruned_loss=0.1263, over 28972.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3946, pruned_loss=0.1429, over 5666433.74 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3933, pruned_loss=0.1413, over 5724395.62 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3945, pruned_loss=0.1424, over 5661002.12 frames. ], batch size: 136, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:04:26,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 19:04:59,739 INFO [train.py:968] (1/2) Epoch 5, batch 25200, giga_loss[loss=0.29, simple_loss=0.3546, pruned_loss=0.1127, over 28941.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3948, pruned_loss=0.1439, over 5675099.04 frames. ], libri_tot_loss[loss=0.3377, simple_loss=0.393, pruned_loss=0.1412, over 5726984.30 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.395, pruned_loss=0.1437, over 5667732.33 frames. ], batch size: 213, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:05:10,377 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:05:10,661 INFO [optim.py:369] (1/2) 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 19:05:12,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:05:16,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-02 19:05:16,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3974, 2.0307, 2.0168, 1.7961], device='cuda:1'), covar=tensor([0.0924, 0.2011, 0.1426, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0748, 0.0645, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 19:05:27,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5542, 1.8016, 1.3544, 1.0599], device='cuda:1'), covar=tensor([0.1387, 0.0922, 0.0771, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.1405, 0.1201, 0.1195, 0.1280], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 19:05:40,197 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 25250, giga_loss[loss=0.2919, simple_loss=0.3653, pruned_loss=0.1093, over 28890.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3934, pruned_loss=0.1433, over 5662626.20 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3934, pruned_loss=0.1415, over 5718605.82 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3933, pruned_loss=0.1429, over 5663589.97 frames. ], batch size: 164, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:06:34,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-02 19:06:36,643 INFO [train.py:968] (1/2) Epoch 5, batch 25300, giga_loss[loss=0.4064, simple_loss=0.4254, pruned_loss=0.1937, over 26746.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3934, pruned_loss=0.1436, over 5659069.90 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.3938, pruned_loss=0.1418, over 5713253.45 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.393, pruned_loss=0.143, over 5664000.07 frames. ], batch size: 555, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:06:47,049 INFO [optim.py:369] (1/2) 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 19:07:02,867 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=207227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:07:24,892 INFO [train.py:968] (1/2) Epoch 5, batch 25350, giga_loss[loss=0.3123, simple_loss=0.3841, pruned_loss=0.1203, over 28958.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3941, pruned_loss=0.1441, over 5650188.32 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3942, pruned_loss=0.1423, over 5703799.32 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3934, pruned_loss=0.1433, over 5661061.11 frames. ], batch size: 164, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:07:53,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-02 19:08:14,153 INFO [train.py:968] (1/2) Epoch 5, batch 25400, giga_loss[loss=0.3893, simple_loss=0.4162, pruned_loss=0.1812, over 26650.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3941, pruned_loss=0.1432, over 5654012.44 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.3943, pruned_loss=0.1426, over 5706818.30 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3934, pruned_loss=0.1423, over 5659029.37 frames. ], batch size: 555, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:08:22,793 INFO [optim.py:369] (1/2) 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 19:08:59,468 INFO [train.py:968] (1/2) Epoch 5, batch 25450, giga_loss[loss=0.3561, simple_loss=0.3901, pruned_loss=0.161, over 23830.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3937, pruned_loss=0.1424, over 5651144.99 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.394, pruned_loss=0.1428, over 5700685.43 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3934, pruned_loss=0.1415, over 5658905.13 frames. ], batch size: 705, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:09:27,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4696, 4.2775, 4.1076, 1.8234], device='cuda:1'), covar=tensor([0.0495, 0.0687, 0.0985, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.0868, 0.0796, 0.0816, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:09:45,661 INFO [train.py:968] (1/2) Epoch 5, batch 25500, giga_loss[loss=0.3321, simple_loss=0.3865, pruned_loss=0.1389, over 28816.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.394, pruned_loss=0.1426, over 5648680.42 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.3939, pruned_loss=0.1428, over 5698409.76 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3938, pruned_loss=0.1419, over 5655280.56 frames. ], batch size: 99, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:09:52,372 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 25550, giga_loss[loss=0.3989, simple_loss=0.4378, pruned_loss=0.18, over 27981.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3953, pruned_loss=0.1441, over 5654163.87 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3938, pruned_loss=0.1428, over 5702397.92 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3952, pruned_loss=0.1435, over 5655143.62 frames. ], batch size: 412, lr: 6.30e-03, grad_scale: 2.0
+2023-03-02 19:10:52,532 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-02 19:11:17,501 INFO [train.py:968] (1/2) Epoch 5, batch 25600, giga_loss[loss=0.3164, simple_loss=0.3726, pruned_loss=0.1301, over 28740.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3969, pruned_loss=0.1461, over 5652774.10 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.394, pruned_loss=0.143, over 5704981.26 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3967, pruned_loss=0.1455, over 5649870.79 frames. ], batch size: 99, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:11:26,483 INFO [optim.py:369] (1/2) 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 19:11:54,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5130, 1.8348, 1.8034, 1.6441], device='cuda:1'), covar=tensor([0.1474, 0.1866, 0.1140, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0743, 0.0771, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 19:12:03,022 INFO [train.py:968] (1/2) Epoch 5, batch 25650, giga_loss[loss=0.3721, simple_loss=0.4153, pruned_loss=0.1645, over 28583.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3971, pruned_loss=0.1469, over 5667776.55 frames. ], libri_tot_loss[loss=0.34, simple_loss=0.3939, pruned_loss=0.1431, over 5711877.81 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3971, pruned_loss=0.1465, over 5657107.43 frames. ], batch size: 307, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:12:24,557 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=207569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:12:28,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-02 19:12:40,581 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-02 19:12:43,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1487, 1.4562, 1.4027, 1.3808], device='cuda:1'), covar=tensor([0.1031, 0.0888, 0.1032, 0.0995], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0747, 0.0641, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 19:12:54,098 INFO [train.py:968] (1/2) Epoch 5, batch 25700, giga_loss[loss=0.3373, simple_loss=0.3947, pruned_loss=0.1399, over 28876.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.3981, pruned_loss=0.1487, over 5665824.55 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.3938, pruned_loss=0.1429, over 5718136.60 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.3984, pruned_loss=0.1487, over 5650336.42 frames. ], batch size: 186, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:12:56,539 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:13:14,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5808, 5.4136, 5.1879, 2.5645], device='cuda:1'), covar=tensor([0.0343, 0.0396, 0.0596, 0.1626], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0801, 0.0825, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:13:38,199 INFO [train.py:968] (1/2) Epoch 5, batch 25750, giga_loss[loss=0.4758, simple_loss=0.4853, pruned_loss=0.2331, over 24189.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.3989, pruned_loss=0.1496, over 5662644.83 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3934, pruned_loss=0.1426, over 5721398.19 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.3997, pruned_loss=0.15, over 5646210.25 frames. ], batch size: 705, lr: 6.30e-03, grad_scale: 4.0
+2023-03-02 19:13:46,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7451, 2.1800, 2.0012, 1.8101], device='cuda:1'), covar=tensor([0.1614, 0.1730, 0.1196, 0.1276], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0747, 0.0775, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 19:14:25,220 INFO [train.py:968] (1/2) Epoch 5, batch 25800, giga_loss[loss=0.3445, simple_loss=0.394, pruned_loss=0.1475, over 28322.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.3984, pruned_loss=0.1493, over 5662836.64 frames. ], libri_tot_loss[loss=0.3398, simple_loss=0.3936, pruned_loss=0.1429, over 5724267.26 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.3988, pruned_loss=0.1494, over 5646244.96 frames. ], batch size: 368, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:14:34,332 INFO [optim.py:369] (1/2) 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 19:15:06,312 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=207745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:15:08,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=207748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:15:09,623 INFO [train.py:968] (1/2) Epoch 5, batch 25850, giga_loss[loss=0.2913, simple_loss=0.3698, pruned_loss=0.1064, over 28872.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3983, pruned_loss=0.1472, over 5674930.43 frames. ], libri_tot_loss[loss=0.3401, simple_loss=0.394, pruned_loss=0.1431, over 5723229.50 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3984, pruned_loss=0.1473, over 5661800.89 frames. ], batch size: 145, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:15:33,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=207777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:15:56,203 INFO [train.py:968] (1/2) Epoch 5, batch 25900, giga_loss[loss=0.3409, simple_loss=0.3905, pruned_loss=0.1457, over 28951.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3958, pruned_loss=0.1455, over 5662426.42 frames. ], libri_tot_loss[loss=0.3414, simple_loss=0.395, pruned_loss=0.1439, over 5720319.48 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.395, pruned_loss=0.1449, over 5652119.45 frames. ], batch size: 213, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:16:04,535 INFO [optim.py:369] (1/2) 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 19:16:06,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 19:16:43,207 INFO [train.py:968] (1/2) Epoch 5, batch 25950, libri_loss[loss=0.4834, simple_loss=0.4777, pruned_loss=0.2446, over 19433.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3938, pruned_loss=0.1444, over 5652929.17 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3952, pruned_loss=0.1442, over 5703962.24 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.393, pruned_loss=0.1436, over 5659061.70 frames. ], batch size: 186, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:17:26,833 INFO [train.py:968] (1/2) Epoch 5, batch 26000, giga_loss[loss=0.3604, simple_loss=0.4111, pruned_loss=0.1549, over 28982.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3928, pruned_loss=0.1444, over 5666138.63 frames. ], libri_tot_loss[loss=0.3415, simple_loss=0.395, pruned_loss=0.1441, over 5706207.97 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3923, pruned_loss=0.1439, over 5667737.82 frames. ], batch size: 136, lr: 6.29e-03, grad_scale: 8.0
+2023-03-02 19:17:37,055 INFO [optim.py:369] (1/2) 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 19:18:11,210 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 26050, giga_loss[loss=0.3248, simple_loss=0.3902, pruned_loss=0.1297, over 28996.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3926, pruned_loss=0.1443, over 5661587.19 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3952, pruned_loss=0.1444, over 5696449.03 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3919, pruned_loss=0.1435, over 5670155.50 frames. ], batch size: 164, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:18:25,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-02 19:19:01,714 INFO [train.py:968] (1/2) Epoch 5, batch 26100, giga_loss[loss=0.3737, simple_loss=0.4267, pruned_loss=0.1603, over 27572.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3961, pruned_loss=0.1455, over 5661348.27 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3957, pruned_loss=0.1449, over 5690092.47 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3951, pruned_loss=0.1445, over 5672544.24 frames. ], batch size: 472, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:19:13,646 INFO [optim.py:369] (1/2) 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 19:19:19,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4476, 1.5610, 1.3645, 1.0835], device='cuda:1'), covar=tensor([0.1098, 0.1058, 0.0738, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1226, 0.1221, 0.1308], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 19:19:38,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-02 19:19:41,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 19:19:51,851 INFO [train.py:968] (1/2) Epoch 5, batch 26150, giga_loss[loss=0.3704, simple_loss=0.4163, pruned_loss=0.1623, over 26817.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3972, pruned_loss=0.1425, over 5667406.44 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3956, pruned_loss=0.1449, over 5694237.74 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3965, pruned_loss=0.1418, over 5672389.23 frames. ], batch size: 555, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:20:30,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:20:32,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:20:41,036 INFO [train.py:968] (1/2) Epoch 5, batch 26200, giga_loss[loss=0.3617, simple_loss=0.4151, pruned_loss=0.1541, over 28585.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3974, pruned_loss=0.1418, over 5674483.11 frames. ], libri_tot_loss[loss=0.3417, simple_loss=0.3948, pruned_loss=0.1443, over 5696282.19 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3976, pruned_loss=0.1417, over 5676189.05 frames. ], batch size: 307, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:20:52,381 INFO [optim.py:369] (1/2) 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 19:20:54,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-02 19:21:00,774 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:21:31,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-02 19:21:31,310 INFO [train.py:968] (1/2) Epoch 5, batch 26250, giga_loss[loss=0.3898, simple_loss=0.4264, pruned_loss=0.1766, over 27844.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3995, pruned_loss=0.1442, over 5672935.03 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3944, pruned_loss=0.1441, over 5697793.77 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.4001, pruned_loss=0.1443, over 5672783.10 frames. ], batch size: 412, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:22:14,954 INFO [train.py:968] (1/2) Epoch 5, batch 26300, libri_loss[loss=0.3836, simple_loss=0.4306, pruned_loss=0.1683, over 25877.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.4016, pruned_loss=0.1464, over 5664612.64 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3951, pruned_loss=0.1447, over 5688071.99 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.4016, pruned_loss=0.146, over 5672677.61 frames. ], batch size: 136, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:22:26,846 INFO [optim.py:369] (1/2) 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 19:23:03,503 INFO [train.py:968] (1/2) Epoch 5, batch 26350, giga_loss[loss=0.2995, simple_loss=0.3703, pruned_loss=0.1144, over 29051.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4022, pruned_loss=0.1481, over 5669843.27 frames. ], libri_tot_loss[loss=0.3416, simple_loss=0.3945, pruned_loss=0.1443, over 5689305.88 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4028, pruned_loss=0.1481, over 5674967.32 frames. ], batch size: 155, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:23:21,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4819, 1.9006, 1.7679, 1.6024], device='cuda:1'), covar=tensor([0.1513, 0.1896, 0.1147, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0747, 0.0782, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 19:23:27,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2476, 1.5473, 1.1974, 1.8255], device='cuda:1'), covar=tensor([0.2173, 0.2046, 0.2142, 0.2105], device='cuda:1'), in_proj_covar=tensor([0.1115, 0.0876, 0.0991, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 19:23:53,034 INFO [train.py:968] (1/2) Epoch 5, batch 26400, libri_loss[loss=0.4457, simple_loss=0.4698, pruned_loss=0.2109, over 29281.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3999, pruned_loss=0.147, over 5681766.12 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3948, pruned_loss=0.1446, over 5694823.09 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4002, pruned_loss=0.1468, over 5680439.92 frames. ], batch size: 94, lr: 6.29e-03, grad_scale: 4.0
+2023-03-02 19:24:04,426 INFO [optim.py:369] (1/2) 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 19:24:26,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2684, 1.2296, 1.0706, 1.3198], device='cuda:1'), covar=tensor([0.0790, 0.0345, 0.0347, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0125, 0.0128, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:1')
+2023-03-02 19:24:42,085 INFO [train.py:968] (1/2) Epoch 5, batch 26450, giga_loss[loss=0.3509, simple_loss=0.3959, pruned_loss=0.1529, over 28564.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3985, pruned_loss=0.147, over 5678802.67 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.395, pruned_loss=0.1448, over 5689779.00 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3987, pruned_loss=0.1467, over 5682322.46 frames. ], batch size: 336, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:24:47,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2295, 3.0768, 2.9438, 1.3438], device='cuda:1'), covar=tensor([0.0718, 0.0710, 0.0989, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0809, 0.0825, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:25:34,561 INFO [train.py:968] (1/2) Epoch 5, batch 26500, giga_loss[loss=0.3697, simple_loss=0.426, pruned_loss=0.1567, over 28840.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.3974, pruned_loss=0.1465, over 5674980.47 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.395, pruned_loss=0.1448, over 5692972.56 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3976, pruned_loss=0.1463, over 5674763.22 frames. ], batch size: 243, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:25:45,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4972, 1.8012, 1.8108, 1.6892], device='cuda:1'), covar=tensor([0.1457, 0.1802, 0.1098, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0747, 0.0781, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 19:25:47,549 INFO [optim.py:369] (1/2) 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 19:25:47,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3694, 1.3920, 1.4170, 1.3600], device='cuda:1'), covar=tensor([0.1011, 0.1425, 0.1601, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0755, 0.0642, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 19:25:58,270 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 26550, giga_loss[loss=0.3908, simple_loss=0.432, pruned_loss=0.1748, over 28957.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.3993, pruned_loss=0.1485, over 5682343.92 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3949, pruned_loss=0.1449, over 5698576.16 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.3996, pruned_loss=0.1483, over 5676522.86 frames. ], batch size: 213, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:26:27,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-02 19:27:03,466 INFO [train.py:968] (1/2) Epoch 5, batch 26600, giga_loss[loss=0.3347, simple_loss=0.3928, pruned_loss=0.1383, over 28982.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3979, pruned_loss=0.1487, over 5656846.67 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3952, pruned_loss=0.1452, over 5684244.07 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.398, pruned_loss=0.1484, over 5663689.86 frames. ], batch size: 213, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:27:14,065 INFO [optim.py:369] (1/2) 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 19:27:50,903 INFO [train.py:968] (1/2) Epoch 5, batch 26650, giga_loss[loss=0.3105, simple_loss=0.3669, pruned_loss=0.1271, over 28665.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3946, pruned_loss=0.1462, over 5654346.09 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3952, pruned_loss=0.145, over 5686826.81 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3947, pruned_loss=0.1461, over 5656904.93 frames. ], batch size: 85, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:28:05,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3028, 3.1686, 1.4897, 1.3367], device='cuda:1'), covar=tensor([0.0904, 0.0329, 0.0817, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0484, 0.0309, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 19:28:38,422 INFO [train.py:968] (1/2) Epoch 5, batch 26700, libri_loss[loss=0.3548, simple_loss=0.3887, pruned_loss=0.1604, over 29393.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3938, pruned_loss=0.1446, over 5657368.92 frames. ], libri_tot_loss[loss=0.3422, simple_loss=0.3947, pruned_loss=0.1448, over 5687346.14 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3942, pruned_loss=0.1447, over 5658056.44 frames. ], batch size: 71, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:28:48,243 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 26750, giga_loss[loss=0.352, simple_loss=0.4089, pruned_loss=0.1476, over 28588.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3962, pruned_loss=0.1455, over 5664096.70 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3947, pruned_loss=0.1449, over 5692733.38 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3965, pruned_loss=0.1455, over 5659197.87 frames. ], batch size: 307, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:29:30,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=208658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:30:11,921 INFO [train.py:968] (1/2) Epoch 5, batch 26800, giga_loss[loss=0.2982, simple_loss=0.3717, pruned_loss=0.1124, over 28980.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3969, pruned_loss=0.1466, over 5663491.54 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3948, pruned_loss=0.145, over 5695977.24 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3971, pruned_loss=0.1465, over 5656303.85 frames. ], batch size: 164, lr: 6.28e-03, grad_scale: 8.0
+2023-03-02 19:30:23,739 INFO [optim.py:369] (1/2) 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 19:30:26,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8452, 1.1293, 3.3778, 2.9715], device='cuda:1'), covar=tensor([0.1718, 0.2310, 0.0477, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0528, 0.0754, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 19:30:54,942 INFO [train.py:968] (1/2) Epoch 5, batch 26850, giga_loss[loss=0.3047, simple_loss=0.3926, pruned_loss=0.1084, over 28924.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.397, pruned_loss=0.1441, over 5682614.71 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3949, pruned_loss=0.1451, over 5702018.52 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3971, pruned_loss=0.144, over 5670552.65 frames. ], batch size: 112, lr: 6.28e-03, grad_scale: 8.0
+2023-03-02 19:31:45,052 INFO [train.py:968] (1/2) Epoch 5, batch 26900, giga_loss[loss=0.325, simple_loss=0.3972, pruned_loss=0.1264, over 28706.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3977, pruned_loss=0.1416, over 5678650.24 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.3952, pruned_loss=0.1454, over 5694108.48 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3976, pruned_loss=0.1412, over 5675981.12 frames. ], batch size: 262, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:31:46,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=208800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:31:49,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5490, 2.4238, 1.2632, 1.1350], device='cuda:1'), covar=tensor([0.1755, 0.0963, 0.1238, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1226, 0.1213, 0.1314], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 19:31:57,954 INFO [optim.py:369] (1/2) 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 19:32:31,112 INFO [train.py:968] (1/2) Epoch 5, batch 26950, giga_loss[loss=0.3787, simple_loss=0.4265, pruned_loss=0.1654, over 28988.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.4012, pruned_loss=0.1432, over 5684420.80 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3953, pruned_loss=0.1455, over 5697282.38 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.4011, pruned_loss=0.1428, over 5679208.41 frames. ], batch size: 213, lr: 6.28e-03, grad_scale: 4.0
+2023-03-02 19:32:32,057 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-02 19:32:36,997 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-02 19:33:17,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5806, 1.7004, 1.8363, 1.7188], device='cuda:1'), covar=tensor([0.1206, 0.1566, 0.0952, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0748, 0.0780, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 19:33:20,912 INFO [train.py:968] (1/2) Epoch 5, batch 27000, giga_loss[loss=0.3381, simple_loss=0.3991, pruned_loss=0.1386, over 28409.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.4051, pruned_loss=0.1473, over 5669128.51 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3959, pruned_loss=0.1461, over 5684590.05 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4046, pruned_loss=0.1465, over 5675529.07 frames. ], batch size: 71, lr: 6.28e-03, grad_scale: 2.0
+2023-03-02 19:33:20,912 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 19:33:26,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6121, 1.5453, 1.2290, 1.2954], device='cuda:1'), covar=tensor([0.0671, 0.0524, 0.0986, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0451, 0.0497, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:33:28,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0644, 1.4253, 3.3158, 3.0712], device='cuda:1'), covar=tensor([0.1622, 0.1985, 0.0443, 0.0600], device='cuda:1'), in_proj_covar=tensor([0.0577, 0.0530, 0.0765, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 19:33:29,451 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-02 19:33:41,039 INFO [optim.py:369] (1/2) 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 19:33:57,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9510, 1.3221, 3.6877, 2.9464], device='cuda:1'), covar=tensor([0.1654, 0.2041, 0.0440, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0527, 0.0760, 0.0612], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 19:34:07,049 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=208943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:34:10,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=208946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:34:14,833 INFO [train.py:968] (1/2) Epoch 5, batch 27050, libri_loss[loss=0.2986, simple_loss=0.3629, pruned_loss=0.1171, over 29587.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4068, pruned_loss=0.1503, over 5668126.77 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3957, pruned_loss=0.146, over 5693177.28 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.4069, pruned_loss=0.1498, over 5664559.63 frames. ], batch size: 75, lr: 6.28e-03, grad_scale: 2.0
+2023-03-02 19:34:20,066 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 19:34:35,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=208969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:34:41,200 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=208975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:34:59,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7452, 5.5735, 5.3430, 2.5351], device='cuda:1'), covar=tensor([0.0359, 0.0412, 0.0706, 0.1755], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0803, 0.0816, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:35:01,701 INFO [train.py:968] (1/2) Epoch 5, batch 27100, giga_loss[loss=0.3409, simple_loss=0.3976, pruned_loss=0.1421, over 28649.00 frames. ], tot_loss[loss=0.355, simple_loss=0.4067, pruned_loss=0.1516, over 5661053.02 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3956, pruned_loss=0.146, over 5699196.12 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4074, pruned_loss=0.1514, over 5651492.29 frames. ], batch size: 242, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:35:16,305 INFO [optim.py:369] (1/2) 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 19:35:18,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1615, 1.7054, 1.4213, 1.3148], device='cuda:1'), covar=tensor([0.0896, 0.0295, 0.0309, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0124, 0.0128, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0042, 0.0038, 0.0063], device='cuda:1')
+2023-03-02 19:35:25,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2285, 1.1585, 1.0738, 0.9667], device='cuda:1'), covar=tensor([0.0640, 0.0519, 0.0960, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0456, 0.0504, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:35:35,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:35:42,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-02 19:35:54,764 INFO [train.py:968] (1/2) Epoch 5, batch 27150, giga_loss[loss=0.4043, simple_loss=0.425, pruned_loss=0.1918, over 23519.00 frames. ], tot_loss[loss=0.355, simple_loss=0.4063, pruned_loss=0.1518, over 5650636.38 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3955, pruned_loss=0.1462, over 5699411.51 frames. ], giga_tot_loss[loss=0.3551, simple_loss=0.407, pruned_loss=0.1516, over 5642293.14 frames. ], batch size: 705, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:36:36,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-02 19:36:40,297 INFO [train.py:968] (1/2) Epoch 5, batch 27200, giga_loss[loss=0.3888, simple_loss=0.4252, pruned_loss=0.1762, over 28584.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4049, pruned_loss=0.1491, over 5656432.28 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3954, pruned_loss=0.1462, over 5699966.63 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4058, pruned_loss=0.149, over 5648411.66 frames. ], batch size: 336, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:36:51,162 INFO [optim.py:369] (1/2) 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 19:37:26,816 INFO [train.py:968] (1/2) Epoch 5, batch 27250, giga_loss[loss=0.3597, simple_loss=0.4195, pruned_loss=0.1499, over 28971.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4046, pruned_loss=0.1471, over 5670707.64 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3959, pruned_loss=0.1467, over 5703881.78 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4051, pruned_loss=0.1466, over 5659783.28 frames. ], batch size: 213, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:37:51,743 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:37:54,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:38:13,400 INFO [train.py:968] (1/2) Epoch 5, batch 27300, giga_loss[loss=0.3368, simple_loss=0.394, pruned_loss=0.1398, over 28675.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.4044, pruned_loss=0.1469, over 5668704.45 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3962, pruned_loss=0.1468, over 5705610.42 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4048, pruned_loss=0.1464, over 5657431.53 frames. ], batch size: 242, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:38:22,885 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:38:28,172 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 27350, giga_loss[loss=0.3747, simple_loss=0.4206, pruned_loss=0.1644, over 28997.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4041, pruned_loss=0.1471, over 5674899.42 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3961, pruned_loss=0.1467, over 5708732.92 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.4047, pruned_loss=0.1468, over 5662246.99 frames. ], batch size: 128, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:39:47,759 INFO [train.py:968] (1/2) Epoch 5, batch 27400, giga_loss[loss=0.3461, simple_loss=0.3978, pruned_loss=0.1472, over 29020.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.4014, pruned_loss=0.1455, over 5685715.42 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3962, pruned_loss=0.1468, over 5714241.22 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.4019, pruned_loss=0.1451, over 5669770.37 frames. ], batch size: 106, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:40:05,314 INFO [optim.py:369] (1/2) 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 19:40:31,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:40:36,424 INFO [train.py:968] (1/2) Epoch 5, batch 27450, giga_loss[loss=0.4149, simple_loss=0.4307, pruned_loss=0.1995, over 26609.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.3996, pruned_loss=0.1464, over 5667293.99 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.3958, pruned_loss=0.1466, over 5719247.73 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.4005, pruned_loss=0.1464, over 5648656.93 frames. ], batch size: 555, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:41:26,053 INFO [train.py:968] (1/2) Epoch 5, batch 27500, libri_loss[loss=0.393, simple_loss=0.4307, pruned_loss=0.1776, over 29571.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3991, pruned_loss=0.1467, over 5657185.66 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3964, pruned_loss=0.147, over 5720090.45 frames. ], giga_tot_loss[loss=0.346, simple_loss=0.3995, pruned_loss=0.1462, over 5639021.91 frames. ], batch size: 77, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:41:41,611 INFO [optim.py:369] (1/2) 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 19:42:14,672 INFO [train.py:968] (1/2) Epoch 5, batch 27550, libri_loss[loss=0.3231, simple_loss=0.3864, pruned_loss=0.1299, over 29757.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3975, pruned_loss=0.1465, over 5668309.67 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3963, pruned_loss=0.147, over 5724621.66 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3979, pruned_loss=0.1461, over 5648421.16 frames. ], batch size: 87, lr: 6.27e-03, grad_scale: 2.0
+2023-03-02 19:42:50,225 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=209487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:42:53,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=209490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:42:54,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5804, 1.9346, 1.8484, 1.6993], device='cuda:1'), covar=tensor([0.1477, 0.1750, 0.1061, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0752, 0.0778, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 19:43:00,261 INFO [train.py:968] (1/2) Epoch 5, batch 27600, giga_loss[loss=0.3742, simple_loss=0.4166, pruned_loss=0.1658, over 27922.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3975, pruned_loss=0.1471, over 5656075.29 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3965, pruned_loss=0.1471, over 5717361.28 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.3977, pruned_loss=0.1467, over 5645951.07 frames. ], batch size: 412, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:43:15,720 INFO [optim.py:369] (1/2) 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 19:43:18,506 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=209519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:43:19,413 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 19:43:37,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 19:43:39,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4186, 1.4229, 1.4786, 1.3650], device='cuda:1'), covar=tensor([0.0915, 0.1274, 0.1490, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0757, 0.0650, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 19:43:40,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5721, 2.1423, 1.9569, 1.7748], device='cuda:1'), covar=tensor([0.1681, 0.1909, 0.1244, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0759, 0.0781, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 19:43:44,549 INFO [train.py:968] (1/2) Epoch 5, batch 27650, giga_loss[loss=0.2961, simple_loss=0.3598, pruned_loss=0.1162, over 28674.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3956, pruned_loss=0.1453, over 5657719.16 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.396, pruned_loss=0.1467, over 5717601.21 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3961, pruned_loss=0.1453, over 5648269.77 frames. ], batch size: 92, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:44:02,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-02 19:44:25,075 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:44:28,248 INFO [train.py:968] (1/2) Epoch 5, batch 27700, giga_loss[loss=0.3161, simple_loss=0.3627, pruned_loss=0.1348, over 23825.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3919, pruned_loss=0.1408, over 5663684.66 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.3964, pruned_loss=0.147, over 5716870.14 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3919, pruned_loss=0.1403, over 5653713.32 frames. ], batch size: 705, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:44:38,537 INFO [optim.py:369] (1/2) 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 19:44:44,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4643, 3.6304, 1.5708, 1.4662], device='cuda:1'), covar=tensor([0.0810, 0.0312, 0.0793, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0485, 0.0310, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 19:45:09,946 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 27750, libri_loss[loss=0.3004, simple_loss=0.3555, pruned_loss=0.1227, over 29350.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3908, pruned_loss=0.139, over 5674747.10 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3962, pruned_loss=0.1469, over 5720632.71 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3907, pruned_loss=0.1385, over 5660516.08 frames. ], batch size: 67, lr: 6.27e-03, grad_scale: 4.0
+2023-03-02 19:45:47,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8831, 1.6267, 1.3507, 1.4602], device='cuda:1'), covar=tensor([0.0689, 0.0705, 0.0976, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0461, 0.0505, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:45:54,732 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 27800, giga_loss[loss=0.3421, simple_loss=0.3958, pruned_loss=0.1442, over 28719.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3913, pruned_loss=0.1403, over 5659514.88 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3964, pruned_loss=0.1469, over 5724473.94 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.391, pruned_loss=0.1397, over 5643520.78 frames. ], batch size: 99, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:46:20,524 INFO [optim.py:369] (1/2) 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 19:46:36,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0052, 1.5954, 1.3185, 1.4408], device='cuda:1'), covar=tensor([0.0672, 0.0764, 0.0982, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0457, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:47:00,385 INFO [train.py:968] (1/2) Epoch 5, batch 27850, giga_loss[loss=0.3086, simple_loss=0.3601, pruned_loss=0.1285, over 28785.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3866, pruned_loss=0.1376, over 5672863.39 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3963, pruned_loss=0.1469, over 5725531.57 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3863, pruned_loss=0.1371, over 5659170.24 frames. ], batch size: 99, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:47:50,373 INFO [train.py:968] (1/2) Epoch 5, batch 27900, giga_loss[loss=0.3265, simple_loss=0.4001, pruned_loss=0.1265, over 28990.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3884, pruned_loss=0.1394, over 5673790.98 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3964, pruned_loss=0.1469, over 5728965.69 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3879, pruned_loss=0.1389, over 5658328.37 frames. ], batch size: 155, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:48:05,073 INFO [optim.py:369] (1/2) 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 19:48:36,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6022, 1.9068, 1.5133, 1.1595], device='cuda:1'), covar=tensor([0.0999, 0.0743, 0.0556, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.1427, 0.1243, 0.1215, 0.1301], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 19:48:37,678 INFO [train.py:968] (1/2) Epoch 5, batch 27950, giga_loss[loss=0.3726, simple_loss=0.4203, pruned_loss=0.1625, over 28281.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3917, pruned_loss=0.1415, over 5655891.11 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3963, pruned_loss=0.1467, over 5720904.73 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3913, pruned_loss=0.141, over 5649795.56 frames. ], batch size: 77, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:48:40,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4868, 1.6923, 1.4181, 1.4913], device='cuda:1'), covar=tensor([0.0747, 0.0298, 0.0306, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0124, 0.0127, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:1')
+2023-03-02 19:48:46,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:49:25,256 INFO [train.py:968] (1/2) Epoch 5, batch 28000, giga_loss[loss=0.3079, simple_loss=0.3713, pruned_loss=0.1223, over 28954.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3924, pruned_loss=0.1419, over 5652778.37 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3961, pruned_loss=0.1466, over 5724703.15 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3921, pruned_loss=0.1415, over 5643359.76 frames. ], batch size: 199, lr: 6.26e-03, grad_scale: 8.0
+2023-03-02 19:49:28,159 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:49:39,612 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 19:50:10,079 INFO [train.py:968] (1/2) Epoch 5, batch 28050, giga_loss[loss=0.3157, simple_loss=0.3794, pruned_loss=0.1261, over 28445.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3928, pruned_loss=0.1419, over 5646380.45 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.397, pruned_loss=0.1472, over 5711112.41 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3917, pruned_loss=0.1409, over 5649594.31 frames. ], batch size: 65, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:50:15,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5533, 3.3560, 1.6441, 1.4360], device='cuda:1'), covar=tensor([0.0809, 0.0323, 0.0747, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0485, 0.0312, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 19:50:22,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2449, 1.5028, 1.3012, 1.3933], device='cuda:1'), covar=tensor([0.0788, 0.0308, 0.0320, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0124, 0.0127, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0042, 0.0037, 0.0063], device='cuda:1')
+2023-03-02 19:50:28,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=209970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:50:53,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=209999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:50:54,329 INFO [train.py:968] (1/2) Epoch 5, batch 28100, giga_loss[loss=0.3646, simple_loss=0.4072, pruned_loss=0.161, over 28593.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3935, pruned_loss=0.143, over 5638463.47 frames. ], libri_tot_loss[loss=0.3463, simple_loss=0.3975, pruned_loss=0.1476, over 5701502.93 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.392, pruned_loss=0.1418, over 5648363.79 frames. ], batch size: 307, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:51:07,311 INFO [optim.py:369] (1/2) 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 19:51:13,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 28150, giga_loss[loss=0.3843, simple_loss=0.4309, pruned_loss=0.1689, over 28593.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3955, pruned_loss=0.1444, over 5648538.02 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.397, pruned_loss=0.1474, over 5698990.41 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3947, pruned_loss=0.1435, over 5656810.84 frames. ], batch size: 336, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:51:54,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:52:09,375 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:52:25,964 INFO [train.py:968] (1/2) Epoch 5, batch 28200, giga_loss[loss=0.3673, simple_loss=0.4172, pruned_loss=0.1587, over 28777.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3973, pruned_loss=0.145, over 5656428.08 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3975, pruned_loss=0.1477, over 5701670.22 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3962, pruned_loss=0.1439, over 5660014.66 frames. ], batch size: 284, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:52:28,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5571, 1.9038, 1.8588, 1.7143], device='cuda:1'), covar=tensor([0.1504, 0.1675, 0.1099, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0751, 0.0776, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 19:52:40,144 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:52:42,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:53:15,197 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 28250, giga_loss[loss=0.3224, simple_loss=0.383, pruned_loss=0.1309, over 28191.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3983, pruned_loss=0.1458, over 5651945.87 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3975, pruned_loss=0.1477, over 5701372.23 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3975, pruned_loss=0.1449, over 5654465.63 frames. ], batch size: 65, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:53:29,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3242, 1.8333, 1.7474, 1.5367], device='cuda:1'), covar=tensor([0.1431, 0.1780, 0.1063, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0746, 0.0772, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-02 19:53:35,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:53:38,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:53:49,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-02 19:53:58,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5122, 3.3048, 3.2159, 1.8250], device='cuda:1'), covar=tensor([0.0674, 0.0680, 0.0857, 0.2116], device='cuda:1'), in_proj_covar=tensor([0.0880, 0.0809, 0.0822, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 19:54:08,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:12,002 INFO [train.py:968] (1/2) Epoch 5, batch 28300, giga_loss[loss=0.4265, simple_loss=0.4305, pruned_loss=0.2113, over 23564.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3996, pruned_loss=0.1478, over 5647211.73 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.3973, pruned_loss=0.1477, over 5704252.04 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.3991, pruned_loss=0.1471, over 5646108.36 frames. ], batch size: 705, lr: 6.26e-03, grad_scale: 2.0
+2023-03-02 19:54:22,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:28,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:31,092 INFO [optim.py:369] (1/2) 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 19:54:49,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:54:54,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:55:05,838 INFO [train.py:968] (1/2) Epoch 5, batch 28350, giga_loss[loss=0.3028, simple_loss=0.3731, pruned_loss=0.1163, over 27588.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3996, pruned_loss=0.1457, over 5655708.52 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3969, pruned_loss=0.1475, over 5706576.35 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3997, pruned_loss=0.1454, over 5651348.78 frames. ], batch size: 472, lr: 6.26e-03, grad_scale: 2.0
+2023-03-02 19:55:29,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:55:42,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4195, 3.2071, 1.4883, 1.4204], device='cuda:1'), covar=tensor([0.0834, 0.0316, 0.0822, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0487, 0.0315, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 19:55:46,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 19:55:48,946 INFO [train.py:968] (1/2) Epoch 5, batch 28400, libri_loss[loss=0.3189, simple_loss=0.3757, pruned_loss=0.1311, over 27911.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3993, pruned_loss=0.1452, over 5656851.88 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3968, pruned_loss=0.1476, over 5699726.18 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3996, pruned_loss=0.1447, over 5657866.32 frames. ], batch size: 116, lr: 6.26e-03, grad_scale: 4.0
+2023-03-02 19:56:09,109 INFO [optim.py:369] (1/2) 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 19:56:15,946 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 19:56:39,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9773, 1.0476, 3.8123, 3.0929], device='cuda:1'), covar=tensor([0.1632, 0.2303, 0.0407, 0.0675], device='cuda:1'), in_proj_covar=tensor([0.0576, 0.0535, 0.0770, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 19:56:41,023 INFO [train.py:968] (1/2) Epoch 5, batch 28450, giga_loss[loss=0.325, simple_loss=0.3896, pruned_loss=0.1302, over 28807.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3995, pruned_loss=0.1468, over 5663762.60 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.397, pruned_loss=0.1479, over 5702497.22 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3996, pruned_loss=0.1461, over 5661344.09 frames. ], batch size: 119, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:56:54,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-02 19:57:04,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:10,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:14,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:33,913 INFO [train.py:968] (1/2) Epoch 5, batch 28500, libri_loss[loss=0.4675, simple_loss=0.4788, pruned_loss=0.228, over 20325.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.399, pruned_loss=0.1469, over 5655832.29 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.3972, pruned_loss=0.1481, over 5689956.69 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.3989, pruned_loss=0.1461, over 5664381.51 frames. ], batch size: 186, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:57:44,851 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:57:53,996 INFO [optim.py:369] (1/2) 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 19:57:57,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2887, 2.0346, 2.0773, 1.9308], device='cuda:1'), covar=tensor([0.1083, 0.2139, 0.1530, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0742, 0.0640, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 19:57:59,051 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:01,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:12,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1161, 1.6157, 1.5507, 1.3786], device='cuda:1'), covar=tensor([0.1327, 0.1982, 0.1067, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0745, 0.0774, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 19:58:32,668 INFO [train.py:968] (1/2) Epoch 5, batch 28550, giga_loss[loss=0.331, simple_loss=0.3876, pruned_loss=0.1372, over 28979.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3963, pruned_loss=0.145, over 5656239.27 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3972, pruned_loss=0.1479, over 5685137.38 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3963, pruned_loss=0.1445, over 5666371.97 frames. ], batch size: 213, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:58:32,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=210450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:34,728 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:58:36,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:59:19,239 INFO [train.py:968] (1/2) Epoch 5, batch 28600, giga_loss[loss=0.3182, simple_loss=0.3806, pruned_loss=0.1279, over 28895.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3948, pruned_loss=0.144, over 5658747.78 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.397, pruned_loss=0.1478, over 5677066.37 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3949, pruned_loss=0.1436, over 5672849.47 frames. ], batch size: 145, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 19:59:37,481 INFO [optim.py:369] (1/2) 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 19:59:37,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 19:59:40,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:00:10,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:00:11,208 INFO [train.py:968] (1/2) Epoch 5, batch 28650, giga_loss[loss=0.3161, simple_loss=0.378, pruned_loss=0.1271, over 28851.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3952, pruned_loss=0.1451, over 5644245.95 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3974, pruned_loss=0.1482, over 5678051.03 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3949, pruned_loss=0.1445, over 5654124.69 frames. ], batch size: 174, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:01:01,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:01:01,716 INFO [train.py:968] (1/2) Epoch 5, batch 28700, giga_loss[loss=0.3483, simple_loss=0.3795, pruned_loss=0.1585, over 23443.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3973, pruned_loss=0.1472, over 5636343.17 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3978, pruned_loss=0.1486, over 5669569.91 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3967, pruned_loss=0.1463, over 5651839.75 frames. ], batch size: 705, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:01:04,092 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:01:35,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=210631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:01:45,388 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 20:01:49,738 INFO [train.py:968] (1/2) Epoch 5, batch 28750, giga_loss[loss=0.3471, simple_loss=0.4038, pruned_loss=0.1452, over 28701.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3988, pruned_loss=0.1488, over 5643070.46 frames. ], libri_tot_loss[loss=0.3478, simple_loss=0.3981, pruned_loss=0.1487, over 5672545.77 frames. ], giga_tot_loss[loss=0.347, simple_loss=0.3981, pruned_loss=0.1479, over 5651915.54 frames. ], batch size: 284, lr: 6.25e-03, grad_scale: 2.0
+2023-03-02 20:02:38,427 INFO [train.py:968] (1/2) Epoch 5, batch 28800, giga_loss[loss=0.3403, simple_loss=0.3989, pruned_loss=0.1409, over 28747.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4004, pruned_loss=0.1503, over 5632064.97 frames. ], libri_tot_loss[loss=0.3485, simple_loss=0.3987, pruned_loss=0.1492, over 5668200.75 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.3992, pruned_loss=0.1492, over 5642106.90 frames. ], batch size: 284, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:02:54,907 INFO [optim.py:369] (1/2) 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 20:03:19,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3198, 1.6781, 1.6049, 1.4977], device='cuda:1'), covar=tensor([0.1148, 0.1650, 0.0957, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0751, 0.0776, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 20:03:24,987 INFO [train.py:968] (1/2) Epoch 5, batch 28850, giga_loss[loss=0.3747, simple_loss=0.4143, pruned_loss=0.1675, over 29022.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.3999, pruned_loss=0.1505, over 5629064.86 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.3984, pruned_loss=0.1489, over 5666632.92 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.3993, pruned_loss=0.1499, over 5637414.07 frames. ], batch size: 145, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:04:03,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2725, 1.4497, 1.1948, 1.4502], device='cuda:1'), covar=tensor([0.0768, 0.0335, 0.0349, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0124, 0.0127, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0057, 0.0042, 0.0038, 0.0063], device='cuda:1')
+2023-03-02 20:04:09,592 INFO [train.py:968] (1/2) Epoch 5, batch 28900, giga_loss[loss=0.3391, simple_loss=0.3922, pruned_loss=0.1431, over 28271.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.3994, pruned_loss=0.1502, over 5645779.93 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3979, pruned_loss=0.1485, over 5670480.64 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.3995, pruned_loss=0.1502, over 5647867.24 frames. ], batch size: 368, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:04:23,910 INFO [optim.py:369] (1/2) 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 20:04:30,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=210825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:04:37,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3542, 1.7473, 1.2971, 0.6243], device='cuda:1'), covar=tensor([0.1683, 0.1080, 0.1162, 0.2310], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1334, 0.1381, 0.1167], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 20:04:40,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0923, 1.3412, 3.4336, 3.1456], device='cuda:1'), covar=tensor([0.1360, 0.1976, 0.0404, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0538, 0.0771, 0.0622], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 20:04:53,562 INFO [train.py:968] (1/2) Epoch 5, batch 28950, giga_loss[loss=0.3722, simple_loss=0.4206, pruned_loss=0.1619, over 28580.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.3991, pruned_loss=0.1503, over 5643173.79 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3967, pruned_loss=0.1476, over 5678467.22 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4003, pruned_loss=0.1512, over 5636942.99 frames. ], batch size: 92, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:05:42,277 INFO [train.py:968] (1/2) Epoch 5, batch 29000, giga_loss[loss=0.3464, simple_loss=0.4087, pruned_loss=0.142, over 28524.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4002, pruned_loss=0.1504, over 5642621.60 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.3972, pruned_loss=0.1481, over 5672121.23 frames. ], giga_tot_loss[loss=0.3511, simple_loss=0.4007, pruned_loss=0.1507, over 5641902.68 frames. ], batch size: 60, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:06:00,190 INFO [optim.py:369] (1/2) 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 20:06:30,194 INFO [train.py:968] (1/2) Epoch 5, batch 29050, giga_loss[loss=0.2826, simple_loss=0.3523, pruned_loss=0.1065, over 28527.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4003, pruned_loss=0.1498, over 5649433.42 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3972, pruned_loss=0.1483, over 5674827.37 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4008, pruned_loss=0.15, over 5646156.57 frames. ], batch size: 85, lr: 6.25e-03, grad_scale: 4.0
+2023-03-02 20:06:46,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9087, 3.7450, 3.6156, 1.8943], device='cuda:1'), covar=tensor([0.0540, 0.0645, 0.0764, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0820, 0.0828, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 20:06:48,349 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=210968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:06:50,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=210971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:07:14,384 INFO [train.py:968] (1/2) Epoch 5, batch 29100, giga_loss[loss=0.3363, simple_loss=0.3971, pruned_loss=0.1378, over 28297.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.401, pruned_loss=0.1502, over 5666028.09 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3975, pruned_loss=0.1485, over 5680632.62 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4012, pruned_loss=0.1501, over 5657739.03 frames. ], batch size: 65, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:07:14,700 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:07:31,607 INFO [optim.py:369] (1/2) 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 20:07:34,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-02 20:07:37,799 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:07:56,553 INFO [train.py:968] (1/2) Epoch 5, batch 29150, giga_loss[loss=0.3462, simple_loss=0.4071, pruned_loss=0.1426, over 28739.00 frames. ], tot_loss[loss=0.3527, simple_loss=0.4024, pruned_loss=0.1515, over 5673669.62 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3975, pruned_loss=0.1485, over 5680989.01 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4028, pruned_loss=0.1516, over 5666359.48 frames. ], batch size: 262, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:08:04,198 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:08:19,946 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:08:43,618 INFO [train.py:968] (1/2) Epoch 5, batch 29200, giga_loss[loss=0.353, simple_loss=0.407, pruned_loss=0.1495, over 28944.00 frames. ], tot_loss[loss=0.355, simple_loss=0.4042, pruned_loss=0.1529, over 5667299.83 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3976, pruned_loss=0.1485, over 5682960.93 frames. ], giga_tot_loss[loss=0.3552, simple_loss=0.4044, pruned_loss=0.153, over 5659763.15 frames. ], batch size: 227, lr: 6.24e-03, grad_scale: 8.0
+2023-03-02 20:08:53,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-02 20:09:03,247 INFO [optim.py:369] (1/2) 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 20:09:34,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0566, 4.7205, 1.8162, 1.9913], device='cuda:1'), covar=tensor([0.0763, 0.0266, 0.0836, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0485, 0.0311, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 20:09:37,470 INFO [train.py:968] (1/2) Epoch 5, batch 29250, giga_loss[loss=0.3639, simple_loss=0.4189, pruned_loss=0.1545, over 28328.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4036, pruned_loss=0.1505, over 5670221.07 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.3978, pruned_loss=0.1485, over 5684794.97 frames. ], giga_tot_loss[loss=0.3525, simple_loss=0.4037, pruned_loss=0.1506, over 5662366.54 frames. ], batch size: 368, lr: 6.24e-03, grad_scale: 8.0
+2023-03-02 20:09:45,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:10:22,833 INFO [train.py:968] (1/2) Epoch 5, batch 29300, giga_loss[loss=0.3004, simple_loss=0.3728, pruned_loss=0.114, over 28924.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.4015, pruned_loss=0.1482, over 5657064.73 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.3971, pruned_loss=0.1481, over 5676562.75 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4023, pruned_loss=0.1487, over 5657707.93 frames. ], batch size: 174, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:10:40,382 INFO [optim.py:369] (1/2) 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 20:11:08,606 INFO [train.py:968] (1/2) Epoch 5, batch 29350, giga_loss[loss=0.3987, simple_loss=0.4295, pruned_loss=0.184, over 27980.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.3995, pruned_loss=0.1471, over 5654544.63 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3973, pruned_loss=0.1482, over 5671669.79 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.4, pruned_loss=0.1473, over 5659408.75 frames. ], batch size: 412, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:11:52,654 INFO [train.py:968] (1/2) Epoch 5, batch 29400, giga_loss[loss=0.3717, simple_loss=0.419, pruned_loss=0.1622, over 28215.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.4, pruned_loss=0.1478, over 5644624.23 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.398, pruned_loss=0.1489, over 5665763.49 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.3998, pruned_loss=0.1474, over 5652521.11 frames. ], batch size: 368, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:11:53,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7850, 2.7797, 1.9321, 0.7724], device='cuda:1'), covar=tensor([0.3577, 0.1481, 0.1890, 0.3420], device='cuda:1'), in_proj_covar=tensor([0.1402, 0.1319, 0.1371, 0.1167], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 20:12:10,649 INFO [optim.py:369] (1/2) 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 20:12:37,401 INFO [train.py:968] (1/2) Epoch 5, batch 29450, giga_loss[loss=0.3659, simple_loss=0.4102, pruned_loss=0.1608, over 28209.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.4006, pruned_loss=0.148, over 5653366.77 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.3971, pruned_loss=0.148, over 5672314.09 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4015, pruned_loss=0.1484, over 5653459.85 frames. ], batch size: 368, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:13:06,609 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:13:27,066 INFO [train.py:968] (1/2) Epoch 5, batch 29500, giga_loss[loss=0.3626, simple_loss=0.4094, pruned_loss=0.1579, over 28552.00 frames. ], tot_loss[loss=0.3494, simple_loss=0.4012, pruned_loss=0.1488, over 5656100.60 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.3973, pruned_loss=0.1481, over 5676539.26 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.4017, pruned_loss=0.1491, over 5651932.69 frames. ], batch size: 336, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:13:27,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:13:36,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8746, 1.7348, 1.2545, 1.4087], device='cuda:1'), covar=tensor([0.0605, 0.0595, 0.1021, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0453, 0.0505, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 20:13:45,121 INFO [optim.py:369] (1/2) 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 20:14:00,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:14:13,332 INFO [train.py:968] (1/2) Epoch 5, batch 29550, libri_loss[loss=0.3504, simple_loss=0.4119, pruned_loss=0.1445, over 27785.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4012, pruned_loss=0.1498, over 5662172.37 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.3977, pruned_loss=0.1485, over 5680002.81 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4014, pruned_loss=0.1497, over 5654934.19 frames. ], batch size: 116, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:14:13,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:14:44,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3370, 2.9440, 1.3891, 1.3503], device='cuda:1'), covar=tensor([0.0813, 0.0334, 0.0803, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0486, 0.0309, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 20:14:58,873 INFO [train.py:968] (1/2) Epoch 5, batch 29600, giga_loss[loss=0.3282, simple_loss=0.3879, pruned_loss=0.1343, over 28795.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.4017, pruned_loss=0.1505, over 5644896.16 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.3981, pruned_loss=0.1488, over 5664610.52 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.4016, pruned_loss=0.1502, over 5652094.30 frames. ], batch size: 199, lr: 6.24e-03, grad_scale: 8.0
+2023-03-02 20:15:17,234 INFO [optim.py:369] (1/2) 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 20:15:19,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-02 20:15:20,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3260, 4.1322, 3.9918, 1.6534], device='cuda:1'), covar=tensor([0.0460, 0.0529, 0.0718, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0825, 0.0841, 0.0618], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-02 20:15:29,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:15:38,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:15:41,308 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 29650, giga_loss[loss=0.3196, simple_loss=0.38, pruned_loss=0.1296, over 28794.00 frames. ], tot_loss[loss=0.3514, simple_loss=0.402, pruned_loss=0.1504, over 5651455.27 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.3983, pruned_loss=0.1489, over 5667216.90 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.4017, pruned_loss=0.1501, over 5654436.50 frames. ], batch size: 119, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:16:09,872 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:11,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:15,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:24,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1641, 1.4224, 1.1185, 1.1998], device='cuda:1'), covar=tensor([0.2087, 0.2029, 0.2103, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.1130, 0.0884, 0.1001, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 20:16:26,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:30,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:16:33,221 INFO [train.py:968] (1/2) Epoch 5, batch 29700, giga_loss[loss=0.3324, simple_loss=0.3931, pruned_loss=0.1358, over 28947.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4029, pruned_loss=0.1512, over 5651051.00 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.3985, pruned_loss=0.1489, over 5671220.91 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4027, pruned_loss=0.151, over 5649431.34 frames. ], batch size: 112, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:16:39,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9268, 2.5880, 1.5863, 1.2958], device='cuda:1'), covar=tensor([0.1271, 0.0797, 0.1034, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.1418, 0.1267, 0.1226, 0.1308], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 20:16:43,938 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:16:57,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:17:24,470 INFO [train.py:968] (1/2) Epoch 5, batch 29750, giga_loss[loss=0.3779, simple_loss=0.421, pruned_loss=0.1674, over 28507.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.4002, pruned_loss=0.1479, over 5670369.17 frames. ], libri_tot_loss[loss=0.3479, simple_loss=0.3984, pruned_loss=0.1487, over 5674978.78 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4002, pruned_loss=0.148, over 5665641.41 frames. ], batch size: 307, lr: 6.24e-03, grad_scale: 4.0
+2023-03-02 20:17:48,041 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:17:51,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:18:10,490 INFO [train.py:968] (1/2) Epoch 5, batch 29800, giga_loss[loss=0.4401, simple_loss=0.4416, pruned_loss=0.2193, over 23932.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.3998, pruned_loss=0.1475, over 5664808.98 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.3984, pruned_loss=0.1488, over 5681498.98 frames. ], giga_tot_loss[loss=0.3473, simple_loss=0.3997, pruned_loss=0.1475, over 5654675.89 frames. ], batch size: 705, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:18:17,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:18:29,152 INFO [optim.py:369] (1/2) 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 20:18:49,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8591, 1.7122, 1.2755, 1.3969], device='cuda:1'), covar=tensor([0.0616, 0.0594, 0.0946, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0456, 0.0504, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 20:18:55,497 INFO [train.py:968] (1/2) Epoch 5, batch 29850, giga_loss[loss=0.3685, simple_loss=0.4138, pruned_loss=0.1616, over 28754.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.398, pruned_loss=0.1455, over 5668882.26 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.3981, pruned_loss=0.1483, over 5681726.26 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.3982, pruned_loss=0.1457, over 5660178.06 frames. ], batch size: 243, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:18:59,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=211752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:19:23,806 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=211779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:19:42,095 INFO [train.py:968] (1/2) Epoch 5, batch 29900, giga_loss[loss=0.3468, simple_loss=0.4008, pruned_loss=0.1464, over 28905.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3969, pruned_loss=0.1453, over 5666299.17 frames. ], libri_tot_loss[loss=0.3472, simple_loss=0.3979, pruned_loss=0.1483, over 5684143.45 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3972, pruned_loss=0.1454, over 5656968.36 frames. ], batch size: 174, lr: 6.23e-03, grad_scale: 2.0
+2023-03-02 20:19:45,601 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:20:07,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2606, 1.4759, 1.2075, 1.4676], device='cuda:1'), covar=tensor([0.1940, 0.1867, 0.1834, 0.1803], device='cuda:1'), in_proj_covar=tensor([0.1127, 0.0884, 0.1001, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 20:20:17,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-02 20:20:25,976 INFO [train.py:968] (1/2) Epoch 5, batch 29950, giga_loss[loss=0.3656, simple_loss=0.3941, pruned_loss=0.1685, over 23592.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.395, pruned_loss=0.1443, over 5664824.54 frames. ], libri_tot_loss[loss=0.347, simple_loss=0.3977, pruned_loss=0.1481, over 5687488.21 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3953, pruned_loss=0.1444, over 5654169.27 frames. ], batch size: 705, lr: 6.23e-03, grad_scale: 2.0
+2023-03-02 20:21:01,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1711, 0.9774, 0.7984, 1.4394], device='cuda:1'), covar=tensor([0.0783, 0.0324, 0.0375, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0124, 0.0129, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0058, 0.0042, 0.0038, 0.0064], device='cuda:1')
+2023-03-02 20:21:06,776 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=211895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:21:13,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=211898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:21:13,888 INFO [train.py:968] (1/2) Epoch 5, batch 30000, giga_loss[loss=0.3766, simple_loss=0.4124, pruned_loss=0.1704, over 28267.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3919, pruned_loss=0.1425, over 5661916.84 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3981, pruned_loss=0.1485, over 5681188.29 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3918, pruned_loss=0.1422, over 5658837.36 frames. ], batch size: 368, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:21:13,889 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 20:21:22,279 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-02 20:21:25,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5363, 1.7410, 1.3003, 1.0518], device='cuda:1'), covar=tensor([0.1229, 0.0913, 0.0795, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1267, 0.1227, 0.1308], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 20:21:26,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5057, 3.1812, 1.4923, 1.4761], device='cuda:1'), covar=tensor([0.0811, 0.0357, 0.0807, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0487, 0.0311, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 20:21:39,581 INFO [optim.py:369] (1/2) 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 20:21:44,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=211927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:22:04,579 INFO [train.py:968] (1/2) Epoch 5, batch 30050, giga_loss[loss=0.3636, simple_loss=0.4151, pruned_loss=0.156, over 28942.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3898, pruned_loss=0.1416, over 5677779.47 frames. ], libri_tot_loss[loss=0.3478, simple_loss=0.3984, pruned_loss=0.1486, over 5684083.60 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3893, pruned_loss=0.1411, over 5672624.08 frames. ], batch size: 164, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:22:54,818 INFO [train.py:968] (1/2) Epoch 5, batch 30100, giga_loss[loss=0.3066, simple_loss=0.3734, pruned_loss=0.12, over 29012.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3889, pruned_loss=0.1415, over 5691291.85 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3983, pruned_loss=0.1484, over 5687440.37 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3885, pruned_loss=0.1413, over 5684161.22 frames. ], batch size: 155, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:23:12,515 INFO [optim.py:369] (1/2) 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 20:23:43,230 INFO [train.py:968] (1/2) Epoch 5, batch 30150, giga_loss[loss=0.3512, simple_loss=0.408, pruned_loss=0.1472, over 28932.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3881, pruned_loss=0.1403, over 5690353.68 frames. ], libri_tot_loss[loss=0.3472, simple_loss=0.3982, pruned_loss=0.1481, over 5691274.72 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3877, pruned_loss=0.1402, over 5681630.32 frames. ], batch size: 174, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:24:30,192 INFO [train.py:968] (1/2) Epoch 5, batch 30200, giga_loss[loss=0.3067, simple_loss=0.3753, pruned_loss=0.1191, over 28946.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3864, pruned_loss=0.1376, over 5676112.17 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3981, pruned_loss=0.1483, over 5685299.89 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3858, pruned_loss=0.1372, over 5674636.69 frames. ], batch size: 213, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:24:52,755 INFO [optim.py:369] (1/2) 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 20:24:54,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1138, 0.8997, 0.7245, 1.2501], device='cuda:1'), covar=tensor([0.0807, 0.0368, 0.0384, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0127, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0058, 0.0042, 0.0038, 0.0064], device='cuda:1')
+2023-03-02 20:25:24,165 INFO [train.py:968] (1/2) Epoch 5, batch 30250, giga_loss[loss=0.2928, simple_loss=0.3673, pruned_loss=0.1091, over 28844.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3831, pruned_loss=0.1337, over 5660705.30 frames. ], libri_tot_loss[loss=0.3466, simple_loss=0.3975, pruned_loss=0.1479, over 5679950.08 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.383, pruned_loss=0.1334, over 5663704.13 frames. ], batch size: 227, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:25:29,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:25:54,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:26:15,064 INFO [train.py:968] (1/2) Epoch 5, batch 30300, libri_loss[loss=0.3227, simple_loss=0.3655, pruned_loss=0.14, over 29551.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.378, pruned_loss=0.1288, over 5655506.74 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.3973, pruned_loss=0.1478, over 5679491.06 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1284, over 5657601.68 frames. ], batch size: 78, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:26:36,630 INFO [optim.py:369] (1/2) 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 20:27:04,351 INFO [train.py:968] (1/2) Epoch 5, batch 30350, giga_loss[loss=0.2881, simple_loss=0.3606, pruned_loss=0.1078, over 29104.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5655031.01 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3963, pruned_loss=0.1472, over 5684020.18 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3738, pruned_loss=0.1246, over 5652085.28 frames. ], batch size: 155, lr: 6.23e-03, grad_scale: 4.0
+2023-03-02 20:27:36,725 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-02 20:27:52,094 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:27:53,681 INFO [train.py:968] (1/2) Epoch 5, batch 30400, giga_loss[loss=0.2663, simple_loss=0.356, pruned_loss=0.08824, over 28907.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3716, pruned_loss=0.1212, over 5646048.35 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3963, pruned_loss=0.1473, over 5675709.55 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3715, pruned_loss=0.1207, over 5650618.20 frames. ], batch size: 164, lr: 6.23e-03, grad_scale: 8.0
+2023-03-02 20:27:54,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:00,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2151, 2.4520, 1.1800, 1.2814], device='cuda:1'), covar=tensor([0.0880, 0.0325, 0.0896, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0480, 0.0312, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 20:28:16,995 INFO [optim.py:369] (1/2) 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 20:28:21,142 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=212323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:23,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=212326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:27,680 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:28:50,037 INFO [train.py:968] (1/2) Epoch 5, batch 30450, giga_loss[loss=0.357, simple_loss=0.4015, pruned_loss=0.1562, over 26673.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3705, pruned_loss=0.1196, over 5633846.27 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.3964, pruned_loss=0.1475, over 5678076.94 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3701, pruned_loss=0.1187, over 5635045.60 frames. ], batch size: 555, lr: 6.23e-03, grad_scale: 8.0
+2023-03-02 20:28:55,580 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=212355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:29:42,638 INFO [train.py:968] (1/2) Epoch 5, batch 30500, giga_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 27617.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3698, pruned_loss=0.1193, over 5636811.03 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3957, pruned_loss=0.1473, over 5682564.12 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3696, pruned_loss=0.1183, over 5632976.92 frames. ], batch size: 472, lr: 6.22e-03, grad_scale: 8.0
+2023-03-02 20:30:06,424 INFO [optim.py:369] (1/2) 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 20:30:34,617 INFO [train.py:968] (1/2) Epoch 5, batch 30550, giga_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 28872.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3662, pruned_loss=0.1165, over 5628461.05 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3955, pruned_loss=0.1471, over 5674617.15 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.366, pruned_loss=0.1155, over 5632573.70 frames. ], batch size: 227, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:30:56,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4411, 2.9967, 1.4577, 1.4434], device='cuda:1'), covar=tensor([0.0771, 0.0330, 0.0806, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0480, 0.0310, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 20:31:14,091 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7867, 1.9527, 1.7876, 1.7544], device='cuda:1'), covar=tensor([0.1013, 0.1536, 0.1361, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0728, 0.0628, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 20:31:19,412 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212497.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 20:31:22,904 INFO [train.py:968] (1/2) Epoch 5, batch 30600, giga_loss[loss=0.3443, simple_loss=0.407, pruned_loss=0.1409, over 28798.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3651, pruned_loss=0.1166, over 5636594.44 frames. ], libri_tot_loss[loss=0.3442, simple_loss=0.3948, pruned_loss=0.1468, over 5680245.65 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3645, pruned_loss=0.115, over 5633392.79 frames. ], batch size: 243, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:31:45,342 INFO [optim.py:369] (1/2) 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 20:32:08,331 INFO [train.py:968] (1/2) Epoch 5, batch 30650, giga_loss[loss=0.2871, simple_loss=0.3565, pruned_loss=0.1089, over 27866.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3642, pruned_loss=0.1158, over 5635685.70 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.3938, pruned_loss=0.1463, over 5676621.90 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3638, pruned_loss=0.1142, over 5635240.43 frames. ], batch size: 412, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:32:58,480 INFO [train.py:968] (1/2) Epoch 5, batch 30700, giga_loss[loss=0.3546, simple_loss=0.403, pruned_loss=0.1532, over 27943.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3642, pruned_loss=0.1156, over 5644489.06 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3931, pruned_loss=0.146, over 5679857.87 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.364, pruned_loss=0.1139, over 5640457.12 frames. ], batch size: 412, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:33:18,665 INFO [optim.py:369] (1/2) 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 20:33:42,741 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:33:45,241 INFO [train.py:968] (1/2) Epoch 5, batch 30750, giga_loss[loss=0.2935, simple_loss=0.3695, pruned_loss=0.1088, over 28783.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3623, pruned_loss=0.1141, over 5650673.88 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3929, pruned_loss=0.1462, over 5683422.52 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3612, pruned_loss=0.1115, over 5643013.94 frames. ], batch size: 284, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:34:14,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-02 20:34:33,887 INFO [train.py:968] (1/2) Epoch 5, batch 30800, giga_loss[loss=0.2734, simple_loss=0.3462, pruned_loss=0.1003, over 28986.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3575, pruned_loss=0.1106, over 5639849.46 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3921, pruned_loss=0.1459, over 5679473.85 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3565, pruned_loss=0.108, over 5636195.02 frames. ], batch size: 213, lr: 6.22e-03, grad_scale: 8.0
+2023-03-02 20:34:55,311 INFO [optim.py:369] (1/2) 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 20:35:18,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4575, 1.5746, 1.3549, 1.8314], device='cuda:1'), covar=tensor([0.2210, 0.1951, 0.1910, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.1130, 0.0870, 0.1003, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 20:35:22,924 INFO [train.py:968] (1/2) Epoch 5, batch 30850, libri_loss[loss=0.3052, simple_loss=0.3584, pruned_loss=0.126, over 29553.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3541, pruned_loss=0.1091, over 5641731.67 frames. ], libri_tot_loss[loss=0.3408, simple_loss=0.391, pruned_loss=0.1452, over 5679231.02 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3528, pruned_loss=0.106, over 5637592.05 frames. ], batch size: 79, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:35:41,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-02 20:35:47,067 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-02 20:36:12,872 INFO [train.py:968] (1/2) Epoch 5, batch 30900, giga_loss[loss=0.2663, simple_loss=0.3378, pruned_loss=0.0974, over 28558.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3531, pruned_loss=0.1091, over 5638337.32 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3908, pruned_loss=0.1453, over 5672684.09 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3518, pruned_loss=0.1061, over 5640206.14 frames. ], batch size: 336, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:36:36,596 INFO [optim.py:369] (1/2) 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 20:36:46,693 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=212832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:37:04,052 INFO [train.py:968] (1/2) Epoch 5, batch 30950, giga_loss[loss=0.2904, simple_loss=0.3627, pruned_loss=0.1091, over 27977.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3512, pruned_loss=0.1082, over 5614539.51 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3898, pruned_loss=0.1448, over 5666497.59 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3498, pruned_loss=0.105, over 5620730.52 frames. ], batch size: 412, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:37:27,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=212872.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 20:37:52,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0849, 2.1370, 1.9570, 1.8489], device='cuda:1'), covar=tensor([0.0960, 0.1798, 0.1299, 0.1427], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0717, 0.0615, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 20:38:00,463 INFO [train.py:968] (1/2) Epoch 5, batch 31000, giga_loss[loss=0.3111, simple_loss=0.3849, pruned_loss=0.1187, over 28810.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3547, pruned_loss=0.1093, over 5620749.52 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3895, pruned_loss=0.1447, over 5662169.64 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3531, pruned_loss=0.1061, over 5628393.46 frames. ], batch size: 227, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:38:24,297 INFO [optim.py:369] (1/2) 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 20:38:59,058 INFO [train.py:968] (1/2) Epoch 5, batch 31050, giga_loss[loss=0.2959, simple_loss=0.3652, pruned_loss=0.1133, over 28882.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3573, pruned_loss=0.11, over 5627378.04 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3891, pruned_loss=0.1447, over 5656855.77 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3559, pruned_loss=0.1069, over 5637268.52 frames. ], batch size: 186, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:39:58,556 INFO [train.py:968] (1/2) Epoch 5, batch 31100, giga_loss[loss=0.272, simple_loss=0.3534, pruned_loss=0.09532, over 29003.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3585, pruned_loss=0.1104, over 5644756.26 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.388, pruned_loss=0.1439, over 5653669.34 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3572, pruned_loss=0.1074, over 5654853.11 frames. ], batch size: 155, lr: 6.22e-03, grad_scale: 4.0
+2023-03-02 20:40:15,741 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 20:40:19,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213018.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 20:40:22,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:40:23,661 INFO [optim.py:369] (1/2) 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 20:40:53,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213047.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 20:40:55,539 INFO [train.py:968] (1/2) Epoch 5, batch 31150, giga_loss[loss=0.2635, simple_loss=0.3397, pruned_loss=0.09363, over 28613.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3561, pruned_loss=0.1092, over 5650367.97 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3875, pruned_loss=0.1437, over 5657346.71 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3544, pruned_loss=0.1058, over 5655170.23 frames. ], batch size: 307, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:41:43,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:41:47,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3737, 1.5286, 1.2942, 1.6637], device='cuda:1'), covar=tensor([0.2323, 0.2088, 0.2155, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.1128, 0.0867, 0.1004, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 20:41:52,401 INFO [train.py:968] (1/2) Epoch 5, batch 31200, giga_loss[loss=0.2666, simple_loss=0.3476, pruned_loss=0.09282, over 28815.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3548, pruned_loss=0.108, over 5659442.06 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3861, pruned_loss=0.1429, over 5668274.59 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3531, pruned_loss=0.1041, over 5652572.83 frames. ], batch size: 119, lr: 6.21e-03, grad_scale: 8.0
+2023-03-02 20:41:56,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:42:20,557 INFO [optim.py:369] (1/2) 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 20:42:49,851 INFO [train.py:968] (1/2) Epoch 5, batch 31250, giga_loss[loss=0.2953, simple_loss=0.3629, pruned_loss=0.1138, over 28712.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3538, pruned_loss=0.1066, over 5653452.80 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3861, pruned_loss=0.143, over 5661160.28 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3517, pruned_loss=0.1026, over 5654676.94 frames. ], batch size: 243, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:43:05,965 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:09,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:09,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:32,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1525, 1.5320, 1.1873, 0.9600], device='cuda:1'), covar=tensor([0.1161, 0.0787, 0.0545, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.1401, 0.1192, 0.1157, 0.1244], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 20:43:42,884 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:43:46,925 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0839, 1.3072, 1.0209, 0.9469], device='cuda:1'), covar=tensor([0.0858, 0.0681, 0.0494, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.1401, 0.1193, 0.1157, 0.1243], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 20:43:48,325 INFO [train.py:968] (1/2) Epoch 5, batch 31300, giga_loss[loss=0.2512, simple_loss=0.3246, pruned_loss=0.08888, over 28491.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3495, pruned_loss=0.105, over 5664609.69 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3849, pruned_loss=0.1423, over 5666957.76 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3482, pruned_loss=0.1016, over 5660524.97 frames. ], batch size: 369, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:43:55,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6263, 1.6905, 1.6161, 1.5242], device='cuda:1'), covar=tensor([0.0764, 0.1047, 0.1186, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0728, 0.0623, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 20:43:59,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:44:19,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-02 20:44:20,005 INFO [optim.py:369] (1/2) 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 20:44:48,586 INFO [train.py:968] (1/2) Epoch 5, batch 31350, giga_loss[loss=0.2484, simple_loss=0.3242, pruned_loss=0.08627, over 28747.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3503, pruned_loss=0.1061, over 5665132.41 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3851, pruned_loss=0.1426, over 5672444.85 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3481, pruned_loss=0.1021, over 5656700.73 frames. ], batch size: 243, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:45:48,083 INFO [train.py:968] (1/2) Epoch 5, batch 31400, giga_loss[loss=0.2756, simple_loss=0.3471, pruned_loss=0.102, over 27550.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3506, pruned_loss=0.106, over 5660348.29 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3849, pruned_loss=0.1425, over 5664727.03 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3488, pruned_loss=0.1027, over 5659751.98 frames. ], batch size: 472, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:46:12,861 INFO [optim.py:369] (1/2) 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 20:46:25,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-02 20:46:42,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5461, 2.3170, 1.6638, 0.6800], device='cuda:1'), covar=tensor([0.2154, 0.1185, 0.1966, 0.2502], device='cuda:1'), in_proj_covar=tensor([0.1379, 0.1299, 0.1365, 0.1155], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 20:46:45,725 INFO [train.py:968] (1/2) Epoch 5, batch 31450, giga_loss[loss=0.2999, simple_loss=0.3811, pruned_loss=0.1094, over 28901.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3522, pruned_loss=0.1059, over 5665462.79 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3841, pruned_loss=0.1419, over 5671687.30 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3504, pruned_loss=0.1025, over 5658581.75 frames. ], batch size: 186, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:46:46,016 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:46:49,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:47:04,784 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-02 20:47:27,159 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:47:46,856 INFO [train.py:968] (1/2) Epoch 5, batch 31500, giga_loss[loss=0.2397, simple_loss=0.323, pruned_loss=0.07818, over 28974.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3526, pruned_loss=0.1059, over 5662073.90 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3835, pruned_loss=0.1418, over 5667652.58 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3509, pruned_loss=0.1026, over 5660056.53 frames. ], batch size: 136, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:48:16,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-02 20:48:21,712 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 31550, giga_loss[loss=0.2757, simple_loss=0.3426, pruned_loss=0.1044, over 28203.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3498, pruned_loss=0.1042, over 5662987.91 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3834, pruned_loss=0.1417, over 5666509.73 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3483, pruned_loss=0.1013, over 5662120.41 frames. ], batch size: 412, lr: 6.21e-03, grad_scale: 2.0
+2023-03-02 20:49:20,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:49:38,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:50:05,810 INFO [train.py:968] (1/2) Epoch 5, batch 31600, giga_loss[loss=0.3473, simple_loss=0.417, pruned_loss=0.1389, over 28398.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3528, pruned_loss=0.1058, over 5656963.17 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3834, pruned_loss=0.1418, over 5660669.91 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3511, pruned_loss=0.1028, over 5660815.67 frames. ], batch size: 368, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:50:37,330 INFO [optim.py:369] (1/2) 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 20:51:00,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=213542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:51:09,840 INFO [train.py:968] (1/2) Epoch 5, batch 31650, giga_loss[loss=0.2395, simple_loss=0.3355, pruned_loss=0.07171, over 28489.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3562, pruned_loss=0.1047, over 5651234.03 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3833, pruned_loss=0.1417, over 5661359.00 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3547, pruned_loss=0.1021, over 5653527.56 frames. ], batch size: 370, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:52:11,824 INFO [train.py:968] (1/2) Epoch 5, batch 31700, libri_loss[loss=0.2583, simple_loss=0.3222, pruned_loss=0.09717, over 28537.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3567, pruned_loss=0.1036, over 5660040.33 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3823, pruned_loss=0.1411, over 5666234.98 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3557, pruned_loss=0.101, over 5657103.90 frames. ], batch size: 63, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:52:20,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:52:25,976 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:52:37,720 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:52:39,938 INFO [optim.py:369] (1/2) 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 20:52:40,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:01,543 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 31750, giga_loss[loss=0.2962, simple_loss=0.3595, pruned_loss=0.1164, over 27095.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3555, pruned_loss=0.102, over 5636276.67 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3826, pruned_loss=0.1414, over 5649430.15 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3541, pruned_loss=0.09923, over 5648679.94 frames. ], batch size: 555, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:53:16,066 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=213654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:51,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=213685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:53:54,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=213688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 20:54:05,082 INFO [train.py:968] (1/2) Epoch 5, batch 31800, giga_loss[loss=0.2641, simple_loss=0.3459, pruned_loss=0.09118, over 28855.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3558, pruned_loss=0.103, over 5640989.86 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3819, pruned_loss=0.1412, over 5645901.86 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3544, pruned_loss=0.09957, over 5653781.27 frames. ], batch size: 227, lr: 6.21e-03, grad_scale: 4.0
+2023-03-02 20:54:21,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3144, 1.8564, 1.6271, 1.5055], device='cuda:1'), covar=tensor([0.1403, 0.1799, 0.1142, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0726, 0.0775, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 20:54:26,007 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:55:04,356 INFO [train.py:968] (1/2) Epoch 5, batch 31850, giga_loss[loss=0.2726, simple_loss=0.3438, pruned_loss=0.1007, over 28153.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3565, pruned_loss=0.1053, over 5633801.41 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3815, pruned_loss=0.141, over 5642689.17 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3548, pruned_loss=0.1013, over 5647403.34 frames. ], batch size: 412, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:56:15,783 INFO [train.py:968] (1/2) Epoch 5, batch 31900, giga_loss[loss=0.3106, simple_loss=0.3786, pruned_loss=0.1213, over 28488.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3587, pruned_loss=0.1076, over 5643891.60 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3818, pruned_loss=0.1413, over 5638521.37 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3564, pruned_loss=0.1032, over 5658777.86 frames. ], batch size: 336, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:56:58,746 INFO [optim.py:369] (1/2) 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 20:57:31,498 INFO [train.py:968] (1/2) Epoch 5, batch 31950, giga_loss[loss=0.2472, simple_loss=0.3068, pruned_loss=0.09382, over 24295.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3563, pruned_loss=0.1064, over 5645866.44 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3815, pruned_loss=0.1412, over 5631101.58 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3544, pruned_loss=0.1025, over 5664619.37 frames. ], batch size: 705, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:58:05,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2825, 3.1324, 2.9395, 1.4008], device='cuda:1'), covar=tensor([0.0700, 0.0640, 0.0921, 0.2060], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0778, 0.0782, 0.0588], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 20:58:38,288 INFO [train.py:968] (1/2) Epoch 5, batch 32000, giga_loss[loss=0.267, simple_loss=0.3452, pruned_loss=0.09442, over 28772.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3532, pruned_loss=0.1044, over 5648007.78 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3816, pruned_loss=0.1413, over 5628359.39 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1004, over 5665362.83 frames. ], batch size: 243, lr: 6.20e-03, grad_scale: 8.0
+2023-03-02 20:59:09,632 INFO [optim.py:369] (1/2) 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 20:59:41,814 INFO [train.py:968] (1/2) Epoch 5, batch 32050, giga_loss[loss=0.2373, simple_loss=0.3131, pruned_loss=0.08074, over 29043.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3499, pruned_loss=0.1027, over 5652584.93 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3811, pruned_loss=0.1409, over 5635484.37 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3478, pruned_loss=0.09895, over 5660378.64 frames. ], batch size: 100, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 20:59:52,941 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=213959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:00:18,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3830, 3.2463, 1.3728, 1.4079], device='cuda:1'), covar=tensor([0.0849, 0.0261, 0.0820, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0478, 0.0311, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 21:00:46,911 INFO [train.py:968] (1/2) Epoch 5, batch 32100, giga_loss[loss=0.281, simple_loss=0.3662, pruned_loss=0.09794, over 28973.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3525, pruned_loss=0.1041, over 5659988.33 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3809, pruned_loss=0.1407, over 5637914.88 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3509, pruned_loss=0.1009, over 5664233.80 frames. ], batch size: 164, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:01:20,400 INFO [optim.py:369] (1/2) 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 21:01:48,321 INFO [train.py:968] (1/2) Epoch 5, batch 32150, giga_loss[loss=0.3238, simple_loss=0.3873, pruned_loss=0.1302, over 28722.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3563, pruned_loss=0.1063, over 5666497.28 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3811, pruned_loss=0.1411, over 5643628.18 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3541, pruned_loss=0.1027, over 5665247.86 frames. ], batch size: 99, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:02:51,224 INFO [train.py:968] (1/2) Epoch 5, batch 32200, giga_loss[loss=0.2572, simple_loss=0.3161, pruned_loss=0.09912, over 24617.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3534, pruned_loss=0.1059, over 5666480.83 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3806, pruned_loss=0.1408, over 5652803.49 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3513, pruned_loss=0.102, over 5657648.73 frames. ], batch size: 705, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:03:22,895 INFO [optim.py:369] (1/2) 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 21:03:50,703 INFO [train.py:968] (1/2) Epoch 5, batch 32250, giga_loss[loss=0.2758, simple_loss=0.3261, pruned_loss=0.1127, over 24401.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3542, pruned_loss=0.1072, over 5667220.29 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3805, pruned_loss=0.1408, over 5655340.87 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3522, pruned_loss=0.1036, over 5658046.23 frames. ], batch size: 705, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:04:44,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 21:04:58,247 INFO [train.py:968] (1/2) Epoch 5, batch 32300, giga_loss[loss=0.2411, simple_loss=0.3058, pruned_loss=0.08818, over 24592.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3544, pruned_loss=0.1069, over 5671764.18 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.38, pruned_loss=0.1405, over 5659260.81 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3529, pruned_loss=0.1039, over 5661293.74 frames. ], batch size: 705, lr: 6.20e-03, grad_scale: 2.0
+2023-03-02 21:05:34,171 INFO [optim.py:369] (1/2) 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 21:06:09,181 INFO [train.py:968] (1/2) Epoch 5, batch 32350, giga_loss[loss=0.2602, simple_loss=0.3401, pruned_loss=0.09009, over 28965.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3555, pruned_loss=0.106, over 5676334.71 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3797, pruned_loss=0.1403, over 5662961.58 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1034, over 5665052.42 frames. ], batch size: 227, lr: 6.20e-03, grad_scale: 2.0
+2023-03-02 21:07:20,830 INFO [train.py:968] (1/2) Epoch 5, batch 32400, giga_loss[loss=0.2912, simple_loss=0.3612, pruned_loss=0.1106, over 28139.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5678293.45 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3791, pruned_loss=0.1398, over 5670210.59 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3535, pruned_loss=0.1024, over 5663091.53 frames. ], batch size: 412, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:07:56,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-02 21:07:58,266 INFO [optim.py:369] (1/2) 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 21:08:09,176 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:08:27,734 INFO [train.py:968] (1/2) Epoch 5, batch 32450, giga_loss[loss=0.2625, simple_loss=0.3327, pruned_loss=0.09613, over 28725.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3513, pruned_loss=0.1041, over 5682817.63 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3788, pruned_loss=0.1396, over 5671422.30 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3502, pruned_loss=0.1015, over 5670078.46 frames. ], batch size: 262, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:08:32,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2409, 1.5412, 1.2710, 1.3289], device='cuda:1'), covar=tensor([0.2041, 0.1778, 0.1811, 0.1707], device='cuda:1'), in_proj_covar=tensor([0.1130, 0.0867, 0.1007, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 21:08:54,390 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 21:09:33,949 INFO [train.py:968] (1/2) Epoch 5, batch 32500, giga_loss[loss=0.223, simple_loss=0.2981, pruned_loss=0.07396, over 29096.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3449, pruned_loss=0.1012, over 5681224.68 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3786, pruned_loss=0.1394, over 5675279.70 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3436, pruned_loss=0.09866, over 5667695.21 frames. ], batch size: 113, lr: 6.20e-03, grad_scale: 4.0
+2023-03-02 21:10:11,714 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 32550, giga_loss[loss=0.2455, simple_loss=0.3255, pruned_loss=0.08271, over 28307.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3447, pruned_loss=0.1018, over 5661878.92 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3788, pruned_loss=0.1396, over 5665213.42 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.343, pruned_loss=0.09907, over 5660012.33 frames. ], batch size: 368, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:10:41,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-02 21:11:10,670 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=214477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:11:13,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=214480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:11:31,406 INFO [train.py:968] (1/2) Epoch 5, batch 32600, libri_loss[loss=0.3187, simple_loss=0.373, pruned_loss=0.1322, over 29640.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3478, pruned_loss=0.1049, over 5664445.77 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3775, pruned_loss=0.1388, over 5673847.82 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3459, pruned_loss=0.1016, over 5654475.77 frames. ], batch size: 91, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:11:40,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=214509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:11:56,627 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=214522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:12:01,346 INFO [optim.py:369] (1/2) 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 21:12:32,339 INFO [train.py:968] (1/2) Epoch 5, batch 32650, giga_loss[loss=0.2377, simple_loss=0.3248, pruned_loss=0.07524, over 28847.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3457, pruned_loss=0.103, over 5662312.41 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.377, pruned_loss=0.1384, over 5676338.46 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3442, pruned_loss=0.1002, over 5652270.41 frames. ], batch size: 174, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:13:19,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-02 21:13:33,775 INFO [train.py:968] (1/2) Epoch 5, batch 32700, giga_loss[loss=0.2552, simple_loss=0.3334, pruned_loss=0.08849, over 28843.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3439, pruned_loss=0.1006, over 5668238.08 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3764, pruned_loss=0.1378, over 5682129.20 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3423, pruned_loss=0.09789, over 5654370.41 frames. ], batch size: 186, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:13:47,374 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=214611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:14:09,612 INFO [optim.py:369] (1/2) 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 21:14:35,794 INFO [train.py:968] (1/2) Epoch 5, batch 32750, giga_loss[loss=0.2583, simple_loss=0.3359, pruned_loss=0.09031, over 28667.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3427, pruned_loss=0.1002, over 5672291.22 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3766, pruned_loss=0.1379, over 5686307.89 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3406, pruned_loss=0.09722, over 5657077.84 frames. ], batch size: 307, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:15:36,868 INFO [train.py:968] (1/2) Epoch 5, batch 32800, giga_loss[loss=0.2911, simple_loss=0.3678, pruned_loss=0.1072, over 29041.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3427, pruned_loss=0.1005, over 5677968.73 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3766, pruned_loss=0.1379, over 5695912.43 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3395, pruned_loss=0.09626, over 5656066.14 frames. ], batch size: 285, lr: 6.19e-03, grad_scale: 8.0
+2023-03-02 21:16:11,223 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 32850, giga_loss[loss=0.2459, simple_loss=0.3276, pruned_loss=0.08208, over 28122.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.343, pruned_loss=0.1008, over 5652337.67 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3761, pruned_loss=0.1377, over 5677699.86 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3401, pruned_loss=0.09663, over 5650295.66 frames. ], batch size: 412, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:17:03,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0101, 4.8541, 4.6438, 1.9260], device='cuda:1'), covar=tensor([0.0351, 0.0368, 0.0631, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0785, 0.0780, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:17:41,110 INFO [train.py:968] (1/2) Epoch 5, batch 32900, giga_loss[loss=0.3, simple_loss=0.3629, pruned_loss=0.1185, over 28659.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3447, pruned_loss=0.1028, over 5660332.78 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.376, pruned_loss=0.1377, over 5684840.65 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3412, pruned_loss=0.09805, over 5651300.51 frames. ], batch size: 242, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:18:12,029 INFO [optim.py:369] (1/2) 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 21:18:42,512 INFO [train.py:968] (1/2) Epoch 5, batch 32950, giga_loss[loss=0.2164, simple_loss=0.2799, pruned_loss=0.07643, over 24583.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.344, pruned_loss=0.1025, over 5663585.64 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.376, pruned_loss=0.1377, over 5689039.09 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3406, pruned_loss=0.09814, over 5652496.81 frames. ], batch size: 705, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:19:03,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 21:19:18,091 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 21:19:39,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:19:42,142 INFO [train.py:968] (1/2) Epoch 5, batch 33000, giga_loss[loss=0.2478, simple_loss=0.3333, pruned_loss=0.08113, over 28979.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3443, pruned_loss=0.1007, over 5664541.41 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3763, pruned_loss=0.1379, over 5688720.96 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3409, pruned_loss=0.09663, over 5655597.70 frames. ], batch size: 100, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:19:42,142 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 21:19:50,441 INFO [train.py:1012] (1/2) Epoch 5, validation: loss=0.2195, simple_loss=0.317, pruned_loss=0.06099, over 944034.00 frames. 
+2023-03-02 21:19:50,442 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-02 21:20:21,976 INFO [optim.py:369] (1/2) 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 21:20:23,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4588, 2.0122, 1.6672, 1.8059], device='cuda:1'), covar=tensor([0.0688, 0.0243, 0.0292, 0.0757], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0124, 0.0129, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-02 21:20:47,510 INFO [train.py:968] (1/2) Epoch 5, batch 33050, giga_loss[loss=0.2351, simple_loss=0.3279, pruned_loss=0.07109, over 28927.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3461, pruned_loss=0.1004, over 5666809.46 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3763, pruned_loss=0.1379, over 5692058.31 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3427, pruned_loss=0.09621, over 5656234.48 frames. ], batch size: 186, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:21:30,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=214986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:21:46,615 INFO [train.py:968] (1/2) Epoch 5, batch 33100, giga_loss[loss=0.2884, simple_loss=0.368, pruned_loss=0.1044, over 28967.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3476, pruned_loss=0.1015, over 5648659.45 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3766, pruned_loss=0.1385, over 5683455.59 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.344, pruned_loss=0.09679, over 5646333.83 frames. ], batch size: 155, lr: 6.19e-03, grad_scale: 2.0
+2023-03-02 21:22:27,117 INFO [optim.py:369] (1/2) 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 21:22:41,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:22:45,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:22:52,492 INFO [train.py:968] (1/2) Epoch 5, batch 33150, giga_loss[loss=0.2934, simple_loss=0.3709, pruned_loss=0.1079, over 28903.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3493, pruned_loss=0.1028, over 5652544.11 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.376, pruned_loss=0.1381, over 5686909.58 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3466, pruned_loss=0.09895, over 5646877.87 frames. ], batch size: 227, lr: 6.19e-03, grad_scale: 2.0
+2023-03-02 21:22:54,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2308, 1.2550, 1.0895, 0.9941], device='cuda:1'), covar=tensor([0.0659, 0.0462, 0.1005, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0442, 0.0503, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:23:16,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:23:45,110 INFO [train.py:968] (1/2) Epoch 5, batch 33200, giga_loss[loss=0.2621, simple_loss=0.3364, pruned_loss=0.09385, over 28821.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3486, pruned_loss=0.1026, over 5662791.29 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3756, pruned_loss=0.1378, over 5691723.91 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3457, pruned_loss=0.09846, over 5652807.82 frames. ], batch size: 243, lr: 6.19e-03, grad_scale: 4.0
+2023-03-02 21:23:52,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4935, 4.3363, 4.1223, 2.0547], device='cuda:1'), covar=tensor([0.0410, 0.0481, 0.0703, 0.2057], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0770, 0.0768, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:24:17,501 INFO [optim.py:369] (1/2) 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 21:24:19,367 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=215129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:24:22,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=215132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:24:40,199 INFO [train.py:968] (1/2) Epoch 5, batch 33250, giga_loss[loss=0.2586, simple_loss=0.3352, pruned_loss=0.09096, over 28651.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3463, pruned_loss=0.1008, over 5664343.91 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3755, pruned_loss=0.1375, over 5694454.78 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3432, pruned_loss=0.09653, over 5653016.17 frames. ], batch size: 262, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:24:40,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5122, 2.2233, 1.5075, 0.5959], device='cuda:1'), covar=tensor([0.2702, 0.1418, 0.2515, 0.3050], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1328, 0.1367, 0.1162], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 21:24:58,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=215161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:25:42,153 INFO [train.py:968] (1/2) Epoch 5, batch 33300, giga_loss[loss=0.2602, simple_loss=0.3327, pruned_loss=0.09383, over 28921.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3444, pruned_loss=0.1006, over 5667697.07 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3749, pruned_loss=0.1372, over 5697519.54 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.342, pruned_loss=0.0968, over 5655351.13 frames. ], batch size: 213, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:26:16,900 INFO [optim.py:369] (1/2) 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 21:26:34,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5757, 1.5577, 1.2270, 1.2965], device='cuda:1'), covar=tensor([0.0642, 0.0504, 0.0852, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0443, 0.0502, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:26:37,528 INFO [train.py:968] (1/2) Epoch 5, batch 33350, giga_loss[loss=0.2753, simple_loss=0.3655, pruned_loss=0.0925, over 28840.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3451, pruned_loss=0.1008, over 5674322.32 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3748, pruned_loss=0.1371, over 5696572.56 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3421, pruned_loss=0.09647, over 5663293.09 frames. ], batch size: 174, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:27:38,861 INFO [train.py:968] (1/2) Epoch 5, batch 33400, giga_loss[loss=0.2658, simple_loss=0.3407, pruned_loss=0.09538, over 28866.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3499, pruned_loss=0.1041, over 5666876.95 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3746, pruned_loss=0.137, over 5692772.21 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3467, pruned_loss=0.09963, over 5659968.60 frames. ], batch size: 112, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:27:39,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0454, 3.8990, 3.7283, 1.7743], device='cuda:1'), covar=tensor([0.0458, 0.0516, 0.0696, 0.2279], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0777, 0.0775, 0.0585], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:28:15,965 INFO [optim.py:369] (1/2) 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 21:28:41,904 INFO [train.py:968] (1/2) Epoch 5, batch 33450, giga_loss[loss=0.2862, simple_loss=0.3565, pruned_loss=0.1079, over 28679.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3503, pruned_loss=0.1046, over 5665975.02 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3743, pruned_loss=0.1369, over 5697921.45 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3475, pruned_loss=0.1004, over 5655387.76 frames. ], batch size: 262, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:29:17,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4526, 2.4237, 1.9740, 1.6905], device='cuda:1'), covar=tensor([0.0763, 0.0202, 0.0252, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0128, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-02 21:29:46,465 INFO [train.py:968] (1/2) Epoch 5, batch 33500, giga_loss[loss=0.2845, simple_loss=0.3694, pruned_loss=0.09982, over 28912.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3531, pruned_loss=0.1061, over 5681041.37 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3741, pruned_loss=0.1368, over 5703824.95 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3504, pruned_loss=0.102, over 5666480.52 frames. ], batch size: 164, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:30:05,921 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 21:30:28,250 INFO [optim.py:369] (1/2) 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 21:30:45,715 INFO [train.py:968] (1/2) Epoch 5, batch 33550, giga_loss[loss=0.2836, simple_loss=0.3641, pruned_loss=0.1016, over 28905.00 frames. ], tot_loss[loss=0.284, simple_loss=0.355, pruned_loss=0.1065, over 5679379.50 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3736, pruned_loss=0.1366, over 5707801.63 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.353, pruned_loss=0.1027, over 5663383.56 frames. ], batch size: 213, lr: 6.18e-03, grad_scale: 2.0
+2023-03-02 21:30:53,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4027, 1.4824, 1.2529, 1.6985], device='cuda:1'), covar=tensor([0.2313, 0.2080, 0.2069, 0.2261], device='cuda:1'), in_proj_covar=tensor([0.1105, 0.0853, 0.0993, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 21:31:49,828 INFO [train.py:968] (1/2) Epoch 5, batch 33600, giga_loss[loss=0.2877, simple_loss=0.3589, pruned_loss=0.1083, over 29107.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5671217.15 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3738, pruned_loss=0.1368, over 5700884.76 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3524, pruned_loss=0.1013, over 5664108.60 frames. ], batch size: 200, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:32:27,304 INFO [optim.py:369] (1/2) 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 21:32:56,596 INFO [train.py:968] (1/2) Epoch 5, batch 33650, giga_loss[loss=0.2889, simple_loss=0.3431, pruned_loss=0.1174, over 27007.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3531, pruned_loss=0.1045, over 5664422.93 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3735, pruned_loss=0.1365, over 5704211.93 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 5655568.28 frames. ], batch size: 555, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:34:04,683 INFO [train.py:968] (1/2) Epoch 5, batch 33700, giga_loss[loss=0.2654, simple_loss=0.3374, pruned_loss=0.09669, over 28653.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3526, pruned_loss=0.1052, over 5656267.93 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.374, pruned_loss=0.1369, over 5696957.29 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3504, pruned_loss=0.1018, over 5654803.89 frames. ], batch size: 307, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:34:32,592 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215623.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:34:39,039 INFO [optim.py:369] (1/2) 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 21:35:06,821 INFO [train.py:968] (1/2) Epoch 5, batch 33750, giga_loss[loss=0.2874, simple_loss=0.3578, pruned_loss=0.1085, over 28951.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3509, pruned_loss=0.1043, over 5658054.27 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3733, pruned_loss=0.1367, over 5701704.32 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3493, pruned_loss=0.101, over 5651854.70 frames. ], batch size: 285, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:36:10,347 INFO [train.py:968] (1/2) Epoch 5, batch 33800, giga_loss[loss=0.2577, simple_loss=0.3354, pruned_loss=0.08995, over 28886.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3513, pruned_loss=0.1055, over 5646433.99 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3739, pruned_loss=0.1372, over 5687254.93 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3491, pruned_loss=0.1018, over 5653987.18 frames. ], batch size: 227, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:36:14,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8127, 1.7034, 1.2746, 1.3997], device='cuda:1'), covar=tensor([0.0612, 0.0593, 0.0899, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0439, 0.0499, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:36:32,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9273, 1.2087, 3.2242, 2.7376], device='cuda:1'), covar=tensor([0.1518, 0.2154, 0.0402, 0.1452], device='cuda:1'), in_proj_covar=tensor([0.0562, 0.0524, 0.0741, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 21:36:35,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215718.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:36:47,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:36:51,037 INFO [optim.py:369] (1/2) 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 21:37:16,879 INFO [train.py:968] (1/2) Epoch 5, batch 33850, giga_loss[loss=0.2603, simple_loss=0.3324, pruned_loss=0.09407, over 28957.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3485, pruned_loss=0.1042, over 5640778.85 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3737, pruned_loss=0.1371, over 5688414.65 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3469, pruned_loss=0.1013, over 5645620.31 frames. ], batch size: 136, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:38:18,096 INFO [train.py:968] (1/2) Epoch 5, batch 33900, libri_loss[loss=0.3437, simple_loss=0.3795, pruned_loss=0.154, over 29568.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3473, pruned_loss=0.1022, over 5647838.80 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3736, pruned_loss=0.1371, over 5692396.74 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3456, pruned_loss=0.09926, over 5647204.04 frames. ], batch size: 76, lr: 6.18e-03, grad_scale: 4.0
+2023-03-02 21:38:34,726 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-02 21:38:43,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-02 21:38:56,634 INFO [optim.py:369] (1/2) 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 21:39:04,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-02 21:39:20,464 INFO [train.py:968] (1/2) Epoch 5, batch 33950, giga_loss[loss=0.2688, simple_loss=0.3572, pruned_loss=0.09015, over 28401.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3466, pruned_loss=0.09983, over 5662779.70 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3736, pruned_loss=0.1371, over 5694538.34 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.345, pruned_loss=0.09719, over 5660089.72 frames. ], batch size: 336, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:39:23,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8434, 4.6915, 4.4536, 1.9060], device='cuda:1'), covar=tensor([0.0304, 0.0345, 0.0569, 0.1949], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0763, 0.0760, 0.0579], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:40:07,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9589, 5.8981, 5.4928, 2.3775], device='cuda:1'), covar=tensor([0.0333, 0.0395, 0.0716, 0.1895], device='cuda:1'), in_proj_covar=tensor([0.0828, 0.0763, 0.0761, 0.0580], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:40:13,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:40:18,460 INFO [train.py:968] (1/2) Epoch 5, batch 34000, giga_loss[loss=0.2745, simple_loss=0.3574, pruned_loss=0.09579, over 28348.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3482, pruned_loss=0.09854, over 5664785.86 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3735, pruned_loss=0.1371, over 5694012.20 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3467, pruned_loss=0.09596, over 5662582.42 frames. ], batch size: 368, lr: 6.17e-03, grad_scale: 8.0
+2023-03-02 21:40:55,438 INFO [optim.py:369] (1/2) 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 21:41:07,047 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215941.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:41:17,040 INFO [train.py:968] (1/2) Epoch 5, batch 34050, libri_loss[loss=0.2848, simple_loss=0.3306, pruned_loss=0.1194, over 29353.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3484, pruned_loss=0.09816, over 5661143.88 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3733, pruned_loss=0.1371, over 5693411.33 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.347, pruned_loss=0.09567, over 5659327.94 frames. ], batch size: 71, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:41:28,738 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=215958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:42:17,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9780, 1.9887, 1.3763, 1.6838], device='cuda:1'), covar=tensor([0.0592, 0.0442, 0.0864, 0.0784], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0438, 0.0498, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:42:23,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=215998.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:42:24,862 INFO [train.py:968] (1/2) Epoch 5, batch 34100, giga_loss[loss=0.2582, simple_loss=0.3384, pruned_loss=0.08899, over 28626.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3491, pruned_loss=0.09926, over 5658349.43 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3733, pruned_loss=0.1372, over 5689985.02 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3473, pruned_loss=0.09602, over 5659648.28 frames. ], batch size: 262, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:42:33,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4404, 1.8654, 1.7503, 1.6321], device='cuda:1'), covar=tensor([0.1483, 0.1705, 0.1126, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0719, 0.0773, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 21:43:06,139 INFO [optim.py:369] (1/2) 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 21:43:26,957 INFO [train.py:968] (1/2) Epoch 5, batch 34150, giga_loss[loss=0.3057, simple_loss=0.3757, pruned_loss=0.1179, over 28386.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3503, pruned_loss=0.1001, over 5667540.90 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.373, pruned_loss=0.137, over 5693114.18 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3486, pruned_loss=0.09688, over 5665224.32 frames. ], batch size: 336, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:44:24,785 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-02 21:44:27,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216093.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:44:35,037 INFO [train.py:968] (1/2) Epoch 5, batch 34200, giga_loss[loss=0.3021, simple_loss=0.3723, pruned_loss=0.116, over 28897.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3498, pruned_loss=0.09948, over 5662616.66 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3727, pruned_loss=0.1367, over 5693288.77 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3482, pruned_loss=0.09629, over 5659863.22 frames. ], batch size: 284, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:44:36,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:45:20,667 INFO [optim.py:369] (1/2) 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 21:45:35,704 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216141.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:45:40,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216144.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:45:47,574 INFO [train.py:968] (1/2) Epoch 5, batch 34250, giga_loss[loss=0.3194, simple_loss=0.3955, pruned_loss=0.1216, over 28648.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3503, pruned_loss=0.09943, over 5665967.96 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3724, pruned_loss=0.1365, over 5698151.42 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3488, pruned_loss=0.09623, over 5659167.63 frames. ], batch size: 307, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:46:16,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216173.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:46:50,613 INFO [train.py:968] (1/2) Epoch 5, batch 34300, libri_loss[loss=0.307, simple_loss=0.3621, pruned_loss=0.1259, over 29526.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3539, pruned_loss=0.1018, over 5673245.36 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3718, pruned_loss=0.1359, over 5705392.51 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3526, pruned_loss=0.09854, over 5660073.64 frames. ], batch size: 83, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:47:18,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:47:24,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8650, 2.3431, 1.9309, 2.0551], device='cuda:1'), covar=tensor([0.0548, 0.0230, 0.0234, 0.0590], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0124, 0.0128, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-02 21:47:30,930 INFO [optim.py:369] (1/2) 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 21:47:38,818 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216236.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:47:42,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216239.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:47:47,874 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:47:51,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:47:54,438 INFO [train.py:968] (1/2) Epoch 5, batch 34350, giga_loss[loss=0.2859, simple_loss=0.3616, pruned_loss=0.105, over 28072.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3543, pruned_loss=0.1015, over 5676346.14 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3716, pruned_loss=0.1359, over 5700450.71 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.353, pruned_loss=0.09824, over 5669686.40 frames. ], batch size: 412, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:48:21,528 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216268.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 21:48:23,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:48:32,077 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:49:01,509 INFO [train.py:968] (1/2) Epoch 5, batch 34400, giga_loss[loss=0.3031, simple_loss=0.3522, pruned_loss=0.127, over 24768.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3535, pruned_loss=0.1023, over 5676059.55 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3717, pruned_loss=0.1357, over 5694523.58 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3517, pruned_loss=0.09843, over 5674371.40 frames. ], batch size: 705, lr: 6.17e-03, grad_scale: 8.0
+2023-03-02 21:49:21,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216316.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:49:24,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 21:49:43,218 INFO [optim.py:369] (1/2) 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 21:49:45,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:50:02,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2219, 1.8419, 1.3336, 0.4829], device='cuda:1'), covar=tensor([0.2013, 0.1399, 0.2669, 0.2535], device='cuda:1'), in_proj_covar=tensor([0.1404, 0.1333, 0.1393, 0.1168], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 21:50:08,586 INFO [train.py:968] (1/2) Epoch 5, batch 34450, libri_loss[loss=0.3482, simple_loss=0.3796, pruned_loss=0.1584, over 29678.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3503, pruned_loss=0.1007, over 5684271.36 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3711, pruned_loss=0.1354, over 5699710.60 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3489, pruned_loss=0.09706, over 5677831.89 frames. ], batch size: 73, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:51:20,672 INFO [train.py:968] (1/2) Epoch 5, batch 34500, giga_loss[loss=0.2783, simple_loss=0.3513, pruned_loss=0.1027, over 27640.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3472, pruned_loss=0.09707, over 5691671.26 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3712, pruned_loss=0.1354, over 5700945.05 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.346, pruned_loss=0.09411, over 5685608.15 frames. ], batch size: 472, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:51:38,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:51:41,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:51:50,285 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:52:02,332 INFO [optim.py:369] (1/2) 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 21:52:19,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:52:23,457 INFO [train.py:968] (1/2) Epoch 5, batch 34550, giga_loss[loss=0.2629, simple_loss=0.3434, pruned_loss=0.09125, over 28961.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3483, pruned_loss=0.09832, over 5694616.45 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3718, pruned_loss=0.1358, over 5705200.02 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3461, pruned_loss=0.09464, over 5685836.26 frames. ], batch size: 199, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:52:23,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3589, 1.8388, 1.3221, 0.6356], device='cuda:1'), covar=tensor([0.2424, 0.1381, 0.1946, 0.2639], device='cuda:1'), in_proj_covar=tensor([0.1402, 0.1329, 0.1385, 0.1168], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 21:52:37,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216459.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:52:40,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216462.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:52:56,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:53:00,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:53:15,543 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216491.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 21:53:25,904 INFO [train.py:968] (1/2) Epoch 5, batch 34600, giga_loss[loss=0.3242, simple_loss=0.389, pruned_loss=0.1297, over 27548.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.351, pruned_loss=0.1001, over 5679109.66 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3718, pruned_loss=0.1358, over 5697957.19 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.349, pruned_loss=0.09681, over 5678033.15 frames. ], batch size: 472, lr: 6.17e-03, grad_scale: 4.0
+2023-03-02 21:53:36,334 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:53:40,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 21:54:04,430 INFO [optim.py:369] (1/2) 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 21:54:24,481 INFO [train.py:968] (1/2) Epoch 5, batch 34650, giga_loss[loss=0.2721, simple_loss=0.3485, pruned_loss=0.0978, over 28415.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3538, pruned_loss=0.1027, over 5671826.59 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3717, pruned_loss=0.1358, over 5699857.79 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3519, pruned_loss=0.0993, over 5668384.31 frames. ], batch size: 368, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:54:40,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3211, 1.8534, 1.6543, 1.4974], device='cuda:1'), covar=tensor([0.1478, 0.1758, 0.1144, 0.1226], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0717, 0.0775, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 21:55:17,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:55:21,385 INFO [train.py:968] (1/2) Epoch 5, batch 34700, giga_loss[loss=0.2888, simple_loss=0.3575, pruned_loss=0.11, over 28996.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3501, pruned_loss=0.1019, over 5674852.14 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3711, pruned_loss=0.1353, over 5707143.29 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3485, pruned_loss=0.09848, over 5664572.32 frames. ], batch size: 285, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:55:30,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4410, 1.6842, 1.3127, 1.0897], device='cuda:1'), covar=tensor([0.1059, 0.0862, 0.0607, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.1417, 0.1185, 0.1164, 0.1254], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 21:55:58,406 INFO [optim.py:369] (1/2) 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 21:56:01,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4468, 1.8240, 1.7501, 1.6945], device='cuda:1'), covar=tensor([0.1333, 0.1508, 0.1086, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0712, 0.0774, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 21:56:16,941 INFO [train.py:968] (1/2) Epoch 5, batch 34750, giga_loss[loss=0.3108, simple_loss=0.358, pruned_loss=0.1318, over 26866.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3504, pruned_loss=0.1027, over 5678759.27 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3712, pruned_loss=0.1353, over 5709697.59 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3487, pruned_loss=0.09951, over 5667841.28 frames. ], batch size: 555, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:57:06,853 INFO [train.py:968] (1/2) Epoch 5, batch 34800, libri_loss[loss=0.2801, simple_loss=0.3347, pruned_loss=0.1127, over 29493.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3536, pruned_loss=0.1055, over 5671641.26 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.371, pruned_loss=0.1351, over 5711523.19 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3517, pruned_loss=0.102, over 5659884.38 frames. ], batch size: 70, lr: 6.16e-03, grad_scale: 8.0
+2023-03-02 21:57:36,235 INFO [optim.py:369] (1/2) 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 21:57:40,219 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-02 21:57:43,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:57:45,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:57:51,034 INFO [train.py:968] (1/2) Epoch 5, batch 34850, libri_loss[loss=0.2968, simple_loss=0.3631, pruned_loss=0.1152, over 27603.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3638, pruned_loss=0.1122, over 5670708.51 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3708, pruned_loss=0.1349, over 5702424.01 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3622, pruned_loss=0.1088, over 5668594.28 frames. ], batch size: 116, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:57:54,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8482, 1.6012, 1.1732, 1.3469], device='cuda:1'), covar=tensor([0.0634, 0.0667, 0.0994, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0443, 0.0509, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 21:58:12,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:58:33,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=216796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 21:58:37,300 INFO [train.py:968] (1/2) Epoch 5, batch 34900, giga_loss[loss=0.336, simple_loss=0.3956, pruned_loss=0.1383, over 29123.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.371, pruned_loss=0.1167, over 5669736.57 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.371, pruned_loss=0.1349, over 5697835.65 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3694, pruned_loss=0.1135, over 5671825.56 frames. ], batch size: 155, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:58:44,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8767, 1.8387, 1.7059, 1.7175], device='cuda:1'), covar=tensor([0.0786, 0.1156, 0.1177, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0733, 0.0628, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 21:59:05,082 INFO [optim.py:369] (1/2) 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 21:59:16,783 INFO [train.py:968] (1/2) Epoch 5, batch 34950, libri_loss[loss=0.3809, simple_loss=0.4112, pruned_loss=0.1754, over 19996.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3723, pruned_loss=0.119, over 5667346.88 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3712, pruned_loss=0.135, over 5692943.50 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3708, pruned_loss=0.1157, over 5673335.58 frames. ], batch size: 187, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 21:59:46,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1436, 1.4616, 1.1926, 0.5031], device='cuda:1'), covar=tensor([0.1237, 0.0832, 0.1226, 0.2094], device='cuda:1'), in_proj_covar=tensor([0.1396, 0.1324, 0.1377, 0.1155], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 22:00:01,316 INFO [train.py:968] (1/2) Epoch 5, batch 35000, giga_loss[loss=0.3127, simple_loss=0.371, pruned_loss=0.1272, over 28781.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.366, pruned_loss=0.1163, over 5671000.26 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3712, pruned_loss=0.1349, over 5694751.28 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3648, pruned_loss=0.1136, over 5673707.71 frames. ], batch size: 284, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:00:07,992 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-02 22:00:09,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1926, 1.6991, 1.4774, 1.4501], device='cuda:1'), covar=tensor([0.0751, 0.0403, 0.0318, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0125, 0.0130, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0039, 0.0066], device='cuda:1')
+2023-03-02 22:00:29,095 INFO [optim.py:369] (1/2) 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 22:00:34,962 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=216939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:00:36,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=216942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:00:43,308 INFO [train.py:968] (1/2) Epoch 5, batch 35050, giga_loss[loss=0.2421, simple_loss=0.3042, pruned_loss=0.08995, over 28379.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3585, pruned_loss=0.1127, over 5673392.64 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3717, pruned_loss=0.1349, over 5693811.77 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3569, pruned_loss=0.11, over 5675822.42 frames. ], batch size: 65, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:01:00,946 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=216971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:01:11,119 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=216984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:01:25,117 INFO [train.py:968] (1/2) Epoch 5, batch 35100, giga_loss[loss=0.259, simple_loss=0.3227, pruned_loss=0.09761, over 28894.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.351, pruned_loss=0.1094, over 5681213.25 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3718, pruned_loss=0.1349, over 5696734.64 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3491, pruned_loss=0.1065, over 5680145.55 frames. ], batch size: 199, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:01:32,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6726, 1.7348, 1.6575, 1.6871], device='cuda:1'), covar=tensor([0.1198, 0.2019, 0.1545, 0.1654], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0739, 0.0625, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 22:01:50,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1828, 1.6756, 1.2550, 0.4647], device='cuda:1'), covar=tensor([0.1936, 0.1131, 0.2118, 0.2644], device='cuda:1'), in_proj_covar=tensor([0.1373, 0.1304, 0.1356, 0.1147], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 22:01:52,665 INFO [optim.py:369] (1/2) 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 22:01:53,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6735, 2.0350, 1.9608, 1.7397], device='cuda:1'), covar=tensor([0.1460, 0.1629, 0.1081, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0731, 0.0784, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-02 22:01:54,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1418, 1.8154, 1.4414, 0.4000], device='cuda:1'), covar=tensor([0.2376, 0.1541, 0.2318, 0.2742], device='cuda:1'), in_proj_covar=tensor([0.1373, 0.1304, 0.1357, 0.1147], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 22:02:03,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217048.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:02:05,853 INFO [train.py:968] (1/2) Epoch 5, batch 35150, giga_loss[loss=0.2442, simple_loss=0.3063, pruned_loss=0.09098, over 28633.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3436, pruned_loss=0.1059, over 5679238.21 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3725, pruned_loss=0.1352, over 5696381.26 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3409, pruned_loss=0.1028, over 5678496.80 frames. ], batch size: 85, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:02:49,039 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:02:50,310 INFO [train.py:968] (1/2) Epoch 5, batch 35200, giga_loss[loss=0.2409, simple_loss=0.3154, pruned_loss=0.08316, over 29052.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.337, pruned_loss=0.1023, over 5688000.90 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3724, pruned_loss=0.1351, over 5699569.32 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3344, pruned_loss=0.09952, over 5684345.64 frames. ], batch size: 155, lr: 6.16e-03, grad_scale: 8.0
+2023-03-02 22:03:15,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6230, 1.6798, 1.6580, 1.7075], device='cuda:1'), covar=tensor([0.1208, 0.1635, 0.1587, 0.1474], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0743, 0.0630, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 22:03:19,040 INFO [optim.py:369] (1/2) 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 22:03:33,122 INFO [train.py:968] (1/2) Epoch 5, batch 35250, giga_loss[loss=0.2513, simple_loss=0.3198, pruned_loss=0.09139, over 29021.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3358, pruned_loss=0.1024, over 5685617.05 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3732, pruned_loss=0.1355, over 5691453.34 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3318, pruned_loss=0.09858, over 5690454.24 frames. ], batch size: 128, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:04:15,041 INFO [train.py:968] (1/2) Epoch 5, batch 35300, libri_loss[loss=0.3279, simple_loss=0.3871, pruned_loss=0.1343, over 29639.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3326, pruned_loss=0.1007, over 5680419.24 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3737, pruned_loss=0.1356, over 5687865.47 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3275, pruned_loss=0.09639, over 5686175.42 frames. ], batch size: 91, lr: 6.16e-03, grad_scale: 4.0
+2023-03-02 22:04:43,773 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 35350, giga_loss[loss=0.2476, simple_loss=0.3031, pruned_loss=0.09604, over 28723.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3311, pruned_loss=0.1007, over 5667784.50 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.374, pruned_loss=0.1358, over 5684704.42 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3254, pruned_loss=0.09602, over 5674641.21 frames. ], batch size: 78, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:05:07,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6980, 1.5136, 1.3384, 1.2446], device='cuda:1'), covar=tensor([0.0604, 0.0534, 0.0941, 0.0898], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0443, 0.0510, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 22:05:17,066 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-02 22:05:40,500 INFO [train.py:968] (1/2) Epoch 5, batch 35400, libri_loss[loss=0.3293, simple_loss=0.3898, pruned_loss=0.1344, over 29732.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3286, pruned_loss=0.09944, over 5679396.43 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3747, pruned_loss=0.136, over 5692269.20 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3219, pruned_loss=0.09428, over 5677417.01 frames. ], batch size: 87, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:06:00,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-02 22:06:07,935 INFO [optim.py:369] (1/2) 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 22:06:21,560 INFO [train.py:968] (1/2) Epoch 5, batch 35450, giga_loss[loss=0.2487, simple_loss=0.3156, pruned_loss=0.09088, over 28658.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3253, pruned_loss=0.09712, over 5685434.40 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.375, pruned_loss=0.1361, over 5693735.07 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3191, pruned_loss=0.09241, over 5682681.33 frames. ], batch size: 307, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:06:29,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:06:30,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:06:37,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6524, 1.7165, 1.6025, 2.1756], device='cuda:1'), covar=tensor([0.2188, 0.2034, 0.2077, 0.2020], device='cuda:1'), in_proj_covar=tensor([0.1122, 0.0872, 0.1003, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 22:07:01,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0072, 1.1687, 4.0893, 3.2593], device='cuda:1'), covar=tensor([0.1600, 0.2405, 0.0335, 0.0661], device='cuda:1'), in_proj_covar=tensor([0.0563, 0.0528, 0.0742, 0.0603], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 22:07:04,260 INFO [train.py:968] (1/2) Epoch 5, batch 35500, giga_loss[loss=0.2336, simple_loss=0.3002, pruned_loss=0.08347, over 28682.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3226, pruned_loss=0.09611, over 5680263.93 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3758, pruned_loss=0.1367, over 5688223.96 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3162, pruned_loss=0.09124, over 5682318.56 frames. ], batch size: 92, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:07:23,458 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217423.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:07:30,996 INFO [optim.py:369] (1/2) 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 22:07:44,743 INFO [train.py:968] (1/2) Epoch 5, batch 35550, giga_loss[loss=0.2107, simple_loss=0.278, pruned_loss=0.07167, over 28157.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3199, pruned_loss=0.09463, over 5682747.51 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3759, pruned_loss=0.1368, over 5689526.49 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3133, pruned_loss=0.08961, over 5682972.68 frames. ], batch size: 77, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:08:04,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:08:12,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5701, 1.6684, 1.3362, 1.0217], device='cuda:1'), covar=tensor([0.1093, 0.0985, 0.0670, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.1405, 0.1187, 0.1181, 0.1259], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 22:08:25,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6807, 1.7027, 1.7161, 1.5722], device='cuda:1'), covar=tensor([0.1263, 0.1840, 0.1560, 0.1581], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0739, 0.0634, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 22:08:28,784 INFO [train.py:968] (1/2) Epoch 5, batch 35600, giga_loss[loss=0.2383, simple_loss=0.3015, pruned_loss=0.08757, over 28861.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3164, pruned_loss=0.09291, over 5667778.92 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3768, pruned_loss=0.1374, over 5681982.10 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3094, pruned_loss=0.08767, over 5673947.01 frames. ], batch size: 186, lr: 6.15e-03, grad_scale: 8.0
+2023-03-02 22:08:32,046 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:08:34,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:09:01,963 INFO [optim.py:369] (1/2) 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 22:09:02,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 35650, giga_loss[loss=0.3132, simple_loss=0.3712, pruned_loss=0.1277, over 29030.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3203, pruned_loss=0.09562, over 5664444.37 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.377, pruned_loss=0.1375, over 5678739.44 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3136, pruned_loss=0.09063, over 5671576.31 frames. ], batch size: 106, lr: 6.15e-03, grad_scale: 8.0
+2023-03-02 22:09:31,615 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217566.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:09:33,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217569.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:09:39,401 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-02 22:09:58,851 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217598.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:10:00,038 INFO [train.py:968] (1/2) Epoch 5, batch 35700, giga_loss[loss=0.2997, simple_loss=0.3702, pruned_loss=0.1146, over 28866.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3341, pruned_loss=0.1028, over 5669343.61 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3772, pruned_loss=0.1377, over 5672582.21 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3277, pruned_loss=0.09806, over 5681461.44 frames. ], batch size: 199, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:10:14,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:10:16,416 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:10:30,971 INFO [optim.py:369] (1/2) 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 22:10:41,727 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:10:42,860 INFO [train.py:968] (1/2) Epoch 5, batch 35750, giga_loss[loss=0.3189, simple_loss=0.3912, pruned_loss=0.1233, over 29008.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.349, pruned_loss=0.1119, over 5673929.49 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3777, pruned_loss=0.1378, over 5679230.58 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3422, pruned_loss=0.1069, over 5677495.25 frames. ], batch size: 164, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:11:18,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3177, 1.2985, 1.2604, 1.5033], device='cuda:1'), covar=tensor([0.0788, 0.0337, 0.0316, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0123, 0.0127, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0058, 0.0043, 0.0038, 0.0064], device='cuda:1')
+2023-03-02 22:11:25,885 INFO [train.py:968] (1/2) Epoch 5, batch 35800, giga_loss[loss=0.3095, simple_loss=0.3826, pruned_loss=0.1182, over 28652.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3586, pruned_loss=0.1167, over 5665578.52 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3784, pruned_loss=0.1383, over 5671539.62 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3521, pruned_loss=0.1118, over 5675545.41 frames. ], batch size: 336, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:11:52,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=217733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:11:54,633 INFO [optim.py:369] (1/2) 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 22:12:04,480 INFO [train.py:968] (1/2) Epoch 5, batch 35850, libri_loss[loss=0.2965, simple_loss=0.3551, pruned_loss=0.1189, over 29592.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3636, pruned_loss=0.1176, over 5678534.65 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3786, pruned_loss=0.1381, over 5677806.80 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3577, pruned_loss=0.1133, over 5680813.31 frames. ], batch size: 74, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:12:33,591 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:12:49,362 INFO [train.py:968] (1/2) Epoch 5, batch 35900, giga_loss[loss=0.3295, simple_loss=0.3854, pruned_loss=0.1368, over 29073.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.365, pruned_loss=0.1172, over 5663306.69 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3791, pruned_loss=0.1384, over 5674159.73 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3594, pruned_loss=0.1129, over 5668793.36 frames. ], batch size: 128, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:13:12,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3833, 3.1749, 1.4233, 1.3777], device='cuda:1'), covar=tensor([0.1164, 0.0388, 0.1053, 0.1641], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0471, 0.0306, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 22:13:24,980 INFO [optim.py:369] (1/2) 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 22:13:36,472 INFO [train.py:968] (1/2) Epoch 5, batch 35950, giga_loss[loss=0.2882, simple_loss=0.3539, pruned_loss=0.1113, over 28719.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3661, pruned_loss=0.1172, over 5663880.21 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3792, pruned_loss=0.1384, over 5676695.30 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3615, pruned_loss=0.1136, over 5665812.41 frames. ], batch size: 92, lr: 6.15e-03, grad_scale: 2.0
+2023-03-02 22:13:57,733 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=217876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:13:59,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=217879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:14:04,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=217885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:14:16,690 INFO [train.py:968] (1/2) Epoch 5, batch 36000, giga_loss[loss=0.3199, simple_loss=0.3713, pruned_loss=0.1342, over 28718.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3676, pruned_loss=0.1182, over 5672327.60 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3793, pruned_loss=0.1383, over 5671313.83 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3636, pruned_loss=0.1149, over 5679166.92 frames. ], batch size: 92, lr: 6.15e-03, grad_scale: 4.0
+2023-03-02 22:14:16,690 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 22:14:26,800 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-02 22:14:28,049 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-02 22:14:33,302 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=217908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:14:56,496 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 36050, giga_loss[loss=0.3065, simple_loss=0.3704, pruned_loss=0.1213, over 28308.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3704, pruned_loss=0.1202, over 5676820.58 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3797, pruned_loss=0.1384, over 5675611.12 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3666, pruned_loss=0.1171, over 5678334.41 frames. ], batch size: 65, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:15:48,527 INFO [train.py:968] (1/2) Epoch 5, batch 36100, giga_loss[loss=0.315, simple_loss=0.3807, pruned_loss=0.1247, over 28892.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3735, pruned_loss=0.1214, over 5682206.56 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3797, pruned_loss=0.1383, over 5671369.44 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3702, pruned_loss=0.1186, over 5686655.57 frames. ], batch size: 145, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:16:16,367 INFO [optim.py:369] (1/2) 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 22:16:27,854 INFO [train.py:968] (1/2) Epoch 5, batch 36150, giga_loss[loss=0.2998, simple_loss=0.3716, pruned_loss=0.114, over 28847.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3764, pruned_loss=0.1225, over 5676886.30 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3802, pruned_loss=0.1387, over 5664345.97 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3734, pruned_loss=0.1197, over 5687476.86 frames. ], batch size: 106, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:16:33,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:16:43,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9608, 1.8444, 1.7690, 1.6756], device='cuda:1'), covar=tensor([0.0961, 0.1489, 0.1405, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0733, 0.0627, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 22:16:45,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5080, 1.6030, 1.3732, 1.7305], device='cuda:1'), covar=tensor([0.1896, 0.1619, 0.1518, 0.1648], device='cuda:1'), in_proj_covar=tensor([0.1105, 0.0858, 0.0986, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 22:16:50,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3948, 1.7195, 1.4002, 1.5615], device='cuda:1'), covar=tensor([0.1951, 0.1739, 0.1839, 0.1616], device='cuda:1'), in_proj_covar=tensor([0.1106, 0.0858, 0.0987, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 22:17:10,174 INFO [train.py:968] (1/2) Epoch 5, batch 36200, giga_loss[loss=0.3357, simple_loss=0.402, pruned_loss=0.1347, over 29004.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3774, pruned_loss=0.1224, over 5683852.95 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3809, pruned_loss=0.1391, over 5668457.29 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3743, pruned_loss=0.1193, over 5689268.54 frames. ], batch size: 136, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:17:37,561 INFO [optim.py:369] (1/2) 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 22:17:50,452 INFO [train.py:968] (1/2) Epoch 5, batch 36250, giga_loss[loss=0.3922, simple_loss=0.4266, pruned_loss=0.1789, over 26709.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3779, pruned_loss=0.1218, over 5687267.74 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3811, pruned_loss=0.1393, over 5673047.64 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3751, pruned_loss=0.1189, over 5687849.47 frames. ], batch size: 555, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:17:51,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 22:17:54,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:18:13,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9862, 2.3698, 2.1521, 1.6340], device='cuda:1'), covar=tensor([0.0676, 0.0236, 0.0240, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0122, 0.0126, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0058, 0.0043, 0.0038, 0.0064], device='cuda:1')
+2023-03-02 22:18:28,141 INFO [train.py:968] (1/2) Epoch 5, batch 36300, giga_loss[loss=0.2888, simple_loss=0.3613, pruned_loss=0.1081, over 28405.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.378, pruned_loss=0.1212, over 5690844.02 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3822, pruned_loss=0.1399, over 5669418.32 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3746, pruned_loss=0.1177, over 5696102.66 frames. ], batch size: 71, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:18:50,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3185, 4.0710, 4.0014, 1.6830], device='cuda:1'), covar=tensor([0.0407, 0.0460, 0.0629, 0.2174], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0761, 0.0765, 0.0581], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 22:18:58,964 INFO [optim.py:369] (1/2) 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 22:19:09,139 INFO [train.py:968] (1/2) Epoch 5, batch 36350, giga_loss[loss=0.258, simple_loss=0.3449, pruned_loss=0.08551, over 28910.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.374, pruned_loss=0.1176, over 5693707.84 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3821, pruned_loss=0.1397, over 5673708.56 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3713, pruned_loss=0.1146, over 5694434.20 frames. ], batch size: 174, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:19:17,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:19:32,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0753, 1.3452, 1.0576, 0.2753], device='cuda:1'), covar=tensor([0.1662, 0.1431, 0.2416, 0.2629], device='cuda:1'), in_proj_covar=tensor([0.1391, 0.1321, 0.1381, 0.1169], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 22:19:43,640 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:19:50,865 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:19:51,133 INFO [train.py:968] (1/2) Epoch 5, batch 36400, giga_loss[loss=0.3174, simple_loss=0.387, pruned_loss=0.124, over 28730.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3751, pruned_loss=0.1189, over 5683178.94 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3828, pruned_loss=0.1399, over 5671353.32 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3722, pruned_loss=0.1159, over 5686292.88 frames. ], batch size: 242, lr: 6.14e-03, grad_scale: 8.0
+2023-03-02 22:19:52,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:20:18,529 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:20:36,339 INFO [train.py:968] (1/2) Epoch 5, batch 36450, giga_loss[loss=0.3933, simple_loss=0.4264, pruned_loss=0.1801, over 29071.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3771, pruned_loss=0.1225, over 5686814.36 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3827, pruned_loss=0.1397, over 5676217.44 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3747, pruned_loss=0.1199, over 5685296.89 frames. ], batch size: 155, lr: 6.14e-03, grad_scale: 8.0
+2023-03-02 22:20:48,142 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-02 22:20:51,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9264, 1.2460, 3.5721, 3.1094], device='cuda:1'), covar=tensor([0.1307, 0.1834, 0.0336, 0.0580], device='cuda:1'), in_proj_covar=tensor([0.0564, 0.0528, 0.0738, 0.0605], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 22:21:19,899 INFO [train.py:968] (1/2) Epoch 5, batch 36500, giga_loss[loss=0.3419, simple_loss=0.3946, pruned_loss=0.1446, over 29012.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3792, pruned_loss=0.1263, over 5678717.84 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3828, pruned_loss=0.1398, over 5669080.27 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3772, pruned_loss=0.1241, over 5683896.41 frames. ], batch size: 128, lr: 6.14e-03, grad_scale: 8.0
+2023-03-02 22:21:22,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:24,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:47,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:50,214 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:21:51,468 INFO [optim.py:369] (1/2) 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 22:22:02,488 INFO [train.py:968] (1/2) Epoch 5, batch 36550, giga_loss[loss=0.3594, simple_loss=0.4094, pruned_loss=0.1547, over 28939.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3792, pruned_loss=0.1278, over 5681203.75 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3831, pruned_loss=0.1399, over 5671199.89 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3772, pruned_loss=0.1256, over 5683704.39 frames. ], batch size: 145, lr: 6.14e-03, grad_scale: 4.0
+2023-03-02 22:22:38,941 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:22:43,508 INFO [train.py:968] (1/2) Epoch 5, batch 36600, giga_loss[loss=0.2812, simple_loss=0.3482, pruned_loss=0.1071, over 28568.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3772, pruned_loss=0.1266, over 5679155.91 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3837, pruned_loss=0.1403, over 5657158.79 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3749, pruned_loss=0.124, over 5694252.51 frames. ], batch size: 85, lr: 6.14e-03, grad_scale: 2.0
+2023-03-02 22:23:16,050 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 36650, giga_loss[loss=0.3194, simple_loss=0.3829, pruned_loss=0.128, over 28921.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3771, pruned_loss=0.1267, over 5686319.58 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3839, pruned_loss=0.1403, over 5661016.44 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.375, pruned_loss=0.1245, over 5695432.09 frames. ], batch size: 145, lr: 6.14e-03, grad_scale: 2.0
+2023-03-02 22:23:48,948 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:23:49,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:23:52,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:24:10,596 INFO [train.py:968] (1/2) Epoch 5, batch 36700, giga_loss[loss=0.3115, simple_loss=0.3768, pruned_loss=0.1231, over 28928.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3754, pruned_loss=0.1248, over 5683273.34 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3843, pruned_loss=0.1405, over 5660403.52 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3733, pruned_loss=0.1226, over 5691211.59 frames. ], batch size: 199, lr: 6.14e-03, grad_scale: 2.0
+2023-03-02 22:24:15,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:24:18,066 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:24:45,671 INFO [optim.py:369] (1/2) 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 22:24:59,870 INFO [train.py:968] (1/2) Epoch 5, batch 36750, giga_loss[loss=0.266, simple_loss=0.3465, pruned_loss=0.09275, over 28855.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 5683882.02 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3844, pruned_loss=0.1406, over 5663906.78 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.37, pruned_loss=0.1203, over 5687493.70 frames. ], batch size: 174, lr: 6.13e-03, grad_scale: 2.0
+2023-03-02 22:25:14,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:25:47,690 INFO [train.py:968] (1/2) Epoch 5, batch 36800, giga_loss[loss=0.2507, simple_loss=0.3325, pruned_loss=0.08443, over 28587.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3644, pruned_loss=0.1172, over 5692460.93 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3845, pruned_loss=0.1406, over 5664915.16 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5694657.61 frames. ], batch size: 336, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:26:24,645 INFO [optim.py:369] (1/2) 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 22:26:37,244 INFO [train.py:968] (1/2) Epoch 5, batch 36850, giga_loss[loss=0.2252, simple_loss=0.2996, pruned_loss=0.07539, over 28781.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.358, pruned_loss=0.1139, over 5667731.36 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3855, pruned_loss=0.1413, over 5659521.64 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3553, pruned_loss=0.1114, over 5675368.36 frames. ], batch size: 66, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:27:27,278 INFO [train.py:968] (1/2) Epoch 5, batch 36900, giga_loss[loss=0.304, simple_loss=0.3747, pruned_loss=0.1166, over 29036.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3534, pruned_loss=0.1107, over 5673063.69 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3852, pruned_loss=0.141, over 5664809.69 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.351, pruned_loss=0.1084, over 5674745.04 frames. ], batch size: 155, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:27:34,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218809.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:27:36,769 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:28:02,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218841.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:28:09,460 INFO [train.py:968] (1/2) Epoch 5, batch 36950, giga_loss[loss=0.2817, simple_loss=0.3495, pruned_loss=0.1069, over 28223.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.356, pruned_loss=0.1122, over 5668774.20 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.3863, pruned_loss=0.1416, over 5659387.21 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.352, pruned_loss=0.109, over 5676155.79 frames. ], batch size: 77, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:28:25,661 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 37000, giga_loss[loss=0.34, simple_loss=0.3882, pruned_loss=0.1459, over 27647.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3545, pruned_loss=0.1108, over 5674451.82 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3867, pruned_loss=0.1418, over 5652190.06 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3508, pruned_loss=0.1078, over 5687384.58 frames. ], batch size: 472, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:28:53,080 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:29:19,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5324, 1.6126, 1.4289, 1.0672], device='cuda:1'), covar=tensor([0.1064, 0.0892, 0.0653, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.1429, 0.1214, 0.1201, 0.1295], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 22:29:24,590 INFO [optim.py:369] (1/2) 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 22:29:34,350 INFO [train.py:968] (1/2) Epoch 5, batch 37050, libri_loss[loss=0.3664, simple_loss=0.4136, pruned_loss=0.1596, over 29661.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3526, pruned_loss=0.1102, over 5679525.53 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3867, pruned_loss=0.1418, over 5654682.67 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3493, pruned_loss=0.1075, over 5687641.02 frames. ], batch size: 73, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:29:34,622 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=218967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:29:55,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:08,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2898, 1.9191, 1.4320, 0.4614], device='cuda:1'), covar=tensor([0.2230, 0.1212, 0.2131, 0.3049], device='cuda:1'), in_proj_covar=tensor([0.1367, 0.1292, 0.1354, 0.1143], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 22:30:10,994 INFO [train.py:968] (1/2) Epoch 5, batch 37100, giga_loss[loss=0.333, simple_loss=0.3864, pruned_loss=0.1398, over 28613.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3534, pruned_loss=0.1115, over 5689168.49 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3883, pruned_loss=0.1428, over 5660083.88 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.348, pruned_loss=0.1073, over 5691853.04 frames. ], batch size: 307, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:30:20,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:23,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:41,822 INFO [optim.py:369] (1/2) 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 22:30:46,958 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:30:51,202 INFO [train.py:968] (1/2) Epoch 5, batch 37150, giga_loss[loss=0.2477, simple_loss=0.3194, pruned_loss=0.08803, over 28936.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3497, pruned_loss=0.109, over 5690209.86 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3888, pruned_loss=0.143, over 5651219.36 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3445, pruned_loss=0.1051, over 5700892.94 frames. ], batch size: 186, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:31:26,421 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:28,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-02 22:31:29,124 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 37200, libri_loss[loss=0.384, simple_loss=0.4375, pruned_loss=0.1653, over 29194.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3468, pruned_loss=0.1073, over 5692892.39 frames. ], libri_tot_loss[loss=0.3383, simple_loss=0.3897, pruned_loss=0.1435, over 5644528.01 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3414, pruned_loss=0.1033, over 5708379.68 frames. ], batch size: 97, lr: 6.13e-03, grad_scale: 8.0
+2023-03-02 22:31:49,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:52,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:53,815 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:31:55,885 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-02 22:32:01,874 INFO [optim.py:369] (1/2) 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 22:32:11,013 INFO [train.py:968] (1/2) Epoch 5, batch 37250, libri_loss[loss=0.3887, simple_loss=0.428, pruned_loss=0.1747, over 29568.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.346, pruned_loss=0.1074, over 5692476.65 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3906, pruned_loss=0.144, over 5647712.73 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.34, pruned_loss=0.1031, over 5702498.76 frames. ], batch size: 79, lr: 6.13e-03, grad_scale: 8.0
+2023-03-02 22:32:13,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:32:14,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-02 22:32:48,926 INFO [train.py:968] (1/2) Epoch 5, batch 37300, giga_loss[loss=0.291, simple_loss=0.35, pruned_loss=0.1159, over 28978.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3461, pruned_loss=0.1079, over 5695078.72 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3919, pruned_loss=0.1447, over 5644054.44 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3384, pruned_loss=0.1023, over 5707361.06 frames. ], batch size: 136, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:32:52,946 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 22:33:21,641 INFO [optim.py:369] (1/2) 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 22:33:31,242 INFO [train.py:968] (1/2) Epoch 5, batch 37350, giga_loss[loss=0.2342, simple_loss=0.3091, pruned_loss=0.07963, over 28923.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3427, pruned_loss=0.1059, over 5701454.10 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3925, pruned_loss=0.145, over 5645117.58 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3352, pruned_loss=0.1005, over 5711135.26 frames. ], batch size: 213, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:33:51,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:34:03,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2172, 2.8252, 1.2883, 1.3017], device='cuda:1'), covar=tensor([0.0958, 0.0330, 0.0918, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0467, 0.0304, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 22:34:09,896 INFO [train.py:968] (1/2) Epoch 5, batch 37400, giga_loss[loss=0.2438, simple_loss=0.3066, pruned_loss=0.09048, over 28679.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3408, pruned_loss=0.1043, over 5712113.78 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3928, pruned_loss=0.1449, over 5648786.53 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3326, pruned_loss=0.09861, over 5718877.20 frames. ], batch size: 85, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:34:39,564 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:34:47,714 INFO [train.py:968] (1/2) Epoch 5, batch 37450, giga_loss[loss=0.2486, simple_loss=0.3231, pruned_loss=0.08709, over 29039.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3391, pruned_loss=0.1029, over 5721385.28 frames. ], libri_tot_loss[loss=0.3421, simple_loss=0.3936, pruned_loss=0.1452, over 5654274.54 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3306, pruned_loss=0.09707, over 5723251.21 frames. ], batch size: 155, lr: 6.13e-03, grad_scale: 4.0
+2023-03-02 22:34:55,619 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:35:29,073 INFO [train.py:968] (1/2) Epoch 5, batch 37500, giga_loss[loss=0.2197, simple_loss=0.3042, pruned_loss=0.06766, over 28862.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3408, pruned_loss=0.1042, over 5722326.25 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3941, pruned_loss=0.1455, over 5660692.42 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3324, pruned_loss=0.09834, over 5719714.31 frames. ], batch size: 174, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:35:42,024 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:35:46,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:35:51,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:04,111 INFO [optim.py:369] (1/2) 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 22:36:14,689 INFO [train.py:968] (1/2) Epoch 5, batch 37550, giga_loss[loss=0.2876, simple_loss=0.3576, pruned_loss=0.1088, over 28557.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3461, pruned_loss=0.1075, over 5719038.83 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3944, pruned_loss=0.1455, over 5664865.47 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3382, pruned_loss=0.1022, over 5714272.97 frames. ], batch size: 336, lr: 6.12e-03, grad_scale: 2.0
+2023-03-02 22:36:16,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:28,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:47,019 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:36:50,242 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 37600, giga_loss[loss=0.3098, simple_loss=0.3683, pruned_loss=0.1257, over 28555.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3532, pruned_loss=0.1123, over 5704827.98 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.3946, pruned_loss=0.1457, over 5668332.19 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.346, pruned_loss=0.1072, over 5698838.05 frames. ], batch size: 85, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:37:20,646 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:37:42,415 INFO [optim.py:369] (1/2) 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 22:37:51,476 INFO [train.py:968] (1/2) Epoch 5, batch 37650, giga_loss[loss=0.2813, simple_loss=0.3517, pruned_loss=0.1055, over 28829.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3617, pruned_loss=0.1183, over 5696133.32 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.3949, pruned_loss=0.1459, over 5670811.48 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3554, pruned_loss=0.1138, over 5689527.39 frames. ], batch size: 99, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:38:08,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2436, 2.8405, 1.4387, 1.2263], device='cuda:1'), covar=tensor([0.0840, 0.0323, 0.0779, 0.1276], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0469, 0.0305, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 22:38:36,165 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:38:40,689 INFO [train.py:968] (1/2) Epoch 5, batch 37700, giga_loss[loss=0.3333, simple_loss=0.397, pruned_loss=0.1348, over 29057.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3674, pruned_loss=0.1209, over 5678198.09 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3956, pruned_loss=0.1463, over 5667254.31 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3609, pruned_loss=0.1163, over 5676044.65 frames. ], batch size: 128, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:39:14,175 INFO [optim.py:369] (1/2) 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 22:39:23,218 INFO [train.py:968] (1/2) Epoch 5, batch 37750, giga_loss[loss=0.2984, simple_loss=0.371, pruned_loss=0.1129, over 28836.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3723, pruned_loss=0.1232, over 5673769.62 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.396, pruned_loss=0.1465, over 5659533.16 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3661, pruned_loss=0.1187, over 5678424.85 frames. ], batch size: 199, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:40:09,144 INFO [train.py:968] (1/2) Epoch 5, batch 37800, giga_loss[loss=0.3034, simple_loss=0.3788, pruned_loss=0.114, over 28588.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3779, pruned_loss=0.1268, over 5673994.79 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3956, pruned_loss=0.1463, over 5661839.42 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3731, pruned_loss=0.1232, over 5675713.01 frames. ], batch size: 60, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:40:37,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:40:40,215 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:40:43,542 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 37850, giga_loss[loss=0.2643, simple_loss=0.3379, pruned_loss=0.09539, over 28680.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3736, pruned_loss=0.1241, over 5674016.22 frames. ], libri_tot_loss[loss=0.3438, simple_loss=0.3953, pruned_loss=0.1461, over 5663203.22 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.37, pruned_loss=0.1214, over 5674297.80 frames. ], batch size: 262, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:40:54,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:41:26,987 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:41:38,459 INFO [train.py:968] (1/2) Epoch 5, batch 37900, giga_loss[loss=0.2584, simple_loss=0.3401, pruned_loss=0.08829, over 28530.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3695, pruned_loss=0.12, over 5679400.70 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3952, pruned_loss=0.1461, over 5663791.21 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3668, pruned_loss=0.1178, over 5679199.58 frames. ], batch size: 71, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:41:41,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6937, 1.0884, 2.8102, 2.6579], device='cuda:1'), covar=tensor([0.1632, 0.2215, 0.0528, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0571, 0.0527, 0.0735, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 22:42:10,711 INFO [optim.py:369] (1/2) 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 22:42:11,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:19,723 INFO [train.py:968] (1/2) Epoch 5, batch 37950, giga_loss[loss=0.2613, simple_loss=0.3392, pruned_loss=0.09171, over 28694.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3695, pruned_loss=0.1198, over 5678320.89 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3959, pruned_loss=0.1467, over 5660106.15 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3658, pruned_loss=0.1166, over 5681637.55 frames. ], batch size: 92, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:42:41,882 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:44,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:46,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-02 22:42:47,274 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=219883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:42:58,924 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:02,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:03,109 INFO [train.py:968] (1/2) Epoch 5, batch 38000, giga_loss[loss=0.3113, simple_loss=0.3838, pruned_loss=0.1194, over 28875.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3698, pruned_loss=0.1192, over 5684205.13 frames. ], libri_tot_loss[loss=0.3448, simple_loss=0.3961, pruned_loss=0.1468, over 5660899.62 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3667, pruned_loss=0.1165, over 5686189.76 frames. ], batch size: 106, lr: 6.12e-03, grad_scale: 8.0
+2023-03-02 22:43:11,247 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:29,100 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:32,990 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:35,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:43:39,350 INFO [optim.py:369] (1/2) 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 22:43:46,927 INFO [train.py:968] (1/2) Epoch 5, batch 38050, giga_loss[loss=0.2648, simple_loss=0.3395, pruned_loss=0.09501, over 28523.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3717, pruned_loss=0.1199, over 5684747.01 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3963, pruned_loss=0.1469, over 5663099.79 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3688, pruned_loss=0.1175, over 5684525.37 frames. ], batch size: 85, lr: 6.12e-03, grad_scale: 8.0
+2023-03-02 22:44:00,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:06,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:15,776 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:20,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:44:33,165 INFO [train.py:968] (1/2) Epoch 5, batch 38100, giga_loss[loss=0.2975, simple_loss=0.372, pruned_loss=0.1115, over 28905.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.375, pruned_loss=0.1228, over 5680412.97 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3967, pruned_loss=0.1472, over 5658963.03 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3719, pruned_loss=0.1201, over 5684179.58 frames. ], batch size: 186, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:44:46,338 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:45:04,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-02 22:45:06,999 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 38150, giga_loss[loss=0.298, simple_loss=0.3687, pruned_loss=0.1137, over 28850.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3767, pruned_loss=0.1241, over 5686946.26 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.3971, pruned_loss=0.1472, over 5660964.34 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3734, pruned_loss=0.1214, over 5688739.90 frames. ], batch size: 145, lr: 6.12e-03, grad_scale: 4.0
+2023-03-02 22:45:58,957 INFO [train.py:968] (1/2) Epoch 5, batch 38200, giga_loss[loss=0.2724, simple_loss=0.3488, pruned_loss=0.09796, over 29071.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3775, pruned_loss=0.1251, over 5687472.27 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.3975, pruned_loss=0.1474, over 5663343.78 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.374, pruned_loss=0.1223, over 5686943.09 frames. ], batch size: 128, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:46:06,675 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:46:09,782 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:46:11,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:46:34,141 INFO [optim.py:369] (1/2) 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 22:46:38,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:46:40,649 INFO [train.py:968] (1/2) Epoch 5, batch 38250, giga_loss[loss=0.3216, simple_loss=0.3898, pruned_loss=0.1267, over 28740.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3791, pruned_loss=0.1261, over 5694030.66 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3986, pruned_loss=0.1481, over 5664415.28 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3748, pruned_loss=0.1226, over 5693543.97 frames. ], batch size: 92, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:46:44,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:47:18,182 INFO [train.py:968] (1/2) Epoch 5, batch 38300, giga_loss[loss=0.2764, simple_loss=0.3554, pruned_loss=0.0987, over 28271.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3769, pruned_loss=0.1237, over 5699979.53 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3974, pruned_loss=0.1473, over 5665300.26 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3738, pruned_loss=0.1208, over 5699641.96 frames. ], batch size: 65, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:47:43,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9915, 1.3021, 1.0424, 0.2413], device='cuda:1'), covar=tensor([0.1796, 0.1363, 0.2467, 0.2909], device='cuda:1'), in_proj_covar=tensor([0.1361, 0.1277, 0.1350, 0.1138], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-02 22:47:54,569 INFO [optim.py:369] (1/2) 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 22:48:00,842 INFO [train.py:968] (1/2) Epoch 5, batch 38350, giga_loss[loss=0.2837, simple_loss=0.3668, pruned_loss=0.1003, over 28942.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3772, pruned_loss=0.1224, over 5702214.30 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.3976, pruned_loss=0.1473, over 5668347.40 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3742, pruned_loss=0.1196, over 5699764.64 frames. ], batch size: 164, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:48:01,349 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:48:03,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220253.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 22:48:05,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220256.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 22:48:07,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:48:29,253 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 38400, giga_loss[loss=0.2771, simple_loss=0.3516, pruned_loss=0.1013, over 28700.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3769, pruned_loss=0.1217, over 5707672.70 frames. ], libri_tot_loss[loss=0.347, simple_loss=0.3982, pruned_loss=0.1479, over 5671685.95 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3737, pruned_loss=0.1186, over 5703114.36 frames. ], batch size: 262, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:48:53,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1381, 1.9036, 1.4382, 1.7933], device='cuda:1'), covar=tensor([0.0657, 0.0647, 0.0968, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0441, 0.0506, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 22:49:03,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7372, 1.8217, 1.7346, 1.6371], device='cuda:1'), covar=tensor([0.1281, 0.1729, 0.1637, 0.1681], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0728, 0.0627, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 22:49:12,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3553, 1.6726, 1.2098, 1.4938], device='cuda:1'), covar=tensor([0.0706, 0.0271, 0.0342, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0122, 0.0125, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-02 22:49:15,401 INFO [optim.py:369] (1/2) 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 22:49:24,915 INFO [train.py:968] (1/2) Epoch 5, batch 38450, giga_loss[loss=0.2858, simple_loss=0.3575, pruned_loss=0.107, over 28621.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3735, pruned_loss=0.12, over 5711099.58 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3978, pruned_loss=0.1476, over 5678741.59 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3706, pruned_loss=0.117, over 5701889.90 frames. ], batch size: 78, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:49:25,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2514, 2.5359, 1.1901, 1.2563], device='cuda:1'), covar=tensor([0.0864, 0.0272, 0.0850, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0463, 0.0303, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0024, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 22:49:34,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-02 22:50:04,770 INFO [train.py:968] (1/2) Epoch 5, batch 38500, giga_loss[loss=0.2805, simple_loss=0.3577, pruned_loss=0.1016, over 28629.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3719, pruned_loss=0.1191, over 5710018.32 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.398, pruned_loss=0.1477, over 5672996.95 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.369, pruned_loss=0.1161, over 5708012.96 frames. ], batch size: 85, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:50:05,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:50:08,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:50:14,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2006, 1.2723, 1.1747, 0.8663], device='cuda:1'), covar=tensor([0.1169, 0.1023, 0.0739, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.1415, 0.1223, 0.1212, 0.1305], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 22:50:31,688 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:50:37,441 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 38550, giga_loss[loss=0.3033, simple_loss=0.3711, pruned_loss=0.1177, over 29034.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3691, pruned_loss=0.1175, over 5713686.85 frames. ], libri_tot_loss[loss=0.3462, simple_loss=0.3975, pruned_loss=0.1475, over 5675312.49 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.367, pruned_loss=0.1151, over 5710464.91 frames. ], batch size: 155, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:51:11,132 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:51:19,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:51:27,134 INFO [train.py:968] (1/2) Epoch 5, batch 38600, giga_loss[loss=0.2817, simple_loss=0.3584, pruned_loss=0.1025, over 28697.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.371, pruned_loss=0.1196, over 5717915.43 frames. ], libri_tot_loss[loss=0.347, simple_loss=0.3981, pruned_loss=0.148, over 5681376.14 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.368, pruned_loss=0.1164, over 5710736.00 frames. ], batch size: 60, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:51:52,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:52:01,775 INFO [optim.py:369] (1/2) 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 22:52:08,193 INFO [train.py:968] (1/2) Epoch 5, batch 38650, giga_loss[loss=0.2847, simple_loss=0.3632, pruned_loss=0.1031, over 28908.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3707, pruned_loss=0.1191, over 5717546.41 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3979, pruned_loss=0.1479, over 5684085.15 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.368, pruned_loss=0.1161, over 5710055.27 frames. ], batch size: 174, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:52:45,646 INFO [train.py:968] (1/2) Epoch 5, batch 38700, libri_loss[loss=0.349, simple_loss=0.4042, pruned_loss=0.1469, over 29267.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3703, pruned_loss=0.1176, over 5714116.16 frames. ], libri_tot_loss[loss=0.3467, simple_loss=0.3978, pruned_loss=0.1478, over 5683303.33 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3678, pruned_loss=0.1149, over 5709546.46 frames. ], batch size: 94, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:53:02,055 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:04,107 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 5, batch 38750, giga_loss[loss=0.2869, simple_loss=0.356, pruned_loss=0.1089, over 28927.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3689, pruned_loss=0.1163, over 5718615.62 frames. ], libri_tot_loss[loss=0.3464, simple_loss=0.3977, pruned_loss=0.1476, over 5689752.93 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3663, pruned_loss=0.1134, over 5709972.69 frames. ], batch size: 106, lr: 6.11e-03, grad_scale: 4.0
+2023-03-02 22:53:24,191 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:42,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:44,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:53:59,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-02 22:54:06,188 INFO [train.py:968] (1/2) Epoch 5, batch 38800, giga_loss[loss=0.2605, simple_loss=0.3359, pruned_loss=0.09255, over 28448.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3675, pruned_loss=0.1158, over 5715883.26 frames. ], libri_tot_loss[loss=0.3461, simple_loss=0.3973, pruned_loss=0.1474, over 5690825.24 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3653, pruned_loss=0.1133, over 5708521.01 frames. ], batch size: 85, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:54:10,183 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:54:46,632 INFO [train.py:968] (1/2) Epoch 5, batch 38850, giga_loss[loss=0.2832, simple_loss=0.355, pruned_loss=0.1057, over 28880.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3651, pruned_loss=0.1146, over 5711999.47 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.398, pruned_loss=0.1479, over 5691785.13 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3624, pruned_loss=0.1118, over 5705812.96 frames. ], batch size: 284, lr: 6.11e-03, grad_scale: 8.0
+2023-03-02 22:55:03,358 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:55:06,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:55:28,130 INFO [train.py:968] (1/2) Epoch 5, batch 38900, giga_loss[loss=0.2607, simple_loss=0.3292, pruned_loss=0.09613, over 29083.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3625, pruned_loss=0.1136, over 5702722.93 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3985, pruned_loss=0.1483, over 5683647.61 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3589, pruned_loss=0.11, over 5706138.82 frames. ], batch size: 106, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:55:28,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:55:32,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-02 22:55:45,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6009, 2.8938, 1.8001, 1.6803], device='cuda:1'), covar=tensor([0.1128, 0.0724, 0.0879, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1222, 0.1203, 0.1304], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 22:56:00,070 INFO [optim.py:369] (1/2) 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 22:56:07,115 INFO [train.py:968] (1/2) Epoch 5, batch 38950, giga_loss[loss=0.2368, simple_loss=0.315, pruned_loss=0.07934, over 28471.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3616, pruned_loss=0.1134, over 5700825.86 frames. ], libri_tot_loss[loss=0.348, simple_loss=0.3988, pruned_loss=0.1486, over 5682623.84 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.358, pruned_loss=0.1098, over 5704652.44 frames. ], batch size: 60, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:56:12,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:56:22,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:56:34,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3464, 1.5812, 1.3561, 1.4964], device='cuda:1'), covar=tensor([0.0775, 0.0319, 0.0314, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0121, 0.0125, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0058, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-02 22:56:51,051 INFO [train.py:968] (1/2) Epoch 5, batch 39000, giga_loss[loss=0.2445, simple_loss=0.3244, pruned_loss=0.08236, over 28897.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3615, pruned_loss=0.1139, over 5697576.42 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.3989, pruned_loss=0.1486, over 5684705.89 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3583, pruned_loss=0.1108, over 5698888.06 frames. ], batch size: 174, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:56:51,051 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 22:57:00,201 INFO [train.py:1012] (1/2) Epoch 5, validation: loss=0.2426, simple_loss=0.3456, pruned_loss=0.0698, over 944034.00 frames. 
+2023-03-02 22:57:00,201 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-02 22:57:02,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3152, 4.1184, 3.9331, 1.7746], device='cuda:1'), covar=tensor([0.0362, 0.0453, 0.0590, 0.2287], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0789, 0.0775, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 22:57:31,903 INFO [optim.py:369] (1/2) 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 22:57:36,918 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=220948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:57:38,377 INFO [train.py:968] (1/2) Epoch 5, batch 39050, giga_loss[loss=0.2913, simple_loss=0.3519, pruned_loss=0.1153, over 28463.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5702964.90 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3983, pruned_loss=0.1484, over 5688270.74 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.357, pruned_loss=0.1106, over 5701163.38 frames. ], batch size: 65, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:58:16,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:17,358 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:18,432 INFO [train.py:968] (1/2) Epoch 5, batch 39100, giga_loss[loss=0.24, simple_loss=0.3123, pruned_loss=0.08383, over 28987.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3574, pruned_loss=0.1126, over 5710922.49 frames. ], libri_tot_loss[loss=0.3476, simple_loss=0.3984, pruned_loss=0.1484, over 5691735.53 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3542, pruned_loss=0.1095, over 5706884.03 frames. ], batch size: 106, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 22:58:19,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:25,513 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:27,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:37,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:41,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:50,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 22:58:51,118 INFO [optim.py:369] (1/2) 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 22:58:51,912 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 39150, giga_loss[loss=0.2568, simple_loss=0.328, pruned_loss=0.09282, over 28987.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.356, pruned_loss=0.1124, over 5707983.10 frames. ], libri_tot_loss[loss=0.3477, simple_loss=0.3985, pruned_loss=0.1484, over 5693205.18 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3525, pruned_loss=0.1091, over 5703717.68 frames. ], batch size: 213, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 22:59:07,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5147, 3.7655, 1.6236, 1.4190], device='cuda:1'), covar=tensor([0.0799, 0.0315, 0.0858, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0466, 0.0305, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 22:59:40,277 INFO [train.py:968] (1/2) Epoch 5, batch 39200, giga_loss[loss=0.326, simple_loss=0.3768, pruned_loss=0.1376, over 23428.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3548, pruned_loss=0.1117, over 5707725.98 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3983, pruned_loss=0.1483, over 5694065.64 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3514, pruned_loss=0.1085, over 5703626.17 frames. ], batch size: 705, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:00:00,329 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-02 23:00:11,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9222, 1.0984, 0.9023, 0.7041], device='cuda:1'), covar=tensor([0.1061, 0.0980, 0.0677, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1243, 0.1227, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 23:00:14,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:16,829 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:00:25,117 INFO [train.py:968] (1/2) Epoch 5, batch 39250, giga_loss[loss=0.3051, simple_loss=0.3722, pruned_loss=0.119, over 29015.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3572, pruned_loss=0.1126, over 5705393.71 frames. ], libri_tot_loss[loss=0.3477, simple_loss=0.3985, pruned_loss=0.1485, over 5697044.40 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3538, pruned_loss=0.1095, over 5699851.35 frames. ], batch size: 155, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:00:42,351 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:45,032 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:45,057 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:00:47,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2488, 1.9711, 1.5318, 1.7679], device='cuda:1'), covar=tensor([0.0613, 0.0693, 0.0946, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0444, 0.0501, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 23:01:09,652 INFO [train.py:968] (1/2) Epoch 5, batch 39300, giga_loss[loss=0.2481, simple_loss=0.335, pruned_loss=0.08055, over 28921.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3599, pruned_loss=0.114, over 5700454.21 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.3982, pruned_loss=0.1483, over 5700210.23 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3569, pruned_loss=0.1112, over 5693104.29 frames. ], batch size: 174, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:01:10,775 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:01:22,718 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:01:45,673 INFO [optim.py:369] (1/2) 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 23:01:51,518 INFO [train.py:968] (1/2) Epoch 5, batch 39350, giga_loss[loss=0.2978, simple_loss=0.3574, pruned_loss=0.1191, over 28582.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3633, pruned_loss=0.1156, over 5698955.04 frames. ], libri_tot_loss[loss=0.3476, simple_loss=0.3984, pruned_loss=0.1483, over 5702530.35 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3595, pruned_loss=0.1122, over 5690790.23 frames. ], batch size: 92, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:01:53,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7422, 2.2121, 1.8251, 2.2632], device='cuda:1'), covar=tensor([0.0498, 0.0538, 0.0806, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0445, 0.0501, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 23:02:32,899 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:02:34,702 INFO [train.py:968] (1/2) Epoch 5, batch 39400, libri_loss[loss=0.418, simple_loss=0.445, pruned_loss=0.1955, over 19341.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3636, pruned_loss=0.1151, over 5693532.45 frames. ], libri_tot_loss[loss=0.347, simple_loss=0.398, pruned_loss=0.148, over 5698613.94 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.36, pruned_loss=0.1117, over 5691380.96 frames. ], batch size: 188, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:02:54,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:03:11,963 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 39450, giga_loss[loss=0.2905, simple_loss=0.3607, pruned_loss=0.1101, over 28929.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3642, pruned_loss=0.1153, over 5692822.01 frames. ], libri_tot_loss[loss=0.3481, simple_loss=0.3989, pruned_loss=0.1486, over 5693030.95 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3596, pruned_loss=0.1111, over 5695848.54 frames. ], batch size: 145, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:03:56,710 INFO [train.py:968] (1/2) Epoch 5, batch 39500, giga_loss[loss=0.2653, simple_loss=0.339, pruned_loss=0.09576, over 28456.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3637, pruned_loss=0.1155, over 5681550.57 frames. ], libri_tot_loss[loss=0.3477, simple_loss=0.3986, pruned_loss=0.1485, over 5681618.49 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3592, pruned_loss=0.1112, over 5694825.73 frames. ], batch size: 65, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:04:12,410 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:04:31,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1949, 1.4382, 1.2258, 0.8807], device='cuda:1'), covar=tensor([0.1172, 0.0893, 0.0629, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.1416, 0.1223, 0.1219, 0.1313], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 23:04:32,396 INFO [optim.py:369] (1/2) 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 23:04:36,143 INFO [train.py:968] (1/2) Epoch 5, batch 39550, giga_loss[loss=0.279, simple_loss=0.3498, pruned_loss=0.1041, over 29013.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3635, pruned_loss=0.1149, over 5700979.83 frames. ], libri_tot_loss[loss=0.3469, simple_loss=0.3981, pruned_loss=0.1478, over 5689168.15 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3588, pruned_loss=0.1107, over 5705351.36 frames. ], batch size: 128, lr: 6.10e-03, grad_scale: 4.0
+2023-03-02 23:04:48,914 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:04:51,279 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:05:19,300 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:05:20,359 INFO [train.py:968] (1/2) Epoch 5, batch 39600, libri_loss[loss=0.35, simple_loss=0.4029, pruned_loss=0.1485, over 29533.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3642, pruned_loss=0.1153, over 5705360.41 frames. ], libri_tot_loss[loss=0.3471, simple_loss=0.3982, pruned_loss=0.1479, over 5683293.19 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3598, pruned_loss=0.1113, over 5713472.53 frames. ], batch size: 82, lr: 6.10e-03, grad_scale: 8.0
+2023-03-02 23:05:23,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2926, 2.9123, 1.3959, 1.2471], device='cuda:1'), covar=tensor([0.0861, 0.0356, 0.0835, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0330, 0.0475, 0.0305, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 23:05:56,772 INFO [optim.py:369] (1/2) 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 23:06:02,134 INFO [train.py:968] (1/2) Epoch 5, batch 39650, giga_loss[loss=0.3029, simple_loss=0.3741, pruned_loss=0.1158, over 28939.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3702, pruned_loss=0.1197, over 5699305.09 frames. ], libri_tot_loss[loss=0.3474, simple_loss=0.3982, pruned_loss=0.1482, over 5681668.99 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3653, pruned_loss=0.115, over 5708094.65 frames. ], batch size: 136, lr: 6.09e-03, grad_scale: 8.0
+2023-03-02 23:06:11,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:13,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:33,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:36,356 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:06:40,358 INFO [train.py:968] (1/2) Epoch 5, batch 39700, giga_loss[loss=0.2919, simple_loss=0.3695, pruned_loss=0.1071, over 28937.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3714, pruned_loss=0.1197, over 5706289.69 frames. ], libri_tot_loss[loss=0.347, simple_loss=0.3981, pruned_loss=0.1479, over 5686137.05 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3671, pruned_loss=0.1157, over 5709519.24 frames. ], batch size: 227, lr: 6.09e-03, grad_scale: 8.0
+2023-03-02 23:06:53,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2525, 1.4565, 1.2041, 1.3157], device='cuda:1'), covar=tensor([0.2073, 0.2001, 0.2156, 0.1980], device='cuda:1'), in_proj_covar=tensor([0.1126, 0.0864, 0.0995, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 23:07:14,344 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 39750, giga_loss[loss=0.2966, simple_loss=0.3713, pruned_loss=0.1109, over 28837.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3725, pruned_loss=0.1201, over 5707599.05 frames. ], libri_tot_loss[loss=0.3475, simple_loss=0.3988, pruned_loss=0.1481, over 5690098.89 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3677, pruned_loss=0.1159, over 5707084.31 frames. ], batch size: 199, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:07:37,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:08:01,241 INFO [train.py:968] (1/2) Epoch 5, batch 39800, giga_loss[loss=0.2964, simple_loss=0.371, pruned_loss=0.1109, over 28635.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.373, pruned_loss=0.1201, over 5699245.98 frames. ], libri_tot_loss[loss=0.3483, simple_loss=0.3995, pruned_loss=0.1486, over 5682580.55 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3681, pruned_loss=0.1158, over 5706133.64 frames. ], batch size: 336, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:08:20,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3558, 1.2067, 4.8211, 3.5354], device='cuda:1'), covar=tensor([0.1584, 0.2411, 0.0290, 0.0615], device='cuda:1'), in_proj_covar=tensor([0.0577, 0.0533, 0.0755, 0.0624], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 23:08:27,417 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:08:29,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:08:36,104 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 39850, giga_loss[loss=0.3031, simple_loss=0.3682, pruned_loss=0.119, over 29087.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3741, pruned_loss=0.1209, over 5705132.92 frames. ], libri_tot_loss[loss=0.3488, simple_loss=0.4, pruned_loss=0.1488, over 5686129.82 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.369, pruned_loss=0.1166, over 5708272.92 frames. ], batch size: 128, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:08:51,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:09:20,197 INFO [train.py:968] (1/2) Epoch 5, batch 39900, giga_loss[loss=0.2958, simple_loss=0.3562, pruned_loss=0.1177, over 28688.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3724, pruned_loss=0.1199, over 5709867.02 frames. ], libri_tot_loss[loss=0.3487, simple_loss=0.4001, pruned_loss=0.1487, over 5688156.77 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3681, pruned_loss=0.1163, over 5710742.43 frames. ], batch size: 92, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:09:33,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:09:35,067 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:09:59,619 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 39950, libri_loss[loss=0.3594, simple_loss=0.4098, pruned_loss=0.1546, over 29392.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3687, pruned_loss=0.118, over 5706674.47 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.4003, pruned_loss=0.1489, over 5680256.68 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3649, pruned_loss=0.1147, over 5714126.37 frames. ], batch size: 92, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:10:38,116 INFO [train.py:968] (1/2) Epoch 5, batch 40000, giga_loss[loss=0.2904, simple_loss=0.346, pruned_loss=0.1174, over 23883.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3685, pruned_loss=0.1188, over 5700622.58 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.4014, pruned_loss=0.1497, over 5680399.34 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3624, pruned_loss=0.1135, over 5707774.16 frames. ], batch size: 705, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:10:38,362 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=221900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:11:15,239 INFO [optim.py:369] (1/2) 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 23:11:18,839 INFO [train.py:968] (1/2) Epoch 5, batch 40050, giga_loss[loss=0.2968, simple_loss=0.3744, pruned_loss=0.1096, over 28884.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3671, pruned_loss=0.1172, over 5698900.95 frames. ], libri_tot_loss[loss=0.3511, simple_loss=0.402, pruned_loss=0.1502, over 5671399.29 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3615, pruned_loss=0.1123, over 5713356.01 frames. ], batch size: 199, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:11:40,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0702, 1.1456, 3.9270, 3.1748], device='cuda:1'), covar=tensor([0.1576, 0.2272, 0.0337, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0567, 0.0525, 0.0746, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 23:11:52,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8692, 1.1253, 3.4971, 3.0111], device='cuda:1'), covar=tensor([0.1633, 0.2312, 0.0361, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.0565, 0.0523, 0.0743, 0.0612], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-02 23:11:59,592 INFO [train.py:968] (1/2) Epoch 5, batch 40100, giga_loss[loss=0.2973, simple_loss=0.3753, pruned_loss=0.1096, over 28733.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3687, pruned_loss=0.1158, over 5703008.03 frames. ], libri_tot_loss[loss=0.3513, simple_loss=0.4021, pruned_loss=0.1502, over 5673209.57 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3637, pruned_loss=0.1115, over 5713235.16 frames. ], batch size: 242, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:12:35,031 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 40150, giga_loss[loss=0.3214, simple_loss=0.3901, pruned_loss=0.1264, over 28559.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3698, pruned_loss=0.1167, over 5697377.24 frames. ], libri_tot_loss[loss=0.3509, simple_loss=0.4019, pruned_loss=0.15, over 5670817.41 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3647, pruned_loss=0.1122, over 5709158.53 frames. ], batch size: 336, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:12:44,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2548, 3.1068, 2.9287, 1.3361], device='cuda:1'), covar=tensor([0.0772, 0.0773, 0.1038, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.0844, 0.0785, 0.0776, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 23:13:19,322 INFO [train.py:968] (1/2) Epoch 5, batch 40200, giga_loss[loss=0.332, simple_loss=0.3927, pruned_loss=0.1357, over 28942.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3682, pruned_loss=0.1168, over 5701613.32 frames. ], libri_tot_loss[loss=0.3508, simple_loss=0.4019, pruned_loss=0.1499, over 5673171.41 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3639, pruned_loss=0.113, over 5709039.83 frames. ], batch size: 227, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:13:42,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4703, 3.6591, 1.4744, 1.4852], device='cuda:1'), covar=tensor([0.0879, 0.0369, 0.0877, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0328, 0.0472, 0.0304, 0.0372], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 23:13:53,683 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 40250, giga_loss[loss=0.3245, simple_loss=0.391, pruned_loss=0.129, over 28737.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3681, pruned_loss=0.1183, over 5698604.31 frames. ], libri_tot_loss[loss=0.3501, simple_loss=0.4014, pruned_loss=0.1494, over 5670241.53 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3637, pruned_loss=0.1143, over 5709109.59 frames. ], batch size: 242, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:13:57,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:14:41,613 INFO [train.py:968] (1/2) Epoch 5, batch 40300, giga_loss[loss=0.2522, simple_loss=0.3272, pruned_loss=0.0886, over 28980.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3663, pruned_loss=0.1186, over 5698266.08 frames. ], libri_tot_loss[loss=0.3498, simple_loss=0.4011, pruned_loss=0.1492, over 5671257.33 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3628, pruned_loss=0.1155, over 5705679.30 frames. ], batch size: 164, lr: 6.09e-03, grad_scale: 4.0
+2023-03-02 23:14:51,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5186, 1.7515, 1.4333, 2.3341], device='cuda:1'), covar=tensor([0.2198, 0.2013, 0.2035, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.0856, 0.0988, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 23:15:17,721 INFO [optim.py:369] (1/2) 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 23:15:21,236 INFO [train.py:968] (1/2) Epoch 5, batch 40350, giga_loss[loss=0.2747, simple_loss=0.3304, pruned_loss=0.1095, over 28522.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3638, pruned_loss=0.1179, over 5709772.92 frames. ], libri_tot_loss[loss=0.3496, simple_loss=0.4009, pruned_loss=0.1492, over 5675808.80 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3607, pruned_loss=0.115, over 5712236.46 frames. ], batch size: 92, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:15:42,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:15:49,133 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 40400, giga_loss[loss=0.2597, simple_loss=0.3356, pruned_loss=0.0919, over 29022.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3624, pruned_loss=0.1171, over 5704733.24 frames. ], libri_tot_loss[loss=0.3501, simple_loss=0.4012, pruned_loss=0.1495, over 5667260.28 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3592, pruned_loss=0.1141, over 5715646.19 frames. ], batch size: 128, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:16:39,479 INFO [optim.py:369] (1/2) 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 23:16:42,676 INFO [train.py:968] (1/2) Epoch 5, batch 40450, giga_loss[loss=0.2367, simple_loss=0.3177, pruned_loss=0.07789, over 29063.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3588, pruned_loss=0.1152, over 5703472.18 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.4015, pruned_loss=0.1497, over 5662843.95 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.355, pruned_loss=0.1118, over 5717963.45 frames. ], batch size: 155, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:17:22,580 INFO [train.py:968] (1/2) Epoch 5, batch 40500, giga_loss[loss=0.2489, simple_loss=0.3199, pruned_loss=0.08898, over 28807.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3551, pruned_loss=0.1133, over 5704495.97 frames. ], libri_tot_loss[loss=0.3504, simple_loss=0.4014, pruned_loss=0.1497, over 5660816.21 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3501, pruned_loss=0.109, over 5720162.20 frames. ], batch size: 119, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:17:24,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5811, 3.5721, 1.6226, 1.5391], device='cuda:1'), covar=tensor([0.0771, 0.0316, 0.0802, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0483, 0.0310, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0026, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 23:17:25,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:17:38,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:17:40,384 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 5, batch 40550, giga_loss[loss=0.2536, simple_loss=0.3352, pruned_loss=0.08605, over 28915.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3524, pruned_loss=0.1113, over 5703506.93 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4012, pruned_loss=0.1494, over 5665616.39 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3474, pruned_loss=0.1072, over 5712499.01 frames. ], batch size: 213, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:18:03,142 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 40600, giga_loss[loss=0.2685, simple_loss=0.3485, pruned_loss=0.09427, over 28886.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3558, pruned_loss=0.1132, over 5708895.56 frames. ], libri_tot_loss[loss=0.3497, simple_loss=0.4009, pruned_loss=0.1493, over 5675150.77 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3503, pruned_loss=0.1086, over 5708921.08 frames. ], batch size: 174, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:18:53,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8091, 2.1599, 1.8564, 1.8624], device='cuda:1'), covar=tensor([0.0714, 0.0252, 0.0291, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0121, 0.0126, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0058, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-02 23:19:00,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:19:16,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-02 23:19:18,770 INFO [optim.py:369] (1/2) 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 23:19:21,051 INFO [train.py:968] (1/2) Epoch 5, batch 40650, giga_loss[loss=0.3348, simple_loss=0.3984, pruned_loss=0.1356, over 28208.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3596, pruned_loss=0.1148, over 5701596.49 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.4007, pruned_loss=0.1492, over 5666130.35 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3547, pruned_loss=0.1107, over 5709725.86 frames. ], batch size: 367, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:19:35,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2072, 1.3460, 1.4762, 1.4110], device='cuda:1'), covar=tensor([0.0875, 0.0890, 0.1141, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0737, 0.0641, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 23:19:45,256 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:20:01,170 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:20:02,346 INFO [train.py:968] (1/2) Epoch 5, batch 40700, giga_loss[loss=0.3094, simple_loss=0.3718, pruned_loss=0.1235, over 28687.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1154, over 5693842.00 frames. ], libri_tot_loss[loss=0.3494, simple_loss=0.4005, pruned_loss=0.1491, over 5655980.98 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3578, pruned_loss=0.1116, over 5710743.54 frames. ], batch size: 99, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:20:43,406 INFO [optim.py:369] (1/2) 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 23:20:45,394 INFO [train.py:968] (1/2) Epoch 5, batch 40750, libri_loss[loss=0.3118, simple_loss=0.3581, pruned_loss=0.1328, over 28674.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3645, pruned_loss=0.1161, over 5706992.49 frames. ], libri_tot_loss[loss=0.3496, simple_loss=0.4007, pruned_loss=0.1492, over 5658730.88 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3605, pruned_loss=0.1126, over 5718510.19 frames. ], batch size: 63, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:20:51,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:01,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:03,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:27,463 INFO [train.py:968] (1/2) Epoch 5, batch 40800, giga_loss[loss=0.2764, simple_loss=0.354, pruned_loss=0.09935, over 28898.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3685, pruned_loss=0.1188, over 5703446.05 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.4008, pruned_loss=0.1491, over 5663573.98 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3645, pruned_loss=0.1155, over 5709156.28 frames. ], batch size: 145, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:21:28,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:21:44,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2561, 1.2374, 0.9946, 1.4335], device='cuda:1'), covar=tensor([0.0780, 0.0340, 0.0373, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0123, 0.0127, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0039, 0.0065], device='cuda:1')
+2023-03-02 23:22:12,560 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 40850, giga_loss[loss=0.3395, simple_loss=0.3979, pruned_loss=0.1405, over 28924.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3729, pruned_loss=0.1227, over 5700706.23 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.401, pruned_loss=0.149, over 5667860.69 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3689, pruned_loss=0.1195, over 5702077.95 frames. ], batch size: 227, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:22:43,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:22:43,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-02 23:23:04,407 INFO [train.py:968] (1/2) Epoch 5, batch 40900, giga_loss[loss=0.3485, simple_loss=0.3966, pruned_loss=0.1502, over 28098.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3815, pruned_loss=0.1306, over 5679599.99 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4013, pruned_loss=0.1493, over 5669071.56 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3775, pruned_loss=0.1273, over 5680126.54 frames. ], batch size: 77, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:23:05,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:23:07,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:23:38,828 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:23:50,347 INFO [optim.py:369] (1/2) 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 23:23:54,002 INFO [train.py:968] (1/2) Epoch 5, batch 40950, giga_loss[loss=0.3654, simple_loss=0.4126, pruned_loss=0.1591, over 27876.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3887, pruned_loss=0.1359, over 5684204.27 frames. ], libri_tot_loss[loss=0.3499, simple_loss=0.4013, pruned_loss=0.1493, over 5670309.20 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3855, pruned_loss=0.1332, over 5683485.93 frames. ], batch size: 412, lr: 6.08e-03, grad_scale: 8.0
+2023-03-02 23:24:39,036 INFO [train.py:968] (1/2) Epoch 5, batch 41000, giga_loss[loss=0.3556, simple_loss=0.4054, pruned_loss=0.1529, over 29073.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3948, pruned_loss=0.1422, over 5676947.66 frames. ], libri_tot_loss[loss=0.3493, simple_loss=0.4006, pruned_loss=0.149, over 5677303.67 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3926, pruned_loss=0.14, over 5670344.50 frames. ], batch size: 155, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:24:52,479 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:24:55,479 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:24:58,255 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:25:21,880 INFO [train.py:968] (1/2) Epoch 5, batch 41050, giga_loss[loss=0.4128, simple_loss=0.4447, pruned_loss=0.1904, over 27910.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.4004, pruned_loss=0.147, over 5680109.87 frames. ], libri_tot_loss[loss=0.3492, simple_loss=0.4007, pruned_loss=0.1488, over 5676968.80 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.3984, pruned_loss=0.1452, over 5674501.94 frames. ], batch size: 412, lr: 6.08e-03, grad_scale: 4.0
+2023-03-02 23:25:24,545 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:26,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:33,848 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=222963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:25:35,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-02 23:25:41,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:26:09,826 INFO [train.py:968] (1/2) Epoch 5, batch 41100, giga_loss[loss=0.3987, simple_loss=0.4294, pruned_loss=0.1839, over 27449.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4052, pruned_loss=0.1514, over 5655479.61 frames. ], libri_tot_loss[loss=0.3495, simple_loss=0.4009, pruned_loss=0.149, over 5669622.19 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4033, pruned_loss=0.1498, over 5658168.21 frames. ], batch size: 472, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:26:58,658 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 41150, giga_loss[loss=0.3103, simple_loss=0.3774, pruned_loss=0.1216, over 29025.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.405, pruned_loss=0.1514, over 5653951.81 frames. ], libri_tot_loss[loss=0.349, simple_loss=0.4006, pruned_loss=0.1487, over 5662752.41 frames. ], giga_tot_loss[loss=0.3524, simple_loss=0.404, pruned_loss=0.1504, over 5662631.81 frames. ], batch size: 128, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:27:42,474 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-02 23:27:55,575 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:00,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:00,824 INFO [train.py:968] (1/2) Epoch 5, batch 41200, giga_loss[loss=0.3596, simple_loss=0.4175, pruned_loss=0.1509, over 28776.00 frames. ], tot_loss[loss=0.3615, simple_loss=0.4094, pruned_loss=0.1567, over 5625398.81 frames. ], libri_tot_loss[loss=0.3489, simple_loss=0.4005, pruned_loss=0.1487, over 5664561.56 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4088, pruned_loss=0.1561, over 5630217.91 frames. ], batch size: 174, lr: 6.07e-03, grad_scale: 8.0
+2023-03-02 23:28:16,880 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:19,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:28,304 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:28:46,426 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:28:47,520 INFO [train.py:968] (1/2) Epoch 5, batch 41250, giga_loss[loss=0.3751, simple_loss=0.4199, pruned_loss=0.1651, over 28740.00 frames. ], tot_loss[loss=0.3671, simple_loss=0.4126, pruned_loss=0.1607, over 5611571.25 frames. ], libri_tot_loss[loss=0.35, simple_loss=0.4012, pruned_loss=0.1494, over 5656187.05 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.412, pruned_loss=0.16, over 5621013.23 frames. ], batch size: 284, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:29:21,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8106, 4.7381, 1.9441, 1.8483], device='cuda:1'), covar=tensor([0.0845, 0.0251, 0.0759, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0484, 0.0307, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0026, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 23:29:41,080 INFO [train.py:968] (1/2) Epoch 5, batch 41300, giga_loss[loss=0.4081, simple_loss=0.4534, pruned_loss=0.1814, over 28950.00 frames. ], tot_loss[loss=0.371, simple_loss=0.4159, pruned_loss=0.163, over 5619630.65 frames. ], libri_tot_loss[loss=0.3493, simple_loss=0.4007, pruned_loss=0.149, over 5659171.39 frames. ], giga_tot_loss[loss=0.371, simple_loss=0.416, pruned_loss=0.163, over 5623617.73 frames. ], batch size: 164, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:30:35,276 INFO [optim.py:369] (1/2) 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 23:30:36,841 INFO [train.py:968] (1/2) Epoch 5, batch 41350, giga_loss[loss=0.3658, simple_loss=0.4167, pruned_loss=0.1574, over 28844.00 frames. ], tot_loss[loss=0.3719, simple_loss=0.4164, pruned_loss=0.1636, over 5631104.73 frames. ], libri_tot_loss[loss=0.3492, simple_loss=0.4007, pruned_loss=0.1488, over 5663695.92 frames. ], giga_tot_loss[loss=0.3725, simple_loss=0.4168, pruned_loss=0.1641, over 5629620.70 frames. ], batch size: 284, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:31:13,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:31:20,385 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-02 23:31:20,621 INFO [train.py:968] (1/2) Epoch 5, batch 41400, giga_loss[loss=0.3439, simple_loss=0.3975, pruned_loss=0.1451, over 28841.00 frames. ], tot_loss[loss=0.3683, simple_loss=0.4129, pruned_loss=0.1619, over 5637846.86 frames. ], libri_tot_loss[loss=0.3484, simple_loss=0.4, pruned_loss=0.1484, over 5671773.16 frames. ], giga_tot_loss[loss=0.3703, simple_loss=0.4143, pruned_loss=0.1631, over 5627885.84 frames. ], batch size: 99, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:31:51,609 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:31:56,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:32:05,991 INFO [optim.py:369] (1/2) 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 23:32:06,550 INFO [train.py:968] (1/2) Epoch 5, batch 41450, giga_loss[loss=0.3195, simple_loss=0.3823, pruned_loss=0.1284, over 28925.00 frames. ], tot_loss[loss=0.3675, simple_loss=0.412, pruned_loss=0.1615, over 5620525.20 frames. ], libri_tot_loss[loss=0.3477, simple_loss=0.3993, pruned_loss=0.1481, over 5656737.48 frames. ], giga_tot_loss[loss=0.3705, simple_loss=0.4143, pruned_loss=0.1633, over 5623933.29 frames. ], batch size: 199, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:32:19,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:32:34,501 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:32:56,428 INFO [train.py:968] (1/2) Epoch 5, batch 41500, giga_loss[loss=0.3282, simple_loss=0.3972, pruned_loss=0.1296, over 28606.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.4113, pruned_loss=0.1596, over 5610169.56 frames. ], libri_tot_loss[loss=0.3468, simple_loss=0.3986, pruned_loss=0.1475, over 5652914.51 frames. ], giga_tot_loss[loss=0.369, simple_loss=0.4141, pruned_loss=0.1619, over 5614947.08 frames. ], batch size: 307, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:33:31,238 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:33:33,229 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:33:45,326 INFO [train.py:968] (1/2) Epoch 5, batch 41550, giga_loss[loss=0.3515, simple_loss=0.4027, pruned_loss=0.1501, over 28629.00 frames. ], tot_loss[loss=0.3689, simple_loss=0.4138, pruned_loss=0.162, over 5612784.12 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3978, pruned_loss=0.1469, over 5659019.99 frames. ], giga_tot_loss[loss=0.3732, simple_loss=0.4171, pruned_loss=0.1646, over 5610100.57 frames. ], batch size: 85, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:33:55,107 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:05,647 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:09,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3801, 2.0331, 1.3973, 0.6446], device='cuda:1'), covar=tensor([0.2294, 0.1349, 0.2210, 0.2574], device='cuda:1'), in_proj_covar=tensor([0.1417, 0.1335, 0.1391, 0.1171], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 23:34:17,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:22,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:34:41,417 INFO [train.py:968] (1/2) Epoch 5, batch 41600, giga_loss[loss=0.3353, simple_loss=0.3933, pruned_loss=0.1386, over 28596.00 frames. ], tot_loss[loss=0.366, simple_loss=0.4116, pruned_loss=0.1602, over 5593595.52 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3973, pruned_loss=0.1467, over 5651438.17 frames. ], giga_tot_loss[loss=0.3702, simple_loss=0.4148, pruned_loss=0.1628, over 5597056.42 frames. ], batch size: 307, lr: 6.07e-03, grad_scale: 8.0
+2023-03-02 23:34:48,784 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-02 23:34:53,041 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:35:30,860 INFO [optim.py:369] (1/2) 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 23:35:30,872 INFO [train.py:968] (1/2) Epoch 5, batch 41650, giga_loss[loss=0.3069, simple_loss=0.3825, pruned_loss=0.1156, over 28992.00 frames. ], tot_loss[loss=0.3604, simple_loss=0.409, pruned_loss=0.156, over 5613120.20 frames. ], libri_tot_loss[loss=0.3457, simple_loss=0.3976, pruned_loss=0.1469, over 5655901.39 frames. ], giga_tot_loss[loss=0.3638, simple_loss=0.4116, pruned_loss=0.158, over 5611263.27 frames. ], batch size: 155, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:35:47,639 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223566.0, num_to_drop=1, layers_to_drop={1}
+2023-03-02 23:36:18,132 INFO [train.py:968] (1/2) Epoch 5, batch 41700, giga_loss[loss=0.3411, simple_loss=0.3986, pruned_loss=0.1418, over 28798.00 frames. ], tot_loss[loss=0.3562, simple_loss=0.4065, pruned_loss=0.153, over 5628869.14 frames. ], libri_tot_loss[loss=0.346, simple_loss=0.3977, pruned_loss=0.1471, over 5660426.79 frames. ], giga_tot_loss[loss=0.3591, simple_loss=0.4087, pruned_loss=0.1547, over 5622213.06 frames. ], batch size: 284, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:36:32,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-02 23:36:44,978 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5767, 3.7231, 1.6321, 1.5096], device='cuda:1'), covar=tensor([0.0856, 0.0310, 0.0830, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0477, 0.0307, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 23:37:05,501 INFO [optim.py:369] (1/2) 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 23:37:05,519 INFO [train.py:968] (1/2) Epoch 5, batch 41750, giga_loss[loss=0.3604, simple_loss=0.4104, pruned_loss=0.1552, over 28876.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.4036, pruned_loss=0.1507, over 5633163.38 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3977, pruned_loss=0.1471, over 5667124.05 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4056, pruned_loss=0.1522, over 5620706.09 frames. ], batch size: 112, lr: 6.07e-03, grad_scale: 4.0
+2023-03-02 23:37:25,576 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7730, 1.8171, 1.6805, 1.6512], device='cuda:1'), covar=tensor([0.0901, 0.1307, 0.1355, 0.1207], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0737, 0.0630, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 23:37:37,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4910, 2.1120, 1.4996, 0.6136], device='cuda:1'), covar=tensor([0.2478, 0.1275, 0.2191, 0.2933], device='cuda:1'), in_proj_covar=tensor([0.1396, 0.1302, 0.1368, 0.1154], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-02 23:37:43,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2455, 4.0761, 3.8885, 1.6702], device='cuda:1'), covar=tensor([0.0462, 0.0513, 0.0730, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0816, 0.0806, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 23:37:54,218 INFO [train.py:968] (1/2) Epoch 5, batch 41800, giga_loss[loss=0.3944, simple_loss=0.4314, pruned_loss=0.1788, over 27940.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.401, pruned_loss=0.1483, over 5637877.76 frames. ], libri_tot_loss[loss=0.3453, simple_loss=0.3973, pruned_loss=0.1466, over 5672402.79 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4031, pruned_loss=0.15, over 5621957.88 frames. ], batch size: 412, lr: 6.07e-03, grad_scale: 2.0
+2023-03-02 23:38:01,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:38:31,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:38:44,543 INFO [train.py:968] (1/2) Epoch 5, batch 41850, giga_loss[loss=0.3326, simple_loss=0.3897, pruned_loss=0.1377, over 28718.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4, pruned_loss=0.1475, over 5637585.92 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3973, pruned_loss=0.1468, over 5663411.89 frames. ], giga_tot_loss[loss=0.3496, simple_loss=0.4017, pruned_loss=0.1488, over 5631941.75 frames. ], batch size: 242, lr: 6.06e-03, grad_scale: 2.0
+2023-03-02 23:38:45,200 INFO [optim.py:369] (1/2) 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 23:38:46,149 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:39:32,417 INFO [train.py:968] (1/2) Epoch 5, batch 41900, libri_loss[loss=0.3592, simple_loss=0.412, pruned_loss=0.1532, over 29527.00 frames. ], tot_loss[loss=0.3473, simple_loss=0.3999, pruned_loss=0.1474, over 5644975.47 frames. ], libri_tot_loss[loss=0.3454, simple_loss=0.3974, pruned_loss=0.1467, over 5665851.59 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4013, pruned_loss=0.1484, over 5637971.36 frames. ], batch size: 83, lr: 6.06e-03, grad_scale: 2.0
+2023-03-02 23:40:03,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:23,073 INFO [train.py:968] (1/2) Epoch 5, batch 41950, libri_loss[loss=0.3046, simple_loss=0.3593, pruned_loss=0.125, over 29501.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3971, pruned_loss=0.145, over 5641495.77 frames. ], libri_tot_loss[loss=0.3451, simple_loss=0.3971, pruned_loss=0.1465, over 5671625.71 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3985, pruned_loss=0.146, over 5629582.95 frames. ], batch size: 70, lr: 6.06e-03, grad_scale: 2.0
+2023-03-02 23:40:23,829 INFO [optim.py:369] (1/2) 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 23:40:24,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:28,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:54,124 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:57,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:40:57,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:12,257 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:15,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:16,172 INFO [train.py:968] (1/2) Epoch 5, batch 42000, giga_loss[loss=0.3636, simple_loss=0.4226, pruned_loss=0.1524, over 28490.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3968, pruned_loss=0.1421, over 5647086.76 frames. ], libri_tot_loss[loss=0.3455, simple_loss=0.3975, pruned_loss=0.1468, over 5672600.11 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3976, pruned_loss=0.1426, over 5636160.00 frames. ], batch size: 336, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:41:16,172 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-02 23:41:24,563 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-02 23:41:31,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0491, 1.0336, 0.9318, 1.2217], device='cuda:1'), covar=tensor([0.0813, 0.0417, 0.0327, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0121, 0.0126, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0039, 0.0065], device='cuda:1')
+2023-03-02 23:41:35,872 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:41:55,328 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:42:06,736 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 42050, giga_loss[loss=0.3794, simple_loss=0.4441, pruned_loss=0.1573, over 29060.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3997, pruned_loss=0.1426, over 5650088.32 frames. ], libri_tot_loss[loss=0.3458, simple_loss=0.3977, pruned_loss=0.147, over 5666447.61 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.4, pruned_loss=0.1427, over 5647632.75 frames. ], batch size: 136, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:42:15,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.31 vs. limit=2.0
+2023-03-02 23:42:16,012 INFO [optim.py:369] (1/2) 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 23:42:34,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=223971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:42:38,278 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:42:41,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:42:46,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-02 23:43:04,782 INFO [train.py:968] (1/2) Epoch 5, batch 42100, giga_loss[loss=0.3315, simple_loss=0.3843, pruned_loss=0.1394, over 28918.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.4006, pruned_loss=0.1439, over 5652099.86 frames. ], libri_tot_loss[loss=0.3459, simple_loss=0.3977, pruned_loss=0.1471, over 5660330.08 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.401, pruned_loss=0.1439, over 5654516.56 frames. ], batch size: 112, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:43:10,563 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 42150, giga_loss[loss=0.3815, simple_loss=0.4261, pruned_loss=0.1685, over 28749.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3998, pruned_loss=0.144, over 5644426.96 frames. ], libri_tot_loss[loss=0.3452, simple_loss=0.397, pruned_loss=0.1467, over 5658115.78 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.4008, pruned_loss=0.1442, over 5648360.72 frames. ], batch size: 262, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:43:49,419 INFO [optim.py:369] (1/2) 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 23:44:18,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224084.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 23:44:21,033 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 42200, giga_loss[loss=0.3275, simple_loss=0.3838, pruned_loss=0.1357, over 28645.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3978, pruned_loss=0.1435, over 5665112.78 frames. ], libri_tot_loss[loss=0.3448, simple_loss=0.3967, pruned_loss=0.1464, over 5662174.33 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3988, pruned_loss=0.1439, over 5664437.30 frames. ], batch size: 242, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:44:48,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224116.0, num_to_drop=1, layers_to_drop={0}
+2023-03-02 23:45:17,343 INFO [train.py:968] (1/2) Epoch 5, batch 42250, giga_loss[loss=0.29, simple_loss=0.366, pruned_loss=0.107, over 29022.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3975, pruned_loss=0.1451, over 5660082.40 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3967, pruned_loss=0.1464, over 5663141.58 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3985, pruned_loss=0.1454, over 5657744.53 frames. ], batch size: 155, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:45:17,988 INFO [optim.py:369] (1/2) 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 23:45:25,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6576, 1.6206, 1.7237, 1.4942], device='cuda:1'), covar=tensor([0.1982, 0.2832, 0.1473, 0.1504], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0741, 0.0781, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-02 23:45:52,487 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3322, 1.4258, 1.2418, 1.5112], device='cuda:1'), covar=tensor([0.2216, 0.2181, 0.2142, 0.2035], device='cuda:1'), in_proj_covar=tensor([0.1135, 0.0877, 0.1005, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-02 23:46:04,340 INFO [train.py:968] (1/2) Epoch 5, batch 42300, giga_loss[loss=0.3049, simple_loss=0.3885, pruned_loss=0.1106, over 28582.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3978, pruned_loss=0.145, over 5664132.87 frames. ], libri_tot_loss[loss=0.3447, simple_loss=0.3967, pruned_loss=0.1464, over 5670547.32 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3986, pruned_loss=0.1452, over 5655874.68 frames. ], batch size: 60, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:46:39,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7705, 1.5925, 1.2015, 1.3597], device='cuda:1'), covar=tensor([0.0579, 0.0586, 0.0916, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0453, 0.0505, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 23:46:51,401 INFO [train.py:968] (1/2) Epoch 5, batch 42350, giga_loss[loss=0.3172, simple_loss=0.3873, pruned_loss=0.1235, over 28653.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3973, pruned_loss=0.1431, over 5676499.67 frames. ], libri_tot_loss[loss=0.3446, simple_loss=0.3965, pruned_loss=0.1463, over 5674680.18 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3981, pruned_loss=0.1433, over 5666340.72 frames. ], batch size: 242, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:46:52,046 INFO [optim.py:369] (1/2) 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 23:47:24,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-02 23:47:40,950 INFO [train.py:968] (1/2) Epoch 5, batch 42400, giga_loss[loss=0.318, simple_loss=0.3861, pruned_loss=0.1249, over 28684.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3978, pruned_loss=0.1434, over 5674602.83 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.3965, pruned_loss=0.1463, over 5676671.90 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3985, pruned_loss=0.1435, over 5665063.63 frames. ], batch size: 307, lr: 6.06e-03, grad_scale: 8.0
+2023-03-02 23:48:09,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1022, 1.2570, 4.0315, 3.2082], device='cuda:1'), covar=tensor([0.1659, 0.2203, 0.0411, 0.0615], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0536, 0.0774, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-02 23:48:21,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:48:24,750 INFO [train.py:968] (1/2) Epoch 5, batch 42450, libri_loss[loss=0.35, simple_loss=0.4028, pruned_loss=0.1486, over 27591.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3976, pruned_loss=0.1433, over 5665800.61 frames. ], libri_tot_loss[loss=0.3445, simple_loss=0.3966, pruned_loss=0.1462, over 5670534.12 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3982, pruned_loss=0.1434, over 5663188.40 frames. ], batch size: 116, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:48:27,168 INFO [optim.py:369] (1/2) 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 23:49:07,687 INFO [train.py:968] (1/2) Epoch 5, batch 42500, giga_loss[loss=0.2996, simple_loss=0.3672, pruned_loss=0.116, over 28940.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3948, pruned_loss=0.1415, over 5676673.28 frames. ], libri_tot_loss[loss=0.3449, simple_loss=0.3968, pruned_loss=0.1465, over 5673694.90 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3951, pruned_loss=0.1412, over 5671788.64 frames. ], batch size: 164, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:49:58,013 INFO [train.py:968] (1/2) Epoch 5, batch 42550, giga_loss[loss=0.3507, simple_loss=0.3821, pruned_loss=0.1596, over 23730.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3935, pruned_loss=0.1415, over 5659684.62 frames. ], libri_tot_loss[loss=0.3443, simple_loss=0.3963, pruned_loss=0.1462, over 5665583.53 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3941, pruned_loss=0.1414, over 5662704.54 frames. ], batch size: 705, lr: 6.06e-03, grad_scale: 4.0
+2023-03-02 23:49:59,395 INFO [optim.py:369] (1/2) 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 23:50:35,522 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:50:39,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:50:46,204 INFO [train.py:968] (1/2) Epoch 5, batch 42600, giga_loss[loss=0.3009, simple_loss=0.3691, pruned_loss=0.1164, over 28879.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3929, pruned_loss=0.1418, over 5676910.06 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3958, pruned_loss=0.1457, over 5672913.24 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3938, pruned_loss=0.1421, over 5672753.08 frames. ], batch size: 112, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:51:07,975 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 42650, giga_loss[loss=0.3124, simple_loss=0.3711, pruned_loss=0.1268, over 28904.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3914, pruned_loss=0.1411, over 5672581.49 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3961, pruned_loss=0.146, over 5668063.84 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3918, pruned_loss=0.141, over 5673395.87 frames. ], batch size: 186, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:51:36,787 INFO [optim.py:369] (1/2) 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 23:52:19,665 INFO [train.py:968] (1/2) Epoch 5, batch 42700, giga_loss[loss=0.3723, simple_loss=0.4183, pruned_loss=0.1631, over 28886.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3913, pruned_loss=0.1417, over 5665065.07 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3961, pruned_loss=0.1459, over 5664873.48 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3914, pruned_loss=0.1415, over 5668474.78 frames. ], batch size: 199, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:52:27,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=224605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:53:12,859 INFO [train.py:968] (1/2) Epoch 5, batch 42750, giga_loss[loss=0.3358, simple_loss=0.3939, pruned_loss=0.1389, over 28796.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3916, pruned_loss=0.1424, over 5649673.73 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3964, pruned_loss=0.1459, over 5663717.85 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3914, pruned_loss=0.1421, over 5653089.45 frames. ], batch size: 284, lr: 6.05e-03, grad_scale: 4.0
+2023-03-02 23:53:14,397 INFO [optim.py:369] (1/2) 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 23:53:58,398 INFO [train.py:968] (1/2) Epoch 5, batch 42800, giga_loss[loss=0.3468, simple_loss=0.3988, pruned_loss=0.1474, over 28509.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3922, pruned_loss=0.142, over 5659642.74 frames. ], libri_tot_loss[loss=0.3441, simple_loss=0.3965, pruned_loss=0.1459, over 5669214.98 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3918, pruned_loss=0.1418, over 5657154.73 frames. ], batch size: 336, lr: 6.05e-03, grad_scale: 8.0
+2023-03-02 23:54:14,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4700, 1.9310, 1.4682, 1.5192], device='cuda:1'), covar=tensor([0.0763, 0.0280, 0.0316, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0122, 0.0125, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-02 23:54:34,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4924, 1.8403, 1.2531, 1.3525], device='cuda:1'), covar=tensor([0.1288, 0.0955, 0.0972, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.1445, 0.1257, 0.1241, 0.1341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 23:54:44,089 INFO [train.py:968] (1/2) Epoch 5, batch 42850, giga_loss[loss=0.3418, simple_loss=0.3761, pruned_loss=0.1538, over 23466.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.393, pruned_loss=0.1414, over 5667179.07 frames. ], libri_tot_loss[loss=0.344, simple_loss=0.3964, pruned_loss=0.1458, over 5672593.66 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3927, pruned_loss=0.1412, over 5661980.46 frames. ], batch size: 705, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:54:47,703 INFO [optim.py:369] (1/2) 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 23:55:29,078 INFO [train.py:968] (1/2) Epoch 5, batch 42900, giga_loss[loss=0.3743, simple_loss=0.4224, pruned_loss=0.1631, over 28580.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3931, pruned_loss=0.1409, over 5681726.54 frames. ], libri_tot_loss[loss=0.3439, simple_loss=0.3962, pruned_loss=0.1458, over 5680851.30 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3929, pruned_loss=0.1406, over 5669865.24 frames. ], batch size: 307, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:56:09,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5750, 1.9077, 1.4289, 1.1610], device='cuda:1'), covar=tensor([0.1289, 0.0782, 0.0763, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.1453, 0.1253, 0.1241, 0.1353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-02 23:56:26,182 INFO [train.py:968] (1/2) Epoch 5, batch 42950, giga_loss[loss=0.3652, simple_loss=0.4141, pruned_loss=0.1582, over 28505.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3941, pruned_loss=0.1419, over 5684294.57 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.396, pruned_loss=0.1456, over 5682023.04 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3942, pruned_loss=0.1418, over 5673981.96 frames. ], batch size: 336, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:56:29,868 INFO [optim.py:369] (1/2) 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 23:56:45,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0831, 1.2104, 1.2249, 1.2198], device='cuda:1'), covar=tensor([0.0735, 0.0774, 0.1217, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0735, 0.0635, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-02 23:57:13,816 INFO [train.py:968] (1/2) Epoch 5, batch 43000, giga_loss[loss=0.2975, simple_loss=0.365, pruned_loss=0.115, over 28670.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3952, pruned_loss=0.1431, over 5692234.13 frames. ], libri_tot_loss[loss=0.343, simple_loss=0.3954, pruned_loss=0.1453, over 5685995.56 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3957, pruned_loss=0.1433, over 5680619.19 frames. ], batch size: 242, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:57:16,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-02 23:58:06,395 INFO [train.py:968] (1/2) Epoch 5, batch 43050, giga_loss[loss=0.3449, simple_loss=0.3964, pruned_loss=0.1467, over 28948.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3968, pruned_loss=0.1458, over 5689511.64 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3956, pruned_loss=0.1452, over 5690650.89 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3971, pruned_loss=0.1459, over 5676086.64 frames. ], batch size: 145, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:58:10,168 INFO [optim.py:369] (1/2) 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 23:58:11,015 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=224955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:58:36,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-02 23:58:56,963 INFO [train.py:968] (1/2) Epoch 5, batch 43100, giga_loss[loss=0.4678, simple_loss=0.4828, pruned_loss=0.2264, over 24199.00 frames. ], tot_loss[loss=0.348, simple_loss=0.3989, pruned_loss=0.1486, over 5681195.03 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3957, pruned_loss=0.1454, over 5693932.88 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.3991, pruned_loss=0.1486, over 5667776.31 frames. ], batch size: 705, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:59:01,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6743, 3.7065, 1.6460, 1.6380], device='cuda:1'), covar=tensor([0.0735, 0.0316, 0.0828, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0489, 0.0310, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0026, 0.0016, 0.0021], device='cuda:1')
+2023-03-02 23:59:07,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8214, 3.6359, 3.4527, 2.0110], device='cuda:1'), covar=tensor([0.0475, 0.0602, 0.0812, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0817, 0.0812, 0.0609], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-02 23:59:41,066 INFO [train.py:968] (1/2) Epoch 5, batch 43150, giga_loss[loss=0.3061, simple_loss=0.3653, pruned_loss=0.1234, over 28925.00 frames. ], tot_loss[loss=0.349, simple_loss=0.3994, pruned_loss=0.1493, over 5674810.22 frames. ], libri_tot_loss[loss=0.3434, simple_loss=0.3959, pruned_loss=0.1454, over 5698035.90 frames. ], giga_tot_loss[loss=0.3492, simple_loss=0.3996, pruned_loss=0.1494, over 5659514.51 frames. ], batch size: 199, lr: 6.05e-03, grad_scale: 2.0
+2023-03-02 23:59:43,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3282, 1.7441, 1.5073, 1.4809], device='cuda:1'), covar=tensor([0.0768, 0.0290, 0.0301, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0122, 0.0125, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0044, 0.0039, 0.0065], device='cuda:1')
+2023-03-02 23:59:43,702 INFO [optim.py:369] (1/2) 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 23:59:53,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3354, 1.4703, 1.2393, 1.3457], device='cuda:1'), covar=tensor([0.2079, 0.2090, 0.2199, 0.1780], device='cuda:1'), in_proj_covar=tensor([0.1124, 0.0869, 0.0996, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 00:00:26,818 INFO [train.py:968] (1/2) Epoch 5, batch 43200, giga_loss[loss=0.3109, simple_loss=0.3662, pruned_loss=0.1278, over 28801.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3977, pruned_loss=0.148, over 5663518.70 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.3958, pruned_loss=0.1453, over 5690505.91 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3979, pruned_loss=0.1482, over 5658792.38 frames. ], batch size: 92, lr: 6.05e-03, grad_scale: 4.0
+2023-03-03 00:00:44,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:00:47,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:00:50,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:00:57,094 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:01:11,341 INFO [train.py:968] (1/2) Epoch 5, batch 43250, giga_loss[loss=0.3453, simple_loss=0.3906, pruned_loss=0.1501, over 27864.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3971, pruned_loss=0.1459, over 5667404.35 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3958, pruned_loss=0.1451, over 5693172.26 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3973, pruned_loss=0.1463, over 5660966.68 frames. ], batch size: 412, lr: 6.05e-03, grad_scale: 4.0
+2023-03-03 00:01:17,546 INFO [optim.py:369] (1/2) 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 00:01:18,394 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:01:34,776 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9430, 1.1380, 3.7303, 3.0574], device='cuda:1'), covar=tensor([0.1708, 0.2280, 0.0365, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0541, 0.0770, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 00:01:58,766 INFO [train.py:968] (1/2) Epoch 5, batch 43300, giga_loss[loss=0.3285, simple_loss=0.383, pruned_loss=0.137, over 28199.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3935, pruned_loss=0.1425, over 5657481.92 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3954, pruned_loss=0.1449, over 5688166.19 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.394, pruned_loss=0.143, over 5655076.88 frames. ], batch size: 368, lr: 6.05e-03, grad_scale: 4.0
+2023-03-03 00:02:43,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3434, 1.6083, 1.3573, 1.0885], device='cuda:1'), covar=tensor([0.1041, 0.0776, 0.0466, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.1462, 0.1266, 0.1246, 0.1357], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 00:02:43,903 INFO [train.py:968] (1/2) Epoch 5, batch 43350, libri_loss[loss=0.3392, simple_loss=0.3926, pruned_loss=0.1429, over 29547.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3919, pruned_loss=0.1417, over 5671405.33 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3956, pruned_loss=0.1451, over 5692704.66 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3921, pruned_loss=0.1418, over 5664430.46 frames. ], batch size: 78, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:02:46,600 INFO [optim.py:369] (1/2) 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 00:03:10,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4211, 2.8924, 1.5655, 1.3550], device='cuda:1'), covar=tensor([0.0736, 0.0323, 0.0701, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0482, 0.0309, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 00:03:11,189 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-03 00:03:29,802 INFO [train.py:968] (1/2) Epoch 5, batch 43400, giga_loss[loss=0.4097, simple_loss=0.4223, pruned_loss=0.1986, over 23585.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3909, pruned_loss=0.142, over 5667637.79 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.396, pruned_loss=0.1452, over 5695979.09 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3906, pruned_loss=0.1419, over 5658587.71 frames. ], batch size: 705, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:04:00,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:04:17,652 INFO [train.py:968] (1/2) Epoch 5, batch 43450, giga_loss[loss=0.3644, simple_loss=0.4146, pruned_loss=0.1571, over 28800.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3922, pruned_loss=0.1429, over 5666670.11 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3959, pruned_loss=0.1452, over 5687660.52 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.392, pruned_loss=0.1428, over 5665851.27 frames. ], batch size: 243, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:04:20,982 INFO [optim.py:369] (1/2) 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 00:05:03,046 INFO [train.py:968] (1/2) Epoch 5, batch 43500, giga_loss[loss=0.3201, simple_loss=0.4059, pruned_loss=0.1172, over 28991.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3957, pruned_loss=0.1444, over 5663382.49 frames. ], libri_tot_loss[loss=0.3432, simple_loss=0.3959, pruned_loss=0.1452, over 5689741.94 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3955, pruned_loss=0.1443, over 5660530.58 frames. ], batch size: 136, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:05:47,155 INFO [train.py:968] (1/2) Epoch 5, batch 43550, giga_loss[loss=0.3408, simple_loss=0.413, pruned_loss=0.1343, over 28737.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3966, pruned_loss=0.1419, over 5678322.93 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3948, pruned_loss=0.1445, over 5698542.87 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3974, pruned_loss=0.1425, over 5666853.91 frames. ], batch size: 262, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:05:50,589 INFO [optim.py:369] (1/2) 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 00:06:05,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4476, 3.1496, 1.4821, 1.4720], device='cuda:1'), covar=tensor([0.0839, 0.0308, 0.0829, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0481, 0.0309, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 00:06:14,608 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:17,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:33,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:34,412 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:37,430 INFO [train.py:968] (1/2) Epoch 5, batch 43600, giga_loss[loss=0.3609, simple_loss=0.4169, pruned_loss=0.1524, over 28318.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3972, pruned_loss=0.1421, over 5667660.75 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.394, pruned_loss=0.1439, over 5691005.57 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3988, pruned_loss=0.143, over 5663725.77 frames. ], batch size: 368, lr: 6.04e-03, grad_scale: 8.0
+2023-03-03 00:06:43,605 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:06:46,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:07:11,087 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:07:24,347 INFO [train.py:968] (1/2) Epoch 5, batch 43650, giga_loss[loss=0.3683, simple_loss=0.4102, pruned_loss=0.1632, over 28276.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3998, pruned_loss=0.1442, over 5663101.36 frames. ], libri_tot_loss[loss=0.3405, simple_loss=0.3936, pruned_loss=0.1437, over 5687351.62 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.4015, pruned_loss=0.1451, over 5663068.15 frames. ], batch size: 368, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:07:28,848 INFO [optim.py:369] (1/2) 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 00:07:38,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2874, 1.6334, 1.5870, 1.4637], device='cuda:1'), covar=tensor([0.1221, 0.1716, 0.0982, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0744, 0.0783, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 00:08:13,520 INFO [train.py:968] (1/2) Epoch 5, batch 43700, giga_loss[loss=0.3469, simple_loss=0.4094, pruned_loss=0.1422, over 28880.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.4012, pruned_loss=0.1461, over 5659012.60 frames. ], libri_tot_loss[loss=0.3403, simple_loss=0.3934, pruned_loss=0.1436, over 5691519.61 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4028, pruned_loss=0.1469, over 5654925.57 frames. ], batch size: 174, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:08:44,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:47,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:55,691 INFO [train.py:968] (1/2) Epoch 5, batch 43750, giga_loss[loss=0.3415, simple_loss=0.3961, pruned_loss=0.1434, over 29074.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.4004, pruned_loss=0.1469, over 5673049.70 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3929, pruned_loss=0.1433, over 5699023.69 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4025, pruned_loss=0.1479, over 5662115.70 frames. ], batch size: 155, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:08:56,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:58,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:08:58,857 INFO [optim.py:369] (1/2) 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 00:09:15,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:09:28,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:09:44,214 INFO [train.py:968] (1/2) Epoch 5, batch 43800, giga_loss[loss=0.3339, simple_loss=0.3951, pruned_loss=0.1363, over 28796.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3983, pruned_loss=0.1461, over 5671619.47 frames. ], libri_tot_loss[loss=0.3394, simple_loss=0.3926, pruned_loss=0.1431, over 5701223.85 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.4005, pruned_loss=0.1473, over 5660316.38 frames. ], batch size: 284, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:10:30,009 INFO [train.py:968] (1/2) Epoch 5, batch 43850, giga_loss[loss=0.3482, simple_loss=0.3988, pruned_loss=0.1488, over 28660.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3972, pruned_loss=0.1458, over 5669833.93 frames. ], libri_tot_loss[loss=0.3407, simple_loss=0.3936, pruned_loss=0.1439, over 5695781.00 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3981, pruned_loss=0.146, over 5665000.94 frames. ], batch size: 336, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:10:34,358 INFO [optim.py:369] (1/2) 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 00:10:45,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7329, 4.7717, 1.9617, 1.8800], device='cuda:1'), covar=tensor([0.0839, 0.0300, 0.0751, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0485, 0.0309, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 00:11:22,150 INFO [train.py:968] (1/2) Epoch 5, batch 43900, giga_loss[loss=0.3397, simple_loss=0.3916, pruned_loss=0.1439, over 28741.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3968, pruned_loss=0.1465, over 5670213.98 frames. ], libri_tot_loss[loss=0.341, simple_loss=0.3937, pruned_loss=0.1441, over 5697853.42 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3975, pruned_loss=0.1466, over 5664146.42 frames. ], batch size: 284, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:11:41,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2638, 1.3865, 1.2049, 1.4384], device='cuda:1'), covar=tensor([0.2386, 0.2309, 0.2381, 0.2102], device='cuda:1'), in_proj_covar=tensor([0.1127, 0.0872, 0.1000, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 00:11:54,220 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:10,440 INFO [train.py:968] (1/2) Epoch 5, batch 43950, giga_loss[loss=0.3137, simple_loss=0.3746, pruned_loss=0.1264, over 28911.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3975, pruned_loss=0.147, over 5672549.68 frames. ], libri_tot_loss[loss=0.341, simple_loss=0.3938, pruned_loss=0.1442, over 5701149.45 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.398, pruned_loss=0.147, over 5664505.44 frames. ], batch size: 213, lr: 6.04e-03, grad_scale: 4.0
+2023-03-03 00:12:16,523 INFO [optim.py:369] (1/2) 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 00:12:20,276 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:33,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:56,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:12:57,434 INFO [train.py:968] (1/2) Epoch 5, batch 44000, giga_loss[loss=0.3061, simple_loss=0.3715, pruned_loss=0.1204, over 28676.00 frames. ], tot_loss[loss=0.3458, simple_loss=0.3973, pruned_loss=0.1471, over 5677721.05 frames. ], libri_tot_loss[loss=0.3413, simple_loss=0.3941, pruned_loss=0.1442, over 5707034.95 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3977, pruned_loss=0.1473, over 5664492.65 frames. ], batch size: 242, lr: 6.04e-03, grad_scale: 8.0
+2023-03-03 00:13:06,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:13:09,606 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:13:42,308 INFO [train.py:968] (1/2) Epoch 5, batch 44050, giga_loss[loss=0.368, simple_loss=0.4145, pruned_loss=0.1607, over 28873.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3957, pruned_loss=0.1465, over 5681889.45 frames. ], libri_tot_loss[loss=0.3409, simple_loss=0.3938, pruned_loss=0.144, over 5710626.11 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3963, pruned_loss=0.1469, over 5667633.32 frames. ], batch size: 227, lr: 6.03e-03, grad_scale: 8.0
+2023-03-03 00:13:46,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=225952.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 00:13:47,809 INFO [optim.py:369] (1/2) 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 00:14:23,910 INFO [train.py:968] (1/2) Epoch 5, batch 44100, giga_loss[loss=0.3404, simple_loss=0.3966, pruned_loss=0.1421, over 28612.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3956, pruned_loss=0.1462, over 5671402.34 frames. ], libri_tot_loss[loss=0.3418, simple_loss=0.3945, pruned_loss=0.1446, over 5702555.11 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3956, pruned_loss=0.1461, over 5665914.67 frames. ], batch size: 336, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:14:36,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:14:39,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:14:48,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2493, 1.4329, 1.4798, 1.4807], device='cuda:1'), covar=tensor([0.0801, 0.0893, 0.1181, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0747, 0.0645, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 00:14:53,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9779, 4.6574, 1.8458, 1.8958], device='cuda:1'), covar=tensor([0.0837, 0.0258, 0.0828, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0487, 0.0312, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 00:15:15,465 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:15:19,494 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-03 00:15:20,277 INFO [train.py:968] (1/2) Epoch 5, batch 44150, libri_loss[loss=0.3486, simple_loss=0.4053, pruned_loss=0.146, over 27946.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3984, pruned_loss=0.1473, over 5661467.54 frames. ], libri_tot_loss[loss=0.3419, simple_loss=0.3946, pruned_loss=0.1446, over 5693727.37 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.3982, pruned_loss=0.1472, over 5664369.11 frames. ], batch size: 116, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:15:22,446 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:15:22,672 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-03 00:15:24,126 INFO [optim.py:369] (1/2) 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 00:15:24,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:15:52,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:16:08,705 INFO [train.py:968] (1/2) Epoch 5, batch 44200, giga_loss[loss=0.3054, simple_loss=0.3774, pruned_loss=0.1168, over 29012.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.3989, pruned_loss=0.1473, over 5671074.88 frames. ], libri_tot_loss[loss=0.3421, simple_loss=0.3949, pruned_loss=0.1447, over 5696668.11 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3986, pruned_loss=0.1472, over 5670184.94 frames. ], batch size: 155, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:16:58,335 INFO [train.py:968] (1/2) Epoch 5, batch 44250, giga_loss[loss=0.3109, simple_loss=0.378, pruned_loss=0.122, over 29030.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3989, pruned_loss=0.1478, over 5662370.10 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.395, pruned_loss=0.1448, over 5695960.42 frames. ], giga_tot_loss[loss=0.347, simple_loss=0.3986, pruned_loss=0.1477, over 5661708.35 frames. ], batch size: 128, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:17:03,846 INFO [optim.py:369] (1/2) 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 00:17:17,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5970, 1.4908, 1.2506, 1.3867], device='cuda:1'), covar=tensor([0.0659, 0.0572, 0.0959, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0455, 0.0505, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 00:17:46,150 INFO [train.py:968] (1/2) Epoch 5, batch 44300, libri_loss[loss=0.3334, simple_loss=0.3919, pruned_loss=0.1374, over 29657.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.4006, pruned_loss=0.1462, over 5663348.85 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3951, pruned_loss=0.145, over 5690647.85 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.4004, pruned_loss=0.1459, over 5666048.64 frames. ], batch size: 88, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:17:52,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:18:13,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:18:24,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3966, 1.6292, 1.2063, 1.5259], device='cuda:1'), covar=tensor([0.0724, 0.0316, 0.0347, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0120, 0.0124, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0038, 0.0065], device='cuda:1')
+2023-03-03 00:18:31,912 INFO [train.py:968] (1/2) Epoch 5, batch 44350, giga_loss[loss=0.4252, simple_loss=0.4316, pruned_loss=0.2094, over 23624.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.4017, pruned_loss=0.1447, over 5653496.74 frames. ], libri_tot_loss[loss=0.3431, simple_loss=0.3954, pruned_loss=0.1454, over 5672460.10 frames. ], giga_tot_loss[loss=0.3448, simple_loss=0.4013, pruned_loss=0.1442, over 5671812.61 frames. ], batch size: 705, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:18:33,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-03 00:18:36,482 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:18:37,566 INFO [optim.py:369] (1/2) 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 00:18:59,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:19:11,034 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:19:12,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:19:16,040 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 00:19:22,646 INFO [train.py:968] (1/2) Epoch 5, batch 44400, giga_loss[loss=0.3652, simple_loss=0.4183, pruned_loss=0.1561, over 29033.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.4043, pruned_loss=0.146, over 5674741.69 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3952, pruned_loss=0.1453, over 5678362.04 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.4044, pruned_loss=0.1456, over 5683713.45 frames. ], batch size: 164, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:19:49,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226327.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 00:20:08,773 INFO [train.py:968] (1/2) Epoch 5, batch 44450, giga_loss[loss=0.3122, simple_loss=0.3795, pruned_loss=0.1224, over 28877.00 frames. ], tot_loss[loss=0.3533, simple_loss=0.4069, pruned_loss=0.1498, over 5666916.94 frames. ], libri_tot_loss[loss=0.3427, simple_loss=0.3949, pruned_loss=0.1452, over 5680461.77 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4075, pruned_loss=0.1497, over 5671887.89 frames. ], batch size: 145, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:20:11,875 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:13,874 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:20:32,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:37,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:43,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 00:20:44,824 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:20:58,748 INFO [train.py:968] (1/2) Epoch 5, batch 44500, giga_loss[loss=0.3868, simple_loss=0.4045, pruned_loss=0.1846, over 23256.00 frames. ], tot_loss[loss=0.355, simple_loss=0.4072, pruned_loss=0.1514, over 5652485.00 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3946, pruned_loss=0.1449, over 5686028.91 frames. ], giga_tot_loss[loss=0.3558, simple_loss=0.4083, pruned_loss=0.1516, over 5650831.94 frames. ], batch size: 705, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:21:02,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:14,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:18,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:29,965 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:30,444 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:32,729 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:44,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:21:46,066 INFO [train.py:968] (1/2) Epoch 5, batch 44550, giga_loss[loss=0.3299, simple_loss=0.3951, pruned_loss=0.1324, over 28548.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4076, pruned_loss=0.1517, over 5658591.89 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3947, pruned_loss=0.145, over 5687355.76 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4084, pruned_loss=0.1519, over 5656111.10 frames. ], batch size: 307, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:21:51,680 INFO [optim.py:369] (1/2) 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 00:22:00,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:22:02,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1399, 1.4637, 1.2587, 1.3078], device='cuda:1'), covar=tensor([0.0774, 0.0370, 0.0309, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0121, 0.0125, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0043, 0.0039, 0.0066], device='cuda:1')
+2023-03-03 00:22:03,744 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226470.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 00:22:06,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226473.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 00:22:31,042 INFO [train.py:968] (1/2) Epoch 5, batch 44600, giga_loss[loss=0.3185, simple_loss=0.3859, pruned_loss=0.1256, over 28693.00 frames. ], tot_loss[loss=0.3518, simple_loss=0.4053, pruned_loss=0.1492, over 5663529.12 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3946, pruned_loss=0.145, over 5692076.91 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4063, pruned_loss=0.1496, over 5656745.95 frames. ], batch size: 284, lr: 6.03e-03, grad_scale: 4.0
+2023-03-03 00:22:32,471 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226502.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 00:22:43,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2758, 1.5598, 1.2530, 1.4451], device='cuda:1'), covar=tensor([0.0662, 0.0436, 0.0323, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0122, 0.0125, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0059, 0.0044, 0.0039, 0.0066], device='cuda:1')
+2023-03-03 00:23:16,730 INFO [train.py:968] (1/2) Epoch 5, batch 44650, giga_loss[loss=0.3535, simple_loss=0.4169, pruned_loss=0.145, over 28938.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4056, pruned_loss=0.1476, over 5669404.90 frames. ], libri_tot_loss[loss=0.342, simple_loss=0.3943, pruned_loss=0.1448, over 5695775.89 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4069, pruned_loss=0.148, over 5660341.04 frames. ], batch size: 213, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:23:24,471 INFO [optim.py:369] (1/2) 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 00:23:34,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5050, 1.9171, 1.7968, 1.6216], device='cuda:1'), covar=tensor([0.1671, 0.1911, 0.1250, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0747, 0.0788, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 00:23:59,564 INFO [train.py:968] (1/2) Epoch 5, batch 44700, giga_loss[loss=0.393, simple_loss=0.436, pruned_loss=0.175, over 28592.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4068, pruned_loss=0.1482, over 5664857.41 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3947, pruned_loss=0.1452, over 5688005.29 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4079, pruned_loss=0.1483, over 5663328.40 frames. ], batch size: 336, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:24:25,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=226626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:24:28,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:24:47,534 INFO [train.py:968] (1/2) Epoch 5, batch 44750, giga_loss[loss=0.3078, simple_loss=0.3705, pruned_loss=0.1225, over 28903.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4078, pruned_loss=0.1498, over 5662601.12 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3949, pruned_loss=0.1454, over 5685806.49 frames. ], giga_tot_loss[loss=0.3543, simple_loss=0.4088, pruned_loss=0.1499, over 5663573.54 frames. ], batch size: 227, lr: 6.03e-03, grad_scale: 2.0
+2023-03-03 00:24:57,260 INFO [optim.py:369] (1/2) 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 00:25:03,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:25:34,168 INFO [train.py:968] (1/2) Epoch 5, batch 44800, giga_loss[loss=0.3512, simple_loss=0.4091, pruned_loss=0.1467, over 28622.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4059, pruned_loss=0.1488, over 5676417.55 frames. ], libri_tot_loss[loss=0.3423, simple_loss=0.3946, pruned_loss=0.145, over 5688978.26 frames. ], giga_tot_loss[loss=0.3528, simple_loss=0.4071, pruned_loss=0.1492, over 5674181.36 frames. ], batch size: 307, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:26:20,301 INFO [train.py:968] (1/2) Epoch 5, batch 44850, giga_loss[loss=0.3149, simple_loss=0.3751, pruned_loss=0.1273, over 29131.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4038, pruned_loss=0.1486, over 5654858.61 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3949, pruned_loss=0.145, over 5687162.60 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4048, pruned_loss=0.1491, over 5653428.60 frames. ], batch size: 128, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:26:28,134 INFO [optim.py:369] (1/2) 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 00:26:45,872 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:26:47,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:26:50,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 00:27:09,407 INFO [train.py:968] (1/2) Epoch 5, batch 44900, giga_loss[loss=0.3297, simple_loss=0.3637, pruned_loss=0.1479, over 23596.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4008, pruned_loss=0.1472, over 5651898.11 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.395, pruned_loss=0.145, over 5685974.76 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.4017, pruned_loss=0.1477, over 5651332.88 frames. ], batch size: 705, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:27:14,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:15,222 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:18,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:18,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:27:47,128 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 44950, giga_loss[loss=0.2821, simple_loss=0.352, pruned_loss=0.1061, over 28519.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3973, pruned_loss=0.1448, over 5659917.36 frames. ], libri_tot_loss[loss=0.3424, simple_loss=0.3951, pruned_loss=0.1448, over 5690480.15 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.398, pruned_loss=0.1453, over 5654784.93 frames. ], batch size: 71, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:28:06,165 INFO [optim.py:369] (1/2) 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 00:28:40,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4920, 1.4906, 1.5201, 1.4321], device='cuda:1'), covar=tensor([0.0985, 0.1561, 0.1398, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0745, 0.0638, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 00:28:47,173 INFO [train.py:968] (1/2) Epoch 5, batch 45000, giga_loss[loss=0.3286, simple_loss=0.3858, pruned_loss=0.1357, over 28663.00 frames. ], tot_loss[loss=0.3436, simple_loss=0.3967, pruned_loss=0.1452, over 5656742.38 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3955, pruned_loss=0.145, over 5683867.93 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.397, pruned_loss=0.1455, over 5657873.09 frames. ], batch size: 242, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:28:47,173 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 00:28:55,866 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 00:29:09,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2021, 1.3497, 1.0802, 0.9052], device='cuda:1'), covar=tensor([0.0958, 0.0919, 0.0721, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.1475, 0.1288, 0.1266, 0.1361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 00:29:42,246 INFO [train.py:968] (1/2) Epoch 5, batch 45050, giga_loss[loss=0.3216, simple_loss=0.3842, pruned_loss=0.1295, over 27553.00 frames. ], tot_loss[loss=0.341, simple_loss=0.395, pruned_loss=0.1435, over 5663680.05 frames. ], libri_tot_loss[loss=0.3425, simple_loss=0.3953, pruned_loss=0.1449, over 5688017.98 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3954, pruned_loss=0.1438, over 5660677.67 frames. ], batch size: 472, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:29:44,042 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:29:46,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:29:47,498 INFO [optim.py:369] (1/2) 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 00:30:14,446 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:30:29,040 INFO [train.py:968] (1/2) Epoch 5, batch 45100, giga_loss[loss=0.3535, simple_loss=0.4018, pruned_loss=0.1526, over 27662.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.392, pruned_loss=0.1401, over 5650528.20 frames. ], libri_tot_loss[loss=0.3426, simple_loss=0.3954, pruned_loss=0.1449, over 5677606.39 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3922, pruned_loss=0.1403, over 5657123.99 frames. ], batch size: 472, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:30:29,974 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:30:48,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6753, 1.4984, 1.2163, 1.3034], device='cuda:1'), covar=tensor([0.0547, 0.0510, 0.0951, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0455, 0.0508, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 00:30:50,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4171, 3.2555, 1.5251, 1.4070], device='cuda:1'), covar=tensor([0.0842, 0.0346, 0.0826, 0.1279], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0488, 0.0310, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 00:31:15,398 INFO [train.py:968] (1/2) Epoch 5, batch 45150, giga_loss[loss=0.3313, simple_loss=0.3684, pruned_loss=0.1471, over 23690.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3913, pruned_loss=0.139, over 5648930.22 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3955, pruned_loss=0.1451, over 5669026.31 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3913, pruned_loss=0.1388, over 5661584.56 frames. ], batch size: 705, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:31:22,255 INFO [optim.py:369] (1/2) 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 00:31:34,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6830, 4.5245, 4.2909, 1.9802], device='cuda:1'), covar=tensor([0.0345, 0.0434, 0.0617, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0842, 0.0817, 0.0608], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 00:31:40,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-03 00:32:01,957 INFO [train.py:968] (1/2) Epoch 5, batch 45200, giga_loss[loss=0.3561, simple_loss=0.4035, pruned_loss=0.1543, over 27864.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3901, pruned_loss=0.1381, over 5642343.77 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3959, pruned_loss=0.1453, over 5670758.26 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3895, pruned_loss=0.1375, over 5650337.72 frames. ], batch size: 412, lr: 6.02e-03, grad_scale: 8.0
+2023-03-03 00:32:18,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4086, 1.9794, 1.4909, 0.7570], device='cuda:1'), covar=tensor([0.1907, 0.1068, 0.1630, 0.2309], device='cuda:1'), in_proj_covar=tensor([0.1421, 0.1343, 0.1383, 0.1189], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 00:32:44,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:32:49,747 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 5, batch 45250, giga_loss[loss=0.519, simple_loss=0.4879, pruned_loss=0.2751, over 26713.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.388, pruned_loss=0.1384, over 5629951.25 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3962, pruned_loss=0.1455, over 5676788.34 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3871, pruned_loss=0.1376, over 5630493.81 frames. ], batch size: 555, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:32:59,489 INFO [optim.py:369] (1/2) 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 00:33:15,340 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:33:20,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5370, 1.8512, 1.8215, 1.6681], device='cuda:1'), covar=tensor([0.1519, 0.1963, 0.1172, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0750, 0.0790, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 00:33:36,717 INFO [train.py:968] (1/2) Epoch 5, batch 45300, giga_loss[loss=0.376, simple_loss=0.4137, pruned_loss=0.1691, over 27571.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3879, pruned_loss=0.1388, over 5637292.00 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3956, pruned_loss=0.1451, over 5674038.87 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3874, pruned_loss=0.1382, over 5638347.61 frames. ], batch size: 472, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:33:59,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 00:34:19,292 INFO [train.py:968] (1/2) Epoch 5, batch 45350, giga_loss[loss=0.3315, simple_loss=0.3896, pruned_loss=0.1367, over 28623.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3899, pruned_loss=0.139, over 5644117.33 frames. ], libri_tot_loss[loss=0.3428, simple_loss=0.3956, pruned_loss=0.1449, over 5677167.44 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3893, pruned_loss=0.1386, over 5641327.60 frames. ], batch size: 307, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:34:27,994 INFO [optim.py:369] (1/2) 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 00:35:09,940 INFO [train.py:968] (1/2) Epoch 5, batch 45400, giga_loss[loss=0.3648, simple_loss=0.3928, pruned_loss=0.1684, over 23680.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3909, pruned_loss=0.1395, over 5635252.60 frames. ], libri_tot_loss[loss=0.3429, simple_loss=0.3959, pruned_loss=0.1449, over 5678240.84 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3901, pruned_loss=0.1391, over 5631555.68 frames. ], batch size: 705, lr: 6.02e-03, grad_scale: 4.0
+2023-03-03 00:35:57,070 INFO [train.py:968] (1/2) Epoch 5, batch 45450, giga_loss[loss=0.3329, simple_loss=0.3965, pruned_loss=0.1346, over 28961.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3897, pruned_loss=0.1388, over 5628082.73 frames. ], libri_tot_loss[loss=0.3433, simple_loss=0.3964, pruned_loss=0.1451, over 5682485.53 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3885, pruned_loss=0.1382, over 5620104.45 frames. ], batch size: 112, lr: 6.02e-03, grad_scale: 2.0
+2023-03-03 00:36:04,895 INFO [optim.py:369] (1/2) 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 00:36:26,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=227382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:36:36,449 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-03 00:36:43,973 INFO [train.py:968] (1/2) Epoch 5, batch 45500, giga_loss[loss=0.3483, simple_loss=0.4037, pruned_loss=0.1465, over 29014.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3894, pruned_loss=0.1386, over 5640229.13 frames. ], libri_tot_loss[loss=0.3435, simple_loss=0.3966, pruned_loss=0.1452, over 5684457.86 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3882, pruned_loss=0.1379, over 5631704.53 frames. ], batch size: 155, lr: 6.02e-03, grad_scale: 2.0
+2023-03-03 00:36:59,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7242, 1.6023, 1.6920, 1.4754], device='cuda:1'), covar=tensor([0.1067, 0.1843, 0.1565, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0747, 0.0645, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 00:37:32,162 INFO [train.py:968] (1/2) Epoch 5, batch 45550, giga_loss[loss=0.3532, simple_loss=0.4109, pruned_loss=0.1478, over 28908.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3926, pruned_loss=0.1414, over 5639991.81 frames. ], libri_tot_loss[loss=0.3436, simple_loss=0.3966, pruned_loss=0.1453, over 5677880.40 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3915, pruned_loss=0.1408, over 5638938.18 frames. ], batch size: 164, lr: 6.02e-03, grad_scale: 2.0
+2023-03-03 00:37:41,445 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 5, batch 45600, giga_loss[loss=0.3299, simple_loss=0.3799, pruned_loss=0.1399, over 28584.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.394, pruned_loss=0.1419, over 5638330.13 frames. ], libri_tot_loss[loss=0.3437, simple_loss=0.3968, pruned_loss=0.1453, over 5668005.08 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.393, pruned_loss=0.1413, over 5646756.02 frames. ], batch size: 85, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:39:09,197 INFO [train.py:968] (1/2) Epoch 5, batch 45650, libri_loss[loss=0.4268, simple_loss=0.4571, pruned_loss=0.1982, over 19208.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3976, pruned_loss=0.145, over 5616037.47 frames. ], libri_tot_loss[loss=0.3443, simple_loss=0.3972, pruned_loss=0.1457, over 5641733.35 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3964, pruned_loss=0.1442, over 5647263.79 frames. ], batch size: 188, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:39:19,310 INFO [optim.py:369] (1/2) 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 00:39:49,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1114, 1.3458, 1.1208, 0.8808], device='cuda:1'), covar=tensor([0.2076, 0.1989, 0.2016, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.1135, 0.0882, 0.1004, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 00:39:58,560 INFO [train.py:968] (1/2) Epoch 5, batch 45700, giga_loss[loss=0.3031, simple_loss=0.3778, pruned_loss=0.1142, over 28921.00 frames. ], tot_loss[loss=0.3464, simple_loss=0.3991, pruned_loss=0.1468, over 5594051.33 frames. ], libri_tot_loss[loss=0.345, simple_loss=0.3977, pruned_loss=0.1461, over 5606873.79 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3978, pruned_loss=0.1458, over 5648914.20 frames. ], batch size: 199, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:40:50,088 INFO [train.py:968] (1/2) Epoch 5, batch 45750, giga_loss[loss=0.3221, simple_loss=0.3933, pruned_loss=0.1255, over 28886.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3994, pruned_loss=0.1451, over 5561884.99 frames. ], libri_tot_loss[loss=0.3465, simple_loss=0.3986, pruned_loss=0.1472, over 5557068.02 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3975, pruned_loss=0.1434, over 5649334.37 frames. ], batch size: 174, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:41:01,007 INFO [optim.py:369] (1/2) 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 00:41:08,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3186, 1.6387, 1.6249, 1.4814], device='cuda:1'), covar=tensor([0.1560, 0.2234, 0.1282, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0746, 0.0789, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 00:41:39,115 INFO [train.py:968] (1/2) Epoch 5, batch 45800, giga_loss[loss=0.3517, simple_loss=0.403, pruned_loss=0.1502, over 28917.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3995, pruned_loss=0.1452, over 5544378.04 frames. ], libri_tot_loss[loss=0.3473, simple_loss=0.3991, pruned_loss=0.1477, over 5515793.29 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3976, pruned_loss=0.1433, over 5649630.48 frames. ], batch size: 213, lr: 6.01e-03, grad_scale: 4.0
+2023-03-03 00:41:53,203 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-03 00:43:14,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7466, 1.6435, 1.2483, 1.3765], device='cuda:1'), covar=tensor([0.0693, 0.0637, 0.0981, 0.1000], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0456, 0.0514, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 00:43:15,093 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:43:18,226 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 50, giga_loss[loss=0.3372, simple_loss=0.4117, pruned_loss=0.1314, over 28619.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3983, pruned_loss=0.1295, over 1265971.41 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3829, pruned_loss=0.1157, over 59280.57 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.399, pruned_loss=0.1301, over 1219103.32 frames. ], batch size: 262, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:44:00,854 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 100, giga_loss[loss=0.3023, simple_loss=0.3762, pruned_loss=0.1142, over 28864.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3879, pruned_loss=0.1255, over 2233083.36 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3711, pruned_loss=0.1156, over 221489.77 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3896, pruned_loss=0.1265, over 2094471.51 frames. ], batch size: 199, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:44:52,273 INFO [optim.py:369] (1/2) 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 00:44:52,881 INFO [train.py:968] (1/2) Epoch 6, batch 150, giga_loss[loss=0.2757, simple_loss=0.3333, pruned_loss=0.109, over 28870.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3691, pruned_loss=0.1155, over 2998008.62 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3735, pruned_loss=0.1157, over 274420.27 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3693, pruned_loss=0.1157, over 2863146.52 frames. ], batch size: 99, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:45:24,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:45:26,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:45:31,370 INFO [train.py:968] (1/2) Epoch 6, batch 200, giga_loss[loss=0.2358, simple_loss=0.3124, pruned_loss=0.07954, over 28591.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3557, pruned_loss=0.1085, over 3606136.57 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3649, pruned_loss=0.1105, over 575741.12 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3555, pruned_loss=0.1088, over 3366548.99 frames. ], batch size: 336, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:45:48,093 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:45:49,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 00:46:12,576 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 250, giga_loss[loss=0.2293, simple_loss=0.3022, pruned_loss=0.07818, over 28224.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3441, pruned_loss=0.1024, over 4067051.16 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3624, pruned_loss=0.1085, over 705427.28 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3433, pruned_loss=0.1024, over 3831947.28 frames. ], batch size: 77, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:47:00,223 INFO [train.py:968] (1/2) Epoch 6, batch 300, giga_loss[loss=0.2498, simple_loss=0.3209, pruned_loss=0.08929, over 29022.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3338, pruned_loss=0.09788, over 4423797.21 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3637, pruned_loss=0.1094, over 731168.15 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3324, pruned_loss=0.0975, over 4232732.86 frames. ], batch size: 164, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:47:42,601 INFO [optim.py:369] (1/2) 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 00:47:43,232 INFO [train.py:968] (1/2) Epoch 6, batch 350, giga_loss[loss=0.2299, simple_loss=0.3075, pruned_loss=0.07611, over 28773.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3272, pruned_loss=0.09474, over 4704917.54 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3652, pruned_loss=0.1115, over 883346.94 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3246, pruned_loss=0.09361, over 4517143.78 frames. ], batch size: 174, lr: 5.61e-03, grad_scale: 4.0
+2023-03-03 00:48:03,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:48:23,086 INFO [train.py:968] (1/2) Epoch 6, batch 400, giga_loss[loss=0.2076, simple_loss=0.2878, pruned_loss=0.06365, over 28439.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3227, pruned_loss=0.09194, over 4931212.38 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3685, pruned_loss=0.1124, over 1007348.35 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3189, pruned_loss=0.09033, over 4757442.45 frames. ], batch size: 78, lr: 5.61e-03, grad_scale: 8.0
+2023-03-03 00:49:00,948 INFO [optim.py:369] (1/2) 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 00:49:01,635 INFO [train.py:968] (1/2) Epoch 6, batch 450, giga_loss[loss=0.3315, simple_loss=0.362, pruned_loss=0.1505, over 26649.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.32, pruned_loss=0.09074, over 5102930.03 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3666, pruned_loss=0.1113, over 1127329.07 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3162, pruned_loss=0.08914, over 4945606.73 frames. ], batch size: 555, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:49:20,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:49:49,773 INFO [train.py:968] (1/2) Epoch 6, batch 500, giga_loss[loss=0.214, simple_loss=0.2801, pruned_loss=0.07392, over 28712.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3161, pruned_loss=0.08899, over 5224403.86 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3655, pruned_loss=0.1107, over 1151370.42 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.313, pruned_loss=0.08774, over 5097957.27 frames. ], batch size: 92, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:49:57,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:50:11,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-03 00:50:16,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5134, 2.2709, 1.6927, 0.8142], device='cuda:1'), covar=tensor([0.3261, 0.1531, 0.2382, 0.3198], device='cuda:1'), in_proj_covar=tensor([0.1411, 0.1335, 0.1374, 0.1179], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 00:50:26,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7293, 2.7809, 1.9177, 0.7650], device='cuda:1'), covar=tensor([0.4293, 0.1953, 0.2307, 0.4019], device='cuda:1'), in_proj_covar=tensor([0.1408, 0.1331, 0.1370, 0.1174], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 00:50:31,512 INFO [optim.py:369] (1/2) 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 00:50:32,419 INFO [train.py:968] (1/2) Epoch 6, batch 550, giga_loss[loss=0.2282, simple_loss=0.3021, pruned_loss=0.07717, over 29009.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3138, pruned_loss=0.08754, over 5329926.18 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3672, pruned_loss=0.1114, over 1244305.08 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.31, pruned_loss=0.08597, over 5219585.59 frames. ], batch size: 136, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:51:19,354 INFO [train.py:968] (1/2) Epoch 6, batch 600, giga_loss[loss=0.2104, simple_loss=0.2841, pruned_loss=0.06836, over 29047.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3121, pruned_loss=0.08634, over 5417333.62 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3687, pruned_loss=0.112, over 1336378.44 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3077, pruned_loss=0.08446, over 5320566.52 frames. ], batch size: 136, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:51:29,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:51:31,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:51:59,176 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:52:04,670 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 650, giga_loss[loss=0.227, simple_loss=0.2957, pruned_loss=0.07919, over 28923.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3101, pruned_loss=0.08566, over 5476370.77 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3676, pruned_loss=0.1119, over 1440014.63 frames. ], giga_tot_loss[loss=0.2364, simple_loss=0.3056, pruned_loss=0.0836, over 5397063.35 frames. ], batch size: 106, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:52:40,976 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:52:48,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4528, 1.6168, 1.1347, 1.2269], device='cuda:1'), covar=tensor([0.0780, 0.0585, 0.1220, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0454, 0.0506, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 00:52:48,614 INFO [train.py:968] (1/2) Epoch 6, batch 700, giga_loss[loss=0.2395, simple_loss=0.3056, pruned_loss=0.08671, over 28485.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3076, pruned_loss=0.08419, over 5522554.07 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3672, pruned_loss=0.1112, over 1528877.67 frames. ], giga_tot_loss[loss=0.2337, simple_loss=0.303, pruned_loss=0.0822, over 5452007.70 frames. ], batch size: 71, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:53:35,411 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 750, giga_loss[loss=0.2444, simple_loss=0.3084, pruned_loss=0.09021, over 28629.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3057, pruned_loss=0.08328, over 5554213.47 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3671, pruned_loss=0.1107, over 1615237.44 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3009, pruned_loss=0.08131, over 5492167.44 frames. ], batch size: 262, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:53:35,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:53:45,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:54:17,311 INFO [train.py:968] (1/2) Epoch 6, batch 800, giga_loss[loss=0.3127, simple_loss=0.3687, pruned_loss=0.1283, over 28862.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3062, pruned_loss=0.08407, over 5586936.19 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3663, pruned_loss=0.1103, over 1699367.94 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.3015, pruned_loss=0.0821, over 5532667.60 frames. ], batch size: 199, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:55:08,446 INFO [optim.py:369] (1/2) 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 00:55:08,468 INFO [train.py:968] (1/2) Epoch 6, batch 850, giga_loss[loss=0.3326, simple_loss=0.3899, pruned_loss=0.1376, over 27910.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3186, pruned_loss=0.09113, over 5607298.84 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3672, pruned_loss=0.1112, over 1782764.47 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3137, pruned_loss=0.08895, over 5558758.74 frames. ], batch size: 412, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 00:55:39,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:55:47,190 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:55:49,143 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 900, libri_loss[loss=0.3271, simple_loss=0.3963, pruned_loss=0.1289, over 29560.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3337, pruned_loss=0.0993, over 5633180.74 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3683, pruned_loss=0.1122, over 1923510.94 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3282, pruned_loss=0.09679, over 5586198.27 frames. ], batch size: 89, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:55:57,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4449, 1.6370, 1.3616, 1.6871], device='cuda:1'), covar=tensor([0.2326, 0.2128, 0.2206, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.1139, 0.0886, 0.1015, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 00:56:12,485 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:56:33,797 INFO [train.py:968] (1/2) Epoch 6, batch 950, giga_loss[loss=0.289, simple_loss=0.3667, pruned_loss=0.1057, over 29000.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3451, pruned_loss=0.1056, over 5642889.80 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3682, pruned_loss=0.1121, over 2095039.64 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3394, pruned_loss=0.1032, over 5597709.05 frames. ], batch size: 227, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:56:34,566 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 1000, giga_loss[loss=0.2942, simple_loss=0.3643, pruned_loss=0.1121, over 28542.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3514, pruned_loss=0.1075, over 5651222.12 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3675, pruned_loss=0.1117, over 2151023.02 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3469, pruned_loss=0.1057, over 5612819.40 frames. ], batch size: 78, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:57:25,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:28,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:38,402 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:40,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:57:58,526 INFO [train.py:968] (1/2) Epoch 6, batch 1050, giga_loss[loss=0.2628, simple_loss=0.3372, pruned_loss=0.09419, over 28796.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3549, pruned_loss=0.1079, over 5662088.07 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3678, pruned_loss=0.1118, over 2169787.91 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3513, pruned_loss=0.1065, over 5630935.91 frames. ], batch size: 99, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:58:00,582 INFO [optim.py:369] (1/2) 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 00:58:07,940 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:58:13,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:58:35,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0683, 1.1693, 1.2664, 1.0648], device='cuda:1'), covar=tensor([0.1164, 0.1124, 0.1628, 0.1198], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0730, 0.0634, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 00:58:39,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0306, 1.2797, 4.0062, 3.1446], device='cuda:1'), covar=tensor([0.1618, 0.2284, 0.0360, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0525, 0.0745, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 00:58:40,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8692, 1.7482, 1.6932, 1.7303], device='cuda:1'), covar=tensor([0.1096, 0.1728, 0.1573, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0731, 0.0635, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 00:58:41,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-03 00:58:42,294 INFO [train.py:968] (1/2) Epoch 6, batch 1100, giga_loss[loss=0.307, simple_loss=0.377, pruned_loss=0.1185, over 28883.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3566, pruned_loss=0.1081, over 5659247.25 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3688, pruned_loss=0.1125, over 2236575.56 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3531, pruned_loss=0.1066, over 5636324.37 frames. ], batch size: 199, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:58:53,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=228822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 00:59:02,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5753, 1.8162, 1.9097, 1.6941], device='cuda:1'), covar=tensor([0.1326, 0.1585, 0.0993, 0.1116], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0733, 0.0796, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 00:59:14,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 00:59:25,760 INFO [train.py:968] (1/2) Epoch 6, batch 1150, giga_loss[loss=0.3061, simple_loss=0.3761, pruned_loss=0.118, over 28759.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3592, pruned_loss=0.1103, over 5665378.56 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3687, pruned_loss=0.1128, over 2308131.01 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3563, pruned_loss=0.109, over 5643914.37 frames. ], batch size: 284, lr: 5.60e-03, grad_scale: 4.0
+2023-03-03 00:59:27,276 INFO [optim.py:369] (1/2) 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 00:59:32,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 01:00:13,580 INFO [train.py:968] (1/2) Epoch 6, batch 1200, giga_loss[loss=0.2745, simple_loss=0.3505, pruned_loss=0.0993, over 28987.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3625, pruned_loss=0.1126, over 5671105.75 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3693, pruned_loss=0.1129, over 2343642.78 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.36, pruned_loss=0.1115, over 5652471.50 frames. ], batch size: 213, lr: 5.60e-03, grad_scale: 8.0
+2023-03-03 01:00:21,874 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:00:23,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:00:43,199 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9553, 1.9391, 1.4009, 1.6577], device='cuda:1'), covar=tensor([0.0601, 0.0484, 0.0935, 0.0771], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0457, 0.0514, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:00:46,701 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 1250, giga_loss[loss=0.2989, simple_loss=0.3728, pruned_loss=0.1124, over 28504.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.366, pruned_loss=0.1145, over 5681000.96 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3695, pruned_loss=0.1129, over 2414103.31 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3638, pruned_loss=0.1137, over 5662292.48 frames. ], batch size: 336, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:00:55,680 INFO [optim.py:369] (1/2) 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 01:01:07,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-03 01:01:24,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:01:26,282 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 1300, giga_loss[loss=0.2919, simple_loss=0.366, pruned_loss=0.1089, over 28686.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3686, pruned_loss=0.1151, over 5682276.22 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.369, pruned_loss=0.1125, over 2464888.17 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3671, pruned_loss=0.1146, over 5665815.99 frames. ], batch size: 78, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:01:51,544 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 1350, giga_loss[loss=0.3098, simple_loss=0.3824, pruned_loss=0.1186, over 28441.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3712, pruned_loss=0.1163, over 5687124.55 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3685, pruned_loss=0.1118, over 2565792.54 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3703, pruned_loss=0.1164, over 5668915.30 frames. ], batch size: 60, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:02:24,775 INFO [optim.py:369] (1/2) 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 01:02:30,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 01:02:38,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2554, 1.4148, 1.1394, 0.8756], device='cuda:1'), covar=tensor([0.1168, 0.1115, 0.0778, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.1445, 0.1272, 0.1234, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 01:02:54,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:02:56,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:03:04,344 INFO [train.py:968] (1/2) Epoch 6, batch 1400, giga_loss[loss=0.2918, simple_loss=0.377, pruned_loss=0.1032, over 28607.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3714, pruned_loss=0.1151, over 5694088.60 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3694, pruned_loss=0.1124, over 2630675.45 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3704, pruned_loss=0.1149, over 5677041.66 frames. ], batch size: 60, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:03:45,962 INFO [train.py:968] (1/2) Epoch 6, batch 1450, giga_loss[loss=0.2914, simple_loss=0.3693, pruned_loss=0.1067, over 28840.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3691, pruned_loss=0.1125, over 5697914.70 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3679, pruned_loss=0.1115, over 2677875.33 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3689, pruned_loss=0.1128, over 5683088.28 frames. ], batch size: 186, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:03:47,274 INFO [optim.py:369] (1/2) 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 01:04:14,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229197.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:04:25,428 INFO [train.py:968] (1/2) Epoch 6, batch 1500, giga_loss[loss=0.2547, simple_loss=0.3425, pruned_loss=0.08344, over 28632.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3666, pruned_loss=0.1099, over 5712088.81 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3671, pruned_loss=0.111, over 2788922.19 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3668, pruned_loss=0.1103, over 5693986.54 frames. ], batch size: 119, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:04:52,452 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:04:56,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:04:58,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:05:03,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8590, 2.4202, 1.6346, 1.4471], device='cuda:1'), covar=tensor([0.1601, 0.0938, 0.1122, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.1458, 0.1278, 0.1248, 0.1336], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 01:05:09,773 INFO [train.py:968] (1/2) Epoch 6, batch 1550, giga_loss[loss=0.2852, simple_loss=0.3578, pruned_loss=0.1063, over 29019.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3666, pruned_loss=0.1101, over 5705915.91 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3675, pruned_loss=0.1111, over 2819745.12 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3666, pruned_loss=0.1103, over 5690185.25 frames. ], batch size: 136, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:05:11,523 INFO [optim.py:369] (1/2) 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 01:05:19,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:05:19,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.00 vs. limit=2.0
+2023-03-03 01:05:23,572 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:05:26,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-03 01:05:28,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5897, 1.8006, 1.3646, 1.2022], device='cuda:1'), covar=tensor([0.1161, 0.0962, 0.0798, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.1458, 0.1282, 0.1249, 0.1340], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 01:05:51,942 INFO [train.py:968] (1/2) Epoch 6, batch 1600, giga_loss[loss=0.2889, simple_loss=0.3668, pruned_loss=0.1054, over 28573.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3683, pruned_loss=0.113, over 5700529.11 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3675, pruned_loss=0.1113, over 2902074.20 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3682, pruned_loss=0.1131, over 5690284.04 frames. ], batch size: 71, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:06:08,814 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:06:19,739 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229340.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:06:23,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229343.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 01:06:36,995 INFO [train.py:968] (1/2) Epoch 6, batch 1650, libri_loss[loss=0.2975, simple_loss=0.3781, pruned_loss=0.1084, over 29525.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3712, pruned_loss=0.1176, over 5709129.45 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.367, pruned_loss=0.111, over 2990918.61 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3715, pruned_loss=0.118, over 5695405.58 frames. ], batch size: 84, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:06:38,502 INFO [optim.py:369] (1/2) 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 01:06:45,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229372.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:07:19,029 INFO [train.py:968] (1/2) Epoch 6, batch 1700, libri_loss[loss=0.2382, simple_loss=0.3127, pruned_loss=0.08183, over 28164.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3704, pruned_loss=0.1179, over 5713923.30 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3658, pruned_loss=0.1101, over 3089989.51 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3715, pruned_loss=0.1189, over 5697943.18 frames. ], batch size: 62, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:07:22,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-03 01:08:01,862 INFO [train.py:968] (1/2) Epoch 6, batch 1750, giga_loss[loss=0.2942, simple_loss=0.3639, pruned_loss=0.1122, over 28573.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3689, pruned_loss=0.1177, over 5700251.16 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3651, pruned_loss=0.1095, over 3171778.49 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3702, pruned_loss=0.1191, over 5683666.27 frames. ], batch size: 336, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:08:03,807 INFO [optim.py:369] (1/2) 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 01:08:43,570 INFO [train.py:968] (1/2) Epoch 6, batch 1800, giga_loss[loss=0.2865, simple_loss=0.3629, pruned_loss=0.105, over 29088.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3676, pruned_loss=0.1177, over 5701910.64 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3635, pruned_loss=0.1085, over 3239088.58 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3696, pruned_loss=0.1195, over 5684118.60 frames. ], batch size: 155, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:09:26,529 INFO [train.py:968] (1/2) Epoch 6, batch 1850, giga_loss[loss=0.3044, simple_loss=0.3807, pruned_loss=0.1141, over 28932.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3666, pruned_loss=0.1165, over 5694674.69 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.363, pruned_loss=0.1082, over 3292123.25 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3685, pruned_loss=0.1182, over 5676882.26 frames. ], batch size: 213, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:09:28,566 INFO [optim.py:369] (1/2) 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 01:10:08,852 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-03 01:10:09,770 INFO [train.py:968] (1/2) Epoch 6, batch 1900, giga_loss[loss=0.2837, simple_loss=0.3516, pruned_loss=0.1079, over 28791.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3644, pruned_loss=0.1139, over 5703532.46 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3634, pruned_loss=0.108, over 3381881.30 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3657, pruned_loss=0.1157, over 5683028.98 frames. ], batch size: 112, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:10:34,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8121, 1.6629, 1.3384, 1.4056], device='cuda:1'), covar=tensor([0.0627, 0.0574, 0.0901, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0446, 0.0502, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:10:35,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8669, 1.7191, 1.3687, 1.4463], device='cuda:1'), covar=tensor([0.0633, 0.0617, 0.0922, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0446, 0.0502, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:10:44,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4262, 3.1847, 1.4859, 1.4565], device='cuda:1'), covar=tensor([0.0896, 0.0253, 0.0851, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0473, 0.0307, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 01:10:48,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3957, 1.4886, 1.5405, 1.4965], device='cuda:1'), covar=tensor([0.1250, 0.1411, 0.1669, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0741, 0.0641, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 01:10:53,398 INFO [train.py:968] (1/2) Epoch 6, batch 1950, giga_loss[loss=0.2649, simple_loss=0.3396, pruned_loss=0.09511, over 28651.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3588, pruned_loss=0.1104, over 5683495.92 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3631, pruned_loss=0.1078, over 3445046.60 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.36, pruned_loss=0.1121, over 5672213.33 frames. ], batch size: 262, lr: 5.59e-03, grad_scale: 4.0
+2023-03-03 01:10:56,486 INFO [optim.py:369] (1/2) 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 01:11:23,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:11:33,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:11:39,646 INFO [train.py:968] (1/2) Epoch 6, batch 2000, libri_loss[loss=0.2762, simple_loss=0.3431, pruned_loss=0.1046, over 29586.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3534, pruned_loss=0.1078, over 5680735.47 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3635, pruned_loss=0.1082, over 3541710.96 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3539, pruned_loss=0.1089, over 5663615.43 frames. ], batch size: 74, lr: 5.59e-03, grad_scale: 8.0
+2023-03-03 01:12:25,621 INFO [train.py:968] (1/2) Epoch 6, batch 2050, giga_loss[loss=0.274, simple_loss=0.3293, pruned_loss=0.1093, over 27522.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3483, pruned_loss=0.1051, over 5681412.72 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3636, pruned_loss=0.1083, over 3611333.28 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3482, pruned_loss=0.1058, over 5661995.36 frames. ], batch size: 472, lr: 5.58e-03, grad_scale: 8.0
+2023-03-03 01:12:28,496 INFO [optim.py:369] (1/2) 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 01:13:13,546 INFO [train.py:968] (1/2) Epoch 6, batch 2100, giga_loss[loss=0.274, simple_loss=0.3552, pruned_loss=0.09642, over 28870.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3481, pruned_loss=0.105, over 5673093.12 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.364, pruned_loss=0.1086, over 3666061.20 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3474, pruned_loss=0.1053, over 5654155.77 frames. ], batch size: 174, lr: 5.58e-03, grad_scale: 8.0
+2023-03-03 01:13:17,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3003, 3.1876, 1.4235, 1.4301], device='cuda:1'), covar=tensor([0.0899, 0.0261, 0.0813, 0.1297], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0466, 0.0302, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 01:13:35,123 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:13:43,179 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:13:45,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:13:54,643 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 01:13:54,898 INFO [train.py:968] (1/2) Epoch 6, batch 2150, giga_loss[loss=0.2877, simple_loss=0.3602, pruned_loss=0.1076, over 28356.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3491, pruned_loss=0.1049, over 5687897.41 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.364, pruned_loss=0.1085, over 3699487.78 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3483, pruned_loss=0.1052, over 5669939.57 frames. ], batch size: 368, lr: 5.58e-03, grad_scale: 8.0
+2023-03-03 01:13:57,206 INFO [optim.py:369] (1/2) 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 01:14:02,479 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:14:08,900 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 2200, libri_loss[loss=0.3179, simple_loss=0.3926, pruned_loss=0.1216, over 29533.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3483, pruned_loss=0.1047, over 5691963.67 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3653, pruned_loss=0.1092, over 3753848.99 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3465, pruned_loss=0.1044, over 5672823.73 frames. ], batch size: 80, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:15:14,878 INFO [train.py:968] (1/2) Epoch 6, batch 2250, giga_loss[loss=0.269, simple_loss=0.3433, pruned_loss=0.09737, over 28540.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3447, pruned_loss=0.1025, over 5700101.00 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3655, pruned_loss=0.1091, over 3804644.75 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3428, pruned_loss=0.1022, over 5682687.06 frames. ], batch size: 336, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:15:17,446 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=229970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:15:33,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-03 01:15:40,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-03 01:15:55,224 INFO [train.py:968] (1/2) Epoch 6, batch 2300, giga_loss[loss=0.2556, simple_loss=0.3256, pruned_loss=0.09279, over 28422.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3426, pruned_loss=0.101, over 5709322.58 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3666, pruned_loss=0.1095, over 3856341.55 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3398, pruned_loss=0.1004, over 5691056.52 frames. ], batch size: 71, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:16:27,910 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230052.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:16:31,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9501, 2.4280, 2.2014, 2.0822], device='cuda:1'), covar=tensor([0.0509, 0.0663, 0.0802, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0443, 0.0501, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:16:33,658 INFO [train.py:968] (1/2) Epoch 6, batch 2350, giga_loss[loss=0.2308, simple_loss=0.3098, pruned_loss=0.07589, over 28316.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3404, pruned_loss=0.09996, over 5704401.95 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3675, pruned_loss=0.11, over 3897525.68 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.337, pruned_loss=0.09889, over 5693653.95 frames. ], batch size: 65, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:16:36,890 INFO [optim.py:369] (1/2) 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 01:16:39,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:17:11,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-03 01:17:11,983 INFO [train.py:968] (1/2) Epoch 6, batch 2400, giga_loss[loss=0.2533, simple_loss=0.3231, pruned_loss=0.09174, over 28849.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3382, pruned_loss=0.09935, over 5699140.84 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3679, pruned_loss=0.11, over 3927623.71 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3348, pruned_loss=0.09833, over 5695114.90 frames. ], batch size: 284, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:17:31,050 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 01:17:49,955 INFO [train.py:968] (1/2) Epoch 6, batch 2450, giga_loss[loss=0.2461, simple_loss=0.3079, pruned_loss=0.09216, over 28632.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3358, pruned_loss=0.09783, over 5708911.81 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3678, pruned_loss=0.1095, over 3994590.06 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3322, pruned_loss=0.09697, over 5700553.42 frames. ], batch size: 92, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:17:52,894 INFO [optim.py:369] (1/2) 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 01:18:31,057 INFO [train.py:968] (1/2) Epoch 6, batch 2500, giga_loss[loss=0.2316, simple_loss=0.303, pruned_loss=0.08016, over 28425.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.333, pruned_loss=0.0966, over 5716624.59 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.368, pruned_loss=0.1096, over 4022533.03 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3296, pruned_loss=0.09569, over 5707883.12 frames. ], batch size: 71, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:18:31,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:18:31,328 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:18:34,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:18:59,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:19:00,548 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 2550, giga_loss[loss=0.2385, simple_loss=0.3176, pruned_loss=0.07973, over 28747.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3327, pruned_loss=0.09672, over 5724366.05 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3688, pruned_loss=0.1098, over 4050885.73 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3289, pruned_loss=0.09567, over 5714572.39 frames. ], batch size: 262, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:19:16,568 INFO [optim.py:369] (1/2) 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 01:19:34,575 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:19:50,878 INFO [train.py:968] (1/2) Epoch 6, batch 2600, giga_loss[loss=0.2366, simple_loss=0.3115, pruned_loss=0.08091, over 28940.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3319, pruned_loss=0.0961, over 5727325.84 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3694, pruned_loss=0.1098, over 4096430.29 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3276, pruned_loss=0.09493, over 5715576.18 frames. ], batch size: 186, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:20:17,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:24,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:26,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:29,299 INFO [train.py:968] (1/2) Epoch 6, batch 2650, giga_loss[loss=0.2939, simple_loss=0.3638, pruned_loss=0.112, over 27953.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3325, pruned_loss=0.09664, over 5720479.98 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3698, pruned_loss=0.1099, over 4130880.09 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3279, pruned_loss=0.09533, over 5715673.07 frames. ], batch size: 412, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:20:34,251 INFO [optim.py:369] (1/2) 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 01:20:44,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:50,950 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:52,322 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:20:56,016 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 2700, giga_loss[loss=0.2945, simple_loss=0.3729, pruned_loss=0.108, over 29023.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3371, pruned_loss=0.09983, over 5720242.47 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3702, pruned_loss=0.1102, over 4139744.69 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3332, pruned_loss=0.0986, over 5715705.41 frames. ], batch size: 164, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:21:20,962 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:21:28,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230427.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:21:55,827 INFO [train.py:968] (1/2) Epoch 6, batch 2750, giga_loss[loss=0.3069, simple_loss=0.3725, pruned_loss=0.1206, over 28761.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3433, pruned_loss=0.1037, over 5719221.70 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3695, pruned_loss=0.1098, over 4199530.90 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3396, pruned_loss=0.1026, over 5710558.83 frames. ], batch size: 99, lr: 5.58e-03, grad_scale: 2.0
+2023-03-03 01:22:00,311 INFO [optim.py:369] (1/2) 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 01:22:11,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2325, 1.7372, 1.3072, 0.4030], device='cuda:1'), covar=tensor([0.1560, 0.0978, 0.1524, 0.2291], device='cuda:1'), in_proj_covar=tensor([0.1391, 0.1321, 0.1376, 0.1160], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 01:22:16,789 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:22:20,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:22:33,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5044, 5.1195, 4.9991, 2.6416], device='cuda:1'), covar=tensor([0.0350, 0.0433, 0.0650, 0.1765], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0801, 0.0775, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:22:37,409 INFO [train.py:968] (1/2) Epoch 6, batch 2800, giga_loss[loss=0.3065, simple_loss=0.3772, pruned_loss=0.1179, over 28851.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3524, pruned_loss=0.11, over 5705016.35 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3698, pruned_loss=0.11, over 4238296.33 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3487, pruned_loss=0.1089, over 5703112.09 frames. ], batch size: 174, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:22:44,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:23:04,019 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:23:14,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5764, 3.2629, 1.6470, 1.4693], device='cuda:1'), covar=tensor([0.0824, 0.0272, 0.0795, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0473, 0.0306, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 01:23:19,369 INFO [train.py:968] (1/2) Epoch 6, batch 2850, giga_loss[loss=0.2884, simple_loss=0.3559, pruned_loss=0.1105, over 28764.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3574, pruned_loss=0.1124, over 5704065.06 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3689, pruned_loss=0.1095, over 4301286.66 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3545, pruned_loss=0.112, over 5697472.79 frames. ], batch size: 92, lr: 5.58e-03, grad_scale: 4.0
+2023-03-03 01:23:25,167 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230570.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:23:30,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230573.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:23:59,340 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230602.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 01:24:06,206 INFO [train.py:968] (1/2) Epoch 6, batch 2900, giga_loss[loss=0.3098, simple_loss=0.3817, pruned_loss=0.1189, over 28913.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3629, pruned_loss=0.1143, over 5707539.42 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3692, pruned_loss=0.1097, over 4359731.05 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1139, over 5699410.62 frames. ], batch size: 145, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:24:25,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:24:49,563 INFO [train.py:968] (1/2) Epoch 6, batch 2950, giga_loss[loss=0.3718, simple_loss=0.4199, pruned_loss=0.1619, over 28575.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3692, pruned_loss=0.1183, over 5706612.57 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.369, pruned_loss=0.1096, over 4390036.90 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.367, pruned_loss=0.1182, over 5697062.98 frames. ], batch size: 336, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:24:50,531 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:24:54,351 INFO [optim.py:369] (1/2) 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 01:25:34,484 INFO [train.py:968] (1/2) Epoch 6, batch 3000, giga_loss[loss=0.2577, simple_loss=0.339, pruned_loss=0.08819, over 28772.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3731, pruned_loss=0.121, over 5686152.73 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3689, pruned_loss=0.1095, over 4430111.48 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3714, pruned_loss=0.1213, over 5677287.38 frames. ], batch size: 99, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:25:34,484 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 01:25:41,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3087, 1.4012, 1.0930, 1.4387], device='cuda:1'), covar=tensor([0.0734, 0.0279, 0.0319, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0123, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0060, 0.0043, 0.0039, 0.0066], device='cuda:1')
+2023-03-03 01:25:42,838 INFO [train.py:1012] (1/2) Epoch 6, validation: loss=0.2441, simple_loss=0.3436, pruned_loss=0.07229, over 944034.00 frames. 
+2023-03-03 01:25:42,839 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 01:25:56,398 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3534, 1.4986, 1.4722, 1.3871], device='cuda:1'), covar=tensor([0.1796, 0.2582, 0.1470, 0.1470], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0732, 0.0793, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 01:26:19,823 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 3050, libri_loss[loss=0.2563, simple_loss=0.3343, pruned_loss=0.0891, over 29382.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3676, pruned_loss=0.1169, over 5690300.41 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3688, pruned_loss=0.1095, over 4454063.47 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3664, pruned_loss=0.1173, over 5684055.25 frames. ], batch size: 67, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:26:28,972 INFO [optim.py:369] (1/2) 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 01:27:03,157 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:05,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:07,678 INFO [train.py:968] (1/2) Epoch 6, batch 3100, giga_loss[loss=0.2924, simple_loss=0.3561, pruned_loss=0.1144, over 28580.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3648, pruned_loss=0.1142, over 5691253.70 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3689, pruned_loss=0.1097, over 4471688.04 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3637, pruned_loss=0.1145, over 5692239.67 frames. ], batch size: 85, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:27:14,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5889, 1.8857, 1.4751, 2.0096], device='cuda:1'), covar=tensor([0.2176, 0.1910, 0.2015, 0.2040], device='cuda:1'), in_proj_covar=tensor([0.1131, 0.0873, 0.1007, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 01:27:32,745 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:49,882 INFO [train.py:968] (1/2) Epoch 6, batch 3150, giga_loss[loss=0.2761, simple_loss=0.3489, pruned_loss=0.1016, over 28808.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3637, pruned_loss=0.1129, over 5699382.48 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3689, pruned_loss=0.1097, over 4499747.63 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3627, pruned_loss=0.1132, over 5696762.13 frames. ], batch size: 119, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:27:50,193 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:27:55,155 INFO [optim.py:369] (1/2) 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 01:28:21,941 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:24,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:31,357 INFO [train.py:968] (1/2) Epoch 6, batch 3200, giga_loss[loss=0.3409, simple_loss=0.3986, pruned_loss=0.1416, over 28847.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3664, pruned_loss=0.1143, over 5706124.26 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3689, pruned_loss=0.1097, over 4520882.58 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3656, pruned_loss=0.1146, over 5701118.47 frames. ], batch size: 119, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:28:37,887 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:37,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:42,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=230926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:28:46,543 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 3250, giga_loss[loss=0.2764, simple_loss=0.3362, pruned_loss=0.1083, over 23633.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3683, pruned_loss=0.1151, over 5708418.21 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3688, pruned_loss=0.1095, over 4572459.59 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3677, pruned_loss=0.1156, over 5699926.04 frames. ], batch size: 705, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:29:15,860 INFO [optim.py:369] (1/2) 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 01:29:18,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3374, 1.4418, 1.2165, 1.4604], device='cuda:1'), covar=tensor([0.2236, 0.2108, 0.2141, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.1131, 0.0878, 0.1009, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 01:29:51,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:29:52,788 INFO [train.py:968] (1/2) Epoch 6, batch 3300, giga_loss[loss=0.2911, simple_loss=0.3607, pruned_loss=0.1108, over 29058.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3706, pruned_loss=0.117, over 5699296.84 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3694, pruned_loss=0.1099, over 4595835.23 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3696, pruned_loss=0.1173, over 5697332.24 frames. ], batch size: 128, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:30:35,209 INFO [train.py:968] (1/2) Epoch 6, batch 3350, giga_loss[loss=0.283, simple_loss=0.3503, pruned_loss=0.1079, over 28546.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3718, pruned_loss=0.1184, over 5700216.68 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3704, pruned_loss=0.1107, over 4617836.35 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3702, pruned_loss=0.1182, over 5698561.16 frames. ], batch size: 85, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:30:35,535 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:30:37,911 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:30:44,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4352, 3.3204, 1.6109, 1.2962], device='cuda:1'), covar=tensor([0.0866, 0.0292, 0.0774, 0.1363], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0471, 0.0304, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 01:31:06,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5070, 1.5859, 1.4587, 1.5189], device='cuda:1'), covar=tensor([0.1159, 0.1578, 0.1664, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0743, 0.0645, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 01:31:07,253 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:31:16,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4868, 1.8003, 1.7513, 1.5753], device='cuda:1'), covar=tensor([0.1447, 0.1821, 0.1084, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0735, 0.0795, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 01:31:19,818 INFO [train.py:968] (1/2) Epoch 6, batch 3400, libri_loss[loss=0.3043, simple_loss=0.3806, pruned_loss=0.114, over 29490.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.372, pruned_loss=0.1187, over 5714161.45 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3705, pruned_loss=0.1108, over 4643296.63 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3707, pruned_loss=0.1186, over 5709229.59 frames. ], batch size: 85, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:31:25,487 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3448, 1.4422, 1.4132, 1.4079], device='cuda:1'), covar=tensor([0.0985, 0.1194, 0.1445, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0741, 0.0645, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 01:31:47,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-03 01:31:53,599 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:31:55,804 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 3450, giga_loss[loss=0.463, simple_loss=0.4818, pruned_loss=0.2221, over 27888.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.373, pruned_loss=0.1196, over 5710926.36 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3704, pruned_loss=0.1108, over 4652590.79 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3722, pruned_loss=0.1196, over 5713232.29 frames. ], batch size: 412, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:32:07,411 INFO [optim.py:369] (1/2) 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 01:32:20,351 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:32:22,626 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:32:42,627 INFO [train.py:968] (1/2) Epoch 6, batch 3500, libri_loss[loss=0.2831, simple_loss=0.3638, pruned_loss=0.1012, over 29557.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3728, pruned_loss=0.1184, over 5707808.61 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3705, pruned_loss=0.1108, over 4661839.04 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3721, pruned_loss=0.1185, over 5710434.63 frames. ], batch size: 83, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:32:58,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9734, 1.7365, 1.3271, 1.3896], device='cuda:1'), covar=tensor([0.0558, 0.0582, 0.0919, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0329, 0.0434, 0.0497, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:33:00,101 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 3550, giga_loss[loss=0.2733, simple_loss=0.3473, pruned_loss=0.0996, over 29010.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3729, pruned_loss=0.1173, over 5713375.40 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3703, pruned_loss=0.1105, over 4708791.05 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3726, pruned_loss=0.1178, over 5710220.25 frames. ], batch size: 106, lr: 5.57e-03, grad_scale: 4.0
+2023-03-03 01:33:27,779 INFO [optim.py:369] (1/2) 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 01:33:40,031 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:33:49,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:33:55,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:34:03,658 INFO [train.py:968] (1/2) Epoch 6, batch 3600, giga_loss[loss=0.3007, simple_loss=0.3709, pruned_loss=0.1152, over 29015.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3722, pruned_loss=0.1162, over 5719040.48 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3705, pruned_loss=0.1108, over 4744488.12 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3718, pruned_loss=0.1166, over 5711274.10 frames. ], batch size: 128, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:34:15,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3769, 1.5356, 1.3095, 1.3426], device='cuda:1'), covar=tensor([0.2046, 0.1868, 0.1888, 0.1809], device='cuda:1'), in_proj_covar=tensor([0.1134, 0.0881, 0.1011, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 01:34:40,736 INFO [train.py:968] (1/2) Epoch 6, batch 3650, giga_loss[loss=0.2718, simple_loss=0.347, pruned_loss=0.09825, over 28864.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.37, pruned_loss=0.1148, over 5723698.93 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3709, pruned_loss=0.1108, over 4767691.49 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3694, pruned_loss=0.1152, over 5718585.49 frames. ], batch size: 186, lr: 5.57e-03, grad_scale: 8.0
+2023-03-03 01:34:48,482 INFO [optim.py:369] (1/2) 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 01:34:57,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:34:59,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:23,002 INFO [train.py:968] (1/2) Epoch 6, batch 3700, giga_loss[loss=0.2698, simple_loss=0.3459, pruned_loss=0.09689, over 28966.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3679, pruned_loss=0.1139, over 5713434.32 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3708, pruned_loss=0.1109, over 4795952.43 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3675, pruned_loss=0.1143, over 5713831.81 frames. ], batch size: 145, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:35:23,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:42,508 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:44,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:48,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:35:51,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:36:02,093 INFO [train.py:968] (1/2) Epoch 6, batch 3750, giga_loss[loss=0.3922, simple_loss=0.4397, pruned_loss=0.1724, over 28943.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3669, pruned_loss=0.1135, over 5722344.14 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3706, pruned_loss=0.1109, over 4823491.34 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3666, pruned_loss=0.1139, over 5718547.66 frames. ], batch size: 145, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:36:08,390 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:36:13,885 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:36:42,507 INFO [train.py:968] (1/2) Epoch 6, batch 3800, giga_loss[loss=0.2877, simple_loss=0.3551, pruned_loss=0.1101, over 28725.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3668, pruned_loss=0.1131, over 5732666.65 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3706, pruned_loss=0.1108, over 4857911.69 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3665, pruned_loss=0.1136, over 5726624.14 frames. ], batch size: 99, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:37:23,493 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 3850, giga_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 28568.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3685, pruned_loss=0.1146, over 5722660.36 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3708, pruned_loss=0.1108, over 4861226.09 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3681, pruned_loss=0.115, over 5722883.46 frames. ], batch size: 71, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:37:28,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5643, 1.7415, 1.4031, 1.0826], device='cuda:1'), covar=tensor([0.1216, 0.0942, 0.0834, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1257, 0.1239, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 01:37:30,035 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 3900, giga_loss[loss=0.2753, simple_loss=0.3499, pruned_loss=0.1003, over 28658.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3677, pruned_loss=0.1134, over 5710415.51 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3709, pruned_loss=0.111, over 4878852.43 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3673, pruned_loss=0.1137, over 5716949.44 frames. ], batch size: 242, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:38:30,723 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=231644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:38:42,236 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 3950, giga_loss[loss=0.2579, simple_loss=0.3384, pruned_loss=0.08864, over 28434.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3656, pruned_loss=0.1117, over 5713070.18 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3705, pruned_loss=0.111, over 4900476.80 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3655, pruned_loss=0.112, over 5717523.24 frames. ], batch size: 60, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:38:50,389 INFO [optim.py:369] (1/2) 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 01:39:13,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9531, 4.6735, 4.5631, 2.0532], device='cuda:1'), covar=tensor([0.0337, 0.0452, 0.0583, 0.1983], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0804, 0.0771, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:39:18,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:39:20,508 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 4000, giga_loss[loss=0.3098, simple_loss=0.3742, pruned_loss=0.1227, over 28578.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3639, pruned_loss=0.1112, over 5715657.16 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3701, pruned_loss=0.1108, over 4915614.46 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3641, pruned_loss=0.1116, over 5716380.11 frames. ], batch size: 85, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:39:44,223 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:39:47,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3305, 1.4147, 1.4759, 1.3813], device='cuda:1'), covar=tensor([0.1094, 0.1390, 0.1424, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0728, 0.0631, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 01:39:50,959 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-03 01:40:03,152 INFO [train.py:968] (1/2) Epoch 6, batch 4050, libri_loss[loss=0.3128, simple_loss=0.3852, pruned_loss=0.1202, over 29519.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3607, pruned_loss=0.1094, over 5712941.43 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3697, pruned_loss=0.1106, over 4935138.50 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3609, pruned_loss=0.1098, over 5709975.97 frames. ], batch size: 84, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:40:11,251 INFO [optim.py:369] (1/2) 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 01:40:35,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:40:37,814 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 4100, giga_loss[loss=0.3125, simple_loss=0.3654, pruned_loss=0.1298, over 28876.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3576, pruned_loss=0.1081, over 5713098.78 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3698, pruned_loss=0.1108, over 4953843.29 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3576, pruned_loss=0.1083, over 5707763.90 frames. ], batch size: 186, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:40:48,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3370, 1.4966, 1.2972, 1.3064], device='cuda:1'), covar=tensor([0.2138, 0.2049, 0.2114, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.1133, 0.0879, 0.1007, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 01:41:01,033 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:41:21,345 INFO [train.py:968] (1/2) Epoch 6, batch 4150, giga_loss[loss=0.2879, simple_loss=0.3579, pruned_loss=0.1089, over 28875.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.358, pruned_loss=0.1091, over 5698532.71 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3699, pruned_loss=0.111, over 4961074.10 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3576, pruned_loss=0.109, over 5704825.75 frames. ], batch size: 186, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:41:29,538 INFO [optim.py:369] (1/2) 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 01:41:34,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 01:42:01,344 INFO [train.py:968] (1/2) Epoch 6, batch 4200, giga_loss[loss=0.306, simple_loss=0.3721, pruned_loss=0.1199, over 28420.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3582, pruned_loss=0.1103, over 5698051.32 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3701, pruned_loss=0.1112, over 4990137.72 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3574, pruned_loss=0.11, over 5699785.92 frames. ], batch size: 369, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:42:42,675 INFO [train.py:968] (1/2) Epoch 6, batch 4250, giga_loss[loss=0.2597, simple_loss=0.3262, pruned_loss=0.09664, over 28768.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3555, pruned_loss=0.1091, over 5694287.61 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3701, pruned_loss=0.1112, over 5001540.16 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3546, pruned_loss=0.1089, over 5698119.41 frames. ], batch size: 99, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:42:50,465 INFO [optim.py:369] (1/2) 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 01:43:22,632 INFO [train.py:968] (1/2) Epoch 6, batch 4300, giga_loss[loss=0.2604, simple_loss=0.3321, pruned_loss=0.09433, over 28949.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3529, pruned_loss=0.108, over 5704271.91 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.37, pruned_loss=0.111, over 5013527.31 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3521, pruned_loss=0.108, over 5705977.30 frames. ], batch size: 136, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:43:30,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:43:47,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:44:02,143 INFO [train.py:968] (1/2) Epoch 6, batch 4350, libri_loss[loss=0.3368, simple_loss=0.405, pruned_loss=0.1343, over 29523.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3508, pruned_loss=0.1073, over 5704332.21 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3704, pruned_loss=0.1113, over 5026070.92 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3495, pruned_loss=0.107, over 5703444.36 frames. ], batch size: 79, lr: 5.56e-03, grad_scale: 2.0
+2023-03-03 01:44:10,358 INFO [optim.py:369] (1/2) 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 01:44:10,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2263, 1.3259, 4.8305, 3.3322], device='cuda:1'), covar=tensor([0.1620, 0.2412, 0.0276, 0.0726], device='cuda:1'), in_proj_covar=tensor([0.0569, 0.0529, 0.0748, 0.0624], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 01:44:41,057 INFO [train.py:968] (1/2) Epoch 6, batch 4400, giga_loss[loss=0.2385, simple_loss=0.3139, pruned_loss=0.08156, over 28774.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3497, pruned_loss=0.1067, over 5711115.67 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3704, pruned_loss=0.1113, over 5054822.17 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.348, pruned_loss=0.1062, over 5705438.62 frames. ], batch size: 119, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:45:22,004 INFO [train.py:968] (1/2) Epoch 6, batch 4450, giga_loss[loss=0.3012, simple_loss=0.3729, pruned_loss=0.1148, over 28265.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3516, pruned_loss=0.1072, over 5716380.54 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3702, pruned_loss=0.1112, over 5080051.35 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3498, pruned_loss=0.1068, over 5706550.99 frames. ], batch size: 368, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:45:22,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:45:25,029 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:45:30,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-03 01:45:48,100 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:46:01,718 INFO [train.py:968] (1/2) Epoch 6, batch 4500, giga_loss[loss=0.2755, simple_loss=0.3521, pruned_loss=0.09947, over 28963.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3551, pruned_loss=0.1089, over 5703222.40 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3704, pruned_loss=0.1113, over 5101499.12 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3529, pruned_loss=0.1084, over 5698606.27 frames. ], batch size: 164, lr: 5.56e-03, grad_scale: 4.0
+2023-03-03 01:46:41,216 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-03 01:46:42,072 INFO [train.py:968] (1/2) Epoch 6, batch 4550, libri_loss[loss=0.2738, simple_loss=0.3472, pruned_loss=0.1002, over 29631.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3579, pruned_loss=0.1098, over 5707151.59 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3703, pruned_loss=0.1112, over 5119116.38 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3558, pruned_loss=0.1094, over 5701136.45 frames. ], batch size: 69, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:46:50,284 INFO [optim.py:369] (1/2) 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 01:47:01,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4673, 4.3582, 1.5443, 1.5655], device='cuda:1'), covar=tensor([0.0889, 0.0196, 0.0959, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0328, 0.0477, 0.0308, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 01:47:19,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-03 01:47:20,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1864, 2.5019, 1.1534, 1.3320], device='cuda:1'), covar=tensor([0.0855, 0.0303, 0.0902, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0327, 0.0476, 0.0307, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0025, 0.0016, 0.0021], device='cuda:1')
+2023-03-03 01:47:25,107 INFO [train.py:968] (1/2) Epoch 6, batch 4600, giga_loss[loss=0.2393, simple_loss=0.3279, pruned_loss=0.07536, over 28837.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3591, pruned_loss=0.1097, over 5699084.83 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3707, pruned_loss=0.1114, over 5142337.87 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3568, pruned_loss=0.1092, over 5691314.39 frames. ], batch size: 174, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:48:06,381 INFO [train.py:968] (1/2) Epoch 6, batch 4650, giga_loss[loss=0.3396, simple_loss=0.3914, pruned_loss=0.1439, over 26744.00 frames. ], tot_loss[loss=0.289, simple_loss=0.359, pruned_loss=0.1095, over 5700757.92 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3708, pruned_loss=0.1116, over 5165038.31 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3567, pruned_loss=0.1088, over 5688740.86 frames. ], batch size: 555, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:48:13,711 INFO [optim.py:369] (1/2) 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 01:48:42,641 INFO [train.py:968] (1/2) Epoch 6, batch 4700, giga_loss[loss=0.2922, simple_loss=0.3675, pruned_loss=0.1085, over 28766.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3587, pruned_loss=0.1094, over 5717396.09 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3705, pruned_loss=0.1116, over 5201455.06 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3566, pruned_loss=0.1088, over 5698560.11 frames. ], batch size: 242, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:48:46,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:48:54,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6818, 1.7727, 1.5447, 2.2840], device='cuda:1'), covar=tensor([0.2010, 0.1968, 0.2058, 0.1853], device='cuda:1'), in_proj_covar=tensor([0.1143, 0.0882, 0.1015, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 01:49:26,375 INFO [train.py:968] (1/2) Epoch 6, batch 4750, giga_loss[loss=0.2966, simple_loss=0.3703, pruned_loss=0.1115, over 28979.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3594, pruned_loss=0.1103, over 5698459.01 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.371, pruned_loss=0.112, over 5199323.19 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3572, pruned_loss=0.1095, over 5692065.58 frames. ], batch size: 174, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:49:34,185 INFO [optim.py:369] (1/2) 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 01:50:04,859 INFO [train.py:968] (1/2) Epoch 6, batch 4800, giga_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 27931.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3607, pruned_loss=0.1114, over 5705465.95 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.371, pruned_loss=0.1119, over 5222762.93 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3585, pruned_loss=0.1108, over 5694685.37 frames. ], batch size: 412, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:50:37,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1436, 3.9492, 3.7129, 2.0194], device='cuda:1'), covar=tensor([0.0490, 0.0537, 0.0743, 0.1833], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0790, 0.0768, 0.0591], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:50:42,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:50:44,919 INFO [train.py:968] (1/2) Epoch 6, batch 4850, giga_loss[loss=0.3198, simple_loss=0.3833, pruned_loss=0.1282, over 28689.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3636, pruned_loss=0.1132, over 5696562.07 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3714, pruned_loss=0.1123, over 5233057.18 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3613, pruned_loss=0.1124, over 5693107.15 frames. ], batch size: 307, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:50:45,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:50:51,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9005, 3.6953, 3.5590, 1.6961], device='cuda:1'), covar=tensor([0.0570, 0.0726, 0.0969, 0.1937], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0790, 0.0768, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:50:51,648 INFO [optim.py:369] (1/2) 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 01:51:05,983 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-03 01:51:08,224 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:51:08,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1766, 1.4369, 1.0735, 0.7365], device='cuda:1'), covar=tensor([0.1337, 0.0916, 0.0790, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.1441, 0.1270, 0.1251, 0.1348], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 01:51:11,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5200, 1.4685, 1.2191, 1.7442], device='cuda:1'), covar=tensor([0.1943, 0.2021, 0.2130, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.1135, 0.0872, 0.1011, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 01:51:23,705 INFO [train.py:968] (1/2) Epoch 6, batch 4900, giga_loss[loss=0.3029, simple_loss=0.3656, pruned_loss=0.1201, over 28561.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3657, pruned_loss=0.1138, over 5698897.78 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3717, pruned_loss=0.1125, over 5242629.32 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3634, pruned_loss=0.113, over 5700668.94 frames. ], batch size: 71, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:51:45,910 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:51:57,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4449, 1.6543, 1.7280, 1.5767], device='cuda:1'), covar=tensor([0.1229, 0.1520, 0.0994, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0724, 0.0784, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 01:52:00,464 INFO [train.py:968] (1/2) Epoch 6, batch 4950, giga_loss[loss=0.2866, simple_loss=0.3494, pruned_loss=0.1119, over 28795.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3666, pruned_loss=0.1141, over 5705383.84 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3709, pruned_loss=0.112, over 5266472.64 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3651, pruned_loss=0.1139, over 5701347.84 frames. ], batch size: 99, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:52:09,215 INFO [optim.py:369] (1/2) 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 01:52:11,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 01:52:18,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4508, 4.2285, 4.1260, 1.7970], device='cuda:1'), covar=tensor([0.0498, 0.0651, 0.0942, 0.2006], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0794, 0.0772, 0.0593], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:52:35,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8866, 1.1388, 3.4151, 2.8582], device='cuda:1'), covar=tensor([0.1622, 0.2236, 0.0444, 0.0707], device='cuda:1'), in_proj_covar=tensor([0.0579, 0.0532, 0.0762, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 01:52:40,068 INFO [train.py:968] (1/2) Epoch 6, batch 5000, giga_loss[loss=0.2772, simple_loss=0.3543, pruned_loss=0.1001, over 28921.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3673, pruned_loss=0.1141, over 5713495.16 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3716, pruned_loss=0.1126, over 5290214.11 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3654, pruned_loss=0.1135, over 5706347.27 frames. ], batch size: 145, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:53:20,988 INFO [train.py:968] (1/2) Epoch 6, batch 5050, giga_loss[loss=0.292, simple_loss=0.3682, pruned_loss=0.1079, over 28889.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3678, pruned_loss=0.1146, over 5709079.29 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3723, pruned_loss=0.1131, over 5288492.97 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3656, pruned_loss=0.1137, over 5713509.31 frames. ], batch size: 186, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:53:29,654 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:54:00,911 INFO [train.py:968] (1/2) Epoch 6, batch 5100, giga_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08915, over 28576.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3667, pruned_loss=0.1142, over 5716929.42 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3721, pruned_loss=0.1129, over 5304070.40 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.365, pruned_loss=0.1137, over 5715999.69 frames. ], batch size: 307, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:54:42,182 INFO [train.py:968] (1/2) Epoch 6, batch 5150, giga_loss[loss=0.2584, simple_loss=0.3387, pruned_loss=0.08906, over 28882.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3647, pruned_loss=0.1133, over 5716504.32 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3725, pruned_loss=0.1133, over 5310136.68 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.363, pruned_loss=0.1127, over 5714086.32 frames. ], batch size: 174, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:54:51,471 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:54:52,450 INFO [optim.py:369] (1/2) 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 01:55:23,146 INFO [train.py:968] (1/2) Epoch 6, batch 5200, giga_loss[loss=0.3152, simple_loss=0.3811, pruned_loss=0.1247, over 28943.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3613, pruned_loss=0.1116, over 5718755.77 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3729, pruned_loss=0.1135, over 5318584.47 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3594, pruned_loss=0.1109, over 5716775.17 frames. ], batch size: 213, lr: 5.55e-03, grad_scale: 8.0
+2023-03-03 01:55:23,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=232911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:55:35,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-03 01:55:46,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5596, 1.5313, 1.1251, 1.3426], device='cuda:1'), covar=tensor([0.0665, 0.0621, 0.1059, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0441, 0.0505, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:56:01,289 INFO [train.py:968] (1/2) Epoch 6, batch 5250, giga_loss[loss=0.2544, simple_loss=0.332, pruned_loss=0.08839, over 28801.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3595, pruned_loss=0.1103, over 5718629.76 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.373, pruned_loss=0.1135, over 5331289.44 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3576, pruned_loss=0.1096, over 5714675.86 frames. ], batch size: 119, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:56:09,941 INFO [optim.py:369] (1/2) 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 01:56:29,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4086, 1.8529, 1.7811, 1.5612], device='cuda:1'), covar=tensor([0.1516, 0.1808, 0.1225, 0.1294], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0725, 0.0785, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 01:56:40,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1083, 1.2259, 0.9918, 1.0279], device='cuda:1'), covar=tensor([0.0706, 0.0509, 0.1135, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0438, 0.0504, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 01:56:42,356 INFO [train.py:968] (1/2) Epoch 6, batch 5300, giga_loss[loss=0.26, simple_loss=0.3475, pruned_loss=0.08625, over 28805.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3605, pruned_loss=0.1092, over 5717318.61 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.373, pruned_loss=0.1135, over 5347806.99 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3586, pruned_loss=0.1086, over 5709788.79 frames. ], batch size: 243, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:56:46,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:56:53,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3473, 1.4066, 1.4049, 1.3038], device='cuda:1'), covar=tensor([0.1029, 0.1590, 0.1612, 0.1524], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0717, 0.0629, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 01:57:24,215 INFO [train.py:968] (1/2) Epoch 6, batch 5350, giga_loss[loss=0.3094, simple_loss=0.3797, pruned_loss=0.1195, over 28785.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3623, pruned_loss=0.1101, over 5712295.09 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3729, pruned_loss=0.1134, over 5358223.76 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3607, pruned_loss=0.1095, over 5703461.15 frames. ], batch size: 174, lr: 5.55e-03, grad_scale: 4.0
+2023-03-03 01:57:33,092 INFO [optim.py:369] (1/2) 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 01:58:02,841 INFO [train.py:968] (1/2) Epoch 6, batch 5400, giga_loss[loss=0.3406, simple_loss=0.3971, pruned_loss=0.1421, over 28060.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3631, pruned_loss=0.1117, over 5707190.58 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3734, pruned_loss=0.1139, over 5369405.74 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3611, pruned_loss=0.1108, over 5702356.83 frames. ], batch size: 412, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 01:58:31,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:58:38,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:58:40,438 INFO [train.py:968] (1/2) Epoch 6, batch 5450, giga_loss[loss=0.3075, simple_loss=0.3605, pruned_loss=0.1273, over 28814.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3619, pruned_loss=0.1119, over 5708581.21 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3731, pruned_loss=0.1137, over 5386303.26 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3601, pruned_loss=0.1113, over 5701451.31 frames. ], batch size: 119, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 01:58:41,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:58:49,398 INFO [optim.py:369] (1/2) 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 01:59:05,095 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:59:21,235 INFO [train.py:968] (1/2) Epoch 6, batch 5500, giga_loss[loss=0.2694, simple_loss=0.3401, pruned_loss=0.09937, over 28878.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3614, pruned_loss=0.113, over 5706765.27 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3735, pruned_loss=0.1138, over 5395550.64 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5699928.08 frames. ], batch size: 186, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 01:59:47,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 01:59:59,682 INFO [train.py:968] (1/2) Epoch 6, batch 5550, giga_loss[loss=0.3095, simple_loss=0.3657, pruned_loss=0.1266, over 28655.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1127, over 5700458.51 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3737, pruned_loss=0.1141, over 5396760.77 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3573, pruned_loss=0.1119, over 5700818.65 frames. ], batch size: 336, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:00:09,293 INFO [optim.py:369] (1/2) 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 02:00:20,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:00:25,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:00:28,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:00:40,830 INFO [train.py:968] (1/2) Epoch 6, batch 5600, giga_loss[loss=0.3087, simple_loss=0.3611, pruned_loss=0.1281, over 28851.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.358, pruned_loss=0.1124, over 5706381.46 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3739, pruned_loss=0.1142, over 5411501.42 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3556, pruned_loss=0.1115, over 5702815.70 frames. ], batch size: 112, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:00:53,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:20,631 INFO [train.py:968] (1/2) Epoch 6, batch 5650, giga_loss[loss=0.2809, simple_loss=0.3463, pruned_loss=0.1078, over 28917.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3576, pruned_loss=0.1122, over 5708411.39 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3748, pruned_loss=0.115, over 5421084.96 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3543, pruned_loss=0.1108, over 5707544.12 frames. ], batch size: 227, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:01:26,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-03 02:01:28,912 INFO [optim.py:369] (1/2) 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 02:01:35,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0383, 3.8910, 3.6321, 1.6403], device='cuda:1'), covar=tensor([0.0493, 0.0517, 0.0784, 0.2267], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0804, 0.0781, 0.0604], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 02:01:36,855 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:41,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:43,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:01:55,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5705, 2.2680, 2.2260, 2.0048], device='cuda:1'), covar=tensor([0.1137, 0.2100, 0.1565, 0.1779], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0738, 0.0648, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 02:01:57,623 INFO [train.py:968] (1/2) Epoch 6, batch 5700, giga_loss[loss=0.2358, simple_loss=0.3124, pruned_loss=0.07954, over 29110.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3526, pruned_loss=0.1095, over 5719847.09 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3748, pruned_loss=0.1151, over 5434372.21 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3493, pruned_loss=0.1081, over 5714919.15 frames. ], batch size: 155, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:01:57,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:04,276 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:12,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:15,104 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 5750, libri_loss[loss=0.3179, simple_loss=0.3845, pruned_loss=0.1257, over 29361.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3489, pruned_loss=0.1077, over 5724556.24 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3749, pruned_loss=0.1155, over 5449457.66 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3455, pruned_loss=0.1061, over 5715861.74 frames. ], batch size: 92, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:02:37,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:02:45,327 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 5800, giga_loss[loss=0.2902, simple_loss=0.3532, pruned_loss=0.1136, over 28830.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3486, pruned_loss=0.1069, over 5725093.78 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3746, pruned_loss=0.1151, over 5460611.23 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3454, pruned_loss=0.1057, over 5715098.79 frames. ], batch size: 99, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:03:56,008 INFO [train.py:968] (1/2) Epoch 6, batch 5850, giga_loss[loss=0.2669, simple_loss=0.3472, pruned_loss=0.09326, over 28952.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.351, pruned_loss=0.1074, over 5730310.05 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3745, pruned_loss=0.1152, over 5469664.78 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.348, pruned_loss=0.1062, over 5718980.42 frames. ], batch size: 145, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:04:03,671 INFO [optim.py:369] (1/2) 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 02:04:04,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3731, 1.5281, 1.3380, 1.4743], device='cuda:1'), covar=tensor([0.2201, 0.2104, 0.2249, 0.1842], device='cuda:1'), in_proj_covar=tensor([0.1141, 0.0871, 0.1007, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 02:04:36,917 INFO [train.py:968] (1/2) Epoch 6, batch 5900, giga_loss[loss=0.2932, simple_loss=0.3675, pruned_loss=0.1095, over 28636.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3545, pruned_loss=0.109, over 5719681.63 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3747, pruned_loss=0.1154, over 5468019.68 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3516, pruned_loss=0.1077, over 5717458.82 frames. ], batch size: 262, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:05:08,937 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=233652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:05:17,168 INFO [train.py:968] (1/2) Epoch 6, batch 5950, giga_loss[loss=0.3943, simple_loss=0.4194, pruned_loss=0.1846, over 23905.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3585, pruned_loss=0.1108, over 5712651.21 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3744, pruned_loss=0.1152, over 5474193.77 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3562, pruned_loss=0.1099, over 5708826.95 frames. ], batch size: 705, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:05:26,218 INFO [optim.py:369] (1/2) 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 02:05:57,739 INFO [train.py:968] (1/2) Epoch 6, batch 6000, giga_loss[loss=0.3117, simple_loss=0.384, pruned_loss=0.1197, over 28863.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3606, pruned_loss=0.1121, over 5717829.73 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3738, pruned_loss=0.1149, over 5484898.75 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.359, pruned_loss=0.1116, over 5710369.18 frames. ], batch size: 199, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:05:57,739 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 02:06:04,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6852, 1.0417, 3.0925, 2.7107], device='cuda:1'), covar=tensor([0.1549, 0.2067, 0.0427, 0.0630], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0533, 0.0761, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 02:06:06,026 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 02:06:14,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.28 vs. limit=5.0
+2023-03-03 02:06:24,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3227, 2.9208, 1.5321, 1.3304], device='cuda:1'), covar=tensor([0.0804, 0.0361, 0.0800, 0.1217], device='cuda:1'), in_proj_covar=tensor([0.0330, 0.0484, 0.0309, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0026, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 02:06:40,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2608, 1.7073, 1.2378, 0.5815], device='cuda:1'), covar=tensor([0.2051, 0.1170, 0.1260, 0.2608], device='cuda:1'), in_proj_covar=tensor([0.1399, 0.1321, 0.1371, 0.1170], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 02:06:50,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:06:51,720 INFO [train.py:968] (1/2) Epoch 6, batch 6050, giga_loss[loss=0.2873, simple_loss=0.3504, pruned_loss=0.1121, over 28830.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3636, pruned_loss=0.1145, over 5709288.41 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.374, pruned_loss=0.115, over 5483136.26 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3622, pruned_loss=0.114, over 5705932.11 frames. ], batch size: 119, lr: 5.54e-03, grad_scale: 8.0
+2023-03-03 02:06:59,752 INFO [optim.py:369] (1/2) 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 02:07:12,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:07:32,878 INFO [train.py:968] (1/2) Epoch 6, batch 6100, giga_loss[loss=0.3007, simple_loss=0.369, pruned_loss=0.1162, over 29069.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3711, pruned_loss=0.1205, over 5710348.40 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3745, pruned_loss=0.1153, over 5500429.92 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3693, pruned_loss=0.12, over 5700842.72 frames. ], batch size: 155, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:08:14,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3700, 1.4846, 1.4127, 1.4294], device='cuda:1'), covar=tensor([0.0965, 0.1211, 0.1373, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0733, 0.0643, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 02:08:20,235 INFO [train.py:968] (1/2) Epoch 6, batch 6150, giga_loss[loss=0.3943, simple_loss=0.4402, pruned_loss=0.1742, over 28646.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3782, pruned_loss=0.1265, over 5691869.13 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3747, pruned_loss=0.1154, over 5506268.94 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3767, pruned_loss=0.1262, over 5682415.57 frames. ], batch size: 307, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:08:31,053 INFO [optim.py:369] (1/2) 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 02:08:34,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4466, 1.9984, 1.4537, 1.2692], device='cuda:1'), covar=tensor([0.1370, 0.0894, 0.1068, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.1444, 0.1271, 0.1238, 0.1321], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 02:08:53,667 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:08:55,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:02,421 INFO [train.py:968] (1/2) Epoch 6, batch 6200, giga_loss[loss=0.3291, simple_loss=0.3933, pruned_loss=0.1324, over 28628.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3848, pruned_loss=0.1318, over 5689656.34 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3741, pruned_loss=0.1152, over 5520369.78 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3842, pruned_loss=0.1323, over 5675699.95 frames. ], batch size: 262, lr: 5.54e-03, grad_scale: 4.0
+2023-03-03 02:09:21,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:24,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:25,883 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:09:49,081 INFO [train.py:968] (1/2) Epoch 6, batch 6250, giga_loss[loss=0.3243, simple_loss=0.3786, pruned_loss=0.135, over 28924.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3899, pruned_loss=0.1369, over 5684167.05 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3743, pruned_loss=0.1153, over 5527800.96 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3895, pruned_loss=0.1375, over 5669286.60 frames. ], batch size: 106, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:09:49,305 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:10:00,278 INFO [optim.py:369] (1/2) 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 02:10:01,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2325, 2.4746, 1.2866, 1.2862], device='cuda:1'), covar=tensor([0.0855, 0.0390, 0.0787, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0331, 0.0482, 0.0308, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0026, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 02:10:27,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2372, 1.5248, 1.0935, 1.0316], device='cuda:1'), covar=tensor([0.1073, 0.0849, 0.0776, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1270, 0.1238, 0.1324], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 02:10:32,546 INFO [train.py:968] (1/2) Epoch 6, batch 6300, giga_loss[loss=0.3596, simple_loss=0.4084, pruned_loss=0.1554, over 28265.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3961, pruned_loss=0.1419, over 5692200.51 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3743, pruned_loss=0.1152, over 5536816.03 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3962, pruned_loss=0.143, over 5675965.25 frames. ], batch size: 368, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:10:38,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2567, 2.8065, 1.3836, 1.2185], device='cuda:1'), covar=tensor([0.0826, 0.0341, 0.0798, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0482, 0.0309, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0026, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 02:10:49,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:10:49,983 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-03 02:10:53,706 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:11:21,294 INFO [train.py:968] (1/2) Epoch 6, batch 6350, giga_loss[loss=0.3879, simple_loss=0.4265, pruned_loss=0.1747, over 28789.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3996, pruned_loss=0.1453, over 5667964.95 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3739, pruned_loss=0.1149, over 5543281.77 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.4005, pruned_loss=0.1471, over 5652370.54 frames. ], batch size: 186, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:11:31,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4450, 2.0175, 1.4980, 1.4109], device='cuda:1'), covar=tensor([0.0731, 0.0286, 0.0300, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0120, 0.0124, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0060, 0.0044, 0.0039, 0.0066], device='cuda:1')
+2023-03-03 02:11:36,845 INFO [optim.py:369] (1/2) 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 02:12:03,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:12:09,465 INFO [train.py:968] (1/2) Epoch 6, batch 6400, giga_loss[loss=0.3457, simple_loss=0.3999, pruned_loss=0.1457, over 28977.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4014, pruned_loss=0.1479, over 5663402.93 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.374, pruned_loss=0.1147, over 5551943.20 frames. ], giga_tot_loss[loss=0.3516, simple_loss=0.4028, pruned_loss=0.1502, over 5645744.76 frames. ], batch size: 164, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:12:54,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-03 02:13:03,031 INFO [train.py:968] (1/2) Epoch 6, batch 6450, giga_loss[loss=0.3273, simple_loss=0.3948, pruned_loss=0.1299, over 29015.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.4045, pruned_loss=0.1518, over 5646655.43 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3737, pruned_loss=0.1145, over 5557621.19 frames. ], giga_tot_loss[loss=0.3578, simple_loss=0.4064, pruned_loss=0.1546, over 5628989.87 frames. ], batch size: 164, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:13:13,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1294, 0.9136, 0.8295, 1.4149], device='cuda:1'), covar=tensor([0.0755, 0.0346, 0.0349, 0.0778], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0120, 0.0124, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0060, 0.0044, 0.0039, 0.0067], device='cuda:1')
+2023-03-03 02:13:16,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4833, 1.6607, 1.6777, 1.6278], device='cuda:1'), covar=tensor([0.1041, 0.1208, 0.1099, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0738, 0.0647, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 02:13:16,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:13:18,467 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:13:19,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-03 02:13:49,675 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:13:58,058 INFO [train.py:968] (1/2) Epoch 6, batch 6500, giga_loss[loss=0.4571, simple_loss=0.4772, pruned_loss=0.2185, over 28889.00 frames. ], tot_loss[loss=0.3638, simple_loss=0.4109, pruned_loss=0.1583, over 5619315.03 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3739, pruned_loss=0.1146, over 5561250.47 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.4127, pruned_loss=0.1611, over 5602936.32 frames. ], batch size: 136, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:14:23,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4102, 3.2052, 1.4478, 1.4397], device='cuda:1'), covar=tensor([0.0831, 0.0283, 0.0793, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0491, 0.0314, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 02:14:46,900 INFO [train.py:968] (1/2) Epoch 6, batch 6550, giga_loss[loss=0.3453, simple_loss=0.4043, pruned_loss=0.1431, over 28845.00 frames. ], tot_loss[loss=0.3656, simple_loss=0.4124, pruned_loss=0.1594, over 5629809.96 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.374, pruned_loss=0.1146, over 5567748.41 frames. ], giga_tot_loss[loss=0.3696, simple_loss=0.4145, pruned_loss=0.1624, over 5612394.52 frames. ], batch size: 119, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:14:59,104 INFO [optim.py:369] (1/2) 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 02:15:00,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5112, 2.2909, 1.6865, 0.7302], device='cuda:1'), covar=tensor([0.3029, 0.1685, 0.2436, 0.3052], device='cuda:1'), in_proj_covar=tensor([0.1414, 0.1339, 0.1388, 0.1181], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 02:15:29,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4135, 1.6800, 1.4178, 1.4565], device='cuda:1'), covar=tensor([0.0775, 0.0300, 0.0302, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0120, 0.0124, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0060, 0.0044, 0.0039, 0.0066], device='cuda:1')
+2023-03-03 02:15:32,928 INFO [train.py:968] (1/2) Epoch 6, batch 6600, giga_loss[loss=0.4141, simple_loss=0.4424, pruned_loss=0.1929, over 28337.00 frames. ], tot_loss[loss=0.3654, simple_loss=0.4114, pruned_loss=0.1597, over 5648742.51 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3742, pruned_loss=0.1149, over 5576370.41 frames. ], giga_tot_loss[loss=0.3697, simple_loss=0.4138, pruned_loss=0.1628, over 5628748.49 frames. ], batch size: 368, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:16:13,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1540, 1.5982, 1.5724, 1.3931], device='cuda:1'), covar=tensor([0.1376, 0.1984, 0.1106, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0733, 0.0784, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 02:16:23,839 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 02:16:23,949 INFO [train.py:968] (1/2) Epoch 6, batch 6650, giga_loss[loss=0.3158, simple_loss=0.3731, pruned_loss=0.1292, over 28489.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4091, pruned_loss=0.158, over 5630988.79 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3745, pruned_loss=0.1151, over 5564486.31 frames. ], giga_tot_loss[loss=0.3669, simple_loss=0.4114, pruned_loss=0.1612, over 5626843.47 frames. ], batch size: 78, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:16:28,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0791, 1.1801, 3.8653, 3.2323], device='cuda:1'), covar=tensor([0.2142, 0.2768, 0.0697, 0.1555], device='cuda:1'), in_proj_covar=tensor([0.0580, 0.0539, 0.0762, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 02:16:36,719 INFO [optim.py:369] (1/2) 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 02:17:03,363 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 6700, giga_loss[loss=0.3876, simple_loss=0.4265, pruned_loss=0.1743, over 27582.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4082, pruned_loss=0.1557, over 5638239.94 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3749, pruned_loss=0.1153, over 5575232.85 frames. ], giga_tot_loss[loss=0.3652, simple_loss=0.4111, pruned_loss=0.1597, over 5627252.60 frames. ], batch size: 472, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:17:15,414 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:17:54,067 INFO [train.py:968] (1/2) Epoch 6, batch 6750, giga_loss[loss=0.4081, simple_loss=0.4188, pruned_loss=0.1987, over 23299.00 frames. ], tot_loss[loss=0.3602, simple_loss=0.409, pruned_loss=0.1557, over 5632199.11 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3745, pruned_loss=0.1151, over 5572829.42 frames. ], giga_tot_loss[loss=0.3661, simple_loss=0.4123, pruned_loss=0.1599, over 5626726.91 frames. ], batch size: 705, lr: 5.53e-03, grad_scale: 2.0
+2023-03-03 02:18:09,034 INFO [optim.py:369] (1/2) 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 02:18:11,996 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 6800, giga_loss[loss=0.3167, simple_loss=0.3851, pruned_loss=0.1241, over 28891.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.408, pruned_loss=0.1547, over 5620329.44 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3742, pruned_loss=0.115, over 5581042.31 frames. ], giga_tot_loss[loss=0.3651, simple_loss=0.4118, pruned_loss=0.1592, over 5609565.72 frames. ], batch size: 186, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:18:58,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5947, 2.4086, 1.7074, 2.3443], device='cuda:1'), covar=tensor([0.0575, 0.0574, 0.0917, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0453, 0.0507, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 02:19:19,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:19:23,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:19:31,883 INFO [train.py:968] (1/2) Epoch 6, batch 6850, giga_loss[loss=0.2587, simple_loss=0.3462, pruned_loss=0.08564, over 28417.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4035, pruned_loss=0.1504, over 5617502.90 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3739, pruned_loss=0.1149, over 5583529.85 frames. ], giga_tot_loss[loss=0.3595, simple_loss=0.408, pruned_loss=0.1555, over 5607531.95 frames. ], batch size: 65, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:19:40,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4533, 1.7673, 1.7767, 1.6261], device='cuda:1'), covar=tensor([0.1385, 0.1779, 0.1108, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0729, 0.0783, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 02:19:47,253 INFO [optim.py:369] (1/2) 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 02:19:54,243 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:06,095 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:21,418 INFO [train.py:968] (1/2) Epoch 6, batch 6900, giga_loss[loss=0.3449, simple_loss=0.4026, pruned_loss=0.1436, over 28891.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.4022, pruned_loss=0.1476, over 5629951.14 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3743, pruned_loss=0.1151, over 5588315.50 frames. ], giga_tot_loss[loss=0.3549, simple_loss=0.4058, pruned_loss=0.152, over 5618232.96 frames. ], batch size: 284, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:20:32,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:35,650 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=234624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:35,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:20:47,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 02:21:02,966 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:21:08,332 INFO [train.py:968] (1/2) Epoch 6, batch 6950, giga_loss[loss=0.3177, simple_loss=0.3748, pruned_loss=0.1303, over 28778.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3996, pruned_loss=0.1452, over 5635193.57 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3745, pruned_loss=0.1151, over 5585961.27 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.403, pruned_loss=0.1495, over 5629159.02 frames. ], batch size: 99, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:21:21,823 INFO [optim.py:369] (1/2) 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 02:21:54,454 INFO [train.py:968] (1/2) Epoch 6, batch 7000, giga_loss[loss=0.3451, simple_loss=0.3991, pruned_loss=0.1456, over 28562.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3967, pruned_loss=0.1428, over 5639863.22 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3736, pruned_loss=0.1147, over 5589527.55 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.4007, pruned_loss=0.1472, over 5632661.87 frames. ], batch size: 336, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:22:31,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3241, 2.7968, 1.4308, 1.3648], device='cuda:1'), covar=tensor([0.0804, 0.0365, 0.0765, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0485, 0.0311, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 02:22:43,129 INFO [train.py:968] (1/2) Epoch 6, batch 7050, giga_loss[loss=0.3509, simple_loss=0.4131, pruned_loss=0.1444, over 28814.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3946, pruned_loss=0.1415, over 5651200.94 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3732, pruned_loss=0.1144, over 5596471.08 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3986, pruned_loss=0.1459, over 5640761.27 frames. ], batch size: 119, lr: 5.53e-03, grad_scale: 4.0
+2023-03-03 02:22:55,699 INFO [optim.py:369] (1/2) 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 02:23:10,967 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:23:29,918 INFO [train.py:968] (1/2) Epoch 6, batch 7100, libri_loss[loss=0.3273, simple_loss=0.3968, pruned_loss=0.129, over 27873.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3951, pruned_loss=0.1418, over 5647411.94 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3731, pruned_loss=0.1144, over 5589792.07 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3986, pruned_loss=0.1457, over 5645659.87 frames. ], batch size: 116, lr: 5.52e-03, grad_scale: 2.0
+2023-03-03 02:24:18,869 INFO [train.py:968] (1/2) Epoch 6, batch 7150, giga_loss[loss=0.3058, simple_loss=0.3748, pruned_loss=0.1184, over 28878.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3925, pruned_loss=0.1389, over 5658654.76 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3728, pruned_loss=0.1142, over 5601985.20 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3965, pruned_loss=0.1432, over 5648254.09 frames. ], batch size: 174, lr: 5.52e-03, grad_scale: 2.0
+2023-03-03 02:24:33,351 INFO [optim.py:369] (1/2) 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 02:24:50,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3605, 2.0153, 1.4687, 0.5066], device='cuda:1'), covar=tensor([0.2617, 0.1444, 0.2568, 0.3400], device='cuda:1'), in_proj_covar=tensor([0.1413, 0.1325, 0.1375, 0.1174], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 02:25:04,508 INFO [train.py:968] (1/2) Epoch 6, batch 7200, giga_loss[loss=0.3363, simple_loss=0.4116, pruned_loss=0.1305, over 28884.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3903, pruned_loss=0.1357, over 5657910.33 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3732, pruned_loss=0.1144, over 5593396.26 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3938, pruned_loss=0.1396, over 5660651.02 frames. ], batch size: 213, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:25:36,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:25:38,979 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:02,235 INFO [train.py:968] (1/2) Epoch 6, batch 7250, giga_loss[loss=0.3581, simple_loss=0.4178, pruned_loss=0.1491, over 28356.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.392, pruned_loss=0.1352, over 5644391.04 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3734, pruned_loss=0.1146, over 5587969.22 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.395, pruned_loss=0.1386, over 5653365.64 frames. ], batch size: 368, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:26:07,671 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:10,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:14,161 INFO [optim.py:369] (1/2) 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 02:26:29,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4514, 1.5636, 1.5812, 1.5285], device='cuda:1'), covar=tensor([0.1078, 0.1502, 0.1262, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0718, 0.0634, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 02:26:37,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:26:46,837 INFO [train.py:968] (1/2) Epoch 6, batch 7300, giga_loss[loss=0.3391, simple_loss=0.3972, pruned_loss=0.1405, over 28874.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3938, pruned_loss=0.1363, over 5644586.89 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3733, pruned_loss=0.1147, over 5592631.18 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3971, pruned_loss=0.1399, over 5650394.00 frames. ], batch size: 213, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:27:12,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2629, 2.4812, 1.2629, 1.3389], device='cuda:1'), covar=tensor([0.0874, 0.0366, 0.0805, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0485, 0.0314, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 02:27:31,876 INFO [train.py:968] (1/2) Epoch 6, batch 7350, giga_loss[loss=0.3324, simple_loss=0.392, pruned_loss=0.1364, over 28638.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3933, pruned_loss=0.1366, over 5667289.37 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3727, pruned_loss=0.1144, over 5602042.33 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3971, pruned_loss=0.1404, over 5665196.69 frames. ], batch size: 307, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:27:34,217 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:27:48,798 INFO [optim.py:369] (1/2) 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 02:28:11,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5837, 1.7856, 1.4337, 2.1925], device='cuda:1'), covar=tensor([0.2041, 0.1989, 0.2033, 0.1977], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.0883, 0.1012, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 02:28:19,223 INFO [train.py:968] (1/2) Epoch 6, batch 7400, giga_loss[loss=0.3011, simple_loss=0.369, pruned_loss=0.1166, over 28713.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3916, pruned_loss=0.136, over 5671064.88 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3723, pruned_loss=0.1141, over 5608000.68 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3953, pruned_loss=0.1396, over 5665109.38 frames. ], batch size: 262, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:28:20,999 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:21,119 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-03 02:28:24,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:27,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:50,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:51,895 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:28:51,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:29:06,584 INFO [train.py:968] (1/2) Epoch 6, batch 7450, giga_loss[loss=0.4395, simple_loss=0.4559, pruned_loss=0.2115, over 28353.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3905, pruned_loss=0.1365, over 5662432.98 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3724, pruned_loss=0.114, over 5613917.18 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.394, pruned_loss=0.1403, over 5653681.21 frames. ], batch size: 368, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:29:06,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-03 02:29:10,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-03 02:29:18,829 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:29:19,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:29:19,915 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 7500, giga_loss[loss=0.328, simple_loss=0.393, pruned_loss=0.1314, over 28914.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3891, pruned_loss=0.1356, over 5679316.57 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3724, pruned_loss=0.1139, over 5620749.86 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3923, pruned_loss=0.1392, over 5667402.64 frames. ], batch size: 145, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:30:37,802 INFO [train.py:968] (1/2) Epoch 6, batch 7550, giga_loss[loss=0.3331, simple_loss=0.3948, pruned_loss=0.1357, over 28709.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.389, pruned_loss=0.1342, over 5690502.93 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3724, pruned_loss=0.1137, over 5625690.80 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3921, pruned_loss=0.1379, over 5678155.74 frames. ], batch size: 242, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:30:50,817 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.95 vs. limit=5.0
+2023-03-03 02:30:52,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:31:00,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3387, 1.4859, 1.2502, 1.7101], device='cuda:1'), covar=tensor([0.2548, 0.2468, 0.2548, 0.2214], device='cuda:1'), in_proj_covar=tensor([0.1147, 0.0885, 0.1014, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 02:31:24,450 INFO [train.py:968] (1/2) Epoch 6, batch 7600, giga_loss[loss=0.3287, simple_loss=0.392, pruned_loss=0.1327, over 28685.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3892, pruned_loss=0.1334, over 5688416.04 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3723, pruned_loss=0.1137, over 5620565.74 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.392, pruned_loss=0.1368, over 5685091.37 frames. ], batch size: 92, lr: 5.52e-03, grad_scale: 8.0
+2023-03-03 02:31:58,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:32:00,574 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 7650, giga_loss[loss=0.3779, simple_loss=0.4161, pruned_loss=0.1699, over 27966.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3895, pruned_loss=0.134, over 5687683.55 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3724, pruned_loss=0.1139, over 5628135.88 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3924, pruned_loss=0.1373, over 5681292.76 frames. ], batch size: 412, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:32:21,655 INFO [optim.py:369] (1/2) 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 02:32:24,000 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-03 02:32:50,644 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-03 02:32:56,028 INFO [train.py:968] (1/2) Epoch 6, batch 7700, giga_loss[loss=0.3406, simple_loss=0.3946, pruned_loss=0.1433, over 28953.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3883, pruned_loss=0.134, over 5690348.37 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3723, pruned_loss=0.1139, over 5628560.20 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3906, pruned_loss=0.1367, over 5685255.52 frames. ], batch size: 136, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:33:20,979 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:33:36,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-03 02:33:46,408 INFO [train.py:968] (1/2) Epoch 6, batch 7750, giga_loss[loss=0.3672, simple_loss=0.4089, pruned_loss=0.1627, over 27964.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3874, pruned_loss=0.1343, over 5687897.80 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3721, pruned_loss=0.1139, over 5634077.60 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3897, pruned_loss=0.1368, over 5680179.86 frames. ], batch size: 412, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:33:52,350 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-03 02:34:01,007 INFO [optim.py:369] (1/2) 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 02:34:15,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:34:32,049 INFO [train.py:968] (1/2) Epoch 6, batch 7800, giga_loss[loss=0.2803, simple_loss=0.3511, pruned_loss=0.1048, over 28917.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3882, pruned_loss=0.1359, over 5682639.18 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3725, pruned_loss=0.1141, over 5631616.05 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3901, pruned_loss=0.1381, over 5679215.34 frames. ], batch size: 106, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:35:08,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:35:20,123 INFO [train.py:968] (1/2) Epoch 6, batch 7850, giga_loss[loss=0.3223, simple_loss=0.3818, pruned_loss=0.1314, over 28519.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3867, pruned_loss=0.1347, over 5694955.23 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3723, pruned_loss=0.1139, over 5635601.77 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3888, pruned_loss=0.1373, over 5690227.88 frames. ], batch size: 60, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:35:34,440 INFO [optim.py:369] (1/2) 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 02:35:37,616 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:35:40,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:04,705 INFO [train.py:968] (1/2) Epoch 6, batch 7900, libri_loss[loss=0.3776, simple_loss=0.4313, pruned_loss=0.162, over 27716.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3851, pruned_loss=0.1344, over 5695407.77 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3727, pruned_loss=0.1143, over 5638568.60 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3866, pruned_loss=0.1365, over 5690361.55 frames. ], batch size: 115, lr: 5.52e-03, grad_scale: 4.0
+2023-03-03 02:36:07,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:30,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:32,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:40,176 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:44,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:36:48,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2196, 1.2868, 1.3931, 1.2269], device='cuda:1'), covar=tensor([0.1131, 0.1376, 0.1526, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0747, 0.0649, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 02:36:53,641 INFO [train.py:968] (1/2) Epoch 6, batch 7950, giga_loss[loss=0.3123, simple_loss=0.375, pruned_loss=0.1248, over 28656.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3841, pruned_loss=0.1341, over 5700009.75 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3725, pruned_loss=0.1141, over 5640792.79 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3856, pruned_loss=0.1361, over 5694532.66 frames. ], batch size: 242, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:37:01,759 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:02,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-03 02:37:08,237 INFO [optim.py:369] (1/2) 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 02:37:23,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:25,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:35,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3324, 1.4619, 1.2457, 1.4382], device='cuda:1'), covar=tensor([0.2149, 0.2122, 0.2210, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.1145, 0.0879, 0.1012, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 02:37:40,767 INFO [train.py:968] (1/2) Epoch 6, batch 8000, giga_loss[loss=0.4071, simple_loss=0.4311, pruned_loss=0.1915, over 27685.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3854, pruned_loss=0.1351, over 5694060.39 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3728, pruned_loss=0.1144, over 5647024.79 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3867, pruned_loss=0.1369, over 5685286.93 frames. ], batch size: 472, lr: 5.51e-03, grad_scale: 8.0
+2023-03-03 02:37:53,837 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:54,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:37:56,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:38:20,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=235753.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 02:38:26,547 INFO [train.py:968] (1/2) Epoch 6, batch 8050, giga_loss[loss=0.3589, simple_loss=0.4139, pruned_loss=0.1519, over 28640.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3858, pruned_loss=0.1343, over 5692569.08 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3732, pruned_loss=0.1145, over 5653022.38 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3868, pruned_loss=0.1361, over 5680910.72 frames. ], batch size: 92, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:38:40,981 INFO [optim.py:369] (1/2) 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 02:38:54,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:38:55,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7008, 1.7220, 1.7495, 1.5211], device='cuda:1'), covar=tensor([0.0994, 0.1430, 0.1315, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0738, 0.0644, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 02:38:56,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:39:09,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-03 02:39:10,685 INFO [train.py:968] (1/2) Epoch 6, batch 8100, giga_loss[loss=0.2978, simple_loss=0.3742, pruned_loss=0.1107, over 29012.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3851, pruned_loss=0.1329, over 5686081.91 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3726, pruned_loss=0.114, over 5658585.95 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3866, pruned_loss=0.1353, over 5672335.59 frames. ], batch size: 155, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:39:25,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:39:38,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7589, 2.3862, 1.7374, 0.8589], device='cuda:1'), covar=tensor([0.2791, 0.1536, 0.2332, 0.3206], device='cuda:1'), in_proj_covar=tensor([0.1420, 0.1339, 0.1386, 0.1181], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 02:39:56,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 02:40:00,076 INFO [train.py:968] (1/2) Epoch 6, batch 8150, giga_loss[loss=0.3074, simple_loss=0.3712, pruned_loss=0.1218, over 29034.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3865, pruned_loss=0.1344, over 5687898.90 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3728, pruned_loss=0.1141, over 5661064.88 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3876, pruned_loss=0.1363, over 5674990.34 frames. ], batch size: 128, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:40:07,795 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:09,601 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:10,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:11,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:15,342 INFO [optim.py:369] (1/2) 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 02:40:24,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-03 02:40:40,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:41,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:40:49,485 INFO [train.py:968] (1/2) Epoch 6, batch 8200, giga_loss[loss=0.3755, simple_loss=0.4319, pruned_loss=0.1595, over 28982.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3895, pruned_loss=0.1371, over 5691376.85 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3727, pruned_loss=0.114, over 5667035.93 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1392, over 5676350.33 frames. ], batch size: 145, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:40:53,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2557, 5.1301, 4.8142, 2.3404], device='cuda:1'), covar=tensor([0.0343, 0.0438, 0.0658, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0850, 0.0823, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 02:41:29,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3162, 1.7236, 1.3668, 1.5007], device='cuda:1'), covar=tensor([0.0678, 0.0341, 0.0307, 0.0737], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0122, 0.0125, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0067], device='cuda:1')
+2023-03-03 02:41:43,017 INFO [train.py:968] (1/2) Epoch 6, batch 8250, giga_loss[loss=0.3343, simple_loss=0.3873, pruned_loss=0.1407, over 28760.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3908, pruned_loss=0.1393, over 5690600.09 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3726, pruned_loss=0.1139, over 5669583.93 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3921, pruned_loss=0.1413, over 5676780.31 frames. ], batch size: 119, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:41:58,465 INFO [optim.py:369] (1/2) 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 02:42:33,197 INFO [train.py:968] (1/2) Epoch 6, batch 8300, giga_loss[loss=0.3679, simple_loss=0.4109, pruned_loss=0.1624, over 28536.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3923, pruned_loss=0.142, over 5679245.69 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3723, pruned_loss=0.1138, over 5674327.44 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.394, pruned_loss=0.1441, over 5664415.48 frames. ], batch size: 336, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:42:39,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-03 02:42:45,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:43:24,921 INFO [train.py:968] (1/2) Epoch 6, batch 8350, giga_loss[loss=0.2942, simple_loss=0.3599, pruned_loss=0.1143, over 29026.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.394, pruned_loss=0.1443, over 5673821.76 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3725, pruned_loss=0.1139, over 5675592.81 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3952, pruned_loss=0.146, over 5661099.16 frames. ], batch size: 128, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:43:36,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:43:41,621 INFO [optim.py:369] (1/2) 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 02:43:50,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5620, 2.2485, 1.5749, 0.9107], device='cuda:1'), covar=tensor([0.2445, 0.1280, 0.2199, 0.2674], device='cuda:1'), in_proj_covar=tensor([0.1407, 0.1329, 0.1381, 0.1176], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 02:43:59,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6635, 1.3386, 5.1218, 3.7519], device='cuda:1'), covar=tensor([0.2222, 0.2976, 0.0675, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0583, 0.0538, 0.0766, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 02:44:08,813 INFO [train.py:968] (1/2) Epoch 6, batch 8400, giga_loss[loss=0.349, simple_loss=0.3974, pruned_loss=0.1504, over 28563.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3918, pruned_loss=0.1423, over 5678672.89 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3728, pruned_loss=0.1143, over 5684999.80 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3934, pruned_loss=0.1446, over 5659659.26 frames. ], batch size: 336, lr: 5.51e-03, grad_scale: 8.0
+2023-03-03 02:44:22,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236128.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 02:44:29,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5689, 1.9468, 1.9387, 1.6963], device='cuda:1'), covar=tensor([0.1647, 0.1946, 0.1233, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0740, 0.0792, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 02:44:52,176 INFO [train.py:968] (1/2) Epoch 6, batch 8450, giga_loss[loss=0.3929, simple_loss=0.4173, pruned_loss=0.1842, over 26455.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3915, pruned_loss=0.1411, over 5680799.26 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3729, pruned_loss=0.1144, over 5687837.24 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3932, pruned_loss=0.1436, over 5663154.98 frames. ], batch size: 555, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:44:56,589 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:44:58,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:45:05,284 INFO [optim.py:369] (1/2) 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 02:45:11,763 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:45:23,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:45:35,469 INFO [train.py:968] (1/2) Epoch 6, batch 8500, giga_loss[loss=0.3084, simple_loss=0.3728, pruned_loss=0.122, over 28564.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3893, pruned_loss=0.1383, over 5678173.95 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3733, pruned_loss=0.1146, over 5695152.33 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3909, pruned_loss=0.1409, over 5657126.37 frames. ], batch size: 307, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:46:18,662 INFO [train.py:968] (1/2) Epoch 6, batch 8550, giga_loss[loss=0.3134, simple_loss=0.3796, pruned_loss=0.1236, over 28775.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.387, pruned_loss=0.1367, over 5683392.01 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3734, pruned_loss=0.1148, over 5696655.78 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3883, pruned_loss=0.1389, over 5665313.97 frames. ], batch size: 66, lr: 5.51e-03, grad_scale: 4.0
+2023-03-03 02:46:29,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236271.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 02:46:32,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236274.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 02:46:37,974 INFO [optim.py:369] (1/2) 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 02:47:00,307 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236303.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 02:47:06,297 INFO [train.py:968] (1/2) Epoch 6, batch 8600, giga_loss[loss=0.3491, simple_loss=0.4013, pruned_loss=0.1485, over 28595.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1354, over 5688884.67 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3738, pruned_loss=0.1149, over 5702149.78 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3858, pruned_loss=0.1377, over 5668971.05 frames. ], batch size: 307, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:47:17,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5428, 1.6150, 1.6172, 1.4992], device='cuda:1'), covar=tensor([0.1061, 0.1643, 0.1466, 0.1459], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0739, 0.0643, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 02:47:46,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 02:47:53,121 INFO [train.py:968] (1/2) Epoch 6, batch 8650, giga_loss[loss=0.2984, simple_loss=0.3615, pruned_loss=0.1176, over 28116.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3846, pruned_loss=0.1361, over 5672763.90 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3736, pruned_loss=0.1148, over 5698275.10 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3859, pruned_loss=0.1386, over 5659270.35 frames. ], batch size: 77, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:48:11,672 INFO [optim.py:369] (1/2) 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 02:48:42,612 INFO [train.py:968] (1/2) Epoch 6, batch 8700, giga_loss[loss=0.3321, simple_loss=0.4026, pruned_loss=0.1307, over 29020.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3881, pruned_loss=0.1384, over 5667287.59 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3736, pruned_loss=0.1147, over 5702291.53 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3895, pruned_loss=0.141, over 5652230.77 frames. ], batch size: 155, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:48:42,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=236411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:48:56,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5736, 1.9377, 1.9240, 1.7133], device='cuda:1'), covar=tensor([0.1695, 0.1888, 0.1225, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0740, 0.0796, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 02:49:08,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2146, 1.6227, 1.3526, 1.4488], device='cuda:1'), covar=tensor([0.0723, 0.0343, 0.0320, 0.0762], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0121, 0.0125, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0060, 0.0044, 0.0040, 0.0067], device='cuda:1')
+2023-03-03 02:49:15,580 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:49:28,877 INFO [train.py:968] (1/2) Epoch 6, batch 8750, giga_loss[loss=0.3136, simple_loss=0.3878, pruned_loss=0.1197, over 28707.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3907, pruned_loss=0.1378, over 5668934.99 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3738, pruned_loss=0.1149, over 5705117.31 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3922, pruned_loss=0.1406, over 5653176.53 frames. ], batch size: 242, lr: 5.51e-03, grad_scale: 2.0
+2023-03-03 02:49:46,439 INFO [optim.py:369] (1/2) 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 02:49:55,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4350, 1.7781, 1.3428, 1.0802], device='cuda:1'), covar=tensor([0.1481, 0.1083, 0.0832, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.1468, 0.1302, 0.1265, 0.1370], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 02:49:58,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7071, 2.0539, 1.9553, 1.7876], device='cuda:1'), covar=tensor([0.1603, 0.1832, 0.1198, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0738, 0.0796, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 02:50:15,930 INFO [train.py:968] (1/2) Epoch 6, batch 8800, giga_loss[loss=0.3061, simple_loss=0.3756, pruned_loss=0.1183, over 28599.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3924, pruned_loss=0.1367, over 5681908.61 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3735, pruned_loss=0.1149, over 5709265.39 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3942, pruned_loss=0.1394, over 5665090.44 frames. ], batch size: 307, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:50:54,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2702, 1.4372, 1.2382, 1.4440], device='cuda:1'), covar=tensor([0.0670, 0.0442, 0.0318, 0.0695], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0121, 0.0125, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0067], device='cuda:1')
+2023-03-03 02:50:59,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:51:00,763 INFO [train.py:968] (1/2) Epoch 6, batch 8850, giga_loss[loss=0.3662, simple_loss=0.4198, pruned_loss=0.1563, over 28745.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3932, pruned_loss=0.1372, over 5683548.97 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3726, pruned_loss=0.1144, over 5718321.12 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3963, pruned_loss=0.1408, over 5660427.49 frames. ], batch size: 284, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:51:16,145 INFO [optim.py:369] (1/2) 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 02:51:25,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:51:28,227 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 8900, libri_loss[loss=0.297, simple_loss=0.3837, pruned_loss=0.1052, over 29262.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3959, pruned_loss=0.1401, over 5673214.38 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3723, pruned_loss=0.1141, over 5721002.25 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.399, pruned_loss=0.1436, over 5652023.75 frames. ], batch size: 94, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:51:55,842 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:52:31,223 INFO [train.py:968] (1/2) Epoch 6, batch 8950, giga_loss[loss=0.3269, simple_loss=0.3887, pruned_loss=0.1326, over 28576.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3953, pruned_loss=0.14, over 5674590.31 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3725, pruned_loss=0.1142, over 5723273.74 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3981, pruned_loss=0.1432, over 5654752.26 frames. ], batch size: 307, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:52:50,422 INFO [optim.py:369] (1/2) 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 02:52:50,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-03 02:53:09,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:53:12,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:53:18,995 INFO [train.py:968] (1/2) Epoch 6, batch 9000, giga_loss[loss=0.343, simple_loss=0.3915, pruned_loss=0.1472, over 28903.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3946, pruned_loss=0.141, over 5662644.34 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3722, pruned_loss=0.1139, over 5727019.79 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3976, pruned_loss=0.1444, over 5642318.33 frames. ], batch size: 227, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:53:18,995 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 02:53:27,484 INFO [train.py:1012] (1/2) Epoch 6, validation: loss=0.2325, simple_loss=0.3368, pruned_loss=0.0641, over 944034.00 frames. 
+2023-03-03 02:53:27,484 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 02:53:36,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1743, 1.3722, 1.0064, 0.8712], device='cuda:1'), covar=tensor([0.1078, 0.0941, 0.0725, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.1456, 0.1289, 0.1253, 0.1351], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 02:53:42,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3606, 2.9421, 1.4824, 1.3744], device='cuda:1'), covar=tensor([0.0857, 0.0346, 0.0787, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0487, 0.0313, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 02:53:43,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9047, 1.0750, 3.4476, 2.9120], device='cuda:1'), covar=tensor([0.1653, 0.2350, 0.0425, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0587, 0.0544, 0.0779, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 02:53:50,906 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:54:15,667 INFO [train.py:968] (1/2) Epoch 6, batch 9050, giga_loss[loss=0.31, simple_loss=0.3704, pruned_loss=0.1248, over 28891.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3921, pruned_loss=0.1398, over 5664067.44 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3717, pruned_loss=0.1136, over 5726494.70 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3953, pruned_loss=0.1433, over 5646882.52 frames. ], batch size: 136, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:54:35,009 INFO [optim.py:369] (1/2) 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 02:54:40,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:55:03,549 INFO [train.py:968] (1/2) Epoch 6, batch 9100, giga_loss[loss=0.316, simple_loss=0.3802, pruned_loss=0.1259, over 28884.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3901, pruned_loss=0.139, over 5668374.33 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3714, pruned_loss=0.1135, over 5731978.99 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3935, pruned_loss=0.1427, over 5647984.51 frames. ], batch size: 186, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:55:16,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5474, 2.9911, 1.6015, 1.5155], device='cuda:1'), covar=tensor([0.0756, 0.0290, 0.0727, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0488, 0.0314, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 02:55:52,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4133, 1.8648, 1.3169, 1.5095], device='cuda:1'), covar=tensor([0.0699, 0.0271, 0.0303, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0122, 0.0125, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0068], device='cuda:1')
+2023-03-03 02:55:55,834 INFO [train.py:968] (1/2) Epoch 6, batch 9150, giga_loss[loss=0.3317, simple_loss=0.3898, pruned_loss=0.1368, over 28935.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3914, pruned_loss=0.1408, over 5665482.91 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3716, pruned_loss=0.1137, over 5734831.97 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3942, pruned_loss=0.1439, over 5646066.87 frames. ], batch size: 186, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 02:56:16,073 INFO [optim.py:369] (1/2) 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 02:56:43,316 INFO [train.py:968] (1/2) Epoch 6, batch 9200, giga_loss[loss=0.3012, simple_loss=0.3628, pruned_loss=0.1198, over 28990.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3912, pruned_loss=0.1407, over 5660213.60 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3721, pruned_loss=0.1139, over 5736563.26 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3936, pruned_loss=0.1438, over 5640878.14 frames. ], batch size: 164, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:57:00,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:57:03,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:57:28,338 INFO [train.py:968] (1/2) Epoch 6, batch 9250, giga_loss[loss=0.3355, simple_loss=0.3942, pruned_loss=0.1384, over 28957.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3882, pruned_loss=0.139, over 5662924.91 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3717, pruned_loss=0.1136, over 5734331.14 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3912, pruned_loss=0.1429, over 5645941.08 frames. ], batch size: 164, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:57:28,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 02:57:45,204 INFO [optim.py:369] (1/2) 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 02:58:13,816 INFO [train.py:968] (1/2) Epoch 6, batch 9300, libri_loss[loss=0.3054, simple_loss=0.3852, pruned_loss=0.1128, over 29532.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.388, pruned_loss=0.1388, over 5653454.34 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3723, pruned_loss=0.1139, over 5728252.37 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3902, pruned_loss=0.1423, over 5642690.68 frames. ], batch size: 83, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:58:22,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4109, 1.0315, 4.8840, 3.5456], device='cuda:1'), covar=tensor([0.1580, 0.2484, 0.0326, 0.0631], device='cuda:1'), in_proj_covar=tensor([0.0580, 0.0538, 0.0774, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 02:59:01,734 INFO [train.py:968] (1/2) Epoch 6, batch 9350, giga_loss[loss=0.4664, simple_loss=0.471, pruned_loss=0.2309, over 26542.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.39, pruned_loss=0.1391, over 5660752.87 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3729, pruned_loss=0.1142, over 5731694.57 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3916, pruned_loss=0.1423, over 5647289.84 frames. ], batch size: 555, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 02:59:21,744 INFO [optim.py:369] (1/2) 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 02:59:48,170 INFO [train.py:968] (1/2) Epoch 6, batch 9400, giga_loss[loss=0.3225, simple_loss=0.3916, pruned_loss=0.1267, over 29056.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3918, pruned_loss=0.1397, over 5667257.90 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3731, pruned_loss=0.1142, over 5734463.75 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3934, pruned_loss=0.1429, over 5652453.79 frames. ], batch size: 155, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 03:00:05,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8486, 2.5419, 1.6253, 2.3155], device='cuda:1'), covar=tensor([0.0489, 0.0504, 0.0879, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0449, 0.0498, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 03:00:30,340 INFO [train.py:968] (1/2) Epoch 6, batch 9450, giga_loss[loss=0.3791, simple_loss=0.3958, pruned_loss=0.1811, over 23660.00 frames. ], tot_loss[loss=0.335, simple_loss=0.391, pruned_loss=0.1395, over 5648829.47 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3729, pruned_loss=0.114, over 5723347.14 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3932, pruned_loss=0.1432, over 5644920.89 frames. ], batch size: 705, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 03:00:50,330 INFO [optim.py:369] (1/2) 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 03:01:18,382 INFO [train.py:968] (1/2) Epoch 6, batch 9500, giga_loss[loss=0.3297, simple_loss=0.3918, pruned_loss=0.1339, over 29007.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3918, pruned_loss=0.1382, over 5658182.28 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3726, pruned_loss=0.1138, over 5725168.50 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3941, pruned_loss=0.1417, over 5652209.83 frames. ], batch size: 128, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 03:01:24,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2849, 1.7733, 1.2520, 0.5627], device='cuda:1'), covar=tensor([0.2582, 0.1478, 0.1508, 0.2902], device='cuda:1'), in_proj_covar=tensor([0.1419, 0.1345, 0.1374, 0.1182], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 03:01:37,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0417, 1.2053, 1.0049, 0.7670], device='cuda:1'), covar=tensor([0.0868, 0.0746, 0.0580, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.1476, 0.1302, 0.1265, 0.1362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 03:02:01,219 INFO [train.py:968] (1/2) Epoch 6, batch 9550, libri_loss[loss=0.2978, simple_loss=0.3685, pruned_loss=0.1135, over 29464.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3917, pruned_loss=0.1359, over 5665217.57 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.372, pruned_loss=0.1136, over 5723231.63 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3951, pruned_loss=0.1401, over 5658790.87 frames. ], batch size: 85, lr: 5.50e-03, grad_scale: 2.0
+2023-03-03 03:02:18,331 INFO [optim.py:369] (1/2) 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 03:02:21,051 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=237285.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:02:44,095 INFO [train.py:968] (1/2) Epoch 6, batch 9600, giga_loss[loss=0.2923, simple_loss=0.373, pruned_loss=0.1058, over 28940.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3964, pruned_loss=0.138, over 5674591.90 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3724, pruned_loss=0.1138, over 5723151.55 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3992, pruned_loss=0.1417, over 5668623.48 frames. ], batch size: 199, lr: 5.50e-03, grad_scale: 4.0
+2023-03-03 03:03:35,492 INFO [train.py:968] (1/2) Epoch 6, batch 9650, giga_loss[loss=0.3313, simple_loss=0.3938, pruned_loss=0.1345, over 28573.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3998, pruned_loss=0.1416, over 5663660.67 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3726, pruned_loss=0.1139, over 5721280.17 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.4021, pruned_loss=0.1447, over 5660146.16 frames. ], batch size: 307, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:03:58,053 INFO [optim.py:369] (1/2) 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 03:04:24,486 INFO [train.py:968] (1/2) Epoch 6, batch 9700, giga_loss[loss=0.3672, simple_loss=0.4171, pruned_loss=0.1586, over 28870.00 frames. ], tot_loss[loss=0.347, simple_loss=0.4034, pruned_loss=0.1453, over 5675000.83 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3731, pruned_loss=0.1142, over 5723715.72 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4052, pruned_loss=0.1479, over 5669206.48 frames. ], batch size: 186, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:04:27,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-03 03:05:08,763 INFO [train.py:968] (1/2) Epoch 6, batch 9750, libri_loss[loss=0.2612, simple_loss=0.3401, pruned_loss=0.0911, over 29572.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.403, pruned_loss=0.146, over 5666450.30 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3725, pruned_loss=0.1139, over 5729610.00 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4065, pruned_loss=0.1499, over 5653593.57 frames. ], batch size: 78, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:05:28,191 INFO [optim.py:369] (1/2) 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 03:05:39,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0194, 1.2088, 3.4398, 2.9676], device='cuda:1'), covar=tensor([0.1449, 0.2156, 0.0429, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0585, 0.0539, 0.0778, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 03:05:52,977 INFO [train.py:968] (1/2) Epoch 6, batch 9800, giga_loss[loss=0.3512, simple_loss=0.4068, pruned_loss=0.1478, over 29043.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.4005, pruned_loss=0.1443, over 5669560.89 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3716, pruned_loss=0.1136, over 5731185.25 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.405, pruned_loss=0.1487, over 5656026.57 frames. ], batch size: 155, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:06:37,155 INFO [train.py:968] (1/2) Epoch 6, batch 9850, giga_loss[loss=0.3672, simple_loss=0.415, pruned_loss=0.1597, over 28870.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3998, pruned_loss=0.1422, over 5677508.87 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3717, pruned_loss=0.1136, over 5734677.85 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4039, pruned_loss=0.1464, over 5662421.54 frames. ], batch size: 285, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:06:56,219 INFO [optim.py:369] (1/2) 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 03:07:25,934 INFO [train.py:968] (1/2) Epoch 6, batch 9900, giga_loss[loss=0.3263, simple_loss=0.4005, pruned_loss=0.1261, over 29041.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.4008, pruned_loss=0.1418, over 5668742.57 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3717, pruned_loss=0.1136, over 5727111.31 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.4044, pruned_loss=0.1454, over 5662628.17 frames. ], batch size: 136, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:07:44,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2090, 1.4820, 1.2184, 1.2321], device='cuda:1'), covar=tensor([0.2106, 0.2005, 0.2155, 0.1863], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.0889, 0.1018, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 03:08:07,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=237660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:08:07,628 INFO [train.py:968] (1/2) Epoch 6, batch 9950, giga_loss[loss=0.3698, simple_loss=0.4222, pruned_loss=0.1587, over 28650.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3991, pruned_loss=0.1401, over 5675533.35 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3712, pruned_loss=0.1133, over 5730245.79 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.4037, pruned_loss=0.1444, over 5665555.18 frames. ], batch size: 242, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:08:29,335 INFO [optim.py:369] (1/2) 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 03:08:58,011 INFO [train.py:968] (1/2) Epoch 6, batch 10000, libri_loss[loss=0.2948, simple_loss=0.3752, pruned_loss=0.1072, over 29270.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3996, pruned_loss=0.1416, over 5668019.57 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3707, pruned_loss=0.1129, over 5733664.18 frames. ], giga_tot_loss[loss=0.3481, simple_loss=0.4042, pruned_loss=0.146, over 5656023.25 frames. ], batch size: 94, lr: 5.49e-03, grad_scale: 8.0
+2023-03-03 03:09:21,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3733, 1.4946, 1.5863, 1.4188], device='cuda:1'), covar=tensor([0.1046, 0.1183, 0.1379, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0744, 0.0643, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 03:09:22,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 03:09:31,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3410, 1.3920, 1.4004, 1.2978], device='cuda:1'), covar=tensor([0.1047, 0.1316, 0.1522, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0742, 0.0642, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 03:09:45,751 INFO [train.py:968] (1/2) Epoch 6, batch 10050, giga_loss[loss=0.2989, simple_loss=0.3627, pruned_loss=0.1175, over 28953.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3975, pruned_loss=0.1412, over 5665922.66 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3707, pruned_loss=0.1129, over 5736108.30 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.4015, pruned_loss=0.145, over 5653441.66 frames. ], batch size: 145, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:10:10,246 INFO [optim.py:369] (1/2) 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 03:10:30,104 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237803.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:10:30,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3584, 1.5422, 1.2274, 0.9551], device='cuda:1'), covar=tensor([0.1029, 0.0918, 0.0794, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.1490, 0.1307, 0.1285, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 03:10:33,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=237806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:10:36,959 INFO [train.py:968] (1/2) Epoch 6, batch 10100, giga_loss[loss=0.3845, simple_loss=0.4327, pruned_loss=0.1682, over 28766.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3974, pruned_loss=0.1428, over 5664866.31 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3707, pruned_loss=0.113, over 5739752.05 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.401, pruned_loss=0.1463, over 5650437.57 frames. ], batch size: 284, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:11:01,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237835.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:11:26,935 INFO [train.py:968] (1/2) Epoch 6, batch 10150, giga_loss[loss=0.3272, simple_loss=0.3868, pruned_loss=0.1339, over 28944.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3945, pruned_loss=0.1416, over 5664930.64 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3706, pruned_loss=0.1128, over 5742392.85 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3979, pruned_loss=0.145, over 5649837.54 frames. ], batch size: 186, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:11:43,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8029, 1.1383, 3.2846, 2.8573], device='cuda:1'), covar=tensor([0.1631, 0.2236, 0.0452, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0538, 0.0776, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 03:11:49,416 INFO [optim.py:369] (1/2) 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 03:11:49,726 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=237884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:12:18,978 INFO [train.py:968] (1/2) Epoch 6, batch 10200, giga_loss[loss=0.2791, simple_loss=0.352, pruned_loss=0.1031, over 28987.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3941, pruned_loss=0.1419, over 5660134.48 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3712, pruned_loss=0.1132, over 5745003.25 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3969, pruned_loss=0.1452, over 5643385.27 frames. ], batch size: 128, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:13:00,943 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=237956.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:13:05,531 INFO [train.py:968] (1/2) Epoch 6, batch 10250, giga_loss[loss=0.2779, simple_loss=0.3475, pruned_loss=0.1041, over 29013.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3932, pruned_loss=0.1415, over 5670898.03 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3714, pruned_loss=0.1132, over 5745257.98 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3958, pruned_loss=0.1447, over 5655238.50 frames. ], batch size: 136, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:13:23,375 INFO [optim.py:369] (1/2) 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 03:13:30,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8680, 3.7052, 3.5109, 1.8294], device='cuda:1'), covar=tensor([0.0568, 0.0607, 0.0755, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0864, 0.0824, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 03:13:52,108 INFO [train.py:968] (1/2) Epoch 6, batch 10300, giga_loss[loss=0.2994, simple_loss=0.373, pruned_loss=0.1129, over 28658.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3895, pruned_loss=0.1381, over 5667892.37 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3708, pruned_loss=0.1128, over 5747335.13 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3924, pruned_loss=0.1415, over 5652460.40 frames. ], batch size: 262, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:14:35,816 INFO [train.py:968] (1/2) Epoch 6, batch 10350, libri_loss[loss=0.2801, simple_loss=0.359, pruned_loss=0.1007, over 29258.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3874, pruned_loss=0.1349, over 5666022.80 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3713, pruned_loss=0.113, over 5750435.78 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3899, pruned_loss=0.1381, over 5648045.48 frames. ], batch size: 97, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:14:46,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2327, 1.7110, 1.5618, 1.3722], device='cuda:1'), covar=tensor([0.1440, 0.1863, 0.1178, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0744, 0.0800, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 03:15:02,332 INFO [optim.py:369] (1/2) 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 03:15:25,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6987, 4.7095, 1.9940, 1.5702], device='cuda:1'), covar=tensor([0.0838, 0.0239, 0.0742, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0487, 0.0314, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 03:15:28,601 INFO [train.py:968] (1/2) Epoch 6, batch 10400, giga_loss[loss=0.3719, simple_loss=0.4129, pruned_loss=0.1654, over 28790.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.386, pruned_loss=0.1332, over 5672413.86 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3714, pruned_loss=0.1131, over 5752091.78 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.388, pruned_loss=0.1359, over 5656011.96 frames. ], batch size: 85, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:15:51,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5310, 2.9976, 1.5636, 1.4608], device='cuda:1'), covar=tensor([0.0745, 0.0308, 0.0743, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0488, 0.0315, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 03:16:17,384 INFO [train.py:968] (1/2) Epoch 6, batch 10450, giga_loss[loss=0.2887, simple_loss=0.3512, pruned_loss=0.1131, over 28906.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3856, pruned_loss=0.1337, over 5668018.71 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3719, pruned_loss=0.1134, over 5748128.85 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.387, pruned_loss=0.1359, over 5657026.21 frames. ], batch size: 186, lr: 5.49e-03, grad_scale: 4.0
+2023-03-03 03:16:44,543 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:16:49,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3068, 1.8773, 1.3815, 1.5182], device='cuda:1'), covar=tensor([0.0711, 0.0290, 0.0307, 0.0772], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0121, 0.0125, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0068], device='cuda:1')
+2023-03-03 03:17:09,552 INFO [train.py:968] (1/2) Epoch 6, batch 10500, giga_loss[loss=0.3142, simple_loss=0.352, pruned_loss=0.1382, over 23484.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3822, pruned_loss=0.1328, over 5668832.57 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3716, pruned_loss=0.1132, over 5750573.53 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3838, pruned_loss=0.1351, over 5656460.33 frames. ], batch size: 705, lr: 5.49e-03, grad_scale: 2.0
+2023-03-03 03:17:26,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8704, 3.7029, 3.5353, 1.8745], device='cuda:1'), covar=tensor([0.0479, 0.0540, 0.0679, 0.2092], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0854, 0.0816, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 03:17:38,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9802, 1.9027, 1.4957, 2.1794], device='cuda:1'), covar=tensor([0.1854, 0.1943, 0.2037, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.1154, 0.0889, 0.1021, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 03:17:57,751 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 10550, giga_loss[loss=0.3021, simple_loss=0.3685, pruned_loss=0.1178, over 28613.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3834, pruned_loss=0.1336, over 5669893.91 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3715, pruned_loss=0.1131, over 5752082.06 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3848, pruned_loss=0.1356, over 5658091.34 frames. ], batch size: 92, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:18:21,081 INFO [optim.py:369] (1/2) 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 03:18:29,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 03:18:40,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:18:43,539 INFO [train.py:968] (1/2) Epoch 6, batch 10600, libri_loss[loss=0.335, simple_loss=0.4113, pruned_loss=0.1294, over 25795.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3848, pruned_loss=0.1335, over 5668025.86 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3719, pruned_loss=0.1133, over 5751032.38 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3859, pruned_loss=0.1354, over 5657757.66 frames. ], batch size: 136, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:19:06,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238331.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:19:35,235 INFO [train.py:968] (1/2) Epoch 6, batch 10650, giga_loss[loss=0.3421, simple_loss=0.3952, pruned_loss=0.1444, over 28501.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3847, pruned_loss=0.1335, over 5664480.11 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3721, pruned_loss=0.1135, over 5754732.78 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3857, pruned_loss=0.1355, over 5650802.63 frames. ], batch size: 336, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:19:57,516 INFO [optim.py:369] (1/2) 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 03:20:14,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:20:18,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:20:22,242 INFO [train.py:968] (1/2) Epoch 6, batch 10700, giga_loss[loss=0.3144, simple_loss=0.3788, pruned_loss=0.125, over 28869.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3837, pruned_loss=0.1327, over 5660548.74 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3722, pruned_loss=0.1134, over 5756831.36 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3847, pruned_loss=0.1348, over 5645475.68 frames. ], batch size: 186, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:20:45,051 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:20:51,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-03 03:21:04,478 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:21:09,072 INFO [train.py:968] (1/2) Epoch 6, batch 10750, giga_loss[loss=0.3832, simple_loss=0.4288, pruned_loss=0.1688, over 28670.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3848, pruned_loss=0.1339, over 5665415.83 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3723, pruned_loss=0.1133, over 5759988.62 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3857, pruned_loss=0.136, over 5648758.56 frames. ], batch size: 262, lr: 5.48e-03, grad_scale: 2.0
+2023-03-03 03:21:17,142 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:21:25,196 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238474.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:21:27,098 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:21:56,285 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238506.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:22:00,628 INFO [train.py:968] (1/2) Epoch 6, batch 10800, giga_loss[loss=0.4123, simple_loss=0.4437, pruned_loss=0.1904, over 28354.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3874, pruned_loss=0.1359, over 5664931.04 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.372, pruned_loss=0.1131, over 5763728.87 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3888, pruned_loss=0.1382, over 5646115.74 frames. ], batch size: 368, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:22:05,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5129, 4.3618, 4.1939, 1.6738], device='cuda:1'), covar=tensor([0.0446, 0.0515, 0.0720, 0.2202], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0860, 0.0823, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 03:22:08,693 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:22:28,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 03:22:45,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:22:47,389 INFO [train.py:968] (1/2) Epoch 6, batch 10850, libri_loss[loss=0.3445, simple_loss=0.4163, pruned_loss=0.1364, over 29514.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3896, pruned_loss=0.1368, over 5669294.21 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3725, pruned_loss=0.1133, over 5763310.44 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3906, pruned_loss=0.1393, over 5651788.76 frames. ], batch size: 84, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:23:07,594 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 10900, giga_loss[loss=0.384, simple_loss=0.4208, pruned_loss=0.1736, over 28556.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3909, pruned_loss=0.1378, over 5685499.71 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3723, pruned_loss=0.1132, over 5768076.31 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3926, pruned_loss=0.1407, over 5663814.16 frames. ], batch size: 336, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:24:19,788 INFO [train.py:968] (1/2) Epoch 6, batch 10950, giga_loss[loss=0.3669, simple_loss=0.4266, pruned_loss=0.1536, over 28750.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3921, pruned_loss=0.1394, over 5688742.32 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3721, pruned_loss=0.1131, over 5770997.32 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3941, pruned_loss=0.1423, over 5667549.54 frames. ], batch size: 284, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:24:39,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:24:45,090 INFO [optim.py:369] (1/2) 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 03:24:58,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3654, 1.6883, 1.2973, 1.1638], device='cuda:1'), covar=tensor([0.1317, 0.0858, 0.0791, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.1477, 0.1316, 0.1284, 0.1369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 03:25:01,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:25:04,012 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 11000, libri_loss[loss=0.3612, simple_loss=0.4154, pruned_loss=0.1535, over 19906.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3947, pruned_loss=0.1402, over 5667392.09 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3723, pruned_loss=0.1132, over 5761797.03 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3965, pruned_loss=0.143, over 5657001.98 frames. ], batch size: 186, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:25:32,614 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 11050, giga_loss[loss=0.3492, simple_loss=0.4071, pruned_loss=0.1456, over 28840.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3937, pruned_loss=0.1395, over 5650729.72 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3721, pruned_loss=0.1131, over 5752257.51 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.396, pruned_loss=0.1427, over 5647993.31 frames. ], batch size: 174, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:26:25,205 INFO [optim.py:369] (1/2) 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 03:26:49,573 INFO [train.py:968] (1/2) Epoch 6, batch 11100, libri_loss[loss=0.2664, simple_loss=0.3507, pruned_loss=0.09106, over 29539.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3937, pruned_loss=0.1405, over 5653237.85 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3723, pruned_loss=0.1131, over 5756204.64 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3959, pruned_loss=0.1437, over 5644931.85 frames. ], batch size: 83, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:26:51,337 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=238813.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:27:04,959 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:07,912 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:12,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:28,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:27:44,780 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 11150, giga_loss[loss=0.3365, simple_loss=0.3903, pruned_loss=0.1414, over 28887.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3931, pruned_loss=0.1408, over 5638448.28 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3728, pruned_loss=0.1134, over 5748956.89 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3948, pruned_loss=0.1436, over 5635891.70 frames. ], batch size: 112, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:28:12,129 INFO [optim.py:369] (1/2) 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 03:28:22,112 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 11200, giga_loss[loss=0.3304, simple_loss=0.3848, pruned_loss=0.1381, over 28995.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3904, pruned_loss=0.1396, over 5642890.29 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3724, pruned_loss=0.1133, over 5751592.83 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3924, pruned_loss=0.1423, over 5637354.20 frames. ], batch size: 106, lr: 5.48e-03, grad_scale: 8.0
+2023-03-03 03:28:46,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3782, 1.7810, 1.7430, 1.5444], device='cuda:1'), covar=tensor([0.1441, 0.1859, 0.1117, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0745, 0.0799, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 03:29:09,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7366, 1.5860, 1.3907, 1.9403], device='cuda:1'), covar=tensor([0.1943, 0.2068, 0.2060, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1148, 0.0887, 0.1020, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 03:29:12,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2869, 1.3432, 1.1697, 1.4431], device='cuda:1'), covar=tensor([0.0769, 0.0326, 0.0333, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0122, 0.0126, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0062, 0.0045, 0.0040, 0.0068], device='cuda:1')
+2023-03-03 03:29:25,226 INFO [train.py:968] (1/2) Epoch 6, batch 11250, giga_loss[loss=0.3343, simple_loss=0.3941, pruned_loss=0.1372, over 28938.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3893, pruned_loss=0.139, over 5655415.53 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3722, pruned_loss=0.1132, over 5753079.33 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3913, pruned_loss=0.1419, over 5647420.98 frames. ], batch size: 164, lr: 5.48e-03, grad_scale: 8.0
+2023-03-03 03:29:40,202 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:29:42,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:29:51,048 INFO [optim.py:369] (1/2) 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 03:29:53,524 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:29:55,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:14,728 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:17,479 INFO [train.py:968] (1/2) Epoch 6, batch 11300, giga_loss[loss=0.3172, simple_loss=0.3858, pruned_loss=0.1243, over 29010.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3892, pruned_loss=0.1392, over 5655854.84 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3725, pruned_loss=0.1133, over 5754727.73 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3907, pruned_loss=0.1416, over 5647231.00 frames. ], batch size: 164, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:30:25,304 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:49,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:30:52,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:31:09,231 INFO [train.py:968] (1/2) Epoch 6, batch 11350, giga_loss[loss=0.3777, simple_loss=0.4164, pruned_loss=0.1695, over 28317.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3908, pruned_loss=0.1408, over 5646435.68 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3728, pruned_loss=0.1134, over 5743989.87 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3921, pruned_loss=0.1431, over 5647601.84 frames. ], batch size: 368, lr: 5.48e-03, grad_scale: 4.0
+2023-03-03 03:31:15,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:31:20,327 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=239074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:31:27,153 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-03 03:31:31,235 INFO [optim.py:369] (1/2) 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 03:31:49,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5806, 1.5256, 1.2801, 1.3165], device='cuda:1'), covar=tensor([0.0584, 0.0481, 0.0843, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0455, 0.0501, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 03:31:56,235 INFO [train.py:968] (1/2) Epoch 6, batch 11400, giga_loss[loss=0.4569, simple_loss=0.4664, pruned_loss=0.2237, over 26600.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3931, pruned_loss=0.1428, over 5646719.44 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3725, pruned_loss=0.1132, over 5744695.29 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3948, pruned_loss=0.1455, over 5644821.38 frames. ], batch size: 555, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:32:12,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6357, 1.6092, 1.2313, 1.3226], device='cuda:1'), covar=tensor([0.0597, 0.0486, 0.0869, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0454, 0.0501, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 03:32:27,959 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3734, 1.5879, 1.4156, 1.5501], device='cuda:1'), covar=tensor([0.0611, 0.0252, 0.0261, 0.0616], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0121, 0.0125, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0061, 0.0045, 0.0040, 0.0067], device='cuda:1')
+2023-03-03 03:32:42,532 INFO [train.py:968] (1/2) Epoch 6, batch 11450, libri_loss[loss=0.3212, simple_loss=0.3851, pruned_loss=0.1286, over 20331.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3948, pruned_loss=0.1439, over 5635709.80 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3727, pruned_loss=0.1133, over 5738775.47 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.3966, pruned_loss=0.1468, over 5637468.63 frames. ], batch size: 188, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:32:48,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4117, 1.7617, 1.7328, 1.5600], device='cuda:1'), covar=tensor([0.0889, 0.1163, 0.0720, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0744, 0.0802, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 03:33:08,898 INFO [optim.py:369] (1/2) 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 03:33:09,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8778, 4.7092, 4.4992, 2.0320], device='cuda:1'), covar=tensor([0.0356, 0.0427, 0.0594, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0855, 0.0823, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 03:33:10,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239188.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:33:24,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-03 03:33:30,299 INFO [train.py:968] (1/2) Epoch 6, batch 11500, giga_loss[loss=0.312, simple_loss=0.3752, pruned_loss=0.1244, over 28646.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3958, pruned_loss=0.1453, over 5644189.05 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3732, pruned_loss=0.1138, over 5743807.12 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.3975, pruned_loss=0.1482, over 5637941.00 frames. ], batch size: 307, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:34:10,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3372, 1.9415, 1.3728, 0.4858], device='cuda:1'), covar=tensor([0.1908, 0.1021, 0.1774, 0.2696], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1358, 0.1411, 0.1195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 03:34:19,178 INFO [train.py:968] (1/2) Epoch 6, batch 11550, libri_loss[loss=0.285, simple_loss=0.3628, pruned_loss=0.1035, over 29372.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3959, pruned_loss=0.1455, over 5656593.16 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.373, pruned_loss=0.1137, over 5745993.93 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.3977, pruned_loss=0.1483, over 5648256.11 frames. ], batch size: 92, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:34:42,100 INFO [optim.py:369] (1/2) 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 03:35:06,146 INFO [train.py:968] (1/2) Epoch 6, batch 11600, giga_loss[loss=0.3141, simple_loss=0.3739, pruned_loss=0.1271, over 28868.00 frames. ], tot_loss[loss=0.3441, simple_loss=0.3968, pruned_loss=0.1457, over 5645599.61 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3732, pruned_loss=0.1138, over 5740011.95 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3986, pruned_loss=0.1485, over 5641506.22 frames. ], batch size: 112, lr: 5.47e-03, grad_scale: 8.0
+2023-03-03 03:35:24,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=239329.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:35:26,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239331.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:35:28,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239334.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:35:41,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1130, 2.4146, 1.1818, 1.1532], device='cuda:1'), covar=tensor([0.0964, 0.0424, 0.0870, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0491, 0.0317, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 03:35:52,839 INFO [train.py:968] (1/2) Epoch 6, batch 11650, giga_loss[loss=0.3641, simple_loss=0.416, pruned_loss=0.1561, over 28669.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3958, pruned_loss=0.1438, over 5666097.51 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3728, pruned_loss=0.1136, over 5745023.83 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3984, pruned_loss=0.1473, over 5655528.86 frames. ], batch size: 284, lr: 5.47e-03, grad_scale: 8.0
+2023-03-03 03:35:55,876 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239363.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:36:20,620 INFO [optim.py:369] (1/2) 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 03:36:44,380 INFO [train.py:968] (1/2) Epoch 6, batch 11700, giga_loss[loss=0.3578, simple_loss=0.4072, pruned_loss=0.1542, over 29028.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3972, pruned_loss=0.1451, over 5655899.75 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3726, pruned_loss=0.1134, over 5745627.90 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4, pruned_loss=0.1488, over 5645163.94 frames. ], batch size: 128, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:37:18,214 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-03 03:37:21,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:37:35,099 INFO [train.py:968] (1/2) Epoch 6, batch 11750, libri_loss[loss=0.3149, simple_loss=0.3829, pruned_loss=0.1235, over 19316.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3986, pruned_loss=0.1466, over 5647103.06 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3724, pruned_loss=0.1133, over 5740197.13 frames. ], giga_tot_loss[loss=0.3515, simple_loss=0.4018, pruned_loss=0.1506, over 5642003.24 frames. ], batch size: 186, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:37:58,276 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 11800, giga_loss[loss=0.3215, simple_loss=0.3847, pruned_loss=0.1291, over 28591.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3989, pruned_loss=0.1475, over 5643368.69 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3727, pruned_loss=0.1135, over 5731058.78 frames. ], giga_tot_loss[loss=0.3513, simple_loss=0.4013, pruned_loss=0.1507, over 5646383.84 frames. ], batch size: 307, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:39:08,624 INFO [train.py:968] (1/2) Epoch 6, batch 11850, giga_loss[loss=0.3722, simple_loss=0.4124, pruned_loss=0.166, over 27486.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3982, pruned_loss=0.1456, over 5645816.00 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3723, pruned_loss=0.1132, over 5733334.72 frames. ], giga_tot_loss[loss=0.3497, simple_loss=0.4011, pruned_loss=0.1492, over 5644442.92 frames. ], batch size: 472, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:39:12,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2054, 1.8313, 1.3872, 0.3994], device='cuda:1'), covar=tensor([0.2083, 0.1614, 0.2397, 0.2854], device='cuda:1'), in_proj_covar=tensor([0.1427, 0.1356, 0.1394, 0.1186], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 03:39:32,449 INFO [optim.py:369] (1/2) 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 03:39:37,698 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:39:39,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:39:53,199 INFO [train.py:968] (1/2) Epoch 6, batch 11900, giga_loss[loss=0.3495, simple_loss=0.3985, pruned_loss=0.1503, over 28599.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3987, pruned_loss=0.1453, over 5640644.97 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3723, pruned_loss=0.1134, over 5728194.49 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4017, pruned_loss=0.149, over 5642181.41 frames. ], batch size: 85, lr: 5.47e-03, grad_scale: 2.0
+2023-03-03 03:40:04,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:40:39,380 INFO [train.py:968] (1/2) Epoch 6, batch 11950, giga_loss[loss=0.3409, simple_loss=0.3939, pruned_loss=0.1439, over 28004.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3959, pruned_loss=0.1424, over 5650940.00 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.372, pruned_loss=0.1131, over 5733118.22 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3991, pruned_loss=0.1462, over 5646367.35 frames. ], batch size: 412, lr: 5.47e-03, grad_scale: 2.0
+2023-03-03 03:40:57,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5605, 3.3977, 1.5371, 1.5941], device='cuda:1'), covar=tensor([0.0800, 0.0283, 0.0801, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0491, 0.0314, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 03:41:07,064 INFO [optim.py:369] (1/2) 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 03:41:21,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239704.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:41:26,910 INFO [train.py:968] (1/2) Epoch 6, batch 12000, giga_loss[loss=0.338, simple_loss=0.3938, pruned_loss=0.1411, over 28832.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3942, pruned_loss=0.1417, over 5642842.19 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3718, pruned_loss=0.113, over 5726506.75 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3974, pruned_loss=0.1453, over 5643179.72 frames. ], batch size: 119, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:41:26,910 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 03:41:35,219 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 03:42:27,874 INFO [train.py:968] (1/2) Epoch 6, batch 12050, giga_loss[loss=0.3702, simple_loss=0.4149, pruned_loss=0.1628, over 28571.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3957, pruned_loss=0.1428, over 5653067.44 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3719, pruned_loss=0.1131, over 5729452.40 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3985, pruned_loss=0.146, over 5649701.42 frames. ], batch size: 336, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:42:53,635 INFO [optim.py:369] (1/2) 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 03:43:04,345 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-03 03:43:12,981 INFO [train.py:968] (1/2) Epoch 6, batch 12100, libri_loss[loss=0.2739, simple_loss=0.338, pruned_loss=0.1048, over 29350.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3946, pruned_loss=0.1422, over 5655748.42 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.371, pruned_loss=0.1127, over 5735849.73 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3986, pruned_loss=0.1464, over 5644059.93 frames. ], batch size: 67, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:43:45,607 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239847.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:43:50,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239850.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 03:43:59,972 INFO [train.py:968] (1/2) Epoch 6, batch 12150, giga_loss[loss=0.3393, simple_loss=0.3962, pruned_loss=0.1412, over 28988.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3928, pruned_loss=0.141, over 5671400.42 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3708, pruned_loss=0.1124, over 5738242.33 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3969, pruned_loss=0.1453, over 5658121.74 frames. ], batch size: 155, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:44:17,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239879.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 03:44:24,744 INFO [optim.py:369] (1/2) 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 03:44:25,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8198, 2.0024, 1.4578, 1.2817], device='cuda:1'), covar=tensor([0.1068, 0.0970, 0.0873, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.1490, 0.1332, 0.1275, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 03:44:46,644 INFO [train.py:968] (1/2) Epoch 6, batch 12200, giga_loss[loss=0.393, simple_loss=0.4223, pruned_loss=0.1818, over 27596.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3924, pruned_loss=0.1411, over 5671467.61 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3709, pruned_loss=0.1125, over 5740972.88 frames. ], giga_tot_loss[loss=0.3436, simple_loss=0.3962, pruned_loss=0.1455, over 5655986.41 frames. ], batch size: 472, lr: 5.47e-03, grad_scale: 4.0
+2023-03-03 03:45:32,556 INFO [train.py:968] (1/2) Epoch 6, batch 12250, giga_loss[loss=0.3623, simple_loss=0.423, pruned_loss=0.1508, over 29105.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3946, pruned_loss=0.1428, over 5678133.57 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3709, pruned_loss=0.1125, over 5745670.97 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.3985, pruned_loss=0.1474, over 5659216.19 frames. ], batch size: 164, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:45:39,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 03:45:58,116 INFO [optim.py:369] (1/2) 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 03:46:18,418 INFO [train.py:968] (1/2) Epoch 6, batch 12300, giga_loss[loss=0.353, simple_loss=0.4031, pruned_loss=0.1514, over 28862.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3959, pruned_loss=0.1441, over 5658496.36 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3712, pruned_loss=0.1129, over 5728954.01 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.3992, pruned_loss=0.148, over 5656522.53 frames. ], batch size: 145, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:46:21,822 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=240015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:47:05,904 INFO [train.py:968] (1/2) Epoch 6, batch 12350, giga_loss[loss=0.3048, simple_loss=0.3709, pruned_loss=0.1193, over 28961.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3951, pruned_loss=0.1432, over 5668854.20 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3712, pruned_loss=0.1128, over 5731550.62 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.3985, pruned_loss=0.1473, over 5662911.68 frames. ], batch size: 227, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:47:35,233 INFO [optim.py:369] (1/2) 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 03:47:54,184 INFO [train.py:968] (1/2) Epoch 6, batch 12400, giga_loss[loss=0.3289, simple_loss=0.3894, pruned_loss=0.1342, over 28991.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3949, pruned_loss=0.1423, over 5663589.76 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3714, pruned_loss=0.1128, over 5734917.61 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3978, pruned_loss=0.1461, over 5654599.05 frames. ], batch size: 145, lr: 5.46e-03, grad_scale: 8.0
+2023-03-03 03:48:42,159 INFO [train.py:968] (1/2) Epoch 6, batch 12450, giga_loss[loss=0.3638, simple_loss=0.3922, pruned_loss=0.1678, over 23687.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3953, pruned_loss=0.1419, over 5661740.38 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3719, pruned_loss=0.1131, over 5725833.44 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3977, pruned_loss=0.1453, over 5662001.67 frames. ], batch size: 705, lr: 5.46e-03, grad_scale: 8.0
+2023-03-03 03:49:10,142 INFO [optim.py:369] (1/2) 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 03:49:12,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5235, 1.6713, 1.4059, 2.0936], device='cuda:1'), covar=tensor([0.2212, 0.2294, 0.2281, 0.1983], device='cuda:1'), in_proj_covar=tensor([0.1157, 0.0901, 0.1025, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 03:49:34,707 INFO [train.py:968] (1/2) Epoch 6, batch 12500, giga_loss[loss=0.3289, simple_loss=0.3821, pruned_loss=0.1379, over 28886.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3939, pruned_loss=0.1405, over 5675573.66 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.372, pruned_loss=0.1131, over 5726755.27 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3958, pruned_loss=0.1432, over 5674734.59 frames. ], batch size: 186, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:50:20,277 INFO [train.py:968] (1/2) Epoch 6, batch 12550, giga_loss[loss=0.2945, simple_loss=0.3566, pruned_loss=0.1162, over 28104.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3922, pruned_loss=0.1397, over 5670510.12 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3717, pruned_loss=0.1128, over 5731233.63 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3946, pruned_loss=0.1431, over 5664056.74 frames. ], batch size: 77, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:50:52,107 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 12600, giga_loss[loss=0.3122, simple_loss=0.3664, pruned_loss=0.1289, over 28475.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3905, pruned_loss=0.1388, over 5672137.62 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3723, pruned_loss=0.113, over 5734001.45 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3923, pruned_loss=0.1418, over 5663174.25 frames. ], batch size: 71, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:51:56,096 INFO [train.py:968] (1/2) Epoch 6, batch 12650, giga_loss[loss=0.3753, simple_loss=0.4115, pruned_loss=0.1695, over 27893.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3879, pruned_loss=0.1379, over 5681578.17 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3726, pruned_loss=0.1131, over 5738365.51 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3895, pruned_loss=0.1409, over 5668994.29 frames. ], batch size: 412, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:52:26,472 INFO [optim.py:369] (1/2) 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 03:52:26,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:52:39,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6950, 4.4958, 4.2566, 2.1564], device='cuda:1'), covar=tensor([0.0528, 0.0759, 0.0981, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0882, 0.0835, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-03 03:52:42,957 INFO [train.py:968] (1/2) Epoch 6, batch 12700, giga_loss[loss=0.2995, simple_loss=0.3608, pruned_loss=0.1191, over 28838.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.385, pruned_loss=0.1367, over 5686294.75 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3727, pruned_loss=0.1134, over 5738465.10 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3866, pruned_loss=0.1393, over 5674976.90 frames. ], batch size: 199, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:52:43,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-03 03:53:30,446 INFO [train.py:968] (1/2) Epoch 6, batch 12750, giga_loss[loss=0.3339, simple_loss=0.3716, pruned_loss=0.1481, over 28407.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3846, pruned_loss=0.1365, over 5691504.14 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3727, pruned_loss=0.1134, over 5739197.41 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3861, pruned_loss=0.1393, over 5680646.78 frames. ], batch size: 65, lr: 5.46e-03, grad_scale: 2.0
+2023-03-03 03:54:00,601 INFO [optim.py:369] (1/2) 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 03:54:21,865 INFO [train.py:968] (1/2) Epoch 6, batch 12800, libri_loss[loss=0.2641, simple_loss=0.3444, pruned_loss=0.09189, over 29526.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3848, pruned_loss=0.1357, over 5680321.67 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3729, pruned_loss=0.1135, over 5732317.89 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3861, pruned_loss=0.1385, over 5677339.39 frames. ], batch size: 84, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:54:43,397 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=240533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:54:45,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=240536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:55:07,761 INFO [train.py:968] (1/2) Epoch 6, batch 12850, giga_loss[loss=0.2959, simple_loss=0.368, pruned_loss=0.112, over 28613.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3827, pruned_loss=0.1326, over 5685340.31 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3723, pruned_loss=0.1134, over 5738777.53 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3848, pruned_loss=0.1356, over 5675216.23 frames. ], batch size: 262, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:55:11,866 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=240565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 03:55:17,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.10 vs. limit=5.0
+2023-03-03 03:55:36,389 INFO [optim.py:369] (1/2) 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 03:55:56,273 INFO [train.py:968] (1/2) Epoch 6, batch 12900, giga_loss[loss=0.3407, simple_loss=0.382, pruned_loss=0.1497, over 26604.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3795, pruned_loss=0.1291, over 5681754.87 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3715, pruned_loss=0.1131, over 5744871.13 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3824, pruned_loss=0.1327, over 5665259.82 frames. ], batch size: 555, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:56:04,250 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-03 03:56:48,856 INFO [train.py:968] (1/2) Epoch 6, batch 12950, giga_loss[loss=0.2398, simple_loss=0.3001, pruned_loss=0.08972, over 23906.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3757, pruned_loss=0.1257, over 5667654.28 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3711, pruned_loss=0.1131, over 5738922.99 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3784, pruned_loss=0.1288, over 5659123.17 frames. ], batch size: 705, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:57:19,171 INFO [optim.py:369] (1/2) 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 03:57:43,477 INFO [train.py:968] (1/2) Epoch 6, batch 13000, giga_loss[loss=0.2765, simple_loss=0.3562, pruned_loss=0.09839, over 28790.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3728, pruned_loss=0.1227, over 5665750.33 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3712, pruned_loss=0.1132, over 5740223.34 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3749, pruned_loss=0.1252, over 5657164.48 frames. ], batch size: 243, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:58:06,563 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 13050, giga_loss[loss=0.2712, simple_loss=0.3359, pruned_loss=0.1033, over 26706.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3699, pruned_loss=0.1189, over 5659397.09 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3708, pruned_loss=0.113, over 5733920.74 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3719, pruned_loss=0.1211, over 5656945.16 frames. ], batch size: 555, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 03:58:40,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3580, 1.8406, 1.7646, 1.5958], device='cuda:1'), covar=tensor([0.1645, 0.1952, 0.1229, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0733, 0.0799, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 03:59:04,538 INFO [optim.py:369] (1/2) 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 03:59:25,745 INFO [train.py:968] (1/2) Epoch 6, batch 13100, giga_loss[loss=0.3448, simple_loss=0.3809, pruned_loss=0.1544, over 26681.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3689, pruned_loss=0.1171, over 5654757.17 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3705, pruned_loss=0.1129, over 5735554.71 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3708, pruned_loss=0.1191, over 5649641.83 frames. ], batch size: 555, lr: 5.46e-03, grad_scale: 4.0
+2023-03-03 04:00:10,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-03 04:00:18,611 INFO [train.py:968] (1/2) Epoch 6, batch 13150, giga_loss[loss=0.3151, simple_loss=0.3781, pruned_loss=0.126, over 28025.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3676, pruned_loss=0.1154, over 5662163.71 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3702, pruned_loss=0.1129, over 5737371.14 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3693, pruned_loss=0.1171, over 5655911.51 frames. ], batch size: 412, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:00:48,284 INFO [optim.py:369] (1/2) 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 04:00:58,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4008, 1.5667, 1.5535, 1.4804], device='cuda:1'), covar=tensor([0.1029, 0.1339, 0.1250, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0724, 0.0635, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 04:00:59,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 04:01:09,953 INFO [train.py:968] (1/2) Epoch 6, batch 13200, giga_loss[loss=0.2442, simple_loss=0.3231, pruned_loss=0.08266, over 28131.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3647, pruned_loss=0.1138, over 5659946.16 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3698, pruned_loss=0.1127, over 5736288.39 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3664, pruned_loss=0.1152, over 5655085.96 frames. ], batch size: 77, lr: 5.45e-03, grad_scale: 8.0
+2023-03-03 04:01:53,433 INFO [train.py:968] (1/2) Epoch 6, batch 13250, giga_loss[loss=0.3227, simple_loss=0.3833, pruned_loss=0.131, over 29008.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3627, pruned_loss=0.1122, over 5676438.53 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3689, pruned_loss=0.1122, over 5744574.83 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3645, pruned_loss=0.1138, over 5660768.53 frames. ], batch size: 213, lr: 5.45e-03, grad_scale: 8.0
+2023-03-03 04:02:25,999 INFO [optim.py:369] (1/2) 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 04:02:45,679 INFO [train.py:968] (1/2) Epoch 6, batch 13300, giga_loss[loss=0.2958, simple_loss=0.3694, pruned_loss=0.1111, over 28987.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3633, pruned_loss=0.1125, over 5671939.52 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3688, pruned_loss=0.1122, over 5746216.87 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3648, pruned_loss=0.1139, over 5657473.96 frames. ], batch size: 213, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:03:36,364 INFO [train.py:968] (1/2) Epoch 6, batch 13350, giga_loss[loss=0.2677, simple_loss=0.3421, pruned_loss=0.09665, over 28780.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3612, pruned_loss=0.1108, over 5666823.45 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3685, pruned_loss=0.1121, over 5745913.78 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3626, pruned_loss=0.112, over 5654374.26 frames. ], batch size: 92, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:03:39,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-03 04:04:05,874 INFO [optim.py:369] (1/2) 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 04:04:23,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:04:25,740 INFO [train.py:968] (1/2) Epoch 6, batch 13400, giga_loss[loss=0.255, simple_loss=0.343, pruned_loss=0.08352, over 28882.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3592, pruned_loss=0.1088, over 5676866.50 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3679, pruned_loss=0.1119, over 5748379.01 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3606, pruned_loss=0.1098, over 5661707.73 frames. ], batch size: 145, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:05:16,134 INFO [train.py:968] (1/2) Epoch 6, batch 13450, giga_loss[loss=0.2601, simple_loss=0.334, pruned_loss=0.09311, over 28891.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.355, pruned_loss=0.106, over 5661576.77 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3678, pruned_loss=0.1119, over 5741089.63 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.356, pruned_loss=0.1067, over 5655590.52 frames. ], batch size: 199, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:05:46,259 INFO [optim.py:369] (1/2) 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 04:06:07,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2843, 1.5533, 1.2615, 1.5114], device='cuda:1'), covar=tensor([0.2293, 0.2040, 0.2201, 0.1883], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.0871, 0.1012, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 04:06:07,337 INFO [train.py:968] (1/2) Epoch 6, batch 13500, giga_loss[loss=0.263, simple_loss=0.3328, pruned_loss=0.09664, over 28954.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3514, pruned_loss=0.1046, over 5653917.27 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3665, pruned_loss=0.1113, over 5743124.22 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3529, pruned_loss=0.1056, over 5644571.53 frames. ], batch size: 213, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:06:48,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:06:50,758 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 13550, giga_loss[loss=0.285, simple_loss=0.3532, pruned_loss=0.1084, over 28250.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3509, pruned_loss=0.1051, over 5657382.08 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3661, pruned_loss=0.111, over 5749113.22 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3519, pruned_loss=0.1059, over 5641169.02 frames. ], batch size: 368, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:07:00,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5985, 1.6250, 1.2451, 1.3817], device='cuda:1'), covar=tensor([0.0617, 0.0421, 0.0862, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0450, 0.0502, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:07:21,955 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:07:29,057 INFO [optim.py:369] (1/2) 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 04:07:30,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7616, 1.7513, 1.2315, 1.4441], device='cuda:1'), covar=tensor([0.0649, 0.0497, 0.0905, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0449, 0.0503, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:07:52,922 INFO [train.py:968] (1/2) Epoch 6, batch 13600, giga_loss[loss=0.3518, simple_loss=0.3937, pruned_loss=0.155, over 26645.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3515, pruned_loss=0.106, over 5643666.81 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3658, pruned_loss=0.111, over 5751705.59 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3523, pruned_loss=0.1066, over 5627232.21 frames. ], batch size: 555, lr: 5.45e-03, grad_scale: 8.0
+2023-03-03 04:08:08,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241325.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:08:48,335 INFO [train.py:968] (1/2) Epoch 6, batch 13650, giga_loss[loss=0.2739, simple_loss=0.3548, pruned_loss=0.09656, over 29086.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3527, pruned_loss=0.1055, over 5654274.04 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3644, pruned_loss=0.1103, over 5756628.12 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3543, pruned_loss=0.1065, over 5633467.77 frames. ], batch size: 128, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:09:11,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:09:28,627 INFO [optim.py:369] (1/2) 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 04:09:50,762 INFO [train.py:968] (1/2) Epoch 6, batch 13700, giga_loss[loss=0.2478, simple_loss=0.3304, pruned_loss=0.08258, over 28962.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3535, pruned_loss=0.1051, over 5654632.87 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3643, pruned_loss=0.1102, over 5758081.05 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3547, pruned_loss=0.1059, over 5636115.79 frames. ], batch size: 227, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:10:48,882 INFO [train.py:968] (1/2) Epoch 6, batch 13750, giga_loss[loss=0.2947, simple_loss=0.3617, pruned_loss=0.1139, over 28727.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3536, pruned_loss=0.1057, over 5640963.83 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3637, pruned_loss=0.11, over 5751685.49 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3549, pruned_loss=0.1063, over 5629060.77 frames. ], batch size: 243, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:11:28,176 INFO [optim.py:369] (1/2) 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 04:11:49,635 INFO [train.py:968] (1/2) Epoch 6, batch 13800, giga_loss[loss=0.2686, simple_loss=0.3466, pruned_loss=0.09527, over 28927.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3516, pruned_loss=0.1041, over 5646585.00 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3635, pruned_loss=0.11, over 5744291.52 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3526, pruned_loss=0.1046, over 5643009.50 frames. ], batch size: 199, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:11:57,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1968, 1.7404, 1.5775, 1.3923], device='cuda:1'), covar=tensor([0.1459, 0.1903, 0.1158, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0721, 0.0791, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 04:12:50,211 INFO [train.py:968] (1/2) Epoch 6, batch 13850, giga_loss[loss=0.2697, simple_loss=0.3474, pruned_loss=0.09602, over 28992.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3499, pruned_loss=0.102, over 5641863.64 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.363, pruned_loss=0.1099, over 5747561.98 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3509, pruned_loss=0.1024, over 5634781.05 frames. ], batch size: 285, lr: 5.45e-03, grad_scale: 2.0
+2023-03-03 04:13:32,135 INFO [optim.py:369] (1/2) 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 04:13:53,811 INFO [train.py:968] (1/2) Epoch 6, batch 13900, giga_loss[loss=0.2785, simple_loss=0.346, pruned_loss=0.1054, over 29033.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3468, pruned_loss=0.1001, over 5651398.29 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3628, pruned_loss=0.1098, over 5748628.95 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3475, pruned_loss=0.1003, over 5643522.10 frames. ], batch size: 285, lr: 5.45e-03, grad_scale: 2.0
+2023-03-03 04:14:12,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3999, 2.5500, 1.5188, 1.5622], device='cuda:1'), covar=tensor([0.0696, 0.0320, 0.0700, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0482, 0.0317, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 04:14:24,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3267, 1.8427, 1.4161, 0.3575], device='cuda:1'), covar=tensor([0.1958, 0.1542, 0.2479, 0.2877], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1323, 0.1385, 0.1172], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 04:14:52,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8735, 2.8217, 1.9444, 0.8730], device='cuda:1'), covar=tensor([0.3675, 0.1747, 0.2151, 0.3385], device='cuda:1'), in_proj_covar=tensor([0.1402, 0.1331, 0.1392, 0.1178], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 04:14:55,820 INFO [train.py:968] (1/2) Epoch 6, batch 13950, giga_loss[loss=0.3077, simple_loss=0.3702, pruned_loss=0.1226, over 28430.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.345, pruned_loss=0.1001, over 5655526.42 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3624, pruned_loss=0.1096, over 5751030.97 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3457, pruned_loss=0.1003, over 5645917.36 frames. ], batch size: 336, lr: 5.45e-03, grad_scale: 2.0
+2023-03-03 04:15:00,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-03 04:15:32,246 INFO [optim.py:369] (1/2) 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 04:15:42,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241700.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:15:54,469 INFO [train.py:968] (1/2) Epoch 6, batch 14000, giga_loss[loss=0.2767, simple_loss=0.3556, pruned_loss=0.09887, over 28947.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3448, pruned_loss=0.1003, over 5660306.93 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3624, pruned_loss=0.1097, over 5750597.08 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3451, pruned_loss=0.1003, over 5651424.72 frames. ], batch size: 284, lr: 5.45e-03, grad_scale: 4.0
+2023-03-03 04:16:47,296 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 14050, giga_loss[loss=0.2425, simple_loss=0.3381, pruned_loss=0.07347, over 28850.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3476, pruned_loss=0.1013, over 5667278.65 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3622, pruned_loss=0.1097, over 5752498.81 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3476, pruned_loss=0.101, over 5656606.48 frames. ], batch size: 174, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:17:28,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2992, 4.1604, 3.9155, 1.7526], device='cuda:1'), covar=tensor([0.0428, 0.0502, 0.0692, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0819, 0.0779, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:17:36,924 INFO [optim.py:369] (1/2) 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 04:17:45,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=241799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:17:58,990 INFO [train.py:968] (1/2) Epoch 6, batch 14100, giga_loss[loss=0.261, simple_loss=0.3293, pruned_loss=0.09632, over 26769.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3494, pruned_loss=0.1013, over 5679101.54 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.362, pruned_loss=0.1098, over 5754972.54 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3494, pruned_loss=0.101, over 5667227.50 frames. ], batch size: 555, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:18:21,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4803, 1.9433, 1.3436, 1.2806], device='cuda:1'), covar=tensor([0.1289, 0.0855, 0.0919, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.1467, 0.1264, 0.1235, 0.1317], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 04:18:41,866 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241843.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:18:46,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241846.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:19:05,381 INFO [train.py:968] (1/2) Epoch 6, batch 14150, giga_loss[loss=0.3256, simple_loss=0.3804, pruned_loss=0.1354, over 28131.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3457, pruned_loss=0.09912, over 5679332.56 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3619, pruned_loss=0.1097, over 5754560.78 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09858, over 5667774.78 frames. ], batch size: 412, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:19:24,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241875.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:19:31,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 04:19:37,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-03 04:19:45,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6135, 3.4502, 3.2894, 1.5717], device='cuda:1'), covar=tensor([0.0580, 0.0624, 0.0737, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0813, 0.0778, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:19:46,282 INFO [optim.py:369] (1/2) 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 04:19:52,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:19:57,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:20:11,101 INFO [train.py:968] (1/2) Epoch 6, batch 14200, giga_loss[loss=0.2737, simple_loss=0.3532, pruned_loss=0.09706, over 28473.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3473, pruned_loss=0.1005, over 5676707.56 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3618, pruned_loss=0.1097, over 5756083.25 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3469, pruned_loss=0.09986, over 5664906.35 frames. ], batch size: 336, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:20:36,776 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:20:42,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5661, 1.6972, 1.5464, 1.6770], device='cuda:1'), covar=tensor([0.1046, 0.1710, 0.1512, 0.1363], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0717, 0.0632, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 04:21:18,681 INFO [train.py:968] (1/2) Epoch 6, batch 14250, giga_loss[loss=0.3122, simple_loss=0.3884, pruned_loss=0.118, over 28838.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.35, pruned_loss=0.1007, over 5661099.82 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3612, pruned_loss=0.1093, over 5753057.28 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3499, pruned_loss=0.1005, over 5652612.04 frames. ], batch size: 213, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:21:58,392 INFO [optim.py:369] (1/2) 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 04:22:19,400 INFO [train.py:968] (1/2) Epoch 6, batch 14300, giga_loss[loss=0.2608, simple_loss=0.3453, pruned_loss=0.08816, over 28472.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3519, pruned_loss=0.09924, over 5654269.67 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.361, pruned_loss=0.1091, over 5745110.01 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3518, pruned_loss=0.09902, over 5652386.49 frames. ], batch size: 336, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:22:24,757 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:23:19,715 INFO [train.py:968] (1/2) Epoch 6, batch 14350, giga_loss[loss=0.2647, simple_loss=0.3495, pruned_loss=0.08998, over 28975.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3516, pruned_loss=0.09814, over 5649378.68 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3604, pruned_loss=0.1088, over 5749087.57 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3519, pruned_loss=0.09804, over 5642019.48 frames. ], batch size: 186, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:23:20,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5234, 3.3741, 3.1741, 1.8225], device='cuda:1'), covar=tensor([0.0530, 0.0640, 0.0791, 0.1938], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0812, 0.0774, 0.0598], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:23:56,336 INFO [optim.py:369] (1/2) 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 04:24:18,848 INFO [train.py:968] (1/2) Epoch 6, batch 14400, giga_loss[loss=0.3023, simple_loss=0.3734, pruned_loss=0.1157, over 28131.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3519, pruned_loss=0.09808, over 5660998.65 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3602, pruned_loss=0.1087, over 5751879.98 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3522, pruned_loss=0.09777, over 5650308.97 frames. ], batch size: 412, lr: 5.44e-03, grad_scale: 8.0
+2023-03-03 04:25:23,333 INFO [train.py:968] (1/2) Epoch 6, batch 14450, giga_loss[loss=0.3075, simple_loss=0.355, pruned_loss=0.13, over 24683.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3531, pruned_loss=0.1003, over 5665066.36 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3597, pruned_loss=0.1085, over 5751373.78 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3537, pruned_loss=0.1001, over 5656242.55 frames. ], batch size: 705, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:25:37,893 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:26:05,500 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:26:08,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0371, 1.1927, 1.3405, 1.0277], device='cuda:1'), covar=tensor([0.1013, 0.1018, 0.1500, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0722, 0.0633, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 04:26:26,265 INFO [train.py:968] (1/2) Epoch 6, batch 14500, giga_loss[loss=0.3092, simple_loss=0.3815, pruned_loss=0.1184, over 28414.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3527, pruned_loss=0.1013, over 5663538.07 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3591, pruned_loss=0.1082, over 5748614.42 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3535, pruned_loss=0.1013, over 5656306.12 frames. ], batch size: 336, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:26:44,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242224.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 04:27:40,184 INFO [train.py:968] (1/2) Epoch 6, batch 14550, giga_loss[loss=0.2293, simple_loss=0.3177, pruned_loss=0.07039, over 28897.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3531, pruned_loss=0.1018, over 5670838.30 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3591, pruned_loss=0.1083, over 5751723.84 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3537, pruned_loss=0.1016, over 5660743.81 frames. ], batch size: 174, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:28:34,907 INFO [optim.py:369] (1/2) 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 04:28:59,254 INFO [train.py:968] (1/2) Epoch 6, batch 14600, giga_loss[loss=0.2478, simple_loss=0.3276, pruned_loss=0.08393, over 28140.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3474, pruned_loss=0.09817, over 5659740.37 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3591, pruned_loss=0.1083, over 5736514.62 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3476, pruned_loss=0.09778, over 5661956.51 frames. ], batch size: 412, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:29:08,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:29:10,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:29:47,688 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:30:01,568 INFO [train.py:968] (1/2) Epoch 6, batch 14650, giga_loss[loss=0.2671, simple_loss=0.3489, pruned_loss=0.09267, over 28771.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3462, pruned_loss=0.09771, over 5659235.11 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3588, pruned_loss=0.1081, over 5739473.77 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09721, over 5655570.42 frames. ], batch size: 243, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:30:37,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:30:44,243 INFO [optim.py:369] (1/2) 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 04:30:51,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-03 04:31:08,251 INFO [train.py:968] (1/2) Epoch 6, batch 14700, giga_loss[loss=0.2959, simple_loss=0.3578, pruned_loss=0.117, over 26749.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3444, pruned_loss=0.09719, over 5669993.38 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3581, pruned_loss=0.1077, over 5739608.79 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3446, pruned_loss=0.09685, over 5665163.20 frames. ], batch size: 555, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:32:12,915 INFO [train.py:968] (1/2) Epoch 6, batch 14750, giga_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1144, over 27695.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3501, pruned_loss=0.1, over 5677642.57 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3583, pruned_loss=0.1077, over 5741821.89 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.35, pruned_loss=0.09958, over 5671062.90 frames. ], batch size: 472, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:32:55,156 INFO [optim.py:369] (1/2) 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 04:33:16,367 INFO [train.py:968] (1/2) Epoch 6, batch 14800, giga_loss[loss=0.2836, simple_loss=0.3465, pruned_loss=0.1103, over 28024.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.09984, over 5675529.48 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.358, pruned_loss=0.1076, over 5741022.18 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3488, pruned_loss=0.09951, over 5670233.81 frames. ], batch size: 412, lr: 5.44e-03, grad_scale: 8.0
+2023-03-03 04:33:27,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 04:33:28,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-03 04:33:44,720 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:01,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-03 04:34:02,638 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:23,939 INFO [train.py:968] (1/2) Epoch 6, batch 14850, giga_loss[loss=0.2719, simple_loss=0.3473, pruned_loss=0.09826, over 28565.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3489, pruned_loss=0.1011, over 5668299.94 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3577, pruned_loss=0.1075, over 5742671.93 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3491, pruned_loss=0.1009, over 5662080.73 frames. ], batch size: 336, lr: 5.44e-03, grad_scale: 8.0
+2023-03-03 04:34:27,426 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:33,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:34:52,774 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-03 04:34:58,984 INFO [optim.py:369] (1/2) 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 04:35:04,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242599.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:35:18,347 INFO [train.py:968] (1/2) Epoch 6, batch 14900, libri_loss[loss=0.2867, simple_loss=0.3545, pruned_loss=0.1095, over 29164.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3488, pruned_loss=0.1014, over 5671664.96 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3573, pruned_loss=0.1074, over 5742289.04 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.349, pruned_loss=0.1011, over 5663989.88 frames. ], batch size: 97, lr: 5.44e-03, grad_scale: 4.0
+2023-03-03 04:35:27,697 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:35:27,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4014, 1.8999, 1.7300, 1.5708], device='cuda:1'), covar=tensor([0.1517, 0.1901, 0.1167, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0714, 0.0789, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 04:35:40,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0927, 1.2560, 3.4081, 2.9814], device='cuda:1'), covar=tensor([0.1501, 0.2206, 0.0479, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0581, 0.0541, 0.0756, 0.0622], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 04:36:20,170 INFO [train.py:968] (1/2) Epoch 6, batch 14950, giga_loss[loss=0.258, simple_loss=0.3479, pruned_loss=0.08403, over 28907.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3508, pruned_loss=0.102, over 5670133.92 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.357, pruned_loss=0.1075, over 5736770.75 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3511, pruned_loss=0.1014, over 5666862.76 frames. ], batch size: 284, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:37:11,386 INFO [optim.py:369] (1/2) 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 04:37:23,984 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 15000, giga_loss[loss=0.2648, simple_loss=0.3448, pruned_loss=0.09235, over 28981.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.353, pruned_loss=0.1029, over 5672930.48 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3568, pruned_loss=0.1076, over 5741545.25 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1022, over 5663878.07 frames. ], batch size: 199, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:37:32,478 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 04:37:41,145 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 04:37:43,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:37:48,052 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:38:27,232 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:38:35,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242745.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:38:37,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:38:57,801 INFO [train.py:968] (1/2) Epoch 6, batch 15050, giga_loss[loss=0.2476, simple_loss=0.3211, pruned_loss=0.0871, over 29050.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3504, pruned_loss=0.1014, over 5677401.21 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3567, pruned_loss=0.1075, over 5744769.35 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3506, pruned_loss=0.1009, over 5665857.24 frames. ], batch size: 285, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:39:00,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6044, 1.1946, 5.2405, 3.8028], device='cuda:1'), covar=tensor([0.1509, 0.2463, 0.0308, 0.0538], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0539, 0.0749, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 04:39:22,472 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242774.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 04:39:32,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7371, 1.8459, 1.8811, 1.7286], device='cuda:1'), covar=tensor([0.0993, 0.1421, 0.1232, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0717, 0.0625, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 04:39:48,743 INFO [optim.py:369] (1/2) 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 04:40:08,882 INFO [train.py:968] (1/2) Epoch 6, batch 15100, giga_loss[loss=0.2451, simple_loss=0.318, pruned_loss=0.08609, over 28951.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3456, pruned_loss=0.09974, over 5682526.85 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3566, pruned_loss=0.1074, over 5738059.34 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3458, pruned_loss=0.0993, over 5679264.15 frames. ], batch size: 213, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:41:03,681 INFO [train.py:968] (1/2) Epoch 6, batch 15150, giga_loss[loss=0.2645, simple_loss=0.335, pruned_loss=0.097, over 28987.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3419, pruned_loss=0.09801, over 5682278.16 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3573, pruned_loss=0.1079, over 5737934.27 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3406, pruned_loss=0.0967, over 5675705.98 frames. ], batch size: 213, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:41:25,324 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:41:47,650 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 15200, giga_loss[loss=0.245, simple_loss=0.3224, pruned_loss=0.08377, over 28927.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3431, pruned_loss=0.09937, over 5677228.51 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3571, pruned_loss=0.1079, over 5737805.74 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3419, pruned_loss=0.09822, over 5670644.92 frames. ], batch size: 145, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:42:13,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3328, 1.7537, 1.6685, 1.5355], device='cuda:1'), covar=tensor([0.1286, 0.1698, 0.1040, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0710, 0.0789, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 04:42:18,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:42:48,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:42:57,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6676, 3.5007, 3.3166, 1.8208], device='cuda:1'), covar=tensor([0.0578, 0.0745, 0.0839, 0.2179], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0819, 0.0762, 0.0600], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:43:03,295 INFO [train.py:968] (1/2) Epoch 6, batch 15250, giga_loss[loss=0.2525, simple_loss=0.3318, pruned_loss=0.08657, over 28674.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3437, pruned_loss=0.09977, over 5673757.67 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.357, pruned_loss=0.1078, over 5738022.13 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3427, pruned_loss=0.09883, over 5667597.33 frames. ], batch size: 307, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:43:07,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5231, 4.3616, 4.1266, 2.0782], device='cuda:1'), covar=tensor([0.0366, 0.0522, 0.0716, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0816, 0.0759, 0.0598], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:43:44,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-03 04:43:45,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:43:46,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=242995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:43:47,252 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 15300, giga_loss[loss=0.2905, simple_loss=0.3666, pruned_loss=0.1072, over 28505.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3415, pruned_loss=0.09809, over 5670172.20 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3568, pruned_loss=0.1077, over 5742671.66 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3405, pruned_loss=0.09709, over 5658758.51 frames. ], batch size: 370, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:45:08,035 INFO [train.py:968] (1/2) Epoch 6, batch 15350, giga_loss[loss=0.2306, simple_loss=0.3146, pruned_loss=0.07324, over 28917.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3403, pruned_loss=0.09668, over 5672615.72 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3571, pruned_loss=0.108, over 5744414.30 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3391, pruned_loss=0.09551, over 5661486.42 frames. ], batch size: 145, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:45:12,652 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:45:18,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:45:34,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:45:59,974 INFO [optim.py:369] (1/2) 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 04:46:00,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:46:20,412 INFO [train.py:968] (1/2) Epoch 6, batch 15400, giga_loss[loss=0.2438, simple_loss=0.324, pruned_loss=0.08181, over 28932.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3391, pruned_loss=0.09641, over 5671504.14 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.357, pruned_loss=0.108, over 5744414.45 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3379, pruned_loss=0.09533, over 5661117.21 frames. ], batch size: 145, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:46:41,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-03 04:46:58,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:47:00,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:47:24,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4661, 1.6885, 1.4794, 1.5231], device='cuda:1'), covar=tensor([0.1273, 0.1757, 0.1677, 0.1577], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0715, 0.0629, 0.0628], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 04:47:28,668 INFO [train.py:968] (1/2) Epoch 6, batch 15450, giga_loss[loss=0.2624, simple_loss=0.3423, pruned_loss=0.09129, over 28662.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3406, pruned_loss=0.09667, over 5678163.62 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3567, pruned_loss=0.1079, over 5737864.23 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3396, pruned_loss=0.09562, over 5674266.50 frames. ], batch size: 307, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:47:33,295 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:47:38,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:48:16,324 INFO [optim.py:369] (1/2) 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 04:48:31,821 INFO [train.py:968] (1/2) Epoch 6, batch 15500, giga_loss[loss=0.2343, simple_loss=0.3039, pruned_loss=0.08234, over 28703.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3412, pruned_loss=0.09727, over 5687125.50 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3566, pruned_loss=0.1079, over 5738841.44 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3402, pruned_loss=0.09627, over 5682426.79 frames. ], batch size: 92, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:48:41,820 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:48:46,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:49:28,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:49:29,640 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 15550, giga_loss[loss=0.2413, simple_loss=0.3172, pruned_loss=0.08266, over 28156.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3424, pruned_loss=0.09874, over 5685043.03 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3565, pruned_loss=0.1079, over 5738423.66 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3414, pruned_loss=0.09783, over 5681073.43 frames. ], batch size: 412, lr: 5.43e-03, grad_scale: 2.0
+2023-03-03 04:50:20,883 INFO [optim.py:369] (1/2) 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 04:50:35,924 INFO [train.py:968] (1/2) Epoch 6, batch 15600, giga_loss[loss=0.2635, simple_loss=0.3563, pruned_loss=0.08533, over 28936.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3411, pruned_loss=0.09703, over 5681827.40 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3561, pruned_loss=0.1075, over 5741313.69 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3404, pruned_loss=0.09637, over 5674549.48 frames. ], batch size: 145, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:50:39,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:50:50,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:51:14,065 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 04:51:15,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2699, 1.1785, 4.2411, 3.2243], device='cuda:1'), covar=tensor([0.1476, 0.2290, 0.0361, 0.0718], device='cuda:1'), in_proj_covar=tensor([0.0568, 0.0532, 0.0744, 0.0607], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 04:51:33,065 INFO [train.py:968] (1/2) Epoch 6, batch 15650, giga_loss[loss=0.2604, simple_loss=0.3171, pruned_loss=0.1018, over 24721.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.343, pruned_loss=0.09707, over 5664359.78 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3557, pruned_loss=0.1073, over 5742296.80 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3424, pruned_loss=0.09652, over 5655679.28 frames. ], batch size: 705, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:51:43,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:52:11,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3298, 4.1594, 3.9357, 1.7579], device='cuda:1'), covar=tensor([0.0464, 0.0569, 0.0794, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0812, 0.0763, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:52:12,654 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:52:14,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:52:27,397 INFO [train.py:968] (1/2) Epoch 6, batch 15700, giga_loss[loss=0.2402, simple_loss=0.332, pruned_loss=0.07422, over 29006.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3456, pruned_loss=0.09851, over 5673535.20 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3549, pruned_loss=0.1067, over 5748752.55 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3453, pruned_loss=0.09809, over 5657163.10 frames. ], batch size: 175, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:52:44,537 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:52:59,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1434, 1.3290, 3.2775, 2.9662], device='cuda:1'), covar=tensor([0.1387, 0.2220, 0.0392, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0576, 0.0539, 0.0754, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 04:53:23,213 INFO [train.py:968] (1/2) Epoch 6, batch 15750, giga_loss[loss=0.2613, simple_loss=0.3354, pruned_loss=0.09361, over 28080.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3454, pruned_loss=0.09792, over 5675244.50 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3537, pruned_loss=0.106, over 5754950.95 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.09793, over 5653909.35 frames. ], batch size: 412, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:53:29,615 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:53:32,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:54:03,686 INFO [optim.py:369] (1/2) 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 04:54:04,477 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 15800, libri_loss[loss=0.275, simple_loss=0.3429, pruned_loss=0.1035, over 29568.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3444, pruned_loss=0.09773, over 5654036.61 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3532, pruned_loss=0.1058, over 5740779.91 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.345, pruned_loss=0.09762, over 5645281.13 frames. ], batch size: 78, lr: 5.43e-03, grad_scale: 4.0
+2023-03-03 04:54:23,304 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:54:27,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:54:55,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:55:22,165 INFO [train.py:968] (1/2) Epoch 6, batch 15850, giga_loss[loss=0.2218, simple_loss=0.3037, pruned_loss=0.0699, over 27680.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3411, pruned_loss=0.09579, over 5655563.59 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3528, pruned_loss=0.1057, over 5742713.26 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3417, pruned_loss=0.09573, over 5646209.43 frames. ], batch size: 472, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 04:55:27,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8651, 2.6334, 2.9129, 2.3302], device='cuda:1'), covar=tensor([0.0854, 0.1612, 0.1037, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0717, 0.0631, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 04:55:38,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6511, 1.6636, 1.3394, 1.3617], device='cuda:1'), covar=tensor([0.0616, 0.0426, 0.0850, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0441, 0.0505, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:56:05,755 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 15900, giga_loss[loss=0.2617, simple_loss=0.3338, pruned_loss=0.0948, over 28943.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09584, over 5663523.43 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3528, pruned_loss=0.1055, over 5747111.04 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3416, pruned_loss=0.09567, over 5649831.68 frames. ], batch size: 112, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 04:57:23,530 INFO [train.py:968] (1/2) Epoch 6, batch 15950, giga_loss[loss=0.2536, simple_loss=0.333, pruned_loss=0.08711, over 28826.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3398, pruned_loss=0.09559, over 5673408.46 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3527, pruned_loss=0.1054, over 5750226.88 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3399, pruned_loss=0.09535, over 5658280.43 frames. ], batch size: 164, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 04:57:51,710 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:57:54,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=243686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:57:54,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:57:58,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=243689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:58:05,781 INFO [optim.py:369] (1/2) 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 04:58:22,624 INFO [train.py:968] (1/2) Epoch 6, batch 16000, giga_loss[loss=0.266, simple_loss=0.3488, pruned_loss=0.09161, over 28714.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09583, over 5681447.05 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3523, pruned_loss=0.105, over 5754329.76 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3413, pruned_loss=0.09568, over 5662856.60 frames. ], batch size: 307, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 04:58:27,215 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 04:58:52,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5574, 4.3935, 4.1697, 1.8247], device='cuda:1'), covar=tensor([0.0483, 0.0574, 0.0802, 0.2068], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0821, 0.0771, 0.0604], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:58:54,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3878, 1.9462, 1.3948, 0.6165], device='cuda:1'), covar=tensor([0.2399, 0.1324, 0.2217, 0.2952], device='cuda:1'), in_proj_covar=tensor([0.1427, 0.1365, 0.1397, 0.1187], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 04:59:27,952 INFO [train.py:968] (1/2) Epoch 6, batch 16050, giga_loss[loss=0.3214, simple_loss=0.3752, pruned_loss=0.1339, over 26884.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3431, pruned_loss=0.09698, over 5674870.16 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3524, pruned_loss=0.105, over 5755648.94 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3428, pruned_loss=0.09666, over 5657650.23 frames. ], batch size: 555, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 04:59:33,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2143, 3.0777, 2.8895, 1.3562], device='cuda:1'), covar=tensor([0.0806, 0.0775, 0.0915, 0.2258], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0815, 0.0767, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 04:59:54,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5414, 4.4275, 4.1487, 1.8596], device='cuda:1'), covar=tensor([0.0450, 0.0525, 0.0697, 0.2015], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0815, 0.0768, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:00:17,122 INFO [optim.py:369] (1/2) 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 05:00:34,802 INFO [train.py:968] (1/2) Epoch 6, batch 16100, giga_loss[loss=0.2723, simple_loss=0.3513, pruned_loss=0.09667, over 28966.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3441, pruned_loss=0.09804, over 5670616.48 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3523, pruned_loss=0.105, over 5756235.23 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3438, pruned_loss=0.09778, over 5655753.60 frames. ], batch size: 284, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:00:52,528 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 05:00:55,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=243832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:00:58,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=243835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:01:27,002 INFO [train.py:968] (1/2) Epoch 6, batch 16150, giga_loss[loss=0.2617, simple_loss=0.3467, pruned_loss=0.08833, over 28707.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3481, pruned_loss=0.1004, over 5666166.96 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3521, pruned_loss=0.105, over 5759719.70 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3479, pruned_loss=0.09997, over 5648453.66 frames. ], batch size: 307, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:01:31,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=243864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:02:08,272 INFO [optim.py:369] (1/2) 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 05:02:24,715 INFO [train.py:968] (1/2) Epoch 6, batch 16200, giga_loss[loss=0.2515, simple_loss=0.3367, pruned_loss=0.08312, over 28993.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3492, pruned_loss=0.1, over 5664334.11 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3517, pruned_loss=0.1048, over 5762077.40 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3494, pruned_loss=0.09987, over 5647096.36 frames. ], batch size: 155, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:03:28,992 INFO [train.py:968] (1/2) Epoch 6, batch 16250, giga_loss[loss=0.2496, simple_loss=0.324, pruned_loss=0.08756, over 29012.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 5662515.00 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3511, pruned_loss=0.1046, over 5766081.73 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3501, pruned_loss=0.1006, over 5642229.60 frames. ], batch size: 186, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:04:18,462 INFO [optim.py:369] (1/2) 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 05:04:22,764 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-03 05:04:36,954 INFO [train.py:968] (1/2) Epoch 6, batch 16300, giga_loss[loss=0.2595, simple_loss=0.334, pruned_loss=0.09248, over 28902.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3468, pruned_loss=0.09938, over 5669387.69 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3506, pruned_loss=0.1044, over 5769667.40 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3477, pruned_loss=0.09942, over 5647881.14 frames. ], batch size: 120, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:05:45,041 INFO [train.py:968] (1/2) Epoch 6, batch 16350, giga_loss[loss=0.2308, simple_loss=0.3141, pruned_loss=0.07378, over 28921.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3464, pruned_loss=0.09897, over 5676034.71 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3511, pruned_loss=0.1046, over 5771506.06 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3465, pruned_loss=0.09868, over 5656196.53 frames. ], batch size: 112, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:05:45,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=244061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:06:32,533 INFO [optim.py:369] (1/2) 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 05:06:48,649 INFO [train.py:968] (1/2) Epoch 6, batch 16400, giga_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1198, over 28934.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3462, pruned_loss=0.09963, over 5674906.78 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3509, pruned_loss=0.1046, over 5771853.34 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3464, pruned_loss=0.09941, over 5657604.58 frames. ], batch size: 213, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:07:47,612 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-03 05:07:49,076 INFO [train.py:968] (1/2) Epoch 6, batch 16450, libri_loss[loss=0.2529, simple_loss=0.3282, pruned_loss=0.08879, over 29524.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3437, pruned_loss=0.09894, over 5669210.49 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3509, pruned_loss=0.1045, over 5774771.40 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3439, pruned_loss=0.09873, over 5649984.15 frames. ], batch size: 80, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:08:31,342 INFO [optim.py:369] (1/2) 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 05:08:37,837 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=244204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:08:41,267 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 16500, giga_loss[loss=0.2034, simple_loss=0.2846, pruned_loss=0.06108, over 28562.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3424, pruned_loss=0.09815, over 5659931.33 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3512, pruned_loss=0.1048, over 5769177.71 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3419, pruned_loss=0.09751, over 5645401.30 frames. ], batch size: 71, lr: 5.42e-03, grad_scale: 8.0
+2023-03-03 05:09:15,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=244236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:09:43,655 INFO [train.py:968] (1/2) Epoch 6, batch 16550, giga_loss[loss=0.22, simple_loss=0.3141, pruned_loss=0.06296, over 28964.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3414, pruned_loss=0.09633, over 5673971.08 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3505, pruned_loss=0.1044, over 5765854.47 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3414, pruned_loss=0.09594, over 5661587.98 frames. ], batch size: 145, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:09:44,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4284, 3.8183, 1.6078, 1.3694], device='cuda:1'), covar=tensor([0.0884, 0.0250, 0.0845, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0480, 0.0316, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 05:10:26,230 INFO [optim.py:369] (1/2) 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 05:10:39,476 INFO [train.py:968] (1/2) Epoch 6, batch 16600, libri_loss[loss=0.238, simple_loss=0.3114, pruned_loss=0.0823, over 29556.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3416, pruned_loss=0.09477, over 5679312.29 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.35, pruned_loss=0.1041, over 5768579.03 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3418, pruned_loss=0.09445, over 5664315.47 frames. ], batch size: 76, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:11:37,609 INFO [train.py:968] (1/2) Epoch 6, batch 16650, giga_loss[loss=0.2621, simple_loss=0.3358, pruned_loss=0.09419, over 26646.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3428, pruned_loss=0.09335, over 5685488.33 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3498, pruned_loss=0.104, over 5766885.33 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.343, pruned_loss=0.09299, over 5673989.80 frames. ], batch size: 555, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:12:22,619 INFO [optim.py:369] (1/2) 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 05:12:35,646 INFO [train.py:968] (1/2) Epoch 6, batch 16700, giga_loss[loss=0.2429, simple_loss=0.3273, pruned_loss=0.07925, over 28000.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3433, pruned_loss=0.09379, over 5666245.67 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3497, pruned_loss=0.1041, over 5755685.34 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3433, pruned_loss=0.09317, over 5664788.22 frames. ], batch size: 412, lr: 5.42e-03, grad_scale: 4.0
+2023-03-03 05:13:34,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4137, 4.2534, 4.0311, 1.9512], device='cuda:1'), covar=tensor([0.0479, 0.0564, 0.0774, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.0862, 0.0811, 0.0757, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:13:45,111 INFO [train.py:968] (1/2) Epoch 6, batch 16750, giga_loss[loss=0.2927, simple_loss=0.353, pruned_loss=0.1162, over 26759.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3429, pruned_loss=0.09374, over 5659814.00 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3496, pruned_loss=0.104, over 5757205.61 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.343, pruned_loss=0.09327, over 5656411.54 frames. ], batch size: 555, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:14:36,528 INFO [optim.py:369] (1/2) 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 05:14:53,346 INFO [train.py:968] (1/2) Epoch 6, batch 16800, giga_loss[loss=0.2803, simple_loss=0.3574, pruned_loss=0.1016, over 28900.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3436, pruned_loss=0.09455, over 5655688.11 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3496, pruned_loss=0.1039, over 5757054.99 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3435, pruned_loss=0.09396, over 5650899.17 frames. ], batch size: 284, lr: 5.41e-03, grad_scale: 8.0
+2023-03-03 05:15:36,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-03 05:15:57,684 INFO [train.py:968] (1/2) Epoch 6, batch 16850, libri_loss[loss=0.2563, simple_loss=0.3415, pruned_loss=0.08561, over 29212.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3439, pruned_loss=0.09379, over 5654939.79 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3493, pruned_loss=0.1037, over 5748461.49 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3438, pruned_loss=0.09317, over 5654360.35 frames. ], batch size: 94, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:16:52,686 INFO [optim.py:369] (1/2) 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 05:17:09,089 INFO [train.py:968] (1/2) Epoch 6, batch 16900, giga_loss[loss=0.3363, simple_loss=0.3999, pruned_loss=0.1364, over 28433.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3449, pruned_loss=0.09461, over 5653068.47 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3489, pruned_loss=0.1034, over 5751493.18 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3452, pruned_loss=0.09414, over 5647986.87 frames. ], batch size: 336, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:18:17,207 INFO [train.py:968] (1/2) Epoch 6, batch 16950, giga_loss[loss=0.2854, simple_loss=0.364, pruned_loss=0.1034, over 28155.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3485, pruned_loss=0.09654, over 5659522.21 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3484, pruned_loss=0.1031, over 5754932.87 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3492, pruned_loss=0.09626, over 5650463.73 frames. ], batch size: 412, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:19:05,120 INFO [optim.py:369] (1/2) 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 05:19:19,710 INFO [train.py:968] (1/2) Epoch 6, batch 17000, giga_loss[loss=0.3052, simple_loss=0.3681, pruned_loss=0.1211, over 27646.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3474, pruned_loss=0.09597, over 5671465.38 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3476, pruned_loss=0.1027, over 5752587.77 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3486, pruned_loss=0.0959, over 5663845.44 frames. ], batch size: 472, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:20:09,114 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 17050, giga_loss[loss=0.2768, simple_loss=0.3364, pruned_loss=0.1086, over 24924.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3458, pruned_loss=0.0959, over 5668806.42 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3472, pruned_loss=0.1025, over 5753214.45 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3471, pruned_loss=0.09602, over 5661395.20 frames. ], batch size: 705, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:20:39,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3339, 1.6510, 1.3553, 1.6913], device='cuda:1'), covar=tensor([0.0764, 0.0282, 0.0329, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0126, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0063, 0.0045, 0.0041, 0.0069], device='cuda:1')
+2023-03-03 05:21:23,565 INFO [optim.py:369] (1/2) 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 05:21:40,580 INFO [train.py:968] (1/2) Epoch 6, batch 17100, libri_loss[loss=0.2543, simple_loss=0.3304, pruned_loss=0.08911, over 29555.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3451, pruned_loss=0.09515, over 5678465.85 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3477, pruned_loss=0.1027, over 5753498.19 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3456, pruned_loss=0.09482, over 5669340.95 frames. ], batch size: 78, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:22:44,563 INFO [train.py:968] (1/2) Epoch 6, batch 17150, giga_loss[loss=0.2554, simple_loss=0.319, pruned_loss=0.09592, over 24356.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3441, pruned_loss=0.09473, over 5666700.31 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3476, pruned_loss=0.1027, over 5748623.59 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3445, pruned_loss=0.09424, over 5661626.74 frames. ], batch size: 705, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:22:47,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7291, 4.5522, 4.3008, 1.8533], device='cuda:1'), covar=tensor([0.0367, 0.0484, 0.0654, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0803, 0.0760, 0.0597], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:23:32,070 INFO [optim.py:369] (1/2) 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 05:23:46,276 INFO [train.py:968] (1/2) Epoch 6, batch 17200, giga_loss[loss=0.2722, simple_loss=0.3493, pruned_loss=0.09755, over 28916.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3437, pruned_loss=0.09412, over 5673297.70 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3471, pruned_loss=0.1023, over 5750383.22 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3443, pruned_loss=0.09383, over 5665971.41 frames. ], batch size: 227, lr: 5.41e-03, grad_scale: 8.0
+2023-03-03 05:24:46,675 INFO [train.py:968] (1/2) Epoch 6, batch 17250, giga_loss[loss=0.2754, simple_loss=0.3567, pruned_loss=0.0971, over 28889.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3468, pruned_loss=0.09599, over 5674519.81 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3468, pruned_loss=0.1022, over 5753472.87 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3476, pruned_loss=0.0958, over 5664273.42 frames. ], batch size: 284, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:25:27,910 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 17300, giga_loss[loss=0.2619, simple_loss=0.3407, pruned_loss=0.09154, over 28873.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3463, pruned_loss=0.09615, over 5681212.88 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3465, pruned_loss=0.102, over 5755999.43 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3473, pruned_loss=0.09589, over 5667318.00 frames. ], batch size: 106, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:26:35,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8007, 3.6460, 3.4354, 1.6749], device='cuda:1'), covar=tensor([0.0576, 0.0599, 0.0744, 0.2026], device='cuda:1'), in_proj_covar=tensor([0.0870, 0.0799, 0.0753, 0.0588], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:26:36,143 INFO [train.py:968] (1/2) Epoch 6, batch 17350, giga_loss[loss=0.2863, simple_loss=0.3586, pruned_loss=0.107, over 28792.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3438, pruned_loss=0.09622, over 5673263.03 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3467, pruned_loss=0.1021, over 5756416.47 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3443, pruned_loss=0.09586, over 5660909.09 frames. ], batch size: 263, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:26:46,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5543, 2.0905, 1.6715, 1.7680], device='cuda:1'), covar=tensor([0.0761, 0.0244, 0.0315, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0127, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0063, 0.0045, 0.0041, 0.0069], device='cuda:1')
+2023-03-03 05:27:24,984 INFO [optim.py:369] (1/2) 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 05:27:36,427 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245110.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:27:36,838 INFO [train.py:968] (1/2) Epoch 6, batch 17400, giga_loss[loss=0.3086, simple_loss=0.3661, pruned_loss=0.1255, over 28742.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3443, pruned_loss=0.09748, over 5661680.39 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3471, pruned_loss=0.1024, over 5754797.43 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3443, pruned_loss=0.09683, over 5652370.71 frames. ], batch size: 92, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:27:38,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-03 05:27:50,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:27:50,482 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-03 05:28:07,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3063, 1.6252, 1.1850, 1.0832], device='cuda:1'), covar=tensor([0.1236, 0.0994, 0.0846, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.1463, 0.1270, 0.1221, 0.1321], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 05:28:31,732 INFO [train.py:968] (1/2) Epoch 6, batch 17450, giga_loss[loss=0.34, simple_loss=0.4051, pruned_loss=0.1374, over 28936.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3505, pruned_loss=0.1019, over 5661673.65 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3468, pruned_loss=0.1022, over 5757375.61 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3508, pruned_loss=0.1016, over 5650064.53 frames. ], batch size: 112, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:29:00,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 05:29:10,176 INFO [optim.py:369] (1/2) 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 05:29:17,788 INFO [train.py:968] (1/2) Epoch 6, batch 17500, giga_loss[loss=0.3354, simple_loss=0.4064, pruned_loss=0.1322, over 29109.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3603, pruned_loss=0.1083, over 5655286.02 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3466, pruned_loss=0.1021, over 5739915.01 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3612, pruned_loss=0.1082, over 5657100.78 frames. ], batch size: 155, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:29:23,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2532, 1.4077, 1.1242, 1.6397], device='cuda:1'), covar=tensor([0.2349, 0.2113, 0.2391, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.1125, 0.0856, 0.1001, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 05:29:58,118 INFO [train.py:968] (1/2) Epoch 6, batch 17550, giga_loss[loss=0.2849, simple_loss=0.3621, pruned_loss=0.1038, over 28855.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3658, pruned_loss=0.1118, over 5660988.37 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3466, pruned_loss=0.1021, over 5735113.29 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3671, pruned_loss=0.112, over 5664215.51 frames. ], batch size: 285, lr: 5.41e-03, grad_scale: 4.0
+2023-03-03 05:30:01,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:30:02,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.89 vs. limit=5.0
+2023-03-03 05:30:03,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:30:16,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:30:28,778 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 6, batch 17600, giga_loss[loss=0.2557, simple_loss=0.3326, pruned_loss=0.08937, over 28716.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3639, pruned_loss=0.1118, over 5667599.85 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3463, pruned_loss=0.102, over 5741146.67 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3657, pruned_loss=0.1124, over 5662761.58 frames. ], batch size: 284, lr: 5.41e-03, grad_scale: 8.0
+2023-03-03 05:30:44,343 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-03 05:31:14,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:31:26,870 INFO [train.py:968] (1/2) Epoch 6, batch 17650, giga_loss[loss=0.2632, simple_loss=0.3326, pruned_loss=0.09687, over 28303.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3574, pruned_loss=0.109, over 5678159.98 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3466, pruned_loss=0.1021, over 5743887.68 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3588, pruned_loss=0.1096, over 5671065.63 frames. ], batch size: 368, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:32:04,739 INFO [optim.py:369] (1/2) 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 05:32:11,827 INFO [train.py:968] (1/2) Epoch 6, batch 17700, giga_loss[loss=0.2326, simple_loss=0.309, pruned_loss=0.0781, over 29010.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3511, pruned_loss=0.1064, over 5683992.26 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3471, pruned_loss=0.1023, over 5745391.66 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3519, pruned_loss=0.1067, over 5675505.59 frames. ], batch size: 155, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:32:19,898 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-03 05:32:56,078 INFO [train.py:968] (1/2) Epoch 6, batch 17750, giga_loss[loss=0.2302, simple_loss=0.3003, pruned_loss=0.08012, over 28768.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3425, pruned_loss=0.1024, over 5692441.89 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3469, pruned_loss=0.1021, over 5749378.37 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3433, pruned_loss=0.1028, over 5680852.14 frames. ], batch size: 243, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:33:16,495 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:33:38,282 INFO [train.py:968] (1/2) Epoch 6, batch 17800, giga_loss[loss=0.2216, simple_loss=0.3029, pruned_loss=0.07013, over 29013.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3342, pruned_loss=0.09815, over 5695891.57 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3467, pruned_loss=0.1019, over 5751751.83 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3349, pruned_loss=0.09873, over 5683430.31 frames. ], batch size: 136, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:34:02,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2897, 1.5216, 1.4295, 1.4440], device='cuda:1'), covar=tensor([0.1327, 0.1442, 0.1377, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0734, 0.0638, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 05:34:04,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-03 05:34:13,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2786, 0.9098, 0.7797, 1.4109], device='cuda:1'), covar=tensor([0.0759, 0.0362, 0.0369, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0126, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0063, 0.0045, 0.0041, 0.0069], device='cuda:1')
+2023-03-03 05:34:16,995 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-03 05:34:19,588 INFO [train.py:968] (1/2) Epoch 6, batch 17850, giga_loss[loss=0.2151, simple_loss=0.2926, pruned_loss=0.06882, over 28827.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3297, pruned_loss=0.09586, over 5691896.14 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3474, pruned_loss=0.1021, over 5745241.57 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3293, pruned_loss=0.09599, over 5685683.88 frames. ], batch size: 174, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:34:53,387 INFO [optim.py:369] (1/2) 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 05:35:01,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2825, 2.7536, 1.4231, 1.3394], device='cuda:1'), covar=tensor([0.0856, 0.0341, 0.0825, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0476, 0.0310, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 05:35:02,022 INFO [train.py:968] (1/2) Epoch 6, batch 17900, giga_loss[loss=0.2555, simple_loss=0.3247, pruned_loss=0.09311, over 28201.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3267, pruned_loss=0.09476, over 5695447.89 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3474, pruned_loss=0.1021, over 5746323.12 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3261, pruned_loss=0.0948, over 5689180.63 frames. ], batch size: 368, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:35:19,918 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245628.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:35:21,821 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245631.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:35:38,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5615, 3.0745, 2.0121, 1.6024], device='cuda:1'), covar=tensor([0.1198, 0.0690, 0.0825, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.1451, 0.1264, 0.1211, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 05:35:42,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:35:45,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:35:46,158 INFO [train.py:968] (1/2) Epoch 6, batch 17950, giga_loss[loss=0.2323, simple_loss=0.3086, pruned_loss=0.07798, over 28803.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3242, pruned_loss=0.09333, over 5695312.42 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3479, pruned_loss=0.1021, over 5746911.92 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3227, pruned_loss=0.09311, over 5688380.96 frames. ], batch size: 145, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:36:20,354 INFO [optim.py:369] (1/2) 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 05:36:27,673 INFO [train.py:968] (1/2) Epoch 6, batch 18000, giga_loss[loss=0.2354, simple_loss=0.3068, pruned_loss=0.08199, over 28746.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3209, pruned_loss=0.09158, over 5693699.70 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3479, pruned_loss=0.102, over 5750296.27 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3191, pruned_loss=0.0913, over 5683897.05 frames. ], batch size: 262, lr: 5.40e-03, grad_scale: 8.0
+2023-03-03 05:36:27,673 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 05:36:36,326 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 05:36:46,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=245722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:37:05,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2463, 1.3994, 1.1232, 0.9870], device='cuda:1'), covar=tensor([0.1116, 0.1021, 0.0679, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.1456, 0.1273, 0.1221, 0.1333], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 05:37:18,150 INFO [train.py:968] (1/2) Epoch 6, batch 18050, giga_loss[loss=0.2563, simple_loss=0.3199, pruned_loss=0.09635, over 28759.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3192, pruned_loss=0.09075, over 5705827.30 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3488, pruned_loss=0.1025, over 5755348.36 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3161, pruned_loss=0.08973, over 5691894.62 frames. ], batch size: 92, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:37:48,326 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:37:51,389 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:37:59,708 INFO [train.py:968] (1/2) Epoch 6, batch 18100, libri_loss[loss=0.309, simple_loss=0.3894, pruned_loss=0.1143, over 29209.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3179, pruned_loss=0.08999, over 5698799.21 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3501, pruned_loss=0.1031, over 5759714.42 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3129, pruned_loss=0.08811, over 5681511.07 frames. ], batch size: 97, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:38:15,177 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:38:34,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:38:38,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5433, 1.0554, 5.2435, 3.7024], device='cuda:1'), covar=tensor([0.1524, 0.2426, 0.0294, 0.0649], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0535, 0.0752, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 05:38:39,310 INFO [train.py:968] (1/2) Epoch 6, batch 18150, giga_loss[loss=0.2079, simple_loss=0.273, pruned_loss=0.07143, over 28553.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3158, pruned_loss=0.08915, over 5702346.60 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.351, pruned_loss=0.1036, over 5762032.38 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3098, pruned_loss=0.08674, over 5684266.36 frames. ], batch size: 78, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:38:42,332 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=245865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:38:45,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=245868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:39:11,666 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=245897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:39:16,981 INFO [optim.py:369] (1/2) 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 05:39:25,151 INFO [train.py:968] (1/2) Epoch 6, batch 18200, giga_loss[loss=0.231, simple_loss=0.2944, pruned_loss=0.08377, over 28508.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.312, pruned_loss=0.08683, over 5710348.33 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3512, pruned_loss=0.1036, over 5764042.17 frames. ], giga_tot_loss[loss=0.238, simple_loss=0.3066, pruned_loss=0.08471, over 5693422.61 frames. ], batch size: 85, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:39:53,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=245939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:40:01,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6072, 2.2082, 1.4654, 0.7248], device='cuda:1'), covar=tensor([0.3910, 0.2215, 0.2181, 0.3805], device='cuda:1'), in_proj_covar=tensor([0.1406, 0.1332, 0.1377, 0.1172], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 05:40:10,624 INFO [train.py:968] (1/2) Epoch 6, batch 18250, giga_loss[loss=0.2356, simple_loss=0.3039, pruned_loss=0.08361, over 28984.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3106, pruned_loss=0.0867, over 5708887.95 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3518, pruned_loss=0.1039, over 5765329.48 frames. ], giga_tot_loss[loss=0.2362, simple_loss=0.3042, pruned_loss=0.08411, over 5692021.41 frames. ], batch size: 227, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:40:49,202 INFO [optim.py:369] (1/2) 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 05:40:57,817 INFO [train.py:968] (1/2) Epoch 6, batch 18300, giga_loss[loss=0.3123, simple_loss=0.3718, pruned_loss=0.1263, over 27588.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3176, pruned_loss=0.09092, over 5710010.45 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3518, pruned_loss=0.1038, over 5768741.11 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3113, pruned_loss=0.08842, over 5691809.28 frames. ], batch size: 472, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:41:46,961 INFO [train.py:968] (1/2) Epoch 6, batch 18350, giga_loss[loss=0.3089, simple_loss=0.3825, pruned_loss=0.1177, over 28570.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3311, pruned_loss=0.09808, over 5709822.96 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3518, pruned_loss=0.1038, over 5770757.36 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3257, pruned_loss=0.09602, over 5692737.98 frames. ], batch size: 92, lr: 5.40e-03, grad_scale: 2.0
+2023-03-03 05:42:01,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:42:20,915 INFO [optim.py:369] (1/2) 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 05:42:25,471 INFO [train.py:968] (1/2) Epoch 6, batch 18400, libri_loss[loss=0.2743, simple_loss=0.3426, pruned_loss=0.103, over 29644.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3449, pruned_loss=0.1063, over 5712172.57 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3526, pruned_loss=0.1043, over 5775035.13 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3395, pruned_loss=0.1042, over 5692559.71 frames. ], batch size: 73, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:43:05,884 INFO [train.py:968] (1/2) Epoch 6, batch 18450, giga_loss[loss=0.29, simple_loss=0.3656, pruned_loss=0.1072, over 28764.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3525, pruned_loss=0.1094, over 5705131.24 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3525, pruned_loss=0.1042, over 5767742.50 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3483, pruned_loss=0.1079, over 5694851.81 frames. ], batch size: 119, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:43:27,874 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:43:48,078 INFO [train.py:968] (1/2) Epoch 6, batch 18500, giga_loss[loss=0.2879, simple_loss=0.3613, pruned_loss=0.1073, over 29037.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3567, pruned_loss=0.1104, over 5703738.12 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3529, pruned_loss=0.1044, over 5770280.57 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3531, pruned_loss=0.1091, over 5691997.55 frames. ], batch size: 128, lr: 5.40e-03, grad_scale: 4.0
+2023-03-03 05:44:02,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:44:31,482 INFO [train.py:968] (1/2) Epoch 6, batch 18550, giga_loss[loss=0.3239, simple_loss=0.3859, pruned_loss=0.131, over 28830.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3583, pruned_loss=0.11, over 5696076.63 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3531, pruned_loss=0.1044, over 5769497.01 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3553, pruned_loss=0.1091, over 5686098.61 frames. ], batch size: 186, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:44:53,302 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:45:18,411 INFO [train.py:968] (1/2) Epoch 6, batch 18600, giga_loss[loss=0.2986, simple_loss=0.369, pruned_loss=0.1142, over 28611.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3601, pruned_loss=0.111, over 5691376.66 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3535, pruned_loss=0.1046, over 5767987.24 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3575, pruned_loss=0.1102, over 5683489.20 frames. ], batch size: 307, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:45:20,511 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 18650, giga_loss[loss=0.3304, simple_loss=0.4019, pruned_loss=0.1295, over 28912.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3634, pruned_loss=0.1133, over 5690771.19 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3544, pruned_loss=0.1049, over 5767459.63 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3607, pruned_loss=0.1127, over 5681920.33 frames. ], batch size: 174, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:46:08,626 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:46:10,765 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:46:38,340 INFO [optim.py:369] (1/2) 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 05:46:44,156 INFO [train.py:968] (1/2) Epoch 6, batch 18700, giga_loss[loss=0.285, simple_loss=0.3584, pruned_loss=0.1058, over 28889.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3663, pruned_loss=0.1154, over 5699173.00 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3547, pruned_loss=0.105, over 5769739.29 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3641, pruned_loss=0.1149, over 5689388.02 frames. ], batch size: 199, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:47:19,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246454.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:47:22,042 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:47:24,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:47:24,667 INFO [train.py:968] (1/2) Epoch 6, batch 18750, giga_loss[loss=0.3108, simple_loss=0.3904, pruned_loss=0.1156, over 28982.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3695, pruned_loss=0.1166, over 5702682.12 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3552, pruned_loss=0.1052, over 5769274.53 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3675, pruned_loss=0.1162, over 5693689.93 frames. ], batch size: 155, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:47:47,911 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:47:58,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5030, 1.4301, 1.4680, 1.3472], device='cuda:1'), covar=tensor([0.1929, 0.2751, 0.1501, 0.1524], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0722, 0.0799, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 05:47:59,943 INFO [optim.py:369] (1/2) 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 05:48:01,094 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:48:06,882 INFO [train.py:968] (1/2) Epoch 6, batch 18800, giga_loss[loss=0.3052, simple_loss=0.3748, pruned_loss=0.1178, over 28455.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3716, pruned_loss=0.1168, over 5705859.87 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3554, pruned_loss=0.1052, over 5769717.67 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3702, pruned_loss=0.1167, over 5697388.01 frames. ], batch size: 71, lr: 5.39e-03, grad_scale: 8.0
+2023-03-03 05:48:11,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4140, 1.9487, 1.4125, 0.5422], device='cuda:1'), covar=tensor([0.2532, 0.1453, 0.1986, 0.2978], device='cuda:1'), in_proj_covar=tensor([0.1403, 0.1321, 0.1376, 0.1170], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 05:48:33,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3516, 3.2929, 1.5218, 1.3770], device='cuda:1'), covar=tensor([0.0950, 0.0252, 0.0863, 0.1360], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0475, 0.0311, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0025, 0.0017, 0.0021], device='cuda:1')
+2023-03-03 05:48:46,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:48:46,814 INFO [train.py:968] (1/2) Epoch 6, batch 18850, giga_loss[loss=0.2991, simple_loss=0.3729, pruned_loss=0.1126, over 28294.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3733, pruned_loss=0.1169, over 5702191.56 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3558, pruned_loss=0.1055, over 5762385.88 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3719, pruned_loss=0.1167, over 5701257.61 frames. ], batch size: 368, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:49:10,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:49:15,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246597.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:49:17,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246600.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:49:22,473 INFO [optim.py:369] (1/2) 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 05:49:26,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9897, 1.3168, 1.0584, 0.1522], device='cuda:1'), covar=tensor([0.1915, 0.1513, 0.2573, 0.3229], device='cuda:1'), in_proj_covar=tensor([0.1421, 0.1336, 0.1391, 0.1180], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 05:49:26,474 INFO [train.py:968] (1/2) Epoch 6, batch 18900, giga_loss[loss=0.2915, simple_loss=0.3733, pruned_loss=0.1049, over 28806.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3726, pruned_loss=0.1156, over 5702676.79 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3564, pruned_loss=0.1059, over 5765465.26 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3715, pruned_loss=0.1154, over 5697359.79 frames. ], batch size: 174, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:49:29,528 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=246614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:49:41,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246629.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:49:41,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 05:50:06,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:50:07,166 INFO [train.py:968] (1/2) Epoch 6, batch 18950, giga_loss[loss=0.3065, simple_loss=0.3791, pruned_loss=0.117, over 29075.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3711, pruned_loss=0.1138, over 5704452.34 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3568, pruned_loss=0.106, over 5768678.96 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3702, pruned_loss=0.1137, over 5695725.48 frames. ], batch size: 113, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:50:12,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4296, 1.0025, 2.8795, 2.4745], device='cuda:1'), covar=tensor([0.1781, 0.2259, 0.0489, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0579, 0.0540, 0.0757, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 05:50:42,052 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:50:42,960 INFO [optim.py:369] (1/2) 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 05:50:44,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=246706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:50:47,504 INFO [train.py:968] (1/2) Epoch 6, batch 19000, giga_loss[loss=0.3035, simple_loss=0.3723, pruned_loss=0.1173, over 29011.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3695, pruned_loss=0.1121, over 5711138.75 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3569, pruned_loss=0.106, over 5771608.08 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3689, pruned_loss=0.1122, over 5700369.36 frames. ], batch size: 213, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:51:08,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:51:29,913 INFO [train.py:968] (1/2) Epoch 6, batch 19050, giga_loss[loss=0.3011, simple_loss=0.3735, pruned_loss=0.1143, over 28599.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3705, pruned_loss=0.1144, over 5700316.22 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3569, pruned_loss=0.1058, over 5774585.19 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3704, pruned_loss=0.1147, over 5687921.25 frames. ], batch size: 78, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:52:07,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=246802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:52:10,837 INFO [optim.py:369] (1/2) 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 05:52:11,074 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 19100, giga_loss[loss=0.3368, simple_loss=0.4043, pruned_loss=0.1347, over 28524.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3738, pruned_loss=0.1198, over 5685254.29 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3569, pruned_loss=0.1058, over 5770329.39 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.374, pruned_loss=0.1203, over 5677794.17 frames. ], batch size: 60, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:52:36,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=246834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:52:39,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-03 05:52:58,254 INFO [train.py:968] (1/2) Epoch 6, batch 19150, giga_loss[loss=0.2847, simple_loss=0.3503, pruned_loss=0.1096, over 28495.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3747, pruned_loss=0.1217, over 5695724.54 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5773813.60 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3751, pruned_loss=0.1225, over 5684659.91 frames. ], batch size: 71, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:53:12,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:53:19,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5078, 0.9496, 2.7252, 2.5084], device='cuda:1'), covar=tensor([0.1369, 0.1857, 0.0436, 0.0745], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0534, 0.0748, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 05:53:31,957 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 19200, giga_loss[loss=0.2905, simple_loss=0.3408, pruned_loss=0.1201, over 23788.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3723, pruned_loss=0.1211, over 5697772.97 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.357, pruned_loss=0.1058, over 5773823.97 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.373, pruned_loss=0.1221, over 5687187.92 frames. ], batch size: 705, lr: 5.39e-03, grad_scale: 8.0
+2023-03-03 05:53:48,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4738, 2.1673, 1.5803, 0.6195], device='cuda:1'), covar=tensor([0.3162, 0.1720, 0.2703, 0.3367], device='cuda:1'), in_proj_covar=tensor([0.1417, 0.1325, 0.1385, 0.1173], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 05:54:19,753 INFO [train.py:968] (1/2) Epoch 6, batch 19250, giga_loss[loss=0.2796, simple_loss=0.3537, pruned_loss=0.1027, over 29106.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3713, pruned_loss=0.1204, over 5702909.55 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3576, pruned_loss=0.106, over 5776308.06 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5691057.14 frames. ], batch size: 128, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:54:23,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246964.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:54:38,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8371, 3.6690, 3.4650, 1.8048], device='cuda:1'), covar=tensor([0.0554, 0.0598, 0.0714, 0.2226], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0821, 0.0771, 0.0612], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:54:45,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=246989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:54:49,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5223, 1.2882, 5.2197, 3.6624], device='cuda:1'), covar=tensor([0.1471, 0.2232, 0.0266, 0.0596], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0539, 0.0752, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 05:54:58,211 INFO [optim.py:369] (1/2) 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 05:55:03,247 INFO [train.py:968] (1/2) Epoch 6, batch 19300, giga_loss[loss=0.2916, simple_loss=0.3614, pruned_loss=0.1109, over 28834.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.371, pruned_loss=0.1199, over 5696951.04 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3577, pruned_loss=0.1059, over 5780800.76 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3716, pruned_loss=0.1212, over 5681158.47 frames. ], batch size: 186, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:55:12,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5369, 1.7067, 1.7751, 1.6078], device='cuda:1'), covar=tensor([0.1433, 0.1814, 0.1107, 0.1217], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0727, 0.0798, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 05:55:13,164 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:55:16,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3445, 4.1578, 4.0401, 1.8069], device='cuda:1'), covar=tensor([0.0486, 0.0541, 0.0647, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0819, 0.0771, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:55:38,245 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 19350, giga_loss[loss=0.2554, simple_loss=0.3327, pruned_loss=0.08901, over 28887.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3692, pruned_loss=0.1181, over 5692676.16 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3582, pruned_loss=0.1062, over 5771934.26 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3694, pruned_loss=0.1191, over 5687013.61 frames. ], batch size: 145, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:55:51,074 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:56:03,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4334, 1.5059, 1.3761, 1.4663], device='cuda:1'), covar=tensor([0.1742, 0.1599, 0.1477, 0.1546], device='cuda:1'), in_proj_covar=tensor([0.1133, 0.0862, 0.1004, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 05:56:23,439 INFO [optim.py:369] (1/2) 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 05:56:25,953 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247107.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 05:56:28,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247110.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 05:56:29,132 INFO [train.py:968] (1/2) Epoch 6, batch 19400, giga_loss[loss=0.2607, simple_loss=0.3334, pruned_loss=0.094, over 28708.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3635, pruned_loss=0.1139, over 5688053.98 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3582, pruned_loss=0.1061, over 5775047.18 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.364, pruned_loss=0.115, over 5679122.66 frames. ], batch size: 119, lr: 5.39e-03, grad_scale: 4.0
+2023-03-03 05:56:44,998 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:56:50,810 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 19450, giga_loss[loss=0.2599, simple_loss=0.3263, pruned_loss=0.09674, over 28313.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3576, pruned_loss=0.1105, over 5687881.52 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3584, pruned_loss=0.1062, over 5777440.61 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3578, pruned_loss=0.1115, over 5676447.99 frames. ], batch size: 368, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 05:57:16,337 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 05:57:18,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3891, 2.1756, 1.6099, 0.6292], device='cuda:1'), covar=tensor([0.3009, 0.1391, 0.2203, 0.3179], device='cuda:1'), in_proj_covar=tensor([0.1387, 0.1297, 0.1353, 0.1144], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1')
+2023-03-03 05:57:51,929 INFO [optim.py:369] (1/2) 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 05:57:55,965 INFO [train.py:968] (1/2) Epoch 6, batch 19500, giga_loss[loss=0.2405, simple_loss=0.321, pruned_loss=0.08003, over 29024.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3515, pruned_loss=0.1071, over 5688490.16 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3581, pruned_loss=0.1059, over 5780849.97 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3519, pruned_loss=0.1082, over 5674273.89 frames. ], batch size: 155, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 05:58:04,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9105, 2.5776, 1.5798, 1.4529], device='cuda:1'), covar=tensor([0.1503, 0.0867, 0.1086, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.1482, 0.1305, 0.1247, 0.1371], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 05:58:25,841 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 6, batch 19550, giga_loss[loss=0.2628, simple_loss=0.317, pruned_loss=0.1043, over 23508.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3484, pruned_loss=0.1049, over 5691260.56 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3581, pruned_loss=0.1059, over 5780245.19 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3485, pruned_loss=0.1058, over 5679837.88 frames. ], batch size: 705, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 05:58:59,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8655, 1.7690, 1.3747, 1.3799], device='cuda:1'), covar=tensor([0.0664, 0.0645, 0.0925, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0435, 0.0490, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:59:16,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-03 05:59:23,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0657, 3.9042, 3.7001, 1.8925], device='cuda:1'), covar=tensor([0.0497, 0.0558, 0.0654, 0.2032], device='cuda:1'), in_proj_covar=tensor([0.0855, 0.0801, 0.0758, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 05:59:29,620 INFO [optim.py:369] (1/2) 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 05:59:34,880 INFO [train.py:968] (1/2) Epoch 6, batch 19600, giga_loss[loss=0.255, simple_loss=0.3345, pruned_loss=0.08774, over 29066.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3501, pruned_loss=0.1056, over 5700323.01 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3589, pruned_loss=0.1064, over 5780730.56 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3494, pruned_loss=0.1058, over 5690088.04 frames. ], batch size: 136, lr: 5.38e-03, grad_scale: 8.0
+2023-03-03 05:59:46,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:00:15,955 INFO [train.py:968] (1/2) Epoch 6, batch 19650, giga_loss[loss=0.263, simple_loss=0.3315, pruned_loss=0.09728, over 28837.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3503, pruned_loss=0.1054, over 5688698.79 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3601, pruned_loss=0.1069, over 5764220.76 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3484, pruned_loss=0.1052, over 5692516.88 frames. ], batch size: 119, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:00:56,136 INFO [optim.py:369] (1/2) 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 06:00:59,881 INFO [train.py:968] (1/2) Epoch 6, batch 19700, giga_loss[loss=0.2699, simple_loss=0.3368, pruned_loss=0.1015, over 28952.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3486, pruned_loss=0.105, over 5685325.37 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3607, pruned_loss=0.1072, over 5746592.22 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3464, pruned_loss=0.1045, over 5703445.42 frames. ], batch size: 145, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:01:16,445 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-03 06:01:25,640 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 19750, giga_loss[loss=0.2363, simple_loss=0.2993, pruned_loss=0.08663, over 28537.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3454, pruned_loss=0.1032, over 5696845.36 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3614, pruned_loss=0.1074, over 5746396.47 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3428, pruned_loss=0.1025, over 5710412.22 frames. ], batch size: 85, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:02:01,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 06:02:17,213 INFO [optim.py:369] (1/2) 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 06:02:19,077 INFO [train.py:968] (1/2) Epoch 6, batch 19800, giga_loss[loss=0.2447, simple_loss=0.3161, pruned_loss=0.08668, over 28790.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3428, pruned_loss=0.1023, over 5704003.30 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3613, pruned_loss=0.1072, over 5748723.31 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3406, pruned_loss=0.1019, over 5711965.39 frames. ], batch size: 119, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:03:00,166 INFO [train.py:968] (1/2) Epoch 6, batch 19850, giga_loss[loss=0.2448, simple_loss=0.3142, pruned_loss=0.08775, over 28630.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3419, pruned_loss=0.1018, over 5712321.19 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3622, pruned_loss=0.1076, over 5751827.68 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3387, pruned_loss=0.101, over 5714497.98 frames. ], batch size: 85, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:03:19,424 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:03:23,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=247591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:03:34,940 INFO [optim.py:369] (1/2) 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 06:03:37,170 INFO [train.py:968] (1/2) Epoch 6, batch 19900, giga_loss[loss=0.2973, simple_loss=0.356, pruned_loss=0.1193, over 28834.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3401, pruned_loss=0.1011, over 5718761.94 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3627, pruned_loss=0.1078, over 5755085.96 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3368, pruned_loss=0.1002, over 5716989.58 frames. ], batch size: 112, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:03:45,289 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:04:08,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5067, 2.0976, 1.5336, 0.7402], device='cuda:1'), covar=tensor([0.2829, 0.1384, 0.2108, 0.3021], device='cuda:1'), in_proj_covar=tensor([0.1413, 0.1324, 0.1389, 0.1170], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 06:04:16,680 INFO [train.py:968] (1/2) Epoch 6, batch 19950, giga_loss[loss=0.2399, simple_loss=0.3142, pruned_loss=0.0828, over 28908.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3388, pruned_loss=0.1003, over 5707423.95 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3634, pruned_loss=0.1079, over 5746094.11 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3347, pruned_loss=0.09911, over 5712663.77 frames. ], batch size: 112, lr: 5.38e-03, grad_scale: 2.0
+2023-03-03 06:04:49,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=247702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:04:52,728 INFO [optim.py:369] (1/2) 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 06:04:55,754 INFO [train.py:968] (1/2) Epoch 6, batch 20000, giga_loss[loss=0.2397, simple_loss=0.3099, pruned_loss=0.08473, over 28446.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3365, pruned_loss=0.09909, over 5714656.57 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3641, pruned_loss=0.1083, over 5747628.06 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3321, pruned_loss=0.0977, over 5716878.42 frames. ], batch size: 78, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:05:12,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4802, 2.2336, 2.1618, 2.0728], device='cuda:1'), covar=tensor([0.1178, 0.2072, 0.1600, 0.1709], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0739, 0.0647, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 06:05:34,966 INFO [train.py:968] (1/2) Epoch 6, batch 20050, giga_loss[loss=0.2805, simple_loss=0.3536, pruned_loss=0.1036, over 28273.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3335, pruned_loss=0.0969, over 5723914.99 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3644, pruned_loss=0.1084, over 5749748.80 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3294, pruned_loss=0.09561, over 5723461.30 frames. ], batch size: 368, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:05:40,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7404, 1.3941, 5.5783, 3.9732], device='cuda:1'), covar=tensor([0.1340, 0.2131, 0.0231, 0.0493], device='cuda:1'), in_proj_covar=tensor([0.0575, 0.0541, 0.0749, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 06:05:45,736 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:06:11,775 INFO [optim.py:369] (1/2) 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 06:06:13,855 INFO [train.py:968] (1/2) Epoch 6, batch 20100, giga_loss[loss=0.2463, simple_loss=0.3178, pruned_loss=0.08736, over 28893.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3329, pruned_loss=0.09653, over 5730324.20 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3654, pruned_loss=0.1091, over 5754507.40 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3279, pruned_loss=0.09456, over 5725038.21 frames. ], batch size: 199, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:06:20,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:06:26,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2641, 1.3152, 1.2218, 1.5232], device='cuda:1'), covar=tensor([0.0781, 0.0340, 0.0319, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0123, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0062, 0.0045, 0.0040, 0.0068], device='cuda:1')
+2023-03-03 06:06:38,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2940, 4.1472, 3.9292, 2.1127], device='cuda:1'), covar=tensor([0.0435, 0.0485, 0.0637, 0.1865], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0810, 0.0765, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:06:40,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=247845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:06:44,621 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 20150, libri_loss[loss=0.3222, simple_loss=0.398, pruned_loss=0.1232, over 29521.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3368, pruned_loss=0.09872, over 5734956.69 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3665, pruned_loss=0.1093, over 5758400.33 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3302, pruned_loss=0.09632, over 5725871.52 frames. ], batch size: 81, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:07:10,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1361, 1.4076, 3.1522, 2.9305], device='cuda:1'), covar=tensor([0.1293, 0.2004, 0.0393, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0538, 0.0748, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 06:07:10,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=247877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:07:23,881 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:07:37,792 INFO [optim.py:369] (1/2) 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 06:07:41,605 INFO [train.py:968] (1/2) Epoch 6, batch 20200, giga_loss[loss=0.3033, simple_loss=0.3693, pruned_loss=0.1187, over 28951.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.343, pruned_loss=0.1034, over 5721377.49 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3665, pruned_loss=0.1094, over 5758999.33 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3377, pruned_loss=0.1014, over 5713633.12 frames. ], batch size: 164, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:07:50,177 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=247947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:08:21,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3968, 1.4433, 1.3952, 1.4254], device='cuda:1'), covar=tensor([0.0975, 0.1335, 0.1463, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0739, 0.0651, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 06:08:31,691 INFO [train.py:968] (1/2) Epoch 6, batch 20250, giga_loss[loss=0.4645, simple_loss=0.4746, pruned_loss=0.2272, over 26411.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3533, pruned_loss=0.111, over 5695775.38 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3665, pruned_loss=0.1095, over 5750933.04 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3487, pruned_loss=0.1093, over 5695749.77 frames. ], batch size: 555, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:09:05,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2547, 1.6931, 1.2539, 0.4005], device='cuda:1'), covar=tensor([0.1672, 0.1089, 0.1843, 0.2564], device='cuda:1'), in_proj_covar=tensor([0.1404, 0.1326, 0.1379, 0.1163], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 06:09:18,308 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 20300, giga_loss[loss=0.2939, simple_loss=0.3771, pruned_loss=0.1053, over 28530.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3596, pruned_loss=0.1147, over 5694207.28 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3662, pruned_loss=0.1093, over 5752063.61 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3562, pruned_loss=0.1136, over 5692542.26 frames. ], batch size: 307, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:09:36,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8154, 2.5846, 1.7652, 0.9166], device='cuda:1'), covar=tensor([0.3453, 0.1789, 0.2050, 0.3361], device='cuda:1'), in_proj_covar=tensor([0.1404, 0.1328, 0.1382, 0.1165], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 06:10:05,698 INFO [train.py:968] (1/2) Epoch 6, batch 20350, giga_loss[loss=0.2863, simple_loss=0.3643, pruned_loss=0.1041, over 28883.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1163, over 5674005.51 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3671, pruned_loss=0.1098, over 5745903.74 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3605, pruned_loss=0.1151, over 5676205.12 frames. ], batch size: 186, lr: 5.38e-03, grad_scale: 4.0
+2023-03-03 06:10:48,911 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 20400, giga_loss[loss=0.2933, simple_loss=0.3691, pruned_loss=0.1087, over 28832.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3684, pruned_loss=0.1181, over 5677933.15 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3672, pruned_loss=0.11, over 5748103.66 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3654, pruned_loss=0.1172, over 5676372.27 frames. ], batch size: 186, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:11:25,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:11:35,101 INFO [train.py:968] (1/2) Epoch 6, batch 20450, giga_loss[loss=0.3125, simple_loss=0.3752, pruned_loss=0.1249, over 28754.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3755, pruned_loss=0.1233, over 5666635.11 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3677, pruned_loss=0.1104, over 5741527.56 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3727, pruned_loss=0.1224, over 5670229.62 frames. ], batch size: 92, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:11:50,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:12:06,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:12:19,849 INFO [optim.py:369] (1/2) 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 06:12:21,154 INFO [train.py:968] (1/2) Epoch 6, batch 20500, giga_loss[loss=0.2425, simple_loss=0.3192, pruned_loss=0.08294, over 28595.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3694, pruned_loss=0.1189, over 5671211.30 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3677, pruned_loss=0.1105, over 5742214.46 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3673, pruned_loss=0.1181, over 5672762.63 frames. ], batch size: 262, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:13:02,168 INFO [train.py:968] (1/2) Epoch 6, batch 20550, giga_loss[loss=0.3116, simple_loss=0.3751, pruned_loss=0.124, over 28879.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.366, pruned_loss=0.1157, over 5676008.77 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3685, pruned_loss=0.1111, over 5736906.34 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3635, pruned_loss=0.1147, over 5680976.81 frames. ], batch size: 119, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:13:07,987 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:16,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6298, 1.6134, 1.6280, 1.4928], device='cuda:1'), covar=tensor([0.0969, 0.1407, 0.1495, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0736, 0.0639, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 06:13:26,791 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:29,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:29,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:43,788 INFO [optim.py:369] (1/2) 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 06:13:45,145 INFO [train.py:968] (1/2) Epoch 6, batch 20600, giga_loss[loss=0.3353, simple_loss=0.3903, pruned_loss=0.1402, over 28879.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3642, pruned_loss=0.1142, over 5683310.84 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3686, pruned_loss=0.1113, over 5737436.77 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3621, pruned_loss=0.1133, over 5685679.77 frames. ], batch size: 112, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:13:55,912 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:13:55,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:08,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:11,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:29,116 INFO [train.py:968] (1/2) Epoch 6, batch 20650, giga_loss[loss=0.308, simple_loss=0.3779, pruned_loss=0.1191, over 28945.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.364, pruned_loss=0.1131, over 5683435.73 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.369, pruned_loss=0.1116, over 5739288.02 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3619, pruned_loss=0.1121, over 5683020.27 frames. ], batch size: 186, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:14:34,710 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:14:42,110 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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] (1/2) Epoch 6, batch 20700, giga_loss[loss=0.2724, simple_loss=0.3509, pruned_loss=0.09694, over 28883.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.366, pruned_loss=0.1144, over 5691914.30 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3686, pruned_loss=0.1114, over 5743374.86 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3646, pruned_loss=0.1139, over 5686727.57 frames. ], batch size: 112, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:15:12,539 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:14,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:36,641 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:39,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:41,834 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:15:55,894 INFO [train.py:968] (1/2) Epoch 6, batch 20750, giga_loss[loss=0.3106, simple_loss=0.3825, pruned_loss=0.1193, over 28887.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3677, pruned_loss=0.1156, over 5704741.86 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3686, pruned_loss=0.1112, over 5746415.39 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3665, pruned_loss=0.1154, over 5697003.87 frames. ], batch size: 112, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:15:59,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:03,056 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:03,088 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:16:14,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-03 06:16:31,744 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:16:43,925 INFO [train.py:968] (1/2) Epoch 6, batch 20800, giga_loss[loss=0.2843, simple_loss=0.3567, pruned_loss=0.106, over 28581.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3686, pruned_loss=0.117, over 5687002.36 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3687, pruned_loss=0.1114, over 5747238.31 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3675, pruned_loss=0.1168, over 5679828.80 frames. ], batch size: 262, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:17:19,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3367, 3.1294, 3.0592, 1.3798], device='cuda:1'), covar=tensor([0.0828, 0.0993, 0.1095, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0825, 0.0771, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:17:19,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:17:26,675 INFO [train.py:968] (1/2) Epoch 6, batch 20850, giga_loss[loss=0.283, simple_loss=0.3541, pruned_loss=0.1059, over 28702.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1192, over 5689554.57 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3697, pruned_loss=0.112, over 5747119.93 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3691, pruned_loss=0.1185, over 5682189.36 frames. ], batch size: 262, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:18:06,968 INFO [optim.py:369] (1/2) 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 06:18:08,249 INFO [train.py:968] (1/2) Epoch 6, batch 20900, giga_loss[loss=0.2948, simple_loss=0.3625, pruned_loss=0.1136, over 28962.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3705, pruned_loss=0.1185, over 5691498.21 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3698, pruned_loss=0.1122, over 5739493.27 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.369, pruned_loss=0.1179, over 5691260.82 frames. ], batch size: 145, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:18:47,749 INFO [train.py:968] (1/2) Epoch 6, batch 20950, giga_loss[loss=0.2557, simple_loss=0.3334, pruned_loss=0.08899, over 28661.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3703, pruned_loss=0.1175, over 5692454.60 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3703, pruned_loss=0.1126, over 5740317.85 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3686, pruned_loss=0.1168, over 5690725.60 frames. ], batch size: 92, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:19:15,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:19:18,079 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:19:31,784 INFO [train.py:968] (1/2) Epoch 6, batch 21000, giga_loss[loss=0.3283, simple_loss=0.3932, pruned_loss=0.1317, over 28727.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.371, pruned_loss=0.116, over 5698192.12 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3706, pruned_loss=0.1126, over 5742214.42 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3695, pruned_loss=0.1155, over 5694925.98 frames. ], batch size: 92, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:19:31,784 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 06:19:39,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2595, 1.3757, 1.0704, 1.4163], device='cuda:1'), covar=tensor([0.0760, 0.0290, 0.0359, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0122, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0062, 0.0045, 0.0040, 0.0068], device='cuda:1')
+2023-03-03 06:19:40,437 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 06:19:52,030 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:19:53,343 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:14,696 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:21,382 INFO [train.py:968] (1/2) Epoch 6, batch 21050, giga_loss[loss=0.3316, simple_loss=0.3969, pruned_loss=0.1332, over 28812.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3706, pruned_loss=0.116, over 5690887.37 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3708, pruned_loss=0.1129, over 5736499.37 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3692, pruned_loss=0.1154, over 5692688.57 frames. ], batch size: 284, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:20:23,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=248764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:25,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7317, 1.6696, 1.2712, 1.3104], device='cuda:1'), covar=tensor([0.0625, 0.0554, 0.0903, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0438, 0.0497, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:20:33,128 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=248776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:20:48,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4961, 1.7849, 0.9948, 1.2966], device='cuda:1'), covar=tensor([0.1006, 0.0756, 0.1661, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0435, 0.0494, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:20:58,827 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 21100, giga_loss[loss=0.3618, simple_loss=0.4104, pruned_loss=0.1566, over 28324.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3676, pruned_loss=0.1143, over 5704648.51 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.371, pruned_loss=0.1131, over 5738069.54 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3664, pruned_loss=0.1136, over 5704206.84 frames. ], batch size: 368, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:21:01,587 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-03 06:21:30,536 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 21150, giga_loss[loss=0.3463, simple_loss=0.3867, pruned_loss=0.1529, over 26538.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3665, pruned_loss=0.1141, over 5708626.01 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3713, pruned_loss=0.1135, over 5741972.28 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3651, pruned_loss=0.1133, over 5703982.92 frames. ], batch size: 555, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:22:06,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:08,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:11,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8022, 1.1358, 2.6731, 2.5354], device='cuda:1'), covar=tensor([0.1360, 0.1896, 0.0493, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0529, 0.0743, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 06:22:14,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:17,532 INFO [optim.py:369] (1/2) 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 06:22:17,838 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 21200, giga_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.09883, over 28965.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3653, pruned_loss=0.1136, over 5705389.04 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.372, pruned_loss=0.1141, over 5737323.23 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3634, pruned_loss=0.1124, over 5705600.81 frames. ], batch size: 112, lr: 5.37e-03, grad_scale: 8.0
+2023-03-03 06:22:32,634 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:39,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3386, 1.4615, 1.4997, 1.3673], device='cuda:1'), covar=tensor([0.1022, 0.1263, 0.1496, 0.1209], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0732, 0.0637, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 06:22:42,407 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=248939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:22:58,429 INFO [train.py:968] (1/2) Epoch 6, batch 21250, giga_loss[loss=0.2898, simple_loss=0.3583, pruned_loss=0.1106, over 28849.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3681, pruned_loss=0.1164, over 5705480.53 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3731, pruned_loss=0.1153, over 5738318.11 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3653, pruned_loss=0.1144, over 5703390.53 frames. ], batch size: 112, lr: 5.37e-03, grad_scale: 4.0
+2023-03-03 06:23:24,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=248993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:23:26,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=248996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:23:30,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-03 06:23:38,283 INFO [optim.py:369] (1/2) 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 06:23:38,296 INFO [train.py:968] (1/2) Epoch 6, batch 21300, giga_loss[loss=0.2905, simple_loss=0.3663, pruned_loss=0.1074, over 28539.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3682, pruned_loss=0.116, over 5713056.58 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3729, pruned_loss=0.1155, over 5742122.25 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3659, pruned_loss=0.1141, over 5707169.21 frames. ], batch size: 307, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:23:49,153 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:24:08,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-03 06:24:12,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-03 06:24:16,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3246, 1.8280, 1.4522, 0.5041], device='cuda:1'), covar=tensor([0.1971, 0.1263, 0.2114, 0.2681], device='cuda:1'), in_proj_covar=tensor([0.1411, 0.1299, 0.1367, 0.1157], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 06:24:17,878 INFO [train.py:968] (1/2) Epoch 6, batch 21350, giga_loss[loss=0.3095, simple_loss=0.382, pruned_loss=0.1184, over 28720.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3671, pruned_loss=0.1146, over 5700942.42 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3728, pruned_loss=0.1155, over 5735172.73 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3654, pruned_loss=0.1132, over 5701916.37 frames. ], batch size: 284, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:24:23,153 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 06:24:30,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7568, 1.6416, 1.2452, 1.3503], device='cuda:1'), covar=tensor([0.0701, 0.0643, 0.0986, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0437, 0.0498, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:24:38,484 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:24:51,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:24:56,760 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:24:58,953 INFO [train.py:968] (1/2) Epoch 6, batch 21400, giga_loss[loss=0.2786, simple_loss=0.3553, pruned_loss=0.1009, over 28888.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3648, pruned_loss=0.1124, over 5704544.49 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3724, pruned_loss=0.1155, over 5727091.55 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3636, pruned_loss=0.1112, over 5711955.67 frames. ], batch size: 227, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:25:32,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:25:39,414 INFO [train.py:968] (1/2) Epoch 6, batch 21450, giga_loss[loss=0.3202, simple_loss=0.3831, pruned_loss=0.1286, over 28865.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3653, pruned_loss=0.1131, over 5713299.04 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.373, pruned_loss=0.1164, over 5728464.64 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3634, pruned_loss=0.1111, over 5717012.57 frames. ], batch size: 112, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:26:18,497 INFO [optim.py:369] (1/2) 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 06:26:18,509 INFO [train.py:968] (1/2) Epoch 6, batch 21500, giga_loss[loss=0.2898, simple_loss=0.3551, pruned_loss=0.1123, over 28855.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3639, pruned_loss=0.1127, over 5720764.49 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3738, pruned_loss=0.1172, over 5733249.06 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3614, pruned_loss=0.1103, over 5719083.88 frames. ], batch size: 99, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:26:43,334 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:26:46,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:26:57,342 INFO [train.py:968] (1/2) Epoch 6, batch 21550, giga_loss[loss=0.279, simple_loss=0.344, pruned_loss=0.107, over 28618.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3604, pruned_loss=0.1108, over 5718393.83 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3742, pruned_loss=0.1177, over 5736205.54 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3577, pruned_loss=0.1083, over 5713962.45 frames. ], batch size: 78, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:26:58,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 06:27:08,060 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:27:21,709 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:27:24,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:27:30,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1865, 1.1682, 1.0231, 0.9769], device='cuda:1'), covar=tensor([0.0626, 0.0444, 0.0947, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0437, 0.0500, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:27:36,463 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 21600, giga_loss[loss=0.2658, simple_loss=0.3397, pruned_loss=0.09594, over 28665.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3597, pruned_loss=0.1107, over 5728133.78 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.374, pruned_loss=0.1178, over 5742946.13 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.357, pruned_loss=0.1082, over 5717726.45 frames. ], batch size: 60, lr: 5.36e-03, grad_scale: 8.0
+2023-03-03 06:27:47,469 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:27:54,692 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 06:28:00,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3490, 1.5965, 1.3744, 1.1975], device='cuda:1'), covar=tensor([0.2087, 0.1937, 0.2058, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.1137, 0.0868, 0.1002, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 06:28:15,273 INFO [train.py:968] (1/2) Epoch 6, batch 21650, giga_loss[loss=0.3387, simple_loss=0.3942, pruned_loss=0.1416, over 27678.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3606, pruned_loss=0.1122, over 5728609.51 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3743, pruned_loss=0.1182, over 5749205.38 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3578, pruned_loss=0.1096, over 5713900.54 frames. ], batch size: 472, lr: 5.36e-03, grad_scale: 8.0
+2023-03-03 06:28:54,906 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 21700, libri_loss[loss=0.3807, simple_loss=0.4, pruned_loss=0.1807, over 29657.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3598, pruned_loss=0.1131, over 5725944.64 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3747, pruned_loss=0.119, over 5748688.25 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3568, pruned_loss=0.1101, over 5714513.50 frames. ], batch size: 69, lr: 5.36e-03, grad_scale: 8.0
+2023-03-03 06:28:57,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-03 06:28:57,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-03 06:29:31,787 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 21750, giga_loss[loss=0.3153, simple_loss=0.3692, pruned_loss=0.1307, over 28688.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3581, pruned_loss=0.1127, over 5718759.46 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3746, pruned_loss=0.1194, over 5743786.62 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3552, pruned_loss=0.1096, over 5712223.81 frames. ], batch size: 262, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:29:48,659 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:30:10,445 INFO [train.py:968] (1/2) Epoch 6, batch 21800, giga_loss[loss=0.2743, simple_loss=0.3465, pruned_loss=0.1011, over 28585.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3562, pruned_loss=0.1119, over 5714859.84 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3751, pruned_loss=0.1199, over 5744366.73 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3527, pruned_loss=0.1087, over 5708274.41 frames. ], batch size: 336, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:30:11,750 INFO [optim.py:369] (1/2) 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 06:30:29,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4382, 3.5711, 1.5687, 1.4169], device='cuda:1'), covar=tensor([0.0838, 0.0319, 0.0825, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0477, 0.0308, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 06:30:48,721 INFO [train.py:968] (1/2) Epoch 6, batch 21850, giga_loss[loss=0.2293, simple_loss=0.3089, pruned_loss=0.07487, over 29013.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3527, pruned_loss=0.1097, over 5717400.00 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3755, pruned_loss=0.1202, over 5746567.65 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3492, pruned_loss=0.1068, over 5709803.16 frames. ], batch size: 66, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:31:25,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:27,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:30,235 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=249608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:32,028 INFO [train.py:968] (1/2) Epoch 6, batch 21900, giga_loss[loss=0.246, simple_loss=0.3205, pruned_loss=0.0858, over 28435.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3508, pruned_loss=0.1084, over 5711979.41 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3755, pruned_loss=0.1202, over 5747371.81 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3481, pruned_loss=0.1061, over 5705279.29 frames. ], batch size: 71, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:31:33,364 INFO [optim.py:369] (1/2) 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 06:31:45,184 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=249626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:47,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=249629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:31:51,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:32:11,267 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=249658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:32:12,969 INFO [train.py:968] (1/2) Epoch 6, batch 21950, giga_loss[loss=0.3125, simple_loss=0.3786, pruned_loss=0.1232, over 29031.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3538, pruned_loss=0.1099, over 5711026.53 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3759, pruned_loss=0.1209, over 5743726.14 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3504, pruned_loss=0.107, over 5707350.69 frames. ], batch size: 128, lr: 5.36e-03, grad_scale: 2.0
+2023-03-03 06:32:52,922 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 6, batch 22000, giga_loss[loss=0.2741, simple_loss=0.3569, pruned_loss=0.09567, over 28485.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3561, pruned_loss=0.1102, over 5712064.60 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3763, pruned_loss=0.1212, over 5745456.80 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3528, pruned_loss=0.1076, over 5707150.72 frames. ], batch size: 336, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:33:00,301 INFO [optim.py:369] (1/2) 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 06:33:17,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3090, 2.0569, 1.6529, 0.5412], device='cuda:1'), covar=tensor([0.2533, 0.1255, 0.1839, 0.3012], device='cuda:1'), in_proj_covar=tensor([0.1425, 0.1316, 0.1384, 0.1170], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 06:33:39,068 INFO [train.py:968] (1/2) Epoch 6, batch 22050, giga_loss[loss=0.2747, simple_loss=0.3478, pruned_loss=0.1007, over 29034.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.358, pruned_loss=0.1106, over 5712378.58 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3764, pruned_loss=0.1216, over 5748864.09 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3547, pruned_loss=0.1077, over 5704303.67 frames. ], batch size: 128, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:34:13,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5890, 2.1940, 1.3543, 0.6973], device='cuda:1'), covar=tensor([0.4040, 0.2177, 0.2297, 0.3714], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1320, 0.1387, 0.1169], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 06:34:21,831 INFO [train.py:968] (1/2) Epoch 6, batch 22100, giga_loss[loss=0.3193, simple_loss=0.3708, pruned_loss=0.134, over 28924.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3572, pruned_loss=0.1096, over 5707936.32 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1216, over 5750554.55 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3546, pruned_loss=0.1072, over 5699615.72 frames. ], batch size: 112, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:34:23,703 INFO [optim.py:369] (1/2) 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 06:35:06,183 INFO [train.py:968] (1/2) Epoch 6, batch 22150, giga_loss[loss=0.2909, simple_loss=0.3555, pruned_loss=0.1131, over 28939.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3576, pruned_loss=0.1099, over 5701809.25 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3766, pruned_loss=0.1219, over 5752814.64 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3549, pruned_loss=0.1076, over 5692693.35 frames. ], batch size: 106, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:35:44,894 INFO [train.py:968] (1/2) Epoch 6, batch 22200, giga_loss[loss=0.2301, simple_loss=0.3108, pruned_loss=0.07472, over 28559.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.359, pruned_loss=0.1109, over 5713235.80 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3772, pruned_loss=0.1225, over 5756011.38 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3557, pruned_loss=0.1082, over 5701707.95 frames. ], batch size: 71, lr: 5.36e-03, grad_scale: 4.0
+2023-03-03 06:35:45,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7860, 3.5889, 1.7883, 1.6165], device='cuda:1'), covar=tensor([0.0725, 0.0305, 0.0729, 0.1166], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0481, 0.0309, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 06:35:46,144 INFO [optim.py:369] (1/2) 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 06:36:25,295 INFO [train.py:968] (1/2) Epoch 6, batch 22250, giga_loss[loss=0.2905, simple_loss=0.3627, pruned_loss=0.1092, over 28525.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3608, pruned_loss=0.1125, over 5692554.60 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3783, pruned_loss=0.1233, over 5737511.75 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3567, pruned_loss=0.1092, over 5699734.60 frames. ], batch size: 336, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:36:45,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=249983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:37:07,218 INFO [train.py:968] (1/2) Epoch 6, batch 22300, giga_loss[loss=0.2962, simple_loss=0.3662, pruned_loss=0.1131, over 28613.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3631, pruned_loss=0.1139, over 5692748.91 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3784, pruned_loss=0.1234, over 5738450.77 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3595, pruned_loss=0.1111, over 5697035.57 frames. ], batch size: 85, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:37:08,493 INFO [optim.py:369] (1/2) 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 06:37:31,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-03 06:37:45,601 INFO [train.py:968] (1/2) Epoch 6, batch 22350, libri_loss[loss=0.3663, simple_loss=0.4144, pruned_loss=0.1591, over 29093.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3659, pruned_loss=0.1155, over 5686773.38 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3793, pruned_loss=0.1241, over 5725113.19 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3617, pruned_loss=0.1123, over 5700933.42 frames. ], batch size: 101, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:38:26,702 INFO [train.py:968] (1/2) Epoch 6, batch 22400, giga_loss[loss=0.2872, simple_loss=0.3646, pruned_loss=0.105, over 28873.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3668, pruned_loss=0.1153, over 5675976.82 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3798, pruned_loss=0.1244, over 5710082.47 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3627, pruned_loss=0.1122, over 5700822.72 frames. ], batch size: 199, lr: 5.35e-03, grad_scale: 8.0
+2023-03-03 06:38:28,345 INFO [optim.py:369] (1/2) 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 06:38:38,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:38:40,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:39:03,309 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 22450, giga_loss[loss=0.2673, simple_loss=0.348, pruned_loss=0.09329, over 28858.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3673, pruned_loss=0.1152, over 5684172.30 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3801, pruned_loss=0.1246, over 5705103.74 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3635, pruned_loss=0.1124, over 5707187.30 frames. ], batch size: 186, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:39:12,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4729, 1.5902, 1.3863, 1.7877], device='cuda:1'), covar=tensor([0.1672, 0.1576, 0.1456, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.0868, 0.1002, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 06:39:18,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8914, 1.0206, 3.5794, 2.9892], device='cuda:1'), covar=tensor([0.1686, 0.2498, 0.0430, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0587, 0.0540, 0.0767, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 06:39:35,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8954, 2.2032, 2.3797, 2.2587], device='cuda:1'), covar=tensor([0.0876, 0.2045, 0.1314, 0.1455], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0746, 0.0649, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 06:39:44,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:39:47,796 INFO [train.py:968] (1/2) Epoch 6, batch 22500, giga_loss[loss=0.2894, simple_loss=0.3613, pruned_loss=0.1088, over 28697.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3686, pruned_loss=0.1164, over 5693656.49 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3804, pruned_loss=0.1249, over 5708861.84 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3651, pruned_loss=0.1137, over 5708330.89 frames. ], batch size: 242, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:39:50,321 INFO [optim.py:369] (1/2) 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 06:40:08,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5118, 1.7171, 1.4599, 2.0260], device='cuda:1'), covar=tensor([0.1983, 0.1946, 0.1946, 0.1868], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.0869, 0.1001, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 06:40:28,293 INFO [train.py:968] (1/2) Epoch 6, batch 22550, giga_loss[loss=0.2745, simple_loss=0.3535, pruned_loss=0.09781, over 29023.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3682, pruned_loss=0.1164, over 5695684.20 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.381, pruned_loss=0.1255, over 5711876.56 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3647, pruned_loss=0.1136, over 5704279.21 frames. ], batch size: 155, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:40:33,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-03 06:41:12,641 INFO [train.py:968] (1/2) Epoch 6, batch 22600, giga_loss[loss=0.2859, simple_loss=0.3496, pruned_loss=0.1111, over 29005.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3652, pruned_loss=0.1142, over 5706467.95 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3812, pruned_loss=0.1257, over 5715963.98 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3619, pruned_loss=0.1116, over 5709400.88 frames. ], batch size: 106, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:41:15,231 INFO [optim.py:369] (1/2) 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 06:41:15,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4898, 4.3282, 4.0884, 1.8237], device='cuda:1'), covar=tensor([0.0431, 0.0517, 0.0727, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.0880, 0.0829, 0.0776, 0.0601], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:41:42,970 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250348.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 06:41:51,734 INFO [train.py:968] (1/2) Epoch 6, batch 22650, giga_loss[loss=0.2887, simple_loss=0.3581, pruned_loss=0.1096, over 29064.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3623, pruned_loss=0.1131, over 5695694.15 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3817, pruned_loss=0.1263, over 5706414.83 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.359, pruned_loss=0.1103, over 5706554.01 frames. ], batch size: 136, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:41:58,581 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=250369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:42:18,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-03 06:42:31,709 INFO [train.py:968] (1/2) Epoch 6, batch 22700, giga_loss[loss=0.2869, simple_loss=0.3631, pruned_loss=0.1053, over 28596.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3607, pruned_loss=0.1124, over 5688646.49 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3816, pruned_loss=0.1266, over 5695267.67 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3575, pruned_loss=0.1095, over 5706103.82 frames. ], batch size: 262, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:42:34,506 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 22750, giga_loss[loss=0.2608, simple_loss=0.34, pruned_loss=0.09084, over 29009.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3611, pruned_loss=0.1114, over 5691813.09 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3813, pruned_loss=0.1264, over 5702715.06 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.358, pruned_loss=0.1086, over 5699290.92 frames. ], batch size: 136, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:43:32,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0931, 3.9092, 3.7036, 1.8221], device='cuda:1'), covar=tensor([0.0504, 0.0571, 0.0742, 0.2029], device='cuda:1'), in_proj_covar=tensor([0.0881, 0.0825, 0.0777, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 06:43:55,754 INFO [train.py:968] (1/2) Epoch 6, batch 22800, giga_loss[loss=0.247, simple_loss=0.3262, pruned_loss=0.08387, over 29035.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3628, pruned_loss=0.111, over 5694481.61 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.381, pruned_loss=0.1264, over 5704941.72 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.36, pruned_loss=0.1084, over 5698250.28 frames. ], batch size: 128, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:43:58,227 INFO [optim.py:369] (1/2) 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 06:44:17,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-03 06:44:34,000 INFO [train.py:968] (1/2) Epoch 6, batch 22850, giga_loss[loss=0.2654, simple_loss=0.3257, pruned_loss=0.1026, over 28909.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3613, pruned_loss=0.1108, over 5695761.58 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3811, pruned_loss=0.1265, over 5707129.41 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3589, pruned_loss=0.1086, over 5696748.78 frames. ], batch size: 106, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:44:34,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4244, 3.3687, 1.6074, 1.4570], device='cuda:1'), covar=tensor([0.0833, 0.0302, 0.0831, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0488, 0.0311, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 06:44:51,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:45:14,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1958, 1.5448, 1.2544, 1.0032], device='cuda:1'), covar=tensor([0.1930, 0.1870, 0.2075, 0.1808], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.0874, 0.1008, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 06:45:16,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 06:45:16,397 INFO [train.py:968] (1/2) Epoch 6, batch 22900, giga_loss[loss=0.2764, simple_loss=0.3453, pruned_loss=0.1037, over 28767.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3609, pruned_loss=0.1125, over 5697042.28 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3818, pruned_loss=0.1271, over 5706834.51 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3581, pruned_loss=0.11, over 5698144.13 frames. ], batch size: 284, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:45:20,706 INFO [optim.py:369] (1/2) 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 06:45:58,451 INFO [train.py:968] (1/2) Epoch 6, batch 22950, giga_loss[loss=0.2777, simple_loss=0.3481, pruned_loss=0.1037, over 28762.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3587, pruned_loss=0.1124, over 5699804.95 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3821, pruned_loss=0.1273, over 5704253.07 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3557, pruned_loss=0.1099, over 5703091.13 frames. ], batch size: 243, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:46:37,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3327, 1.4411, 1.4507, 1.4499], device='cuda:1'), covar=tensor([0.1013, 0.1273, 0.1505, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0744, 0.0642, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 06:46:38,210 INFO [train.py:968] (1/2) Epoch 6, batch 23000, giga_loss[loss=0.3049, simple_loss=0.3537, pruned_loss=0.1281, over 28790.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3565, pruned_loss=0.1124, over 5706508.42 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3818, pruned_loss=0.1273, over 5708217.79 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3539, pruned_loss=0.1101, over 5705617.96 frames. ], batch size: 99, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:46:43,118 INFO [optim.py:369] (1/2) 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 06:46:49,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=250723.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 06:46:51,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:46:54,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:47:05,773 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:47:18,790 INFO [train.py:968] (1/2) Epoch 6, batch 23050, giga_loss[loss=0.2808, simple_loss=0.3477, pruned_loss=0.107, over 28996.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3594, pruned_loss=0.1146, over 5711996.31 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3828, pruned_loss=0.1282, over 5711220.61 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3555, pruned_loss=0.1114, over 5708495.38 frames. ], batch size: 136, lr: 5.35e-03, grad_scale: 4.0
+2023-03-03 06:47:56,489 INFO [train.py:968] (1/2) Epoch 6, batch 23100, giga_loss[loss=0.2622, simple_loss=0.3283, pruned_loss=0.09804, over 28910.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3556, pruned_loss=0.1122, over 5713439.71 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3833, pruned_loss=0.1287, over 5711119.88 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3515, pruned_loss=0.1089, over 5710969.01 frames. ], batch size: 186, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:48:01,385 INFO [optim.py:369] (1/2) 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 06:48:35,830 INFO [train.py:968] (1/2) Epoch 6, batch 23150, giga_loss[loss=0.3185, simple_loss=0.3681, pruned_loss=0.1344, over 28679.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3525, pruned_loss=0.111, over 5696325.70 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3838, pruned_loss=0.1292, over 5699131.07 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3479, pruned_loss=0.1074, over 5704626.75 frames. ], batch size: 262, lr: 5.35e-03, grad_scale: 2.0
+2023-03-03 06:48:39,504 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250866.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 06:48:41,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250869.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 06:48:48,904 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 06:48:56,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=250887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:48:58,247 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=250890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:49:04,500 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250898.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 06:49:12,535 INFO [train.py:968] (1/2) Epoch 6, batch 23200, giga_loss[loss=0.2905, simple_loss=0.3633, pruned_loss=0.1089, over 28650.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3493, pruned_loss=0.1093, over 5697729.41 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.384, pruned_loss=0.1296, over 5695348.50 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3445, pruned_loss=0.1054, over 5707925.92 frames. ], batch size: 262, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:49:16,098 INFO [optim.py:369] (1/2) 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 06:49:18,278 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=250919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:49:52,078 INFO [train.py:968] (1/2) Epoch 6, batch 23250, giga_loss[loss=0.289, simple_loss=0.3618, pruned_loss=0.1081, over 28711.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3504, pruned_loss=0.1097, over 5702404.07 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.384, pruned_loss=0.1299, over 5698683.27 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3458, pruned_loss=0.1059, over 5707517.18 frames. ], batch size: 307, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:50:34,704 INFO [train.py:968] (1/2) Epoch 6, batch 23300, giga_loss[loss=0.2816, simple_loss=0.3495, pruned_loss=0.1069, over 28785.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3531, pruned_loss=0.1103, over 5697955.80 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3845, pruned_loss=0.1302, over 5692529.62 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3483, pruned_loss=0.1066, over 5708371.59 frames. ], batch size: 99, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:50:38,007 INFO [optim.py:369] (1/2) 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 06:51:09,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 06:51:11,613 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:51:17,077 INFO [train.py:968] (1/2) Epoch 6, batch 23350, giga_loss[loss=0.2812, simple_loss=0.3567, pruned_loss=0.1029, over 29037.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3576, pruned_loss=0.1124, over 5700435.98 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3849, pruned_loss=0.1306, over 5694377.84 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3526, pruned_loss=0.1086, over 5707262.07 frames. ], batch size: 174, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:51:27,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7253, 1.0885, 2.7709, 2.6002], device='cuda:1'), covar=tensor([0.1556, 0.2143, 0.0537, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0535, 0.0768, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 06:51:32,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8019, 2.1474, 1.4777, 1.4647], device='cuda:1'), covar=tensor([0.1042, 0.0678, 0.0770, 0.0784], device='cuda:1'), in_proj_covar=tensor([0.1483, 0.1297, 0.1270, 0.1369], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 06:51:56,107 INFO [train.py:968] (1/2) Epoch 6, batch 23400, giga_loss[loss=0.2953, simple_loss=0.3696, pruned_loss=0.1105, over 28929.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1138, over 5703149.39 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3852, pruned_loss=0.1307, over 5699634.49 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3565, pruned_loss=0.1103, over 5704146.38 frames. ], batch size: 145, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:52:01,782 INFO [optim.py:369] (1/2) 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 06:52:38,743 INFO [train.py:968] (1/2) Epoch 6, batch 23450, giga_loss[loss=0.3467, simple_loss=0.3973, pruned_loss=0.148, over 26637.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3644, pruned_loss=0.1153, over 5702142.57 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3847, pruned_loss=0.1304, over 5707181.54 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3599, pruned_loss=0.1121, over 5696306.61 frames. ], batch size: 555, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:53:21,627 INFO [train.py:968] (1/2) Epoch 6, batch 23500, giga_loss[loss=0.3077, simple_loss=0.3658, pruned_loss=0.1248, over 28760.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.368, pruned_loss=0.1183, over 5696438.32 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3853, pruned_loss=0.1309, over 5700993.43 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3638, pruned_loss=0.1151, over 5696444.00 frames. ], batch size: 92, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:53:30,061 INFO [optim.py:369] (1/2) 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 06:53:38,619 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:53:48,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:54:14,106 INFO [train.py:968] (1/2) Epoch 6, batch 23550, giga_loss[loss=0.3231, simple_loss=0.3937, pruned_loss=0.1263, over 28946.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3748, pruned_loss=0.1248, over 5684617.01 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3852, pruned_loss=0.1309, over 5702199.25 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3713, pruned_loss=0.1221, over 5683465.12 frames. ], batch size: 136, lr: 5.34e-03, grad_scale: 2.0
+2023-03-03 06:54:58,469 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-03 06:55:08,285 INFO [train.py:968] (1/2) Epoch 6, batch 23600, giga_loss[loss=0.3569, simple_loss=0.4073, pruned_loss=0.1532, over 27927.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3815, pruned_loss=0.1299, over 5684377.48 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3853, pruned_loss=0.131, over 5699834.26 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3787, pruned_loss=0.1278, over 5685648.31 frames. ], batch size: 412, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:55:12,600 INFO [optim.py:369] (1/2) 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 06:55:39,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0335, 1.2419, 4.0243, 3.1890], device='cuda:1'), covar=tensor([0.1725, 0.2282, 0.0366, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0584, 0.0537, 0.0767, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 06:55:46,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 06:55:58,013 INFO [train.py:968] (1/2) Epoch 6, batch 23650, giga_loss[loss=0.3193, simple_loss=0.3804, pruned_loss=0.1291, over 28191.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3878, pruned_loss=0.1354, over 5679479.70 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3853, pruned_loss=0.131, over 5702425.85 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3855, pruned_loss=0.1337, over 5678054.82 frames. ], batch size: 77, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:56:46,917 INFO [train.py:968] (1/2) Epoch 6, batch 23700, giga_loss[loss=0.3667, simple_loss=0.4229, pruned_loss=0.1553, over 28580.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3959, pruned_loss=0.1432, over 5671482.06 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3859, pruned_loss=0.1314, over 5702902.16 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3937, pruned_loss=0.1417, over 5669203.35 frames. ], batch size: 78, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:56:53,049 INFO [optim.py:369] (1/2) 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 06:57:04,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:57:31,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 06:57:34,934 INFO [train.py:968] (1/2) Epoch 6, batch 23750, giga_loss[loss=0.3763, simple_loss=0.4247, pruned_loss=0.1639, over 28823.00 frames. ], tot_loss[loss=0.3492, simple_loss=0.4021, pruned_loss=0.1482, over 5651862.49 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3864, pruned_loss=0.1319, over 5694040.32 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.4001, pruned_loss=0.1469, over 5656126.64 frames. ], batch size: 199, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:57:59,663 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-03 06:58:15,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3575, 1.5659, 1.2747, 1.4811], device='cuda:1'), covar=tensor([0.1955, 0.1808, 0.1822, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.1139, 0.0879, 0.1005, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 06:58:24,278 INFO [train.py:968] (1/2) Epoch 6, batch 23800, giga_loss[loss=0.3991, simple_loss=0.4306, pruned_loss=0.1838, over 28226.00 frames. ], tot_loss[loss=0.3507, simple_loss=0.4029, pruned_loss=0.1492, over 5659016.85 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3862, pruned_loss=0.1318, over 5698006.38 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4019, pruned_loss=0.1485, over 5658238.17 frames. ], batch size: 368, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:58:29,951 INFO [optim.py:369] (1/2) 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 06:58:55,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2228, 1.6330, 1.2602, 1.5173], device='cuda:1'), covar=tensor([0.0683, 0.0375, 0.0326, 0.0716], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0119, 0.0122, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0062, 0.0045, 0.0041, 0.0068], device='cuda:1')
+2023-03-03 06:59:12,396 INFO [train.py:968] (1/2) Epoch 6, batch 23850, giga_loss[loss=0.4866, simple_loss=0.4865, pruned_loss=0.2434, over 28285.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4038, pruned_loss=0.151, over 5665912.58 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3858, pruned_loss=0.1318, over 5702891.09 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.4039, pruned_loss=0.151, over 5659540.35 frames. ], batch size: 368, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 06:59:26,199 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:59:29,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:59:43,591 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 06:59:53,198 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:00:03,055 INFO [train.py:968] (1/2) Epoch 6, batch 23900, giga_loss[loss=0.3795, simple_loss=0.4235, pruned_loss=0.1678, over 28884.00 frames. ], tot_loss[loss=0.3575, simple_loss=0.4065, pruned_loss=0.1543, over 5647978.81 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3858, pruned_loss=0.1319, over 5704389.85 frames. ], giga_tot_loss[loss=0.358, simple_loss=0.4068, pruned_loss=0.1546, over 5640827.27 frames. ], batch size: 199, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 07:00:06,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:00:09,792 INFO [optim.py:369] (1/2) 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 07:00:40,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5432, 1.6021, 1.3884, 1.8136], device='cuda:1'), covar=tensor([0.1808, 0.1709, 0.1677, 0.1737], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.0877, 0.1008, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0008, 0.0007], device='cuda:1')
+2023-03-03 07:01:01,783 INFO [train.py:968] (1/2) Epoch 6, batch 23950, giga_loss[loss=0.4611, simple_loss=0.4716, pruned_loss=0.2253, over 27533.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4115, pruned_loss=0.1592, over 5636519.01 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3858, pruned_loss=0.132, over 5708384.02 frames. ], giga_tot_loss[loss=0.366, simple_loss=0.4123, pruned_loss=0.1599, over 5626198.49 frames. ], batch size: 472, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 07:01:36,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:01:45,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 07:01:56,221 INFO [train.py:968] (1/2) Epoch 6, batch 24000, giga_loss[loss=0.336, simple_loss=0.3865, pruned_loss=0.1427, over 28672.00 frames. ], tot_loss[loss=0.3659, simple_loss=0.4114, pruned_loss=0.1602, over 5618999.06 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3857, pruned_loss=0.1321, over 5711611.90 frames. ], giga_tot_loss[loss=0.3674, simple_loss=0.4126, pruned_loss=0.1611, over 5606228.43 frames. ], batch size: 92, lr: 5.34e-03, grad_scale: 8.0
+2023-03-03 07:01:56,221 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 07:02:04,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0967, 1.7606, 1.4704, 1.2344], device='cuda:1'), covar=tensor([0.1363, 0.2219, 0.1298, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0723, 0.0791, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 07:02:04,973 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 07:02:10,027 INFO [optim.py:369] (1/2) 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 07:02:38,357 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:41,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:47,376 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=251756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:50,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=251759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:02:51,231 INFO [train.py:968] (1/2) Epoch 6, batch 24050, giga_loss[loss=0.2927, simple_loss=0.353, pruned_loss=0.1163, over 28720.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.4093, pruned_loss=0.1596, over 5628089.78 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3854, pruned_loss=0.1321, over 5706401.03 frames. ], giga_tot_loss[loss=0.3665, simple_loss=0.4111, pruned_loss=0.1609, over 5621113.35 frames. ], batch size: 99, lr: 5.34e-03, grad_scale: 4.0
+2023-03-03 07:03:06,458 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:03:17,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=251788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:03:23,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=251797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:03:31,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8272, 3.6805, 3.4817, 1.6246], device='cuda:1'), covar=tensor([0.0599, 0.0668, 0.0813, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0865, 0.0810, 0.0623], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:03:38,390 INFO [train.py:968] (1/2) Epoch 6, batch 24100, giga_loss[loss=0.4886, simple_loss=0.4809, pruned_loss=0.2482, over 26503.00 frames. ], tot_loss[loss=0.3601, simple_loss=0.4065, pruned_loss=0.1568, over 5637295.96 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.3854, pruned_loss=0.1323, over 5711739.02 frames. ], giga_tot_loss[loss=0.3625, simple_loss=0.4084, pruned_loss=0.1583, over 5625346.34 frames. ], batch size: 555, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:03:43,873 INFO [optim.py:369] (1/2) 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 07:04:05,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9261, 4.7600, 4.4770, 2.2169], device='cuda:1'), covar=tensor([0.0355, 0.0464, 0.0634, 0.1932], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0861, 0.0808, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:04:24,562 INFO [train.py:968] (1/2) Epoch 6, batch 24150, giga_loss[loss=0.3414, simple_loss=0.3995, pruned_loss=0.1417, over 28220.00 frames. ], tot_loss[loss=0.3603, simple_loss=0.4073, pruned_loss=0.1566, over 5620377.88 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.386, pruned_loss=0.1328, over 5707538.07 frames. ], giga_tot_loss[loss=0.3624, simple_loss=0.4089, pruned_loss=0.158, over 5611776.56 frames. ], batch size: 368, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:05:07,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2565, 1.5271, 1.1552, 1.6668], device='cuda:1'), covar=tensor([0.2127, 0.1924, 0.2020, 0.1695], device='cuda:1'), in_proj_covar=tensor([0.1143, 0.0886, 0.1015, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 07:05:14,634 INFO [train.py:968] (1/2) Epoch 6, batch 24200, giga_loss[loss=0.3687, simple_loss=0.3985, pruned_loss=0.1694, over 23426.00 frames. ], tot_loss[loss=0.362, simple_loss=0.4094, pruned_loss=0.1573, over 5621857.16 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3861, pruned_loss=0.133, over 5707837.84 frames. ], giga_tot_loss[loss=0.3642, simple_loss=0.411, pruned_loss=0.1587, over 5613019.32 frames. ], batch size: 705, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:05:19,677 INFO [optim.py:369] (1/2) 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 07:05:41,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9815, 1.1947, 1.0145, 0.8502], device='cuda:1'), covar=tensor([0.0826, 0.0848, 0.0560, 0.0708], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1323, 0.1285, 0.1386], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 07:06:03,668 INFO [train.py:968] (1/2) Epoch 6, batch 24250, libri_loss[loss=0.2913, simple_loss=0.3606, pruned_loss=0.111, over 29528.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4078, pruned_loss=0.1558, over 5632636.75 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3856, pruned_loss=0.1329, over 5715382.40 frames. ], giga_tot_loss[loss=0.3631, simple_loss=0.4104, pruned_loss=0.1579, over 5616061.73 frames. ], batch size: 79, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:06:05,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-03 07:06:07,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=251966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:06:54,117 INFO [train.py:968] (1/2) Epoch 6, batch 24300, giga_loss[loss=0.3606, simple_loss=0.4158, pruned_loss=0.1527, over 28248.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4053, pruned_loss=0.1531, over 5629344.65 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3864, pruned_loss=0.1337, over 5715068.45 frames. ], giga_tot_loss[loss=0.3579, simple_loss=0.407, pruned_loss=0.1545, over 5614949.97 frames. ], batch size: 368, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:07:00,411 INFO [optim.py:369] (1/2) 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 07:07:01,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8651, 1.8386, 1.3358, 1.5128], device='cuda:1'), covar=tensor([0.0627, 0.0547, 0.0963, 0.0895], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0449, 0.0500, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:07:16,583 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 24350, giga_loss[loss=0.3146, simple_loss=0.38, pruned_loss=0.1246, over 28673.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4033, pruned_loss=0.1501, over 5636303.77 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3867, pruned_loss=0.134, over 5713768.49 frames. ], giga_tot_loss[loss=0.3538, simple_loss=0.4049, pruned_loss=0.1514, over 5623816.51 frames. ], batch size: 242, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:07:48,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:08:22,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:08:28,430 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 24400, libri_loss[loss=0.2687, simple_loss=0.3294, pruned_loss=0.104, over 29659.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.4002, pruned_loss=0.1478, over 5635633.84 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3859, pruned_loss=0.1336, over 5717827.04 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4025, pruned_loss=0.1495, over 5620504.33 frames. ], batch size: 69, lr: 5.33e-03, grad_scale: 8.0
+2023-03-03 07:08:30,489 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:08:59,605 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 24450, giga_loss[loss=0.2934, simple_loss=0.3633, pruned_loss=0.1118, over 29003.00 frames. ], tot_loss[loss=0.3444, simple_loss=0.3973, pruned_loss=0.1457, over 5633230.05 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3859, pruned_loss=0.1338, over 5710268.32 frames. ], giga_tot_loss[loss=0.3467, simple_loss=0.3993, pruned_loss=0.147, over 5626764.15 frames. ], batch size: 155, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:09:31,368 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 24500, giga_loss[loss=0.3209, simple_loss=0.3766, pruned_loss=0.1326, over 28565.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3975, pruned_loss=0.146, over 5635554.08 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.386, pruned_loss=0.1339, over 5713930.28 frames. ], giga_tot_loss[loss=0.3469, simple_loss=0.3992, pruned_loss=0.1473, over 5625065.24 frames. ], batch size: 92, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:10:08,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:10:12,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:10:15,306 INFO [optim.py:369] (1/2) 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 07:10:44,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:11:02,625 INFO [train.py:968] (1/2) Epoch 6, batch 24550, giga_loss[loss=0.3753, simple_loss=0.4228, pruned_loss=0.1639, over 28599.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3963, pruned_loss=0.1448, over 5637641.64 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3858, pruned_loss=0.1338, over 5715665.59 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.398, pruned_loss=0.146, over 5626963.28 frames. ], batch size: 85, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:11:04,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7994, 1.9329, 1.6220, 1.7864], device='cuda:1'), covar=tensor([0.1191, 0.1691, 0.1694, 0.1470], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0743, 0.0650, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 07:11:16,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6321, 1.8244, 1.5212, 1.5973], device='cuda:1'), covar=tensor([0.0687, 0.0281, 0.0297, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0119, 0.0123, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0062, 0.0046, 0.0041, 0.0068], device='cuda:1')
+2023-03-03 07:11:52,948 INFO [train.py:968] (1/2) Epoch 6, batch 24600, giga_loss[loss=0.3385, simple_loss=0.4066, pruned_loss=0.1352, over 28788.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3946, pruned_loss=0.1427, over 5654886.38 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3858, pruned_loss=0.1339, over 5717519.04 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.396, pruned_loss=0.1437, over 5643803.18 frames. ], batch size: 243, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:11:58,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:12:03,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:12:04,681 INFO [optim.py:369] (1/2) 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 07:12:30,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:12:43,317 INFO [train.py:968] (1/2) Epoch 6, batch 24650, libri_loss[loss=0.3076, simple_loss=0.3713, pruned_loss=0.1219, over 29532.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3942, pruned_loss=0.1403, over 5662609.18 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3851, pruned_loss=0.1336, over 5724450.01 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3964, pruned_loss=0.1417, over 5644240.29 frames. ], batch size: 82, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:13:04,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6230, 2.0403, 1.9634, 1.6683], device='cuda:1'), covar=tensor([0.1659, 0.1769, 0.1222, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0727, 0.0798, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 07:13:32,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:13:36,654 INFO [train.py:968] (1/2) Epoch 6, batch 24700, giga_loss[loss=0.3693, simple_loss=0.3957, pruned_loss=0.1714, over 23530.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3966, pruned_loss=0.1403, over 5665316.13 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3854, pruned_loss=0.134, over 5724592.10 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3982, pruned_loss=0.1413, over 5649539.04 frames. ], batch size: 705, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:13:45,051 INFO [optim.py:369] (1/2) 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 07:13:48,186 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.97 vs. limit=5.0
+2023-03-03 07:14:30,476 INFO [train.py:968] (1/2) Epoch 6, batch 24750, giga_loss[loss=0.3461, simple_loss=0.4022, pruned_loss=0.145, over 28932.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3969, pruned_loss=0.1413, over 5666858.27 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3855, pruned_loss=0.1341, over 5725626.14 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3981, pruned_loss=0.142, over 5653455.11 frames. ], batch size: 145, lr: 5.33e-03, grad_scale: 2.0
+2023-03-03 07:14:44,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=252477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:15:14,111 INFO [train.py:968] (1/2) Epoch 6, batch 24800, libri_loss[loss=0.3766, simple_loss=0.4143, pruned_loss=0.1695, over 18397.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3963, pruned_loss=0.141, over 5672711.46 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.385, pruned_loss=0.134, over 5718534.06 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3981, pruned_loss=0.1418, over 5667841.53 frames. ], batch size: 187, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:15:24,566 INFO [optim.py:369] (1/2) 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 07:15:37,692 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 07:15:39,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1839, 1.9354, 1.4327, 1.7441], device='cuda:1'), covar=tensor([0.0652, 0.0754, 0.0987, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0453, 0.0502, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:15:54,737 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:15:58,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:16:04,293 INFO [train.py:968] (1/2) Epoch 6, batch 24850, giga_loss[loss=0.3872, simple_loss=0.4143, pruned_loss=0.1801, over 26702.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.394, pruned_loss=0.1401, over 5680934.88 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3852, pruned_loss=0.134, over 5722280.19 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3955, pruned_loss=0.1408, over 5673198.39 frames. ], batch size: 555, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:16:25,901 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:16:37,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8438, 1.0068, 3.3795, 2.8767], device='cuda:1'), covar=tensor([0.1664, 0.2403, 0.0477, 0.0736], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0545, 0.0777, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 07:16:47,604 INFO [train.py:968] (1/2) Epoch 6, batch 24900, giga_loss[loss=0.3663, simple_loss=0.417, pruned_loss=0.1577, over 28338.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3919, pruned_loss=0.1394, over 5672272.30 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3855, pruned_loss=0.1343, over 5714930.01 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3929, pruned_loss=0.1398, over 5672439.22 frames. ], batch size: 368, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:16:56,158 INFO [optim.py:369] (1/2) 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 07:16:58,071 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=252620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:17:00,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=252623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:17:17,417 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-03 07:17:26,698 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=252652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:17:34,369 INFO [train.py:968] (1/2) Epoch 6, batch 24950, giga_loss[loss=0.2864, simple_loss=0.3678, pruned_loss=0.1025, over 28812.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3919, pruned_loss=0.1386, over 5674009.97 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3856, pruned_loss=0.1344, over 5716040.36 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3926, pruned_loss=0.1389, over 5673094.30 frames. ], batch size: 119, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:18:24,583 INFO [train.py:968] (1/2) Epoch 6, batch 25000, giga_loss[loss=0.3089, simple_loss=0.3803, pruned_loss=0.1188, over 28877.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3903, pruned_loss=0.136, over 5683215.92 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3856, pruned_loss=0.1344, over 5717008.16 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.391, pruned_loss=0.1362, over 5681561.70 frames. ], batch size: 186, lr: 5.33e-03, grad_scale: 4.0
+2023-03-03 07:18:33,043 INFO [optim.py:369] (1/2) 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 07:19:08,126 INFO [train.py:968] (1/2) Epoch 6, batch 25050, giga_loss[loss=0.3325, simple_loss=0.3939, pruned_loss=0.1356, over 28628.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3902, pruned_loss=0.1363, over 5671626.24 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3856, pruned_loss=0.1346, over 5711165.62 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3908, pruned_loss=0.1364, over 5674977.77 frames. ], batch size: 307, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:19:56,988 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 25100, giga_loss[loss=0.323, simple_loss=0.3966, pruned_loss=0.1247, over 28961.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3897, pruned_loss=0.1364, over 5670394.58 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3859, pruned_loss=0.135, over 5703907.84 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3901, pruned_loss=0.1362, over 5677934.68 frames. ], batch size: 164, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:20:05,376 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:20:16,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3522, 3.1985, 1.4397, 1.3995], device='cuda:1'), covar=tensor([0.0910, 0.0305, 0.0831, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0493, 0.0316, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 07:20:18,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4598, 4.3106, 4.0974, 1.7547], device='cuda:1'), covar=tensor([0.0441, 0.0520, 0.0661, 0.2068], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0863, 0.0809, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:20:44,659 INFO [train.py:968] (1/2) Epoch 6, batch 25150, giga_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 29105.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3899, pruned_loss=0.1378, over 5657761.95 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3863, pruned_loss=0.1353, over 5704981.69 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3898, pruned_loss=0.1373, over 5662007.18 frames. ], batch size: 128, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:21:31,690 INFO [train.py:968] (1/2) Epoch 6, batch 25200, giga_loss[loss=0.3159, simple_loss=0.3811, pruned_loss=0.1254, over 29003.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3894, pruned_loss=0.1383, over 5664252.21 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3861, pruned_loss=0.1353, over 5707406.19 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3896, pruned_loss=0.138, over 5664594.93 frames. ], batch size: 136, lr: 5.32e-03, grad_scale: 8.0
+2023-03-03 07:21:40,964 INFO [optim.py:369] (1/2) 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 07:22:00,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1857, 1.4651, 1.2933, 1.0001], device='cuda:1'), covar=tensor([0.2018, 0.1951, 0.2001, 0.1615], device='cuda:1'), in_proj_covar=tensor([0.1145, 0.0884, 0.1016, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 07:22:19,287 INFO [train.py:968] (1/2) Epoch 6, batch 25250, giga_loss[loss=0.3785, simple_loss=0.417, pruned_loss=0.17, over 26677.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3892, pruned_loss=0.139, over 5665611.92 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3864, pruned_loss=0.1355, over 5710605.13 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3892, pruned_loss=0.1387, over 5662614.28 frames. ], batch size: 555, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:22:43,730 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=252985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:23:02,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:23:06,939 INFO [train.py:968] (1/2) Epoch 6, batch 25300, giga_loss[loss=0.3002, simple_loss=0.3659, pruned_loss=0.1172, over 28800.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3875, pruned_loss=0.1378, over 5672912.24 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3863, pruned_loss=0.1354, over 5714557.20 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3876, pruned_loss=0.1376, over 5666318.63 frames. ], batch size: 99, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:23:15,349 INFO [optim.py:369] (1/2) 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 07:23:56,271 INFO [train.py:968] (1/2) Epoch 6, batch 25350, giga_loss[loss=0.305, simple_loss=0.3684, pruned_loss=0.1208, over 29019.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3883, pruned_loss=0.1389, over 5665794.11 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.387, pruned_loss=0.1358, over 5717958.21 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3878, pruned_loss=0.1384, over 5656552.29 frames. ], batch size: 155, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:23:56,504 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:24:45,313 INFO [train.py:968] (1/2) Epoch 6, batch 25400, giga_loss[loss=0.3672, simple_loss=0.3887, pruned_loss=0.1728, over 23597.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3871, pruned_loss=0.1376, over 5667727.84 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3868, pruned_loss=0.1358, over 5720857.53 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3869, pruned_loss=0.1373, over 5657161.84 frames. ], batch size: 705, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:24:54,552 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 25450, giga_loss[loss=0.2857, simple_loss=0.3647, pruned_loss=0.1033, over 29034.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.387, pruned_loss=0.1363, over 5673994.49 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3865, pruned_loss=0.1358, over 5726087.84 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3872, pruned_loss=0.1362, over 5659246.01 frames. ], batch size: 155, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:25:45,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7336, 1.7526, 1.7060, 1.6375], device='cuda:1'), covar=tensor([0.1185, 0.2045, 0.1530, 0.1600], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0743, 0.0648, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 07:25:50,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:26:04,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:26:11,636 INFO [train.py:968] (1/2) Epoch 6, batch 25500, giga_loss[loss=0.2873, simple_loss=0.3622, pruned_loss=0.1062, over 28810.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3871, pruned_loss=0.1358, over 5676081.80 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3868, pruned_loss=0.1361, over 5731164.66 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3869, pruned_loss=0.1354, over 5658396.90 frames. ], batch size: 66, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:26:20,680 INFO [optim.py:369] (1/2) 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 07:26:48,821 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 25550, giga_loss[loss=0.3313, simple_loss=0.393, pruned_loss=0.1348, over 28919.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3875, pruned_loss=0.1364, over 5655987.83 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.387, pruned_loss=0.1364, over 5713315.70 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3872, pruned_loss=0.1358, over 5657073.46 frames. ], batch size: 164, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:27:46,140 INFO [train.py:968] (1/2) Epoch 6, batch 25600, libri_loss[loss=0.3734, simple_loss=0.4241, pruned_loss=0.1614, over 25809.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3907, pruned_loss=0.1395, over 5653786.57 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3872, pruned_loss=0.1365, over 5713421.57 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3902, pruned_loss=0.139, over 5653853.74 frames. ], batch size: 136, lr: 5.32e-03, grad_scale: 8.0
+2023-03-03 07:27:56,396 INFO [optim.py:369] (1/2) 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 07:28:04,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:06,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:18,870 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:21,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:36,637 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:36,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:37,008 INFO [train.py:968] (1/2) Epoch 6, batch 25650, giga_loss[loss=0.2862, simple_loss=0.3513, pruned_loss=0.1105, over 28926.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3926, pruned_loss=0.1425, over 5640217.33 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3871, pruned_loss=0.1364, over 5705534.15 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3925, pruned_loss=0.1422, over 5646616.47 frames. ], batch size: 227, lr: 5.32e-03, grad_scale: 4.0
+2023-03-03 07:28:43,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9487, 3.7657, 3.5858, 1.7572], device='cuda:1'), covar=tensor([0.0576, 0.0708, 0.0812, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0871, 0.0817, 0.0623], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:28:52,311 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:55,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:28:59,593 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:29:28,458 INFO [train.py:968] (1/2) Epoch 6, batch 25700, giga_loss[loss=0.3787, simple_loss=0.4199, pruned_loss=0.1687, over 28921.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3934, pruned_loss=0.1441, over 5655002.64 frames. ], libri_tot_loss[loss=0.3298, simple_loss=0.3868, pruned_loss=0.1364, over 5702058.58 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3937, pruned_loss=0.144, over 5661091.52 frames. ], batch size: 112, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:29:39,435 INFO [optim.py:369] (1/2) 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 07:29:55,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:30:19,011 INFO [train.py:968] (1/2) Epoch 6, batch 25750, giga_loss[loss=0.3418, simple_loss=0.3936, pruned_loss=0.145, over 28013.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.3954, pruned_loss=0.1465, over 5645496.01 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3869, pruned_loss=0.1365, over 5706127.13 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3956, pruned_loss=0.1465, over 5645677.88 frames. ], batch size: 412, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:30:55,012 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:30:56,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:02,694 INFO [train.py:968] (1/2) Epoch 6, batch 25800, giga_loss[loss=0.3094, simple_loss=0.3832, pruned_loss=0.1178, over 28977.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3946, pruned_loss=0.1458, over 5658345.70 frames. ], libri_tot_loss[loss=0.3303, simple_loss=0.3871, pruned_loss=0.1368, over 5711159.54 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3949, pruned_loss=0.1459, over 5652345.38 frames. ], batch size: 164, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:31:13,923 INFO [optim.py:369] (1/2) 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 07:31:15,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:15,754 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-03 07:31:19,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:29,104 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:47,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:31:51,281 INFO [train.py:968] (1/2) Epoch 6, batch 25850, giga_loss[loss=0.3219, simple_loss=0.3897, pruned_loss=0.127, over 28825.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3941, pruned_loss=0.1451, over 5656776.65 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3872, pruned_loss=0.1368, over 5711442.63 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3944, pruned_loss=0.1452, over 5651211.55 frames. ], batch size: 119, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:32:07,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:32:08,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5119, 2.1325, 1.5982, 0.7154], device='cuda:1'), covar=tensor([0.2671, 0.1771, 0.2074, 0.3095], device='cuda:1'), in_proj_covar=tensor([0.1469, 0.1369, 0.1410, 0.1198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 07:32:10,571 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:32:24,143 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 07:32:35,440 INFO [train.py:968] (1/2) Epoch 6, batch 25900, giga_loss[loss=0.2972, simple_loss=0.3636, pruned_loss=0.1154, over 28681.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3912, pruned_loss=0.1411, over 5671357.84 frames. ], libri_tot_loss[loss=0.3303, simple_loss=0.3871, pruned_loss=0.1368, over 5714947.20 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3916, pruned_loss=0.1413, over 5662881.47 frames. ], batch size: 262, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:32:36,228 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:32:53,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:33:20,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 07:33:20,866 INFO [train.py:968] (1/2) Epoch 6, batch 25950, giga_loss[loss=0.2941, simple_loss=0.3664, pruned_loss=0.1109, over 28372.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3887, pruned_loss=0.1395, over 5667588.59 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3875, pruned_loss=0.1372, over 5719336.64 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3887, pruned_loss=0.1394, over 5655051.54 frames. ], batch size: 65, lr: 5.32e-03, grad_scale: 2.0
+2023-03-03 07:33:33,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-03 07:34:08,325 INFO [train.py:968] (1/2) Epoch 6, batch 26000, giga_loss[loss=0.3203, simple_loss=0.379, pruned_loss=0.1308, over 28912.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3854, pruned_loss=0.1376, over 5670277.92 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3873, pruned_loss=0.1369, over 5722108.38 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3856, pruned_loss=0.1377, over 5656914.98 frames. ], batch size: 227, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:34:19,584 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:34:19,963 INFO [optim.py:369] (1/2) 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 07:34:52,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-03 07:35:00,925 INFO [train.py:968] (1/2) Epoch 6, batch 26050, giga_loss[loss=0.3092, simple_loss=0.3749, pruned_loss=0.1218, over 28711.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3838, pruned_loss=0.1361, over 5683555.63 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3872, pruned_loss=0.1368, over 5722668.65 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.384, pruned_loss=0.1364, over 5671950.31 frames. ], batch size: 262, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:35:01,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=253761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:06,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4453, 2.1737, 1.5839, 0.6794], device='cuda:1'), covar=tensor([0.2622, 0.1482, 0.2267, 0.2865], device='cuda:1'), in_proj_covar=tensor([0.1449, 0.1351, 0.1392, 0.1185], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 07:35:06,637 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-03 07:35:08,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:10,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:13,150 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:39,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:46,601 INFO [train.py:968] (1/2) Epoch 6, batch 26100, giga_loss[loss=0.2811, simple_loss=0.3556, pruned_loss=0.1033, over 29077.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3872, pruned_loss=0.138, over 5685958.46 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3874, pruned_loss=0.137, over 5724672.90 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3871, pruned_loss=0.1381, over 5674065.77 frames. ], batch size: 155, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:35:47,563 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:35:58,947 INFO [optim.py:369] (1/2) 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 07:36:21,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4878, 1.5073, 1.3318, 1.6862], device='cuda:1'), covar=tensor([0.2262, 0.2236, 0.2354, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.0882, 0.1017, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 07:36:32,655 INFO [train.py:968] (1/2) Epoch 6, batch 26150, giga_loss[loss=0.3221, simple_loss=0.4031, pruned_loss=0.1205, over 28955.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3907, pruned_loss=0.1375, over 5686995.35 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3878, pruned_loss=0.1372, over 5729008.88 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3904, pruned_loss=0.1374, over 5672711.71 frames. ], batch size: 164, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:37:12,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=253904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:37:14,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=253907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:37:18,813 INFO [train.py:968] (1/2) Epoch 6, batch 26200, giga_loss[loss=0.3255, simple_loss=0.3878, pruned_loss=0.1316, over 28476.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3924, pruned_loss=0.1369, over 5690979.44 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3876, pruned_loss=0.1372, over 5731074.19 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3924, pruned_loss=0.1368, over 5677042.41 frames. ], batch size: 85, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:37:21,672 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:37:42,152 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=253936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:38:05,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=253960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:38:06,304 INFO [train.py:968] (1/2) Epoch 6, batch 26250, giga_loss[loss=0.3911, simple_loss=0.4342, pruned_loss=0.174, over 28742.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3943, pruned_loss=0.1389, over 5689871.71 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3877, pruned_loss=0.1375, over 5730242.57 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3944, pruned_loss=0.1386, over 5677864.22 frames. ], batch size: 284, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:38:53,342 INFO [train.py:968] (1/2) Epoch 6, batch 26300, giga_loss[loss=0.36, simple_loss=0.4188, pruned_loss=0.1506, over 28938.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3968, pruned_loss=0.1413, over 5692228.12 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3876, pruned_loss=0.1375, over 5731815.76 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3971, pruned_loss=0.1411, over 5680898.46 frames. ], batch size: 174, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:39:04,640 INFO [optim.py:369] (1/2) 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 07:39:06,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2288, 1.3863, 1.4758, 1.4075], device='cuda:1'), covar=tensor([0.0903, 0.0857, 0.1236, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0748, 0.0653, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 07:39:24,990 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:39:42,886 INFO [train.py:968] (1/2) Epoch 6, batch 26350, giga_loss[loss=0.3199, simple_loss=0.383, pruned_loss=0.1284, over 28629.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3972, pruned_loss=0.1429, over 5679958.68 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3876, pruned_loss=0.1375, over 5732131.21 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3976, pruned_loss=0.1428, over 5669758.20 frames. ], batch size: 307, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:40:17,345 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:40:30,124 INFO [train.py:968] (1/2) Epoch 6, batch 26400, giga_loss[loss=0.3203, simple_loss=0.3828, pruned_loss=0.1289, over 28573.00 frames. ], tot_loss[loss=0.3425, simple_loss=0.3974, pruned_loss=0.1438, over 5674938.38 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 5715524.02 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3973, pruned_loss=0.1433, over 5679666.82 frames. ], batch size: 336, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:40:41,032 INFO [optim.py:369] (1/2) 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 07:41:03,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:41:14,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2429, 4.0721, 3.8227, 1.7005], device='cuda:1'), covar=tensor([0.0523, 0.0647, 0.0788, 0.2023], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0865, 0.0807, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:41:15,380 INFO [train.py:968] (1/2) Epoch 6, batch 26450, giga_loss[loss=0.307, simple_loss=0.3685, pruned_loss=0.1228, over 28777.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3946, pruned_loss=0.1427, over 5680474.65 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3881, pruned_loss=0.138, over 5719729.25 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.395, pruned_loss=0.1425, over 5679625.58 frames. ], batch size: 99, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:41:20,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6147, 2.3457, 1.6069, 0.7608], device='cuda:1'), covar=tensor([0.2573, 0.1493, 0.2490, 0.2773], device='cuda:1'), in_proj_covar=tensor([0.1439, 0.1347, 0.1380, 0.1173], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 07:41:29,653 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 07:41:37,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:42:02,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3357, 2.1141, 1.5291, 0.5587], device='cuda:1'), covar=tensor([0.2901, 0.1364, 0.2131, 0.3148], device='cuda:1'), in_proj_covar=tensor([0.1438, 0.1343, 0.1378, 0.1169], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 07:42:04,403 INFO [train.py:968] (1/2) Epoch 6, batch 26500, giga_loss[loss=0.3013, simple_loss=0.3708, pruned_loss=0.1159, over 28854.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3934, pruned_loss=0.1425, over 5684781.20 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3888, pruned_loss=0.1386, over 5719494.54 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3932, pruned_loss=0.142, over 5683640.92 frames. ], batch size: 145, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:42:09,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2969, 1.2457, 1.1393, 0.9721], device='cuda:1'), covar=tensor([0.0684, 0.0523, 0.1027, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0457, 0.0509, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:42:16,330 INFO [optim.py:369] (1/2) 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 07:42:32,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:42:34,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:42:41,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 07:42:48,697 INFO [train.py:968] (1/2) Epoch 6, batch 26550, giga_loss[loss=0.3667, simple_loss=0.3949, pruned_loss=0.1693, over 23620.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3931, pruned_loss=0.1429, over 5682322.99 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3878, pruned_loss=0.138, over 5726966.24 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3941, pruned_loss=0.1432, over 5672550.05 frames. ], batch size: 705, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:42:58,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:12,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:15,049 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:16,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:31,037 INFO [train.py:968] (1/2) Epoch 6, batch 26600, giga_loss[loss=0.4102, simple_loss=0.4312, pruned_loss=0.1946, over 26746.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3927, pruned_loss=0.1429, over 5680730.64 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3879, pruned_loss=0.1381, over 5722097.06 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3936, pruned_loss=0.1431, over 5676240.94 frames. ], batch size: 555, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:43:42,213 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:43,172 INFO [optim.py:369] (1/2) 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 07:43:49,446 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:52,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:43:54,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=254335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:44:18,799 INFO [train.py:968] (1/2) Epoch 6, batch 26650, giga_loss[loss=0.348, simple_loss=0.4048, pruned_loss=0.1456, over 28924.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.391, pruned_loss=0.1427, over 5665549.09 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3874, pruned_loss=0.1378, over 5725035.44 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3921, pruned_loss=0.1433, over 5658775.13 frames. ], batch size: 213, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:44:19,677 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:45:07,374 INFO [train.py:968] (1/2) Epoch 6, batch 26700, giga_loss[loss=0.3308, simple_loss=0.3874, pruned_loss=0.1371, over 27602.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3904, pruned_loss=0.1423, over 5663278.69 frames. ], libri_tot_loss[loss=0.331, simple_loss=0.3871, pruned_loss=0.1375, over 5727731.30 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3917, pruned_loss=0.1431, over 5654499.10 frames. ], batch size: 472, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:45:13,732 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:45:26,450 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:45:28,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:45:39,274 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-03 07:45:53,510 INFO [train.py:968] (1/2) Epoch 6, batch 26750, giga_loss[loss=0.2909, simple_loss=0.3726, pruned_loss=0.1046, over 28919.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.391, pruned_loss=0.1411, over 5670756.24 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3877, pruned_loss=0.1381, over 5726992.77 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3915, pruned_loss=0.1413, over 5663198.80 frames. ], batch size: 174, lr: 5.31e-03, grad_scale: 2.0
+2023-03-03 07:45:57,889 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:12,853 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:15,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:18,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8672, 3.6944, 3.5092, 1.9059], device='cuda:1'), covar=tensor([0.0579, 0.0666, 0.0760, 0.1975], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0874, 0.0817, 0.0620], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:46:45,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:46:45,435 INFO [train.py:968] (1/2) Epoch 6, batch 26800, giga_loss[loss=0.3906, simple_loss=0.4266, pruned_loss=0.1773, over 28193.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3953, pruned_loss=0.1448, over 5658197.20 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3879, pruned_loss=0.1383, over 5726897.02 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3956, pruned_loss=0.1449, over 5651681.20 frames. ], batch size: 368, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:46:58,820 INFO [optim.py:369] (1/2) 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 07:47:10,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3438, 1.4324, 1.4309, 1.3696], device='cuda:1'), covar=tensor([0.0978, 0.1205, 0.1522, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0740, 0.0649, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 07:47:12,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5074, 1.4871, 1.6606, 1.4614], device='cuda:1'), covar=tensor([0.1688, 0.2683, 0.1325, 0.1399], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0740, 0.0808, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 07:47:30,709 INFO [train.py:968] (1/2) Epoch 6, batch 26850, giga_loss[loss=0.3775, simple_loss=0.4173, pruned_loss=0.1688, over 28080.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3943, pruned_loss=0.1441, over 5663125.32 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3879, pruned_loss=0.1382, over 5722224.12 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3948, pruned_loss=0.1444, over 5659838.69 frames. ], batch size: 412, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:47:31,076 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=254561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:47:34,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=254564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:47:58,600 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=254593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:48:15,790 INFO [train.py:968] (1/2) Epoch 6, batch 26900, giga_loss[loss=0.2925, simple_loss=0.3714, pruned_loss=0.1068, over 28737.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3934, pruned_loss=0.1403, over 5672906.30 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3876, pruned_loss=0.1381, over 5726982.24 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3941, pruned_loss=0.1407, over 5664728.91 frames. ], batch size: 119, lr: 5.31e-03, grad_scale: 4.0
+2023-03-03 07:48:26,492 INFO [optim.py:369] (1/2) 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 07:48:48,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2520, 1.7113, 1.6149, 1.3788], device='cuda:1'), covar=tensor([0.1648, 0.2091, 0.1373, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0736, 0.0803, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 07:49:01,692 INFO [train.py:968] (1/2) Epoch 6, batch 26950, giga_loss[loss=0.3083, simple_loss=0.3799, pruned_loss=0.1184, over 28627.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3954, pruned_loss=0.1393, over 5683569.73 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.388, pruned_loss=0.1385, over 5729265.03 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3957, pruned_loss=0.1394, over 5674618.96 frames. ], batch size: 85, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:49:49,211 INFO [train.py:968] (1/2) Epoch 6, batch 27000, giga_loss[loss=0.3774, simple_loss=0.4272, pruned_loss=0.1638, over 28537.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3984, pruned_loss=0.1409, over 5685573.94 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3881, pruned_loss=0.1385, over 5729815.71 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3986, pruned_loss=0.1409, over 5677844.20 frames. ], batch size: 336, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:49:49,211 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 07:49:56,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2834, 1.5037, 1.2117, 1.0965], device='cuda:1'), covar=tensor([0.0910, 0.0854, 0.0591, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.1524, 0.1343, 0.1295, 0.1412], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 07:49:57,828 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 07:50:09,970 INFO [optim.py:369] (1/2) 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 07:50:46,751 INFO [train.py:968] (1/2) Epoch 6, batch 27050, giga_loss[loss=0.5389, simple_loss=0.5171, pruned_loss=0.2803, over 27651.00 frames. ], tot_loss[loss=0.346, simple_loss=0.402, pruned_loss=0.145, over 5680132.59 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3882, pruned_loss=0.1385, over 5730553.63 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.4022, pruned_loss=0.145, over 5673201.66 frames. ], batch size: 472, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:51:37,263 INFO [train.py:968] (1/2) Epoch 6, batch 27100, giga_loss[loss=0.3819, simple_loss=0.4333, pruned_loss=0.1652, over 28560.00 frames. ], tot_loss[loss=0.3487, simple_loss=0.4029, pruned_loss=0.1473, over 5660884.30 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3879, pruned_loss=0.1384, over 5735824.73 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4037, pruned_loss=0.1476, over 5648993.18 frames. ], batch size: 60, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:51:49,234 INFO [optim.py:369] (1/2) 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 07:52:26,231 INFO [train.py:968] (1/2) Epoch 6, batch 27150, giga_loss[loss=0.3693, simple_loss=0.4219, pruned_loss=0.1584, over 28610.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.4012, pruned_loss=0.1462, over 5645242.79 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3879, pruned_loss=0.1387, over 5708057.50 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.402, pruned_loss=0.1464, over 5658458.44 frames. ], batch size: 307, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:53:14,816 INFO [train.py:968] (1/2) Epoch 6, batch 27200, libri_loss[loss=0.3769, simple_loss=0.4223, pruned_loss=0.1657, over 29513.00 frames. ], tot_loss[loss=0.346, simple_loss=0.4007, pruned_loss=0.1457, over 5636708.63 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3881, pruned_loss=0.1389, over 5709087.98 frames. ], giga_tot_loss[loss=0.3465, simple_loss=0.4014, pruned_loss=0.1458, over 5644190.49 frames. ], batch size: 80, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 07:53:25,231 INFO [optim.py:369] (1/2) 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 07:53:29,699 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:53:59,640 INFO [train.py:968] (1/2) Epoch 6, batch 27250, giga_loss[loss=0.317, simple_loss=0.3914, pruned_loss=0.1213, over 28918.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.4004, pruned_loss=0.1433, over 5652116.01 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3884, pruned_loss=0.1391, over 5712355.73 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.4009, pruned_loss=0.1433, over 5654088.96 frames. ], batch size: 186, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 07:54:06,267 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=254967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:54:46,614 INFO [train.py:968] (1/2) Epoch 6, batch 27300, giga_loss[loss=0.3253, simple_loss=0.3901, pruned_loss=0.1303, over 28975.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.4, pruned_loss=0.1422, over 5658958.79 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3884, pruned_loss=0.139, over 5718239.04 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.4007, pruned_loss=0.1424, over 5653689.49 frames. ], batch size: 213, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:55:02,364 INFO [optim.py:369] (1/2) 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 07:55:37,488 INFO [train.py:968] (1/2) Epoch 6, batch 27350, giga_loss[loss=0.369, simple_loss=0.4172, pruned_loss=0.1604, over 28762.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.4003, pruned_loss=0.143, over 5663528.99 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3879, pruned_loss=0.1386, over 5721585.86 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.4016, pruned_loss=0.1435, over 5655491.96 frames. ], batch size: 284, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:56:23,750 INFO [train.py:968] (1/2) Epoch 6, batch 27400, giga_loss[loss=0.2796, simple_loss=0.3484, pruned_loss=0.1054, over 28847.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3993, pruned_loss=0.1428, over 5664440.25 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3883, pruned_loss=0.1389, over 5714042.11 frames. ], giga_tot_loss[loss=0.343, simple_loss=0.4, pruned_loss=0.143, over 5665280.53 frames. ], batch size: 112, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:56:26,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5116, 1.8517, 1.8682, 1.6071], device='cuda:1'), covar=tensor([0.1532, 0.1988, 0.1176, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0737, 0.0807, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 07:56:39,576 INFO [optim.py:369] (1/2) 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 07:57:01,130 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 27450, giga_loss[loss=0.3945, simple_loss=0.4243, pruned_loss=0.1823, over 28260.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3975, pruned_loss=0.1436, over 5653442.70 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3882, pruned_loss=0.1388, over 5718134.27 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3985, pruned_loss=0.144, over 5649025.30 frames. ], batch size: 368, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:57:40,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9516, 2.4860, 1.6047, 1.5382], device='cuda:1'), covar=tensor([0.1558, 0.0957, 0.1154, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.1504, 0.1347, 0.1290, 0.1402], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 07:57:43,184 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 27500, giga_loss[loss=0.3742, simple_loss=0.4173, pruned_loss=0.1656, over 27554.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3949, pruned_loss=0.1428, over 5636138.95 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.388, pruned_loss=0.1387, over 5712135.47 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.396, pruned_loss=0.1433, over 5636381.06 frames. ], batch size: 472, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:58:18,549 INFO [optim.py:369] (1/2) 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 07:58:39,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:58:55,813 INFO [train.py:968] (1/2) Epoch 6, batch 27550, giga_loss[loss=0.3299, simple_loss=0.3825, pruned_loss=0.1387, over 28480.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3925, pruned_loss=0.1418, over 5640755.98 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3882, pruned_loss=0.1389, over 5705002.35 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3932, pruned_loss=0.142, over 5646698.88 frames. ], batch size: 71, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 07:59:04,849 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-03 07:59:33,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8367, 5.6776, 5.3742, 2.8891], device='cuda:1'), covar=tensor([0.0344, 0.0463, 0.0635, 0.1526], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0885, 0.0824, 0.0617], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 07:59:33,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 07:59:35,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6334, 2.4962, 1.8198, 0.7761], device='cuda:1'), covar=tensor([0.2647, 0.1304, 0.2051, 0.2679], device='cuda:1'), in_proj_covar=tensor([0.1456, 0.1379, 0.1402, 0.1191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 07:59:38,783 INFO [train.py:968] (1/2) Epoch 6, batch 27600, giga_loss[loss=0.4457, simple_loss=0.4648, pruned_loss=0.2133, over 28909.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3931, pruned_loss=0.1434, over 5632640.02 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3886, pruned_loss=0.1393, over 5695225.14 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3935, pruned_loss=0.1434, over 5644484.44 frames. ], batch size: 145, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 07:59:52,538 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:00:22,887 INFO [train.py:968] (1/2) Epoch 6, batch 27650, giga_loss[loss=0.26, simple_loss=0.3392, pruned_loss=0.09042, over 28803.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3903, pruned_loss=0.141, over 5641851.41 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3885, pruned_loss=0.1393, over 5701582.86 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3908, pruned_loss=0.1411, over 5643796.43 frames. ], batch size: 112, lr: 5.30e-03, grad_scale: 8.0
+2023-03-03 08:01:10,055 INFO [train.py:968] (1/2) Epoch 6, batch 27700, giga_loss[loss=0.2807, simple_loss=0.3564, pruned_loss=0.1025, over 28695.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.387, pruned_loss=0.1364, over 5647998.16 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3892, pruned_loss=0.1398, over 5692041.98 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3868, pruned_loss=0.136, over 5657882.63 frames. ], batch size: 119, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 08:01:24,447 INFO [optim.py:369] (1/2) 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 08:01:46,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:01:49,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:01,054 INFO [train.py:968] (1/2) Epoch 6, batch 27750, giga_loss[loss=0.347, simple_loss=0.4034, pruned_loss=0.1453, over 28715.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3853, pruned_loss=0.1345, over 5645054.49 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3892, pruned_loss=0.1398, over 5692041.98 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3852, pruned_loss=0.1342, over 5652747.69 frames. ], batch size: 262, lr: 5.30e-03, grad_scale: 4.0
+2023-03-03 08:02:20,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:27,825 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:29,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:02:52,188 INFO [train.py:968] (1/2) Epoch 6, batch 27800, libri_loss[loss=0.4052, simple_loss=0.43, pruned_loss=0.1902, over 29552.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3842, pruned_loss=0.1343, over 5638234.56 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3895, pruned_loss=0.1401, over 5695683.47 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3837, pruned_loss=0.1336, over 5640066.97 frames. ], batch size: 79, lr: 5.30e-03, grad_scale: 2.0
+2023-03-03 08:03:01,513 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:03:06,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:03:10,924 INFO [optim.py:369] (1/2) 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 08:03:47,746 INFO [train.py:968] (1/2) Epoch 6, batch 27850, giga_loss[loss=0.2707, simple_loss=0.3445, pruned_loss=0.0985, over 28852.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3803, pruned_loss=0.132, over 5662562.24 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3897, pruned_loss=0.1403, over 5698094.62 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3795, pruned_loss=0.1312, over 5661270.67 frames. ], batch size: 112, lr: 5.30e-03, grad_scale: 2.0
+2023-03-03 08:03:52,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=255565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:04:38,209 INFO [train.py:968] (1/2) Epoch 6, batch 27900, giga_loss[loss=0.32, simple_loss=0.3851, pruned_loss=0.1275, over 28767.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.383, pruned_loss=0.1343, over 5654134.26 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3901, pruned_loss=0.1406, over 5689911.04 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3819, pruned_loss=0.1333, over 5659815.35 frames. ], batch size: 284, lr: 5.30e-03, grad_scale: 2.0
+2023-03-03 08:04:43,475 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:05:23,895 INFO [train.py:968] (1/2) Epoch 6, batch 27950, giga_loss[loss=0.3333, simple_loss=0.3983, pruned_loss=0.1342, over 28715.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3862, pruned_loss=0.1362, over 5650939.93 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3899, pruned_loss=0.1404, over 5695507.34 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3853, pruned_loss=0.1354, over 5650195.12 frames. ], batch size: 242, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:05:31,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:05:34,078 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 08:05:35,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:02,253 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:12,487 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:14,799 INFO [train.py:968] (1/2) Epoch 6, batch 28000, giga_loss[loss=0.2935, simple_loss=0.3585, pruned_loss=0.1143, over 28819.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3868, pruned_loss=0.1367, over 5648948.78 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3902, pruned_loss=0.1405, over 5696448.63 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3858, pruned_loss=0.1358, over 5647063.12 frames. ], batch size: 99, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:06:15,072 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:06:40,620 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=255740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:06:43,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4615, 1.8135, 1.8165, 1.5877], device='cuda:1'), covar=tensor([0.1515, 0.1950, 0.1120, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0744, 0.0815, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 08:06:59,005 INFO [train.py:968] (1/2) Epoch 6, batch 28050, giga_loss[loss=0.3238, simple_loss=0.3857, pruned_loss=0.1309, over 28633.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3871, pruned_loss=0.1368, over 5647659.38 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3909, pruned_loss=0.1409, over 5691092.53 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3855, pruned_loss=0.1356, over 5648619.05 frames. ], batch size: 307, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:06:59,860 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=255762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:07:02,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=255765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:07:17,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=255781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:07:27,975 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 28100, giga_loss[loss=0.3798, simple_loss=0.4217, pruned_loss=0.1689, over 28937.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3882, pruned_loss=0.1379, over 5654167.90 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.391, pruned_loss=0.1409, over 5693938.41 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3867, pruned_loss=0.1369, over 5650769.42 frames. ], batch size: 213, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:07:55,532 INFO [optim.py:369] (1/2) 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 08:08:23,055 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-03 08:08:28,546 INFO [train.py:968] (1/2) Epoch 6, batch 28150, giga_loss[loss=0.3801, simple_loss=0.4261, pruned_loss=0.167, over 27977.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3919, pruned_loss=0.1408, over 5655644.77 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3918, pruned_loss=0.1416, over 5692356.21 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.39, pruned_loss=0.1393, over 5653994.77 frames. ], batch size: 412, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:08:48,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0520, 1.3281, 4.1240, 3.4057], device='cuda:1'), covar=tensor([0.1662, 0.2280, 0.0350, 0.0598], device='cuda:1'), in_proj_covar=tensor([0.0590, 0.0547, 0.0783, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 08:09:15,792 INFO [train.py:968] (1/2) Epoch 6, batch 28200, giga_loss[loss=0.3317, simple_loss=0.3924, pruned_loss=0.1356, over 28686.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3935, pruned_loss=0.1414, over 5654728.72 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.3925, pruned_loss=0.1423, over 5685391.93 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3915, pruned_loss=0.1397, over 5659382.04 frames. ], batch size: 307, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:09:29,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6006, 1.8885, 1.4504, 1.2380], device='cuda:1'), covar=tensor([0.1657, 0.1165, 0.0989, 0.1331], device='cuda:1'), in_proj_covar=tensor([0.1523, 0.1355, 0.1313, 0.1417], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 08:09:30,910 INFO [optim.py:369] (1/2) 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 08:09:40,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 08:10:06,410 INFO [train.py:968] (1/2) Epoch 6, batch 28250, giga_loss[loss=0.4505, simple_loss=0.4544, pruned_loss=0.2234, over 27673.00 frames. ], tot_loss[loss=0.3425, simple_loss=0.3963, pruned_loss=0.1444, over 5647021.20 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.393, pruned_loss=0.1427, over 5689465.40 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3942, pruned_loss=0.1427, over 5646434.39 frames. ], batch size: 472, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:10:53,566 INFO [train.py:968] (1/2) Epoch 6, batch 28300, giga_loss[loss=0.3456, simple_loss=0.4077, pruned_loss=0.1418, over 27655.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3959, pruned_loss=0.1442, over 5640679.01 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3931, pruned_loss=0.1427, over 5684545.14 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3942, pruned_loss=0.1428, over 5644001.02 frames. ], batch size: 472, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:11:09,506 INFO [optim.py:369] (1/2) 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 08:11:45,974 INFO [train.py:968] (1/2) Epoch 6, batch 28350, giga_loss[loss=0.3577, simple_loss=0.4204, pruned_loss=0.1475, over 28695.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3958, pruned_loss=0.1419, over 5650419.01 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.3925, pruned_loss=0.1423, over 5686996.36 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.395, pruned_loss=0.1411, over 5650501.08 frames. ], batch size: 307, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:12:21,199 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1281, 1.3137, 3.6176, 3.0968], device='cuda:1'), covar=tensor([0.1494, 0.2235, 0.0430, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0597, 0.0549, 0.0791, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 08:12:33,010 INFO [train.py:968] (1/2) Epoch 6, batch 28400, giga_loss[loss=0.299, simple_loss=0.3651, pruned_loss=0.1164, over 28582.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3962, pruned_loss=0.1424, over 5654027.10 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3922, pruned_loss=0.1423, over 5679620.79 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3959, pruned_loss=0.1418, over 5659513.42 frames. ], batch size: 78, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:12:38,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7251, 3.5630, 3.3678, 1.8057], device='cuda:1'), covar=tensor([0.0566, 0.0698, 0.0780, 0.2139], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0876, 0.0822, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:12:48,716 INFO [optim.py:369] (1/2) 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 08:13:13,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:13:17,597 INFO [train.py:968] (1/2) Epoch 6, batch 28450, libri_loss[loss=0.3431, simple_loss=0.3935, pruned_loss=0.1464, over 29531.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3946, pruned_loss=0.1422, over 5645807.96 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3925, pruned_loss=0.1427, over 5669342.87 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3942, pruned_loss=0.1413, over 5658084.01 frames. ], batch size: 81, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:13:47,764 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-03 08:13:56,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6078, 5.4575, 5.1518, 2.5928], device='cuda:1'), covar=tensor([0.0330, 0.0497, 0.0671, 0.1698], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0875, 0.0822, 0.0612], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:14:08,508 INFO [train.py:968] (1/2) Epoch 6, batch 28500, giga_loss[loss=0.3111, simple_loss=0.3675, pruned_loss=0.1274, over 28938.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3941, pruned_loss=0.1421, over 5658185.12 frames. ], libri_tot_loss[loss=0.3388, simple_loss=0.3924, pruned_loss=0.1425, over 5673700.78 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3939, pruned_loss=0.1415, over 5663731.44 frames. ], batch size: 112, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:14:28,859 INFO [optim.py:369] (1/2) 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 08:15:01,069 INFO [train.py:968] (1/2) Epoch 6, batch 28550, giga_loss[loss=0.371, simple_loss=0.4096, pruned_loss=0.1662, over 28081.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3908, pruned_loss=0.1399, over 5660753.07 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3917, pruned_loss=0.142, over 5671503.72 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3913, pruned_loss=0.1398, over 5666121.11 frames. ], batch size: 412, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:15:33,798 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:15:36,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:15:43,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9687, 1.1094, 0.9382, 0.7501], device='cuda:1'), covar=tensor([0.0973, 0.0961, 0.0697, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.1529, 0.1350, 0.1309, 0.1419], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 08:15:43,701 INFO [train.py:968] (1/2) Epoch 6, batch 28600, giga_loss[loss=0.3619, simple_loss=0.4094, pruned_loss=0.1572, over 28957.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3912, pruned_loss=0.1405, over 5658838.68 frames. ], libri_tot_loss[loss=0.338, simple_loss=0.3918, pruned_loss=0.1421, over 5661448.36 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3916, pruned_loss=0.1403, over 5671807.09 frames. ], batch size: 227, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:16:00,212 INFO [optim.py:369] (1/2) 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 08:16:00,975 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:16:15,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:16:26,320 INFO [train.py:968] (1/2) Epoch 6, batch 28650, giga_loss[loss=0.3117, simple_loss=0.3774, pruned_loss=0.123, over 28890.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3907, pruned_loss=0.1409, over 5647008.50 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3909, pruned_loss=0.1414, over 5664698.38 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3918, pruned_loss=0.1413, over 5655123.67 frames. ], batch size: 112, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:16:38,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3129, 4.1742, 3.9993, 1.8905], device='cuda:1'), covar=tensor([0.0490, 0.0596, 0.0698, 0.1924], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0889, 0.0827, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0009], device='cuda:1')
+2023-03-03 08:17:05,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6240, 1.7407, 1.4561, 2.1574], device='cuda:1'), covar=tensor([0.2228, 0.2248, 0.2274, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.1154, 0.0887, 0.1020, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 08:17:14,789 INFO [train.py:968] (1/2) Epoch 6, batch 28700, giga_loss[loss=0.3014, simple_loss=0.3725, pruned_loss=0.1152, over 28654.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3906, pruned_loss=0.1404, over 5656460.02 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3908, pruned_loss=0.1413, over 5668135.35 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3915, pruned_loss=0.1408, over 5659584.20 frames. ], batch size: 92, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:17:18,552 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5922, 3.9162, 1.7674, 1.5245], device='cuda:1'), covar=tensor([0.0780, 0.0287, 0.0813, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0491, 0.0315, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 08:17:32,692 INFO [optim.py:369] (1/2) 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 08:17:33,684 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 28750, libri_loss[loss=0.3191, simple_loss=0.3713, pruned_loss=0.1334, over 29552.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3924, pruned_loss=0.1427, over 5656245.74 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.39, pruned_loss=0.141, over 5673882.71 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3938, pruned_loss=0.1433, over 5653197.51 frames. ], batch size: 76, lr: 5.29e-03, grad_scale: 2.0
+2023-03-03 08:18:06,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-03 08:18:15,518 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:18:18,576 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 28800, libri_loss[loss=0.3436, simple_loss=0.3769, pruned_loss=0.1551, over 29358.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3939, pruned_loss=0.1439, over 5653250.27 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3899, pruned_loss=0.1409, over 5671019.55 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3953, pruned_loss=0.1446, over 5651989.45 frames. ], batch size: 71, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:19:03,584 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:19:07,677 INFO [optim.py:369] (1/2) 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 08:19:11,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2844, 1.4101, 1.0918, 0.9542], device='cuda:1'), covar=tensor([0.1191, 0.1007, 0.0789, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.1517, 0.1359, 0.1306, 0.1416], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 08:19:20,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0280, 1.4566, 1.3583, 1.1449], device='cuda:1'), covar=tensor([0.1111, 0.1795, 0.1010, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0739, 0.0808, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 08:19:37,636 INFO [train.py:968] (1/2) Epoch 6, batch 28850, giga_loss[loss=0.3559, simple_loss=0.4017, pruned_loss=0.1551, over 28655.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3935, pruned_loss=0.1443, over 5638990.96 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3898, pruned_loss=0.1408, over 5669134.73 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3947, pruned_loss=0.1449, over 5639061.67 frames. ], batch size: 262, lr: 5.29e-03, grad_scale: 4.0
+2023-03-03 08:20:21,309 INFO [train.py:968] (1/2) Epoch 6, batch 28900, giga_loss[loss=0.3163, simple_loss=0.3736, pruned_loss=0.1295, over 28789.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3944, pruned_loss=0.1456, over 5646652.76 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3894, pruned_loss=0.1405, over 5672527.19 frames. ], giga_tot_loss[loss=0.3444, simple_loss=0.3958, pruned_loss=0.1465, over 5642691.18 frames. ], batch size: 199, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:20:38,779 INFO [optim.py:369] (1/2) 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 08:20:42,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0103, 1.7833, 1.3481, 1.5522], device='cuda:1'), covar=tensor([0.0619, 0.0619, 0.0978, 0.0954], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0457, 0.0511, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:20:44,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5952, 3.4234, 1.6947, 1.4243], device='cuda:1'), covar=tensor([0.0819, 0.0306, 0.0804, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0495, 0.0317, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 08:21:09,416 INFO [train.py:968] (1/2) Epoch 6, batch 28950, giga_loss[loss=0.3472, simple_loss=0.4036, pruned_loss=0.1453, over 28908.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3927, pruned_loss=0.1442, over 5647835.55 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3894, pruned_loss=0.1405, over 5677164.70 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.394, pruned_loss=0.145, over 5640472.33 frames. ], batch size: 227, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:21:11,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8833, 2.8329, 1.7664, 1.1866], device='cuda:1'), covar=tensor([0.3618, 0.1637, 0.2216, 0.3265], device='cuda:1'), in_proj_covar=tensor([0.1458, 0.1369, 0.1408, 0.1197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 08:21:24,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:21:57,665 INFO [train.py:968] (1/2) Epoch 6, batch 29000, libri_loss[loss=0.3862, simple_loss=0.4448, pruned_loss=0.1639, over 29238.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.394, pruned_loss=0.1444, over 5648881.85 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.39, pruned_loss=0.1408, over 5682718.94 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3945, pruned_loss=0.1449, over 5637257.81 frames. ], batch size: 97, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:22:07,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-03 08:22:07,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7310, 1.6606, 1.5785, 1.5661], device='cuda:1'), covar=tensor([0.1119, 0.1831, 0.1695, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0746, 0.0653, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 08:22:09,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:22:12,679 INFO [optim.py:369] (1/2) 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 08:22:40,025 INFO [train.py:968] (1/2) Epoch 6, batch 29050, giga_loss[loss=0.3748, simple_loss=0.4096, pruned_loss=0.17, over 27623.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.393, pruned_loss=0.1428, over 5661055.14 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3895, pruned_loss=0.1402, over 5689854.76 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3941, pruned_loss=0.1439, over 5643954.94 frames. ], batch size: 472, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:22:44,324 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:23:21,576 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 29100, giga_loss[loss=0.3317, simple_loss=0.3995, pruned_loss=0.132, over 28481.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3946, pruned_loss=0.1438, over 5668336.51 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3896, pruned_loss=0.1404, over 5690000.65 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3954, pruned_loss=0.1445, over 5654104.01 frames. ], batch size: 60, lr: 5.28e-03, grad_scale: 2.0
+2023-03-03 08:23:42,150 INFO [optim.py:369] (1/2) 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 08:23:46,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4106, 1.8873, 1.7879, 1.5115], device='cuda:1'), covar=tensor([0.1392, 0.1857, 0.1095, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0740, 0.0807, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 08:23:57,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3837, 2.2382, 1.5808, 0.5033], device='cuda:1'), covar=tensor([0.3075, 0.1519, 0.2272, 0.3556], device='cuda:1'), in_proj_covar=tensor([0.1441, 0.1358, 0.1398, 0.1190], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 08:23:58,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:00,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=256852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:01,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8604, 1.7947, 1.7617, 1.7356], device='cuda:1'), covar=tensor([0.1056, 0.1700, 0.1496, 0.1503], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0745, 0.0650, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 08:24:09,756 INFO [train.py:968] (1/2) Epoch 6, batch 29150, giga_loss[loss=0.3147, simple_loss=0.3808, pruned_loss=0.1243, over 28927.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.396, pruned_loss=0.1451, over 5678459.78 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3893, pruned_loss=0.1402, over 5694103.84 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.397, pruned_loss=0.146, over 5663315.17 frames. ], batch size: 66, lr: 5.28e-03, grad_scale: 2.0
+2023-03-03 08:24:11,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=256864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:15,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:18,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:20,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3422, 2.0539, 1.5545, 1.5011], device='cuda:1'), covar=tensor([0.0709, 0.0313, 0.0293, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0119, 0.0122, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0063, 0.0046, 0.0041, 0.0069], device='cuda:1')
+2023-03-03 08:24:42,934 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:24:43,974 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 29200, libri_loss[loss=0.2697, simple_loss=0.3337, pruned_loss=0.1028, over 29367.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3961, pruned_loss=0.1453, over 5679549.45 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3894, pruned_loss=0.1403, over 5697329.81 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.397, pruned_loss=0.146, over 5664103.74 frames. ], batch size: 67, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:24:56,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-03 08:25:14,893 INFO [optim.py:369] (1/2) 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 08:25:32,958 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:25:36,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:25:45,025 INFO [train.py:968] (1/2) Epoch 6, batch 29250, giga_loss[loss=0.3456, simple_loss=0.4127, pruned_loss=0.1392, over 28618.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.397, pruned_loss=0.1447, over 5679024.86 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3891, pruned_loss=0.1401, over 5701700.71 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.3981, pruned_loss=0.1456, over 5662365.51 frames. ], batch size: 307, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:26:04,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=256981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:16,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:20,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:20,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=256995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:22,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=256998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:32,121 INFO [train.py:968] (1/2) Epoch 6, batch 29300, giga_loss[loss=0.2941, simple_loss=0.3685, pruned_loss=0.1098, over 28967.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3971, pruned_loss=0.1438, over 5665672.20 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3895, pruned_loss=0.1403, over 5694892.27 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3977, pruned_loss=0.1443, over 5658417.55 frames. ], batch size: 213, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:26:43,164 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:46,511 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:26:49,785 INFO [optim.py:369] (1/2) 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 08:26:59,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:01,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:07,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7913, 1.1376, 3.4015, 2.9468], device='cuda:1'), covar=tensor([0.1638, 0.2173, 0.0455, 0.0709], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0545, 0.0781, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 08:27:09,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:10,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 08:27:17,067 INFO [train.py:968] (1/2) Epoch 6, batch 29350, giga_loss[loss=0.418, simple_loss=0.4356, pruned_loss=0.2002, over 26500.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3955, pruned_loss=0.143, over 5655211.86 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.39, pruned_loss=0.1406, over 5687255.75 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3958, pruned_loss=0.1432, over 5655377.18 frames. ], batch size: 555, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:27:21,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2724, 1.8063, 1.4706, 0.3645], device='cuda:1'), covar=tensor([0.2160, 0.1482, 0.2146, 0.2847], device='cuda:1'), in_proj_covar=tensor([0.1429, 0.1356, 0.1387, 0.1182], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 08:27:27,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:27:44,411 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-03 08:28:01,558 INFO [train.py:968] (1/2) Epoch 6, batch 29400, giga_loss[loss=0.3223, simple_loss=0.3847, pruned_loss=0.1299, over 28993.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3956, pruned_loss=0.1434, over 5656494.37 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3896, pruned_loss=0.1405, over 5689563.15 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3962, pruned_loss=0.1437, over 5654546.94 frames. ], batch size: 136, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:28:21,840 INFO [optim.py:369] (1/2) 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 08:28:30,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:28:52,153 INFO [train.py:968] (1/2) Epoch 6, batch 29450, giga_loss[loss=0.3872, simple_loss=0.4231, pruned_loss=0.1756, over 27718.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3965, pruned_loss=0.1437, over 5663254.55 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3899, pruned_loss=0.1406, over 5693750.92 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3968, pruned_loss=0.1439, over 5657373.67 frames. ], batch size: 474, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:29:26,368 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:29:28,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:29:39,897 INFO [train.py:968] (1/2) Epoch 6, batch 29500, giga_loss[loss=0.3412, simple_loss=0.3982, pruned_loss=0.1421, over 28672.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3972, pruned_loss=0.1449, over 5650610.92 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3901, pruned_loss=0.1406, over 5687924.31 frames. ], giga_tot_loss[loss=0.3438, simple_loss=0.3975, pruned_loss=0.1451, over 5650683.00 frames. ], batch size: 262, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:29:48,987 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257220.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:29:57,992 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:30:05,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:30:24,479 INFO [train.py:968] (1/2) Epoch 6, batch 29550, giga_loss[loss=0.3136, simple_loss=0.3841, pruned_loss=0.1216, over 28514.00 frames. ], tot_loss[loss=0.343, simple_loss=0.3964, pruned_loss=0.1448, over 5660509.61 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3902, pruned_loss=0.1407, over 5694590.49 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3966, pruned_loss=0.145, over 5653529.43 frames. ], batch size: 71, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:30:36,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257276.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:30:43,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:30:44,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:30:48,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7088, 1.6890, 1.2424, 1.3979], device='cuda:1'), covar=tensor([0.0650, 0.0553, 0.0964, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0454, 0.0509, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:31:06,241 INFO [train.py:968] (1/2) Epoch 6, batch 29600, giga_loss[loss=0.3256, simple_loss=0.3877, pruned_loss=0.1318, over 28753.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3964, pruned_loss=0.1457, over 5642402.79 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3904, pruned_loss=0.1411, over 5682556.14 frames. ], giga_tot_loss[loss=0.344, simple_loss=0.3967, pruned_loss=0.1457, over 5646618.15 frames. ], batch size: 119, lr: 5.28e-03, grad_scale: 8.0
+2023-03-03 08:31:11,104 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:31:24,878 INFO [optim.py:369] (1/2) 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 08:31:44,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4412, 4.2670, 4.0416, 1.9483], device='cuda:1'), covar=tensor([0.0456, 0.0548, 0.0630, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0885, 0.0824, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:31:54,446 INFO [train.py:968] (1/2) Epoch 6, batch 29650, giga_loss[loss=0.4035, simple_loss=0.4388, pruned_loss=0.1841, over 28733.00 frames. ], tot_loss[loss=0.3442, simple_loss=0.397, pruned_loss=0.1457, over 5650858.07 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.391, pruned_loss=0.1414, over 5679389.80 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3968, pruned_loss=0.1455, over 5655591.77 frames. ], batch size: 262, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:32:14,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:32:18,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:32:39,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 08:32:39,723 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:32:44,401 INFO [train.py:968] (1/2) Epoch 6, batch 29700, giga_loss[loss=0.3536, simple_loss=0.4089, pruned_loss=0.1491, over 28753.00 frames. ], tot_loss[loss=0.3454, simple_loss=0.3977, pruned_loss=0.1465, over 5642265.94 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3908, pruned_loss=0.1412, over 5682282.07 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.3978, pruned_loss=0.1466, over 5643173.10 frames. ], batch size: 284, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:32:46,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:33:04,190 INFO [optim.py:369] (1/2) 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 08:33:31,394 INFO [train.py:968] (1/2) Epoch 6, batch 29750, giga_loss[loss=0.3468, simple_loss=0.4026, pruned_loss=0.1455, over 28626.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3953, pruned_loss=0.1434, over 5664447.87 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3909, pruned_loss=0.1414, over 5683404.66 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3953, pruned_loss=0.1433, over 5664010.51 frames. ], batch size: 242, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:34:13,679 INFO [train.py:968] (1/2) Epoch 6, batch 29800, giga_loss[loss=0.2917, simple_loss=0.3629, pruned_loss=0.1102, over 28921.00 frames. ], tot_loss[loss=0.3418, simple_loss=0.3958, pruned_loss=0.1439, over 5653017.21 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3908, pruned_loss=0.1415, over 5674986.75 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3961, pruned_loss=0.1439, over 5659264.37 frames. ], batch size: 164, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:34:17,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3557, 1.4965, 1.2172, 1.6775], device='cuda:1'), covar=tensor([0.2135, 0.2125, 0.2199, 0.2044], device='cuda:1'), in_proj_covar=tensor([0.1151, 0.0886, 0.1025, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 08:34:34,728 INFO [optim.py:369] (1/2) 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 08:34:57,548 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:35:01,124 INFO [train.py:968] (1/2) Epoch 6, batch 29850, giga_loss[loss=0.3102, simple_loss=0.3762, pruned_loss=0.1221, over 28697.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3952, pruned_loss=0.1433, over 5661279.36 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3904, pruned_loss=0.1412, over 5679605.90 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3959, pruned_loss=0.1436, over 5661658.99 frames. ], batch size: 242, lr: 5.28e-03, grad_scale: 4.0
+2023-03-03 08:35:33,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257595.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:35:40,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7831, 5.6225, 5.3150, 2.7878], device='cuda:1'), covar=tensor([0.0338, 0.0449, 0.0653, 0.1510], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0883, 0.0824, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:35:47,586 INFO [train.py:968] (1/2) Epoch 6, batch 29900, giga_loss[loss=0.2943, simple_loss=0.3585, pruned_loss=0.1151, over 28128.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3929, pruned_loss=0.1416, over 5663964.37 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3905, pruned_loss=0.1412, over 5683760.81 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3934, pruned_loss=0.1418, over 5660192.76 frames. ], batch size: 77, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:35:51,419 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:36:05,880 INFO [optim.py:369] (1/2) 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 08:36:08,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:36:22,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257651.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:36:33,074 INFO [train.py:968] (1/2) Epoch 6, batch 29950, giga_loss[loss=0.3056, simple_loss=0.3627, pruned_loss=0.1242, over 28404.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3919, pruned_loss=0.1415, over 5663379.55 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3901, pruned_loss=0.1408, over 5687880.84 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3928, pruned_loss=0.1421, over 5656533.54 frames. ], batch size: 60, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:37:16,946 INFO [train.py:968] (1/2) Epoch 6, batch 30000, giga_loss[loss=0.3056, simple_loss=0.3664, pruned_loss=0.1224, over 28865.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3898, pruned_loss=0.1407, over 5663093.78 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3903, pruned_loss=0.1411, over 5690706.69 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3905, pruned_loss=0.1409, over 5653903.72 frames. ], batch size: 174, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:37:16,947 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 08:37:26,866 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 08:37:48,618 INFO [optim.py:369] (1/2) 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 08:37:53,956 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:37:58,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257741.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:38:08,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3327, 3.0599, 1.4084, 1.4449], device='cuda:1'), covar=tensor([0.0848, 0.0349, 0.0876, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0493, 0.0316, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 08:38:14,612 INFO [train.py:968] (1/2) Epoch 6, batch 30050, giga_loss[loss=0.2743, simple_loss=0.3415, pruned_loss=0.1035, over 28976.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3856, pruned_loss=0.1381, over 5676626.99 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3904, pruned_loss=0.1413, over 5690543.48 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.386, pruned_loss=0.1381, over 5668576.18 frames. ], batch size: 164, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:38:21,501 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257770.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:38:30,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:38:42,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257794.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:38:44,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 08:38:58,227 INFO [train.py:968] (1/2) Epoch 6, batch 30100, giga_loss[loss=0.3461, simple_loss=0.3933, pruned_loss=0.1494, over 27915.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3857, pruned_loss=0.1389, over 5690816.35 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3906, pruned_loss=0.1413, over 5694798.06 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3857, pruned_loss=0.1389, over 5680241.55 frames. ], batch size: 412, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:39:15,589 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:39:46,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 30150, giga_loss[loss=0.2592, simple_loss=0.3334, pruned_loss=0.09256, over 28531.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3857, pruned_loss=0.1389, over 5692728.49 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3907, pruned_loss=0.1414, over 5696045.08 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3855, pruned_loss=0.1388, over 5683271.01 frames. ], batch size: 60, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:40:36,125 INFO [train.py:968] (1/2) Epoch 6, batch 30200, giga_loss[loss=0.3539, simple_loss=0.4118, pruned_loss=0.1481, over 28811.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3848, pruned_loss=0.1365, over 5686033.78 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3908, pruned_loss=0.1414, over 5698762.68 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3845, pruned_loss=0.1364, over 5676089.74 frames. ], batch size: 284, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:40:48,775 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=257924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:40:52,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=257927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:40:56,359 INFO [optim.py:369] (1/2) 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 08:40:56,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:41:18,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=257956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:41:22,808 INFO [train.py:968] (1/2) Epoch 6, batch 30250, giga_loss[loss=0.3039, simple_loss=0.3835, pruned_loss=0.1122, over 28623.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3815, pruned_loss=0.1325, over 5681976.48 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3902, pruned_loss=0.1413, over 5697750.20 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3815, pruned_loss=0.1322, over 5674438.29 frames. ], batch size: 262, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:41:32,485 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=257971.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:41:46,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3308, 1.5753, 1.3296, 1.4752], device='cuda:1'), covar=tensor([0.1962, 0.1632, 0.1672, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.0893, 0.1033, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 08:41:52,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=257990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:42:08,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:42:09,882 INFO [train.py:968] (1/2) Epoch 6, batch 30300, giga_loss[loss=0.2791, simple_loss=0.3528, pruned_loss=0.1027, over 28942.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3783, pruned_loss=0.1295, over 5677658.23 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3902, pruned_loss=0.1414, over 5704497.99 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.378, pruned_loss=0.1288, over 5664734.94 frames. ], batch size: 213, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:42:14,600 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 08:42:29,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4732, 3.9405, 1.6028, 1.4128], device='cuda:1'), covar=tensor([0.0871, 0.0265, 0.0869, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0494, 0.0319, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 08:42:31,685 INFO [optim.py:369] (1/2) 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 08:43:00,007 INFO [train.py:968] (1/2) Epoch 6, batch 30350, giga_loss[loss=0.2603, simple_loss=0.345, pruned_loss=0.08776, over 29052.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3742, pruned_loss=0.1254, over 5664791.45 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3901, pruned_loss=0.1413, over 5704842.83 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3737, pruned_loss=0.1246, over 5653742.13 frames. ], batch size: 136, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:43:13,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:43:17,262 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:43:42,151 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:43:43,818 INFO [train.py:968] (1/2) Epoch 6, batch 30400, giga_loss[loss=0.2907, simple_loss=0.3682, pruned_loss=0.1066, over 28821.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3703, pruned_loss=0.1216, over 5661294.36 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3897, pruned_loss=0.1412, over 5702378.27 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3696, pruned_loss=0.1204, over 5653002.34 frames. ], batch size: 227, lr: 5.27e-03, grad_scale: 8.0
+2023-03-03 08:43:58,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-03 08:44:01,370 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:01,658 INFO [optim.py:369] (1/2) 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 08:44:05,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:25,036 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:27,322 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 30450, giga_loss[loss=0.2523, simple_loss=0.3295, pruned_loss=0.08755, over 28304.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.367, pruned_loss=0.1173, over 5636864.44 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.389, pruned_loss=0.1412, over 5687335.95 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3666, pruned_loss=0.116, over 5641976.74 frames. ], batch size: 71, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:44:37,341 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:44:58,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:45:25,983 INFO [train.py:968] (1/2) Epoch 6, batch 30500, giga_loss[loss=0.2874, simple_loss=0.3563, pruned_loss=0.1092, over 27960.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5633030.47 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.3885, pruned_loss=0.141, over 5681342.39 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.368, pruned_loss=0.1169, over 5640499.95 frames. ], batch size: 412, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:45:54,080 INFO [optim.py:369] (1/2) 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 08:45:55,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:46:09,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5325, 1.9234, 1.3491, 1.2793], device='cuda:1'), covar=tensor([0.1448, 0.0930, 0.0897, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.1479, 0.1312, 0.1257, 0.1358], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 08:46:12,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 30550, giga_loss[loss=0.2616, simple_loss=0.3362, pruned_loss=0.09349, over 28431.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.1159, over 5629605.60 frames. ], libri_tot_loss[loss=0.335, simple_loss=0.3884, pruned_loss=0.1408, over 5682447.68 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3654, pruned_loss=0.115, over 5634319.88 frames. ], batch size: 85, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:46:38,371 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 30600, giga_loss[loss=0.2885, simple_loss=0.3489, pruned_loss=0.114, over 27523.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.362, pruned_loss=0.1134, over 5624753.95 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.388, pruned_loss=0.1407, over 5675458.21 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3619, pruned_loss=0.1125, over 5634308.94 frames. ], batch size: 472, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:47:21,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3141, 1.7564, 1.4946, 1.5542], device='cuda:1'), covar=tensor([0.0754, 0.0302, 0.0317, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0120, 0.0123, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0070], device='cuda:1')
+2023-03-03 08:47:34,559 INFO [optim.py:369] (1/2) 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 08:47:44,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:47:47,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258346.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:48:01,301 INFO [train.py:968] (1/2) Epoch 6, batch 30650, giga_loss[loss=0.3016, simple_loss=0.3744, pruned_loss=0.1144, over 28692.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3608, pruned_loss=0.1126, over 5625196.55 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.3878, pruned_loss=0.1406, over 5670855.32 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3605, pruned_loss=0.1115, over 5635643.92 frames. ], batch size: 242, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:48:17,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:48:20,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:48:48,893 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:48:49,318 INFO [train.py:968] (1/2) Epoch 6, batch 30700, giga_loss[loss=0.2971, simple_loss=0.3732, pruned_loss=0.1105, over 28307.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3607, pruned_loss=0.112, over 5633784.31 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3877, pruned_loss=0.1407, over 5674456.74 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3602, pruned_loss=0.1107, over 5638333.48 frames. ], batch size: 368, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:49:09,575 INFO [optim.py:369] (1/2) 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 08:49:34,118 INFO [train.py:968] (1/2) Epoch 6, batch 30750, giga_loss[loss=0.2688, simple_loss=0.3436, pruned_loss=0.09698, over 28894.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3591, pruned_loss=0.1104, over 5646074.35 frames. ], libri_tot_loss[loss=0.3339, simple_loss=0.3871, pruned_loss=0.1404, over 5679524.39 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3583, pruned_loss=0.1087, over 5644039.78 frames. ], batch size: 227, lr: 5.27e-03, grad_scale: 4.0
+2023-03-03 08:50:03,975 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258489.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:50:07,841 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 30800, giga_loss[loss=0.2599, simple_loss=0.3354, pruned_loss=0.09219, over 28567.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3554, pruned_loss=0.1071, over 5641988.16 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.387, pruned_loss=0.1403, over 5680785.77 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3548, pruned_loss=0.1056, over 5639254.96 frames. ], batch size: 307, lr: 5.27e-03, grad_scale: 8.0
+2023-03-03 08:50:38,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258521.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 08:50:51,959 INFO [optim.py:369] (1/2) 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 08:51:19,885 INFO [train.py:968] (1/2) Epoch 6, batch 30850, giga_loss[loss=0.2388, simple_loss=0.3201, pruned_loss=0.07878, over 29027.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3509, pruned_loss=0.1047, over 5634149.02 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3867, pruned_loss=0.1402, over 5674916.58 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3502, pruned_loss=0.1031, over 5636319.58 frames. ], batch size: 136, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:51:44,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9672, 1.2612, 0.9607, 0.1967], device='cuda:1'), covar=tensor([0.1493, 0.1382, 0.1937, 0.2607], device='cuda:1'), in_proj_covar=tensor([0.1400, 0.1325, 0.1361, 0.1165], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 08:52:05,553 INFO [train.py:968] (1/2) Epoch 6, batch 30900, giga_loss[loss=0.2482, simple_loss=0.3298, pruned_loss=0.08334, over 28839.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3505, pruned_loss=0.1055, over 5642586.51 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3858, pruned_loss=0.1397, over 5680246.55 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3497, pruned_loss=0.1036, over 5638552.52 frames. ], batch size: 174, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:52:20,767 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:52:46,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:52:56,055 INFO [train.py:968] (1/2) Epoch 6, batch 30950, libri_loss[loss=0.3208, simple_loss=0.3736, pruned_loss=0.134, over 28719.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3485, pruned_loss=0.1047, over 5626837.40 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3851, pruned_loss=0.1393, over 5676141.20 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3474, pruned_loss=0.1026, over 5625932.46 frames. ], batch size: 106, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:53:50,069 INFO [train.py:968] (1/2) Epoch 6, batch 31000, giga_loss[loss=0.2735, simple_loss=0.3584, pruned_loss=0.09433, over 29014.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3513, pruned_loss=0.1064, over 5621133.67 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3849, pruned_loss=0.1395, over 5672717.92 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3497, pruned_loss=0.1037, over 5621562.00 frames. ], batch size: 155, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:54:00,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=258718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:54:18,469 INFO [optim.py:369] (1/2) 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 08:54:35,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3076, 1.5173, 1.4699, 1.4892], device='cuda:1'), covar=tensor([0.1112, 0.1408, 0.1357, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0725, 0.0635, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 08:54:45,443 INFO [train.py:968] (1/2) Epoch 6, batch 31050, libri_loss[loss=0.2895, simple_loss=0.3509, pruned_loss=0.1141, over 29517.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3535, pruned_loss=0.1063, over 5638286.95 frames. ], libri_tot_loss[loss=0.3308, simple_loss=0.384, pruned_loss=0.1388, over 5677526.16 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3522, pruned_loss=0.1037, over 5632908.72 frames. ], batch size: 84, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:54:56,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:00,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:25,957 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:29,311 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:38,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:55:38,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3538, 1.8354, 1.7275, 1.4891], device='cuda:1'), covar=tensor([0.1509, 0.1800, 0.1167, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0714, 0.0796, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 08:55:52,224 INFO [train.py:968] (1/2) Epoch 6, batch 31100, giga_loss[loss=0.3117, simple_loss=0.371, pruned_loss=0.1262, over 28928.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3545, pruned_loss=0.1061, over 5657155.19 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3838, pruned_loss=0.1387, over 5680820.38 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3532, pruned_loss=0.1038, over 5649722.97 frames. ], batch size: 145, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:55:53,499 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-03 08:56:07,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:56:21,295 INFO [optim.py:369] (1/2) 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 08:56:36,835 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 31150, giga_loss[loss=0.2779, simple_loss=0.3517, pruned_loss=0.102, over 28993.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3523, pruned_loss=0.1048, over 5669478.03 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3834, pruned_loss=0.1385, over 5688510.29 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3507, pruned_loss=0.1021, over 5655933.39 frames. ], batch size: 285, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:56:52,649 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=258861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:56:56,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=258864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:57:28,572 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=258893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:57:42,607 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=258904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 08:57:52,037 INFO [train.py:968] (1/2) Epoch 6, batch 31200, giga_loss[loss=0.2583, simple_loss=0.3387, pruned_loss=0.08894, over 28424.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3517, pruned_loss=0.1043, over 5666596.96 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3826, pruned_loss=0.138, over 5692782.91 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3503, pruned_loss=0.1018, over 5651796.90 frames. ], batch size: 336, lr: 5.26e-03, grad_scale: 8.0
+2023-03-03 08:58:23,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2152, 1.1877, 1.0395, 0.9772], device='cuda:1'), covar=tensor([0.0700, 0.0486, 0.0997, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0438, 0.0494, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:58:25,353 INFO [optim.py:369] (1/2) 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 08:58:50,202 INFO [train.py:968] (1/2) Epoch 6, batch 31250, giga_loss[loss=0.2628, simple_loss=0.3402, pruned_loss=0.09265, over 28962.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3503, pruned_loss=0.1033, over 5669537.04 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3817, pruned_loss=0.1379, over 5695377.91 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3491, pruned_loss=0.1004, over 5654483.93 frames. ], batch size: 199, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 08:59:01,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7723, 2.5984, 2.0503, 2.3648], device='cuda:1'), covar=tensor([0.0563, 0.0539, 0.0777, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0439, 0.0494, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 08:59:18,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4355, 1.6415, 1.6173, 1.4702], device='cuda:1'), covar=tensor([0.1192, 0.1577, 0.1484, 0.1486], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0726, 0.0633, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 08:59:47,724 INFO [train.py:968] (1/2) Epoch 6, batch 31300, giga_loss[loss=0.3193, simple_loss=0.3718, pruned_loss=0.1334, over 28209.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3466, pruned_loss=0.102, over 5678767.02 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3811, pruned_loss=0.1375, over 5701174.96 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3453, pruned_loss=0.09908, over 5661022.33 frames. ], batch size: 412, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:00:20,435 INFO [optim.py:369] (1/2) 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 09:00:51,844 INFO [train.py:968] (1/2) Epoch 6, batch 31350, giga_loss[loss=0.2674, simple_loss=0.3373, pruned_loss=0.09877, over 28925.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3459, pruned_loss=0.1019, over 5672005.26 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3809, pruned_loss=0.1374, over 5702936.95 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3446, pruned_loss=0.09919, over 5656234.90 frames. ], batch size: 186, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:01:16,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1172, 1.3656, 1.1255, 0.9783], device='cuda:1'), covar=tensor([0.2127, 0.1993, 0.2175, 0.1837], device='cuda:1'), in_proj_covar=tensor([0.1145, 0.0866, 0.1016, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 09:01:49,341 INFO [train.py:968] (1/2) Epoch 6, batch 31400, giga_loss[loss=0.2957, simple_loss=0.3645, pruned_loss=0.1134, over 28350.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3449, pruned_loss=0.1008, over 5672600.79 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3807, pruned_loss=0.1373, over 5702611.89 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3436, pruned_loss=0.09815, over 5659763.32 frames. ], batch size: 368, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:02:20,219 INFO [optim.py:369] (1/2) 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 09:02:50,527 INFO [train.py:968] (1/2) Epoch 6, batch 31450, giga_loss[loss=0.2412, simple_loss=0.3227, pruned_loss=0.07983, over 27621.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3465, pruned_loss=0.1005, over 5671560.25 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3807, pruned_loss=0.1374, over 5705644.62 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.345, pruned_loss=0.09793, over 5658264.15 frames. ], batch size: 472, lr: 5.26e-03, grad_scale: 2.0
+2023-03-03 09:03:53,100 INFO [train.py:968] (1/2) Epoch 6, batch 31500, giga_loss[loss=0.227, simple_loss=0.3093, pruned_loss=0.07232, over 29060.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3457, pruned_loss=0.09979, over 5672627.03 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3803, pruned_loss=0.1372, over 5709689.91 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3442, pruned_loss=0.09717, over 5657686.52 frames. ], batch size: 128, lr: 5.26e-03, grad_scale: 2.0
+2023-03-03 09:04:03,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:04:24,219 INFO [optim.py:369] (1/2) 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 09:04:40,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5358, 3.4868, 1.5791, 1.5686], device='cuda:1'), covar=tensor([0.0787, 0.0280, 0.0807, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0486, 0.0319, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 09:04:53,495 INFO [train.py:968] (1/2) Epoch 6, batch 31550, giga_loss[loss=0.2816, simple_loss=0.3434, pruned_loss=0.1099, over 28217.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3439, pruned_loss=0.09939, over 5671616.70 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3796, pruned_loss=0.1372, over 5706357.24 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3419, pruned_loss=0.09583, over 5661015.60 frames. ], batch size: 412, lr: 5.26e-03, grad_scale: 2.0
+2023-03-03 09:05:18,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:05:49,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3453, 1.9628, 1.5175, 1.5353], device='cuda:1'), covar=tensor([0.0783, 0.0286, 0.0317, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0120, 0.0123, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 09:05:56,465 INFO [train.py:968] (1/2) Epoch 6, batch 31600, giga_loss[loss=0.27, simple_loss=0.3591, pruned_loss=0.09042, over 28679.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3472, pruned_loss=0.1017, over 5680383.93 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3795, pruned_loss=0.1374, over 5711214.82 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.0973, over 5666431.02 frames. ], batch size: 307, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:06:00,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4003, 3.9710, 1.3890, 1.5494], device='cuda:1'), covar=tensor([0.1115, 0.0270, 0.1057, 0.1498], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0488, 0.0321, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 09:06:07,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 09:06:29,785 INFO [optim.py:369] (1/2) 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 09:06:54,100 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:06:58,230 INFO [train.py:968] (1/2) Epoch 6, batch 31650, giga_loss[loss=0.1955, simple_loss=0.2714, pruned_loss=0.05981, over 24590.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.35, pruned_loss=0.1014, over 5663157.94 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3792, pruned_loss=0.1374, over 5712680.47 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3474, pruned_loss=0.09706, over 5649249.07 frames. ], batch size: 705, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:07:03,288 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:07:05,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:07:37,780 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 31700, giga_loss[loss=0.2638, simple_loss=0.3573, pruned_loss=0.08514, over 29109.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3533, pruned_loss=0.1021, over 5663111.14 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3789, pruned_loss=0.1374, over 5709972.07 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.35, pruned_loss=0.09677, over 5651546.04 frames. ], batch size: 200, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:08:09,646 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:08:13,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:08:25,912 INFO [optim.py:369] (1/2) 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 09:08:28,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-03 09:08:47,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:08:52,626 INFO [train.py:968] (1/2) Epoch 6, batch 31750, giga_loss[loss=0.2588, simple_loss=0.3497, pruned_loss=0.08398, over 28653.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3541, pruned_loss=0.1015, over 5664335.38 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3786, pruned_loss=0.1371, over 5713816.28 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3514, pruned_loss=0.09686, over 5650622.58 frames. ], batch size: 262, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:09:54,020 INFO [train.py:968] (1/2) Epoch 6, batch 31800, giga_loss[loss=0.2938, simple_loss=0.3708, pruned_loss=0.1084, over 28902.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3531, pruned_loss=0.1005, over 5665134.93 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3788, pruned_loss=0.1371, over 5715548.45 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3503, pruned_loss=0.0962, over 5652209.76 frames. ], batch size: 227, lr: 5.26e-03, grad_scale: 4.0
+2023-03-03 09:10:24,195 INFO [optim.py:369] (1/2) 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 09:10:55,974 INFO [train.py:968] (1/2) Epoch 6, batch 31850, giga_loss[loss=0.3347, simple_loss=0.3944, pruned_loss=0.1375, over 28356.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3535, pruned_loss=0.1018, over 5664089.44 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3784, pruned_loss=0.1369, over 5719032.34 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3512, pruned_loss=0.09796, over 5650124.16 frames. ], batch size: 368, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:12:12,646 INFO [train.py:968] (1/2) Epoch 6, batch 31900, giga_loss[loss=0.2669, simple_loss=0.3476, pruned_loss=0.09312, over 28893.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3536, pruned_loss=0.1027, over 5663209.96 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3783, pruned_loss=0.1368, over 5712281.66 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3516, pruned_loss=0.09921, over 5656586.63 frames. ], batch size: 145, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:12:23,211 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:12:54,475 INFO [optim.py:369] (1/2) 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 09:13:06,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2567, 1.8767, 1.3517, 0.3943], device='cuda:1'), covar=tensor([0.2580, 0.1444, 0.2251, 0.3056], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1342, 0.1391, 0.1186], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 09:13:21,661 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:13:25,542 INFO [train.py:968] (1/2) Epoch 6, batch 31950, giga_loss[loss=0.2243, simple_loss=0.3087, pruned_loss=0.06995, over 29136.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3523, pruned_loss=0.102, over 5675326.02 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3784, pruned_loss=0.1368, over 5713430.85 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.35, pruned_loss=0.09861, over 5668386.08 frames. ], batch size: 146, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:14:17,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2204, 4.0764, 3.8562, 1.7732], device='cuda:1'), covar=tensor([0.0493, 0.0590, 0.0708, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0832, 0.0766, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 09:14:39,708 INFO [train.py:968] (1/2) Epoch 6, batch 32000, giga_loss[loss=0.2607, simple_loss=0.3404, pruned_loss=0.09048, over 28592.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3488, pruned_loss=0.09983, over 5671302.67 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3783, pruned_loss=0.1368, over 5714454.69 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.347, pruned_loss=0.09695, over 5664684.83 frames. ], batch size: 307, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:15:08,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:15:13,137 INFO [optim.py:369] (1/2) 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 09:15:32,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5254, 1.4906, 1.3604, 1.6931], device='cuda:1'), covar=tensor([0.2152, 0.2081, 0.2040, 0.2218], device='cuda:1'), in_proj_covar=tensor([0.1151, 0.0876, 0.1025, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 09:15:43,794 INFO [train.py:968] (1/2) Epoch 6, batch 32050, libri_loss[loss=0.2768, simple_loss=0.3365, pruned_loss=0.1085, over 29546.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3468, pruned_loss=0.09945, over 5668892.36 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3779, pruned_loss=0.1366, over 5718098.22 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.345, pruned_loss=0.0965, over 5659212.87 frames. ], batch size: 77, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:16:44,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=259807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:16:48,071 INFO [train.py:968] (1/2) Epoch 6, batch 32100, giga_loss[loss=0.294, simple_loss=0.3729, pruned_loss=0.1075, over 29072.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.349, pruned_loss=0.1012, over 5673968.48 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3777, pruned_loss=0.1365, over 5722010.29 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3472, pruned_loss=0.09815, over 5661823.90 frames. ], batch size: 155, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:16:51,343 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-03 09:17:18,092 INFO [optim.py:369] (1/2) 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 09:17:42,780 INFO [train.py:968] (1/2) Epoch 6, batch 32150, giga_loss[loss=0.2849, simple_loss=0.3576, pruned_loss=0.1061, over 28102.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3527, pruned_loss=0.1033, over 5669871.77 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3774, pruned_loss=0.1366, over 5715450.26 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3506, pruned_loss=0.09959, over 5663893.32 frames. ], batch size: 412, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:18:02,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=259876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:18:05,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=259879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:18:49,343 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=259908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:18:51,466 INFO [train.py:968] (1/2) Epoch 6, batch 32200, giga_loss[loss=0.2796, simple_loss=0.3585, pruned_loss=0.1004, over 28838.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3511, pruned_loss=0.1031, over 5666892.63 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3771, pruned_loss=0.1365, over 5716249.01 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3495, pruned_loss=0.1001, over 5661338.42 frames. ], batch size: 243, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:19:22,473 INFO [optim.py:369] (1/2) 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 09:19:52,207 INFO [train.py:968] (1/2) Epoch 6, batch 32250, giga_loss[loss=0.3086, simple_loss=0.3723, pruned_loss=0.1225, over 28451.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3527, pruned_loss=0.1054, over 5668058.53 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3773, pruned_loss=0.1367, over 5717437.59 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.351, pruned_loss=0.1025, over 5661955.00 frames. ], batch size: 336, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:20:32,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=259994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:20:52,326 INFO [train.py:968] (1/2) Epoch 6, batch 32300, giga_loss[loss=0.2783, simple_loss=0.3573, pruned_loss=0.09961, over 28960.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3534, pruned_loss=0.1059, over 5672296.62 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3767, pruned_loss=0.1362, over 5719873.76 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3519, pruned_loss=0.1029, over 5663582.56 frames. ], batch size: 199, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:21:25,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:21:25,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 09:21:32,995 INFO [optim.py:369] (1/2) 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 09:22:07,971 INFO [train.py:968] (1/2) Epoch 6, batch 32350, giga_loss[loss=0.2734, simple_loss=0.3541, pruned_loss=0.09635, over 28912.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5669480.42 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3768, pruned_loss=0.1363, over 5719038.39 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3532, pruned_loss=0.1025, over 5663143.47 frames. ], batch size: 186, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:23:25,935 INFO [train.py:968] (1/2) Epoch 6, batch 32400, giga_loss[loss=0.2541, simple_loss=0.3364, pruned_loss=0.08593, over 28139.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3528, pruned_loss=0.1029, over 5671074.16 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3766, pruned_loss=0.1363, over 5721831.12 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3515, pruned_loss=0.1006, over 5663091.53 frames. ], batch size: 412, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:24:02,388 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:24:06,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:08,269 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:24:34,804 INFO [train.py:968] (1/2) Epoch 6, batch 32450, giga_loss[loss=0.3002, simple_loss=0.3662, pruned_loss=0.1172, over 28180.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.35, pruned_loss=0.1027, over 5676952.28 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3765, pruned_loss=0.1363, over 5723772.70 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3488, pruned_loss=0.1005, over 5668556.47 frames. ], batch size: 412, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:24:47,358 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:55,981 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:24:56,066 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:25:00,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:25:04,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:25:32,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-03 09:25:40,212 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 32500, giga_loss[loss=0.2354, simple_loss=0.3139, pruned_loss=0.07846, over 28615.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.343, pruned_loss=0.09919, over 5674185.20 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3766, pruned_loss=0.1363, over 5721056.48 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3418, pruned_loss=0.09715, over 5669536.91 frames. ], batch size: 307, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:26:12,175 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-03 09:26:23,864 INFO [optim.py:369] (1/2) 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 09:26:50,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:26:51,733 INFO [train.py:968] (1/2) Epoch 6, batch 32550, giga_loss[loss=0.2919, simple_loss=0.356, pruned_loss=0.1139, over 27567.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3421, pruned_loss=0.09894, over 5660594.84 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3767, pruned_loss=0.1363, over 5721535.20 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3409, pruned_loss=0.09716, over 5656397.66 frames. ], batch size: 472, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:26:59,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3767, 3.3241, 1.5161, 1.5042], device='cuda:1'), covar=tensor([0.0870, 0.0310, 0.0881, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0482, 0.0320, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 09:27:51,054 INFO [train.py:968] (1/2) Epoch 6, batch 32600, giga_loss[loss=0.2858, simple_loss=0.3445, pruned_loss=0.1136, over 27006.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3441, pruned_loss=0.1009, over 5657681.21 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.376, pruned_loss=0.136, over 5724246.17 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3432, pruned_loss=0.09917, over 5650725.13 frames. ], batch size: 555, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:28:08,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:28:11,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:28:22,879 INFO [optim.py:369] (1/2) 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 09:28:46,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:28:49,190 INFO [train.py:968] (1/2) Epoch 6, batch 32650, giga_loss[loss=0.2505, simple_loss=0.3235, pruned_loss=0.08872, over 27570.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3421, pruned_loss=0.09942, over 5646456.59 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3761, pruned_loss=0.1361, over 5709784.28 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3404, pruned_loss=0.09702, over 5651755.20 frames. ], batch size: 472, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:29:51,094 INFO [train.py:968] (1/2) Epoch 6, batch 32700, giga_loss[loss=0.2404, simple_loss=0.3265, pruned_loss=0.07714, over 28414.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3402, pruned_loss=0.0974, over 5657542.71 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3753, pruned_loss=0.1358, over 5714157.04 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3389, pruned_loss=0.09505, over 5656645.47 frames. ], batch size: 336, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:30:08,445 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:30:30,599 INFO [optim.py:369] (1/2) 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 09:30:59,553 INFO [train.py:968] (1/2) Epoch 6, batch 32750, giga_loss[loss=0.2821, simple_loss=0.347, pruned_loss=0.1086, over 27582.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3392, pruned_loss=0.0972, over 5660739.16 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3749, pruned_loss=0.1356, over 5714376.37 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.338, pruned_loss=0.09493, over 5658908.25 frames. ], batch size: 472, lr: 5.25e-03, grad_scale: 4.0
+2023-03-03 09:31:44,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5772, 1.7364, 1.4840, 1.7623], device='cuda:1'), covar=tensor([0.2082, 0.1734, 0.1802, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.1143, 0.0869, 0.1026, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 09:32:08,392 INFO [train.py:968] (1/2) Epoch 6, batch 32800, giga_loss[loss=0.2568, simple_loss=0.3385, pruned_loss=0.08758, over 28900.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3388, pruned_loss=0.09616, over 5653182.08 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3753, pruned_loss=0.1359, over 5713425.44 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.337, pruned_loss=0.09355, over 5651802.20 frames. ], batch size: 164, lr: 5.25e-03, grad_scale: 8.0
+2023-03-03 09:32:18,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260517.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:32:50,134 INFO [optim.py:369] (1/2) 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 09:33:02,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:33:16,140 INFO [train.py:968] (1/2) Epoch 6, batch 32850, giga_loss[loss=0.3358, simple_loss=0.3919, pruned_loss=0.1398, over 28657.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3386, pruned_loss=0.0959, over 5656624.66 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3749, pruned_loss=0.1356, over 5715440.46 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3371, pruned_loss=0.09372, over 5653142.75 frames. ], batch size: 262, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:34:12,804 INFO [train.py:968] (1/2) Epoch 6, batch 32900, giga_loss[loss=0.2779, simple_loss=0.3444, pruned_loss=0.1057, over 28131.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3406, pruned_loss=0.0985, over 5655029.99 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3737, pruned_loss=0.1349, over 5709807.53 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3391, pruned_loss=0.09583, over 5654403.06 frames. ], batch size: 412, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:34:22,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=260617.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:34:41,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:34:46,725 INFO [optim.py:369] (1/2) 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 09:35:15,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260660.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:35:15,530 INFO [train.py:968] (1/2) Epoch 6, batch 32950, giga_loss[loss=0.2891, simple_loss=0.3658, pruned_loss=0.1062, over 28888.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3384, pruned_loss=0.09677, over 5655114.67 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3734, pruned_loss=0.1347, over 5713694.37 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.337, pruned_loss=0.0942, over 5650318.41 frames. ], batch size: 186, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:35:18,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260663.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:35:47,316 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260692.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:35:48,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:35:52,628 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 33000, giga_loss[loss=0.2536, simple_loss=0.3435, pruned_loss=0.08181, over 28673.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3408, pruned_loss=0.09636, over 5663054.96 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3729, pruned_loss=0.1343, over 5718168.39 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3389, pruned_loss=0.0934, over 5653798.11 frames. ], batch size: 262, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:36:09,857 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 09:36:18,234 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 09:36:34,251 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:36:50,273 INFO [optim.py:369] (1/2) 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 09:37:15,042 INFO [train.py:968] (1/2) Epoch 6, batch 33050, giga_loss[loss=0.3278, simple_loss=0.3776, pruned_loss=0.139, over 26888.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3435, pruned_loss=0.09734, over 5657694.78 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3726, pruned_loss=0.1342, over 5713840.31 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3416, pruned_loss=0.09427, over 5652556.16 frames. ], batch size: 555, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:37:36,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:37:41,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:38:05,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:38:18,951 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:38:19,922 INFO [train.py:968] (1/2) Epoch 6, batch 33100, giga_loss[loss=0.2758, simple_loss=0.3533, pruned_loss=0.09917, over 28690.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3446, pruned_loss=0.09794, over 5649432.37 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3727, pruned_loss=0.1342, over 5714820.11 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3429, pruned_loss=0.09537, over 5644113.88 frames. ], batch size: 262, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:39:00,518 INFO [optim.py:369] (1/2) 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 09:39:22,509 INFO [train.py:968] (1/2) Epoch 6, batch 33150, libri_loss[loss=0.3079, simple_loss=0.3659, pruned_loss=0.1249, over 27711.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.346, pruned_loss=0.09941, over 5657102.72 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.373, pruned_loss=0.1346, over 5717797.16 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3437, pruned_loss=0.09611, over 5648704.73 frames. ], batch size: 116, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:39:30,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4507, 1.4797, 1.2167, 1.9704], device='cuda:1'), covar=tensor([0.2191, 0.2139, 0.2349, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1133, 0.0864, 0.1012, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 09:40:18,352 INFO [train.py:968] (1/2) Epoch 6, batch 33200, giga_loss[loss=0.2616, simple_loss=0.3422, pruned_loss=0.09052, over 28690.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3433, pruned_loss=0.09742, over 5665113.25 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3723, pruned_loss=0.1342, over 5720788.90 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3416, pruned_loss=0.09453, over 5654724.28 frames. ], batch size: 262, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:40:49,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4244, 1.5505, 1.0847, 1.2831], device='cuda:1'), covar=tensor([0.0701, 0.0530, 0.1208, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0437, 0.0499, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 09:40:54,286 INFO [optim.py:369] (1/2) 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 09:40:56,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-03 09:40:58,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=260941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:41:01,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=260944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:41:18,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-03 09:41:19,521 INFO [train.py:968] (1/2) Epoch 6, batch 33250, giga_loss[loss=0.2653, simple_loss=0.3402, pruned_loss=0.09523, over 28606.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3424, pruned_loss=0.09733, over 5659331.49 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.372, pruned_loss=0.134, over 5711169.93 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3407, pruned_loss=0.09442, over 5657839.42 frames. ], batch size: 307, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:41:30,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0662, 1.2211, 3.2895, 2.8692], device='cuda:1'), covar=tensor([0.1468, 0.2269, 0.0459, 0.1491], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0538, 0.0759, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 09:41:34,125 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=260973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:41:55,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=260992.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:42:06,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-03 09:42:13,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5412, 3.6005, 1.5495, 1.4278], device='cuda:1'), covar=tensor([0.0848, 0.0265, 0.0880, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0485, 0.0321, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 09:42:17,271 INFO [train.py:968] (1/2) Epoch 6, batch 33300, giga_loss[loss=0.2533, simple_loss=0.3365, pruned_loss=0.08506, over 29024.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3404, pruned_loss=0.09679, over 5656789.67 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3721, pruned_loss=0.1341, over 5704772.32 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3381, pruned_loss=0.09346, over 5660440.24 frames. ], batch size: 285, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:42:55,508 INFO [optim.py:369] (1/2) 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 09:43:21,875 INFO [train.py:968] (1/2) Epoch 6, batch 33350, giga_loss[loss=0.2742, simple_loss=0.3536, pruned_loss=0.09737, over 28664.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3436, pruned_loss=0.09819, over 5660537.45 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3717, pruned_loss=0.1338, over 5705416.57 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3419, pruned_loss=0.09548, over 5662438.31 frames. ], batch size: 242, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:44:24,806 INFO [train.py:968] (1/2) Epoch 6, batch 33400, giga_loss[loss=0.2968, simple_loss=0.3691, pruned_loss=0.1122, over 28509.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3441, pruned_loss=0.0984, over 5659206.00 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3714, pruned_loss=0.1336, over 5698431.43 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3426, pruned_loss=0.09586, over 5665404.89 frames. ], batch size: 336, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:44:37,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3317, 1.5572, 1.0938, 1.0931], device='cuda:1'), covar=tensor([0.1068, 0.0871, 0.0766, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.1489, 0.1278, 0.1238, 0.1353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 09:44:39,119 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:44:52,776 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261135.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 09:44:56,947 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261138.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:45:00,562 INFO [optim.py:369] (1/2) 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 09:45:03,575 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:45:23,790 INFO [train.py:968] (1/2) Epoch 6, batch 33450, giga_loss[loss=0.2657, simple_loss=0.3437, pruned_loss=0.09391, over 29053.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3447, pruned_loss=0.09909, over 5653996.05 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3704, pruned_loss=0.1328, over 5691512.69 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3435, pruned_loss=0.09668, over 5663875.60 frames. ], batch size: 136, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:45:25,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3417, 1.5415, 1.1186, 1.1654], device='cuda:1'), covar=tensor([0.1123, 0.0882, 0.0837, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.1489, 0.1276, 0.1236, 0.1353], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 09:45:26,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-03 09:45:30,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 33500, giga_loss[loss=0.2516, simple_loss=0.332, pruned_loss=0.08561, over 28786.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3478, pruned_loss=0.1003, over 5654631.17 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3704, pruned_loss=0.1328, over 5693669.09 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3467, pruned_loss=0.09812, over 5660378.15 frames. ], batch size: 99, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:46:42,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1039, 1.2315, 3.4631, 2.8217], device='cuda:1'), covar=tensor([0.1465, 0.2279, 0.0418, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0532, 0.0753, 0.0624], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 09:46:50,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8174, 1.7481, 1.3946, 1.6139], device='cuda:1'), covar=tensor([0.0474, 0.0443, 0.0749, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0442, 0.0503, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 09:46:59,532 INFO [optim.py:369] (1/2) 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 09:47:10,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5121, 1.6194, 1.4499, 1.8392], device='cuda:1'), covar=tensor([0.2089, 0.2032, 0.2068, 0.1883], device='cuda:1'), in_proj_covar=tensor([0.1134, 0.0861, 0.1013, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 09:47:11,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0246, 1.2789, 0.9848, 0.4176], device='cuda:1'), covar=tensor([0.1563, 0.1376, 0.2232, 0.2414], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1351, 0.1383, 0.1183], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 09:47:21,710 INFO [train.py:968] (1/2) Epoch 6, batch 33550, giga_loss[loss=0.2664, simple_loss=0.3438, pruned_loss=0.09446, over 28928.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3497, pruned_loss=0.1007, over 5656395.26 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3709, pruned_loss=0.1332, over 5695194.94 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3479, pruned_loss=0.09787, over 5658659.79 frames. ], batch size: 136, lr: 5.24e-03, grad_scale: 4.0
+2023-03-03 09:47:43,512 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 33600, giga_loss[loss=0.2804, simple_loss=0.3273, pruned_loss=0.1168, over 24465.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3485, pruned_loss=0.1003, over 5651165.58 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3706, pruned_loss=0.133, over 5695769.88 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.09768, over 5651535.45 frames. ], batch size: 705, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:48:49,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-03 09:49:15,907 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 33650, giga_loss[loss=0.2874, simple_loss=0.3581, pruned_loss=0.1083, over 28551.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3463, pruned_loss=0.0992, over 5664075.81 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3705, pruned_loss=0.1329, over 5698631.18 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.09671, over 5661271.18 frames. ], batch size: 307, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:50:48,537 INFO [train.py:968] (1/2) Epoch 6, batch 33700, libri_loss[loss=0.2824, simple_loss=0.3348, pruned_loss=0.115, over 29378.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3465, pruned_loss=0.09968, over 5646181.37 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3703, pruned_loss=0.1329, over 5691364.36 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3452, pruned_loss=0.09728, over 5649916.12 frames. ], batch size: 71, lr: 5.24e-03, grad_scale: 8.0
+2023-03-03 09:51:27,696 INFO [optim.py:369] (1/2) 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 09:51:35,305 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5446, 1.5396, 1.2635, 1.2816], device='cuda:1'), covar=tensor([0.0625, 0.0467, 0.0949, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0439, 0.0502, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 09:51:53,214 INFO [train.py:968] (1/2) Epoch 6, batch 33750, giga_loss[loss=0.2541, simple_loss=0.3282, pruned_loss=0.09002, over 28624.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3463, pruned_loss=0.1002, over 5649903.81 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3702, pruned_loss=0.1326, over 5693976.31 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.345, pruned_loss=0.09791, over 5649800.80 frames. ], batch size: 262, lr: 5.24e-03, grad_scale: 2.0
+2023-03-03 09:52:21,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4078, 4.2600, 3.9973, 1.7933], device='cuda:1'), covar=tensor([0.0496, 0.0699, 0.0817, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0831, 0.0768, 0.0596], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 09:52:23,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7369, 2.3716, 1.5465, 1.4190], device='cuda:1'), covar=tensor([0.1737, 0.1052, 0.1120, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.1506, 0.1277, 0.1239, 0.1362], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 09:52:37,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:52:52,890 INFO [train.py:968] (1/2) Epoch 6, batch 33800, giga_loss[loss=0.2414, simple_loss=0.3278, pruned_loss=0.07747, over 28832.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3441, pruned_loss=0.1, over 5643788.29 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3691, pruned_loss=0.132, over 5688557.31 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3431, pruned_loss=0.09768, over 5646562.74 frames. ], batch size: 174, lr: 5.24e-03, grad_scale: 2.0
+2023-03-03 09:52:53,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-03 09:52:54,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6373, 3.2296, 2.0664, 1.9316], device='cuda:1'), covar=tensor([0.1050, 0.0646, 0.0799, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.1505, 0.1278, 0.1240, 0.1364], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 09:53:04,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:53:17,530 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261529.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 09:53:33,434 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 33850, giga_loss[loss=0.2329, simple_loss=0.3217, pruned_loss=0.07199, over 28861.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3449, pruned_loss=0.1005, over 5632118.64 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3693, pruned_loss=0.1322, over 5681061.21 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3435, pruned_loss=0.0978, over 5640410.52 frames. ], batch size: 164, lr: 5.23e-03, grad_scale: 2.0
+2023-03-03 09:54:53,731 INFO [train.py:968] (1/2) Epoch 6, batch 33900, giga_loss[loss=0.3244, simple_loss=0.3853, pruned_loss=0.1318, over 28663.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3436, pruned_loss=0.09866, over 5658779.46 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3689, pruned_loss=0.1319, over 5687791.61 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3418, pruned_loss=0.09557, over 5658006.77 frames. ], batch size: 242, lr: 5.23e-03, grad_scale: 2.0
+2023-03-03 09:55:11,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7995, 2.7087, 1.7878, 0.7123], device='cuda:1'), covar=tensor([0.4159, 0.1855, 0.2471, 0.4041], device='cuda:1'), in_proj_covar=tensor([0.1414, 0.1344, 0.1381, 0.1178], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 09:55:26,279 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:55:30,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:55:33,732 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0541, 1.2564, 4.0171, 2.9907], device='cuda:1'), covar=tensor([0.1665, 0.2279, 0.0308, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0577, 0.0536, 0.0755, 0.0624], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 09:55:37,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=261649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:55:51,177 INFO [train.py:968] (1/2) Epoch 6, batch 33950, giga_loss[loss=0.244, simple_loss=0.341, pruned_loss=0.07355, over 28996.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3454, pruned_loss=0.09719, over 5671699.24 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3686, pruned_loss=0.1318, over 5690548.25 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.344, pruned_loss=0.09444, over 5668272.71 frames. ], batch size: 145, lr: 5.23e-03, grad_scale: 2.0
+2023-03-03 09:55:52,163 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:55:55,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:56:05,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:56:09,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3026, 1.6345, 1.2404, 0.7093], device='cuda:1'), covar=tensor([0.2563, 0.1512, 0.1741, 0.2624], device='cuda:1'), in_proj_covar=tensor([0.1414, 0.1345, 0.1380, 0.1176], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 09:56:27,947 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:56:44,644 INFO [train.py:968] (1/2) Epoch 6, batch 34000, giga_loss[loss=0.2789, simple_loss=0.3554, pruned_loss=0.1012, over 27977.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.09772, over 5659061.49 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3688, pruned_loss=0.1319, over 5685031.49 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3451, pruned_loss=0.09417, over 5660830.10 frames. ], batch size: 412, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 09:57:13,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1775, 1.4885, 1.2039, 1.0023], device='cuda:1'), covar=tensor([0.1342, 0.0947, 0.0805, 0.1057], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1285, 0.1254, 0.1374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 09:57:24,177 INFO [optim.py:369] (1/2) 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 09:57:48,496 INFO [train.py:968] (1/2) Epoch 6, batch 34050, giga_loss[loss=0.2416, simple_loss=0.3351, pruned_loss=0.07408, over 28878.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3464, pruned_loss=0.09685, over 5654114.78 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3683, pruned_loss=0.1316, over 5677953.13 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3448, pruned_loss=0.09367, over 5661376.27 frames. ], batch size: 174, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 09:58:31,461 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=261792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:58:35,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=261795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:58:41,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-03 09:58:57,384 INFO [train.py:968] (1/2) Epoch 6, batch 34100, giga_loss[loss=0.2488, simple_loss=0.3374, pruned_loss=0.08007, over 28759.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3472, pruned_loss=0.09768, over 5656500.91 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3683, pruned_loss=0.1318, over 5673779.57 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3455, pruned_loss=0.09424, over 5664774.03 frames. ], batch size: 243, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 09:59:14,323 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=261824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 09:59:34,463 INFO [optim.py:369] (1/2) 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 09:59:45,799 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=261849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:00:03,770 INFO [train.py:968] (1/2) Epoch 6, batch 34150, giga_loss[loss=0.2879, simple_loss=0.3626, pruned_loss=0.1066, over 29035.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.09769, over 5658045.13 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3683, pruned_loss=0.1316, over 5678203.01 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3453, pruned_loss=0.09433, over 5660497.71 frames. ], batch size: 200, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:01:01,685 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 34200, giga_loss[loss=0.2692, simple_loss=0.355, pruned_loss=0.09174, over 28783.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3466, pruned_loss=0.09695, over 5661741.78 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3678, pruned_loss=0.1313, over 5683116.25 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.345, pruned_loss=0.0936, over 5658749.98 frames. ], batch size: 243, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:01:58,900 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 34250, giga_loss[loss=0.2569, simple_loss=0.3424, pruned_loss=0.08572, over 29088.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3475, pruned_loss=0.09718, over 5657009.53 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3676, pruned_loss=0.131, over 5685684.51 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.346, pruned_loss=0.09411, over 5651764.39 frames. ], batch size: 93, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:02:50,493 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3069, 3.1560, 2.9600, 1.3567], device='cuda:1'), covar=tensor([0.0758, 0.0784, 0.0938, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0823, 0.0758, 0.0592], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 10:03:22,818 INFO [train.py:968] (1/2) Epoch 6, batch 34300, giga_loss[loss=0.262, simple_loss=0.3488, pruned_loss=0.08759, over 28848.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3506, pruned_loss=0.0986, over 5675202.94 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3669, pruned_loss=0.1305, over 5693554.67 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3495, pruned_loss=0.09557, over 5663080.87 frames. ], batch size: 243, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:04:06,765 INFO [optim.py:369] (1/2) 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 10:04:12,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262047.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:04:19,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262050.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:04:30,964 INFO [train.py:968] (1/2) Epoch 6, batch 34350, giga_loss[loss=0.2596, simple_loss=0.3346, pruned_loss=0.09234, over 29196.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3505, pruned_loss=0.099, over 5685229.66 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3666, pruned_loss=0.1302, over 5697179.17 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3496, pruned_loss=0.0963, over 5672066.79 frames. ], batch size: 200, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:04:45,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0083, 5.8026, 5.4790, 2.6999], device='cuda:1'), covar=tensor([0.0435, 0.0655, 0.0843, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0814, 0.0752, 0.0587], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 10:04:58,286 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:04:58,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1347, 2.3499, 1.2707, 1.2172], device='cuda:1'), covar=tensor([0.0843, 0.0437, 0.0790, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0477, 0.0318, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 10:05:35,248 INFO [train.py:968] (1/2) Epoch 6, batch 34400, giga_loss[loss=0.2832, simple_loss=0.3546, pruned_loss=0.1059, over 27755.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3488, pruned_loss=0.09886, over 5695764.11 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3671, pruned_loss=0.1304, over 5704744.99 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.347, pruned_loss=0.0953, over 5678099.47 frames. ], batch size: 476, lr: 5.23e-03, grad_scale: 8.0
+2023-03-03 10:06:02,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5917, 1.7497, 1.4364, 1.3249], device='cuda:1'), covar=tensor([0.1425, 0.1037, 0.0831, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.1473, 0.1243, 0.1213, 0.1339], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 10:06:24,072 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 34450, giga_loss[loss=0.3007, simple_loss=0.3738, pruned_loss=0.1138, over 28890.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3461, pruned_loss=0.09633, over 5694482.85 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.367, pruned_loss=0.1303, over 5703431.89 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3444, pruned_loss=0.09306, over 5681269.10 frames. ], batch size: 164, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:07:01,150 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-03 10:07:51,634 INFO [train.py:968] (1/2) Epoch 6, batch 34500, giga_loss[loss=0.2408, simple_loss=0.3227, pruned_loss=0.07948, over 28976.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3452, pruned_loss=0.09573, over 5694712.64 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3674, pruned_loss=0.1306, over 5694056.35 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.343, pruned_loss=0.09198, over 5691537.61 frames. ], batch size: 136, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:08:00,086 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:08:10,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:08:12,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:08:32,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4230, 1.5796, 1.3278, 1.6314], device='cuda:1'), covar=tensor([0.2134, 0.1953, 0.2140, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.1140, 0.0865, 0.1018, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 10:08:33,542 INFO [optim.py:369] (1/2) 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 10:08:53,984 INFO [train.py:968] (1/2) Epoch 6, batch 34550, giga_loss[loss=0.23, simple_loss=0.3173, pruned_loss=0.0714, over 29056.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3457, pruned_loss=0.09617, over 5686954.80 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.367, pruned_loss=0.1303, over 5695597.79 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3441, pruned_loss=0.09297, over 5683075.94 frames. ], batch size: 128, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:09:48,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4033, 1.9000, 1.3474, 0.7413], device='cuda:1'), covar=tensor([0.3061, 0.1747, 0.2181, 0.3095], device='cuda:1'), in_proj_covar=tensor([0.1409, 0.1338, 0.1371, 0.1168], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 10:09:58,121 INFO [train.py:968] (1/2) Epoch 6, batch 34600, giga_loss[loss=0.2844, simple_loss=0.3584, pruned_loss=0.1052, over 28247.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3488, pruned_loss=0.09797, over 5672745.47 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3667, pruned_loss=0.1301, over 5696697.80 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3476, pruned_loss=0.09548, over 5668707.44 frames. ], batch size: 412, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:10:17,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3783, 3.1910, 3.0565, 1.8772], device='cuda:1'), covar=tensor([0.0668, 0.0766, 0.0839, 0.1730], device='cuda:1'), in_proj_covar=tensor([0.0878, 0.0812, 0.0748, 0.0583], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0009, 0.0009], device='cuda:1')
+2023-03-03 10:10:33,737 INFO [optim.py:369] (1/2) 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 10:10:54,138 INFO [train.py:968] (1/2) Epoch 6, batch 34650, giga_loss[loss=0.2356, simple_loss=0.3128, pruned_loss=0.07921, over 28994.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3487, pruned_loss=0.09925, over 5676090.13 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.367, pruned_loss=0.1304, over 5701737.77 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3471, pruned_loss=0.09633, over 5667841.91 frames. ], batch size: 106, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:11:04,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:07,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=262370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:07,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:20,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2306, 2.4630, 1.2490, 1.2886], device='cuda:1'), covar=tensor([0.0902, 0.0374, 0.0842, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0477, 0.0319, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 10:11:37,336 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:11:52,284 INFO [train.py:968] (1/2) Epoch 6, batch 34700, giga_loss[loss=0.2607, simple_loss=0.3425, pruned_loss=0.08941, over 28862.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3463, pruned_loss=0.09865, over 5676064.10 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3667, pruned_loss=0.1302, over 5702544.27 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3451, pruned_loss=0.09608, over 5668471.01 frames. ], batch size: 227, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:12:27,230 INFO [optim.py:369] (1/2) 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 10:12:45,976 INFO [train.py:968] (1/2) Epoch 6, batch 34750, giga_loss[loss=0.3068, simple_loss=0.3832, pruned_loss=0.1152, over 28888.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3476, pruned_loss=0.1003, over 5666210.21 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3666, pruned_loss=0.13, over 5695614.86 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3462, pruned_loss=0.09761, over 5664975.61 frames. ], batch size: 284, lr: 5.23e-03, grad_scale: 4.0
+2023-03-03 10:13:35,209 INFO [train.py:968] (1/2) Epoch 6, batch 34800, giga_loss[loss=0.3024, simple_loss=0.3761, pruned_loss=0.1144, over 28856.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3565, pruned_loss=0.1067, over 5663857.76 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3665, pruned_loss=0.1301, over 5696336.10 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3551, pruned_loss=0.1039, over 5661529.63 frames. ], batch size: 99, lr: 5.23e-03, grad_scale: 8.0
+2023-03-03 10:14:01,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-03 10:14:06,520 INFO [optim.py:369] (1/2) 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 10:14:10,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3803, 1.9714, 1.4532, 0.7410], device='cuda:1'), covar=tensor([0.2830, 0.1597, 0.1982, 0.2956], device='cuda:1'), in_proj_covar=tensor([0.1429, 0.1360, 0.1389, 0.1180], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 10:14:21,978 INFO [train.py:968] (1/2) Epoch 6, batch 34850, giga_loss[loss=0.3029, simple_loss=0.3836, pruned_loss=0.1111, over 28923.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3665, pruned_loss=0.1126, over 5678800.44 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3667, pruned_loss=0.1302, over 5700301.12 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3652, pruned_loss=0.1099, over 5672961.40 frames. ], batch size: 227, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:14:51,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:14:59,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:15:06,853 INFO [train.py:968] (1/2) Epoch 6, batch 34900, giga_loss[loss=0.2618, simple_loss=0.3346, pruned_loss=0.09454, over 28907.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3682, pruned_loss=0.1141, over 5670553.26 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3668, pruned_loss=0.1303, over 5689307.84 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3671, pruned_loss=0.1118, over 5675250.61 frames. ], batch size: 145, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:15:36,885 INFO [optim.py:369] (1/2) 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 10:15:49,883 INFO [train.py:968] (1/2) Epoch 6, batch 34950, giga_loss[loss=0.253, simple_loss=0.319, pruned_loss=0.09346, over 28608.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3635, pruned_loss=0.1127, over 5676202.85 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3673, pruned_loss=0.1306, over 5694251.33 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3622, pruned_loss=0.1102, over 5675076.06 frames. ], batch size: 78, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:16:23,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3424, 2.3461, 2.2592, 2.1857], device='cuda:1'), covar=tensor([0.0378, 0.0513, 0.0663, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0441, 0.0503, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 10:16:31,951 INFO [train.py:968] (1/2) Epoch 6, batch 35000, giga_loss[loss=0.2569, simple_loss=0.3227, pruned_loss=0.09554, over 28272.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3564, pruned_loss=0.1094, over 5677715.79 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3676, pruned_loss=0.1306, over 5698285.38 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3548, pruned_loss=0.1068, over 5672709.85 frames. ], batch size: 77, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:16:36,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2634, 3.0919, 2.9545, 1.4107], device='cuda:1'), covar=tensor([0.0745, 0.0843, 0.0867, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0830, 0.0770, 0.0599], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 10:16:46,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-03 10:16:52,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:16:54,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:00,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:00,526 INFO [optim.py:369] (1/2) 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 10:17:00,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=262745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:02,595 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 35050, giga_loss[loss=0.2251, simple_loss=0.2951, pruned_loss=0.0775, over 28995.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3486, pruned_loss=0.1059, over 5689543.23 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3672, pruned_loss=0.1302, over 5701014.46 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3474, pruned_loss=0.1037, over 5682850.88 frames. ], batch size: 106, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:17:17,035 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4965, 2.2071, 1.6315, 1.9958], device='cuda:1'), covar=tensor([0.0629, 0.0225, 0.0275, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0120, 0.0123, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 10:17:19,771 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:27,456 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:17:48,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5195, 1.6967, 1.2918, 1.2002], device='cuda:1'), covar=tensor([0.1366, 0.1026, 0.0829, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.1486, 0.1268, 0.1251, 0.1374], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 10:17:55,731 INFO [train.py:968] (1/2) Epoch 6, batch 35100, giga_loss[loss=0.2673, simple_loss=0.3321, pruned_loss=0.1012, over 29044.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3405, pruned_loss=0.1025, over 5689179.34 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.367, pruned_loss=0.13, over 5704400.52 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3394, pruned_loss=0.1005, over 5680786.31 frames. ], batch size: 128, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:18:24,089 INFO [optim.py:369] (1/2) 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 10:18:38,003 INFO [train.py:968] (1/2) Epoch 6, batch 35150, giga_loss[loss=0.235, simple_loss=0.3014, pruned_loss=0.08427, over 28872.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3347, pruned_loss=0.09978, over 5683552.66 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3677, pruned_loss=0.1305, over 5701970.70 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3326, pruned_loss=0.09715, over 5678883.76 frames. ], batch size: 186, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:19:03,589 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=262888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:19:06,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=262891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:19:19,710 INFO [train.py:968] (1/2) Epoch 6, batch 35200, giga_loss[loss=0.2832, simple_loss=0.3422, pruned_loss=0.1121, over 28298.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3318, pruned_loss=0.09864, over 5696872.40 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3685, pruned_loss=0.1311, over 5704882.58 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3288, pruned_loss=0.09552, over 5690319.93 frames. ], batch size: 368, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:19:29,390 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=262920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:19:51,413 INFO [optim.py:369] (1/2) 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 10:20:02,903 INFO [train.py:968] (1/2) Epoch 6, batch 35250, giga_loss[loss=0.2244, simple_loss=0.2975, pruned_loss=0.07562, over 28937.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.328, pruned_loss=0.09664, over 5689208.41 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3691, pruned_loss=0.1315, over 5699585.96 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3244, pruned_loss=0.09321, over 5689165.87 frames. ], batch size: 136, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:20:11,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-03 10:20:42,882 INFO [train.py:968] (1/2) Epoch 6, batch 35300, giga_loss[loss=0.2225, simple_loss=0.3004, pruned_loss=0.07226, over 28727.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3267, pruned_loss=0.09638, over 5686341.75 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3697, pruned_loss=0.1316, over 5703313.13 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.322, pruned_loss=0.09241, over 5682531.94 frames. ], batch size: 262, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:20:54,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3947, 1.6794, 1.3446, 1.5631], device='cuda:1'), covar=tensor([0.2118, 0.2069, 0.2178, 0.2024], device='cuda:1'), in_proj_covar=tensor([0.1136, 0.0874, 0.1013, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 10:21:02,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2964, 1.6049, 1.2962, 1.5257], device='cuda:1'), covar=tensor([0.0743, 0.0356, 0.0331, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0119, 0.0123, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 10:21:14,992 INFO [optim.py:369] (1/2) 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 10:21:26,273 INFO [train.py:968] (1/2) Epoch 6, batch 35350, giga_loss[loss=0.2022, simple_loss=0.2748, pruned_loss=0.06484, over 28885.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3241, pruned_loss=0.09529, over 5682544.72 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3703, pruned_loss=0.1319, over 5706020.33 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3185, pruned_loss=0.09087, over 5676187.45 frames. ], batch size: 99, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:21:57,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-03 10:22:07,924 INFO [train.py:968] (1/2) Epoch 6, batch 35400, giga_loss[loss=0.2064, simple_loss=0.2787, pruned_loss=0.06707, over 28466.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3207, pruned_loss=0.09322, over 5691217.72 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3705, pruned_loss=0.1318, over 5710294.57 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3153, pruned_loss=0.0891, over 5681929.37 frames. ], batch size: 78, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:22:35,979 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:22:49,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1933, 1.3165, 3.4952, 3.1571], device='cuda:1'), covar=tensor([0.1311, 0.2134, 0.0381, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0577, 0.0536, 0.0767, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 10:22:51,486 INFO [train.py:968] (1/2) Epoch 6, batch 35450, giga_loss[loss=0.1967, simple_loss=0.276, pruned_loss=0.05866, over 28890.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3167, pruned_loss=0.09109, over 5688712.70 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3708, pruned_loss=0.1319, over 5710489.64 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3117, pruned_loss=0.08744, over 5681195.43 frames. ], batch size: 145, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:23:34,460 INFO [train.py:968] (1/2) Epoch 6, batch 35500, giga_loss[loss=0.2053, simple_loss=0.283, pruned_loss=0.06377, over 28964.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3142, pruned_loss=0.08989, over 5677884.09 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3713, pruned_loss=0.1322, over 5698390.09 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3089, pruned_loss=0.08602, over 5683231.84 frames. ], batch size: 213, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:24:07,325 INFO [optim.py:369] (1/2) 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 10:24:20,975 INFO [train.py:968] (1/2) Epoch 6, batch 35550, giga_loss[loss=0.2036, simple_loss=0.272, pruned_loss=0.06758, over 28309.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3111, pruned_loss=0.08865, over 5672918.99 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.372, pruned_loss=0.1327, over 5702576.86 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.305, pruned_loss=0.08419, over 5672976.16 frames. ], batch size: 71, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:25:04,810 INFO [train.py:968] (1/2) Epoch 6, batch 35600, giga_loss[loss=0.2782, simple_loss=0.3528, pruned_loss=0.1018, over 28917.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3155, pruned_loss=0.09165, over 5680455.95 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3726, pruned_loss=0.1331, over 5708388.39 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3082, pruned_loss=0.08641, over 5674439.09 frames. ], batch size: 164, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:25:26,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0507, 1.2486, 1.2606, 1.2000], device='cuda:1'), covar=tensor([0.1170, 0.1171, 0.1725, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0738, 0.0647, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 10:25:36,285 INFO [optim.py:369] (1/2) 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 10:25:50,332 INFO [train.py:968] (1/2) Epoch 6, batch 35650, giga_loss[loss=0.2849, simple_loss=0.3628, pruned_loss=0.1036, over 28465.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.33, pruned_loss=0.09969, over 5686396.10 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3732, pruned_loss=0.1334, over 5709804.41 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3231, pruned_loss=0.0948, over 5680161.89 frames. ], batch size: 60, lr: 5.22e-03, grad_scale: 8.0
+2023-03-03 10:26:35,292 INFO [train.py:968] (1/2) Epoch 6, batch 35700, giga_loss[loss=0.3289, simple_loss=0.3844, pruned_loss=0.1367, over 28348.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3446, pruned_loss=0.1082, over 5685375.97 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3736, pruned_loss=0.1335, over 5713130.38 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3379, pruned_loss=0.1036, over 5676871.61 frames. ], batch size: 71, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:27:05,035 INFO [optim.py:369] (1/2) 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 10:27:19,840 INFO [train.py:968] (1/2) Epoch 6, batch 35750, giga_loss[loss=0.3021, simple_loss=0.3738, pruned_loss=0.1152, over 28938.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3537, pruned_loss=0.1127, over 5689572.48 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3739, pruned_loss=0.1337, over 5714880.65 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3473, pruned_loss=0.1082, over 5680177.65 frames. ], batch size: 227, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:27:21,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-03 10:27:43,413 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:27:54,832 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:28:00,258 INFO [train.py:968] (1/2) Epoch 6, batch 35800, giga_loss[loss=0.2919, simple_loss=0.3618, pruned_loss=0.111, over 28447.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3597, pruned_loss=0.1147, over 5678947.28 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3747, pruned_loss=0.1342, over 5704545.14 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3539, pruned_loss=0.1104, over 5680856.25 frames. ], batch size: 71, lr: 5.22e-03, grad_scale: 4.0
+2023-03-03 10:28:07,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:28:20,599 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:28:33,030 INFO [optim.py:369] (1/2) 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 10:28:44,188 INFO [train.py:968] (1/2) Epoch 6, batch 35850, giga_loss[loss=0.3101, simple_loss=0.3745, pruned_loss=0.1228, over 28747.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3605, pruned_loss=0.1139, over 5671572.82 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3745, pruned_loss=0.134, over 5707738.77 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3555, pruned_loss=0.1101, over 5669196.39 frames. ], batch size: 119, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:29:29,915 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=263607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:29:32,382 INFO [train.py:968] (1/2) Epoch 6, batch 35900, libri_loss[loss=0.3778, simple_loss=0.4285, pruned_loss=0.1636, over 27928.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.362, pruned_loss=0.1142, over 5668783.59 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.375, pruned_loss=0.1344, over 5707960.20 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3575, pruned_loss=0.1106, over 5666508.35 frames. ], batch size: 116, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:29:55,242 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:30:14,304 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-03 10:30:15,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=263660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:30:15,825 INFO [train.py:968] (1/2) Epoch 6, batch 35950, libri_loss[loss=0.3894, simple_loss=0.4363, pruned_loss=0.1712, over 29234.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3652, pruned_loss=0.1162, over 5682851.02 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3757, pruned_loss=0.1347, over 5708746.76 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3607, pruned_loss=0.1126, over 5679445.08 frames. ], batch size: 97, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:30:17,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=263663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:30:42,829 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 36000, giga_loss[loss=0.2988, simple_loss=0.3685, pruned_loss=0.1145, over 28579.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3674, pruned_loss=0.1181, over 5682863.65 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3756, pruned_loss=0.1347, over 5711563.20 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3638, pruned_loss=0.1151, over 5677259.90 frames. ], batch size: 307, lr: 5.21e-03, grad_scale: 8.0
+2023-03-03 10:30:59,315 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 10:31:08,351 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 10:31:36,371 INFO [optim.py:369] (1/2) 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 10:31:48,796 INFO [train.py:968] (1/2) Epoch 6, batch 36050, giga_loss[loss=0.3371, simple_loss=0.3974, pruned_loss=0.1384, over 28925.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3703, pruned_loss=0.1187, over 5686685.68 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3756, pruned_loss=0.1347, over 5702598.37 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3674, pruned_loss=0.1162, over 5689381.32 frames. ], batch size: 186, lr: 5.21e-03, grad_scale: 8.0
+2023-03-03 10:32:29,213 INFO [train.py:968] (1/2) Epoch 6, batch 36100, giga_loss[loss=0.3019, simple_loss=0.3742, pruned_loss=0.1149, over 29016.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3725, pruned_loss=0.119, over 5686611.86 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3766, pruned_loss=0.1353, over 5703250.17 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3692, pruned_loss=0.1161, over 5688133.86 frames. ], batch size: 66, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:33:00,285 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 36150, giga_loss[loss=0.3019, simple_loss=0.374, pruned_loss=0.1149, over 28917.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3741, pruned_loss=0.1191, over 5695587.14 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3772, pruned_loss=0.1355, over 5709172.81 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3708, pruned_loss=0.1162, over 5691097.41 frames. ], batch size: 106, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:33:13,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:33:24,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:33:50,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:33:52,725 INFO [train.py:968] (1/2) Epoch 6, batch 36200, giga_loss[loss=0.3072, simple_loss=0.3885, pruned_loss=0.1129, over 28715.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3747, pruned_loss=0.1187, over 5692235.78 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.378, pruned_loss=0.1359, over 5709297.77 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3713, pruned_loss=0.1156, over 5688125.86 frames. ], batch size: 284, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:34:20,692 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 36250, giga_loss[loss=0.2671, simple_loss=0.356, pruned_loss=0.08905, over 28604.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.373, pruned_loss=0.1162, over 5695888.84 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3781, pruned_loss=0.1358, over 5703260.45 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3702, pruned_loss=0.1135, over 5697974.47 frames. ], batch size: 60, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:34:50,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=263982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:08,789 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:11,084 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 36300, libri_loss[loss=0.3893, simple_loss=0.4255, pruned_loss=0.1766, over 19322.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3712, pruned_loss=0.1145, over 5687051.92 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.379, pruned_loss=0.1363, over 5696967.50 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.368, pruned_loss=0.1115, over 5695092.80 frames. ], batch size: 187, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:35:14,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:21,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:23,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:35,511 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:35:46,244 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:48,840 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:48,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:35:53,781 INFO [train.py:968] (1/2) Epoch 6, batch 36350, giga_loss[loss=0.2885, simple_loss=0.3542, pruned_loss=0.1114, over 28999.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3725, pruned_loss=0.1165, over 5677357.85 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3792, pruned_loss=0.1363, over 5691512.55 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3696, pruned_loss=0.1136, over 5688979.32 frames. ], batch size: 136, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:36:14,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:36:38,075 INFO [train.py:968] (1/2) Epoch 6, batch 36400, giga_loss[loss=0.3477, simple_loss=0.3972, pruned_loss=0.1491, over 28799.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3752, pruned_loss=0.1208, over 5683222.86 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3791, pruned_loss=0.1361, over 5699208.90 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3727, pruned_loss=0.1182, over 5685057.16 frames. ], batch size: 112, lr: 5.21e-03, grad_scale: 8.0
+2023-03-03 10:36:52,349 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:36:55,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:12,442 INFO [optim.py:369] (1/2) 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 10:37:19,139 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:19,195 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:20,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:21,327 INFO [train.py:968] (1/2) Epoch 6, batch 36450, giga_loss[loss=0.3147, simple_loss=0.3727, pruned_loss=0.1283, over 28939.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3772, pruned_loss=0.1246, over 5677223.48 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3791, pruned_loss=0.1361, over 5690577.52 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3753, pruned_loss=0.1224, over 5686351.40 frames. ], batch size: 213, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:37:31,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3367, 1.4808, 1.3957, 1.3523], device='cuda:1'), covar=tensor([0.1400, 0.1718, 0.1733, 0.1646], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0739, 0.0644, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 10:37:44,849 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:37:58,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4867, 2.8842, 1.8869, 1.6575], device='cuda:1'), covar=tensor([0.1129, 0.0753, 0.1046, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.1498, 0.1293, 0.1289, 0.1402], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 10:38:00,796 INFO [train.py:968] (1/2) Epoch 6, batch 36500, giga_loss[loss=0.2772, simple_loss=0.3495, pruned_loss=0.1024, over 28956.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3766, pruned_loss=0.1252, over 5671748.58 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3804, pruned_loss=0.1368, over 5685014.31 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3737, pruned_loss=0.1223, over 5684241.87 frames. ], batch size: 106, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:38:12,726 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:38:31,362 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 36550, libri_loss[loss=0.3521, simple_loss=0.4111, pruned_loss=0.1465, over 29521.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3751, pruned_loss=0.1249, over 5692105.78 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3804, pruned_loss=0.1369, over 5693563.02 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3725, pruned_loss=0.1219, over 5694155.29 frames. ], batch size: 82, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:39:23,584 INFO [train.py:968] (1/2) Epoch 6, batch 36600, giga_loss[loss=0.2757, simple_loss=0.353, pruned_loss=0.09915, over 28837.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1235, over 5697133.13 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3807, pruned_loss=0.1371, over 5698537.82 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3707, pruned_loss=0.1207, over 5694345.75 frames. ], batch size: 145, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:39:56,459 INFO [optim.py:369] (1/2) 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 10:39:57,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6410, 4.5047, 4.2871, 1.8634], device='cuda:1'), covar=tensor([0.0448, 0.0500, 0.0609, 0.2038], device='cuda:1'), in_proj_covar=tensor([0.0896, 0.0836, 0.0770, 0.0602], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 10:39:59,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5302, 1.5863, 1.3913, 1.9731], device='cuda:1'), covar=tensor([0.1748, 0.1666, 0.1568, 0.1646], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.0880, 0.1016, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 10:40:00,820 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 36650, giga_loss[loss=0.3178, simple_loss=0.3935, pruned_loss=0.121, over 29000.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3725, pruned_loss=0.1223, over 5678060.72 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3812, pruned_loss=0.1373, over 5682175.36 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3698, pruned_loss=0.1193, over 5690578.54 frames. ], batch size: 174, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:40:56,304 INFO [train.py:968] (1/2) Epoch 6, batch 36700, giga_loss[loss=0.2615, simple_loss=0.3382, pruned_loss=0.09245, over 28624.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1187, over 5680664.25 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3808, pruned_loss=0.1371, over 5684184.29 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3662, pruned_loss=0.1164, over 5688788.88 frames. ], batch size: 92, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:41:29,708 INFO [optim.py:369] (1/2) 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 10:41:41,855 INFO [train.py:968] (1/2) Epoch 6, batch 36750, giga_loss[loss=0.2387, simple_loss=0.3122, pruned_loss=0.08259, over 28996.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1142, over 5690975.59 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3805, pruned_loss=0.1367, over 5687864.90 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3599, pruned_loss=0.1125, over 5694191.68 frames. ], batch size: 227, lr: 5.21e-03, grad_scale: 2.0
+2023-03-03 10:42:34,974 INFO [train.py:968] (1/2) Epoch 6, batch 36800, giga_loss[loss=0.2639, simple_loss=0.3119, pruned_loss=0.108, over 23409.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3561, pruned_loss=0.1117, over 5678085.95 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3812, pruned_loss=0.1372, over 5690886.02 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3539, pruned_loss=0.1094, over 5677941.23 frames. ], batch size: 705, lr: 5.21e-03, grad_scale: 4.0
+2023-03-03 10:43:14,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-03 10:43:17,356 INFO [optim.py:369] (1/2) 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 10:43:27,060 INFO [train.py:968] (1/2) Epoch 6, batch 36850, giga_loss[loss=0.2977, simple_loss=0.3586, pruned_loss=0.1184, over 27616.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3518, pruned_loss=0.1092, over 5667007.28 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3818, pruned_loss=0.1376, over 5683970.47 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3492, pruned_loss=0.1068, over 5672746.09 frames. ], batch size: 472, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:44:02,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264601.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:44:10,264 INFO [train.py:968] (1/2) Epoch 6, batch 36900, giga_loss[loss=0.2609, simple_loss=0.3393, pruned_loss=0.09124, over 29060.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3529, pruned_loss=0.1097, over 5672878.08 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3815, pruned_loss=0.1372, over 5685372.84 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3505, pruned_loss=0.1074, over 5675969.21 frames. ], batch size: 145, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:44:22,764 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:44:38,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-03 10:44:40,077 INFO [optim.py:369] (1/2) 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 10:44:48,682 INFO [train.py:968] (1/2) Epoch 6, batch 36950, giga_loss[loss=0.2896, simple_loss=0.3557, pruned_loss=0.1118, over 28964.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3535, pruned_loss=0.1095, over 5685977.53 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3825, pruned_loss=0.1377, over 5685317.38 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3495, pruned_loss=0.1061, over 5687613.09 frames. ], batch size: 227, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:45:30,649 INFO [train.py:968] (1/2) Epoch 6, batch 37000, giga_loss[loss=0.2497, simple_loss=0.3164, pruned_loss=0.09148, over 28594.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3525, pruned_loss=0.1095, over 5684190.10 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3826, pruned_loss=0.1377, over 5686174.54 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3487, pruned_loss=0.1063, over 5684606.82 frames. ], batch size: 60, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:45:45,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=264728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:45:52,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3886, 1.5774, 1.5375, 1.5285], device='cuda:1'), covar=tensor([0.0897, 0.0893, 0.1268, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0749, 0.0655, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 10:45:55,597 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264744.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:45:58,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264747.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:46:01,807 INFO [optim.py:369] (1/2) 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 10:46:10,580 INFO [train.py:968] (1/2) Epoch 6, batch 37050, giga_loss[loss=0.2465, simple_loss=0.3234, pruned_loss=0.08482, over 28953.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3504, pruned_loss=0.1084, over 5688953.83 frames. ], libri_tot_loss[loss=0.3291, simple_loss=0.383, pruned_loss=0.1376, over 5687261.90 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3462, pruned_loss=0.1052, over 5688324.10 frames. ], batch size: 136, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:46:24,017 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264776.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 10:46:51,214 INFO [train.py:968] (1/2) Epoch 6, batch 37100, giga_loss[loss=0.2573, simple_loss=0.3275, pruned_loss=0.09356, over 29072.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3469, pruned_loss=0.1062, over 5705631.90 frames. ], libri_tot_loss[loss=0.3293, simple_loss=0.3832, pruned_loss=0.1377, over 5692320.34 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3426, pruned_loss=0.1029, over 5700686.07 frames. ], batch size: 128, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:47:23,171 INFO [optim.py:369] (1/2) 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 10:47:29,966 INFO [train.py:968] (1/2) Epoch 6, batch 37150, giga_loss[loss=0.2496, simple_loss=0.3221, pruned_loss=0.08854, over 28696.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3446, pruned_loss=0.1049, over 5712657.42 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3838, pruned_loss=0.1378, over 5695282.40 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3395, pruned_loss=0.1011, over 5706378.82 frames. ], batch size: 242, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:47:38,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2491, 1.7428, 1.3484, 0.4495], device='cuda:1'), covar=tensor([0.2257, 0.1485, 0.2497, 0.3164], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1328, 0.1386, 0.1183], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 10:47:38,971 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=264871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:47:39,441 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:47:40,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=264874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:47:55,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=264893.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:48:05,083 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=264903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:48:10,654 INFO [train.py:968] (1/2) Epoch 6, batch 37200, giga_loss[loss=0.2845, simple_loss=0.3477, pruned_loss=0.1107, over 28581.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3433, pruned_loss=0.1047, over 5711066.00 frames. ], libri_tot_loss[loss=0.3302, simple_loss=0.3843, pruned_loss=0.138, over 5698482.46 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3383, pruned_loss=0.1011, over 5703547.48 frames. ], batch size: 307, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:48:43,676 INFO [optim.py:369] (1/2) 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 10:48:51,600 INFO [train.py:968] (1/2) Epoch 6, batch 37250, giga_loss[loss=0.2256, simple_loss=0.2959, pruned_loss=0.07762, over 28853.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3398, pruned_loss=0.1026, over 5703670.91 frames. ], libri_tot_loss[loss=0.3306, simple_loss=0.3847, pruned_loss=0.1383, over 5689929.94 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3351, pruned_loss=0.09933, over 5705589.96 frames. ], batch size: 112, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:49:09,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-03 10:49:23,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:49:30,036 INFO [train.py:968] (1/2) Epoch 6, batch 37300, giga_loss[loss=0.2387, simple_loss=0.3151, pruned_loss=0.08115, over 29051.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3384, pruned_loss=0.1019, over 5711793.05 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3854, pruned_loss=0.1384, over 5692948.94 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3332, pruned_loss=0.09841, over 5710759.90 frames. ], batch size: 136, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:50:02,385 INFO [optim.py:369] (1/2) 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 10:50:09,916 INFO [train.py:968] (1/2) Epoch 6, batch 37350, giga_loss[loss=0.2518, simple_loss=0.3234, pruned_loss=0.09012, over 28536.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3358, pruned_loss=0.1002, over 5722135.90 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3854, pruned_loss=0.1384, over 5696078.58 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3309, pruned_loss=0.09697, over 5718792.16 frames. ], batch size: 71, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:50:49,437 INFO [train.py:968] (1/2) Epoch 6, batch 37400, giga_loss[loss=0.254, simple_loss=0.3185, pruned_loss=0.09474, over 28788.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3354, pruned_loss=0.09987, over 5721015.10 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3863, pruned_loss=0.1388, over 5691177.13 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3296, pruned_loss=0.09614, over 5722826.34 frames. ], batch size: 99, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:51:00,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2406, 1.6601, 1.2355, 1.5479], device='cuda:1'), covar=tensor([0.0786, 0.0302, 0.0339, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0120, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0046, 0.0041, 0.0070], device='cuda:1')
+2023-03-03 10:51:18,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:51:20,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:51:23,375 INFO [optim.py:369] (1/2) 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 10:51:31,282 INFO [train.py:968] (1/2) Epoch 6, batch 37450, giga_loss[loss=0.2445, simple_loss=0.3102, pruned_loss=0.0894, over 28708.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3382, pruned_loss=0.1019, over 5722052.36 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3874, pruned_loss=0.1394, over 5696316.13 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3315, pruned_loss=0.09759, over 5719449.55 frames. ], batch size: 92, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:51:43,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4770, 1.8050, 1.7472, 1.4855], device='cuda:1'), covar=tensor([0.1435, 0.1727, 0.1142, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0719, 0.0815, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 10:51:44,773 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:52:03,329 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:52:14,023 INFO [train.py:968] (1/2) Epoch 6, batch 37500, giga_loss[loss=0.2573, simple_loss=0.3243, pruned_loss=0.09516, over 28857.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3434, pruned_loss=0.1053, over 5714075.92 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.388, pruned_loss=0.1398, over 5695283.25 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3362, pruned_loss=0.1004, over 5713893.30 frames. ], batch size: 119, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:52:45,965 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:52:59,451 INFO [train.py:968] (1/2) Epoch 6, batch 37550, giga_loss[loss=0.3691, simple_loss=0.414, pruned_loss=0.1621, over 28344.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3517, pruned_loss=0.1111, over 5686571.89 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3885, pruned_loss=0.1402, over 5681370.25 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3445, pruned_loss=0.1062, over 5697807.70 frames. ], batch size: 369, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:53:07,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265268.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:53:46,011 INFO [train.py:968] (1/2) Epoch 6, batch 37600, giga_loss[loss=0.3108, simple_loss=0.379, pruned_loss=0.1213, over 28765.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3591, pruned_loss=0.1163, over 5688157.14 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.3878, pruned_loss=0.1398, over 5685635.00 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.353, pruned_loss=0.1119, over 5693436.30 frames. ], batch size: 199, lr: 5.20e-03, grad_scale: 4.0
+2023-03-03 10:54:00,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2541, 1.1191, 4.5644, 3.3393], device='cuda:1'), covar=tensor([0.1647, 0.2537, 0.0314, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0579, 0.0536, 0.0764, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 10:54:25,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4688, 3.0360, 1.7818, 1.7190], device='cuda:1'), covar=tensor([0.1292, 0.0696, 0.1093, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.1497, 0.1312, 0.1325, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 10:54:30,942 INFO [optim.py:369] (1/2) 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 10:54:37,538 INFO [train.py:968] (1/2) Epoch 6, batch 37650, giga_loss[loss=0.3056, simple_loss=0.372, pruned_loss=0.1196, over 28688.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3637, pruned_loss=0.1183, over 5665176.84 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3881, pruned_loss=0.1399, over 5677225.86 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.358, pruned_loss=0.1141, over 5677335.80 frames. ], batch size: 92, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:54:58,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2349, 2.7610, 1.3275, 1.2706], device='cuda:1'), covar=tensor([0.0955, 0.0289, 0.0873, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0479, 0.0312, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 10:55:02,872 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:55:06,204 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 37700, giga_loss[loss=0.3245, simple_loss=0.3953, pruned_loss=0.1268, over 28714.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.368, pruned_loss=0.1195, over 5670404.74 frames. ], libri_tot_loss[loss=0.334, simple_loss=0.3882, pruned_loss=0.1399, over 5678439.06 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.363, pruned_loss=0.1159, over 5678820.42 frames. ], batch size: 262, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:55:22,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265411.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:55:25,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265414.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:55:33,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:55:53,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265443.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 10:56:00,984 INFO [optim.py:369] (1/2) 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 10:56:07,536 INFO [train.py:968] (1/2) Epoch 6, batch 37750, giga_loss[loss=0.3522, simple_loss=0.414, pruned_loss=0.1452, over 29083.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.1239, over 5672045.98 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3885, pruned_loss=0.1399, over 5683893.75 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.37, pruned_loss=0.1207, over 5673804.81 frames. ], batch size: 155, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:56:48,666 INFO [train.py:968] (1/2) Epoch 6, batch 37800, giga_loss[loss=0.2413, simple_loss=0.3216, pruned_loss=0.08051, over 28920.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3732, pruned_loss=0.1232, over 5669163.27 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3884, pruned_loss=0.1401, over 5682848.38 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3694, pruned_loss=0.1202, over 5671149.67 frames. ], batch size: 227, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:57:21,960 INFO [optim.py:369] (1/2) 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 10:57:28,067 INFO [train.py:968] (1/2) Epoch 6, batch 37850, giga_loss[loss=0.2893, simple_loss=0.3565, pruned_loss=0.111, over 28950.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3688, pruned_loss=0.1194, over 5674748.01 frames. ], libri_tot_loss[loss=0.3344, simple_loss=0.3884, pruned_loss=0.1402, over 5680184.38 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3652, pruned_loss=0.1163, over 5679280.20 frames. ], batch size: 106, lr: 5.20e-03, grad_scale: 2.0
+2023-03-03 10:57:37,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=265572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:57:51,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-03 10:58:12,965 INFO [train.py:968] (1/2) Epoch 6, batch 37900, giga_loss[loss=0.2399, simple_loss=0.3184, pruned_loss=0.08074, over 28625.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3676, pruned_loss=0.1179, over 5671694.17 frames. ], libri_tot_loss[loss=0.3351, simple_loss=0.3889, pruned_loss=0.1406, over 5674090.58 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3641, pruned_loss=0.1149, over 5680185.52 frames. ], batch size: 60, lr: 5.19e-03, grad_scale: 2.0
+2023-03-03 10:58:50,061 INFO [optim.py:369] (1/2) 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 10:58:55,631 INFO [train.py:968] (1/2) Epoch 6, batch 37950, giga_loss[loss=0.2601, simple_loss=0.3391, pruned_loss=0.09053, over 29026.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3669, pruned_loss=0.117, over 5682347.74 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3888, pruned_loss=0.1404, over 5683646.91 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3633, pruned_loss=0.1138, over 5680626.96 frames. ], batch size: 213, lr: 5.19e-03, grad_scale: 2.0
+2023-03-03 10:59:03,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:59:37,556 INFO [train.py:968] (1/2) Epoch 6, batch 38000, giga_loss[loss=0.3194, simple_loss=0.3825, pruned_loss=0.1281, over 28618.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3695, pruned_loss=0.1186, over 5686110.98 frames. ], libri_tot_loss[loss=0.3352, simple_loss=0.389, pruned_loss=0.1407, over 5687488.92 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3659, pruned_loss=0.1152, over 5681346.36 frames. ], batch size: 336, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 10:59:42,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=265715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 10:59:44,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=265718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:00:08,377 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=265747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:00:13,369 INFO [optim.py:369] (1/2) 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 11:00:16,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6595, 1.8893, 1.9450, 1.6657], device='cuda:1'), covar=tensor([0.1513, 0.1801, 0.1124, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0722, 0.0816, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 11:00:19,762 INFO [train.py:968] (1/2) Epoch 6, batch 38050, giga_loss[loss=0.2823, simple_loss=0.3535, pruned_loss=0.1055, over 28558.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3724, pruned_loss=0.1206, over 5690216.65 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3892, pruned_loss=0.1407, over 5690192.65 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3688, pruned_loss=0.1173, over 5683923.04 frames. ], batch size: 71, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:01:05,639 INFO [train.py:968] (1/2) Epoch 6, batch 38100, giga_loss[loss=0.3041, simple_loss=0.3729, pruned_loss=0.1177, over 28547.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3736, pruned_loss=0.1217, over 5697097.34 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3888, pruned_loss=0.1404, over 5693857.57 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3709, pruned_loss=0.119, over 5688948.68 frames. ], batch size: 85, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:01:42,519 INFO [optim.py:369] (1/2) 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 11:01:49,799 INFO [train.py:968] (1/2) Epoch 6, batch 38150, giga_loss[loss=0.3483, simple_loss=0.3982, pruned_loss=0.1492, over 27586.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3745, pruned_loss=0.1227, over 5691746.70 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3895, pruned_loss=0.1408, over 5695268.82 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3714, pruned_loss=0.1198, over 5684048.00 frames. ], batch size: 472, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:02:16,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4495, 1.4255, 5.0907, 3.5079], device='cuda:1'), covar=tensor([0.1594, 0.2208, 0.0273, 0.0613], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0533, 0.0759, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:02:20,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8349, 2.2648, 2.2368, 1.8431], device='cuda:1'), covar=tensor([0.1576, 0.1697, 0.1125, 0.1276], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0721, 0.0815, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 11:02:28,808 INFO [train.py:968] (1/2) Epoch 6, batch 38200, giga_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.115, over 28793.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3745, pruned_loss=0.1228, over 5693833.02 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3896, pruned_loss=0.1408, over 5690204.76 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3716, pruned_loss=0.1202, over 5691612.39 frames. ], batch size: 186, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:02:39,882 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=265923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:02:56,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6199, 1.8563, 1.5507, 2.0123], device='cuda:1'), covar=tensor([0.2109, 0.2065, 0.2132, 0.2102], device='cuda:1'), in_proj_covar=tensor([0.1151, 0.0883, 0.1022, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:03:05,517 INFO [optim.py:369] (1/2) 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 11:03:11,595 INFO [train.py:968] (1/2) Epoch 6, batch 38250, giga_loss[loss=0.3189, simple_loss=0.3877, pruned_loss=0.1251, over 28614.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3741, pruned_loss=0.1216, over 5699888.43 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3899, pruned_loss=0.1409, over 5694440.90 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3713, pruned_loss=0.1191, over 5694572.94 frames. ], batch size: 307, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:03:33,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4288, 1.7151, 1.3649, 1.9387], device='cuda:1'), covar=tensor([0.2140, 0.2015, 0.2107, 0.1967], device='cuda:1'), in_proj_covar=tensor([0.1145, 0.0879, 0.1018, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 11:03:34,694 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-03 11:03:47,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:03:51,061 INFO [train.py:968] (1/2) Epoch 6, batch 38300, giga_loss[loss=0.3316, simple_loss=0.3968, pruned_loss=0.1331, over 28279.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3744, pruned_loss=0.1204, over 5697791.83 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3905, pruned_loss=0.1412, over 5687639.60 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3711, pruned_loss=0.1175, over 5699914.70 frames. ], batch size: 368, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:04:24,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:04:29,368 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 38350, giga_loss[loss=0.3028, simple_loss=0.3708, pruned_loss=0.1174, over 28530.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3749, pruned_loss=0.1199, over 5702122.87 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3904, pruned_loss=0.141, over 5692046.10 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3721, pruned_loss=0.1173, over 5699990.77 frames. ], batch size: 71, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:04:59,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-03 11:05:01,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3792, 5.1937, 4.9394, 2.5529], device='cuda:1'), covar=tensor([0.0302, 0.0384, 0.0505, 0.1668], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0834, 0.0756, 0.0594], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:05:16,185 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 38400, libri_loss[loss=0.4489, simple_loss=0.4699, pruned_loss=0.214, over 19538.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3723, pruned_loss=0.118, over 5694862.27 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3905, pruned_loss=0.1412, over 5685374.10 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3698, pruned_loss=0.1155, over 5700414.91 frames. ], batch size: 186, lr: 5.19e-03, grad_scale: 8.0
+2023-03-03 11:05:35,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 11:05:52,861 INFO [optim.py:369] (1/2) 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 11:05:58,811 INFO [train.py:968] (1/2) Epoch 6, batch 38450, giga_loss[loss=0.2883, simple_loss=0.3561, pruned_loss=0.1103, over 28014.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.37, pruned_loss=0.1168, over 5702610.42 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3908, pruned_loss=0.1415, over 5686972.60 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3674, pruned_loss=0.1143, over 5705628.41 frames. ], batch size: 412, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:06:23,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:06:25,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:06:42,008 INFO [train.py:968] (1/2) Epoch 6, batch 38500, giga_loss[loss=0.3028, simple_loss=0.3676, pruned_loss=0.119, over 28546.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3673, pruned_loss=0.1151, over 5694235.26 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3912, pruned_loss=0.1418, over 5670283.81 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3646, pruned_loss=0.1123, over 5712533.55 frames. ], batch size: 78, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:06:48,812 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:06:59,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1197, 1.3747, 1.1567, 0.9398], device='cuda:1'), covar=tensor([0.1921, 0.1857, 0.1792, 0.1712], device='cuda:1'), in_proj_covar=tensor([0.1159, 0.0889, 0.1027, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:07:09,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-03 11:07:15,176 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 38550, giga_loss[loss=0.3072, simple_loss=0.3815, pruned_loss=0.1165, over 28981.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3671, pruned_loss=0.1157, over 5697889.14 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3908, pruned_loss=0.1416, over 5676977.38 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3646, pruned_loss=0.1129, over 5707298.31 frames. ], batch size: 145, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:07:24,320 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:07:47,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 11:07:50,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:07:52,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8494, 0.9878, 3.4939, 2.9423], device='cuda:1'), covar=tensor([0.1524, 0.2130, 0.0383, 0.0732], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0529, 0.0755, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:07:59,875 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:07:59,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5894, 3.6814, 1.7249, 1.5100], device='cuda:1'), covar=tensor([0.0865, 0.0224, 0.0765, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0469, 0.0308, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:08:00,442 INFO [train.py:968] (1/2) Epoch 6, batch 38600, giga_loss[loss=0.2836, simple_loss=0.3593, pruned_loss=0.104, over 28686.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3674, pruned_loss=0.116, over 5700093.99 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3906, pruned_loss=0.1416, over 5672338.59 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3649, pruned_loss=0.1131, over 5712280.05 frames. ], batch size: 60, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:08:32,241 INFO [optim.py:369] (1/2) 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 11:08:38,015 INFO [train.py:968] (1/2) Epoch 6, batch 38650, giga_loss[loss=0.2728, simple_loss=0.3544, pruned_loss=0.09562, over 28875.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3681, pruned_loss=0.1161, over 5701058.04 frames. ], libri_tot_loss[loss=0.3373, simple_loss=0.391, pruned_loss=0.1418, over 5669382.80 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3652, pruned_loss=0.113, over 5713829.19 frames. ], batch size: 174, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:08:55,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:09:05,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5518, 3.3500, 1.7406, 1.6273], device='cuda:1'), covar=tensor([0.0864, 0.0252, 0.0793, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0467, 0.0308, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0025, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:09:24,362 INFO [train.py:968] (1/2) Epoch 6, batch 38700, libri_loss[loss=0.3377, simple_loss=0.3951, pruned_loss=0.1401, over 25889.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3669, pruned_loss=0.1144, over 5700413.23 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3913, pruned_loss=0.1418, over 5672317.01 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3637, pruned_loss=0.1111, over 5708970.29 frames. ], batch size: 136, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:09:47,555 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:09:49,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:09:57,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6231, 4.3553, 1.7465, 1.6283], device='cuda:1'), covar=tensor([0.0893, 0.0186, 0.0834, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0468, 0.0309, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:09:57,471 INFO [optim.py:369] (1/2) 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 11:10:02,178 INFO [train.py:968] (1/2) Epoch 6, batch 38750, giga_loss[loss=0.2538, simple_loss=0.3378, pruned_loss=0.08489, over 28568.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3662, pruned_loss=0.1139, over 5708992.72 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3913, pruned_loss=0.142, over 5673526.45 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3631, pruned_loss=0.1107, over 5715352.17 frames. ], batch size: 60, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:10:13,607 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:10:21,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:10:44,041 INFO [train.py:968] (1/2) Epoch 6, batch 38800, giga_loss[loss=0.2936, simple_loss=0.3634, pruned_loss=0.1119, over 28565.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3642, pruned_loss=0.1132, over 5698450.05 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3911, pruned_loss=0.142, over 5667743.87 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3615, pruned_loss=0.1102, over 5708697.60 frames. ], batch size: 336, lr: 5.19e-03, grad_scale: 8.0
+2023-03-03 11:10:54,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:10:57,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:11:14,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6797, 4.4330, 1.8568, 1.5098], device='cuda:1'), covar=tensor([0.0928, 0.0210, 0.0863, 0.1457], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0471, 0.0312, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:11:17,611 INFO [optim.py:369] (1/2) 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 11:11:19,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1712, 0.9523, 4.1992, 3.3250], device='cuda:1'), covar=tensor([0.1390, 0.2188, 0.0303, 0.0682], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0527, 0.0749, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:11:20,338 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:11:22,918 INFO [train.py:968] (1/2) Epoch 6, batch 38850, giga_loss[loss=0.2337, simple_loss=0.3006, pruned_loss=0.0834, over 23367.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3614, pruned_loss=0.1121, over 5698656.06 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.391, pruned_loss=0.1419, over 5675557.74 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3583, pruned_loss=0.1086, over 5701044.74 frames. ], batch size: 710, lr: 5.19e-03, grad_scale: 4.0
+2023-03-03 11:11:44,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-03 11:11:57,971 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:02,833 INFO [train.py:968] (1/2) Epoch 6, batch 38900, giga_loss[loss=0.3936, simple_loss=0.4263, pruned_loss=0.1805, over 26759.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3597, pruned_loss=0.1116, over 5706895.61 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3909, pruned_loss=0.1419, over 5681192.32 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3564, pruned_loss=0.108, over 5704421.17 frames. ], batch size: 555, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:12:08,429 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266619.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 11:12:14,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:15,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:24,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:30,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3711, 1.9549, 1.4657, 0.5888], device='cuda:1'), covar=tensor([0.2448, 0.1435, 0.2318, 0.2891], device='cuda:1'), in_proj_covar=tensor([0.1428, 0.1327, 0.1396, 0.1176], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 11:12:36,679 INFO [optim.py:369] (1/2) 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 11:12:39,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:12:41,798 INFO [train.py:968] (1/2) Epoch 6, batch 38950, giga_loss[loss=0.2696, simple_loss=0.3403, pruned_loss=0.09943, over 28819.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3596, pruned_loss=0.1119, over 5702550.76 frames. ], libri_tot_loss[loss=0.3376, simple_loss=0.3912, pruned_loss=0.142, over 5675247.34 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3562, pruned_loss=0.1084, over 5706467.39 frames. ], batch size: 199, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:12:51,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-03 11:13:03,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:13:13,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 11:13:21,567 INFO [train.py:968] (1/2) Epoch 6, batch 39000, giga_loss[loss=0.2715, simple_loss=0.3424, pruned_loss=0.1003, over 28716.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5693666.00 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.3909, pruned_loss=0.1417, over 5673668.66 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3555, pruned_loss=0.1087, over 5699349.04 frames. ], batch size: 284, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:13:21,567 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 11:13:30,554 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 11:13:52,283 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:14:06,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3704, 3.4018, 1.3026, 1.4185], device='cuda:1'), covar=tensor([0.1082, 0.0397, 0.1103, 0.1540], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0475, 0.0311, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:14:07,290 INFO [optim.py:369] (1/2) 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 11:14:11,450 INFO [train.py:968] (1/2) Epoch 6, batch 39050, giga_loss[loss=0.3063, simple_loss=0.3716, pruned_loss=0.1205, over 28286.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3565, pruned_loss=0.111, over 5701803.75 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3911, pruned_loss=0.1419, over 5677099.08 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3528, pruned_loss=0.1075, over 5703541.21 frames. ], batch size: 368, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:14:29,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:14:31,261 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 39100, giga_loss[loss=0.3087, simple_loss=0.3685, pruned_loss=0.1244, over 28613.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.355, pruned_loss=0.1106, over 5708514.13 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3906, pruned_loss=0.1414, over 5681296.08 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3508, pruned_loss=0.1068, over 5707218.73 frames. ], batch size: 336, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:14:54,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:04,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=266828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:05,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=266831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:20,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8529, 2.3312, 1.8358, 2.0586], device='cuda:1'), covar=tensor([0.0516, 0.0225, 0.0228, 0.0548], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0121, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0070], device='cuda:1')
+2023-03-03 11:15:27,623 INFO [optim.py:369] (1/2) 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 11:15:28,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=266857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:31,122 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=266860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:15:31,625 INFO [train.py:968] (1/2) Epoch 6, batch 39150, giga_loss[loss=0.2484, simple_loss=0.3222, pruned_loss=0.08737, over 28999.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3527, pruned_loss=0.1098, over 5692475.12 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.3909, pruned_loss=0.1416, over 5669137.88 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3487, pruned_loss=0.1062, over 5703274.96 frames. ], batch size: 164, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:16:10,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-03 11:16:16,544 INFO [train.py:968] (1/2) Epoch 6, batch 39200, giga_loss[loss=0.2754, simple_loss=0.3537, pruned_loss=0.09857, over 28991.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3505, pruned_loss=0.1079, over 5700533.08 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3906, pruned_loss=0.1414, over 5673709.73 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3468, pruned_loss=0.1046, over 5705451.63 frames. ], batch size: 164, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:16:44,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5403, 5.3038, 5.0666, 2.4221], device='cuda:1'), covar=tensor([0.0309, 0.0449, 0.0573, 0.1799], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0836, 0.0764, 0.0595], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:16:48,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2302, 1.1964, 1.0625, 0.9633], device='cuda:1'), covar=tensor([0.0598, 0.0440, 0.0965, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0441, 0.0500, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:16:55,575 INFO [optim.py:369] (1/2) 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 11:16:59,063 INFO [train.py:968] (1/2) Epoch 6, batch 39250, giga_loss[loss=0.2893, simple_loss=0.364, pruned_loss=0.1074, over 28569.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3531, pruned_loss=0.1089, over 5701216.68 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3907, pruned_loss=0.1412, over 5679616.59 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3492, pruned_loss=0.1056, over 5700398.50 frames. ], batch size: 336, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:17:20,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:17:25,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1427, 1.3502, 4.0297, 3.2302], device='cuda:1'), covar=tensor([0.1611, 0.2274, 0.0316, 0.0646], device='cuda:1'), in_proj_covar=tensor([0.0574, 0.0526, 0.0751, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:17:27,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-03 11:17:28,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=266994.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 11:17:45,964 INFO [train.py:968] (1/2) Epoch 6, batch 39300, giga_loss[loss=0.2945, simple_loss=0.3526, pruned_loss=0.1182, over 23786.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3564, pruned_loss=0.1106, over 5692178.46 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3907, pruned_loss=0.1412, over 5682717.03 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3528, pruned_loss=0.1076, over 5689001.82 frames. ], batch size: 705, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:18:13,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1086, 1.2644, 3.6885, 3.1084], device='cuda:1'), covar=tensor([0.2154, 0.2857, 0.0704, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0572, 0.0525, 0.0751, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:18:21,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-03 11:18:25,190 INFO [optim.py:369] (1/2) 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 11:18:27,885 INFO [train.py:968] (1/2) Epoch 6, batch 39350, giga_loss[loss=0.3072, simple_loss=0.388, pruned_loss=0.1132, over 28804.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5689816.58 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3905, pruned_loss=0.1411, over 5677719.73 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3548, pruned_loss=0.1081, over 5691901.51 frames. ], batch size: 285, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:18:37,763 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:02,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:06,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-03 11:19:13,422 INFO [train.py:968] (1/2) Epoch 6, batch 39400, giga_loss[loss=0.2712, simple_loss=0.3461, pruned_loss=0.09816, over 28852.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3591, pruned_loss=0.1108, over 5688498.54 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3907, pruned_loss=0.1414, over 5677300.13 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3559, pruned_loss=0.108, over 5690532.05 frames. ], batch size: 199, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:19:16,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:25,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:26,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:19:37,245 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267137.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 11:19:39,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267140.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 11:19:41,375 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-03 11:19:51,545 INFO [optim.py:369] (1/2) 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 11:19:51,744 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 39450, giga_loss[loss=0.2569, simple_loss=0.3294, pruned_loss=0.09223, over 28604.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3583, pruned_loss=0.1103, over 5700618.83 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3903, pruned_loss=0.1411, over 5684452.74 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3549, pruned_loss=0.1071, over 5696228.52 frames. ], batch size: 71, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:19:59,474 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267169.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 11:20:31,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1259, 1.7277, 1.4514, 1.5916], device='cuda:1'), covar=tensor([0.0480, 0.0499, 0.0800, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0439, 0.0494, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:20:34,596 INFO [train.py:968] (1/2) Epoch 6, batch 39500, giga_loss[loss=0.286, simple_loss=0.3591, pruned_loss=0.1065, over 28553.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3589, pruned_loss=0.111, over 5703926.92 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3905, pruned_loss=0.1411, over 5687815.85 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3553, pruned_loss=0.1078, over 5697616.04 frames. ], batch size: 307, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:20:41,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2895, 0.9987, 0.9447, 1.4299], device='cuda:1'), covar=tensor([0.0662, 0.0295, 0.0315, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0121, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0070], device='cuda:1')
+2023-03-03 11:20:50,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:21:10,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-03 11:21:12,778 INFO [optim.py:369] (1/2) 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 11:21:13,101 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:21:14,867 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 39550, giga_loss[loss=0.3177, simple_loss=0.3863, pruned_loss=0.1246, over 28898.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3596, pruned_loss=0.1113, over 5709960.23 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3904, pruned_loss=0.1409, over 5685099.28 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.356, pruned_loss=0.1082, over 5708385.50 frames. ], batch size: 199, lr: 5.18e-03, grad_scale: 2.0
+2023-03-03 11:21:39,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:21:59,279 INFO [train.py:968] (1/2) Epoch 6, batch 39600, giga_loss[loss=0.3204, simple_loss=0.3841, pruned_loss=0.1283, over 28711.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3618, pruned_loss=0.1126, over 5708371.69 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3912, pruned_loss=0.1414, over 5682722.08 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3578, pruned_loss=0.1092, over 5709529.73 frames. ], batch size: 119, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:22:01,056 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:22:12,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7992, 1.4517, 1.2973, 1.3400], device='cuda:1'), covar=tensor([0.0601, 0.0659, 0.0957, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0445, 0.0502, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:22:37,882 INFO [optim.py:369] (1/2) 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 11:22:41,424 INFO [train.py:968] (1/2) Epoch 6, batch 39650, giga_loss[loss=0.3002, simple_loss=0.3752, pruned_loss=0.1126, over 28711.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3655, pruned_loss=0.1147, over 5697978.90 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3912, pruned_loss=0.1414, over 5678984.82 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3617, pruned_loss=0.1114, over 5702412.21 frames. ], batch size: 242, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:22:53,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:22:55,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:23:18,245 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 39700, giga_loss[loss=0.2958, simple_loss=0.3606, pruned_loss=0.1155, over 28411.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3676, pruned_loss=0.116, over 5704066.91 frames. ], libri_tot_loss[loss=0.3374, simple_loss=0.3916, pruned_loss=0.1416, over 5679236.19 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3637, pruned_loss=0.1126, over 5708015.90 frames. ], batch size: 71, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:23:43,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6139, 2.3023, 1.5376, 0.7207], device='cuda:1'), covar=tensor([0.4313, 0.1998, 0.2480, 0.4164], device='cuda:1'), in_proj_covar=tensor([0.1421, 0.1320, 0.1386, 0.1175], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 11:23:49,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:23:57,304 INFO [optim.py:369] (1/2) 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 11:24:01,373 INFO [train.py:968] (1/2) Epoch 6, batch 39750, giga_loss[loss=0.3063, simple_loss=0.3633, pruned_loss=0.1246, over 23657.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3675, pruned_loss=0.1155, over 5702806.23 frames. ], libri_tot_loss[loss=0.3382, simple_loss=0.3923, pruned_loss=0.142, over 5682730.59 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3633, pruned_loss=0.112, over 5703247.30 frames. ], batch size: 705, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:24:11,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:24:13,590 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:24:27,420 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:24:43,860 INFO [train.py:968] (1/2) Epoch 6, batch 39800, giga_loss[loss=0.3118, simple_loss=0.3828, pruned_loss=0.1204, over 28903.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3694, pruned_loss=0.1163, over 5697112.43 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3927, pruned_loss=0.1421, over 5677035.34 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3653, pruned_loss=0.113, over 5703205.98 frames. ], batch size: 199, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:24:54,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4664, 2.2245, 1.7190, 0.5389], device='cuda:1'), covar=tensor([0.2054, 0.0983, 0.1474, 0.2548], device='cuda:1'), in_proj_covar=tensor([0.1445, 0.1346, 0.1404, 0.1191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 11:25:10,389 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 11:25:10,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-03 11:25:21,471 INFO [optim.py:369] (1/2) 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 11:25:23,965 INFO [train.py:968] (1/2) Epoch 6, batch 39850, giga_loss[loss=0.2688, simple_loss=0.34, pruned_loss=0.09881, over 28272.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3686, pruned_loss=0.1158, over 5695365.51 frames. ], libri_tot_loss[loss=0.3386, simple_loss=0.3929, pruned_loss=0.1422, over 5669356.43 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3651, pruned_loss=0.1129, over 5706975.01 frames. ], batch size: 77, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:25:47,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:25:49,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:00,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1520, 1.1520, 4.2073, 3.2834], device='cuda:1'), covar=tensor([0.1573, 0.2405, 0.0301, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0538, 0.0773, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:26:03,444 INFO [train.py:968] (1/2) Epoch 6, batch 39900, giga_loss[loss=0.2864, simple_loss=0.3563, pruned_loss=0.1083, over 29078.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3665, pruned_loss=0.1149, over 5702095.98 frames. ], libri_tot_loss[loss=0.3387, simple_loss=0.393, pruned_loss=0.1422, over 5670770.77 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3634, pruned_loss=0.1124, over 5710252.20 frames. ], batch size: 128, lr: 5.18e-03, grad_scale: 4.0
+2023-03-03 11:26:06,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:09,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:12,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:26:31,006 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:26:43,123 INFO [train.py:968] (1/2) Epoch 6, batch 39950, giga_loss[loss=0.2271, simple_loss=0.3038, pruned_loss=0.07515, over 28785.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3636, pruned_loss=0.1133, over 5708070.55 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.3931, pruned_loss=0.142, over 5677168.42 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3599, pruned_loss=0.1103, over 5710468.47 frames. ], batch size: 99, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:26:44,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2296, 3.0023, 1.4253, 1.3080], device='cuda:1'), covar=tensor([0.0866, 0.0310, 0.0851, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0331, 0.0482, 0.0311, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:27:03,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:27:20,969 INFO [train.py:968] (1/2) Epoch 6, batch 40000, giga_loss[loss=0.2454, simple_loss=0.3186, pruned_loss=0.08608, over 28425.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3608, pruned_loss=0.1119, over 5710087.81 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3935, pruned_loss=0.1423, over 5679047.01 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3569, pruned_loss=0.1086, over 5711109.93 frames. ], batch size: 71, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:27:37,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-03 11:27:41,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2278, 2.6038, 1.1973, 1.3116], device='cuda:1'), covar=tensor([0.0833, 0.0321, 0.0884, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0483, 0.0312, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:27:56,861 INFO [optim.py:369] (1/2) 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 11:27:59,296 INFO [train.py:968] (1/2) Epoch 6, batch 40050, giga_loss[loss=0.2647, simple_loss=0.3517, pruned_loss=0.08881, over 28273.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3618, pruned_loss=0.111, over 5716252.76 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3934, pruned_loss=0.1422, over 5682413.18 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.358, pruned_loss=0.1079, over 5714999.31 frames. ], batch size: 368, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:28:04,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6836, 2.4192, 1.5303, 1.3752], device='cuda:1'), covar=tensor([0.2093, 0.0976, 0.1292, 0.1617], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1327, 0.1316, 0.1392], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 11:28:28,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2841, 1.7153, 1.3296, 1.5272], device='cuda:1'), covar=tensor([0.0737, 0.0312, 0.0332, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0121, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0046, 0.0042, 0.0070], device='cuda:1')
+2023-03-03 11:28:41,931 INFO [train.py:968] (1/2) Epoch 6, batch 40100, giga_loss[loss=0.2748, simple_loss=0.3373, pruned_loss=0.1061, over 28602.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3626, pruned_loss=0.1105, over 5709522.71 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3929, pruned_loss=0.1419, over 5685735.10 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3596, pruned_loss=0.1078, over 5706178.62 frames. ], batch size: 92, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:28:58,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:00,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:13,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:18,740 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 40150, giga_loss[loss=0.2815, simple_loss=0.3518, pruned_loss=0.1056, over 29032.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3629, pruned_loss=0.1117, over 5703261.98 frames. ], libri_tot_loss[loss=0.3385, simple_loss=0.393, pruned_loss=0.142, over 5678015.80 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.36, pruned_loss=0.1089, over 5708583.95 frames. ], batch size: 136, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:29:22,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=267863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:25,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=267866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:42,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=267887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:29:51,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4671, 3.3351, 1.5231, 1.4748], device='cuda:1'), covar=tensor([0.0809, 0.0341, 0.0855, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0488, 0.0314, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:30:02,060 INFO [train.py:968] (1/2) Epoch 6, batch 40200, giga_loss[loss=0.2848, simple_loss=0.3533, pruned_loss=0.1082, over 28629.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3636, pruned_loss=0.114, over 5701648.00 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3935, pruned_loss=0.1424, over 5677773.02 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3602, pruned_loss=0.111, over 5706237.62 frames. ], batch size: 262, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:30:04,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-03 11:30:40,800 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 40250, giga_loss[loss=0.3042, simple_loss=0.3678, pruned_loss=0.1203, over 27979.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3618, pruned_loss=0.1143, over 5699214.29 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.394, pruned_loss=0.1427, over 5679029.15 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3586, pruned_loss=0.1114, over 5701658.07 frames. ], batch size: 412, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:31:10,695 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=267993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:12,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=267996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:23,603 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:24,716 INFO [train.py:968] (1/2) Epoch 6, batch 40300, giga_loss[loss=0.2445, simple_loss=0.3078, pruned_loss=0.09059, over 28420.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3591, pruned_loss=0.1134, over 5693406.13 frames. ], libri_tot_loss[loss=0.3404, simple_loss=0.3945, pruned_loss=0.1431, over 5662381.35 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3558, pruned_loss=0.1105, over 5710207.19 frames. ], batch size: 71, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:31:26,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:38,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:31:50,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:32:05,393 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 40350, giga_loss[loss=0.3221, simple_loss=0.3845, pruned_loss=0.1298, over 28717.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3584, pruned_loss=0.1134, over 5706640.62 frames. ], libri_tot_loss[loss=0.3406, simple_loss=0.3946, pruned_loss=0.1432, over 5666954.36 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.355, pruned_loss=0.1105, over 5716504.75 frames. ], batch size: 262, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:32:47,099 INFO [train.py:968] (1/2) Epoch 6, batch 40400, giga_loss[loss=0.2859, simple_loss=0.3615, pruned_loss=0.1052, over 28726.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3544, pruned_loss=0.1108, over 5712076.54 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3941, pruned_loss=0.1429, over 5672809.58 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3513, pruned_loss=0.1081, over 5715687.99 frames. ], batch size: 262, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:33:24,703 INFO [optim.py:369] (1/2) 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 11:33:26,583 INFO [train.py:968] (1/2) Epoch 6, batch 40450, giga_loss[loss=0.2309, simple_loss=0.3008, pruned_loss=0.08047, over 29077.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3506, pruned_loss=0.109, over 5715124.06 frames. ], libri_tot_loss[loss=0.3397, simple_loss=0.3941, pruned_loss=0.1427, over 5675297.45 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3469, pruned_loss=0.1061, over 5716885.80 frames. ], batch size: 128, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:34:01,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 11:34:09,388 INFO [train.py:968] (1/2) Epoch 6, batch 40500, giga_loss[loss=0.2777, simple_loss=0.3512, pruned_loss=0.1021, over 29086.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3471, pruned_loss=0.107, over 5716205.09 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3941, pruned_loss=0.1428, over 5679079.71 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3436, pruned_loss=0.1041, over 5714552.07 frames. ], batch size: 155, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:34:12,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3851, 3.0207, 1.5070, 1.4462], device='cuda:1'), covar=tensor([0.0844, 0.0362, 0.0853, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0483, 0.0312, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 11:34:34,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9873, 1.0458, 3.4397, 3.0573], device='cuda:1'), covar=tensor([0.1601, 0.2464, 0.0411, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0534, 0.0770, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:34:45,777 INFO [optim.py:369] (1/2) 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 11:34:47,244 INFO [train.py:968] (1/2) Epoch 6, batch 40550, giga_loss[loss=0.2754, simple_loss=0.3518, pruned_loss=0.09951, over 28861.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3495, pruned_loss=0.1082, over 5711934.13 frames. ], libri_tot_loss[loss=0.3392, simple_loss=0.3936, pruned_loss=0.1424, over 5683868.66 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3454, pruned_loss=0.105, over 5707559.37 frames. ], batch size: 186, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:34:48,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=268262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:35:02,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-03 11:35:22,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5428, 1.4727, 1.5571, 1.3887], device='cuda:1'), covar=tensor([0.1051, 0.1824, 0.1595, 0.1624], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0736, 0.0656, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 11:35:29,777 INFO [train.py:968] (1/2) Epoch 6, batch 40600, giga_loss[loss=0.3031, simple_loss=0.3736, pruned_loss=0.1163, over 28885.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3533, pruned_loss=0.1101, over 5701168.81 frames. ], libri_tot_loss[loss=0.3399, simple_loss=0.3941, pruned_loss=0.1428, over 5671158.09 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3485, pruned_loss=0.1062, over 5710380.10 frames. ], batch size: 199, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:36:05,287 INFO [optim.py:369] (1/2) 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 11:36:06,991 INFO [train.py:968] (1/2) Epoch 6, batch 40650, giga_loss[loss=0.3191, simple_loss=0.3853, pruned_loss=0.1265, over 28946.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3568, pruned_loss=0.1116, over 5705873.78 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3935, pruned_loss=0.1425, over 5678013.33 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3522, pruned_loss=0.1078, over 5708257.38 frames. ], batch size: 213, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:36:41,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1754, 1.2214, 4.0789, 2.9789], device='cuda:1'), covar=tensor([0.1581, 0.2399, 0.0365, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0537, 0.0773, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:36:43,548 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=268405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:36:47,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=268408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:36:49,136 INFO [train.py:968] (1/2) Epoch 6, batch 40700, giga_loss[loss=0.2905, simple_loss=0.3654, pruned_loss=0.1078, over 28711.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3601, pruned_loss=0.1126, over 5715169.32 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3934, pruned_loss=0.1424, over 5682053.10 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3559, pruned_loss=0.1091, over 5713999.48 frames. ], batch size: 242, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:37:11,424 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=268437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:37:28,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4599, 1.4778, 1.1665, 1.2118], device='cuda:1'), covar=tensor([0.0595, 0.0467, 0.0902, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0446, 0.0498, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:37:30,830 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 40750, giga_loss[loss=0.287, simple_loss=0.3533, pruned_loss=0.1103, over 28467.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3618, pruned_loss=0.1134, over 5714443.41 frames. ], libri_tot_loss[loss=0.3391, simple_loss=0.3934, pruned_loss=0.1424, over 5682053.10 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3586, pruned_loss=0.1106, over 5713532.91 frames. ], batch size: 60, lr: 5.17e-03, grad_scale: 4.0
+2023-03-03 11:38:16,120 INFO [train.py:968] (1/2) Epoch 6, batch 40800, giga_loss[loss=0.305, simple_loss=0.3704, pruned_loss=0.1198, over 28933.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.365, pruned_loss=0.1161, over 5709995.98 frames. ], libri_tot_loss[loss=0.3393, simple_loss=0.3936, pruned_loss=0.1425, over 5687889.80 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3615, pruned_loss=0.1132, over 5704931.89 frames. ], batch size: 174, lr: 5.17e-03, grad_scale: 8.0
+2023-03-03 11:39:00,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-03 11:39:05,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2826, 1.3224, 1.2469, 1.4686], device='cuda:1'), covar=tensor([0.0781, 0.0327, 0.0319, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0118, 0.0122, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 11:39:05,839 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 40850, giga_loss[loss=0.3329, simple_loss=0.3898, pruned_loss=0.138, over 28880.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3725, pruned_loss=0.123, over 5684041.99 frames. ], libri_tot_loss[loss=0.339, simple_loss=0.3934, pruned_loss=0.1423, over 5684399.26 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3692, pruned_loss=0.1201, over 5684008.01 frames. ], batch size: 119, lr: 5.17e-03, grad_scale: 2.0
+2023-03-03 11:39:44,132 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:39:49,001 INFO [train.py:968] (1/2) Epoch 6, batch 40900, giga_loss[loss=0.3626, simple_loss=0.4265, pruned_loss=0.1494, over 28941.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3792, pruned_loss=0.1282, over 5689086.37 frames. ], libri_tot_loss[loss=0.3389, simple_loss=0.3933, pruned_loss=0.1423, over 5689099.77 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1254, over 5684606.93 frames. ], batch size: 136, lr: 5.17e-03, grad_scale: 2.0
+2023-03-03 11:40:01,654 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-03 11:40:31,986 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 40950, giga_loss[loss=0.4255, simple_loss=0.4328, pruned_loss=0.2091, over 23603.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3847, pruned_loss=0.1328, over 5677766.75 frames. ], libri_tot_loss[loss=0.3384, simple_loss=0.3929, pruned_loss=0.142, over 5694557.14 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3822, pruned_loss=0.1304, over 5669151.04 frames. ], batch size: 705, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:40:38,648 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:41:09,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7864, 1.6809, 1.6730, 1.7363], device='cuda:1'), covar=tensor([0.0990, 0.1646, 0.1529, 0.1245], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0734, 0.0650, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 11:41:15,165 INFO [train.py:968] (1/2) Epoch 6, batch 41000, giga_loss[loss=0.375, simple_loss=0.4187, pruned_loss=0.1656, over 28534.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3914, pruned_loss=0.1387, over 5676823.35 frames. ], libri_tot_loss[loss=0.3379, simple_loss=0.3925, pruned_loss=0.1416, over 5694047.75 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3898, pruned_loss=0.1371, over 5670270.08 frames. ], batch size: 336, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:41:43,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-03 11:42:02,111 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 41050, giga_loss[loss=0.3558, simple_loss=0.406, pruned_loss=0.1528, over 29062.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3973, pruned_loss=0.1434, over 5675331.49 frames. ], libri_tot_loss[loss=0.3375, simple_loss=0.3923, pruned_loss=0.1413, over 5695171.14 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3961, pruned_loss=0.1424, over 5669037.65 frames. ], batch size: 128, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:42:30,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2823, 1.2355, 1.1369, 1.4031], device='cuda:1'), covar=tensor([0.0636, 0.0427, 0.0317, 0.0678], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0118, 0.0122, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 11:42:54,205 INFO [train.py:968] (1/2) Epoch 6, batch 41100, giga_loss[loss=0.3484, simple_loss=0.3948, pruned_loss=0.151, over 28748.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.4002, pruned_loss=0.1468, over 5669010.61 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.392, pruned_loss=0.141, over 5700503.52 frames. ], giga_tot_loss[loss=0.3462, simple_loss=0.3997, pruned_loss=0.1464, over 5658469.95 frames. ], batch size: 119, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:43:46,000 INFO [optim.py:369] (1/2) 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 11:43:46,013 INFO [train.py:968] (1/2) Epoch 6, batch 41150, giga_loss[loss=0.3124, simple_loss=0.377, pruned_loss=0.1239, over 29062.00 frames. ], tot_loss[loss=0.3531, simple_loss=0.4037, pruned_loss=0.1513, over 5640897.32 frames. ], libri_tot_loss[loss=0.3378, simple_loss=0.3925, pruned_loss=0.1416, over 5693482.22 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4032, pruned_loss=0.1506, over 5638300.18 frames. ], batch size: 155, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:44:04,867 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 11:44:33,206 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 11:44:37,302 INFO [train.py:968] (1/2) Epoch 6, batch 41200, giga_loss[loss=0.4038, simple_loss=0.4386, pruned_loss=0.1845, over 28924.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4062, pruned_loss=0.1545, over 5631790.86 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3917, pruned_loss=0.141, over 5698199.59 frames. ], giga_tot_loss[loss=0.3581, simple_loss=0.4068, pruned_loss=0.1547, over 5624065.30 frames. ], batch size: 213, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:45:26,652 INFO [optim.py:369] (1/2) 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 11:45:26,665 INFO [train.py:968] (1/2) Epoch 6, batch 41250, giga_loss[loss=0.3585, simple_loss=0.4123, pruned_loss=0.1523, over 28732.00 frames. ], tot_loss[loss=0.366, simple_loss=0.412, pruned_loss=0.16, over 5632200.47 frames. ], libri_tot_loss[loss=0.3371, simple_loss=0.392, pruned_loss=0.1411, over 5701950.76 frames. ], giga_tot_loss[loss=0.3668, simple_loss=0.4126, pruned_loss=0.1605, over 5621095.00 frames. ], batch size: 99, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:45:49,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=268980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:46:03,366 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=268994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:46:10,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6593, 1.5763, 1.2754, 1.3590], device='cuda:1'), covar=tensor([0.0558, 0.0450, 0.0860, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0447, 0.0501, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:46:18,151 INFO [train.py:968] (1/2) Epoch 6, batch 41300, giga_loss[loss=0.4011, simple_loss=0.4324, pruned_loss=0.1848, over 29117.00 frames. ], tot_loss[loss=0.3674, simple_loss=0.4132, pruned_loss=0.1608, over 5644120.64 frames. ], libri_tot_loss[loss=0.3366, simple_loss=0.3915, pruned_loss=0.1409, over 5708015.44 frames. ], giga_tot_loss[loss=0.3694, simple_loss=0.4147, pruned_loss=0.162, over 5627784.78 frames. ], batch size: 113, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:46:51,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:47:08,218 INFO [optim.py:369] (1/2) 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 11:47:08,230 INFO [train.py:968] (1/2) Epoch 6, batch 41350, giga_loss[loss=0.4287, simple_loss=0.438, pruned_loss=0.2096, over 23427.00 frames. ], tot_loss[loss=0.365, simple_loss=0.4106, pruned_loss=0.1597, over 5638489.58 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3911, pruned_loss=0.1404, over 5709726.63 frames. ], giga_tot_loss[loss=0.368, simple_loss=0.4128, pruned_loss=0.1616, over 5622361.44 frames. ], batch size: 705, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:47:55,879 INFO [train.py:968] (1/2) Epoch 6, batch 41400, giga_loss[loss=0.2671, simple_loss=0.3461, pruned_loss=0.09408, over 28171.00 frames. ], tot_loss[loss=0.3614, simple_loss=0.408, pruned_loss=0.1574, over 5641823.73 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3909, pruned_loss=0.1401, over 5714437.01 frames. ], giga_tot_loss[loss=0.3649, simple_loss=0.4104, pruned_loss=0.1597, over 5622857.98 frames. ], batch size: 65, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:48:02,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-03 11:48:07,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:48:10,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:48:34,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6710, 1.7797, 1.1956, 1.4107], device='cuda:1'), covar=tensor([0.0734, 0.0680, 0.1092, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0449, 0.0504, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 11:48:37,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:48:43,097 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 41450, giga_loss[loss=0.3411, simple_loss=0.4001, pruned_loss=0.141, over 28657.00 frames. ], tot_loss[loss=0.3584, simple_loss=0.4063, pruned_loss=0.1553, over 5630181.34 frames. ], libri_tot_loss[loss=0.3353, simple_loss=0.3906, pruned_loss=0.14, over 5706891.51 frames. ], giga_tot_loss[loss=0.3623, simple_loss=0.4091, pruned_loss=0.1577, over 5619118.39 frames. ], batch size: 60, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:49:11,021 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:49:14,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:49:32,915 INFO [train.py:968] (1/2) Epoch 6, batch 41500, giga_loss[loss=0.3781, simple_loss=0.431, pruned_loss=0.1626, over 28205.00 frames. ], tot_loss[loss=0.3625, simple_loss=0.4097, pruned_loss=0.1577, over 5621444.60 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3908, pruned_loss=0.1401, over 5708543.26 frames. ], giga_tot_loss[loss=0.3656, simple_loss=0.4118, pruned_loss=0.1597, over 5610253.53 frames. ], batch size: 368, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:49:40,874 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:50:26,827 INFO [optim.py:369] (1/2) 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 11:50:26,840 INFO [train.py:968] (1/2) Epoch 6, batch 41550, giga_loss[loss=0.3284, simple_loss=0.3885, pruned_loss=0.1341, over 28707.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4077, pruned_loss=0.1559, over 5607116.52 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3908, pruned_loss=0.1401, over 5710555.24 frames. ], giga_tot_loss[loss=0.3627, simple_loss=0.4097, pruned_loss=0.1578, over 5594969.35 frames. ], batch size: 262, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:50:50,229 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:51:13,670 INFO [train.py:968] (1/2) Epoch 6, batch 41600, giga_loss[loss=0.3604, simple_loss=0.4079, pruned_loss=0.1564, over 28263.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4048, pruned_loss=0.1525, over 5625092.68 frames. ], libri_tot_loss[loss=0.3367, simple_loss=0.3915, pruned_loss=0.141, over 5715342.68 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4062, pruned_loss=0.1537, over 5607862.78 frames. ], batch size: 368, lr: 5.16e-03, grad_scale: 8.0
+2023-03-03 11:51:58,802 INFO [train.py:968] (1/2) Epoch 6, batch 41650, giga_loss[loss=0.3543, simple_loss=0.4106, pruned_loss=0.149, over 28773.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.4039, pruned_loss=0.1504, over 5643582.60 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3915, pruned_loss=0.1411, over 5720052.39 frames. ], giga_tot_loss[loss=0.3541, simple_loss=0.4054, pruned_loss=0.1514, over 5623253.50 frames. ], batch size: 243, lr: 5.16e-03, grad_scale: 4.0
+2023-03-03 11:51:59,471 INFO [optim.py:369] (1/2) 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 11:52:05,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5908, 1.7233, 1.5557, 1.5521], device='cuda:1'), covar=tensor([0.1106, 0.1761, 0.1527, 0.1469], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0729, 0.0646, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 11:52:07,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:52:10,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5617, 1.9587, 1.8092, 1.5658], device='cuda:1'), covar=tensor([0.1488, 0.1904, 0.1201, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0725, 0.0806, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 11:52:20,881 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:52:44,022 INFO [train.py:968] (1/2) Epoch 6, batch 41700, giga_loss[loss=0.285, simple_loss=0.3597, pruned_loss=0.1051, over 28971.00 frames. ], tot_loss[loss=0.3476, simple_loss=0.3997, pruned_loss=0.1477, over 5643889.00 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3904, pruned_loss=0.1406, over 5726852.66 frames. ], giga_tot_loss[loss=0.3507, simple_loss=0.4025, pruned_loss=0.1494, over 5617790.21 frames. ], batch size: 136, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:53:30,418 INFO [train.py:968] (1/2) Epoch 6, batch 41750, giga_loss[loss=0.3466, simple_loss=0.4047, pruned_loss=0.1442, over 28545.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.398, pruned_loss=0.1457, over 5640752.09 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3904, pruned_loss=0.1406, over 5731209.17 frames. ], giga_tot_loss[loss=0.3475, simple_loss=0.4004, pruned_loss=0.1473, over 5613973.14 frames. ], batch size: 336, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:53:32,072 INFO [optim.py:369] (1/2) 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 11:54:17,839 INFO [train.py:968] (1/2) Epoch 6, batch 41800, giga_loss[loss=0.2922, simple_loss=0.3658, pruned_loss=0.1093, over 28963.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3955, pruned_loss=0.1438, over 5646794.10 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.39, pruned_loss=0.1407, over 5717500.63 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3979, pruned_loss=0.1451, over 5634315.15 frames. ], batch size: 227, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:54:18,710 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:54:20,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:54:48,946 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:55:01,068 INFO [train.py:968] (1/2) Epoch 6, batch 41850, giga_loss[loss=0.407, simple_loss=0.4472, pruned_loss=0.1834, over 28971.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3952, pruned_loss=0.1435, over 5643384.78 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.39, pruned_loss=0.1407, over 5707988.41 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3973, pruned_loss=0.1446, over 5639646.87 frames. ], batch size: 213, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:55:03,227 INFO [optim.py:369] (1/2) 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 11:55:25,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1492, 1.5904, 1.4612, 1.2265], device='cuda:1'), covar=tensor([0.1508, 0.2054, 0.1214, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0728, 0.0812, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 11:55:41,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8795, 1.0805, 0.9671, 0.6708], device='cuda:1'), covar=tensor([0.1273, 0.1234, 0.0752, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.1515, 0.1350, 0.1328, 0.1416], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 11:55:50,593 INFO [train.py:968] (1/2) Epoch 6, batch 41900, giga_loss[loss=0.314, simple_loss=0.3828, pruned_loss=0.1226, over 28649.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3945, pruned_loss=0.1429, over 5633080.31 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3904, pruned_loss=0.141, over 5700537.42 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3959, pruned_loss=0.1435, over 5636713.98 frames. ], batch size: 99, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:56:18,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:56:42,061 INFO [train.py:968] (1/2) Epoch 6, batch 41950, giga_loss[loss=0.3774, simple_loss=0.4303, pruned_loss=0.1623, over 28045.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3923, pruned_loss=0.1399, over 5632433.31 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.39, pruned_loss=0.1408, over 5703442.50 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3938, pruned_loss=0.1406, over 5630766.14 frames. ], batch size: 412, lr: 5.16e-03, grad_scale: 2.0
+2023-03-03 11:56:42,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:57:07,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-03 11:57:08,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0079, 1.0642, 3.5452, 2.9876], device='cuda:1'), covar=tensor([0.1645, 0.2487, 0.0464, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0585, 0.0540, 0.0773, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 11:57:39,329 INFO [train.py:968] (1/2) Epoch 6, batch 42000, giga_loss[loss=0.3716, simple_loss=0.4345, pruned_loss=0.1544, over 28943.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3942, pruned_loss=0.1383, over 5648077.81 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3899, pruned_loss=0.1407, over 5705109.86 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3955, pruned_loss=0.1389, over 5644757.89 frames. ], batch size: 213, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 11:57:39,329 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 11:57:47,704 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 11:58:18,792 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:58:33,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=269759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:58:34,273 INFO [train.py:968] (1/2) Epoch 6, batch 42050, giga_loss[loss=0.3348, simple_loss=0.3947, pruned_loss=0.1375, over 28515.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.396, pruned_loss=0.1398, over 5653160.21 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3898, pruned_loss=0.1407, over 5699153.57 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3972, pruned_loss=0.1402, over 5654836.41 frames. ], batch size: 307, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 11:58:36,613 INFO [optim.py:369] (1/2) 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 11:59:21,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:59:24,943 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:59:26,512 INFO [train.py:968] (1/2) Epoch 6, batch 42100, giga_loss[loss=0.3893, simple_loss=0.4272, pruned_loss=0.1757, over 27586.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3965, pruned_loss=0.1407, over 5654270.05 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3897, pruned_loss=0.1406, over 5701193.90 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3976, pruned_loss=0.1412, over 5653273.51 frames. ], batch size: 472, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 11:59:46,432 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-03 11:59:49,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=269834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 11:59:51,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:00:14,074 INFO [train.py:968] (1/2) Epoch 6, batch 42150, giga_loss[loss=0.4097, simple_loss=0.436, pruned_loss=0.1917, over 28929.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.395, pruned_loss=0.1402, over 5657506.94 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3898, pruned_loss=0.1406, over 5694478.08 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.3958, pruned_loss=0.1405, over 5661981.60 frames. ], batch size: 145, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:00:16,385 INFO [optim.py:369] (1/2) 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 12:00:55,394 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=269902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:00:58,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=269905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:01:03,073 INFO [train.py:968] (1/2) Epoch 6, batch 42200, libri_loss[loss=0.3116, simple_loss=0.3688, pruned_loss=0.1272, over 29548.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3927, pruned_loss=0.1402, over 5657233.51 frames. ], libri_tot_loss[loss=0.3349, simple_loss=0.3893, pruned_loss=0.1403, over 5699102.72 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3939, pruned_loss=0.1408, over 5655592.84 frames. ], batch size: 76, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:01:14,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9214, 3.7262, 3.5405, 1.7641], device='cuda:1'), covar=tensor([0.0548, 0.0719, 0.0752, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0880, 0.0796, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 12:01:25,904 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=269934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:01:52,878 INFO [train.py:968] (1/2) Epoch 6, batch 42250, libri_loss[loss=0.3068, simple_loss=0.3692, pruned_loss=0.1222, over 29548.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3913, pruned_loss=0.139, over 5661037.53 frames. ], libri_tot_loss[loss=0.3345, simple_loss=0.389, pruned_loss=0.14, over 5702575.65 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3926, pruned_loss=0.1397, over 5656074.09 frames. ], batch size: 77, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:01:54,461 INFO [optim.py:369] (1/2) 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 12:02:12,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.18 vs. limit=5.0
+2023-03-03 12:02:40,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:02:41,521 INFO [train.py:968] (1/2) Epoch 6, batch 42300, giga_loss[loss=0.3801, simple_loss=0.4256, pruned_loss=0.1673, over 29035.00 frames. ], tot_loss[loss=0.334, simple_loss=0.392, pruned_loss=0.138, over 5667808.15 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3891, pruned_loss=0.14, over 5699870.48 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.393, pruned_loss=0.1385, over 5665710.67 frames. ], batch size: 213, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:02:53,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-03 12:03:33,945 INFO [train.py:968] (1/2) Epoch 6, batch 42350, giga_loss[loss=0.4545, simple_loss=0.4613, pruned_loss=0.2238, over 26669.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3926, pruned_loss=0.1381, over 5669164.97 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3894, pruned_loss=0.1401, over 5703344.55 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3932, pruned_loss=0.1384, over 5663876.46 frames. ], batch size: 555, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:03:35,332 INFO [optim.py:369] (1/2) 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 12:04:23,784 INFO [train.py:968] (1/2) Epoch 6, batch 42400, giga_loss[loss=0.3288, simple_loss=0.3917, pruned_loss=0.1329, over 28593.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3937, pruned_loss=0.1394, over 5671468.09 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3898, pruned_loss=0.1405, over 5705982.20 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3938, pruned_loss=0.1392, over 5664659.45 frames. ], batch size: 307, lr: 5.15e-03, grad_scale: 8.0
+2023-03-03 12:04:33,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:01,254 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:03,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:10,190 INFO [train.py:968] (1/2) Epoch 6, batch 42450, giga_loss[loss=0.3101, simple_loss=0.3695, pruned_loss=0.1254, over 28546.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3907, pruned_loss=0.1378, over 5682862.92 frames. ], libri_tot_loss[loss=0.3354, simple_loss=0.3897, pruned_loss=0.1406, over 5710622.52 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3909, pruned_loss=0.1375, over 5672175.35 frames. ], batch size: 71, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:05:12,073 INFO [optim.py:369] (1/2) 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 12:05:14,593 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:23,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4699, 1.5258, 1.4165, 1.5436], device='cuda:1'), covar=tensor([0.1111, 0.1537, 0.1732, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0732, 0.0645, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 12:05:32,425 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:44,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1591, 1.2968, 4.2765, 3.2686], device='cuda:1'), covar=tensor([0.1585, 0.2176, 0.0361, 0.0639], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0542, 0.0784, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:05:57,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:05:58,498 INFO [train.py:968] (1/2) Epoch 6, batch 42500, giga_loss[loss=0.294, simple_loss=0.3676, pruned_loss=0.1102, over 28984.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3896, pruned_loss=0.1379, over 5675504.19 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3896, pruned_loss=0.1407, over 5713214.30 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3899, pruned_loss=0.1375, over 5664010.30 frames. ], batch size: 164, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:06:17,984 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 42550, libri_loss[loss=0.3072, simple_loss=0.3749, pruned_loss=0.1198, over 29741.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3896, pruned_loss=0.1387, over 5675329.17 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3903, pruned_loss=0.1411, over 5703596.78 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3892, pruned_loss=0.1379, over 5673606.85 frames. ], batch size: 87, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:06:52,117 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:06:54,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:07:22,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:07:38,141 INFO [train.py:968] (1/2) Epoch 6, batch 42600, giga_loss[loss=0.2916, simple_loss=0.3606, pruned_loss=0.1113, over 28925.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3885, pruned_loss=0.1383, over 5675513.16 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3904, pruned_loss=0.1411, over 5703423.34 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3881, pruned_loss=0.1377, over 5674006.84 frames. ], batch size: 174, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:08:19,230 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:08:22,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:08:27,423 INFO [train.py:968] (1/2) Epoch 6, batch 42650, giga_loss[loss=0.2946, simple_loss=0.3551, pruned_loss=0.1171, over 28619.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3887, pruned_loss=0.1395, over 5670739.16 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3903, pruned_loss=0.141, over 5708737.68 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3883, pruned_loss=0.139, over 5663887.07 frames. ], batch size: 85, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:08:33,906 INFO [optim.py:369] (1/2) 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 12:08:53,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:09:22,909 INFO [train.py:968] (1/2) Epoch 6, batch 42700, giga_loss[loss=0.3384, simple_loss=0.394, pruned_loss=0.1414, over 28880.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3892, pruned_loss=0.1406, over 5654683.86 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3904, pruned_loss=0.1411, over 5705983.99 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3889, pruned_loss=0.1401, over 5651336.39 frames. ], batch size: 227, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:09:23,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-03 12:09:38,668 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-03 12:10:07,697 INFO [train.py:968] (1/2) Epoch 6, batch 42750, giga_loss[loss=0.3143, simple_loss=0.3862, pruned_loss=0.1212, over 28791.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3889, pruned_loss=0.1394, over 5669150.41 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3908, pruned_loss=0.1414, over 5710986.27 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3883, pruned_loss=0.1387, over 5660850.80 frames. ], batch size: 119, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:10:11,532 INFO [optim.py:369] (1/2) 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 12:10:34,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-03 12:10:51,294 INFO [train.py:968] (1/2) Epoch 6, batch 42800, libri_loss[loss=0.2961, simple_loss=0.3674, pruned_loss=0.1124, over 29391.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3881, pruned_loss=0.1376, over 5677589.96 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3903, pruned_loss=0.1412, over 5717724.36 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.388, pruned_loss=0.1372, over 5663655.16 frames. ], batch size: 92, lr: 5.15e-03, grad_scale: 8.0
+2023-03-03 12:11:16,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:11:21,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1364, 1.6358, 1.4560, 1.1966], device='cuda:1'), covar=tensor([0.1182, 0.1976, 0.1091, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0734, 0.0815, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 12:11:36,791 INFO [train.py:968] (1/2) Epoch 6, batch 42850, giga_loss[loss=0.3627, simple_loss=0.4142, pruned_loss=0.1556, over 29048.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3887, pruned_loss=0.1374, over 5685296.33 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3908, pruned_loss=0.1416, over 5719249.29 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3881, pruned_loss=0.1365, over 5671120.79 frames. ], batch size: 128, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:11:39,832 INFO [optim.py:369] (1/2) 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 12:12:21,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-03 12:12:22,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=270604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:12:28,727 INFO [train.py:968] (1/2) Epoch 6, batch 42900, giga_loss[loss=0.4428, simple_loss=0.4684, pruned_loss=0.2086, over 27590.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3896, pruned_loss=0.1381, over 5670374.12 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3906, pruned_loss=0.1415, over 5703092.56 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3892, pruned_loss=0.1374, over 5673212.26 frames. ], batch size: 472, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:13:15,138 INFO [train.py:968] (1/2) Epoch 6, batch 42950, giga_loss[loss=0.3367, simple_loss=0.3885, pruned_loss=0.1425, over 28734.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3922, pruned_loss=0.1404, over 5681046.33 frames. ], libri_tot_loss[loss=0.3372, simple_loss=0.391, pruned_loss=0.1417, over 5705141.62 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3916, pruned_loss=0.1396, over 5680344.60 frames. ], batch size: 284, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:13:20,288 INFO [optim.py:369] (1/2) 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 12:13:30,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270673.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 12:13:32,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3182, 1.6299, 1.5495, 1.5212], device='cuda:1'), covar=tensor([0.0744, 0.0297, 0.0281, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0120, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0070], device='cuda:1')
+2023-03-03 12:13:39,794 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:13:42,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:14:07,397 INFO [train.py:968] (1/2) Epoch 6, batch 43000, giga_loss[loss=0.4614, simple_loss=0.4659, pruned_loss=0.2284, over 26527.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.393, pruned_loss=0.1424, over 5681336.50 frames. ], libri_tot_loss[loss=0.3362, simple_loss=0.3903, pruned_loss=0.1411, over 5709949.56 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3932, pruned_loss=0.1424, over 5675839.50 frames. ], batch size: 555, lr: 5.15e-03, grad_scale: 4.0
+2023-03-03 12:14:15,609 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:14:45,831 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=270747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:14:49,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=270750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:14:59,463 INFO [train.py:968] (1/2) Epoch 6, batch 43050, giga_loss[loss=0.4498, simple_loss=0.4672, pruned_loss=0.2163, over 24150.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3943, pruned_loss=0.1444, over 5680328.86 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3903, pruned_loss=0.1412, over 5713930.81 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3944, pruned_loss=0.1442, over 5671678.69 frames. ], batch size: 705, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:15:02,988 INFO [optim.py:369] (1/2) 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 12:15:11,318 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-03 12:15:17,807 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=270779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:15:28,479 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:15:44,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 12:15:48,370 INFO [train.py:968] (1/2) Epoch 6, batch 43100, giga_loss[loss=0.3264, simple_loss=0.3888, pruned_loss=0.132, over 28757.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3963, pruned_loss=0.1464, over 5666269.72 frames. ], libri_tot_loss[loss=0.3358, simple_loss=0.3898, pruned_loss=0.1409, over 5713948.15 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.397, pruned_loss=0.1467, over 5658883.93 frames. ], batch size: 242, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:16:07,954 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=270832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:16:34,039 INFO [train.py:968] (1/2) Epoch 6, batch 43150, giga_loss[loss=0.3271, simple_loss=0.3857, pruned_loss=0.1343, over 28483.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3948, pruned_loss=0.1452, over 5666101.20 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3897, pruned_loss=0.1407, over 5715049.84 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3956, pruned_loss=0.1457, over 5658248.57 frames. ], batch size: 336, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:16:39,580 INFO [optim.py:369] (1/2) 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 12:17:22,421 INFO [train.py:968] (1/2) Epoch 6, batch 43200, giga_loss[loss=0.4027, simple_loss=0.4269, pruned_loss=0.1893, over 24117.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3938, pruned_loss=0.1427, over 5669214.65 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3898, pruned_loss=0.1407, over 5716293.94 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3943, pruned_loss=0.1431, over 5661409.73 frames. ], batch size: 705, lr: 5.14e-03, grad_scale: 8.0
+2023-03-03 12:17:35,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2058, 1.3871, 1.1942, 1.0290], device='cuda:1'), covar=tensor([0.2562, 0.2352, 0.2590, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.0890, 0.1015, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:17:39,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2653, 1.8609, 1.3145, 0.4236], device='cuda:1'), covar=tensor([0.2114, 0.1673, 0.2298, 0.2483], device='cuda:1'), in_proj_covar=tensor([0.1446, 0.1372, 0.1405, 0.1200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 12:18:12,841 INFO [train.py:968] (1/2) Epoch 6, batch 43250, giga_loss[loss=0.3721, simple_loss=0.4003, pruned_loss=0.1719, over 26601.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.392, pruned_loss=0.1413, over 5657123.34 frames. ], libri_tot_loss[loss=0.3361, simple_loss=0.3901, pruned_loss=0.141, over 5711252.30 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3922, pruned_loss=0.1413, over 5655153.65 frames. ], batch size: 555, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:18:17,149 INFO [optim.py:369] (1/2) 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 12:18:52,438 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:18:55,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9062, 1.1578, 3.4040, 2.9098], device='cuda:1'), covar=tensor([0.1739, 0.2337, 0.0464, 0.0778], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0545, 0.0786, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:18:57,292 INFO [train.py:968] (1/2) Epoch 6, batch 43300, giga_loss[loss=0.3065, simple_loss=0.3715, pruned_loss=0.1208, over 29064.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3895, pruned_loss=0.1394, over 5666674.04 frames. ], libri_tot_loss[loss=0.3365, simple_loss=0.3905, pruned_loss=0.1412, over 5711105.80 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3893, pruned_loss=0.1392, over 5663885.88 frames. ], batch size: 136, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:19:30,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:19:41,482 INFO [train.py:968] (1/2) Epoch 6, batch 43350, giga_loss[loss=0.3045, simple_loss=0.3676, pruned_loss=0.1207, over 28925.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3884, pruned_loss=0.1397, over 5663082.00 frames. ], libri_tot_loss[loss=0.337, simple_loss=0.3908, pruned_loss=0.1416, over 5706494.90 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3878, pruned_loss=0.139, over 5663313.88 frames. ], batch size: 145, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:19:47,484 INFO [optim.py:369] (1/2) 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 12:20:12,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8422, 1.8101, 1.7230, 1.6194], device='cuda:1'), covar=tensor([0.1162, 0.1913, 0.1588, 0.1730], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0731, 0.0646, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 12:20:29,469 INFO [train.py:968] (1/2) Epoch 6, batch 43400, giga_loss[loss=0.4895, simple_loss=0.4935, pruned_loss=0.2427, over 27926.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3884, pruned_loss=0.1397, over 5667167.55 frames. ], libri_tot_loss[loss=0.3369, simple_loss=0.3907, pruned_loss=0.1415, over 5709587.62 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3879, pruned_loss=0.1393, over 5663907.94 frames. ], batch size: 412, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:20:37,644 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:20:45,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3125, 1.4027, 1.3834, 1.3459], device='cuda:1'), covar=tensor([0.1179, 0.1383, 0.1778, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0734, 0.0651, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 12:21:17,229 INFO [train.py:968] (1/2) Epoch 6, batch 43450, giga_loss[loss=0.3095, simple_loss=0.3821, pruned_loss=0.1185, over 28622.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3931, pruned_loss=0.1423, over 5672464.51 frames. ], libri_tot_loss[loss=0.3368, simple_loss=0.3907, pruned_loss=0.1414, over 5714724.37 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3928, pruned_loss=0.1421, over 5664057.50 frames. ], batch size: 242, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:21:20,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:21:22,319 INFO [optim.py:369] (1/2) 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 12:21:45,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271191.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:21:48,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271194.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:22:02,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:22:06,256 INFO [train.py:968] (1/2) Epoch 6, batch 43500, giga_loss[loss=0.3154, simple_loss=0.3941, pruned_loss=0.1184, over 28652.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3945, pruned_loss=0.14, over 5678293.07 frames. ], libri_tot_loss[loss=0.3363, simple_loss=0.3902, pruned_loss=0.1412, over 5718850.79 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3948, pruned_loss=0.14, over 5667171.64 frames. ], batch size: 262, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:22:06,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2088, 2.5824, 1.1854, 1.3318], device='cuda:1'), covar=tensor([0.0868, 0.0380, 0.0907, 0.1297], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0482, 0.0315, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0017, 0.0022], device='cuda:1')
+2023-03-03 12:22:20,937 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271223.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 12:23:00,437 INFO [train.py:968] (1/2) Epoch 6, batch 43550, giga_loss[loss=0.3781, simple_loss=0.4028, pruned_loss=0.1767, over 23810.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3952, pruned_loss=0.1399, over 5671491.60 frames. ], libri_tot_loss[loss=0.3364, simple_loss=0.3903, pruned_loss=0.1412, over 5720703.96 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3954, pruned_loss=0.1399, over 5660529.91 frames. ], batch size: 705, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:23:04,158 INFO [optim.py:369] (1/2) 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 12:23:43,358 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:23:45,344 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:23:45,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:23:45,915 INFO [train.py:968] (1/2) Epoch 6, batch 43600, giga_loss[loss=0.2859, simple_loss=0.3543, pruned_loss=0.1088, over 28574.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3969, pruned_loss=0.1414, over 5673594.29 frames. ], libri_tot_loss[loss=0.336, simple_loss=0.3898, pruned_loss=0.1411, over 5724705.15 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3976, pruned_loss=0.1415, over 5659970.41 frames. ], batch size: 85, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:24:15,870 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:26,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:30,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:36,668 INFO [train.py:968] (1/2) Epoch 6, batch 43650, giga_loss[loss=0.4784, simple_loss=0.4739, pruned_loss=0.2414, over 26559.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3978, pruned_loss=0.1432, over 5669509.45 frames. ], libri_tot_loss[loss=0.3356, simple_loss=0.3895, pruned_loss=0.1409, over 5728098.03 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3988, pruned_loss=0.1435, over 5654865.21 frames. ], batch size: 555, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:24:41,918 INFO [optim.py:369] (1/2) 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 12:24:54,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:24:54,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:25:08,861 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-03 12:25:18,552 INFO [train.py:968] (1/2) Epoch 6, batch 43700, giga_loss[loss=0.4007, simple_loss=0.4222, pruned_loss=0.1896, over 26537.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3958, pruned_loss=0.1423, over 5681074.49 frames. ], libri_tot_loss[loss=0.3355, simple_loss=0.3894, pruned_loss=0.1408, over 5732992.16 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3969, pruned_loss=0.1426, over 5663415.63 frames. ], batch size: 555, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:25:59,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5646, 2.0111, 1.5963, 1.8534], device='cuda:1'), covar=tensor([0.0589, 0.0707, 0.0933, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0454, 0.0505, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 12:26:10,621 INFO [train.py:968] (1/2) Epoch 6, batch 43750, giga_loss[loss=0.3757, simple_loss=0.4239, pruned_loss=0.1638, over 28549.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3942, pruned_loss=0.1422, over 5669711.29 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3896, pruned_loss=0.1409, over 5727478.26 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3949, pruned_loss=0.1424, over 5660631.34 frames. ], batch size: 336, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:26:18,430 INFO [optim.py:369] (1/2) 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 12:26:43,971 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 43800, giga_loss[loss=0.345, simple_loss=0.396, pruned_loss=0.147, over 28667.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3916, pruned_loss=0.1404, over 5674559.89 frames. ], libri_tot_loss[loss=0.3357, simple_loss=0.3897, pruned_loss=0.1408, over 5728867.10 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3922, pruned_loss=0.1407, over 5664781.63 frames. ], batch size: 262, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:27:11,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:27:13,925 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4466, 1.7354, 1.7088, 1.4743], device='cuda:1'), covar=tensor([0.1480, 0.1923, 0.1149, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0735, 0.0816, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 12:27:15,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:27:45,285 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 43850, giga_loss[loss=0.3424, simple_loss=0.3987, pruned_loss=0.143, over 28650.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3909, pruned_loss=0.1404, over 5678016.13 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3891, pruned_loss=0.1402, over 5733253.20 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3919, pruned_loss=0.1411, over 5665104.77 frames. ], batch size: 307, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:27:55,726 INFO [optim.py:369] (1/2) 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 12:28:38,143 INFO [train.py:968] (1/2) Epoch 6, batch 43900, giga_loss[loss=0.3733, simple_loss=0.4203, pruned_loss=0.1631, over 28916.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3922, pruned_loss=0.1418, over 5672369.46 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3887, pruned_loss=0.1398, over 5729019.64 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3935, pruned_loss=0.1427, over 5664150.91 frames. ], batch size: 199, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:29:02,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:06,163 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:16,108 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:23,060 INFO [train.py:968] (1/2) Epoch 6, batch 43950, giga_loss[loss=0.3864, simple_loss=0.4207, pruned_loss=0.176, over 27824.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3915, pruned_loss=0.1417, over 5674065.08 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3887, pruned_loss=0.1398, over 5728229.16 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3925, pruned_loss=0.1425, over 5667437.23 frames. ], batch size: 412, lr: 5.14e-03, grad_scale: 2.0
+2023-03-03 12:29:28,678 INFO [optim.py:369] (1/2) 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 12:29:29,718 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:29:49,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=271685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:30:09,198 INFO [train.py:968] (1/2) Epoch 6, batch 44000, giga_loss[loss=0.2969, simple_loss=0.3654, pruned_loss=0.1142, over 29027.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3881, pruned_loss=0.1396, over 5683149.04 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3881, pruned_loss=0.1394, over 5733473.81 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3894, pruned_loss=0.1406, over 5671470.25 frames. ], batch size: 155, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:30:53,095 INFO [train.py:968] (1/2) Epoch 6, batch 44050, giga_loss[loss=0.4409, simple_loss=0.4542, pruned_loss=0.2138, over 26594.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3883, pruned_loss=0.139, over 5681873.49 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3879, pruned_loss=0.1391, over 5736726.89 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3896, pruned_loss=0.1401, over 5668059.95 frames. ], batch size: 555, lr: 5.14e-03, grad_scale: 4.0
+2023-03-03 12:31:01,655 INFO [optim.py:369] (1/2) 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 12:31:02,184 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 44100, giga_loss[loss=0.3096, simple_loss=0.3742, pruned_loss=0.1225, over 28831.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3904, pruned_loss=0.1399, over 5674561.58 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3877, pruned_loss=0.1389, over 5739772.12 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3916, pruned_loss=0.141, over 5659981.55 frames. ], batch size: 106, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:31:56,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=271819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:03,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=271828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:05,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=271831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:33,461 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=271860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:32:33,927 INFO [train.py:968] (1/2) Epoch 6, batch 44150, giga_loss[loss=0.2988, simple_loss=0.3693, pruned_loss=0.1141, over 29017.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3904, pruned_loss=0.1396, over 5686165.43 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3877, pruned_loss=0.1388, over 5742878.45 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3915, pruned_loss=0.1405, over 5670807.28 frames. ], batch size: 164, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:32:44,059 INFO [optim.py:369] (1/2) 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 12:32:52,501 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-03 12:32:53,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-03 12:33:24,617 INFO [train.py:968] (1/2) Epoch 6, batch 44200, giga_loss[loss=0.3222, simple_loss=0.3635, pruned_loss=0.1404, over 23835.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3904, pruned_loss=0.139, over 5666359.89 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3875, pruned_loss=0.1387, over 5734394.10 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3915, pruned_loss=0.1399, over 5659949.63 frames. ], batch size: 705, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:34:08,702 INFO [train.py:968] (1/2) Epoch 6, batch 44250, giga_loss[loss=0.294, simple_loss=0.3725, pruned_loss=0.1078, over 28980.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3927, pruned_loss=0.1382, over 5668452.86 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3881, pruned_loss=0.1393, over 5726795.94 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3931, pruned_loss=0.1383, over 5668125.10 frames. ], batch size: 155, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:34:16,002 INFO [optim.py:369] (1/2) 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 12:34:54,019 INFO [train.py:968] (1/2) Epoch 6, batch 44300, giga_loss[loss=0.2932, simple_loss=0.371, pruned_loss=0.1077, over 28590.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3956, pruned_loss=0.1387, over 5659095.89 frames. ], libri_tot_loss[loss=0.3341, simple_loss=0.3886, pruned_loss=0.1399, over 5709178.21 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3956, pruned_loss=0.1383, over 5672993.73 frames. ], batch size: 78, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:35:15,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:35:42,744 INFO [train.py:968] (1/2) Epoch 6, batch 44350, giga_loss[loss=0.3728, simple_loss=0.4184, pruned_loss=0.1636, over 28909.00 frames. ], tot_loss[loss=0.3397, simple_loss=0.3982, pruned_loss=0.1406, over 5671486.91 frames. ], libri_tot_loss[loss=0.3348, simple_loss=0.3891, pruned_loss=0.1403, over 5706802.18 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.398, pruned_loss=0.1399, over 5683007.62 frames. ], batch size: 227, lr: 5.13e-03, grad_scale: 2.0
+2023-03-03 12:35:47,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4220, 1.5749, 1.3253, 1.5426], device='cuda:1'), covar=tensor([0.2105, 0.2074, 0.2124, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.1156, 0.0887, 0.1013, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:35:52,041 INFO [optim.py:369] (1/2) 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 12:36:19,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-03 12:36:34,227 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-03 12:36:36,466 INFO [train.py:968] (1/2) Epoch 6, batch 44400, giga_loss[loss=0.3514, simple_loss=0.4066, pruned_loss=0.1481, over 28537.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.4012, pruned_loss=0.1446, over 5661781.56 frames. ], libri_tot_loss[loss=0.3347, simple_loss=0.389, pruned_loss=0.1402, over 5707840.17 frames. ], giga_tot_loss[loss=0.3447, simple_loss=0.4012, pruned_loss=0.1441, over 5669730.17 frames. ], batch size: 336, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:37:00,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4388, 3.3520, 1.4900, 1.4236], device='cuda:1'), covar=tensor([0.0875, 0.0292, 0.0824, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0493, 0.0317, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 12:37:07,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:37:23,549 INFO [train.py:968] (1/2) Epoch 6, batch 44450, giga_loss[loss=0.3385, simple_loss=0.4053, pruned_loss=0.1358, over 28792.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.4015, pruned_loss=0.1461, over 5649880.27 frames. ], libri_tot_loss[loss=0.3342, simple_loss=0.3885, pruned_loss=0.14, over 5711618.04 frames. ], giga_tot_loss[loss=0.347, simple_loss=0.4021, pruned_loss=0.1459, over 5652173.28 frames. ], batch size: 186, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:37:32,524 INFO [optim.py:369] (1/2) 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 12:37:34,933 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:37:38,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:37:54,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:38:03,460 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:38:07,663 INFO [train.py:968] (1/2) Epoch 6, batch 44500, giga_loss[loss=0.3249, simple_loss=0.3981, pruned_loss=0.1258, over 28825.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.4011, pruned_loss=0.1457, over 5646068.61 frames. ], libri_tot_loss[loss=0.3343, simple_loss=0.3886, pruned_loss=0.14, over 5699033.54 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.402, pruned_loss=0.1458, over 5656655.55 frames. ], batch size: 174, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:38:37,837 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:38:49,971 INFO [train.py:968] (1/2) Epoch 6, batch 44550, giga_loss[loss=0.3198, simple_loss=0.3852, pruned_loss=0.1272, over 28693.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3978, pruned_loss=0.1427, over 5654107.93 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3874, pruned_loss=0.1393, over 5704159.85 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3999, pruned_loss=0.1435, over 5656745.95 frames. ], batch size: 284, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:38:52,493 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4883, 1.6106, 1.3234, 1.7417], device='cuda:1'), covar=tensor([0.2241, 0.2226, 0.2385, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1162, 0.0891, 0.1018, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:38:56,214 INFO [optim.py:369] (1/2) 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 12:39:13,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:39:14,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:39:35,959 INFO [train.py:968] (1/2) Epoch 6, batch 44600, giga_loss[loss=0.3463, simple_loss=0.4109, pruned_loss=0.1408, over 28938.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3983, pruned_loss=0.141, over 5658481.51 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3873, pruned_loss=0.1391, over 5703480.51 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.4003, pruned_loss=0.1419, over 5660341.04 frames. ], batch size: 213, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:39:43,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:39:52,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:40:00,296 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:40:03,607 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 44650, giga_loss[loss=0.3713, simple_loss=0.4257, pruned_loss=0.1584, over 28592.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3977, pruned_loss=0.1403, over 5669609.44 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3869, pruned_loss=0.139, over 5710319.12 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.4002, pruned_loss=0.1412, over 5663328.40 frames. ], batch size: 336, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:40:32,815 INFO [optim.py:369] (1/2) 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 12:40:33,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:41:14,326 INFO [train.py:968] (1/2) Epoch 6, batch 44700, giga_loss[loss=0.2962, simple_loss=0.3646, pruned_loss=0.1139, over 28447.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3976, pruned_loss=0.1406, over 5671777.90 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3868, pruned_loss=0.1389, over 5712538.78 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3997, pruned_loss=0.1414, over 5664276.98 frames. ], batch size: 71, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:42:00,314 INFO [train.py:968] (1/2) Epoch 6, batch 44750, giga_loss[loss=0.3378, simple_loss=0.3737, pruned_loss=0.1509, over 23085.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3965, pruned_loss=0.1403, over 5677673.32 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3869, pruned_loss=0.139, over 5713727.83 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3982, pruned_loss=0.1409, over 5670341.23 frames. ], batch size: 705, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:42:06,832 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 44800, giga_loss[loss=0.3008, simple_loss=0.3701, pruned_loss=0.1158, over 28902.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3955, pruned_loss=0.1413, over 5653705.09 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.387, pruned_loss=0.1391, over 5702294.62 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3972, pruned_loss=0.1418, over 5655341.73 frames. ], batch size: 186, lr: 5.13e-03, grad_scale: 8.0
+2023-03-03 12:42:55,889 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:43:29,605 INFO [train.py:968] (1/2) Epoch 6, batch 44850, giga_loss[loss=0.3006, simple_loss=0.3672, pruned_loss=0.1171, over 28906.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3934, pruned_loss=0.1402, over 5659693.47 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3874, pruned_loss=0.1392, over 5706408.41 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3948, pruned_loss=0.1406, over 5655514.54 frames. ], batch size: 112, lr: 5.13e-03, grad_scale: 8.0
+2023-03-03 12:43:39,253 INFO [optim.py:369] (1/2) 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 12:43:44,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272574.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:43:55,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7441, 2.4795, 1.4518, 1.3169], device='cuda:1'), covar=tensor([0.1905, 0.1080, 0.1435, 0.1581], device='cuda:1'), in_proj_covar=tensor([0.1534, 0.1373, 0.1349, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 12:44:19,440 INFO [train.py:968] (1/2) Epoch 6, batch 44900, giga_loss[loss=0.3325, simple_loss=0.3831, pruned_loss=0.141, over 28765.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3914, pruned_loss=0.1395, over 5661818.96 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3876, pruned_loss=0.1394, over 5711128.41 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3923, pruned_loss=0.1396, over 5653264.02 frames. ], batch size: 119, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:44:25,114 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:44:31,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:44:34,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1507, 1.5874, 1.5746, 1.2796], device='cuda:1'), covar=tensor([0.1332, 0.1957, 0.1047, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0734, 0.0810, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 12:44:38,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3410, 1.4649, 1.2623, 1.5442], device='cuda:1'), covar=tensor([0.2031, 0.1940, 0.1911, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1169, 0.0894, 0.1024, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:44:47,586 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:45:07,508 INFO [train.py:968] (1/2) Epoch 6, batch 44950, giga_loss[loss=0.3272, simple_loss=0.384, pruned_loss=0.1352, over 28933.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3908, pruned_loss=0.14, over 5656085.60 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3878, pruned_loss=0.1396, over 5703412.76 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3914, pruned_loss=0.1399, over 5656286.53 frames. ], batch size: 145, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:45:16,316 INFO [optim.py:369] (1/2) 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 12:45:43,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-03 12:45:45,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:45:53,254 INFO [train.py:968] (1/2) Epoch 6, batch 45000, libri_loss[loss=0.3171, simple_loss=0.3813, pruned_loss=0.1265, over 25830.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3893, pruned_loss=0.1386, over 5651286.64 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3871, pruned_loss=0.1391, over 5694874.43 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3905, pruned_loss=0.139, over 5658428.38 frames. ], batch size: 136, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:45:53,255 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 12:46:02,093 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 12:46:44,893 INFO [train.py:968] (1/2) Epoch 6, batch 45050, libri_loss[loss=0.363, simple_loss=0.4143, pruned_loss=0.1558, over 29492.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3862, pruned_loss=0.1354, over 5646505.35 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3872, pruned_loss=0.1392, over 5691102.83 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3871, pruned_loss=0.1356, over 5653535.02 frames. ], batch size: 85, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:46:49,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:46:50,038 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=272766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:46:52,162 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:47:19,027 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:47:30,899 INFO [train.py:968] (1/2) Epoch 6, batch 45100, giga_loss[loss=0.4043, simple_loss=0.4368, pruned_loss=0.1859, over 26672.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3851, pruned_loss=0.1336, over 5652816.53 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3878, pruned_loss=0.1395, over 5683914.58 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3852, pruned_loss=0.1334, over 5663572.30 frames. ], batch size: 555, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:48:10,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=272845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:48:13,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=272848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:48:23,812 INFO [train.py:968] (1/2) Epoch 6, batch 45150, giga_loss[loss=0.3364, simple_loss=0.3942, pruned_loss=0.1393, over 28900.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3851, pruned_loss=0.134, over 5646147.50 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3879, pruned_loss=0.1396, over 5684115.01 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.385, pruned_loss=0.1337, over 5654281.69 frames. ], batch size: 174, lr: 5.13e-03, grad_scale: 4.0
+2023-03-03 12:48:32,572 INFO [optim.py:369] (1/2) 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 12:48:38,340 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=272877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:48:56,210 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 45200, libri_loss[loss=0.3394, simple_loss=0.3977, pruned_loss=0.1405, over 29253.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3824, pruned_loss=0.1341, over 5629604.41 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.388, pruned_loss=0.1396, over 5688266.13 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3822, pruned_loss=0.1337, over 5631517.10 frames. ], batch size: 97, lr: 5.12e-03, grad_scale: 8.0
+2023-03-03 12:49:20,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4755, 1.7917, 1.7817, 1.5018], device='cuda:1'), covar=tensor([0.1677, 0.2117, 0.1265, 0.1437], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0734, 0.0812, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 12:49:23,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0356, 1.2279, 3.5982, 3.0102], device='cuda:1'), covar=tensor([0.1612, 0.2244, 0.0414, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0590, 0.0545, 0.0790, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:49:47,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272949.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:49:56,939 INFO [train.py:968] (1/2) Epoch 6, batch 45250, giga_loss[loss=0.3114, simple_loss=0.369, pruned_loss=0.1269, over 28765.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3817, pruned_loss=0.1339, over 5648064.44 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3879, pruned_loss=0.1396, over 5693113.73 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3814, pruned_loss=0.1333, over 5642920.79 frames. ], batch size: 92, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:50:04,902 INFO [optim.py:369] (1/2) 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 12:50:22,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=272991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:50:33,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=273006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:50:37,749 INFO [train.py:968] (1/2) Epoch 6, batch 45300, giga_loss[loss=0.3349, simple_loss=0.4031, pruned_loss=0.1334, over 28692.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3835, pruned_loss=0.1345, over 5651604.15 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3877, pruned_loss=0.1395, over 5699292.27 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3832, pruned_loss=0.1339, over 5639020.63 frames. ], batch size: 242, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:50:41,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:50:52,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1709, 1.2939, 4.0527, 3.2969], device='cuda:1'), covar=tensor([0.1695, 0.2409, 0.0366, 0.0704], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0544, 0.0791, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 12:51:01,717 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:51:04,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 6, batch 45350, giga_loss[loss=0.4252, simple_loss=0.4459, pruned_loss=0.2023, over 26520.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3842, pruned_loss=0.1342, over 5656480.51 frames. ], libri_tot_loss[loss=0.3337, simple_loss=0.388, pruned_loss=0.1397, over 5703827.01 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3836, pruned_loss=0.1333, over 5641329.57 frames. ], batch size: 555, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:51:34,814 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:51:35,151 INFO [optim.py:369] (1/2) 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 12:51:44,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4489, 3.2296, 1.5309, 1.4357], device='cuda:1'), covar=tensor([0.0865, 0.0263, 0.0832, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0493, 0.0315, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 12:51:56,397 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273092.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:51:58,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273095.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 12:52:14,564 INFO [train.py:968] (1/2) Epoch 6, batch 45400, giga_loss[loss=0.3282, simple_loss=0.3869, pruned_loss=0.1348, over 29013.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3837, pruned_loss=0.1343, over 5644997.77 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3877, pruned_loss=0.1395, over 5705864.94 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3834, pruned_loss=0.1338, over 5630954.49 frames. ], batch size: 136, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:52:26,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273124.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 12:52:35,955 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:38,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:41,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:58,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:52:59,658 INFO [train.py:968] (1/2) Epoch 6, batch 45450, libri_loss[loss=0.3813, simple_loss=0.4183, pruned_loss=0.1721, over 19167.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3833, pruned_loss=0.134, over 5644323.39 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3879, pruned_loss=0.1394, over 5703892.49 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3827, pruned_loss=0.1334, over 5632819.52 frames. ], batch size: 187, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:52:59,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:53:03,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:53:26,005 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:53:33,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3268, 1.1804, 1.0938, 1.3728], device='cuda:1'), covar=tensor([0.0749, 0.0333, 0.0328, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0117, 0.0120, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0064, 0.0047, 0.0042, 0.0070], device='cuda:1')
+2023-03-03 12:53:46,104 INFO [train.py:968] (1/2) Epoch 6, batch 45500, giga_loss[loss=0.3181, simple_loss=0.3784, pruned_loss=0.1289, over 28877.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3866, pruned_loss=0.137, over 5648573.16 frames. ], libri_tot_loss[loss=0.3336, simple_loss=0.3881, pruned_loss=0.1395, over 5705170.63 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3859, pruned_loss=0.1363, over 5637354.99 frames. ], batch size: 112, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:54:21,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 12:54:34,152 INFO [train.py:968] (1/2) Epoch 6, batch 45550, giga_loss[loss=0.3346, simple_loss=0.397, pruned_loss=0.1361, over 28836.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3888, pruned_loss=0.138, over 5663442.37 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3877, pruned_loss=0.1393, over 5710233.01 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3886, pruned_loss=0.1377, over 5648709.92 frames. ], batch size: 227, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:54:44,018 INFO [optim.py:369] (1/2) 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 12:54:54,875 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:54:57,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:55:21,149 INFO [train.py:968] (1/2) Epoch 6, batch 45600, giga_loss[loss=0.3302, simple_loss=0.402, pruned_loss=0.1292, over 28871.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3912, pruned_loss=0.1398, over 5666352.73 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3874, pruned_loss=0.1391, over 5714371.26 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3913, pruned_loss=0.1397, over 5650098.04 frames. ], batch size: 174, lr: 5.12e-03, grad_scale: 8.0
+2023-03-03 12:55:24,971 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:56:13,177 INFO [train.py:968] (1/2) Epoch 6, batch 45650, giga_loss[loss=0.2794, simple_loss=0.3446, pruned_loss=0.1071, over 28911.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3931, pruned_loss=0.142, over 5653056.11 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.388, pruned_loss=0.1395, over 5707947.64 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3927, pruned_loss=0.1417, over 5644594.69 frames. ], batch size: 112, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:56:23,449 INFO [optim.py:369] (1/2) 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 12:56:32,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=273381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:57:02,038 INFO [train.py:968] (1/2) Epoch 6, batch 45700, giga_loss[loss=0.3308, simple_loss=0.3957, pruned_loss=0.133, over 28939.00 frames. ], tot_loss[loss=0.3389, simple_loss=0.3941, pruned_loss=0.1419, over 5651585.13 frames. ], libri_tot_loss[loss=0.3329, simple_loss=0.3876, pruned_loss=0.1391, over 5704152.92 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3943, pruned_loss=0.142, over 5647238.40 frames. ], batch size: 213, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:57:08,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4515, 1.9709, 1.4028, 1.6602], device='cuda:1'), covar=tensor([0.0729, 0.0265, 0.0325, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0118, 0.0121, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 12:57:53,570 INFO [train.py:968] (1/2) Epoch 6, batch 45750, giga_loss[loss=0.3121, simple_loss=0.3841, pruned_loss=0.12, over 28558.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3936, pruned_loss=0.1403, over 5658762.05 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3869, pruned_loss=0.1387, over 5709908.30 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3946, pruned_loss=0.1408, over 5648720.09 frames. ], batch size: 336, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:58:03,654 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 6, batch 45800, giga_loss[loss=0.3074, simple_loss=0.3714, pruned_loss=0.1217, over 28862.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3938, pruned_loss=0.1413, over 5624140.11 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3878, pruned_loss=0.1394, over 5681903.88 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.394, pruned_loss=0.1411, over 5641379.44 frames. ], batch size: 199, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:58:52,274 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=273524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:58:56,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=273527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:59:20,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1804, 2.0511, 1.5132, 1.6897], device='cuda:1'), covar=tensor([0.0731, 0.0745, 0.1023, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0455, 0.0505, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 12:59:23,921 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=273556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 12:59:27,241 INFO [train.py:968] (1/2) Epoch 6, batch 45850, giga_loss[loss=0.3037, simple_loss=0.363, pruned_loss=0.1222, over 28870.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3926, pruned_loss=0.1411, over 5585906.65 frames. ], libri_tot_loss[loss=0.3338, simple_loss=0.3882, pruned_loss=0.1398, over 5630711.38 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3925, pruned_loss=0.1406, over 5646070.50 frames. ], batch size: 112, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 12:59:40,215 INFO [optim.py:369] (1/2) 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 12:59:44,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9918, 1.1536, 3.6495, 3.0201], device='cuda:1'), covar=tensor([0.1718, 0.2433, 0.0433, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0598, 0.0549, 0.0793, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:00:24,067 INFO [train.py:968] (1/2) Epoch 6, batch 45900, libri_loss[loss=0.3685, simple_loss=0.4073, pruned_loss=0.1649, over 18943.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.394, pruned_loss=0.1436, over 5556293.05 frames. ], libri_tot_loss[loss=0.3346, simple_loss=0.3887, pruned_loss=0.1402, over 5596672.23 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3935, pruned_loss=0.1428, over 5634686.41 frames. ], batch size: 186, lr: 5.12e-03, grad_scale: 4.0
+2023-03-03 13:01:03,756 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-03 13:01:50,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1445, 1.3889, 1.1823, 1.0017], device='cuda:1'), covar=tensor([0.2206, 0.2126, 0.2180, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.0892, 0.1024, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:01:59,071 INFO [optim.py:369] (1/2) 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 13:02:19,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1633, 2.0427, 1.5047, 1.3938], device='cuda:1'), covar=tensor([0.0888, 0.0287, 0.0313, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0118, 0.0122, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0065, 0.0047, 0.0043, 0.0071], device='cuda:1')
+2023-03-03 13:02:22,723 INFO [train.py:968] (1/2) Epoch 7, batch 50, giga_loss[loss=0.3281, simple_loss=0.4081, pruned_loss=0.1241, over 28453.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3924, pruned_loss=0.1258, over 1262166.29 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3565, pruned_loss=0.1039, over 139927.51 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3966, pruned_loss=0.1283, over 1150694.68 frames. ], batch size: 71, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:03:12,016 INFO [train.py:968] (1/2) Epoch 7, batch 100, giga_loss[loss=0.263, simple_loss=0.3324, pruned_loss=0.09682, over 28479.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3815, pruned_loss=0.1203, over 2238558.45 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3521, pruned_loss=0.1005, over 254920.16 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3848, pruned_loss=0.1224, over 2075987.45 frames. ], batch size: 71, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:03:32,065 INFO [optim.py:369] (1/2) 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 13:03:56,684 INFO [train.py:968] (1/2) Epoch 7, batch 150, giga_loss[loss=0.2741, simple_loss=0.3486, pruned_loss=0.09981, over 28640.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3646, pruned_loss=0.1112, over 3008663.42 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3476, pruned_loss=0.09775, over 422354.50 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3673, pruned_loss=0.1132, over 2790548.87 frames. ], batch size: 242, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:03:58,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=273802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:04:21,172 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 200, giga_loss[loss=0.2161, simple_loss=0.2917, pruned_loss=0.07029, over 28842.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3508, pruned_loss=0.1044, over 3610088.11 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3497, pruned_loss=0.09913, over 557614.55 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3518, pruned_loss=0.1055, over 3378558.25 frames. ], batch size: 66, lr: 4.80e-03, grad_scale: 2.0
+2023-03-03 13:04:56,428 INFO [optim.py:369] (1/2) 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 13:05:20,593 INFO [train.py:968] (1/2) Epoch 7, batch 250, giga_loss[loss=0.2392, simple_loss=0.3107, pruned_loss=0.08385, over 29016.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3393, pruned_loss=0.09819, over 4071961.12 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3498, pruned_loss=0.09857, over 689356.29 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3391, pruned_loss=0.09874, over 3841803.54 frames. ], batch size: 128, lr: 4.79e-03, grad_scale: 2.0
+2023-03-03 13:05:21,306 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-03 13:05:40,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2079, 1.2978, 3.8863, 3.0477], device='cuda:1'), covar=tensor([0.2029, 0.2767, 0.0619, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0594, 0.0547, 0.0787, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:06:02,750 INFO [train.py:968] (1/2) Epoch 7, batch 300, giga_loss[loss=0.2497, simple_loss=0.305, pruned_loss=0.09723, over 26604.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3303, pruned_loss=0.0945, over 4429567.36 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3492, pruned_loss=0.09806, over 793275.80 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3295, pruned_loss=0.09477, over 4217050.29 frames. ], batch size: 555, lr: 4.79e-03, grad_scale: 2.0
+2023-03-03 13:06:07,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8288, 4.5676, 2.0200, 1.8449], device='cuda:1'), covar=tensor([0.0845, 0.0238, 0.0785, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0488, 0.0316, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 13:06:20,450 INFO [optim.py:369] (1/2) 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 13:06:45,831 INFO [train.py:968] (1/2) Epoch 7, batch 350, giga_loss[loss=0.2443, simple_loss=0.3128, pruned_loss=0.08791, over 28738.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.324, pruned_loss=0.09119, over 4710305.08 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3518, pruned_loss=0.09943, over 992664.87 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3215, pruned_loss=0.09065, over 4491531.76 frames. ], batch size: 262, lr: 4.79e-03, grad_scale: 2.0
+2023-03-03 13:07:23,928 INFO [train.py:968] (1/2) Epoch 7, batch 400, giga_loss[loss=0.229, simple_loss=0.2983, pruned_loss=0.07982, over 29007.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3193, pruned_loss=0.08845, over 4937302.94 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3513, pruned_loss=0.09862, over 1138451.37 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3162, pruned_loss=0.08776, over 4732709.73 frames. ], batch size: 145, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:07:38,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9230, 3.7205, 3.5098, 1.8442], device='cuda:1'), covar=tensor([0.0674, 0.0849, 0.0970, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0871, 0.0797, 0.0612], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:07:46,546 INFO [optim.py:369] (1/2) 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 13:07:55,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-03 13:08:05,953 INFO [train.py:968] (1/2) Epoch 7, batch 450, giga_loss[loss=0.2286, simple_loss=0.2988, pruned_loss=0.07914, over 28847.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.318, pruned_loss=0.08766, over 5111823.40 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3537, pruned_loss=0.09994, over 1325014.89 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3136, pruned_loss=0.08632, over 4915080.49 frames. ], batch size: 112, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:08:27,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2785, 2.1796, 1.6034, 1.8622], device='cuda:1'), covar=tensor([0.0661, 0.0676, 0.0923, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0442, 0.0495, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:08:47,769 INFO [train.py:968] (1/2) Epoch 7, batch 500, giga_loss[loss=0.2296, simple_loss=0.2994, pruned_loss=0.07987, over 29075.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3166, pruned_loss=0.08742, over 5239237.85 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3534, pruned_loss=0.09999, over 1503860.47 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3116, pruned_loss=0.0858, over 5054557.39 frames. ], batch size: 128, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:09:09,241 INFO [optim.py:369] (1/2) 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 13:09:12,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:09:33,476 INFO [train.py:968] (1/2) Epoch 7, batch 550, giga_loss[loss=0.201, simple_loss=0.2855, pruned_loss=0.05827, over 29050.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3147, pruned_loss=0.0869, over 5341160.31 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3546, pruned_loss=0.1008, over 1569695.15 frames. ], giga_tot_loss[loss=0.2401, simple_loss=0.3098, pruned_loss=0.0852, over 5186853.66 frames. ], batch size: 155, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:09:35,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-03 13:09:36,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=274203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:10:09,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 13:10:18,647 INFO [train.py:968] (1/2) Epoch 7, batch 600, giga_loss[loss=0.2264, simple_loss=0.2967, pruned_loss=0.07808, over 28686.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3119, pruned_loss=0.08532, over 5418180.70 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3558, pruned_loss=0.1013, over 1634056.89 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3071, pruned_loss=0.08361, over 5289471.93 frames. ], batch size: 242, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:10:25,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7732, 3.5899, 3.3654, 1.6908], device='cuda:1'), covar=tensor([0.0644, 0.0777, 0.0775, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0867, 0.0788, 0.0606], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:10:42,847 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 650, giga_loss[loss=0.244, simple_loss=0.3064, pruned_loss=0.09074, over 28842.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3097, pruned_loss=0.08442, over 5484093.24 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3575, pruned_loss=0.1027, over 1718622.85 frames. ], giga_tot_loss[loss=0.2345, simple_loss=0.3045, pruned_loss=0.0823, over 5373872.34 frames. ], batch size: 92, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:11:26,838 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:31,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:50,833 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:11:55,507 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 700, giga_loss[loss=0.2207, simple_loss=0.2875, pruned_loss=0.0769, over 28989.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3065, pruned_loss=0.08273, over 5533393.11 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3557, pruned_loss=0.102, over 1821388.62 frames. ], giga_tot_loss[loss=0.2316, simple_loss=0.3017, pruned_loss=0.08078, over 5437565.45 frames. ], batch size: 227, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:11:57,848 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:12:08,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6685, 1.6714, 1.3022, 1.2868], device='cuda:1'), covar=tensor([0.0695, 0.0576, 0.0975, 0.0994], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0447, 0.0500, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:12:14,734 INFO [optim.py:369] (1/2) 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 13:12:21,217 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:12:36,172 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:12:40,643 INFO [train.py:968] (1/2) Epoch 7, batch 750, giga_loss[loss=0.2215, simple_loss=0.2885, pruned_loss=0.07723, over 28761.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3035, pruned_loss=0.08154, over 5564403.60 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3555, pruned_loss=0.1024, over 1862215.98 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.2993, pruned_loss=0.07973, over 5485973.87 frames. ], batch size: 262, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:13:26,505 INFO [train.py:968] (1/2) Epoch 7, batch 800, libri_loss[loss=0.2592, simple_loss=0.3304, pruned_loss=0.09399, over 29342.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3029, pruned_loss=0.08161, over 5597022.16 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3569, pruned_loss=0.1034, over 1981611.95 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.2976, pruned_loss=0.07924, over 5525723.21 frames. ], batch size: 71, lr: 4.79e-03, grad_scale: 8.0
+2023-03-03 13:13:49,158 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 850, giga_loss[loss=0.2814, simple_loss=0.3503, pruned_loss=0.1063, over 28767.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3124, pruned_loss=0.08728, over 5607548.45 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3574, pruned_loss=0.104, over 2048842.46 frames. ], giga_tot_loss[loss=0.2385, simple_loss=0.3072, pruned_loss=0.08493, over 5555473.36 frames. ], batch size: 119, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:14:36,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 13:14:38,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:15:00,840 INFO [train.py:968] (1/2) Epoch 7, batch 900, giga_loss[loss=0.2881, simple_loss=0.3629, pruned_loss=0.1067, over 28773.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3258, pruned_loss=0.09398, over 5632226.47 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3569, pruned_loss=0.1034, over 2200430.00 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3205, pruned_loss=0.09186, over 5579499.35 frames. ], batch size: 284, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:15:23,209 INFO [optim.py:369] (1/2) 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 13:15:43,995 INFO [train.py:968] (1/2) Epoch 7, batch 950, giga_loss[loss=0.298, simple_loss=0.3721, pruned_loss=0.112, over 28601.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3374, pruned_loss=0.1002, over 5643254.50 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3563, pruned_loss=0.1032, over 2274179.83 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.333, pruned_loss=0.09853, over 5596394.13 frames. ], batch size: 60, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:16:25,439 INFO [train.py:968] (1/2) Epoch 7, batch 1000, giga_loss[loss=0.3277, simple_loss=0.383, pruned_loss=0.1362, over 26602.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3455, pruned_loss=0.1034, over 5660347.21 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3579, pruned_loss=0.1041, over 2381813.31 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3411, pruned_loss=0.1017, over 5615899.96 frames. ], batch size: 555, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:16:38,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:16:46,540 INFO [optim.py:369] (1/2) 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 13:17:04,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:17:07,102 INFO [train.py:968] (1/2) Epoch 7, batch 1050, giga_loss[loss=0.2689, simple_loss=0.3495, pruned_loss=0.09416, over 28866.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3496, pruned_loss=0.1039, over 5673707.52 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3584, pruned_loss=0.1042, over 2468745.59 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3456, pruned_loss=0.1025, over 5632900.37 frames. ], batch size: 174, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:17:56,786 INFO [train.py:968] (1/2) Epoch 7, batch 1100, giga_loss[loss=0.3537, simple_loss=0.4129, pruned_loss=0.1473, over 28798.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3518, pruned_loss=0.1045, over 5667889.36 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3584, pruned_loss=0.1042, over 2468745.59 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3488, pruned_loss=0.1034, over 5636128.64 frames. ], batch size: 199, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:18:12,572 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:18:39,087 INFO [train.py:968] (1/2) Epoch 7, batch 1150, giga_loss[loss=0.2849, simple_loss=0.3611, pruned_loss=0.1043, over 28942.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3549, pruned_loss=0.1069, over 5661529.78 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3588, pruned_loss=0.1044, over 2526426.59 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3523, pruned_loss=0.106, over 5640841.69 frames. ], batch size: 227, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:19:00,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4187, 1.6172, 1.2807, 1.7591], device='cuda:1'), covar=tensor([0.2229, 0.2156, 0.2182, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.1167, 0.0893, 0.1029, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:19:24,750 INFO [train.py:968] (1/2) Epoch 7, batch 1200, giga_loss[loss=0.266, simple_loss=0.343, pruned_loss=0.09449, over 29010.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3578, pruned_loss=0.109, over 5672034.46 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3585, pruned_loss=0.1041, over 2593583.16 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3558, pruned_loss=0.1085, over 5651185.25 frames. ], batch size: 164, lr: 4.79e-03, grad_scale: 8.0
+2023-03-03 13:19:46,182 INFO [optim.py:369] (1/2) 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 13:20:10,796 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 1250, giga_loss[loss=0.3177, simple_loss=0.3599, pruned_loss=0.1377, over 23737.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3613, pruned_loss=0.111, over 5671910.75 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.359, pruned_loss=0.1042, over 2622480.09 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3595, pruned_loss=0.1106, over 5656928.44 frames. ], batch size: 705, lr: 4.79e-03, grad_scale: 8.0
+2023-03-03 13:20:20,039 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=274912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:22,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=274915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:39,863 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=274933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:40,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8821, 1.2381, 4.3754, 3.2548], device='cuda:1'), covar=tensor([0.1856, 0.2429, 0.0288, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0588, 0.0541, 0.0769, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:20:49,668 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=274944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:20:53,578 INFO [train.py:968] (1/2) Epoch 7, batch 1300, giga_loss[loss=0.3865, simple_loss=0.4211, pruned_loss=0.176, over 26573.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3647, pruned_loss=0.112, over 5685969.17 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3591, pruned_loss=0.1039, over 2764591.03 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3634, pruned_loss=0.1121, over 5667190.73 frames. ], batch size: 555, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:21:12,732 INFO [optim.py:369] (1/2) 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 13:21:19,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8547, 1.8184, 1.3566, 1.4332], device='cuda:1'), covar=tensor([0.0701, 0.0634, 0.0992, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0448, 0.0504, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:21:24,721 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4357, 1.6130, 1.3400, 1.5555], device='cuda:1'), covar=tensor([0.2151, 0.2090, 0.2113, 0.2032], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.0900, 0.1028, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:21:33,366 INFO [train.py:968] (1/2) Epoch 7, batch 1350, giga_loss[loss=0.3122, simple_loss=0.3604, pruned_loss=0.132, over 23519.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3658, pruned_loss=0.112, over 5684309.92 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3587, pruned_loss=0.1036, over 2825822.19 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.365, pruned_loss=0.1123, over 5666511.38 frames. ], batch size: 705, lr: 4.79e-03, grad_scale: 4.0
+2023-03-03 13:22:08,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:12,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:14,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:18,444 INFO [train.py:968] (1/2) Epoch 7, batch 1400, giga_loss[loss=0.2853, simple_loss=0.3619, pruned_loss=0.1043, over 28255.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3671, pruned_loss=0.1117, over 5692673.07 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3587, pruned_loss=0.1034, over 2884182.25 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3666, pruned_loss=0.1123, over 5677206.55 frames. ], batch size: 368, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:22:22,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-03 13:22:35,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:22:38,742 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:23:02,846 INFO [train.py:968] (1/2) Epoch 7, batch 1450, giga_loss[loss=0.255, simple_loss=0.3444, pruned_loss=0.08282, over 28539.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3662, pruned_loss=0.1101, over 5685181.16 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3592, pruned_loss=0.1038, over 2919871.47 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3657, pruned_loss=0.1105, over 5678197.75 frames. ], batch size: 78, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:23:26,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2266, 1.5304, 1.2606, 1.2465], device='cuda:1'), covar=tensor([0.2357, 0.2179, 0.2319, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.1166, 0.0892, 0.1024, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:23:40,779 INFO [train.py:968] (1/2) Epoch 7, batch 1500, giga_loss[loss=0.2551, simple_loss=0.3432, pruned_loss=0.08346, over 28737.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3649, pruned_loss=0.1082, over 5691749.98 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3593, pruned_loss=0.1038, over 3012865.27 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3647, pruned_loss=0.1087, over 5688408.09 frames. ], batch size: 99, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:24:01,155 INFO [optim.py:369] (1/2) 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 13:24:05,518 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:07,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:19,661 INFO [train.py:968] (1/2) Epoch 7, batch 1550, giga_loss[loss=0.2715, simple_loss=0.3505, pruned_loss=0.09623, over 28652.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3638, pruned_loss=0.1072, over 5705726.90 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3592, pruned_loss=0.1037, over 3126651.28 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3638, pruned_loss=0.1077, over 5695389.26 frames. ], batch size: 262, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:24:32,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:34,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:24:35,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:25:02,495 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 1600, giga_loss[loss=0.3838, simple_loss=0.4085, pruned_loss=0.1796, over 27636.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3668, pruned_loss=0.1111, over 5691193.20 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3599, pruned_loss=0.1041, over 3136887.02 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3666, pruned_loss=0.1113, over 5685303.55 frames. ], batch size: 472, lr: 4.78e-03, grad_scale: 8.0
+2023-03-03 13:25:27,629 INFO [optim.py:369] (1/2) 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 13:25:40,665 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:25:51,326 INFO [train.py:968] (1/2) Epoch 7, batch 1650, libri_loss[loss=0.3569, simple_loss=0.4128, pruned_loss=0.1505, over 25859.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3695, pruned_loss=0.1151, over 5699582.64 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3608, pruned_loss=0.1047, over 3201959.41 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.369, pruned_loss=0.1152, over 5693687.81 frames. ], batch size: 136, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:25:58,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:26:06,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4520, 1.5173, 1.6243, 1.5048], device='cuda:1'), covar=tensor([0.1134, 0.1386, 0.1594, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0738, 0.0647, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 13:26:34,338 INFO [train.py:968] (1/2) Epoch 7, batch 1700, libri_loss[loss=0.2442, simple_loss=0.3262, pruned_loss=0.08113, over 29590.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3705, pruned_loss=0.1171, over 5707220.51 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3611, pruned_loss=0.1052, over 3328111.29 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3704, pruned_loss=0.1174, over 5698859.44 frames. ], batch size: 74, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:26:36,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4311, 1.6088, 1.3043, 1.0707], device='cuda:1'), covar=tensor([0.1226, 0.1010, 0.0852, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1359, 0.1335, 0.1444], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 13:26:45,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7249, 1.6865, 1.8279, 1.6313], device='cuda:1'), covar=tensor([0.1067, 0.1525, 0.1403, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0737, 0.0646, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 13:26:57,060 INFO [optim.py:369] (1/2) 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 13:27:07,912 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 13:27:20,281 INFO [train.py:968] (1/2) Epoch 7, batch 1750, giga_loss[loss=0.2697, simple_loss=0.3401, pruned_loss=0.09961, over 28749.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3694, pruned_loss=0.1178, over 5699316.12 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3609, pruned_loss=0.1051, over 3366471.69 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3695, pruned_loss=0.1183, over 5690215.06 frames. ], batch size: 284, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:27:22,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1515, 1.3665, 1.1630, 1.0776], device='cuda:1'), covar=tensor([0.1914, 0.1935, 0.1912, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.0894, 0.1028, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:28:04,468 INFO [train.py:968] (1/2) Epoch 7, batch 1800, giga_loss[loss=0.2993, simple_loss=0.364, pruned_loss=0.1173, over 28382.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3665, pruned_loss=0.1166, over 5694720.33 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3607, pruned_loss=0.1049, over 3416475.86 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3669, pruned_loss=0.1173, over 5684207.71 frames. ], batch size: 65, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:28:05,929 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:28:08,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:28:11,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3704, 1.7585, 1.7499, 1.4190], device='cuda:1'), covar=tensor([0.1618, 0.1941, 0.1239, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0726, 0.0814, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 13:28:22,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-03 13:28:26,943 INFO [optim.py:369] (1/2) 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 13:28:30,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1695, 1.5443, 1.2489, 0.5810], device='cuda:1'), covar=tensor([0.1988, 0.1160, 0.1404, 0.2788], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1353, 0.1399, 0.1183], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 13:28:33,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=275483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:28:48,210 INFO [train.py:968] (1/2) Epoch 7, batch 1850, giga_loss[loss=0.2662, simple_loss=0.3441, pruned_loss=0.09413, over 28964.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3652, pruned_loss=0.1152, over 5679318.58 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3613, pruned_loss=0.1052, over 3443004.67 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3652, pruned_loss=0.1157, over 5678532.59 frames. ], batch size: 106, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:29:34,999 INFO [train.py:968] (1/2) Epoch 7, batch 1900, giga_loss[loss=0.2597, simple_loss=0.3385, pruned_loss=0.09043, over 28728.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3631, pruned_loss=0.113, over 5689190.63 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3614, pruned_loss=0.105, over 3515358.76 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3632, pruned_loss=0.1137, over 5683062.49 frames. ], batch size: 92, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:29:35,967 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:29:36,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4168, 1.4863, 1.2644, 1.6135], device='cuda:1'), covar=tensor([0.2219, 0.2131, 0.2246, 0.2029], device='cuda:1'), in_proj_covar=tensor([0.1164, 0.0887, 0.1021, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 13:30:01,536 INFO [optim.py:369] (1/2) 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 13:30:20,301 INFO [train.py:968] (1/2) Epoch 7, batch 1950, giga_loss[loss=0.2683, simple_loss=0.3379, pruned_loss=0.09936, over 28870.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3583, pruned_loss=0.1095, over 5690680.05 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3614, pruned_loss=0.1052, over 3595360.23 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3582, pruned_loss=0.1102, over 5680830.67 frames. ], batch size: 199, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:31:05,873 INFO [train.py:968] (1/2) Epoch 7, batch 2000, giga_loss[loss=0.2345, simple_loss=0.3103, pruned_loss=0.07936, over 28771.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.352, pruned_loss=0.1061, over 5677575.38 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.361, pruned_loss=0.105, over 3663919.75 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.352, pruned_loss=0.1068, over 5672736.42 frames. ], batch size: 99, lr: 4.78e-03, grad_scale: 8.0
+2023-03-03 13:31:18,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:31:31,314 INFO [optim.py:369] (1/2) 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 13:31:49,578 INFO [train.py:968] (1/2) Epoch 7, batch 2050, giga_loss[loss=0.2527, simple_loss=0.3261, pruned_loss=0.08965, over 29056.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3464, pruned_loss=0.1027, over 5682340.75 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3605, pruned_loss=0.105, over 3741492.45 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3463, pruned_loss=0.1033, over 5670626.14 frames. ], batch size: 136, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:31:49,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:32:27,241 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275740.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 13:32:36,408 INFO [train.py:968] (1/2) Epoch 7, batch 2100, giga_loss[loss=0.2655, simple_loss=0.3369, pruned_loss=0.09701, over 28567.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3443, pruned_loss=0.1022, over 5655324.51 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3611, pruned_loss=0.1055, over 3793431.54 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3435, pruned_loss=0.1023, over 5650574.25 frames. ], batch size: 307, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:33:00,025 INFO [optim.py:369] (1/2) 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 13:33:00,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=275778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:33:18,614 INFO [train.py:968] (1/2) Epoch 7, batch 2150, giga_loss[loss=0.2691, simple_loss=0.344, pruned_loss=0.0971, over 27879.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3465, pruned_loss=0.1027, over 5666953.22 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3621, pruned_loss=0.1062, over 3859495.90 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3447, pruned_loss=0.1023, over 5661613.70 frames. ], batch size: 412, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:33:22,783 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=275805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:33:25,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=275808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:33:48,099 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 2200, giga_loss[loss=0.3086, simple_loss=0.3558, pruned_loss=0.1307, over 28865.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3461, pruned_loss=0.1024, over 5683195.77 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3622, pruned_loss=0.1061, over 3898564.20 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3443, pruned_loss=0.1021, over 5676248.86 frames. ], batch size: 92, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:34:25,335 INFO [optim.py:369] (1/2) 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 13:34:39,832 INFO [train.py:968] (1/2) Epoch 7, batch 2250, giga_loss[loss=0.2714, simple_loss=0.3446, pruned_loss=0.09916, over 28257.00 frames. ], tot_loss[loss=0.273, simple_loss=0.344, pruned_loss=0.101, over 5690182.50 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3631, pruned_loss=0.1062, over 3955490.02 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3414, pruned_loss=0.1005, over 5680757.88 frames. ], batch size: 368, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:35:01,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8797, 1.1140, 3.4280, 2.9951], device='cuda:1'), covar=tensor([0.1647, 0.2436, 0.0371, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.0578, 0.0534, 0.0756, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:35:01,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=275926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:35:25,082 INFO [train.py:968] (1/2) Epoch 7, batch 2300, giga_loss[loss=0.2492, simple_loss=0.3142, pruned_loss=0.09215, over 29121.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.341, pruned_loss=0.09984, over 5699746.53 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3631, pruned_loss=0.1062, over 3965306.57 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.339, pruned_loss=0.09944, over 5691422.34 frames. ], batch size: 113, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:35:47,722 INFO [optim.py:369] (1/2) 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 13:36:06,912 INFO [train.py:968] (1/2) Epoch 7, batch 2350, giga_loss[loss=0.2163, simple_loss=0.2976, pruned_loss=0.06754, over 28942.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3377, pruned_loss=0.09792, over 5702813.57 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3628, pruned_loss=0.1058, over 4013307.03 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3357, pruned_loss=0.09766, over 5691892.66 frames. ], batch size: 174, lr: 4.78e-03, grad_scale: 2.0
+2023-03-03 13:36:49,952 INFO [train.py:968] (1/2) Epoch 7, batch 2400, giga_loss[loss=0.252, simple_loss=0.3198, pruned_loss=0.09205, over 28874.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3344, pruned_loss=0.09621, over 5701807.39 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3627, pruned_loss=0.1056, over 4057397.03 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3323, pruned_loss=0.09591, over 5690964.26 frames. ], batch size: 112, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:37:05,812 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:08,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:10,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:12,491 INFO [optim.py:369] (1/2) 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 13:37:31,373 INFO [train.py:968] (1/2) Epoch 7, batch 2450, giga_loss[loss=0.2235, simple_loss=0.3036, pruned_loss=0.07173, over 29002.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3322, pruned_loss=0.09529, over 5709937.06 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3633, pruned_loss=0.106, over 4075869.66 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3299, pruned_loss=0.09478, over 5699669.57 frames. ], batch size: 164, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:37:32,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:37:42,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276115.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 13:38:09,918 INFO [train.py:968] (1/2) Epoch 7, batch 2500, giga_loss[loss=0.2478, simple_loss=0.3155, pruned_loss=0.09006, over 28960.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3307, pruned_loss=0.09426, over 5719618.89 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3633, pruned_loss=0.1056, over 4147635.94 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3277, pruned_loss=0.09367, over 5704747.68 frames. ], batch size: 227, lr: 4.78e-03, grad_scale: 4.0
+2023-03-03 13:38:13,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:38:31,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-03 13:38:32,867 INFO [optim.py:369] (1/2) 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 13:38:35,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5669, 1.7510, 1.3710, 1.1300], device='cuda:1'), covar=tensor([0.1625, 0.1065, 0.0894, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.1503, 0.1321, 0.1305, 0.1409], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 13:38:50,692 INFO [train.py:968] (1/2) Epoch 7, batch 2550, libri_loss[loss=0.2979, simple_loss=0.383, pruned_loss=0.1064, over 29526.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3286, pruned_loss=0.09277, over 5718624.60 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3639, pruned_loss=0.1058, over 4172630.85 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3252, pruned_loss=0.09196, over 5712706.98 frames. ], batch size: 83, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:39:05,952 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:08,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:12,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9979, 1.1678, 3.2140, 3.1019], device='cuda:1'), covar=tensor([0.1963, 0.2821, 0.0736, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0583, 0.0538, 0.0763, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:39:31,817 INFO [train.py:968] (1/2) Epoch 7, batch 2600, giga_loss[loss=0.2703, simple_loss=0.3401, pruned_loss=0.1002, over 28285.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.328, pruned_loss=0.09248, over 5698531.40 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3652, pruned_loss=0.1063, over 4190655.56 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3239, pruned_loss=0.09123, over 5711685.84 frames. ], batch size: 368, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:39:31,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:34,322 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:39:37,707 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276258.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 13:39:40,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276261.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 13:39:53,143 INFO [optim.py:369] (1/2) 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 13:40:02,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276290.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 13:40:06,585 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:09,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:09,335 INFO [train.py:968] (1/2) Epoch 7, batch 2650, giga_loss[loss=0.2391, simple_loss=0.3139, pruned_loss=0.08214, over 28744.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3277, pruned_loss=0.09202, over 5711285.89 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3661, pruned_loss=0.1064, over 4240194.66 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3226, pruned_loss=0.09055, over 5717497.88 frames. ], batch size: 284, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:40:14,823 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 13:40:30,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5190, 1.8957, 1.5562, 1.4887], device='cuda:1'), covar=tensor([0.0729, 0.0270, 0.0290, 0.0779], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0121, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 13:40:33,133 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:47,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2527, 2.4644, 1.2762, 1.3067], device='cuda:1'), covar=tensor([0.0865, 0.0371, 0.0835, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0479, 0.0313, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 13:40:49,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:40:53,778 INFO [train.py:968] (1/2) Epoch 7, batch 2700, giga_loss[loss=0.2944, simple_loss=0.3598, pruned_loss=0.1145, over 28837.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.332, pruned_loss=0.09525, over 5714882.53 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3664, pruned_loss=0.1066, over 4264798.02 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3273, pruned_loss=0.09377, over 5717499.53 frames. ], batch size: 186, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:40:59,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5503, 3.3874, 1.6348, 1.5299], device='cuda:1'), covar=tensor([0.0840, 0.0309, 0.0783, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0478, 0.0313, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 13:41:17,417 INFO [optim.py:369] (1/2) 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 13:41:20,496 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-03 13:41:36,630 INFO [train.py:968] (1/2) Epoch 7, batch 2750, giga_loss[loss=0.3039, simple_loss=0.3704, pruned_loss=0.1187, over 29019.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3384, pruned_loss=0.0991, over 5714656.51 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3668, pruned_loss=0.1065, over 4313324.09 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3335, pruned_loss=0.09773, over 5711708.69 frames. ], batch size: 164, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:41:46,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6293, 1.5644, 1.6446, 1.5686], device='cuda:1'), covar=tensor([0.1324, 0.1868, 0.1844, 0.1621], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0738, 0.0651, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 13:42:19,717 INFO [train.py:968] (1/2) Epoch 7, batch 2800, giga_loss[loss=0.3102, simple_loss=0.3782, pruned_loss=0.1211, over 28973.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3454, pruned_loss=0.1033, over 5718999.73 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3666, pruned_loss=0.1063, over 4374906.24 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3407, pruned_loss=0.1021, over 5710637.03 frames. ], batch size: 145, lr: 4.77e-03, grad_scale: 8.0
+2023-03-03 13:42:42,834 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 2850, giga_loss[loss=0.2997, simple_loss=0.3745, pruned_loss=0.1124, over 28669.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3524, pruned_loss=0.1079, over 5708712.58 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3661, pruned_loss=0.106, over 4442229.89 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3481, pruned_loss=0.1071, over 5694135.66 frames. ], batch size: 262, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:43:48,311 INFO [train.py:968] (1/2) Epoch 7, batch 2900, libri_loss[loss=0.2819, simple_loss=0.3651, pruned_loss=0.09933, over 29312.00 frames. ], tot_loss[loss=0.287, simple_loss=0.356, pruned_loss=0.109, over 5717895.59 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3657, pruned_loss=0.1059, over 4478409.34 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3526, pruned_loss=0.1086, over 5701965.16 frames. ], batch size: 94, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:44:15,615 INFO [optim.py:369] (1/2) 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 13:44:27,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-03 13:44:34,613 INFO [train.py:968] (1/2) Epoch 7, batch 2950, giga_loss[loss=0.269, simple_loss=0.3461, pruned_loss=0.09594, over 28942.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3616, pruned_loss=0.1118, over 5712631.21 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3659, pruned_loss=0.1059, over 4499575.21 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3588, pruned_loss=0.1115, over 5697814.19 frames. ], batch size: 112, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:44:57,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:45:19,835 INFO [train.py:968] (1/2) Epoch 7, batch 3000, giga_loss[loss=0.3515, simple_loss=0.4058, pruned_loss=0.1486, over 28636.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3681, pruned_loss=0.1169, over 5697841.59 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3651, pruned_loss=0.1056, over 4558432.00 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3663, pruned_loss=0.1173, over 5680611.33 frames. ], batch size: 307, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:45:19,835 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 13:45:28,383 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 13:45:55,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 13:45:55,759 INFO [optim.py:369] (1/2) 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 13:46:12,264 INFO [train.py:968] (1/2) Epoch 7, batch 3050, giga_loss[loss=0.2396, simple_loss=0.3244, pruned_loss=0.07738, over 28896.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3666, pruned_loss=0.1157, over 5700019.36 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3656, pruned_loss=0.1063, over 4578767.14 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3649, pruned_loss=0.1157, over 5683270.05 frames. ], batch size: 174, lr: 4.77e-03, grad_scale: 2.0
+2023-03-03 13:46:21,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=276710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:46:30,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=276720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:46:37,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1769, 1.2323, 4.4660, 3.3410], device='cuda:1'), covar=tensor([0.1754, 0.2541, 0.0321, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0537, 0.0758, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:46:54,299 INFO [train.py:968] (1/2) Epoch 7, batch 3100, giga_loss[loss=0.313, simple_loss=0.38, pruned_loss=0.123, over 28895.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3627, pruned_loss=0.1127, over 5702970.17 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3655, pruned_loss=0.1063, over 4602307.62 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3614, pruned_loss=0.1128, over 5688455.74 frames. ], batch size: 145, lr: 4.77e-03, grad_scale: 2.0
+2023-03-03 13:47:13,914 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:47:16,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:47:20,594 INFO [optim.py:369] (1/2) 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 13:47:40,773 INFO [train.py:968] (1/2) Epoch 7, batch 3150, giga_loss[loss=0.2639, simple_loss=0.3379, pruned_loss=0.09493, over 28926.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3614, pruned_loss=0.1115, over 5709960.84 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3652, pruned_loss=0.1062, over 4615183.72 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3605, pruned_loss=0.1116, over 5696856.94 frames. ], batch size: 106, lr: 4.77e-03, grad_scale: 2.0
+2023-03-03 13:47:42,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:47:43,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6448, 4.4883, 4.2622, 1.8678], device='cuda:1'), covar=tensor([0.0431, 0.0590, 0.0635, 0.2052], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0851, 0.0773, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:48:23,571 INFO [train.py:968] (1/2) Epoch 7, batch 3200, libri_loss[loss=0.3008, simple_loss=0.3756, pruned_loss=0.113, over 29662.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3609, pruned_loss=0.1106, over 5712126.21 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3646, pruned_loss=0.106, over 4650739.64 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3605, pruned_loss=0.111, over 5698766.18 frames. ], batch size: 91, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:48:34,086 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=276863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:48:36,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=276866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:48:49,080 INFO [optim.py:369] (1/2) 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 13:49:03,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=276895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:49:06,733 INFO [train.py:968] (1/2) Epoch 7, batch 3250, giga_loss[loss=0.3196, simple_loss=0.3807, pruned_loss=0.1293, over 28782.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3633, pruned_loss=0.1114, over 5715363.11 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3645, pruned_loss=0.106, over 4669260.01 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.363, pruned_loss=0.1118, over 5702604.58 frames. ], batch size: 99, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:49:51,751 INFO [train.py:968] (1/2) Epoch 7, batch 3300, giga_loss[loss=0.3103, simple_loss=0.3775, pruned_loss=0.1215, over 28638.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3654, pruned_loss=0.1129, over 5711189.52 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3646, pruned_loss=0.1059, over 4686066.60 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3651, pruned_loss=0.1133, over 5700121.59 frames. ], batch size: 336, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:50:17,390 INFO [optim.py:369] (1/2) 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 13:50:34,125 INFO [train.py:968] (1/2) Epoch 7, batch 3350, giga_loss[loss=0.3057, simple_loss=0.3721, pruned_loss=0.1196, over 29052.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3665, pruned_loss=0.1141, over 5706169.68 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3646, pruned_loss=0.1059, over 4734246.49 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3663, pruned_loss=0.1148, over 5699417.24 frames. ], batch size: 128, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:50:37,757 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:50:47,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1931, 1.4448, 1.2366, 0.9468], device='cuda:1'), covar=tensor([0.2030, 0.1956, 0.2003, 0.1833], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.0894, 0.1032, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:50:51,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6011, 2.2712, 1.6720, 0.7802], device='cuda:1'), covar=tensor([0.2288, 0.1374, 0.2241, 0.2808], device='cuda:1'), in_proj_covar=tensor([0.1421, 0.1337, 0.1394, 0.1172], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 13:51:16,989 INFO [train.py:968] (1/2) Epoch 7, batch 3400, giga_loss[loss=0.273, simple_loss=0.3528, pruned_loss=0.09662, over 28851.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3686, pruned_loss=0.116, over 5712370.73 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3644, pruned_loss=0.1058, over 4745971.60 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3686, pruned_loss=0.1168, over 5705347.41 frames. ], batch size: 174, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:51:21,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-03 13:51:28,038 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:51:39,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7352, 4.5729, 4.3054, 1.9500], device='cuda:1'), covar=tensor([0.0441, 0.0559, 0.0678, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0846, 0.0772, 0.0611], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:51:43,191 INFO [optim.py:369] (1/2) 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 13:51:48,944 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 3450, giga_loss[loss=0.2713, simple_loss=0.3445, pruned_loss=0.09899, over 28296.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3682, pruned_loss=0.1155, over 5719365.54 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3646, pruned_loss=0.1057, over 4776472.95 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3682, pruned_loss=0.1165, over 5711945.15 frames. ], batch size: 77, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:52:21,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2341, 1.2098, 1.1105, 0.9015], device='cuda:1'), covar=tensor([0.0676, 0.0508, 0.0964, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0438, 0.0496, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 13:52:23,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3910, 1.8718, 1.6868, 1.3315], device='cuda:1'), covar=tensor([0.1603, 0.2076, 0.1276, 0.1424], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0729, 0.0816, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 13:52:42,963 INFO [train.py:968] (1/2) Epoch 7, batch 3500, giga_loss[loss=0.2903, simple_loss=0.3445, pruned_loss=0.118, over 23936.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.37, pruned_loss=0.1165, over 5718106.51 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3651, pruned_loss=0.1061, over 4804082.52 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3697, pruned_loss=0.1172, over 5708415.13 frames. ], batch size: 705, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:53:06,510 INFO [optim.py:369] (1/2) 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 13:53:10,056 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:53:25,543 INFO [train.py:968] (1/2) Epoch 7, batch 3550, giga_loss[loss=0.2998, simple_loss=0.3733, pruned_loss=0.1131, over 28456.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3697, pruned_loss=0.1148, over 5717949.44 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3654, pruned_loss=0.1062, over 4808626.11 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3693, pruned_loss=0.1153, over 5710392.55 frames. ], batch size: 78, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:53:51,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:53:52,922 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 3600, giga_loss[loss=0.2588, simple_loss=0.3373, pruned_loss=0.09015, over 28993.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3705, pruned_loss=0.1147, over 5723261.93 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3655, pruned_loss=0.1062, over 4824704.20 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3702, pruned_loss=0.1152, over 5715145.91 frames. ], batch size: 136, lr: 4.77e-03, grad_scale: 8.0
+2023-03-03 13:54:15,292 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:54:18,922 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:54:50,017 INFO [train.py:968] (1/2) Epoch 7, batch 3650, libri_loss[loss=0.3993, simple_loss=0.442, pruned_loss=0.1783, over 29514.00 frames. ], tot_loss[loss=0.297, simple_loss=0.368, pruned_loss=0.113, over 5725702.21 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3659, pruned_loss=0.1065, over 4841028.89 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3674, pruned_loss=0.1133, over 5716454.62 frames. ], batch size: 83, lr: 4.77e-03, grad_scale: 8.0
+2023-03-03 13:55:10,327 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:55:33,486 INFO [train.py:968] (1/2) Epoch 7, batch 3700, giga_loss[loss=0.2595, simple_loss=0.3454, pruned_loss=0.08678, over 28994.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3668, pruned_loss=0.1131, over 5722433.76 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3665, pruned_loss=0.1069, over 4856757.53 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3659, pruned_loss=0.1131, over 5712917.07 frames. ], batch size: 164, lr: 4.77e-03, grad_scale: 4.0
+2023-03-03 13:55:55,625 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:55:56,662 INFO [optim.py:369] (1/2) 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 13:56:10,618 INFO [train.py:968] (1/2) Epoch 7, batch 3750, giga_loss[loss=0.2884, simple_loss=0.3646, pruned_loss=0.1061, over 28833.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3638, pruned_loss=0.111, over 5728296.40 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3663, pruned_loss=0.1067, over 4872404.03 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3633, pruned_loss=0.1112, over 5718392.71 frames. ], batch size: 199, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:56:15,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:56:44,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:56:53,664 INFO [train.py:968] (1/2) Epoch 7, batch 3800, giga_loss[loss=0.267, simple_loss=0.3384, pruned_loss=0.09776, over 28600.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3626, pruned_loss=0.1102, over 5734498.75 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3657, pruned_loss=0.1065, over 4894635.89 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3625, pruned_loss=0.1105, over 5725847.39 frames. ], batch size: 85, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:57:07,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-03 13:57:19,425 INFO [optim.py:369] (1/2) 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 13:57:35,793 INFO [train.py:968] (1/2) Epoch 7, batch 3850, giga_loss[loss=0.2988, simple_loss=0.3643, pruned_loss=0.1167, over 28677.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3644, pruned_loss=0.1116, over 5725013.52 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3656, pruned_loss=0.1064, over 4905360.71 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3644, pruned_loss=0.112, over 5722716.34 frames. ], batch size: 85, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:57:54,818 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:57:56,659 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 3900, giga_loss[loss=0.2801, simple_loss=0.3574, pruned_loss=0.1014, over 28604.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.364, pruned_loss=0.1105, over 5720004.17 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3659, pruned_loss=0.1066, over 4915249.72 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3637, pruned_loss=0.1108, over 5716587.75 frames. ], batch size: 336, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:58:22,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:26,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:32,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-03 13:58:37,714 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:43,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4786, 1.6653, 1.3456, 2.1625], device='cuda:1'), covar=tensor([0.2261, 0.2106, 0.2288, 0.1822], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.0897, 0.1032, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 13:58:44,665 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:58:45,783 INFO [optim.py:369] (1/2) 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 13:58:47,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:59:01,257 INFO [train.py:968] (1/2) Epoch 7, batch 3950, giga_loss[loss=0.2613, simple_loss=0.3395, pruned_loss=0.09156, over 28434.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3634, pruned_loss=0.1098, over 5722512.11 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3664, pruned_loss=0.1068, over 4929703.15 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3628, pruned_loss=0.1098, over 5717523.24 frames. ], batch size: 60, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 13:59:11,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:59:12,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 13:59:26,717 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 4000, giga_loss[loss=0.2555, simple_loss=0.3315, pruned_loss=0.08971, over 28681.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3626, pruned_loss=0.11, over 5721604.84 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3661, pruned_loss=0.1067, over 4950379.20 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3622, pruned_loss=0.1102, over 5716238.23 frames. ], batch size: 60, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:00:09,753 INFO [optim.py:369] (1/2) 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 14:00:13,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4014, 1.5411, 1.2586, 1.5102], device='cuda:1'), covar=tensor([0.2179, 0.2100, 0.2221, 0.2008], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.0895, 0.1032, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:00:21,560 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:22,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:22,535 INFO [train.py:968] (1/2) Epoch 7, batch 4050, giga_loss[loss=0.2924, simple_loss=0.3632, pruned_loss=0.1108, over 27990.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3602, pruned_loss=0.1088, over 5717260.97 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3659, pruned_loss=0.1066, over 4972364.20 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.36, pruned_loss=0.1091, over 5710105.38 frames. ], batch size: 412, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:00:23,538 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:25,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:00:34,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6622, 1.7241, 1.8056, 1.6234], device='cuda:1'), covar=tensor([0.1227, 0.1686, 0.1640, 0.1594], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0729, 0.0648, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 14:00:49,237 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 4100, giga_loss[loss=0.2829, simple_loss=0.3508, pruned_loss=0.1075, over 28956.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3574, pruned_loss=0.1075, over 5716510.76 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3662, pruned_loss=0.1069, over 4999001.60 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3568, pruned_loss=0.1076, over 5706603.80 frames. ], batch size: 106, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:01:22,479 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:01:25,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:01:26,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:01:28,050 INFO [optim.py:369] (1/2) 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 14:01:43,651 INFO [train.py:968] (1/2) Epoch 7, batch 4150, libri_loss[loss=0.2932, simple_loss=0.3652, pruned_loss=0.1106, over 29617.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3555, pruned_loss=0.1068, over 5715441.93 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3663, pruned_loss=0.1069, over 5017266.08 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3548, pruned_loss=0.1069, over 5703807.20 frames. ], batch size: 91, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:01:48,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:15,923 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:18,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:21,872 INFO [train.py:968] (1/2) Epoch 7, batch 4200, giga_loss[loss=0.3716, simple_loss=0.4012, pruned_loss=0.171, over 28809.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3548, pruned_loss=0.1071, over 5709374.42 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3666, pruned_loss=0.1071, over 5031407.96 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3538, pruned_loss=0.1071, over 5699865.25 frames. ], batch size: 99, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:02:40,959 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=277874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:02:46,116 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 4250, giga_loss[loss=0.2513, simple_loss=0.3252, pruned_loss=0.08869, over 29005.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3524, pruned_loss=0.1061, over 5708690.63 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3663, pruned_loss=0.1068, over 5048791.78 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3515, pruned_loss=0.1062, over 5697840.61 frames. ], batch size: 164, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:03:08,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-03 14:03:24,947 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=277922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:26,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=277925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:29,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8003, 1.7644, 1.2688, 1.4601], device='cuda:1'), covar=tensor([0.0642, 0.0600, 0.0986, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0436, 0.0495, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:03:42,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:43,574 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:03:45,371 INFO [train.py:968] (1/2) Epoch 7, batch 4300, giga_loss[loss=0.2703, simple_loss=0.3268, pruned_loss=0.1069, over 28634.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3495, pruned_loss=0.1052, over 5716919.48 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3666, pruned_loss=0.1072, over 5061754.62 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3483, pruned_loss=0.1049, over 5705556.09 frames. ], batch size: 85, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:03:48,459 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=277963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:04:10,800 INFO [optim.py:369] (1/2) 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 14:04:14,345 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=277987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:04:25,311 INFO [train.py:968] (1/2) Epoch 7, batch 4350, giga_loss[loss=0.2549, simple_loss=0.3209, pruned_loss=0.09444, over 28480.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3469, pruned_loss=0.1043, over 5711974.05 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3667, pruned_loss=0.1072, over 5070406.83 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3456, pruned_loss=0.1041, over 5703934.03 frames. ], batch size: 71, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:05:01,991 INFO [train.py:968] (1/2) Epoch 7, batch 4400, giga_loss[loss=0.2819, simple_loss=0.3544, pruned_loss=0.1047, over 28255.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3468, pruned_loss=0.1041, over 5704158.66 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3671, pruned_loss=0.1074, over 5091961.36 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3447, pruned_loss=0.1036, over 5704360.23 frames. ], batch size: 368, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:05:20,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:27,245 INFO [optim.py:369] (1/2) 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 14:05:30,333 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:33,682 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:35,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:05:41,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4238, 1.6157, 1.3496, 1.4009], device='cuda:1'), covar=tensor([0.2064, 0.2019, 0.2189, 0.1849], device='cuda:1'), in_proj_covar=tensor([0.1164, 0.0884, 0.1025, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0007], device='cuda:1')
+2023-03-03 14:05:41,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4115, 1.5791, 1.3022, 1.2223], device='cuda:1'), covar=tensor([0.1568, 0.1213, 0.0972, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.1498, 0.1334, 0.1323, 0.1403], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 14:05:42,848 INFO [train.py:968] (1/2) Epoch 7, batch 4450, giga_loss[loss=0.3045, simple_loss=0.3681, pruned_loss=0.1204, over 29057.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3481, pruned_loss=0.1042, over 5708303.20 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3671, pruned_loss=0.1075, over 5112984.00 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3458, pruned_loss=0.1036, over 5706221.54 frames. ], batch size: 155, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:06:01,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:06:10,181 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:06:13,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:06:21,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-03 14:06:26,329 INFO [train.py:968] (1/2) Epoch 7, batch 4500, giga_loss[loss=0.3201, simple_loss=0.3984, pruned_loss=0.1209, over 28632.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3515, pruned_loss=0.1061, over 5702056.14 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3669, pruned_loss=0.1074, over 5123399.78 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3497, pruned_loss=0.1057, over 5699131.27 frames. ], batch size: 307, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:06:28,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2268, 1.7548, 1.2728, 0.4950], device='cuda:1'), covar=tensor([0.2259, 0.1201, 0.1920, 0.2923], device='cuda:1'), in_proj_covar=tensor([0.1428, 0.1337, 0.1393, 0.1173], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 14:06:38,829 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:07:08,972 INFO [train.py:968] (1/2) Epoch 7, batch 4550, giga_loss[loss=0.2541, simple_loss=0.3387, pruned_loss=0.08477, over 28975.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3549, pruned_loss=0.1075, over 5702182.91 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3668, pruned_loss=0.1073, over 5131519.50 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3533, pruned_loss=0.1073, over 5700442.81 frames. ], batch size: 164, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:07:14,185 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 14:07:24,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:07:28,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:07:53,860 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:07:55,076 INFO [train.py:968] (1/2) Epoch 7, batch 4600, giga_loss[loss=0.282, simple_loss=0.3564, pruned_loss=0.1038, over 28693.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3557, pruned_loss=0.1071, over 5696810.25 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.367, pruned_loss=0.1075, over 5139242.90 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3543, pruned_loss=0.1067, over 5693650.59 frames. ], batch size: 242, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:08:24,524 INFO [optim.py:369] (1/2) 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 14:08:38,765 INFO [train.py:968] (1/2) Epoch 7, batch 4650, libri_loss[loss=0.247, simple_loss=0.3196, pruned_loss=0.08721, over 29359.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3544, pruned_loss=0.1058, over 5697616.91 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3668, pruned_loss=0.1075, over 5156319.97 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3531, pruned_loss=0.1055, over 5691970.56 frames. ], batch size: 67, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:08:56,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:09:10,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:09:19,444 INFO [train.py:968] (1/2) Epoch 7, batch 4700, libri_loss[loss=0.2869, simple_loss=0.3682, pruned_loss=0.1028, over 29261.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.354, pruned_loss=0.1061, over 5697517.10 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3669, pruned_loss=0.1076, over 5172225.89 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3525, pruned_loss=0.1058, over 5695245.01 frames. ], batch size: 97, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:09:28,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:09:43,870 INFO [optim.py:369] (1/2) 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 14:09:59,151 INFO [train.py:968] (1/2) Epoch 7, batch 4750, libri_loss[loss=0.2905, simple_loss=0.3699, pruned_loss=0.1055, over 29533.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.355, pruned_loss=0.1069, over 5701249.13 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3664, pruned_loss=0.1074, over 5197412.81 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3538, pruned_loss=0.1068, over 5693005.24 frames. ], batch size: 82, lr: 4.76e-03, grad_scale: 4.0
+2023-03-03 14:10:39,545 INFO [train.py:968] (1/2) Epoch 7, batch 4800, giga_loss[loss=0.3078, simple_loss=0.3775, pruned_loss=0.1191, over 28277.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3576, pruned_loss=0.1088, over 5700323.32 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3667, pruned_loss=0.1076, over 5210070.55 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3562, pruned_loss=0.1086, over 5691517.99 frames. ], batch size: 368, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:10:45,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 14:10:51,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:10:55,966 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:10:58,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:11:07,180 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:11:09,185 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:11:20,226 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:11:22,280 INFO [train.py:968] (1/2) Epoch 7, batch 4850, libri_loss[loss=0.2404, simple_loss=0.3256, pruned_loss=0.07757, over 29577.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3612, pruned_loss=0.1108, over 5703873.68 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3667, pruned_loss=0.1075, over 5227157.65 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3599, pruned_loss=0.1107, over 5692321.02 frames. ], batch size: 76, lr: 4.76e-03, grad_scale: 8.0
+2023-03-03 14:11:34,357 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 4900, libri_loss[loss=0.3207, simple_loss=0.3887, pruned_loss=0.1263, over 29533.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3637, pruned_loss=0.1121, over 5717008.91 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3667, pruned_loss=0.1077, over 5247556.20 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3627, pruned_loss=0.112, over 5701952.86 frames. ], batch size: 83, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:12:19,050 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-03 14:12:30,544 INFO [optim.py:369] (1/2) 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 14:12:32,465 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 14:12:45,073 INFO [train.py:968] (1/2) Epoch 7, batch 4950, giga_loss[loss=0.3172, simple_loss=0.382, pruned_loss=0.1261, over 28631.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3642, pruned_loss=0.1121, over 5719345.08 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3664, pruned_loss=0.1076, over 5257554.83 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3635, pruned_loss=0.1122, over 5704959.59 frames. ], batch size: 336, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:12:47,535 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=278602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:12:50,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:13:12,268 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:13:25,613 INFO [train.py:968] (1/2) Epoch 7, batch 5000, giga_loss[loss=0.3173, simple_loss=0.3825, pruned_loss=0.1261, over 28903.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3641, pruned_loss=0.1116, over 5729121.89 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3663, pruned_loss=0.1074, over 5283355.45 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3636, pruned_loss=0.112, over 5710605.20 frames. ], batch size: 186, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:13:51,785 INFO [optim.py:369] (1/2) 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 14:14:05,546 INFO [train.py:968] (1/2) Epoch 7, batch 5050, giga_loss[loss=0.2653, simple_loss=0.3494, pruned_loss=0.09055, over 28975.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3649, pruned_loss=0.1122, over 5734874.26 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3669, pruned_loss=0.1078, over 5304292.25 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.364, pruned_loss=0.1123, over 5715035.79 frames. ], batch size: 213, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:14:18,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:14:37,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=278739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:14:39,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-03 14:14:45,880 INFO [train.py:968] (1/2) Epoch 7, batch 5100, giga_loss[loss=0.2834, simple_loss=0.3474, pruned_loss=0.1098, over 28703.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3637, pruned_loss=0.1118, over 5721376.20 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3676, pruned_loss=0.1083, over 5310680.80 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3622, pruned_loss=0.1115, over 5712028.39 frames. ], batch size: 85, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:14:59,550 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:15:27,990 INFO [train.py:968] (1/2) Epoch 7, batch 5150, giga_loss[loss=0.2771, simple_loss=0.3521, pruned_loss=0.1011, over 28890.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.359, pruned_loss=0.1088, over 5727639.34 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3664, pruned_loss=0.1077, over 5327058.04 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3588, pruned_loss=0.1092, over 5716240.40 frames. ], batch size: 227, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:16:08,404 INFO [train.py:968] (1/2) Epoch 7, batch 5200, giga_loss[loss=0.2439, simple_loss=0.3238, pruned_loss=0.08201, over 28878.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3544, pruned_loss=0.106, over 5729482.77 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3666, pruned_loss=0.1078, over 5332797.90 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3538, pruned_loss=0.1062, over 5718977.93 frames. ], batch size: 186, lr: 4.75e-03, grad_scale: 8.0
+2023-03-03 14:16:24,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3019, 1.7631, 1.2581, 0.5365], device='cuda:1'), covar=tensor([0.2207, 0.1190, 0.1836, 0.3039], device='cuda:1'), in_proj_covar=tensor([0.1423, 0.1336, 0.1390, 0.1169], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 14:16:34,091 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:16:36,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=278885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:16:46,993 INFO [train.py:968] (1/2) Epoch 7, batch 5250, giga_loss[loss=0.2564, simple_loss=0.3267, pruned_loss=0.09305, over 28642.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3542, pruned_loss=0.1057, over 5722703.00 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3665, pruned_loss=0.1079, over 5345344.85 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3534, pruned_loss=0.1057, over 5716567.76 frames. ], batch size: 92, lr: 4.75e-03, grad_scale: 8.0
+2023-03-03 14:16:50,700 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=278904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:17:00,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=278914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:17:13,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2028, 1.3273, 1.0734, 1.3524], device='cuda:1'), covar=tensor([0.0731, 0.0301, 0.0335, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 14:17:23,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6671, 2.2110, 1.9360, 1.7684], device='cuda:1'), covar=tensor([0.0689, 0.0231, 0.0266, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0065, 0.0047, 0.0042, 0.0071], device='cuda:1')
+2023-03-03 14:17:31,032 INFO [train.py:968] (1/2) Epoch 7, batch 5300, libri_loss[loss=0.3131, simple_loss=0.3917, pruned_loss=0.1173, over 29204.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3565, pruned_loss=0.1056, over 5716371.73 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3662, pruned_loss=0.1077, over 5361855.50 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3558, pruned_loss=0.1057, over 5705966.64 frames. ], batch size: 97, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:18:01,779 INFO [optim.py:369] (1/2) 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 14:18:13,719 INFO [train.py:968] (1/2) Epoch 7, batch 5350, giga_loss[loss=0.2932, simple_loss=0.3547, pruned_loss=0.1158, over 28843.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3577, pruned_loss=0.1067, over 5694265.39 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3666, pruned_loss=0.1081, over 5351958.62 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3567, pruned_loss=0.1064, over 5701132.09 frames. ], batch size: 112, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:18:52,290 INFO [train.py:968] (1/2) Epoch 7, batch 5400, giga_loss[loss=0.2617, simple_loss=0.333, pruned_loss=0.09526, over 28876.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3587, pruned_loss=0.1081, over 5699760.14 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3677, pruned_loss=0.1088, over 5367151.69 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3567, pruned_loss=0.1072, over 5705333.68 frames. ], batch size: 199, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:19:22,960 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 7, batch 5450, giga_loss[loss=0.3813, simple_loss=0.4083, pruned_loss=0.1772, over 24107.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3585, pruned_loss=0.1093, over 5696732.34 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3679, pruned_loss=0.1089, over 5377421.65 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3566, pruned_loss=0.1085, over 5697830.97 frames. ], batch size: 705, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:19:44,813 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-03 14:19:55,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5425, 1.5747, 1.4900, 1.3460], device='cuda:1'), covar=tensor([0.0912, 0.1261, 0.1439, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0721, 0.0641, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 14:20:09,365 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:20:15,517 INFO [train.py:968] (1/2) Epoch 7, batch 5500, giga_loss[loss=0.2736, simple_loss=0.3462, pruned_loss=0.1006, over 28927.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3569, pruned_loss=0.1097, over 5706383.60 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3677, pruned_loss=0.1089, over 5390391.14 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3553, pruned_loss=0.1092, over 5702958.64 frames. ], batch size: 164, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:20:15,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5800, 1.5447, 1.1434, 1.3802], device='cuda:1'), covar=tensor([0.0553, 0.0485, 0.0956, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0450, 0.0501, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:20:27,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8118, 0.9199, 0.7186, 0.7911], device='cuda:1'), covar=tensor([0.0875, 0.0996, 0.0673, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.1528, 0.1352, 0.1348, 0.1446], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 14:20:40,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3843, 2.1639, 1.5731, 0.6309], device='cuda:1'), covar=tensor([0.3489, 0.1493, 0.2382, 0.3753], device='cuda:1'), in_proj_covar=tensor([0.1430, 0.1328, 0.1394, 0.1176], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 14:20:43,199 INFO [optim.py:369] (1/2) 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 14:20:55,460 INFO [train.py:968] (1/2) Epoch 7, batch 5550, libri_loss[loss=0.2524, simple_loss=0.332, pruned_loss=0.08642, over 29577.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3563, pruned_loss=0.1103, over 5692399.89 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3685, pruned_loss=0.1094, over 5393419.74 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3539, pruned_loss=0.1095, over 5698932.79 frames. ], batch size: 74, lr: 4.75e-03, grad_scale: 2.0
+2023-03-03 14:21:26,741 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:21:29,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:21:39,472 INFO [train.py:968] (1/2) Epoch 7, batch 5600, giga_loss[loss=0.2645, simple_loss=0.339, pruned_loss=0.095, over 28920.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3547, pruned_loss=0.1091, over 5705299.02 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3684, pruned_loss=0.1094, over 5405787.41 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3526, pruned_loss=0.1085, over 5705899.81 frames. ], batch size: 227, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:21:54,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:04,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:08,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:08,426 INFO [optim.py:369] (1/2) 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 14:22:10,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:12,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7057, 2.5406, 1.7694, 1.0001], device='cuda:1'), covar=tensor([0.4287, 0.1861, 0.2360, 0.3791], device='cuda:1'), in_proj_covar=tensor([0.1438, 0.1337, 0.1399, 0.1182], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 14:22:18,615 INFO [train.py:968] (1/2) Epoch 7, batch 5650, giga_loss[loss=0.2194, simple_loss=0.2876, pruned_loss=0.07562, over 28470.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3524, pruned_loss=0.108, over 5712818.61 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3686, pruned_loss=0.1095, over 5420581.48 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3501, pruned_loss=0.1074, over 5709264.73 frames. ], batch size: 71, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:22:33,561 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:22:33,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2602, 1.3802, 1.1549, 1.1862], device='cuda:1'), covar=tensor([0.0655, 0.0435, 0.0992, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0448, 0.0499, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:22:35,233 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 14:22:59,212 INFO [train.py:968] (1/2) Epoch 7, batch 5700, libri_loss[loss=0.262, simple_loss=0.347, pruned_loss=0.08852, over 29542.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3477, pruned_loss=0.1054, over 5715730.47 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3691, pruned_loss=0.1097, over 5427228.29 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3449, pruned_loss=0.1046, over 5715704.13 frames. ], batch size: 81, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:23:29,655 INFO [optim.py:369] (1/2) 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 14:23:32,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-03 14:23:39,397 INFO [train.py:968] (1/2) Epoch 7, batch 5750, giga_loss[loss=0.2888, simple_loss=0.3572, pruned_loss=0.1102, over 28731.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3442, pruned_loss=0.1034, over 5717928.15 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.369, pruned_loss=0.1099, over 5435054.00 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3415, pruned_loss=0.1025, over 5716026.30 frames. ], batch size: 307, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:23:59,435 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=279422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:24:01,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=279425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:24:20,098 INFO [train.py:968] (1/2) Epoch 7, batch 5800, giga_loss[loss=0.2378, simple_loss=0.3137, pruned_loss=0.08093, over 29057.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3447, pruned_loss=0.1032, over 5720299.83 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.369, pruned_loss=0.1099, over 5439510.46 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3423, pruned_loss=0.1025, over 5717326.42 frames. ], batch size: 128, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:24:20,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1718, 1.6423, 1.5654, 1.2132], device='cuda:1'), covar=tensor([0.1367, 0.1918, 0.1145, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0715, 0.0807, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 14:24:23,852 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=279454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:24:48,346 INFO [optim.py:369] (1/2) 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 14:24:58,386 INFO [train.py:968] (1/2) Epoch 7, batch 5850, giga_loss[loss=0.2563, simple_loss=0.3363, pruned_loss=0.08817, over 28979.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3484, pruned_loss=0.1051, over 5723478.59 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3685, pruned_loss=0.1097, over 5453416.14 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3463, pruned_loss=0.1044, over 5716080.26 frames. ], batch size: 164, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:25:18,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:25:40,617 INFO [train.py:968] (1/2) Epoch 7, batch 5900, giga_loss[loss=0.3275, simple_loss=0.3992, pruned_loss=0.1278, over 29093.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3523, pruned_loss=0.1066, over 5724867.66 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3684, pruned_loss=0.1096, over 5466993.31 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3502, pruned_loss=0.1061, over 5714060.50 frames. ], batch size: 128, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:26:12,227 INFO [optim.py:369] (1/2) 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 14:26:23,962 INFO [train.py:968] (1/2) Epoch 7, batch 5950, giga_loss[loss=0.2828, simple_loss=0.3558, pruned_loss=0.1049, over 28876.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3559, pruned_loss=0.1083, over 5722034.35 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3685, pruned_loss=0.1096, over 5473428.41 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.354, pruned_loss=0.1079, over 5711197.29 frames. ], batch size: 199, lr: 4.75e-03, grad_scale: 4.0
+2023-03-03 14:27:07,465 INFO [train.py:968] (1/2) Epoch 7, batch 6000, libri_loss[loss=0.2958, simple_loss=0.3657, pruned_loss=0.1129, over 29550.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3588, pruned_loss=0.1099, over 5725355.06 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.369, pruned_loss=0.11, over 5491852.36 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3564, pruned_loss=0.1092, over 5709340.90 frames. ], batch size: 77, lr: 4.75e-03, grad_scale: 8.0
+2023-03-03 14:27:07,466 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 14:27:12,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7631, 3.5176, 3.4455, 1.5366], device='cuda:1'), covar=tensor([0.0750, 0.0842, 0.0805, 0.2310], device='cuda:1'), in_proj_covar=tensor([0.0913, 0.0855, 0.0775, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:27:16,122 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 14:27:49,948 INFO [optim.py:369] (1/2) 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 14:28:02,623 INFO [train.py:968] (1/2) Epoch 7, batch 6050, giga_loss[loss=0.3455, simple_loss=0.4003, pruned_loss=0.1453, over 28333.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3632, pruned_loss=0.1135, over 5719275.79 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.369, pruned_loss=0.11, over 5493532.10 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3612, pruned_loss=0.113, over 5706279.77 frames. ], batch size: 71, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:28:07,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3620, 2.5327, 1.4293, 1.4393], device='cuda:1'), covar=tensor([0.0678, 0.0351, 0.0614, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0484, 0.0314, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 14:28:51,453 INFO [train.py:968] (1/2) Epoch 7, batch 6100, giga_loss[loss=0.3419, simple_loss=0.3953, pruned_loss=0.1443, over 28939.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3708, pruned_loss=0.1207, over 5712292.81 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3695, pruned_loss=0.1104, over 5500062.59 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.12, over 5699354.47 frames. ], batch size: 145, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:29:27,221 INFO [optim.py:369] (1/2) 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 14:29:38,886 INFO [train.py:968] (1/2) Epoch 7, batch 6150, giga_loss[loss=0.3317, simple_loss=0.397, pruned_loss=0.1332, over 28735.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.378, pruned_loss=0.1263, over 5691571.52 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3701, pruned_loss=0.1108, over 5513633.59 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3761, pruned_loss=0.1259, over 5674976.15 frames. ], batch size: 284, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:30:21,902 INFO [train.py:968] (1/2) Epoch 7, batch 6200, giga_loss[loss=0.4341, simple_loss=0.4617, pruned_loss=0.2032, over 27548.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3845, pruned_loss=0.1312, over 5681396.44 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3707, pruned_loss=0.1114, over 5513176.81 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3829, pruned_loss=0.1311, over 5676169.12 frames. ], batch size: 472, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:30:59,386 INFO [optim.py:369] (1/2) 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 14:31:11,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=279899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:31:11,625 INFO [train.py:968] (1/2) Epoch 7, batch 6250, giga_loss[loss=0.3583, simple_loss=0.412, pruned_loss=0.1522, over 28788.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.389, pruned_loss=0.1356, over 5670997.67 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.371, pruned_loss=0.1115, over 5509299.45 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3876, pruned_loss=0.1357, over 5673842.32 frames. ], batch size: 145, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:31:32,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:31:40,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-03 14:31:42,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4859, 1.7065, 1.3048, 1.6779], device='cuda:1'), covar=tensor([0.2076, 0.1961, 0.2107, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.1174, 0.0893, 0.1029, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:31:43,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=279935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:32:00,028 INFO [train.py:968] (1/2) Epoch 7, batch 6300, giga_loss[loss=0.3384, simple_loss=0.398, pruned_loss=0.1394, over 28945.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3941, pruned_loss=0.1396, over 5672216.97 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3708, pruned_loss=0.1114, over 5518975.53 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3937, pruned_loss=0.1405, over 5669419.68 frames. ], batch size: 227, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:32:35,064 INFO [optim.py:369] (1/2) 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 14:32:47,104 INFO [train.py:968] (1/2) Epoch 7, batch 6350, giga_loss[loss=0.3606, simple_loss=0.4038, pruned_loss=0.1587, over 28651.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3967, pruned_loss=0.1429, over 5657042.76 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3702, pruned_loss=0.111, over 5530948.03 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3978, pruned_loss=0.145, over 5649043.19 frames. ], batch size: 262, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:32:55,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-03 14:33:27,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:33:29,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:33:34,819 INFO [train.py:968] (1/2) Epoch 7, batch 6400, giga_loss[loss=0.3782, simple_loss=0.4232, pruned_loss=0.1666, over 28767.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3967, pruned_loss=0.1434, over 5650595.21 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3703, pruned_loss=0.1111, over 5537632.69 frames. ], giga_tot_loss[loss=0.3456, simple_loss=0.3986, pruned_loss=0.1463, over 5643333.15 frames. ], batch size: 119, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:34:00,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:34:08,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2705, 1.6362, 1.3078, 1.7468], device='cuda:1'), covar=tensor([0.2448, 0.2201, 0.2367, 0.1894], device='cuda:1'), in_proj_covar=tensor([0.1173, 0.0892, 0.1030, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:34:14,402 INFO [optim.py:369] (1/2) 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 14:34:26,424 INFO [train.py:968] (1/2) Epoch 7, batch 6450, giga_loss[loss=0.3664, simple_loss=0.4095, pruned_loss=0.1616, over 28970.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.4002, pruned_loss=0.1476, over 5639328.57 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3702, pruned_loss=0.1111, over 5550285.45 frames. ], giga_tot_loss[loss=0.3526, simple_loss=0.4029, pruned_loss=0.1512, over 5625221.06 frames. ], batch size: 106, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:35:20,900 INFO [train.py:968] (1/2) Epoch 7, batch 6500, giga_loss[loss=0.3187, simple_loss=0.3888, pruned_loss=0.1243, over 28720.00 frames. ], tot_loss[loss=0.3563, simple_loss=0.4058, pruned_loss=0.1534, over 5619906.33 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3703, pruned_loss=0.1112, over 5555674.38 frames. ], giga_tot_loss[loss=0.3609, simple_loss=0.4083, pruned_loss=0.1568, over 5604623.03 frames. ], batch size: 119, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:35:26,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9993, 1.0314, 4.1692, 3.1753], device='cuda:1'), covar=tensor([0.1825, 0.2528, 0.0343, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0594, 0.0548, 0.0783, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:35:34,388 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 14:36:02,910 INFO [optim.py:369] (1/2) 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 14:36:14,139 INFO [train.py:968] (1/2) Epoch 7, batch 6550, giga_loss[loss=0.3335, simple_loss=0.3886, pruned_loss=0.1392, over 28925.00 frames. ], tot_loss[loss=0.3566, simple_loss=0.4061, pruned_loss=0.1536, over 5627177.89 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.37, pruned_loss=0.1111, over 5558863.67 frames. ], giga_tot_loss[loss=0.3619, simple_loss=0.4092, pruned_loss=0.1573, over 5613257.55 frames. ], batch size: 164, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:36:22,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 14:37:01,914 INFO [train.py:968] (1/2) Epoch 7, batch 6600, giga_loss[loss=0.3511, simple_loss=0.4057, pruned_loss=0.1483, over 29085.00 frames. ], tot_loss[loss=0.3555, simple_loss=0.4047, pruned_loss=0.1532, over 5645770.35 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3699, pruned_loss=0.1111, over 5565117.41 frames. ], giga_tot_loss[loss=0.3609, simple_loss=0.4079, pruned_loss=0.1569, over 5630552.67 frames. ], batch size: 155, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:37:42,759 INFO [optim.py:369] (1/2) 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 14:37:51,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:37:55,064 INFO [train.py:968] (1/2) Epoch 7, batch 6650, giga_loss[loss=0.3652, simple_loss=0.4192, pruned_loss=0.1556, over 28930.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4031, pruned_loss=0.1524, over 5638587.61 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3695, pruned_loss=0.1109, over 5570578.55 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.4069, pruned_loss=0.1566, over 5623089.90 frames. ], batch size: 199, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:38:07,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:38:19,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3090, 1.7882, 1.4237, 1.4997], device='cuda:1'), covar=tensor([0.0682, 0.0394, 0.0298, 0.0703], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0117, 0.0121, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0065, 0.0047, 0.0043, 0.0072], device='cuda:1')
+2023-03-03 14:38:34,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5477, 1.4763, 1.1479, 1.1832], device='cuda:1'), covar=tensor([0.0592, 0.0493, 0.0930, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0451, 0.0504, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:38:41,907 INFO [train.py:968] (1/2) Epoch 7, batch 6700, giga_loss[loss=0.3065, simple_loss=0.3723, pruned_loss=0.1204, over 28445.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.403, pruned_loss=0.1508, over 5636629.74 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3693, pruned_loss=0.1108, over 5569199.61 frames. ], giga_tot_loss[loss=0.3592, simple_loss=0.4072, pruned_loss=0.1556, over 5626809.37 frames. ], batch size: 77, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:39:20,183 INFO [optim.py:369] (1/2) 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 14:39:30,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 14:39:31,062 INFO [train.py:968] (1/2) Epoch 7, batch 6750, giga_loss[loss=0.3756, simple_loss=0.4202, pruned_loss=0.1655, over 28900.00 frames. ], tot_loss[loss=0.3538, simple_loss=0.4045, pruned_loss=0.1516, over 5627153.66 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3695, pruned_loss=0.1108, over 5565654.45 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.4083, pruned_loss=0.156, over 5624080.99 frames. ], batch size: 112, lr: 4.74e-03, grad_scale: 2.0
+2023-03-03 14:40:13,454 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:15,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:16,081 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 6800, giga_loss[loss=0.345, simple_loss=0.3972, pruned_loss=0.1464, over 28281.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4052, pruned_loss=0.1523, over 5609906.86 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3694, pruned_loss=0.1109, over 5572593.76 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4089, pruned_loss=0.1565, over 5602615.94 frames. ], batch size: 368, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:40:26,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:30,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:35,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9920, 1.2003, 3.4826, 3.1023], device='cuda:1'), covar=tensor([0.1625, 0.2258, 0.0410, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0547, 0.0784, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:40:43,039 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:51,860 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:55,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:40:59,289 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 6850, libri_loss[loss=0.2838, simple_loss=0.3566, pruned_loss=0.1055, over 29658.00 frames. ], tot_loss[loss=0.3483, simple_loss=0.4008, pruned_loss=0.1479, over 5618278.19 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3689, pruned_loss=0.1106, over 5582099.01 frames. ], giga_tot_loss[loss=0.3561, simple_loss=0.4057, pruned_loss=0.1532, over 5605628.70 frames. ], batch size: 88, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:41:54,631 INFO [train.py:968] (1/2) Epoch 7, batch 6900, giga_loss[loss=0.2782, simple_loss=0.3589, pruned_loss=0.09872, over 28447.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3972, pruned_loss=0.1435, over 5639032.46 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3683, pruned_loss=0.1104, over 5597646.94 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4032, pruned_loss=0.1498, over 5616107.63 frames. ], batch size: 65, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:42:03,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4443, 3.4038, 1.8964, 1.8130], device='cuda:1'), covar=tensor([0.1555, 0.0680, 0.1167, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.1511, 0.1361, 0.1333, 0.1424], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 14:42:36,419 INFO [optim.py:369] (1/2) 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 14:42:47,337 INFO [train.py:968] (1/2) Epoch 7, batch 6950, giga_loss[loss=0.3318, simple_loss=0.3913, pruned_loss=0.1361, over 28783.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3944, pruned_loss=0.1409, over 5650103.63 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.368, pruned_loss=0.1102, over 5600566.30 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3997, pruned_loss=0.1462, over 5629796.23 frames. ], batch size: 284, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:43:36,982 INFO [train.py:968] (1/2) Epoch 7, batch 7000, giga_loss[loss=0.3518, simple_loss=0.4095, pruned_loss=0.1471, over 28749.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3923, pruned_loss=0.139, over 5650627.96 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3677, pruned_loss=0.1101, over 5603402.98 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3972, pruned_loss=0.144, over 5633247.56 frames. ], batch size: 284, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:44:04,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8211, 1.6944, 1.3502, 1.4559], device='cuda:1'), covar=tensor([0.0645, 0.0652, 0.0887, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0452, 0.0502, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:44:12,275 INFO [optim.py:369] (1/2) 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 14:44:23,021 INFO [train.py:968] (1/2) Epoch 7, batch 7050, giga_loss[loss=0.3209, simple_loss=0.3766, pruned_loss=0.1327, over 29019.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3914, pruned_loss=0.1385, over 5649103.54 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3686, pruned_loss=0.1107, over 5598526.96 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3953, pruned_loss=0.1428, over 5640148.01 frames. ], batch size: 106, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:44:36,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9805, 0.9927, 3.8564, 3.0775], device='cuda:1'), covar=tensor([0.1770, 0.2532, 0.0406, 0.0723], device='cuda:1'), in_proj_covar=tensor([0.0589, 0.0548, 0.0786, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:44:42,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:45:01,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-03 14:45:12,612 INFO [train.py:968] (1/2) Epoch 7, batch 7100, giga_loss[loss=0.4275, simple_loss=0.4443, pruned_loss=0.2054, over 27583.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3911, pruned_loss=0.1383, over 5656200.14 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3681, pruned_loss=0.1105, over 5602892.93 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3949, pruned_loss=0.1423, over 5645947.61 frames. ], batch size: 472, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:45:34,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-03 14:45:58,008 INFO [optim.py:369] (1/2) 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 14:46:09,108 INFO [train.py:968] (1/2) Epoch 7, batch 7150, giga_loss[loss=0.3112, simple_loss=0.3767, pruned_loss=0.1229, over 28952.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3896, pruned_loss=0.1365, over 5653694.91 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3673, pruned_loss=0.1101, over 5600223.89 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3939, pruned_loss=0.1408, over 5648964.82 frames. ], batch size: 174, lr: 4.74e-03, grad_scale: 4.0
+2023-03-03 14:46:29,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:46:57,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3841, 3.0096, 1.4969, 1.3445], device='cuda:1'), covar=tensor([0.0775, 0.0282, 0.0742, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0489, 0.0317, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 14:47:00,841 INFO [train.py:968] (1/2) Epoch 7, batch 7200, giga_loss[loss=0.2705, simple_loss=0.3435, pruned_loss=0.09873, over 28590.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3883, pruned_loss=0.134, over 5666992.51 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3674, pruned_loss=0.1101, over 5604230.61 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.392, pruned_loss=0.1377, over 5660551.00 frames. ], batch size: 92, lr: 4.74e-03, grad_scale: 8.0
+2023-03-03 14:47:09,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2257, 1.6655, 1.2244, 0.3568], device='cuda:1'), covar=tensor([0.1776, 0.1084, 0.1822, 0.2670], device='cuda:1'), in_proj_covar=tensor([0.1443, 0.1354, 0.1406, 0.1192], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 14:47:12,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=280856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:47:46,681 INFO [optim.py:369] (1/2) 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 14:47:55,355 INFO [train.py:968] (1/2) Epoch 7, batch 7250, libri_loss[loss=0.2576, simple_loss=0.3317, pruned_loss=0.09176, over 29580.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3894, pruned_loss=0.1329, over 5663528.78 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3673, pruned_loss=0.11, over 5607980.43 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.393, pruned_loss=0.1364, over 5656111.50 frames. ], batch size: 75, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:48:04,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:48:20,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:48:32,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=280940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:48:41,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1172, 1.1602, 3.7606, 3.1619], device='cuda:1'), covar=tensor([0.1583, 0.2327, 0.0433, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0591, 0.0553, 0.0790, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:48:42,167 INFO [train.py:968] (1/2) Epoch 7, batch 7300, giga_loss[loss=0.3118, simple_loss=0.3802, pruned_loss=0.1217, over 28950.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3916, pruned_loss=0.1346, over 5657656.36 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3671, pruned_loss=0.1099, over 5609547.48 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3955, pruned_loss=0.1384, over 5652469.61 frames. ], batch size: 145, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:48:57,573 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:01,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=280965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:07,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9928, 1.2041, 0.9159, 0.1909], device='cuda:1'), covar=tensor([0.1282, 0.1204, 0.1621, 0.2694], device='cuda:1'), in_proj_covar=tensor([0.1456, 0.1374, 0.1411, 0.1200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 14:49:22,794 INFO [optim.py:369] (1/2) 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 14:49:29,014 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=280994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:34,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=280999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:35,082 INFO [train.py:968] (1/2) Epoch 7, batch 7350, giga_loss[loss=0.3256, simple_loss=0.3895, pruned_loss=0.1309, over 28589.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3915, pruned_loss=0.1349, over 5671471.91 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3669, pruned_loss=0.11, over 5615391.95 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3953, pruned_loss=0.1385, over 5663381.89 frames. ], batch size: 307, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:49:36,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:49:41,497 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:50:01,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:50:18,230 INFO [train.py:968] (1/2) Epoch 7, batch 7400, libri_loss[loss=0.3341, simple_loss=0.3958, pruned_loss=0.1362, over 19680.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3911, pruned_loss=0.1354, over 5671438.85 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3673, pruned_loss=0.1102, over 5618967.55 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3949, pruned_loss=0.1392, over 5664975.09 frames. ], batch size: 186, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:50:57,073 INFO [optim.py:369] (1/2) 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 14:51:03,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:51:03,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:51:07,791 INFO [train.py:968] (1/2) Epoch 7, batch 7450, giga_loss[loss=0.2692, simple_loss=0.3482, pruned_loss=0.09509, over 28879.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3898, pruned_loss=0.136, over 5656527.50 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3672, pruned_loss=0.1102, over 5615577.60 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3934, pruned_loss=0.1396, over 5655203.42 frames. ], batch size: 174, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:51:40,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4681, 1.4931, 1.1751, 1.1739], device='cuda:1'), covar=tensor([0.0657, 0.0503, 0.0950, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0453, 0.0506, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:51:52,818 INFO [train.py:968] (1/2) Epoch 7, batch 7500, giga_loss[loss=0.3134, simple_loss=0.3782, pruned_loss=0.1243, over 28568.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3882, pruned_loss=0.135, over 5670043.99 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3675, pruned_loss=0.1103, over 5616073.92 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3911, pruned_loss=0.138, over 5669360.83 frames. ], batch size: 307, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:52:32,733 INFO [optim.py:369] (1/2) 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 14:52:44,349 INFO [train.py:968] (1/2) Epoch 7, batch 7550, giga_loss[loss=0.3199, simple_loss=0.3813, pruned_loss=0.1292, over 28266.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3871, pruned_loss=0.1325, over 5684351.40 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3677, pruned_loss=0.1104, over 5622511.29 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3899, pruned_loss=0.1357, over 5680075.49 frames. ], batch size: 368, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:53:20,650 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:53:22,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:53:28,537 INFO [train.py:968] (1/2) Epoch 7, batch 7600, libri_loss[loss=0.3023, simple_loss=0.3745, pruned_loss=0.1151, over 29558.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3869, pruned_loss=0.1312, over 5695652.13 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3678, pruned_loss=0.1104, over 5631123.92 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3897, pruned_loss=0.1344, over 5686448.93 frames. ], batch size: 79, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:53:40,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-03 14:53:47,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8387, 2.2533, 1.7583, 2.0680], device='cuda:1'), covar=tensor([0.0591, 0.0256, 0.0253, 0.0619], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0118, 0.0121, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0072], device='cuda:1')
+2023-03-03 14:53:47,867 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:54:01,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:54:05,160 INFO [optim.py:369] (1/2) 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 14:54:12,556 INFO [train.py:968] (1/2) Epoch 7, batch 7650, libri_loss[loss=0.3243, simple_loss=0.3993, pruned_loss=0.1246, over 29161.00 frames. ], tot_loss[loss=0.325, simple_loss=0.387, pruned_loss=0.1315, over 5684700.35 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3681, pruned_loss=0.1106, over 5627062.79 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3896, pruned_loss=0.1347, over 5682798.11 frames. ], batch size: 97, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:54:13,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:54:20,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6193, 2.3618, 1.9950, 1.6039], device='cuda:1'), covar=tensor([0.1613, 0.1769, 0.1259, 0.1438], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0727, 0.0813, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 14:54:25,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:55:00,271 INFO [train.py:968] (1/2) Epoch 7, batch 7700, giga_loss[loss=0.3118, simple_loss=0.3691, pruned_loss=0.1272, over 28738.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3859, pruned_loss=0.1315, over 5689539.21 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3683, pruned_loss=0.1109, over 5630056.63 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3881, pruned_loss=0.1341, over 5686120.18 frames. ], batch size: 92, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:55:26,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0147, 1.2931, 0.9963, 0.1594], device='cuda:1'), covar=tensor([0.1401, 0.1227, 0.1908, 0.2849], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1360, 0.1402, 0.1194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 14:55:29,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:55:37,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3465, 1.5752, 1.2987, 1.6251], device='cuda:1'), covar=tensor([0.2163, 0.2010, 0.2044, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.1177, 0.0891, 0.1034, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 14:55:38,905 INFO [optim.py:369] (1/2) 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 14:55:39,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5606, 1.5553, 1.2327, 1.2813], device='cuda:1'), covar=tensor([0.0694, 0.0540, 0.0971, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0447, 0.0504, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 14:55:51,983 INFO [train.py:968] (1/2) Epoch 7, batch 7750, giga_loss[loss=0.4162, simple_loss=0.4365, pruned_loss=0.1979, over 26734.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3852, pruned_loss=0.1319, over 5683902.97 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3681, pruned_loss=0.1106, over 5634710.87 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3875, pruned_loss=0.1346, over 5678512.96 frames. ], batch size: 555, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:56:21,178 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:23,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:33,087 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 14:56:35,088 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:38,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:42,075 INFO [train.py:968] (1/2) Epoch 7, batch 7800, giga_loss[loss=0.3168, simple_loss=0.3762, pruned_loss=0.1287, over 28990.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.386, pruned_loss=0.1336, over 5680134.93 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.368, pruned_loss=0.1107, over 5630062.63 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3882, pruned_loss=0.1361, over 5680410.17 frames. ], batch size: 155, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:56:52,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:54,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:56:54,625 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:02,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:06,498 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-03 14:57:08,837 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:23,839 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:57:24,190 INFO [optim.py:369] (1/2) 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 14:57:34,082 INFO [train.py:968] (1/2) Epoch 7, batch 7850, giga_loss[loss=0.3533, simple_loss=0.3875, pruned_loss=0.1595, over 23795.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3839, pruned_loss=0.1327, over 5689543.19 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3678, pruned_loss=0.1106, over 5632722.92 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3859, pruned_loss=0.1349, over 5687980.38 frames. ], batch size: 705, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:57:54,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3097, 1.5446, 1.3989, 1.4257], device='cuda:1'), covar=tensor([0.0769, 0.0316, 0.0307, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0118, 0.0121, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0072], device='cuda:1')
+2023-03-03 14:57:58,859 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:58:01,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:58:20,454 INFO [train.py:968] (1/2) Epoch 7, batch 7900, giga_loss[loss=0.2897, simple_loss=0.3601, pruned_loss=0.1096, over 28705.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3823, pruned_loss=0.1319, over 5685660.09 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.368, pruned_loss=0.1108, over 5630840.66 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3841, pruned_loss=0.1341, over 5687558.74 frames. ], batch size: 307, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:58:29,305 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:58:35,303 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:59:06,142 INFO [train.py:968] (1/2) Epoch 7, batch 7950, giga_loss[loss=0.3293, simple_loss=0.3861, pruned_loss=0.1362, over 28656.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3809, pruned_loss=0.1307, over 5688856.52 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3684, pruned_loss=0.111, over 5628876.49 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3824, pruned_loss=0.1329, over 5694532.66 frames. ], batch size: 242, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 14:59:19,386 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:59:21,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:59:51,939 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 14:59:55,850 INFO [train.py:968] (1/2) Epoch 7, batch 8000, libri_loss[loss=0.2858, simple_loss=0.3628, pruned_loss=0.1044, over 29519.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3817, pruned_loss=0.1314, over 5683873.24 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3683, pruned_loss=0.1109, over 5633395.59 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3833, pruned_loss=0.1337, over 5685286.93 frames. ], batch size: 81, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 14:59:56,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3490, 2.6395, 1.4493, 1.3719], device='cuda:1'), covar=tensor([0.0758, 0.0357, 0.0718, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0493, 0.0316, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 15:00:35,315 INFO [optim.py:369] (1/2) 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 15:00:44,546 INFO [train.py:968] (1/2) Epoch 7, batch 8050, giga_loss[loss=0.2764, simple_loss=0.3551, pruned_loss=0.09888, over 28952.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3819, pruned_loss=0.1306, over 5683013.45 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3683, pruned_loss=0.1109, over 5634883.62 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3832, pruned_loss=0.1325, over 5683098.49 frames. ], batch size: 112, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:01:17,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 15:01:30,202 INFO [train.py:968] (1/2) Epoch 7, batch 8100, giga_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.09823, over 28569.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3815, pruned_loss=0.1299, over 5676439.83 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3677, pruned_loss=0.1105, over 5642230.85 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3835, pruned_loss=0.1323, over 5671021.34 frames. ], batch size: 85, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:02:06,068 INFO [optim.py:369] (1/2) 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 15:02:13,839 INFO [train.py:968] (1/2) Epoch 7, batch 8150, giga_loss[loss=0.3482, simple_loss=0.4031, pruned_loss=0.1466, over 28943.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3825, pruned_loss=0.1305, over 5677950.66 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3677, pruned_loss=0.1105, over 5643638.26 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3846, pruned_loss=0.1332, over 5673156.33 frames. ], batch size: 164, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:02:31,842 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=281817.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:02:33,250 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:03:04,746 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=281849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:03:05,261 INFO [train.py:968] (1/2) Epoch 7, batch 8200, giga_loss[loss=0.2886, simple_loss=0.3522, pruned_loss=0.1125, over 28515.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3847, pruned_loss=0.1324, over 5682983.22 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3678, pruned_loss=0.1105, over 5647497.95 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3866, pruned_loss=0.1349, over 5676178.96 frames. ], batch size: 71, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:03:50,023 INFO [optim.py:369] (1/2) 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 15:03:58,216 INFO [train.py:968] (1/2) Epoch 7, batch 8250, giga_loss[loss=0.3895, simple_loss=0.4264, pruned_loss=0.1763, over 28929.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3864, pruned_loss=0.1347, over 5682830.91 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3678, pruned_loss=0.1103, over 5651907.50 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3883, pruned_loss=0.1374, over 5674270.38 frames. ], batch size: 213, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:04:35,460 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 8300, libri_loss[loss=0.2674, simple_loss=0.3522, pruned_loss=0.09133, over 29503.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3884, pruned_loss=0.1373, over 5677148.33 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.368, pruned_loss=0.1102, over 5656957.82 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3902, pruned_loss=0.1402, over 5666116.65 frames. ], batch size: 81, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:04:57,660 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=281960.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:04:59,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=281963.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:05:28,701 INFO [optim.py:369] (1/2) 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 15:05:28,987 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=281992.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:05:31,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-03 15:05:36,840 INFO [train.py:968] (1/2) Epoch 7, batch 8350, giga_loss[loss=0.4862, simple_loss=0.4937, pruned_loss=0.2393, over 26459.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3906, pruned_loss=0.1406, over 5660405.58 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3681, pruned_loss=0.1105, over 5645661.82 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3926, pruned_loss=0.1436, over 5662922.48 frames. ], batch size: 555, lr: 4.73e-03, grad_scale: 4.0
+2023-03-03 15:06:23,761 INFO [train.py:968] (1/2) Epoch 7, batch 8400, libri_loss[loss=0.2651, simple_loss=0.346, pruned_loss=0.09205, over 29529.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.389, pruned_loss=0.1393, over 5663808.76 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3681, pruned_loss=0.1105, over 5650890.30 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3911, pruned_loss=0.1423, over 5661166.22 frames. ], batch size: 80, lr: 4.73e-03, grad_scale: 8.0
+2023-03-03 15:06:48,309 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:06:50,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:06:57,609 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 8450, libri_loss[loss=0.2566, simple_loss=0.3239, pruned_loss=0.09469, over 29365.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3879, pruned_loss=0.1378, over 5672489.98 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3679, pruned_loss=0.1103, over 5661035.26 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3906, pruned_loss=0.1413, over 5661642.53 frames. ], batch size: 67, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:07:15,011 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 8500, giga_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1154, over 28767.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3857, pruned_loss=0.1353, over 5668115.53 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3676, pruned_loss=0.1102, over 5665972.79 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3884, pruned_loss=0.1387, over 5655523.32 frames. ], batch size: 284, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:08:25,584 INFO [optim.py:369] (1/2) 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 15:08:28,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:08:30,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5784, 4.4065, 4.1544, 1.7827], device='cuda:1'), covar=tensor([0.0483, 0.0617, 0.0682, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0896, 0.0808, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 15:08:32,170 INFO [train.py:968] (1/2) Epoch 7, batch 8550, giga_loss[loss=0.348, simple_loss=0.3981, pruned_loss=0.149, over 28743.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3833, pruned_loss=0.1329, over 5677348.76 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3683, pruned_loss=0.1106, over 5665735.80 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3855, pruned_loss=0.1361, over 5666949.74 frames. ], batch size: 284, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:08:56,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:09:00,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3798, 1.6025, 1.3006, 1.5480], device='cuda:1'), covar=tensor([0.2201, 0.2093, 0.2164, 0.2057], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.0904, 0.1040, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 15:09:14,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:09:19,826 INFO [train.py:968] (1/2) Epoch 7, batch 8600, giga_loss[loss=0.3187, simple_loss=0.3862, pruned_loss=0.1256, over 28771.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.381, pruned_loss=0.132, over 5675633.80 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3682, pruned_loss=0.1105, over 5660418.87 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.383, pruned_loss=0.135, over 5672301.83 frames. ], batch size: 119, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:09:37,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3255, 4.1416, 3.9448, 1.8083], device='cuda:1'), covar=tensor([0.0461, 0.0606, 0.0610, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0890, 0.0802, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 15:10:04,458 INFO [optim.py:369] (1/2) 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 15:10:11,643 INFO [train.py:968] (1/2) Epoch 7, batch 8650, giga_loss[loss=0.3183, simple_loss=0.3794, pruned_loss=0.1286, over 28684.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3817, pruned_loss=0.1337, over 5662986.23 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3682, pruned_loss=0.1104, over 5662543.12 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3835, pruned_loss=0.1364, over 5658294.85 frames. ], batch size: 307, lr: 4.72e-03, grad_scale: 2.0
+2023-03-03 15:10:13,966 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282303.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:10:49,978 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:10:50,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:10:53,149 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 8700, giga_loss[loss=0.3658, simple_loss=0.4181, pruned_loss=0.1567, over 28632.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3834, pruned_loss=0.1348, over 5658482.63 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3682, pruned_loss=0.1106, over 5667892.73 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.385, pruned_loss=0.1373, over 5650374.10 frames. ], batch size: 307, lr: 4.72e-03, grad_scale: 2.0
+2023-03-03 15:11:17,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3260, 1.4069, 1.4478, 1.3165], device='cuda:1'), covar=tensor([0.1110, 0.1292, 0.1609, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0736, 0.0650, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 15:11:19,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:23,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:24,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:47,720 INFO [optim.py:369] (1/2) 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 15:11:51,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2336, 1.3644, 0.9391, 1.1337], device='cuda:1'), covar=tensor([0.0906, 0.0836, 0.0765, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.1556, 0.1390, 0.1355, 0.1466], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 15:11:53,929 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:11:54,304 INFO [train.py:968] (1/2) Epoch 7, batch 8750, giga_loss[loss=0.3306, simple_loss=0.4044, pruned_loss=0.1284, over 28979.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3868, pruned_loss=0.1345, over 5662963.56 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.368, pruned_loss=0.1104, over 5670314.76 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3885, pruned_loss=0.1368, over 5654317.61 frames. ], batch size: 164, lr: 4.72e-03, grad_scale: 2.0
+2023-03-03 15:12:18,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1641, 2.5558, 1.1857, 1.2746], device='cuda:1'), covar=tensor([0.0896, 0.0369, 0.0900, 0.1351], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0489, 0.0315, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 15:12:42,050 INFO [train.py:968] (1/2) Epoch 7, batch 8800, giga_loss[loss=0.3351, simple_loss=0.3925, pruned_loss=0.1388, over 28923.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3875, pruned_loss=0.1326, over 5679096.12 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.368, pruned_loss=0.1104, over 5676988.91 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3894, pruned_loss=0.1351, over 5665687.99 frames. ], batch size: 86, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:13:13,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8664, 1.7353, 1.2755, 1.3751], device='cuda:1'), covar=tensor([0.0649, 0.0631, 0.0998, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0450, 0.0505, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 15:13:18,225 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:13:22,592 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 8850, giga_loss[loss=0.2999, simple_loss=0.3713, pruned_loss=0.1143, over 28643.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3915, pruned_loss=0.136, over 5672355.51 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3683, pruned_loss=0.1107, over 5674789.32 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.393, pruned_loss=0.1381, over 5663754.17 frames. ], batch size: 78, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:13:38,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7202, 4.5340, 1.8640, 1.7064], device='cuda:1'), covar=tensor([0.0821, 0.0224, 0.0767, 0.1231], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0494, 0.0316, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 15:13:46,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2847, 1.6131, 1.6406, 1.2717], device='cuda:1'), covar=tensor([0.1338, 0.1983, 0.1076, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0730, 0.0815, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 15:14:10,766 INFO [train.py:968] (1/2) Epoch 7, batch 8900, giga_loss[loss=0.3479, simple_loss=0.4083, pruned_loss=0.1437, over 28982.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3939, pruned_loss=0.1387, over 5664876.25 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3677, pruned_loss=0.1102, over 5679787.98 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3963, pruned_loss=0.1415, over 5653475.84 frames. ], batch size: 164, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:14:47,322 INFO [optim.py:369] (1/2) 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 15:14:51,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6208, 4.4654, 4.2028, 1.8196], device='cuda:1'), covar=tensor([0.0427, 0.0575, 0.0661, 0.2108], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0891, 0.0798, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 15:14:53,082 INFO [train.py:968] (1/2) Epoch 7, batch 8950, giga_loss[loss=0.3329, simple_loss=0.3932, pruned_loss=0.1363, over 28576.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3919, pruned_loss=0.1374, over 5669062.32 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3679, pruned_loss=0.1104, over 5684914.88 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3948, pruned_loss=0.1407, over 5654752.26 frames. ], batch size: 307, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:15:12,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:15:21,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2290, 4.0303, 3.7686, 1.8004], device='cuda:1'), covar=tensor([0.0636, 0.0850, 0.1006, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0891, 0.0797, 0.0622], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 15:15:42,799 INFO [train.py:968] (1/2) Epoch 7, batch 9000, giga_loss[loss=0.2963, simple_loss=0.3652, pruned_loss=0.1137, over 28843.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3904, pruned_loss=0.1375, over 5659147.07 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3679, pruned_loss=0.1104, over 5690859.86 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3933, pruned_loss=0.1409, over 5641623.45 frames. ], batch size: 145, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:15:42,800 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 15:15:51,394 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 15:16:18,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282678.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:16:32,030 INFO [optim.py:369] (1/2) 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 15:16:37,293 INFO [train.py:968] (1/2) Epoch 7, batch 9050, giga_loss[loss=0.4078, simple_loss=0.4346, pruned_loss=0.1905, over 27669.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3882, pruned_loss=0.1363, over 5664522.36 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3673, pruned_loss=0.11, over 5695127.26 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3915, pruned_loss=0.1398, over 5646222.64 frames. ], batch size: 472, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:16:52,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:17:14,698 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-03 15:17:26,177 INFO [train.py:968] (1/2) Epoch 7, batch 9100, giga_loss[loss=0.3467, simple_loss=0.4053, pruned_loss=0.1441, over 28884.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.387, pruned_loss=0.1363, over 5668612.78 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3671, pruned_loss=0.1097, over 5700947.52 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3903, pruned_loss=0.1401, over 5647984.51 frames. ], batch size: 186, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:17:39,312 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:17:42,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:18:09,223 INFO [optim.py:369] (1/2) 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 15:18:11,228 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:18:14,598 INFO [train.py:968] (1/2) Epoch 7, batch 9150, giga_loss[loss=0.3171, simple_loss=0.3798, pruned_loss=0.1272, over 28939.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3876, pruned_loss=0.1373, over 5665280.74 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3671, pruned_loss=0.1097, over 5702507.19 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3907, pruned_loss=0.1409, over 5647085.09 frames. ], batch size: 213, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:18:32,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9696, 1.0959, 0.8948, 0.8337], device='cuda:1'), covar=tensor([0.0993, 0.1145, 0.0741, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.1524, 0.1375, 0.1339, 0.1434], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 15:18:37,341 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282821.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:18:41,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282824.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:19:04,463 INFO [train.py:968] (1/2) Epoch 7, batch 9200, libri_loss[loss=0.2606, simple_loss=0.335, pruned_loss=0.09311, over 29649.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.388, pruned_loss=0.1376, over 5659814.59 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3672, pruned_loss=0.1096, over 5707474.47 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.391, pruned_loss=0.1415, over 5639460.94 frames. ], batch size: 73, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:19:07,207 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282853.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:19:09,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=282856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:12,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=282859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:17,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=282864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:34,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=282882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:41,105 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=282888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:19:45,807 INFO [optim.py:369] (1/2) 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 15:19:50,644 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6969, 1.7410, 1.5480, 1.6942], device='cuda:1'), covar=tensor([0.1284, 0.1961, 0.1924, 0.1601], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0740, 0.0650, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 15:19:50,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4747, 1.6597, 1.4235, 1.5659], device='cuda:1'), covar=tensor([0.2035, 0.2055, 0.2098, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.1179, 0.0903, 0.1037, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 15:19:52,621 INFO [train.py:968] (1/2) Epoch 7, batch 9250, giga_loss[loss=0.2764, simple_loss=0.3515, pruned_loss=0.1007, over 28779.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3844, pruned_loss=0.1358, over 5667820.55 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3668, pruned_loss=0.1093, over 5711273.95 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3876, pruned_loss=0.1396, over 5647755.42 frames. ], batch size: 119, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:20:40,364 INFO [train.py:968] (1/2) Epoch 7, batch 9300, giga_loss[loss=0.3272, simple_loss=0.3905, pruned_loss=0.132, over 28887.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3836, pruned_loss=0.1353, over 5665972.88 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3666, pruned_loss=0.1093, over 5715926.08 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3867, pruned_loss=0.1389, over 5645075.44 frames. ], batch size: 174, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:20:45,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 15:21:23,010 INFO [optim.py:369] (1/2) 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 15:21:30,265 INFO [train.py:968] (1/2) Epoch 7, batch 9350, giga_loss[loss=0.3142, simple_loss=0.3805, pruned_loss=0.124, over 28945.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3843, pruned_loss=0.1346, over 5672410.30 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3661, pruned_loss=0.109, over 5721585.17 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3877, pruned_loss=0.1385, over 5649082.12 frames. ], batch size: 227, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:21:37,964 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:21:41,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:21:57,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0314, 1.1647, 3.4359, 2.9744], device='cuda:1'), covar=tensor([0.1523, 0.2340, 0.0415, 0.1541], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0552, 0.0787, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 15:21:58,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3327, 1.4098, 1.2423, 1.4428], device='cuda:1'), covar=tensor([0.0743, 0.0304, 0.0317, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0118, 0.0121, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0066, 0.0048, 0.0043, 0.0073], device='cuda:1')
+2023-03-03 15:22:07,314 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 9400, giga_loss[loss=0.3035, simple_loss=0.3694, pruned_loss=0.1188, over 28891.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3869, pruned_loss=0.136, over 5671368.03 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.366, pruned_loss=0.109, over 5718958.66 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3903, pruned_loss=0.14, over 5653082.52 frames. ], batch size: 145, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:22:18,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:22:37,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:22:42,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 15:22:55,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283091.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:22:59,336 INFO [optim.py:369] (1/2) 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 15:23:03,391 INFO [train.py:968] (1/2) Epoch 7, batch 9450, giga_loss[loss=0.3349, simple_loss=0.3864, pruned_loss=0.1417, over 28315.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3877, pruned_loss=0.1375, over 5664843.94 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3663, pruned_loss=0.1092, over 5722004.48 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3905, pruned_loss=0.141, over 5646856.51 frames. ], batch size: 77, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:23:11,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3152, 1.7428, 1.6785, 1.3312], device='cuda:1'), covar=tensor([0.1497, 0.1991, 0.1194, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0729, 0.0816, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 15:23:30,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4641, 1.5504, 1.5627, 1.4291], device='cuda:1'), covar=tensor([0.1175, 0.1460, 0.1574, 0.1499], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0730, 0.0643, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 15:23:52,358 INFO [train.py:968] (1/2) Epoch 7, batch 9500, giga_loss[loss=0.2926, simple_loss=0.3882, pruned_loss=0.09852, over 29010.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3886, pruned_loss=0.1356, over 5665808.30 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3661, pruned_loss=0.1092, over 5716442.09 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3915, pruned_loss=0.1388, over 5655128.13 frames. ], batch size: 155, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:24:32,744 INFO [optim.py:369] (1/2) 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 15:24:34,949 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:24:36,017 INFO [train.py:968] (1/2) Epoch 7, batch 9550, giga_loss[loss=0.2917, simple_loss=0.376, pruned_loss=0.1037, over 29036.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3897, pruned_loss=0.1345, over 5662564.60 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3661, pruned_loss=0.1093, over 5711010.01 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3924, pruned_loss=0.1374, over 5657292.39 frames. ], batch size: 136, lr: 4.72e-03, grad_scale: 4.0
+2023-03-03 15:25:25,995 INFO [train.py:968] (1/2) Epoch 7, batch 9600, giga_loss[loss=0.3683, simple_loss=0.4165, pruned_loss=0.16, over 27923.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.392, pruned_loss=0.1349, over 5667241.44 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3659, pruned_loss=0.1093, over 5711601.58 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3948, pruned_loss=0.1379, over 5661409.79 frames. ], batch size: 412, lr: 4.72e-03, grad_scale: 8.0
+2023-03-03 15:25:31,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:26:10,266 INFO [optim.py:369] (1/2) 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 15:26:13,078 INFO [train.py:968] (1/2) Epoch 7, batch 9650, giga_loss[loss=0.3498, simple_loss=0.4076, pruned_loss=0.146, over 28896.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3948, pruned_loss=0.1378, over 5672678.36 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3659, pruned_loss=0.1095, over 5714758.74 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3977, pruned_loss=0.1406, over 5664328.04 frames. ], batch size: 174, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:26:14,543 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 15:26:58,135 INFO [train.py:968] (1/2) Epoch 7, batch 9700, giga_loss[loss=0.3488, simple_loss=0.4025, pruned_loss=0.1475, over 29043.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3967, pruned_loss=0.1402, over 5682741.14 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3661, pruned_loss=0.1096, over 5719795.74 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3996, pruned_loss=0.143, over 5670625.53 frames. ], batch size: 136, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:27:44,585 INFO [optim.py:369] (1/2) 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 15:27:47,657 INFO [train.py:968] (1/2) Epoch 7, batch 9750, giga_loss[loss=0.2826, simple_loss=0.3631, pruned_loss=0.101, over 28705.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3963, pruned_loss=0.1409, over 5666352.36 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3658, pruned_loss=0.1094, over 5723234.17 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3996, pruned_loss=0.1441, over 5652474.25 frames. ], batch size: 60, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:27:49,000 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:27:52,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:00,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2599, 2.9768, 1.4374, 1.3085], device='cuda:1'), covar=tensor([0.0880, 0.0318, 0.0803, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0489, 0.0315, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 15:28:00,973 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 15:28:12,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:20,092 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:27,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9431, 3.0128, 2.0008, 1.0178], device='cuda:1'), covar=tensor([0.4090, 0.1712, 0.2216, 0.3842], device='cuda:1'), in_proj_covar=tensor([0.1453, 0.1374, 0.1413, 0.1185], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 15:28:31,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:28:34,601 INFO [train.py:968] (1/2) Epoch 7, batch 9800, giga_loss[loss=0.2899, simple_loss=0.3657, pruned_loss=0.1071, over 28820.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3949, pruned_loss=0.1394, over 5665258.78 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3659, pruned_loss=0.1095, over 5717087.67 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3979, pruned_loss=0.1423, over 5659295.81 frames. ], batch size: 99, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:28:49,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283466.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:29:14,644 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-03 15:29:14,723 INFO [optim.py:369] (1/2) 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 15:29:17,435 INFO [train.py:968] (1/2) Epoch 7, batch 9850, libri_loss[loss=0.309, simple_loss=0.39, pruned_loss=0.114, over 29744.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3936, pruned_loss=0.1365, over 5664452.59 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.366, pruned_loss=0.1097, over 5712873.46 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.397, pruned_loss=0.1398, over 5661911.52 frames. ], batch size: 87, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:29:25,049 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:29:49,707 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:29:59,990 INFO [train.py:968] (1/2) Epoch 7, batch 9900, giga_loss[loss=0.3219, simple_loss=0.3872, pruned_loss=0.1283, over 28260.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3946, pruned_loss=0.1363, over 5671040.05 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3663, pruned_loss=0.1099, over 5718384.62 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3978, pruned_loss=0.1394, over 5662938.96 frames. ], batch size: 368, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:30:17,467 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:20,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1158, 2.5363, 1.2138, 1.2379], device='cuda:1'), covar=tensor([0.0987, 0.0370, 0.0832, 0.1421], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0492, 0.0316, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 15:30:20,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:21,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:37,000 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:42,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:46,026 INFO [optim.py:369] (1/2) 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 15:30:49,662 INFO [train.py:968] (1/2) Epoch 7, batch 9950, giga_loss[loss=0.3375, simple_loss=0.3968, pruned_loss=0.1391, over 28726.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3954, pruned_loss=0.1369, over 5669345.22 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3665, pruned_loss=0.1099, over 5719865.93 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3982, pruned_loss=0.1399, over 5660711.20 frames. ], batch size: 262, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:30:50,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:30:57,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283609.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:31:00,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283612.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:31:09,849 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:31:29,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283641.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 15:31:36,421 INFO [train.py:968] (1/2) Epoch 7, batch 10000, giga_loss[loss=0.3226, simple_loss=0.3893, pruned_loss=0.1279, over 28880.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3955, pruned_loss=0.1377, over 5676041.50 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3662, pruned_loss=0.1098, over 5727001.86 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3992, pruned_loss=0.1413, over 5660619.41 frames. ], batch size: 199, lr: 4.71e-03, grad_scale: 8.0
+2023-03-03 15:32:01,257 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:32:20,460 INFO [optim.py:369] (1/2) 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 15:32:22,812 INFO [train.py:968] (1/2) Epoch 7, batch 10050, giga_loss[loss=0.2826, simple_loss=0.3478, pruned_loss=0.1087, over 28629.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3936, pruned_loss=0.1379, over 5666431.02 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3662, pruned_loss=0.1098, over 5730592.18 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3972, pruned_loss=0.1413, over 5649913.04 frames. ], batch size: 85, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:32:42,343 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=283716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:32:44,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=283719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:33:13,183 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=283748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:33:14,134 INFO [train.py:968] (1/2) Epoch 7, batch 10100, giga_loss[loss=0.3156, simple_loss=0.3755, pruned_loss=0.1279, over 28962.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3918, pruned_loss=0.1374, over 5674311.98 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3661, pruned_loss=0.1096, over 5735566.27 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3955, pruned_loss=0.1412, over 5654541.74 frames. ], batch size: 136, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:33:37,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7261, 4.2040, 1.7774, 1.7017], device='cuda:1'), covar=tensor([0.0772, 0.0214, 0.0730, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0494, 0.0318, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 15:34:02,809 INFO [optim.py:369] (1/2) 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 15:34:04,353 INFO [train.py:968] (1/2) Epoch 7, batch 10150, giga_loss[loss=0.2973, simple_loss=0.3663, pruned_loss=0.1141, over 28558.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3896, pruned_loss=0.1371, over 5670827.22 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3661, pruned_loss=0.1096, over 5738032.79 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3928, pruned_loss=0.1404, over 5652248.14 frames. ], batch size: 60, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:34:57,720 INFO [train.py:968] (1/2) Epoch 7, batch 10200, giga_loss[loss=0.3715, simple_loss=0.4059, pruned_loss=0.1685, over 27597.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3889, pruned_loss=0.1376, over 5665429.47 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3662, pruned_loss=0.1096, over 5740121.59 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3917, pruned_loss=0.1406, over 5648156.66 frames. ], batch size: 472, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:35:27,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:35:40,323 INFO [optim.py:369] (1/2) 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 15:35:41,521 INFO [train.py:968] (1/2) Epoch 7, batch 10250, giga_loss[loss=0.3136, simple_loss=0.3798, pruned_loss=0.1237, over 28940.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3867, pruned_loss=0.1361, over 5672488.93 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.366, pruned_loss=0.1095, over 5744618.46 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3899, pruned_loss=0.1395, over 5652249.14 frames. ], batch size: 227, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:35:55,421 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=283912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:36:19,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=283937.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:36:30,504 INFO [train.py:968] (1/2) Epoch 7, batch 10300, giga_loss[loss=0.2855, simple_loss=0.3532, pruned_loss=0.1089, over 28780.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3841, pruned_loss=0.1332, over 5664298.15 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3662, pruned_loss=0.1096, over 5733353.43 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3867, pruned_loss=0.1361, over 5657621.32 frames. ], batch size: 119, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:36:34,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 15:37:19,115 INFO [optim.py:369] (1/2) 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 15:37:21,383 INFO [train.py:968] (1/2) Epoch 7, batch 10350, giga_loss[loss=0.3099, simple_loss=0.3696, pruned_loss=0.1251, over 28688.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3802, pruned_loss=0.1293, over 5656384.02 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3661, pruned_loss=0.1095, over 5734248.66 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3824, pruned_loss=0.1317, over 5649967.46 frames. ], batch size: 92, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:37:47,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:37:50,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:09,112 INFO [train.py:968] (1/2) Epoch 7, batch 10400, giga_loss[loss=0.2929, simple_loss=0.3609, pruned_loss=0.1125, over 28905.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3796, pruned_loss=0.1276, over 5668566.94 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3663, pruned_loss=0.1095, over 5740131.34 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3816, pruned_loss=0.1303, over 5655713.75 frames. ], batch size: 186, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:38:10,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:13,236 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:16,529 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:16,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:38:44,194 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:38:57,189 INFO [train.py:968] (1/2) Epoch 7, batch 10450, giga_loss[loss=0.2983, simple_loss=0.3547, pruned_loss=0.1209, over 28879.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3774, pruned_loss=0.1269, over 5662430.43 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3659, pruned_loss=0.1091, over 5734003.47 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3798, pruned_loss=0.1299, over 5654819.97 frames. ], batch size: 112, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:39:31,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3115, 3.1187, 2.9297, 1.4427], device='cuda:1'), covar=tensor([0.0829, 0.0959, 0.0943, 0.2173], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0892, 0.0806, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 15:39:44,496 INFO [train.py:968] (1/2) Epoch 7, batch 10500, giga_loss[loss=0.3249, simple_loss=0.3561, pruned_loss=0.1469, over 23524.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3747, pruned_loss=0.1255, over 5666267.73 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3662, pruned_loss=0.1091, over 5740102.66 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3767, pruned_loss=0.1286, over 5652125.97 frames. ], batch size: 705, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:39:58,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8995, 1.8021, 1.3301, 1.4379], device='cuda:1'), covar=tensor([0.0696, 0.0654, 0.1034, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0451, 0.0503, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 15:40:26,380 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:40:28,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:40:29,325 INFO [optim.py:369] (1/2) 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 15:40:29,997 INFO [train.py:968] (1/2) Epoch 7, batch 10550, giga_loss[loss=0.2985, simple_loss=0.3659, pruned_loss=0.1156, over 28613.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.376, pruned_loss=0.1263, over 5670615.96 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3659, pruned_loss=0.1089, over 5739182.91 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3782, pruned_loss=0.1295, over 5658091.34 frames. ], batch size: 92, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:40:43,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8330, 1.8921, 1.3558, 1.5351], device='cuda:1'), covar=tensor([0.0608, 0.0487, 0.0918, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0446, 0.0496, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 15:40:43,628 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-03 15:40:52,981 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:41:00,368 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:41:12,857 INFO [train.py:968] (1/2) Epoch 7, batch 10600, giga_loss[loss=0.2897, simple_loss=0.3651, pruned_loss=0.1071, over 28987.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3785, pruned_loss=0.1274, over 5676124.44 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.366, pruned_loss=0.1088, over 5747072.99 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3808, pruned_loss=0.131, over 5655409.48 frames. ], batch size: 136, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:41:52,920 INFO [optim.py:369] (1/2) 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 15:41:53,578 INFO [train.py:968] (1/2) Epoch 7, batch 10650, giga_loss[loss=0.3223, simple_loss=0.3782, pruned_loss=0.1332, over 27896.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3806, pruned_loss=0.1287, over 5671178.41 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3668, pruned_loss=0.1093, over 5752350.71 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3824, pruned_loss=0.1321, over 5645878.16 frames. ], batch size: 412, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:42:07,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284312.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:42:21,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4393, 2.3256, 2.1197, 2.1269], device='cuda:1'), covar=tensor([0.1182, 0.1846, 0.1614, 0.1546], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0736, 0.0650, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 15:42:23,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9790, 1.2579, 0.8479, 1.0017], device='cuda:1'), covar=tensor([0.0928, 0.0578, 0.1578, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0447, 0.0496, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 15:42:36,732 INFO [train.py:968] (1/2) Epoch 7, batch 10700, giga_loss[loss=0.3042, simple_loss=0.3686, pruned_loss=0.1199, over 28889.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3793, pruned_loss=0.128, over 5660041.38 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3671, pruned_loss=0.1098, over 5738268.57 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3811, pruned_loss=0.1312, over 5648060.78 frames. ], batch size: 145, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:42:45,592 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:43:17,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2650, 1.7271, 1.2552, 0.5502], device='cuda:1'), covar=tensor([0.2303, 0.1112, 0.1470, 0.2458], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1359, 0.1400, 0.1178], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 15:43:21,060 INFO [optim.py:369] (1/2) 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 15:43:22,903 INFO [train.py:968] (1/2) Epoch 7, batch 10750, giga_loss[loss=0.4257, simple_loss=0.4502, pruned_loss=0.2006, over 27553.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3792, pruned_loss=0.1285, over 5651759.20 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.367, pruned_loss=0.1097, over 5733580.33 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.381, pruned_loss=0.1316, over 5645208.90 frames. ], batch size: 472, lr: 4.71e-03, grad_scale: 2.0
+2023-03-03 15:43:34,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5069, 4.3493, 4.1472, 2.0382], device='cuda:1'), covar=tensor([0.0489, 0.0586, 0.0744, 0.1880], device='cuda:1'), in_proj_covar=tensor([0.0954, 0.0906, 0.0820, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 15:44:11,330 INFO [train.py:968] (1/2) Epoch 7, batch 10800, giga_loss[loss=0.3406, simple_loss=0.3989, pruned_loss=0.1412, over 28976.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3821, pruned_loss=0.1307, over 5660615.63 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3667, pruned_loss=0.1094, over 5736292.02 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3841, pruned_loss=0.1339, over 5651089.25 frames. ], batch size: 213, lr: 4.71e-03, grad_scale: 4.0
+2023-03-03 15:44:17,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284455.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:44:20,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284458.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:44:46,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284487.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 15:44:55,541 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 10850, giga_loss[loss=0.3617, simple_loss=0.4074, pruned_loss=0.158, over 27900.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3839, pruned_loss=0.1313, over 5664176.40 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3668, pruned_loss=0.1093, over 5738902.89 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3861, pruned_loss=0.1349, over 5651140.30 frames. ], batch size: 412, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:45:12,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2540, 4.0391, 3.7847, 1.8210], device='cuda:1'), covar=tensor([0.0660, 0.0890, 0.1111, 0.2096], device='cuda:1'), in_proj_covar=tensor([0.0950, 0.0901, 0.0814, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 15:45:41,581 INFO [train.py:968] (1/2) Epoch 7, batch 10900, giga_loss[loss=0.3338, simple_loss=0.3952, pruned_loss=0.1362, over 28993.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3848, pruned_loss=0.1321, over 5665954.81 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3666, pruned_loss=0.1092, over 5733402.18 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.387, pruned_loss=0.1355, over 5658486.55 frames. ], batch size: 145, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:46:32,444 INFO [optim.py:369] (1/2) 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 15:46:33,190 INFO [train.py:968] (1/2) Epoch 7, batch 10950, giga_loss[loss=0.349, simple_loss=0.4058, pruned_loss=0.146, over 28327.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.386, pruned_loss=0.1339, over 5670527.25 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3664, pruned_loss=0.1091, over 5735486.76 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3882, pruned_loss=0.1369, over 5662290.93 frames. ], batch size: 368, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:46:45,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:47:27,208 INFO [train.py:968] (1/2) Epoch 7, batch 11000, giga_loss[loss=0.2875, simple_loss=0.362, pruned_loss=0.1066, over 28941.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3882, pruned_loss=0.1347, over 5662384.80 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3662, pruned_loss=0.109, over 5737164.52 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3903, pruned_loss=0.1374, over 5653791.69 frames. ], batch size: 145, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:48:18,943 INFO [optim.py:369] (1/2) 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 15:48:20,605 INFO [train.py:968] (1/2) Epoch 7, batch 11050, giga_loss[loss=0.3415, simple_loss=0.3791, pruned_loss=0.1519, over 23550.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3876, pruned_loss=0.134, over 5664009.94 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3662, pruned_loss=0.1091, over 5738871.75 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3894, pruned_loss=0.1363, over 5655193.43 frames. ], batch size: 705, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:48:36,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-03 15:48:55,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=284734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:49:09,526 INFO [train.py:968] (1/2) Epoch 7, batch 11100, giga_loss[loss=0.3604, simple_loss=0.3898, pruned_loss=0.1655, over 23457.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3874, pruned_loss=0.135, over 5650289.46 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.366, pruned_loss=0.1091, over 5732735.02 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3896, pruned_loss=0.1374, over 5645980.35 frames. ], batch size: 705, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:49:14,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:49:17,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4035, 2.0173, 1.4829, 0.5743], device='cuda:1'), covar=tensor([0.2335, 0.1382, 0.2126, 0.2996], device='cuda:1'), in_proj_covar=tensor([0.1458, 0.1377, 0.1403, 0.1188], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 15:49:18,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:49:23,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-03 15:49:50,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:50:05,709 INFO [optim.py:369] (1/2) 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 15:50:06,733 INFO [train.py:968] (1/2) Epoch 7, batch 11150, giga_loss[loss=0.3109, simple_loss=0.377, pruned_loss=0.1224, over 28779.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3867, pruned_loss=0.135, over 5648664.29 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.366, pruned_loss=0.1089, over 5735247.89 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3889, pruned_loss=0.1376, over 5641262.34 frames. ], batch size: 174, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:50:19,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1768, 0.9092, 0.8555, 1.3081], device='cuda:1'), covar=tensor([0.0740, 0.0381, 0.0355, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0118, 0.0122, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:1')
+2023-03-03 15:50:37,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5464, 2.3715, 1.8239, 0.8123], device='cuda:1'), covar=tensor([0.3405, 0.1480, 0.2110, 0.3244], device='cuda:1'), in_proj_covar=tensor([0.1468, 0.1386, 0.1416, 0.1198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 15:50:45,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=284836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:50:58,310 INFO [train.py:968] (1/2) Epoch 7, batch 11200, giga_loss[loss=0.272, simple_loss=0.3352, pruned_loss=0.1044, over 28896.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3839, pruned_loss=0.1336, over 5651473.97 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3656, pruned_loss=0.1085, over 5738747.18 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3864, pruned_loss=0.1364, over 5640913.50 frames. ], batch size: 112, lr: 4.70e-03, grad_scale: 8.0
+2023-03-03 15:51:00,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 15:51:25,637 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=284877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:51:27,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=284880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:51:44,291 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 11250, giga_loss[loss=0.3816, simple_loss=0.4099, pruned_loss=0.1767, over 26535.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3837, pruned_loss=0.1339, over 5657130.43 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3653, pruned_loss=0.1084, over 5743172.75 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3864, pruned_loss=0.137, over 5642309.86 frames. ], batch size: 555, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:51:51,452 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=284909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:52:33,084 INFO [train.py:968] (1/2) Epoch 7, batch 11300, giga_loss[loss=0.3703, simple_loss=0.4108, pruned_loss=0.1649, over 28799.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3819, pruned_loss=0.1324, over 5656132.35 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3653, pruned_loss=0.1083, over 5735685.98 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3843, pruned_loss=0.1354, over 5649015.33 frames. ], batch size: 284, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:52:35,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9188, 3.7492, 3.5261, 1.7055], device='cuda:1'), covar=tensor([0.0627, 0.0742, 0.0847, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0892, 0.0810, 0.0622], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 15:53:02,753 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-03 15:53:08,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0315, 3.1359, 2.2945, 0.7102], device='cuda:1'), covar=tensor([0.3785, 0.1525, 0.1989, 0.4429], device='cuda:1'), in_proj_covar=tensor([0.1465, 0.1380, 0.1419, 0.1198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 15:53:25,148 INFO [optim.py:369] (1/2) 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 15:53:25,160 INFO [train.py:968] (1/2) Epoch 7, batch 11350, giga_loss[loss=0.3232, simple_loss=0.387, pruned_loss=0.1297, over 29052.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.383, pruned_loss=0.134, over 5656126.14 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3649, pruned_loss=0.1081, over 5738963.39 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3858, pruned_loss=0.1371, over 5645557.99 frames. ], batch size: 164, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:54:10,246 INFO [train.py:968] (1/2) Epoch 7, batch 11400, giga_loss[loss=0.4209, simple_loss=0.4508, pruned_loss=0.1955, over 28211.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3851, pruned_loss=0.1361, over 5658020.87 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3651, pruned_loss=0.1083, over 5743849.97 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3876, pruned_loss=0.1391, over 5642885.70 frames. ], batch size: 368, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:54:52,477 INFO [train.py:968] (1/2) Epoch 7, batch 11450, giga_loss[loss=0.3541, simple_loss=0.4084, pruned_loss=0.1499, over 28873.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3869, pruned_loss=0.1371, over 5665944.14 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3646, pruned_loss=0.108, over 5741439.00 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3903, pruned_loss=0.1411, over 5651852.60 frames. ], batch size: 186, lr: 4.70e-03, grad_scale: 2.0
+2023-03-03 15:54:53,626 INFO [optim.py:369] (1/2) 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 15:55:40,161 INFO [train.py:968] (1/2) Epoch 7, batch 11500, giga_loss[loss=0.3492, simple_loss=0.4025, pruned_loss=0.148, over 28997.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.387, pruned_loss=0.1382, over 5654999.84 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3641, pruned_loss=0.1077, over 5744947.14 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3906, pruned_loss=0.1421, over 5639346.60 frames. ], batch size: 155, lr: 4.70e-03, grad_scale: 2.0
+2023-03-03 15:55:56,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4554, 1.7390, 1.7812, 1.4471], device='cuda:1'), covar=tensor([0.1327, 0.1862, 0.1061, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0733, 0.0820, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 15:56:26,164 INFO [train.py:968] (1/2) Epoch 7, batch 11550, giga_loss[loss=0.316, simple_loss=0.3833, pruned_loss=0.1244, over 28897.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3872, pruned_loss=0.1384, over 5662870.92 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.364, pruned_loss=0.1078, over 5746339.45 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3909, pruned_loss=0.1425, over 5646360.51 frames. ], batch size: 227, lr: 4.70e-03, grad_scale: 2.0
+2023-03-03 15:56:27,936 INFO [optim.py:369] (1/2) 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 15:56:35,583 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:56:44,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2582, 1.3630, 1.4129, 1.3612], device='cuda:1'), covar=tensor([0.1130, 0.1291, 0.1648, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0736, 0.0653, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 15:57:10,911 INFO [train.py:968] (1/2) Epoch 7, batch 11600, giga_loss[loss=0.3397, simple_loss=0.4039, pruned_loss=0.1378, over 29049.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3879, pruned_loss=0.1384, over 5665781.88 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3642, pruned_loss=0.1077, over 5752748.12 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3919, pruned_loss=0.1432, over 5642754.08 frames. ], batch size: 155, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:57:32,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3136, 1.8279, 1.6646, 1.3043], device='cuda:1'), covar=tensor([0.1548, 0.1975, 0.1219, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0736, 0.0823, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 15:57:54,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 15:58:01,514 INFO [train.py:968] (1/2) Epoch 7, batch 11650, giga_loss[loss=0.3267, simple_loss=0.3821, pruned_loss=0.1356, over 28734.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3882, pruned_loss=0.1378, over 5663741.78 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3643, pruned_loss=0.1077, over 5746755.21 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3917, pruned_loss=0.1422, over 5649282.69 frames. ], batch size: 99, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:58:02,174 INFO [optim.py:369] (1/2) 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 15:58:30,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:58:47,641 INFO [train.py:968] (1/2) Epoch 7, batch 11700, giga_loss[loss=0.3116, simple_loss=0.3752, pruned_loss=0.124, over 28849.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3881, pruned_loss=0.137, over 5670164.18 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3645, pruned_loss=0.1077, over 5750291.56 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3913, pruned_loss=0.1412, over 5653849.83 frames. ], batch size: 199, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:58:51,690 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:58:53,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:59:21,365 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 15:59:33,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6034, 2.2157, 1.4614, 1.3381], device='cuda:1'), covar=tensor([0.1482, 0.0953, 0.1255, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.1526, 0.1393, 0.1341, 0.1441], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 15:59:33,883 INFO [train.py:968] (1/2) Epoch 7, batch 11750, giga_loss[loss=0.3389, simple_loss=0.3956, pruned_loss=0.1411, over 28889.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3893, pruned_loss=0.1381, over 5668394.66 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3642, pruned_loss=0.1074, over 5749140.63 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.393, pruned_loss=0.1428, over 5653158.72 frames. ], batch size: 285, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 15:59:36,300 INFO [optim.py:369] (1/2) 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 16:00:10,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=285435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:00:23,187 INFO [train.py:968] (1/2) Epoch 7, batch 11800, libri_loss[loss=0.2962, simple_loss=0.3796, pruned_loss=0.1064, over 29185.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3914, pruned_loss=0.1401, over 5666040.22 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3645, pruned_loss=0.1074, over 5752114.10 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3946, pruned_loss=0.1445, over 5649634.47 frames. ], batch size: 101, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:00:32,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2976, 1.4998, 1.3442, 1.5230], device='cuda:1'), covar=tensor([0.0613, 0.0285, 0.0284, 0.0639], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0118, 0.0121, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:1')
+2023-03-03 16:00:40,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-03 16:01:05,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4609, 1.7459, 1.8517, 1.4633], device='cuda:1'), covar=tensor([0.1361, 0.1765, 0.1058, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0733, 0.0820, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 16:01:10,633 INFO [train.py:968] (1/2) Epoch 7, batch 11850, giga_loss[loss=0.3039, simple_loss=0.3775, pruned_loss=0.1152, over 28717.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3916, pruned_loss=0.1406, over 5658746.43 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3643, pruned_loss=0.1074, over 5755592.76 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3951, pruned_loss=0.145, over 5639956.40 frames. ], batch size: 60, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:01:11,233 INFO [optim.py:369] (1/2) 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 16:01:17,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2933, 3.1177, 2.8924, 1.4598], device='cuda:1'), covar=tensor([0.0789, 0.0927, 0.1022, 0.2252], device='cuda:1'), in_proj_covar=tensor([0.0951, 0.0901, 0.0813, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 16:01:18,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-03 16:01:58,603 INFO [train.py:968] (1/2) Epoch 7, batch 11900, giga_loss[loss=0.3446, simple_loss=0.4021, pruned_loss=0.1435, over 28920.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3916, pruned_loss=0.1391, over 5662412.26 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3643, pruned_loss=0.1074, over 5758216.11 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3949, pruned_loss=0.1432, over 5643530.77 frames. ], batch size: 213, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:02:32,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:02:45,062 INFO [train.py:968] (1/2) Epoch 7, batch 11950, libri_loss[loss=0.3344, simple_loss=0.4058, pruned_loss=0.1316, over 29080.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3911, pruned_loss=0.1382, over 5664515.00 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3643, pruned_loss=0.1074, over 5758640.83 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3944, pruned_loss=0.1422, over 5646569.16 frames. ], batch size: 101, lr: 4.70e-03, grad_scale: 4.0
+2023-03-03 16:02:45,665 INFO [optim.py:369] (1/2) 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 16:03:27,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6415, 1.9688, 1.9447, 1.5737], device='cuda:1'), covar=tensor([0.1440, 0.1723, 0.1072, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0732, 0.0823, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 16:03:30,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5054, 1.7605, 1.4246, 1.2456], device='cuda:1'), covar=tensor([0.1497, 0.1235, 0.0962, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.1530, 0.1397, 0.1334, 0.1442], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 16:03:32,043 INFO [train.py:968] (1/2) Epoch 7, batch 12000, giga_loss[loss=0.3547, simple_loss=0.4077, pruned_loss=0.1509, over 28952.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3903, pruned_loss=0.1377, over 5656643.76 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3643, pruned_loss=0.1073, over 5759590.28 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3934, pruned_loss=0.1416, over 5639903.61 frames. ], batch size: 186, lr: 4.70e-03, grad_scale: 8.0
+2023-03-03 16:03:32,043 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 16:03:41,419 INFO [train.py:1012] (1/2) Epoch 7, validation: loss=0.2307, simple_loss=0.3348, pruned_loss=0.06333, over 944034.00 frames. 
+2023-03-03 16:03:41,419 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 16:04:28,424 INFO [train.py:968] (1/2) Epoch 7, batch 12050, giga_loss[loss=0.3642, simple_loss=0.4131, pruned_loss=0.1577, over 28676.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3895, pruned_loss=0.1371, over 5661889.71 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3643, pruned_loss=0.1074, over 5751421.74 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3924, pruned_loss=0.1406, over 5653930.44 frames. ], batch size: 262, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:04:29,005 INFO [optim.py:369] (1/2) 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 16:04:37,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:04:55,813 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:04:57,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:05:00,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7330, 1.5962, 1.1833, 1.3490], device='cuda:1'), covar=tensor([0.0627, 0.0653, 0.0896, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0456, 0.0502, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 16:05:17,233 INFO [train.py:968] (1/2) Epoch 7, batch 12100, giga_loss[loss=0.2991, simple_loss=0.3688, pruned_loss=0.1147, over 28914.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3906, pruned_loss=0.1382, over 5644247.23 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3644, pruned_loss=0.1075, over 5749369.08 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.393, pruned_loss=0.1411, over 5638979.67 frames. ], batch size: 174, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:05:26,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:05:30,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:06:03,913 INFO [train.py:968] (1/2) Epoch 7, batch 12150, giga_loss[loss=0.3094, simple_loss=0.3732, pruned_loss=0.1228, over 28675.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3897, pruned_loss=0.1377, over 5659559.32 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3644, pruned_loss=0.1073, over 5752971.52 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3924, pruned_loss=0.141, over 5649468.67 frames. ], batch size: 262, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:06:06,010 INFO [optim.py:369] (1/2) 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 16:06:16,889 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=285810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:06:20,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5386, 2.2004, 1.5197, 0.6079], device='cuda:1'), covar=tensor([0.2581, 0.1415, 0.2505, 0.3215], device='cuda:1'), in_proj_covar=tensor([0.1451, 0.1382, 0.1421, 0.1189], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 16:06:54,813 INFO [train.py:968] (1/2) Epoch 7, batch 12200, giga_loss[loss=0.3706, simple_loss=0.3885, pruned_loss=0.1763, over 23575.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3883, pruned_loss=0.1375, over 5666532.32 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3643, pruned_loss=0.1073, over 5755710.29 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.391, pruned_loss=0.1407, over 5654514.01 frames. ], batch size: 705, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:06:55,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:06:57,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:07:24,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:07:41,485 INFO [train.py:968] (1/2) Epoch 7, batch 12250, giga_loss[loss=0.3041, simple_loss=0.3698, pruned_loss=0.1192, over 29052.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3897, pruned_loss=0.1385, over 5665549.03 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3646, pruned_loss=0.1075, over 5749464.51 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.392, pruned_loss=0.1416, over 5659598.26 frames. ], batch size: 155, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:07:43,468 INFO [optim.py:369] (1/2) 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 16:07:49,176 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:07:51,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:08:15,622 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-03 16:08:19,077 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:08:30,363 INFO [train.py:968] (1/2) Epoch 7, batch 12300, giga_loss[loss=0.4231, simple_loss=0.438, pruned_loss=0.2041, over 26488.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3902, pruned_loss=0.1386, over 5664151.83 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3644, pruned_loss=0.1073, over 5747024.04 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3928, pruned_loss=0.1419, over 5659663.74 frames. ], batch size: 555, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:08:33,784 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=285953.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:08:35,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=285956.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:09:05,125 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=285985.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:09:17,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1432, 1.8929, 1.8365, 1.7010], device='cuda:1'), covar=tensor([0.1222, 0.2322, 0.1826, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0738, 0.0654, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 16:09:18,863 INFO [train.py:968] (1/2) Epoch 7, batch 12350, giga_loss[loss=0.3196, simple_loss=0.3801, pruned_loss=0.1296, over 28596.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3905, pruned_loss=0.1391, over 5666634.28 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3646, pruned_loss=0.1074, over 5749415.19 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3928, pruned_loss=0.1421, over 5659944.08 frames. ], batch size: 78, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:09:21,552 INFO [optim.py:369] (1/2) 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 16:09:36,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:10:05,180 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-03 16:10:09,819 INFO [train.py:968] (1/2) Epoch 7, batch 12400, giga_loss[loss=0.3281, simple_loss=0.3844, pruned_loss=0.1359, over 28915.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3887, pruned_loss=0.137, over 5676014.47 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3644, pruned_loss=0.1072, over 5754125.34 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3914, pruned_loss=0.1404, over 5663863.20 frames. ], batch size: 213, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:10:56,343 INFO [train.py:968] (1/2) Epoch 7, batch 12450, giga_loss[loss=0.3224, simple_loss=0.3845, pruned_loss=0.1302, over 28976.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3893, pruned_loss=0.1369, over 5671212.64 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3649, pruned_loss=0.1076, over 5747774.46 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3914, pruned_loss=0.1398, over 5665876.33 frames. ], batch size: 227, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:10:57,753 INFO [optim.py:369] (1/2) 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 16:11:42,073 INFO [train.py:968] (1/2) Epoch 7, batch 12500, giga_loss[loss=0.3308, simple_loss=0.3983, pruned_loss=0.1316, over 28930.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.388, pruned_loss=0.1351, over 5679256.70 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3651, pruned_loss=0.1076, over 5748955.75 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3899, pruned_loss=0.138, over 5672660.98 frames. ], batch size: 145, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:12:32,983 INFO [train.py:968] (1/2) Epoch 7, batch 12550, giga_loss[loss=0.4167, simple_loss=0.4389, pruned_loss=0.1973, over 26660.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3871, pruned_loss=0.1354, over 5671836.18 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3654, pruned_loss=0.1078, over 5752359.67 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3888, pruned_loss=0.138, over 5662342.55 frames. ], batch size: 555, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:12:37,235 INFO [optim.py:369] (1/2) 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 16:12:53,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5076, 1.8791, 1.2805, 1.2034], device='cuda:1'), covar=tensor([0.1562, 0.1098, 0.1107, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1389, 0.1326, 0.1446], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 16:13:20,393 INFO [train.py:968] (1/2) Epoch 7, batch 12600, giga_loss[loss=0.3171, simple_loss=0.378, pruned_loss=0.1281, over 28645.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3852, pruned_loss=0.1341, over 5670696.25 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3654, pruned_loss=0.1077, over 5754608.89 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3872, pruned_loss=0.1371, over 5658522.39 frames. ], batch size: 242, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:14:07,736 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 16:14:09,571 INFO [train.py:968] (1/2) Epoch 7, batch 12650, giga_loss[loss=0.3087, simple_loss=0.3658, pruned_loss=0.1258, over 28743.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3812, pruned_loss=0.1318, over 5669189.06 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3651, pruned_loss=0.1075, over 5745655.34 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3834, pruned_loss=0.1349, over 5666337.99 frames. ], batch size: 284, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:14:12,634 INFO [optim.py:369] (1/2) 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 16:14:59,888 INFO [train.py:968] (1/2) Epoch 7, batch 12700, giga_loss[loss=0.4089, simple_loss=0.4155, pruned_loss=0.2011, over 23638.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3787, pruned_loss=0.1312, over 5676951.61 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3651, pruned_loss=0.1075, over 5747231.11 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3805, pruned_loss=0.1338, over 5672503.22 frames. ], batch size: 705, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:15:39,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:15:49,436 INFO [train.py:968] (1/2) Epoch 7, batch 12750, giga_loss[loss=0.3651, simple_loss=0.4001, pruned_loss=0.1651, over 26698.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3781, pruned_loss=0.131, over 5688221.98 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3655, pruned_loss=0.1077, over 5750576.88 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3794, pruned_loss=0.1333, over 5680600.68 frames. ], batch size: 555, lr: 4.69e-03, grad_scale: 4.0
+2023-03-03 16:15:51,637 INFO [optim.py:369] (1/2) 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 16:15:58,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2881, 1.2368, 1.1323, 1.3723], device='cuda:1'), covar=tensor([0.0746, 0.0337, 0.0322, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0118, 0.0120, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:1')
+2023-03-03 16:16:03,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-03 16:16:13,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8194, 2.2436, 1.5396, 1.5860], device='cuda:1'), covar=tensor([0.1316, 0.0977, 0.1164, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.1530, 0.1401, 0.1356, 0.1465], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 16:16:35,932 INFO [train.py:968] (1/2) Epoch 7, batch 12800, giga_loss[loss=0.2846, simple_loss=0.3526, pruned_loss=0.1083, over 29052.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3782, pruned_loss=0.1305, over 5677892.86 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3652, pruned_loss=0.1076, over 5745014.80 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.38, pruned_loss=0.1332, over 5675322.60 frames. ], batch size: 128, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:16:54,757 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2016, 1.6693, 1.2746, 0.4413], device='cuda:1'), covar=tensor([0.1822, 0.1290, 0.1718, 0.2580], device='cuda:1'), in_proj_covar=tensor([0.1449, 0.1376, 0.1404, 0.1187], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 16:17:24,385 INFO [train.py:968] (1/2) Epoch 7, batch 12850, libri_loss[loss=0.2842, simple_loss=0.3669, pruned_loss=0.1008, over 29205.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3765, pruned_loss=0.1271, over 5671894.63 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3647, pruned_loss=0.1074, over 5737556.87 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3786, pruned_loss=0.13, over 5674984.04 frames. ], batch size: 101, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:17:27,394 INFO [optim.py:369] (1/2) 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 16:17:29,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 16:17:38,182 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286516.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:17:44,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:17:56,440 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=286533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:18:00,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=286536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:18:14,812 INFO [train.py:968] (1/2) Epoch 7, batch 12900, giga_loss[loss=0.243, simple_loss=0.3005, pruned_loss=0.09275, over 24330.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3745, pruned_loss=0.125, over 5665647.13 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3645, pruned_loss=0.1075, over 5742822.61 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1279, over 5661263.53 frames. ], batch size: 705, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:18:28,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=286565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:19:03,441 INFO [train.py:968] (1/2) Epoch 7, batch 12950, giga_loss[loss=0.2302, simple_loss=0.2939, pruned_loss=0.08322, over 23906.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3707, pruned_loss=0.1211, over 5667850.23 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3639, pruned_loss=0.1073, over 5746340.12 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3733, pruned_loss=0.1241, over 5659123.17 frames. ], batch size: 705, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:19:05,687 INFO [optim.py:369] (1/2) 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 16:19:29,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-03 16:19:55,889 INFO [train.py:968] (1/2) Epoch 7, batch 13000, giga_loss[loss=0.2753, simple_loss=0.3598, pruned_loss=0.09542, over 28918.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3671, pruned_loss=0.1177, over 5669670.30 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3639, pruned_loss=0.1074, over 5750416.05 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5657026.59 frames. ], batch size: 136, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:20:34,765 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:20:47,077 INFO [train.py:968] (1/2) Epoch 7, batch 13050, libri_loss[loss=0.2897, simple_loss=0.3576, pruned_loss=0.1109, over 29542.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3642, pruned_loss=0.1137, over 5670360.39 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3637, pruned_loss=0.1073, over 5752529.54 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3662, pruned_loss=0.116, over 5656810.62 frames. ], batch size: 83, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:20:50,106 INFO [optim.py:369] (1/2) 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 16:21:06,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0414, 1.3852, 1.2678, 1.2726], device='cuda:1'), covar=tensor([0.1145, 0.0992, 0.1329, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0721, 0.0635, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 16:21:23,977 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 13100, giga_loss[loss=0.2988, simple_loss=0.3436, pruned_loss=0.127, over 24201.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3627, pruned_loss=0.1114, over 5664030.23 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3631, pruned_loss=0.1072, over 5753006.10 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3648, pruned_loss=0.1134, over 5651115.43 frames. ], batch size: 705, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:21:48,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4379, 1.6699, 1.7705, 1.5392], device='cuda:1'), covar=tensor([0.1187, 0.1346, 0.1344, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0718, 0.0631, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 16:22:28,262 INFO [train.py:968] (1/2) Epoch 7, batch 13150, giga_loss[loss=0.2798, simple_loss=0.3515, pruned_loss=0.1041, over 28991.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.362, pruned_loss=0.1108, over 5668389.23 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3629, pruned_loss=0.1072, over 5755069.59 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3638, pruned_loss=0.1124, over 5655177.88 frames. ], batch size: 136, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:22:31,576 INFO [optim.py:369] (1/2) 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 16:22:34,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7304, 1.8895, 1.7139, 1.6663], device='cuda:1'), covar=tensor([0.1032, 0.1725, 0.1360, 0.1529], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0716, 0.0631, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 16:23:16,320 INFO [train.py:968] (1/2) Epoch 7, batch 13200, giga_loss[loss=0.2862, simple_loss=0.3577, pruned_loss=0.1073, over 28907.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3594, pruned_loss=0.1091, over 5651290.11 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.363, pruned_loss=0.1074, over 5739052.13 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3607, pruned_loss=0.1103, over 5653509.89 frames. ], batch size: 227, lr: 4.69e-03, grad_scale: 8.0
+2023-03-03 16:23:58,531 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286891.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:24:03,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=286897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:24:05,544 INFO [train.py:968] (1/2) Epoch 7, batch 13250, giga_loss[loss=0.3118, simple_loss=0.377, pruned_loss=0.1233, over 29008.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3561, pruned_loss=0.1068, over 5665988.69 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3619, pruned_loss=0.1069, over 5744367.90 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.358, pruned_loss=0.1081, over 5660768.53 frames. ], batch size: 213, lr: 4.69e-03, grad_scale: 2.0
+2023-03-03 16:24:10,594 INFO [optim.py:369] (1/2) 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 16:24:51,649 INFO [train.py:968] (1/2) Epoch 7, batch 13300, giga_loss[loss=0.3023, simple_loss=0.3721, pruned_loss=0.1163, over 28777.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3563, pruned_loss=0.1067, over 5657883.33 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3616, pruned_loss=0.1068, over 5736227.20 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3579, pruned_loss=0.1079, over 5657712.00 frames. ], batch size: 262, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:25:19,761 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=286981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:25:35,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8199, 4.6161, 4.3984, 2.1267], device='cuda:1'), covar=tensor([0.0515, 0.0750, 0.0844, 0.1936], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0870, 0.0781, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 16:25:37,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2813, 1.7709, 1.6612, 1.3197], device='cuda:1'), covar=tensor([0.1638, 0.2060, 0.1290, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0716, 0.0814, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 16:25:39,939 INFO [train.py:968] (1/2) Epoch 7, batch 13350, giga_loss[loss=0.2523, simple_loss=0.3332, pruned_loss=0.08568, over 28563.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3553, pruned_loss=0.1058, over 5666571.65 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3605, pruned_loss=0.1062, over 5741189.62 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3574, pruned_loss=0.1073, over 5659775.66 frames. ], batch size: 60, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:25:43,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8490, 1.4113, 5.5567, 3.6258], device='cuda:1'), covar=tensor([0.1415, 0.2242, 0.0284, 0.0753], device='cuda:1'), in_proj_covar=tensor([0.0596, 0.0553, 0.0795, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 16:25:45,039 INFO [optim.py:369] (1/2) 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 16:25:56,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2246, 1.8024, 1.6079, 1.2486], device='cuda:1'), covar=tensor([0.1714, 0.2185, 0.1360, 0.1548], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0712, 0.0811, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 16:26:06,653 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287034.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 16:26:08,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287037.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:26:10,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:15,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:23,141 INFO [train.py:968] (1/2) Epoch 7, batch 13400, giga_loss[loss=0.2636, simple_loss=0.3365, pruned_loss=0.09528, over 27571.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3538, pruned_loss=0.1049, over 5670042.99 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3592, pruned_loss=0.1057, over 5743685.49 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3565, pruned_loss=0.1065, over 5657840.13 frames. ], batch size: 472, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:26:37,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:38,616 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:26:41,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:26:45,947 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:27:15,380 INFO [train.py:968] (1/2) Epoch 7, batch 13450, giga_loss[loss=0.2349, simple_loss=0.3216, pruned_loss=0.07409, over 29007.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3511, pruned_loss=0.1028, over 5674788.79 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3592, pruned_loss=0.1059, over 5746459.77 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3531, pruned_loss=0.1039, over 5661489.84 frames. ], batch size: 155, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:27:21,786 INFO [optim.py:369] (1/2) 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 16:27:28,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4922, 2.8972, 1.4532, 1.4787], device='cuda:1'), covar=tensor([0.0806, 0.0281, 0.0846, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0495, 0.0322, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 16:27:31,415 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:28:02,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-03 16:28:05,025 INFO [train.py:968] (1/2) Epoch 7, batch 13500, giga_loss[loss=0.2813, simple_loss=0.3526, pruned_loss=0.105, over 28676.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3476, pruned_loss=0.1011, over 5655883.69 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3587, pruned_loss=0.1057, over 5737532.78 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3493, pruned_loss=0.1021, over 5650948.37 frames. ], batch size: 262, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:28:13,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1547, 1.1452, 3.7956, 3.0482], device='cuda:1'), covar=tensor([0.1557, 0.2438, 0.0365, 0.0901], device='cuda:1'), in_proj_covar=tensor([0.0591, 0.0550, 0.0787, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 16:28:22,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:28:59,095 INFO [train.py:968] (1/2) Epoch 7, batch 13550, libri_loss[loss=0.2241, simple_loss=0.3028, pruned_loss=0.07274, over 29676.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3457, pruned_loss=0.1009, over 5639141.29 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3581, pruned_loss=0.1053, over 5730855.84 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3474, pruned_loss=0.1019, over 5639541.46 frames. ], batch size: 73, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:29:01,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5336, 1.9559, 1.4138, 1.3370], device='cuda:1'), covar=tensor([0.1383, 0.0853, 0.0876, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1348, 0.1288, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 16:29:04,973 INFO [optim.py:369] (1/2) 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 16:29:05,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:29:09,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:29:36,163 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 13600, libri_loss[loss=0.2337, simple_loss=0.3124, pruned_loss=0.07751, over 29647.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3457, pruned_loss=0.1014, over 5636822.48 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3572, pruned_loss=0.1049, over 5725854.82 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3474, pruned_loss=0.1025, over 5637224.17 frames. ], batch size: 73, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:29:51,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:29:55,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:30:27,426 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:30:36,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8799, 1.2575, 1.2871, 1.0706], device='cuda:1'), covar=tensor([0.1270, 0.0944, 0.1603, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0714, 0.0633, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 16:30:41,093 INFO [train.py:968] (1/2) Epoch 7, batch 13650, giga_loss[loss=0.2804, simple_loss=0.3568, pruned_loss=0.102, over 28035.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3477, pruned_loss=0.1022, over 5639688.10 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.357, pruned_loss=0.1049, over 5729536.31 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3491, pruned_loss=0.1029, over 5635016.73 frames. ], batch size: 412, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:30:45,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1617, 1.4860, 1.2699, 0.9700], device='cuda:1'), covar=tensor([0.2286, 0.2057, 0.2278, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.1181, 0.0889, 0.1042, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 16:30:46,406 INFO [optim.py:369] (1/2) 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 16:31:13,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1033, 1.3972, 1.1252, 0.9599], device='cuda:1'), covar=tensor([0.2170, 0.1910, 0.2162, 0.1778], device='cuda:1'), in_proj_covar=tensor([0.1179, 0.0887, 0.1040, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 16:31:35,946 INFO [train.py:968] (1/2) Epoch 7, batch 13700, giga_loss[loss=0.2774, simple_loss=0.3567, pruned_loss=0.09902, over 28967.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3496, pruned_loss=0.1019, over 5648434.95 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3566, pruned_loss=0.1049, over 5734875.54 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3508, pruned_loss=0.1024, over 5637810.00 frames. ], batch size: 285, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:31:45,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:31:45,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2968, 1.4657, 1.6410, 1.4658], device='cuda:1'), covar=tensor([0.0946, 0.0909, 0.0948, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0710, 0.0630, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 16:32:33,380 INFO [train.py:968] (1/2) Epoch 7, batch 13750, giga_loss[loss=0.2767, simple_loss=0.3534, pruned_loss=0.1001, over 28917.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3509, pruned_loss=0.1028, over 5641139.02 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3564, pruned_loss=0.1049, over 5729640.69 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3518, pruned_loss=0.1032, over 5634572.31 frames. ], batch size: 145, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:32:41,476 INFO [optim.py:369] (1/2) 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 16:33:21,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:33:31,587 INFO [train.py:968] (1/2) Epoch 7, batch 13800, giga_loss[loss=0.2447, simple_loss=0.3258, pruned_loss=0.08182, over 28642.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3494, pruned_loss=0.1022, over 5647538.92 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3559, pruned_loss=0.1046, over 5730797.00 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3505, pruned_loss=0.1028, over 5638124.78 frames. ], batch size: 307, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:34:30,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:34:31,143 INFO [train.py:968] (1/2) Epoch 7, batch 13850, giga_loss[loss=0.2573, simple_loss=0.3258, pruned_loss=0.09436, over 24321.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.347, pruned_loss=0.0998, over 5650208.97 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3553, pruned_loss=0.1042, over 5734361.02 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3482, pruned_loss=0.1005, over 5637317.60 frames. ], batch size: 705, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:34:33,418 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:34:48,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4030, 1.9030, 1.7821, 1.4090], device='cuda:1'), covar=tensor([0.1884, 0.2093, 0.1384, 0.1564], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0706, 0.0809, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 16:35:07,909 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:35:16,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=287538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:35:31,513 INFO [train.py:968] (1/2) Epoch 7, batch 13900, giga_loss[loss=0.2601, simple_loss=0.3316, pruned_loss=0.0943, over 27820.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3452, pruned_loss=0.09763, over 5648850.88 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3554, pruned_loss=0.1044, over 5735768.92 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.09794, over 5636755.76 frames. ], batch size: 474, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:36:10,295 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:36:14,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=287585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:36:31,717 INFO [train.py:968] (1/2) Epoch 7, batch 13950, giga_loss[loss=0.2439, simple_loss=0.3117, pruned_loss=0.0881, over 28893.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3424, pruned_loss=0.09754, over 5657479.18 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3544, pruned_loss=0.1042, over 5740605.60 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3435, pruned_loss=0.09772, over 5639855.27 frames. ], batch size: 106, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:36:39,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7901, 2.4692, 1.7737, 0.9327], device='cuda:1'), covar=tensor([0.3832, 0.2060, 0.2187, 0.3651], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1370, 0.1417, 0.1186], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 16:36:39,639 INFO [optim.py:369] (1/2) 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 16:36:48,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:37:32,063 INFO [train.py:968] (1/2) Epoch 7, batch 14000, giga_loss[loss=0.2504, simple_loss=0.3296, pruned_loss=0.08565, over 28850.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3427, pruned_loss=0.09805, over 5661239.77 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3542, pruned_loss=0.104, over 5739427.07 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3435, pruned_loss=0.09818, over 5647055.03 frames. ], batch size: 284, lr: 4.68e-03, grad_scale: 8.0
+2023-03-03 16:38:07,250 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=287681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:38:09,373 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 14050, libri_loss[loss=0.2551, simple_loss=0.3304, pruned_loss=0.08994, over 29522.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3437, pruned_loss=0.09862, over 5674640.64 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3544, pruned_loss=0.1043, over 5743801.74 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3439, pruned_loss=0.0983, over 5657169.20 frames. ], batch size: 80, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:38:36,114 INFO [optim.py:369] (1/2) 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 16:38:44,345 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=287713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:39:04,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-03 16:39:31,290 INFO [train.py:968] (1/2) Epoch 7, batch 14100, giga_loss[loss=0.2796, simple_loss=0.3598, pruned_loss=0.09972, over 28694.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3448, pruned_loss=0.09816, over 5676210.47 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3543, pruned_loss=0.1043, over 5745196.89 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.345, pruned_loss=0.09786, over 5660597.69 frames. ], batch size: 243, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:40:35,649 INFO [train.py:968] (1/2) Epoch 7, batch 14150, giga_loss[loss=0.2524, simple_loss=0.3309, pruned_loss=0.08697, over 28170.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3465, pruned_loss=0.09873, over 5672675.24 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3543, pruned_loss=0.1047, over 5738260.04 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3464, pruned_loss=0.09802, over 5665683.82 frames. ], batch size: 412, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:40:40,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:40:43,204 INFO [optim.py:369] (1/2) 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 16:41:38,367 INFO [train.py:968] (1/2) Epoch 7, batch 14200, giga_loss[loss=0.2724, simple_loss=0.3463, pruned_loss=0.09926, over 28711.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3432, pruned_loss=0.09724, over 5684174.37 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1043, over 5744903.42 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3434, pruned_loss=0.09675, over 5670265.06 frames. ], batch size: 307, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:42:17,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4236, 2.1067, 1.5040, 1.6694], device='cuda:1'), covar=tensor([0.0657, 0.0219, 0.0293, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0116, 0.0122, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0048, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 16:42:33,952 INFO [train.py:968] (1/2) Epoch 7, batch 14250, giga_loss[loss=0.2669, simple_loss=0.3484, pruned_loss=0.0927, over 28790.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3442, pruned_loss=0.09791, over 5683939.72 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1044, over 5748774.29 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.344, pruned_loss=0.09711, over 5666534.64 frames. ], batch size: 174, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:42:45,488 INFO [optim.py:369] (1/2) 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 16:43:18,901 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:43:44,083 INFO [train.py:968] (1/2) Epoch 7, batch 14300, giga_loss[loss=0.2931, simple_loss=0.3781, pruned_loss=0.1041, over 28704.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3485, pruned_loss=0.09896, over 5666785.62 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3536, pruned_loss=0.1044, over 5750845.24 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3482, pruned_loss=0.09822, over 5650160.09 frames. ], batch size: 242, lr: 4.68e-03, grad_scale: 4.0
+2023-03-03 16:43:50,966 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:43:55,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 16:44:19,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=287976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:44:19,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2990, 1.4466, 1.2022, 1.7583], device='cuda:1'), covar=tensor([0.2526, 0.2394, 0.2526, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.1189, 0.0902, 0.1056, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 16:44:50,408 INFO [train.py:968] (1/2) Epoch 7, batch 14350, giga_loss[loss=0.2688, simple_loss=0.3533, pruned_loss=0.09217, over 28449.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3499, pruned_loss=0.09717, over 5665633.67 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1044, over 5751593.01 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3497, pruned_loss=0.0966, over 5651699.62 frames. ], batch size: 369, lr: 4.68e-03, grad_scale: 1.0
+2023-03-03 16:45:00,752 INFO [optim.py:369] (1/2) 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 16:45:49,503 INFO [train.py:968] (1/2) Epoch 7, batch 14400, giga_loss[loss=0.2512, simple_loss=0.3428, pruned_loss=0.07981, over 29104.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3481, pruned_loss=0.09486, over 5654937.97 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3531, pruned_loss=0.1043, over 5746879.81 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3482, pruned_loss=0.09433, over 5644896.70 frames. ], batch size: 285, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:46:44,968 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 16:46:49,267 INFO [train.py:968] (1/2) Epoch 7, batch 14450, giga_loss[loss=0.25, simple_loss=0.3379, pruned_loss=0.08104, over 28381.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3493, pruned_loss=0.09564, over 5657883.34 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3535, pruned_loss=0.1044, over 5740144.75 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.349, pruned_loss=0.09482, over 5653554.60 frames. ], batch size: 368, lr: 4.68e-03, grad_scale: 2.0
+2023-03-03 16:47:02,360 INFO [optim.py:369] (1/2) 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 16:47:08,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:47:24,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9091, 2.2637, 2.1030, 2.0061], device='cuda:1'), covar=tensor([0.0675, 0.0222, 0.0253, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0116, 0.0121, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0048, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 16:47:54,869 INFO [train.py:968] (1/2) Epoch 7, batch 14500, giga_loss[loss=0.3102, simple_loss=0.3813, pruned_loss=0.1196, over 28621.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3503, pruned_loss=0.09792, over 5660704.11 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3532, pruned_loss=0.1043, over 5739027.82 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3502, pruned_loss=0.09729, over 5657268.44 frames. ], batch size: 307, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:48:15,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:48:33,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:48:38,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-03 16:49:05,924 INFO [train.py:968] (1/2) Epoch 7, batch 14550, giga_loss[loss=0.2833, simple_loss=0.3604, pruned_loss=0.1031, over 28732.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3506, pruned_loss=0.0989, over 5657813.48 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3536, pruned_loss=0.1046, over 5737178.73 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3501, pruned_loss=0.09808, over 5656002.29 frames. ], batch size: 119, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:49:19,738 INFO [optim.py:369] (1/2) 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 16:50:24,762 INFO [train.py:968] (1/2) Epoch 7, batch 14600, giga_loss[loss=0.2293, simple_loss=0.3123, pruned_loss=0.0731, over 28658.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3477, pruned_loss=0.09722, over 5674991.19 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3526, pruned_loss=0.104, over 5742492.12 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3481, pruned_loss=0.09696, over 5666282.81 frames. ], batch size: 307, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:50:46,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4690, 2.0859, 1.5553, 0.7169], device='cuda:1'), covar=tensor([0.3052, 0.1613, 0.2513, 0.3396], device='cuda:1'), in_proj_covar=tensor([0.1449, 0.1372, 0.1433, 0.1199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 16:51:15,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8638, 3.6725, 3.5289, 2.0183], device='cuda:1'), covar=tensor([0.0492, 0.0644, 0.0745, 0.1759], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0857, 0.0769, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 16:51:39,845 INFO [train.py:968] (1/2) Epoch 7, batch 14650, giga_loss[loss=0.2665, simple_loss=0.3395, pruned_loss=0.09674, over 27531.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3431, pruned_loss=0.09482, over 5666170.65 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3523, pruned_loss=0.1039, over 5743803.64 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3436, pruned_loss=0.09448, over 5656922.49 frames. ], batch size: 472, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:51:46,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:51:50,926 INFO [optim.py:369] (1/2) 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 16:52:08,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:52:11,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:52:19,040 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 14700, giga_loss[loss=0.2358, simple_loss=0.313, pruned_loss=0.07926, over 28752.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3414, pruned_loss=0.09424, over 5669768.53 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3521, pruned_loss=0.1038, over 5744256.68 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3418, pruned_loss=0.09398, over 5661245.04 frames. ], batch size: 119, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:52:50,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:52:54,302 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:53:06,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5311, 2.0736, 1.6751, 1.6510], device='cuda:1'), covar=tensor([0.0754, 0.0248, 0.0293, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0117, 0.0122, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0048, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 16:53:26,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6062, 1.6331, 1.2450, 1.2366], device='cuda:1'), covar=tensor([0.0667, 0.0464, 0.0932, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0443, 0.0498, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 16:53:45,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:53:49,808 INFO [train.py:968] (1/2) Epoch 7, batch 14750, giga_loss[loss=0.3249, simple_loss=0.3929, pruned_loss=0.1285, over 28626.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3428, pruned_loss=0.09569, over 5685983.76 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3513, pruned_loss=0.1035, over 5751836.46 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3433, pruned_loss=0.09533, over 5669256.08 frames. ], batch size: 242, lr: 4.67e-03, grad_scale: 2.0
+2023-03-03 16:53:58,899 INFO [optim.py:369] (1/2) 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 16:54:10,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9638, 3.7797, 3.5708, 1.6251], device='cuda:1'), covar=tensor([0.0677, 0.0890, 0.1077, 0.2240], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0859, 0.0773, 0.0611], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 16:54:48,088 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:54:50,419 INFO [train.py:968] (1/2) Epoch 7, batch 14800, giga_loss[loss=0.2675, simple_loss=0.3457, pruned_loss=0.09471, over 28986.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.09774, over 5686561.72 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3508, pruned_loss=0.1032, over 5752463.41 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3478, pruned_loss=0.09755, over 5670414.27 frames. ], batch size: 136, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:54:51,450 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:17,976 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:22,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:27,191 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:38,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:46,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:49,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:55:53,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288499.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:55:53,853 INFO [train.py:968] (1/2) Epoch 7, batch 14850, giga_loss[loss=0.288, simple_loss=0.3437, pruned_loss=0.1161, over 28957.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3453, pruned_loss=0.09794, over 5687940.20 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3508, pruned_loss=0.1033, over 5754600.43 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3459, pruned_loss=0.09762, over 5672356.89 frames. ], batch size: 213, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:55:59,091 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-03 16:55:59,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:56:05,557 INFO [optim.py:369] (1/2) 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 16:56:27,175 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288545.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 16:57:01,433 INFO [train.py:968] (1/2) Epoch 7, batch 14900, giga_loss[loss=0.2602, simple_loss=0.3381, pruned_loss=0.09117, over 28982.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3466, pruned_loss=0.09969, over 5675144.13 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3509, pruned_loss=0.1033, over 5754480.42 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3469, pruned_loss=0.09937, over 5662770.12 frames. ], batch size: 155, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:58:03,719 INFO [train.py:968] (1/2) Epoch 7, batch 14950, giga_loss[loss=0.2549, simple_loss=0.3404, pruned_loss=0.08473, over 28998.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3465, pruned_loss=0.09947, over 5672927.67 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3505, pruned_loss=0.1031, over 5755414.56 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.347, pruned_loss=0.09939, over 5661706.84 frames. ], batch size: 145, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:58:06,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0528, 1.4271, 1.1030, 0.1233], device='cuda:1'), covar=tensor([0.1736, 0.1543, 0.2627, 0.3101], device='cuda:1'), in_proj_covar=tensor([0.1429, 0.1356, 0.1421, 0.1189], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 16:58:16,046 INFO [optim.py:369] (1/2) 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 16:58:48,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:58:53,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:58:54,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8690, 5.7144, 5.3466, 2.1779], device='cuda:1'), covar=tensor([0.0324, 0.0445, 0.0614, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0856, 0.0767, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 16:59:15,303 INFO [train.py:968] (1/2) Epoch 7, batch 15000, giga_loss[loss=0.2602, simple_loss=0.3437, pruned_loss=0.08839, over 28831.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3489, pruned_loss=0.09986, over 5673251.01 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3505, pruned_loss=0.1032, over 5757018.01 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3493, pruned_loss=0.09965, over 5662452.54 frames. ], batch size: 99, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 16:59:15,303 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 16:59:23,626 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 16:59:45,442 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 16:59:52,302 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-03 17:00:15,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9438, 1.1776, 0.9518, 0.8114], device='cuda:1'), covar=tensor([0.1161, 0.1112, 0.0689, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.1538, 0.1366, 0.1306, 0.1432], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 17:00:17,798 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:00:21,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:00:40,593 INFO [train.py:968] (1/2) Epoch 7, batch 15050, giga_loss[loss=0.253, simple_loss=0.341, pruned_loss=0.08254, over 28934.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3503, pruned_loss=0.1005, over 5673966.04 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3505, pruned_loss=0.1032, over 5759726.47 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3505, pruned_loss=0.1003, over 5661420.25 frames. ], batch size: 164, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:00:54,510 INFO [optim.py:369] (1/2) 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 17:01:08,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:01:14,209 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:01:20,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-03 17:01:41,833 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:01:57,703 INFO [train.py:968] (1/2) Epoch 7, batch 15100, giga_loss[loss=0.2298, simple_loss=0.3098, pruned_loss=0.0749, over 28770.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.09817, over 5691484.37 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.35, pruned_loss=0.1029, over 5761721.87 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3458, pruned_loss=0.09819, over 5678516.41 frames. ], batch size: 119, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:02:26,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:02:52,384 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=288789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:03:05,571 INFO [train.py:968] (1/2) Epoch 7, batch 15150, giga_loss[loss=0.2427, simple_loss=0.3228, pruned_loss=0.08127, over 28036.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3395, pruned_loss=0.09559, over 5691840.05 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3499, pruned_loss=0.1029, over 5764911.84 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3399, pruned_loss=0.09558, over 5677552.46 frames. ], batch size: 412, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:03:14,277 INFO [optim.py:369] (1/2) 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 17:04:05,840 INFO [train.py:968] (1/2) Epoch 7, batch 15200, giga_loss[loss=0.2624, simple_loss=0.3397, pruned_loss=0.09259, over 28656.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3386, pruned_loss=0.09545, over 5674727.83 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3499, pruned_loss=0.1029, over 5755292.88 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3389, pruned_loss=0.09536, over 5670252.28 frames. ], batch size: 242, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:04:34,659 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288874.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:05:00,415 INFO [train.py:968] (1/2) Epoch 7, batch 15250, giga_loss[loss=0.2842, simple_loss=0.3577, pruned_loss=0.1054, over 28851.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.34, pruned_loss=0.09646, over 5682046.36 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.35, pruned_loss=0.1028, over 5761280.23 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3397, pruned_loss=0.09624, over 5670117.01 frames. ], batch size: 186, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:05:09,889 INFO [optim.py:369] (1/2) 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 17:05:15,358 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=288914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:05:16,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8069, 3.6447, 3.3710, 1.8028], device='cuda:1'), covar=tensor([0.0571, 0.0826, 0.0910, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.0912, 0.0861, 0.0767, 0.0612], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 17:05:17,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=288917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:05:21,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=288920.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:05:56,348 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=288946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:05:59,029 INFO [train.py:968] (1/2) Epoch 7, batch 15300, giga_loss[loss=0.2644, simple_loss=0.3416, pruned_loss=0.09366, over 28925.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.338, pruned_loss=0.09548, over 5666544.86 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3494, pruned_loss=0.1027, over 5762599.38 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.338, pruned_loss=0.09521, over 5654426.71 frames. ], batch size: 136, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:07:02,710 INFO [train.py:968] (1/2) Epoch 7, batch 15350, giga_loss[loss=0.2304, simple_loss=0.3109, pruned_loss=0.07497, over 28407.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3358, pruned_loss=0.09301, over 5669952.21 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.349, pruned_loss=0.1025, over 5762763.88 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3361, pruned_loss=0.09282, over 5658593.45 frames. ], batch size: 71, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:07:06,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:07:15,530 INFO [optim.py:369] (1/2) 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 17:07:24,046 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289017.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:07:27,234 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289020.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:08:12,283 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 15400, giga_loss[loss=0.2643, simple_loss=0.3341, pruned_loss=0.09725, over 28958.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3349, pruned_loss=0.09279, over 5669784.97 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3491, pruned_loss=0.1025, over 5764147.97 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3348, pruned_loss=0.09252, over 5658886.75 frames. ], batch size: 213, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:08:32,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289063.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:08:38,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:08:46,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0217, 1.1245, 3.9048, 3.1558], device='cuda:1'), covar=tensor([0.1707, 0.2597, 0.0342, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0583, 0.0550, 0.0766, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 17:09:13,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7410, 1.7516, 1.2133, 1.4539], device='cuda:1'), covar=tensor([0.0590, 0.0470, 0.0864, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0439, 0.0496, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 17:09:15,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289095.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:09:15,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:09:21,098 INFO [train.py:968] (1/2) Epoch 7, batch 15450, giga_loss[loss=0.3043, simple_loss=0.3729, pruned_loss=0.1179, over 28924.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3362, pruned_loss=0.09317, over 5680619.30 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3496, pruned_loss=0.1029, over 5764558.91 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3356, pruned_loss=0.09249, over 5670867.17 frames. ], batch size: 227, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:09:32,867 INFO [optim.py:369] (1/2) 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 17:09:38,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2706, 1.0181, 0.9179, 1.4234], device='cuda:1'), covar=tensor([0.0754, 0.0310, 0.0340, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0116, 0.0123, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0048, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 17:09:41,123 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 15500, giga_loss[loss=0.2477, simple_loss=0.3264, pruned_loss=0.08449, over 28467.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3367, pruned_loss=0.09308, over 5692904.09 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.349, pruned_loss=0.1026, over 5768950.16 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3363, pruned_loss=0.09249, over 5678817.84 frames. ], batch size: 336, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:10:46,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:11:17,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:11:27,700 INFO [train.py:968] (1/2) Epoch 7, batch 15550, giga_loss[loss=0.2289, simple_loss=0.3134, pruned_loss=0.0722, over 28885.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3383, pruned_loss=0.09487, over 5686861.11 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3486, pruned_loss=0.1024, over 5762317.60 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3382, pruned_loss=0.09447, over 5680435.76 frames. ], batch size: 174, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:11:42,882 INFO [optim.py:369] (1/2) 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 17:12:20,176 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:12:22,082 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 15600, giga_loss[loss=0.2609, simple_loss=0.3438, pruned_loss=0.089, over 28948.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3369, pruned_loss=0.09409, over 5682788.62 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3484, pruned_loss=0.1023, over 5764984.88 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3368, pruned_loss=0.09373, over 5673673.32 frames. ], batch size: 145, lr: 4.67e-03, grad_scale: 8.0
+2023-03-03 17:12:38,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:12:41,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:12:55,722 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:18,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:31,508 INFO [train.py:968] (1/2) Epoch 7, batch 15650, giga_loss[loss=0.2715, simple_loss=0.3232, pruned_loss=0.1099, over 24146.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.338, pruned_loss=0.09336, over 5670264.69 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3482, pruned_loss=0.1023, over 5767639.93 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3378, pruned_loss=0.0929, over 5659164.33 frames. ], batch size: 705, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:13:41,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:45,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:13:46,070 INFO [optim.py:369] (1/2) 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 17:14:21,882 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 15700, giga_loss[loss=0.2466, simple_loss=0.3357, pruned_loss=0.07877, over 28929.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3414, pruned_loss=0.09464, over 5670374.72 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3483, pruned_loss=0.1023, over 5771116.40 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.341, pruned_loss=0.09409, over 5655979.63 frames. ], batch size: 227, lr: 4.67e-03, grad_scale: 4.0
+2023-03-03 17:15:08,582 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:15:35,148 INFO [train.py:968] (1/2) Epoch 7, batch 15750, giga_loss[loss=0.2721, simple_loss=0.3414, pruned_loss=0.1014, over 27810.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3427, pruned_loss=0.09492, over 5670873.14 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3483, pruned_loss=0.1022, over 5772723.08 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3423, pruned_loss=0.09444, over 5656805.64 frames. ], batch size: 476, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:15:46,783 INFO [optim.py:369] (1/2) 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 17:15:51,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 17:16:20,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289439.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:16:32,396 INFO [train.py:968] (1/2) Epoch 7, batch 15800, giga_loss[loss=0.3122, simple_loss=0.3677, pruned_loss=0.1284, over 26721.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3427, pruned_loss=0.09546, over 5649780.50 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3485, pruned_loss=0.1024, over 5763019.10 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3419, pruned_loss=0.0947, over 5644353.06 frames. ], batch size: 555, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:17:22,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4400, 1.7963, 1.4214, 1.6376], device='cuda:1'), covar=tensor([0.0721, 0.0423, 0.0345, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0117, 0.0123, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0048, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 17:17:29,035 INFO [train.py:968] (1/2) Epoch 7, batch 15850, giga_loss[loss=0.2387, simple_loss=0.3201, pruned_loss=0.07868, over 28981.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3401, pruned_loss=0.09432, over 5653265.96 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3481, pruned_loss=0.1022, over 5763013.89 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3398, pruned_loss=0.09374, over 5646478.63 frames. ], batch size: 136, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:17:42,690 INFO [optim.py:369] (1/2) 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 17:17:49,361 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:17:52,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:18:01,561 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=289529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:18:25,623 INFO [train.py:968] (1/2) Epoch 7, batch 15900, giga_loss[loss=0.2289, simple_loss=0.2943, pruned_loss=0.0817, over 24389.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3386, pruned_loss=0.09318, over 5659712.71 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3479, pruned_loss=0.102, over 5766222.61 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3381, pruned_loss=0.09245, over 5646529.60 frames. ], batch size: 705, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:18:32,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:18:45,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:19:25,054 INFO [train.py:968] (1/2) Epoch 7, batch 15950, giga_loss[loss=0.2872, simple_loss=0.362, pruned_loss=0.1061, over 28831.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3382, pruned_loss=0.09386, over 5675771.09 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.348, pruned_loss=0.1022, over 5770509.73 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3375, pruned_loss=0.09285, over 5658669.40 frames. ], batch size: 243, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:19:39,049 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 16000, giga_loss[loss=0.2462, simple_loss=0.3287, pruned_loss=0.08183, over 28619.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3384, pruned_loss=0.0937, over 5668717.86 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3478, pruned_loss=0.1021, over 5762160.94 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3379, pruned_loss=0.0929, over 5661244.86 frames. ], batch size: 307, lr: 4.66e-03, grad_scale: 8.0
+2023-03-03 17:21:28,421 INFO [train.py:968] (1/2) Epoch 7, batch 16050, giga_loss[loss=0.256, simple_loss=0.3379, pruned_loss=0.08709, over 28910.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3398, pruned_loss=0.09403, over 5671728.89 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3475, pruned_loss=0.102, over 5765747.78 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3394, pruned_loss=0.09329, over 5660633.19 frames. ], batch size: 284, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:21:42,511 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:21:44,807 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:22:21,035 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:22:27,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5452, 1.4871, 1.1871, 1.1739], device='cuda:1'), covar=tensor([0.0589, 0.0429, 0.0832, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0435, 0.0496, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 17:22:32,007 INFO [train.py:968] (1/2) Epoch 7, batch 16100, giga_loss[loss=0.2886, simple_loss=0.3607, pruned_loss=0.1083, over 28443.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.341, pruned_loss=0.09568, over 5665126.60 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3473, pruned_loss=0.1018, over 5767516.20 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3406, pruned_loss=0.09505, over 5652017.35 frames. ], batch size: 336, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:23:29,068 INFO [train.py:968] (1/2) Epoch 7, batch 16150, giga_loss[loss=0.2873, simple_loss=0.3703, pruned_loss=0.1021, over 28713.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3437, pruned_loss=0.09731, over 5666296.56 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3464, pruned_loss=0.1013, over 5770981.29 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3442, pruned_loss=0.0972, over 5650456.69 frames. ], batch size: 242, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:23:41,467 INFO [optim.py:369] (1/2) 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 17:23:44,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289814.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:24:23,881 INFO [train.py:968] (1/2) Epoch 7, batch 16200, libri_loss[loss=0.2821, simple_loss=0.3418, pruned_loss=0.1112, over 29554.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3453, pruned_loss=0.09714, over 5665942.42 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3462, pruned_loss=0.1011, over 5773830.75 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3457, pruned_loss=0.09708, over 5647715.42 frames. ], batch size: 78, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:25:19,467 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 16250, giga_loss[loss=0.2568, simple_loss=0.3389, pruned_loss=0.08733, over 28592.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.09667, over 5663199.16 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3458, pruned_loss=0.101, over 5775650.92 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3456, pruned_loss=0.0966, over 5643960.10 frames. ], batch size: 92, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:25:29,762 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=289904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:25:40,389 INFO [optim.py:369] (1/2) 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 17:26:31,061 INFO [train.py:968] (1/2) Epoch 7, batch 16300, giga_loss[loss=0.2428, simple_loss=0.3243, pruned_loss=0.08059, over 28953.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3438, pruned_loss=0.09645, over 5668431.00 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3455, pruned_loss=0.1008, over 5778938.80 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3447, pruned_loss=0.09654, over 5647297.32 frames. ], batch size: 213, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:26:38,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3086, 1.5712, 1.5004, 1.3364], device='cuda:1'), covar=tensor([0.0999, 0.1198, 0.1574, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0712, 0.0626, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 17:26:41,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=289957.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:26:45,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=289960.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:26:45,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-03 17:27:04,200 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 17:27:24,171 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=289989.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:27:38,159 INFO [train.py:968] (1/2) Epoch 7, batch 16350, libri_loss[loss=0.2963, simple_loss=0.3636, pruned_loss=0.1145, over 29537.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3416, pruned_loss=0.09547, over 5675523.32 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.345, pruned_loss=0.1005, over 5782207.37 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09567, over 5653207.83 frames. ], batch size: 84, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:27:53,985 INFO [optim.py:369] (1/2) 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 17:28:38,244 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 16400, giga_loss[loss=0.3106, simple_loss=0.3762, pruned_loss=0.1224, over 28904.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.341, pruned_loss=0.09524, over 5683838.45 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3448, pruned_loss=0.1004, over 5784205.68 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.342, pruned_loss=0.09537, over 5661906.46 frames. ], batch size: 284, lr: 4.66e-03, grad_scale: 8.0
+2023-03-03 17:28:41,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:29:20,002 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 16450, giga_loss[loss=0.2122, simple_loss=0.2988, pruned_loss=0.06282, over 28918.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3404, pruned_loss=0.09622, over 5672059.43 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3445, pruned_loss=0.1003, over 5787510.51 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3413, pruned_loss=0.09631, over 5648356.35 frames. ], batch size: 174, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:29:59,676 INFO [optim.py:369] (1/2) 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 17:30:02,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0911, 1.2809, 3.6527, 3.0443], device='cuda:1'), covar=tensor([0.1613, 0.2332, 0.0424, 0.0723], device='cuda:1'), in_proj_covar=tensor([0.0589, 0.0551, 0.0775, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 17:30:41,529 INFO [train.py:968] (1/2) Epoch 7, batch 16500, giga_loss[loss=0.2337, simple_loss=0.3245, pruned_loss=0.07142, over 28987.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3383, pruned_loss=0.09493, over 5672040.27 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3444, pruned_loss=0.1002, over 5790719.17 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.339, pruned_loss=0.09496, over 5647474.94 frames. ], batch size: 155, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:31:41,337 INFO [train.py:968] (1/2) Epoch 7, batch 16550, giga_loss[loss=0.2647, simple_loss=0.3401, pruned_loss=0.09465, over 28694.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3387, pruned_loss=0.09434, over 5678711.81 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3447, pruned_loss=0.1004, over 5791728.49 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3388, pruned_loss=0.09402, over 5655014.14 frames. ], batch size: 307, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:31:58,576 INFO [optim.py:369] (1/2) 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 17:32:03,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4995, 1.7676, 1.5664, 1.5575], device='cuda:1'), covar=tensor([0.1098, 0.1529, 0.1572, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0714, 0.0629, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 17:32:39,572 INFO [train.py:968] (1/2) Epoch 7, batch 16600, giga_loss[loss=0.2509, simple_loss=0.3402, pruned_loss=0.08078, over 28939.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.337, pruned_loss=0.09141, over 5686066.91 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3446, pruned_loss=0.1004, over 5793114.07 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3371, pruned_loss=0.09107, over 5664632.84 frames. ], batch size: 227, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:33:03,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:33:26,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-03 17:33:34,459 INFO [train.py:968] (1/2) Epoch 7, batch 16650, giga_loss[loss=0.2472, simple_loss=0.3322, pruned_loss=0.08105, over 28896.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3388, pruned_loss=0.09064, over 5692342.53 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3442, pruned_loss=0.1001, over 5792312.94 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.339, pruned_loss=0.09023, over 5671883.40 frames. ], batch size: 186, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:33:49,159 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290313.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:33:49,462 INFO [optim.py:369] (1/2) 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 17:34:31,338 INFO [train.py:968] (1/2) Epoch 7, batch 16700, giga_loss[loss=0.243, simple_loss=0.307, pruned_loss=0.08946, over 24528.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3405, pruned_loss=0.09132, over 5686162.15 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.344, pruned_loss=0.09997, over 5794301.13 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3408, pruned_loss=0.09102, over 5666977.18 frames. ], batch size: 705, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:35:24,614 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:35:36,103 INFO [train.py:968] (1/2) Epoch 7, batch 16750, giga_loss[loss=0.2679, simple_loss=0.3419, pruned_loss=0.0969, over 28716.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3398, pruned_loss=0.09098, over 5678761.70 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.344, pruned_loss=0.09999, over 5795211.27 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.34, pruned_loss=0.09057, over 5660812.31 frames. ], batch size: 99, lr: 4.66e-03, grad_scale: 2.0
+2023-03-03 17:35:59,258 INFO [optim.py:369] (1/2) 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 17:35:59,962 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:03,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:36,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:41,882 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:36:46,084 INFO [train.py:968] (1/2) Epoch 7, batch 16800, giga_loss[loss=0.28, simple_loss=0.3463, pruned_loss=0.1068, over 26977.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3399, pruned_loss=0.09135, over 5663039.43 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3437, pruned_loss=0.09983, over 5796143.39 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3403, pruned_loss=0.09109, over 5646938.98 frames. ], batch size: 555, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:37:22,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1476, 1.4811, 1.1773, 1.0316], device='cuda:1'), covar=tensor([0.2250, 0.2063, 0.2315, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.1168, 0.0879, 0.1037, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 17:37:51,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 17:37:51,974 INFO [train.py:968] (1/2) Epoch 7, batch 16850, giga_loss[loss=0.2587, simple_loss=0.3471, pruned_loss=0.08517, over 28650.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3411, pruned_loss=0.09185, over 5665927.81 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3438, pruned_loss=0.1001, over 5789841.59 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3412, pruned_loss=0.09123, over 5655594.41 frames. ], batch size: 307, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:38:11,792 INFO [optim.py:369] (1/2) 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 17:39:01,851 INFO [train.py:968] (1/2) Epoch 7, batch 16900, giga_loss[loss=0.2514, simple_loss=0.3349, pruned_loss=0.08397, over 28058.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3414, pruned_loss=0.09173, over 5661901.98 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3436, pruned_loss=0.0998, over 5789515.99 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3416, pruned_loss=0.09125, over 5650565.05 frames. ], batch size: 412, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:39:40,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-03 17:40:13,024 INFO [train.py:968] (1/2) Epoch 7, batch 16950, giga_loss[loss=0.2437, simple_loss=0.3351, pruned_loss=0.07615, over 28852.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3454, pruned_loss=0.09337, over 5669072.87 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3437, pruned_loss=0.09984, over 5790858.12 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3454, pruned_loss=0.09289, over 5657750.75 frames. ], batch size: 164, lr: 4.66e-03, grad_scale: 4.0
+2023-03-03 17:40:35,305 INFO [optim.py:369] (1/2) 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 17:41:04,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-03-03 17:41:05,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4472, 2.1285, 1.4314, 0.7219], device='cuda:1'), covar=tensor([0.2835, 0.1539, 0.2680, 0.3260], device='cuda:1'), in_proj_covar=tensor([0.1457, 0.1394, 0.1433, 0.1201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 17:41:26,494 INFO [train.py:968] (1/2) Epoch 7, batch 17000, giga_loss[loss=0.2498, simple_loss=0.3366, pruned_loss=0.08149, over 28167.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3451, pruned_loss=0.09306, over 5675054.41 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.344, pruned_loss=0.1, over 5791390.01 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3449, pruned_loss=0.09248, over 5664575.93 frames. ], batch size: 412, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:41:40,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6720, 1.6192, 1.2110, 1.3010], device='cuda:1'), covar=tensor([0.0700, 0.0640, 0.0999, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0433, 0.0498, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 17:42:01,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3004, 1.5655, 1.2262, 1.2155], device='cuda:1'), covar=tensor([0.1445, 0.0865, 0.0872, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.1518, 0.1323, 0.1284, 0.1398], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 17:42:18,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290688.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:42:28,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6927, 2.4600, 1.6862, 1.4850], device='cuda:1'), covar=tensor([0.2281, 0.0830, 0.1128, 0.1458], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1317, 0.1279, 0.1395], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 17:42:33,581 INFO [train.py:968] (1/2) Epoch 7, batch 17050, giga_loss[loss=0.2142, simple_loss=0.3035, pruned_loss=0.06242, over 28930.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3443, pruned_loss=0.09388, over 5676300.80 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3437, pruned_loss=0.09977, over 5794159.07 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3444, pruned_loss=0.09351, over 5663363.93 frames. ], batch size: 145, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:42:50,878 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=290711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:42:53,126 INFO [optim.py:369] (1/2) 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 17:43:44,669 INFO [train.py:968] (1/2) Epoch 7, batch 17100, libri_loss[loss=0.2875, simple_loss=0.3631, pruned_loss=0.1059, over 29495.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3424, pruned_loss=0.09269, over 5680839.25 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3444, pruned_loss=0.1002, over 5793642.19 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3419, pruned_loss=0.09178, over 5668334.94 frames. ], batch size: 85, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:44:10,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:44:15,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1382, 1.9774, 1.3883, 1.5789], device='cuda:1'), covar=tensor([0.0557, 0.0530, 0.0894, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0430, 0.0494, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 17:44:53,196 INFO [train.py:968] (1/2) Epoch 7, batch 17150, giga_loss[loss=0.2787, simple_loss=0.3555, pruned_loss=0.1009, over 28891.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3402, pruned_loss=0.09108, over 5674335.25 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3441, pruned_loss=0.1001, over 5794385.47 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3399, pruned_loss=0.09034, over 5662209.60 frames. ], batch size: 227, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:45:08,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2413, 1.4833, 1.2287, 1.0641], device='cuda:1'), covar=tensor([0.2191, 0.2024, 0.2282, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.1165, 0.0880, 0.1038, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 17:45:09,436 INFO [optim.py:369] (1/2) 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 17:45:16,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=290818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:45:32,221 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290831.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:45:35,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290834.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 17:45:44,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2296, 1.0756, 4.3916, 3.3661], device='cuda:1'), covar=tensor([0.1554, 0.2587, 0.0298, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0582, 0.0546, 0.0767, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 17:45:52,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-03 17:45:52,861 INFO [train.py:968] (1/2) Epoch 7, batch 17200, libri_loss[loss=0.2227, simple_loss=0.3018, pruned_loss=0.07184, over 29394.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3406, pruned_loss=0.0912, over 5679996.18 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3438, pruned_loss=0.0999, over 5795260.63 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3407, pruned_loss=0.09064, over 5666559.56 frames. ], batch size: 67, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:46:06,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290863.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 17:46:20,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-03 17:46:48,695 INFO [train.py:968] (1/2) Epoch 7, batch 17250, giga_loss[loss=0.2722, simple_loss=0.3532, pruned_loss=0.09566, over 28666.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3442, pruned_loss=0.09361, over 5679069.52 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3437, pruned_loss=0.09978, over 5796807.57 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3443, pruned_loss=0.09306, over 5663702.94 frames. ], batch size: 262, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:46:59,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-03 17:46:59,994 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:47:04,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:47:06,678 INFO [optim.py:369] (1/2) 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 17:47:24,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6485, 1.8303, 1.7271, 1.6505], device='cuda:1'), covar=tensor([0.1130, 0.1901, 0.1538, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0714, 0.0633, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 17:47:34,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:47:46,299 INFO [train.py:968] (1/2) Epoch 7, batch 17300, giga_loss[loss=0.2691, simple_loss=0.3164, pruned_loss=0.1109, over 24363.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3438, pruned_loss=0.09393, over 5678877.96 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3438, pruned_loss=0.09981, over 5797745.38 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3438, pruned_loss=0.09341, over 5664900.68 frames. ], batch size: 705, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:47:50,849 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-03 17:47:57,929 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=290961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:48:01,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=290964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:48:31,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-03 17:48:34,312 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=290993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:48:41,006 INFO [train.py:968] (1/2) Epoch 7, batch 17350, giga_loss[loss=0.2782, simple_loss=0.328, pruned_loss=0.1142, over 24258.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3413, pruned_loss=0.09404, over 5672070.07 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3435, pruned_loss=0.09959, over 5797776.72 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3416, pruned_loss=0.09367, over 5657394.90 frames. ], batch size: 705, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:49:00,293 INFO [optim.py:369] (1/2) 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 17:49:13,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8973, 2.7738, 1.8493, 0.8237], device='cuda:1'), covar=tensor([0.4104, 0.1907, 0.2528, 0.3726], device='cuda:1'), in_proj_covar=tensor([0.1454, 0.1390, 0.1422, 0.1191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 17:49:38,918 INFO [train.py:968] (1/2) Epoch 7, batch 17400, giga_loss[loss=0.2697, simple_loss=0.3488, pruned_loss=0.09536, over 28616.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3412, pruned_loss=0.09483, over 5666956.87 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3433, pruned_loss=0.09937, over 5799485.97 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3415, pruned_loss=0.09466, over 5652033.20 frames. ], batch size: 307, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:50:12,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5117, 1.8099, 1.8288, 1.4428], device='cuda:1'), covar=tensor([0.1518, 0.1982, 0.1192, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0702, 0.0812, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 17:50:16,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=291086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:50:23,434 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-03 17:50:30,721 INFO [train.py:968] (1/2) Epoch 7, batch 17450, giga_loss[loss=0.3159, simple_loss=0.3895, pruned_loss=0.1211, over 28936.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3472, pruned_loss=0.09887, over 5667947.45 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3429, pruned_loss=0.09915, over 5801169.72 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3478, pruned_loss=0.09886, over 5650064.53 frames. ], batch size: 112, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:50:47,693 INFO [optim.py:369] (1/2) 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 17:51:17,521 INFO [train.py:968] (1/2) Epoch 7, batch 17500, giga_loss[loss=0.3157, simple_loss=0.3901, pruned_loss=0.1207, over 28830.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3566, pruned_loss=0.1043, over 5669442.13 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3432, pruned_loss=0.09926, over 5793964.08 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3571, pruned_loss=0.1043, over 5658263.67 frames. ], batch size: 284, lr: 4.65e-03, grad_scale: 2.0
+2023-03-03 17:51:57,886 INFO [train.py:968] (1/2) Epoch 7, batch 17550, giga_loss[loss=0.287, simple_loss=0.3634, pruned_loss=0.1052, over 28855.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.362, pruned_loss=0.1078, over 5666848.63 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3436, pruned_loss=0.09938, over 5781973.81 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3626, pruned_loss=0.1079, over 5664215.51 frames. ], batch size: 285, lr: 4.65e-03, grad_scale: 2.0
+2023-03-03 17:52:14,295 INFO [optim.py:369] (1/2) 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 17:52:24,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=291229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:52:27,360 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 17600, giga_loss[loss=0.2707, simple_loss=0.3303, pruned_loss=0.1055, over 28725.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3617, pruned_loss=0.1092, over 5666001.26 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3434, pruned_loss=0.09927, over 5779012.63 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3626, pruned_loss=0.1095, over 5665093.14 frames. ], batch size: 92, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:52:54,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=291261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:53:26,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4579, 2.2048, 1.5921, 0.5195], device='cuda:1'), covar=tensor([0.2396, 0.1399, 0.2317, 0.2973], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1375, 0.1401, 0.1188], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 17:53:26,385 INFO [train.py:968] (1/2) Epoch 7, batch 17650, libri_loss[loss=0.2409, simple_loss=0.3134, pruned_loss=0.08422, over 29667.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3552, pruned_loss=0.1066, over 5675099.72 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3434, pruned_loss=0.09922, over 5778608.95 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3563, pruned_loss=0.1071, over 5671922.99 frames. ], batch size: 73, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:53:42,353 INFO [optim.py:369] (1/2) 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 17:54:01,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2678, 2.4478, 1.2809, 1.3149], device='cuda:1'), covar=tensor([0.0881, 0.0339, 0.0829, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0480, 0.0316, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 17:54:10,706 INFO [train.py:968] (1/2) Epoch 7, batch 17700, libri_loss[loss=0.3191, simple_loss=0.3798, pruned_loss=0.1292, over 29502.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3482, pruned_loss=0.1037, over 5684804.20 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3437, pruned_loss=0.09934, over 5781563.59 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.349, pruned_loss=0.1041, over 5677054.91 frames. ], batch size: 81, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:54:44,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 17:54:53,405 INFO [train.py:968] (1/2) Epoch 7, batch 17750, giga_loss[loss=0.2427, simple_loss=0.3086, pruned_loss=0.08845, over 28471.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3408, pruned_loss=0.1005, over 5689069.43 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3441, pruned_loss=0.09944, over 5781782.29 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.341, pruned_loss=0.1008, over 5680210.18 frames. ], batch size: 85, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:55:07,328 INFO [optim.py:369] (1/2) 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 17:55:38,134 INFO [train.py:968] (1/2) Epoch 7, batch 17800, giga_loss[loss=0.2444, simple_loss=0.2978, pruned_loss=0.09552, over 23657.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3334, pruned_loss=0.09713, over 5689690.83 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3446, pruned_loss=0.0997, over 5783625.53 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3329, pruned_loss=0.0971, over 5679878.22 frames. ], batch size: 705, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:55:47,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=291462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 17:56:16,497 INFO [train.py:968] (1/2) Epoch 7, batch 17850, giga_loss[loss=0.2678, simple_loss=0.3209, pruned_loss=0.1073, over 28964.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3282, pruned_loss=0.09447, over 5697302.88 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3449, pruned_loss=0.09972, over 5786212.30 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3272, pruned_loss=0.09434, over 5685203.77 frames. ], batch size: 136, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:56:32,754 INFO [optim.py:369] (1/2) 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 17:56:59,368 INFO [train.py:968] (1/2) Epoch 7, batch 17900, giga_loss[loss=0.2384, simple_loss=0.3109, pruned_loss=0.08297, over 28970.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3258, pruned_loss=0.09314, over 5701498.42 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3456, pruned_loss=0.09988, over 5785706.17 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3239, pruned_loss=0.09273, over 5690099.84 frames. ], batch size: 164, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:57:45,904 INFO [train.py:968] (1/2) Epoch 7, batch 17950, giga_loss[loss=0.282, simple_loss=0.3359, pruned_loss=0.114, over 26663.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3219, pruned_loss=0.09143, over 5685414.47 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3456, pruned_loss=0.09987, over 5777632.01 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3203, pruned_loss=0.09105, over 5683372.68 frames. ], batch size: 555, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 17:57:57,514 INFO [optim.py:369] (1/2) 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 17:58:24,953 INFO [train.py:968] (1/2) Epoch 7, batch 18000, libri_loss[loss=0.2984, simple_loss=0.375, pruned_loss=0.1109, over 29549.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3198, pruned_loss=0.09023, over 5682000.06 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3467, pruned_loss=0.1004, over 5768658.29 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3166, pruned_loss=0.08903, over 5684558.44 frames. ], batch size: 83, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:58:24,953 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 17:58:33,310 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 17:59:18,795 INFO [train.py:968] (1/2) Epoch 7, batch 18050, giga_loss[loss=0.2086, simple_loss=0.2843, pruned_loss=0.0664, over 28718.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3156, pruned_loss=0.08785, over 5689185.27 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3468, pruned_loss=0.1004, over 5770173.70 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3126, pruned_loss=0.08682, over 5688967.05 frames. ], batch size: 92, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 17:59:29,034 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 18100, libri_loss[loss=0.2268, simple_loss=0.3103, pruned_loss=0.07166, over 29599.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3135, pruned_loss=0.08707, over 5688580.63 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3468, pruned_loss=0.1002, over 5772780.68 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3101, pruned_loss=0.086, over 5683741.90 frames. ], batch size: 74, lr: 4.65e-03, grad_scale: 8.0
+2023-03-03 18:00:17,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4359, 1.8082, 1.7579, 1.4078], device='cuda:1'), covar=tensor([0.1513, 0.1995, 0.1215, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0712, 0.0823, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 18:00:39,484 INFO [train.py:968] (1/2) Epoch 7, batch 18150, giga_loss[loss=0.2292, simple_loss=0.3065, pruned_loss=0.07601, over 29048.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3112, pruned_loss=0.08545, over 5694478.08 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3474, pruned_loss=0.1002, over 5774262.87 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3068, pruned_loss=0.0842, over 5686633.00 frames. ], batch size: 155, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 18:00:53,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 18:00:54,791 INFO [optim.py:369] (1/2) 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 18:01:06,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 18:01:16,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=291837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:01:29,763 INFO [train.py:968] (1/2) Epoch 7, batch 18200, giga_loss[loss=0.1915, simple_loss=0.2672, pruned_loss=0.05787, over 28563.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3074, pruned_loss=0.08359, over 5700460.48 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3475, pruned_loss=0.1002, over 5774872.56 frames. ], giga_tot_loss[loss=0.2344, simple_loss=0.3036, pruned_loss=0.08254, over 5693399.10 frames. ], batch size: 307, lr: 4.65e-03, grad_scale: 4.0
+2023-03-03 18:01:32,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-03 18:02:13,134 INFO [train.py:968] (1/2) Epoch 7, batch 18250, giga_loss[loss=0.2884, simple_loss=0.3462, pruned_loss=0.1153, over 28606.00 frames. ], tot_loss[loss=0.2378, simple_loss=0.3071, pruned_loss=0.08425, over 5693886.62 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3477, pruned_loss=0.1003, over 5768385.51 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3033, pruned_loss=0.08303, over 5692112.16 frames. ], batch size: 262, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:02:28,348 INFO [optim.py:369] (1/2) 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 18:03:03,352 INFO [train.py:968] (1/2) Epoch 7, batch 18300, giga_loss[loss=0.2978, simple_loss=0.3682, pruned_loss=0.1137, over 28727.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3175, pruned_loss=0.09008, over 5693694.47 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3478, pruned_loss=0.1003, over 5770960.81 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3138, pruned_loss=0.08889, over 5688539.71 frames. ], batch size: 99, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:03:31,922 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=291980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:03:36,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=291983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:03:51,572 INFO [train.py:968] (1/2) Epoch 7, batch 18350, giga_loss[loss=0.2713, simple_loss=0.3565, pruned_loss=0.09299, over 28923.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3322, pruned_loss=0.09825, over 5692579.95 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3475, pruned_loss=0.1001, over 5772187.64 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3293, pruned_loss=0.09747, over 5686654.91 frames. ], batch size: 174, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:04:01,205 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:04:05,853 INFO [optim.py:369] (1/2) 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 18:04:17,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-03 18:04:29,937 INFO [train.py:968] (1/2) Epoch 7, batch 18400, giga_loss[loss=0.3199, simple_loss=0.3781, pruned_loss=0.1309, over 28800.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3437, pruned_loss=0.1038, over 5703717.71 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.348, pruned_loss=0.1003, over 5773542.44 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3408, pruned_loss=0.1031, over 5696421.70 frames. ], batch size: 99, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:04:52,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8358, 2.3307, 2.1855, 1.7601], device='cuda:1'), covar=tensor([0.1615, 0.1639, 0.1126, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0706, 0.0816, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 18:05:16,060 INFO [train.py:968] (1/2) Epoch 7, batch 18450, giga_loss[loss=0.2819, simple_loss=0.3606, pruned_loss=0.1016, over 29003.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3509, pruned_loss=0.1069, over 5697590.06 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.348, pruned_loss=0.1003, over 5774222.73 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3487, pruned_loss=0.1063, over 5690853.85 frames. ], batch size: 155, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:05:29,464 INFO [optim.py:369] (1/2) 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 18:05:48,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3706, 1.5423, 1.3255, 1.7156], device='cuda:1'), covar=tensor([0.2371, 0.2263, 0.2333, 0.2048], device='cuda:1'), in_proj_covar=tensor([0.1180, 0.0898, 0.1046, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:05:54,964 INFO [train.py:968] (1/2) Epoch 7, batch 18500, libri_loss[loss=0.2475, simple_loss=0.3182, pruned_loss=0.08842, over 29650.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3528, pruned_loss=0.1062, over 5698046.48 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3479, pruned_loss=0.1002, over 5774946.28 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3512, pruned_loss=0.1059, over 5689833.75 frames. ], batch size: 69, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:06:13,937 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=292169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:06:41,260 INFO [train.py:968] (1/2) Epoch 7, batch 18550, giga_loss[loss=0.2858, simple_loss=0.354, pruned_loss=0.1088, over 28573.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3547, pruned_loss=0.1066, over 5692741.44 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3483, pruned_loss=0.1004, over 5775220.87 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3532, pruned_loss=0.1065, over 5683197.82 frames. ], batch size: 85, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:06:57,529 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 18600, giga_loss[loss=0.2981, simple_loss=0.3576, pruned_loss=0.1193, over 28473.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3573, pruned_loss=0.1088, over 5694789.33 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3488, pruned_loss=0.1008, over 5778021.46 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3559, pruned_loss=0.1085, over 5682843.47 frames. ], batch size: 71, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:07:28,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-03 18:07:42,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3736, 1.5760, 1.5692, 1.5489], device='cuda:1'), covar=tensor([0.1121, 0.1241, 0.1239, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0723, 0.0641, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 18:08:09,559 INFO [train.py:968] (1/2) Epoch 7, batch 18650, giga_loss[loss=0.3366, simple_loss=0.3923, pruned_loss=0.1404, over 27584.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3602, pruned_loss=0.1108, over 5694605.40 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3487, pruned_loss=0.1007, over 5772120.32 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3594, pruned_loss=0.1109, over 5688309.06 frames. ], batch size: 472, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:08:25,007 INFO [optim.py:369] (1/2) 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 18:08:54,141 INFO [train.py:968] (1/2) Epoch 7, batch 18700, giga_loss[loss=0.2794, simple_loss=0.3631, pruned_loss=0.09788, over 29008.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3636, pruned_loss=0.113, over 5695148.52 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3489, pruned_loss=0.1008, over 5773002.72 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3629, pruned_loss=0.113, over 5689162.19 frames. ], batch size: 155, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:09:34,540 INFO [train.py:968] (1/2) Epoch 7, batch 18750, giga_loss[loss=0.2871, simple_loss=0.3567, pruned_loss=0.1088, over 28562.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3662, pruned_loss=0.1133, over 5702321.62 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3489, pruned_loss=0.1007, over 5773378.19 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3658, pruned_loss=0.1136, over 5696551.31 frames. ], batch size: 92, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:09:49,711 INFO [optim.py:369] (1/2) 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 18:10:15,147 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 18800, giga_loss[loss=0.3203, simple_loss=0.3891, pruned_loss=0.1257, over 28224.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3673, pruned_loss=0.1132, over 5700891.44 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3492, pruned_loss=0.1009, over 5771885.07 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3669, pruned_loss=0.1133, over 5696714.14 frames. ], batch size: 368, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:10:58,496 INFO [train.py:968] (1/2) Epoch 7, batch 18850, giga_loss[loss=0.2879, simple_loss=0.3704, pruned_loss=0.1027, over 29083.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3691, pruned_loss=0.1138, over 5699901.13 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3498, pruned_loss=0.1011, over 5773736.93 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3687, pruned_loss=0.114, over 5693100.09 frames. ], batch size: 136, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:10:58,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0881, 2.9409, 1.9403, 1.5000], device='cuda:1'), covar=tensor([0.1520, 0.0692, 0.0940, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.1523, 0.1354, 0.1317, 0.1430], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 18:11:08,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1966, 1.4266, 1.1782, 1.4158], device='cuda:1'), covar=tensor([0.2326, 0.2114, 0.2338, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.1180, 0.0894, 0.1044, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:11:10,504 INFO [optim.py:369] (1/2) 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 18:11:30,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=292544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:11:34,682 INFO [train.py:968] (1/2) Epoch 7, batch 18900, giga_loss[loss=0.2584, simple_loss=0.3397, pruned_loss=0.08855, over 28589.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3683, pruned_loss=0.112, over 5696805.76 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3505, pruned_loss=0.1014, over 5765943.50 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3679, pruned_loss=0.1122, over 5697121.32 frames. ], batch size: 336, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:12:16,326 INFO [train.py:968] (1/2) Epoch 7, batch 18950, giga_loss[loss=0.2806, simple_loss=0.3611, pruned_loss=0.1001, over 28821.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3661, pruned_loss=0.1096, over 5699270.44 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3508, pruned_loss=0.1015, over 5763849.16 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3657, pruned_loss=0.1097, over 5700573.41 frames. ], batch size: 199, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:12:33,305 INFO [optim.py:369] (1/2) 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 18:12:56,937 INFO [train.py:968] (1/2) Epoch 7, batch 19000, giga_loss[loss=0.3352, simple_loss=0.3978, pruned_loss=0.1363, over 28930.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3668, pruned_loss=0.1099, over 5707790.59 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3517, pruned_loss=0.1019, over 5768881.95 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3662, pruned_loss=0.1099, over 5701791.44 frames. ], batch size: 199, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:13:31,357 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=292687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:13:33,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=292690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:13:42,404 INFO [train.py:968] (1/2) Epoch 7, batch 19050, libri_loss[loss=0.3117, simple_loss=0.3942, pruned_loss=0.1146, over 25745.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3685, pruned_loss=0.1135, over 5694863.29 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3518, pruned_loss=0.1019, over 5769222.31 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3683, pruned_loss=0.1138, over 5687943.17 frames. ], batch size: 136, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:13:58,557 INFO [optim.py:369] (1/2) 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 18:14:00,210 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:14:07,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5673, 1.8336, 1.8492, 1.4896], device='cuda:1'), covar=tensor([0.1423, 0.1915, 0.1124, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0704, 0.0815, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 18:14:27,254 INFO [train.py:968] (1/2) Epoch 7, batch 19100, giga_loss[loss=0.2719, simple_loss=0.3459, pruned_loss=0.09897, over 28613.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3709, pruned_loss=0.1175, over 5691006.90 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3519, pruned_loss=0.102, over 5771371.97 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3711, pruned_loss=0.118, over 5681914.14 frames. ], batch size: 262, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:14:34,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3804, 2.9043, 2.0132, 1.7693], device='cuda:1'), covar=tensor([0.1359, 0.0942, 0.1139, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.1541, 0.1371, 0.1334, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 18:15:05,067 INFO [train.py:968] (1/2) Epoch 7, batch 19150, giga_loss[loss=0.4102, simple_loss=0.4417, pruned_loss=0.1893, over 28022.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3704, pruned_loss=0.1177, over 5701448.63 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3525, pruned_loss=0.1022, over 5773848.44 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3706, pruned_loss=0.1185, over 5689405.03 frames. ], batch size: 412, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:15:14,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2289, 4.0009, 3.8246, 1.9271], device='cuda:1'), covar=tensor([0.0549, 0.0709, 0.0817, 0.2029], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0853, 0.0769, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 18:15:21,529 INFO [optim.py:369] (1/2) 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 18:15:24,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=292822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:15:37,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3132, 1.4014, 1.4578, 1.3785], device='cuda:1'), covar=tensor([0.1152, 0.1378, 0.1696, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0725, 0.0642, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 18:15:48,840 INFO [train.py:968] (1/2) Epoch 7, batch 19200, giga_loss[loss=0.3361, simple_loss=0.3923, pruned_loss=0.1399, over 27935.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.368, pruned_loss=0.1167, over 5705548.30 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3528, pruned_loss=0.1023, over 5777121.18 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3683, pruned_loss=0.1176, over 5691382.10 frames. ], batch size: 412, lr: 4.64e-03, grad_scale: 8.0
+2023-03-03 18:15:59,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1940, 0.9284, 0.9356, 1.3993], device='cuda:1'), covar=tensor([0.0748, 0.0334, 0.0329, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0120, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:1')
+2023-03-03 18:15:59,192 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-03 18:16:32,656 INFO [train.py:968] (1/2) Epoch 7, batch 19250, giga_loss[loss=0.2954, simple_loss=0.3705, pruned_loss=0.1101, over 28186.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3678, pruned_loss=0.1168, over 5701346.42 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3531, pruned_loss=0.1024, over 5780161.39 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3681, pruned_loss=0.1178, over 5685992.34 frames. ], batch size: 77, lr: 4.64e-03, grad_scale: 4.0
+2023-03-03 18:16:39,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5600, 3.3373, 1.6156, 1.5721], device='cuda:1'), covar=tensor([0.0883, 0.0216, 0.0785, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0482, 0.0315, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 18:16:44,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9627, 1.1705, 3.6733, 3.2046], device='cuda:1'), covar=tensor([0.1685, 0.2385, 0.0374, 0.0634], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0547, 0.0775, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:16:50,250 INFO [optim.py:369] (1/2) 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 18:17:13,134 INFO [train.py:968] (1/2) Epoch 7, batch 19300, giga_loss[loss=0.3267, simple_loss=0.3792, pruned_loss=0.1371, over 26658.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3668, pruned_loss=0.1155, over 5699843.35 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3533, pruned_loss=0.1023, over 5782527.87 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3673, pruned_loss=0.1168, over 5682751.88 frames. ], batch size: 555, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:17:26,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=292965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:17:28,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=292968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:17:35,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7919, 1.7201, 1.7130, 1.5936], device='cuda:1'), covar=tensor([0.1315, 0.2152, 0.1857, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0728, 0.0645, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 18:17:54,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=292997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:17:56,379 INFO [train.py:968] (1/2) Epoch 7, batch 19350, giga_loss[loss=0.2548, simple_loss=0.336, pruned_loss=0.08676, over 28921.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3633, pruned_loss=0.1123, over 5705202.15 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3533, pruned_loss=0.1022, over 5786713.78 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3641, pruned_loss=0.1138, over 5685266.49 frames. ], batch size: 174, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:18:08,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8470, 1.8466, 1.2353, 1.5428], device='cuda:1'), covar=tensor([0.0689, 0.0645, 0.1026, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0435, 0.0496, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 18:18:13,883 INFO [optim.py:369] (1/2) 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 18:18:40,292 INFO [train.py:968] (1/2) Epoch 7, batch 19400, giga_loss[loss=0.2718, simple_loss=0.3389, pruned_loss=0.1024, over 28861.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3591, pruned_loss=0.1096, over 5698199.87 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3538, pruned_loss=0.1025, over 5786906.70 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3595, pruned_loss=0.1109, over 5678646.27 frames. ], batch size: 199, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:18:41,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 18:19:15,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0483, 1.1960, 3.4894, 2.9975], device='cuda:1'), covar=tensor([0.1442, 0.2336, 0.0362, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0550, 0.0778, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:19:28,278 INFO [train.py:968] (1/2) Epoch 7, batch 19450, giga_loss[loss=0.311, simple_loss=0.358, pruned_loss=0.132, over 26499.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3526, pruned_loss=0.1064, over 5693438.49 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3539, pruned_loss=0.1025, over 5786075.00 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3529, pruned_loss=0.1074, over 5677862.58 frames. ], batch size: 555, lr: 4.64e-03, grad_scale: 2.0
+2023-03-03 18:19:35,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 18:19:46,857 INFO [optim.py:369] (1/2) 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 18:20:18,627 INFO [train.py:968] (1/2) Epoch 7, batch 19500, giga_loss[loss=0.2403, simple_loss=0.3151, pruned_loss=0.08277, over 28691.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3469, pruned_loss=0.1035, over 5693096.93 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3543, pruned_loss=0.1027, over 5787336.53 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3467, pruned_loss=0.1041, over 5679075.33 frames. ], batch size: 242, lr: 4.63e-03, grad_scale: 2.0
+2023-03-03 18:20:49,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4619, 1.9254, 1.2708, 1.1752], device='cuda:1'), covar=tensor([0.1708, 0.1132, 0.1338, 0.1655], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1346, 0.1318, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 18:21:05,936 INFO [train.py:968] (1/2) Epoch 7, batch 19550, giga_loss[loss=0.2814, simple_loss=0.3516, pruned_loss=0.1056, over 27918.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3468, pruned_loss=0.1032, over 5691997.38 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3546, pruned_loss=0.103, over 5786073.34 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3462, pruned_loss=0.1034, over 5679893.38 frames. ], batch size: 412, lr: 4.63e-03, grad_scale: 2.0
+2023-03-03 18:21:21,299 INFO [optim.py:369] (1/2) 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 18:21:45,746 INFO [train.py:968] (1/2) Epoch 7, batch 19600, giga_loss[loss=0.2633, simple_loss=0.341, pruned_loss=0.09278, over 28936.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3475, pruned_loss=0.1027, over 5697528.18 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3553, pruned_loss=0.1033, over 5777007.71 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3462, pruned_loss=0.1027, over 5692571.20 frames. ], batch size: 213, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:22:28,394 INFO [train.py:968] (1/2) Epoch 7, batch 19650, libri_loss[loss=0.3423, simple_loss=0.4109, pruned_loss=0.1369, over 29530.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3471, pruned_loss=0.1028, over 5701191.99 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3559, pruned_loss=0.1037, over 5779855.48 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3453, pruned_loss=0.1024, over 5692633.68 frames. ], batch size: 89, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:22:39,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7714, 4.6366, 4.4179, 2.1218], device='cuda:1'), covar=tensor([0.0426, 0.0482, 0.0556, 0.1958], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0846, 0.0761, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 18:22:44,828 INFO [optim.py:369] (1/2) 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 18:23:10,104 INFO [train.py:968] (1/2) Epoch 7, batch 19700, giga_loss[loss=0.2319, simple_loss=0.3149, pruned_loss=0.07444, over 29050.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3442, pruned_loss=0.101, over 5698094.24 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3563, pruned_loss=0.1038, over 5763529.08 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3422, pruned_loss=0.1006, over 5703761.43 frames. ], batch size: 164, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:23:16,761 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:23:51,253 INFO [train.py:968] (1/2) Epoch 7, batch 19750, giga_loss[loss=0.3256, simple_loss=0.3749, pruned_loss=0.1382, over 26694.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3414, pruned_loss=0.1, over 5708240.50 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3565, pruned_loss=0.1038, over 5764864.37 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3395, pruned_loss=0.09957, over 5710848.80 frames. ], batch size: 555, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:24:08,823 INFO [optim.py:369] (1/2) 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 18:24:18,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5967, 2.1087, 1.6700, 1.8243], device='cuda:1'), covar=tensor([0.0729, 0.0255, 0.0281, 0.0722], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0119, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0048, 0.0043, 0.0073], device='cuda:1')
+2023-03-03 18:24:24,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-03 18:24:30,052 INFO [train.py:968] (1/2) Epoch 7, batch 19800, giga_loss[loss=0.2336, simple_loss=0.3072, pruned_loss=0.07998, over 28921.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3394, pruned_loss=0.09889, over 5716364.16 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3575, pruned_loss=0.1041, over 5769047.30 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3365, pruned_loss=0.09816, over 5713174.36 frames. ], batch size: 227, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:24:33,596 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-03 18:24:49,977 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:25:11,538 INFO [train.py:968] (1/2) Epoch 7, batch 19850, libri_loss[loss=0.2739, simple_loss=0.3679, pruned_loss=0.08991, over 29644.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3373, pruned_loss=0.09766, over 5720749.43 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3582, pruned_loss=0.1042, over 5770255.52 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3339, pruned_loss=0.09691, over 5716059.26 frames. ], batch size: 88, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:25:18,262 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:25:31,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9723, 1.0876, 0.8902, 0.5957], device='cuda:1'), covar=tensor([0.1247, 0.1262, 0.0818, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.1528, 0.1356, 0.1336, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 18:25:38,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-03 18:25:45,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6215, 2.3440, 2.5020, 2.2215], device='cuda:1'), covar=tensor([0.1147, 0.1831, 0.1439, 0.1670], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0734, 0.0650, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 18:25:50,323 INFO [train.py:968] (1/2) Epoch 7, batch 19900, giga_loss[loss=0.2177, simple_loss=0.2965, pruned_loss=0.06946, over 28688.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3361, pruned_loss=0.09727, over 5716767.32 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.359, pruned_loss=0.1044, over 5768402.08 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3323, pruned_loss=0.09635, over 5713197.02 frames. ], batch size: 262, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:25:58,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-03 18:26:28,456 INFO [train.py:968] (1/2) Epoch 7, batch 19950, giga_loss[loss=0.3572, simple_loss=0.3975, pruned_loss=0.1584, over 27620.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3359, pruned_loss=0.09766, over 5713183.42 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3596, pruned_loss=0.1046, over 5763580.08 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3315, pruned_loss=0.09655, over 5712628.58 frames. ], batch size: 472, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:26:46,562 INFO [optim.py:369] (1/2) 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 18:27:06,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6383, 2.0556, 1.3864, 1.2512], device='cuda:1'), covar=tensor([0.1714, 0.1079, 0.1250, 0.1626], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1345, 0.1327, 0.1446], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 18:27:12,407 INFO [train.py:968] (1/2) Epoch 7, batch 20000, giga_loss[loss=0.2966, simple_loss=0.3497, pruned_loss=0.1218, over 28877.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3327, pruned_loss=0.09606, over 5719715.59 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3596, pruned_loss=0.1045, over 5764292.18 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3292, pruned_loss=0.09522, over 5718489.12 frames. ], batch size: 186, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:27:50,631 INFO [train.py:968] (1/2) Epoch 7, batch 20050, giga_loss[loss=0.2348, simple_loss=0.3093, pruned_loss=0.0801, over 28973.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3308, pruned_loss=0.09513, over 5716270.01 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3603, pruned_loss=0.1048, over 5757467.25 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3269, pruned_loss=0.09403, over 5721564.40 frames. ], batch size: 227, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:28:04,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-03 18:28:04,489 INFO [optim.py:369] (1/2) 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 18:28:15,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:28:20,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-03 18:28:27,336 INFO [train.py:968] (1/2) Epoch 7, batch 20100, giga_loss[loss=0.2362, simple_loss=0.3124, pruned_loss=0.08003, over 29040.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3299, pruned_loss=0.09443, over 5728135.23 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3607, pruned_loss=0.1049, over 5760908.16 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3257, pruned_loss=0.09325, over 5728455.75 frames. ], batch size: 136, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:28:30,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:28:32,332 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=293757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:29:11,799 INFO [train.py:968] (1/2) Epoch 7, batch 20150, giga_loss[loss=0.2786, simple_loss=0.3464, pruned_loss=0.1054, over 28842.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3331, pruned_loss=0.0968, over 5722064.80 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3608, pruned_loss=0.105, over 5762437.16 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3295, pruned_loss=0.09573, over 5720728.14 frames. ], batch size: 112, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:29:29,691 INFO [optim.py:369] (1/2) 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 18:29:59,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=293849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:29:59,530 INFO [train.py:968] (1/2) Epoch 7, batch 20200, giga_loss[loss=0.3717, simple_loss=0.4176, pruned_loss=0.1629, over 27638.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3405, pruned_loss=0.1012, over 5719861.55 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3614, pruned_loss=0.1051, over 5765950.10 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3365, pruned_loss=0.1001, over 5714726.02 frames. ], batch size: 472, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:30:27,394 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:30:33,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:30:37,994 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 20250, giga_loss[loss=0.3245, simple_loss=0.3841, pruned_loss=0.1324, over 28845.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3515, pruned_loss=0.1091, over 5700733.13 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3617, pruned_loss=0.1052, over 5767432.32 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3479, pruned_loss=0.1081, over 5695100.60 frames. ], batch size: 99, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:31:00,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=293910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:31:10,244 INFO [optim.py:369] (1/2) 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 18:31:32,841 INFO [train.py:968] (1/2) Epoch 7, batch 20300, giga_loss[loss=0.2647, simple_loss=0.3489, pruned_loss=0.09022, over 28876.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3565, pruned_loss=0.1114, over 5700443.20 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3612, pruned_loss=0.1047, over 5770625.79 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3539, pruned_loss=0.1113, over 5690756.30 frames. ], batch size: 174, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:32:11,702 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=293992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:16,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=293995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:19,881 INFO [train.py:968] (1/2) Epoch 7, batch 20350, giga_loss[loss=0.3087, simple_loss=0.3875, pruned_loss=0.115, over 28710.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3606, pruned_loss=0.1128, over 5686654.85 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3609, pruned_loss=0.1045, over 5774101.89 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3586, pruned_loss=0.113, over 5674470.69 frames. ], batch size: 262, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:32:21,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5997, 2.0708, 1.9326, 1.5805], device='cuda:1'), covar=tensor([0.1468, 0.1927, 0.1235, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0710, 0.0816, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 18:32:36,710 INFO [optim.py:369] (1/2) 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 18:32:42,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:45,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:32:47,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:33:06,638 INFO [train.py:968] (1/2) Epoch 7, batch 20400, libri_loss[loss=0.2643, simple_loss=0.3413, pruned_loss=0.09369, over 29667.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3649, pruned_loss=0.115, over 5681383.27 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3607, pruned_loss=0.1043, over 5775394.24 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1156, over 5668524.29 frames. ], batch size: 73, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:33:15,336 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:33:28,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 18:33:47,078 INFO [train.py:968] (1/2) Epoch 7, batch 20450, libri_loss[loss=0.3065, simple_loss=0.3784, pruned_loss=0.1173, over 29676.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3701, pruned_loss=0.1189, over 5678358.49 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3609, pruned_loss=0.1045, over 5775952.93 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.369, pruned_loss=0.1194, over 5665945.66 frames. ], batch size: 88, lr: 4.63e-03, grad_scale: 8.0
+2023-03-03 18:34:05,911 INFO [optim.py:369] (1/2) 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 18:34:12,063 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:34:14,700 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 20500, giga_loss[loss=0.2981, simple_loss=0.363, pruned_loss=0.1167, over 28744.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3641, pruned_loss=0.1143, over 5661622.31 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3615, pruned_loss=0.1051, over 5748660.80 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3628, pruned_loss=0.1143, over 5674573.99 frames. ], batch size: 284, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:35:10,713 INFO [train.py:968] (1/2) Epoch 7, batch 20550, giga_loss[loss=0.2812, simple_loss=0.3584, pruned_loss=0.102, over 28883.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3619, pruned_loss=0.1119, over 5680016.86 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3617, pruned_loss=0.1051, over 5752205.67 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3608, pruned_loss=0.1121, over 5685431.28 frames. ], batch size: 186, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:35:22,119 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-03 18:35:28,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9898, 2.4564, 0.9821, 1.1802], device='cuda:1'), covar=tensor([0.1112, 0.0484, 0.0973, 0.1453], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0471, 0.0310, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0026, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 18:35:29,983 INFO [optim.py:369] (1/2) 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 18:35:50,910 INFO [train.py:968] (1/2) Epoch 7, batch 20600, giga_loss[loss=0.2876, simple_loss=0.3548, pruned_loss=0.1102, over 28788.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3619, pruned_loss=0.1114, over 5688170.10 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3624, pruned_loss=0.1056, over 5757454.65 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3603, pruned_loss=0.1113, over 5685349.52 frames. ], batch size: 99, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:36:09,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:11,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-03 18:36:11,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:11,649 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:15,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:29,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5291, 1.7586, 1.8900, 1.4593], device='cuda:1'), covar=tensor([0.1565, 0.2046, 0.1203, 0.1373], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0711, 0.0817, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 18:36:30,265 INFO [train.py:968] (1/2) Epoch 7, batch 20650, giga_loss[loss=0.3031, simple_loss=0.3752, pruned_loss=0.1155, over 28977.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3635, pruned_loss=0.112, over 5692231.71 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3625, pruned_loss=0.1059, over 5761120.04 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3622, pruned_loss=0.112, over 5684041.02 frames. ], batch size: 136, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:36:33,047 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:34,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:36:50,143 INFO [optim.py:369] (1/2) 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 18:37:02,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-03 18:37:12,806 INFO [train.py:968] (1/2) Epoch 7, batch 20700, giga_loss[loss=0.2994, simple_loss=0.3669, pruned_loss=0.116, over 28883.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3668, pruned_loss=0.1142, over 5698299.28 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3636, pruned_loss=0.1065, over 5763630.78 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3648, pruned_loss=0.1138, over 5686727.57 frames. ], batch size: 112, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:37:18,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-03 18:37:20,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3960, 1.5660, 1.2858, 1.0305], device='cuda:1'), covar=tensor([0.1350, 0.1208, 0.1044, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1351, 0.1343, 0.1433], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 18:37:32,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0933, 1.2520, 0.9612, 0.9348], device='cuda:1'), covar=tensor([0.0960, 0.0966, 0.0809, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1351, 0.1343, 0.1434], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 18:37:40,827 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:37:56,112 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 18:37:56,323 INFO [train.py:968] (1/2) Epoch 7, batch 20750, giga_loss[loss=0.2756, simple_loss=0.3588, pruned_loss=0.09621, over 28906.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3678, pruned_loss=0.1151, over 5708798.62 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3636, pruned_loss=0.1065, over 5765091.81 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3663, pruned_loss=0.1149, over 5698018.85 frames. ], batch size: 174, lr: 4.63e-03, grad_scale: 4.0
+2023-03-03 18:38:18,823 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 20800, giga_loss[loss=0.3726, simple_loss=0.415, pruned_loss=0.1651, over 27545.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3687, pruned_loss=0.1166, over 5685506.40 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3637, pruned_loss=0.1066, over 5762033.35 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3674, pruned_loss=0.1164, over 5679334.70 frames. ], batch size: 472, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:39:03,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1601, 1.5304, 1.5181, 1.1771], device='cuda:1'), covar=tensor([0.1191, 0.1684, 0.0955, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0716, 0.0816, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 18:39:24,221 INFO [train.py:968] (1/2) Epoch 7, batch 20850, giga_loss[loss=0.2974, simple_loss=0.3658, pruned_loss=0.1145, over 28848.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3709, pruned_loss=0.1189, over 5693302.85 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3643, pruned_loss=0.1072, over 5765772.79 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3695, pruned_loss=0.1186, over 5682685.17 frames. ], batch size: 186, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:39:28,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5800, 2.0460, 1.8592, 1.8897], device='cuda:1'), covar=tensor([0.0729, 0.0257, 0.0262, 0.0745], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0120, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0048, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 18:39:43,350 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 20900, giga_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09381, over 28765.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3698, pruned_loss=0.1176, over 5693901.76 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3647, pruned_loss=0.1075, over 5757061.79 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3686, pruned_loss=0.1174, over 5690752.58 frames. ], batch size: 119, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:40:08,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 18:40:13,085 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:40:13,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8493, 1.2645, 3.6576, 2.8815], device='cuda:1'), covar=tensor([0.1790, 0.2342, 0.0397, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0595, 0.0552, 0.0776, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:40:20,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2007, 1.2038, 1.0298, 0.9356], device='cuda:1'), covar=tensor([0.0587, 0.0427, 0.0885, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0437, 0.0498, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 18:40:41,575 INFO [train.py:968] (1/2) Epoch 7, batch 20950, giga_loss[loss=0.2593, simple_loss=0.3388, pruned_loss=0.08987, over 28436.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3696, pruned_loss=0.1167, over 5697487.47 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3658, pruned_loss=0.1085, over 5760748.89 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3678, pruned_loss=0.116, over 5689903.38 frames. ], batch size: 71, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:41:00,416 INFO [optim.py:369] (1/2) 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 18:41:21,876 INFO [train.py:968] (1/2) Epoch 7, batch 21000, giga_loss[loss=0.3009, simple_loss=0.3697, pruned_loss=0.1161, over 28663.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3702, pruned_loss=0.1155, over 5704804.40 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3661, pruned_loss=0.1089, over 5765225.38 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3685, pruned_loss=0.1148, over 5693168.76 frames. ], batch size: 85, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:41:21,877 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 18:41:30,269 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 18:42:08,111 INFO [train.py:968] (1/2) Epoch 7, batch 21050, giga_loss[loss=0.2488, simple_loss=0.3335, pruned_loss=0.08207, over 29088.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3695, pruned_loss=0.1149, over 5689650.36 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3663, pruned_loss=0.1091, over 5750270.32 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3681, pruned_loss=0.1142, over 5692742.94 frames. ], batch size: 155, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:42:12,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4471, 1.5909, 1.3383, 1.6903], device='cuda:1'), covar=tensor([0.2072, 0.1849, 0.1860, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.1186, 0.0899, 0.1045, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:42:15,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-03 18:42:25,832 INFO [optim.py:369] (1/2) 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 18:42:34,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:42:45,886 INFO [train.py:968] (1/2) Epoch 7, batch 21100, giga_loss[loss=0.2553, simple_loss=0.3412, pruned_loss=0.08469, over 28940.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3654, pruned_loss=0.1122, over 5701753.51 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3661, pruned_loss=0.109, over 5751684.55 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3645, pruned_loss=0.1118, over 5702513.38 frames. ], batch size: 145, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:42:50,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:43:24,422 INFO [train.py:968] (1/2) Epoch 7, batch 21150, libri_loss[loss=0.2901, simple_loss=0.3524, pruned_loss=0.1139, over 29626.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3635, pruned_loss=0.1113, over 5697374.83 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3664, pruned_loss=0.1094, over 5743318.81 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3624, pruned_loss=0.1107, over 5704895.04 frames. ], batch size: 73, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:43:37,186 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:43:42,171 INFO [optim.py:369] (1/2) 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 18:44:02,697 INFO [train.py:968] (1/2) Epoch 7, batch 21200, libri_loss[loss=0.3181, simple_loss=0.3648, pruned_loss=0.1357, over 29511.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3627, pruned_loss=0.1115, over 5703834.49 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3668, pruned_loss=0.11, over 5748049.11 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3613, pruned_loss=0.1105, over 5704495.21 frames. ], batch size: 70, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:44:07,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7925, 1.2665, 5.2105, 3.5548], device='cuda:1'), covar=tensor([0.1506, 0.2552, 0.0280, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0589, 0.0550, 0.0774, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:44:29,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:44:33,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7069, 1.6126, 1.2265, 1.2709], device='cuda:1'), covar=tensor([0.0646, 0.0552, 0.0938, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0335, 0.0429, 0.0492, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 18:44:41,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=294898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:44:41,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-03 18:44:42,666 INFO [train.py:968] (1/2) Epoch 7, batch 21250, giga_loss[loss=0.3147, simple_loss=0.3768, pruned_loss=0.1263, over 28710.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3636, pruned_loss=0.1125, over 5703734.59 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3673, pruned_loss=0.1105, over 5749268.04 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.362, pruned_loss=0.1113, over 5702236.27 frames. ], batch size: 60, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:44:43,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=294901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:45:03,863 INFO [optim.py:369] (1/2) 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 18:45:07,261 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=294930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:45:13,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=294938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:45:21,258 INFO [train.py:968] (1/2) Epoch 7, batch 21300, giga_loss[loss=0.2971, simple_loss=0.3691, pruned_loss=0.1126, over 28764.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3632, pruned_loss=0.1118, over 5717218.66 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3672, pruned_loss=0.1106, over 5752494.07 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.362, pruned_loss=0.1108, over 5711610.85 frames. ], batch size: 119, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:45:33,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=294964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:46:03,815 INFO [train.py:968] (1/2) Epoch 7, batch 21350, libri_loss[loss=0.3109, simple_loss=0.3839, pruned_loss=0.119, over 29652.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3633, pruned_loss=0.1112, over 5708747.47 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3678, pruned_loss=0.1112, over 5754920.17 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3617, pruned_loss=0.1099, over 5701373.76 frames. ], batch size: 88, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:46:12,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-03 18:46:23,888 INFO [optim.py:369] (1/2) 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 18:46:45,020 INFO [train.py:968] (1/2) Epoch 7, batch 21400, giga_loss[loss=0.2681, simple_loss=0.35, pruned_loss=0.09311, over 28755.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3626, pruned_loss=0.1101, over 5722471.79 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3687, pruned_loss=0.112, over 5759238.89 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3603, pruned_loss=0.1083, over 5711218.21 frames. ], batch size: 242, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:46:46,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3543, 2.1142, 1.6192, 0.5786], device='cuda:1'), covar=tensor([0.2929, 0.1416, 0.2138, 0.3211], device='cuda:1'), in_proj_covar=tensor([0.1415, 0.1324, 0.1388, 0.1169], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 18:47:09,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:47:12,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:47:19,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2678, 1.6249, 1.5809, 1.2224], device='cuda:1'), covar=tensor([0.1427, 0.1982, 0.1137, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0712, 0.0814, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 18:47:23,267 INFO [train.py:968] (1/2) Epoch 7, batch 21450, giga_loss[loss=0.2933, simple_loss=0.3645, pruned_loss=0.1111, over 28538.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3627, pruned_loss=0.1104, over 5729502.13 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3694, pruned_loss=0.1128, over 5762821.36 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3601, pruned_loss=0.1082, over 5716618.10 frames. ], batch size: 85, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:47:30,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:47:31,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-03 18:47:36,841 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:47:57,978 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-03 18:47:59,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4302, 1.2195, 4.9015, 3.7981], device='cuda:1'), covar=tensor([0.2149, 0.2982, 0.0510, 0.0710], device='cuda:1'), in_proj_covar=tensor([0.0586, 0.0544, 0.0765, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:48:05,235 INFO [train.py:968] (1/2) Epoch 7, batch 21500, giga_loss[loss=0.2427, simple_loss=0.3271, pruned_loss=0.07918, over 28838.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3603, pruned_loss=0.1094, over 5734094.66 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3697, pruned_loss=0.1131, over 5766874.95 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3577, pruned_loss=0.1073, over 5719302.53 frames. ], batch size: 174, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:48:36,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:48:43,397 INFO [train.py:968] (1/2) Epoch 7, batch 21550, giga_loss[loss=0.2421, simple_loss=0.3196, pruned_loss=0.08231, over 28461.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3569, pruned_loss=0.1077, over 5712564.68 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3699, pruned_loss=0.1134, over 5750859.62 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3544, pruned_loss=0.1057, over 5714544.84 frames. ], batch size: 65, lr: 4.62e-03, grad_scale: 2.0
+2023-03-03 18:48:47,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0041, 1.1999, 3.7798, 3.0922], device='cuda:1'), covar=tensor([0.1614, 0.2364, 0.0355, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0590, 0.0547, 0.0770, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 18:49:05,445 INFO [optim.py:369] (1/2) 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 18:49:25,417 INFO [train.py:968] (1/2) Epoch 7, batch 21600, giga_loss[loss=0.2653, simple_loss=0.3412, pruned_loss=0.09469, over 28836.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3567, pruned_loss=0.1081, over 5719144.30 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3701, pruned_loss=0.1137, over 5752776.17 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3544, pruned_loss=0.1062, over 5718626.13 frames. ], batch size: 186, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:49:25,690 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:49:27,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:49:29,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:49:37,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1747, 1.8140, 1.4930, 0.3584], device='cuda:1'), covar=tensor([0.2577, 0.1466, 0.2299, 0.3103], device='cuda:1'), in_proj_covar=tensor([0.1412, 0.1324, 0.1389, 0.1163], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 18:49:50,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:50:03,797 INFO [train.py:968] (1/2) Epoch 7, batch 21650, giga_loss[loss=0.2264, simple_loss=0.308, pruned_loss=0.07238, over 28506.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3565, pruned_loss=0.109, over 5718465.23 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3702, pruned_loss=0.1141, over 5756065.71 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3543, pruned_loss=0.1069, over 5714386.42 frames. ], batch size: 60, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:50:25,847 INFO [optim.py:369] (1/2) 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 18:50:31,321 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:50:34,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:50:45,392 INFO [train.py:968] (1/2) Epoch 7, batch 21700, giga_loss[loss=0.2872, simple_loss=0.3508, pruned_loss=0.1117, over 28311.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3558, pruned_loss=0.1096, over 5712428.53 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3705, pruned_loss=0.1145, over 5748758.61 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3536, pruned_loss=0.1076, over 5715337.29 frames. ], batch size: 368, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:50:56,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:09,806 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:22,187 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:22,581 INFO [train.py:968] (1/2) Epoch 7, batch 21750, giga_loss[loss=0.3516, simple_loss=0.404, pruned_loss=0.1496, over 28075.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3547, pruned_loss=0.1095, over 5708210.27 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3706, pruned_loss=0.1148, over 5743476.75 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3524, pruned_loss=0.1075, over 5713770.62 frames. ], batch size: 412, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:51:24,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:51:24,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6233, 1.5913, 1.1182, 1.3096], device='cuda:1'), covar=tensor([0.0600, 0.0546, 0.0939, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0431, 0.0491, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 18:51:42,382 INFO [optim.py:369] (1/2) 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 18:51:48,739 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:52:01,907 INFO [train.py:968] (1/2) Epoch 7, batch 21800, giga_loss[loss=0.282, simple_loss=0.3489, pruned_loss=0.1075, over 29067.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3518, pruned_loss=0.1085, over 5700782.73 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3714, pruned_loss=0.1155, over 5736531.56 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3489, pruned_loss=0.1062, over 5710091.76 frames. ], batch size: 155, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:52:10,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:52:25,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:52:29,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:52:41,780 INFO [train.py:968] (1/2) Epoch 7, batch 21850, giga_loss[loss=0.2288, simple_loss=0.3052, pruned_loss=0.0762, over 28338.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3502, pruned_loss=0.1077, over 5700343.29 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3727, pruned_loss=0.1167, over 5737676.77 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3462, pruned_loss=0.1045, over 5705625.49 frames. ], batch size: 77, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:52:52,084 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:53:02,741 INFO [optim.py:369] (1/2) 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 18:53:23,781 INFO [train.py:968] (1/2) Epoch 7, batch 21900, giga_loss[loss=0.2805, simple_loss=0.3573, pruned_loss=0.1018, over 28704.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.35, pruned_loss=0.1073, over 5699966.75 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3724, pruned_loss=0.1167, over 5740173.74 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3468, pruned_loss=0.1046, over 5701644.73 frames. ], batch size: 262, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:54:06,773 INFO [train.py:968] (1/2) Epoch 7, batch 21950, giga_loss[loss=0.3379, simple_loss=0.3986, pruned_loss=0.1386, over 28773.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3528, pruned_loss=0.108, over 5703642.56 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3728, pruned_loss=0.117, over 5741424.90 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3497, pruned_loss=0.1056, over 5703497.30 frames. ], batch size: 284, lr: 4.62e-03, grad_scale: 4.0
+2023-03-03 18:54:25,421 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:54:30,719 INFO [optim.py:369] (1/2) 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 18:54:37,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3517, 1.4526, 1.4966, 1.3848], device='cuda:1'), covar=tensor([0.1103, 0.1483, 0.1451, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0724, 0.0635, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 18:54:51,397 INFO [train.py:968] (1/2) Epoch 7, batch 22000, giga_loss[loss=0.2731, simple_loss=0.3477, pruned_loss=0.09921, over 29059.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3555, pruned_loss=0.1091, over 5713176.29 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3733, pruned_loss=0.1177, over 5745226.02 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3519, pruned_loss=0.1063, over 5708457.77 frames. ], batch size: 128, lr: 4.62e-03, grad_scale: 8.0
+2023-03-03 18:55:10,789 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:55:33,992 INFO [train.py:968] (1/2) Epoch 7, batch 22050, giga_loss[loss=0.2597, simple_loss=0.3482, pruned_loss=0.08557, over 28945.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3586, pruned_loss=0.1105, over 5707710.59 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3738, pruned_loss=0.1183, over 5747571.09 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3549, pruned_loss=0.1074, over 5700897.87 frames. ], batch size: 186, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:55:54,598 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:55:55,056 INFO [optim.py:369] (1/2) 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 18:56:14,303 INFO [train.py:968] (1/2) Epoch 7, batch 22100, giga_loss[loss=0.2649, simple_loss=0.3458, pruned_loss=0.09195, over 28656.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3578, pruned_loss=0.1095, over 5699891.32 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3745, pruned_loss=0.1191, over 5747684.37 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3539, pruned_loss=0.1062, over 5693379.06 frames. ], batch size: 262, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:56:16,046 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:56:22,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:56:58,855 INFO [train.py:968] (1/2) Epoch 7, batch 22150, giga_loss[loss=0.2927, simple_loss=0.359, pruned_loss=0.1132, over 28894.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3581, pruned_loss=0.1095, over 5705029.56 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.375, pruned_loss=0.1195, over 5751435.23 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3541, pruned_loss=0.1062, over 5695406.79 frames. ], batch size: 199, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:57:23,651 INFO [optim.py:369] (1/2) 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 18:57:29,678 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 22200, giga_loss[loss=0.2777, simple_loss=0.3439, pruned_loss=0.1058, over 28798.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3592, pruned_loss=0.1104, over 5707233.98 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3754, pruned_loss=0.1198, over 5753203.82 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3556, pruned_loss=0.1075, over 5697810.76 frames. ], batch size: 99, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:57:49,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5753, 4.3821, 4.1533, 1.7542], device='cuda:1'), covar=tensor([0.0409, 0.0513, 0.0628, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0852, 0.0770, 0.0613], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 18:58:00,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9337, 1.2800, 1.2275, 1.1105], device='cuda:1'), covar=tensor([0.1364, 0.1119, 0.1793, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0733, 0.0644, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 18:58:25,863 INFO [train.py:968] (1/2) Epoch 7, batch 22250, giga_loss[loss=0.3654, simple_loss=0.4151, pruned_loss=0.1578, over 27616.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3598, pruned_loss=0.1107, over 5709577.83 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3756, pruned_loss=0.12, over 5753987.80 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3568, pruned_loss=0.1083, over 5701413.73 frames. ], batch size: 472, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:58:27,882 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:58:29,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:58:38,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-03 18:58:45,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-03 18:58:46,757 INFO [optim.py:369] (1/2) 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 18:58:51,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=295932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:07,366 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:07,880 INFO [train.py:968] (1/2) Epoch 7, batch 22300, giga_loss[loss=0.2913, simple_loss=0.3584, pruned_loss=0.1121, over 28473.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3624, pruned_loss=0.1124, over 5697865.59 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3764, pruned_loss=0.1207, over 5749922.92 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3587, pruned_loss=0.1095, over 5694095.75 frames. ], batch size: 78, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:59:08,709 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=295951.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 18:59:32,711 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=295979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:35,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=295982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:42,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=295993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:48,663 INFO [train.py:968] (1/2) Epoch 7, batch 22350, giga_loss[loss=0.3064, simple_loss=0.3778, pruned_loss=0.1175, over 29105.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3644, pruned_loss=0.1132, over 5703280.04 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3771, pruned_loss=0.1211, over 5750054.98 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3607, pruned_loss=0.1104, over 5699579.53 frames. ], batch size: 155, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 18:59:55,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-03 18:59:56,301 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 18:59:58,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 19:00:05,275 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296022.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:00:08,866 INFO [optim.py:369] (1/2) 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 19:00:22,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:00:23,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7770, 2.2669, 1.5844, 1.3424], device='cuda:1'), covar=tensor([0.1584, 0.1012, 0.1419, 0.1524], device='cuda:1'), in_proj_covar=tensor([0.1520, 0.1361, 0.1351, 0.1434], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 19:00:25,463 INFO [train.py:968] (1/2) Epoch 7, batch 22400, giga_loss[loss=0.3269, simple_loss=0.3898, pruned_loss=0.132, over 27597.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.365, pruned_loss=0.1132, over 5713165.98 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3781, pruned_loss=0.1221, over 5754523.06 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3606, pruned_loss=0.1098, over 5704672.10 frames. ], batch size: 472, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:00:28,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2831, 1.5012, 1.2181, 1.0821], device='cuda:1'), covar=tensor([0.1463, 0.1047, 0.1034, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.1521, 0.1361, 0.1354, 0.1435], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 19:00:48,014 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 22450, giga_loss[loss=0.2824, simple_loss=0.3589, pruned_loss=0.1029, over 28836.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3654, pruned_loss=0.1132, over 5714844.05 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3783, pruned_loss=0.1223, over 5755100.41 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3614, pruned_loss=0.1101, over 5706471.05 frames. ], batch size: 227, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:01:07,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:29,325 INFO [optim.py:369] (1/2) 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 19:01:30,544 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:38,200 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:40,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:01:49,808 INFO [train.py:968] (1/2) Epoch 7, batch 22500, giga_loss[loss=0.2943, simple_loss=0.3599, pruned_loss=0.1143, over 28783.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3663, pruned_loss=0.1141, over 5715971.04 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3786, pruned_loss=0.1224, over 5754305.84 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3629, pruned_loss=0.1115, over 5709686.05 frames. ], batch size: 119, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:02:04,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:22,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:25,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:30,677 INFO [train.py:968] (1/2) Epoch 7, batch 22550, giga_loss[loss=0.2601, simple_loss=0.3314, pruned_loss=0.09434, over 28318.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3645, pruned_loss=0.1129, over 5715893.58 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3784, pruned_loss=0.1224, over 5758810.78 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3616, pruned_loss=0.1106, over 5705995.46 frames. ], batch size: 71, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:02:39,236 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:41,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-03 19:02:43,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0885, 1.6107, 1.4402, 1.0934], device='cuda:1'), covar=tensor([0.1204, 0.1929, 0.1094, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0706, 0.0812, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 19:02:44,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9949, 1.1250, 0.9602, 0.8445], device='cuda:1'), covar=tensor([0.1113, 0.1146, 0.0684, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.1517, 0.1361, 0.1358, 0.1437], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 19:02:45,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:02:46,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2247, 4.0107, 3.8165, 1.8378], device='cuda:1'), covar=tensor([0.0557, 0.0660, 0.0776, 0.2066], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0863, 0.0780, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 19:02:48,224 INFO [optim.py:369] (1/2) 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 19:02:54,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4699, 1.6918, 1.8063, 1.4549], device='cuda:1'), covar=tensor([0.1591, 0.1840, 0.1242, 0.1363], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0707, 0.0813, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 19:03:04,296 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:07,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:09,288 INFO [train.py:968] (1/2) Epoch 7, batch 22600, libri_loss[loss=0.2861, simple_loss=0.3357, pruned_loss=0.1182, over 28651.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3626, pruned_loss=0.1121, over 5721382.23 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3795, pruned_loss=0.1234, over 5762358.43 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3586, pruned_loss=0.1089, over 5708662.51 frames. ], batch size: 63, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:03:21,862 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:24,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:26,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:28,833 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:03:43,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2508, 2.4830, 1.3327, 1.3164], device='cuda:1'), covar=tensor([0.0846, 0.0334, 0.0855, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0484, 0.0316, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 19:03:48,927 INFO [train.py:968] (1/2) Epoch 7, batch 22650, giga_loss[loss=0.2569, simple_loss=0.3268, pruned_loss=0.09356, over 28495.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3585, pruned_loss=0.11, over 5712160.01 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.38, pruned_loss=0.1239, over 5755455.19 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3545, pruned_loss=0.1067, over 5706554.28 frames. ], batch size: 85, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:03:50,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:04:07,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:04:08,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:04:09,240 INFO [optim.py:369] (1/2) 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 19:04:27,909 INFO [train.py:968] (1/2) Epoch 7, batch 22700, giga_loss[loss=0.3085, simple_loss=0.3859, pruned_loss=0.1155, over 28924.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3571, pruned_loss=0.1087, over 5712517.47 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3801, pruned_loss=0.1243, over 5759071.85 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3532, pruned_loss=0.1053, over 5703726.84 frames. ], batch size: 227, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:05:11,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296397.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:05:14,593 INFO [train.py:968] (1/2) Epoch 7, batch 22750, giga_loss[loss=0.2993, simple_loss=0.369, pruned_loss=0.1148, over 28956.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3587, pruned_loss=0.108, over 5709030.32 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3801, pruned_loss=0.1244, over 5760371.24 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3554, pruned_loss=0.1051, over 5700451.83 frames. ], batch size: 213, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:05:30,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-03 19:05:37,695 INFO [optim.py:369] (1/2) 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 19:05:42,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3113, 1.5839, 1.2808, 1.3049], device='cuda:1'), covar=tensor([0.2421, 0.2248, 0.2477, 0.1967], device='cuda:1'), in_proj_covar=tensor([0.1179, 0.0889, 0.1037, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:05:48,238 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 22800, giga_loss[loss=0.3338, simple_loss=0.3947, pruned_loss=0.1365, over 27611.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3598, pruned_loss=0.1083, over 5706550.41 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3803, pruned_loss=0.1246, over 5763488.44 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3566, pruned_loss=0.1054, over 5695883.51 frames. ], batch size: 472, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:05:56,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:01,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3754, 3.1124, 1.4967, 1.3564], device='cuda:1'), covar=tensor([0.0860, 0.0342, 0.0821, 0.1302], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0487, 0.0317, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 19:06:07,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:08,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296469.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:06:09,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:10,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296472.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:06:33,483 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:06:33,959 INFO [train.py:968] (1/2) Epoch 7, batch 22850, giga_loss[loss=0.295, simple_loss=0.3638, pruned_loss=0.1131, over 28590.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3565, pruned_loss=0.1076, over 5704655.37 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.38, pruned_loss=0.1245, over 5763391.48 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3538, pruned_loss=0.1051, over 5695320.57 frames. ], batch size: 307, lr: 4.61e-03, grad_scale: 8.0
+2023-03-03 19:06:35,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:07:08,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296540.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:07:09,967 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:07:14,961 INFO [train.py:968] (1/2) Epoch 7, batch 22900, giga_loss[loss=0.3489, simple_loss=0.3948, pruned_loss=0.1515, over 28996.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3566, pruned_loss=0.1094, over 5698939.10 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3803, pruned_loss=0.1249, over 5753248.58 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3533, pruned_loss=0.1065, over 5698596.25 frames. ], batch size: 136, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:07:33,011 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296572.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:07:37,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8896, 1.6825, 1.3149, 1.3664], device='cuda:1'), covar=tensor([0.0614, 0.0618, 0.0906, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0442, 0.0494, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 19:07:45,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:07:53,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:07:56,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:07:57,329 INFO [train.py:968] (1/2) Epoch 7, batch 22950, giga_loss[loss=0.2496, simple_loss=0.313, pruned_loss=0.0931, over 28537.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3549, pruned_loss=0.1097, over 5711049.47 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3803, pruned_loss=0.1251, over 5755654.15 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3519, pruned_loss=0.107, over 5707960.30 frames. ], batch size: 71, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:08:06,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2497, 1.7918, 1.3425, 0.4444], device='cuda:1'), covar=tensor([0.2489, 0.1455, 0.2643, 0.3322], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1336, 0.1403, 0.1193], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 19:08:19,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:08:19,566 INFO [optim.py:369] (1/2) 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 19:08:28,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:08:36,934 INFO [train.py:968] (1/2) Epoch 7, batch 23000, giga_loss[loss=0.3061, simple_loss=0.3689, pruned_loss=0.1216, over 27987.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3547, pruned_loss=0.111, over 5708517.89 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3803, pruned_loss=0.1252, over 5758202.14 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3519, pruned_loss=0.1085, over 5702992.69 frames. ], batch size: 412, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:08:38,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-03 19:08:49,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-03 19:09:14,918 INFO [train.py:968] (1/2) Epoch 7, batch 23050, libri_loss[loss=0.3138, simple_loss=0.3808, pruned_loss=0.1234, over 29691.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3543, pruned_loss=0.1107, over 5715822.64 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3808, pruned_loss=0.1257, over 5754822.09 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3511, pruned_loss=0.1079, over 5713311.13 frames. ], batch size: 91, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:09:21,766 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:09:39,077 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:09:41,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:09:53,773 INFO [train.py:968] (1/2) Epoch 7, batch 23100, giga_loss[loss=0.246, simple_loss=0.3177, pruned_loss=0.08716, over 28901.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3493, pruned_loss=0.1081, over 5712042.12 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3805, pruned_loss=0.1258, over 5756522.51 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3464, pruned_loss=0.1055, over 5707658.86 frames. ], batch size: 213, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:10:03,115 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:15,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-03 19:10:19,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:22,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:32,317 INFO [train.py:968] (1/2) Epoch 7, batch 23150, giga_loss[loss=0.2636, simple_loss=0.3281, pruned_loss=0.09955, over 28992.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3448, pruned_loss=0.1056, over 5703333.88 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3805, pruned_loss=0.1259, over 5748450.85 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3418, pruned_loss=0.103, over 5706758.54 frames. ], batch size: 106, lr: 4.61e-03, grad_scale: 2.0
+2023-03-03 19:10:44,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6721, 1.8029, 1.4303, 1.9913], device='cuda:1'), covar=tensor([0.2104, 0.2106, 0.2196, 0.1988], device='cuda:1'), in_proj_covar=tensor([0.1181, 0.0891, 0.1037, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:10:45,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:46,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=296817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:10:52,562 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:11:10,291 INFO [train.py:968] (1/2) Epoch 7, batch 23200, giga_loss[loss=0.3318, simple_loss=0.381, pruned_loss=0.1413, over 26633.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3451, pruned_loss=0.1058, over 5704537.71 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3807, pruned_loss=0.1263, over 5747981.29 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.341, pruned_loss=0.1025, over 5706140.45 frames. ], batch size: 555, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:11:40,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:11:43,544 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:11:51,531 INFO [train.py:968] (1/2) Epoch 7, batch 23250, giga_loss[loss=0.309, simple_loss=0.3853, pruned_loss=0.1164, over 28720.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3469, pruned_loss=0.106, over 5709787.77 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3806, pruned_loss=0.1264, over 5750546.21 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3433, pruned_loss=0.103, over 5708167.49 frames. ], batch size: 284, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:11:56,246 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=296907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:12:13,359 INFO [optim.py:369] (1/2) 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 19:12:31,936 INFO [train.py:968] (1/2) Epoch 7, batch 23300, giga_loss[loss=0.2737, simple_loss=0.345, pruned_loss=0.1012, over 28234.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3512, pruned_loss=0.1082, over 5713615.56 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3805, pruned_loss=0.1267, over 5753019.86 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3474, pruned_loss=0.1049, over 5708879.20 frames. ], batch size: 77, lr: 4.61e-03, grad_scale: 4.0
+2023-03-03 19:12:39,918 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=296960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:12:42,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:13:06,695 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=296992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:13:08,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6878, 1.3665, 5.4193, 3.6105], device='cuda:1'), covar=tensor([0.1663, 0.2484, 0.0297, 0.0653], device='cuda:1'), in_proj_covar=tensor([0.0596, 0.0552, 0.0784, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:13:12,663 INFO [train.py:968] (1/2) Epoch 7, batch 23350, giga_loss[loss=0.3192, simple_loss=0.3905, pruned_loss=0.124, over 28727.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3553, pruned_loss=0.1099, over 5711174.09 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3806, pruned_loss=0.1268, over 5752660.33 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.352, pruned_loss=0.1071, over 5707291.35 frames. ], batch size: 284, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:13:18,081 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:13:37,386 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 23400, giga_loss[loss=0.3013, simple_loss=0.3588, pruned_loss=0.1219, over 23749.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3589, pruned_loss=0.1115, over 5696408.81 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3811, pruned_loss=0.1273, over 5746405.70 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3553, pruned_loss=0.1085, over 5698595.06 frames. ], batch size: 705, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:14:13,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-03 19:14:16,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-03 19:14:21,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:14:23,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:14:36,702 INFO [train.py:968] (1/2) Epoch 7, batch 23450, libri_loss[loss=0.3145, simple_loss=0.3659, pruned_loss=0.1316, over 29678.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3607, pruned_loss=0.112, over 5696080.44 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3809, pruned_loss=0.1272, over 5748514.60 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3578, pruned_loss=0.1095, over 5695153.69 frames. ], batch size: 73, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:14:40,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2925, 1.4929, 1.2039, 1.3228], device='cuda:1'), covar=tensor([0.2033, 0.2047, 0.2165, 0.1920], device='cuda:1'), in_proj_covar=tensor([0.1186, 0.0898, 0.1041, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:14:42,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7421, 1.6638, 1.2492, 1.3996], device='cuda:1'), covar=tensor([0.0570, 0.0535, 0.0927, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0441, 0.0494, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 19:15:05,479 INFO [optim.py:369] (1/2) 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 19:15:16,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4577, 3.0015, 1.9186, 1.7346], device='cuda:1'), covar=tensor([0.1452, 0.0835, 0.1174, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1377, 0.1367, 0.1447], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 19:15:21,538 INFO [train.py:968] (1/2) Epoch 7, batch 23500, giga_loss[loss=0.28, simple_loss=0.3564, pruned_loss=0.1018, over 28900.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3675, pruned_loss=0.1185, over 5693056.75 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3813, pruned_loss=0.128, over 5748332.07 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3642, pruned_loss=0.1153, over 5690632.94 frames. ], batch size: 136, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:15:59,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 19:16:12,024 INFO [train.py:968] (1/2) Epoch 7, batch 23550, giga_loss[loss=0.3702, simple_loss=0.4238, pruned_loss=0.1583, over 28861.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3742, pruned_loss=0.1241, over 5690079.82 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3814, pruned_loss=0.1281, over 5751712.50 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3713, pruned_loss=0.1214, over 5684286.64 frames. ], batch size: 99, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:16:14,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:16:14,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2889, 1.8402, 1.3839, 0.4381], device='cuda:1'), covar=tensor([0.2051, 0.1440, 0.2378, 0.2855], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1347, 0.1408, 0.1189], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 19:16:36,877 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:16:40,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:16:40,638 INFO [optim.py:369] (1/2) 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 19:16:59,554 INFO [train.py:968] (1/2) Epoch 7, batch 23600, giga_loss[loss=0.3318, simple_loss=0.387, pruned_loss=0.1383, over 28825.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3806, pruned_loss=0.1292, over 5684884.94 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.381, pruned_loss=0.1281, over 5750465.57 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3786, pruned_loss=0.127, over 5679718.00 frames. ], batch size: 99, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:17:08,955 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:10,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:17,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:18,038 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:31,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:17:44,143 INFO [train.py:968] (1/2) Epoch 7, batch 23650, giga_loss[loss=0.3207, simple_loss=0.3805, pruned_loss=0.1305, over 28672.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3861, pruned_loss=0.1339, over 5683567.74 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3812, pruned_loss=0.1281, over 5749901.22 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3844, pruned_loss=0.1322, over 5677201.97 frames. ], batch size: 92, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:18:15,731 INFO [optim.py:369] (1/2) 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 19:18:16,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=297331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:18:32,429 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:18:34,721 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 23700, giga_loss[loss=0.4083, simple_loss=0.443, pruned_loss=0.1868, over 28931.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.394, pruned_loss=0.1412, over 5668841.41 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3809, pruned_loss=0.128, over 5750416.82 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.393, pruned_loss=0.1401, over 5662733.24 frames. ], batch size: 213, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:19:01,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:05,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:23,714 INFO [train.py:968] (1/2) Epoch 7, batch 23750, giga_loss[loss=0.4599, simple_loss=0.4697, pruned_loss=0.2251, over 26541.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3987, pruned_loss=0.1449, over 5670365.94 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3814, pruned_loss=0.1283, over 5754241.21 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.398, pruned_loss=0.1443, over 5659241.74 frames. ], batch size: 555, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:19:26,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:28,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:31,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:32,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:45,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:49,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:51,012 INFO [optim.py:369] (1/2) 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 19:19:55,843 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:19:59,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:20:09,714 INFO [train.py:968] (1/2) Epoch 7, batch 23800, giga_loss[loss=0.3246, simple_loss=0.3838, pruned_loss=0.1327, over 28668.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3997, pruned_loss=0.1463, over 5668546.15 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3811, pruned_loss=0.1282, over 5757647.70 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3999, pruned_loss=0.1464, over 5654304.09 frames. ], batch size: 262, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:20:15,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:20:15,939 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:20:25,040 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 19:20:31,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-03 19:21:00,351 INFO [train.py:968] (1/2) Epoch 7, batch 23850, libri_loss[loss=0.3456, simple_loss=0.4015, pruned_loss=0.1449, over 29222.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.4003, pruned_loss=0.1479, over 5659123.00 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3812, pruned_loss=0.1283, over 5760459.32 frames. ], giga_tot_loss[loss=0.3487, simple_loss=0.4008, pruned_loss=0.1483, over 5643489.70 frames. ], batch size: 97, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:21:24,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:21:26,434 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:21:48,793 INFO [train.py:968] (1/2) Epoch 7, batch 23900, giga_loss[loss=0.4306, simple_loss=0.4489, pruned_loss=0.2062, over 27669.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4031, pruned_loss=0.151, over 5650830.55 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3813, pruned_loss=0.1285, over 5756385.86 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4038, pruned_loss=0.1516, over 5639517.62 frames. ], batch size: 472, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:21:54,142 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:22:39,269 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:22:39,595 INFO [train.py:968] (1/2) Epoch 7, batch 23950, giga_loss[loss=0.364, simple_loss=0.4103, pruned_loss=0.1589, over 28941.00 frames. ], tot_loss[loss=0.36, simple_loss=0.4082, pruned_loss=0.1559, over 5628478.10 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3821, pruned_loss=0.1293, over 5738974.54 frames. ], giga_tot_loss[loss=0.3612, simple_loss=0.409, pruned_loss=0.1567, over 5629575.29 frames. ], batch size: 227, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:22:41,329 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:22:56,496 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:23:15,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:23:22,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2747, 1.4749, 1.1405, 1.0225], device='cuda:1'), covar=tensor([0.1216, 0.1119, 0.0890, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.1536, 0.1375, 0.1365, 0.1452], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 19:23:24,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:23:32,892 INFO [train.py:968] (1/2) Epoch 7, batch 24000, giga_loss[loss=0.3449, simple_loss=0.3962, pruned_loss=0.1468, over 28938.00 frames. ], tot_loss[loss=0.361, simple_loss=0.4078, pruned_loss=0.1571, over 5606061.90 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.382, pruned_loss=0.1293, over 5739226.94 frames. ], giga_tot_loss[loss=0.3628, simple_loss=0.4091, pruned_loss=0.1583, over 5603865.12 frames. ], batch size: 227, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:23:32,892 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 19:23:41,158 INFO [train.py:1012] (1/2) Epoch 7, validation: loss=0.2251, simple_loss=0.3303, pruned_loss=0.06, over 944034.00 frames. 
+2023-03-03 19:23:41,159 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19464MB
+2023-03-03 19:24:27,496 INFO [train.py:968] (1/2) Epoch 7, batch 24050, giga_loss[loss=0.4302, simple_loss=0.4455, pruned_loss=0.2074, over 26600.00 frames. ], tot_loss[loss=0.3598, simple_loss=0.4062, pruned_loss=0.1567, over 5620112.45 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3823, pruned_loss=0.1297, over 5739557.40 frames. ], giga_tot_loss[loss=0.3616, simple_loss=0.4076, pruned_loss=0.1578, over 5615733.01 frames. ], batch size: 555, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:24:33,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:24:54,548 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 24100, giga_loss[loss=0.3363, simple_loss=0.3979, pruned_loss=0.1373, over 28604.00 frames. ], tot_loss[loss=0.3569, simple_loss=0.4043, pruned_loss=0.1547, over 5639196.68 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3823, pruned_loss=0.1299, over 5745290.16 frames. ], giga_tot_loss[loss=0.3596, simple_loss=0.4063, pruned_loss=0.1565, over 5626643.76 frames. ], batch size: 242, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:25:43,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=297784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:25:45,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:25:55,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5066, 1.4429, 1.1550, 1.0887], device='cuda:1'), covar=tensor([0.0623, 0.0488, 0.0943, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0447, 0.0497, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 19:25:57,585 INFO [train.py:968] (1/2) Epoch 7, batch 24150, giga_loss[loss=0.3146, simple_loss=0.379, pruned_loss=0.1251, over 28424.00 frames. ], tot_loss[loss=0.3542, simple_loss=0.4032, pruned_loss=0.1526, over 5629978.29 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.382, pruned_loss=0.1298, over 5749323.80 frames. ], giga_tot_loss[loss=0.3577, simple_loss=0.4057, pruned_loss=0.1548, over 5613016.40 frames. ], batch size: 71, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:26:16,437 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:26:33,029 INFO [optim.py:369] (1/2) 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 19:26:50,051 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 24200, libri_loss[loss=0.3169, simple_loss=0.3726, pruned_loss=0.1306, over 29562.00 frames. ], tot_loss[loss=0.3551, simple_loss=0.4045, pruned_loss=0.1529, over 5624629.04 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3818, pruned_loss=0.1298, over 5745681.12 frames. ], giga_tot_loss[loss=0.3586, simple_loss=0.4071, pruned_loss=0.155, over 5612404.09 frames. ], batch size: 76, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:26:50,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5465, 3.2289, 1.5060, 1.4909], device='cuda:1'), covar=tensor([0.0824, 0.0269, 0.0833, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0485, 0.0316, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 19:26:53,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=297852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:27:21,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=297881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:27:36,874 INFO [train.py:968] (1/2) Epoch 7, batch 24250, giga_loss[loss=0.3055, simple_loss=0.3722, pruned_loss=0.1195, over 28950.00 frames. ], tot_loss[loss=0.3541, simple_loss=0.404, pruned_loss=0.1521, over 5633020.64 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3822, pruned_loss=0.1302, over 5746885.58 frames. ], giga_tot_loss[loss=0.3576, simple_loss=0.4065, pruned_loss=0.1544, over 5617807.71 frames. ], batch size: 164, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:28:08,055 INFO [optim.py:369] (1/2) 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 19:28:24,525 INFO [train.py:968] (1/2) Epoch 7, batch 24300, giga_loss[loss=0.3337, simple_loss=0.4024, pruned_loss=0.1325, over 28796.00 frames. ], tot_loss[loss=0.3499, simple_loss=0.4012, pruned_loss=0.1493, over 5634332.63 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3821, pruned_loss=0.1304, over 5750904.13 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.404, pruned_loss=0.1516, over 5614775.85 frames. ], batch size: 174, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:29:04,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=297990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:29:15,704 INFO [train.py:968] (1/2) Epoch 7, batch 24350, giga_loss[loss=0.3261, simple_loss=0.3916, pruned_loss=0.1303, over 28902.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3989, pruned_loss=0.1461, over 5643118.65 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3824, pruned_loss=0.1305, over 5752559.68 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.401, pruned_loss=0.1479, over 5625234.56 frames. ], batch size: 174, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:29:45,467 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 24400, giga_loss[loss=0.3118, simple_loss=0.3745, pruned_loss=0.1245, over 28888.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3947, pruned_loss=0.1427, over 5642715.45 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3821, pruned_loss=0.1304, over 5756578.40 frames. ], giga_tot_loss[loss=0.3431, simple_loss=0.3971, pruned_loss=0.1446, over 5622150.96 frames. ], batch size: 199, lr: 4.60e-03, grad_scale: 8.0
+2023-03-03 19:30:54,862 INFO [train.py:968] (1/2) Epoch 7, batch 24450, giga_loss[loss=0.3081, simple_loss=0.3736, pruned_loss=0.1213, over 28661.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3935, pruned_loss=0.142, over 5646578.06 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.382, pruned_loss=0.1306, over 5758153.76 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3955, pruned_loss=0.1435, over 5627537.27 frames. ], batch size: 307, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:31:26,329 INFO [optim.py:369] (1/2) 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 19:31:26,667 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:31:29,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:31:41,221 INFO [train.py:968] (1/2) Epoch 7, batch 24500, giga_loss[loss=0.3837, simple_loss=0.4209, pruned_loss=0.1732, over 27580.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3932, pruned_loss=0.1422, over 5637224.78 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3816, pruned_loss=0.1304, over 5750674.46 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3955, pruned_loss=0.1438, over 5625609.63 frames. ], batch size: 472, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:31:54,562 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:32:29,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2231, 2.5490, 1.2491, 1.3317], device='cuda:1'), covar=tensor([0.0921, 0.0328, 0.0865, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0491, 0.0318, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 19:32:33,709 INFO [train.py:968] (1/2) Epoch 7, batch 24550, giga_loss[loss=0.4546, simple_loss=0.4585, pruned_loss=0.2253, over 26605.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3924, pruned_loss=0.1412, over 5639987.70 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3815, pruned_loss=0.1303, over 5753932.52 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3945, pruned_loss=0.1428, over 5625433.47 frames. ], batch size: 555, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:32:39,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5056, 1.8449, 1.8615, 1.4381], device='cuda:1'), covar=tensor([0.1453, 0.1836, 0.1111, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0713, 0.0808, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 19:32:43,619 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-03 19:32:45,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-03 19:33:07,604 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 24600, libri_loss[loss=0.3088, simple_loss=0.3593, pruned_loss=0.1291, over 29652.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3898, pruned_loss=0.1385, over 5660163.88 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.381, pruned_loss=0.1302, over 5756969.81 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3921, pruned_loss=0.1401, over 5643803.18 frames. ], batch size: 73, lr: 4.60e-03, grad_scale: 4.0
+2023-03-03 19:34:14,958 INFO [train.py:968] (1/2) Epoch 7, batch 24650, giga_loss[loss=0.3271, simple_loss=0.4018, pruned_loss=0.1262, over 28910.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.391, pruned_loss=0.1371, over 5664502.83 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3811, pruned_loss=0.1302, over 5759381.36 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.393, pruned_loss=0.1384, over 5647696.39 frames. ], batch size: 145, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:34:38,433 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-03 19:34:50,329 INFO [optim.py:369] (1/2) 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 19:35:08,772 INFO [train.py:968] (1/2) Epoch 7, batch 24700, giga_loss[loss=0.3476, simple_loss=0.3972, pruned_loss=0.149, over 27578.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3932, pruned_loss=0.1375, over 5655882.09 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3816, pruned_loss=0.1308, over 5749181.61 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3945, pruned_loss=0.1382, over 5649894.67 frames. ], batch size: 472, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:35:09,051 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=298350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:35:55,095 INFO [train.py:968] (1/2) Epoch 7, batch 24750, giga_loss[loss=0.3292, simple_loss=0.3915, pruned_loss=0.1334, over 28821.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3937, pruned_loss=0.1384, over 5661603.12 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3819, pruned_loss=0.131, over 5750145.58 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3949, pruned_loss=0.139, over 5652226.18 frames. ], batch size: 284, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:36:09,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-03 19:36:21,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5480, 1.0344, 2.8625, 2.6648], device='cuda:1'), covar=tensor([0.1790, 0.2330, 0.0530, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0599, 0.0563, 0.0791, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:36:28,933 INFO [optim.py:369] (1/2) 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 19:36:40,608 INFO [train.py:968] (1/2) Epoch 7, batch 24800, giga_loss[loss=0.3257, simple_loss=0.3846, pruned_loss=0.1334, over 28573.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3927, pruned_loss=0.1376, over 5679033.71 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3818, pruned_loss=0.1311, over 5752737.17 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3939, pruned_loss=0.1381, over 5668075.32 frames. ], batch size: 307, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:36:41,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3681, 1.6192, 1.3303, 1.5773], device='cuda:1'), covar=tensor([0.2158, 0.2172, 0.2262, 0.1871], device='cuda:1'), in_proj_covar=tensor([0.1183, 0.0899, 0.1047, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:37:28,149 INFO [train.py:968] (1/2) Epoch 7, batch 24850, giga_loss[loss=0.3272, simple_loss=0.3786, pruned_loss=0.1379, over 28676.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3898, pruned_loss=0.1361, over 5687634.82 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3815, pruned_loss=0.1308, over 5755023.79 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3913, pruned_loss=0.1369, over 5674870.75 frames. ], batch size: 307, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:37:59,735 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 24900, giga_loss[loss=0.31, simple_loss=0.3763, pruned_loss=0.1218, over 29059.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3886, pruned_loss=0.1364, over 5685086.11 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3816, pruned_loss=0.1309, over 5756562.73 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3898, pruned_loss=0.137, over 5673136.02 frames. ], batch size: 155, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:38:58,431 INFO [train.py:968] (1/2) Epoch 7, batch 24950, giga_loss[loss=0.3214, simple_loss=0.3768, pruned_loss=0.133, over 28292.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3879, pruned_loss=0.1354, over 5676209.76 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3816, pruned_loss=0.131, over 5748418.96 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3891, pruned_loss=0.1359, over 5671807.08 frames. ], batch size: 368, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:39:13,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5106, 4.2831, 4.0520, 1.7217], device='cuda:1'), covar=tensor([0.0555, 0.0845, 0.0902, 0.2146], device='cuda:1'), in_proj_covar=tensor([0.0951, 0.0901, 0.0802, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 19:39:28,854 INFO [optim.py:369] (1/2) 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 19:39:43,055 INFO [train.py:968] (1/2) Epoch 7, batch 25000, giga_loss[loss=0.379, simple_loss=0.4236, pruned_loss=0.1672, over 27918.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3876, pruned_loss=0.1335, over 5686897.55 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.382, pruned_loss=0.1313, over 5750986.51 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3882, pruned_loss=0.1337, over 5679566.27 frames. ], batch size: 412, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:39:43,465 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-03 19:40:11,539 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-03 19:40:19,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3504, 1.5537, 1.3198, 1.5336], device='cuda:1'), covar=tensor([0.0656, 0.0381, 0.0307, 0.0672], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0120, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 19:40:27,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4869, 1.5850, 1.2867, 1.8679], device='cuda:1'), covar=tensor([0.2422, 0.2398, 0.2560, 0.2251], device='cuda:1'), in_proj_covar=tensor([0.1194, 0.0906, 0.1052, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:40:31,772 INFO [train.py:968] (1/2) Epoch 7, batch 25050, giga_loss[loss=0.3311, simple_loss=0.3894, pruned_loss=0.1364, over 28821.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3879, pruned_loss=0.1339, over 5685153.37 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3822, pruned_loss=0.1315, over 5754790.67 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3883, pruned_loss=0.134, over 5674282.36 frames. ], batch size: 186, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:40:54,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=298725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:41:04,158 INFO [optim.py:369] (1/2) 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 19:41:17,644 INFO [train.py:968] (1/2) Epoch 7, batch 25100, giga_loss[loss=0.3477, simple_loss=0.3984, pruned_loss=0.1485, over 28783.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3862, pruned_loss=0.1331, over 5682414.66 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3823, pruned_loss=0.1316, over 5749987.81 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3867, pruned_loss=0.1331, over 5675257.74 frames. ], batch size: 284, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:41:48,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5121, 2.1421, 1.5525, 0.5553], device='cuda:1'), covar=tensor([0.2724, 0.1495, 0.2270, 0.3505], device='cuda:1'), in_proj_covar=tensor([0.1459, 0.1369, 0.1425, 0.1197], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 19:41:51,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3428, 1.5942, 1.4158, 1.4657], device='cuda:1'), covar=tensor([0.0721, 0.0320, 0.0303, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0117, 0.0119, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0049, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 19:42:05,920 INFO [train.py:968] (1/2) Epoch 7, batch 25150, giga_loss[loss=0.2884, simple_loss=0.3528, pruned_loss=0.1119, over 28909.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3842, pruned_loss=0.1324, over 5678144.16 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3818, pruned_loss=0.1314, over 5753379.01 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3852, pruned_loss=0.1327, over 5667803.97 frames. ], batch size: 106, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:42:36,001 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 25200, giga_loss[loss=0.2845, simple_loss=0.3541, pruned_loss=0.1074, over 28972.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3852, pruned_loss=0.1342, over 5674941.18 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3819, pruned_loss=0.1317, over 5755454.82 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3859, pruned_loss=0.1342, over 5661002.12 frames. ], batch size: 136, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:43:04,047 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=298868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:43:08,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=298871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:43:11,449 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 25250, giga_loss[loss=0.3574, simple_loss=0.4124, pruned_loss=0.1512, over 29044.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3862, pruned_loss=0.1358, over 5678005.53 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3826, pruned_loss=0.1322, over 5753984.79 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3862, pruned_loss=0.1353, over 5666711.52 frames. ], batch size: 155, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:43:32,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=298900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:43:48,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8350, 1.1044, 1.0540, 1.0583], device='cuda:1'), covar=tensor([0.1060, 0.0970, 0.1499, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0735, 0.0656, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 19:43:56,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=298923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:44:06,129 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 25300, giga_loss[loss=0.3584, simple_loss=0.4053, pruned_loss=0.1558, over 28681.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3844, pruned_loss=0.1352, over 5672954.35 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.382, pruned_loss=0.1318, over 5754013.79 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3851, pruned_loss=0.1353, over 5662404.00 frames. ], batch size: 262, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:45:06,720 INFO [train.py:968] (1/2) Epoch 7, batch 25350, giga_loss[loss=0.2822, simple_loss=0.3512, pruned_loss=0.1066, over 28899.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3837, pruned_loss=0.1348, over 5679917.11 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3821, pruned_loss=0.132, over 5753536.84 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3842, pruned_loss=0.1348, over 5670592.94 frames. ], batch size: 227, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:45:31,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299022.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:45:43,965 INFO [optim.py:369] (1/2) 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 19:45:56,321 INFO [train.py:968] (1/2) Epoch 7, batch 25400, giga_loss[loss=0.3138, simple_loss=0.3772, pruned_loss=0.1252, over 28659.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3834, pruned_loss=0.1345, over 5670541.99 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3822, pruned_loss=0.132, over 5755249.16 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3837, pruned_loss=0.1345, over 5660405.13 frames. ], batch size: 242, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:46:26,764 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:46:41,264 INFO [train.py:968] (1/2) Epoch 7, batch 25450, libri_loss[loss=0.2571, simple_loss=0.3229, pruned_loss=0.09568, over 28587.00 frames. ], tot_loss[loss=0.326, simple_loss=0.384, pruned_loss=0.134, over 5660409.95 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3821, pruned_loss=0.132, over 5743085.91 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3844, pruned_loss=0.1341, over 5660672.36 frames. ], batch size: 63, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:47:11,528 INFO [optim.py:369] (1/2) 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 19:47:17,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3530, 1.4276, 1.2384, 1.4961], device='cuda:1'), covar=tensor([0.0768, 0.0323, 0.0335, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0117, 0.0119, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0049, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 19:47:24,799 INFO [train.py:968] (1/2) Epoch 7, batch 25500, giga_loss[loss=0.2988, simple_loss=0.3712, pruned_loss=0.1131, over 28794.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3831, pruned_loss=0.1327, over 5667734.42 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3816, pruned_loss=0.1318, over 5744821.21 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3839, pruned_loss=0.133, over 5664879.68 frames. ], batch size: 119, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:47:47,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2511, 1.9340, 1.4198, 0.4184], device='cuda:1'), covar=tensor([0.2710, 0.1452, 0.2234, 0.3426], device='cuda:1'), in_proj_covar=tensor([0.1455, 0.1371, 0.1414, 0.1192], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 19:47:55,376 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-03 19:48:11,490 INFO [train.py:968] (1/2) Epoch 7, batch 25550, giga_loss[loss=0.3514, simple_loss=0.4058, pruned_loss=0.1485, over 28816.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3841, pruned_loss=0.1334, over 5650061.05 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.382, pruned_loss=0.1322, over 5734928.11 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3844, pruned_loss=0.1333, over 5655280.56 frames. ], batch size: 99, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:48:26,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-03 19:48:43,770 INFO [optim.py:369] (1/2) 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 19:48:55,721 INFO [train.py:968] (1/2) Epoch 7, batch 25600, giga_loss[loss=0.3265, simple_loss=0.3864, pruned_loss=0.1333, over 27945.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3862, pruned_loss=0.1357, over 5655461.21 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.382, pruned_loss=0.1323, over 5737707.99 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3865, pruned_loss=0.1356, over 5655439.82 frames. ], batch size: 412, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:48:57,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:49:43,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:49:45,776 INFO [train.py:968] (1/2) Epoch 7, batch 25650, giga_loss[loss=0.352, simple_loss=0.4091, pruned_loss=0.1475, over 28849.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3892, pruned_loss=0.1393, over 5651772.09 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.382, pruned_loss=0.1323, over 5737726.29 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3895, pruned_loss=0.1393, over 5651038.23 frames. ], batch size: 174, lr: 4.59e-03, grad_scale: 4.0
+2023-03-03 19:50:06,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6610, 1.0047, 2.8455, 2.6873], device='cuda:1'), covar=tensor([0.1689, 0.2287, 0.0556, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0594, 0.0554, 0.0783, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 19:50:19,246 INFO [optim.py:369] (1/2) 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 19:50:34,389 INFO [train.py:968] (1/2) Epoch 7, batch 25700, giga_loss[loss=0.3242, simple_loss=0.3803, pruned_loss=0.1341, over 28689.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3905, pruned_loss=0.1411, over 5663108.36 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3825, pruned_loss=0.1328, over 5741390.58 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3906, pruned_loss=0.1409, over 5656889.08 frames. ], batch size: 262, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:51:22,656 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:51:24,385 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.81 vs. limit=5.0
+2023-03-03 19:51:24,821 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:51:26,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299397.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:51:29,321 INFO [train.py:968] (1/2) Epoch 7, batch 25750, giga_loss[loss=0.3686, simple_loss=0.4161, pruned_loss=0.1606, over 27941.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3925, pruned_loss=0.1437, over 5646692.01 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3825, pruned_loss=0.133, over 5741598.76 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3926, pruned_loss=0.1434, over 5640909.06 frames. ], batch size: 412, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:51:50,348 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:52:04,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:07,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:13,501 INFO [train.py:968] (1/2) Epoch 7, batch 25800, giga_loss[loss=0.3194, simple_loss=0.3688, pruned_loss=0.135, over 28720.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.393, pruned_loss=0.1446, over 5654879.25 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3826, pruned_loss=0.1331, over 5743480.05 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3931, pruned_loss=0.1444, over 5647872.33 frames. ], batch size: 92, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:52:20,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:37,160 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:52:42,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 19:52:59,699 INFO [train.py:968] (1/2) Epoch 7, batch 25850, giga_loss[loss=0.3126, simple_loss=0.3833, pruned_loss=0.121, over 28581.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3917, pruned_loss=0.1428, over 5660921.44 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3829, pruned_loss=0.1332, over 5746857.42 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3917, pruned_loss=0.1427, over 5650639.79 frames. ], batch size: 71, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:53:16,213 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:53:31,727 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299540.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:53:38,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 19:53:44,124 INFO [train.py:968] (1/2) Epoch 7, batch 25900, giga_loss[loss=0.3232, simple_loss=0.3866, pruned_loss=0.1299, over 28970.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3895, pruned_loss=0.1392, over 5674812.62 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3834, pruned_loss=0.1335, over 5749197.32 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3891, pruned_loss=0.1391, over 5663230.82 frames. ], batch size: 136, lr: 4.59e-03, grad_scale: 2.0
+2023-03-03 19:54:06,649 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299572.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 19:54:17,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:54:29,736 INFO [train.py:968] (1/2) Epoch 7, batch 25950, giga_loss[loss=0.3695, simple_loss=0.3912, pruned_loss=0.1739, over 23895.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3858, pruned_loss=0.1366, over 5652875.69 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3836, pruned_loss=0.1336, over 5738074.50 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3854, pruned_loss=0.1364, over 5652168.63 frames. ], batch size: 705, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:54:30,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=299600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:54:32,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=299603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:54:40,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4164, 1.6332, 1.2414, 1.0963], device='cuda:1'), covar=tensor([0.1473, 0.1311, 0.1003, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.1556, 0.1402, 0.1375, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 19:55:01,160 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=299632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:55:04,249 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:55:04,574 INFO [optim.py:369] (1/2) 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 19:55:14,720 INFO [train.py:968] (1/2) Epoch 7, batch 26000, giga_loss[loss=0.3452, simple_loss=0.3952, pruned_loss=0.1476, over 28901.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3832, pruned_loss=0.1348, over 5667494.14 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3836, pruned_loss=0.1336, over 5740140.06 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3828, pruned_loss=0.1347, over 5663061.05 frames. ], batch size: 199, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 19:55:34,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3191, 1.8520, 1.6808, 1.2689], device='cuda:1'), covar=tensor([0.1550, 0.1913, 0.1190, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0722, 0.0818, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 19:55:51,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-03 19:56:05,125 INFO [train.py:968] (1/2) Epoch 7, batch 26050, giga_loss[loss=0.3221, simple_loss=0.3784, pruned_loss=0.1328, over 28780.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3819, pruned_loss=0.1341, over 5679654.28 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3833, pruned_loss=0.1335, over 5740636.24 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3818, pruned_loss=0.1342, over 5673663.54 frames. ], batch size: 284, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 19:56:20,455 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=299716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 19:56:25,501 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 19:56:38,963 INFO [optim.py:369] (1/2) 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 19:56:51,314 INFO [train.py:968] (1/2) Epoch 7, batch 26100, giga_loss[loss=0.3197, simple_loss=0.3818, pruned_loss=0.1288, over 28549.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3851, pruned_loss=0.1364, over 5678653.08 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.384, pruned_loss=0.1339, over 5742274.91 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3845, pruned_loss=0.136, over 5671762.20 frames. ], batch size: 336, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 19:57:34,681 INFO [train.py:968] (1/2) Epoch 7, batch 26150, giga_loss[loss=0.3171, simple_loss=0.3994, pruned_loss=0.1174, over 29027.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3892, pruned_loss=0.1368, over 5689956.43 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3844, pruned_loss=0.1342, over 5746110.73 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3884, pruned_loss=0.1364, over 5679732.19 frames. ], batch size: 128, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:58:12,126 INFO [optim.py:369] (1/2) 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 19:58:21,877 INFO [train.py:968] (1/2) Epoch 7, batch 26200, libri_loss[loss=0.3009, simple_loss=0.3681, pruned_loss=0.1169, over 29530.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3913, pruned_loss=0.1363, over 5686058.95 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3844, pruned_loss=0.1343, over 5748557.01 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3908, pruned_loss=0.1358, over 5674224.16 frames. ], batch size: 84, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:58:58,334 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 26250, giga_loss[loss=0.3125, simple_loss=0.3813, pruned_loss=0.1218, over 28983.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.393, pruned_loss=0.1381, over 5688161.73 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3844, pruned_loss=0.1346, over 5743985.07 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3929, pruned_loss=0.1375, over 5680051.32 frames. ], batch size: 145, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 19:59:43,034 INFO [optim.py:369] (1/2) 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 19:59:53,771 INFO [train.py:968] (1/2) Epoch 7, batch 26300, giga_loss[loss=0.365, simple_loss=0.4135, pruned_loss=0.1582, over 28648.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.395, pruned_loss=0.1401, over 5685168.85 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3842, pruned_loss=0.1344, over 5746158.53 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3952, pruned_loss=0.1398, over 5676150.73 frames. ], batch size: 336, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:00:00,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=299959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:00:39,982 INFO [train.py:968] (1/2) Epoch 7, batch 26350, giga_loss[loss=0.3499, simple_loss=0.3992, pruned_loss=0.1504, over 28873.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3957, pruned_loss=0.1415, over 5678067.40 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1343, over 5741037.77 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3965, pruned_loss=0.1415, over 5672992.54 frames. ], batch size: 112, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:00:51,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:01:16,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:01:16,765 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:01:33,182 INFO [train.py:968] (1/2) Epoch 7, batch 26400, giga_loss[loss=0.3017, simple_loss=0.3643, pruned_loss=0.1195, over 28854.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3945, pruned_loss=0.1411, over 5682164.18 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.384, pruned_loss=0.1344, over 5741885.58 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.395, pruned_loss=0.1412, over 5677075.20 frames. ], batch size: 112, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:01:48,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:01:49,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-03 20:02:08,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:14,119 INFO [train.py:968] (1/2) Epoch 7, batch 26450, giga_loss[loss=0.3208, simple_loss=0.3804, pruned_loss=0.1306, over 28656.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3914, pruned_loss=0.1393, over 5688255.51 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3839, pruned_loss=0.1343, over 5742096.25 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3921, pruned_loss=0.1396, over 5682180.01 frames. ], batch size: 307, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:02:18,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:20,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:37,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2563, 1.3392, 1.1300, 1.0251], device='cuda:1'), covar=tensor([0.0600, 0.0354, 0.0894, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0450, 0.0496, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 20:02:46,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:02:48,991 INFO [optim.py:369] (1/2) 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 20:02:52,950 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-03 20:02:54,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2716, 1.6371, 1.3625, 1.4680], device='cuda:1'), covar=tensor([0.0716, 0.0317, 0.0298, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0117, 0.0120, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0067, 0.0049, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 20:03:01,747 INFO [train.py:968] (1/2) Epoch 7, batch 26500, giga_loss[loss=0.3431, simple_loss=0.3913, pruned_loss=0.1474, over 28706.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3909, pruned_loss=0.1399, over 5685881.12 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3851, pruned_loss=0.1354, over 5737392.23 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3906, pruned_loss=0.1393, over 5684460.70 frames. ], batch size: 85, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:03:08,718 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:10,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:15,831 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:35,613 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:03:40,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300191.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 20:03:44,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-03 20:03:50,028 INFO [train.py:968] (1/2) Epoch 7, batch 26550, giga_loss[loss=0.3734, simple_loss=0.4193, pruned_loss=0.1637, over 27598.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3892, pruned_loss=0.1389, over 5681586.05 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3848, pruned_loss=0.1351, over 5742865.09 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3894, pruned_loss=0.1389, over 5674164.70 frames. ], batch size: 472, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:04:20,821 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:04:25,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:04:25,706 INFO [optim.py:369] (1/2) 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 20:04:34,195 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 20:04:35,252 INFO [train.py:968] (1/2) Epoch 7, batch 26600, giga_loss[loss=0.2884, simple_loss=0.3499, pruned_loss=0.1135, over 28214.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3888, pruned_loss=0.1384, over 5683458.21 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3848, pruned_loss=0.1351, over 5741825.72 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.389, pruned_loss=0.1384, over 5677111.25 frames. ], batch size: 77, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:04:50,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:05:21,162 INFO [train.py:968] (1/2) Epoch 7, batch 26650, giga_loss[loss=0.3068, simple_loss=0.3624, pruned_loss=0.1256, over 28515.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3884, pruned_loss=0.1394, over 5669116.87 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3846, pruned_loss=0.1351, over 5741432.56 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3888, pruned_loss=0.1395, over 5663475.36 frames. ], batch size: 65, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:05:56,009 INFO [optim.py:369] (1/2) 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 20:06:00,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5414, 4.4874, 1.8924, 1.5809], device='cuda:1'), covar=tensor([0.0918, 0.0221, 0.0764, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0489, 0.0319, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 20:06:06,775 INFO [train.py:968] (1/2) Epoch 7, batch 26700, giga_loss[loss=0.2913, simple_loss=0.3556, pruned_loss=0.1135, over 28665.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3871, pruned_loss=0.1389, over 5657875.11 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3854, pruned_loss=0.1357, over 5734287.75 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3868, pruned_loss=0.1386, over 5656904.93 frames. ], batch size: 85, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:06:55,719 INFO [train.py:968] (1/2) Epoch 7, batch 26750, giga_loss[loss=0.3184, simple_loss=0.3857, pruned_loss=0.1255, over 28856.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3867, pruned_loss=0.1375, over 5659916.41 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3852, pruned_loss=0.1355, over 5736141.60 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3867, pruned_loss=0.1374, over 5656828.88 frames. ], batch size: 285, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:07:32,914 INFO [optim.py:369] (1/2) 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 20:07:43,493 INFO [train.py:968] (1/2) Epoch 7, batch 26800, giga_loss[loss=0.2958, simple_loss=0.3703, pruned_loss=0.1107, over 28851.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3888, pruned_loss=0.1383, over 5667713.88 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3845, pruned_loss=0.1351, over 5739510.52 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3894, pruned_loss=0.1386, over 5661176.98 frames. ], batch size: 174, lr: 4.58e-03, grad_scale: 8.0
+2023-03-03 20:08:01,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=300466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:08:08,497 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9505, 1.6257, 1.4197, 1.3332], device='cuda:1'), covar=tensor([0.0570, 0.0579, 0.0909, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0446, 0.0493, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 20:08:33,076 INFO [train.py:968] (1/2) Epoch 7, batch 26850, giga_loss[loss=0.3426, simple_loss=0.4037, pruned_loss=0.1407, over 28997.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3897, pruned_loss=0.1397, over 5652569.66 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3847, pruned_loss=0.1353, over 5725876.77 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3901, pruned_loss=0.1399, over 5657019.33 frames. ], batch size: 128, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:09:07,100 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:09:17,900 INFO [train.py:968] (1/2) Epoch 7, batch 26900, giga_loss[loss=0.313, simple_loss=0.3922, pruned_loss=0.1169, over 28725.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3888, pruned_loss=0.1365, over 5665529.23 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3846, pruned_loss=0.1352, over 5727495.54 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3893, pruned_loss=0.1367, over 5666926.28 frames. ], batch size: 99, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:09:32,280 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 20:10:06,632 INFO [train.py:968] (1/2) Epoch 7, batch 26950, giga_loss[loss=0.4577, simple_loss=0.4673, pruned_loss=0.2241, over 26620.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3898, pruned_loss=0.1346, over 5661280.23 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3846, pruned_loss=0.1354, over 5711235.16 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3903, pruned_loss=0.1347, over 5674794.43 frames. ], batch size: 555, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:10:21,576 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8419, 4.6085, 4.3578, 2.2401], device='cuda:1'), covar=tensor([0.0565, 0.0824, 0.0918, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0953, 0.0905, 0.0802, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 20:10:40,544 INFO [optim.py:369] (1/2) 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 20:10:49,795 INFO [train.py:968] (1/2) Epoch 7, batch 27000, giga_loss[loss=0.3382, simple_loss=0.3981, pruned_loss=0.1392, over 29003.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3935, pruned_loss=0.1364, over 5671278.04 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1353, over 5712825.14 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3941, pruned_loss=0.1365, over 5679815.37 frames. ], batch size: 136, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:10:49,795 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 20:10:58,246 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-03 20:11:02,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5139, 1.7910, 1.9077, 1.4370], device='cuda:1'), covar=tensor([0.1605, 0.1969, 0.1269, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0720, 0.0814, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 20:11:27,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:11:29,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:11:45,762 INFO [train.py:968] (1/2) Epoch 7, batch 27050, giga_loss[loss=0.3324, simple_loss=0.3915, pruned_loss=0.1367, over 28409.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3963, pruned_loss=0.1394, over 5666786.81 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3847, pruned_loss=0.1356, over 5709914.26 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3968, pruned_loss=0.1393, over 5675529.07 frames. ], batch size: 71, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:11:56,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300709.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 20:11:58,573 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:11:58,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300712.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 20:12:26,182 INFO [optim.py:369] (1/2) 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 20:12:28,830 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300741.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 20:12:35,322 INFO [train.py:968] (1/2) Epoch 7, batch 27100, giga_loss[loss=0.3459, simple_loss=0.4012, pruned_loss=0.1453, over 29026.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3988, pruned_loss=0.143, over 5653787.51 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3847, pruned_loss=0.1356, over 5703941.49 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3996, pruned_loss=0.1431, over 5665244.35 frames. ], batch size: 128, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:13:24,714 INFO [train.py:968] (1/2) Epoch 7, batch 27150, libri_loss[loss=0.2594, simple_loss=0.3252, pruned_loss=0.09678, over 29611.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3969, pruned_loss=0.1429, over 5640129.17 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3835, pruned_loss=0.1349, over 5700044.59 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3991, pruned_loss=0.1438, over 5651396.08 frames. ], batch size: 69, lr: 4.58e-03, grad_scale: 2.0
+2023-03-03 20:14:05,683 INFO [optim.py:369] (1/2) 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 20:14:07,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=300841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:14:13,607 INFO [train.py:968] (1/2) Epoch 7, batch 27200, giga_loss[loss=0.3631, simple_loss=0.396, pruned_loss=0.1651, over 23519.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.397, pruned_loss=0.1431, over 5639382.70 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3837, pruned_loss=0.1351, over 5704862.10 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3989, pruned_loss=0.1439, over 5642293.14 frames. ], batch size: 705, lr: 4.58e-03, grad_scale: 4.0
+2023-03-03 20:14:57,583 INFO [train.py:968] (1/2) Epoch 7, batch 27250, giga_loss[loss=0.2761, simple_loss=0.359, pruned_loss=0.09654, over 28933.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3964, pruned_loss=0.1414, over 5642012.95 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3838, pruned_loss=0.1353, over 5699094.46 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3981, pruned_loss=0.1421, over 5647440.20 frames. ], batch size: 164, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:15:35,375 INFO [optim.py:369] (1/2) 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 20:15:44,656 INFO [train.py:968] (1/2) Epoch 7, batch 27300, libri_loss[loss=0.3417, simple_loss=0.3944, pruned_loss=0.1446, over 27921.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3965, pruned_loss=0.1394, over 5658562.44 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3836, pruned_loss=0.1351, over 5701806.98 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3983, pruned_loss=0.1402, over 5659783.28 frames. ], batch size: 116, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:16:03,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4633, 1.7842, 1.5402, 1.5770], device='cuda:1'), covar=tensor([0.0616, 0.0253, 0.0253, 0.0620], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0116, 0.0120, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0074], device='cuda:1')
+2023-03-03 20:16:17,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=300984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:16:20,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=300987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:16:32,532 INFO [train.py:968] (1/2) Epoch 7, batch 27350, giga_loss[loss=0.2972, simple_loss=0.3723, pruned_loss=0.111, over 29048.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3965, pruned_loss=0.1395, over 5662483.84 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3834, pruned_loss=0.1349, over 5706688.58 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3984, pruned_loss=0.1404, over 5658192.37 frames. ], batch size: 128, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:16:43,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:16:50,726 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:17:14,676 INFO [optim.py:369] (1/2) 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 20:17:16,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1979, 2.5418, 1.2207, 1.2457], device='cuda:1'), covar=tensor([0.0964, 0.0429, 0.0858, 0.1371], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0491, 0.0319, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 20:17:21,311 INFO [train.py:968] (1/2) Epoch 7, batch 27400, giga_loss[loss=0.3761, simple_loss=0.4225, pruned_loss=0.1649, over 28928.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.396, pruned_loss=0.1395, over 5669499.12 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3837, pruned_loss=0.1351, over 5708602.39 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3975, pruned_loss=0.1401, over 5663439.63 frames. ], batch size: 227, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:18:07,950 INFO [train.py:968] (1/2) Epoch 7, batch 27450, giga_loss[loss=0.3007, simple_loss=0.3692, pruned_loss=0.1161, over 29024.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3938, pruned_loss=0.1384, over 5674512.11 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3842, pruned_loss=0.1354, over 5707793.16 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3951, pruned_loss=0.1389, over 5669095.85 frames. ], batch size: 128, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:18:23,856 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-03 20:18:28,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3592, 1.5468, 1.3155, 1.6787], device='cuda:1'), covar=tensor([0.2000, 0.1921, 0.1948, 0.1739], device='cuda:1'), in_proj_covar=tensor([0.1195, 0.0907, 0.1051, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 20:18:49,601 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 27500, giga_loss[loss=0.3007, simple_loss=0.3667, pruned_loss=0.1174, over 28757.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3899, pruned_loss=0.1373, over 5659603.43 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3833, pruned_loss=0.1347, over 5711782.03 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.392, pruned_loss=0.1384, over 5650299.42 frames. ], batch size: 119, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:19:41,278 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:19:44,910 INFO [train.py:968] (1/2) Epoch 7, batch 27550, giga_loss[loss=0.3492, simple_loss=0.3918, pruned_loss=0.1532, over 27556.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3886, pruned_loss=0.137, over 5641508.24 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3835, pruned_loss=0.1348, over 5703672.12 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3903, pruned_loss=0.138, over 5639021.91 frames. ], batch size: 472, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:20:22,074 INFO [optim.py:369] (1/2) 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 20:20:29,847 INFO [train.py:968] (1/2) Epoch 7, batch 27600, giga_loss[loss=0.3012, simple_loss=0.3661, pruned_loss=0.1181, over 28853.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3873, pruned_loss=0.1366, over 5651169.34 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3828, pruned_loss=0.1341, over 5703185.58 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3896, pruned_loss=0.1381, over 5646227.27 frames. ], batch size: 199, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:21:13,423 INFO [train.py:968] (1/2) Epoch 7, batch 27650, giga_loss[loss=0.3366, simple_loss=0.3937, pruned_loss=0.1397, over 28909.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3873, pruned_loss=0.1375, over 5654412.21 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.383, pruned_loss=0.1342, over 5708344.52 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.389, pruned_loss=0.1387, over 5644484.44 frames. ], batch size: 145, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:21:51,129 INFO [optim.py:369] (1/2) 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 20:21:58,624 INFO [train.py:968] (1/2) Epoch 7, batch 27700, giga_loss[loss=0.3551, simple_loss=0.4039, pruned_loss=0.1531, over 28697.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3857, pruned_loss=0.136, over 5641524.20 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3831, pruned_loss=0.1343, over 5695756.86 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.387, pruned_loss=0.1369, over 5645164.00 frames. ], batch size: 262, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:22:18,519 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-03 20:22:31,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:22:45,242 INFO [train.py:968] (1/2) Epoch 7, batch 27750, giga_loss[loss=0.2821, simple_loss=0.3472, pruned_loss=0.1085, over 28586.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3828, pruned_loss=0.1321, over 5652205.93 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3835, pruned_loss=0.1346, over 5690467.25 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3836, pruned_loss=0.1325, over 5658179.22 frames. ], batch size: 85, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:22:55,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-03 20:23:24,532 INFO [optim.py:369] (1/2) 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 20:23:33,609 INFO [train.py:968] (1/2) Epoch 7, batch 27800, giga_loss[loss=0.3074, simple_loss=0.3791, pruned_loss=0.1178, over 28883.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3821, pruned_loss=0.1312, over 5656722.20 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3842, pruned_loss=0.1353, over 5693714.66 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3821, pruned_loss=0.1309, over 5657930.59 frames. ], batch size: 227, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:23:54,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5882, 1.4680, 1.2025, 1.1307], device='cuda:1'), covar=tensor([0.0702, 0.0596, 0.0983, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0450, 0.0498, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 20:24:27,136 INFO [train.py:968] (1/2) Epoch 7, batch 27850, giga_loss[loss=0.2869, simple_loss=0.3599, pruned_loss=0.1069, over 28856.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3814, pruned_loss=0.1314, over 5630905.31 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3844, pruned_loss=0.1356, over 5686359.57 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3811, pruned_loss=0.1308, over 5638519.24 frames. ], batch size: 174, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:24:54,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:24:58,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:25:00,001 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=301533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:25:08,263 INFO [optim.py:369] (1/2) 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 20:25:19,755 INFO [train.py:968] (1/2) Epoch 7, batch 27900, giga_loss[loss=0.3624, simple_loss=0.4023, pruned_loss=0.1613, over 27576.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3777, pruned_loss=0.1298, over 5660598.20 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3843, pruned_loss=0.1356, over 5693936.39 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3775, pruned_loss=0.1292, over 5658678.63 frames. ], batch size: 472, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:25:29,645 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:25:41,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:25:53,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 20:26:10,134 INFO [train.py:968] (1/2) Epoch 7, batch 27950, giga_loss[loss=0.3513, simple_loss=0.4214, pruned_loss=0.1406, over 28985.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3788, pruned_loss=0.131, over 5664730.48 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3848, pruned_loss=0.136, over 5699221.74 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.378, pruned_loss=0.13, over 5657223.62 frames. ], batch size: 136, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:26:36,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7555, 1.0099, 3.4334, 2.7584], device='cuda:1'), covar=tensor([0.1764, 0.2479, 0.0466, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0606, 0.0566, 0.0807, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 20:26:44,362 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 28000, giga_loss[loss=0.3434, simple_loss=0.3986, pruned_loss=0.1441, over 27442.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3808, pruned_loss=0.1314, over 5661842.48 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3843, pruned_loss=0.1355, over 5706126.77 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3805, pruned_loss=0.131, over 5648564.38 frames. ], batch size: 472, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:27:21,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9516, 1.1380, 3.7446, 3.0091], device='cuda:1'), covar=tensor([0.1739, 0.2520, 0.0409, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0605, 0.0566, 0.0806, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 20:27:41,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7461, 5.5064, 5.2012, 2.4222], device='cuda:1'), covar=tensor([0.0458, 0.0763, 0.0862, 0.1879], device='cuda:1'), in_proj_covar=tensor([0.0968, 0.0914, 0.0818, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-03 20:27:43,459 INFO [train.py:968] (1/2) Epoch 7, batch 28050, giga_loss[loss=0.3457, simple_loss=0.3989, pruned_loss=0.1463, over 27617.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3826, pruned_loss=0.1328, over 5656029.32 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3846, pruned_loss=0.1356, over 5709245.57 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.382, pruned_loss=0.1323, over 5641999.86 frames. ], batch size: 472, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:27:57,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=301713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:27:58,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=301716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:28:19,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-03 20:28:20,585 INFO [optim.py:369] (1/2) 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 20:28:25,342 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=301745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:28:30,425 INFO [train.py:968] (1/2) Epoch 7, batch 28100, giga_loss[loss=0.3384, simple_loss=0.377, pruned_loss=0.1498, over 23264.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3823, pruned_loss=0.1323, over 5651462.39 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3851, pruned_loss=0.1358, over 5703136.12 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3814, pruned_loss=0.1316, over 5644666.08 frames. ], batch size: 705, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:28:54,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-03 20:29:14,697 INFO [train.py:968] (1/2) Epoch 7, batch 28150, giga_loss[loss=0.2984, simple_loss=0.3634, pruned_loss=0.1167, over 28767.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3827, pruned_loss=0.1329, over 5657952.53 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3849, pruned_loss=0.1356, over 5705430.14 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3821, pruned_loss=0.1325, over 5649157.80 frames. ], batch size: 284, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:29:36,248 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 7, batch 28200, giga_loss[loss=0.338, simple_loss=0.3724, pruned_loss=0.1518, over 23519.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3858, pruned_loss=0.1354, over 5661225.47 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3849, pruned_loss=0.1356, over 5708609.03 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3853, pruned_loss=0.1351, over 5650289.93 frames. ], batch size: 705, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:30:47,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2104, 1.1708, 1.0896, 0.9628], device='cuda:1'), covar=tensor([0.0712, 0.0506, 0.0955, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0446, 0.0499, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 20:30:48,887 INFO [train.py:968] (1/2) Epoch 7, batch 28250, giga_loss[loss=0.3067, simple_loss=0.376, pruned_loss=0.1187, over 28805.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3871, pruned_loss=0.1356, over 5670792.16 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.385, pruned_loss=0.1356, over 5709089.84 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3867, pruned_loss=0.1353, over 5660787.17 frames. ], batch size: 119, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:30:56,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=301908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:31:32,495 INFO [optim.py:369] (1/2) 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 20:31:42,950 INFO [train.py:968] (1/2) Epoch 7, batch 28300, giga_loss[loss=0.403, simple_loss=0.4156, pruned_loss=0.1952, over 23356.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3888, pruned_loss=0.1371, over 5657401.62 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3854, pruned_loss=0.1359, over 5708258.81 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3882, pruned_loss=0.1366, over 5649618.62 frames. ], batch size: 705, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:31:59,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5400, 4.4492, 1.7823, 1.6448], device='cuda:1'), covar=tensor([0.0910, 0.0226, 0.0802, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0497, 0.0320, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 20:32:33,659 INFO [train.py:968] (1/2) Epoch 7, batch 28350, giga_loss[loss=0.3918, simple_loss=0.416, pruned_loss=0.1837, over 23800.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3891, pruned_loss=0.1382, over 5650658.69 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3857, pruned_loss=0.1361, over 5710023.90 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3884, pruned_loss=0.1377, over 5642364.83 frames. ], batch size: 705, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:33:02,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2586, 1.7039, 1.5755, 1.1843], device='cuda:1'), covar=tensor([0.1398, 0.2146, 0.1199, 0.1399], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0721, 0.0810, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 20:33:17,065 INFO [optim.py:369] (1/2) 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 20:33:22,431 INFO [train.py:968] (1/2) Epoch 7, batch 28400, giga_loss[loss=0.3458, simple_loss=0.4015, pruned_loss=0.1451, over 28600.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3904, pruned_loss=0.1374, over 5665446.81 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3848, pruned_loss=0.1355, over 5715436.34 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3908, pruned_loss=0.1375, over 5652322.13 frames. ], batch size: 78, lr: 4.57e-03, grad_scale: 8.0
+2023-03-03 20:33:23,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:33:27,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:33:57,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:34:02,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0570, 5.2202, 2.0568, 2.4611], device='cuda:1'), covar=tensor([0.0789, 0.0196, 0.0726, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0499, 0.0322, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 20:34:10,526 INFO [train.py:968] (1/2) Epoch 7, batch 28450, giga_loss[loss=0.3693, simple_loss=0.4181, pruned_loss=0.1602, over 28933.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.391, pruned_loss=0.1375, over 5673023.72 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3848, pruned_loss=0.1357, over 5717662.89 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3915, pruned_loss=0.1375, over 5658647.74 frames. ], batch size: 227, lr: 4.57e-03, grad_scale: 4.0
+2023-03-03 20:34:37,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6440, 5.4998, 5.2011, 2.3564], device='cuda:1'), covar=tensor([0.0341, 0.0520, 0.0626, 0.1819], device='cuda:1'), in_proj_covar=tensor([0.0971, 0.0918, 0.0823, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-03 20:34:50,979 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 7, batch 28500, giga_loss[loss=0.3514, simple_loss=0.3965, pruned_loss=0.1532, over 28931.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3894, pruned_loss=0.1376, over 5664403.02 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.385, pruned_loss=0.1358, over 5708896.57 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3896, pruned_loss=0.1375, over 5660156.25 frames. ], batch size: 145, lr: 4.57e-03, grad_scale: 2.0
+2023-03-03 20:35:55,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:35:56,937 INFO [train.py:968] (1/2) Epoch 7, batch 28550, libri_loss[loss=0.3087, simple_loss=0.3839, pruned_loss=0.1168, over 29615.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3894, pruned_loss=0.1379, over 5675183.12 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3847, pruned_loss=0.1354, over 5715447.20 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.39, pruned_loss=0.1383, over 5663986.97 frames. ], batch size: 91, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:36:51,179 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 28600, giga_loss[loss=0.3341, simple_loss=0.3864, pruned_loss=0.1409, over 27564.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3875, pruned_loss=0.1369, over 5679449.17 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3847, pruned_loss=0.1353, over 5716439.96 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.388, pruned_loss=0.1373, over 5669606.07 frames. ], batch size: 472, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:37:26,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4666, 1.8204, 1.8240, 1.3973], device='cuda:1'), covar=tensor([0.1518, 0.1863, 0.1137, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0719, 0.0812, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 20:37:33,918 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:37:38,270 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:37:42,410 INFO [train.py:968] (1/2) Epoch 7, batch 28650, giga_loss[loss=0.3748, simple_loss=0.3979, pruned_loss=0.1759, over 23444.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3864, pruned_loss=0.1364, over 5679906.44 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3843, pruned_loss=0.135, over 5718218.04 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3871, pruned_loss=0.1371, over 5669466.55 frames. ], batch size: 705, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:38:22,688 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:38:25,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:38:29,454 INFO [train.py:968] (1/2) Epoch 7, batch 28700, giga_loss[loss=0.2762, simple_loss=0.35, pruned_loss=0.1012, over 29076.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3859, pruned_loss=0.1363, over 5657722.42 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3845, pruned_loss=0.1349, over 5709879.22 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3864, pruned_loss=0.137, over 5655762.27 frames. ], batch size: 128, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:38:54,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:39:19,448 INFO [train.py:968] (1/2) Epoch 7, batch 28750, libri_loss[loss=0.3279, simple_loss=0.3908, pruned_loss=0.1325, over 27845.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3863, pruned_loss=0.1369, over 5655171.37 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3842, pruned_loss=0.1346, over 5710984.93 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.387, pruned_loss=0.1376, over 5651808.74 frames. ], batch size: 115, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:39:47,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4812, 4.3127, 4.1087, 1.8278], device='cuda:1'), covar=tensor([0.0521, 0.0628, 0.0689, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.0976, 0.0921, 0.0825, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-03 20:40:01,820 INFO [optim.py:369] (1/2) 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 20:40:02,651 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-03 20:40:09,858 INFO [train.py:968] (1/2) Epoch 7, batch 28800, giga_loss[loss=0.4204, simple_loss=0.434, pruned_loss=0.2034, over 26607.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3883, pruned_loss=0.1391, over 5650242.21 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3844, pruned_loss=0.1347, over 5712620.27 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3887, pruned_loss=0.1397, over 5645728.30 frames. ], batch size: 555, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:40:57,970 INFO [train.py:968] (1/2) Epoch 7, batch 28850, giga_loss[loss=0.3378, simple_loss=0.3836, pruned_loss=0.1459, over 28907.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3887, pruned_loss=0.1393, over 5644670.15 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3846, pruned_loss=0.1349, over 5706417.21 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3889, pruned_loss=0.1396, over 5645053.90 frames. ], batch size: 106, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:41:12,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:41:43,265 INFO [optim.py:369] (1/2) 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 20:41:48,341 INFO [train.py:968] (1/2) Epoch 7, batch 28900, giga_loss[loss=0.3955, simple_loss=0.4192, pruned_loss=0.1859, over 26549.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3894, pruned_loss=0.1408, over 5642632.21 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3843, pruned_loss=0.1348, over 5706425.12 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3898, pruned_loss=0.1412, over 5642608.48 frames. ], batch size: 555, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:42:29,075 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:42:34,711 INFO [train.py:968] (1/2) Epoch 7, batch 28950, libri_loss[loss=0.3308, simple_loss=0.3717, pruned_loss=0.1449, over 29377.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3897, pruned_loss=0.1416, over 5646523.30 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3847, pruned_loss=0.1354, over 5707313.15 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3898, pruned_loss=0.1415, over 5644495.86 frames. ], batch size: 67, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:43:15,848 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 29000, libri_loss[loss=0.391, simple_loss=0.431, pruned_loss=0.1755, over 29765.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3902, pruned_loss=0.1413, over 5627444.05 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3851, pruned_loss=0.1358, over 5691882.74 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.39, pruned_loss=0.141, over 5638134.11 frames. ], batch size: 87, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:43:25,873 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:43:30,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:43:40,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:43:42,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4978, 1.8288, 1.8094, 1.4007], device='cuda:1'), covar=tensor([0.1690, 0.2093, 0.1294, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0728, 0.0821, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 20:43:45,109 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:44:00,372 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 29050, libri_loss[loss=0.3787, simple_loss=0.4294, pruned_loss=0.164, over 29646.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3906, pruned_loss=0.1406, over 5641186.76 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3858, pruned_loss=0.136, over 5696217.72 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.39, pruned_loss=0.1402, over 5644282.52 frames. ], batch size: 88, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:44:49,201 INFO [optim.py:369] (1/2) 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 20:44:52,652 INFO [train.py:968] (1/2) Epoch 7, batch 29100, giga_loss[loss=0.3413, simple_loss=0.3971, pruned_loss=0.1428, over 28949.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3913, pruned_loss=0.1407, over 5655317.41 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3857, pruned_loss=0.136, over 5702020.03 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3909, pruned_loss=0.1406, over 5650354.46 frames. ], batch size: 227, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:44:59,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-03 20:45:10,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1485, 1.2350, 1.1535, 0.9541], device='cuda:1'), covar=tensor([0.0971, 0.0985, 0.0614, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1420, 0.1390, 0.1483], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-03 20:45:31,802 INFO [train.py:968] (1/2) Epoch 7, batch 29150, libri_loss[loss=0.3102, simple_loss=0.3673, pruned_loss=0.1266, over 29545.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3911, pruned_loss=0.1409, over 5663941.07 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3852, pruned_loss=0.1358, over 5702046.50 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3917, pruned_loss=0.1414, over 5656779.43 frames. ], batch size: 79, lr: 4.56e-03, grad_scale: 2.0
+2023-03-03 20:45:41,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:45:43,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:45:46,594 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=302816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:45:48,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=302819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:09,879 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:12,144 INFO [optim.py:369] (1/2) 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 20:46:14,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=302848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:16,034 INFO [train.py:968] (1/2) Epoch 7, batch 29200, giga_loss[loss=0.3473, simple_loss=0.4009, pruned_loss=0.1469, over 28919.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3918, pruned_loss=0.1415, over 5666424.89 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3849, pruned_loss=0.1357, over 5693132.50 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3926, pruned_loss=0.142, over 5667943.56 frames. ], batch size: 227, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:46:33,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=302870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:46:49,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 20:47:01,793 INFO [train.py:968] (1/2) Epoch 7, batch 29250, giga_loss[loss=0.302, simple_loss=0.3765, pruned_loss=0.1138, over 28542.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3926, pruned_loss=0.1416, over 5662625.36 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3851, pruned_loss=0.1356, over 5696793.80 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3933, pruned_loss=0.1423, over 5659733.27 frames. ], batch size: 85, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:47:41,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-03 20:47:51,143 INFO [optim.py:369] (1/2) 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 20:47:56,166 INFO [train.py:968] (1/2) Epoch 7, batch 29300, giga_loss[loss=0.3123, simple_loss=0.3827, pruned_loss=0.121, over 28873.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3931, pruned_loss=0.1406, over 5663400.03 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3851, pruned_loss=0.1355, over 5698825.93 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3938, pruned_loss=0.1412, over 5659034.08 frames. ], batch size: 174, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:48:17,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=302968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:48:44,229 INFO [train.py:968] (1/2) Epoch 7, batch 29350, giga_loss[loss=0.317, simple_loss=0.378, pruned_loss=0.128, over 28803.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3913, pruned_loss=0.1385, over 5665503.44 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3853, pruned_loss=0.1356, over 5699908.78 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3917, pruned_loss=0.1389, over 5660566.74 frames. ], batch size: 119, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:48:58,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9096, 1.9335, 1.6061, 2.2976], device='cuda:1'), covar=tensor([0.2121, 0.2173, 0.2278, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1200, 0.0907, 0.1058, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 20:49:08,996 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:49:27,617 INFO [optim.py:369] (1/2) 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 20:49:28,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4456, 1.7607, 1.6979, 1.3271], device='cuda:1'), covar=tensor([0.1455, 0.1998, 0.1179, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0722, 0.0818, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 20:49:30,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 20:49:30,938 INFO [train.py:968] (1/2) Epoch 7, batch 29400, giga_loss[loss=0.3744, simple_loss=0.4193, pruned_loss=0.1648, over 28934.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3892, pruned_loss=0.1374, over 5665423.02 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3851, pruned_loss=0.1355, over 5702776.20 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3898, pruned_loss=0.1379, over 5658254.95 frames. ], batch size: 227, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:49:45,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:50:05,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5386, 3.2584, 1.6469, 1.6913], device='cuda:1'), covar=tensor([0.0783, 0.0384, 0.0762, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0497, 0.0321, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 20:50:19,611 INFO [train.py:968] (1/2) Epoch 7, batch 29450, giga_loss[loss=0.3651, simple_loss=0.4137, pruned_loss=0.1582, over 28620.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3906, pruned_loss=0.1386, over 5659945.01 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3853, pruned_loss=0.1356, over 5701052.37 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.391, pruned_loss=0.139, over 5655481.62 frames. ], batch size: 307, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:50:29,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:50:32,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:51:04,459 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:51:05,657 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 29500, giga_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 28613.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3909, pruned_loss=0.139, over 5646941.86 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.385, pruned_loss=0.1355, over 5691006.14 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3915, pruned_loss=0.1394, over 5652661.76 frames. ], batch size: 92, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:51:24,586 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 20:51:58,368 INFO [train.py:968] (1/2) Epoch 7, batch 29550, libri_loss[loss=0.3424, simple_loss=0.4, pruned_loss=0.1424, over 29534.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.391, pruned_loss=0.1399, over 5653769.19 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3849, pruned_loss=0.1355, over 5689249.71 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3918, pruned_loss=0.1404, over 5658039.67 frames. ], batch size: 82, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:52:05,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:08,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:29,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5898, 2.3205, 1.7926, 0.6880], device='cuda:1'), covar=tensor([0.3202, 0.1745, 0.2386, 0.3754], device='cuda:1'), in_proj_covar=tensor([0.1473, 0.1387, 0.1441, 0.1211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 20:52:33,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:39,281 INFO [optim.py:369] (1/2) 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 20:52:39,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:52:45,282 INFO [train.py:968] (1/2) Epoch 7, batch 29600, giga_loss[loss=0.33, simple_loss=0.3932, pruned_loss=0.1334, over 28507.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3917, pruned_loss=0.1415, over 5648203.98 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3851, pruned_loss=0.1356, over 5692759.81 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3922, pruned_loss=0.1418, over 5647919.87 frames. ], batch size: 336, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:53:08,253 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:53:34,029 INFO [train.py:968] (1/2) Epoch 7, batch 29650, giga_loss[loss=0.2966, simple_loss=0.3672, pruned_loss=0.113, over 28791.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3912, pruned_loss=0.1408, over 5659375.19 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3849, pruned_loss=0.1354, over 5694751.74 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3919, pruned_loss=0.1412, over 5656940.03 frames. ], batch size: 263, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:54:17,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4835, 1.7999, 1.8175, 1.4069], device='cuda:1'), covar=tensor([0.1488, 0.1956, 0.1142, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0725, 0.0816, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 20:54:19,319 INFO [optim.py:369] (1/2) 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 20:54:20,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4452, 3.5844, 1.5404, 1.5074], device='cuda:1'), covar=tensor([0.0920, 0.0326, 0.0840, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0497, 0.0321, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 20:54:24,255 INFO [train.py:968] (1/2) Epoch 7, batch 29700, giga_loss[loss=0.3644, simple_loss=0.4142, pruned_loss=0.1573, over 27913.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3933, pruned_loss=0.1425, over 5639772.07 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3853, pruned_loss=0.1358, over 5683250.97 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.3935, pruned_loss=0.1425, over 5648553.33 frames. ], batch size: 412, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:54:35,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-03 20:54:42,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3420, 1.8639, 1.4646, 0.4480], device='cuda:1'), covar=tensor([0.1595, 0.1169, 0.1810, 0.2420], device='cuda:1'), in_proj_covar=tensor([0.1471, 0.1389, 0.1440, 0.1210], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 20:55:00,883 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:03,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:10,314 INFO [train.py:968] (1/2) Epoch 7, batch 29750, giga_loss[loss=0.3814, simple_loss=0.4302, pruned_loss=0.1663, over 28773.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3924, pruned_loss=0.1412, over 5661146.11 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3858, pruned_loss=0.1361, over 5690901.81 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3925, pruned_loss=0.1412, over 5660028.53 frames. ], batch size: 284, lr: 4.56e-03, grad_scale: 8.0
+2023-03-03 20:55:10,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:29,529 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:54,265 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:55:54,659 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 29800, giga_loss[loss=0.2901, simple_loss=0.3663, pruned_loss=0.1069, over 28922.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3916, pruned_loss=0.14, over 5665118.59 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3856, pruned_loss=0.1359, over 5693032.57 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3919, pruned_loss=0.1403, over 5661671.43 frames. ], batch size: 145, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:56:47,323 INFO [train.py:968] (1/2) Epoch 7, batch 29850, giga_loss[loss=0.3229, simple_loss=0.3862, pruned_loss=0.1297, over 28856.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3917, pruned_loss=0.1397, over 5658569.85 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3859, pruned_loss=0.1361, over 5693023.73 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3918, pruned_loss=0.1398, over 5655127.80 frames. ], batch size: 186, lr: 4.56e-03, grad_scale: 4.0
+2023-03-03 20:57:08,816 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-03 20:57:31,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:57:33,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:57:33,420 INFO [optim.py:369] (1/2) 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 20:57:35,321 INFO [train.py:968] (1/2) Epoch 7, batch 29900, giga_loss[loss=0.3305, simple_loss=0.3793, pruned_loss=0.1409, over 28894.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3898, pruned_loss=0.1387, over 5660811.90 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3862, pruned_loss=0.1366, over 5695208.29 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3897, pruned_loss=0.1384, over 5655551.83 frames. ], batch size: 186, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 20:57:58,404 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:58:10,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-03 20:58:21,038 INFO [train.py:968] (1/2) Epoch 7, batch 29950, giga_loss[loss=0.3041, simple_loss=0.3675, pruned_loss=0.1204, over 28859.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.388, pruned_loss=0.1377, over 5672449.28 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3856, pruned_loss=0.1361, over 5699712.22 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3885, pruned_loss=0.138, over 5663165.83 frames. ], batch size: 174, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 20:59:04,272 INFO [optim.py:369] (1/2) 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 20:59:07,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 20:59:08,155 INFO [train.py:968] (1/2) Epoch 7, batch 30000, giga_loss[loss=0.3386, simple_loss=0.3845, pruned_loss=0.1464, over 28876.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3862, pruned_loss=0.137, over 5662884.09 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3853, pruned_loss=0.1358, over 5700072.16 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3869, pruned_loss=0.1375, over 5654248.84 frames. ], batch size: 99, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 20:59:08,155 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 20:59:16,782 INFO [train.py:1012] (1/2) Epoch 7, validation: loss=0.2258, simple_loss=0.3317, pruned_loss=0.05999, over 944034.00 frames. 
+2023-03-03 20:59:16,783 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-03 20:59:17,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:00:06,436 INFO [train.py:968] (1/2) Epoch 7, batch 30050, giga_loss[loss=0.2984, simple_loss=0.3605, pruned_loss=0.1182, over 28894.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3818, pruned_loss=0.1342, over 5666465.37 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3849, pruned_loss=0.1355, over 5694287.13 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3827, pruned_loss=0.1348, over 5664489.35 frames. ], batch size: 213, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:00:48,307 INFO [optim.py:369] (1/2) 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 21:00:50,965 INFO [train.py:968] (1/2) Epoch 7, batch 30100, libri_loss[loss=0.3657, simple_loss=0.4192, pruned_loss=0.1561, over 29476.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3802, pruned_loss=0.1333, over 5679694.32 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3852, pruned_loss=0.1355, over 5691426.48 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3805, pruned_loss=0.1337, over 5679366.34 frames. ], batch size: 85, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:01:31,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:01:33,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:01:34,345 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:01:38,007 INFO [train.py:968] (1/2) Epoch 7, batch 30150, giga_loss[loss=0.3465, simple_loss=0.395, pruned_loss=0.1489, over 28961.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3815, pruned_loss=0.135, over 5681122.63 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.385, pruned_loss=0.1354, over 5683649.28 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3817, pruned_loss=0.1353, over 5686891.36 frames. ], batch size: 213, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:01:59,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=303820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:02:03,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:02:25,040 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 30200, giga_loss[loss=0.3556, simple_loss=0.4039, pruned_loss=0.1536, over 28273.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3817, pruned_loss=0.1344, over 5678572.63 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3854, pruned_loss=0.1358, over 5685028.67 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3816, pruned_loss=0.1343, over 5681782.32 frames. ], batch size: 368, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:03:08,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-03 21:03:17,144 INFO [train.py:968] (1/2) Epoch 7, batch 30250, giga_loss[loss=0.3166, simple_loss=0.3818, pruned_loss=0.1257, over 28740.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3801, pruned_loss=0.1316, over 5673224.03 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3851, pruned_loss=0.1359, over 5681280.51 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3801, pruned_loss=0.1314, over 5678366.46 frames. ], batch size: 262, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:03:32,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1371, 1.1631, 3.4765, 3.1052], device='cuda:1'), covar=tensor([0.1565, 0.2529, 0.0394, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0610, 0.0569, 0.0811, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 21:04:06,601 INFO [optim.py:369] (1/2) 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 21:04:08,572 INFO [train.py:968] (1/2) Epoch 7, batch 30300, giga_loss[loss=0.257, simple_loss=0.3398, pruned_loss=0.0871, over 29008.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3776, pruned_loss=0.1291, over 5663701.33 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3851, pruned_loss=0.136, over 5688221.63 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3776, pruned_loss=0.1287, over 5661135.64 frames. ], batch size: 128, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:04:23,206 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=303963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:25,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=303966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:25,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=303967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:53,338 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=303995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:04:57,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6551, 1.7913, 1.7311, 1.6265], device='cuda:1'), covar=tensor([0.1205, 0.1708, 0.1475, 0.1574], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0720, 0.0641, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 21:04:59,908 INFO [train.py:968] (1/2) Epoch 7, batch 30350, giga_loss[loss=0.302, simple_loss=0.3708, pruned_loss=0.1166, over 28861.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3729, pruned_loss=0.1246, over 5662994.85 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3846, pruned_loss=0.1359, over 5693305.35 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.373, pruned_loss=0.1241, over 5655512.92 frames. ], batch size: 199, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:05:24,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:05:47,805 INFO [optim.py:369] (1/2) 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 21:05:49,229 INFO [train.py:968] (1/2) Epoch 7, batch 30400, giga_loss[loss=0.3374, simple_loss=0.3834, pruned_loss=0.1457, over 26806.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3684, pruned_loss=0.1203, over 5659738.62 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3841, pruned_loss=0.1357, over 5696237.78 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1199, over 5650574.51 frames. ], batch size: 555, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:06:21,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.35 vs. limit=5.0
+2023-03-03 21:06:40,578 INFO [train.py:968] (1/2) Epoch 7, batch 30450, giga_loss[loss=0.3027, simple_loss=0.3756, pruned_loss=0.1149, over 28818.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3664, pruned_loss=0.1167, over 5653367.32 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.384, pruned_loss=0.1357, over 5699126.44 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3664, pruned_loss=0.116, over 5642918.63 frames. ], batch size: 199, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:07:32,899 INFO [optim.py:369] (1/2) 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 21:07:34,382 INFO [train.py:968] (1/2) Epoch 7, batch 30500, giga_loss[loss=0.2718, simple_loss=0.3501, pruned_loss=0.09681, over 28940.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3664, pruned_loss=0.116, over 5651406.59 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3838, pruned_loss=0.1356, over 5702321.72 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3664, pruned_loss=0.1153, over 5639559.30 frames. ], batch size: 145, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:07:55,174 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:07:58,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:07:59,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:08:24,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9172, 2.9918, 2.0956, 0.8392], device='cuda:1'), covar=tensor([0.3988, 0.1684, 0.2118, 0.3848], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1365, 0.1422, 0.1194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 21:08:25,166 INFO [train.py:968] (1/2) Epoch 7, batch 30550, giga_loss[loss=0.2427, simple_loss=0.3301, pruned_loss=0.07766, over 28998.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3647, pruned_loss=0.1148, over 5650308.76 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3834, pruned_loss=0.1354, over 5706850.89 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3647, pruned_loss=0.1139, over 5635509.19 frames. ], batch size: 155, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:08:27,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:08:29,807 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:08:30,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 21:09:15,369 INFO [optim.py:369] (1/2) 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 21:09:15,963 INFO [train.py:968] (1/2) Epoch 7, batch 30600, giga_loss[loss=0.27, simple_loss=0.3446, pruned_loss=0.0977, over 28566.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3609, pruned_loss=0.1121, over 5632948.63 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3834, pruned_loss=0.1356, over 5689672.08 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3606, pruned_loss=0.111, over 5635850.85 frames. ], batch size: 307, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:09:40,245 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-03 21:09:57,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:04,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5486, 2.2833, 1.6728, 0.6014], device='cuda:1'), covar=tensor([0.3230, 0.1669, 0.2438, 0.3534], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1369, 0.1424, 0.1194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 21:10:05,846 INFO [train.py:968] (1/2) Epoch 7, batch 30650, giga_loss[loss=0.2803, simple_loss=0.3524, pruned_loss=0.1041, over 28325.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3588, pruned_loss=0.1109, over 5638713.56 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3829, pruned_loss=0.1354, over 5694620.96 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3585, pruned_loss=0.1097, over 5635623.31 frames. ], batch size: 71, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:10:19,312 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:21,575 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:45,500 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:10:50,995 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 30700, libri_loss[loss=0.426, simple_loss=0.4443, pruned_loss=0.2039, over 19209.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3599, pruned_loss=0.1116, over 5635057.01 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3828, pruned_loss=0.1358, over 5687024.63 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3592, pruned_loss=0.1096, over 5638795.53 frames. ], batch size: 188, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:11:10,226 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304367.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:11:22,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2590, 2.5892, 1.2092, 1.3269], device='cuda:1'), covar=tensor([0.0931, 0.0335, 0.0931, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0495, 0.0323, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 21:11:26,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6534, 1.5992, 1.1873, 1.3054], device='cuda:1'), covar=tensor([0.0627, 0.0482, 0.0875, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0334, 0.0436, 0.0486, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 21:11:41,246 INFO [train.py:968] (1/2) Epoch 7, batch 30750, giga_loss[loss=0.2426, simple_loss=0.322, pruned_loss=0.08163, over 28766.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3586, pruned_loss=0.1101, over 5638924.68 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3829, pruned_loss=0.136, over 5686548.62 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3575, pruned_loss=0.108, over 5641698.22 frames. ], batch size: 119, lr: 4.55e-03, grad_scale: 2.0
+2023-03-03 21:11:43,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2186, 3.1024, 1.2826, 1.3677], device='cuda:1'), covar=tensor([0.1149, 0.0377, 0.1073, 0.1557], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0494, 0.0322, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 21:12:30,696 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 30800, giga_loss[loss=0.2577, simple_loss=0.3422, pruned_loss=0.08662, over 28847.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3559, pruned_loss=0.1077, over 5630550.39 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3827, pruned_loss=0.136, over 5671066.07 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3548, pruned_loss=0.1054, over 5645310.20 frames. ], batch size: 174, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:13:07,378 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:13:07,501 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-03 21:13:10,360 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 30850, giga_loss[loss=0.2692, simple_loss=0.3259, pruned_loss=0.1062, over 26692.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.351, pruned_loss=0.1048, over 5630027.40 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3816, pruned_loss=0.1353, over 5677017.75 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3504, pruned_loss=0.103, over 5635599.36 frames. ], batch size: 555, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:13:38,209 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:14:08,303 INFO [train.py:968] (1/2) Epoch 7, batch 30900, giga_loss[loss=0.2554, simple_loss=0.3307, pruned_loss=0.09003, over 28855.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3486, pruned_loss=0.1038, over 5639288.04 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3802, pruned_loss=0.1345, over 5680577.71 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3484, pruned_loss=0.102, over 5639234.60 frames. ], batch size: 145, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:14:37,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:14:59,340 INFO [train.py:968] (1/2) Epoch 7, batch 30950, giga_loss[loss=0.2553, simple_loss=0.3393, pruned_loss=0.08571, over 28981.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3467, pruned_loss=0.1031, over 5632740.59 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.38, pruned_loss=0.1345, over 5681596.94 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.346, pruned_loss=0.1011, over 5631231.99 frames. ], batch size: 164, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:15:50,326 INFO [optim.py:369] (1/2) 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 21:15:52,062 INFO [train.py:968] (1/2) Epoch 7, batch 31000, giga_loss[loss=0.348, simple_loss=0.3917, pruned_loss=0.1522, over 26512.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3466, pruned_loss=0.1031, over 5627706.51 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3792, pruned_loss=0.134, over 5687589.91 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3459, pruned_loss=0.1011, over 5619601.72 frames. ], batch size: 555, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:16:06,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:16:44,987 INFO [train.py:968] (1/2) Epoch 7, batch 31050, libri_loss[loss=0.3115, simple_loss=0.3643, pruned_loss=0.1294, over 27539.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3512, pruned_loss=0.1052, over 5642869.48 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.379, pruned_loss=0.1341, over 5693078.43 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3497, pruned_loss=0.1024, over 5629175.77 frames. ], batch size: 115, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:17:12,105 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:17:15,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:17:37,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=304742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:17:46,181 INFO [optim.py:369] (1/2) 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 21:17:47,116 INFO [train.py:968] (1/2) Epoch 7, batch 31100, giga_loss[loss=0.2306, simple_loss=0.3183, pruned_loss=0.07148, over 28939.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3522, pruned_loss=0.105, over 5657039.08 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3785, pruned_loss=0.1339, over 5695932.20 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.351, pruned_loss=0.1025, over 5643070.13 frames. ], batch size: 174, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:17:51,152 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:18:50,086 INFO [train.py:968] (1/2) Epoch 7, batch 31150, giga_loss[loss=0.2788, simple_loss=0.3534, pruned_loss=0.1021, over 28608.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3522, pruned_loss=0.1049, over 5672442.16 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.378, pruned_loss=0.1335, over 5701038.02 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3511, pruned_loss=0.1026, over 5655884.79 frames. ], batch size: 307, lr: 4.55e-03, grad_scale: 4.0
+2023-03-03 21:19:01,449 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:19:06,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:19:42,217 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:19:54,271 INFO [optim.py:369] (1/2) 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 21:19:55,372 INFO [train.py:968] (1/2) Epoch 7, batch 31200, giga_loss[loss=0.3308, simple_loss=0.3792, pruned_loss=0.1412, over 26879.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3489, pruned_loss=0.1021, over 5664515.91 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3776, pruned_loss=0.1333, over 5702200.83 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3482, pruned_loss=0.1003, over 5650457.24 frames. ], batch size: 555, lr: 4.55e-03, grad_scale: 8.0
+2023-03-03 21:20:43,039 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=304885.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:20:46,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=304888.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:20:59,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2819, 1.5747, 1.4644, 1.4348], device='cuda:1'), covar=tensor([0.1348, 0.1542, 0.1729, 0.1480], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0710, 0.0639, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 21:20:59,384 INFO [train.py:968] (1/2) Epoch 7, batch 31250, giga_loss[loss=0.2594, simple_loss=0.3412, pruned_loss=0.08877, over 28443.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.347, pruned_loss=0.09999, over 5672309.01 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.376, pruned_loss=0.1325, over 5708270.65 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.347, pruned_loss=0.09826, over 5654281.51 frames. ], batch size: 336, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:21:04,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4053, 2.0829, 1.5434, 0.6756], device='cuda:1'), covar=tensor([0.2615, 0.1497, 0.2265, 0.2984], device='cuda:1'), in_proj_covar=tensor([0.1452, 0.1370, 0.1431, 0.1201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 21:21:14,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=304911.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:21:20,233 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=304917.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:22:04,483 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 31300, giga_loss[loss=0.2525, simple_loss=0.3184, pruned_loss=0.09328, over 27611.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3443, pruned_loss=0.09917, over 5669674.23 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.376, pruned_loss=0.1324, over 5705584.07 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.344, pruned_loss=0.09754, over 5657839.59 frames. ], batch size: 472, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:22:08,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5845, 4.4300, 4.1919, 1.7972], device='cuda:1'), covar=tensor([0.0436, 0.0590, 0.0687, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0881, 0.0784, 0.0617], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 21:23:02,650 INFO [train.py:968] (1/2) Epoch 7, batch 31350, giga_loss[loss=0.2406, simple_loss=0.3221, pruned_loss=0.07954, over 28878.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3439, pruned_loss=0.0994, over 5662888.34 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3761, pruned_loss=0.1326, over 5699574.73 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3429, pruned_loss=0.09717, over 5658146.23 frames. ], batch size: 164, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:23:15,644 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:23:22,347 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:23:48,392 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 7, batch 31400, giga_loss[loss=0.2474, simple_loss=0.3312, pruned_loss=0.0818, over 28980.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.343, pruned_loss=0.09923, over 5667128.41 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3751, pruned_loss=0.1321, over 5699750.89 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3424, pruned_loss=0.09719, over 5662192.68 frames. ], batch size: 285, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:24:14,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:24:24,388 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:24:59,025 INFO [train.py:968] (1/2) Epoch 7, batch 31450, giga_loss[loss=0.2511, simple_loss=0.3419, pruned_loss=0.08013, over 29012.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.344, pruned_loss=0.09882, over 5666694.04 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.375, pruned_loss=0.132, over 5703162.07 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3431, pruned_loss=0.09666, over 5659090.45 frames. ], batch size: 136, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:26:00,917 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 31500, giga_loss[loss=0.2481, simple_loss=0.3088, pruned_loss=0.09375, over 24650.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3458, pruned_loss=0.09975, over 5665565.18 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3741, pruned_loss=0.1315, over 5706226.37 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3448, pruned_loss=0.09748, over 5655741.93 frames. ], batch size: 705, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:26:45,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:27:02,226 INFO [train.py:968] (1/2) Epoch 7, batch 31550, giga_loss[loss=0.2633, simple_loss=0.3391, pruned_loss=0.0937, over 28887.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3425, pruned_loss=0.09784, over 5675230.68 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3739, pruned_loss=0.1314, over 5712351.73 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3411, pruned_loss=0.09519, over 5660820.55 frames. ], batch size: 213, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:28:11,752 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 31600, giga_loss[loss=0.2682, simple_loss=0.3411, pruned_loss=0.09769, over 28155.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.345, pruned_loss=0.09967, over 5678731.08 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3738, pruned_loss=0.1314, over 5715060.29 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3437, pruned_loss=0.09719, over 5664590.21 frames. ], batch size: 412, lr: 4.54e-03, grad_scale: 8.0
+2023-03-03 21:28:49,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305286.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:29:06,300 INFO [train.py:968] (1/2) Epoch 7, batch 31650, giga_loss[loss=0.3255, simple_loss=0.3848, pruned_loss=0.1331, over 26953.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3496, pruned_loss=0.1014, over 5655837.34 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3742, pruned_loss=0.1318, over 5700466.91 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.09764, over 5654655.89 frames. ], batch size: 555, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:30:11,251 INFO [train.py:968] (1/2) Epoch 7, batch 31700, giga_loss[loss=0.2577, simple_loss=0.3501, pruned_loss=0.08267, over 28464.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3529, pruned_loss=0.1009, over 5647356.53 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3741, pruned_loss=0.132, over 5684852.59 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3505, pruned_loss=0.09728, over 5657866.04 frames. ], batch size: 336, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:30:12,993 INFO [optim.py:369] (1/2) 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 21:30:52,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:30:55,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1864, 1.4221, 1.3904, 1.2812], device='cuda:1'), covar=tensor([0.1168, 0.1413, 0.1712, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0710, 0.0638, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 21:30:57,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305390.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:31:09,020 INFO [train.py:968] (1/2) Epoch 7, batch 31750, giga_loss[loss=0.3223, simple_loss=0.3924, pruned_loss=0.1261, over 28897.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3517, pruned_loss=0.09953, over 5648136.51 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3732, pruned_loss=0.1314, over 5692016.12 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.35, pruned_loss=0.09609, over 5648880.70 frames. ], batch size: 213, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:31:23,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:31:28,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7107, 3.6854, 1.7072, 1.5516], device='cuda:1'), covar=tensor([0.0787, 0.0204, 0.0769, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0491, 0.0324, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 21:31:43,213 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:31:47,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305432.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:31:53,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:32:00,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:32:05,690 INFO [train.py:968] (1/2) Epoch 7, batch 31800, giga_loss[loss=0.2607, simple_loss=0.3529, pruned_loss=0.08428, over 28411.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3504, pruned_loss=0.098, over 5657224.07 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.373, pruned_loss=0.1313, over 5698824.22 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3485, pruned_loss=0.09444, over 5650634.40 frames. ], batch size: 60, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:32:08,845 INFO [optim.py:369] (1/2) 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 21:32:22,115 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305461.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:32:54,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4262, 2.1064, 1.4874, 0.5482], device='cuda:1'), covar=tensor([0.2885, 0.1384, 0.2144, 0.3274], device='cuda:1'), in_proj_covar=tensor([0.1457, 0.1376, 0.1438, 0.1208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 21:33:11,189 INFO [train.py:968] (1/2) Epoch 7, batch 31850, giga_loss[loss=0.3325, simple_loss=0.3899, pruned_loss=0.1376, over 27641.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3517, pruned_loss=0.09955, over 5659641.32 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3729, pruned_loss=0.1313, over 5701719.67 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.35, pruned_loss=0.0963, over 5651383.55 frames. ], batch size: 472, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:33:22,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-03 21:33:44,662 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:33:47,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:33:52,191 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305533.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:33:57,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305536.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:34:17,062 INFO [train.py:968] (1/2) Epoch 7, batch 31900, giga_loss[loss=0.282, simple_loss=0.3618, pruned_loss=0.1011, over 28747.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3522, pruned_loss=0.101, over 5664753.00 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3732, pruned_loss=0.1315, over 5703441.67 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3499, pruned_loss=0.09727, over 5655302.93 frames. ], batch size: 119, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:34:19,387 INFO [optim.py:369] (1/2) 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 21:34:26,906 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:34:33,952 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=305562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:34:44,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305565.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:35:02,532 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:05,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:13,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:20,729 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:23,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:35:37,502 INFO [train.py:968] (1/2) Epoch 7, batch 31950, giga_loss[loss=0.2527, simple_loss=0.3341, pruned_loss=0.08565, over 28422.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3524, pruned_loss=0.1016, over 5674205.04 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3733, pruned_loss=0.1317, over 5706268.86 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3503, pruned_loss=0.0981, over 5663907.88 frames. ], batch size: 336, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:35:53,095 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:36:04,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:36:19,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3328, 1.5839, 1.4707, 1.6396], device='cuda:1'), covar=tensor([0.0766, 0.0302, 0.0320, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0122, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0075], device='cuda:1')
+2023-03-03 21:36:32,328 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:36:40,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2158, 1.2972, 4.1595, 3.2701], device='cuda:1'), covar=tensor([0.1915, 0.2643, 0.0597, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0591, 0.0556, 0.0785, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 21:36:48,956 INFO [train.py:968] (1/2) Epoch 7, batch 32000, giga_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09472, over 28838.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3465, pruned_loss=0.0981, over 5677820.31 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3724, pruned_loss=0.1312, over 5710000.74 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3451, pruned_loss=0.0951, over 5665708.31 frames. ], batch size: 145, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:36:51,093 INFO [optim.py:369] (1/2) 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 21:37:51,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8459, 4.6603, 4.4061, 1.9968], device='cuda:1'), covar=tensor([0.0401, 0.0517, 0.0654, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0876, 0.0781, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 21:37:54,021 INFO [train.py:968] (1/2) Epoch 7, batch 32050, giga_loss[loss=0.2233, simple_loss=0.3108, pruned_loss=0.06786, over 28841.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3459, pruned_loss=0.09775, over 5677222.38 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3724, pruned_loss=0.1312, over 5712653.69 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3444, pruned_loss=0.09495, over 5664807.47 frames. ], batch size: 174, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:38:04,704 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=305705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:09,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=305708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:15,482 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=305715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:20,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-03 21:38:40,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=305737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:38:57,159 INFO [train.py:968] (1/2) Epoch 7, batch 32100, giga_loss[loss=0.2667, simple_loss=0.3329, pruned_loss=0.1002, over 28534.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.345, pruned_loss=0.09814, over 5674291.29 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3718, pruned_loss=0.1309, over 5713979.94 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3436, pruned_loss=0.09527, over 5661974.54 frames. ], batch size: 71, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:39:00,656 INFO [optim.py:369] (1/2) 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 21:39:59,372 INFO [train.py:968] (1/2) Epoch 7, batch 32150, giga_loss[loss=0.3052, simple_loss=0.3746, pruned_loss=0.1179, over 28774.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3487, pruned_loss=0.0999, over 5680811.77 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3714, pruned_loss=0.1307, over 5717795.82 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3476, pruned_loss=0.0972, over 5666893.53 frames. ], batch size: 243, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:40:32,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4388, 3.8296, 1.4896, 1.6044], device='cuda:1'), covar=tensor([0.0873, 0.0356, 0.0859, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0489, 0.0323, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 21:40:57,757 INFO [train.py:968] (1/2) Epoch 7, batch 32200, giga_loss[loss=0.2427, simple_loss=0.3185, pruned_loss=0.08348, over 29007.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3486, pruned_loss=0.1007, over 5678312.03 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3707, pruned_loss=0.1303, over 5720933.21 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3476, pruned_loss=0.09798, over 5663303.82 frames. ], batch size: 112, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:41:01,692 INFO [optim.py:369] (1/2) 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 21:42:00,991 INFO [train.py:968] (1/2) Epoch 7, batch 32250, giga_loss[loss=0.2497, simple_loss=0.3256, pruned_loss=0.0869, over 28926.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3477, pruned_loss=0.1013, over 5679017.79 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3707, pruned_loss=0.1305, over 5725019.27 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3463, pruned_loss=0.09819, over 5662430.55 frames. ], batch size: 120, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:43:02,967 INFO [train.py:968] (1/2) Epoch 7, batch 32300, giga_loss[loss=0.2744, simple_loss=0.3554, pruned_loss=0.09671, over 28690.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3475, pruned_loss=0.1015, over 5668274.88 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3706, pruned_loss=0.1307, over 5715668.78 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.346, pruned_loss=0.09824, over 5662014.94 frames. ], batch size: 262, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:43:06,151 INFO [optim.py:369] (1/2) 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 21:43:14,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-03 21:44:11,951 INFO [train.py:968] (1/2) Epoch 7, batch 32350, giga_loss[loss=0.3043, simple_loss=0.3805, pruned_loss=0.114, over 28975.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3488, pruned_loss=0.1016, over 5664359.00 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3705, pruned_loss=0.1307, over 5712960.27 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3472, pruned_loss=0.09828, over 5660632.40 frames. ], batch size: 213, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:44:29,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:44:37,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:45:25,031 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:45:31,646 INFO [train.py:968] (1/2) Epoch 7, batch 32400, giga_loss[loss=0.2244, simple_loss=0.2948, pruned_loss=0.07704, over 24373.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.349, pruned_loss=0.1001, over 5664225.13 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3705, pruned_loss=0.1307, over 5710387.47 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3477, pruned_loss=0.09739, over 5663521.46 frames. ], batch size: 705, lr: 4.54e-03, grad_scale: 4.0
+2023-03-03 21:45:36,303 INFO [optim.py:369] (1/2) 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 21:45:50,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3669, 1.5774, 1.2334, 1.9616], device='cuda:1'), covar=tensor([0.2312, 0.2164, 0.2312, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.1189, 0.0888, 0.1055, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 21:46:10,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8980, 1.1492, 1.0161, 0.7022], device='cuda:1'), covar=tensor([0.1188, 0.1188, 0.0689, 0.1086], device='cuda:1'), in_proj_covar=tensor([0.1540, 0.1359, 0.1318, 0.1441], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 21:46:32,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:46:43,264 INFO [train.py:968] (1/2) Epoch 7, batch 32450, giga_loss[loss=0.2644, simple_loss=0.3328, pruned_loss=0.09803, over 29001.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3468, pruned_loss=0.09921, over 5667830.79 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3699, pruned_loss=0.1303, over 5712945.18 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3458, pruned_loss=0.09688, over 5664320.45 frames. ], batch size: 199, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:47:42,818 INFO [train.py:968] (1/2) Epoch 7, batch 32500, giga_loss[loss=0.2102, simple_loss=0.289, pruned_loss=0.06572, over 28709.00 frames. ], tot_loss[loss=0.27, simple_loss=0.343, pruned_loss=0.09847, over 5661641.67 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3695, pruned_loss=0.13, over 5702265.38 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3415, pruned_loss=0.09549, over 5666347.85 frames. ], batch size: 262, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:47:46,427 INFO [optim.py:369] (1/2) 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 21:47:47,548 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:47:50,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:48:28,351 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:48:46,903 INFO [train.py:968] (1/2) Epoch 7, batch 32550, giga_loss[loss=0.2628, simple_loss=0.3377, pruned_loss=0.09398, over 28927.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3376, pruned_loss=0.09622, over 5650873.45 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3688, pruned_loss=0.1297, over 5697233.37 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3363, pruned_loss=0.09325, over 5658372.96 frames. ], batch size: 155, lr: 4.54e-03, grad_scale: 2.0
+2023-03-03 21:48:55,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3791, 1.8930, 1.3850, 1.5760], device='cuda:1'), covar=tensor([0.0707, 0.0374, 0.0330, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0122, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0076], device='cuda:1')
+2023-03-03 21:49:30,577 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:49:32,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:49:43,112 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:49:48,915 INFO [train.py:968] (1/2) Epoch 7, batch 32600, giga_loss[loss=0.2694, simple_loss=0.3406, pruned_loss=0.09908, over 28012.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3385, pruned_loss=0.0969, over 5651279.98 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3686, pruned_loss=0.1295, over 5699627.35 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.337, pruned_loss=0.09407, over 5654201.22 frames. ], batch size: 412, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:49:52,686 INFO [optim.py:369] (1/2) 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 21:50:00,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4861, 1.5108, 4.5671, 3.2520], device='cuda:1'), covar=tensor([0.1537, 0.2313, 0.0330, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0593, 0.0558, 0.0790, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 21:50:03,704 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:50:10,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9800, 1.3589, 1.1195, 0.1767], device='cuda:1'), covar=tensor([0.1843, 0.1753, 0.2609, 0.3252], device='cuda:1'), in_proj_covar=tensor([0.1459, 0.1389, 0.1441, 0.1205], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 21:50:46,915 INFO [train.py:968] (1/2) Epoch 7, batch 32650, giga_loss[loss=0.2263, simple_loss=0.3178, pruned_loss=0.06736, over 28940.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3397, pruned_loss=0.09766, over 5654389.93 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3687, pruned_loss=0.1296, over 5705341.93 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3377, pruned_loss=0.09452, over 5650863.84 frames. ], batch size: 155, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:51:49,871 INFO [train.py:968] (1/2) Epoch 7, batch 32700, giga_loss[loss=0.2875, simple_loss=0.3622, pruned_loss=0.1064, over 28674.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3377, pruned_loss=0.09538, over 5655971.71 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3686, pruned_loss=0.1295, over 5707237.27 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3357, pruned_loss=0.09237, over 5650528.88 frames. ], batch size: 243, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:51:55,158 INFO [optim.py:369] (1/2) 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 21:52:08,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4869, 4.3485, 4.0494, 1.8779], device='cuda:1'), covar=tensor([0.0442, 0.0614, 0.0816, 0.2087], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0875, 0.0775, 0.0610], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 21:52:41,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306391.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:52:53,450 INFO [train.py:968] (1/2) Epoch 7, batch 32750, giga_loss[loss=0.2574, simple_loss=0.3317, pruned_loss=0.09155, over 28976.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3371, pruned_loss=0.09449, over 5655326.93 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3689, pruned_loss=0.1297, over 5697903.31 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3347, pruned_loss=0.09128, over 5658657.37 frames. ], batch size: 213, lr: 4.53e-03, grad_scale: 2.0
+2023-03-03 21:53:21,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:53:33,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2718, 1.4184, 1.4227, 1.3314], device='cuda:1'), covar=tensor([0.0932, 0.1194, 0.1344, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0706, 0.0636, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 21:54:00,203 INFO [train.py:968] (1/2) Epoch 7, batch 32800, giga_loss[loss=0.2716, simple_loss=0.3492, pruned_loss=0.09697, over 28673.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3372, pruned_loss=0.09546, over 5646517.30 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3687, pruned_loss=0.1296, over 5694005.16 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3345, pruned_loss=0.09189, over 5651489.55 frames. ], batch size: 242, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:54:05,651 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 32850, giga_loss[loss=0.2528, simple_loss=0.3305, pruned_loss=0.0875, over 29094.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3373, pruned_loss=0.09505, over 5655401.02 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3683, pruned_loss=0.1296, over 5699731.41 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3344, pruned_loss=0.09107, over 5652991.46 frames. ], batch size: 113, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:55:50,030 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306537.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 21:56:10,806 INFO [train.py:968] (1/2) Epoch 7, batch 32900, giga_loss[loss=0.3437, simple_loss=0.3968, pruned_loss=0.1454, over 28801.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.337, pruned_loss=0.09498, over 5659354.85 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3681, pruned_loss=0.1295, over 5702591.01 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3344, pruned_loss=0.09148, over 5654358.90 frames. ], batch size: 243, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:56:15,656 INFO [optim.py:369] (1/2) 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 21:56:26,200 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:56:31,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 21:56:31,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:56:55,234 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:57:04,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:57:07,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1964, 1.8590, 1.5008, 1.4395], device='cuda:1'), covar=tensor([0.0835, 0.0297, 0.0299, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0122, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0076], device='cuda:1')
+2023-03-03 21:57:09,923 INFO [train.py:968] (1/2) Epoch 7, batch 32950, giga_loss[loss=0.2642, simple_loss=0.3416, pruned_loss=0.09346, over 28750.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3387, pruned_loss=0.0966, over 5667781.95 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3679, pruned_loss=0.1292, over 5709440.90 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3358, pruned_loss=0.09291, over 5656163.49 frames. ], batch size: 243, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:57:36,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 21:58:03,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3511, 1.4713, 1.2341, 1.5619], device='cuda:1'), covar=tensor([0.0716, 0.0433, 0.0349, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0122, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0076], device='cuda:1')
+2023-03-03 21:58:10,154 INFO [train.py:968] (1/2) Epoch 7, batch 33000, giga_loss[loss=0.278, simple_loss=0.3489, pruned_loss=0.1036, over 28065.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3364, pruned_loss=0.0948, over 5654718.65 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3675, pruned_loss=0.1289, over 5702338.12 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3338, pruned_loss=0.09138, over 5651034.09 frames. ], batch size: 412, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:58:10,154 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 21:58:18,664 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-03 21:58:23,225 INFO [optim.py:369] (1/2) 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 21:59:06,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3070, 1.5438, 1.2493, 1.4441], device='cuda:1'), covar=tensor([0.2170, 0.2047, 0.2161, 0.1831], device='cuda:1'), in_proj_covar=tensor([0.1185, 0.0895, 0.1052, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 21:59:11,106 INFO [train.py:968] (1/2) Epoch 7, batch 33050, giga_loss[loss=0.2478, simple_loss=0.337, pruned_loss=0.07928, over 28913.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3386, pruned_loss=0.09498, over 5662156.91 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3668, pruned_loss=0.1287, over 5704965.67 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3361, pruned_loss=0.09117, over 5655063.01 frames. ], batch size: 112, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 21:59:15,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1019, 1.1715, 3.8320, 3.2228], device='cuda:1'), covar=tensor([0.1613, 0.2429, 0.0359, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0591, 0.0557, 0.0784, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 21:59:16,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-03 22:00:07,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7347, 1.8942, 1.6522, 1.6214], device='cuda:1'), covar=tensor([0.1055, 0.1596, 0.1635, 0.1545], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0703, 0.0637, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 22:00:10,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1321, 1.4353, 1.1853, 0.9712], device='cuda:1'), covar=tensor([0.1383, 0.1157, 0.0833, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.1555, 0.1370, 0.1330, 0.1458], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 22:00:11,983 INFO [train.py:968] (1/2) Epoch 7, batch 33100, giga_loss[loss=0.2812, simple_loss=0.3635, pruned_loss=0.09947, over 28710.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.09647, over 5663994.01 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3669, pruned_loss=0.1286, over 5710010.77 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3393, pruned_loss=0.09271, over 5653022.50 frames. ], batch size: 243, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:00:16,310 INFO [optim.py:369] (1/2) 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 22:00:20,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5916, 1.5887, 1.3053, 1.7506], device='cuda:1'), covar=tensor([0.0648, 0.0266, 0.0298, 0.0698], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0122, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0049, 0.0045, 0.0076], device='cuda:1')
+2023-03-03 22:00:25,813 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=306763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:00:28,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=306766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:00:29,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5369, 3.5506, 1.6730, 1.4464], device='cuda:1'), covar=tensor([0.0835, 0.0349, 0.0829, 0.1267], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0494, 0.0326, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-03 22:00:43,987 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=306777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:01:08,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=306795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:01:14,879 INFO [train.py:968] (1/2) Epoch 7, batch 33150, giga_loss[loss=0.2885, simple_loss=0.3598, pruned_loss=0.1086, over 29010.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3428, pruned_loss=0.09735, over 5654502.20 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3668, pruned_loss=0.1288, over 5711972.20 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.34, pruned_loss=0.09332, over 5642665.09 frames. ], batch size: 199, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:02:16,884 INFO [train.py:968] (1/2) Epoch 7, batch 33200, giga_loss[loss=0.2434, simple_loss=0.3269, pruned_loss=0.07997, over 28984.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3425, pruned_loss=0.09728, over 5657822.56 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3663, pruned_loss=0.1285, over 5707223.65 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3402, pruned_loss=0.0936, over 5651681.29 frames. ], batch size: 213, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:02:21,429 INFO [optim.py:369] (1/2) 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 22:03:10,298 INFO [train.py:968] (1/2) Epoch 7, batch 33250, giga_loss[loss=0.251, simple_loss=0.3291, pruned_loss=0.08645, over 28660.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3403, pruned_loss=0.09582, over 5666034.64 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3657, pruned_loss=0.1281, over 5711885.89 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3382, pruned_loss=0.09215, over 5655252.68 frames. ], batch size: 262, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:04:12,380 INFO [train.py:968] (1/2) Epoch 7, batch 33300, giga_loss[loss=0.28, simple_loss=0.3533, pruned_loss=0.1033, over 28957.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3384, pruned_loss=0.09465, over 5668466.34 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3648, pruned_loss=0.1275, over 5715773.67 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.337, pruned_loss=0.09165, over 5655700.52 frames. ], batch size: 186, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:04:16,907 INFO [optim.py:369] (1/2) 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 22:04:30,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=306962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:05:14,493 INFO [train.py:968] (1/2) Epoch 7, batch 33350, giga_loss[loss=0.2243, simple_loss=0.313, pruned_loss=0.06785, over 28836.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3363, pruned_loss=0.09401, over 5671199.00 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.365, pruned_loss=0.1278, over 5716877.35 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3346, pruned_loss=0.09092, over 5659401.07 frames. ], batch size: 174, lr: 4.53e-03, grad_scale: 8.0
+2023-03-03 22:05:38,322 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 22:06:12,330 INFO [train.py:968] (1/2) Epoch 7, batch 33400, giga_loss[loss=0.2797, simple_loss=0.36, pruned_loss=0.09973, over 28905.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3402, pruned_loss=0.09602, over 5678113.56 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3652, pruned_loss=0.128, over 5721051.50 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3381, pruned_loss=0.0926, over 5664144.82 frames. ], batch size: 174, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:06:18,789 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:07:16,344 INFO [train.py:968] (1/2) Epoch 7, batch 33450, giga_loss[loss=0.2794, simple_loss=0.3502, pruned_loss=0.1043, over 28991.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3423, pruned_loss=0.09717, over 5677346.20 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.365, pruned_loss=0.1278, over 5723742.90 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3402, pruned_loss=0.09395, over 5662935.54 frames. ], batch size: 213, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:07:22,328 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:07:26,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:07:45,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2247, 1.6625, 1.5655, 1.1870], device='cuda:1'), covar=tensor([0.1402, 0.2099, 0.1154, 0.1414], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0690, 0.0800, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 22:08:06,076 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 33500, giga_loss[loss=0.3074, simple_loss=0.3786, pruned_loss=0.1181, over 28497.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3439, pruned_loss=0.09829, over 5673390.36 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3652, pruned_loss=0.128, over 5722554.60 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3418, pruned_loss=0.09517, over 5662446.28 frames. ], batch size: 336, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:08:27,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=307152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:08:29,488 INFO [optim.py:369] (1/2) 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 22:09:22,905 INFO [train.py:968] (1/2) Epoch 7, batch 33550, giga_loss[loss=0.2607, simple_loss=0.3211, pruned_loss=0.1002, over 24412.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3468, pruned_loss=0.1, over 5661293.50 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3644, pruned_loss=0.1276, over 5708395.58 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3451, pruned_loss=0.09674, over 5662016.05 frames. ], batch size: 705, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:10:13,582 INFO [train.py:968] (1/2) Epoch 7, batch 33600, giga_loss[loss=0.2749, simple_loss=0.3387, pruned_loss=0.1055, over 24605.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.349, pruned_loss=0.1013, over 5659855.18 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3638, pruned_loss=0.1273, over 5711163.85 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3473, pruned_loss=0.09742, over 5655612.87 frames. ], batch size: 705, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:10:20,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-03 22:10:20,980 INFO [optim.py:369] (1/2) 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 22:11:18,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:11:22,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:11:23,581 INFO [train.py:968] (1/2) Epoch 7, batch 33650, giga_loss[loss=0.2435, simple_loss=0.3229, pruned_loss=0.08204, over 29060.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3488, pruned_loss=0.1007, over 5662353.09 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3637, pruned_loss=0.1272, over 5713791.90 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09741, over 5655889.37 frames. ], batch size: 186, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:12:01,035 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 33700, giga_loss[loss=0.2561, simple_loss=0.3315, pruned_loss=0.09031, over 28834.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3467, pruned_loss=0.1001, over 5658156.56 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.364, pruned_loss=0.1277, over 5707877.42 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3448, pruned_loss=0.09633, over 5657682.27 frames. ], batch size: 164, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:12:40,906 INFO [optim.py:369] (1/2) 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 22:13:34,184 INFO [train.py:968] (1/2) Epoch 7, batch 33750, giga_loss[loss=0.2785, simple_loss=0.3559, pruned_loss=0.1006, over 28886.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3466, pruned_loss=0.1003, over 5663607.02 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3639, pruned_loss=0.1276, over 5712392.17 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3448, pruned_loss=0.09675, over 5658451.15 frames. ], batch size: 136, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:14:41,559 INFO [train.py:968] (1/2) Epoch 7, batch 33800, giga_loss[loss=0.2908, simple_loss=0.3566, pruned_loss=0.1125, over 28495.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3457, pruned_loss=0.09998, over 5655419.78 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3638, pruned_loss=0.1276, over 5709958.33 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.344, pruned_loss=0.09666, over 5652337.89 frames. ], batch size: 336, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:14:51,906 INFO [optim.py:369] (1/2) 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 22:15:00,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=307460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:15:53,461 INFO [train.py:968] (1/2) Epoch 7, batch 33850, libri_loss[loss=0.314, simple_loss=0.381, pruned_loss=0.1235, over 29757.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3436, pruned_loss=0.0995, over 5650078.37 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3638, pruned_loss=0.1275, over 5712256.27 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3421, pruned_loss=0.09669, over 5644949.89 frames. ], batch size: 87, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:16:53,210 INFO [train.py:968] (1/2) Epoch 7, batch 33900, giga_loss[loss=0.3829, simple_loss=0.4147, pruned_loss=0.1755, over 26788.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3433, pruned_loss=0.0994, over 5637958.02 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3642, pruned_loss=0.1277, over 5702677.62 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3415, pruned_loss=0.09662, over 5640770.55 frames. ], batch size: 555, lr: 4.53e-03, grad_scale: 4.0
+2023-03-03 22:16:59,274 INFO [optim.py:369] (1/2) 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 22:17:44,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 22:17:49,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3675, 1.3773, 4.5170, 3.5839], device='cuda:1'), covar=tensor([0.1538, 0.2412, 0.0312, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0592, 0.0556, 0.0781, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 22:17:53,117 INFO [train.py:968] (1/2) Epoch 7, batch 33950, giga_loss[loss=0.2314, simple_loss=0.3201, pruned_loss=0.07139, over 28932.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3425, pruned_loss=0.09776, over 5649932.43 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3638, pruned_loss=0.1275, over 5699867.77 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3408, pruned_loss=0.09496, over 5653280.44 frames. ], batch size: 284, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:17:57,381 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=307603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:18:02,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=307606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:18:33,556 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=307635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:18:40,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4716, 1.9245, 1.8480, 1.4504], device='cuda:1'), covar=tensor([0.1669, 0.1948, 0.1310, 0.1416], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0696, 0.0809, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 22:18:49,195 INFO [train.py:968] (1/2) Epoch 7, batch 34000, giga_loss[loss=0.2557, simple_loss=0.3412, pruned_loss=0.08509, over 28955.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09571, over 5667839.24 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3633, pruned_loss=0.1271, over 5704862.62 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3412, pruned_loss=0.09308, over 5665081.27 frames. ], batch size: 199, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:18:57,057 INFO [optim.py:369] (1/2) 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 22:19:47,198 INFO [train.py:968] (1/2) Epoch 7, batch 34050, giga_loss[loss=0.265, simple_loss=0.3503, pruned_loss=0.08981, over 28644.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3456, pruned_loss=0.09649, over 5658610.56 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3636, pruned_loss=0.1274, over 5700430.69 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3436, pruned_loss=0.09319, over 5659499.22 frames. ], batch size: 307, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:20:14,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8919, 2.9760, 2.1301, 1.0843], device='cuda:1'), covar=tensor([0.3980, 0.1807, 0.2195, 0.3537], device='cuda:1'), in_proj_covar=tensor([0.1460, 0.1392, 0.1430, 0.1201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 22:20:26,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:20:44,526 INFO [train.py:968] (1/2) Epoch 7, batch 34100, giga_loss[loss=0.2991, simple_loss=0.3557, pruned_loss=0.1213, over 26815.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3446, pruned_loss=0.09563, over 5665380.11 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3627, pruned_loss=0.1269, over 5705296.31 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3434, pruned_loss=0.0928, over 5660903.09 frames. ], batch size: 555, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:20:51,395 INFO [optim.py:369] (1/2) 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 22:21:56,434 INFO [train.py:968] (1/2) Epoch 7, batch 34150, giga_loss[loss=0.2944, simple_loss=0.3756, pruned_loss=0.1066, over 28638.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3444, pruned_loss=0.0951, over 5670423.48 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3628, pruned_loss=0.1269, over 5707287.31 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3433, pruned_loss=0.09261, over 5664999.91 frames. ], batch size: 307, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:22:25,014 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-03 22:23:05,928 INFO [train.py:968] (1/2) Epoch 7, batch 34200, giga_loss[loss=0.2347, simple_loss=0.3011, pruned_loss=0.08413, over 25081.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3432, pruned_loss=0.09405, over 5661037.74 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3627, pruned_loss=0.1268, over 5701545.48 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.342, pruned_loss=0.09157, over 5660244.19 frames. ], batch size: 705, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:23:17,192 INFO [optim.py:369] (1/2) 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 22:24:19,266 INFO [train.py:968] (1/2) Epoch 7, batch 34250, giga_loss[loss=0.2525, simple_loss=0.3342, pruned_loss=0.08541, over 27774.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3427, pruned_loss=0.09325, over 5659264.40 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3621, pruned_loss=0.1264, over 5703605.58 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3421, pruned_loss=0.09121, over 5656424.63 frames. ], batch size: 474, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:25:27,330 INFO [train.py:968] (1/2) Epoch 7, batch 34300, giga_loss[loss=0.2792, simple_loss=0.3651, pruned_loss=0.09665, over 28952.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3446, pruned_loss=0.09468, over 5659871.96 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3618, pruned_loss=0.1262, over 5704847.86 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3438, pruned_loss=0.09217, over 5655730.09 frames. ], batch size: 186, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:25:35,183 INFO [optim.py:369] (1/2) 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 22:26:21,661 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:26:27,290 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=307997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:26:28,922 INFO [train.py:968] (1/2) Epoch 7, batch 34350, libri_loss[loss=0.3756, simple_loss=0.4105, pruned_loss=0.1703, over 19786.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3499, pruned_loss=0.09793, over 5651670.72 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3627, pruned_loss=0.127, over 5691626.83 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3479, pruned_loss=0.09424, over 5658632.87 frames. ], batch size: 186, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:26:46,364 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:26:56,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1453, 1.5299, 1.2251, 0.9843], device='cuda:1'), covar=tensor([0.1780, 0.1099, 0.0767, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1351, 0.1312, 0.1438], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 22:27:00,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8620, 1.7774, 1.2935, 1.3593], device='cuda:1'), covar=tensor([0.0694, 0.0650, 0.1004, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0434, 0.0491, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 22:27:32,683 INFO [train.py:968] (1/2) Epoch 7, batch 34400, libri_loss[loss=0.3312, simple_loss=0.3856, pruned_loss=0.1384, over 29400.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3505, pruned_loss=0.09884, over 5664466.73 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.363, pruned_loss=0.1272, over 5687916.73 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3482, pruned_loss=0.09462, over 5671525.20 frames. ], batch size: 92, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:27:44,597 INFO [optim.py:369] (1/2) 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 22:28:41,741 INFO [train.py:968] (1/2) Epoch 7, batch 34450, giga_loss[loss=0.2326, simple_loss=0.3213, pruned_loss=0.07192, over 29046.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3489, pruned_loss=0.09837, over 5677337.59 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3632, pruned_loss=0.1272, over 5694143.01 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3464, pruned_loss=0.09421, over 5676712.21 frames. ], batch size: 155, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:28:54,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:29:03,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8037, 0.9653, 3.4261, 2.8900], device='cuda:1'), covar=tensor([0.1689, 0.2534, 0.0415, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0595, 0.0557, 0.0782, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 22:29:54,519 INFO [train.py:968] (1/2) Epoch 7, batch 34500, giga_loss[loss=0.2502, simple_loss=0.3392, pruned_loss=0.08064, over 28958.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3456, pruned_loss=0.09584, over 5679612.73 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3626, pruned_loss=0.1267, over 5694405.95 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3438, pruned_loss=0.09237, over 5678646.90 frames. ], batch size: 106, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:30:06,945 INFO [optim.py:369] (1/2) 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 22:31:04,986 INFO [train.py:968] (1/2) Epoch 7, batch 34550, giga_loss[loss=0.256, simple_loss=0.3403, pruned_loss=0.08591, over 28823.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3429, pruned_loss=0.09309, over 5692387.39 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3628, pruned_loss=0.1267, over 5695462.92 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3413, pruned_loss=0.09019, over 5690565.70 frames. ], batch size: 243, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:32:04,241 INFO [train.py:968] (1/2) Epoch 7, batch 34600, giga_loss[loss=0.2381, simple_loss=0.3272, pruned_loss=0.07455, over 28930.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3437, pruned_loss=0.09432, over 5691185.76 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3628, pruned_loss=0.1267, over 5702674.28 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3416, pruned_loss=0.09078, over 5683192.06 frames. ], batch size: 174, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:32:10,341 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:32:12,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:32:13,806 INFO [optim.py:369] (1/2) 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 22:32:47,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 34650, giga_loss[loss=0.2879, simple_loss=0.3686, pruned_loss=0.1036, over 28586.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3461, pruned_loss=0.09568, over 5685118.90 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.363, pruned_loss=0.1269, over 5704240.72 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.344, pruned_loss=0.09215, over 5677030.19 frames. ], batch size: 307, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:34:06,599 INFO [train.py:968] (1/2) Epoch 7, batch 34700, giga_loss[loss=0.2292, simple_loss=0.312, pruned_loss=0.0732, over 29025.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3457, pruned_loss=0.0959, over 5678536.79 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3629, pruned_loss=0.1268, over 5706278.18 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.344, pruned_loss=0.09294, over 5670096.06 frames. ], batch size: 136, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:34:17,218 INFO [optim.py:369] (1/2) 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 22:34:29,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:34:36,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:34:53,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:34:58,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-03 22:35:03,762 INFO [train.py:968] (1/2) Epoch 7, batch 34750, giga_loss[loss=0.2551, simple_loss=0.3405, pruned_loss=0.08483, over 28925.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3426, pruned_loss=0.09517, over 5677262.81 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3628, pruned_loss=0.1268, over 5709093.93 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.341, pruned_loss=0.09239, over 5667593.39 frames. ], batch size: 145, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:35:25,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-03 22:36:02,884 INFO [train.py:968] (1/2) Epoch 7, batch 34800, libri_loss[loss=0.328, simple_loss=0.3882, pruned_loss=0.1339, over 26071.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3428, pruned_loss=0.09586, over 5675653.67 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3626, pruned_loss=0.1264, over 5710509.87 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3413, pruned_loss=0.09323, over 5666321.22 frames. ], batch size: 136, lr: 4.52e-03, grad_scale: 8.0
+2023-03-03 22:36:11,444 INFO [optim.py:369] (1/2) 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 22:36:54,829 INFO [train.py:968] (1/2) Epoch 7, batch 34850, libri_loss[loss=0.2561, simple_loss=0.3223, pruned_loss=0.09497, over 29587.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3504, pruned_loss=0.1017, over 5655072.89 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3627, pruned_loss=0.1267, over 5696640.16 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3487, pruned_loss=0.09854, over 5657355.69 frames. ], batch size: 74, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:37:03,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:04,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:07,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:09,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:21,095 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=308531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:23,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=308534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:31,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:36,704 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:37:38,386 INFO [train.py:968] (1/2) Epoch 7, batch 34900, giga_loss[loss=0.332, simple_loss=0.3996, pruned_loss=0.1322, over 28776.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3607, pruned_loss=0.1078, over 5672941.22 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3631, pruned_loss=0.1272, over 5701166.93 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3588, pruned_loss=0.1044, over 5670127.25 frames. ], batch size: 284, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:37:48,261 INFO [optim.py:369] (1/2) 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 22:37:51,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=308563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:38:17,230 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:38:21,251 INFO [train.py:968] (1/2) Epoch 7, batch 34950, giga_loss[loss=0.3267, simple_loss=0.3891, pruned_loss=0.1322, over 28638.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3654, pruned_loss=0.1114, over 5674013.18 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3631, pruned_loss=0.1273, over 5701263.14 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3639, pruned_loss=0.1081, over 5670509.13 frames. ], batch size: 242, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:38:29,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1929, 1.4178, 1.1820, 1.0283], device='cuda:1'), covar=tensor([0.2066, 0.1981, 0.2138, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.1176, 0.0889, 0.1045, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 22:39:06,552 INFO [train.py:968] (1/2) Epoch 7, batch 35000, giga_loss[loss=0.2495, simple_loss=0.3259, pruned_loss=0.0865, over 28923.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3628, pruned_loss=0.1112, over 5671427.31 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3634, pruned_loss=0.1275, over 5694044.18 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3613, pruned_loss=0.1081, over 5675188.37 frames. ], batch size: 186, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:39:14,687 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 35050, giga_loss[loss=0.2569, simple_loss=0.3062, pruned_loss=0.1038, over 23856.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3558, pruned_loss=0.1083, over 5675373.12 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3642, pruned_loss=0.1281, over 5697543.59 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3537, pruned_loss=0.1047, over 5674255.35 frames. ], batch size: 705, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:39:49,320 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:40:09,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7811, 2.2924, 1.5362, 1.4452], device='cuda:1'), covar=tensor([0.1724, 0.1133, 0.1358, 0.1527], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1375, 0.1326, 0.1456], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 22:40:17,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7324, 1.5198, 5.1385, 3.4268], device='cuda:1'), covar=tensor([0.1429, 0.2255, 0.0278, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0595, 0.0556, 0.0788, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 22:40:30,695 INFO [train.py:968] (1/2) Epoch 7, batch 35100, giga_loss[loss=0.221, simple_loss=0.2923, pruned_loss=0.07485, over 28442.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3478, pruned_loss=0.1043, over 5682604.67 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.364, pruned_loss=0.1278, over 5698730.16 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3461, pruned_loss=0.1014, over 5680138.44 frames. ], batch size: 78, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:40:39,213 INFO [optim.py:369] (1/2) 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 22:40:40,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4811, 1.5544, 1.5653, 1.5182], device='cuda:1'), covar=tensor([0.1077, 0.1496, 0.1575, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0713, 0.0640, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 22:41:12,874 INFO [train.py:968] (1/2) Epoch 7, batch 35150, giga_loss[loss=0.26, simple_loss=0.32, pruned_loss=0.09999, over 28631.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3409, pruned_loss=0.1018, over 5683466.34 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3643, pruned_loss=0.1281, over 5704099.72 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3387, pruned_loss=0.09851, over 5675922.32 frames. ], batch size: 92, lr: 4.52e-03, grad_scale: 2.0
+2023-03-03 22:41:25,881 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2897, 2.4968, 1.3045, 1.3154], device='cuda:1'), covar=tensor([0.0892, 0.0343, 0.0865, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0483, 0.0320, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 22:41:46,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-03 22:41:55,264 INFO [train.py:968] (1/2) Epoch 7, batch 35200, giga_loss[loss=0.2112, simple_loss=0.283, pruned_loss=0.06972, over 28878.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3329, pruned_loss=0.09759, over 5689535.55 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3646, pruned_loss=0.1282, over 5706739.10 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3305, pruned_loss=0.09449, over 5680918.99 frames. ], batch size: 145, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:42:03,747 INFO [optim.py:369] (1/2) 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 22:42:09,477 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 22:42:30,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4206, 1.5670, 1.4071, 1.2544], device='cuda:1'), covar=tensor([0.1513, 0.1282, 0.1019, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1361, 0.1307, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-03 22:42:31,801 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-03 22:42:37,285 INFO [train.py:968] (1/2) Epoch 7, batch 35250, giga_loss[loss=0.2448, simple_loss=0.3165, pruned_loss=0.08653, over 28787.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3286, pruned_loss=0.09555, over 5696766.94 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3646, pruned_loss=0.1281, over 5709237.27 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3259, pruned_loss=0.0924, over 5687379.15 frames. ], batch size: 119, lr: 4.52e-03, grad_scale: 4.0
+2023-03-03 22:42:37,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8664, 1.9156, 1.8202, 1.7109], device='cuda:1'), covar=tensor([0.1018, 0.1448, 0.1438, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0712, 0.0637, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 22:43:18,780 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=308945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:43:19,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7805, 1.8051, 1.3633, 1.4297], device='cuda:1'), covar=tensor([0.0600, 0.0508, 0.0891, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0437, 0.0496, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 22:43:21,967 INFO [train.py:968] (1/2) Epoch 7, batch 35300, libri_loss[loss=0.3194, simple_loss=0.3855, pruned_loss=0.1266, over 27709.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3259, pruned_loss=0.09446, over 5697784.44 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3645, pruned_loss=0.128, over 5713196.56 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3227, pruned_loss=0.09113, over 5686565.77 frames. ], batch size: 115, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:43:30,181 INFO [optim.py:369] (1/2) 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 22:43:37,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=308970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:44:01,323 INFO [train.py:968] (1/2) Epoch 7, batch 35350, giga_loss[loss=0.2857, simple_loss=0.3319, pruned_loss=0.1197, over 26745.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3229, pruned_loss=0.09291, over 5689363.08 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3649, pruned_loss=0.128, over 5712861.89 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3189, pruned_loss=0.08932, over 5680206.74 frames. ], batch size: 555, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:44:47,709 INFO [train.py:968] (1/2) Epoch 7, batch 35400, giga_loss[loss=0.2131, simple_loss=0.2838, pruned_loss=0.07115, over 28939.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3198, pruned_loss=0.0916, over 5682087.63 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3653, pruned_loss=0.1281, over 5716398.35 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3154, pruned_loss=0.08799, over 5671324.68 frames. ], batch size: 227, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:44:53,970 INFO [optim.py:369] (1/2) 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 22:44:57,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:11,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:28,398 INFO [train.py:968] (1/2) Epoch 7, batch 35450, giga_loss[loss=0.2605, simple_loss=0.3161, pruned_loss=0.1025, over 27567.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3174, pruned_loss=0.09007, over 5690913.95 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3658, pruned_loss=0.1281, over 5719534.98 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3124, pruned_loss=0.08637, over 5678975.37 frames. ], batch size: 472, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:45:41,212 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:43,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:45:52,341 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:46:08,498 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:46:11,994 INFO [train.py:968] (1/2) Epoch 7, batch 35500, giga_loss[loss=0.2319, simple_loss=0.3043, pruned_loss=0.07976, over 28488.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3157, pruned_loss=0.0894, over 5693671.79 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3665, pruned_loss=0.1285, over 5719507.53 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3101, pruned_loss=0.08516, over 5683267.26 frames. ], batch size: 336, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:46:19,231 INFO [optim.py:369] (1/2) 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 22:46:51,796 INFO [train.py:968] (1/2) Epoch 7, batch 35550, giga_loss[loss=0.2451, simple_loss=0.3171, pruned_loss=0.08653, over 28601.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3131, pruned_loss=0.08802, over 5698035.85 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3664, pruned_loss=0.1282, over 5724995.62 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3069, pruned_loss=0.08363, over 5684069.53 frames. ], batch size: 85, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:47:08,643 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:47:11,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:47:34,081 INFO [train.py:968] (1/2) Epoch 7, batch 35600, giga_loss[loss=0.218, simple_loss=0.2872, pruned_loss=0.07436, over 28652.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3106, pruned_loss=0.08705, over 5695246.23 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3662, pruned_loss=0.128, over 5730606.84 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3042, pruned_loss=0.08257, over 5678267.46 frames. ], batch size: 336, lr: 4.51e-03, grad_scale: 8.0
+2023-03-03 22:47:36,243 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 7, batch 35650, giga_loss[loss=0.2513, simple_loss=0.3176, pruned_loss=0.0925, over 28787.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3101, pruned_loss=0.08733, over 5682936.58 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3672, pruned_loss=0.1287, over 5723117.31 frames. ], giga_tot_loss[loss=0.2337, simple_loss=0.3029, pruned_loss=0.08222, over 5675510.92 frames. ], batch size: 119, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:48:36,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:49:03,202 INFO [train.py:968] (1/2) Epoch 7, batch 35700, giga_loss[loss=0.2698, simple_loss=0.3417, pruned_loss=0.09895, over 28363.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3188, pruned_loss=0.09218, over 5685705.64 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3672, pruned_loss=0.1285, over 5725938.25 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.311, pruned_loss=0.08681, over 5675976.38 frames. ], batch size: 60, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:49:09,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-03 22:49:11,833 INFO [optim.py:369] (1/2) 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 22:49:44,609 INFO [train.py:968] (1/2) Epoch 7, batch 35750, giga_loss[loss=0.3161, simple_loss=0.3831, pruned_loss=0.1246, over 29030.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3331, pruned_loss=0.1007, over 5691354.33 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3678, pruned_loss=0.1287, over 5732589.65 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3236, pruned_loss=0.09405, over 5675072.67 frames. ], batch size: 106, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:50:14,106 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:50:26,147 INFO [train.py:968] (1/2) Epoch 7, batch 35800, giga_loss[loss=0.3298, simple_loss=0.4004, pruned_loss=0.1296, over 28876.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3459, pruned_loss=0.1078, over 5671873.23 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3685, pruned_loss=0.1291, over 5713760.15 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.337, pruned_loss=0.1015, over 5674128.85 frames. ], batch size: 186, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:50:36,035 INFO [optim.py:369] (1/2) 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 22:50:37,785 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:50:39,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:51:05,731 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:51:08,482 INFO [train.py:968] (1/2) Epoch 7, batch 35850, giga_loss[loss=0.3015, simple_loss=0.3714, pruned_loss=0.1158, over 28805.00 frames. ], tot_loss[loss=0.287, simple_loss=0.353, pruned_loss=0.1105, over 5674721.99 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3685, pruned_loss=0.1289, over 5709814.13 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3454, pruned_loss=0.1051, over 5678443.72 frames. ], batch size: 119, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:51:09,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=309500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:51:50,330 INFO [train.py:968] (1/2) Epoch 7, batch 35900, libri_loss[loss=0.2633, simple_loss=0.3234, pruned_loss=0.1016, over 29615.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3558, pruned_loss=0.1102, over 5685225.13 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3685, pruned_loss=0.1289, over 5714835.54 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3494, pruned_loss=0.1056, over 5682836.52 frames. ], batch size: 69, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:51:51,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0026, 3.7959, 3.5816, 1.8324], device='cuda:1'), covar=tensor([0.0523, 0.0632, 0.0667, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0880, 0.0775, 0.0618], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 22:51:59,592 INFO [optim.py:369] (1/2) 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 22:52:20,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:52:23,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=309583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:52:39,012 INFO [train.py:968] (1/2) Epoch 7, batch 35950, giga_loss[loss=0.2902, simple_loss=0.3629, pruned_loss=0.1087, over 28577.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3573, pruned_loss=0.1098, over 5668057.38 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3692, pruned_loss=0.1294, over 5712735.10 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3515, pruned_loss=0.1055, over 5667834.97 frames. ], batch size: 85, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:52:49,599 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:53:16,971 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=309643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:53:18,696 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 36000, giga_loss[loss=0.3349, simple_loss=0.3901, pruned_loss=0.1398, over 27968.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3597, pruned_loss=0.1115, over 5654914.18 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3697, pruned_loss=0.1298, over 5692738.31 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3544, pruned_loss=0.1074, over 5670339.24 frames. ], batch size: 412, lr: 4.51e-03, grad_scale: 8.0
+2023-03-03 22:53:21,336 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 22:53:25,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1955, 1.5323, 1.4884, 1.3839], device='cuda:1'), covar=tensor([0.1321, 0.1374, 0.1789, 0.1471], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0723, 0.0648, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 22:53:30,038 INFO [train.py:1012] (1/2) Epoch 7, validation: loss=0.2225, simple_loss=0.3279, pruned_loss=0.05855, over 944034.00 frames. 
+2023-03-03 22:53:30,038 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-03 22:53:39,190 INFO [optim.py:369] (1/2) 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 22:53:53,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=309675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:54:03,542 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-03 22:54:13,664 INFO [train.py:968] (1/2) Epoch 7, batch 36050, giga_loss[loss=0.2759, simple_loss=0.3558, pruned_loss=0.09797, over 28776.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3606, pruned_loss=0.112, over 5659412.15 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3698, pruned_loss=0.1299, over 5684126.14 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3563, pruned_loss=0.1085, over 5678303.25 frames. ], batch size: 119, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:54:55,511 INFO [train.py:968] (1/2) Epoch 7, batch 36100, giga_loss[loss=0.3099, simple_loss=0.3806, pruned_loss=0.1196, over 28701.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3647, pruned_loss=0.1154, over 5664347.91 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3698, pruned_loss=0.1298, over 5685582.00 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3611, pruned_loss=0.1123, over 5677478.19 frames. ], batch size: 242, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:55:05,115 INFO [optim.py:369] (1/2) 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 22:55:37,327 INFO [train.py:968] (1/2) Epoch 7, batch 36150, giga_loss[loss=0.3286, simple_loss=0.3967, pruned_loss=0.1302, over 28879.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3673, pruned_loss=0.1157, over 5687836.23 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3701, pruned_loss=0.1298, over 5692387.31 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.364, pruned_loss=0.1127, over 5691743.08 frames. ], batch size: 186, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:55:39,834 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:56:15,848 INFO [train.py:968] (1/2) Epoch 7, batch 36200, giga_loss[loss=0.2736, simple_loss=0.3575, pruned_loss=0.09482, over 28981.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3689, pruned_loss=0.1154, over 5691770.01 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3702, pruned_loss=0.1299, over 5696401.57 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3662, pruned_loss=0.1128, over 5691380.35 frames. ], batch size: 106, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:56:28,272 INFO [optim.py:369] (1/2) 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 22:57:00,016 INFO [train.py:968] (1/2) Epoch 7, batch 36250, giga_loss[loss=0.3239, simple_loss=0.3876, pruned_loss=0.1301, over 28841.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3702, pruned_loss=0.1153, over 5695921.86 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3706, pruned_loss=0.1301, over 5696822.61 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3677, pruned_loss=0.1129, over 5695133.19 frames. ], batch size: 199, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:57:01,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4576, 1.5243, 1.4099, 1.3710], device='cuda:1'), covar=tensor([0.1180, 0.1842, 0.1811, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0727, 0.0648, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 22:57:39,573 INFO [train.py:968] (1/2) Epoch 7, batch 36300, giga_loss[loss=0.3253, simple_loss=0.3879, pruned_loss=0.1313, over 28732.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3706, pruned_loss=0.1145, over 5688171.43 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3712, pruned_loss=0.1304, over 5691291.60 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3681, pruned_loss=0.112, over 5692191.22 frames. ], batch size: 242, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:57:47,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0881, 1.4085, 1.1272, 1.1069], device='cuda:1'), covar=tensor([0.2187, 0.2125, 0.2240, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.1182, 0.0892, 0.1042, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 22:57:48,335 INFO [optim.py:369] (1/2) 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 22:57:53,341 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=309966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 22:58:20,498 INFO [train.py:968] (1/2) Epoch 7, batch 36350, giga_loss[loss=0.2781, simple_loss=0.3576, pruned_loss=0.09934, over 28647.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3686, pruned_loss=0.112, over 5687244.09 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3717, pruned_loss=0.1306, over 5684544.90 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3662, pruned_loss=0.1096, over 5696260.39 frames. ], batch size: 307, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:58:41,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7410, 4.5318, 4.4058, 1.9772], device='cuda:1'), covar=tensor([0.0460, 0.0639, 0.0700, 0.1647], device='cuda:1'), in_proj_covar=tensor([0.0903, 0.0860, 0.0760, 0.0611], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 22:59:01,120 INFO [train.py:968] (1/2) Epoch 7, batch 36400, giga_loss[loss=0.312, simple_loss=0.3767, pruned_loss=0.1236, over 28257.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3664, pruned_loss=0.1108, over 5687729.92 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3715, pruned_loss=0.1305, over 5689684.54 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3645, pruned_loss=0.1086, over 5690390.29 frames. ], batch size: 368, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 22:59:11,356 INFO [optim.py:369] (1/2) 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 22:59:40,431 INFO [train.py:968] (1/2) Epoch 7, batch 36450, giga_loss[loss=0.3444, simple_loss=0.3646, pruned_loss=0.1621, over 23346.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3678, pruned_loss=0.113, over 5678603.10 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3718, pruned_loss=0.1305, over 5685619.21 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3658, pruned_loss=0.1104, over 5684766.81 frames. ], batch size: 705, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:00:26,253 INFO [train.py:968] (1/2) Epoch 7, batch 36500, giga_loss[loss=0.3247, simple_loss=0.3839, pruned_loss=0.1327, over 28728.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3717, pruned_loss=0.1181, over 5677611.01 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3718, pruned_loss=0.1304, over 5680136.81 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3701, pruned_loss=0.1159, over 5687937.76 frames. ], batch size: 284, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:00:37,503 INFO [optim.py:369] (1/2) 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 23:00:52,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:01:09,376 INFO [train.py:968] (1/2) Epoch 7, batch 36550, giga_loss[loss=0.3512, simple_loss=0.3944, pruned_loss=0.1541, over 26564.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3726, pruned_loss=0.1206, over 5680151.73 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3721, pruned_loss=0.1306, over 5681971.27 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.371, pruned_loss=0.1186, over 5686611.79 frames. ], batch size: 555, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:01:45,222 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:01:53,767 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:01:55,022 INFO [train.py:968] (1/2) Epoch 7, batch 36600, libri_loss[loss=0.27, simple_loss=0.3298, pruned_loss=0.1051, over 29354.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3712, pruned_loss=0.1206, over 5673237.60 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3728, pruned_loss=0.1309, over 5674088.73 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3693, pruned_loss=0.1183, over 5685153.83 frames. ], batch size: 71, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:02:06,840 INFO [optim.py:369] (1/2) 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 23:02:37,375 INFO [train.py:968] (1/2) Epoch 7, batch 36650, giga_loss[loss=0.2791, simple_loss=0.3432, pruned_loss=0.1076, over 28707.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3708, pruned_loss=0.1211, over 5681004.71 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3733, pruned_loss=0.1312, over 5671666.37 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3688, pruned_loss=0.1188, over 5692928.30 frames. ], batch size: 92, lr: 4.51e-03, grad_scale: 4.0
+2023-03-03 23:03:00,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:03:02,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:03:15,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:03:24,192 INFO [train.py:968] (1/2) Epoch 7, batch 36700, giga_loss[loss=0.2657, simple_loss=0.3432, pruned_loss=0.0941, over 28579.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3688, pruned_loss=0.1194, over 5679076.01 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3736, pruned_loss=0.1314, over 5663172.25 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3669, pruned_loss=0.1173, over 5696527.17 frames. ], batch size: 71, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:03:27,205 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:03:43,335 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:04:06,373 INFO [train.py:968] (1/2) Epoch 7, batch 36750, giga_loss[loss=0.2873, simple_loss=0.3574, pruned_loss=0.1086, over 27606.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3677, pruned_loss=0.1175, over 5687241.31 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3739, pruned_loss=0.1311, over 5672149.87 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3656, pruned_loss=0.1156, over 5693920.22 frames. ], batch size: 472, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:04:49,961 INFO [train.py:968] (1/2) Epoch 7, batch 36800, libri_loss[loss=0.338, simple_loss=0.3895, pruned_loss=0.1432, over 29596.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3624, pruned_loss=0.1139, over 5679128.22 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.374, pruned_loss=0.1311, over 5667381.96 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3606, pruned_loss=0.1121, over 5688280.44 frames. ], batch size: 74, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:05:02,044 INFO [optim.py:369] (1/2) 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 23:05:23,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:05:28,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:05:30,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-03 23:05:37,874 INFO [train.py:968] (1/2) Epoch 7, batch 36850, giga_loss[loss=0.2586, simple_loss=0.3247, pruned_loss=0.09625, over 28792.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3555, pruned_loss=0.1093, over 5684494.91 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3744, pruned_loss=0.1312, over 5660301.93 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3534, pruned_loss=0.1074, over 5697548.26 frames. ], batch size: 99, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:05:52,125 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:06:18,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3425, 1.6990, 1.3284, 1.6983], device='cuda:1'), covar=tensor([0.2197, 0.2134, 0.2217, 0.1932], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.0899, 0.1047, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 23:06:21,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3046, 3.1152, 2.9668, 1.4654], device='cuda:1'), covar=tensor([0.0760, 0.0846, 0.0814, 0.2248], device='cuda:1'), in_proj_covar=tensor([0.0926, 0.0877, 0.0772, 0.0624], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 23:06:29,122 INFO [train.py:968] (1/2) Epoch 7, batch 36900, giga_loss[loss=0.252, simple_loss=0.3223, pruned_loss=0.09081, over 28860.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3495, pruned_loss=0.1062, over 5664317.50 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.375, pruned_loss=0.1316, over 5657396.81 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3467, pruned_loss=0.1038, over 5677956.63 frames. ], batch size: 99, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:06:43,766 INFO [optim.py:369] (1/2) 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 23:07:15,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4227, 1.6616, 1.1898, 1.6181], device='cuda:1'), covar=tensor([0.0665, 0.0265, 0.0318, 0.0731], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0119, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-03 23:07:17,073 INFO [train.py:968] (1/2) Epoch 7, batch 36950, libri_loss[loss=0.3593, simple_loss=0.4112, pruned_loss=0.1538, over 29505.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3467, pruned_loss=0.1043, over 5670148.51 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3749, pruned_loss=0.1315, over 5661522.25 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3439, pruned_loss=0.1017, over 5677300.22 frames. ], batch size: 85, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:07:28,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:07:37,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:07:58,745 INFO [train.py:968] (1/2) Epoch 7, batch 37000, giga_loss[loss=0.2615, simple_loss=0.3343, pruned_loss=0.0943, over 28936.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3473, pruned_loss=0.1045, over 5673194.48 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3749, pruned_loss=0.1315, over 5663051.68 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3448, pruned_loss=0.1022, over 5677622.43 frames. ], batch size: 112, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:08:12,220 INFO [optim.py:369] (1/2) 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 23:08:15,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-03 23:08:25,964 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=310680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:08:41,679 INFO [train.py:968] (1/2) Epoch 7, batch 37050, giga_loss[loss=0.2692, simple_loss=0.3338, pruned_loss=0.1024, over 28862.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3476, pruned_loss=0.1047, over 5688952.07 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3757, pruned_loss=0.132, over 5666628.12 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3444, pruned_loss=0.102, over 5689569.95 frames. ], batch size: 112, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:09:23,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=310749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:23,951 INFO [train.py:968] (1/2) Epoch 7, batch 37100, giga_loss[loss=0.2519, simple_loss=0.3322, pruned_loss=0.08573, over 28761.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3464, pruned_loss=0.1045, over 5685710.80 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3761, pruned_loss=0.132, over 5671816.03 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3426, pruned_loss=0.1014, over 5682145.11 frames. ], batch size: 284, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:09:28,563 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:31,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:35,046 INFO [optim.py:369] (1/2) 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 23:09:36,665 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:38,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=310769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:09:41,740 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-03 23:09:53,760 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:10:02,599 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:10:03,501 INFO [train.py:968] (1/2) Epoch 7, batch 37150, giga_loss[loss=0.2425, simple_loss=0.3152, pruned_loss=0.08491, over 28999.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3437, pruned_loss=0.103, over 5697629.03 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3764, pruned_loss=0.132, over 5674186.83 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3398, pruned_loss=0.09993, over 5693024.21 frames. ], batch size: 128, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:10:20,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 23:10:41,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5800, 3.2715, 1.5690, 1.5601], device='cuda:1'), covar=tensor([0.0863, 0.0303, 0.0844, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0484, 0.0314, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 23:10:44,555 INFO [train.py:968] (1/2) Epoch 7, batch 37200, libri_loss[loss=0.3396, simple_loss=0.4098, pruned_loss=0.1347, over 29269.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3405, pruned_loss=0.1009, over 5708583.71 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3769, pruned_loss=0.1321, over 5677124.40 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3364, pruned_loss=0.09798, over 5702573.71 frames. ], batch size: 94, lr: 4.50e-03, grad_scale: 8.0
+2023-03-03 23:10:55,168 INFO [optim.py:369] (1/2) 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 23:11:16,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=310892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:11:17,954 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 37250, giga_loss[loss=0.2276, simple_loss=0.3021, pruned_loss=0.07654, over 28556.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3392, pruned_loss=0.1003, over 5715746.07 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3775, pruned_loss=0.1322, over 5680233.74 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3343, pruned_loss=0.097, over 5709102.95 frames. ], batch size: 78, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:11:40,627 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=310924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:11:53,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 23:12:00,781 INFO [train.py:968] (1/2) Epoch 7, batch 37300, giga_loss[loss=0.2627, simple_loss=0.333, pruned_loss=0.09622, over 28489.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3389, pruned_loss=0.1009, over 5698837.51 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3786, pruned_loss=0.1329, over 5670891.68 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3331, pruned_loss=0.09682, over 5702586.22 frames. ], batch size: 71, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:12:13,084 INFO [optim.py:369] (1/2) 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 23:12:41,002 INFO [train.py:968] (1/2) Epoch 7, batch 37350, giga_loss[loss=0.2146, simple_loss=0.2929, pruned_loss=0.06817, over 28449.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3353, pruned_loss=0.09865, over 5706705.79 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3787, pruned_loss=0.1329, over 5674182.20 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3301, pruned_loss=0.095, over 5707033.40 frames. ], batch size: 65, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:13:13,508 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:13:18,789 INFO [train.py:968] (1/2) Epoch 7, batch 37400, giga_loss[loss=0.2262, simple_loss=0.3006, pruned_loss=0.07592, over 28518.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3364, pruned_loss=0.09969, over 5714230.65 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3809, pruned_loss=0.1339, over 5679578.77 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3278, pruned_loss=0.09404, over 5711097.59 frames. ], batch size: 78, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:13:22,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:13:30,123 INFO [optim.py:369] (1/2) 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 23:13:41,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-03 23:13:58,591 INFO [train.py:968] (1/2) Epoch 7, batch 37450, libri_loss[loss=0.3117, simple_loss=0.3893, pruned_loss=0.1171, over 29565.00 frames. ], tot_loss[loss=0.265, simple_loss=0.334, pruned_loss=0.098, over 5721279.57 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3816, pruned_loss=0.1341, over 5680439.35 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3256, pruned_loss=0.09258, over 5718653.66 frames. ], batch size: 77, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:14:37,704 INFO [train.py:968] (1/2) Epoch 7, batch 37500, libri_loss[loss=0.3384, simple_loss=0.4072, pruned_loss=0.1348, over 29537.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3336, pruned_loss=0.09751, over 5731264.94 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3823, pruned_loss=0.1341, over 5688736.23 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3245, pruned_loss=0.09181, over 5723348.95 frames. ], batch size: 84, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:14:38,912 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-03 23:14:51,801 INFO [optim.py:369] (1/2) 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 23:15:15,943 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:15:17,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:15:18,409 INFO [train.py:968] (1/2) Epoch 7, batch 37550, giga_loss[loss=0.3292, simple_loss=0.3919, pruned_loss=0.1333, over 28270.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3369, pruned_loss=0.09987, over 5728792.56 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3827, pruned_loss=0.1342, over 5692815.89 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3281, pruned_loss=0.09443, over 5719385.73 frames. ], batch size: 368, lr: 4.50e-03, grad_scale: 2.0
+2023-03-03 23:15:19,365 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:15:45,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=311230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:16:04,002 INFO [train.py:968] (1/2) Epoch 7, batch 37600, giga_loss[loss=0.3443, simple_loss=0.3954, pruned_loss=0.1466, over 28557.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3414, pruned_loss=0.1029, over 5708986.89 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3831, pruned_loss=0.1344, over 5678257.02 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3332, pruned_loss=0.09771, over 5714272.97 frames. ], batch size: 336, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:16:16,646 INFO [optim.py:369] (1/2) 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 23:16:47,062 INFO [train.py:968] (1/2) Epoch 7, batch 37650, libri_loss[loss=0.3165, simple_loss=0.3837, pruned_loss=0.1246, over 29145.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3492, pruned_loss=0.1083, over 5698437.73 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3832, pruned_loss=0.1344, over 5683606.54 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3411, pruned_loss=0.103, over 5698289.66 frames. ], batch size: 101, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:17:35,048 INFO [train.py:968] (1/2) Epoch 7, batch 37700, giga_loss[loss=0.3078, simple_loss=0.3685, pruned_loss=0.1235, over 28908.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.358, pruned_loss=0.1148, over 5686817.28 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3837, pruned_loss=0.1349, over 5677936.24 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3506, pruned_loss=0.1097, over 5692321.75 frames. ], batch size: 112, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:17:49,896 INFO [optim.py:369] (1/2) 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 23:18:23,652 INFO [train.py:968] (1/2) Epoch 7, batch 37750, giga_loss[loss=0.275, simple_loss=0.3557, pruned_loss=0.09712, over 28922.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3619, pruned_loss=0.1163, over 5668339.35 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3837, pruned_loss=0.1348, over 5672193.64 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3551, pruned_loss=0.1117, over 5677871.12 frames. ], batch size: 213, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:18:38,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:18:41,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:19:09,575 INFO [train.py:968] (1/2) Epoch 7, batch 37800, giga_loss[loss=0.2899, simple_loss=0.3663, pruned_loss=0.1068, over 28302.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3665, pruned_loss=0.1183, over 5676468.51 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3835, pruned_loss=0.1346, over 5678126.29 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3609, pruned_loss=0.1144, over 5678498.91 frames. ], batch size: 65, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:19:14,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5980, 2.2189, 1.5444, 0.8307], device='cuda:1'), covar=tensor([0.2849, 0.1583, 0.2417, 0.3161], device='cuda:1'), in_proj_covar=tensor([0.1432, 0.1370, 0.1427, 0.1183], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 23:19:22,586 INFO [optim.py:369] (1/2) 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 23:19:53,392 INFO [train.py:968] (1/2) Epoch 7, batch 37850, giga_loss[loss=0.3221, simple_loss=0.3846, pruned_loss=0.1298, over 28693.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3714, pruned_loss=0.1215, over 5674110.14 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3832, pruned_loss=0.1346, over 5681389.40 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.118, over 5673087.25 frames. ], batch size: 262, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:19:53,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4987, 2.1062, 1.4748, 0.6557], device='cuda:1'), covar=tensor([0.3021, 0.1712, 0.2595, 0.3418], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1386, 0.1444, 0.1195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 23:20:37,237 INFO [train.py:968] (1/2) Epoch 7, batch 37900, giga_loss[loss=0.2543, simple_loss=0.3321, pruned_loss=0.08823, over 28521.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3687, pruned_loss=0.1192, over 5673734.51 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3826, pruned_loss=0.1343, over 5683298.82 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3655, pruned_loss=0.1167, over 5671314.92 frames. ], batch size: 85, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:20:44,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1700, 2.5463, 1.3171, 1.2674], device='cuda:1'), covar=tensor([0.0948, 0.0314, 0.0865, 0.1453], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0478, 0.0309, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 23:20:47,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:20:49,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:20:50,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4027, 3.2606, 1.4364, 1.5783], device='cuda:1'), covar=tensor([0.0902, 0.0280, 0.0842, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0478, 0.0309, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 23:20:51,372 INFO [optim.py:369] (1/2) 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 23:20:55,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:20:56,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3293, 1.4650, 1.5413, 1.3384], device='cuda:1'), covar=tensor([0.1217, 0.1441, 0.1674, 0.1466], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0724, 0.0646, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 23:21:11,516 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=311592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:21:17,694 INFO [train.py:968] (1/2) Epoch 7, batch 37950, giga_loss[loss=0.2819, simple_loss=0.3605, pruned_loss=0.1017, over 29127.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3644, pruned_loss=0.1155, over 5689447.97 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.383, pruned_loss=0.1346, over 5691176.29 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3609, pruned_loss=0.1125, over 5680010.80 frames. ], batch size: 128, lr: 4.50e-03, grad_scale: 4.0
+2023-03-03 23:21:59,549 INFO [train.py:968] (1/2) Epoch 7, batch 38000, giga_loss[loss=0.2525, simple_loss=0.3351, pruned_loss=0.08492, over 29039.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3637, pruned_loss=0.1143, over 5689180.36 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3835, pruned_loss=0.1351, over 5693046.51 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3601, pruned_loss=0.1111, over 5679960.32 frames. ], batch size: 155, lr: 4.50e-03, grad_scale: 8.0
+2023-03-03 23:22:14,927 INFO [optim.py:369] (1/2) 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 23:22:42,967 INFO [train.py:968] (1/2) Epoch 7, batch 38050, giga_loss[loss=0.308, simple_loss=0.3834, pruned_loss=0.1163, over 28737.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3628, pruned_loss=0.1131, over 5692405.89 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3835, pruned_loss=0.135, over 5695023.66 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3598, pruned_loss=0.1104, over 5683483.62 frames. ], batch size: 119, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:22:54,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:22:57,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:23:04,535 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:23:09,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-03 23:23:22,385 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 38100, giga_loss[loss=0.2875, simple_loss=0.3601, pruned_loss=0.1074, over 28588.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3657, pruned_loss=0.1149, over 5695326.51 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3834, pruned_loss=0.1351, over 5700399.80 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3629, pruned_loss=0.1122, over 5683535.74 frames. ], batch size: 60, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:23:37,348 INFO [optim.py:369] (1/2) 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 23:23:56,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8422, 1.9934, 1.6786, 2.3176], device='cuda:1'), covar=tensor([0.2033, 0.1932, 0.2056, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.1187, 0.0897, 0.1047, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 23:24:03,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=311796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:24:05,911 INFO [train.py:968] (1/2) Epoch 7, batch 38150, giga_loss[loss=0.2963, simple_loss=0.3643, pruned_loss=0.1141, over 28882.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3688, pruned_loss=0.1169, over 5693661.79 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3842, pruned_loss=0.1354, over 5696945.81 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3654, pruned_loss=0.1139, over 5686387.04 frames. ], batch size: 106, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:24:52,164 INFO [train.py:968] (1/2) Epoch 7, batch 38200, giga_loss[loss=0.2996, simple_loss=0.371, pruned_loss=0.1141, over 28988.00 frames. ], tot_loss[loss=0.303, simple_loss=0.37, pruned_loss=0.118, over 5692384.33 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3846, pruned_loss=0.1356, over 5694073.24 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3667, pruned_loss=0.1152, over 5689439.20 frames. ], batch size: 164, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:24:54,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3992, 1.6125, 1.3758, 1.6365], device='cuda:1'), covar=tensor([0.0752, 0.0306, 0.0301, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0119, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-03 23:25:05,353 INFO [optim.py:369] (1/2) 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 23:25:09,017 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-03 23:25:18,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2460, 1.8204, 1.4020, 0.4051], device='cuda:1'), covar=tensor([0.2322, 0.1403, 0.2537, 0.3040], device='cuda:1'), in_proj_covar=tensor([0.1439, 0.1367, 0.1430, 0.1186], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 23:25:21,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4444, 1.5119, 0.9683, 1.2678], device='cuda:1'), covar=tensor([0.0780, 0.0676, 0.1454, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0436, 0.0494, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 23:25:36,056 INFO [train.py:968] (1/2) Epoch 7, batch 38250, giga_loss[loss=0.3046, simple_loss=0.3707, pruned_loss=0.1192, over 28845.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3711, pruned_loss=0.1193, over 5674997.60 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3849, pruned_loss=0.1358, over 5679491.93 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3679, pruned_loss=0.1165, over 5684997.00 frames. ], batch size: 99, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:26:06,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=311939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:08,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=311942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:15,101 INFO [train.py:968] (1/2) Epoch 7, batch 38300, libri_loss[loss=0.3532, simple_loss=0.4031, pruned_loss=0.1517, over 25866.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3718, pruned_loss=0.12, over 5679300.10 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3859, pruned_loss=0.1366, over 5673937.31 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3678, pruned_loss=0.1165, over 5692588.03 frames. ], batch size: 137, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:26:23,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=311960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:28,718 INFO [optim.py:369] (1/2) 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 23:26:32,661 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=311971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:26:56,326 INFO [train.py:968] (1/2) Epoch 7, batch 38350, giga_loss[loss=0.2791, simple_loss=0.3538, pruned_loss=0.1022, over 28890.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3718, pruned_loss=0.1194, over 5687136.83 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3859, pruned_loss=0.1364, over 5677392.03 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3684, pruned_loss=0.1164, over 5694990.08 frames. ], batch size: 112, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:27:35,104 INFO [train.py:968] (1/2) Epoch 7, batch 38400, libri_loss[loss=0.3051, simple_loss=0.3728, pruned_loss=0.1187, over 29511.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3719, pruned_loss=0.1186, over 5694015.57 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3868, pruned_loss=0.1373, over 5676967.26 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3676, pruned_loss=0.1145, over 5700737.07 frames. ], batch size: 81, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:27:48,492 INFO [optim.py:369] (1/2) 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 23:27:57,773 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=312078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:28:15,179 INFO [train.py:968] (1/2) Epoch 7, batch 38450, giga_loss[loss=0.2969, simple_loss=0.3708, pruned_loss=0.1115, over 28852.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3724, pruned_loss=0.118, over 5678661.28 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3872, pruned_loss=0.1375, over 5661342.60 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3683, pruned_loss=0.1141, over 5698883.32 frames. ], batch size: 112, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:28:15,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312100.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:28:31,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=312119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:28:41,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-03 23:28:58,076 INFO [train.py:968] (1/2) Epoch 7, batch 38500, giga_loss[loss=0.2981, simple_loss=0.3446, pruned_loss=0.1258, over 23488.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3703, pruned_loss=0.1165, over 5679610.72 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3877, pruned_loss=0.138, over 5657541.27 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3662, pruned_loss=0.1125, over 5699763.10 frames. ], batch size: 705, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:29:12,699 INFO [optim.py:369] (1/2) 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 23:29:40,860 INFO [train.py:968] (1/2) Epoch 7, batch 38550, giga_loss[loss=0.2645, simple_loss=0.3465, pruned_loss=0.0913, over 29006.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3681, pruned_loss=0.1154, over 5689344.27 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3874, pruned_loss=0.1378, over 5658342.43 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3649, pruned_loss=0.1121, over 5704790.25 frames. ], batch size: 164, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:30:16,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312243.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:30:18,582 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312246.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:30:19,180 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=312247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:30:20,895 INFO [train.py:968] (1/2) Epoch 7, batch 38600, giga_loss[loss=0.2992, simple_loss=0.3663, pruned_loss=0.1161, over 28931.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3641, pruned_loss=0.1126, over 5702118.36 frames. ], libri_tot_loss[loss=0.3309, simple_loss=0.387, pruned_loss=0.1374, over 5661833.05 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3616, pruned_loss=0.1101, over 5711680.00 frames. ], batch size: 227, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:30:30,292 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-03 23:30:35,406 INFO [optim.py:369] (1/2) 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 23:30:40,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312275.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:31:01,683 INFO [train.py:968] (1/2) Epoch 7, batch 38650, giga_loss[loss=0.3223, simple_loss=0.3816, pruned_loss=0.1315, over 28964.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3644, pruned_loss=0.1134, over 5698719.13 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3878, pruned_loss=0.1379, over 5660148.47 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.361, pruned_loss=0.1101, over 5709319.93 frames. ], batch size: 112, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:31:31,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:31:42,382 INFO [train.py:968] (1/2) Epoch 7, batch 38700, giga_loss[loss=0.337, simple_loss=0.3974, pruned_loss=0.1383, over 28554.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3652, pruned_loss=0.1143, over 5705454.12 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3874, pruned_loss=0.1376, over 5662563.99 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3626, pruned_loss=0.1117, over 5711971.68 frames. ], batch size: 71, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:31:56,465 INFO [optim.py:369] (1/2) 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 23:32:21,401 INFO [train.py:968] (1/2) Epoch 7, batch 38750, giga_loss[loss=0.276, simple_loss=0.3498, pruned_loss=0.1011, over 28451.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3657, pruned_loss=0.1141, over 5709245.26 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3876, pruned_loss=0.1378, over 5664193.96 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3629, pruned_loss=0.1114, over 5713521.80 frames. ], batch size: 60, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:32:59,961 INFO [train.py:968] (1/2) Epoch 7, batch 38800, giga_loss[loss=0.2716, simple_loss=0.3518, pruned_loss=0.09568, over 28896.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3647, pruned_loss=0.1129, over 5707077.51 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3877, pruned_loss=0.1379, over 5666138.05 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3622, pruned_loss=0.1103, over 5709500.69 frames. ], batch size: 199, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:33:03,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:13,612 INFO [optim.py:369] (1/2) 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 23:33:23,401 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:26,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:35,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:33:39,355 INFO [train.py:968] (1/2) Epoch 7, batch 38850, libri_loss[loss=0.3699, simple_loss=0.4158, pruned_loss=0.162, over 29473.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3646, pruned_loss=0.1129, over 5717994.70 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3881, pruned_loss=0.1382, over 5673215.32 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3614, pruned_loss=0.1097, over 5715078.37 frames. ], batch size: 85, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:33:44,313 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-03 23:33:47,728 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:34:22,373 INFO [train.py:968] (1/2) Epoch 7, batch 38900, giga_loss[loss=0.2634, simple_loss=0.3357, pruned_loss=0.09555, over 28731.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3633, pruned_loss=0.1125, over 5715370.94 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3876, pruned_loss=0.1378, over 5678289.05 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3605, pruned_loss=0.1096, over 5709274.63 frames. ], batch size: 85, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:34:35,982 INFO [optim.py:369] (1/2) 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 23:34:59,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:02,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:02,619 INFO [train.py:968] (1/2) Epoch 7, batch 38950, giga_loss[loss=0.2544, simple_loss=0.3166, pruned_loss=0.0961, over 23367.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3602, pruned_loss=0.1109, over 5710048.44 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3879, pruned_loss=0.138, over 5684085.98 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3572, pruned_loss=0.1079, over 5701044.74 frames. ], batch size: 710, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:35:13,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-03 23:35:20,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=312622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:26,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:30,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7895, 1.9074, 1.5372, 2.4476], device='cuda:1'), covar=tensor([0.2121, 0.2086, 0.2172, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.1188, 0.0899, 0.1049, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 23:35:32,466 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:34,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=312640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:35:43,220 INFO [train.py:968] (1/2) Epoch 7, batch 39000, giga_loss[loss=0.2486, simple_loss=0.3284, pruned_loss=0.08442, over 29131.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3561, pruned_loss=0.1085, over 5715521.92 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3873, pruned_loss=0.1376, over 5687512.68 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3538, pruned_loss=0.1061, over 5705839.50 frames. ], batch size: 128, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:35:43,221 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-03 23:35:47,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3027, 3.0390, 1.3857, 1.4029], device='cuda:1'), covar=tensor([0.0944, 0.0337, 0.0948, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0481, 0.0312, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0022], device='cuda:1')
+2023-03-03 23:35:52,207 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-03 23:35:55,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2068, 1.0251, 4.3243, 3.3653], device='cuda:1'), covar=tensor([0.1644, 0.2724, 0.0339, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0553, 0.0785, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 23:35:58,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-03 23:36:05,753 INFO [optim.py:369] (1/2) 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 23:36:06,030 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 39050, giga_loss[loss=0.3375, simple_loss=0.3989, pruned_loss=0.138, over 28583.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3573, pruned_loss=0.1095, over 5716071.70 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3874, pruned_loss=0.1375, over 5691597.23 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3538, pruned_loss=0.1061, over 5705700.31 frames. ], batch size: 336, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:37:11,661 INFO [train.py:968] (1/2) Epoch 7, batch 39100, giga_loss[loss=0.2923, simple_loss=0.3508, pruned_loss=0.1169, over 28660.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.358, pruned_loss=0.1111, over 5708721.43 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3872, pruned_loss=0.1376, over 5693801.76 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3545, pruned_loss=0.1075, over 5699128.68 frames. ], batch size: 92, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:37:25,012 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=312765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:37:26,983 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:37:50,391 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-03 23:37:52,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=312797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:37:54,349 INFO [train.py:968] (1/2) Epoch 7, batch 39150, giga_loss[loss=0.2878, simple_loss=0.3542, pruned_loss=0.1107, over 28286.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3545, pruned_loss=0.1091, over 5712472.24 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3874, pruned_loss=0.1377, over 5695496.62 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3513, pruned_loss=0.106, over 5703541.21 frames. ], batch size: 368, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:38:33,414 INFO [train.py:968] (1/2) Epoch 7, batch 39200, giga_loss[loss=0.2625, simple_loss=0.3312, pruned_loss=0.09696, over 28642.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3543, pruned_loss=0.1097, over 5718259.82 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3875, pruned_loss=0.1376, over 5700783.48 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3499, pruned_loss=0.1058, over 5707141.44 frames. ], batch size: 71, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:38:42,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4056, 1.4055, 1.0214, 1.1263], device='cuda:1'), covar=tensor([0.0667, 0.0559, 0.1142, 0.0959], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0441, 0.0496, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 23:38:48,323 INFO [optim.py:369] (1/2) 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 23:39:14,617 INFO [train.py:968] (1/2) Epoch 7, batch 39250, giga_loss[loss=0.2947, simple_loss=0.3583, pruned_loss=0.1155, over 28688.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3517, pruned_loss=0.1085, over 5709505.51 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3875, pruned_loss=0.1376, over 5698168.56 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3482, pruned_loss=0.1054, over 5703446.58 frames. ], batch size: 242, lr: 4.49e-03, grad_scale: 8.0
+2023-03-03 23:39:36,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2943, 3.1055, 2.9589, 1.3444], device='cuda:1'), covar=tensor([0.0765, 0.0873, 0.0875, 0.2284], device='cuda:1'), in_proj_covar=tensor([0.0941, 0.0878, 0.0778, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 23:39:57,662 INFO [train.py:968] (1/2) Epoch 7, batch 39300, giga_loss[loss=0.2752, simple_loss=0.3482, pruned_loss=0.1011, over 28840.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3505, pruned_loss=0.1075, over 5697637.23 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3878, pruned_loss=0.1379, over 5685290.07 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3463, pruned_loss=0.1039, over 5704578.45 frames. ], batch size: 199, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:40:12,121 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:40:41,146 INFO [train.py:968] (1/2) Epoch 7, batch 39350, giga_loss[loss=0.3218, simple_loss=0.392, pruned_loss=0.1258, over 28646.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.352, pruned_loss=0.1075, over 5698690.07 frames. ], libri_tot_loss[loss=0.3314, simple_loss=0.3876, pruned_loss=0.1376, over 5689717.63 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3481, pruned_loss=0.1043, over 5700331.16 frames. ], batch size: 262, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:40:52,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6798, 1.7255, 1.6865, 1.5970], device='cuda:1'), covar=tensor([0.1223, 0.1788, 0.1835, 0.1694], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0725, 0.0646, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 23:40:56,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1917, 3.9781, 3.8183, 1.9947], device='cuda:1'), covar=tensor([0.0406, 0.0564, 0.0594, 0.2200], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0877, 0.0774, 0.0617], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 23:41:07,703 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:41:27,532 INFO [train.py:968] (1/2) Epoch 7, batch 39400, giga_loss[loss=0.3295, simple_loss=0.3769, pruned_loss=0.1411, over 23786.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3557, pruned_loss=0.1099, over 5688254.57 frames. ], libri_tot_loss[loss=0.3317, simple_loss=0.3879, pruned_loss=0.1378, over 5690340.73 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3522, pruned_loss=0.1069, over 5689001.82 frames. ], batch size: 705, lr: 4.49e-03, grad_scale: 4.0
+2023-03-03 23:41:45,643 INFO [optim.py:369] (1/2) 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 23:42:11,333 INFO [train.py:968] (1/2) Epoch 7, batch 39450, giga_loss[loss=0.2824, simple_loss=0.3587, pruned_loss=0.1031, over 29054.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3578, pruned_loss=0.1103, over 5695479.44 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.388, pruned_loss=0.138, over 5692606.35 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3546, pruned_loss=0.1075, over 5694026.89 frames. ], batch size: 128, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:42:54,794 INFO [train.py:968] (1/2) Epoch 7, batch 39500, giga_loss[loss=0.3067, simple_loss=0.3753, pruned_loss=0.1191, over 28629.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3579, pruned_loss=0.1096, over 5695444.22 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3882, pruned_loss=0.1382, over 5696098.00 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3546, pruned_loss=0.1067, over 5691255.98 frames. ], batch size: 336, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:43:11,034 INFO [optim.py:369] (1/2) 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 23:43:32,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2903, 1.8478, 1.3396, 0.5241], device='cuda:1'), covar=tensor([0.2944, 0.1385, 0.2157, 0.3673], device='cuda:1'), in_proj_covar=tensor([0.1451, 0.1361, 0.1427, 0.1188], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 23:43:36,270 INFO [train.py:968] (1/2) Epoch 7, batch 39550, giga_loss[loss=0.2731, simple_loss=0.3472, pruned_loss=0.09948, over 27908.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3572, pruned_loss=0.1092, over 5705215.48 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3882, pruned_loss=0.1382, over 5702804.29 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3532, pruned_loss=0.1057, over 5695655.38 frames. ], batch size: 412, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:43:47,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2253, 1.6938, 1.2696, 1.4726], device='cuda:1'), covar=tensor([0.0750, 0.0297, 0.0341, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0119, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0049, 0.0045, 0.0075], device='cuda:1')
+2023-03-03 23:43:52,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3778, 3.3126, 1.5576, 1.3689], device='cuda:1'), covar=tensor([0.0824, 0.0358, 0.0822, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0487, 0.0313, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 23:44:13,952 INFO [train.py:968] (1/2) Epoch 7, batch 39600, giga_loss[loss=0.3428, simple_loss=0.391, pruned_loss=0.1473, over 28907.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3566, pruned_loss=0.1089, over 5697816.32 frames. ], libri_tot_loss[loss=0.3318, simple_loss=0.3878, pruned_loss=0.1378, over 5695265.82 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3527, pruned_loss=0.1053, over 5696642.19 frames. ], batch size: 186, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:44:31,427 INFO [optim.py:369] (1/2) 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 23:44:54,420 INFO [train.py:968] (1/2) Epoch 7, batch 39650, giga_loss[loss=0.2837, simple_loss=0.35, pruned_loss=0.1087, over 29103.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3581, pruned_loss=0.1099, over 5713228.18 frames. ], libri_tot_loss[loss=0.3321, simple_loss=0.3882, pruned_loss=0.138, over 5701018.28 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3538, pruned_loss=0.1062, over 5707413.07 frames. ], batch size: 113, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:45:16,494 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-03 23:45:30,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=313342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:45:37,764 INFO [train.py:968] (1/2) Epoch 7, batch 39700, giga_loss[loss=0.3272, simple_loss=0.3847, pruned_loss=0.1349, over 28726.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3591, pruned_loss=0.1105, over 5718411.84 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3878, pruned_loss=0.1377, over 5705063.96 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3554, pruned_loss=0.1072, over 5710639.62 frames. ], batch size: 99, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:45:52,196 INFO [optim.py:369] (1/2) 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 23:45:55,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6853, 2.5124, 1.6733, 0.6045], device='cuda:1'), covar=tensor([0.4465, 0.2262, 0.2522, 0.4519], device='cuda:1'), in_proj_covar=tensor([0.1451, 0.1362, 0.1432, 0.1193], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 23:46:00,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8334, 4.5104, 1.8974, 1.8446], device='cuda:1'), covar=tensor([0.0780, 0.0302, 0.0775, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0487, 0.0313, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 23:46:00,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-03 23:46:10,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:46:17,782 INFO [train.py:968] (1/2) Epoch 7, batch 39750, giga_loss[loss=0.3007, simple_loss=0.3745, pruned_loss=0.1134, over 29068.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3624, pruned_loss=0.1121, over 5715562.92 frames. ], libri_tot_loss[loss=0.3315, simple_loss=0.3877, pruned_loss=0.1376, over 5708437.17 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3591, pruned_loss=0.109, over 5706692.99 frames. ], batch size: 155, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:46:23,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=313406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:46:54,304 INFO [train.py:968] (1/2) Epoch 7, batch 39800, giga_loss[loss=0.2884, simple_loss=0.3591, pruned_loss=0.1089, over 28965.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3659, pruned_loss=0.1146, over 5714387.84 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 5703853.11 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3615, pruned_loss=0.1106, over 5711255.35 frames. ], batch size: 112, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:47:00,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2854, 1.5504, 1.2285, 1.7203], device='cuda:1'), covar=tensor([0.2092, 0.2094, 0.2331, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1178, 0.0885, 0.1041, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 23:47:10,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3451, 1.4839, 1.2862, 1.3002], device='cuda:1'), covar=tensor([0.1333, 0.1721, 0.1797, 0.1700], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0727, 0.0651, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 23:47:10,509 INFO [optim.py:369] (1/2) 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 23:47:22,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=313485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:47:25,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=313488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:47:35,129 INFO [train.py:968] (1/2) Epoch 7, batch 39850, giga_loss[loss=0.2809, simple_loss=0.3615, pruned_loss=0.1001, over 28893.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3664, pruned_loss=0.1148, over 5713685.76 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 5706064.60 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3628, pruned_loss=0.1114, over 5709384.37 frames. ], batch size: 227, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:47:52,850 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=313517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:48:20,257 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 39900, giga_loss[loss=0.2575, simple_loss=0.335, pruned_loss=0.08994, over 28452.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3677, pruned_loss=0.1155, over 5712464.40 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 5706859.34 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3646, pruned_loss=0.1126, over 5708292.43 frames. ], batch size: 65, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:48:22,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2955, 4.1076, 3.8718, 1.8105], device='cuda:1'), covar=tensor([0.0443, 0.0582, 0.0672, 0.2028], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0871, 0.0777, 0.0614], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-03 23:48:22,296 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:48:42,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=313581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:48:57,651 INFO [train.py:968] (1/2) Epoch 7, batch 39950, giga_loss[loss=0.2621, simple_loss=0.3317, pruned_loss=0.09628, over 28738.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3684, pruned_loss=0.1163, over 5702322.68 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3886, pruned_loss=0.1383, over 5698914.76 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3652, pruned_loss=0.1133, over 5705643.01 frames. ], batch size: 99, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:49:05,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-03 23:49:35,298 INFO [train.py:968] (1/2) Epoch 7, batch 40000, giga_loss[loss=0.3454, simple_loss=0.3902, pruned_loss=0.1503, over 26805.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3667, pruned_loss=0.1155, over 5710144.36 frames. ], libri_tot_loss[loss=0.3319, simple_loss=0.3881, pruned_loss=0.1379, over 5704186.63 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3641, pruned_loss=0.113, over 5708455.09 frames. ], batch size: 555, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:49:45,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:49:46,350 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313664.0, num_to_drop=1, layers_to_drop={0}
+2023-03-03 23:49:51,146 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 40050, giga_loss[loss=0.3566, simple_loss=0.3963, pruned_loss=0.1584, over 24184.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3632, pruned_loss=0.1138, over 5712683.70 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3885, pruned_loss=0.1381, over 5706377.22 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3604, pruned_loss=0.1112, over 5709630.96 frames. ], batch size: 705, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:50:54,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=313743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:50:59,442 INFO [train.py:968] (1/2) Epoch 7, batch 40100, giga_loss[loss=0.2536, simple_loss=0.3325, pruned_loss=0.08735, over 28633.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3576, pruned_loss=0.1099, over 5713527.88 frames. ], libri_tot_loss[loss=0.3323, simple_loss=0.3885, pruned_loss=0.1381, over 5706377.22 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3555, pruned_loss=0.1079, over 5711151.89 frames. ], batch size: 336, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:51:14,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=313766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:51:16,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4389, 1.4774, 1.6238, 1.3057], device='cuda:1'), covar=tensor([0.1254, 0.1761, 0.1664, 0.1766], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0725, 0.0648, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 23:51:17,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8126, 2.2307, 2.1308, 1.6631], device='cuda:1'), covar=tensor([0.1612, 0.1757, 0.1243, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0707, 0.0813, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 23:51:18,368 INFO [optim.py:369] (1/2) 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 23:51:40,812 INFO [train.py:968] (1/2) Epoch 7, batch 40150, giga_loss[loss=0.3106, simple_loss=0.3886, pruned_loss=0.1163, over 28659.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3589, pruned_loss=0.1091, over 5718776.14 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3887, pruned_loss=0.1382, over 5708485.39 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3567, pruned_loss=0.1071, over 5715083.31 frames. ], batch size: 336, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:51:53,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-03 23:52:06,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9659, 2.5104, 2.3354, 1.8215], device='cuda:1'), covar=tensor([0.1546, 0.1672, 0.1165, 0.1294], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0704, 0.0812, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 23:52:15,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7132, 1.7039, 1.7135, 1.5075], device='cuda:1'), covar=tensor([0.1126, 0.1829, 0.1617, 0.1798], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0720, 0.0642, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 23:52:21,562 INFO [train.py:968] (1/2) Epoch 7, batch 40200, giga_loss[loss=0.2863, simple_loss=0.3693, pruned_loss=0.1016, over 28345.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3602, pruned_loss=0.1089, over 5713516.83 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3886, pruned_loss=0.1382, over 5714447.90 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3577, pruned_loss=0.1064, over 5705572.19 frames. ], batch size: 368, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:52:38,738 INFO [optim.py:369] (1/2) 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 23:53:01,835 INFO [train.py:968] (1/2) Epoch 7, batch 40250, giga_loss[loss=0.3172, simple_loss=0.3774, pruned_loss=0.1285, over 28820.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3593, pruned_loss=0.1088, over 5716011.58 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3891, pruned_loss=0.1385, over 5716491.30 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3564, pruned_loss=0.1062, over 5707884.84 frames. ], batch size: 186, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:53:03,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3888, 3.5701, 1.4638, 1.4231], device='cuda:1'), covar=tensor([0.0934, 0.0405, 0.0912, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0493, 0.0316, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-03 23:53:07,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-03 23:53:09,742 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=313909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:53:11,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=313912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:53:38,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=313941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:53:44,895 INFO [train.py:968] (1/2) Epoch 7, batch 40300, giga_loss[loss=0.2599, simple_loss=0.3351, pruned_loss=0.09231, over 29033.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3588, pruned_loss=0.1102, over 5713723.23 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3888, pruned_loss=0.1383, over 5717278.84 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3567, pruned_loss=0.1081, over 5706739.43 frames. ], batch size: 164, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:54:00,953 INFO [optim.py:369] (1/2) 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 23:54:23,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2312, 0.9448, 0.9038, 1.3671], device='cuda:1'), covar=tensor([0.0705, 0.0312, 0.0333, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0115, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0075], device='cuda:1')
+2023-03-03 23:54:25,824 INFO [train.py:968] (1/2) Epoch 7, batch 40350, libri_loss[loss=0.3534, simple_loss=0.4106, pruned_loss=0.1481, over 25738.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3595, pruned_loss=0.1123, over 5710683.35 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3889, pruned_loss=0.1384, over 5719486.73 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3569, pruned_loss=0.1099, over 5703395.56 frames. ], batch size: 136, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:54:41,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3050, 1.4586, 1.2573, 1.4887], device='cuda:1'), covar=tensor([0.0657, 0.0284, 0.0300, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0068, 0.0049, 0.0044, 0.0075], device='cuda:1')
+2023-03-03 23:54:55,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:54:56,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314039.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:55:04,531 INFO [train.py:968] (1/2) Epoch 7, batch 40400, giga_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 28389.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3582, pruned_loss=0.1125, over 5713663.48 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3889, pruned_loss=0.1383, over 5716556.74 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3553, pruned_loss=0.11, over 5710119.72 frames. ], batch size: 368, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:55:07,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-03 23:55:13,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0737, 1.7251, 1.3354, 0.2960], device='cuda:1'), covar=tensor([0.2319, 0.1191, 0.2047, 0.2693], device='cuda:1'), in_proj_covar=tensor([0.1463, 0.1362, 0.1431, 0.1199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-03 23:55:19,075 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4415, 1.7575, 1.7997, 1.3735], device='cuda:1'), covar=tensor([0.1436, 0.1980, 0.1227, 0.1418], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0704, 0.0811, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-03 23:55:23,328 INFO [optim.py:369] (1/2) 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 23:55:35,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4772, 1.7781, 1.3770, 1.9276], device='cuda:1'), covar=tensor([0.2361, 0.2239, 0.2465, 0.2336], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.0892, 0.1045, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 23:55:47,616 INFO [train.py:968] (1/2) Epoch 7, batch 40450, giga_loss[loss=0.2483, simple_loss=0.3112, pruned_loss=0.09274, over 28748.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3567, pruned_loss=0.1119, over 5720155.82 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.3891, pruned_loss=0.1383, over 5718296.95 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.354, pruned_loss=0.1097, over 5715801.20 frames. ], batch size: 66, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:56:03,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:56:28,074 INFO [train.py:968] (1/2) Epoch 7, batch 40500, giga_loss[loss=0.2468, simple_loss=0.3244, pruned_loss=0.08456, over 28900.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3534, pruned_loss=0.1101, over 5719662.01 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3892, pruned_loss=0.1385, over 5719338.85 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3503, pruned_loss=0.1074, over 5715427.49 frames. ], batch size: 199, lr: 4.48e-03, grad_scale: 8.0
+2023-03-03 23:56:45,803 INFO [optim.py:369] (1/2) 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 23:56:53,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:56:54,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314182.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:56:55,476 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:56:56,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314185.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:56:57,943 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:08,902 INFO [train.py:968] (1/2) Epoch 7, batch 40550, libri_loss[loss=0.2659, simple_loss=0.3281, pruned_loss=0.1018, over 28561.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3495, pruned_loss=0.1079, over 5713662.88 frames. ], libri_tot_loss[loss=0.3332, simple_loss=0.3893, pruned_loss=0.1385, over 5713903.56 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3461, pruned_loss=0.1051, over 5715252.19 frames. ], batch size: 63, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:57:19,482 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:20,783 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314214.0, num_to_drop=1, layers_to_drop={1}
+2023-03-03 23:57:22,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:50,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-03 23:57:50,335 INFO [train.py:968] (1/2) Epoch 7, batch 40600, libri_loss[loss=0.3306, simple_loss=0.3927, pruned_loss=0.1342, over 29658.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3463, pruned_loss=0.1063, over 5714698.02 frames. ], libri_tot_loss[loss=0.3327, simple_loss=0.3889, pruned_loss=0.1383, over 5716758.32 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3433, pruned_loss=0.1039, over 5713328.30 frames. ], batch size: 91, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:57:51,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6158, 1.7037, 1.7337, 1.5542], device='cuda:1'), covar=tensor([0.1287, 0.1785, 0.1674, 0.1703], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0730, 0.0655, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-03 23:57:54,918 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:57,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:57:59,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:58:06,564 INFO [optim.py:369] (1/2) 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 23:58:22,537 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-03 23:58:29,251 INFO [train.py:968] (1/2) Epoch 7, batch 40650, giga_loss[loss=0.2423, simple_loss=0.3204, pruned_loss=0.08207, over 28892.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3481, pruned_loss=0.1071, over 5708165.57 frames. ], libri_tot_loss[loss=0.3324, simple_loss=0.3887, pruned_loss=0.1381, over 5717172.01 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3449, pruned_loss=0.1045, over 5706909.08 frames. ], batch size: 112, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:59:12,103 INFO [train.py:968] (1/2) Epoch 7, batch 40700, giga_loss[loss=0.2937, simple_loss=0.3682, pruned_loss=0.1096, over 28589.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.35, pruned_loss=0.1069, over 5713320.22 frames. ], libri_tot_loss[loss=0.3322, simple_loss=0.3885, pruned_loss=0.1379, over 5719098.42 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3473, pruned_loss=0.1046, over 5710494.91 frames. ], batch size: 307, lr: 4.48e-03, grad_scale: 4.0
+2023-03-03 23:59:25,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3904, 1.7907, 1.3685, 1.6535], device='cuda:1'), covar=tensor([0.2171, 0.1956, 0.2130, 0.1975], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.0892, 0.1041, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-03 23:59:29,002 INFO [optim.py:369] (1/2) 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 23:59:49,871 INFO [train.py:968] (1/2) Epoch 7, batch 40750, giga_loss[loss=0.2959, simple_loss=0.3692, pruned_loss=0.1113, over 28946.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3537, pruned_loss=0.1085, over 5707331.61 frames. ], libri_tot_loss[loss=0.3325, simple_loss=0.3887, pruned_loss=0.1382, over 5715386.66 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3498, pruned_loss=0.1053, over 5708257.38 frames. ], batch size: 213, lr: 4.48e-03, grad_scale: 4.0
+2023-03-04 00:00:09,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 00:00:15,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3804, 1.6457, 1.4183, 1.5947], device='cuda:1'), covar=tensor([0.0707, 0.0279, 0.0317, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0116, 0.0119, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0075], device='cuda:1')
+2023-03-04 00:00:30,310 INFO [train.py:968] (1/2) Epoch 7, batch 40800, giga_loss[loss=0.2954, simple_loss=0.3703, pruned_loss=0.1102, over 28767.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3576, pruned_loss=0.1101, over 5714553.08 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.389, pruned_loss=0.1383, over 5716801.41 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3534, pruned_loss=0.1067, over 5713857.77 frames. ], batch size: 119, lr: 4.48e-03, grad_scale: 8.0
+2023-03-04 00:00:49,059 INFO [optim.py:369] (1/2) 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 00:01:11,376 INFO [train.py:968] (1/2) Epoch 7, batch 40850, libri_loss[loss=0.2962, simple_loss=0.3629, pruned_loss=0.1148, over 29556.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3593, pruned_loss=0.1109, over 5718895.08 frames. ], libri_tot_loss[loss=0.3326, simple_loss=0.3888, pruned_loss=0.1381, over 5720756.13 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3556, pruned_loss=0.1078, over 5714795.24 frames. ], batch size: 80, lr: 4.47e-03, grad_scale: 8.0
+2023-03-04 00:01:55,871 INFO [train.py:968] (1/2) Epoch 7, batch 40900, giga_loss[loss=0.4136, simple_loss=0.4238, pruned_loss=0.2017, over 23591.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3626, pruned_loss=0.1138, over 5709348.96 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3892, pruned_loss=0.1384, over 5723215.95 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.359, pruned_loss=0.1108, over 5703895.71 frames. ], batch size: 705, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:02:10,023 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:02:23,497 INFO [optim.py:369] (1/2) 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 00:02:34,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 00:02:41,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:02:44,324 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:02:50,399 INFO [train.py:968] (1/2) Epoch 7, batch 40950, giga_loss[loss=0.3419, simple_loss=0.3989, pruned_loss=0.1424, over 28731.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1223, over 5686564.13 frames. ], libri_tot_loss[loss=0.3328, simple_loss=0.389, pruned_loss=0.1384, over 5720980.49 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3688, pruned_loss=0.1197, over 5684318.97 frames. ], batch size: 284, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:03:20,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:03:30,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2505, 1.1114, 4.2698, 3.2417], device='cuda:1'), covar=tensor([0.1572, 0.2494, 0.0366, 0.0926], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0555, 0.0791, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:03:37,982 INFO [train.py:968] (1/2) Epoch 7, batch 41000, giga_loss[loss=0.326, simple_loss=0.4003, pruned_loss=0.1259, over 28917.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3789, pruned_loss=0.1275, over 5688626.31 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3892, pruned_loss=0.1387, over 5722429.39 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3762, pruned_loss=0.125, over 5685116.67 frames. ], batch size: 145, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:03:59,627 INFO [optim.py:369] (1/2) 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 00:04:26,941 INFO [train.py:968] (1/2) Epoch 7, batch 41050, giga_loss[loss=0.4109, simple_loss=0.4388, pruned_loss=0.1915, over 27632.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3852, pruned_loss=0.1332, over 5670453.66 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.389, pruned_loss=0.1385, over 5723953.01 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3832, pruned_loss=0.1313, over 5666139.68 frames. ], batch size: 472, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:04:31,667 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:34,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:54,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-04 00:04:56,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:58,559 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:04:59,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:10,429 INFO [train.py:968] (1/2) Epoch 7, batch 41100, giga_loss[loss=0.317, simple_loss=0.3815, pruned_loss=0.1263, over 28703.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3911, pruned_loss=0.1383, over 5669258.03 frames. ], libri_tot_loss[loss=0.3335, simple_loss=0.3895, pruned_loss=0.1387, over 5717543.58 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.389, pruned_loss=0.1365, over 5671083.00 frames. ], batch size: 66, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:05:26,848 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:34,253 INFO [optim.py:369] (1/2) 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 00:05:34,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=314775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:37,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=314778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:05:59,458 INFO [train.py:968] (1/2) Epoch 7, batch 41150, giga_loss[loss=0.3384, simple_loss=0.395, pruned_loss=0.1408, over 28815.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3973, pruned_loss=0.1433, over 5665935.69 frames. ], libri_tot_loss[loss=0.3333, simple_loss=0.3894, pruned_loss=0.1386, over 5717876.78 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3958, pruned_loss=0.142, over 5666175.61 frames. ], batch size: 112, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:06:07,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=314807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:06:52,993 INFO [train.py:968] (1/2) Epoch 7, batch 41200, giga_loss[loss=0.3492, simple_loss=0.4124, pruned_loss=0.143, over 28867.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3989, pruned_loss=0.1455, over 5662750.77 frames. ], libri_tot_loss[loss=0.3334, simple_loss=0.3895, pruned_loss=0.1387, over 5720161.94 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3978, pruned_loss=0.1445, over 5659957.79 frames. ], batch size: 145, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:07:21,348 INFO [optim.py:369] (1/2) 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 00:07:49,325 INFO [train.py:968] (1/2) Epoch 7, batch 41250, giga_loss[loss=0.3438, simple_loss=0.3944, pruned_loss=0.1466, over 28798.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4008, pruned_loss=0.1486, over 5631091.76 frames. ], libri_tot_loss[loss=0.3331, simple_loss=0.3892, pruned_loss=0.1385, over 5713878.65 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.4003, pruned_loss=0.1482, over 5633635.74 frames. ], batch size: 284, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:08:17,625 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=314929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:08:39,196 INFO [train.py:968] (1/2) Epoch 7, batch 41300, giga_loss[loss=0.408, simple_loss=0.4415, pruned_loss=0.1872, over 28552.00 frames. ], tot_loss[loss=0.354, simple_loss=0.4038, pruned_loss=0.1521, over 5609296.88 frames. ], libri_tot_loss[loss=0.333, simple_loss=0.3891, pruned_loss=0.1385, over 5699865.03 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4039, pruned_loss=0.1521, over 5621352.55 frames. ], batch size: 336, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:08:41,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3689, 1.5193, 1.5137, 1.4404], device='cuda:1'), covar=tensor([0.1097, 0.1158, 0.1355, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0731, 0.0653, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 00:08:41,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8147, 2.5660, 1.6879, 1.1636], device='cuda:1'), covar=tensor([0.3886, 0.1934, 0.2273, 0.3360], device='cuda:1'), in_proj_covar=tensor([0.1480, 0.1391, 0.1444, 0.1211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 00:08:56,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=314968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:08:59,977 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 00:09:02,416 INFO [optim.py:369] (1/2) 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 00:09:12,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-04 00:09:26,419 INFO [train.py:968] (1/2) Epoch 7, batch 41350, giga_loss[loss=0.3828, simple_loss=0.4451, pruned_loss=0.1602, over 28846.00 frames. ], tot_loss[loss=0.3579, simple_loss=0.4064, pruned_loss=0.1547, over 5621140.89 frames. ], libri_tot_loss[loss=0.332, simple_loss=0.3881, pruned_loss=0.1379, over 5704933.11 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4084, pruned_loss=0.1561, over 5621835.52 frames. ], batch size: 174, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:10:12,721 INFO [train.py:968] (1/2) Epoch 7, batch 41400, giga_loss[loss=0.4651, simple_loss=0.4707, pruned_loss=0.2298, over 26645.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.4074, pruned_loss=0.1555, over 5632065.13 frames. ], libri_tot_loss[loss=0.3316, simple_loss=0.3878, pruned_loss=0.1377, over 5708438.53 frames. ], giga_tot_loss[loss=0.3626, simple_loss=0.4102, pruned_loss=0.1575, over 5626051.62 frames. ], batch size: 555, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:10:29,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4206, 3.2176, 1.4566, 1.3374], device='cuda:1'), covar=tensor([0.0868, 0.0285, 0.0823, 0.1369], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0494, 0.0317, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 00:10:42,162 INFO [optim.py:369] (1/2) 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 00:11:03,835 INFO [train.py:968] (1/2) Epoch 7, batch 41450, giga_loss[loss=0.3576, simple_loss=0.4051, pruned_loss=0.155, over 28610.00 frames. ], tot_loss[loss=0.3585, simple_loss=0.4062, pruned_loss=0.1554, over 5633229.97 frames. ], libri_tot_loss[loss=0.3313, simple_loss=0.3874, pruned_loss=0.1376, over 5702981.75 frames. ], giga_tot_loss[loss=0.3627, simple_loss=0.4095, pruned_loss=0.1579, over 5629928.65 frames. ], batch size: 307, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:11:04,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8265, 1.6663, 1.7952, 1.6201], device='cuda:1'), covar=tensor([0.1188, 0.1931, 0.1682, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0735, 0.0658, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 00:11:15,925 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:11:18,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:11:47,043 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:11:54,866 INFO [train.py:968] (1/2) Epoch 7, batch 41500, giga_loss[loss=0.4689, simple_loss=0.4718, pruned_loss=0.233, over 26516.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4054, pruned_loss=0.1553, over 5625531.00 frames. ], libri_tot_loss[loss=0.3311, simple_loss=0.3873, pruned_loss=0.1374, over 5703418.28 frames. ], giga_tot_loss[loss=0.362, simple_loss=0.4084, pruned_loss=0.1578, over 5621166.71 frames. ], batch size: 555, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:12:07,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1974, 1.2897, 1.0599, 1.0006], device='cuda:1'), covar=tensor([0.0638, 0.0403, 0.0906, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0450, 0.0503, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 00:12:19,297 INFO [optim.py:369] (1/2) 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 00:12:43,628 INFO [train.py:968] (1/2) Epoch 7, batch 41550, giga_loss[loss=0.3169, simple_loss=0.3917, pruned_loss=0.1211, over 28986.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.4047, pruned_loss=0.1535, over 5624616.25 frames. ], libri_tot_loss[loss=0.3312, simple_loss=0.3876, pruned_loss=0.1375, over 5705055.01 frames. ], giga_tot_loss[loss=0.3596, simple_loss=0.4073, pruned_loss=0.1559, over 5617845.39 frames. ], batch size: 155, lr: 4.47e-03, grad_scale: 2.0
+2023-03-04 00:13:38,783 INFO [train.py:968] (1/2) Epoch 7, batch 41600, giga_loss[loss=0.3226, simple_loss=0.388, pruned_loss=0.1286, over 28880.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.4052, pruned_loss=0.1539, over 5613820.19 frames. ], libri_tot_loss[loss=0.3304, simple_loss=0.3868, pruned_loss=0.137, over 5700038.58 frames. ], giga_tot_loss[loss=0.361, simple_loss=0.4086, pruned_loss=0.1567, over 5610061.56 frames. ], batch size: 213, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:14:06,504 INFO [optim.py:369] (1/2) 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 00:14:14,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2833, 2.8456, 1.4423, 1.3125], device='cuda:1'), covar=tensor([0.0871, 0.0307, 0.0811, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0494, 0.0317, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 00:14:18,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-04 00:14:30,351 INFO [train.py:968] (1/2) Epoch 7, batch 41650, giga_loss[loss=0.2871, simple_loss=0.3614, pruned_loss=0.1064, over 29028.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.4053, pruned_loss=0.1537, over 5600428.53 frames. ], libri_tot_loss[loss=0.3307, simple_loss=0.3872, pruned_loss=0.1372, over 5703422.90 frames. ], giga_tot_loss[loss=0.3603, simple_loss=0.4081, pruned_loss=0.1562, over 5592406.89 frames. ], batch size: 155, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:14:34,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:15:00,350 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:15:23,810 INFO [train.py:968] (1/2) Epoch 7, batch 41700, giga_loss[loss=0.3795, simple_loss=0.4302, pruned_loss=0.1644, over 28584.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4013, pruned_loss=0.1492, over 5616016.35 frames. ], libri_tot_loss[loss=0.3301, simple_loss=0.3866, pruned_loss=0.1368, over 5704546.18 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4041, pruned_loss=0.1516, over 5607944.67 frames. ], batch size: 336, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:15:29,244 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:15:50,001 INFO [optim.py:369] (1/2) 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 00:16:11,752 INFO [train.py:968] (1/2) Epoch 7, batch 41750, giga_loss[loss=0.2946, simple_loss=0.3663, pruned_loss=0.1115, over 28779.00 frames. ], tot_loss[loss=0.3446, simple_loss=0.3984, pruned_loss=0.1454, over 5622141.82 frames. ], libri_tot_loss[loss=0.33, simple_loss=0.3864, pruned_loss=0.1367, over 5696082.04 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4012, pruned_loss=0.1477, over 5620284.42 frames. ], batch size: 99, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:16:33,141 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:16:58,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:17:00,052 INFO [train.py:968] (1/2) Epoch 7, batch 41800, giga_loss[loss=0.3187, simple_loss=0.3806, pruned_loss=0.1285, over 28817.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3961, pruned_loss=0.1439, over 5627472.03 frames. ], libri_tot_loss[loss=0.3296, simple_loss=0.3861, pruned_loss=0.1366, over 5703654.82 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3992, pruned_loss=0.1462, over 5616671.09 frames. ], batch size: 284, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:17:00,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:17:02,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5294, 4.4042, 1.7299, 1.6839], device='cuda:1'), covar=tensor([0.0895, 0.0293, 0.0794, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0492, 0.0316, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 00:17:23,351 INFO [optim.py:369] (1/2) 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 00:17:26,009 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:17:46,345 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 00:17:47,383 INFO [train.py:968] (1/2) Epoch 7, batch 41850, giga_loss[loss=0.3398, simple_loss=0.3962, pruned_loss=0.1417, over 29100.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3939, pruned_loss=0.1418, over 5621374.01 frames. ], libri_tot_loss[loss=0.3297, simple_loss=0.3861, pruned_loss=0.1366, over 5701854.26 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3966, pruned_loss=0.1438, over 5612214.37 frames. ], batch size: 113, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:18:34,188 INFO [train.py:968] (1/2) Epoch 7, batch 41900, giga_loss[loss=0.3254, simple_loss=0.3911, pruned_loss=0.1299, over 28982.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3915, pruned_loss=0.14, over 5646661.97 frames. ], libri_tot_loss[loss=0.3294, simple_loss=0.3858, pruned_loss=0.1365, over 5706797.54 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3941, pruned_loss=0.1418, over 5633056.03 frames. ], batch size: 164, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:18:51,329 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:19:02,294 INFO [optim.py:369] (1/2) 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 00:19:21,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.29 vs. limit=5.0
+2023-03-04 00:19:22,525 INFO [train.py:968] (1/2) Epoch 7, batch 41950, giga_loss[loss=0.3026, simple_loss=0.3757, pruned_loss=0.1147, over 29082.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3928, pruned_loss=0.1412, over 5653058.78 frames. ], libri_tot_loss[loss=0.3299, simple_loss=0.3862, pruned_loss=0.1369, over 5709651.03 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3947, pruned_loss=0.1424, over 5638870.22 frames. ], batch size: 128, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:20:12,756 INFO [train.py:968] (1/2) Epoch 7, batch 42000, giga_loss[loss=0.3371, simple_loss=0.3916, pruned_loss=0.1413, over 28028.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3906, pruned_loss=0.1395, over 5646136.85 frames. ], libri_tot_loss[loss=0.3292, simple_loss=0.3855, pruned_loss=0.1365, over 5707658.17 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3929, pruned_loss=0.1409, over 5634467.16 frames. ], batch size: 412, lr: 4.47e-03, grad_scale: 8.0
+2023-03-04 00:20:12,757 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 00:20:20,972 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 00:20:47,638 INFO [optim.py:369] (1/2) 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 00:21:10,923 INFO [train.py:968] (1/2) Epoch 7, batch 42050, giga_loss[loss=0.2972, simple_loss=0.3777, pruned_loss=0.1084, over 28991.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3877, pruned_loss=0.1358, over 5645967.96 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3853, pruned_loss=0.1364, over 5708716.57 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3898, pruned_loss=0.1371, over 5634065.52 frames. ], batch size: 164, lr: 4.47e-03, grad_scale: 8.0
+2023-03-04 00:21:14,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:21:33,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:21:44,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:22:10,249 INFO [train.py:968] (1/2) Epoch 7, batch 42100, giga_loss[loss=0.3168, simple_loss=0.3877, pruned_loss=0.1229, over 28706.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3884, pruned_loss=0.1334, over 5653808.42 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3849, pruned_loss=0.1362, over 5710480.64 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3905, pruned_loss=0.1346, over 5642241.65 frames. ], batch size: 262, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:22:24,018 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:22:27,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 00:22:32,625 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=315775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:22:33,738 INFO [optim.py:369] (1/2) 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 00:22:46,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4847, 3.0093, 1.5649, 1.4532], device='cuda:1'), covar=tensor([0.0807, 0.0336, 0.0776, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0498, 0.0319, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 00:22:54,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:22:55,144 INFO [train.py:968] (1/2) Epoch 7, batch 42150, giga_loss[loss=0.3433, simple_loss=0.4045, pruned_loss=0.1411, over 29071.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3897, pruned_loss=0.1338, over 5667139.87 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.3848, pruned_loss=0.1362, over 5716182.81 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3916, pruned_loss=0.1347, over 5651388.98 frames. ], batch size: 155, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:23:38,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:23:42,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:23:43,097 INFO [train.py:968] (1/2) Epoch 7, batch 42200, giga_loss[loss=0.2901, simple_loss=0.3629, pruned_loss=0.1086, over 28900.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3918, pruned_loss=0.1362, over 5659454.80 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3848, pruned_loss=0.1361, over 5711129.80 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3935, pruned_loss=0.137, over 5650478.18 frames. ], batch size: 112, lr: 4.47e-03, grad_scale: 4.0
+2023-03-04 00:24:02,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:24:04,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:24:06,541 INFO [optim.py:369] (1/2) 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 00:24:06,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:24:10,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6883, 1.7502, 1.6264, 1.6361], device='cuda:1'), covar=tensor([0.1080, 0.1511, 0.1625, 0.1435], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0732, 0.0655, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 00:24:27,315 INFO [train.py:968] (1/2) Epoch 7, batch 42250, giga_loss[loss=0.2905, simple_loss=0.3565, pruned_loss=0.1123, over 28868.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3905, pruned_loss=0.136, over 5665484.18 frames. ], libri_tot_loss[loss=0.3286, simple_loss=0.385, pruned_loss=0.1361, over 5710154.35 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3918, pruned_loss=0.1366, over 5657968.00 frames. ], batch size: 99, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:24:31,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:06,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=315941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:06,778 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=315941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:09,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=315944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:13,281 INFO [train.py:968] (1/2) Epoch 7, batch 42300, giga_loss[loss=0.3241, simple_loss=0.3791, pruned_loss=0.1345, over 28948.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3898, pruned_loss=0.1371, over 5668724.89 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3852, pruned_loss=0.1364, over 5713247.09 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3906, pruned_loss=0.1373, over 5659181.19 frames. ], batch size: 227, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:25:34,536 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=315973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:25:37,940 INFO [optim.py:369] (1/2) 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 00:25:41,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2575, 1.5287, 1.1571, 1.5562], device='cuda:1'), covar=tensor([0.2170, 0.1993, 0.2252, 0.1863], device='cuda:1'), in_proj_covar=tensor([0.1192, 0.0898, 0.1049, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:25:49,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4219, 1.6910, 1.7102, 1.3539], device='cuda:1'), covar=tensor([0.1238, 0.1822, 0.1004, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0714, 0.0809, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 00:26:00,523 INFO [train.py:968] (1/2) Epoch 7, batch 42350, giga_loss[loss=0.3447, simple_loss=0.4039, pruned_loss=0.1427, over 28267.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3883, pruned_loss=0.1366, over 5667712.34 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3849, pruned_loss=0.1361, over 5715331.00 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3895, pruned_loss=0.1371, over 5657044.49 frames. ], batch size: 368, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:26:08,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4354, 1.6396, 1.4486, 1.6225], device='cuda:1'), covar=tensor([0.1288, 0.1624, 0.1925, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0738, 0.0659, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 00:26:15,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0162, 1.1719, 3.3740, 2.9610], device='cuda:1'), covar=tensor([0.1583, 0.2432, 0.0432, 0.1560], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0556, 0.0799, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:26:48,665 INFO [train.py:968] (1/2) Epoch 7, batch 42400, giga_loss[loss=0.3076, simple_loss=0.3827, pruned_loss=0.1163, over 28940.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3885, pruned_loss=0.1354, over 5666814.34 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.385, pruned_loss=0.136, over 5709372.59 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3894, pruned_loss=0.1358, over 5662264.63 frames. ], batch size: 227, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:27:13,784 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-04 00:27:13,960 INFO [optim.py:369] (1/2) 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 00:27:21,662 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:27:24,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:27:27,967 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:27:34,123 INFO [train.py:968] (1/2) Epoch 7, batch 42450, giga_loss[loss=0.3253, simple_loss=0.3815, pruned_loss=0.1346, over 28665.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3877, pruned_loss=0.134, over 5678263.10 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3848, pruned_loss=0.1358, over 5711564.75 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3887, pruned_loss=0.1345, over 5671747.49 frames. ], batch size: 119, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:27:51,801 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:28:16,432 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 42500, giga_loss[loss=0.264, simple_loss=0.3446, pruned_loss=0.09171, over 28798.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3883, pruned_loss=0.1349, over 5664321.84 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3844, pruned_loss=0.1357, over 5704937.26 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3895, pruned_loss=0.1353, over 5664985.12 frames. ], batch size: 119, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:28:24,564 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:28:55,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5653, 1.0108, 2.8594, 2.5930], device='cuda:1'), covar=tensor([0.1691, 0.2217, 0.0554, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0556, 0.0801, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:29:10,687 INFO [train.py:968] (1/2) Epoch 7, batch 42550, libri_loss[loss=0.3456, simple_loss=0.4083, pruned_loss=0.1415, over 25642.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3859, pruned_loss=0.1334, over 5667655.37 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3846, pruned_loss=0.1357, over 5704078.24 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3867, pruned_loss=0.1337, over 5668583.19 frames. ], batch size: 136, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:29:42,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:29:45,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:29:56,501 INFO [train.py:968] (1/2) Epoch 7, batch 42600, giga_loss[loss=0.2865, simple_loss=0.349, pruned_loss=0.112, over 28343.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3846, pruned_loss=0.133, over 5667394.27 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3842, pruned_loss=0.1354, over 5707341.35 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3855, pruned_loss=0.1334, over 5664541.36 frames. ], batch size: 78, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:30:15,311 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:28,543 INFO [optim.py:369] (1/2) 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 00:30:33,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:35,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:41,159 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:45,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:30:47,699 INFO [train.py:968] (1/2) Epoch 7, batch 42650, giga_loss[loss=0.3326, simple_loss=0.3847, pruned_loss=0.1402, over 27600.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3826, pruned_loss=0.1322, over 5673728.46 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3842, pruned_loss=0.1354, over 5709208.01 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3834, pruned_loss=0.1326, over 5669485.42 frames. ], batch size: 472, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:30:56,369 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=316306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:31:07,682 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:31:13,999 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 7, batch 42700, giga_loss[loss=0.2888, simple_loss=0.3492, pruned_loss=0.1141, over 28888.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3813, pruned_loss=0.1315, over 5680499.34 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3846, pruned_loss=0.1357, over 5709583.44 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3815, pruned_loss=0.1314, over 5676017.73 frames. ], batch size: 99, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:32:02,021 INFO [optim.py:369] (1/2) 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 00:32:22,814 INFO [train.py:968] (1/2) Epoch 7, batch 42750, libri_loss[loss=0.3208, simple_loss=0.386, pruned_loss=0.1278, over 29232.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3816, pruned_loss=0.1327, over 5672340.63 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3846, pruned_loss=0.1357, over 5705009.19 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3816, pruned_loss=0.1326, over 5671456.38 frames. ], batch size: 97, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:32:30,516 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=316407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:32:57,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-04 00:33:12,667 INFO [train.py:968] (1/2) Epoch 7, batch 42800, giga_loss[loss=0.29, simple_loss=0.3589, pruned_loss=0.1105, over 28806.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3824, pruned_loss=0.1342, over 5658876.11 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3847, pruned_loss=0.1357, over 5707547.91 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3823, pruned_loss=0.134, over 5654882.55 frames. ], batch size: 66, lr: 4.46e-03, grad_scale: 8.0
+2023-03-04 00:33:13,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3600, 3.1405, 1.5152, 1.3897], device='cuda:1'), covar=tensor([0.0879, 0.0304, 0.0824, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0498, 0.0318, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 00:33:38,962 INFO [optim.py:369] (1/2) 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 00:33:53,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9039, 4.7105, 4.4657, 1.9956], device='cuda:1'), covar=tensor([0.0373, 0.0580, 0.0642, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.0970, 0.0918, 0.0807, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 00:33:57,846 INFO [train.py:968] (1/2) Epoch 7, batch 42850, giga_loss[loss=0.2962, simple_loss=0.3686, pruned_loss=0.1119, over 28988.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3825, pruned_loss=0.1336, over 5656855.94 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3846, pruned_loss=0.1357, over 5700755.05 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3825, pruned_loss=0.1334, over 5658401.18 frames. ], batch size: 213, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:34:42,926 INFO [train.py:968] (1/2) Epoch 7, batch 42900, giga_loss[loss=0.4319, simple_loss=0.446, pruned_loss=0.2089, over 26571.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3832, pruned_loss=0.1332, over 5664567.29 frames. ], libri_tot_loss[loss=0.3277, simple_loss=0.3844, pruned_loss=0.1355, over 5701314.08 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3833, pruned_loss=0.1332, over 5664533.49 frames. ], batch size: 555, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:34:45,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1334, 1.3001, 0.9859, 0.9717], device='cuda:1'), covar=tensor([0.0817, 0.0477, 0.1240, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0447, 0.0504, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 00:35:09,379 INFO [optim.py:369] (1/2) 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 00:35:25,781 INFO [train.py:968] (1/2) Epoch 7, batch 42950, giga_loss[loss=0.3085, simple_loss=0.3726, pruned_loss=0.1221, over 28331.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3831, pruned_loss=0.1327, over 5662706.07 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.384, pruned_loss=0.1354, over 5696307.00 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3835, pruned_loss=0.1327, over 5666754.32 frames. ], batch size: 368, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:36:03,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8285, 2.5150, 1.6490, 2.3055], device='cuda:1'), covar=tensor([0.0528, 0.0575, 0.0927, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0442, 0.0498, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 00:36:06,182 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-04 00:36:13,699 INFO [train.py:968] (1/2) Epoch 7, batch 43000, giga_loss[loss=0.4401, simple_loss=0.4532, pruned_loss=0.2136, over 27575.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3849, pruned_loss=0.134, over 5656302.73 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3847, pruned_loss=0.136, over 5679292.09 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3845, pruned_loss=0.1334, over 5673828.67 frames. ], batch size: 472, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:36:46,427 INFO [optim.py:369] (1/2) 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 00:36:48,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:37:04,831 INFO [train.py:968] (1/2) Epoch 7, batch 43050, giga_loss[loss=0.3988, simple_loss=0.4342, pruned_loss=0.1817, over 28053.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3881, pruned_loss=0.137, over 5656992.57 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3849, pruned_loss=0.1363, over 5673858.61 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3876, pruned_loss=0.1363, over 5675231.10 frames. ], batch size: 412, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:37:17,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9016, 4.7013, 4.4902, 2.1511], device='cuda:1'), covar=tensor([0.0370, 0.0533, 0.0597, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.0970, 0.0919, 0.0809, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 00:37:55,329 INFO [train.py:968] (1/2) Epoch 7, batch 43100, giga_loss[loss=0.345, simple_loss=0.395, pruned_loss=0.1475, over 28674.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3873, pruned_loss=0.1376, over 5659349.43 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3843, pruned_loss=0.1358, over 5664898.15 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3876, pruned_loss=0.1374, over 5681632.61 frames. ], batch size: 119, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:38:27,038 INFO [optim.py:369] (1/2) 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 00:38:31,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=316782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:38:41,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-04 00:38:47,073 INFO [train.py:968] (1/2) Epoch 7, batch 43150, libri_loss[loss=0.3269, simple_loss=0.3887, pruned_loss=0.1325, over 29388.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3889, pruned_loss=0.14, over 5652492.23 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3845, pruned_loss=0.1359, over 5659506.45 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3891, pruned_loss=0.1399, over 5674560.18 frames. ], batch size: 92, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:39:11,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:39:14,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:39:36,606 INFO [train.py:968] (1/2) Epoch 7, batch 43200, giga_loss[loss=0.3256, simple_loss=0.378, pruned_loss=0.1366, over 27662.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3912, pruned_loss=0.1425, over 5641890.41 frames. ], libri_tot_loss[loss=0.3284, simple_loss=0.3846, pruned_loss=0.1361, over 5661498.65 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3913, pruned_loss=0.1423, over 5657386.44 frames. ], batch size: 474, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:39:41,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=316856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:39:57,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.72 vs. limit=5.0
+2023-03-04 00:40:02,625 INFO [optim.py:369] (1/2) 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 00:40:21,135 INFO [train.py:968] (1/2) Epoch 7, batch 43250, giga_loss[loss=0.3352, simple_loss=0.3883, pruned_loss=0.141, over 28726.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3898, pruned_loss=0.1415, over 5655053.95 frames. ], libri_tot_loss[loss=0.3279, simple_loss=0.3843, pruned_loss=0.1357, over 5669299.09 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3903, pruned_loss=0.1419, over 5660127.34 frames. ], batch size: 284, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:40:30,668 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=316909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:40:44,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=316925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:40:47,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=316928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:41:07,424 INFO [train.py:968] (1/2) Epoch 7, batch 43300, giga_loss[loss=0.3682, simple_loss=0.4112, pruned_loss=0.1626, over 27605.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3893, pruned_loss=0.1403, over 5660909.99 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3844, pruned_loss=0.1358, over 5674203.16 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3897, pruned_loss=0.1405, over 5660304.51 frames. ], batch size: 472, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:41:12,031 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:41:52,410 INFO [train.py:968] (1/2) Epoch 7, batch 43350, giga_loss[loss=0.2976, simple_loss=0.3679, pruned_loss=0.1137, over 28995.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3867, pruned_loss=0.1366, over 5662530.04 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3841, pruned_loss=0.1355, over 5678218.19 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3873, pruned_loss=0.1371, over 5658351.01 frames. ], batch size: 213, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:42:36,429 INFO [train.py:968] (1/2) Epoch 7, batch 43400, giga_loss[loss=0.3711, simple_loss=0.4086, pruned_loss=0.1668, over 28622.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3861, pruned_loss=0.1365, over 5662332.79 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3847, pruned_loss=0.1358, over 5681665.47 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3861, pruned_loss=0.1366, over 5655627.85 frames. ], batch size: 307, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:43:02,088 INFO [optim.py:369] (1/2) 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 00:43:11,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9923, 1.1779, 0.8183, 0.8700], device='cuda:1'), covar=tensor([0.0900, 0.0566, 0.1570, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0447, 0.0499, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 00:43:24,719 INFO [train.py:968] (1/2) Epoch 7, batch 43450, giga_loss[loss=0.3069, simple_loss=0.367, pruned_loss=0.1234, over 28813.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3845, pruned_loss=0.1359, over 5672420.63 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3841, pruned_loss=0.1355, over 5685061.73 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3851, pruned_loss=0.1363, over 5664055.78 frames. ], batch size: 99, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:44:13,396 INFO [train.py:968] (1/2) Epoch 7, batch 43500, giga_loss[loss=0.3333, simple_loss=0.3853, pruned_loss=0.1406, over 28653.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3841, pruned_loss=0.1358, over 5673649.03 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.384, pruned_loss=0.1354, over 5685206.90 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3846, pruned_loss=0.1363, over 5666771.99 frames. ], batch size: 92, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:44:25,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-04 00:44:39,729 INFO [optim.py:369] (1/2) 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 00:45:00,229 INFO [train.py:968] (1/2) Epoch 7, batch 43550, giga_loss[loss=0.3708, simple_loss=0.4356, pruned_loss=0.1529, over 28542.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3893, pruned_loss=0.1394, over 5669744.65 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3841, pruned_loss=0.1355, over 5689432.25 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3897, pruned_loss=0.1397, over 5660282.90 frames. ], batch size: 336, lr: 4.46e-03, grad_scale: 2.0
+2023-03-04 00:45:36,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7824, 1.6740, 1.5917, 1.4379], device='cuda:1'), covar=tensor([0.1136, 0.1914, 0.1750, 0.1893], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0732, 0.0650, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 00:45:41,294 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-04 00:45:47,307 INFO [train.py:968] (1/2) Epoch 7, batch 43600, giga_loss[loss=0.2943, simple_loss=0.3733, pruned_loss=0.1076, over 28687.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.392, pruned_loss=0.1382, over 5674294.74 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3844, pruned_loss=0.1356, over 5694550.50 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3923, pruned_loss=0.1384, over 5661639.97 frames. ], batch size: 92, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:46:15,218 INFO [optim.py:369] (1/2) 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 00:46:17,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=317284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:46:32,196 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:46:35,774 INFO [train.py:968] (1/2) Epoch 7, batch 43650, giga_loss[loss=0.3724, simple_loss=0.4218, pruned_loss=0.1616, over 29044.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3916, pruned_loss=0.1367, over 5682604.48 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3833, pruned_loss=0.135, over 5702180.88 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.393, pruned_loss=0.1375, over 5664907.74 frames. ], batch size: 128, lr: 4.46e-03, grad_scale: 4.0
+2023-03-04 00:47:21,505 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 00:47:22,589 INFO [train.py:968] (1/2) Epoch 7, batch 43700, giga_loss[loss=0.3504, simple_loss=0.4061, pruned_loss=0.1474, over 27975.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.3946, pruned_loss=0.1394, over 5677904.28 frames. ], libri_tot_loss[loss=0.327, simple_loss=0.3835, pruned_loss=0.1352, over 5706863.00 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3958, pruned_loss=0.1399, over 5658792.61 frames. ], batch size: 412, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:47:51,793 INFO [optim.py:369] (1/2) 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 00:48:08,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:48:10,077 INFO [train.py:968] (1/2) Epoch 7, batch 43750, giga_loss[loss=0.2953, simple_loss=0.3666, pruned_loss=0.112, over 28826.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3944, pruned_loss=0.1393, over 5680359.46 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3837, pruned_loss=0.1353, over 5707730.46 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3954, pruned_loss=0.1396, over 5663679.58 frames. ], batch size: 112, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:48:35,605 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=317427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:48:37,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=317430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:48:53,485 INFO [train.py:968] (1/2) Epoch 7, batch 43800, giga_loss[loss=0.3212, simple_loss=0.3847, pruned_loss=0.1289, over 28978.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.395, pruned_loss=0.1407, over 5681260.08 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3837, pruned_loss=0.1352, over 5711766.61 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.396, pruned_loss=0.1412, over 5663675.73 frames. ], batch size: 174, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:49:02,146 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:49:37,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5447, 1.7903, 1.9264, 1.4669], device='cuda:1'), covar=tensor([0.1522, 0.1899, 0.1135, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0721, 0.0817, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 00:49:37,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6314, 1.7019, 1.4265, 1.8263], device='cuda:1'), covar=tensor([0.2157, 0.2191, 0.2290, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.1195, 0.0907, 0.1052, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:49:39,691 INFO [train.py:968] (1/2) Epoch 7, batch 43850, libri_loss[loss=0.3437, simple_loss=0.3991, pruned_loss=0.1442, over 29534.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3926, pruned_loss=0.14, over 5671590.75 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3836, pruned_loss=0.1353, over 5709057.59 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3938, pruned_loss=0.1405, over 5659111.51 frames. ], batch size: 81, lr: 4.45e-03, grad_scale: 2.0
+2023-03-04 00:49:49,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4078, 1.6323, 1.3661, 1.6197], device='cuda:1'), covar=tensor([0.1800, 0.1716, 0.1671, 0.1678], device='cuda:1'), in_proj_covar=tensor([0.1195, 0.0907, 0.1051, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:50:26,804 INFO [train.py:968] (1/2) Epoch 7, batch 43900, giga_loss[loss=0.3461, simple_loss=0.4098, pruned_loss=0.1412, over 28895.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3905, pruned_loss=0.1396, over 5673664.39 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3831, pruned_loss=0.1351, over 5712200.76 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3921, pruned_loss=0.1403, over 5660215.22 frames. ], batch size: 145, lr: 4.45e-03, grad_scale: 2.0
+2023-03-04 00:50:44,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5392, 1.6349, 1.4491, 1.3929], device='cuda:1'), covar=tensor([0.1865, 0.1438, 0.1182, 0.1429], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1419, 0.1380, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 00:50:57,987 INFO [optim.py:369] (1/2) 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 00:51:06,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4468, 1.4253, 0.9605, 1.1743], device='cuda:1'), covar=tensor([0.0900, 0.0860, 0.1475, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0448, 0.0502, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 00:51:15,598 INFO [train.py:968] (1/2) Epoch 7, batch 43950, giga_loss[loss=0.4151, simple_loss=0.4354, pruned_loss=0.1974, over 28504.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3893, pruned_loss=0.1394, over 5679214.22 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3831, pruned_loss=0.1353, over 5714306.98 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3907, pruned_loss=0.1399, over 5666255.55 frames. ], batch size: 336, lr: 4.45e-03, grad_scale: 2.0
+2023-03-04 00:52:02,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317643.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 00:52:08,650 INFO [train.py:968] (1/2) Epoch 7, batch 44000, giga_loss[loss=0.3554, simple_loss=0.4057, pruned_loss=0.1526, over 28677.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3908, pruned_loss=0.1412, over 5677461.00 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.383, pruned_loss=0.1353, over 5708830.94 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3922, pruned_loss=0.1416, over 5670720.66 frames. ], batch size: 262, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:52:12,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3407, 1.5933, 1.3200, 1.5887], device='cuda:1'), covar=tensor([0.1841, 0.1778, 0.1827, 0.1618], device='cuda:1'), in_proj_covar=tensor([0.1196, 0.0906, 0.1055, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:52:30,382 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:52:49,855 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=317691.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 00:52:59,743 INFO [train.py:968] (1/2) Epoch 7, batch 44050, giga_loss[loss=0.341, simple_loss=0.3969, pruned_loss=0.1426, over 28688.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3913, pruned_loss=0.1422, over 5670356.39 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.383, pruned_loss=0.1352, over 5711719.20 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3924, pruned_loss=0.1427, over 5662200.88 frames. ], batch size: 284, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:53:27,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6852, 1.6087, 1.2768, 1.3388], device='cuda:1'), covar=tensor([0.0711, 0.0618, 0.0999, 0.0954], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0448, 0.0499, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 00:53:40,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9034, 1.7996, 1.3439, 1.4660], device='cuda:1'), covar=tensor([0.0626, 0.0546, 0.0938, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0448, 0.0499, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 00:53:40,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 00:53:41,764 INFO [train.py:968] (1/2) Epoch 7, batch 44100, giga_loss[loss=0.2947, simple_loss=0.3496, pruned_loss=0.1199, over 28522.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3879, pruned_loss=0.14, over 5678339.43 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3823, pruned_loss=0.1345, over 5718401.72 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3897, pruned_loss=0.1412, over 5664551.85 frames. ], batch size: 71, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:53:44,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3915, 1.9440, 1.3925, 1.7350], device='cuda:1'), covar=tensor([0.0723, 0.0264, 0.0300, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0117, 0.0118, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-04 00:54:01,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=317772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:54:09,303 INFO [optim.py:369] (1/2) 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 00:54:16,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2560, 1.3714, 1.1775, 1.5281], device='cuda:1'), covar=tensor([0.0692, 0.0303, 0.0314, 0.0730], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0117, 0.0118, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-04 00:54:27,087 INFO [train.py:968] (1/2) Epoch 7, batch 44150, giga_loss[loss=0.361, simple_loss=0.4023, pruned_loss=0.1598, over 27590.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3864, pruned_loss=0.1384, over 5678728.22 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3824, pruned_loss=0.1345, over 5719800.83 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3878, pruned_loss=0.1394, over 5665699.32 frames. ], batch size: 472, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:54:37,439 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=317813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:54:40,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=317816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:55:10,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=317845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:55:14,533 INFO [train.py:968] (1/2) Epoch 7, batch 44200, giga_loss[loss=0.3775, simple_loss=0.4172, pruned_loss=0.1689, over 28921.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3875, pruned_loss=0.1382, over 5677454.14 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.382, pruned_loss=0.1342, over 5724167.41 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3891, pruned_loss=0.1393, over 5662181.81 frames. ], batch size: 106, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:55:45,943 INFO [optim.py:369] (1/2) 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 00:56:03,981 INFO [train.py:968] (1/2) Epoch 7, batch 44250, giga_loss[loss=0.3385, simple_loss=0.3937, pruned_loss=0.1416, over 28891.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3879, pruned_loss=0.138, over 5682284.47 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3817, pruned_loss=0.134, over 5725102.13 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3895, pruned_loss=0.1392, over 5669076.60 frames. ], batch size: 227, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:56:17,348 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=317915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:56:21,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=317918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:56:51,183 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=317947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 00:56:54,385 INFO [train.py:968] (1/2) Epoch 7, batch 44300, giga_loss[loss=0.3711, simple_loss=0.4144, pruned_loss=0.1639, over 28309.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3885, pruned_loss=0.1388, over 5673250.38 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3822, pruned_loss=0.1343, over 5726763.60 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3893, pruned_loss=0.1395, over 5660917.85 frames. ], batch size: 368, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:57:23,618 INFO [optim.py:369] (1/2) 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 00:57:38,686 INFO [train.py:968] (1/2) Epoch 7, batch 44350, giga_loss[loss=0.2738, simple_loss=0.3627, pruned_loss=0.09247, over 28874.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3899, pruned_loss=0.1369, over 5674424.32 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3825, pruned_loss=0.1344, over 5725387.18 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3905, pruned_loss=0.1375, over 5664472.71 frames. ], batch size: 106, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:57:46,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2423, 3.0251, 1.3160, 1.3704], device='cuda:1'), covar=tensor([0.0946, 0.0378, 0.0888, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0497, 0.0319, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 00:57:53,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=318018.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 00:58:21,440 INFO [train.py:968] (1/2) Epoch 7, batch 44400, giga_loss[loss=0.3002, simple_loss=0.3819, pruned_loss=0.1092, over 28929.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.392, pruned_loss=0.1365, over 5678843.26 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3829, pruned_loss=0.1349, over 5719637.42 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3924, pruned_loss=0.1365, over 5674239.64 frames. ], batch size: 174, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:58:23,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0425, 2.0265, 1.9974, 1.8368], device='cuda:1'), covar=tensor([0.0902, 0.1285, 0.1198, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0729, 0.0650, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 00:58:37,110 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:58:57,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0272, 1.0357, 3.4262, 3.0118], device='cuda:1'), covar=tensor([0.1527, 0.2443, 0.0494, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0605, 0.0560, 0.0809, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 00:59:08,347 INFO [train.py:968] (1/2) Epoch 7, batch 44450, giga_loss[loss=0.3131, simple_loss=0.378, pruned_loss=0.1241, over 28797.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.395, pruned_loss=0.1379, over 5676993.84 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3828, pruned_loss=0.1349, over 5710865.97 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3957, pruned_loss=0.1381, over 5680160.94 frames. ], batch size: 99, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 00:59:46,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3100, 3.1256, 2.9891, 1.4482], device='cuda:1'), covar=tensor([0.0773, 0.0947, 0.0891, 0.2101], device='cuda:1'), in_proj_covar=tensor([0.0972, 0.0925, 0.0811, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 00:59:58,190 INFO [train.py:968] (1/2) Epoch 7, batch 44500, giga_loss[loss=0.3599, simple_loss=0.4069, pruned_loss=0.1565, over 28596.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3984, pruned_loss=0.1418, over 5674522.25 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3826, pruned_loss=0.1347, over 5712030.17 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3993, pruned_loss=0.1421, over 5675434.61 frames. ], batch size: 92, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:00:08,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3357, 3.1469, 3.0313, 1.4938], device='cuda:1'), covar=tensor([0.0798, 0.0962, 0.0890, 0.2213], device='cuda:1'), in_proj_covar=tensor([0.0970, 0.0924, 0.0813, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 01:00:09,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318161.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:00:13,545 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:00:35,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:00:37,162 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 01:00:41,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=318193.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:00:46,981 INFO [train.py:968] (1/2) Epoch 7, batch 44550, giga_loss[loss=0.442, simple_loss=0.4587, pruned_loss=0.2127, over 27516.00 frames. ], tot_loss[loss=0.3444, simple_loss=0.4, pruned_loss=0.1444, over 5647153.20 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3828, pruned_loss=0.1348, over 5705483.27 frames. ], giga_tot_loss[loss=0.3451, simple_loss=0.4008, pruned_loss=0.1447, over 5653585.70 frames. ], batch size: 472, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:00:54,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318209.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:00:56,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=318212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:01:14,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6515, 1.6411, 1.6719, 1.5519], device='cuda:1'), covar=tensor([0.1158, 0.1702, 0.1551, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0728, 0.0649, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:01:23,725 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=318241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:01:32,315 INFO [train.py:968] (1/2) Epoch 7, batch 44600, giga_loss[loss=0.2961, simple_loss=0.3689, pruned_loss=0.1116, over 28937.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3992, pruned_loss=0.1442, over 5658724.62 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3827, pruned_loss=0.1348, over 5710537.83 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.4004, pruned_loss=0.1447, over 5658373.17 frames. ], batch size: 174, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:01:54,503 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:01:59,003 INFO [optim.py:369] (1/2) 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 01:02:14,936 INFO [train.py:968] (1/2) Epoch 7, batch 44650, giga_loss[loss=0.3073, simple_loss=0.3736, pruned_loss=0.1205, over 28939.00 frames. ], tot_loss[loss=0.341, simple_loss=0.3973, pruned_loss=0.1423, over 5663133.75 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3828, pruned_loss=0.1348, over 5714999.56 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3986, pruned_loss=0.1429, over 5657769.53 frames. ], batch size: 136, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:02:27,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0937, 1.5185, 1.4600, 1.0698], device='cuda:1'), covar=tensor([0.1663, 0.2310, 0.1395, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0721, 0.0818, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 01:02:56,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0621, 1.3268, 1.2985, 1.2397], device='cuda:1'), covar=tensor([0.1077, 0.0946, 0.1526, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0724, 0.0645, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:03:01,004 INFO [train.py:968] (1/2) Epoch 7, batch 44700, libri_loss[loss=0.2747, simple_loss=0.3423, pruned_loss=0.1035, over 29574.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3969, pruned_loss=0.1396, over 5671199.54 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3824, pruned_loss=0.1346, over 5718140.33 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3984, pruned_loss=0.1404, over 5663342.65 frames. ], batch size: 75, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:03:23,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3442, 1.6595, 1.2048, 1.1608], device='cuda:1'), covar=tensor([0.1450, 0.1238, 0.1159, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.1587, 0.1416, 0.1393, 0.1496], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 01:03:29,828 INFO [optim.py:369] (1/2) 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 01:03:32,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4278, 1.6918, 1.7371, 1.3496], device='cuda:1'), covar=tensor([0.1579, 0.2044, 0.1224, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0721, 0.0821, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 01:03:46,156 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 01:03:46,388 INFO [train.py:968] (1/2) Epoch 7, batch 44750, giga_loss[loss=0.2911, simple_loss=0.3675, pruned_loss=0.1074, over 28307.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.398, pruned_loss=0.1401, over 5670848.88 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3825, pruned_loss=0.1347, over 5721074.66 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3995, pruned_loss=0.1407, over 5661049.60 frames. ], batch size: 65, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:04:36,787 INFO [train.py:968] (1/2) Epoch 7, batch 44800, giga_loss[loss=0.3657, simple_loss=0.4175, pruned_loss=0.157, over 29010.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3981, pruned_loss=0.1408, over 5676362.62 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3826, pruned_loss=0.1348, over 5724335.24 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3995, pruned_loss=0.1414, over 5664647.71 frames. ], batch size: 128, lr: 4.45e-03, grad_scale: 8.0
+2023-03-04 01:04:39,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-04 01:05:06,407 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 7, batch 44850, giga_loss[loss=0.3839, simple_loss=0.4272, pruned_loss=0.1703, over 28195.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3974, pruned_loss=0.1405, over 5678478.45 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3832, pruned_loss=0.1349, over 5717345.83 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3985, pruned_loss=0.1411, over 5673257.08 frames. ], batch size: 368, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:06:04,563 INFO [train.py:968] (1/2) Epoch 7, batch 44900, giga_loss[loss=0.3429, simple_loss=0.3996, pruned_loss=0.1431, over 29022.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3959, pruned_loss=0.1413, over 5667606.06 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3827, pruned_loss=0.1347, over 5724638.51 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3978, pruned_loss=0.1421, over 5655311.70 frames. ], batch size: 155, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:06:15,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=318560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:06:26,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 01:06:39,595 INFO [optim.py:369] (1/2) 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 01:06:51,453 INFO [train.py:968] (1/2) Epoch 7, batch 44950, giga_loss[loss=0.3724, simple_loss=0.4169, pruned_loss=0.1639, over 27471.00 frames. ], tot_loss[loss=0.3391, simple_loss=0.3948, pruned_loss=0.1417, over 5654218.12 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3831, pruned_loss=0.135, over 5716925.96 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3961, pruned_loss=0.1422, over 5650912.25 frames. ], batch size: 472, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:07:17,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4255, 1.6009, 1.4819, 1.1765], device='cuda:1'), covar=tensor([0.1571, 0.1403, 0.0901, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.1605, 0.1425, 0.1416, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 01:07:40,244 INFO [train.py:968] (1/2) Epoch 7, batch 45000, giga_loss[loss=0.2949, simple_loss=0.3665, pruned_loss=0.1117, over 28819.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3915, pruned_loss=0.1399, over 5630694.40 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3835, pruned_loss=0.1354, over 5690916.39 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3923, pruned_loss=0.14, over 5650422.74 frames. ], batch size: 174, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:07:40,244 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 01:07:49,542 INFO [train.py:1012] (1/2) Epoch 7, validation: loss=0.2285, simple_loss=0.3357, pruned_loss=0.06063, over 944034.00 frames. 
+2023-03-04 01:07:49,543 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 01:07:51,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=318652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:07:55,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5815, 1.9042, 1.9318, 1.4770], device='cuda:1'), covar=tensor([0.1615, 0.1906, 0.1171, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0718, 0.0817, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 01:08:16,373 INFO [optim.py:369] (1/2) 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 01:08:29,602 INFO [train.py:968] (1/2) Epoch 7, batch 45050, giga_loss[loss=0.3507, simple_loss=0.3978, pruned_loss=0.1518, over 28633.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3903, pruned_loss=0.1397, over 5584204.61 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3846, pruned_loss=0.1363, over 5621223.57 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3902, pruned_loss=0.1391, over 5658219.53 frames. ], batch size: 307, lr: 4.45e-03, grad_scale: 4.0
+2023-03-04 01:08:33,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:08:33,907 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:08:35,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=318706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:08:47,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3056, 2.9464, 1.4457, 1.3587], device='cuda:1'), covar=tensor([0.0913, 0.0347, 0.0844, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0498, 0.0321, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 01:09:02,777 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=318735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:09:11,487 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:09:12,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4018, 2.0901, 1.5436, 0.5966], device='cuda:1'), covar=tensor([0.3252, 0.1619, 0.2333, 0.3723], device='cuda:1'), in_proj_covar=tensor([0.1477, 0.1396, 0.1446, 0.1212], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 01:09:15,595 INFO [train.py:968] (1/2) Epoch 7, batch 45100, giga_loss[loss=0.2716, simple_loss=0.3453, pruned_loss=0.09892, over 29058.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3894, pruned_loss=0.1387, over 5558507.58 frames. ], libri_tot_loss[loss=0.3295, simple_loss=0.3853, pruned_loss=0.1369, over 5568924.46 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3887, pruned_loss=0.1378, over 5663708.84 frames. ], batch size: 155, lr: 4.44e-03, grad_scale: 2.0
+2023-03-04 01:09:26,609 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=318760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:09:43,859 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-04 01:10:23,104 INFO [optim.py:369] (1/2) 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 01:10:32,101 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=318795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:10:34,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=318798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:10:59,342 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 50, giga_loss[loss=0.3139, simple_loss=0.3901, pruned_loss=0.1188, over 28933.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3844, pruned_loss=0.1192, over 1262219.23 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3696, pruned_loss=0.1099, over 223508.51 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3873, pruned_loss=0.1209, over 1081430.82 frames. ], batch size: 213, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:11:50,667 INFO [train.py:968] (1/2) Epoch 8, batch 100, giga_loss[loss=0.2988, simple_loss=0.3681, pruned_loss=0.1147, over 27581.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3787, pruned_loss=0.1173, over 2235017.78 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3634, pruned_loss=0.1054, over 360463.67 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3813, pruned_loss=0.1192, over 2001648.86 frames. ], batch size: 472, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:11:54,581 INFO [optim.py:369] (1/2) 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 01:12:25,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1825, 3.9697, 3.7231, 1.9540], device='cuda:1'), covar=tensor([0.0493, 0.0715, 0.0747, 0.2036], device='cuda:1'), in_proj_covar=tensor([0.0960, 0.0914, 0.0807, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 01:12:31,824 INFO [train.py:968] (1/2) Epoch 8, batch 150, giga_loss[loss=0.257, simple_loss=0.323, pruned_loss=0.09549, over 27681.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3639, pruned_loss=0.11, over 3012110.09 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3619, pruned_loss=0.1059, over 578370.12 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3649, pruned_loss=0.111, over 2705641.19 frames. ], batch size: 472, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:13:15,579 INFO [train.py:968] (1/2) Epoch 8, batch 200, libri_loss[loss=0.2481, simple_loss=0.3334, pruned_loss=0.08145, over 29592.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3504, pruned_loss=0.1031, over 3610358.51 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3599, pruned_loss=0.104, over 760760.41 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3501, pruned_loss=0.1036, over 3283243.17 frames. ], batch size: 74, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:13:19,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5053, 4.2953, 4.0526, 1.9507], device='cuda:1'), covar=tensor([0.0433, 0.0680, 0.0705, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.0963, 0.0917, 0.0811, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 01:13:19,399 INFO [optim.py:369] (1/2) 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 01:13:41,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5357, 1.8004, 1.2743, 1.5981], device='cuda:1'), covar=tensor([0.0726, 0.0295, 0.0338, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0116, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0075], device='cuda:1')
+2023-03-04 01:13:54,318 INFO [train.py:968] (1/2) Epoch 8, batch 250, giga_loss[loss=0.2231, simple_loss=0.3014, pruned_loss=0.0724, over 28755.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3397, pruned_loss=0.09721, over 4078068.68 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3583, pruned_loss=0.1022, over 933546.74 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3383, pruned_loss=0.09734, over 3754088.15 frames. ], batch size: 60, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:14:15,200 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-04 01:14:38,966 INFO [train.py:968] (1/2) Epoch 8, batch 300, libri_loss[loss=0.2641, simple_loss=0.3514, pruned_loss=0.08843, over 29501.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3306, pruned_loss=0.09381, over 4431286.60 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3589, pruned_loss=0.1027, over 1008011.63 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3285, pruned_loss=0.09346, over 4152977.12 frames. ], batch size: 85, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:14:39,236 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:15:16,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:15:21,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=319125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:15:24,945 INFO [train.py:968] (1/2) Epoch 8, batch 350, giga_loss[loss=0.2285, simple_loss=0.3028, pruned_loss=0.07713, over 28791.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3227, pruned_loss=0.09, over 4703693.49 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3581, pruned_loss=0.1019, over 1104003.37 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3201, pruned_loss=0.08946, over 4460057.94 frames. ], batch size: 284, lr: 4.19e-03, grad_scale: 2.0
+2023-03-04 01:15:30,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:16:03,270 INFO [train.py:968] (1/2) Epoch 8, batch 400, giga_loss[loss=0.1992, simple_loss=0.2735, pruned_loss=0.06239, over 28568.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3185, pruned_loss=0.08758, over 4930194.12 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3608, pruned_loss=0.1043, over 1246030.96 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3145, pruned_loss=0.08607, over 4706161.31 frames. ], batch size: 60, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:16:07,634 INFO [optim.py:369] (1/2) 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 01:16:21,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0979, 1.2470, 1.0955, 0.9258], device='cuda:1'), covar=tensor([0.1012, 0.1216, 0.0735, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1405, 0.1401, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 01:16:39,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319222.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:16:42,588 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319225.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:16:45,248 INFO [train.py:968] (1/2) Epoch 8, batch 450, libri_loss[loss=0.3077, simple_loss=0.3765, pruned_loss=0.1194, over 29518.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3161, pruned_loss=0.08661, over 5104236.67 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3611, pruned_loss=0.1047, over 1338123.91 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3119, pruned_loss=0.08493, over 4910787.43 frames. ], batch size: 84, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:17:09,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:17:15,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:18,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:28,205 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:28,550 INFO [train.py:968] (1/2) Epoch 8, batch 500, giga_loss[loss=0.2178, simple_loss=0.2902, pruned_loss=0.07273, over 28971.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3128, pruned_loss=0.08509, over 5233151.59 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3602, pruned_loss=0.1048, over 1451428.97 frames. ], giga_tot_loss[loss=0.2374, simple_loss=0.3084, pruned_loss=0.08321, over 5061829.86 frames. ], batch size: 128, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:17:31,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9094, 2.0072, 1.8554, 1.9508], device='cuda:1'), covar=tensor([0.1399, 0.1810, 0.1826, 0.1555], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0729, 0.0648, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:17:31,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:33,623 INFO [optim.py:369] (1/2) 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 01:17:44,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:17:55,498 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:18:12,613 INFO [train.py:968] (1/2) Epoch 8, batch 550, giga_loss[loss=0.2358, simple_loss=0.309, pruned_loss=0.0813, over 28855.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.3121, pruned_loss=0.08507, over 5334792.19 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.36, pruned_loss=0.1045, over 1580172.62 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.3073, pruned_loss=0.0831, over 5183722.27 frames. ], batch size: 145, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:18:20,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3015, 1.0661, 4.8107, 3.4442], device='cuda:1'), covar=tensor([0.1635, 0.2654, 0.0298, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0554, 0.0805, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 01:18:25,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0011, 3.7973, 3.5604, 1.8677], device='cuda:1'), covar=tensor([0.0518, 0.0695, 0.0680, 0.2083], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0902, 0.0796, 0.0622], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 01:19:00,179 INFO [train.py:968] (1/2) Epoch 8, batch 600, giga_loss[loss=0.2239, simple_loss=0.297, pruned_loss=0.07539, over 28686.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3087, pruned_loss=0.08349, over 5409554.83 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3599, pruned_loss=0.1045, over 1619682.50 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3044, pruned_loss=0.08172, over 5289471.93 frames. ], batch size: 242, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:19:04,075 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 650, giga_loss[loss=0.2337, simple_loss=0.304, pruned_loss=0.08165, over 28911.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3055, pruned_loss=0.08183, over 5476176.03 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3596, pruned_loss=0.1045, over 1662393.83 frames. ], giga_tot_loss[loss=0.2311, simple_loss=0.3017, pruned_loss=0.08023, over 5377806.11 frames. ], batch size: 174, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:20:23,453 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 700, giga_loss[loss=0.2077, simple_loss=0.279, pruned_loss=0.06819, over 28989.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3039, pruned_loss=0.0806, over 5528327.86 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3598, pruned_loss=0.1046, over 1807597.05 frames. ], giga_tot_loss[loss=0.2279, simple_loss=0.2989, pruned_loss=0.07848, over 5437565.45 frames. ], batch size: 227, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:20:38,347 INFO [optim.py:369] (1/2) 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 01:20:50,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:21:10,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3110, 1.4959, 1.5408, 1.5047], device='cuda:1'), covar=tensor([0.1191, 0.1089, 0.1411, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0727, 0.0647, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:21:15,879 INFO [train.py:968] (1/2) Epoch 8, batch 750, giga_loss[loss=0.2013, simple_loss=0.2739, pruned_loss=0.06432, over 28589.00 frames. ], tot_loss[loss=0.23, simple_loss=0.3015, pruned_loss=0.07927, over 5558952.24 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.36, pruned_loss=0.1048, over 1905256.42 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.2962, pruned_loss=0.07693, over 5481778.71 frames. ], batch size: 307, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:21:36,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=319551.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:21:39,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8123, 1.6487, 1.8903, 1.5290], device='cuda:1'), covar=tensor([0.2015, 0.3428, 0.1584, 0.1754], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0729, 0.0838, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 01:21:40,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4100, 2.1497, 1.4277, 1.6998], device='cuda:1'), covar=tensor([0.0753, 0.0248, 0.0312, 0.0775], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-04 01:21:59,322 INFO [train.py:968] (1/2) Epoch 8, batch 800, giga_loss[loss=0.191, simple_loss=0.2646, pruned_loss=0.05868, over 29006.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3008, pruned_loss=0.07914, over 5592691.15 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3588, pruned_loss=0.1037, over 2042882.41 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.2949, pruned_loss=0.07678, over 5520289.01 frames. ], batch size: 136, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:22:03,600 INFO [optim.py:369] (1/2) 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 01:22:44,314 INFO [train.py:968] (1/2) Epoch 8, batch 850, giga_loss[loss=0.2758, simple_loss=0.356, pruned_loss=0.09781, over 29084.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3102, pruned_loss=0.08458, over 5615511.90 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3605, pruned_loss=0.1045, over 2175260.57 frames. ], giga_tot_loss[loss=0.2334, simple_loss=0.3033, pruned_loss=0.08173, over 5548553.32 frames. ], batch size: 155, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:22:51,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4453, 1.9688, 1.4008, 1.5478], device='cuda:1'), covar=tensor([0.0761, 0.0270, 0.0326, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0075], device='cuda:1')
+2023-03-04 01:22:59,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:23:02,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:23:14,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3068, 1.5672, 0.9347, 1.2699], device='cuda:1'), covar=tensor([0.1050, 0.0830, 0.1720, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0441, 0.0498, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 01:23:29,316 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=319675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:23:33,255 INFO [train.py:968] (1/2) Epoch 8, batch 900, giga_loss[loss=0.2926, simple_loss=0.3643, pruned_loss=0.1105, over 29023.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3233, pruned_loss=0.09142, over 5625906.65 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3597, pruned_loss=0.1039, over 2239781.04 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3174, pruned_loss=0.08916, over 5576593.89 frames. ], batch size: 128, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:23:39,089 INFO [optim.py:369] (1/2) 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 01:24:05,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6945, 1.8032, 1.6676, 1.6236], device='cuda:1'), covar=tensor([0.1389, 0.1750, 0.1775, 0.1695], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0724, 0.0643, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:24:14,257 INFO [train.py:968] (1/2) Epoch 8, batch 950, giga_loss[loss=0.3309, simple_loss=0.398, pruned_loss=0.1319, over 27955.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3356, pruned_loss=0.09816, over 5645657.73 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3592, pruned_loss=0.1037, over 2348109.05 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3304, pruned_loss=0.09623, over 5599315.17 frames. ], batch size: 412, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:24:59,269 INFO [train.py:968] (1/2) Epoch 8, batch 1000, giga_loss[loss=0.2837, simple_loss=0.3559, pruned_loss=0.1058, over 28678.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3449, pruned_loss=0.1026, over 5658400.95 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3594, pruned_loss=0.1039, over 2436249.51 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3402, pruned_loss=0.1009, over 5615682.66 frames. ], batch size: 85, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:25:05,505 INFO [optim.py:369] (1/2) 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 01:25:38,869 INFO [train.py:968] (1/2) Epoch 8, batch 1050, giga_loss[loss=0.2638, simple_loss=0.3475, pruned_loss=0.09009, over 28866.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3488, pruned_loss=0.1031, over 5667386.62 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3595, pruned_loss=0.104, over 2453574.26 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3451, pruned_loss=0.1018, over 5632900.37 frames. ], batch size: 174, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:25:54,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:26:17,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4297, 1.7674, 1.7660, 1.3114], device='cuda:1'), covar=tensor([0.1599, 0.2044, 0.1259, 0.1480], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0716, 0.0827, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 01:26:28,183 INFO [train.py:968] (1/2) Epoch 8, batch 1100, giga_loss[loss=0.2675, simple_loss=0.3506, pruned_loss=0.09222, over 29025.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3501, pruned_loss=0.1025, over 5665101.49 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3599, pruned_loss=0.1039, over 2521638.00 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3468, pruned_loss=0.1015, over 5633611.77 frames. ], batch size: 128, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:26:29,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4234, 1.5533, 1.3054, 1.7747], device='cuda:1'), covar=tensor([0.2236, 0.2183, 0.2339, 0.1933], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.0912, 0.1063, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 01:26:32,157 INFO [optim.py:369] (1/2) 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 01:27:08,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=319926.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:27:10,478 INFO [train.py:968] (1/2) Epoch 8, batch 1150, giga_loss[loss=0.2667, simple_loss=0.3435, pruned_loss=0.09493, over 28901.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3531, pruned_loss=0.1047, over 5663423.55 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3604, pruned_loss=0.104, over 2568233.91 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3502, pruned_loss=0.1038, over 5639072.25 frames. ], batch size: 136, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:27:21,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 01:27:53,623 INFO [train.py:968] (1/2) Epoch 8, batch 1200, giga_loss[loss=0.4203, simple_loss=0.4389, pruned_loss=0.2008, over 26564.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3556, pruned_loss=0.1067, over 5664883.31 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3615, pruned_loss=0.1048, over 2639500.17 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3527, pruned_loss=0.1057, over 5649873.89 frames. ], batch size: 555, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:27:59,314 INFO [optim.py:369] (1/2) 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 01:28:01,159 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=319987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:28:04,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=319990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:28:29,377 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 1250, giga_loss[loss=0.3229, simple_loss=0.3852, pruned_loss=0.1303, over 28301.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3582, pruned_loss=0.1083, over 5675318.15 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3612, pruned_loss=0.1044, over 2704479.81 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.356, pruned_loss=0.1077, over 5659065.19 frames. ], batch size: 368, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:28:54,547 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-04 01:29:12,145 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320069.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:29:14,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=320072.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:29:20,663 INFO [train.py:968] (1/2) Epoch 8, batch 1300, giga_loss[loss=0.2781, simple_loss=0.3552, pruned_loss=0.1005, over 28526.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3615, pruned_loss=0.1092, over 5684421.77 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3613, pruned_loss=0.1042, over 2796423.59 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3596, pruned_loss=0.1089, over 5668065.20 frames. ], batch size: 85, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:29:26,461 INFO [optim.py:369] (1/2) 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 01:29:34,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5398, 1.9072, 1.3278, 1.8182], device='cuda:1'), covar=tensor([0.0694, 0.0244, 0.0301, 0.0753], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0069, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-04 01:29:38,885 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=320101.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:30:01,685 INFO [train.py:968] (1/2) Epoch 8, batch 1350, giga_loss[loss=0.2578, simple_loss=0.341, pruned_loss=0.08733, over 28880.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3636, pruned_loss=0.11, over 5677942.52 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3607, pruned_loss=0.1038, over 2862204.98 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3624, pruned_loss=0.1101, over 5670395.02 frames. ], batch size: 112, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:30:24,877 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 1400, giga_loss[loss=0.264, simple_loss=0.3486, pruned_loss=0.08967, over 28580.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3637, pruned_loss=0.1094, over 5687024.21 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3593, pruned_loss=0.1032, over 2952177.21 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3635, pruned_loss=0.11, over 5675109.80 frames. ], batch size: 78, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:30:47,452 INFO [optim.py:369] (1/2) 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 01:31:17,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 01:31:27,165 INFO [train.py:968] (1/2) Epoch 8, batch 1450, giga_loss[loss=0.2676, simple_loss=0.3495, pruned_loss=0.09282, over 28916.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3626, pruned_loss=0.1074, over 5689259.13 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3594, pruned_loss=0.103, over 2981904.89 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3625, pruned_loss=0.108, over 5677933.31 frames. ], batch size: 213, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:31:33,597 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:32:06,108 INFO [train.py:968] (1/2) Epoch 8, batch 1500, giga_loss[loss=0.2507, simple_loss=0.343, pruned_loss=0.07921, over 28923.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3611, pruned_loss=0.1052, over 5700081.43 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3595, pruned_loss=0.1031, over 3025677.23 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3609, pruned_loss=0.1056, over 5688111.65 frames. ], batch size: 164, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:32:11,981 INFO [optim.py:369] (1/2) 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 01:32:42,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9603, 1.9725, 1.2983, 1.5607], device='cuda:1'), covar=tensor([0.0683, 0.0632, 0.1007, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0435, 0.0493, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 01:32:45,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6261, 1.9308, 1.9029, 1.5203], device='cuda:1'), covar=tensor([0.1505, 0.1787, 0.1124, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0711, 0.0825, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 01:32:46,073 INFO [train.py:968] (1/2) Epoch 8, batch 1550, giga_loss[loss=0.3095, simple_loss=0.3741, pruned_loss=0.1225, over 28158.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3594, pruned_loss=0.1039, over 5709632.99 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3584, pruned_loss=0.1027, over 3110488.51 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3598, pruned_loss=0.1045, over 5695070.31 frames. ], batch size: 77, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:33:28,024 INFO [train.py:968] (1/2) Epoch 8, batch 1600, giga_loss[loss=0.3056, simple_loss=0.3742, pruned_loss=0.1185, over 29079.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3605, pruned_loss=0.1055, over 5696477.59 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3579, pruned_loss=0.1023, over 3224722.03 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3611, pruned_loss=0.1063, over 5684099.97 frames. ], batch size: 128, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:33:34,269 INFO [optim.py:369] (1/2) 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 01:34:08,964 INFO [train.py:968] (1/2) Epoch 8, batch 1650, giga_loss[loss=0.3256, simple_loss=0.378, pruned_loss=0.1366, over 28785.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3634, pruned_loss=0.1096, over 5705726.28 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3581, pruned_loss=0.1024, over 3328253.13 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.364, pruned_loss=0.1104, over 5689909.58 frames. ], batch size: 99, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:34:29,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 01:34:30,818 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:34:50,268 INFO [train.py:968] (1/2) Epoch 8, batch 1700, giga_loss[loss=0.3396, simple_loss=0.3823, pruned_loss=0.1484, over 28279.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3652, pruned_loss=0.1126, over 5714817.66 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3574, pruned_loss=0.1022, over 3403620.25 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3662, pruned_loss=0.1136, over 5697765.64 frames. ], batch size: 77, lr: 4.18e-03, grad_scale: 8.0
+2023-03-04 01:34:57,829 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 01:34:58,655 INFO [optim.py:369] (1/2) 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 01:35:08,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0673, 3.3723, 2.4507, 1.0483], device='cuda:1'), covar=tensor([0.4491, 0.1574, 0.2096, 0.3982], device='cuda:1'), in_proj_covar=tensor([0.1460, 0.1378, 0.1428, 0.1199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 01:35:35,363 INFO [train.py:968] (1/2) Epoch 8, batch 1750, giga_loss[loss=0.2932, simple_loss=0.3433, pruned_loss=0.1216, over 23717.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.364, pruned_loss=0.1129, over 5706013.47 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3573, pruned_loss=0.1021, over 3451671.79 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.365, pruned_loss=0.1139, over 5690534.76 frames. ], batch size: 705, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:35:38,276 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 8, batch 1800, giga_loss[loss=0.3031, simple_loss=0.372, pruned_loss=0.1171, over 28925.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.362, pruned_loss=0.1124, over 5697039.54 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3576, pruned_loss=0.1021, over 3509979.58 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3627, pruned_loss=0.1135, over 5682188.24 frames. ], batch size: 186, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:36:26,945 INFO [optim.py:369] (1/2) 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 01:36:45,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 1850, giga_loss[loss=0.2899, simple_loss=0.3537, pruned_loss=0.113, over 23692.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3607, pruned_loss=0.1113, over 5693986.11 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.357, pruned_loss=0.1018, over 3545891.86 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3617, pruned_loss=0.1124, over 5679376.24 frames. ], batch size: 705, lr: 4.18e-03, grad_scale: 4.0
+2023-03-04 01:37:10,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4632, 2.1987, 1.6595, 0.5736], device='cuda:1'), covar=tensor([0.3275, 0.1587, 0.2410, 0.3794], device='cuda:1'), in_proj_covar=tensor([0.1451, 0.1372, 0.1420, 0.1194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 01:37:21,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=320654.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:37:39,448 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:37:43,592 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 1900, giga_loss[loss=0.2937, simple_loss=0.3558, pruned_loss=0.1158, over 28637.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3597, pruned_loss=0.1099, over 5689008.18 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3579, pruned_loss=0.1025, over 3593662.49 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.36, pruned_loss=0.1106, over 5683169.55 frames. ], batch size: 92, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:37:53,124 INFO [optim.py:369] (1/2) 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 01:37:54,969 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-04 01:38:12,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=320706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:38:32,877 INFO [train.py:968] (1/2) Epoch 8, batch 1950, giga_loss[loss=0.2445, simple_loss=0.319, pruned_loss=0.08495, over 28870.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3547, pruned_loss=0.1067, over 5685374.95 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3579, pruned_loss=0.1025, over 3593662.49 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3549, pruned_loss=0.1073, over 5680830.67 frames. ], batch size: 199, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:38:52,916 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 01:39:01,306 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:39:04,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=320758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:39:23,637 INFO [train.py:968] (1/2) Epoch 8, batch 2000, giga_loss[loss=0.2355, simple_loss=0.311, pruned_loss=0.07995, over 28599.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3479, pruned_loss=0.1032, over 5672399.77 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3585, pruned_loss=0.1029, over 3627226.55 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3476, pruned_loss=0.1034, over 5666640.09 frames. ], batch size: 307, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:39:32,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=320787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:39:32,834 INFO [optim.py:369] (1/2) 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 01:40:08,138 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 2050, giga_loss[loss=0.3093, simple_loss=0.3573, pruned_loss=0.1306, over 26605.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3432, pruned_loss=0.1004, over 5663799.91 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.359, pruned_loss=0.1031, over 3672125.54 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3424, pruned_loss=0.1004, over 5664665.35 frames. ], batch size: 555, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:40:45,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2624, 1.3689, 1.1960, 1.1009], device='cuda:1'), covar=tensor([0.0650, 0.0390, 0.0913, 0.0696], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0435, 0.0495, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 01:40:55,056 INFO [train.py:968] (1/2) Epoch 8, batch 2100, libri_loss[loss=0.3199, simple_loss=0.3885, pruned_loss=0.1256, over 29741.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3415, pruned_loss=0.0996, over 5658291.67 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3586, pruned_loss=0.1029, over 3745859.13 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3405, pruned_loss=0.09958, over 5653955.17 frames. ], batch size: 87, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:41:02,773 INFO [optim.py:369] (1/2) 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 01:41:34,104 INFO [train.py:968] (1/2) Epoch 8, batch 2150, giga_loss[loss=0.2472, simple_loss=0.3308, pruned_loss=0.08176, over 28689.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3431, pruned_loss=0.09987, over 5665136.83 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3586, pruned_loss=0.103, over 3821432.63 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3417, pruned_loss=0.09967, over 5662192.93 frames. ], batch size: 242, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:41:35,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=320931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:42:07,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=320971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:42:09,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 01:42:09,942 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 2200, giga_loss[loss=0.2382, simple_loss=0.3184, pruned_loss=0.07907, over 29077.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3419, pruned_loss=0.09878, over 5675181.02 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3592, pruned_loss=0.1031, over 3852830.70 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3401, pruned_loss=0.09849, over 5677816.34 frames. ], batch size: 155, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:42:21,215 INFO [optim.py:369] (1/2) 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 01:42:34,211 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-04 01:42:37,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=321003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:42:56,992 INFO [train.py:968] (1/2) Epoch 8, batch 2250, giga_loss[loss=0.3954, simple_loss=0.425, pruned_loss=0.1828, over 26768.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3398, pruned_loss=0.09805, over 5677440.44 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3594, pruned_loss=0.1032, over 3878828.49 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3381, pruned_loss=0.09776, over 5680351.25 frames. ], batch size: 555, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:42:57,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321029.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:43:21,110 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:43:35,299 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=321074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:43:37,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3418, 4.1686, 3.9163, 1.8039], device='cuda:1'), covar=tensor([0.0481, 0.0584, 0.0610, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0877, 0.0775, 0.0620], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 01:43:37,231 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 2300, giga_loss[loss=0.2476, simple_loss=0.3207, pruned_loss=0.08719, over 28539.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3374, pruned_loss=0.09655, over 5694900.74 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3598, pruned_loss=0.103, over 3929031.27 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3353, pruned_loss=0.0963, over 5692376.42 frames. ], batch size: 71, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:43:46,605 INFO [optim.py:369] (1/2) 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 01:43:59,029 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=321106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:44:15,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-04 01:44:16,164 INFO [train.py:968] (1/2) Epoch 8, batch 2350, giga_loss[loss=0.2788, simple_loss=0.3473, pruned_loss=0.1052, over 28561.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3362, pruned_loss=0.09563, over 5702623.36 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3601, pruned_loss=0.103, over 4033708.75 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3331, pruned_loss=0.09506, over 5691328.47 frames. ], batch size: 336, lr: 4.17e-03, grad_scale: 2.0
+2023-03-04 01:44:20,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1402, 2.5292, 1.1890, 1.3393], device='cuda:1'), covar=tensor([0.0972, 0.0329, 0.0832, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0487, 0.0318, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 01:44:37,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4217, 1.4448, 1.5214, 1.3637], device='cuda:1'), covar=tensor([0.1350, 0.1748, 0.1741, 0.1721], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0729, 0.0651, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 01:44:52,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=321172.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 01:44:56,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=321175.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 01:44:58,953 INFO [train.py:968] (1/2) Epoch 8, batch 2400, giga_loss[loss=0.2275, simple_loss=0.3055, pruned_loss=0.07474, over 29065.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3342, pruned_loss=0.0948, over 5703369.04 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3605, pruned_loss=0.103, over 4070950.87 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3309, pruned_loss=0.09414, over 5690398.97 frames. ], batch size: 136, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:45:06,607 INFO [optim.py:369] (1/2) 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 01:45:16,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3008, 1.4945, 1.1791, 1.0004], device='cuda:1'), covar=tensor([0.1677, 0.1316, 0.1136, 0.1517], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1412, 0.1397, 0.1512], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 01:45:17,959 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 2450, giga_loss[loss=0.2176, simple_loss=0.2928, pruned_loss=0.0712, over 28937.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3335, pruned_loss=0.0944, over 5713762.10 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3618, pruned_loss=0.1034, over 4151296.82 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3289, pruned_loss=0.09327, over 5695824.74 frames. ], batch size: 174, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:45:44,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3066, 1.4451, 1.5760, 1.3392], device='cuda:1'), covar=tensor([0.1112, 0.1252, 0.1510, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0731, 0.0652, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 01:46:08,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-04 01:46:13,019 INFO [train.py:968] (1/2) Epoch 8, batch 2500, giga_loss[loss=0.2451, simple_loss=0.3173, pruned_loss=0.08641, over 28973.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3315, pruned_loss=0.09362, over 5722556.80 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3614, pruned_loss=0.103, over 4202949.95 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3272, pruned_loss=0.09271, over 5703633.28 frames. ], batch size: 227, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:46:21,540 INFO [optim.py:369] (1/2) 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 01:46:35,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4598, 2.1257, 2.1650, 1.9216], device='cuda:1'), covar=tensor([0.1235, 0.2249, 0.1687, 0.1902], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0729, 0.0650, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:46:54,871 INFO [train.py:968] (1/2) Epoch 8, batch 2550, giga_loss[loss=0.2168, simple_loss=0.29, pruned_loss=0.07182, over 28518.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3277, pruned_loss=0.09161, over 5727091.48 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3615, pruned_loss=0.103, over 4211604.20 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3241, pruned_loss=0.09086, over 5711431.45 frames. ], batch size: 78, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:47:28,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5134, 2.1736, 1.7089, 0.6966], device='cuda:1'), covar=tensor([0.3008, 0.1371, 0.2416, 0.3509], device='cuda:1'), in_proj_covar=tensor([0.1469, 0.1379, 0.1430, 0.1200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 01:47:33,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.36 vs. limit=5.0
+2023-03-04 01:47:33,898 INFO [train.py:968] (1/2) Epoch 8, batch 2600, libri_loss[loss=0.3006, simple_loss=0.3854, pruned_loss=0.1079, over 29488.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3268, pruned_loss=0.09092, over 5731254.49 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3624, pruned_loss=0.1032, over 4261676.52 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3223, pruned_loss=0.0898, over 5713822.11 frames. ], batch size: 85, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:47:42,872 INFO [optim.py:369] (1/2) 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 01:47:59,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2852, 1.3854, 1.2219, 1.1808], device='cuda:1'), covar=tensor([0.1489, 0.1210, 0.1115, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.1569, 0.1404, 0.1395, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 01:48:07,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5510, 1.9390, 1.8985, 1.4608], device='cuda:1'), covar=tensor([0.1595, 0.1993, 0.1214, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0718, 0.0832, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 01:48:13,036 INFO [train.py:968] (1/2) Epoch 8, batch 2650, giga_loss[loss=0.2429, simple_loss=0.315, pruned_loss=0.08539, over 28744.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3247, pruned_loss=0.09018, over 5731747.34 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3621, pruned_loss=0.103, over 4269742.14 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3212, pruned_loss=0.08938, over 5717497.88 frames. ], batch size: 284, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:48:17,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:48:32,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5977, 2.0556, 1.4751, 1.3671], device='cuda:1'), covar=tensor([0.1834, 0.1254, 0.1445, 0.1531], device='cuda:1'), in_proj_covar=tensor([0.1578, 0.1411, 0.1405, 0.1518], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 01:48:56,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2010, 5.0096, 4.7420, 2.0167], device='cuda:1'), covar=tensor([0.0349, 0.0467, 0.0514, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0888, 0.0785, 0.0623], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 01:48:57,312 INFO [train.py:968] (1/2) Epoch 8, batch 2700, giga_loss[loss=0.2538, simple_loss=0.3325, pruned_loss=0.08752, over 28837.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3282, pruned_loss=0.09253, over 5723077.22 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3624, pruned_loss=0.1031, over 4284094.60 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3248, pruned_loss=0.09166, over 5717499.53 frames. ], batch size: 186, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:49:08,383 INFO [optim.py:369] (1/2) 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 01:49:29,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:49:31,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:49:39,784 INFO [train.py:968] (1/2) Epoch 8, batch 2750, giga_loss[loss=0.2868, simple_loss=0.3589, pruned_loss=0.1074, over 29019.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3338, pruned_loss=0.09583, over 5718117.20 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3628, pruned_loss=0.1032, over 4328894.41 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.33, pruned_loss=0.09485, over 5711708.69 frames. ], batch size: 164, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:50:01,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:50:21,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=321577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:50:22,001 INFO [train.py:968] (1/2) Epoch 8, batch 2800, giga_loss[loss=0.3084, simple_loss=0.3763, pruned_loss=0.1202, over 28973.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3417, pruned_loss=0.1005, over 5721998.89 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3629, pruned_loss=0.1032, over 4390246.18 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3376, pruned_loss=0.09952, over 5710637.03 frames. ], batch size: 145, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:50:24,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=321580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:50:32,967 INFO [optim.py:369] (1/2) 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 01:50:40,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4443, 1.8318, 1.8229, 1.3722], device='cuda:1'), covar=tensor([0.1513, 0.1926, 0.1132, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0719, 0.0830, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 01:50:49,839 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=321609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:51:06,411 INFO [train.py:968] (1/2) Epoch 8, batch 2850, giga_loss[loss=0.2928, simple_loss=0.3598, pruned_loss=0.1129, over 28663.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3504, pruned_loss=0.1067, over 5708728.29 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.363, pruned_loss=0.1033, over 4434475.40 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3466, pruned_loss=0.1059, over 5695020.03 frames. ], batch size: 71, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:51:49,974 INFO [train.py:968] (1/2) Epoch 8, batch 2900, giga_loss[loss=0.3119, simple_loss=0.3784, pruned_loss=0.1227, over 28538.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3548, pruned_loss=0.108, over 5710751.59 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3633, pruned_loss=0.1036, over 4468208.99 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3511, pruned_loss=0.1072, over 5702607.42 frames. ], batch size: 71, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:52:03,381 INFO [optim.py:369] (1/2) 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 01:52:27,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:52:36,597 INFO [train.py:968] (1/2) Epoch 8, batch 2950, giga_loss[loss=0.3159, simple_loss=0.3861, pruned_loss=0.1228, over 28902.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3589, pruned_loss=0.1098, over 5699658.73 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3632, pruned_loss=0.1035, over 4486891.81 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3561, pruned_loss=0.1093, over 5698227.12 frames. ], batch size: 199, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:53:23,929 INFO [train.py:968] (1/2) Epoch 8, batch 3000, giga_loss[loss=0.3168, simple_loss=0.3854, pruned_loss=0.1241, over 28508.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3665, pruned_loss=0.1153, over 5682965.68 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3637, pruned_loss=0.1038, over 4511585.53 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3638, pruned_loss=0.1149, over 5681403.11 frames. ], batch size: 336, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:53:23,929 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 01:53:30,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2743, 1.5286, 1.5398, 1.3651], device='cuda:1'), covar=tensor([0.1129, 0.1068, 0.1511, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0726, 0.0649, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:53:32,726 INFO [train.py:1012] (1/2) Epoch 8, validation: loss=0.2338, simple_loss=0.3367, pruned_loss=0.06548, over 944034.00 frames. 
+2023-03-04 01:53:32,727 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 01:53:43,397 INFO [optim.py:369] (1/2) 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 01:53:48,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6288, 2.2775, 1.9814, 1.9181], device='cuda:1'), covar=tensor([0.0748, 0.0232, 0.0259, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-04 01:54:04,258 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:54:13,262 INFO [train.py:968] (1/2) Epoch 8, batch 3050, giga_loss[loss=0.2499, simple_loss=0.327, pruned_loss=0.08647, over 28687.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3635, pruned_loss=0.1129, over 5689647.93 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3635, pruned_loss=0.104, over 4546251.45 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3616, pruned_loss=0.1126, over 5683061.61 frames. ], batch size: 242, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:54:44,161 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=321866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:54:54,759 INFO [train.py:968] (1/2) Epoch 8, batch 3100, giga_loss[loss=0.2987, simple_loss=0.3739, pruned_loss=0.1117, over 28700.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3588, pruned_loss=0.109, over 5686162.68 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3635, pruned_loss=0.1041, over 4560247.85 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3572, pruned_loss=0.1088, over 5689545.09 frames. ], batch size: 242, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:55:04,266 INFO [optim.py:369] (1/2) 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 01:55:05,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:55:05,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6187, 1.6771, 1.6923, 1.4481], device='cuda:1'), covar=tensor([0.1334, 0.1790, 0.1642, 0.1764], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0726, 0.0649, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 01:55:06,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:55:18,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-04 01:55:31,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3251, 1.3710, 1.2588, 1.3870], device='cuda:1'), covar=tensor([0.0784, 0.0316, 0.0309, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0050, 0.0045, 0.0076], device='cuda:1')
+2023-03-04 01:55:37,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=321928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:55:37,758 INFO [train.py:968] (1/2) Epoch 8, batch 3150, libri_loss[loss=0.2762, simple_loss=0.345, pruned_loss=0.1037, over 29537.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.357, pruned_loss=0.107, over 5693271.69 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3628, pruned_loss=0.1038, over 4604076.72 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.356, pruned_loss=0.1073, over 5695765.83 frames. ], batch size: 78, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:56:07,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-04 01:56:21,623 INFO [train.py:968] (1/2) Epoch 8, batch 3200, giga_loss[loss=0.2753, simple_loss=0.3514, pruned_loss=0.09961, over 28678.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3578, pruned_loss=0.1069, over 5698953.64 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3629, pruned_loss=0.1039, over 4617040.34 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3569, pruned_loss=0.107, over 5699207.23 frames. ], batch size: 284, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:56:29,734 INFO [optim.py:369] (1/2) 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 01:57:02,085 INFO [train.py:968] (1/2) Epoch 8, batch 3250, giga_loss[loss=0.302, simple_loss=0.3682, pruned_loss=0.1179, over 28851.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3606, pruned_loss=0.1084, over 5706803.35 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3624, pruned_loss=0.1036, over 4654115.41 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3601, pruned_loss=0.1088, over 5702334.25 frames. ], batch size: 199, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:57:07,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:08,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:10,176 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:10,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:15,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8194, 3.6312, 3.3518, 2.1292], device='cuda:1'), covar=tensor([0.0478, 0.0668, 0.0680, 0.1671], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0876, 0.0767, 0.0611], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 01:57:33,887 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:34,685 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:36,593 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:39,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:57:42,498 INFO [train.py:968] (1/2) Epoch 8, batch 3300, giga_loss[loss=0.2954, simple_loss=0.3676, pruned_loss=0.1116, over 28881.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3624, pruned_loss=0.1096, over 5705317.72 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3628, pruned_loss=0.1038, over 4676818.15 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3618, pruned_loss=0.11, over 5699937.20 frames. ], batch size: 186, lr: 4.17e-03, grad_scale: 8.0
+2023-03-04 01:57:53,754 INFO [optim.py:369] (1/2) 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 01:57:55,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2100, 1.7781, 1.3719, 0.4117], device='cuda:1'), covar=tensor([0.2547, 0.1530, 0.2734, 0.3216], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1359, 0.1425, 0.1188], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 01:57:56,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:58:03,974 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:58:04,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4030, 3.2132, 2.0288, 1.7230], device='cuda:1'), covar=tensor([0.1426, 0.0685, 0.0965, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1415, 0.1404, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 01:58:08,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7988, 2.8727, 2.1207, 0.8704], device='cuda:1'), covar=tensor([0.3943, 0.1519, 0.1974, 0.3731], device='cuda:1'), in_proj_covar=tensor([0.1443, 0.1355, 0.1421, 0.1184], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 01:58:13,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4273, 1.6192, 1.7953, 1.3926], device='cuda:1'), covar=tensor([0.1363, 0.1634, 0.1016, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0714, 0.0827, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 01:58:25,348 INFO [train.py:968] (1/2) Epoch 8, batch 3350, giga_loss[loss=0.2989, simple_loss=0.3675, pruned_loss=0.1152, over 28485.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3636, pruned_loss=0.1111, over 5704355.23 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3627, pruned_loss=0.1037, over 4697802.05 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3632, pruned_loss=0.1116, over 5699511.51 frames. ], batch size: 78, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:59:08,204 INFO [train.py:968] (1/2) Epoch 8, batch 3400, giga_loss[loss=0.3145, simple_loss=0.38, pruned_loss=0.1245, over 28645.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.364, pruned_loss=0.1118, over 5709756.11 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3627, pruned_loss=0.1036, over 4718176.35 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3638, pruned_loss=0.1124, over 5705671.90 frames. ], batch size: 307, lr: 4.17e-03, grad_scale: 4.0
+2023-03-04 01:59:17,994 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:59:18,447 INFO [optim.py:369] (1/2) 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 01:59:47,723 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-04 01:59:48,731 INFO [train.py:968] (1/2) Epoch 8, batch 3450, giga_loss[loss=0.2695, simple_loss=0.3454, pruned_loss=0.09679, over 28296.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3636, pruned_loss=0.1109, over 5704733.77 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3635, pruned_loss=0.1039, over 4745924.76 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3628, pruned_loss=0.1114, over 5711945.15 frames. ], batch size: 77, lr: 4.16e-03, grad_scale: 2.0
+2023-03-04 01:59:55,636 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:59:57,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 01:59:58,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:00:14,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0576, 1.2430, 3.6801, 3.0950], device='cuda:1'), covar=tensor([0.1683, 0.2431, 0.0421, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0599, 0.0553, 0.0788, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:00:22,267 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:00:31,549 INFO [train.py:968] (1/2) Epoch 8, batch 3500, giga_loss[loss=0.2795, simple_loss=0.3376, pruned_loss=0.1107, over 23936.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3642, pruned_loss=0.1111, over 5706903.96 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3636, pruned_loss=0.104, over 4774397.92 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3634, pruned_loss=0.1115, over 5708415.13 frames. ], batch size: 705, lr: 4.16e-03, grad_scale: 2.0
+2023-03-04 02:00:40,997 INFO [optim.py:369] (1/2) 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 02:01:07,573 INFO [train.py:968] (1/2) Epoch 8, batch 3550, giga_loss[loss=0.3124, simple_loss=0.3911, pruned_loss=0.1168, over 28815.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3642, pruned_loss=0.11, over 5694579.79 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.364, pruned_loss=0.1044, over 4805265.42 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3633, pruned_loss=0.1104, over 5708691.87 frames. ], batch size: 174, lr: 4.16e-03, grad_scale: 2.0
+2023-03-04 02:01:07,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=322329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:12,641 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:16,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:42,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:43,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6328, 1.6106, 1.2036, 1.2626], device='cuda:1'), covar=tensor([0.0718, 0.0603, 0.1040, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0442, 0.0501, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:01:52,384 INFO [train.py:968] (1/2) Epoch 8, batch 3600, giga_loss[loss=0.2537, simple_loss=0.3406, pruned_loss=0.08342, over 28896.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3642, pruned_loss=0.1091, over 5704175.86 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3635, pruned_loss=0.1041, over 4815104.07 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3639, pruned_loss=0.1097, over 5714442.62 frames. ], batch size: 145, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:01:56,467 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:01:58,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:02:03,590 INFO [optim.py:369] (1/2) 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 02:02:21,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:02:31,131 INFO [train.py:968] (1/2) Epoch 8, batch 3650, giga_loss[loss=0.2604, simple_loss=0.3392, pruned_loss=0.09081, over 28889.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3628, pruned_loss=0.1085, over 5711037.79 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3637, pruned_loss=0.1042, over 4837208.90 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3625, pruned_loss=0.1089, over 5714983.54 frames. ], batch size: 174, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:02:52,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1204, 1.3990, 3.6838, 3.1424], device='cuda:1'), covar=tensor([0.1626, 0.2189, 0.0405, 0.0694], device='cuda:1'), in_proj_covar=tensor([0.0594, 0.0548, 0.0783, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:03:13,256 INFO [train.py:968] (1/2) Epoch 8, batch 3700, giga_loss[loss=0.2621, simple_loss=0.3384, pruned_loss=0.09286, over 28962.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3597, pruned_loss=0.1071, over 5713229.52 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3631, pruned_loss=0.1038, over 4852852.80 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3598, pruned_loss=0.1078, over 5714114.09 frames. ], batch size: 106, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:03:26,313 INFO [optim.py:369] (1/2) 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 02:03:54,146 INFO [train.py:968] (1/2) Epoch 8, batch 3750, giga_loss[loss=0.2927, simple_loss=0.3656, pruned_loss=0.1099, over 28333.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3574, pruned_loss=0.1058, over 5717756.09 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3627, pruned_loss=0.1036, over 4868708.97 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3577, pruned_loss=0.1066, over 5715920.56 frames. ], batch size: 65, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:04:37,726 INFO [train.py:968] (1/2) Epoch 8, batch 3800, giga_loss[loss=0.3024, simple_loss=0.3745, pruned_loss=0.1152, over 28911.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3566, pruned_loss=0.1053, over 5729465.48 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3617, pruned_loss=0.1031, over 4894528.75 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3574, pruned_loss=0.1063, over 5724021.58 frames. ], batch size: 213, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:04:47,441 INFO [optim.py:369] (1/2) 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 02:05:04,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5585, 2.4471, 2.4130, 2.1509], device='cuda:1'), covar=tensor([0.1117, 0.1500, 0.1249, 0.1373], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0719, 0.0644, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 02:05:17,413 INFO [train.py:968] (1/2) Epoch 8, batch 3850, giga_loss[loss=0.3028, simple_loss=0.3774, pruned_loss=0.1141, over 28668.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3581, pruned_loss=0.1064, over 5724790.85 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3617, pruned_loss=0.103, over 4920113.60 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3587, pruned_loss=0.1073, over 5722975.64 frames. ], batch size: 307, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:05:56,067 INFO [train.py:968] (1/2) Epoch 8, batch 3900, giga_loss[loss=0.3008, simple_loss=0.3758, pruned_loss=0.1129, over 28733.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3582, pruned_loss=0.1061, over 5714210.01 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3614, pruned_loss=0.103, over 4929806.02 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3588, pruned_loss=0.1069, over 5718912.68 frames. ], batch size: 242, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:06:07,692 INFO [optim.py:369] (1/2) 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 02:06:15,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=322704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:06:36,228 INFO [train.py:968] (1/2) Epoch 8, batch 3950, libri_loss[loss=0.2505, simple_loss=0.3292, pruned_loss=0.08593, over 29375.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3564, pruned_loss=0.1043, over 5714935.17 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3607, pruned_loss=0.1027, over 4955394.38 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3573, pruned_loss=0.1053, over 5716476.64 frames. ], batch size: 67, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:06:38,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1454, 2.0281, 1.9203, 1.9023], device='cuda:1'), covar=tensor([0.1270, 0.2113, 0.1701, 0.1746], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0720, 0.0642, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 02:07:14,028 INFO [train.py:968] (1/2) Epoch 8, batch 4000, giga_loss[loss=0.2949, simple_loss=0.3551, pruned_loss=0.1173, over 28400.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3568, pruned_loss=0.1048, over 5721197.79 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3609, pruned_loss=0.1029, over 4987125.56 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3573, pruned_loss=0.1055, over 5717432.96 frames. ], batch size: 65, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:07:25,387 INFO [optim.py:369] (1/2) 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 02:07:55,616 INFO [train.py:968] (1/2) Epoch 8, batch 4050, libri_loss[loss=0.3215, simple_loss=0.4046, pruned_loss=0.1192, over 25571.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3552, pruned_loss=0.1045, over 5711559.83 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3613, pruned_loss=0.1031, over 5001650.39 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3552, pruned_loss=0.1049, over 5709029.02 frames. ], batch size: 136, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:07:56,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9661, 1.9862, 1.2902, 1.7079], device='cuda:1'), covar=tensor([0.0681, 0.0576, 0.0992, 0.0881], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0437, 0.0498, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:08:09,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=322847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:08:11,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=322850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:08:33,529 INFO [train.py:968] (1/2) Epoch 8, batch 4100, giga_loss[loss=0.2618, simple_loss=0.3369, pruned_loss=0.09335, over 28773.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3524, pruned_loss=0.1032, over 5717033.67 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3607, pruned_loss=0.1029, over 5028364.06 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3526, pruned_loss=0.1037, over 5709850.90 frames. ], batch size: 242, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:08:33,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=322879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:08:42,632 INFO [optim.py:369] (1/2) 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 02:09:09,390 INFO [train.py:968] (1/2) Epoch 8, batch 4150, giga_loss[loss=0.2852, simple_loss=0.3443, pruned_loss=0.113, over 28832.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3518, pruned_loss=0.1035, over 5714681.71 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3605, pruned_loss=0.1029, over 5062532.72 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3519, pruned_loss=0.1039, over 5702265.11 frames. ], batch size: 99, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:09:49,977 INFO [train.py:968] (1/2) Epoch 8, batch 4200, libri_loss[loss=0.2997, simple_loss=0.3731, pruned_loss=0.1131, over 27831.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3525, pruned_loss=0.1048, over 5712023.81 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.361, pruned_loss=0.1033, over 5084891.19 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.352, pruned_loss=0.1048, over 5699634.50 frames. ], batch size: 116, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:10:00,206 INFO [optim.py:369] (1/2) 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 02:10:21,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 02:10:27,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7907, 1.7208, 1.2318, 1.5177], device='cuda:1'), covar=tensor([0.0600, 0.0595, 0.0911, 0.0995], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0436, 0.0497, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:10:27,549 INFO [train.py:968] (1/2) Epoch 8, batch 4250, giga_loss[loss=0.2821, simple_loss=0.3534, pruned_loss=0.1054, over 27847.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3514, pruned_loss=0.1046, over 5708031.35 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.361, pruned_loss=0.1033, over 5100746.30 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3506, pruned_loss=0.1047, over 5702931.40 frames. ], batch size: 412, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:11:11,309 INFO [train.py:968] (1/2) Epoch 8, batch 4300, giga_loss[loss=0.2576, simple_loss=0.3249, pruned_loss=0.0951, over 28199.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3485, pruned_loss=0.1032, over 5708549.46 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3609, pruned_loss=0.1032, over 5118276.10 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3477, pruned_loss=0.1034, over 5701956.98 frames. ], batch size: 77, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:11:21,640 INFO [optim.py:369] (1/2) 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 02:11:50,787 INFO [train.py:968] (1/2) Epoch 8, batch 4350, giga_loss[loss=0.2787, simple_loss=0.3475, pruned_loss=0.1049, over 28587.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3464, pruned_loss=0.1028, over 5691374.29 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.361, pruned_loss=0.1033, over 5117022.11 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3454, pruned_loss=0.1028, over 5701578.80 frames. ], batch size: 307, lr: 4.16e-03, grad_scale: 4.0
+2023-03-04 02:12:27,977 INFO [train.py:968] (1/2) Epoch 8, batch 4400, giga_loss[loss=0.2798, simple_loss=0.3469, pruned_loss=0.1063, over 28884.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3434, pruned_loss=0.1008, over 5704052.61 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3609, pruned_loss=0.1031, over 5143779.40 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3422, pruned_loss=0.101, over 5706181.06 frames. ], batch size: 174, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:12:40,120 INFO [optim.py:369] (1/2) 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 02:12:42,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5391, 1.5489, 1.2914, 1.9310], device='cuda:1'), covar=tensor([0.2439, 0.2408, 0.2613, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1205, 0.0901, 0.1057, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:13:07,564 INFO [train.py:968] (1/2) Epoch 8, batch 4450, giga_loss[loss=0.2938, simple_loss=0.3626, pruned_loss=0.1125, over 28859.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3419, pruned_loss=0.09988, over 5711232.10 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3606, pruned_loss=0.1029, over 5162472.97 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3407, pruned_loss=0.1001, over 5708835.94 frames. ], batch size: 285, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:13:07,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3961, 1.6143, 1.2863, 1.4572], device='cuda:1'), covar=tensor([0.2054, 0.1962, 0.2149, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.1199, 0.0895, 0.1051, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:13:14,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 02:13:17,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6782, 1.0293, 2.8804, 2.6906], device='cuda:1'), covar=tensor([0.1671, 0.2280, 0.0484, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0597, 0.0552, 0.0787, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:13:28,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 02:13:35,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 02:13:47,440 INFO [train.py:968] (1/2) Epoch 8, batch 4500, libri_loss[loss=0.3075, simple_loss=0.3832, pruned_loss=0.1159, over 29529.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3466, pruned_loss=0.1019, over 5710928.71 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3604, pruned_loss=0.1027, over 5187010.24 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3452, pruned_loss=0.1022, over 5708488.98 frames. ], batch size: 89, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:14:03,135 INFO [optim.py:369] (1/2) 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 02:14:30,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3661, 2.8991, 1.4540, 1.4471], device='cuda:1'), covar=tensor([0.0790, 0.0291, 0.0836, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0492, 0.0319, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 02:14:31,734 INFO [train.py:968] (1/2) Epoch 8, batch 4550, giga_loss[loss=0.2713, simple_loss=0.35, pruned_loss=0.09626, over 28744.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3493, pruned_loss=0.1034, over 5697108.35 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3605, pruned_loss=0.1028, over 5198817.54 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3477, pruned_loss=0.1035, over 5698154.50 frames. ], batch size: 242, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:15:10,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1758, 2.0806, 1.5193, 1.8377], device='cuda:1'), covar=tensor([0.0694, 0.0685, 0.0950, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0435, 0.0495, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:15:12,224 INFO [train.py:968] (1/2) Epoch 8, batch 4600, giga_loss[loss=0.2781, simple_loss=0.3527, pruned_loss=0.1018, over 28930.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3515, pruned_loss=0.1041, over 5701589.04 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3608, pruned_loss=0.1031, over 5209005.38 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3498, pruned_loss=0.104, over 5702126.79 frames. ], batch size: 199, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:15:13,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=323380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:15:27,519 INFO [optim.py:369] (1/2) 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 02:15:57,347 INFO [train.py:968] (1/2) Epoch 8, batch 4650, giga_loss[loss=0.2549, simple_loss=0.3404, pruned_loss=0.08464, over 28629.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3533, pruned_loss=0.1043, over 5697473.06 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3609, pruned_loss=0.1033, over 5232324.72 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3515, pruned_loss=0.1041, over 5693911.50 frames. ], batch size: 242, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:16:09,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9796, 1.1728, 3.1941, 2.8444], device='cuda:1'), covar=tensor([0.1545, 0.2383, 0.0447, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0556, 0.0793, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:16:42,203 INFO [train.py:968] (1/2) Epoch 8, batch 4700, giga_loss[loss=0.2751, simple_loss=0.3521, pruned_loss=0.09903, over 28942.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3534, pruned_loss=0.1044, over 5700376.98 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3611, pruned_loss=0.1034, over 5248544.11 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3517, pruned_loss=0.1042, over 5693491.97 frames. ], batch size: 213, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:16:53,167 INFO [optim.py:369] (1/2) 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 02:17:16,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-04 02:17:21,642 INFO [train.py:968] (1/2) Epoch 8, batch 4750, giga_loss[loss=0.2688, simple_loss=0.3469, pruned_loss=0.09535, over 28871.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.353, pruned_loss=0.1049, over 5707621.89 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3613, pruned_loss=0.1035, over 5265103.70 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3512, pruned_loss=0.1046, over 5697359.44 frames. ], batch size: 227, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:18:02,559 INFO [train.py:968] (1/2) Epoch 8, batch 4800, giga_loss[loss=0.3095, simple_loss=0.3674, pruned_loss=0.1258, over 28651.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3557, pruned_loss=0.1075, over 5706539.16 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3615, pruned_loss=0.1037, over 5280869.46 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.354, pruned_loss=0.1073, over 5694119.43 frames. ], batch size: 92, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:18:15,115 INFO [optim.py:369] (1/2) 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 02:18:45,128 INFO [train.py:968] (1/2) Epoch 8, batch 4850, giga_loss[loss=0.2957, simple_loss=0.3713, pruned_loss=0.1101, over 28723.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3569, pruned_loss=0.108, over 5698036.26 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3617, pruned_loss=0.1039, over 5284342.40 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3552, pruned_loss=0.1077, over 5692750.49 frames. ], batch size: 284, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:19:30,539 INFO [train.py:968] (1/2) Epoch 8, batch 4900, giga_loss[loss=0.274, simple_loss=0.3496, pruned_loss=0.09919, over 28561.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3594, pruned_loss=0.109, over 5697871.65 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3622, pruned_loss=0.1041, over 5293030.78 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3576, pruned_loss=0.1087, over 5693410.90 frames. ], batch size: 71, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:19:41,165 INFO [optim.py:369] (1/2) 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 02:19:56,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-04 02:20:09,086 INFO [train.py:968] (1/2) Epoch 8, batch 4950, giga_loss[loss=0.279, simple_loss=0.3606, pruned_loss=0.09867, over 28958.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3615, pruned_loss=0.1095, over 5711652.86 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3623, pruned_loss=0.1041, over 5305311.80 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.36, pruned_loss=0.1094, over 5704549.57 frames. ], batch size: 227, lr: 4.16e-03, grad_scale: 8.0
+2023-03-04 02:20:31,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=323755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:20:36,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=323762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:20:38,573 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-04 02:20:49,915 INFO [train.py:968] (1/2) Epoch 8, batch 5000, giga_loss[loss=0.3197, simple_loss=0.3905, pruned_loss=0.1245, over 28837.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3617, pruned_loss=0.1093, over 5715668.93 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3629, pruned_loss=0.1044, over 5314039.84 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3601, pruned_loss=0.109, over 5707678.83 frames. ], batch size: 174, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:20:54,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3952, 1.3799, 1.5043, 1.1635], device='cuda:1'), covar=tensor([0.1817, 0.2647, 0.1498, 0.1483], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0709, 0.0820, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 02:21:01,928 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 5050, giga_loss[loss=0.3088, simple_loss=0.3838, pruned_loss=0.1169, over 28869.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3614, pruned_loss=0.1091, over 5719390.17 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3628, pruned_loss=0.1043, over 5319739.65 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3602, pruned_loss=0.1089, over 5711499.28 frames. ], batch size: 145, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:21:47,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=323850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:22:09,995 INFO [train.py:968] (1/2) Epoch 8, batch 5100, giga_loss[loss=0.2756, simple_loss=0.3563, pruned_loss=0.09744, over 29013.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3614, pruned_loss=0.1092, over 5710366.57 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3632, pruned_loss=0.1048, over 5319901.15 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3599, pruned_loss=0.1088, over 5716945.28 frames. ], batch size: 164, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:22:22,079 INFO [optim.py:369] (1/2) 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 02:22:25,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=323898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:22:27,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=323901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:22:37,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3024, 1.3680, 1.4981, 1.2457], device='cuda:1'), covar=tensor([0.1258, 0.1528, 0.1800, 0.1523], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0728, 0.0651, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 02:22:49,857 INFO [train.py:968] (1/2) Epoch 8, batch 5150, giga_loss[loss=0.2988, simple_loss=0.3666, pruned_loss=0.1155, over 29021.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.36, pruned_loss=0.1088, over 5710690.56 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3631, pruned_loss=0.1047, over 5334075.30 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.359, pruned_loss=0.1087, over 5711778.76 frames. ], batch size: 136, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:22:50,773 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=323930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:23:04,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9431, 1.6839, 1.3877, 1.4454], device='cuda:1'), covar=tensor([0.0609, 0.0698, 0.0930, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0444, 0.0497, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:23:31,243 INFO [train.py:968] (1/2) Epoch 8, batch 5200, libri_loss[loss=0.3402, simple_loss=0.4048, pruned_loss=0.1378, over 29142.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.357, pruned_loss=0.1071, over 5711102.62 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3637, pruned_loss=0.1052, over 5339584.27 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3555, pruned_loss=0.1068, over 5718409.95 frames. ], batch size: 101, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:23:44,580 INFO [optim.py:369] (1/2) 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 02:24:11,591 INFO [train.py:968] (1/2) Epoch 8, batch 5250, giga_loss[loss=0.3125, simple_loss=0.3784, pruned_loss=0.1233, over 28224.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3541, pruned_loss=0.1057, over 5715489.08 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3637, pruned_loss=0.1052, over 5345197.22 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3528, pruned_loss=0.1054, over 5719508.42 frames. ], batch size: 368, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:24:25,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2363, 1.2219, 1.1226, 0.9779], device='cuda:1'), covar=tensor([0.0644, 0.0503, 0.0942, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0443, 0.0495, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:24:34,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:24:49,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8359, 3.6656, 3.4235, 1.7507], device='cuda:1'), covar=tensor([0.0564, 0.0667, 0.0724, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.0891, 0.0790, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 02:24:50,986 INFO [train.py:968] (1/2) Epoch 8, batch 5300, giga_loss[loss=0.2877, simple_loss=0.3654, pruned_loss=0.105, over 27962.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3547, pruned_loss=0.1053, over 5704908.59 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3638, pruned_loss=0.1054, over 5358418.86 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3532, pruned_loss=0.1048, over 5711076.00 frames. ], batch size: 412, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:25:05,399 INFO [optim.py:369] (1/2) 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 02:25:18,281 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.01 vs. limit=5.0
+2023-03-04 02:25:25,801 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:25:35,165 INFO [train.py:968] (1/2) Epoch 8, batch 5350, giga_loss[loss=0.2631, simple_loss=0.3428, pruned_loss=0.09165, over 28966.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3571, pruned_loss=0.1057, over 5703769.98 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3637, pruned_loss=0.1054, over 5366697.93 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.356, pruned_loss=0.1053, over 5705590.85 frames. ], batch size: 128, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:25:41,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:26:03,691 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:26:15,826 INFO [train.py:968] (1/2) Epoch 8, batch 5400, giga_loss[loss=0.2791, simple_loss=0.3523, pruned_loss=0.103, over 29054.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3572, pruned_loss=0.1059, over 5702317.47 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.364, pruned_loss=0.1059, over 5384058.76 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3558, pruned_loss=0.1053, over 5700175.27 frames. ], batch size: 155, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:26:25,539 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-04 02:26:25,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4842, 1.4596, 1.1877, 1.2086], device='cuda:1'), covar=tensor([0.0744, 0.0568, 0.1049, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0337, 0.0439, 0.0490, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:26:27,685 INFO [optim.py:369] (1/2) 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 02:26:52,376 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 5450, giga_loss[loss=0.2939, simple_loss=0.362, pruned_loss=0.1129, over 28721.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3565, pruned_loss=0.1067, over 5704176.66 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.364, pruned_loss=0.1058, over 5390625.23 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3552, pruned_loss=0.1062, over 5702574.13 frames. ], batch size: 242, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:27:07,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-04 02:27:16,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0271, 1.1609, 3.7655, 3.0739], device='cuda:1'), covar=tensor([0.1674, 0.2431, 0.0410, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0598, 0.0556, 0.0795, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:27:28,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0430, 4.8593, 4.5643, 2.1205], device='cuda:1'), covar=tensor([0.0410, 0.0571, 0.0707, 0.2092], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.0889, 0.0786, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 02:27:35,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4473, 1.6828, 1.7283, 1.3486], device='cuda:1'), covar=tensor([0.1525, 0.1908, 0.1261, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0707, 0.0818, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 02:27:38,267 INFO [train.py:968] (1/2) Epoch 8, batch 5500, giga_loss[loss=0.256, simple_loss=0.326, pruned_loss=0.09299, over 28863.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3553, pruned_loss=0.1075, over 5704995.75 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3642, pruned_loss=0.1059, over 5408693.32 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3537, pruned_loss=0.1071, over 5697084.05 frames. ], batch size: 186, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:27:39,429 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:27:41,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:27:50,217 INFO [optim.py:369] (1/2) 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 02:28:04,178 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:28:13,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 02:28:17,336 INFO [train.py:968] (1/2) Epoch 8, batch 5550, giga_loss[loss=0.2493, simple_loss=0.3186, pruned_loss=0.08998, over 28258.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3537, pruned_loss=0.1074, over 5707863.00 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3644, pruned_loss=0.1058, over 5423898.84 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3522, pruned_loss=0.1072, over 5695952.28 frames. ], batch size: 65, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:28:21,066 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 02:28:46,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2245, 2.0488, 2.0176, 1.8786], device='cuda:1'), covar=tensor([0.1099, 0.2025, 0.1568, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0719, 0.0644, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 02:28:50,361 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:28:53,852 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 5600, giga_loss[loss=0.3222, simple_loss=0.3661, pruned_loss=0.1391, over 28547.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3511, pruned_loss=0.1062, over 5711652.56 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3644, pruned_loss=0.1057, over 5430999.47 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3497, pruned_loss=0.1062, over 5699788.52 frames. ], batch size: 85, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:29:00,497 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:29:20,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:29:39,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2260, 1.2547, 4.1027, 3.2410], device='cuda:1'), covar=tensor([0.1536, 0.2430, 0.0364, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0603, 0.0563, 0.0803, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:29:45,893 INFO [train.py:968] (1/2) Epoch 8, batch 5650, giga_loss[loss=0.2455, simple_loss=0.3126, pruned_loss=0.08924, over 28660.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3495, pruned_loss=0.1055, over 5714002.73 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3646, pruned_loss=0.1059, over 5429679.47 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3482, pruned_loss=0.1054, over 5707017.02 frames. ], batch size: 92, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:29:49,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:29:57,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-04 02:30:24,873 INFO [train.py:968] (1/2) Epoch 8, batch 5700, giga_loss[loss=0.241, simple_loss=0.3094, pruned_loss=0.08635, over 28924.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3454, pruned_loss=0.1032, over 5706870.79 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3654, pruned_loss=0.1063, over 5422604.57 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3433, pruned_loss=0.1026, over 5714866.16 frames. ], batch size: 145, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:30:35,569 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:30:38,922 INFO [optim.py:369] (1/2) 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 02:31:01,210 INFO [train.py:968] (1/2) Epoch 8, batch 5750, giga_loss[loss=0.2751, simple_loss=0.3476, pruned_loss=0.1013, over 27936.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3413, pruned_loss=0.1005, over 5712694.85 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3655, pruned_loss=0.1064, over 5441614.65 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3386, pruned_loss=0.09981, over 5713331.22 frames. ], batch size: 412, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:31:01,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5058, 1.6821, 1.4204, 1.4571], device='cuda:1'), covar=tensor([0.2240, 0.2161, 0.2390, 0.2024], device='cuda:1'), in_proj_covar=tensor([0.1200, 0.0897, 0.1053, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:31:09,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:31:10,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 02:31:40,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:31:42,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:31:42,962 INFO [train.py:968] (1/2) Epoch 8, batch 5800, giga_loss[loss=0.2694, simple_loss=0.3494, pruned_loss=0.09473, over 28926.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3409, pruned_loss=0.1004, over 5715215.98 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3651, pruned_loss=0.1062, over 5448662.02 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3387, pruned_loss=0.09988, over 5713256.12 frames. ], batch size: 186, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:31:55,891 INFO [optim.py:369] (1/2) 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 02:32:05,085 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:32:21,940 INFO [train.py:968] (1/2) Epoch 8, batch 5850, libri_loss[loss=0.3035, simple_loss=0.3806, pruned_loss=0.1132, over 29161.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3449, pruned_loss=0.1022, over 5714550.20 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3659, pruned_loss=0.1068, over 5450044.88 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3419, pruned_loss=0.1012, over 5718024.86 frames. ], batch size: 101, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:32:28,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:32:30,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:32:47,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7194, 1.7909, 1.4688, 1.8389], device='cuda:1'), covar=tensor([0.2017, 0.2093, 0.2274, 0.2288], device='cuda:1'), in_proj_covar=tensor([0.1201, 0.0896, 0.1051, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:32:54,287 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:04,043 INFO [train.py:968] (1/2) Epoch 8, batch 5900, giga_loss[loss=0.261, simple_loss=0.3413, pruned_loss=0.09039, over 29027.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3477, pruned_loss=0.1031, over 5716229.27 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3655, pruned_loss=0.1066, over 5456640.70 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3454, pruned_loss=0.1025, over 5716694.64 frames. ], batch size: 155, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:33:05,656 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=324681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:07,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:19,878 INFO [optim.py:369] (1/2) 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 02:33:29,517 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=324709.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 02:33:32,166 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:33:45,656 INFO [train.py:968] (1/2) Epoch 8, batch 5950, giga_loss[loss=0.2847, simple_loss=0.3619, pruned_loss=0.1038, over 28877.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3511, pruned_loss=0.1046, over 5716555.55 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3659, pruned_loss=0.1069, over 5466838.45 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3486, pruned_loss=0.1037, over 5713405.25 frames. ], batch size: 186, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:34:05,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=324754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:34:17,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8083, 5.1280, 1.9223, 1.9273], device='cuda:1'), covar=tensor([0.0831, 0.0156, 0.0829, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0498, 0.0321, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 02:34:22,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8637, 1.6998, 1.3107, 1.4330], device='cuda:1'), covar=tensor([0.0663, 0.0691, 0.1019, 0.1005], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0447, 0.0498, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:34:27,749 INFO [train.py:968] (1/2) Epoch 8, batch 6000, giga_loss[loss=0.3427, simple_loss=0.4027, pruned_loss=0.1413, over 28764.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3544, pruned_loss=0.1065, over 5717406.06 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3658, pruned_loss=0.1068, over 5482185.35 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.352, pruned_loss=0.1058, over 5708868.42 frames. ], batch size: 284, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:34:27,750 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 02:34:36,370 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 02:34:44,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6782, 1.6685, 1.3726, 2.1298], device='cuda:1'), covar=tensor([0.2041, 0.2149, 0.2295, 0.2020], device='cuda:1'), in_proj_covar=tensor([0.1205, 0.0900, 0.1053, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:34:52,728 INFO [optim.py:369] (1/2) 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 02:35:19,170 INFO [train.py:968] (1/2) Epoch 8, batch 6050, giga_loss[loss=0.4563, simple_loss=0.4624, pruned_loss=0.2251, over 27627.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.357, pruned_loss=0.1082, over 5711348.96 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3657, pruned_loss=0.1066, over 5487558.52 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3549, pruned_loss=0.1078, over 5705684.40 frames. ], batch size: 472, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:35:21,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6696, 3.4811, 3.2893, 1.7468], device='cuda:1'), covar=tensor([0.0609, 0.0799, 0.0806, 0.2309], device='cuda:1'), in_proj_covar=tensor([0.0956, 0.0900, 0.0800, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 02:35:50,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2655, 1.5776, 1.2456, 1.6621], device='cuda:1'), covar=tensor([0.2032, 0.1767, 0.1998, 0.1634], device='cuda:1'), in_proj_covar=tensor([0.1211, 0.0904, 0.1058, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:36:02,889 INFO [train.py:968] (1/2) Epoch 8, batch 6100, giga_loss[loss=0.2999, simple_loss=0.366, pruned_loss=0.1169, over 28460.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3629, pruned_loss=0.1134, over 5703745.49 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3655, pruned_loss=0.1068, over 5488869.21 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3612, pruned_loss=0.113, over 5704217.08 frames. ], batch size: 78, lr: 4.15e-03, grad_scale: 8.0
+2023-03-04 02:36:19,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9782, 1.7510, 1.5019, 1.5046], device='cuda:1'), covar=tensor([0.0709, 0.0747, 0.0911, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0447, 0.0496, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:36:21,989 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:36:23,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=324900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:36:29,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5562, 3.4937, 1.5651, 1.6645], device='cuda:1'), covar=tensor([0.0868, 0.0351, 0.0860, 0.1184], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0496, 0.0319, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 02:36:51,592 INFO [train.py:968] (1/2) Epoch 8, batch 6150, giga_loss[loss=0.3384, simple_loss=0.4039, pruned_loss=0.1364, over 28542.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3685, pruned_loss=0.1182, over 5693444.04 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3651, pruned_loss=0.1065, over 5495331.15 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3675, pruned_loss=0.1182, over 5690887.16 frames. ], batch size: 60, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:36:51,838 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=324929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:37:19,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9421, 4.7423, 4.5144, 2.1665], device='cuda:1'), covar=tensor([0.0382, 0.0571, 0.0582, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.0954, 0.0902, 0.0797, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 02:37:40,403 INFO [train.py:968] (1/2) Epoch 8, batch 6200, libri_loss[loss=0.2968, simple_loss=0.3678, pruned_loss=0.1129, over 29558.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3764, pruned_loss=0.1242, over 5686143.20 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3653, pruned_loss=0.1067, over 5501691.64 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3756, pruned_loss=0.1244, over 5681058.25 frames. ], batch size: 76, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:37:57,996 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 6250, giga_loss[loss=0.3324, simple_loss=0.3926, pruned_loss=0.1361, over 28742.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3817, pruned_loss=0.1289, over 5679488.28 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3661, pruned_loss=0.1072, over 5508215.58 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3809, pruned_loss=0.1294, over 5676089.47 frames. ], batch size: 284, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:38:45,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8582, 1.1913, 3.4064, 2.9458], device='cuda:1'), covar=tensor([0.1594, 0.2187, 0.0446, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0600, 0.0556, 0.0804, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:39:12,590 INFO [train.py:968] (1/2) Epoch 8, batch 6300, giga_loss[loss=0.3451, simple_loss=0.4162, pruned_loss=0.137, over 29039.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3872, pruned_loss=0.1339, over 5678834.45 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3655, pruned_loss=0.1068, over 5514943.85 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3873, pruned_loss=0.1351, over 5673183.64 frames. ], batch size: 128, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:39:16,914 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=325084.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 02:39:25,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3346, 4.1405, 3.9660, 1.7488], device='cuda:1'), covar=tensor([0.0491, 0.0688, 0.0687, 0.2195], device='cuda:1'), in_proj_covar=tensor([0.0959, 0.0902, 0.0799, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 02:39:33,262 INFO [optim.py:369] (1/2) 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 02:39:34,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6996, 1.5494, 1.1817, 1.3090], device='cuda:1'), covar=tensor([0.0640, 0.0581, 0.0944, 0.0939], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0450, 0.0500, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:40:00,812 INFO [train.py:968] (1/2) Epoch 8, batch 6350, giga_loss[loss=0.3142, simple_loss=0.3827, pruned_loss=0.1229, over 29009.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3934, pruned_loss=0.1398, over 5660807.12 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3662, pruned_loss=0.1074, over 5520287.82 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3934, pruned_loss=0.1408, over 5654267.63 frames. ], batch size: 227, lr: 4.15e-03, grad_scale: 2.0
+2023-03-04 02:40:36,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9019, 1.7362, 1.3364, 1.4611], device='cuda:1'), covar=tensor([0.0607, 0.0663, 0.0907, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0451, 0.0501, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:40:56,976 INFO [train.py:968] (1/2) Epoch 8, batch 6400, giga_loss[loss=0.3345, simple_loss=0.3927, pruned_loss=0.1381, over 28956.00 frames. ], tot_loss[loss=0.3398, simple_loss=0.3951, pruned_loss=0.1423, over 5654775.39 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3663, pruned_loss=0.1074, over 5521816.38 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3952, pruned_loss=0.1432, over 5648968.69 frames. ], batch size: 106, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:41:17,373 INFO [optim.py:369] (1/2) 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 02:41:50,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=325227.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 02:41:51,772 INFO [train.py:968] (1/2) Epoch 8, batch 6450, giga_loss[loss=0.3398, simple_loss=0.4012, pruned_loss=0.1392, over 29008.00 frames. ], tot_loss[loss=0.3474, simple_loss=0.3995, pruned_loss=0.1476, over 5634745.50 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3662, pruned_loss=0.1072, over 5527695.04 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4001, pruned_loss=0.1489, over 5626836.67 frames. ], batch size: 155, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:41:53,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=325230.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 02:41:59,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3448, 1.4195, 1.0728, 1.2303], device='cuda:1'), covar=tensor([0.0979, 0.1022, 0.0887, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1457, 0.1429, 0.1529], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 02:42:23,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=325259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 02:42:30,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5745, 1.6898, 1.8912, 1.4862], device='cuda:1'), covar=tensor([0.1092, 0.1500, 0.0898, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0710, 0.0812, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 02:42:43,718 INFO [train.py:968] (1/2) Epoch 8, batch 6500, giga_loss[loss=0.335, simple_loss=0.3993, pruned_loss=0.1353, over 29032.00 frames. ], tot_loss[loss=0.353, simple_loss=0.4034, pruned_loss=0.1513, over 5629291.37 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3663, pruned_loss=0.1072, over 5542265.02 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.405, pruned_loss=0.154, over 5613062.13 frames. ], batch size: 106, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:43:05,105 INFO [optim.py:369] (1/2) 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 02:43:33,574 INFO [train.py:968] (1/2) Epoch 8, batch 6550, giga_loss[loss=0.4055, simple_loss=0.4387, pruned_loss=0.1861, over 28931.00 frames. ], tot_loss[loss=0.3564, simple_loss=0.406, pruned_loss=0.1535, over 5626912.46 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3666, pruned_loss=0.1074, over 5549164.61 frames. ], giga_tot_loss[loss=0.3602, simple_loss=0.4078, pruned_loss=0.1563, over 5609226.83 frames. ], batch size: 174, lr: 4.15e-03, grad_scale: 4.0
+2023-03-04 02:44:24,618 INFO [train.py:968] (1/2) Epoch 8, batch 6600, giga_loss[loss=0.3378, simple_loss=0.3821, pruned_loss=0.1468, over 28677.00 frames. ], tot_loss[loss=0.355, simple_loss=0.4045, pruned_loss=0.1527, over 5643082.66 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3667, pruned_loss=0.1074, over 5552388.06 frames. ], giga_tot_loss[loss=0.3588, simple_loss=0.4064, pruned_loss=0.1556, over 5627465.87 frames. ], batch size: 60, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:44:43,815 INFO [optim.py:369] (1/2) 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 02:44:56,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4077, 1.6103, 1.2647, 1.5697], device='cuda:1'), covar=tensor([0.0737, 0.0298, 0.0317, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0119, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0050, 0.0046, 0.0076], device='cuda:1')
+2023-03-04 02:45:18,041 INFO [train.py:968] (1/2) Epoch 8, batch 6650, giga_loss[loss=0.34, simple_loss=0.3974, pruned_loss=0.1413, over 28648.00 frames. ], tot_loss[loss=0.3529, simple_loss=0.4027, pruned_loss=0.1515, over 5641911.97 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3669, pruned_loss=0.1075, over 5557646.95 frames. ], giga_tot_loss[loss=0.3565, simple_loss=0.4044, pruned_loss=0.1543, over 5626474.07 frames. ], batch size: 307, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:45:46,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-04 02:46:10,406 INFO [train.py:968] (1/2) Epoch 8, batch 6700, giga_loss[loss=0.42, simple_loss=0.439, pruned_loss=0.2005, over 26540.00 frames. ], tot_loss[loss=0.3521, simple_loss=0.4027, pruned_loss=0.1507, over 5640888.26 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3668, pruned_loss=0.1074, over 5560817.49 frames. ], giga_tot_loss[loss=0.3557, simple_loss=0.4046, pruned_loss=0.1534, over 5626496.85 frames. ], batch size: 555, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:46:32,333 INFO [optim.py:369] (1/2) 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 02:47:03,342 INFO [train.py:968] (1/2) Epoch 8, batch 6750, giga_loss[loss=0.3475, simple_loss=0.4036, pruned_loss=0.1457, over 28848.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.404, pruned_loss=0.1512, over 5638218.21 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3668, pruned_loss=0.1074, over 5563820.95 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4059, pruned_loss=0.1538, over 5624808.58 frames. ], batch size: 186, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:47:26,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 02:47:46,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5554, 1.5903, 1.1695, 1.4185], device='cuda:1'), covar=tensor([0.0583, 0.0448, 0.0792, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0450, 0.0498, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 02:47:58,659 INFO [train.py:968] (1/2) Epoch 8, batch 6800, giga_loss[loss=0.2759, simple_loss=0.348, pruned_loss=0.1019, over 28896.00 frames. ], tot_loss[loss=0.3539, simple_loss=0.4045, pruned_loss=0.1517, over 5616190.75 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3667, pruned_loss=0.1074, over 5565566.84 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4062, pruned_loss=0.1539, over 5604424.24 frames. ], batch size: 106, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 02:48:14,803 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3169, 1.9343, 1.4284, 0.3771], device='cuda:1'), covar=tensor([0.2674, 0.1418, 0.2159, 0.3493], device='cuda:1'), in_proj_covar=tensor([0.1490, 0.1395, 0.1450, 0.1217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 02:48:15,052 INFO [optim.py:369] (1/2) 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 02:48:15,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5126, 3.5449, 1.5281, 1.5982], device='cuda:1'), covar=tensor([0.0881, 0.0282, 0.0872, 0.1267], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0503, 0.0323, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 02:48:50,819 INFO [train.py:968] (1/2) Epoch 8, batch 6850, giga_loss[loss=0.3022, simple_loss=0.3773, pruned_loss=0.1136, over 28430.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.4009, pruned_loss=0.1473, over 5619537.40 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3671, pruned_loss=0.1078, over 5566942.27 frames. ], giga_tot_loss[loss=0.351, simple_loss=0.4027, pruned_loss=0.1496, over 5609892.68 frames. ], batch size: 60, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 02:49:38,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6202, 2.3715, 1.6442, 0.7007], device='cuda:1'), covar=tensor([0.3453, 0.1741, 0.2658, 0.3891], device='cuda:1'), in_proj_covar=tensor([0.1481, 0.1385, 0.1441, 0.1209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 02:49:41,315 INFO [train.py:968] (1/2) Epoch 8, batch 6900, libri_loss[loss=0.243, simple_loss=0.3162, pruned_loss=0.08486, over 28561.00 frames. ], tot_loss[loss=0.3415, simple_loss=0.3972, pruned_loss=0.1429, over 5631115.20 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3668, pruned_loss=0.1077, over 5572612.38 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3995, pruned_loss=0.1455, over 5619495.22 frames. ], batch size: 63, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:49:53,384 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=325693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:49:58,697 INFO [optim.py:369] (1/2) 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 02:50:05,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 02:50:31,327 INFO [train.py:968] (1/2) Epoch 8, batch 6950, giga_loss[loss=0.315, simple_loss=0.3848, pruned_loss=0.1226, over 28024.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.393, pruned_loss=0.1391, over 5646074.67 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3663, pruned_loss=0.1074, over 5576753.84 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3956, pruned_loss=0.1418, over 5634205.52 frames. ], batch size: 412, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:51:18,468 INFO [train.py:968] (1/2) Epoch 8, batch 7000, giga_loss[loss=0.33, simple_loss=0.3883, pruned_loss=0.1359, over 28943.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3903, pruned_loss=0.1369, over 5649108.11 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3659, pruned_loss=0.1071, over 5585585.06 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3936, pruned_loss=0.1403, over 5633709.55 frames. ], batch size: 112, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:51:24,844 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=325787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:51:37,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3115, 1.6436, 1.3334, 1.4608], device='cuda:1'), covar=tensor([0.0735, 0.0288, 0.0312, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0051, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 02:51:38,235 INFO [optim.py:369] (1/2) 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 02:52:04,741 INFO [train.py:968] (1/2) Epoch 8, batch 7050, giga_loss[loss=0.3007, simple_loss=0.377, pruned_loss=0.1122, over 28935.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3887, pruned_loss=0.1357, over 5660742.35 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.366, pruned_loss=0.1073, over 5595137.15 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3919, pruned_loss=0.1392, over 5641655.60 frames. ], batch size: 174, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:52:07,750 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-04 02:52:54,883 INFO [train.py:968] (1/2) Epoch 8, batch 7100, giga_loss[loss=0.2942, simple_loss=0.3699, pruned_loss=0.1092, over 28868.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3886, pruned_loss=0.1354, over 5669072.20 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3659, pruned_loss=0.1073, over 5602960.11 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3917, pruned_loss=0.1387, over 5648256.50 frames. ], batch size: 186, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:53:15,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0785, 1.0807, 4.0602, 3.1718], device='cuda:1'), covar=tensor([0.1707, 0.2573, 0.0420, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0562, 0.0812, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 02:53:17,245 INFO [optim.py:369] (1/2) 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 02:53:50,497 INFO [train.py:968] (1/2) Epoch 8, batch 7150, giga_loss[loss=0.2711, simple_loss=0.3564, pruned_loss=0.09288, over 28855.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3854, pruned_loss=0.1322, over 5675860.71 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3655, pruned_loss=0.107, over 5609369.86 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3888, pruned_loss=0.1358, over 5655061.70 frames. ], batch size: 174, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:54:03,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 02:54:22,230 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=325960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:54:45,876 INFO [train.py:968] (1/2) Epoch 8, batch 7200, libri_loss[loss=0.2948, simple_loss=0.3624, pruned_loss=0.1136, over 29521.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3843, pruned_loss=0.1297, over 5676816.52 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3651, pruned_loss=0.1068, over 5616717.41 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3877, pruned_loss=0.1333, over 5654968.39 frames. ], batch size: 81, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 02:55:08,446 INFO [optim.py:369] (1/2) 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 02:55:35,290 INFO [train.py:968] (1/2) Epoch 8, batch 7250, giga_loss[loss=0.3399, simple_loss=0.4039, pruned_loss=0.138, over 29003.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3863, pruned_loss=0.1295, over 5672595.60 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.365, pruned_loss=0.1068, over 5611960.81 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3895, pruned_loss=0.1327, over 5659876.54 frames. ], batch size: 136, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 02:55:59,383 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=326052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:56:09,873 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-04 02:56:16,543 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 7300, giga_loss[loss=0.3173, simple_loss=0.3818, pruned_loss=0.1264, over 29047.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3869, pruned_loss=0.1303, over 5671018.14 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3649, pruned_loss=0.1065, over 5616214.98 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3899, pruned_loss=0.1335, over 5658099.26 frames. ], batch size: 128, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:56:50,554 INFO [optim.py:369] (1/2) 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 02:57:10,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2609, 1.3760, 1.2490, 1.0936], device='cuda:1'), covar=tensor([0.1240, 0.1183, 0.0746, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.1594, 0.1436, 0.1419, 0.1519], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 02:57:16,607 INFO [train.py:968] (1/2) Epoch 8, batch 7350, giga_loss[loss=0.3295, simple_loss=0.3905, pruned_loss=0.1343, over 28258.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.386, pruned_loss=0.1303, over 5670130.83 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3646, pruned_loss=0.1064, over 5613252.93 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3892, pruned_loss=0.1335, over 5663960.31 frames. ], batch size: 77, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:57:45,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:58:06,554 INFO [train.py:968] (1/2) Epoch 8, batch 7400, giga_loss[loss=0.3332, simple_loss=0.3878, pruned_loss=0.1393, over 27897.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3847, pruned_loss=0.1304, over 5666181.50 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3645, pruned_loss=0.1063, over 5623966.36 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3881, pruned_loss=0.134, over 5653929.05 frames. ], batch size: 412, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:58:25,031 INFO [optim.py:369] (1/2) 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 02:58:35,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:58:37,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:58:53,100 INFO [train.py:968] (1/2) Epoch 8, batch 7450, giga_loss[loss=0.3243, simple_loss=0.3844, pruned_loss=0.1321, over 28862.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3837, pruned_loss=0.1312, over 5675455.26 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3642, pruned_loss=0.1062, over 5627041.61 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3869, pruned_loss=0.1345, over 5663780.02 frames. ], batch size: 186, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:59:05,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 02:59:13,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8730, 2.6312, 1.7229, 1.1116], device='cuda:1'), covar=tensor([0.3809, 0.1983, 0.2406, 0.3509], device='cuda:1'), in_proj_covar=tensor([0.1481, 0.1388, 0.1446, 0.1211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 02:59:40,840 INFO [train.py:968] (1/2) Epoch 8, batch 7500, giga_loss[loss=0.3521, simple_loss=0.4149, pruned_loss=0.1447, over 28605.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3833, pruned_loss=0.1307, over 5679802.04 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3641, pruned_loss=0.106, over 5631173.97 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3865, pruned_loss=0.1341, over 5667913.64 frames. ], batch size: 78, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 02:59:48,355 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=326287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:00:04,504 INFO [optim.py:369] (1/2) 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 03:00:08,987 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:00:13,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:00:30,820 INFO [train.py:968] (1/2) Epoch 8, batch 7550, libri_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08795, over 29585.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3812, pruned_loss=0.1274, over 5694480.51 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3635, pruned_loss=0.1057, over 5636605.34 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3847, pruned_loss=0.1309, over 5680908.84 frames. ], batch size: 75, lr: 4.14e-03, grad_scale: 2.0
+2023-03-04 03:00:35,745 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:01:18,716 INFO [train.py:968] (1/2) Epoch 8, batch 7600, giga_loss[loss=0.3243, simple_loss=0.3907, pruned_loss=0.129, over 28612.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3813, pruned_loss=0.127, over 5695135.06 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3633, pruned_loss=0.1056, over 5636867.81 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3844, pruned_loss=0.1301, over 5685307.51 frames. ], batch size: 307, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:01:39,472 INFO [optim.py:369] (1/2) 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 03:02:00,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:02:01,426 INFO [train.py:968] (1/2) Epoch 8, batch 7650, giga_loss[loss=0.2873, simple_loss=0.3571, pruned_loss=0.1088, over 29036.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3806, pruned_loss=0.1266, over 5702116.02 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3634, pruned_loss=0.1057, over 5644969.57 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3835, pruned_loss=0.1297, over 5688139.21 frames. ], batch size: 128, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:02:33,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-04 03:02:53,015 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 7700, giga_loss[loss=0.3173, simple_loss=0.3571, pruned_loss=0.1388, over 23784.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3792, pruned_loss=0.1269, over 5695997.71 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3639, pruned_loss=0.1061, over 5646816.18 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3813, pruned_loss=0.1292, over 5683860.44 frames. ], batch size: 705, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:02:56,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:03:19,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4069, 1.5445, 1.5877, 1.5476], device='cuda:1'), covar=tensor([0.1105, 0.1139, 0.1228, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0725, 0.0651, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 03:03:20,794 INFO [optim.py:369] (1/2) 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 03:03:27,245 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:03:32,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6171, 1.6943, 1.5411, 1.6034], device='cuda:1'), covar=tensor([0.1359, 0.1734, 0.1902, 0.1556], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0726, 0.0652, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 03:03:34,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6088, 1.7823, 1.3978, 1.2905], device='cuda:1'), covar=tensor([0.1606, 0.1285, 0.1120, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.1595, 0.1451, 0.1420, 0.1533], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 03:03:42,496 INFO [train.py:968] (1/2) Epoch 8, batch 7750, giga_loss[loss=0.2969, simple_loss=0.3633, pruned_loss=0.1152, over 28641.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3795, pruned_loss=0.1279, over 5694817.95 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3632, pruned_loss=0.1057, over 5654449.44 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3824, pruned_loss=0.1309, over 5678874.11 frames. ], batch size: 242, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:04:24,547 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:04:26,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:04:27,764 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-04 03:04:32,712 INFO [train.py:968] (1/2) Epoch 8, batch 7800, giga_loss[loss=0.307, simple_loss=0.3681, pruned_loss=0.123, over 28290.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3798, pruned_loss=0.1288, over 5693654.12 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3631, pruned_loss=0.1056, over 5649641.27 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3825, pruned_loss=0.1316, over 5685881.04 frames. ], batch size: 368, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:04:57,015 INFO [optim.py:369] (1/2) 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 03:04:57,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:05:15,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=326620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:05:23,813 INFO [train.py:968] (1/2) Epoch 8, batch 7850, libri_loss[loss=0.289, simple_loss=0.3616, pruned_loss=0.1082, over 29561.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3773, pruned_loss=0.1277, over 5698542.37 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3629, pruned_loss=0.1055, over 5652074.23 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3798, pruned_loss=0.1302, over 5690551.87 frames. ], batch size: 79, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:05:55,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:06:08,070 INFO [train.py:968] (1/2) Epoch 8, batch 7900, libri_loss[loss=0.3037, simple_loss=0.374, pruned_loss=0.1168, over 25934.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5700603.80 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3629, pruned_loss=0.1055, over 5655298.84 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3792, pruned_loss=0.1304, over 5692718.32 frames. ], batch size: 136, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:06:29,066 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 7950, giga_loss[loss=0.3, simple_loss=0.3735, pruned_loss=0.1133, over 28915.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3777, pruned_loss=0.1283, over 5698349.11 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3635, pruned_loss=0.1059, over 5661013.26 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3795, pruned_loss=0.1309, over 5687959.51 frames. ], batch size: 199, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:07:09,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2131, 1.2953, 1.1096, 1.0440], device='cuda:1'), covar=tensor([0.0703, 0.0471, 0.0963, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0445, 0.0492, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 03:07:34,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3286, 1.6046, 1.3728, 1.4306], device='cuda:1'), covar=tensor([0.0763, 0.0305, 0.0314, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0116, 0.0120, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0050, 0.0046, 0.0076], device='cuda:1')
+2023-03-04 03:07:41,745 INFO [train.py:968] (1/2) Epoch 8, batch 8000, giga_loss[loss=0.3474, simple_loss=0.3922, pruned_loss=0.1513, over 27455.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3794, pruned_loss=0.1294, over 5692648.80 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3636, pruned_loss=0.106, over 5664675.70 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.381, pruned_loss=0.1316, over 5681716.82 frames. ], batch size: 472, lr: 4.14e-03, grad_scale: 8.0
+2023-03-04 03:08:07,103 INFO [optim.py:369] (1/2) 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 03:08:09,061 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=326805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:08:11,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=326808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:08:31,070 INFO [train.py:968] (1/2) Epoch 8, batch 8050, giga_loss[loss=0.3127, simple_loss=0.3819, pruned_loss=0.1218, over 28831.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3792, pruned_loss=0.1286, over 5683005.77 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3634, pruned_loss=0.1059, over 5665404.75 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3808, pruned_loss=0.1306, over 5674058.01 frames. ], batch size: 174, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:08:33,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3015, 1.6295, 1.3007, 1.4898], device='cuda:1'), covar=tensor([0.0755, 0.0319, 0.0331, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0116, 0.0120, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0051, 0.0046, 0.0076], device='cuda:1')
+2023-03-04 03:08:37,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=326837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:09:19,023 INFO [train.py:968] (1/2) Epoch 8, batch 8100, giga_loss[loss=0.2962, simple_loss=0.3616, pruned_loss=0.1154, over 28372.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3809, pruned_loss=0.1297, over 5678382.39 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3636, pruned_loss=0.1062, over 5670703.13 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3822, pruned_loss=0.1315, over 5666931.02 frames. ], batch size: 71, lr: 4.14e-03, grad_scale: 4.0
+2023-03-04 03:09:42,386 INFO [optim.py:369] (1/2) 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 03:10:04,148 INFO [train.py:968] (1/2) Epoch 8, batch 8150, giga_loss[loss=0.3841, simple_loss=0.4186, pruned_loss=0.1748, over 28987.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3815, pruned_loss=0.1301, over 5692040.62 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3636, pruned_loss=0.1061, over 5677754.44 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3831, pruned_loss=0.1322, over 5676842.47 frames. ], batch size: 213, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:10:53,687 INFO [train.py:968] (1/2) Epoch 8, batch 8200, giga_loss[loss=0.2955, simple_loss=0.3637, pruned_loss=0.1136, over 29058.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3834, pruned_loss=0.1322, over 5682194.79 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.363, pruned_loss=0.1057, over 5681088.39 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3857, pruned_loss=0.1351, over 5667740.51 frames. ], batch size: 155, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:11:13,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=326995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:11:21,768 INFO [optim.py:369] (1/2) 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 03:11:23,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1849, 1.8341, 1.3676, 0.3082], device='cuda:1'), covar=tensor([0.2419, 0.1543, 0.2406, 0.3262], device='cuda:1'), in_proj_covar=tensor([0.1477, 0.1392, 0.1441, 0.1205], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 03:11:46,558 INFO [train.py:968] (1/2) Epoch 8, batch 8250, giga_loss[loss=0.3348, simple_loss=0.3848, pruned_loss=0.1424, over 28869.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3839, pruned_loss=0.1336, over 5687212.17 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3628, pruned_loss=0.1058, over 5687048.17 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3864, pruned_loss=0.1364, over 5670369.11 frames. ], batch size: 119, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:11:50,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-04 03:12:37,720 INFO [train.py:968] (1/2) Epoch 8, batch 8300, giga_loss[loss=0.2853, simple_loss=0.3568, pruned_loss=0.1069, over 28972.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3871, pruned_loss=0.1371, over 5675514.32 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3629, pruned_loss=0.1058, over 5691360.47 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3895, pruned_loss=0.14, over 5658401.80 frames. ], batch size: 136, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:12:58,794 INFO [optim.py:369] (1/2) 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 03:13:24,538 INFO [train.py:968] (1/2) Epoch 8, batch 8350, giga_loss[loss=0.3144, simple_loss=0.3793, pruned_loss=0.1247, over 28765.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3858, pruned_loss=0.1364, over 5675007.16 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3624, pruned_loss=0.1056, over 5695271.22 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3891, pruned_loss=0.1401, over 5657119.15 frames. ], batch size: 284, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:13:33,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=327138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:13:35,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=327141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:14:02,760 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=327170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:14:03,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-04 03:14:08,456 INFO [train.py:968] (1/2) Epoch 8, batch 8400, giga_loss[loss=0.3396, simple_loss=0.3946, pruned_loss=0.1423, over 28859.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1354, over 5682005.25 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3627, pruned_loss=0.1056, over 5702347.76 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3879, pruned_loss=0.1393, over 5660503.82 frames. ], batch size: 145, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:14:12,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5660, 4.9598, 1.7311, 1.7993], device='cuda:1'), covar=tensor([0.1149, 0.0299, 0.0954, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0503, 0.0322, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0028, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 03:14:29,954 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 8450, giga_loss[loss=0.3133, simple_loss=0.3593, pruned_loss=0.1337, over 26502.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3844, pruned_loss=0.1343, over 5666811.90 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3626, pruned_loss=0.1057, over 5685343.82 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3871, pruned_loss=0.1376, over 5665280.96 frames. ], batch size: 555, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:15:02,344 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-04 03:15:41,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3791, 1.6598, 1.6793, 1.2789], device='cuda:1'), covar=tensor([0.1597, 0.2280, 0.1274, 0.1459], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0721, 0.0823, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 03:15:41,918 INFO [train.py:968] (1/2) Epoch 8, batch 8500, giga_loss[loss=0.311, simple_loss=0.3806, pruned_loss=0.1208, over 28956.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3816, pruned_loss=0.1314, over 5665540.04 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.363, pruned_loss=0.1058, over 5688172.66 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1344, over 5661651.52 frames. ], batch size: 145, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:15:59,444 INFO [optim.py:369] (1/2) 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 03:16:24,761 INFO [train.py:968] (1/2) Epoch 8, batch 8550, giga_loss[loss=0.3096, simple_loss=0.3742, pruned_loss=0.1225, over 29090.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3815, pruned_loss=0.1316, over 5667870.80 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3638, pruned_loss=0.1064, over 5686950.13 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3832, pruned_loss=0.1346, over 5665244.04 frames. ], batch size: 155, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:17:06,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1948, 1.1634, 4.2118, 3.3646], device='cuda:1'), covar=tensor([0.1649, 0.2550, 0.0359, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0564, 0.0807, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 03:17:12,906 INFO [train.py:968] (1/2) Epoch 8, batch 8600, giga_loss[loss=0.3391, simple_loss=0.3937, pruned_loss=0.1422, over 28881.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3806, pruned_loss=0.1321, over 5672040.50 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3637, pruned_loss=0.1063, over 5689403.48 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3823, pruned_loss=0.1348, over 5667587.14 frames. ], batch size: 186, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:17:14,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4158, 2.0559, 1.6522, 1.7375], device='cuda:1'), covar=tensor([0.0604, 0.0655, 0.0874, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0450, 0.0499, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 03:17:32,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=327398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:17:39,677 INFO [optim.py:369] (1/2) 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 03:18:05,309 INFO [train.py:968] (1/2) Epoch 8, batch 8650, giga_loss[loss=0.4124, simple_loss=0.4287, pruned_loss=0.1981, over 23375.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3806, pruned_loss=0.1329, over 5645609.62 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3634, pruned_loss=0.1061, over 5686256.51 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3825, pruned_loss=0.1358, over 5644649.48 frames. ], batch size: 705, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:18:38,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4845, 2.3431, 2.3425, 1.9037], device='cuda:1'), covar=tensor([0.1191, 0.1890, 0.1566, 0.1867], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0727, 0.0651, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 03:18:51,222 INFO [train.py:968] (1/2) Epoch 8, batch 8700, libri_loss[loss=0.337, simple_loss=0.4049, pruned_loss=0.1346, over 27768.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3819, pruned_loss=0.1321, over 5650801.32 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3629, pruned_loss=0.1058, over 5682226.68 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3844, pruned_loss=0.1354, over 5653634.72 frames. ], batch size: 116, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:19:18,006 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 8750, giga_loss[loss=0.3326, simple_loss=0.3975, pruned_loss=0.1338, over 28563.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3853, pruned_loss=0.1317, over 5646117.17 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.363, pruned_loss=0.1059, over 5674260.61 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3873, pruned_loss=0.1344, over 5655237.78 frames. ], batch size: 336, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:19:48,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 03:19:54,195 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 8800, giga_loss[loss=0.3416, simple_loss=0.3985, pruned_loss=0.1424, over 28925.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3844, pruned_loss=0.1298, over 5660359.51 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3619, pruned_loss=0.1054, over 5676981.80 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3882, pruned_loss=0.1338, over 5664019.22 frames. ], batch size: 119, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:20:48,696 INFO [optim.py:369] (1/2) 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 03:21:08,594 INFO [train.py:968] (1/2) Epoch 8, batch 8850, libri_loss[loss=0.2887, simple_loss=0.3636, pruned_loss=0.1069, over 29536.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3868, pruned_loss=0.1317, over 5667197.19 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.362, pruned_loss=0.1053, over 5677754.07 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3902, pruned_loss=0.1354, over 5669198.68 frames. ], batch size: 80, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:21:54,300 INFO [train.py:968] (1/2) Epoch 8, batch 8900, giga_loss[loss=0.4352, simple_loss=0.4591, pruned_loss=0.2056, over 27573.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.389, pruned_loss=0.1342, over 5657600.46 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3619, pruned_loss=0.1052, over 5681351.31 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3923, pruned_loss=0.1378, over 5655678.53 frames. ], batch size: 472, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:21:57,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3466, 1.4312, 1.0472, 1.2588], device='cuda:1'), covar=tensor([0.1093, 0.1026, 0.0877, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.1606, 0.1456, 0.1434, 0.1544], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 03:22:19,026 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 8950, giga_loss[loss=0.4392, simple_loss=0.4385, pruned_loss=0.2199, over 23487.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3883, pruned_loss=0.1347, over 5654221.09 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3621, pruned_loss=0.1053, over 5685327.73 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3912, pruned_loss=0.138, over 5648706.25 frames. ], batch size: 705, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:23:25,508 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 9000, giga_loss[loss=0.3035, simple_loss=0.3671, pruned_loss=0.12, over 28902.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3874, pruned_loss=0.1352, over 5643986.12 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3619, pruned_loss=0.1051, over 5690252.01 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3907, pruned_loss=0.1389, over 5634156.82 frames. ], batch size: 186, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:23:30,975 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 03:23:39,746 INFO [train.py:1012] (1/2) Epoch 8, validation: loss=0.2244, simple_loss=0.3301, pruned_loss=0.05937, over 944034.00 frames. 
+2023-03-04 03:23:39,746 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 03:23:59,994 INFO [optim.py:369] (1/2) 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 03:24:07,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3717, 5.2011, 4.8971, 2.2639], device='cuda:1'), covar=tensor([0.0331, 0.0487, 0.0574, 0.1791], device='cuda:1'), in_proj_covar=tensor([0.0965, 0.0925, 0.0815, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 03:24:24,821 INFO [train.py:968] (1/2) Epoch 8, batch 9050, giga_loss[loss=0.2813, simple_loss=0.3516, pruned_loss=0.1055, over 28785.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3858, pruned_loss=0.1342, over 5660148.96 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3623, pruned_loss=0.1051, over 5697613.07 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3889, pruned_loss=0.1381, over 5644484.94 frames. ], batch size: 174, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:24:58,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2205, 1.3234, 1.0967, 1.0706], device='cuda:1'), covar=tensor([0.1052, 0.1153, 0.0805, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1436, 0.1420, 0.1529], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 03:25:12,558 INFO [train.py:968] (1/2) Epoch 8, batch 9100, giga_loss[loss=0.3226, simple_loss=0.3795, pruned_loss=0.1328, over 28682.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3869, pruned_loss=0.136, over 5667462.56 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3626, pruned_loss=0.1053, over 5701662.09 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3896, pruned_loss=0.1395, over 5650617.38 frames. ], batch size: 242, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:25:42,025 INFO [optim.py:369] (1/2) 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 03:25:44,059 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=327907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:25:52,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=327915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:25:52,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=327916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:25:55,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=327919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:25:58,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4007, 1.7288, 1.4595, 1.5361], device='cuda:1'), covar=tensor([0.0761, 0.0295, 0.0323, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0116, 0.0120, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0051, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 03:26:05,140 INFO [train.py:968] (1/2) Epoch 8, batch 9150, giga_loss[loss=0.3653, simple_loss=0.4144, pruned_loss=0.1581, over 28786.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3877, pruned_loss=0.1375, over 5656391.36 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3623, pruned_loss=0.1052, over 5706003.92 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3905, pruned_loss=0.1409, over 5638458.14 frames. ], batch size: 243, lr: 4.13e-03, grad_scale: 2.0
+2023-03-04 03:26:23,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1915, 0.9183, 0.8144, 1.4312], device='cuda:1'), covar=tensor([0.0754, 0.0336, 0.0338, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0116, 0.0119, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0070, 0.0051, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 03:26:23,642 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=327948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:26:30,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=327955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:26:53,760 INFO [train.py:968] (1/2) Epoch 8, batch 9200, libri_loss[loss=0.355, simple_loss=0.41, pruned_loss=0.15, over 29311.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3855, pruned_loss=0.1364, over 5658834.99 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3624, pruned_loss=0.1055, over 5700853.98 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3882, pruned_loss=0.1396, over 5647506.76 frames. ], batch size: 94, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:27:21,293 INFO [optim.py:369] (1/2) 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 03:27:26,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9579, 4.7869, 4.5083, 2.2076], device='cuda:1'), covar=tensor([0.0342, 0.0503, 0.0579, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.0961, 0.0916, 0.0809, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 03:27:37,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5769, 1.5565, 1.1678, 1.2205], device='cuda:1'), covar=tensor([0.0592, 0.0472, 0.0870, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0448, 0.0496, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 03:27:44,809 INFO [train.py:968] (1/2) Epoch 8, batch 9250, giga_loss[loss=0.3386, simple_loss=0.3933, pruned_loss=0.1419, over 28933.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3825, pruned_loss=0.1345, over 5657804.45 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3623, pruned_loss=0.1053, over 5699006.14 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3851, pruned_loss=0.1377, over 5650015.95 frames. ], batch size: 186, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:28:15,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:28:17,427 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 9300, giga_loss[loss=0.3063, simple_loss=0.3791, pruned_loss=0.1168, over 28872.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3816, pruned_loss=0.1329, over 5649098.42 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3621, pruned_loss=0.1052, over 5693609.66 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3843, pruned_loss=0.1362, over 5646937.74 frames. ], batch size: 186, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:28:41,639 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:28:55,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8814, 1.7624, 1.2341, 1.5281], device='cuda:1'), covar=tensor([0.0656, 0.0693, 0.0969, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0450, 0.0498, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 03:28:59,238 INFO [optim.py:369] (1/2) 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 03:29:20,481 INFO [train.py:968] (1/2) Epoch 8, batch 9350, giga_loss[loss=0.4786, simple_loss=0.48, pruned_loss=0.2386, over 26488.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3843, pruned_loss=0.1338, over 5652394.90 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3626, pruned_loss=0.1054, over 5693573.21 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3867, pruned_loss=0.1371, over 5649463.23 frames. ], batch size: 555, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:30:08,098 INFO [train.py:968] (1/2) Epoch 8, batch 9400, libri_loss[loss=0.278, simple_loss=0.3571, pruned_loss=0.09946, over 27974.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3854, pruned_loss=0.1346, over 5649064.52 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3626, pruned_loss=0.1053, over 5697560.31 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.388, pruned_loss=0.1382, over 5641912.43 frames. ], batch size: 116, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:30:26,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-04 03:30:31,427 INFO [optim.py:369] (1/2) 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 03:30:46,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3632, 1.5715, 1.2926, 1.5170], device='cuda:1'), covar=tensor([0.2244, 0.2187, 0.2433, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.0913, 0.1067, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 03:30:53,187 INFO [train.py:968] (1/2) Epoch 8, batch 9450, giga_loss[loss=0.3589, simple_loss=0.4063, pruned_loss=0.1558, over 27572.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3854, pruned_loss=0.1339, over 5659952.07 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3624, pruned_loss=0.1052, over 5700560.09 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.388, pruned_loss=0.1376, over 5650400.68 frames. ], batch size: 472, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:31:40,034 INFO [train.py:968] (1/2) Epoch 8, batch 9500, giga_loss[loss=0.3667, simple_loss=0.4275, pruned_loss=0.1529, over 28881.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3862, pruned_loss=0.1321, over 5665088.06 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3623, pruned_loss=0.1051, over 5704597.49 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3889, pruned_loss=0.1355, over 5653228.35 frames. ], batch size: 199, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:31:44,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:31:48,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4305, 1.8013, 1.6819, 1.2570], device='cuda:1'), covar=tensor([0.1793, 0.2219, 0.1429, 0.1619], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0722, 0.0825, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 03:32:06,573 INFO [optim.py:369] (1/2) 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 03:32:20,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 03:32:26,092 INFO [train.py:968] (1/2) Epoch 8, batch 9550, giga_loss[loss=0.3091, simple_loss=0.3877, pruned_loss=0.1152, over 29138.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3887, pruned_loss=0.1321, over 5665384.69 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3624, pruned_loss=0.1053, over 5693152.45 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3911, pruned_loss=0.1351, over 5665353.95 frames. ], batch size: 155, lr: 4.13e-03, grad_scale: 4.0
+2023-03-04 03:32:27,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:15,019 INFO [train.py:968] (1/2) Epoch 8, batch 9600, giga_loss[loss=0.3767, simple_loss=0.4159, pruned_loss=0.1687, over 28275.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.391, pruned_loss=0.1337, over 5659551.10 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3621, pruned_loss=0.1052, over 5689333.16 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3937, pruned_loss=0.1368, over 5662588.30 frames. ], batch size: 368, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:33:20,029 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:26,674 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:31,538 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:33:41,933 INFO [optim.py:369] (1/2) 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 03:33:58,665 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=328423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:00,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:02,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:03,266 INFO [train.py:968] (1/2) Epoch 8, batch 9650, giga_loss[loss=0.3229, simple_loss=0.3826, pruned_loss=0.1317, over 28783.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3932, pruned_loss=0.1362, over 5667176.20 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3621, pruned_loss=0.1052, over 5690416.49 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3955, pruned_loss=0.1388, over 5668508.62 frames. ], batch size: 99, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:34:12,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5615, 1.5387, 1.5023, 1.4803], device='cuda:1'), covar=tensor([0.1039, 0.1672, 0.1565, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0727, 0.0651, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 03:34:30,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:35,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3452, 4.1773, 3.9547, 1.8280], device='cuda:1'), covar=tensor([0.0519, 0.0690, 0.0732, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.0972, 0.0929, 0.0817, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 03:34:46,293 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:49,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:34:51,418 INFO [train.py:968] (1/2) Epoch 8, batch 9700, giga_loss[loss=0.3513, simple_loss=0.4079, pruned_loss=0.1474, over 28962.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.395, pruned_loss=0.1389, over 5660006.18 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3622, pruned_loss=0.1053, over 5696715.13 frames. ], giga_tot_loss[loss=0.3409, simple_loss=0.3978, pruned_loss=0.1419, over 5654597.47 frames. ], batch size: 164, lr: 4.13e-03, grad_scale: 8.0
+2023-03-04 03:34:54,004 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7275, 1.7403, 1.5882, 1.6849], device='cuda:1'), covar=tensor([0.1214, 0.1888, 0.1804, 0.1551], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0726, 0.0649, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 03:35:18,270 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:35:37,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.19 vs. limit=2.0
+2023-03-04 03:35:38,898 INFO [train.py:968] (1/2) Epoch 8, batch 9750, giga_loss[loss=0.3296, simple_loss=0.3909, pruned_loss=0.1341, over 28249.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.393, pruned_loss=0.1372, over 5657738.33 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3621, pruned_loss=0.1053, over 5693247.55 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3964, pruned_loss=0.1407, over 5655901.13 frames. ], batch size: 77, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:36:25,091 INFO [train.py:968] (1/2) Epoch 8, batch 9800, giga_loss[loss=0.2782, simple_loss=0.3628, pruned_loss=0.09683, over 28854.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3911, pruned_loss=0.1345, over 5669122.63 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3616, pruned_loss=0.105, over 5696322.35 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3946, pruned_loss=0.138, over 5664455.85 frames. ], batch size: 99, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:36:52,404 INFO [optim.py:369] (1/2) 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 03:37:12,147 INFO [train.py:968] (1/2) Epoch 8, batch 9850, giga_loss[loss=0.3993, simple_loss=0.4288, pruned_loss=0.1849, over 26703.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3921, pruned_loss=0.1339, over 5670161.12 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3616, pruned_loss=0.105, over 5698420.84 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3952, pruned_loss=0.1369, over 5664438.36 frames. ], batch size: 555, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:38:00,400 INFO [train.py:968] (1/2) Epoch 8, batch 9900, libri_loss[loss=0.2546, simple_loss=0.3242, pruned_loss=0.09247, over 29384.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3926, pruned_loss=0.134, over 5673569.32 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3614, pruned_loss=0.1048, over 5701597.71 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3958, pruned_loss=0.137, over 5665555.18 frames. ], batch size: 71, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:38:13,868 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 03:38:22,587 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-04 03:38:29,289 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 9950, giga_loss[loss=0.301, simple_loss=0.3653, pruned_loss=0.1184, over 28711.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3937, pruned_loss=0.1362, over 5664620.29 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3613, pruned_loss=0.1048, over 5703366.09 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3968, pruned_loss=0.1391, over 5656023.25 frames. ], batch size: 262, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:39:04,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 03:39:20,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:39:26,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:39:32,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:39:36,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5684, 1.4608, 1.2294, 1.2369], device='cuda:1'), covar=tensor([0.0464, 0.0321, 0.0720, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0447, 0.0493, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 03:39:37,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0950, 1.0839, 4.0475, 3.3570], device='cuda:1'), covar=tensor([0.1709, 0.2535, 0.0382, 0.0653], device='cuda:1'), in_proj_covar=tensor([0.0611, 0.0568, 0.0816, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 03:39:40,613 INFO [train.py:968] (1/2) Epoch 8, batch 10000, giga_loss[loss=0.3186, simple_loss=0.3814, pruned_loss=0.1279, over 28580.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3907, pruned_loss=0.1348, over 5662935.81 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3612, pruned_loss=0.1048, over 5706531.03 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3938, pruned_loss=0.1377, over 5652752.42 frames. ], batch size: 307, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:40:00,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=328798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:40:10,487 INFO [optim.py:369] (1/2) 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 03:40:31,586 INFO [train.py:968] (1/2) Epoch 8, batch 10050, giga_loss[loss=0.3112, simple_loss=0.3732, pruned_loss=0.1246, over 28959.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3886, pruned_loss=0.1346, over 5655192.80 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3613, pruned_loss=0.1049, over 5702580.32 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3916, pruned_loss=0.1375, over 5649264.50 frames. ], batch size: 227, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:41:12,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-04 03:41:17,750 INFO [train.py:968] (1/2) Epoch 8, batch 10100, giga_loss[loss=0.2973, simple_loss=0.3572, pruned_loss=0.1186, over 28974.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3855, pruned_loss=0.1331, over 5665938.18 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3612, pruned_loss=0.1049, over 5708406.80 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3888, pruned_loss=0.1363, over 5654935.19 frames. ], batch size: 200, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:41:43,199 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:43,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 03:41:46,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:48,493 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:41:51,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:54,246 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:41:59,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:12,964 INFO [train.py:968] (1/2) Epoch 8, batch 10150, libri_loss[loss=0.3, simple_loss=0.3771, pruned_loss=0.1114, over 29155.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3839, pruned_loss=0.1329, over 5656024.16 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3613, pruned_loss=0.1049, over 5713240.38 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.387, pruned_loss=0.1361, over 5641715.92 frames. ], batch size: 101, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:42:17,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:21,023 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:22,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=328941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:27,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=328944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:28,468 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:49,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7384, 3.5565, 3.3745, 1.7565], device='cuda:1'), covar=tensor([0.0703, 0.0799, 0.0803, 0.2060], device='cuda:1'), in_proj_covar=tensor([0.0982, 0.0936, 0.0824, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 03:42:54,917 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=328973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:42:58,966 INFO [train.py:968] (1/2) Epoch 8, batch 10200, libri_loss[loss=0.3232, simple_loss=0.3905, pruned_loss=0.1279, over 29748.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3838, pruned_loss=0.1331, over 5674004.39 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3621, pruned_loss=0.1055, over 5721239.87 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3864, pruned_loss=0.1361, over 5653278.49 frames. ], batch size: 87, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:43:24,648 INFO [optim.py:369] (1/2) 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 03:43:29,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 03:43:29,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0890, 1.2007, 1.2930, 1.0494], device='cuda:1'), covar=tensor([0.0957, 0.1000, 0.1426, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0737, 0.0655, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 03:43:35,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7842, 1.7426, 1.2649, 1.4415], device='cuda:1'), covar=tensor([0.0689, 0.0665, 0.0996, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0338, 0.0444, 0.0490, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 03:43:43,821 INFO [train.py:968] (1/2) Epoch 8, batch 10250, giga_loss[loss=0.2621, simple_loss=0.3465, pruned_loss=0.08889, over 29022.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3825, pruned_loss=0.1321, over 5670122.75 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.362, pruned_loss=0.1054, over 5725566.83 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3852, pruned_loss=0.1352, over 5648604.28 frames. ], batch size: 155, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:44:30,010 INFO [train.py:968] (1/2) Epoch 8, batch 10300, giga_loss[loss=0.3167, simple_loss=0.3839, pruned_loss=0.1247, over 29030.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3784, pruned_loss=0.1274, over 5668593.07 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3619, pruned_loss=0.1053, over 5728451.04 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3809, pruned_loss=0.1304, over 5648019.54 frames. ], batch size: 155, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:45:00,876 INFO [optim.py:369] (1/2) 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 03:45:15,807 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:45:22,559 INFO [train.py:968] (1/2) Epoch 8, batch 10350, giga_loss[loss=0.2882, simple_loss=0.3737, pruned_loss=0.1013, over 28898.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3776, pruned_loss=0.126, over 5673517.60 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3623, pruned_loss=0.1056, over 5731027.94 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3794, pruned_loss=0.1285, over 5654371.09 frames. ], batch size: 174, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:45:33,472 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:45:55,672 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 03:46:08,471 INFO [train.py:968] (1/2) Epoch 8, batch 10400, giga_loss[loss=0.3032, simple_loss=0.3622, pruned_loss=0.1221, over 28933.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3773, pruned_loss=0.1257, over 5678629.05 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3629, pruned_loss=0.1057, over 5734758.76 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3788, pruned_loss=0.1282, over 5657925.49 frames. ], batch size: 227, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:46:39,649 INFO [optim.py:369] (1/2) 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 03:46:53,638 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 10450, giga_loss[loss=0.2925, simple_loss=0.356, pruned_loss=0.1144, over 28770.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3747, pruned_loss=0.1256, over 5676971.59 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3629, pruned_loss=0.1056, over 5736312.78 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3759, pruned_loss=0.1278, over 5658886.41 frames. ], batch size: 243, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:47:04,930 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7756, 1.7606, 1.8979, 1.5842], device='cuda:1'), covar=tensor([0.1238, 0.1681, 0.1463, 0.1510], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0727, 0.0649, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 03:47:18,551 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 03:47:26,124 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:47:48,430 INFO [train.py:968] (1/2) Epoch 8, batch 10500, libri_loss[loss=0.2833, simple_loss=0.3518, pruned_loss=0.1075, over 29347.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3755, pruned_loss=0.1265, over 5666833.62 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3629, pruned_loss=0.1057, over 5730367.63 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3767, pruned_loss=0.1287, over 5656045.21 frames. ], batch size: 71, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:48:15,653 INFO [optim.py:369] (1/2) 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 03:48:36,338 INFO [train.py:968] (1/2) Epoch 8, batch 10550, giga_loss[loss=0.3107, simple_loss=0.3552, pruned_loss=0.1331, over 23554.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3789, pruned_loss=0.1284, over 5665251.79 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3631, pruned_loss=0.1058, over 5731853.43 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3799, pruned_loss=0.1303, over 5654695.96 frames. ], batch size: 705, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:49:27,736 INFO [train.py:968] (1/2) Epoch 8, batch 10600, giga_loss[loss=0.2416, simple_loss=0.3218, pruned_loss=0.08076, over 28544.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3793, pruned_loss=0.1289, over 5650845.76 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3631, pruned_loss=0.1058, over 5723970.45 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1305, over 5649680.82 frames. ], batch size: 60, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:49:57,510 INFO [optim.py:369] (1/2) 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 03:50:16,480 INFO [train.py:968] (1/2) Epoch 8, batch 10650, giga_loss[loss=0.2851, simple_loss=0.3548, pruned_loss=0.1077, over 28939.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3779, pruned_loss=0.1276, over 5655487.64 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3628, pruned_loss=0.1055, over 5727896.51 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3792, pruned_loss=0.1296, over 5649620.94 frames. ], batch size: 106, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:50:53,997 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:51:01,101 INFO [train.py:968] (1/2) Epoch 8, batch 10700, giga_loss[loss=0.4122, simple_loss=0.4352, pruned_loss=0.1946, over 26469.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3796, pruned_loss=0.1297, over 5656651.77 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.363, pruned_loss=0.1057, over 5731446.02 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3807, pruned_loss=0.1316, over 5647604.23 frames. ], batch size: 555, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:51:23,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:51:34,191 INFO [optim.py:369] (1/2) 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 03:51:44,304 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 10750, giga_loss[loss=0.3519, simple_loss=0.4054, pruned_loss=0.1492, over 28354.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.383, pruned_loss=0.132, over 5656777.22 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3631, pruned_loss=0.1057, over 5734061.75 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3841, pruned_loss=0.1339, over 5646115.74 frames. ], batch size: 368, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:52:43,775 INFO [train.py:968] (1/2) Epoch 8, batch 10800, giga_loss[loss=0.2867, simple_loss=0.3602, pruned_loss=0.1066, over 28831.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3854, pruned_loss=0.1335, over 5664505.09 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3632, pruned_loss=0.1057, over 5735425.43 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3863, pruned_loss=0.1351, over 5654286.48 frames. ], batch size: 174, lr: 4.12e-03, grad_scale: 8.0
+2023-03-04 03:52:59,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:11,857 INFO [optim.py:369] (1/2) 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 03:53:31,604 INFO [train.py:968] (1/2) Epoch 8, batch 10850, giga_loss[loss=0.3281, simple_loss=0.3958, pruned_loss=0.1302, over 28864.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3866, pruned_loss=0.1346, over 5677808.29 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3629, pruned_loss=0.1056, over 5739542.65 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3881, pruned_loss=0.1366, over 5664683.30 frames. ], batch size: 227, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:53:33,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:45,805 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:47,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:53:59,087 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:04,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:06,188 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:06,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:16,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=329673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:54:20,934 INFO [train.py:968] (1/2) Epoch 8, batch 10900, giga_loss[loss=0.299, simple_loss=0.367, pruned_loss=0.1155, over 28954.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3874, pruned_loss=0.1353, over 5673166.25 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3637, pruned_loss=0.106, over 5734524.41 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3887, pruned_loss=0.1376, over 5664320.41 frames. ], batch size: 145, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:54:32,592 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:55:09,747 INFO [train.py:968] (1/2) Epoch 8, batch 10950, giga_loss[loss=0.3474, simple_loss=0.3825, pruned_loss=0.1562, over 23864.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.388, pruned_loss=0.1344, over 5666835.79 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3633, pruned_loss=0.1059, over 5739041.50 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3901, pruned_loss=0.1373, over 5653358.56 frames. ], batch size: 705, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:55:14,596 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329733.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 03:55:20,303 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:24,720 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=329771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:54,718 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:55,743 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:56,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:55:59,173 INFO [train.py:968] (1/2) Epoch 8, batch 11000, giga_loss[loss=0.3363, simple_loss=0.389, pruned_loss=0.1418, over 28829.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3876, pruned_loss=0.1342, over 5658492.82 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3635, pruned_loss=0.106, over 5733597.89 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.39, pruned_loss=0.1372, over 5649645.16 frames. ], batch size: 119, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:56:25,617 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=329804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:56:30,732 INFO [optim.py:369] (1/2) 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 03:56:52,376 INFO [train.py:968] (1/2) Epoch 8, batch 11050, libri_loss[loss=0.2722, simple_loss=0.351, pruned_loss=0.09668, over 29539.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3873, pruned_loss=0.135, over 5651253.04 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3631, pruned_loss=0.1057, over 5736310.83 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3901, pruned_loss=0.1383, over 5640244.71 frames. ], batch size: 80, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:57:10,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=329845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:57:50,311 INFO [train.py:968] (1/2) Epoch 8, batch 11100, giga_loss[loss=0.2674, simple_loss=0.3356, pruned_loss=0.09963, over 28707.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3861, pruned_loss=0.1348, over 5643688.09 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3632, pruned_loss=0.1059, over 5733572.00 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3885, pruned_loss=0.1377, over 5636229.90 frames. ], batch size: 92, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:58:05,406 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:58:09,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-04 03:58:13,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2795, 1.5183, 1.2897, 1.0966], device='cuda:1'), covar=tensor([0.1355, 0.1195, 0.0760, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1452, 0.1427, 0.1530], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 03:58:19,888 INFO [optim.py:369] (1/2) 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 03:58:20,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3321, 1.5882, 1.2786, 1.2927], device='cuda:1'), covar=tensor([0.1777, 0.1688, 0.1729, 0.1638], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.0914, 0.1068, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 03:58:35,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2573, 1.4343, 1.2611, 1.4052], device='cuda:1'), covar=tensor([0.0717, 0.0344, 0.0313, 0.0748], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0119, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 03:58:36,590 INFO [train.py:968] (1/2) Epoch 8, batch 11150, giga_loss[loss=0.2866, simple_loss=0.354, pruned_loss=0.1096, over 28869.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3837, pruned_loss=0.1333, over 5651271.32 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3631, pruned_loss=0.1058, over 5738018.22 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3862, pruned_loss=0.1363, over 5639193.12 frames. ], batch size: 186, lr: 4.12e-03, grad_scale: 2.0
+2023-03-04 03:58:42,797 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:58:55,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=329947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:59:17,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6183, 1.9849, 1.9969, 1.5316], device='cuda:1'), covar=tensor([0.1647, 0.1916, 0.1205, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0722, 0.0827, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 03:59:23,831 INFO [train.py:968] (1/2) Epoch 8, batch 11200, giga_loss[loss=0.3376, simple_loss=0.3884, pruned_loss=0.1434, over 28303.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3826, pruned_loss=0.1328, over 5656474.43 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3631, pruned_loss=0.1057, over 5740883.78 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3849, pruned_loss=0.1358, over 5642867.87 frames. ], batch size: 368, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 03:59:34,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=329988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:59:37,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=329991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 03:59:55,837 INFO [optim.py:369] (1/2) 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 03:59:57,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2157, 1.0580, 4.1331, 3.2725], device='cuda:1'), covar=tensor([0.1702, 0.2587, 0.0390, 0.0703], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0566, 0.0815, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 04:00:04,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:00:10,874 INFO [train.py:968] (1/2) Epoch 8, batch 11250, giga_loss[loss=0.3135, simple_loss=0.3743, pruned_loss=0.1264, over 29050.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3818, pruned_loss=0.1325, over 5660543.67 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3633, pruned_loss=0.1057, over 5742564.84 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3838, pruned_loss=0.1353, over 5646528.31 frames. ], batch size: 106, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 04:00:15,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:00:20,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:00:46,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3159, 2.0035, 1.4282, 0.5339], device='cuda:1'), covar=tensor([0.2890, 0.1580, 0.2476, 0.3308], device='cuda:1'), in_proj_covar=tensor([0.1495, 0.1411, 0.1453, 0.1215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 04:01:00,978 INFO [train.py:968] (1/2) Epoch 8, batch 11300, giga_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1206, over 28533.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3823, pruned_loss=0.1331, over 5662614.30 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3632, pruned_loss=0.1056, over 5744335.08 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3842, pruned_loss=0.1357, over 5649311.73 frames. ], batch size: 336, lr: 4.12e-03, grad_scale: 4.0
+2023-03-04 04:01:31,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330108.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 04:01:32,941 INFO [optim.py:369] (1/2) 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 04:01:50,527 INFO [train.py:968] (1/2) Epoch 8, batch 11350, libri_loss[loss=0.34, simple_loss=0.4019, pruned_loss=0.139, over 29642.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3848, pruned_loss=0.1354, over 5663157.82 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3634, pruned_loss=0.1059, over 5748223.27 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3866, pruned_loss=0.1379, over 5646903.21 frames. ], batch size: 91, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:02:12,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:38,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:40,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:42,786 INFO [train.py:968] (1/2) Epoch 8, batch 11400, giga_loss[loss=0.3884, simple_loss=0.4203, pruned_loss=0.1783, over 28603.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3866, pruned_loss=0.1373, over 5652074.94 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3637, pruned_loss=0.1061, over 5749820.98 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3879, pruned_loss=0.1394, over 5637017.14 frames. ], batch size: 307, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:02:44,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:02:47,247 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:03:09,404 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:03:15,418 INFO [optim.py:369] (1/2) 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 04:03:17,268 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:03:34,946 INFO [train.py:968] (1/2) Epoch 8, batch 11450, giga_loss[loss=0.283, simple_loss=0.3573, pruned_loss=0.1043, over 28477.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3873, pruned_loss=0.1384, over 5647400.77 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3642, pruned_loss=0.1064, over 5741978.08 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3882, pruned_loss=0.1401, over 5641013.04 frames. ], batch size: 65, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:03:54,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330251.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 04:03:57,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330254.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 04:04:01,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4298, 1.7473, 1.8123, 1.4463], device='cuda:1'), covar=tensor([0.1030, 0.1478, 0.0836, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0723, 0.0826, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 04:04:02,109 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=330261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:04:10,247 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 11500, giga_loss[loss=0.3317, simple_loss=0.394, pruned_loss=0.1347, over 28936.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3865, pruned_loss=0.1372, over 5656936.71 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3643, pruned_loss=0.1065, over 5744370.11 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3877, pruned_loss=0.1394, over 5647444.48 frames. ], batch size: 136, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:04:21,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7973, 4.6200, 4.3275, 2.0815], device='cuda:1'), covar=tensor([0.0526, 0.0710, 0.0883, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.0971, 0.0931, 0.0824, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 04:04:24,127 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330283.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 04:04:33,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:04:36,455 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:04:52,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:04:56,733 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.98 vs. limit=5.0
+2023-03-04 04:05:00,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:05:02,271 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:05:06,350 INFO [train.py:968] (1/2) Epoch 8, batch 11550, libri_loss[loss=0.4291, simple_loss=0.4605, pruned_loss=0.1988, over 19162.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3882, pruned_loss=0.1384, over 5636039.64 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3647, pruned_loss=0.1068, over 5727797.15 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3895, pruned_loss=0.1407, over 5641660.43 frames. ], batch size: 187, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:05:10,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7419, 1.7969, 1.2587, 1.4664], device='cuda:1'), covar=tensor([0.0793, 0.0649, 0.1090, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0450, 0.0499, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 04:05:51,724 INFO [train.py:968] (1/2) Epoch 8, batch 11600, giga_loss[loss=0.2924, simple_loss=0.3657, pruned_loss=0.1096, over 28886.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3874, pruned_loss=0.1369, over 5656665.95 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3645, pruned_loss=0.1066, over 5729608.52 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3891, pruned_loss=0.1395, over 5657908.44 frames. ], batch size: 174, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:05:57,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 04:06:00,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=330386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:06:18,775 INFO [optim.py:369] (1/2) 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 04:06:19,874 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:06:25,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:06:40,258 INFO [train.py:968] (1/2) Epoch 8, batch 11650, giga_loss[loss=0.3356, simple_loss=0.3927, pruned_loss=0.1392, over 28741.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3885, pruned_loss=0.1381, over 5648130.41 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3646, pruned_loss=0.1067, over 5732991.05 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3906, pruned_loss=0.1412, over 5643602.42 frames. ], batch size: 284, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:06:52,323 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:02,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:04,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:13,081 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:16,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:28,097 INFO [train.py:968] (1/2) Epoch 8, batch 11700, giga_loss[loss=0.315, simple_loss=0.3766, pruned_loss=0.1266, over 28658.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3897, pruned_loss=0.1386, over 5652195.63 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3647, pruned_loss=0.1067, over 5735003.54 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3923, pruned_loss=0.1424, over 5643960.21 frames. ], batch size: 307, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:07:33,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7615, 1.7299, 1.2219, 1.3502], device='cuda:1'), covar=tensor([0.0710, 0.0591, 0.1033, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0448, 0.0497, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 04:07:35,546 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:44,778 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:07:59,088 INFO [optim.py:369] (1/2) 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 04:08:14,296 INFO [train.py:968] (1/2) Epoch 8, batch 11750, giga_loss[loss=0.3727, simple_loss=0.3961, pruned_loss=0.1747, over 23474.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3898, pruned_loss=0.1397, over 5650889.21 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3642, pruned_loss=0.1064, over 5737850.57 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3926, pruned_loss=0.1434, over 5640663.84 frames. ], batch size: 705, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:08:59,251 INFO [train.py:968] (1/2) Epoch 8, batch 11800, giga_loss[loss=0.341, simple_loss=0.4001, pruned_loss=0.141, over 28290.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.389, pruned_loss=0.1377, over 5657529.78 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3636, pruned_loss=0.1061, over 5740321.41 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3926, pruned_loss=0.1421, over 5644510.71 frames. ], batch size: 368, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:09:33,352 INFO [optim.py:369] (1/2) 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 04:09:49,292 INFO [train.py:968] (1/2) Epoch 8, batch 11850, giga_loss[loss=0.3152, simple_loss=0.376, pruned_loss=0.1272, over 28514.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3879, pruned_loss=0.1359, over 5649733.97 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3635, pruned_loss=0.106, over 5733541.65 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3912, pruned_loss=0.1398, over 5644999.57 frames. ], batch size: 71, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:09:56,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:10:36,083 INFO [train.py:968] (1/2) Epoch 8, batch 11900, giga_loss[loss=0.2915, simple_loss=0.3616, pruned_loss=0.1107, over 28295.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3874, pruned_loss=0.1352, over 5650254.06 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3637, pruned_loss=0.1062, over 5731982.88 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3905, pruned_loss=0.139, over 5645343.15 frames. ], batch size: 71, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:11:07,971 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 11950, giga_loss[loss=0.3444, simple_loss=0.3997, pruned_loss=0.1445, over 28877.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3853, pruned_loss=0.1342, over 5653350.58 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3633, pruned_loss=0.106, over 5734516.07 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3884, pruned_loss=0.1377, over 5646106.08 frames. ], batch size: 145, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:11:57,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=330761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:12,503 INFO [train.py:968] (1/2) Epoch 8, batch 12000, giga_loss[loss=0.3254, simple_loss=0.3918, pruned_loss=0.1295, over 28877.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3862, pruned_loss=0.1343, over 5662267.91 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3635, pruned_loss=0.1059, over 5738798.41 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.389, pruned_loss=0.1378, over 5650977.74 frames. ], batch size: 186, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:12:12,504 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 04:12:21,416 INFO [train.py:1012] (1/2) Epoch 8, validation: loss=0.2256, simple_loss=0.3306, pruned_loss=0.06027, over 944034.00 frames. 
+2023-03-04 04:12:21,416 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 04:12:21,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:23,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:50,447 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:12:51,882 INFO [optim.py:369] (1/2) 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 04:13:08,724 INFO [train.py:968] (1/2) Epoch 8, batch 12050, giga_loss[loss=0.3965, simple_loss=0.4152, pruned_loss=0.1889, over 23640.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3867, pruned_loss=0.1349, over 5655469.05 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3634, pruned_loss=0.1058, over 5742168.30 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3895, pruned_loss=0.1383, over 5641944.38 frames. ], batch size: 705, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:13:53,487 INFO [train.py:968] (1/2) Epoch 8, batch 12100, giga_loss[loss=0.3929, simple_loss=0.4271, pruned_loss=0.1793, over 28748.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3848, pruned_loss=0.1339, over 5678733.43 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.363, pruned_loss=0.1056, over 5747355.25 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3881, pruned_loss=0.1378, over 5660712.40 frames. ], batch size: 284, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:14:20,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=330904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:14:22,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=330907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:14:28,057 INFO [optim.py:369] (1/2) 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 04:14:37,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=330924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:14:41,452 INFO [train.py:968] (1/2) Epoch 8, batch 12150, giga_loss[loss=0.3062, simple_loss=0.3731, pruned_loss=0.1196, over 28910.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3852, pruned_loss=0.1345, over 5675105.38 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3633, pruned_loss=0.1057, over 5748025.02 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3878, pruned_loss=0.1379, over 5659179.71 frames. ], batch size: 227, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:14:51,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=330936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:15:30,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-04 04:15:30,220 INFO [train.py:968] (1/2) Epoch 8, batch 12200, giga_loss[loss=0.3198, simple_loss=0.387, pruned_loss=0.1263, over 28919.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.388, pruned_loss=0.1368, over 5673561.30 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3637, pruned_loss=0.106, over 5747219.88 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3902, pruned_loss=0.1398, over 5660476.57 frames. ], batch size: 164, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:16:00,972 INFO [optim.py:369] (1/2) 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 04:16:13,873 INFO [train.py:968] (1/2) Epoch 8, batch 12250, giga_loss[loss=0.2839, simple_loss=0.3613, pruned_loss=0.1032, over 29066.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3868, pruned_loss=0.1359, over 5670239.11 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.363, pruned_loss=0.1057, over 5748109.41 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3902, pruned_loss=0.1399, over 5656051.88 frames. ], batch size: 128, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:17:04,082 INFO [train.py:968] (1/2) Epoch 8, batch 12300, giga_loss[loss=0.3222, simple_loss=0.3823, pruned_loss=0.131, over 28226.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3868, pruned_loss=0.1353, over 5680859.87 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3627, pruned_loss=0.1053, over 5750213.46 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3902, pruned_loss=0.1392, over 5666484.44 frames. ], batch size: 77, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:17:37,673 INFO [optim.py:369] (1/2) 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 04:17:52,735 INFO [train.py:968] (1/2) Epoch 8, batch 12350, libri_loss[loss=0.3226, simple_loss=0.3983, pruned_loss=0.1234, over 29375.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3862, pruned_loss=0.1346, over 5671602.79 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3628, pruned_loss=0.1053, over 5754936.41 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3895, pruned_loss=0.1387, over 5653593.22 frames. ], batch size: 92, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:18:25,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5645, 1.6246, 1.6807, 1.4705], device='cuda:1'), covar=tensor([0.1300, 0.1709, 0.1714, 0.1531], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0732, 0.0657, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 04:18:34,585 INFO [train.py:968] (1/2) Epoch 8, batch 12400, giga_loss[loss=0.3678, simple_loss=0.4244, pruned_loss=0.1557, over 28858.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3859, pruned_loss=0.1338, over 5674015.32 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3623, pruned_loss=0.1051, over 5747172.48 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3896, pruned_loss=0.138, over 5663796.96 frames. ], batch size: 199, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:18:35,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 04:18:43,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0105, 1.9449, 1.3868, 1.6848], device='cuda:1'), covar=tensor([0.0674, 0.0542, 0.0950, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0449, 0.0496, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 04:18:55,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9008, 1.6792, 1.5535, 1.1745], device='cuda:1'), covar=tensor([0.0780, 0.0238, 0.0238, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0119, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 04:19:03,256 INFO [optim.py:369] (1/2) 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 04:19:18,378 INFO [train.py:968] (1/2) Epoch 8, batch 12450, giga_loss[loss=0.3282, simple_loss=0.3845, pruned_loss=0.1359, over 28230.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3851, pruned_loss=0.1325, over 5683285.37 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3622, pruned_loss=0.105, over 5744266.21 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3894, pruned_loss=0.1374, over 5673890.14 frames. ], batch size: 368, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:19:24,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 04:20:01,552 INFO [train.py:968] (1/2) Epoch 8, batch 12500, giga_loss[loss=0.3803, simple_loss=0.428, pruned_loss=0.1663, over 28858.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.384, pruned_loss=0.1324, over 5668623.14 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3621, pruned_loss=0.1049, over 5739025.27 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3884, pruned_loss=0.1374, over 5663133.92 frames. ], batch size: 145, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:20:21,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=331299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:20:38,942 INFO [optim.py:369] (1/2) 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 04:20:50,299 INFO [train.py:968] (1/2) Epoch 8, batch 12550, giga_loss[loss=0.3215, simple_loss=0.383, pruned_loss=0.13, over 28865.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3821, pruned_loss=0.1316, over 5667233.97 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3619, pruned_loss=0.1049, over 5738923.08 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3863, pruned_loss=0.1364, over 5661266.65 frames. ], batch size: 227, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:21:38,710 INFO [train.py:968] (1/2) Epoch 8, batch 12600, giga_loss[loss=0.2577, simple_loss=0.3292, pruned_loss=0.09313, over 28809.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3786, pruned_loss=0.1299, over 5674891.08 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3624, pruned_loss=0.1052, over 5739458.69 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3818, pruned_loss=0.1338, over 5668807.57 frames. ], batch size: 112, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:21:44,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0057, 2.4847, 1.0292, 1.2458], device='cuda:1'), covar=tensor([0.1201, 0.0485, 0.1006, 0.1542], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0503, 0.0327, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 04:22:14,360 INFO [optim.py:369] (1/2) 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 04:22:17,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5669, 1.9128, 1.9248, 1.4461], device='cuda:1'), covar=tensor([0.1394, 0.1928, 0.1122, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0721, 0.0826, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 04:22:25,984 INFO [train.py:968] (1/2) Epoch 8, batch 12650, giga_loss[loss=0.2738, simple_loss=0.3424, pruned_loss=0.1026, over 28831.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3774, pruned_loss=0.1299, over 5678844.31 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3628, pruned_loss=0.1056, over 5738341.87 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3801, pruned_loss=0.1334, over 5672971.70 frames. ], batch size: 119, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:22:40,377 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=331442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:22:42,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=331445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:23:12,942 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 12700, giga_loss[loss=0.2656, simple_loss=0.3387, pruned_loss=0.09629, over 28956.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3769, pruned_loss=0.1302, over 5686734.77 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3627, pruned_loss=0.1055, over 5739225.16 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3791, pruned_loss=0.1332, over 5681199.54 frames. ], batch size: 136, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:23:39,797 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-03-04 04:23:54,148 INFO [optim.py:369] (1/2) 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 04:24:07,904 INFO [train.py:968] (1/2) Epoch 8, batch 12750, giga_loss[loss=0.3038, simple_loss=0.3566, pruned_loss=0.1255, over 28595.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3787, pruned_loss=0.1306, over 5683738.24 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3631, pruned_loss=0.1056, over 5739961.77 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3804, pruned_loss=0.1331, over 5677955.29 frames. ], batch size: 78, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:24:08,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3678, 1.7248, 1.7075, 1.3234], device='cuda:1'), covar=tensor([0.1550, 0.2112, 0.1280, 0.1492], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0718, 0.0824, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 04:24:24,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 04:24:51,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3684, 4.1897, 3.9605, 2.0816], device='cuda:1'), covar=tensor([0.0498, 0.0685, 0.0756, 0.1901], device='cuda:1'), in_proj_covar=tensor([0.0973, 0.0924, 0.0815, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 04:24:55,115 INFO [train.py:968] (1/2) Epoch 8, batch 12800, giga_loss[loss=0.3099, simple_loss=0.3826, pruned_loss=0.1187, over 28859.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3764, pruned_loss=0.1269, over 5687118.07 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3622, pruned_loss=0.1053, over 5744884.99 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.379, pruned_loss=0.13, over 5676057.03 frames. ], batch size: 145, lr: 4.11e-03, grad_scale: 8.0
+2023-03-04 04:25:13,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-04 04:25:15,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3359, 1.8916, 1.3606, 1.5244], device='cuda:1'), covar=tensor([0.0708, 0.0384, 0.0334, 0.0738], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0116, 0.0119, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 04:25:34,454 INFO [optim.py:369] (1/2) 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 04:25:38,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 04:25:48,237 INFO [train.py:968] (1/2) Epoch 8, batch 12850, giga_loss[loss=0.3179, simple_loss=0.3769, pruned_loss=0.1294, over 27505.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3729, pruned_loss=0.1233, over 5671488.29 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3615, pruned_loss=0.105, over 5744181.93 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.376, pruned_loss=0.1266, over 5661859.60 frames. ], batch size: 472, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:26:11,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3785, 2.9870, 1.3770, 1.4142], device='cuda:1'), covar=tensor([0.0869, 0.0313, 0.0904, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0502, 0.0326, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 04:26:33,824 INFO [train.py:968] (1/2) Epoch 8, batch 12900, giga_loss[loss=0.28, simple_loss=0.3543, pruned_loss=0.1029, over 28004.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 5658066.80 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3612, pruned_loss=0.1049, over 5731619.84 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3728, pruned_loss=0.1234, over 5658718.58 frames. ], batch size: 412, lr: 4.11e-03, grad_scale: 4.0
+2023-03-04 04:26:42,488 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-04 04:26:51,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=331695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:27:12,646 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 12950, giga_loss[loss=0.2877, simple_loss=0.3554, pruned_loss=0.11, over 28707.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3662, pruned_loss=0.1169, over 5663256.70 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3612, pruned_loss=0.1053, over 5735968.85 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 5658141.44 frames. ], batch size: 262, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:28:14,333 INFO [train.py:968] (1/2) Epoch 8, batch 13000, giga_loss[loss=0.272, simple_loss=0.3367, pruned_loss=0.1037, over 26706.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3643, pruned_loss=0.114, over 5667323.57 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3606, pruned_loss=0.1051, over 5740718.23 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3671, pruned_loss=0.1166, over 5656945.16 frames. ], batch size: 555, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:28:51,514 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 13050, giga_loss[loss=0.2586, simple_loss=0.339, pruned_loss=0.08907, over 28980.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3636, pruned_loss=0.1125, over 5660982.64 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3601, pruned_loss=0.1049, over 5741289.84 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3664, pruned_loss=0.1149, over 5650373.62 frames. ], batch size: 106, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:29:41,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4064, 1.7753, 1.4821, 1.4528], device='cuda:1'), covar=tensor([0.0724, 0.0268, 0.0306, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0120, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 04:29:51,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2787, 1.8886, 1.4695, 0.3546], device='cuda:1'), covar=tensor([0.2323, 0.1660, 0.2631, 0.3306], device='cuda:1'), in_proj_covar=tensor([0.1459, 0.1385, 0.1434, 0.1203], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 04:29:53,910 INFO [train.py:968] (1/2) Epoch 8, batch 13100, giga_loss[loss=0.2602, simple_loss=0.3386, pruned_loss=0.09092, over 28991.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3613, pruned_loss=0.1103, over 5670901.43 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3592, pruned_loss=0.1045, over 5746301.43 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3645, pruned_loss=0.1129, over 5655177.88 frames. ], batch size: 136, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:30:26,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4286, 2.0195, 1.5152, 0.5693], device='cuda:1'), covar=tensor([0.2217, 0.1394, 0.2192, 0.3009], device='cuda:1'), in_proj_covar=tensor([0.1453, 0.1379, 0.1427, 0.1200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 04:30:31,231 INFO [optim.py:369] (1/2) 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 04:30:45,729 INFO [train.py:968] (1/2) Epoch 8, batch 13150, giga_loss[loss=0.2411, simple_loss=0.3247, pruned_loss=0.07877, over 28780.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3595, pruned_loss=0.1093, over 5671595.34 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3594, pruned_loss=0.105, over 5750820.44 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.362, pruned_loss=0.1111, over 5652867.23 frames. ], batch size: 243, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:31:35,414 INFO [train.py:968] (1/2) Epoch 8, batch 13200, libri_loss[loss=0.2881, simple_loss=0.3626, pruned_loss=0.1068, over 29531.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3567, pruned_loss=0.1071, over 5678311.18 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.359, pruned_loss=0.1049, over 5752475.78 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.359, pruned_loss=0.1087, over 5659525.46 frames. ], batch size: 84, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:31:41,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3003, 4.1249, 3.8801, 2.0157], device='cuda:1'), covar=tensor([0.0498, 0.0656, 0.0779, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.0947, 0.0900, 0.0791, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 04:32:06,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-04 04:32:10,071 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 13250, giga_loss[loss=0.2599, simple_loss=0.3466, pruned_loss=0.08658, over 28992.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3561, pruned_loss=0.1064, over 5670123.04 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3586, pruned_loss=0.1047, over 5747799.08 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3581, pruned_loss=0.108, over 5657221.10 frames. ], batch size: 155, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:32:40,468 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:33:12,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1957, 1.2063, 3.9887, 3.3801], device='cuda:1'), covar=tensor([0.1548, 0.2500, 0.0370, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0603, 0.0558, 0.0803, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 04:33:12,925 INFO [train.py:968] (1/2) Epoch 8, batch 13300, giga_loss[loss=0.296, simple_loss=0.3676, pruned_loss=0.1122, over 27899.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3563, pruned_loss=0.1066, over 5668653.98 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3584, pruned_loss=0.1047, over 5746532.26 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3581, pruned_loss=0.1078, over 5658329.07 frames. ], batch size: 412, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:33:28,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9589, 1.2387, 1.3046, 1.0854], device='cuda:1'), covar=tensor([0.1248, 0.0940, 0.1596, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0714, 0.0639, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 04:33:49,415 INFO [optim.py:369] (1/2) 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 04:34:01,972 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 13350, giga_loss[loss=0.2438, simple_loss=0.331, pruned_loss=0.07828, over 28566.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3531, pruned_loss=0.1038, over 5674065.25 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3576, pruned_loss=0.1042, over 5751014.35 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3552, pruned_loss=0.1052, over 5659528.83 frames. ], batch size: 307, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:34:54,669 INFO [train.py:968] (1/2) Epoch 8, batch 13400, giga_loss[loss=0.2619, simple_loss=0.3433, pruned_loss=0.09024, over 28975.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3492, pruned_loss=0.101, over 5661583.96 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3574, pruned_loss=0.1042, over 5740792.73 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3509, pruned_loss=0.1021, over 5658246.37 frames. ], batch size: 136, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:35:06,565 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:35:27,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:35:30,359 INFO [optim.py:369] (1/2) 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 04:35:32,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:35:40,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3774, 5.1760, 4.8924, 2.4914], device='cuda:1'), covar=tensor([0.0350, 0.0577, 0.0636, 0.1670], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.0898, 0.0788, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 04:35:44,442 INFO [train.py:968] (1/2) Epoch 8, batch 13450, giga_loss[loss=0.2833, simple_loss=0.3339, pruned_loss=0.1163, over 24013.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3461, pruned_loss=0.09959, over 5655868.01 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3573, pruned_loss=0.1042, over 5743501.85 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3473, pruned_loss=0.1003, over 5647827.33 frames. ], batch size: 705, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:36:05,612 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:36:38,489 INFO [train.py:968] (1/2) Epoch 8, batch 13500, libri_loss[loss=0.2858, simple_loss=0.3578, pruned_loss=0.1069, over 29252.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3457, pruned_loss=0.1002, over 5651881.83 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3574, pruned_loss=0.1043, over 5745235.50 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3462, pruned_loss=0.1006, over 5641570.07 frames. ], batch size: 94, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:36:40,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4475, 3.5174, 1.6801, 1.3800], device='cuda:1'), covar=tensor([0.0883, 0.0260, 0.0828, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0496, 0.0328, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 04:37:07,987 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:37:13,870 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 13550, giga_loss[loss=0.2492, simple_loss=0.33, pruned_loss=0.08424, over 28829.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3466, pruned_loss=0.1015, over 5644542.58 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.357, pruned_loss=0.1042, over 5744025.79 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3471, pruned_loss=0.1018, over 5634998.78 frames. ], batch size: 186, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:38:24,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:38:27,477 INFO [train.py:968] (1/2) Epoch 8, batch 13600, giga_loss[loss=0.2721, simple_loss=0.3565, pruned_loss=0.09384, over 28607.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.348, pruned_loss=0.1016, over 5644342.00 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3561, pruned_loss=0.104, over 5747901.06 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3489, pruned_loss=0.102, over 5631116.06 frames. ], batch size: 242, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:39:14,956 INFO [optim.py:369] (1/2) 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 04:39:18,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4817, 2.4522, 2.3203, 2.0947], device='cuda:1'), covar=tensor([0.1140, 0.1764, 0.1413, 0.1634], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0703, 0.0635, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 04:39:23,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:39:31,648 INFO [train.py:968] (1/2) Epoch 8, batch 13650, giga_loss[loss=0.2709, simple_loss=0.3416, pruned_loss=0.1001, over 28105.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3492, pruned_loss=0.1013, over 5644907.75 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.356, pruned_loss=0.1038, over 5747306.71 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.35, pruned_loss=0.1018, over 5633955.54 frames. ], batch size: 412, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:40:29,298 INFO [train.py:968] (1/2) Epoch 8, batch 13700, libri_loss[loss=0.3093, simple_loss=0.3674, pruned_loss=0.1257, over 29577.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3502, pruned_loss=0.1026, over 5648282.35 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3555, pruned_loss=0.1037, over 5751501.03 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3512, pruned_loss=0.1029, over 5632591.96 frames. ], batch size: 76, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:41:00,192 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:41:30,238 INFO [train.py:968] (1/2) Epoch 8, batch 13750, giga_loss[loss=0.2898, simple_loss=0.364, pruned_loss=0.1078, over 27498.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3473, pruned_loss=0.1002, over 5657083.95 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3555, pruned_loss=0.1036, over 5754014.96 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3479, pruned_loss=0.1005, over 5641131.14 frames. ], batch size: 472, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:41:49,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=332543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:16,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:18,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:19,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:42:32,655 INFO [train.py:968] (1/2) Epoch 8, batch 13800, giga_loss[loss=0.252, simple_loss=0.3383, pruned_loss=0.08285, over 28375.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3456, pruned_loss=0.09799, over 5643986.38 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3552, pruned_loss=0.1035, over 5748472.33 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3462, pruned_loss=0.09826, over 5634433.26 frames. ], batch size: 368, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:42:55,101 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:43:22,424 INFO [optim.py:369] (1/2) 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 04:43:38,683 INFO [train.py:968] (1/2) Epoch 8, batch 13850, giga_loss[loss=0.2294, simple_loss=0.3076, pruned_loss=0.07559, over 28984.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3424, pruned_loss=0.0961, over 5650335.92 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3547, pruned_loss=0.1033, over 5749772.49 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3432, pruned_loss=0.09643, over 5640862.67 frames. ], batch size: 213, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:43:54,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4474, 3.0680, 1.4615, 1.5986], device='cuda:1'), covar=tensor([0.0834, 0.0366, 0.0897, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0491, 0.0328, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 04:43:54,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1881, 1.8916, 1.4407, 0.3240], device='cuda:1'), covar=tensor([0.2067, 0.1228, 0.2128, 0.2471], device='cuda:1'), in_proj_covar=tensor([0.1461, 0.1379, 0.1433, 0.1204], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 04:43:58,988 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:44:01,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:44:18,056 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-04 04:44:28,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8547, 1.7836, 1.4738, 1.5076], device='cuda:1'), covar=tensor([0.0710, 0.0610, 0.0844, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0441, 0.0494, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 04:44:35,559 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:44:36,589 INFO [train.py:968] (1/2) Epoch 8, batch 13900, libri_loss[loss=0.2746, simple_loss=0.3514, pruned_loss=0.09892, over 29468.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3405, pruned_loss=0.09599, over 5659928.45 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3545, pruned_loss=0.1032, over 5755319.61 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3408, pruned_loss=0.09603, over 5643122.32 frames. ], batch size: 85, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:44:43,224 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:45:13,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:45:18,132 INFO [optim.py:369] (1/2) 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 04:45:28,640 INFO [train.py:968] (1/2) Epoch 8, batch 13950, giga_loss[loss=0.2534, simple_loss=0.3273, pruned_loss=0.08975, over 28097.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3413, pruned_loss=0.09732, over 5663786.32 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3542, pruned_loss=0.1035, over 5757185.81 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3412, pruned_loss=0.09688, over 5645186.92 frames. ], batch size: 412, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:45:44,547 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:45:56,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:46:28,611 INFO [train.py:968] (1/2) Epoch 8, batch 14000, giga_loss[loss=0.2687, simple_loss=0.3475, pruned_loss=0.09496, over 29022.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3417, pruned_loss=0.09665, over 5675264.25 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3541, pruned_loss=0.1033, over 5760168.54 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3415, pruned_loss=0.0963, over 5656344.01 frames. ], batch size: 199, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:46:35,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9630, 1.1811, 1.3050, 0.9259], device='cuda:1'), covar=tensor([0.1147, 0.1093, 0.1649, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0703, 0.0632, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 04:47:18,466 INFO [optim.py:369] (1/2) 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 04:47:32,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:47:33,003 INFO [train.py:968] (1/2) Epoch 8, batch 14050, giga_loss[loss=0.2826, simple_loss=0.3508, pruned_loss=0.1072, over 27537.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3449, pruned_loss=0.09736, over 5681406.98 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.354, pruned_loss=0.1033, over 5761448.32 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3447, pruned_loss=0.09708, over 5664702.94 frames. ], batch size: 472, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:47:38,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:48:18,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:48:43,239 INFO [train.py:968] (1/2) Epoch 8, batch 14100, giga_loss[loss=0.2321, simple_loss=0.3154, pruned_loss=0.07442, over 28425.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3422, pruned_loss=0.09535, over 5680718.46 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.354, pruned_loss=0.1033, over 5762931.97 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.342, pruned_loss=0.095, over 5665385.57 frames. ], batch size: 336, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:49:06,057 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=332895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:07,912 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=332898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:32,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6854, 1.8742, 1.5950, 1.4929], device='cuda:1'), covar=tensor([0.1627, 0.1182, 0.1060, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.1548, 0.1376, 0.1337, 0.1453], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 04:49:33,932 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:49:35,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=332918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:38,543 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 04:49:48,717 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=332927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:49:50,075 INFO [train.py:968] (1/2) Epoch 8, batch 14150, giga_loss[loss=0.2913, simple_loss=0.3692, pruned_loss=0.1067, over 28855.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3433, pruned_loss=0.09696, over 5682719.05 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3537, pruned_loss=0.1034, over 5760848.29 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3431, pruned_loss=0.09645, over 5670597.54 frames. ], batch size: 199, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:50:56,091 INFO [train.py:968] (1/2) Epoch 8, batch 14200, giga_loss[loss=0.2683, simple_loss=0.352, pruned_loss=0.09234, over 27721.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3469, pruned_loss=0.09914, over 5667513.76 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1035, over 5765263.74 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3464, pruned_loss=0.09839, over 5651560.08 frames. ], batch size: 472, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:51:45,230 INFO [optim.py:369] (1/2) 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 04:51:59,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2095, 1.3950, 1.1781, 1.0769], device='cuda:1'), covar=tensor([0.1193, 0.1161, 0.0834, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.1561, 0.1386, 0.1348, 0.1467], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 04:51:59,646 INFO [train.py:968] (1/2) Epoch 8, batch 14250, giga_loss[loss=0.2878, simple_loss=0.3752, pruned_loss=0.1002, over 28926.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3498, pruned_loss=0.0979, over 5654972.23 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3537, pruned_loss=0.1037, over 5753898.01 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3493, pruned_loss=0.0971, over 5651537.22 frames. ], batch size: 284, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:52:34,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6655, 4.5108, 4.3123, 1.7820], device='cuda:1'), covar=tensor([0.0429, 0.0546, 0.0642, 0.2161], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0875, 0.0777, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 04:52:40,462 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:52:44,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:53:00,836 INFO [train.py:968] (1/2) Epoch 8, batch 14300, giga_loss[loss=0.2843, simple_loss=0.3636, pruned_loss=0.1025, over 29127.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3479, pruned_loss=0.09578, over 5646952.11 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3533, pruned_loss=0.1035, over 5757722.83 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3478, pruned_loss=0.09508, over 5638089.11 frames. ], batch size: 200, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:53:17,300 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:53:45,429 INFO [optim.py:369] (1/2) 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 04:53:57,776 INFO [train.py:968] (1/2) Epoch 8, batch 14350, giga_loss[loss=0.2552, simple_loss=0.3418, pruned_loss=0.0843, over 28914.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3473, pruned_loss=0.09382, over 5661925.96 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3531, pruned_loss=0.1034, over 5761288.48 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3472, pruned_loss=0.09319, over 5649293.97 frames. ], batch size: 213, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:55:02,944 INFO [train.py:968] (1/2) Epoch 8, batch 14400, giga_loss[loss=0.2302, simple_loss=0.3158, pruned_loss=0.07229, over 28210.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3488, pruned_loss=0.09564, over 5669783.40 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.353, pruned_loss=0.1033, over 5763480.42 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3488, pruned_loss=0.09509, over 5656225.77 frames. ], batch size: 71, lr: 4.10e-03, grad_scale: 8.0
+2023-03-04 04:55:52,727 INFO [optim.py:369] (1/2) 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 04:55:56,873 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 04:56:05,906 INFO [train.py:968] (1/2) Epoch 8, batch 14450, giga_loss[loss=0.2729, simple_loss=0.3457, pruned_loss=0.1001, over 28960.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3477, pruned_loss=0.09646, over 5673792.57 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.353, pruned_loss=0.1032, over 5766007.01 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3476, pruned_loss=0.09593, over 5658998.87 frames. ], batch size: 199, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:56:12,086 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 04:56:21,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6915, 1.6809, 1.4892, 2.0703], device='cuda:1'), covar=tensor([0.2243, 0.2285, 0.2335, 0.2032], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.0900, 0.1063, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 04:57:19,908 INFO [train.py:968] (1/2) Epoch 8, batch 14500, giga_loss[loss=0.2546, simple_loss=0.3367, pruned_loss=0.08625, over 28203.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3493, pruned_loss=0.09808, over 5664720.14 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3521, pruned_loss=0.1027, over 5756806.46 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3499, pruned_loss=0.09795, over 5657787.76 frames. ], batch size: 412, lr: 4.10e-03, grad_scale: 4.0
+2023-03-04 04:57:43,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333292.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 04:58:31,980 INFO [optim.py:369] (1/2) 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 04:58:40,439 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-04 04:58:47,099 INFO [train.py:968] (1/2) Epoch 8, batch 14550, giga_loss[loss=0.2284, simple_loss=0.3129, pruned_loss=0.07202, over 28524.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3445, pruned_loss=0.09506, over 5663757.23 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3521, pruned_loss=0.1028, over 5749232.16 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3449, pruned_loss=0.0948, over 5663371.75 frames. ], batch size: 85, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 04:59:54,697 INFO [train.py:968] (1/2) Epoch 8, batch 14600, giga_loss[loss=0.2578, simple_loss=0.3387, pruned_loss=0.08841, over 28510.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3423, pruned_loss=0.0939, over 5662139.67 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3523, pruned_loss=0.1029, over 5751761.21 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3424, pruned_loss=0.09348, over 5658518.58 frames. ], batch size: 370, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:00:19,726 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333395.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:00:47,564 INFO [optim.py:369] (1/2) 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 05:00:48,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:01:01,185 INFO [train.py:968] (1/2) Epoch 8, batch 14650, giga_loss[loss=0.2546, simple_loss=0.3412, pruned_loss=0.08394, over 28835.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3398, pruned_loss=0.0928, over 5667486.42 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3518, pruned_loss=0.1029, over 5747169.81 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.34, pruned_loss=0.09227, over 5665720.31 frames. ], batch size: 243, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:01:10,382 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:01:15,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:01:44,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333467.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:02:00,776 INFO [train.py:968] (1/2) Epoch 8, batch 14700, giga_loss[loss=0.3304, simple_loss=0.3921, pruned_loss=0.1343, over 28183.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3454, pruned_loss=0.09612, over 5682904.60 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3509, pruned_loss=0.1024, over 5752717.80 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.346, pruned_loss=0.09588, over 5674052.43 frames. ], batch size: 412, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:02:47,162 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 14750, giga_loss[loss=0.2949, simple_loss=0.3668, pruned_loss=0.1115, over 28791.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3465, pruned_loss=0.09763, over 5672859.50 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3507, pruned_loss=0.1024, over 5746534.91 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3472, pruned_loss=0.09732, over 5668903.90 frames. ], batch size: 243, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:03:09,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4227, 1.6548, 1.2290, 1.4076], device='cuda:1'), covar=tensor([0.1308, 0.1068, 0.1055, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.1568, 0.1395, 0.1357, 0.1479], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 05:04:00,603 INFO [train.py:968] (1/2) Epoch 8, batch 14800, giga_loss[loss=0.2744, simple_loss=0.3457, pruned_loss=0.1016, over 27673.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3446, pruned_loss=0.09732, over 5664262.37 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3503, pruned_loss=0.1022, over 5732621.46 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3453, pruned_loss=0.09707, over 5671101.26 frames. ], batch size: 472, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:04:26,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:04:48,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:05:02,232 INFO [train.py:968] (1/2) Epoch 8, batch 14850, giga_loss[loss=0.2221, simple_loss=0.3079, pruned_loss=0.06812, over 29033.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3445, pruned_loss=0.09767, over 5663087.80 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3499, pruned_loss=0.102, over 5733636.92 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3454, pruned_loss=0.09763, over 5665741.25 frames. ], batch size: 155, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:06:06,284 INFO [train.py:968] (1/2) Epoch 8, batch 14900, giga_loss[loss=0.2607, simple_loss=0.3527, pruned_loss=0.0843, over 28981.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.09757, over 5668390.13 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3494, pruned_loss=0.1017, over 5737274.06 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3463, pruned_loss=0.09766, over 5665641.68 frames. ], batch size: 128, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:07:01,268 INFO [optim.py:369] (1/2) 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 05:07:16,141 INFO [train.py:968] (1/2) Epoch 8, batch 14950, giga_loss[loss=0.2586, simple_loss=0.3395, pruned_loss=0.08886, over 28500.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3464, pruned_loss=0.09731, over 5670686.97 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3489, pruned_loss=0.1016, over 5739589.62 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3476, pruned_loss=0.09745, over 5665163.67 frames. ], batch size: 336, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:07:32,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:07:37,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:08:16,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333770.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:08:22,055 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:08:36,515 INFO [train.py:968] (1/2) Epoch 8, batch 15000, giga_loss[loss=0.2265, simple_loss=0.3085, pruned_loss=0.07222, over 29056.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3448, pruned_loss=0.09625, over 5675016.34 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3482, pruned_loss=0.1013, over 5742951.11 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3463, pruned_loss=0.09646, over 5666171.14 frames. ], batch size: 165, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:08:36,515 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 05:08:45,239 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 05:09:08,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:09:14,309 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=333799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:09:39,146 INFO [optim.py:369] (1/2) 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 05:09:50,334 INFO [train.py:968] (1/2) Epoch 8, batch 15050, giga_loss[loss=0.2271, simple_loss=0.298, pruned_loss=0.07807, over 28779.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3406, pruned_loss=0.09485, over 5692473.54 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3478, pruned_loss=0.1011, over 5747045.60 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3421, pruned_loss=0.09511, over 5680238.00 frames. ], batch size: 99, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:10:53,950 INFO [train.py:968] (1/2) Epoch 8, batch 15100, giga_loss[loss=0.2819, simple_loss=0.3457, pruned_loss=0.1091, over 28905.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3347, pruned_loss=0.09181, over 5690352.83 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3476, pruned_loss=0.1008, over 5751414.37 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3357, pruned_loss=0.09206, over 5674971.73 frames. ], batch size: 186, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:11:34,046 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333913.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:11:37,608 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:11:52,413 INFO [train.py:968] (1/2) Epoch 8, batch 15150, giga_loss[loss=0.3431, simple_loss=0.3874, pruned_loss=0.1494, over 26892.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3354, pruned_loss=0.09278, over 5686856.97 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3472, pruned_loss=0.1006, over 5755282.06 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3363, pruned_loss=0.09293, over 5669566.30 frames. ], batch size: 555, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:12:07,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=333938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:12:11,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=333941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:12:14,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333945.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:12:33,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-04 05:12:38,821 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=333970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:12:47,572 INFO [train.py:968] (1/2) Epoch 8, batch 15200, giga_loss[loss=0.2149, simple_loss=0.2968, pruned_loss=0.06655, over 28677.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3356, pruned_loss=0.09285, over 5688361.30 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3466, pruned_loss=0.1002, over 5756190.00 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3364, pruned_loss=0.09309, over 5670844.74 frames. ], batch size: 60, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:12:49,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1689, 1.3013, 3.1825, 3.0590], device='cuda:1'), covar=tensor([0.1434, 0.2357, 0.0474, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0596, 0.0556, 0.0787, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 05:12:59,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=333990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:13:39,447 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 15250, giga_loss[loss=0.2584, simple_loss=0.3351, pruned_loss=0.09082, over 28534.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3336, pruned_loss=0.0914, over 5674381.06 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3463, pruned_loss=0.0999, over 5759316.25 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3343, pruned_loss=0.09167, over 5656159.03 frames. ], batch size: 336, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:14:48,709 INFO [train.py:968] (1/2) Epoch 8, batch 15300, giga_loss[loss=0.2688, simple_loss=0.3413, pruned_loss=0.09814, over 28917.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3335, pruned_loss=0.09077, over 5673712.11 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3466, pruned_loss=0.1002, over 5758679.33 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3335, pruned_loss=0.09052, over 5657829.68 frames. ], batch size: 199, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:15:15,151 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-04 05:15:34,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3428, 3.2096, 1.5081, 1.4727], device='cuda:1'), covar=tensor([0.0907, 0.0348, 0.0849, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0484, 0.0324, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 05:15:49,813 INFO [optim.py:369] (1/2) 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 05:15:59,785 INFO [train.py:968] (1/2) Epoch 8, batch 15350, libri_loss[loss=0.2799, simple_loss=0.3563, pruned_loss=0.1018, over 28550.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.332, pruned_loss=0.09018, over 5671732.73 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3465, pruned_loss=0.1001, over 5758115.32 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3318, pruned_loss=0.08989, over 5658418.84 frames. ], batch size: 106, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:16:06,461 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=334133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:16:11,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=334136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:16:49,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=334165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:17:01,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=334174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:17:05,642 INFO [train.py:968] (1/2) Epoch 8, batch 15400, giga_loss[loss=0.2581, simple_loss=0.3379, pruned_loss=0.08922, over 28648.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.333, pruned_loss=0.09015, over 5685965.53 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3465, pruned_loss=0.1001, over 5760335.63 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3325, pruned_loss=0.08972, over 5671918.69 frames. ], batch size: 307, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:18:00,927 INFO [optim.py:369] (1/2) 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 05:18:04,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 05:18:13,367 INFO [train.py:968] (1/2) Epoch 8, batch 15450, giga_loss[loss=0.2576, simple_loss=0.3337, pruned_loss=0.09068, over 28911.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3331, pruned_loss=0.08995, over 5694346.65 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3464, pruned_loss=0.1001, over 5762746.64 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3327, pruned_loss=0.08954, over 5680095.36 frames. ], batch size: 284, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:18:13,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3296, 1.9166, 1.4515, 0.3646], device='cuda:1'), covar=tensor([0.2086, 0.1629, 0.2719, 0.3015], device='cuda:1'), in_proj_covar=tensor([0.1466, 0.1396, 0.1456, 0.1209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 05:18:50,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3543, 3.0370, 1.4925, 1.4186], device='cuda:1'), covar=tensor([0.0891, 0.0379, 0.0834, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0486, 0.0324, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 05:19:17,817 INFO [train.py:968] (1/2) Epoch 8, batch 15500, giga_loss[loss=0.2466, simple_loss=0.322, pruned_loss=0.0856, over 28799.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3349, pruned_loss=0.09197, over 5696738.56 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3461, pruned_loss=0.09991, over 5766625.72 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3346, pruned_loss=0.09159, over 5680407.37 frames. ], batch size: 263, lr: 4.09e-03, grad_scale: 2.0
+2023-03-04 05:19:59,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9871, 4.8283, 4.4922, 2.0069], device='cuda:1'), covar=tensor([0.0341, 0.0522, 0.0590, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0883, 0.0776, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 05:20:07,941 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=334317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:20:10,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=334320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:20:12,960 INFO [optim.py:369] (1/2) 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 05:20:18,925 INFO [train.py:968] (1/2) Epoch 8, batch 15550, libri_loss[loss=0.3182, simple_loss=0.3779, pruned_loss=0.1292, over 19828.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3332, pruned_loss=0.09109, over 5679264.55 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3456, pruned_loss=0.09968, over 5759706.07 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3332, pruned_loss=0.09084, over 5671968.03 frames. ], batch size: 186, lr: 4.09e-03, grad_scale: 2.0
+2023-03-04 05:20:42,183 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=334349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:20:56,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7215, 3.5341, 3.3544, 1.9251], device='cuda:1'), covar=tensor([0.0584, 0.0824, 0.0861, 0.1805], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0885, 0.0777, 0.0615], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 05:21:11,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2331, 3.0466, 2.8692, 1.3920], device='cuda:1'), covar=tensor([0.1006, 0.1113, 0.1102, 0.2455], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0888, 0.0778, 0.0617], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 05:21:19,527 INFO [train.py:968] (1/2) Epoch 8, batch 15600, giga_loss[loss=0.2856, simple_loss=0.3592, pruned_loss=0.106, over 27581.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3341, pruned_loss=0.09005, over 5664409.85 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3456, pruned_loss=0.09963, over 5761929.45 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3339, pruned_loss=0.08976, over 5655541.71 frames. ], batch size: 472, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:21:55,632 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=334409.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:22:08,744 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 15650, giga_loss[loss=0.272, simple_loss=0.3499, pruned_loss=0.09705, over 28979.00 frames. ], tot_loss[loss=0.261, simple_loss=0.338, pruned_loss=0.092, over 5662041.03 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3455, pruned_loss=0.09941, over 5754894.09 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3376, pruned_loss=0.09168, over 5657293.80 frames. ], batch size: 186, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:23:16,178 INFO [train.py:968] (1/2) Epoch 8, batch 15700, giga_loss[loss=0.2717, simple_loss=0.3521, pruned_loss=0.0957, over 29009.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3397, pruned_loss=0.09266, over 5659268.11 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3456, pruned_loss=0.09938, over 5755182.01 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3393, pruned_loss=0.09234, over 5654099.85 frames. ], batch size: 128, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:24:04,767 INFO [optim.py:369] (1/2) 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 05:24:10,543 INFO [train.py:968] (1/2) Epoch 8, batch 15750, giga_loss[loss=0.2828, simple_loss=0.3483, pruned_loss=0.1087, over 28736.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.34, pruned_loss=0.0938, over 5656046.89 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.345, pruned_loss=0.09909, over 5758023.96 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.34, pruned_loss=0.09356, over 5644867.29 frames. ], batch size: 99, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:25:10,446 INFO [train.py:968] (1/2) Epoch 8, batch 15800, giga_loss[loss=0.236, simple_loss=0.319, pruned_loss=0.07655, over 28776.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3365, pruned_loss=0.0916, over 5658817.85 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3442, pruned_loss=0.09868, over 5759556.81 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3371, pruned_loss=0.09167, over 5646830.81 frames. ], batch size: 119, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:26:07,540 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 15850, giga_loss[loss=0.2633, simple_loss=0.3435, pruned_loss=0.0916, over 28985.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3359, pruned_loss=0.09119, over 5660732.66 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3442, pruned_loss=0.09869, over 5760521.64 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3362, pruned_loss=0.09113, over 5649134.35 frames. ], batch size: 284, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:27:13,069 INFO [train.py:968] (1/2) Epoch 8, batch 15900, giga_loss[loss=0.2374, simple_loss=0.3162, pruned_loss=0.07931, over 28821.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.334, pruned_loss=0.09062, over 5672659.44 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3438, pruned_loss=0.09862, over 5763927.52 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3344, pruned_loss=0.09045, over 5658352.91 frames. ], batch size: 119, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:27:26,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 05:28:02,265 INFO [optim.py:369] (1/2) 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 05:28:10,111 INFO [train.py:968] (1/2) Epoch 8, batch 15950, giga_loss[loss=0.286, simple_loss=0.3589, pruned_loss=0.1065, over 27648.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3353, pruned_loss=0.09114, over 5681274.50 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3427, pruned_loss=0.09799, over 5768704.62 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3363, pruned_loss=0.09132, over 5662082.29 frames. ], batch size: 474, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:29:14,319 INFO [train.py:968] (1/2) Epoch 8, batch 16000, giga_loss[loss=0.2831, simple_loss=0.3542, pruned_loss=0.106, over 28939.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3366, pruned_loss=0.09163, over 5677899.14 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3426, pruned_loss=0.09777, over 5771799.48 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3373, pruned_loss=0.09182, over 5657727.19 frames. ], batch size: 213, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:29:22,353 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=334784.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:29:58,637 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-04 05:30:09,691 INFO [optim.py:369] (1/2) 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 05:30:19,682 INFO [train.py:968] (1/2) Epoch 8, batch 16050, giga_loss[loss=0.252, simple_loss=0.3312, pruned_loss=0.08644, over 28749.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3382, pruned_loss=0.0933, over 5676691.55 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3427, pruned_loss=0.09783, over 5775322.97 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3385, pruned_loss=0.09327, over 5655058.21 frames. ], batch size: 262, lr: 4.09e-03, grad_scale: 8.0
+2023-03-04 05:31:13,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3593, 1.6073, 1.3178, 1.3895], device='cuda:1'), covar=tensor([0.2377, 0.2170, 0.2436, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.1197, 0.0890, 0.1058, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 05:31:18,640 INFO [train.py:968] (1/2) Epoch 8, batch 16100, giga_loss[loss=0.2636, simple_loss=0.349, pruned_loss=0.08905, over 28026.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3429, pruned_loss=0.09594, over 5658523.33 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.343, pruned_loss=0.09797, over 5766779.02 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.343, pruned_loss=0.09577, over 5647986.59 frames. ], batch size: 412, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:31:36,035 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 05:32:09,583 INFO [optim.py:369] (1/2) 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 05:32:14,540 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=334927.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:32:15,516 INFO [train.py:968] (1/2) Epoch 8, batch 16150, giga_loss[loss=0.3077, simple_loss=0.3886, pruned_loss=0.1134, over 28691.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3447, pruned_loss=0.09625, over 5659156.11 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3427, pruned_loss=0.09794, over 5769579.66 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.345, pruned_loss=0.09612, over 5646170.89 frames. ], batch size: 243, lr: 4.09e-03, grad_scale: 4.0
+2023-03-04 05:32:17,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=334930.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:32:30,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5875, 1.9121, 1.5650, 1.4616], device='cuda:1'), covar=tensor([0.1781, 0.1181, 0.0913, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1392, 0.1346, 0.1498], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 05:32:50,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 05:32:51,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=334959.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:33:16,454 INFO [train.py:968] (1/2) Epoch 8, batch 16200, giga_loss[loss=0.3388, simple_loss=0.3926, pruned_loss=0.1425, over 27635.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3457, pruned_loss=0.09724, over 5657894.91 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3426, pruned_loss=0.09803, over 5773074.96 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3462, pruned_loss=0.097, over 5640359.20 frames. ], batch size: 474, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:33:50,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335002.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:34:16,455 INFO [optim.py:369] (1/2) 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 05:34:24,314 INFO [train.py:968] (1/2) Epoch 8, batch 16250, giga_loss[loss=0.2518, simple_loss=0.324, pruned_loss=0.08978, over 28146.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3432, pruned_loss=0.09619, over 5664561.22 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3422, pruned_loss=0.09786, over 5775211.32 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.344, pruned_loss=0.09613, over 5646826.25 frames. ], batch size: 412, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:34:26,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 05:34:33,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2365, 1.3075, 1.1541, 1.0826], device='cuda:1'), covar=tensor([0.0680, 0.0415, 0.0976, 0.0779], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0437, 0.0494, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 05:35:27,425 INFO [train.py:968] (1/2) Epoch 8, batch 16300, giga_loss[loss=0.2401, simple_loss=0.3233, pruned_loss=0.07844, over 28019.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3419, pruned_loss=0.09511, over 5676097.96 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.342, pruned_loss=0.09782, over 5779055.32 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3427, pruned_loss=0.09504, over 5655703.32 frames. ], batch size: 412, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:36:20,800 INFO [optim.py:369] (1/2) 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 05:36:27,901 INFO [train.py:968] (1/2) Epoch 8, batch 16350, giga_loss[loss=0.2825, simple_loss=0.3532, pruned_loss=0.106, over 28648.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3417, pruned_loss=0.0958, over 5677315.31 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3417, pruned_loss=0.09775, over 5781070.59 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09574, over 5655653.73 frames. ], batch size: 262, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:37:28,779 INFO [train.py:968] (1/2) Epoch 8, batch 16400, giga_loss[loss=0.2306, simple_loss=0.312, pruned_loss=0.07457, over 28721.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3391, pruned_loss=0.09519, over 5663633.22 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3415, pruned_loss=0.09759, over 5776470.96 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.34, pruned_loss=0.09525, over 5648254.70 frames. ], batch size: 243, lr: 4.08e-03, grad_scale: 8.0
+2023-03-04 05:37:47,340 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 05:38:21,690 INFO [optim.py:369] (1/2) 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 05:38:26,846 INFO [train.py:968] (1/2) Epoch 8, batch 16450, giga_loss[loss=0.2951, simple_loss=0.3795, pruned_loss=0.1053, over 28663.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3369, pruned_loss=0.09363, over 5652601.79 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3411, pruned_loss=0.09747, over 5761165.53 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3379, pruned_loss=0.09373, over 5650073.53 frames. ], batch size: 262, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:39:20,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6050, 1.9650, 1.5585, 1.3423], device='cuda:1'), covar=tensor([0.1621, 0.1178, 0.0872, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1380, 0.1333, 0.1483], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 05:39:28,020 INFO [train.py:968] (1/2) Epoch 8, batch 16500, giga_loss[loss=0.2253, simple_loss=0.3165, pruned_loss=0.06705, over 28955.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3352, pruned_loss=0.09128, over 5665255.12 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3404, pruned_loss=0.09698, over 5763820.11 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3366, pruned_loss=0.0917, over 5658606.75 frames. ], batch size: 155, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:39:55,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7922, 2.0453, 1.6357, 2.0776], device='cuda:1'), covar=tensor([0.2128, 0.1980, 0.2143, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.1202, 0.0893, 0.1066, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 05:40:17,893 INFO [optim.py:369] (1/2) 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 05:40:23,732 INFO [train.py:968] (1/2) Epoch 8, batch 16550, giga_loss[loss=0.2575, simple_loss=0.3476, pruned_loss=0.08376, over 28956.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.335, pruned_loss=0.08974, over 5669071.58 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3402, pruned_loss=0.0969, over 5757589.83 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3361, pruned_loss=0.08995, over 5666086.55 frames. ], batch size: 284, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:40:31,804 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:40:40,247 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:41:17,938 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 16600, giga_loss[loss=0.2292, simple_loss=0.3231, pruned_loss=0.06762, over 28903.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3373, pruned_loss=0.08897, over 5682757.47 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3399, pruned_loss=0.09671, over 5762285.07 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3383, pruned_loss=0.08907, over 5673682.47 frames. ], batch size: 213, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:41:20,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-04 05:41:51,102 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 8, batch 16650, giga_loss[loss=0.2506, simple_loss=0.3131, pruned_loss=0.09401, over 24528.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3382, pruned_loss=0.08949, over 5670379.30 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3394, pruned_loss=0.09642, over 5754817.22 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3394, pruned_loss=0.08959, over 5666977.18 frames. ], batch size: 705, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:43:16,131 INFO [train.py:968] (1/2) Epoch 8, batch 16700, giga_loss[loss=0.2464, simple_loss=0.3326, pruned_loss=0.0801, over 28629.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3389, pruned_loss=0.09034, over 5666818.30 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3392, pruned_loss=0.09626, over 5756955.22 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3401, pruned_loss=0.09047, over 5661137.25 frames. ], batch size: 242, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:44:10,782 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335520.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 05:44:15,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:44:16,471 INFO [optim.py:369] (1/2) 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 05:44:23,028 INFO [train.py:968] (1/2) Epoch 8, batch 16750, libri_loss[loss=0.3331, simple_loss=0.378, pruned_loss=0.1441, over 29536.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3392, pruned_loss=0.09076, over 5660668.19 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3395, pruned_loss=0.09652, over 5758438.18 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3399, pruned_loss=0.09043, over 5652229.08 frames. ], batch size: 81, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:44:36,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0149, 1.3281, 1.0762, 0.1998], device='cuda:1'), covar=tensor([0.1688, 0.1483, 0.2503, 0.2986], device='cuda:1'), in_proj_covar=tensor([0.1459, 0.1389, 0.1444, 0.1199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 05:44:56,995 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335552.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 05:45:10,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4114, 1.6555, 1.3197, 1.2222], device='cuda:1'), covar=tensor([0.1549, 0.1162, 0.0979, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1376, 0.1340, 0.1486], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 05:45:34,875 INFO [train.py:968] (1/2) Epoch 8, batch 16800, giga_loss[loss=0.2524, simple_loss=0.3408, pruned_loss=0.08204, over 28650.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3401, pruned_loss=0.09089, over 5664636.53 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3397, pruned_loss=0.09664, over 5759701.07 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3405, pruned_loss=0.09046, over 5655594.41 frames. ], batch size: 307, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:45:40,649 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=335582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:46:23,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-04 05:46:41,417 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 16850, giga_loss[loss=0.3123, simple_loss=0.3858, pruned_loss=0.1194, over 29051.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3401, pruned_loss=0.09045, over 5661656.71 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3394, pruned_loss=0.09637, over 5762505.00 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3406, pruned_loss=0.09023, over 5650056.96 frames. ], batch size: 285, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:47:51,341 INFO [train.py:968] (1/2) Epoch 8, batch 16900, giga_loss[loss=0.3276, simple_loss=0.3884, pruned_loss=0.1334, over 28068.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3447, pruned_loss=0.09304, over 5656408.36 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3395, pruned_loss=0.09635, over 5747653.00 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3451, pruned_loss=0.09272, over 5656078.16 frames. ], batch size: 412, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:48:16,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1805, 1.2123, 4.1421, 3.1558], device='cuda:1'), covar=tensor([0.2069, 0.2745, 0.0588, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0601, 0.0559, 0.0787, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 05:48:37,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:48:47,676 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:49:01,065 INFO [train.py:968] (1/2) Epoch 8, batch 16950, giga_loss[loss=0.2744, simple_loss=0.3504, pruned_loss=0.09915, over 28869.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3447, pruned_loss=0.09263, over 5664432.36 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3398, pruned_loss=0.09649, over 5748386.16 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3449, pruned_loss=0.09221, over 5662576.53 frames. ], batch size: 174, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:50:08,495 INFO [train.py:968] (1/2) Epoch 8, batch 17000, giga_loss[loss=0.2704, simple_loss=0.3488, pruned_loss=0.09594, over 28815.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3434, pruned_loss=0.09291, over 5672918.41 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3394, pruned_loss=0.09625, over 5754103.64 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.344, pruned_loss=0.09265, over 5664081.09 frames. ], batch size: 243, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:50:14,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:50:47,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-04 05:51:12,819 INFO [optim.py:369] (1/2) 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 05:51:18,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-04 05:51:18,592 INFO [train.py:968] (1/2) Epoch 8, batch 17050, giga_loss[loss=0.2523, simple_loss=0.3414, pruned_loss=0.08159, over 28615.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3415, pruned_loss=0.09182, over 5669788.25 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3397, pruned_loss=0.09642, over 5746606.06 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3418, pruned_loss=0.09138, over 5667209.17 frames. ], batch size: 262, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:51:28,543 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 05:51:58,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:02,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:10,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:15,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:29,850 INFO [train.py:968] (1/2) Epoch 8, batch 17100, giga_loss[loss=0.239, simple_loss=0.3283, pruned_loss=0.0748, over 28968.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3395, pruned_loss=0.09005, over 5668758.37 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3398, pruned_loss=0.09645, over 5748751.02 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3396, pruned_loss=0.08961, over 5663737.90 frames. ], batch size: 213, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:52:40,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:52:50,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:53:26,715 INFO [optim.py:369] (1/2) 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 05:53:27,043 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=335925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:53:32,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=335928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:53:32,661 INFO [train.py:968] (1/2) Epoch 8, batch 17150, giga_loss[loss=0.2437, simple_loss=0.3228, pruned_loss=0.08234, over 28776.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3392, pruned_loss=0.08992, over 5674640.79 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3398, pruned_loss=0.09641, over 5752308.32 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3393, pruned_loss=0.08945, over 5665474.60 frames. ], batch size: 119, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:54:06,384 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=335957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:54:06,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=335957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:54:06,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5490, 1.9771, 1.8878, 1.4454], device='cuda:1'), covar=tensor([0.1671, 0.1945, 0.1256, 0.1486], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0688, 0.0812, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 05:54:27,283 INFO [train.py:968] (1/2) Epoch 8, batch 17200, giga_loss[loss=0.2586, simple_loss=0.3373, pruned_loss=0.08999, over 28937.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.341, pruned_loss=0.09095, over 5679152.31 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3389, pruned_loss=0.09593, over 5757975.76 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3419, pruned_loss=0.09082, over 5664256.52 frames. ], batch size: 106, lr: 4.08e-03, grad_scale: 8.0
+2023-03-04 05:55:22,485 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 17250, libri_loss[loss=0.2184, simple_loss=0.2952, pruned_loss=0.07076, over 29351.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3413, pruned_loss=0.09129, over 5681705.40 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3388, pruned_loss=0.09585, over 5760366.67 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3422, pruned_loss=0.09119, over 5666172.98 frames. ], batch size: 67, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:55:33,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4559, 3.4298, 1.5165, 1.6123], device='cuda:1'), covar=tensor([0.0887, 0.0369, 0.0887, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0486, 0.0326, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 05:56:24,780 INFO [train.py:968] (1/2) Epoch 8, batch 17300, giga_loss[loss=0.3117, simple_loss=0.3598, pruned_loss=0.1318, over 26919.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.339, pruned_loss=0.09162, over 5663869.38 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.339, pruned_loss=0.09593, over 5751045.44 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3395, pruned_loss=0.09142, over 5658886.48 frames. ], batch size: 555, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 05:56:49,578 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=336100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:56:52,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=336103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:57:19,566 INFO [optim.py:369] (1/2) 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 05:57:21,326 INFO [train.py:968] (1/2) Epoch 8, batch 17350, giga_loss[loss=0.2598, simple_loss=0.3356, pruned_loss=0.09205, over 28623.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3398, pruned_loss=0.09324, over 5660119.74 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3391, pruned_loss=0.09601, over 5752941.59 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3401, pruned_loss=0.09293, over 5652531.19 frames. ], batch size: 242, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:57:24,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=336132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:57:51,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4761, 2.0043, 1.8100, 1.3737], device='cuda:1'), covar=tensor([0.1551, 0.1898, 0.1213, 0.1391], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0687, 0.0811, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0012, 0.0011], device='cuda:1')
+2023-03-04 05:58:15,140 INFO [train.py:968] (1/2) Epoch 8, batch 17400, giga_loss[loss=0.3036, simple_loss=0.3775, pruned_loss=0.1148, over 28141.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3423, pruned_loss=0.09504, over 5660600.20 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3391, pruned_loss=0.09597, over 5754883.37 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3426, pruned_loss=0.09473, over 5650058.78 frames. ], batch size: 412, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:58:36,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2136, 2.5160, 1.2260, 1.2791], device='cuda:1'), covar=tensor([0.0931, 0.0328, 0.0902, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0487, 0.0326, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 05:58:40,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-04 05:59:09,608 INFO [optim.py:369] (1/2) 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 05:59:10,540 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=336228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 05:59:10,922 INFO [train.py:968] (1/2) Epoch 8, batch 17450, giga_loss[loss=0.3819, simple_loss=0.4395, pruned_loss=0.1622, over 29047.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3537, pruned_loss=0.102, over 5665086.01 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.339, pruned_loss=0.09587, over 5755593.96 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.354, pruned_loss=0.1019, over 5655845.38 frames. ], batch size: 136, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 05:59:55,010 INFO [train.py:968] (1/2) Epoch 8, batch 17500, giga_loss[loss=0.2731, simple_loss=0.3446, pruned_loss=0.1008, over 28838.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.359, pruned_loss=0.1053, over 5676201.49 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3393, pruned_loss=0.09594, over 5759129.20 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3595, pruned_loss=0.1053, over 5663678.90 frames. ], batch size: 119, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 06:00:22,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6009, 1.0957, 5.0274, 3.4678], device='cuda:1'), covar=tensor([0.1600, 0.2643, 0.0286, 0.0690], device='cuda:1'), in_proj_covar=tensor([0.0603, 0.0558, 0.0793, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 06:00:34,750 INFO [optim.py:369] (1/2) 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 06:00:36,643 INFO [train.py:968] (1/2) Epoch 8, batch 17550, giga_loss[loss=0.26, simple_loss=0.3291, pruned_loss=0.09545, over 28869.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.357, pruned_loss=0.1054, over 5670433.95 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3389, pruned_loss=0.09571, over 5753094.82 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3586, pruned_loss=0.106, over 5661858.60 frames. ], batch size: 199, lr: 4.08e-03, grad_scale: 2.0
+2023-03-04 06:01:22,892 INFO [train.py:968] (1/2) Epoch 8, batch 17600, giga_loss[loss=0.2483, simple_loss=0.3156, pruned_loss=0.09044, over 28852.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3502, pruned_loss=0.1021, over 5678015.85 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.339, pruned_loss=0.09569, over 5750987.03 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3517, pruned_loss=0.1029, over 5671458.71 frames. ], batch size: 112, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:02:07,474 INFO [optim.py:369] (1/2) 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 06:02:09,958 INFO [train.py:968] (1/2) Epoch 8, batch 17650, libri_loss[loss=0.2823, simple_loss=0.3648, pruned_loss=0.09991, over 29361.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3436, pruned_loss=0.09951, over 5684419.72 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3393, pruned_loss=0.09577, over 5749533.44 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3445, pruned_loss=0.1001, over 5679251.19 frames. ], batch size: 92, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:02:48,400 INFO [train.py:968] (1/2) Epoch 8, batch 17700, giga_loss[loss=0.2342, simple_loss=0.3039, pruned_loss=0.08225, over 28686.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.337, pruned_loss=0.09644, over 5686624.31 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3399, pruned_loss=0.09605, over 5743457.17 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3373, pruned_loss=0.09673, over 5685581.63 frames. ], batch size: 92, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:03:23,150 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-04 06:03:27,336 INFO [optim.py:369] (1/2) 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 06:03:28,829 INFO [train.py:968] (1/2) Epoch 8, batch 17750, giga_loss[loss=0.2059, simple_loss=0.2795, pruned_loss=0.06615, over 28207.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.33, pruned_loss=0.09317, over 5690230.47 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3399, pruned_loss=0.09576, over 5745975.32 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3301, pruned_loss=0.09368, over 5684893.21 frames. ], batch size: 77, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:03:53,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2468, 1.4904, 1.1509, 0.8931], device='cuda:1'), covar=tensor([0.1557, 0.1362, 0.0964, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.1596, 0.1398, 0.1367, 0.1502], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 06:04:08,238 INFO [train.py:968] (1/2) Epoch 8, batch 17800, giga_loss[loss=0.2484, simple_loss=0.3207, pruned_loss=0.08803, over 28850.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3244, pruned_loss=0.09041, over 5692025.73 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.34, pruned_loss=0.0958, over 5749905.76 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.324, pruned_loss=0.09069, over 5682305.90 frames. ], batch size: 199, lr: 4.08e-03, grad_scale: 4.0
+2023-03-04 06:04:29,720 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 8, batch 17850, giga_loss[loss=0.2412, simple_loss=0.3063, pruned_loss=0.08801, over 28657.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3222, pruned_loss=0.08992, over 5695526.40 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3403, pruned_loss=0.09596, over 5749082.53 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3213, pruned_loss=0.08988, over 5687739.88 frames. ], batch size: 92, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:05:33,109 INFO [train.py:968] (1/2) Epoch 8, batch 17900, giga_loss[loss=0.2257, simple_loss=0.2958, pruned_loss=0.07784, over 28968.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3203, pruned_loss=0.08903, over 5694441.50 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.341, pruned_loss=0.09615, over 5751244.77 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3183, pruned_loss=0.08861, over 5684707.16 frames. ], batch size: 227, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:06:12,282 INFO [optim.py:369] (1/2) 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 06:06:14,726 INFO [train.py:968] (1/2) Epoch 8, batch 17950, libri_loss[loss=0.2683, simple_loss=0.3428, pruned_loss=0.09691, over 29552.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3183, pruned_loss=0.08818, over 5684775.22 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3417, pruned_loss=0.09644, over 5746512.54 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3152, pruned_loss=0.08727, over 5679545.16 frames. ], batch size: 79, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:06:28,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=336746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:06:29,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=336749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:06:54,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=336778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:06:54,734 INFO [train.py:968] (1/2) Epoch 8, batch 18000, giga_loss[loss=0.2272, simple_loss=0.3025, pruned_loss=0.07595, over 28847.00 frames. ], tot_loss[loss=0.244, simple_loss=0.315, pruned_loss=0.08647, over 5699068.18 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.09651, over 5748989.81 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3118, pruned_loss=0.08552, over 5691801.35 frames. ], batch size: 186, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:06:54,735 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 06:07:04,418 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 06:07:21,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=336797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:07:31,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 06:07:47,779 INFO [optim.py:369] (1/2) 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 06:07:48,423 INFO [train.py:968] (1/2) Epoch 8, batch 18050, giga_loss[loss=0.2228, simple_loss=0.2864, pruned_loss=0.07956, over 27673.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3131, pruned_loss=0.0858, over 5685268.62 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3429, pruned_loss=0.09705, over 5740821.07 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3093, pruned_loss=0.08433, over 5686658.38 frames. ], batch size: 472, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:07:57,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8763, 4.7497, 2.0327, 2.0914], device='cuda:1'), covar=tensor([0.0839, 0.0229, 0.0793, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0483, 0.0321, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 06:08:34,240 INFO [train.py:968] (1/2) Epoch 8, batch 18100, giga_loss[loss=0.2268, simple_loss=0.2879, pruned_loss=0.08284, over 28722.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3092, pruned_loss=0.08412, over 5680659.27 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3431, pruned_loss=0.09714, over 5738863.46 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3057, pruned_loss=0.08277, over 5683134.61 frames. ], batch size: 92, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:09:17,270 INFO [optim.py:369] (1/2) 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 06:09:17,844 INFO [train.py:968] (1/2) Epoch 8, batch 18150, giga_loss[loss=0.2194, simple_loss=0.2897, pruned_loss=0.07454, over 28528.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3066, pruned_loss=0.08234, over 5693452.65 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3435, pruned_loss=0.09719, over 5742667.24 frames. ], giga_tot_loss[loss=0.2322, simple_loss=0.3027, pruned_loss=0.0809, over 5690967.90 frames. ], batch size: 336, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:10:03,571 INFO [train.py:968] (1/2) Epoch 8, batch 18200, giga_loss[loss=0.2002, simple_loss=0.2742, pruned_loss=0.0631, over 28944.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3036, pruned_loss=0.08101, over 5694670.30 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09736, over 5744387.27 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.2992, pruned_loss=0.07941, over 5690366.33 frames. ], batch size: 213, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:10:45,752 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 18250, giga_loss[loss=0.2608, simple_loss=0.3305, pruned_loss=0.09554, over 28962.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3092, pruned_loss=0.08453, over 5692692.81 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3441, pruned_loss=0.09746, over 5739682.92 frames. ], giga_tot_loss[loss=0.2349, simple_loss=0.3044, pruned_loss=0.08269, over 5691371.43 frames. ], batch size: 106, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:11:35,283 INFO [train.py:968] (1/2) Epoch 8, batch 18300, giga_loss[loss=0.2997, simple_loss=0.3719, pruned_loss=0.1137, over 28801.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3221, pruned_loss=0.09143, over 5696864.23 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3439, pruned_loss=0.09722, over 5744013.31 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3178, pruned_loss=0.08993, over 5691147.71 frames. ], batch size: 119, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:11:57,362 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:12:15,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1222, 2.5364, 1.2666, 1.2348], device='cuda:1'), covar=tensor([0.0977, 0.0311, 0.0872, 0.1363], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0485, 0.0323, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 06:12:16,773 INFO [optim.py:369] (1/2) 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 06:12:17,326 INFO [train.py:968] (1/2) Epoch 8, batch 18350, giga_loss[loss=0.3572, simple_loss=0.4011, pruned_loss=0.1566, over 28788.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3363, pruned_loss=0.09965, over 5689747.20 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3437, pruned_loss=0.09708, over 5736158.98 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3328, pruned_loss=0.09856, over 5691736.12 frames. ], batch size: 92, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:12:17,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-04 06:12:53,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=337172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:12:57,989 INFO [train.py:968] (1/2) Epoch 8, batch 18400, giga_loss[loss=0.3142, simple_loss=0.3872, pruned_loss=0.1206, over 28760.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3459, pruned_loss=0.1041, over 5695836.44 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3438, pruned_loss=0.09706, over 5739063.16 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.343, pruned_loss=0.1034, over 5693493.03 frames. ], batch size: 242, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:13:40,959 INFO [optim.py:369] (1/2) 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 06:13:40,978 INFO [train.py:968] (1/2) Epoch 8, batch 18450, giga_loss[loss=0.3672, simple_loss=0.4205, pruned_loss=0.1569, over 27561.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3512, pruned_loss=0.1058, over 5695996.67 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.344, pruned_loss=0.09726, over 5742188.14 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3488, pruned_loss=0.1052, over 5690397.14 frames. ], batch size: 472, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:13:45,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3001, 1.4656, 1.2104, 1.1023], device='cuda:1'), covar=tensor([0.1357, 0.1325, 0.0886, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.1604, 0.1425, 0.1389, 0.1518], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 06:14:20,349 INFO [train.py:968] (1/2) Epoch 8, batch 18500, giga_loss[loss=0.2667, simple_loss=0.3469, pruned_loss=0.09326, over 28825.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3533, pruned_loss=0.1054, over 5694840.55 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.09729, over 5742304.61 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3512, pruned_loss=0.1052, over 5688580.52 frames. ], batch size: 199, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:14:31,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7539, 1.7997, 1.8213, 1.6227], device='cuda:1'), covar=tensor([0.1259, 0.1619, 0.1547, 0.1500], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0729, 0.0651, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 06:14:32,380 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:14:49,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1969, 1.6056, 1.4983, 1.3748], device='cuda:1'), covar=tensor([0.0875, 0.0307, 0.0290, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0120, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0051, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 06:14:56,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=337315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:14:58,575 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=337318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:15:06,858 INFO [optim.py:369] (1/2) 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 06:15:06,870 INFO [train.py:968] (1/2) Epoch 8, batch 18550, libri_loss[loss=0.2625, simple_loss=0.3405, pruned_loss=0.0922, over 29555.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3543, pruned_loss=0.1057, over 5691549.74 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.345, pruned_loss=0.09752, over 5745876.85 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3524, pruned_loss=0.1056, over 5681419.70 frames. ], batch size: 79, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:15:22,340 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=337347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:15:50,850 INFO [train.py:968] (1/2) Epoch 8, batch 18600, giga_loss[loss=0.3045, simple_loss=0.3538, pruned_loss=0.1277, over 23390.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3571, pruned_loss=0.1079, over 5686469.23 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3451, pruned_loss=0.09756, over 5742752.46 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3556, pruned_loss=0.1079, over 5680943.97 frames. ], batch size: 705, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:16:33,352 INFO [optim.py:369] (1/2) 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 06:16:33,364 INFO [train.py:968] (1/2) Epoch 8, batch 18650, giga_loss[loss=0.3367, simple_loss=0.3861, pruned_loss=0.1437, over 28579.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3603, pruned_loss=0.1102, over 5689659.26 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.09801, over 5737602.29 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3587, pruned_loss=0.1101, over 5688506.93 frames. ], batch size: 71, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:17:11,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=337476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:17:14,500 INFO [train.py:968] (1/2) Epoch 8, batch 18700, giga_loss[loss=0.2923, simple_loss=0.3688, pruned_loss=0.1079, over 28721.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3635, pruned_loss=0.1117, over 5685957.67 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3462, pruned_loss=0.09814, over 5729600.12 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3622, pruned_loss=0.1118, over 5691168.07 frames. ], batch size: 284, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:17:23,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-04 06:17:56,742 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 18750, giga_loss[loss=0.3264, simple_loss=0.3952, pruned_loss=0.1288, over 29075.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3645, pruned_loss=0.111, over 5690974.05 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09794, over 5723051.52 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.364, pruned_loss=0.1114, over 5699552.66 frames. ], batch size: 155, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:18:31,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3315, 1.4966, 1.2625, 1.2480], device='cuda:1'), covar=tensor([0.2261, 0.2104, 0.2286, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.1210, 0.0911, 0.1070, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 06:18:37,155 INFO [train.py:968] (1/2) Epoch 8, batch 18800, giga_loss[loss=0.2978, simple_loss=0.3761, pruned_loss=0.1097, over 28930.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3657, pruned_loss=0.1109, over 5698753.34 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3461, pruned_loss=0.09791, over 5728691.81 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3658, pruned_loss=0.1117, over 5699874.29 frames. ], batch size: 174, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:19:07,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3225, 1.7092, 1.6288, 1.2634], device='cuda:1'), covar=tensor([0.1573, 0.1980, 0.1223, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0697, 0.0823, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 06:19:10,317 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=337619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:12,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=337622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:16,200 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 18850, giga_loss[loss=0.2711, simple_loss=0.359, pruned_loss=0.09158, over 28953.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3664, pruned_loss=0.1107, over 5694036.57 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09792, over 5729413.06 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.367, pruned_loss=0.1116, over 5693430.61 frames. ], batch size: 174, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:19:34,543 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=337651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:49,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=337666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:19:59,526 INFO [train.py:968] (1/2) Epoch 8, batch 18900, giga_loss[loss=0.3164, simple_loss=0.3919, pruned_loss=0.1205, over 28462.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3655, pruned_loss=0.109, over 5693691.13 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3463, pruned_loss=0.09805, over 5730541.00 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3658, pruned_loss=0.1097, over 5692147.10 frames. ], batch size: 71, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:20:09,801 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:20:10,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9850, 1.2308, 1.2252, 1.2135], device='cuda:1'), covar=tensor([0.1163, 0.0996, 0.1698, 0.1122], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0724, 0.0645, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 06:20:40,881 INFO [train.py:968] (1/2) Epoch 8, batch 18950, giga_loss[loss=0.258, simple_loss=0.3368, pruned_loss=0.0896, over 28931.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3637, pruned_loss=0.1073, over 5707377.59 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3465, pruned_loss=0.09804, over 5733698.44 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3642, pruned_loss=0.1081, over 5702835.94 frames. ], batch size: 112, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:20:41,631 INFO [optim.py:369] (1/2) 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 06:21:05,266 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-04 06:21:21,411 INFO [train.py:968] (1/2) Epoch 8, batch 19000, giga_loss[loss=0.2644, simple_loss=0.3478, pruned_loss=0.09049, over 28394.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.365, pruned_loss=0.1087, over 5702022.35 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3473, pruned_loss=0.09847, over 5735951.06 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.365, pruned_loss=0.1091, over 5695730.99 frames. ], batch size: 60, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:21:28,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-04 06:21:49,873 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=337809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:21:52,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=337812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:21:58,399 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 19050, libri_loss[loss=0.2942, simple_loss=0.3768, pruned_loss=0.1059, over 29650.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3683, pruned_loss=0.1136, over 5696489.97 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.348, pruned_loss=0.09865, over 5741365.84 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3683, pruned_loss=0.1142, over 5685170.37 frames. ], batch size: 88, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:22:08,550 INFO [optim.py:369] (1/2) 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 06:22:19,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=337841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:22:51,915 INFO [train.py:968] (1/2) Epoch 8, batch 19100, libri_loss[loss=0.2754, simple_loss=0.3575, pruned_loss=0.09669, over 29500.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3698, pruned_loss=0.1163, over 5698772.57 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3486, pruned_loss=0.09891, over 5744685.89 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3698, pruned_loss=0.1169, over 5685477.20 frames. ], batch size: 81, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:22:56,667 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-04 06:23:04,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2958, 4.0378, 3.8735, 1.9775], device='cuda:1'), covar=tensor([0.0536, 0.0721, 0.0679, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0886, 0.0784, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 06:23:32,198 INFO [train.py:968] (1/2) Epoch 8, batch 19150, giga_loss[loss=0.3309, simple_loss=0.381, pruned_loss=0.1404, over 27705.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3686, pruned_loss=0.1164, over 5699857.62 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3488, pruned_loss=0.09897, over 5746342.29 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3687, pruned_loss=0.1172, over 5687236.53 frames. ], batch size: 472, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:23:32,886 INFO [optim.py:369] (1/2) 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 06:24:14,449 INFO [train.py:968] (1/2) Epoch 8, batch 19200, giga_loss[loss=0.2673, simple_loss=0.3496, pruned_loss=0.09248, over 28775.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3664, pruned_loss=0.1152, over 5704672.16 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3495, pruned_loss=0.09919, over 5746054.15 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3663, pruned_loss=0.116, over 5693714.56 frames. ], batch size: 145, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:24:16,355 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:24:25,493 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=337991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:24:43,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2285, 4.0146, 3.8162, 1.8567], device='cuda:1'), covar=tensor([0.0549, 0.0674, 0.0716, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0884, 0.0783, 0.0628], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 06:24:43,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4168, 1.5817, 1.3887, 1.2422], device='cuda:1'), covar=tensor([0.1490, 0.1293, 0.0890, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.1606, 0.1421, 0.1397, 0.1519], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 06:24:55,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6357, 1.5530, 1.1416, 1.2595], device='cuda:1'), covar=tensor([0.0725, 0.0606, 0.1025, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0438, 0.0495, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 06:25:02,755 INFO [train.py:968] (1/2) Epoch 8, batch 19250, giga_loss[loss=0.2743, simple_loss=0.3515, pruned_loss=0.09853, over 29053.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3655, pruned_loss=0.1146, over 5688049.21 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3495, pruned_loss=0.09919, over 5746054.15 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3655, pruned_loss=0.1151, over 5679520.77 frames. ], batch size: 155, lr: 4.07e-03, grad_scale: 8.0
+2023-03-04 06:25:03,514 INFO [optim.py:369] (1/2) 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 06:25:17,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4494, 1.5984, 1.4974, 1.5213], device='cuda:1'), covar=tensor([0.1206, 0.1621, 0.1740, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0722, 0.0648, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 06:25:32,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:25:35,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2034, 2.5148, 2.4140, 2.0059], device='cuda:1'), covar=tensor([0.1438, 0.1650, 0.1075, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0702, 0.0825, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 06:25:43,842 INFO [train.py:968] (1/2) Epoch 8, batch 19300, giga_loss[loss=0.3076, simple_loss=0.3677, pruned_loss=0.1237, over 28290.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3629, pruned_loss=0.1119, over 5694227.33 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.35, pruned_loss=0.09947, over 5747252.62 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3625, pruned_loss=0.1123, over 5686056.21 frames. ], batch size: 368, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:26:27,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 06:26:31,154 INFO [train.py:968] (1/2) Epoch 8, batch 19350, giga_loss[loss=0.2391, simple_loss=0.2985, pruned_loss=0.08982, over 23480.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3582, pruned_loss=0.1087, over 5685133.32 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3504, pruned_loss=0.0996, over 5746391.22 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3578, pruned_loss=0.109, over 5678466.93 frames. ], batch size: 705, lr: 4.07e-03, grad_scale: 4.0
+2023-03-04 06:26:33,885 INFO [optim.py:369] (1/2) 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 06:27:19,576 INFO [train.py:968] (1/2) Epoch 8, batch 19400, giga_loss[loss=0.2339, simple_loss=0.3105, pruned_loss=0.07868, over 28906.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3517, pruned_loss=0.1048, over 5678369.65 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3509, pruned_loss=0.09984, over 5739863.43 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3509, pruned_loss=0.105, over 5678350.33 frames. ], batch size: 227, lr: 4.07e-03, grad_scale: 2.0
+2023-03-04 06:27:29,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:27:43,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:27:46,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:27:48,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8433, 3.6602, 3.4094, 1.9299], device='cuda:1'), covar=tensor([0.0543, 0.0673, 0.0656, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0880, 0.0777, 0.0627], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 06:28:02,410 INFO [train.py:968] (1/2) Epoch 8, batch 19450, giga_loss[loss=0.2559, simple_loss=0.3243, pruned_loss=0.0937, over 28693.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3474, pruned_loss=0.1028, over 5675677.17 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3518, pruned_loss=0.1001, over 5741988.56 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.346, pruned_loss=0.1028, over 5671986.08 frames. ], batch size: 284, lr: 4.07e-03, grad_scale: 2.0
+2023-03-04 06:28:04,769 INFO [optim.py:369] (1/2) 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 06:28:15,959 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:28:16,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5793, 1.5257, 1.1525, 1.2010], device='cuda:1'), covar=tensor([0.0650, 0.0487, 0.0953, 0.0921], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0439, 0.0496, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 06:28:16,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 06:28:18,993 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 19500, giga_loss[loss=0.2572, simple_loss=0.3279, pruned_loss=0.09325, over 28563.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.344, pruned_loss=0.1005, over 5689967.63 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.352, pruned_loss=0.1002, over 5745863.66 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3425, pruned_loss=0.1005, over 5682553.73 frames. ], batch size: 92, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:29:08,184 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:29:35,368 INFO [train.py:968] (1/2) Epoch 8, batch 19550, giga_loss[loss=0.2565, simple_loss=0.3339, pruned_loss=0.08958, over 28956.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3447, pruned_loss=0.1003, over 5696793.38 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3518, pruned_loss=0.09988, over 5749059.84 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3435, pruned_loss=0.1005, over 5686640.02 frames. ], batch size: 213, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:29:37,737 INFO [optim.py:369] (1/2) 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 06:29:43,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:29:45,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:29:52,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 06:30:01,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:30:11,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:30:12,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:30:20,095 INFO [train.py:968] (1/2) Epoch 8, batch 19600, giga_loss[loss=0.2661, simple_loss=0.3377, pruned_loss=0.09724, over 28923.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3441, pruned_loss=0.09975, over 5690591.01 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3524, pruned_loss=0.1002, over 5739723.21 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3424, pruned_loss=0.09966, over 5690544.35 frames. ], batch size: 186, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:31:01,790 INFO [train.py:968] (1/2) Epoch 8, batch 19650, giga_loss[loss=0.251, simple_loss=0.3312, pruned_loss=0.08538, over 28521.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3425, pruned_loss=0.09935, over 5696664.44 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3535, pruned_loss=0.1008, over 5733817.25 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.34, pruned_loss=0.09873, over 5700639.33 frames. ], batch size: 336, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:31:04,515 INFO [optim.py:369] (1/2) 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 06:31:09,568 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:31:36,328 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:31:41,257 INFO [train.py:968] (1/2) Epoch 8, batch 19700, giga_loss[loss=0.2455, simple_loss=0.3127, pruned_loss=0.0892, over 28800.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3401, pruned_loss=0.09827, over 5708226.44 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3541, pruned_loss=0.1011, over 5736921.55 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3373, pruned_loss=0.09744, over 5708070.04 frames. ], batch size: 99, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:31:57,753 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:31:59,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:32:05,317 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:32:08,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:32:21,167 INFO [train.py:968] (1/2) Epoch 8, batch 19750, libri_loss[loss=0.3203, simple_loss=0.4045, pruned_loss=0.118, over 26063.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3379, pruned_loss=0.09723, over 5706311.84 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.355, pruned_loss=0.1014, over 5729228.15 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3345, pruned_loss=0.09621, over 5712143.05 frames. ], batch size: 136, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:32:22,073 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:32:30,622 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 19800, giga_loss[loss=0.2594, simple_loss=0.3301, pruned_loss=0.09436, over 28892.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.336, pruned_loss=0.09639, over 5711272.51 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3555, pruned_loss=0.1017, over 5731928.96 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3325, pruned_loss=0.09526, over 5712924.49 frames. ], batch size: 186, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:33:29,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8703, 1.8165, 1.3617, 1.6004], device='cuda:1'), covar=tensor([0.0685, 0.0569, 0.0886, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0447, 0.0500, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 06:33:37,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:33:43,940 INFO [train.py:968] (1/2) Epoch 8, batch 19850, giga_loss[loss=0.2555, simple_loss=0.33, pruned_loss=0.09047, over 28838.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3337, pruned_loss=0.09499, over 5717346.56 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3564, pruned_loss=0.1019, over 5735835.66 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3296, pruned_loss=0.09378, over 5714818.68 frames. ], batch size: 284, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:33:46,617 INFO [optim.py:369] (1/2) 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 06:34:06,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-04 06:34:16,341 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:34:23,239 INFO [train.py:968] (1/2) Epoch 8, batch 19900, libri_loss[loss=0.3454, simple_loss=0.4096, pruned_loss=0.1406, over 19987.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3333, pruned_loss=0.09547, over 5710247.22 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3571, pruned_loss=0.1021, over 5732440.31 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3286, pruned_loss=0.09396, over 5711394.93 frames. ], batch size: 187, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:34:40,056 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-04 06:35:04,007 INFO [train.py:968] (1/2) Epoch 8, batch 19950, giga_loss[loss=0.253, simple_loss=0.3224, pruned_loss=0.09185, over 28888.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3307, pruned_loss=0.09401, over 5709569.18 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3571, pruned_loss=0.1021, over 5726008.47 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3265, pruned_loss=0.09272, over 5715842.43 frames. ], batch size: 186, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:35:06,524 INFO [optim.py:369] (1/2) 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 06:35:29,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:35:32,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:35:42,636 INFO [train.py:968] (1/2) Epoch 8, batch 20000, giga_loss[loss=0.2236, simple_loss=0.2989, pruned_loss=0.0742, over 28924.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3291, pruned_loss=0.09279, over 5721200.23 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3578, pruned_loss=0.1023, over 5731348.84 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3244, pruned_loss=0.09137, over 5721202.59 frames. ], batch size: 213, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:35:56,249 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:01,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:09,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:10,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=338814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:11,767 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 20050, libri_loss[loss=0.2318, simple_loss=0.3256, pruned_loss=0.06897, over 28482.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3283, pruned_loss=0.09242, over 5723128.56 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3581, pruned_loss=0.1023, over 5730053.44 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3237, pruned_loss=0.09107, over 5724319.12 frames. ], batch size: 63, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:36:23,291 INFO [optim.py:369] (1/2) 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 06:36:32,718 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:36:34,805 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 20100, giga_loss[loss=0.2601, simple_loss=0.3313, pruned_loss=0.0945, over 29053.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3293, pruned_loss=0.09284, over 5727976.10 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3583, pruned_loss=0.1022, over 5732521.62 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3243, pruned_loss=0.0915, over 5726814.83 frames. ], batch size: 128, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:37:40,454 INFO [train.py:968] (1/2) Epoch 8, batch 20150, giga_loss[loss=0.2921, simple_loss=0.3619, pruned_loss=0.1111, over 28596.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3359, pruned_loss=0.09736, over 5710483.93 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3591, pruned_loss=0.1026, over 5729009.89 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3303, pruned_loss=0.09572, over 5713099.34 frames. ], batch size: 336, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:37:44,121 INFO [optim.py:369] (1/2) 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 06:38:09,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:13,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:31,049 INFO [train.py:968] (1/2) Epoch 8, batch 20200, libri_loss[loss=0.302, simple_loss=0.3811, pruned_loss=0.1114, over 29141.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3438, pruned_loss=0.1025, over 5712333.39 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3592, pruned_loss=0.1026, over 5731988.81 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.339, pruned_loss=0.1011, over 5711391.81 frames. ], batch size: 101, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:38:39,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=338989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:40,517 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=338991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:40,604 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=338991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:45,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=338994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:38:54,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6622, 2.1594, 1.9734, 1.5606], device='cuda:1'), covar=tensor([0.1623, 0.1987, 0.1265, 0.1514], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0704, 0.0828, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 06:39:09,315 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:39:10,245 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 06:39:15,433 INFO [train.py:968] (1/2) Epoch 8, batch 20250, giga_loss[loss=0.3943, simple_loss=0.4314, pruned_loss=0.1786, over 26677.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3534, pruned_loss=0.1096, over 5701659.72 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3591, pruned_loss=0.1025, over 5737958.98 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3491, pruned_loss=0.1087, over 5694158.58 frames. ], batch size: 555, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:39:19,622 INFO [optim.py:369] (1/2) 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 06:39:44,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:39:59,797 INFO [train.py:968] (1/2) Epoch 8, batch 20300, giga_loss[loss=0.3218, simple_loss=0.392, pruned_loss=0.1258, over 28544.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3576, pruned_loss=0.1107, over 5686010.55 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3597, pruned_loss=0.1029, over 5722357.97 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3535, pruned_loss=0.1098, over 5693973.79 frames. ], batch size: 336, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:40:08,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:40:19,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:40:43,627 INFO [train.py:968] (1/2) Epoch 8, batch 20350, libri_loss[loss=0.3062, simple_loss=0.3828, pruned_loss=0.1148, over 29217.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3612, pruned_loss=0.1121, over 5670235.23 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3598, pruned_loss=0.1029, over 5724102.65 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3578, pruned_loss=0.1116, over 5673220.35 frames. ], batch size: 97, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:40:46,327 INFO [optim.py:369] (1/2) 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 06:41:22,667 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:41:30,901 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 06:41:32,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:41:33,524 INFO [train.py:968] (1/2) Epoch 8, batch 20400, libri_loss[loss=0.2844, simple_loss=0.3681, pruned_loss=0.1003, over 27592.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3672, pruned_loss=0.1161, over 5667445.05 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3601, pruned_loss=0.1031, over 5722285.59 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3643, pruned_loss=0.1157, over 5671023.28 frames. ], batch size: 115, lr: 4.06e-03, grad_scale: 8.0
+2023-03-04 06:42:18,631 INFO [train.py:968] (1/2) Epoch 8, batch 20450, libri_loss[loss=0.3086, simple_loss=0.3875, pruned_loss=0.1148, over 29146.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3672, pruned_loss=0.1158, over 5670330.65 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3602, pruned_loss=0.1032, over 5723737.12 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3648, pruned_loss=0.1156, over 5671405.29 frames. ], batch size: 101, lr: 4.06e-03, grad_scale: 8.0
+2023-03-04 06:42:22,004 INFO [optim.py:369] (1/2) 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 06:42:44,483 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:42:55,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 06:43:03,298 INFO [train.py:968] (1/2) Epoch 8, batch 20500, giga_loss[loss=0.2507, simple_loss=0.334, pruned_loss=0.0837, over 28548.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3617, pruned_loss=0.1118, over 5666300.95 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3604, pruned_loss=0.1033, over 5716530.62 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3597, pruned_loss=0.1115, over 5673753.36 frames. ], batch size: 60, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:43:17,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 06:43:18,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8447, 3.6575, 3.4475, 1.7264], device='cuda:1'), covar=tensor([0.0644, 0.0714, 0.0732, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0886, 0.0783, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 06:43:38,653 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:43:40,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:43:45,747 INFO [train.py:968] (1/2) Epoch 8, batch 20550, giga_loss[loss=0.2471, simple_loss=0.3312, pruned_loss=0.08153, over 29076.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3593, pruned_loss=0.109, over 5686989.55 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3605, pruned_loss=0.1033, over 5719169.52 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3577, pruned_loss=0.1089, over 5690000.47 frames. ], batch size: 155, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:43:50,801 INFO [optim.py:369] (1/2) 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 06:43:55,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2826, 1.8743, 1.3283, 0.6128], device='cuda:1'), covar=tensor([0.3406, 0.1600, 0.2238, 0.3656], device='cuda:1'), in_proj_covar=tensor([0.1456, 0.1384, 0.1439, 0.1190], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 06:43:57,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4299, 1.7841, 1.3518, 1.3069], device='cuda:1'), covar=tensor([0.1684, 0.1064, 0.1053, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.1600, 0.1430, 0.1417, 0.1535], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 06:44:04,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:44:16,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:44:26,335 INFO [train.py:968] (1/2) Epoch 8, batch 20600, giga_loss[loss=0.2961, simple_loss=0.3631, pruned_loss=0.1146, over 28566.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3594, pruned_loss=0.1089, over 5673376.57 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3604, pruned_loss=0.1035, over 5709541.58 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3581, pruned_loss=0.1089, over 5682831.48 frames. ], batch size: 85, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:45:08,769 INFO [train.py:968] (1/2) Epoch 8, batch 20650, giga_loss[loss=0.3158, simple_loss=0.3764, pruned_loss=0.1276, over 28905.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3626, pruned_loss=0.111, over 5674012.95 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3611, pruned_loss=0.1041, over 5705369.95 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3609, pruned_loss=0.1106, over 5684188.89 frames. ], batch size: 213, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:45:12,987 INFO [optim.py:369] (1/2) 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 06:45:14,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:45:34,155 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:45:34,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:45:47,410 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 20700, giga_loss[loss=0.2989, simple_loss=0.3638, pruned_loss=0.117, over 28766.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3647, pruned_loss=0.1124, over 5681288.28 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3616, pruned_loss=0.1044, over 5700903.70 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3629, pruned_loss=0.1121, over 5692452.80 frames. ], batch size: 284, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:45:59,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4497, 1.9960, 1.2906, 0.7923], device='cuda:1'), covar=tensor([0.3944, 0.1890, 0.2103, 0.3687], device='cuda:1'), in_proj_covar=tensor([0.1457, 0.1388, 0.1442, 0.1194], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 06:46:09,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 06:46:13,673 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:46:17,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:46:34,517 INFO [train.py:968] (1/2) Epoch 8, batch 20750, giga_loss[loss=0.2988, simple_loss=0.3594, pruned_loss=0.1191, over 28742.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3666, pruned_loss=0.1142, over 5678028.53 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3616, pruned_loss=0.1043, over 5701952.18 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3652, pruned_loss=0.114, over 5685811.62 frames. ], batch size: 92, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:46:40,599 INFO [optim.py:369] (1/2) 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 06:46:44,947 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:46:46,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:16,232 INFO [train.py:968] (1/2) Epoch 8, batch 20800, libri_loss[loss=0.2398, simple_loss=0.324, pruned_loss=0.07786, over 29569.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3671, pruned_loss=0.1153, over 5681448.26 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3607, pruned_loss=0.1037, over 5711149.68 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3671, pruned_loss=0.1162, over 5678111.99 frames. ], batch size: 74, lr: 4.06e-03, grad_scale: 4.0
+2023-03-04 06:47:17,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:19,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:37,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:39,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:42,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:46,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:49,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:47:55,932 INFO [train.py:968] (1/2) Epoch 8, batch 20850, libri_loss[loss=0.3005, simple_loss=0.3756, pruned_loss=0.1127, over 20032.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3664, pruned_loss=0.1146, over 5677850.48 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3611, pruned_loss=0.104, over 5700401.43 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3662, pruned_loss=0.1155, over 5684528.23 frames. ], batch size: 186, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:47:58,246 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:48:01,094 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:11,072 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:15,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3251, 1.4110, 1.2141, 1.4960], device='cuda:1'), covar=tensor([0.0749, 0.0329, 0.0325, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0119, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0078], device='cuda:1')
+2023-03-04 06:48:31,959 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4337, 3.6376, 1.4807, 1.5321], device='cuda:1'), covar=tensor([0.0924, 0.0226, 0.0852, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0484, 0.0320, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0018, 0.0023], device='cuda:1')
+2023-03-04 06:48:34,734 INFO [train.py:968] (1/2) Epoch 8, batch 20900, giga_loss[loss=0.2833, simple_loss=0.351, pruned_loss=0.1078, over 28783.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3666, pruned_loss=0.1142, over 5689503.57 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3614, pruned_loss=0.1044, over 5700397.10 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3663, pruned_loss=0.1147, over 5694314.24 frames. ], batch size: 119, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:48:39,809 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:41,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:48:43,274 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:04,450 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=339718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:07,872 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 06:49:13,908 INFO [train.py:968] (1/2) Epoch 8, batch 20950, giga_loss[loss=0.2683, simple_loss=0.3514, pruned_loss=0.09261, over 28657.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3665, pruned_loss=0.1133, over 5689420.44 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3618, pruned_loss=0.1046, over 5704559.80 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.366, pruned_loss=0.1137, over 5689342.85 frames. ], batch size: 242, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:49:20,387 INFO [optim.py:369] (1/2) 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 06:49:53,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:56,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:49:56,732 INFO [train.py:968] (1/2) Epoch 8, batch 21000, giga_loss[loss=0.3057, simple_loss=0.3715, pruned_loss=0.1199, over 27935.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3663, pruned_loss=0.1118, over 5695090.46 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3618, pruned_loss=0.1046, over 5705579.00 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.366, pruned_loss=0.1122, over 5694075.94 frames. ], batch size: 412, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:49:56,732 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 06:50:06,714 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 06:50:11,669 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:50:28,319 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 21050, giga_loss[loss=0.2961, simple_loss=0.3711, pruned_loss=0.1105, over 28882.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3647, pruned_loss=0.1109, over 5701910.17 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3622, pruned_loss=0.1049, over 5710670.72 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3641, pruned_loss=0.1111, over 5696492.67 frames. ], batch size: 112, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:50:51,596 INFO [optim.py:369] (1/2) 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 06:50:51,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=339836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:51:25,021 INFO [train.py:968] (1/2) Epoch 8, batch 21100, giga_loss[loss=0.2751, simple_loss=0.3448, pruned_loss=0.1028, over 28345.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3616, pruned_loss=0.1093, over 5707278.65 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3624, pruned_loss=0.1051, over 5711916.63 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3611, pruned_loss=0.1093, over 5701742.70 frames. ], batch size: 65, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:51:55,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9594, 1.1200, 1.0760, 0.8452], device='cuda:1'), covar=tensor([0.1188, 0.1350, 0.0784, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1427, 0.1405, 0.1517], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 06:52:04,562 INFO [train.py:968] (1/2) Epoch 8, batch 21150, giga_loss[loss=0.2387, simple_loss=0.3241, pruned_loss=0.07671, over 28567.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3608, pruned_loss=0.1091, over 5712070.17 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3623, pruned_loss=0.1052, over 5713126.71 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3604, pruned_loss=0.1092, over 5706382.26 frames. ], batch size: 78, lr: 4.06e-03, grad_scale: 2.0
+2023-03-04 06:52:05,387 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=339930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:52:09,262 INFO [optim.py:369] (1/2) 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 06:52:47,573 INFO [train.py:968] (1/2) Epoch 8, batch 21200, giga_loss[loss=0.272, simple_loss=0.3499, pruned_loss=0.0971, over 29046.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3602, pruned_loss=0.1093, over 5708775.66 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3628, pruned_loss=0.1055, over 5716282.10 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3594, pruned_loss=0.1091, over 5701475.05 frames. ], batch size: 128, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:52:47,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=339979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:52:50,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=339982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:53:13,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:53:29,287 INFO [train.py:968] (1/2) Epoch 8, batch 21250, giga_loss[loss=0.304, simple_loss=0.3712, pruned_loss=0.1184, over 28779.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3614, pruned_loss=0.1099, over 5717248.36 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3633, pruned_loss=0.106, over 5719012.99 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3602, pruned_loss=0.1095, over 5708927.39 frames. ], batch size: 284, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:53:34,084 INFO [optim.py:369] (1/2) 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 06:53:46,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4496, 1.5645, 1.2743, 1.2181], device='cuda:1'), covar=tensor([0.1489, 0.1357, 0.1056, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.1600, 0.1437, 0.1416, 0.1531], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 06:53:57,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:54:07,040 INFO [train.py:968] (1/2) Epoch 8, batch 21300, giga_loss[loss=0.2455, simple_loss=0.3323, pruned_loss=0.07933, over 28426.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3615, pruned_loss=0.1099, over 5713386.87 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3637, pruned_loss=0.1064, over 5722692.79 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3603, pruned_loss=0.1092, over 5703380.35 frames. ], batch size: 71, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:54:29,269 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=340107.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 06:54:47,451 INFO [train.py:968] (1/2) Epoch 8, batch 21350, giga_loss[loss=0.2927, simple_loss=0.3656, pruned_loss=0.1099, over 29052.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3596, pruned_loss=0.1078, over 5717513.37 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3639, pruned_loss=0.1067, over 5724145.03 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3583, pruned_loss=0.107, over 5708313.34 frames. ], batch size: 155, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:54:54,598 INFO [optim.py:369] (1/2) 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 06:55:03,756 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=340145.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 06:55:15,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:55:29,643 INFO [train.py:968] (1/2) Epoch 8, batch 21400, giga_loss[loss=0.2461, simple_loss=0.3292, pruned_loss=0.08153, over 28678.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3587, pruned_loss=0.1071, over 5726757.26 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3641, pruned_loss=0.1069, over 5727487.67 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3574, pruned_loss=0.1064, over 5716296.90 frames. ], batch size: 78, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:55:32,258 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 06:55:55,354 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:55:58,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:56:10,137 INFO [train.py:968] (1/2) Epoch 8, batch 21450, giga_loss[loss=0.2726, simple_loss=0.3388, pruned_loss=0.1032, over 28941.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3567, pruned_loss=0.1067, over 5726221.35 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3642, pruned_loss=0.1071, over 5727708.67 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3556, pruned_loss=0.1058, over 5717670.35 frames. ], batch size: 106, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:56:10,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 06:56:15,677 INFO [optim.py:369] (1/2) 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 06:56:21,944 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:56:53,706 INFO [train.py:968] (1/2) Epoch 8, batch 21500, giga_loss[loss=0.2609, simple_loss=0.3405, pruned_loss=0.09067, over 28907.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3535, pruned_loss=0.105, over 5721460.86 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3646, pruned_loss=0.1075, over 5728590.13 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3521, pruned_loss=0.1039, over 5713914.02 frames. ], batch size: 174, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:57:13,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:57:14,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:57:15,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:57:29,843 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 06:57:32,487 INFO [train.py:968] (1/2) Epoch 8, batch 21550, giga_loss[loss=0.2779, simple_loss=0.3446, pruned_loss=0.1056, over 28665.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3538, pruned_loss=0.1057, over 5714752.81 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3655, pruned_loss=0.1083, over 5718634.93 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3515, pruned_loss=0.1041, over 5717726.45 frames. ], batch size: 60, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:57:37,383 INFO [optim.py:369] (1/2) 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 06:57:37,638 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:58:13,029 INFO [train.py:968] (1/2) Epoch 8, batch 21600, giga_loss[loss=0.2928, simple_loss=0.3596, pruned_loss=0.113, over 29016.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3543, pruned_loss=0.1067, over 5714121.83 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3654, pruned_loss=0.1083, over 5720787.55 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3524, pruned_loss=0.1053, over 5714509.92 frames. ], batch size: 106, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 06:58:41,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9008, 1.7916, 1.7352, 1.5874], device='cuda:1'), covar=tensor([0.1137, 0.2125, 0.1700, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0722, 0.0647, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 06:58:52,816 INFO [train.py:968] (1/2) Epoch 8, batch 21650, giga_loss[loss=0.2671, simple_loss=0.3344, pruned_loss=0.09986, over 28949.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3531, pruned_loss=0.107, over 5711194.25 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3655, pruned_loss=0.1085, over 5717081.20 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3512, pruned_loss=0.1056, over 5714889.93 frames. ], batch size: 136, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 06:58:58,285 INFO [optim.py:369] (1/2) 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 06:58:59,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5214, 0.9259, 2.9031, 2.7000], device='cuda:1'), covar=tensor([0.1735, 0.2380, 0.0477, 0.0706], device='cuda:1'), in_proj_covar=tensor([0.0599, 0.0552, 0.0787, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0007, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 06:59:07,189 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:59:09,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:59:24,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2779, 2.5193, 1.2775, 1.3331], device='cuda:1'), covar=tensor([0.0828, 0.0388, 0.0819, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0486, 0.0321, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0027, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 06:59:30,569 INFO [train.py:968] (1/2) Epoch 8, batch 21700, giga_loss[loss=0.2622, simple_loss=0.3319, pruned_loss=0.09626, over 29066.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3527, pruned_loss=0.1073, over 5710659.98 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3662, pruned_loss=0.1093, over 5717553.27 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3499, pruned_loss=0.1056, over 5712728.69 frames. ], batch size: 128, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 06:59:31,541 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 06:59:33,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340482.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:00:02,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=340520.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:00:10,654 INFO [train.py:968] (1/2) Epoch 8, batch 21750, giga_loss[loss=0.3114, simple_loss=0.369, pruned_loss=0.1269, over 28817.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3508, pruned_loss=0.1068, over 5708607.00 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3671, pruned_loss=0.1101, over 5719229.46 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3475, pruned_loss=0.1046, over 5708701.45 frames. ], batch size: 112, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:00:16,039 INFO [optim.py:369] (1/2) 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 07:00:48,753 INFO [train.py:968] (1/2) Epoch 8, batch 21800, giga_loss[loss=0.24, simple_loss=0.3173, pruned_loss=0.08138, over 28547.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3481, pruned_loss=0.1053, over 5703088.42 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3677, pruned_loss=0.1106, over 5711188.00 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3446, pruned_loss=0.103, over 5709801.14 frames. ], batch size: 60, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:01:29,536 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340625.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:01:32,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340628.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:01:32,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2599, 1.5211, 1.3257, 1.0599], device='cuda:1'), covar=tensor([0.2165, 0.2088, 0.2302, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.1213, 0.0911, 0.1065, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 07:01:32,587 INFO [train.py:968] (1/2) Epoch 8, batch 21850, giga_loss[loss=0.289, simple_loss=0.3497, pruned_loss=0.1141, over 28970.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3468, pruned_loss=0.1045, over 5699789.94 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3677, pruned_loss=0.1108, over 5710640.64 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3438, pruned_loss=0.1025, over 5705371.15 frames. ], batch size: 106, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:01:38,892 INFO [optim.py:369] (1/2) 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 07:01:59,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340657.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:02:03,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=340663.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:02:07,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=340666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:02:15,830 INFO [train.py:968] (1/2) Epoch 8, batch 21900, giga_loss[loss=0.3073, simple_loss=0.3792, pruned_loss=0.1177, over 27922.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3495, pruned_loss=0.1057, over 5691397.13 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3681, pruned_loss=0.1111, over 5703245.37 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3467, pruned_loss=0.1038, over 5702926.29 frames. ], batch size: 412, lr: 4.05e-03, grad_scale: 2.0
+2023-03-04 07:02:30,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=340695.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:03:01,437 INFO [train.py:968] (1/2) Epoch 8, batch 21950, giga_loss[loss=0.2873, simple_loss=0.3623, pruned_loss=0.1062, over 27939.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3521, pruned_loss=0.1065, over 5700588.41 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3677, pruned_loss=0.111, over 5706373.40 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3499, pruned_loss=0.1051, over 5706840.19 frames. ], batch size: 412, lr: 4.05e-03, grad_scale: 2.0
+2023-03-04 07:03:08,310 INFO [optim.py:369] (1/2) 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 07:03:16,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6660, 5.4235, 5.1858, 2.5305], device='cuda:1'), covar=tensor([0.0341, 0.0519, 0.0609, 0.1742], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0888, 0.0786, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 07:03:22,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2882, 1.8528, 1.3963, 0.6688], device='cuda:1'), covar=tensor([0.2965, 0.1543, 0.2453, 0.3348], device='cuda:1'), in_proj_covar=tensor([0.1463, 0.1381, 0.1443, 0.1190], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:03:24,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-04 07:03:41,996 INFO [train.py:968] (1/2) Epoch 8, batch 22000, giga_loss[loss=0.2741, simple_loss=0.3529, pruned_loss=0.09765, over 28852.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3548, pruned_loss=0.1074, over 5706216.81 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3681, pruned_loss=0.1115, over 5712975.55 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3521, pruned_loss=0.1056, over 5705158.39 frames. ], batch size: 186, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:03:52,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4849, 1.6382, 1.8194, 1.4304], device='cuda:1'), covar=tensor([0.1214, 0.1620, 0.0988, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0702, 0.0819, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 07:04:21,753 INFO [train.py:968] (1/2) Epoch 8, batch 22050, giga_loss[loss=0.26, simple_loss=0.3438, pruned_loss=0.08807, over 28945.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3551, pruned_loss=0.1071, over 5697196.93 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3686, pruned_loss=0.1121, over 5708049.09 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3523, pruned_loss=0.105, over 5700062.65 frames. ], batch size: 174, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:04:30,093 INFO [optim.py:369] (1/2) 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 07:05:06,501 INFO [train.py:968] (1/2) Epoch 8, batch 22100, giga_loss[loss=0.349, simple_loss=0.4, pruned_loss=0.149, over 28690.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3546, pruned_loss=0.1066, over 5691225.72 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3692, pruned_loss=0.1127, over 5708590.91 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3516, pruned_loss=0.1043, over 5692919.88 frames. ], batch size: 242, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:05:39,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=340918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:05:47,266 INFO [train.py:968] (1/2) Epoch 8, batch 22150, giga_loss[loss=0.2428, simple_loss=0.3198, pruned_loss=0.08288, over 28433.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3544, pruned_loss=0.1065, over 5700611.94 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3693, pruned_loss=0.1129, over 5708028.30 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3516, pruned_loss=0.1043, over 5701954.25 frames. ], batch size: 60, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:05:54,821 INFO [optim.py:369] (1/2) 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 07:05:59,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3236, 1.5524, 1.3157, 1.5478], device='cuda:1'), covar=tensor([0.0717, 0.0297, 0.0315, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0118, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0050, 0.0046, 0.0077], device='cuda:1')
+2023-03-04 07:06:28,229 INFO [train.py:968] (1/2) Epoch 8, batch 22200, giga_loss[loss=0.446, simple_loss=0.4756, pruned_loss=0.2081, over 27587.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3556, pruned_loss=0.1078, over 5702364.49 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3696, pruned_loss=0.1133, over 5712186.31 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3528, pruned_loss=0.1056, over 5699806.25 frames. ], batch size: 472, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:06:31,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 07:06:41,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3666, 4.1915, 3.9229, 1.9357], device='cuda:1'), covar=tensor([0.0441, 0.0542, 0.0608, 0.1920], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0883, 0.0787, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 07:07:07,749 INFO [train.py:968] (1/2) Epoch 8, batch 22250, giga_loss[loss=0.2827, simple_loss=0.3554, pruned_loss=0.105, over 28613.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3584, pruned_loss=0.1095, over 5703748.76 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3699, pruned_loss=0.1137, over 5717153.52 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3555, pruned_loss=0.1073, over 5697035.57 frames. ], batch size: 85, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:07:13,675 INFO [optim.py:369] (1/2) 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 07:07:44,165 INFO [train.py:968] (1/2) Epoch 8, batch 22300, giga_loss[loss=0.3446, simple_loss=0.405, pruned_loss=0.1421, over 28049.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3611, pruned_loss=0.1112, over 5711008.98 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3703, pruned_loss=0.1142, over 5722654.31 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3581, pruned_loss=0.1088, over 5700477.75 frames. ], batch size: 412, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:08:26,382 INFO [train.py:968] (1/2) Epoch 8, batch 22350, giga_loss[loss=0.28, simple_loss=0.3552, pruned_loss=0.1024, over 28912.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3632, pruned_loss=0.1123, over 5711959.28 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3711, pruned_loss=0.1149, over 5724775.64 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3599, pruned_loss=0.1098, over 5701628.08 frames. ], batch size: 145, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:08:33,155 INFO [optim.py:369] (1/2) 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 07:08:42,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0894, 1.8305, 1.7429, 1.5980], device='cuda:1'), covar=tensor([0.1262, 0.2457, 0.1999, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0724, 0.0652, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 07:08:52,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4739, 1.7055, 1.7467, 1.3344], device='cuda:1'), covar=tensor([0.1594, 0.1941, 0.1263, 0.1378], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0703, 0.0818, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 07:09:05,548 INFO [train.py:968] (1/2) Epoch 8, batch 22400, giga_loss[loss=0.2884, simple_loss=0.3613, pruned_loss=0.1077, over 28932.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3636, pruned_loss=0.112, over 5717507.39 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3714, pruned_loss=0.115, over 5725410.76 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.3609, pruned_loss=0.1099, over 5708815.80 frames. ], batch size: 136, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:09:06,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0042, 1.9091, 1.7908, 1.7094], device='cuda:1'), covar=tensor([0.1334, 0.2346, 0.1785, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0723, 0.0650, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 07:09:13,741 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-04 07:09:27,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 07:09:49,613 INFO [train.py:968] (1/2) Epoch 8, batch 22450, giga_loss[loss=0.3009, simple_loss=0.359, pruned_loss=0.1214, over 28694.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.365, pruned_loss=0.1132, over 5713304.93 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3718, pruned_loss=0.1155, over 5724615.49 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3624, pruned_loss=0.1112, over 5706708.99 frames. ], batch size: 92, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:09:57,927 INFO [optim.py:369] (1/2) 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 07:10:30,921 INFO [train.py:968] (1/2) Epoch 8, batch 22500, giga_loss[loss=0.326, simple_loss=0.3805, pruned_loss=0.1358, over 27940.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3644, pruned_loss=0.1128, over 5714554.44 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3723, pruned_loss=0.1158, over 5729436.01 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3616, pruned_loss=0.1108, over 5704541.89 frames. ], batch size: 412, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:10:43,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=341293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:10:43,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4662, 1.6336, 1.5777, 1.4564], device='cuda:1'), covar=tensor([0.1284, 0.1669, 0.1769, 0.1657], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0723, 0.0651, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 07:11:13,323 INFO [train.py:968] (1/2) Epoch 8, batch 22550, giga_loss[loss=0.2621, simple_loss=0.3391, pruned_loss=0.0926, over 28770.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3606, pruned_loss=0.1105, over 5717686.87 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3726, pruned_loss=0.1161, over 5732831.73 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.358, pruned_loss=0.1086, over 5706541.74 frames. ], batch size: 242, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:11:22,074 INFO [optim.py:369] (1/2) 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 07:11:24,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5248, 1.6159, 1.3557, 1.9342], device='cuda:1'), covar=tensor([0.2065, 0.2149, 0.2308, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.1212, 0.0911, 0.1066, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 07:11:55,105 INFO [train.py:968] (1/2) Epoch 8, batch 22600, giga_loss[loss=0.2635, simple_loss=0.3423, pruned_loss=0.09234, over 28721.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3562, pruned_loss=0.1082, over 5719068.78 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3729, pruned_loss=0.1164, over 5734819.78 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3537, pruned_loss=0.1063, over 5708286.34 frames. ], batch size: 262, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:12:05,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3478, 1.1995, 4.8887, 3.3854], device='cuda:1'), covar=tensor([0.1627, 0.2566, 0.0330, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0603, 0.0557, 0.0798, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 07:12:33,979 INFO [train.py:968] (1/2) Epoch 8, batch 22650, giga_loss[loss=0.2849, simple_loss=0.3645, pruned_loss=0.1026, over 28941.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.355, pruned_loss=0.1071, over 5711314.30 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3733, pruned_loss=0.1169, over 5728960.07 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3523, pruned_loss=0.1049, over 5707672.79 frames. ], batch size: 174, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:12:39,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9822, 1.2752, 0.9527, 0.2438], device='cuda:1'), covar=tensor([0.1736, 0.1492, 0.2398, 0.3395], device='cuda:1'), in_proj_covar=tensor([0.1450, 0.1374, 0.1435, 0.1184], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:12:40,748 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=341436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:12:43,294 INFO [optim.py:369] (1/2) 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 07:12:43,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=341439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:12:45,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0015, 1.8250, 1.3953, 1.4917], device='cuda:1'), covar=tensor([0.0563, 0.0573, 0.0897, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0442, 0.0492, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 07:13:10,991 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=341468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:13:18,752 INFO [train.py:968] (1/2) Epoch 8, batch 22700, libri_loss[loss=0.3356, simple_loss=0.3975, pruned_loss=0.1369, over 29479.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3559, pruned_loss=0.1058, over 5705141.67 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3738, pruned_loss=0.1173, over 5732324.64 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3529, pruned_loss=0.1035, over 5698761.16 frames. ], batch size: 85, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:13:58,321 INFO [train.py:968] (1/2) Epoch 8, batch 22750, giga_loss[loss=0.3779, simple_loss=0.4212, pruned_loss=0.1673, over 26811.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3578, pruned_loss=0.1064, over 5710038.40 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3736, pruned_loss=0.1175, over 5736602.68 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 5700144.79 frames. ], batch size: 555, lr: 4.05e-03, grad_scale: 4.0
+2023-03-04 07:14:05,805 INFO [optim.py:369] (1/2) 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 07:14:38,276 INFO [train.py:968] (1/2) Epoch 8, batch 22800, giga_loss[loss=0.3158, simple_loss=0.3612, pruned_loss=0.1352, over 28330.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3557, pruned_loss=0.1064, over 5701777.98 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3738, pruned_loss=0.1177, over 5734677.75 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3531, pruned_loss=0.1043, over 5695387.78 frames. ], batch size: 65, lr: 4.05e-03, grad_scale: 8.0
+2023-03-04 07:15:21,419 INFO [train.py:968] (1/2) Epoch 8, batch 22850, giga_loss[loss=0.2554, simple_loss=0.3187, pruned_loss=0.09603, over 28703.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3543, pruned_loss=0.1072, over 5701663.26 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3742, pruned_loss=0.118, over 5732648.58 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3519, pruned_loss=0.1052, over 5698090.70 frames. ], batch size: 99, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:15:27,962 INFO [optim.py:369] (1/2) 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 07:16:00,789 INFO [train.py:968] (1/2) Epoch 8, batch 22900, giga_loss[loss=0.2717, simple_loss=0.346, pruned_loss=0.09869, over 29078.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3526, pruned_loss=0.1071, over 5714401.48 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3743, pruned_loss=0.1182, over 5736912.94 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3501, pruned_loss=0.105, over 5707232.11 frames. ], batch size: 155, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:16:40,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2412, 1.8217, 1.4481, 0.3535], device='cuda:1'), covar=tensor([0.2404, 0.1491, 0.2491, 0.3479], device='cuda:1'), in_proj_covar=tensor([0.1450, 0.1371, 0.1434, 0.1191], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:16:43,736 INFO [train.py:968] (1/2) Epoch 8, batch 22950, giga_loss[loss=0.3201, simple_loss=0.3762, pruned_loss=0.132, over 28911.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3521, pruned_loss=0.1086, over 5710761.45 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3744, pruned_loss=0.1186, over 5738680.13 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3497, pruned_loss=0.1064, over 5702926.29 frames. ], batch size: 174, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:16:50,888 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 23000, giga_loss[loss=0.2303, simple_loss=0.3049, pruned_loss=0.07784, over 28265.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3521, pruned_loss=0.1082, over 5723148.77 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3745, pruned_loss=0.119, over 5743555.10 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3493, pruned_loss=0.1058, over 5711652.31 frames. ], batch size: 77, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:17:32,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2379, 1.8380, 1.4303, 1.4749], device='cuda:1'), covar=tensor([0.0765, 0.0296, 0.0335, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0078], device='cuda:1')
+2023-03-04 07:17:58,773 INFO [train.py:968] (1/2) Epoch 8, batch 23050, giga_loss[loss=0.2387, simple_loss=0.3088, pruned_loss=0.08428, over 28893.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3487, pruned_loss=0.107, over 5721936.05 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3747, pruned_loss=0.1195, over 5747721.86 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3455, pruned_loss=0.1042, over 5708376.15 frames. ], batch size: 112, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:18:06,752 INFO [optim.py:369] (1/2) 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 07:18:13,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3046, 3.1196, 2.9466, 1.3434], device='cuda:1'), covar=tensor([0.0755, 0.0888, 0.0855, 0.2181], device='cuda:1'), in_proj_covar=tensor([0.0952, 0.0898, 0.0796, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 07:18:32,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2913, 2.6040, 1.2716, 1.3917], device='cuda:1'), covar=tensor([0.0884, 0.0357, 0.0895, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0500, 0.0324, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 07:18:38,210 INFO [train.py:968] (1/2) Epoch 8, batch 23100, libri_loss[loss=0.3404, simple_loss=0.4061, pruned_loss=0.1373, over 25710.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3455, pruned_loss=0.1054, over 5714942.65 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.375, pruned_loss=0.1199, over 5745501.66 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.342, pruned_loss=0.1026, over 5705756.44 frames. ], batch size: 136, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:18:54,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4639, 3.1959, 1.4027, 1.4577], device='cuda:1'), covar=tensor([0.0867, 0.0321, 0.0938, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0500, 0.0324, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 07:19:15,985 INFO [train.py:968] (1/2) Epoch 8, batch 23150, giga_loss[loss=0.2566, simple_loss=0.334, pruned_loss=0.08966, over 28709.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3422, pruned_loss=0.103, over 5714658.63 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3749, pruned_loss=0.1199, over 5744620.06 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3392, pruned_loss=0.1005, over 5707714.99 frames. ], batch size: 242, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:19:25,542 INFO [optim.py:369] (1/2) 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 07:19:29,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3422, 1.5700, 1.2359, 1.4958], device='cuda:1'), covar=tensor([0.2208, 0.2111, 0.2416, 0.2105], device='cuda:1'), in_proj_covar=tensor([0.1214, 0.0903, 0.1061, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 07:19:57,972 INFO [train.py:968] (1/2) Epoch 8, batch 23200, giga_loss[loss=0.3001, simple_loss=0.3732, pruned_loss=0.1136, over 28695.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3439, pruned_loss=0.1034, over 5716202.30 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3753, pruned_loss=0.1205, over 5748222.04 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3403, pruned_loss=0.1005, over 5706788.11 frames. ], batch size: 262, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:20:40,443 INFO [train.py:968] (1/2) Epoch 8, batch 23250, giga_loss[loss=0.2716, simple_loss=0.352, pruned_loss=0.09561, over 28712.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3469, pruned_loss=0.1043, over 5717094.95 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3754, pruned_loss=0.1205, over 5748894.93 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.344, pruned_loss=0.102, over 5709046.80 frames. ], batch size: 262, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:20:49,513 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 23300, giga_loss[loss=0.2995, simple_loss=0.37, pruned_loss=0.1145, over 29030.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3509, pruned_loss=0.1059, over 5715798.29 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3758, pruned_loss=0.1209, over 5750458.04 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3479, pruned_loss=0.1036, over 5707784.90 frames. ], batch size: 155, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:21:53,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4226, 3.4212, 1.5229, 1.4910], device='cuda:1'), covar=tensor([0.0897, 0.0321, 0.0900, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0501, 0.0325, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 07:22:00,727 INFO [train.py:968] (1/2) Epoch 8, batch 23350, giga_loss[loss=0.3819, simple_loss=0.4268, pruned_loss=0.1685, over 28261.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3551, pruned_loss=0.1084, over 5712657.30 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.376, pruned_loss=0.1214, over 5752211.45 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3518, pruned_loss=0.1055, over 5703362.88 frames. ], batch size: 368, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:22:01,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5630, 1.8972, 1.9442, 1.4123], device='cuda:1'), covar=tensor([0.1564, 0.1835, 0.1186, 0.1349], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0705, 0.0826, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 07:22:10,458 INFO [optim.py:369] (1/2) 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 07:22:44,009 INFO [train.py:968] (1/2) Epoch 8, batch 23400, giga_loss[loss=0.2604, simple_loss=0.3465, pruned_loss=0.08713, over 28892.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3576, pruned_loss=0.1097, over 5704254.53 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.376, pruned_loss=0.1215, over 5753661.81 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3548, pruned_loss=0.1073, over 5695153.69 frames. ], batch size: 174, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:23:26,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6017, 2.3300, 1.6524, 0.6688], device='cuda:1'), covar=tensor([0.2918, 0.1581, 0.2527, 0.3308], device='cuda:1'), in_proj_covar=tensor([0.1458, 0.1380, 0.1441, 0.1193], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:23:30,360 INFO [train.py:968] (1/2) Epoch 8, batch 23450, giga_loss[loss=0.3386, simple_loss=0.3908, pruned_loss=0.1432, over 28833.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3641, pruned_loss=0.1157, over 5703131.50 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.376, pruned_loss=0.1216, over 5757158.62 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3616, pruned_loss=0.1135, over 5691137.98 frames. ], batch size: 112, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:23:37,292 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:23:42,639 INFO [optim.py:369] (1/2) 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 07:24:23,444 INFO [train.py:968] (1/2) Epoch 8, batch 23500, giga_loss[loss=0.348, simple_loss=0.3966, pruned_loss=0.1497, over 28861.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3711, pruned_loss=0.1215, over 5696879.16 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3762, pruned_loss=0.1217, over 5758508.54 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3689, pruned_loss=0.1197, over 5685692.23 frames. ], batch size: 136, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:25:15,819 INFO [train.py:968] (1/2) Epoch 8, batch 23550, giga_loss[loss=0.3029, simple_loss=0.3717, pruned_loss=0.1171, over 28290.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3792, pruned_loss=0.128, over 5686609.44 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3764, pruned_loss=0.1221, over 5755585.78 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3773, pruned_loss=0.1262, over 5679793.63 frames. ], batch size: 77, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:25:28,139 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 23600, giga_loss[loss=0.3226, simple_loss=0.3789, pruned_loss=0.1331, over 28756.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3851, pruned_loss=0.1336, over 5681340.23 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3763, pruned_loss=0.1221, over 5755362.73 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3837, pruned_loss=0.1323, over 5675629.53 frames. ], batch size: 92, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:26:04,353 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:26:08,764 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 07:26:49,967 INFO [train.py:968] (1/2) Epoch 8, batch 23650, giga_loss[loss=0.4692, simple_loss=0.4784, pruned_loss=0.2301, over 27543.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3916, pruned_loss=0.1395, over 5670702.07 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3761, pruned_loss=0.1221, over 5759667.86 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3911, pruned_loss=0.1391, over 5659132.07 frames. ], batch size: 472, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:27:01,918 INFO [optim.py:369] (1/2) 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 07:27:40,040 INFO [train.py:968] (1/2) Epoch 8, batch 23700, giga_loss[loss=0.3419, simple_loss=0.3978, pruned_loss=0.143, over 28658.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3948, pruned_loss=0.1418, over 5665656.32 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3764, pruned_loss=0.1223, over 5749289.71 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3945, pruned_loss=0.1415, over 5663790.64 frames. ], batch size: 262, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:27:54,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1031, 3.9284, 3.7529, 1.7931], device='cuda:1'), covar=tensor([0.0475, 0.0648, 0.0627, 0.2147], device='cuda:1'), in_proj_covar=tensor([0.0956, 0.0901, 0.0795, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 07:28:08,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2633, 1.6508, 1.2218, 0.7510], device='cuda:1'), covar=tensor([0.2147, 0.1517, 0.1558, 0.2713], device='cuda:1'), in_proj_covar=tensor([0.1478, 0.1403, 0.1452, 0.1206], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:28:29,566 INFO [train.py:968] (1/2) Epoch 8, batch 23750, libri_loss[loss=0.3689, simple_loss=0.4173, pruned_loss=0.1603, over 29494.00 frames. ], tot_loss[loss=0.3431, simple_loss=0.3973, pruned_loss=0.1445, over 5662607.50 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3768, pruned_loss=0.1227, over 5751438.59 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3969, pruned_loss=0.1442, over 5658144.48 frames. ], batch size: 85, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:28:42,717 INFO [optim.py:369] (1/2) 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 07:28:50,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5555, 2.1907, 1.6319, 0.6084], device='cuda:1'), covar=tensor([0.2837, 0.1478, 0.2019, 0.3521], device='cuda:1'), in_proj_covar=tensor([0.1478, 0.1403, 0.1453, 0.1207], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:29:19,711 INFO [train.py:968] (1/2) Epoch 8, batch 23800, giga_loss[loss=0.3908, simple_loss=0.4315, pruned_loss=0.1751, over 28644.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3994, pruned_loss=0.1472, over 5644642.64 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3774, pruned_loss=0.123, over 5753036.24 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.3991, pruned_loss=0.1472, over 5637451.29 frames. ], batch size: 307, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:29:47,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=342609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:29:51,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 07:30:07,921 INFO [train.py:968] (1/2) Epoch 8, batch 23850, giga_loss[loss=0.3776, simple_loss=0.4163, pruned_loss=0.1694, over 28928.00 frames. ], tot_loss[loss=0.35, simple_loss=0.4012, pruned_loss=0.1494, over 5646234.49 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3771, pruned_loss=0.123, over 5746486.43 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4023, pruned_loss=0.1506, over 5642313.90 frames. ], batch size: 227, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:30:12,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3162, 1.2641, 1.1295, 1.0673], device='cuda:1'), covar=tensor([0.0563, 0.0429, 0.0937, 0.0735], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0448, 0.0495, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 07:30:19,518 INFO [optim.py:369] (1/2) 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 07:30:27,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2671, 1.4305, 1.2526, 1.0681], device='cuda:1'), covar=tensor([0.1436, 0.1378, 0.0912, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1465, 0.1427, 0.1537], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 07:31:01,242 INFO [train.py:968] (1/2) Epoch 8, batch 23900, giga_loss[loss=0.4424, simple_loss=0.4412, pruned_loss=0.2218, over 23713.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.4067, pruned_loss=0.1548, over 5623466.55 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3778, pruned_loss=0.1238, over 5751045.51 frames. ], giga_tot_loss[loss=0.3601, simple_loss=0.4079, pruned_loss=0.1561, over 5612282.33 frames. ], batch size: 705, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:31:28,055 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:31:53,546 INFO [train.py:968] (1/2) Epoch 8, batch 23950, giga_loss[loss=0.3223, simple_loss=0.3823, pruned_loss=0.1311, over 28867.00 frames. ], tot_loss[loss=0.3582, simple_loss=0.4059, pruned_loss=0.1553, over 5621077.55 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.378, pruned_loss=0.1241, over 5752340.92 frames. ], giga_tot_loss[loss=0.36, simple_loss=0.4071, pruned_loss=0.1565, over 5609563.04 frames. ], batch size: 186, lr: 4.04e-03, grad_scale: 2.0
+2023-03-04 07:32:09,957 INFO [optim.py:369] (1/2) 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 07:32:18,884 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=342752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:32:21,166 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=342755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:32:22,889 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 24000, giga_loss[loss=0.2995, simple_loss=0.3568, pruned_loss=0.1211, over 28654.00 frames. ], tot_loss[loss=0.3549, simple_loss=0.4031, pruned_loss=0.1534, over 5628585.09 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3782, pruned_loss=0.1243, over 5745811.46 frames. ], giga_tot_loss[loss=0.3568, simple_loss=0.4043, pruned_loss=0.1546, over 5623758.79 frames. ], batch size: 85, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:32:44,807 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 07:32:49,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2341, 1.8474, 1.3487, 0.4297], device='cuda:1'), covar=tensor([0.2799, 0.1820, 0.3120, 0.3572], device='cuda:1'), in_proj_covar=tensor([0.1472, 0.1400, 0.1446, 0.1200], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:32:53,830 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 07:32:58,297 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=342784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:33:36,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3168, 1.9216, 1.4146, 0.5096], device='cuda:1'), covar=tensor([0.3133, 0.1580, 0.2366, 0.3615], device='cuda:1'), in_proj_covar=tensor([0.1467, 0.1400, 0.1444, 0.1195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:33:38,673 INFO [train.py:968] (1/2) Epoch 8, batch 24050, giga_loss[loss=0.3709, simple_loss=0.4198, pruned_loss=0.161, over 28723.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.4033, pruned_loss=0.1526, over 5630410.80 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3782, pruned_loss=0.1244, over 5749809.77 frames. ], giga_tot_loss[loss=0.3567, simple_loss=0.4049, pruned_loss=0.1543, over 5620203.78 frames. ], batch size: 284, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:33:52,013 INFO [optim.py:369] (1/2) 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 07:34:10,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-04 07:34:28,648 INFO [train.py:968] (1/2) Epoch 8, batch 24100, giga_loss[loss=0.3654, simple_loss=0.4181, pruned_loss=0.1563, over 28294.00 frames. ], tot_loss[loss=0.3547, simple_loss=0.4041, pruned_loss=0.1527, over 5622596.29 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3782, pruned_loss=0.1246, over 5754135.39 frames. ], giga_tot_loss[loss=0.3576, simple_loss=0.406, pruned_loss=0.1546, over 5607293.70 frames. ], batch size: 368, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:34:46,667 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 07:34:48,105 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=342899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:34:50,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=342902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:35:15,285 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=342928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:35:15,732 INFO [train.py:968] (1/2) Epoch 8, batch 24150, giga_loss[loss=0.3012, simple_loss=0.3709, pruned_loss=0.1157, over 28775.00 frames. ], tot_loss[loss=0.3536, simple_loss=0.4038, pruned_loss=0.1516, over 5630231.10 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3781, pruned_loss=0.1246, over 5756490.06 frames. ], giga_tot_loss[loss=0.357, simple_loss=0.4062, pruned_loss=0.1539, over 5612583.58 frames. ], batch size: 99, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:35:17,707 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=342931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:35:20,824 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:35:35,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6142, 1.6999, 1.2338, 1.3641], device='cuda:1'), covar=tensor([0.0586, 0.0393, 0.0920, 0.0760], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0448, 0.0495, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 07:36:10,339 INFO [train.py:968] (1/2) Epoch 8, batch 24200, giga_loss[loss=0.3797, simple_loss=0.4041, pruned_loss=0.1777, over 24020.00 frames. ], tot_loss[loss=0.3517, simple_loss=0.4025, pruned_loss=0.1505, over 5627548.25 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3784, pruned_loss=0.1249, over 5757368.46 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4044, pruned_loss=0.1524, over 5611171.14 frames. ], batch size: 705, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:36:24,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9337, 5.1434, 2.0271, 2.3292], device='cuda:1'), covar=tensor([0.0862, 0.0220, 0.0785, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0497, 0.0323, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 07:36:58,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 07:36:58,985 INFO [train.py:968] (1/2) Epoch 8, batch 24250, giga_loss[loss=0.3773, simple_loss=0.4123, pruned_loss=0.1711, over 27579.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3993, pruned_loss=0.1468, over 5637981.09 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3786, pruned_loss=0.1251, over 5760562.29 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4011, pruned_loss=0.1485, over 5619637.07 frames. ], batch size: 472, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:37:11,209 INFO [optim.py:369] (1/2) 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 07:37:45,810 INFO [train.py:968] (1/2) Epoch 8, batch 24300, giga_loss[loss=0.3953, simple_loss=0.4167, pruned_loss=0.187, over 23677.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3959, pruned_loss=0.1438, over 5636042.79 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3779, pruned_loss=0.1251, over 5763138.65 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3986, pruned_loss=0.1459, over 5615494.38 frames. ], batch size: 705, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:37:46,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:37:51,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=343086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:38:18,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4341, 2.0949, 1.5519, 0.5939], device='cuda:1'), covar=tensor([0.3217, 0.1679, 0.2532, 0.4102], device='cuda:1'), in_proj_covar=tensor([0.1481, 0.1410, 0.1455, 0.1213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 07:38:33,478 INFO [train.py:968] (1/2) Epoch 8, batch 24350, giga_loss[loss=0.3298, simple_loss=0.3969, pruned_loss=0.1314, over 28716.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3923, pruned_loss=0.1408, over 5640188.71 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3777, pruned_loss=0.1252, over 5761271.18 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.3948, pruned_loss=0.1425, over 5623459.60 frames. ], batch size: 242, lr: 4.04e-03, grad_scale: 4.0
+2023-03-04 07:38:45,644 INFO [optim.py:369] (1/2) 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 07:38:58,181 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-04 07:39:22,111 INFO [train.py:968] (1/2) Epoch 8, batch 24400, giga_loss[loss=0.3168, simple_loss=0.3826, pruned_loss=0.1255, over 28913.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3896, pruned_loss=0.1391, over 5638522.85 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3774, pruned_loss=0.1252, over 5763414.59 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.392, pruned_loss=0.1407, over 5621913.73 frames. ], batch size: 136, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:39:48,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=343206.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:40:03,455 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343222.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:40:06,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343225.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:40:11,826 INFO [train.py:968] (1/2) Epoch 8, batch 24450, giga_loss[loss=0.3377, simple_loss=0.3952, pruned_loss=0.1401, over 28920.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3896, pruned_loss=0.139, over 5644926.90 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3772, pruned_loss=0.1251, over 5764472.11 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3919, pruned_loss=0.1405, over 5629275.50 frames. ], batch size: 227, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:40:25,767 INFO [optim.py:369] (1/2) 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 07:40:30,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3404, 1.7567, 1.5111, 1.5726], device='cuda:1'), covar=tensor([0.0631, 0.0256, 0.0257, 0.0654], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0078], device='cuda:1')
+2023-03-04 07:40:32,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3379, 1.6811, 1.3243, 1.4141], device='cuda:1'), covar=tensor([0.2132, 0.1849, 0.2034, 0.1786], device='cuda:1'), in_proj_covar=tensor([0.1209, 0.0901, 0.1058, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 07:40:40,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 07:40:50,393 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 07:40:53,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-04 07:41:04,763 INFO [train.py:968] (1/2) Epoch 8, batch 24500, giga_loss[loss=0.3414, simple_loss=0.3895, pruned_loss=0.1467, over 27568.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3888, pruned_loss=0.1378, over 5651369.80 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3774, pruned_loss=0.1253, over 5767582.18 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3907, pruned_loss=0.1392, over 5633600.50 frames. ], batch size: 472, lr: 4.04e-03, grad_scale: 8.0
+2023-03-04 07:41:31,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:41:34,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:41:37,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 07:41:54,365 INFO [train.py:968] (1/2) Epoch 8, batch 24550, giga_loss[loss=0.4039, simple_loss=0.4467, pruned_loss=0.1806, over 28126.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3856, pruned_loss=0.1338, over 5664597.61 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3776, pruned_loss=0.1256, over 5768457.69 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3873, pruned_loss=0.135, over 5646532.27 frames. ], batch size: 412, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:42:11,057 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 24600, giga_loss[loss=0.3109, simple_loss=0.3834, pruned_loss=0.1192, over 28691.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3876, pruned_loss=0.1327, over 5671050.78 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3776, pruned_loss=0.1257, over 5767130.04 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3891, pruned_loss=0.1336, over 5655652.11 frames. ], batch size: 71, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:43:11,733 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-04 07:43:35,139 INFO [train.py:968] (1/2) Epoch 8, batch 24650, giga_loss[loss=0.3647, simple_loss=0.4072, pruned_loss=0.1611, over 27584.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3884, pruned_loss=0.1334, over 5661370.72 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.377, pruned_loss=0.1254, over 5770380.70 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3906, pruned_loss=0.1347, over 5643202.90 frames. ], batch size: 472, lr: 4.03e-03, grad_scale: 1.0
+2023-03-04 07:43:49,712 INFO [optim.py:369] (1/2) 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 07:43:50,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:43:52,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:43:55,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:43:58,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:06,811 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=343461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:06,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:10,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3997, 1.5057, 1.2696, 1.1497], device='cuda:1'), covar=tensor([0.1333, 0.1190, 0.0926, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1468, 0.1432, 0.1533], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 07:44:23,365 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:23,718 INFO [train.py:968] (1/2) Epoch 8, batch 24700, giga_loss[loss=0.358, simple_loss=0.4166, pruned_loss=0.1497, over 28967.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3892, pruned_loss=0.1344, over 5673784.05 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3767, pruned_loss=0.1253, over 5768074.56 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3913, pruned_loss=0.1356, over 5659494.01 frames. ], batch size: 227, lr: 4.03e-03, grad_scale: 1.0
+2023-03-04 07:44:26,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:44:42,431 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-04 07:45:14,136 INFO [train.py:968] (1/2) Epoch 8, batch 24750, giga_loss[loss=0.3409, simple_loss=0.391, pruned_loss=0.1454, over 27588.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3886, pruned_loss=0.1342, over 5686583.13 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.377, pruned_loss=0.1255, over 5769589.71 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3901, pruned_loss=0.135, over 5673384.94 frames. ], batch size: 472, lr: 4.03e-03, grad_scale: 1.0
+2023-03-04 07:45:28,288 INFO [optim.py:369] (1/2) 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 07:45:59,966 INFO [train.py:968] (1/2) Epoch 8, batch 24800, giga_loss[loss=0.3918, simple_loss=0.4262, pruned_loss=0.1787, over 28028.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3873, pruned_loss=0.1348, over 5672540.56 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3771, pruned_loss=0.1258, over 5758945.02 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3886, pruned_loss=0.1354, over 5670392.13 frames. ], batch size: 412, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:46:01,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:46:21,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:46:24,347 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 24850, giga_loss[loss=0.3147, simple_loss=0.3871, pruned_loss=0.1211, over 28973.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3862, pruned_loss=0.1346, over 5664366.78 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3772, pruned_loss=0.126, over 5752298.29 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3873, pruned_loss=0.135, over 5667823.67 frames. ], batch size: 155, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:46:51,383 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:46:58,925 INFO [optim.py:369] (1/2) 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 07:47:24,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4981, 1.7793, 1.8197, 1.3404], device='cuda:1'), covar=tensor([0.1710, 0.2180, 0.1384, 0.1542], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0714, 0.0826, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 07:47:28,545 INFO [train.py:968] (1/2) Epoch 8, batch 24900, giga_loss[loss=0.3241, simple_loss=0.3936, pruned_loss=0.1273, over 28907.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3858, pruned_loss=0.1326, over 5667097.68 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3774, pruned_loss=0.1261, over 5744664.38 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3866, pruned_loss=0.1329, over 5675156.39 frames. ], batch size: 227, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:48:13,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:48:16,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343727.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:48:17,364 INFO [train.py:968] (1/2) Epoch 8, batch 24950, giga_loss[loss=0.2861, simple_loss=0.3668, pruned_loss=0.1027, over 28987.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.386, pruned_loss=0.1319, over 5674226.62 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3782, pruned_loss=0.1269, over 5748277.39 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3861, pruned_loss=0.1316, over 5675860.24 frames. ], batch size: 128, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:48:32,731 INFO [optim.py:369] (1/2) 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 07:48:41,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=343756.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 07:48:44,791 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5316, 1.6804, 1.3853, 1.7267], device='cuda:1'), covar=tensor([0.2152, 0.2135, 0.2374, 0.2076], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.0908, 0.1067, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 07:49:01,524 INFO [train.py:968] (1/2) Epoch 8, batch 25000, giga_loss[loss=0.3191, simple_loss=0.3818, pruned_loss=0.1282, over 28708.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.385, pruned_loss=0.1318, over 5664521.73 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.1271, over 5741596.88 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3856, pruned_loss=0.1316, over 5669492.85 frames. ], batch size: 242, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:49:52,334 INFO [train.py:968] (1/2) Epoch 8, batch 25050, giga_loss[loss=0.3485, simple_loss=0.394, pruned_loss=0.1515, over 27554.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3841, pruned_loss=0.1316, over 5671923.07 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3781, pruned_loss=0.1272, over 5744025.32 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3844, pruned_loss=0.1313, over 5672982.90 frames. ], batch size: 472, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:49:59,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=343836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:50:07,390 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 07:50:10,550 INFO [optim.py:369] (1/2) 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 07:50:41,020 INFO [train.py:968] (1/2) Epoch 8, batch 25100, giga_loss[loss=0.3955, simple_loss=0.4306, pruned_loss=0.1802, over 28737.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3832, pruned_loss=0.1318, over 5665937.79 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3781, pruned_loss=0.1271, over 5748557.73 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3837, pruned_loss=0.1318, over 5659973.33 frames. ], batch size: 284, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:51:27,341 INFO [train.py:968] (1/2) Epoch 8, batch 25150, giga_loss[loss=0.401, simple_loss=0.4351, pruned_loss=0.1834, over 26661.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3825, pruned_loss=0.1321, over 5666563.82 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3779, pruned_loss=0.1271, over 5749193.09 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3831, pruned_loss=0.1321, over 5660256.83 frames. ], batch size: 555, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 07:51:41,590 INFO [optim.py:369] (1/2) 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 07:52:10,864 INFO [train.py:968] (1/2) Epoch 8, batch 25200, giga_loss[loss=0.2692, simple_loss=0.3351, pruned_loss=0.1017, over 28760.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3824, pruned_loss=0.1324, over 5677033.30 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3777, pruned_loss=0.1268, over 5752849.38 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3832, pruned_loss=0.1329, over 5666650.40 frames. ], batch size: 119, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:52:12,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=343979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:52:14,915 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=343982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:52:22,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0812, 5.8368, 5.5003, 2.7439], device='cuda:1'), covar=tensor([0.0462, 0.0776, 0.0854, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.0980, 0.0933, 0.0819, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 07:52:38,789 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=344011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:52:43,292 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-04 07:52:53,899 INFO [train.py:968] (1/2) Epoch 8, batch 25250, giga_loss[loss=0.2981, simple_loss=0.3631, pruned_loss=0.1165, over 28798.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3808, pruned_loss=0.1314, over 5682659.32 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3785, pruned_loss=0.1275, over 5755340.20 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3809, pruned_loss=0.1314, over 5669177.36 frames. ], batch size: 284, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:52:56,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 07:53:10,792 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 25300, giga_loss[loss=0.3231, simple_loss=0.3892, pruned_loss=0.1284, over 28578.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3812, pruned_loss=0.1324, over 5664026.83 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3785, pruned_loss=0.1275, over 5746530.31 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3812, pruned_loss=0.1324, over 5660993.71 frames. ], batch size: 60, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:53:58,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-04 07:54:24,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5952, 4.6455, 1.7521, 1.7613], device='cuda:1'), covar=tensor([0.0927, 0.0279, 0.0836, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0504, 0.0327, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 07:54:37,364 INFO [train.py:968] (1/2) Epoch 8, batch 25350, giga_loss[loss=0.3027, simple_loss=0.3776, pruned_loss=0.1138, over 28772.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3814, pruned_loss=0.1319, over 5658987.47 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3783, pruned_loss=0.1274, over 5747281.65 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3816, pruned_loss=0.132, over 5655604.69 frames. ], batch size: 119, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:54:51,491 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 25400, giga_loss[loss=0.2806, simple_loss=0.362, pruned_loss=0.09957, over 28882.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3816, pruned_loss=0.1313, over 5668746.87 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.379, pruned_loss=0.1281, over 5750818.55 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3812, pruned_loss=0.1307, over 5661236.14 frames. ], batch size: 174, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:56:05,549 INFO [train.py:968] (1/2) Epoch 8, batch 25450, libri_loss[loss=0.3355, simple_loss=0.387, pruned_loss=0.142, over 29534.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3813, pruned_loss=0.1308, over 5657719.71 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3789, pruned_loss=0.1282, over 5741586.66 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3811, pruned_loss=0.1303, over 5657612.12 frames. ], batch size: 80, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:56:21,868 INFO [optim.py:369] (1/2) 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 07:56:36,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3763, 5.2154, 4.9441, 2.0463], device='cuda:1'), covar=tensor([0.0389, 0.0560, 0.0638, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.0977, 0.0930, 0.0817, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 07:56:49,774 INFO [train.py:968] (1/2) Epoch 8, batch 25500, giga_loss[loss=0.3, simple_loss=0.3649, pruned_loss=0.1175, over 28980.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.382, pruned_loss=0.1317, over 5659320.83 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3789, pruned_loss=0.1283, over 5736476.18 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3819, pruned_loss=0.1313, over 5661963.62 frames. ], batch size: 136, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:56:51,908 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=344281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 07:57:36,886 INFO [train.py:968] (1/2) Epoch 8, batch 25550, giga_loss[loss=0.4196, simple_loss=0.4487, pruned_loss=0.1953, over 27598.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.386, pruned_loss=0.1355, over 5641672.76 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3795, pruned_loss=0.1289, over 5731130.14 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3855, pruned_loss=0.1348, over 5646810.87 frames. ], batch size: 472, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:57:53,052 INFO [optim.py:369] (1/2) 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 07:58:29,244 INFO [train.py:968] (1/2) Epoch 8, batch 25600, giga_loss[loss=0.3285, simple_loss=0.3877, pruned_loss=0.1346, over 28880.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3865, pruned_loss=0.1369, over 5648917.14 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3796, pruned_loss=0.129, over 5732628.31 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3861, pruned_loss=0.1364, over 5650910.55 frames. ], batch size: 227, lr: 4.03e-03, grad_scale: 8.0
+2023-03-04 07:58:32,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5590, 1.6148, 1.5778, 1.4515], device='cuda:1'), covar=tensor([0.1140, 0.1621, 0.1617, 0.1581], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0731, 0.0659, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 07:59:18,842 INFO [train.py:968] (1/2) Epoch 8, batch 25650, giga_loss[loss=0.4245, simple_loss=0.4416, pruned_loss=0.2037, over 28794.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3881, pruned_loss=0.1393, over 5656321.33 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3797, pruned_loss=0.1291, over 5732805.07 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3878, pruned_loss=0.1389, over 5656147.44 frames. ], batch size: 119, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 07:59:39,038 INFO [optim.py:369] (1/2) 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 08:00:04,921 INFO [train.py:968] (1/2) Epoch 8, batch 25700, giga_loss[loss=0.3224, simple_loss=0.3839, pruned_loss=0.1304, over 28849.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3887, pruned_loss=0.1399, over 5651212.18 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3801, pruned_loss=0.1294, over 5735456.81 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3884, pruned_loss=0.1396, over 5646704.66 frames. ], batch size: 174, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:00:53,906 INFO [train.py:968] (1/2) Epoch 8, batch 25750, giga_loss[loss=0.2925, simple_loss=0.3628, pruned_loss=0.1111, over 28889.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3874, pruned_loss=0.1391, over 5646547.00 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3801, pruned_loss=0.1296, over 5726315.93 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3872, pruned_loss=0.1388, over 5651185.66 frames. ], batch size: 174, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:01:07,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3895, 3.0157, 1.4018, 1.4694], device='cuda:1'), covar=tensor([0.0877, 0.0317, 0.0847, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0502, 0.0325, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 08:01:10,724 INFO [optim.py:369] (1/2) 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 08:01:25,085 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 25800, libri_loss[loss=0.3064, simple_loss=0.3519, pruned_loss=0.1304, over 29635.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3866, pruned_loss=0.1372, over 5655380.73 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3803, pruned_loss=0.1297, over 5725854.72 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3863, pruned_loss=0.1369, over 5658210.36 frames. ], batch size: 69, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:02:25,930 INFO [train.py:968] (1/2) Epoch 8, batch 25850, giga_loss[loss=0.2897, simple_loss=0.3544, pruned_loss=0.1125, over 28529.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3829, pruned_loss=0.1337, over 5656075.83 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3803, pruned_loss=0.1298, over 5728887.79 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3829, pruned_loss=0.1336, over 5654222.99 frames. ], batch size: 336, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:02:27,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 08:02:41,941 INFO [optim.py:369] (1/2) 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 08:02:49,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=344656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:03:04,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5078, 1.7175, 1.3348, 1.8569], device='cuda:1'), covar=tensor([0.2293, 0.2223, 0.2369, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.1224, 0.0919, 0.1071, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 08:03:12,700 INFO [train.py:968] (1/2) Epoch 8, batch 25900, giga_loss[loss=0.3191, simple_loss=0.3808, pruned_loss=0.1286, over 28803.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3811, pruned_loss=0.1324, over 5662157.97 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3805, pruned_loss=0.1299, over 5732007.64 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3809, pruned_loss=0.1323, over 5656621.64 frames. ], batch size: 119, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:03:55,631 INFO [train.py:968] (1/2) Epoch 8, batch 25950, giga_loss[loss=0.2804, simple_loss=0.3507, pruned_loss=0.1051, over 28946.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3795, pruned_loss=0.1322, over 5658205.39 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3803, pruned_loss=0.13, over 5718284.25 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3795, pruned_loss=0.132, over 5665091.23 frames. ], batch size: 164, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 08:04:09,006 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:04:45,383 INFO [train.py:968] (1/2) Epoch 8, batch 26000, giga_loss[loss=0.3224, simple_loss=0.3843, pruned_loss=0.1302, over 28332.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3805, pruned_loss=0.1327, over 5664299.91 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3811, pruned_loss=0.1303, over 5719130.88 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3797, pruned_loss=0.1323, over 5666936.78 frames. ], batch size: 368, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:05:02,343 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=344799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:05:04,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=344802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:05:27,904 INFO [train.py:968] (1/2) Epoch 8, batch 26050, giga_loss[loss=0.3815, simple_loss=0.4326, pruned_loss=0.1652, over 28725.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3833, pruned_loss=0.1339, over 5664816.75 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3814, pruned_loss=0.1306, over 5712608.48 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3824, pruned_loss=0.1334, over 5671756.28 frames. ], batch size: 284, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:05:29,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=344831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:05:31,368 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 08:05:34,885 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-04 08:05:42,263 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 26100, giga_loss[loss=0.3277, simple_loss=0.4006, pruned_loss=0.1275, over 28878.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3851, pruned_loss=0.1317, over 5675565.09 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3809, pruned_loss=0.1302, over 5717243.31 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3849, pruned_loss=0.1317, over 5675928.68 frames. ], batch size: 112, lr: 4.03e-03, grad_scale: 4.0
+2023-03-04 08:06:44,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-04 08:06:57,839 INFO [train.py:968] (1/2) Epoch 8, batch 26150, giga_loss[loss=0.3318, simple_loss=0.3982, pruned_loss=0.1327, over 29041.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3859, pruned_loss=0.1312, over 5683464.43 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3806, pruned_loss=0.1302, over 5723887.05 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3861, pruned_loss=0.1313, over 5676438.06 frames. ], batch size: 155, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 08:07:09,940 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:07:25,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=344960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:07:32,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2922, 2.9564, 1.3489, 1.3326], device='cuda:1'), covar=tensor([0.0899, 0.0402, 0.0864, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0498, 0.0325, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 08:07:40,639 INFO [train.py:968] (1/2) Epoch 8, batch 26200, giga_loss[loss=0.3888, simple_loss=0.4323, pruned_loss=0.1726, over 28742.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3864, pruned_loss=0.1326, over 5689772.69 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3796, pruned_loss=0.13, over 5728204.87 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3879, pruned_loss=0.1329, over 5677864.22 frames. ], batch size: 284, lr: 4.03e-03, grad_scale: 2.0
+2023-03-04 08:07:42,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3342, 1.5697, 1.2852, 1.4948], device='cuda:1'), covar=tensor([0.0767, 0.0306, 0.0330, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0118, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0071, 0.0051, 0.0046, 0.0078], device='cuda:1')
+2023-03-04 08:08:26,825 INFO [train.py:968] (1/2) Epoch 8, batch 26250, giga_loss[loss=0.3453, simple_loss=0.405, pruned_loss=0.1427, over 29064.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3881, pruned_loss=0.1344, over 5682893.16 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3796, pruned_loss=0.1303, over 5720492.65 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3895, pruned_loss=0.1345, over 5679289.96 frames. ], batch size: 155, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:08:44,843 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 26300, giga_loss[loss=0.3043, simple_loss=0.3662, pruned_loss=0.1212, over 28910.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3894, pruned_loss=0.1366, over 5673460.02 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3795, pruned_loss=0.1303, over 5721326.18 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3906, pruned_loss=0.1367, over 5669476.58 frames. ], batch size: 106, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:09:20,065 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:22,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:50,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 08:09:51,749 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:56,116 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:09:56,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:10:03,129 INFO [train.py:968] (1/2) Epoch 8, batch 26350, giga_loss[loss=0.3049, simple_loss=0.3743, pruned_loss=0.1178, over 28573.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3879, pruned_loss=0.1359, over 5676919.02 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3795, pruned_loss=0.1304, over 5714486.03 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3891, pruned_loss=0.1361, over 5679666.82 frames. ], batch size: 336, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:10:19,227 INFO [optim.py:369] (1/2) 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 08:10:39,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3756, 1.5688, 1.4903, 1.3949], device='cuda:1'), covar=tensor([0.1058, 0.1027, 0.1562, 0.1198], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0730, 0.0661, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 08:10:45,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3907, 1.6893, 1.2509, 1.9166], device='cuda:1'), covar=tensor([0.2254, 0.2113, 0.2372, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1222, 0.0917, 0.1074, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 08:10:45,355 INFO [train.py:968] (1/2) Epoch 8, batch 26400, giga_loss[loss=0.2872, simple_loss=0.3541, pruned_loss=0.1102, over 28910.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3857, pruned_loss=0.135, over 5677449.31 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3793, pruned_loss=0.1302, over 5715292.50 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3871, pruned_loss=0.1354, over 5678055.45 frames. ], batch size: 227, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:10:46,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7987, 5.6610, 5.2928, 2.4479], device='cuda:1'), covar=tensor([0.0340, 0.0483, 0.0601, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0979, 0.0930, 0.0818, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 08:11:26,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5088, 2.4284, 1.8110, 2.1424], device='cuda:1'), covar=tensor([0.0628, 0.0581, 0.0870, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0448, 0.0493, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 08:11:33,255 INFO [train.py:968] (1/2) Epoch 8, batch 26450, giga_loss[loss=0.3036, simple_loss=0.3584, pruned_loss=0.1244, over 28991.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3845, pruned_loss=0.1344, over 5692472.16 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3799, pruned_loss=0.1305, over 5720400.76 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3853, pruned_loss=0.1345, over 5687087.72 frames. ], batch size: 106, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:11:36,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7437, 1.9366, 1.9836, 1.5912], device='cuda:1'), covar=tensor([0.1568, 0.1897, 0.1160, 0.1351], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0717, 0.0828, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 08:11:48,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4586, 1.7148, 1.8226, 1.4398], device='cuda:1'), covar=tensor([0.1272, 0.1604, 0.0960, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0717, 0.0827, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 08:11:51,109 INFO [optim.py:369] (1/2) 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 08:12:02,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:12:04,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:12:17,817 INFO [train.py:968] (1/2) Epoch 8, batch 26500, giga_loss[loss=0.3582, simple_loss=0.4169, pruned_loss=0.1497, over 28935.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3855, pruned_loss=0.1355, over 5684555.60 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3799, pruned_loss=0.1305, over 5723519.30 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3863, pruned_loss=0.1358, over 5676475.56 frames. ], batch size: 227, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:12:30,006 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:12:49,238 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:13:00,725 INFO [train.py:968] (1/2) Epoch 8, batch 26550, giga_loss[loss=0.3132, simple_loss=0.3728, pruned_loss=0.1268, over 29029.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3847, pruned_loss=0.135, over 5677188.29 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3796, pruned_loss=0.1304, over 5713632.37 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3857, pruned_loss=0.1354, over 5678040.57 frames. ], batch size: 128, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:13:07,917 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:13:20,213 INFO [optim.py:369] (1/2) 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 08:13:36,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3640, 1.6178, 1.3177, 1.4592], device='cuda:1'), covar=tensor([0.2027, 0.1949, 0.2056, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.1223, 0.0921, 0.1075, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 08:13:45,062 INFO [train.py:968] (1/2) Epoch 8, batch 26600, giga_loss[loss=0.3218, simple_loss=0.3739, pruned_loss=0.1348, over 28237.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3837, pruned_loss=0.1354, over 5671102.40 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3795, pruned_loss=0.1305, over 5718416.05 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3847, pruned_loss=0.1358, over 5666496.58 frames. ], batch size: 60, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:14:32,230 INFO [train.py:968] (1/2) Epoch 8, batch 26650, giga_loss[loss=0.3328, simple_loss=0.3933, pruned_loss=0.1361, over 28838.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.383, pruned_loss=0.1352, over 5662767.63 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3793, pruned_loss=0.1303, over 5723357.74 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.384, pruned_loss=0.1358, over 5653255.99 frames. ], batch size: 186, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:14:49,087 INFO [optim.py:369] (1/2) 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 08:14:49,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5152, 1.7148, 1.7821, 1.3669], device='cuda:1'), covar=tensor([0.1475, 0.1970, 0.1143, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0717, 0.0826, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 08:15:17,002 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:15:17,454 INFO [train.py:968] (1/2) Epoch 8, batch 26700, giga_loss[loss=0.2682, simple_loss=0.3425, pruned_loss=0.09695, over 28510.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3846, pruned_loss=0.1351, over 5662927.65 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.38, pruned_loss=0.1308, over 5718258.95 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.385, pruned_loss=0.1353, over 5659411.96 frames. ], batch size: 60, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:15:20,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:15:30,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:15:46,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=345510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:16:05,303 INFO [train.py:968] (1/2) Epoch 8, batch 26750, giga_loss[loss=0.3177, simple_loss=0.3828, pruned_loss=0.1263, over 28768.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3884, pruned_loss=0.1377, over 5654977.19 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.38, pruned_loss=0.1308, over 5715556.66 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3886, pruned_loss=0.1378, over 5654322.84 frames. ], batch size: 262, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:16:26,894 INFO [optim.py:369] (1/2) 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 08:16:55,432 INFO [train.py:968] (1/2) Epoch 8, batch 26800, giga_loss[loss=0.2797, simple_loss=0.3505, pruned_loss=0.1044, over 28607.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3872, pruned_loss=0.1372, over 5661058.39 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1308, over 5716495.87 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3874, pruned_loss=0.1373, over 5659451.24 frames. ], batch size: 60, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:17:34,828 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 26850, giga_loss[loss=0.2934, simple_loss=0.3702, pruned_loss=0.1082, over 28737.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3874, pruned_loss=0.1341, over 5660779.13 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.1309, over 5709423.81 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3874, pruned_loss=0.1342, over 5664728.91 frames. ], batch size: 119, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:17:49,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:17:51,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:18:00,986 INFO [optim.py:369] (1/2) 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 08:18:04,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 08:18:17,916 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 26900, giga_loss[loss=0.3094, simple_loss=0.3863, pruned_loss=0.1163, over 28992.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3884, pruned_loss=0.1328, over 5654748.56 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3798, pruned_loss=0.1308, over 5690620.59 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.389, pruned_loss=0.1331, over 5673553.63 frames. ], batch size: 136, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:18:37,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 26950, giga_loss[loss=0.2809, simple_loss=0.3569, pruned_loss=0.1025, over 29032.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3915, pruned_loss=0.1343, over 5674754.65 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3804, pruned_loss=0.1311, over 5696159.20 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3918, pruned_loss=0.1343, over 5684163.50 frames. ], batch size: 164, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:19:28,565 INFO [optim.py:369] (1/2) 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 08:19:55,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 08:19:58,319 INFO [train.py:968] (1/2) Epoch 8, batch 27000, giga_loss[loss=0.3246, simple_loss=0.3861, pruned_loss=0.1315, over 28663.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3947, pruned_loss=0.1381, over 5663525.80 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3804, pruned_loss=0.1311, over 5686919.60 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3951, pruned_loss=0.1382, over 5678542.54 frames. ], batch size: 242, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:19:58,320 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 08:20:07,677 INFO [train.py:1012] (1/2) Epoch 8, validation: loss=0.2229, simple_loss=0.3279, pruned_loss=0.05895, over 944034.00 frames. 
+2023-03-04 08:20:07,678 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 08:20:49,424 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345824.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:20:52,537 INFO [train.py:968] (1/2) Epoch 8, batch 27050, giga_loss[loss=0.325, simple_loss=0.3898, pruned_loss=0.1301, over 28795.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3976, pruned_loss=0.1422, over 5651406.05 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3805, pruned_loss=0.1313, over 5692331.15 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3981, pruned_loss=0.1423, over 5657351.04 frames. ], batch size: 199, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:20:58,033 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=345832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:21:00,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=345835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:21:14,225 INFO [optim.py:369] (1/2) 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 08:21:28,018 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 27100, libri_loss[loss=0.3415, simple_loss=0.3934, pruned_loss=0.1448, over 29469.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3958, pruned_loss=0.1415, over 5658089.48 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3805, pruned_loss=0.1314, over 5698757.85 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3967, pruned_loss=0.1418, over 5655780.83 frames. ], batch size: 85, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:22:28,043 INFO [train.py:968] (1/2) Epoch 8, batch 27150, giga_loss[loss=0.3005, simple_loss=0.3808, pruned_loss=0.1101, over 29040.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3936, pruned_loss=0.1397, over 5652994.50 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3799, pruned_loss=0.131, over 5702650.11 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3953, pruned_loss=0.1405, over 5646255.76 frames. ], batch size: 155, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:22:36,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:22:42,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2924, 1.3303, 1.1751, 1.4909], device='cuda:1'), covar=tensor([0.0658, 0.0428, 0.0324, 0.0670], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0051, 0.0046, 0.0078], device='cuda:1')
+2023-03-04 08:22:45,008 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 08:22:46,490 INFO [optim.py:369] (1/2) 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 08:23:01,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6742, 2.3604, 1.5177, 0.8724], device='cuda:1'), covar=tensor([0.4371, 0.2500, 0.2691, 0.4130], device='cuda:1'), in_proj_covar=tensor([0.1481, 0.1403, 0.1435, 0.1201], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 08:23:11,656 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=345977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:23:12,855 INFO [train.py:968] (1/2) Epoch 8, batch 27200, giga_loss[loss=0.309, simple_loss=0.3829, pruned_loss=0.1176, over 29050.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3937, pruned_loss=0.1383, over 5660032.16 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3804, pruned_loss=0.1316, over 5704676.74 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.395, pruned_loss=0.1387, over 5651213.50 frames. ], batch size: 128, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:23:31,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=345997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:23:59,225 INFO [train.py:968] (1/2) Epoch 8, batch 27250, giga_loss[loss=0.3723, simple_loss=0.4186, pruned_loss=0.163, over 27957.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3927, pruned_loss=0.1357, over 5677002.12 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.38, pruned_loss=0.1313, over 5709059.50 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3944, pruned_loss=0.1363, over 5665360.46 frames. ], batch size: 412, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:24:19,740 INFO [optim.py:369] (1/2) 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 08:24:49,758 INFO [train.py:968] (1/2) Epoch 8, batch 27300, giga_loss[loss=0.3309, simple_loss=0.385, pruned_loss=0.1384, over 27949.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.394, pruned_loss=0.1372, over 5662423.64 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3795, pruned_loss=0.1311, over 5702431.67 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3959, pruned_loss=0.138, over 5657988.24 frames. ], batch size: 412, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:24:53,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4513, 4.0302, 1.5030, 1.5887], device='cuda:1'), covar=tensor([0.0900, 0.0282, 0.0844, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0501, 0.0326, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 08:25:36,058 INFO [train.py:968] (1/2) Epoch 8, batch 27350, giga_loss[loss=0.3448, simple_loss=0.4013, pruned_loss=0.1442, over 29056.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3929, pruned_loss=0.1367, over 5666533.87 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3799, pruned_loss=0.1313, over 5697794.99 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3944, pruned_loss=0.1372, over 5665754.54 frames. ], batch size: 145, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:25:45,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:25:48,450 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:25:59,498 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-04 08:26:17,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=346172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:26:24,068 INFO [train.py:968] (1/2) Epoch 8, batch 27400, giga_loss[loss=0.3985, simple_loss=0.4318, pruned_loss=0.1826, over 27930.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3908, pruned_loss=0.1368, over 5656373.78 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3797, pruned_loss=0.1312, over 5697774.75 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3924, pruned_loss=0.1374, over 5655096.93 frames. ], batch size: 412, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:26:44,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346199.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:26:56,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4541, 1.6853, 1.8913, 1.4382], device='cuda:1'), covar=tensor([0.1203, 0.1706, 0.0957, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0716, 0.0827, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 08:27:02,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6204, 4.3957, 1.6904, 1.7398], device='cuda:1'), covar=tensor([0.0868, 0.0220, 0.0779, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0502, 0.0326, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 08:27:10,873 INFO [train.py:968] (1/2) Epoch 8, batch 27450, libri_loss[loss=0.289, simple_loss=0.3539, pruned_loss=0.112, over 29548.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3892, pruned_loss=0.1371, over 5640184.19 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3798, pruned_loss=0.1314, over 5694657.56 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3907, pruned_loss=0.1377, over 5640296.83 frames. ], batch size: 75, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:27:34,699 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 27500, giga_loss[loss=0.3029, simple_loss=0.3699, pruned_loss=0.1179, over 28712.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3873, pruned_loss=0.136, over 5648417.20 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3802, pruned_loss=0.1316, over 5695807.47 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3884, pruned_loss=0.1364, over 5646376.16 frames. ], batch size: 262, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:28:29,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346311.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:28:46,012 INFO [train.py:968] (1/2) Epoch 8, batch 27550, giga_loss[loss=0.3708, simple_loss=0.4062, pruned_loss=0.1677, over 28697.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3867, pruned_loss=0.1369, over 5646340.00 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3801, pruned_loss=0.1315, over 5698908.53 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3877, pruned_loss=0.1373, over 5641353.25 frames. ], batch size: 284, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:28:49,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4779, 2.1019, 1.5351, 0.8127], device='cuda:1'), covar=tensor([0.3059, 0.1597, 0.2400, 0.3445], device='cuda:1'), in_proj_covar=tensor([0.1484, 0.1416, 0.1444, 0.1213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 08:28:56,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3945, 2.1655, 2.0160, 2.0453], device='cuda:1'), covar=tensor([0.1202, 0.2132, 0.1666, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0733, 0.0657, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 08:28:58,647 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346342.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:29:01,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=346345.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:29:05,954 INFO [optim.py:369] (1/2) 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 08:29:07,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:29:27,249 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 27600, giga_loss[loss=0.3001, simple_loss=0.3696, pruned_loss=0.1152, over 29040.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3858, pruned_loss=0.1364, over 5644107.61 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3801, pruned_loss=0.1315, over 5693548.13 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3867, pruned_loss=0.1368, over 5643792.46 frames. ], batch size: 164, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:29:49,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-04 08:30:15,788 INFO [train.py:968] (1/2) Epoch 8, batch 27650, giga_loss[loss=0.2814, simple_loss=0.3568, pruned_loss=0.1029, over 28625.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3822, pruned_loss=0.132, over 5652307.45 frames. ], libri_tot_loss[loss=0.3215, simple_loss=0.3801, pruned_loss=0.1315, over 5694456.06 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3829, pruned_loss=0.1324, over 5650861.24 frames. ], batch size: 78, lr: 4.02e-03, grad_scale: 8.0
+2023-03-04 08:30:36,283 INFO [optim.py:369] (1/2) 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 08:30:39,620 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346454.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:30:41,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=346457.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:31:01,962 INFO [train.py:968] (1/2) Epoch 8, batch 27700, giga_loss[loss=0.2923, simple_loss=0.3664, pruned_loss=0.109, over 28536.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3788, pruned_loss=0.1283, over 5659480.73 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3803, pruned_loss=0.1316, over 5694752.60 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3792, pruned_loss=0.1285, over 5657418.93 frames. ], batch size: 60, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:31:06,774 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=346486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:31:06,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4350, 2.2519, 1.6495, 0.5922], device='cuda:1'), covar=tensor([0.2742, 0.1699, 0.2421, 0.3207], device='cuda:1'), in_proj_covar=tensor([0.1484, 0.1425, 0.1451, 0.1219], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 08:31:15,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5270, 3.6718, 1.6022, 1.6170], device='cuda:1'), covar=tensor([0.0920, 0.0294, 0.0864, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0500, 0.0323, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 08:31:15,193 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=346495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:31:19,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=346498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:31:47,832 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=346527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:31:47,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5757, 2.1690, 1.5155, 1.2234], device='cuda:1'), covar=tensor([0.1992, 0.1356, 0.1689, 0.1901], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1469, 0.1452, 0.1548], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 08:31:49,021 INFO [train.py:968] (1/2) Epoch 8, batch 27750, giga_loss[loss=0.2942, simple_loss=0.3638, pruned_loss=0.1123, over 28961.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3792, pruned_loss=0.1288, over 5647569.74 frames. ], libri_tot_loss[loss=0.3219, simple_loss=0.3804, pruned_loss=0.1317, over 5688697.94 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3794, pruned_loss=0.1288, over 5649675.09 frames. ], batch size: 164, lr: 4.02e-03, grad_scale: 4.0
+2023-03-04 08:31:55,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8181, 2.8253, 1.8926, 0.8859], device='cuda:1'), covar=tensor([0.4056, 0.1922, 0.2309, 0.4102], device='cuda:1'), in_proj_covar=tensor([0.1477, 0.1420, 0.1447, 0.1214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 08:32:06,170 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:32:09,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5107, 1.9173, 1.6028, 1.4043], device='cuda:1'), covar=tensor([0.1375, 0.0948, 0.0850, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.1626, 0.1467, 0.1448, 0.1545], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 08:32:09,504 INFO [optim.py:369] (1/2) 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 08:32:18,293 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 08:32:34,860 INFO [train.py:968] (1/2) Epoch 8, batch 27800, giga_loss[loss=0.2973, simple_loss=0.3504, pruned_loss=0.1221, over 28638.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3757, pruned_loss=0.1271, over 5658627.20 frames. ], libri_tot_loss[loss=0.322, simple_loss=0.3805, pruned_loss=0.1318, over 5694896.87 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3757, pruned_loss=0.1269, over 5652986.20 frames. ], batch size: 85, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:33:03,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2323, 0.8228, 0.8497, 1.3502], device='cuda:1'), covar=tensor([0.0750, 0.0354, 0.0340, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0052, 0.0046, 0.0078], device='cuda:1')
+2023-03-04 08:33:22,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:33:23,832 INFO [train.py:968] (1/2) Epoch 8, batch 27850, giga_loss[loss=0.3217, simple_loss=0.3927, pruned_loss=0.1254, over 28654.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3742, pruned_loss=0.127, over 5659053.65 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3803, pruned_loss=0.1317, over 5699571.21 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3741, pruned_loss=0.1267, over 5649172.21 frames. ], batch size: 242, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:33:46,199 INFO [optim.py:369] (1/2) 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 08:34:09,126 INFO [train.py:968] (1/2) Epoch 8, batch 27900, giga_loss[loss=0.3696, simple_loss=0.4132, pruned_loss=0.163, over 28289.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1281, over 5676282.28 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3803, pruned_loss=0.1315, over 5702065.48 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3772, pruned_loss=0.128, over 5665632.63 frames. ], batch size: 368, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:34:56,719 INFO [train.py:968] (1/2) Epoch 8, batch 27950, giga_loss[loss=0.2918, simple_loss=0.367, pruned_loss=0.1083, over 28943.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3791, pruned_loss=0.1292, over 5660407.93 frames. ], libri_tot_loss[loss=0.3221, simple_loss=0.3806, pruned_loss=0.1318, over 5697402.85 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3786, pruned_loss=0.1288, over 5655007.75 frames. ], batch size: 145, lr: 4.02e-03, grad_scale: 2.0
+2023-03-04 08:35:19,273 INFO [optim.py:369] (1/2) 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 08:35:43,188 INFO [train.py:968] (1/2) Epoch 8, batch 28000, giga_loss[loss=0.3177, simple_loss=0.3749, pruned_loss=0.1302, over 28704.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3808, pruned_loss=0.1306, over 5655328.81 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3813, pruned_loss=0.1322, over 5701418.10 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3799, pruned_loss=0.1299, over 5646687.20 frames. ], batch size: 92, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:36:29,438 INFO [train.py:968] (1/2) Epoch 8, batch 28050, giga_loss[loss=0.3073, simple_loss=0.375, pruned_loss=0.1198, over 28935.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3815, pruned_loss=0.1317, over 5644761.94 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3823, pruned_loss=0.133, over 5694745.29 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3799, pruned_loss=0.1303, over 5643040.30 frames. ], batch size: 155, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:36:43,898 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:36:48,485 INFO [optim.py:369] (1/2) 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 08:37:14,260 INFO [train.py:968] (1/2) Epoch 8, batch 28100, giga_loss[loss=0.3375, simple_loss=0.4013, pruned_loss=0.1368, over 28607.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3826, pruned_loss=0.1323, over 5657221.12 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3823, pruned_loss=0.1331, over 5687753.65 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3813, pruned_loss=0.1312, over 5661344.17 frames. ], batch size: 307, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:37:14,994 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-04 08:37:20,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=346886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:37:31,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3039, 1.4518, 1.4975, 1.3150], device='cuda:1'), covar=tensor([0.1225, 0.1382, 0.1620, 0.1498], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0737, 0.0658, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 08:37:53,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=346922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:37:55,323 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 28150, giga_loss[loss=0.3456, simple_loss=0.403, pruned_loss=0.1442, over 28867.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3844, pruned_loss=0.1335, over 5648996.66 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3821, pruned_loss=0.1331, over 5684132.33 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3835, pruned_loss=0.1325, over 5653963.05 frames. ], batch size: 145, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:38:11,892 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.30 vs. limit=2.0
+2023-03-04 08:38:18,200 INFO [optim.py:369] (1/2) 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 08:38:28,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7244, 1.9990, 2.0730, 1.6136], device='cuda:1'), covar=tensor([0.1633, 0.1909, 0.1174, 0.1360], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0718, 0.0831, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 08:38:42,892 INFO [train.py:968] (1/2) Epoch 8, batch 28200, giga_loss[loss=0.3334, simple_loss=0.3989, pruned_loss=0.1339, over 28894.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3864, pruned_loss=0.1348, over 5650710.69 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3822, pruned_loss=0.1333, over 5676554.89 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3856, pruned_loss=0.1339, over 5660377.00 frames. ], batch size: 227, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:39:06,960 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 28250, libri_loss[loss=0.3358, simple_loss=0.3967, pruned_loss=0.1375, over 29637.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3876, pruned_loss=0.1366, over 5632955.76 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3824, pruned_loss=0.1333, over 5671132.65 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3869, pruned_loss=0.1359, over 5644372.85 frames. ], batch size: 91, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:39:53,996 INFO [optim.py:369] (1/2) 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 08:40:06,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:40:10,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:40:20,941 INFO [train.py:968] (1/2) Epoch 8, batch 28300, giga_loss[loss=0.3166, simple_loss=0.3822, pruned_loss=0.1256, over 28024.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3888, pruned_loss=0.1368, over 5642774.70 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3824, pruned_loss=0.1332, over 5676906.24 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3884, pruned_loss=0.1364, over 5646216.76 frames. ], batch size: 412, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:40:42,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:40:43,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0053, 1.3493, 1.0838, 0.2124], device='cuda:1'), covar=tensor([0.1773, 0.1616, 0.2360, 0.2765], device='cuda:1'), in_proj_covar=tensor([0.1486, 0.1418, 0.1447, 0.1220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 08:40:55,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 08:41:13,657 INFO [train.py:968] (1/2) Epoch 8, batch 28350, giga_loss[loss=0.3424, simple_loss=0.3989, pruned_loss=0.143, over 28564.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3883, pruned_loss=0.1349, over 5650080.23 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3823, pruned_loss=0.1331, over 5677845.15 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3882, pruned_loss=0.1348, over 5651116.35 frames. ], batch size: 71, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:41:28,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:41:30,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:41:36,265 INFO [optim.py:369] (1/2) 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 08:42:01,086 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 28400, giga_loss[loss=0.3436, simple_loss=0.3998, pruned_loss=0.1437, over 28909.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3877, pruned_loss=0.1349, over 5661914.85 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3821, pruned_loss=0.133, over 5680676.19 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3878, pruned_loss=0.1349, over 5659864.93 frames. ], batch size: 227, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:42:44,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:42:53,927 INFO [train.py:968] (1/2) Epoch 8, batch 28450, giga_loss[loss=0.3187, simple_loss=0.3785, pruned_loss=0.1295, over 28760.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3865, pruned_loss=0.1349, over 5664325.17 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3825, pruned_loss=0.1332, over 5680825.28 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3864, pruned_loss=0.1347, over 5662294.08 frames. ], batch size: 284, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:43:24,040 INFO [optim.py:369] (1/2) 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 08:43:31,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:43:51,936 INFO [train.py:968] (1/2) Epoch 8, batch 28500, giga_loss[loss=0.2858, simple_loss=0.3568, pruned_loss=0.1074, over 28691.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3844, pruned_loss=0.1338, over 5671235.88 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3825, pruned_loss=0.1333, over 5683161.25 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3843, pruned_loss=0.1336, over 5667511.41 frames. ], batch size: 262, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:44:12,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:44:28,436 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=347317.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:44:38,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2786, 1.2783, 1.1317, 1.4993], device='cuda:1'), covar=tensor([0.0679, 0.0385, 0.0312, 0.0691], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0052, 0.0046, 0.0078], device='cuda:1')
+2023-03-04 08:44:39,066 INFO [train.py:968] (1/2) Epoch 8, batch 28550, giga_loss[loss=0.3784, simple_loss=0.4156, pruned_loss=0.1706, over 27600.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3848, pruned_loss=0.1345, over 5670600.07 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3829, pruned_loss=0.1335, over 5681746.41 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3844, pruned_loss=0.1342, over 5668662.49 frames. ], batch size: 472, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:44:43,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1308, 2.6036, 1.1784, 1.2762], device='cuda:1'), covar=tensor([0.0912, 0.0383, 0.0861, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0502, 0.0326, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 08:45:02,900 INFO [optim.py:369] (1/2) 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 08:45:10,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:14,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:25,929 INFO [train.py:968] (1/2) Epoch 8, batch 28600, giga_loss[loss=0.3314, simple_loss=0.3656, pruned_loss=0.1486, over 23401.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3847, pruned_loss=0.1351, over 5660778.14 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3832, pruned_loss=0.1337, over 5687564.08 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3842, pruned_loss=0.1347, over 5654067.54 frames. ], batch size: 705, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:45:39,361 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:48,495 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:45:50,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:08,942 INFO [train.py:968] (1/2) Epoch 8, batch 28650, libri_loss[loss=0.3464, simple_loss=0.4073, pruned_loss=0.1427, over 29503.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3844, pruned_loss=0.135, over 5668923.24 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3834, pruned_loss=0.1338, over 5693138.70 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3839, pruned_loss=0.1346, over 5657901.95 frames. ], batch size: 85, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:46:15,176 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:19,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:21,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:22,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 08:46:30,427 INFO [optim.py:369] (1/2) 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 08:46:48,700 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:46:51,753 INFO [train.py:968] (1/2) Epoch 8, batch 28700, libri_loss[loss=0.3769, simple_loss=0.4209, pruned_loss=0.1665, over 29513.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3855, pruned_loss=0.1361, over 5653571.11 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.384, pruned_loss=0.1343, over 5682337.98 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1353, over 5652424.72 frames. ], batch size: 84, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:46:57,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-04 08:47:35,774 INFO [train.py:968] (1/2) Epoch 8, batch 28750, giga_loss[loss=0.3211, simple_loss=0.373, pruned_loss=0.1346, over 28561.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3866, pruned_loss=0.1372, over 5656565.54 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3833, pruned_loss=0.1337, over 5687261.45 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3865, pruned_loss=0.1372, over 5650362.14 frames. ], batch size: 92, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:47:41,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5505, 4.3817, 4.1660, 1.7922], device='cuda:1'), covar=tensor([0.0474, 0.0634, 0.0680, 0.2101], device='cuda:1'), in_proj_covar=tensor([0.0994, 0.0946, 0.0837, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 08:47:43,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=347536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:48:01,413 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 28800, giga_loss[loss=0.3544, simple_loss=0.404, pruned_loss=0.1524, over 28891.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3883, pruned_loss=0.1392, over 5643505.59 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3837, pruned_loss=0.134, over 5684232.88 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.388, pruned_loss=0.1391, over 5640115.66 frames. ], batch size: 186, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:49:03,183 INFO [train.py:968] (1/2) Epoch 8, batch 28850, giga_loss[loss=0.3406, simple_loss=0.397, pruned_loss=0.1421, over 28248.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3898, pruned_loss=0.1412, over 5638932.28 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.384, pruned_loss=0.1344, over 5679389.56 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3893, pruned_loss=0.1409, over 5639510.17 frames. ], batch size: 77, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:49:26,805 INFO [optim.py:369] (1/2) 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 08:49:46,515 INFO [train.py:968] (1/2) Epoch 8, batch 28900, giga_loss[loss=0.3075, simple_loss=0.3687, pruned_loss=0.1231, over 28753.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3894, pruned_loss=0.1408, over 5641624.58 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.384, pruned_loss=0.1344, over 5673081.92 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3891, pruned_loss=0.1407, over 5647632.03 frames. ], batch size: 119, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:50:00,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347692.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:50:36,612 INFO [train.py:968] (1/2) Epoch 8, batch 28950, giga_loss[loss=0.3108, simple_loss=0.3787, pruned_loss=0.1215, over 28724.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3883, pruned_loss=0.1392, over 5638567.86 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3839, pruned_loss=0.1343, over 5677641.96 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3883, pruned_loss=0.1393, over 5638464.69 frames. ], batch size: 243, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:50:58,156 INFO [optim.py:369] (1/2) 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 08:51:19,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2432, 1.2977, 1.1914, 1.4366], device='cuda:1'), covar=tensor([0.0702, 0.0368, 0.0313, 0.0737], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0078], device='cuda:1')
+2023-03-04 08:51:19,406 INFO [train.py:968] (1/2) Epoch 8, batch 29000, giga_loss[loss=0.3527, simple_loss=0.4019, pruned_loss=0.1518, over 28268.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3879, pruned_loss=0.138, over 5650959.29 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3832, pruned_loss=0.1338, over 5681315.12 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3887, pruned_loss=0.1386, over 5646687.22 frames. ], batch size: 368, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:51:27,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 08:52:06,202 INFO [train.py:968] (1/2) Epoch 8, batch 29050, giga_loss[loss=0.3161, simple_loss=0.3769, pruned_loss=0.1277, over 28981.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.39, pruned_loss=0.1396, over 5656944.94 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3831, pruned_loss=0.1338, over 5682784.52 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3907, pruned_loss=0.1402, over 5651851.42 frames. ], batch size: 106, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:52:12,947 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=347835.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 08:52:15,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=347838.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:52:28,497 INFO [optim.py:369] (1/2) 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 08:52:33,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2217, 1.4800, 1.1803, 1.3480], device='cuda:1'), covar=tensor([0.2121, 0.2006, 0.2172, 0.1730], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.0924, 0.1076, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 08:52:39,546 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=347867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 08:52:49,136 INFO [train.py:968] (1/2) Epoch 8, batch 29100, giga_loss[loss=0.3522, simple_loss=0.405, pruned_loss=0.1497, over 28680.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3905, pruned_loss=0.1402, over 5674092.57 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3832, pruned_loss=0.1339, over 5688801.91 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3914, pruned_loss=0.1409, over 5664065.84 frames. ], batch size: 284, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:52:50,368 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-04 08:53:01,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6133, 3.3942, 1.5532, 1.5786], device='cuda:1'), covar=tensor([0.0870, 0.0427, 0.0841, 0.1267], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0502, 0.0325, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 08:53:18,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=347911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:53:33,793 INFO [train.py:968] (1/2) Epoch 8, batch 29150, giga_loss[loss=0.3069, simple_loss=0.3734, pruned_loss=0.1202, over 28860.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3901, pruned_loss=0.1398, over 5671761.10 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3831, pruned_loss=0.1338, over 5686184.31 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.391, pruned_loss=0.1405, over 5665976.98 frames. ], batch size: 213, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:53:34,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0744, 1.1749, 3.6811, 3.1369], device='cuda:1'), covar=tensor([0.1683, 0.2536, 0.0439, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0572, 0.0823, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 08:54:02,033 INFO [optim.py:369] (1/2) 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 08:54:17,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1364, 1.4683, 1.2040, 0.9381], device='cuda:1'), covar=tensor([0.1573, 0.1271, 0.1021, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.1639, 0.1480, 0.1471, 0.1563], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 08:54:27,305 INFO [train.py:968] (1/2) Epoch 8, batch 29200, giga_loss[loss=0.3436, simple_loss=0.4054, pruned_loss=0.1409, over 28794.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3918, pruned_loss=0.1402, over 5667871.10 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3828, pruned_loss=0.1335, over 5688443.81 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3929, pruned_loss=0.1411, over 5660992.28 frames. ], batch size: 199, lr: 4.01e-03, grad_scale: 8.0
+2023-03-04 08:55:16,323 INFO [train.py:968] (1/2) Epoch 8, batch 29250, giga_loss[loss=0.3114, simple_loss=0.3735, pruned_loss=0.1246, over 28721.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3908, pruned_loss=0.1383, over 5655239.71 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3834, pruned_loss=0.1341, over 5682679.23 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3913, pruned_loss=0.1386, over 5654271.63 frames. ], batch size: 99, lr: 4.01e-03, grad_scale: 8.0
+2023-03-04 08:55:20,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4636, 1.7862, 1.8145, 1.3654], device='cuda:1'), covar=tensor([0.1613, 0.2098, 0.1220, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0721, 0.0833, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 08:55:39,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=348054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:55:40,173 INFO [optim.py:369] (1/2) 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 08:55:41,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=348057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:55:59,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7262, 1.6195, 1.3256, 1.3163], device='cuda:1'), covar=tensor([0.0656, 0.0521, 0.0894, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0449, 0.0500, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 08:56:03,184 INFO [train.py:968] (1/2) Epoch 8, batch 29300, giga_loss[loss=0.3041, simple_loss=0.3715, pruned_loss=0.1183, over 28765.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3874, pruned_loss=0.1355, over 5658760.55 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3834, pruned_loss=0.134, over 5684777.51 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3878, pruned_loss=0.1359, over 5655945.60 frames. ], batch size: 262, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:56:09,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=348086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 08:56:26,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-04 08:56:47,604 INFO [train.py:968] (1/2) Epoch 8, batch 29350, giga_loss[loss=0.4024, simple_loss=0.4393, pruned_loss=0.1828, over 27885.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3885, pruned_loss=0.1369, over 5659987.77 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3837, pruned_loss=0.1341, over 5688501.23 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3887, pruned_loss=0.137, over 5654217.59 frames. ], batch size: 412, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:57:12,696 INFO [optim.py:369] (1/2) 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 08:57:34,094 INFO [train.py:968] (1/2) Epoch 8, batch 29400, giga_loss[loss=0.3449, simple_loss=0.4018, pruned_loss=0.144, over 28886.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3893, pruned_loss=0.1372, over 5667311.73 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3837, pruned_loss=0.1343, over 5692517.10 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3896, pruned_loss=0.1373, over 5658517.78 frames. ], batch size: 227, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:57:51,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 08:58:19,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4164, 1.6714, 1.3555, 1.5681], device='cuda:1'), covar=tensor([0.2049, 0.1907, 0.2024, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.1228, 0.0921, 0.1080, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 08:58:22,139 INFO [train.py:968] (1/2) Epoch 8, batch 29450, giga_loss[loss=0.344, simple_loss=0.4004, pruned_loss=0.1438, over 28614.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3899, pruned_loss=0.1383, over 5655612.23 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3833, pruned_loss=0.1341, over 5685111.46 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3906, pruned_loss=0.1387, over 5654094.59 frames. ], batch size: 307, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 08:58:36,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-04 08:58:46,039 INFO [optim.py:369] (1/2) 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 08:58:48,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9348, 1.8824, 1.6671, 1.6191], device='cuda:1'), covar=tensor([0.1315, 0.2206, 0.1799, 0.1823], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0735, 0.0654, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 08:59:06,064 INFO [train.py:968] (1/2) Epoch 8, batch 29500, giga_loss[loss=0.3009, simple_loss=0.3646, pruned_loss=0.1186, over 28791.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3874, pruned_loss=0.1371, over 5658563.27 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3825, pruned_loss=0.1334, over 5687671.39 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.389, pruned_loss=0.1382, over 5653492.46 frames. ], batch size: 119, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 08:59:38,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2086, 1.0942, 4.2071, 3.3212], device='cuda:1'), covar=tensor([0.1684, 0.2636, 0.0347, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0572, 0.0823, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 08:59:47,694 INFO [train.py:968] (1/2) Epoch 8, batch 29550, giga_loss[loss=0.365, simple_loss=0.3901, pruned_loss=0.1699, over 23554.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3872, pruned_loss=0.1376, over 5643716.50 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3819, pruned_loss=0.133, over 5681356.70 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3893, pruned_loss=0.139, over 5644047.47 frames. ], batch size: 705, lr: 4.01e-03, grad_scale: 2.0
+2023-03-04 09:00:14,395 INFO [optim.py:369] (1/2) 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 09:00:22,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4633, 1.6240, 1.7616, 1.4078], device='cuda:1'), covar=tensor([0.1188, 0.1730, 0.0953, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0716, 0.0829, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 09:00:31,334 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:00:33,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:00:35,890 INFO [train.py:968] (1/2) Epoch 8, batch 29600, giga_loss[loss=0.3358, simple_loss=0.3981, pruned_loss=0.1368, over 28747.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3878, pruned_loss=0.1375, over 5660069.94 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3816, pruned_loss=0.1329, over 5682934.87 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3897, pruned_loss=0.1389, over 5658497.53 frames. ], batch size: 199, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 09:01:25,830 INFO [train.py:968] (1/2) Epoch 8, batch 29650, libri_loss[loss=0.3252, simple_loss=0.3857, pruned_loss=0.1323, over 29535.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3889, pruned_loss=0.1386, over 5645580.97 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3818, pruned_loss=0.1329, over 5685399.09 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3904, pruned_loss=0.1397, over 5641395.02 frames. ], batch size: 84, lr: 4.01e-03, grad_scale: 4.0
+2023-03-04 09:01:45,903 INFO [optim.py:369] (1/2) 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 09:02:05,827 INFO [train.py:968] (1/2) Epoch 8, batch 29700, giga_loss[loss=0.4658, simple_loss=0.4674, pruned_loss=0.2321, over 26484.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3882, pruned_loss=0.1373, over 5653335.56 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3822, pruned_loss=0.1332, over 5673087.22 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3892, pruned_loss=0.138, over 5659678.53 frames. ], batch size: 555, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:02:31,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5725, 1.6902, 1.3107, 2.0489], device='cuda:1'), covar=tensor([0.2218, 0.2233, 0.2428, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.0922, 0.1083, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 09:02:55,556 INFO [train.py:968] (1/2) Epoch 8, batch 29750, giga_loss[loss=0.3222, simple_loss=0.3952, pruned_loss=0.1246, over 28443.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3876, pruned_loss=0.1362, over 5657094.50 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3824, pruned_loss=0.1332, over 5676585.97 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3884, pruned_loss=0.1369, over 5658815.97 frames. ], batch size: 71, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:02:58,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-04 09:03:08,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4695, 3.7767, 1.5576, 1.6392], device='cuda:1'), covar=tensor([0.0839, 0.0313, 0.0851, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0504, 0.0326, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 09:03:19,881 INFO [optim.py:369] (1/2) 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 09:03:21,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-04 09:03:39,460 INFO [train.py:968] (1/2) Epoch 8, batch 29800, giga_loss[loss=0.2956, simple_loss=0.3641, pruned_loss=0.1135, over 28961.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3878, pruned_loss=0.136, over 5663564.70 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3823, pruned_loss=0.1331, over 5684566.35 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3887, pruned_loss=0.1368, over 5657187.91 frames. ], batch size: 227, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:04:26,073 INFO [train.py:968] (1/2) Epoch 8, batch 29850, giga_loss[loss=0.3228, simple_loss=0.3823, pruned_loss=0.1316, over 28857.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3867, pruned_loss=0.1355, over 5646645.60 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.3826, pruned_loss=0.1332, over 5667267.02 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3874, pruned_loss=0.136, over 5656131.07 frames. ], batch size: 186, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:04:26,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2457, 1.3131, 1.1359, 1.0592], device='cuda:1'), covar=tensor([0.0730, 0.0470, 0.0991, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0449, 0.0503, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 09:04:51,028 INFO [optim.py:369] (1/2) 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 09:05:11,505 INFO [train.py:968] (1/2) Epoch 8, batch 29900, giga_loss[loss=0.2891, simple_loss=0.3609, pruned_loss=0.1086, over 28907.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3858, pruned_loss=0.1352, over 5661669.55 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3828, pruned_loss=0.1333, over 5671942.70 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3861, pruned_loss=0.1355, over 5664843.67 frames. ], batch size: 174, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:05:55,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:05:56,926 INFO [train.py:968] (1/2) Epoch 8, batch 29950, giga_loss[loss=0.3242, simple_loss=0.3762, pruned_loss=0.1361, over 28619.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3845, pruned_loss=0.135, over 5655711.81 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3829, pruned_loss=0.1334, over 5672047.40 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3847, pruned_loss=0.1353, over 5657796.06 frames. ], batch size: 307, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:06:21,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=348749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:06:24,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=348752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:06:30,494 INFO [optim.py:369] (1/2) 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 09:06:33,500 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 30000, giga_loss[loss=0.3241, simple_loss=0.3766, pruned_loss=0.1359, over 27878.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3809, pruned_loss=0.1332, over 5658074.73 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3827, pruned_loss=0.1333, over 5665483.05 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3813, pruned_loss=0.1336, over 5666195.69 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:06:48,546 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 09:06:57,951 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 09:07:17,970 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=348805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:07:41,481 INFO [train.py:968] (1/2) Epoch 8, batch 30050, giga_loss[loss=0.2658, simple_loss=0.3345, pruned_loss=0.09855, over 28494.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.378, pruned_loss=0.1317, over 5677012.67 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3819, pruned_loss=0.1329, over 5670195.36 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3789, pruned_loss=0.1323, over 5679271.36 frames. ], batch size: 71, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:08:09,873 INFO [optim.py:369] (1/2) 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 09:08:32,507 INFO [train.py:968] (1/2) Epoch 8, batch 30100, giga_loss[loss=0.3045, simple_loss=0.3715, pruned_loss=0.1188, over 29014.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.378, pruned_loss=0.1316, over 5682939.54 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.382, pruned_loss=0.1328, over 5671665.79 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3785, pruned_loss=0.132, over 5683394.64 frames. ], batch size: 145, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:08:47,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=348892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:08:49,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=348895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:08:49,404 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=348895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:08:51,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=348898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:09:14,331 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=348924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:09:17,105 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=348927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:09:18,228 INFO [train.py:968] (1/2) Epoch 8, batch 30150, giga_loss[loss=0.2704, simple_loss=0.344, pruned_loss=0.09836, over 28872.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3783, pruned_loss=0.1304, over 5683888.21 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3825, pruned_loss=0.1333, over 5677245.21 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3782, pruned_loss=0.1303, over 5679390.79 frames. ], batch size: 199, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:09:33,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-04 09:09:48,313 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 30200, giga_loss[loss=0.2815, simple_loss=0.3642, pruned_loss=0.09943, over 28623.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3757, pruned_loss=0.1269, over 5681926.72 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3815, pruned_loss=0.1329, over 5681536.24 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.1271, over 5674438.29 frames. ], batch size: 262, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:10:59,319 INFO [train.py:968] (1/2) Epoch 8, batch 30250, giga_loss[loss=0.2622, simple_loss=0.3463, pruned_loss=0.08909, over 28824.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3714, pruned_loss=0.1231, over 5663883.82 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3806, pruned_loss=0.1325, over 5673020.98 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3726, pruned_loss=0.1234, over 5665235.27 frames. ], batch size: 174, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:11:26,808 INFO [optim.py:369] (1/2) 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 09:11:48,081 INFO [train.py:968] (1/2) Epoch 8, batch 30300, libri_loss[loss=0.2668, simple_loss=0.3292, pruned_loss=0.1023, over 29569.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3666, pruned_loss=0.1193, over 5660011.78 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3793, pruned_loss=0.1321, over 5680904.89 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1194, over 5653110.01 frames. ], batch size: 77, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:12:06,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:12:31,986 INFO [train.py:968] (1/2) Epoch 8, batch 30350, giga_loss[loss=0.259, simple_loss=0.3412, pruned_loss=0.08838, over 28542.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.1161, over 5665598.87 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3787, pruned_loss=0.1319, over 5688935.44 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3649, pruned_loss=0.1158, over 5652035.61 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:12:39,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:12:52,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6433, 1.7086, 1.2901, 2.1384], device='cuda:1'), covar=tensor([0.2382, 0.2259, 0.2496, 0.2114], device='cuda:1'), in_proj_covar=tensor([0.1229, 0.0919, 0.1087, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 09:12:56,180 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:13:17,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1658, 1.2599, 1.1088, 1.0974], device='cuda:1'), covar=tensor([0.1083, 0.1064, 0.0681, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1471, 0.1430, 0.1542], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 09:13:24,252 INFO [train.py:968] (1/2) Epoch 8, batch 30400, giga_loss[loss=0.2928, simple_loss=0.3736, pruned_loss=0.106, over 28675.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3609, pruned_loss=0.1121, over 5644041.07 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3785, pruned_loss=0.132, over 5680124.35 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3621, pruned_loss=0.1117, over 5641193.11 frames. ], batch size: 262, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:13:25,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:13:36,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9299, 1.2561, 1.3336, 1.1516], device='cuda:1'), covar=tensor([0.1312, 0.1054, 0.1580, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0718, 0.0642, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 09:14:18,098 INFO [train.py:968] (1/2) Epoch 8, batch 30450, giga_loss[loss=0.2761, simple_loss=0.3486, pruned_loss=0.1018, over 27960.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3617, pruned_loss=0.1121, over 5637282.95 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3788, pruned_loss=0.1322, over 5673959.96 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3622, pruned_loss=0.1113, over 5640499.95 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:14:36,418 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:14:38,376 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 09:14:38,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:14:42,060 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-04 09:14:50,605 INFO [optim.py:369] (1/2) 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 09:15:06,289 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:08,826 INFO [train.py:968] (1/2) Epoch 8, batch 30500, giga_loss[loss=0.2977, simple_loss=0.3654, pruned_loss=0.1151, over 28980.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3608, pruned_loss=0.1118, over 5641542.72 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3781, pruned_loss=0.1319, over 5679113.07 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3615, pruned_loss=0.111, over 5638480.43 frames. ], batch size: 213, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:15:09,097 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:12,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:37,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0492, 2.9588, 2.2649, 0.8275], device='cuda:1'), covar=tensor([0.4060, 0.1911, 0.1924, 0.4137], device='cuda:1'), in_proj_covar=tensor([0.1479, 0.1410, 0.1454, 0.1214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 09:15:39,592 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:52,289 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:55,166 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:15:57,737 INFO [train.py:968] (1/2) Epoch 8, batch 30550, giga_loss[loss=0.302, simple_loss=0.3612, pruned_loss=0.1214, over 27520.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3578, pruned_loss=0.1099, over 5637747.48 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3776, pruned_loss=0.1318, over 5681035.98 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3584, pruned_loss=0.1091, over 5632863.56 frames. ], batch size: 472, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:16:24,278 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:16:27,630 INFO [optim.py:369] (1/2) 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 09:16:46,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1749, 1.4852, 1.1383, 0.9909], device='cuda:1'), covar=tensor([0.2197, 0.2011, 0.2267, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.0911, 0.1079, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 09:16:48,383 INFO [train.py:968] (1/2) Epoch 8, batch 30600, giga_loss[loss=0.3025, simple_loss=0.3719, pruned_loss=0.1166, over 28928.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3561, pruned_loss=0.1086, over 5644838.23 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3772, pruned_loss=0.1315, over 5685095.76 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3565, pruned_loss=0.1078, over 5636899.07 frames. ], batch size: 227, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:17:35,064 INFO [train.py:968] (1/2) Epoch 8, batch 30650, libri_loss[loss=0.353, simple_loss=0.401, pruned_loss=0.1525, over 28855.00 frames. ], tot_loss[loss=0.286, simple_loss=0.356, pruned_loss=0.108, over 5638008.79 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3772, pruned_loss=0.1316, over 5678695.86 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3559, pruned_loss=0.1067, over 5636036.34 frames. ], batch size: 107, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:18:05,408 INFO [optim.py:369] (1/2) 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 09:18:23,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 09:18:24,845 INFO [train.py:968] (1/2) Epoch 8, batch 30700, giga_loss[loss=0.282, simple_loss=0.357, pruned_loss=0.1035, over 27980.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3538, pruned_loss=0.106, over 5652600.89 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3764, pruned_loss=0.1313, over 5686287.90 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3536, pruned_loss=0.1044, over 5642559.74 frames. ], batch size: 412, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:18:38,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5159, 1.5458, 1.1809, 1.1926], device='cuda:1'), covar=tensor([0.0585, 0.0382, 0.0838, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0442, 0.0496, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 09:19:08,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 09:19:15,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=349528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:19:15,839 INFO [train.py:968] (1/2) Epoch 8, batch 30750, giga_loss[loss=0.229, simple_loss=0.3211, pruned_loss=0.0684, over 28791.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3513, pruned_loss=0.1039, over 5659548.38 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.376, pruned_loss=0.1313, over 5691743.00 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 5645911.62 frames. ], batch size: 174, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:19:43,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4764, 1.8409, 1.8372, 1.4062], device='cuda:1'), covar=tensor([0.1740, 0.2080, 0.1362, 0.1558], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0702, 0.0827, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 09:19:47,691 INFO [optim.py:369] (1/2) 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 09:20:08,777 INFO [train.py:968] (1/2) Epoch 8, batch 30800, giga_loss[loss=0.235, simple_loss=0.3161, pruned_loss=0.07693, over 28884.00 frames. ], tot_loss[loss=0.275, simple_loss=0.347, pruned_loss=0.1015, over 5646588.66 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3761, pruned_loss=0.1314, over 5693890.13 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3463, pruned_loss=0.09958, over 5633421.28 frames. ], batch size: 145, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:20:26,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1761, 1.4075, 3.5442, 3.1810], device='cuda:1'), covar=tensor([0.1560, 0.2338, 0.0441, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0562, 0.0805, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 09:20:56,184 INFO [train.py:968] (1/2) Epoch 8, batch 30850, giga_loss[loss=0.308, simple_loss=0.3602, pruned_loss=0.128, over 26698.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3453, pruned_loss=0.1008, over 5653580.93 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3754, pruned_loss=0.131, over 5694835.16 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3447, pruned_loss=0.09886, over 5640915.30 frames. ], batch size: 555, lr: 4.00e-03, grad_scale: 8.0
+2023-03-04 09:21:26,893 INFO [optim.py:369] (1/2) 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 09:21:37,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=349671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:21:40,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=349674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:21:46,367 INFO [train.py:968] (1/2) Epoch 8, batch 30900, giga_loss[loss=0.3007, simple_loss=0.3764, pruned_loss=0.1125, over 28473.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3438, pruned_loss=0.1003, over 5643539.16 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3753, pruned_loss=0.1309, over 5698245.70 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3425, pruned_loss=0.09806, over 5629218.48 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:22:03,607 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349693.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:22:14,089 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=349703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:22:26,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:22:39,600 INFO [train.py:968] (1/2) Epoch 8, batch 30950, giga_loss[loss=0.278, simple_loss=0.3527, pruned_loss=0.1017, over 27728.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3457, pruned_loss=0.1016, over 5630157.89 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.375, pruned_loss=0.1308, over 5692075.57 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3442, pruned_loss=0.09916, over 5622491.77 frames. ], batch size: 472, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:23:01,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 09:23:16,237 INFO [optim.py:369] (1/2) 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 09:23:35,086 INFO [train.py:968] (1/2) Epoch 8, batch 31000, giga_loss[loss=0.3155, simple_loss=0.3708, pruned_loss=0.1301, over 26773.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3493, pruned_loss=0.1028, over 5642523.32 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3746, pruned_loss=0.1307, over 5698078.85 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3478, pruned_loss=0.1001, over 5629414.62 frames. ], batch size: 555, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:24:09,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3570, 1.5938, 1.2528, 1.2176], device='cuda:1'), covar=tensor([0.1408, 0.1130, 0.0876, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1425, 0.1372, 0.1493], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 09:24:35,743 INFO [train.py:968] (1/2) Epoch 8, batch 31050, giga_loss[loss=0.2699, simple_loss=0.3443, pruned_loss=0.09782, over 28903.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3499, pruned_loss=0.1025, over 5654365.73 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3745, pruned_loss=0.1308, over 5694972.37 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3483, pruned_loss=0.0999, over 5645551.60 frames. ], batch size: 106, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:25:13,745 INFO [optim.py:369] (1/2) 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 09:25:25,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:25:37,008 INFO [train.py:968] (1/2) Epoch 8, batch 31100, libri_loss[loss=0.3799, simple_loss=0.4131, pruned_loss=0.1733, over 27637.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3481, pruned_loss=0.1016, over 5667247.10 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3731, pruned_loss=0.1301, over 5701105.42 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.347, pruned_loss=0.09888, over 5653211.60 frames. ], batch size: 115, lr: 4.00e-03, grad_scale: 2.0
+2023-03-04 09:26:30,175 INFO [train.py:968] (1/2) Epoch 8, batch 31150, giga_loss[loss=0.2466, simple_loss=0.3314, pruned_loss=0.08091, over 29026.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3458, pruned_loss=0.1001, over 5665557.35 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3724, pruned_loss=0.1298, over 5704773.32 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3445, pruned_loss=0.09694, over 5649548.21 frames. ], batch size: 155, lr: 4.00e-03, grad_scale: 2.0
+2023-03-04 09:27:15,567 INFO [optim.py:369] (1/2) 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 09:27:20,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3175, 3.0288, 1.3702, 1.3174], device='cuda:1'), covar=tensor([0.0907, 0.0317, 0.0913, 0.1399], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0492, 0.0324, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 09:27:35,912 INFO [train.py:968] (1/2) Epoch 8, batch 31200, giga_loss[loss=0.2867, simple_loss=0.3615, pruned_loss=0.1059, over 28443.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3443, pruned_loss=0.09765, over 5668615.02 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.372, pruned_loss=0.1295, over 5706418.37 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3433, pruned_loss=0.09512, over 5654281.51 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:27:47,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=349989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:28:28,890 INFO [train.py:968] (1/2) Epoch 8, batch 31250, giga_loss[loss=0.2515, simple_loss=0.3296, pruned_loss=0.08675, over 28547.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.343, pruned_loss=0.09813, over 5669579.76 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3713, pruned_loss=0.1293, over 5704166.81 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3414, pruned_loss=0.0945, over 5657657.88 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:29:06,603 INFO [optim.py:369] (1/2) 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 09:29:14,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4586, 3.2239, 1.4811, 1.5801], device='cuda:1'), covar=tensor([0.0889, 0.0347, 0.0876, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0496, 0.0326, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 09:29:16,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350068.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 09:29:19,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 31300, giga_loss[loss=0.2749, simple_loss=0.3489, pruned_loss=0.1004, over 28493.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.34, pruned_loss=0.09704, over 5669345.00 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3702, pruned_loss=0.1287, over 5708887.74 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.339, pruned_loss=0.09392, over 5654897.03 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:29:40,072 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 31350, giga_loss[loss=0.3027, simple_loss=0.3774, pruned_loss=0.114, over 28530.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3399, pruned_loss=0.09723, over 5663162.86 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3701, pruned_loss=0.1288, over 5700698.46 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3385, pruned_loss=0.09401, over 5658559.02 frames. ], batch size: 336, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:30:42,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2836, 4.1052, 3.8930, 1.8365], device='cuda:1'), covar=tensor([0.0519, 0.0654, 0.0820, 0.2010], device='cuda:1'), in_proj_covar=tensor([0.0966, 0.0904, 0.0804, 0.0628], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 09:31:02,680 INFO [optim.py:369] (1/2) 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 09:31:22,051 INFO [train.py:968] (1/2) Epoch 8, batch 31400, giga_loss[loss=0.2539, simple_loss=0.3402, pruned_loss=0.08382, over 28701.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3414, pruned_loss=0.09747, over 5646797.11 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3698, pruned_loss=0.1288, over 5685882.18 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3399, pruned_loss=0.0943, over 5655636.19 frames. ], batch size: 262, lr: 4.00e-03, grad_scale: 4.0
+2023-03-04 09:32:03,654 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350211.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:32:07,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350214.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:32:21,472 INFO [train.py:968] (1/2) Epoch 8, batch 31450, giga_loss[loss=0.2976, simple_loss=0.3681, pruned_loss=0.1135, over 28445.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3441, pruned_loss=0.09877, over 5656416.89 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3693, pruned_loss=0.1286, over 5689952.29 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3423, pruned_loss=0.09501, over 5658548.25 frames. ], batch size: 368, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:32:23,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 09:32:27,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:32:29,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:32:39,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350243.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:32:42,114 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:33:02,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:33:19,994 INFO [train.py:968] (1/2) Epoch 8, batch 31500, giga_loss[loss=0.2135, simple_loss=0.2979, pruned_loss=0.06456, over 28730.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3409, pruned_loss=0.09678, over 5667971.01 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3679, pruned_loss=0.1277, over 5695441.82 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3398, pruned_loss=0.09355, over 5663913.48 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:33:31,555 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 09:34:24,220 INFO [train.py:968] (1/2) Epoch 8, batch 31550, giga_loss[loss=0.2966, simple_loss=0.3739, pruned_loss=0.1097, over 28699.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3425, pruned_loss=0.09807, over 5669583.63 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3677, pruned_loss=0.1274, over 5694406.99 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3408, pruned_loss=0.0946, over 5666385.77 frames. ], batch size: 262, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:35:03,851 INFO [optim.py:369] (1/2) 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 09:35:10,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:35:28,253 INFO [train.py:968] (1/2) Epoch 8, batch 31600, giga_loss[loss=0.2562, simple_loss=0.3395, pruned_loss=0.08646, over 28059.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3455, pruned_loss=0.09885, over 5667310.49 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3679, pruned_loss=0.1276, over 5697558.96 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3437, pruned_loss=0.09557, over 5661677.02 frames. ], batch size: 412, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:35:40,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:35:43,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:36:20,418 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 8, batch 31650, giga_loss[loss=0.2688, simple_loss=0.3525, pruned_loss=0.09259, over 27709.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3483, pruned_loss=0.09759, over 5663433.29 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3677, pruned_loss=0.1275, over 5699401.48 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3469, pruned_loss=0.09483, over 5657109.01 frames. ], batch size: 472, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:36:57,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:37:38,186 INFO [train.py:968] (1/2) Epoch 8, batch 31700, giga_loss[loss=0.2672, simple_loss=0.3573, pruned_loss=0.08859, over 28950.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3487, pruned_loss=0.09626, over 5654838.17 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3677, pruned_loss=0.1275, over 5700091.48 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3475, pruned_loss=0.09396, over 5649085.92 frames. ], batch size: 106, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:38:07,281 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:38:10,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:38:33,374 INFO [train.py:968] (1/2) Epoch 8, batch 31750, giga_loss[loss=0.284, simple_loss=0.3571, pruned_loss=0.1055, over 28989.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3488, pruned_loss=0.09623, over 5659704.40 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3681, pruned_loss=0.1278, over 5705074.81 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3468, pruned_loss=0.09309, over 5649296.75 frames. ], batch size: 213, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:38:39,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8161, 2.3831, 1.7843, 1.5153], device='cuda:1'), covar=tensor([0.1508, 0.0933, 0.1091, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.1587, 0.1432, 0.1396, 0.1528], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 09:38:45,470 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:39:35,425 INFO [train.py:968] (1/2) Epoch 8, batch 31800, libri_loss[loss=0.3362, simple_loss=0.3867, pruned_loss=0.1428, over 29087.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.349, pruned_loss=0.09692, over 5662555.21 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3679, pruned_loss=0.1277, over 5709671.96 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3471, pruned_loss=0.09377, over 5648961.95 frames. ], batch size: 101, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:39:37,882 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:39:47,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:39:52,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:40:29,633 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:40:43,994 INFO [train.py:968] (1/2) Epoch 8, batch 31850, giga_loss[loss=0.2909, simple_loss=0.3637, pruned_loss=0.109, over 28975.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3477, pruned_loss=0.09707, over 5668168.76 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3677, pruned_loss=0.1276, over 5710375.62 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3462, pruned_loss=0.0943, over 5656262.39 frames. ], batch size: 199, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:41:30,997 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:42:00,225 INFO [train.py:968] (1/2) Epoch 8, batch 31900, giga_loss[loss=0.3234, simple_loss=0.357, pruned_loss=0.1449, over 27244.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3489, pruned_loss=0.09904, over 5680227.66 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3671, pruned_loss=0.1273, over 5717259.06 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3476, pruned_loss=0.09612, over 5663142.41 frames. ], batch size: 555, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:42:31,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9652, 1.1465, 3.7248, 3.0861], device='cuda:1'), covar=tensor([0.1649, 0.2469, 0.0380, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0607, 0.0564, 0.0800, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 09:42:43,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-04 09:43:10,467 INFO [train.py:968] (1/2) Epoch 8, batch 31950, giga_loss[loss=0.2192, simple_loss=0.3095, pruned_loss=0.06449, over 28695.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3438, pruned_loss=0.09587, over 5670586.44 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3666, pruned_loss=0.127, over 5709952.41 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3427, pruned_loss=0.09307, over 5663065.88 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:43:50,142 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.10 vs. limit=2.0
+2023-03-04 09:43:54,048 INFO [optim.py:369] (1/2) 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 09:44:13,352 INFO [train.py:968] (1/2) Epoch 8, batch 32000, giga_loss[loss=0.2437, simple_loss=0.3228, pruned_loss=0.08232, over 28912.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3422, pruned_loss=0.0948, over 5674273.75 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3664, pruned_loss=0.1269, over 5713007.74 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.341, pruned_loss=0.09195, over 5664458.73 frames. ], batch size: 186, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:44:38,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:44:46,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350803.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 09:44:48,762 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:44:49,421 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 09:45:15,514 INFO [train.py:968] (1/2) Epoch 8, batch 32050, giga_loss[loss=0.3682, simple_loss=0.4292, pruned_loss=0.1536, over 28668.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3409, pruned_loss=0.09504, over 5668883.97 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3661, pruned_loss=0.127, over 5707495.47 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3394, pruned_loss=0.09158, over 5663883.00 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:45:25,054 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:46:21,792 INFO [train.py:968] (1/2) Epoch 8, batch 32100, giga_loss[loss=0.2973, simple_loss=0.3599, pruned_loss=0.1173, over 26823.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3457, pruned_loss=0.09785, over 5668856.09 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3662, pruned_loss=0.1271, over 5710420.26 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3441, pruned_loss=0.09456, over 5661796.46 frames. ], batch size: 555, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:47:21,575 INFO [train.py:968] (1/2) Epoch 8, batch 32150, giga_loss[loss=0.2563, simple_loss=0.3411, pruned_loss=0.08573, over 28829.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3478, pruned_loss=0.09935, over 5665634.85 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3661, pruned_loss=0.127, over 5703311.58 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3463, pruned_loss=0.09632, over 5665519.52 frames. ], batch size: 174, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:47:38,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=350941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:47:45,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=350944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:48:01,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=350957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:48:07,579 INFO [optim.py:369] (1/2) 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 09:48:16,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8496, 1.8757, 1.4050, 1.6481], device='cuda:1'), covar=tensor([0.0641, 0.0599, 0.0835, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0438, 0.0497, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 09:48:18,272 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=350973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:48:26,997 INFO [train.py:968] (1/2) Epoch 8, batch 32200, giga_loss[loss=0.2895, simple_loss=0.3579, pruned_loss=0.1106, over 28923.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3468, pruned_loss=0.09997, over 5667119.65 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3656, pruned_loss=0.1267, over 5709558.15 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3455, pruned_loss=0.09707, over 5660405.03 frames. ], batch size: 227, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:48:39,071 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=350992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:48:59,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7494, 1.0902, 2.8823, 2.6735], device='cuda:1'), covar=tensor([0.1562, 0.2291, 0.0578, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0612, 0.0571, 0.0813, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 09:49:01,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3360, 1.5473, 1.2750, 1.2345], device='cuda:1'), covar=tensor([0.1623, 0.1239, 0.1075, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.1601, 0.1443, 0.1400, 0.1528], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 09:49:24,402 INFO [train.py:968] (1/2) Epoch 8, batch 32250, giga_loss[loss=0.2253, simple_loss=0.3106, pruned_loss=0.07003, over 28792.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3468, pruned_loss=0.1007, over 5662579.80 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3652, pruned_loss=0.1265, over 5702279.53 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.0977, over 5662257.58 frames. ], batch size: 119, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:50:05,499 INFO [optim.py:369] (1/2) 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 09:50:15,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6075, 2.2073, 1.4622, 0.7906], device='cuda:1'), covar=tensor([0.4851, 0.2631, 0.2833, 0.4181], device='cuda:1'), in_proj_covar=tensor([0.1485, 0.1413, 0.1454, 0.1217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 09:50:25,713 INFO [train.py:968] (1/2) Epoch 8, batch 32300, giga_loss[loss=0.2842, simple_loss=0.3613, pruned_loss=0.1036, over 28652.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3471, pruned_loss=0.1005, over 5671575.64 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3642, pruned_loss=0.1259, over 5710827.97 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3463, pruned_loss=0.09752, over 5661896.15 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:50:33,938 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:50:54,584 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:50:59,451 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 32350, libri_loss[loss=0.2972, simple_loss=0.358, pruned_loss=0.1182, over 27671.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3478, pruned_loss=0.09933, over 5666482.58 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3641, pruned_loss=0.1258, over 5699836.15 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3469, pruned_loss=0.09652, over 5667459.97 frames. ], batch size: 116, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:51:40,657 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:52:19,543 INFO [optim.py:369] (1/2) 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 09:52:42,498 INFO [train.py:968] (1/2) Epoch 8, batch 32400, giga_loss[loss=0.2765, simple_loss=0.3556, pruned_loss=0.09868, over 28363.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3462, pruned_loss=0.09838, over 5664638.64 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3628, pruned_loss=0.125, over 5699749.34 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3458, pruned_loss=0.09548, over 5663139.07 frames. ], batch size: 368, lr: 3.99e-03, grad_scale: 8.0
+2023-03-04 09:52:43,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:53:46,640 INFO [train.py:968] (1/2) Epoch 8, batch 32450, giga_loss[loss=0.2377, simple_loss=0.3146, pruned_loss=0.08043, over 28725.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3436, pruned_loss=0.098, over 5673372.13 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3632, pruned_loss=0.1253, over 5701357.40 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3426, pruned_loss=0.09491, over 5670078.46 frames. ], batch size: 262, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:53:54,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8921, 1.8378, 1.3335, 1.4128], device='cuda:1'), covar=tensor([0.0635, 0.0533, 0.0930, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0439, 0.0500, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 09:54:36,737 INFO [optim.py:369] (1/2) 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 09:54:38,652 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:54:49,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5943, 3.5969, 1.6177, 1.6013], device='cuda:1'), covar=tensor([0.0853, 0.0277, 0.0827, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0491, 0.0326, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 09:54:57,162 INFO [train.py:968] (1/2) Epoch 8, batch 32500, giga_loss[loss=0.2954, simple_loss=0.3559, pruned_loss=0.1174, over 28950.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3374, pruned_loss=0.09528, over 5672832.34 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3632, pruned_loss=0.1253, over 5702382.61 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3364, pruned_loss=0.09269, over 5669268.25 frames. ], batch size: 174, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 09:54:59,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 09:55:49,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:55:51,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:55:53,377 INFO [train.py:968] (1/2) Epoch 8, batch 32550, giga_loss[loss=0.2763, simple_loss=0.3481, pruned_loss=0.1023, over 29042.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3377, pruned_loss=0.09611, over 5659915.26 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3638, pruned_loss=0.1258, over 5696722.05 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3352, pruned_loss=0.09228, over 5660521.32 frames. ], batch size: 285, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:56:04,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:56:24,863 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:56:35,071 INFO [optim.py:369] (1/2) 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 09:56:37,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:56:49,519 INFO [train.py:968] (1/2) Epoch 8, batch 32600, giga_loss[loss=0.2331, simple_loss=0.3154, pruned_loss=0.07538, over 28517.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3388, pruned_loss=0.09703, over 5655799.11 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3635, pruned_loss=0.1255, over 5696936.56 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3364, pruned_loss=0.09337, over 5654720.39 frames. ], batch size: 71, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:57:13,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4427, 1.8378, 1.6021, 1.6421], device='cuda:1'), covar=tensor([0.0747, 0.0281, 0.0304, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0119, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 09:57:52,553 INFO [train.py:968] (1/2) Epoch 8, batch 32650, giga_loss[loss=0.2237, simple_loss=0.3112, pruned_loss=0.06812, over 28963.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.337, pruned_loss=0.09514, over 5654672.53 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3635, pruned_loss=0.1256, over 5696215.88 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3351, pruned_loss=0.09213, over 5654233.66 frames. ], batch size: 106, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:58:27,007 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:58:51,883 INFO [train.py:968] (1/2) Epoch 8, batch 32700, giga_loss[loss=0.2452, simple_loss=0.3265, pruned_loss=0.08201, over 28960.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3362, pruned_loss=0.09417, over 5663898.87 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3626, pruned_loss=0.1251, over 5703894.28 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3345, pruned_loss=0.09101, over 5655058.55 frames. ], batch size: 120, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 09:59:31,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:59:35,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 09:59:46,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6526, 1.6010, 1.1993, 1.3890], device='cuda:1'), covar=tensor([0.0631, 0.0529, 0.0875, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0438, 0.0496, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 09:59:55,002 INFO [train.py:968] (1/2) Epoch 8, batch 32750, giga_loss[loss=0.2641, simple_loss=0.3406, pruned_loss=0.09385, over 28667.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3347, pruned_loss=0.09348, over 5655261.34 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3625, pruned_loss=0.125, over 5693588.31 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3328, pruned_loss=0.09042, over 5657077.84 frames. ], batch size: 307, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:00:14,362 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:00:32,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6935, 1.6732, 1.2820, 1.3208], device='cuda:1'), covar=tensor([0.0721, 0.0565, 0.0987, 0.1010], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0437, 0.0496, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 10:00:43,869 INFO [optim.py:369] (1/2) 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 10:01:04,832 INFO [train.py:968] (1/2) Epoch 8, batch 32800, giga_loss[loss=0.2252, simple_loss=0.2937, pruned_loss=0.07832, over 24506.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.334, pruned_loss=0.09275, over 5653182.69 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3626, pruned_loss=0.1251, over 5697510.43 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3319, pruned_loss=0.08961, over 5650440.88 frames. ], batch size: 705, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:01:36,008 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:01:39,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:01:56,465 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351617.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:02:11,375 INFO [train.py:968] (1/2) Epoch 8, batch 32850, giga_loss[loss=0.2275, simple_loss=0.313, pruned_loss=0.07104, over 29029.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3342, pruned_loss=0.09284, over 5655519.94 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3622, pruned_loss=0.1248, over 5701599.98 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3322, pruned_loss=0.08981, over 5648783.51 frames. ], batch size: 175, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:02:14,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:02:23,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:02:56,260 INFO [optim.py:369] (1/2) 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 10:03:08,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-04 10:03:11,984 INFO [train.py:968] (1/2) Epoch 8, batch 32900, giga_loss[loss=0.2728, simple_loss=0.3489, pruned_loss=0.09832, over 28819.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3354, pruned_loss=0.09438, over 5649257.35 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3622, pruned_loss=0.1251, over 5689715.82 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3328, pruned_loss=0.09063, over 5651215.12 frames. ], batch size: 243, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:03:12,618 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:03:48,785 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 10:03:53,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:04:11,322 INFO [train.py:968] (1/2) Epoch 8, batch 32950, libri_loss[loss=0.258, simple_loss=0.3247, pruned_loss=0.09563, over 29524.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3342, pruned_loss=0.09381, over 5659355.36 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3619, pruned_loss=0.1249, over 5695899.22 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3316, pruned_loss=0.09004, over 5654122.18 frames. ], batch size: 81, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:04:40,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3590, 1.5631, 1.3560, 1.1513], device='cuda:1'), covar=tensor([0.1996, 0.1601, 0.1270, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1448, 0.1399, 0.1520], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 10:04:56,524 INFO [optim.py:369] (1/2) 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 10:05:10,319 INFO [train.py:968] (1/2) Epoch 8, batch 33000, giga_loss[loss=0.268, simple_loss=0.3443, pruned_loss=0.09584, over 27494.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3352, pruned_loss=0.09283, over 5663991.70 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3615, pruned_loss=0.1247, over 5699452.62 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3328, pruned_loss=0.08926, over 5655804.77 frames. ], batch size: 472, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:05:10,319 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 10:05:18,695 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 10:05:22,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:05:24,117 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=351785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:05:27,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:05:34,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-04 10:06:01,408 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:06:12,326 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 10:06:14,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 33050, giga_loss[loss=0.2811, simple_loss=0.3584, pruned_loss=0.1019, over 28448.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3386, pruned_loss=0.09383, over 5666429.60 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3613, pruned_loss=0.1247, over 5702668.01 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3365, pruned_loss=0.09056, over 5656411.22 frames. ], batch size: 336, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:06:34,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4587, 1.5846, 1.3120, 1.2349], device='cuda:1'), covar=tensor([0.1440, 0.1307, 0.0963, 0.1136], device='cuda:1'), in_proj_covar=tensor([0.1598, 0.1433, 0.1392, 0.1512], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 10:06:51,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=351856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:06:53,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=351859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:07:03,060 INFO [optim.py:369] (1/2) 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 10:07:17,059 INFO [train.py:968] (1/2) Epoch 8, batch 33100, giga_loss[loss=0.2984, simple_loss=0.3795, pruned_loss=0.1087, over 28874.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3402, pruned_loss=0.09492, over 5660151.56 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3612, pruned_loss=0.1245, over 5706408.58 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.338, pruned_loss=0.09152, over 5647309.88 frames. ], batch size: 227, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:07:20,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5823, 1.8244, 1.6570, 1.6684], device='cuda:1'), covar=tensor([0.1138, 0.1744, 0.1474, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0711, 0.0638, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 10:07:24,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0250, 3.8526, 3.6534, 1.7510], device='cuda:1'), covar=tensor([0.0560, 0.0700, 0.0723, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0890, 0.0788, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 10:07:25,011 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=351888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:07:51,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6486, 1.7894, 1.7157, 1.6858], device='cuda:1'), covar=tensor([0.1013, 0.1412, 0.1534, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0709, 0.0637, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 10:08:23,107 INFO [train.py:968] (1/2) Epoch 8, batch 33150, giga_loss[loss=0.2573, simple_loss=0.3307, pruned_loss=0.092, over 29054.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3411, pruned_loss=0.09556, over 5659048.27 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3613, pruned_loss=0.1246, over 5706560.77 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3389, pruned_loss=0.09227, over 5647932.91 frames. ], batch size: 128, lr: 3.99e-03, grad_scale: 2.0
+2023-03-04 10:09:07,347 INFO [optim.py:369] (1/2) 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 10:09:23,333 INFO [train.py:968] (1/2) Epoch 8, batch 33200, giga_loss[loss=0.2718, simple_loss=0.3508, pruned_loss=0.09637, over 28481.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3396, pruned_loss=0.09442, over 5665983.58 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3613, pruned_loss=0.1247, over 5708251.83 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3376, pruned_loss=0.09145, over 5655211.60 frames. ], batch size: 369, lr: 3.99e-03, grad_scale: 4.0
+2023-03-04 10:09:40,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=351992.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:10:03,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3879, 1.6139, 1.4078, 1.1754], device='cuda:1'), covar=tensor([0.1780, 0.1378, 0.1066, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.1591, 0.1426, 0.1378, 0.1498], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 10:10:24,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4463, 1.7256, 1.3085, 1.5787], device='cuda:1'), covar=tensor([0.0766, 0.0273, 0.0326, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0119, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:1')
+2023-03-04 10:10:24,550 INFO [train.py:968] (1/2) Epoch 8, batch 33250, giga_loss[loss=0.2375, simple_loss=0.3202, pruned_loss=0.07738, over 28960.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3373, pruned_loss=0.09279, over 5663348.04 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3609, pruned_loss=0.1244, over 5712526.87 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3357, pruned_loss=0.09014, over 5650360.99 frames. ], batch size: 136, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:10:49,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:11:03,240 INFO [optim.py:369] (1/2) 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 10:11:13,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4992, 1.8410, 1.2995, 0.9331], device='cuda:1'), covar=tensor([0.3528, 0.2363, 0.2251, 0.3222], device='cuda:1'), in_proj_covar=tensor([0.1481, 0.1417, 0.1449, 0.1216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 10:11:18,597 INFO [train.py:968] (1/2) Epoch 8, batch 33300, giga_loss[loss=0.2562, simple_loss=0.3375, pruned_loss=0.08746, over 28666.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3356, pruned_loss=0.0929, over 5671798.87 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3608, pruned_loss=0.1242, over 5718969.42 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3337, pruned_loss=0.08992, over 5654303.70 frames. ], batch size: 262, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:12:17,595 INFO [train.py:968] (1/2) Epoch 8, batch 33350, giga_loss[loss=0.3287, simple_loss=0.3945, pruned_loss=0.1314, over 28719.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3361, pruned_loss=0.09283, over 5671617.56 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3599, pruned_loss=0.1238, over 5712464.37 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3347, pruned_loss=0.09001, over 5662675.30 frames. ], batch size: 307, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:12:26,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352135.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:12:29,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352138.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:12:54,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:13:03,091 INFO [optim.py:369] (1/2) 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 10:13:04,625 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352167.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:13:20,564 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 33400, giga_loss[loss=0.2726, simple_loss=0.3506, pruned_loss=0.09732, over 27655.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3395, pruned_loss=0.09492, over 5663507.71 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3599, pruned_loss=0.1238, over 5707044.09 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3377, pruned_loss=0.09173, over 5659882.51 frames. ], batch size: 472, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:13:44,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:13:46,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:13:47,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:14:05,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 10:14:23,303 INFO [train.py:968] (1/2) Epoch 8, batch 33450, giga_loss[loss=0.2871, simple_loss=0.3595, pruned_loss=0.1074, over 28717.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.34, pruned_loss=0.09535, over 5661458.00 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.36, pruned_loss=0.1238, over 5709145.97 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3383, pruned_loss=0.09255, over 5656423.03 frames. ], batch size: 242, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:14:26,326 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:14:29,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 10:15:12,306 INFO [optim.py:369] (1/2) 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 10:15:29,580 INFO [train.py:968] (1/2) Epoch 8, batch 33500, giga_loss[loss=0.3452, simple_loss=0.3932, pruned_loss=0.1486, over 26707.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3432, pruned_loss=0.09753, over 5668435.65 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3598, pruned_loss=0.1239, over 5711570.77 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3414, pruned_loss=0.09432, over 5661344.58 frames. ], batch size: 555, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:15:59,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:03,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:25,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 33550, giga_loss[loss=0.2412, simple_loss=0.3317, pruned_loss=0.0754, over 28968.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3449, pruned_loss=0.097, over 5664339.52 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3595, pruned_loss=0.1237, over 5710255.92 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3437, pruned_loss=0.09454, over 5659836.36 frames. ], batch size: 155, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:16:34,534 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:44,810 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:16:47,656 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=352376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:17:27,930 INFO [train.py:968] (1/2) Epoch 8, batch 33600, giga_loss[loss=0.2454, simple_loss=0.3302, pruned_loss=0.08029, over 28439.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3446, pruned_loss=0.097, over 5668514.50 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3591, pruned_loss=0.1235, over 5715293.38 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3433, pruned_loss=0.09399, over 5658501.50 frames. ], batch size: 369, lr: 3.98e-03, grad_scale: 8.0
+2023-03-04 10:17:43,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 10:18:37,663 INFO [train.py:968] (1/2) Epoch 8, batch 33650, giga_loss[loss=0.2484, simple_loss=0.3208, pruned_loss=0.08801, over 27534.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.0971, over 5658894.83 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3586, pruned_loss=0.1233, over 5706391.98 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3427, pruned_loss=0.09423, over 5657429.52 frames. ], batch size: 472, lr: 3.98e-03, grad_scale: 8.0
+2023-03-04 10:19:03,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-04 10:19:30,289 INFO [optim.py:369] (1/2) 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 10:19:44,353 INFO [train.py:968] (1/2) Epoch 8, batch 33700, giga_loss[loss=0.2307, simple_loss=0.3207, pruned_loss=0.07032, over 28894.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3428, pruned_loss=0.09702, over 5661754.61 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3587, pruned_loss=0.1234, over 5708864.40 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3417, pruned_loss=0.09431, over 5657623.04 frames. ], batch size: 164, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:20:06,576 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3193, 1.8138, 1.5304, 1.4425], device='cuda:1'), covar=tensor([0.0725, 0.0307, 0.0295, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0118, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 10:20:41,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3095, 1.5990, 1.2863, 1.4035], device='cuda:1'), covar=tensor([0.2456, 0.2331, 0.2625, 0.2018], device='cuda:1'), in_proj_covar=tensor([0.1203, 0.0899, 0.1069, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 10:20:51,465 INFO [train.py:968] (1/2) Epoch 8, batch 33750, giga_loss[loss=0.2395, simple_loss=0.3203, pruned_loss=0.07939, over 28956.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3427, pruned_loss=0.09693, over 5647816.60 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3594, pruned_loss=0.124, over 5699041.87 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3411, pruned_loss=0.0939, over 5652811.05 frames. ], batch size: 199, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:21:23,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:21:41,689 INFO [optim.py:369] (1/2) 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 10:21:57,507 INFO [train.py:968] (1/2) Epoch 8, batch 33800, giga_loss[loss=0.2447, simple_loss=0.3208, pruned_loss=0.08429, over 28782.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3419, pruned_loss=0.09712, over 5650883.76 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3595, pruned_loss=0.1242, over 5692436.06 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3402, pruned_loss=0.09407, over 5659218.38 frames. ], batch size: 263, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:22:57,094 INFO [train.py:968] (1/2) Epoch 8, batch 33850, giga_loss[loss=0.2568, simple_loss=0.3395, pruned_loss=0.08707, over 28920.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3403, pruned_loss=0.09683, over 5630883.26 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3594, pruned_loss=0.1241, over 5685862.43 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3384, pruned_loss=0.09369, over 5641726.18 frames. ], batch size: 186, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:23:42,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4240, 1.8956, 1.4597, 1.2978], device='cuda:1'), covar=tensor([0.1593, 0.1111, 0.1243, 0.1395], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1420, 0.1373, 0.1501], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1')
+2023-03-04 10:23:42,908 INFO [optim.py:369] (1/2) 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 10:23:55,126 INFO [train.py:968] (1/2) Epoch 8, batch 33900, giga_loss[loss=0.2582, simple_loss=0.3439, pruned_loss=0.08626, over 29008.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3388, pruned_loss=0.0949, over 5647116.03 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3587, pruned_loss=0.1236, over 5691976.31 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3374, pruned_loss=0.09205, over 5648576.03 frames. ], batch size: 284, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:24:14,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:24:18,588 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:24:21,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=352701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:24:52,772 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 33950, giga_loss[loss=0.2387, simple_loss=0.3315, pruned_loss=0.07296, over 28968.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3381, pruned_loss=0.09282, over 5659573.55 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3589, pruned_loss=0.1237, over 5691306.29 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3363, pruned_loss=0.08971, over 5659988.66 frames. ], batch size: 284, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:25:37,664 INFO [optim.py:369] (1/2) 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 10:25:46,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3114, 4.1457, 3.8736, 1.6999], device='cuda:1'), covar=tensor([0.0463, 0.0602, 0.0678, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.0953, 0.0892, 0.0788, 0.0622], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 10:25:49,100 INFO [train.py:968] (1/2) Epoch 8, batch 34000, libri_loss[loss=0.2858, simple_loss=0.3446, pruned_loss=0.1135, over 29568.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3402, pruned_loss=0.09218, over 5669017.48 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3586, pruned_loss=0.1235, over 5696544.15 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3386, pruned_loss=0.08917, over 5663994.57 frames. ], batch size: 78, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:26:23,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4021, 1.6855, 1.6944, 1.3609], device='cuda:1'), covar=tensor([0.1570, 0.2033, 0.1260, 0.1467], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0693, 0.0821, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 10:26:38,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2203, 1.2402, 3.9981, 3.3623], device='cuda:1'), covar=tensor([0.1604, 0.2562, 0.0341, 0.1045], device='cuda:1'), in_proj_covar=tensor([0.0606, 0.0560, 0.0788, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 10:26:45,683 INFO [train.py:968] (1/2) Epoch 8, batch 34050, giga_loss[loss=0.2552, simple_loss=0.3413, pruned_loss=0.08454, over 28610.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.341, pruned_loss=0.09218, over 5663041.49 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3587, pruned_loss=0.1235, over 5692376.72 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3392, pruned_loss=0.08897, over 5662179.04 frames. ], batch size: 262, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:27:01,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=352844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:27:04,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=352847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:27:34,948 INFO [optim.py:369] (1/2) 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 10:27:35,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=352868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:27:45,808 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 34100, giga_loss[loss=0.2488, simple_loss=0.3268, pruned_loss=0.08538, over 27545.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3401, pruned_loss=0.09212, over 5650021.57 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3573, pruned_loss=0.1226, over 5677848.47 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3391, pruned_loss=0.08898, over 5660869.15 frames. ], batch size: 472, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:28:58,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7980, 1.8659, 1.2726, 1.5593], device='cuda:1'), covar=tensor([0.0769, 0.0614, 0.0982, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0434, 0.0494, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 10:28:58,848 INFO [train.py:968] (1/2) Epoch 8, batch 34150, giga_loss[loss=0.2373, simple_loss=0.323, pruned_loss=0.07579, over 28049.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3407, pruned_loss=0.09251, over 5649565.64 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3573, pruned_loss=0.1228, over 5669677.18 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3395, pruned_loss=0.0892, over 5664980.10 frames. ], batch size: 412, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:29:27,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3744, 1.8197, 1.6769, 1.2760], device='cuda:1'), covar=tensor([0.1668, 0.2082, 0.1309, 0.1503], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0691, 0.0819, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 10:29:48,582 INFO [optim.py:369] (1/2) 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 10:30:03,476 INFO [train.py:968] (1/2) Epoch 8, batch 34200, giga_loss[loss=0.2524, simple_loss=0.3409, pruned_loss=0.08192, over 28587.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3417, pruned_loss=0.09329, over 5657296.75 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3575, pruned_loss=0.1231, over 5674197.34 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3403, pruned_loss=0.08974, over 5665289.93 frames. ], batch size: 242, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:30:27,669 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=352998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:31:11,708 INFO [train.py:968] (1/2) Epoch 8, batch 34250, libri_loss[loss=0.2935, simple_loss=0.3524, pruned_loss=0.1173, over 29544.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.342, pruned_loss=0.09344, over 5664951.92 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3573, pruned_loss=0.123, over 5683519.93 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3404, pruned_loss=0.08939, over 5662248.75 frames. ], batch size: 80, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:32:05,798 INFO [optim.py:369] (1/2) 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 10:32:18,449 INFO [train.py:968] (1/2) Epoch 8, batch 34300, giga_loss[loss=0.319, simple_loss=0.3861, pruned_loss=0.126, over 28580.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3453, pruned_loss=0.09519, over 5661544.52 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3576, pruned_loss=0.1232, over 5687116.12 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3436, pruned_loss=0.09139, over 5655831.28 frames. ], batch size: 370, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:33:02,016 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 34350, giga_loss[loss=0.27, simple_loss=0.3463, pruned_loss=0.09678, over 29173.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3467, pruned_loss=0.09534, over 5676539.61 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3577, pruned_loss=0.1232, over 5691067.64 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3451, pruned_loss=0.09182, over 5668037.43 frames. ], batch size: 285, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:33:46,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:33:56,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4968, 2.1344, 1.5319, 0.6232], device='cuda:1'), covar=tensor([0.2411, 0.1708, 0.2483, 0.3039], device='cuda:1'), in_proj_covar=tensor([0.1482, 0.1424, 0.1455, 0.1214], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 10:34:03,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4728, 4.3048, 4.0811, 1.9640], device='cuda:1'), covar=tensor([0.0487, 0.0624, 0.0692, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.0960, 0.0894, 0.0795, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 10:34:23,561 INFO [optim.py:369] (1/2) 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 10:34:35,417 INFO [train.py:968] (1/2) Epoch 8, batch 34400, giga_loss[loss=0.2402, simple_loss=0.3224, pruned_loss=0.07905, over 28933.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3449, pruned_loss=0.09456, over 5686468.18 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3577, pruned_loss=0.1231, over 5692345.57 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3433, pruned_loss=0.0912, over 5678234.82 frames. ], batch size: 106, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:35:40,637 INFO [train.py:968] (1/2) Epoch 8, batch 34450, giga_loss[loss=0.2232, simple_loss=0.3129, pruned_loss=0.06676, over 28731.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.342, pruned_loss=0.09354, over 5685157.61 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3572, pruned_loss=0.1228, over 5696496.97 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3407, pruned_loss=0.09022, over 5674704.62 frames. ], batch size: 243, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:35:58,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6061, 4.4757, 4.1657, 1.8391], device='cuda:1'), covar=tensor([0.0413, 0.0499, 0.0626, 0.2092], device='cuda:1'), in_proj_covar=tensor([0.0955, 0.0886, 0.0791, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 10:36:02,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:36:28,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7809, 1.6885, 1.3043, 1.3588], device='cuda:1'), covar=tensor([0.0655, 0.0664, 0.0944, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0432, 0.0489, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 10:36:42,942 INFO [optim.py:369] (1/2) 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 10:36:52,444 INFO [train.py:968] (1/2) Epoch 8, batch 34500, libri_loss[loss=0.2398, simple_loss=0.2992, pruned_loss=0.09026, over 29382.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3398, pruned_loss=0.09134, over 5695276.70 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3568, pruned_loss=0.1223, over 5702612.10 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3387, pruned_loss=0.08823, over 5681215.84 frames. ], batch size: 67, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:37:55,366 INFO [train.py:968] (1/2) Epoch 8, batch 34550, giga_loss[loss=0.2585, simple_loss=0.3367, pruned_loss=0.09016, over 28858.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3395, pruned_loss=0.09174, over 5704807.07 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3564, pruned_loss=0.1221, over 5708848.08 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3385, pruned_loss=0.08861, over 5688035.70 frames. ], batch size: 227, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:38:03,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:38:42,886 INFO [optim.py:369] (1/2) 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 10:38:48,928 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 34600, giga_loss[loss=0.2543, simple_loss=0.3333, pruned_loss=0.08767, over 28959.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3417, pruned_loss=0.09283, over 5683530.01 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3564, pruned_loss=0.122, over 5698475.67 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3406, pruned_loss=0.08993, over 5679287.24 frames. ], batch size: 136, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:39:03,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:39:07,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:39:27,265 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353404.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:39:27,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8784, 1.0209, 3.2888, 2.7610], device='cuda:1'), covar=tensor([0.2078, 0.2850, 0.0773, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0606, 0.0557, 0.0790, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 10:39:45,143 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:39:56,363 INFO [train.py:968] (1/2) Epoch 8, batch 34650, libri_loss[loss=0.2956, simple_loss=0.3409, pruned_loss=0.1252, over 29366.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3458, pruned_loss=0.09579, over 5675287.89 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3564, pruned_loss=0.122, over 5701478.41 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3447, pruned_loss=0.09312, over 5668988.26 frames. ], batch size: 71, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:40:32,619 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 10:40:49,187 INFO [optim.py:369] (1/2) 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 10:40:58,950 INFO [train.py:968] (1/2) Epoch 8, batch 34700, giga_loss[loss=0.2598, simple_loss=0.3411, pruned_loss=0.08922, over 28993.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3434, pruned_loss=0.09495, over 5669620.69 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3566, pruned_loss=0.1222, over 5693308.02 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3424, pruned_loss=0.09257, over 5672620.91 frames. ], batch size: 285, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:40:59,982 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=353480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:07,588 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:36,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-04 10:41:40,512 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:43,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:46,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:41:53,279 INFO [train.py:968] (1/2) Epoch 8, batch 34750, giga_loss[loss=0.286, simple_loss=0.3593, pruned_loss=0.1064, over 28881.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3435, pruned_loss=0.09634, over 5664650.88 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3567, pruned_loss=0.1225, over 5692364.86 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3423, pruned_loss=0.09366, over 5667078.60 frames. ], batch size: 284, lr: 3.98e-03, grad_scale: 2.0
+2023-03-04 10:42:14,073 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:42:40,540 INFO [optim.py:369] (1/2) 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 10:42:52,197 INFO [train.py:968] (1/2) Epoch 8, batch 34800, giga_loss[loss=0.2308, simple_loss=0.3152, pruned_loss=0.07318, over 28901.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3456, pruned_loss=0.09834, over 5655996.58 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3565, pruned_loss=0.1223, over 5691085.69 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3446, pruned_loss=0.09588, over 5658809.77 frames. ], batch size: 164, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:43:11,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9379, 3.0356, 2.0181, 1.0609], device='cuda:1'), covar=tensor([0.3420, 0.1396, 0.1878, 0.2757], device='cuda:1'), in_proj_covar=tensor([0.1468, 0.1406, 0.1440, 0.1199], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 10:43:39,978 INFO [train.py:968] (1/2) Epoch 8, batch 34850, giga_loss[loss=0.3614, simple_loss=0.4224, pruned_loss=0.1502, over 28592.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3557, pruned_loss=0.1042, over 5668755.01 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3566, pruned_loss=0.1223, over 5693374.75 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3548, pruned_loss=0.102, over 5668843.30 frames. ], batch size: 60, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:43:41,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:43:44,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:14,889 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:16,108 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:19,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:21,401 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 34900, giga_loss[loss=0.316, simple_loss=0.3811, pruned_loss=0.1255, over 28449.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3636, pruned_loss=0.109, over 5675212.13 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3567, pruned_loss=0.1222, over 5696378.70 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3628, pruned_loss=0.1071, over 5671915.43 frames. ], batch size: 71, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:44:38,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4869, 1.8008, 1.4104, 1.3728], device='cuda:1'), covar=tensor([0.2068, 0.2024, 0.2231, 0.1868], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.0898, 0.1069, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 10:44:44,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:44:56,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:45:08,873 INFO [train.py:968] (1/2) Epoch 8, batch 34950, giga_loss[loss=0.2398, simple_loss=0.321, pruned_loss=0.07929, over 28707.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3626, pruned_loss=0.1095, over 5675113.68 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.357, pruned_loss=0.1222, over 5694836.07 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.362, pruned_loss=0.1074, over 5673335.58 frames. ], batch size: 242, lr: 3.98e-03, grad_scale: 4.0
+2023-03-04 10:45:42,843 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 35000, giga_loss[loss=0.2412, simple_loss=0.3108, pruned_loss=0.08578, over 28879.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3578, pruned_loss=0.1081, over 5683772.15 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3571, pruned_loss=0.1221, over 5700053.72 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3573, pruned_loss=0.1061, over 5677115.18 frames. ], batch size: 99, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:45:50,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353779.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:46:25,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4953, 2.1230, 1.5598, 0.7919], device='cuda:1'), covar=tensor([0.3220, 0.1743, 0.2622, 0.3399], device='cuda:1'), in_proj_covar=tensor([0.1465, 0.1403, 0.1435, 0.1198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 10:46:35,853 INFO [train.py:968] (1/2) Epoch 8, batch 35050, giga_loss[loss=0.2401, simple_loss=0.3035, pruned_loss=0.08835, over 28379.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3502, pruned_loss=0.1046, over 5682898.60 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3574, pruned_loss=0.122, over 5701740.85 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3495, pruned_loss=0.1029, over 5675822.42 frames. ], batch size: 65, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:46:56,583 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:46:58,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=353855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:47:00,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:47:06,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3678, 3.3647, 1.5429, 1.3923], device='cuda:1'), covar=tensor([0.0934, 0.0315, 0.0862, 0.1314], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0487, 0.0325, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 10:47:09,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 10:47:09,728 INFO [optim.py:369] (1/2) 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 10:47:17,446 INFO [train.py:968] (1/2) Epoch 8, batch 35100, giga_loss[loss=0.2422, simple_loss=0.3058, pruned_loss=0.08929, over 28898.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3419, pruned_loss=0.1008, over 5686282.91 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3577, pruned_loss=0.1222, over 5701453.89 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3409, pruned_loss=0.09888, over 5680855.03 frames. ], batch size: 112, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:47:22,425 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:47:54,582 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353922.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:47:56,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=353925.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 10:47:58,978 INFO [train.py:968] (1/2) Epoch 8, batch 35150, giga_loss[loss=0.2408, simple_loss=0.3085, pruned_loss=0.0866, over 28702.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3351, pruned_loss=0.09787, over 5686678.54 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3582, pruned_loss=0.1227, over 5705221.98 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3334, pruned_loss=0.09555, over 5678789.91 frames. ], batch size: 119, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:48:18,622 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=353954.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 10:48:23,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-04 10:48:30,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-04 10:48:32,480 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 35200, giga_loss[loss=0.3164, simple_loss=0.3713, pruned_loss=0.1307, over 28863.00 frames. ], tot_loss[loss=0.262, simple_loss=0.331, pruned_loss=0.09647, over 5683504.06 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3586, pruned_loss=0.1229, over 5697480.58 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3286, pruned_loss=0.09355, over 5682660.41 frames. ], batch size: 174, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:48:58,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=353998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:00,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:03,062 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:24,817 INFO [train.py:968] (1/2) Epoch 8, batch 35250, giga_loss[loss=0.2184, simple_loss=0.2864, pruned_loss=0.07524, over 28661.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3272, pruned_loss=0.09448, over 5693867.11 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3589, pruned_loss=0.1231, over 5699380.48 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3246, pruned_loss=0.09171, over 5691362.75 frames. ], batch size: 92, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:49:25,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:38,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:49:50,272 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:50:06,399 INFO [train.py:968] (1/2) Epoch 8, batch 35300, giga_loss[loss=0.2139, simple_loss=0.2863, pruned_loss=0.07081, over 27640.00 frames. ], tot_loss[loss=0.255, simple_loss=0.324, pruned_loss=0.09304, over 5689426.80 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3596, pruned_loss=0.1235, over 5702112.99 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3205, pruned_loss=0.0898, over 5684827.36 frames. ], batch size: 472, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:50:17,005 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:50:45,304 INFO [train.py:968] (1/2) Epoch 8, batch 35350, giga_loss[loss=0.249, simple_loss=0.3128, pruned_loss=0.09253, over 29010.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3238, pruned_loss=0.09368, over 5676793.20 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3612, pruned_loss=0.1244, over 5697932.58 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3179, pruned_loss=0.08882, over 5675778.28 frames. ], batch size: 128, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:50:59,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4328, 1.5469, 1.3380, 1.5908], device='cuda:1'), covar=tensor([0.0725, 0.0319, 0.0311, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0118, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 10:51:24,221 INFO [optim.py:369] (1/2) 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 10:51:29,456 INFO [train.py:968] (1/2) Epoch 8, batch 35400, giga_loss[loss=0.2559, simple_loss=0.3125, pruned_loss=0.09968, over 28031.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3195, pruned_loss=0.09118, over 5683266.99 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3613, pruned_loss=0.1244, over 5702126.23 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.314, pruned_loss=0.08673, over 5678368.93 frames. ], batch size: 77, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:52:14,638 INFO [train.py:968] (1/2) Epoch 8, batch 35450, giga_loss[loss=0.2326, simple_loss=0.3055, pruned_loss=0.07983, over 28786.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3157, pruned_loss=0.08881, over 5686443.84 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3616, pruned_loss=0.1246, over 5703257.60 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.3108, pruned_loss=0.08499, over 5681445.39 frames. ], batch size: 99, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:52:47,989 INFO [optim.py:369] (1/2) 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 10:52:54,027 INFO [train.py:968] (1/2) Epoch 8, batch 35500, giga_loss[loss=0.1908, simple_loss=0.2662, pruned_loss=0.0577, over 29087.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3127, pruned_loss=0.08748, over 5685435.65 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3621, pruned_loss=0.1248, over 5699679.13 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3073, pruned_loss=0.08336, over 5684355.98 frames. ], batch size: 136, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:53:34,215 INFO [train.py:968] (1/2) Epoch 8, batch 35550, libri_loss[loss=0.3487, simple_loss=0.4042, pruned_loss=0.1466, over 28577.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3126, pruned_loss=0.08781, over 5690905.83 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3629, pruned_loss=0.1251, over 5706540.95 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3056, pruned_loss=0.08278, over 5683390.38 frames. ], batch size: 106, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 10:54:13,999 INFO [optim.py:369] (1/2) 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 10:54:19,552 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 35600, giga_loss[loss=0.2078, simple_loss=0.289, pruned_loss=0.0633, over 29076.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3097, pruned_loss=0.0866, over 5683864.88 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3632, pruned_loss=0.125, over 5709101.62 frames. ], giga_tot_loss[loss=0.2331, simple_loss=0.3026, pruned_loss=0.08175, over 5674869.45 frames. ], batch size: 155, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:54:56,495 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=354420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:55:05,525 INFO [train.py:968] (1/2) Epoch 8, batch 35650, libri_loss[loss=0.311, simple_loss=0.3723, pruned_loss=0.1248, over 29529.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3135, pruned_loss=0.089, over 5664372.26 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3634, pruned_loss=0.1252, over 5691608.68 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3071, pruned_loss=0.0846, over 5672229.08 frames. ], batch size: 79, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:55:05,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-04 10:55:07,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=354432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:55:40,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=354465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:55:41,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3990, 1.5506, 1.2946, 1.7710], device='cuda:1'), covar=tensor([0.2134, 0.2098, 0.2166, 0.1829], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.0908, 0.1076, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 10:55:46,395 INFO [optim.py:369] (1/2) 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 10:55:52,103 INFO [train.py:968] (1/2) Epoch 8, batch 35700, giga_loss[loss=0.3121, simple_loss=0.3836, pruned_loss=0.1203, over 29081.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.328, pruned_loss=0.0967, over 5677569.21 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3638, pruned_loss=0.1253, over 5694506.69 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.322, pruned_loss=0.09267, over 5680755.17 frames. ], batch size: 128, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:56:19,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4446, 1.5956, 1.3028, 1.8254], device='cuda:1'), covar=tensor([0.2224, 0.2224, 0.2375, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.1208, 0.0911, 0.1076, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 10:56:32,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:56:35,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:56:39,012 INFO [train.py:968] (1/2) Epoch 8, batch 35750, giga_loss[loss=0.3152, simple_loss=0.3842, pruned_loss=0.1231, over 29101.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3414, pruned_loss=0.1043, over 5673137.23 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3641, pruned_loss=0.1254, over 5696581.78 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3362, pruned_loss=0.1008, over 5673620.96 frames. ], batch size: 128, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:57:01,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:09,789 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:13,999 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:57:21,254 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:23,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:23,713 INFO [train.py:968] (1/2) Epoch 8, batch 35800, libri_loss[loss=0.3258, simple_loss=0.393, pruned_loss=0.1293, over 29668.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3498, pruned_loss=0.1075, over 5681816.08 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3647, pruned_loss=0.1257, over 5699644.63 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3447, pruned_loss=0.1041, over 5679000.50 frames. ], batch size: 88, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:57:37,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5049, 2.0774, 1.4668, 0.6937], device='cuda:1'), covar=tensor([0.3569, 0.1868, 0.2873, 0.3765], device='cuda:1'), in_proj_covar=tensor([0.1480, 0.1404, 0.1443, 0.1202], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 10:57:38,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:44,205 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-04 10:57:49,004 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=354607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:49,694 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=354608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:57:52,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=354611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 10:58:07,347 INFO [train.py:968] (1/2) Epoch 8, batch 35850, giga_loss[loss=0.2588, simple_loss=0.3467, pruned_loss=0.08547, over 28694.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3535, pruned_loss=0.108, over 5684023.69 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3647, pruned_loss=0.1256, over 5698688.64 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3494, pruned_loss=0.1052, over 5682632.32 frames. ], batch size: 262, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:58:18,469 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:58:52,023 INFO [train.py:968] (1/2) Epoch 8, batch 35900, giga_loss[loss=0.3216, simple_loss=0.381, pruned_loss=0.1311, over 28778.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3545, pruned_loss=0.1073, over 5654113.13 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3649, pruned_loss=0.1258, over 5682180.90 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3507, pruned_loss=0.1046, over 5668273.79 frames. ], batch size: 99, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 10:58:56,725 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-04 10:59:35,312 INFO [train.py:968] (1/2) Epoch 8, batch 35950, giga_loss[loss=0.3101, simple_loss=0.3764, pruned_loss=0.1219, over 28660.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3574, pruned_loss=0.1094, over 5655555.95 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.366, pruned_loss=0.1266, over 5677519.43 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3534, pruned_loss=0.1061, over 5670647.55 frames. ], batch size: 242, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:00:10,855 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:00:14,343 INFO [optim.py:369] (1/2) 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 11:00:19,561 INFO [train.py:968] (1/2) Epoch 8, batch 36000, libri_loss[loss=0.3526, simple_loss=0.4045, pruned_loss=0.1503, over 25755.00 frames. ], tot_loss[loss=0.29, simple_loss=0.359, pruned_loss=0.1105, over 5663697.56 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3662, pruned_loss=0.1268, over 5675667.20 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3553, pruned_loss=0.1074, over 5677653.10 frames. ], batch size: 137, lr: 3.97e-03, grad_scale: 8.0
+2023-03-04 11:00:19,561 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 11:00:28,897 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 11:01:08,019 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=354823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:01:12,495 INFO [train.py:968] (1/2) Epoch 8, batch 36050, giga_loss[loss=0.2959, simple_loss=0.3705, pruned_loss=0.1106, over 28954.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3627, pruned_loss=0.1135, over 5669597.21 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3674, pruned_loss=0.1277, over 5679558.93 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3586, pruned_loss=0.1099, over 5677163.01 frames. ], batch size: 227, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:01:47,650 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 36100, giga_loss[loss=0.3453, simple_loss=0.4026, pruned_loss=0.144, over 28983.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.366, pruned_loss=0.1144, over 5686562.74 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3684, pruned_loss=0.1282, over 5681109.87 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3617, pruned_loss=0.1105, over 5691070.03 frames. ], batch size: 128, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:02:32,259 INFO [train.py:968] (1/2) Epoch 8, batch 36150, giga_loss[loss=0.2758, simple_loss=0.3629, pruned_loss=0.09433, over 29029.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3682, pruned_loss=0.115, over 5690415.97 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3682, pruned_loss=0.1281, over 5686682.90 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3648, pruned_loss=0.1116, over 5689348.23 frames. ], batch size: 164, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:03:01,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5639, 3.1487, 1.5728, 1.4989], device='cuda:1'), covar=tensor([0.0826, 0.0248, 0.0794, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0487, 0.0323, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 11:03:10,248 INFO [optim.py:369] (1/2) 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 11:03:13,372 INFO [train.py:968] (1/2) Epoch 8, batch 36200, giga_loss[loss=0.2961, simple_loss=0.3799, pruned_loss=0.1062, over 28855.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3687, pruned_loss=0.1141, over 5688472.31 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3687, pruned_loss=0.1283, over 5681096.50 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3656, pruned_loss=0.111, over 5693138.03 frames. ], batch size: 119, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:03:17,227 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-04 11:03:42,508 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-04 11:03:51,431 INFO [train.py:968] (1/2) Epoch 8, batch 36250, giga_loss[loss=0.2639, simple_loss=0.3531, pruned_loss=0.08733, over 29040.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3693, pruned_loss=0.1139, over 5686100.04 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3693, pruned_loss=0.1286, over 5680504.24 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3662, pruned_loss=0.1107, over 5689570.49 frames. ], batch size: 136, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:04:27,127 INFO [optim.py:369] (1/2) 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 11:04:31,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 11:04:31,516 INFO [train.py:968] (1/2) Epoch 8, batch 36300, giga_loss[loss=0.2513, simple_loss=0.343, pruned_loss=0.07984, over 28819.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3687, pruned_loss=0.1124, over 5699860.61 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.37, pruned_loss=0.1288, over 5685887.78 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3657, pruned_loss=0.1093, over 5698188.30 frames. ], batch size: 174, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:04:44,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-04 11:05:13,686 INFO [train.py:968] (1/2) Epoch 8, batch 36350, giga_loss[loss=0.2725, simple_loss=0.3455, pruned_loss=0.09976, over 27975.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3651, pruned_loss=0.1096, over 5696955.70 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3701, pruned_loss=0.1288, over 5688019.10 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3626, pruned_loss=0.1069, over 5693786.81 frames. ], batch size: 412, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:05:25,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:05:51,867 INFO [optim.py:369] (1/2) 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 11:05:54,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-04 11:05:56,445 INFO [train.py:968] (1/2) Epoch 8, batch 36400, libri_loss[loss=0.377, simple_loss=0.4232, pruned_loss=0.1654, over 25884.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3665, pruned_loss=0.1115, over 5683161.15 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3706, pruned_loss=0.129, over 5679325.77 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3639, pruned_loss=0.1084, over 5688979.32 frames. ], batch size: 136, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:06:09,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2699, 1.8124, 1.3670, 0.4439], device='cuda:1'), covar=tensor([0.2323, 0.1648, 0.2574, 0.3063], device='cuda:1'), in_proj_covar=tensor([0.1479, 0.1397, 0.1445, 0.1196], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 11:06:13,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:06:25,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8303, 1.8538, 1.7462, 1.6045], device='cuda:1'), covar=tensor([0.1095, 0.1550, 0.1580, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0723, 0.0641, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 11:06:37,307 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8486, 3.5556, 3.4262, 1.7125], device='cuda:1'), covar=tensor([0.0695, 0.0946, 0.0883, 0.2212], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0887, 0.0785, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 11:06:40,000 INFO [train.py:968] (1/2) Epoch 8, batch 36450, giga_loss[loss=0.2737, simple_loss=0.3507, pruned_loss=0.09841, over 28630.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3686, pruned_loss=0.1152, over 5676088.75 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3707, pruned_loss=0.129, over 5673786.24 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3662, pruned_loss=0.1124, over 5685261.87 frames. ], batch size: 71, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:06:40,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:07:19,109 INFO [optim.py:369] (1/2) 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 11:07:23,103 INFO [train.py:968] (1/2) Epoch 8, batch 36500, giga_loss[loss=0.2923, simple_loss=0.3554, pruned_loss=0.1147, over 28765.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3707, pruned_loss=0.1189, over 5681718.61 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3707, pruned_loss=0.1291, over 5676304.95 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3689, pruned_loss=0.1164, over 5686684.64 frames. ], batch size: 99, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:07:31,033 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:07:33,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:00,463 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 36550, giga_loss[loss=0.2756, simple_loss=0.3473, pruned_loss=0.1019, over 27927.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.37, pruned_loss=0.1196, over 5684564.16 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3708, pruned_loss=0.1292, over 5679182.05 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3685, pruned_loss=0.1175, over 5685949.67 frames. ], batch size: 412, lr: 3.97e-03, grad_scale: 4.0
+2023-03-04 11:08:19,977 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:23,907 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:23,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:45,782 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=355373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:08:46,882 INFO [optim.py:369] (1/2) 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 11:08:49,570 INFO [train.py:968] (1/2) Epoch 8, batch 36600, libri_loss[loss=0.2788, simple_loss=0.3486, pruned_loss=0.1045, over 29576.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3693, pruned_loss=0.12, over 5686786.59 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3719, pruned_loss=0.1299, over 5671415.61 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3671, pruned_loss=0.1173, over 5695720.11 frames. ], batch size: 75, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:09:07,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2648, 1.3795, 1.4049, 1.0951], device='cuda:1'), covar=tensor([0.1591, 0.2569, 0.1300, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0694, 0.0821, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 11:09:32,592 INFO [train.py:968] (1/2) Epoch 8, batch 36650, giga_loss[loss=0.276, simple_loss=0.3446, pruned_loss=0.1037, over 28638.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3677, pruned_loss=0.1183, over 5686874.01 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3723, pruned_loss=0.1301, over 5670888.99 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3656, pruned_loss=0.116, over 5694729.04 frames. ], batch size: 78, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:09:43,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6441, 1.6557, 1.3049, 1.3802], device='cuda:1'), covar=tensor([0.0565, 0.0406, 0.0874, 0.0731], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0439, 0.0498, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 11:09:43,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4474, 1.6787, 1.6771, 1.4201], device='cuda:1'), covar=tensor([0.0988, 0.0952, 0.1348, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0723, 0.0641, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 11:10:12,589 INFO [optim.py:369] (1/2) 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 11:10:15,556 INFO [train.py:968] (1/2) Epoch 8, batch 36700, giga_loss[loss=0.2507, simple_loss=0.3276, pruned_loss=0.08693, over 28898.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3657, pruned_loss=0.116, over 5693041.49 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3729, pruned_loss=0.1302, over 5677161.97 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3633, pruned_loss=0.1136, over 5694159.83 frames. ], batch size: 186, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:10:58,950 INFO [train.py:968] (1/2) Epoch 8, batch 36750, giga_loss[loss=0.2592, simple_loss=0.3307, pruned_loss=0.09382, over 28334.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.363, pruned_loss=0.1143, over 5680199.46 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3737, pruned_loss=0.1307, over 5670529.51 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3599, pruned_loss=0.1114, over 5686855.45 frames. ], batch size: 368, lr: 3.97e-03, grad_scale: 2.0
+2023-03-04 11:11:09,142 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:11:20,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:11:37,398 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 36800, giga_loss[loss=0.2308, simple_loss=0.3108, pruned_loss=0.07536, over 28770.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3576, pruned_loss=0.1108, over 5696837.29 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3737, pruned_loss=0.1305, over 5678954.81 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3548, pruned_loss=0.1081, over 5695066.66 frames. ], batch size: 284, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:11:58,236 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:12:04,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:12:33,345 INFO [train.py:968] (1/2) Epoch 8, batch 36850, giga_loss[loss=0.2553, simple_loss=0.3305, pruned_loss=0.09005, over 29109.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3502, pruned_loss=0.1066, over 5685135.89 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3738, pruned_loss=0.1305, over 5684242.19 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3474, pruned_loss=0.1039, over 5679269.75 frames. ], batch size: 113, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:13:19,501 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 36900, giga_loss[loss=0.2892, simple_loss=0.3546, pruned_loss=0.1119, over 27616.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3457, pruned_loss=0.1039, over 5681342.32 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3744, pruned_loss=0.1307, over 5688576.08 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3423, pruned_loss=0.101, over 5672746.09 frames. ], batch size: 472, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:13:57,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:14:05,111 INFO [train.py:968] (1/2) Epoch 8, batch 36950, giga_loss[loss=0.2554, simple_loss=0.3352, pruned_loss=0.08775, over 29060.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3473, pruned_loss=0.1043, over 5667704.61 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3747, pruned_loss=0.1308, over 5671789.15 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3437, pruned_loss=0.1013, over 5675969.21 frames. ], batch size: 145, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:14:21,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:14:25,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:14:37,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2913, 3.0962, 2.9331, 1.3686], device='cuda:1'), covar=tensor([0.0802, 0.0914, 0.0831, 0.2329], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0886, 0.0786, 0.0625], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 11:14:45,477 INFO [optim.py:369] (1/2) 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 11:14:47,569 INFO [train.py:968] (1/2) Epoch 8, batch 37000, libri_loss[loss=0.3075, simple_loss=0.3728, pruned_loss=0.1212, over 29553.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3474, pruned_loss=0.104, over 5685163.85 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3754, pruned_loss=0.131, over 5676074.83 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.343, pruned_loss=0.1006, over 5687892.81 frames. ], batch size: 79, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:14:47,758 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=355779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:15:18,399 WARNING [optim.py:389] (1/2) Scaling gradients by 0.07581179589033127, model_norm_threshold=2250.47265625
+2023-03-04 11:15:18,485 INFO [optim.py:451] (1/2) 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 11:15:20,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-04 11:15:28,940 INFO [train.py:968] (1/2) Epoch 8, batch 37050, giga_loss[loss=0.2989, simple_loss=0.3572, pruned_loss=0.1203, over 28866.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3482, pruned_loss=0.1052, over 5688711.87 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3766, pruned_loss=0.1315, over 5683537.83 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3427, pruned_loss=0.1011, over 5684435.00 frames. ], batch size: 186, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:15:54,582 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=355862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:15:56,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=355865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:05,347 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 37100, libri_loss[loss=0.3189, simple_loss=0.3802, pruned_loss=0.1288, over 29560.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3477, pruned_loss=0.1054, over 5688683.83 frames. ], libri_tot_loss[loss=0.321, simple_loss=0.3779, pruned_loss=0.1321, over 5681594.25 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3412, pruned_loss=0.1008, over 5687659.45 frames. ], batch size: 78, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:16:13,970 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=355888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:15,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4109, 1.6527, 1.3941, 1.7177], device='cuda:1'), covar=tensor([0.0755, 0.0296, 0.0309, 0.0748], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0051, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 11:16:18,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=355894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:35,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:16:47,188 INFO [train.py:968] (1/2) Epoch 8, batch 37150, giga_loss[loss=0.2247, simple_loss=0.3095, pruned_loss=0.06991, over 28822.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3454, pruned_loss=0.1041, over 5702919.31 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3791, pruned_loss=0.1328, over 5685248.66 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3381, pruned_loss=0.09884, over 5699509.20 frames. ], batch size: 174, lr: 3.96e-03, grad_scale: 2.0
+2023-03-04 11:16:48,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:17:20,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=355971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:17:23,540 INFO [optim.py:369] (1/2) 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 11:17:25,450 INFO [train.py:968] (1/2) Epoch 8, batch 37200, giga_loss[loss=0.2207, simple_loss=0.3035, pruned_loss=0.06897, over 29029.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3415, pruned_loss=0.1016, over 5711917.49 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3796, pruned_loss=0.1329, over 5688806.02 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3347, pruned_loss=0.09687, over 5706213.89 frames. ], batch size: 136, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:18:07,150 INFO [train.py:968] (1/2) Epoch 8, batch 37250, giga_loss[loss=0.2795, simple_loss=0.3437, pruned_loss=0.1076, over 28837.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3405, pruned_loss=0.1013, over 5701568.95 frames. ], libri_tot_loss[loss=0.3228, simple_loss=0.3799, pruned_loss=0.1328, over 5683084.14 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3332, pruned_loss=0.09637, over 5702239.91 frames. ], batch size: 199, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:18:28,809 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:31,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:41,969 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:42,973 INFO [optim.py:369] (1/2) 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 11:18:44,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:18:47,198 INFO [train.py:968] (1/2) Epoch 8, batch 37300, giga_loss[loss=0.2086, simple_loss=0.2879, pruned_loss=0.0647, over 28605.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3369, pruned_loss=0.09911, over 5688131.71 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3802, pruned_loss=0.1329, over 5667401.60 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3303, pruned_loss=0.09472, over 5702679.73 frames. ], batch size: 71, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:18:55,441 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:10,958 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:18,066 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:19,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:19:25,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9557, 1.9425, 1.7927, 1.7079], device='cuda:1'), covar=tensor([0.1304, 0.2134, 0.1844, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0729, 0.0653, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 11:19:27,561 INFO [train.py:968] (1/2) Epoch 8, batch 37350, giga_loss[loss=0.2336, simple_loss=0.3062, pruned_loss=0.08047, over 28460.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.334, pruned_loss=0.09727, over 5694017.72 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3808, pruned_loss=0.1332, over 5660601.40 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3277, pruned_loss=0.09309, over 5711955.72 frames. ], batch size: 65, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:19:41,306 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:20:03,457 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 37400, giga_loss[loss=0.2873, simple_loss=0.3485, pruned_loss=0.1131, over 28943.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3339, pruned_loss=0.09741, over 5705949.54 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3814, pruned_loss=0.1331, over 5665821.30 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3265, pruned_loss=0.09281, over 5716993.45 frames. ], batch size: 106, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:20:19,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0855, 2.5941, 1.7519, 1.4617], device='cuda:1'), covar=tensor([0.1945, 0.1112, 0.1474, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1459, 0.1433, 0.1543], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 11:20:27,962 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-04 11:20:47,460 INFO [train.py:968] (1/2) Epoch 8, batch 37450, giga_loss[loss=0.2254, simple_loss=0.2986, pruned_loss=0.07612, over 28921.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3314, pruned_loss=0.09597, over 5711719.48 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3815, pruned_loss=0.1331, over 5664146.11 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3251, pruned_loss=0.09203, over 5722518.40 frames. ], batch size: 145, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:21:16,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=356263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:21:27,407 INFO [optim.py:369] (1/2) 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 11:21:30,773 INFO [train.py:968] (1/2) Epoch 8, batch 37500, giga_loss[loss=0.3363, simple_loss=0.3955, pruned_loss=0.1385, over 27651.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3335, pruned_loss=0.09737, over 5702470.13 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3821, pruned_loss=0.1332, over 5657843.33 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3269, pruned_loss=0.09332, over 5717515.62 frames. ], batch size: 472, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:21:47,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=356299.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 11:22:03,336 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-04 11:22:11,707 INFO [train.py:968] (1/2) Epoch 8, batch 37550, giga_loss[loss=0.3009, simple_loss=0.3658, pruned_loss=0.118, over 28529.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3387, pruned_loss=0.1007, over 5702143.02 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3828, pruned_loss=0.1333, over 5661094.62 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3316, pruned_loss=0.0965, over 5712362.52 frames. ], batch size: 336, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:22:57,532 INFO [optim.py:369] (1/2) 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 11:22:59,745 INFO [train.py:968] (1/2) Epoch 8, batch 37600, giga_loss[loss=0.3164, simple_loss=0.3803, pruned_loss=0.1262, over 28588.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3446, pruned_loss=0.1044, over 5701165.52 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3829, pruned_loss=0.1334, over 5663382.31 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3384, pruned_loss=0.1006, over 5707496.27 frames. ], batch size: 336, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:23:29,646 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:23:31,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:23:49,601 INFO [train.py:968] (1/2) Epoch 8, batch 37650, giga_loss[loss=0.3083, simple_loss=0.3713, pruned_loss=0.1227, over 29115.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3546, pruned_loss=0.1118, over 5687042.80 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3832, pruned_loss=0.1336, over 5662262.37 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3492, pruned_loss=0.1085, over 5693136.99 frames. ], batch size: 128, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:23:53,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2368, 1.4836, 1.2140, 1.0295], device='cuda:1'), covar=tensor([0.1373, 0.1255, 0.0867, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.1616, 0.1472, 0.1444, 0.1547], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 11:23:59,367 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:24:39,511 INFO [train.py:968] (1/2) Epoch 8, batch 37700, giga_loss[loss=0.3032, simple_loss=0.3712, pruned_loss=0.1176, over 28688.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3591, pruned_loss=0.114, over 5676344.68 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.3829, pruned_loss=0.1335, over 5666908.80 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3546, pruned_loss=0.111, over 5677335.80 frames. ], batch size: 92, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:25:20,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 11:25:24,246 INFO [train.py:968] (1/2) Epoch 8, batch 37750, giga_loss[loss=0.2918, simple_loss=0.3649, pruned_loss=0.1094, over 28860.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3637, pruned_loss=0.1158, over 5680704.13 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3831, pruned_loss=0.1336, over 5669346.55 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3597, pruned_loss=0.1132, over 5679286.32 frames. ], batch size: 119, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:25:33,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8433, 2.4653, 2.0771, 2.3516], device='cuda:1'), covar=tensor([0.0595, 0.0617, 0.0850, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0442, 0.0501, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 11:25:58,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 11:26:07,396 INFO [optim.py:369] (1/2) 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 11:26:09,547 INFO [train.py:968] (1/2) Epoch 8, batch 37800, giga_loss[loss=0.4009, simple_loss=0.4392, pruned_loss=0.1813, over 27605.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3697, pruned_loss=0.1196, over 5678810.56 frames. ], libri_tot_loss[loss=0.3254, simple_loss=0.3834, pruned_loss=0.1337, over 5674801.74 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3659, pruned_loss=0.117, over 5673040.79 frames. ], batch size: 472, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:26:50,487 INFO [train.py:968] (1/2) Epoch 8, batch 37850, giga_loss[loss=0.2834, simple_loss=0.3476, pruned_loss=0.1096, over 28521.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3677, pruned_loss=0.1182, over 5679954.37 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3835, pruned_loss=0.134, over 5679217.94 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3644, pruned_loss=0.1157, over 5671314.92 frames. ], batch size: 85, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:27:27,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=356674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 11:27:28,960 INFO [optim.py:369] (1/2) 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 11:27:31,332 INFO [train.py:968] (1/2) Epoch 8, batch 37900, giga_loss[loss=0.2915, simple_loss=0.3416, pruned_loss=0.1207, over 23821.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3621, pruned_loss=0.1132, over 5686940.40 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3831, pruned_loss=0.1338, over 5682587.35 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3595, pruned_loss=0.1111, over 5677054.37 frames. ], batch size: 705, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:27:57,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-04 11:28:12,426 INFO [train.py:968] (1/2) Epoch 8, batch 37950, libri_loss[loss=0.3519, simple_loss=0.4191, pruned_loss=0.1424, over 29639.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3612, pruned_loss=0.112, over 5685228.92 frames. ], libri_tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1344, over 5678187.51 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3577, pruned_loss=0.1091, over 5680727.59 frames. ], batch size: 91, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:28:52,590 INFO [optim.py:369] (1/2) 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 11:28:53,922 INFO [train.py:968] (1/2) Epoch 8, batch 38000, giga_loss[loss=0.2898, simple_loss=0.3644, pruned_loss=0.1076, over 28753.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3612, pruned_loss=0.1118, over 5688881.99 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3838, pruned_loss=0.1343, over 5680356.61 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.358, pruned_loss=0.109, over 5683162.43 frames. ], batch size: 284, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:28:55,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1344, 2.4820, 1.2108, 1.2884], device='cuda:1'), covar=tensor([0.0980, 0.0326, 0.0852, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0484, 0.0320, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 11:29:27,717 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=356817.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 11:29:29,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=356820.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 11:29:37,204 INFO [train.py:968] (1/2) Epoch 8, batch 38050, giga_loss[loss=0.2649, simple_loss=0.3441, pruned_loss=0.09289, over 28588.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3641, pruned_loss=0.1132, over 5683075.42 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3841, pruned_loss=0.1344, over 5674908.56 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3609, pruned_loss=0.1105, over 5683535.74 frames. ], batch size: 60, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:29:48,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 11:29:52,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1273, 0.8793, 0.8348, 1.3308], device='cuda:1'), covar=tensor([0.0796, 0.0359, 0.0345, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 11:29:54,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=356849.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 11:30:18,729 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 38100, libri_loss[loss=0.3106, simple_loss=0.3708, pruned_loss=0.1252, over 28647.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.366, pruned_loss=0.1147, over 5677020.49 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3841, pruned_loss=0.1345, over 5668358.20 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3632, pruned_loss=0.1122, over 5682610.12 frames. ], batch size: 63, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:30:35,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=356896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:31:05,485 INFO [train.py:968] (1/2) Epoch 8, batch 38150, giga_loss[loss=0.2848, simple_loss=0.3563, pruned_loss=0.1066, over 28634.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3668, pruned_loss=0.1152, over 5690226.97 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3841, pruned_loss=0.1343, over 5673756.84 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3641, pruned_loss=0.1129, over 5690192.96 frames. ], batch size: 262, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:31:18,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6068, 1.8720, 1.9715, 1.4898], device='cuda:1'), covar=tensor([0.1483, 0.1973, 0.1100, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0702, 0.0828, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 11:31:46,953 INFO [optim.py:369] (1/2) 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 11:31:48,151 INFO [train.py:968] (1/2) Epoch 8, batch 38200, giga_loss[loss=0.3107, simple_loss=0.3771, pruned_loss=0.1221, over 27923.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3684, pruned_loss=0.1171, over 5688947.14 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3843, pruned_loss=0.1343, over 5677609.69 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3659, pruned_loss=0.1151, over 5685645.96 frames. ], batch size: 412, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:32:28,677 INFO [train.py:968] (1/2) Epoch 8, batch 38250, giga_loss[loss=0.2694, simple_loss=0.3511, pruned_loss=0.09387, over 28998.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3689, pruned_loss=0.1172, over 5697569.49 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3844, pruned_loss=0.1343, over 5679303.44 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3664, pruned_loss=0.1151, over 5694024.72 frames. ], batch size: 136, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:33:06,302 INFO [optim.py:369] (1/2) 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 11:33:07,864 INFO [train.py:968] (1/2) Epoch 8, batch 38300, giga_loss[loss=0.2869, simple_loss=0.363, pruned_loss=0.1054, over 29027.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3689, pruned_loss=0.1166, over 5701012.25 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3847, pruned_loss=0.1348, over 5680076.47 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3664, pruned_loss=0.1143, over 5697771.45 frames. ], batch size: 128, lr: 3.96e-03, grad_scale: 8.0
+2023-03-04 11:33:37,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 11:33:47,736 INFO [train.py:968] (1/2) Epoch 8, batch 38350, giga_loss[loss=0.2411, simple_loss=0.3278, pruned_loss=0.07723, over 28564.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3681, pruned_loss=0.1145, over 5694313.73 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.385, pruned_loss=0.1349, over 5669354.91 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3656, pruned_loss=0.1122, over 5702555.50 frames. ], batch size: 60, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:34:26,873 INFO [optim.py:369] (1/2) 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 11:34:26,885 INFO [train.py:968] (1/2) Epoch 8, batch 38400, giga_loss[loss=0.2767, simple_loss=0.3569, pruned_loss=0.09828, over 28895.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3681, pruned_loss=0.1139, over 5689103.63 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3852, pruned_loss=0.1352, over 5662442.88 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3658, pruned_loss=0.1116, over 5701722.25 frames. ], batch size: 227, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:34:50,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2278, 1.4772, 1.3099, 1.4218], device='cuda:1'), covar=tensor([0.0778, 0.0338, 0.0317, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0051, 0.0046, 0.0079], device='cuda:1')
+2023-03-04 11:35:08,108 INFO [train.py:968] (1/2) Epoch 8, batch 38450, libri_loss[loss=0.3361, simple_loss=0.3959, pruned_loss=0.1382, over 29239.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3651, pruned_loss=0.1122, over 5686438.23 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3848, pruned_loss=0.1349, over 5659666.80 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3628, pruned_loss=0.1097, over 5701150.12 frames. ], batch size: 94, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:35:43,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=357271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:35:48,718 INFO [optim.py:369] (1/2) 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 11:35:48,730 INFO [train.py:968] (1/2) Epoch 8, batch 38500, giga_loss[loss=0.3189, simple_loss=0.3855, pruned_loss=0.1262, over 28649.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.364, pruned_loss=0.1118, over 5698404.67 frames. ], libri_tot_loss[loss=0.3272, simple_loss=0.3847, pruned_loss=0.1348, over 5665210.85 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3616, pruned_loss=0.1092, over 5706132.18 frames. ], batch size: 307, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:36:27,296 INFO [train.py:968] (1/2) Epoch 8, batch 38550, giga_loss[loss=0.3098, simple_loss=0.3718, pruned_loss=0.1239, over 28931.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.362, pruned_loss=0.1107, over 5710771.32 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3849, pruned_loss=0.1349, over 5670018.75 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3595, pruned_loss=0.1081, over 5713321.00 frames. ], batch size: 106, lr: 3.96e-03, grad_scale: 4.0
+2023-03-04 11:36:33,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 11:36:42,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-04 11:37:05,870 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 38600, giga_loss[loss=0.328, simple_loss=0.3801, pruned_loss=0.138, over 28769.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3613, pruned_loss=0.111, over 5707385.74 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3844, pruned_loss=0.1347, over 5672397.64 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3592, pruned_loss=0.1086, over 5708240.64 frames. ], batch size: 92, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:37:06,068 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=357379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:37:28,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6572, 3.4661, 3.2291, 1.5515], device='cuda:1'), covar=tensor([0.0704, 0.0816, 0.0842, 0.2416], device='cuda:1'), in_proj_covar=tensor([0.0963, 0.0902, 0.0800, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 11:37:33,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:37:35,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=357417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:37:44,356 INFO [train.py:968] (1/2) Epoch 8, batch 38650, giga_loss[loss=0.3006, simple_loss=0.3663, pruned_loss=0.1175, over 28831.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3614, pruned_loss=0.1111, over 5717436.85 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.383, pruned_loss=0.1337, over 5679595.01 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3601, pruned_loss=0.1092, over 5712956.21 frames. ], batch size: 119, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:37:58,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=357446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:38:21,962 INFO [optim.py:369] (1/2) 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 11:38:21,974 INFO [train.py:968] (1/2) Epoch 8, batch 38700, giga_loss[loss=0.2477, simple_loss=0.3333, pruned_loss=0.08107, over 28906.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3614, pruned_loss=0.1101, over 5708760.39 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.383, pruned_loss=0.1337, over 5670370.35 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3601, pruned_loss=0.1083, over 5714553.17 frames. ], batch size: 227, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:38:27,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=357486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:38:58,492 INFO [train.py:968] (1/2) Epoch 8, batch 38750, libri_loss[loss=0.3402, simple_loss=0.3978, pruned_loss=0.1414, over 26013.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3616, pruned_loss=0.11, over 5691641.45 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3824, pruned_loss=0.1333, over 5658888.99 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3599, pruned_loss=0.1077, over 5708970.29 frames. ], batch size: 136, lr: 3.95e-03, grad_scale: 2.0
+2023-03-04 11:39:35,561 INFO [train.py:968] (1/2) Epoch 8, batch 38800, giga_loss[loss=0.2479, simple_loss=0.3343, pruned_loss=0.08072, over 28568.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3606, pruned_loss=0.1092, over 5705329.10 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3823, pruned_loss=0.1331, over 5663697.85 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.359, pruned_loss=0.1071, over 5715352.17 frames. ], batch size: 60, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:39:36,999 INFO [optim.py:369] (1/2) 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 11:40:14,324 INFO [train.py:968] (1/2) Epoch 8, batch 38850, giga_loss[loss=0.2721, simple_loss=0.349, pruned_loss=0.09766, over 28963.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3594, pruned_loss=0.1092, over 5692618.13 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3818, pruned_loss=0.1328, over 5657109.68 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3577, pruned_loss=0.1068, over 5708414.95 frames. ], batch size: 164, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:40:38,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 11:40:52,036 INFO [train.py:968] (1/2) Epoch 8, batch 38900, giga_loss[loss=0.274, simple_loss=0.3478, pruned_loss=0.1001, over 28595.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3572, pruned_loss=0.1083, over 5687464.44 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3823, pruned_loss=0.1332, over 5652662.24 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3547, pruned_loss=0.1053, over 5705070.35 frames. ], batch size: 71, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:40:52,562 INFO [optim.py:369] (1/2) 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 11:41:30,216 INFO [train.py:968] (1/2) Epoch 8, batch 38950, giga_loss[loss=0.2581, simple_loss=0.3329, pruned_loss=0.09166, over 28231.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3538, pruned_loss=0.1067, over 5696587.52 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3815, pruned_loss=0.1327, over 5657729.49 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.352, pruned_loss=0.1043, over 5706589.58 frames. ], batch size: 77, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:41:48,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4964, 1.7204, 1.3947, 1.5293], device='cuda:1'), covar=tensor([0.2262, 0.2189, 0.2381, 0.2205], device='cuda:1'), in_proj_covar=tensor([0.1211, 0.0912, 0.1075, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 11:41:49,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=357754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:42:08,305 INFO [train.py:968] (1/2) Epoch 8, batch 39000, giga_loss[loss=0.2578, simple_loss=0.3347, pruned_loss=0.09039, over 29104.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3549, pruned_loss=0.1079, over 5693187.66 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.382, pruned_loss=0.1331, over 5654700.29 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3523, pruned_loss=0.105, over 5705645.54 frames. ], batch size: 155, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:42:08,305 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 11:42:15,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1865, 1.7273, 1.5346, 1.1140], device='cuda:1'), covar=tensor([0.1714, 0.2529, 0.1414, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0706, 0.0832, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 11:42:16,877 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 11:42:18,393 INFO [optim.py:369] (1/2) 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 11:42:57,042 INFO [train.py:968] (1/2) Epoch 8, batch 39050, giga_loss[loss=0.2471, simple_loss=0.3201, pruned_loss=0.08706, over 28660.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3543, pruned_loss=0.1081, over 5693078.36 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3818, pruned_loss=0.1328, over 5660013.02 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3518, pruned_loss=0.1054, over 5699128.68 frames. ], batch size: 92, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:43:21,897 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 39100, libri_loss[loss=0.3996, simple_loss=0.441, pruned_loss=0.1791, over 29640.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3514, pruned_loss=0.1068, over 5694103.83 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3818, pruned_loss=0.133, over 5655169.42 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3488, pruned_loss=0.104, over 5703347.82 frames. ], batch size: 91, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:43:37,728 INFO [optim.py:369] (1/2) 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 11:43:51,064 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:43:53,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=357900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:44:04,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3145, 1.7660, 1.3977, 1.5999], device='cuda:1'), covar=tensor([0.0750, 0.0271, 0.0321, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 11:44:11,966 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 11:44:15,314 INFO [train.py:968] (1/2) Epoch 8, batch 39150, libri_loss[loss=0.3483, simple_loss=0.409, pruned_loss=0.1438, over 29358.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3495, pruned_loss=0.1059, over 5705635.41 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3822, pruned_loss=0.1331, over 5663208.57 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.346, pruned_loss=0.1026, over 5707218.73 frames. ], batch size: 92, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:44:15,641 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 39200, giga_loss[loss=0.2399, simple_loss=0.3126, pruned_loss=0.08358, over 28913.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3468, pruned_loss=0.1047, over 5708124.16 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3821, pruned_loss=0.1333, over 5669656.30 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3433, pruned_loss=0.1013, over 5704591.86 frames. ], batch size: 106, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:44:53,198 INFO [optim.py:369] (1/2) 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 11:45:14,647 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:45:16,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:45:17,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0067, 1.8176, 1.4557, 1.5774], device='cuda:1'), covar=tensor([0.0654, 0.0673, 0.0990, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0436, 0.0494, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 11:45:22,023 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:45:35,684 INFO [train.py:968] (1/2) Epoch 8, batch 39250, giga_loss[loss=0.2785, simple_loss=0.3476, pruned_loss=0.1047, over 28840.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3461, pruned_loss=0.1043, over 5702152.01 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3823, pruned_loss=0.1335, over 5663920.58 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3426, pruned_loss=0.1009, over 5704578.45 frames. ], batch size: 199, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:45:41,183 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:46:18,265 INFO [train.py:968] (1/2) Epoch 8, batch 39300, giga_loss[loss=0.3228, simple_loss=0.3869, pruned_loss=0.1294, over 28569.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3492, pruned_loss=0.1056, over 5697922.38 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3821, pruned_loss=0.1333, over 5664138.83 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3461, pruned_loss=0.1027, over 5700398.50 frames. ], batch size: 336, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:46:19,540 INFO [optim.py:369] (1/2) 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 11:46:58,391 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 11:47:02,552 INFO [train.py:968] (1/2) Epoch 8, batch 39350, giga_loss[loss=0.3805, simple_loss=0.4122, pruned_loss=0.1743, over 23786.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3528, pruned_loss=0.1076, over 5686162.17 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.382, pruned_loss=0.1333, over 5663929.95 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3499, pruned_loss=0.105, over 5689001.82 frames. ], batch size: 705, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:47:07,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6006, 1.5335, 1.2395, 1.2456], device='cuda:1'), covar=tensor([0.0722, 0.0555, 0.1035, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0435, 0.0495, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 11:47:17,478 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:47:17,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-04 11:47:42,306 INFO [train.py:968] (1/2) Epoch 8, batch 39400, giga_loss[loss=0.2732, simple_loss=0.3463, pruned_loss=0.1001, over 28863.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3547, pruned_loss=0.108, over 5695066.60 frames. ], libri_tot_loss[loss=0.3245, simple_loss=0.382, pruned_loss=0.1335, over 5671867.54 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3512, pruned_loss=0.1046, over 5691199.28 frames. ], batch size: 186, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:47:43,538 INFO [optim.py:369] (1/2) 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 11:47:53,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2794, 1.4509, 1.4431, 1.4270], device='cuda:1'), covar=tensor([0.1143, 0.1383, 0.1584, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0725, 0.0653, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 11:48:21,811 INFO [train.py:968] (1/2) Epoch 8, batch 39450, giga_loss[loss=0.2432, simple_loss=0.3184, pruned_loss=0.08403, over 28011.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3569, pruned_loss=0.1089, over 5691035.80 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3827, pruned_loss=0.1339, over 5669018.00 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3526, pruned_loss=0.1049, over 5691380.96 frames. ], batch size: 77, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:49:03,824 INFO [train.py:968] (1/2) Epoch 8, batch 39500, giga_loss[loss=0.2778, simple_loss=0.3485, pruned_loss=0.1036, over 28952.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3565, pruned_loss=0.1082, over 5696818.20 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3831, pruned_loss=0.1342, over 5670670.59 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.352, pruned_loss=0.1041, over 5696321.30 frames. ], batch size: 106, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:49:05,216 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 39550, giga_loss[loss=0.3363, simple_loss=0.3938, pruned_loss=0.1393, over 27945.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3563, pruned_loss=0.1085, over 5680671.65 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3833, pruned_loss=0.1343, over 5653012.19 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3523, pruned_loss=0.1049, over 5695756.05 frames. ], batch size: 412, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:49:45,095 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:50:10,455 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 39600, giga_loss[loss=0.2824, simple_loss=0.3537, pruned_loss=0.1056, over 28844.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3572, pruned_loss=0.1092, over 5690393.13 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3834, pruned_loss=0.1344, over 5650881.91 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.353, pruned_loss=0.1055, over 5706481.32 frames. ], batch size: 119, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:50:24,576 INFO [optim.py:369] (1/2) 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 11:50:29,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:50:37,396 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:51:03,880 INFO [train.py:968] (1/2) Epoch 8, batch 39650, libri_loss[loss=0.2863, simple_loss=0.3427, pruned_loss=0.1149, over 29656.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.358, pruned_loss=0.1096, over 5689265.69 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.3834, pruned_loss=0.1345, over 5644078.18 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3538, pruned_loss=0.1057, over 5709547.40 frames. ], batch size: 69, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:51:32,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2160, 1.3794, 1.0743, 1.0447], device='cuda:1'), covar=tensor([0.1348, 0.1132, 0.1026, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1457, 0.1438, 0.1530], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 11:51:32,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-04 11:51:44,104 INFO [train.py:968] (1/2) Epoch 8, batch 39700, giga_loss[loss=0.2813, simple_loss=0.3623, pruned_loss=0.1002, over 28803.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3615, pruned_loss=0.1111, over 5674116.24 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3837, pruned_loss=0.1346, over 5628015.03 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3577, pruned_loss=0.1076, over 5705262.25 frames. ], batch size: 284, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:51:45,450 INFO [optim.py:369] (1/2) 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 11:52:17,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:52:23,262 INFO [train.py:968] (1/2) Epoch 8, batch 39750, giga_loss[loss=0.3243, simple_loss=0.3785, pruned_loss=0.135, over 23752.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3635, pruned_loss=0.1126, over 5685761.76 frames. ], libri_tot_loss[loss=0.3267, simple_loss=0.3839, pruned_loss=0.1347, over 5634009.76 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3598, pruned_loss=0.1092, over 5706451.07 frames. ], batch size: 705, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:52:24,948 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:52:27,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:52:48,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3378, 1.4870, 1.3630, 1.3514], device='cuda:1'), covar=tensor([0.1197, 0.1624, 0.1683, 0.1533], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0728, 0.0654, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 11:52:49,695 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 39800, giga_loss[loss=0.279, simple_loss=0.3558, pruned_loss=0.101, over 28787.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3633, pruned_loss=0.112, over 5697295.39 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3838, pruned_loss=0.1346, over 5640940.51 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3599, pruned_loss=0.1089, over 5708849.81 frames. ], batch size: 284, lr: 3.95e-03, grad_scale: 8.0
+2023-03-04 11:53:03,103 INFO [optim.py:369] (1/2) 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 11:53:07,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=358586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:53:41,672 INFO [train.py:968] (1/2) Epoch 8, batch 39850, giga_loss[loss=0.3052, simple_loss=0.3775, pruned_loss=0.1165, over 28607.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.366, pruned_loss=0.1139, over 5692951.67 frames. ], libri_tot_loss[loss=0.3274, simple_loss=0.3846, pruned_loss=0.1351, over 5638856.23 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3617, pruned_loss=0.1101, over 5707104.86 frames. ], batch size: 336, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:54:10,992 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:54:13,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:54:20,808 INFO [train.py:968] (1/2) Epoch 8, batch 39900, giga_loss[loss=0.3181, simple_loss=0.3864, pruned_loss=0.1249, over 28926.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3654, pruned_loss=0.1135, over 5694878.73 frames. ], libri_tot_loss[loss=0.3275, simple_loss=0.3847, pruned_loss=0.1352, over 5640289.95 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.362, pruned_loss=0.1103, over 5704888.57 frames. ], batch size: 174, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:54:23,137 INFO [optim.py:369] (1/2) 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 11:54:37,456 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:54:45,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:55:00,374 INFO [train.py:968] (1/2) Epoch 8, batch 39950, giga_loss[loss=0.2554, simple_loss=0.328, pruned_loss=0.09138, over 28779.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3637, pruned_loss=0.1127, over 5704866.51 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.385, pruned_loss=0.1353, over 5644662.33 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3603, pruned_loss=0.1096, over 5709883.52 frames. ], batch size: 78, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:55:01,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3179, 2.7888, 1.4452, 1.3484], device='cuda:1'), covar=tensor([0.0800, 0.0300, 0.0829, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0493, 0.0323, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 11:55:05,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:55:34,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:55:38,474 INFO [train.py:968] (1/2) Epoch 8, batch 40000, libri_loss[loss=0.4229, simple_loss=0.4504, pruned_loss=0.1978, over 19722.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3618, pruned_loss=0.1121, over 5696067.21 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3853, pruned_loss=0.1356, over 5633821.44 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3576, pruned_loss=0.1083, over 5714126.37 frames. ], batch size: 186, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:55:41,073 INFO [optim.py:369] (1/2) 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 11:56:22,095 INFO [train.py:968] (1/2) Epoch 8, batch 40050, giga_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09872, over 28862.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.358, pruned_loss=0.1097, over 5683608.65 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3855, pruned_loss=0.1358, over 5622649.85 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3544, pruned_loss=0.1063, over 5708697.70 frames. ], batch size: 199, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:56:37,413 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:56:39,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:56:59,403 INFO [train.py:968] (1/2) Epoch 8, batch 40100, giga_loss[loss=0.3161, simple_loss=0.3923, pruned_loss=0.1199, over 27615.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3582, pruned_loss=0.1092, over 5696191.51 frames. ], libri_tot_loss[loss=0.3288, simple_loss=0.3856, pruned_loss=0.136, over 5628464.71 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3545, pruned_loss=0.1057, over 5712941.06 frames. ], batch size: 472, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:57:00,872 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:57:02,646 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:02,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:23,914 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:28,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=358916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:31,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=358919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:57:40,969 INFO [train.py:968] (1/2) Epoch 8, batch 40150, giga_loss[loss=0.2757, simple_loss=0.3435, pruned_loss=0.104, over 28793.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3593, pruned_loss=0.1081, over 5694401.00 frames. ], libri_tot_loss[loss=0.3285, simple_loss=0.3854, pruned_loss=0.1358, over 5632672.70 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3562, pruned_loss=0.1051, over 5704935.23 frames. ], batch size: 99, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:57:55,474 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=358948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:58:00,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 11:58:07,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=358961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 11:58:20,620 INFO [train.py:968] (1/2) Epoch 8, batch 40200, giga_loss[loss=0.2841, simple_loss=0.3594, pruned_loss=0.1043, over 28573.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3575, pruned_loss=0.1068, over 5701004.43 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3853, pruned_loss=0.1356, over 5635303.90 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3549, pruned_loss=0.1043, over 5707512.67 frames. ], batch size: 307, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:58:23,209 INFO [optim.py:369] (1/2) 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 11:59:00,252 INFO [train.py:968] (1/2) Epoch 8, batch 40250, giga_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 28815.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3579, pruned_loss=0.1085, over 5703915.14 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3853, pruned_loss=0.1355, over 5640617.70 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3552, pruned_loss=0.1061, over 5705437.27 frames. ], batch size: 186, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:59:36,677 INFO [train.py:968] (1/2) Epoch 8, batch 40300, libri_loss[loss=0.2783, simple_loss=0.3366, pruned_loss=0.11, over 29471.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3583, pruned_loss=0.1105, over 5710904.30 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3852, pruned_loss=0.1355, over 5651168.72 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3551, pruned_loss=0.1075, over 5705069.80 frames. ], batch size: 70, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 11:59:40,117 INFO [optim.py:369] (1/2) 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 11:59:57,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=359104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:00,569 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=359107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:18,243 INFO [train.py:968] (1/2) Epoch 8, batch 40350, giga_loss[loss=0.2298, simple_loss=0.3001, pruned_loss=0.0797, over 28493.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3573, pruned_loss=0.1111, over 5712954.53 frames. ], libri_tot_loss[loss=0.3289, simple_loss=0.3859, pruned_loss=0.1359, over 5653918.87 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3536, pruned_loss=0.1079, over 5706971.73 frames. ], batch size: 60, lr: 3.95e-03, grad_scale: 4.0
+2023-03-04 12:00:22,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=359136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:22,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=359136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:00:26,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4493, 1.6896, 1.7631, 1.3404], device='cuda:1'), covar=tensor([0.1463, 0.1880, 0.1211, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0696, 0.0822, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 12:00:56,324 INFO [train.py:968] (1/2) Epoch 8, batch 40400, giga_loss[loss=0.2624, simple_loss=0.3291, pruned_loss=0.09788, over 29009.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3555, pruned_loss=0.1103, over 5717749.97 frames. ], libri_tot_loss[loss=0.329, simple_loss=0.3861, pruned_loss=0.1359, over 5655549.60 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3518, pruned_loss=0.1072, over 5712778.68 frames. ], batch size: 128, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:00:57,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4151, 1.6823, 1.4051, 1.5296], device='cuda:1'), covar=tensor([0.0731, 0.0299, 0.0316, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 12:00:58,725 INFO [optim.py:369] (1/2) 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 12:01:35,120 INFO [train.py:968] (1/2) Epoch 8, batch 40450, giga_loss[loss=0.2533, simple_loss=0.3316, pruned_loss=0.08752, over 28686.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3528, pruned_loss=0.1086, over 5715907.39 frames. ], libri_tot_loss[loss=0.3287, simple_loss=0.3859, pruned_loss=0.1358, over 5650504.81 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3496, pruned_loss=0.1058, over 5717193.53 frames. ], batch size: 262, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:02:15,138 INFO [train.py:968] (1/2) Epoch 8, batch 40500, giga_loss[loss=0.2265, simple_loss=0.3002, pruned_loss=0.07639, over 28898.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3475, pruned_loss=0.1057, over 5717545.84 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3852, pruned_loss=0.1354, over 5651315.05 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3449, pruned_loss=0.1034, over 5719073.48 frames. ], batch size: 136, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:02:18,144 INFO [optim.py:369] (1/2) 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 12:02:18,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5691, 1.8993, 1.8156, 1.3914], device='cuda:1'), covar=tensor([0.1394, 0.2096, 0.1249, 0.1435], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0699, 0.0824, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 12:02:42,445 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 12:02:44,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2270, 1.4937, 1.2483, 1.0272], device='cuda:1'), covar=tensor([0.1745, 0.1752, 0.1677, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.1218, 0.0916, 0.1082, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 12:02:52,962 INFO [train.py:968] (1/2) Epoch 8, batch 40550, giga_loss[loss=0.3436, simple_loss=0.3938, pruned_loss=0.1467, over 27645.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3434, pruned_loss=0.1035, over 5700001.61 frames. ], libri_tot_loss[loss=0.3281, simple_loss=0.3854, pruned_loss=0.1355, over 5638797.98 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.34, pruned_loss=0.1006, over 5714788.86 frames. ], batch size: 472, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:03:29,174 INFO [train.py:968] (1/2) Epoch 8, batch 40600, giga_loss[loss=0.3099, simple_loss=0.3756, pruned_loss=0.1221, over 28889.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.343, pruned_loss=0.1027, over 5703073.77 frames. ], libri_tot_loss[loss=0.3283, simple_loss=0.3856, pruned_loss=0.1355, over 5640573.02 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3393, pruned_loss=0.09986, over 5714400.11 frames. ], batch size: 186, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:03:33,207 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 40650, giga_loss[loss=0.2852, simple_loss=0.3589, pruned_loss=0.1057, over 28543.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3463, pruned_loss=0.1039, over 5707156.55 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3855, pruned_loss=0.1353, over 5647260.94 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3426, pruned_loss=0.101, over 5711459.04 frames. ], batch size: 307, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:04:52,536 INFO [train.py:968] (1/2) Epoch 8, batch 40700, giga_loss[loss=0.3352, simple_loss=0.3837, pruned_loss=0.1434, over 23999.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3492, pruned_loss=0.1049, over 5704438.25 frames. ], libri_tot_loss[loss=0.3278, simple_loss=0.3853, pruned_loss=0.1351, over 5651658.57 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3457, pruned_loss=0.1022, over 5704882.52 frames. ], batch size: 705, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:04:55,700 INFO [optim.py:369] (1/2) 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 12:05:16,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=359511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:05:29,352 INFO [train.py:968] (1/2) Epoch 8, batch 40750, giga_loss[loss=0.2638, simple_loss=0.3464, pruned_loss=0.09062, over 29034.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3527, pruned_loss=0.1062, over 5714455.75 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3854, pruned_loss=0.1353, over 5654012.80 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.349, pruned_loss=0.1031, over 5714172.47 frames. ], batch size: 155, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:06:10,894 INFO [train.py:968] (1/2) Epoch 8, batch 40800, giga_loss[loss=0.314, simple_loss=0.376, pruned_loss=0.126, over 28847.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3564, pruned_loss=0.1086, over 5710166.02 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3856, pruned_loss=0.1354, over 5651522.25 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3524, pruned_loss=0.1053, over 5713899.40 frames. ], batch size: 112, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:06:15,172 INFO [optim.py:369] (1/2) 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 12:06:36,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3670, 1.7084, 1.3448, 1.5114], device='cuda:1'), covar=tensor([0.0758, 0.0288, 0.0317, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0072, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 12:06:42,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-04 12:06:52,634 INFO [train.py:968] (1/2) Epoch 8, batch 40850, giga_loss[loss=0.2715, simple_loss=0.349, pruned_loss=0.09705, over 28871.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3583, pruned_loss=0.1099, over 5704745.86 frames. ], libri_tot_loss[loss=0.328, simple_loss=0.3855, pruned_loss=0.1353, over 5654613.37 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3548, pruned_loss=0.107, over 5706234.88 frames. ], batch size: 145, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:07:05,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9870, 1.0550, 3.6687, 2.9893], device='cuda:1'), covar=tensor([0.1640, 0.2467, 0.0409, 0.0762], device='cuda:1'), in_proj_covar=tensor([0.0614, 0.0563, 0.0816, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 12:07:17,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=359654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:07:19,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=359657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:07:42,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3395, 1.5335, 1.2087, 1.2227], device='cuda:1'), covar=tensor([0.1422, 0.1184, 0.1239, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1489, 0.1454, 0.1557], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 12:07:42,620 INFO [train.py:968] (1/2) Epoch 8, batch 40900, giga_loss[loss=0.3966, simple_loss=0.433, pruned_loss=0.18, over 27929.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3643, pruned_loss=0.1157, over 5689400.84 frames. ], libri_tot_loss[loss=0.3276, simple_loss=0.3851, pruned_loss=0.1351, over 5655741.11 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3614, pruned_loss=0.1131, over 5690506.47 frames. ], batch size: 412, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:07:45,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4262, 4.2117, 4.0056, 1.8277], device='cuda:1'), covar=tensor([0.0486, 0.0710, 0.0720, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.0980, 0.0912, 0.0810, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 12:07:47,920 INFO [optim.py:369] (1/2) 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 12:07:48,711 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=359686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:08:00,242 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 40950, giga_loss[loss=0.3176, simple_loss=0.3839, pruned_loss=0.1256, over 28761.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3713, pruned_loss=0.1214, over 5686273.54 frames. ], libri_tot_loss[loss=0.3282, simple_loss=0.3856, pruned_loss=0.1354, over 5662058.18 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3676, pruned_loss=0.1183, over 5683056.97 frames. ], batch size: 284, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:09:06,438 INFO [train.py:968] (1/2) Epoch 8, batch 41000, giga_loss[loss=0.3468, simple_loss=0.404, pruned_loss=0.1447, over 28908.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3768, pruned_loss=0.1253, over 5691890.29 frames. ], libri_tot_loss[loss=0.3273, simple_loss=0.3848, pruned_loss=0.1349, over 5666918.04 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3744, pruned_loss=0.123, over 5685328.95 frames. ], batch size: 227, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:09:12,783 INFO [optim.py:369] (1/2) 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 12:09:23,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 41050, giga_loss[loss=0.3254, simple_loss=0.3872, pruned_loss=0.1318, over 28318.00 frames. ], tot_loss[loss=0.32, simple_loss=0.381, pruned_loss=0.1295, over 5670800.01 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3843, pruned_loss=0.1347, over 5661177.05 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3793, pruned_loss=0.1276, over 5670818.50 frames. ], batch size: 65, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:10:34,506 INFO [train.py:968] (1/2) Epoch 8, batch 41100, giga_loss[loss=0.4844, simple_loss=0.4832, pruned_loss=0.2428, over 26571.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3877, pruned_loss=0.1354, over 5679790.38 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.384, pruned_loss=0.1344, over 5668760.11 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3866, pruned_loss=0.1341, over 5673116.43 frames. ], batch size: 555, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:10:39,009 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=359882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:10:40,849 INFO [optim.py:369] (1/2) 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 12:10:50,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2756, 1.4967, 1.3435, 1.4076], device='cuda:1'), covar=tensor([0.0650, 0.0405, 0.0298, 0.0659], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:1')
+2023-03-04 12:11:26,635 INFO [train.py:968] (1/2) Epoch 8, batch 41150, giga_loss[loss=0.33, simple_loss=0.3894, pruned_loss=0.1353, over 29051.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3931, pruned_loss=0.1401, over 5663438.60 frames. ], libri_tot_loss[loss=0.3269, simple_loss=0.3845, pruned_loss=0.1347, over 5671878.30 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3919, pruned_loss=0.1389, over 5655318.15 frames. ], batch size: 128, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:12:19,644 INFO [train.py:968] (1/2) Epoch 8, batch 41200, libri_loss[loss=0.3305, simple_loss=0.3951, pruned_loss=0.1329, over 27727.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3955, pruned_loss=0.1429, over 5655747.39 frames. ], libri_tot_loss[loss=0.3271, simple_loss=0.3847, pruned_loss=0.1348, over 5668334.15 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3945, pruned_loss=0.1419, over 5652472.46 frames. ], batch size: 116, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:12:26,417 INFO [optim.py:369] (1/2) 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 12:13:12,944 INFO [train.py:968] (1/2) Epoch 8, batch 41250, giga_loss[loss=0.4176, simple_loss=0.4309, pruned_loss=0.2021, over 23446.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3986, pruned_loss=0.1468, over 5628782.84 frames. ], libri_tot_loss[loss=0.3268, simple_loss=0.3845, pruned_loss=0.1346, over 5671939.33 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.3982, pruned_loss=0.1464, over 5622391.65 frames. ], batch size: 705, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:13:36,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.76 vs. limit=5.0
+2023-03-04 12:13:57,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:14:02,629 INFO [train.py:968] (1/2) Epoch 8, batch 41300, giga_loss[loss=0.42, simple_loss=0.4289, pruned_loss=0.2055, over 23567.00 frames. ], tot_loss[loss=0.353, simple_loss=0.403, pruned_loss=0.1515, over 5618877.85 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3842, pruned_loss=0.1344, over 5665509.05 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4031, pruned_loss=0.1515, over 5619030.98 frames. ], batch size: 705, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:14:08,087 INFO [optim.py:369] (1/2) 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 12:14:51,551 INFO [train.py:968] (1/2) Epoch 8, batch 41350, giga_loss[loss=0.3553, simple_loss=0.404, pruned_loss=0.1533, over 27876.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.4067, pruned_loss=0.1542, over 5623909.94 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3839, pruned_loss=0.1342, over 5659403.98 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4075, pruned_loss=0.1548, over 5629184.21 frames. ], batch size: 412, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:14:52,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2946, 2.0454, 1.6681, 1.9740], device='cuda:1'), covar=tensor([0.0648, 0.0628, 0.0896, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0449, 0.0499, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 12:15:36,280 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 41400, giga_loss[loss=0.3067, simple_loss=0.3724, pruned_loss=0.1205, over 28464.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.4068, pruned_loss=0.1553, over 5618561.60 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.3841, pruned_loss=0.1344, over 5655505.56 frames. ], giga_tot_loss[loss=0.3599, simple_loss=0.4077, pruned_loss=0.1561, over 5625694.72 frames. ], batch size: 65, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:15:47,789 INFO [optim.py:369] (1/2) 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 12:16:18,886 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:16:21,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:16:29,857 INFO [train.py:968] (1/2) Epoch 8, batch 41450, giga_loss[loss=0.2853, simple_loss=0.3624, pruned_loss=0.1041, over 29021.00 frames. ], tot_loss[loss=0.3568, simple_loss=0.4048, pruned_loss=0.1544, over 5616263.04 frames. ], libri_tot_loss[loss=0.3261, simple_loss=0.3839, pruned_loss=0.1341, over 5650766.03 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.406, pruned_loss=0.1555, over 5625167.18 frames. ], batch size: 136, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:16:45,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:16:51,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:17:10,669 INFO [train.py:968] (1/2) Epoch 8, batch 41500, giga_loss[loss=0.3649, simple_loss=0.3919, pruned_loss=0.1689, over 23494.00 frames. ], tot_loss[loss=0.356, simple_loss=0.4047, pruned_loss=0.1537, over 5608165.98 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3843, pruned_loss=0.1344, over 5640288.70 frames. ], giga_tot_loss[loss=0.3585, simple_loss=0.4063, pruned_loss=0.1553, over 5622109.86 frames. ], batch size: 705, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:17:18,546 INFO [optim.py:369] (1/2) 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 12:17:49,939 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:17:54,794 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 41550, giga_loss[loss=0.4107, simple_loss=0.4404, pruned_loss=0.1905, over 28561.00 frames. ], tot_loss[loss=0.3577, simple_loss=0.4062, pruned_loss=0.1546, over 5590132.09 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3844, pruned_loss=0.1344, over 5631770.26 frames. ], giga_tot_loss[loss=0.3597, simple_loss=0.4075, pruned_loss=0.1559, over 5607858.46 frames. ], batch size: 336, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:18:16,938 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=360343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:18:19,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8024, 1.8423, 1.2665, 1.4712], device='cuda:1'), covar=tensor([0.0732, 0.0629, 0.1064, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0450, 0.0500, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 12:18:20,338 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:18:53,104 INFO [train.py:968] (1/2) Epoch 8, batch 41600, giga_loss[loss=0.4423, simple_loss=0.4666, pruned_loss=0.209, over 28736.00 frames. ], tot_loss[loss=0.358, simple_loss=0.4064, pruned_loss=0.1548, over 5572738.80 frames. ], libri_tot_loss[loss=0.3265, simple_loss=0.3842, pruned_loss=0.1344, over 5626901.94 frames. ], giga_tot_loss[loss=0.3605, simple_loss=0.4081, pruned_loss=0.1564, over 5590430.22 frames. ], batch size: 284, lr: 3.94e-03, grad_scale: 8.0
+2023-03-04 12:18:55,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4328, 1.6441, 1.6377, 1.2418], device='cuda:1'), covar=tensor([0.1360, 0.2142, 0.1139, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0701, 0.0820, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 12:18:59,930 INFO [optim.py:369] (1/2) 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 12:19:04,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 12:19:14,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:19:17,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:19:27,934 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=360411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:19:41,552 INFO [train.py:968] (1/2) Epoch 8, batch 41650, libri_loss[loss=0.2978, simple_loss=0.3572, pruned_loss=0.1192, over 29576.00 frames. ], tot_loss[loss=0.3504, simple_loss=0.4013, pruned_loss=0.1497, over 5594224.50 frames. ], libri_tot_loss[loss=0.3262, simple_loss=0.384, pruned_loss=0.1342, over 5632395.78 frames. ], giga_tot_loss[loss=0.3536, simple_loss=0.4036, pruned_loss=0.1518, over 5601869.15 frames. ], batch size: 77, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:19:44,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:20:02,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3651, 1.4879, 1.5453, 1.4266], device='cuda:1'), covar=tensor([0.0996, 0.0933, 0.1134, 0.0962], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0732, 0.0657, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 12:20:24,775 INFO [train.py:968] (1/2) Epoch 8, batch 41700, giga_loss[loss=0.3171, simple_loss=0.3889, pruned_loss=0.1226, over 28785.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3973, pruned_loss=0.145, over 5617511.77 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3827, pruned_loss=0.1335, over 5641065.14 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.401, pruned_loss=0.1479, over 5614680.38 frames. ], batch size: 99, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:20:32,879 INFO [optim.py:369] (1/2) 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 12:21:12,191 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-04 12:21:15,505 INFO [train.py:968] (1/2) Epoch 8, batch 41750, giga_loss[loss=0.2923, simple_loss=0.3648, pruned_loss=0.1098, over 28838.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3949, pruned_loss=0.1427, over 5621090.78 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3825, pruned_loss=0.1334, over 5643156.96 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3982, pruned_loss=0.1452, over 5616404.71 frames. ], batch size: 284, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:21:56,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4431, 1.6502, 1.7931, 1.4136], device='cuda:1'), covar=tensor([0.1218, 0.1726, 0.0989, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0702, 0.0824, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 12:21:58,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1721, 3.2605, 2.1994, 1.1493], device='cuda:1'), covar=tensor([0.2101, 0.1042, 0.1554, 0.2673], device='cuda:1'), in_proj_covar=tensor([0.1511, 0.1429, 0.1460, 0.1228], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 12:22:02,182 INFO [train.py:968] (1/2) Epoch 8, batch 41800, libri_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1084, over 29564.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3924, pruned_loss=0.1408, over 5612986.27 frames. ], libri_tot_loss[loss=0.3241, simple_loss=0.3821, pruned_loss=0.1331, over 5646095.46 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3958, pruned_loss=0.1434, over 5605446.48 frames. ], batch size: 74, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:22:08,524 INFO [optim.py:369] (1/2) 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 12:22:50,481 INFO [train.py:968] (1/2) Epoch 8, batch 41850, giga_loss[loss=0.3523, simple_loss=0.4129, pruned_loss=0.1458, over 28773.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3893, pruned_loss=0.1376, over 5637482.85 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3823, pruned_loss=0.1332, over 5648608.98 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3919, pruned_loss=0.1397, over 5628936.61 frames. ], batch size: 284, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:23:19,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=360659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:23:33,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2837, 1.9330, 1.4502, 0.4602], device='cuda:1'), covar=tensor([0.2882, 0.1547, 0.2324, 0.3643], device='cuda:1'), in_proj_covar=tensor([0.1511, 0.1431, 0.1459, 0.1226], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 12:23:38,174 INFO [train.py:968] (1/2) Epoch 8, batch 41900, giga_loss[loss=0.3247, simple_loss=0.3882, pruned_loss=0.1306, over 28574.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3901, pruned_loss=0.1384, over 5644704.68 frames. ], libri_tot_loss[loss=0.3242, simple_loss=0.3822, pruned_loss=0.1331, over 5652679.44 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3923, pruned_loss=0.1401, over 5634116.77 frames. ], batch size: 65, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:23:43,322 INFO [optim.py:369] (1/2) 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 12:24:13,339 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 41950, giga_loss[loss=0.283, simple_loss=0.3559, pruned_loss=0.1051, over 28876.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3891, pruned_loss=0.1374, over 5631951.48 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3823, pruned_loss=0.1331, over 5643729.61 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.391, pruned_loss=0.1389, over 5631450.41 frames. ], batch size: 112, lr: 3.94e-03, grad_scale: 2.0
+2023-03-04 12:25:11,141 INFO [train.py:968] (1/2) Epoch 8, batch 42000, giga_loss[loss=0.3077, simple_loss=0.3763, pruned_loss=0.1196, over 28937.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3863, pruned_loss=0.1344, over 5632058.87 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3824, pruned_loss=0.1332, over 5641082.26 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.388, pruned_loss=0.1356, over 5634387.85 frames. ], batch size: 213, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:25:11,142 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 12:25:20,263 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 12:25:25,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=360786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:25:26,501 INFO [optim.py:369] (1/2) 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 12:26:05,800 INFO [train.py:968] (1/2) Epoch 8, batch 42050, giga_loss[loss=0.3036, simple_loss=0.3785, pruned_loss=0.1144, over 28897.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3874, pruned_loss=0.1324, over 5635464.31 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3821, pruned_loss=0.133, over 5637871.80 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3891, pruned_loss=0.1337, over 5640253.38 frames. ], batch size: 145, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:26:37,427 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:26:40,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:26:54,337 INFO [train.py:968] (1/2) Epoch 8, batch 42100, giga_loss[loss=0.374, simple_loss=0.4207, pruned_loss=0.1637, over 28858.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3896, pruned_loss=0.1333, over 5648627.74 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3815, pruned_loss=0.1326, over 5644368.94 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3916, pruned_loss=0.1346, over 5646772.94 frames. ], batch size: 199, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:27:01,617 INFO [optim.py:369] (1/2) 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 12:27:05,924 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 42150, giga_loss[loss=0.301, simple_loss=0.3754, pruned_loss=0.1133, over 28904.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3897, pruned_loss=0.134, over 5653026.03 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3816, pruned_loss=0.1326, over 5646988.82 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3913, pruned_loss=0.135, over 5649408.66 frames. ], batch size: 164, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:27:42,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=360929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:27:44,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=360932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:28:12,263 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=360961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:28:27,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0904, 1.4209, 1.0980, 0.1701], device='cuda:1'), covar=tensor([0.1590, 0.1584, 0.2313, 0.2962], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1438, 0.1465, 0.1232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 12:28:27,799 INFO [train.py:968] (1/2) Epoch 8, batch 42200, libri_loss[loss=0.4306, simple_loss=0.4495, pruned_loss=0.2059, over 29268.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3882, pruned_loss=0.1337, over 5662311.32 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3819, pruned_loss=0.133, over 5652743.05 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3894, pruned_loss=0.1341, over 5654010.08 frames. ], batch size: 94, lr: 3.94e-03, grad_scale: 4.0
+2023-03-04 12:28:35,453 INFO [optim.py:369] (1/2) 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 12:29:15,270 INFO [train.py:968] (1/2) Epoch 8, batch 42250, giga_loss[loss=0.3186, simple_loss=0.3801, pruned_loss=0.1285, over 28882.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3854, pruned_loss=0.1329, over 5669035.24 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3814, pruned_loss=0.1326, over 5655173.98 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3868, pruned_loss=0.1336, over 5660506.38 frames. ], batch size: 186, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:29:19,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:29:39,979 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-04 12:29:58,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2424, 1.3848, 1.3631, 1.2608], device='cuda:1'), covar=tensor([0.1099, 0.1269, 0.1686, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0742, 0.0663, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 12:29:58,731 INFO [train.py:968] (1/2) Epoch 8, batch 42300, giga_loss[loss=0.3708, simple_loss=0.4109, pruned_loss=0.1654, over 27888.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3857, pruned_loss=0.1343, over 5659370.92 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3813, pruned_loss=0.1325, over 5649685.03 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3871, pruned_loss=0.135, over 5658521.16 frames. ], batch size: 412, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:30:07,495 INFO [optim.py:369] (1/2) 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 12:30:45,664 INFO [train.py:968] (1/2) Epoch 8, batch 42350, giga_loss[loss=0.3174, simple_loss=0.3849, pruned_loss=0.1249, over 28570.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3854, pruned_loss=0.133, over 5658636.21 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3815, pruned_loss=0.1327, over 5649159.66 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3864, pruned_loss=0.1334, over 5659442.51 frames. ], batch size: 336, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:30:52,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=361137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:31:27,991 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 42400, giga_loss[loss=0.3006, simple_loss=0.3721, pruned_loss=0.1146, over 28653.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3851, pruned_loss=0.1316, over 5674856.36 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3813, pruned_loss=0.1325, over 5655402.47 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3861, pruned_loss=0.132, over 5670090.80 frames. ], batch size: 119, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:31:30,840 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:31:59,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=361209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:32:16,937 INFO [train.py:968] (1/2) Epoch 8, batch 42450, giga_loss[loss=0.3347, simple_loss=0.3826, pruned_loss=0.1434, over 27608.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3856, pruned_loss=0.132, over 5672138.03 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3815, pruned_loss=0.1327, over 5658915.42 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3863, pruned_loss=0.1322, over 5665547.88 frames. ], batch size: 472, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:32:47,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6340, 2.3521, 2.3756, 2.1109], device='cuda:1'), covar=tensor([0.1188, 0.1678, 0.1364, 0.1447], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0737, 0.0657, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 12:32:55,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3424, 1.5228, 1.1858, 1.6422], device='cuda:1'), covar=tensor([0.2412, 0.2324, 0.2643, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.1216, 0.0912, 0.1080, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 12:33:00,800 INFO [train.py:968] (1/2) Epoch 8, batch 42500, libri_loss[loss=0.3798, simple_loss=0.4256, pruned_loss=0.167, over 29521.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3846, pruned_loss=0.1319, over 5668589.02 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3817, pruned_loss=0.1327, over 5651821.98 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3851, pruned_loss=0.132, over 5669635.19 frames. ], batch size: 80, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:33:09,456 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 42550, giga_loss[loss=0.2656, simple_loss=0.3388, pruned_loss=0.09614, over 28743.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3826, pruned_loss=0.1313, over 5669343.21 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3814, pruned_loss=0.1326, over 5657103.37 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3832, pruned_loss=0.1315, over 5665626.94 frames. ], batch size: 119, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:34:30,844 INFO [train.py:968] (1/2) Epoch 8, batch 42600, giga_loss[loss=0.2906, simple_loss=0.3606, pruned_loss=0.1103, over 28975.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3815, pruned_loss=0.1313, over 5661668.39 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.382, pruned_loss=0.1329, over 5648908.50 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3816, pruned_loss=0.1311, over 5665571.19 frames. ], batch size: 227, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:34:40,479 INFO [optim.py:369] (1/2) 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 12:34:47,000 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 42650, giga_loss[loss=0.3335, simple_loss=0.3757, pruned_loss=0.1456, over 23678.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3818, pruned_loss=0.1318, over 5671117.03 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3823, pruned_loss=0.1332, over 5653039.82 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3816, pruned_loss=0.1314, over 5671112.73 frames. ], batch size: 705, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:35:24,056 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=361433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:35:42,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-04 12:35:55,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6596, 1.7826, 1.9079, 1.5127], device='cuda:1'), covar=tensor([0.1581, 0.2077, 0.1231, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0706, 0.0825, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 12:36:05,397 INFO [train.py:968] (1/2) Epoch 8, batch 42700, giga_loss[loss=0.3291, simple_loss=0.3857, pruned_loss=0.1362, over 28941.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3802, pruned_loss=0.1316, over 5665736.06 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3822, pruned_loss=0.1333, over 5652277.16 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.38, pruned_loss=0.1311, over 5667366.26 frames. ], batch size: 213, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:36:13,246 INFO [optim.py:369] (1/2) 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 12:36:37,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:36:53,728 INFO [train.py:968] (1/2) Epoch 8, batch 42750, giga_loss[loss=0.3395, simple_loss=0.3913, pruned_loss=0.1438, over 28247.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3808, pruned_loss=0.1327, over 5658518.69 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3824, pruned_loss=0.1334, over 5655934.98 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3804, pruned_loss=0.1322, over 5656741.11 frames. ], batch size: 368, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:37:01,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8234, 2.4284, 1.6620, 1.4459], device='cuda:1'), covar=tensor([0.1910, 0.1223, 0.1475, 0.1799], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1497, 0.1472, 0.1570], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 12:37:35,189 INFO [train.py:968] (1/2) Epoch 8, batch 42800, libri_loss[loss=0.2874, simple_loss=0.3504, pruned_loss=0.1122, over 29549.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3817, pruned_loss=0.1333, over 5649060.84 frames. ], libri_tot_loss[loss=0.3247, simple_loss=0.3824, pruned_loss=0.1335, over 5647254.36 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3813, pruned_loss=0.1328, over 5656830.22 frames. ], batch size: 77, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:37:45,040 INFO [optim.py:369] (1/2) 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 12:38:17,175 INFO [train.py:968] (1/2) Epoch 8, batch 42850, giga_loss[loss=0.3035, simple_loss=0.3701, pruned_loss=0.1185, over 28629.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3808, pruned_loss=0.1312, over 5665688.28 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3815, pruned_loss=0.1328, over 5655724.35 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3812, pruned_loss=0.1313, over 5664481.61 frames. ], batch size: 85, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:38:39,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=361655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:38:42,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=361658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:38:59,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2544, 3.0484, 2.8859, 1.5079], device='cuda:1'), covar=tensor([0.0887, 0.1029, 0.0941, 0.2254], device='cuda:1'), in_proj_covar=tensor([0.0993, 0.0936, 0.0825, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 12:38:59,992 INFO [train.py:968] (1/2) Epoch 8, batch 42900, giga_loss[loss=0.3767, simple_loss=0.4208, pruned_loss=0.1663, over 28680.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3816, pruned_loss=0.1309, over 5666579.74 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3816, pruned_loss=0.1329, over 5654152.86 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3818, pruned_loss=0.1309, over 5667173.99 frames. ], batch size: 307, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:39:08,114 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:39:25,782 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-04 12:39:45,923 INFO [train.py:968] (1/2) Epoch 8, batch 42950, giga_loss[loss=0.3434, simple_loss=0.4002, pruned_loss=0.1433, over 28854.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.382, pruned_loss=0.1312, over 5670622.26 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3812, pruned_loss=0.1327, over 5656813.41 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3826, pruned_loss=0.1313, over 5669197.79 frames. ], batch size: 186, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:39:55,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2062, 5.0229, 4.7384, 2.5859], device='cuda:1'), covar=tensor([0.0446, 0.0593, 0.0680, 0.1646], device='cuda:1'), in_proj_covar=tensor([0.0993, 0.0934, 0.0824, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 12:40:21,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3960, 3.0002, 1.5636, 1.4582], device='cuda:1'), covar=tensor([0.0820, 0.0297, 0.0770, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0501, 0.0326, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 12:40:28,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:40:33,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 12:40:34,832 INFO [train.py:968] (1/2) Epoch 8, batch 43000, giga_loss[loss=0.3541, simple_loss=0.4061, pruned_loss=0.151, over 29038.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3837, pruned_loss=0.1323, over 5683768.90 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3813, pruned_loss=0.1326, over 5661788.76 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.384, pruned_loss=0.1325, over 5678391.89 frames. ], batch size: 136, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:40:45,803 INFO [optim.py:369] (1/2) 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 12:41:03,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=361808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:41:22,451 INFO [train.py:968] (1/2) Epoch 8, batch 43050, giga_loss[loss=0.38, simple_loss=0.4182, pruned_loss=0.1709, over 28532.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3863, pruned_loss=0.1357, over 5684173.77 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.381, pruned_loss=0.1324, over 5667547.53 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.387, pruned_loss=0.136, over 5675059.24 frames. ], batch size: 71, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:42:14,926 INFO [train.py:968] (1/2) Epoch 8, batch 43100, giga_loss[loss=0.3099, simple_loss=0.3746, pruned_loss=0.1227, over 29050.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3871, pruned_loss=0.1376, over 5674407.40 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3813, pruned_loss=0.1328, over 5664394.66 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3876, pruned_loss=0.1377, over 5670310.73 frames. ], batch size: 164, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:42:24,180 INFO [optim.py:369] (1/2) 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 12:42:44,580 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=361912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:42:48,421 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 43150, giga_loss[loss=0.431, simple_loss=0.4579, pruned_loss=0.2021, over 24096.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.389, pruned_loss=0.1398, over 5669443.39 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3813, pruned_loss=0.1327, over 5669924.11 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3895, pruned_loss=0.14, over 5661784.49 frames. ], batch size: 705, lr: 3.93e-03, grad_scale: 2.0
+2023-03-04 12:43:16,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=361944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:22,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=361951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:23,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7613, 2.0953, 2.0985, 1.6326], device='cuda:1'), covar=tensor([0.1637, 0.1925, 0.1215, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0706, 0.0821, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 12:43:25,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=361954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:45,213 INFO [train.py:968] (1/2) Epoch 8, batch 43200, libri_loss[loss=0.3039, simple_loss=0.3691, pruned_loss=0.1194, over 29528.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.388, pruned_loss=0.1392, over 5670977.46 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3812, pruned_loss=0.1325, over 5673352.87 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3887, pruned_loss=0.1398, over 5661554.07 frames. ], batch size: 83, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:43:48,528 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=361983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:43:55,635 INFO [optim.py:369] (1/2) 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 12:44:01,813 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=361998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:44:15,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1661, 0.8977, 0.9146, 1.3492], device='cuda:1'), covar=tensor([0.0758, 0.0355, 0.0338, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 12:44:20,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 12:44:31,052 INFO [train.py:968] (1/2) Epoch 8, batch 43250, giga_loss[loss=0.3034, simple_loss=0.3839, pruned_loss=0.1114, over 28801.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3867, pruned_loss=0.1381, over 5669973.42 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3811, pruned_loss=0.1325, over 5678394.72 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3874, pruned_loss=0.1388, over 5657922.22 frames. ], batch size: 174, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:44:59,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-04 12:45:16,147 INFO [train.py:968] (1/2) Epoch 8, batch 43300, giga_loss[loss=0.2801, simple_loss=0.357, pruned_loss=0.1016, over 28785.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3866, pruned_loss=0.1363, over 5671489.33 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3822, pruned_loss=0.1333, over 5679902.56 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3863, pruned_loss=0.1361, over 5660465.44 frames. ], batch size: 119, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:45:22,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4440, 1.6031, 1.2909, 1.7336], device='cuda:1'), covar=tensor([0.2403, 0.2383, 0.2535, 0.2222], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.0920, 0.1081, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 12:45:26,952 INFO [optim.py:369] (1/2) 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 12:45:40,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4036, 1.6857, 1.3583, 1.2815], device='cuda:1'), covar=tensor([0.1788, 0.1329, 0.1149, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1488, 0.1456, 0.1561], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 12:45:50,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-04 12:46:02,727 INFO [train.py:968] (1/2) Epoch 8, batch 43350, giga_loss[loss=0.3205, simple_loss=0.3749, pruned_loss=0.1331, over 28773.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3835, pruned_loss=0.1338, over 5666195.75 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3824, pruned_loss=0.1334, over 5682189.75 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3831, pruned_loss=0.1335, over 5655309.74 frames. ], batch size: 99, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:46:28,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5074, 1.6804, 1.2202, 1.2828], device='cuda:1'), covar=tensor([0.1766, 0.1496, 0.1384, 0.1555], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1493, 0.1461, 0.1567], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 12:46:43,100 INFO [train.py:968] (1/2) Epoch 8, batch 43400, giga_loss[loss=0.2952, simple_loss=0.3635, pruned_loss=0.1135, over 28929.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3815, pruned_loss=0.1325, over 5672979.15 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.3829, pruned_loss=0.1337, over 5678341.47 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3807, pruned_loss=0.132, over 5667679.47 frames. ], batch size: 213, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:46:49,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 12:46:55,195 INFO [optim.py:369] (1/2) 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 12:47:03,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2873, 1.5191, 1.2487, 1.0949], device='cuda:1'), covar=tensor([0.1744, 0.1366, 0.0912, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.1634, 0.1494, 0.1463, 0.1566], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 12:47:04,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 12:47:07,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=362204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:47:28,525 INFO [train.py:968] (1/2) Epoch 8, batch 43450, libri_loss[loss=0.3672, simple_loss=0.422, pruned_loss=0.1562, over 29544.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3809, pruned_loss=0.133, over 5667072.98 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3831, pruned_loss=0.1337, over 5680216.56 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.38, pruned_loss=0.1326, over 5661025.90 frames. ], batch size: 89, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:48:09,395 INFO [train.py:968] (1/2) Epoch 8, batch 43500, giga_loss[loss=0.2939, simple_loss=0.369, pruned_loss=0.1095, over 28937.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3823, pruned_loss=0.1336, over 5680453.16 frames. ], libri_tot_loss[loss=0.3249, simple_loss=0.3829, pruned_loss=0.1335, over 5685904.96 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3816, pruned_loss=0.1334, over 5670318.72 frames. ], batch size: 119, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:48:21,044 INFO [optim.py:369] (1/2) 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 12:48:21,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=362292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:48:21,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6160, 1.5153, 1.6735, 1.3122], device='cuda:1'), covar=tensor([0.1933, 0.2939, 0.1487, 0.1588], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0706, 0.0828, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 12:48:56,970 INFO [train.py:968] (1/2) Epoch 8, batch 43550, giga_loss[loss=0.3147, simple_loss=0.3934, pruned_loss=0.118, over 28842.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3865, pruned_loss=0.1351, over 5670626.17 frames. ], libri_tot_loss[loss=0.3248, simple_loss=0.3829, pruned_loss=0.1334, over 5687982.56 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3861, pruned_loss=0.135, over 5660480.04 frames. ], batch size: 99, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:49:05,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4693, 1.6533, 1.3458, 1.7309], device='cuda:1'), covar=tensor([0.2030, 0.1920, 0.1978, 0.2005], device='cuda:1'), in_proj_covar=tensor([0.1222, 0.0921, 0.1081, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 12:49:41,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=362373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:49:44,549 INFO [train.py:968] (1/2) Epoch 8, batch 43600, giga_loss[loss=0.3367, simple_loss=0.4038, pruned_loss=0.1347, over 28660.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3882, pruned_loss=0.1336, over 5679974.16 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3834, pruned_loss=0.1339, over 5693698.34 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3876, pruned_loss=0.1331, over 5665968.45 frames. ], batch size: 92, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:49:57,388 INFO [optim.py:369] (1/2) 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 12:50:30,624 INFO [train.py:968] (1/2) Epoch 8, batch 43650, giga_loss[loss=0.3098, simple_loss=0.3779, pruned_loss=0.1209, over 28850.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3901, pruned_loss=0.135, over 5675310.04 frames. ], libri_tot_loss[loss=0.3255, simple_loss=0.3833, pruned_loss=0.1338, over 5696192.40 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3897, pruned_loss=0.1347, over 5661518.60 frames. ], batch size: 119, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:51:02,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-04 12:51:16,697 INFO [train.py:968] (1/2) Epoch 8, batch 43700, giga_loss[loss=0.357, simple_loss=0.4088, pruned_loss=0.1525, over 28758.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3919, pruned_loss=0.1366, over 5669007.07 frames. ], libri_tot_loss[loss=0.3256, simple_loss=0.3834, pruned_loss=0.1339, over 5699616.63 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3917, pruned_loss=0.1364, over 5654409.07 frames. ], batch size: 262, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:51:28,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 12:51:30,836 INFO [optim.py:369] (1/2) 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 12:51:54,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=362516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:51:54,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5130, 1.6496, 1.3683, 1.6717], device='cuda:1'), covar=tensor([0.2144, 0.2151, 0.2292, 0.1831], device='cuda:1'), in_proj_covar=tensor([0.1218, 0.0921, 0.1078, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 12:51:56,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=362519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:51:57,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8920, 3.6991, 3.4951, 1.6128], device='cuda:1'), covar=tensor([0.0604, 0.0812, 0.0778, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.0986, 0.0930, 0.0822, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 12:52:04,457 INFO [train.py:968] (1/2) Epoch 8, batch 43750, giga_loss[loss=0.3206, simple_loss=0.379, pruned_loss=0.1311, over 28779.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3922, pruned_loss=0.1377, over 5670245.31 frames. ], libri_tot_loss[loss=0.3253, simple_loss=0.3831, pruned_loss=0.1337, over 5700499.65 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3924, pruned_loss=0.1377, over 5657644.84 frames. ], batch size: 186, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:52:20,520 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 43800, giga_loss[loss=0.3094, simple_loss=0.3814, pruned_loss=0.1187, over 29070.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3914, pruned_loss=0.138, over 5660831.84 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3835, pruned_loss=0.134, over 5697037.52 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3915, pruned_loss=0.1379, over 5653108.50 frames. ], batch size: 155, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:52:49,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=362579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:53:00,579 INFO [optim.py:369] (1/2) 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 12:53:08,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5786, 3.7838, 1.6719, 1.6033], device='cuda:1'), covar=tensor([0.0914, 0.0260, 0.0806, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0500, 0.0325, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 12:53:35,202 INFO [train.py:968] (1/2) Epoch 8, batch 43850, giga_loss[loss=0.3371, simple_loss=0.3936, pruned_loss=0.1403, over 28714.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3884, pruned_loss=0.1366, over 5662681.98 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3837, pruned_loss=0.134, over 5699906.29 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3884, pruned_loss=0.1365, over 5653753.97 frames. ], batch size: 284, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:54:06,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-04 12:54:09,840 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=362669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:54:22,138 INFO [train.py:968] (1/2) Epoch 8, batch 43900, giga_loss[loss=0.405, simple_loss=0.4324, pruned_loss=0.1888, over 27541.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3858, pruned_loss=0.1351, over 5672313.55 frames. ], libri_tot_loss[loss=0.3258, simple_loss=0.3837, pruned_loss=0.134, over 5702751.46 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3859, pruned_loss=0.1351, over 5662430.95 frames. ], batch size: 472, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:54:33,211 INFO [optim.py:369] (1/2) 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 12:54:54,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3556, 2.8355, 1.4833, 1.3990], device='cuda:1'), covar=tensor([0.0806, 0.0309, 0.0760, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0501, 0.0326, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 12:55:05,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=362722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:55:07,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=362725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:55:11,549 INFO [train.py:968] (1/2) Epoch 8, batch 43950, giga_loss[loss=0.3262, simple_loss=0.3871, pruned_loss=0.1327, over 28931.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3867, pruned_loss=0.1362, over 5675244.67 frames. ], libri_tot_loss[loss=0.3266, simple_loss=0.3842, pruned_loss=0.1345, over 5700320.66 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3863, pruned_loss=0.1358, over 5669216.25 frames. ], batch size: 227, lr: 3.93e-03, grad_scale: 4.0
+2023-03-04 12:55:35,945 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-04 12:55:36,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=362754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:55:42,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6760, 3.4945, 3.3258, 1.7561], device='cuda:1'), covar=tensor([0.0663, 0.0781, 0.0759, 0.1995], device='cuda:1'), in_proj_covar=tensor([0.0982, 0.0928, 0.0822, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 12:55:57,411 INFO [train.py:968] (1/2) Epoch 8, batch 44000, giga_loss[loss=0.3123, simple_loss=0.3692, pruned_loss=0.1277, over 29044.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3877, pruned_loss=0.1377, over 5672596.46 frames. ], libri_tot_loss[loss=0.3259, simple_loss=0.3836, pruned_loss=0.1341, over 5703689.68 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.388, pruned_loss=0.1378, over 5663939.47 frames. ], batch size: 128, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:56:10,477 INFO [optim.py:369] (1/2) 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 12:56:26,956 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=362810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:56:28,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=362813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:56:44,094 INFO [train.py:968] (1/2) Epoch 8, batch 44050, giga_loss[loss=0.3196, simple_loss=0.3814, pruned_loss=0.1289, over 28279.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3846, pruned_loss=0.136, over 5671912.15 frames. ], libri_tot_loss[loss=0.326, simple_loss=0.3838, pruned_loss=0.1341, over 5704718.23 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3847, pruned_loss=0.136, over 5664203.44 frames. ], batch size: 368, lr: 3.93e-03, grad_scale: 8.0
+2023-03-04 12:56:56,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=362842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 12:57:27,711 INFO [train.py:968] (1/2) Epoch 8, batch 44100, giga_loss[loss=0.3027, simple_loss=0.3634, pruned_loss=0.121, over 28590.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3836, pruned_loss=0.1353, over 5677381.80 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3835, pruned_loss=0.1339, over 5704903.95 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.384, pruned_loss=0.1356, over 5670499.24 frames. ], batch size: 85, lr: 3.92e-03, grad_scale: 8.0
+2023-03-04 12:57:40,567 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 44150, giga_loss[loss=0.3562, simple_loss=0.4103, pruned_loss=0.1511, over 28901.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3852, pruned_loss=0.1357, over 5667733.94 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3835, pruned_loss=0.134, over 5705691.43 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3855, pruned_loss=0.1358, over 5661568.65 frames. ], batch size: 145, lr: 3.92e-03, grad_scale: 8.0
+2023-03-04 12:58:32,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-04 12:59:09,486 INFO [train.py:968] (1/2) Epoch 8, batch 44200, giga_loss[loss=0.2916, simple_loss=0.3574, pruned_loss=0.1129, over 28803.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3858, pruned_loss=0.1357, over 5674332.73 frames. ], libri_tot_loss[loss=0.3257, simple_loss=0.3835, pruned_loss=0.134, over 5705691.43 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3861, pruned_loss=0.1358, over 5669534.21 frames. ], batch size: 119, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 12:59:15,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2558, 3.0756, 2.8957, 1.3974], device='cuda:1'), covar=tensor([0.0856, 0.0961, 0.0940, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.0986, 0.0932, 0.0826, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 12:59:21,848 INFO [optim.py:369] (1/2) 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 12:59:22,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3930, 1.5580, 1.5705, 1.4260], device='cuda:1'), covar=tensor([0.1205, 0.1328, 0.1628, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0736, 0.0656, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0009], device='cuda:1')
+2023-03-04 12:59:31,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3589, 1.5529, 1.2575, 1.3871], device='cuda:1'), covar=tensor([0.0691, 0.0291, 0.0301, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0053, 0.0047, 0.0080], device='cuda:1')
+2023-03-04 12:59:53,529 INFO [train.py:968] (1/2) Epoch 8, batch 44250, giga_loss[loss=0.3493, simple_loss=0.3809, pruned_loss=0.1589, over 23811.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3854, pruned_loss=0.1358, over 5667926.04 frames. ], libri_tot_loss[loss=0.3252, simple_loss=0.3832, pruned_loss=0.1336, over 5710723.72 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.386, pruned_loss=0.1363, over 5658141.98 frames. ], batch size: 705, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:00:05,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:00:34,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 13:00:36,542 INFO [train.py:968] (1/2) Epoch 8, batch 44300, giga_loss[loss=0.3701, simple_loss=0.427, pruned_loss=0.1566, over 28748.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3865, pruned_loss=0.1341, over 5666267.41 frames. ], libri_tot_loss[loss=0.3251, simple_loss=0.383, pruned_loss=0.1336, over 5704901.44 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3871, pruned_loss=0.1345, over 5662602.75 frames. ], batch size: 119, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:00:40,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3768, 2.0464, 1.4778, 0.6541], device='cuda:1'), covar=tensor([0.3154, 0.1855, 0.2573, 0.3993], device='cuda:1'), in_proj_covar=tensor([0.1501, 0.1444, 0.1449, 0.1222], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 13:00:48,954 INFO [optim.py:369] (1/2) 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 13:01:19,078 INFO [train.py:968] (1/2) Epoch 8, batch 44350, giga_loss[loss=0.3657, simple_loss=0.421, pruned_loss=0.1552, over 28736.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3872, pruned_loss=0.1318, over 5685940.07 frames. ], libri_tot_loss[loss=0.325, simple_loss=0.383, pruned_loss=0.1335, over 5708255.40 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3877, pruned_loss=0.1322, over 5679121.41 frames. ], batch size: 284, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:02:05,679 INFO [train.py:968] (1/2) Epoch 8, batch 44400, giga_loss[loss=0.4053, simple_loss=0.4528, pruned_loss=0.1789, over 28796.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3903, pruned_loss=0.1337, over 5677430.25 frames. ], libri_tot_loss[loss=0.3246, simple_loss=0.3828, pruned_loss=0.1332, over 5699712.38 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3912, pruned_loss=0.1343, over 5678976.85 frames. ], batch size: 119, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:02:14,046 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=363187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:02:15,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=363190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:02:18,076 INFO [optim.py:369] (1/2) 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 13:02:44,068 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=363219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:02:53,136 INFO [train.py:968] (1/2) Epoch 8, batch 44450, giga_loss[loss=0.3186, simple_loss=0.3885, pruned_loss=0.1243, over 28409.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3931, pruned_loss=0.1366, over 5670624.94 frames. ], libri_tot_loss[loss=0.3244, simple_loss=0.3826, pruned_loss=0.1331, over 5694674.93 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3942, pruned_loss=0.1372, over 5675680.36 frames. ], batch size: 71, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:03:13,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-04 13:03:42,005 INFO [train.py:968] (1/2) Epoch 8, batch 44500, giga_loss[loss=0.3275, simple_loss=0.3903, pruned_loss=0.1323, over 28994.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3956, pruned_loss=0.1406, over 5655103.30 frames. ], libri_tot_loss[loss=0.3239, simple_loss=0.3821, pruned_loss=0.1329, over 5699948.32 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3973, pruned_loss=0.1415, over 5653120.38 frames. ], batch size: 164, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:03:53,324 INFO [optim.py:369] (1/2) 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 13:04:22,011 INFO [train.py:968] (1/2) Epoch 8, batch 44550, giga_loss[loss=0.2867, simple_loss=0.3649, pruned_loss=0.1042, over 28869.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.395, pruned_loss=0.1408, over 5654715.82 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3819, pruned_loss=0.1327, over 5699057.57 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3972, pruned_loss=0.1421, over 5652157.11 frames. ], batch size: 174, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:04:46,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1414, 1.5522, 1.4832, 1.1320], device='cuda:1'), covar=tensor([0.1046, 0.1568, 0.0889, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0710, 0.0830, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 13:05:04,122 INFO [train.py:968] (1/2) Epoch 8, batch 44600, giga_loss[loss=0.3383, simple_loss=0.3902, pruned_loss=0.1432, over 27527.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3936, pruned_loss=0.1395, over 5660042.44 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3813, pruned_loss=0.1323, over 5701782.34 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3962, pruned_loss=0.1411, over 5654395.25 frames. ], batch size: 472, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:05:17,484 INFO [optim.py:369] (1/2) 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 13:05:47,763 INFO [train.py:968] (1/2) Epoch 8, batch 44650, giga_loss[loss=0.4012, simple_loss=0.4273, pruned_loss=0.1875, over 26643.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3913, pruned_loss=0.1358, over 5658142.91 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3812, pruned_loss=0.1323, over 5696783.38 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3937, pruned_loss=0.1371, over 5657233.19 frames. ], batch size: 555, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:06:05,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:06:14,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:06:31,468 INFO [train.py:968] (1/2) Epoch 8, batch 44700, giga_loss[loss=0.3096, simple_loss=0.3907, pruned_loss=0.1142, over 28556.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3919, pruned_loss=0.1349, over 5654920.87 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3813, pruned_loss=0.1325, over 5691581.57 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.394, pruned_loss=0.1358, over 5658875.91 frames. ], batch size: 60, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:06:43,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8916, 2.6041, 1.6103, 1.0903], device='cuda:1'), covar=tensor([0.3699, 0.2032, 0.2617, 0.3724], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1453, 0.1459, 0.1224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 13:06:44,221 INFO [optim.py:369] (1/2) 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 13:06:55,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-04 13:07:12,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8176, 1.6418, 1.2357, 1.3626], device='cuda:1'), covar=tensor([0.0639, 0.0593, 0.0922, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0445, 0.0495, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:07:18,868 INFO [train.py:968] (1/2) Epoch 8, batch 44750, giga_loss[loss=0.3452, simple_loss=0.4061, pruned_loss=0.1421, over 28704.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3925, pruned_loss=0.1351, over 5661324.90 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3816, pruned_loss=0.1327, over 5690725.78 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3939, pruned_loss=0.1358, over 5664978.52 frames. ], batch size: 242, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:07:30,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:07:31,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 13:07:47,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363556.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:08:04,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1314, 1.0364, 3.8962, 3.1614], device='cuda:1'), covar=tensor([0.1642, 0.2595, 0.0384, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0573, 0.0823, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:08:08,181 INFO [train.py:968] (1/2) Epoch 8, batch 44800, giga_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1253, over 28050.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3923, pruned_loss=0.1358, over 5669397.41 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3816, pruned_loss=0.1326, over 5691797.15 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3935, pruned_loss=0.1363, over 5671217.41 frames. ], batch size: 412, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:08:21,629 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 44850, giga_loss[loss=0.3883, simple_loss=0.4251, pruned_loss=0.1758, over 28802.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3914, pruned_loss=0.1367, over 5660900.76 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3816, pruned_loss=0.1325, over 5695360.49 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3926, pruned_loss=0.1373, over 5658764.85 frames. ], batch size: 262, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:09:29,994 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 13:09:40,387 INFO [train.py:968] (1/2) Epoch 8, batch 44900, giga_loss[loss=0.3592, simple_loss=0.4095, pruned_loss=0.1545, over 28633.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3904, pruned_loss=0.1371, over 5652212.42 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3817, pruned_loss=0.1325, over 5690114.73 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3914, pruned_loss=0.1376, over 5654303.99 frames. ], batch size: 242, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:09:55,271 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 8, batch 44950, giga_loss[loss=0.2971, simple_loss=0.3664, pruned_loss=0.1138, over 29035.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3869, pruned_loss=0.1352, over 5649663.22 frames. ], libri_tot_loss[loss=0.3234, simple_loss=0.3817, pruned_loss=0.1325, over 5688658.83 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3878, pruned_loss=0.1357, over 5651878.19 frames. ], batch size: 164, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:11:00,542 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=363768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:11:08,977 INFO [train.py:968] (1/2) Epoch 8, batch 45000, giga_loss[loss=0.3832, simple_loss=0.4108, pruned_loss=0.1778, over 27609.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3855, pruned_loss=0.135, over 5643748.32 frames. ], libri_tot_loss[loss=0.3238, simple_loss=0.3821, pruned_loss=0.1328, over 5677903.47 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3861, pruned_loss=0.1353, over 5654249.32 frames. ], batch size: 472, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:11:08,977 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 13:11:17,316 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 13:11:30,936 INFO [optim.py:369] (1/2) 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 13:11:35,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-04 13:11:51,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-04 13:11:53,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:11:59,690 INFO [train.py:968] (1/2) Epoch 8, batch 45050, giga_loss[loss=0.3427, simple_loss=0.3797, pruned_loss=0.1528, over 23477.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3859, pruned_loss=0.1354, over 5638296.10 frames. ], libri_tot_loss[loss=0.324, simple_loss=0.3822, pruned_loss=0.1329, over 5663153.34 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3864, pruned_loss=0.1356, over 5658468.25 frames. ], batch size: 705, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:12:05,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:12:24,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0232, 1.8833, 1.3952, 1.6654], device='cuda:1'), covar=tensor([0.0639, 0.0543, 0.0887, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0440, 0.0492, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:12:29,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3754, 1.5962, 1.3267, 1.7151], device='cuda:1'), covar=tensor([0.2310, 0.2378, 0.2476, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.1228, 0.0925, 0.1093, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 13:12:40,914 INFO [train.py:968] (1/2) Epoch 8, batch 45100, giga_loss[loss=0.3698, simple_loss=0.4106, pruned_loss=0.1645, over 26651.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3834, pruned_loss=0.1329, over 5652313.49 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3816, pruned_loss=0.1324, over 5670138.40 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3844, pruned_loss=0.1335, over 5661492.48 frames. ], batch size: 555, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:12:52,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3163, 1.3056, 1.1753, 1.4314], device='cuda:1'), covar=tensor([0.0744, 0.0343, 0.0333, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0118, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 13:13:00,013 INFO [optim.py:369] (1/2) 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 13:13:16,070 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 45150, giga_loss[loss=0.3103, simple_loss=0.3819, pruned_loss=0.1194, over 29001.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3798, pruned_loss=0.129, over 5654916.70 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3816, pruned_loss=0.1324, over 5670415.72 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3806, pruned_loss=0.1295, over 5661468.87 frames. ], batch size: 164, lr: 3.92e-03, grad_scale: 2.0
+2023-03-04 13:13:29,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=363931.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:14:03,410 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=363967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:06,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=363970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:15,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=363978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:15,978 INFO [train.py:968] (1/2) Epoch 8, batch 45200, giga_loss[loss=0.3078, simple_loss=0.3733, pruned_loss=0.1211, over 28650.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3793, pruned_loss=0.1281, over 5649258.91 frames. ], libri_tot_loss[loss=0.3235, simple_loss=0.3819, pruned_loss=0.1326, over 5663605.16 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3796, pruned_loss=0.1284, over 5659487.36 frames. ], batch size: 71, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:14:18,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=363981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:22,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2605, 2.0554, 1.6926, 1.5229], device='cuda:1'), covar=tensor([0.0669, 0.0221, 0.0238, 0.0748], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 13:14:26,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1967, 1.2705, 4.0875, 3.0777], device='cuda:1'), covar=tensor([0.1628, 0.2432, 0.0428, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0624, 0.0574, 0.0826, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:14:30,287 INFO [optim.py:369] (1/2) 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 13:14:32,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=363999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:14:43,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4634, 2.6892, 1.6033, 1.5529], device='cuda:1'), covar=tensor([0.0671, 0.0262, 0.0590, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0507, 0.0330, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 13:14:44,360 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 8, batch 45250, giga_loss[loss=0.2742, simple_loss=0.3439, pruned_loss=0.1022, over 28887.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3776, pruned_loss=0.128, over 5646399.52 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3821, pruned_loss=0.1327, over 5669681.35 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3777, pruned_loss=0.128, over 5648624.78 frames. ], batch size: 174, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:15:03,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3277, 2.9783, 1.4879, 1.3236], device='cuda:1'), covar=tensor([0.0889, 0.0297, 0.0788, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0505, 0.0330, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 13:15:30,933 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=364058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:15:34,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=364061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:15:44,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=364074.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:15:46,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=364077.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:15:48,426 INFO [train.py:968] (1/2) Epoch 8, batch 45300, giga_loss[loss=0.3809, simple_loss=0.4106, pruned_loss=0.1756, over 26656.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3751, pruned_loss=0.1274, over 5613754.39 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3821, pruned_loss=0.1327, over 5644085.58 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3749, pruned_loss=0.1272, over 5639264.09 frames. ], batch size: 555, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:15:57,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=364090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:16:02,021 INFO [optim.py:369] (1/2) 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 13:16:03,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-04 13:16:10,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8564, 1.1664, 3.4303, 2.9093], device='cuda:1'), covar=tensor([0.1786, 0.2486, 0.0488, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0626, 0.0573, 0.0828, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:16:11,199 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=364106.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:16:18,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1444, 1.2769, 3.4043, 2.9423], device='cuda:1'), covar=tensor([0.1499, 0.2377, 0.0431, 0.1436], device='cuda:1'), in_proj_covar=tensor([0.0625, 0.0574, 0.0827, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:16:32,601 INFO [train.py:968] (1/2) Epoch 8, batch 45350, giga_loss[loss=0.2951, simple_loss=0.3712, pruned_loss=0.1095, over 28947.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3772, pruned_loss=0.1287, over 5584678.51 frames. ], libri_tot_loss[loss=0.3243, simple_loss=0.3824, pruned_loss=0.1331, over 5609121.25 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3767, pruned_loss=0.1282, over 5635975.17 frames. ], batch size: 213, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:16:42,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=364143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:17:03,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0492, 2.5967, 1.9808, 1.4931], device='cuda:1'), covar=tensor([0.2578, 0.1862, 0.1837, 0.2708], device='cuda:1'), in_proj_covar=tensor([0.1511, 0.1443, 0.1457, 0.1216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 13:17:10,956 INFO [train.py:968] (1/2) Epoch 8, batch 45400, giga_loss[loss=0.3337, simple_loss=0.3919, pruned_loss=0.1377, over 28888.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3804, pruned_loss=0.1305, over 5523954.10 frames. ], libri_tot_loss[loss=0.3264, simple_loss=0.384, pruned_loss=0.1344, over 5525958.35 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3784, pruned_loss=0.1287, over 5641850.83 frames. ], batch size: 213, lr: 3.92e-03, grad_scale: 4.0
+2023-03-04 13:17:25,347 INFO [optim.py:369] (1/2) 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 13:17:28,555 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-04 13:18:40,353 INFO [train.py:968] (1/2) Epoch 9, batch 50, giga_loss[loss=0.3016, simple_loss=0.3808, pruned_loss=0.1112, over 28692.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3851, pruned_loss=0.1189, over 1267680.83 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3764, pruned_loss=0.1147, over 201562.16 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3865, pruned_loss=0.1196, over 1104715.67 frames. ], batch size: 262, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:18:48,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3109, 1.4637, 1.4374, 1.3122], device='cuda:1'), covar=tensor([0.1344, 0.1427, 0.1839, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0733, 0.0650, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 13:19:17,982 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=364286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:19:19,843 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 9, batch 100, giga_loss[loss=0.2491, simple_loss=0.3296, pruned_loss=0.0843, over 28870.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3793, pruned_loss=0.1177, over 2233017.79 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3721, pruned_loss=0.1112, over 304367.62 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3804, pruned_loss=0.1186, over 2039002.44 frames. ], batch size: 213, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:19:43,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=364318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:20:05,404 INFO [train.py:968] (1/2) Epoch 9, batch 150, giga_loss[loss=0.2912, simple_loss=0.3505, pruned_loss=0.116, over 27681.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.364, pruned_loss=0.1102, over 3011439.55 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3647, pruned_loss=0.109, over 577173.87 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3646, pruned_loss=0.1107, over 2705641.19 frames. ], batch size: 472, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:20:16,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4986, 2.0274, 1.5546, 1.3515], device='cuda:1'), covar=tensor([0.2369, 0.1436, 0.1497, 0.1879], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1476, 0.1442, 0.1558], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 13:20:43,299 INFO [optim.py:369] (1/2) 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 13:20:44,628 INFO [train.py:968] (1/2) Epoch 9, batch 200, libri_loss[loss=0.2897, simple_loss=0.3676, pruned_loss=0.106, over 29355.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3492, pruned_loss=0.1023, over 3605828.64 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3593, pruned_loss=0.1044, over 730178.58 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3489, pruned_loss=0.1027, over 3295702.95 frames. ], batch size: 92, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:20:47,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9591, 2.7988, 1.9192, 0.8812], device='cuda:1'), covar=tensor([0.4703, 0.2132, 0.2524, 0.4475], device='cuda:1'), in_proj_covar=tensor([0.1484, 0.1419, 0.1439, 0.1198], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 13:21:20,250 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-04 13:21:21,240 INFO [train.py:968] (1/2) Epoch 9, batch 250, giga_loss[loss=0.2358, simple_loss=0.3106, pruned_loss=0.08046, over 28755.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3387, pruned_loss=0.09685, over 4076620.78 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3558, pruned_loss=0.1026, over 930215.16 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3376, pruned_loss=0.09684, over 3754088.15 frames. ], batch size: 60, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:21:30,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-04 13:22:01,664 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:22:04,676 INFO [train.py:968] (1/2) Epoch 9, batch 300, giga_loss[loss=0.2435, simple_loss=0.3029, pruned_loss=0.09205, over 27575.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3285, pruned_loss=0.09216, over 4429707.94 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3551, pruned_loss=0.1027, over 1004325.60 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3268, pruned_loss=0.09168, over 4152977.12 frames. ], batch size: 472, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:22:14,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1279, 2.1938, 2.0495, 1.9524], device='cuda:1'), covar=tensor([0.1458, 0.2167, 0.1737, 0.1949], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0728, 0.0649, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 13:22:50,644 INFO [train.py:968] (1/2) Epoch 9, batch 350, giga_loss[loss=0.2169, simple_loss=0.273, pruned_loss=0.08036, over 23972.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3212, pruned_loss=0.08865, over 4704304.92 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3558, pruned_loss=0.103, over 1125802.12 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3184, pruned_loss=0.08764, over 4453534.61 frames. ], batch size: 705, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:22:56,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=364556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:23:27,095 INFO [optim.py:369] (1/2) 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 13:23:28,327 INFO [train.py:968] (1/2) Epoch 9, batch 400, giga_loss[loss=0.2494, simple_loss=0.3194, pruned_loss=0.08966, over 28640.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3156, pruned_loss=0.08555, over 4928879.84 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3552, pruned_loss=0.1023, over 1196858.35 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3128, pruned_loss=0.08458, over 4716621.80 frames. ], batch size: 307, lr: 3.71e-03, grad_scale: 8.0
+2023-03-04 13:23:31,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-04 13:24:09,309 INFO [train.py:968] (1/2) Epoch 9, batch 450, giga_loss[loss=0.2121, simple_loss=0.2869, pruned_loss=0.06865, over 28847.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3142, pruned_loss=0.08504, over 5104520.31 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3568, pruned_loss=0.1028, over 1313503.10 frames. ], giga_tot_loss[loss=0.2389, simple_loss=0.3105, pruned_loss=0.08366, over 4915080.49 frames. ], batch size: 112, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:24:09,750 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 13:24:49,154 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 500, giga_loss[loss=0.1998, simple_loss=0.2746, pruned_loss=0.06247, over 28830.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3131, pruned_loss=0.08492, over 5230428.08 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3586, pruned_loss=0.1039, over 1424305.00 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3087, pruned_loss=0.08313, over 5065350.71 frames. ], batch size: 92, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:25:17,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 13:25:32,101 INFO [train.py:968] (1/2) Epoch 9, batch 550, giga_loss[loss=0.2073, simple_loss=0.2847, pruned_loss=0.06493, over 28868.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3118, pruned_loss=0.08446, over 5326419.06 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3595, pruned_loss=0.1042, over 1525779.68 frames. ], giga_tot_loss[loss=0.236, simple_loss=0.3069, pruned_loss=0.0825, over 5189787.39 frames. ], batch size: 285, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:25:49,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 13:26:12,130 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 600, libri_loss[loss=0.2459, simple_loss=0.3268, pruned_loss=0.08254, over 28551.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3093, pruned_loss=0.08336, over 5407231.15 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3581, pruned_loss=0.1032, over 1633043.33 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3046, pruned_loss=0.0816, over 5287222.04 frames. ], batch size: 63, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:26:44,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2452, 0.9341, 0.9631, 1.4195], device='cuda:1'), covar=tensor([0.0776, 0.0367, 0.0345, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 13:26:58,535 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-04 13:26:58,606 INFO [train.py:968] (1/2) Epoch 9, batch 650, giga_loss[loss=0.209, simple_loss=0.2846, pruned_loss=0.06668, over 28771.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3064, pruned_loss=0.08192, over 5473944.97 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3584, pruned_loss=0.103, over 1675342.60 frames. ], giga_tot_loss[loss=0.2315, simple_loss=0.3022, pruned_loss=0.08039, over 5375773.98 frames. ], batch size: 119, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:27:17,133 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:27:39,760 INFO [train.py:968] (1/2) Epoch 9, batch 700, giga_loss[loss=0.213, simple_loss=0.2834, pruned_loss=0.07125, over 28936.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3037, pruned_loss=0.08058, over 5524032.09 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3585, pruned_loss=0.1029, over 1736948.50 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.2996, pruned_loss=0.07908, over 5442481.57 frames. ], batch size: 213, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:28:04,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=364931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:28:17,370 INFO [train.py:968] (1/2) Epoch 9, batch 750, libri_loss[loss=0.2928, simple_loss=0.3732, pruned_loss=0.1062, over 29116.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3025, pruned_loss=0.07982, over 5559721.62 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3573, pruned_loss=0.1022, over 1899164.28 frames. ], giga_tot_loss[loss=0.2266, simple_loss=0.2973, pruned_loss=0.07795, over 5481778.71 frames. ], batch size: 101, lr: 3.71e-03, grad_scale: 4.0
+2023-03-04 13:28:52,390 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:28:56,732 INFO [train.py:968] (1/2) Epoch 9, batch 800, giga_loss[loss=0.2225, simple_loss=0.2961, pruned_loss=0.07443, over 27946.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3005, pruned_loss=0.07884, over 5587800.77 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3561, pruned_loss=0.1013, over 2049993.40 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.2948, pruned_loss=0.07678, over 5518877.40 frames. ], batch size: 412, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:29:08,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9383, 1.8877, 1.4065, 1.6386], device='cuda:1'), covar=tensor([0.0616, 0.0504, 0.0854, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0443, 0.0496, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:29:08,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:29:10,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:29:17,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4626, 1.7503, 1.7582, 1.3529], device='cuda:1'), covar=tensor([0.1360, 0.1919, 0.1135, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0715, 0.0850, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-04 13:29:39,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:29:41,884 INFO [train.py:968] (1/2) Epoch 9, batch 850, libri_loss[loss=0.2763, simple_loss=0.3381, pruned_loss=0.1073, over 29360.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3069, pruned_loss=0.08246, over 5609562.97 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3559, pruned_loss=0.1016, over 2126102.90 frames. ], giga_tot_loss[loss=0.2311, simple_loss=0.3014, pruned_loss=0.08037, over 5549780.38 frames. ], batch size: 71, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:30:04,881 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:30:06,750 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 9, batch 900, libri_loss[loss=0.3024, simple_loss=0.381, pruned_loss=0.1119, over 29286.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3205, pruned_loss=0.08935, over 5630573.09 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3562, pruned_loss=0.1018, over 2182328.85 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3157, pruned_loss=0.08749, over 5579499.35 frames. ], batch size: 94, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:30:31,327 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:31:05,200 INFO [train.py:968] (1/2) Epoch 9, batch 950, giga_loss[loss=0.3191, simple_loss=0.3935, pruned_loss=0.1224, over 28725.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3333, pruned_loss=0.09669, over 5642551.43 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3559, pruned_loss=0.1017, over 2274663.29 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.329, pruned_loss=0.0951, over 5595771.99 frames. ], batch size: 262, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:31:18,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8689, 1.7159, 1.2586, 1.5534], device='cuda:1'), covar=tensor([0.0700, 0.0675, 0.1058, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0439, 0.0493, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:31:45,344 INFO [optim.py:369] (1/2) 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 13:31:46,059 INFO [train.py:968] (1/2) Epoch 9, batch 1000, giga_loss[loss=0.2624, simple_loss=0.3525, pruned_loss=0.08617, over 28533.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.341, pruned_loss=0.09975, over 5658606.52 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3548, pruned_loss=0.1012, over 2347190.02 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3377, pruned_loss=0.09864, over 5616353.46 frames. ], batch size: 60, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:32:08,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2789, 3.0654, 2.8901, 1.4392], device='cuda:1'), covar=tensor([0.0774, 0.0893, 0.0878, 0.2303], device='cuda:1'), in_proj_covar=tensor([0.0960, 0.0902, 0.0802, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 13:32:23,895 INFO [train.py:968] (1/2) Epoch 9, batch 1050, giga_loss[loss=0.2674, simple_loss=0.3405, pruned_loss=0.09709, over 28673.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3457, pruned_loss=0.1007, over 5670055.11 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3551, pruned_loss=0.1014, over 2399152.44 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3429, pruned_loss=0.09973, over 5634425.85 frames. ], batch size: 92, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:32:30,158 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:33:09,286 INFO [optim.py:369] (1/2) 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 13:33:10,462 INFO [train.py:968] (1/2) Epoch 9, batch 1100, giga_loss[loss=0.2716, simple_loss=0.3418, pruned_loss=0.1007, over 28592.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3469, pruned_loss=0.1003, over 5666817.16 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3547, pruned_loss=0.1012, over 2467155.09 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3447, pruned_loss=0.09967, over 5635508.18 frames. ], batch size: 78, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:33:19,084 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:33:49,355 INFO [train.py:968] (1/2) Epoch 9, batch 1150, giga_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09314, over 28721.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3505, pruned_loss=0.1028, over 5670239.42 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3555, pruned_loss=0.1015, over 2584861.66 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3482, pruned_loss=0.1022, over 5638363.07 frames. ], batch size: 284, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:33:49,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6116, 1.6858, 1.4777, 2.2042], device='cuda:1'), covar=tensor([0.2125, 0.2148, 0.2132, 0.1896], device='cuda:1'), in_proj_covar=tensor([0.1229, 0.0923, 0.1088, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 13:34:00,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:34:04,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:34:30,479 INFO [optim.py:369] (1/2) 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 13:34:31,584 INFO [train.py:968] (1/2) Epoch 9, batch 1200, giga_loss[loss=0.3848, simple_loss=0.4187, pruned_loss=0.1755, over 26564.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3525, pruned_loss=0.1045, over 5679239.94 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3549, pruned_loss=0.1011, over 2650703.51 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3508, pruned_loss=0.1042, over 5649873.89 frames. ], batch size: 555, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:34:45,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 13:35:12,909 INFO [train.py:968] (1/2) Epoch 9, batch 1250, giga_loss[loss=0.2981, simple_loss=0.3668, pruned_loss=0.1146, over 28807.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3559, pruned_loss=0.1069, over 5685907.86 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.355, pruned_loss=0.1009, over 2699204.70 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3546, pruned_loss=0.1068, over 5659576.87 frames. ], batch size: 99, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:35:53,219 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 1300, giga_loss[loss=0.2869, simple_loss=0.3694, pruned_loss=0.1022, over 28616.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3571, pruned_loss=0.1062, over 5696674.97 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3531, pruned_loss=0.1, over 2839543.46 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.357, pruned_loss=0.1068, over 5667535.27 frames. ], batch size: 307, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:36:02,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:36:04,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:36:27,189 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:36:29,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8055, 1.8159, 1.7394, 1.6187], device='cuda:1'), covar=tensor([0.1331, 0.1819, 0.1830, 0.1648], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0728, 0.0654, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 13:36:30,856 INFO [train.py:968] (1/2) Epoch 9, batch 1350, giga_loss[loss=0.3222, simple_loss=0.3882, pruned_loss=0.1281, over 28246.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3596, pruned_loss=0.1072, over 5696393.98 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.353, pruned_loss=0.09989, over 2914055.37 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3597, pruned_loss=0.1079, over 5669621.45 frames. ], batch size: 368, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:37:05,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5866, 1.7758, 1.7818, 1.4102], device='cuda:1'), covar=tensor([0.1775, 0.2215, 0.1367, 0.1590], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0706, 0.0837, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 13:37:11,670 INFO [optim.py:369] (1/2) 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 13:37:11,682 INFO [train.py:968] (1/2) Epoch 9, batch 1400, giga_loss[loss=0.3093, simple_loss=0.3799, pruned_loss=0.1193, over 28842.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3613, pruned_loss=0.1074, over 5704531.46 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3534, pruned_loss=0.1, over 3059371.47 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3615, pruned_loss=0.1082, over 5674167.89 frames. ], batch size: 119, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:37:23,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9263, 2.7786, 2.6534, 2.6240], device='cuda:1'), covar=tensor([0.1065, 0.1752, 0.1384, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0721, 0.0647, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 13:37:25,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-04 13:37:34,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:37:50,565 INFO [train.py:968] (1/2) Epoch 9, batch 1450, giga_loss[loss=0.2688, simple_loss=0.36, pruned_loss=0.08877, over 28701.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3604, pruned_loss=0.1054, over 5710799.90 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3527, pruned_loss=0.09951, over 3130005.01 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.361, pruned_loss=0.1065, over 5682189.68 frames. ], batch size: 60, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:37:56,415 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 13:38:11,007 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=365672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:38:20,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:38:30,590 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 1500, giga_loss[loss=0.2823, simple_loss=0.3585, pruned_loss=0.1031, over 28889.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3599, pruned_loss=0.1042, over 5712018.75 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3529, pruned_loss=0.09966, over 3171295.25 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3605, pruned_loss=0.1051, over 5687084.32 frames. ], batch size: 186, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:38:49,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6732, 4.3417, 1.7778, 1.8122], device='cuda:1'), covar=tensor([0.0908, 0.0162, 0.0820, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0485, 0.0320, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 13:39:00,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=365736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:39:09,511 INFO [train.py:968] (1/2) Epoch 9, batch 1550, giga_loss[loss=0.3009, simple_loss=0.3789, pruned_loss=0.1115, over 28585.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3584, pruned_loss=0.1024, over 5721463.15 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3536, pruned_loss=0.09991, over 3226008.56 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3585, pruned_loss=0.103, over 5697885.57 frames. ], batch size: 307, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:39:30,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:39:32,549 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:39:51,876 INFO [train.py:968] (1/2) Epoch 9, batch 1600, libri_loss[loss=0.2444, simple_loss=0.3251, pruned_loss=0.08181, over 29544.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3604, pruned_loss=0.1055, over 5706227.63 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3535, pruned_loss=0.1001, over 3292305.30 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3607, pruned_loss=0.106, over 5683123.69 frames. ], batch size: 77, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:39:56,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:02,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5396, 1.7276, 1.5458, 1.6333], device='cuda:1'), covar=tensor([0.1206, 0.1413, 0.1753, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0720, 0.0649, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 13:40:14,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:18,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:32,402 INFO [train.py:968] (1/2) Epoch 9, batch 1650, giga_loss[loss=0.2887, simple_loss=0.3613, pruned_loss=0.1081, over 28796.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3622, pruned_loss=0.1086, over 5712458.68 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3537, pruned_loss=0.1, over 3344426.13 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3625, pruned_loss=0.1091, over 5690256.03 frames. ], batch size: 119, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:40:34,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9363, 2.9754, 2.2344, 0.9347], device='cuda:1'), covar=tensor([0.4089, 0.1912, 0.2056, 0.4037], device='cuda:1'), in_proj_covar=tensor([0.1506, 0.1421, 0.1463, 0.1223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 13:40:34,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6483, 2.2136, 2.0082, 1.5501], device='cuda:1'), covar=tensor([0.1669, 0.2028, 0.1388, 0.1533], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0704, 0.0837, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 13:40:42,717 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:40:58,897 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=365879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:41:01,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=365882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:41:14,926 INFO [train.py:968] (1/2) Epoch 9, batch 1700, giga_loss[loss=0.281, simple_loss=0.3543, pruned_loss=0.1039, over 28616.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3637, pruned_loss=0.1112, over 5718100.28 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3541, pruned_loss=0.1001, over 3393279.85 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.364, pruned_loss=0.1118, over 5698797.48 frames. ], batch size: 336, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:41:15,515 INFO [optim.py:369] (1/2) 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 13:41:24,526 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=365911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:41:54,104 INFO [train.py:968] (1/2) Epoch 9, batch 1750, giga_loss[loss=0.2758, simple_loss=0.3466, pruned_loss=0.1025, over 28566.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3622, pruned_loss=0.1113, over 5711840.41 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3535, pruned_loss=0.09961, over 3468193.24 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.363, pruned_loss=0.1124, over 5690648.08 frames. ], batch size: 71, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:42:27,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1238, 0.9723, 3.8710, 3.2419], device='cuda:1'), covar=tensor([0.2082, 0.3023, 0.0695, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0612, 0.0562, 0.0807, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 13:42:33,802 INFO [train.py:968] (1/2) Epoch 9, batch 1800, giga_loss[loss=0.2704, simple_loss=0.3405, pruned_loss=0.1001, over 28764.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3604, pruned_loss=0.1111, over 5702530.23 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3538, pruned_loss=0.09984, over 3550013.74 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3611, pruned_loss=0.1122, over 5681671.60 frames. ], batch size: 99, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:42:34,363 INFO [optim.py:369] (1/2) 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 13:43:14,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:43:14,924 INFO [train.py:968] (1/2) Epoch 9, batch 1850, giga_loss[loss=0.3071, simple_loss=0.3653, pruned_loss=0.1245, over 23692.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.359, pruned_loss=0.1096, over 5694972.80 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3543, pruned_loss=0.09999, over 3570630.47 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3593, pruned_loss=0.1105, over 5679376.24 frames. ], batch size: 705, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:43:55,455 INFO [train.py:968] (1/2) Epoch 9, batch 1900, giga_loss[loss=0.3029, simple_loss=0.3735, pruned_loss=0.1161, over 28688.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3569, pruned_loss=0.1073, over 5700152.43 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3541, pruned_loss=0.0997, over 3638608.84 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3573, pruned_loss=0.1084, over 5682947.29 frames. ], batch size: 99, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:43:56,124 INFO [optim.py:369] (1/2) 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 13:44:09,753 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-04 13:44:28,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366135.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:44:36,851 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 1950, giga_loss[loss=0.2322, simple_loss=0.3144, pruned_loss=0.07497, over 28613.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3518, pruned_loss=0.1038, over 5696065.51 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.354, pruned_loss=0.09956, over 3683506.48 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3523, pruned_loss=0.1049, over 5678935.21 frames. ], batch size: 92, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:45:17,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:45:19,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:45:23,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-04 13:45:23,751 INFO [train.py:968] (1/2) Epoch 9, batch 2000, giga_loss[loss=0.238, simple_loss=0.3144, pruned_loss=0.0808, over 28748.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3451, pruned_loss=0.09997, over 5687450.82 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3539, pruned_loss=0.09944, over 3725381.61 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3454, pruned_loss=0.1009, over 5672215.13 frames. ], batch size: 262, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:45:24,746 INFO [optim.py:369] (1/2) 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 13:45:46,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:46:10,284 INFO [train.py:968] (1/2) Epoch 9, batch 2050, giga_loss[loss=0.2512, simple_loss=0.3047, pruned_loss=0.09881, over 23363.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3394, pruned_loss=0.0971, over 5679050.72 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3539, pruned_loss=0.09945, over 3736414.70 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3396, pruned_loss=0.09783, over 5665956.09 frames. ], batch size: 705, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:46:19,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1240, 1.1654, 3.7669, 2.9964], device='cuda:1'), covar=tensor([0.1884, 0.2658, 0.0757, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0609, 0.0562, 0.0800, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 13:46:23,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3649, 3.3201, 1.5632, 1.4774], device='cuda:1'), covar=tensor([0.0872, 0.0275, 0.0803, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0485, 0.0322, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 13:46:32,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-04 13:46:51,657 INFO [train.py:968] (1/2) Epoch 9, batch 2100, giga_loss[loss=0.2811, simple_loss=0.3551, pruned_loss=0.1036, over 28918.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3389, pruned_loss=0.09732, over 5670266.47 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.355, pruned_loss=0.1004, over 3800760.57 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3379, pruned_loss=0.09728, over 5653286.60 frames. ], batch size: 145, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:46:52,308 INFO [optim.py:369] (1/2) 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 13:46:58,877 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:47:11,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:47:32,197 INFO [train.py:968] (1/2) Epoch 9, batch 2150, libri_loss[loss=0.2794, simple_loss=0.363, pruned_loss=0.09786, over 29267.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3402, pruned_loss=0.09722, over 5680536.63 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3556, pruned_loss=0.1006, over 3850674.50 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3387, pruned_loss=0.097, over 5663468.15 frames. ], batch size: 94, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:47:37,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-04 13:47:53,792 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:48:09,700 INFO [train.py:968] (1/2) Epoch 9, batch 2200, giga_loss[loss=0.2898, simple_loss=0.3439, pruned_loss=0.1179, over 23946.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3403, pruned_loss=0.09745, over 5689990.08 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3562, pruned_loss=0.1008, over 3880501.54 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3386, pruned_loss=0.09711, over 5674379.72 frames. ], batch size: 705, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:48:10,368 INFO [optim.py:369] (1/2) 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 13:48:48,120 INFO [train.py:968] (1/2) Epoch 9, batch 2250, libri_loss[loss=0.3017, simple_loss=0.3821, pruned_loss=0.1107, over 29342.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3394, pruned_loss=0.0969, over 5698015.45 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3572, pruned_loss=0.1011, over 3946772.52 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3367, pruned_loss=0.09629, over 5681993.21 frames. ], batch size: 92, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:49:02,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=366468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:49:26,048 INFO [train.py:968] (1/2) Epoch 9, batch 2300, giga_loss[loss=0.2446, simple_loss=0.3231, pruned_loss=0.08304, over 29095.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3363, pruned_loss=0.095, over 5709012.10 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3576, pruned_loss=0.101, over 3994596.11 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3335, pruned_loss=0.09443, over 5692298.91 frames. ], batch size: 155, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:49:26,741 INFO [optim.py:369] (1/2) 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 13:49:31,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4139, 3.2994, 1.5589, 1.5747], device='cuda:1'), covar=tensor([0.0890, 0.0284, 0.0811, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0485, 0.0321, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 13:49:34,970 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366510.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 13:49:43,050 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:49:45,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6249, 2.7323, 1.7819, 0.7625], device='cuda:1'), covar=tensor([0.4591, 0.1537, 0.2321, 0.3668], device='cuda:1'), in_proj_covar=tensor([0.1492, 0.1404, 0.1445, 0.1213], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 13:49:54,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2366, 1.4471, 1.0286, 1.1407], device='cuda:1'), covar=tensor([0.1672, 0.1183, 0.1378, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1474, 0.1457, 0.1564], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 13:50:03,455 INFO [train.py:968] (1/2) Epoch 9, batch 2350, giga_loss[loss=0.2366, simple_loss=0.306, pruned_loss=0.08362, over 28914.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3333, pruned_loss=0.09406, over 5706823.35 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3574, pruned_loss=0.1009, over 4003986.13 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3311, pruned_loss=0.09367, over 5692974.01 frames. ], batch size: 106, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:50:39,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6003, 4.6157, 1.8429, 1.6684], device='cuda:1'), covar=tensor([0.0898, 0.0231, 0.0810, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0487, 0.0323, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 13:50:42,963 INFO [train.py:968] (1/2) Epoch 9, batch 2400, giga_loss[loss=0.2604, simple_loss=0.3243, pruned_loss=0.09821, over 28970.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3311, pruned_loss=0.093, over 5706922.71 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3584, pruned_loss=0.1013, over 4041827.04 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3282, pruned_loss=0.09231, over 5692543.00 frames. ], batch size: 106, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:50:43,697 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 2450, giga_loss[loss=0.3383, simple_loss=0.383, pruned_loss=0.1468, over 26828.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3284, pruned_loss=0.0916, over 5710136.25 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3589, pruned_loss=0.1014, over 4074269.55 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3254, pruned_loss=0.09082, over 5699356.69 frames. ], batch size: 555, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:51:23,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:51:25,585 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366656.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:51:30,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:32,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:43,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:46,876 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366685.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 13:51:54,014 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:51:55,915 INFO [train.py:968] (1/2) Epoch 9, batch 2500, giga_loss[loss=0.2366, simple_loss=0.308, pruned_loss=0.08263, over 29115.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3262, pruned_loss=0.09035, over 5719341.49 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3587, pruned_loss=0.1011, over 4135571.30 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3229, pruned_loss=0.08962, over 5706465.33 frames. ], batch size: 128, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:51:56,171 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:52:36,942 INFO [train.py:968] (1/2) Epoch 9, batch 2550, giga_loss[loss=0.2178, simple_loss=0.2972, pruned_loss=0.06923, over 28558.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3254, pruned_loss=0.08981, over 5728364.35 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3593, pruned_loss=0.1013, over 4188053.47 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3214, pruned_loss=0.08881, over 5713094.72 frames. ], batch size: 60, lr: 3.70e-03, grad_scale: 8.0
+2023-03-04 13:52:41,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:16,001 INFO [train.py:968] (1/2) Epoch 9, batch 2600, giga_loss[loss=0.2465, simple_loss=0.3198, pruned_loss=0.08664, over 29064.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3246, pruned_loss=0.08956, over 5728774.57 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3595, pruned_loss=0.1013, over 4222388.15 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3207, pruned_loss=0.08851, over 5713295.59 frames. ], batch size: 136, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:53:17,381 INFO [optim.py:369] (1/2) 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 13:53:26,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7684, 1.0460, 3.2803, 2.6757], device='cuda:1'), covar=tensor([0.1800, 0.2593, 0.0451, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0561, 0.0804, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 13:53:35,441 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:37,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:46,872 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:48,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=366843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:48,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:53:51,405 INFO [train.py:968] (1/2) Epoch 9, batch 2650, giga_loss[loss=0.2549, simple_loss=0.3238, pruned_loss=0.09303, over 28816.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.325, pruned_loss=0.09011, over 5730254.41 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3604, pruned_loss=0.1015, over 4268792.23 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3202, pruned_loss=0.08881, over 5716187.92 frames. ], batch size: 119, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:53:57,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3623, 2.1927, 2.0789, 2.0444], device='cuda:1'), covar=tensor([0.1242, 0.2102, 0.1732, 0.1773], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0730, 0.0657, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 13:53:58,843 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:12,415 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:31,046 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:32,020 INFO [train.py:968] (1/2) Epoch 9, batch 2700, giga_loss[loss=0.2754, simple_loss=0.3429, pruned_loss=0.104, over 29059.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.33, pruned_loss=0.09324, over 5711485.60 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.361, pruned_loss=0.1018, over 4297526.50 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3249, pruned_loss=0.0917, over 5713040.31 frames. ], batch size: 128, lr: 3.70e-03, grad_scale: 4.0
+2023-03-04 13:54:33,575 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:34,067 INFO [optim.py:369] (1/2) 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 13:54:42,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6906, 1.0180, 2.9677, 2.7812], device='cuda:1'), covar=tensor([0.1588, 0.2204, 0.0502, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0614, 0.0558, 0.0799, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 13:54:56,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=366928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:54:58,508 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-04 13:55:11,688 INFO [train.py:968] (1/2) Epoch 9, batch 2750, giga_loss[loss=0.3042, simple_loss=0.3725, pruned_loss=0.1179, over 28594.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3355, pruned_loss=0.09641, over 5713521.85 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3612, pruned_loss=0.1018, over 4328709.31 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3308, pruned_loss=0.09508, over 5711986.98 frames. ], batch size: 307, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 13:55:39,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6039, 1.6815, 1.2250, 1.3548], device='cuda:1'), covar=tensor([0.0690, 0.0549, 0.1003, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0441, 0.0496, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 13:55:45,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=366986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:55:47,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=366989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:55:53,855 INFO [train.py:968] (1/2) Epoch 9, batch 2800, giga_loss[loss=0.3475, simple_loss=0.3999, pruned_loss=0.1475, over 28832.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3436, pruned_loss=0.1016, over 5713681.76 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3614, pruned_loss=0.1018, over 4390163.42 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3388, pruned_loss=0.1005, over 5705866.10 frames. ], batch size: 174, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:55:55,282 INFO [optim.py:369] (1/2) 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 13:56:09,924 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=367018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:56:29,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 13:56:35,275 INFO [train.py:968] (1/2) Epoch 9, batch 2850, giga_loss[loss=0.3021, simple_loss=0.3746, pruned_loss=0.1148, over 29006.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3492, pruned_loss=0.1053, over 5702888.69 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3606, pruned_loss=0.1014, over 4430300.99 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3455, pruned_loss=0.1047, over 5695774.83 frames. ], batch size: 136, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:56:36,591 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-04 13:57:18,096 INFO [train.py:968] (1/2) Epoch 9, batch 2900, giga_loss[loss=0.2911, simple_loss=0.3691, pruned_loss=0.1065, over 28973.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3532, pruned_loss=0.1063, over 5712145.44 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3607, pruned_loss=0.1014, over 4471157.14 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3499, pruned_loss=0.1059, over 5703750.89 frames. ], batch size: 145, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:57:19,405 INFO [optim.py:369] (1/2) 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 13:57:33,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6361, 4.4274, 4.1824, 1.9174], device='cuda:1'), covar=tensor([0.0475, 0.0701, 0.0695, 0.2223], device='cuda:1'), in_proj_covar=tensor([0.0964, 0.0899, 0.0803, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 13:57:44,638 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:57:47,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0853, 1.3573, 1.0581, 0.9691], device='cuda:1'), covar=tensor([0.1294, 0.1176, 0.1019, 0.1292], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1479, 0.1471, 0.1570], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 13:57:54,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5169, 5.2891, 5.0192, 2.3490], device='cuda:1'), covar=tensor([0.0353, 0.0559, 0.0546, 0.2029], device='cuda:1'), in_proj_covar=tensor([0.0964, 0.0901, 0.0804, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 13:57:58,451 INFO [train.py:968] (1/2) Epoch 9, batch 2950, giga_loss[loss=0.2913, simple_loss=0.3696, pruned_loss=0.1065, over 29035.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3585, pruned_loss=0.1091, over 5703366.60 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3606, pruned_loss=0.1016, over 4511241.28 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3558, pruned_loss=0.1089, over 5698770.98 frames. ], batch size: 155, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:58:04,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 13:58:43,848 INFO [train.py:968] (1/2) Epoch 9, batch 3000, giga_loss[loss=0.3105, simple_loss=0.3814, pruned_loss=0.1198, over 28508.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3636, pruned_loss=0.1129, over 5686968.40 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.36, pruned_loss=0.1013, over 4542758.22 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3619, pruned_loss=0.1131, over 5681403.11 frames. ], batch size: 336, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 13:58:43,848 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 13:58:48,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2482, 1.4963, 1.2167, 1.1480], device='cuda:1'), covar=tensor([0.1453, 0.1282, 0.0989, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1473, 0.1466, 0.1568], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 13:58:52,131 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 13:58:53,395 INFO [optim.py:369] (1/2) 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 13:59:31,084 INFO [train.py:968] (1/2) Epoch 9, batch 3050, giga_loss[loss=0.2492, simple_loss=0.3306, pruned_loss=0.08394, over 28288.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3605, pruned_loss=0.1102, over 5692209.67 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3594, pruned_loss=0.101, over 4587663.88 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3597, pruned_loss=0.1111, over 5682651.81 frames. ], batch size: 77, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:00:10,445 INFO [train.py:968] (1/2) Epoch 9, batch 3100, giga_loss[loss=0.259, simple_loss=0.339, pruned_loss=0.08954, over 28683.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3566, pruned_loss=0.1071, over 5700473.26 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3593, pruned_loss=0.1011, over 4630897.44 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.356, pruned_loss=0.1079, over 5688000.75 frames. ], batch size: 242, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:00:11,808 INFO [optim.py:369] (1/2) 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 14:00:53,198 INFO [train.py:968] (1/2) Epoch 9, batch 3150, giga_loss[loss=0.2586, simple_loss=0.3368, pruned_loss=0.09017, over 28926.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3542, pruned_loss=0.1049, over 5707390.39 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3593, pruned_loss=0.1012, over 4637133.95 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3537, pruned_loss=0.1055, over 5696856.94 frames. ], batch size: 106, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:01:20,427 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:01:32,793 INFO [train.py:968] (1/2) Epoch 9, batch 3200, giga_loss[loss=0.3103, simple_loss=0.3891, pruned_loss=0.1158, over 28891.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3563, pruned_loss=0.1057, over 5686878.64 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3597, pruned_loss=0.1014, over 4651544.95 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3555, pruned_loss=0.1062, over 5698333.85 frames. ], batch size: 174, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:01:34,797 INFO [optim.py:369] (1/2) 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 14:02:09,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4299, 2.0494, 1.7944, 1.6436], device='cuda:1'), covar=tensor([0.0750, 0.0261, 0.0263, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:1')
+2023-03-04 14:02:10,584 INFO [train.py:968] (1/2) Epoch 9, batch 3250, giga_loss[loss=0.2663, simple_loss=0.3436, pruned_loss=0.0945, over 28851.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3587, pruned_loss=0.1068, over 5693221.10 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3599, pruned_loss=0.1015, over 4667216.76 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3579, pruned_loss=0.1071, over 5702334.25 frames. ], batch size: 199, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:02:50,647 INFO [train.py:968] (1/2) Epoch 9, batch 3300, giga_loss[loss=0.3021, simple_loss=0.3779, pruned_loss=0.1132, over 28881.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3601, pruned_loss=0.1079, over 5695065.74 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3601, pruned_loss=0.1016, over 4690191.98 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3593, pruned_loss=0.1081, over 5699937.20 frames. ], batch size: 186, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:02:52,657 INFO [optim.py:369] (1/2) 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 14:02:56,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:03:06,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3422, 3.1644, 1.5271, 1.3764], device='cuda:1'), covar=tensor([0.0887, 0.0298, 0.0765, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0487, 0.0321, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 14:03:15,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1195, 1.3922, 1.2822, 1.0077], device='cuda:1'), covar=tensor([0.1736, 0.1381, 0.1114, 0.1465], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1477, 0.1460, 0.1553], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 14:03:17,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:03:30,506 INFO [train.py:968] (1/2) Epoch 9, batch 3350, giga_loss[loss=0.3073, simple_loss=0.3738, pruned_loss=0.1204, over 28788.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3607, pruned_loss=0.1086, over 5698426.62 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3595, pruned_loss=0.1011, over 4723251.06 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3605, pruned_loss=0.1094, over 5699547.87 frames. ], batch size: 284, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:03:32,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:03:47,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2332, 1.4711, 1.2722, 1.3057], device='cuda:1'), covar=tensor([0.1665, 0.1815, 0.1971, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0728, 0.0655, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 14:04:01,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 14:04:14,100 INFO [train.py:968] (1/2) Epoch 9, batch 3400, giga_loss[loss=0.3116, simple_loss=0.3679, pruned_loss=0.1277, over 28872.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3607, pruned_loss=0.1091, over 5707837.13 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3585, pruned_loss=0.1007, over 4751399.31 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3613, pruned_loss=0.1102, over 5705021.52 frames. ], batch size: 112, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:04:16,331 INFO [optim.py:369] (1/2) 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 14:04:18,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-04 14:04:34,185 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=367622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:04:54,100 INFO [train.py:968] (1/2) Epoch 9, batch 3450, giga_loss[loss=0.2601, simple_loss=0.3422, pruned_loss=0.08904, over 28328.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3609, pruned_loss=0.109, over 5717356.40 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3588, pruned_loss=0.1008, over 4774238.74 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3612, pruned_loss=0.1099, over 5711713.43 frames. ], batch size: 77, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:04:54,415 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=367648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:04:56,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=367651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:04,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3678, 1.5256, 1.3054, 1.4178], device='cuda:1'), covar=tensor([0.2043, 0.2051, 0.2120, 0.1934], device='cuda:1'), in_proj_covar=tensor([0.1224, 0.0915, 0.1080, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 14:05:14,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=367675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:16,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=367678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:17,629 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 3500, giga_loss[loss=0.3138, simple_loss=0.3915, pruned_loss=0.118, over 28698.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3608, pruned_loss=0.1083, over 5720447.16 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3585, pruned_loss=0.1006, over 4814705.46 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3613, pruned_loss=0.1094, over 5712996.33 frames. ], batch size: 284, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:05:35,287 INFO [optim.py:369] (1/2) 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 14:05:39,883 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=367707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:05:41,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4129, 4.2315, 3.9815, 1.8379], device='cuda:1'), covar=tensor([0.0469, 0.0699, 0.0719, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.0966, 0.0902, 0.0807, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 14:05:56,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2577, 3.4608, 2.4952, 1.2331], device='cuda:1'), covar=tensor([0.4307, 0.1231, 0.2105, 0.3842], device='cuda:1'), in_proj_covar=tensor([0.1479, 0.1391, 0.1436, 0.1195], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 14:06:13,120 INFO [train.py:968] (1/2) Epoch 9, batch 3550, giga_loss[loss=0.2693, simple_loss=0.3572, pruned_loss=0.09073, over 28771.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3605, pruned_loss=0.1072, over 5718682.31 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3582, pruned_loss=0.1005, over 4836021.67 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3612, pruned_loss=0.1083, over 5709727.04 frames. ], batch size: 119, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:06:21,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:06:53,359 INFO [train.py:968] (1/2) Epoch 9, batch 3600, giga_loss[loss=0.2532, simple_loss=0.3344, pruned_loss=0.08597, over 28642.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3612, pruned_loss=0.1067, over 5726686.87 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3582, pruned_loss=0.1005, over 4872436.71 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3619, pruned_loss=0.1078, over 5714117.21 frames. ], batch size: 71, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:06:57,276 INFO [optim.py:369] (1/2) 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 14:07:31,440 INFO [train.py:968] (1/2) Epoch 9, batch 3650, giga_loss[loss=0.3, simple_loss=0.3698, pruned_loss=0.1151, over 28533.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.361, pruned_loss=0.1067, over 5727392.22 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3594, pruned_loss=0.1014, over 4901502.15 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3607, pruned_loss=0.1071, over 5713746.77 frames. ], batch size: 336, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:07:42,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-04 14:07:47,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7678, 4.5594, 4.3989, 2.0631], device='cuda:1'), covar=tensor([0.0511, 0.0760, 0.0841, 0.1826], device='cuda:1'), in_proj_covar=tensor([0.0967, 0.0900, 0.0805, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 14:08:04,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9033, 1.7982, 1.7034, 1.6033], device='cuda:1'), covar=tensor([0.1335, 0.2386, 0.1891, 0.1958], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0720, 0.0646, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 14:08:11,670 INFO [train.py:968] (1/2) Epoch 9, batch 3700, giga_loss[loss=0.2597, simple_loss=0.3306, pruned_loss=0.09444, over 28522.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3596, pruned_loss=0.1066, over 5733975.64 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3593, pruned_loss=0.1013, over 4936178.02 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3594, pruned_loss=0.1071, over 5717486.39 frames. ], batch size: 71, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:08:14,043 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:08:16,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=367904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:08:33,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:08:39,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=367933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:08:50,164 INFO [train.py:968] (1/2) Epoch 9, batch 3750, giga_loss[loss=0.2732, simple_loss=0.3469, pruned_loss=0.09975, over 28932.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3571, pruned_loss=0.1052, over 5730735.68 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3588, pruned_loss=0.101, over 4950808.42 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3574, pruned_loss=0.106, over 5715250.85 frames. ], batch size: 145, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:08:50,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7795, 0.9309, 3.3891, 2.9703], device='cuda:1'), covar=tensor([0.1827, 0.2675, 0.0439, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0562, 0.0803, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 14:09:21,868 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 14:09:28,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=367997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:09:29,364 INFO [train.py:968] (1/2) Epoch 9, batch 3800, giga_loss[loss=0.2413, simple_loss=0.3249, pruned_loss=0.07879, over 28863.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.356, pruned_loss=0.1044, over 5741344.05 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3588, pruned_loss=0.101, over 4983657.24 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3561, pruned_loss=0.1051, over 5723411.17 frames. ], batch size: 66, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:09:32,942 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:09:45,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4098, 1.6530, 1.6597, 1.3048], device='cuda:1'), covar=tensor([0.1460, 0.2039, 0.1160, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0704, 0.0835, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 14:10:11,594 INFO [train.py:968] (1/2) Epoch 9, batch 3850, giga_loss[loss=0.277, simple_loss=0.3544, pruned_loss=0.09982, over 28966.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3571, pruned_loss=0.1056, over 5736456.14 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3588, pruned_loss=0.101, over 4983657.24 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3571, pruned_loss=0.1062, over 5722498.75 frames. ], batch size: 227, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:10:24,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:10:27,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:10:30,500 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 3900, giga_loss[loss=0.3408, simple_loss=0.3973, pruned_loss=0.1422, over 27568.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.357, pruned_loss=0.1054, over 5735223.42 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3582, pruned_loss=0.1009, over 5019569.81 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3575, pruned_loss=0.1061, over 5718144.28 frames. ], batch size: 472, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:10:49,649 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:11:20,468 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:11:23,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:11:24,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-04 14:11:26,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368146.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:11:27,118 INFO [train.py:968] (1/2) Epoch 9, batch 3950, libri_loss[loss=0.2419, simple_loss=0.3201, pruned_loss=0.08182, over 29568.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3554, pruned_loss=0.1037, over 5731580.55 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3573, pruned_loss=0.1006, over 5043845.89 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3564, pruned_loss=0.1046, over 5714680.09 frames. ], batch size: 78, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:11:44,080 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:12:04,081 INFO [train.py:968] (1/2) Epoch 9, batch 4000, giga_loss[loss=0.3457, simple_loss=0.3955, pruned_loss=0.1479, over 28883.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3552, pruned_loss=0.104, over 5731649.15 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3574, pruned_loss=0.1006, over 5058712.12 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.356, pruned_loss=0.1048, over 5717287.45 frames. ], batch size: 213, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:12:07,412 INFO [optim.py:369] (1/2) 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 14:12:19,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:12:41,141 INFO [train.py:968] (1/2) Epoch 9, batch 4050, giga_loss[loss=0.2554, simple_loss=0.3315, pruned_loss=0.0896, over 28597.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3537, pruned_loss=0.1037, over 5714944.45 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3575, pruned_loss=0.1005, over 5068558.94 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3542, pruned_loss=0.1045, over 5708685.23 frames. ], batch size: 85, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:13:04,716 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-04 14:13:12,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5521, 1.8355, 1.6446, 1.7448], device='cuda:1'), covar=tensor([0.1647, 0.1887, 0.2073, 0.1773], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0720, 0.0645, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 14:13:18,203 INFO [train.py:968] (1/2) Epoch 9, batch 4100, giga_loss[loss=0.2539, simple_loss=0.3328, pruned_loss=0.08751, over 29002.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3509, pruned_loss=0.102, over 5716791.24 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3575, pruned_loss=0.1006, over 5094083.60 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1027, over 5708891.42 frames. ], batch size: 164, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:13:22,626 INFO [optim.py:369] (1/2) 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 14:13:34,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3676, 1.5458, 1.2332, 1.6250], device='cuda:1'), covar=tensor([0.2333, 0.2165, 0.2362, 0.2160], device='cuda:1'), in_proj_covar=tensor([0.1227, 0.0907, 0.1084, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 14:13:41,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 14:13:56,518 INFO [train.py:968] (1/2) Epoch 9, batch 4150, giga_loss[loss=0.344, simple_loss=0.4044, pruned_loss=0.1418, over 28725.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3488, pruned_loss=0.1008, over 5715750.21 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3572, pruned_loss=0.1003, over 5121908.91 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3489, pruned_loss=0.1016, over 5706352.01 frames. ], batch size: 242, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:14:03,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6059, 4.4114, 4.2260, 1.8990], device='cuda:1'), covar=tensor([0.0545, 0.0703, 0.0818, 0.1886], device='cuda:1'), in_proj_covar=tensor([0.0972, 0.0899, 0.0800, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 14:14:16,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:14:22,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-04 14:14:33,165 INFO [train.py:968] (1/2) Epoch 9, batch 4200, giga_loss[loss=0.2732, simple_loss=0.3396, pruned_loss=0.1034, over 28548.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.349, pruned_loss=0.1018, over 5713050.53 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3576, pruned_loss=0.1006, over 5137556.07 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3486, pruned_loss=0.1022, over 5702188.45 frames. ], batch size: 71, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:14:39,016 INFO [optim.py:369] (1/2) 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 14:15:11,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-04 14:15:14,235 INFO [train.py:968] (1/2) Epoch 9, batch 4250, giga_loss[loss=0.3542, simple_loss=0.4012, pruned_loss=0.1537, over 27847.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3484, pruned_loss=0.1023, over 5714702.93 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3579, pruned_loss=0.1008, over 5152311.77 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3476, pruned_loss=0.1025, over 5702931.40 frames. ], batch size: 412, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:15:15,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:15:22,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-04 14:15:38,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:15:52,851 INFO [train.py:968] (1/2) Epoch 9, batch 4300, giga_loss[loss=0.2619, simple_loss=0.3286, pruned_loss=0.09761, over 28635.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3459, pruned_loss=0.1014, over 5715270.84 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3578, pruned_loss=0.101, over 5173022.63 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.345, pruned_loss=0.1014, over 5702269.33 frames. ], batch size: 85, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:15:57,289 INFO [optim.py:369] (1/2) 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 14:15:59,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3657, 1.5913, 1.3850, 1.5012], device='cuda:1'), covar=tensor([0.0732, 0.0304, 0.0317, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:1')
+2023-03-04 14:16:07,547 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:09,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368521.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:16:10,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:22,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:27,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2911, 1.5157, 1.2676, 1.0029], device='cuda:1'), covar=tensor([0.2131, 0.2157, 0.2419, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.0908, 0.1082, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 14:16:30,075 INFO [train.py:968] (1/2) Epoch 9, batch 4350, libri_loss[loss=0.3045, simple_loss=0.3853, pruned_loss=0.1119, over 27828.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3441, pruned_loss=0.1011, over 5712961.05 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3578, pruned_loss=0.1009, over 5191169.67 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3431, pruned_loss=0.1012, over 5700923.92 frames. ], batch size: 116, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:16:32,340 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:16:49,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5283, 3.7191, 1.5761, 1.5916], device='cuda:1'), covar=tensor([0.0847, 0.0330, 0.0848, 0.1206], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0490, 0.0323, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 14:17:05,525 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:06,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:07,612 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 4400, giga_loss[loss=0.2655, simple_loss=0.3381, pruned_loss=0.09646, over 28884.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3423, pruned_loss=0.09986, over 5718354.62 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3584, pruned_loss=0.101, over 5207010.99 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3406, pruned_loss=0.0998, over 5706181.06 frames. ], batch size: 174, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:17:12,457 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:46,339 INFO [train.py:968] (1/2) Epoch 9, batch 4450, giga_loss[loss=0.2612, simple_loss=0.3377, pruned_loss=0.0923, over 28826.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3413, pruned_loss=0.09928, over 5720907.20 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3584, pruned_loss=0.1011, over 5214061.19 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3398, pruned_loss=0.09917, over 5709618.59 frames. ], batch size: 199, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:17:55,527 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:17:59,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368664.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:18:01,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368667.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:18:03,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:18:14,446 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:18:27,159 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368696.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:18:29,784 INFO [train.py:968] (1/2) Epoch 9, batch 4500, giga_loss[loss=0.3336, simple_loss=0.3829, pruned_loss=0.1422, over 26786.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3452, pruned_loss=0.1012, over 5702964.97 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3582, pruned_loss=0.1011, over 5217172.15 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3439, pruned_loss=0.1011, over 5700802.72 frames. ], batch size: 555, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:18:33,795 INFO [optim.py:369] (1/2) 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 14:18:59,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0387, 1.1719, 3.3684, 2.9722], device='cuda:1'), covar=tensor([0.1576, 0.2448, 0.0387, 0.1679], device='cuda:1'), in_proj_covar=tensor([0.0616, 0.0561, 0.0801, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 14:19:00,101 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:02,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=368740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:09,247 INFO [train.py:968] (1/2) Epoch 9, batch 4550, giga_loss[loss=0.3447, simple_loss=0.419, pruned_loss=0.1352, over 28678.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3488, pruned_loss=0.103, over 5701965.81 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3585, pruned_loss=0.1013, over 5230093.73 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3474, pruned_loss=0.1028, over 5697477.28 frames. ], batch size: 262, lr: 3.69e-03, grad_scale: 8.0
+2023-03-04 14:19:18,279 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:25,611 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=368769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:19:49,619 INFO [train.py:968] (1/2) Epoch 9, batch 4600, giga_loss[loss=0.277, simple_loss=0.3594, pruned_loss=0.09731, over 28898.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3519, pruned_loss=0.1042, over 5706792.86 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3585, pruned_loss=0.1013, over 5246120.51 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3506, pruned_loss=0.104, over 5699405.97 frames. ], batch size: 174, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:19:56,830 INFO [optim.py:369] (1/2) 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 14:20:13,757 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4421, 2.2798, 1.5925, 0.5694], device='cuda:1'), covar=tensor([0.3255, 0.1644, 0.2354, 0.3115], device='cuda:1'), in_proj_covar=tensor([0.1490, 0.1400, 0.1442, 0.1208], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 14:20:34,817 INFO [train.py:968] (1/2) Epoch 9, batch 4650, giga_loss[loss=0.2562, simple_loss=0.3266, pruned_loss=0.09291, over 28544.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3513, pruned_loss=0.1031, over 5696753.39 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3585, pruned_loss=0.1012, over 5254977.47 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3502, pruned_loss=0.1031, over 5689077.09 frames. ], batch size: 92, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:20:39,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:21:14,271 INFO [train.py:968] (1/2) Epoch 9, batch 4700, giga_loss[loss=0.3564, simple_loss=0.3909, pruned_loss=0.1609, over 28542.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3501, pruned_loss=0.1025, over 5696050.41 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3586, pruned_loss=0.1014, over 5251452.88 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3491, pruned_loss=0.1024, over 5696250.95 frames. ], batch size: 78, lr: 3.69e-03, grad_scale: 4.0
+2023-03-04 14:21:18,821 INFO [optim.py:369] (1/2) 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 14:21:19,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2292, 1.6745, 1.3264, 1.4225], device='cuda:1'), covar=tensor([0.0720, 0.0327, 0.0323, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0080], device='cuda:1')
+2023-03-04 14:21:23,758 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:21:25,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=368913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:21:29,792 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:21:37,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8834, 1.9062, 1.4281, 1.7436], device='cuda:1'), covar=tensor([0.0700, 0.0552, 0.0938, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0340, 0.0438, 0.0492, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 14:21:53,521 INFO [train.py:968] (1/2) Epoch 9, batch 4750, giga_loss[loss=0.2779, simple_loss=0.3534, pruned_loss=0.1011, over 29015.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3507, pruned_loss=0.1032, over 5698508.88 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3585, pruned_loss=0.1015, over 5267292.08 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3497, pruned_loss=0.103, over 5694304.97 frames. ], batch size: 155, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:22:12,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=368972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:22:17,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4186, 3.7437, 1.5874, 1.5852], device='cuda:1'), covar=tensor([0.0919, 0.0340, 0.0795, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0493, 0.0323, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 14:22:33,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=368997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:22:33,982 INFO [train.py:968] (1/2) Epoch 9, batch 4800, giga_loss[loss=0.3463, simple_loss=0.3956, pruned_loss=0.1485, over 26586.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3523, pruned_loss=0.1045, over 5698955.20 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3585, pruned_loss=0.1015, over 5281469.07 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3513, pruned_loss=0.1043, over 5692562.20 frames. ], batch size: 555, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:22:35,545 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:22:57,793 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:03,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:11,324 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 4850, giga_loss[loss=0.3314, simple_loss=0.3932, pruned_loss=0.1348, over 28872.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3557, pruned_loss=0.1064, over 5691961.94 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3589, pruned_loss=0.102, over 5287737.98 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3544, pruned_loss=0.106, over 5690954.20 frames. ], batch size: 145, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:23:18,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:20,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:21,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:44,363 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:23:52,817 INFO [train.py:968] (1/2) Epoch 9, batch 4900, giga_loss[loss=0.2789, simple_loss=0.3625, pruned_loss=0.09766, over 28844.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3581, pruned_loss=0.1074, over 5700973.29 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3591, pruned_loss=0.1021, over 5290459.29 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.357, pruned_loss=0.107, over 5699635.48 frames. ], batch size: 227, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:23:58,144 INFO [optim.py:369] (1/2) 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 14:24:20,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 4950, giga_loss[loss=0.2617, simple_loss=0.3329, pruned_loss=0.09528, over 28795.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3587, pruned_loss=0.1074, over 5706501.07 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3589, pruned_loss=0.1019, over 5305098.12 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.358, pruned_loss=0.1074, over 5701347.84 frames. ], batch size: 99, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:24:57,397 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:24:59,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:04,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:06,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:12,120 INFO [train.py:968] (1/2) Epoch 9, batch 5000, giga_loss[loss=0.3337, simple_loss=0.394, pruned_loss=0.1367, over 28297.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3593, pruned_loss=0.1072, over 5713709.59 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.359, pruned_loss=0.1019, over 5317916.63 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3586, pruned_loss=0.1074, over 5706921.92 frames. ], batch size: 368, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:25:14,620 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:16,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:16,756 INFO [optim.py:369] (1/2) 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 14:25:19,040 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:20,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:29,573 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:40,016 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:25:41,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4956, 2.2029, 1.9163, 1.4342], device='cuda:1'), covar=tensor([0.1761, 0.2080, 0.1395, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0700, 0.0830, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 14:25:46,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3965, 3.2996, 1.4077, 1.4512], device='cuda:1'), covar=tensor([0.0883, 0.0327, 0.0877, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0493, 0.0323, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 14:25:49,647 INFO [train.py:968] (1/2) Epoch 9, batch 5050, giga_loss[loss=0.3101, simple_loss=0.378, pruned_loss=0.1211, over 28889.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3598, pruned_loss=0.1076, over 5721305.75 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3595, pruned_loss=0.1022, over 5336019.45 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3588, pruned_loss=0.1077, over 5713509.31 frames. ], batch size: 186, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:26:01,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:02,016 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 14:26:11,886 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:15,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:19,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:24,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:28,835 INFO [train.py:968] (1/2) Epoch 9, batch 5100, libri_loss[loss=0.2737, simple_loss=0.3546, pruned_loss=0.0964, over 29582.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3584, pruned_loss=0.1067, over 5718700.12 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3589, pruned_loss=0.1018, over 5348422.58 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3581, pruned_loss=0.1073, over 5716055.02 frames. ], batch size: 75, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:26:34,449 INFO [optim.py:369] (1/2) 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 14:26:35,844 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:26:37,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:27:05,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:27:06,147 INFO [train.py:968] (1/2) Epoch 9, batch 5150, giga_loss[loss=0.2693, simple_loss=0.3439, pruned_loss=0.09739, over 28985.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3564, pruned_loss=0.1058, over 5707354.58 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.359, pruned_loss=0.1019, over 5343462.51 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3561, pruned_loss=0.1063, over 5714300.55 frames. ], batch size: 145, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:27:45,643 INFO [train.py:968] (1/2) Epoch 9, batch 5200, giga_loss[loss=0.2191, simple_loss=0.3016, pruned_loss=0.06827, over 28406.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3529, pruned_loss=0.1039, over 5715320.80 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3589, pruned_loss=0.1019, over 5360126.12 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3526, pruned_loss=0.1043, over 5715509.86 frames. ], batch size: 71, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:27:52,125 INFO [optim.py:369] (1/2) 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 14:28:11,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:13,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:17,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6312, 2.1674, 1.3161, 0.7243], device='cuda:1'), covar=tensor([0.3996, 0.2207, 0.2524, 0.4420], device='cuda:1'), in_proj_covar=tensor([0.1489, 0.1406, 0.1451, 0.1215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 14:28:17,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:20,582 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:26,277 INFO [train.py:968] (1/2) Epoch 9, batch 5250, giga_loss[loss=0.2503, simple_loss=0.3274, pruned_loss=0.08658, over 28985.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3513, pruned_loss=0.1032, over 5715856.17 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3591, pruned_loss=0.102, over 5365841.54 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3508, pruned_loss=0.1035, over 5714098.79 frames. ], batch size: 136, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:28:36,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:28:37,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2128, 2.0703, 1.9341, 1.8620], device='cuda:1'), covar=tensor([0.1150, 0.2149, 0.1672, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0724, 0.0650, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 14:28:42,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:29:00,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:29:02,185 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:29:06,120 INFO [train.py:968] (1/2) Epoch 9, batch 5300, giga_loss[loss=0.358, simple_loss=0.4132, pruned_loss=0.1514, over 26633.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.354, pruned_loss=0.1036, over 5713313.00 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3596, pruned_loss=0.1022, over 5378118.74 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.353, pruned_loss=0.1036, over 5708582.60 frames. ], batch size: 555, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:29:12,579 INFO [optim.py:369] (1/2) 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 14:29:24,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8720, 2.8310, 1.8030, 0.8969], device='cuda:1'), covar=tensor([0.4704, 0.2067, 0.2632, 0.4556], device='cuda:1'), in_proj_covar=tensor([0.1489, 0.1404, 0.1450, 0.1216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 14:29:25,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:29:32,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5664, 1.8554, 1.4062, 1.5251], device='cuda:1'), covar=tensor([0.0689, 0.0257, 0.0309, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0116, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 14:29:46,840 INFO [train.py:968] (1/2) Epoch 9, batch 5350, giga_loss[loss=0.3103, simple_loss=0.3807, pruned_loss=0.1199, over 27671.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3551, pruned_loss=0.1035, over 5709161.97 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3595, pruned_loss=0.1022, over 5386019.20 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1036, over 5702732.77 frames. ], batch size: 472, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:29:53,789 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:30:06,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6757, 1.9958, 1.9422, 1.4752], device='cuda:1'), covar=tensor([0.1487, 0.1829, 0.1206, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0698, 0.0830, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 14:30:12,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2832, 1.4456, 1.1818, 1.4570], device='cuda:1'), covar=tensor([0.0669, 0.0368, 0.0329, 0.0729], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0116, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0073, 0.0052, 0.0047, 0.0079], device='cuda:1')
+2023-03-04 14:30:13,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:30:24,136 INFO [train.py:968] (1/2) Epoch 9, batch 5400, giga_loss[loss=0.2705, simple_loss=0.3416, pruned_loss=0.09967, over 28987.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.355, pruned_loss=0.1046, over 5710986.32 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3591, pruned_loss=0.102, over 5396477.51 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3547, pruned_loss=0.1048, over 5702319.84 frames. ], batch size: 145, lr: 3.68e-03, grad_scale: 8.0
+2023-03-04 14:30:29,312 INFO [optim.py:369] (1/2) 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 14:30:54,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:31:02,282 INFO [train.py:968] (1/2) Epoch 9, batch 5450, giga_loss[loss=0.251, simple_loss=0.3237, pruned_loss=0.08916, over 29032.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3542, pruned_loss=0.1054, over 5704357.97 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3593, pruned_loss=0.1022, over 5397225.38 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3537, pruned_loss=0.1055, over 5702398.21 frames. ], batch size: 128, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:31:27,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:31:33,473 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369689.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:31:40,010 INFO [train.py:968] (1/2) Epoch 9, batch 5500, giga_loss[loss=0.2507, simple_loss=0.3243, pruned_loss=0.08858, over 28994.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3533, pruned_loss=0.1059, over 5693173.97 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3593, pruned_loss=0.1023, over 5405405.64 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3527, pruned_loss=0.1061, over 5699547.82 frames. ], batch size: 128, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:31:46,042 INFO [optim.py:369] (1/2) 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 14:31:48,079 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:02,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:05,538 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:18,736 INFO [train.py:968] (1/2) Epoch 9, batch 5550, libri_loss[loss=0.3264, simple_loss=0.393, pruned_loss=0.1299, over 29526.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3514, pruned_loss=0.1059, over 5699244.45 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3594, pruned_loss=0.1026, over 5416568.87 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3506, pruned_loss=0.1059, over 5700666.99 frames. ], batch size: 79, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:32:26,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:37,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369773.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:32:43,950 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:46,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:32:58,954 INFO [train.py:968] (1/2) Epoch 9, batch 5600, giga_loss[loss=0.2563, simple_loss=0.3262, pruned_loss=0.09325, over 28572.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3496, pruned_loss=0.1051, over 5704742.74 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3597, pruned_loss=0.1028, over 5427150.55 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3485, pruned_loss=0.105, over 5702873.62 frames. ], batch size: 71, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:33:05,687 INFO [optim.py:369] (1/2) 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 14:33:10,618 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:33:10,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5494, 2.1782, 1.8696, 1.9290], device='cuda:1'), covar=tensor([0.0643, 0.0738, 0.0900, 0.1076], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0448, 0.0501, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 14:33:21,438 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=369825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:33:23,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=369828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:33:39,097 INFO [train.py:968] (1/2) Epoch 9, batch 5650, giga_loss[loss=0.2331, simple_loss=0.3056, pruned_loss=0.08031, over 28727.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3472, pruned_loss=0.104, over 5710451.63 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3598, pruned_loss=0.1028, over 5429693.80 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3463, pruned_loss=0.1039, over 5708059.00 frames. ], batch size: 92, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:33:47,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=369857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:34:03,562 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=369880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:34:16,182 INFO [train.py:968] (1/2) Epoch 9, batch 5700, giga_loss[loss=0.2324, simple_loss=0.3157, pruned_loss=0.07453, over 28326.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3413, pruned_loss=0.1005, over 5702228.84 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3603, pruned_loss=0.103, over 5418651.61 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3399, pruned_loss=0.1002, over 5715316.09 frames. ], batch size: 368, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:34:23,389 INFO [optim.py:369] (1/2) 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 14:34:44,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=369932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:34:56,318 INFO [train.py:968] (1/2) Epoch 9, batch 5750, giga_loss[loss=0.277, simple_loss=0.3414, pruned_loss=0.1063, over 28856.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3374, pruned_loss=0.0982, over 5710718.02 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3604, pruned_loss=0.1033, over 5433130.63 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3355, pruned_loss=0.09758, over 5715734.96 frames. ], batch size: 199, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:34:56,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7608, 4.7483, 1.8752, 1.9717], device='cuda:1'), covar=tensor([0.0850, 0.0242, 0.0828, 0.1163], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0496, 0.0325, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 14:35:33,103 INFO [train.py:968] (1/2) Epoch 9, batch 5800, giga_loss[loss=0.2318, simple_loss=0.32, pruned_loss=0.07181, over 28921.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3391, pruned_loss=0.09867, over 5710760.91 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3607, pruned_loss=0.1035, over 5436280.67 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3371, pruned_loss=0.09797, over 5715862.84 frames. ], batch size: 174, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:35:41,842 INFO [optim.py:369] (1/2) 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 14:35:54,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:01,829 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 5850, giga_loss[loss=0.2641, simple_loss=0.3397, pruned_loss=0.09428, over 28510.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3411, pruned_loss=0.09878, over 5717136.87 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3604, pruned_loss=0.1034, over 5446034.66 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3394, pruned_loss=0.09828, over 5717409.99 frames. ], batch size: 71, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:36:28,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370064.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:36:37,421 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:39,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:43,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:36:55,189 INFO [train.py:968] (1/2) Epoch 9, batch 5900, giga_loss[loss=0.303, simple_loss=0.3588, pruned_loss=0.1236, over 23753.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3454, pruned_loss=0.1008, over 5718576.93 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3609, pruned_loss=0.1038, over 5458864.37 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3431, pruned_loss=0.1, over 5714365.23 frames. ], batch size: 705, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:37:01,483 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:37:01,801 INFO [optim.py:369] (1/2) 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 14:37:10,913 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 5950, giga_loss[loss=0.2786, simple_loss=0.3588, pruned_loss=0.09919, over 28888.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3487, pruned_loss=0.1024, over 5716547.85 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3605, pruned_loss=0.1034, over 5465558.59 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3471, pruned_loss=0.102, over 5710481.56 frames. ], batch size: 186, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:37:36,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370148.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:37:43,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:38:03,400 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 6000, giga_loss[loss=0.275, simple_loss=0.3496, pruned_loss=0.1002, over 28402.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3509, pruned_loss=0.1033, over 5715372.14 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.361, pruned_loss=0.1036, over 5473083.74 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.349, pruned_loss=0.1028, over 5709196.19 frames. ], batch size: 65, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:38:16,864 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 14:38:23,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2931, 1.9167, 1.4234, 0.3951], device='cuda:1'), covar=tensor([0.2583, 0.1642, 0.3029, 0.3670], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1413, 0.1450, 0.1216], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 14:38:25,196 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 14:38:32,867 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:38:35,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370210.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:38:48,522 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:38:50,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:00,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370239.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:39:05,960 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=370246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:07,084 INFO [train.py:968] (1/2) Epoch 9, batch 6050, giga_loss[loss=0.3841, simple_loss=0.4269, pruned_loss=0.1706, over 28689.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3568, pruned_loss=0.1081, over 5707292.52 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.362, pruned_loss=0.1044, over 5476956.76 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3542, pruned_loss=0.1071, over 5705597.66 frames. ], batch size: 262, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:39:13,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:16,813 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:20,299 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:22,161 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:47,104 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:39:48,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:39:49,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370294.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 14:39:53,289 INFO [train.py:968] (1/2) Epoch 9, batch 6100, giga_loss[loss=0.344, simple_loss=0.4023, pruned_loss=0.1429, over 28638.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3635, pruned_loss=0.1137, over 5706865.91 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3623, pruned_loss=0.1046, over 5483209.54 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3612, pruned_loss=0.1128, over 5702922.38 frames. ], batch size: 307, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:40:03,100 INFO [optim.py:369] (1/2) 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 14:40:15,293 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370323.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 14:40:39,658 INFO [train.py:968] (1/2) Epoch 9, batch 6150, giga_loss[loss=0.3916, simple_loss=0.4348, pruned_loss=0.1742, over 28735.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3707, pruned_loss=0.1195, over 5680548.32 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3625, pruned_loss=0.1047, over 5489026.47 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3687, pruned_loss=0.119, over 5674976.15 frames. ], batch size: 284, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:40:53,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0768, 1.1807, 3.7103, 2.9425], device='cuda:1'), covar=tensor([0.1641, 0.2421, 0.0419, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0621, 0.0566, 0.0816, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 14:41:25,260 INFO [train.py:968] (1/2) Epoch 9, batch 6200, giga_loss[loss=0.3432, simple_loss=0.3989, pruned_loss=0.1437, over 28742.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3782, pruned_loss=0.1258, over 5675889.30 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3625, pruned_loss=0.1048, over 5491508.77 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3767, pruned_loss=0.1254, over 5671696.99 frames. ], batch size: 242, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:41:26,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:26,810 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:29,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:32,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:41:35,503 INFO [optim.py:369] (1/2) 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 14:41:40,131 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 14:41:52,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:42:06,875 INFO [train.py:968] (1/2) Epoch 9, batch 6250, giga_loss[loss=0.3101, simple_loss=0.3801, pruned_loss=0.1201, over 28814.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3829, pruned_loss=0.1299, over 5684096.85 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3628, pruned_loss=0.1048, over 5502738.30 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3821, pruned_loss=0.1302, over 5675620.58 frames. ], batch size: 66, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:42:14,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3273, 5.1255, 4.8483, 2.5018], device='cuda:1'), covar=tensor([0.0364, 0.0545, 0.0581, 0.1658], device='cuda:1'), in_proj_covar=tensor([0.0985, 0.0915, 0.0809, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 14:42:46,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0084, 1.7093, 1.4866, 1.4053], device='cuda:1'), covar=tensor([0.0675, 0.0703, 0.0856, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0442, 0.0495, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 14:42:52,081 INFO [train.py:968] (1/2) Epoch 9, batch 6300, giga_loss[loss=0.3263, simple_loss=0.3859, pruned_loss=0.1333, over 28829.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3894, pruned_loss=0.1356, over 5675831.68 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3633, pruned_loss=0.1051, over 5511180.04 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3887, pruned_loss=0.1361, over 5664614.96 frames. ], batch size: 186, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:43:03,304 INFO [optim.py:369] (1/2) 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 14:43:16,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0266, 1.3211, 1.0364, 0.2794], device='cuda:1'), covar=tensor([0.2111, 0.1835, 0.2891, 0.3574], device='cuda:1'), in_proj_covar=tensor([0.1495, 0.1417, 0.1454, 0.1218], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 14:43:25,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:34,691 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:37,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:41,200 INFO [train.py:968] (1/2) Epoch 9, batch 6350, giga_loss[loss=0.394, simple_loss=0.4271, pruned_loss=0.1805, over 27438.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3927, pruned_loss=0.1395, over 5660136.62 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3631, pruned_loss=0.105, over 5519926.57 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3929, pruned_loss=0.1407, over 5646664.20 frames. ], batch size: 472, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:43:43,452 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:46,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:43:50,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:44:04,638 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:44:06,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3292, 3.1267, 2.9795, 1.3579], device='cuda:1'), covar=tensor([0.0758, 0.0953, 0.0903, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.0981, 0.0917, 0.0809, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 14:44:12,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:44:27,953 INFO [train.py:968] (1/2) Epoch 9, batch 6400, libri_loss[loss=0.2765, simple_loss=0.3636, pruned_loss=0.09473, over 29268.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.394, pruned_loss=0.1417, over 5650519.46 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3632, pruned_loss=0.1051, over 5529968.63 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3952, pruned_loss=0.1438, over 5634710.58 frames. ], batch size: 94, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:44:38,758 INFO [optim.py:369] (1/2) 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 14:44:45,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9580, 1.7969, 1.4127, 1.4125], device='cuda:1'), covar=tensor([0.0623, 0.0591, 0.0910, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0442, 0.0497, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 14:44:50,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=370621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:45:18,320 INFO [train.py:968] (1/2) Epoch 9, batch 6450, giga_loss[loss=0.3887, simple_loss=0.4283, pruned_loss=0.1746, over 28545.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3987, pruned_loss=0.1472, over 5621622.33 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3632, pruned_loss=0.1052, over 5524436.10 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.3999, pruned_loss=0.1491, over 5614327.06 frames. ], batch size: 307, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:45:47,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:45:49,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:08,343 INFO [train.py:968] (1/2) Epoch 9, batch 6500, giga_loss[loss=0.5133, simple_loss=0.5016, pruned_loss=0.2625, over 27557.00 frames. ], tot_loss[loss=0.3505, simple_loss=0.4019, pruned_loss=0.1495, over 5617000.85 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3633, pruned_loss=0.1052, over 5530575.23 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.4037, pruned_loss=0.1521, over 5607717.13 frames. ], batch size: 472, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:46:11,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:16,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:17,567 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:19,432 INFO [optim.py:369] (1/2) 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 14:46:43,853 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=370733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:46:56,338 INFO [train.py:968] (1/2) Epoch 9, batch 6550, giga_loss[loss=0.4593, simple_loss=0.4703, pruned_loss=0.2241, over 26620.00 frames. ], tot_loss[loss=0.3513, simple_loss=0.4024, pruned_loss=0.1501, over 5628412.61 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3633, pruned_loss=0.1052, over 5536562.97 frames. ], giga_tot_loss[loss=0.3553, simple_loss=0.4045, pruned_loss=0.1531, over 5617358.91 frames. ], batch size: 555, lr: 3.68e-03, grad_scale: 4.0
+2023-03-04 14:47:11,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=370764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:47:14,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=370767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 14:47:44,431 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 6600, giga_loss[loss=0.5396, simple_loss=0.5042, pruned_loss=0.2875, over 23536.00 frames. ], tot_loss[loss=0.3512, simple_loss=0.4015, pruned_loss=0.1504, over 5631884.94 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3633, pruned_loss=0.105, over 5540433.13 frames. ], giga_tot_loss[loss=0.3556, simple_loss=0.4039, pruned_loss=0.1537, over 5621567.58 frames. ], batch size: 705, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:47:56,128 INFO [optim.py:369] (1/2) 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 14:48:33,785 INFO [train.py:968] (1/2) Epoch 9, batch 6650, giga_loss[loss=0.3709, simple_loss=0.421, pruned_loss=0.1604, over 27885.00 frames. ], tot_loss[loss=0.3485, simple_loss=0.3999, pruned_loss=0.1486, over 5623417.93 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3635, pruned_loss=0.1051, over 5536238.25 frames. ], giga_tot_loss[loss=0.3533, simple_loss=0.4024, pruned_loss=0.1521, over 5621470.73 frames. ], batch size: 412, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:49:17,944 INFO [train.py:968] (1/2) Epoch 9, batch 6700, giga_loss[loss=0.351, simple_loss=0.409, pruned_loss=0.1465, over 28715.00 frames. ], tot_loss[loss=0.3479, simple_loss=0.4002, pruned_loss=0.1478, over 5635038.04 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3632, pruned_loss=0.1049, over 5543542.09 frames. ], giga_tot_loss[loss=0.3534, simple_loss=0.4033, pruned_loss=0.1517, over 5628898.03 frames. ], batch size: 262, lr: 3.68e-03, grad_scale: 2.0
+2023-03-04 14:49:28,211 INFO [optim.py:369] (1/2) 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 14:49:42,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3124, 1.4006, 1.1230, 1.1916], device='cuda:1'), covar=tensor([0.0965, 0.0922, 0.0827, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1506, 0.1475, 0.1575], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 14:50:00,846 INFO [train.py:968] (1/2) Epoch 9, batch 6750, giga_loss[loss=0.3281, simple_loss=0.3868, pruned_loss=0.1347, over 28859.00 frames. ], tot_loss[loss=0.3475, simple_loss=0.4005, pruned_loss=0.1473, over 5619128.50 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3632, pruned_loss=0.1049, over 5542279.74 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4043, pruned_loss=0.152, over 5619122.06 frames. ], batch size: 186, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:50:49,125 INFO [train.py:968] (1/2) Epoch 9, batch 6800, giga_loss[loss=0.3487, simple_loss=0.3986, pruned_loss=0.1494, over 27924.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3986, pruned_loss=0.1457, over 5611409.97 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3631, pruned_loss=0.1048, over 5547530.32 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4026, pruned_loss=0.1506, over 5608463.08 frames. ], batch size: 412, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:51:00,412 INFO [optim.py:369] (1/2) 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 14:51:02,322 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.18 vs. limit=5.0
+2023-03-04 14:51:25,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-03-04 14:51:38,754 INFO [train.py:968] (1/2) Epoch 9, batch 6850, libri_loss[loss=0.281, simple_loss=0.3583, pruned_loss=0.1019, over 29668.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3965, pruned_loss=0.143, over 5616840.60 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3631, pruned_loss=0.1048, over 5551073.91 frames. ], giga_tot_loss[loss=0.3472, simple_loss=0.4, pruned_loss=0.1472, over 5611678.18 frames. ], batch size: 88, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:52:22,793 INFO [train.py:968] (1/2) Epoch 9, batch 6900, giga_loss[loss=0.3106, simple_loss=0.3706, pruned_loss=0.1253, over 28919.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3928, pruned_loss=0.1389, over 5632301.70 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3629, pruned_loss=0.1047, over 5555658.65 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3962, pruned_loss=0.1429, over 5624975.62 frames. ], batch size: 106, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:52:30,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7235, 1.8337, 1.4382, 1.5694], device='cuda:1'), covar=tensor([0.1350, 0.1776, 0.1800, 0.1764], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0721, 0.0647, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 14:52:34,772 INFO [optim.py:369] (1/2) 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 14:53:07,684 INFO [train.py:968] (1/2) Epoch 9, batch 6950, giga_loss[loss=0.3081, simple_loss=0.3719, pruned_loss=0.1221, over 28895.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3905, pruned_loss=0.137, over 5641658.10 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3631, pruned_loss=0.1048, over 5560054.78 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3935, pruned_loss=0.1405, over 5632799.55 frames. ], batch size: 145, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:53:53,017 INFO [train.py:968] (1/2) Epoch 9, batch 7000, giga_loss[loss=0.3095, simple_loss=0.3695, pruned_loss=0.1247, over 28816.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.387, pruned_loss=0.134, over 5654099.07 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3632, pruned_loss=0.1049, over 5570511.81 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3901, pruned_loss=0.1377, over 5639883.95 frames. ], batch size: 186, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:54:04,591 INFO [optim.py:369] (1/2) 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 14:54:34,971 INFO [train.py:968] (1/2) Epoch 9, batch 7050, giga_loss[loss=0.3204, simple_loss=0.3822, pruned_loss=0.1293, over 28553.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3855, pruned_loss=0.1329, over 5652565.38 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3631, pruned_loss=0.105, over 5569819.92 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3887, pruned_loss=0.1365, over 5642871.57 frames. ], batch size: 307, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:55:25,783 INFO [train.py:968] (1/2) Epoch 9, batch 7100, giga_loss[loss=0.3348, simple_loss=0.4068, pruned_loss=0.1314, over 28673.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3861, pruned_loss=0.1333, over 5650928.91 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3632, pruned_loss=0.1051, over 5564742.51 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3888, pruned_loss=0.1364, over 5648541.77 frames. ], batch size: 92, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:55:39,644 INFO [optim.py:369] (1/2) 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 14:55:50,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8263, 1.9320, 1.5899, 2.1062], device='cuda:1'), covar=tensor([0.2167, 0.2179, 0.2455, 0.2156], device='cuda:1'), in_proj_covar=tensor([0.1222, 0.0906, 0.1081, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 14:56:00,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0536, 1.1935, 3.7407, 3.1134], device='cuda:1'), covar=tensor([0.1627, 0.2401, 0.0387, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0567, 0.0815, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 14:56:14,344 INFO [train.py:968] (1/2) Epoch 9, batch 7150, giga_loss[loss=0.3209, simple_loss=0.4029, pruned_loss=0.1194, over 28879.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3834, pruned_loss=0.1302, over 5665114.89 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3634, pruned_loss=0.1052, over 5574931.17 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3859, pruned_loss=0.1333, over 5656295.81 frames. ], batch size: 174, lr: 3.67e-03, grad_scale: 2.0
+2023-03-04 14:57:00,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3218, 1.4596, 1.1883, 1.1006], device='cuda:1'), covar=tensor([0.1413, 0.1352, 0.1055, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1501, 0.1461, 0.1567], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 14:57:04,746 INFO [train.py:968] (1/2) Epoch 9, batch 7200, giga_loss[loss=0.3445, simple_loss=0.4027, pruned_loss=0.1431, over 28712.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3844, pruned_loss=0.1289, over 5651008.18 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3631, pruned_loss=0.1051, over 5564639.28 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3871, pruned_loss=0.1321, over 5656745.63 frames. ], batch size: 262, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:57:20,469 INFO [optim.py:369] (1/2) 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 14:57:50,683 INFO [train.py:968] (1/2) Epoch 9, batch 7250, giga_loss[loss=0.3247, simple_loss=0.3706, pruned_loss=0.1393, over 23662.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3863, pruned_loss=0.1293, over 5655693.99 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3629, pruned_loss=0.1049, over 5572292.69 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3892, pruned_loss=0.1325, over 5655321.56 frames. ], batch size: 705, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:58:43,107 INFO [train.py:968] (1/2) Epoch 9, batch 7300, giga_loss[loss=0.3373, simple_loss=0.3931, pruned_loss=0.1408, over 28805.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3874, pruned_loss=0.131, over 5663729.84 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3628, pruned_loss=0.1049, over 5574003.23 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3898, pruned_loss=0.1336, over 5662266.68 frames. ], batch size: 199, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:58:55,251 INFO [optim.py:369] (1/2) 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 14:58:56,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-04 14:59:26,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 14:59:28,690 INFO [train.py:968] (1/2) Epoch 9, batch 7350, giga_loss[loss=0.3015, simple_loss=0.3691, pruned_loss=0.117, over 28706.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3868, pruned_loss=0.131, over 5667927.95 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3633, pruned_loss=0.1052, over 5578897.12 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3887, pruned_loss=0.1332, over 5663521.97 frames. ], batch size: 242, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 14:59:50,065 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.23 vs. limit=5.0
+2023-03-04 15:00:18,615 INFO [train.py:968] (1/2) Epoch 9, batch 7400, giga_loss[loss=0.3193, simple_loss=0.3559, pruned_loss=0.1413, over 23640.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.385, pruned_loss=0.1317, over 5658941.28 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3634, pruned_loss=0.1054, over 5583554.98 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3868, pruned_loss=0.1336, over 5652625.09 frames. ], batch size: 705, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:00:29,472 INFO [optim.py:369] (1/2) 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 15:00:37,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-04 15:00:54,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1827, 1.4271, 1.2492, 1.1042], device='cuda:1'), covar=tensor([0.1499, 0.1326, 0.0982, 0.1314], device='cuda:1'), in_proj_covar=tensor([0.1625, 0.1502, 0.1461, 0.1572], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 15:01:03,748 INFO [train.py:968] (1/2) Epoch 9, batch 7450, giga_loss[loss=0.3136, simple_loss=0.3721, pruned_loss=0.1276, over 28704.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3832, pruned_loss=0.1308, over 5672345.30 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3634, pruned_loss=0.1054, over 5583554.98 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3846, pruned_loss=0.1323, over 5667429.33 frames. ], batch size: 71, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:01:27,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-04 15:01:33,110 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=371679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:01:47,233 INFO [train.py:968] (1/2) Epoch 9, batch 7500, giga_loss[loss=0.3081, simple_loss=0.3814, pruned_loss=0.1175, over 28709.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.382, pruned_loss=0.1284, over 5681081.36 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3635, pruned_loss=0.1056, over 5585712.21 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3838, pruned_loss=0.1305, over 5678155.74 frames. ], batch size: 242, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:01:59,105 INFO [optim.py:369] (1/2) 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 15:02:00,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-04 15:02:30,463 INFO [train.py:968] (1/2) Epoch 9, batch 7550, giga_loss[loss=0.3056, simple_loss=0.3777, pruned_loss=0.1168, over 28974.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3805, pruned_loss=0.1261, over 5690133.72 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3634, pruned_loss=0.1055, over 5591564.99 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3825, pruned_loss=0.1284, over 5684942.23 frames. ], batch size: 128, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:02:35,795 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=371755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:03:12,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-04 15:03:12,380 INFO [train.py:968] (1/2) Epoch 9, batch 7600, giga_loss[loss=0.3973, simple_loss=0.4238, pruned_loss=0.1854, over 26668.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3805, pruned_loss=0.1262, over 5685034.21 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3638, pruned_loss=0.1059, over 5595051.13 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3823, pruned_loss=0.1284, over 5681735.44 frames. ], batch size: 555, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:03:22,858 INFO [optim.py:369] (1/2) 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 15:03:55,764 INFO [train.py:968] (1/2) Epoch 9, batch 7650, giga_loss[loss=0.3126, simple_loss=0.3753, pruned_loss=0.125, over 28937.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3803, pruned_loss=0.1265, over 5688235.92 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3642, pruned_loss=0.1061, over 5599045.63 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3817, pruned_loss=0.1285, over 5683729.17 frames. ], batch size: 227, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:04:44,290 INFO [train.py:968] (1/2) Epoch 9, batch 7700, giga_loss[loss=0.3268, simple_loss=0.3911, pruned_loss=0.1313, over 28650.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3788, pruned_loss=0.1261, over 5694008.82 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3643, pruned_loss=0.1061, over 5604902.98 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3802, pruned_loss=0.1281, over 5686746.59 frames. ], batch size: 307, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:04:56,915 INFO [optim.py:369] (1/2) 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 15:05:30,215 INFO [train.py:968] (1/2) Epoch 9, batch 7750, giga_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 28985.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3786, pruned_loss=0.1272, over 5686223.19 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.364, pruned_loss=0.106, over 5607824.21 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.38, pruned_loss=0.129, over 5678691.80 frames. ], batch size: 128, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:06:13,439 INFO [train.py:968] (1/2) Epoch 9, batch 7800, giga_loss[loss=0.3101, simple_loss=0.3739, pruned_loss=0.1232, over 28893.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3774, pruned_loss=0.1265, over 5698814.74 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3638, pruned_loss=0.1057, over 5613730.73 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3792, pruned_loss=0.1288, over 5689991.63 frames. ], batch size: 285, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:06:26,950 INFO [optim.py:369] (1/2) 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 15:06:57,096 INFO [train.py:968] (1/2) Epoch 9, batch 7850, giga_loss[loss=0.3411, simple_loss=0.3962, pruned_loss=0.143, over 28491.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3749, pruned_loss=0.1254, over 5699303.19 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3632, pruned_loss=0.1053, over 5618807.06 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.377, pruned_loss=0.1281, over 5689474.17 frames. ], batch size: 336, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:07:02,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:07:24,207 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 15:07:40,381 INFO [train.py:968] (1/2) Epoch 9, batch 7900, giga_loss[loss=0.2897, simple_loss=0.3592, pruned_loss=0.1101, over 29162.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3752, pruned_loss=0.1264, over 5704420.50 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3631, pruned_loss=0.1053, over 5626346.10 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3774, pruned_loss=0.1291, over 5691987.48 frames. ], batch size: 113, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:07:46,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-04 15:07:51,722 INFO [optim.py:369] (1/2) 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 15:08:07,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:08:25,248 INFO [train.py:968] (1/2) Epoch 9, batch 7950, giga_loss[loss=0.3249, simple_loss=0.3858, pruned_loss=0.132, over 28774.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.375, pruned_loss=0.126, over 5700384.34 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.363, pruned_loss=0.1051, over 5633176.32 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3772, pruned_loss=0.1288, over 5686086.93 frames. ], batch size: 284, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:09:10,393 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=372197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:09:10,667 INFO [train.py:968] (1/2) Epoch 9, batch 8000, giga_loss[loss=0.306, simple_loss=0.3701, pruned_loss=0.1209, over 28819.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3761, pruned_loss=0.1261, over 5693632.97 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3631, pruned_loss=0.1052, over 5637252.14 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3779, pruned_loss=0.1285, over 5679636.07 frames. ], batch size: 243, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:09:13,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=372200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:09:22,934 INFO [optim.py:369] (1/2) 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 15:09:38,188 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 8050, giga_loss[loss=0.4065, simple_loss=0.429, pruned_loss=0.192, over 26506.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3758, pruned_loss=0.1248, over 5690099.25 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3632, pruned_loss=0.1052, over 5642860.10 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3775, pruned_loss=0.1274, over 5674995.01 frames. ], batch size: 555, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:10:18,708 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=372273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:20,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=372276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:38,662 INFO [train.py:968] (1/2) Epoch 9, batch 8100, libri_loss[loss=0.2691, simple_loss=0.3593, pruned_loss=0.08941, over 29546.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3764, pruned_loss=0.125, over 5678265.60 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3631, pruned_loss=0.105, over 5637893.53 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3781, pruned_loss=0.1276, over 5671673.61 frames. ], batch size: 84, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:10:40,854 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=372301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:45,615 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=372305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:10:51,232 INFO [optim.py:369] (1/2) 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 15:11:25,331 INFO [train.py:968] (1/2) Epoch 9, batch 8150, giga_loss[loss=0.3885, simple_loss=0.4284, pruned_loss=0.1744, over 28601.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3782, pruned_loss=0.1266, over 5680261.33 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3632, pruned_loss=0.1052, over 5636829.77 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3798, pruned_loss=0.1292, over 5677142.28 frames. ], batch size: 336, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:12:17,829 INFO [train.py:968] (1/2) Epoch 9, batch 8200, giga_loss[loss=0.3235, simple_loss=0.3803, pruned_loss=0.1333, over 28966.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3806, pruned_loss=0.1301, over 5676727.59 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3629, pruned_loss=0.105, over 5639715.09 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3824, pruned_loss=0.1325, over 5671969.67 frames. ], batch size: 227, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:12:31,790 INFO [optim.py:369] (1/2) 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 15:13:01,054 INFO [train.py:968] (1/2) Epoch 9, batch 8250, giga_loss[loss=0.3391, simple_loss=0.3909, pruned_loss=0.1436, over 28653.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3816, pruned_loss=0.1318, over 5678035.88 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3625, pruned_loss=0.1047, over 5644979.12 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3838, pruned_loss=0.1344, over 5669980.32 frames. ], batch size: 307, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:13:12,216 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=372458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:13:48,883 INFO [train.py:968] (1/2) Epoch 9, batch 8300, giga_loss[loss=0.3237, simple_loss=0.3792, pruned_loss=0.1341, over 28889.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3838, pruned_loss=0.1342, over 5674212.01 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3628, pruned_loss=0.1049, over 5648899.12 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3858, pruned_loss=0.137, over 5664535.09 frames. ], batch size: 112, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:13:58,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1195, 1.1184, 3.5574, 2.9760], device='cuda:1'), covar=tensor([0.1546, 0.2407, 0.0455, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0626, 0.0572, 0.0827, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 15:14:02,151 INFO [optim.py:369] (1/2) 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 15:14:18,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6244, 1.9427, 1.9474, 1.4737], device='cuda:1'), covar=tensor([0.1423, 0.1832, 0.1097, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0707, 0.0832, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 15:14:21,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2353, 1.9084, 1.5306, 0.3590], device='cuda:1'), covar=tensor([0.2363, 0.1592, 0.2305, 0.3146], device='cuda:1'), in_proj_covar=tensor([0.1513, 0.1429, 0.1469, 0.1232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 15:14:33,424 INFO [train.py:968] (1/2) Epoch 9, batch 8350, giga_loss[loss=0.271, simple_loss=0.3483, pruned_loss=0.09683, over 28897.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3825, pruned_loss=0.1337, over 5675777.38 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3625, pruned_loss=0.1048, over 5655297.20 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3849, pruned_loss=0.1367, over 5662392.79 frames. ], batch size: 174, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:15:14,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2141, 1.5020, 1.2148, 1.0376], device='cuda:1'), covar=tensor([0.2375, 0.2267, 0.2622, 0.2071], device='cuda:1'), in_proj_covar=tensor([0.1235, 0.0922, 0.1092, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 15:15:17,451 INFO [train.py:968] (1/2) Epoch 9, batch 8400, giga_loss[loss=0.3088, simple_loss=0.3864, pruned_loss=0.1156, over 28847.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3819, pruned_loss=0.1325, over 5676120.11 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3623, pruned_loss=0.1047, over 5657775.21 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3842, pruned_loss=0.1353, over 5663462.88 frames. ], batch size: 186, lr: 3.67e-03, grad_scale: 8.0
+2023-03-04 15:15:29,865 INFO [optim.py:369] (1/2) 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 15:15:39,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1663, 1.5839, 1.5125, 1.0945], device='cuda:1'), covar=tensor([0.1563, 0.2160, 0.1298, 0.1489], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0708, 0.0833, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 15:16:00,697 INFO [train.py:968] (1/2) Epoch 9, batch 8450, giga_loss[loss=0.3365, simple_loss=0.3919, pruned_loss=0.1406, over 28767.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3807, pruned_loss=0.1305, over 5671292.08 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.362, pruned_loss=0.1043, over 5664346.22 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3836, pruned_loss=0.134, over 5655523.32 frames. ], batch size: 284, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:16:23,947 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:16:42,877 INFO [train.py:968] (1/2) Epoch 9, batch 8500, giga_loss[loss=0.3781, simple_loss=0.4097, pruned_loss=0.1733, over 27640.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3796, pruned_loss=0.1296, over 5678211.66 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3624, pruned_loss=0.1045, over 5663014.71 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3816, pruned_loss=0.1324, over 5666997.77 frames. ], batch size: 474, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:16:58,714 INFO [optim.py:369] (1/2) 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 15:17:24,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1447, 2.0525, 1.5125, 1.7329], device='cuda:1'), covar=tensor([0.0648, 0.0606, 0.0921, 0.0926], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0444, 0.0493, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 15:17:25,221 INFO [train.py:968] (1/2) Epoch 9, batch 8550, giga_loss[loss=0.2789, simple_loss=0.3465, pruned_loss=0.1056, over 29011.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3776, pruned_loss=0.1292, over 5673527.52 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3626, pruned_loss=0.1046, over 5658166.83 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3793, pruned_loss=0.1318, over 5669619.48 frames. ], batch size: 164, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:18:09,220 INFO [train.py:968] (1/2) Epoch 9, batch 8600, giga_loss[loss=0.2971, simple_loss=0.3645, pruned_loss=0.1148, over 28709.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3766, pruned_loss=0.1292, over 5664487.95 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.362, pruned_loss=0.1043, over 5661227.60 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3789, pruned_loss=0.1322, over 5658712.37 frames. ], batch size: 262, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:18:25,730 INFO [optim.py:369] (1/2) 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 15:18:31,776 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=372822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:18:42,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=372833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:18:47,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 15:18:57,260 INFO [train.py:968] (1/2) Epoch 9, batch 8650, giga_loss[loss=0.3112, simple_loss=0.3869, pruned_loss=0.1177, over 29054.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3787, pruned_loss=0.1306, over 5659951.81 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.362, pruned_loss=0.1043, over 5666099.38 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3808, pruned_loss=0.1335, over 5650814.18 frames. ], batch size: 128, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:18:59,546 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=372851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:19:41,980 INFO [train.py:968] (1/2) Epoch 9, batch 8700, giga_loss[loss=0.3273, simple_loss=0.4043, pruned_loss=0.1251, over 28979.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3822, pruned_loss=0.1304, over 5665284.23 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3623, pruned_loss=0.1045, over 5670053.84 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.384, pruned_loss=0.133, over 5654317.61 frames. ], batch size: 164, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:19:54,456 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 8750, libri_loss[loss=0.2838, simple_loss=0.3598, pruned_loss=0.1039, over 29551.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3843, pruned_loss=0.1295, over 5678875.91 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3623, pruned_loss=0.1044, over 5674009.10 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3861, pruned_loss=0.1321, over 5666529.94 frames. ], batch size: 83, lr: 3.67e-03, grad_scale: 4.0
+2023-03-04 15:20:33,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1357, 1.5480, 1.5026, 1.0800], device='cuda:1'), covar=tensor([0.1403, 0.2011, 0.1139, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0708, 0.0837, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 15:20:53,855 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=372976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:20:55,698 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 8800, giga_loss[loss=0.2816, simple_loss=0.3585, pruned_loss=0.1023, over 28933.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3864, pruned_loss=0.1314, over 5677277.02 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3624, pruned_loss=0.1045, over 5676331.28 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.388, pruned_loss=0.1337, over 5665451.91 frames. ], batch size: 199, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:21:18,500 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=373008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:21:23,594 INFO [optim.py:369] (1/2) 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 15:21:54,138 INFO [train.py:968] (1/2) Epoch 9, batch 8850, giga_loss[loss=0.3232, simple_loss=0.3852, pruned_loss=0.1306, over 29013.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3883, pruned_loss=0.1334, over 5667930.85 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3624, pruned_loss=0.1044, over 5679445.60 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3902, pruned_loss=0.1361, over 5655564.03 frames. ], batch size: 213, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:22:26,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:22:33,714 INFO [train.py:968] (1/2) Epoch 9, batch 8900, libri_loss[loss=0.2844, simple_loss=0.3602, pruned_loss=0.1044, over 29542.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.387, pruned_loss=0.133, over 5672539.57 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3619, pruned_loss=0.1041, over 5685085.70 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3899, pruned_loss=0.1364, over 5657109.15 frames. ], batch size: 78, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:22:45,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8028, 1.7787, 1.6367, 1.5779], device='cuda:1'), covar=tensor([0.1164, 0.1959, 0.1672, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0731, 0.0658, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 15:22:49,029 INFO [optim.py:369] (1/2) 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 15:23:00,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 15:23:19,561 INFO [train.py:968] (1/2) Epoch 9, batch 8950, giga_loss[loss=0.2975, simple_loss=0.3642, pruned_loss=0.1154, over 28797.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3854, pruned_loss=0.1332, over 5655403.23 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3616, pruned_loss=0.1039, over 5690261.39 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3886, pruned_loss=0.1368, over 5637824.85 frames. ], batch size: 186, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:24:02,679 INFO [train.py:968] (1/2) Epoch 9, batch 9000, giga_loss[loss=0.315, simple_loss=0.377, pruned_loss=0.1265, over 28812.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3833, pruned_loss=0.132, over 5659960.33 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.362, pruned_loss=0.1041, over 5692691.66 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3861, pruned_loss=0.1356, over 5642686.69 frames. ], batch size: 112, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:24:02,679 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 15:24:10,971 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 15:24:24,251 INFO [optim.py:369] (1/2) 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 15:24:54,075 INFO [train.py:968] (1/2) Epoch 9, batch 9050, giga_loss[loss=0.3794, simple_loss=0.4129, pruned_loss=0.173, over 28486.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3823, pruned_loss=0.1319, over 5665065.04 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3618, pruned_loss=0.104, over 5695273.61 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3851, pruned_loss=0.1353, over 5648261.71 frames. ], batch size: 85, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:25:00,977 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 15:25:32,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-04 15:25:37,579 INFO [train.py:968] (1/2) Epoch 9, batch 9100, giga_loss[loss=0.4147, simple_loss=0.4436, pruned_loss=0.1929, over 26509.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3828, pruned_loss=0.1329, over 5666016.69 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3616, pruned_loss=0.104, over 5701940.85 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.386, pruned_loss=0.1366, over 5645358.66 frames. ], batch size: 555, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:25:44,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3281, 5.1529, 4.8558, 2.5578], device='cuda:1'), covar=tensor([0.0402, 0.0550, 0.0609, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.1000, 0.0937, 0.0828, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 15:25:53,552 INFO [optim.py:369] (1/2) 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 15:25:59,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-04 15:26:05,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:26:22,185 INFO [train.py:968] (1/2) Epoch 9, batch 9150, giga_loss[loss=0.343, simple_loss=0.3916, pruned_loss=0.1472, over 28748.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3822, pruned_loss=0.1329, over 5654036.84 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3615, pruned_loss=0.104, over 5695674.80 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3851, pruned_loss=0.1362, over 5642244.22 frames. ], batch size: 262, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:27:06,912 INFO [train.py:968] (1/2) Epoch 9, batch 9200, giga_loss[loss=0.3049, simple_loss=0.3662, pruned_loss=0.1217, over 28414.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3798, pruned_loss=0.1313, over 5656214.78 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3622, pruned_loss=0.1043, over 5692283.76 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3821, pruned_loss=0.1346, over 5648585.57 frames. ], batch size: 65, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:27:13,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0547, 1.7762, 1.4516, 1.4638], device='cuda:1'), covar=tensor([0.0681, 0.0700, 0.0888, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0451, 0.0499, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 15:27:23,595 INFO [optim.py:369] (1/2) 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 15:27:43,946 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:27:53,017 INFO [train.py:968] (1/2) Epoch 9, batch 9250, giga_loss[loss=0.2789, simple_loss=0.3538, pruned_loss=0.102, over 29024.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3782, pruned_loss=0.13, over 5640730.91 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3622, pruned_loss=0.1043, over 5683595.67 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3802, pruned_loss=0.1329, over 5642716.87 frames. ], batch size: 136, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:28:04,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=373462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:28:34,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5608, 1.7815, 1.9434, 1.4223], device='cuda:1'), covar=tensor([0.1425, 0.2119, 0.1158, 0.1483], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0709, 0.0837, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 15:28:37,597 INFO [train.py:968] (1/2) Epoch 9, batch 9300, libri_loss[loss=0.2227, simple_loss=0.2997, pruned_loss=0.07285, over 29377.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3783, pruned_loss=0.1284, over 5656976.37 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3618, pruned_loss=0.1039, over 5691111.80 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3808, pruned_loss=0.132, over 5650573.95 frames. ], batch size: 67, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:28:52,364 INFO [optim.py:369] (1/2) 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 15:29:17,996 INFO [train.py:968] (1/2) Epoch 9, batch 9350, giga_loss[loss=0.2903, simple_loss=0.3589, pruned_loss=0.1108, over 28896.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3813, pruned_loss=0.1304, over 5655154.16 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3616, pruned_loss=0.1039, over 5685724.59 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.384, pruned_loss=0.1338, over 5653476.44 frames. ], batch size: 186, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:29:18,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1352, 2.5880, 1.1534, 1.2741], device='cuda:1'), covar=tensor([0.0978, 0.0310, 0.0877, 0.1408], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0500, 0.0327, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 15:29:20,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5572, 1.7512, 1.4412, 1.7801], device='cuda:1'), covar=tensor([0.2256, 0.2187, 0.2289, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.1237, 0.0926, 0.1091, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 15:29:39,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3322, 3.0678, 1.4797, 1.4522], device='cuda:1'), covar=tensor([0.0877, 0.0299, 0.0775, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0500, 0.0328, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 15:30:03,357 INFO [train.py:968] (1/2) Epoch 9, batch 9400, libri_loss[loss=0.3044, simple_loss=0.3929, pruned_loss=0.108, over 29505.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3817, pruned_loss=0.1312, over 5656321.03 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3614, pruned_loss=0.1037, over 5691463.40 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3845, pruned_loss=0.1349, over 5648592.92 frames. ], batch size: 84, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:30:10,608 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=373605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:30:12,540 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:30:38,304 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=373637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:30:46,661 INFO [train.py:968] (1/2) Epoch 9, batch 9450, giga_loss[loss=0.2871, simple_loss=0.371, pruned_loss=0.1017, over 29056.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3837, pruned_loss=0.1302, over 5665345.53 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3613, pruned_loss=0.1036, over 5695065.63 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3866, pruned_loss=0.1338, over 5655057.26 frames. ], batch size: 128, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:31:27,809 INFO [train.py:968] (1/2) Epoch 9, batch 9500, giga_loss[loss=0.3237, simple_loss=0.3915, pruned_loss=0.128, over 28695.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.385, pruned_loss=0.1293, over 5671977.67 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3617, pruned_loss=0.1038, over 5700532.86 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3877, pruned_loss=0.1328, over 5657700.92 frames. ], batch size: 242, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:31:33,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=373704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:31:33,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3818, 1.6129, 1.6714, 1.2698], device='cuda:1'), covar=tensor([0.1419, 0.1956, 0.1157, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0707, 0.0835, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 15:31:43,500 INFO [optim.py:369] (1/2) 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 15:31:53,425 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-04 15:32:04,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-04 15:32:11,523 INFO [train.py:968] (1/2) Epoch 9, batch 9550, giga_loss[loss=0.29, simple_loss=0.373, pruned_loss=0.1035, over 28758.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.389, pruned_loss=0.1313, over 5668350.17 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3621, pruned_loss=0.1042, over 5693734.67 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.391, pruned_loss=0.1341, over 5662868.38 frames. ], batch size: 92, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:32:57,374 INFO [train.py:968] (1/2) Epoch 9, batch 9600, giga_loss[loss=0.3341, simple_loss=0.3905, pruned_loss=0.1389, over 28973.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3929, pruned_loss=0.1353, over 5672683.25 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3624, pruned_loss=0.1044, over 5695946.09 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3945, pruned_loss=0.1377, over 5666223.18 frames. ], batch size: 227, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:33:08,110 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:33:38,407 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=373847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:33:38,742 INFO [train.py:968] (1/2) Epoch 9, batch 9650, libri_loss[loss=0.3404, simple_loss=0.3961, pruned_loss=0.1424, over 27770.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3946, pruned_loss=0.138, over 5663379.22 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3624, pruned_loss=0.1046, over 5697817.06 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3969, pruned_loss=0.1407, over 5655721.87 frames. ], batch size: 116, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:33:40,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=373850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:33:47,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1022, 1.1958, 3.4599, 3.0149], device='cuda:1'), covar=tensor([0.1496, 0.2319, 0.0424, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0626, 0.0573, 0.0832, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 15:33:54,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1953, 1.5738, 1.3025, 0.9269], device='cuda:1'), covar=tensor([0.1703, 0.1619, 0.1632, 0.1549], device='cuda:1'), in_proj_covar=tensor([0.1238, 0.0928, 0.1093, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 15:34:06,247 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=373879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:34:10,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3913, 2.5606, 1.5045, 1.5413], device='cuda:1'), covar=tensor([0.0734, 0.0320, 0.0636, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0502, 0.0328, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 15:34:21,752 INFO [train.py:968] (1/2) Epoch 9, batch 9700, giga_loss[loss=0.3718, simple_loss=0.4206, pruned_loss=0.1615, over 28604.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3935, pruned_loss=0.1379, over 5665551.69 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3619, pruned_loss=0.1043, over 5702197.26 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3968, pruned_loss=0.1413, over 5654272.28 frames. ], batch size: 307, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:34:39,664 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:34:59,395 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=373938.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:35:06,132 INFO [train.py:968] (1/2) Epoch 9, batch 9750, libri_loss[loss=0.2733, simple_loss=0.3546, pruned_loss=0.09604, over 29507.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3916, pruned_loss=0.1363, over 5670999.95 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3618, pruned_loss=0.1043, over 5705240.94 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3946, pruned_loss=0.1394, over 5658767.46 frames. ], batch size: 81, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:35:11,081 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=373955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:35:14,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=373958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:35:23,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5571, 4.5604, 1.6898, 1.6399], device='cuda:1'), covar=tensor([0.0942, 0.0276, 0.0883, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0501, 0.0326, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 15:35:40,004 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 9800, giga_loss[loss=0.3466, simple_loss=0.4094, pruned_loss=0.1419, over 28726.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3902, pruned_loss=0.1332, over 5668256.91 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3618, pruned_loss=0.1044, over 5698017.48 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3929, pruned_loss=0.1359, over 5664303.10 frames. ], batch size: 262, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:35:56,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:36:05,086 INFO [optim.py:369] (1/2) 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 15:36:25,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-04 15:36:30,725 INFO [train.py:968] (1/2) Epoch 9, batch 9850, libri_loss[loss=0.3586, simple_loss=0.4087, pruned_loss=0.1543, over 19824.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3901, pruned_loss=0.1321, over 5661867.83 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3621, pruned_loss=0.1046, over 5690899.72 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3925, pruned_loss=0.1346, over 5665322.73 frames. ], batch size: 186, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:37:01,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3731, 3.1838, 3.0014, 1.9780], device='cuda:1'), covar=tensor([0.0703, 0.0849, 0.0774, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.0999, 0.0941, 0.0829, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 15:37:13,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9740, 1.8297, 1.3943, 1.4409], device='cuda:1'), covar=tensor([0.0627, 0.0595, 0.0920, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0448, 0.0498, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 15:37:16,841 INFO [train.py:968] (1/2) Epoch 9, batch 9900, giga_loss[loss=0.2782, simple_loss=0.3571, pruned_loss=0.09965, over 28427.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.39, pruned_loss=0.1328, over 5664571.59 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3611, pruned_loss=0.104, over 5696386.68 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3935, pruned_loss=0.136, over 5661558.59 frames. ], batch size: 71, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:37:24,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4063, 1.6198, 1.3732, 1.4847], device='cuda:1'), covar=tensor([0.0746, 0.0306, 0.0318, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0074, 0.0053, 0.0048, 0.0080], device='cuda:1')
+2023-03-04 15:37:27,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7683, 1.7571, 1.2327, 1.3699], device='cuda:1'), covar=tensor([0.0702, 0.0622, 0.0983, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0450, 0.0499, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 15:37:28,224 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:37:34,220 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:37:37,350 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 9950, giga_loss[loss=0.3488, simple_loss=0.394, pruned_loss=0.1518, over 27625.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.389, pruned_loss=0.1329, over 5659881.20 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3605, pruned_loss=0.1035, over 5699492.79 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3931, pruned_loss=0.1366, over 5653801.41 frames. ], batch size: 472, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:38:37,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-04 15:38:48,112 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-04 15:38:49,134 INFO [train.py:968] (1/2) Epoch 9, batch 10000, libri_loss[loss=0.2945, simple_loss=0.3691, pruned_loss=0.1099, over 29535.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.388, pruned_loss=0.1336, over 5654517.92 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3607, pruned_loss=0.1036, over 5701925.61 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3916, pruned_loss=0.137, over 5646538.53 frames. ], batch size: 82, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:39:04,982 INFO [optim.py:369] (1/2) 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 15:39:24,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-04 15:39:28,300 INFO [train.py:968] (1/2) Epoch 9, batch 10050, giga_loss[loss=0.2827, simple_loss=0.353, pruned_loss=0.1062, over 28934.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3851, pruned_loss=0.1317, over 5666134.18 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3606, pruned_loss=0.1036, over 5702244.97 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3895, pruned_loss=0.136, over 5657238.87 frames. ], batch size: 145, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:39:58,980 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374282.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:40:12,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:40:12,531 INFO [train.py:968] (1/2) Epoch 9, batch 10100, giga_loss[loss=0.3423, simple_loss=0.3955, pruned_loss=0.1445, over 28586.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.384, pruned_loss=0.1318, over 5662201.75 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3607, pruned_loss=0.1036, over 5708388.64 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3881, pruned_loss=0.136, over 5648580.36 frames. ], batch size: 336, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:40:26,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374313.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:40:31,289 INFO [optim.py:369] (1/2) 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 15:40:54,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8105, 2.3807, 1.6183, 1.1312], device='cuda:1'), covar=tensor([0.3717, 0.2356, 0.2219, 0.3394], device='cuda:1'), in_proj_covar=tensor([0.1504, 0.1429, 0.1458, 0.1223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 15:40:59,176 INFO [train.py:968] (1/2) Epoch 9, batch 10150, giga_loss[loss=0.3062, simple_loss=0.3617, pruned_loss=0.1254, over 28508.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3837, pruned_loss=0.1323, over 5655612.01 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3611, pruned_loss=0.1038, over 5701633.14 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3869, pruned_loss=0.1359, over 5650228.97 frames. ], batch size: 85, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:41:00,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:41:10,730 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-04 15:41:30,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:41:33,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0578, 1.4564, 1.3888, 1.0387], device='cuda:1'), covar=tensor([0.1131, 0.1693, 0.0943, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0712, 0.0841, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 15:41:43,557 INFO [train.py:968] (1/2) Epoch 9, batch 10200, libri_loss[loss=0.2854, simple_loss=0.3537, pruned_loss=0.1086, over 29569.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3827, pruned_loss=0.1318, over 5651352.05 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3614, pruned_loss=0.1041, over 5695223.96 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3854, pruned_loss=0.1349, over 5651281.53 frames. ], batch size: 76, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:42:01,512 INFO [optim.py:369] (1/2) 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 15:42:23,944 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:42:25,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:42:29,624 INFO [train.py:968] (1/2) Epoch 9, batch 10250, giga_loss[loss=0.3316, simple_loss=0.3926, pruned_loss=0.1354, over 28945.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3803, pruned_loss=0.1285, over 5661477.16 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3616, pruned_loss=0.1043, over 5696300.13 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3827, pruned_loss=0.1314, over 5659755.26 frames. ], batch size: 164, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:42:35,214 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-04 15:42:37,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374456.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:42:39,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374459.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:42:51,740 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:00,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:04,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374488.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:43:06,587 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 10300, giga_loss[loss=0.2844, simple_loss=0.3519, pruned_loss=0.1085, over 28688.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3755, pruned_loss=0.1242, over 5658201.17 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3611, pruned_loss=0.1038, over 5702728.12 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3784, pruned_loss=0.1276, over 5649967.46 frames. ], batch size: 92, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:43:28,083 INFO [optim.py:369] (1/2) 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 15:43:35,856 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:39,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:43:44,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9137, 2.8687, 1.9386, 0.8950], device='cuda:1'), covar=tensor([0.4727, 0.2183, 0.2625, 0.4613], device='cuda:1'), in_proj_covar=tensor([0.1515, 0.1436, 0.1471, 0.1232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 15:43:57,564 INFO [train.py:968] (1/2) Epoch 9, batch 10350, giga_loss[loss=0.2845, simple_loss=0.3503, pruned_loss=0.1094, over 28761.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3754, pruned_loss=0.1236, over 5663267.15 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3611, pruned_loss=0.1038, over 5701952.23 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3781, pruned_loss=0.1269, over 5656196.19 frames. ], batch size: 99, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:44:02,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1457, 3.5304, 2.3986, 1.1919], device='cuda:1'), covar=tensor([0.4153, 0.1550, 0.2228, 0.4112], device='cuda:1'), in_proj_covar=tensor([0.1515, 0.1434, 0.1471, 0.1231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 15:44:03,845 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:44:05,159 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:44:43,340 INFO [train.py:968] (1/2) Epoch 9, batch 10400, giga_loss[loss=0.2768, simple_loss=0.3432, pruned_loss=0.1052, over 28947.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.1241, over 5658352.72 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3608, pruned_loss=0.1036, over 5696617.68 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3768, pruned_loss=0.1272, over 5656631.50 frames. ], batch size: 213, lr: 3.66e-03, grad_scale: 8.0
+2023-03-04 15:45:04,953 INFO [optim.py:369] (1/2) 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 15:45:09,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:11,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 15:45:13,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:16,893 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:20,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:30,143 INFO [train.py:968] (1/2) Epoch 9, batch 10450, giga_loss[loss=0.2873, simple_loss=0.3541, pruned_loss=0.1102, over 28499.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5662504.79 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3606, pruned_loss=0.1033, over 5702114.32 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3735, pruned_loss=0.1256, over 5655168.81 frames. ], batch size: 65, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:45:40,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374657.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:45:41,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:45:45,640 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:46:06,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:46:08,172 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=374694.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:46:10,546 INFO [train.py:968] (1/2) Epoch 9, batch 10500, giga_loss[loss=0.384, simple_loss=0.4261, pruned_loss=0.171, over 27564.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.372, pruned_loss=0.1224, over 5669229.95 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3607, pruned_loss=0.1033, over 5706563.52 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3744, pruned_loss=0.1257, over 5658180.72 frames. ], batch size: 472, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:46:28,861 INFO [optim.py:369] (1/2) 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 15:46:29,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 15:46:34,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:46:53,858 INFO [train.py:968] (1/2) Epoch 9, batch 10550, giga_loss[loss=0.3355, simple_loss=0.3754, pruned_loss=0.1478, over 23431.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3738, pruned_loss=0.1229, over 5660949.03 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3606, pruned_loss=0.1031, over 5703250.04 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3762, pruned_loss=0.1263, over 5654198.51 frames. ], batch size: 705, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:47:43,160 INFO [train.py:968] (1/2) Epoch 9, batch 10600, giga_loss[loss=0.308, simple_loss=0.3628, pruned_loss=0.1267, over 28683.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3744, pruned_loss=0.124, over 5638793.78 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3606, pruned_loss=0.1032, over 5685376.33 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3764, pruned_loss=0.1268, over 5648394.73 frames. ], batch size: 92, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:47:45,367 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:47:47,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374803.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:48:04,673 INFO [optim.py:369] (1/2) 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 15:48:15,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374832.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:48:29,539 INFO [train.py:968] (1/2) Epoch 9, batch 10650, giga_loss[loss=0.3036, simple_loss=0.3682, pruned_loss=0.1195, over 28879.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3736, pruned_loss=0.1236, over 5638904.05 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3609, pruned_loss=0.1034, over 5682498.68 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3754, pruned_loss=0.1263, over 5647780.37 frames. ], batch size: 186, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:48:40,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4067, 3.0425, 1.4999, 1.5370], device='cuda:1'), covar=tensor([0.0831, 0.0287, 0.0772, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0500, 0.0330, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 15:48:43,146 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=374867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:48:46,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=374870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:48:48,404 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4912, 1.7011, 1.4117, 1.8524], device='cuda:1'), covar=tensor([0.1945, 0.1827, 0.1824, 0.1825], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.0924, 0.1087, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 15:49:12,208 INFO [train.py:968] (1/2) Epoch 9, batch 10700, giga_loss[loss=0.3651, simple_loss=0.4178, pruned_loss=0.1561, over 28670.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3754, pruned_loss=0.1253, over 5642193.67 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.361, pruned_loss=0.1035, over 5681895.28 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1282, over 5648758.56 frames. ], batch size: 262, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:49:14,850 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=374899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:49:34,513 INFO [optim.py:369] (1/2) 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 15:49:43,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=374930.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:49:58,433 INFO [train.py:968] (1/2) Epoch 9, batch 10750, giga_loss[loss=0.3603, simple_loss=0.4041, pruned_loss=0.1583, over 27497.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3776, pruned_loss=0.1267, over 5635412.97 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3608, pruned_loss=0.1034, over 5677063.00 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3799, pruned_loss=0.1301, over 5644036.69 frames. ], batch size: 472, lr: 3.66e-03, grad_scale: 2.0
+2023-03-04 15:50:43,657 INFO [train.py:968] (1/2) Epoch 9, batch 10800, giga_loss[loss=0.3333, simple_loss=0.3897, pruned_loss=0.1384, over 28804.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3787, pruned_loss=0.1269, over 5655048.68 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3609, pruned_loss=0.1034, over 5683248.65 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.381, pruned_loss=0.1303, over 5655344.34 frames. ], batch size: 99, lr: 3.66e-03, grad_scale: 4.0
+2023-03-04 15:50:59,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2701, 1.5045, 1.4410, 1.4372], device='cuda:1'), covar=tensor([0.0762, 0.0313, 0.0298, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0118, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0074, 0.0053, 0.0048, 0.0080], device='cuda:1')
+2023-03-04 15:51:04,128 INFO [optim.py:369] (1/2) 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 15:51:29,880 INFO [train.py:968] (1/2) Epoch 9, batch 10850, giga_loss[loss=0.3352, simple_loss=0.3933, pruned_loss=0.1385, over 28771.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.381, pruned_loss=0.1289, over 5666798.84 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.361, pruned_loss=0.1035, over 5687476.47 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3833, pruned_loss=0.1323, over 5662518.07 frames. ], batch size: 262, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:51:33,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:51:49,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:51:51,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375069.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:51:55,309 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375076.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:52:18,926 INFO [train.py:968] (1/2) Epoch 9, batch 10900, giga_loss[loss=0.3174, simple_loss=0.3835, pruned_loss=0.1256, over 28902.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3819, pruned_loss=0.1301, over 5672890.68 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3607, pruned_loss=0.1034, over 5689834.03 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3842, pruned_loss=0.1332, over 5667135.22 frames. ], batch size: 145, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:52:26,928 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375105.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:52:39,958 INFO [optim.py:369] (1/2) 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 15:53:06,799 INFO [train.py:968] (1/2) Epoch 9, batch 10950, giga_loss[loss=0.3073, simple_loss=0.3809, pruned_loss=0.1168, over 28634.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3828, pruned_loss=0.1297, over 5666808.67 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3603, pruned_loss=0.1032, over 5694922.16 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3856, pruned_loss=0.133, over 5657295.46 frames. ], batch size: 307, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:53:53,915 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:53:57,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:53:57,710 INFO [train.py:968] (1/2) Epoch 9, batch 11000, giga_loss[loss=0.2825, simple_loss=0.357, pruned_loss=0.104, over 28729.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3808, pruned_loss=0.1286, over 5660902.44 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3597, pruned_loss=0.1029, over 5699023.54 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.384, pruned_loss=0.132, over 5649061.47 frames. ], batch size: 60, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:54:04,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0190, 1.1901, 1.2895, 1.1078], device='cuda:1'), covar=tensor([0.1304, 0.1315, 0.1840, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0739, 0.0656, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 15:54:05,224 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:09,856 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:13,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:54:14,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:15,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375215.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 15:54:19,501 INFO [optim.py:369] (1/2) 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 15:54:25,371 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:42,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:54:43,950 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375244.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 15:54:47,287 INFO [train.py:968] (1/2) Epoch 9, batch 11050, giga_loss[loss=0.3756, simple_loss=0.4008, pruned_loss=0.1752, over 23701.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3818, pruned_loss=0.1304, over 5652670.51 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3596, pruned_loss=0.1029, over 5701966.05 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3847, pruned_loss=0.1335, over 5640244.71 frames. ], batch size: 705, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:55:44,816 INFO [train.py:968] (1/2) Epoch 9, batch 11100, giga_loss[loss=0.3475, simple_loss=0.4003, pruned_loss=0.1474, over 28395.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3819, pruned_loss=0.1315, over 5650212.02 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3593, pruned_loss=0.1027, over 5706348.15 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3849, pruned_loss=0.1346, over 5635701.41 frames. ], batch size: 369, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:55:51,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 15:56:01,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9086, 1.1330, 3.3385, 2.9878], device='cuda:1'), covar=tensor([0.1636, 0.2439, 0.0451, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0620, 0.0569, 0.0821, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 15:56:06,493 INFO [optim.py:369] (1/2) 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 15:56:31,172 INFO [train.py:968] (1/2) Epoch 9, batch 11150, giga_loss[loss=0.2699, simple_loss=0.3448, pruned_loss=0.09747, over 28869.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.38, pruned_loss=0.1305, over 5655071.73 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3596, pruned_loss=0.1031, over 5710218.53 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3825, pruned_loss=0.1332, over 5639193.12 frames. ], batch size: 186, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 15:56:42,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 15:56:49,718 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 15:57:13,654 INFO [train.py:968] (1/2) Epoch 9, batch 11200, giga_loss[loss=0.3343, simple_loss=0.3902, pruned_loss=0.1392, over 27923.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3791, pruned_loss=0.13, over 5661501.86 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3595, pruned_loss=0.103, over 5716129.13 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3817, pruned_loss=0.133, over 5642166.59 frames. ], batch size: 412, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:57:25,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5024, 4.2625, 4.0244, 2.0885], device='cuda:1'), covar=tensor([0.0608, 0.0838, 0.1015, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.1011, 0.0956, 0.0840, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 15:57:35,462 INFO [optim.py:369] (1/2) 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 15:57:57,917 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 11250, libri_loss[loss=0.2467, simple_loss=0.329, pruned_loss=0.08224, over 29565.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3788, pruned_loss=0.1299, over 5666910.02 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3597, pruned_loss=0.103, over 5719949.86 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3811, pruned_loss=0.1328, over 5647231.00 frames. ], batch size: 77, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:58:29,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5208, 1.5389, 1.3107, 1.5554], device='cuda:1'), covar=tensor([0.0711, 0.0298, 0.0314, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0074, 0.0053, 0.0048, 0.0081], device='cuda:1')
+2023-03-04 15:58:48,804 INFO [train.py:968] (1/2) Epoch 9, batch 11300, libri_loss[loss=0.2732, simple_loss=0.3519, pruned_loss=0.09721, over 29523.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3802, pruned_loss=0.1313, over 5666353.98 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3599, pruned_loss=0.1031, over 5722240.42 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3822, pruned_loss=0.134, over 5647601.84 frames. ], batch size: 81, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:59:11,319 INFO [optim.py:369] (1/2) 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 15:59:34,740 INFO [train.py:968] (1/2) Epoch 9, batch 11350, giga_loss[loss=0.437, simple_loss=0.4495, pruned_loss=0.2123, over 26600.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3821, pruned_loss=0.1332, over 5662285.79 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3598, pruned_loss=0.103, over 5723137.11 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3843, pruned_loss=0.1362, over 5644821.38 frames. ], batch size: 555, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 15:59:59,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:00:02,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:00:19,078 INFO [train.py:968] (1/2) Epoch 9, batch 11400, giga_loss[loss=0.371, simple_loss=0.4153, pruned_loss=0.1633, over 28532.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3828, pruned_loss=0.1333, over 5647978.67 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3602, pruned_loss=0.1033, over 5718390.59 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3851, pruned_loss=0.1365, over 5636717.22 frames. ], batch size: 336, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:00:42,490 INFO [optim.py:369] (1/2) 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 16:00:46,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6213, 2.3916, 1.5724, 0.7539], device='cuda:1'), covar=tensor([0.3434, 0.1860, 0.3132, 0.3820], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1430, 0.1467, 0.1230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 16:01:09,739 INFO [train.py:968] (1/2) Epoch 9, batch 11450, giga_loss[loss=0.3662, simple_loss=0.4157, pruned_loss=0.1584, over 27898.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3836, pruned_loss=0.1349, over 5647866.63 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3599, pruned_loss=0.1032, over 5720639.26 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3858, pruned_loss=0.1379, over 5636348.83 frames. ], batch size: 412, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:01:30,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1226, 1.8432, 1.8436, 1.7191], device='cuda:1'), covar=tensor([0.1319, 0.2418, 0.1853, 0.1918], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0746, 0.0663, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 16:01:55,433 INFO [train.py:968] (1/2) Epoch 9, batch 11500, giga_loss[loss=0.2967, simple_loss=0.3687, pruned_loss=0.1123, over 28887.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3821, pruned_loss=0.1333, over 5664032.51 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.36, pruned_loss=0.1033, over 5725735.23 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3845, pruned_loss=0.1365, over 5648256.11 frames. ], batch size: 174, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:02:19,911 INFO [optim.py:369] (1/2) 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 16:02:23,627 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:02:25,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:02:41,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:02:44,689 INFO [train.py:968] (1/2) Epoch 9, batch 11550, giga_loss[loss=0.3484, simple_loss=0.3972, pruned_loss=0.1497, over 28257.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3835, pruned_loss=0.1344, over 5657093.34 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3598, pruned_loss=0.1032, over 5727516.49 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.386, pruned_loss=0.1376, over 5641555.69 frames. ], batch size: 368, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:02:52,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:03:29,105 INFO [train.py:968] (1/2) Epoch 9, batch 11600, giga_loss[loss=0.3233, simple_loss=0.383, pruned_loss=0.1318, over 28270.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3833, pruned_loss=0.1331, over 5667358.75 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3599, pruned_loss=0.1032, over 5724085.18 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3858, pruned_loss=0.1364, over 5656037.97 frames. ], batch size: 368, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:03:46,086 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4900, 1.6932, 1.8318, 1.4088], device='cuda:1'), covar=tensor([0.1320, 0.1759, 0.1040, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0719, 0.0842, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 16:03:49,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:03:51,954 INFO [optim.py:369] (1/2) 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 16:03:54,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4777, 2.1543, 1.4703, 0.7547], device='cuda:1'), covar=tensor([0.3006, 0.1574, 0.2641, 0.3457], device='cuda:1'), in_proj_covar=tensor([0.1524, 0.1442, 0.1475, 0.1240], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 16:04:06,891 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:04:08,664 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=375838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:04:19,666 INFO [train.py:968] (1/2) Epoch 9, batch 11650, libri_loss[loss=0.2261, simple_loss=0.3031, pruned_loss=0.07459, over 29503.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3849, pruned_loss=0.1347, over 5655008.68 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3595, pruned_loss=0.103, over 5724600.35 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3879, pruned_loss=0.1384, over 5644046.22 frames. ], batch size: 70, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:04:56,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:04:59,929 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 11700, giga_loss[loss=0.4045, simple_loss=0.4312, pruned_loss=0.1889, over 28332.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3871, pruned_loss=0.1364, over 5655398.37 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3595, pruned_loss=0.1029, over 5726691.86 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3902, pruned_loss=0.1403, over 5642900.31 frames. ], batch size: 368, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:05:23,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-04 16:05:25,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=375919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:05:27,221 INFO [optim.py:369] (1/2) 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 16:05:40,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-04 16:05:52,668 INFO [train.py:968] (1/2) Epoch 9, batch 11750, giga_loss[loss=0.3287, simple_loss=0.3886, pruned_loss=0.1344, over 28591.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3866, pruned_loss=0.1362, over 5660480.65 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3596, pruned_loss=0.1029, over 5730108.37 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3895, pruned_loss=0.14, over 5646383.84 frames. ], batch size: 307, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:05:56,889 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=375952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:06:04,308 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=375962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:06:06,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=375965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:06:13,334 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-04 16:06:29,862 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 11800, giga_loss[loss=0.3037, simple_loss=0.3742, pruned_loss=0.1166, over 27869.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3867, pruned_loss=0.1349, over 5652702.32 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3595, pruned_loss=0.1028, over 5721928.99 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3902, pruned_loss=0.1393, over 5645643.23 frames. ], batch size: 412, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:06:38,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 16:06:59,170 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 11850, giga_loss[loss=0.3218, simple_loss=0.3777, pruned_loss=0.1329, over 28599.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3877, pruned_loss=0.1354, over 5651210.61 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3595, pruned_loss=0.1028, over 5724434.32 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.391, pruned_loss=0.1393, over 5642181.41 frames. ], batch size: 85, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:08:03,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:08:06,321 INFO [train.py:968] (1/2) Epoch 9, batch 11900, giga_loss[loss=0.3537, simple_loss=0.4033, pruned_loss=0.1521, over 28004.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3859, pruned_loss=0.1334, over 5642354.62 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3595, pruned_loss=0.1029, over 5709466.31 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.389, pruned_loss=0.1371, over 5646367.35 frames. ], batch size: 412, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:08:06,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:08:30,013 INFO [optim.py:369] (1/2) 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 16:08:35,636 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 11950, giga_loss[loss=0.2725, simple_loss=0.345, pruned_loss=0.09995, over 28456.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3829, pruned_loss=0.1313, over 5643214.96 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3594, pruned_loss=0.1028, over 5711995.68 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.386, pruned_loss=0.1351, over 5642560.52 frames. ], batch size: 85, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:09:38,983 INFO [train.py:968] (1/2) Epoch 9, batch 12000, giga_loss[loss=0.3101, simple_loss=0.3736, pruned_loss=0.1233, over 28904.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3836, pruned_loss=0.1317, over 5656004.69 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3591, pruned_loss=0.1025, over 5718229.25 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3872, pruned_loss=0.1358, over 5648147.48 frames. ], batch size: 199, lr: 3.65e-03, grad_scale: 8.0
+2023-03-04 16:09:38,984 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 16:09:47,545 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 16:10:00,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=376211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:10:01,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=376213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:10:11,910 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 12050, giga_loss[loss=0.331, simple_loss=0.3898, pruned_loss=0.1361, over 28751.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3851, pruned_loss=0.1334, over 5654113.14 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3588, pruned_loss=0.1024, over 5720909.95 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3886, pruned_loss=0.1372, over 5644590.63 frames. ], batch size: 284, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:11:24,681 INFO [train.py:968] (1/2) Epoch 9, batch 12100, giga_loss[loss=0.2949, simple_loss=0.3506, pruned_loss=0.1196, over 28579.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3844, pruned_loss=0.1336, over 5670032.13 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.359, pruned_loss=0.1024, over 5722892.54 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3871, pruned_loss=0.1368, over 5660265.73 frames. ], batch size: 78, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:11:44,073 INFO [optim.py:369] (1/2) 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 16:11:56,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5223, 4.3579, 4.1174, 2.0539], device='cuda:1'), covar=tensor([0.0473, 0.0633, 0.0660, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.1017, 0.0954, 0.0844, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 16:12:05,504 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=376342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:10,183 INFO [train.py:968] (1/2) Epoch 9, batch 12150, giga_loss[loss=0.3086, simple_loss=0.3717, pruned_loss=0.1227, over 28666.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3847, pruned_loss=0.1341, over 5663801.39 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3601, pruned_loss=0.1033, over 5718215.88 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3866, pruned_loss=0.1367, over 5658210.48 frames. ], batch size: 242, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:12:11,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2945, 3.0922, 2.9263, 1.4213], device='cuda:1'), covar=tensor([0.0867, 0.1098, 0.1067, 0.2260], device='cuda:1'), in_proj_covar=tensor([0.1015, 0.0953, 0.0843, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 16:12:15,426 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:16,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:17,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:18,919 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:46,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=376386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:48,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=376388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:12:56,229 INFO [train.py:968] (1/2) Epoch 9, batch 12200, giga_loss[loss=0.3928, simple_loss=0.4389, pruned_loss=0.1733, over 28313.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.386, pruned_loss=0.1348, over 5654619.94 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3608, pruned_loss=0.1037, over 5706824.79 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.388, pruned_loss=0.138, over 5658401.41 frames. ], batch size: 369, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:13:19,763 INFO [optim.py:369] (1/2) 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 16:13:26,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1957, 1.7567, 1.3983, 0.3204], device='cuda:1'), covar=tensor([0.2363, 0.1604, 0.2443, 0.3531], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1447, 0.1477, 0.1238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 16:13:38,986 INFO [train.py:968] (1/2) Epoch 9, batch 12250, giga_loss[loss=0.2954, simple_loss=0.3648, pruned_loss=0.1129, over 29010.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3861, pruned_loss=0.1345, over 5648227.83 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3609, pruned_loss=0.1037, over 5697871.94 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3884, pruned_loss=0.138, over 5657249.92 frames. ], batch size: 164, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:13:40,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2305, 1.5057, 1.2549, 0.9923], device='cuda:1'), covar=tensor([0.1656, 0.1646, 0.1681, 0.1533], device='cuda:1'), in_proj_covar=tensor([0.1242, 0.0928, 0.1096, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 16:14:29,876 INFO [train.py:968] (1/2) Epoch 9, batch 12300, giga_loss[loss=0.3141, simple_loss=0.3885, pruned_loss=0.1199, over 28464.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3855, pruned_loss=0.134, over 5659179.36 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3609, pruned_loss=0.1036, over 5698903.58 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3875, pruned_loss=0.1369, over 5665096.44 frames. ], batch size: 71, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:14:57,224 INFO [optim.py:369] (1/2) 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 16:14:58,790 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-04 16:15:11,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-04 16:15:17,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9691, 1.0914, 3.4817, 2.9404], device='cuda:1'), covar=tensor([0.1658, 0.2487, 0.0442, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0572, 0.0833, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 16:15:19,816 INFO [train.py:968] (1/2) Epoch 9, batch 12350, giga_loss[loss=0.2947, simple_loss=0.3679, pruned_loss=0.1108, over 28656.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3852, pruned_loss=0.1335, over 5650558.01 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.361, pruned_loss=0.1036, over 5701441.68 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.387, pruned_loss=0.1362, over 5652382.19 frames. ], batch size: 242, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:16:05,538 INFO [train.py:968] (1/2) Epoch 9, batch 12400, giga_loss[loss=0.2878, simple_loss=0.3578, pruned_loss=0.1089, over 28327.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3846, pruned_loss=0.1323, over 5657485.53 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3613, pruned_loss=0.1037, over 5695620.66 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3861, pruned_loss=0.1348, over 5663804.97 frames. ], batch size: 65, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:16:23,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3100, 3.3200, 1.4340, 1.4109], device='cuda:1'), covar=tensor([0.0953, 0.0335, 0.0839, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0503, 0.0329, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 16:16:29,263 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 12450, giga_loss[loss=0.3357, simple_loss=0.3996, pruned_loss=0.1358, over 28871.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.383, pruned_loss=0.1303, over 5676766.69 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3613, pruned_loss=0.1036, over 5701200.25 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.385, pruned_loss=0.1334, over 5675789.84 frames. ], batch size: 174, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:16:53,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4063, 1.0751, 4.3435, 3.3953], device='cuda:1'), covar=tensor([0.1568, 0.2595, 0.0357, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0574, 0.0832, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 16:16:56,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6068, 1.5213, 1.3311, 1.2401], device='cuda:1'), covar=tensor([0.0586, 0.0460, 0.0753, 0.0899], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0442, 0.0493, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 16:17:09,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 16:17:36,967 INFO [train.py:968] (1/2) Epoch 9, batch 12500, giga_loss[loss=0.3261, simple_loss=0.383, pruned_loss=0.1346, over 28806.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3824, pruned_loss=0.131, over 5672555.92 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3613, pruned_loss=0.1037, over 5705464.75 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3844, pruned_loss=0.1338, over 5667097.75 frames. ], batch size: 243, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:18:00,461 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:18:10,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4507, 3.2573, 1.4357, 1.5508], device='cuda:1'), covar=tensor([0.0851, 0.0274, 0.0830, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0504, 0.0330, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 16:18:11,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=376729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:18:27,866 INFO [train.py:968] (1/2) Epoch 9, batch 12550, giga_loss[loss=0.2729, simple_loss=0.3421, pruned_loss=0.1018, over 28850.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3795, pruned_loss=0.1297, over 5668369.04 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3614, pruned_loss=0.1038, over 5706194.43 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3811, pruned_loss=0.1319, over 5663400.05 frames. ], batch size: 199, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:19:16,174 INFO [train.py:968] (1/2) Epoch 9, batch 12600, giga_loss[loss=0.2827, simple_loss=0.3399, pruned_loss=0.1128, over 28633.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3755, pruned_loss=0.1275, over 5680419.01 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3608, pruned_loss=0.1034, over 5710206.23 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3776, pruned_loss=0.1301, over 5672131.24 frames. ], batch size: 85, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:19:40,595 INFO [optim.py:369] (1/2) 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 16:20:02,926 INFO [train.py:968] (1/2) Epoch 9, batch 12650, giga_loss[loss=0.2836, simple_loss=0.3498, pruned_loss=0.1087, over 28943.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3743, pruned_loss=0.1274, over 5687821.56 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3607, pruned_loss=0.1035, over 5712851.18 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3763, pruned_loss=0.1298, over 5678338.08 frames. ], batch size: 186, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:20:16,903 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=376860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:20:18,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=376863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:20:48,096 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 12700, giga_loss[loss=0.2976, simple_loss=0.3673, pruned_loss=0.114, over 29049.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3742, pruned_loss=0.1277, over 5685646.01 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3605, pruned_loss=0.1033, over 5714748.66 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.376, pruned_loss=0.13, over 5676218.29 frames. ], batch size: 136, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:21:19,290 INFO [optim.py:369] (1/2) 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 16:21:42,273 INFO [train.py:968] (1/2) Epoch 9, batch 12750, giga_loss[loss=0.3329, simple_loss=0.4002, pruned_loss=0.1327, over 28630.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3732, pruned_loss=0.1255, over 5682814.06 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3594, pruned_loss=0.1027, over 5715941.19 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3759, pruned_loss=0.1284, over 5673761.61 frames. ], batch size: 307, lr: 3.65e-03, grad_scale: 2.0
+2023-03-04 16:22:24,959 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6848, 1.5781, 1.3679, 1.3610], device='cuda:1'), covar=tensor([0.0545, 0.0430, 0.0704, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0440, 0.0494, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 16:22:33,543 INFO [train.py:968] (1/2) Epoch 9, batch 12800, giga_loss[loss=0.2536, simple_loss=0.3294, pruned_loss=0.08888, over 28809.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3719, pruned_loss=0.1229, over 5672567.46 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3595, pruned_loss=0.1027, over 5718076.89 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3741, pruned_loss=0.1253, over 5663253.72 frames. ], batch size: 92, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:23:00,172 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 12850, giga_loss[loss=0.2837, simple_loss=0.3418, pruned_loss=0.1128, over 26459.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5671245.32 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3594, pruned_loss=0.1028, over 5721000.43 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3714, pruned_loss=0.1223, over 5660697.53 frames. ], batch size: 555, lr: 3.65e-03, grad_scale: 4.0
+2023-03-04 16:23:51,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-04 16:24:01,036 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 12900, giga_loss[loss=0.2673, simple_loss=0.3229, pruned_loss=0.1059, over 24078.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3659, pruned_loss=0.1167, over 5653500.43 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3595, pruned_loss=0.1029, over 5709853.58 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3675, pruned_loss=0.1184, over 5654629.25 frames. ], batch size: 705, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:24:24,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:24:40,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:24:45,139 INFO [optim.py:369] (1/2) 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 16:25:04,881 INFO [train.py:968] (1/2) Epoch 9, batch 12950, giga_loss[loss=0.281, simple_loss=0.3452, pruned_loss=0.1083, over 26685.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3627, pruned_loss=0.1131, over 5667922.45 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3592, pruned_loss=0.1031, over 5715214.01 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3644, pruned_loss=0.1147, over 5662498.32 frames. ], batch size: 555, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:25:13,401 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:25:55,600 INFO [train.py:968] (1/2) Epoch 9, batch 13000, giga_loss[loss=0.3538, simple_loss=0.3908, pruned_loss=0.1584, over 26594.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3606, pruned_loss=0.1096, over 5665522.00 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3584, pruned_loss=0.1027, over 5718013.50 frames. ], giga_tot_loss[loss=0.2926, simple_loss=0.3627, pruned_loss=0.1112, over 5658134.59 frames. ], batch size: 555, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:26:23,286 INFO [optim.py:369] (1/2) 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 16:26:44,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:26:45,128 INFO [train.py:968] (1/2) Epoch 9, batch 13050, libri_loss[loss=0.2702, simple_loss=0.3372, pruned_loss=0.1016, over 29582.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3602, pruned_loss=0.1094, over 5664604.23 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.358, pruned_loss=0.1026, over 5722859.82 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3623, pruned_loss=0.111, over 5653163.15 frames. ], batch size: 74, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:26:47,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=377250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:27:02,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 16:27:18,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:27:34,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3576, 1.3714, 1.5123, 1.1876], device='cuda:1'), covar=tensor([0.1607, 0.2595, 0.1372, 0.1469], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0698, 0.0832, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 16:27:37,497 INFO [train.py:968] (1/2) Epoch 9, batch 13100, giga_loss[loss=0.2864, simple_loss=0.3455, pruned_loss=0.1136, over 26555.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3577, pruned_loss=0.1076, over 5662598.08 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3582, pruned_loss=0.1028, over 5724443.85 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3592, pruned_loss=0.1087, over 5651602.32 frames. ], batch size: 555, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:28:02,323 INFO [optim.py:369] (1/2) 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 16:28:26,716 INFO [train.py:968] (1/2) Epoch 9, batch 13150, giga_loss[loss=0.2518, simple_loss=0.3285, pruned_loss=0.08748, over 27994.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3545, pruned_loss=0.105, over 5670467.15 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3583, pruned_loss=0.1029, over 5726558.63 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3555, pruned_loss=0.1058, over 5658831.45 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:28:32,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8400, 1.0695, 2.8254, 2.6644], device='cuda:1'), covar=tensor([0.1601, 0.2330, 0.0575, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0623, 0.0568, 0.0822, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 16:29:17,194 INFO [train.py:968] (1/2) Epoch 9, batch 13200, giga_loss[loss=0.3035, simple_loss=0.3715, pruned_loss=0.1177, over 28735.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3542, pruned_loss=0.1048, over 5669336.43 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3581, pruned_loss=0.1029, over 5725718.28 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3551, pruned_loss=0.1055, over 5659947.60 frames. ], batch size: 242, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:29:43,977 INFO [optim.py:369] (1/2) 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 16:30:07,082 INFO [train.py:968] (1/2) Epoch 9, batch 13250, giga_loss[loss=0.2295, simple_loss=0.3141, pruned_loss=0.07249, over 28563.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3531, pruned_loss=0.1035, over 5670490.90 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3576, pruned_loss=0.1026, over 5728515.82 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3543, pruned_loss=0.1043, over 5659775.66 frames. ], batch size: 60, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:30:17,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:30:53,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:30:59,054 INFO [train.py:968] (1/2) Epoch 9, batch 13300, giga_loss[loss=0.2692, simple_loss=0.353, pruned_loss=0.09276, over 29069.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3508, pruned_loss=0.1015, over 5672206.18 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.357, pruned_loss=0.1024, over 5732579.97 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3521, pruned_loss=0.1024, over 5658617.21 frames. ], batch size: 128, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:31:23,024 INFO [optim.py:369] (1/2) 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 16:31:29,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:31:46,073 INFO [train.py:968] (1/2) Epoch 9, batch 13350, libri_loss[loss=0.2672, simple_loss=0.3441, pruned_loss=0.0951, over 29621.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3474, pruned_loss=0.09915, over 5676022.05 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.356, pruned_loss=0.1021, over 5732048.32 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3491, pruned_loss=0.1001, over 5663022.03 frames. ], batch size: 91, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:32:24,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7597, 4.5556, 4.3309, 1.7501], device='cuda:1'), covar=tensor([0.0408, 0.0615, 0.0658, 0.2329], device='cuda:1'), in_proj_covar=tensor([0.0972, 0.0913, 0.0811, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 16:32:41,317 INFO [train.py:968] (1/2) Epoch 9, batch 13400, giga_loss[loss=0.2506, simple_loss=0.3088, pruned_loss=0.09625, over 24013.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3439, pruned_loss=0.09758, over 5659388.62 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3562, pruned_loss=0.1023, over 5733013.45 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.345, pruned_loss=0.09813, over 5647827.33 frames. ], batch size: 705, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:32:43,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:32:48,507 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:32:48,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=377603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:32:57,265 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-04 16:33:12,029 INFO [optim.py:369] (1/2) 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 16:33:18,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:20,262 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5009, 2.0638, 1.4817, 0.5845], device='cuda:1'), covar=tensor([0.2949, 0.1732, 0.2515, 0.3677], device='cuda:1'), in_proj_covar=tensor([0.1504, 0.1416, 0.1450, 0.1224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 16:33:24,670 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:27,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=377640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:33,357 INFO [train.py:968] (1/2) Epoch 9, batch 13450, giga_loss[loss=0.259, simple_loss=0.3321, pruned_loss=0.09293, over 27953.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3427, pruned_loss=0.0978, over 5654990.54 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3559, pruned_loss=0.1021, over 5736302.43 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3437, pruned_loss=0.09828, over 5641315.22 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:33:53,241 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:55,283 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:33:59,575 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=377674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:34:02,227 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 13500, giga_loss[loss=0.2528, simple_loss=0.3374, pruned_loss=0.08414, over 28877.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3432, pruned_loss=0.09876, over 5644937.74 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3555, pruned_loss=0.102, over 5736779.96 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3441, pruned_loss=0.09923, over 5632230.67 frames. ], batch size: 174, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:34:39,023 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=377706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:35:00,429 INFO [optim.py:369] (1/2) 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 16:35:16,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-04 16:35:23,323 INFO [train.py:968] (1/2) Epoch 9, batch 13550, libri_loss[loss=0.2367, simple_loss=0.3027, pruned_loss=0.08538, over 28582.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3456, pruned_loss=0.09907, over 5646991.57 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3552, pruned_loss=0.102, over 5739246.92 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3464, pruned_loss=0.09939, over 5632544.49 frames. ], batch size: 63, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:35:24,163 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 16:35:42,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6922, 1.9948, 1.7410, 1.7993], device='cuda:1'), covar=tensor([0.1282, 0.1649, 0.1690, 0.1466], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0726, 0.0652, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 16:35:46,616 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 13600, giga_loss[loss=0.2696, simple_loss=0.3421, pruned_loss=0.09857, over 28105.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3465, pruned_loss=0.0987, over 5649928.85 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3549, pruned_loss=0.1019, over 5742125.43 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3471, pruned_loss=0.09895, over 5633955.54 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:36:54,685 INFO [optim.py:369] (1/2) 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 16:37:24,577 INFO [train.py:968] (1/2) Epoch 9, batch 13650, giga_loss[loss=0.3108, simple_loss=0.3691, pruned_loss=0.1262, over 27646.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3468, pruned_loss=0.09938, over 5643457.91 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3542, pruned_loss=0.1016, over 5743469.92 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3478, pruned_loss=0.09983, over 5628476.15 frames. ], batch size: 472, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:38:26,395 INFO [train.py:968] (1/2) Epoch 9, batch 13700, giga_loss[loss=0.2743, simple_loss=0.351, pruned_loss=0.0988, over 28778.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3444, pruned_loss=0.09748, over 5656280.29 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3541, pruned_loss=0.1016, over 5744100.57 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3454, pruned_loss=0.09783, over 5643572.46 frames. ], batch size: 243, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:39:02,186 INFO [optim.py:369] (1/2) 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 16:39:06,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9170, 1.1462, 1.0289, 0.7941], device='cuda:1'), covar=tensor([0.1377, 0.1342, 0.0739, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1467, 0.1419, 0.1533], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 16:39:26,967 INFO [train.py:968] (1/2) Epoch 9, batch 13750, giga_loss[loss=0.2446, simple_loss=0.3298, pruned_loss=0.07973, over 28992.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3425, pruned_loss=0.09543, over 5647474.52 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3531, pruned_loss=0.1011, over 5744535.86 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3439, pruned_loss=0.09602, over 5634781.05 frames. ], batch size: 285, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:39:49,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3257, 1.7263, 1.4018, 1.4909], device='cuda:1'), covar=tensor([0.0718, 0.0310, 0.0325, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0119, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:1')
+2023-03-04 16:39:58,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2537, 1.5699, 1.2663, 1.1202], device='cuda:1'), covar=tensor([0.1923, 0.1327, 0.0990, 0.1428], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1471, 0.1421, 0.1537], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 16:40:02,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=377978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:40:14,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-04 16:40:23,996 INFO [train.py:968] (1/2) Epoch 9, batch 13800, giga_loss[loss=0.2633, simple_loss=0.3294, pruned_loss=0.09861, over 28984.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3401, pruned_loss=0.09403, over 5657941.28 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3524, pruned_loss=0.1008, over 5748869.53 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3414, pruned_loss=0.09453, over 5640862.67 frames. ], batch size: 213, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:40:24,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=377998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:40:59,655 INFO [optim.py:369] (1/2) 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 16:41:17,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:41:24,514 INFO [train.py:968] (1/2) Epoch 9, batch 13850, giga_loss[loss=0.2583, simple_loss=0.3332, pruned_loss=0.09172, over 27791.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3387, pruned_loss=0.09433, over 5660745.57 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3528, pruned_loss=0.1012, over 5751438.55 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3392, pruned_loss=0.09435, over 5643884.57 frames. ], batch size: 474, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:42:20,691 INFO [train.py:968] (1/2) Epoch 9, batch 13900, giga_loss[loss=0.2138, simple_loss=0.2996, pruned_loss=0.06401, over 29018.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3376, pruned_loss=0.09386, over 5667126.35 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3521, pruned_loss=0.1008, over 5755402.00 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3382, pruned_loss=0.09395, over 5647733.53 frames. ], batch size: 120, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:42:46,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:42:50,097 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:43:15,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:43:18,417 INFO [train.py:968] (1/2) Epoch 9, batch 13950, giga_loss[loss=0.284, simple_loss=0.3599, pruned_loss=0.1041, over 28881.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3395, pruned_loss=0.0942, over 5677759.23 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.352, pruned_loss=0.1008, over 5757559.28 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3399, pruned_loss=0.09423, over 5659416.91 frames. ], batch size: 227, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:43:23,950 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=378153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:44:00,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:44:03,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=378187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:44:12,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2101, 1.4312, 1.4106, 1.2750], device='cuda:1'), covar=tensor([0.1146, 0.1309, 0.1585, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0716, 0.0646, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 16:44:13,781 INFO [train.py:968] (1/2) Epoch 9, batch 14000, giga_loss[loss=0.2702, simple_loss=0.3493, pruned_loss=0.09558, over 28438.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3422, pruned_loss=0.09492, over 5687168.03 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3517, pruned_loss=0.1007, over 5762729.16 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3425, pruned_loss=0.09477, over 5665406.52 frames. ], batch size: 368, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:44:36,634 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:45:15,673 INFO [train.py:968] (1/2) Epoch 9, batch 14050, giga_loss[loss=0.2161, simple_loss=0.3034, pruned_loss=0.06439, over 28425.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3384, pruned_loss=0.09224, over 5687508.41 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3515, pruned_loss=0.1006, over 5765647.89 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3385, pruned_loss=0.09208, over 5665385.57 frames. ], batch size: 336, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:46:03,883 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:46:07,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=378291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:46:13,480 INFO [train.py:968] (1/2) Epoch 9, batch 14100, giga_loss[loss=0.2875, simple_loss=0.3536, pruned_loss=0.1107, over 28476.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3387, pruned_loss=0.09295, over 5688872.03 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.351, pruned_loss=0.1005, over 5759977.40 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09265, over 5673864.32 frames. ], batch size: 369, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:46:23,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-04 16:46:46,829 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:47:24,078 INFO [train.py:968] (1/2) Epoch 9, batch 14150, giga_loss[loss=0.2846, simple_loss=0.3615, pruned_loss=0.1039, over 27721.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3418, pruned_loss=0.0945, over 5667773.24 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3515, pruned_loss=0.1008, over 5762919.82 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3413, pruned_loss=0.09386, over 5651560.08 frames. ], batch size: 472, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:47:55,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:48:24,792 INFO [train.py:968] (1/2) Epoch 9, batch 14200, giga_loss[loss=0.2625, simple_loss=0.356, pruned_loss=0.08452, over 29022.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3443, pruned_loss=0.09288, over 5669430.69 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.351, pruned_loss=0.1005, over 5766621.50 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3441, pruned_loss=0.0925, over 5650865.55 frames. ], batch size: 128, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:48:55,198 INFO [optim.py:369] (1/2) 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 16:49:23,819 INFO [train.py:968] (1/2) Epoch 9, batch 14250, giga_loss[loss=0.2679, simple_loss=0.3564, pruned_loss=0.08963, over 28780.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3432, pruned_loss=0.09137, over 5656198.19 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3505, pruned_loss=0.1001, over 5768895.67 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3434, pruned_loss=0.09117, over 5636942.57 frames. ], batch size: 243, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:50:21,750 INFO [train.py:968] (1/2) Epoch 9, batch 14300, libri_loss[loss=0.2717, simple_loss=0.3432, pruned_loss=0.1001, over 29533.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3442, pruned_loss=0.09097, over 5667446.07 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3504, pruned_loss=0.1001, over 5770340.67 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3443, pruned_loss=0.09072, over 5649293.97 frames. ], batch size: 81, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:50:45,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=378516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:50:48,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=378519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:51:01,517 INFO [optim.py:369] (1/2) 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 16:51:25,616 INFO [train.py:968] (1/2) Epoch 9, batch 14350, giga_loss[loss=0.3124, simple_loss=0.3714, pruned_loss=0.1267, over 27010.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3464, pruned_loss=0.0933, over 5668108.24 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3503, pruned_loss=0.09999, over 5767124.92 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3466, pruned_loss=0.09312, over 5655618.18 frames. ], batch size: 555, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:51:26,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=378548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:51:53,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 16:52:16,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3550, 1.2887, 4.5548, 3.3992], device='cuda:1'), covar=tensor([0.1545, 0.2383, 0.0305, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0616, 0.0569, 0.0815, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 16:52:28,668 INFO [train.py:968] (1/2) Epoch 9, batch 14400, libri_loss[loss=0.2297, simple_loss=0.3025, pruned_loss=0.07847, over 29583.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3451, pruned_loss=0.09396, over 5674231.27 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3502, pruned_loss=0.1, over 5770338.52 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3452, pruned_loss=0.09365, over 5659125.88 frames. ], batch size: 75, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:52:46,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:52:53,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:53:09,823 INFO [optim.py:369] (1/2) 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 16:53:40,578 INFO [train.py:968] (1/2) Epoch 9, batch 14450, giga_loss[loss=0.2672, simple_loss=0.3473, pruned_loss=0.0936, over 28928.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3467, pruned_loss=0.09545, over 5675247.31 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3501, pruned_loss=0.09996, over 5772953.79 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3468, pruned_loss=0.09516, over 5658911.44 frames. ], batch size: 213, lr: 3.64e-03, grad_scale: 8.0
+2023-03-04 16:53:51,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4807, 1.6855, 1.3191, 1.9316], device='cuda:1'), covar=tensor([0.2473, 0.2316, 0.2614, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.1233, 0.0914, 0.1093, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 16:54:37,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-04 16:54:44,450 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:54:58,123 INFO [train.py:968] (1/2) Epoch 9, batch 14500, giga_loss[loss=0.2429, simple_loss=0.3192, pruned_loss=0.08325, over 28524.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3413, pruned_loss=0.09257, over 5680006.43 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3491, pruned_loss=0.09953, over 5774757.95 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3422, pruned_loss=0.09259, over 5663371.75 frames. ], batch size: 85, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 16:55:43,360 INFO [optim.py:369] (1/2) 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 16:56:04,016 INFO [train.py:968] (1/2) Epoch 9, batch 14550, libri_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.0896, over 29484.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3397, pruned_loss=0.09174, over 5673389.73 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3492, pruned_loss=0.09961, over 5774672.15 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.34, pruned_loss=0.09147, over 5657565.31 frames. ], batch size: 85, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:56:16,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4185, 2.0043, 1.3761, 0.7239], device='cuda:1'), covar=tensor([0.3470, 0.1758, 0.2827, 0.3484], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1411, 0.1454, 0.1215], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 16:57:05,079 INFO [train.py:968] (1/2) Epoch 9, batch 14600, giga_loss[loss=0.2829, simple_loss=0.3444, pruned_loss=0.1107, over 26777.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3367, pruned_loss=0.09064, over 5679811.03 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3483, pruned_loss=0.09914, over 5775287.43 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3375, pruned_loss=0.09062, over 5663843.69 frames. ], batch size: 555, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:57:41,509 INFO [optim.py:369] (1/2) 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 16:58:00,791 INFO [train.py:968] (1/2) Epoch 9, batch 14650, libri_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1119, over 27961.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3416, pruned_loss=0.09335, over 5684106.82 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3487, pruned_loss=0.09945, over 5768571.62 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3415, pruned_loss=0.09275, over 5673324.24 frames. ], batch size: 116, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:58:07,197 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:58:13,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=378855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 16:59:05,753 INFO [train.py:968] (1/2) Epoch 9, batch 14700, giga_loss[loss=0.2391, simple_loss=0.32, pruned_loss=0.07905, over 28711.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3427, pruned_loss=0.09467, over 5678816.61 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3484, pruned_loss=0.09932, over 5769816.23 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3429, pruned_loss=0.09427, over 5668455.17 frames. ], batch size: 307, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 16:59:45,038 INFO [optim.py:369] (1/2) 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 17:00:05,563 INFO [train.py:968] (1/2) Epoch 9, batch 14750, giga_loss[loss=0.2624, simple_loss=0.3228, pruned_loss=0.101, over 24371.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.342, pruned_loss=0.0954, over 5678869.94 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3481, pruned_loss=0.09912, over 5771934.62 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3423, pruned_loss=0.09519, over 5667116.76 frames. ], batch size: 705, lr: 3.64e-03, grad_scale: 2.0
+2023-03-04 17:00:07,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8997, 1.0892, 0.9985, 0.7813], device='cuda:1'), covar=tensor([0.1234, 0.1282, 0.0788, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1459, 0.1410, 0.1533], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 17:00:52,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-04 17:00:53,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:01:00,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=378991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:01:06,716 INFO [train.py:968] (1/2) Epoch 9, batch 14800, libri_loss[loss=0.2193, simple_loss=0.2953, pruned_loss=0.07167, over 29507.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3419, pruned_loss=0.09554, over 5674155.87 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3474, pruned_loss=0.09881, over 5767160.28 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.09559, over 5665512.54 frames. ], batch size: 70, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:01:21,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 17:01:46,361 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 14850, giga_loss[loss=0.3212, simple_loss=0.3929, pruned_loss=0.1248, over 28495.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3428, pruned_loss=0.09534, over 5677677.55 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3472, pruned_loss=0.09888, over 5770771.92 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3434, pruned_loss=0.09525, over 5665808.47 frames. ], batch size: 369, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:02:12,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2303, 1.2096, 1.0532, 0.9148], device='cuda:1'), covar=tensor([0.0700, 0.0480, 0.0960, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0435, 0.0493, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 17:02:28,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:02:46,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8615, 1.7901, 1.2468, 1.5136], device='cuda:1'), covar=tensor([0.0713, 0.0675, 0.0967, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0436, 0.0494, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 17:02:58,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2961, 2.5434, 1.3162, 1.3869], device='cuda:1'), covar=tensor([0.0832, 0.0344, 0.0849, 0.1226], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0487, 0.0326, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 17:03:15,847 INFO [train.py:968] (1/2) Epoch 9, batch 14900, giga_loss[loss=0.234, simple_loss=0.3187, pruned_loss=0.07463, over 28864.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3446, pruned_loss=0.09553, over 5680598.24 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3465, pruned_loss=0.09846, over 5774636.69 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3456, pruned_loss=0.09575, over 5664988.61 frames. ], batch size: 213, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:04:04,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:06,775 INFO [optim.py:369] (1/2) 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 17:04:09,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:12,131 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:16,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:04:37,785 INFO [train.py:968] (1/2) Epoch 9, batch 14950, giga_loss[loss=0.2765, simple_loss=0.3443, pruned_loss=0.1044, over 28140.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3431, pruned_loss=0.09442, over 5679854.85 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3464, pruned_loss=0.09845, over 5775308.51 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.344, pruned_loss=0.09458, over 5666751.32 frames. ], batch size: 412, lr: 3.64e-03, grad_scale: 4.0
+2023-03-04 17:04:51,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:05:02,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:05:41,806 INFO [train.py:968] (1/2) Epoch 9, batch 15000, giga_loss[loss=0.2393, simple_loss=0.3166, pruned_loss=0.08093, over 28807.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3389, pruned_loss=0.09315, over 5686593.83 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3465, pruned_loss=0.0987, over 5767931.93 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3394, pruned_loss=0.09295, over 5680974.47 frames. ], batch size: 174, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:05:41,806 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 17:05:50,586 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 17:06:00,460 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:05,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:31,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:34,391 INFO [optim.py:369] (1/2) 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 17:06:35,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:06:44,251 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 15050, giga_loss[loss=0.3109, simple_loss=0.3669, pruned_loss=0.1275, over 28987.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3338, pruned_loss=0.09135, over 5675673.22 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3464, pruned_loss=0.09868, over 5761669.55 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3342, pruned_loss=0.09108, over 5675705.98 frames. ], batch size: 213, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:07:08,110 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:07:38,292 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:07:54,780 INFO [train.py:968] (1/2) Epoch 9, batch 15100, giga_loss[loss=0.2518, simple_loss=0.3269, pruned_loss=0.08836, over 28975.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3359, pruned_loss=0.09344, over 5673959.47 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3465, pruned_loss=0.09896, over 5763092.64 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.336, pruned_loss=0.09289, over 5671846.56 frames. ], batch size: 227, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:08:13,557 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:08:46,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 17:08:46,974 INFO [train.py:968] (1/2) Epoch 9, batch 15150, giga_loss[loss=0.2752, simple_loss=0.3436, pruned_loss=0.1034, over 28572.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3372, pruned_loss=0.09467, over 5672006.06 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3461, pruned_loss=0.09885, over 5763596.25 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3374, pruned_loss=0.09421, over 5667259.63 frames. ], batch size: 242, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:09:15,190 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:18,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:20,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:24,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:44,219 INFO [train.py:968] (1/2) Epoch 9, batch 15200, giga_loss[loss=0.274, simple_loss=0.3498, pruned_loss=0.09913, over 28381.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3336, pruned_loss=0.09168, over 5668247.70 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3454, pruned_loss=0.09841, over 5767078.44 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.334, pruned_loss=0.09154, over 5657919.84 frames. ], batch size: 368, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:09:50,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:09:56,271 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:10:21,263 INFO [optim.py:369] (1/2) 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 17:10:41,253 INFO [train.py:968] (1/2) Epoch 9, batch 15250, giga_loss[loss=0.2452, simple_loss=0.3274, pruned_loss=0.08147, over 28167.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3327, pruned_loss=0.09034, over 5667878.94 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3449, pruned_loss=0.09814, over 5765040.43 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3331, pruned_loss=0.09024, over 5658269.53 frames. ], batch size: 412, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:11:52,430 INFO [train.py:968] (1/2) Epoch 9, batch 15300, giga_loss[loss=0.2693, simple_loss=0.3374, pruned_loss=0.1007, over 28420.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3319, pruned_loss=0.09049, over 5669297.32 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3447, pruned_loss=0.098, over 5766498.40 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3322, pruned_loss=0.09046, over 5659634.23 frames. ], batch size: 336, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:12:31,481 INFO [optim.py:369] (1/2) 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 17:12:37,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-04 17:12:51,961 INFO [train.py:968] (1/2) Epoch 9, batch 15350, libri_loss[loss=0.2093, simple_loss=0.2937, pruned_loss=0.06244, over 28574.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3316, pruned_loss=0.08924, over 5686031.33 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3442, pruned_loss=0.09769, over 5767702.71 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3319, pruned_loss=0.08926, over 5674349.34 frames. ], batch size: 63, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:13:27,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 17:13:53,090 INFO [train.py:968] (1/2) Epoch 9, batch 15400, libri_loss[loss=0.2785, simple_loss=0.3504, pruned_loss=0.1033, over 29211.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3316, pruned_loss=0.08914, over 5696691.24 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3435, pruned_loss=0.09735, over 5772753.83 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.332, pruned_loss=0.08916, over 5679582.27 frames. ], batch size: 97, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:14:32,129 INFO [optim.py:369] (1/2) 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 17:14:35,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:14:54,678 INFO [train.py:968] (1/2) Epoch 9, batch 15450, giga_loss[loss=0.2339, simple_loss=0.3144, pruned_loss=0.07668, over 28799.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3339, pruned_loss=0.09122, over 5694406.76 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3438, pruned_loss=0.09748, over 5771473.23 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3338, pruned_loss=0.091, over 5680407.37 frames. ], batch size: 263, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:15:05,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=379657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:15:19,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379670.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 17:15:44,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2649, 1.8708, 1.4200, 1.6219], device='cuda:1'), covar=tensor([0.0794, 0.0309, 0.0333, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0120, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0053, 0.0049, 0.0082], device='cuda:1')
+2023-03-04 17:15:48,280 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 15500, giga_loss[loss=0.2892, simple_loss=0.3704, pruned_loss=0.104, over 28871.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3319, pruned_loss=0.08981, over 5687225.20 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3434, pruned_loss=0.09726, over 5774434.01 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.332, pruned_loss=0.0897, over 5671968.03 frames. ], batch size: 227, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:16:22,321 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:16:52,095 INFO [train.py:968] (1/2) Epoch 9, batch 15550, giga_loss[loss=0.2915, simple_loss=0.3642, pruned_loss=0.1093, over 28675.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3343, pruned_loss=0.09017, over 5671413.58 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3433, pruned_loss=0.09708, over 5776196.27 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3343, pruned_loss=0.09007, over 5655911.48 frames. ], batch size: 262, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:17:25,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:17:28,615 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:17:49,008 INFO [train.py:968] (1/2) Epoch 9, batch 15600, giga_loss[loss=0.2765, simple_loss=0.3527, pruned_loss=0.1002, over 28979.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3382, pruned_loss=0.09187, over 5658154.15 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3435, pruned_loss=0.09725, over 5760781.29 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3378, pruned_loss=0.09148, over 5657293.80 frames. ], batch size: 186, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:17:49,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-04 17:17:51,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:17:55,121 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:27,883 INFO [optim.py:369] (1/2) 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 17:18:29,202 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:29,479 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-04 17:18:30,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=379834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:33,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=379837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:18:47,688 INFO [train.py:968] (1/2) Epoch 9, batch 15650, giga_loss[loss=0.2441, simple_loss=0.3025, pruned_loss=0.0928, over 24321.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3396, pruned_loss=0.09252, over 5659594.88 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3434, pruned_loss=0.09716, over 5764252.45 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3393, pruned_loss=0.09217, over 5653514.21 frames. ], batch size: 705, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:19:07,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=379866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:19:32,781 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=379887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:19:42,262 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 15700, giga_loss[loss=0.265, simple_loss=0.3308, pruned_loss=0.09957, over 28736.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3397, pruned_loss=0.09327, over 5654603.26 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.343, pruned_loss=0.09694, over 5765875.86 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3396, pruned_loss=0.09307, over 5644867.29 frames. ], batch size: 99, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:20:04,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1921, 3.0236, 2.8191, 1.3924], device='cuda:1'), covar=tensor([0.0918, 0.0992, 0.1008, 0.2401], device='cuda:1'), in_proj_covar=tensor([0.0955, 0.0891, 0.0790, 0.0618], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 17:20:22,910 INFO [optim.py:369] (1/2) 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 17:20:40,377 INFO [train.py:968] (1/2) Epoch 9, batch 15750, giga_loss[loss=0.2358, simple_loss=0.3161, pruned_loss=0.07772, over 28981.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3369, pruned_loss=0.09177, over 5660307.30 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3432, pruned_loss=0.09714, over 5769119.95 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3366, pruned_loss=0.09129, over 5646478.63 frames. ], batch size: 136, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:21:44,284 INFO [train.py:968] (1/2) Epoch 9, batch 15800, giga_loss[loss=0.2706, simple_loss=0.3392, pruned_loss=0.101, over 27605.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3355, pruned_loss=0.09079, over 5653520.08 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3431, pruned_loss=0.09706, over 5760335.39 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3352, pruned_loss=0.09034, over 5647520.89 frames. ], batch size: 472, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:22:21,590 INFO [optim.py:369] (1/2) 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 17:22:30,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1217, 1.2742, 3.3537, 2.8709], device='cuda:1'), covar=tensor([0.1535, 0.2346, 0.0451, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0611, 0.0567, 0.0806, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 17:22:37,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380045.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 17:22:41,075 INFO [train.py:968] (1/2) Epoch 9, batch 15850, giga_loss[loss=0.2411, simple_loss=0.3023, pruned_loss=0.08996, over 24621.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3344, pruned_loss=0.09108, over 5665345.28 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3425, pruned_loss=0.09698, over 5762291.91 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3345, pruned_loss=0.09062, over 5656353.32 frames. ], batch size: 705, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:23:41,714 INFO [train.py:968] (1/2) Epoch 9, batch 15900, giga_loss[loss=0.2379, simple_loss=0.3316, pruned_loss=0.07213, over 28683.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3357, pruned_loss=0.09118, over 5671713.85 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3425, pruned_loss=0.09693, over 5763944.32 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3357, pruned_loss=0.0908, over 5662454.87 frames. ], batch size: 307, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:23:44,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:23:54,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6998, 1.6588, 1.2312, 1.3605], device='cuda:1'), covar=tensor([0.0688, 0.0584, 0.0899, 0.1032], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0436, 0.0494, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 17:24:23,461 INFO [optim.py:369] (1/2) 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 17:24:31,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5920, 1.6956, 1.2072, 1.3170], device='cuda:1'), covar=tensor([0.0669, 0.0472, 0.0887, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0437, 0.0495, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 17:24:46,406 INFO [train.py:968] (1/2) Epoch 9, batch 15950, giga_loss[loss=0.2607, simple_loss=0.3379, pruned_loss=0.09174, over 28964.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3363, pruned_loss=0.09147, over 5670917.33 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3423, pruned_loss=0.09685, over 5767590.25 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3363, pruned_loss=0.09111, over 5658402.56 frames. ], batch size: 213, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:25:33,066 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380188.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 17:25:36,118 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 16000, giga_loss[loss=0.2712, simple_loss=0.3527, pruned_loss=0.09489, over 29041.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.338, pruned_loss=0.09335, over 5669791.05 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.342, pruned_loss=0.09663, over 5770031.74 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3382, pruned_loss=0.09311, over 5654582.12 frames. ], batch size: 128, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:26:08,048 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=380220.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 17:26:23,134 INFO [optim.py:369] (1/2) 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 17:26:33,944 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:26:39,063 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=380245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:26:40,992 INFO [train.py:968] (1/2) Epoch 9, batch 16050, giga_loss[loss=0.2884, simple_loss=0.3681, pruned_loss=0.1044, over 28893.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3427, pruned_loss=0.09604, over 5659145.63 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3417, pruned_loss=0.09648, over 5763076.38 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.343, pruned_loss=0.09592, over 5652003.36 frames. ], batch size: 284, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:26:57,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:26:58,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 17:26:59,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-04 17:27:07,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:27:10,912 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=380274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:27:37,049 INFO [train.py:968] (1/2) Epoch 9, batch 16100, giga_loss[loss=0.2828, simple_loss=0.3614, pruned_loss=0.1021, over 28718.00 frames. ], tot_loss[loss=0.268, simple_loss=0.344, pruned_loss=0.09601, over 5662896.42 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3412, pruned_loss=0.09621, over 5767924.17 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3447, pruned_loss=0.09616, over 5649378.43 frames. ], batch size: 307, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:27:38,698 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-04 17:28:21,010 INFO [optim.py:369] (1/2) 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 17:28:41,857 INFO [train.py:968] (1/2) Epoch 9, batch 16150, giga_loss[loss=0.2397, simple_loss=0.296, pruned_loss=0.09172, over 24227.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3437, pruned_loss=0.09572, over 5655124.18 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3408, pruned_loss=0.0961, over 5769613.30 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3446, pruned_loss=0.09593, over 5640808.88 frames. ], batch size: 705, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:29:48,773 INFO [train.py:968] (1/2) Epoch 9, batch 16200, giga_loss[loss=0.246, simple_loss=0.3222, pruned_loss=0.08494, over 28153.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3428, pruned_loss=0.09521, over 5666343.78 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3415, pruned_loss=0.09647, over 5772301.28 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.343, pruned_loss=0.09498, over 5649919.74 frames. ], batch size: 412, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:29:57,862 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:00,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=380408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:06,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=380414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:10,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:30:38,172 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=380437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:30:47,390 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 16250, giga_loss[loss=0.2679, simple_loss=0.3468, pruned_loss=0.09447, over 28457.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3408, pruned_loss=0.09395, over 5663828.69 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3415, pruned_loss=0.09644, over 5765419.45 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.341, pruned_loss=0.09375, over 5655416.22 frames. ], batch size: 336, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:31:52,751 INFO [train.py:968] (1/2) Epoch 9, batch 16300, giga_loss[loss=0.2443, simple_loss=0.3237, pruned_loss=0.08246, over 29015.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3402, pruned_loss=0.09388, over 5664021.52 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3416, pruned_loss=0.0964, over 5764077.00 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3402, pruned_loss=0.09373, over 5656964.31 frames. ], batch size: 285, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:32:28,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6953, 2.4187, 1.7371, 1.4446], device='cuda:1'), covar=tensor([0.2322, 0.1208, 0.1414, 0.1834], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1462, 0.1412, 0.1538], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 17:32:34,638 INFO [optim.py:369] (1/2) 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 17:32:53,256 INFO [train.py:968] (1/2) Epoch 9, batch 16350, giga_loss[loss=0.2164, simple_loss=0.3018, pruned_loss=0.06548, over 28721.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3373, pruned_loss=0.09324, over 5661541.22 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3408, pruned_loss=0.09593, over 5768768.62 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.338, pruned_loss=0.09346, over 5648254.70 frames. ], batch size: 243, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:33:00,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1205, 1.3030, 3.2886, 2.8710], device='cuda:1'), covar=tensor([0.1261, 0.2011, 0.0398, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0618, 0.0570, 0.0811, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 17:33:08,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-04 17:33:25,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:33:29,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9647, 1.1550, 3.1144, 2.7398], device='cuda:1'), covar=tensor([0.2194, 0.2836, 0.1005, 0.2003], device='cuda:1'), in_proj_covar=tensor([0.0619, 0.0572, 0.0813, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 17:33:47,019 INFO [train.py:968] (1/2) Epoch 9, batch 16400, giga_loss[loss=0.2871, simple_loss=0.3587, pruned_loss=0.1077, over 28102.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3362, pruned_loss=0.09261, over 5660614.90 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3407, pruned_loss=0.09592, over 5764880.85 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3366, pruned_loss=0.09269, over 5648995.81 frames. ], batch size: 412, lr: 3.63e-03, grad_scale: 8.0
+2023-03-04 17:34:16,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7096, 1.5246, 5.0601, 3.4629], device='cuda:1'), covar=tensor([0.1499, 0.2394, 0.0318, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0620, 0.0573, 0.0813, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 17:34:29,715 INFO [optim.py:369] (1/2) 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 17:34:45,290 INFO [train.py:968] (1/2) Epoch 9, batch 16450, giga_loss[loss=0.2325, simple_loss=0.3197, pruned_loss=0.07266, over 28607.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3363, pruned_loss=0.09189, over 5660401.95 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3405, pruned_loss=0.09592, over 5756267.95 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3367, pruned_loss=0.09187, over 5657941.45 frames. ], batch size: 85, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:35:02,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:35:45,218 INFO [train.py:968] (1/2) Epoch 9, batch 16500, giga_loss[loss=0.233, simple_loss=0.3287, pruned_loss=0.06859, over 28576.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09024, over 5673536.84 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3406, pruned_loss=0.09594, over 5759271.17 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3364, pruned_loss=0.0901, over 5667017.24 frames. ], batch size: 307, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:36:14,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1620, 3.9924, 3.7582, 1.7955], device='cuda:1'), covar=tensor([0.0475, 0.0602, 0.0670, 0.2127], device='cuda:1'), in_proj_covar=tensor([0.0955, 0.0889, 0.0788, 0.0619], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 17:36:24,415 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 16550, giga_loss[loss=0.271, simple_loss=0.3606, pruned_loss=0.09071, over 28134.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3385, pruned_loss=0.08982, over 5678035.77 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3397, pruned_loss=0.0955, over 5754844.41 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3394, pruned_loss=0.08996, over 5673751.12 frames. ], batch size: 412, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:37:21,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1129, 1.2453, 3.7625, 3.2376], device='cuda:1'), covar=tensor([0.1563, 0.2478, 0.0371, 0.1457], device='cuda:1'), in_proj_covar=tensor([0.0614, 0.0570, 0.0805, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 17:37:34,037 INFO [train.py:968] (1/2) Epoch 9, batch 16600, giga_loss[loss=0.2642, simple_loss=0.3458, pruned_loss=0.09127, over 29053.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.34, pruned_loss=0.09077, over 5672409.95 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3398, pruned_loss=0.0956, over 5755684.90 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3406, pruned_loss=0.09072, over 5667187.42 frames. ], batch size: 128, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:38:20,992 INFO [optim.py:369] (1/2) 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 17:38:39,636 INFO [train.py:968] (1/2) Epoch 9, batch 16650, giga_loss[loss=0.2514, simple_loss=0.3409, pruned_loss=0.08098, over 28989.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3391, pruned_loss=0.09038, over 5663238.00 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3397, pruned_loss=0.09559, over 5756876.02 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3396, pruned_loss=0.09031, over 5657225.06 frames. ], batch size: 145, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:38:44,619 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:39:11,955 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:39:46,151 INFO [train.py:968] (1/2) Epoch 9, batch 16700, giga_loss[loss=0.2683, simple_loss=0.3447, pruned_loss=0.0959, over 29058.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3385, pruned_loss=0.09011, over 5658689.91 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3397, pruned_loss=0.0955, over 5759454.80 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.339, pruned_loss=0.09005, over 5649898.61 frames. ], batch size: 214, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:40:17,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=380921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:40:32,462 INFO [optim.py:369] (1/2) 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 17:40:50,691 INFO [train.py:968] (1/2) Epoch 9, batch 16750, giga_loss[loss=0.2662, simple_loss=0.3521, pruned_loss=0.09011, over 28742.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3384, pruned_loss=0.08922, over 5659876.07 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3394, pruned_loss=0.09541, over 5753757.15 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.339, pruned_loss=0.08911, over 5655030.50 frames. ], batch size: 243, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:40:56,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=380951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:41:57,467 INFO [train.py:968] (1/2) Epoch 9, batch 16800, giga_loss[loss=0.2546, simple_loss=0.3353, pruned_loss=0.08694, over 28731.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3386, pruned_loss=0.08938, over 5658774.08 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3394, pruned_loss=0.09525, over 5757676.57 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3391, pruned_loss=0.08922, over 5647392.29 frames. ], batch size: 307, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:42:44,858 INFO [optim.py:369] (1/2) 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 17:42:45,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:42:50,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3750, 1.7031, 1.3347, 1.4420], device='cuda:1'), covar=tensor([0.0754, 0.0284, 0.0330, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0119, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0053, 0.0049, 0.0082], device='cuda:1')
+2023-03-04 17:43:03,059 INFO [train.py:968] (1/2) Epoch 9, batch 16850, giga_loss[loss=0.2845, simple_loss=0.3606, pruned_loss=0.1042, over 27814.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3424, pruned_loss=0.09103, over 5655156.38 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3391, pruned_loss=0.09517, over 5747150.94 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.343, pruned_loss=0.09089, over 5653525.69 frames. ], batch size: 474, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:44:07,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:44:11,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:44:12,221 INFO [train.py:968] (1/2) Epoch 9, batch 16900, giga_loss[loss=0.2853, simple_loss=0.3558, pruned_loss=0.1073, over 27646.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3425, pruned_loss=0.09105, over 5668663.85 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3393, pruned_loss=0.09533, over 5749710.51 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3429, pruned_loss=0.09073, over 5663845.44 frames. ], batch size: 472, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:44:51,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:45:03,085 INFO [optim.py:369] (1/2) 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 17:45:12,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-04 17:45:18,609 INFO [train.py:968] (1/2) Epoch 9, batch 16950, giga_loss[loss=0.2719, simple_loss=0.3464, pruned_loss=0.09876, over 28886.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3413, pruned_loss=0.09114, over 5671670.29 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3395, pruned_loss=0.09535, over 5752167.46 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3415, pruned_loss=0.09078, over 5663933.11 frames. ], batch size: 164, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:45:49,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 17:46:01,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4134, 1.6389, 1.6555, 1.2776], device='cuda:1'), covar=tensor([0.1441, 0.2023, 0.1175, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0688, 0.0825, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 17:46:05,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:46:09,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:46:14,716 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:46:30,848 INFO [train.py:968] (1/2) Epoch 9, batch 17000, giga_loss[loss=0.2339, simple_loss=0.3293, pruned_loss=0.06922, over 29076.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3397, pruned_loss=0.08997, over 5675592.62 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3395, pruned_loss=0.09534, over 5750225.59 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3398, pruned_loss=0.08965, over 5670611.82 frames. ], batch size: 128, lr: 3.63e-03, grad_scale: 4.0
+2023-03-04 17:46:50,536 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:47:07,988 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:47:24,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-04 17:47:34,563 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 17050, giga_loss[loss=0.2634, simple_loss=0.3437, pruned_loss=0.09157, over 28483.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3385, pruned_loss=0.08925, over 5663153.04 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3393, pruned_loss=0.09525, over 5744131.22 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3389, pruned_loss=0.08893, over 5662381.55 frames. ], batch size: 336, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:48:31,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:48:33,485 INFO [train.py:968] (1/2) Epoch 9, batch 17100, giga_loss[loss=0.2918, simple_loss=0.3573, pruned_loss=0.1131, over 26775.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3393, pruned_loss=0.0897, over 5669745.35 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.339, pruned_loss=0.09509, over 5746505.50 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3398, pruned_loss=0.08949, over 5665696.47 frames. ], batch size: 555, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:49:16,205 INFO [optim.py:369] (1/2) 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 17:49:32,146 INFO [train.py:968] (1/2) Epoch 9, batch 17150, giga_loss[loss=0.3142, simple_loss=0.3838, pruned_loss=0.1223, over 28094.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3433, pruned_loss=0.09227, over 5661512.98 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3392, pruned_loss=0.09525, over 5738962.21 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3435, pruned_loss=0.0919, over 5664402.47 frames. ], batch size: 412, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:49:49,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3161, 1.4489, 1.0373, 1.1440], device='cuda:1'), covar=tensor([0.0862, 0.0553, 0.1169, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0436, 0.0497, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 17:49:55,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:49:59,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:17,089 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:26,033 INFO [train.py:968] (1/2) Epoch 9, batch 17200, giga_loss[loss=0.295, simple_loss=0.3564, pruned_loss=0.1167, over 28137.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3423, pruned_loss=0.0927, over 5666456.47 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.339, pruned_loss=0.09513, over 5742912.82 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3428, pruned_loss=0.09248, over 5663235.10 frames. ], batch size: 412, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:50:29,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:56,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:50:56,217 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:51:05,813 INFO [optim.py:369] (1/2) 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 17:51:11,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:51:14,271 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 17250, giga_loss[loss=0.3497, simple_loss=0.3911, pruned_loss=0.1541, over 28038.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3395, pruned_loss=0.09261, over 5661335.22 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3391, pruned_loss=0.09528, over 5741469.42 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3398, pruned_loss=0.09224, over 5658162.52 frames. ], batch size: 412, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:51:48,489 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:51:59,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5459, 1.6935, 1.4338, 1.6614], device='cuda:1'), covar=tensor([0.2403, 0.2253, 0.2492, 0.2109], device='cuda:1'), in_proj_covar=tensor([0.1232, 0.0918, 0.1096, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 17:52:17,554 INFO [train.py:968] (1/2) Epoch 9, batch 17300, giga_loss[loss=0.2382, simple_loss=0.3234, pruned_loss=0.07647, over 28714.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3397, pruned_loss=0.09335, over 5658748.85 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3391, pruned_loss=0.09526, over 5743330.03 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.34, pruned_loss=0.09305, over 5653466.04 frames. ], batch size: 307, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:52:35,785 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:52:57,966 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 17350, giga_loss[loss=0.3343, simple_loss=0.3982, pruned_loss=0.1352, over 28382.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3464, pruned_loss=0.09763, over 5662184.05 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.339, pruned_loss=0.09511, over 5747436.45 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3468, pruned_loss=0.09754, over 5652244.54 frames. ], batch size: 65, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:53:23,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:53:53,462 INFO [train.py:968] (1/2) Epoch 9, batch 17400, giga_loss[loss=0.3198, simple_loss=0.396, pruned_loss=0.1218, over 28665.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3543, pruned_loss=0.1023, over 5673534.93 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3383, pruned_loss=0.09466, over 5752313.29 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3558, pruned_loss=0.1028, over 5658768.28 frames. ], batch size: 307, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:54:13,951 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3360, 1.4201, 1.2279, 1.3022], device='cuda:1'), covar=tensor([0.1619, 0.1467, 0.1330, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.1624, 0.1455, 0.1401, 0.1530], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 17:54:14,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4346, 1.5624, 1.2937, 1.6273], device='cuda:1'), covar=tensor([0.2250, 0.2179, 0.2445, 0.2002], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.0913, 0.1091, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 17:54:23,855 INFO [optim.py:369] (1/2) 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 17:54:33,409 INFO [train.py:968] (1/2) Epoch 9, batch 17450, giga_loss[loss=0.2764, simple_loss=0.3639, pruned_loss=0.0945, over 28971.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.358, pruned_loss=0.1047, over 5682717.18 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3381, pruned_loss=0.09453, over 5756635.33 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3598, pruned_loss=0.1055, over 5664956.49 frames. ], batch size: 164, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:54:49,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7094, 1.8329, 1.4106, 1.3840], device='cuda:1'), covar=tensor([0.1560, 0.1328, 0.1213, 0.1297], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1457, 0.1403, 0.1532], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 17:55:15,549 INFO [train.py:968] (1/2) Epoch 9, batch 17500, giga_loss[loss=0.2754, simple_loss=0.3462, pruned_loss=0.1023, over 28866.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3552, pruned_loss=0.104, over 5684294.81 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3383, pruned_loss=0.09454, over 5758233.31 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3571, pruned_loss=0.105, over 5666046.52 frames. ], batch size: 227, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:55:16,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2936, 1.4769, 1.1881, 1.2846], device='cuda:1'), covar=tensor([0.1561, 0.1239, 0.1143, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.1631, 0.1461, 0.1407, 0.1536], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 17:55:20,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:55:22,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:55:40,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 17:55:46,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:55:46,997 INFO [optim.py:369] (1/2) 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 17:55:56,714 INFO [train.py:968] (1/2) Epoch 9, batch 17550, giga_loss[loss=0.2809, simple_loss=0.3431, pruned_loss=0.1093, over 28652.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3486, pruned_loss=0.1011, over 5690890.30 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3383, pruned_loss=0.09438, over 5761967.18 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3505, pruned_loss=0.1022, over 5670805.00 frames. ], batch size: 262, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:55:56,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3617, 1.4921, 1.5490, 1.3181], device='cuda:1'), covar=tensor([0.1199, 0.1252, 0.1633, 0.1377], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0711, 0.0640, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 17:56:38,762 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:56:39,093 INFO [train.py:968] (1/2) Epoch 9, batch 17600, giga_loss[loss=0.2449, simple_loss=0.3184, pruned_loss=0.08573, over 28866.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3421, pruned_loss=0.09845, over 5694779.38 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3384, pruned_loss=0.09443, over 5760814.97 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3437, pruned_loss=0.09947, over 5677535.64 frames. ], batch size: 186, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:56:48,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-04 17:57:09,775 INFO [optim.py:369] (1/2) 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 17:57:21,065 INFO [train.py:968] (1/2) Epoch 9, batch 17650, giga_loss[loss=0.2386, simple_loss=0.3061, pruned_loss=0.08553, over 28471.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3349, pruned_loss=0.09531, over 5697746.90 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3385, pruned_loss=0.09438, over 5762568.34 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.336, pruned_loss=0.09623, over 5680210.18 frames. ], batch size: 85, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:57:52,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=381887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:57:52,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=381887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:58:00,479 INFO [train.py:968] (1/2) Epoch 9, batch 17700, giga_loss[loss=0.2103, simple_loss=0.2913, pruned_loss=0.06462, over 29013.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3279, pruned_loss=0.09189, over 5699547.77 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3387, pruned_loss=0.09437, over 5762712.95 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3283, pruned_loss=0.09259, over 5683430.31 frames. ], batch size: 136, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 17:58:31,656 INFO [optim.py:369] (1/2) 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 17:58:34,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=381940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:58:35,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=381943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:58:39,000 INFO [train.py:968] (1/2) Epoch 9, batch 17750, giga_loss[loss=0.2391, simple_loss=0.3014, pruned_loss=0.0884, over 28779.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3239, pruned_loss=0.09023, over 5702834.61 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.339, pruned_loss=0.09457, over 5765506.18 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3237, pruned_loss=0.09053, over 5686453.93 frames. ], batch size: 92, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:58:57,163 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=381972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:16,798 INFO [train.py:968] (1/2) Epoch 9, batch 17800, libri_loss[loss=0.2551, simple_loss=0.3371, pruned_loss=0.08653, over 29518.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3218, pruned_loss=0.08904, over 5710162.49 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3397, pruned_loss=0.09486, over 5769695.54 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3204, pruned_loss=0.08885, over 5690729.04 frames. ], batch size: 84, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 17:59:34,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:44,897 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:47,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 17:59:51,438 INFO [optim.py:369] (1/2) 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 18:00:02,583 INFO [train.py:968] (1/2) Epoch 9, batch 17850, giga_loss[loss=0.2142, simple_loss=0.282, pruned_loss=0.07323, over 28461.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3187, pruned_loss=0.0877, over 5702992.82 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.34, pruned_loss=0.09485, over 5768268.60 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.317, pruned_loss=0.08744, over 5687638.24 frames. ], batch size: 78, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:00:14,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=382062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:00:44,090 INFO [train.py:968] (1/2) Epoch 9, batch 17900, giga_loss[loss=0.2163, simple_loss=0.2941, pruned_loss=0.06925, over 28947.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3154, pruned_loss=0.08646, over 5697470.61 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3402, pruned_loss=0.09493, over 5769185.26 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3138, pruned_loss=0.08616, over 5684424.56 frames. ], batch size: 164, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:00:52,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0642, 2.7123, 2.1017, 1.6481], device='cuda:1'), covar=tensor([0.1713, 0.0960, 0.1020, 0.1527], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1454, 0.1403, 0.1527], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:01:20,540 INFO [optim.py:369] (1/2) 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 18:01:30,720 INFO [train.py:968] (1/2) Epoch 9, batch 17950, giga_loss[loss=0.2246, simple_loss=0.2961, pruned_loss=0.07655, over 28718.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3113, pruned_loss=0.08433, over 5700036.91 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3402, pruned_loss=0.09486, over 5769879.33 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3099, pruned_loss=0.08411, over 5688967.05 frames. ], batch size: 92, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:02:04,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 18:02:09,832 INFO [train.py:968] (1/2) Epoch 9, batch 18000, giga_loss[loss=0.1972, simple_loss=0.2658, pruned_loss=0.06427, over 28541.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3101, pruned_loss=0.08344, over 5696759.69 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3409, pruned_loss=0.09501, over 5772065.49 frames. ], giga_tot_loss[loss=0.236, simple_loss=0.3069, pruned_loss=0.08259, over 5682342.22 frames. ], batch size: 85, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:02:09,832 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 18:02:18,282 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 18:02:44,330 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-04 18:02:49,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 18:02:49,406 INFO [optim.py:369] (1/2) 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 18:02:57,766 INFO [train.py:968] (1/2) Epoch 9, batch 18050, giga_loss[loss=0.2083, simple_loss=0.2858, pruned_loss=0.06534, over 27832.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3075, pruned_loss=0.08209, over 5696701.41 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3416, pruned_loss=0.09522, over 5769417.99 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3034, pruned_loss=0.08089, over 5685407.11 frames. ], batch size: 412, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:03:11,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:03:40,673 INFO [train.py:968] (1/2) Epoch 9, batch 18100, giga_loss[loss=0.233, simple_loss=0.2953, pruned_loss=0.0853, over 28508.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3071, pruned_loss=0.08179, over 5701709.08 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3424, pruned_loss=0.09563, over 5764925.15 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3016, pruned_loss=0.07988, over 5693422.61 frames. ], batch size: 85, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:04:12,738 INFO [optim.py:369] (1/2) 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 18:04:15,096 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 18150, giga_loss[loss=0.2574, simple_loss=0.3257, pruned_loss=0.09454, over 29057.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3053, pruned_loss=0.08174, over 5699036.93 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3429, pruned_loss=0.09595, over 5762960.03 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.2996, pruned_loss=0.07963, over 5692999.28 frames. ], batch size: 155, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:05:09,128 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 18200, giga_loss[loss=0.3205, simple_loss=0.3761, pruned_loss=0.1325, over 26674.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3124, pruned_loss=0.08607, over 5699773.83 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3429, pruned_loss=0.09601, over 5765025.25 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3075, pruned_loss=0.08419, over 5692159.63 frames. ], batch size: 555, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:05:19,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:05:22,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:05:45,091 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-04 18:05:48,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=382437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:05:48,497 INFO [optim.py:369] (1/2) 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 18:05:57,161 INFO [train.py:968] (1/2) Epoch 9, batch 18250, giga_loss[loss=0.3305, simple_loss=0.3944, pruned_loss=0.1333, over 28633.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3254, pruned_loss=0.09254, over 5702560.56 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3429, pruned_loss=0.0959, over 5766183.46 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3205, pruned_loss=0.09084, over 5692907.29 frames. ], batch size: 242, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:06:03,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382456.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:06:29,079 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:06:36,073 INFO [train.py:968] (1/2) Epoch 9, batch 18300, giga_loss[loss=0.269, simple_loss=0.3514, pruned_loss=0.09328, over 29026.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3372, pruned_loss=0.09903, over 5707326.66 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3431, pruned_loss=0.09589, over 5770906.48 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3328, pruned_loss=0.09773, over 5693122.91 frames. ], batch size: 155, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:06:57,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-04 18:06:59,843 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-04 18:07:04,585 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:07:06,234 INFO [optim.py:369] (1/2) 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 18:07:06,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:07:12,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 18:07:13,707 INFO [train.py:968] (1/2) Epoch 9, batch 18350, giga_loss[loss=0.2849, simple_loss=0.3669, pruned_loss=0.1015, over 28023.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3448, pruned_loss=0.1021, over 5701684.07 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3429, pruned_loss=0.09568, over 5764384.73 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3413, pruned_loss=0.1014, over 5693462.11 frames. ], batch size: 412, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:07:30,754 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=382567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:07:57,248 INFO [train.py:968] (1/2) Epoch 9, batch 18400, giga_loss[loss=0.3084, simple_loss=0.3726, pruned_loss=0.1221, over 28280.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.348, pruned_loss=0.1023, over 5699609.27 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3428, pruned_loss=0.09557, over 5765532.24 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3454, pruned_loss=0.102, over 5690680.56 frames. ], batch size: 77, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:07:59,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:08:30,174 INFO [optim.py:369] (1/2) 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 18:08:39,263 INFO [train.py:968] (1/2) Epoch 9, batch 18450, giga_loss[loss=0.2613, simple_loss=0.3394, pruned_loss=0.09167, over 28790.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3502, pruned_loss=0.1025, over 5694639.36 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3433, pruned_loss=0.09582, over 5767153.72 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3479, pruned_loss=0.1021, over 5684824.27 frames. ], batch size: 119, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:09:22,080 INFO [train.py:968] (1/2) Epoch 9, batch 18500, giga_loss[loss=0.2923, simple_loss=0.3623, pruned_loss=0.1112, over 29064.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3519, pruned_loss=0.1034, over 5694884.71 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.344, pruned_loss=0.09621, over 5770440.41 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3496, pruned_loss=0.103, over 5682045.69 frames. ], batch size: 128, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:09:35,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:09:53,828 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=382735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:09:56,161 INFO [optim.py:369] (1/2) 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 18:10:03,822 INFO [train.py:968] (1/2) Epoch 9, batch 18550, giga_loss[loss=0.3064, simple_loss=0.3731, pruned_loss=0.1198, over 28941.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3541, pruned_loss=0.1056, over 5695572.37 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3435, pruned_loss=0.0959, over 5771823.43 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3529, pruned_loss=0.1057, over 5682783.65 frames. ], batch size: 213, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:10:37,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3609, 1.3970, 1.4105, 1.3864], device='cuda:1'), covar=tensor([0.1316, 0.1554, 0.1916, 0.1465], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0727, 0.0649, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 18:10:46,905 INFO [train.py:968] (1/2) Epoch 9, batch 18600, libri_loss[loss=0.2969, simple_loss=0.3684, pruned_loss=0.1127, over 29520.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3584, pruned_loss=0.1086, over 5705068.48 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09648, over 5774814.91 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.357, pruned_loss=0.1085, over 5689687.48 frames. ], batch size: 82, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:11:12,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382831.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:11:18,634 INFO [optim.py:369] (1/2) 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 18:11:27,178 INFO [train.py:968] (1/2) Epoch 9, batch 18650, giga_loss[loss=0.2843, simple_loss=0.3614, pruned_loss=0.1036, over 28980.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3615, pruned_loss=0.11, over 5707927.31 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3448, pruned_loss=0.09668, over 5778571.74 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3605, pruned_loss=0.1102, over 5690313.50 frames. ], batch size: 186, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:11:35,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:11:37,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:11:38,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:11:45,423 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-04 18:12:00,935 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 18700, giga_loss[loss=0.2784, simple_loss=0.3545, pruned_loss=0.1012, over 28958.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.363, pruned_loss=0.1098, over 5717916.66 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3447, pruned_loss=0.09656, over 5780728.97 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3626, pruned_loss=0.1103, over 5700843.92 frames. ], batch size: 213, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:12:29,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5756, 2.2269, 1.6344, 0.8386], device='cuda:1'), covar=tensor([0.3476, 0.1904, 0.2838, 0.3498], device='cuda:1'), in_proj_covar=tensor([0.1495, 0.1426, 0.1462, 0.1225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 18:12:38,587 INFO [optim.py:369] (1/2) 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 18:12:44,700 INFO [train.py:968] (1/2) Epoch 9, batch 18750, giga_loss[loss=0.2656, simple_loss=0.3537, pruned_loss=0.08878, over 29102.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3637, pruned_loss=0.1092, over 5718800.24 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3449, pruned_loss=0.0965, over 5784013.63 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3637, pruned_loss=0.1099, over 5700806.12 frames. ], batch size: 128, lr: 3.62e-03, grad_scale: 4.0
+2023-03-04 18:13:06,176 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=382974.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:13:07,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=382976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:08,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=382977.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:13:22,280 INFO [train.py:968] (1/2) Epoch 9, batch 18800, giga_loss[loss=0.285, simple_loss=0.3589, pruned_loss=0.1055, over 28869.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3634, pruned_loss=0.1081, over 5699081.81 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3452, pruned_loss=0.09656, over 5770987.98 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3637, pruned_loss=0.109, over 5694859.07 frames. ], batch size: 186, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:13:30,441 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:31,158 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383006.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:13:32,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:35,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-04 18:13:36,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3814, 1.5632, 1.3100, 1.5416], device='cuda:1'), covar=tensor([0.2346, 0.2306, 0.2458, 0.2184], device='cuda:1'), in_proj_covar=tensor([0.1227, 0.0920, 0.1090, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:13:40,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:50,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:55,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:13:57,217 INFO [optim.py:369] (1/2) 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 18:14:04,316 INFO [train.py:968] (1/2) Epoch 9, batch 18850, giga_loss[loss=0.2653, simple_loss=0.3449, pruned_loss=0.09289, over 28888.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3622, pruned_loss=0.1065, over 5698176.24 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3455, pruned_loss=0.09664, over 5771545.34 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3625, pruned_loss=0.1074, over 5692145.83 frames. ], batch size: 213, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:14:44,437 INFO [train.py:968] (1/2) Epoch 9, batch 18900, giga_loss[loss=0.2793, simple_loss=0.3561, pruned_loss=0.1013, over 28349.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3598, pruned_loss=0.1041, over 5707943.07 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3455, pruned_loss=0.09664, over 5771545.34 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3601, pruned_loss=0.1048, over 5703249.53 frames. ], batch size: 65, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:14:53,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:15:03,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:15:03,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1944, 1.5532, 1.5060, 1.1188], device='cuda:1'), covar=tensor([0.1508, 0.2027, 0.1213, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0693, 0.0839, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 18:15:05,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:15:08,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4420, 2.0948, 1.5622, 0.6697], device='cuda:1'), covar=tensor([0.3443, 0.1889, 0.2933, 0.3911], device='cuda:1'), in_proj_covar=tensor([0.1507, 0.1430, 0.1470, 0.1234], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 18:15:18,567 INFO [optim.py:369] (1/2) 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 18:15:25,926 INFO [train.py:968] (1/2) Epoch 9, batch 18950, giga_loss[loss=0.3211, simple_loss=0.3797, pruned_loss=0.1312, over 28355.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3616, pruned_loss=0.1061, over 5704637.66 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.346, pruned_loss=0.0967, over 5775225.94 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3618, pruned_loss=0.1068, over 5696029.56 frames. ], batch size: 71, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:15:29,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:16:09,755 INFO [train.py:968] (1/2) Epoch 9, batch 19000, giga_loss[loss=0.3471, simple_loss=0.4018, pruned_loss=0.1462, over 28227.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3644, pruned_loss=0.1108, over 5687562.09 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3461, pruned_loss=0.09684, over 5766375.89 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3648, pruned_loss=0.1115, over 5686025.91 frames. ], batch size: 368, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:16:17,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4404, 2.0782, 1.6409, 0.5590], device='cuda:1'), covar=tensor([0.3400, 0.1702, 0.2391, 0.3655], device='cuda:1'), in_proj_covar=tensor([0.1505, 0.1431, 0.1466, 0.1231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 18:16:39,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5885, 1.6912, 1.3818, 1.9682], device='cuda:1'), covar=tensor([0.2250, 0.2216, 0.2344, 0.2124], device='cuda:1'), in_proj_covar=tensor([0.1221, 0.0915, 0.1084, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:16:43,223 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 19050, giga_loss[loss=0.3166, simple_loss=0.3724, pruned_loss=0.1304, over 28944.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3663, pruned_loss=0.1136, over 5687241.41 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3467, pruned_loss=0.09713, over 5768928.16 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3665, pruned_loss=0.1143, over 5681855.97 frames. ], batch size: 106, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:16:55,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:16:57,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:17:21,389 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:17:28,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6004, 1.5386, 1.1796, 1.1571], device='cuda:1'), covar=tensor([0.0661, 0.0558, 0.0901, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0438, 0.0497, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 18:17:31,170 INFO [train.py:968] (1/2) Epoch 9, batch 19100, giga_loss[loss=0.2689, simple_loss=0.3423, pruned_loss=0.09769, over 28730.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3654, pruned_loss=0.1141, over 5687869.74 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3468, pruned_loss=0.09717, over 5761935.12 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3659, pruned_loss=0.115, over 5688079.62 frames. ], batch size: 60, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:17:42,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:17:56,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:18:03,180 INFO [optim.py:369] (1/2) 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 18:18:11,161 INFO [train.py:968] (1/2) Epoch 9, batch 19150, giga_loss[loss=0.2741, simple_loss=0.3487, pruned_loss=0.09972, over 29043.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3632, pruned_loss=0.1129, over 5692822.76 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3468, pruned_loss=0.09714, over 5760142.88 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3639, pruned_loss=0.1139, over 5693268.11 frames. ], batch size: 128, lr: 3.62e-03, grad_scale: 8.0
+2023-03-04 18:18:29,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6655, 1.7370, 1.9025, 1.5008], device='cuda:1'), covar=tensor([0.1744, 0.1931, 0.1318, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0698, 0.0838, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 18:18:52,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:18:54,913 INFO [train.py:968] (1/2) Epoch 9, batch 19200, libri_loss[loss=0.2253, simple_loss=0.3105, pruned_loss=0.06999, over 29577.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3636, pruned_loss=0.1131, over 5681319.57 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09742, over 5755653.21 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3641, pruned_loss=0.1141, over 5683411.14 frames. ], batch size: 75, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:19:02,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:19:04,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5621, 1.1966, 5.1205, 3.6312], device='cuda:1'), covar=tensor([0.1655, 0.2648, 0.0335, 0.0725], device='cuda:1'), in_proj_covar=tensor([0.0617, 0.0566, 0.0811, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:19:26,842 INFO [optim.py:369] (1/2) 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 18:19:33,286 INFO [train.py:968] (1/2) Epoch 9, batch 19250, giga_loss[loss=0.3386, simple_loss=0.3873, pruned_loss=0.145, over 28309.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3611, pruned_loss=0.1102, over 5694958.39 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3482, pruned_loss=0.09772, over 5762008.27 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3614, pruned_loss=0.1113, over 5688004.67 frames. ], batch size: 77, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:19:40,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7898, 2.1402, 1.8685, 1.7210], device='cuda:1'), covar=tensor([0.1823, 0.1389, 0.1483, 0.1597], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1493, 0.1439, 0.1562], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:20:11,109 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:20:16,971 INFO [train.py:968] (1/2) Epoch 9, batch 19300, giga_loss[loss=0.2549, simple_loss=0.3288, pruned_loss=0.09053, over 28644.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3565, pruned_loss=0.1069, over 5695459.47 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3482, pruned_loss=0.09773, over 5764995.20 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3569, pruned_loss=0.1079, over 5685822.42 frames. ], batch size: 307, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:20:30,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-04 18:20:35,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5031, 1.6605, 1.7822, 1.3912], device='cuda:1'), covar=tensor([0.1473, 0.1905, 0.1141, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0694, 0.0834, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 18:20:53,771 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:20:55,741 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 19350, giga_loss[loss=0.2988, simple_loss=0.3662, pruned_loss=0.1157, over 28904.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3511, pruned_loss=0.1042, over 5687061.53 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3483, pruned_loss=0.09788, over 5766100.21 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3514, pruned_loss=0.105, over 5677459.96 frames. ], batch size: 186, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:21:03,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:07,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:08,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8593, 1.8568, 1.7835, 1.6973], device='cuda:1'), covar=tensor([0.1310, 0.2000, 0.1820, 0.1759], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0728, 0.0657, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 18:21:22,179 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:33,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:21:48,157 INFO [train.py:968] (1/2) Epoch 9, batch 19400, giga_loss[loss=0.2542, simple_loss=0.3226, pruned_loss=0.09294, over 28698.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3456, pruned_loss=0.1017, over 5682490.48 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.09782, over 5769477.37 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3458, pruned_loss=0.1025, over 5670068.27 frames. ], batch size: 262, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:21:49,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4324, 3.5012, 1.6043, 1.5495], device='cuda:1'), covar=tensor([0.0985, 0.0220, 0.0871, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0487, 0.0325, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 18:22:09,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-04 18:22:24,109 INFO [optim.py:369] (1/2) 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 18:22:31,758 INFO [train.py:968] (1/2) Epoch 9, batch 19450, giga_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 28579.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.342, pruned_loss=0.09933, over 5696652.69 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3491, pruned_loss=0.09831, over 5774189.99 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3413, pruned_loss=0.09964, over 5679271.86 frames. ], batch size: 336, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:23:07,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:23:15,407 INFO [train.py:968] (1/2) Epoch 9, batch 19500, libri_loss[loss=0.3041, simple_loss=0.3837, pruned_loss=0.1123, over 26013.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3436, pruned_loss=0.09983, over 5697007.47 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3495, pruned_loss=0.09846, over 5772549.92 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3425, pruned_loss=0.09995, over 5683928.26 frames. ], batch size: 136, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:23:22,123 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:23:59,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 18:24:01,296 INFO [train.py:968] (1/2) Epoch 9, batch 19550, giga_loss[loss=0.2703, simple_loss=0.3465, pruned_loss=0.097, over 28887.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3431, pruned_loss=0.09896, over 5701028.78 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3498, pruned_loss=0.09841, over 5773255.88 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.342, pruned_loss=0.09913, over 5688795.37 frames. ], batch size: 145, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:24:08,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6326, 4.3375, 1.7830, 1.7512], device='cuda:1'), covar=tensor([0.0873, 0.0233, 0.0797, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0482, 0.0321, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 18:24:35,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4931, 1.6383, 1.4170, 1.6481], device='cuda:1'), covar=tensor([0.2103, 0.2198, 0.2332, 0.2123], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.0916, 0.1089, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:24:38,890 INFO [train.py:968] (1/2) Epoch 9, batch 19600, giga_loss[loss=0.2856, simple_loss=0.3489, pruned_loss=0.1111, over 28801.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3427, pruned_loss=0.09882, over 5711659.46 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3506, pruned_loss=0.09864, over 5775187.30 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3407, pruned_loss=0.09877, over 5697480.55 frames. ], batch size: 99, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:24:59,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2995, 1.3373, 1.1177, 1.1385], device='cuda:1'), covar=tensor([0.1552, 0.1407, 0.1098, 0.1428], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1478, 0.1436, 0.1562], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:25:06,813 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:08,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:13,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 18:25:13,681 INFO [optim.py:369] (1/2) 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 18:25:18,483 INFO [train.py:968] (1/2) Epoch 9, batch 19650, giga_loss[loss=0.3225, simple_loss=0.3818, pruned_loss=0.1317, over 28026.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.34, pruned_loss=0.09714, over 5720080.16 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3512, pruned_loss=0.09863, over 5775211.03 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3375, pruned_loss=0.09706, over 5706883.89 frames. ], batch size: 412, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:25:19,644 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=383849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:22,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=383852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:29,705 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:31,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=383864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:34,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3281, 3.3114, 1.5845, 1.2776], device='cuda:1'), covar=tensor([0.0969, 0.0297, 0.0836, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0482, 0.0321, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0023], device='cuda:1')
+2023-03-04 18:25:45,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2917, 1.7644, 1.1995, 0.6976], device='cuda:1'), covar=tensor([0.3555, 0.1640, 0.2264, 0.3958], device='cuda:1'), in_proj_covar=tensor([0.1498, 0.1417, 0.1462, 0.1225], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 18:25:46,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=383881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:25:49,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4738, 1.5738, 1.3458, 1.6627], device='cuda:1'), covar=tensor([0.2216, 0.2224, 0.2359, 0.2086], device='cuda:1'), in_proj_covar=tensor([0.1228, 0.0918, 0.1090, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:26:00,646 INFO [train.py:968] (1/2) Epoch 9, batch 19700, giga_loss[loss=0.286, simple_loss=0.3457, pruned_loss=0.1132, over 28975.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3372, pruned_loss=0.09608, over 5721078.64 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3513, pruned_loss=0.09862, over 5775966.62 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3349, pruned_loss=0.09601, over 5709487.32 frames. ], batch size: 213, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:26:05,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3912, 1.7043, 1.6815, 1.2628], device='cuda:1'), covar=tensor([0.1512, 0.2021, 0.1179, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0698, 0.0839, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-04 18:26:08,883 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-04 18:26:22,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:26:33,598 INFO [optim.py:369] (1/2) 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 18:26:36,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7585, 1.8523, 2.0098, 1.5985], device='cuda:1'), covar=tensor([0.1734, 0.2029, 0.1290, 0.1457], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0699, 0.0841, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-04 18:26:39,546 INFO [train.py:968] (1/2) Epoch 9, batch 19750, giga_loss[loss=0.2986, simple_loss=0.3598, pruned_loss=0.1187, over 28252.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3361, pruned_loss=0.09583, over 5723685.30 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3519, pruned_loss=0.09875, over 5779094.92 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3334, pruned_loss=0.0956, over 5710528.49 frames. ], batch size: 368, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:26:48,308 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:26:54,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=383965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:01,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9753, 4.7805, 4.4809, 2.0090], device='cuda:1'), covar=tensor([0.0366, 0.0504, 0.0550, 0.2214], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0884, 0.0785, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 18:27:17,430 INFO [train.py:968] (1/2) Epoch 9, batch 19800, giga_loss[loss=0.2477, simple_loss=0.313, pruned_loss=0.0912, over 28125.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3338, pruned_loss=0.09472, over 5731268.79 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3524, pruned_loss=0.09891, over 5781880.70 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3308, pruned_loss=0.09429, over 5716930.37 frames. ], batch size: 77, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:27:26,193 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:28,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:43,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-04 18:27:50,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:27:51,672 INFO [optim.py:369] (1/2) 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 18:27:56,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384046.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:27:57,208 INFO [train.py:968] (1/2) Epoch 9, batch 19850, giga_loss[loss=0.2366, simple_loss=0.3068, pruned_loss=0.08315, over 28621.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3335, pruned_loss=0.09445, over 5730249.93 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3534, pruned_loss=0.09901, over 5781899.01 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3294, pruned_loss=0.09382, over 5716253.40 frames. ], batch size: 85, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:28:30,299 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:28:37,211 INFO [train.py:968] (1/2) Epoch 9, batch 19900, giga_loss[loss=0.2946, simple_loss=0.3576, pruned_loss=0.1159, over 27853.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3325, pruned_loss=0.0943, over 5726524.79 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3535, pruned_loss=0.09894, over 5784473.05 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3287, pruned_loss=0.09379, over 5712440.05 frames. ], batch size: 412, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:28:42,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7864, 2.3025, 1.6591, 1.3131], device='cuda:1'), covar=tensor([0.2207, 0.1316, 0.1590, 0.2125], device='cuda:1'), in_proj_covar=tensor([0.1609, 0.1473, 0.1435, 0.1566], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:29:08,617 INFO [optim.py:369] (1/2) 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 18:29:13,278 INFO [train.py:968] (1/2) Epoch 9, batch 19950, giga_loss[loss=0.2458, simple_loss=0.3182, pruned_loss=0.08669, over 28655.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3316, pruned_loss=0.09347, over 5738846.46 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3546, pruned_loss=0.09946, over 5790387.25 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3266, pruned_loss=0.09237, over 5719680.59 frames. ], batch size: 242, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:29:23,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5246, 1.6714, 1.5920, 1.6139], device='cuda:1'), covar=tensor([0.1498, 0.1912, 0.2022, 0.1703], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0739, 0.0662, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:29:51,537 INFO [train.py:968] (1/2) Epoch 9, batch 20000, giga_loss[loss=0.2303, simple_loss=0.3063, pruned_loss=0.07716, over 28973.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3312, pruned_loss=0.09317, over 5730572.53 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3562, pruned_loss=0.1004, over 5781079.45 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3246, pruned_loss=0.09112, over 5721564.40 frames. ], batch size: 227, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:30:26,748 INFO [optim.py:369] (1/2) 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 18:30:31,121 INFO [train.py:968] (1/2) Epoch 9, batch 20050, giga_loss[loss=0.2448, simple_loss=0.3136, pruned_loss=0.08801, over 28745.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3293, pruned_loss=0.09202, over 5734650.49 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.357, pruned_loss=0.1007, over 5779288.63 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.323, pruned_loss=0.09001, over 5728558.47 frames. ], batch size: 99, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:30:54,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:01,732 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:11,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3320, 1.4199, 1.0315, 1.2438], device='cuda:1'), covar=tensor([0.1281, 0.1257, 0.1243, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1460, 0.1419, 0.1552], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:31:13,156 INFO [train.py:968] (1/2) Epoch 9, batch 20100, giga_loss[loss=0.2786, simple_loss=0.3492, pruned_loss=0.104, over 28619.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.333, pruned_loss=0.09468, over 5727480.35 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3572, pruned_loss=0.1006, over 5780949.33 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3273, pruned_loss=0.09304, over 5720488.58 frames. ], batch size: 284, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:31:15,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9129, 4.7206, 4.4313, 2.1695], device='cuda:1'), covar=tensor([0.0367, 0.0488, 0.0506, 0.1990], device='cuda:1'), in_proj_covar=tensor([0.0950, 0.0888, 0.0785, 0.0631], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 18:31:16,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:47,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:31:52,343 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:31:58,039 INFO [train.py:968] (1/2) Epoch 9, batch 20150, libri_loss[loss=0.2833, simple_loss=0.3643, pruned_loss=0.1011, over 29498.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.338, pruned_loss=0.09782, over 5725644.13 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.357, pruned_loss=0.1005, over 5782604.99 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3328, pruned_loss=0.09644, over 5716527.67 frames. ], batch size: 81, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:32:46,761 INFO [train.py:968] (1/2) Epoch 9, batch 20200, giga_loss[loss=0.3154, simple_loss=0.378, pruned_loss=0.1264, over 28726.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3473, pruned_loss=0.1047, over 5705409.33 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3562, pruned_loss=0.1, over 5786216.92 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3434, pruned_loss=0.104, over 5692897.43 frames. ], batch size: 284, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:33:09,272 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:33:28,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:33:31,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:33:32,048 INFO [train.py:968] (1/2) Epoch 9, batch 20250, giga_loss[loss=0.2646, simple_loss=0.3498, pruned_loss=0.08967, over 28955.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3528, pruned_loss=0.1076, over 5701938.98 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3566, pruned_loss=0.1002, over 5788103.97 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3493, pruned_loss=0.107, over 5689219.49 frames. ], batch size: 145, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:33:47,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:00,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:02,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:05,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:07,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3907, 1.5576, 1.2596, 1.1823], device='cuda:1'), covar=tensor([0.1550, 0.1452, 0.1208, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1468, 0.1431, 0.1556], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:34:07,209 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:09,489 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 20300, giga_loss[loss=0.3074, simple_loss=0.3886, pruned_loss=0.1131, over 28710.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3565, pruned_loss=0.1089, over 5685192.41 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3564, pruned_loss=0.1002, over 5788620.45 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3539, pruned_loss=0.1085, over 5674470.69 frames. ], batch size: 262, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:34:30,289 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:33,732 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:34:58,647 INFO [optim.py:369] (1/2) 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 18:35:04,695 INFO [train.py:968] (1/2) Epoch 9, batch 20350, giga_loss[loss=0.3021, simple_loss=0.374, pruned_loss=0.1151, over 28620.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3612, pruned_loss=0.1115, over 5676983.34 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3564, pruned_loss=0.1001, over 5786271.13 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3591, pruned_loss=0.1114, over 5668801.67 frames. ], batch size: 262, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:35:19,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384564.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:35:21,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384567.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:35:43,003 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384596.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:35:44,474 INFO [train.py:968] (1/2) Epoch 9, batch 20400, giga_loss[loss=0.3403, simple_loss=0.3775, pruned_loss=0.1515, over 23455.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3652, pruned_loss=0.1142, over 5680090.30 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3559, pruned_loss=0.09994, over 5791146.60 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3643, pruned_loss=0.1148, over 5665337.90 frames. ], batch size: 705, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:35:50,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:35:52,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:20,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:24,763 INFO [optim.py:369] (1/2) 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 18:36:28,946 INFO [train.py:968] (1/2) Epoch 9, batch 20450, libri_loss[loss=0.2673, simple_loss=0.3431, pruned_loss=0.09577, over 29588.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3592, pruned_loss=0.1095, over 5686250.05 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3561, pruned_loss=0.1001, over 5790321.80 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3585, pruned_loss=0.1101, over 5673212.40 frames. ], batch size: 74, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:36:32,176 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:39,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=384661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:36:47,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4118, 1.6399, 1.3240, 1.4568], device='cuda:1'), covar=tensor([0.2556, 0.2456, 0.2666, 0.2310], device='cuda:1'), in_proj_covar=tensor([0.1232, 0.0920, 0.1092, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:37:12,543 INFO [train.py:968] (1/2) Epoch 9, batch 20500, giga_loss[loss=0.27, simple_loss=0.3453, pruned_loss=0.09732, over 28951.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3559, pruned_loss=0.1066, over 5697081.86 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3559, pruned_loss=0.09998, over 5790932.19 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3555, pruned_loss=0.1071, over 5685955.12 frames. ], batch size: 106, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:37:35,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:37:48,567 INFO [optim.py:369] (1/2) 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 18:37:54,698 INFO [train.py:968] (1/2) Epoch 9, batch 20550, giga_loss[loss=0.2764, simple_loss=0.3493, pruned_loss=0.1018, over 28910.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3559, pruned_loss=0.1063, over 5696874.15 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3557, pruned_loss=0.0999, over 5793137.68 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3558, pruned_loss=0.107, over 5684436.54 frames. ], batch size: 227, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:38:03,496 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=384760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:31,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:34,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:34,682 INFO [train.py:968] (1/2) Epoch 9, batch 20600, giga_loss[loss=0.3201, simple_loss=0.3839, pruned_loss=0.1281, over 28341.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3581, pruned_loss=0.1072, over 5685160.56 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3559, pruned_loss=0.1, over 5776702.36 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3578, pruned_loss=0.1077, over 5687424.81 frames. ], batch size: 368, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:38:39,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=384804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:40,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3172, 3.0462, 1.3712, 1.4678], device='cuda:1'), covar=tensor([0.0923, 0.0257, 0.0768, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0487, 0.0322, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 18:38:42,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=384807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:38:59,864 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:39:00,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7123, 1.9343, 2.0279, 1.5588], device='cuda:1'), covar=tensor([0.1532, 0.1898, 0.1137, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0696, 0.0836, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 18:39:07,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=384836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:39:12,667 INFO [optim.py:369] (1/2) 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 18:39:18,230 INFO [train.py:968] (1/2) Epoch 9, batch 20650, giga_loss[loss=0.2974, simple_loss=0.3759, pruned_loss=0.1095, over 28883.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3607, pruned_loss=0.1093, over 5674598.01 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3563, pruned_loss=0.1003, over 5759299.01 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3602, pruned_loss=0.1096, over 5689468.25 frames. ], batch size: 199, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:39:31,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-04 18:40:01,630 INFO [train.py:968] (1/2) Epoch 9, batch 20700, giga_loss[loss=0.3049, simple_loss=0.36, pruned_loss=0.1248, over 23808.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3627, pruned_loss=0.1109, over 5680956.57 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3566, pruned_loss=0.1004, over 5762537.63 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3622, pruned_loss=0.1113, over 5688695.65 frames. ], batch size: 705, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:40:29,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2012, 4.0152, 3.7806, 1.6048], device='cuda:1'), covar=tensor([0.0514, 0.0656, 0.0647, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.0954, 0.0899, 0.0792, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 18:40:47,073 INFO [optim.py:369] (1/2) 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 18:40:48,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3102, 3.3838, 1.4841, 1.4273], device='cuda:1'), covar=tensor([0.0929, 0.0254, 0.0792, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0487, 0.0324, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 18:40:50,586 INFO [train.py:968] (1/2) Epoch 9, batch 20750, giga_loss[loss=0.2437, simple_loss=0.3264, pruned_loss=0.08046, over 28436.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3653, pruned_loss=0.1138, over 5673403.93 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3569, pruned_loss=0.1006, over 5762849.94 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3646, pruned_loss=0.114, over 5678838.63 frames. ], batch size: 60, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:41:33,353 INFO [train.py:968] (1/2) Epoch 9, batch 20800, giga_loss[loss=0.3104, simple_loss=0.3721, pruned_loss=0.1244, over 28308.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3653, pruned_loss=0.1143, over 5681641.92 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3572, pruned_loss=0.1007, over 5763276.20 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3646, pruned_loss=0.1145, over 5684413.59 frames. ], batch size: 77, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:41:52,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:42:08,700 INFO [optim.py:369] (1/2) 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 18:42:13,099 INFO [train.py:968] (1/2) Epoch 9, batch 20850, giga_loss[loss=0.2719, simple_loss=0.3492, pruned_loss=0.09728, over 29103.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3645, pruned_loss=0.1126, over 5693795.09 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3575, pruned_loss=0.1008, over 5762897.54 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3639, pruned_loss=0.1129, over 5694945.67 frames. ], batch size: 113, lr: 3.61e-03, grad_scale: 8.0
+2023-03-04 18:42:26,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5062, 1.6154, 1.2780, 1.9041], device='cuda:1'), covar=tensor([0.2573, 0.2496, 0.2755, 0.2242], device='cuda:1'), in_proj_covar=tensor([0.1237, 0.0928, 0.1098, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:42:49,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-04 18:42:52,703 INFO [train.py:968] (1/2) Epoch 9, batch 20900, giga_loss[loss=0.2842, simple_loss=0.3542, pruned_loss=0.1071, over 28490.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3633, pruned_loss=0.111, over 5690687.74 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3573, pruned_loss=0.1008, over 5764201.80 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3631, pruned_loss=0.1114, over 5689838.67 frames. ], batch size: 78, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:42:56,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:43:21,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3881, 1.7989, 1.4609, 1.6065], device='cuda:1'), covar=tensor([0.0761, 0.0280, 0.0306, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:1')
+2023-03-04 18:43:22,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:43:28,527 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 20950, giga_loss[loss=0.2715, simple_loss=0.3455, pruned_loss=0.09872, over 28792.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3637, pruned_loss=0.1098, over 5697255.80 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3579, pruned_loss=0.101, over 5765696.26 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.363, pruned_loss=0.11, over 5694397.56 frames. ], batch size: 99, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:44:10,101 INFO [train.py:968] (1/2) Epoch 9, batch 21000, giga_loss[loss=0.2648, simple_loss=0.3411, pruned_loss=0.09427, over 28956.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3613, pruned_loss=0.1084, over 5703807.01 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3582, pruned_loss=0.1013, over 5769089.25 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3606, pruned_loss=0.1085, over 5696966.20 frames. ], batch size: 155, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:44:10,102 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 18:44:18,485 INFO [train.py:1012] (1/2) Epoch 9, validation: loss=0.2261, simple_loss=0.331, pruned_loss=0.0606, over 944034.00 frames. 
+2023-03-04 18:44:18,485 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 18:44:52,431 INFO [optim.py:369] (1/2) 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 18:44:52,681 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:44:54,949 INFO [train.py:968] (1/2) Epoch 9, batch 21050, giga_loss[loss=0.2804, simple_loss=0.3323, pruned_loss=0.1142, over 23414.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3582, pruned_loss=0.1064, over 5710704.37 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3586, pruned_loss=0.1016, over 5771027.07 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3574, pruned_loss=0.1064, over 5701907.23 frames. ], batch size: 705, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:44:55,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=385248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:07,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-04 18:45:10,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:16,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=385277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:17,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:45:20,642 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 21100, giga_loss[loss=0.2408, simple_loss=0.3181, pruned_loss=0.08175, over 28427.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3566, pruned_loss=0.1054, over 5718174.45 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3584, pruned_loss=0.1015, over 5774746.85 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.356, pruned_loss=0.1057, over 5704989.30 frames. ], batch size: 65, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:45:41,774 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=385310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:46:10,395 INFO [optim.py:369] (1/2) 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 18:46:12,492 INFO [train.py:968] (1/2) Epoch 9, batch 21150, giga_loss[loss=0.3923, simple_loss=0.419, pruned_loss=0.1828, over 26547.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3561, pruned_loss=0.1055, over 5718314.31 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3586, pruned_loss=0.1016, over 5775224.98 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3555, pruned_loss=0.1057, over 5705672.94 frames. ], batch size: 555, lr: 3.61e-03, grad_scale: 2.0
+2023-03-04 18:46:31,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 18:46:50,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5427, 2.2293, 1.6429, 0.6143], device='cuda:1'), covar=tensor([0.3245, 0.1516, 0.2481, 0.3863], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1402, 0.1451, 0.1209], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 18:46:52,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:46:53,681 INFO [train.py:968] (1/2) Epoch 9, batch 21200, libri_loss[loss=0.2359, simple_loss=0.3144, pruned_loss=0.0787, over 29500.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.358, pruned_loss=0.1071, over 5719401.82 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3583, pruned_loss=0.1014, over 5777675.90 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3577, pruned_loss=0.1076, over 5706026.81 frames. ], batch size: 70, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:47:23,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6156, 1.6818, 1.4295, 1.3901], device='cuda:1'), covar=tensor([0.1658, 0.1428, 0.1286, 0.1501], device='cuda:1'), in_proj_covar=tensor([0.1616, 0.1497, 0.1462, 0.1569], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:47:30,687 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 21250, giga_loss[loss=0.3224, simple_loss=0.3794, pruned_loss=0.1328, over 26648.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5721193.34 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3581, pruned_loss=0.1015, over 5780613.10 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3569, pruned_loss=0.1064, over 5705934.03 frames. ], batch size: 555, lr: 3.61e-03, grad_scale: 4.0
+2023-03-04 18:48:12,227 INFO [train.py:968] (1/2) Epoch 9, batch 21300, giga_loss[loss=0.3268, simple_loss=0.3936, pruned_loss=0.13, over 28557.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3564, pruned_loss=0.1048, over 5720383.47 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3582, pruned_loss=0.1015, over 5782862.41 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3563, pruned_loss=0.1053, over 5704422.80 frames. ], batch size: 336, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:48:17,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 18:48:18,720 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-04 18:48:47,244 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:48:49,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=385542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:48:51,426 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 21350, giga_loss[loss=0.2741, simple_loss=0.3463, pruned_loss=0.101, over 28685.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3561, pruned_loss=0.1045, over 5728526.43 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3582, pruned_loss=0.1016, over 5784962.56 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3559, pruned_loss=0.1048, over 5713283.82 frames. ], batch size: 92, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:49:01,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-04 18:49:10,627 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 21400, giga_loss[loss=0.2625, simple_loss=0.3311, pruned_loss=0.097, over 28870.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3548, pruned_loss=0.1042, over 5735070.46 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3587, pruned_loss=0.1021, over 5786946.73 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3542, pruned_loss=0.104, over 5719742.34 frames. ], batch size: 112, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:49:35,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2470, 2.5379, 1.2403, 1.3728], device='cuda:1'), covar=tensor([0.0915, 0.0298, 0.0856, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0486, 0.0324, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0028, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 18:50:04,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=385642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:50:05,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:50:06,879 INFO [optim.py:369] (1/2) 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 18:50:09,360 INFO [train.py:968] (1/2) Epoch 9, batch 21450, giga_loss[loss=0.2466, simple_loss=0.3077, pruned_loss=0.09275, over 23416.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3518, pruned_loss=0.1029, over 5721178.68 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3588, pruned_loss=0.1023, over 5783971.94 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1026, over 5711136.60 frames. ], batch size: 705, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:50:47,170 INFO [train.py:968] (1/2) Epoch 9, batch 21500, giga_loss[loss=0.2489, simple_loss=0.3302, pruned_loss=0.08379, over 28929.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3496, pruned_loss=0.1017, over 5726999.62 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3586, pruned_loss=0.1024, over 5785608.96 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3489, pruned_loss=0.1013, over 5715748.16 frames. ], batch size: 227, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:51:12,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385730.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:51:25,054 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 21550, giga_loss[loss=0.2316, simple_loss=0.3058, pruned_loss=0.07873, over 28499.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3491, pruned_loss=0.1019, over 5722038.71 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3588, pruned_loss=0.1026, over 5776615.91 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3484, pruned_loss=0.1015, over 5720255.77 frames. ], batch size: 85, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:51:55,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=385785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:51:57,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=385788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:52:07,153 INFO [train.py:968] (1/2) Epoch 9, batch 21600, giga_loss[loss=0.2675, simple_loss=0.3433, pruned_loss=0.0959, over 28842.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3488, pruned_loss=0.1026, over 5718629.61 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3587, pruned_loss=0.1024, over 5778348.25 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3482, pruned_loss=0.1024, over 5714946.88 frames. ], batch size: 119, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:52:22,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=385817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:52:25,567 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=385822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:52:39,373 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:52:46,152 INFO [train.py:968] (1/2) Epoch 9, batch 21650, giga_loss[loss=0.2636, simple_loss=0.3416, pruned_loss=0.09285, over 28720.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3478, pruned_loss=0.1027, over 5717259.27 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3591, pruned_loss=0.1026, over 5780912.15 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3468, pruned_loss=0.1024, over 5710685.11 frames. ], batch size: 284, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:53:21,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5504, 1.6870, 1.5177, 1.4678], device='cuda:1'), covar=tensor([0.1826, 0.1457, 0.1196, 0.1374], device='cuda:1'), in_proj_covar=tensor([0.1611, 0.1486, 0.1455, 0.1558], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:53:23,536 INFO [train.py:968] (1/2) Epoch 9, batch 21700, libri_loss[loss=0.2805, simple_loss=0.3614, pruned_loss=0.0998, over 29631.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3455, pruned_loss=0.1018, over 5721181.54 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3592, pruned_loss=0.1028, over 5781637.72 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3441, pruned_loss=0.1013, over 5713040.48 frames. ], batch size: 91, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:54:01,965 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 21750, giga_loss[loss=0.2957, simple_loss=0.361, pruned_loss=0.1152, over 27598.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3439, pruned_loss=0.1018, over 5716928.30 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3592, pruned_loss=0.103, over 5784384.34 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3425, pruned_loss=0.1012, over 5706918.34 frames. ], batch size: 472, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:54:41,771 INFO [train.py:968] (1/2) Epoch 9, batch 21800, giga_loss[loss=0.2333, simple_loss=0.3124, pruned_loss=0.07712, over 28373.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3431, pruned_loss=0.1015, over 5715418.86 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3593, pruned_loss=0.1031, over 5787642.86 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3417, pruned_loss=0.1009, over 5702940.96 frames. ], batch size: 65, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:54:58,284 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 9, batch 21850, libri_loss[loss=0.3409, simple_loss=0.401, pruned_loss=0.1405, over 29530.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3452, pruned_loss=0.1025, over 5718730.98 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3599, pruned_loss=0.1037, over 5790121.09 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.343, pruned_loss=0.1014, over 5704490.65 frames. ], batch size: 84, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:56:02,390 INFO [train.py:968] (1/2) Epoch 9, batch 21900, giga_loss[loss=0.2602, simple_loss=0.3451, pruned_loss=0.08762, over 28944.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3482, pruned_loss=0.1035, over 5719990.31 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3605, pruned_loss=0.1043, over 5791881.74 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3457, pruned_loss=0.1022, over 5705792.51 frames. ], batch size: 213, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:56:08,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8916, 2.4910, 1.8581, 1.6963], device='cuda:1'), covar=tensor([0.2211, 0.1302, 0.1499, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1494, 0.1459, 0.1561], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:56:09,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=386105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:56:21,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3641, 1.4924, 1.5117, 1.3992], device='cuda:1'), covar=tensor([0.1159, 0.1499, 0.1584, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0717, 0.0649, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 18:56:38,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7657, 1.7897, 1.8221, 1.6789], device='cuda:1'), covar=tensor([0.1364, 0.1953, 0.1751, 0.1739], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0717, 0.0650, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 18:56:43,169 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 21950, giga_loss[loss=0.2901, simple_loss=0.3648, pruned_loss=0.1077, over 28927.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3517, pruned_loss=0.105, over 5717506.67 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3617, pruned_loss=0.1052, over 5792272.12 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3482, pruned_loss=0.1031, over 5703496.51 frames. ], batch size: 174, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 18:56:55,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:56:57,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:57:19,580 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:57:20,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8747, 2.6046, 1.9516, 1.7324], device='cuda:1'), covar=tensor([0.2174, 0.1246, 0.1422, 0.1605], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1488, 0.1451, 0.1552], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 18:57:22,690 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 22000, giga_loss[loss=0.3123, simple_loss=0.3846, pruned_loss=0.12, over 27485.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.353, pruned_loss=0.1049, over 5711053.30 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3625, pruned_loss=0.106, over 5791877.12 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3492, pruned_loss=0.1027, over 5697536.78 frames. ], batch size: 472, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:57:26,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-04 18:57:36,952 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:58:05,665 INFO [train.py:968] (1/2) Epoch 9, batch 22050, libri_loss[loss=0.2453, simple_loss=0.315, pruned_loss=0.08779, over 29645.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3521, pruned_loss=0.1044, over 5706736.77 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3622, pruned_loss=0.1061, over 5793818.12 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3491, pruned_loss=0.1024, over 5692335.16 frames. ], batch size: 73, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:58:06,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386248.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:58:08,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386251.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 18:58:15,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5451, 1.3206, 4.9435, 3.4667], device='cuda:1'), covar=tensor([0.1596, 0.2544, 0.0342, 0.0709], device='cuda:1'), in_proj_covar=tensor([0.0615, 0.0559, 0.0807, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 18:58:33,085 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386280.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 18:58:46,704 INFO [train.py:968] (1/2) Epoch 9, batch 22100, giga_loss[loss=0.336, simple_loss=0.4008, pruned_loss=0.1356, over 28258.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3522, pruned_loss=0.1044, over 5707055.02 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3627, pruned_loss=0.1065, over 5791071.46 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3493, pruned_loss=0.1025, over 5697361.64 frames. ], batch size: 368, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:59:18,839 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:21,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:24,117 INFO [optim.py:369] (1/2) 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 18:59:25,325 INFO [train.py:968] (1/2) Epoch 9, batch 22150, giga_loss[loss=0.2841, simple_loss=0.3481, pruned_loss=0.1101, over 28726.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3534, pruned_loss=0.1055, over 5705346.88 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3634, pruned_loss=0.1071, over 5791150.82 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3502, pruned_loss=0.1034, over 5695256.57 frames. ], batch size: 99, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 18:59:33,235 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=386358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:35,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=386361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:41,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8584, 3.1944, 2.1618, 0.9087], device='cuda:1'), covar=tensor([0.4380, 0.1857, 0.2230, 0.3742], device='cuda:1'), in_proj_covar=tensor([0.1503, 0.1416, 0.1471, 0.1220], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 18:59:43,946 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 18:59:44,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1361, 1.3865, 1.1716, 1.4230], device='cuda:1'), covar=tensor([0.0809, 0.0324, 0.0321, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:1')
+2023-03-04 18:59:59,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=386390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:00:04,266 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-04 19:00:05,224 INFO [train.py:968] (1/2) Epoch 9, batch 22200, giga_loss[loss=0.2957, simple_loss=0.3686, pruned_loss=0.1114, over 28289.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3547, pruned_loss=0.1062, over 5714621.43 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3634, pruned_loss=0.1072, over 5793722.50 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3519, pruned_loss=0.1043, over 5702707.80 frames. ], batch size: 368, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:00:41,014 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 22250, giga_loss[loss=0.2776, simple_loss=0.3543, pruned_loss=0.1005, over 28703.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3576, pruned_loss=0.1078, over 5705595.99 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3638, pruned_loss=0.1075, over 5791007.76 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3547, pruned_loss=0.106, over 5695587.57 frames. ], batch size: 60, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:01:00,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-04 19:01:19,562 INFO [train.py:968] (1/2) Epoch 9, batch 22300, giga_loss[loss=0.2669, simple_loss=0.3497, pruned_loss=0.09207, over 28872.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.36, pruned_loss=0.1091, over 5704142.24 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3636, pruned_loss=0.1077, over 5781897.75 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3578, pruned_loss=0.1076, over 5702731.02 frames. ], batch size: 145, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:01:44,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2209, 1.9093, 1.8106, 1.7573], device='cuda:1'), covar=tensor([0.1236, 0.2326, 0.1888, 0.1956], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0726, 0.0660, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:01:57,822 INFO [optim.py:369] (1/2) 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 19:01:59,254 INFO [train.py:968] (1/2) Epoch 9, batch 22350, giga_loss[loss=0.266, simple_loss=0.3381, pruned_loss=0.09699, over 28374.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3601, pruned_loss=0.1088, over 5700703.70 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3638, pruned_loss=0.1078, over 5773861.26 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3581, pruned_loss=0.1075, over 5706701.57 frames. ], batch size: 60, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:02:18,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-04 19:02:40,602 INFO [train.py:968] (1/2) Epoch 9, batch 22400, giga_loss[loss=0.2493, simple_loss=0.3197, pruned_loss=0.08947, over 28575.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3611, pruned_loss=0.1093, over 5708469.74 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3648, pruned_loss=0.1086, over 5776721.54 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3586, pruned_loss=0.1076, over 5709389.49 frames. ], batch size: 78, lr: 3.60e-03, grad_scale: 8.0
+2023-03-04 19:02:56,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3763, 2.1002, 1.8537, 1.8714], device='cuda:1'), covar=tensor([0.1169, 0.2252, 0.1953, 0.1963], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0725, 0.0662, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:03:19,010 INFO [optim.py:369] (1/2) 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 19:03:19,022 INFO [train.py:968] (1/2) Epoch 9, batch 22450, giga_loss[loss=0.2821, simple_loss=0.355, pruned_loss=0.1046, over 27914.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3622, pruned_loss=0.1107, over 5710111.19 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3651, pruned_loss=0.109, over 5779458.01 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.36, pruned_loss=0.109, over 5706856.71 frames. ], batch size: 412, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:03:30,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:03:51,774 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386685.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 19:04:00,729 INFO [train.py:968] (1/2) Epoch 9, batch 22500, giga_loss[loss=0.3182, simple_loss=0.3793, pruned_loss=0.1286, over 28848.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3595, pruned_loss=0.109, over 5716685.80 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3655, pruned_loss=0.1093, over 5782055.69 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3572, pruned_loss=0.1074, over 5710431.18 frames. ], batch size: 199, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:04:41,738 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 22550, giga_loss[loss=0.2974, simple_loss=0.3658, pruned_loss=0.1145, over 28220.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3561, pruned_loss=0.1071, over 5711986.65 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3658, pruned_loss=0.1097, over 5785367.82 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3538, pruned_loss=0.1054, over 5702033.05 frames. ], batch size: 368, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:04:42,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-04 19:05:07,635 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:05:21,190 INFO [train.py:968] (1/2) Epoch 9, batch 22600, giga_loss[loss=0.2782, simple_loss=0.3612, pruned_loss=0.09765, over 28607.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3521, pruned_loss=0.1048, over 5715026.79 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3658, pruned_loss=0.1099, over 5787116.73 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3501, pruned_loss=0.1033, over 5704831.98 frames. ], batch size: 336, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:05:29,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 19:05:33,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4618, 1.4939, 1.0839, 1.1841], device='cuda:1'), covar=tensor([0.0719, 0.0641, 0.1076, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0343, 0.0439, 0.0495, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 19:05:37,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0519, 1.5102, 1.3620, 1.0388], device='cuda:1'), covar=tensor([0.1204, 0.1608, 0.1047, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0695, 0.0834, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 19:05:50,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2849, 1.5797, 1.3076, 1.4806], device='cuda:1'), covar=tensor([0.0731, 0.0306, 0.0332, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0082], device='cuda:1')
+2023-03-04 19:05:55,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0839, 1.2361, 0.9401, 1.0359], device='cuda:1'), covar=tensor([0.1169, 0.0977, 0.0792, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.1606, 0.1491, 0.1453, 0.1558], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 19:05:59,226 INFO [optim.py:369] (1/2) 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 19:05:59,238 INFO [train.py:968] (1/2) Epoch 9, batch 22650, giga_loss[loss=0.2641, simple_loss=0.3506, pruned_loss=0.08883, over 28840.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3526, pruned_loss=0.1043, over 5712978.03 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3659, pruned_loss=0.1102, over 5790227.69 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3506, pruned_loss=0.1027, over 5700533.35 frames. ], batch size: 174, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:06:00,001 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:06:34,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3987, 1.6642, 1.5079, 1.2312], device='cuda:1'), covar=tensor([0.2272, 0.1498, 0.1164, 0.1694], device='cuda:1'), in_proj_covar=tensor([0.1603, 0.1488, 0.1451, 0.1557], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 19:06:42,437 INFO [train.py:968] (1/2) Epoch 9, batch 22700, giga_loss[loss=0.271, simple_loss=0.3395, pruned_loss=0.1012, over 24067.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3547, pruned_loss=0.1043, over 5709367.46 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3659, pruned_loss=0.1103, over 5791545.97 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3531, pruned_loss=0.1029, over 5697823.67 frames. ], batch size: 705, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:07:20,400 INFO [optim.py:369] (1/2) 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 19:07:20,412 INFO [train.py:968] (1/2) Epoch 9, batch 22750, giga_loss[loss=0.287, simple_loss=0.3599, pruned_loss=0.1071, over 27866.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3535, pruned_loss=0.1041, over 5709251.06 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3659, pruned_loss=0.1107, over 5794537.10 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3519, pruned_loss=0.1025, over 5695325.01 frames. ], batch size: 412, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:07:35,985 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=386965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:07:53,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3257, 1.5979, 1.3023, 1.0577], device='cuda:1'), covar=tensor([0.2063, 0.2057, 0.2331, 0.1910], device='cuda:1'), in_proj_covar=tensor([0.1237, 0.0921, 0.1089, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:08:03,246 INFO [train.py:968] (1/2) Epoch 9, batch 22800, giga_loss[loss=0.2616, simple_loss=0.3207, pruned_loss=0.1013, over 28608.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3525, pruned_loss=0.1053, over 5710216.45 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3658, pruned_loss=0.1109, over 5795594.77 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.351, pruned_loss=0.1037, over 5696697.31 frames. ], batch size: 85, lr: 3.60e-03, grad_scale: 8.0
+2023-03-04 19:08:09,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 19:08:21,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5800, 0.9488, 2.7944, 2.7684], device='cuda:1'), covar=tensor([0.2200, 0.2745, 0.0970, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0615, 0.0561, 0.0810, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:08:33,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:08:42,073 INFO [optim.py:369] (1/2) 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 19:08:42,085 INFO [train.py:968] (1/2) Epoch 9, batch 22850, giga_loss[loss=0.2817, simple_loss=0.3546, pruned_loss=0.1044, over 28977.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.352, pruned_loss=0.1064, over 5712817.96 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3659, pruned_loss=0.1111, over 5794257.49 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3505, pruned_loss=0.1049, over 5702041.54 frames. ], batch size: 164, lr: 3.60e-03, grad_scale: 8.0
+2023-03-04 19:08:52,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387060.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 19:09:23,018 INFO [train.py:968] (1/2) Epoch 9, batch 22900, giga_loss[loss=0.3009, simple_loss=0.3626, pruned_loss=0.1196, over 28664.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3503, pruned_loss=0.1064, over 5700043.31 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3665, pruned_loss=0.1116, over 5777827.11 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3482, pruned_loss=0.1046, over 5704533.28 frames. ], batch size: 263, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:09:29,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5930, 1.7753, 1.4222, 1.4008], device='cuda:1'), covar=tensor([0.1883, 0.1460, 0.1316, 0.1488], device='cuda:1'), in_proj_covar=tensor([0.1606, 0.1493, 0.1452, 0.1563], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 19:09:38,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4396, 1.2347, 4.9669, 3.4341], device='cuda:1'), covar=tensor([0.1699, 0.2637, 0.0341, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0615, 0.0564, 0.0811, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:10:01,943 INFO [train.py:968] (1/2) Epoch 9, batch 22950, giga_loss[loss=0.2719, simple_loss=0.3424, pruned_loss=0.1007, over 28969.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3496, pruned_loss=0.106, over 5702452.63 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3667, pruned_loss=0.1118, over 5774970.47 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3477, pruned_loss=0.1044, over 5707678.74 frames. ], batch size: 136, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:10:02,647 INFO [optim.py:369] (1/2) 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 19:10:08,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:25,799 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:27,876 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:39,201 INFO [train.py:968] (1/2) Epoch 9, batch 23000, giga_loss[loss=0.228, simple_loss=0.2956, pruned_loss=0.08018, over 28497.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3494, pruned_loss=0.1068, over 5705066.83 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3679, pruned_loss=0.1131, over 5768842.65 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3461, pruned_loss=0.1041, over 5711510.13 frames. ], batch size: 71, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:10:42,656 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387203.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 19:10:45,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387206.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 19:10:49,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:10:59,466 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:11:06,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387235.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 19:11:15,688 INFO [train.py:968] (1/2) Epoch 9, batch 23050, giga_loss[loss=0.3273, simple_loss=0.3756, pruned_loss=0.1395, over 26753.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3453, pruned_loss=0.1047, over 5711274.52 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3679, pruned_loss=0.1135, over 5774660.63 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3417, pruned_loss=0.1018, over 5708648.47 frames. ], batch size: 555, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:11:17,149 INFO [optim.py:369] (1/2) 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 19:11:28,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.29 vs. limit=5.0
+2023-03-04 19:11:55,451 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:11:58,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8091, 1.9156, 1.6932, 1.7295], device='cuda:1'), covar=tensor([0.1448, 0.2308, 0.2014, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0727, 0.0660, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:11:58,532 INFO [train.py:968] (1/2) Epoch 9, batch 23100, giga_loss[loss=0.2371, simple_loss=0.3165, pruned_loss=0.07879, over 28710.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3408, pruned_loss=0.1019, over 5701847.05 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3684, pruned_loss=0.1139, over 5766492.31 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3372, pruned_loss=0.09903, over 5706721.71 frames. ], batch size: 262, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:11:59,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:01,912 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:05,454 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-04 19:12:05,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7079, 1.7358, 1.6492, 1.6258], device='cuda:1'), covar=tensor([0.1329, 0.1980, 0.1863, 0.1866], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0725, 0.0659, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:12:24,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:30,858 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 23150, giga_loss[loss=0.299, simple_loss=0.3688, pruned_loss=0.1146, over 28786.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.341, pruned_loss=0.1013, over 5709143.47 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3685, pruned_loss=0.1142, over 5766554.10 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3374, pruned_loss=0.09846, over 5711413.80 frames. ], batch size: 119, lr: 3.60e-03, grad_scale: 2.0
+2023-03-04 19:12:36,645 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:37,764 INFO [optim.py:369] (1/2) 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 19:12:42,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:44,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-04 19:12:50,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:12:52,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:13:14,840 INFO [train.py:968] (1/2) Epoch 9, batch 23200, giga_loss[loss=0.2594, simple_loss=0.3322, pruned_loss=0.09331, over 29045.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.344, pruned_loss=0.1026, over 5709170.69 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3687, pruned_loss=0.1146, over 5768967.63 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3399, pruned_loss=0.09948, over 5706829.59 frames. ], batch size: 106, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:13:15,435 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 19:13:15,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:13:56,768 INFO [train.py:968] (1/2) Epoch 9, batch 23250, giga_loss[loss=0.2646, simple_loss=0.3391, pruned_loss=0.09502, over 28455.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3483, pruned_loss=0.1044, over 5709648.86 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3692, pruned_loss=0.1151, over 5770300.15 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3441, pruned_loss=0.1012, over 5705083.71 frames. ], batch size: 71, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:13:58,160 INFO [optim.py:369] (1/2) 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 19:14:24,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:27,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:36,874 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:38,503 INFO [train.py:968] (1/2) Epoch 9, batch 23300, libri_loss[loss=0.2741, simple_loss=0.3511, pruned_loss=0.09849, over 29761.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3523, pruned_loss=0.106, over 5708507.17 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3691, pruned_loss=0.115, over 5772016.09 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3489, pruned_loss=0.1035, over 5702708.21 frames. ], batch size: 87, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:14:44,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=387506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:14:50,928 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 23350, giga_loss[loss=0.3576, simple_loss=0.4038, pruned_loss=0.1557, over 26637.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3558, pruned_loss=0.1079, over 5702478.00 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3695, pruned_loss=0.1155, over 5771662.54 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3524, pruned_loss=0.1054, over 5696306.61 frames. ], batch size: 555, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:15:19,859 INFO [optim.py:369] (1/2) 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 19:15:58,044 INFO [train.py:968] (1/2) Epoch 9, batch 23400, giga_loss[loss=0.577, simple_loss=0.5358, pruned_loss=0.3092, over 26692.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3584, pruned_loss=0.1099, over 5702492.31 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3694, pruned_loss=0.1156, over 5771710.45 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3556, pruned_loss=0.1076, over 5695889.27 frames. ], batch size: 555, lr: 3.60e-03, grad_scale: 4.0
+2023-03-04 19:16:41,992 INFO [train.py:968] (1/2) Epoch 9, batch 23450, libri_loss[loss=0.3151, simple_loss=0.3813, pruned_loss=0.1245, over 29107.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3655, pruned_loss=0.1164, over 5693038.87 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3701, pruned_loss=0.1163, over 5766448.86 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3623, pruned_loss=0.1138, over 5690210.04 frames. ], batch size: 101, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:16:44,328 INFO [optim.py:369] (1/2) 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 19:17:00,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:17:18,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5035, 4.2849, 4.0590, 1.9351], device='cuda:1'), covar=tensor([0.0513, 0.0721, 0.0729, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.0989, 0.0928, 0.0818, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 19:17:29,132 INFO [train.py:968] (1/2) Epoch 9, batch 23500, giga_loss[loss=0.3607, simple_loss=0.4104, pruned_loss=0.1555, over 29086.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3708, pruned_loss=0.1207, over 5692695.36 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.37, pruned_loss=0.1164, over 5770030.43 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3683, pruned_loss=0.1186, over 5685354.22 frames. ], batch size: 136, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:17:35,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3050, 2.8417, 1.4209, 1.4083], device='cuda:1'), covar=tensor([0.0872, 0.0370, 0.0817, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0502, 0.0329, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0019, 0.0024], device='cuda:1')
+2023-03-04 19:17:51,585 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:18:02,836 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 23550, giga_loss[loss=0.3535, simple_loss=0.4077, pruned_loss=0.1497, over 28531.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.379, pruned_loss=0.1269, over 5682967.73 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3704, pruned_loss=0.1168, over 5761402.37 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3767, pruned_loss=0.1251, over 5681533.77 frames. ], batch size: 71, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:18:17,793 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 23600, giga_loss[loss=0.4329, simple_loss=0.4366, pruned_loss=0.2146, over 23616.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3844, pruned_loss=0.1323, over 5674158.05 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3704, pruned_loss=0.1168, over 5762269.40 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3828, pruned_loss=0.1311, over 5670785.46 frames. ], batch size: 705, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:19:18,091 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:19:21,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:19:44,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 19:19:49,119 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:19:53,439 INFO [train.py:968] (1/2) Epoch 9, batch 23650, giga_loss[loss=0.3188, simple_loss=0.3873, pruned_loss=0.1252, over 28811.00 frames. ], tot_loss[loss=0.334, simple_loss=0.3912, pruned_loss=0.1384, over 5658195.15 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3705, pruned_loss=0.1171, over 5754361.08 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3902, pruned_loss=0.1375, over 5661072.38 frames. ], batch size: 112, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:19:56,081 INFO [optim.py:369] (1/2) 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 19:20:08,433 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:11,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:11,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:17,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=387875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:21,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:24,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=387881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:36,448 INFO [train.py:968] (1/2) Epoch 9, batch 23700, giga_loss[loss=0.2951, simple_loss=0.3692, pruned_loss=0.1105, over 28866.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3918, pruned_loss=0.1384, over 5669482.31 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3701, pruned_loss=0.1169, over 5758248.23 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3919, pruned_loss=0.1384, over 5665951.42 frames. ], batch size: 174, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:20:36,677 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:45,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=387907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:20:54,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4403, 1.7297, 1.7239, 1.3081], device='cuda:1'), covar=tensor([0.1474, 0.1964, 0.1182, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0692, 0.0826, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 19:21:15,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8202, 4.6245, 1.7978, 1.9729], device='cuda:1'), covar=tensor([0.0815, 0.0239, 0.0805, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0505, 0.0331, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 19:21:26,538 INFO [train.py:968] (1/2) Epoch 9, batch 23750, giga_loss[loss=0.3428, simple_loss=0.3976, pruned_loss=0.1441, over 28970.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3942, pruned_loss=0.1415, over 5662179.36 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3702, pruned_loss=0.117, over 5757284.16 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3945, pruned_loss=0.1417, over 5659135.39 frames. ], batch size: 227, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:21:29,903 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 23800, giga_loss[loss=0.3263, simple_loss=0.3884, pruned_loss=0.1321, over 28817.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3957, pruned_loss=0.1437, over 5645800.16 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3703, pruned_loss=0.1172, over 5758360.33 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3963, pruned_loss=0.1442, over 5640288.26 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:22:22,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:23,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7540, 2.7609, 1.8067, 0.7974], device='cuda:1'), covar=tensor([0.4803, 0.2210, 0.2647, 0.4548], device='cuda:1'), in_proj_covar=tensor([0.1513, 0.1444, 0.1473, 0.1234], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 19:22:35,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:39,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:48,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:22:52,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:23:11,235 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:23:14,927 INFO [train.py:968] (1/2) Epoch 9, batch 23850, giga_loss[loss=0.3672, simple_loss=0.4132, pruned_loss=0.1606, over 28694.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.4007, pruned_loss=0.1488, over 5639001.77 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3701, pruned_loss=0.1172, over 5759018.53 frames. ], giga_tot_loss[loss=0.3501, simple_loss=0.4015, pruned_loss=0.1493, over 5633565.48 frames. ], batch size: 262, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:23:21,134 INFO [optim.py:369] (1/2) 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 19:23:28,728 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:24:04,949 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:24:11,532 INFO [train.py:968] (1/2) Epoch 9, batch 23900, giga_loss[loss=0.3047, simple_loss=0.3762, pruned_loss=0.1166, over 28858.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.4025, pruned_loss=0.1513, over 5617400.17 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3704, pruned_loss=0.1174, over 5762658.66 frames. ], giga_tot_loss[loss=0.3544, simple_loss=0.4038, pruned_loss=0.1525, over 5605968.15 frames. ], batch size: 145, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:24:15,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:24:36,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0630, 3.8557, 3.6874, 1.8160], device='cuda:1'), covar=tensor([0.0581, 0.0731, 0.0794, 0.2101], device='cuda:1'), in_proj_covar=tensor([0.0990, 0.0936, 0.0828, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 19:24:59,171 INFO [train.py:968] (1/2) Epoch 9, batch 23950, giga_loss[loss=0.3946, simple_loss=0.4331, pruned_loss=0.178, over 28716.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.402, pruned_loss=0.1518, over 5614451.88 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3714, pruned_loss=0.1184, over 5748731.86 frames. ], giga_tot_loss[loss=0.3546, simple_loss=0.4033, pruned_loss=0.153, over 5612037.85 frames. ], batch size: 307, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:25:01,954 INFO [optim.py:369] (1/2) 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 19:25:38,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7151, 4.5498, 4.3393, 2.0972], device='cuda:1'), covar=tensor([0.0516, 0.0666, 0.0740, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.0995, 0.0937, 0.0829, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 19:25:44,389 INFO [train.py:968] (1/2) Epoch 9, batch 24000, giga_loss[loss=0.3986, simple_loss=0.4173, pruned_loss=0.1899, over 23671.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.3996, pruned_loss=0.1503, over 5628493.33 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3714, pruned_loss=0.1186, over 5748232.15 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4011, pruned_loss=0.1518, over 5624690.94 frames. ], batch size: 705, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:25:44,389 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 19:25:53,061 INFO [train.py:1012] (1/2) Epoch 9, validation: loss=0.22, simple_loss=0.3265, pruned_loss=0.0568, over 944034.00 frames. 
+2023-03-04 19:25:53,062 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 19:26:39,664 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-04 19:26:41,025 INFO [train.py:968] (1/2) Epoch 9, batch 24050, giga_loss[loss=0.417, simple_loss=0.4568, pruned_loss=0.1886, over 28588.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.3995, pruned_loss=0.1492, over 5627499.19 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3711, pruned_loss=0.1186, over 5750422.39 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4013, pruned_loss=0.1507, over 5621183.66 frames. ], batch size: 307, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:26:44,204 INFO [optim.py:369] (1/2) 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 19:27:28,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 19:27:34,891 INFO [train.py:968] (1/2) Epoch 9, batch 24100, giga_loss[loss=0.3401, simple_loss=0.3985, pruned_loss=0.1409, over 29021.00 frames. ], tot_loss[loss=0.349, simple_loss=0.4, pruned_loss=0.1491, over 5622362.65 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3711, pruned_loss=0.1186, over 5752469.52 frames. ], giga_tot_loss[loss=0.3522, simple_loss=0.4022, pruned_loss=0.1511, over 5612090.72 frames. ], batch size: 155, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:28:29,661 INFO [train.py:968] (1/2) Epoch 9, batch 24150, giga_loss[loss=0.3277, simple_loss=0.3887, pruned_loss=0.1334, over 28349.00 frames. ], tot_loss[loss=0.3497, simple_loss=0.401, pruned_loss=0.1492, over 5628788.17 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3713, pruned_loss=0.1189, over 5755522.74 frames. ], giga_tot_loss[loss=0.3527, simple_loss=0.4031, pruned_loss=0.1511, over 5615605.61 frames. ], batch size: 368, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:28:34,398 INFO [optim.py:369] (1/2) 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 19:28:40,262 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:28:40,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4165, 1.6133, 1.2918, 1.5883], device='cuda:1'), covar=tensor([0.2238, 0.2185, 0.2381, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.1243, 0.0930, 0.1098, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:28:44,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3792, 2.1825, 1.7346, 1.4825], device='cuda:1'), covar=tensor([0.0825, 0.0241, 0.0269, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0117, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0053, 0.0048, 0.0081], device='cuda:1')
+2023-03-04 19:28:59,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:29:23,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-04 19:29:24,874 INFO [train.py:968] (1/2) Epoch 9, batch 24200, giga_loss[loss=0.3333, simple_loss=0.3959, pruned_loss=0.1353, over 28881.00 frames. ], tot_loss[loss=0.3451, simple_loss=0.3979, pruned_loss=0.1461, over 5625207.17 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3712, pruned_loss=0.1188, over 5757093.61 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.4, pruned_loss=0.1479, over 5612589.84 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:29:51,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1343, 1.4597, 1.4617, 1.0591], device='cuda:1'), covar=tensor([0.1377, 0.2130, 0.1153, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0694, 0.0825, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 19:30:15,427 INFO [train.py:968] (1/2) Epoch 9, batch 24250, giga_loss[loss=0.3151, simple_loss=0.3617, pruned_loss=0.1343, over 23803.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3966, pruned_loss=0.1438, over 5634328.81 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3712, pruned_loss=0.1188, over 5759329.96 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.3985, pruned_loss=0.1456, over 5620577.18 frames. ], batch size: 705, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:30:19,414 INFO [optim.py:369] (1/2) 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 19:30:28,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-04 19:30:35,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:30:45,319 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 24300, giga_loss[loss=0.2853, simple_loss=0.3522, pruned_loss=0.1092, over 28678.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3918, pruned_loss=0.1397, over 5632975.04 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3719, pruned_loss=0.1196, over 5759723.87 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3932, pruned_loss=0.1409, over 5618604.27 frames. ], batch size: 92, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:31:13,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0484, 1.1387, 3.4414, 2.9711], device='cuda:1'), covar=tensor([0.1546, 0.2498, 0.0416, 0.1715], device='cuda:1'), in_proj_covar=tensor([0.0627, 0.0572, 0.0823, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 19:31:25,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:31:30,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:31:35,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 19:31:46,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1523, 4.9464, 4.7181, 2.2310], device='cuda:1'), covar=tensor([0.0387, 0.0548, 0.0587, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.0992, 0.0937, 0.0827, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 19:31:55,146 INFO [train.py:968] (1/2) Epoch 9, batch 24350, giga_loss[loss=0.3932, simple_loss=0.4219, pruned_loss=0.1823, over 26684.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3893, pruned_loss=0.1377, over 5638177.65 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3722, pruned_loss=0.1201, over 5757759.66 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3905, pruned_loss=0.1387, over 5625844.41 frames. ], batch size: 555, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:32:00,045 INFO [optim.py:369] (1/2) 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 19:32:00,238 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:32:22,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6270, 1.8008, 1.5657, 1.6873], device='cuda:1'), covar=tensor([0.1097, 0.1710, 0.1615, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0727, 0.0660, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:32:26,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5413, 1.6313, 1.6054, 1.5164], device='cuda:1'), covar=tensor([0.1290, 0.1712, 0.1827, 0.1609], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0728, 0.0661, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:32:45,688 INFO [train.py:968] (1/2) Epoch 9, batch 24400, giga_loss[loss=0.324, simple_loss=0.384, pruned_loss=0.132, over 27605.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3883, pruned_loss=0.1373, over 5636530.66 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3722, pruned_loss=0.1201, over 5758929.05 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3895, pruned_loss=0.1382, over 5624716.24 frames. ], batch size: 472, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:32:49,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=388603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:32:57,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:00,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:05,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:07,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:29,363 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:29,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3131, 1.6431, 1.3396, 1.5326], device='cuda:1'), covar=tensor([0.0746, 0.0294, 0.0310, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0118, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0054, 0.0049, 0.0082], device='cuda:1')
+2023-03-04 19:33:36,947 INFO [train.py:968] (1/2) Epoch 9, batch 24450, giga_loss[loss=0.3834, simple_loss=0.4078, pruned_loss=0.1795, over 23761.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3871, pruned_loss=0.1363, over 5644154.67 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3724, pruned_loss=0.1204, over 5761976.44 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3884, pruned_loss=0.1373, over 5627584.12 frames. ], batch size: 705, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:33:43,401 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:33:43,713 INFO [optim.py:369] (1/2) 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 19:34:12,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-04 19:34:28,443 INFO [train.py:968] (1/2) Epoch 9, batch 24500, giga_loss[loss=0.3099, simple_loss=0.3708, pruned_loss=0.1245, over 28838.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.386, pruned_loss=0.1347, over 5654446.02 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3722, pruned_loss=0.1204, over 5760818.45 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3875, pruned_loss=0.1359, over 5640172.02 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:34:49,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2151, 1.4620, 1.1753, 1.0622], device='cuda:1'), covar=tensor([0.2441, 0.2342, 0.2684, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.1243, 0.0930, 0.1100, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:35:02,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:35:17,157 INFO [train.py:968] (1/2) Epoch 9, batch 24550, giga_loss[loss=0.3115, simple_loss=0.3822, pruned_loss=0.1204, over 28859.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3849, pruned_loss=0.1318, over 5663111.29 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3725, pruned_loss=0.1208, over 5765918.30 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3862, pruned_loss=0.1328, over 5643857.09 frames. ], batch size: 186, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:35:21,926 INFO [optim.py:369] (1/2) 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 19:35:41,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 19:36:08,443 INFO [train.py:968] (1/2) Epoch 9, batch 24600, giga_loss[loss=0.3528, simple_loss=0.4084, pruned_loss=0.1486, over 28851.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3876, pruned_loss=0.1322, over 5664403.44 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3727, pruned_loss=0.1211, over 5760061.91 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3888, pruned_loss=0.1329, over 5652257.86 frames. ], batch size: 174, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:36:57,017 INFO [train.py:968] (1/2) Epoch 9, batch 24650, giga_loss[loss=0.3211, simple_loss=0.3839, pruned_loss=0.1292, over 28924.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3881, pruned_loss=0.1328, over 5656406.08 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3731, pruned_loss=0.1216, over 5753114.53 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.389, pruned_loss=0.1332, over 5650259.03 frames. ], batch size: 106, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:37:03,399 INFO [optim.py:369] (1/2) 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 19:37:24,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4297, 1.9853, 1.4545, 0.6885], device='cuda:1'), covar=tensor([0.2960, 0.1567, 0.2209, 0.3533], device='cuda:1'), in_proj_covar=tensor([0.1499, 0.1426, 0.1464, 0.1224], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 19:37:29,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=388876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:37:30,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=388879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:37:46,522 INFO [train.py:968] (1/2) Epoch 9, batch 24700, giga_loss[loss=0.2922, simple_loss=0.3568, pruned_loss=0.1138, over 28826.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.388, pruned_loss=0.1329, over 5664545.07 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3733, pruned_loss=0.1218, over 5744359.60 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.389, pruned_loss=0.1334, over 5665190.03 frames. ], batch size: 99, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:37:54,383 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=388908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:37:54,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2952, 1.3802, 1.3814, 1.2882], device='cuda:1'), covar=tensor([0.1230, 0.1508, 0.1864, 0.1485], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0733, 0.0665, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 19:38:04,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2579, 2.8609, 1.9393, 1.6422], device='cuda:1'), covar=tensor([0.1738, 0.1104, 0.1478, 0.1775], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1516, 0.1474, 0.1581], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 19:38:34,038 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-04 19:38:34,771 INFO [train.py:968] (1/2) Epoch 9, batch 24750, libri_loss[loss=0.3202, simple_loss=0.3873, pruned_loss=0.1266, over 29367.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3867, pruned_loss=0.1326, over 5676995.74 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.122, over 5748484.74 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3878, pruned_loss=0.1331, over 5672126.55 frames. ], batch size: 92, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:38:39,690 INFO [optim.py:369] (1/2) 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 19:39:03,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=388978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:39:13,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3673, 2.0463, 1.5327, 0.5342], device='cuda:1'), covar=tensor([0.3222, 0.1850, 0.2472, 0.3932], device='cuda:1'), in_proj_covar=tensor([0.1509, 0.1438, 0.1474, 0.1231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 19:39:20,150 INFO [train.py:968] (1/2) Epoch 9, batch 24800, giga_loss[loss=0.3302, simple_loss=0.3837, pruned_loss=0.1383, over 28677.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.386, pruned_loss=0.1338, over 5681220.09 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.122, over 5752025.15 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3871, pruned_loss=0.1344, over 5672787.62 frames. ], batch size: 242, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:40:00,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4899, 1.6407, 1.3191, 1.7688], device='cuda:1'), covar=tensor([0.2189, 0.2207, 0.2375, 0.2280], device='cuda:1'), in_proj_covar=tensor([0.1248, 0.0929, 0.1100, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 19:40:03,882 INFO [train.py:968] (1/2) Epoch 9, batch 24850, giga_loss[loss=0.3192, simple_loss=0.3841, pruned_loss=0.1271, over 29045.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3843, pruned_loss=0.1329, over 5675918.23 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3733, pruned_loss=0.1222, over 5745866.58 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3856, pruned_loss=0.1335, over 5671857.36 frames. ], batch size: 128, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:40:09,334 INFO [optim.py:369] (1/2) 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 19:40:34,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9117, 3.6789, 3.4356, 1.6471], device='cuda:1'), covar=tensor([0.0655, 0.0829, 0.0890, 0.2251], device='cuda:1'), in_proj_covar=tensor([0.0994, 0.0934, 0.0823, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 19:40:48,235 INFO [train.py:968] (1/2) Epoch 9, batch 24900, libri_loss[loss=0.3701, simple_loss=0.4248, pruned_loss=0.1577, over 29385.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3841, pruned_loss=0.1313, over 5683279.01 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3735, pruned_loss=0.1226, over 5751655.29 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3853, pruned_loss=0.1318, over 5672264.56 frames. ], batch size: 92, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:41:11,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=389121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:41:13,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=389124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:41:36,830 INFO [train.py:968] (1/2) Epoch 9, batch 24950, giga_loss[loss=0.2935, simple_loss=0.3555, pruned_loss=0.1157, over 28635.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3837, pruned_loss=0.1304, over 5685444.39 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3735, pruned_loss=0.1229, over 5754633.37 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3848, pruned_loss=0.1306, over 5672873.93 frames. ], batch size: 85, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:41:42,759 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:42:21,984 INFO [train.py:968] (1/2) Epoch 9, batch 25000, giga_loss[loss=0.3156, simple_loss=0.3754, pruned_loss=0.1279, over 28424.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3834, pruned_loss=0.1304, over 5683411.06 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3739, pruned_loss=0.1233, over 5755049.82 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3843, pruned_loss=0.1305, over 5670550.51 frames. ], batch size: 71, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:42:52,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0445, 1.1801, 1.3394, 1.0099], device='cuda:1'), covar=tensor([0.1180, 0.1254, 0.1773, 0.1425], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0732, 0.0664, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 19:43:12,804 INFO [train.py:968] (1/2) Epoch 9, batch 25050, giga_loss[loss=0.2956, simple_loss=0.368, pruned_loss=0.1116, over 28926.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3819, pruned_loss=0.1299, over 5686341.45 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3742, pruned_loss=0.1234, over 5756751.93 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3825, pruned_loss=0.1299, over 5673816.22 frames. ], batch size: 174, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:43:18,228 INFO [optim.py:369] (1/2) 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 19:43:59,571 INFO [train.py:968] (1/2) Epoch 9, batch 25100, giga_loss[loss=0.2811, simple_loss=0.3534, pruned_loss=0.1044, over 29050.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3804, pruned_loss=0.1295, over 5677402.86 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.374, pruned_loss=0.1234, over 5760232.35 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3813, pruned_loss=0.1298, over 5661888.04 frames. ], batch size: 136, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:44:06,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:44:15,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0426, 2.4867, 1.7871, 1.7019], device='cuda:1'), covar=tensor([0.1839, 0.1308, 0.1627, 0.1668], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1508, 0.1477, 0.1579], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 19:44:40,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2555, 1.6649, 1.2010, 0.4857], device='cuda:1'), covar=tensor([0.1893, 0.1095, 0.1898, 0.3116], device='cuda:1'), in_proj_covar=tensor([0.1498, 0.1432, 0.1472, 0.1228], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 19:44:46,254 INFO [train.py:968] (1/2) Epoch 9, batch 25150, giga_loss[loss=0.3625, simple_loss=0.4133, pruned_loss=0.1558, over 28755.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.381, pruned_loss=0.1307, over 5675762.86 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5759953.67 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3816, pruned_loss=0.1309, over 5662079.87 frames. ], batch size: 284, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:44:50,875 INFO [optim.py:369] (1/2) 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 19:45:01,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4336, 1.4848, 4.5972, 3.4857], device='cuda:1'), covar=tensor([0.1569, 0.2286, 0.0333, 0.0735], device='cuda:1'), in_proj_covar=tensor([0.0628, 0.0574, 0.0824, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 19:45:28,483 INFO [train.py:968] (1/2) Epoch 9, batch 25200, libri_loss[loss=0.35, simple_loss=0.4034, pruned_loss=0.1483, over 26419.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3798, pruned_loss=0.1302, over 5676577.08 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3747, pruned_loss=0.1239, over 5760785.84 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3801, pruned_loss=0.1303, over 5661979.15 frames. ], batch size: 136, lr: 3.59e-03, grad_scale: 8.0
+2023-03-04 19:46:16,305 INFO [train.py:968] (1/2) Epoch 9, batch 25250, giga_loss[loss=0.3897, simple_loss=0.4306, pruned_loss=0.1744, over 27516.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3774, pruned_loss=0.1285, over 5671923.03 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3749, pruned_loss=0.1241, over 5754044.00 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5664765.34 frames. ], batch size: 472, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:46:22,309 INFO [optim.py:369] (1/2) 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 19:47:03,170 INFO [train.py:968] (1/2) Epoch 9, batch 25300, giga_loss[loss=0.3079, simple_loss=0.374, pruned_loss=0.1209, over 28663.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3784, pruned_loss=0.13, over 5671290.59 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5756686.30 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3786, pruned_loss=0.1301, over 5660097.80 frames. ], batch size: 242, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:47:51,612 INFO [train.py:968] (1/2) Epoch 9, batch 25350, giga_loss[loss=0.3433, simple_loss=0.411, pruned_loss=0.1378, over 28904.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3802, pruned_loss=0.1308, over 5669420.65 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3748, pruned_loss=0.1241, over 5756487.47 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3804, pruned_loss=0.1311, over 5658791.01 frames. ], batch size: 199, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:47:59,164 INFO [optim.py:369] (1/2) 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 19:48:11,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2603, 1.9368, 1.3952, 1.4533], device='cuda:1'), covar=tensor([0.0785, 0.0282, 0.0315, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:1')
+2023-03-04 19:48:21,997 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-04 19:48:22,021 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-04 19:48:34,397 INFO [train.py:968] (1/2) Epoch 9, batch 25400, libri_loss[loss=0.2647, simple_loss=0.3408, pruned_loss=0.09427, over 29567.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.379, pruned_loss=0.1288, over 5669163.06 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.1239, over 5750150.42 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3799, pruned_loss=0.1295, over 5662163.87 frames. ], batch size: 75, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:48:35,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:48:46,232 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 19:49:19,232 INFO [train.py:968] (1/2) Epoch 9, batch 25450, giga_loss[loss=0.2856, simple_loss=0.3591, pruned_loss=0.106, over 28552.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3788, pruned_loss=0.1282, over 5668655.40 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3741, pruned_loss=0.1238, over 5750261.95 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3799, pruned_loss=0.1291, over 5660050.62 frames. ], batch size: 336, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:49:25,440 INFO [optim.py:369] (1/2) 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 19:49:50,325 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 25500, giga_loss[loss=0.3253, simple_loss=0.3829, pruned_loss=0.1339, over 28769.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.379, pruned_loss=0.1288, over 5662941.77 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1239, over 5744348.35 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3802, pruned_loss=0.1295, over 5658939.78 frames. ], batch size: 284, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:50:31,508 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:50:50,494 INFO [train.py:968] (1/2) Epoch 9, batch 25550, giga_loss[loss=0.2997, simple_loss=0.3676, pruned_loss=0.1159, over 29092.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3814, pruned_loss=0.1315, over 5663182.84 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.374, pruned_loss=0.124, over 5747752.75 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3825, pruned_loss=0.1322, over 5655116.00 frames. ], batch size: 155, lr: 3.59e-03, grad_scale: 2.0
+2023-03-04 19:50:59,629 INFO [optim.py:369] (1/2) 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 19:51:00,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3732, 1.6653, 1.3941, 1.5563], device='cuda:1'), covar=tensor([0.0764, 0.0298, 0.0307, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0117, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:1')
+2023-03-04 19:51:39,266 INFO [train.py:968] (1/2) Epoch 9, batch 25600, giga_loss[loss=0.3896, simple_loss=0.4225, pruned_loss=0.1784, over 27593.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3832, pruned_loss=0.1341, over 5658497.80 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3742, pruned_loss=0.1244, over 5752044.74 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3841, pruned_loss=0.1345, over 5645920.07 frames. ], batch size: 474, lr: 3.59e-03, grad_scale: 4.0
+2023-03-04 19:52:04,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=389824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:52:08,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=389827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:52:09,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-04 19:52:31,748 INFO [train.py:968] (1/2) Epoch 9, batch 25650, giga_loss[loss=0.2704, simple_loss=0.3409, pruned_loss=0.09999, over 28910.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3848, pruned_loss=0.1364, over 5664940.35 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1245, over 5742705.52 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3854, pruned_loss=0.1365, over 5662608.42 frames. ], batch size: 199, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:52:41,331 INFO [optim.py:369] (1/2) 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 19:52:41,616 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=389856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:53:21,979 INFO [train.py:968] (1/2) Epoch 9, batch 25700, libri_loss[loss=0.3139, simple_loss=0.3794, pruned_loss=0.1242, over 29533.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3874, pruned_loss=0.139, over 5649397.20 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3746, pruned_loss=0.1247, over 5745526.17 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3879, pruned_loss=0.1393, over 5642479.35 frames. ], batch size: 84, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:54:03,813 INFO [train.py:968] (1/2) Epoch 9, batch 25750, giga_loss[loss=0.284, simple_loss=0.3571, pruned_loss=0.1054, over 28995.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3856, pruned_loss=0.1376, over 5658051.09 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.375, pruned_loss=0.1253, over 5740608.84 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3861, pruned_loss=0.1378, over 5653078.65 frames. ], batch size: 155, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:54:11,825 INFO [optim.py:369] (1/2) 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 19:54:12,233 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:54:30,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=389974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:54:37,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2834, 3.2212, 1.3779, 1.3746], device='cuda:1'), covar=tensor([0.0987, 0.0388, 0.0882, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0505, 0.0331, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 19:54:38,827 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=389983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:54:51,395 INFO [train.py:968] (1/2) Epoch 9, batch 25800, giga_loss[loss=0.307, simple_loss=0.3762, pruned_loss=0.1189, over 28645.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3843, pruned_loss=0.1363, over 5658197.27 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3741, pruned_loss=0.1248, over 5743010.50 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3856, pruned_loss=0.1371, over 5650639.75 frames. ], batch size: 262, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:55:35,462 INFO [train.py:968] (1/2) Epoch 9, batch 25850, giga_loss[loss=0.2785, simple_loss=0.3527, pruned_loss=0.1022, over 29039.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3816, pruned_loss=0.1324, over 5669605.33 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3743, pruned_loss=0.1252, over 5743262.89 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3826, pruned_loss=0.1329, over 5661800.03 frames. ], batch size: 128, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:55:44,181 INFO [optim.py:369] (1/2) 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 19:56:20,936 INFO [train.py:968] (1/2) Epoch 9, batch 25900, giga_loss[loss=0.3089, simple_loss=0.3769, pruned_loss=0.1205, over 29020.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3799, pruned_loss=0.1313, over 5666600.59 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3749, pruned_loss=0.1256, over 5746776.69 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3803, pruned_loss=0.1314, over 5655796.29 frames. ], batch size: 164, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:56:27,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-04 19:56:28,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:56:41,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:56:44,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:57:02,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1814, 1.7250, 1.2395, 0.3732], device='cuda:1'), covar=tensor([0.2217, 0.1417, 0.2196, 0.3253], device='cuda:1'), in_proj_covar=tensor([0.1498, 0.1442, 0.1466, 0.1232], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 19:57:08,675 INFO [train.py:968] (1/2) Epoch 9, batch 25950, giga_loss[loss=0.2936, simple_loss=0.363, pruned_loss=0.1121, over 28904.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3783, pruned_loss=0.1309, over 5670076.94 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3752, pruned_loss=0.1257, over 5749634.52 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3784, pruned_loss=0.131, over 5657534.40 frames. ], batch size: 112, lr: 3.58e-03, grad_scale: 2.0
+2023-03-04 19:57:09,483 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:57:15,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:57:16,007 INFO [optim.py:369] (1/2) 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 19:58:03,319 INFO [train.py:968] (1/2) Epoch 9, batch 26000, giga_loss[loss=0.3826, simple_loss=0.403, pruned_loss=0.1812, over 23443.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3766, pruned_loss=0.1293, over 5678384.62 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3752, pruned_loss=0.1257, over 5749634.52 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3766, pruned_loss=0.1293, over 5668622.61 frames. ], batch size: 705, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:58:45,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:58:45,502 INFO [train.py:968] (1/2) Epoch 9, batch 26050, libri_loss[loss=0.3035, simple_loss=0.37, pruned_loss=0.1185, over 29672.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3802, pruned_loss=0.1318, over 5681390.67 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3753, pruned_loss=0.1261, over 5746474.73 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3804, pruned_loss=0.1318, over 5672299.20 frames. ], batch size: 88, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:58:47,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:58:52,885 INFO [optim.py:369] (1/2) 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 19:59:13,710 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 19:59:15,785 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-04 19:59:33,524 INFO [train.py:968] (1/2) Epoch 9, batch 26100, giga_loss[loss=0.324, simple_loss=0.41, pruned_loss=0.1189, over 28570.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3828, pruned_loss=0.1308, over 5672421.40 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1262, over 5737853.71 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3828, pruned_loss=0.1307, over 5672117.30 frames. ], batch size: 336, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 19:59:39,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-04 19:59:48,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:00:06,012 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 26150, giga_loss[loss=0.2723, simple_loss=0.3565, pruned_loss=0.094, over 29034.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3851, pruned_loss=0.1303, over 5672643.40 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3757, pruned_loss=0.1265, over 5730751.20 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3851, pruned_loss=0.1301, over 5676951.17 frames. ], batch size: 136, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:00:29,888 INFO [optim.py:369] (1/2) 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 20:00:33,050 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:01:10,668 INFO [train.py:968] (1/2) Epoch 9, batch 26200, giga_loss[loss=0.3409, simple_loss=0.4028, pruned_loss=0.1395, over 29043.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.387, pruned_loss=0.1321, over 5677950.70 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3761, pruned_loss=0.1269, over 5732928.84 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3869, pruned_loss=0.1317, over 5678496.20 frames. ], batch size: 155, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:01:26,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3413, 1.5578, 1.2391, 1.5677], device='cuda:1'), covar=tensor([0.2139, 0.2014, 0.2202, 0.1859], device='cuda:1'), in_proj_covar=tensor([0.1246, 0.0929, 0.1098, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 20:01:36,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0609, 1.2580, 1.3535, 1.0925], device='cuda:1'), covar=tensor([0.1192, 0.1126, 0.1767, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0730, 0.0655, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 20:01:55,252 INFO [train.py:968] (1/2) Epoch 9, batch 26250, giga_loss[loss=0.2996, simple_loss=0.3756, pruned_loss=0.1118, over 29113.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3882, pruned_loss=0.1335, over 5686072.70 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3756, pruned_loss=0.1267, over 5737876.63 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3889, pruned_loss=0.1335, over 5680362.27 frames. ], batch size: 155, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:02:03,540 INFO [optim.py:369] (1/2) 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 20:02:22,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:26,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:44,341 INFO [train.py:968] (1/2) Epoch 9, batch 26300, libri_loss[loss=0.3181, simple_loss=0.3797, pruned_loss=0.1283, over 29641.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.389, pruned_loss=0.1355, over 5679254.92 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3753, pruned_loss=0.1264, over 5742858.45 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3901, pruned_loss=0.1359, over 5668663.46 frames. ], batch size: 91, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:02:47,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:49,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:02:52,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:03:16,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:03:19,051 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:03:31,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-04 20:03:32,741 INFO [train.py:968] (1/2) Epoch 9, batch 26350, libri_loss[loss=0.2663, simple_loss=0.3315, pruned_loss=0.1005, over 29557.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3876, pruned_loss=0.135, over 5693065.86 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3753, pruned_loss=0.1263, over 5747024.23 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3889, pruned_loss=0.1357, over 5679491.28 frames. ], batch size: 75, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:03:39,606 INFO [optim.py:369] (1/2) 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 20:03:43,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3182, 1.5615, 1.2486, 1.3656], device='cuda:1'), covar=tensor([0.2382, 0.2281, 0.2539, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.1246, 0.0928, 0.1101, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 20:04:16,982 INFO [train.py:968] (1/2) Epoch 9, batch 26400, libri_loss[loss=0.3226, simple_loss=0.3895, pruned_loss=0.1279, over 29148.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3854, pruned_loss=0.1343, over 5694378.37 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3748, pruned_loss=0.1262, over 5751531.51 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3872, pruned_loss=0.1352, over 5678055.45 frames. ], batch size: 101, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:05:08,376 INFO [train.py:968] (1/2) Epoch 9, batch 26450, giga_loss[loss=0.3279, simple_loss=0.3875, pruned_loss=0.1341, over 28879.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3837, pruned_loss=0.1335, over 5697899.99 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1264, over 5752553.39 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3849, pruned_loss=0.1342, over 5683202.93 frames. ], batch size: 186, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:05:18,340 INFO [optim.py:369] (1/2) 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 20:05:35,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=390674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:05:39,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:05:49,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=390690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:05:57,375 INFO [train.py:968] (1/2) Epoch 9, batch 26500, giga_loss[loss=0.2979, simple_loss=0.3666, pruned_loss=0.1146, over 29041.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.383, pruned_loss=0.1334, over 5679707.62 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3748, pruned_loss=0.1262, over 5745249.33 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3844, pruned_loss=0.1341, over 5674103.90 frames. ], batch size: 155, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:06:04,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:06:40,356 INFO [train.py:968] (1/2) Epoch 9, batch 26550, libri_loss[loss=0.2885, simple_loss=0.3419, pruned_loss=0.1175, over 29502.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3831, pruned_loss=0.1341, over 5683364.88 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3746, pruned_loss=0.1262, over 5746488.82 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3847, pruned_loss=0.135, over 5676039.05 frames. ], batch size: 70, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:06:49,895 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 26600, giga_loss[loss=0.3542, simple_loss=0.3865, pruned_loss=0.161, over 23697.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3814, pruned_loss=0.1341, over 5664610.57 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3749, pruned_loss=0.1264, over 5749415.43 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3825, pruned_loss=0.1348, over 5655213.72 frames. ], batch size: 705, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:07:30,799 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:08:00,479 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=390836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:08:13,032 INFO [train.py:968] (1/2) Epoch 9, batch 26650, giga_loss[loss=0.3071, simple_loss=0.3718, pruned_loss=0.1213, over 28651.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3815, pruned_loss=0.1341, over 5658604.96 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3749, pruned_loss=0.1264, over 5743406.28 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3826, pruned_loss=0.1348, over 5655337.80 frames. ], batch size: 242, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:08:22,341 INFO [optim.py:369] (1/2) 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 20:08:28,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=390865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:08:43,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7015, 1.9276, 2.0377, 1.4962], device='cuda:1'), covar=tensor([0.1751, 0.2160, 0.1328, 0.1574], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0711, 0.0842, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 20:08:56,640 INFO [train.py:968] (1/2) Epoch 9, batch 26700, giga_loss[loss=0.2954, simple_loss=0.374, pruned_loss=0.1084, over 28947.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3812, pruned_loss=0.1323, over 5669494.52 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3753, pruned_loss=0.1267, over 5745408.66 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3818, pruned_loss=0.1327, over 5662592.76 frames. ], batch size: 164, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:09:45,979 INFO [train.py:968] (1/2) Epoch 9, batch 26750, giga_loss[loss=0.33, simple_loss=0.3886, pruned_loss=0.1357, over 28977.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3845, pruned_loss=0.1348, over 5658481.18 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1268, over 5745839.94 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.385, pruned_loss=0.1352, over 5651394.65 frames. ], batch size: 213, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:09:57,364 INFO [optim.py:369] (1/2) 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 20:10:22,730 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=390983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:10:35,163 INFO [train.py:968] (1/2) Epoch 9, batch 26800, giga_loss[loss=0.2857, simple_loss=0.3553, pruned_loss=0.108, over 28607.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3838, pruned_loss=0.1345, over 5670771.14 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3754, pruned_loss=0.1268, over 5749691.55 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3844, pruned_loss=0.135, over 5659451.24 frames. ], batch size: 60, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:10:37,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 20:10:51,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3848, 1.5269, 1.1612, 1.5192], device='cuda:1'), covar=tensor([0.2467, 0.2471, 0.2841, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.1254, 0.0934, 0.1107, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 20:11:15,067 INFO [train.py:968] (1/2) Epoch 9, batch 26850, libri_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1208, over 29554.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.385, pruned_loss=0.133, over 5676507.83 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3758, pruned_loss=0.127, over 5750760.08 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3857, pruned_loss=0.1335, over 5663156.88 frames. ], batch size: 80, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:11:24,252 INFO [optim.py:369] (1/2) 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 20:11:27,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:00,235 INFO [train.py:968] (1/2) Epoch 9, batch 26900, giga_loss[loss=0.3161, simple_loss=0.3928, pruned_loss=0.1197, over 28812.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3859, pruned_loss=0.1313, over 5686755.69 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5747779.57 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.387, pruned_loss=0.1318, over 5676171.13 frames. ], batch size: 174, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:12:07,057 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:13,435 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:17,973 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:12:30,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3891, 1.5787, 1.5407, 1.4183], device='cuda:1'), covar=tensor([0.1405, 0.1655, 0.1776, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0730, 0.0659, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 20:12:36,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-04 20:12:41,823 INFO [train.py:968] (1/2) Epoch 9, batch 26950, giga_loss[loss=0.3021, simple_loss=0.3761, pruned_loss=0.1141, over 29003.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3885, pruned_loss=0.1324, over 5693262.58 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3759, pruned_loss=0.1275, over 5750968.17 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3895, pruned_loss=0.1327, over 5679815.37 frames. ], batch size: 136, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:12:50,735 INFO [optim.py:369] (1/2) 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 20:13:09,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:13:09,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 20:13:16,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8045, 1.7801, 1.5661, 1.5981], device='cuda:1'), covar=tensor([0.1066, 0.1842, 0.1598, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0729, 0.0658, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 20:13:28,319 INFO [train.py:968] (1/2) Epoch 9, batch 27000, giga_loss[loss=0.5064, simple_loss=0.4932, pruned_loss=0.2598, over 26510.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3904, pruned_loss=0.1346, over 5690456.14 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1273, over 5754434.05 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3919, pruned_loss=0.1351, over 5674980.46 frames. ], batch size: 555, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:13:28,319 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 20:13:36,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3671, 2.8853, 1.4706, 1.4962], device='cuda:1'), covar=tensor([0.0885, 0.0399, 0.0879, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0507, 0.0330, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 20:13:37,446 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 20:14:14,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:14:29,448 INFO [train.py:968] (1/2) Epoch 9, batch 27050, giga_loss[loss=0.3045, simple_loss=0.3687, pruned_loss=0.1202, over 28897.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3943, pruned_loss=0.139, over 5668879.65 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3757, pruned_loss=0.1274, over 5745411.88 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3954, pruned_loss=0.1393, over 5664135.35 frames. ], batch size: 106, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:14:43,148 INFO [optim.py:369] (1/2) 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 20:15:22,094 INFO [train.py:968] (1/2) Epoch 9, batch 27100, giga_loss[loss=0.389, simple_loss=0.4339, pruned_loss=0.172, over 27639.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3935, pruned_loss=0.1392, over 5658966.94 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.376, pruned_loss=0.1277, over 5746667.78 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3942, pruned_loss=0.1393, over 5653287.92 frames. ], batch size: 474, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:15:48,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:15:51,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:16:01,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:16:12,994 INFO [train.py:968] (1/2) Epoch 9, batch 27150, libri_loss[loss=0.356, simple_loss=0.4165, pruned_loss=0.1477, over 25998.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.392, pruned_loss=0.138, over 5649183.61 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3762, pruned_loss=0.1279, over 5745512.57 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3927, pruned_loss=0.138, over 5644661.36 frames. ], batch size: 136, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:16:17,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-04 20:16:19,272 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:16:22,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:16:42,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4199, 1.5302, 1.0634, 1.1551], device='cuda:1'), covar=tensor([0.0769, 0.0561, 0.1156, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0443, 0.0497, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 20:16:56,975 INFO [train.py:968] (1/2) Epoch 9, batch 27200, giga_loss[loss=0.3705, simple_loss=0.4237, pruned_loss=0.1586, over 27923.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.392, pruned_loss=0.1364, over 5654949.94 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5744508.61 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3926, pruned_loss=0.1365, over 5650415.58 frames. ], batch size: 412, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:17:29,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4556, 3.5379, 1.5415, 1.4792], device='cuda:1'), covar=tensor([0.0914, 0.0347, 0.0914, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0508, 0.0332, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 20:17:35,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:17:46,401 INFO [train.py:968] (1/2) Epoch 9, batch 27250, giga_loss[loss=0.4186, simple_loss=0.4537, pruned_loss=0.1917, over 27957.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3913, pruned_loss=0.134, over 5661121.77 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3765, pruned_loss=0.1282, over 5733288.22 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3919, pruned_loss=0.1341, over 5665360.46 frames. ], batch size: 412, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:17:55,834 INFO [optim.py:369] (1/2) 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 20:18:12,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4542, 1.6911, 1.3707, 1.7134], device='cuda:1'), covar=tensor([0.1973, 0.1883, 0.1945, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.1257, 0.0936, 0.1113, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 20:18:18,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:18:24,877 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=391486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:18:29,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:18:35,311 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 27300, giga_loss[loss=0.3777, simple_loss=0.4155, pruned_loss=0.17, over 27949.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3918, pruned_loss=0.135, over 5658965.48 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3762, pruned_loss=0.1279, over 5736541.44 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3929, pruned_loss=0.1354, over 5657988.24 frames. ], batch size: 412, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:18:41,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:18:44,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:19:10,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:19:22,929 INFO [train.py:968] (1/2) Epoch 9, batch 27350, giga_loss[loss=0.3454, simple_loss=0.3979, pruned_loss=0.1464, over 28982.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3912, pruned_loss=0.1347, over 5661120.17 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3764, pruned_loss=0.1281, over 5728672.74 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.392, pruned_loss=0.135, over 5666407.77 frames. ], batch size: 213, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:19:32,485 INFO [optim.py:369] (1/2) 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 20:19:52,178 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:19:55,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:04,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0621, 1.1950, 1.3346, 1.1399], device='cuda:1'), covar=tensor([0.1292, 0.1205, 0.1797, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0741, 0.0665, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 20:20:11,371 INFO [train.py:968] (1/2) Epoch 9, batch 27400, giga_loss[loss=0.3071, simple_loss=0.3638, pruned_loss=0.1252, over 28671.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3897, pruned_loss=0.1355, over 5656891.68 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3768, pruned_loss=0.1284, over 5733225.58 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3904, pruned_loss=0.1357, over 5655492.44 frames. ], batch size: 92, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:20:21,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:25,528 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:34,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:36,031 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:46,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:47,012 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:50,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=391640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:20:57,662 INFO [train.py:968] (1/2) Epoch 9, batch 27450, giga_loss[loss=0.3225, simple_loss=0.39, pruned_loss=0.1275, over 28794.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3876, pruned_loss=0.1352, over 5649371.83 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3773, pruned_loss=0.1288, over 5738706.93 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3881, pruned_loss=0.1351, over 5640889.35 frames. ], batch size: 243, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:21:03,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:21:08,732 INFO [optim.py:369] (1/2) 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 20:21:15,191 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:21:19,899 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:21:49,263 INFO [train.py:968] (1/2) Epoch 9, batch 27500, giga_loss[loss=0.3128, simple_loss=0.3736, pruned_loss=0.126, over 28732.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3848, pruned_loss=0.1335, over 5655026.74 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3774, pruned_loss=0.1289, over 5737782.05 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3852, pruned_loss=0.1334, over 5647971.69 frames. ], batch size: 119, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:22:01,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:22:36,043 INFO [train.py:968] (1/2) Epoch 9, batch 27550, giga_loss[loss=0.3327, simple_loss=0.3858, pruned_loss=0.1398, over 28303.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3845, pruned_loss=0.1349, over 5650038.24 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3769, pruned_loss=0.1285, over 5739141.48 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3854, pruned_loss=0.1354, over 5641449.48 frames. ], batch size: 368, lr: 3.58e-03, grad_scale: 4.0
+2023-03-04 20:22:38,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:22:42,264 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:23:06,938 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 27600, giga_loss[loss=0.3221, simple_loss=0.3809, pruned_loss=0.1317, over 28904.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3833, pruned_loss=0.1341, over 5656185.15 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3772, pruned_loss=0.1286, over 5742659.63 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3839, pruned_loss=0.1345, over 5644477.50 frames. ], batch size: 199, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:24:05,276 INFO [train.py:968] (1/2) Epoch 9, batch 27650, giga_loss[loss=0.2853, simple_loss=0.3612, pruned_loss=0.1047, over 28830.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3794, pruned_loss=0.1299, over 5656270.63 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3772, pruned_loss=0.1288, over 5736127.55 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3799, pruned_loss=0.1301, over 5650442.90 frames. ], batch size: 199, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:24:11,546 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=391853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:24:14,443 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:24:19,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=391861.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:24:39,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=391885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:24:52,660 INFO [train.py:968] (1/2) Epoch 9, batch 27700, giga_loss[loss=0.2898, simple_loss=0.3649, pruned_loss=0.1074, over 29006.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3766, pruned_loss=0.1268, over 5665636.24 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3773, pruned_loss=0.1288, over 5739358.20 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.377, pruned_loss=0.1269, over 5656738.47 frames. ], batch size: 164, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:25:09,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 27750, giga_loss[loss=0.3821, simple_loss=0.4253, pruned_loss=0.1694, over 28952.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3777, pruned_loss=0.128, over 5658313.71 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3776, pruned_loss=0.129, over 5740109.94 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3776, pruned_loss=0.1278, over 5648958.89 frames. ], batch size: 285, lr: 3.58e-03, grad_scale: 8.0
+2023-03-04 20:25:54,238 INFO [optim.py:369] (1/2) 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 20:26:21,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-04 20:26:35,464 INFO [train.py:968] (1/2) Epoch 9, batch 27800, giga_loss[loss=0.239, simple_loss=0.3217, pruned_loss=0.07816, over 28888.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3729, pruned_loss=0.1252, over 5662918.94 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3774, pruned_loss=0.1289, over 5738805.39 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3731, pruned_loss=0.1252, over 5656416.22 frames. ], batch size: 145, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:26:39,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1778, 3.9733, 3.8208, 1.6432], device='cuda:1'), covar=tensor([0.0589, 0.0746, 0.0735, 0.2237], device='cuda:1'), in_proj_covar=tensor([0.1003, 0.0952, 0.0836, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 20:26:41,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=392004.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:26:44,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=392007.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:27:14,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=392036.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:27:23,089 INFO [train.py:968] (1/2) Epoch 9, batch 27850, giga_loss[loss=0.3247, simple_loss=0.3915, pruned_loss=0.129, over 28849.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3722, pruned_loss=0.1254, over 5664843.29 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3772, pruned_loss=0.1288, over 5744053.56 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3722, pruned_loss=0.1252, over 5651250.93 frames. ], batch size: 174, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:27:35,465 INFO [optim.py:369] (1/2) 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 20:27:50,788 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-04 20:28:03,002 INFO [train.py:968] (1/2) Epoch 9, batch 27900, giga_loss[loss=0.3262, simple_loss=0.3905, pruned_loss=0.131, over 28837.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3748, pruned_loss=0.1263, over 5677583.11 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.377, pruned_loss=0.1287, over 5741521.76 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3748, pruned_loss=0.1262, over 5665440.19 frames. ], batch size: 199, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:28:03,185 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=392098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:28:16,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9424, 3.7507, 3.5452, 1.6835], device='cuda:1'), covar=tensor([0.0585, 0.0700, 0.0719, 0.2285], device='cuda:1'), in_proj_covar=tensor([0.1002, 0.0952, 0.0833, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 20:28:47,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 20:28:47,952 INFO [train.py:968] (1/2) Epoch 9, batch 27950, giga_loss[loss=0.2892, simple_loss=0.3583, pruned_loss=0.11, over 28702.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3765, pruned_loss=0.1274, over 5646949.36 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.377, pruned_loss=0.1288, over 5720369.57 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3765, pruned_loss=0.1272, over 5652800.88 frames. ], batch size: 262, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:28:58,919 INFO [optim.py:369] (1/2) 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 20:29:23,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6059, 2.0540, 1.9286, 1.6570], device='cuda:1'), covar=tensor([0.0728, 0.0252, 0.0247, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:1')
+2023-03-04 20:29:32,695 INFO [train.py:968] (1/2) Epoch 9, batch 28000, giga_loss[loss=0.3046, simple_loss=0.3672, pruned_loss=0.121, over 28854.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1282, over 5654456.32 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3776, pruned_loss=0.129, over 5726242.12 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3772, pruned_loss=0.1277, over 5651860.28 frames. ], batch size: 199, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:29:54,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9074, 1.0735, 3.2547, 2.7149], device='cuda:1'), covar=tensor([0.1599, 0.2408, 0.0430, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0573, 0.0823, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 20:29:54,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1268, 1.2777, 1.1816, 1.0399], device='cuda:1'), covar=tensor([0.1146, 0.1162, 0.0823, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.1631, 0.1522, 0.1487, 0.1595], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 20:30:19,373 INFO [train.py:968] (1/2) Epoch 9, batch 28050, giga_loss[loss=0.3002, simple_loss=0.3592, pruned_loss=0.1206, over 28620.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3793, pruned_loss=0.1299, over 5639661.05 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.378, pruned_loss=0.1293, over 5715846.97 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3786, pruned_loss=0.1293, over 5645390.49 frames. ], batch size: 92, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:30:31,077 INFO [optim.py:369] (1/2) 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 20:30:54,872 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 28100, giga_loss[loss=0.3126, simple_loss=0.3854, pruned_loss=0.1199, over 28659.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3814, pruned_loss=0.1313, over 5633044.74 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3794, pruned_loss=0.1305, over 5691824.14 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3796, pruned_loss=0.1298, over 5656925.17 frames. ], batch size: 307, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:31:42,319 INFO [train.py:968] (1/2) Epoch 9, batch 28150, giga_loss[loss=0.314, simple_loss=0.3724, pruned_loss=0.1278, over 29005.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.384, pruned_loss=0.1331, over 5636479.31 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3796, pruned_loss=0.1306, over 5688237.81 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3824, pruned_loss=0.1319, over 5656445.99 frames. ], batch size: 100, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:31:57,324 INFO [optim.py:369] (1/2) 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 20:31:58,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5093, 1.6427, 1.4455, 1.2909], device='cuda:1'), covar=tensor([0.1754, 0.1550, 0.1161, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.1628, 0.1522, 0.1481, 0.1594], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 20:32:31,546 INFO [train.py:968] (1/2) Epoch 9, batch 28200, giga_loss[loss=0.3334, simple_loss=0.3917, pruned_loss=0.1376, over 28520.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3847, pruned_loss=0.1331, over 5644640.98 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3795, pruned_loss=0.1305, over 5690380.00 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3837, pruned_loss=0.1323, over 5657917.68 frames. ], batch size: 336, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:33:02,196 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=392433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:33:05,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 28250, giga_loss[loss=0.3219, simple_loss=0.3834, pruned_loss=0.1302, over 29070.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3862, pruned_loss=0.1349, over 5644483.56 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3794, pruned_loss=0.1304, over 5696028.90 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3857, pruned_loss=0.1344, over 5648022.28 frames. ], batch size: 155, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:33:29,291 INFO [optim.py:369] (1/2) 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 20:33:30,870 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=392465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:33:38,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 28300, giga_loss[loss=0.3683, simple_loss=0.4019, pruned_loss=0.1673, over 24000.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3869, pruned_loss=0.1357, over 5633229.27 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3795, pruned_loss=0.1306, over 5689367.88 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3865, pruned_loss=0.1352, over 5640996.72 frames. ], batch size: 705, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:34:51,716 INFO [train.py:968] (1/2) Epoch 9, batch 28350, giga_loss[loss=0.3168, simple_loss=0.3886, pruned_loss=0.1225, over 28560.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3868, pruned_loss=0.134, over 5640994.25 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3798, pruned_loss=0.131, over 5682754.15 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3864, pruned_loss=0.1333, over 5651267.76 frames. ], batch size: 336, lr: 3.57e-03, grad_scale: 1.0
+2023-03-04 20:35:08,988 INFO [optim.py:369] (1/2) 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 20:35:40,420 INFO [train.py:968] (1/2) Epoch 9, batch 28400, giga_loss[loss=0.3065, simple_loss=0.3718, pruned_loss=0.1205, over 28914.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3862, pruned_loss=0.1338, over 5651856.71 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3795, pruned_loss=0.1308, over 5688336.92 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3863, pruned_loss=0.1335, over 5654047.41 frames. ], batch size: 213, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:35:56,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=392616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:36:00,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=392619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:36:27,510 INFO [train.py:968] (1/2) Epoch 9, batch 28450, giga_loss[loss=0.3477, simple_loss=0.3844, pruned_loss=0.1555, over 23721.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3841, pruned_loss=0.1331, over 5659639.14 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3794, pruned_loss=0.1308, over 5691814.73 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3844, pruned_loss=0.1329, over 5657512.01 frames. ], batch size: 705, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:36:28,013 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=392648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:36:28,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-04 20:36:28,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-04 20:36:45,021 INFO [optim.py:369] (1/2) 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 20:37:23,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4839, 1.7144, 1.2845, 1.2761], device='cuda:1'), covar=tensor([0.1840, 0.1497, 0.1462, 0.1589], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1525, 0.1484, 0.1593], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 20:37:28,654 INFO [train.py:968] (1/2) Epoch 9, batch 28500, giga_loss[loss=0.3635, simple_loss=0.4053, pruned_loss=0.1608, over 29010.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.383, pruned_loss=0.1325, over 5671521.24 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3792, pruned_loss=0.1307, over 5694082.65 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3834, pruned_loss=0.1325, over 5667659.26 frames. ], batch size: 213, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:37:33,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-04 20:38:16,313 INFO [train.py:968] (1/2) Epoch 9, batch 28550, giga_loss[loss=0.3578, simple_loss=0.3985, pruned_loss=0.1585, over 28880.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3808, pruned_loss=0.1311, over 5675348.91 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.379, pruned_loss=0.1305, over 5695865.75 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3814, pruned_loss=0.1313, over 5669889.32 frames. ], batch size: 99, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:38:31,204 INFO [optim.py:369] (1/2) 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 20:38:40,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=392774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:39:00,256 INFO [train.py:968] (1/2) Epoch 9, batch 28600, giga_loss[loss=0.2913, simple_loss=0.3671, pruned_loss=0.1078, over 28682.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3814, pruned_loss=0.132, over 5671504.56 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3793, pruned_loss=0.1307, over 5691543.69 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3816, pruned_loss=0.132, over 5670866.74 frames. ], batch size: 242, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:39:43,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4695, 2.1867, 1.9980, 1.8833], device='cuda:1'), covar=tensor([0.1164, 0.2323, 0.1910, 0.2086], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0744, 0.0665, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 20:39:46,256 INFO [train.py:968] (1/2) Epoch 9, batch 28650, giga_loss[loss=0.3265, simple_loss=0.3813, pruned_loss=0.1359, over 28818.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.381, pruned_loss=0.1323, over 5658891.41 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3788, pruned_loss=0.1303, over 5698104.74 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3818, pruned_loss=0.1327, over 5651369.83 frames. ], batch size: 99, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:40:02,615 INFO [optim.py:369] (1/2) 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 20:40:34,339 INFO [train.py:968] (1/2) Epoch 9, batch 28700, giga_loss[loss=0.3459, simple_loss=0.3932, pruned_loss=0.1494, over 27550.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3815, pruned_loss=0.1329, over 5645095.96 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3786, pruned_loss=0.1301, over 5687732.56 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3823, pruned_loss=0.1335, over 5647200.05 frames. ], batch size: 472, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:40:51,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=392917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:41:21,028 INFO [train.py:968] (1/2) Epoch 9, batch 28750, giga_loss[loss=0.3505, simple_loss=0.4043, pruned_loss=0.1483, over 28943.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3849, pruned_loss=0.1362, over 5647361.89 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3787, pruned_loss=0.1302, over 5685811.06 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3856, pruned_loss=0.1366, over 5650305.53 frames. ], batch size: 186, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:41:35,560 INFO [optim.py:369] (1/2) 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 20:42:10,619 INFO [train.py:968] (1/2) Epoch 9, batch 28800, giga_loss[loss=0.3402, simple_loss=0.3978, pruned_loss=0.1413, over 28722.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3845, pruned_loss=0.136, over 5645294.08 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3781, pruned_loss=0.1299, over 5691579.58 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3857, pruned_loss=0.1368, over 5641663.03 frames. ], batch size: 262, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:42:57,373 INFO [train.py:968] (1/2) Epoch 9, batch 28850, libri_loss[loss=0.3278, simple_loss=0.3875, pruned_loss=0.1341, over 19612.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3841, pruned_loss=0.1364, over 5636662.94 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3777, pruned_loss=0.1296, over 5683538.52 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3855, pruned_loss=0.1374, over 5640876.47 frames. ], batch size: 186, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:43:12,274 INFO [optim.py:369] (1/2) 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 20:43:42,952 INFO [train.py:968] (1/2) Epoch 9, batch 28900, giga_loss[loss=0.35, simple_loss=0.4021, pruned_loss=0.1489, over 28928.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3845, pruned_loss=0.1367, over 5643499.44 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3778, pruned_loss=0.1296, over 5684635.82 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3856, pruned_loss=0.1375, over 5645635.69 frames. ], batch size: 174, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:44:31,012 INFO [train.py:968] (1/2) Epoch 9, batch 28950, giga_loss[loss=0.3198, simple_loss=0.3782, pruned_loss=0.1307, over 28325.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3849, pruned_loss=0.1364, over 5639704.88 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3782, pruned_loss=0.1298, over 5688967.83 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3855, pruned_loss=0.137, over 5636479.35 frames. ], batch size: 368, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:44:32,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:44:45,271 INFO [optim.py:369] (1/2) 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 20:44:50,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3401, 3.5945, 2.5708, 1.1439], device='cuda:1'), covar=tensor([0.3748, 0.1395, 0.1940, 0.3921], device='cuda:1'), in_proj_covar=tensor([0.1535, 0.1467, 0.1479, 0.1244], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 20:44:56,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-04 20:45:10,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9968, 1.0261, 3.3382, 2.9065], device='cuda:1'), covar=tensor([0.1623, 0.2529, 0.0486, 0.0962], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0579, 0.0835, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 20:45:19,095 INFO [train.py:968] (1/2) Epoch 9, batch 29000, giga_loss[loss=0.3336, simple_loss=0.3722, pruned_loss=0.1475, over 23309.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3852, pruned_loss=0.1359, over 5647277.07 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3782, pruned_loss=0.1298, over 5691823.68 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3857, pruned_loss=0.1364, over 5641762.78 frames. ], batch size: 705, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:46:02,688 INFO [train.py:968] (1/2) Epoch 9, batch 29050, giga_loss[loss=0.3544, simple_loss=0.4034, pruned_loss=0.1527, over 28937.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.387, pruned_loss=0.1374, over 5642673.16 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3786, pruned_loss=0.1302, over 5677268.86 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3872, pruned_loss=0.1376, over 5650452.31 frames. ], batch size: 186, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:46:17,543 INFO [optim.py:369] (1/2) 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 20:46:42,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=393292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:42,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:42,808 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:44,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:46:46,534 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-04 20:46:46,716 INFO [train.py:968] (1/2) Epoch 9, batch 29100, giga_loss[loss=0.3112, simple_loss=0.3769, pruned_loss=0.1228, over 28801.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3873, pruned_loss=0.1374, over 5662659.34 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3787, pruned_loss=0.1303, over 5680879.36 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3875, pruned_loss=0.1376, over 5665170.05 frames. ], batch size: 119, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:46:52,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3407, 3.8141, 1.4237, 1.5286], device='cuda:1'), covar=tensor([0.0919, 0.0354, 0.0888, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0510, 0.0333, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 20:47:09,928 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=393324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:47:19,246 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=393334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:47:30,683 INFO [train.py:968] (1/2) Epoch 9, batch 29150, giga_loss[loss=0.3553, simple_loss=0.41, pruned_loss=0.1503, over 28276.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3874, pruned_loss=0.1376, over 5670194.30 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3786, pruned_loss=0.1301, over 5687934.03 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3881, pruned_loss=0.1382, over 5665444.20 frames. ], batch size: 369, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:47:43,691 INFO [optim.py:369] (1/2) 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 20:47:44,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2749, 1.1124, 4.3612, 3.3207], device='cuda:1'), covar=tensor([0.1691, 0.2743, 0.0373, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0636, 0.0577, 0.0833, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 20:47:58,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8180, 1.7672, 1.2896, 1.4453], device='cuda:1'), covar=tensor([0.0690, 0.0626, 0.0981, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0441, 0.0497, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 20:48:17,236 INFO [train.py:968] (1/2) Epoch 9, batch 29200, giga_loss[loss=0.3007, simple_loss=0.3742, pruned_loss=0.1136, over 28779.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.388, pruned_loss=0.1371, over 5670662.48 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.378, pruned_loss=0.1298, over 5693068.78 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3893, pruned_loss=0.1381, over 5662018.76 frames. ], batch size: 99, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:48:20,698 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 20:48:28,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-04 20:48:52,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:48:57,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:49:07,608 INFO [train.py:968] (1/2) Epoch 9, batch 29250, giga_loss[loss=0.3009, simple_loss=0.3705, pruned_loss=0.1157, over 28781.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3871, pruned_loss=0.1354, over 5664337.81 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3779, pruned_loss=0.1295, over 5699120.07 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3884, pruned_loss=0.1366, over 5651404.18 frames. ], batch size: 285, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:49:18,910 INFO [optim.py:369] (1/2) 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 20:49:19,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3869, 1.7085, 1.3184, 1.7920], device='cuda:1'), covar=tensor([0.2375, 0.2242, 0.2474, 0.2063], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.0930, 0.1104, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 20:49:22,569 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 29300, giga_loss[loss=0.352, simple_loss=0.3802, pruned_loss=0.1619, over 23885.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3856, pruned_loss=0.1341, over 5672780.89 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3779, pruned_loss=0.1296, over 5700175.41 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3867, pruned_loss=0.135, over 5661305.44 frames. ], batch size: 705, lr: 3.57e-03, grad_scale: 8.0
+2023-03-04 20:50:00,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1856, 1.3164, 3.3187, 2.9944], device='cuda:1'), covar=tensor([0.1434, 0.2295, 0.0455, 0.1447], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0571, 0.0824, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 20:50:12,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=393524.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:50:32,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7071, 2.0102, 1.6119, 1.6279], device='cuda:1'), covar=tensor([0.1640, 0.1226, 0.1081, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.1622, 0.1512, 0.1465, 0.1578], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 20:50:32,717 INFO [train.py:968] (1/2) Epoch 9, batch 29350, libri_loss[loss=0.2894, simple_loss=0.3489, pruned_loss=0.115, over 29567.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3857, pruned_loss=0.1349, over 5667803.92 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3787, pruned_loss=0.1303, over 5707896.84 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3862, pruned_loss=0.1351, over 5649976.31 frames. ], batch size: 75, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:50:43,910 INFO [optim.py:369] (1/2) 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 20:51:15,868 INFO [train.py:968] (1/2) Epoch 9, batch 29400, giga_loss[loss=0.305, simple_loss=0.3745, pruned_loss=0.1177, over 28916.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3855, pruned_loss=0.1341, over 5677183.44 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3786, pruned_loss=0.1304, over 5710818.95 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3863, pruned_loss=0.1344, over 5658975.00 frames. ], batch size: 227, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:52:08,155 INFO [train.py:968] (1/2) Epoch 9, batch 29450, giga_loss[loss=0.3213, simple_loss=0.3772, pruned_loss=0.1327, over 28733.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3877, pruned_loss=0.1361, over 5666549.65 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3786, pruned_loss=0.1304, over 5710818.95 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3883, pruned_loss=0.1363, over 5652377.80 frames. ], batch size: 99, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:52:22,463 INFO [optim.py:369] (1/2) 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 20:52:24,398 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 29500, giga_loss[loss=0.2908, simple_loss=0.3612, pruned_loss=0.1102, over 29035.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3873, pruned_loss=0.1372, over 5660573.02 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3789, pruned_loss=0.1307, over 5703667.98 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3876, pruned_loss=0.1372, over 5654693.36 frames. ], batch size: 164, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:53:05,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:53:41,554 INFO [train.py:968] (1/2) Epoch 9, batch 29550, giga_loss[loss=0.3186, simple_loss=0.3851, pruned_loss=0.1261, over 28753.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3887, pruned_loss=0.139, over 5651642.23 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3789, pruned_loss=0.1308, over 5703742.01 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3891, pruned_loss=0.139, over 5646618.15 frames. ], batch size: 119, lr: 3.57e-03, grad_scale: 2.0
+2023-03-04 20:53:57,801 INFO [optim.py:369] (1/2) 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 20:54:25,453 INFO [train.py:968] (1/2) Epoch 9, batch 29600, giga_loss[loss=0.3606, simple_loss=0.4116, pruned_loss=0.1548, over 28733.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.389, pruned_loss=0.1389, over 5662223.54 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3782, pruned_loss=0.1305, over 5705368.38 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3901, pruned_loss=0.1394, over 5655591.77 frames. ], batch size: 262, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:54:25,691 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4305, 1.3724, 4.9624, 3.5867], device='cuda:1'), covar=tensor([0.1788, 0.2551, 0.0348, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0576, 0.0829, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 20:54:36,407 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:54:38,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:04,448 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 29650, libri_loss[loss=0.2547, simple_loss=0.3232, pruned_loss=0.09313, over 29476.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3885, pruned_loss=0.1383, over 5652132.77 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3779, pruned_loss=0.13, over 5708890.94 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3901, pruned_loss=0.1395, over 5641395.02 frames. ], batch size: 70, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:55:13,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=393852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:16,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5224, 4.3298, 4.0958, 1.8805], device='cuda:1'), covar=tensor([0.0496, 0.0627, 0.0672, 0.2001], device='cuda:1'), in_proj_covar=tensor([0.1011, 0.0956, 0.0838, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-04 20:55:16,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=393855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:55:24,416 INFO [optim.py:369] (1/2) 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 20:55:40,839 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 29700, giga_loss[loss=0.2991, simple_loss=0.3657, pruned_loss=0.1163, over 28912.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3866, pruned_loss=0.1359, over 5674383.04 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3776, pruned_loss=0.1297, over 5712808.65 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3884, pruned_loss=0.1373, over 5661534.20 frames. ], batch size: 213, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:55:54,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=393899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 20:56:15,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 20:56:40,558 INFO [train.py:968] (1/2) Epoch 9, batch 29750, giga_loss[loss=0.3496, simple_loss=0.4082, pruned_loss=0.1454, over 28942.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3868, pruned_loss=0.1359, over 5667995.71 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1296, over 5711360.68 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3885, pruned_loss=0.1371, over 5658636.55 frames. ], batch size: 186, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:56:57,943 INFO [optim.py:369] (1/2) 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 20:57:24,927 INFO [train.py:968] (1/2) Epoch 9, batch 29800, giga_loss[loss=0.3082, simple_loss=0.3796, pruned_loss=0.1184, over 28915.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3865, pruned_loss=0.1353, over 5661063.45 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3782, pruned_loss=0.1301, over 5700975.02 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3874, pruned_loss=0.136, over 5660631.49 frames. ], batch size: 136, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:58:00,669 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=394042.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:58:03,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=394045.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:58:06,794 INFO [train.py:968] (1/2) Epoch 9, batch 29850, giga_loss[loss=0.3003, simple_loss=0.3644, pruned_loss=0.1181, over 28935.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3848, pruned_loss=0.1344, over 5655343.23 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3776, pruned_loss=0.1298, over 5696471.28 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3864, pruned_loss=0.1355, over 5656733.11 frames. ], batch size: 227, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:58:22,816 INFO [optim.py:369] (1/2) 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 20:58:29,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=394074.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 20:58:50,285 INFO [train.py:968] (1/2) Epoch 9, batch 29900, giga_loss[loss=0.3015, simple_loss=0.3646, pruned_loss=0.1192, over 28981.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3831, pruned_loss=0.1335, over 5660298.42 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3777, pruned_loss=0.1298, over 5696867.17 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3845, pruned_loss=0.1344, over 5660274.42 frames. ], batch size: 213, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:58:54,833 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-04 20:59:34,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=394145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 20:59:35,642 INFO [train.py:968] (1/2) Epoch 9, batch 29950, giga_loss[loss=0.3236, simple_loss=0.3762, pruned_loss=0.1355, over 28619.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3818, pruned_loss=0.133, over 5651017.27 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3783, pruned_loss=0.1303, over 5687943.90 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3825, pruned_loss=0.1334, over 5657796.06 frames. ], batch size: 307, lr: 3.57e-03, grad_scale: 4.0
+2023-03-04 20:59:51,452 INFO [optim.py:369] (1/2) 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 21:00:22,003 INFO [train.py:968] (1/2) Epoch 9, batch 30000, giga_loss[loss=0.3302, simple_loss=0.382, pruned_loss=0.1392, over 28756.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3779, pruned_loss=0.1308, over 5665067.68 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3785, pruned_loss=0.1305, over 5691613.58 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3784, pruned_loss=0.131, over 5666650.95 frames. ], batch size: 262, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:00:22,003 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 21:00:25,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0243, 1.2353, 3.4098, 3.0766], device='cuda:1'), covar=tensor([0.1952, 0.2769, 0.0500, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0636, 0.0573, 0.0828, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 21:00:30,274 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 21:00:51,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4181, 2.3553, 1.9050, 1.5004], device='cuda:1'), covar=tensor([0.0804, 0.0223, 0.0257, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0116, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0082], device='cuda:1')
+2023-03-04 21:00:58,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-04 21:01:14,696 INFO [train.py:968] (1/2) Epoch 9, batch 30050, giga_loss[loss=0.2772, simple_loss=0.3408, pruned_loss=0.1068, over 28494.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3765, pruned_loss=0.1305, over 5680650.90 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3783, pruned_loss=0.1303, over 5694358.13 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3769, pruned_loss=0.1309, over 5679271.36 frames. ], batch size: 71, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:01:33,138 INFO [optim.py:369] (1/2) 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 21:02:02,385 INFO [train.py:968] (1/2) Epoch 9, batch 30100, giga_loss[loss=0.328, simple_loss=0.3896, pruned_loss=0.1332, over 28946.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3768, pruned_loss=0.1309, over 5692067.19 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3782, pruned_loss=0.1302, over 5698449.43 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3772, pruned_loss=0.1313, over 5687223.73 frames. ], batch size: 174, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:02:19,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5423, 1.5164, 1.3076, 1.1881], device='cuda:1'), covar=tensor([0.0718, 0.0543, 0.0960, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0441, 0.0498, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 21:02:49,269 INFO [train.py:968] (1/2) Epoch 9, batch 30150, giga_loss[loss=0.2684, simple_loss=0.3445, pruned_loss=0.09616, over 28006.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3766, pruned_loss=0.1293, over 5683306.69 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3785, pruned_loss=0.1304, over 5698605.79 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3766, pruned_loss=0.1294, over 5678913.36 frames. ], batch size: 412, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:03:08,629 INFO [optim.py:369] (1/2) 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 21:03:26,772 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-04 21:03:33,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-04 21:03:42,471 INFO [train.py:968] (1/2) Epoch 9, batch 30200, giga_loss[loss=0.2822, simple_loss=0.3458, pruned_loss=0.1094, over 26704.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3736, pruned_loss=0.1252, over 5666042.64 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3784, pruned_loss=0.1305, over 5685523.64 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3736, pruned_loss=0.1251, over 5674186.22 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:04:11,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-04 21:04:27,196 INFO [train.py:968] (1/2) Epoch 9, batch 30250, giga_loss[loss=0.2642, simple_loss=0.3457, pruned_loss=0.09137, over 28988.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3703, pruned_loss=0.122, over 5662921.08 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3781, pruned_loss=0.1304, over 5689632.76 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5664502.02 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:04:39,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-04 21:04:43,169 INFO [optim.py:369] (1/2) 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 21:05:13,513 INFO [train.py:968] (1/2) Epoch 9, batch 30300, giga_loss[loss=0.2534, simple_loss=0.335, pruned_loss=0.08592, over 29038.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3668, pruned_loss=0.119, over 5645667.33 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3778, pruned_loss=0.1305, over 5682980.91 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3668, pruned_loss=0.1184, over 5651832.77 frames. ], batch size: 113, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:05:34,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=394520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:06:02,791 INFO [train.py:968] (1/2) Epoch 9, batch 30350, giga_loss[loss=0.269, simple_loss=0.3403, pruned_loss=0.09882, over 27611.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3628, pruned_loss=0.1147, over 5648737.15 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3777, pruned_loss=0.1304, over 5683924.39 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3628, pruned_loss=0.1142, over 5652443.56 frames. ], batch size: 472, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:06:19,833 INFO [optim.py:369] (1/2) 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 21:06:32,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=394579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:06:52,312 INFO [train.py:968] (1/2) Epoch 9, batch 30400, giga_loss[loss=0.2555, simple_loss=0.3461, pruned_loss=0.08247, over 28987.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.36, pruned_loss=0.1105, over 5635734.22 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3775, pruned_loss=0.1304, over 5677740.10 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3599, pruned_loss=0.1098, over 5642753.85 frames. ], batch size: 164, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:07:43,679 INFO [train.py:968] (1/2) Epoch 9, batch 30450, giga_loss[loss=0.2805, simple_loss=0.3568, pruned_loss=0.1021, over 28909.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3602, pruned_loss=0.1105, over 5640745.27 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3768, pruned_loss=0.1301, over 5682325.73 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3605, pruned_loss=0.1099, over 5641206.45 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:07:59,911 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=394663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:08:01,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5488, 1.5826, 1.1964, 1.2792], device='cuda:1'), covar=tensor([0.0636, 0.0419, 0.0881, 0.0976], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0438, 0.0498, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 21:08:03,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=394666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:08:04,109 INFO [optim.py:369] (1/2) 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 21:08:06,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 21:08:21,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-04 21:08:32,265 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 30500, giga_loss[loss=0.2603, simple_loss=0.339, pruned_loss=0.09078, over 28431.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3592, pruned_loss=0.1099, over 5636335.63 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3768, pruned_loss=0.1301, over 5684518.85 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3593, pruned_loss=0.1092, over 5634319.88 frames. ], batch size: 85, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:08:44,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3452, 1.8323, 1.3317, 1.5651], device='cuda:1'), covar=tensor([0.0719, 0.0316, 0.0330, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:1')
+2023-03-04 21:09:23,817 INFO [train.py:968] (1/2) Epoch 9, batch 30550, giga_loss[loss=0.2573, simple_loss=0.3377, pruned_loss=0.08848, over 28724.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3557, pruned_loss=0.1074, over 5641533.52 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3767, pruned_loss=0.1302, over 5689375.07 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3554, pruned_loss=0.1063, over 5634802.75 frames. ], batch size: 284, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:09:41,454 INFO [optim.py:369] (1/2) 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 21:10:06,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-04 21:10:09,101 INFO [train.py:968] (1/2) Epoch 9, batch 30600, giga_loss[loss=0.2579, simple_loss=0.3374, pruned_loss=0.08918, over 28679.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3538, pruned_loss=0.1063, over 5650659.41 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3755, pruned_loss=0.1296, over 5697456.51 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3538, pruned_loss=0.1052, over 5636151.83 frames. ], batch size: 262, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:10:54,714 INFO [train.py:968] (1/2) Epoch 9, batch 30650, giga_loss[loss=0.3683, simple_loss=0.4007, pruned_loss=0.1679, over 26776.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3546, pruned_loss=0.1068, over 5639818.74 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3755, pruned_loss=0.1298, over 5688549.63 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3542, pruned_loss=0.1051, over 5635567.18 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:11:15,134 INFO [optim.py:369] (1/2) 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 21:11:43,413 INFO [train.py:968] (1/2) Epoch 9, batch 30700, giga_loss[loss=0.2572, simple_loss=0.337, pruned_loss=0.08868, over 29010.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3533, pruned_loss=0.1053, over 5648314.83 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3752, pruned_loss=0.1296, over 5690978.49 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.353, pruned_loss=0.1038, over 5642198.69 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:12:31,088 INFO [train.py:968] (1/2) Epoch 9, batch 30750, giga_loss[loss=0.2713, simple_loss=0.3492, pruned_loss=0.09668, over 28926.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3504, pruned_loss=0.1029, over 5640013.36 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3748, pruned_loss=0.1296, over 5676395.25 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3499, pruned_loss=0.1011, over 5647283.11 frames. ], batch size: 213, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:12:37,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=394954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:12:49,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3253, 4.1046, 3.9290, 1.8880], device='cuda:1'), covar=tensor([0.0676, 0.0888, 0.0972, 0.2005], device='cuda:1'), in_proj_covar=tensor([0.0985, 0.0929, 0.0810, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 21:12:49,476 INFO [optim.py:369] (1/2) 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 21:13:01,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=394981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:13:16,801 INFO [train.py:968] (1/2) Epoch 9, batch 30800, giga_loss[loss=0.2496, simple_loss=0.3198, pruned_loss=0.08972, over 27577.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3454, pruned_loss=0.09994, over 5639282.01 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3741, pruned_loss=0.129, over 5685599.89 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3448, pruned_loss=0.09802, over 5635461.04 frames. ], batch size: 472, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:14:03,651 INFO [train.py:968] (1/2) Epoch 9, batch 30850, libri_loss[loss=0.2288, simple_loss=0.2898, pruned_loss=0.08385, over 29500.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3442, pruned_loss=0.09982, over 5648261.70 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3728, pruned_loss=0.1284, over 5688514.06 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3441, pruned_loss=0.09819, over 5641503.87 frames. ], batch size: 70, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:14:04,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 21:14:22,666 INFO [optim.py:369] (1/2) 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 21:14:50,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:14:50,975 INFO [train.py:968] (1/2) Epoch 9, batch 30900, giga_loss[loss=0.2276, simple_loss=0.3089, pruned_loss=0.07312, over 28896.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3426, pruned_loss=0.09965, over 5635192.12 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3721, pruned_loss=0.1279, over 5684503.46 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3424, pruned_loss=0.09798, over 5632884.08 frames. ], batch size: 145, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:14:54,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:15:27,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=395129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:15:46,085 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 30950, giga_loss[loss=0.3119, simple_loss=0.3757, pruned_loss=0.1241, over 27728.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3434, pruned_loss=0.09969, over 5627334.92 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.372, pruned_loss=0.1278, over 5686971.70 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3431, pruned_loss=0.09817, over 5622491.77 frames. ], batch size: 472, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:16:10,980 INFO [optim.py:369] (1/2) 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 21:16:29,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 21:16:40,283 INFO [train.py:968] (1/2) Epoch 9, batch 31000, giga_loss[loss=0.3102, simple_loss=0.368, pruned_loss=0.1261, over 26773.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3466, pruned_loss=0.1003, over 5632094.75 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3717, pruned_loss=0.1278, over 5683018.62 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3459, pruned_loss=0.09847, over 5629414.62 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:17:41,388 INFO [train.py:968] (1/2) Epoch 9, batch 31050, giga_loss[loss=0.2489, simple_loss=0.3306, pruned_loss=0.08353, over 28915.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3474, pruned_loss=0.1001, over 5640039.66 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3719, pruned_loss=0.1281, over 5672798.22 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3463, pruned_loss=0.09796, over 5646238.84 frames. ], batch size: 284, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:17:56,615 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:18:36,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3248, 2.9345, 1.4471, 1.4163], device='cuda:1'), covar=tensor([0.0825, 0.0446, 0.0801, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0503, 0.0334, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 21:18:37,426 INFO [train.py:968] (1/2) Epoch 9, batch 31100, libri_loss[loss=0.3499, simple_loss=0.3954, pruned_loss=0.1523, over 26166.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3471, pruned_loss=0.1005, over 5637087.83 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3718, pruned_loss=0.1282, over 5658313.46 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3452, pruned_loss=0.09734, over 5653084.02 frames. ], batch size: 136, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:19:34,915 INFO [train.py:968] (1/2) Epoch 9, batch 31150, giga_loss[loss=0.282, simple_loss=0.3482, pruned_loss=0.1079, over 26879.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3444, pruned_loss=0.09824, over 5641841.56 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3713, pruned_loss=0.1279, over 5662247.61 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.343, pruned_loss=0.09556, over 5650457.24 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 2.0
+2023-03-04 21:19:46,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=395356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:20:06,423 INFO [optim.py:369] (1/2) 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 21:20:37,368 INFO [train.py:968] (1/2) Epoch 9, batch 31200, giga_loss[loss=0.3169, simple_loss=0.3766, pruned_loss=0.1286, over 28992.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3437, pruned_loss=0.0968, over 5641480.81 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3712, pruned_loss=0.1279, over 5656378.70 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3421, pruned_loss=0.09413, over 5653710.82 frames. ], batch size: 285, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:21:35,280 INFO [train.py:968] (1/2) Epoch 9, batch 31250, libri_loss[loss=0.3114, simple_loss=0.3721, pruned_loss=0.1254, over 29380.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3402, pruned_loss=0.09557, over 5654018.53 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3706, pruned_loss=0.1278, over 5662558.71 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3387, pruned_loss=0.09283, over 5657839.59 frames. ], batch size: 92, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:21:48,964 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=395461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:21:55,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-04 21:21:58,956 INFO [optim.py:369] (1/2) 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 21:22:24,859 INFO [train.py:968] (1/2) Epoch 9, batch 31300, giga_loss[loss=0.2461, simple_loss=0.3266, pruned_loss=0.08282, over 28196.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3399, pruned_loss=0.09658, over 5662420.14 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3694, pruned_loss=0.1272, over 5673772.28 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3382, pruned_loss=0.09308, over 5654687.27 frames. ], batch size: 412, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:22:26,088 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:22:29,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:22:33,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6232, 1.8361, 1.5384, 1.4934], device='cuda:1'), covar=tensor([0.1472, 0.1163, 0.1044, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1478, 0.1433, 0.1543], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 21:22:39,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1966, 1.3961, 1.4394, 1.2817], device='cuda:1'), covar=tensor([0.1045, 0.1080, 0.1426, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0715, 0.0646, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 21:22:53,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=395522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:23:04,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=395531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:23:22,295 INFO [train.py:968] (1/2) Epoch 9, batch 31350, giga_loss[loss=0.2709, simple_loss=0.3522, pruned_loss=0.09476, over 28930.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3392, pruned_loss=0.09612, over 5665503.45 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3694, pruned_loss=0.1273, over 5673158.93 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3374, pruned_loss=0.09287, over 5659480.25 frames. ], batch size: 186, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:23:45,267 INFO [optim.py:369] (1/2) 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 21:23:55,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8589, 2.5832, 1.6457, 1.0149], device='cuda:1'), covar=tensor([0.4681, 0.2380, 0.2891, 0.3942], device='cuda:1'), in_proj_covar=tensor([0.1501, 0.1426, 0.1460, 0.1229], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 21:24:13,420 INFO [train.py:968] (1/2) Epoch 9, batch 31400, giga_loss[loss=0.256, simple_loss=0.338, pruned_loss=0.08697, over 28362.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3413, pruned_loss=0.09716, over 5660207.26 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3689, pruned_loss=0.1271, over 5672315.37 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3393, pruned_loss=0.09359, over 5656541.70 frames. ], batch size: 368, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:24:59,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=395636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:25:14,561 INFO [train.py:968] (1/2) Epoch 9, batch 31450, giga_loss[loss=0.2532, simple_loss=0.3351, pruned_loss=0.08567, over 28837.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.343, pruned_loss=0.09726, over 5668481.08 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3685, pruned_loss=0.1269, over 5676923.35 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3412, pruned_loss=0.09389, over 5661163.24 frames. ], batch size: 284, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:25:35,191 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:25:38,100 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:26:04,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4602, 3.8171, 1.5127, 1.6073], device='cuda:1'), covar=tensor([0.0841, 0.0310, 0.0860, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0499, 0.0334, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 21:26:09,693 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 31500, giga_loss[loss=0.252, simple_loss=0.3377, pruned_loss=0.08314, over 28942.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3393, pruned_loss=0.09539, over 5672008.84 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3673, pruned_loss=0.1263, over 5683224.29 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3378, pruned_loss=0.09189, over 5659717.92 frames. ], batch size: 155, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:27:15,874 INFO [train.py:968] (1/2) Epoch 9, batch 31550, giga_loss[loss=0.2668, simple_loss=0.3443, pruned_loss=0.09463, over 28692.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3406, pruned_loss=0.09622, over 5675905.99 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.367, pruned_loss=0.1262, over 5686344.81 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3393, pruned_loss=0.09316, over 5663269.29 frames. ], batch size: 307, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:27:40,967 INFO [optim.py:369] (1/2) 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 21:27:52,299 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:27:54,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:28:08,207 INFO [train.py:968] (1/2) Epoch 9, batch 31600, giga_loss[loss=0.2841, simple_loss=0.3802, pruned_loss=0.09397, over 28847.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.343, pruned_loss=0.09683, over 5671952.68 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.366, pruned_loss=0.1256, over 5684184.91 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3416, pruned_loss=0.09349, over 5662707.78 frames. ], batch size: 227, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:28:26,195 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=395811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:28:57,071 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 31650, giga_loss[loss=0.2557, simple_loss=0.3452, pruned_loss=0.08306, over 28216.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3456, pruned_loss=0.09544, over 5667726.49 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3659, pruned_loss=0.1255, over 5687681.65 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3444, pruned_loss=0.0925, over 5657039.46 frames. ], batch size: 412, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:29:35,352 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3898, 1.4925, 1.2335, 1.2667], device='cuda:1'), covar=tensor([0.1280, 0.1201, 0.1098, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1487, 0.1439, 0.1558], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 21:29:39,165 INFO [optim.py:369] (1/2) 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 21:30:11,716 INFO [train.py:968] (1/2) Epoch 9, batch 31700, giga_loss[loss=0.2848, simple_loss=0.3655, pruned_loss=0.102, over 28472.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3462, pruned_loss=0.09471, over 5660067.86 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3659, pruned_loss=0.1256, over 5689255.17 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3448, pruned_loss=0.09167, over 5649827.97 frames. ], batch size: 336, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:31:09,891 INFO [train.py:968] (1/2) Epoch 9, batch 31750, giga_loss[loss=0.2545, simple_loss=0.3417, pruned_loss=0.08362, over 28383.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3452, pruned_loss=0.09378, over 5662392.99 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3656, pruned_loss=0.1254, over 5695471.16 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3438, pruned_loss=0.09057, over 5647739.20 frames. ], batch size: 336, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:31:34,119 INFO [optim.py:369] (1/2) 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 21:31:41,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1804, 0.8381, 0.8165, 1.3810], device='cuda:1'), covar=tensor([0.0751, 0.0344, 0.0349, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0117, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-04 21:31:43,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=395979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:31:46,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=395982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:32:04,770 INFO [train.py:968] (1/2) Epoch 9, batch 31800, giga_loss[loss=0.2858, simple_loss=0.3388, pruned_loss=0.1165, over 24501.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3467, pruned_loss=0.09565, over 5657603.09 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3652, pruned_loss=0.1253, over 5692627.40 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3451, pruned_loss=0.09191, over 5647461.62 frames. ], batch size: 705, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:32:18,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:33:05,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9625, 1.2312, 1.2634, 1.0353], device='cuda:1'), covar=tensor([0.1160, 0.1109, 0.1618, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0719, 0.0646, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 21:33:06,081 INFO [train.py:968] (1/2) Epoch 9, batch 31850, giga_loss[loss=0.2841, simple_loss=0.3571, pruned_loss=0.1055, over 28686.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3468, pruned_loss=0.09697, over 5640884.84 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3653, pruned_loss=0.1255, over 5668496.60 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.345, pruned_loss=0.09316, over 5654509.94 frames. ], batch size: 307, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:33:39,850 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 31900, giga_loss[loss=0.2778, simple_loss=0.3505, pruned_loss=0.1026, over 28765.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3473, pruned_loss=0.09762, over 5649323.49 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3646, pruned_loss=0.1251, over 5664182.47 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3461, pruned_loss=0.09431, over 5662945.86 frames. ], batch size: 284, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:35:32,841 INFO [train.py:968] (1/2) Epoch 9, batch 31950, giga_loss[loss=0.2884, simple_loss=0.3431, pruned_loss=0.1168, over 26970.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3444, pruned_loss=0.09643, over 5654571.67 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3641, pruned_loss=0.1248, over 5664060.86 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3435, pruned_loss=0.09346, over 5665351.10 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:35:39,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:35:41,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9473, 1.9700, 1.5059, 1.5624], device='cuda:1'), covar=tensor([0.0695, 0.0587, 0.0928, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0437, 0.0500, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 21:35:45,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-04 21:35:48,343 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:36:01,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4601, 1.8706, 1.8128, 1.3182], device='cuda:1'), covar=tensor([0.1789, 0.2128, 0.1394, 0.1579], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0690, 0.0834, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 21:36:10,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:36:26,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2924, 1.5156, 1.2334, 1.0458], device='cuda:1'), covar=tensor([0.2442, 0.2196, 0.2585, 0.1808], device='cuda:1'), in_proj_covar=tensor([0.1248, 0.0923, 0.1107, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 21:36:35,714 INFO [train.py:968] (1/2) Epoch 9, batch 32000, giga_loss[loss=0.3438, simple_loss=0.3862, pruned_loss=0.1507, over 26850.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3428, pruned_loss=0.09589, over 5657729.73 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3638, pruned_loss=0.1246, over 5670037.67 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3418, pruned_loss=0.09284, over 5660817.51 frames. ], batch size: 555, lr: 3.56e-03, grad_scale: 8.0
+2023-03-04 21:37:33,749 INFO [train.py:968] (1/2) Epoch 9, batch 32050, giga_loss[loss=0.2906, simple_loss=0.3621, pruned_loss=0.1095, over 28885.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3397, pruned_loss=0.09449, over 5666522.73 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3627, pruned_loss=0.124, over 5675076.21 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3392, pruned_loss=0.09169, over 5664153.63 frames. ], batch size: 174, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:37:50,534 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:38:09,517 INFO [optim.py:369] (1/2) 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 21:38:18,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:38:24,705 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:38:38,994 INFO [train.py:968] (1/2) Epoch 9, batch 32100, giga_loss[loss=0.2658, simple_loss=0.3509, pruned_loss=0.09032, over 28078.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3431, pruned_loss=0.09655, over 5671211.56 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3625, pruned_loss=0.1239, over 5680476.57 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3424, pruned_loss=0.09367, over 5663977.90 frames. ], batch size: 412, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:39:15,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3733, 1.6298, 1.3752, 1.4189], device='cuda:1'), covar=tensor([0.1944, 0.1812, 0.1903, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.1247, 0.0922, 0.1106, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 21:39:37,808 INFO [train.py:968] (1/2) Epoch 9, batch 32150, giga_loss[loss=0.2355, simple_loss=0.3108, pruned_loss=0.08014, over 28887.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3456, pruned_loss=0.0979, over 5672466.74 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3624, pruned_loss=0.124, over 5685098.26 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3448, pruned_loss=0.09509, over 5662568.19 frames. ], batch size: 213, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:40:05,018 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 32200, giga_loss[loss=0.2478, simple_loss=0.3242, pruned_loss=0.0857, over 28907.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3434, pruned_loss=0.09783, over 5673944.20 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3617, pruned_loss=0.1236, over 5689808.20 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3426, pruned_loss=0.09477, over 5660935.71 frames. ], batch size: 227, lr: 3.56e-03, grad_scale: 4.0
+2023-03-04 21:40:40,479 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:41:16,451 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=396429.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 21:41:41,354 INFO [train.py:968] (1/2) Epoch 9, batch 32250, giga_loss[loss=0.2906, simple_loss=0.3669, pruned_loss=0.1071, over 27994.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3444, pruned_loss=0.0987, over 5672236.84 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3617, pruned_loss=0.1236, over 5690909.16 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3436, pruned_loss=0.0962, over 5661082.35 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:42:07,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5966, 1.8010, 1.5984, 1.6011], device='cuda:1'), covar=tensor([0.1225, 0.1767, 0.1689, 0.1662], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0717, 0.0649, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 21:42:13,888 INFO [optim.py:369] (1/2) 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 21:42:50,707 INFO [train.py:968] (1/2) Epoch 9, batch 32300, giga_loss[loss=0.2496, simple_loss=0.3338, pruned_loss=0.08274, over 28652.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3455, pruned_loss=0.09883, over 5664487.49 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3617, pruned_loss=0.1237, over 5683007.92 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3448, pruned_loss=0.09655, over 5661896.15 frames. ], batch size: 307, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:43:23,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3746, 1.7211, 1.6960, 1.2436], device='cuda:1'), covar=tensor([0.1574, 0.2148, 0.1245, 0.1483], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0688, 0.0832, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 21:43:27,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:43:37,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:43:40,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4105, 1.6122, 1.2860, 1.4709], device='cuda:1'), covar=tensor([0.2601, 0.2370, 0.2673, 0.2274], device='cuda:1'), in_proj_covar=tensor([0.1247, 0.0922, 0.1104, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 21:43:46,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-04 21:43:52,877 INFO [train.py:968] (1/2) Epoch 9, batch 32350, giga_loss[loss=0.2583, simple_loss=0.3415, pruned_loss=0.08759, over 28965.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3472, pruned_loss=0.09918, over 5677409.69 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.361, pruned_loss=0.1232, over 5693153.79 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3465, pruned_loss=0.09646, over 5665052.42 frames. ], batch size: 227, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:43:57,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:44:30,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4381, 1.6835, 1.7527, 1.3671], device='cuda:1'), covar=tensor([0.1534, 0.1900, 0.1214, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0688, 0.0833, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 21:44:35,010 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 32400, giga_loss[loss=0.2567, simple_loss=0.3451, pruned_loss=0.0841, over 28139.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.09765, over 5670300.77 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3603, pruned_loss=0.1227, over 5689290.15 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3454, pruned_loss=0.09528, over 5663091.53 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:45:52,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:13,947 INFO [train.py:968] (1/2) Epoch 9, batch 32450, giga_loss[loss=0.248, simple_loss=0.3265, pruned_loss=0.08469, over 28725.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3415, pruned_loss=0.09576, over 5679656.63 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3602, pruned_loss=0.1227, over 5693556.60 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3411, pruned_loss=0.09346, over 5670078.46 frames. ], batch size: 262, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:46:22,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:30,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:44,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:46,019 INFO [optim.py:369] (1/2) 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 21:46:46,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:52,484 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:46:55,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:13,582 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:15,814 INFO [train.py:968] (1/2) Epoch 9, batch 32500, giga_loss[loss=0.2097, simple_loss=0.2867, pruned_loss=0.06633, over 28653.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3373, pruned_loss=0.09442, over 5672735.21 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3607, pruned_loss=0.1229, over 5692184.50 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3358, pruned_loss=0.09145, over 5665686.73 frames. ], batch size: 99, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:47:17,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:21,479 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:28,276 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:47:54,117 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:48:18,985 INFO [train.py:968] (1/2) Epoch 9, batch 32550, libri_loss[loss=0.3441, simple_loss=0.3926, pruned_loss=0.1478, over 29269.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3361, pruned_loss=0.09419, over 5659412.43 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3608, pruned_loss=0.123, over 5684034.08 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3342, pruned_loss=0.09108, over 5659778.21 frames. ], batch size: 94, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:48:45,795 INFO [optim.py:369] (1/2) 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 21:48:48,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:48:49,450 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:48:51,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:13,281 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:13,697 INFO [train.py:968] (1/2) Epoch 9, batch 32600, giga_loss[loss=0.2694, simple_loss=0.3383, pruned_loss=0.1002, over 28517.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3381, pruned_loss=0.09569, over 5647009.31 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3609, pruned_loss=0.1231, over 5674017.44 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.336, pruned_loss=0.09257, over 5654720.39 frames. ], batch size: 71, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 21:49:16,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:17,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0426, 2.9629, 2.1430, 1.0444], device='cuda:1'), covar=tensor([0.3816, 0.1883, 0.2153, 0.3740], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1447, 0.1472, 0.1242], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 21:49:18,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:20,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=396804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 21:49:21,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:26,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:33,498 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 21:49:51,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:49:57,414 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:50:11,689 INFO [train.py:968] (1/2) Epoch 9, batch 32650, libri_loss[loss=0.2708, simple_loss=0.3278, pruned_loss=0.1069, over 29557.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3374, pruned_loss=0.09585, over 5648231.91 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.361, pruned_loss=0.1236, over 5676472.08 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3349, pruned_loss=0.09193, over 5650826.25 frames. ], batch size: 78, lr: 3.55e-03, grad_scale: 1.0
+2023-03-04 21:50:12,159 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 21:50:38,262 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4898, 1.6803, 1.7995, 1.3736], device='cuda:1'), covar=tensor([0.1794, 0.2227, 0.1333, 0.1608], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0690, 0.0836, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 21:50:44,343 INFO [optim.py:369] (1/2) 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 21:51:11,964 INFO [train.py:968] (1/2) Epoch 9, batch 32700, giga_loss[loss=0.2268, simple_loss=0.3217, pruned_loss=0.06594, over 28824.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3356, pruned_loss=0.09354, over 5647501.17 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3612, pruned_loss=0.1238, over 5672107.02 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3329, pruned_loss=0.08959, over 5653474.76 frames. ], batch size: 174, lr: 3.55e-03, grad_scale: 1.0
+2023-03-04 21:51:22,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7366, 3.5587, 3.3494, 1.5514], device='cuda:1'), covar=tensor([0.0696, 0.0844, 0.0850, 0.2460], device='cuda:1'), in_proj_covar=tensor([0.0977, 0.0913, 0.0800, 0.0623], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 21:51:38,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:51:38,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4277, 1.6868, 1.3349, 1.6408], device='cuda:1'), covar=tensor([0.2471, 0.2259, 0.2555, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1239, 0.0916, 0.1102, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 21:51:43,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=396921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:52:14,710 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=396947.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 21:52:14,990 INFO [train.py:968] (1/2) Epoch 9, batch 32750, giga_loss[loss=0.2807, simple_loss=0.3549, pruned_loss=0.1032, over 28956.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3341, pruned_loss=0.09302, over 5656249.07 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3604, pruned_loss=0.1234, over 5677754.67 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.332, pruned_loss=0.08954, over 5655264.80 frames. ], batch size: 145, lr: 3.55e-03, grad_scale: 1.0
+2023-03-04 21:52:18,813 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:52:18,871 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:52:59,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=396979.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 21:53:20,885 INFO [train.py:968] (1/2) Epoch 9, batch 32800, giga_loss[loss=0.2264, simple_loss=0.3173, pruned_loss=0.06777, over 28900.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3337, pruned_loss=0.09264, over 5654240.83 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3608, pruned_loss=0.1238, over 5674027.44 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3309, pruned_loss=0.08864, over 5656685.81 frames. ], batch size: 145, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:54:30,654 INFO [train.py:968] (1/2) Epoch 9, batch 32850, giga_loss[loss=0.2165, simple_loss=0.2841, pruned_loss=0.07447, over 24247.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3335, pruned_loss=0.09219, over 5649292.51 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3602, pruned_loss=0.1233, over 5678443.80 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3313, pruned_loss=0.0889, over 5646833.92 frames. ], batch size: 705, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:55:02,090 INFO [optim.py:369] (1/2) 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 21:55:33,238 INFO [train.py:968] (1/2) Epoch 9, batch 32900, giga_loss[loss=0.2548, simple_loss=0.3262, pruned_loss=0.09169, over 28970.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3351, pruned_loss=0.09367, over 5654587.86 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3604, pruned_loss=0.1234, over 5680690.61 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3331, pruned_loss=0.09074, over 5650555.87 frames. ], batch size: 186, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:56:37,554 INFO [train.py:968] (1/2) Epoch 9, batch 32950, giga_loss[loss=0.2226, simple_loss=0.3023, pruned_loss=0.07146, over 27605.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3331, pruned_loss=0.09204, over 5658501.40 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3603, pruned_loss=0.1233, over 5683021.18 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3313, pruned_loss=0.0895, over 5653196.93 frames. ], batch size: 472, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:57:09,980 INFO [optim.py:369] (1/2) 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 21:57:37,712 INFO [train.py:968] (1/2) Epoch 9, batch 33000, giga_loss[loss=0.2407, simple_loss=0.337, pruned_loss=0.07222, over 28999.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3346, pruned_loss=0.09126, over 5652823.49 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3605, pruned_loss=0.1235, over 5673326.96 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3326, pruned_loss=0.08857, over 5656687.05 frames. ], batch size: 136, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:57:37,713 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 21:57:46,369 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 21:58:10,726 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=397219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 21:58:36,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4940, 1.7844, 1.7664, 1.3202], device='cuda:1'), covar=tensor([0.1794, 0.2145, 0.1357, 0.1624], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0693, 0.0836, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 21:58:48,002 INFO [train.py:968] (1/2) Epoch 9, batch 33050, giga_loss[loss=0.2458, simple_loss=0.3299, pruned_loss=0.08087, over 28115.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3375, pruned_loss=0.09183, over 5655136.03 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3603, pruned_loss=0.1233, over 5675648.73 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.336, pruned_loss=0.0896, over 5655933.80 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 21:59:22,091 INFO [optim.py:369] (1/2) 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 21:59:50,801 INFO [train.py:968] (1/2) Epoch 9, batch 33100, giga_loss[loss=0.2537, simple_loss=0.3368, pruned_loss=0.08535, over 28878.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3381, pruned_loss=0.09226, over 5647126.56 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3599, pruned_loss=0.1231, over 5679262.46 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3369, pruned_loss=0.09024, over 5644227.66 frames. ], batch size: 145, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:00:50,878 INFO [train.py:968] (1/2) Epoch 9, batch 33150, giga_loss[loss=0.2679, simple_loss=0.3476, pruned_loss=0.09411, over 28441.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3384, pruned_loss=0.09302, over 5660941.03 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3592, pruned_loss=0.1228, over 5685774.68 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3371, pruned_loss=0.09038, over 5651350.95 frames. ], batch size: 336, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:01:20,070 INFO [optim.py:369] (1/2) 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 22:01:46,796 INFO [train.py:968] (1/2) Epoch 9, batch 33200, giga_loss[loss=0.2503, simple_loss=0.3366, pruned_loss=0.082, over 28616.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.338, pruned_loss=0.09307, over 5656153.04 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3593, pruned_loss=0.1231, over 5678894.42 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3365, pruned_loss=0.09018, over 5653693.28 frames. ], batch size: 242, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:02:47,402 INFO [train.py:968] (1/2) Epoch 9, batch 33250, giga_loss[loss=0.2439, simple_loss=0.3199, pruned_loss=0.08401, over 28685.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3368, pruned_loss=0.09229, over 5641112.26 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3598, pruned_loss=0.1234, over 5662747.91 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3349, pruned_loss=0.08926, over 5652522.47 frames. ], batch size: 119, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:03:17,570 INFO [optim.py:369] (1/2) 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 22:03:41,438 INFO [train.py:968] (1/2) Epoch 9, batch 33300, giga_loss[loss=0.2523, simple_loss=0.3268, pruned_loss=0.0889, over 28842.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3355, pruned_loss=0.09295, over 5646004.77 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3594, pruned_loss=0.1233, over 5661024.70 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3331, pruned_loss=0.08917, over 5655596.04 frames. ], batch size: 112, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:04:39,808 INFO [train.py:968] (1/2) Epoch 9, batch 33350, giga_loss[loss=0.2401, simple_loss=0.3276, pruned_loss=0.07633, over 28970.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3367, pruned_loss=0.09338, over 5663178.10 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3589, pruned_loss=0.1231, over 5668652.19 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3346, pruned_loss=0.08978, over 5663933.53 frames. ], batch size: 199, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:05:15,564 INFO [optim.py:369] (1/2) 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 22:05:42,984 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 33400, giga_loss[loss=0.2808, simple_loss=0.3522, pruned_loss=0.1047, over 28077.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3391, pruned_loss=0.09448, over 5664446.35 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3589, pruned_loss=0.123, over 5671132.79 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3374, pruned_loss=0.09141, over 5662731.95 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:05:57,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2256, 1.4956, 1.4148, 1.3398], device='cuda:1'), covar=tensor([0.1129, 0.1292, 0.1608, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0712, 0.0644, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 22:06:02,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7880, 1.8649, 1.8689, 1.7865], device='cuda:1'), covar=tensor([0.0811, 0.0948, 0.1056, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0712, 0.0643, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 22:06:52,054 INFO [train.py:968] (1/2) Epoch 9, batch 33450, giga_loss[loss=0.2919, simple_loss=0.3668, pruned_loss=0.1085, over 28709.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3404, pruned_loss=0.09601, over 5647709.58 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3591, pruned_loss=0.1234, over 5656249.74 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3384, pruned_loss=0.09271, over 5658250.10 frames. ], batch size: 307, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:07:33,683 INFO [optim.py:369] (1/2) 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 22:07:54,437 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=397692.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 22:08:00,253 INFO [train.py:968] (1/2) Epoch 9, batch 33500, giga_loss[loss=0.2474, simple_loss=0.3307, pruned_loss=0.08206, over 28928.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3442, pruned_loss=0.09776, over 5648972.62 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.359, pruned_loss=0.1234, over 5650060.51 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3425, pruned_loss=0.09478, over 5663327.50 frames. ], batch size: 136, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:08:45,288 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=397737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:08:48,418 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 33550, giga_loss[loss=0.2976, simple_loss=0.3738, pruned_loss=0.1107, over 28073.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3455, pruned_loss=0.09737, over 5647922.20 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3588, pruned_loss=0.1233, over 5653958.86 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3441, pruned_loss=0.09475, over 5656029.61 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:09:02,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4468, 1.0974, 5.1595, 3.3687], device='cuda:1'), covar=tensor([0.1653, 0.2704, 0.0335, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0627, 0.0571, 0.0815, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 22:09:12,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8584, 3.6604, 3.4388, 1.7120], device='cuda:1'), covar=tensor([0.0666, 0.0798, 0.0843, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.0966, 0.0905, 0.0794, 0.0621], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 22:09:24,830 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:10:02,056 INFO [train.py:968] (1/2) Epoch 9, batch 33600, libri_loss[loss=0.3484, simple_loss=0.3933, pruned_loss=0.1518, over 25637.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.345, pruned_loss=0.09694, over 5651983.52 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3591, pruned_loss=0.1234, over 5658330.29 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3431, pruned_loss=0.09386, over 5654672.26 frames. ], batch size: 137, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:10:32,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3552, 1.6012, 1.2670, 1.5298], device='cuda:1'), covar=tensor([0.2425, 0.2182, 0.2561, 0.1986], device='cuda:1'), in_proj_covar=tensor([0.1238, 0.0916, 0.1102, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 22:10:33,984 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 22:10:47,672 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-04 22:11:10,548 INFO [train.py:968] (1/2) Epoch 9, batch 33650, giga_loss[loss=0.2803, simple_loss=0.3581, pruned_loss=0.1013, over 28941.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3435, pruned_loss=0.09641, over 5658455.38 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3594, pruned_loss=0.1237, over 5658235.60 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3414, pruned_loss=0.09302, over 5660663.26 frames. ], batch size: 145, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:11:14,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5570, 3.3954, 3.1821, 1.7010], device='cuda:1'), covar=tensor([0.0709, 0.0803, 0.0824, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.0967, 0.0903, 0.0797, 0.0620], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 22:11:27,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8753, 1.1409, 1.0283, 0.7025], device='cuda:1'), covar=tensor([0.1443, 0.1385, 0.0851, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.1653, 0.1495, 0.1441, 0.1571], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 22:11:46,564 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 33700, giga_loss[loss=0.2414, simple_loss=0.3306, pruned_loss=0.07616, over 29037.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3417, pruned_loss=0.09567, over 5652813.63 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3591, pruned_loss=0.1236, over 5654360.91 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3399, pruned_loss=0.09252, over 5657627.55 frames. ], batch size: 285, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:12:34,205 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-04 22:12:50,603 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=397925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:13:20,995 INFO [train.py:968] (1/2) Epoch 9, batch 33750, giga_loss[loss=0.2867, simple_loss=0.3543, pruned_loss=0.1095, over 28015.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3403, pruned_loss=0.0949, over 5648234.26 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3587, pruned_loss=0.1233, over 5654411.72 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.339, pruned_loss=0.0923, over 5652103.41 frames. ], batch size: 412, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:14:00,809 INFO [optim.py:369] (1/2) 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 22:14:06,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-04 22:14:28,472 INFO [train.py:968] (1/2) Epoch 9, batch 33800, giga_loss[loss=0.2682, simple_loss=0.345, pruned_loss=0.0957, over 28971.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3397, pruned_loss=0.09552, over 5652686.92 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3582, pruned_loss=0.1229, over 5658912.47 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3388, pruned_loss=0.09325, over 5651622.99 frames. ], batch size: 284, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:15:33,720 INFO [train.py:968] (1/2) Epoch 9, batch 33850, giga_loss[loss=0.2229, simple_loss=0.3073, pruned_loss=0.06923, over 29102.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3387, pruned_loss=0.09521, over 5646538.46 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3584, pruned_loss=0.123, over 5663888.48 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3375, pruned_loss=0.0928, over 5640861.85 frames. ], batch size: 113, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:15:56,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398067.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 22:16:05,889 INFO [optim.py:369] (1/2) 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 22:16:32,688 INFO [train.py:968] (1/2) Epoch 9, batch 33900, giga_loss[loss=0.2217, simple_loss=0.3099, pruned_loss=0.06673, over 28746.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.338, pruned_loss=0.09373, over 5657762.14 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3581, pruned_loss=0.1229, over 5665359.15 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3367, pruned_loss=0.09111, over 5651491.83 frames. ], batch size: 243, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:16:36,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4110, 1.6130, 1.2576, 1.8206], device='cuda:1'), covar=tensor([0.2489, 0.2349, 0.2663, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.1239, 0.0919, 0.1103, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 22:17:30,876 INFO [train.py:968] (1/2) Epoch 9, batch 33950, giga_loss[loss=0.2171, simple_loss=0.3118, pruned_loss=0.06125, over 28519.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3384, pruned_loss=0.09175, over 5672669.98 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3576, pruned_loss=0.1225, over 5668883.94 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3376, pruned_loss=0.08956, over 5664448.51 frames. ], batch size: 65, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:17:47,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7587, 1.8104, 1.3596, 1.4642], device='cuda:1'), covar=tensor([0.0766, 0.0621, 0.0984, 0.1048], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0434, 0.0498, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 22:18:04,314 INFO [optim.py:369] (1/2) 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 22:18:30,890 INFO [train.py:968] (1/2) Epoch 9, batch 34000, giga_loss[loss=0.234, simple_loss=0.3279, pruned_loss=0.07006, over 28897.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3401, pruned_loss=0.09124, over 5670238.13 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.357, pruned_loss=0.1223, over 5672427.91 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3397, pruned_loss=0.08932, over 5660634.54 frames. ], batch size: 136, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:18:42,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=398210.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 22:18:47,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=398213.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 22:19:22,690 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=398242.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 22:19:30,744 INFO [train.py:968] (1/2) Epoch 9, batch 34050, giga_loss[loss=0.2598, simple_loss=0.3476, pruned_loss=0.08596, over 29028.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3402, pruned_loss=0.09112, over 5668536.60 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.357, pruned_loss=0.1222, over 5675813.24 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3397, pruned_loss=0.08927, over 5657904.14 frames. ], batch size: 128, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:20:14,365 INFO [optim.py:369] (1/2) 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 22:20:36,385 INFO [train.py:968] (1/2) Epoch 9, batch 34100, giga_loss[loss=0.2847, simple_loss=0.3715, pruned_loss=0.09893, over 28906.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3409, pruned_loss=0.09208, over 5676717.40 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3573, pruned_loss=0.1223, over 5679816.10 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3396, pruned_loss=0.08924, over 5664390.29 frames. ], batch size: 164, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:20:39,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:21:15,245 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:21:40,424 INFO [train.py:968] (1/2) Epoch 9, batch 34150, giga_loss[loss=0.2908, simple_loss=0.3594, pruned_loss=0.1111, over 28946.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3406, pruned_loss=0.09205, over 5677535.54 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3566, pruned_loss=0.1218, over 5686272.35 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3397, pruned_loss=0.0894, over 5662026.98 frames. ], batch size: 199, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:22:20,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2874, 1.6267, 1.5828, 1.2025], device='cuda:1'), covar=tensor([0.1454, 0.2015, 0.1177, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0685, 0.0830, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0013, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 22:22:21,424 INFO [optim.py:369] (1/2) 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 22:22:47,537 INFO [train.py:968] (1/2) Epoch 9, batch 34200, giga_loss[loss=0.2467, simple_loss=0.3387, pruned_loss=0.07742, over 28368.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.341, pruned_loss=0.09197, over 5679809.21 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3564, pruned_loss=0.1217, over 5693119.64 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.34, pruned_loss=0.08909, over 5660970.70 frames. ], batch size: 369, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:23:51,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6494, 3.4418, 3.2149, 1.9848], device='cuda:1'), covar=tensor([0.0614, 0.0814, 0.0811, 0.1803], device='cuda:1'), in_proj_covar=tensor([0.0967, 0.0899, 0.0793, 0.0618], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 22:23:57,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=398443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:24:00,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=398446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:24:01,656 INFO [train.py:968] (1/2) Epoch 9, batch 34250, giga_loss[loss=0.2635, simple_loss=0.3537, pruned_loss=0.08664, over 28836.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3408, pruned_loss=0.09143, over 5672168.49 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3561, pruned_loss=0.1216, over 5696696.24 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.34, pruned_loss=0.08876, over 5653804.29 frames. ], batch size: 164, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:24:22,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6019, 1.7878, 1.6567, 1.4082], device='cuda:1'), covar=tensor([0.2066, 0.1401, 0.1079, 0.1464], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1471, 0.1416, 0.1553], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 22:24:37,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=398475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:24:40,600 INFO [optim.py:369] (1/2) 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 22:24:46,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1684, 4.9698, 4.7089, 2.1902], device='cuda:1'), covar=tensor([0.0412, 0.0553, 0.0647, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.0966, 0.0896, 0.0790, 0.0616], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 22:24:53,816 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:25:07,046 INFO [train.py:968] (1/2) Epoch 9, batch 34300, giga_loss[loss=0.2803, simple_loss=0.3637, pruned_loss=0.09849, over 28714.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3447, pruned_loss=0.0938, over 5667564.21 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3556, pruned_loss=0.1214, over 5692385.05 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.344, pruned_loss=0.09088, over 5656158.06 frames. ], batch size: 262, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:25:42,430 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:26:06,133 INFO [train.py:968] (1/2) Epoch 9, batch 34350, giga_loss[loss=0.2795, simple_loss=0.3532, pruned_loss=0.1029, over 28777.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3445, pruned_loss=0.09335, over 5683244.62 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3548, pruned_loss=0.1207, over 5694580.52 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3443, pruned_loss=0.09064, over 5671402.89 frames. ], batch size: 243, lr: 3.55e-03, grad_scale: 2.0
+2023-03-04 22:26:50,650 INFO [optim.py:369] (1/2) 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 22:27:20,860 INFO [train.py:968] (1/2) Epoch 9, batch 34400, libri_loss[loss=0.2662, simple_loss=0.3374, pruned_loss=0.09752, over 29560.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3426, pruned_loss=0.09304, over 5689416.75 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3545, pruned_loss=0.1205, over 5698827.31 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3425, pruned_loss=0.09058, over 5676053.71 frames. ], batch size: 78, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:27:37,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0121, 1.2675, 1.2642, 1.0278], device='cuda:1'), covar=tensor([0.0976, 0.0991, 0.1661, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0710, 0.0644, 0.0633], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 22:27:59,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-04 22:28:26,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3140, 1.5542, 1.2580, 1.4685], device='cuda:1'), covar=tensor([0.0749, 0.0315, 0.0335, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0118, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0077, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-04 22:28:28,059 INFO [train.py:968] (1/2) Epoch 9, batch 34450, giga_loss[loss=0.2557, simple_loss=0.3447, pruned_loss=0.0833, over 28749.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.341, pruned_loss=0.09236, over 5689231.02 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3546, pruned_loss=0.1206, over 5695340.09 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3404, pruned_loss=0.08941, over 5680927.49 frames. ], batch size: 243, lr: 3.55e-03, grad_scale: 4.0
+2023-03-04 22:29:10,249 INFO [optim.py:369] (1/2) 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 22:29:14,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1555, 1.1599, 3.9832, 2.9880], device='cuda:1'), covar=tensor([0.1682, 0.2531, 0.0379, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0571, 0.0817, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 22:29:18,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 34500, giga_loss[loss=0.2425, simple_loss=0.3225, pruned_loss=0.08122, over 28952.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3392, pruned_loss=0.09046, over 5696072.29 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3552, pruned_loss=0.121, over 5697714.03 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3377, pruned_loss=0.08687, over 5687117.53 frames. ], batch size: 213, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:29:38,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:29:42,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7111, 1.7656, 1.2207, 1.3647], device='cuda:1'), covar=tensor([0.0700, 0.0569, 0.0991, 0.1032], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0431, 0.0496, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 22:30:18,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.05 vs. limit=2.0
+2023-03-04 22:30:39,426 INFO [train.py:968] (1/2) Epoch 9, batch 34550, giga_loss[loss=0.2831, simple_loss=0.3601, pruned_loss=0.103, over 28419.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.339, pruned_loss=0.09056, over 5692671.82 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3551, pruned_loss=0.1209, over 5697216.12 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3376, pruned_loss=0.08727, over 5686216.04 frames. ], batch size: 369, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:31:13,235 INFO [optim.py:369] (1/2) 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 22:31:36,045 INFO [train.py:968] (1/2) Epoch 9, batch 34600, giga_loss[loss=0.2927, simple_loss=0.3761, pruned_loss=0.1046, over 28758.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3408, pruned_loss=0.09182, over 5685634.51 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3553, pruned_loss=0.1212, over 5699250.65 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.339, pruned_loss=0.08798, over 5678400.98 frames. ], batch size: 262, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:31:55,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4816, 1.6811, 1.3355, 1.8294], device='cuda:1'), covar=tensor([0.2390, 0.2257, 0.2561, 0.2308], device='cuda:1'), in_proj_covar=tensor([0.1235, 0.0917, 0.1103, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 22:32:29,073 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:32:33,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=398845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:32:36,978 INFO [train.py:968] (1/2) Epoch 9, batch 34650, giga_loss[loss=0.2455, simple_loss=0.3292, pruned_loss=0.08092, over 28795.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3429, pruned_loss=0.09347, over 5679627.09 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3554, pruned_loss=0.1211, over 5704283.94 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3412, pruned_loss=0.08982, over 5668837.39 frames. ], batch size: 174, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:32:38,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=398848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:01,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:15,232 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=398877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:15,611 INFO [optim.py:369] (1/2) 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 22:33:38,923 INFO [train.py:968] (1/2) Epoch 9, batch 34700, giga_loss[loss=0.2607, simple_loss=0.3358, pruned_loss=0.09286, over 28890.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3398, pruned_loss=0.09257, over 5677714.15 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3554, pruned_loss=0.121, over 5705881.33 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3382, pruned_loss=0.08952, over 5667567.51 frames. ], batch size: 186, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:33:44,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=398903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:33:56,224 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=398913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:34:11,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8757, 1.8336, 1.3239, 1.5378], device='cuda:1'), covar=tensor([0.0651, 0.0538, 0.0888, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0433, 0.0498, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 22:34:36,130 INFO [train.py:968] (1/2) Epoch 9, batch 34750, giga_loss[loss=0.2442, simple_loss=0.3217, pruned_loss=0.08333, over 28679.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3385, pruned_loss=0.0925, over 5667555.66 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3549, pruned_loss=0.1207, over 5697877.26 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3374, pruned_loss=0.08985, over 5666908.23 frames. ], batch size: 242, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:35:14,288 INFO [optim.py:369] (1/2) 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 22:35:32,287 INFO [train.py:968] (1/2) Epoch 9, batch 34800, giga_loss[loss=0.33, simple_loss=0.4058, pruned_loss=0.1271, over 28678.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3434, pruned_loss=0.09584, over 5664761.07 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3546, pruned_loss=0.1206, over 5700245.67 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3427, pruned_loss=0.09353, over 5661688.57 frames. ], batch size: 243, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:35:41,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6891, 1.7616, 1.6459, 1.6384], device='cuda:1'), covar=tensor([0.1300, 0.2008, 0.1844, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0711, 0.0640, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 22:35:43,521 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:35:46,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:09,384 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:17,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:18,325 INFO [train.py:968] (1/2) Epoch 9, batch 34850, giga_loss[loss=0.2837, simple_loss=0.3726, pruned_loss=0.09746, over 28964.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3532, pruned_loss=0.1018, over 5676483.50 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3547, pruned_loss=0.1206, over 5705415.15 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3523, pruned_loss=0.0994, over 5668699.01 frames. ], batch size: 164, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:36:20,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:31,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:36:47,962 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:37:06,044 INFO [train.py:968] (1/2) Epoch 9, batch 34900, giga_loss[loss=0.2775, simple_loss=0.3609, pruned_loss=0.09709, over 28898.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3597, pruned_loss=0.1061, over 5679992.21 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.355, pruned_loss=0.1207, over 5704398.94 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3588, pruned_loss=0.1037, over 5673879.02 frames. ], batch size: 174, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:37:47,718 INFO [train.py:968] (1/2) Epoch 9, batch 34950, giga_loss[loss=0.283, simple_loss=0.3466, pruned_loss=0.1097, over 28974.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3575, pruned_loss=0.1061, over 5681315.75 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3553, pruned_loss=0.1209, over 5706806.97 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3566, pruned_loss=0.1036, over 5673458.36 frames. ], batch size: 227, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:37:52,273 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:38:13,046 INFO [optim.py:369] (1/2) 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 22:38:21,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4755, 1.6283, 1.4000, 1.2256], device='cuda:1'), covar=tensor([0.1928, 0.1640, 0.1217, 0.1606], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1474, 0.1423, 0.1564], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 22:38:30,926 INFO [train.py:968] (1/2) Epoch 9, batch 35000, giga_loss[loss=0.2187, simple_loss=0.2905, pruned_loss=0.07347, over 29106.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3498, pruned_loss=0.1028, over 5686964.44 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3551, pruned_loss=0.1207, over 5710923.42 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3492, pruned_loss=0.1007, over 5676754.83 frames. ], batch size: 128, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:38:37,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:38:39,137 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:38:45,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:39:05,286 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:39:15,559 INFO [train.py:968] (1/2) Epoch 9, batch 35050, giga_loss[loss=0.2819, simple_loss=0.3345, pruned_loss=0.1146, over 26652.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3416, pruned_loss=0.09887, over 5687267.51 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3548, pruned_loss=0.1205, over 5711566.71 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3413, pruned_loss=0.09725, over 5678546.63 frames. ], batch size: 555, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:39:23,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:39:39,569 INFO [optim.py:369] (1/2) 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 22:39:46,787 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 35100, giga_loss[loss=0.2539, simple_loss=0.2983, pruned_loss=0.1047, over 23853.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3348, pruned_loss=0.09614, over 5687317.81 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3554, pruned_loss=0.1207, over 5717275.75 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3336, pruned_loss=0.09398, over 5674751.25 frames. ], batch size: 705, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:40:35,454 INFO [train.py:968] (1/2) Epoch 9, batch 35150, giga_loss[loss=0.2279, simple_loss=0.2961, pruned_loss=0.07981, over 28905.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3284, pruned_loss=0.09327, over 5685979.66 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3559, pruned_loss=0.121, over 5712999.98 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3262, pruned_loss=0.09056, over 5678863.46 frames. ], batch size: 66, lr: 3.54e-03, grad_scale: 2.0
+2023-03-04 22:40:39,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7121, 4.6449, 1.8066, 1.9061], device='cuda:1'), covar=tensor([0.0879, 0.0218, 0.0827, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0501, 0.0333, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 22:40:45,811 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:40:48,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:40:56,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7019, 1.8090, 1.5924, 1.6028], device='cuda:1'), covar=tensor([0.2106, 0.1667, 0.1340, 0.1409], device='cuda:1'), in_proj_covar=tensor([0.1646, 0.1486, 0.1437, 0.1571], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 22:41:04,778 INFO [optim.py:369] (1/2) 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 22:41:14,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:41:20,705 INFO [train.py:968] (1/2) Epoch 9, batch 35200, libri_loss[loss=0.3287, simple_loss=0.3881, pruned_loss=0.1346, over 27908.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.324, pruned_loss=0.09153, over 5692299.11 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3563, pruned_loss=0.1213, over 5713424.54 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3217, pruned_loss=0.08882, over 5686107.13 frames. ], batch size: 116, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:41:25,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4930, 1.7181, 1.6859, 1.3328], device='cuda:1'), covar=tensor([0.1630, 0.2184, 0.1301, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0697, 0.0843, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-04 22:41:49,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:41:51,531 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:41:56,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3130, 1.6875, 1.3535, 1.4675], device='cuda:1'), covar=tensor([0.0727, 0.0376, 0.0339, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0118, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-04 22:42:03,239 INFO [train.py:968] (1/2) Epoch 9, batch 35250, giga_loss[loss=0.2442, simple_loss=0.313, pruned_loss=0.08769, over 27930.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3216, pruned_loss=0.09071, over 5691247.47 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3564, pruned_loss=0.1212, over 5714248.98 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3188, pruned_loss=0.08794, over 5685230.38 frames. ], batch size: 412, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:42:15,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:42:23,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0325, 1.2142, 3.5042, 2.9233], device='cuda:1'), covar=tensor([0.1582, 0.2228, 0.0466, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0630, 0.0571, 0.0826, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 22:42:28,979 INFO [optim.py:369] (1/2) 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 22:42:30,557 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:42:33,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-04 22:42:43,873 INFO [train.py:968] (1/2) Epoch 9, batch 35300, libri_loss[loss=0.355, simple_loss=0.3967, pruned_loss=0.1567, over 29574.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.319, pruned_loss=0.08971, over 5690346.34 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3571, pruned_loss=0.1218, over 5717982.68 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3152, pruned_loss=0.08619, over 5681514.67 frames. ], batch size: 78, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:43:08,632 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 35350, giga_loss[loss=0.1831, simple_loss=0.2554, pruned_loss=0.05537, over 28375.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3153, pruned_loss=0.08782, over 5679042.74 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3575, pruned_loss=0.1219, over 5718838.94 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3114, pruned_loss=0.08446, over 5670739.38 frames. ], batch size: 71, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:43:53,538 INFO [optim.py:369] (1/2) 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 22:44:08,647 INFO [train.py:968] (1/2) Epoch 9, batch 35400, libri_loss[loss=0.2723, simple_loss=0.3369, pruned_loss=0.1038, over 29362.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.311, pruned_loss=0.08516, over 5679903.10 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3577, pruned_loss=0.1221, over 5711483.44 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3072, pruned_loss=0.08195, over 5679481.50 frames. ], batch size: 67, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:44:34,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5744, 3.4969, 1.6724, 1.6049], device='cuda:1'), covar=tensor([0.0873, 0.0303, 0.0811, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0497, 0.0332, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 22:44:38,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:44:51,559 INFO [train.py:968] (1/2) Epoch 9, batch 35450, libri_loss[loss=0.2685, simple_loss=0.3258, pruned_loss=0.1056, over 29657.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3096, pruned_loss=0.08483, over 5682089.29 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3578, pruned_loss=0.1221, over 5710227.58 frames. ], giga_tot_loss[loss=0.2344, simple_loss=0.3057, pruned_loss=0.0816, over 5682345.28 frames. ], batch size: 69, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:45:06,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2974, 4.1036, 3.9142, 1.9638], device='cuda:1'), covar=tensor([0.0539, 0.0703, 0.0665, 0.2106], device='cuda:1'), in_proj_covar=tensor([0.0975, 0.0912, 0.0799, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 22:45:10,783 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:45:13,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7238, 1.7064, 1.3091, 1.3353], device='cuda:1'), covar=tensor([0.0694, 0.0577, 0.1043, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0435, 0.0498, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 22:45:13,159 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:45:31,802 INFO [train.py:968] (1/2) Epoch 9, batch 35500, giga_loss[loss=0.2429, simple_loss=0.3082, pruned_loss=0.08885, over 28803.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3083, pruned_loss=0.08452, over 5692220.12 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3583, pruned_loss=0.1223, over 5717857.40 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3028, pruned_loss=0.08031, over 5684452.19 frames. ], batch size: 243, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:45:36,631 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:46:18,182 INFO [train.py:968] (1/2) Epoch 9, batch 35550, giga_loss[loss=0.2044, simple_loss=0.2778, pruned_loss=0.06552, over 28947.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3052, pruned_loss=0.08294, over 5682079.01 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3586, pruned_loss=0.1223, over 5719854.09 frames. ], giga_tot_loss[loss=0.2292, simple_loss=0.3, pruned_loss=0.07917, over 5673633.07 frames. ], batch size: 145, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:46:33,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=399765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:46:43,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:46:46,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=399778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:46:47,485 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 35600, giga_loss[loss=0.2267, simple_loss=0.291, pruned_loss=0.08116, over 28337.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.304, pruned_loss=0.08249, over 5682650.14 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3589, pruned_loss=0.1224, over 5722796.97 frames. ], giga_tot_loss[loss=0.2282, simple_loss=0.2988, pruned_loss=0.07883, over 5672726.96 frames. ], batch size: 77, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:47:04,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-04 22:47:06,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4949, 1.7464, 1.4172, 1.6439], device='cuda:1'), covar=tensor([0.2193, 0.2167, 0.2375, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.1247, 0.0926, 0.1109, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 22:47:10,307 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=399807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:47:15,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1834, 2.0388, 2.0579, 1.8552], device='cuda:1'), covar=tensor([0.1286, 0.2201, 0.1716, 0.1885], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0722, 0.0649, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 22:47:47,924 INFO [train.py:968] (1/2) Epoch 9, batch 35650, giga_loss[loss=0.3079, simple_loss=0.3769, pruned_loss=0.1195, over 28552.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3162, pruned_loss=0.08887, over 5690857.77 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3597, pruned_loss=0.1228, over 5724420.79 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.31, pruned_loss=0.0846, over 5680167.04 frames. ], batch size: 307, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:47:53,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=399856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:47:59,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1767, 2.6262, 1.2411, 1.2971], device='cuda:1'), covar=tensor([0.0889, 0.0358, 0.0863, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0498, 0.0330, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 22:48:15,304 INFO [optim.py:369] (1/2) 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 22:48:24,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1776, 1.4631, 1.1598, 1.0122], device='cuda:1'), covar=tensor([0.2134, 0.2043, 0.2319, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.1245, 0.0924, 0.1107, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 22:48:30,777 INFO [train.py:968] (1/2) Epoch 9, batch 35700, giga_loss[loss=0.269, simple_loss=0.3454, pruned_loss=0.09624, over 28381.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3313, pruned_loss=0.09775, over 5689406.14 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3602, pruned_loss=0.123, over 5726788.42 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3247, pruned_loss=0.09323, over 5677728.79 frames. ], batch size: 71, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:48:33,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 22:49:09,928 INFO [train.py:968] (1/2) Epoch 9, batch 35750, giga_loss[loss=0.3262, simple_loss=0.3904, pruned_loss=0.131, over 29035.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3415, pruned_loss=0.1031, over 5684867.70 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3601, pruned_loss=0.1228, over 5721712.74 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3355, pruned_loss=0.09888, over 5679099.37 frames. ], batch size: 106, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:49:38,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7245, 1.9172, 1.9657, 1.5303], device='cuda:1'), covar=tensor([0.1571, 0.1978, 0.1200, 0.1468], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0699, 0.0842, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-04 22:49:39,093 INFO [optim.py:369] (1/2) 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 22:49:55,028 INFO [train.py:968] (1/2) Epoch 9, batch 35800, giga_loss[loss=0.2791, simple_loss=0.3629, pruned_loss=0.09768, over 28195.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3486, pruned_loss=0.1056, over 5682834.10 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3602, pruned_loss=0.1228, over 5722586.17 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3438, pruned_loss=0.1022, over 5677456.34 frames. ], batch size: 368, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:49:57,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=399999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:49:59,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:50:24,139 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:50:37,931 INFO [train.py:968] (1/2) Epoch 9, batch 35850, giga_loss[loss=0.2857, simple_loss=0.3617, pruned_loss=0.1048, over 28821.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.351, pruned_loss=0.1057, over 5681280.88 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3606, pruned_loss=0.1229, over 5726433.23 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3466, pruned_loss=0.1025, over 5672463.43 frames. ], batch size: 199, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:51:06,558 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 35900, giga_loss[loss=0.2528, simple_loss=0.3311, pruned_loss=0.08725, over 28426.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3514, pruned_loss=0.1046, over 5682776.57 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3608, pruned_loss=0.1228, over 5730793.87 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3475, pruned_loss=0.1017, over 5670812.52 frames. ], batch size: 65, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:51:38,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5400, 1.8294, 1.7621, 1.3865], device='cuda:1'), covar=tensor([0.1466, 0.2036, 0.1189, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0696, 0.0843, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 22:51:58,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:52:05,602 INFO [train.py:968] (1/2) Epoch 9, batch 35950, giga_loss[loss=0.3763, simple_loss=0.4224, pruned_loss=0.1651, over 27961.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.354, pruned_loss=0.1066, over 5677117.60 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3607, pruned_loss=0.1228, over 5722979.38 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3508, pruned_loss=0.1042, over 5674591.73 frames. ], batch size: 412, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:52:35,517 INFO [optim.py:369] (1/2) 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 22:52:48,851 INFO [train.py:968] (1/2) Epoch 9, batch 36000, giga_loss[loss=0.2994, simple_loss=0.3684, pruned_loss=0.1153, over 28335.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3582, pruned_loss=0.1102, over 5677153.42 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3614, pruned_loss=0.1234, over 5722951.66 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.355, pruned_loss=0.1075, over 5674409.03 frames. ], batch size: 368, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:52:48,851 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 22:52:57,485 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 22:53:26,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4568, 1.8439, 1.7182, 1.3013], device='cuda:1'), covar=tensor([0.1388, 0.2082, 0.1192, 0.1421], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0699, 0.0843, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 22:53:35,309 INFO [train.py:968] (1/2) Epoch 9, batch 36050, giga_loss[loss=0.289, simple_loss=0.3677, pruned_loss=0.1052, over 28929.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3619, pruned_loss=0.1126, over 5691127.08 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.362, pruned_loss=0.1234, over 5731382.15 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3588, pruned_loss=0.1098, over 5679471.70 frames. ], batch size: 213, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:53:53,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4741, 3.1396, 1.5043, 1.4998], device='cuda:1'), covar=tensor([0.0919, 0.0268, 0.0862, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0497, 0.0330, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 22:53:59,914 INFO [optim.py:369] (1/2) 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 22:54:01,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:54:03,601 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 36100, giga_loss[loss=0.2683, simple_loss=0.3516, pruned_loss=0.09249, over 28646.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3648, pruned_loss=0.1131, over 5697749.68 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.363, pruned_loss=0.124, over 5724772.81 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3612, pruned_loss=0.1099, over 5692438.11 frames. ], batch size: 85, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:54:25,353 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:54:44,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4989, 1.6949, 1.8273, 1.3871], device='cuda:1'), covar=tensor([0.1645, 0.2145, 0.1252, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0696, 0.0842, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 22:54:52,936 INFO [train.py:968] (1/2) Epoch 9, batch 36150, giga_loss[loss=0.301, simple_loss=0.3756, pruned_loss=0.1132, over 28647.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3659, pruned_loss=0.1128, over 5691158.70 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3633, pruned_loss=0.124, over 5725047.19 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3628, pruned_loss=0.11, over 5686262.56 frames. ], batch size: 92, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:55:18,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:55:18,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9833, 1.2222, 1.2779, 1.1502], device='cuda:1'), covar=tensor([0.1323, 0.1156, 0.1843, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0725, 0.0653, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 22:55:20,641 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:55:33,789 INFO [train.py:968] (1/2) Epoch 9, batch 36200, giga_loss[loss=0.2916, simple_loss=0.3677, pruned_loss=0.1078, over 28539.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3671, pruned_loss=0.1124, over 5705186.03 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3639, pruned_loss=0.1242, over 5729343.77 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3642, pruned_loss=0.1097, over 5696753.21 frames. ], batch size: 336, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:55:49,088 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:56:14,376 INFO [train.py:968] (1/2) Epoch 9, batch 36250, giga_loss[loss=0.2917, simple_loss=0.3645, pruned_loss=0.1095, over 28666.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3664, pruned_loss=0.111, over 5698825.10 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3641, pruned_loss=0.1242, over 5726674.41 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3639, pruned_loss=0.1085, over 5693631.59 frames. ], batch size: 262, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:56:18,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:56:20,568 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=400455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 22:56:41,569 INFO [optim.py:369] (1/2) 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 22:56:56,781 INFO [train.py:968] (1/2) Epoch 9, batch 36300, giga_loss[loss=0.2629, simple_loss=0.3431, pruned_loss=0.09137, over 28897.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3639, pruned_loss=0.1079, over 5701352.55 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3641, pruned_loss=0.1242, over 5726674.41 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3619, pruned_loss=0.106, over 5697310.38 frames. ], batch size: 186, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:57:37,366 INFO [train.py:968] (1/2) Epoch 9, batch 36350, giga_loss[loss=0.2648, simple_loss=0.349, pruned_loss=0.09032, over 28742.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.363, pruned_loss=0.1076, over 5683668.24 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3647, pruned_loss=0.1244, over 5720820.35 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3609, pruned_loss=0.1055, over 5684874.33 frames. ], batch size: 119, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:58:04,878 INFO [optim.py:369] (1/2) 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 22:58:21,179 INFO [train.py:968] (1/2) Epoch 9, batch 36400, giga_loss[loss=0.2885, simple_loss=0.3652, pruned_loss=0.1059, over 28899.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3661, pruned_loss=0.1116, over 5681162.29 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3653, pruned_loss=0.1248, over 5721307.01 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.364, pruned_loss=0.1095, over 5681398.90 frames. ], batch size: 145, lr: 3.54e-03, grad_scale: 8.0
+2023-03-04 22:59:02,298 INFO [train.py:968] (1/2) Epoch 9, batch 36450, giga_loss[loss=0.3278, simple_loss=0.3798, pruned_loss=0.1379, over 27624.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3684, pruned_loss=0.1151, over 5691624.65 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3657, pruned_loss=0.125, over 5727786.84 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3664, pruned_loss=0.1128, over 5684898.38 frames. ], batch size: 472, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:59:23,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0135, 1.1115, 1.0629, 0.8984], device='cuda:1'), covar=tensor([0.1194, 0.1579, 0.0839, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.1638, 0.1493, 0.1447, 0.1573], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 22:59:30,947 INFO [optim.py:369] (1/2) 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 22:59:44,649 INFO [train.py:968] (1/2) Epoch 9, batch 36500, libri_loss[loss=0.3418, simple_loss=0.4, pruned_loss=0.1418, over 27602.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3692, pruned_loss=0.1173, over 5692608.54 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3659, pruned_loss=0.1251, over 5728162.37 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3675, pruned_loss=0.1153, over 5686566.69 frames. ], batch size: 115, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 22:59:50,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3785, 1.5285, 1.4620, 1.4883], device='cuda:1'), covar=tensor([0.1117, 0.1459, 0.1649, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0733, 0.0657, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 23:00:30,097 INFO [train.py:968] (1/2) Epoch 9, batch 36550, giga_loss[loss=0.2803, simple_loss=0.3485, pruned_loss=0.1061, over 28573.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3674, pruned_loss=0.1169, over 5698219.62 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3661, pruned_loss=0.125, over 5731425.39 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3658, pruned_loss=0.1152, over 5689599.70 frames. ], batch size: 71, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:00:33,039 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400755.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:00:47,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-04 23:00:58,299 INFO [optim.py:369] (1/2) 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 23:01:05,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:01:09,638 INFO [train.py:968] (1/2) Epoch 9, batch 36600, giga_loss[loss=0.3354, simple_loss=0.3905, pruned_loss=0.1401, over 28743.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3666, pruned_loss=0.1171, over 5699790.19 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3662, pruned_loss=0.125, over 5732754.50 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3653, pruned_loss=0.1156, over 5690646.86 frames. ], batch size: 92, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:01:33,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:01:35,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=400830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:01:50,909 INFO [train.py:968] (1/2) Epoch 9, batch 36650, giga_loss[loss=0.2599, simple_loss=0.3373, pruned_loss=0.09125, over 29067.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3641, pruned_loss=0.1147, over 5689753.73 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3661, pruned_loss=0.1249, over 5723719.04 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3631, pruned_loss=0.1133, over 5690459.01 frames. ], batch size: 128, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:02:13,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3317, 1.5481, 1.2550, 1.4726], device='cuda:1'), covar=tensor([0.2546, 0.2498, 0.2762, 0.2171], device='cuda:1'), in_proj_covar=tensor([0.1244, 0.0922, 0.1102, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 23:02:24,103 INFO [optim.py:369] (1/2) 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 23:02:35,175 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:02:36,857 INFO [train.py:968] (1/2) Epoch 9, batch 36700, giga_loss[loss=0.2626, simple_loss=0.3506, pruned_loss=0.08733, over 28675.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3608, pruned_loss=0.1116, over 5698409.20 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.366, pruned_loss=0.1249, over 5725434.08 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.36, pruned_loss=0.1105, over 5697259.16 frames. ], batch size: 284, lr: 3.54e-03, grad_scale: 4.0
+2023-03-04 23:02:37,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:02:37,991 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400898.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:02:39,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400901.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:03:06,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:09,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400930.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:03:15,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:17,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:18,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5366, 1.6133, 1.3318, 1.5694], device='cuda:1'), covar=tensor([0.2309, 0.2343, 0.2512, 0.2141], device='cuda:1'), in_proj_covar=tensor([0.1244, 0.0924, 0.1103, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 23:03:19,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-04 23:03:25,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8653, 2.1000, 1.7160, 1.6642], device='cuda:1'), covar=tensor([0.1660, 0.1455, 0.1541, 0.1671], device='cuda:1'), in_proj_covar=tensor([0.1644, 0.1503, 0.1454, 0.1588], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 23:03:26,002 INFO [train.py:968] (1/2) Epoch 9, batch 36750, giga_loss[loss=0.2285, simple_loss=0.3054, pruned_loss=0.0758, over 28837.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.355, pruned_loss=0.1081, over 5692073.97 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3659, pruned_loss=0.1248, over 5724420.91 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3545, pruned_loss=0.1073, over 5691872.66 frames. ], batch size: 243, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:03:45,981 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=400967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:48,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:49,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=400973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:50,419 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:51,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=400976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:03:57,048 INFO [optim.py:369] (1/2) 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 23:04:10,784 INFO [train.py:968] (1/2) Epoch 9, batch 36800, giga_loss[loss=0.2412, simple_loss=0.3151, pruned_loss=0.08365, over 28754.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.35, pruned_loss=0.1055, over 5695065.56 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3677, pruned_loss=0.1261, over 5726288.91 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3475, pruned_loss=0.103, over 5692310.51 frames. ], batch size: 284, lr: 3.53e-03, grad_scale: 8.0
+2023-03-04 23:04:17,068 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:04:18,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:04:24,238 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:04:49,265 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-04 23:05:01,713 INFO [train.py:968] (1/2) Epoch 9, batch 36850, giga_loss[loss=0.2347, simple_loss=0.313, pruned_loss=0.07824, over 28913.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3437, pruned_loss=0.102, over 5679289.56 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.368, pruned_loss=0.1261, over 5722636.94 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3408, pruned_loss=0.09936, over 5679252.70 frames. ], batch size: 227, lr: 3.53e-03, grad_scale: 8.0
+2023-03-04 23:05:13,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-04 23:05:17,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:05:23,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1255, 1.2763, 3.7366, 3.0077], device='cuda:1'), covar=tensor([0.2159, 0.2930, 0.0700, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0622, 0.0564, 0.0811, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 23:05:31,656 INFO [optim.py:369] (1/2) 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 23:05:45,102 INFO [train.py:968] (1/2) Epoch 9, batch 36900, giga_loss[loss=0.2718, simple_loss=0.3268, pruned_loss=0.1084, over 23768.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3441, pruned_loss=0.1018, over 5677289.26 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3686, pruned_loss=0.1264, over 5727738.66 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3406, pruned_loss=0.09881, over 5671483.86 frames. ], batch size: 705, lr: 3.53e-03, grad_scale: 8.0
+2023-03-04 23:05:46,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-04 23:06:23,512 INFO [train.py:968] (1/2) Epoch 9, batch 36950, giga_loss[loss=0.3052, simple_loss=0.3689, pruned_loss=0.1207, over 28569.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3445, pruned_loss=0.102, over 5674049.87 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3685, pruned_loss=0.1263, over 5711965.41 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3412, pruned_loss=0.09903, over 5681861.96 frames. ], batch size: 307, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:06:45,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-04 23:06:53,304 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 37000, giga_loss[loss=0.235, simple_loss=0.3077, pruned_loss=0.08119, over 28546.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3427, pruned_loss=0.1007, over 5683357.38 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3685, pruned_loss=0.1261, over 5713771.77 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.34, pruned_loss=0.09833, over 5687610.40 frames. ], batch size: 71, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:07:37,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:07:46,166 INFO [train.py:968] (1/2) Epoch 9, batch 37050, giga_loss[loss=0.2326, simple_loss=0.3121, pruned_loss=0.07656, over 28861.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.34, pruned_loss=0.09967, over 5687559.04 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3685, pruned_loss=0.1261, over 5715741.03 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3375, pruned_loss=0.09763, over 5689001.98 frames. ], batch size: 174, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:08:06,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401274.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:08:13,215 INFO [optim.py:369] (1/2) 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 23:08:24,915 INFO [train.py:968] (1/2) Epoch 9, batch 37100, giga_loss[loss=0.25, simple_loss=0.3234, pruned_loss=0.08829, over 28358.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3379, pruned_loss=0.09859, over 5700380.12 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3691, pruned_loss=0.1262, over 5718484.14 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3351, pruned_loss=0.09645, over 5698743.83 frames. ], batch size: 77, lr: 3.53e-03, grad_scale: 1.0
+2023-03-04 23:08:52,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5007, 2.1541, 1.6071, 0.8048], device='cuda:1'), covar=tensor([0.3707, 0.1749, 0.2664, 0.3733], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1419, 0.1450, 0.1211], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 23:09:00,024 INFO [train.py:968] (1/2) Epoch 9, batch 37150, giga_loss[loss=0.227, simple_loss=0.3072, pruned_loss=0.07344, over 28903.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3396, pruned_loss=0.1003, over 5713290.40 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3709, pruned_loss=0.1272, over 5726516.89 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3342, pruned_loss=0.09649, over 5704083.87 frames. ], batch size: 145, lr: 3.53e-03, grad_scale: 1.0
+2023-03-04 23:09:06,818 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:09:10,722 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-04 23:09:27,887 INFO [optim.py:369] (1/2) 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 23:09:28,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:09:39,276 INFO [train.py:968] (1/2) Epoch 9, batch 37200, giga_loss[loss=0.2553, simple_loss=0.3285, pruned_loss=0.09101, over 28904.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3362, pruned_loss=0.09844, over 5711518.45 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3707, pruned_loss=0.1269, over 5729436.86 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3316, pruned_loss=0.09522, over 5701527.00 frames. ], batch size: 145, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:10:09,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:10:17,761 INFO [train.py:968] (1/2) Epoch 9, batch 37250, libri_loss[loss=0.4112, simple_loss=0.4535, pruned_loss=0.1844, over 19848.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3337, pruned_loss=0.09688, over 5706250.36 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3711, pruned_loss=0.127, over 5724314.49 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3289, pruned_loss=0.0936, over 5703498.77 frames. ], batch size: 188, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:10:35,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-04 23:10:46,617 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 37300, giga_loss[loss=0.2438, simple_loss=0.3182, pruned_loss=0.08473, over 28917.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3343, pruned_loss=0.09747, over 5706003.02 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3727, pruned_loss=0.1277, over 5716851.57 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3273, pruned_loss=0.09276, over 5710183.55 frames. ], batch size: 112, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:11:19,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:11:22,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:11:33,462 INFO [train.py:968] (1/2) Epoch 9, batch 37350, giga_loss[loss=0.2511, simple_loss=0.3204, pruned_loss=0.09092, over 29125.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3314, pruned_loss=0.09549, over 5708493.94 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3732, pruned_loss=0.1279, over 5711905.41 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3245, pruned_loss=0.09091, over 5716134.45 frames. ], batch size: 128, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:11:35,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1191, 3.4303, 1.4046, 1.4688], device='cuda:1'), covar=tensor([0.1369, 0.0364, 0.1025, 0.1670], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0494, 0.0328, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 23:11:35,395 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-04 23:11:44,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:00,246 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:02,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:03,293 INFO [optim.py:369] (1/2) 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 23:12:11,452 INFO [train.py:968] (1/2) Epoch 9, batch 37400, giga_loss[loss=0.2456, simple_loss=0.314, pruned_loss=0.0886, over 28838.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3307, pruned_loss=0.09508, over 5718755.02 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3738, pruned_loss=0.128, over 5716001.13 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3237, pruned_loss=0.09056, over 5721268.51 frames. ], batch size: 112, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:12:22,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:23,516 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:12:52,712 INFO [train.py:968] (1/2) Epoch 9, batch 37450, giga_loss[loss=0.2223, simple_loss=0.3019, pruned_loss=0.07136, over 28985.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3308, pruned_loss=0.09539, over 5719680.98 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3746, pruned_loss=0.1284, over 5718296.53 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3234, pruned_loss=0.09063, over 5719773.81 frames. ], batch size: 164, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:12:53,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=401649.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:13:02,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1785, 1.7008, 1.2901, 1.5306], device='cuda:1'), covar=tensor([0.0814, 0.0306, 0.0331, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0048, 0.0082], device='cuda:1')
+2023-03-04 23:13:05,024 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=401663.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:13:24,378 INFO [optim.py:369] (1/2) 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 23:13:33,112 INFO [train.py:968] (1/2) Epoch 9, batch 37500, giga_loss[loss=0.3039, simple_loss=0.3695, pruned_loss=0.1191, over 28816.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3359, pruned_loss=0.09845, over 5722358.88 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3749, pruned_loss=0.1281, over 5725643.28 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3279, pruned_loss=0.09358, over 5715452.77 frames. ], batch size: 186, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:14:05,080 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 37550, giga_loss[loss=0.3337, simple_loss=0.3812, pruned_loss=0.1431, over 23760.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3407, pruned_loss=0.1016, over 5714406.65 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3751, pruned_loss=0.1281, over 5729121.98 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3335, pruned_loss=0.09722, over 5705729.72 frames. ], batch size: 705, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:14:24,737 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:14:26,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:14:52,581 INFO [optim.py:369] (1/2) 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 23:14:55,146 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:14:58,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401792.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:15:00,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401795.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:15:01,995 INFO [train.py:968] (1/2) Epoch 9, batch 37600, giga_loss[loss=0.3207, simple_loss=0.3851, pruned_loss=0.1281, over 29017.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3504, pruned_loss=0.1084, over 5703039.40 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3765, pruned_loss=0.129, over 5725410.68 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3421, pruned_loss=0.1032, over 5698761.99 frames. ], batch size: 164, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:15:23,534 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401824.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:15:45,588 INFO [train.py:968] (1/2) Epoch 9, batch 37650, giga_loss[loss=0.2737, simple_loss=0.346, pruned_loss=0.1007, over 29017.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3573, pruned_loss=0.1129, over 5687501.94 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.377, pruned_loss=0.1293, over 5727014.37 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3496, pruned_loss=0.1079, over 5681598.54 frames. ], batch size: 128, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:16:00,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5495, 1.7185, 1.6128, 1.6257], device='cuda:1'), covar=tensor([0.0750, 0.0292, 0.0292, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0075, 0.0054, 0.0048, 0.0082], device='cuda:1')
+2023-03-04 23:16:13,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=401876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:16:15,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=401879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:16:20,343 INFO [optim.py:369] (1/2) 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 23:16:29,597 INFO [train.py:968] (1/2) Epoch 9, batch 37700, giga_loss[loss=0.2954, simple_loss=0.3661, pruned_loss=0.1124, over 28615.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5688194.52 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.377, pruned_loss=0.1293, over 5729962.52 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.354, pruned_loss=0.1097, over 5678707.18 frames. ], batch size: 85, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:16:36,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=401908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:17:14,830 INFO [train.py:968] (1/2) Epoch 9, batch 37750, giga_loss[loss=0.2917, simple_loss=0.3639, pruned_loss=0.1097, over 29076.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3647, pruned_loss=0.1155, over 5681985.67 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.377, pruned_loss=0.1292, over 5731730.16 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3588, pruned_loss=0.1116, over 5672073.90 frames. ], batch size: 128, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:17:45,926 INFO [optim.py:369] (1/2) 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 23:17:48,852 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-04 23:17:53,675 INFO [train.py:968] (1/2) Epoch 9, batch 37800, libri_loss[loss=0.3448, simple_loss=0.3995, pruned_loss=0.145, over 29510.00 frames. ], tot_loss[loss=0.303, simple_loss=0.369, pruned_loss=0.1185, over 5683980.49 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3766, pruned_loss=0.1291, over 5728318.84 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.364, pruned_loss=0.1147, over 5675502.52 frames. ], batch size: 84, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:18:26,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2198, 1.4646, 1.5376, 1.3163], device='cuda:1'), covar=tensor([0.1420, 0.1463, 0.1815, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0726, 0.0655, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-04 23:18:27,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=402038.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:18:37,786 INFO [train.py:968] (1/2) Epoch 9, batch 37850, giga_loss[loss=0.2439, simple_loss=0.3257, pruned_loss=0.08103, over 28832.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3656, pruned_loss=0.1161, over 5682250.67 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3765, pruned_loss=0.1289, over 5730144.98 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3616, pruned_loss=0.1131, over 5673794.71 frames. ], batch size: 199, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:19:05,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=402081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:19:09,759 INFO [optim.py:369] (1/2) 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 23:19:17,587 INFO [train.py:968] (1/2) Epoch 9, batch 37900, giga_loss[loss=0.2495, simple_loss=0.3298, pruned_loss=0.08463, over 28835.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3607, pruned_loss=0.1116, over 5680339.27 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3763, pruned_loss=0.129, over 5722595.84 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3575, pruned_loss=0.1089, over 5679936.18 frames. ], batch size: 186, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:19:59,800 INFO [train.py:968] (1/2) Epoch 9, batch 37950, giga_loss[loss=0.2952, simple_loss=0.3685, pruned_loss=0.1109, over 28996.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3592, pruned_loss=0.1103, over 5680318.53 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3762, pruned_loss=0.1292, over 5718137.47 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3562, pruned_loss=0.1073, over 5682954.02 frames. ], batch size: 128, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:20:29,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=402181.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:20:31,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=402184.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:20:33,497 INFO [optim.py:369] (1/2) 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 23:20:42,072 INFO [train.py:968] (1/2) Epoch 9, batch 38000, giga_loss[loss=0.2844, simple_loss=0.3621, pruned_loss=0.1033, over 28930.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.359, pruned_loss=0.1097, over 5685680.52 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.376, pruned_loss=0.1291, over 5720232.84 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3565, pruned_loss=0.1071, over 5685190.87 frames. ], batch size: 145, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:20:45,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4012, 1.4498, 1.3711, 1.3008], device='cuda:1'), covar=tensor([0.1731, 0.1649, 0.1106, 0.1374], device='cuda:1'), in_proj_covar=tensor([0.1640, 0.1511, 0.1478, 0.1598], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 23:20:56,051 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 38050, giga_loss[loss=0.2849, simple_loss=0.363, pruned_loss=0.1034, over 28607.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3624, pruned_loss=0.1119, over 5690934.34 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3766, pruned_loss=0.1295, over 5724971.30 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3595, pruned_loss=0.109, over 5685478.20 frames. ], batch size: 307, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:21:49,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-04 23:21:51,448 INFO [scaling.py:679] (1/2) 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] (1/2) 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 23:22:04,453 INFO [train.py:968] (1/2) Epoch 9, batch 38100, giga_loss[loss=0.3219, simple_loss=0.3861, pruned_loss=0.1289, over 28949.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3658, pruned_loss=0.1146, over 5677661.78 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3775, pruned_loss=0.1302, over 5711641.19 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3621, pruned_loss=0.111, over 5684187.68 frames. ], batch size: 174, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:22:48,568 INFO [train.py:968] (1/2) Epoch 9, batch 38150, giga_loss[loss=0.3429, simple_loss=0.3946, pruned_loss=0.1456, over 27917.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3672, pruned_loss=0.1158, over 5688301.57 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3774, pruned_loss=0.1299, over 5714756.45 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.3641, pruned_loss=0.1128, over 5689659.00 frames. ], batch size: 412, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:22:58,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3258, 3.0633, 2.0666, 2.0415], device='cuda:1'), covar=tensor([0.1580, 0.0792, 0.1190, 0.1206], device='cuda:1'), in_proj_covar=tensor([0.1638, 0.1508, 0.1478, 0.1599], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 23:23:00,321 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=402365.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:23:18,246 INFO [optim.py:369] (1/2) 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 23:23:20,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5357, 2.2319, 1.6027, 0.7246], device='cuda:1'), covar=tensor([0.2368, 0.1152, 0.1913, 0.2466], device='cuda:1'), in_proj_covar=tensor([0.1508, 0.1434, 0.1470, 0.1223], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 23:23:27,957 INFO [train.py:968] (1/2) Epoch 9, batch 38200, giga_loss[loss=0.295, simple_loss=0.3595, pruned_loss=0.1152, over 28845.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.368, pruned_loss=0.1166, over 5682044.64 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3771, pruned_loss=0.1294, over 5711827.84 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3654, pruned_loss=0.1141, over 5684997.00 frames. ], batch size: 99, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:24:09,219 INFO [train.py:968] (1/2) Epoch 9, batch 38250, giga_loss[loss=0.3061, simple_loss=0.3778, pruned_loss=0.1172, over 28631.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3681, pruned_loss=0.1166, over 5693688.30 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3772, pruned_loss=0.1294, over 5714662.60 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3656, pruned_loss=0.1143, over 5692889.48 frames. ], batch size: 336, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:24:14,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5777, 1.7298, 1.3741, 1.8346], device='cuda:1'), covar=tensor([0.2275, 0.2355, 0.2516, 0.2173], device='cuda:1'), in_proj_covar=tensor([0.1254, 0.0930, 0.1110, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 23:24:15,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=402456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:24:43,569 INFO [optim.py:369] (1/2) 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 23:24:50,783 INFO [train.py:968] (1/2) Epoch 9, batch 38300, giga_loss[loss=0.2893, simple_loss=0.3645, pruned_loss=0.107, over 28811.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3678, pruned_loss=0.1157, over 5699262.02 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3773, pruned_loss=0.1293, over 5717345.73 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3656, pruned_loss=0.1136, over 5695769.94 frames. ], batch size: 199, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:25:30,297 INFO [train.py:968] (1/2) Epoch 9, batch 38350, giga_loss[loss=0.2944, simple_loss=0.3715, pruned_loss=0.1087, over 29040.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3666, pruned_loss=0.1133, over 5702511.91 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3774, pruned_loss=0.1294, over 5715447.32 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3646, pruned_loss=0.1113, over 5701201.89 frames. ], batch size: 155, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:25:59,052 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-04 23:26:01,871 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 38400, giga_loss[loss=0.251, simple_loss=0.3356, pruned_loss=0.08323, over 29004.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3673, pruned_loss=0.1134, over 5697732.45 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1295, over 5711347.03 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3653, pruned_loss=0.1113, over 5700130.31 frames. ], batch size: 155, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:26:10,565 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=402599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:26:13,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=402602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:26:36,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=402631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:26:49,940 INFO [train.py:968] (1/2) Epoch 9, batch 38450, giga_loss[loss=0.2431, simple_loss=0.3276, pruned_loss=0.07929, over 28606.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3632, pruned_loss=0.1106, over 5703172.26 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3768, pruned_loss=0.129, over 5716014.94 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.362, pruned_loss=0.1089, over 5700553.59 frames. ], batch size: 60, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:26:51,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1707, 2.6457, 1.9708, 1.5891], device='cuda:1'), covar=tensor([0.1607, 0.1218, 0.1335, 0.1716], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1496, 0.1461, 0.1582], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 23:27:22,407 INFO [optim.py:369] (1/2) 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 23:27:22,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8320, 1.7858, 1.3325, 1.3666], device='cuda:1'), covar=tensor([0.0733, 0.0673, 0.0993, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0342, 0.0433, 0.0499, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 23:27:30,435 INFO [train.py:968] (1/2) Epoch 9, batch 38500, giga_loss[loss=0.2689, simple_loss=0.3392, pruned_loss=0.09931, over 28014.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3615, pruned_loss=0.1095, over 5706873.88 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3771, pruned_loss=0.1291, over 5714577.78 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3599, pruned_loss=0.1077, over 5705628.41 frames. ], batch size: 412, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:28:05,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=402740.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:28:11,380 INFO [train.py:968] (1/2) Epoch 9, batch 38550, giga_loss[loss=0.2609, simple_loss=0.3402, pruned_loss=0.09075, over 28897.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.359, pruned_loss=0.1079, over 5715533.95 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3775, pruned_loss=0.1293, over 5716258.94 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3572, pruned_loss=0.106, over 5712954.78 frames. ], batch size: 213, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:28:44,088 INFO [optim.py:369] (1/2) 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 23:28:51,135 INFO [train.py:968] (1/2) Epoch 9, batch 38600, giga_loss[loss=0.2978, simple_loss=0.3628, pruned_loss=0.1164, over 28868.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3588, pruned_loss=0.1085, over 5711881.65 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3771, pruned_loss=0.1291, over 5717815.33 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3574, pruned_loss=0.1069, over 5708567.44 frames. ], batch size: 99, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:29:33,209 INFO [train.py:968] (1/2) Epoch 9, batch 38650, giga_loss[loss=0.2654, simple_loss=0.339, pruned_loss=0.09586, over 28704.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3598, pruned_loss=0.1092, over 5716821.37 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1293, over 5720403.19 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3583, pruned_loss=0.1075, over 5711865.90 frames. ], batch size: 60, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:29:59,919 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=402883.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:30:01,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=402886.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:30:03,577 INFO [optim.py:369] (1/2) 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 23:30:10,637 INFO [train.py:968] (1/2) Epoch 9, batch 38700, libri_loss[loss=0.4054, simple_loss=0.4452, pruned_loss=0.1828, over 29756.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3604, pruned_loss=0.1092, over 5705311.14 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3776, pruned_loss=0.1297, over 5706507.89 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3583, pruned_loss=0.1067, over 5714398.40 frames. ], batch size: 87, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:30:24,075 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=402915.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:30:46,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=402945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:30:48,537 INFO [train.py:968] (1/2) Epoch 9, batch 38750, giga_loss[loss=0.311, simple_loss=0.384, pruned_loss=0.119, over 28003.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3595, pruned_loss=0.1081, over 5699857.75 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3772, pruned_loss=0.1295, over 5706048.48 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3579, pruned_loss=0.106, over 5707271.57 frames. ], batch size: 412, lr: 3.53e-03, grad_scale: 2.0
+2023-03-04 23:31:12,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-04 23:31:18,608 INFO [optim.py:369] (1/2) 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 23:31:25,042 INFO [train.py:968] (1/2) Epoch 9, batch 38800, giga_loss[loss=0.3112, simple_loss=0.3554, pruned_loss=0.1335, over 23461.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.359, pruned_loss=0.108, over 5708303.75 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3773, pruned_loss=0.1296, over 5710488.57 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3572, pruned_loss=0.1057, over 5710228.55 frames. ], batch size: 705, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:32:05,778 INFO [train.py:968] (1/2) Epoch 9, batch 38850, giga_loss[loss=0.2692, simple_loss=0.3435, pruned_loss=0.09741, over 28714.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3582, pruned_loss=0.108, over 5714493.00 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3772, pruned_loss=0.1294, over 5717914.28 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3561, pruned_loss=0.1055, over 5709204.87 frames. ], batch size: 262, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:32:15,730 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4074, 2.9325, 1.4068, 1.4710], device='cuda:1'), covar=tensor([0.0881, 0.0250, 0.0808, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0491, 0.0329, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0028, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 23:32:37,801 INFO [optim.py:369] (1/2) 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 23:32:38,775 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:32:44,357 INFO [train.py:968] (1/2) Epoch 9, batch 38900, giga_loss[loss=0.2546, simple_loss=0.3293, pruned_loss=0.08995, over 28951.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3553, pruned_loss=0.1067, over 5706887.73 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.377, pruned_loss=0.1291, over 5717598.43 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3532, pruned_loss=0.1042, over 5702334.56 frames. ], batch size: 119, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:33:22,887 INFO [train.py:968] (1/2) Epoch 9, batch 38950, giga_loss[loss=0.2902, simple_loss=0.3511, pruned_loss=0.1146, over 28979.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3524, pruned_loss=0.1052, over 5713530.96 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3764, pruned_loss=0.1286, over 5722877.50 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3503, pruned_loss=0.1028, over 5704878.07 frames. ], batch size: 106, lr: 3.53e-03, grad_scale: 4.0
+2023-03-04 23:33:54,117 INFO [optim.py:369] (1/2) 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 23:33:56,658 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:34:02,012 INFO [train.py:968] (1/2) Epoch 9, batch 39000, giga_loss[loss=0.3017, simple_loss=0.373, pruned_loss=0.1152, over 28307.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3524, pruned_loss=0.1053, over 5714987.58 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3767, pruned_loss=0.1287, over 5724529.32 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3498, pruned_loss=0.1026, over 5706179.28 frames. ], batch size: 368, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:34:02,012 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-04 23:34:10,205 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-04 23:34:37,786 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 39050, giga_loss[loss=0.2147, simple_loss=0.2946, pruned_loss=0.06738, over 28481.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.351, pruned_loss=0.1052, over 5710350.39 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3765, pruned_loss=0.1287, over 5727879.05 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3486, pruned_loss=0.1026, over 5700195.02 frames. ], batch size: 60, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:34:50,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4027, 1.6901, 1.3836, 1.5779], device='cuda:1'), covar=tensor([0.2312, 0.2275, 0.2552, 0.2205], device='cuda:1'), in_proj_covar=tensor([0.1249, 0.0930, 0.1110, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 23:34:58,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3353, 1.2445, 1.1053, 1.5120], device='cuda:1'), covar=tensor([0.0674, 0.0296, 0.0308, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-04 23:35:21,195 INFO [optim.py:369] (1/2) 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 23:35:29,016 INFO [train.py:968] (1/2) Epoch 9, batch 39100, giga_loss[loss=0.2663, simple_loss=0.3404, pruned_loss=0.09612, over 28286.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3484, pruned_loss=0.104, over 5712718.35 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3766, pruned_loss=0.1288, over 5728861.35 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3457, pruned_loss=0.1012, over 5703541.21 frames. ], batch size: 368, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:35:46,743 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 39150, giga_loss[loss=0.2637, simple_loss=0.3357, pruned_loss=0.09582, over 28989.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.347, pruned_loss=0.1039, over 5711007.40 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.377, pruned_loss=0.129, over 5722028.75 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3442, pruned_loss=0.1012, over 5708897.35 frames. ], batch size: 213, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:36:11,207 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7876, 1.6198, 1.3056, 1.2783], device='cuda:1'), covar=tensor([0.0626, 0.0607, 0.0987, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0435, 0.0499, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 23:36:37,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2353, 1.2460, 1.1054, 0.9464], device='cuda:1'), covar=tensor([0.0603, 0.0434, 0.0909, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0346, 0.0437, 0.0501, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 23:36:41,386 INFO [optim.py:369] (1/2) 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 23:36:47,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-04 23:36:49,400 INFO [train.py:968] (1/2) Epoch 9, batch 39200, giga_loss[loss=0.2586, simple_loss=0.3334, pruned_loss=0.09189, over 28809.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3447, pruned_loss=0.1029, over 5703578.39 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3768, pruned_loss=0.1289, over 5723986.84 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3421, pruned_loss=0.1005, over 5699981.60 frames. ], batch size: 284, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:36:56,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6152, 1.8299, 1.8670, 1.4564], device='cuda:1'), covar=tensor([0.1709, 0.2161, 0.1321, 0.1439], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0701, 0.0840, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 23:36:59,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6102, 1.7149, 1.3831, 1.9492], device='cuda:1'), covar=tensor([0.2214, 0.2321, 0.2598, 0.2279], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.0931, 0.1113, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 23:37:15,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-04 23:37:31,520 INFO [train.py:968] (1/2) Epoch 9, batch 39250, giga_loss[loss=0.246, simple_loss=0.3159, pruned_loss=0.08803, over 28494.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3425, pruned_loss=0.1013, over 5711379.95 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3769, pruned_loss=0.1291, over 5726565.97 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.34, pruned_loss=0.09882, over 5706035.57 frames. ], batch size: 71, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:37:45,615 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:37:46,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:37:47,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=403466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:03,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2492, 1.9653, 1.4596, 0.6130], device='cuda:1'), covar=tensor([0.3466, 0.1881, 0.2911, 0.3681], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1443, 0.1477, 0.1238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 23:38:07,415 INFO [optim.py:369] (1/2) 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 23:38:14,867 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:14,888 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=403495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:15,416 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:38:17,557 INFO [train.py:968] (1/2) Epoch 9, batch 39300, giga_loss[loss=0.2403, simple_loss=0.3143, pruned_loss=0.08315, over 28601.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3465, pruned_loss=0.103, over 5702922.82 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.377, pruned_loss=0.129, over 5728997.09 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3439, pruned_loss=0.1008, over 5696322.11 frames. ], batch size: 85, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:38:20,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-04 23:39:00,002 INFO [train.py:968] (1/2) Epoch 9, batch 39350, giga_loss[loss=0.319, simple_loss=0.3812, pruned_loss=0.1284, over 28844.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3495, pruned_loss=0.1043, over 5700910.23 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3767, pruned_loss=0.1286, over 5732876.24 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3472, pruned_loss=0.1024, over 5691388.02 frames. ], batch size: 145, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:39:15,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2502, 0.8967, 0.9289, 1.3396], device='cuda:1'), covar=tensor([0.0705, 0.0328, 0.0340, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-04 23:39:15,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:39:30,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8662, 1.1063, 3.3783, 2.9230], device='cuda:1'), covar=tensor([0.1716, 0.2568, 0.0431, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0626, 0.0570, 0.0819, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-04 23:39:35,492 INFO [optim.py:369] (1/2) 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 23:39:36,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1679, 1.4291, 1.1512, 1.0119], device='cuda:1'), covar=tensor([0.1851, 0.1715, 0.1138, 0.1458], device='cuda:1'), in_proj_covar=tensor([0.1643, 0.1524, 0.1496, 0.1594], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 23:39:40,882 INFO [train.py:968] (1/2) Epoch 9, batch 39400, giga_loss[loss=0.2934, simple_loss=0.3658, pruned_loss=0.1105, over 28685.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.352, pruned_loss=0.1051, over 5706674.42 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3771, pruned_loss=0.129, over 5736225.32 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3489, pruned_loss=0.1024, over 5694911.72 frames. ], batch size: 262, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:39:50,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:39:50,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:39:55,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=403611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:40:18,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=403640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:40:21,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.45 vs. limit=5.0
+2023-03-04 23:40:25,017 INFO [train.py:968] (1/2) Epoch 9, batch 39450, giga_loss[loss=0.2432, simple_loss=0.3238, pruned_loss=0.08131, over 28831.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3521, pruned_loss=0.1043, over 5705243.30 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3771, pruned_loss=0.1288, over 5738792.16 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.349, pruned_loss=0.1018, over 5692527.57 frames. ], batch size: 119, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:40:26,407 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-04 23:40:28,952 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=403653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:41:00,977 INFO [optim.py:369] (1/2) 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 23:41:06,137 INFO [train.py:968] (1/2) Epoch 9, batch 39500, giga_loss[loss=0.2387, simple_loss=0.3211, pruned_loss=0.07817, over 29023.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.351, pruned_loss=0.1037, over 5710406.33 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3771, pruned_loss=0.1288, over 5740230.77 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3483, pruned_loss=0.1014, over 5698791.64 frames. ], batch size: 155, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:41:15,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:41:17,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=403712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:41:40,605 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 39550, giga_loss[loss=0.3481, simple_loss=0.4034, pruned_loss=0.1464, over 28312.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3515, pruned_loss=0.1046, over 5713950.06 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3767, pruned_loss=0.1285, over 5745264.30 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3491, pruned_loss=0.1024, over 5699430.14 frames. ], batch size: 368, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:41:47,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=403750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:41:49,833 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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 23:42:28,975 INFO [train.py:968] (1/2) Epoch 9, batch 39600, libri_loss[loss=0.4142, simple_loss=0.4505, pruned_loss=0.189, over 19807.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3522, pruned_loss=0.1052, over 5713472.69 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3773, pruned_loss=0.1291, over 5736750.61 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3495, pruned_loss=0.1028, over 5710403.11 frames. ], batch size: 187, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:42:56,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2051, 2.9956, 2.7934, 1.3678], device='cuda:1'), covar=tensor([0.0907, 0.0974, 0.0905, 0.2446], device='cuda:1'), in_proj_covar=tensor([0.0979, 0.0910, 0.0812, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-04 23:43:08,898 INFO [train.py:968] (1/2) Epoch 9, batch 39650, giga_loss[loss=0.2594, simple_loss=0.3437, pruned_loss=0.08758, over 29019.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3551, pruned_loss=0.1065, over 5713128.75 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.377, pruned_loss=0.1288, over 5739462.60 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.353, pruned_loss=0.1045, over 5708136.59 frames. ], batch size: 164, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:43:25,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:43:25,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=403871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:43:37,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2438, 1.6331, 1.4122, 1.4847], device='cuda:1'), covar=tensor([0.0730, 0.0298, 0.0304, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-04 23:43:41,477 INFO [optim.py:369] (1/2) 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 23:43:47,119 INFO [train.py:968] (1/2) Epoch 9, batch 39700, giga_loss[loss=0.2845, simple_loss=0.3591, pruned_loss=0.105, over 28836.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3595, pruned_loss=0.1092, over 5709696.37 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3776, pruned_loss=0.1292, over 5739201.01 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3565, pruned_loss=0.1066, over 5704874.80 frames. ], batch size: 199, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:44:25,654 INFO [train.py:968] (1/2) Epoch 9, batch 39750, giga_loss[loss=0.287, simple_loss=0.3615, pruned_loss=0.1063, over 28622.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3606, pruned_loss=0.1097, over 5704828.98 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3776, pruned_loss=0.1291, over 5732765.34 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3578, pruned_loss=0.1072, over 5705917.55 frames. ], batch size: 242, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:44:33,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3299, 1.5908, 1.4476, 1.4640], device='cuda:1'), covar=tensor([0.1340, 0.1599, 0.1786, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0733, 0.0659, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-04 23:44:59,670 INFO [optim.py:369] (1/2) 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 23:45:04,320 INFO [train.py:968] (1/2) Epoch 9, batch 39800, giga_loss[loss=0.2805, simple_loss=0.3695, pruned_loss=0.09573, over 28942.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3618, pruned_loss=0.1105, over 5710664.63 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3779, pruned_loss=0.1293, over 5737857.50 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3587, pruned_loss=0.1076, over 5705438.16 frames. ], batch size: 174, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:45:16,269 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:18,100 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:19,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:19,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:27,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404028.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:45:42,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:45:43,098 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 39850, giga_loss[loss=0.2658, simple_loss=0.3365, pruned_loss=0.09758, over 28720.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.363, pruned_loss=0.1112, over 5705010.91 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3781, pruned_loss=0.1294, over 5733678.03 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3598, pruned_loss=0.1081, over 5704093.23 frames. ], batch size: 99, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:45:45,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7677, 2.7146, 1.7433, 0.8619], device='cuda:1'), covar=tensor([0.5417, 0.2323, 0.3213, 0.5005], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1447, 0.1479, 0.1238], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-04 23:46:15,999 INFO [optim.py:369] (1/2) 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 23:46:21,436 INFO [train.py:968] (1/2) Epoch 9, batch 39900, giga_loss[loss=0.309, simple_loss=0.3796, pruned_loss=0.1192, over 29054.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3618, pruned_loss=0.11, over 5708507.02 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3782, pruned_loss=0.1293, over 5733838.68 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3586, pruned_loss=0.107, over 5706759.01 frames. ], batch size: 128, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:46:48,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-04 23:46:56,892 INFO [train.py:968] (1/2) Epoch 9, batch 39950, giga_loss[loss=0.3015, simple_loss=0.3733, pruned_loss=0.1149, over 28852.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3606, pruned_loss=0.11, over 5706464.53 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3782, pruned_loss=0.1294, over 5728359.55 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3576, pruned_loss=0.107, over 5709819.33 frames. ], batch size: 174, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:47:09,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-04 23:47:16,837 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:47:18,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404174.0, num_to_drop=1, layers_to_drop={0}
+2023-03-04 23:47:32,505 INFO [optim.py:369] (1/2) 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 23:47:38,553 INFO [train.py:968] (1/2) Epoch 9, batch 40000, giga_loss[loss=0.2527, simple_loss=0.3295, pruned_loss=0.08798, over 28953.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3574, pruned_loss=0.1086, over 5708407.76 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3785, pruned_loss=0.1296, over 5730122.80 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3542, pruned_loss=0.1054, over 5709343.94 frames. ], batch size: 186, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:47:42,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404203.0, num_to_drop=1, layers_to_drop={1}
+2023-03-04 23:48:18,643 INFO [train.py:968] (1/2) Epoch 9, batch 40050, giga_loss[loss=0.3129, simple_loss=0.3787, pruned_loss=0.1236, over 28633.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3542, pruned_loss=0.1065, over 5705642.16 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3789, pruned_loss=0.1298, over 5723910.30 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.351, pruned_loss=0.1035, over 5711151.89 frames. ], batch size: 336, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:48:52,521 INFO [optim.py:369] (1/2) 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 23:48:57,918 INFO [train.py:968] (1/2) Epoch 9, batch 40100, giga_loss[loss=0.3085, simple_loss=0.3912, pruned_loss=0.1129, over 28659.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3551, pruned_loss=0.1053, over 5704277.70 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.379, pruned_loss=0.1299, over 5716109.61 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3523, pruned_loss=0.1027, over 5715083.31 frames. ], batch size: 336, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:49:24,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0571, 1.2404, 3.8391, 3.1365], device='cuda:1'), covar=tensor([0.1687, 0.2453, 0.0379, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0572, 0.0826, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 23:49:39,160 INFO [train.py:968] (1/2) Epoch 9, batch 40150, giga_loss[loss=0.269, simple_loss=0.3456, pruned_loss=0.0962, over 28791.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.356, pruned_loss=0.1049, over 5701221.37 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3785, pruned_loss=0.1294, over 5720276.72 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3536, pruned_loss=0.1025, over 5705835.33 frames. ], batch size: 186, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:49:41,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-04 23:50:03,424 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:50:12,141 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 40200, giga_loss[loss=0.3976, simple_loss=0.4281, pruned_loss=0.1836, over 26752.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3567, pruned_loss=0.1062, over 5699527.57 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3792, pruned_loss=0.1299, over 5716573.15 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3535, pruned_loss=0.1032, over 5707106.57 frames. ], batch size: 555, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:50:47,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:50:56,428 INFO [train.py:968] (1/2) Epoch 9, batch 40250, giga_loss[loss=0.3053, simple_loss=0.3735, pruned_loss=0.1185, over 29019.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3563, pruned_loss=0.1075, over 5695888.75 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3789, pruned_loss=0.1299, over 5712274.47 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3533, pruned_loss=0.1044, over 5704574.03 frames. ], batch size: 213, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:51:06,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-04 23:51:31,102 INFO [optim.py:369] (1/2) 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 23:51:34,772 INFO [train.py:968] (1/2) Epoch 9, batch 40300, giga_loss[loss=0.2891, simple_loss=0.3487, pruned_loss=0.1147, over 28881.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3554, pruned_loss=0.1082, over 5690375.75 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3797, pruned_loss=0.1303, over 5707100.37 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.352, pruned_loss=0.105, over 5702201.10 frames. ], batch size: 112, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:52:17,126 INFO [train.py:968] (1/2) Epoch 9, batch 40350, giga_loss[loss=0.2718, simple_loss=0.3381, pruned_loss=0.1027, over 28913.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3538, pruned_loss=0.1083, over 5702620.88 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3796, pruned_loss=0.1302, over 5709093.34 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3509, pruned_loss=0.1057, over 5709952.89 frames. ], batch size: 145, lr: 3.52e-03, grad_scale: 2.0
+2023-03-04 23:52:34,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9662, 1.1528, 1.0272, 0.7694], device='cuda:1'), covar=tensor([0.1555, 0.1607, 0.0901, 0.1595], device='cuda:1'), in_proj_covar=tensor([0.1663, 0.1529, 0.1513, 0.1614], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-04 23:52:41,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4203, 3.2521, 1.5929, 1.4206], device='cuda:1'), covar=tensor([0.0838, 0.0298, 0.0866, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0499, 0.0328, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0021, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-04 23:52:51,045 INFO [optim.py:369] (1/2) 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 23:52:55,073 INFO [train.py:968] (1/2) Epoch 9, batch 40400, giga_loss[loss=0.2933, simple_loss=0.3538, pruned_loss=0.1164, over 29033.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3516, pruned_loss=0.1076, over 5711810.59 frames. ], libri_tot_loss[loss=0.3199, simple_loss=0.3794, pruned_loss=0.1302, over 5712452.67 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3488, pruned_loss=0.1048, over 5714618.63 frames. ], batch size: 136, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:53:08,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:53:36,011 INFO [train.py:968] (1/2) Epoch 9, batch 40450, giga_loss[loss=0.2232, simple_loss=0.3005, pruned_loss=0.07299, over 28901.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3489, pruned_loss=0.1056, over 5710955.85 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3797, pruned_loss=0.1303, over 5706337.30 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3457, pruned_loss=0.1029, over 5718950.00 frames. ], batch size: 66, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:54:13,232 INFO [optim.py:369] (1/2) 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 23:54:17,779 INFO [train.py:968] (1/2) Epoch 9, batch 40500, giga_loss[loss=0.2381, simple_loss=0.3154, pruned_loss=0.08035, over 28786.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3448, pruned_loss=0.1038, over 5714052.14 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3797, pruned_loss=0.1304, over 5711403.10 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3417, pruned_loss=0.101, over 5716097.95 frames. ], batch size: 199, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:54:36,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-04 23:54:55,632 INFO [train.py:968] (1/2) Epoch 9, batch 40550, giga_loss[loss=0.2326, simple_loss=0.3121, pruned_loss=0.07658, over 28944.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3402, pruned_loss=0.1011, over 5716217.01 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3799, pruned_loss=0.1306, over 5712600.50 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3366, pruned_loss=0.09805, over 5717118.58 frames. ], batch size: 213, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:55:00,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:55:08,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-04 23:55:31,233 INFO [optim.py:369] (1/2) 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 23:55:36,217 INFO [train.py:968] (1/2) Epoch 9, batch 40600, giga_loss[loss=0.263, simple_loss=0.3477, pruned_loss=0.08911, over 29012.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3407, pruned_loss=0.1008, over 5711045.37 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1308, over 5711121.92 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3373, pruned_loss=0.09797, over 5713013.94 frames. ], batch size: 164, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:55:48,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:56:16,092 INFO [train.py:968] (1/2) Epoch 9, batch 40650, giga_loss[loss=0.2877, simple_loss=0.3641, pruned_loss=0.1056, over 28559.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3441, pruned_loss=0.1021, over 5697245.10 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3803, pruned_loss=0.131, over 5694466.78 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3402, pruned_loss=0.09886, over 5713306.21 frames. ], batch size: 336, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:56:52,859 INFO [optim.py:369] (1/2) 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 23:56:56,385 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:56:56,726 INFO [train.py:968] (1/2) Epoch 9, batch 40700, giga_loss[loss=0.2633, simple_loss=0.3331, pruned_loss=0.09678, over 28721.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.348, pruned_loss=0.1041, over 5699122.49 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3802, pruned_loss=0.131, over 5700034.14 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.344, pruned_loss=0.1006, over 5707015.29 frames. ], batch size: 99, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:56:58,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:18,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4723, 1.7264, 1.7189, 1.3150], device='cuda:1'), covar=tensor([0.1628, 0.2092, 0.1328, 0.1494], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0694, 0.0835, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-04 23:57:22,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:34,156 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 40750, giga_loss[loss=0.3066, simple_loss=0.3722, pruned_loss=0.1205, over 28924.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3515, pruned_loss=0.1053, over 5709102.69 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3798, pruned_loss=0.1306, over 5702446.65 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3481, pruned_loss=0.1025, over 5713297.99 frames. ], batch size: 66, lr: 3.52e-03, grad_scale: 4.0
+2023-03-04 23:57:45,083 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=404956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:46,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=404959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:57:52,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-04 23:57:53,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=404967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:58:09,230 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=404988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:58:09,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=404988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-04 23:58:12,573 INFO [optim.py:369] (1/2) 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 23:58:18,179 INFO [train.py:968] (1/2) Epoch 9, batch 40800, giga_loss[loss=0.2728, simple_loss=0.3514, pruned_loss=0.09709, over 28870.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.354, pruned_loss=0.1064, over 5710006.02 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3795, pruned_loss=0.1303, over 5702467.95 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3509, pruned_loss=0.1038, over 5713737.54 frames. ], batch size: 199, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:58:41,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-04 23:58:58,181 INFO [train.py:968] (1/2) Epoch 9, batch 40850, giga_loss[loss=0.2706, simple_loss=0.343, pruned_loss=0.09908, over 28903.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3558, pruned_loss=0.1078, over 5710218.17 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3785, pruned_loss=0.1297, over 5709339.78 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3533, pruned_loss=0.1055, over 5707270.13 frames. ], batch size: 106, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:59:20,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6008, 1.3035, 5.2027, 3.3967], device='cuda:1'), covar=tensor([0.1606, 0.2627, 0.0312, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0575, 0.0832, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-04 23:59:36,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6584, 2.0349, 1.5707, 2.1262], device='cuda:1'), covar=tensor([0.2201, 0.2029, 0.2291, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.1255, 0.0927, 0.1103, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-04 23:59:43,324 INFO [optim.py:369] (1/2) 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 23:59:46,798 INFO [train.py:968] (1/2) Epoch 9, batch 40900, giga_loss[loss=0.3222, simple_loss=0.3837, pruned_loss=0.1303, over 28603.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.363, pruned_loss=0.1149, over 5692991.17 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.378, pruned_loss=0.1293, over 5713755.73 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.361, pruned_loss=0.1129, over 5686697.13 frames. ], batch size: 71, lr: 3.52e-03, grad_scale: 8.0
+2023-03-04 23:59:49,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-05 00:00:18,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=405131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:00:21,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=405134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:00:32,089 INFO [train.py:968] (1/2) Epoch 9, batch 40950, giga_loss[loss=0.3503, simple_loss=0.3824, pruned_loss=0.1591, over 23644.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3694, pruned_loss=0.12, over 5687947.79 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3777, pruned_loss=0.1291, over 5717435.13 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3677, pruned_loss=0.1182, over 5678826.47 frames. ], batch size: 705, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:00:34,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3141, 4.1165, 3.8982, 1.7387], device='cuda:1'), covar=tensor([0.0500, 0.0669, 0.0673, 0.2138], device='cuda:1'), in_proj_covar=tensor([0.1002, 0.0937, 0.0826, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 00:00:45,147 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:01:05,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0294, 1.3158, 1.2750, 1.1447], device='cuda:1'), covar=tensor([0.1117, 0.0999, 0.1710, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0739, 0.0668, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 00:01:07,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2605, 1.4734, 1.2519, 1.0283], device='cuda:1'), covar=tensor([0.1622, 0.1565, 0.1054, 0.1527], device='cuda:1'), in_proj_covar=tensor([0.1689, 0.1555, 0.1533, 0.1632], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 00:01:09,832 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 41000, giga_loss[loss=0.3327, simple_loss=0.3916, pruned_loss=0.1369, over 28691.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.376, pruned_loss=0.1244, over 5686461.50 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3783, pruned_loss=0.1293, over 5714127.63 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.374, pruned_loss=0.1226, over 5681010.00 frames. ], batch size: 242, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:01:39,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3992, 1.6451, 1.3183, 1.5510], device='cuda:1'), covar=tensor([0.2344, 0.2286, 0.2539, 0.2096], device='cuda:1'), in_proj_covar=tensor([0.1248, 0.0926, 0.1101, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 00:01:57,329 INFO [train.py:968] (1/2) Epoch 9, batch 41050, giga_loss[loss=0.4029, simple_loss=0.4332, pruned_loss=0.1863, over 27580.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3835, pruned_loss=0.1314, over 5675817.28 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3786, pruned_loss=0.1296, over 5715482.01 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3817, pruned_loss=0.1297, over 5669332.24 frames. ], batch size: 472, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:02:39,039 INFO [optim.py:369] (1/2) 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 00:02:43,355 INFO [train.py:968] (1/2) Epoch 9, batch 41100, giga_loss[loss=0.4142, simple_loss=0.4489, pruned_loss=0.1898, over 28253.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3899, pruned_loss=0.1366, over 5678321.38 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3787, pruned_loss=0.1294, over 5717161.69 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3885, pruned_loss=0.1355, over 5671061.74 frames. ], batch size: 368, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:02:47,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1735, 0.8504, 0.8109, 1.3858], device='cuda:1'), covar=tensor([0.0784, 0.0352, 0.0351, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0116, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0082], device='cuda:1')
+2023-03-05 00:03:01,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-05 00:03:05,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=405319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:03:28,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=405342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:03:33,321 INFO [train.py:968] (1/2) Epoch 9, batch 41150, giga_loss[loss=0.3633, simple_loss=0.4194, pruned_loss=0.1536, over 28989.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3943, pruned_loss=0.1413, over 5659865.58 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3792, pruned_loss=0.1297, over 5719582.21 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3931, pruned_loss=0.1403, over 5650371.74 frames. ], batch size: 136, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:04:23,738 INFO [optim.py:369] (1/2) 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 00:04:27,760 INFO [train.py:968] (1/2) Epoch 9, batch 41200, giga_loss[loss=0.3137, simple_loss=0.3785, pruned_loss=0.1244, over 28888.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3966, pruned_loss=0.1439, over 5659256.52 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3793, pruned_loss=0.1298, over 5722534.44 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3957, pruned_loss=0.1433, over 5648416.39 frames. ], batch size: 199, lr: 3.52e-03, grad_scale: 8.0
+2023-03-05 00:05:19,461 INFO [train.py:968] (1/2) Epoch 9, batch 41250, giga_loss[loss=0.3456, simple_loss=0.4076, pruned_loss=0.1418, over 28974.00 frames. ], tot_loss[loss=0.345, simple_loss=0.398, pruned_loss=0.146, over 5637264.04 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3791, pruned_loss=0.1297, over 5725743.95 frames. ], giga_tot_loss[loss=0.3449, simple_loss=0.398, pruned_loss=0.1459, over 5624053.18 frames. ], batch size: 164, lr: 3.52e-03, grad_scale: 4.0
+2023-03-05 00:05:35,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=405462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:05:38,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=405465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:05:58,147 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=405485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:06:00,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=405488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:06:07,225 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:06:09,797 INFO [train.py:968] (1/2) Epoch 9, batch 41300, giga_loss[loss=0.5439, simple_loss=0.518, pruned_loss=0.2849, over 26570.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4023, pruned_loss=0.1511, over 5634407.52 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3787, pruned_loss=0.1294, over 5730513.61 frames. ], giga_tot_loss[loss=0.3535, simple_loss=0.4033, pruned_loss=0.1519, over 5616813.02 frames. ], batch size: 555, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:06:31,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=405517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:06:38,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 00:06:53,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2443, 1.3739, 1.1773, 1.3676], device='cuda:1'), covar=tensor([0.0737, 0.0310, 0.0304, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-05 00:07:01,813 INFO [train.py:968] (1/2) Epoch 9, batch 41350, giga_loss[loss=0.3266, simple_loss=0.3847, pruned_loss=0.1342, over 28860.00 frames. ], tot_loss[loss=0.3534, simple_loss=0.4038, pruned_loss=0.1516, over 5634055.63 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3786, pruned_loss=0.1293, over 5719302.34 frames. ], giga_tot_loss[loss=0.355, simple_loss=0.4049, pruned_loss=0.1525, over 5629829.49 frames. ], batch size: 199, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:07:52,580 INFO [optim.py:369] (1/2) 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 00:07:55,669 INFO [train.py:968] (1/2) Epoch 9, batch 41400, giga_loss[loss=0.2946, simple_loss=0.3581, pruned_loss=0.1155, over 28523.00 frames. ], tot_loss[loss=0.3535, simple_loss=0.4032, pruned_loss=0.1518, over 5631696.91 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3786, pruned_loss=0.1294, over 5716734.83 frames. ], giga_tot_loss[loss=0.3547, simple_loss=0.4042, pruned_loss=0.1527, over 5630302.01 frames. ], batch size: 78, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:08:28,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-05 00:08:41,879 INFO [train.py:968] (1/2) Epoch 9, batch 41450, giga_loss[loss=0.335, simple_loss=0.3991, pruned_loss=0.1355, over 29015.00 frames. ], tot_loss[loss=0.3523, simple_loss=0.4018, pruned_loss=0.1514, over 5622219.48 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3789, pruned_loss=0.1295, over 5711652.23 frames. ], giga_tot_loss[loss=0.354, simple_loss=0.403, pruned_loss=0.1525, over 5622764.53 frames. ], batch size: 164, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:09:29,831 INFO [optim.py:369] (1/2) 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 00:09:32,892 INFO [train.py:968] (1/2) Epoch 9, batch 41500, giga_loss[loss=0.3169, simple_loss=0.3902, pruned_loss=0.1218, over 28657.00 frames. ], tot_loss[loss=0.3502, simple_loss=0.401, pruned_loss=0.1498, over 5620821.65 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3791, pruned_loss=0.1296, over 5712765.90 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4021, pruned_loss=0.1509, over 5619118.39 frames. ], batch size: 60, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:10:21,278 INFO [train.py:968] (1/2) Epoch 9, batch 41550, giga_loss[loss=0.4016, simple_loss=0.4304, pruned_loss=0.1865, over 26561.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4037, pruned_loss=0.1519, over 5594586.67 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3793, pruned_loss=0.1298, over 5691710.44 frames. ], giga_tot_loss[loss=0.3554, simple_loss=0.4048, pruned_loss=0.153, over 5609402.77 frames. ], batch size: 555, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:11:07,191 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 41600, giga_loss[loss=0.2972, simple_loss=0.3653, pruned_loss=0.1145, over 29028.00 frames. ], tot_loss[loss=0.3524, simple_loss=0.4027, pruned_loss=0.151, over 5564228.90 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3796, pruned_loss=0.1302, over 5669135.60 frames. ], giga_tot_loss[loss=0.3542, simple_loss=0.4041, pruned_loss=0.1522, over 5592406.89 frames. ], batch size: 155, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:12:01,040 INFO [train.py:968] (1/2) Epoch 9, batch 41650, giga_loss[loss=0.3121, simple_loss=0.3812, pruned_loss=0.1215, over 28589.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3991, pruned_loss=0.1469, over 5588579.17 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3798, pruned_loss=0.1304, over 5672340.24 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4004, pruned_loss=0.1481, over 5606597.64 frames. ], batch size: 85, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:12:36,000 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-05 00:12:46,970 INFO [optim.py:369] (1/2) 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 00:12:48,327 INFO [train.py:968] (1/2) Epoch 9, batch 41700, giga_loss[loss=0.2698, simple_loss=0.3454, pruned_loss=0.09712, over 28915.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3979, pruned_loss=0.1447, over 5609776.87 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3792, pruned_loss=0.1301, over 5673933.60 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.3997, pruned_loss=0.1461, over 5621658.76 frames. ], batch size: 136, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:13:03,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3891, 1.7434, 1.3487, 1.5442], device='cuda:1'), covar=tensor([0.2376, 0.2169, 0.2436, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.1251, 0.0929, 0.1105, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 00:13:14,250 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-05 00:13:32,286 INFO [train.py:968] (1/2) Epoch 9, batch 41750, giga_loss[loss=0.3156, simple_loss=0.3797, pruned_loss=0.1257, over 28817.00 frames. ], tot_loss[loss=0.338, simple_loss=0.3933, pruned_loss=0.1413, over 5610970.37 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3781, pruned_loss=0.1294, over 5675173.48 frames. ], giga_tot_loss[loss=0.3419, simple_loss=0.3965, pruned_loss=0.1436, over 5616671.09 frames. ], batch size: 284, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:14:16,914 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 41800, giga_loss[loss=0.2969, simple_loss=0.3694, pruned_loss=0.1122, over 29022.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3903, pruned_loss=0.1385, over 5604942.45 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3781, pruned_loss=0.1295, over 5670153.15 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3932, pruned_loss=0.1405, over 5613495.62 frames. ], batch size: 164, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:15:08,116 INFO [train.py:968] (1/2) Epoch 9, batch 41850, giga_loss[loss=0.3275, simple_loss=0.3886, pruned_loss=0.1332, over 28820.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3884, pruned_loss=0.137, over 5625841.43 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3781, pruned_loss=0.1296, over 5666018.60 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3908, pruned_loss=0.1387, over 5635054.54 frames. ], batch size: 199, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:15:50,466 INFO [optim.py:369] (1/2) 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 00:15:51,915 INFO [train.py:968] (1/2) Epoch 9, batch 41900, giga_loss[loss=0.3139, simple_loss=0.3842, pruned_loss=0.1217, over 29030.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3871, pruned_loss=0.1358, over 5638508.36 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3778, pruned_loss=0.1293, over 5671331.05 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3895, pruned_loss=0.1375, over 5640478.64 frames. ], batch size: 136, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:16:10,990 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:16:41,845 INFO [train.py:968] (1/2) Epoch 9, batch 41950, giga_loss[loss=0.2789, simple_loss=0.3505, pruned_loss=0.1036, over 28599.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3862, pruned_loss=0.1351, over 5631055.91 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3782, pruned_loss=0.1297, over 5665779.10 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.388, pruned_loss=0.1363, over 5636452.47 frames. ], batch size: 307, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:17:10,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 00:17:30,761 INFO [optim.py:369] (1/2) 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 00:17:32,217 INFO [train.py:968] (1/2) Epoch 9, batch 42000, giga_loss[loss=0.3662, simple_loss=0.4277, pruned_loss=0.1523, over 27984.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3853, pruned_loss=0.133, over 5623888.60 frames. ], libri_tot_loss[loss=0.3191, simple_loss=0.3786, pruned_loss=0.1298, over 5658026.15 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3866, pruned_loss=0.134, over 5634325.62 frames. ], batch size: 412, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:17:32,217 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 00:17:36,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2601, 1.3166, 1.0957, 1.4499], device='cuda:1'), covar=tensor([0.0861, 0.0337, 0.0356, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-05 00:17:41,245 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 00:18:32,232 INFO [train.py:968] (1/2) Epoch 9, batch 42050, giga_loss[loss=0.3715, simple_loss=0.4263, pruned_loss=0.1583, over 28726.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3872, pruned_loss=0.1323, over 5635223.38 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3781, pruned_loss=0.1298, over 5659402.00 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3888, pruned_loss=0.1332, over 5641744.37 frames. ], batch size: 262, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:18:48,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-05 00:19:17,765 INFO [optim.py:369] (1/2) 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 00:19:18,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3089, 1.5953, 1.2736, 1.3323], device='cuda:1'), covar=tensor([0.2106, 0.2076, 0.2230, 0.1792], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.0937, 0.1111, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 00:19:18,639 INFO [train.py:968] (1/2) Epoch 9, batch 42100, giga_loss[loss=0.3778, simple_loss=0.4265, pruned_loss=0.1646, over 28510.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3881, pruned_loss=0.1325, over 5651350.13 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3777, pruned_loss=0.1295, over 5664710.34 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3899, pruned_loss=0.1335, over 5651642.03 frames. ], batch size: 336, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:19:32,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5849, 1.9055, 1.5347, 1.4328], device='cuda:1'), covar=tensor([0.1729, 0.1449, 0.1493, 0.1489], device='cuda:1'), in_proj_covar=tensor([0.1663, 0.1543, 0.1499, 0.1609], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 00:19:33,136 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 00:20:05,189 INFO [train.py:968] (1/2) Epoch 9, batch 42150, giga_loss[loss=0.3113, simple_loss=0.3801, pruned_loss=0.1213, over 28765.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.388, pruned_loss=0.1329, over 5656678.42 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3773, pruned_loss=0.1292, over 5670695.00 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3901, pruned_loss=0.1341, over 5650990.79 frames. ], batch size: 284, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:20:51,506 INFO [optim.py:369] (1/2) 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 00:20:52,136 INFO [train.py:968] (1/2) Epoch 9, batch 42200, giga_loss[loss=0.3697, simple_loss=0.4209, pruned_loss=0.1593, over 28287.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3866, pruned_loss=0.1323, over 5664788.90 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1293, over 5672496.11 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3882, pruned_loss=0.1331, over 5658527.70 frames. ], batch size: 368, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:21:37,968 INFO [train.py:968] (1/2) Epoch 9, batch 42250, giga_loss[loss=0.3944, simple_loss=0.4177, pruned_loss=0.1855, over 26631.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3856, pruned_loss=0.1333, over 5658032.77 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1293, over 5672746.06 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3871, pruned_loss=0.1341, over 5653011.02 frames. ], batch size: 555, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:22:19,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:22:21,241 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:22:26,040 INFO [optim.py:369] (1/2) 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 00:22:26,749 INFO [train.py:968] (1/2) Epoch 9, batch 42300, giga_loss[loss=0.298, simple_loss=0.3704, pruned_loss=0.1128, over 28605.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3846, pruned_loss=0.1328, over 5658922.79 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3776, pruned_loss=0.1294, over 5673040.33 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3856, pruned_loss=0.1333, over 5654680.94 frames. ], batch size: 78, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:23:11,791 INFO [train.py:968] (1/2) Epoch 9, batch 42350, libri_loss[loss=0.2787, simple_loss=0.3392, pruned_loss=0.1091, over 29368.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3838, pruned_loss=0.1307, over 5675196.02 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3775, pruned_loss=0.1291, over 5682401.76 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3851, pruned_loss=0.1315, over 5662610.10 frames. ], batch size: 71, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:23:12,012 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:23:55,438 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 9, batch 42400, giga_loss[loss=0.2884, simple_loss=0.3577, pruned_loss=0.1096, over 28625.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3836, pruned_loss=0.1297, over 5684400.28 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3778, pruned_loss=0.1292, over 5684712.98 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3844, pruned_loss=0.1302, over 5672256.99 frames. ], batch size: 85, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:24:12,236 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=406614.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:24:25,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3794, 4.2169, 3.9556, 1.9350], device='cuda:1'), covar=tensor([0.0531, 0.0652, 0.0679, 0.2048], device='cuda:1'), in_proj_covar=tensor([0.1008, 0.0944, 0.0830, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 00:24:29,946 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=406635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:24:33,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=406638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:24:40,911 INFO [train.py:968] (1/2) Epoch 9, batch 42450, giga_loss[loss=0.3065, simple_loss=0.3481, pruned_loss=0.1324, over 23513.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3845, pruned_loss=0.1309, over 5675740.45 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3775, pruned_loss=0.1289, over 5690261.76 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3857, pruned_loss=0.1316, over 5660546.40 frames. ], batch size: 705, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:24:43,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-05 00:24:59,176 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=406667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:25:23,698 INFO [optim.py:369] (1/2) 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 00:25:23,711 INFO [train.py:968] (1/2) Epoch 9, batch 42500, giga_loss[loss=0.2647, simple_loss=0.3444, pruned_loss=0.09252, over 28902.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3816, pruned_loss=0.1294, over 5685422.36 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3767, pruned_loss=0.1284, over 5694274.23 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3833, pruned_loss=0.1305, over 5669623.94 frames. ], batch size: 174, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:26:13,505 INFO [train.py:968] (1/2) Epoch 9, batch 42550, libri_loss[loss=0.3046, simple_loss=0.3818, pruned_loss=0.1137, over 29647.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.381, pruned_loss=0.1298, over 5674679.18 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3767, pruned_loss=0.1283, over 5690693.74 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3826, pruned_loss=0.1308, over 5665114.65 frames. ], batch size: 88, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:26:17,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5555, 1.8078, 1.4104, 1.3080], device='cuda:1'), covar=tensor([0.1731, 0.1455, 0.1182, 0.1475], device='cuda:1'), in_proj_covar=tensor([0.1669, 0.1556, 0.1504, 0.1611], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 00:26:35,374 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:26:58,880 INFO [train.py:968] (1/2) Epoch 9, batch 42600, giga_loss[loss=0.3003, simple_loss=0.3699, pruned_loss=0.1153, over 28854.00 frames. ], tot_loss[loss=0.32, simple_loss=0.38, pruned_loss=0.13, over 5683212.95 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3768, pruned_loss=0.1285, over 5697036.82 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3812, pruned_loss=0.1307, over 5669295.11 frames. ], batch size: 174, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:27:21,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5608, 1.6546, 1.8869, 1.4616], device='cuda:1'), covar=tensor([0.1371, 0.1671, 0.1025, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0697, 0.0836, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-05 00:27:44,454 INFO [train.py:968] (1/2) Epoch 9, batch 42650, giga_loss[loss=0.2723, simple_loss=0.3349, pruned_loss=0.1048, over 28505.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3778, pruned_loss=0.1284, over 5688910.77 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3768, pruned_loss=0.1283, over 5700949.91 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3789, pruned_loss=0.1292, over 5674092.78 frames. ], batch size: 85, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:27:55,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-05 00:28:05,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:28:21,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2844, 1.4356, 1.4521, 1.3288], device='cuda:1'), covar=tensor([0.1308, 0.1550, 0.1923, 0.1537], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0739, 0.0666, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 00:28:27,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4718, 3.6472, 1.6535, 1.3981], device='cuda:1'), covar=tensor([0.0892, 0.0344, 0.0814, 0.1354], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0507, 0.0332, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 00:28:32,334 INFO [optim.py:369] (1/2) 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 00:28:32,347 INFO [train.py:968] (1/2) Epoch 9, batch 42700, giga_loss[loss=0.2842, simple_loss=0.3581, pruned_loss=0.1052, over 28837.00 frames. ], tot_loss[loss=0.317, simple_loss=0.377, pruned_loss=0.1285, over 5687073.64 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3766, pruned_loss=0.128, over 5706825.92 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.378, pruned_loss=0.1294, over 5669259.79 frames. ], batch size: 186, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:28:53,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:29:14,179 INFO [train.py:968] (1/2) Epoch 9, batch 42750, giga_loss[loss=0.2884, simple_loss=0.3562, pruned_loss=0.1103, over 28733.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3772, pruned_loss=0.1298, over 5653044.07 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3769, pruned_loss=0.1286, over 5683573.12 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3777, pruned_loss=0.13, over 5657837.45 frames. ], batch size: 262, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:29:33,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2499, 4.0179, 3.7797, 1.8718], device='cuda:1'), covar=tensor([0.0591, 0.0783, 0.0785, 0.2115], device='cuda:1'), in_proj_covar=tensor([0.1016, 0.0956, 0.0838, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 00:29:45,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3378, 1.4563, 1.5858, 1.1265], device='cuda:1'), covar=tensor([0.1883, 0.2882, 0.1529, 0.1603], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0700, 0.0837, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-05 00:29:52,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=406989.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:29:52,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0172, 1.9431, 1.7954, 1.6752], device='cuda:1'), covar=tensor([0.1401, 0.1993, 0.1781, 0.1958], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0736, 0.0663, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 00:29:58,753 INFO [optim.py:369] (1/2) 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 00:29:58,766 INFO [train.py:968] (1/2) Epoch 9, batch 42800, giga_loss[loss=0.2781, simple_loss=0.3629, pruned_loss=0.09666, over 28812.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3758, pruned_loss=0.1282, over 5659900.09 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3761, pruned_loss=0.1281, over 5687849.59 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3771, pruned_loss=0.1288, over 5659133.61 frames. ], batch size: 199, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:30:07,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407005.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:30:12,615 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:16,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:20,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 00:30:34,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:42,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:30:46,794 INFO [train.py:968] (1/2) Epoch 9, batch 42850, giga_loss[loss=0.2742, simple_loss=0.348, pruned_loss=0.1002, over 28607.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3764, pruned_loss=0.1275, over 5665852.62 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.376, pruned_loss=0.1279, over 5689043.90 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3774, pruned_loss=0.1282, over 5663971.52 frames. ], batch size: 92, lr: 3.51e-03, grad_scale: 8.0
+2023-03-05 00:31:04,009 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:31:05,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:31:25,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4826, 4.3133, 4.0736, 2.2707], device='cuda:1'), covar=tensor([0.0520, 0.0674, 0.0721, 0.1871], device='cuda:1'), in_proj_covar=tensor([0.1020, 0.0957, 0.0842, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 00:31:33,376 INFO [train.py:968] (1/2) Epoch 9, batch 42900, libri_loss[loss=0.3169, simple_loss=0.3851, pruned_loss=0.1243, over 29265.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3769, pruned_loss=0.127, over 5669703.66 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3761, pruned_loss=0.1279, over 5690951.99 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3777, pruned_loss=0.1275, over 5666297.60 frames. ], batch size: 94, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:31:33,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:31:34,052 INFO [optim.py:369] (1/2) 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 00:32:06,359 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407132.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:32:08,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407135.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:32:18,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407145.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:32:20,724 INFO [train.py:968] (1/2) Epoch 9, batch 42950, giga_loss[loss=0.317, simple_loss=0.3839, pruned_loss=0.1251, over 29085.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3794, pruned_loss=0.1286, over 5671401.26 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3768, pruned_loss=0.1284, over 5684493.11 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3794, pruned_loss=0.1285, over 5673848.59 frames. ], batch size: 155, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:32:38,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407164.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:32:44,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7656, 2.7074, 1.7445, 0.8305], device='cuda:1'), covar=tensor([0.4802, 0.2334, 0.2671, 0.4724], device='cuda:1'), in_proj_covar=tensor([0.1522, 0.1455, 0.1472, 0.1241], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 00:33:08,770 INFO [train.py:968] (1/2) Epoch 9, batch 43000, giga_loss[loss=0.3045, simple_loss=0.3725, pruned_loss=0.1183, over 28753.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3826, pruned_loss=0.1317, over 5676648.20 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3763, pruned_loss=0.1281, over 5688646.97 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3831, pruned_loss=0.132, over 5674557.89 frames. ], batch size: 99, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:33:09,817 INFO [optim.py:369] (1/2) 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 00:33:57,913 INFO [train.py:968] (1/2) Epoch 9, batch 43050, giga_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 28921.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3842, pruned_loss=0.1339, over 5676207.77 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3767, pruned_loss=0.1284, over 5681150.94 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3844, pruned_loss=0.1339, over 5680577.41 frames. ], batch size: 174, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:34:38,360 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407288.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:34:43,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:34:49,742 INFO [train.py:968] (1/2) Epoch 9, batch 43100, giga_loss[loss=0.3748, simple_loss=0.4155, pruned_loss=0.1671, over 28831.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3857, pruned_loss=0.1364, over 5667013.33 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3764, pruned_loss=0.1281, over 5679780.36 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3865, pruned_loss=0.137, over 5671312.01 frames. ], batch size: 186, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:34:51,071 INFO [optim.py:369] (1/2) 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 00:35:12,959 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407320.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:35:41,285 INFO [train.py:968] (1/2) Epoch 9, batch 43150, giga_loss[loss=0.3307, simple_loss=0.3899, pruned_loss=0.1358, over 28716.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3873, pruned_loss=0.1385, over 5654394.44 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.376, pruned_loss=0.1279, over 5681401.63 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3885, pruned_loss=0.1394, over 5655988.19 frames. ], batch size: 262, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:36:09,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407380.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:36:23,075 INFO [train.py:968] (1/2) Epoch 9, batch 43200, giga_loss[loss=0.3306, simple_loss=0.383, pruned_loss=0.1392, over 28687.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3857, pruned_loss=0.137, over 5657757.08 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.376, pruned_loss=0.1277, over 5682510.01 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3869, pruned_loss=0.1382, over 5657286.98 frames. ], batch size: 119, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:36:25,194 INFO [optim.py:369] (1/2) 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 00:36:34,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:37:08,866 INFO [train.py:968] (1/2) Epoch 9, batch 43250, giga_loss[loss=0.2824, simple_loss=0.3526, pruned_loss=0.1061, over 28874.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3837, pruned_loss=0.1348, over 5664378.13 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3759, pruned_loss=0.1275, over 5684430.20 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.385, pruned_loss=0.1361, over 5661723.03 frames. ], batch size: 199, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:37:41,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:37:54,424 INFO [train.py:968] (1/2) Epoch 9, batch 43300, giga_loss[loss=0.3442, simple_loss=0.3987, pruned_loss=0.1448, over 29023.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3824, pruned_loss=0.1326, over 5659993.06 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3759, pruned_loss=0.1275, over 5686664.21 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3834, pruned_loss=0.1336, over 5655746.24 frames. ], batch size: 128, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:37:55,866 INFO [optim.py:369] (1/2) 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 00:38:18,943 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:38:19,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0890, 1.4967, 1.3889, 1.0443], device='cuda:1'), covar=tensor([0.1394, 0.2143, 0.1151, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0700, 0.0838, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-05 00:38:20,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407526.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:38:39,821 INFO [train.py:968] (1/2) Epoch 9, batch 43350, libri_loss[loss=0.3083, simple_loss=0.3655, pruned_loss=0.1256, over 28212.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3802, pruned_loss=0.1308, over 5650004.54 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3761, pruned_loss=0.1276, over 5678406.22 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3809, pruned_loss=0.1317, over 5654353.66 frames. ], batch size: 62, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:38:44,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=407553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:38:46,094 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:38:46,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=407556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:39:10,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=407585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:39:23,699 INFO [train.py:968] (1/2) Epoch 9, batch 43400, giga_loss[loss=0.3469, simple_loss=0.3862, pruned_loss=0.1538, over 26545.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3787, pruned_loss=0.1301, over 5642748.82 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3763, pruned_loss=0.1277, over 5658246.05 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3791, pruned_loss=0.1306, over 5664121.62 frames. ], batch size: 555, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:39:24,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 00:39:26,946 INFO [optim.py:369] (1/2) 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 00:39:43,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0111, 1.2211, 1.2415, 1.1073], device='cuda:1'), covar=tensor([0.1015, 0.0874, 0.1412, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0735, 0.0663, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 00:39:50,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2708, 1.5308, 1.3191, 1.0267], device='cuda:1'), covar=tensor([0.1675, 0.1492, 0.0997, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.1660, 0.1556, 0.1505, 0.1598], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 00:40:09,674 INFO [train.py:968] (1/2) Epoch 9, batch 43450, giga_loss[loss=0.311, simple_loss=0.3709, pruned_loss=0.1255, over 28752.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3783, pruned_loss=0.1306, over 5655610.20 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3758, pruned_loss=0.1273, over 5664858.74 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3793, pruned_loss=0.1316, over 5666298.45 frames. ], batch size: 119, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:40:49,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2221, 1.3149, 1.4246, 1.2763], device='cuda:1'), covar=tensor([0.1087, 0.0982, 0.1419, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0735, 0.0662, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 00:40:54,403 INFO [train.py:968] (1/2) Epoch 9, batch 43500, giga_loss[loss=0.453, simple_loss=0.4593, pruned_loss=0.2233, over 26522.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.382, pruned_loss=0.1332, over 5662162.10 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3762, pruned_loss=0.1277, over 5671679.20 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3825, pruned_loss=0.1338, over 5664526.35 frames. ], batch size: 555, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:40:56,838 INFO [optim.py:369] (1/2) 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 00:41:42,895 INFO [train.py:968] (1/2) Epoch 9, batch 43550, giga_loss[loss=0.3261, simple_loss=0.4042, pruned_loss=0.124, over 28999.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3861, pruned_loss=0.1337, over 5651296.23 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3762, pruned_loss=0.1278, over 5664220.29 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3864, pruned_loss=0.134, over 5660564.81 frames. ], batch size: 145, lr: 3.51e-03, grad_scale: 2.0
+2023-03-05 00:42:01,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-05 00:42:26,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 43600, giga_loss[loss=0.2985, simple_loss=0.3647, pruned_loss=0.1161, over 28858.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.387, pruned_loss=0.1328, over 5661059.57 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.1281, over 5668530.52 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3874, pruned_loss=0.1329, over 5664184.66 frames. ], batch size: 112, lr: 3.51e-03, grad_scale: 4.0
+2023-03-05 00:42:38,116 INFO [optim.py:369] (1/2) 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 00:43:24,524 INFO [train.py:968] (1/2) Epoch 9, batch 43650, giga_loss[loss=0.3661, simple_loss=0.4176, pruned_loss=0.1573, over 28962.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3902, pruned_loss=0.1357, over 5660324.20 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3763, pruned_loss=0.128, over 5672124.04 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3909, pruned_loss=0.1359, over 5659460.65 frames. ], batch size: 213, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:43:34,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=407859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:43:45,275 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:44:02,299 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 43700, giga_loss[loss=0.3703, simple_loss=0.4159, pruned_loss=0.1624, over 28940.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3895, pruned_loss=0.1351, over 5658894.40 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3765, pruned_loss=0.1281, over 5667024.35 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3902, pruned_loss=0.1355, over 5662030.35 frames. ], batch size: 213, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:44:09,616 INFO [optim.py:369] (1/2) 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 00:44:12,322 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-05 00:44:16,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407908.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:44:49,729 INFO [train.py:968] (1/2) Epoch 9, batch 43750, giga_loss[loss=0.3209, simple_loss=0.3853, pruned_loss=0.1283, over 29127.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3882, pruned_loss=0.1352, over 5665568.84 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3758, pruned_loss=0.1277, over 5674408.42 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.39, pruned_loss=0.1361, over 5660940.84 frames. ], batch size: 155, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:45:40,694 INFO [train.py:968] (1/2) Epoch 9, batch 43800, giga_loss[loss=0.347, simple_loss=0.3992, pruned_loss=0.1474, over 28701.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3863, pruned_loss=0.1347, over 5661455.79 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3758, pruned_loss=0.1277, over 5674408.42 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3877, pruned_loss=0.1354, over 5657853.75 frames. ], batch size: 242, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:45:40,877 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=407998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:45:42,992 INFO [optim.py:369] (1/2) 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 00:45:44,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:45:46,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:46:13,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:46:18,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 00:46:26,927 INFO [train.py:968] (1/2) Epoch 9, batch 43850, giga_loss[loss=0.3416, simple_loss=0.4007, pruned_loss=0.1412, over 28876.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3834, pruned_loss=0.1329, over 5668069.34 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1277, over 5678713.53 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3845, pruned_loss=0.1336, over 5661168.17 frames. ], batch size: 174, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:46:58,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-05 00:47:14,826 INFO [train.py:968] (1/2) Epoch 9, batch 43900, giga_loss[loss=0.3763, simple_loss=0.3881, pruned_loss=0.1822, over 23445.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3824, pruned_loss=0.133, over 5670651.89 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3761, pruned_loss=0.1278, over 5682336.12 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3832, pruned_loss=0.1335, over 5661958.03 frames. ], batch size: 705, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:47:17,722 INFO [optim.py:369] (1/2) 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 00:48:04,860 INFO [train.py:968] (1/2) Epoch 9, batch 43950, giga_loss[loss=0.305, simple_loss=0.3689, pruned_loss=0.1205, over 29047.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3835, pruned_loss=0.1339, over 5679624.28 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3757, pruned_loss=0.1274, over 5682941.41 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3847, pruned_loss=0.1348, over 5671660.21 frames. ], batch size: 186, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:48:23,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:48:52,131 INFO [train.py:968] (1/2) Epoch 9, batch 44000, giga_loss[loss=0.3436, simple_loss=0.3979, pruned_loss=0.1447, over 28688.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3849, pruned_loss=0.1359, over 5664268.73 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3755, pruned_loss=0.1273, over 5678515.68 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3862, pruned_loss=0.1368, over 5662200.88 frames. ], batch size: 284, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:48:54,171 INFO [optim.py:369] (1/2) 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 00:49:08,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3846, 3.1749, 2.9658, 1.9542], device='cuda:1'), covar=tensor([0.0681, 0.0847, 0.0788, 0.1753], device='cuda:1'), in_proj_covar=tensor([0.1014, 0.0961, 0.0837, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 00:49:37,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:49:38,155 INFO [train.py:968] (1/2) Epoch 9, batch 44050, giga_loss[loss=0.3095, simple_loss=0.3785, pruned_loss=0.1202, over 28898.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3823, pruned_loss=0.1343, over 5672031.73 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3754, pruned_loss=0.1273, over 5682644.08 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3835, pruned_loss=0.1353, over 5666163.79 frames. ], batch size: 199, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:49:51,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:50:05,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408283.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:50:16,886 INFO [train.py:968] (1/2) Epoch 9, batch 44100, giga_loss[loss=0.301, simple_loss=0.3756, pruned_loss=0.1131, over 28916.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3813, pruned_loss=0.1328, over 5667284.53 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3754, pruned_loss=0.127, over 5678736.46 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3826, pruned_loss=0.1341, over 5666441.53 frames. ], batch size: 174, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:50:19,085 INFO [optim.py:369] (1/2) 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 00:50:25,646 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:50:27,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:50:56,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:04,506 INFO [train.py:968] (1/2) Epoch 9, batch 44150, libri_loss[loss=0.3012, simple_loss=0.367, pruned_loss=0.1177, over 29539.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3831, pruned_loss=0.1333, over 5660201.15 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3753, pruned_loss=0.1269, over 5676435.78 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3845, pruned_loss=0.1346, over 5660660.82 frames. ], batch size: 81, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:51:12,022 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:27,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:38,578 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:42,204 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 44200, libri_loss[loss=0.3599, simple_loss=0.4079, pruned_loss=0.1559, over 29647.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3837, pruned_loss=0.1334, over 5676064.96 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.375, pruned_loss=0.1268, over 5685543.02 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3855, pruned_loss=0.1349, over 5667652.22 frames. ], batch size: 88, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:51:49,002 INFO [optim.py:369] (1/2) 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 00:51:56,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:51:59,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:52:08,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:52:12,172 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408426.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:52:15,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408429.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 00:52:26,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:52:32,080 INFO [train.py:968] (1/2) Epoch 9, batch 44250, libri_loss[loss=0.3545, simple_loss=0.4111, pruned_loss=0.149, over 29674.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3836, pruned_loss=0.1341, over 5664636.96 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3751, pruned_loss=0.1269, over 5682572.09 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.385, pruned_loss=0.1353, over 5660358.54 frames. ], batch size: 91, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:52:35,344 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-05 00:52:42,648 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408458.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:53:15,393 INFO [train.py:968] (1/2) Epoch 9, batch 44300, libri_loss[loss=0.2903, simple_loss=0.3598, pruned_loss=0.1104, over 29545.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3845, pruned_loss=0.1326, over 5672355.06 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3747, pruned_loss=0.1267, over 5685678.58 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3865, pruned_loss=0.1341, over 5665132.52 frames. ], batch size: 79, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:53:19,398 INFO [optim.py:369] (1/2) 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 00:53:28,242 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408511.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:53:31,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:53:33,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:53:55,980 INFO [train.py:968] (1/2) Epoch 9, batch 44350, giga_loss[loss=0.2949, simple_loss=0.3737, pruned_loss=0.108, over 28943.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.385, pruned_loss=0.1303, over 5688471.02 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3751, pruned_loss=0.1269, over 5693113.84 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3866, pruned_loss=0.1315, over 5675383.25 frames. ], batch size: 227, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 00:53:56,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:54:06,613 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:54:09,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5008, 1.7378, 1.6416, 1.3945], device='cuda:1'), covar=tensor([0.2212, 0.1627, 0.1230, 0.1653], device='cuda:1'), in_proj_covar=tensor([0.1654, 0.1548, 0.1496, 0.1601], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 00:54:14,055 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:54:46,913 INFO [train.py:968] (1/2) Epoch 9, batch 44400, giga_loss[loss=0.3218, simple_loss=0.3818, pruned_loss=0.1309, over 28544.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3872, pruned_loss=0.1304, over 5690646.95 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3752, pruned_loss=0.1268, over 5695540.03 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3886, pruned_loss=0.1314, over 5678165.33 frames. ], batch size: 336, lr: 3.50e-03, grad_scale: 8.0
+2023-03-05 00:54:50,220 INFO [optim.py:369] (1/2) 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 00:54:54,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6205, 1.5203, 1.2181, 1.1656], device='cuda:1'), covar=tensor([0.0639, 0.0506, 0.0953, 0.0954], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0442, 0.0498, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 00:55:05,878 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-05 00:55:35,109 INFO [train.py:968] (1/2) Epoch 9, batch 44450, giga_loss[loss=0.3391, simple_loss=0.4014, pruned_loss=0.1384, over 28790.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3904, pruned_loss=0.1338, over 5687784.72 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.375, pruned_loss=0.1267, over 5698407.58 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3919, pruned_loss=0.1348, over 5675197.32 frames. ], batch size: 199, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:55:41,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4183, 3.1600, 1.5301, 1.5428], device='cuda:1'), covar=tensor([0.0836, 0.0364, 0.0875, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0508, 0.0334, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 00:56:24,727 INFO [train.py:968] (1/2) Epoch 9, batch 44500, libri_loss[loss=0.2659, simple_loss=0.3315, pruned_loss=0.1002, over 29488.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.391, pruned_loss=0.1359, over 5682787.11 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3742, pruned_loss=0.1263, over 5704858.11 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3936, pruned_loss=0.1374, over 5666101.16 frames. ], batch size: 70, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:56:32,711 INFO [optim.py:369] (1/2) 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 00:56:55,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:57:13,164 INFO [train.py:968] (1/2) Epoch 9, batch 44550, giga_loss[loss=0.3812, simple_loss=0.4191, pruned_loss=0.1716, over 28971.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3928, pruned_loss=0.1385, over 5664751.66 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3749, pruned_loss=0.1269, over 5704454.16 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3945, pruned_loss=0.1394, over 5651227.98 frames. ], batch size: 213, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:57:13,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7101, 1.6202, 1.2082, 1.2743], device='cuda:1'), covar=tensor([0.0668, 0.0563, 0.1007, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0444, 0.0500, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 00:57:52,037 INFO [train.py:968] (1/2) Epoch 9, batch 44600, libri_loss[loss=0.3271, simple_loss=0.3886, pruned_loss=0.1328, over 29738.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3911, pruned_loss=0.1367, over 5671168.70 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3749, pruned_loss=0.1269, over 5704233.69 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3933, pruned_loss=0.138, over 5658758.72 frames. ], batch size: 87, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:57:56,719 INFO [optim.py:369] (1/2) 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 00:58:29,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=408841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:58:34,971 INFO [train.py:968] (1/2) Epoch 9, batch 44650, libri_loss[loss=0.2528, simple_loss=0.3194, pruned_loss=0.09307, over 29424.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3896, pruned_loss=0.1343, over 5674628.46 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.375, pruned_loss=0.1271, over 5706645.72 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3916, pruned_loss=0.1354, over 5661833.63 frames. ], batch size: 67, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:58:57,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2299, 1.1890, 3.7327, 3.1907], device='cuda:1'), covar=tensor([0.1487, 0.2530, 0.0412, 0.1631], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0577, 0.0837, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 00:58:59,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5238, 1.5648, 1.1711, 1.2442], device='cuda:1'), covar=tensor([0.0663, 0.0465, 0.0934, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0441, 0.0498, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 00:59:00,435 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=408874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:59:03,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=408877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:59:03,945 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 00:59:13,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408886.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 00:59:23,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3852, 1.2947, 4.7855, 3.5048], device='cuda:1'), covar=tensor([0.1709, 0.2602, 0.0394, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0640, 0.0580, 0.0840, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 00:59:23,701 INFO [train.py:968] (1/2) Epoch 9, batch 44700, giga_loss[loss=0.2968, simple_loss=0.3745, pruned_loss=0.1096, over 29026.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3909, pruned_loss=0.1334, over 5671476.88 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3747, pruned_loss=0.1268, over 5707777.50 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3928, pruned_loss=0.1344, over 5660407.01 frames. ], batch size: 155, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 00:59:28,535 INFO [optim.py:369] (1/2) 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 00:59:31,144 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=408906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 00:59:53,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7423, 1.6003, 1.7780, 1.4162], device='cuda:1'), covar=tensor([0.1724, 0.2659, 0.1315, 0.1469], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0704, 0.0841, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 00:59:58,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=408934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:00:04,998 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 44750, libri_loss[loss=0.316, simple_loss=0.3837, pruned_loss=0.1241, over 29556.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3915, pruned_loss=0.1342, over 5672555.94 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3749, pruned_loss=0.1271, over 5702915.54 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3934, pruned_loss=0.135, over 5667587.36 frames. ], batch size: 89, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:00:57,266 INFO [train.py:968] (1/2) Epoch 9, batch 44800, giga_loss[loss=0.2762, simple_loss=0.3505, pruned_loss=0.1009, over 28644.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3908, pruned_loss=0.1345, over 5676394.53 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3745, pruned_loss=0.1268, over 5704390.26 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3929, pruned_loss=0.1356, over 5670680.67 frames. ], batch size: 92, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:01:00,941 INFO [optim.py:369] (1/2) 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 01:01:22,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409029.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 01:01:24,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409032.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:01:37,450 INFO [train.py:968] (1/2) Epoch 9, batch 44850, giga_loss[loss=0.3141, simple_loss=0.3762, pruned_loss=0.126, over 28970.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3914, pruned_loss=0.1362, over 5643238.50 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.375, pruned_loss=0.1272, over 5678150.87 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3932, pruned_loss=0.137, over 5660688.68 frames. ], batch size: 227, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:01:45,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3965, 3.4549, 1.4966, 1.5623], device='cuda:1'), covar=tensor([0.0931, 0.0338, 0.0908, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0509, 0.0336, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 01:01:49,753 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409061.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:01:58,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-05 01:02:06,361 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:08,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:13,399 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:16,100 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 44900, giga_loss[loss=0.2824, simple_loss=0.3539, pruned_loss=0.1054, over 29035.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3892, pruned_loss=0.1363, over 5637830.78 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.375, pruned_loss=0.1271, over 5681018.54 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3911, pruned_loss=0.1372, over 5648676.76 frames. ], batch size: 106, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:02:31,897 INFO [optim.py:369] (1/2) 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 01:02:38,564 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 01:02:38,963 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:02:43,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8718, 3.6923, 3.5132, 1.7691], device='cuda:1'), covar=tensor([0.0588, 0.0684, 0.0720, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.1027, 0.0965, 0.0847, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 01:02:44,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:03:16,358 INFO [train.py:968] (1/2) Epoch 9, batch 44950, giga_loss[loss=0.269, simple_loss=0.3396, pruned_loss=0.09922, over 28550.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3863, pruned_loss=0.1346, over 5643452.25 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3753, pruned_loss=0.1272, over 5680935.74 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3878, pruned_loss=0.1354, over 5651605.15 frames. ], batch size: 60, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:04:04,860 INFO [train.py:968] (1/2) Epoch 9, batch 45000, giga_loss[loss=0.3512, simple_loss=0.3799, pruned_loss=0.1612, over 23534.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3841, pruned_loss=0.1337, over 5641931.10 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3754, pruned_loss=0.1274, over 5676765.98 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3854, pruned_loss=0.1344, over 5651236.13 frames. ], batch size: 705, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:04:04,861 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 01:04:13,178 INFO [train.py:1012] (1/2) Epoch 9, validation: loss=0.2212, simple_loss=0.3284, pruned_loss=0.05699, over 944034.00 frames. 
+2023-03-05 01:04:13,178 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 01:04:18,633 INFO [optim.py:369] (1/2) 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 01:04:28,904 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 9, batch 45050, giga_loss[loss=0.2922, simple_loss=0.3572, pruned_loss=0.1136, over 28943.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3841, pruned_loss=0.1341, over 5662214.36 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3754, pruned_loss=0.1274, over 5682025.91 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3853, pruned_loss=0.1347, over 5664348.45 frames. ], batch size: 186, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:05:34,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4987, 3.5191, 1.5989, 1.5803], device='cuda:1'), covar=tensor([0.0919, 0.0316, 0.0871, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0509, 0.0334, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 01:05:42,152 INFO [train.py:968] (1/2) Epoch 9, batch 45100, giga_loss[loss=0.3495, simple_loss=0.3986, pruned_loss=0.1502, over 27579.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3816, pruned_loss=0.1312, over 5667364.24 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3758, pruned_loss=0.1275, over 5687445.98 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3824, pruned_loss=0.1318, over 5664044.87 frames. ], batch size: 472, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:05:49,016 INFO [optim.py:369] (1/2) 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 01:06:25,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3760, 2.1061, 1.6502, 0.6663], device='cuda:1'), covar=tensor([0.3735, 0.2051, 0.2869, 0.4071], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1450, 0.1473, 0.1245], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 01:06:28,565 INFO [train.py:968] (1/2) Epoch 9, batch 45150, giga_loss[loss=0.2806, simple_loss=0.3617, pruned_loss=0.09973, over 28920.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.378, pruned_loss=0.1275, over 5667633.94 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3764, pruned_loss=0.1281, over 5690739.58 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3782, pruned_loss=0.1275, over 5661498.64 frames. ], batch size: 213, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:06:39,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:06:42,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:06:49,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=409369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:07:10,704 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:07:17,206 INFO [train.py:968] (1/2) Epoch 9, batch 45200, giga_loss[loss=0.2621, simple_loss=0.3416, pruned_loss=0.09125, over 28761.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3778, pruned_loss=0.1269, over 5669608.71 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3765, pruned_loss=0.1281, over 5693370.37 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3779, pruned_loss=0.1268, over 5661957.55 frames. ], batch size: 66, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:07:25,446 INFO [optim.py:369] (1/2) 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 01:07:32,232 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5123, 3.4223, 1.5112, 1.5807], device='cuda:1'), covar=tensor([0.0832, 0.0315, 0.0891, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0506, 0.0334, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 01:08:04,444 INFO [train.py:968] (1/2) Epoch 9, batch 45250, giga_loss[loss=0.3364, simple_loss=0.3942, pruned_loss=0.1393, over 28862.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3752, pruned_loss=0.1258, over 5650255.86 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3763, pruned_loss=0.1279, over 5685316.91 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3755, pruned_loss=0.1259, over 5650297.25 frames. ], batch size: 186, lr: 3.50e-03, grad_scale: 4.0
+2023-03-05 01:08:55,607 INFO [train.py:968] (1/2) Epoch 9, batch 45300, giga_loss[loss=0.3441, simple_loss=0.3948, pruned_loss=0.1467, over 27639.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3738, pruned_loss=0.1262, over 5591650.36 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3773, pruned_loss=0.1288, over 5632575.23 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.373, pruned_loss=0.1253, over 5638716.64 frames. ], batch size: 472, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:09:01,170 INFO [optim.py:369] (1/2) 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 01:09:23,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2927, 1.2344, 4.3718, 3.3229], device='cuda:1'), covar=tensor([0.1697, 0.2684, 0.0387, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0641, 0.0584, 0.0839, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 01:09:36,237 INFO [train.py:968] (1/2) Epoch 9, batch 45350, libri_loss[loss=0.3545, simple_loss=0.3994, pruned_loss=0.1548, over 19893.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3763, pruned_loss=0.1281, over 5537280.37 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3784, pruned_loss=0.1297, over 5564600.94 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3745, pruned_loss=0.1265, over 5637214.57 frames. ], batch size: 189, lr: 3.50e-03, grad_scale: 2.0
+2023-03-05 01:10:20,806 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-05 01:11:05,577 INFO [optim.py:369] (1/2) 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 01:11:09,590 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=409608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:11:43,031 INFO [train.py:968] (1/2) Epoch 10, batch 50, giga_loss[loss=0.348, simple_loss=0.4159, pruned_loss=0.14, over 28426.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3865, pruned_loss=0.1189, over 1266448.95 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3646, pruned_loss=0.1037, over 88054.62 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.388, pruned_loss=0.12, over 1196466.66 frames. ], batch size: 368, lr: 3.33e-03, grad_scale: 2.0
+2023-03-05 01:12:28,865 INFO [train.py:968] (1/2) Epoch 10, batch 100, giga_loss[loss=0.2698, simple_loss=0.3528, pruned_loss=0.09342, over 28870.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3757, pruned_loss=0.1151, over 2243564.72 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3439, pruned_loss=0.09337, over 316214.65 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3803, pruned_loss=0.1181, over 2039002.44 frames. ], batch size: 213, lr: 3.33e-03, grad_scale: 2.0
+2023-03-05 01:12:36,945 INFO [optim.py:369] (1/2) 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 01:13:12,345 INFO [train.py:968] (1/2) Epoch 10, batch 150, giga_loss[loss=0.2372, simple_loss=0.3099, pruned_loss=0.08223, over 28396.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3606, pruned_loss=0.1079, over 3007861.45 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3512, pruned_loss=0.09754, over 477243.98 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3624, pruned_loss=0.1096, over 2760598.80 frames. ], batch size: 65, lr: 3.33e-03, grad_scale: 2.0
+2023-03-05 01:13:13,586 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=409744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:13:52,954 INFO [train.py:968] (1/2) Epoch 10, batch 200, giga_loss[loss=0.2843, simple_loss=0.3367, pruned_loss=0.1159, over 26776.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3471, pruned_loss=0.1008, over 3609701.85 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3448, pruned_loss=0.09464, over 713925.69 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3488, pruned_loss=0.1024, over 3306000.44 frames. ], batch size: 555, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:14:01,819 INFO [optim.py:369] (1/2) 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 01:14:32,956 INFO [train.py:968] (1/2) Epoch 10, batch 250, giga_loss[loss=0.2101, simple_loss=0.2831, pruned_loss=0.06856, over 28447.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3357, pruned_loss=0.09464, over 4077241.76 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3453, pruned_loss=0.09451, over 837901.56 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3358, pruned_loss=0.09544, over 3793370.47 frames. ], batch size: 60, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:14:38,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4847, 2.0356, 1.6546, 1.6779], device='cuda:1'), covar=tensor([0.0733, 0.0264, 0.0294, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:1')
+2023-03-05 01:14:42,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-05 01:14:45,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4401, 1.5715, 1.5311, 1.4272], device='cuda:1'), covar=tensor([0.1350, 0.1627, 0.1596, 0.1464], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0738, 0.0662, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 01:14:53,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 01:15:07,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5146, 1.6342, 1.3611, 1.8918], device='cuda:1'), covar=tensor([0.2285, 0.2230, 0.2332, 0.2292], device='cuda:1'), in_proj_covar=tensor([0.1264, 0.0940, 0.1122, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 01:15:12,927 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=409887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:15:16,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=409890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:15:19,436 INFO [train.py:968] (1/2) Epoch 10, batch 300, giga_loss[loss=0.203, simple_loss=0.2841, pruned_loss=0.06096, over 28864.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3255, pruned_loss=0.09028, over 4416996.53 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3456, pruned_loss=0.09488, over 879041.69 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3248, pruned_loss=0.09052, over 4187846.87 frames. ], batch size: 174, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:15:30,794 INFO [optim.py:369] (1/2) 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 01:15:42,284 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=409919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:16:06,451 INFO [train.py:968] (1/2) Epoch 10, batch 350, giga_loss[loss=0.2049, simple_loss=0.2728, pruned_loss=0.06846, over 28934.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3173, pruned_loss=0.08653, over 4695918.57 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3448, pruned_loss=0.09455, over 952857.39 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3161, pruned_loss=0.08648, over 4498008.10 frames. ], batch size: 106, lr: 3.32e-03, grad_scale: 1.0
+2023-03-05 01:16:37,733 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 01:16:38,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=409983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:16:45,210 INFO [train.py:968] (1/2) Epoch 10, batch 400, giga_loss[loss=0.2017, simple_loss=0.279, pruned_loss=0.06224, over 29007.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.314, pruned_loss=0.08436, over 4915263.75 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3477, pruned_loss=0.0955, over 1111478.74 frames. ], giga_tot_loss[loss=0.2393, simple_loss=0.3113, pruned_loss=0.08368, over 4732709.73 frames. ], batch size: 145, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:16:56,603 INFO [optim.py:369] (1/2) 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 01:17:00,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=410010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:17:25,657 INFO [train.py:968] (1/2) Epoch 10, batch 450, giga_loss[loss=0.2469, simple_loss=0.311, pruned_loss=0.09142, over 29044.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3127, pruned_loss=0.08395, over 5083511.40 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3486, pruned_loss=0.09632, over 1264083.68 frames. ], giga_tot_loss[loss=0.2374, simple_loss=0.3091, pruned_loss=0.08283, over 4919549.09 frames. ], batch size: 106, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:17:54,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 01:18:09,590 INFO [train.py:968] (1/2) Epoch 10, batch 500, giga_loss[loss=0.2003, simple_loss=0.2731, pruned_loss=0.06381, over 28429.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3112, pruned_loss=0.08358, over 5213260.25 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3507, pruned_loss=0.09706, over 1331216.40 frames. ], giga_tot_loss[loss=0.2361, simple_loss=0.3074, pruned_loss=0.08237, over 5075474.44 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:18:22,629 INFO [optim.py:369] (1/2) 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 01:18:39,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:18:42,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:18:56,123 INFO [train.py:968] (1/2) Epoch 10, batch 550, giga_loss[loss=0.2159, simple_loss=0.2881, pruned_loss=0.07189, over 28802.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.308, pruned_loss=0.08169, over 5321842.56 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3502, pruned_loss=0.09699, over 1421913.18 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3043, pruned_loss=0.08045, over 5201253.79 frames. ], batch size: 99, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:19:07,194 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=410158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:19:13,310 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=410163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:19:24,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2378, 0.8444, 0.8430, 1.4581], device='cuda:1'), covar=tensor([0.0724, 0.0332, 0.0336, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:1')
+2023-03-05 01:19:30,310 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=410185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:19:30,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5608, 1.5655, 1.3186, 1.2217], device='cuda:1'), covar=tensor([0.0750, 0.0562, 0.1015, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0439, 0.0498, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 01:19:37,748 INFO [train.py:968] (1/2) Epoch 10, batch 600, giga_loss[loss=0.2403, simple_loss=0.3063, pruned_loss=0.08711, over 28686.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.307, pruned_loss=0.08143, over 5400311.27 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3498, pruned_loss=0.09766, over 1590945.25 frames. ], giga_tot_loss[loss=0.2309, simple_loss=0.3024, pruned_loss=0.07964, over 5289471.93 frames. ], batch size: 242, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:19:42,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-05 01:19:50,698 INFO [optim.py:369] (1/2) 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 01:20:23,773 INFO [train.py:968] (1/2) Epoch 10, batch 650, giga_loss[loss=0.1945, simple_loss=0.2689, pruned_loss=0.06007, over 29062.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3055, pruned_loss=0.08052, over 5467680.02 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.35, pruned_loss=0.09734, over 1694158.73 frames. ], giga_tot_loss[loss=0.2292, simple_loss=0.3007, pruned_loss=0.07879, over 5371889.79 frames. ], batch size: 136, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:21:10,119 INFO [train.py:968] (1/2) Epoch 10, batch 700, giga_loss[loss=0.1897, simple_loss=0.2677, pruned_loss=0.05584, over 28885.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3009, pruned_loss=0.07784, over 5519733.19 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3497, pruned_loss=0.09715, over 1736221.82 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.2968, pruned_loss=0.07633, over 5440749.32 frames. ], batch size: 112, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:21:21,174 INFO [optim.py:369] (1/2) 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 01:21:22,216 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 01:21:55,029 INFO [train.py:968] (1/2) Epoch 10, batch 750, giga_loss[loss=0.2008, simple_loss=0.2726, pruned_loss=0.06451, over 28761.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.2978, pruned_loss=0.07621, over 5553623.28 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3494, pruned_loss=0.09698, over 1817154.38 frames. ], giga_tot_loss[loss=0.2215, simple_loss=0.2937, pruned_loss=0.07469, over 5485973.87 frames. ], batch size: 262, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:22:34,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=410385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:22:41,527 INFO [train.py:968] (1/2) Epoch 10, batch 800, libri_loss[loss=0.2137, simple_loss=0.3016, pruned_loss=0.06288, over 29499.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.2963, pruned_loss=0.07597, over 5574459.06 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3494, pruned_loss=0.09694, over 1888765.28 frames. ], giga_tot_loss[loss=0.2206, simple_loss=0.2922, pruned_loss=0.07444, over 5523117.06 frames. ], batch size: 70, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:22:52,780 INFO [optim.py:369] (1/2) 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 01:23:04,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7297, 1.7162, 1.3300, 1.3835], device='cuda:1'), covar=tensor([0.0723, 0.0575, 0.0978, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0438, 0.0499, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 01:23:29,627 INFO [train.py:968] (1/2) Epoch 10, batch 850, giga_loss[loss=0.2841, simple_loss=0.3529, pruned_loss=0.1077, over 28675.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3061, pruned_loss=0.08145, over 5603800.21 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3504, pruned_loss=0.09733, over 1968337.91 frames. ], giga_tot_loss[loss=0.2306, simple_loss=0.3017, pruned_loss=0.0798, over 5557531.79 frames. ], batch size: 307, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:24:11,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4353, 1.5821, 1.3815, 1.3251], device='cuda:1'), covar=tensor([0.1603, 0.1538, 0.1230, 0.1452], device='cuda:1'), in_proj_covar=tensor([0.1645, 0.1542, 0.1486, 0.1585], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 01:24:20,047 INFO [train.py:968] (1/2) Epoch 10, batch 900, giga_loss[loss=0.3043, simple_loss=0.3766, pruned_loss=0.116, over 28973.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3211, pruned_loss=0.08947, over 5624994.80 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3508, pruned_loss=0.09727, over 2008054.03 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3173, pruned_loss=0.08813, over 5585739.43 frames. ], batch size: 136, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:24:29,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3485, 1.8385, 1.3382, 0.5768], device='cuda:1'), covar=tensor([0.2717, 0.1552, 0.2455, 0.3493], device='cuda:1'), in_proj_covar=tensor([0.1503, 0.1446, 0.1470, 0.1230], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 01:24:34,188 INFO [optim.py:369] (1/2) 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 01:24:50,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:24:54,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:25:00,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=410538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:25:04,328 INFO [train.py:968] (1/2) Epoch 10, batch 950, giga_loss[loss=0.3114, simple_loss=0.3895, pruned_loss=0.1167, over 29000.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3325, pruned_loss=0.09542, over 5629248.50 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3513, pruned_loss=0.09783, over 2062880.09 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.329, pruned_loss=0.09415, over 5597709.05 frames. ], batch size: 227, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:25:17,978 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=410560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:25:17,992 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 1000, giga_loss[loss=0.246, simple_loss=0.3223, pruned_loss=0.08489, over 23648.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3393, pruned_loss=0.09786, over 5642177.89 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3516, pruned_loss=0.09836, over 2193771.03 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3359, pruned_loss=0.09662, over 5610050.53 frames. ], batch size: 705, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:25:55,132 INFO [optim.py:369] (1/2) 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 01:26:15,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0645, 1.0343, 4.0764, 3.1950], device='cuda:1'), covar=tensor([0.1760, 0.2810, 0.0375, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0569, 0.0823, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 01:26:20,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0447, 1.1960, 3.1844, 2.8045], device='cuda:1'), covar=tensor([0.1488, 0.2470, 0.0453, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.0629, 0.0569, 0.0823, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 01:26:27,060 INFO [train.py:968] (1/2) Epoch 10, batch 1050, giga_loss[loss=0.2884, simple_loss=0.3672, pruned_loss=0.1048, over 28904.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3421, pruned_loss=0.09735, over 5660752.83 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3519, pruned_loss=0.09827, over 2248858.87 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3393, pruned_loss=0.0964, over 5632431.93 frames. ], batch size: 227, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:27:01,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:03,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:12,223 INFO [train.py:968] (1/2) Epoch 10, batch 1100, giga_loss[loss=0.2687, simple_loss=0.3428, pruned_loss=0.09731, over 29065.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3434, pruned_loss=0.09739, over 5659505.33 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3519, pruned_loss=0.09824, over 2339236.31 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3409, pruned_loss=0.09662, over 5631802.38 frames. ], batch size: 128, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:27:20,418 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=410703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:23,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=410706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:23,317 INFO [optim.py:369] (1/2) 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 01:27:29,861 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=410735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:27:57,422 INFO [train.py:968] (1/2) Epoch 10, batch 1150, giga_loss[loss=0.2614, simple_loss=0.3394, pruned_loss=0.09169, over 28942.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3462, pruned_loss=0.09951, over 5659274.65 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3514, pruned_loss=0.098, over 2381889.57 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3444, pruned_loss=0.099, over 5643738.76 frames. ], batch size: 164, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:28:41,054 INFO [train.py:968] (1/2) Epoch 10, batch 1200, giga_loss[loss=0.2806, simple_loss=0.3616, pruned_loss=0.0998, over 29010.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3503, pruned_loss=0.1022, over 5660697.79 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3527, pruned_loss=0.09861, over 2539223.71 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3481, pruned_loss=0.1017, over 5651185.25 frames. ], batch size: 164, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:28:45,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-05 01:28:52,880 INFO [optim.py:369] (1/2) 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 01:29:24,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-05 01:29:24,812 INFO [train.py:968] (1/2) Epoch 10, batch 1250, giga_loss[loss=0.2903, simple_loss=0.3702, pruned_loss=0.1052, over 28872.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3534, pruned_loss=0.1042, over 5671805.60 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3527, pruned_loss=0.09857, over 2588995.73 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3517, pruned_loss=0.1039, over 5661498.94 frames. ], batch size: 174, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:29:30,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-05 01:29:50,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-05 01:30:10,361 INFO [train.py:968] (1/2) Epoch 10, batch 1300, giga_loss[loss=0.2966, simple_loss=0.3683, pruned_loss=0.1124, over 28277.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.357, pruned_loss=0.1054, over 5678568.77 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.353, pruned_loss=0.09866, over 2605796.75 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3556, pruned_loss=0.1052, over 5669313.86 frames. ], batch size: 368, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:30:14,065 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:30:50,760 INFO [train.py:968] (1/2) Epoch 10, batch 1350, giga_loss[loss=0.2559, simple_loss=0.3451, pruned_loss=0.08336, over 29009.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3592, pruned_loss=0.1061, over 5676241.20 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3534, pruned_loss=0.09886, over 2653773.08 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3579, pruned_loss=0.1059, over 5667107.32 frames. ], batch size: 155, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:31:29,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8618, 3.5901, 3.4567, 1.6013], device='cuda:1'), covar=tensor([0.0651, 0.0884, 0.0839, 0.2281], device='cuda:1'), in_proj_covar=tensor([0.0977, 0.0914, 0.0805, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 01:31:31,554 INFO [train.py:968] (1/2) Epoch 10, batch 1400, giga_loss[loss=0.2855, simple_loss=0.3628, pruned_loss=0.1041, over 29040.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3596, pruned_loss=0.1054, over 5693611.18 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3529, pruned_loss=0.09892, over 2781254.89 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.359, pruned_loss=0.1055, over 5678520.07 frames. ], batch size: 128, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:31:43,194 INFO [optim.py:369] (1/2) 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 01:32:01,642 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:32:05,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4729, 3.4397, 1.5812, 1.5676], device='cuda:1'), covar=tensor([0.0937, 0.0227, 0.0861, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0498, 0.0330, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 01:32:16,218 INFO [train.py:968] (1/2) Epoch 10, batch 1450, giga_loss[loss=0.2633, simple_loss=0.3444, pruned_loss=0.0911, over 28715.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3594, pruned_loss=0.1044, over 5691098.72 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3531, pruned_loss=0.09883, over 2823902.62 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3589, pruned_loss=0.1046, over 5679786.66 frames. ], batch size: 92, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:32:16,400 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:32:29,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3466, 1.4395, 1.3180, 1.3601], device='cuda:1'), covar=tensor([0.0781, 0.0326, 0.0311, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0055, 0.0049, 0.0083], device='cuda:1')
+2023-03-05 01:32:35,012 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:32:56,075 INFO [train.py:968] (1/2) Epoch 10, batch 1500, giga_loss[loss=0.3025, simple_loss=0.3785, pruned_loss=0.1133, over 28646.00 frames. ], tot_loss[loss=0.28, simple_loss=0.357, pruned_loss=0.1015, over 5693285.57 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3533, pruned_loss=0.09877, over 2888753.41 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3566, pruned_loss=0.1018, over 5690543.35 frames. ], batch size: 85, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:33:05,443 INFO [optim.py:369] (1/2) 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 01:33:36,465 INFO [train.py:968] (1/2) Epoch 10, batch 1550, giga_loss[loss=0.2274, simple_loss=0.3218, pruned_loss=0.06655, over 28414.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3561, pruned_loss=0.1007, over 5697300.35 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3536, pruned_loss=0.09868, over 2948409.14 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3557, pruned_loss=0.101, over 5691633.75 frames. ], batch size: 71, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:33:44,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4795, 1.7389, 1.7916, 1.3698], device='cuda:1'), covar=tensor([0.1746, 0.2189, 0.1349, 0.1566], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0706, 0.0856, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 01:34:15,408 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 1600, giga_loss[loss=0.2768, simple_loss=0.3524, pruned_loss=0.1006, over 28645.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3583, pruned_loss=0.1036, over 5685918.11 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3542, pruned_loss=0.09894, over 3025027.66 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3579, pruned_loss=0.1038, over 5686120.61 frames. ], batch size: 92, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:34:22,984 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 01:34:32,051 INFO [optim.py:369] (1/2) 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 01:35:07,608 INFO [train.py:968] (1/2) Epoch 10, batch 1650, giga_loss[loss=0.2842, simple_loss=0.3533, pruned_loss=0.1075, over 29051.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3608, pruned_loss=0.1078, over 5697324.49 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3541, pruned_loss=0.09885, over 3081373.49 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3606, pruned_loss=0.1081, over 5694608.55 frames. ], batch size: 136, lr: 3.32e-03, grad_scale: 8.0
+2023-03-05 01:35:07,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3793, 1.9301, 1.4089, 0.6294], device='cuda:1'), covar=tensor([0.2865, 0.1456, 0.2118, 0.3524], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1446, 0.1476, 0.1244], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 01:35:26,058 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 01:35:35,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411274.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:35:52,306 INFO [train.py:968] (1/2) Epoch 10, batch 1700, giga_loss[loss=0.2523, simple_loss=0.3333, pruned_loss=0.08565, over 28592.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3608, pruned_loss=0.1089, over 5703374.49 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3538, pruned_loss=0.09867, over 3134019.32 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3609, pruned_loss=0.1095, over 5700712.71 frames. ], batch size: 60, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:36:05,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6660, 1.9079, 1.6181, 1.4260], device='cuda:1'), covar=tensor([0.1724, 0.1244, 0.1157, 0.1418], device='cuda:1'), in_proj_covar=tensor([0.1631, 0.1530, 0.1482, 0.1584], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 01:36:06,751 INFO [optim.py:369] (1/2) 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 01:36:39,619 INFO [train.py:968] (1/2) Epoch 10, batch 1750, giga_loss[loss=0.2619, simple_loss=0.3389, pruned_loss=0.09242, over 28960.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3587, pruned_loss=0.1089, over 5686380.34 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3535, pruned_loss=0.09852, over 3161377.97 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.359, pruned_loss=0.1095, over 5682754.00 frames. ], batch size: 174, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:36:55,525 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:12,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-05 01:37:21,558 INFO [train.py:968] (1/2) Epoch 10, batch 1800, giga_loss[loss=0.266, simple_loss=0.34, pruned_loss=0.09595, over 28862.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3573, pruned_loss=0.1086, over 5689991.14 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3531, pruned_loss=0.0983, over 3215864.82 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3578, pruned_loss=0.1093, over 5682999.83 frames. ], batch size: 145, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:37:27,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:29,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:32,472 INFO [optim.py:369] (1/2) 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 01:37:39,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 01:37:40,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:37:41,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411420.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 01:38:01,405 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:38:01,869 INFO [train.py:968] (1/2) Epoch 10, batch 1850, libri_loss[loss=0.2417, simple_loss=0.3276, pruned_loss=0.07785, over 29536.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3567, pruned_loss=0.1076, over 5684192.07 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3534, pruned_loss=0.09866, over 3286598.32 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3571, pruned_loss=0.1084, over 5680251.04 frames. ], batch size: 80, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:38:06,271 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411449.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 01:38:15,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3522, 1.6068, 1.2906, 1.1387], device='cuda:1'), covar=tensor([0.1923, 0.1484, 0.1174, 0.1590], device='cuda:1'), in_proj_covar=tensor([0.1644, 0.1543, 0.1496, 0.1597], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 01:38:46,388 INFO [train.py:968] (1/2) Epoch 10, batch 1900, giga_loss[loss=0.2902, simple_loss=0.3598, pruned_loss=0.1103, over 27894.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3534, pruned_loss=0.1045, over 5687204.29 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3523, pruned_loss=0.09788, over 3384750.01 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3543, pruned_loss=0.1058, over 5680424.62 frames. ], batch size: 412, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:39:04,684 INFO [optim.py:369] (1/2) 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 01:39:32,613 INFO [train.py:968] (1/2) Epoch 10, batch 1950, giga_loss[loss=0.3026, simple_loss=0.3637, pruned_loss=0.1207, over 28541.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3498, pruned_loss=0.1028, over 5673405.31 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3523, pruned_loss=0.09804, over 3438257.63 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3506, pruned_loss=0.1039, over 5670817.17 frames. ], batch size: 85, lr: 3.32e-03, grad_scale: 2.0
+2023-03-05 01:39:34,203 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:39:52,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:39:52,472 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:39:55,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:06,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=411577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:06,649 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:15,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:17,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:21,983 INFO [train.py:968] (1/2) Epoch 10, batch 2000, giga_loss[loss=0.2469, simple_loss=0.3047, pruned_loss=0.09454, over 23525.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3443, pruned_loss=0.1001, over 5668646.71 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3527, pruned_loss=0.09831, over 3485788.94 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3445, pruned_loss=0.1008, over 5663615.43 frames. ], batch size: 705, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:40:22,172 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:40:35,658 INFO [optim.py:369] (1/2) 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 01:40:43,290 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:41:01,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2406, 1.3026, 1.1788, 0.9522], device='cuda:1'), covar=tensor([0.0725, 0.0465, 0.0998, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0441, 0.0499, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 01:41:07,380 INFO [train.py:968] (1/2) Epoch 10, batch 2050, giga_loss[loss=0.2275, simple_loss=0.286, pruned_loss=0.08449, over 23543.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3381, pruned_loss=0.09643, over 5664680.56 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3522, pruned_loss=0.09788, over 3543670.49 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3383, pruned_loss=0.09724, over 5657228.39 frames. ], batch size: 705, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:41:49,755 INFO [train.py:968] (1/2) Epoch 10, batch 2100, giga_loss[loss=0.2643, simple_loss=0.3376, pruned_loss=0.09549, over 28709.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3394, pruned_loss=0.09718, over 5667594.29 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3522, pruned_loss=0.09784, over 3658033.46 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3389, pruned_loss=0.09783, over 5651390.18 frames. ], batch size: 119, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:41:58,227 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:00,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:01,365 INFO [optim.py:369] (1/2) 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 01:42:24,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:25,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:42:28,937 INFO [train.py:968] (1/2) Epoch 10, batch 2150, giga_loss[loss=0.2662, simple_loss=0.3389, pruned_loss=0.09679, over 28857.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3412, pruned_loss=0.09765, over 5674749.25 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.352, pruned_loss=0.09767, over 3702067.53 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3406, pruned_loss=0.09826, over 5667234.94 frames. ], batch size: 99, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:42:43,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6306, 1.7390, 1.1490, 1.2077], device='cuda:1'), covar=tensor([0.0850, 0.0628, 0.1172, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0440, 0.0498, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 01:42:58,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:43:12,805 INFO [train.py:968] (1/2) Epoch 10, batch 2200, giga_loss[loss=0.3141, simple_loss=0.3767, pruned_loss=0.1257, over 27674.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3395, pruned_loss=0.09683, over 5670479.14 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3523, pruned_loss=0.09786, over 3714174.42 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3388, pruned_loss=0.09719, over 5670923.95 frames. ], batch size: 472, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:43:27,401 INFO [optim.py:369] (1/2) 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 01:43:52,428 INFO [train.py:968] (1/2) Epoch 10, batch 2250, giga_loss[loss=0.2292, simple_loss=0.3142, pruned_loss=0.07209, over 29020.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3365, pruned_loss=0.0953, over 5677482.28 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3533, pruned_loss=0.09832, over 3747352.68 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3352, pruned_loss=0.09528, over 5682792.82 frames. ], batch size: 164, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:44:26,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:44:29,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:44:30,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 01:44:35,441 INFO [train.py:968] (1/2) Epoch 10, batch 2300, giga_loss[loss=0.2654, simple_loss=0.3333, pruned_loss=0.09879, over 28852.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3347, pruned_loss=0.09427, over 5688455.74 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3544, pruned_loss=0.09873, over 3797581.30 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3326, pruned_loss=0.09394, over 5690522.58 frames. ], batch size: 262, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:44:49,428 INFO [optim.py:369] (1/2) 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 01:44:52,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:01,390 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=411923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:03,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=411926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:17,912 INFO [train.py:968] (1/2) Epoch 10, batch 2350, giga_loss[loss=0.2849, simple_loss=0.3459, pruned_loss=0.1119, over 28316.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3315, pruned_loss=0.09271, over 5693882.19 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3548, pruned_loss=0.09887, over 3827732.76 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3293, pruned_loss=0.09228, over 5693653.95 frames. ], batch size: 65, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:45:24,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=411952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:26,944 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=411955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:45:59,452 INFO [train.py:968] (1/2) Epoch 10, batch 2400, giga_loss[loss=0.2307, simple_loss=0.3093, pruned_loss=0.07608, over 29002.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3284, pruned_loss=0.09134, over 5689563.00 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3554, pruned_loss=0.09915, over 3838470.68 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3262, pruned_loss=0.0908, over 5695968.11 frames. ], batch size: 164, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:46:13,600 INFO [optim.py:369] (1/2) 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 01:46:37,269 INFO [train.py:968] (1/2) Epoch 10, batch 2450, giga_loss[loss=0.2591, simple_loss=0.3311, pruned_loss=0.0936, over 28903.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3271, pruned_loss=0.09091, over 5699314.32 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.356, pruned_loss=0.09913, over 3869579.63 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3246, pruned_loss=0.09038, over 5701343.76 frames. ], batch size: 145, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:46:53,448 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 01:47:17,720 INFO [train.py:968] (1/2) Epoch 10, batch 2500, libri_loss[loss=0.2891, simple_loss=0.3793, pruned_loss=0.09943, over 27791.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3248, pruned_loss=0.0897, over 5709139.33 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3561, pruned_loss=0.09887, over 3908153.54 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3222, pruned_loss=0.08927, over 5708948.16 frames. ], batch size: 116, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:47:18,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:47:21,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:47:29,918 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:47:44,587 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:47:47,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-05 01:47:57,764 INFO [train.py:968] (1/2) Epoch 10, batch 2550, giga_loss[loss=0.2319, simple_loss=0.311, pruned_loss=0.0764, over 28618.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3241, pruned_loss=0.08889, over 5719590.57 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3572, pruned_loss=0.09912, over 3957461.51 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3206, pruned_loss=0.08817, over 5715293.71 frames. ], batch size: 336, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:48:10,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2727, 1.4725, 1.4375, 1.5099], device='cuda:1'), covar=tensor([0.0771, 0.0328, 0.0303, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0077, 0.0055, 0.0049, 0.0084], device='cuda:1')
+2023-03-05 01:48:36,390 INFO [train.py:968] (1/2) Epoch 10, batch 2600, libri_loss[loss=0.2573, simple_loss=0.3422, pruned_loss=0.08622, over 29571.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3242, pruned_loss=0.08878, over 5725711.14 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3582, pruned_loss=0.09947, over 4033719.68 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3194, pruned_loss=0.08756, over 5715576.18 frames. ], batch size: 75, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:48:38,686 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:48:49,020 INFO [optim.py:369] (1/2) 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 01:49:09,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-05 01:49:15,096 INFO [train.py:968] (1/2) Epoch 10, batch 2650, giga_loss[loss=0.2679, simple_loss=0.3353, pruned_loss=0.1003, over 28884.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.326, pruned_loss=0.09004, over 5729466.82 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3582, pruned_loss=0.09939, over 4105662.09 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3209, pruned_loss=0.08874, over 5716622.05 frames. ], batch size: 186, lr: 3.32e-03, grad_scale: 4.0
+2023-03-05 01:50:00,932 INFO [train.py:968] (1/2) Epoch 10, batch 2700, giga_loss[loss=0.2562, simple_loss=0.3257, pruned_loss=0.09339, over 28886.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3301, pruned_loss=0.09278, over 5726187.35 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3581, pruned_loss=0.09926, over 4123659.16 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3259, pruned_loss=0.09178, over 5714459.42 frames. ], batch size: 106, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:50:01,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:50:13,896 INFO [optim.py:369] (1/2) 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 01:50:45,985 INFO [train.py:968] (1/2) Epoch 10, batch 2750, giga_loss[loss=0.2468, simple_loss=0.3217, pruned_loss=0.0859, over 28611.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3351, pruned_loss=0.09584, over 5722890.36 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3577, pruned_loss=0.09915, over 4159066.59 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3316, pruned_loss=0.09501, over 5710349.58 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:51:31,613 INFO [train.py:968] (1/2) Epoch 10, batch 2800, giga_loss[loss=0.3705, simple_loss=0.4202, pruned_loss=0.1604, over 28629.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3437, pruned_loss=0.1017, over 5697699.96 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3583, pruned_loss=0.0995, over 4189231.74 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.34, pruned_loss=0.1008, over 5702774.96 frames. ], batch size: 307, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:51:47,900 INFO [optim.py:369] (1/2) 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 01:52:13,201 INFO [train.py:968] (1/2) Epoch 10, batch 2850, giga_loss[loss=0.3295, simple_loss=0.3831, pruned_loss=0.1379, over 28764.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.35, pruned_loss=0.1053, over 5695094.65 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3584, pruned_loss=0.09957, over 4242650.39 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3467, pruned_loss=0.1047, over 5697472.79 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:52:23,068 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:52:48,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=412482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:52:59,914 INFO [train.py:968] (1/2) Epoch 10, batch 2900, giga_loss[loss=0.3146, simple_loss=0.3809, pruned_loss=0.1242, over 28493.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3543, pruned_loss=0.1063, over 5703475.13 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3585, pruned_loss=0.09979, over 4299647.53 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3513, pruned_loss=0.1059, over 5699406.57 frames. ], batch size: 60, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:53:12,692 INFO [optim.py:369] (1/2) 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 01:53:18,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 01:53:38,475 INFO [train.py:968] (1/2) Epoch 10, batch 2950, giga_loss[loss=0.3179, simple_loss=0.3889, pruned_loss=0.1234, over 28283.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3576, pruned_loss=0.1076, over 5712331.15 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3573, pruned_loss=0.09922, over 4384614.25 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3558, pruned_loss=0.108, over 5700494.10 frames. ], batch size: 368, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:54:04,058 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-05 01:54:09,196 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 3000, giga_loss[loss=0.2495, simple_loss=0.3378, pruned_loss=0.08056, over 28847.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.364, pruned_loss=0.1126, over 5687218.20 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3574, pruned_loss=0.09934, over 4413209.49 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3626, pruned_loss=0.113, over 5675344.93 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:54:28,022 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 01:54:36,632 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 01:54:49,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5258, 1.7844, 1.4796, 1.7233], device='cuda:1'), covar=tensor([0.2019, 0.1786, 0.1793, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.0933, 0.1113, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 01:54:50,573 INFO [optim.py:369] (1/2) 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 01:55:01,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:55:03,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:55:17,811 INFO [train.py:968] (1/2) Epoch 10, batch 3050, giga_loss[loss=0.2617, simple_loss=0.3313, pruned_loss=0.09607, over 27522.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3588, pruned_loss=0.1086, over 5699406.00 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3571, pruned_loss=0.09933, over 4464501.24 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3581, pruned_loss=0.1094, over 5683510.83 frames. ], batch size: 472, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:55:28,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:55:41,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=412670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:56:00,440 INFO [train.py:968] (1/2) Epoch 10, batch 3100, giga_loss[loss=0.3012, simple_loss=0.3614, pruned_loss=0.1205, over 28195.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3551, pruned_loss=0.1055, over 5705750.22 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3568, pruned_loss=0.09915, over 4499011.69 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3547, pruned_loss=0.1064, over 5689581.63 frames. ], batch size: 368, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:56:06,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5556, 1.7863, 1.8813, 1.4326], device='cuda:1'), covar=tensor([0.1719, 0.2103, 0.1299, 0.1506], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0698, 0.0849, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 01:56:17,212 INFO [optim.py:369] (1/2) 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 01:56:19,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:56:21,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:56:22,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-05 01:56:42,607 INFO [train.py:968] (1/2) Epoch 10, batch 3150, giga_loss[loss=0.2658, simple_loss=0.346, pruned_loss=0.09284, over 29031.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3537, pruned_loss=0.104, over 5718102.55 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.357, pruned_loss=0.09947, over 4534129.36 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3531, pruned_loss=0.1046, over 5700525.77 frames. ], batch size: 155, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 01:56:45,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:19,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=412783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:26,843 INFO [train.py:968] (1/2) Epoch 10, batch 3200, giga_loss[loss=0.3034, simple_loss=0.3792, pruned_loss=0.1138, over 28889.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3545, pruned_loss=0.1039, over 5718900.22 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3566, pruned_loss=0.09914, over 4567974.99 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3543, pruned_loss=0.1047, over 5700670.98 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 01:57:44,186 INFO [optim.py:369] (1/2) 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 01:57:45,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:47,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=412816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:57:58,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=412829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:58:06,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4340, 1.4853, 1.5654, 1.4654], device='cuda:1'), covar=tensor([0.1468, 0.1772, 0.1878, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0728, 0.0658, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 01:58:11,233 INFO [train.py:968] (1/2) Epoch 10, batch 3250, giga_loss[loss=0.2977, simple_loss=0.3717, pruned_loss=0.1119, over 28671.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3573, pruned_loss=0.1056, over 5719381.51 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3565, pruned_loss=0.09901, over 4581296.87 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3571, pruned_loss=0.1064, over 5703366.16 frames. ], batch size: 262, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 01:58:12,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=412845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 01:58:41,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2520, 1.5339, 1.1967, 1.3597], device='cuda:1'), covar=tensor([0.2363, 0.2228, 0.2483, 0.2007], device='cuda:1'), in_proj_covar=tensor([0.1263, 0.0936, 0.1116, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 01:58:45,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2327, 2.5767, 1.1729, 1.3556], device='cuda:1'), covar=tensor([0.0896, 0.0300, 0.0831, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0493, 0.0328, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0028, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 01:58:46,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7740, 4.6231, 4.3298, 1.8726], device='cuda:1'), covar=tensor([0.0474, 0.0578, 0.0726, 0.2117], device='cuda:1'), in_proj_covar=tensor([0.0991, 0.0917, 0.0811, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 01:58:55,566 INFO [train.py:968] (1/2) Epoch 10, batch 3300, giga_loss[loss=0.2848, simple_loss=0.359, pruned_loss=0.1053, over 28901.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3587, pruned_loss=0.1068, over 5713773.89 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3563, pruned_loss=0.09887, over 4607111.14 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5698138.65 frames. ], batch size: 199, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 01:59:11,055 INFO [optim.py:369] (1/2) 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 01:59:37,506 INFO [train.py:968] (1/2) Epoch 10, batch 3350, giga_loss[loss=0.3242, simple_loss=0.3922, pruned_loss=0.1281, over 28936.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3596, pruned_loss=0.1079, over 5703544.18 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3562, pruned_loss=0.09888, over 4634227.86 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3598, pruned_loss=0.1088, over 5696815.59 frames. ], batch size: 227, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:00:00,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=412972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:00:03,538 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 3400, libri_loss[loss=0.2726, simple_loss=0.3599, pruned_loss=0.09263, over 29756.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3602, pruned_loss=0.1083, over 5716718.99 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3559, pruned_loss=0.09865, over 4686847.06 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3607, pruned_loss=0.1096, over 5706425.46 frames. ], batch size: 87, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:00:28,535 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=413004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:00:30,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 02:00:34,001 INFO [optim.py:369] (1/2) 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 02:00:47,650 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3015, 1.4761, 1.4637, 1.4088], device='cuda:1'), covar=tensor([0.1293, 0.1364, 0.1779, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0729, 0.0658, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 02:00:49,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5500, 2.2556, 1.5061, 0.6758], device='cuda:1'), covar=tensor([0.4956, 0.2184, 0.2574, 0.4472], device='cuda:1'), in_proj_covar=tensor([0.1505, 0.1427, 0.1466, 0.1231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 02:01:02,866 INFO [train.py:968] (1/2) Epoch 10, batch 3450, giga_loss[loss=0.273, simple_loss=0.3515, pruned_loss=0.0973, over 28705.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3599, pruned_loss=0.1078, over 5722081.09 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3559, pruned_loss=0.09846, over 4723560.50 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3605, pruned_loss=0.1093, over 5712958.74 frames. ], batch size: 284, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:01:43,230 INFO [train.py:968] (1/2) Epoch 10, batch 3500, giga_loss[loss=0.2725, simple_loss=0.3513, pruned_loss=0.09682, over 28984.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3595, pruned_loss=0.1071, over 5721724.30 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3547, pruned_loss=0.09775, over 4768513.30 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.361, pruned_loss=0.1091, over 5709456.10 frames. ], batch size: 128, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:01:58,728 INFO [optim.py:369] (1/2) 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 02:02:24,241 INFO [train.py:968] (1/2) Epoch 10, batch 3550, giga_loss[loss=0.2674, simple_loss=0.3486, pruned_loss=0.09312, over 29061.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3601, pruned_loss=0.1066, over 5720931.02 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3544, pruned_loss=0.09755, over 4792531.15 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3616, pruned_loss=0.1086, over 5710488.99 frames. ], batch size: 128, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:02:36,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=413158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:03:06,121 INFO [train.py:968] (1/2) Epoch 10, batch 3600, giga_loss[loss=0.2525, simple_loss=0.3376, pruned_loss=0.08366, over 28964.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3605, pruned_loss=0.1062, over 5729162.84 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3539, pruned_loss=0.09723, over 4830989.80 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3624, pruned_loss=0.1084, over 5714877.64 frames. ], batch size: 155, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:03:19,270 INFO [optim.py:369] (1/2) 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 02:03:43,454 INFO [train.py:968] (1/2) Epoch 10, batch 3650, giga_loss[loss=0.2657, simple_loss=0.3343, pruned_loss=0.09853, over 28397.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3587, pruned_loss=0.1054, over 5722075.90 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3538, pruned_loss=0.09735, over 4847429.63 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3604, pruned_loss=0.1072, over 5715085.30 frames. ], batch size: 60, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:04:26,920 INFO [train.py:968] (1/2) Epoch 10, batch 3700, giga_loss[loss=0.2587, simple_loss=0.3351, pruned_loss=0.09119, over 28962.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3571, pruned_loss=0.1051, over 5723204.20 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3538, pruned_loss=0.09736, over 4873080.24 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3585, pruned_loss=0.1067, over 5714114.09 frames. ], batch size: 106, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:04:33,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6295, 1.7015, 1.4355, 1.8433], device='cuda:1'), covar=tensor([0.2406, 0.2305, 0.2513, 0.2173], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.0938, 0.1118, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 02:04:34,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=413301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:04:37,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=413304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:04:42,781 INFO [optim.py:369] (1/2) 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 02:04:54,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9780, 1.2978, 5.4080, 3.5484], device='cuda:1'), covar=tensor([0.1314, 0.2441, 0.0278, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0636, 0.0572, 0.0823, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 02:04:58,383 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=413333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:05:06,674 INFO [train.py:968] (1/2) Epoch 10, batch 3750, giga_loss[loss=0.239, simple_loss=0.3193, pruned_loss=0.07937, over 28821.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3553, pruned_loss=0.1042, over 5716899.46 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3536, pruned_loss=0.09724, over 4874477.13 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3565, pruned_loss=0.1056, over 5716161.95 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:05:49,143 INFO [train.py:968] (1/2) Epoch 10, batch 3800, giga_loss[loss=0.2914, simple_loss=0.3645, pruned_loss=0.1091, over 28903.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3548, pruned_loss=0.1036, over 5720972.94 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3534, pruned_loss=0.09707, over 4900087.51 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3561, pruned_loss=0.1051, over 5723729.22 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:06:02,519 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 3850, giga_loss[loss=0.2696, simple_loss=0.3504, pruned_loss=0.0944, over 29036.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3559, pruned_loss=0.1045, over 5723337.26 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3536, pruned_loss=0.09724, over 4931204.15 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3568, pruned_loss=0.1058, over 5722922.25 frames. ], batch size: 155, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:06:48,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6519, 1.6639, 1.2439, 1.3041], device='cuda:1'), covar=tensor([0.0806, 0.0601, 0.1059, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0435, 0.0498, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:07:06,204 INFO [train.py:968] (1/2) Epoch 10, batch 3900, giga_loss[loss=0.2833, simple_loss=0.3663, pruned_loss=0.1002, over 28590.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3569, pruned_loss=0.1048, over 5720844.39 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3538, pruned_loss=0.0974, over 4953151.43 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3575, pruned_loss=0.1059, over 5718308.09 frames. ], batch size: 262, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:07:23,643 INFO [optim.py:369] (1/2) 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 02:07:47,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9368, 3.0501, 2.1622, 0.9769], device='cuda:1'), covar=tensor([0.4587, 0.1638, 0.2248, 0.4405], device='cuda:1'), in_proj_covar=tensor([0.1489, 0.1397, 0.1444, 0.1217], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 02:07:50,196 INFO [train.py:968] (1/2) Epoch 10, batch 3950, libri_loss[loss=0.2547, simple_loss=0.3283, pruned_loss=0.09053, over 29571.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3548, pruned_loss=0.1026, over 5722408.86 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3539, pruned_loss=0.09751, over 4976550.15 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3553, pruned_loss=0.1035, over 5716247.88 frames. ], batch size: 74, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:08:18,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5784, 4.3602, 4.0912, 1.8305], device='cuda:1'), covar=tensor([0.0424, 0.0578, 0.0679, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.0990, 0.0924, 0.0814, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 02:08:28,869 INFO [train.py:968] (1/2) Epoch 10, batch 4000, giga_loss[loss=0.2725, simple_loss=0.3404, pruned_loss=0.1023, over 28400.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3543, pruned_loss=0.1027, over 5724070.79 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.354, pruned_loss=0.09773, over 5000703.08 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3546, pruned_loss=0.1034, over 5717432.96 frames. ], batch size: 65, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:08:30,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5461, 1.7983, 1.5414, 1.3643], device='cuda:1'), covar=tensor([0.1752, 0.1269, 0.1007, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.1603, 0.1504, 0.1468, 0.1576], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:08:43,946 INFO [optim.py:369] (1/2) 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 02:09:09,558 INFO [train.py:968] (1/2) Epoch 10, batch 4050, giga_loss[loss=0.2561, simple_loss=0.3298, pruned_loss=0.09121, over 28976.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3523, pruned_loss=0.1022, over 5716954.01 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3541, pruned_loss=0.09771, over 5014085.60 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3525, pruned_loss=0.1028, over 5709459.88 frames. ], batch size: 106, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:09:33,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3813, 1.4961, 1.6190, 1.4656], device='cuda:1'), covar=tensor([0.1278, 0.1330, 0.1349, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0727, 0.0660, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 02:09:49,462 INFO [train.py:968] (1/2) Epoch 10, batch 4100, giga_loss[loss=0.2389, simple_loss=0.3027, pruned_loss=0.08752, over 23326.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3494, pruned_loss=0.1008, over 5713450.83 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3542, pruned_loss=0.09774, over 5026865.39 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3494, pruned_loss=0.1013, over 5705332.05 frames. ], batch size: 705, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:10:03,650 INFO [optim.py:369] (1/2) 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 02:10:28,268 INFO [train.py:968] (1/2) Epoch 10, batch 4150, giga_loss[loss=0.2563, simple_loss=0.3357, pruned_loss=0.0885, over 28168.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3484, pruned_loss=0.1005, over 5712797.29 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3543, pruned_loss=0.09754, over 5057430.98 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3482, pruned_loss=0.1012, over 5700314.74 frames. ], batch size: 77, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:10:41,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 02:11:08,661 INFO [train.py:968] (1/2) Epoch 10, batch 4200, giga_loss[loss=0.2697, simple_loss=0.3352, pruned_loss=0.1021, over 28681.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3484, pruned_loss=0.1014, over 5712498.46 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3541, pruned_loss=0.09751, over 5065762.19 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3483, pruned_loss=0.1019, over 5701068.22 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:11:13,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2260, 1.3239, 1.1064, 0.9917], device='cuda:1'), covar=tensor([0.0641, 0.0403, 0.0941, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0435, 0.0500, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:11:17,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3701, 1.6345, 1.3591, 1.4652], device='cuda:1'), covar=tensor([0.2054, 0.1891, 0.2003, 0.1721], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.0937, 0.1113, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 02:11:18,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 02:11:23,441 INFO [optim.py:369] (1/2) 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 02:11:49,543 INFO [train.py:968] (1/2) Epoch 10, batch 4250, giga_loss[loss=0.2532, simple_loss=0.3337, pruned_loss=0.08631, over 28838.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3468, pruned_loss=0.1012, over 5701436.67 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3541, pruned_loss=0.09754, over 5068510.28 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3467, pruned_loss=0.1017, over 5699248.14 frames. ], batch size: 145, lr: 3.31e-03, grad_scale: 2.0
+2023-03-05 02:11:59,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8086, 1.6357, 1.3799, 1.2962], device='cuda:1'), covar=tensor([0.0660, 0.0632, 0.0937, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0436, 0.0499, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:12:15,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0014, 1.1218, 0.9417, 0.9004], device='cuda:1'), covar=tensor([0.1382, 0.1575, 0.0929, 0.1231], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1512, 0.1480, 0.1588], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:12:33,339 INFO [train.py:968] (1/2) Epoch 10, batch 4300, giga_loss[loss=0.2716, simple_loss=0.3329, pruned_loss=0.1051, over 28724.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3429, pruned_loss=0.09897, over 5711021.70 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3543, pruned_loss=0.0976, over 5080883.72 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3425, pruned_loss=0.09932, over 5706672.07 frames. ], batch size: 92, lr: 3.31e-03, grad_scale: 2.0
+2023-03-05 02:12:40,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9597, 1.6409, 5.4119, 3.6102], device='cuda:1'), covar=tensor([0.1267, 0.2154, 0.0279, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0631, 0.0571, 0.0819, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 02:12:49,337 INFO [optim.py:369] (1/2) 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 02:13:11,832 INFO [train.py:968] (1/2) Epoch 10, batch 4350, giga_loss[loss=0.237, simple_loss=0.3164, pruned_loss=0.07879, over 28908.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3427, pruned_loss=0.09975, over 5705343.05 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3543, pruned_loss=0.09779, over 5099946.17 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3419, pruned_loss=0.09991, over 5705048.34 frames. ], batch size: 136, lr: 3.31e-03, grad_scale: 2.0
+2023-03-05 02:13:49,013 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=413989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:13:51,814 INFO [train.py:968] (1/2) Epoch 10, batch 4400, giga_loss[loss=0.293, simple_loss=0.3545, pruned_loss=0.1158, over 28987.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3407, pruned_loss=0.09886, over 5711482.04 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3541, pruned_loss=0.09771, over 5114935.11 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.34, pruned_loss=0.09909, over 5708577.73 frames. ], batch size: 213, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:14:08,069 INFO [optim.py:369] (1/2) 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 02:14:26,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0205, 0.9733, 3.7477, 3.0667], device='cuda:1'), covar=tensor([0.1651, 0.2654, 0.0340, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0633, 0.0571, 0.0820, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 02:14:34,373 INFO [train.py:968] (1/2) Epoch 10, batch 4450, giga_loss[loss=0.279, simple_loss=0.347, pruned_loss=0.1055, over 28431.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3395, pruned_loss=0.09756, over 5709854.60 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3543, pruned_loss=0.09791, over 5121223.20 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3387, pruned_loss=0.0976, over 5707272.73 frames. ], batch size: 60, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:15:20,037 INFO [train.py:968] (1/2) Epoch 10, batch 4500, giga_loss[loss=0.2419, simple_loss=0.3278, pruned_loss=0.07798, over 28986.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3434, pruned_loss=0.09952, over 5703076.25 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3542, pruned_loss=0.09775, over 5132736.21 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3427, pruned_loss=0.09971, over 5698520.24 frames. ], batch size: 136, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:15:23,867 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:15:38,182 INFO [optim.py:369] (1/2) 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 02:16:02,988 INFO [train.py:968] (1/2) Epoch 10, batch 4550, giga_loss[loss=0.2678, simple_loss=0.3508, pruned_loss=0.09237, over 28836.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3455, pruned_loss=0.09976, over 5707098.75 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3544, pruned_loss=0.09781, over 5140698.83 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3448, pruned_loss=0.09987, over 5701664.56 frames. ], batch size: 199, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:16:37,581 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:16:41,731 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:16:43,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5165, 1.5837, 1.2502, 1.1891], device='cuda:1'), covar=tensor([0.0728, 0.0490, 0.0930, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0435, 0.0498, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:16:49,938 INFO [train.py:968] (1/2) Epoch 10, batch 4600, giga_loss[loss=0.2981, simple_loss=0.3813, pruned_loss=0.1074, over 28590.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3475, pruned_loss=0.1003, over 5702476.64 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3543, pruned_loss=0.09787, over 5163544.66 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3467, pruned_loss=0.1004, over 5692138.32 frames. ], batch size: 307, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:17:09,237 INFO [optim.py:369] (1/2) 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 02:17:34,751 INFO [train.py:968] (1/2) Epoch 10, batch 4650, giga_loss[loss=0.3224, simple_loss=0.3802, pruned_loss=0.1324, over 26744.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3485, pruned_loss=0.1009, over 5699854.13 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3543, pruned_loss=0.09787, over 5174655.68 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3478, pruned_loss=0.101, over 5688740.86 frames. ], batch size: 555, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:18:16,925 INFO [train.py:968] (1/2) Epoch 10, batch 4700, giga_loss[loss=0.2565, simple_loss=0.3252, pruned_loss=0.09393, over 28541.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3473, pruned_loss=0.1003, over 5705365.16 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3541, pruned_loss=0.09794, over 5188933.48 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3468, pruned_loss=0.1004, over 5695353.68 frames. ], batch size: 78, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:18:33,888 INFO [optim.py:369] (1/2) 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 02:19:00,188 INFO [train.py:968] (1/2) Epoch 10, batch 4750, giga_loss[loss=0.3079, simple_loss=0.3633, pruned_loss=0.1262, over 28642.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3488, pruned_loss=0.1019, over 5701448.87 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3542, pruned_loss=0.09811, over 5195780.82 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3482, pruned_loss=0.1019, over 5692042.15 frames. ], batch size: 60, lr: 3.31e-03, grad_scale: 4.0
+2023-03-05 02:19:15,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:19:40,005 INFO [train.py:968] (1/2) Epoch 10, batch 4800, giga_loss[loss=0.2922, simple_loss=0.3632, pruned_loss=0.1107, over 27951.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3509, pruned_loss=0.1036, over 5695467.43 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3546, pruned_loss=0.09833, over 5207138.18 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3501, pruned_loss=0.1036, over 5691513.88 frames. ], batch size: 412, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:20:00,307 INFO [optim.py:369] (1/2) 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 02:20:24,858 INFO [train.py:968] (1/2) Epoch 10, batch 4850, giga_loss[loss=0.2665, simple_loss=0.3444, pruned_loss=0.09428, over 28804.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3541, pruned_loss=0.1053, over 5697484.05 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3547, pruned_loss=0.09848, over 5214081.25 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3533, pruned_loss=0.1051, over 5692613.46 frames. ], batch size: 119, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:20:38,175 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:20:47,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:20:56,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:20:59,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1419, 1.3773, 1.2195, 1.0461], device='cuda:1'), covar=tensor([0.1822, 0.1498, 0.1001, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1538, 0.1506, 0.1606], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:21:03,975 INFO [train.py:968] (1/2) Epoch 10, batch 4900, giga_loss[loss=0.3204, simple_loss=0.3889, pruned_loss=0.1259, over 28879.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3561, pruned_loss=0.1055, over 5714077.96 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3547, pruned_loss=0.09839, over 5249857.77 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3555, pruned_loss=0.1058, over 5701244.29 frames. ], batch size: 186, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:21:15,305 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:17,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:19,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9402, 2.9211, 1.9321, 0.9887], device='cuda:1'), covar=tensor([0.4735, 0.1928, 0.2797, 0.4406], device='cuda:1'), in_proj_covar=tensor([0.1500, 0.1421, 0.1465, 0.1231], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 02:21:20,753 INFO [optim.py:369] (1/2) 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 02:21:42,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=414539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:47,068 INFO [train.py:968] (1/2) Epoch 10, batch 4950, giga_loss[loss=0.2803, simple_loss=0.3547, pruned_loss=0.103, over 28709.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3574, pruned_loss=0.1059, over 5716513.28 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3551, pruned_loss=0.0986, over 5264520.63 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3566, pruned_loss=0.1062, over 5703401.57 frames. ], batch size: 99, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:21:53,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:21:57,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:22:03,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6910, 1.9367, 1.6640, 1.5191], device='cuda:1'), covar=tensor([0.1487, 0.1262, 0.1081, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1538, 0.1505, 0.1604], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:22:18,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414585.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:22:26,703 INFO [train.py:968] (1/2) Epoch 10, batch 5000, giga_loss[loss=0.3181, simple_loss=0.3923, pruned_loss=0.1219, over 28903.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3575, pruned_loss=0.1056, over 5723730.75 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3551, pruned_loss=0.0985, over 5277523.89 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.357, pruned_loss=0.1061, over 5709695.83 frames. ], batch size: 174, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:22:43,829 INFO [optim.py:369] (1/2) 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 02:22:44,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:22:47,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:22:48,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-05 02:22:48,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0494, 1.1344, 3.6564, 2.9761], device='cuda:1'), covar=tensor([0.1618, 0.2554, 0.0391, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0576, 0.0828, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:23:07,270 INFO [train.py:968] (1/2) Epoch 10, batch 5050, giga_loss[loss=0.4045, simple_loss=0.4286, pruned_loss=0.1902, over 26647.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3577, pruned_loss=0.1057, over 5726658.27 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3558, pruned_loss=0.09876, over 5292669.93 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3567, pruned_loss=0.106, over 5714127.51 frames. ], batch size: 555, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:23:10,033 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=414648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:24,274 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=414666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:23:47,349 INFO [train.py:968] (1/2) Epoch 10, batch 5100, giga_loss[loss=0.2527, simple_loss=0.326, pruned_loss=0.0897, over 28373.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3559, pruned_loss=0.1047, over 5721163.48 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3565, pruned_loss=0.09912, over 5298512.83 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3546, pruned_loss=0.1047, over 5711884.81 frames. ], batch size: 71, lr: 3.31e-03, grad_scale: 8.0
+2023-03-05 02:23:48,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:52,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:54,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:23:56,004 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:24:17,900 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=414728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:24:22,535 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 5150, giga_loss[loss=0.2459, simple_loss=0.3257, pruned_loss=0.08302, over 29044.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3526, pruned_loss=0.1028, over 5725270.81 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3563, pruned_loss=0.09901, over 5301673.27 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3516, pruned_loss=0.1029, over 5717160.12 frames. ], batch size: 128, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:25:09,749 INFO [train.py:968] (1/2) Epoch 10, batch 5200, giga_loss[loss=0.2379, simple_loss=0.3181, pruned_loss=0.07887, over 28878.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.35, pruned_loss=0.1012, over 5722291.75 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3574, pruned_loss=0.0997, over 5309874.94 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3482, pruned_loss=0.1008, over 5718977.93 frames. ], batch size: 186, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:25:18,588 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 02:25:19,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5873, 1.7536, 1.5485, 1.3265], device='cuda:1'), covar=tensor([0.2293, 0.1725, 0.1487, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.1633, 0.1552, 0.1515, 0.1616], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:25:29,190 INFO [optim.py:369] (1/2) 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 02:25:41,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:25:48,366 INFO [train.py:968] (1/2) Epoch 10, batch 5250, libri_loss[loss=0.2688, simple_loss=0.3406, pruned_loss=0.09852, over 29553.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3495, pruned_loss=0.1008, over 5725559.40 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3577, pruned_loss=0.1, over 5332943.28 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3476, pruned_loss=0.1003, over 5716567.76 frames. ], batch size: 79, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:25:52,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3253, 1.6283, 1.2630, 1.5502], device='cuda:1'), covar=tensor([0.0733, 0.0292, 0.0318, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0076, 0.0054, 0.0049, 0.0083], device='cuda:1')
+2023-03-05 02:26:00,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:26:25,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5953, 3.9143, 1.6823, 1.6788], device='cuda:1'), covar=tensor([0.0850, 0.0267, 0.0855, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0502, 0.0332, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 02:26:27,595 INFO [train.py:968] (1/2) Epoch 10, batch 5300, libri_loss[loss=0.38, simple_loss=0.4212, pruned_loss=0.1694, over 20204.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3521, pruned_loss=0.1011, over 5709255.46 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3583, pruned_loss=0.1006, over 5345862.42 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3497, pruned_loss=0.1002, over 5706498.20 frames. ], batch size: 187, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:26:48,194 INFO [optim.py:369] (1/2) 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 02:27:09,731 INFO [train.py:968] (1/2) Epoch 10, batch 5350, giga_loss[loss=0.2685, simple_loss=0.3493, pruned_loss=0.09383, over 28596.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3528, pruned_loss=0.1014, over 5706766.79 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3583, pruned_loss=0.1006, over 5359594.79 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3508, pruned_loss=0.1006, over 5700292.51 frames. ], batch size: 307, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:27:13,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4865, 1.7799, 1.7893, 1.3340], device='cuda:1'), covar=tensor([0.1588, 0.1993, 0.1327, 0.1443], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0695, 0.0846, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 02:27:22,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=414960.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:27:36,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=414977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:38,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=414980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:46,244 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-05 02:27:48,628 INFO [train.py:968] (1/2) Epoch 10, batch 5400, giga_loss[loss=0.273, simple_loss=0.3436, pruned_loss=0.1011, over 28949.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3522, pruned_loss=0.1019, over 5715046.60 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3585, pruned_loss=0.1008, over 5375550.45 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3503, pruned_loss=0.1012, over 5704982.59 frames. ], batch size: 164, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:27:58,049 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:59,772 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:27:59,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:28:02,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:28:09,129 INFO [optim.py:369] (1/2) 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 02:28:11,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1982, 1.2588, 4.0861, 3.3418], device='cuda:1'), covar=tensor([0.1646, 0.2481, 0.0357, 0.0716], device='cuda:1'), in_proj_covar=tensor([0.0632, 0.0572, 0.0829, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:28:18,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9764, 1.8086, 1.3425, 1.5201], device='cuda:1'), covar=tensor([0.0666, 0.0655, 0.0984, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0345, 0.0437, 0.0494, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:28:23,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:28:29,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=415041.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:28:32,498 INFO [train.py:968] (1/2) Epoch 10, batch 5450, giga_loss[loss=0.3615, simple_loss=0.3976, pruned_loss=0.1627, over 24107.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3518, pruned_loss=0.1033, over 5705577.06 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3587, pruned_loss=0.1008, over 5380215.13 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.35, pruned_loss=0.1028, over 5697830.97 frames. ], batch size: 705, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:28:33,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-05 02:28:54,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4597, 1.7498, 1.4330, 1.5922], device='cuda:1'), covar=tensor([0.2175, 0.2175, 0.2418, 0.2001], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.0935, 0.1112, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 02:29:13,294 INFO [train.py:968] (1/2) Epoch 10, batch 5500, giga_loss[loss=0.2807, simple_loss=0.3562, pruned_loss=0.1026, over 28895.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3506, pruned_loss=0.1041, over 5710708.69 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3587, pruned_loss=0.1008, over 5382822.05 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3491, pruned_loss=0.1037, over 5703699.59 frames. ], batch size: 145, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:29:22,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415103.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:29:25,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415106.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:29:29,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-05 02:29:32,832 INFO [optim.py:369] (1/2) 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 02:29:50,139 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415135.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:29:55,763 INFO [train.py:968] (1/2) Epoch 10, batch 5550, giga_loss[loss=0.247, simple_loss=0.3133, pruned_loss=0.09034, over 28759.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3479, pruned_loss=0.1033, over 5702191.01 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3586, pruned_loss=0.1007, over 5385814.03 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3466, pruned_loss=0.1031, over 5701670.50 frames. ], batch size: 92, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:30:32,307 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415184.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:30:37,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415187.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 02:30:41,870 INFO [train.py:968] (1/2) Epoch 10, batch 5600, giga_loss[loss=0.3287, simple_loss=0.3875, pruned_loss=0.1349, over 28326.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3465, pruned_loss=0.1027, over 5709119.50 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3585, pruned_loss=0.1007, over 5393647.98 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3454, pruned_loss=0.1025, over 5705947.61 frames. ], batch size: 368, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:31:00,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:31:00,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415216.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 02:31:02,027 INFO [optim.py:369] (1/2) 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 02:31:23,679 INFO [train.py:968] (1/2) Epoch 10, batch 5650, giga_loss[loss=0.2461, simple_loss=0.3245, pruned_loss=0.08379, over 28891.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3432, pruned_loss=0.1012, over 5716508.93 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3588, pruned_loss=0.1009, over 5398658.98 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.342, pruned_loss=0.1009, over 5712407.58 frames. ], batch size: 174, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:31:55,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7118, 1.6441, 1.3714, 1.3685], device='cuda:1'), covar=tensor([0.0742, 0.0630, 0.0975, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0438, 0.0495, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:32:02,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7560, 4.5537, 4.3383, 2.3516], device='cuda:1'), covar=tensor([0.0494, 0.0744, 0.0889, 0.1704], device='cuda:1'), in_proj_covar=tensor([0.0994, 0.0926, 0.0818, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 02:32:04,537 INFO [train.py:968] (1/2) Epoch 10, batch 5700, giga_loss[loss=0.2373, simple_loss=0.3157, pruned_loss=0.07944, over 28924.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3379, pruned_loss=0.09805, over 5718775.58 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3589, pruned_loss=0.1009, over 5410138.52 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3364, pruned_loss=0.09783, over 5712411.43 frames. ], batch size: 186, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:32:16,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3112, 1.5332, 1.3677, 1.1575], device='cuda:1'), covar=tensor([0.2022, 0.1631, 0.1121, 0.1702], device='cuda:1'), in_proj_covar=tensor([0.1651, 0.1553, 0.1511, 0.1621], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:32:24,642 INFO [optim.py:369] (1/2) 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 02:32:31,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5595, 3.5789, 1.5491, 1.6453], device='cuda:1'), covar=tensor([0.0816, 0.0242, 0.0846, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0503, 0.0333, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 02:32:47,204 INFO [train.py:968] (1/2) Epoch 10, batch 5750, giga_loss[loss=0.2464, simple_loss=0.3252, pruned_loss=0.08376, over 28995.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3373, pruned_loss=0.09779, over 5712552.17 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3589, pruned_loss=0.101, over 5410789.69 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3358, pruned_loss=0.09747, over 5711629.46 frames. ], batch size: 155, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:33:05,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1584, 0.8777, 0.8889, 1.4089], device='cuda:1'), covar=tensor([0.0731, 0.0366, 0.0348, 0.0778], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0077, 0.0055, 0.0050, 0.0084], device='cuda:1')
+2023-03-05 02:33:07,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:33:14,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=415380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:33:25,643 INFO [train.py:968] (1/2) Epoch 10, batch 5800, giga_loss[loss=0.2856, simple_loss=0.3503, pruned_loss=0.1105, over 28641.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3394, pruned_loss=0.09855, over 5722855.49 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3587, pruned_loss=0.101, over 5425117.89 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3378, pruned_loss=0.09823, over 5717308.11 frames. ], batch size: 92, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:33:44,085 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 5850, giga_loss[loss=0.2778, simple_loss=0.3434, pruned_loss=0.1061, over 28903.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3431, pruned_loss=0.09979, over 5729476.89 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3586, pruned_loss=0.101, over 5441857.94 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3414, pruned_loss=0.09948, over 5719118.94 frames. ], batch size: 66, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:34:17,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 02:34:20,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0856, 1.1750, 3.5203, 3.0993], device='cuda:1'), covar=tensor([0.1592, 0.2622, 0.0413, 0.1479], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0576, 0.0830, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:34:36,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9847, 1.0999, 3.6710, 3.0123], device='cuda:1'), covar=tensor([0.1714, 0.2651, 0.0398, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0576, 0.0829, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:34:45,991 INFO [train.py:968] (1/2) Epoch 10, batch 5900, giga_loss[loss=0.2836, simple_loss=0.3471, pruned_loss=0.11, over 23806.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3463, pruned_loss=0.1013, over 5712028.96 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3581, pruned_loss=0.1009, over 5439329.22 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.345, pruned_loss=0.1012, over 5711686.01 frames. ], batch size: 705, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:35:06,856 INFO [optim.py:369] (1/2) 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 02:35:13,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:35:14,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:35:15,946 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 02:35:32,253 INFO [train.py:968] (1/2) Epoch 10, batch 5950, giga_loss[loss=0.2459, simple_loss=0.3247, pruned_loss=0.08355, over 28945.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3493, pruned_loss=0.1026, over 5714405.02 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3582, pruned_loss=0.1009, over 5443695.10 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3481, pruned_loss=0.1024, over 5712577.68 frames. ], batch size: 128, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:35:41,887 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:36:14,072 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 6000, libri_loss[loss=0.3128, simple_loss=0.3885, pruned_loss=0.1186, over 29389.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3522, pruned_loss=0.1047, over 5709478.35 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3582, pruned_loss=0.1009, over 5449316.08 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3513, pruned_loss=0.1046, over 5706472.56 frames. ], batch size: 92, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:36:17,410 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 02:36:22,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8995, 1.1425, 3.3865, 2.9998], device='cuda:1'), covar=tensor([0.1893, 0.2873, 0.0492, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0639, 0.0578, 0.0833, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:36:25,634 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 02:36:48,494 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 6050, giga_loss[loss=0.2996, simple_loss=0.3742, pruned_loss=0.1125, over 29063.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3592, pruned_loss=0.1107, over 5690896.96 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3581, pruned_loss=0.101, over 5443253.28 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3584, pruned_loss=0.1108, over 5703593.89 frames. ], batch size: 155, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:37:41,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-05 02:37:50,508 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 6100, giga_loss[loss=0.3311, simple_loss=0.3944, pruned_loss=0.134, over 28912.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3644, pruned_loss=0.1152, over 5683934.31 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.358, pruned_loss=0.1012, over 5458019.77 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.364, pruned_loss=0.1155, over 5689619.08 frames. ], batch size: 213, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:38:06,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5738, 3.3706, 3.2073, 1.9815], device='cuda:1'), covar=tensor([0.0623, 0.0874, 0.0774, 0.1725], device='cuda:1'), in_proj_covar=tensor([0.1008, 0.0935, 0.0825, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 02:38:12,010 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 02:38:19,697 INFO [optim.py:369] (1/2) 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 02:38:35,232 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:38:36,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1907, 0.9085, 0.9352, 1.4293], device='cuda:1'), covar=tensor([0.0696, 0.0402, 0.0330, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0116, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0077, 0.0055, 0.0050, 0.0084], device='cuda:1')
+2023-03-05 02:38:37,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:38:42,549 INFO [train.py:968] (1/2) Epoch 10, batch 6150, giga_loss[loss=0.3117, simple_loss=0.3797, pruned_loss=0.1218, over 28950.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.12, over 5676953.86 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3582, pruned_loss=0.1014, over 5465900.02 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3707, pruned_loss=0.1203, over 5678467.00 frames. ], batch size: 213, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:38:44,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=415746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:39:05,627 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=415766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:39:20,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1928, 1.4968, 1.1830, 0.5000], device='cuda:1'), covar=tensor([0.1532, 0.0995, 0.1521, 0.2705], device='cuda:1'), in_proj_covar=tensor([0.1509, 0.1426, 0.1474, 0.1233], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 02:39:32,916 INFO [train.py:968] (1/2) Epoch 10, batch 6200, giga_loss[loss=0.3105, simple_loss=0.3687, pruned_loss=0.1262, over 28606.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3774, pruned_loss=0.1254, over 5673830.05 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3584, pruned_loss=0.1015, over 5470935.46 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3772, pruned_loss=0.1259, over 5674242.78 frames. ], batch size: 85, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:39:44,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 02:39:56,197 INFO [optim.py:369] (1/2) 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 02:40:12,627 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=415837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:40:18,208 INFO [train.py:968] (1/2) Epoch 10, batch 6250, giga_loss[loss=0.3442, simple_loss=0.4004, pruned_loss=0.144, over 28731.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3827, pruned_loss=0.1301, over 5681149.00 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.358, pruned_loss=0.1012, over 5483419.40 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3835, pruned_loss=0.1315, over 5675459.02 frames. ], batch size: 119, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:41:01,694 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=415889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:41:05,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=415892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:41:06,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3295, 1.7132, 1.6459, 1.1912], device='cuda:1'), covar=tensor([0.1667, 0.2412, 0.1442, 0.1649], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0690, 0.0838, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 02:41:06,516 INFO [train.py:968] (1/2) Epoch 10, batch 6300, libri_loss[loss=0.2867, simple_loss=0.3669, pruned_loss=0.1032, over 29673.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3861, pruned_loss=0.1329, over 5669714.58 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3581, pruned_loss=0.1012, over 5492060.86 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3875, pruned_loss=0.135, over 5662126.36 frames. ], batch size: 88, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:41:32,143 INFO [optim.py:369] (1/2) 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 02:41:33,756 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 6350, giga_loss[loss=0.2851, simple_loss=0.3523, pruned_loss=0.1089, over 28136.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3882, pruned_loss=0.1356, over 5643327.50 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3583, pruned_loss=0.1014, over 5485597.98 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3895, pruned_loss=0.1375, over 5645338.28 frames. ], batch size: 77, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:42:15,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5283, 1.5695, 1.6138, 1.5289], device='cuda:1'), covar=tensor([0.0990, 0.1190, 0.1180, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0727, 0.0661, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 02:42:51,616 INFO [train.py:968] (1/2) Epoch 10, batch 6400, giga_loss[loss=0.414, simple_loss=0.4474, pruned_loss=0.1903, over 27837.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3917, pruned_loss=0.1401, over 5621457.36 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.358, pruned_loss=0.1013, over 5482607.15 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.3933, pruned_loss=0.1422, over 5626800.26 frames. ], batch size: 412, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:43:17,447 INFO [optim.py:369] (1/2) 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 02:43:41,706 INFO [train.py:968] (1/2) Epoch 10, batch 6450, libri_loss[loss=0.2613, simple_loss=0.3383, pruned_loss=0.09218, over 29557.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3948, pruned_loss=0.1439, over 5611332.36 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3579, pruned_loss=0.1013, over 5496792.40 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3975, pruned_loss=0.147, over 5606656.69 frames. ], batch size: 75, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:44:03,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:44:36,303 INFO [train.py:968] (1/2) Epoch 10, batch 6500, giga_loss[loss=0.3386, simple_loss=0.3899, pruned_loss=0.1437, over 28302.00 frames. ], tot_loss[loss=0.346, simple_loss=0.3988, pruned_loss=0.1466, over 5616204.79 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3578, pruned_loss=0.1012, over 5498855.44 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4011, pruned_loss=0.1492, over 5611361.10 frames. ], batch size: 369, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:45:03,682 INFO [optim.py:369] (1/2) 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 02:45:13,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:45:28,977 INFO [train.py:968] (1/2) Epoch 10, batch 6550, giga_loss[loss=0.4119, simple_loss=0.4372, pruned_loss=0.1933, over 28706.00 frames. ], tot_loss[loss=0.3477, simple_loss=0.3993, pruned_loss=0.148, over 5629455.48 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.358, pruned_loss=0.1013, over 5503017.11 frames. ], giga_tot_loss[loss=0.3511, simple_loss=0.4014, pruned_loss=0.1504, over 5622965.65 frames. ], batch size: 307, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:45:31,137 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 02:46:18,009 INFO [train.py:968] (1/2) Epoch 10, batch 6600, giga_loss[loss=0.2922, simple_loss=0.3619, pruned_loss=0.1112, over 29014.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.3984, pruned_loss=0.1479, over 5632294.68 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3583, pruned_loss=0.1015, over 5511133.93 frames. ], giga_tot_loss[loss=0.3508, simple_loss=0.4005, pruned_loss=0.1505, over 5622110.03 frames. ], batch size: 128, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:46:33,533 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:46:36,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:46:37,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:46:46,543 INFO [optim.py:369] (1/2) 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 02:47:04,512 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 6650, giga_loss[loss=0.3072, simple_loss=0.3719, pruned_loss=0.1213, over 28453.00 frames. ], tot_loss[loss=0.3467, simple_loss=0.3988, pruned_loss=0.1473, over 5637704.03 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3582, pruned_loss=0.1016, over 5516985.30 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4012, pruned_loss=0.15, over 5626016.48 frames. ], batch size: 85, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:47:29,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0364, 1.1958, 3.5098, 2.9620], device='cuda:1'), covar=tensor([0.1607, 0.2484, 0.0434, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0639, 0.0574, 0.0838, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:47:55,105 INFO [train.py:968] (1/2) Epoch 10, batch 6700, giga_loss[loss=0.3641, simple_loss=0.4167, pruned_loss=0.1558, over 28558.00 frames. ], tot_loss[loss=0.3419, simple_loss=0.3961, pruned_loss=0.1439, over 5645965.87 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3577, pruned_loss=0.1012, over 5526996.78 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3997, pruned_loss=0.1479, over 5631408.19 frames. ], batch size: 336, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:47:58,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1834, 1.2104, 3.9261, 3.1509], device='cuda:1'), covar=tensor([0.1598, 0.2487, 0.0389, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0640, 0.0575, 0.0839, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:48:17,381 INFO [optim.py:369] (1/2) 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 02:48:35,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 02:48:43,387 INFO [train.py:968] (1/2) Epoch 10, batch 6750, giga_loss[loss=0.2882, simple_loss=0.362, pruned_loss=0.1072, over 28776.00 frames. ], tot_loss[loss=0.3425, simple_loss=0.3968, pruned_loss=0.1441, over 5632016.57 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3578, pruned_loss=0.1012, over 5536800.62 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4007, pruned_loss=0.1484, over 5614594.43 frames. ], batch size: 119, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:48:55,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:48:58,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:49:12,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7391, 2.8925, 1.8887, 0.8483], device='cuda:1'), covar=tensor([0.4888, 0.2189, 0.2634, 0.4864], device='cuda:1'), in_proj_covar=tensor([0.1531, 0.1454, 0.1487, 0.1255], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 02:49:26,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=416387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:49:28,626 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:49:33,860 INFO [train.py:968] (1/2) Epoch 10, batch 6800, giga_loss[loss=0.2975, simple_loss=0.372, pruned_loss=0.1115, over 28824.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3951, pruned_loss=0.1426, over 5624868.81 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3583, pruned_loss=0.1016, over 5540272.43 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3984, pruned_loss=0.1465, over 5609781.68 frames. ], batch size: 99, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:49:58,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7133, 2.2837, 2.0577, 1.5609], device='cuda:1'), covar=tensor([0.1767, 0.2110, 0.1449, 0.1647], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0695, 0.0835, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0011], device='cuda:1')
+2023-03-05 02:50:01,046 INFO [optim.py:369] (1/2) 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 02:50:20,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7085, 1.8255, 1.7133, 1.5725], device='cuda:1'), covar=tensor([0.1297, 0.1749, 0.1722, 0.1763], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0724, 0.0661, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 02:50:24,167 INFO [train.py:968] (1/2) Epoch 10, batch 6850, libri_loss[loss=0.2259, simple_loss=0.3005, pruned_loss=0.07569, over 29418.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3908, pruned_loss=0.1381, over 5632766.67 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3571, pruned_loss=0.1009, over 5552757.50 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3957, pruned_loss=0.1431, over 5612040.73 frames. ], batch size: 67, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:50:39,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-05 02:51:11,530 INFO [train.py:968] (1/2) Epoch 10, batch 6900, giga_loss[loss=0.3299, simple_loss=0.3868, pruned_loss=0.1364, over 27542.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3866, pruned_loss=0.1339, over 5647082.82 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3571, pruned_loss=0.101, over 5559473.83 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3914, pruned_loss=0.1387, over 5626836.55 frames. ], batch size: 472, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:51:20,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:51:25,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3634, 1.7371, 1.3833, 1.5218], device='cuda:1'), covar=tensor([0.2211, 0.2126, 0.2355, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.1265, 0.0938, 0.1120, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 02:51:37,073 INFO [optim.py:369] (1/2) 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 02:51:44,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:51:59,043 INFO [train.py:968] (1/2) Epoch 10, batch 6950, giga_loss[loss=0.308, simple_loss=0.377, pruned_loss=0.1195, over 28752.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3848, pruned_loss=0.1322, over 5646581.86 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3575, pruned_loss=0.1014, over 5563034.78 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3887, pruned_loss=0.1363, over 5628855.36 frames. ], batch size: 284, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:52:01,950 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:52:03,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=416548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:52:47,197 INFO [train.py:968] (1/2) Epoch 10, batch 7000, giga_loss[loss=0.3627, simple_loss=0.4027, pruned_loss=0.1614, over 26827.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3833, pruned_loss=0.1313, over 5654065.78 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3573, pruned_loss=0.1013, over 5570708.20 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3871, pruned_loss=0.1353, over 5635035.46 frames. ], batch size: 555, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:53:09,687 INFO [optim.py:369] (1/2) 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 02:53:18,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 02:53:30,135 INFO [train.py:968] (1/2) Epoch 10, batch 7050, giga_loss[loss=0.2775, simple_loss=0.3495, pruned_loss=0.1027, over 28496.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3822, pruned_loss=0.13, over 5653410.11 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3577, pruned_loss=0.1018, over 5564388.87 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3861, pruned_loss=0.134, over 5646933.63 frames. ], batch size: 60, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:53:32,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:53:34,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1106, 1.0473, 4.2007, 3.2796], device='cuda:1'), covar=tensor([0.1698, 0.2688, 0.0371, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0641, 0.0578, 0.0840, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 02:53:34,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:53:59,327 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-05 02:54:05,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=416679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:54:20,795 INFO [train.py:968] (1/2) Epoch 10, batch 7100, giga_loss[loss=0.2866, simple_loss=0.364, pruned_loss=0.1046, over 28889.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3827, pruned_loss=0.1299, over 5653988.00 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3577, pruned_loss=0.1019, over 5563485.24 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3864, pruned_loss=0.1337, over 5652240.83 frames. ], batch size: 145, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:54:51,718 INFO [optim.py:369] (1/2) 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 02:55:05,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-05 02:55:08,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1909, 1.3968, 1.1822, 0.9945], device='cuda:1'), covar=tensor([0.1610, 0.1530, 0.1021, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.1649, 0.1549, 0.1502, 0.1607], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:55:15,138 INFO [train.py:968] (1/2) Epoch 10, batch 7150, giga_loss[loss=0.3887, simple_loss=0.4403, pruned_loss=0.1686, over 27990.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3816, pruned_loss=0.1282, over 5663513.93 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3579, pruned_loss=0.102, over 5570485.08 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3848, pruned_loss=0.1317, over 5657486.27 frames. ], batch size: 412, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:55:38,169 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:55:57,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-05 02:56:08,282 INFO [train.py:968] (1/2) Epoch 10, batch 7200, libri_loss[loss=0.2463, simple_loss=0.3195, pruned_loss=0.08658, over 29537.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3819, pruned_loss=0.1262, over 5672141.52 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3574, pruned_loss=0.1017, over 5582637.06 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3857, pruned_loss=0.13, over 5659011.91 frames. ], batch size: 79, lr: 3.30e-03, grad_scale: 8.0
+2023-03-05 02:56:32,406 INFO [optim.py:369] (1/2) 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 02:56:37,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4264, 4.2417, 4.0127, 2.1156], device='cuda:1'), covar=tensor([0.0525, 0.0674, 0.0753, 0.1913], device='cuda:1'), in_proj_covar=tensor([0.1020, 0.0958, 0.0838, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 02:56:52,787 INFO [train.py:968] (1/2) Epoch 10, batch 7250, giga_loss[loss=0.3218, simple_loss=0.3994, pruned_loss=0.1221, over 28966.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3823, pruned_loss=0.1261, over 5668396.61 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3572, pruned_loss=0.1016, over 5592027.29 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3864, pruned_loss=0.13, over 5651507.77 frames. ], batch size: 164, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:57:40,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3663, 2.9494, 1.4843, 1.4598], device='cuda:1'), covar=tensor([0.0830, 0.0310, 0.0784, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0505, 0.0334, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 02:57:43,122 INFO [train.py:968] (1/2) Epoch 10, batch 7300, libri_loss[loss=0.2823, simple_loss=0.3626, pruned_loss=0.101, over 29391.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3823, pruned_loss=0.1263, over 5678130.77 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3572, pruned_loss=0.1015, over 5597370.53 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3863, pruned_loss=0.1304, over 5662266.68 frames. ], batch size: 92, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 02:57:50,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:57:53,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=416907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:57:55,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=416910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:57:57,765 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-05 02:58:08,882 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:58:09,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=416923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:58:22,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=416939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 02:58:27,297 INFO [train.py:968] (1/2) Epoch 10, batch 7350, giga_loss[loss=0.3151, simple_loss=0.3736, pruned_loss=0.1283, over 28872.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3805, pruned_loss=0.1255, over 5680193.68 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3567, pruned_loss=0.1013, over 5605163.54 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3847, pruned_loss=0.1295, over 5662496.11 frames. ], batch size: 174, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:58:29,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4916, 2.2410, 1.7295, 0.6632], device='cuda:1'), covar=tensor([0.3590, 0.2026, 0.2758, 0.3938], device='cuda:1'), in_proj_covar=tensor([0.1524, 0.1440, 0.1478, 0.1252], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 02:59:00,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3192, 4.1068, 3.9517, 1.6424], device='cuda:1'), covar=tensor([0.0607, 0.0733, 0.0812, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.1018, 0.0956, 0.0835, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 02:59:02,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 02:59:16,110 INFO [train.py:968] (1/2) Epoch 10, batch 7400, giga_loss[loss=0.3145, simple_loss=0.3772, pruned_loss=0.1259, over 28889.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3793, pruned_loss=0.1259, over 5667453.44 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3572, pruned_loss=0.1015, over 5602571.13 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3829, pruned_loss=0.1296, over 5657673.63 frames. ], batch size: 174, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 02:59:36,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2918, 2.8983, 1.4115, 1.3230], device='cuda:1'), covar=tensor([0.0899, 0.0344, 0.0843, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0506, 0.0336, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 02:59:40,472 INFO [optim.py:369] (1/2) 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 02:59:58,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3797, 2.8201, 1.9220, 1.7947], device='cuda:1'), covar=tensor([0.1870, 0.1351, 0.1608, 0.1806], device='cuda:1'), in_proj_covar=tensor([0.1666, 0.1560, 0.1516, 0.1623], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 02:59:58,963 INFO [train.py:968] (1/2) Epoch 10, batch 7450, giga_loss[loss=0.2738, simple_loss=0.3434, pruned_loss=0.1021, over 28511.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3777, pruned_loss=0.1251, over 5679255.50 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3573, pruned_loss=0.1014, over 5612089.64 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3812, pruned_loss=0.1288, over 5665015.80 frames. ], batch size: 85, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 03:00:01,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:03,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:16,395 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:18,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:20,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:21,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:32,817 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:34,763 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:37,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6750, 1.8868, 1.9784, 1.4676], device='cuda:1'), covar=tensor([0.1865, 0.2252, 0.1462, 0.1669], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0701, 0.0845, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 03:00:46,690 INFO [train.py:968] (1/2) Epoch 10, batch 7500, giga_loss[loss=0.3123, simple_loss=0.3834, pruned_loss=0.1206, over 28802.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3777, pruned_loss=0.1245, over 5692140.92 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3577, pruned_loss=0.1018, over 5622265.52 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3808, pruned_loss=0.1279, over 5673856.74 frames. ], batch size: 112, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 03:00:50,555 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:00:51,191 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:01:12,734 INFO [optim.py:369] (1/2) 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 03:01:31,684 INFO [train.py:968] (1/2) Epoch 10, batch 7550, giga_loss[loss=0.3073, simple_loss=0.3782, pruned_loss=0.1182, over 28482.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3773, pruned_loss=0.1228, over 5687460.73 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3576, pruned_loss=0.1017, over 5611549.36 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3803, pruned_loss=0.1262, over 5683829.76 frames. ], batch size: 71, lr: 3.30e-03, grad_scale: 2.0
+2023-03-05 03:02:13,895 INFO [train.py:968] (1/2) Epoch 10, batch 7600, giga_loss[loss=0.3411, simple_loss=0.4008, pruned_loss=0.1407, over 29009.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3764, pruned_loss=0.1217, over 5698071.17 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3573, pruned_loss=0.1016, over 5621851.86 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3797, pruned_loss=0.1252, over 5688258.98 frames. ], batch size: 145, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 03:02:38,560 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:02:55,497 INFO [train.py:968] (1/2) Epoch 10, batch 7650, giga_loss[loss=0.322, simple_loss=0.3835, pruned_loss=0.1302, over 28854.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3766, pruned_loss=0.1226, over 5691795.75 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3577, pruned_loss=0.1018, over 5621193.23 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3797, pruned_loss=0.126, over 5686721.44 frames. ], batch size: 106, lr: 3.30e-03, grad_scale: 4.0
+2023-03-05 03:03:03,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-05 03:03:46,378 INFO [train.py:968] (1/2) Epoch 10, batch 7700, giga_loss[loss=0.3173, simple_loss=0.3816, pruned_loss=0.1265, over 28877.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3754, pruned_loss=0.1228, over 5678577.75 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3576, pruned_loss=0.1018, over 5611490.58 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3781, pruned_loss=0.1256, over 5684404.22 frames. ], batch size: 186, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:04:18,048 INFO [optim.py:369] (1/2) 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 03:04:36,022 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:04:36,440 INFO [train.py:968] (1/2) Epoch 10, batch 7750, giga_loss[loss=0.2729, simple_loss=0.3507, pruned_loss=0.09754, over 29021.00 frames. ], tot_loss[loss=0.312, simple_loss=0.376, pruned_loss=0.124, over 5679458.01 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3578, pruned_loss=0.1019, over 5617386.22 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3783, pruned_loss=0.1267, over 5679929.87 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:05:27,484 INFO [train.py:968] (1/2) Epoch 10, batch 7800, giga_loss[loss=0.3214, simple_loss=0.375, pruned_loss=0.1339, over 27992.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3756, pruned_loss=0.1247, over 5685328.21 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3579, pruned_loss=0.1019, over 5618036.12 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3775, pruned_loss=0.127, over 5686179.91 frames. ], batch size: 412, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:05:36,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4079, 2.0778, 1.5593, 0.5801], device='cuda:1'), covar=tensor([0.3900, 0.1883, 0.2681, 0.4411], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1456, 0.1490, 0.1267], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 03:05:41,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8302, 2.3984, 2.0305, 1.4713], device='cuda:1'), covar=tensor([0.2476, 0.1575, 0.1567, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.1666, 0.1560, 0.1519, 0.1629], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 03:05:55,333 INFO [optim.py:369] (1/2) 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 03:06:17,269 INFO [train.py:968] (1/2) Epoch 10, batch 7850, giga_loss[loss=0.3358, simple_loss=0.3678, pruned_loss=0.1519, over 23795.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3718, pruned_loss=0.1224, over 5692047.74 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3575, pruned_loss=0.1016, over 5625537.50 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3741, pruned_loss=0.125, over 5687709.67 frames. ], batch size: 705, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:06:27,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=417456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:06:58,369 INFO [train.py:968] (1/2) Epoch 10, batch 7900, giga_loss[loss=0.3345, simple_loss=0.387, pruned_loss=0.141, over 28569.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3718, pruned_loss=0.1224, over 5703781.58 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.358, pruned_loss=0.1016, over 5635686.02 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3738, pruned_loss=0.1253, over 5693778.50 frames. ], batch size: 307, lr: 3.29e-03, grad_scale: 2.0
+2023-03-05 03:07:28,023 INFO [optim.py:369] (1/2) 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 03:07:46,225 INFO [train.py:968] (1/2) Epoch 10, batch 7950, giga_loss[loss=0.3627, simple_loss=0.4117, pruned_loss=0.1569, over 27931.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3723, pruned_loss=0.1231, over 5686618.68 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3582, pruned_loss=0.1018, over 5632524.38 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3741, pruned_loss=0.1259, over 5683050.89 frames. ], batch size: 412, lr: 3.29e-03, grad_scale: 2.0
+2023-03-05 03:08:07,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-05 03:08:25,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 03:08:29,582 INFO [train.py:968] (1/2) Epoch 10, batch 8000, giga_loss[loss=0.3927, simple_loss=0.4277, pruned_loss=0.1788, over 27973.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3736, pruned_loss=0.1235, over 5690362.93 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3584, pruned_loss=0.1018, over 5639811.05 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3753, pruned_loss=0.1264, over 5682675.20 frames. ], batch size: 412, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:08:32,396 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:08:34,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:08:38,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:08:52,435 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:09:05,899 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:09:16,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-05 03:09:17,172 INFO [train.py:968] (1/2) Epoch 10, batch 8050, giga_loss[loss=0.32, simple_loss=0.3813, pruned_loss=0.1294, over 28506.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.374, pruned_loss=0.1234, over 5682592.96 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3584, pruned_loss=0.1018, over 5644825.16 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3757, pruned_loss=0.1262, over 5672726.57 frames. ], batch size: 336, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:09:50,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7461, 1.8167, 1.6793, 1.6004], device='cuda:1'), covar=tensor([0.1236, 0.1694, 0.1709, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0737, 0.0667, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 03:10:02,681 INFO [train.py:968] (1/2) Epoch 10, batch 8100, giga_loss[loss=0.3218, simple_loss=0.3855, pruned_loss=0.129, over 28971.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3737, pruned_loss=0.1228, over 5676820.57 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3577, pruned_loss=0.1013, over 5649202.97 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.376, pruned_loss=0.1258, over 5665979.29 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:10:28,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=417718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:10:32,520 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:10:32,935 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 8150, giga_loss[loss=0.333, simple_loss=0.3833, pruned_loss=0.1413, over 28666.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3744, pruned_loss=0.1233, over 5681995.32 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3576, pruned_loss=0.1013, over 5646630.94 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3767, pruned_loss=0.1262, over 5675908.07 frames. ], batch size: 92, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:11:07,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:11:09,575 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:11:34,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:11:40,088 INFO [train.py:968] (1/2) Epoch 10, batch 8200, giga_loss[loss=0.3245, simple_loss=0.3789, pruned_loss=0.135, over 28785.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3764, pruned_loss=0.1255, over 5677132.21 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3578, pruned_loss=0.1015, over 5653575.21 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3785, pruned_loss=0.1284, over 5666763.41 frames. ], batch size: 119, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:11:48,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8010, 2.0164, 2.0714, 1.6227], device='cuda:1'), covar=tensor([0.1709, 0.1986, 0.1273, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0704, 0.0847, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 03:12:11,591 INFO [optim.py:369] (1/2) 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 03:12:29,628 INFO [train.py:968] (1/2) Epoch 10, batch 8250, giga_loss[loss=0.3024, simple_loss=0.3606, pruned_loss=0.1221, over 28914.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.377, pruned_loss=0.1267, over 5688383.46 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3579, pruned_loss=0.1015, over 5660240.56 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.379, pruned_loss=0.1297, over 5674740.87 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:12:48,784 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=417861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:12:52,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=417864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:13:16,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=417893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:13:16,502 INFO [train.py:968] (1/2) Epoch 10, batch 8300, giga_loss[loss=0.3371, simple_loss=0.389, pruned_loss=0.1427, over 28235.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.379, pruned_loss=0.1295, over 5680914.76 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.358, pruned_loss=0.1016, over 5665122.69 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3812, pruned_loss=0.1326, over 5665750.42 frames. ], batch size: 368, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:13:45,034 INFO [optim.py:369] (1/2) 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 03:14:04,454 INFO [train.py:968] (1/2) Epoch 10, batch 8350, giga_loss[loss=0.3119, simple_loss=0.3769, pruned_loss=0.1234, over 28847.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3798, pruned_loss=0.1305, over 5677692.87 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3583, pruned_loss=0.1018, over 5670915.16 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3819, pruned_loss=0.1336, over 5660964.37 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:14:29,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=417971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:14:47,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=417991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:14:49,925 INFO [train.py:968] (1/2) Epoch 10, batch 8400, giga_loss[loss=0.2995, simple_loss=0.368, pruned_loss=0.1155, over 28657.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.378, pruned_loss=0.1292, over 5669691.75 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3583, pruned_loss=0.1018, over 5666979.33 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3801, pruned_loss=0.1323, over 5659944.55 frames. ], batch size: 242, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:15:16,391 INFO [optim.py:369] (1/2) 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 03:15:36,010 INFO [train.py:968] (1/2) Epoch 10, batch 8450, giga_loss[loss=0.3134, simple_loss=0.3602, pruned_loss=0.1333, over 26502.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3785, pruned_loss=0.1286, over 5674713.35 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3585, pruned_loss=0.1018, over 5669089.34 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3803, pruned_loss=0.1312, over 5665280.96 frames. ], batch size: 555, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:15:51,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1796, 1.1597, 3.4330, 3.0271], device='cuda:1'), covar=tensor([0.1517, 0.2410, 0.0426, 0.1647], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0580, 0.0845, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 03:15:53,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-05 03:16:21,585 INFO [train.py:968] (1/2) Epoch 10, batch 8500, giga_loss[loss=0.291, simple_loss=0.3635, pruned_loss=0.1093, over 28956.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3762, pruned_loss=0.1263, over 5670545.42 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3585, pruned_loss=0.1019, over 5670691.56 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3778, pruned_loss=0.1287, over 5661651.52 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:16:25,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:16:39,436 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:16:41,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:16:46,388 INFO [optim.py:369] (1/2) 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 03:17:08,502 INFO [train.py:968] (1/2) Epoch 10, batch 8550, giga_loss[loss=0.3045, simple_loss=0.3639, pruned_loss=0.1226, over 28887.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3742, pruned_loss=0.1255, over 5680404.70 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3581, pruned_loss=0.1017, over 5677701.71 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3763, pruned_loss=0.1282, over 5667238.82 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:17:09,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:17:28,998 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-05 03:17:37,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:17:52,672 INFO [train.py:968] (1/2) Epoch 10, batch 8600, giga_loss[loss=0.2768, simple_loss=0.3411, pruned_loss=0.1063, over 28578.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3728, pruned_loss=0.1255, over 5682563.51 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3577, pruned_loss=0.1014, over 5682493.80 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3752, pruned_loss=0.1285, over 5667827.20 frames. ], batch size: 85, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:18:06,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3331, 3.1319, 2.9712, 1.4448], device='cuda:1'), covar=tensor([0.0863, 0.0979, 0.0884, 0.2141], device='cuda:1'), in_proj_covar=tensor([0.1018, 0.0951, 0.0843, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 03:18:23,859 INFO [optim.py:369] (1/2) 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 03:18:39,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:18:43,703 INFO [train.py:968] (1/2) Epoch 10, batch 8650, giga_loss[loss=0.41, simple_loss=0.4285, pruned_loss=0.1957, over 23375.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3735, pruned_loss=0.1266, over 5660735.13 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3576, pruned_loss=0.1013, over 5686781.10 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.376, pruned_loss=0.1298, over 5644649.48 frames. ], batch size: 705, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:18:44,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:19:01,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-05 03:19:09,292 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:19:24,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5828, 2.9007, 1.6197, 1.7299], device='cuda:1'), covar=tensor([0.0680, 0.0294, 0.0683, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0505, 0.0333, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 03:19:30,417 INFO [train.py:968] (1/2) Epoch 10, batch 8700, giga_loss[loss=0.3221, simple_loss=0.3951, pruned_loss=0.1245, over 28881.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 5664901.55 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3574, pruned_loss=0.1013, over 5684884.31 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3783, pruned_loss=0.13, over 5653634.72 frames. ], batch size: 106, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:19:46,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7254, 1.7381, 1.3348, 1.3944], device='cuda:1'), covar=tensor([0.0764, 0.0666, 0.0964, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0446, 0.0500, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 03:19:58,912 INFO [optim.py:369] (1/2) 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 03:20:07,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 03:20:18,052 INFO [train.py:968] (1/2) Epoch 10, batch 8750, giga_loss[loss=0.312, simple_loss=0.3822, pruned_loss=0.1209, over 28736.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3783, pruned_loss=0.1257, over 5667029.45 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3571, pruned_loss=0.1012, over 5688333.62 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3809, pruned_loss=0.1288, over 5654949.52 frames. ], batch size: 262, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:20:38,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:21:03,374 INFO [train.py:968] (1/2) Epoch 10, batch 8800, giga_loss[loss=0.3263, simple_loss=0.3863, pruned_loss=0.1332, over 28865.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3796, pruned_loss=0.1254, over 5673266.63 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.357, pruned_loss=0.1011, over 5685992.50 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3827, pruned_loss=0.1288, over 5665039.64 frames. ], batch size: 112, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:21:12,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418402.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:21:32,416 INFO [optim.py:369] (1/2) 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 03:21:38,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:21:47,675 INFO [train.py:968] (1/2) Epoch 10, batch 8850, giga_loss[loss=0.3904, simple_loss=0.4233, pruned_loss=0.1787, over 27567.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3817, pruned_loss=0.1272, over 5670824.88 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3567, pruned_loss=0.1008, over 5686131.17 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3848, pruned_loss=0.1307, over 5663884.48 frames. ], batch size: 472, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:22:34,775 INFO [train.py:968] (1/2) Epoch 10, batch 8900, giga_loss[loss=0.3743, simple_loss=0.3994, pruned_loss=0.1745, over 23618.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3839, pruned_loss=0.1296, over 5650890.86 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3569, pruned_loss=0.101, over 5678489.20 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3866, pruned_loss=0.1326, over 5651700.32 frames. ], batch size: 705, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:22:47,103 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:22:50,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:22:53,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:23:05,236 INFO [optim.py:369] (1/2) 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 03:23:22,348 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:23:24,762 INFO [train.py:968] (1/2) Epoch 10, batch 8950, giga_loss[loss=0.2933, simple_loss=0.3488, pruned_loss=0.1189, over 28822.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3822, pruned_loss=0.1295, over 5650628.26 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3571, pruned_loss=0.1013, over 5678727.58 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3844, pruned_loss=0.1319, over 5650653.56 frames. ], batch size: 92, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:23:33,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:24:16,818 INFO [train.py:968] (1/2) Epoch 10, batch 9000, giga_loss[loss=0.286, simple_loss=0.3616, pruned_loss=0.1052, over 28882.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3816, pruned_loss=0.1302, over 5642052.80 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.357, pruned_loss=0.1012, over 5683553.58 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.384, pruned_loss=0.1327, over 5636962.73 frames. ], batch size: 174, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:24:16,818 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 03:24:23,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9532, 3.6994, 3.5495, 1.6659], device='cuda:1'), covar=tensor([0.0653, 0.0887, 0.0815, 0.2400], device='cuda:1'), in_proj_covar=tensor([0.1024, 0.0960, 0.0848, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 03:24:25,362 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 03:24:51,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 03:24:55,960 INFO [optim.py:369] (1/2) 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 03:25:12,645 INFO [train.py:968] (1/2) Epoch 10, batch 9050, giga_loss[loss=0.2827, simple_loss=0.3585, pruned_loss=0.1034, over 28966.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3789, pruned_loss=0.1288, over 5656770.58 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3569, pruned_loss=0.101, over 5688747.50 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3814, pruned_loss=0.1316, over 5647410.87 frames. ], batch size: 164, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:26:03,909 INFO [train.py:968] (1/2) Epoch 10, batch 9100, giga_loss[loss=0.2989, simple_loss=0.3649, pruned_loss=0.1165, over 28860.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3785, pruned_loss=0.129, over 5665178.27 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3567, pruned_loss=0.1009, over 5689883.77 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3806, pruned_loss=0.1314, over 5656825.00 frames. ], batch size: 186, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:26:05,678 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:26:08,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:26:33,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:26:35,575 INFO [optim.py:369] (1/2) 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 03:26:37,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 9150, giga_loss[loss=0.3039, simple_loss=0.3634, pruned_loss=0.1222, over 28496.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.38, pruned_loss=0.1309, over 5645435.22 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3571, pruned_loss=0.1014, over 5695085.59 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3819, pruned_loss=0.1331, over 5633359.36 frames. ], batch size: 71, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:26:54,323 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:23,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:25,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418777.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 03:27:36,290 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 9200, giga_loss[loss=0.2939, simple_loss=0.3554, pruned_loss=0.1162, over 28706.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3783, pruned_loss=0.1301, over 5665325.83 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3577, pruned_loss=0.1018, over 5699987.24 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3801, pruned_loss=0.1325, over 5649595.29 frames. ], batch size: 66, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:27:52,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=418808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:27:54,477 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:28:05,917 INFO [optim.py:369] (1/2) 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 03:28:25,218 INFO [train.py:968] (1/2) Epoch 10, batch 9250, giga_loss[loss=0.3139, simple_loss=0.3762, pruned_loss=0.1258, over 28836.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3762, pruned_loss=0.1285, over 5655706.96 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.358, pruned_loss=0.1018, over 5700632.30 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3778, pruned_loss=0.1313, over 5641786.31 frames. ], batch size: 199, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:28:55,950 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 9300, giga_loss[loss=0.3908, simple_loss=0.4216, pruned_loss=0.18, over 26591.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.376, pruned_loss=0.1274, over 5664443.52 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3579, pruned_loss=0.1016, over 5705270.18 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3778, pruned_loss=0.1303, over 5648089.18 frames. ], batch size: 555, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:29:32,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418920.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 03:29:36,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418923.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 03:29:38,423 INFO [optim.py:369] (1/2) 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 03:29:53,611 INFO [train.py:968] (1/2) Epoch 10, batch 9350, giga_loss[loss=0.2692, simple_loss=0.3472, pruned_loss=0.09564, over 28901.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.379, pruned_loss=0.1294, over 5668194.36 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3575, pruned_loss=0.1015, over 5709088.97 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3812, pruned_loss=0.1324, over 5650948.51 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:30:00,095 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=418951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:01,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418952.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:30:02,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=418954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:29,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=418982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:29,641 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=418983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:30:38,822 INFO [train.py:968] (1/2) Epoch 10, batch 9400, giga_loss[loss=0.3161, simple_loss=0.3722, pruned_loss=0.1299, over 28921.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3804, pruned_loss=0.1306, over 5662653.30 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3578, pruned_loss=0.1016, over 5713763.04 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3824, pruned_loss=0.1335, over 5643867.34 frames. ], batch size: 186, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:31:05,747 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:31:08,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:31:08,486 INFO [optim.py:369] (1/2) 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 03:31:23,952 INFO [train.py:968] (1/2) Epoch 10, batch 9450, giga_loss[loss=0.2967, simple_loss=0.3781, pruned_loss=0.1076, over 28944.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3807, pruned_loss=0.1293, over 5666318.79 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3578, pruned_loss=0.1017, over 5716709.86 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3826, pruned_loss=0.132, over 5648281.08 frames. ], batch size: 199, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:31:31,689 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:32:04,480 INFO [train.py:968] (1/2) Epoch 10, batch 9500, giga_loss[loss=0.2559, simple_loss=0.3487, pruned_loss=0.08152, over 28632.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3806, pruned_loss=0.1271, over 5674779.79 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.357, pruned_loss=0.1012, over 5720143.59 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3836, pruned_loss=0.1306, over 5655774.07 frames. ], batch size: 60, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:32:08,476 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:32:19,123 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:32:19,711 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419112.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 03:32:25,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:32:30,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-05 03:32:31,901 INFO [optim.py:369] (1/2) 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 03:32:45,300 INFO [train.py:968] (1/2) Epoch 10, batch 9550, giga_loss[loss=0.2992, simple_loss=0.3845, pruned_loss=0.107, over 28999.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3844, pruned_loss=0.1281, over 5688440.02 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3571, pruned_loss=0.1012, over 5725488.25 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3877, pruned_loss=0.1318, over 5666766.26 frames. ], batch size: 106, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:32:49,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:33:06,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:33:12,387 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:33:13,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2966, 1.5478, 1.5325, 1.4282], device='cuda:1'), covar=tensor([0.1292, 0.1195, 0.1541, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0730, 0.0660, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 03:33:20,884 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 9600, giga_loss[loss=0.4313, simple_loss=0.4651, pruned_loss=0.1988, over 28538.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3872, pruned_loss=0.1303, over 5681504.33 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3572, pruned_loss=0.1014, over 5725608.73 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3904, pruned_loss=0.1339, over 5663175.60 frames. ], batch size: 336, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:33:34,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6144, 2.2845, 1.5702, 0.7665], device='cuda:1'), covar=tensor([0.4861, 0.2595, 0.2195, 0.4352], device='cuda:1'), in_proj_covar=tensor([0.1523, 0.1446, 0.1472, 0.1251], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 03:34:02,086 INFO [optim.py:369] (1/2) 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 03:34:05,142 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:12,845 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 9650, giga_loss[loss=0.3289, simple_loss=0.3897, pruned_loss=0.134, over 28905.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3886, pruned_loss=0.1323, over 5687091.69 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3569, pruned_loss=0.1012, over 5730053.75 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3922, pruned_loss=0.1361, over 5667270.05 frames. ], batch size: 145, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:34:15,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:34,312 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:36,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:34:45,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:04,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:05,232 INFO [train.py:968] (1/2) Epoch 10, batch 9700, giga_loss[loss=0.3812, simple_loss=0.4318, pruned_loss=0.1653, over 28803.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3904, pruned_loss=0.1348, over 5663328.05 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.357, pruned_loss=0.1012, over 5721705.70 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3934, pruned_loss=0.1381, over 5655571.26 frames. ], batch size: 174, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:35:05,423 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:06,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:18,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:20,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:20,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3985, 1.6146, 1.3454, 1.4651], device='cuda:1'), covar=tensor([0.2765, 0.2568, 0.2855, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.0936, 0.1120, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 03:35:30,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:32,937 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:34,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:34,910 INFO [optim.py:369] (1/2) 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 03:35:38,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:40,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:49,321 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:35:50,593 INFO [train.py:968] (1/2) Epoch 10, batch 9750, giga_loss[loss=0.3321, simple_loss=0.3856, pruned_loss=0.1393, over 28751.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3884, pruned_loss=0.1331, over 5669688.62 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3572, pruned_loss=0.1012, over 5727223.77 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3919, pruned_loss=0.137, over 5656372.63 frames. ], batch size: 119, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:36:03,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:36:04,635 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:36:08,529 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:36:32,753 INFO [train.py:968] (1/2) Epoch 10, batch 9800, libri_loss[loss=0.2382, simple_loss=0.3237, pruned_loss=0.07641, over 29565.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3869, pruned_loss=0.1312, over 5677282.13 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3566, pruned_loss=0.101, over 5728984.51 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3911, pruned_loss=0.1353, over 5663567.29 frames. ], batch size: 78, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:36:59,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2156, 1.2981, 1.1520, 1.3942], device='cuda:1'), covar=tensor([0.0668, 0.0417, 0.0327, 0.0709], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0116, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0077, 0.0055, 0.0049, 0.0084], device='cuda:1')
+2023-03-05 03:37:02,548 INFO [optim.py:369] (1/2) 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 03:37:16,063 INFO [train.py:968] (1/2) Epoch 10, batch 9850, giga_loss[loss=0.3537, simple_loss=0.407, pruned_loss=0.1503, over 28761.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3862, pruned_loss=0.1292, over 5675445.77 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3564, pruned_loss=0.1009, over 5728230.80 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3904, pruned_loss=0.1332, over 5664291.94 frames. ], batch size: 242, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:37:29,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0858, 5.2917, 2.2743, 2.4118], device='cuda:1'), covar=tensor([0.0802, 0.0171, 0.0766, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0509, 0.0335, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 03:37:50,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:37:51,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419487.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:37:57,205 INFO [train.py:968] (1/2) Epoch 10, batch 9900, giga_loss[loss=0.3359, simple_loss=0.4033, pruned_loss=0.1342, over 28766.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3869, pruned_loss=0.1291, over 5681405.70 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3565, pruned_loss=0.101, over 5733515.60 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3914, pruned_loss=0.1333, over 5665646.71 frames. ], batch size: 284, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:38:01,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:38:05,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:38:27,790 INFO [optim.py:369] (1/2) 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 03:38:30,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7715, 1.7495, 1.6204, 1.5837], device='cuda:1'), covar=tensor([0.1352, 0.2043, 0.1956, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0730, 0.0662, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 03:38:31,453 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 9950, giga_loss[loss=0.3628, simple_loss=0.3922, pruned_loss=0.1667, over 23202.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3888, pruned_loss=0.1317, over 5674759.97 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3561, pruned_loss=0.1007, over 5738393.28 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3938, pruned_loss=0.1363, over 5656074.63 frames. ], batch size: 705, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:38:47,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:39:30,799 INFO [train.py:968] (1/2) Epoch 10, batch 10000, giga_loss[loss=0.2893, simple_loss=0.3619, pruned_loss=0.1084, over 29023.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3874, pruned_loss=0.1312, over 5672606.05 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.356, pruned_loss=0.1008, over 5742548.99 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3921, pruned_loss=0.1355, over 5652544.41 frames. ], batch size: 128, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:39:41,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:39:44,497 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:40:04,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:40:05,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419630.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:40:09,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:40:10,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419633.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 03:40:20,772 INFO [train.py:968] (1/2) Epoch 10, batch 10050, giga_loss[loss=0.3481, simple_loss=0.4065, pruned_loss=0.1448, over 28314.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3863, pruned_loss=0.1319, over 5661054.20 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3557, pruned_loss=0.1005, over 5744847.86 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3908, pruned_loss=0.136, over 5642197.85 frames. ], batch size: 71, lr: 3.29e-03, grad_scale: 8.0
+2023-03-05 03:40:35,240 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:40:35,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419662.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 03:41:04,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:06,257 INFO [train.py:968] (1/2) Epoch 10, batch 10100, giga_loss[loss=0.4212, simple_loss=0.4459, pruned_loss=0.1983, over 27980.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3834, pruned_loss=0.1305, over 5672153.17 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3556, pruned_loss=0.1004, over 5746231.54 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.388, pruned_loss=0.1348, over 5653486.16 frames. ], batch size: 412, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:41:06,504 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:20,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:38,308 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:42,101 INFO [optim.py:369] (1/2) 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 03:41:52,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:41:55,272 INFO [train.py:968] (1/2) Epoch 10, batch 10150, giga_loss[loss=0.341, simple_loss=0.4, pruned_loss=0.141, over 29093.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3808, pruned_loss=0.1291, over 5654544.45 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.356, pruned_loss=0.1006, over 5737117.67 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3848, pruned_loss=0.1332, over 5646133.44 frames. ], batch size: 155, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:41:59,334 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:02,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:30,565 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:33,283 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:42:34,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 03:42:42,415 INFO [train.py:968] (1/2) Epoch 10, batch 10200, giga_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1248, over 28757.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3812, pruned_loss=0.1305, over 5658590.50 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3565, pruned_loss=0.101, over 5739831.17 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3843, pruned_loss=0.1339, over 5648130.36 frames. ], batch size: 284, lr: 3.29e-03, grad_scale: 4.0
+2023-03-05 03:43:11,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4913, 1.7290, 1.4525, 1.5778], device='cuda:1'), covar=tensor([0.1857, 0.1738, 0.1688, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.0939, 0.1118, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 03:43:15,040 INFO [optim.py:369] (1/2) 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 03:43:25,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:29,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:29,547 INFO [train.py:968] (1/2) Epoch 10, batch 10250, giga_loss[loss=0.284, simple_loss=0.3557, pruned_loss=0.1061, over 29017.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3796, pruned_loss=0.1294, over 5662558.07 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3563, pruned_loss=0.1008, over 5741876.28 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3826, pruned_loss=0.1327, over 5651509.69 frames. ], batch size: 128, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:43:37,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:39,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4695, 1.8295, 1.4279, 1.5681], device='cuda:1'), covar=tensor([0.2651, 0.2432, 0.2736, 0.2421], device='cuda:1'), in_proj_covar=tensor([0.1267, 0.0938, 0.1119, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 03:43:39,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:43:44,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8677, 2.5775, 1.7173, 1.0411], device='cuda:1'), covar=tensor([0.4493, 0.2372, 0.2623, 0.4200], device='cuda:1'), in_proj_covar=tensor([0.1533, 0.1455, 0.1478, 0.1256], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 03:43:56,376 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:07,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:07,839 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=419883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:10,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=419886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:17,273 INFO [train.py:968] (1/2) Epoch 10, batch 10300, libri_loss[loss=0.2718, simple_loss=0.3466, pruned_loss=0.09854, over 29534.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3761, pruned_loss=0.1252, over 5666196.37 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3562, pruned_loss=0.1007, over 5743298.54 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.379, pruned_loss=0.1283, over 5654534.81 frames. ], batch size: 77, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:44:22,605 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=419900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:38,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=419915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:44:53,722 INFO [optim.py:369] (1/2) 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 03:45:08,706 INFO [train.py:968] (1/2) Epoch 10, batch 10350, giga_loss[loss=0.2777, simple_loss=0.3538, pruned_loss=0.1008, over 27713.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3738, pruned_loss=0.123, over 5662193.52 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3562, pruned_loss=0.1007, over 5743812.05 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3761, pruned_loss=0.1256, over 5652368.33 frames. ], batch size: 472, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:45:51,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=419985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:45:58,927 INFO [train.py:968] (1/2) Epoch 10, batch 10400, giga_loss[loss=0.2794, simple_loss=0.3465, pruned_loss=0.1061, over 28933.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3739, pruned_loss=0.1233, over 5666679.92 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3563, pruned_loss=0.1007, over 5744563.99 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3757, pruned_loss=0.1254, over 5657925.49 frames. ], batch size: 227, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 03:46:29,439 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:46:36,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5691, 4.2979, 1.8010, 1.7112], device='cuda:1'), covar=tensor([0.0868, 0.0217, 0.0761, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0507, 0.0334, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 03:46:44,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5165, 3.7690, 1.6518, 1.6048], device='cuda:1'), covar=tensor([0.0833, 0.0290, 0.0766, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0508, 0.0334, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 03:46:47,406 INFO [train.py:968] (1/2) Epoch 10, batch 10450, giga_loss[loss=0.2623, simple_loss=0.333, pruned_loss=0.09576, over 28942.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3703, pruned_loss=0.1218, over 5670002.70 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3564, pruned_loss=0.1007, over 5748836.22 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.372, pruned_loss=0.124, over 5657568.28 frames. ], batch size: 106, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:46:58,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7873, 1.0893, 2.8377, 2.7290], device='cuda:1'), covar=tensor([0.1562, 0.2309, 0.0535, 0.1297], device='cuda:1'), in_proj_covar=tensor([0.0646, 0.0582, 0.0846, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 03:47:32,088 INFO [train.py:968] (1/2) Epoch 10, batch 10500, libri_loss[loss=0.2776, simple_loss=0.3603, pruned_loss=0.09746, over 29530.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.37, pruned_loss=0.1216, over 5676358.07 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3567, pruned_loss=0.1008, over 5753814.49 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3715, pruned_loss=0.1239, over 5658963.52 frames. ], batch size: 81, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:47:33,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.39 vs. limit=5.0
+2023-03-05 03:48:03,584 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:48:04,585 INFO [optim.py:369] (1/2) 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 03:48:05,674 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 10550, giga_loss[loss=0.267, simple_loss=0.347, pruned_loss=0.09355, over 29013.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3725, pruned_loss=0.1229, over 5673118.68 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3566, pruned_loss=0.1008, over 5755087.57 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.1249, over 5657588.26 frames. ], batch size: 136, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:48:33,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:48:35,386 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:48:37,352 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3301, 3.3951, 1.4728, 1.5346], device='cuda:1'), covar=tensor([0.0912, 0.0226, 0.0854, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0505, 0.0333, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 03:49:04,163 INFO [train.py:968] (1/2) Epoch 10, batch 10600, giga_loss[loss=0.2742, simple_loss=0.3468, pruned_loss=0.1007, over 29077.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.374, pruned_loss=0.1238, over 5663805.91 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3564, pruned_loss=0.1006, over 5758619.62 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3758, pruned_loss=0.1263, over 5645841.43 frames. ], batch size: 128, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:49:22,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-05 03:49:37,641 INFO [optim.py:369] (1/2) 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 03:49:51,892 INFO [train.py:968] (1/2) Epoch 10, batch 10650, giga_loss[loss=0.3265, simple_loss=0.382, pruned_loss=0.1355, over 29028.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3731, pruned_loss=0.123, over 5670112.94 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3568, pruned_loss=0.1007, over 5762274.63 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3747, pruned_loss=0.1256, over 5649337.71 frames. ], batch size: 128, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:50:19,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:50:36,635 INFO [train.py:968] (1/2) Epoch 10, batch 10700, giga_loss[loss=0.3954, simple_loss=0.4217, pruned_loss=0.1845, over 26610.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3747, pruned_loss=0.1248, over 5662371.55 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3574, pruned_loss=0.1009, over 5761441.52 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.376, pruned_loss=0.1274, over 5643592.86 frames. ], batch size: 555, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:50:44,185 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:50:46,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:50:47,705 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:50:59,413 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-05 03:51:12,264 INFO [optim.py:369] (1/2) 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 03:51:17,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:51:27,549 INFO [train.py:968] (1/2) Epoch 10, batch 10750, giga_loss[loss=0.3001, simple_loss=0.3749, pruned_loss=0.1126, over 28796.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3777, pruned_loss=0.1271, over 5664508.31 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3578, pruned_loss=0.1012, over 5763323.85 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3786, pruned_loss=0.1292, over 5646811.95 frames. ], batch size: 284, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:51:48,832 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:51:53,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9201, 5.7353, 5.4243, 2.7507], device='cuda:1'), covar=tensor([0.0423, 0.0584, 0.0730, 0.1748], device='cuda:1'), in_proj_covar=tensor([0.1032, 0.0973, 0.0858, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 03:52:09,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:15,900 INFO [train.py:968] (1/2) Epoch 10, batch 10800, giga_loss[loss=0.3247, simple_loss=0.3876, pruned_loss=0.1309, over 28520.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3794, pruned_loss=0.1279, over 5670007.68 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.358, pruned_loss=0.1014, over 5765296.62 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3802, pruned_loss=0.1298, over 5653004.48 frames. ], batch size: 336, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 03:52:18,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:25,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5405, 1.7024, 1.8332, 1.3881], device='cuda:1'), covar=tensor([0.1461, 0.1966, 0.1167, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0706, 0.0852, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 03:52:34,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:36,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:43,603 INFO [optim.py:369] (1/2) 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 03:52:53,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:52:55,514 INFO [train.py:968] (1/2) Epoch 10, batch 10850, libri_loss[loss=0.2778, simple_loss=0.3628, pruned_loss=0.0964, over 29480.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3806, pruned_loss=0.1283, over 5664511.66 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3591, pruned_loss=0.1021, over 5752308.20 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3812, pruned_loss=0.1304, over 5657726.39 frames. ], batch size: 85, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:53:02,660 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:53:03,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7270, 4.4750, 4.2091, 2.3257], device='cuda:1'), covar=tensor([0.0638, 0.0928, 0.1025, 0.1878], device='cuda:1'), in_proj_covar=tensor([0.1041, 0.0980, 0.0863, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 03:53:07,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-05 03:53:13,519 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-05 03:53:45,547 INFO [train.py:968] (1/2) Epoch 10, batch 10900, giga_loss[loss=0.2908, simple_loss=0.3484, pruned_loss=0.1167, over 28571.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3815, pruned_loss=0.1299, over 5671692.31 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3594, pruned_loss=0.1024, over 5754572.66 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.382, pruned_loss=0.1316, over 5663294.69 frames. ], batch size: 85, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:54:04,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2400, 1.5736, 1.5034, 1.1318], device='cuda:1'), covar=tensor([0.1296, 0.1960, 0.1120, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0702, 0.0848, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 03:54:18,063 INFO [optim.py:369] (1/2) 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 03:54:28,695 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:54:30,298 INFO [train.py:968] (1/2) Epoch 10, batch 10950, libri_loss[loss=0.3921, simple_loss=0.43, pruned_loss=0.1771, over 19839.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3833, pruned_loss=0.13, over 5664709.82 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3595, pruned_loss=0.1027, over 5751376.42 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3845, pruned_loss=0.1322, over 5658304.39 frames. ], batch size: 186, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:54:30,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:54:44,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6266, 1.6142, 1.2798, 1.2391], device='cuda:1'), covar=tensor([0.0704, 0.0584, 0.0969, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0445, 0.0503, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 03:54:58,136 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:55:19,217 INFO [train.py:968] (1/2) Epoch 10, batch 11000, giga_loss[loss=0.3237, simple_loss=0.387, pruned_loss=0.1302, over 28909.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3827, pruned_loss=0.1291, over 5657828.73 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3595, pruned_loss=0.1027, over 5743279.75 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3844, pruned_loss=0.1317, over 5657202.22 frames. ], batch size: 285, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:55:26,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6412, 1.8418, 1.5384, 1.3428], device='cuda:1'), covar=tensor([0.2159, 0.1799, 0.1517, 0.1826], device='cuda:1'), in_proj_covar=tensor([0.1647, 0.1565, 0.1505, 0.1609], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 03:56:00,131 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 11050, giga_loss[loss=0.3106, simple_loss=0.3701, pruned_loss=0.1256, over 28618.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3836, pruned_loss=0.1312, over 5651710.27 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3598, pruned_loss=0.1029, over 5745254.34 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.385, pruned_loss=0.1334, over 5648096.47 frames. ], batch size: 307, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:56:48,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:57:03,258 INFO [train.py:968] (1/2) Epoch 10, batch 11100, libri_loss[loss=0.2823, simple_loss=0.3637, pruned_loss=0.1005, over 29231.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3817, pruned_loss=0.1301, over 5648968.20 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3594, pruned_loss=0.1027, over 5752374.03 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3841, pruned_loss=0.1334, over 5635455.97 frames. ], batch size: 97, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:57:41,125 INFO [optim.py:369] (1/2) 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 03:57:52,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:57:54,010 INFO [train.py:968] (1/2) Epoch 10, batch 11150, giga_loss[loss=0.2749, simple_loss=0.3446, pruned_loss=0.1025, over 28874.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3793, pruned_loss=0.1284, over 5652607.81 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3595, pruned_loss=0.1026, over 5753271.12 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3818, pruned_loss=0.1317, over 5638460.38 frames. ], batch size: 112, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:58:11,591 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 11200, libri_loss[loss=0.2443, simple_loss=0.319, pruned_loss=0.08479, over 29489.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5645604.36 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3587, pruned_loss=0.1023, over 5746928.89 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3812, pruned_loss=0.132, over 5638084.73 frames. ], batch size: 70, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 03:59:02,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=420816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:59:11,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:59:14,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 03:59:17,776 INFO [optim.py:369] (1/2) 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 03:59:32,485 INFO [train.py:968] (1/2) Epoch 10, batch 11250, giga_loss[loss=0.3437, simple_loss=0.3718, pruned_loss=0.1578, over 23588.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.377, pruned_loss=0.1279, over 5657669.15 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3582, pruned_loss=0.1019, over 5750127.54 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3803, pruned_loss=0.1316, over 5647315.36 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 03:59:43,965 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:11,957 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:15,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:23,274 INFO [train.py:968] (1/2) Epoch 10, batch 11300, giga_loss[loss=0.324, simple_loss=0.3844, pruned_loss=0.1318, over 28549.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3769, pruned_loss=0.1281, over 5652925.52 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3582, pruned_loss=0.1018, over 5752467.93 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3798, pruned_loss=0.1315, over 5641423.30 frames. ], batch size: 307, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:00:34,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:36,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:40,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:00:44,691 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:01:03,808 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=420937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:01:10,527 INFO [train.py:968] (1/2) Epoch 10, batch 11350, giga_loss[loss=0.3075, simple_loss=0.3807, pruned_loss=0.1171, over 28910.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3777, pruned_loss=0.129, over 5656923.94 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3583, pruned_loss=0.1018, over 5755003.02 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3802, pruned_loss=0.1322, over 5643920.05 frames. ], batch size: 174, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:01:25,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=420959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:01:27,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=420962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:01:37,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7747, 4.6616, 1.8580, 1.8726], device='cuda:1'), covar=tensor([0.0855, 0.0247, 0.0806, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0509, 0.0334, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 04:01:52,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4602, 2.1817, 1.6732, 0.6518], device='cuda:1'), covar=tensor([0.3415, 0.2014, 0.2831, 0.3981], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1466, 0.1486, 0.1263], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 04:01:55,207 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 11400, giga_loss[loss=0.3061, simple_loss=0.3726, pruned_loss=0.1198, over 28884.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3808, pruned_loss=0.1316, over 5656976.75 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3587, pruned_loss=0.102, over 5749019.02 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3829, pruned_loss=0.1344, over 5649917.81 frames. ], batch size: 186, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:01:58,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=420995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:02:38,013 INFO [optim.py:369] (1/2) 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 04:02:48,710 INFO [train.py:968] (1/2) Epoch 10, batch 11450, giga_loss[loss=0.297, simple_loss=0.3557, pruned_loss=0.1191, over 28675.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3822, pruned_loss=0.1334, over 5645456.05 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.359, pruned_loss=0.1022, over 5750345.30 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3837, pruned_loss=0.1357, over 5637737.27 frames. ], batch size: 92, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:03:41,452 INFO [train.py:968] (1/2) Epoch 10, batch 11500, giga_loss[loss=0.3108, simple_loss=0.3719, pruned_loss=0.1248, over 28312.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3835, pruned_loss=0.1347, over 5654694.63 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3592, pruned_loss=0.1023, over 5751350.57 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3847, pruned_loss=0.1366, over 5647280.68 frames. ], batch size: 368, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:03:52,170 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:04:18,241 INFO [optim.py:369] (1/2) 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 04:04:28,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3839, 1.8896, 1.3081, 0.5772], device='cuda:1'), covar=tensor([0.2252, 0.1422, 0.2316, 0.3389], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1464, 0.1485, 0.1267], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 04:04:32,953 INFO [train.py:968] (1/2) Epoch 10, batch 11550, giga_loss[loss=0.3534, simple_loss=0.3846, pruned_loss=0.1611, over 23566.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3839, pruned_loss=0.1349, over 5637079.58 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3594, pruned_loss=0.1025, over 5744329.33 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3852, pruned_loss=0.137, over 5635783.40 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:05:13,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-05 04:05:18,269 INFO [train.py:968] (1/2) Epoch 10, batch 11600, giga_loss[loss=0.3784, simple_loss=0.4161, pruned_loss=0.1704, over 27585.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3835, pruned_loss=0.1338, over 5655024.09 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3592, pruned_loss=0.1024, over 5746834.94 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3853, pruned_loss=0.1363, over 5649461.50 frames. ], batch size: 472, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:05:56,843 INFO [optim.py:369] (1/2) 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 04:06:07,051 INFO [train.py:968] (1/2) Epoch 10, batch 11650, giga_loss[loss=0.4325, simple_loss=0.45, pruned_loss=0.2075, over 28636.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3833, pruned_loss=0.133, over 5664830.77 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3591, pruned_loss=0.1022, over 5748670.11 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3853, pruned_loss=0.1358, over 5657072.53 frames. ], batch size: 307, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:06:54,802 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 11700, giga_loss[loss=0.2936, simple_loss=0.3672, pruned_loss=0.1099, over 29138.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3849, pruned_loss=0.135, over 5659405.77 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3587, pruned_loss=0.1021, over 5753287.56 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3876, pruned_loss=0.1382, over 5646574.61 frames. ], batch size: 155, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:07:27,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-05 04:07:28,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 04:07:35,232 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 11750, giga_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1146, over 28757.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3852, pruned_loss=0.1356, over 5657853.68 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3587, pruned_loss=0.102, over 5755597.44 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3877, pruned_loss=0.1387, over 5644416.08 frames. ], batch size: 99, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:07:59,872 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-05 04:08:11,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:08:23,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-05 04:08:34,167 INFO [train.py:968] (1/2) Epoch 10, batch 11800, giga_loss[loss=0.3402, simple_loss=0.3993, pruned_loss=0.1406, over 28695.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3861, pruned_loss=0.1358, over 5660974.51 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3584, pruned_loss=0.1018, over 5757865.14 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3888, pruned_loss=0.1388, over 5647181.31 frames. ], batch size: 307, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:09:15,088 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:09:17,407 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=421435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:09:24,491 INFO [train.py:968] (1/2) Epoch 10, batch 11850, giga_loss[loss=0.385, simple_loss=0.4133, pruned_loss=0.1783, over 26596.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3871, pruned_loss=0.1354, over 5656555.10 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3586, pruned_loss=0.102, over 5760498.36 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3894, pruned_loss=0.1382, over 5641838.38 frames. ], batch size: 555, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:09:48,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=421464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:01,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:14,204 INFO [train.py:968] (1/2) Epoch 10, batch 11900, giga_loss[loss=0.305, simple_loss=0.3707, pruned_loss=0.1196, over 28660.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3856, pruned_loss=0.1338, over 5660775.42 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3586, pruned_loss=0.1019, over 5761964.72 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3877, pruned_loss=0.1363, over 5646962.14 frames. ], batch size: 284, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:10:34,051 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=421513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:36,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=421516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:10:48,897 INFO [optim.py:369] (1/2) 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 04:10:53,049 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 04:11:00,807 INFO [train.py:968] (1/2) Epoch 10, batch 11950, giga_loss[loss=0.2987, simple_loss=0.3586, pruned_loss=0.1194, over 28823.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3838, pruned_loss=0.1323, over 5654719.83 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3587, pruned_loss=0.1019, over 5761004.35 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3861, pruned_loss=0.1351, over 5641563.16 frames. ], batch size: 119, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:11:01,722 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=421545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:11:37,689 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:11:49,403 INFO [train.py:968] (1/2) Epoch 10, batch 12000, libri_loss[loss=0.2634, simple_loss=0.3496, pruned_loss=0.0886, over 29228.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3843, pruned_loss=0.1323, over 5664494.91 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3593, pruned_loss=0.1022, over 5765183.45 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3865, pruned_loss=0.1353, over 5646979.63 frames. ], batch size: 94, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 04:11:49,403 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 04:11:57,828 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 04:12:12,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:23,322 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:28,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=421623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:30,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=421626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:12:37,001 INFO [optim.py:369] (1/2) 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 04:12:48,668 INFO [train.py:968] (1/2) Epoch 10, batch 12050, giga_loss[loss=0.2989, simple_loss=0.3646, pruned_loss=0.1166, over 28755.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3854, pruned_loss=0.1336, over 5658776.21 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3594, pruned_loss=0.1023, over 5766769.40 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3873, pruned_loss=0.1362, over 5642610.62 frames. ], batch size: 119, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 04:12:49,645 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421645.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 04:12:59,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=421655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:13:37,429 INFO [train.py:968] (1/2) Epoch 10, batch 12100, giga_loss[loss=0.3117, simple_loss=0.3712, pruned_loss=0.1261, over 28896.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3844, pruned_loss=0.1333, over 5675243.21 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3598, pruned_loss=0.1025, over 5769774.50 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3861, pruned_loss=0.1359, over 5657420.84 frames. ], batch size: 112, lr: 3.28e-03, grad_scale: 8.0
+2023-03-05 04:13:46,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:13:59,932 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:14:12,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-05 04:14:18,038 INFO [optim.py:369] (1/2) 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 04:14:28,019 INFO [train.py:968] (1/2) Epoch 10, batch 12150, giga_loss[loss=0.3074, simple_loss=0.3767, pruned_loss=0.1191, over 28993.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3843, pruned_loss=0.1338, over 5672069.38 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.36, pruned_loss=0.1025, over 5770091.22 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3858, pruned_loss=0.1364, over 5656049.81 frames. ], batch size: 164, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:14:39,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2577, 1.3916, 1.3933, 1.2003], device='cuda:1'), covar=tensor([0.1052, 0.1220, 0.1590, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0743, 0.0671, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 04:14:54,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1930, 1.2269, 1.1174, 1.4259], device='cuda:1'), covar=tensor([0.0699, 0.0382, 0.0323, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0077, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 04:15:17,174 INFO [train.py:968] (1/2) Epoch 10, batch 12200, giga_loss[loss=0.3328, simple_loss=0.3929, pruned_loss=0.1364, over 28709.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3864, pruned_loss=0.1354, over 5664338.99 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3603, pruned_loss=0.1026, over 5759213.30 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3877, pruned_loss=0.1378, over 5660167.31 frames. ], batch size: 242, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:15:58,419 INFO [optim.py:369] (1/2) 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 04:16:06,259 INFO [train.py:968] (1/2) Epoch 10, batch 12250, giga_loss[loss=0.4116, simple_loss=0.4364, pruned_loss=0.1934, over 27823.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3877, pruned_loss=0.1365, over 5662732.75 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3608, pruned_loss=0.103, over 5761980.89 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3887, pruned_loss=0.1386, over 5655108.73 frames. ], batch size: 412, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:16:34,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4073, 1.6572, 1.3169, 1.1235], device='cuda:1'), covar=tensor([0.1642, 0.1557, 0.1165, 0.1494], device='cuda:1'), in_proj_covar=tensor([0.1674, 0.1578, 0.1526, 0.1632], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 04:16:53,522 INFO [train.py:968] (1/2) Epoch 10, batch 12300, giga_loss[loss=0.3396, simple_loss=0.4006, pruned_loss=0.1393, over 28710.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3847, pruned_loss=0.1331, over 5676573.06 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3611, pruned_loss=0.1031, over 5764297.66 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3861, pruned_loss=0.1359, over 5664957.23 frames. ], batch size: 242, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:17:34,070 INFO [optim.py:369] (1/2) 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 04:17:42,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=421943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:17:43,214 INFO [train.py:968] (1/2) Epoch 10, batch 12350, giga_loss[loss=0.3078, simple_loss=0.3764, pruned_loss=0.1195, over 28877.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3845, pruned_loss=0.1327, over 5666287.54 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3617, pruned_loss=0.1035, over 5768023.72 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3857, pruned_loss=0.1353, over 5651458.27 frames. ], batch size: 112, lr: 3.28e-03, grad_scale: 2.0
+2023-03-05 04:17:58,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:18:23,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:18:30,483 INFO [train.py:968] (1/2) Epoch 10, batch 12400, giga_loss[loss=0.34, simple_loss=0.3937, pruned_loss=0.1431, over 27558.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3834, pruned_loss=0.1311, over 5678056.23 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.361, pruned_loss=0.103, over 5770440.49 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3855, pruned_loss=0.1342, over 5662208.32 frames. ], batch size: 472, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:18:30,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=421994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:18:52,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422020.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 04:19:05,655 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 12450, giga_loss[loss=0.306, simple_loss=0.3705, pruned_loss=0.1207, over 27835.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3811, pruned_loss=0.1284, over 5684484.68 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3612, pruned_loss=0.103, over 5764646.47 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3835, pruned_loss=0.1321, over 5673556.07 frames. ], batch size: 412, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:19:45,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:19:56,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:19:57,786 INFO [train.py:968] (1/2) Epoch 10, batch 12500, giga_loss[loss=0.3719, simple_loss=0.4173, pruned_loss=0.1632, over 28221.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3804, pruned_loss=0.1287, over 5676810.55 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3609, pruned_loss=0.1028, over 5766856.71 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3835, pruned_loss=0.1329, over 5662452.12 frames. ], batch size: 368, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:20:05,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:07,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:27,962 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 04:20:32,514 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:32,778 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-05 04:20:35,304 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:37,572 INFO [optim.py:369] (1/2) 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 04:20:39,535 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:41,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:20:44,963 INFO [train.py:968] (1/2) Epoch 10, batch 12550, libri_loss[loss=0.2307, simple_loss=0.3118, pruned_loss=0.07479, over 29686.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3773, pruned_loss=0.1269, over 5676724.42 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3601, pruned_loss=0.1024, over 5769525.25 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.381, pruned_loss=0.1313, over 5659966.00 frames. ], batch size: 73, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:20:55,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1580, 1.6442, 1.2155, 0.5047], device='cuda:1'), covar=tensor([0.2171, 0.1289, 0.1641, 0.2754], device='cuda:1'), in_proj_covar=tensor([0.1539, 0.1460, 0.1483, 0.1261], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 04:20:58,912 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:21:01,557 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:21:05,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422166.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 04:21:07,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5537, 1.9625, 1.4169, 1.6784], device='cuda:1'), covar=tensor([0.0729, 0.0261, 0.0308, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0116, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 04:21:07,934 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:21:32,674 INFO [train.py:968] (1/2) Epoch 10, batch 12600, giga_loss[loss=0.2814, simple_loss=0.3475, pruned_loss=0.1076, over 28866.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3751, pruned_loss=0.1263, over 5675811.49 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3603, pruned_loss=0.1025, over 5760754.04 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3782, pruned_loss=0.1302, over 5667791.89 frames. ], batch size: 199, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:21:33,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 04:22:01,679 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:04,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:14,126 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:22:16,134 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 12650, giga_loss[loss=0.3363, simple_loss=0.3688, pruned_loss=0.1519, over 23638.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3732, pruned_loss=0.126, over 5683645.76 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3598, pruned_loss=0.1023, over 5763991.97 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3764, pruned_loss=0.1299, over 5672503.22 frames. ], batch size: 705, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:22:29,792 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:43,240 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:52,673 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:22:52,895 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-05 04:23:08,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9772, 1.6989, 5.4035, 3.6607], device='cuda:1'), covar=tensor([0.1388, 0.2282, 0.0324, 0.0702], device='cuda:1'), in_proj_covar=tensor([0.0655, 0.0587, 0.0859, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 04:23:10,515 INFO [train.py:968] (1/2) Epoch 10, batch 12700, giga_loss[loss=0.4148, simple_loss=0.4343, pruned_loss=0.1976, over 26698.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3729, pruned_loss=0.1262, over 5684016.65 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3601, pruned_loss=0.1026, over 5757253.93 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3754, pruned_loss=0.1294, over 5680600.68 frames. ], batch size: 555, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:23:33,327 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:23:34,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:23:51,112 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 12750, giga_loss[loss=0.3052, simple_loss=0.3734, pruned_loss=0.1184, over 28921.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3723, pruned_loss=0.1248, over 5683695.39 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3597, pruned_loss=0.1024, over 5760591.50 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.375, pruned_loss=0.1282, over 5675866.99 frames. ], batch size: 213, lr: 3.28e-03, grad_scale: 4.0
+2023-03-05 04:24:10,876 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:24:48,989 INFO [train.py:968] (1/2) Epoch 10, batch 12800, giga_loss[loss=0.3837, simple_loss=0.4376, pruned_loss=0.1649, over 28897.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5684838.35 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3597, pruned_loss=0.1024, over 5762142.77 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1252, over 5675737.15 frames. ], batch size: 186, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:25:05,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=422410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:25:33,074 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 12850, giga_loss[loss=0.3034, simple_loss=0.3705, pruned_loss=0.1181, over 28948.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3683, pruned_loss=0.1188, over 5662713.76 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3595, pruned_loss=0.1024, over 5753000.16 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3704, pruned_loss=0.1213, over 5662667.94 frames. ], batch size: 145, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:25:59,987 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:26:05,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:26:34,399 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:26:34,806 INFO [train.py:968] (1/2) Epoch 10, batch 12900, giga_loss[loss=0.2818, simple_loss=0.3551, pruned_loss=0.1042, over 28745.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3647, pruned_loss=0.1148, over 5663468.85 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3596, pruned_loss=0.1025, over 5754172.64 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3665, pruned_loss=0.1171, over 5660839.46 frames. ], batch size: 284, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:27:17,879 INFO [optim.py:369] (1/2) 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 04:27:22,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:27:25,004 INFO [train.py:968] (1/2) Epoch 10, batch 12950, giga_loss[loss=0.2587, simple_loss=0.3415, pruned_loss=0.08792, over 28943.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3606, pruned_loss=0.1113, over 5666611.93 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3587, pruned_loss=0.1021, over 5756863.10 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.363, pruned_loss=0.1138, over 5658793.73 frames. ], batch size: 164, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:27:35,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 04:27:49,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1200, 4.9287, 4.6414, 2.2152], device='cuda:1'), covar=tensor([0.0433, 0.0611, 0.0714, 0.1966], device='cuda:1'), in_proj_covar=tensor([0.1025, 0.0963, 0.0842, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 04:28:13,269 INFO [train.py:968] (1/2) Epoch 10, batch 13000, giga_loss[loss=0.2592, simple_loss=0.3466, pruned_loss=0.08588, over 28870.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3581, pruned_loss=0.1077, over 5672344.12 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3579, pruned_loss=0.1017, over 5762021.83 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3607, pruned_loss=0.1104, over 5658531.82 frames. ], batch size: 199, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:28:53,166 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 13050, giga_loss[loss=0.3105, simple_loss=0.3761, pruned_loss=0.1225, over 28284.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.358, pruned_loss=0.1069, over 5668790.81 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3571, pruned_loss=0.1014, over 5766576.93 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3608, pruned_loss=0.1095, over 5651216.91 frames. ], batch size: 368, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:29:10,880 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:42,386 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:45,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:51,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:29:52,860 INFO [train.py:968] (1/2) Epoch 10, batch 13100, giga_loss[loss=0.2444, simple_loss=0.3307, pruned_loss=0.079, over 28774.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3572, pruned_loss=0.1063, over 5666418.69 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3567, pruned_loss=0.1013, over 5761302.53 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3598, pruned_loss=0.1086, over 5654459.17 frames. ], batch size: 174, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:30:00,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5722, 2.0573, 1.5412, 1.4568], device='cuda:1'), covar=tensor([0.1876, 0.1312, 0.1476, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.1634, 0.1534, 0.1490, 0.1594], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 04:30:04,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 04:30:11,201 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:30:23,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:30:23,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0947, 1.5383, 1.3988, 1.0532], device='cuda:1'), covar=tensor([0.1537, 0.2276, 0.1271, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0695, 0.0842, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 04:30:26,966 INFO [optim.py:369] (1/2) 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 04:30:37,218 INFO [train.py:968] (1/2) Epoch 10, batch 13150, giga_loss[loss=0.2362, simple_loss=0.3189, pruned_loss=0.07671, over 28865.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3545, pruned_loss=0.1046, over 5667448.05 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3561, pruned_loss=0.1011, over 5762365.05 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3572, pruned_loss=0.1069, over 5652502.31 frames. ], batch size: 199, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:31:16,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=422785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:31:25,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3331, 1.9510, 1.4097, 0.6613], device='cuda:1'), covar=tensor([0.3843, 0.1998, 0.2594, 0.3788], device='cuda:1'), in_proj_covar=tensor([0.1538, 0.1452, 0.1480, 0.1261], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 04:31:25,828 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:31:26,097 INFO [train.py:968] (1/2) Epoch 10, batch 13200, giga_loss[loss=0.2548, simple_loss=0.335, pruned_loss=0.08732, over 28750.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3516, pruned_loss=0.1026, over 5678274.99 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3552, pruned_loss=0.1008, over 5766699.04 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3545, pruned_loss=0.1048, over 5659897.09 frames. ], batch size: 243, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:31:29,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:31:49,803 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9793, 1.2606, 5.2339, 3.8801], device='cuda:1'), covar=tensor([0.1448, 0.2683, 0.0351, 0.0617], device='cuda:1'), in_proj_covar=tensor([0.0648, 0.0579, 0.0844, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 04:31:59,166 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:08,830 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:32:11,949 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:17,725 INFO [train.py:968] (1/2) Epoch 10, batch 13250, giga_loss[loss=0.2487, simple_loss=0.3392, pruned_loss=0.07913, over 28992.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3509, pruned_loss=0.1021, over 5665576.66 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3549, pruned_loss=0.1007, over 5758764.09 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3534, pruned_loss=0.1039, over 5657221.10 frames. ], batch size: 155, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:32:26,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4353, 1.6839, 1.3631, 1.5874], device='cuda:1'), covar=tensor([0.2316, 0.2113, 0.2325, 0.1951], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.0935, 0.1122, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 04:32:42,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:48,376 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:32:51,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:09,108 INFO [train.py:968] (1/2) Epoch 10, batch 13300, giga_loss[loss=0.2529, simple_loss=0.333, pruned_loss=0.08637, over 27899.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3497, pruned_loss=0.1013, over 5668437.86 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3544, pruned_loss=0.1006, over 5761143.41 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3521, pruned_loss=0.1028, over 5658329.07 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:33:21,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:43,487 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=422928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:48,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=422931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:33:53,208 INFO [optim.py:369] (1/2) 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 04:34:02,474 INFO [train.py:968] (1/2) Epoch 10, batch 13350, giga_loss[loss=0.2871, simple_loss=0.361, pruned_loss=0.1066, over 28680.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3484, pruned_loss=0.09987, over 5672848.83 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3541, pruned_loss=0.1005, over 5764275.72 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3505, pruned_loss=0.1012, over 5659911.18 frames. ], batch size: 262, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:34:19,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=422960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:34:52,636 INFO [train.py:968] (1/2) Epoch 10, batch 13400, libri_loss[loss=0.2922, simple_loss=0.3693, pruned_loss=0.1075, over 29540.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3456, pruned_loss=0.09795, over 5673130.52 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3542, pruned_loss=0.1007, over 5766954.35 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3469, pruned_loss=0.09876, over 5658246.37 frames. ], batch size: 84, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:35:34,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4349, 1.7678, 1.7142, 1.3126], device='cuda:1'), covar=tensor([0.1711, 0.2285, 0.1459, 0.1660], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0689, 0.0842, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 04:35:37,235 INFO [optim.py:369] (1/2) 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 04:35:42,038 INFO [train.py:968] (1/2) Epoch 10, batch 13450, giga_loss[loss=0.233, simple_loss=0.3137, pruned_loss=0.07613, over 28360.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3418, pruned_loss=0.09605, over 5669923.61 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3536, pruned_loss=0.1005, over 5769098.12 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.343, pruned_loss=0.09671, over 5652074.71 frames. ], batch size: 65, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:36:08,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1132, 1.2884, 0.9435, 1.1208], device='cuda:1'), covar=tensor([0.1173, 0.1037, 0.0790, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.1624, 0.1510, 0.1472, 0.1575], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 04:36:14,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3700, 1.5443, 1.3948, 1.3435], device='cuda:1'), covar=tensor([0.0694, 0.0286, 0.0291, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0116, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 04:36:40,064 INFO [train.py:968] (1/2) Epoch 10, batch 13500, giga_loss[loss=0.2571, simple_loss=0.3315, pruned_loss=0.09133, over 27953.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3405, pruned_loss=0.0961, over 5656974.17 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3535, pruned_loss=0.1005, over 5769234.51 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3413, pruned_loss=0.09653, over 5641315.22 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:37:24,323 INFO [optim.py:369] (1/2) 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 04:37:33,894 INFO [train.py:968] (1/2) Epoch 10, batch 13550, giga_loss[loss=0.2367, simple_loss=0.324, pruned_loss=0.0747, over 28888.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3411, pruned_loss=0.0968, over 5653331.90 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3535, pruned_loss=0.1005, over 5771498.65 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3415, pruned_loss=0.09705, over 5635711.79 frames. ], batch size: 145, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:38:26,563 INFO [train.py:968] (1/2) Epoch 10, batch 13600, libri_loss[loss=0.3039, simple_loss=0.3694, pruned_loss=0.1191, over 29674.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3433, pruned_loss=0.09738, over 5649928.54 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3526, pruned_loss=0.1002, over 5772902.49 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.344, pruned_loss=0.09776, over 5630662.37 frames. ], batch size: 91, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:38:47,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4520, 1.8082, 1.7159, 1.6038], device='cuda:1'), covar=tensor([0.1358, 0.1360, 0.1536, 0.1479], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0714, 0.0654, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-05 04:39:02,676 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:39:25,377 INFO [train.py:968] (1/2) Epoch 10, batch 13650, giga_loss[loss=0.246, simple_loss=0.3289, pruned_loss=0.0815, over 28886.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3449, pruned_loss=0.09746, over 5652344.20 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3522, pruned_loss=0.09996, over 5774806.89 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3457, pruned_loss=0.09793, over 5631554.61 frames. ], batch size: 145, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:39:38,018 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=423255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:40:03,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6426, 1.6384, 1.2127, 1.3327], device='cuda:1'), covar=tensor([0.0741, 0.0580, 0.1049, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0434, 0.0497, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 04:40:04,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 04:40:23,803 INFO [train.py:968] (1/2) Epoch 10, batch 13700, giga_loss[loss=0.3121, simple_loss=0.3595, pruned_loss=0.1323, over 26904.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3459, pruned_loss=0.09826, over 5651257.39 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3519, pruned_loss=0.09979, over 5776599.07 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3466, pruned_loss=0.09876, over 5631475.63 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:41:16,939 INFO [optim.py:369] (1/2) 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 04:41:19,354 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 04:41:22,581 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 04:41:23,524 INFO [train.py:968] (1/2) Epoch 10, batch 13750, giga_loss[loss=0.2596, simple_loss=0.3288, pruned_loss=0.09521, over 27651.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3434, pruned_loss=0.0967, over 5660787.46 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3513, pruned_loss=0.09952, over 5776908.88 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.09731, over 5642581.27 frames. ], batch size: 472, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:42:25,791 INFO [train.py:968] (1/2) Epoch 10, batch 13800, libri_loss[loss=0.259, simple_loss=0.3379, pruned_loss=0.09011, over 29478.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3431, pruned_loss=0.09556, over 5653549.47 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3515, pruned_loss=0.09951, over 5779729.30 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3436, pruned_loss=0.09595, over 5632964.44 frames. ], batch size: 85, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:43:15,436 INFO [optim.py:369] (1/2) 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 04:43:22,584 INFO [train.py:968] (1/2) Epoch 10, batch 13850, giga_loss[loss=0.2476, simple_loss=0.3234, pruned_loss=0.08591, over 29051.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3408, pruned_loss=0.09371, over 5663045.02 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3512, pruned_loss=0.09942, over 5782333.85 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3411, pruned_loss=0.09393, over 5639760.26 frames. ], batch size: 128, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:43:32,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5278, 1.7540, 1.4835, 1.5487], device='cuda:1'), covar=tensor([0.2169, 0.1838, 0.1952, 0.1798], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.0933, 0.1121, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 04:43:39,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0731, 1.9498, 1.4447, 1.6810], device='cuda:1'), covar=tensor([0.0751, 0.0687, 0.0990, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0432, 0.0496, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 04:43:51,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=423468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:43:56,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 04:44:04,137 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-05 04:44:06,276 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4725, 1.6977, 1.5932, 1.4351], device='cuda:1'), covar=tensor([0.1190, 0.1517, 0.1857, 0.1578], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0713, 0.0652, 0.0638], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-05 04:44:20,193 INFO [train.py:968] (1/2) Epoch 10, batch 13900, giga_loss[loss=0.2707, simple_loss=0.3457, pruned_loss=0.0978, over 28093.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3383, pruned_loss=0.09356, over 5671775.95 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3502, pruned_loss=0.09898, over 5787871.60 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3389, pruned_loss=0.09387, over 5642682.49 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:45:01,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3695, 1.6856, 1.6664, 1.2664], device='cuda:1'), covar=tensor([0.1350, 0.1935, 0.1120, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0687, 0.0845, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 04:45:10,575 INFO [optim.py:369] (1/2) 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 04:45:18,668 INFO [train.py:968] (1/2) Epoch 10, batch 13950, giga_loss[loss=0.2734, simple_loss=0.3435, pruned_loss=0.1017, over 28887.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3382, pruned_loss=0.09368, over 5675948.19 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3498, pruned_loss=0.09877, over 5790008.74 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3387, pruned_loss=0.09397, over 5647513.14 frames. ], batch size: 164, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:46:15,221 INFO [train.py:968] (1/2) Epoch 10, batch 14000, libri_loss[loss=0.253, simple_loss=0.3317, pruned_loss=0.08716, over 29531.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3374, pruned_loss=0.09267, over 5681935.76 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3492, pruned_loss=0.09847, over 5791355.01 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3381, pruned_loss=0.09306, over 5656083.31 frames. ], batch size: 83, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:46:20,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=423598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:47:01,247 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=423630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:47:11,637 INFO [optim.py:369] (1/2) 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 04:47:19,359 INFO [train.py:968] (1/2) Epoch 10, batch 14050, giga_loss[loss=0.2757, simple_loss=0.3552, pruned_loss=0.09814, over 28812.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3399, pruned_loss=0.09331, over 5684202.15 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3491, pruned_loss=0.09839, over 5790013.90 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3405, pruned_loss=0.09361, over 5663401.02 frames. ], batch size: 119, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:48:24,368 INFO [train.py:968] (1/2) Epoch 10, batch 14100, giga_loss[loss=0.2309, simple_loss=0.3007, pruned_loss=0.08054, over 27581.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3387, pruned_loss=0.09247, over 5684236.05 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3482, pruned_loss=0.09791, over 5792619.65 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3397, pruned_loss=0.09303, over 5662985.61 frames. ], batch size: 472, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:49:16,534 INFO [optim.py:369] (1/2) 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 04:49:22,334 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=423741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:49:22,871 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 14150, giga_loss[loss=0.2702, simple_loss=0.3496, pruned_loss=0.0954, over 28889.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3376, pruned_loss=0.09234, over 5676153.39 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3483, pruned_loss=0.09825, over 5775566.88 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3379, pruned_loss=0.0922, over 5669353.28 frames. ], batch size: 227, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:49:27,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=423744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:50:02,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7904, 4.5967, 4.4007, 1.9642], device='cuda:1'), covar=tensor([0.0412, 0.0553, 0.0601, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.0992, 0.0931, 0.0818, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 04:50:02,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=423773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:50:02,097 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=423773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:50:05,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=423776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:50:31,188 INFO [train.py:968] (1/2) Epoch 10, batch 14200, giga_loss[loss=0.2438, simple_loss=0.3019, pruned_loss=0.09287, over 24616.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3398, pruned_loss=0.09385, over 5664086.86 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3484, pruned_loss=0.09846, over 5775969.07 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3397, pruned_loss=0.09346, over 5656552.89 frames. ], batch size: 705, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:50:48,847 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=423805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:51:30,370 INFO [optim.py:369] (1/2) 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 04:51:36,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=423843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:51:36,645 INFO [train.py:968] (1/2) Epoch 10, batch 14250, giga_loss[loss=0.2982, simple_loss=0.3838, pruned_loss=0.1063, over 28840.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3436, pruned_loss=0.09338, over 5661987.71 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3481, pruned_loss=0.09833, over 5777446.22 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3437, pruned_loss=0.09311, over 5652342.70 frames. ], batch size: 174, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:52:32,602 INFO [train.py:968] (1/2) Epoch 10, batch 14300, libri_loss[loss=0.274, simple_loss=0.3537, pruned_loss=0.09718, over 27855.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3433, pruned_loss=0.09178, over 5655955.35 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3478, pruned_loss=0.09824, over 5779162.06 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3435, pruned_loss=0.09139, over 5642025.74 frames. ], batch size: 115, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:52:50,049 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-05 04:53:30,508 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 14350, giga_loss[loss=0.2398, simple_loss=0.3292, pruned_loss=0.07524, over 29038.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3436, pruned_loss=0.09028, over 5660188.39 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.09823, over 5780206.07 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08991, over 5647099.71 frames. ], batch size: 136, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:53:37,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3380, 1.8628, 1.3764, 1.5991], device='cuda:1'), covar=tensor([0.0762, 0.0285, 0.0323, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 04:54:26,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=423986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:54:30,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=423989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:54:37,000 INFO [train.py:968] (1/2) Epoch 10, batch 14400, giga_loss[loss=0.3579, simple_loss=0.4007, pruned_loss=0.1576, over 26916.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3448, pruned_loss=0.09186, over 5667626.03 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09792, over 5783662.45 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3454, pruned_loss=0.09165, over 5650653.51 frames. ], batch size: 555, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 04:55:05,327 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=424018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:55:06,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7928, 2.5138, 1.6586, 0.8282], device='cuda:1'), covar=tensor([0.4334, 0.2320, 0.2656, 0.4285], device='cuda:1'), in_proj_covar=tensor([0.1535, 0.1459, 0.1486, 0.1266], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 04:55:30,414 INFO [optim.py:369] (1/2) 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 04:55:33,817 INFO [train.py:968] (1/2) Epoch 10, batch 14450, giga_loss[loss=0.2918, simple_loss=0.3648, pruned_loss=0.1094, over 29040.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3434, pruned_loss=0.09194, over 5667331.79 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3469, pruned_loss=0.0978, over 5773768.38 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3441, pruned_loss=0.09171, over 5659139.63 frames. ], batch size: 285, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:56:41,665 INFO [train.py:968] (1/2) Epoch 10, batch 14500, giga_loss[loss=0.355, simple_loss=0.4105, pruned_loss=0.1498, over 28732.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3439, pruned_loss=0.09332, over 5667101.92 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3466, pruned_loss=0.09775, over 5775926.37 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3445, pruned_loss=0.09304, over 5656002.29 frames. ], batch size: 119, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:57:13,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=424117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 04:57:20,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3955, 1.6322, 1.3164, 1.6092], device='cuda:1'), covar=tensor([0.2661, 0.2496, 0.2784, 0.2097], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.0932, 0.1122, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 04:57:43,695 INFO [optim.py:369] (1/2) 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 04:57:53,551 INFO [train.py:968] (1/2) Epoch 10, batch 14550, giga_loss[loss=0.2597, simple_loss=0.339, pruned_loss=0.0902, over 28975.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3434, pruned_loss=0.09335, over 5667468.58 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3467, pruned_loss=0.09785, over 5762693.57 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3438, pruned_loss=0.09289, over 5665367.42 frames. ], batch size: 145, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:59:03,319 INFO [train.py:968] (1/2) Epoch 10, batch 14600, libri_loss[loss=0.2423, simple_loss=0.3205, pruned_loss=0.08201, over 29552.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3392, pruned_loss=0.09116, over 5673641.89 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3462, pruned_loss=0.09762, over 5768847.45 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3398, pruned_loss=0.09075, over 5662314.99 frames. ], batch size: 79, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 04:59:26,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2858, 1.6080, 1.2219, 1.3076], device='cuda:1'), covar=tensor([0.2444, 0.2316, 0.2699, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.0941, 0.1129, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 04:59:30,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 04:59:36,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 04:59:50,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-05 05:00:01,775 INFO [optim.py:369] (1/2) 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 05:00:10,843 INFO [train.py:968] (1/2) Epoch 10, batch 14650, giga_loss[loss=0.2321, simple_loss=0.3173, pruned_loss=0.07341, over 28878.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3378, pruned_loss=0.09064, over 5669066.70 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3463, pruned_loss=0.09777, over 5769040.99 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.338, pruned_loss=0.09005, over 5658191.42 frames. ], batch size: 174, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:00:34,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=424260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:00:36,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=424263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:01:10,311 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 14700, libri_loss[loss=0.2333, simple_loss=0.3099, pruned_loss=0.0784, over 29569.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.337, pruned_loss=0.09073, over 5679690.88 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3455, pruned_loss=0.09744, over 5769444.64 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3375, pruned_loss=0.0903, over 5666568.31 frames. ], batch size: 77, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:02:07,975 INFO [optim.py:369] (1/2) 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 05:02:15,421 INFO [train.py:968] (1/2) Epoch 10, batch 14750, giga_loss[loss=0.2578, simple_loss=0.3197, pruned_loss=0.09789, over 24702.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.342, pruned_loss=0.09335, over 5681657.42 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.09789, over 5770947.27 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.342, pruned_loss=0.09256, over 5668786.12 frames. ], batch size: 705, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:03:16,115 INFO [train.py:968] (1/2) Epoch 10, batch 14800, giga_loss[loss=0.2563, simple_loss=0.3243, pruned_loss=0.0942, over 28758.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3408, pruned_loss=0.0934, over 5688110.67 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09793, over 5774218.74 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3406, pruned_loss=0.09263, over 5672603.93 frames. ], batch size: 243, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 05:03:21,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 05:04:09,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5664, 1.9102, 1.6193, 1.3367], device='cuda:1'), covar=tensor([0.1995, 0.1368, 0.1111, 0.1574], device='cuda:1'), in_proj_covar=tensor([0.1624, 0.1509, 0.1448, 0.1574], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 05:04:11,167 INFO [optim.py:369] (1/2) 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 05:04:17,412 INFO [train.py:968] (1/2) Epoch 10, batch 14850, giga_loss[loss=0.271, simple_loss=0.3456, pruned_loss=0.09818, over 28565.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3411, pruned_loss=0.09452, over 5677838.14 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3458, pruned_loss=0.09773, over 5774751.46 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3409, pruned_loss=0.09394, over 5662080.73 frames. ], batch size: 336, lr: 3.27e-03, grad_scale: 8.0
+2023-03-05 05:04:40,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-05 05:05:14,489 INFO [train.py:968] (1/2) Epoch 10, batch 14900, libri_loss[loss=0.2586, simple_loss=0.3339, pruned_loss=0.0916, over 29752.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3404, pruned_loss=0.09409, over 5685484.72 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3454, pruned_loss=0.09755, over 5779848.93 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3405, pruned_loss=0.09366, over 5664400.93 frames. ], batch size: 87, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:05:36,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4056, 1.5860, 1.2901, 1.6153], device='cuda:1'), covar=tensor([0.2425, 0.2279, 0.2483, 0.2081], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.0933, 0.1122, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 05:06:13,079 INFO [optim.py:369] (1/2) 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 05:06:16,619 INFO [train.py:968] (1/2) Epoch 10, batch 14950, giga_loss[loss=0.2624, simple_loss=0.3461, pruned_loss=0.08942, over 28603.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3419, pruned_loss=0.09391, over 5689970.77 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3453, pruned_loss=0.09759, over 5783384.52 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3419, pruned_loss=0.09342, over 5667131.44 frames. ], batch size: 92, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:07:29,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 05:07:32,500 INFO [train.py:968] (1/2) Epoch 10, batch 15000, giga_loss[loss=0.2472, simple_loss=0.3287, pruned_loss=0.08288, over 28925.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3438, pruned_loss=0.09456, over 5685641.12 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.09772, over 5784323.79 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3436, pruned_loss=0.094, over 5664946.29 frames. ], batch size: 213, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:07:32,501 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 05:07:41,003 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 05:08:49,833 INFO [optim.py:369] (1/2) 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 05:08:52,374 INFO [train.py:968] (1/2) Epoch 10, batch 15050, giga_loss[loss=0.2542, simple_loss=0.3207, pruned_loss=0.09379, over 28791.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3409, pruned_loss=0.09314, over 5685726.68 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3447, pruned_loss=0.09725, over 5783789.72 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3415, pruned_loss=0.09303, over 5666886.81 frames. ], batch size: 99, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:10:00,015 INFO [train.py:968] (1/2) Epoch 10, batch 15100, giga_loss[loss=0.2925, simple_loss=0.3568, pruned_loss=0.1141, over 27753.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3374, pruned_loss=0.0922, over 5689669.13 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.345, pruned_loss=0.09765, over 5777474.91 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3375, pruned_loss=0.09162, over 5678706.68 frames. ], batch size: 474, lr: 3.27e-03, grad_scale: 2.0
+2023-03-05 05:10:01,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3569, 3.3066, 1.5218, 1.4876], device='cuda:1'), covar=tensor([0.0899, 0.0292, 0.0834, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0498, 0.0333, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 05:10:17,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 05:11:01,817 INFO [optim.py:369] (1/2) 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 05:11:03,862 INFO [train.py:968] (1/2) Epoch 10, batch 15150, giga_loss[loss=0.2518, simple_loss=0.3366, pruned_loss=0.08355, over 28863.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3326, pruned_loss=0.09013, over 5686071.56 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3448, pruned_loss=0.09767, over 5778732.97 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3326, pruned_loss=0.08951, over 5674229.29 frames. ], batch size: 244, lr: 3.27e-03, grad_scale: 2.0
+2023-03-05 05:11:19,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6983, 1.6744, 1.4644, 1.7863], device='cuda:1'), covar=tensor([0.2316, 0.2430, 0.2504, 0.2330], device='cuda:1'), in_proj_covar=tensor([0.1268, 0.0941, 0.1130, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 05:12:02,895 INFO [train.py:968] (1/2) Epoch 10, batch 15200, giga_loss[loss=0.2587, simple_loss=0.3317, pruned_loss=0.09288, over 28962.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3342, pruned_loss=0.09156, over 5688405.00 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3447, pruned_loss=0.09763, over 5782587.05 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.334, pruned_loss=0.09086, over 5672449.32 frames. ], batch size: 164, lr: 3.27e-03, grad_scale: 4.0
+2023-03-05 05:12:41,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5215, 1.7643, 0.9889, 1.4782], device='cuda:1'), covar=tensor([0.1000, 0.0759, 0.1626, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0429, 0.0495, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 05:12:49,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0641, 1.5758, 1.3836, 1.0764], device='cuda:1'), covar=tensor([0.1143, 0.1775, 0.1005, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0683, 0.0845, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 05:12:57,471 INFO [optim.py:369] (1/2) 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 05:12:57,490 INFO [train.py:968] (1/2) Epoch 10, batch 15250, giga_loss[loss=0.2737, simple_loss=0.3463, pruned_loss=0.1005, over 28515.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3361, pruned_loss=0.09305, over 5683044.54 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3445, pruned_loss=0.09758, over 5782896.71 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3359, pruned_loss=0.09248, over 5668995.87 frames. ], batch size: 336, lr: 3.27e-03, grad_scale: 1.0
+2023-03-05 05:14:04,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=424892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:14:06,033 INFO [train.py:968] (1/2) Epoch 10, batch 15300, giga_loss[loss=0.23, simple_loss=0.3137, pruned_loss=0.07321, over 28149.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3331, pruned_loss=0.09059, over 5674034.50 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3444, pruned_loss=0.0975, over 5783560.23 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.333, pruned_loss=0.09018, over 5661942.39 frames. ], batch size: 412, lr: 3.27e-03, grad_scale: 1.0
+2023-03-05 05:14:24,329 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-05 05:14:49,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-05 05:15:02,505 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 15350, giga_loss[loss=0.2092, simple_loss=0.2742, pruned_loss=0.07208, over 24205.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3312, pruned_loss=0.08914, over 5671737.41 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3443, pruned_loss=0.09769, over 5784268.43 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3309, pruned_loss=0.08842, over 5658639.03 frames. ], batch size: 705, lr: 3.27e-03, grad_scale: 1.0
+2023-03-05 05:15:29,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0422, 1.1654, 3.4832, 3.0776], device='cuda:1'), covar=tensor([0.1594, 0.2502, 0.0468, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0633, 0.0571, 0.0822, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 05:16:04,834 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 05:16:14,887 INFO [train.py:968] (1/2) Epoch 10, batch 15400, libri_loss[loss=0.2787, simple_loss=0.352, pruned_loss=0.1027, over 27793.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.33, pruned_loss=0.08862, over 5676053.07 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3438, pruned_loss=0.09742, over 5785886.58 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3299, pruned_loss=0.08809, over 5661934.60 frames. ], batch size: 116, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:16:20,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2458, 1.4941, 1.3312, 1.0693], device='cuda:1'), covar=tensor([0.1663, 0.1454, 0.0923, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.1623, 0.1502, 0.1433, 0.1569], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 05:17:21,711 INFO [optim.py:369] (1/2) 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 05:17:21,724 INFO [train.py:968] (1/2) Epoch 10, batch 15450, libri_loss[loss=0.2803, simple_loss=0.3568, pruned_loss=0.1019, over 29216.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3313, pruned_loss=0.0887, over 5692464.41 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3433, pruned_loss=0.0971, over 5789359.98 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3313, pruned_loss=0.08832, over 5675167.39 frames. ], batch size: 97, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:18:18,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-05 05:18:26,208 INFO [train.py:968] (1/2) Epoch 10, batch 15500, giga_loss[loss=0.3063, simple_loss=0.3605, pruned_loss=0.126, over 26876.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3313, pruned_loss=0.08887, over 5697555.02 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.343, pruned_loss=0.09697, over 5791098.11 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3314, pruned_loss=0.08857, over 5681154.20 frames. ], batch size: 555, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:19:05,158 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=425120.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:19:20,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5858, 1.5537, 1.1819, 1.2509], device='cuda:1'), covar=tensor([0.0680, 0.0469, 0.0969, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0430, 0.0499, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 05:19:35,494 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 15550, giga_loss[loss=0.2184, simple_loss=0.3063, pruned_loss=0.06529, over 29017.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3316, pruned_loss=0.08928, over 5688239.00 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3431, pruned_loss=0.09711, over 5782094.62 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3315, pruned_loss=0.0889, over 5683193.22 frames. ], batch size: 155, lr: 3.26e-03, grad_scale: 1.0
+2023-03-05 05:20:35,085 INFO [train.py:968] (1/2) Epoch 10, batch 15600, giga_loss[loss=0.2549, simple_loss=0.3561, pruned_loss=0.07686, over 28949.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3301, pruned_loss=0.08771, over 5673930.24 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3426, pruned_loss=0.09675, over 5781988.16 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3302, pruned_loss=0.08747, over 5667399.45 frames. ], batch size: 174, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:21:02,035 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=425215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:21:38,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3258, 1.6972, 1.6592, 1.2174], device='cuda:1'), covar=tensor([0.1610, 0.2130, 0.1307, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0680, 0.0843, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 05:21:39,141 INFO [optim.py:369] (1/2) 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 05:21:39,156 INFO [train.py:968] (1/2) Epoch 10, batch 15650, giga_loss[loss=0.263, simple_loss=0.351, pruned_loss=0.08747, over 28569.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3333, pruned_loss=0.0885, over 5665146.60 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3427, pruned_loss=0.09695, over 5783071.03 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3331, pruned_loss=0.08805, over 5657977.09 frames. ], batch size: 307, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:21:48,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2682, 1.6237, 1.5602, 1.1761], device='cuda:1'), covar=tensor([0.1518, 0.2062, 0.1230, 0.1486], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0681, 0.0843, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 05:21:57,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4997, 3.1812, 2.3646, 1.8685], device='cuda:1'), covar=tensor([0.1625, 0.0886, 0.1022, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.1653, 0.1526, 0.1456, 0.1596], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 05:22:10,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=425267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:22:41,679 INFO [train.py:968] (1/2) Epoch 10, batch 15700, giga_loss[loss=0.2324, simple_loss=0.3232, pruned_loss=0.07076, over 29033.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3359, pruned_loss=0.0898, over 5667191.23 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3426, pruned_loss=0.09683, over 5785196.88 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3357, pruned_loss=0.08942, over 5657599.02 frames. ], batch size: 155, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:23:41,671 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 15750, libri_loss[loss=0.29, simple_loss=0.3623, pruned_loss=0.1089, over 29204.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3368, pruned_loss=0.09036, over 5654653.43 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3429, pruned_loss=0.09687, over 5776995.38 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3362, pruned_loss=0.08986, over 5650819.51 frames. ], batch size: 97, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:24:34,504 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-05 05:24:42,663 INFO [train.py:968] (1/2) Epoch 10, batch 15800, giga_loss[loss=0.2316, simple_loss=0.3115, pruned_loss=0.07582, over 28820.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3361, pruned_loss=0.09034, over 5652985.56 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3424, pruned_loss=0.09662, over 5779118.32 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3361, pruned_loss=0.09009, over 5646459.31 frames. ], batch size: 119, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:25:02,975 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=425410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:25:06,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=425413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:25:43,337 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=425442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:25:45,507 INFO [optim.py:369] (1/2) 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 05:25:45,525 INFO [train.py:968] (1/2) Epoch 10, batch 15850, giga_loss[loss=0.2539, simple_loss=0.3326, pruned_loss=0.08763, over 28409.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3332, pruned_loss=0.0887, over 5655620.15 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3423, pruned_loss=0.09666, over 5781269.44 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3331, pruned_loss=0.08831, over 5646065.65 frames. ], batch size: 368, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:26:36,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4716, 1.7388, 1.5953, 1.5740], device='cuda:1'), covar=tensor([0.1232, 0.1734, 0.1485, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0706, 0.0639, 0.0629], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-05 05:26:49,281 INFO [train.py:968] (1/2) Epoch 10, batch 15900, giga_loss[loss=0.2627, simple_loss=0.3408, pruned_loss=0.09226, over 28669.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3316, pruned_loss=0.08835, over 5657651.08 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.09648, over 5772943.00 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3317, pruned_loss=0.08814, over 5656048.46 frames. ], batch size: 242, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:26:50,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=425495.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:27:48,039 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 15950, giga_loss[loss=0.2707, simple_loss=0.3509, pruned_loss=0.09529, over 27630.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3321, pruned_loss=0.08843, over 5658696.09 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.09649, over 5767305.58 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3318, pruned_loss=0.08804, over 5659307.58 frames. ], batch size: 472, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:28:25,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3935, 1.6689, 1.2892, 1.7853], device='cuda:1'), covar=tensor([0.2394, 0.2252, 0.2433, 0.2235], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.0923, 0.1118, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 05:28:46,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=425590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:28:52,334 INFO [train.py:968] (1/2) Epoch 10, batch 16000, giga_loss[loss=0.2629, simple_loss=0.3468, pruned_loss=0.08946, over 28606.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3347, pruned_loss=0.08959, over 5666502.73 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3418, pruned_loss=0.0964, over 5767831.15 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3346, pruned_loss=0.08925, over 5665197.32 frames. ], batch size: 307, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:29:12,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-05 05:29:51,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=425638.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:29:56,884 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:29:58,833 INFO [train.py:968] (1/2) Epoch 10, batch 16050, giga_loss[loss=0.2346, simple_loss=0.2926, pruned_loss=0.08825, over 24461.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3351, pruned_loss=0.0907, over 5652266.39 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3416, pruned_loss=0.09623, over 5766185.80 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.335, pruned_loss=0.09044, over 5650145.90 frames. ], batch size: 705, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:30:28,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6889, 1.8773, 1.6444, 1.7158], device='cuda:1'), covar=tensor([0.1200, 0.1918, 0.1596, 0.1586], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0707, 0.0640, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-05 05:30:34,717 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=425670.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:30:47,536 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 16100, giga_loss[loss=0.3052, simple_loss=0.3744, pruned_loss=0.1181, over 28425.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3384, pruned_loss=0.09245, over 5658397.71 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.0965, over 5768321.05 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3379, pruned_loss=0.09191, over 5653190.58 frames. ], batch size: 368, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:31:43,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=425733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:31:46,085 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 10, batch 16150, giga_loss[loss=0.2436, simple_loss=0.3367, pruned_loss=0.07524, over 28795.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3412, pruned_loss=0.09402, over 5654648.48 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3417, pruned_loss=0.09638, over 5771966.55 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.341, pruned_loss=0.09359, over 5642814.21 frames. ], batch size: 174, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:32:19,594 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=425765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:32:53,690 INFO [train.py:968] (1/2) Epoch 10, batch 16200, giga_loss[loss=0.2709, simple_loss=0.3486, pruned_loss=0.09662, over 28915.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3424, pruned_loss=0.09411, over 5660083.24 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3411, pruned_loss=0.09602, over 5774406.46 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3428, pruned_loss=0.09401, over 5644275.33 frames. ], batch size: 145, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:32:54,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.53 vs. limit=5.0
+2023-03-05 05:34:05,494 INFO [optim.py:369] (1/2) 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 05:34:05,506 INFO [train.py:968] (1/2) Epoch 10, batch 16250, libri_loss[loss=0.2126, simple_loss=0.291, pruned_loss=0.06709, over 29491.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3407, pruned_loss=0.09345, over 5658846.86 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3409, pruned_loss=0.09587, over 5776183.63 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3413, pruned_loss=0.09348, over 5642251.67 frames. ], batch size: 70, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:34:47,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3350, 3.1730, 3.0050, 1.5368], device='cuda:1'), covar=tensor([0.0857, 0.0979, 0.0889, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1004, 0.0932, 0.0823, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 05:35:09,396 INFO [train.py:968] (1/2) Epoch 10, batch 16300, giga_loss[loss=0.2574, simple_loss=0.3146, pruned_loss=0.1001, over 24614.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3392, pruned_loss=0.09315, over 5664497.18 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3408, pruned_loss=0.09575, over 5777918.52 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3396, pruned_loss=0.09324, over 5647856.59 frames. ], batch size: 705, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:36:14,560 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 16350, giga_loss[loss=0.269, simple_loss=0.3503, pruned_loss=0.09385, over 28929.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3381, pruned_loss=0.09212, over 5680340.94 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3405, pruned_loss=0.09561, over 5782481.14 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3387, pruned_loss=0.0922, over 5659691.05 frames. ], batch size: 186, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:36:21,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7241, 4.3110, 1.7706, 1.7945], device='cuda:1'), covar=tensor([0.0844, 0.0291, 0.0802, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0500, 0.0333, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 05:37:15,207 INFO [train.py:968] (1/2) Epoch 10, batch 16400, giga_loss[loss=0.2409, simple_loss=0.3231, pruned_loss=0.07939, over 28982.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3379, pruned_loss=0.09301, over 5675098.32 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3399, pruned_loss=0.09529, over 5784182.64 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3389, pruned_loss=0.0933, over 5654276.59 frames. ], batch size: 120, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 05:37:49,884 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:37:59,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3318, 1.5117, 1.2653, 1.3302], device='cuda:1'), covar=tensor([0.1707, 0.1390, 0.1293, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.1648, 0.1510, 0.1450, 0.1588], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 05:38:15,976 INFO [train.py:968] (1/2) Epoch 10, batch 16450, giga_loss[loss=0.2675, simple_loss=0.3445, pruned_loss=0.09521, over 28798.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3357, pruned_loss=0.09228, over 5672106.11 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3398, pruned_loss=0.09515, over 5786106.05 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3365, pruned_loss=0.09257, over 5650905.49 frames. ], batch size: 243, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:38:16,685 INFO [optim.py:369] (1/2) 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 05:38:32,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426056.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:38:59,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0734, 1.1608, 3.8416, 3.0414], device='cuda:1'), covar=tensor([0.1698, 0.2469, 0.0392, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0641, 0.0573, 0.0828, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 05:39:20,187 INFO [train.py:968] (1/2) Epoch 10, batch 16500, giga_loss[loss=0.2433, simple_loss=0.3307, pruned_loss=0.07799, over 28794.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.335, pruned_loss=0.09114, over 5671549.47 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3395, pruned_loss=0.09496, over 5787099.78 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3358, pruned_loss=0.09147, over 5650901.62 frames. ], batch size: 243, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:40:17,462 INFO [train.py:968] (1/2) Epoch 10, batch 16550, giga_loss[loss=0.2642, simple_loss=0.3563, pruned_loss=0.08606, over 29009.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3348, pruned_loss=0.0899, over 5680425.28 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3396, pruned_loss=0.09502, over 5789368.13 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3352, pruned_loss=0.09001, over 5659854.58 frames. ], batch size: 120, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:40:19,493 INFO [optim.py:369] (1/2) 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 05:40:41,270 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:40:55,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1229, 1.3277, 4.0386, 3.0767], device='cuda:1'), covar=tensor([0.1609, 0.2351, 0.0332, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0640, 0.0575, 0.0824, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 05:41:14,845 INFO [train.py:968] (1/2) Epoch 10, batch 16600, giga_loss[loss=0.2626, simple_loss=0.3551, pruned_loss=0.0851, over 28374.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3354, pruned_loss=0.08763, over 5693551.40 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3393, pruned_loss=0.09486, over 5792115.45 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3359, pruned_loss=0.08771, over 5672331.26 frames. ], batch size: 368, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:41:21,795 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426199.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:41:24,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426202.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 05:41:57,125 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:42:10,364 INFO [train.py:968] (1/2) Epoch 10, batch 16650, giga_loss[loss=0.2576, simple_loss=0.3399, pruned_loss=0.08762, over 29053.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3379, pruned_loss=0.08893, over 5693017.66 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3387, pruned_loss=0.09457, over 5796446.93 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3388, pruned_loss=0.08901, over 5668540.80 frames. ], batch size: 155, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:42:11,098 INFO [optim.py:369] (1/2) 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 05:42:34,093 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:43:10,186 INFO [train.py:968] (1/2) Epoch 10, batch 16700, giga_loss[loss=0.247, simple_loss=0.3294, pruned_loss=0.08234, over 28347.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3379, pruned_loss=0.08894, over 5687167.85 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3391, pruned_loss=0.09489, over 5798792.72 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3382, pruned_loss=0.08856, over 5662782.23 frames. ], batch size: 368, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:44:15,257 INFO [train.py:968] (1/2) Epoch 10, batch 16750, giga_loss[loss=0.2516, simple_loss=0.3339, pruned_loss=0.08467, over 28540.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3382, pruned_loss=0.08945, over 5680791.54 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3393, pruned_loss=0.09511, over 5797512.84 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3382, pruned_loss=0.08878, over 5658731.46 frames. ], batch size: 71, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:44:16,028 INFO [optim.py:369] (1/2) 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 05:45:05,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:45:26,467 INFO [train.py:968] (1/2) Epoch 10, batch 16800, giga_loss[loss=0.2463, simple_loss=0.3337, pruned_loss=0.0795, over 28554.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3374, pruned_loss=0.08888, over 5671856.54 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.339, pruned_loss=0.09499, over 5797625.17 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3377, pruned_loss=0.08836, over 5651837.58 frames. ], batch size: 336, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 05:45:33,311 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:46:30,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5576, 1.5078, 1.1914, 1.2374], device='cuda:1'), covar=tensor([0.0607, 0.0371, 0.0803, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0427, 0.0498, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 05:46:42,077 INFO [train.py:968] (1/2) Epoch 10, batch 16850, giga_loss[loss=0.2187, simple_loss=0.3149, pruned_loss=0.06124, over 28773.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3374, pruned_loss=0.08803, over 5666304.81 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3389, pruned_loss=0.09496, over 5798193.05 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3377, pruned_loss=0.08762, over 5649455.02 frames. ], batch size: 307, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 05:46:42,714 INFO [optim.py:369] (1/2) 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 05:47:49,758 INFO [train.py:968] (1/2) Epoch 10, batch 16900, giga_loss[loss=0.2296, simple_loss=0.3144, pruned_loss=0.07246, over 28144.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3395, pruned_loss=0.08907, over 5663431.56 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3384, pruned_loss=0.09471, over 5791566.83 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3402, pruned_loss=0.08877, over 5651782.31 frames. ], batch size: 412, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:48:48,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:52,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=426541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:52,973 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:54,984 INFO [train.py:968] (1/2) Epoch 10, batch 16950, giga_loss[loss=0.238, simple_loss=0.3241, pruned_loss=0.07598, over 27611.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3415, pruned_loss=0.08992, over 5671120.82 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3383, pruned_loss=0.09467, over 5794047.86 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3423, pruned_loss=0.08953, over 5654828.69 frames. ], batch size: 472, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:48:55,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:48:56,369 INFO [optim.py:369] (1/2) 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 05:49:32,546 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:49:35,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3974, 1.6367, 1.3185, 1.9221], device='cuda:1'), covar=tensor([0.2093, 0.2035, 0.2229, 0.1919], device='cuda:1'), in_proj_covar=tensor([0.1248, 0.0925, 0.1119, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 05:49:56,588 INFO [train.py:968] (1/2) Epoch 10, batch 17000, libri_loss[loss=0.1985, simple_loss=0.2784, pruned_loss=0.05931, over 28561.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3399, pruned_loss=0.08932, over 5680041.15 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3378, pruned_loss=0.09441, over 5787024.47 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3411, pruned_loss=0.08908, over 5668924.09 frames. ], batch size: 63, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:50:33,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4615, 1.7221, 1.4801, 1.7410], device='cuda:1'), covar=tensor([0.0710, 0.0276, 0.0306, 0.0787], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0055, 0.0050, 0.0086], device='cuda:1')
+2023-03-05 05:50:35,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3396, 3.1664, 3.0137, 1.3294], device='cuda:1'), covar=tensor([0.0818, 0.0894, 0.0880, 0.2223], device='cuda:1'), in_proj_covar=tensor([0.0987, 0.0918, 0.0810, 0.0630], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 05:50:56,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:51:02,078 INFO [train.py:968] (1/2) Epoch 10, batch 17050, giga_loss[loss=0.2598, simple_loss=0.3405, pruned_loss=0.08958, over 28505.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3378, pruned_loss=0.08896, over 5681223.28 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3375, pruned_loss=0.09406, over 5791053.32 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3391, pruned_loss=0.08894, over 5665020.32 frames. ], batch size: 370, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:51:08,150 INFO [optim.py:369] (1/2) 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 05:51:55,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:51:58,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:52:12,334 INFO [train.py:968] (1/2) Epoch 10, batch 17100, giga_loss[loss=0.217, simple_loss=0.314, pruned_loss=0.05999, over 28724.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3358, pruned_loss=0.08738, over 5687988.21 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.337, pruned_loss=0.09379, over 5795171.45 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3373, pruned_loss=0.08743, over 5667861.10 frames. ], batch size: 307, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:52:40,828 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 17150, giga_loss[loss=0.2223, simple_loss=0.2992, pruned_loss=0.07272, over 28748.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.335, pruned_loss=0.08701, over 5685217.13 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3365, pruned_loss=0.09361, over 5797422.71 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3366, pruned_loss=0.0871, over 5665183.07 frames. ], batch size: 92, lr: 3.26e-03, grad_scale: 2.0
+2023-03-05 05:53:26,469 INFO [optim.py:369] (1/2) 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 05:53:36,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:53:50,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3997, 4.2038, 3.9946, 1.7487], device='cuda:1'), covar=tensor([0.0584, 0.0706, 0.0859, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.0981, 0.0908, 0.0802, 0.0626], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 05:54:11,052 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:54:14,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:54:18,573 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 17200, giga_loss[loss=0.2806, simple_loss=0.3625, pruned_loss=0.09942, over 28912.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3371, pruned_loss=0.08832, over 5684630.27 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3366, pruned_loss=0.09367, over 5798708.50 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3382, pruned_loss=0.08828, over 5666615.85 frames. ], batch size: 155, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:54:40,782 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 05:54:49,123 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:55:24,099 INFO [train.py:968] (1/2) Epoch 10, batch 17250, giga_loss[loss=0.2431, simple_loss=0.3306, pruned_loss=0.07784, over 28956.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3396, pruned_loss=0.08967, over 5672889.37 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3365, pruned_loss=0.09375, over 5791287.41 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3406, pruned_loss=0.08948, over 5663156.53 frames. ], batch size: 164, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:55:26,868 INFO [optim.py:369] (1/2) 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 05:56:05,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-05 05:56:18,490 INFO [train.py:968] (1/2) Epoch 10, batch 17300, giga_loss[loss=0.2607, simple_loss=0.3353, pruned_loss=0.09307, over 28888.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3376, pruned_loss=0.08969, over 5676241.90 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3363, pruned_loss=0.09371, over 5792810.18 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3387, pruned_loss=0.08947, over 5664443.21 frames. ], batch size: 186, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:56:22,855 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=426897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:56:24,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=426900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:56:44,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=426916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:56:47,343 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-05 05:56:58,745 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=426929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:57:17,925 INFO [train.py:968] (1/2) Epoch 10, batch 17350, giga_loss[loss=0.2785, simple_loss=0.3539, pruned_loss=0.1016, over 28921.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3358, pruned_loss=0.08976, over 5661387.20 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3365, pruned_loss=0.0939, over 5784084.89 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3365, pruned_loss=0.08933, over 5658792.43 frames. ], batch size: 213, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:57:20,602 INFO [optim.py:369] (1/2) 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 05:57:40,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6378, 1.7870, 1.2455, 1.4363], device='cuda:1'), covar=tensor([0.0731, 0.0529, 0.0875, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0431, 0.0500, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 05:58:14,411 INFO [train.py:968] (1/2) Epoch 10, batch 17400, giga_loss[loss=0.3196, simple_loss=0.3868, pruned_loss=0.1262, over 28639.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3373, pruned_loss=0.09158, over 5651635.79 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3369, pruned_loss=0.09414, over 5782707.44 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3375, pruned_loss=0.09094, over 5648242.05 frames. ], batch size: 307, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:58:22,025 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427001.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 05:59:11,116 INFO [train.py:968] (1/2) Epoch 10, batch 17450, giga_loss[loss=0.3188, simple_loss=0.3905, pruned_loss=0.1236, over 28457.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3456, pruned_loss=0.09647, over 5663270.14 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3363, pruned_loss=0.09371, over 5785709.98 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3463, pruned_loss=0.09633, over 5655225.19 frames. ], batch size: 65, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 05:59:13,517 INFO [optim.py:369] (1/2) 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 05:59:24,284 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:59:27,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:59:53,621 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 05:59:55,343 INFO [train.py:968] (1/2) Epoch 10, batch 17500, giga_loss[loss=0.3465, simple_loss=0.4207, pruned_loss=0.1362, over 28881.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3547, pruned_loss=0.1018, over 5671596.29 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3361, pruned_loss=0.0935, over 5786768.05 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3558, pruned_loss=0.102, over 5662088.35 frames. ], batch size: 112, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:00:04,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3748, 1.9543, 1.5042, 0.4984], device='cuda:1'), covar=tensor([0.3379, 0.1985, 0.2859, 0.4091], device='cuda:1'), in_proj_covar=tensor([0.1519, 0.1447, 0.1465, 0.1248], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 06:00:38,711 INFO [train.py:968] (1/2) Epoch 10, batch 17550, giga_loss[loss=0.2395, simple_loss=0.307, pruned_loss=0.08603, over 28680.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3564, pruned_loss=0.104, over 5660779.52 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3362, pruned_loss=0.09366, over 5778427.39 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3576, pruned_loss=0.1042, over 5659138.05 frames. ], batch size: 71, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:00:40,529 INFO [optim.py:369] (1/2) 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 06:00:42,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4278, 1.8736, 1.7449, 1.3010], device='cuda:1'), covar=tensor([0.1595, 0.2063, 0.1307, 0.1522], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0679, 0.0839, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 06:00:56,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:01:18,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1229, 4.8833, 4.6237, 1.9496], device='cuda:1'), covar=tensor([0.0353, 0.0537, 0.0581, 0.2078], device='cuda:1'), in_proj_covar=tensor([0.0994, 0.0922, 0.0813, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 06:01:23,593 INFO [train.py:968] (1/2) Epoch 10, batch 17600, libri_loss[loss=0.294, simple_loss=0.3707, pruned_loss=0.1086, over 29298.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3518, pruned_loss=0.1023, over 5671493.48 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3364, pruned_loss=0.09382, over 5781416.62 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.353, pruned_loss=0.1025, over 5664463.89 frames. ], batch size: 94, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:01:57,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-05 06:02:05,556 INFO [train.py:968] (1/2) Epoch 10, batch 17650, giga_loss[loss=0.2271, simple_loss=0.2962, pruned_loss=0.07893, over 28599.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3448, pruned_loss=0.09916, over 5669999.93 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3369, pruned_loss=0.09405, over 5766502.33 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3457, pruned_loss=0.09937, over 5674746.48 frames. ], batch size: 85, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:02:10,159 INFO [optim.py:369] (1/2) 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 06:02:48,636 INFO [train.py:968] (1/2) Epoch 10, batch 17700, giga_loss[loss=0.2408, simple_loss=0.3059, pruned_loss=0.08782, over 27553.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3376, pruned_loss=0.09566, over 5684944.96 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3367, pruned_loss=0.09373, over 5771675.97 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3386, pruned_loss=0.09623, over 5681196.76 frames. ], batch size: 472, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:03:03,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:03:05,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:03:31,199 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 17750, giga_loss[loss=0.2242, simple_loss=0.2963, pruned_loss=0.07604, over 28706.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3301, pruned_loss=0.09244, over 5688658.77 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.337, pruned_loss=0.09388, over 5773741.90 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3306, pruned_loss=0.09274, over 5682223.10 frames. ], batch size: 242, lr: 3.26e-03, grad_scale: 8.0
+2023-03-05 06:03:37,900 INFO [optim.py:369] (1/2) 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 06:04:05,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427376.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:04:06,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3380, 1.5190, 1.5256, 1.3577], device='cuda:1'), covar=tensor([0.1361, 0.1469, 0.1923, 0.1560], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0721, 0.0653, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-05 06:04:08,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:04:18,154 INFO [train.py:968] (1/2) Epoch 10, batch 17800, giga_loss[loss=0.2288, simple_loss=0.2959, pruned_loss=0.0809, over 28720.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3241, pruned_loss=0.08937, over 5692591.66 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3373, pruned_loss=0.09386, over 5776507.11 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3237, pruned_loss=0.0895, over 5681896.04 frames. ], batch size: 99, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:04:21,974 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-05 06:05:00,669 INFO [train.py:968] (1/2) Epoch 10, batch 17850, giga_loss[loss=0.2528, simple_loss=0.319, pruned_loss=0.09333, over 28805.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.321, pruned_loss=0.08818, over 5679552.63 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3381, pruned_loss=0.09428, over 5761661.64 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3198, pruned_loss=0.0878, over 5683348.44 frames. ], batch size: 112, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:05:03,526 INFO [optim.py:369] (1/2) 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 06:05:03,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:05:31,419 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-05 06:05:43,577 INFO [train.py:968] (1/2) Epoch 10, batch 17900, libri_loss[loss=0.2976, simple_loss=0.369, pruned_loss=0.1131, over 28728.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3189, pruned_loss=0.08695, over 5682515.98 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3386, pruned_loss=0.09452, over 5752873.86 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3169, pruned_loss=0.08627, over 5691256.61 frames. ], batch size: 106, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:05:44,040 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-05 06:06:06,222 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427519.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:06:08,479 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427522.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:06:25,791 INFO [train.py:968] (1/2) Epoch 10, batch 17950, giga_loss[loss=0.2438, simple_loss=0.3154, pruned_loss=0.08609, over 28325.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3154, pruned_loss=0.08537, over 5682686.05 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.339, pruned_loss=0.0947, over 5755423.78 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3132, pruned_loss=0.08453, over 5686181.93 frames. ], batch size: 368, lr: 3.26e-03, grad_scale: 4.0
+2023-03-05 06:06:29,985 INFO [optim.py:369] (1/2) 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 06:06:32,293 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427551.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:06:44,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=427564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:07:11,174 INFO [train.py:968] (1/2) Epoch 10, batch 18000, giga_loss[loss=0.2246, simple_loss=0.295, pruned_loss=0.07705, over 28458.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3122, pruned_loss=0.08407, over 5685614.40 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3393, pruned_loss=0.09482, over 5753505.78 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3101, pruned_loss=0.08324, over 5689316.19 frames. ], batch size: 71, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:07:11,174 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 06:07:19,741 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 06:08:03,698 INFO [train.py:968] (1/2) Epoch 10, batch 18050, giga_loss[loss=0.2409, simple_loss=0.3121, pruned_loss=0.08485, over 27933.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3103, pruned_loss=0.08315, over 5681268.39 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.34, pruned_loss=0.09517, over 5745488.41 frames. ], giga_tot_loss[loss=0.2357, simple_loss=0.3074, pruned_loss=0.08197, over 5691193.47 frames. ], batch size: 412, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:08:06,219 INFO [optim.py:369] (1/2) 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 06:08:32,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-05 06:08:34,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-05 06:08:47,925 INFO [train.py:968] (1/2) Epoch 10, batch 18100, giga_loss[loss=0.2249, simple_loss=0.2957, pruned_loss=0.07704, over 28738.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3073, pruned_loss=0.08196, over 5674591.79 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3401, pruned_loss=0.09509, over 5739996.55 frames. ], giga_tot_loss[loss=0.233, simple_loss=0.3043, pruned_loss=0.08081, over 5685646.93 frames. ], batch size: 92, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:08:55,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2716, 1.5318, 1.3223, 1.1928], device='cuda:1'), covar=tensor([0.1548, 0.1331, 0.0968, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.1642, 0.1493, 0.1463, 0.1581], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 06:09:27,707 INFO [train.py:968] (1/2) Epoch 10, batch 18150, giga_loss[loss=0.2192, simple_loss=0.2973, pruned_loss=0.07055, over 29026.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.307, pruned_loss=0.08189, over 5653844.71 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3411, pruned_loss=0.09562, over 5712281.14 frames. ], giga_tot_loss[loss=0.2311, simple_loss=0.3023, pruned_loss=0.07994, over 5685137.76 frames. ], batch size: 128, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:09:33,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-05 06:09:33,753 INFO [optim.py:369] (1/2) 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 06:09:40,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:10:17,698 INFO [train.py:968] (1/2) Epoch 10, batch 18200, libri_loss[loss=0.2525, simple_loss=0.3314, pruned_loss=0.0868, over 29608.00 frames. ], tot_loss[loss=0.232, simple_loss=0.3034, pruned_loss=0.08028, over 5669783.99 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3414, pruned_loss=0.09571, over 5712821.49 frames. ], giga_tot_loss[loss=0.2278, simple_loss=0.2987, pruned_loss=0.07839, over 5693339.26 frames. ], batch size: 74, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:10:43,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=427823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:11:01,385 INFO [train.py:968] (1/2) Epoch 10, batch 18250, giga_loss[loss=0.2665, simple_loss=0.3408, pruned_loss=0.09606, over 28769.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3059, pruned_loss=0.08228, over 5671987.57 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3417, pruned_loss=0.09581, over 5714098.18 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.3014, pruned_loss=0.08048, over 5688726.39 frames. ], batch size: 199, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:11:07,641 INFO [optim.py:369] (1/2) 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 06:11:44,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3061, 2.5733, 1.2918, 1.4504], device='cuda:1'), covar=tensor([0.0911, 0.0314, 0.0919, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0495, 0.0334, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 06:11:54,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7309, 5.4516, 5.1444, 3.0302], device='cuda:1'), covar=tensor([0.0364, 0.0564, 0.0630, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0988, 0.0928, 0.0812, 0.0637], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 06:11:54,654 INFO [train.py:968] (1/2) Epoch 10, batch 18300, giga_loss[loss=0.2744, simple_loss=0.3472, pruned_loss=0.1008, over 28825.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3163, pruned_loss=0.08715, over 5682584.23 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3418, pruned_loss=0.09573, over 5716039.89 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3122, pruned_loss=0.08564, over 5693653.21 frames. ], batch size: 99, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:12:00,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:03,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=427902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:29,003 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:34,898 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 18350, giga_loss[loss=0.3406, simple_loss=0.3925, pruned_loss=0.1444, over 26603.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3303, pruned_loss=0.09472, over 5684962.99 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3419, pruned_loss=0.09559, over 5720048.64 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3266, pruned_loss=0.09354, over 5689630.89 frames. ], batch size: 555, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:12:43,325 INFO [optim.py:369] (1/2) 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 06:12:48,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8086, 4.6949, 1.9036, 2.1726], device='cuda:1'), covar=tensor([0.0914, 0.0169, 0.0809, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0492, 0.0332, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 06:12:50,399 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-05 06:12:55,866 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=427966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:12:59,169 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 18400, giga_loss[loss=0.3021, simple_loss=0.378, pruned_loss=0.1132, over 28901.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3403, pruned_loss=0.09933, over 5693383.08 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3416, pruned_loss=0.09537, over 5723857.01 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3374, pruned_loss=0.09861, over 5693282.90 frames. ], batch size: 174, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:13:23,539 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=427998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:14:04,665 INFO [train.py:968] (1/2) Epoch 10, batch 18450, giga_loss[loss=0.3025, simple_loss=0.3698, pruned_loss=0.1176, over 28776.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3458, pruned_loss=0.101, over 5695908.08 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3423, pruned_loss=0.09575, over 5727329.98 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.343, pruned_loss=0.1001, over 5692117.50 frames. ], batch size: 119, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:14:07,916 INFO [optim.py:369] (1/2) 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 06:14:34,880 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=428082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:14:37,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=428085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:14:45,813 INFO [train.py:968] (1/2) Epoch 10, batch 18500, libri_loss[loss=0.2383, simple_loss=0.3119, pruned_loss=0.08229, over 29320.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3476, pruned_loss=0.1004, over 5701991.25 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3425, pruned_loss=0.09566, over 5733065.22 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3454, pruned_loss=0.09994, over 5692884.30 frames. ], batch size: 71, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:15:04,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=428114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:15:13,667 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:15:23,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8281, 1.8024, 1.2904, 1.4140], device='cuda:1'), covar=tensor([0.0726, 0.0637, 0.1034, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0433, 0.0500, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 06:15:26,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4955, 1.6948, 1.7374, 1.6575], device='cuda:1'), covar=tensor([0.1317, 0.1435, 0.1705, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0722, 0.0655, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 06:15:30,497 INFO [train.py:968] (1/2) Epoch 10, batch 18550, giga_loss[loss=0.284, simple_loss=0.3403, pruned_loss=0.1139, over 23615.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3489, pruned_loss=0.1008, over 5693302.48 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09572, over 5736387.48 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3472, pruned_loss=0.1006, over 5681663.16 frames. ], batch size: 705, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:15:36,706 INFO [optim.py:369] (1/2) 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 06:16:16,622 INFO [train.py:968] (1/2) Epoch 10, batch 18600, giga_loss[loss=0.287, simple_loss=0.3537, pruned_loss=0.1101, over 28945.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3516, pruned_loss=0.103, over 5696842.42 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3428, pruned_loss=0.0958, over 5736943.54 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3501, pruned_loss=0.1028, over 5687060.07 frames. ], batch size: 112, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:17:03,692 INFO [train.py:968] (1/2) Epoch 10, batch 18650, giga_loss[loss=0.2954, simple_loss=0.362, pruned_loss=0.1144, over 28819.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3555, pruned_loss=0.106, over 5698356.76 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3432, pruned_loss=0.09593, over 5739350.39 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3542, pruned_loss=0.1059, over 5687796.77 frames. ], batch size: 112, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:17:08,536 INFO [optim.py:369] (1/2) 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 06:17:43,628 INFO [train.py:968] (1/2) Epoch 10, batch 18700, giga_loss[loss=0.3282, simple_loss=0.3937, pruned_loss=0.1314, over 28921.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3592, pruned_loss=0.1079, over 5700848.46 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3438, pruned_loss=0.09612, over 5740043.77 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3579, pruned_loss=0.1079, over 5690901.58 frames. ], batch size: 213, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:17:57,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3833, 1.7351, 1.7239, 1.2803], device='cuda:1'), covar=tensor([0.1531, 0.2028, 0.1219, 0.1416], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0686, 0.0848, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 06:18:25,184 INFO [train.py:968] (1/2) Epoch 10, batch 18750, giga_loss[loss=0.2663, simple_loss=0.3469, pruned_loss=0.09281, over 28720.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.361, pruned_loss=0.108, over 5709727.86 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3436, pruned_loss=0.09592, over 5743158.46 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3606, pruned_loss=0.1085, over 5698001.66 frames. ], batch size: 85, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:18:32,166 INFO [optim.py:369] (1/2) 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 06:18:42,596 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:19:07,286 INFO [train.py:968] (1/2) Epoch 10, batch 18800, giga_loss[loss=0.2744, simple_loss=0.359, pruned_loss=0.0949, over 28930.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3622, pruned_loss=0.108, over 5711803.24 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3439, pruned_loss=0.09605, over 5745639.85 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3619, pruned_loss=0.1084, over 5699874.29 frames. ], batch size: 174, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:19:46,923 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-05 06:19:48,330 INFO [train.py:968] (1/2) Epoch 10, batch 18850, giga_loss[loss=0.2673, simple_loss=0.3526, pruned_loss=0.09102, over 28686.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3616, pruned_loss=0.1069, over 5706834.33 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3439, pruned_loss=0.096, over 5748346.13 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3617, pruned_loss=0.1075, over 5694115.96 frames. ], batch size: 262, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:19:54,585 INFO [optim.py:369] (1/2) 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 06:20:00,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2170, 1.8151, 1.3830, 1.4579], device='cuda:1'), covar=tensor([0.0777, 0.0342, 0.0328, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 06:20:29,250 INFO [train.py:968] (1/2) Epoch 10, batch 18900, giga_loss[loss=0.244, simple_loss=0.3335, pruned_loss=0.07724, over 28456.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3598, pruned_loss=0.1047, over 5704684.22 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3441, pruned_loss=0.09599, over 5750983.06 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3602, pruned_loss=0.1054, over 5691140.11 frames. ], batch size: 71, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:20:33,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=428499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:21:08,903 INFO [train.py:968] (1/2) Epoch 10, batch 18950, libri_loss[loss=0.256, simple_loss=0.3325, pruned_loss=0.08973, over 29549.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.357, pruned_loss=0.1018, over 5719385.08 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3438, pruned_loss=0.09568, over 5756417.06 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3581, pruned_loss=0.1029, over 5701989.83 frames. ], batch size: 78, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:21:13,481 INFO [optim.py:369] (1/2) 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 06:21:22,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428561.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:21:47,265 INFO [train.py:968] (1/2) Epoch 10, batch 19000, libri_loss[loss=0.2856, simple_loss=0.3725, pruned_loss=0.09937, over 29540.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3574, pruned_loss=0.1021, over 5716873.66 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3442, pruned_loss=0.09576, over 5762310.32 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3583, pruned_loss=0.1032, over 5695922.19 frames. ], batch size: 84, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:22:28,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=428642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:22:29,208 INFO [train.py:968] (1/2) Epoch 10, batch 19050, libri_loss[loss=0.2847, simple_loss=0.3658, pruned_loss=0.1019, over 27519.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3612, pruned_loss=0.1073, over 5701582.76 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.345, pruned_loss=0.09604, over 5761349.80 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3617, pruned_loss=0.1083, over 5683522.35 frames. ], batch size: 115, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:22:30,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=428645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:22:33,319 INFO [optim.py:369] (1/2) 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 06:22:56,365 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=428674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:23:04,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5727, 5.2575, 5.0832, 2.8702], device='cuda:1'), covar=tensor([0.0394, 0.0588, 0.0655, 0.1635], device='cuda:1'), in_proj_covar=tensor([0.0977, 0.0925, 0.0812, 0.0636], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 06:23:13,414 INFO [train.py:968] (1/2) Epoch 10, batch 19100, giga_loss[loss=0.2886, simple_loss=0.3601, pruned_loss=0.1086, over 29033.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3646, pruned_loss=0.1122, over 5701486.41 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3452, pruned_loss=0.09631, over 5764164.92 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3652, pruned_loss=0.113, over 5683657.00 frames. ], batch size: 155, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:23:29,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5420, 1.6812, 1.7889, 1.3824], device='cuda:1'), covar=tensor([0.1551, 0.2164, 0.1248, 0.1453], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0686, 0.0849, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 06:23:48,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=428738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:23:52,313 INFO [train.py:968] (1/2) Epoch 10, batch 19150, giga_loss[loss=0.2973, simple_loss=0.3603, pruned_loss=0.1171, over 27905.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3638, pruned_loss=0.1122, over 5702534.75 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3454, pruned_loss=0.0962, over 5762235.37 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3647, pruned_loss=0.1134, over 5687908.76 frames. ], batch size: 412, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:23:59,419 INFO [optim.py:369] (1/2) 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 06:24:21,757 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7624, 1.2052, 5.0999, 3.4996], device='cuda:1'), covar=tensor([0.1534, 0.2668, 0.0324, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0635, 0.0570, 0.0827, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 06:24:34,304 INFO [train.py:968] (1/2) Epoch 10, batch 19200, giga_loss[loss=0.2669, simple_loss=0.3468, pruned_loss=0.0935, over 28720.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3619, pruned_loss=0.1115, over 5699112.80 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3457, pruned_loss=0.09624, over 5755358.25 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3628, pruned_loss=0.1129, over 5691434.72 frames. ], batch size: 78, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:25:18,194 INFO [train.py:968] (1/2) Epoch 10, batch 19250, giga_loss[loss=0.315, simple_loss=0.38, pruned_loss=0.125, over 28771.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3612, pruned_loss=0.1114, over 5688472.90 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3454, pruned_loss=0.09606, over 5748943.26 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3625, pruned_loss=0.113, over 5687032.74 frames. ], batch size: 242, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:25:26,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9627, 2.0716, 1.7083, 2.3154], device='cuda:1'), covar=tensor([0.2025, 0.2082, 0.2275, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.0939, 0.1122, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 06:25:26,392 INFO [optim.py:369] (1/2) 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 06:25:29,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4128, 1.6204, 1.6805, 1.2955], device='cuda:1'), covar=tensor([0.1486, 0.2003, 0.1192, 0.1404], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0686, 0.0848, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 06:25:44,745 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:25:49,472 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=428881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:25:51,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=428884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:25:55,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 06:26:00,360 INFO [train.py:968] (1/2) Epoch 10, batch 19300, giga_loss[loss=0.2615, simple_loss=0.3424, pruned_loss=0.0903, over 28984.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3604, pruned_loss=0.1101, over 5690056.67 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.09618, over 5751292.84 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3615, pruned_loss=0.1115, over 5685611.75 frames. ], batch size: 106, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:26:17,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=428913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:26:40,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=428936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:26:48,864 INFO [train.py:968] (1/2) Epoch 10, batch 19350, giga_loss[loss=0.2751, simple_loss=0.328, pruned_loss=0.1111, over 23659.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3573, pruned_loss=0.1075, over 5682757.05 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3464, pruned_loss=0.09638, over 5747901.35 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3577, pruned_loss=0.1087, over 5681058.12 frames. ], batch size: 705, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:26:54,734 INFO [optim.py:369] (1/2) 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 06:27:20,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=428979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:27:31,487 INFO [train.py:968] (1/2) Epoch 10, batch 19400, giga_loss[loss=0.2489, simple_loss=0.3226, pruned_loss=0.08763, over 28683.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3528, pruned_loss=0.1047, over 5687117.67 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.347, pruned_loss=0.09656, over 5752158.39 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3529, pruned_loss=0.1059, over 5680316.88 frames. ], batch size: 307, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:28:07,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 06:28:08,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4359, 1.5719, 1.5524, 1.4419], device='cuda:1'), covar=tensor([0.1348, 0.1677, 0.1982, 0.1667], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0728, 0.0659, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 06:28:18,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1006, 1.1725, 3.5723, 3.0103], device='cuda:1'), covar=tensor([0.1608, 0.2647, 0.0435, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0637, 0.0571, 0.0827, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 06:28:21,024 INFO [train.py:968] (1/2) Epoch 10, batch 19450, giga_loss[loss=0.2349, simple_loss=0.3093, pruned_loss=0.08024, over 28787.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3459, pruned_loss=0.101, over 5686970.91 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.347, pruned_loss=0.09649, over 5753720.72 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3461, pruned_loss=0.102, over 5679719.08 frames. ], batch size: 242, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:28:27,985 INFO [optim.py:369] (1/2) 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 06:28:55,755 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:28:58,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429082.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:29:09,025 INFO [train.py:968] (1/2) Epoch 10, batch 19500, libri_loss[loss=0.2512, simple_loss=0.3264, pruned_loss=0.08801, over 29332.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3408, pruned_loss=0.09803, over 5693841.13 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3469, pruned_loss=0.09627, over 5758593.92 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3408, pruned_loss=0.09912, over 5681261.82 frames. ], batch size: 67, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:29:26,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429111.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 06:29:41,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6030, 1.6204, 1.5397, 1.5029], device='cuda:1'), covar=tensor([0.1241, 0.1755, 0.1889, 0.1601], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0729, 0.0661, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 06:29:55,090 INFO [train.py:968] (1/2) Epoch 10, batch 19550, giga_loss[loss=0.2375, simple_loss=0.3158, pruned_loss=0.07956, over 28866.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3414, pruned_loss=0.09802, over 5691605.91 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3471, pruned_loss=0.09636, over 5758038.28 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3411, pruned_loss=0.0988, over 5681692.05 frames. ], batch size: 60, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:30:01,083 INFO [optim.py:369] (1/2) 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 06:30:16,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7318, 1.8390, 1.8085, 1.5836], device='cuda:1'), covar=tensor([0.1295, 0.1682, 0.1645, 0.1577], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0731, 0.0662, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 06:30:41,889 INFO [train.py:968] (1/2) Epoch 10, batch 19600, libri_loss[loss=0.3806, simple_loss=0.4349, pruned_loss=0.1631, over 20266.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.341, pruned_loss=0.09699, over 5692872.24 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3476, pruned_loss=0.09662, over 5751478.04 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3402, pruned_loss=0.09735, over 5690821.88 frames. ], batch size: 187, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:30:52,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-05 06:31:23,057 INFO [train.py:968] (1/2) Epoch 10, batch 19650, giga_loss[loss=0.2582, simple_loss=0.3305, pruned_loss=0.09292, over 28916.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.341, pruned_loss=0.09743, over 5694603.44 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3482, pruned_loss=0.09686, over 5744805.43 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3397, pruned_loss=0.09753, over 5697295.69 frames. ], batch size: 186, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:31:26,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:31:29,010 INFO [optim.py:369] (1/2) 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 06:31:46,947 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:31:57,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3630, 1.5332, 1.3806, 1.2232], device='cuda:1'), covar=tensor([0.1975, 0.1550, 0.1159, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.1643, 0.1516, 0.1500, 0.1617], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 06:32:00,895 INFO [train.py:968] (1/2) Epoch 10, batch 19700, giga_loss[loss=0.2138, simple_loss=0.2896, pruned_loss=0.06905, over 28656.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3382, pruned_loss=0.09618, over 5699770.35 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3489, pruned_loss=0.09723, over 5739024.90 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3364, pruned_loss=0.09591, over 5706244.19 frames. ], batch size: 60, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:32:06,991 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:32:34,663 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:32:38,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1433, 1.1496, 4.2489, 3.2925], device='cuda:1'), covar=tensor([0.1601, 0.2537, 0.0369, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0636, 0.0570, 0.0825, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 06:32:41,048 INFO [train.py:968] (1/2) Epoch 10, batch 19750, giga_loss[loss=0.2465, simple_loss=0.3125, pruned_loss=0.09023, over 28709.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.337, pruned_loss=0.09573, over 5710642.37 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3498, pruned_loss=0.09753, over 5741714.01 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3343, pruned_loss=0.09518, over 5712463.86 frames. ], batch size: 92, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:32:49,698 INFO [optim.py:369] (1/2) 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 06:32:49,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:33:11,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 06:33:21,176 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:33:22,151 INFO [train.py:968] (1/2) Epoch 10, batch 19800, giga_loss[loss=0.2308, simple_loss=0.3057, pruned_loss=0.07793, over 28950.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3341, pruned_loss=0.09484, over 5712107.98 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3498, pruned_loss=0.09751, over 5743409.66 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3319, pruned_loss=0.09441, over 5711888.63 frames. ], batch size: 213, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:33:23,143 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8647, 3.6723, 3.4823, 1.6409], device='cuda:1'), covar=tensor([0.0603, 0.0706, 0.0646, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.0984, 0.0926, 0.0813, 0.0639], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 06:33:23,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:33:49,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:34:02,741 INFO [train.py:968] (1/2) Epoch 10, batch 19850, giga_loss[loss=0.2274, simple_loss=0.3065, pruned_loss=0.07416, over 28888.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3319, pruned_loss=0.09332, over 5724338.51 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3504, pruned_loss=0.09748, over 5748504.65 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.329, pruned_loss=0.09288, over 5718667.46 frames. ], batch size: 174, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:34:11,928 INFO [optim.py:369] (1/2) 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 06:34:43,522 INFO [train.py:968] (1/2) Epoch 10, batch 19900, giga_loss[loss=0.2667, simple_loss=0.3329, pruned_loss=0.1002, over 28596.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3309, pruned_loss=0.09287, over 5724957.71 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3507, pruned_loss=0.09739, over 5752476.83 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3279, pruned_loss=0.09249, over 5716213.47 frames. ], batch size: 307, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:34:45,937 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:34:47,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:35:09,863 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:35:16,752 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=429538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:35:20,693 INFO [train.py:968] (1/2) Epoch 10, batch 19950, giga_loss[loss=0.2214, simple_loss=0.3028, pruned_loss=0.07003, over 28978.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3305, pruned_loss=0.09256, over 5724334.50 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3516, pruned_loss=0.09754, over 5755252.20 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3263, pruned_loss=0.09193, over 5713594.26 frames. ], batch size: 136, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:35:30,609 INFO [optim.py:369] (1/2) 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 06:35:52,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0278, 1.2147, 3.5554, 2.9033], device='cuda:1'), covar=tensor([0.1681, 0.2595, 0.0409, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0639, 0.0571, 0.0827, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 06:36:01,302 INFO [train.py:968] (1/2) Epoch 10, batch 20000, libri_loss[loss=0.2792, simple_loss=0.369, pruned_loss=0.09469, over 25988.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3282, pruned_loss=0.09102, over 5728059.49 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3517, pruned_loss=0.09753, over 5754702.49 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3241, pruned_loss=0.09035, over 5719143.55 frames. ], batch size: 136, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:36:39,433 INFO [train.py:968] (1/2) Epoch 10, batch 20050, giga_loss[loss=0.2273, simple_loss=0.3057, pruned_loss=0.07448, over 28973.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3273, pruned_loss=0.09043, over 5733588.62 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3523, pruned_loss=0.09766, over 5759405.32 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3227, pruned_loss=0.08957, over 5721564.40 frames. ], batch size: 227, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:36:42,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:36:47,334 INFO [optim.py:369] (1/2) 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 06:37:03,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:37:18,047 INFO [train.py:968] (1/2) Epoch 10, batch 20100, giga_loss[loss=0.2363, simple_loss=0.3091, pruned_loss=0.08174, over 28639.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3253, pruned_loss=0.08959, over 5739025.87 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3526, pruned_loss=0.09778, over 5760268.30 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3213, pruned_loss=0.08879, over 5728802.87 frames. ], batch size: 60, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:37:34,783 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:37:57,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4843, 1.7883, 1.6956, 1.3089], device='cuda:1'), covar=tensor([0.1433, 0.2123, 0.1250, 0.1493], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0690, 0.0853, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 06:38:04,542 INFO [train.py:968] (1/2) Epoch 10, batch 20150, giga_loss[loss=0.3035, simple_loss=0.3704, pruned_loss=0.1183, over 28543.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3291, pruned_loss=0.09206, over 5731385.94 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3529, pruned_loss=0.09793, over 5762447.23 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3253, pruned_loss=0.09117, over 5721004.43 frames. ], batch size: 307, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:38:13,211 INFO [optim.py:369] (1/2) 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 06:38:47,508 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:38:49,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:38:49,795 INFO [train.py:968] (1/2) Epoch 10, batch 20200, giga_loss[loss=0.3533, simple_loss=0.4033, pruned_loss=0.1516, over 27638.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3362, pruned_loss=0.09613, over 5730140.25 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3533, pruned_loss=0.09786, over 5768855.20 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3319, pruned_loss=0.0953, over 5714726.02 frames. ], batch size: 472, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:39:08,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4459, 1.6707, 1.4482, 1.4685], device='cuda:1'), covar=tensor([0.1744, 0.1685, 0.1652, 0.1564], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.0942, 0.1121, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 06:39:13,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:17,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:17,955 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:38,348 INFO [train.py:968] (1/2) Epoch 10, batch 20250, giga_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.117, over 28676.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3455, pruned_loss=0.1029, over 5704836.38 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3533, pruned_loss=0.09775, over 5766265.10 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3418, pruned_loss=0.1024, over 5693186.40 frames. ], batch size: 242, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:39:45,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:47,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=429852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:39:47,927 INFO [optim.py:369] (1/2) 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 06:39:50,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=429855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:40:14,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=429884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:40:19,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-05 06:40:21,972 INFO [train.py:968] (1/2) Epoch 10, batch 20300, giga_loss[loss=0.3001, simple_loss=0.3765, pruned_loss=0.1118, over 28902.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3504, pruned_loss=0.1052, over 5698583.06 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3534, pruned_loss=0.09773, over 5764215.94 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3471, pruned_loss=0.1051, over 5688532.69 frames. ], batch size: 145, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:40:37,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=429913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:41:07,300 INFO [train.py:968] (1/2) Epoch 10, batch 20350, giga_loss[loss=0.2892, simple_loss=0.365, pruned_loss=0.1067, over 28909.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3546, pruned_loss=0.1068, over 5693824.33 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3535, pruned_loss=0.09778, over 5769652.08 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3518, pruned_loss=0.107, over 5678925.29 frames. ], batch size: 106, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:41:15,204 INFO [optim.py:369] (1/2) 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 06:41:36,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-05 06:41:52,093 INFO [train.py:968] (1/2) Epoch 10, batch 20400, giga_loss[loss=0.2875, simple_loss=0.3653, pruned_loss=0.1048, over 28920.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3581, pruned_loss=0.1085, over 5695038.44 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.353, pruned_loss=0.09754, over 5772683.58 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3564, pruned_loss=0.1091, over 5678326.42 frames. ], batch size: 186, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:42:35,751 INFO [train.py:968] (1/2) Epoch 10, batch 20450, giga_loss[loss=0.307, simple_loss=0.369, pruned_loss=0.1225, over 27497.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3631, pruned_loss=0.1121, over 5685720.39 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3525, pruned_loss=0.09731, over 5773819.54 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3623, pruned_loss=0.113, over 5669961.33 frames. ], batch size: 472, lr: 3.25e-03, grad_scale: 8.0
+2023-03-05 06:42:37,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7020, 1.9601, 1.9783, 1.5357], device='cuda:1'), covar=tensor([0.1806, 0.2145, 0.1430, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0685, 0.0847, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 06:42:45,188 INFO [optim.py:369] (1/2) 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 06:42:47,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=430056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:42:48,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=430059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:42:51,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 06:43:04,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:43:17,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=430088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:43:18,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4438, 1.5618, 1.7556, 1.4774], device='cuda:1'), covar=tensor([0.1424, 0.1736, 0.1739, 0.1621], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0725, 0.0659, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 06:43:21,605 INFO [train.py:968] (1/2) Epoch 10, batch 20500, giga_loss[loss=0.273, simple_loss=0.3458, pruned_loss=0.1001, over 28015.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3583, pruned_loss=0.1088, over 5687651.68 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3528, pruned_loss=0.09771, over 5775255.01 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3575, pruned_loss=0.1093, over 5673119.50 frames. ], batch size: 412, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:44:03,963 INFO [train.py:968] (1/2) Epoch 10, batch 20550, giga_loss[loss=0.2591, simple_loss=0.3397, pruned_loss=0.0893, over 28915.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3554, pruned_loss=0.1062, over 5690234.02 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3533, pruned_loss=0.09821, over 5767302.57 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3544, pruned_loss=0.1063, over 5683473.16 frames. ], batch size: 199, lr: 3.25e-03, grad_scale: 4.0
+2023-03-05 06:44:14,681 INFO [optim.py:369] (1/2) 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 06:44:48,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:44:48,937 INFO [train.py:968] (1/2) Epoch 10, batch 20600, giga_loss[loss=0.3269, simple_loss=0.3891, pruned_loss=0.1323, over 27595.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3551, pruned_loss=0.1056, over 5682256.15 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3535, pruned_loss=0.09845, over 5756579.19 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3542, pruned_loss=0.1056, over 5684986.45 frames. ], batch size: 472, lr: 3.25e-03, grad_scale: 2.0
+2023-03-05 06:45:02,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2537, 1.2384, 1.1114, 0.8885], device='cuda:1'), covar=tensor([0.0753, 0.0482, 0.0957, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0436, 0.0503, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 06:45:30,642 INFO [train.py:968] (1/2) Epoch 10, batch 20650, giga_loss[loss=0.3206, simple_loss=0.3876, pruned_loss=0.1268, over 28760.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3568, pruned_loss=0.1062, over 5686018.93 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3542, pruned_loss=0.09896, over 5758312.26 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3555, pruned_loss=0.1059, over 5684899.00 frames. ], batch size: 262, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 06:45:40,488 INFO [optim.py:369] (1/2) 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 06:46:10,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7941, 1.7235, 1.3091, 1.3469], device='cuda:1'), covar=tensor([0.0655, 0.0548, 0.0949, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0436, 0.0504, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 06:46:13,529 INFO [train.py:968] (1/2) Epoch 10, batch 20700, giga_loss[loss=0.2707, simple_loss=0.343, pruned_loss=0.09922, over 28827.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.359, pruned_loss=0.1079, over 5693161.54 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3547, pruned_loss=0.09939, over 5760999.84 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3575, pruned_loss=0.1074, over 5688693.46 frames. ], batch size: 99, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 06:46:40,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5528, 1.7052, 1.4768, 1.5754], device='cuda:1'), covar=tensor([0.1166, 0.1590, 0.1637, 0.1467], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0723, 0.0658, 0.0644], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 06:46:55,138 INFO [train.py:968] (1/2) Epoch 10, batch 20750, giga_loss[loss=0.3041, simple_loss=0.3752, pruned_loss=0.1165, over 29082.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3621, pruned_loss=0.1101, over 5704861.21 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3561, pruned_loss=0.1004, over 5762198.41 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3599, pruned_loss=0.1092, over 5697600.85 frames. ], batch size: 155, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 06:47:05,873 INFO [optim.py:369] (1/2) 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 06:47:43,330 INFO [train.py:968] (1/2) Epoch 10, batch 20800, giga_loss[loss=0.2904, simple_loss=0.3603, pruned_loss=0.1102, over 28581.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3624, pruned_loss=0.1108, over 5687431.21 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3562, pruned_loss=0.1004, over 5763756.11 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3606, pruned_loss=0.1101, over 5679828.80 frames. ], batch size: 262, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:48:26,634 INFO [train.py:968] (1/2) Epoch 10, batch 20850, giga_loss[loss=0.2872, simple_loss=0.3586, pruned_loss=0.1079, over 28848.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3638, pruned_loss=0.112, over 5690605.42 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.357, pruned_loss=0.1009, over 5763520.88 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3618, pruned_loss=0.1113, over 5682685.17 frames. ], batch size: 186, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:48:31,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=430448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:48:37,800 INFO [optim.py:369] (1/2) 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 06:49:07,865 INFO [train.py:968] (1/2) Epoch 10, batch 20900, giga_loss[loss=0.2967, simple_loss=0.3674, pruned_loss=0.113, over 28537.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3632, pruned_loss=0.1112, over 5701062.62 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3572, pruned_loss=0.1011, over 5765879.11 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3615, pruned_loss=0.1107, over 5691341.52 frames. ], batch size: 336, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:49:23,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9783, 1.0476, 3.7482, 3.1518], device='cuda:1'), covar=tensor([0.1752, 0.2708, 0.0405, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0639, 0.0571, 0.0827, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 06:49:44,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5633, 4.1615, 1.6070, 1.6715], device='cuda:1'), covar=tensor([0.0893, 0.0193, 0.0820, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0493, 0.0329, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 06:49:47,862 INFO [train.py:968] (1/2) Epoch 10, batch 20950, giga_loss[loss=0.2888, simple_loss=0.3674, pruned_loss=0.1051, over 28881.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3622, pruned_loss=0.1101, over 5703208.93 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.357, pruned_loss=0.1011, over 5768913.32 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3612, pruned_loss=0.1099, over 5690334.86 frames. ], batch size: 227, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:49:51,847 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.08 vs. limit=2.0
+2023-03-05 06:49:58,014 INFO [optim.py:369] (1/2) 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 06:50:07,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=430568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:50:08,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5515, 4.4169, 1.6976, 1.6666], device='cuda:1'), covar=tensor([0.0958, 0.0167, 0.0849, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0492, 0.0328, 0.0382], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 06:50:25,003 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=430591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:50:26,690 INFO [train.py:968] (1/2) Epoch 10, batch 21000, giga_loss[loss=0.2763, simple_loss=0.3536, pruned_loss=0.09945, over 28877.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3633, pruned_loss=0.1092, over 5703045.64 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3579, pruned_loss=0.1018, over 5765677.71 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3617, pruned_loss=0.1086, over 5694266.65 frames. ], batch size: 106, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:50:26,691 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 06:50:35,598 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19481MB
+2023-03-05 06:50:36,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=430594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:50:52,884 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:50:54,189 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 06:51:00,586 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=430623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:51:16,748 INFO [train.py:968] (1/2) Epoch 10, batch 21050, giga_loss[loss=0.2683, simple_loss=0.3419, pruned_loss=0.09734, over 28482.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3627, pruned_loss=0.1087, over 5701275.55 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3585, pruned_loss=0.1023, over 5767393.30 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3611, pruned_loss=0.1079, over 5692150.02 frames. ], batch size: 71, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:51:24,993 INFO [optim.py:369] (1/2) 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 06:51:55,283 INFO [train.py:968] (1/2) Epoch 10, batch 21100, giga_loss[loss=0.2597, simple_loss=0.3322, pruned_loss=0.09361, over 28887.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3597, pruned_loss=0.107, over 5713481.35 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.359, pruned_loss=0.1027, over 5768637.73 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.358, pruned_loss=0.1062, over 5704215.95 frames. ], batch size: 106, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:52:09,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=430711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:52:12,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=430714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:52:36,032 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 21150, giga_loss[loss=0.3443, simple_loss=0.3862, pruned_loss=0.1512, over 26538.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3589, pruned_loss=0.107, over 5709319.54 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3595, pruned_loss=0.1031, over 5765951.17 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3571, pruned_loss=0.106, over 5703982.92 frames. ], batch size: 555, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:52:46,181 INFO [optim.py:369] (1/2) 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 06:52:47,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:52:57,402 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-05 06:53:17,218 INFO [train.py:968] (1/2) Epoch 10, batch 21200, giga_loss[loss=0.2775, simple_loss=0.353, pruned_loss=0.101, over 28879.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3568, pruned_loss=0.1062, over 5704056.27 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3597, pruned_loss=0.1033, over 5758565.84 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3552, pruned_loss=0.1052, over 5705643.35 frames. ], batch size: 227, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:53:35,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4881, 1.6401, 1.6378, 1.5210], device='cuda:1'), covar=tensor([0.1378, 0.1757, 0.1913, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0724, 0.0659, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 06:53:58,153 INFO [train.py:968] (1/2) Epoch 10, batch 21250, giga_loss[loss=0.3351, simple_loss=0.3907, pruned_loss=0.1398, over 28944.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3586, pruned_loss=0.108, over 5707104.28 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3597, pruned_loss=0.1035, over 5761484.70 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3573, pruned_loss=0.1072, over 5704923.03 frames. ], batch size: 106, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:54:10,990 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 21300, giga_loss[loss=0.2572, simple_loss=0.3351, pruned_loss=0.08967, over 28864.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3594, pruned_loss=0.1082, over 5713984.09 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.36, pruned_loss=0.104, over 5764801.00 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3581, pruned_loss=0.1072, over 5707714.69 frames. ], batch size: 112, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:55:20,848 INFO [train.py:968] (1/2) Epoch 10, batch 21350, giga_loss[loss=0.2493, simple_loss=0.3369, pruned_loss=0.0809, over 28442.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3569, pruned_loss=0.106, over 5712186.15 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3597, pruned_loss=0.104, over 5767426.56 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.356, pruned_loss=0.1052, over 5703844.93 frames. ], batch size: 71, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:55:30,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-05 06:55:32,666 INFO [optim.py:369] (1/2) 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 06:55:49,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 06:56:01,479 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 21400, giga_loss[loss=0.272, simple_loss=0.3482, pruned_loss=0.09786, over 28929.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 5722830.67 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3598, pruned_loss=0.1041, over 5770253.76 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3542, pruned_loss=0.1034, over 5712919.19 frames. ], batch size: 145, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:56:07,323 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=430998.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 06:56:44,936 INFO [train.py:968] (1/2) Epoch 10, batch 21450, giga_loss[loss=0.265, simple_loss=0.339, pruned_loss=0.09556, over 29005.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3548, pruned_loss=0.1044, over 5729048.13 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3597, pruned_loss=0.1041, over 5771830.20 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3542, pruned_loss=0.1039, over 5719469.69 frames. ], batch size: 106, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:56:54,769 INFO [optim.py:369] (1/2) 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 06:57:03,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-05 06:57:21,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-05 06:57:23,742 INFO [train.py:968] (1/2) Epoch 10, batch 21500, giga_loss[loss=0.2684, simple_loss=0.3436, pruned_loss=0.09659, over 28643.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3518, pruned_loss=0.1029, over 5726786.44 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3595, pruned_loss=0.1041, over 5774447.31 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3514, pruned_loss=0.1025, over 5715956.89 frames. ], batch size: 85, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:57:55,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:57:57,345 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:58:00,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:58:04,659 INFO [train.py:968] (1/2) Epoch 10, batch 21550, giga_loss[loss=0.2716, simple_loss=0.3539, pruned_loss=0.09469, over 28953.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3497, pruned_loss=0.1019, over 5727292.93 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.36, pruned_loss=0.1045, over 5776552.08 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3488, pruned_loss=0.1011, over 5716122.42 frames. ], batch size: 227, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 06:58:15,943 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 10, batch 21600, giga_loss[loss=0.262, simple_loss=0.3271, pruned_loss=0.09846, over 28499.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3501, pruned_loss=0.1028, over 5730715.87 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3606, pruned_loss=0.1051, over 5777869.29 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3487, pruned_loss=0.1016, over 5720255.77 frames. ], batch size: 85, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:59:26,566 INFO [train.py:968] (1/2) Epoch 10, batch 21650, giga_loss[loss=0.256, simple_loss=0.3309, pruned_loss=0.09051, over 28680.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3502, pruned_loss=0.1036, over 5724665.77 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3615, pruned_loss=0.1059, over 5778610.75 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3481, pruned_loss=0.102, over 5714678.27 frames. ], batch size: 262, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 06:59:37,097 INFO [optim.py:369] (1/2) 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 06:59:53,556 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:59:56,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 06:59:58,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8704, 3.6743, 3.4921, 1.6935], device='cuda:1'), covar=tensor([0.0618, 0.0728, 0.0751, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.0993, 0.0936, 0.0819, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 07:00:07,623 INFO [train.py:968] (1/2) Epoch 10, batch 21700, giga_loss[loss=0.2729, simple_loss=0.3451, pruned_loss=0.1003, over 28720.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3487, pruned_loss=0.1037, over 5711893.88 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3617, pruned_loss=0.1062, over 5770731.95 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3467, pruned_loss=0.1021, over 5710685.11 frames. ], batch size: 284, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:00:19,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=431307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:00:27,726 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 07:00:50,342 INFO [train.py:968] (1/2) Epoch 10, batch 21750, giga_loss[loss=0.3044, simple_loss=0.3617, pruned_loss=0.1236, over 26837.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3457, pruned_loss=0.1025, over 5709481.42 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3617, pruned_loss=0.1062, over 5771419.29 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.344, pruned_loss=0.1012, over 5707687.00 frames. ], batch size: 555, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:01:00,414 INFO [optim.py:369] (1/2) 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 07:01:14,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431373.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 07:01:16,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7816, 1.8556, 1.7451, 1.6913], device='cuda:1'), covar=tensor([0.1286, 0.1910, 0.1865, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0727, 0.0662, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 07:01:30,075 INFO [train.py:968] (1/2) Epoch 10, batch 21800, giga_loss[loss=0.2683, simple_loss=0.3398, pruned_loss=0.09839, over 29020.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3427, pruned_loss=0.1011, over 5713229.09 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3619, pruned_loss=0.1065, over 5773218.09 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3409, pruned_loss=0.09964, over 5709249.09 frames. ], batch size: 128, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:01:34,520 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=431400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:02:09,682 INFO [train.py:968] (1/2) Epoch 10, batch 21850, giga_loss[loss=0.2128, simple_loss=0.2988, pruned_loss=0.06337, over 29075.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3413, pruned_loss=0.09995, over 5712703.05 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3613, pruned_loss=0.1063, over 5775900.88 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3397, pruned_loss=0.0988, over 5705044.44 frames. ], batch size: 155, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:02:20,095 INFO [optim.py:369] (1/2) 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 07:02:34,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=431476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:02:50,828 INFO [train.py:968] (1/2) Epoch 10, batch 21900, giga_loss[loss=0.2885, simple_loss=0.3686, pruned_loss=0.1042, over 28927.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3432, pruned_loss=0.1007, over 5711832.27 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3615, pruned_loss=0.1066, over 5774776.16 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3414, pruned_loss=0.09945, over 5705196.15 frames. ], batch size: 213, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 07:03:10,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431516.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 07:03:12,479 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431519.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 07:03:34,900 INFO [train.py:968] (1/2) Epoch 10, batch 21950, giga_loss[loss=0.3326, simple_loss=0.392, pruned_loss=0.1366, over 28602.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3468, pruned_loss=0.1023, over 5703660.42 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3618, pruned_loss=0.1069, over 5765501.60 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3446, pruned_loss=0.1009, over 5704644.81 frames. ], batch size: 307, lr: 3.24e-03, grad_scale: 2.0
+2023-03-05 07:03:39,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=431548.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 07:03:45,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9290, 3.7287, 3.5397, 1.7358], device='cuda:1'), covar=tensor([0.0600, 0.0703, 0.0696, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.0997, 0.0937, 0.0821, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 07:03:48,692 INFO [optim.py:369] (1/2) 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 07:04:16,594 INFO [train.py:968] (1/2) Epoch 10, batch 22000, giga_loss[loss=0.2512, simple_loss=0.3344, pruned_loss=0.08399, over 28698.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3484, pruned_loss=0.1022, over 5708064.58 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3617, pruned_loss=0.107, over 5768485.60 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3465, pruned_loss=0.1009, over 5704820.67 frames. ], batch size: 284, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:05:00,220 INFO [train.py:968] (1/2) Epoch 10, batch 22050, giga_loss[loss=0.2717, simple_loss=0.3497, pruned_loss=0.09689, over 29046.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3501, pruned_loss=0.1025, over 5707191.24 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3621, pruned_loss=0.1076, over 5771869.66 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3478, pruned_loss=0.1008, over 5699883.78 frames. ], batch size: 164, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:05:13,337 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 22100, giga_loss[loss=0.2431, simple_loss=0.3159, pruned_loss=0.08513, over 28581.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3504, pruned_loss=0.1027, over 5695010.97 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3624, pruned_loss=0.1079, over 5765525.85 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.348, pruned_loss=0.1009, over 5694062.26 frames. ], batch size: 60, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:06:23,098 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.21 vs. limit=5.0
+2023-03-05 07:06:26,260 INFO [train.py:968] (1/2) Epoch 10, batch 22150, giga_loss[loss=0.3265, simple_loss=0.3831, pruned_loss=0.1349, over 28972.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3518, pruned_loss=0.1041, over 5697207.06 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3628, pruned_loss=0.1085, over 5761334.80 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3492, pruned_loss=0.102, over 5698548.58 frames. ], batch size: 136, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:06:40,431 INFO [optim.py:369] (1/2) 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 07:06:53,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:07:00,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-05 07:07:09,752 INFO [train.py:968] (1/2) Epoch 10, batch 22200, giga_loss[loss=0.2298, simple_loss=0.3097, pruned_loss=0.07492, over 28905.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3523, pruned_loss=0.1045, over 5699975.81 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3631, pruned_loss=0.1086, over 5763946.23 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3498, pruned_loss=0.1026, over 5697512.86 frames. ], batch size: 66, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:07:52,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2478, 1.2546, 4.4715, 3.1927], device='cuda:1'), covar=tensor([0.1712, 0.2604, 0.0338, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0573, 0.0826, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 07:07:54,090 INFO [train.py:968] (1/2) Epoch 10, batch 22250, giga_loss[loss=0.2603, simple_loss=0.3421, pruned_loss=0.08925, over 28756.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3542, pruned_loss=0.1057, over 5696987.86 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3634, pruned_loss=0.109, over 5756022.85 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3517, pruned_loss=0.1038, over 5699812.47 frames. ], batch size: 262, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:08:01,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=431851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:08:06,676 INFO [optim.py:369] (1/2) 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 07:08:29,592 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=431885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:08:34,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5096, 4.3148, 4.0816, 2.0310], device='cuda:1'), covar=tensor([0.0485, 0.0664, 0.0651, 0.1898], device='cuda:1'), in_proj_covar=tensor([0.0994, 0.0929, 0.0820, 0.0635], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 07:08:35,775 INFO [train.py:968] (1/2) Epoch 10, batch 22300, libri_loss[loss=0.3316, simple_loss=0.3957, pruned_loss=0.1338, over 27511.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3568, pruned_loss=0.1071, over 5699560.70 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.364, pruned_loss=0.1095, over 5758044.47 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.354, pruned_loss=0.105, over 5698769.15 frames. ], batch size: 115, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:08:56,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:00,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:18,125 INFO [train.py:968] (1/2) Epoch 10, batch 22350, giga_loss[loss=0.2597, simple_loss=0.3345, pruned_loss=0.09246, over 28656.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3578, pruned_loss=0.1071, over 5706338.57 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3643, pruned_loss=0.1098, over 5759062.68 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3553, pruned_loss=0.1052, over 5704452.87 frames. ], batch size: 85, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:09:22,967 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=431950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:26,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0273, 1.8336, 4.4245, 3.8513], device='cuda:1'), covar=tensor([0.1159, 0.2158, 0.0357, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0643, 0.0577, 0.0834, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 07:09:29,686 INFO [optim.py:369] (1/2) 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 07:09:54,957 INFO [train.py:968] (1/2) Epoch 10, batch 22400, giga_loss[loss=0.2948, simple_loss=0.3681, pruned_loss=0.1108, over 28701.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5688801.97 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.365, pruned_loss=0.1104, over 5734712.57 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1054, over 5707423.18 frames. ], batch size: 242, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:09:55,213 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=431994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:09:57,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=431997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:10:10,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-05 07:10:20,066 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 22450, giga_loss[loss=0.3662, simple_loss=0.4134, pruned_loss=0.1595, over 26753.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3594, pruned_loss=0.1081, over 5694932.18 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3655, pruned_loss=0.1109, over 5735550.15 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3565, pruned_loss=0.1058, over 5708022.45 frames. ], batch size: 555, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:10:49,035 INFO [optim.py:369] (1/2) 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 07:10:56,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 07:11:15,223 INFO [train.py:968] (1/2) Epoch 10, batch 22500, libri_loss[loss=0.3344, simple_loss=0.4092, pruned_loss=0.1298, over 29375.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3607, pruned_loss=0.1092, over 5701493.42 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3669, pruned_loss=0.1121, over 5738956.34 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3569, pruned_loss=0.1062, over 5707480.11 frames. ], batch size: 92, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:11:55,250 INFO [train.py:968] (1/2) Epoch 10, batch 22550, giga_loss[loss=0.2614, simple_loss=0.335, pruned_loss=0.09389, over 29044.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3587, pruned_loss=0.1084, over 5709897.48 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3669, pruned_loss=0.1124, over 5743899.59 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3554, pruned_loss=0.1056, over 5709235.79 frames. ], batch size: 128, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:12:06,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0560, 0.9944, 3.4025, 3.0320], device='cuda:1'), covar=tensor([0.1983, 0.3136, 0.0776, 0.2191], device='cuda:1'), in_proj_covar=tensor([0.0648, 0.0581, 0.0843, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 07:12:08,457 INFO [optim.py:369] (1/2) 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 07:12:37,828 INFO [train.py:968] (1/2) Epoch 10, batch 22600, libri_loss[loss=0.2669, simple_loss=0.3353, pruned_loss=0.09927, over 29559.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3559, pruned_loss=0.1071, over 5703111.51 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3673, pruned_loss=0.1129, over 5739186.91 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3526, pruned_loss=0.1042, over 5705882.35 frames. ], batch size: 75, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:12:51,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-05 07:13:16,104 INFO [train.py:968] (1/2) Epoch 10, batch 22650, giga_loss[loss=0.2618, simple_loss=0.338, pruned_loss=0.09278, over 28831.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3523, pruned_loss=0.1051, over 5705514.61 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3676, pruned_loss=0.1134, over 5742740.49 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.349, pruned_loss=0.1022, over 5703491.90 frames. ], batch size: 285, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:13:28,904 INFO [optim.py:369] (1/2) 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 07:13:29,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=432260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:13:55,619 INFO [train.py:968] (1/2) Epoch 10, batch 22700, giga_loss[loss=0.369, simple_loss=0.4135, pruned_loss=0.1622, over 26568.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.352, pruned_loss=0.1038, over 5707581.10 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3683, pruned_loss=0.1138, over 5745404.83 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3483, pruned_loss=0.1008, over 5702759.15 frames. ], batch size: 555, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:14:14,103 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:14:34,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6152, 1.8232, 1.9563, 1.4381], device='cuda:1'), covar=tensor([0.1755, 0.2128, 0.1408, 0.1608], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0688, 0.0846, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 07:14:37,301 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:14:41,241 INFO [train.py:968] (1/2) Epoch 10, batch 22750, giga_loss[loss=0.3151, simple_loss=0.3906, pruned_loss=0.1198, over 28574.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3529, pruned_loss=0.1023, over 5704234.89 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3684, pruned_loss=0.1139, over 5743488.59 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09972, over 5701769.24 frames. ], batch size: 336, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:14:55,263 INFO [optim.py:369] (1/2) 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 07:15:23,392 INFO [train.py:968] (1/2) Epoch 10, batch 22800, giga_loss[loss=0.2595, simple_loss=0.3326, pruned_loss=0.09322, over 28853.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3529, pruned_loss=0.1027, over 5699818.30 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3685, pruned_loss=0.1142, over 5745911.52 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3501, pruned_loss=0.1003, over 5694933.72 frames. ], batch size: 186, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:15:29,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=432403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:15:31,345 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=432406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:15:36,653 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 07:15:57,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=432435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:16:05,079 INFO [train.py:968] (1/2) Epoch 10, batch 22850, giga_loss[loss=0.3556, simple_loss=0.396, pruned_loss=0.1577, over 26727.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3519, pruned_loss=0.104, over 5703093.39 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3687, pruned_loss=0.1145, over 5749908.13 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3491, pruned_loss=0.1015, over 5694628.79 frames. ], batch size: 555, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:16:07,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3871, 1.6252, 1.2871, 1.5934], device='cuda:1'), covar=tensor([0.2249, 0.2206, 0.2528, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.0937, 0.1117, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 07:16:16,979 INFO [optim.py:369] (1/2) 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 07:16:45,603 INFO [train.py:968] (1/2) Epoch 10, batch 22900, giga_loss[loss=0.3123, simple_loss=0.3742, pruned_loss=0.1252, over 27863.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3505, pruned_loss=0.1046, over 5709997.88 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3692, pruned_loss=0.115, over 5752546.37 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3476, pruned_loss=0.1021, over 5700337.74 frames. ], batch size: 412, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:17:22,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-05 07:17:26,251 INFO [train.py:968] (1/2) Epoch 10, batch 22950, giga_loss[loss=0.2844, simple_loss=0.352, pruned_loss=0.1084, over 29033.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3485, pruned_loss=0.1043, over 5712544.59 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3693, pruned_loss=0.1153, over 5748676.38 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3456, pruned_loss=0.1018, over 5707907.71 frames. ], batch size: 128, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:17:39,819 INFO [optim.py:369] (1/2) 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 07:18:07,867 INFO [train.py:968] (1/2) Epoch 10, batch 23000, giga_loss[loss=0.2914, simple_loss=0.3654, pruned_loss=0.1087, over 28574.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3489, pruned_loss=0.1054, over 5704213.01 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3701, pruned_loss=0.1158, over 5741506.30 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3455, pruned_loss=0.1027, over 5706010.96 frames. ], batch size: 336, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:18:21,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:18:25,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2279, 1.7666, 1.3084, 0.4713], device='cuda:1'), covar=tensor([0.2850, 0.1724, 0.3014, 0.3849], device='cuda:1'), in_proj_covar=tensor([0.1530, 0.1434, 0.1469, 0.1259], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 07:18:38,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3098, 1.0523, 4.5836, 3.3890], device='cuda:1'), covar=tensor([0.2106, 0.3200, 0.0569, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0642, 0.0575, 0.0838, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 07:18:40,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5674, 1.6818, 1.8132, 1.3905], device='cuda:1'), covar=tensor([0.1565, 0.2167, 0.1296, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0690, 0.0845, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 07:18:47,074 INFO [train.py:968] (1/2) Epoch 10, batch 23050, giga_loss[loss=0.221, simple_loss=0.2987, pruned_loss=0.07165, over 28741.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3458, pruned_loss=0.1034, over 5713152.05 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3695, pruned_loss=0.1156, over 5745413.15 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3432, pruned_loss=0.1013, over 5710301.45 frames. ], batch size: 119, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:18:47,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 07:19:00,364 INFO [optim.py:369] (1/2) 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 07:19:13,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-05 07:19:20,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=432686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:19:26,148 INFO [train.py:968] (1/2) Epoch 10, batch 23100, giga_loss[loss=0.2815, simple_loss=0.3303, pruned_loss=0.1163, over 28345.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3422, pruned_loss=0.102, over 5714688.49 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3697, pruned_loss=0.1157, over 5747006.87 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3395, pruned_loss=0.09992, over 5710447.71 frames. ], batch size: 65, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:19:42,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=432714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:20:04,576 INFO [train.py:968] (1/2) Epoch 10, batch 23150, giga_loss[loss=0.2564, simple_loss=0.3304, pruned_loss=0.09117, over 28643.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3386, pruned_loss=0.09992, over 5699013.21 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3696, pruned_loss=0.1159, over 5736199.80 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3356, pruned_loss=0.09761, over 5704933.58 frames. ], batch size: 307, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:20:17,481 INFO [optim.py:369] (1/2) 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 07:20:43,975 INFO [train.py:968] (1/2) Epoch 10, batch 23200, giga_loss[loss=0.2802, simple_loss=0.3609, pruned_loss=0.09975, over 28899.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3391, pruned_loss=0.09997, over 5705092.14 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3698, pruned_loss=0.1162, over 5734211.89 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3357, pruned_loss=0.09743, over 5710601.53 frames. ], batch size: 186, lr: 3.24e-03, grad_scale: 8.0
+2023-03-05 07:20:55,565 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-05 07:21:14,179 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=432829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:15,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=432832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:25,894 INFO [train.py:968] (1/2) Epoch 10, batch 23250, giga_loss[loss=0.3087, simple_loss=0.3658, pruned_loss=0.1258, over 24051.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3417, pruned_loss=0.1008, over 5705720.25 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.37, pruned_loss=0.1163, over 5737743.89 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3381, pruned_loss=0.09824, over 5706236.08 frames. ], batch size: 705, lr: 3.24e-03, grad_scale: 4.0
+2023-03-05 07:21:35,765 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:37,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=432857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:39,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=432860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:21:40,282 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:21:51,860 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=432874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:22:04,444 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=432889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:22:08,161 INFO [train.py:968] (1/2) Epoch 10, batch 23300, giga_loss[loss=0.3237, simple_loss=0.386, pruned_loss=0.1307, over 27717.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3458, pruned_loss=0.1025, over 5707952.94 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3705, pruned_loss=0.117, over 5738700.08 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3417, pruned_loss=0.0995, over 5706760.15 frames. ], batch size: 474, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:22:12,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-05 07:22:47,614 INFO [train.py:968] (1/2) Epoch 10, batch 23350, giga_loss[loss=0.2886, simple_loss=0.3488, pruned_loss=0.1142, over 23790.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3493, pruned_loss=0.1042, over 5711083.82 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3706, pruned_loss=0.1173, over 5744244.51 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3452, pruned_loss=0.101, over 5704224.92 frames. ], batch size: 705, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:22:54,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0824, 1.0672, 3.9937, 3.3566], device='cuda:1'), covar=tensor([0.1722, 0.2756, 0.0387, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0644, 0.0575, 0.0840, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 07:23:01,543 INFO [optim.py:369] (1/2) 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 07:23:19,512 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 23400, giga_loss[loss=0.2589, simple_loss=0.3432, pruned_loss=0.08727, over 28683.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3526, pruned_loss=0.1055, over 5710212.95 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3707, pruned_loss=0.1175, over 5748541.08 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3487, pruned_loss=0.1025, over 5700222.62 frames. ], batch size: 242, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:23:49,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1467, 2.4668, 1.9775, 1.8573], device='cuda:1'), covar=tensor([0.1832, 0.1304, 0.1467, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.1680, 0.1562, 0.1548, 0.1650], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 07:24:02,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 07:24:10,655 INFO [train.py:968] (1/2) Epoch 10, batch 23450, giga_loss[loss=0.2872, simple_loss=0.3489, pruned_loss=0.1128, over 28503.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3547, pruned_loss=0.1068, over 5708277.10 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3709, pruned_loss=0.1177, over 5750825.31 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3512, pruned_loss=0.1041, over 5697766.58 frames. ], batch size: 71, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:24:25,723 INFO [optim.py:369] (1/2) 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 07:24:30,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2498, 3.5219, 2.5306, 0.9883], device='cuda:1'), covar=tensor([0.4725, 0.1548, 0.2329, 0.4868], device='cuda:1'), in_proj_covar=tensor([0.1544, 0.1452, 0.1478, 0.1268], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 07:24:57,910 INFO [train.py:968] (1/2) Epoch 10, batch 23500, giga_loss[loss=0.3018, simple_loss=0.375, pruned_loss=0.1143, over 28810.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3632, pruned_loss=0.1144, over 5663642.77 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3719, pruned_loss=0.1186, over 5711681.04 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3591, pruned_loss=0.1112, over 5689902.67 frames. ], batch size: 186, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:25:06,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-05 07:25:14,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4453, 2.0735, 1.5664, 0.6072], device='cuda:1'), covar=tensor([0.2926, 0.1607, 0.2334, 0.3434], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1453, 0.1480, 0.1267], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 07:25:32,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=433127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:25:37,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=433130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:25:51,358 INFO [train.py:968] (1/2) Epoch 10, batch 23550, giga_loss[loss=0.4228, simple_loss=0.4459, pruned_loss=0.1998, over 26635.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3679, pruned_loss=0.118, over 5665236.44 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.372, pruned_loss=0.1186, over 5712649.63 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3647, pruned_loss=0.1155, over 5684691.68 frames. ], batch size: 555, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:26:07,925 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:26:41,579 INFO [train.py:968] (1/2) Epoch 10, batch 23600, giga_loss[loss=0.3001, simple_loss=0.3708, pruned_loss=0.1148, over 28629.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3743, pruned_loss=0.1228, over 5670607.20 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3719, pruned_loss=0.1186, over 5715892.54 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3719, pruned_loss=0.1208, over 5682201.19 frames. ], batch size: 242, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:27:20,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=433231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:27:33,476 INFO [train.py:968] (1/2) Epoch 10, batch 23650, giga_loss[loss=0.4171, simple_loss=0.4517, pruned_loss=0.1913, over 28581.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3819, pruned_loss=0.1302, over 5659274.79 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3715, pruned_loss=0.1183, over 5716912.56 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3802, pruned_loss=0.1289, over 5667228.48 frames. ], batch size: 336, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:27:40,412 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:28:22,612 INFO [train.py:968] (1/2) Epoch 10, batch 23700, giga_loss[loss=0.3992, simple_loss=0.4368, pruned_loss=0.1808, over 27560.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3884, pruned_loss=0.136, over 5660372.82 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3714, pruned_loss=0.1185, over 5721291.89 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3875, pruned_loss=0.1351, over 5661216.12 frames. ], batch size: 472, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:29:09,007 INFO [train.py:968] (1/2) Epoch 10, batch 23750, giga_loss[loss=0.35, simple_loss=0.4048, pruned_loss=0.1476, over 28933.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3908, pruned_loss=0.1379, over 5671002.67 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3713, pruned_loss=0.1187, over 5724545.11 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3906, pruned_loss=0.1375, over 5667532.48 frames. ], batch size: 227, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:29:29,663 INFO [optim.py:369] (1/2) 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 07:29:39,944 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=433374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:29:41,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=433377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:29:57,259 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 23800, giga_loss[loss=0.4223, simple_loss=0.429, pruned_loss=0.2078, over 23361.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3926, pruned_loss=0.1406, over 5659841.73 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3714, pruned_loss=0.1189, over 5726134.98 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3929, pruned_loss=0.1405, over 5654321.98 frames. ], batch size: 705, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:30:01,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=433395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:30:09,109 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=433406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:30:25,570 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=433424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:30:33,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 07:30:45,680 INFO [train.py:968] (1/2) Epoch 10, batch 23850, giga_loss[loss=0.33, simple_loss=0.3905, pruned_loss=0.1348, over 28661.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3944, pruned_loss=0.1431, over 5648140.22 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3717, pruned_loss=0.1194, over 5728503.65 frames. ], giga_tot_loss[loss=0.3411, simple_loss=0.3952, pruned_loss=0.1435, over 5639030.76 frames. ], batch size: 262, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:31:03,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 07:31:05,028 INFO [optim.py:369] (1/2) 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 07:31:19,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-05 07:31:36,265 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-05 07:31:38,643 INFO [train.py:968] (1/2) Epoch 10, batch 23900, giga_loss[loss=0.3911, simple_loss=0.4099, pruned_loss=0.1861, over 23553.00 frames. ], tot_loss[loss=0.345, simple_loss=0.3977, pruned_loss=0.1462, over 5646161.44 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3715, pruned_loss=0.1193, over 5730736.54 frames. ], giga_tot_loss[loss=0.3466, simple_loss=0.399, pruned_loss=0.1471, over 5635269.54 frames. ], batch size: 705, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:32:35,286 INFO [train.py:968] (1/2) Epoch 10, batch 23950, giga_loss[loss=0.3595, simple_loss=0.4052, pruned_loss=0.1569, over 28309.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.3997, pruned_loss=0.1489, over 5620680.76 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3711, pruned_loss=0.1193, over 5730832.69 frames. ], giga_tot_loss[loss=0.3512, simple_loss=0.4017, pruned_loss=0.1504, over 5609807.86 frames. ], batch size: 369, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:32:56,210 INFO [optim.py:369] (1/2) 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 07:33:00,022 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=433567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:33:08,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9182, 2.4280, 1.7937, 1.5633], device='cuda:1'), covar=tensor([0.1987, 0.1406, 0.1739, 0.1907], device='cuda:1'), in_proj_covar=tensor([0.1667, 0.1560, 0.1543, 0.1648], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 07:33:27,149 INFO [train.py:968] (1/2) Epoch 10, batch 24000, giga_loss[loss=0.3733, simple_loss=0.4235, pruned_loss=0.1615, over 28938.00 frames. ], tot_loss[loss=0.3493, simple_loss=0.3993, pruned_loss=0.1496, over 5622155.85 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3717, pruned_loss=0.1198, over 5729127.36 frames. ], giga_tot_loss[loss=0.3517, simple_loss=0.4012, pruned_loss=0.1511, over 5612001.43 frames. ], batch size: 227, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:33:27,150 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 07:33:34,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0377, 1.2400, 3.3020, 3.0225], device='cuda:1'), covar=tensor([0.1710, 0.2529, 0.0508, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0651, 0.0579, 0.0847, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 07:33:35,816 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 07:34:23,983 INFO [train.py:968] (1/2) Epoch 10, batch 24050, giga_loss[loss=0.4378, simple_loss=0.456, pruned_loss=0.2098, over 27412.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.3985, pruned_loss=0.1496, over 5639748.97 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.372, pruned_loss=0.1201, over 5731581.53 frames. ], giga_tot_loss[loss=0.3511, simple_loss=0.4002, pruned_loss=0.151, over 5627727.16 frames. ], batch size: 472, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:34:44,339 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 24100, giga_loss[loss=0.3505, simple_loss=0.4099, pruned_loss=0.1455, over 28703.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3974, pruned_loss=0.1475, over 5634844.08 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1201, over 5734297.73 frames. ], giga_tot_loss[loss=0.3486, simple_loss=0.3992, pruned_loss=0.1489, over 5621475.47 frames. ], batch size: 262, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:35:31,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 07:35:57,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3985, 1.7648, 1.4161, 1.5175], device='cuda:1'), covar=tensor([0.0681, 0.0379, 0.0303, 0.0730], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0077, 0.0055, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 07:36:04,150 INFO [train.py:968] (1/2) Epoch 10, batch 24150, giga_loss[loss=0.4047, simple_loss=0.4397, pruned_loss=0.1848, over 27600.00 frames. ], tot_loss[loss=0.347, simple_loss=0.3985, pruned_loss=0.1477, over 5627232.06 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3722, pruned_loss=0.1204, over 5738388.54 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4003, pruned_loss=0.1493, over 5610522.96 frames. ], batch size: 472, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:36:24,987 INFO [optim.py:369] (1/2) 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 07:36:53,667 INFO [train.py:968] (1/2) Epoch 10, batch 24200, libri_loss[loss=0.3657, simple_loss=0.4128, pruned_loss=0.1593, over 29753.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3989, pruned_loss=0.1474, over 5631758.09 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3723, pruned_loss=0.1208, over 5740269.64 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4009, pruned_loss=0.149, over 5613797.62 frames. ], batch size: 87, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:37:08,882 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=433809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:37:45,832 INFO [train.py:968] (1/2) Epoch 10, batch 24250, giga_loss[loss=0.3442, simple_loss=0.381, pruned_loss=0.1537, over 23926.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3968, pruned_loss=0.1453, over 5627014.57 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3731, pruned_loss=0.1214, over 5741687.46 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.3983, pruned_loss=0.1466, over 5609194.91 frames. ], batch size: 705, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:38:03,516 INFO [optim.py:369] (1/2) 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 07:38:13,912 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-05 07:38:21,524 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=433881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:38:32,858 INFO [train.py:968] (1/2) Epoch 10, batch 24300, libri_loss[loss=0.2967, simple_loss=0.3617, pruned_loss=0.1158, over 19526.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3939, pruned_loss=0.1416, over 5633409.50 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3732, pruned_loss=0.1217, over 5734402.68 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3954, pruned_loss=0.1429, over 5624188.48 frames. ], batch size: 186, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:38:55,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-05 07:39:11,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3717, 1.7937, 1.3773, 1.6510], device='cuda:1'), covar=tensor([0.0705, 0.0268, 0.0307, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 07:39:20,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=433942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:39:22,007 INFO [train.py:968] (1/2) Epoch 10, batch 24350, giga_loss[loss=0.2928, simple_loss=0.3605, pruned_loss=0.1126, over 28279.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3901, pruned_loss=0.1381, over 5632867.67 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3732, pruned_loss=0.1218, over 5736666.92 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3916, pruned_loss=0.1392, over 5622042.66 frames. ], batch size: 368, lr: 3.23e-03, grad_scale: 1.0
+2023-03-05 07:39:40,313 INFO [optim.py:369] (1/2) 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 07:40:08,733 INFO [train.py:968] (1/2) Epoch 10, batch 24400, giga_loss[loss=0.2708, simple_loss=0.3372, pruned_loss=0.1022, over 28789.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3881, pruned_loss=0.1366, over 5622955.51 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.122, over 5722004.19 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3896, pruned_loss=0.1378, over 5623013.15 frames. ], batch size: 99, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:40:38,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-05 07:40:59,318 INFO [train.py:968] (1/2) Epoch 10, batch 24450, giga_loss[loss=0.3163, simple_loss=0.3799, pruned_loss=0.1264, over 28920.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3858, pruned_loss=0.135, over 5617714.86 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3735, pruned_loss=0.1223, over 5711293.21 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3871, pruned_loss=0.136, over 5624713.08 frames. ], batch size: 213, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:41:22,432 INFO [optim.py:369] (1/2) 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 07:41:41,293 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:41:52,467 INFO [train.py:968] (1/2) Epoch 10, batch 24500, giga_loss[loss=0.3069, simple_loss=0.3768, pruned_loss=0.1185, over 28501.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3864, pruned_loss=0.1353, over 5625178.17 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3738, pruned_loss=0.1224, over 5713435.43 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3874, pruned_loss=0.1362, over 5627416.12 frames. ], batch size: 336, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:42:18,046 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:42:26,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2250, 1.3143, 3.8587, 3.1105], device='cuda:1'), covar=tensor([0.1624, 0.2434, 0.0441, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0652, 0.0581, 0.0851, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 07:42:36,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-05 07:42:38,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:42:41,852 INFO [train.py:968] (1/2) Epoch 10, batch 24550, giga_loss[loss=0.3674, simple_loss=0.4213, pruned_loss=0.1567, over 28149.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3861, pruned_loss=0.1345, over 5632521.32 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3741, pruned_loss=0.1227, over 5707961.77 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.387, pruned_loss=0.1352, over 5637325.43 frames. ], batch size: 77, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:43:04,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2670, 3.0471, 2.9203, 1.4567], device='cuda:1'), covar=tensor([0.0836, 0.0971, 0.0939, 0.2205], device='cuda:1'), in_proj_covar=tensor([0.1039, 0.0982, 0.0861, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 07:43:06,256 INFO [optim.py:369] (1/2) 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 07:43:25,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:43:34,441 INFO [train.py:968] (1/2) Epoch 10, batch 24600, libri_loss[loss=0.3291, simple_loss=0.3915, pruned_loss=0.1333, over 29214.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3842, pruned_loss=0.1314, over 5649447.44 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3736, pruned_loss=0.1226, over 5712462.51 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3855, pruned_loss=0.1324, over 5647534.29 frames. ], batch size: 97, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:44:26,010 INFO [train.py:968] (1/2) Epoch 10, batch 24650, giga_loss[loss=0.3014, simple_loss=0.3757, pruned_loss=0.1135, over 28566.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3856, pruned_loss=0.1303, over 5663343.69 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3737, pruned_loss=0.1228, over 5715745.06 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3868, pruned_loss=0.131, over 5657678.18 frames. ], batch size: 60, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:44:31,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5042, 1.6273, 1.4097, 1.6882], device='cuda:1'), covar=tensor([0.1969, 0.1969, 0.1972, 0.1874], device='cuda:1'), in_proj_covar=tensor([0.1263, 0.0942, 0.1123, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 07:44:32,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4141, 1.5578, 1.2442, 1.6597], device='cuda:1'), covar=tensor([0.2310, 0.2351, 0.2539, 0.2114], device='cuda:1'), in_proj_covar=tensor([0.1263, 0.0941, 0.1123, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 07:44:39,326 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:45:01,577 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-05 07:45:14,198 INFO [train.py:968] (1/2) Epoch 10, batch 24700, libri_loss[loss=0.2941, simple_loss=0.3673, pruned_loss=0.1104, over 29279.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3857, pruned_loss=0.1308, over 5649533.85 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3736, pruned_loss=0.1227, over 5712118.28 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3872, pruned_loss=0.1317, over 5645638.37 frames. ], batch size: 94, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:45:48,963 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:45:49,441 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:45:51,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:45:57,508 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:46:02,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5575, 2.4142, 2.4393, 2.0588], device='cuda:1'), covar=tensor([0.1399, 0.2132, 0.1718, 0.2026], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0730, 0.0665, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 07:46:04,502 INFO [train.py:968] (1/2) Epoch 10, batch 24750, giga_loss[loss=0.2802, simple_loss=0.3578, pruned_loss=0.1013, over 29060.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3857, pruned_loss=0.1306, over 5667852.78 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3733, pruned_loss=0.1225, over 5712447.05 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3873, pruned_loss=0.1316, over 5663663.42 frames. ], batch size: 136, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:46:19,789 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:46:20,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:46:24,517 INFO [optim.py:369] (1/2) 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 07:46:53,714 INFO [train.py:968] (1/2) Epoch 10, batch 24800, giga_loss[loss=0.3348, simple_loss=0.3929, pruned_loss=0.1384, over 28569.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3834, pruned_loss=0.1297, over 5684823.79 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.373, pruned_loss=0.1226, over 5718079.67 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3853, pruned_loss=0.1307, over 5675061.14 frames. ], batch size: 336, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:46:59,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:02,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:27,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:37,392 INFO [train.py:968] (1/2) Epoch 10, batch 24850, giga_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1227, over 28297.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3827, pruned_loss=0.1308, over 5683038.04 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1225, over 5720952.50 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3847, pruned_loss=0.1319, over 5671894.73 frames. ], batch size: 65, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:47:41,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3906, 2.0620, 1.4527, 0.5758], device='cuda:1'), covar=tensor([0.2928, 0.1720, 0.2702, 0.3149], device='cuda:1'), in_proj_covar=tensor([0.1552, 0.1468, 0.1489, 0.1276], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 07:47:51,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:47:56,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.32 vs. limit=5.0
+2023-03-05 07:47:56,566 INFO [optim.py:369] (1/2) 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 07:48:00,228 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-05 07:48:21,400 INFO [train.py:968] (1/2) Epoch 10, batch 24900, giga_loss[loss=0.3045, simple_loss=0.3727, pruned_loss=0.1182, over 28494.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3825, pruned_loss=0.1311, over 5663289.13 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3733, pruned_loss=0.123, over 5704211.29 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3837, pruned_loss=0.1316, over 5669387.86 frames. ], batch size: 336, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:48:41,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:49:04,077 INFO [train.py:968] (1/2) Epoch 10, batch 24950, giga_loss[loss=0.2796, simple_loss=0.3576, pruned_loss=0.1008, over 29026.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3816, pruned_loss=0.1293, over 5678401.00 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3734, pruned_loss=0.1233, over 5712117.34 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3829, pruned_loss=0.1297, over 5674622.51 frames. ], batch size: 213, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:49:27,729 INFO [optim.py:369] (1/2) 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 07:49:34,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1943, 1.3677, 1.1247, 1.0384], device='cuda:1'), covar=tensor([0.1391, 0.1412, 0.1008, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.1648, 0.1535, 0.1528, 0.1624], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 07:49:54,709 INFO [train.py:968] (1/2) Epoch 10, batch 25000, giga_loss[loss=0.3016, simple_loss=0.3716, pruned_loss=0.1158, over 28259.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3812, pruned_loss=0.1277, over 5678575.17 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1235, over 5711595.08 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3822, pruned_loss=0.1279, over 5675739.94 frames. ], batch size: 77, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:50:17,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4783, 2.1639, 1.6127, 0.7247], device='cuda:1'), covar=tensor([0.3212, 0.1910, 0.2764, 0.3936], device='cuda:1'), in_proj_covar=tensor([0.1541, 0.1462, 0.1477, 0.1271], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 07:50:41,713 INFO [train.py:968] (1/2) Epoch 10, batch 25050, giga_loss[loss=0.3183, simple_loss=0.3896, pruned_loss=0.1235, over 28393.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3811, pruned_loss=0.1281, over 5676987.89 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1234, over 5716318.53 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3821, pruned_loss=0.1284, over 5669463.88 frames. ], batch size: 368, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:50:57,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:00,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:04,516 INFO [optim.py:369] (1/2) 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 07:51:26,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:30,390 INFO [train.py:968] (1/2) Epoch 10, batch 25100, giga_loss[loss=0.3689, simple_loss=0.4173, pruned_loss=0.1603, over 28646.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.381, pruned_loss=0.1288, over 5680956.63 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.124, over 5717370.27 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3816, pruned_loss=0.1287, over 5673263.98 frames. ], batch size: 262, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:51:40,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:51:49,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:52:09,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=434735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:52:18,328 INFO [train.py:968] (1/2) Epoch 10, batch 25150, giga_loss[loss=0.3218, simple_loss=0.3775, pruned_loss=0.133, over 28013.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3802, pruned_loss=0.1291, over 5666808.98 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3741, pruned_loss=0.1241, over 5717336.45 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3806, pruned_loss=0.129, over 5659686.46 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:52:38,056 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 25200, giga_loss[loss=0.3015, simple_loss=0.3585, pruned_loss=0.1222, over 28479.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3784, pruned_loss=0.1287, over 5663514.18 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3734, pruned_loss=0.124, over 5711481.53 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3796, pruned_loss=0.1289, over 5661915.50 frames. ], batch size: 85, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:53:14,466 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:53:42,241 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 25250, giga_loss[loss=0.2788, simple_loss=0.3424, pruned_loss=0.1076, over 28760.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3789, pruned_loss=0.1297, over 5670017.68 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1243, over 5713410.85 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3796, pruned_loss=0.1296, over 5666650.40 frames. ], batch size: 119, lr: 3.23e-03, grad_scale: 8.0
+2023-03-05 07:53:53,795 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:53:57,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:05,391 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:08,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:14,811 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=434877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:24,389 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:24,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:27,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:35,622 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:54:40,601 INFO [train.py:968] (1/2) Epoch 10, batch 25300, giga_loss[loss=0.2916, simple_loss=0.3577, pruned_loss=0.1127, over 29071.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5669093.58 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3738, pruned_loss=0.1243, over 5713565.79 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3768, pruned_loss=0.1279, over 5666119.94 frames. ], batch size: 136, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:54:55,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=434910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:55:26,393 INFO [train.py:968] (1/2) Epoch 10, batch 25350, giga_loss[loss=0.3008, simple_loss=0.3677, pruned_loss=0.117, over 28922.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3775, pruned_loss=0.1293, over 5664223.89 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3747, pruned_loss=0.125, over 5712492.50 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3771, pruned_loss=0.1288, over 5660719.31 frames. ], batch size: 112, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:55:48,188 INFO [optim.py:369] (1/2) 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 07:55:56,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=434977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:55:59,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=434980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:56:15,119 INFO [train.py:968] (1/2) Epoch 10, batch 25400, giga_loss[loss=0.3734, simple_loss=0.4206, pruned_loss=0.1631, over 28904.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3788, pruned_loss=0.1298, over 5662853.10 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3748, pruned_loss=0.125, over 5712344.36 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3785, pruned_loss=0.1295, over 5658791.01 frames. ], batch size: 199, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:56:27,773 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 07:56:59,082 INFO [train.py:968] (1/2) Epoch 10, batch 25450, giga_loss[loss=0.3234, simple_loss=0.388, pruned_loss=0.1294, over 28971.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3784, pruned_loss=0.1283, over 5661134.72 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5705655.16 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3782, pruned_loss=0.128, over 5663945.47 frames. ], batch size: 227, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:57:21,610 INFO [optim.py:369] (1/2) 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 07:57:45,191 INFO [train.py:968] (1/2) Epoch 10, batch 25500, giga_loss[loss=0.3231, simple_loss=0.3799, pruned_loss=0.1332, over 28665.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3789, pruned_loss=0.1282, over 5660135.30 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3745, pruned_loss=0.1249, over 5709405.75 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3791, pruned_loss=0.1283, over 5658045.07 frames. ], batch size: 92, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:58:32,041 INFO [train.py:968] (1/2) Epoch 10, batch 25550, giga_loss[loss=0.2669, simple_loss=0.3446, pruned_loss=0.09458, over 28960.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3782, pruned_loss=0.128, over 5666880.23 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3741, pruned_loss=0.1247, over 5711337.99 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3788, pruned_loss=0.1283, over 5662613.80 frames. ], batch size: 213, lr: 3.23e-03, grad_scale: 2.0
+2023-03-05 07:58:51,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1455, 1.4579, 1.3863, 1.0483], device='cuda:1'), covar=tensor([0.0993, 0.1583, 0.0867, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0698, 0.0844, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 07:58:55,126 INFO [optim.py:369] (1/2) 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 07:59:05,907 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 25600, giga_loss[loss=0.3014, simple_loss=0.3646, pruned_loss=0.1191, over 28449.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3828, pruned_loss=0.1325, over 5639790.03 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1254, over 5697589.57 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3828, pruned_loss=0.1322, over 5647025.81 frames. ], batch size: 71, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 07:59:28,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3015, 1.4549, 1.6087, 1.4232], device='cuda:1'), covar=tensor([0.0914, 0.0781, 0.1110, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0731, 0.0665, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 08:00:01,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5580, 1.9986, 1.4646, 1.5631], device='cuda:1'), covar=tensor([0.0726, 0.0255, 0.0319, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 08:00:09,431 INFO [train.py:968] (1/2) Epoch 10, batch 25650, giga_loss[loss=0.3233, simple_loss=0.3847, pruned_loss=0.131, over 28738.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3846, pruned_loss=0.135, over 5641217.75 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3749, pruned_loss=0.1255, over 5692301.89 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3845, pruned_loss=0.1348, over 5651394.00 frames. ], batch size: 284, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:00:15,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=435252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:00:26,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0218, 0.9707, 3.9275, 3.1960], device='cuda:1'), covar=tensor([0.1831, 0.2773, 0.0465, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0652, 0.0584, 0.0855, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:00:34,103 INFO [optim.py:369] (1/2) 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 08:00:59,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0360, 1.2119, 1.2470, 1.0242], device='cuda:1'), covar=tensor([0.1041, 0.1029, 0.1526, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0735, 0.0667, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 08:01:00,619 INFO [train.py:968] (1/2) Epoch 10, batch 25700, giga_loss[loss=0.3206, simple_loss=0.3741, pruned_loss=0.1336, over 28865.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3849, pruned_loss=0.1363, over 5651894.59 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3752, pruned_loss=0.1257, over 5695683.03 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3849, pruned_loss=0.1361, over 5656336.95 frames. ], batch size: 199, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:01:33,386 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=435323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:01:35,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=435326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:01:49,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=435342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:01:49,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5385, 1.8191, 1.6248, 1.2490], device='cuda:1'), covar=tensor([0.2289, 0.1675, 0.1433, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1660, 0.1561, 0.1544, 0.1640], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 08:01:50,170 INFO [train.py:968] (1/2) Epoch 10, batch 25750, giga_loss[loss=0.3059, simple_loss=0.3706, pruned_loss=0.1206, over 28993.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3856, pruned_loss=0.1369, over 5648929.94 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.375, pruned_loss=0.1255, over 5699029.99 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3859, pruned_loss=0.1371, over 5648789.11 frames. ], batch size: 145, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:01:59,800 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:02:10,595 INFO [optim.py:369] (1/2) 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 08:02:15,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9146, 3.2193, 2.0726, 1.1151], device='cuda:1'), covar=tensor([0.4079, 0.1512, 0.2173, 0.3210], device='cuda:1'), in_proj_covar=tensor([0.1546, 0.1477, 0.1487, 0.1275], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 08:02:35,760 INFO [train.py:968] (1/2) Epoch 10, batch 25800, giga_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.094, over 28912.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3842, pruned_loss=0.1359, over 5656129.28 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3747, pruned_loss=0.1254, over 5702634.25 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3849, pruned_loss=0.1363, over 5651841.73 frames. ], batch size: 106, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:02:37,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=435395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:02:41,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=435398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:03:06,320 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:03:23,076 INFO [train.py:968] (1/2) Epoch 10, batch 25850, giga_loss[loss=0.3095, simple_loss=0.3745, pruned_loss=0.1223, over 27853.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3832, pruned_loss=0.134, over 5664140.83 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3745, pruned_loss=0.1252, over 5705452.95 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3841, pruned_loss=0.1347, over 5657772.31 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:03:44,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3020, 1.3329, 3.3649, 3.1710], device='cuda:1'), covar=tensor([0.1286, 0.2307, 0.0418, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0646, 0.0581, 0.0847, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:03:45,442 INFO [optim.py:369] (1/2) 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 08:04:02,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2926, 1.6532, 1.4246, 1.5131], device='cuda:1'), covar=tensor([0.0707, 0.0355, 0.0298, 0.0723], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 08:04:08,266 INFO [train.py:968] (1/2) Epoch 10, batch 25900, giga_loss[loss=0.306, simple_loss=0.3678, pruned_loss=0.1221, over 27957.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3813, pruned_loss=0.1318, over 5665395.06 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3755, pruned_loss=0.126, over 5711038.37 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3811, pruned_loss=0.1318, over 5653963.81 frames. ], batch size: 412, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:04:20,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1914, 1.5414, 1.5256, 1.1055], device='cuda:1'), covar=tensor([0.1530, 0.2283, 0.1241, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0702, 0.0848, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 08:04:36,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.83 vs. limit=5.0
+2023-03-05 08:04:53,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4832, 3.4614, 1.5574, 1.5078], device='cuda:1'), covar=tensor([0.0902, 0.0297, 0.0844, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0511, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 08:04:55,369 INFO [train.py:968] (1/2) Epoch 10, batch 25950, giga_loss[loss=0.3271, simple_loss=0.3741, pruned_loss=0.1401, over 28554.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3796, pruned_loss=0.1309, over 5667022.45 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.376, pruned_loss=0.1263, over 5711740.90 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3792, pruned_loss=0.1307, over 5656892.53 frames. ], batch size: 60, lr: 3.23e-03, grad_scale: 4.0
+2023-03-05 08:05:16,081 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=435567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:05:16,559 INFO [optim.py:369] (1/2) 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 08:05:38,859 INFO [train.py:968] (1/2) Epoch 10, batch 26000, libri_loss[loss=0.3185, simple_loss=0.3849, pruned_loss=0.126, over 29551.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3769, pruned_loss=0.1296, over 5680761.82 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3761, pruned_loss=0.1265, over 5718511.45 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3765, pruned_loss=0.1293, over 5665091.23 frames. ], batch size: 89, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:06:28,609 INFO [train.py:968] (1/2) Epoch 10, batch 26050, libri_loss[loss=0.3255, simple_loss=0.3787, pruned_loss=0.1362, over 29537.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3764, pruned_loss=0.1294, over 5678602.12 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3762, pruned_loss=0.1268, over 5715092.09 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3759, pruned_loss=0.1289, over 5666939.48 frames. ], batch size: 80, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:06:50,960 INFO [optim.py:369] (1/2) 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 08:07:15,130 INFO [train.py:968] (1/2) Epoch 10, batch 26100, giga_loss[loss=0.2985, simple_loss=0.3738, pruned_loss=0.1116, over 28870.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3789, pruned_loss=0.1298, over 5684899.88 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3762, pruned_loss=0.1268, over 5716885.11 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3785, pruned_loss=0.1295, over 5673980.41 frames. ], batch size: 112, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:07:35,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=435717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:07:58,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5476, 1.9135, 1.8074, 1.3551], device='cuda:1'), covar=tensor([0.1784, 0.2293, 0.1489, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0699, 0.0848, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 08:08:01,492 INFO [train.py:968] (1/2) Epoch 10, batch 26150, giga_loss[loss=0.2939, simple_loss=0.3755, pruned_loss=0.1061, over 28897.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3823, pruned_loss=0.1291, over 5674330.78 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.1269, over 5701692.19 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3822, pruned_loss=0.1289, over 5677615.95 frames. ], batch size: 227, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:08:28,072 INFO [optim.py:369] (1/2) 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 08:08:30,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2141, 1.6075, 1.2797, 1.4319], device='cuda:1'), covar=tensor([0.0749, 0.0348, 0.0331, 0.0782], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 08:08:52,443 INFO [train.py:968] (1/2) Epoch 10, batch 26200, giga_loss[loss=0.356, simple_loss=0.4125, pruned_loss=0.1498, over 28731.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3839, pruned_loss=0.1295, over 5673210.46 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1269, over 5703679.28 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.384, pruned_loss=0.1293, over 5673761.15 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:09:11,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3254, 3.1818, 1.5336, 1.4300], device='cuda:1'), covar=tensor([0.0935, 0.0345, 0.0820, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0510, 0.0336, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 08:09:40,043 INFO [train.py:968] (1/2) Epoch 10, batch 26250, giga_loss[loss=0.3551, simple_loss=0.4034, pruned_loss=0.1534, over 28291.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3857, pruned_loss=0.1314, over 5669578.45 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3759, pruned_loss=0.1271, over 5695083.95 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3861, pruned_loss=0.1312, over 5677952.05 frames. ], batch size: 368, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:09:41,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1919, 1.3474, 1.2409, 1.0215], device='cuda:1'), covar=tensor([0.1581, 0.1605, 0.1010, 0.1406], device='cuda:1'), in_proj_covar=tensor([0.1652, 0.1554, 0.1541, 0.1639], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 08:09:55,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=435860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:09:58,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=435863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:10:03,127 INFO [optim.py:369] (1/2) 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 08:10:24,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=435892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:10:25,202 INFO [train.py:968] (1/2) Epoch 10, batch 26300, giga_loss[loss=0.3854, simple_loss=0.4221, pruned_loss=0.1743, over 28925.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3867, pruned_loss=0.1326, over 5668397.71 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3758, pruned_loss=0.127, over 5696320.08 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3871, pruned_loss=0.1326, over 5673709.75 frames. ], batch size: 227, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:11:05,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0229, 2.8103, 2.6861, 1.3786], device='cuda:1'), covar=tensor([0.0946, 0.1250, 0.1060, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.1027, 0.0972, 0.0850, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 08:11:13,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=435942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:11:14,644 INFO [train.py:968] (1/2) Epoch 10, batch 26350, giga_loss[loss=0.2822, simple_loss=0.3511, pruned_loss=0.1066, over 28682.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3869, pruned_loss=0.1339, over 5672105.11 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3761, pruned_loss=0.1271, over 5697477.78 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3873, pruned_loss=0.1339, over 5674825.33 frames. ], batch size: 71, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:11:44,258 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 26400, giga_loss[loss=0.2215, simple_loss=0.3034, pruned_loss=0.0698, over 28677.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3848, pruned_loss=0.1328, over 5680656.82 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3762, pruned_loss=0.1272, over 5699804.27 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3851, pruned_loss=0.1328, over 5680363.80 frames. ], batch size: 60, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:12:26,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=436017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:12:43,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3336, 1.2415, 1.2292, 1.3678], device='cuda:1'), covar=tensor([0.0753, 0.0341, 0.0314, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 08:12:52,367 INFO [train.py:968] (1/2) Epoch 10, batch 26450, giga_loss[loss=0.3246, simple_loss=0.3823, pruned_loss=0.1334, over 28772.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3834, pruned_loss=0.1329, over 5683233.27 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3757, pruned_loss=0.1267, over 5701943.71 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3841, pruned_loss=0.1333, over 5681041.23 frames. ], batch size: 243, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:13:21,678 INFO [optim.py:369] (1/2) 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 08:13:35,741 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=436085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:13:38,813 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 26500, giga_loss[loss=0.3364, simple_loss=0.3991, pruned_loss=0.1369, over 29087.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3818, pruned_loss=0.1322, over 5682488.24 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3754, pruned_loss=0.1265, over 5705737.47 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3829, pruned_loss=0.1329, over 5677140.38 frames. ], batch size: 128, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:13:58,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9159, 1.0734, 3.3515, 2.9004], device='cuda:1'), covar=tensor([0.1796, 0.2683, 0.0506, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0649, 0.0583, 0.0848, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:14:08,862 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=436117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:14:29,556 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-05 08:14:30,356 INFO [train.py:968] (1/2) Epoch 10, batch 26550, giga_loss[loss=0.3242, simple_loss=0.38, pruned_loss=0.1342, over 27938.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3821, pruned_loss=0.1326, over 5683032.82 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3754, pruned_loss=0.1268, over 5709388.39 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.383, pruned_loss=0.1331, over 5674580.42 frames. ], batch size: 412, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:14:51,076 INFO [optim.py:369] (1/2) 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 08:15:11,618 INFO [train.py:968] (1/2) Epoch 10, batch 26600, giga_loss[loss=0.3258, simple_loss=0.3748, pruned_loss=0.1384, over 28701.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3814, pruned_loss=0.1327, over 5686056.13 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1267, over 5715708.95 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3826, pruned_loss=0.1334, over 5672544.09 frames. ], batch size: 119, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:16:01,091 INFO [train.py:968] (1/2) Epoch 10, batch 26650, giga_loss[loss=0.3175, simple_loss=0.3837, pruned_loss=0.1257, over 28986.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3809, pruned_loss=0.1335, over 5665424.49 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3753, pruned_loss=0.1267, over 5716774.40 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3818, pruned_loss=0.134, over 5653534.87 frames. ], batch size: 164, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:16:03,570 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:16:30,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6835, 1.8283, 1.6612, 1.4835], device='cuda:1'), covar=tensor([0.1752, 0.1670, 0.1303, 0.1549], device='cuda:1'), in_proj_covar=tensor([0.1662, 0.1571, 0.1547, 0.1653], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 08:16:30,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5979, 2.3977, 1.6458, 1.9481], device='cuda:1'), covar=tensor([0.0696, 0.0208, 0.0292, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 08:16:44,847 INFO [train.py:968] (1/2) Epoch 10, batch 26700, giga_loss[loss=0.3554, simple_loss=0.4071, pruned_loss=0.1518, over 28543.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3804, pruned_loss=0.1323, over 5666758.36 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3754, pruned_loss=0.1268, over 5712934.62 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3813, pruned_loss=0.1328, over 5658137.28 frames. ], batch size: 336, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:17:11,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2733, 1.4333, 1.5128, 1.2751], device='cuda:1'), covar=tensor([0.1325, 0.1509, 0.1821, 0.1609], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0734, 0.0668, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 08:17:35,147 INFO [train.py:968] (1/2) Epoch 10, batch 26750, giga_loss[loss=0.4286, simple_loss=0.4491, pruned_loss=0.2041, over 26691.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3829, pruned_loss=0.1334, over 5666438.00 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3749, pruned_loss=0.1265, over 5714692.43 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3841, pruned_loss=0.1342, over 5657574.65 frames. ], batch size: 555, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:18:01,202 INFO [optim.py:369] (1/2) 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 08:18:01,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5368, 1.6517, 1.3293, 1.9228], device='cuda:1'), covar=tensor([0.2322, 0.2413, 0.2578, 0.2258], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.0954, 0.1129, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 08:18:24,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=436392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:18:25,307 INFO [train.py:968] (1/2) Epoch 10, batch 26800, libri_loss[loss=0.3724, simple_loss=0.4219, pruned_loss=0.1615, over 29533.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3837, pruned_loss=0.1344, over 5654879.05 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3752, pruned_loss=0.1268, over 5709270.93 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3845, pruned_loss=0.1349, over 5652512.25 frames. ], batch size: 83, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:18:50,816 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-05 08:19:09,488 INFO [train.py:968] (1/2) Epoch 10, batch 26850, giga_loss[loss=0.2942, simple_loss=0.3763, pruned_loss=0.1061, over 28925.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3841, pruned_loss=0.1336, over 5665390.16 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3753, pruned_loss=0.127, over 5704245.13 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3849, pruned_loss=0.134, over 5667202.05 frames. ], batch size: 213, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:19:33,248 INFO [optim.py:369] (1/2) 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 08:19:56,213 INFO [train.py:968] (1/2) Epoch 10, batch 26900, giga_loss[loss=0.2745, simple_loss=0.3631, pruned_loss=0.09293, over 28977.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3831, pruned_loss=0.1295, over 5672823.26 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3751, pruned_loss=0.1268, over 5707308.78 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.384, pruned_loss=0.13, over 5671114.35 frames. ], batch size: 155, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:20:32,449 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=436535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:20:35,793 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 26950, giga_loss[loss=0.285, simple_loss=0.3653, pruned_loss=0.1024, over 28148.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3858, pruned_loss=0.1297, over 5681788.15 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3749, pruned_loss=0.1266, over 5711879.63 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.387, pruned_loss=0.1304, over 5675594.25 frames. ], batch size: 77, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:20:51,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8212, 1.8483, 1.2981, 1.4392], device='cuda:1'), covar=tensor([0.0718, 0.0574, 0.0966, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0438, 0.0497, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:20:59,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=436567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:21:01,922 INFO [optim.py:369] (1/2) 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 08:21:07,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3530, 1.6891, 1.3180, 1.6818], device='cuda:1'), covar=tensor([0.2412, 0.2378, 0.2717, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.0951, 0.1130, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 08:21:22,433 INFO [train.py:968] (1/2) Epoch 10, batch 27000, giga_loss[loss=0.363, simple_loss=0.4138, pruned_loss=0.1561, over 29011.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3882, pruned_loss=0.1317, over 5669174.87 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3751, pruned_loss=0.1269, over 5697311.62 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3892, pruned_loss=0.132, over 5677350.55 frames. ], batch size: 136, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:21:22,433 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 08:21:30,946 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 08:21:50,588 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-05 08:21:53,722 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 08:21:56,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=436621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:21:58,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4772, 1.7541, 1.3999, 1.7317], device='cuda:1'), covar=tensor([0.2233, 0.2166, 0.2425, 0.1959], device='cuda:1'), in_proj_covar=tensor([0.1272, 0.0949, 0.1128, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 08:22:19,773 INFO [train.py:968] (1/2) Epoch 10, batch 27050, giga_loss[loss=0.2903, simple_loss=0.3633, pruned_loss=0.1087, over 28930.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3907, pruned_loss=0.1352, over 5668737.41 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.375, pruned_loss=0.1271, over 5701728.89 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3919, pruned_loss=0.1355, over 5670716.29 frames. ], batch size: 213, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:22:36,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4711, 3.2604, 1.5114, 1.5322], device='cuda:1'), covar=tensor([0.0907, 0.0329, 0.0861, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0511, 0.0335, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 08:22:50,347 INFO [optim.py:369] (1/2) 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 08:23:13,765 INFO [train.py:968] (1/2) Epoch 10, batch 27100, giga_loss[loss=0.3742, simple_loss=0.4147, pruned_loss=0.1668, over 28590.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3924, pruned_loss=0.1382, over 5647152.19 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3748, pruned_loss=0.1269, over 5703779.49 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3938, pruned_loss=0.1387, over 5646339.64 frames. ], batch size: 307, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:23:59,448 INFO [train.py:968] (1/2) Epoch 10, batch 27150, giga_loss[loss=0.3096, simple_loss=0.3736, pruned_loss=0.1228, over 28642.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3889, pruned_loss=0.1351, over 5656534.95 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3744, pruned_loss=0.1266, over 5701624.35 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3911, pruned_loss=0.1363, over 5655638.42 frames. ], batch size: 242, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:24:04,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2901, 1.6216, 1.4741, 1.4672], device='cuda:1'), covar=tensor([0.0731, 0.0318, 0.0293, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 08:24:21,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=436764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:24:24,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=436767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:24:27,481 INFO [optim.py:369] (1/2) 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 08:24:47,789 INFO [train.py:968] (1/2) Epoch 10, batch 27200, giga_loss[loss=0.2667, simple_loss=0.3533, pruned_loss=0.09005, over 28635.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.389, pruned_loss=0.1348, over 5646576.67 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3743, pruned_loss=0.1265, over 5700698.14 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3911, pruned_loss=0.1361, over 5645797.82 frames. ], batch size: 92, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:24:49,472 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=436796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:25:36,421 INFO [train.py:968] (1/2) Epoch 10, batch 27250, giga_loss[loss=0.3722, simple_loss=0.422, pruned_loss=0.1612, over 28695.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3882, pruned_loss=0.1323, over 5653028.21 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3746, pruned_loss=0.1267, over 5699918.82 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3897, pruned_loss=0.1331, over 5652622.56 frames. ], batch size: 242, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:25:37,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6440, 5.4219, 5.1344, 2.6074], device='cuda:1'), covar=tensor([0.0451, 0.0684, 0.0757, 0.1789], device='cuda:1'), in_proj_covar=tensor([0.1044, 0.0987, 0.0861, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 08:25:46,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 08:26:05,096 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 27300, giga_loss[loss=0.2937, simple_loss=0.366, pruned_loss=0.1107, over 28626.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3878, pruned_loss=0.1313, over 5658855.48 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3747, pruned_loss=0.1268, over 5702144.22 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3891, pruned_loss=0.1319, over 5655970.42 frames. ], batch size: 92, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:27:19,189 INFO [train.py:968] (1/2) Epoch 10, batch 27350, giga_loss[loss=0.3089, simple_loss=0.3803, pruned_loss=0.1187, over 29014.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3881, pruned_loss=0.132, over 5666874.58 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3747, pruned_loss=0.1268, over 5705865.75 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3894, pruned_loss=0.1326, over 5660528.66 frames. ], batch size: 155, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:27:43,759 INFO [optim.py:369] (1/2) 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 08:28:05,610 INFO [train.py:968] (1/2) Epoch 10, batch 27400, giga_loss[loss=0.3336, simple_loss=0.3866, pruned_loss=0.1403, over 28932.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.386, pruned_loss=0.1308, over 5672383.37 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3748, pruned_loss=0.1268, over 5704130.53 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.387, pruned_loss=0.1313, over 5668413.92 frames. ], batch size: 145, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:28:28,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2248, 1.3841, 3.3494, 3.2350], device='cuda:1'), covar=tensor([0.1332, 0.2236, 0.0434, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0644, 0.0580, 0.0844, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:28:43,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6037, 1.7009, 1.4768, 1.7867], device='cuda:1'), covar=tensor([0.1991, 0.1885, 0.1853, 0.1768], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.0952, 0.1132, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 08:28:48,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5460, 1.7037, 1.4873, 1.3195], device='cuda:1'), covar=tensor([0.1696, 0.1594, 0.1437, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.1663, 0.1573, 0.1557, 0.1657], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 08:28:57,085 INFO [train.py:968] (1/2) Epoch 10, batch 27450, giga_loss[loss=0.3698, simple_loss=0.4055, pruned_loss=0.1671, over 26609.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3838, pruned_loss=0.1313, over 5646520.93 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.375, pruned_loss=0.1269, over 5696652.75 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3846, pruned_loss=0.1317, over 5648656.93 frames. ], batch size: 555, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:29:26,626 INFO [optim.py:369] (1/2) 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 08:29:48,726 INFO [train.py:968] (1/2) Epoch 10, batch 27500, giga_loss[loss=0.2859, simple_loss=0.3581, pruned_loss=0.1068, over 28931.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.382, pruned_loss=0.1308, over 5644375.95 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3754, pruned_loss=0.1272, over 5700508.64 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3825, pruned_loss=0.1309, over 5641489.33 frames. ], batch size: 136, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:30:25,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7232, 2.1720, 1.5679, 1.8468], device='cuda:1'), covar=tensor([0.0677, 0.0234, 0.0282, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 08:30:27,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7436, 2.4843, 2.4390, 2.2367], device='cuda:1'), covar=tensor([0.1331, 0.2074, 0.1694, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0746, 0.0679, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 08:30:35,237 INFO [train.py:968] (1/2) Epoch 10, batch 27550, giga_loss[loss=0.2857, simple_loss=0.3513, pruned_loss=0.11, over 28927.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3817, pruned_loss=0.1316, over 5637267.25 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3758, pruned_loss=0.1275, over 5682674.23 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3818, pruned_loss=0.1315, over 5648706.76 frames. ], batch size: 145, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:31:00,142 INFO [optim.py:369] (1/2) 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 08:31:18,426 INFO [train.py:968] (1/2) Epoch 10, batch 27600, libri_loss[loss=0.3276, simple_loss=0.392, pruned_loss=0.1316, over 29643.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3815, pruned_loss=0.132, over 5636383.33 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3758, pruned_loss=0.1275, over 5679192.79 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3818, pruned_loss=0.1322, over 5646260.38 frames. ], batch size: 91, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:31:58,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:32:00,270 INFO [train.py:968] (1/2) Epoch 10, batch 27650, giga_loss[loss=0.3005, simple_loss=0.3776, pruned_loss=0.1117, over 28730.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1296, over 5643166.98 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3762, pruned_loss=0.1277, over 5682145.99 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3789, pruned_loss=0.1296, over 5647473.31 frames. ], batch size: 242, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:32:19,865 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:32:47,443 INFO [train.py:968] (1/2) Epoch 10, batch 27700, giga_loss[loss=0.2929, simple_loss=0.3621, pruned_loss=0.1118, over 28504.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3749, pruned_loss=0.1251, over 5656719.57 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3761, pruned_loss=0.1276, over 5686765.87 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.375, pruned_loss=0.1251, over 5655596.21 frames. ], batch size: 71, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:33:34,949 INFO [train.py:968] (1/2) Epoch 10, batch 27750, giga_loss[loss=0.3172, simple_loss=0.3831, pruned_loss=0.1256, over 28715.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3743, pruned_loss=0.1241, over 5660170.71 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3761, pruned_loss=0.1275, over 5693356.61 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3743, pruned_loss=0.1242, over 5652747.69 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:34:06,773 INFO [optim.py:369] (1/2) 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 08:34:27,660 INFO [train.py:968] (1/2) Epoch 10, batch 27800, giga_loss[loss=0.2951, simple_loss=0.3556, pruned_loss=0.1173, over 28911.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3726, pruned_loss=0.1237, over 5649353.28 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3757, pruned_loss=0.1273, over 5696502.01 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3729, pruned_loss=0.1238, over 5640200.20 frames. ], batch size: 199, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:34:36,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1212, 3.4387, 1.2918, 1.5019], device='cuda:1'), covar=tensor([0.1214, 0.0382, 0.0986, 0.1501], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0512, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 08:35:17,587 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:35:26,097 INFO [train.py:968] (1/2) Epoch 10, batch 27850, giga_loss[loss=0.3261, simple_loss=0.3702, pruned_loss=0.141, over 23476.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3681, pruned_loss=0.1211, over 5664971.85 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1273, over 5696704.89 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1213, over 5657278.99 frames. ], batch size: 705, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:35:56,763 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 27900, giga_loss[loss=0.2759, simple_loss=0.3433, pruned_loss=0.1043, over 28942.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3717, pruned_loss=0.1241, over 5653555.93 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5681520.41 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1234, over 5660458.27 frames. ], batch size: 106, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:37:03,576 INFO [train.py:968] (1/2) Epoch 10, batch 27950, libri_loss[loss=0.2642, simple_loss=0.3307, pruned_loss=0.09885, over 29640.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1244, over 5648229.39 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3758, pruned_loss=0.1276, over 5685426.02 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3729, pruned_loss=0.1241, over 5649487.14 frames. ], batch size: 73, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:37:31,819 INFO [optim.py:369] (1/2) 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 08:37:54,536 INFO [train.py:968] (1/2) Epoch 10, batch 28000, giga_loss[loss=0.3445, simple_loss=0.4013, pruned_loss=0.1438, over 28904.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5649568.73 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3758, pruned_loss=0.1276, over 5686648.89 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3726, pruned_loss=0.1236, over 5649414.69 frames. ], batch size: 186, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:38:15,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1820, 1.3035, 3.9645, 3.1583], device='cuda:1'), covar=tensor([0.1732, 0.2584, 0.0396, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0650, 0.0585, 0.0850, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:38:15,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=437616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:38:37,533 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 28050, giga_loss[loss=0.4052, simple_loss=0.4303, pruned_loss=0.19, over 26601.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3749, pruned_loss=0.1257, over 5654908.42 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3762, pruned_loss=0.1278, over 5694171.75 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3744, pruned_loss=0.1253, over 5646517.64 frames. ], batch size: 555, lr: 3.22e-03, grad_scale: 8.0
+2023-03-05 08:39:06,566 INFO [optim.py:369] (1/2) 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 08:39:11,894 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:39:22,849 INFO [train.py:968] (1/2) Epoch 10, batch 28100, giga_loss[loss=0.2784, simple_loss=0.3488, pruned_loss=0.104, over 28728.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.1271, over 5662058.31 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5696845.79 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.375, pruned_loss=0.1261, over 5651243.57 frames. ], batch size: 99, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:39:32,389 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:11,400 INFO [train.py:968] (1/2) Epoch 10, batch 28150, giga_loss[loss=0.3557, simple_loss=0.4117, pruned_loss=0.1499, over 28748.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3784, pruned_loss=0.1282, over 5665548.89 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3772, pruned_loss=0.1283, over 5699135.87 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3774, pruned_loss=0.1276, over 5654472.80 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:40:25,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=437759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:29,038 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:40:45,147 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=437782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:49,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=437785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:54,017 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=437791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:40:56,958 INFO [train.py:968] (1/2) Epoch 10, batch 28200, giga_loss[loss=0.3103, simple_loss=0.3747, pruned_loss=0.1229, over 29000.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3808, pruned_loss=0.1296, over 5673194.96 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1283, over 5704831.18 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.38, pruned_loss=0.1291, over 5658288.53 frames. ], batch size: 128, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:41:14,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=437810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:41:15,356 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=437811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:41:18,510 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 28250, giga_loss[loss=0.2998, simple_loss=0.361, pruned_loss=0.1193, over 28290.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3823, pruned_loss=0.1314, over 5661382.99 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3777, pruned_loss=0.1286, over 5706666.85 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3813, pruned_loss=0.1307, over 5647379.27 frames. ], batch size: 77, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:42:15,866 INFO [optim.py:369] (1/2) 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 08:42:31,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-05 08:42:40,466 INFO [train.py:968] (1/2) Epoch 10, batch 28300, giga_loss[loss=0.331, simple_loss=0.4012, pruned_loss=0.1304, over 27655.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.382, pruned_loss=0.1316, over 5643486.62 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5691887.96 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3816, pruned_loss=0.1313, over 5644001.02 frames. ], batch size: 472, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:43:30,438 INFO [train.py:968] (1/2) Epoch 10, batch 28350, giga_loss[loss=0.3075, simple_loss=0.3788, pruned_loss=0.1181, over 28935.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3839, pruned_loss=0.1312, over 5655061.31 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5697859.83 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.383, pruned_loss=0.1304, over 5648922.92 frames. ], batch size: 213, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:43:39,147 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=437953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:43:41,161 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=437956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:43:43,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2220, 1.4171, 3.4268, 3.1824], device='cuda:1'), covar=tensor([0.1374, 0.2298, 0.0498, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0650, 0.0584, 0.0853, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:44:00,609 INFO [optim.py:369] (1/2) 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 08:44:10,033 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=437985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:44:16,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-05 08:44:19,704 INFO [train.py:968] (1/2) Epoch 10, batch 28400, giga_loss[loss=0.3197, simple_loss=0.3795, pruned_loss=0.1299, over 28723.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3847, pruned_loss=0.1318, over 5669493.54 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3783, pruned_loss=0.1291, over 5702125.60 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.384, pruned_loss=0.1312, over 5659938.86 frames. ], batch size: 262, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:44:52,338 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:45:07,504 INFO [train.py:968] (1/2) Epoch 10, batch 28450, giga_loss[loss=0.2797, simple_loss=0.3549, pruned_loss=0.1023, over 29025.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3821, pruned_loss=0.1309, over 5663944.43 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3776, pruned_loss=0.1286, over 5694660.09 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3823, pruned_loss=0.131, over 5661552.31 frames. ], batch size: 155, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:45:18,169 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:45:40,884 INFO [optim.py:369] (1/2) 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 08:45:45,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:45:52,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8854, 2.0744, 2.1036, 1.6872], device='cuda:1'), covar=tensor([0.1631, 0.2029, 0.1258, 0.1457], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0709, 0.0856, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 08:46:03,507 INFO [train.py:968] (1/2) Epoch 10, batch 28500, giga_loss[loss=0.2866, simple_loss=0.3617, pruned_loss=0.1058, over 28913.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3801, pruned_loss=0.1298, over 5666938.70 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3771, pruned_loss=0.1284, over 5691567.85 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3808, pruned_loss=0.1301, over 5666621.97 frames. ], batch size: 174, lr: 3.22e-03, grad_scale: 4.0
+2023-03-05 08:46:17,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6628, 4.4410, 4.2353, 1.9836], device='cuda:1'), covar=tensor([0.0429, 0.0655, 0.0662, 0.2232], device='cuda:1'), in_proj_covar=tensor([0.1048, 0.0991, 0.0870, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 08:46:27,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:46:46,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 08:46:58,365 INFO [train.py:968] (1/2) Epoch 10, batch 28550, giga_loss[loss=0.3415, simple_loss=0.3924, pruned_loss=0.1453, over 27564.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3779, pruned_loss=0.1283, over 5671154.85 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3775, pruned_loss=0.1286, over 5692210.94 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3781, pruned_loss=0.1284, over 5669606.07 frames. ], batch size: 472, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:47:02,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438148.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 08:47:05,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:47:30,017 INFO [optim.py:369] (1/2) 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 08:47:38,638 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 28600, giga_loss[loss=0.2808, simple_loss=0.343, pruned_loss=0.1093, over 28810.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3785, pruned_loss=0.1296, over 5666281.38 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3775, pruned_loss=0.1286, over 5687209.35 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1296, over 5669529.75 frames. ], batch size: 99, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:47:47,891 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-05 08:47:48,952 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:47:51,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:12,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:15,838 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:18,867 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:18,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4038, 1.5622, 1.4695, 1.4520], device='cuda:1'), covar=tensor([0.1312, 0.1749, 0.1815, 0.1641], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0739, 0.0669, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 08:48:33,133 INFO [train.py:968] (1/2) Epoch 10, batch 28650, libri_loss[loss=0.2827, simple_loss=0.355, pruned_loss=0.1052, over 29527.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3775, pruned_loss=0.1292, over 5660858.59 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3769, pruned_loss=0.1281, over 5693693.64 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3783, pruned_loss=0.1297, over 5656927.79 frames. ], batch size: 81, lr: 3.22e-03, grad_scale: 2.0
+2023-03-05 08:48:42,010 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:48:47,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-03-05 08:48:59,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3271, 1.5728, 1.4261, 1.3092], device='cuda:1'), covar=tensor([0.1514, 0.1722, 0.1774, 0.1792], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0738, 0.0668, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 08:49:02,418 INFO [optim.py:369] (1/2) 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 08:49:20,901 INFO [train.py:968] (1/2) Epoch 10, batch 28700, libri_loss[loss=0.3035, simple_loss=0.3748, pruned_loss=0.1161, over 29756.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3786, pruned_loss=0.1303, over 5661578.52 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3759, pruned_loss=0.1274, over 5700200.63 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3801, pruned_loss=0.1314, over 5651808.74 frames. ], batch size: 87, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:49:55,963 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:49:58,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:50:11,181 INFO [train.py:968] (1/2) Epoch 10, batch 28750, giga_loss[loss=0.301, simple_loss=0.3714, pruned_loss=0.1153, over 28888.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3807, pruned_loss=0.1324, over 5655010.01 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5701163.62 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3822, pruned_loss=0.1335, over 5646387.66 frames. ], batch size: 145, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:50:17,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5180, 3.5249, 1.6309, 1.5515], device='cuda:1'), covar=tensor([0.0883, 0.0284, 0.0814, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0511, 0.0335, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 08:50:21,001 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 08:50:25,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:50:40,745 INFO [optim.py:369] (1/2) 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 08:50:57,332 INFO [train.py:968] (1/2) Epoch 10, batch 28800, giga_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 28687.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3808, pruned_loss=0.1321, over 5646719.35 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1272, over 5694905.27 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3821, pruned_loss=0.1331, over 5643813.81 frames. ], batch size: 242, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:51:05,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:51:09,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4767, 2.6929, 1.5683, 1.6180], device='cuda:1'), covar=tensor([0.0714, 0.0298, 0.0657, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0511, 0.0334, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 08:51:41,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5940, 1.9071, 1.5044, 1.6448], device='cuda:1'), covar=tensor([0.2230, 0.2155, 0.2429, 0.2153], device='cuda:1'), in_proj_covar=tensor([0.1274, 0.0945, 0.1130, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 08:51:43,454 INFO [train.py:968] (1/2) Epoch 10, batch 28850, libri_loss[loss=0.254, simple_loss=0.3188, pruned_loss=0.0946, over 29346.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3812, pruned_loss=0.1336, over 5649471.49 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3748, pruned_loss=0.1267, over 5698772.15 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.383, pruned_loss=0.1349, over 5642654.39 frames. ], batch size: 67, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:51:55,665 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:52:13,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1828, 1.5234, 1.2085, 0.9938], device='cuda:1'), covar=tensor([0.2222, 0.2176, 0.2376, 0.1895], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.0948, 0.1133, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 08:52:15,928 INFO [optim.py:369] (1/2) 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 08:52:24,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:52:29,735 INFO [train.py:968] (1/2) Epoch 10, batch 28900, giga_loss[loss=0.3283, simple_loss=0.3876, pruned_loss=0.1345, over 28666.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3825, pruned_loss=0.1347, over 5656037.49 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3752, pruned_loss=0.127, over 5701263.59 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3838, pruned_loss=0.1357, over 5647550.28 frames. ], batch size: 262, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:52:55,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3954, 4.2007, 3.9897, 2.0510], device='cuda:1'), covar=tensor([0.0471, 0.0635, 0.0645, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.1043, 0.0985, 0.0861, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 08:52:55,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 08:52:57,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:02,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2606, 3.0722, 2.9336, 1.4221], device='cuda:1'), covar=tensor([0.0930, 0.1028, 0.0924, 0.2331], device='cuda:1'), in_proj_covar=tensor([0.1043, 0.0985, 0.0861, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 08:53:14,155 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:14,568 INFO [train.py:968] (1/2) Epoch 10, batch 28950, giga_loss[loss=0.3219, simple_loss=0.3872, pruned_loss=0.1283, over 29013.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3831, pruned_loss=0.135, over 5635523.60 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3754, pruned_loss=0.1273, over 5685970.08 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3841, pruned_loss=0.1357, over 5640262.82 frames. ], batch size: 155, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:53:15,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:19,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:48,405 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:53:48,841 INFO [optim.py:369] (1/2) 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 08:54:06,782 INFO [train.py:968] (1/2) Epoch 10, batch 29000, libri_loss[loss=0.3104, simple_loss=0.3699, pruned_loss=0.1255, over 29547.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3832, pruned_loss=0.1344, over 5643195.56 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.375, pruned_loss=0.1271, over 5688403.23 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3846, pruned_loss=0.1353, over 5643367.02 frames. ], batch size: 77, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:54:16,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 08:54:41,071 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:54:44,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:54:54,716 INFO [train.py:968] (1/2) Epoch 10, batch 29050, giga_loss[loss=0.3103, simple_loss=0.3828, pruned_loss=0.119, over 28971.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3834, pruned_loss=0.1342, over 5647862.85 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3751, pruned_loss=0.1272, over 5689473.22 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3844, pruned_loss=0.1348, over 5646878.90 frames. ], batch size: 164, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:55:13,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:55:16,679 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 08:55:18,104 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=438668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:55:18,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438669.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 08:55:19,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:55:23,484 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 29100, giga_loss[loss=0.2892, simple_loss=0.3639, pruned_loss=0.1072, over 28993.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3845, pruned_loss=0.1351, over 5657233.73 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3756, pruned_loss=0.1275, over 5683822.48 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3852, pruned_loss=0.1355, over 5660937.37 frames. ], batch size: 145, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:55:42,100 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438698.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 08:55:43,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=438700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:56:22,684 INFO [train.py:968] (1/2) Epoch 10, batch 29150, libri_loss[loss=0.2478, simple_loss=0.3221, pruned_loss=0.08675, over 29377.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3861, pruned_loss=0.1366, over 5658236.77 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3757, pruned_loss=0.1277, over 5679483.18 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3867, pruned_loss=0.137, over 5664804.22 frames. ], batch size: 71, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 08:56:54,422 INFO [optim.py:369] (1/2) 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 08:57:11,835 INFO [train.py:968] (1/2) Epoch 10, batch 29200, giga_loss[loss=0.2985, simple_loss=0.3772, pruned_loss=0.1099, over 28913.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3858, pruned_loss=0.1358, over 5658047.98 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3753, pruned_loss=0.1276, over 5680034.58 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3869, pruned_loss=0.1364, over 5662120.00 frames. ], batch size: 145, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:57:50,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:58:01,501 INFO [train.py:968] (1/2) Epoch 10, batch 29250, giga_loss[loss=0.3355, simple_loss=0.3917, pruned_loss=0.1397, over 28971.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3856, pruned_loss=0.1341, over 5662702.75 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3757, pruned_loss=0.1278, over 5683841.48 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3865, pruned_loss=0.1346, over 5662061.87 frames. ], batch size: 213, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:58:31,635 INFO [optim.py:369] (1/2) 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 08:58:34,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 08:58:45,861 INFO [train.py:968] (1/2) Epoch 10, batch 29300, giga_loss[loss=0.2727, simple_loss=0.3471, pruned_loss=0.09918, over 28508.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3828, pruned_loss=0.1312, over 5666455.97 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3758, pruned_loss=0.1279, over 5686204.87 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3834, pruned_loss=0.1315, over 5663664.26 frames. ], batch size: 78, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:59:10,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=438918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 08:59:12,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6497, 1.6306, 1.2489, 1.2463], device='cuda:1'), covar=tensor([0.0715, 0.0526, 0.0946, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0445, 0.0500, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 08:59:16,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 08:59:31,440 INFO [train.py:968] (1/2) Epoch 10, batch 29350, giga_loss[loss=0.4905, simple_loss=0.4856, pruned_loss=0.2477, over 26613.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3819, pruned_loss=0.131, over 5660375.53 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1282, over 5685766.52 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3823, pruned_loss=0.1311, over 5658246.56 frames. ], batch size: 555, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 08:59:47,075 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=438959.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 08:59:59,074 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:00:02,093 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=438978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:00:19,635 INFO [train.py:968] (1/2) Epoch 10, batch 29400, giga_loss[loss=0.2861, simple_loss=0.3594, pruned_loss=0.1064, over 28936.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3843, pruned_loss=0.1329, over 5656707.74 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.376, pruned_loss=0.128, over 5683039.60 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3849, pruned_loss=0.1333, over 5656517.91 frames. ], batch size: 186, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:00:28,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4359, 1.5770, 1.4259, 1.5634], device='cuda:1'), covar=tensor([0.0740, 0.0313, 0.0298, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 09:00:31,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:01:10,716 INFO [train.py:968] (1/2) Epoch 10, batch 29450, giga_loss[loss=0.3238, simple_loss=0.3858, pruned_loss=0.1309, over 28843.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3855, pruned_loss=0.1339, over 5656839.93 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1281, over 5686414.83 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3859, pruned_loss=0.1341, over 5653631.18 frames. ], batch size: 186, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:01:28,116 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=439061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:01:31,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=439064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:01:31,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8214, 2.1119, 1.8290, 1.6336], device='cuda:1'), covar=tensor([0.1439, 0.1208, 0.1057, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.1664, 0.1592, 0.1547, 0.1647], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 09:01:38,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-05 09:01:40,822 INFO [optim.py:369] (1/2) 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 09:01:41,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1185, 1.4753, 1.4028, 1.0214], device='cuda:1'), covar=tensor([0.1311, 0.2242, 0.1192, 0.1412], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0707, 0.0856, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 09:01:56,883 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:01:57,272 INFO [train.py:968] (1/2) Epoch 10, batch 29500, giga_loss[loss=0.4081, simple_loss=0.4265, pruned_loss=0.1948, over 26691.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3847, pruned_loss=0.1338, over 5663947.04 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3766, pruned_loss=0.1282, over 5690816.32 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3849, pruned_loss=0.1341, over 5656622.99 frames. ], batch size: 555, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:02:44,037 INFO [train.py:968] (1/2) Epoch 10, batch 29550, giga_loss[loss=0.3336, simple_loss=0.3924, pruned_loss=0.1374, over 28330.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3859, pruned_loss=0.1354, over 5658294.99 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3771, pruned_loss=0.1285, over 5694475.74 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3858, pruned_loss=0.1355, over 5648687.27 frames. ], batch size: 368, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:03:13,247 INFO [optim.py:369] (1/2) 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 09:03:29,740 INFO [train.py:968] (1/2) Epoch 10, batch 29600, giga_loss[loss=0.3116, simple_loss=0.378, pruned_loss=0.1226, over 28639.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3859, pruned_loss=0.1356, over 5666350.58 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3769, pruned_loss=0.1283, over 5697082.48 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3862, pruned_loss=0.136, over 5655144.09 frames. ], batch size: 307, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:03:35,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2162, 1.2794, 4.2724, 3.2203], device='cuda:1'), covar=tensor([0.1751, 0.2596, 0.0377, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0655, 0.0582, 0.0856, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 09:04:18,653 INFO [train.py:968] (1/2) Epoch 10, batch 29650, giga_loss[loss=0.2807, simple_loss=0.347, pruned_loss=0.1072, over 28729.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3851, pruned_loss=0.1348, over 5662188.47 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5699950.26 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3857, pruned_loss=0.1354, over 5650465.43 frames. ], batch size: 99, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:04:20,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0579, 2.4285, 2.0889, 2.4266], device='cuda:1'), covar=tensor([0.1759, 0.1578, 0.1782, 0.1448], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.0949, 0.1136, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 09:04:51,050 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 29700, giga_loss[loss=0.284, simple_loss=0.354, pruned_loss=0.107, over 28943.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3844, pruned_loss=0.1338, over 5660609.66 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3769, pruned_loss=0.1283, over 5687339.62 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3847, pruned_loss=0.1341, over 5661110.04 frames. ], batch size: 106, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:05:49,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=439334.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:05:56,217 INFO [train.py:968] (1/2) Epoch 10, batch 29750, giga_loss[loss=0.3391, simple_loss=0.3937, pruned_loss=0.1422, over 27569.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.383, pruned_loss=0.1321, over 5654758.16 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.377, pruned_loss=0.1284, over 5678378.25 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3833, pruned_loss=0.1324, over 5662282.95 frames. ], batch size: 472, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:06:26,802 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 29800, giga_loss[loss=0.3672, simple_loss=0.4217, pruned_loss=0.1564, over 28856.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3827, pruned_loss=0.1317, over 5647801.00 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.377, pruned_loss=0.1284, over 5675817.75 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3832, pruned_loss=0.1321, over 5655127.80 frames. ], batch size: 186, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:06:46,117 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=439400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:07:23,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3443, 1.7582, 1.3808, 1.3703], device='cuda:1'), covar=tensor([0.2370, 0.2271, 0.2661, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.1272, 0.0946, 0.1132, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 09:07:29,475 INFO [train.py:968] (1/2) Epoch 10, batch 29850, giga_loss[loss=0.2799, simple_loss=0.3457, pruned_loss=0.1071, over 28894.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3811, pruned_loss=0.1307, over 5654972.01 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.377, pruned_loss=0.1284, over 5681158.06 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3817, pruned_loss=0.1311, over 5655551.83 frames. ], batch size: 186, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:07:47,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4909, 1.7935, 1.4982, 1.6337], device='cuda:1'), covar=tensor([0.0608, 0.0247, 0.0258, 0.0604], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 09:08:01,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=439477.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:08:01,505 INFO [optim.py:369] (1/2) 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 09:08:04,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=439480.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:08:16,244 INFO [train.py:968] (1/2) Epoch 10, batch 29900, giga_loss[loss=0.2819, simple_loss=0.3564, pruned_loss=0.1037, over 28957.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3804, pruned_loss=0.1305, over 5664236.66 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3773, pruned_loss=0.1286, over 5681073.54 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3805, pruned_loss=0.1306, over 5664257.96 frames. ], batch size: 145, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:08:29,921 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439509.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:09:01,050 INFO [train.py:968] (1/2) Epoch 10, batch 29950, giga_loss[loss=0.293, simple_loss=0.3573, pruned_loss=0.1144, over 28761.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3794, pruned_loss=0.1306, over 5660357.11 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3773, pruned_loss=0.1286, over 5685109.66 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3795, pruned_loss=0.1308, over 5656311.27 frames. ], batch size: 284, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:09:11,325 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-05 09:09:30,014 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 09:09:37,700 INFO [optim.py:369] (1/2) 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 09:09:46,904 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 09:09:50,993 INFO [train.py:968] (1/2) Epoch 10, batch 30000, giga_loss[loss=0.3833, simple_loss=0.4276, pruned_loss=0.1695, over 28441.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3755, pruned_loss=0.1282, over 5672695.32 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3774, pruned_loss=0.1286, over 5689314.82 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3756, pruned_loss=0.1284, over 5665428.01 frames. ], batch size: 369, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:09:50,994 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 09:09:59,369 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 09:10:47,251 INFO [train.py:968] (1/2) Epoch 10, batch 30050, giga_loss[loss=0.2702, simple_loss=0.3407, pruned_loss=0.09984, over 28992.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3749, pruned_loss=0.1286, over 5687493.56 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3778, pruned_loss=0.1288, over 5692308.82 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3746, pruned_loss=0.1286, over 5679003.24 frames. ], batch size: 106, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:10:51,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5699, 1.6998, 1.8376, 1.4203], device='cuda:1'), covar=tensor([0.1577, 0.2167, 0.1253, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0707, 0.0855, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 09:11:20,388 INFO [optim.py:369] (1/2) 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 09:11:28,977 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=439686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:11:35,872 INFO [train.py:968] (1/2) Epoch 10, batch 30100, giga_loss[loss=0.3101, simple_loss=0.3768, pruned_loss=0.1217, over 28277.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3737, pruned_loss=0.1277, over 5695620.62 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5694728.30 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3731, pruned_loss=0.1274, over 5686711.29 frames. ], batch size: 368, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:12:25,529 INFO [train.py:968] (1/2) Epoch 10, batch 30150, giga_loss[loss=0.2785, simple_loss=0.3529, pruned_loss=0.102, over 28006.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3731, pruned_loss=0.1257, over 5688438.57 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3781, pruned_loss=0.1289, over 5697600.65 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3726, pruned_loss=0.1256, over 5678913.36 frames. ], batch size: 412, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:12:58,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=439775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:13:02,461 INFO [optim.py:369] (1/2) 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 09:13:21,495 INFO [train.py:968] (1/2) Epoch 10, batch 30200, giga_loss[loss=0.3273, simple_loss=0.3759, pruned_loss=0.1394, over 26667.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3707, pruned_loss=0.1223, over 5684139.50 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3782, pruned_loss=0.1292, over 5701695.39 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1218, over 5672733.10 frames. ], batch size: 555, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:14:10,896 INFO [train.py:968] (1/2) Epoch 10, batch 30250, giga_loss[loss=0.2677, simple_loss=0.3536, pruned_loss=0.0909, over 28824.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5678148.62 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3778, pruned_loss=0.1292, over 5705090.11 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3672, pruned_loss=0.1186, over 5665235.27 frames. ], batch size: 174, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:14:40,485 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:14:58,615 INFO [train.py:968] (1/2) Epoch 10, batch 30300, giga_loss[loss=0.2444, simple_loss=0.3282, pruned_loss=0.08027, over 28070.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3635, pruned_loss=0.1161, over 5659867.57 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3767, pruned_loss=0.1289, over 5699102.23 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3637, pruned_loss=0.1154, over 5652914.46 frames. ], batch size: 77, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:15:16,784 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=439918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:15:19,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9950, 2.2077, 1.9451, 1.9485], device='cuda:1'), covar=tensor([0.1312, 0.1723, 0.1629, 0.1664], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0732, 0.0665, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 09:15:19,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=439921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:15:41,899 INFO [train.py:968] (1/2) Epoch 10, batch 30350, giga_loss[loss=0.2607, simple_loss=0.3466, pruned_loss=0.08737, over 28756.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3607, pruned_loss=0.1135, over 5664749.51 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.376, pruned_loss=0.1288, over 5705576.79 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3608, pruned_loss=0.1125, over 5651752.37 frames. ], batch size: 243, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:15:47,719 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:16:13,506 INFO [optim.py:369] (1/2) 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 09:16:21,383 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=439988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:16:28,432 INFO [train.py:968] (1/2) Epoch 10, batch 30400, giga_loss[loss=0.2723, simple_loss=0.3548, pruned_loss=0.09495, over 28759.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3595, pruned_loss=0.1112, over 5650401.16 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3758, pruned_loss=0.1287, over 5701384.31 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3594, pruned_loss=0.1099, over 5642312.19 frames. ], batch size: 284, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:16:36,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-05 09:16:38,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 09:17:08,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 09:17:24,161 INFO [train.py:968] (1/2) Epoch 10, batch 30450, giga_loss[loss=0.2866, simple_loss=0.3591, pruned_loss=0.107, over 28725.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3586, pruned_loss=0.1096, over 5647393.42 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3755, pruned_loss=0.1286, over 5702238.39 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3587, pruned_loss=0.1086, over 5640086.50 frames. ], batch size: 284, lr: 3.21e-03, grad_scale: 8.0
+2023-03-05 09:17:32,674 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=440052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:17:42,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:17:49,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=440068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:17:59,526 INFO [optim.py:369] (1/2) 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 09:18:13,213 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=440092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:18:14,388 INFO [train.py:968] (1/2) Epoch 10, batch 30500, giga_loss[loss=0.2553, simple_loss=0.3384, pruned_loss=0.08614, over 28998.00 frames. ], tot_loss[loss=0.287, simple_loss=0.357, pruned_loss=0.1085, over 5637847.16 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3746, pruned_loss=0.1281, over 5696560.15 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3574, pruned_loss=0.1076, over 5635509.19 frames. ], batch size: 155, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:18:48,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3982, 1.2542, 3.9691, 3.2724], device='cuda:1'), covar=tensor([0.1521, 0.2616, 0.0382, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0648, 0.0579, 0.0849, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 09:19:02,185 INFO [train.py:968] (1/2) Epoch 10, batch 30550, giga_loss[loss=0.2871, simple_loss=0.3576, pruned_loss=0.1083, over 28495.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3546, pruned_loss=0.1072, over 5626706.65 frames. ], libri_tot_loss[loss=0.3159, simple_loss=0.3746, pruned_loss=0.1285, over 5684612.86 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3543, pruned_loss=0.1053, over 5632866.84 frames. ], batch size: 307, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:19:39,929 INFO [optim.py:369] (1/2) 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 09:19:55,730 INFO [train.py:968] (1/2) Epoch 10, batch 30600, giga_loss[loss=0.2569, simple_loss=0.3366, pruned_loss=0.08859, over 28851.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3522, pruned_loss=0.1054, over 5628587.97 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3744, pruned_loss=0.1284, over 5683541.97 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3519, pruned_loss=0.1038, over 5633872.76 frames. ], batch size: 112, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:20:04,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:20:07,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:20:35,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:20:44,267 INFO [train.py:968] (1/2) Epoch 10, batch 30650, giga_loss[loss=0.2419, simple_loss=0.3259, pruned_loss=0.07899, over 28750.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3523, pruned_loss=0.1048, over 5637619.43 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3742, pruned_loss=0.1283, over 5685970.04 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3521, pruned_loss=0.1034, over 5639351.56 frames. ], batch size: 119, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:20:48,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:21:20,471 INFO [optim.py:369] (1/2) 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 09:21:33,144 INFO [train.py:968] (1/2) Epoch 10, batch 30700, giga_loss[loss=0.2276, simple_loss=0.3119, pruned_loss=0.07168, over 28766.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3508, pruned_loss=0.1029, over 5634410.33 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3741, pruned_loss=0.1284, over 5678737.48 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3502, pruned_loss=0.1013, over 5641698.22 frames. ], batch size: 119, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:21:41,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6382, 2.3494, 1.7665, 1.5298], device='cuda:1'), covar=tensor([0.2019, 0.1202, 0.1520, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.1634, 0.1537, 0.1498, 0.1610], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 09:22:25,419 INFO [train.py:968] (1/2) Epoch 10, batch 30750, giga_loss[loss=0.2936, simple_loss=0.3544, pruned_loss=0.1164, over 27575.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.348, pruned_loss=0.1004, over 5643517.85 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3738, pruned_loss=0.1282, over 5683105.64 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3473, pruned_loss=0.09877, over 5644658.65 frames. ], batch size: 472, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:22:44,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:22:47,353 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.52 vs. limit=5.0
+2023-03-05 09:23:02,307 INFO [optim.py:369] (1/2) 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 09:23:04,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3981, 1.7064, 1.3911, 1.6279], device='cuda:1'), covar=tensor([0.2234, 0.1900, 0.2101, 0.1815], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.0940, 0.1133, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 09:23:14,166 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 30800, giga_loss[loss=0.2302, simple_loss=0.3135, pruned_loss=0.07339, over 28643.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3442, pruned_loss=0.09854, over 5640985.79 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3735, pruned_loss=0.1281, over 5688788.54 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3434, pruned_loss=0.0967, over 5636064.36 frames. ], batch size: 307, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:23:17,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:23:47,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:23:49,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:24:05,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:24:05,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4477, 2.2661, 1.5992, 0.5901], device='cuda:1'), covar=tensor([0.3211, 0.1839, 0.2696, 0.3427], device='cuda:1'), in_proj_covar=tensor([0.1539, 0.1461, 0.1467, 0.1264], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 09:24:05,715 INFO [train.py:968] (1/2) Epoch 10, batch 30850, giga_loss[loss=0.2684, simple_loss=0.3396, pruned_loss=0.09861, over 28844.00 frames. ], tot_loss[loss=0.27, simple_loss=0.343, pruned_loss=0.09849, over 5650753.78 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3729, pruned_loss=0.128, over 5694395.42 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3422, pruned_loss=0.09641, over 5640846.10 frames. ], batch size: 112, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:24:31,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=440467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:24:44,476 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 30900, giga_loss[loss=0.2384, simple_loss=0.3225, pruned_loss=0.07708, over 28896.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3416, pruned_loss=0.09834, over 5645065.28 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3725, pruned_loss=0.1279, over 5697745.11 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3406, pruned_loss=0.09615, over 5632884.08 frames. ], batch size: 145, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:25:11,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:25:14,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:25:45,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:25:50,762 INFO [train.py:968] (1/2) Epoch 10, batch 30950, giga_loss[loss=0.2681, simple_loss=0.3502, pruned_loss=0.09298, over 28673.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3429, pruned_loss=0.09886, over 5636212.92 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3726, pruned_loss=0.1279, over 5701178.81 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3414, pruned_loss=0.09651, over 5622392.12 frames. ], batch size: 242, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:26:26,072 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:29,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:32,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 09:26:37,327 INFO [optim.py:369] (1/2) 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 09:26:41,656 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:43,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:26:47,967 INFO [train.py:968] (1/2) Epoch 10, batch 31000, giga_loss[loss=0.2762, simple_loss=0.3617, pruned_loss=0.09538, over 28640.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3455, pruned_loss=0.09919, over 5634234.91 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3722, pruned_loss=0.1279, over 5691736.15 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3441, pruned_loss=0.09674, over 5630795.59 frames. ], batch size: 307, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:26:59,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:09,208 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=440610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:11,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=440613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:17,713 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:46,332 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=440642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:27:47,965 INFO [train.py:968] (1/2) Epoch 10, batch 31050, giga_loss[loss=0.2702, simple_loss=0.3498, pruned_loss=0.0953, over 28974.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3468, pruned_loss=0.09989, over 5646333.19 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3713, pruned_loss=0.1274, over 5688044.89 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3457, pruned_loss=0.0975, over 5644980.05 frames. ], batch size: 164, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:27:58,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3125, 1.5331, 1.3864, 1.2731], device='cuda:1'), covar=tensor([0.1721, 0.1435, 0.1169, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1529, 0.1491, 0.1601], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 09:28:38,019 INFO [optim.py:369] (1/2) 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 09:28:54,360 INFO [train.py:968] (1/2) Epoch 10, batch 31100, giga_loss[loss=0.2591, simple_loss=0.3364, pruned_loss=0.09092, over 29004.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3457, pruned_loss=0.09897, over 5657822.19 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.371, pruned_loss=0.1271, over 5690645.23 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3447, pruned_loss=0.09682, over 5653949.54 frames. ], batch size: 213, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:29:59,144 INFO [train.py:968] (1/2) Epoch 10, batch 31150, giga_loss[loss=0.2887, simple_loss=0.3702, pruned_loss=0.1036, over 28584.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3437, pruned_loss=0.09686, over 5654722.73 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.371, pruned_loss=0.1271, over 5691544.01 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3427, pruned_loss=0.09504, over 5650794.30 frames. ], batch size: 307, lr: 3.21e-03, grad_scale: 2.0
+2023-03-05 09:30:34,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5717, 2.7798, 1.6011, 1.6881], device='cuda:1'), covar=tensor([0.0691, 0.0271, 0.0762, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0507, 0.0336, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 09:30:49,040 INFO [optim.py:369] (1/2) 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 09:31:03,181 INFO [train.py:968] (1/2) Epoch 10, batch 31200, giga_loss[loss=0.2315, simple_loss=0.3122, pruned_loss=0.07541, over 27735.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3416, pruned_loss=0.09437, over 5661705.73 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3707, pruned_loss=0.1271, over 5694879.99 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3408, pruned_loss=0.0926, over 5655367.40 frames. ], batch size: 472, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:31:11,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 09:32:04,117 INFO [train.py:968] (1/2) Epoch 10, batch 31250, giga_loss[loss=0.235, simple_loss=0.3093, pruned_loss=0.08032, over 28779.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3399, pruned_loss=0.09487, over 5659729.68 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3709, pruned_loss=0.1273, over 5688700.86 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3382, pruned_loss=0.09238, over 5659635.06 frames. ], batch size: 243, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:32:46,118 INFO [optim.py:369] (1/2) 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 09:32:59,476 INFO [train.py:968] (1/2) Epoch 10, batch 31300, giga_loss[loss=0.2968, simple_loss=0.3633, pruned_loss=0.1152, over 28560.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3396, pruned_loss=0.09527, over 5648584.23 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3705, pruned_loss=0.1272, over 5676376.05 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3377, pruned_loss=0.09238, over 5658456.56 frames. ], batch size: 307, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:33:49,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-05 09:33:54,288 INFO [train.py:968] (1/2) Epoch 10, batch 31350, giga_loss[loss=0.277, simple_loss=0.3575, pruned_loss=0.0982, over 28952.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3397, pruned_loss=0.09559, over 5658821.06 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3704, pruned_loss=0.1272, over 5679923.58 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3374, pruned_loss=0.09238, over 5662833.71 frames. ], batch size: 213, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:34:40,687 INFO [optim.py:369] (1/2) 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 09:34:50,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1517, 1.6056, 1.4499, 1.0734], device='cuda:1'), covar=tensor([0.1361, 0.2168, 0.1190, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0692, 0.0848, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 09:34:55,747 INFO [train.py:968] (1/2) Epoch 10, batch 31400, giga_loss[loss=0.2799, simple_loss=0.3646, pruned_loss=0.09757, over 28902.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09587, over 5657009.36 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.37, pruned_loss=0.1271, over 5679924.88 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3393, pruned_loss=0.09282, over 5659696.99 frames. ], batch size: 164, lr: 3.21e-03, grad_scale: 4.0
+2023-03-05 09:35:58,614 INFO [train.py:968] (1/2) Epoch 10, batch 31450, libri_loss[loss=0.2788, simple_loss=0.3477, pruned_loss=0.1049, over 27587.00 frames. ], tot_loss[loss=0.267, simple_loss=0.342, pruned_loss=0.09595, over 5648197.78 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3696, pruned_loss=0.1269, over 5671667.99 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3399, pruned_loss=0.09278, over 5656329.22 frames. ], batch size: 116, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:36:14,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2725, 3.1009, 2.9256, 1.3655], device='cuda:1'), covar=tensor([0.0800, 0.0903, 0.0866, 0.2346], device='cuda:1'), in_proj_covar=tensor([0.1005, 0.0944, 0.0828, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 09:36:47,689 INFO [optim.py:369] (1/2) 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 09:37:04,692 INFO [train.py:968] (1/2) Epoch 10, batch 31500, giga_loss[loss=0.2807, simple_loss=0.3569, pruned_loss=0.1023, over 28697.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.338, pruned_loss=0.09327, over 5653682.68 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3696, pruned_loss=0.1269, over 5671247.80 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.336, pruned_loss=0.09029, over 5660288.81 frames. ], batch size: 242, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:37:33,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5765, 1.7791, 1.6227, 1.5979], device='cuda:1'), covar=tensor([0.1200, 0.1918, 0.1508, 0.1590], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0712, 0.0651, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0008, 0.0008], device='cuda:1')
+2023-03-05 09:37:54,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-05 09:38:13,890 INFO [train.py:968] (1/2) Epoch 10, batch 31550, giga_loss[loss=0.2745, simple_loss=0.3525, pruned_loss=0.09823, over 28462.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3405, pruned_loss=0.09495, over 5663195.13 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3698, pruned_loss=0.127, over 5674901.48 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3384, pruned_loss=0.09204, over 5664974.06 frames. ], batch size: 336, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:39:01,494 INFO [optim.py:369] (1/2) 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 09:39:05,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2166, 1.6568, 1.2026, 0.5983], device='cuda:1'), covar=tensor([0.3974, 0.2071, 0.2883, 0.4389], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1469, 0.1479, 0.1272], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 09:39:16,624 INFO [train.py:968] (1/2) Epoch 10, batch 31600, giga_loss[loss=0.2674, simple_loss=0.353, pruned_loss=0.09093, over 27685.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3442, pruned_loss=0.09492, over 5654499.82 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.37, pruned_loss=0.1273, over 5677774.21 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3419, pruned_loss=0.09189, over 5652923.68 frames. ], batch size: 472, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:40:24,937 INFO [train.py:968] (1/2) Epoch 10, batch 31650, giga_loss[loss=0.2581, simple_loss=0.3481, pruned_loss=0.0841, over 28921.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3447, pruned_loss=0.09296, over 5657348.67 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3694, pruned_loss=0.1269, over 5681077.07 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3431, pruned_loss=0.09041, over 5652719.23 frames. ], batch size: 284, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:40:59,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-05 09:41:07,558 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 31700, giga_loss[loss=0.2426, simple_loss=0.3357, pruned_loss=0.0747, over 29041.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3451, pruned_loss=0.09252, over 5659756.52 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.369, pruned_loss=0.1268, over 5683947.71 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3434, pruned_loss=0.08954, over 5652765.98 frames. ], batch size: 285, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:42:20,819 INFO [train.py:968] (1/2) Epoch 10, batch 31750, giga_loss[loss=0.2874, simple_loss=0.3453, pruned_loss=0.1147, over 24702.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3435, pruned_loss=0.09146, over 5656166.83 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3688, pruned_loss=0.1266, over 5684282.91 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.08853, over 5649937.45 frames. ], batch size: 705, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:42:28,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1455, 1.1819, 4.0323, 3.1388], device='cuda:1'), covar=tensor([0.1693, 0.2580, 0.0422, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0644, 0.0578, 0.0842, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 09:43:04,331 INFO [optim.py:369] (1/2) 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 09:43:18,973 INFO [train.py:968] (1/2) Epoch 10, batch 31800, giga_loss[loss=0.2659, simple_loss=0.3455, pruned_loss=0.09315, over 28640.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3438, pruned_loss=0.09298, over 5660235.96 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3682, pruned_loss=0.1262, over 5689016.69 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3421, pruned_loss=0.08971, over 5650018.26 frames. ], batch size: 307, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:43:59,497 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:44:16,162 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-05 09:44:26,900 INFO [train.py:968] (1/2) Epoch 10, batch 31850, giga_loss[loss=0.2516, simple_loss=0.3348, pruned_loss=0.08423, over 28677.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3442, pruned_loss=0.09455, over 5661400.33 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3677, pruned_loss=0.126, over 5686743.77 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3423, pruned_loss=0.09086, over 5653853.25 frames. ], batch size: 262, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:44:56,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 09:45:27,442 INFO [optim.py:369] (1/2) 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 09:45:43,676 INFO [train.py:968] (1/2) Epoch 10, batch 31900, giga_loss[loss=0.2187, simple_loss=0.3077, pruned_loss=0.0649, over 29012.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.344, pruned_loss=0.09485, over 5673873.89 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3679, pruned_loss=0.1262, over 5688369.84 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3418, pruned_loss=0.09099, over 5666031.47 frames. ], batch size: 155, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:46:01,029 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:46:51,762 INFO [train.py:968] (1/2) Epoch 10, batch 31950, libri_loss[loss=0.321, simple_loss=0.3721, pruned_loss=0.135, over 26017.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3409, pruned_loss=0.09331, over 5664763.49 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3677, pruned_loss=0.1262, over 5680825.17 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3385, pruned_loss=0.08932, over 5665532.64 frames. ], batch size: 136, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:47:09,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3089, 4.1145, 3.9049, 1.8251], device='cuda:1'), covar=tensor([0.0551, 0.0658, 0.0747, 0.2207], device='cuda:1'), in_proj_covar=tensor([0.1013, 0.0949, 0.0833, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 09:47:38,578 INFO [optim.py:369] (1/2) 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 09:47:53,683 INFO [train.py:968] (1/2) Epoch 10, batch 32000, giga_loss[loss=0.2315, simple_loss=0.3154, pruned_loss=0.07378, over 28841.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3394, pruned_loss=0.09265, over 5650072.72 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3677, pruned_loss=0.1264, over 5663534.98 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3367, pruned_loss=0.08844, over 5664807.47 frames. ], batch size: 174, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:49:01,294 INFO [train.py:968] (1/2) Epoch 10, batch 32050, giga_loss[loss=0.2495, simple_loss=0.3325, pruned_loss=0.08321, over 29010.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.0928, over 5643438.28 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3677, pruned_loss=0.1264, over 5655286.73 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.336, pruned_loss=0.08882, over 5662456.88 frames. ], batch size: 186, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:49:08,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3309, 1.9685, 1.5878, 1.3284], device='cuda:1'), covar=tensor([0.2535, 0.1457, 0.1554, 0.1820], device='cuda:1'), in_proj_covar=tensor([0.1644, 0.1539, 0.1491, 0.1607], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 09:49:53,533 INFO [optim.py:369] (1/2) 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 09:49:55,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:50:05,872 INFO [train.py:968] (1/2) Epoch 10, batch 32100, giga_loss[loss=0.3029, simple_loss=0.3703, pruned_loss=0.1178, over 29122.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3418, pruned_loss=0.09368, over 5654493.73 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.367, pruned_loss=0.1261, over 5657532.82 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3402, pruned_loss=0.09055, over 5667246.51 frames. ], batch size: 113, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:51:10,424 INFO [train.py:968] (1/2) Epoch 10, batch 32150, giga_loss[loss=0.2334, simple_loss=0.3114, pruned_loss=0.0777, over 29141.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3403, pruned_loss=0.09366, over 5660353.31 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3661, pruned_loss=0.1256, over 5662554.39 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3394, pruned_loss=0.09104, over 5665901.35 frames. ], batch size: 128, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:51:35,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-05 09:51:52,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 09:51:58,395 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 32200, giga_loss[loss=0.2786, simple_loss=0.3537, pruned_loss=0.1018, over 28857.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3399, pruned_loss=0.09448, over 5653276.37 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3659, pruned_loss=0.1255, over 5653014.88 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3388, pruned_loss=0.09176, over 5666115.54 frames. ], batch size: 174, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:52:17,821 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=441800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:52:19,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9150, 1.0664, 1.0044, 0.8482], device='cuda:1'), covar=tensor([0.1250, 0.1395, 0.0769, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.1660, 0.1553, 0.1501, 0.1618], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 09:52:30,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 09:52:38,332 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:52:39,632 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=441819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:52:43,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-05 09:53:09,006 INFO [train.py:968] (1/2) Epoch 10, batch 32250, giga_loss[loss=0.2753, simple_loss=0.3546, pruned_loss=0.09796, over 28672.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3404, pruned_loss=0.09505, over 5656697.99 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3657, pruned_loss=0.1255, over 5656523.61 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.339, pruned_loss=0.09207, over 5663584.05 frames. ], batch size: 307, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:54:08,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=441882.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:54:09,503 INFO [optim.py:369] (1/2) 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 09:54:25,431 INFO [train.py:968] (1/2) Epoch 10, batch 32300, giga_loss[loss=0.2889, simple_loss=0.3644, pruned_loss=0.1067, over 28884.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3417, pruned_loss=0.09457, over 5657079.57 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3656, pruned_loss=0.1255, over 5660135.07 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3403, pruned_loss=0.09175, over 5659231.74 frames. ], batch size: 284, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:54:40,650 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 09:54:56,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4321, 1.5727, 1.4852, 1.4288], device='cuda:1'), covar=tensor([0.1639, 0.1452, 0.1189, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.1646, 0.1540, 0.1489, 0.1607], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 09:55:08,143 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4390, 1.9075, 1.4453, 1.5915], device='cuda:1'), covar=tensor([0.2362, 0.2197, 0.2472, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.1274, 0.0938, 0.1134, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 09:55:41,265 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 32350, giga_loss[loss=0.2531, simple_loss=0.3377, pruned_loss=0.08432, over 28431.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3428, pruned_loss=0.0948, over 5662770.17 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3654, pruned_loss=0.1256, over 5665044.06 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3414, pruned_loss=0.09192, over 5660127.38 frames. ], batch size: 368, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:55:45,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=441946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:56:19,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8166, 3.6511, 3.4357, 1.5190], device='cuda:1'), covar=tensor([0.0678, 0.0801, 0.0785, 0.2334], device='cuda:1'), in_proj_covar=tensor([0.1001, 0.0939, 0.0822, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 09:56:27,267 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=441975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:56:38,111 INFO [optim.py:369] (1/2) 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 09:56:52,941 INFO [train.py:968] (1/2) Epoch 10, batch 32400, giga_loss[loss=0.2282, simple_loss=0.3135, pruned_loss=0.07144, over 28402.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.09409, over 5663893.92 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3648, pruned_loss=0.1252, over 5659882.20 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3398, pruned_loss=0.09159, over 5665891.36 frames. ], batch size: 368, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:57:36,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3250, 1.5005, 1.3323, 1.2287], device='cuda:1'), covar=tensor([0.1687, 0.1501, 0.1090, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.1642, 0.1540, 0.1482, 0.1606], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 09:57:36,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442025.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:57:41,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442028.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 09:57:58,889 INFO [train.py:968] (1/2) Epoch 10, batch 32450, giga_loss[loss=0.253, simple_loss=0.3261, pruned_loss=0.08991, over 28910.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3358, pruned_loss=0.09238, over 5670832.58 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3646, pruned_loss=0.125, over 5663865.15 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3348, pruned_loss=0.09003, over 5668809.94 frames. ], batch size: 227, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 09:58:16,068 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442057.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 09:58:17,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 09:58:54,828 INFO [optim.py:369] (1/2) 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 09:59:07,113 INFO [train.py:968] (1/2) Epoch 10, batch 32500, giga_loss[loss=0.3005, simple_loss=0.3579, pruned_loss=0.1216, over 26812.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3331, pruned_loss=0.09158, over 5645542.31 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3647, pruned_loss=0.1253, over 5647195.07 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3318, pruned_loss=0.08916, over 5657751.37 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 09:59:30,902 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:00:07,689 INFO [train.py:968] (1/2) Epoch 10, batch 32550, giga_loss[loss=0.2837, simple_loss=0.3574, pruned_loss=0.105, over 28943.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3353, pruned_loss=0.09329, over 5643392.71 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3645, pruned_loss=0.1252, over 5648420.09 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3344, pruned_loss=0.09131, over 5651894.02 frames. ], batch size: 199, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:00:52,300 INFO [optim.py:369] (1/2) 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 10:01:02,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:02,774 INFO [train.py:968] (1/2) Epoch 10, batch 32600, giga_loss[loss=0.25, simple_loss=0.3183, pruned_loss=0.09081, over 26663.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3354, pruned_loss=0.09404, over 5652434.73 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3639, pruned_loss=0.1251, over 5658141.78 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3342, pruned_loss=0.0914, over 5650434.80 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:01:03,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:11,863 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:16,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:34,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5877, 2.0269, 1.6888, 1.6291], device='cuda:1'), covar=tensor([0.0743, 0.0251, 0.0301, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0117, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 10:01:50,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:01:51,057 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.93 vs. limit=2.0
+2023-03-05 10:02:00,200 INFO [train.py:968] (1/2) Epoch 10, batch 32650, giga_loss[loss=0.2595, simple_loss=0.3358, pruned_loss=0.09164, over 28110.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3331, pruned_loss=0.09175, over 5652932.87 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3627, pruned_loss=0.1244, over 5661570.51 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3323, pruned_loss=0.08925, over 5647984.53 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:02:51,305 INFO [optim.py:369] (1/2) 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 10:03:03,489 INFO [train.py:968] (1/2) Epoch 10, batch 32700, libri_loss[loss=0.3427, simple_loss=0.3948, pruned_loss=0.1453, over 29308.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3324, pruned_loss=0.09128, over 5665302.94 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3626, pruned_loss=0.1243, over 5668821.20 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3309, pruned_loss=0.08831, over 5654517.08 frames. ], batch size: 94, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:03:31,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:02,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:03,548 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:05,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:06,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:10,406 INFO [train.py:968] (1/2) Epoch 10, batch 32750, giga_loss[loss=0.2848, simple_loss=0.3584, pruned_loss=0.1056, over 28788.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3314, pruned_loss=0.09056, over 5668570.26 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3623, pruned_loss=0.1242, over 5672642.20 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.33, pruned_loss=0.08784, over 5656629.34 frames. ], batch size: 263, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:04:43,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:44,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:04:48,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-05 10:05:06,378 INFO [optim.py:369] (1/2) 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 10:05:16,935 INFO [train.py:968] (1/2) Epoch 10, batch 32800, giga_loss[loss=0.3256, simple_loss=0.3895, pruned_loss=0.1308, over 28346.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3314, pruned_loss=0.09021, over 5658402.64 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3619, pruned_loss=0.1239, over 5672721.27 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3298, pruned_loss=0.0874, over 5648191.91 frames. ], batch size: 368, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 10:06:20,166 INFO [train.py:968] (1/2) Epoch 10, batch 32850, giga_loss[loss=0.2746, simple_loss=0.3491, pruned_loss=0.1, over 28532.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3317, pruned_loss=0.09067, over 5662077.33 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3615, pruned_loss=0.1236, over 5676096.90 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3303, pruned_loss=0.08821, over 5650895.99 frames. ], batch size: 336, lr: 3.20e-03, grad_scale: 8.0
+2023-03-05 10:07:11,898 INFO [optim.py:369] (1/2) 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 10:07:14,775 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 32900, libri_loss[loss=0.2705, simple_loss=0.3295, pruned_loss=0.1058, over 29360.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3316, pruned_loss=0.09106, over 5668698.75 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3615, pruned_loss=0.1238, over 5680463.54 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3299, pruned_loss=0.08826, over 5655552.54 frames. ], batch size: 71, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:08:06,571 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442531.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:08:21,039 INFO [train.py:968] (1/2) Epoch 10, batch 32950, giga_loss[loss=0.2171, simple_loss=0.2918, pruned_loss=0.07122, over 24144.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3316, pruned_loss=0.08979, over 5668326.32 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3611, pruned_loss=0.1236, over 5685361.57 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3298, pruned_loss=0.08695, over 5653225.80 frames. ], batch size: 705, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:08:28,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2189, 1.1960, 3.8668, 3.1020], device='cuda:1'), covar=tensor([0.1664, 0.2714, 0.0385, 0.1590], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0576, 0.0835, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 10:08:44,671 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:08:59,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3882, 1.6222, 1.3326, 1.3800], device='cuda:1'), covar=tensor([0.2229, 0.2013, 0.2124, 0.1789], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.0939, 0.1132, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 10:09:09,541 INFO [optim.py:369] (1/2) 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 10:09:16,662 INFO [train.py:968] (1/2) Epoch 10, batch 33000, giga_loss[loss=0.2192, simple_loss=0.292, pruned_loss=0.07318, over 24277.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3346, pruned_loss=0.09043, over 5660202.09 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.361, pruned_loss=0.1234, over 5679226.52 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3327, pruned_loss=0.08743, over 5652859.87 frames. ], batch size: 705, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:09:16,663 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 10:09:25,195 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 10:10:08,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:11,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:13,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:24,938 INFO [train.py:968] (1/2) Epoch 10, batch 33050, libri_loss[loss=0.2997, simple_loss=0.3627, pruned_loss=0.1184, over 29378.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3372, pruned_loss=0.0919, over 5656755.59 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3611, pruned_loss=0.1235, over 5681667.05 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.335, pruned_loss=0.08884, over 5648160.33 frames. ], batch size: 92, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:10:48,509 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:10:50,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:11:21,270 INFO [optim.py:369] (1/2) 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 10:11:22,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:11:24,579 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=442691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:11:27,740 INFO [train.py:968] (1/2) Epoch 10, batch 33100, giga_loss[loss=0.2432, simple_loss=0.3251, pruned_loss=0.08063, over 28935.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3378, pruned_loss=0.09235, over 5659232.20 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3607, pruned_loss=0.1233, over 5682742.08 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.336, pruned_loss=0.08934, over 5650900.15 frames. ], batch size: 199, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:11:30,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3429, 1.8129, 1.3020, 0.7279], device='cuda:1'), covar=tensor([0.2864, 0.1701, 0.2246, 0.3494], device='cuda:1'), in_proj_covar=tensor([0.1529, 0.1471, 0.1469, 0.1260], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 10:12:08,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4976, 2.2801, 1.6364, 0.7166], device='cuda:1'), covar=tensor([0.4231, 0.1986, 0.2836, 0.4244], device='cuda:1'), in_proj_covar=tensor([0.1534, 0.1477, 0.1472, 0.1265], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 10:12:26,074 INFO [train.py:968] (1/2) Epoch 10, batch 33150, libri_loss[loss=0.3317, simple_loss=0.3834, pruned_loss=0.14, over 25880.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3383, pruned_loss=0.09321, over 5662287.34 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3603, pruned_loss=0.1232, over 5687105.79 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3364, pruned_loss=0.08989, over 5650971.78 frames. ], batch size: 136, lr: 3.20e-03, grad_scale: 1.0
+2023-03-05 10:12:32,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4362, 2.3381, 1.7533, 2.1302], device='cuda:1'), covar=tensor([0.0739, 0.0613, 0.0893, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0436, 0.0497, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 10:13:14,214 INFO [optim.py:369] (1/2) 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 10:13:22,196 INFO [train.py:968] (1/2) Epoch 10, batch 33200, giga_loss[loss=0.2625, simple_loss=0.3505, pruned_loss=0.08728, over 28696.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3355, pruned_loss=0.09139, over 5661141.34 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3592, pruned_loss=0.1226, over 5681586.12 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3343, pruned_loss=0.08826, over 5656555.82 frames. ], batch size: 242, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:14:04,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3523, 1.6750, 1.5114, 1.5950], device='cuda:1'), covar=tensor([0.0745, 0.0300, 0.0307, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0056, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 10:14:10,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=442832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:14:13,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=442835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:14:23,051 INFO [train.py:968] (1/2) Epoch 10, batch 33250, giga_loss[loss=0.2031, simple_loss=0.2721, pruned_loss=0.06707, over 24334.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3347, pruned_loss=0.09181, over 5659341.05 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3594, pruned_loss=0.1227, over 5684267.61 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.333, pruned_loss=0.08866, over 5652986.47 frames. ], batch size: 705, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:14:51,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=442864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:15:05,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4467, 1.6130, 1.3391, 1.9386], device='cuda:1'), covar=tensor([0.2441, 0.2344, 0.2587, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.0929, 0.1123, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 10:15:19,241 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 33300, giga_loss[loss=0.2652, simple_loss=0.3478, pruned_loss=0.09129, over 28367.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3349, pruned_loss=0.09167, over 5670197.39 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3599, pruned_loss=0.123, over 5686072.42 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3329, pruned_loss=0.08855, over 5663343.63 frames. ], batch size: 368, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:15:32,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4872, 3.3275, 3.1442, 1.8347], device='cuda:1'), covar=tensor([0.0715, 0.0882, 0.0821, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.1008, 0.0941, 0.0822, 0.0632], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 10:15:38,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 10:15:44,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442906.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:16:31,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=442940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:16:34,135 INFO [train.py:968] (1/2) Epoch 10, batch 33350, giga_loss[loss=0.3375, simple_loss=0.3864, pruned_loss=0.1442, over 26855.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3376, pruned_loss=0.09274, over 5666535.55 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3599, pruned_loss=0.1231, over 5687476.93 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3358, pruned_loss=0.08995, over 5659659.87 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:16:41,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.44 vs. limit=5.0
+2023-03-05 10:17:29,555 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 33400, giga_loss[loss=0.2511, simple_loss=0.333, pruned_loss=0.08466, over 29017.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3373, pruned_loss=0.0928, over 5662877.85 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3597, pruned_loss=0.1231, over 5691111.65 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3357, pruned_loss=0.09005, over 5654006.80 frames. ], batch size: 285, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:17:51,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:18:30,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:18:39,691 INFO [train.py:968] (1/2) Epoch 10, batch 33450, giga_loss[loss=0.3578, simple_loss=0.406, pruned_loss=0.1548, over 26721.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3406, pruned_loss=0.09484, over 5682434.42 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3596, pruned_loss=0.1232, over 5699339.38 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3385, pruned_loss=0.09141, over 5667110.40 frames. ], batch size: 555, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:18:48,100 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443049.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:18:51,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443052.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:19:05,128 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-05 10:19:09,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:19:24,424 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443081.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:19:26,734 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:19:28,625 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:19:30,187 INFO [optim.py:369] (1/2) 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 10:19:37,240 INFO [train.py:968] (1/2) Epoch 10, batch 33500, giga_loss[loss=0.2576, simple_loss=0.3205, pruned_loss=0.09737, over 24252.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3435, pruned_loss=0.09579, over 5675522.03 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3595, pruned_loss=0.1231, over 5702034.02 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3418, pruned_loss=0.09281, over 5660688.81 frames. ], batch size: 705, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:20:02,676 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:20:31,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-05 10:20:39,041 INFO [train.py:968] (1/2) Epoch 10, batch 33550, giga_loss[loss=0.2651, simple_loss=0.3432, pruned_loss=0.09352, over 29071.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3428, pruned_loss=0.09479, over 5679825.19 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.359, pruned_loss=0.1227, over 5707670.27 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3414, pruned_loss=0.09193, over 5662025.57 frames. ], batch size: 186, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:20:46,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:20:51,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:30,432 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:31,555 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:33,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:21:41,934 INFO [optim.py:369] (1/2) 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 10:21:49,816 INFO [train.py:968] (1/2) Epoch 10, batch 33600, giga_loss[loss=0.2523, simple_loss=0.3345, pruned_loss=0.0851, over 28990.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.341, pruned_loss=0.09409, over 5674701.23 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3588, pruned_loss=0.1227, over 5710477.57 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3399, pruned_loss=0.09151, over 5657694.93 frames. ], batch size: 213, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:21:54,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4670, 1.6699, 1.7608, 1.3696], device='cuda:1'), covar=tensor([0.1551, 0.2113, 0.1238, 0.1506], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0686, 0.0846, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 10:22:12,976 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:15,399 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:17,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:52,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=443241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:22:54,566 INFO [train.py:968] (1/2) Epoch 10, batch 33650, giga_loss[loss=0.2848, simple_loss=0.3511, pruned_loss=0.1092, over 28881.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3408, pruned_loss=0.09525, over 5675320.77 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3587, pruned_loss=0.1228, over 5715335.78 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3396, pruned_loss=0.09245, over 5656682.92 frames. ], batch size: 213, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:23:14,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-05 10:23:30,775 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:23:58,837 INFO [train.py:968] (1/2) Epoch 10, batch 33700, giga_loss[loss=0.2691, simple_loss=0.3445, pruned_loss=0.09688, over 28944.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.34, pruned_loss=0.09486, over 5668999.54 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3585, pruned_loss=0.1227, over 5719680.93 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3387, pruned_loss=0.09211, over 5649534.58 frames. ], batch size: 106, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:24:07,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4160, 1.6755, 1.3325, 1.4383], device='cuda:1'), covar=tensor([0.2396, 0.2306, 0.2668, 0.2044], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.0928, 0.1121, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 10:24:08,076 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-05 10:24:56,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4125, 1.6825, 1.5026, 1.6039], device='cuda:1'), covar=tensor([0.0759, 0.0284, 0.0322, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0056, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 10:24:58,747 INFO [train.py:968] (1/2) Epoch 10, batch 33750, libri_loss[loss=0.2994, simple_loss=0.3631, pruned_loss=0.1179, over 29146.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3403, pruned_loss=0.09626, over 5664783.21 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3581, pruned_loss=0.1225, over 5712621.22 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3389, pruned_loss=0.09307, over 5653368.02 frames. ], batch size: 101, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:25:51,160 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 33800, giga_loss[loss=0.2432, simple_loss=0.3246, pruned_loss=0.08088, over 27548.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3386, pruned_loss=0.09602, over 5653719.42 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3581, pruned_loss=0.1225, over 5713726.89 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.337, pruned_loss=0.09273, over 5642047.06 frames. ], batch size: 472, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:26:57,727 INFO [train.py:968] (1/2) Epoch 10, batch 33850, libri_loss[loss=0.3084, simple_loss=0.3657, pruned_loss=0.1256, over 29279.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3385, pruned_loss=0.09465, over 5657419.84 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3584, pruned_loss=0.1227, over 5716340.67 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3367, pruned_loss=0.09147, over 5644968.44 frames. ], batch size: 94, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:27:49,462 INFO [optim.py:369] (1/2) 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 10:27:57,509 INFO [train.py:968] (1/2) Epoch 10, batch 33900, giga_loss[loss=0.2235, simple_loss=0.2998, pruned_loss=0.07356, over 24489.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3369, pruned_loss=0.09255, over 5663526.45 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3582, pruned_loss=0.1226, over 5710017.69 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3349, pruned_loss=0.08919, over 5657216.39 frames. ], batch size: 705, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:27:57,804 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=443494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:28:50,049 INFO [train.py:968] (1/2) Epoch 10, batch 33950, giga_loss[loss=0.2704, simple_loss=0.3518, pruned_loss=0.09444, over 29080.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3397, pruned_loss=0.09208, over 5679334.51 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3577, pruned_loss=0.1221, over 5714993.87 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3382, pruned_loss=0.089, over 5668762.19 frames. ], batch size: 128, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:29:43,462 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 34000, giga_loss[loss=0.2195, simple_loss=0.3111, pruned_loss=0.064, over 29038.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.34, pruned_loss=0.09118, over 5668505.11 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3575, pruned_loss=0.122, over 5714993.43 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3387, pruned_loss=0.08858, over 5659951.23 frames. ], batch size: 136, lr: 3.20e-03, grad_scale: 4.0
+2023-03-05 10:30:09,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2939, 1.3035, 4.0229, 3.2169], device='cuda:1'), covar=tensor([0.1611, 0.2498, 0.0401, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0635, 0.0570, 0.0825, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 10:30:51,158 INFO [train.py:968] (1/2) Epoch 10, batch 34050, giga_loss[loss=0.2318, simple_loss=0.3206, pruned_loss=0.07153, over 29015.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3395, pruned_loss=0.09115, over 5675453.69 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3568, pruned_loss=0.1216, over 5721320.46 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3386, pruned_loss=0.08848, over 5661614.63 frames. ], batch size: 285, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:30:57,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443647.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:31:56,164 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 34100, giga_loss[loss=0.2831, simple_loss=0.3602, pruned_loss=0.103, over 28049.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3398, pruned_loss=0.09078, over 5677904.85 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3566, pruned_loss=0.1214, over 5723139.43 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3391, pruned_loss=0.08851, over 5664980.10 frames. ], batch size: 412, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:32:06,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 10:33:11,213 INFO [train.py:968] (1/2) Epoch 10, batch 34150, giga_loss[loss=0.2654, simple_loss=0.3505, pruned_loss=0.09015, over 28587.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3406, pruned_loss=0.09116, over 5671094.37 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3564, pruned_loss=0.1212, over 5717704.55 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3398, pruned_loss=0.08888, over 5665289.93 frames. ], batch size: 242, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:34:19,795 INFO [optim.py:369] (1/2) 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 10:34:20,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=443790.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:34:22,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=443793.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:34:22,474 INFO [train.py:968] (1/2) Epoch 10, batch 34200, giga_loss[loss=0.2668, simple_loss=0.3529, pruned_loss=0.09034, over 28954.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3404, pruned_loss=0.09059, over 5668601.36 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3565, pruned_loss=0.1213, over 5720915.66 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3394, pruned_loss=0.08797, over 5659661.19 frames. ], batch size: 136, lr: 3.20e-03, grad_scale: 2.0
+2023-03-05 10:34:59,726 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 34250, giga_loss[loss=0.2902, simple_loss=0.3783, pruned_loss=0.1011, over 29094.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3437, pruned_loss=0.09208, over 5657156.71 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3566, pruned_loss=0.1214, over 5712315.32 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3426, pruned_loss=0.08969, over 5658325.14 frames. ], batch size: 285, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:35:50,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3319, 2.1211, 1.5956, 1.5152], device='cuda:1'), covar=tensor([0.0837, 0.0248, 0.0298, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0113, 0.0118, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0056, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 10:35:57,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=443869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:36:20,726 INFO [optim.py:369] (1/2) 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 10:36:25,360 INFO [train.py:968] (1/2) Epoch 10, batch 34300, giga_loss[loss=0.2527, simple_loss=0.3355, pruned_loss=0.08499, over 29173.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3445, pruned_loss=0.09249, over 5675447.72 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3565, pruned_loss=0.1212, over 5719223.19 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08948, over 5668037.43 frames. ], batch size: 285, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:37:33,864 INFO [train.py:968] (1/2) Epoch 10, batch 34350, giga_loss[loss=0.2288, simple_loss=0.319, pruned_loss=0.06931, over 29016.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3426, pruned_loss=0.09185, over 5687306.84 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3565, pruned_loss=0.1212, over 5721103.15 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3413, pruned_loss=0.08895, over 5678859.65 frames. ], batch size: 136, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:38:19,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=443986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:38:22,971 INFO [optim.py:369] (1/2) 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 10:38:28,136 INFO [train.py:968] (1/2) Epoch 10, batch 34400, libri_loss[loss=0.2921, simple_loss=0.3513, pruned_loss=0.1164, over 29548.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.341, pruned_loss=0.09202, over 5674629.64 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3558, pruned_loss=0.1205, over 5709163.59 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3398, pruned_loss=0.08869, over 5675873.49 frames. ], batch size: 80, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:38:52,643 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:38:55,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:39:35,240 INFO [train.py:968] (1/2) Epoch 10, batch 34450, libri_loss[loss=0.3391, simple_loss=0.3897, pruned_loss=0.1443, over 29649.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3392, pruned_loss=0.09021, over 5677349.98 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3562, pruned_loss=0.1209, over 5702477.69 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3375, pruned_loss=0.08649, over 5682782.37 frames. ], batch size: 91, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:39:36,518 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=444044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:40:26,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444089.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:40:28,886 INFO [optim.py:369] (1/2) 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 10:40:35,025 INFO [train.py:968] (1/2) Epoch 10, batch 34500, giga_loss[loss=0.2768, simple_loss=0.3531, pruned_loss=0.1003, over 28010.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3383, pruned_loss=0.08971, over 5686263.59 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3567, pruned_loss=0.121, over 5704648.79 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3361, pruned_loss=0.08577, over 5688168.99 frames. ], batch size: 412, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:41:36,012 INFO [train.py:968] (1/2) Epoch 10, batch 34550, libri_loss[loss=0.3151, simple_loss=0.3738, pruned_loss=0.1282, over 29541.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3392, pruned_loss=0.09044, over 5680333.88 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3563, pruned_loss=0.1208, over 5707702.07 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3374, pruned_loss=0.08693, over 5678582.96 frames. ], batch size: 89, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:42:01,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-05 10:42:30,196 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 34600, libri_loss[loss=0.3166, simple_loss=0.3677, pruned_loss=0.1327, over 29181.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3419, pruned_loss=0.09235, over 5668028.10 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3567, pruned_loss=0.1212, over 5702907.41 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3396, pruned_loss=0.08833, over 5669521.75 frames. ], batch size: 101, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:42:50,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 10:43:31,603 INFO [train.py:968] (1/2) Epoch 10, batch 34650, giga_loss[loss=0.247, simple_loss=0.3204, pruned_loss=0.08675, over 28354.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3389, pruned_loss=0.09117, over 5670524.95 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3566, pruned_loss=0.1211, over 5705009.09 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3371, pruned_loss=0.08782, over 5669464.42 frames. ], batch size: 368, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:43:58,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8826, 2.2340, 1.8210, 2.1040], device='cuda:1'), covar=tensor([0.0685, 0.0220, 0.0296, 0.0761], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0118, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0056, 0.0052, 0.0087], device='cuda:1')
+2023-03-05 10:44:18,006 INFO [optim.py:369] (1/2) 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 10:44:20,039 INFO [train.py:968] (1/2) Epoch 10, batch 34700, libri_loss[loss=0.312, simple_loss=0.3723, pruned_loss=0.1258, over 29547.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3372, pruned_loss=0.09155, over 5660626.36 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3559, pruned_loss=0.1208, over 5694775.86 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3356, pruned_loss=0.08776, over 5667948.76 frames. ], batch size: 82, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:45:13,244 INFO [train.py:968] (1/2) Epoch 10, batch 34750, giga_loss[loss=0.2952, simple_loss=0.3737, pruned_loss=0.1083, over 28888.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3378, pruned_loss=0.09246, over 5664864.42 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3551, pruned_loss=0.1202, over 5700293.52 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3367, pruned_loss=0.08928, over 5664975.61 frames. ], batch size: 284, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 10:45:20,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7481, 4.5400, 4.2711, 2.0773], device='cuda:1'), covar=tensor([0.0451, 0.0618, 0.0696, 0.2195], device='cuda:1'), in_proj_covar=tensor([0.0998, 0.0929, 0.0813, 0.0628], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 10:45:29,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:45:58,772 INFO [train.py:968] (1/2) Epoch 10, batch 34800, giga_loss[loss=0.303, simple_loss=0.3765, pruned_loss=0.1147, over 28791.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3454, pruned_loss=0.09762, over 5652613.26 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.355, pruned_loss=0.1202, over 5690907.00 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3442, pruned_loss=0.09418, over 5660623.09 frames. ], batch size: 284, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:46:40,904 INFO [train.py:968] (1/2) Epoch 10, batch 34850, giga_loss[loss=0.2846, simple_loss=0.3702, pruned_loss=0.09945, over 29083.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3547, pruned_loss=0.103, over 5662209.56 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3549, pruned_loss=0.1201, over 5686846.80 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3537, pruned_loss=0.09985, over 5670758.25 frames. ], batch size: 155, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:46:56,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=444462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:46:57,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=444464.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:47:21,915 INFO [optim.py:369] (1/2) 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 10:47:24,462 INFO [train.py:968] (1/2) Epoch 10, batch 34900, giga_loss[loss=0.2616, simple_loss=0.3413, pruned_loss=0.09096, over 28558.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3579, pruned_loss=0.1053, over 5667081.44 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3547, pruned_loss=0.12, over 5689471.75 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3574, pruned_loss=0.1026, over 5671247.01 frames. ], batch size: 336, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:47:33,289 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:47:35,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:47:46,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3671, 2.9086, 1.4924, 1.4663], device='cuda:1'), covar=tensor([0.0923, 0.0311, 0.0859, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0500, 0.0336, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 10:47:56,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:48:00,461 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 34950, giga_loss[loss=0.2343, simple_loss=0.2882, pruned_loss=0.09017, over 24135.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3528, pruned_loss=0.1031, over 5672516.76 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3547, pruned_loss=0.12, over 5691941.58 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3524, pruned_loss=0.1007, over 5673097.77 frames. ], batch size: 705, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:48:12,715 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 10:48:35,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9271, 1.1222, 1.0632, 0.8344], device='cuda:1'), covar=tensor([0.1611, 0.1671, 0.0978, 0.1465], device='cuda:1'), in_proj_covar=tensor([0.1664, 0.1534, 0.1499, 0.1622], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 10:48:43,827 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 35000, giga_loss[loss=0.2291, simple_loss=0.3029, pruned_loss=0.07759, over 28993.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3463, pruned_loss=0.1005, over 5681947.22 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3548, pruned_loss=0.1198, over 5697942.09 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3457, pruned_loss=0.09806, over 5676596.09 frames. ], batch size: 227, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:48:56,186 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:48:59,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=444607.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:49:00,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:49:02,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=444610.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:49:21,016 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=444637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:49:22,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=444639.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:49:26,908 INFO [train.py:968] (1/2) Epoch 10, batch 35050, giga_loss[loss=0.2438, simple_loss=0.3217, pruned_loss=0.08298, over 28759.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3401, pruned_loss=0.09784, over 5674984.05 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3559, pruned_loss=0.1204, over 5685610.71 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3383, pruned_loss=0.09496, over 5681423.83 frames. ], batch size: 284, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:49:41,496 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:50:04,819 INFO [optim.py:369] (1/2) 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 10:50:06,543 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 10:50:06,847 INFO [train.py:968] (1/2) Epoch 10, batch 35100, giga_loss[loss=0.2098, simple_loss=0.2857, pruned_loss=0.06695, over 29019.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3342, pruned_loss=0.09567, over 5676630.98 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3566, pruned_loss=0.1207, over 5689767.50 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3314, pruned_loss=0.09219, over 5677506.52 frames. ], batch size: 155, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:50:28,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5266, 1.7700, 1.4136, 1.7428], device='cuda:1'), covar=tensor([0.2392, 0.2310, 0.2594, 0.2174], device='cuda:1'), in_proj_covar=tensor([0.1257, 0.0934, 0.1119, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 10:50:47,146 INFO [train.py:968] (1/2) Epoch 10, batch 35150, giga_loss[loss=0.2177, simple_loss=0.2744, pruned_loss=0.08051, over 23951.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3279, pruned_loss=0.09283, over 5672786.37 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3572, pruned_loss=0.121, over 5685371.90 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3246, pruned_loss=0.08924, over 5676905.59 frames. ], batch size: 705, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:50:52,831 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:50:59,228 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444761.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:51:23,259 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 35200, giga_loss[loss=0.2433, simple_loss=0.3152, pruned_loss=0.08575, over 28787.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3236, pruned_loss=0.09078, over 5690637.09 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.357, pruned_loss=0.1206, over 5692384.99 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3201, pruned_loss=0.08725, over 5687379.15 frames. ], batch size: 119, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:51:39,473 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=444812.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:52:06,606 INFO [train.py:968] (1/2) Epoch 10, batch 35250, giga_loss[loss=0.2229, simple_loss=0.2998, pruned_loss=0.07302, over 29019.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3213, pruned_loss=0.08999, over 5694831.54 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3568, pruned_loss=0.1203, over 5698894.49 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3178, pruned_loss=0.08659, over 5686565.77 frames. ], batch size: 213, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:52:44,401 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 35300, giga_loss[loss=0.2409, simple_loss=0.3143, pruned_loss=0.08378, over 27967.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3187, pruned_loss=0.08889, over 5688370.84 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3571, pruned_loss=0.1204, over 5700536.17 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.315, pruned_loss=0.08563, over 5680200.52 frames. ], batch size: 412, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:52:56,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=444906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:52:58,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3584, 3.2582, 1.5129, 1.4005], device='cuda:1'), covar=tensor([0.0961, 0.0305, 0.0904, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0502, 0.0336, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 10:53:25,958 INFO [train.py:968] (1/2) Epoch 10, batch 35350, giga_loss[loss=0.2675, simple_loss=0.3131, pruned_loss=0.111, over 24004.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3165, pruned_loss=0.08814, over 5669313.00 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3575, pruned_loss=0.1204, over 5689352.26 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3118, pruned_loss=0.08434, over 5670717.97 frames. ], batch size: 705, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:54:02,598 INFO [optim.py:369] (1/2) 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 10:54:04,141 INFO [train.py:968] (1/2) Epoch 10, batch 35400, giga_loss[loss=0.2127, simple_loss=0.2906, pruned_loss=0.06738, over 28901.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3127, pruned_loss=0.0861, over 5680406.08 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3575, pruned_loss=0.1204, over 5691588.36 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3086, pruned_loss=0.08283, over 5679481.50 frames. ], batch size: 174, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:54:28,430 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:54:38,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:54:39,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6640, 2.3701, 1.6449, 0.7598], device='cuda:1'), covar=tensor([0.3501, 0.1986, 0.2582, 0.3683], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1490, 0.1479, 0.1264], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 10:54:42,490 INFO [train.py:968] (1/2) Epoch 10, batch 35450, giga_loss[loss=0.1873, simple_loss=0.267, pruned_loss=0.05382, over 28826.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3128, pruned_loss=0.0863, over 5688345.28 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3585, pruned_loss=0.1207, over 5697406.87 frames. ], giga_tot_loss[loss=0.235, simple_loss=0.3064, pruned_loss=0.08176, over 5681882.19 frames. ], batch size: 199, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:54:47,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:54:49,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:55:13,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:55:23,116 INFO [optim.py:369] (1/2) 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 10:55:24,450 INFO [train.py:968] (1/2) Epoch 10, batch 35500, giga_loss[loss=0.2211, simple_loss=0.2943, pruned_loss=0.07396, over 28642.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.308, pruned_loss=0.08369, over 5690747.03 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3585, pruned_loss=0.1207, over 5698468.83 frames. ], giga_tot_loss[loss=0.2313, simple_loss=0.3027, pruned_loss=0.07994, over 5684671.93 frames. ], batch size: 307, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:55:53,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:56:01,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445136.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:56:07,661 INFO [train.py:968] (1/2) Epoch 10, batch 35550, giga_loss[loss=0.245, simple_loss=0.2935, pruned_loss=0.09831, over 23855.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3059, pruned_loss=0.08292, over 5681794.95 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3592, pruned_loss=0.121, over 5702379.02 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3001, pruned_loss=0.07888, over 5673051.33 frames. ], batch size: 705, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:56:39,354 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:56:41,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:56:41,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6426, 4.1156, 1.8045, 1.6165], device='cuda:1'), covar=tensor([0.0897, 0.0289, 0.0863, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0500, 0.0336, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 10:56:44,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445187.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:56:49,242 INFO [optim.py:369] (1/2) 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 10:56:51,965 INFO [train.py:968] (1/2) Epoch 10, batch 35600, giga_loss[loss=0.2889, simple_loss=0.3561, pruned_loss=0.1109, over 28556.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3051, pruned_loss=0.08295, over 5679463.83 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3595, pruned_loss=0.1212, over 5701624.08 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.2995, pruned_loss=0.0791, over 5672753.73 frames. ], batch size: 336, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:56:57,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-05 10:57:06,483 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:57:06,498 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:57:20,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1927, 1.2646, 1.1263, 0.9632], device='cuda:1'), covar=tensor([0.0779, 0.0496, 0.0975, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0433, 0.0496, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 10:57:21,432 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-05 10:57:28,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5295, 1.6280, 1.5399, 1.3225], device='cuda:1'), covar=tensor([0.1796, 0.1642, 0.1187, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.1673, 0.1542, 0.1513, 0.1640], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 10:57:34,192 INFO [train.py:968] (1/2) Epoch 10, batch 35650, giga_loss[loss=0.3213, simple_loss=0.3903, pruned_loss=0.1262, over 29008.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3163, pruned_loss=0.08876, over 5684068.00 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3601, pruned_loss=0.1215, over 5699069.43 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3097, pruned_loss=0.08424, over 5679854.38 frames. ], batch size: 227, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:57:56,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:57:56,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 10:57:58,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-05 10:58:00,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:58:06,225 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445279.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:58:09,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445282.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:58:16,633 INFO [optim.py:369] (1/2) 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 10:58:17,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8786, 1.3661, 5.3254, 3.8897], device='cuda:1'), covar=tensor([0.1933, 0.3025, 0.0526, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0638, 0.0574, 0.0837, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 10:58:18,238 INFO [train.py:968] (1/2) Epoch 10, batch 35700, libri_loss[loss=0.3632, simple_loss=0.4057, pruned_loss=0.1603, over 25704.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3311, pruned_loss=0.09734, over 5680857.83 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3607, pruned_loss=0.122, over 5698820.81 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3245, pruned_loss=0.09261, over 5677728.79 frames. ], batch size: 136, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:58:23,535 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:58:32,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445311.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 10:58:45,910 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445330.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:58:48,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:58:53,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 10:58:55,517 INFO [train.py:968] (1/2) Epoch 10, batch 35750, libri_loss[loss=0.3218, simple_loss=0.3846, pruned_loss=0.1295, over 29476.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.341, pruned_loss=0.1023, over 5686577.96 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3609, pruned_loss=0.1221, over 5704240.59 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3344, pruned_loss=0.09753, over 5678121.26 frames. ], batch size: 85, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 10:58:56,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4203, 2.1033, 1.6007, 0.5470], device='cuda:1'), covar=tensor([0.4066, 0.2161, 0.3068, 0.4405], device='cuda:1'), in_proj_covar=tensor([0.1544, 0.1491, 0.1477, 0.1265], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 10:59:10,036 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445362.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 10:59:35,555 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 35800, giga_loss[loss=0.285, simple_loss=0.3421, pruned_loss=0.114, over 23527.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3473, pruned_loss=0.1045, over 5685561.79 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3608, pruned_loss=0.1219, over 5707582.52 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3418, pruned_loss=0.1005, over 5675743.72 frames. ], batch size: 705, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 10:59:41,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 10:59:55,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3231, 4.1057, 3.8688, 1.6090], device='cuda:1'), covar=tensor([0.0488, 0.0680, 0.0708, 0.2296], device='cuda:1'), in_proj_covar=tensor([0.0999, 0.0943, 0.0822, 0.0634], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 11:00:20,538 INFO [train.py:968] (1/2) Epoch 10, batch 35850, giga_loss[loss=0.2408, simple_loss=0.3251, pruned_loss=0.07827, over 29012.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3494, pruned_loss=0.1041, over 5687115.91 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.361, pruned_loss=0.122, over 5710547.92 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3447, pruned_loss=0.1006, over 5676612.21 frames. ], batch size: 128, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:00:46,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2062, 2.3179, 2.0960, 2.0952], device='cuda:1'), covar=tensor([0.1449, 0.2017, 0.1808, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0724, 0.0661, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 11:01:07,301 INFO [optim.py:369] (1/2) 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 11:01:09,562 INFO [train.py:968] (1/2) Epoch 10, batch 35900, giga_loss[loss=0.2621, simple_loss=0.3376, pruned_loss=0.09332, over 28426.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.351, pruned_loss=0.1039, over 5680655.62 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3619, pruned_loss=0.1226, over 5712125.16 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3465, pruned_loss=0.1005, over 5670812.52 frames. ], batch size: 65, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:01:39,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4932, 2.1149, 1.4969, 0.7120], device='cuda:1'), covar=tensor([0.3891, 0.2020, 0.2942, 0.3934], device='cuda:1'), in_proj_covar=tensor([0.1539, 0.1485, 0.1477, 0.1263], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 11:01:47,314 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:01:48,270 INFO [train.py:968] (1/2) Epoch 10, batch 35950, libri_loss[loss=0.2617, simple_loss=0.3251, pruned_loss=0.09911, over 29356.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3537, pruned_loss=0.106, over 5688784.12 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3622, pruned_loss=0.1226, over 5718314.40 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3495, pruned_loss=0.1026, over 5673747.71 frames. ], batch size: 67, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:01:49,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:02:14,001 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:02:25,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=445587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:02:30,122 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 36000, giga_loss[loss=0.2698, simple_loss=0.3495, pruned_loss=0.09507, over 28634.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3559, pruned_loss=0.1077, over 5693024.55 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3625, pruned_loss=0.1228, over 5720479.45 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3522, pruned_loss=0.1047, over 5678849.11 frames. ], batch size: 242, lr: 3.19e-03, grad_scale: 8.0
+2023-03-05 11:02:30,668 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 11:02:39,707 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 11:03:17,257 INFO [train.py:968] (1/2) Epoch 10, batch 36050, giga_loss[loss=0.27, simple_loss=0.3482, pruned_loss=0.09591, over 28574.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3584, pruned_loss=0.1089, over 5696685.43 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3629, pruned_loss=0.1227, over 5726543.04 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3548, pruned_loss=0.106, over 5678919.86 frames. ], batch size: 71, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:03:19,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 11:03:31,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 11:03:56,002 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 36100, giga_loss[loss=0.2988, simple_loss=0.3736, pruned_loss=0.112, over 28643.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3615, pruned_loss=0.1096, over 5711124.51 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3632, pruned_loss=0.1228, over 5730850.23 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3584, pruned_loss=0.1069, over 5692618.92 frames. ], batch size: 60, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:04:24,231 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:04:25,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=445730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:04:27,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=445733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:04:35,985 INFO [train.py:968] (1/2) Epoch 10, batch 36150, libri_loss[loss=0.3219, simple_loss=0.376, pruned_loss=0.1339, over 29540.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.363, pruned_loss=0.1099, over 5705914.10 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3638, pruned_loss=0.1231, over 5726640.27 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3599, pruned_loss=0.1068, over 5692887.07 frames. ], batch size: 79, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:04:42,465 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 11:04:49,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5262, 1.6309, 1.4460, 1.3146], device='cuda:1'), covar=tensor([0.1844, 0.1826, 0.1425, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.1658, 0.1533, 0.1511, 0.1624], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 11:04:50,937 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=445762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:05:13,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1259, 1.2818, 0.9586, 0.8556], device='cuda:1'), covar=tensor([0.0938, 0.0490, 0.1260, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0431, 0.0495, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 11:05:15,669 INFO [train.py:968] (1/2) Epoch 10, batch 36200, libri_loss[loss=0.3662, simple_loss=0.4148, pruned_loss=0.1588, over 27727.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3643, pruned_loss=0.1099, over 5699304.46 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3641, pruned_loss=0.1234, over 5718820.31 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3615, pruned_loss=0.1069, over 5696155.11 frames. ], batch size: 115, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:05:16,272 INFO [optim.py:369] (1/2) 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 11:05:52,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2340, 1.7116, 1.4036, 1.4450], device='cuda:1'), covar=tensor([0.0891, 0.0290, 0.0304, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 11:05:55,075 INFO [train.py:968] (1/2) Epoch 10, batch 36250, giga_loss[loss=0.265, simple_loss=0.3536, pruned_loss=0.08821, over 28852.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3638, pruned_loss=0.1086, over 5699433.19 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3643, pruned_loss=0.1235, over 5722810.18 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3613, pruned_loss=0.1059, over 5692796.74 frames. ], batch size: 174, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:06:34,213 INFO [train.py:968] (1/2) Epoch 10, batch 36300, giga_loss[loss=0.2417, simple_loss=0.3285, pruned_loss=0.07743, over 29065.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3611, pruned_loss=0.1057, over 5696400.68 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3641, pruned_loss=0.1232, over 5716517.52 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3593, pruned_loss=0.1033, over 5696844.09 frames. ], batch size: 128, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:06:34,825 INFO [optim.py:369] (1/2) 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 11:06:45,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:07:14,730 INFO [train.py:968] (1/2) Epoch 10, batch 36350, giga_loss[loss=0.2629, simple_loss=0.3421, pruned_loss=0.09184, over 28631.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.36, pruned_loss=0.1051, over 5692857.58 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3644, pruned_loss=0.1234, over 5719715.59 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3583, pruned_loss=0.1028, over 5689968.41 frames. ], batch size: 119, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:07:42,983 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.08 vs. limit=2.0
+2023-03-05 11:07:44,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=445980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:07:56,595 INFO [train.py:968] (1/2) Epoch 10, batch 36400, giga_loss[loss=0.3151, simple_loss=0.3822, pruned_loss=0.124, over 28968.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3625, pruned_loss=0.1082, over 5691140.91 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.365, pruned_loss=0.1237, over 5721625.16 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3605, pruned_loss=0.1058, over 5686880.28 frames. ], batch size: 164, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:07:57,255 INFO [optim.py:369] (1/2) 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 11:08:01,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3337, 1.4242, 1.5466, 1.4596], device='cuda:1'), covar=tensor([0.0904, 0.0851, 0.1217, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0729, 0.0667, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 11:08:40,512 INFO [train.py:968] (1/2) Epoch 10, batch 36450, giga_loss[loss=0.2986, simple_loss=0.368, pruned_loss=0.1145, over 28350.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.366, pruned_loss=0.1129, over 5690794.38 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3663, pruned_loss=0.1246, over 5718737.88 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3632, pruned_loss=0.1098, over 5688906.88 frames. ], batch size: 65, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:08:54,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3300, 1.4181, 1.4359, 1.1685], device='cuda:1'), covar=tensor([0.1312, 0.1489, 0.0906, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.1657, 0.1540, 0.1514, 0.1630], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 11:09:21,301 INFO [train.py:968] (1/2) Epoch 10, batch 36500, giga_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1227, over 28599.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3676, pruned_loss=0.116, over 5692281.32 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3663, pruned_loss=0.1246, over 5722343.02 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3654, pruned_loss=0.1134, over 5687247.48 frames. ], batch size: 307, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:09:22,164 INFO [optim.py:369] (1/2) 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 11:09:29,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:09:37,235 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:10:06,908 INFO [train.py:968] (1/2) Epoch 10, batch 36550, giga_loss[loss=0.2865, simple_loss=0.3511, pruned_loss=0.111, over 27652.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3653, pruned_loss=0.1154, over 5690385.10 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3662, pruned_loss=0.1244, over 5721032.33 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3637, pruned_loss=0.1134, over 5687216.57 frames. ], batch size: 472, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:10:47,838 INFO [train.py:968] (1/2) Epoch 10, batch 36600, giga_loss[loss=0.3629, simple_loss=0.4135, pruned_loss=0.1561, over 28974.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3637, pruned_loss=0.1145, over 5701469.76 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3661, pruned_loss=0.1241, over 5724724.14 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3624, pruned_loss=0.1128, over 5694623.63 frames. ], batch size: 186, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:10:48,464 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 36650, giga_loss[loss=0.3012, simple_loss=0.3632, pruned_loss=0.1196, over 29019.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3618, pruned_loss=0.1126, over 5702202.62 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3662, pruned_loss=0.124, over 5722879.72 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3606, pruned_loss=0.1112, over 5697545.47 frames. ], batch size: 106, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:11:32,490 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:11:34,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=446249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:11:59,915 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:12:04,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:12:13,006 INFO [train.py:968] (1/2) Epoch 10, batch 36700, giga_loss[loss=0.2482, simple_loss=0.3264, pruned_loss=0.08497, over 28884.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3594, pruned_loss=0.11, over 5694126.23 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3667, pruned_loss=0.1242, over 5715144.36 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3579, pruned_loss=0.1084, over 5695766.70 frames. ], batch size: 227, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:12:15,116 INFO [optim.py:369] (1/2) 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 11:12:15,600 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-05 11:12:47,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4681, 2.0345, 1.3842, 0.8242], device='cuda:1'), covar=tensor([0.4202, 0.1937, 0.2898, 0.4120], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1467, 0.1465, 0.1250], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 11:12:57,257 INFO [train.py:968] (1/2) Epoch 10, batch 36750, giga_loss[loss=0.3096, simple_loss=0.3674, pruned_loss=0.1259, over 28004.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3556, pruned_loss=0.1079, over 5692993.00 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3668, pruned_loss=0.1243, over 5721023.35 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3541, pruned_loss=0.1061, over 5688713.84 frames. ], batch size: 412, lr: 3.19e-03, grad_scale: 2.0
+2023-03-05 11:13:06,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:13:32,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3088, 1.5594, 1.4083, 1.5063], device='cuda:1'), covar=tensor([0.0740, 0.0342, 0.0310, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 11:13:41,213 INFO [train.py:968] (1/2) Epoch 10, batch 36800, giga_loss[loss=0.2398, simple_loss=0.309, pruned_loss=0.08534, over 28792.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3498, pruned_loss=0.1044, over 5705796.38 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.367, pruned_loss=0.1243, over 5725697.20 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.348, pruned_loss=0.1023, over 5697548.26 frames. ], batch size: 99, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:13:42,581 INFO [optim.py:369] (1/2) 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 11:14:10,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6749, 1.7578, 1.6760, 1.6249], device='cuda:1'), covar=tensor([0.1324, 0.1840, 0.1886, 0.1584], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0728, 0.0666, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 11:14:12,080 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:14:16,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=446429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:14:31,744 INFO [train.py:968] (1/2) Epoch 10, batch 36850, giga_loss[loss=0.2281, simple_loss=0.3057, pruned_loss=0.07525, over 28901.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3436, pruned_loss=0.1014, over 5680715.54 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3674, pruned_loss=0.1247, over 5719451.39 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3414, pruned_loss=0.09893, over 5678467.85 frames. ], batch size: 145, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:14:44,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:14:54,647 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-05 11:15:07,280 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:09,723 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:15,427 INFO [train.py:968] (1/2) Epoch 10, batch 36900, libri_loss[loss=0.3136, simple_loss=0.3775, pruned_loss=0.1249, over 27543.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3415, pruned_loss=0.09983, over 5682276.89 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3676, pruned_loss=0.1248, over 5721899.74 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.339, pruned_loss=0.09727, over 5677780.72 frames. ], batch size: 115, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:15:17,567 INFO [optim.py:369] (1/2) 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 11:15:19,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:21,707 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:21,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=446501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:24,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0140, 1.3178, 1.0634, 0.1937], device='cuda:1'), covar=tensor([0.2617, 0.2070, 0.3618, 0.4404], device='cuda:1'), in_proj_covar=tensor([0.1522, 0.1457, 0.1456, 0.1248], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 11:15:26,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=5.32 vs. limit=5.0
+2023-03-05 11:15:45,924 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:15:56,618 INFO [train.py:968] (1/2) Epoch 10, batch 36950, giga_loss[loss=0.3215, simple_loss=0.3781, pruned_loss=0.1324, over 28855.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3423, pruned_loss=0.09995, over 5681396.99 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3678, pruned_loss=0.1246, over 5722368.15 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3396, pruned_loss=0.09749, over 5676714.36 frames. ], batch size: 99, lr: 3.19e-03, grad_scale: 4.0
+2023-03-05 11:15:58,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-05 11:16:04,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-05 11:16:34,633 INFO [train.py:968] (1/2) Epoch 10, batch 37000, giga_loss[loss=0.2739, simple_loss=0.3435, pruned_loss=0.1021, over 28957.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3446, pruned_loss=0.1012, over 5697278.95 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3697, pruned_loss=0.1256, over 5726044.36 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3397, pruned_loss=0.0974, over 5688582.58 frames. ], batch size: 213, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:16:35,939 INFO [optim.py:369] (1/2) 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 11:17:01,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=446626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:17:04,229 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 37050, giga_loss[loss=0.2425, simple_loss=0.3182, pruned_loss=0.08341, over 28962.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.344, pruned_loss=0.1016, over 5690494.29 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3705, pruned_loss=0.126, over 5720708.38 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3386, pruned_loss=0.09746, over 5687752.79 frames. ], batch size: 106, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:17:25,110 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 11:17:26,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=446658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:17:51,553 INFO [train.py:968] (1/2) Epoch 10, batch 37100, libri_loss[loss=0.3747, simple_loss=0.4246, pruned_loss=0.1625, over 18885.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3428, pruned_loss=0.1012, over 5681031.74 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3715, pruned_loss=0.1265, over 5706010.79 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3367, pruned_loss=0.09663, over 5691698.54 frames. ], batch size: 187, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:17:52,844 INFO [optim.py:369] (1/2) 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 11:18:21,642 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:18:29,813 INFO [train.py:968] (1/2) Epoch 10, batch 37150, giga_loss[loss=0.2523, simple_loss=0.3307, pruned_loss=0.08691, over 28718.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3392, pruned_loss=0.09895, over 5686167.76 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.372, pruned_loss=0.1268, over 5698283.87 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3335, pruned_loss=0.0947, over 5700459.24 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:18:30,839 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=446745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:18:59,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3279, 1.6456, 1.4400, 1.4356], device='cuda:1'), covar=tensor([0.0736, 0.0331, 0.0302, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0056, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 11:19:08,311 INFO [train.py:968] (1/2) Epoch 10, batch 37200, giga_loss[loss=0.2368, simple_loss=0.318, pruned_loss=0.07782, over 28742.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3358, pruned_loss=0.09711, over 5689946.21 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3721, pruned_loss=0.1268, over 5690109.78 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3309, pruned_loss=0.09359, over 5708379.68 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:19:09,475 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 37250, libri_loss[loss=0.3796, simple_loss=0.414, pruned_loss=0.1727, over 29567.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3357, pruned_loss=0.09795, over 5686085.01 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3735, pruned_loss=0.1277, over 5687632.69 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3294, pruned_loss=0.09331, over 5703287.71 frames. ], batch size: 74, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:19:58,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:20:12,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=446876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:20:25,618 INFO [train.py:968] (1/2) Epoch 10, batch 37300, libri_loss[loss=0.3095, simple_loss=0.3671, pruned_loss=0.1259, over 29670.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3333, pruned_loss=0.09648, over 5690287.55 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3743, pruned_loss=0.1281, over 5684851.69 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3265, pruned_loss=0.09158, over 5707361.06 frames. ], batch size: 69, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:20:27,487 INFO [optim.py:369] (1/2) 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 11:21:01,197 INFO [train.py:968] (1/2) Epoch 10, batch 37350, giga_loss[loss=0.2457, simple_loss=0.314, pruned_loss=0.08868, over 28671.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3331, pruned_loss=0.09622, over 5692758.71 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3748, pruned_loss=0.128, over 5680465.50 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3245, pruned_loss=0.09043, over 5712042.41 frames. ], batch size: 92, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:21:38,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-05 11:21:40,337 INFO [train.py:968] (1/2) Epoch 10, batch 37400, giga_loss[loss=0.2645, simple_loss=0.3394, pruned_loss=0.09478, over 28762.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3312, pruned_loss=0.09482, over 5706148.86 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3751, pruned_loss=0.1278, over 5684699.37 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.323, pruned_loss=0.08956, over 5717915.12 frames. ], batch size: 119, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:21:43,315 INFO [optim.py:369] (1/2) 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 11:21:48,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:21:50,675 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:00,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:02,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:11,933 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:12,539 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:18,916 INFO [train.py:968] (1/2) Epoch 10, batch 37450, giga_loss[loss=0.2163, simple_loss=0.2896, pruned_loss=0.07149, over 28532.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3292, pruned_loss=0.09381, over 5713956.07 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3753, pruned_loss=0.1279, over 5686373.88 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3214, pruned_loss=0.08869, over 5722420.51 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:22:19,106 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:23,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:22:33,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-05 11:22:51,687 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 37500, giga_loss[loss=0.2612, simple_loss=0.3302, pruned_loss=0.09604, over 28794.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.33, pruned_loss=0.09455, over 5715096.36 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3746, pruned_loss=0.1272, over 5690795.05 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3234, pruned_loss=0.0903, over 5718719.75 frames. ], batch size: 99, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:23:00,606 INFO [optim.py:369] (1/2) 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 11:23:09,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:23:18,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5130, 2.1577, 1.8630, 1.6600], device='cuda:1'), covar=tensor([0.0747, 0.0254, 0.0268, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 11:23:18,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:23:38,301 INFO [train.py:968] (1/2) Epoch 10, batch 37550, giga_loss[loss=0.259, simple_loss=0.3321, pruned_loss=0.09294, over 28757.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3362, pruned_loss=0.09843, over 5705631.13 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3749, pruned_loss=0.1272, over 5687210.38 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3292, pruned_loss=0.0941, over 5712893.62 frames. ], batch size: 99, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:24:13,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4970, 1.6361, 1.3250, 1.7811], device='cuda:1'), covar=tensor([0.2428, 0.2306, 0.2505, 0.2183], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.0948, 0.1132, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 11:24:21,171 INFO [train.py:968] (1/2) Epoch 10, batch 37600, giga_loss[loss=0.3397, simple_loss=0.4011, pruned_loss=0.1391, over 28751.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.342, pruned_loss=0.1021, over 5706415.68 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3747, pruned_loss=0.127, over 5692771.15 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3355, pruned_loss=0.09793, over 5707445.50 frames. ], batch size: 284, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:24:25,911 INFO [optim.py:369] (1/2) 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 11:24:35,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:09,939 INFO [train.py:968] (1/2) Epoch 10, batch 37650, giga_loss[loss=0.3398, simple_loss=0.3988, pruned_loss=0.1404, over 27913.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3504, pruned_loss=0.1078, over 5693489.67 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3749, pruned_loss=0.127, over 5694912.06 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3449, pruned_loss=0.1043, over 5692484.41 frames. ], batch size: 412, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:25:17,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:19,393 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:25,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2293, 1.3705, 1.5002, 1.3827], device='cuda:1'), covar=tensor([0.1090, 0.0945, 0.1319, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0730, 0.0667, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 11:25:28,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:31,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:25:41,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4796, 2.1857, 1.5560, 0.7584], device='cuda:1'), covar=tensor([0.4132, 0.2070, 0.2803, 0.4305], device='cuda:1'), in_proj_covar=tensor([0.1526, 0.1465, 0.1466, 0.1247], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 11:25:49,404 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:26:00,548 INFO [train.py:968] (1/2) Epoch 10, batch 37700, giga_loss[loss=0.3158, simple_loss=0.3848, pruned_loss=0.1234, over 29091.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3551, pruned_loss=0.1097, over 5679642.76 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3752, pruned_loss=0.1272, over 5696743.31 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.35, pruned_loss=0.1065, over 5677032.96 frames. ], batch size: 155, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:26:01,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:26:04,164 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:26:44,611 INFO [train.py:968] (1/2) Epoch 10, batch 37750, giga_loss[loss=0.3048, simple_loss=0.3805, pruned_loss=0.1146, over 28862.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3593, pruned_loss=0.1115, over 5676272.58 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3755, pruned_loss=0.1275, over 5692113.23 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3545, pruned_loss=0.1081, over 5678071.46 frames. ], batch size: 174, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:27:31,167 INFO [train.py:968] (1/2) Epoch 10, batch 37800, giga_loss[loss=0.3368, simple_loss=0.4022, pruned_loss=0.1357, over 28729.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3647, pruned_loss=0.1147, over 5656873.81 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3758, pruned_loss=0.1278, over 5675027.04 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3603, pruned_loss=0.1115, over 5672604.19 frames. ], batch size: 85, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:27:34,694 INFO [optim.py:369] (1/2) 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 11:27:38,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1424, 3.9642, 3.7269, 1.6937], device='cuda:1'), covar=tensor([0.0559, 0.0650, 0.0648, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1000, 0.0939, 0.0822, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 11:27:39,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-05 11:27:45,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:27:53,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:28:13,737 INFO [train.py:968] (1/2) Epoch 10, batch 37850, giga_loss[loss=0.2316, simple_loss=0.3149, pruned_loss=0.07417, over 28243.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3643, pruned_loss=0.1144, over 5661075.57 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3756, pruned_loss=0.1276, over 5680136.60 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3608, pruned_loss=0.1117, over 5668831.00 frames. ], batch size: 368, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:28:27,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:28:54,911 INFO [train.py:968] (1/2) Epoch 10, batch 37900, giga_loss[loss=0.3335, simple_loss=0.3924, pruned_loss=0.1373, over 28718.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3588, pruned_loss=0.1098, over 5675674.88 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3759, pruned_loss=0.1279, over 5683311.33 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3555, pruned_loss=0.1071, over 5678723.82 frames. ], batch size: 78, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:28:58,103 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 37950, giga_loss[loss=0.2546, simple_loss=0.3361, pruned_loss=0.0865, over 29039.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.358, pruned_loss=0.1086, over 5679529.54 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3763, pruned_loss=0.1283, over 5685395.62 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3546, pruned_loss=0.1059, over 5679960.32 frames. ], batch size: 155, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:29:43,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4369, 1.6428, 1.3778, 1.4891], device='cuda:1'), covar=tensor([0.0754, 0.0293, 0.0300, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 11:29:48,711 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:29:51,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:29:56,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:29:58,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:14,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=447582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:16,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:17,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2865, 1.5002, 1.5305, 1.1279], device='cuda:1'), covar=tensor([0.1060, 0.2079, 0.0941, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0693, 0.0860, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 11:30:19,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 11:30:22,843 INFO [train.py:968] (1/2) Epoch 10, batch 38000, giga_loss[loss=0.2905, simple_loss=0.3639, pruned_loss=0.1085, over 28737.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3578, pruned_loss=0.1084, over 5685293.95 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1281, over 5687772.46 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3552, pruned_loss=0.1062, over 5683483.62 frames. ], batch size: 119, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:30:23,415 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:26,179 INFO [optim.py:369] (1/2) 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 11:30:31,565 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:30:33,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:31:00,380 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:31:06,760 INFO [train.py:968] (1/2) Epoch 10, batch 38050, giga_loss[loss=0.322, simple_loss=0.3933, pruned_loss=0.1254, over 28230.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3601, pruned_loss=0.1096, over 5687616.92 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1282, over 5689903.31 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.358, pruned_loss=0.1077, over 5684318.75 frames. ], batch size: 368, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:31:14,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5614, 1.7511, 1.8738, 1.3681], device='cuda:1'), covar=tensor([0.1485, 0.2129, 0.1190, 0.1442], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0692, 0.0857, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 11:31:35,438 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 38100, giga_loss[loss=0.2748, simple_loss=0.3548, pruned_loss=0.09736, over 28949.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.362, pruned_loss=0.1112, over 5690539.46 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3758, pruned_loss=0.1281, over 5694223.05 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3602, pruned_loss=0.1095, over 5684187.68 frames. ], batch size: 174, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:31:54,654 INFO [optim.py:369] (1/2) 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 11:32:17,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:32:19,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:32:35,158 INFO [train.py:968] (1/2) Epoch 10, batch 38150, libri_loss[loss=0.357, simple_loss=0.4145, pruned_loss=0.1497, over 28757.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3631, pruned_loss=0.1124, over 5696644.24 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.3761, pruned_loss=0.1284, over 5697989.34 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3611, pruned_loss=0.1104, over 5688213.52 frames. ], batch size: 106, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:32:36,079 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=447745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:32:45,502 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:33:06,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-05 11:33:11,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7343, 2.2112, 2.0067, 1.5503], device='cuda:1'), covar=tensor([0.1648, 0.2103, 0.1339, 0.1605], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0694, 0.0857, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 11:33:14,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7189, 2.6156, 1.5619, 0.8941], device='cuda:1'), covar=tensor([0.4826, 0.2263, 0.3179, 0.4746], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1475, 0.1478, 0.1252], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 11:33:18,305 INFO [train.py:968] (1/2) Epoch 10, batch 38200, giga_loss[loss=0.2484, simple_loss=0.328, pruned_loss=0.08445, over 28655.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3636, pruned_loss=0.1131, over 5687583.93 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3767, pruned_loss=0.1288, over 5693271.63 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3612, pruned_loss=0.1107, over 5685872.73 frames. ], batch size: 60, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:33:22,354 INFO [optim.py:369] (1/2) 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 11:33:40,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=447820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:33:40,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-05 11:33:42,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=447823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:34:01,296 INFO [train.py:968] (1/2) Epoch 10, batch 38250, giga_loss[loss=0.2812, simple_loss=0.3541, pruned_loss=0.1042, over 28536.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.364, pruned_loss=0.1132, over 5696388.78 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3772, pruned_loss=0.1292, over 5694014.87 frames. ], giga_tot_loss[loss=0.2915, simple_loss=0.3614, pruned_loss=0.1108, over 5694337.23 frames. ], batch size: 78, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:34:10,655 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=447852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:34:43,336 INFO [train.py:968] (1/2) Epoch 10, batch 38300, giga_loss[loss=0.2705, simple_loss=0.3532, pruned_loss=0.09392, over 28808.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3639, pruned_loss=0.1121, over 5701605.81 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3775, pruned_loss=0.1293, over 5695433.66 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3614, pruned_loss=0.1097, over 5698690.95 frames. ], batch size: 119, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:34:48,224 INFO [optim.py:369] (1/2) 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 11:35:24,773 INFO [train.py:968] (1/2) Epoch 10, batch 38350, giga_loss[loss=0.2735, simple_loss=0.3514, pruned_loss=0.09781, over 29033.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.364, pruned_loss=0.111, over 5696648.79 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3772, pruned_loss=0.1291, over 5687114.89 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3619, pruned_loss=0.1088, over 5703075.72 frames. ], batch size: 106, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:35:53,839 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-05 11:36:04,000 INFO [train.py:968] (1/2) Epoch 10, batch 38400, giga_loss[loss=0.301, simple_loss=0.3695, pruned_loss=0.1163, over 28863.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.364, pruned_loss=0.1107, over 5694761.94 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3766, pruned_loss=0.1287, over 5685192.51 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3625, pruned_loss=0.1088, over 5701663.71 frames. ], batch size: 99, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:36:07,592 INFO [optim.py:369] (1/2) 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 11:36:27,538 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:36:31,330 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 11:36:32,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2699, 1.6783, 1.4123, 1.5160], device='cuda:1'), covar=tensor([0.0748, 0.0353, 0.0313, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0050, 0.0086], device='cuda:1')
+2023-03-05 11:36:47,408 INFO [train.py:968] (1/2) Epoch 10, batch 38450, giga_loss[loss=0.2651, simple_loss=0.34, pruned_loss=0.09512, over 28919.00 frames. ], tot_loss[loss=0.289, simple_loss=0.361, pruned_loss=0.1085, over 5698334.33 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3765, pruned_loss=0.1286, over 5686447.55 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3598, pruned_loss=0.107, over 5702694.98 frames. ], batch size: 199, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:37:01,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1230, 1.9117, 1.5497, 1.5831], device='cuda:1'), covar=tensor([0.0742, 0.0717, 0.0913, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0434, 0.0498, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 11:37:27,466 INFO [train.py:968] (1/2) Epoch 10, batch 38500, giga_loss[loss=0.251, simple_loss=0.335, pruned_loss=0.08345, over 28940.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3589, pruned_loss=0.1074, over 5707701.06 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3767, pruned_loss=0.1286, over 5691001.34 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3575, pruned_loss=0.1058, over 5707352.04 frames. ], batch size: 164, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:37:32,107 INFO [optim.py:369] (1/2) 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 11:37:49,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=448120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:38:06,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3384, 1.9748, 1.4949, 0.5693], device='cuda:1'), covar=tensor([0.3746, 0.1891, 0.2942, 0.4194], device='cuda:1'), in_proj_covar=tensor([0.1540, 0.1461, 0.1464, 0.1249], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 11:38:07,091 INFO [train.py:968] (1/2) Epoch 10, batch 38550, giga_loss[loss=0.2673, simple_loss=0.3385, pruned_loss=0.09802, over 28570.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3579, pruned_loss=0.107, over 5707001.97 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3771, pruned_loss=0.1288, over 5683043.19 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3559, pruned_loss=0.105, over 5714346.26 frames. ], batch size: 78, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:38:29,975 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 11:38:45,389 INFO [train.py:968] (1/2) Epoch 10, batch 38600, giga_loss[loss=0.2782, simple_loss=0.3545, pruned_loss=0.1009, over 28943.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3585, pruned_loss=0.1079, over 5689802.27 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3773, pruned_loss=0.129, over 5668341.51 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3563, pruned_loss=0.1057, over 5709989.96 frames. ], batch size: 174, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:38:50,583 INFO [optim.py:369] (1/2) 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 11:39:24,748 INFO [train.py:968] (1/2) Epoch 10, batch 38650, giga_loss[loss=0.2726, simple_loss=0.3443, pruned_loss=0.1005, over 28654.00 frames. ], tot_loss[loss=0.288, simple_loss=0.359, pruned_loss=0.1085, over 5695083.10 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3776, pruned_loss=0.1292, over 5672957.60 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3567, pruned_loss=0.1062, over 5707467.22 frames. ], batch size: 92, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:39:40,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:39:42,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=448266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:40:03,574 INFO [train.py:968] (1/2) Epoch 10, batch 38700, giga_loss[loss=0.2744, simple_loss=0.353, pruned_loss=0.09794, over 28569.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3591, pruned_loss=0.1079, over 5693444.62 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3778, pruned_loss=0.1294, over 5667679.00 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3568, pruned_loss=0.1055, over 5708017.65 frames. ], batch size: 78, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:40:04,498 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:40:20,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6903, 1.7534, 1.3192, 1.3630], device='cuda:1'), covar=tensor([0.0788, 0.0577, 0.0970, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0435, 0.0499, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 11:40:42,598 INFO [train.py:968] (1/2) Epoch 10, batch 38750, giga_loss[loss=0.2807, simple_loss=0.3533, pruned_loss=0.104, over 28637.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3568, pruned_loss=0.1055, over 5697772.56 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3776, pruned_loss=0.1293, over 5667489.60 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3551, pruned_loss=0.1037, over 5709442.82 frames. ], batch size: 92, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:41:03,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2917, 1.5850, 1.3225, 1.0225], device='cuda:1'), covar=tensor([0.2159, 0.2023, 0.2289, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.1267, 0.0943, 0.1123, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 11:41:24,307 INFO [train.py:968] (1/2) Epoch 10, batch 38800, giga_loss[loss=0.2682, simple_loss=0.3236, pruned_loss=0.1064, over 23772.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3552, pruned_loss=0.1047, over 5688807.88 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3777, pruned_loss=0.1292, over 5657950.20 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3534, pruned_loss=0.1029, over 5707636.53 frames. ], batch size: 705, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:41:24,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=448394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:41:28,050 INFO [optim.py:369] (1/2) 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 11:42:01,424 INFO [train.py:968] (1/2) Epoch 10, batch 38850, giga_loss[loss=0.3016, simple_loss=0.3663, pruned_loss=0.1184, over 28468.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3541, pruned_loss=0.1048, over 5692902.69 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3781, pruned_loss=0.1296, over 5657334.52 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3517, pruned_loss=0.1023, over 5709914.61 frames. ], batch size: 65, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:42:41,733 INFO [train.py:968] (1/2) Epoch 10, batch 38900, libri_loss[loss=0.3195, simple_loss=0.3868, pruned_loss=0.1261, over 29538.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3519, pruned_loss=0.104, over 5695629.61 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3781, pruned_loss=0.1297, over 5664684.52 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3493, pruned_loss=0.1013, over 5703559.94 frames. ], batch size: 82, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:42:46,712 INFO [optim.py:369] (1/2) 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 11:43:13,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:43:15,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=448540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:43:18,287 INFO [train.py:968] (1/2) Epoch 10, batch 38950, giga_loss[loss=0.2763, simple_loss=0.3381, pruned_loss=0.1072, over 28462.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3491, pruned_loss=0.1027, over 5698645.16 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.378, pruned_loss=0.1296, over 5668104.70 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3466, pruned_loss=0.1001, over 5702606.96 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:43:25,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0711, 2.5375, 1.1312, 1.2961], device='cuda:1'), covar=tensor([0.1012, 0.0407, 0.0927, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0492, 0.0330, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0024], device='cuda:1')
+2023-03-05 11:43:41,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=448569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:43:58,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1658, 0.8295, 0.8873, 1.4052], device='cuda:1'), covar=tensor([0.0759, 0.0340, 0.0343, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0086], device='cuda:1')
+2023-03-05 11:44:03,044 INFO [train.py:968] (1/2) Epoch 10, batch 39000, giga_loss[loss=0.2297, simple_loss=0.3026, pruned_loss=0.07841, over 28701.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3499, pruned_loss=0.1036, over 5696521.96 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3777, pruned_loss=0.1294, over 5670512.97 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3475, pruned_loss=0.1011, over 5698241.02 frames. ], batch size: 85, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:44:03,044 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 11:44:11,824 INFO [train.py:1012] (1/2) Epoch 10, validation: loss=0.2204, simple_loss=0.3259, pruned_loss=0.05747, over 944034.00 frames. 
+2023-03-05 11:44:11,825 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 11:44:17,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2664, 1.4054, 1.2367, 1.4438], device='cuda:1'), covar=tensor([0.0664, 0.0409, 0.0335, 0.0716], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0078, 0.0055, 0.0050, 0.0085], device='cuda:1')
+2023-03-05 11:44:17,389 INFO [optim.py:369] (1/2) 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 11:44:19,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2729, 1.4327, 1.4933, 1.3496], device='cuda:1'), covar=tensor([0.1339, 0.1521, 0.1834, 0.1546], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0726, 0.0662, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 11:44:48,368 INFO [train.py:968] (1/2) Epoch 10, batch 39050, giga_loss[loss=0.3133, simple_loss=0.374, pruned_loss=0.1263, over 28854.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3482, pruned_loss=0.1031, over 5701672.01 frames. ], libri_tot_loss[loss=0.3187, simple_loss=0.3781, pruned_loss=0.1296, over 5674732.57 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3453, pruned_loss=0.1002, over 5699850.79 frames. ], batch size: 145, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:45:29,889 INFO [train.py:968] (1/2) Epoch 10, batch 39100, giga_loss[loss=0.2833, simple_loss=0.3493, pruned_loss=0.1086, over 28911.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3454, pruned_loss=0.1019, over 5706899.06 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.3782, pruned_loss=0.1297, over 5675465.92 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3426, pruned_loss=0.0993, over 5705305.71 frames. ], batch size: 227, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:45:35,566 INFO [optim.py:369] (1/2) 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 11:46:07,657 INFO [train.py:968] (1/2) Epoch 10, batch 39150, giga_loss[loss=0.2534, simple_loss=0.3205, pruned_loss=0.09312, over 28673.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3436, pruned_loss=0.1014, over 5699987.38 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3785, pruned_loss=0.1299, over 5668611.81 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3405, pruned_loss=0.09857, over 5705971.51 frames. ], batch size: 92, lr: 3.18e-03, grad_scale: 2.0
+2023-03-05 11:46:41,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9457, 1.0551, 3.3758, 2.9576], device='cuda:1'), covar=tensor([0.1683, 0.2640, 0.0456, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0640, 0.0574, 0.0837, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 11:46:49,057 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 11:46:50,598 INFO [train.py:968] (1/2) Epoch 10, batch 39200, giga_loss[loss=0.2481, simple_loss=0.3227, pruned_loss=0.08682, over 28893.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3412, pruned_loss=0.0999, over 5700524.92 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3786, pruned_loss=0.13, over 5667274.58 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3384, pruned_loss=0.09738, over 5706387.72 frames. ], batch size: 112, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:46:56,061 INFO [optim.py:369] (1/2) 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 11:47:29,672 INFO [train.py:968] (1/2) Epoch 10, batch 39250, giga_loss[loss=0.2808, simple_loss=0.3597, pruned_loss=0.1009, over 28806.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.344, pruned_loss=0.1019, over 5699553.62 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3787, pruned_loss=0.1303, over 5670153.56 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3407, pruned_loss=0.09881, over 5702341.14 frames. ], batch size: 243, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:47:59,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3994, 4.1874, 3.9962, 1.7490], device='cuda:1'), covar=tensor([0.0470, 0.0660, 0.0649, 0.2200], device='cuda:1'), in_proj_covar=tensor([0.1014, 0.0948, 0.0829, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 11:48:11,675 INFO [train.py:968] (1/2) Epoch 10, batch 39300, giga_loss[loss=0.302, simple_loss=0.3691, pruned_loss=0.1175, over 28846.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3466, pruned_loss=0.1029, over 5691193.36 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3788, pruned_loss=0.1304, over 5666611.35 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3428, pruned_loss=0.09944, over 5697365.85 frames. ], batch size: 112, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:48:16,900 INFO [optim.py:369] (1/2) 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 11:48:19,336 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448904.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 11:48:24,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3852, 1.6455, 1.3329, 1.5230], device='cuda:1'), covar=tensor([0.2184, 0.2097, 0.2242, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.0945, 0.1125, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 11:48:58,189 INFO [train.py:968] (1/2) Epoch 10, batch 39350, giga_loss[loss=0.2813, simple_loss=0.3624, pruned_loss=0.1001, over 28298.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3492, pruned_loss=0.1037, over 5684355.36 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3785, pruned_loss=0.1303, over 5660511.57 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3462, pruned_loss=0.1008, over 5694159.33 frames. ], batch size: 368, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:49:12,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:49:24,328 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=448977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:49:40,087 INFO [train.py:968] (1/2) Epoch 10, batch 39400, giga_loss[loss=0.343, simple_loss=0.4185, pruned_loss=0.1338, over 28716.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3518, pruned_loss=0.105, over 5687546.72 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3784, pruned_loss=0.1303, over 5667964.75 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3486, pruned_loss=0.1019, over 5689513.13 frames. ], batch size: 262, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:49:48,011 INFO [optim.py:369] (1/2) 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 11:49:48,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:49:52,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3913, 1.6092, 1.2285, 1.6045], device='cuda:1'), covar=tensor([0.2443, 0.2439, 0.2712, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.0953, 0.1134, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 11:50:22,264 INFO [train.py:968] (1/2) Epoch 10, batch 39450, giga_loss[loss=0.2686, simple_loss=0.341, pruned_loss=0.09811, over 29036.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3508, pruned_loss=0.1039, over 5694380.27 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3784, pruned_loss=0.1303, over 5669944.27 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3476, pruned_loss=0.1008, over 5694687.37 frames. ], batch size: 128, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:50:49,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9047, 1.0818, 1.0457, 0.8211], device='cuda:1'), covar=tensor([0.1735, 0.1848, 0.1114, 0.1739], device='cuda:1'), in_proj_covar=tensor([0.1643, 0.1552, 0.1525, 0.1635], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 11:51:02,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 11:51:02,546 INFO [train.py:968] (1/2) Epoch 10, batch 39500, giga_loss[loss=0.2895, simple_loss=0.3467, pruned_loss=0.1161, over 28434.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3513, pruned_loss=0.1045, over 5685681.91 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3789, pruned_loss=0.1306, over 5657171.09 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3478, pruned_loss=0.1013, over 5698334.05 frames. ], batch size: 71, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:51:08,875 INFO [optim.py:369] (1/2) 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 11:51:23,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2187, 0.9656, 4.1432, 3.2049], device='cuda:1'), covar=tensor([0.1684, 0.2835, 0.0401, 0.0771], device='cuda:1'), in_proj_covar=tensor([0.0641, 0.0576, 0.0839, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 11:51:44,543 INFO [train.py:968] (1/2) Epoch 10, batch 39550, giga_loss[loss=0.2783, simple_loss=0.3491, pruned_loss=0.1038, over 29075.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3513, pruned_loss=0.1044, over 5688114.64 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3792, pruned_loss=0.1307, over 5652997.65 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3475, pruned_loss=0.1011, over 5703283.58 frames. ], batch size: 155, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:52:27,382 INFO [train.py:968] (1/2) Epoch 10, batch 39600, giga_loss[loss=0.2804, simple_loss=0.3608, pruned_loss=0.1, over 28948.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3526, pruned_loss=0.1052, over 5703029.91 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3792, pruned_loss=0.1307, over 5656971.45 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3491, pruned_loss=0.1021, over 5712343.67 frames. ], batch size: 164, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:52:29,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:52:32,941 INFO [optim.py:369] (1/2) 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 11:53:09,495 INFO [train.py:968] (1/2) Epoch 10, batch 39650, giga_loss[loss=0.2952, simple_loss=0.3699, pruned_loss=0.1102, over 28574.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3567, pruned_loss=0.1074, over 5685842.23 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3801, pruned_loss=0.1312, over 5641082.05 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3527, pruned_loss=0.104, over 5707236.31 frames. ], batch size: 336, lr: 3.18e-03, grad_scale: 8.0
+2023-03-05 11:53:37,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449279.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 11:53:49,315 INFO [train.py:968] (1/2) Epoch 10, batch 39700, giga_loss[loss=0.3163, simple_loss=0.3766, pruned_loss=0.128, over 29006.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3603, pruned_loss=0.1098, over 5695322.50 frames. ], libri_tot_loss[loss=0.3216, simple_loss=0.3803, pruned_loss=0.1315, over 5645171.28 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3567, pruned_loss=0.1065, over 5709127.88 frames. ], batch size: 136, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:53:56,185 INFO [optim.py:369] (1/2) 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 11:54:14,082 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:54:24,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:54:30,040 INFO [train.py:968] (1/2) Epoch 10, batch 39750, giga_loss[loss=0.2707, simple_loss=0.3448, pruned_loss=0.09832, over 28414.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3614, pruned_loss=0.1104, over 5695361.02 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3804, pruned_loss=0.1316, over 5645544.44 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3581, pruned_loss=0.1075, over 5706878.89 frames. ], batch size: 60, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:54:37,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:54:59,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:55:12,933 INFO [train.py:968] (1/2) Epoch 10, batch 39800, libri_loss[loss=0.3881, simple_loss=0.4317, pruned_loss=0.1722, over 27643.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3618, pruned_loss=0.1105, over 5698502.38 frames. ], libri_tot_loss[loss=0.3223, simple_loss=0.3806, pruned_loss=0.1319, over 5647978.90 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3587, pruned_loss=0.1075, over 5706378.91 frames. ], batch size: 116, lr: 3.18e-03, grad_scale: 4.0
+2023-03-05 11:55:18,585 INFO [optim.py:369] (1/2) 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 11:55:35,495 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449422.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 11:55:37,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449425.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 11:55:51,803 INFO [train.py:968] (1/2) Epoch 10, batch 39850, giga_loss[loss=0.2584, simple_loss=0.3354, pruned_loss=0.09076, over 28504.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3624, pruned_loss=0.1107, over 5698869.28 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3809, pruned_loss=0.132, over 5651335.74 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3594, pruned_loss=0.108, over 5702821.89 frames. ], batch size: 85, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 11:55:55,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-05 11:56:00,161 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449454.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 11:56:11,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4740, 1.4008, 1.1352, 1.0497], device='cuda:1'), covar=tensor([0.0640, 0.0519, 0.0948, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0436, 0.0496, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 11:56:21,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:22,949 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:29,741 INFO [train.py:968] (1/2) Epoch 10, batch 39900, giga_loss[loss=0.2497, simple_loss=0.3247, pruned_loss=0.08734, over 28688.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3606, pruned_loss=0.1096, over 5712130.59 frames. ], libri_tot_loss[loss=0.3224, simple_loss=0.3809, pruned_loss=0.132, over 5658101.18 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3575, pruned_loss=0.1067, over 5710742.43 frames. ], batch size: 92, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 11:56:30,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2706, 1.6757, 1.2012, 0.6505], device='cuda:1'), covar=tensor([0.3023, 0.1644, 0.2219, 0.3882], device='cuda:1'), in_proj_covar=tensor([0.1536, 0.1446, 0.1466, 0.1243], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 11:56:30,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:32,880 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:56:42,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:49,479 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:51,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:56:53,815 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:57:08,017 INFO [train.py:968] (1/2) Epoch 10, batch 39950, giga_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.08734, over 28896.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3578, pruned_loss=0.1084, over 5718853.31 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.381, pruned_loss=0.132, over 5665347.26 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3546, pruned_loss=0.1053, over 5713104.10 frames. ], batch size: 136, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 11:57:09,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4666, 1.4278, 3.7496, 3.3387], device='cuda:1'), covar=tensor([0.1716, 0.2697, 0.0717, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0641, 0.0571, 0.0839, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 11:57:16,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:57:32,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:57:51,660 INFO [train.py:968] (1/2) Epoch 10, batch 40000, giga_loss[loss=0.2344, simple_loss=0.3128, pruned_loss=0.07794, over 28712.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3535, pruned_loss=0.1058, over 5713738.45 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.3807, pruned_loss=0.1318, over 5666368.52 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3511, pruned_loss=0.1034, over 5708564.74 frames. ], batch size: 119, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 11:57:55,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4661, 1.5929, 1.3797, 1.2881], device='cuda:1'), covar=tensor([0.2068, 0.1642, 0.1385, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.1653, 0.1568, 0.1541, 0.1637], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 11:58:00,526 INFO [optim.py:369] (1/2) 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 11:58:31,397 INFO [train.py:968] (1/2) Epoch 10, batch 40050, giga_loss[loss=0.256, simple_loss=0.3425, pruned_loss=0.08474, over 28880.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3533, pruned_loss=0.105, over 5721267.86 frames. ], libri_tot_loss[loss=0.3227, simple_loss=0.3813, pruned_loss=0.1321, over 5671180.52 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3503, pruned_loss=0.1022, over 5714112.37 frames. ], batch size: 186, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 11:58:56,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.49 vs. limit=5.0
+2023-03-05 11:59:12,465 INFO [train.py:968] (1/2) Epoch 10, batch 40100, giga_loss[loss=0.2703, simple_loss=0.3428, pruned_loss=0.09885, over 28712.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3559, pruned_loss=0.1046, over 5713897.54 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3816, pruned_loss=0.1323, over 5671314.85 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3529, pruned_loss=0.1018, over 5708917.15 frames. ], batch size: 119, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 11:59:19,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=449699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:59:23,665 INFO [optim.py:369] (1/2) 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 11:59:30,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:59:32,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2872, 3.0453, 1.4003, 1.4019], device='cuda:1'), covar=tensor([0.0898, 0.0293, 0.0914, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0501, 0.0333, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 11:59:33,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 11:59:42,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6667, 1.6593, 1.7712, 1.4220], device='cuda:1'), covar=tensor([0.1194, 0.1699, 0.1588, 0.1554], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0724, 0.0662, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 11:59:53,944 INFO [train.py:968] (1/2) Epoch 10, batch 40150, giga_loss[loss=0.2772, simple_loss=0.3353, pruned_loss=0.1096, over 28496.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3554, pruned_loss=0.1045, over 5705884.31 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3816, pruned_loss=0.1324, over 5669749.18 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3521, pruned_loss=0.1014, over 5705257.85 frames. ], batch size: 78, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 11:59:56,479 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:00:32,647 INFO [train.py:968] (1/2) Epoch 10, batch 40200, giga_loss[loss=0.2589, simple_loss=0.322, pruned_loss=0.0979, over 28956.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3542, pruned_loss=0.1046, over 5713595.97 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3816, pruned_loss=0.1321, over 5674279.36 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3511, pruned_loss=0.1018, over 5709752.09 frames. ], batch size: 106, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:00:43,432 INFO [optim.py:369] (1/2) 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 12:01:12,227 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=449842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:01:13,299 INFO [train.py:968] (1/2) Epoch 10, batch 40250, giga_loss[loss=0.2668, simple_loss=0.3338, pruned_loss=0.09993, over 28857.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3527, pruned_loss=0.1053, over 5710052.74 frames. ], libri_tot_loss[loss=0.3229, simple_loss=0.3816, pruned_loss=0.1321, over 5676070.60 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3498, pruned_loss=0.1027, over 5705924.42 frames. ], batch size: 199, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:01:14,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=449845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:01:40,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=449874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:01:58,457 INFO [train.py:968] (1/2) Epoch 10, batch 40300, libri_loss[loss=0.2823, simple_loss=0.3461, pruned_loss=0.1092, over 29572.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3521, pruned_loss=0.1063, over 5704886.87 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.382, pruned_loss=0.1323, over 5670861.34 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3491, pruned_loss=0.1036, over 5706586.95 frames. ], batch size: 75, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:02:06,251 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=449923.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:02:37,659 INFO [train.py:968] (1/2) Epoch 10, batch 40350, giga_loss[loss=0.2557, simple_loss=0.3358, pruned_loss=0.08777, over 28803.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3512, pruned_loss=0.1065, over 5711766.66 frames. ], libri_tot_loss[loss=0.3236, simple_loss=0.3822, pruned_loss=0.1325, over 5672800.53 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3477, pruned_loss=0.1035, over 5712693.83 frames. ], batch size: 174, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:02:37,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3997, 1.7354, 1.4637, 1.2434], device='cuda:1'), covar=tensor([0.1798, 0.1333, 0.1179, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.1660, 0.1559, 0.1540, 0.1640], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:03:14,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3149, 2.1856, 2.0518, 1.8731], device='cuda:1'), covar=tensor([0.1394, 0.2337, 0.1866, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0731, 0.0668, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 12:03:21,331 INFO [train.py:968] (1/2) Epoch 10, batch 40400, giga_loss[loss=0.2661, simple_loss=0.3379, pruned_loss=0.0971, over 28907.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3486, pruned_loss=0.1046, over 5716761.04 frames. ], libri_tot_loss[loss=0.3237, simple_loss=0.3824, pruned_loss=0.1325, over 5674283.27 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3454, pruned_loss=0.102, over 5716514.48 frames. ], batch size: 227, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:03:28,644 INFO [optim.py:369] (1/2) 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 12:03:42,044 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 12:03:59,401 INFO [train.py:968] (1/2) Epoch 10, batch 40450, libri_loss[loss=0.3491, simple_loss=0.4096, pruned_loss=0.1443, over 26007.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3445, pruned_loss=0.1024, over 5710780.69 frames. ], libri_tot_loss[loss=0.3233, simple_loss=0.3821, pruned_loss=0.1322, over 5665877.14 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3414, pruned_loss=0.09994, over 5719506.25 frames. ], batch size: 136, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:04:10,153 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:04:15,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:04:25,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-05 12:04:40,827 INFO [train.py:968] (1/2) Epoch 10, batch 40500, giga_loss[loss=0.2471, simple_loss=0.317, pruned_loss=0.08855, over 29000.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3412, pruned_loss=0.101, over 5713611.13 frames. ], libri_tot_loss[loss=0.3232, simple_loss=0.3821, pruned_loss=0.1322, over 5673625.90 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3375, pruned_loss=0.098, over 5714993.69 frames. ], batch size: 136, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:04:49,014 INFO [optim.py:369] (1/2) 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 12:04:49,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 12:05:17,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9551, 1.2229, 1.3080, 1.0916], device='cuda:1'), covar=tensor([0.1428, 0.1240, 0.1948, 0.1496], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0731, 0.0669, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 12:05:20,550 INFO [train.py:968] (1/2) Epoch 10, batch 40550, giga_loss[loss=0.2697, simple_loss=0.3425, pruned_loss=0.09846, over 28903.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3397, pruned_loss=0.09985, over 5715073.52 frames. ], libri_tot_loss[loss=0.3231, simple_loss=0.3819, pruned_loss=0.1322, over 5677331.90 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3359, pruned_loss=0.0968, over 5713688.29 frames. ], batch size: 186, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:06:04,288 INFO [train.py:968] (1/2) Epoch 10, batch 40600, giga_loss[loss=0.235, simple_loss=0.3171, pruned_loss=0.07639, over 28522.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3419, pruned_loss=0.1004, over 5714369.30 frames. ], libri_tot_loss[loss=0.323, simple_loss=0.3818, pruned_loss=0.1321, over 5679458.79 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3387, pruned_loss=0.0978, over 5711729.06 frames. ], batch size: 65, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:06:13,680 INFO [optim.py:369] (1/2) 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 12:06:36,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6080, 3.3625, 1.6200, 1.5438], device='cuda:1'), covar=tensor([0.0844, 0.0328, 0.0852, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0500, 0.0332, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 12:06:49,334 INFO [train.py:968] (1/2) Epoch 10, batch 40650, giga_loss[loss=0.2826, simple_loss=0.3467, pruned_loss=0.1092, over 23999.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3453, pruned_loss=0.1019, over 5708743.28 frames. ], libri_tot_loss[loss=0.3225, simple_loss=0.3814, pruned_loss=0.1318, over 5681759.75 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3428, pruned_loss=0.09984, over 5704882.52 frames. ], batch size: 705, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:07:19,228 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 12:07:19,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-05 12:07:27,865 INFO [train.py:968] (1/2) Epoch 10, batch 40700, giga_loss[loss=0.2472, simple_loss=0.3301, pruned_loss=0.08211, over 28962.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3475, pruned_loss=0.102, over 5710197.05 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3815, pruned_loss=0.1318, over 5675139.50 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.345, pruned_loss=0.09993, over 5714221.93 frames. ], batch size: 145, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:07:31,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450298.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:07:34,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-05 12:07:36,833 INFO [optim.py:369] (1/2) 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 12:08:11,575 INFO [train.py:968] (1/2) Epoch 10, batch 40750, giga_loss[loss=0.2873, simple_loss=0.3633, pruned_loss=0.1057, over 28920.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3499, pruned_loss=0.1031, over 5711035.83 frames. ], libri_tot_loss[loss=0.3226, simple_loss=0.3815, pruned_loss=0.1318, over 5676924.25 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3476, pruned_loss=0.1011, over 5713149.47 frames. ], batch size: 213, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:08:29,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4879, 4.3224, 4.0899, 1.9810], device='cuda:1'), covar=tensor([0.0462, 0.0630, 0.0657, 0.1849], device='cuda:1'), in_proj_covar=tensor([0.1019, 0.0952, 0.0839, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 12:08:54,702 INFO [train.py:968] (1/2) Epoch 10, batch 40800, giga_loss[loss=0.3611, simple_loss=0.4159, pruned_loss=0.1531, over 28266.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3517, pruned_loss=0.1045, over 5705826.75 frames. ], libri_tot_loss[loss=0.3217, simple_loss=0.3807, pruned_loss=0.1313, over 5678838.68 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3495, pruned_loss=0.1025, over 5706730.51 frames. ], batch size: 368, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:08:54,861 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:09:05,132 INFO [optim.py:369] (1/2) 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 12:09:35,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:09:40,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:09:45,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450441.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:09:46,997 INFO [train.py:968] (1/2) Epoch 10, batch 40850, giga_loss[loss=0.3833, simple_loss=0.4253, pruned_loss=0.1707, over 28886.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3597, pruned_loss=0.1118, over 5679201.93 frames. ], libri_tot_loss[loss=0.3218, simple_loss=0.3808, pruned_loss=0.1314, over 5672681.72 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3574, pruned_loss=0.1097, over 5686526.76 frames. ], batch size: 199, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:09:47,262 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450444.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:10:03,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4868, 4.2857, 4.0766, 1.9828], device='cuda:1'), covar=tensor([0.0496, 0.0709, 0.0664, 0.2014], device='cuda:1'), in_proj_covar=tensor([0.1023, 0.0961, 0.0843, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 12:10:16,752 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:10:16,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450473.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:10:34,559 INFO [train.py:968] (1/2) Epoch 10, batch 40900, giga_loss[loss=0.3263, simple_loss=0.3684, pruned_loss=0.1421, over 23644.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.118, over 5675888.62 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3805, pruned_loss=0.1312, over 5675834.92 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3655, pruned_loss=0.1162, over 5678826.47 frames. ], batch size: 705, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:10:47,229 INFO [optim.py:369] (1/2) 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 12:10:53,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1308, 1.2636, 3.3550, 2.9203], device='cuda:1'), covar=tensor([0.1524, 0.2381, 0.0458, 0.1829], device='cuda:1'), in_proj_covar=tensor([0.0646, 0.0575, 0.0844, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 12:11:24,914 INFO [train.py:968] (1/2) Epoch 10, batch 40950, giga_loss[loss=0.355, simple_loss=0.403, pruned_loss=0.1535, over 28851.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3729, pruned_loss=0.1218, over 5678431.66 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.38, pruned_loss=0.1309, over 5678114.03 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3717, pruned_loss=0.1206, over 5678725.38 frames. ], batch size: 112, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:11:53,134 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:11:55,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:11:59,382 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:02,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:08,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5115, 4.0795, 1.8019, 1.5976], device='cuda:1'), covar=tensor([0.0896, 0.0332, 0.0787, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0503, 0.0333, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 12:12:11,435 INFO [train.py:968] (1/2) Epoch 10, batch 41000, libri_loss[loss=0.3284, simple_loss=0.3901, pruned_loss=0.1334, over 29747.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3804, pruned_loss=0.1286, over 5675185.02 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3801, pruned_loss=0.1308, over 5680598.22 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3794, pruned_loss=0.1277, over 5672968.00 frames. ], batch size: 87, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:12:23,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:23,617 INFO [optim.py:369] (1/2) 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 12:12:30,177 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:12:45,857 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 41050, giga_loss[loss=0.3492, simple_loss=0.4105, pruned_loss=0.144, over 28847.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3876, pruned_loss=0.1353, over 5670261.34 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3796, pruned_loss=0.1305, over 5681985.49 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3874, pruned_loss=0.1348, over 5667140.77 frames. ], batch size: 145, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:12:59,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2177, 1.2030, 1.1543, 0.9171], device='cuda:1'), covar=tensor([0.0782, 0.0539, 0.0976, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0442, 0.0500, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 12:13:52,612 INFO [train.py:968] (1/2) Epoch 10, batch 41100, giga_loss[loss=0.3101, simple_loss=0.3775, pruned_loss=0.1214, over 28881.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3911, pruned_loss=0.1389, over 5661877.95 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3795, pruned_loss=0.1305, over 5683924.85 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3911, pruned_loss=0.1387, over 5657440.22 frames. ], batch size: 106, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:14:04,060 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 41150, giga_loss[loss=0.3394, simple_loss=0.3931, pruned_loss=0.1429, over 28454.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3935, pruned_loss=0.1421, over 5644536.17 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3789, pruned_loss=0.13, over 5687821.53 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3944, pruned_loss=0.1425, over 5636981.21 frames. ], batch size: 78, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:15:10,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:15:34,497 INFO [train.py:968] (1/2) Epoch 10, batch 41200, giga_loss[loss=0.3245, simple_loss=0.3747, pruned_loss=0.1372, over 28729.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3964, pruned_loss=0.1453, over 5632376.53 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3795, pruned_loss=0.1305, over 5690720.86 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.397, pruned_loss=0.1457, over 5622321.19 frames. ], batch size: 99, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:15:47,472 INFO [optim.py:369] (1/2) 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 12:16:24,865 INFO [train.py:968] (1/2) Epoch 10, batch 41250, giga_loss[loss=0.3307, simple_loss=0.3979, pruned_loss=0.1317, over 28550.00 frames. ], tot_loss[loss=0.3516, simple_loss=0.4021, pruned_loss=0.1505, over 5632961.81 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3799, pruned_loss=0.1308, over 5694508.53 frames. ], giga_tot_loss[loss=0.3523, simple_loss=0.4027, pruned_loss=0.1509, over 5620465.32 frames. ], batch size: 85, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:16:31,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=450848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:17:18,294 INFO [train.py:968] (1/2) Epoch 10, batch 41300, giga_loss[loss=0.302, simple_loss=0.3646, pruned_loss=0.1197, over 28991.00 frames. ], tot_loss[loss=0.3537, simple_loss=0.4038, pruned_loss=0.1518, over 5642347.38 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.38, pruned_loss=0.1308, over 5697587.42 frames. ], giga_tot_loss[loss=0.3548, simple_loss=0.4046, pruned_loss=0.1525, over 5628636.85 frames. ], batch size: 128, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:17:19,799 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=450895.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:17:35,901 INFO [optim.py:369] (1/2) 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 12:17:40,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:17:45,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:18:10,531 INFO [train.py:968] (1/2) Epoch 10, batch 41350, libri_loss[loss=0.3872, simple_loss=0.4267, pruned_loss=0.1738, over 29639.00 frames. ], tot_loss[loss=0.3522, simple_loss=0.4017, pruned_loss=0.1514, over 5630778.06 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3803, pruned_loss=0.1312, over 5691665.99 frames. ], giga_tot_loss[loss=0.3531, simple_loss=0.4024, pruned_loss=0.1519, over 5624569.42 frames. ], batch size: 91, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:18:11,419 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=450944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:18:22,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5103, 2.2199, 1.5729, 0.6710], device='cuda:1'), covar=tensor([0.3653, 0.1964, 0.2955, 0.4428], device='cuda:1'), in_proj_covar=tensor([0.1548, 0.1475, 0.1490, 0.1270], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 12:18:25,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8076, 1.8462, 1.7388, 1.7085], device='cuda:1'), covar=tensor([0.1368, 0.2052, 0.1803, 0.1837], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0733, 0.0670, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 12:18:56,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5160, 2.1588, 1.5910, 0.7926], device='cuda:1'), covar=tensor([0.3127, 0.1897, 0.3009, 0.3696], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1476, 0.1492, 0.1271], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 12:18:58,347 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=450991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:19:01,161 INFO [train.py:968] (1/2) Epoch 10, batch 41400, giga_loss[loss=0.3191, simple_loss=0.3864, pruned_loss=0.1259, over 28964.00 frames. ], tot_loss[loss=0.351, simple_loss=0.4008, pruned_loss=0.1507, over 5632413.32 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3803, pruned_loss=0.1312, over 5695308.80 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4016, pruned_loss=0.1513, over 5623427.98 frames. ], batch size: 136, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:19:01,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=450994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:19:12,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:19:30,233 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:19:50,913 INFO [train.py:968] (1/2) Epoch 10, batch 41450, giga_loss[loss=0.3669, simple_loss=0.4172, pruned_loss=0.1582, over 28565.00 frames. ], tot_loss[loss=0.3498, simple_loss=0.4006, pruned_loss=0.1495, over 5613030.80 frames. ], libri_tot_loss[loss=0.3222, simple_loss=0.381, pruned_loss=0.1317, over 5681633.86 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4011, pruned_loss=0.15, over 5616379.89 frames. ], batch size: 336, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:20:28,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 12:20:32,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6267, 1.8943, 1.9434, 1.4683], device='cuda:1'), covar=tensor([0.1476, 0.1985, 0.1127, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0698, 0.0847, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 12:20:34,945 INFO [train.py:968] (1/2) Epoch 10, batch 41500, libri_loss[loss=0.352, simple_loss=0.3903, pruned_loss=0.1569, over 29503.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.4011, pruned_loss=0.1497, over 5620512.70 frames. ], libri_tot_loss[loss=0.3214, simple_loss=0.3804, pruned_loss=0.1312, over 5692589.77 frames. ], giga_tot_loss[loss=0.353, simple_loss=0.403, pruned_loss=0.1514, over 5610099.02 frames. ], batch size: 70, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:20:46,441 INFO [optim.py:369] (1/2) 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 12:21:23,301 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 41550, giga_loss[loss=0.3065, simple_loss=0.3798, pruned_loss=0.1166, over 29117.00 frames. ], tot_loss[loss=0.3488, simple_loss=0.4, pruned_loss=0.1488, over 5607983.22 frames. ], libri_tot_loss[loss=0.3213, simple_loss=0.3802, pruned_loss=0.1312, over 5695165.39 frames. ], giga_tot_loss[loss=0.3519, simple_loss=0.4023, pruned_loss=0.1507, over 5595057.16 frames. ], batch size: 113, lr: 3.17e-03, grad_scale: 2.0
+2023-03-05 12:21:29,310 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:21:32,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:21:59,620 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 41600, giga_loss[loss=0.3377, simple_loss=0.4098, pruned_loss=0.1328, over 28833.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3979, pruned_loss=0.1459, over 5619248.01 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3801, pruned_loss=0.1311, over 5693333.20 frames. ], giga_tot_loss[loss=0.3482, simple_loss=0.4004, pruned_loss=0.148, over 5607397.65 frames. ], batch size: 186, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:22:16,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0453, 3.8708, 3.6842, 1.9399], device='cuda:1'), covar=tensor([0.0549, 0.0685, 0.0680, 0.2014], device='cuda:1'), in_proj_covar=tensor([0.1033, 0.0972, 0.0852, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 12:22:25,720 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 41650, giga_loss[loss=0.2564, simple_loss=0.3422, pruned_loss=0.08528, over 28472.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3957, pruned_loss=0.1427, over 5637337.90 frames. ], libri_tot_loss[loss=0.3211, simple_loss=0.3801, pruned_loss=0.131, over 5700076.09 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3983, pruned_loss=0.1449, over 5619603.44 frames. ], batch size: 60, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:23:05,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3192, 1.4506, 1.2120, 1.3211], device='cuda:1'), covar=tensor([0.1742, 0.1426, 0.1342, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1583, 0.1549, 0.1653], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:23:17,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=451262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:23:24,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=451270.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:23:32,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 12:23:40,427 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-05 12:23:47,813 INFO [train.py:968] (1/2) Epoch 10, batch 41700, giga_loss[loss=0.2833, simple_loss=0.3589, pruned_loss=0.1039, over 28971.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3921, pruned_loss=0.1399, over 5637241.47 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3798, pruned_loss=0.1309, over 5704653.30 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3948, pruned_loss=0.142, over 5617790.21 frames. ], batch size: 136, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:24:01,458 INFO [optim.py:369] (1/2) 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 12:24:36,345 INFO [train.py:968] (1/2) Epoch 10, batch 41750, giga_loss[loss=0.3444, simple_loss=0.3988, pruned_loss=0.145, over 28776.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3904, pruned_loss=0.1384, over 5633155.59 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3795, pruned_loss=0.1308, over 5705880.33 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.393, pruned_loss=0.1403, over 5615539.09 frames. ], batch size: 284, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:25:25,589 INFO [train.py:968] (1/2) Epoch 10, batch 41800, giga_loss[loss=0.324, simple_loss=0.3725, pruned_loss=0.1378, over 28407.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3869, pruned_loss=0.1353, over 5654553.24 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3791, pruned_loss=0.1306, over 5708987.70 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3894, pruned_loss=0.1371, over 5636980.07 frames. ], batch size: 71, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:25:41,577 INFO [optim.py:369] (1/2) 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 12:25:48,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451413.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:25:50,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451416.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:26:15,465 INFO [train.py:968] (1/2) Epoch 10, batch 41850, giga_loss[loss=0.2884, simple_loss=0.3649, pruned_loss=0.106, over 28847.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3873, pruned_loss=0.1358, over 5650961.98 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3791, pruned_loss=0.1306, over 5707737.67 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3894, pruned_loss=0.1373, over 5637311.42 frames. ], batch size: 186, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:26:17,301 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451445.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:26:22,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4582, 3.9964, 1.6037, 1.5811], device='cuda:1'), covar=tensor([0.0901, 0.0272, 0.0845, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0506, 0.0335, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 12:27:02,486 INFO [train.py:968] (1/2) Epoch 10, batch 41900, giga_loss[loss=0.3116, simple_loss=0.3778, pruned_loss=0.1227, over 28757.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3853, pruned_loss=0.134, over 5646103.35 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3791, pruned_loss=0.1305, over 5702139.89 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3873, pruned_loss=0.1354, over 5637645.04 frames. ], batch size: 284, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:27:22,032 INFO [optim.py:369] (1/2) 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 12:27:26,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 12:27:32,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=451517.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:28:02,756 INFO [train.py:968] (1/2) Epoch 10, batch 41950, giga_loss[loss=0.3108, simple_loss=0.379, pruned_loss=0.1213, over 28741.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.384, pruned_loss=0.1312, over 5638942.42 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3793, pruned_loss=0.1307, over 5704285.76 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3854, pruned_loss=0.1322, over 5629760.57 frames. ], batch size: 92, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:28:53,574 INFO [train.py:968] (1/2) Epoch 10, batch 42000, libri_loss[loss=0.2816, simple_loss=0.3427, pruned_loss=0.1102, over 29663.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3845, pruned_loss=0.1291, over 5649028.23 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3787, pruned_loss=0.1303, over 5697730.98 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3864, pruned_loss=0.1302, over 5645605.65 frames. ], batch size: 73, lr: 3.17e-03, grad_scale: 8.0
+2023-03-05 12:28:53,575 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 12:29:00,260 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2025, 1.6861, 1.5731, 1.1008], device='cuda:1'), covar=tensor([0.1901, 0.2803, 0.1529, 0.1812], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0704, 0.0853, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 12:29:01,940 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 12:29:15,336 INFO [optim.py:369] (1/2) 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 12:29:32,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 12:29:44,161 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=451637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:29:50,415 INFO [train.py:968] (1/2) Epoch 10, batch 42050, giga_loss[loss=0.3647, simple_loss=0.3921, pruned_loss=0.1687, over 23585.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3852, pruned_loss=0.1296, over 5654198.94 frames. ], libri_tot_loss[loss=0.3194, simple_loss=0.3785, pruned_loss=0.1301, over 5697943.05 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3869, pruned_loss=0.1307, over 5650827.63 frames. ], batch size: 705, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:30:06,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451660.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:30:07,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-05 12:30:08,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451663.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:30:35,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451692.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:30:36,875 INFO [train.py:968] (1/2) Epoch 10, batch 42100, giga_loss[loss=0.3679, simple_loss=0.4095, pruned_loss=0.1631, over 27586.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3863, pruned_loss=0.1312, over 5658043.14 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3779, pruned_loss=0.1296, over 5699574.60 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3884, pruned_loss=0.1325, over 5653273.51 frames. ], batch size: 472, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:30:49,127 INFO [optim.py:369] (1/2) 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 12:31:01,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=451723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:31:07,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 12:31:19,992 INFO [train.py:968] (1/2) Epoch 10, batch 42150, giga_loss[loss=0.3332, simple_loss=0.3991, pruned_loss=0.1337, over 28693.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3841, pruned_loss=0.1302, over 5671097.46 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3774, pruned_loss=0.1291, over 5706023.17 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3864, pruned_loss=0.1317, over 5660641.02 frames. ], batch size: 262, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:31:57,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=451780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:32:01,341 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=451783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:32:10,454 INFO [train.py:968] (1/2) Epoch 10, batch 42200, giga_loss[loss=0.2939, simple_loss=0.3568, pruned_loss=0.1155, over 28992.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3829, pruned_loss=0.1313, over 5662706.78 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3777, pruned_loss=0.1294, over 5705089.14 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3845, pruned_loss=0.1323, over 5655037.03 frames. ], batch size: 106, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:32:24,293 INFO [optim.py:369] (1/2) 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 12:32:28,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=451812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:32:34,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2764, 1.2106, 0.9676, 1.4856], device='cuda:1'), covar=tensor([0.0735, 0.0319, 0.0364, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 12:32:58,138 INFO [train.py:968] (1/2) Epoch 10, batch 42250, giga_loss[loss=0.276, simple_loss=0.3537, pruned_loss=0.09913, over 28546.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3816, pruned_loss=0.1302, over 5658014.65 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3774, pruned_loss=0.1291, over 5700148.77 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3834, pruned_loss=0.1313, over 5655423.77 frames. ], batch size: 60, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:33:25,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3592, 1.8101, 1.5815, 1.3080], device='cuda:1'), covar=tensor([0.2202, 0.1501, 0.1599, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.1688, 0.1596, 0.1562, 0.1678], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:33:44,745 INFO [train.py:968] (1/2) Epoch 10, batch 42300, giga_loss[loss=0.2928, simple_loss=0.3674, pruned_loss=0.1091, over 28363.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3805, pruned_loss=0.128, over 5669254.78 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.377, pruned_loss=0.129, over 5702237.02 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3823, pruned_loss=0.129, over 5664453.42 frames. ], batch size: 71, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:33:57,439 INFO [optim.py:369] (1/2) 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 12:34:11,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7305, 2.0450, 2.0115, 1.5479], device='cuda:1'), covar=tensor([0.1854, 0.2170, 0.1409, 0.1631], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0704, 0.0856, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 12:34:30,411 INFO [train.py:968] (1/2) Epoch 10, batch 42350, giga_loss[loss=0.3258, simple_loss=0.3998, pruned_loss=0.1259, over 28970.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3805, pruned_loss=0.1274, over 5680609.32 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3771, pruned_loss=0.129, over 5704956.11 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.382, pruned_loss=0.1281, over 5673526.09 frames. ], batch size: 164, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:34:45,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=451956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:35:19,671 INFO [train.py:968] (1/2) Epoch 10, batch 42400, libri_loss[loss=0.3356, simple_loss=0.3905, pruned_loss=0.1403, over 18980.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3809, pruned_loss=0.1281, over 5654882.43 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3773, pruned_loss=0.1292, over 5688805.97 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3819, pruned_loss=0.1285, over 5663785.92 frames. ], batch size: 187, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:35:35,579 INFO [optim.py:369] (1/2) 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 12:35:48,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2034, 2.6107, 1.2615, 1.2730], device='cuda:1'), covar=tensor([0.0968, 0.0358, 0.0867, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0506, 0.0337, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 12:36:06,247 INFO [train.py:968] (1/2) Epoch 10, batch 42450, giga_loss[loss=0.31, simple_loss=0.373, pruned_loss=0.1236, over 28907.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3789, pruned_loss=0.127, over 5668520.58 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3774, pruned_loss=0.1292, over 5693001.68 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3796, pruned_loss=0.1273, over 5671204.66 frames. ], batch size: 227, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:36:36,505 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:36:53,506 INFO [train.py:968] (1/2) Epoch 10, batch 42500, giga_loss[loss=0.2652, simple_loss=0.3402, pruned_loss=0.0951, over 28622.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3779, pruned_loss=0.1267, over 5657933.00 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3776, pruned_loss=0.1292, over 5684765.20 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3784, pruned_loss=0.1268, over 5666479.09 frames. ], batch size: 85, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:36:56,983 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:36:58,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:37:11,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-05 12:37:11,832 INFO [optim.py:369] (1/2) 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 12:37:44,533 INFO [train.py:968] (1/2) Epoch 10, batch 42550, giga_loss[loss=0.3212, simple_loss=0.3819, pruned_loss=0.1303, over 28879.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3761, pruned_loss=0.1262, over 5669261.78 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3773, pruned_loss=0.1291, over 5688060.35 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3767, pruned_loss=0.1264, over 5672753.08 frames. ], batch size: 112, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:38:22,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3397, 1.6022, 1.3940, 1.1935], device='cuda:1'), covar=tensor([0.1997, 0.1568, 0.1245, 0.1587], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1584, 0.1553, 0.1664], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:38:32,054 INFO [train.py:968] (1/2) Epoch 10, batch 42600, libri_loss[loss=0.3293, simple_loss=0.3936, pruned_loss=0.1325, over 29532.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.1271, over 5675758.11 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3778, pruned_loss=0.1292, over 5693870.26 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3764, pruned_loss=0.127, over 5672856.15 frames. ], batch size: 89, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:38:46,555 INFO [optim.py:369] (1/2) 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 12:38:50,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1464, 2.4239, 1.2559, 1.2708], device='cuda:1'), covar=tensor([0.0907, 0.0384, 0.0817, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0509, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 12:39:19,206 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:39:20,988 INFO [train.py:968] (1/2) Epoch 10, batch 42650, giga_loss[loss=0.3404, simple_loss=0.3836, pruned_loss=0.1486, over 28619.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5666842.48 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.1291, over 5694470.05 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3767, pruned_loss=0.1281, over 5663887.07 frames. ], batch size: 85, lr: 3.17e-03, grad_scale: 4.0
+2023-03-05 12:39:23,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:39:50,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:40:11,173 INFO [train.py:968] (1/2) Epoch 10, batch 42700, giga_loss[loss=0.2893, simple_loss=0.3532, pruned_loss=0.1127, over 28291.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3763, pruned_loss=0.1281, over 5651516.80 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3778, pruned_loss=0.129, over 5689544.60 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3761, pruned_loss=0.1282, over 5653115.00 frames. ], batch size: 77, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:40:20,977 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:40:46,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:40:56,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 12:40:57,072 INFO [train.py:968] (1/2) Epoch 10, batch 42750, libri_loss[loss=0.3489, simple_loss=0.4096, pruned_loss=0.144, over 29534.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.376, pruned_loss=0.1273, over 5657352.47 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3778, pruned_loss=0.1291, over 5689081.46 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3758, pruned_loss=0.1273, over 5658383.71 frames. ], batch size: 89, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:41:40,694 INFO [train.py:968] (1/2) Epoch 10, batch 42800, giga_loss[loss=0.2807, simple_loss=0.3631, pruned_loss=0.09915, over 28916.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3767, pruned_loss=0.1267, over 5658851.12 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3781, pruned_loss=0.1291, over 5681302.22 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3763, pruned_loss=0.1265, over 5666848.70 frames. ], batch size: 145, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:41:57,582 INFO [optim.py:369] (1/2) 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 12:42:04,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452419.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:42:16,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6861, 3.4442, 3.2861, 1.7102], device='cuda:1'), covar=tensor([0.0833, 0.1198, 0.1117, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1047, 0.0985, 0.0860, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 12:42:17,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3730, 1.4742, 1.3517, 1.2825], device='cuda:1'), covar=tensor([0.1833, 0.1506, 0.1311, 0.1551], device='cuda:1'), in_proj_covar=tensor([0.1670, 0.1584, 0.1556, 0.1671], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:42:25,442 INFO [train.py:968] (1/2) Epoch 10, batch 42850, giga_loss[loss=0.3148, simple_loss=0.3796, pruned_loss=0.125, over 28978.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3767, pruned_loss=0.1261, over 5650568.44 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3782, pruned_loss=0.1292, over 5666421.71 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3762, pruned_loss=0.1258, over 5669228.55 frames. ], batch size: 164, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:42:32,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:42:57,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:42:58,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:43:00,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:43:19,854 INFO [train.py:968] (1/2) Epoch 10, batch 42900, libri_loss[loss=0.2567, simple_loss=0.3248, pruned_loss=0.09424, over 29413.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3776, pruned_loss=0.1272, over 5662240.33 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3779, pruned_loss=0.1291, over 5669934.19 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3775, pruned_loss=0.1271, over 5673526.85 frames. ], batch size: 67, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:43:31,382 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:43:37,067 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 10, batch 42950, giga_loss[loss=0.286, simple_loss=0.3553, pruned_loss=0.1084, over 28969.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3792, pruned_loss=0.1286, over 5665003.04 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3779, pruned_loss=0.1291, over 5664068.96 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3792, pruned_loss=0.1286, over 5680010.65 frames. ], batch size: 155, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:44:56,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:44:56,513 INFO [train.py:968] (1/2) Epoch 10, batch 43000, libri_loss[loss=0.3432, simple_loss=0.3986, pruned_loss=0.1439, over 29674.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3805, pruned_loss=0.131, over 5668587.55 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3777, pruned_loss=0.1289, over 5665491.35 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3808, pruned_loss=0.1312, over 5680347.81 frames. ], batch size: 88, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:44:58,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:14,911 INFO [optim.py:369] (1/2) 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 12:45:18,390 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:21,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:28,723 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 43050, giga_loss[loss=0.2981, simple_loss=0.3693, pruned_loss=0.1135, over 29054.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3812, pruned_loss=0.1325, over 5664029.05 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3773, pruned_loss=0.1286, over 5663531.47 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3819, pruned_loss=0.1331, over 5675908.66 frames. ], batch size: 155, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:45:47,539 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:45:57,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8386, 0.9457, 0.7965, 0.8519], device='cuda:1'), covar=tensor([0.1001, 0.1156, 0.0714, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.1673, 0.1588, 0.1561, 0.1667], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:46:04,649 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:46:19,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452680.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:46:32,745 INFO [train.py:968] (1/2) Epoch 10, batch 43100, giga_loss[loss=0.3718, simple_loss=0.4192, pruned_loss=0.1622, over 28936.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3834, pruned_loss=0.1349, over 5647426.10 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3763, pruned_loss=0.1278, over 5662146.88 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3851, pruned_loss=0.1362, over 5657963.33 frames. ], batch size: 227, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 12:46:50,123 INFO [optim.py:369] (1/2) 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 12:46:56,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3395, 1.6887, 1.3605, 1.2944], device='cuda:1'), covar=tensor([0.1923, 0.1774, 0.1906, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.1280, 0.0952, 0.1130, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 12:47:09,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3001, 4.1166, 3.9293, 1.8286], device='cuda:1'), covar=tensor([0.0655, 0.0757, 0.0796, 0.2270], device='cuda:1'), in_proj_covar=tensor([0.1050, 0.0986, 0.0866, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 12:47:15,545 INFO [train.py:968] (1/2) Epoch 10, batch 43150, giga_loss[loss=0.3083, simple_loss=0.3716, pruned_loss=0.1225, over 28941.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3832, pruned_loss=0.1349, over 5641035.77 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3766, pruned_loss=0.128, over 5651393.36 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3845, pruned_loss=0.136, over 5659332.51 frames. ], batch size: 136, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 12:47:21,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:47:27,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2708, 3.3054, 1.4344, 1.3918], device='cuda:1'), covar=tensor([0.0984, 0.0334, 0.0872, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0511, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 12:47:31,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3914, 1.5633, 1.2273, 1.4184], device='cuda:1'), covar=tensor([0.1726, 0.1497, 0.1540, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.1677, 0.1591, 0.1566, 0.1669], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:48:00,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.95 vs. limit=5.0
+2023-03-05 12:48:03,679 INFO [train.py:968] (1/2) Epoch 10, batch 43200, giga_loss[loss=0.348, simple_loss=0.4062, pruned_loss=0.1449, over 28697.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3816, pruned_loss=0.133, over 5652764.90 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1276, over 5656369.75 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3833, pruned_loss=0.1343, over 5662618.40 frames. ], batch size: 307, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:48:04,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=452794.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:48:12,691 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=452807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:48:17,000 INFO [optim.py:369] (1/2) 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 12:48:28,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452823.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:48:30,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452826.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:48:43,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-05 12:48:46,014 INFO [train.py:968] (1/2) Epoch 10, batch 43250, libri_loss[loss=0.3174, simple_loss=0.3789, pruned_loss=0.1279, over 29542.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3798, pruned_loss=0.1302, over 5656181.33 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3762, pruned_loss=0.1277, over 5664621.10 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3811, pruned_loss=0.1313, over 5656326.51 frames. ], batch size: 84, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:48:56,501 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452855.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:49:33,401 INFO [train.py:968] (1/2) Epoch 10, batch 43300, giga_loss[loss=0.2837, simple_loss=0.3529, pruned_loss=0.1073, over 28938.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3776, pruned_loss=0.1286, over 5658462.37 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3766, pruned_loss=0.1282, over 5669350.96 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3783, pruned_loss=0.129, over 5654374.36 frames. ], batch size: 227, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:49:48,922 INFO [optim.py:369] (1/2) 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 12:50:11,646 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=452937.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:50:14,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=452940.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 12:50:19,760 INFO [train.py:968] (1/2) Epoch 10, batch 43350, giga_loss[loss=0.4016, simple_loss=0.4234, pruned_loss=0.1899, over 26653.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3775, pruned_loss=0.1292, over 5660322.39 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3769, pruned_loss=0.1284, over 5663318.31 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3777, pruned_loss=0.1294, over 5662586.74 frames. ], batch size: 555, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:50:38,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8013, 1.9707, 1.5485, 2.2333], device='cuda:1'), covar=tensor([0.2235, 0.2326, 0.2457, 0.2165], device='cuda:1'), in_proj_covar=tensor([0.1282, 0.0954, 0.1130, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 12:50:40,231 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=452969.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 12:50:41,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7491, 1.9313, 2.0283, 1.5361], device='cuda:1'), covar=tensor([0.1631, 0.2067, 0.1259, 0.1493], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0703, 0.0852, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0013, 0.0012], device='cuda:1')
+2023-03-05 12:51:02,097 INFO [train.py:968] (1/2) Epoch 10, batch 43400, giga_loss[loss=0.2856, simple_loss=0.3571, pruned_loss=0.1071, over 28991.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3757, pruned_loss=0.1281, over 5655024.73 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5654063.00 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3763, pruned_loss=0.1286, over 5665292.30 frames. ], batch size: 155, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:51:18,497 INFO [optim.py:369] (1/2) 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 12:51:46,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:51:49,036 INFO [train.py:968] (1/2) Epoch 10, batch 43450, giga_loss[loss=0.4602, simple_loss=0.4603, pruned_loss=0.2301, over 26522.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.379, pruned_loss=0.1306, over 5660838.18 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3763, pruned_loss=0.1279, over 5659248.89 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3796, pruned_loss=0.1311, over 5664526.35 frames. ], batch size: 555, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:52:35,509 INFO [train.py:968] (1/2) Epoch 10, batch 43500, giga_loss[loss=0.2875, simple_loss=0.367, pruned_loss=0.104, over 28956.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3827, pruned_loss=0.1308, over 5661303.75 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.376, pruned_loss=0.1278, over 5664485.47 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3835, pruned_loss=0.1313, over 5659294.61 frames. ], batch size: 106, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:52:54,219 INFO [optim.py:369] (1/2) 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 12:53:08,660 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 43550, giga_loss[loss=0.3222, simple_loss=0.3861, pruned_loss=0.1292, over 28606.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3825, pruned_loss=0.1285, over 5669008.21 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3757, pruned_loss=0.1276, over 5668894.51 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3835, pruned_loss=0.1292, over 5663602.90 frames. ], batch size: 85, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:54:07,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:54:08,746 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:54:10,841 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 43600, giga_loss[loss=0.2941, simple_loss=0.3593, pruned_loss=0.1144, over 28634.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3852, pruned_loss=0.1312, over 5669485.33 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3752, pruned_loss=0.1273, over 5674657.58 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3868, pruned_loss=0.1319, over 5659830.21 frames. ], batch size: 92, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 12:54:21,669 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 12:54:32,495 INFO [optim.py:369] (1/2) 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 12:54:36,300 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 43650, giga_loss[loss=0.3022, simple_loss=0.3742, pruned_loss=0.1151, over 28926.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3865, pruned_loss=0.1324, over 5659550.78 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3751, pruned_loss=0.1273, over 5665568.66 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3883, pruned_loss=0.1333, over 5658494.46 frames. ], batch size: 174, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:55:26,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:55:28,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:55:32,574 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 12:55:41,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:55:46,630 INFO [train.py:968] (1/2) Epoch 10, batch 43700, giga_loss[loss=0.3473, simple_loss=0.4047, pruned_loss=0.1449, over 28657.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3864, pruned_loss=0.1333, over 5663106.00 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3745, pruned_loss=0.1269, over 5669104.64 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3887, pruned_loss=0.1346, over 5658641.11 frames. ], batch size: 242, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:55:52,013 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:02,907 INFO [optim.py:369] (1/2) 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 12:56:03,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5674, 1.8689, 1.5537, 1.3627], device='cuda:1'), covar=tensor([0.2010, 0.1450, 0.1229, 0.1507], device='cuda:1'), in_proj_covar=tensor([0.1681, 0.1590, 0.1561, 0.1670], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 12:56:13,952 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:19,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:32,858 INFO [train.py:968] (1/2) Epoch 10, batch 43750, giga_loss[loss=0.3333, simple_loss=0.3862, pruned_loss=0.1402, over 28667.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3849, pruned_loss=0.1332, over 5654559.44 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3745, pruned_loss=0.1269, over 5665903.82 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.387, pruned_loss=0.1343, over 5654656.44 frames. ], batch size: 99, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:56:42,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:56:46,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 12:57:18,817 INFO [train.py:968] (1/2) Epoch 10, batch 43800, giga_loss[loss=0.3205, simple_loss=0.3831, pruned_loss=0.1289, over 28946.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3827, pruned_loss=0.1324, over 5651603.47 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3746, pruned_loss=0.127, over 5662468.84 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3846, pruned_loss=0.1334, over 5654431.20 frames. ], batch size: 164, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:57:36,418 INFO [optim.py:369] (1/2) 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 12:57:40,371 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-05 12:58:06,559 INFO [train.py:968] (1/2) Epoch 10, batch 43850, giga_loss[loss=0.3181, simple_loss=0.3672, pruned_loss=0.1345, over 28796.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3806, pruned_loss=0.1311, over 5656914.01 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3746, pruned_loss=0.1269, over 5658774.76 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3823, pruned_loss=0.1321, over 5663205.98 frames. ], batch size: 99, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:58:59,679 INFO [train.py:968] (1/2) Epoch 10, batch 43900, giga_loss[loss=0.3285, simple_loss=0.389, pruned_loss=0.134, over 28931.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3826, pruned_loss=0.1334, over 5665606.74 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3746, pruned_loss=0.1268, over 5660361.85 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.384, pruned_loss=0.1342, over 5669216.25 frames. ], batch size: 227, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 12:59:17,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6303, 1.7373, 1.5101, 1.7962], device='cuda:1'), covar=tensor([0.1946, 0.1877, 0.1814, 0.1847], device='cuda:1'), in_proj_covar=tensor([0.1278, 0.0950, 0.1127, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 12:59:17,823 INFO [optim.py:369] (1/2) 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 12:59:50,043 INFO [train.py:968] (1/2) Epoch 10, batch 43950, giga_loss[loss=0.3274, simple_loss=0.384, pruned_loss=0.1354, over 29018.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3834, pruned_loss=0.1343, over 5665613.86 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3746, pruned_loss=0.1268, over 5664969.33 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3847, pruned_loss=0.1352, over 5664637.77 frames. ], batch size: 136, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:00:32,648 INFO [train.py:968] (1/2) Epoch 10, batch 44000, giga_loss[loss=0.3381, simple_loss=0.3862, pruned_loss=0.145, over 28962.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3812, pruned_loss=0.1331, over 5665352.51 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3741, pruned_loss=0.1264, over 5661309.34 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3828, pruned_loss=0.1343, over 5668008.28 frames. ], batch size: 106, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:00:52,151 INFO [optim.py:369] (1/2) 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 13:01:20,782 INFO [train.py:968] (1/2) Epoch 10, batch 44050, giga_loss[loss=0.3159, simple_loss=0.382, pruned_loss=0.1249, over 28777.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3801, pruned_loss=0.1323, over 5669224.89 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.374, pruned_loss=0.1264, over 5662402.79 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3815, pruned_loss=0.1333, over 5670260.54 frames. ], batch size: 285, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:01:37,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:01:58,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:02:07,681 INFO [train.py:968] (1/2) Epoch 10, batch 44100, giga_loss[loss=0.3866, simple_loss=0.4251, pruned_loss=0.174, over 27499.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3806, pruned_loss=0.1316, over 5657347.74 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3741, pruned_loss=0.1264, over 5656033.69 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3818, pruned_loss=0.1326, over 5664575.23 frames. ], batch size: 472, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:02:11,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3292, 1.5145, 1.3640, 1.5457], device='cuda:1'), covar=tensor([0.0746, 0.0325, 0.0310, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 13:02:28,591 INFO [optim.py:369] (1/2) 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 13:02:42,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:02:56,810 INFO [train.py:968] (1/2) Epoch 10, batch 44150, giga_loss[loss=0.3112, simple_loss=0.3739, pruned_loss=0.1243, over 28974.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3836, pruned_loss=0.1334, over 5663706.38 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3746, pruned_loss=0.1268, over 5658060.10 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3842, pruned_loss=0.1339, over 5667652.22 frames. ], batch size: 213, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:03:03,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=453749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:03:46,059 INFO [train.py:968] (1/2) Epoch 10, batch 44200, giga_loss[loss=0.33, simple_loss=0.3863, pruned_loss=0.1368, over 28980.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3834, pruned_loss=0.1343, over 5651134.70 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3749, pruned_loss=0.127, over 5649690.46 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3838, pruned_loss=0.1346, over 5661306.60 frames. ], batch size: 227, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:03:58,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:00,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:03,824 INFO [optim.py:369] (1/2) 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 13:04:18,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:21,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:21,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:26,003 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:32,266 INFO [train.py:968] (1/2) Epoch 10, batch 44250, giga_loss[loss=0.2818, simple_loss=0.362, pruned_loss=0.1008, over 28833.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3852, pruned_loss=0.1332, over 5655893.05 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.375, pruned_loss=0.1271, over 5653755.75 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3857, pruned_loss=0.1336, over 5660384.04 frames. ], batch size: 99, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:04:34,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=453846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:45,760 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:55,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:04:57,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:11,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=453892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:13,019 INFO [train.py:968] (1/2) Epoch 10, batch 44300, giga_loss[loss=0.333, simple_loss=0.4005, pruned_loss=0.1327, over 28976.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3872, pruned_loss=0.1321, over 5662694.25 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3758, pruned_loss=0.1277, over 5642403.67 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3873, pruned_loss=0.1319, over 5677072.65 frames. ], batch size: 213, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 13:05:15,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=453895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:21,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:29,243 INFO [optim.py:369] (1/2) 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 13:05:38,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=453924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:05:58,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2152, 2.1076, 1.4887, 1.8081], device='cuda:1'), covar=tensor([0.0742, 0.0668, 0.1054, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0443, 0.0495, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:05:59,605 INFO [train.py:968] (1/2) Epoch 10, batch 44350, giga_loss[loss=0.3444, simple_loss=0.4103, pruned_loss=0.1392, over 28723.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3892, pruned_loss=0.1324, over 5672551.57 frames. ], libri_tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1277, over 5650442.61 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3896, pruned_loss=0.1325, over 5678011.39 frames. ], batch size: 284, lr: 3.16e-03, grad_scale: 2.0
+2023-03-05 13:06:06,780 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 13:06:48,870 INFO [train.py:968] (1/2) Epoch 10, batch 44400, giga_loss[loss=0.3532, simple_loss=0.4124, pruned_loss=0.147, over 28854.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3924, pruned_loss=0.1354, over 5673456.02 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3761, pruned_loss=0.1278, over 5652993.77 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3927, pruned_loss=0.1354, over 5675675.34 frames. ], batch size: 199, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:07:08,636 INFO [optim.py:369] (1/2) 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 13:07:40,959 INFO [train.py:968] (1/2) Epoch 10, batch 44450, giga_loss[loss=0.3138, simple_loss=0.3815, pruned_loss=0.1231, over 28853.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.395, pruned_loss=0.1394, over 5653055.19 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3763, pruned_loss=0.1279, over 5657001.83 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3953, pruned_loss=0.1395, over 5651576.21 frames. ], batch size: 199, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:08:06,521 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:08:26,757 INFO [train.py:968] (1/2) Epoch 10, batch 44500, giga_loss[loss=0.3433, simple_loss=0.3992, pruned_loss=0.1437, over 28754.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.394, pruned_loss=0.1393, over 5660639.83 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3763, pruned_loss=0.1279, over 5663360.60 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3947, pruned_loss=0.1397, over 5653769.23 frames. ], batch size: 284, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:08:33,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3175, 1.3445, 1.5266, 1.1869], device='cuda:1'), covar=tensor([0.1390, 0.1680, 0.1857, 0.1630], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0733, 0.0668, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 13:08:44,576 INFO [optim.py:369] (1/2) 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 13:08:59,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4209, 1.7292, 1.4195, 1.5486], device='cuda:1'), covar=tensor([0.0750, 0.0295, 0.0304, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 13:09:13,802 INFO [train.py:968] (1/2) Epoch 10, batch 44550, giga_loss[loss=0.3263, simple_loss=0.3936, pruned_loss=0.1295, over 28667.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3922, pruned_loss=0.1377, over 5661819.51 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3762, pruned_loss=0.1279, over 5662304.50 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.393, pruned_loss=0.138, over 5657118.08 frames. ], batch size: 262, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:09:16,652 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:09:58,393 INFO [train.py:968] (1/2) Epoch 10, batch 44600, giga_loss[loss=0.2954, simple_loss=0.3792, pruned_loss=0.1058, over 28595.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3904, pruned_loss=0.134, over 5659882.25 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3762, pruned_loss=0.1279, over 5655559.41 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3913, pruned_loss=0.1345, over 5661370.97 frames. ], batch size: 71, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:10:12,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:10:18,385 INFO [optim.py:369] (1/2) 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 13:10:26,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:10:29,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-05 13:10:43,001 INFO [train.py:968] (1/2) Epoch 10, batch 44650, libri_loss[loss=0.3389, simple_loss=0.3987, pruned_loss=0.1396, over 29759.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3905, pruned_loss=0.133, over 5665032.44 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3764, pruned_loss=0.1279, over 5660853.85 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3916, pruned_loss=0.1336, over 5661539.68 frames. ], batch size: 87, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:10:43,173 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:10:50,410 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:11:31,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5882, 1.6973, 1.5210, 1.5007], device='cuda:1'), covar=tensor([0.1288, 0.1732, 0.1833, 0.1704], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0734, 0.0668, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 13:11:34,752 INFO [train.py:968] (1/2) Epoch 10, batch 44700, giga_loss[loss=0.3521, simple_loss=0.4078, pruned_loss=0.1482, over 28801.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3914, pruned_loss=0.1339, over 5664100.06 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.377, pruned_loss=0.1284, over 5655895.42 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3919, pruned_loss=0.134, over 5665860.56 frames. ], batch size: 243, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:11:56,083 INFO [optim.py:369] (1/2) 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 13:12:18,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 13:12:21,183 INFO [train.py:968] (1/2) Epoch 10, batch 44750, giga_loss[loss=0.3164, simple_loss=0.3767, pruned_loss=0.128, over 28050.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3894, pruned_loss=0.133, over 5674890.89 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3765, pruned_loss=0.1279, over 5661695.58 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3906, pruned_loss=0.1337, over 5671217.41 frames. ], batch size: 412, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:12:22,789 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-05 13:12:28,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:30,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:39,056 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:42,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:47,335 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:12:56,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:13:03,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2553, 1.3193, 3.2333, 3.1512], device='cuda:1'), covar=tensor([0.1295, 0.2230, 0.0428, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0657, 0.0585, 0.0857, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:13:08,243 INFO [train.py:968] (1/2) Epoch 10, batch 44800, giga_loss[loss=0.3375, simple_loss=0.393, pruned_loss=0.141, over 28802.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3881, pruned_loss=0.1333, over 5658508.89 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3762, pruned_loss=0.1276, over 5657769.81 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3896, pruned_loss=0.1342, over 5658764.85 frames. ], batch size: 262, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:13:10,441 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:13:30,266 INFO [optim.py:369] (1/2) 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 13:13:34,806 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:13:56,197 INFO [train.py:968] (1/2) Epoch 10, batch 44850, giga_loss[loss=0.3416, simple_loss=0.3973, pruned_loss=0.1429, over 28900.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3865, pruned_loss=0.1335, over 5658727.79 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3762, pruned_loss=0.1276, over 5663110.81 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3879, pruned_loss=0.1343, over 5653940.69 frames. ], batch size: 213, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:14:00,544 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:14:24,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 13:14:33,384 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 44900, giga_loss[loss=0.2897, simple_loss=0.3549, pruned_loss=0.1122, over 28917.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3833, pruned_loss=0.132, over 5658439.01 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3762, pruned_loss=0.1276, over 5666915.50 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3847, pruned_loss=0.1328, over 5650963.50 frames. ], batch size: 136, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:15:03,038 INFO [optim.py:369] (1/2) 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 13:15:08,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:15:25,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1680, 0.8300, 0.8659, 1.3399], device='cuda:1'), covar=tensor([0.0726, 0.0379, 0.0338, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0057, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 13:15:27,233 INFO [train.py:968] (1/2) Epoch 10, batch 44950, giga_loss[loss=0.2828, simple_loss=0.3509, pruned_loss=0.1074, over 28926.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3817, pruned_loss=0.1314, over 5667220.70 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3762, pruned_loss=0.1276, over 5673557.64 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3831, pruned_loss=0.1322, over 5654914.89 frames. ], batch size: 145, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:15:40,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9031, 2.9254, 1.8687, 1.0781], device='cuda:1'), covar=tensor([0.4488, 0.1685, 0.2654, 0.4176], device='cuda:1'), in_proj_covar=tensor([0.1529, 0.1459, 0.1467, 0.1260], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 13:16:13,742 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:14,830 INFO [train.py:968] (1/2) Epoch 10, batch 45000, giga_loss[loss=0.2957, simple_loss=0.3653, pruned_loss=0.113, over 28956.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.381, pruned_loss=0.1306, over 5675203.67 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3763, pruned_loss=0.1276, over 5678619.44 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3821, pruned_loss=0.1313, over 5660861.25 frames. ], batch size: 164, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:16:14,830 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 13:16:23,769 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 13:16:24,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:39,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:42,123 INFO [optim.py:369] (1/2) 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 13:16:45,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:48,854 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:50,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:16:56,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 13:17:06,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5603, 1.5508, 1.2585, 1.1432], device='cuda:1'), covar=tensor([0.0718, 0.0534, 0.0889, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0443, 0.0495, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:17:06,583 INFO [train.py:968] (1/2) Epoch 10, batch 45050, giga_loss[loss=0.3173, simple_loss=0.3887, pruned_loss=0.123, over 27924.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3795, pruned_loss=0.1289, over 5674352.23 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3767, pruned_loss=0.1278, over 5683563.44 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3801, pruned_loss=0.1293, over 5658330.24 frames. ], batch size: 412, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:17:28,283 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:17:31,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:17:50,756 INFO [train.py:968] (1/2) Epoch 10, batch 45100, giga_loss[loss=0.2728, simple_loss=0.3444, pruned_loss=0.1006, over 28857.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3762, pruned_loss=0.1254, over 5678325.32 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3768, pruned_loss=0.1279, over 5687037.58 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3765, pruned_loss=0.1256, over 5661939.82 frames. ], batch size: 119, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:17:52,991 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:18:03,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2324, 1.7428, 1.3147, 0.3762], device='cuda:1'), covar=tensor([0.2728, 0.1807, 0.2864, 0.4011], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1470, 0.1477, 0.1265], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 13:18:10,328 INFO [optim.py:369] (1/2) 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 13:18:35,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2542, 3.0952, 2.9513, 1.4671], device='cuda:1'), covar=tensor([0.0864, 0.0928, 0.0839, 0.2138], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0987, 0.0863, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 13:18:40,267 INFO [train.py:968] (1/2) Epoch 10, batch 45150, giga_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1235, over 28686.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.124, over 5677521.67 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3764, pruned_loss=0.1276, over 5691333.54 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3752, pruned_loss=0.1243, over 5660526.21 frames. ], batch size: 92, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:18:43,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:18:55,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:18:57,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:02,475 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:05,821 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 45200, giga_loss[loss=0.2853, simple_loss=0.3505, pruned_loss=0.11, over 28830.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5660120.75 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3766, pruned_loss=0.1277, over 5686291.75 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.124, over 5649956.78 frames. ], batch size: 199, lr: 3.16e-03, grad_scale: 8.0
+2023-03-05 13:19:25,188 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:26,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:33,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:19:51,792 INFO [optim.py:369] (1/2) 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 13:20:17,587 INFO [train.py:968] (1/2) Epoch 10, batch 45250, giga_loss[loss=0.3704, simple_loss=0.4052, pruned_loss=0.1678, over 28530.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3714, pruned_loss=0.1241, over 5642654.33 frames. ], libri_tot_loss[loss=0.316, simple_loss=0.3766, pruned_loss=0.1277, over 5676922.29 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3716, pruned_loss=0.1242, over 5642271.44 frames. ], batch size: 60, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:20:30,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:20:55,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-05 13:20:59,091 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:01,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=454892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:03,128 INFO [train.py:968] (1/2) Epoch 10, batch 45300, giga_loss[loss=0.302, simple_loss=0.3794, pruned_loss=0.1123, over 28832.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3738, pruned_loss=0.1256, over 5633169.98 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3772, pruned_loss=0.1282, over 5669972.08 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3732, pruned_loss=0.1251, over 5638124.48 frames. ], batch size: 174, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:21:03,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5850, 1.0232, 2.8804, 2.6000], device='cuda:1'), covar=tensor([0.1913, 0.2579, 0.0563, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0659, 0.0587, 0.0857, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:21:23,130 INFO [optim.py:369] (1/2) 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 13:21:25,131 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 13:21:28,111 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:43,856 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=454937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:44,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=454938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:21:46,541 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 45350, giga_loss[loss=0.323, simple_loss=0.3845, pruned_loss=0.1308, over 28719.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3761, pruned_loss=0.1265, over 5640584.87 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3772, pruned_loss=0.1281, over 5673391.60 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3757, pruned_loss=0.1261, over 5640965.60 frames. ], batch size: 284, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:22:10,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5468, 1.6520, 1.5393, 1.5497], device='cuda:1'), covar=tensor([0.1482, 0.1929, 0.2028, 0.1695], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0737, 0.0672, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 13:22:14,815 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=454970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:31,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=454987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:35,531 INFO [train.py:968] (1/2) Epoch 10, batch 45400, giga_loss[loss=0.4047, simple_loss=0.4329, pruned_loss=0.1882, over 27556.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3765, pruned_loss=0.1268, over 5633879.59 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3776, pruned_loss=0.1282, over 5677582.48 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3757, pruned_loss=0.1265, over 5629615.26 frames. ], batch size: 472, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:22:42,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:44,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:22:55,660 INFO [optim.py:369] (1/2) 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 13:23:08,911 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:23:18,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-05 13:23:19,145 INFO [train.py:968] (1/2) Epoch 10, batch 45450, giga_loss[loss=0.314, simple_loss=0.3766, pruned_loss=0.1257, over 28854.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3758, pruned_loss=0.1265, over 5631356.97 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1284, over 5670740.85 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3749, pruned_loss=0.126, over 5632122.99 frames. ], batch size: 112, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:23:36,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 13:24:03,408 INFO [train.py:968] (1/2) Epoch 10, batch 45500, giga_loss[loss=0.3818, simple_loss=0.4189, pruned_loss=0.1723, over 27538.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3767, pruned_loss=0.1275, over 5628584.77 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5663825.96 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3759, pruned_loss=0.1271, over 5635043.84 frames. ], batch size: 472, lr: 3.16e-03, grad_scale: 4.0
+2023-03-05 13:24:22,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:24:24,754 INFO [optim.py:369] (1/2) 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 13:24:39,221 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:24:42,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:24:50,410 INFO [train.py:968] (1/2) Epoch 10, batch 45550, giga_loss[loss=0.3684, simple_loss=0.4036, pruned_loss=0.1667, over 26647.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3796, pruned_loss=0.1293, over 5647000.77 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3784, pruned_loss=0.1287, over 5670892.96 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3785, pruned_loss=0.1287, over 5645090.91 frames. ], batch size: 555, lr: 3.15e-03, grad_scale: 2.0
+2023-03-05 13:25:09,807 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:25:35,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455193.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:25:36,028 INFO [train.py:968] (1/2) Epoch 10, batch 45600, giga_loss[loss=0.3525, simple_loss=0.3984, pruned_loss=0.1533, over 27974.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3818, pruned_loss=0.131, over 5655133.24 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3787, pruned_loss=0.1289, over 5673604.03 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3807, pruned_loss=0.1303, over 5650499.78 frames. ], batch size: 412, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:25:40,274 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:25:45,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3234, 1.6037, 1.3189, 1.3267], device='cuda:1'), covar=tensor([0.2357, 0.2345, 0.2578, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.0961, 0.1140, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 13:26:00,837 INFO [optim.py:369] (1/2) 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 13:26:21,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4152, 1.5984, 1.4091, 1.3498], device='cuda:1'), covar=tensor([0.1743, 0.1663, 0.1424, 0.1559], device='cuda:1'), in_proj_covar=tensor([0.1683, 0.1600, 0.1565, 0.1670], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 13:26:27,851 INFO [train.py:968] (1/2) Epoch 10, batch 45650, giga_loss[loss=0.3908, simple_loss=0.4142, pruned_loss=0.1837, over 23415.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3838, pruned_loss=0.1332, over 5652687.42 frames. ], libri_tot_loss[loss=0.3184, simple_loss=0.3788, pruned_loss=0.129, over 5676995.85 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3829, pruned_loss=0.1326, over 5645603.61 frames. ], batch size: 705, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:27:13,971 INFO [train.py:968] (1/2) Epoch 10, batch 45700, giga_loss[loss=0.3354, simple_loss=0.3981, pruned_loss=0.1363, over 27991.00 frames. ], tot_loss[loss=0.325, simple_loss=0.384, pruned_loss=0.133, over 5651304.17 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3787, pruned_loss=0.1288, over 5675190.20 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3835, pruned_loss=0.1328, over 5646545.67 frames. ], batch size: 412, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:27:31,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:27:37,813 INFO [optim.py:369] (1/2) 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 13:28:09,936 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 10, batch 45750, giga_loss[loss=0.2708, simple_loss=0.3393, pruned_loss=0.1011, over 29028.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3831, pruned_loss=0.1307, over 5643010.47 frames. ], libri_tot_loss[loss=0.3183, simple_loss=0.3787, pruned_loss=0.129, over 5666650.70 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3827, pruned_loss=0.1304, over 5646467.88 frames. ], batch size: 120, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:28:57,723 INFO [train.py:968] (1/2) Epoch 10, batch 45800, giga_loss[loss=0.314, simple_loss=0.3714, pruned_loss=0.1283, over 28774.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3832, pruned_loss=0.1314, over 5610669.52 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3787, pruned_loss=0.1292, over 5633641.01 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.383, pruned_loss=0.1311, over 5642392.17 frames. ], batch size: 119, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:29:17,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7117, 1.8118, 1.5828, 1.8299], device='cuda:1'), covar=tensor([0.1805, 0.1669, 0.1661, 0.1619], device='cuda:1'), in_proj_covar=tensor([0.1284, 0.0955, 0.1138, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 13:29:20,086 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4508, 4.2822, 4.0694, 1.9192], device='cuda:1'), covar=tensor([0.0534, 0.0688, 0.0739, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.1060, 0.0993, 0.0874, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 13:29:21,941 INFO [optim.py:369] (1/2) 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 13:29:45,976 INFO [train.py:968] (1/2) Epoch 10, batch 45850, giga_loss[loss=0.3125, simple_loss=0.3687, pruned_loss=0.1281, over 28870.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3821, pruned_loss=0.1313, over 5589251.52 frames. ], libri_tot_loss[loss=0.3195, simple_loss=0.3794, pruned_loss=0.1298, over 5598324.73 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3814, pruned_loss=0.1305, over 5646070.50 frames. ], batch size: 112, lr: 3.15e-03, grad_scale: 4.0
+2023-03-05 13:29:59,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:01,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:35,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:39,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:30:41,476 INFO [train.py:968] (1/2) Epoch 10, batch 45900, giga_loss[loss=0.3104, simple_loss=0.3733, pruned_loss=0.1237, over 28711.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3833, pruned_loss=0.1334, over 5547373.03 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.38, pruned_loss=0.1304, over 5546956.83 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3823, pruned_loss=0.1323, over 5636588.34 frames. ], batch size: 284, lr: 3.15e-03, grad_scale: 2.0
+2023-03-05 13:31:07,390 INFO [optim.py:369] (1/2) 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 13:31:30,717 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-05 13:32:31,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455568.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:32:36,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:32:50,179 INFO [train.py:968] (1/2) Epoch 11, batch 50, giga_loss[loss=0.3068, simple_loss=0.387, pruned_loss=0.1133, over 28689.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3816, pruned_loss=0.1157, over 1257796.82 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3689, pruned_loss=0.1073, over 161499.79 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3834, pruned_loss=0.1169, over 1127881.09 frames. ], batch size: 262, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:33:05,972 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:33:16,660 INFO [optim.py:369] (1/2) 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 13:33:31,513 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:33:34,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:33:37,277 INFO [train.py:968] (1/2) Epoch 11, batch 100, giga_loss[loss=0.2872, simple_loss=0.3622, pruned_loss=0.1061, over 28593.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3742, pruned_loss=0.1129, over 2227546.75 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3567, pruned_loss=0.0975, over 322762.99 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3768, pruned_loss=0.1151, over 2020233.61 frames. ], batch size: 307, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:33:50,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6162, 1.8484, 1.8786, 1.4346], device='cuda:1'), covar=tensor([0.1772, 0.2338, 0.1443, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0704, 0.0860, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 13:34:01,681 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:34:21,684 INFO [train.py:968] (1/2) Epoch 11, batch 150, libri_loss[loss=0.2772, simple_loss=0.3523, pruned_loss=0.101, over 29538.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3583, pruned_loss=0.1051, over 2998972.27 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3587, pruned_loss=0.1019, over 433647.80 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3586, pruned_loss=0.1056, over 2775787.81 frames. ], batch size: 89, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:34:39,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7177, 1.7651, 1.2262, 1.5337], device='cuda:1'), covar=tensor([0.0730, 0.0642, 0.0986, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0442, 0.0494, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:34:42,455 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455711.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:34:45,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455714.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:34:46,704 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:34:48,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:34:49,274 INFO [optim.py:369] (1/2) 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 13:34:49,601 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 200, libri_loss[loss=0.2303, simple_loss=0.3068, pruned_loss=0.07687, over 29656.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3445, pruned_loss=0.0984, over 3603515.76 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3557, pruned_loss=0.1, over 595607.17 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3439, pruned_loss=0.09858, over 3354731.65 frames. ], batch size: 73, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:35:06,951 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455743.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 13:35:10,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7671, 2.0781, 1.7161, 2.1952], device='cuda:1'), covar=tensor([0.2286, 0.2225, 0.2402, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.1286, 0.0955, 0.1139, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 13:35:10,757 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:35:26,538 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:35:45,562 INFO [train.py:968] (1/2) Epoch 11, batch 250, libri_loss[loss=0.2445, simple_loss=0.3241, pruned_loss=0.08243, over 28550.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3345, pruned_loss=0.09371, over 4055000.59 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3543, pruned_loss=0.09899, over 738106.75 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3332, pruned_loss=0.09363, over 3812955.24 frames. ], batch size: 63, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:35:58,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 13:36:07,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-05 13:36:10,371 INFO [optim.py:369] (1/2) 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 13:36:29,744 INFO [train.py:968] (1/2) Epoch 11, batch 300, giga_loss[loss=0.2324, simple_loss=0.2989, pruned_loss=0.08296, over 28892.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3242, pruned_loss=0.08934, over 4416652.79 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3519, pruned_loss=0.09763, over 789799.38 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3229, pruned_loss=0.08921, over 4211503.81 frames. ], batch size: 112, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:36:48,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=455861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:36:51,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=455864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:37:15,832 INFO [train.py:968] (1/2) Epoch 11, batch 350, giga_loss[loss=0.2238, simple_loss=0.2963, pruned_loss=0.07568, over 29012.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.317, pruned_loss=0.08562, over 4697984.81 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3517, pruned_loss=0.09713, over 916319.45 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3149, pruned_loss=0.08517, over 4504530.06 frames. ], batch size: 227, lr: 3.01e-03, grad_scale: 2.0
+2023-03-05 13:37:16,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.44 vs. limit=5.0
+2023-03-05 13:37:18,577 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=455893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:37:40,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4757, 2.1410, 1.6466, 0.6862], device='cuda:1'), covar=tensor([0.4145, 0.2245, 0.2986, 0.4332], device='cuda:1'), in_proj_covar=tensor([0.1523, 0.1461, 0.1464, 0.1251], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 13:37:41,525 INFO [optim.py:369] (1/2) 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 13:37:47,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4729, 1.4922, 1.1732, 1.6513], device='cuda:1'), covar=tensor([0.0772, 0.0323, 0.0340, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 13:37:59,488 INFO [train.py:968] (1/2) Epoch 11, batch 400, giga_loss[loss=0.2497, simple_loss=0.3063, pruned_loss=0.09659, over 26681.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3121, pruned_loss=0.08325, over 4922328.76 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3528, pruned_loss=0.09786, over 966196.00 frames. ], giga_tot_loss[loss=0.2374, simple_loss=0.3097, pruned_loss=0.08254, over 4760336.24 frames. ], batch size: 555, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:38:02,341 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=455943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:38:18,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5158, 2.0995, 1.6614, 0.6975], device='cuda:1'), covar=tensor([0.4259, 0.2483, 0.3003, 0.4760], device='cuda:1'), in_proj_covar=tensor([0.1521, 0.1458, 0.1464, 0.1253], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 13:38:20,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3677, 4.1491, 3.9000, 1.7465], device='cuda:1'), covar=tensor([0.0578, 0.0827, 0.0841, 0.2292], device='cuda:1'), in_proj_covar=tensor([0.1040, 0.0977, 0.0858, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 13:38:34,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=455983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:38:40,567 INFO [train.py:968] (1/2) Epoch 11, batch 450, giga_loss[loss=0.2643, simple_loss=0.3066, pruned_loss=0.111, over 24003.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3096, pruned_loss=0.08218, over 5091189.40 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3507, pruned_loss=0.09634, over 1085224.49 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3071, pruned_loss=0.08152, over 4944881.70 frames. ], batch size: 705, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:38:54,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:39:07,937 INFO [optim.py:369] (1/2) 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 13:39:26,874 INFO [train.py:968] (1/2) Epoch 11, batch 500, giga_loss[loss=0.2255, simple_loss=0.2942, pruned_loss=0.07835, over 28555.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3061, pruned_loss=0.08025, over 5220443.97 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3491, pruned_loss=0.09567, over 1157321.90 frames. ], giga_tot_loss[loss=0.2315, simple_loss=0.3038, pruned_loss=0.0796, over 5094718.87 frames. ], batch size: 78, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:40:09,695 INFO [train.py:968] (1/2) Epoch 11, batch 550, giga_loss[loss=0.2381, simple_loss=0.2901, pruned_loss=0.09305, over 23897.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3041, pruned_loss=0.07939, over 5326975.01 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3495, pruned_loss=0.09598, over 1272787.85 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3011, pruned_loss=0.07841, over 5214447.12 frames. ], batch size: 705, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:40:37,816 INFO [optim.py:369] (1/2) 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 13:40:44,988 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:40:47,937 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 600, giga_loss[loss=0.2018, simple_loss=0.2765, pruned_loss=0.06356, over 29047.00 frames. ], tot_loss[loss=0.2289, simple_loss=0.3019, pruned_loss=0.07792, over 5412739.26 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3509, pruned_loss=0.0968, over 1317768.65 frames. ], giga_tot_loss[loss=0.2262, simple_loss=0.2987, pruned_loss=0.07681, over 5320566.52 frames. ], batch size: 136, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:41:03,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:41:14,987 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:41:35,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5317, 2.0699, 1.6249, 0.7008], device='cuda:1'), covar=tensor([0.3671, 0.2097, 0.3006, 0.4164], device='cuda:1'), in_proj_covar=tensor([0.1515, 0.1450, 0.1460, 0.1251], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 13:41:39,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1591, 1.3179, 1.0870, 0.9033], device='cuda:1'), covar=tensor([0.0898, 0.0476, 0.1136, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0438, 0.0491, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:41:44,580 INFO [train.py:968] (1/2) Epoch 11, batch 650, giga_loss[loss=0.1928, simple_loss=0.278, pruned_loss=0.05379, over 28818.00 frames. ], tot_loss[loss=0.228, simple_loss=0.301, pruned_loss=0.07752, over 5480459.08 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3522, pruned_loss=0.09778, over 1474415.24 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.2966, pruned_loss=0.07576, over 5393224.04 frames. ], batch size: 174, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:41:48,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-05 13:42:11,223 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 700, giga_loss[loss=0.2069, simple_loss=0.2681, pruned_loss=0.07289, over 24095.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.2982, pruned_loss=0.07611, over 5525604.54 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3515, pruned_loss=0.09693, over 1562424.90 frames. ], giga_tot_loss[loss=0.2215, simple_loss=0.2939, pruned_loss=0.07454, over 5449237.78 frames. ], batch size: 705, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:42:56,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 13:43:05,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7272, 2.0767, 2.0976, 1.6061], device='cuda:1'), covar=tensor([0.1916, 0.2188, 0.1433, 0.1618], device='cuda:1'), in_proj_covar=tensor([0.0841, 0.0714, 0.0877, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 13:43:16,919 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:20,245 INFO [train.py:968] (1/2) Epoch 11, batch 750, giga_loss[loss=0.1962, simple_loss=0.2694, pruned_loss=0.06153, over 28898.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.2954, pruned_loss=0.07488, over 5555959.40 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3509, pruned_loss=0.09665, over 1627338.01 frames. ], giga_tot_loss[loss=0.2191, simple_loss=0.2914, pruned_loss=0.07343, over 5490993.41 frames. ], batch size: 186, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:43:20,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=456289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:45,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:45,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:43:45,579 INFO [optim.py:369] (1/2) 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 13:44:04,146 INFO [train.py:968] (1/2) Epoch 11, batch 800, libri_loss[loss=0.2845, simple_loss=0.3668, pruned_loss=0.1011, over 29680.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.2964, pruned_loss=0.07592, over 5579815.38 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3519, pruned_loss=0.09754, over 1701982.63 frames. ], giga_tot_loss[loss=0.2201, simple_loss=0.292, pruned_loss=0.07414, over 5531462.91 frames. ], batch size: 88, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:44:16,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2667, 4.0410, 3.8075, 1.8838], device='cuda:1'), covar=tensor([0.0485, 0.0697, 0.0700, 0.2134], device='cuda:1'), in_proj_covar=tensor([0.1022, 0.0960, 0.0846, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 13:44:43,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:44:54,258 INFO [train.py:968] (1/2) Epoch 11, batch 850, giga_loss[loss=0.2722, simple_loss=0.3466, pruned_loss=0.09889, over 28576.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3086, pruned_loss=0.08243, over 5602868.94 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3524, pruned_loss=0.09768, over 1805665.41 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3038, pruned_loss=0.08057, over 5557531.60 frames. ], batch size: 60, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:45:24,986 INFO [optim.py:369] (1/2) 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 13:45:25,427 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 13:45:40,362 INFO [train.py:968] (1/2) Epoch 11, batch 900, giga_loss[loss=0.3069, simple_loss=0.3817, pruned_loss=0.116, over 28849.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3217, pruned_loss=0.08908, over 5628228.82 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.352, pruned_loss=0.09759, over 1886931.55 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3175, pruned_loss=0.08748, over 5587116.28 frames. ], batch size: 186, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:45:59,648 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:02,279 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=456464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:22,314 INFO [train.py:968] (1/2) Epoch 11, batch 950, libri_loss[loss=0.2659, simple_loss=0.3508, pruned_loss=0.09045, over 29535.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3321, pruned_loss=0.09407, over 5639019.31 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3522, pruned_loss=0.09768, over 2004316.25 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.328, pruned_loss=0.09266, over 5599815.06 frames. ], batch size: 84, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:46:27,085 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:29,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0635, 1.0893, 3.9845, 3.1155], device='cuda:1'), covar=tensor([0.1798, 0.2963, 0.0410, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0651, 0.0581, 0.0850, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:46:46,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:46,901 INFO [optim.py:369] (1/2) 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 13:46:48,418 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=456522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:46:50,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=456525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:47:02,181 INFO [train.py:968] (1/2) Epoch 11, batch 1000, giga_loss[loss=0.2934, simple_loss=0.3671, pruned_loss=0.1098, over 28978.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3387, pruned_loss=0.09653, over 5647937.76 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3534, pruned_loss=0.09887, over 2151780.85 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3344, pruned_loss=0.0949, over 5612339.10 frames. ], batch size: 213, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:47:08,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4846, 4.0435, 1.6092, 1.6904], device='cuda:1'), covar=tensor([0.0932, 0.0242, 0.0872, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0503, 0.0336, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 13:47:10,231 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:47:14,600 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=456554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:47:40,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5234, 2.0626, 1.8911, 1.3876], device='cuda:1'), covar=tensor([0.1772, 0.2636, 0.1522, 0.1924], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0708, 0.0870, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 13:47:41,223 INFO [train.py:968] (1/2) Epoch 11, batch 1050, giga_loss[loss=0.262, simple_loss=0.3445, pruned_loss=0.08973, over 29055.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3427, pruned_loss=0.09719, over 5665860.04 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3538, pruned_loss=0.09911, over 2242999.75 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3389, pruned_loss=0.09575, over 5633324.77 frames. ], batch size: 155, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:48:12,308 INFO [optim.py:369] (1/2) 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 13:48:23,134 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 13:48:27,960 INFO [train.py:968] (1/2) Epoch 11, batch 1100, giga_loss[loss=0.3068, simple_loss=0.3794, pruned_loss=0.1171, over 28929.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3447, pruned_loss=0.09797, over 5664021.43 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3543, pruned_loss=0.09954, over 2316043.53 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3413, pruned_loss=0.09662, over 5633347.46 frames. ], batch size: 213, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:48:41,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2455, 1.3839, 1.4136, 1.2552], device='cuda:1'), covar=tensor([0.1276, 0.1504, 0.1774, 0.1488], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0730, 0.0664, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 13:48:45,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4467, 1.9786, 1.4270, 0.6052], device='cuda:1'), covar=tensor([0.3355, 0.2075, 0.3264, 0.3914], device='cuda:1'), in_proj_covar=tensor([0.1522, 0.1453, 0.1475, 0.1258], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 13:49:13,844 INFO [train.py:968] (1/2) Epoch 11, batch 1150, giga_loss[loss=0.2927, simple_loss=0.3651, pruned_loss=0.1102, over 27885.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3469, pruned_loss=0.09973, over 5668986.86 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3538, pruned_loss=0.09919, over 2352026.68 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3443, pruned_loss=0.0988, over 5642598.07 frames. ], batch size: 412, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:49:32,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6064, 1.6274, 1.3706, 1.9895], device='cuda:1'), covar=tensor([0.2279, 0.2395, 0.2430, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1283, 0.0955, 0.1137, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 13:49:41,258 INFO [optim.py:369] (1/2) 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 13:49:58,612 INFO [train.py:968] (1/2) Epoch 11, batch 1200, giga_loss[loss=0.2945, simple_loss=0.3628, pruned_loss=0.1131, over 28980.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3502, pruned_loss=0.1024, over 5674072.96 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3527, pruned_loss=0.0986, over 2439603.30 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3485, pruned_loss=0.102, over 5648356.20 frames. ], batch size: 136, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:50:37,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 13:50:40,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5697, 1.6366, 1.8688, 1.4195], device='cuda:1'), covar=tensor([0.1665, 0.2139, 0.1288, 0.1518], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0705, 0.0870, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 13:50:41,098 INFO [train.py:968] (1/2) Epoch 11, batch 1250, giga_loss[loss=0.3246, simple_loss=0.3846, pruned_loss=0.1323, over 27581.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3537, pruned_loss=0.1045, over 5684204.84 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3529, pruned_loss=0.09875, over 2542169.12 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3522, pruned_loss=0.1042, over 5657613.60 frames. ], batch size: 472, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:51:03,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456815.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:51:07,207 INFO [optim.py:369] (1/2) 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 13:51:21,606 INFO [train.py:968] (1/2) Epoch 11, batch 1300, giga_loss[loss=0.3059, simple_loss=0.3787, pruned_loss=0.1165, over 28713.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3569, pruned_loss=0.1053, over 5693499.78 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.353, pruned_loss=0.09864, over 2653498.73 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3557, pruned_loss=0.1053, over 5669383.78 frames. ], batch size: 284, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:51:29,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6494, 1.7644, 1.5198, 2.0146], device='cuda:1'), covar=tensor([0.2368, 0.2430, 0.2483, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.1287, 0.0956, 0.1142, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 13:52:04,326 INFO [train.py:968] (1/2) Epoch 11, batch 1350, giga_loss[loss=0.3061, simple_loss=0.3797, pruned_loss=0.1162, over 28976.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3589, pruned_loss=0.1062, over 5688075.50 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3537, pruned_loss=0.09924, over 2701638.81 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3578, pruned_loss=0.106, over 5666430.47 frames. ], batch size: 227, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:52:10,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:52:31,842 INFO [optim.py:369] (1/2) 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 13:52:32,086 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=456921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:52:34,169 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=456924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:52:45,005 INFO [train.py:968] (1/2) Epoch 11, batch 1400, giga_loss[loss=0.2804, simple_loss=0.3614, pruned_loss=0.09966, over 28892.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.359, pruned_loss=0.1049, over 5701141.06 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3535, pruned_loss=0.09901, over 2780632.34 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3582, pruned_loss=0.105, over 5679532.29 frames. ], batch size: 112, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:53:29,648 INFO [train.py:968] (1/2) Epoch 11, batch 1450, giga_loss[loss=0.2535, simple_loss=0.3437, pruned_loss=0.08165, over 28114.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3583, pruned_loss=0.1035, over 5700076.04 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3532, pruned_loss=0.09869, over 2842380.20 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.358, pruned_loss=0.1038, over 5679501.73 frames. ], batch size: 77, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:53:49,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3645, 3.7444, 1.5975, 1.4993], device='cuda:1'), covar=tensor([0.0973, 0.0228, 0.0846, 0.1363], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0500, 0.0334, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 13:53:52,295 INFO [optim.py:369] (1/2) 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 13:54:02,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-05 13:54:06,797 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:07,862 INFO [train.py:968] (1/2) Epoch 11, batch 1500, giga_loss[loss=0.2635, simple_loss=0.3469, pruned_loss=0.09005, over 28852.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3561, pruned_loss=0.1008, over 5710731.96 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3529, pruned_loss=0.09819, over 2945243.24 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3561, pruned_loss=0.1014, over 5690349.09 frames. ], batch size: 174, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:54:09,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:21,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=457057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:28,861 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:30,826 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:31,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:54:38,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8625, 1.8289, 1.3612, 1.5268], device='cuda:1'), covar=tensor([0.0775, 0.0658, 0.0989, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0438, 0.0495, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 13:54:47,722 INFO [train.py:968] (1/2) Epoch 11, batch 1550, giga_loss[loss=0.2893, simple_loss=0.36, pruned_loss=0.1093, over 29010.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3556, pruned_loss=0.1003, over 5715501.89 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3532, pruned_loss=0.09832, over 3056583.49 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3555, pruned_loss=0.1008, over 5696443.59 frames. ], batch size: 213, lr: 3.01e-03, grad_scale: 4.0
+2023-03-05 13:54:56,892 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:55:15,499 INFO [optim.py:369] (1/2) 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 13:55:31,316 INFO [train.py:968] (1/2) Epoch 11, batch 1600, giga_loss[loss=0.2749, simple_loss=0.352, pruned_loss=0.09892, over 28840.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3575, pruned_loss=0.1031, over 5708277.66 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3532, pruned_loss=0.0984, over 3155170.59 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3576, pruned_loss=0.1035, over 5686098.04 frames. ], batch size: 119, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:56:21,419 INFO [train.py:968] (1/2) Epoch 11, batch 1650, giga_loss[loss=0.2849, simple_loss=0.3484, pruned_loss=0.1107, over 28593.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3602, pruned_loss=0.1075, over 5712171.71 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3531, pruned_loss=0.0983, over 3168915.74 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3604, pruned_loss=0.1079, over 5693979.60 frames. ], batch size: 85, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:56:22,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=457190.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:56:46,243 INFO [optim.py:369] (1/2) 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 13:56:55,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-05 13:57:00,396 INFO [train.py:968] (1/2) Epoch 11, batch 1700, giga_loss[loss=0.2847, simple_loss=0.3511, pruned_loss=0.1091, over 28738.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3612, pruned_loss=0.1093, over 5720928.69 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3529, pruned_loss=0.09811, over 3287708.45 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3617, pruned_loss=0.1102, over 5700194.02 frames. ], batch size: 242, lr: 3.01e-03, grad_scale: 8.0
+2023-03-05 13:57:44,215 INFO [train.py:968] (1/2) Epoch 11, batch 1750, giga_loss[loss=0.3711, simple_loss=0.4076, pruned_loss=0.1673, over 28749.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3597, pruned_loss=0.1094, over 5711362.13 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3524, pruned_loss=0.09794, over 3376374.88 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3607, pruned_loss=0.1106, over 5690215.06 frames. ], batch size: 284, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 13:57:50,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=457296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:58:09,969 INFO [optim.py:369] (1/2) 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 13:58:20,511 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:58:23,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457336.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:58:25,003 INFO [train.py:968] (1/2) Epoch 11, batch 1800, libri_loss[loss=0.2185, simple_loss=0.3012, pruned_loss=0.06792, over 29640.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3575, pruned_loss=0.1084, over 5706113.00 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.352, pruned_loss=0.09741, over 3439317.57 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3586, pruned_loss=0.1099, over 5684246.94 frames. ], batch size: 69, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 13:58:45,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457365.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 13:58:55,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4399, 2.0138, 1.4423, 0.5913], device='cuda:1'), covar=tensor([0.3690, 0.2012, 0.3087, 0.4457], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1440, 0.1462, 0.1247], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 13:59:04,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-05 13:59:05,078 INFO [train.py:968] (1/2) Epoch 11, batch 1850, giga_loss[loss=0.3324, simple_loss=0.3881, pruned_loss=0.1384, over 28038.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3563, pruned_loss=0.1074, over 5694508.82 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3521, pruned_loss=0.09767, over 3514348.50 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3574, pruned_loss=0.1087, over 5679603.04 frames. ], batch size: 412, lr: 3.00e-03, grad_scale: 2.0
+2023-03-05 13:59:34,132 INFO [optim.py:369] (1/2) 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 13:59:46,001 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 1900, giga_loss[loss=0.3165, simple_loss=0.369, pruned_loss=0.132, over 26589.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3543, pruned_loss=0.1053, over 5697407.03 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3517, pruned_loss=0.09738, over 3550172.02 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3554, pruned_loss=0.1067, over 5682656.52 frames. ], batch size: 555, lr: 3.00e-03, grad_scale: 2.0
+2023-03-05 13:59:52,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 13:59:54,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:00:23,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:00:31,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7209, 2.3835, 1.6654, 0.7992], device='cuda:1'), covar=tensor([0.5419, 0.2708, 0.2747, 0.4956], device='cuda:1'), in_proj_covar=tensor([0.1513, 0.1437, 0.1462, 0.1248], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 14:00:37,402 INFO [train.py:968] (1/2) Epoch 11, batch 1950, giga_loss[loss=0.2419, simple_loss=0.3282, pruned_loss=0.0778, over 28983.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3502, pruned_loss=0.1025, over 5695351.51 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3518, pruned_loss=0.09735, over 3627959.41 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.351, pruned_loss=0.1039, over 5679791.44 frames. ], batch size: 164, lr: 3.00e-03, grad_scale: 2.0
+2023-03-05 14:00:56,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4586, 1.8926, 1.5483, 1.5007], device='cuda:1'), covar=tensor([0.0776, 0.0284, 0.0291, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 14:01:10,210 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 2000, giga_loss[loss=0.2903, simple_loss=0.3424, pruned_loss=0.1191, over 26541.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3437, pruned_loss=0.09907, over 5685425.51 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3512, pruned_loss=0.09692, over 3671433.94 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3446, pruned_loss=0.1004, over 5670913.73 frames. ], batch size: 555, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:01:45,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-05 14:01:59,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=457575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:02:00,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1724, 2.5940, 1.2838, 1.2895], device='cuda:1'), covar=tensor([0.0985, 0.0286, 0.0879, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0496, 0.0333, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 14:02:03,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=457578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:02:13,220 INFO [train.py:968] (1/2) Epoch 11, batch 2050, giga_loss[loss=0.2102, simple_loss=0.2737, pruned_loss=0.07332, over 23363.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3379, pruned_loss=0.09551, over 5681214.32 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3513, pruned_loss=0.09675, over 3715257.41 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3383, pruned_loss=0.09669, over 5665956.09 frames. ], batch size: 705, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:02:30,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=457607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:02:32,388 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-05 14:02:34,030 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:03:01,856 INFO [train.py:968] (1/2) Epoch 11, batch 2100, giga_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09909, over 28803.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3361, pruned_loss=0.09476, over 5671561.55 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3509, pruned_loss=0.09638, over 3790142.28 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3362, pruned_loss=0.09588, over 5652631.76 frames. ], batch size: 243, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:03:09,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 2150, giga_loss[loss=0.2676, simple_loss=0.3492, pruned_loss=0.09303, over 28689.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3371, pruned_loss=0.09455, over 5680155.77 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3505, pruned_loss=0.096, over 3848805.27 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3371, pruned_loss=0.09564, over 5662192.93 frames. ], batch size: 242, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:04:05,974 INFO [optim.py:369] (1/2) 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 14:04:17,924 INFO [train.py:968] (1/2) Epoch 11, batch 2200, giga_loss[loss=0.2602, simple_loss=0.3384, pruned_loss=0.09099, over 28669.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3373, pruned_loss=0.09391, over 5696384.87 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3505, pruned_loss=0.09566, over 3970451.06 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3364, pruned_loss=0.09493, over 5675992.83 frames. ], batch size: 284, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:04:59,382 INFO [train.py:968] (1/2) Epoch 11, batch 2250, giga_loss[loss=0.2603, simple_loss=0.3326, pruned_loss=0.09402, over 28752.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3359, pruned_loss=0.09355, over 5699026.31 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3511, pruned_loss=0.09585, over 4006965.32 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3346, pruned_loss=0.09421, over 5680870.78 frames. ], batch size: 262, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:05:06,035 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=457797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:05:24,840 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 2300, giga_loss[loss=0.2742, simple_loss=0.3349, pruned_loss=0.1067, over 28896.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3349, pruned_loss=0.09329, over 5706748.66 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3514, pruned_loss=0.09583, over 4078206.32 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3332, pruned_loss=0.09377, over 5687891.20 frames. ], batch size: 112, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:05:59,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5384, 1.5501, 1.2457, 1.1854], device='cuda:1'), covar=tensor([0.0787, 0.0547, 0.1019, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0439, 0.0495, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 14:06:22,691 INFO [train.py:968] (1/2) Epoch 11, batch 2350, giga_loss[loss=0.2245, simple_loss=0.3035, pruned_loss=0.07271, over 28593.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3319, pruned_loss=0.09179, over 5709912.76 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.352, pruned_loss=0.09603, over 4113694.88 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3298, pruned_loss=0.09199, over 5692291.18 frames. ], batch size: 307, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:06:51,967 INFO [optim.py:369] (1/2) 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 14:07:07,238 INFO [train.py:968] (1/2) Epoch 11, batch 2400, giga_loss[loss=0.2227, simple_loss=0.3012, pruned_loss=0.07212, over 28837.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3296, pruned_loss=0.09118, over 5708422.08 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3517, pruned_loss=0.09579, over 4131047.91 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.328, pruned_loss=0.09145, over 5693260.36 frames. ], batch size: 174, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:07:17,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5002, 1.6934, 1.4915, 1.2548], device='cuda:1'), covar=tensor([0.1780, 0.1492, 0.1078, 0.1509], device='cuda:1'), in_proj_covar=tensor([0.1663, 0.1569, 0.1540, 0.1668], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 14:07:30,966 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 2450, giga_loss[loss=0.2292, simple_loss=0.3053, pruned_loss=0.07656, over 28408.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3281, pruned_loss=0.09034, over 5704424.54 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3528, pruned_loss=0.09611, over 4181674.77 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3254, pruned_loss=0.0902, over 5694910.11 frames. ], batch size: 71, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:08:12,732 INFO [optim.py:369] (1/2) 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 14:08:13,033 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:08:24,399 INFO [train.py:968] (1/2) Epoch 11, batch 2500, giga_loss[loss=0.2096, simple_loss=0.2897, pruned_loss=0.06477, over 28927.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3267, pruned_loss=0.08975, over 5714837.18 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3539, pruned_loss=0.0966, over 4231998.52 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3231, pruned_loss=0.08915, over 5702666.49 frames. ], batch size: 227, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:09:00,780 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-05 14:09:06,624 INFO [train.py:968] (1/2) Epoch 11, batch 2550, giga_loss[loss=0.2805, simple_loss=0.3451, pruned_loss=0.1079, over 26669.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3242, pruned_loss=0.08854, over 5707175.21 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3547, pruned_loss=0.09707, over 4246433.79 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3203, pruned_loss=0.08763, over 5711576.84 frames. ], batch size: 555, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:09:32,865 INFO [optim.py:369] (1/2) 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 14:09:39,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:09:41,381 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 2600, giga_loss[loss=0.246, simple_loss=0.324, pruned_loss=0.08399, over 28677.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3224, pruned_loss=0.08756, over 5713229.50 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3548, pruned_loss=0.09701, over 4278777.08 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3187, pruned_loss=0.08669, over 5713787.43 frames. ], batch size: 307, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:10:03,068 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:06,403 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:10,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:12,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:14,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:10:26,733 INFO [train.py:968] (1/2) Epoch 11, batch 2650, giga_loss[loss=0.2165, simple_loss=0.2967, pruned_loss=0.06819, over 28916.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3213, pruned_loss=0.08686, over 5711168.69 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3554, pruned_loss=0.09716, over 4308311.71 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3172, pruned_loss=0.08584, over 5717064.97 frames. ], batch size: 186, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:10:34,549 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 2700, giga_loss[loss=0.2722, simple_loss=0.3441, pruned_loss=0.1001, over 29028.00 frames. ], tot_loss[loss=0.252, simple_loss=0.325, pruned_loss=0.08953, over 5691429.84 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3563, pruned_loss=0.09795, over 4333585.09 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3201, pruned_loss=0.08789, over 5718682.71 frames. ], batch size: 136, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:11:30,371 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-05 14:11:47,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:11:55,072 INFO [train.py:968] (1/2) Epoch 11, batch 2750, giga_loss[loss=0.2542, simple_loss=0.3258, pruned_loss=0.09123, over 28792.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3302, pruned_loss=0.09292, over 5689676.25 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3566, pruned_loss=0.09819, over 4353069.60 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3257, pruned_loss=0.09135, over 5711846.72 frames. ], batch size: 66, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:12:19,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:12:21,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:12:26,530 INFO [optim.py:369] (1/2) 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 14:12:40,936 INFO [train.py:968] (1/2) Epoch 11, batch 2800, giga_loss[loss=0.3294, simple_loss=0.3921, pruned_loss=0.1333, over 28478.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3372, pruned_loss=0.0971, over 5696911.42 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3561, pruned_loss=0.09781, over 4389645.82 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3334, pruned_loss=0.09599, over 5711564.87 frames. ], batch size: 336, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:12:45,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:12:48,390 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 2850, giga_loss[loss=0.2894, simple_loss=0.3645, pruned_loss=0.1072, over 28669.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3454, pruned_loss=0.1029, over 5683194.90 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3562, pruned_loss=0.09789, over 4416247.50 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.342, pruned_loss=0.102, over 5694135.66 frames. ], batch size: 262, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:13:31,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3112, 3.1102, 2.9354, 1.3579], device='cuda:1'), covar=tensor([0.0838, 0.1006, 0.0891, 0.2266], device='cuda:1'), in_proj_covar=tensor([0.1019, 0.0948, 0.0836, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 14:13:31,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2907, 1.5788, 1.3254, 0.9621], device='cuda:1'), covar=tensor([0.2236, 0.2207, 0.2459, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1280, 0.0952, 0.1136, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 14:13:43,341 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:13:58,600 INFO [optim.py:369] (1/2) 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 14:14:12,078 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458438.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:14:12,517 INFO [train.py:968] (1/2) Epoch 11, batch 2900, giga_loss[loss=0.3059, simple_loss=0.3763, pruned_loss=0.1177, over 28635.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3493, pruned_loss=0.104, over 5697428.45 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3559, pruned_loss=0.0978, over 4465989.95 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3464, pruned_loss=0.1035, over 5701160.75 frames. ], batch size: 60, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:14:57,725 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 2950, libri_loss[loss=0.2367, simple_loss=0.3201, pruned_loss=0.07672, over 29575.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3534, pruned_loss=0.105, over 5694768.62 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3558, pruned_loss=0.0977, over 4510989.94 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.351, pruned_loss=0.1049, over 5700380.04 frames. ], batch size: 75, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:14:59,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:15:24,110 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:15:32,032 INFO [optim.py:369] (1/2) 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 14:15:40,893 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 3000, giga_loss[loss=0.3169, simple_loss=0.3807, pruned_loss=0.1265, over 28834.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3593, pruned_loss=0.1094, over 5672364.85 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3554, pruned_loss=0.09754, over 4535736.34 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3578, pruned_loss=0.1097, over 5680714.90 frames. ], batch size: 186, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:15:45,347 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 14:15:51,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1881, 1.6747, 1.5215, 1.0943], device='cuda:1'), covar=tensor([0.1768, 0.2401, 0.1492, 0.1732], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0703, 0.0866, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 14:15:53,825 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 14:15:54,093 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:02,711 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:05,868 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:30,370 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:16:34,265 INFO [train.py:968] (1/2) Epoch 11, batch 3050, giga_loss[loss=0.2763, simple_loss=0.3444, pruned_loss=0.1041, over 28295.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3601, pruned_loss=0.1094, over 5673328.71 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3557, pruned_loss=0.09778, over 4567877.51 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3588, pruned_loss=0.1097, over 5681270.54 frames. ], batch size: 368, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:16:41,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 14:17:06,163 INFO [optim.py:369] (1/2) 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 14:17:07,027 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2326, 2.5938, 1.2957, 1.2937], device='cuda:1'), covar=tensor([0.0922, 0.0310, 0.0846, 0.1453], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0498, 0.0332, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 14:17:17,914 INFO [train.py:968] (1/2) Epoch 11, batch 3100, libri_loss[loss=0.2553, simple_loss=0.3297, pruned_loss=0.09051, over 29590.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3565, pruned_loss=0.107, over 5677306.40 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.356, pruned_loss=0.09824, over 4598185.34 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3553, pruned_loss=0.1072, over 5685143.54 frames. ], batch size: 74, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:17:19,684 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 14:17:31,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:17:37,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2575, 1.4631, 1.1932, 1.4713], device='cuda:1'), covar=tensor([0.0765, 0.0325, 0.0325, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0112, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 14:17:49,305 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:17:51,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:18:01,590 INFO [train.py:968] (1/2) Epoch 11, batch 3150, giga_loss[loss=0.3839, simple_loss=0.4339, pruned_loss=0.167, over 28034.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3541, pruned_loss=0.1043, over 5691051.38 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3557, pruned_loss=0.09811, over 4622157.15 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3533, pruned_loss=0.1047, over 5695086.39 frames. ], batch size: 412, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:18:04,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4320, 4.2446, 4.0156, 1.8444], device='cuda:1'), covar=tensor([0.0578, 0.0746, 0.0755, 0.2302], device='cuda:1'), in_proj_covar=tensor([0.1023, 0.0957, 0.0835, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 14:18:15,732 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:18:29,459 INFO [optim.py:369] (1/2) 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 14:18:43,750 INFO [train.py:968] (1/2) Epoch 11, batch 3200, libri_loss[loss=0.2815, simple_loss=0.3627, pruned_loss=0.1002, over 29498.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1035, over 5699657.32 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3557, pruned_loss=0.09814, over 4659378.81 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.353, pruned_loss=0.1039, over 5698094.78 frames. ], batch size: 85, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:18:45,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2559, 1.5224, 1.2619, 1.0364], device='cuda:1'), covar=tensor([0.1963, 0.1880, 0.2006, 0.1874], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.0951, 0.1137, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 14:18:57,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0767, 2.2623, 2.4195, 1.9157], device='cuda:1'), covar=tensor([0.1719, 0.1983, 0.1270, 0.1457], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0703, 0.0870, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 14:19:09,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1811, 3.9992, 3.7421, 1.7959], device='cuda:1'), covar=tensor([0.0502, 0.0649, 0.0645, 0.2465], device='cuda:1'), in_proj_covar=tensor([0.1024, 0.0961, 0.0839, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 14:19:19,779 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=458782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:19:24,370 INFO [train.py:968] (1/2) Epoch 11, batch 3250, giga_loss[loss=0.2901, simple_loss=0.3667, pruned_loss=0.1067, over 28595.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3566, pruned_loss=0.1051, over 5703257.80 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3557, pruned_loss=0.0983, over 4683616.81 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3561, pruned_loss=0.1055, over 5698869.67 frames. ], batch size: 78, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:19:25,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1392, 1.1510, 4.1357, 3.3325], device='cuda:1'), covar=tensor([0.1705, 0.2684, 0.0362, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0650, 0.0575, 0.0841, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 14:19:27,614 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 14:19:33,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:19:35,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:19:44,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458813.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:19:55,091 INFO [optim.py:369] (1/2) 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 14:19:59,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 3300, libri_loss[loss=0.2694, simple_loss=0.3531, pruned_loss=0.09289, over 29548.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.357, pruned_loss=0.105, over 5713212.23 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3549, pruned_loss=0.09767, over 4742890.15 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3572, pruned_loss=0.1061, over 5701211.58 frames. ], batch size: 82, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:20:42,385 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-05 14:20:45,182 INFO [train.py:968] (1/2) Epoch 11, batch 3350, giga_loss[loss=0.2808, simple_loss=0.3515, pruned_loss=0.1051, over 28751.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3572, pruned_loss=0.1053, over 5712646.04 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3545, pruned_loss=0.09765, over 4793964.63 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3577, pruned_loss=0.1066, over 5699421.67 frames. ], batch size: 99, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:21:06,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=458914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:21:16,150 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 3400, giga_loss[loss=0.3145, simple_loss=0.3852, pruned_loss=0.1219, over 29041.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3588, pruned_loss=0.107, over 5714799.96 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3546, pruned_loss=0.09763, over 4820710.17 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3593, pruned_loss=0.1082, over 5700485.86 frames. ], batch size: 155, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:21:42,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=458956.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:21:44,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=458959.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:22:11,045 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=458988.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:22:11,450 INFO [train.py:968] (1/2) Epoch 11, batch 3450, giga_loss[loss=0.296, simple_loss=0.3673, pruned_loss=0.1124, over 28859.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3588, pruned_loss=0.1072, over 5724329.51 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3542, pruned_loss=0.09734, over 4845998.77 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3596, pruned_loss=0.1085, over 5710173.57 frames. ], batch size: 145, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:22:42,069 INFO [optim.py:369] (1/2) 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 14:22:46,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-05 14:22:52,971 INFO [train.py:968] (1/2) Epoch 11, batch 3500, libri_loss[loss=0.2605, simple_loss=0.3393, pruned_loss=0.09086, over 29563.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3585, pruned_loss=0.1068, over 5729537.79 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3538, pruned_loss=0.09711, over 4872062.72 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3595, pruned_loss=0.1083, over 5714343.06 frames. ], batch size: 78, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:23:07,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:23:10,728 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 3550, giga_loss[loss=0.2549, simple_loss=0.3318, pruned_loss=0.089, over 28672.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3577, pruned_loss=0.1054, over 5727379.74 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3529, pruned_loss=0.09653, over 4902626.30 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3593, pruned_loss=0.1073, over 5710434.63 frames. ], batch size: 78, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:23:31,759 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:24:09,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=459131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:24:14,894 INFO [train.py:968] (1/2) Epoch 11, batch 3600, giga_loss[loss=0.2486, simple_loss=0.3375, pruned_loss=0.07986, over 29091.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3597, pruned_loss=0.1057, over 5716499.70 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3531, pruned_loss=0.09652, over 4910233.98 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.361, pruned_loss=0.1073, over 5711570.68 frames. ], batch size: 155, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:24:28,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=459157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:24:54,524 INFO [train.py:968] (1/2) Epoch 11, batch 3650, giga_loss[loss=0.2488, simple_loss=0.3305, pruned_loss=0.08352, over 29025.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3595, pruned_loss=0.1051, over 5719058.56 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3535, pruned_loss=0.09685, over 4941588.79 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3604, pruned_loss=0.1064, over 5712195.77 frames. ], batch size: 136, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:25:23,450 INFO [optim.py:369] (1/2) 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 14:25:28,712 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 3700, giga_loss[loss=0.288, simple_loss=0.3591, pruned_loss=0.1084, over 28966.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3571, pruned_loss=0.1039, over 5729533.62 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3536, pruned_loss=0.09708, over 4970474.78 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3579, pruned_loss=0.1051, over 5719081.02 frames. ], batch size: 199, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:25:39,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2247, 1.4288, 1.4175, 1.2839], device='cuda:1'), covar=tensor([0.1239, 0.1251, 0.1738, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0723, 0.0662, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 14:25:52,064 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 14:26:15,258 INFO [train.py:968] (1/2) Epoch 11, batch 3750, giga_loss[loss=0.2664, simple_loss=0.351, pruned_loss=0.0909, over 28886.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3564, pruned_loss=0.1038, over 5725718.65 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.354, pruned_loss=0.0971, over 4996027.26 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3568, pruned_loss=0.1049, over 5714502.57 frames. ], batch size: 174, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:26:22,916 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:26:25,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=459303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:26:42,289 INFO [optim.py:369] (1/2) 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 14:26:47,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-05 14:26:48,289 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=459332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:26:53,305 INFO [train.py:968] (1/2) Epoch 11, batch 3800, giga_loss[loss=0.2648, simple_loss=0.3417, pruned_loss=0.09393, over 28598.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3548, pruned_loss=0.1032, over 5720662.08 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3542, pruned_loss=0.09744, over 4998358.63 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.355, pruned_loss=0.1039, over 5719476.48 frames. ], batch size: 60, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:27:26,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3232, 1.2265, 1.0857, 1.5251], device='cuda:1'), covar=tensor([0.0783, 0.0343, 0.0324, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 14:27:35,320 INFO [train.py:968] (1/2) Epoch 11, batch 3850, giga_loss[loss=0.2562, simple_loss=0.3327, pruned_loss=0.08987, over 28413.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3544, pruned_loss=0.1029, over 5722275.38 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3536, pruned_loss=0.09701, over 5027451.87 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3551, pruned_loss=0.1041, over 5724477.70 frames. ], batch size: 65, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:28:03,858 INFO [optim.py:369] (1/2) 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 14:28:12,194 INFO [train.py:968] (1/2) Epoch 11, batch 3900, libri_loss[loss=0.2509, simple_loss=0.3341, pruned_loss=0.08386, over 29548.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.356, pruned_loss=0.1041, over 5729696.09 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3532, pruned_loss=0.0969, over 5070615.99 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3569, pruned_loss=0.1055, over 5723091.76 frames. ], batch size: 77, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:28:20,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0160, 1.4362, 1.2911, 1.2200], device='cuda:1'), covar=tensor([0.1750, 0.1430, 0.2023, 0.1645], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0724, 0.0665, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 14:28:51,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5195, 1.1015, 4.7701, 3.4968], device='cuda:1'), covar=tensor([0.1722, 0.2895, 0.0317, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0576, 0.0838, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 14:28:55,566 INFO [train.py:968] (1/2) Epoch 11, batch 3950, giga_loss[loss=0.2578, simple_loss=0.3415, pruned_loss=0.08704, over 28604.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3561, pruned_loss=0.1036, over 5722535.10 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3531, pruned_loss=0.09676, over 5081305.31 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3571, pruned_loss=0.1049, over 5716587.75 frames. ], batch size: 336, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:29:11,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=459506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:29:27,027 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 4000, giga_loss[loss=0.2601, simple_loss=0.3382, pruned_loss=0.09104, over 28898.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3553, pruned_loss=0.1028, over 5716026.90 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.354, pruned_loss=0.09759, over 5096503.64 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3553, pruned_loss=0.1034, over 5717403.68 frames. ], batch size: 213, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:29:52,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4207, 1.5533, 1.4630, 1.4104], device='cuda:1'), covar=tensor([0.1241, 0.1555, 0.1809, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0723, 0.0665, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 14:30:14,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4300, 1.6156, 1.5149, 1.5096], device='cuda:1'), covar=tensor([0.1347, 0.1430, 0.1593, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0720, 0.0662, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 14:30:17,898 INFO [train.py:968] (1/2) Epoch 11, batch 4050, giga_loss[loss=0.3037, simple_loss=0.3781, pruned_loss=0.1147, over 28551.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3548, pruned_loss=0.1031, over 5714741.44 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3542, pruned_loss=0.0975, over 5116236.40 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3548, pruned_loss=0.1037, over 5714848.04 frames. ], batch size: 307, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:30:33,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=459608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:30:44,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5065, 2.1264, 1.5422, 0.7144], device='cuda:1'), covar=tensor([0.3598, 0.1837, 0.2837, 0.4307], device='cuda:1'), in_proj_covar=tensor([0.1513, 0.1425, 0.1450, 0.1245], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 14:30:48,695 INFO [optim.py:369] (1/2) 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 14:30:58,236 INFO [train.py:968] (1/2) Epoch 11, batch 4100, giga_loss[loss=0.2495, simple_loss=0.3249, pruned_loss=0.08704, over 28947.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3526, pruned_loss=0.1023, over 5702728.46 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3544, pruned_loss=0.09784, over 5120613.56 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3523, pruned_loss=0.1026, over 5710440.74 frames. ], batch size: 145, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:31:06,820 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:08,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=459652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:15,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2459, 1.7089, 1.3523, 0.4387], device='cuda:1'), covar=tensor([0.3079, 0.1858, 0.2847, 0.3849], device='cuda:1'), in_proj_covar=tensor([0.1509, 0.1422, 0.1448, 0.1242], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 14:31:21,033 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=459666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:32,779 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=459681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:31:38,054 INFO [train.py:968] (1/2) Epoch 11, batch 4150, giga_loss[loss=0.2377, simple_loss=0.3167, pruned_loss=0.07937, over 28886.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3483, pruned_loss=0.09986, over 5692926.04 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3541, pruned_loss=0.09769, over 5121897.83 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3484, pruned_loss=0.1003, over 5706178.69 frames. ], batch size: 199, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:31:40,757 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7954, 2.6188, 1.6706, 0.9210], device='cuda:1'), covar=tensor([0.5507, 0.2247, 0.3120, 0.5043], device='cuda:1'), in_proj_covar=tensor([0.1513, 0.1424, 0.1450, 0.1244], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 14:32:07,940 INFO [optim.py:369] (1/2) 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 14:32:17,273 INFO [train.py:968] (1/2) Epoch 11, batch 4200, giga_loss[loss=0.2648, simple_loss=0.3363, pruned_loss=0.09664, over 28589.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3484, pruned_loss=0.1006, over 5698256.13 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3539, pruned_loss=0.09752, over 5145071.47 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3485, pruned_loss=0.1011, over 5703290.73 frames. ], batch size: 85, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:32:26,742 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 14:32:28,618 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=459751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:32:30,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=459754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:32:54,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=459783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:32:59,692 INFO [train.py:968] (1/2) Epoch 11, batch 4250, giga_loss[loss=0.2208, simple_loss=0.302, pruned_loss=0.06982, over 28964.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3476, pruned_loss=0.1011, over 5697315.99 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3538, pruned_loss=0.09744, over 5152572.32 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3476, pruned_loss=0.1016, over 5699554.02 frames. ], batch size: 136, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:33:31,592 INFO [optim.py:369] (1/2) 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 14:33:42,125 INFO [train.py:968] (1/2) Epoch 11, batch 4300, giga_loss[loss=0.2524, simple_loss=0.328, pruned_loss=0.08839, over 28947.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3465, pruned_loss=0.1012, over 5703633.66 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3539, pruned_loss=0.09741, over 5172461.97 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3463, pruned_loss=0.1017, over 5702236.91 frames. ], batch size: 164, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:33:52,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3859, 1.4819, 1.5076, 1.3314], device='cuda:1'), covar=tensor([0.1398, 0.1746, 0.1844, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0722, 0.0663, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 14:34:24,511 INFO [train.py:968] (1/2) Epoch 11, batch 4350, giga_loss[loss=0.2888, simple_loss=0.3525, pruned_loss=0.1125, over 27979.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3438, pruned_loss=0.1003, over 5709394.30 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3534, pruned_loss=0.09719, over 5183941.62 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3438, pruned_loss=0.1009, over 5705184.85 frames. ], batch size: 412, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:34:49,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-05 14:34:50,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4957, 4.3157, 4.0932, 1.8887], device='cuda:1'), covar=tensor([0.0519, 0.0689, 0.0723, 0.2151], device='cuda:1'), in_proj_covar=tensor([0.1021, 0.0956, 0.0839, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 14:34:55,161 INFO [optim.py:369] (1/2) 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 14:35:02,859 INFO [train.py:968] (1/2) Epoch 11, batch 4400, giga_loss[loss=0.234, simple_loss=0.3082, pruned_loss=0.07985, over 28873.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3407, pruned_loss=0.09883, over 5708782.74 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3535, pruned_loss=0.09725, over 5196649.71 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3405, pruned_loss=0.09934, over 5703273.57 frames. ], batch size: 106, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:35:44,412 INFO [train.py:968] (1/2) Epoch 11, batch 4450, giga_loss[loss=0.2734, simple_loss=0.3498, pruned_loss=0.09848, over 28942.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3384, pruned_loss=0.09753, over 5711829.55 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3535, pruned_loss=0.0972, over 5206910.04 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3381, pruned_loss=0.098, over 5705190.57 frames. ], batch size: 213, lr: 3.00e-03, grad_scale: 8.0
+2023-03-05 14:36:17,239 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 4500, libri_loss[loss=0.3256, simple_loss=0.3858, pruned_loss=0.1327, over 29532.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3401, pruned_loss=0.09771, over 5716694.78 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3532, pruned_loss=0.09708, over 5229576.31 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3395, pruned_loss=0.09818, over 5706467.43 frames. ], batch size: 81, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:36:29,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=460041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:36:56,128 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-05 14:37:09,982 INFO [train.py:968] (1/2) Epoch 11, batch 4550, giga_loss[loss=0.2873, simple_loss=0.3693, pruned_loss=0.1027, over 28881.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3444, pruned_loss=0.09997, over 5706125.36 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3541, pruned_loss=0.09759, over 5240368.04 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3431, pruned_loss=0.09996, over 5698994.24 frames. ], batch size: 213, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:37:43,918 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 4600, giga_loss[loss=0.2608, simple_loss=0.3509, pruned_loss=0.08534, over 28818.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.347, pruned_loss=0.1006, over 5711189.29 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3539, pruned_loss=0.09752, over 5253549.52 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.346, pruned_loss=0.1007, over 5701867.98 frames. ], batch size: 186, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:38:30,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5853, 2.1873, 1.6645, 1.6664], device='cuda:1'), covar=tensor([0.0704, 0.0222, 0.0290, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 14:38:32,648 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=460184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:38:36,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=460187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:38:37,273 INFO [train.py:968] (1/2) Epoch 11, batch 4650, giga_loss[loss=0.2824, simple_loss=0.3582, pruned_loss=0.1033, over 28570.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3484, pruned_loss=0.1006, over 5705023.78 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3538, pruned_loss=0.0975, over 5276289.99 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3474, pruned_loss=0.1008, over 5691289.94 frames. ], batch size: 336, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:38:44,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-05 14:39:04,121 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 4700, giga_loss[loss=0.2861, simple_loss=0.3569, pruned_loss=0.1076, over 28852.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3473, pruned_loss=0.09966, over 5702382.73 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3535, pruned_loss=0.0974, over 5279539.54 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3467, pruned_loss=0.09991, over 5690714.45 frames. ], batch size: 112, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:40:03,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3170, 1.5408, 1.6040, 1.2156], device='cuda:1'), covar=tensor([0.1324, 0.1781, 0.1117, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0693, 0.0856, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 14:40:07,332 INFO [train.py:968] (1/2) Epoch 11, batch 4750, giga_loss[loss=0.2472, simple_loss=0.3279, pruned_loss=0.08323, over 28997.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3465, pruned_loss=0.09978, over 5706695.72 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3538, pruned_loss=0.09758, over 5285403.42 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3457, pruned_loss=0.09985, over 5696154.60 frames. ], batch size: 155, lr: 3.00e-03, grad_scale: 4.0
+2023-03-05 14:40:41,084 INFO [optim.py:369] (1/2) 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 14:40:48,593 INFO [train.py:968] (1/2) Epoch 11, batch 4800, giga_loss[loss=0.2723, simple_loss=0.3397, pruned_loss=0.1024, over 28642.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3478, pruned_loss=0.1009, over 5704002.59 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3533, pruned_loss=0.09731, over 5293551.52 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3475, pruned_loss=0.1013, over 5693938.12 frames. ], batch size: 85, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:41:24,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9904, 1.2441, 1.0757, 0.9351], device='cuda:1'), covar=tensor([0.1977, 0.1572, 0.1175, 0.1522], device='cuda:1'), in_proj_covar=tensor([0.1680, 0.1600, 0.1564, 0.1678], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 14:41:32,722 INFO [train.py:968] (1/2) Epoch 11, batch 4850, giga_loss[loss=0.2815, simple_loss=0.3599, pruned_loss=0.1015, over 28635.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3499, pruned_loss=0.1019, over 5702104.01 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3541, pruned_loss=0.09764, over 5303461.57 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3489, pruned_loss=0.102, over 5692489.94 frames. ], batch size: 242, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:42:04,454 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 4900, giga_loss[loss=0.2653, simple_loss=0.3501, pruned_loss=0.09026, over 28026.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3531, pruned_loss=0.1038, over 5697416.89 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.354, pruned_loss=0.09764, over 5308758.99 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3523, pruned_loss=0.1039, over 5693074.85 frames. ], batch size: 412, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:42:57,196 INFO [train.py:968] (1/2) Epoch 11, batch 4950, giga_loss[loss=0.2576, simple_loss=0.3342, pruned_loss=0.0905, over 28926.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3558, pruned_loss=0.1048, over 5710772.02 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3543, pruned_loss=0.09769, over 5320592.03 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.355, pruned_loss=0.105, over 5703841.33 frames. ], batch size: 106, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:43:04,508 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:43:28,747 INFO [optim.py:369] (1/2) 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 14:43:36,389 INFO [train.py:968] (1/2) Epoch 11, batch 5000, giga_loss[loss=0.27, simple_loss=0.3515, pruned_loss=0.09424, over 29044.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3572, pruned_loss=0.1054, over 5715461.17 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3548, pruned_loss=0.09794, over 5338079.12 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3561, pruned_loss=0.1056, over 5706234.95 frames. ], batch size: 128, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:44:13,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8102, 1.8977, 1.2919, 1.5343], device='cuda:1'), covar=tensor([0.0720, 0.0503, 0.0988, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0436, 0.0495, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 14:44:19,584 INFO [train.py:968] (1/2) Epoch 11, batch 5050, giga_loss[loss=0.3078, simple_loss=0.367, pruned_loss=0.1243, over 28753.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3565, pruned_loss=0.1045, over 5722027.83 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.355, pruned_loss=0.09797, over 5346625.04 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3555, pruned_loss=0.1047, over 5712176.08 frames. ], batch size: 99, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:44:25,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-05 14:44:34,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:44:49,766 INFO [optim.py:369] (1/2) 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 14:44:58,721 INFO [train.py:968] (1/2) Epoch 11, batch 5100, giga_loss[loss=0.2914, simple_loss=0.3596, pruned_loss=0.1117, over 28437.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3566, pruned_loss=0.1044, over 5727876.26 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3557, pruned_loss=0.09818, over 5363843.95 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3553, pruned_loss=0.1046, over 5716244.24 frames. ], batch size: 71, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:45:12,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4926, 1.7277, 1.6479, 1.5811], device='cuda:1'), covar=tensor([0.1538, 0.1808, 0.1984, 0.1793], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0735, 0.0677, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 14:45:40,212 INFO [train.py:968] (1/2) Epoch 11, batch 5150, giga_loss[loss=0.3317, simple_loss=0.3828, pruned_loss=0.1403, over 26748.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3545, pruned_loss=0.1033, over 5713890.18 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3561, pruned_loss=0.09851, over 5361582.51 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3531, pruned_loss=0.1033, over 5711838.51 frames. ], batch size: 555, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:45:51,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-05 14:46:12,887 INFO [optim.py:369] (1/2) 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 14:46:20,946 INFO [train.py:968] (1/2) Epoch 11, batch 5200, giga_loss[loss=0.2332, simple_loss=0.308, pruned_loss=0.07921, over 28684.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3521, pruned_loss=0.1024, over 5722343.74 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3568, pruned_loss=0.09911, over 5379209.02 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3502, pruned_loss=0.102, over 5717408.92 frames. ], batch size: 92, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:46:31,755 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460754.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:46:57,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:47:01,158 INFO [train.py:968] (1/2) Epoch 11, batch 5250, libri_loss[loss=0.2534, simple_loss=0.3331, pruned_loss=0.08684, over 29589.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3482, pruned_loss=0.09994, over 5724051.12 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3568, pruned_loss=0.09919, over 5388763.40 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3466, pruned_loss=0.09961, over 5720265.67 frames. ], batch size: 76, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:47:21,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:47:31,229 INFO [optim.py:369] (1/2) 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 14:47:39,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=460837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:47:40,394 INFO [train.py:968] (1/2) Epoch 11, batch 5300, giga_loss[loss=0.2702, simple_loss=0.3564, pruned_loss=0.09203, over 28254.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3482, pruned_loss=0.09962, over 5719241.09 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3567, pruned_loss=0.09912, over 5404323.65 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3467, pruned_loss=0.09943, over 5711280.30 frames. ], batch size: 368, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:48:02,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5449, 1.7798, 1.5004, 1.2728], device='cuda:1'), covar=tensor([0.2410, 0.1796, 0.1527, 0.1928], device='cuda:1'), in_proj_covar=tensor([0.1682, 0.1603, 0.1569, 0.1675], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 14:48:11,282 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 5350, giga_loss[loss=0.3022, simple_loss=0.364, pruned_loss=0.1202, over 28705.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3494, pruned_loss=0.09893, over 5714935.24 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3562, pruned_loss=0.09883, over 5411846.63 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3485, pruned_loss=0.09903, over 5706051.22 frames. ], batch size: 99, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:48:58,608 INFO [optim.py:369] (1/2) 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 14:49:09,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-05 14:49:09,273 INFO [train.py:968] (1/2) Epoch 11, batch 5400, giga_loss[loss=0.2584, simple_loss=0.3378, pruned_loss=0.08949, over 29049.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3502, pruned_loss=0.1001, over 5709231.22 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3563, pruned_loss=0.09895, over 5419401.60 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3494, pruned_loss=0.1001, over 5699619.37 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:49:22,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5579, 1.8458, 1.8105, 1.3333], device='cuda:1'), covar=tensor([0.1577, 0.2041, 0.1286, 0.1463], device='cuda:1'), in_proj_covar=tensor([0.0821, 0.0696, 0.0859, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 14:49:42,278 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-05 14:49:47,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=460985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:49:50,330 INFO [train.py:968] (1/2) Epoch 11, batch 5450, giga_loss[loss=0.2584, simple_loss=0.3396, pruned_loss=0.0886, over 28520.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3508, pruned_loss=0.1017, over 5712825.79 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3567, pruned_loss=0.09926, over 5428594.12 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3497, pruned_loss=0.1014, over 5702364.95 frames. ], batch size: 336, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:49:52,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4384, 1.2593, 1.2644, 1.6122], device='cuda:1'), covar=tensor([0.0669, 0.0310, 0.0295, 0.0718], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0112, 0.0115, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 14:50:14,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:50:15,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:50:24,348 INFO [optim.py:369] (1/2) 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 14:50:33,004 INFO [train.py:968] (1/2) Epoch 11, batch 5500, giga_loss[loss=0.2927, simple_loss=0.3556, pruned_loss=0.1149, over 28984.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3504, pruned_loss=0.1034, over 5701915.04 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3572, pruned_loss=0.09967, over 5428641.44 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.349, pruned_loss=0.1029, over 5697723.50 frames. ], batch size: 213, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:50:39,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:51:13,381 INFO [train.py:968] (1/2) Epoch 11, batch 5550, giga_loss[loss=0.2679, simple_loss=0.3398, pruned_loss=0.09795, over 28371.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3476, pruned_loss=0.1025, over 5701476.57 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.357, pruned_loss=0.09971, over 5437129.03 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3465, pruned_loss=0.1022, over 5696695.14 frames. ], batch size: 368, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:51:19,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2279, 1.8040, 1.3574, 0.3507], device='cuda:1'), covar=tensor([0.3226, 0.1845, 0.3343, 0.4622], device='cuda:1'), in_proj_covar=tensor([0.1545, 0.1451, 0.1474, 0.1263], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 14:51:44,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:51:45,159 INFO [optim.py:369] (1/2) 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 14:51:46,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461129.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:51:47,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:51:48,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9028, 3.7159, 3.5001, 1.6671], device='cuda:1'), covar=tensor([0.0653, 0.0814, 0.0824, 0.2282], device='cuda:1'), in_proj_covar=tensor([0.1030, 0.0959, 0.0843, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 14:51:53,445 INFO [train.py:968] (1/2) Epoch 11, batch 5600, libri_loss[loss=0.2846, simple_loss=0.3671, pruned_loss=0.101, over 29640.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3472, pruned_loss=0.1031, over 5707532.53 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3578, pruned_loss=0.1004, over 5447432.47 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3453, pruned_loss=0.1023, over 5700148.58 frames. ], batch size: 88, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 14:52:11,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:11,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:21,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:35,712 INFO [train.py:968] (1/2) Epoch 11, batch 5650, giga_loss[loss=0.2536, simple_loss=0.3196, pruned_loss=0.09384, over 28364.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3461, pruned_loss=0.1026, over 5713257.83 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3582, pruned_loss=0.1008, over 5451901.47 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3442, pruned_loss=0.1018, over 5706891.48 frames. ], batch size: 65, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:52:39,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:52:40,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 14:52:54,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:53:09,713 INFO [optim.py:369] (1/2) 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 14:53:16,838 INFO [train.py:968] (1/2) Epoch 11, batch 5700, giga_loss[loss=0.2422, simple_loss=0.3089, pruned_loss=0.0878, over 28924.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3407, pruned_loss=0.09935, over 5722396.00 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3576, pruned_loss=0.1004, over 5461303.99 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3394, pruned_loss=0.09899, over 5714866.16 frames. ], batch size: 145, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:53:36,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3422, 1.2363, 1.1348, 1.4881], device='cuda:1'), covar=tensor([0.0709, 0.0336, 0.0346, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0079, 0.0056, 0.0051, 0.0086], device='cuda:1')
+2023-03-05 14:53:45,077 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461272.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:53:48,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461275.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 14:53:56,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2677, 1.4349, 1.3057, 1.1443], device='cuda:1'), covar=tensor([0.2148, 0.1895, 0.1371, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1687, 0.1601, 0.1571, 0.1670], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 14:53:59,093 INFO [train.py:968] (1/2) Epoch 11, batch 5750, giga_loss[loss=0.2217, simple_loss=0.3021, pruned_loss=0.07066, over 29052.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3372, pruned_loss=0.09748, over 5721212.86 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3575, pruned_loss=0.1004, over 5466654.28 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3359, pruned_loss=0.09718, over 5715118.19 frames. ], batch size: 136, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:54:09,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6135, 0.9978, 2.9304, 2.7636], device='cuda:1'), covar=tensor([0.1749, 0.2459, 0.0605, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0648, 0.0579, 0.0844, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 14:54:10,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:10,974 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 14:54:11,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-05 14:54:12,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:32,203 INFO [optim.py:369] (1/2) 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 14:54:36,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:36,344 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:38,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:54:38,676 INFO [train.py:968] (1/2) Epoch 11, batch 5800, giga_loss[loss=0.2099, simple_loss=0.2877, pruned_loss=0.06601, over 28960.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3373, pruned_loss=0.09742, over 5720362.35 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3579, pruned_loss=0.1006, over 5474024.00 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3354, pruned_loss=0.09688, over 5714467.35 frames. ], batch size: 106, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:54:50,992 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:55:01,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:55:19,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=461387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:55:20,192 INFO [train.py:968] (1/2) Epoch 11, batch 5850, giga_loss[loss=0.2766, simple_loss=0.3527, pruned_loss=0.1002, over 29061.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.34, pruned_loss=0.09833, over 5717457.13 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3579, pruned_loss=0.1008, over 5472010.12 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3382, pruned_loss=0.09774, over 5718490.98 frames. ], batch size: 155, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:55:24,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3501, 2.2489, 1.7326, 2.0758], device='cuda:1'), covar=tensor([0.0647, 0.0569, 0.0837, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0441, 0.0494, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 14:55:24,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5556, 2.3696, 2.3545, 2.2152], device='cuda:1'), covar=tensor([0.1219, 0.1863, 0.1484, 0.1472], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0732, 0.0672, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 14:55:54,594 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 5900, giga_loss[loss=0.3013, simple_loss=0.3751, pruned_loss=0.1137, over 28658.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3438, pruned_loss=0.09981, over 5718316.39 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3586, pruned_loss=0.1012, over 5480533.73 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3414, pruned_loss=0.09893, over 5717086.10 frames. ], batch size: 242, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:56:21,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 14:56:22,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-05 14:56:40,471 INFO [train.py:968] (1/2) Epoch 11, batch 5950, giga_loss[loss=0.2577, simple_loss=0.3285, pruned_loss=0.09348, over 28513.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3478, pruned_loss=0.1017, over 5711472.67 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3582, pruned_loss=0.101, over 5486664.67 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3458, pruned_loss=0.1012, over 5713093.72 frames. ], batch size: 60, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:57:19,498 INFO [optim.py:369] (1/2) 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 14:57:26,240 INFO [train.py:968] (1/2) Epoch 11, batch 6000, giga_loss[loss=0.2934, simple_loss=0.3618, pruned_loss=0.1125, over 28709.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3509, pruned_loss=0.103, over 5709261.51 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3584, pruned_loss=0.1011, over 5492401.73 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3489, pruned_loss=0.1025, over 5709321.29 frames. ], batch size: 99, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:57:26,240 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 14:57:36,259 INFO [train.py:1012] (1/2) Epoch 11, validation: loss=0.2223, simple_loss=0.3277, pruned_loss=0.05847, over 944034.00 frames. 
+2023-03-05 14:57:36,259 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 14:57:41,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:58:19,647 INFO [train.py:968] (1/2) Epoch 11, batch 6050, giga_loss[loss=0.4183, simple_loss=0.4422, pruned_loss=0.1972, over 26495.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3532, pruned_loss=0.1045, over 5704287.42 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3589, pruned_loss=0.1013, over 5498065.92 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.351, pruned_loss=0.1039, over 5704137.60 frames. ], batch size: 555, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:58:49,357 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:58:56,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:59:00,093 INFO [optim.py:369] (1/2) 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 14:59:07,942 INFO [train.py:968] (1/2) Epoch 11, batch 6100, giga_loss[loss=0.3507, simple_loss=0.4079, pruned_loss=0.1467, over 27980.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3606, pruned_loss=0.111, over 5705870.96 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3587, pruned_loss=0.1011, over 5502251.79 frames. ], giga_tot_loss[loss=0.2903, simple_loss=0.359, pruned_loss=0.1108, over 5704039.52 frames. ], batch size: 412, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 14:59:36,523 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 6150, giga_loss[loss=0.3312, simple_loss=0.3908, pruned_loss=0.1358, over 28662.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3667, pruned_loss=0.1163, over 5693139.06 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3592, pruned_loss=0.1016, over 5513456.38 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3651, pruned_loss=0.116, over 5686881.31 frames. ], batch size: 242, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 14:59:56,094 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=461689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 14:59:57,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=461692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:00:25,948 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:00:40,991 INFO [train.py:968] (1/2) Epoch 11, batch 6200, giga_loss[loss=0.3003, simple_loss=0.3736, pruned_loss=0.1135, over 28958.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3733, pruned_loss=0.1214, over 5673760.73 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3591, pruned_loss=0.1019, over 5506627.60 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3725, pruned_loss=0.1214, over 5681058.25 frames. ], batch size: 136, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:01:31,185 INFO [train.py:968] (1/2) Epoch 11, batch 6250, giga_loss[loss=0.3663, simple_loss=0.4141, pruned_loss=0.1593, over 28850.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3791, pruned_loss=0.127, over 5664848.33 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.359, pruned_loss=0.1018, over 5510439.31 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3787, pruned_loss=0.1273, over 5668538.00 frames. ], batch size: 199, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:02:10,387 INFO [optim.py:369] (1/2) 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 15:02:14,822 INFO [train.py:968] (1/2) Epoch 11, batch 6300, giga_loss[loss=0.3503, simple_loss=0.4115, pruned_loss=0.1446, over 28892.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3836, pruned_loss=0.1303, over 5677276.04 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3591, pruned_loss=0.1019, over 5521145.13 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3838, pruned_loss=0.1312, over 5674801.47 frames. ], batch size: 186, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:02:51,162 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3895, 1.5570, 0.9803, 1.2936], device='cuda:1'), covar=tensor([0.0931, 0.0797, 0.1455, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0440, 0.0494, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 15:03:08,762 INFO [train.py:968] (1/2) Epoch 11, batch 6350, giga_loss[loss=0.3011, simple_loss=0.3728, pruned_loss=0.1147, over 28994.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3884, pruned_loss=0.1348, over 5652250.27 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.359, pruned_loss=0.1018, over 5521254.23 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3889, pruned_loss=0.1359, over 5651840.75 frames. ], batch size: 136, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:03:39,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5953, 1.6225, 1.5991, 1.4994], device='cuda:1'), covar=tensor([0.1314, 0.1679, 0.1743, 0.1554], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0733, 0.0671, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 15:03:53,292 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=461932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:03:53,646 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 6400, giga_loss[loss=0.2924, simple_loss=0.3649, pruned_loss=0.1099, over 29008.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3907, pruned_loss=0.138, over 5634571.32 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3593, pruned_loss=0.1021, over 5511240.61 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3914, pruned_loss=0.1393, over 5644992.72 frames. ], batch size: 128, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:04:44,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3311, 1.6576, 1.3086, 1.2804], device='cuda:1'), covar=tensor([0.2034, 0.1975, 0.2123, 0.1778], device='cuda:1'), in_proj_covar=tensor([0.1274, 0.0953, 0.1132, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 15:04:49,355 INFO [train.py:968] (1/2) Epoch 11, batch 6450, giga_loss[loss=0.3214, simple_loss=0.3882, pruned_loss=0.1273, over 29008.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3928, pruned_loss=0.141, over 5622915.18 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3589, pruned_loss=0.1019, over 5518966.81 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3947, pruned_loss=0.1434, over 5626836.67 frames. ], batch size: 155, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:04:56,842 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=461994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:05:06,791 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 6500, giga_loss[loss=0.3637, simple_loss=0.4157, pruned_loss=0.1558, over 28745.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3978, pruned_loss=0.146, over 5610100.68 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3592, pruned_loss=0.1022, over 5523812.36 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4007, pruned_loss=0.1494, over 5612090.59 frames. ], batch size: 99, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:05:47,311 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 6550, giga_loss[loss=0.409, simple_loss=0.4241, pruned_loss=0.1969, over 23501.00 frames. ], tot_loss[loss=0.3463, simple_loss=0.399, pruned_loss=0.1468, over 5608994.41 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3589, pruned_loss=0.1018, over 5528908.97 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4025, pruned_loss=0.1509, over 5607483.94 frames. ], batch size: 705, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:06:52,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0517, 1.1706, 3.9067, 3.0834], device='cuda:1'), covar=tensor([0.1744, 0.2512, 0.0408, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0651, 0.0583, 0.0849, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 15:07:15,961 INFO [optim.py:369] (1/2) 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 15:07:20,051 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:21,035 INFO [train.py:968] (1/2) Epoch 11, batch 6600, giga_loss[loss=0.341, simple_loss=0.3965, pruned_loss=0.1428, over 28948.00 frames. ], tot_loss[loss=0.3449, simple_loss=0.3978, pruned_loss=0.146, over 5629114.26 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3588, pruned_loss=0.1018, over 5536274.84 frames. ], giga_tot_loss[loss=0.3514, simple_loss=0.4017, pruned_loss=0.1505, over 5623798.83 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:07:22,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:26,278 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:29,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:33,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5962, 2.3975, 2.4205, 2.0658], device='cuda:1'), covar=tensor([0.1352, 0.1988, 0.1617, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0731, 0.0668, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 15:07:50,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:07:57,294 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:09,394 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:10,350 INFO [train.py:968] (1/2) Epoch 11, batch 6650, giga_loss[loss=0.3567, simple_loss=0.4091, pruned_loss=0.1522, over 29014.00 frames. ], tot_loss[loss=0.3428, simple_loss=0.3957, pruned_loss=0.145, over 5631280.60 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3583, pruned_loss=0.1016, over 5544466.24 frames. ], giga_tot_loss[loss=0.3498, simple_loss=0.4001, pruned_loss=0.1497, over 5622110.03 frames. ], batch size: 128, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:08:11,947 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:41,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:08:42,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 15:08:55,121 INFO [optim.py:369] (1/2) 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 15:09:00,884 INFO [train.py:968] (1/2) Epoch 11, batch 6700, giga_loss[loss=0.3775, simple_loss=0.4243, pruned_loss=0.1654, over 27932.00 frames. ], tot_loss[loss=0.3427, simple_loss=0.3961, pruned_loss=0.1446, over 5628582.16 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.358, pruned_loss=0.1014, over 5542377.36 frames. ], giga_tot_loss[loss=0.3495, simple_loss=0.4005, pruned_loss=0.1493, over 5625691.94 frames. ], batch size: 412, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:09:02,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4727, 1.5724, 1.4960, 1.4063], device='cuda:1'), covar=tensor([0.1338, 0.1740, 0.2009, 0.1679], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0733, 0.0670, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 15:09:49,001 INFO [train.py:968] (1/2) Epoch 11, batch 6750, giga_loss[loss=0.3814, simple_loss=0.4256, pruned_loss=0.1686, over 28558.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.396, pruned_loss=0.1434, over 5636691.33 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3578, pruned_loss=0.1012, over 5548594.27 frames. ], giga_tot_loss[loss=0.3488, simple_loss=0.4007, pruned_loss=0.1485, over 5631408.19 frames. ], batch size: 336, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:10:07,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=462307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:10:33,665 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 6800, giga_loss[loss=0.3477, simple_loss=0.397, pruned_loss=0.1492, over 27454.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.397, pruned_loss=0.1439, over 5622956.57 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.358, pruned_loss=0.1013, over 5553958.32 frames. ], giga_tot_loss[loss=0.3491, simple_loss=0.4012, pruned_loss=0.1485, over 5614580.22 frames. ], batch size: 472, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:11:32,907 INFO [train.py:968] (1/2) Epoch 11, batch 6850, libri_loss[loss=0.2765, simple_loss=0.3403, pruned_loss=0.1063, over 29472.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3939, pruned_loss=0.1413, over 5619587.72 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3577, pruned_loss=0.1011, over 5562771.69 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3984, pruned_loss=0.1461, over 5606264.62 frames. ], batch size: 70, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:12:13,045 INFO [optim.py:369] (1/2) 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 15:12:16,256 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=462436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:12:18,464 INFO [train.py:968] (1/2) Epoch 11, batch 6900, giga_loss[loss=0.2571, simple_loss=0.3443, pruned_loss=0.08495, over 28990.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3908, pruned_loss=0.1373, over 5623308.58 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3576, pruned_loss=0.101, over 5564872.70 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.3957, pruned_loss=0.1426, over 5612970.52 frames. ], batch size: 136, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:12:28,660 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:12:32,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:12:48,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5970, 3.8584, 1.7371, 1.5869], device='cuda:1'), covar=tensor([0.0902, 0.0358, 0.0826, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0505, 0.0335, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 15:12:59,976 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:13:09,621 INFO [train.py:968] (1/2) Epoch 11, batch 6950, giga_loss[loss=0.3472, simple_loss=0.4013, pruned_loss=0.1466, over 28042.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3869, pruned_loss=0.1333, over 5641616.46 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3571, pruned_loss=0.1009, over 5573247.64 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3919, pruned_loss=0.1384, over 5627506.35 frames. ], batch size: 412, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:13:52,435 INFO [optim.py:369] (1/2) 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 15:13:58,122 INFO [train.py:968] (1/2) Epoch 11, batch 7000, giga_loss[loss=0.2879, simple_loss=0.3614, pruned_loss=0.1073, over 28896.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3848, pruned_loss=0.1316, over 5636260.65 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3571, pruned_loss=0.1009, over 5568662.85 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3891, pruned_loss=0.136, over 5630678.15 frames. ], batch size: 227, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:14:08,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9333, 1.1353, 1.1425, 0.8685], device='cuda:1'), covar=tensor([0.1687, 0.1732, 0.0939, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.1677, 0.1605, 0.1564, 0.1672], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 15:14:45,998 INFO [train.py:968] (1/2) Epoch 11, batch 7050, giga_loss[loss=0.273, simple_loss=0.3459, pruned_loss=0.1001, over 28960.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.384, pruned_loss=0.1316, over 5646995.18 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.357, pruned_loss=0.1008, over 5576426.45 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3881, pruned_loss=0.1359, over 5637431.87 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:15:24,514 INFO [optim.py:369] (1/2) 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 15:15:27,205 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 7100, giga_loss[loss=0.3288, simple_loss=0.3906, pruned_loss=0.1335, over 28622.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.384, pruned_loss=0.1312, over 5661535.76 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.357, pruned_loss=0.1007, over 5586583.68 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3882, pruned_loss=0.1357, over 5647171.99 frames. ], batch size: 307, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:16:22,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 15:16:31,182 INFO [train.py:968] (1/2) Epoch 11, batch 7150, giga_loss[loss=0.3929, simple_loss=0.4141, pruned_loss=0.1858, over 26562.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.383, pruned_loss=0.1301, over 5661650.73 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3575, pruned_loss=0.1011, over 5589654.77 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3863, pruned_loss=0.1337, over 5648493.30 frames. ], batch size: 555, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:16:37,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3973, 1.7370, 1.4505, 1.2325], device='cuda:1'), covar=tensor([0.2279, 0.1610, 0.1348, 0.1801], device='cuda:1'), in_proj_covar=tensor([0.1681, 0.1602, 0.1562, 0.1673], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 15:16:38,814 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=462695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:16:43,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8909, 1.8904, 1.4417, 1.6238], device='cuda:1'), covar=tensor([0.0814, 0.0725, 0.0967, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0443, 0.0494, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 15:16:54,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-05 15:17:19,350 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 7200, giga_loss[loss=0.3385, simple_loss=0.4118, pruned_loss=0.1326, over 28873.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3813, pruned_loss=0.1271, over 5671561.38 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3577, pruned_loss=0.1012, over 5590610.10 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.384, pruned_loss=0.1302, over 5661220.76 frames. ], batch size: 174, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:18:21,762 INFO [train.py:968] (1/2) Epoch 11, batch 7250, giga_loss[loss=0.3641, simple_loss=0.4191, pruned_loss=0.1546, over 28356.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3828, pruned_loss=0.1268, over 5662940.73 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3578, pruned_loss=0.1012, over 5594308.50 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3856, pruned_loss=0.1299, over 5653365.64 frames. ], batch size: 368, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:18:43,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=462811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:19:05,021 INFO [optim.py:369] (1/2) 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 15:19:10,707 INFO [train.py:968] (1/2) Epoch 11, batch 7300, giga_loss[loss=0.3612, simple_loss=0.4196, pruned_loss=0.1514, over 28907.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3838, pruned_loss=0.1273, over 5662385.22 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3577, pruned_loss=0.1013, over 5599680.58 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3865, pruned_loss=0.1302, over 5651082.92 frames. ], batch size: 164, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:20:01,512 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-05 15:20:02,675 INFO [train.py:968] (1/2) Epoch 11, batch 7350, giga_loss[loss=0.3724, simple_loss=0.4181, pruned_loss=0.1634, over 29056.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3824, pruned_loss=0.1265, over 5674018.93 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3576, pruned_loss=0.1012, over 5603227.40 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3851, pruned_loss=0.1294, over 5663108.43 frames. ], batch size: 155, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:20:45,041 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 7400, giga_loss[loss=0.3247, simple_loss=0.3649, pruned_loss=0.1422, over 23456.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3801, pruned_loss=0.1254, over 5675981.93 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3573, pruned_loss=0.101, over 5612623.26 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3833, pruned_loss=0.1288, over 5660740.15 frames. ], batch size: 705, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:21:08,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=462954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:21:11,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=462957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:21:17,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5834, 4.3901, 4.1884, 1.9897], device='cuda:1'), covar=tensor([0.0547, 0.0749, 0.0725, 0.2063], device='cuda:1'), in_proj_covar=tensor([0.1052, 0.0990, 0.0866, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 15:21:30,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3331, 1.8202, 1.3194, 1.5748], device='cuda:1'), covar=tensor([0.0703, 0.0319, 0.0307, 0.0739], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 15:21:37,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=462986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:21:39,151 INFO [train.py:968] (1/2) Epoch 11, batch 7450, giga_loss[loss=0.2862, simple_loss=0.3542, pruned_loss=0.1091, over 28887.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3791, pruned_loss=0.1261, over 5670030.43 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3575, pruned_loss=0.101, over 5616430.97 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3822, pruned_loss=0.1295, over 5655814.41 frames. ], batch size: 199, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:21:58,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:22:22,186 INFO [optim.py:369] (1/2) 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 15:22:26,936 INFO [train.py:968] (1/2) Epoch 11, batch 7500, giga_loss[loss=0.3023, simple_loss=0.3664, pruned_loss=0.1191, over 28238.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3777, pruned_loss=0.1254, over 5682687.50 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3576, pruned_loss=0.101, over 5619293.70 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3803, pruned_loss=0.1283, over 5669523.85 frames. ], batch size: 368, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:23:01,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9181, 3.6851, 3.5034, 1.6122], device='cuda:1'), covar=tensor([0.0704, 0.0889, 0.0926, 0.2319], device='cuda:1'), in_proj_covar=tensor([0.1045, 0.0980, 0.0859, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 15:23:02,271 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 7550, giga_loss[loss=0.3072, simple_loss=0.3736, pruned_loss=0.1204, over 28736.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3774, pruned_loss=0.1236, over 5695633.91 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3575, pruned_loss=0.101, over 5625215.40 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3799, pruned_loss=0.1265, over 5681368.58 frames. ], batch size: 92, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:23:23,018 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:00,756 INFO [optim.py:369] (1/2) 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 15:24:04,577 INFO [train.py:968] (1/2) Epoch 11, batch 7600, giga_loss[loss=0.3038, simple_loss=0.3659, pruned_loss=0.1209, over 27939.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3777, pruned_loss=0.1231, over 5701087.47 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3576, pruned_loss=0.1011, over 5629796.17 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3802, pruned_loss=0.1259, over 5687143.35 frames. ], batch size: 412, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:24:19,568 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:19,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:21,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:38,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1012, 3.9109, 3.7149, 1.8104], device='cuda:1'), covar=tensor([0.0559, 0.0712, 0.0715, 0.2330], device='cuda:1'), in_proj_covar=tensor([0.1041, 0.0974, 0.0853, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 15:24:47,649 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:24:52,884 INFO [train.py:968] (1/2) Epoch 11, batch 7650, giga_loss[loss=0.2756, simple_loss=0.3537, pruned_loss=0.09875, over 28224.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3779, pruned_loss=0.1236, over 5695242.39 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3579, pruned_loss=0.1012, over 5631932.44 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3799, pruned_loss=0.126, over 5683140.45 frames. ], batch size: 77, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:25:14,870 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:25:18,685 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:25:43,249 INFO [train.py:968] (1/2) Epoch 11, batch 7700, giga_loss[loss=0.2942, simple_loss=0.3642, pruned_loss=0.1121, over 28943.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3761, pruned_loss=0.1232, over 5696379.14 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3574, pruned_loss=0.1008, over 5634149.47 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3783, pruned_loss=0.1256, over 5685789.44 frames. ], batch size: 136, lr: 2.99e-03, grad_scale: 8.0
+2023-03-05 15:25:47,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2628, 5.0748, 4.8089, 2.1999], device='cuda:1'), covar=tensor([0.0380, 0.0526, 0.0557, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.1042, 0.0974, 0.0854, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 15:25:49,952 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:26:34,937 INFO [train.py:968] (1/2) Epoch 11, batch 7750, giga_loss[loss=0.3032, simple_loss=0.363, pruned_loss=0.1217, over 28983.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3756, pruned_loss=0.1238, over 5687223.82 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3573, pruned_loss=0.1008, over 5635512.72 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3775, pruned_loss=0.1259, over 5678082.15 frames. ], batch size: 106, lr: 2.99e-03, grad_scale: 4.0
+2023-03-05 15:27:07,322 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 15:27:18,098 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 7800, giga_loss[loss=0.3221, simple_loss=0.3783, pruned_loss=0.133, over 29100.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3748, pruned_loss=0.1235, over 5694894.91 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3575, pruned_loss=0.1008, over 5640929.03 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3767, pruned_loss=0.126, over 5684040.21 frames. ], batch size: 100, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:27:35,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-05 15:27:35,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1500, 1.6720, 5.3606, 4.0006], device='cuda:1'), covar=tensor([0.1612, 0.2491, 0.0539, 0.0693], device='cuda:1'), in_proj_covar=tensor([0.0654, 0.0588, 0.0855, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 15:27:39,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2874, 3.0951, 2.9419, 1.3927], device='cuda:1'), covar=tensor([0.1015, 0.1132, 0.1165, 0.2335], device='cuda:1'), in_proj_covar=tensor([0.1052, 0.0985, 0.0863, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 15:27:47,472 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 7850, giga_loss[loss=0.2587, simple_loss=0.3331, pruned_loss=0.09208, over 28952.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3727, pruned_loss=0.1228, over 5699315.84 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3568, pruned_loss=0.1005, over 5645795.83 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3752, pruned_loss=0.1255, over 5687320.41 frames. ], batch size: 145, lr: 2.99e-03, grad_scale: 2.0
+2023-03-05 15:28:53,183 INFO [optim.py:369] (1/2) 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 15:28:55,797 INFO [train.py:968] (1/2) Epoch 11, batch 7900, giga_loss[loss=0.2975, simple_loss=0.3669, pruned_loss=0.1141, over 28909.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3707, pruned_loss=0.1216, over 5707594.53 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3567, pruned_loss=0.1004, over 5657148.74 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.1249, over 5689542.54 frames. ], batch size: 174, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:29:09,920 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:29:24,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:29:41,916 INFO [train.py:968] (1/2) Epoch 11, batch 7950, giga_loss[loss=0.2916, simple_loss=0.3535, pruned_loss=0.1149, over 28804.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3701, pruned_loss=0.1214, over 5701556.07 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3568, pruned_loss=0.1004, over 5654577.46 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3726, pruned_loss=0.1245, over 5691537.67 frames. ], batch size: 92, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:29:55,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:30:19,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:30:28,720 INFO [optim.py:369] (1/2) 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 15:30:32,117 INFO [train.py:968] (1/2) Epoch 11, batch 8000, giga_loss[loss=0.3191, simple_loss=0.3815, pruned_loss=0.1284, over 28264.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3717, pruned_loss=0.1226, over 5690638.09 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3567, pruned_loss=0.1004, over 5658115.90 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.374, pruned_loss=0.1254, over 5680231.97 frames. ], batch size: 368, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:31:20,383 INFO [train.py:968] (1/2) Epoch 11, batch 8050, giga_loss[loss=0.3117, simple_loss=0.383, pruned_loss=0.1202, over 28553.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3716, pruned_loss=0.1218, over 5683077.99 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3566, pruned_loss=0.1003, over 5659391.00 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3737, pruned_loss=0.1244, over 5674072.71 frames. ], batch size: 336, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:31:35,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4108, 3.6776, 1.5671, 1.6185], device='cuda:1'), covar=tensor([0.0921, 0.0348, 0.0868, 0.1258], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0508, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 15:31:41,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:31:43,136 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 8100, libri_loss[loss=0.2684, simple_loss=0.3382, pruned_loss=0.09928, over 29588.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3715, pruned_loss=0.1214, over 5677639.47 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3557, pruned_loss=0.0999, over 5663910.02 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3744, pruned_loss=0.1244, over 5666587.01 frames. ], batch size: 74, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:32:11,227 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:12,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:34,032 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:32:37,504 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 8150, giga_loss[loss=0.2772, simple_loss=0.3479, pruned_loss=0.1032, over 28903.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3725, pruned_loss=0.1224, over 5685659.26 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3553, pruned_loss=0.09973, over 5669052.11 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3755, pruned_loss=0.1254, over 5672631.91 frames. ], batch size: 227, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:33:05,808 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:33:14,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2682, 1.4570, 1.2255, 1.5065], device='cuda:1'), covar=tensor([0.0736, 0.0324, 0.0314, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 15:33:39,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 15:33:42,835 INFO [optim.py:369] (1/2) 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 15:33:43,823 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 8200, giga_loss[loss=0.3335, simple_loss=0.387, pruned_loss=0.14, over 29068.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3762, pruned_loss=0.1255, over 5685839.67 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3553, pruned_loss=0.09971, over 5672867.42 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3789, pruned_loss=0.1283, over 5672351.71 frames. ], batch size: 136, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:34:00,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 15:34:39,298 INFO [train.py:968] (1/2) Epoch 11, batch 8250, giga_loss[loss=0.4103, simple_loss=0.431, pruned_loss=0.1948, over 26558.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3779, pruned_loss=0.1279, over 5689248.29 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3553, pruned_loss=0.09957, over 5675783.05 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3803, pruned_loss=0.1306, over 5676202.15 frames. ], batch size: 555, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:35:22,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:35:28,115 INFO [optim.py:369] (1/2) 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 15:35:29,588 INFO [train.py:968] (1/2) Epoch 11, batch 8300, giga_loss[loss=0.2933, simple_loss=0.3572, pruned_loss=0.1147, over 28863.00 frames. ], tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1295, over 5678950.50 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3555, pruned_loss=0.09955, over 5679889.10 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3813, pruned_loss=0.1324, over 5664872.81 frames. ], batch size: 186, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:35:48,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:09,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=463877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:14,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:16,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:20,569 INFO [train.py:968] (1/2) Epoch 11, batch 8350, giga_loss[loss=0.3071, simple_loss=0.3727, pruned_loss=0.1207, over 28989.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3803, pruned_loss=0.1313, over 5672939.72 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3556, pruned_loss=0.09959, over 5681815.44 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3823, pruned_loss=0.1338, over 5660068.45 frames. ], batch size: 213, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:36:43,059 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=463913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:36:54,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5781, 1.8219, 1.8525, 1.3818], device='cuda:1'), covar=tensor([0.1654, 0.2202, 0.1332, 0.1493], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0706, 0.0861, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 15:37:03,016 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 8400, giga_loss[loss=0.2914, simple_loss=0.3565, pruned_loss=0.1131, over 28953.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3784, pruned_loss=0.1298, over 5675290.03 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3554, pruned_loss=0.09944, over 5689724.20 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3809, pruned_loss=0.133, over 5657773.75 frames. ], batch size: 106, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:37:16,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=463951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:37:36,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=463972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:37:39,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=463975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:37:50,963 INFO [train.py:968] (1/2) Epoch 11, batch 8450, giga_loss[loss=0.399, simple_loss=0.4246, pruned_loss=0.1867, over 26610.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3785, pruned_loss=0.1289, over 5682171.88 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3556, pruned_loss=0.09961, over 5692912.84 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3806, pruned_loss=0.1317, over 5665133.94 frames. ], batch size: 555, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:38:06,394 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:19,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:20,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:23,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:38:36,474 INFO [optim.py:369] (1/2) 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 15:38:37,989 INFO [train.py:968] (1/2) Epoch 11, batch 8500, giga_loss[loss=0.3075, simple_loss=0.3754, pruned_loss=0.1198, over 28906.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3765, pruned_loss=0.1272, over 5657895.87 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.356, pruned_loss=0.09998, over 5678239.07 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3782, pruned_loss=0.1296, over 5657265.78 frames. ], batch size: 227, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:38:49,737 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:39:23,138 INFO [train.py:968] (1/2) Epoch 11, batch 8550, giga_loss[loss=0.3013, simple_loss=0.3676, pruned_loss=0.1175, over 29023.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3738, pruned_loss=0.125, over 5668414.61 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3563, pruned_loss=0.1, over 5680922.01 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3754, pruned_loss=0.1276, over 5665136.59 frames. ], batch size: 106, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:39:36,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:40:00,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7895, 1.6668, 1.7078, 1.4813], device='cuda:1'), covar=tensor([0.1466, 0.2314, 0.2059, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0733, 0.0669, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 15:40:07,607 INFO [optim.py:369] (1/2) 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 15:40:08,256 INFO [train.py:968] (1/2) Epoch 11, batch 8600, giga_loss[loss=0.2851, simple_loss=0.3544, pruned_loss=0.1079, over 28576.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1245, over 5676684.61 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3566, pruned_loss=0.1002, over 5683562.37 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3735, pruned_loss=0.1268, over 5671474.24 frames. ], batch size: 307, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:40:22,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-05 15:40:32,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:40:34,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:40:56,868 INFO [train.py:968] (1/2) Epoch 11, batch 8650, giga_loss[loss=0.3452, simple_loss=0.3834, pruned_loss=0.1535, over 23643.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3723, pruned_loss=0.125, over 5660635.47 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3565, pruned_loss=0.09989, over 5687358.64 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.374, pruned_loss=0.1282, over 5651975.00 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:40:57,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:03,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:05,574 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 15:41:08,878 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:40,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:41:46,197 INFO [optim.py:369] (1/2) 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 15:41:46,972 INFO [train.py:968] (1/2) Epoch 11, batch 8700, giga_loss[loss=0.3304, simple_loss=0.3953, pruned_loss=0.1328, over 28317.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3754, pruned_loss=0.1263, over 5667267.30 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3566, pruned_loss=0.09991, over 5690595.84 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.377, pruned_loss=0.1292, over 5656856.96 frames. ], batch size: 368, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:41:47,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-05 15:42:19,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0083, 3.2382, 2.1087, 0.9005], device='cuda:1'), covar=tensor([0.5484, 0.2148, 0.2800, 0.5458], device='cuda:1'), in_proj_covar=tensor([0.1538, 0.1461, 0.1478, 0.1259], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 15:42:34,380 INFO [train.py:968] (1/2) Epoch 11, batch 8750, giga_loss[loss=0.3161, simple_loss=0.3876, pruned_loss=0.1224, over 28583.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3783, pruned_loss=0.1255, over 5670552.91 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3568, pruned_loss=0.1001, over 5694522.83 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3802, pruned_loss=0.1286, over 5657894.64 frames. ], batch size: 85, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:43:07,055 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-05 15:43:07,480 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 8800, giga_loss[loss=0.3513, simple_loss=0.4047, pruned_loss=0.1489, over 27932.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3783, pruned_loss=0.1239, over 5675402.46 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3567, pruned_loss=0.1002, over 5691192.36 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3809, pruned_loss=0.1274, over 5667621.43 frames. ], batch size: 412, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:43:51,022 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:43:53,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:44:01,468 INFO [train.py:968] (1/2) Epoch 11, batch 8850, giga_loss[loss=0.2935, simple_loss=0.3634, pruned_loss=0.1118, over 28933.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3801, pruned_loss=0.1251, over 5674599.52 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3569, pruned_loss=0.1002, over 5693696.25 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3826, pruned_loss=0.1286, over 5665451.91 frames. ], batch size: 199, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:44:15,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:44:46,585 INFO [optim.py:369] (1/2) 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 15:44:47,239 INFO [train.py:968] (1/2) Epoch 11, batch 8900, giga_loss[loss=0.3593, simple_loss=0.4069, pruned_loss=0.1559, over 28684.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3821, pruned_loss=0.1274, over 5666281.31 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3569, pruned_loss=0.1002, over 5695580.02 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3844, pruned_loss=0.1305, over 5657107.28 frames. ], batch size: 307, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:44:54,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9692, 1.0507, 3.3544, 3.0185], device='cuda:1'), covar=tensor([0.1715, 0.2663, 0.0515, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0661, 0.0593, 0.0864, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 15:45:17,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:45:19,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:45:25,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:45:35,939 INFO [train.py:968] (1/2) Epoch 11, batch 8950, giga_loss[loss=0.2802, simple_loss=0.3477, pruned_loss=0.1064, over 28278.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3813, pruned_loss=0.1279, over 5661680.11 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3569, pruned_loss=0.1002, over 5699160.08 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3834, pruned_loss=0.1308, over 5650722.65 frames. ], batch size: 77, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:45:46,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:46:01,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9569, 2.0635, 1.3906, 1.7484], device='cuda:1'), covar=tensor([0.0721, 0.0518, 0.1018, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0445, 0.0499, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 15:46:20,094 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 9000, giga_loss[loss=0.3169, simple_loss=0.3643, pruned_loss=0.1348, over 23430.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3809, pruned_loss=0.1286, over 5651420.65 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3571, pruned_loss=0.1004, over 5706202.64 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3836, pruned_loss=0.1322, over 5633898.74 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:46:20,107 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 15:46:28,969 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 15:46:43,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-05 15:46:51,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:46:53,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4651, 2.0802, 1.5209, 0.6204], device='cuda:1'), covar=tensor([0.3503, 0.1905, 0.2894, 0.4236], device='cuda:1'), in_proj_covar=tensor([0.1535, 0.1464, 0.1478, 0.1259], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 15:47:00,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:47:16,828 INFO [train.py:968] (1/2) Epoch 11, batch 9050, giga_loss[loss=0.336, simple_loss=0.381, pruned_loss=0.1455, over 28883.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3802, pruned_loss=0.1292, over 5650776.08 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3572, pruned_loss=0.1004, over 5696792.62 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3823, pruned_loss=0.132, over 5645319.42 frames. ], batch size: 106, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:47:26,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464597.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 15:47:51,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:47:55,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:48:08,109 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 9100, giga_loss[loss=0.4163, simple_loss=0.4378, pruned_loss=0.1974, over 26579.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3802, pruned_loss=0.1303, over 5655198.56 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3572, pruned_loss=0.1004, over 5700407.65 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3822, pruned_loss=0.133, over 5646892.85 frames. ], batch size: 555, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:48:23,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:48:56,110 INFO [train.py:968] (1/2) Epoch 11, batch 9150, giga_loss[loss=0.4122, simple_loss=0.4358, pruned_loss=0.1943, over 26606.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3794, pruned_loss=0.1295, over 5649880.26 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3573, pruned_loss=0.1004, over 5698818.36 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3817, pruned_loss=0.1326, over 5643544.79 frames. ], batch size: 555, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 15:49:13,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:16,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:23,675 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=464717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:26,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=464720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:29,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4154, 1.8324, 1.5437, 1.4144], device='cuda:1'), covar=tensor([0.1937, 0.1573, 0.1660, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.1692, 0.1611, 0.1572, 0.1688], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 15:49:42,938 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 9200, libri_loss[loss=0.2413, simple_loss=0.325, pruned_loss=0.07883, over 29570.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5659443.63 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3569, pruned_loss=0.1001, over 5706809.80 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3798, pruned_loss=0.1315, over 5645109.06 frames. ], batch size: 76, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:49:43,213 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:49:52,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=464749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:50:28,570 INFO [train.py:968] (1/2) Epoch 11, batch 9250, libri_loss[loss=0.2344, simple_loss=0.3144, pruned_loss=0.07716, over 28519.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3741, pruned_loss=0.1261, over 5663274.68 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3566, pruned_loss=0.09992, over 5708829.07 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3771, pruned_loss=0.1299, over 5649399.77 frames. ], batch size: 63, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:50:50,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6694, 2.0951, 1.5624, 1.4732], device='cuda:1'), covar=tensor([0.1995, 0.1573, 0.1992, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.1702, 0.1623, 0.1585, 0.1699], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 15:50:54,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-05 15:51:15,459 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 9300, giga_loss[loss=0.3474, simple_loss=0.4097, pruned_loss=0.1426, over 28955.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3736, pruned_loss=0.1254, over 5663797.67 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3564, pruned_loss=0.09975, over 5714927.13 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3769, pruned_loss=0.1296, over 5644889.55 frames. ], batch size: 145, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:51:16,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=464840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:51:46,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4711, 1.6381, 1.4453, 1.2688], device='cuda:1'), covar=tensor([0.1704, 0.1638, 0.1208, 0.1487], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1610, 0.1571, 0.1686], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 15:52:03,997 INFO [train.py:968] (1/2) Epoch 11, batch 9350, giga_loss[loss=0.3253, simple_loss=0.3903, pruned_loss=0.1301, over 28587.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3746, pruned_loss=0.1251, over 5671839.25 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3563, pruned_loss=0.09982, over 5718922.01 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3778, pruned_loss=0.129, over 5651747.32 frames. ], batch size: 307, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:52:12,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1343, 1.8167, 1.6285, 1.4134], device='cuda:1'), covar=tensor([0.0841, 0.0273, 0.0261, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0051, 0.0087], device='cuda:1')
+2023-03-05 15:52:37,439 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 9400, giga_loss[loss=0.2992, simple_loss=0.3691, pruned_loss=0.1147, over 29040.00 frames. ], tot_loss[loss=0.314, simple_loss=0.376, pruned_loss=0.126, over 5671127.50 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3562, pruned_loss=0.09971, over 5718130.31 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1296, over 5655305.22 frames. ], batch size: 155, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:53:19,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=464972.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 15:53:34,306 INFO [train.py:968] (1/2) Epoch 11, batch 9450, libri_loss[loss=0.2771, simple_loss=0.3591, pruned_loss=0.09756, over 28604.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3758, pruned_loss=0.1263, over 5657172.32 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3557, pruned_loss=0.09946, over 5710832.73 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3792, pruned_loss=0.1303, over 5649165.14 frames. ], batch size: 106, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:53:52,503 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-05 15:54:16,274 INFO [optim.py:369] (1/2) 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 15:54:16,287 INFO [train.py:968] (1/2) Epoch 11, batch 9500, giga_loss[loss=0.2661, simple_loss=0.3553, pruned_loss=0.08841, over 29056.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3775, pruned_loss=0.1249, over 5671170.60 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3553, pruned_loss=0.09925, over 5718903.83 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3817, pruned_loss=0.1296, over 5655057.26 frames. ], batch size: 128, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:54:17,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5758, 4.3191, 4.0421, 1.9200], device='cuda:1'), covar=tensor([0.0532, 0.0704, 0.0790, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1052, 0.0985, 0.0863, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 15:54:58,649 INFO [train.py:968] (1/2) Epoch 11, batch 9550, giga_loss[loss=0.326, simple_loss=0.4002, pruned_loss=0.1259, over 29036.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.379, pruned_loss=0.1244, over 5671795.38 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3555, pruned_loss=0.09942, over 5718732.05 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3829, pruned_loss=0.1288, over 5657292.39 frames. ], batch size: 136, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:55:22,653 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=465115.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 15:55:24,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465118.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 15:55:45,814 INFO [optim.py:369] (1/2) 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 15:55:45,827 INFO [train.py:968] (1/2) Epoch 11, batch 9600, giga_loss[loss=0.4106, simple_loss=0.4457, pruned_loss=0.1877, over 27923.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3824, pruned_loss=0.1259, over 5678331.24 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3552, pruned_loss=0.09927, over 5723588.79 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3864, pruned_loss=0.1302, over 5661409.79 frames. ], batch size: 412, lr: 2.98e-03, grad_scale: 8.0
+2023-03-05 15:55:52,639 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465147.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 15:56:31,622 INFO [train.py:968] (1/2) Epoch 11, batch 9650, giga_loss[loss=0.3111, simple_loss=0.376, pruned_loss=0.1231, over 28956.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3852, pruned_loss=0.1289, over 5682389.10 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3549, pruned_loss=0.09908, over 5727354.95 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3892, pruned_loss=0.1329, over 5664962.40 frames. ], batch size: 213, lr: 2.98e-03, grad_scale: 8.0
+2023-03-05 15:56:53,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:57:15,646 INFO [train.py:968] (1/2) Epoch 11, batch 9700, giga_loss[loss=0.4609, simple_loss=0.4555, pruned_loss=0.2332, over 23508.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3869, pruned_loss=0.1313, over 5665875.93 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3552, pruned_loss=0.09945, over 5712285.06 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3904, pruned_loss=0.1348, over 5665780.41 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:57:16,308 INFO [optim.py:369] (1/2) 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 15:57:54,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 15:58:05,295 INFO [train.py:968] (1/2) Epoch 11, batch 9750, giga_loss[loss=0.3549, simple_loss=0.4105, pruned_loss=0.1496, over 28604.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3863, pruned_loss=0.1315, over 5655426.38 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.355, pruned_loss=0.09933, over 5712547.28 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3897, pruned_loss=0.1349, over 5654272.28 frames. ], batch size: 307, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:58:15,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:58:41,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6088, 1.8695, 1.8953, 1.4263], device='cuda:1'), covar=tensor([0.1535, 0.2095, 0.1219, 0.1442], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0704, 0.0864, 0.0770], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 15:58:48,026 INFO [train.py:968] (1/2) Epoch 11, batch 9800, giga_loss[loss=0.3103, simple_loss=0.3887, pruned_loss=0.116, over 28820.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3841, pruned_loss=0.1295, over 5665990.86 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3546, pruned_loss=0.09898, over 5716976.74 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3882, pruned_loss=0.1336, over 5659295.81 frames. ], batch size: 99, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:58:48,726 INFO [optim.py:369] (1/2) 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 15:59:04,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=465358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:59:06,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 15:59:09,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 15:59:17,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2469, 3.0541, 2.8945, 1.4170], device='cuda:1'), covar=tensor([0.0975, 0.1109, 0.1013, 0.2375], device='cuda:1'), in_proj_covar=tensor([0.1053, 0.0990, 0.0866, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 15:59:30,181 INFO [train.py:968] (1/2) Epoch 11, batch 9850, giga_loss[loss=0.2813, simple_loss=0.3691, pruned_loss=0.09678, over 29021.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3834, pruned_loss=0.1272, over 5673202.90 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3543, pruned_loss=0.09893, over 5721134.63 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3876, pruned_loss=0.1311, over 5662937.18 frames. ], batch size: 155, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 15:59:31,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:00:14,584 INFO [train.py:968] (1/2) Epoch 11, batch 9900, giga_loss[loss=0.3111, simple_loss=0.383, pruned_loss=0.1196, over 27871.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3842, pruned_loss=0.1268, over 5676449.44 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3542, pruned_loss=0.09895, over 5723929.73 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3881, pruned_loss=0.1304, over 5665145.82 frames. ], batch size: 412, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:00:15,628 INFO [optim.py:369] (1/2) 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 16:00:18,714 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=465444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:00:20,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:00:41,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 16:00:51,345 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:01:02,980 INFO [train.py:968] (1/2) Epoch 11, batch 9950, giga_loss[loss=0.3462, simple_loss=0.4061, pruned_loss=0.1431, over 29027.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3859, pruned_loss=0.1289, over 5672004.34 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.354, pruned_loss=0.09879, over 5725870.91 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3894, pruned_loss=0.1321, over 5660780.69 frames. ], batch size: 128, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:01:03,234 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=465489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:01:54,410 INFO [train.py:968] (1/2) Epoch 11, batch 10000, giga_loss[loss=0.365, simple_loss=0.4063, pruned_loss=0.1618, over 27625.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.386, pruned_loss=0.1299, over 5665042.84 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3541, pruned_loss=0.09878, over 5727644.38 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3892, pruned_loss=0.1329, over 5653801.41 frames. ], batch size: 472, lr: 2.98e-03, grad_scale: 8.0
+2023-03-05 16:01:55,273 INFO [optim.py:369] (1/2) 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 16:02:42,212 INFO [train.py:968] (1/2) Epoch 11, batch 10050, giga_loss[loss=0.3375, simple_loss=0.373, pruned_loss=0.151, over 23623.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3852, pruned_loss=0.1307, over 5659698.99 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3544, pruned_loss=0.09877, over 5731489.31 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3882, pruned_loss=0.1338, over 5645949.28 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:03:28,081 INFO [train.py:968] (1/2) Epoch 11, batch 10100, giga_loss[loss=0.3267, simple_loss=0.3805, pruned_loss=0.1365, over 28284.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3825, pruned_loss=0.1291, over 5677357.21 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3544, pruned_loss=0.09858, over 5738414.29 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3862, pruned_loss=0.1331, over 5657236.67 frames. ], batch size: 368, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:03:31,035 INFO [optim.py:369] (1/2) 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 16:03:43,150 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 10150, giga_loss[loss=0.3066, simple_loss=0.3705, pruned_loss=0.1214, over 28640.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3802, pruned_loss=0.1286, over 5665052.39 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3543, pruned_loss=0.09848, over 5736145.00 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3834, pruned_loss=0.1321, over 5650508.27 frames. ], batch size: 336, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:05:11,468 INFO [train.py:968] (1/2) Epoch 11, batch 10200, giga_loss[loss=0.268, simple_loss=0.3378, pruned_loss=0.09907, over 28998.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3799, pruned_loss=0.1295, over 5664478.00 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3545, pruned_loss=0.09862, over 5740008.80 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.383, pruned_loss=0.133, over 5647429.33 frames. ], batch size: 128, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:05:15,462 INFO [optim.py:369] (1/2) 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 16:05:59,586 INFO [train.py:968] (1/2) Epoch 11, batch 10250, giga_loss[loss=0.3504, simple_loss=0.3873, pruned_loss=0.1567, over 23912.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3787, pruned_loss=0.1285, over 5655605.81 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3548, pruned_loss=0.09878, over 5733247.50 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3813, pruned_loss=0.1316, over 5646516.83 frames. ], batch size: 705, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:06:00,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-05 16:06:27,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:06:37,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 16:06:42,950 INFO [train.py:968] (1/2) Epoch 11, batch 10300, giga_loss[loss=0.2631, simple_loss=0.3383, pruned_loss=0.09396, over 28758.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3737, pruned_loss=0.1232, over 5675297.51 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3541, pruned_loss=0.09847, over 5738165.81 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.377, pruned_loss=0.1267, over 5661582.96 frames. ], batch size: 99, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:06:46,482 INFO [optim.py:369] (1/2) 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 16:07:05,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=465864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:07:30,554 INFO [train.py:968] (1/2) Epoch 11, batch 10350, giga_loss[loss=0.3608, simple_loss=0.4109, pruned_loss=0.1553, over 28711.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3713, pruned_loss=0.1209, over 5667936.29 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3542, pruned_loss=0.09851, over 5743102.84 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3744, pruned_loss=0.1244, over 5650254.57 frames. ], batch size: 242, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:07:35,501 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=465893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:08:20,589 INFO [train.py:968] (1/2) Epoch 11, batch 10400, giga_loss[loss=0.3231, simple_loss=0.3871, pruned_loss=0.1296, over 28474.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3709, pruned_loss=0.1201, over 5675806.22 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3539, pruned_loss=0.09829, over 5744714.66 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3739, pruned_loss=0.1234, over 5659231.53 frames. ], batch size: 60, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:08:23,338 INFO [optim.py:369] (1/2) 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 16:08:43,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=465962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:08:45,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=465965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:10,887 INFO [train.py:968] (1/2) Epoch 11, batch 10450, giga_loss[loss=0.3339, simple_loss=0.3807, pruned_loss=0.1436, over 27566.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3696, pruned_loss=0.1206, over 5674228.91 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3543, pruned_loss=0.09858, over 5747395.93 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5657409.64 frames. ], batch size: 472, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:09:16,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=465994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:21,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4837, 1.8012, 1.4373, 1.6461], device='cuda:1'), covar=tensor([0.2351, 0.2317, 0.2503, 0.2198], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.0961, 0.1145, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 16:09:29,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:33,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:09:50,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:10:02,103 INFO [train.py:968] (1/2) Epoch 11, batch 10500, giga_loss[loss=0.2801, simple_loss=0.3409, pruned_loss=0.1097, over 28830.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3667, pruned_loss=0.1197, over 5666647.67 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3539, pruned_loss=0.09835, over 5748871.19 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3691, pruned_loss=0.1223, over 5651386.54 frames. ], batch size: 284, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:10:03,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:10:05,481 INFO [optim.py:369] (1/2) 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 16:10:43,118 INFO [train.py:968] (1/2) Epoch 11, batch 10550, giga_loss[loss=0.3123, simple_loss=0.3793, pruned_loss=0.1226, over 28661.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1208, over 5672493.49 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3548, pruned_loss=0.0989, over 5751128.80 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.123, over 5656189.92 frames. ], batch size: 242, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:10:44,436 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:11:28,031 INFO [train.py:968] (1/2) Epoch 11, batch 10600, libri_loss[loss=0.252, simple_loss=0.3301, pruned_loss=0.08698, over 29646.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3717, pruned_loss=0.1215, over 5664315.66 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3551, pruned_loss=0.09901, over 5750629.76 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3732, pruned_loss=0.1239, over 5649210.00 frames. ], batch size: 69, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:11:31,414 INFO [optim.py:369] (1/2) 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 16:11:59,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:12:01,641 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 10650, giga_loss[loss=0.3135, simple_loss=0.3765, pruned_loss=0.1253, over 28907.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3713, pruned_loss=0.1215, over 5648679.81 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3552, pruned_loss=0.09901, over 5743203.26 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.1239, over 5641830.82 frames. ], batch size: 227, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:12:26,672 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:12:58,328 INFO [train.py:968] (1/2) Epoch 11, batch 10700, giga_loss[loss=0.3263, simple_loss=0.3863, pruned_loss=0.1332, over 28801.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.122, over 5648103.13 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3552, pruned_loss=0.099, over 5736638.23 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3732, pruned_loss=0.1246, over 5645419.07 frames. ], batch size: 66, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:13:00,889 INFO [optim.py:369] (1/2) 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 16:13:24,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:13:47,434 INFO [train.py:968] (1/2) Epoch 11, batch 10750, giga_loss[loss=0.3487, simple_loss=0.4066, pruned_loss=0.1454, over 28867.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1238, over 5658135.51 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3555, pruned_loss=0.09921, over 5738363.07 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3752, pruned_loss=0.1262, over 5652983.58 frames. ], batch size: 227, lr: 2.98e-03, grad_scale: 2.0
+2023-03-05 16:14:08,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2514, 1.6371, 1.5599, 1.1206], device='cuda:1'), covar=tensor([0.1570, 0.2481, 0.1414, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0705, 0.0867, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 16:14:36,642 INFO [train.py:968] (1/2) Epoch 11, batch 10800, giga_loss[loss=0.3334, simple_loss=0.3937, pruned_loss=0.1365, over 28880.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3764, pruned_loss=0.1252, over 5659510.53 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3558, pruned_loss=0.09936, over 5740894.93 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3775, pruned_loss=0.1273, over 5652023.16 frames. ], batch size: 213, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:14:39,545 INFO [optim.py:369] (1/2) 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 16:14:52,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9843, 1.3357, 1.0768, 0.1506], device='cuda:1'), covar=tensor([0.2241, 0.1853, 0.2862, 0.4056], device='cuda:1'), in_proj_covar=tensor([0.1536, 0.1454, 0.1473, 0.1252], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 16:15:22,811 INFO [train.py:968] (1/2) Epoch 11, batch 10850, giga_loss[loss=0.3321, simple_loss=0.3916, pruned_loss=0.1363, over 28952.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.378, pruned_loss=0.1262, over 5664703.60 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3561, pruned_loss=0.09954, over 5744164.62 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3791, pruned_loss=0.1283, over 5654238.09 frames. ], batch size: 145, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:15:43,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:15:45,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:15:47,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:16:09,714 INFO [train.py:968] (1/2) Epoch 11, batch 10900, giga_loss[loss=0.3169, simple_loss=0.381, pruned_loss=0.1264, over 28890.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3796, pruned_loss=0.1279, over 5674404.43 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3557, pruned_loss=0.09926, over 5747673.41 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3812, pruned_loss=0.1303, over 5661644.52 frames. ], batch size: 174, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:16:14,102 INFO [optim.py:369] (1/2) 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 16:16:14,358 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:16:37,218 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 10950, giga_loss[loss=0.3046, simple_loss=0.3783, pruned_loss=0.1154, over 28849.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3815, pruned_loss=0.1287, over 5676591.10 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3558, pruned_loss=0.09917, over 5750108.03 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3831, pruned_loss=0.1312, over 5663369.70 frames. ], batch size: 199, lr: 2.98e-03, grad_scale: 4.0
+2023-03-05 16:17:01,696 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:17:24,721 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3702, 2.9086, 1.4320, 1.4732], device='cuda:1'), covar=tensor([0.0869, 0.0351, 0.0829, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0512, 0.0338, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 16:17:52,741 INFO [train.py:968] (1/2) Epoch 11, batch 11000, giga_loss[loss=0.3018, simple_loss=0.3646, pruned_loss=0.1195, over 28930.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3811, pruned_loss=0.1276, over 5669435.86 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.356, pruned_loss=0.09928, over 5752670.01 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3826, pruned_loss=0.1298, over 5655994.36 frames. ], batch size: 227, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:17:54,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-05 16:17:58,503 INFO [optim.py:369] (1/2) 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 16:18:11,603 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:18:16,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:18:43,226 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=466588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:18:43,542 INFO [train.py:968] (1/2) Epoch 11, batch 11050, giga_loss[loss=0.3968, simple_loss=0.4235, pruned_loss=0.1851, over 26595.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3829, pruned_loss=0.1301, over 5658455.44 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3566, pruned_loss=0.09981, over 5752403.72 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3839, pruned_loss=0.1319, over 5646505.21 frames. ], batch size: 555, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:19:02,903 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=466608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:19:05,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=466611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:19:20,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-05 16:19:34,615 INFO [train.py:968] (1/2) Epoch 11, batch 11100, giga_loss[loss=0.3242, simple_loss=0.3856, pruned_loss=0.1314, over 28971.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3816, pruned_loss=0.1299, over 5648427.15 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3568, pruned_loss=0.09996, over 5755473.75 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3828, pruned_loss=0.1319, over 5633877.70 frames. ], batch size: 106, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:19:36,443 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:20:13,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=466672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:20:29,773 INFO [train.py:968] (1/2) Epoch 11, batch 11150, giga_loss[loss=0.3274, simple_loss=0.3861, pruned_loss=0.1343, over 28642.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3788, pruned_loss=0.1284, over 5653714.04 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3565, pruned_loss=0.09972, over 5757547.84 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3804, pruned_loss=0.1306, over 5638910.08 frames. ], batch size: 336, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:21:17,280 INFO [train.py:968] (1/2) Epoch 11, batch 11200, libri_loss[loss=0.2391, simple_loss=0.3177, pruned_loss=0.08024, over 29358.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3772, pruned_loss=0.1278, over 5656589.67 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3561, pruned_loss=0.09951, over 5760823.14 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3794, pruned_loss=0.1304, over 5639538.24 frames. ], batch size: 71, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:21:19,813 INFO [optim.py:369] (1/2) 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 16:21:36,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-05 16:22:03,457 INFO [train.py:968] (1/2) Epoch 11, batch 11250, giga_loss[loss=0.3084, simple_loss=0.3689, pruned_loss=0.1239, over 29073.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3769, pruned_loss=0.1274, over 5667495.90 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3564, pruned_loss=0.09967, over 5764594.76 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3788, pruned_loss=0.13, over 5648151.78 frames. ], batch size: 128, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:22:53,072 INFO [train.py:968] (1/2) Epoch 11, batch 11300, giga_loss[loss=0.2907, simple_loss=0.3549, pruned_loss=0.1132, over 29185.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3771, pruned_loss=0.128, over 5660740.44 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3567, pruned_loss=0.09971, over 5766643.77 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3786, pruned_loss=0.1305, over 5642401.18 frames. ], batch size: 113, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:22:58,170 INFO [optim.py:369] (1/2) 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 16:23:16,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=466866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:23:38,966 INFO [train.py:968] (1/2) Epoch 11, batch 11350, giga_loss[loss=0.3069, simple_loss=0.3769, pruned_loss=0.1185, over 28910.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3775, pruned_loss=0.1286, over 5646844.09 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3566, pruned_loss=0.09981, over 5750792.57 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3791, pruned_loss=0.1311, over 5643920.05 frames. ], batch size: 174, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:24:25,044 INFO [train.py:968] (1/2) Epoch 11, batch 11400, giga_loss[loss=0.3774, simple_loss=0.4163, pruned_loss=0.1693, over 27613.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3809, pruned_loss=0.1316, over 5654900.28 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3568, pruned_loss=0.09992, over 5753202.55 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3824, pruned_loss=0.1338, over 5649281.22 frames. ], batch size: 472, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:24:29,029 INFO [optim.py:369] (1/2) 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 16:24:36,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7267, 2.1468, 1.7980, 1.5124], device='cuda:1'), covar=tensor([0.2170, 0.1542, 0.1606, 0.1840], device='cuda:1'), in_proj_covar=tensor([0.1722, 0.1628, 0.1585, 0.1701], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 16:25:15,835 INFO [train.py:968] (1/2) Epoch 11, batch 11450, giga_loss[loss=0.3015, simple_loss=0.3684, pruned_loss=0.1173, over 28918.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3818, pruned_loss=0.133, over 5642525.40 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3568, pruned_loss=0.09987, over 5751424.49 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3832, pruned_loss=0.1351, over 5638466.58 frames. ], batch size: 227, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:25:19,307 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6556, 1.7704, 1.6726, 1.6415], device='cuda:1'), covar=tensor([0.1325, 0.1804, 0.1950, 0.1625], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0741, 0.0677, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 16:25:34,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=467009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:25:37,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=467012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:25:49,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4691, 1.3942, 1.6129, 1.1908], device='cuda:1'), covar=tensor([0.1748, 0.2939, 0.1441, 0.1549], device='cuda:1'), in_proj_covar=tensor([0.0821, 0.0702, 0.0863, 0.0770], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 16:26:01,398 INFO [train.py:968] (1/2) Epoch 11, batch 11500, giga_loss[loss=0.3102, simple_loss=0.3721, pruned_loss=0.1242, over 28920.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.382, pruned_loss=0.1334, over 5644696.06 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3569, pruned_loss=0.1001, over 5743913.57 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3838, pruned_loss=0.1358, over 5644428.52 frames. ], batch size: 112, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:26:03,326 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=467041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:26:06,261 INFO [optim.py:369] (1/2) 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 16:26:08,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=467047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:26:08,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4244, 1.7308, 1.3295, 1.7211], device='cuda:1'), covar=tensor([0.2387, 0.2404, 0.2643, 0.2191], device='cuda:1'), in_proj_covar=tensor([0.1294, 0.0965, 0.1143, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 16:26:19,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6557, 1.6468, 1.2288, 1.2511], device='cuda:1'), covar=tensor([0.0654, 0.0535, 0.0939, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0446, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 16:26:48,438 INFO [train.py:968] (1/2) Epoch 11, batch 11550, giga_loss[loss=0.3472, simple_loss=0.4053, pruned_loss=0.1445, over 28226.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.382, pruned_loss=0.1331, over 5643453.28 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.357, pruned_loss=0.1001, over 5747299.68 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3839, pruned_loss=0.1358, over 5638253.22 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:27:05,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7553, 1.8738, 1.7185, 1.7408], device='cuda:1'), covar=tensor([0.1528, 0.1925, 0.1935, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0742, 0.0677, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 16:27:28,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=467134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:27:34,625 INFO [train.py:968] (1/2) Epoch 11, batch 11600, giga_loss[loss=0.3159, simple_loss=0.374, pruned_loss=0.1288, over 28659.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3817, pruned_loss=0.1319, over 5659148.54 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3571, pruned_loss=0.1003, over 5750898.00 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3837, pruned_loss=0.1346, over 5649695.28 frames. ], batch size: 307, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:27:39,159 INFO [optim.py:369] (1/2) 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 16:27:58,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-05 16:28:15,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2170, 1.4286, 1.3272, 1.2626], device='cuda:1'), covar=tensor([0.1225, 0.1237, 0.1403, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0739, 0.0675, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 16:28:22,268 INFO [train.py:968] (1/2) Epoch 11, batch 11650, giga_loss[loss=0.3305, simple_loss=0.3915, pruned_loss=0.1347, over 28464.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3825, pruned_loss=0.1321, over 5665684.59 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.357, pruned_loss=0.1001, over 5753056.63 frames. ], giga_tot_loss[loss=0.327, simple_loss=0.3845, pruned_loss=0.1348, over 5655242.43 frames. ], batch size: 71, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:28:23,319 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=467190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:28:25,267 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=467222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:28:56,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2635, 2.1560, 1.9403, 1.9267], device='cuda:1'), covar=tensor([0.1287, 0.1948, 0.1699, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0738, 0.0674, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 16:29:15,762 INFO [train.py:968] (1/2) Epoch 11, batch 11700, giga_loss[loss=0.2959, simple_loss=0.3701, pruned_loss=0.1108, over 29138.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3853, pruned_loss=0.1351, over 5652001.77 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3572, pruned_loss=0.1002, over 5748836.68 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3869, pruned_loss=0.1375, over 5646574.61 frames. ], batch size: 155, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:29:22,829 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 11750, giga_loss[loss=0.3177, simple_loss=0.3857, pruned_loss=0.1249, over 28870.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3847, pruned_loss=0.1349, over 5653960.97 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3569, pruned_loss=0.1, over 5752998.71 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3868, pruned_loss=0.1378, over 5643638.66 frames. ], batch size: 174, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:30:20,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4005, 1.1704, 4.0358, 3.2448], device='cuda:1'), covar=tensor([0.1525, 0.2602, 0.0459, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0660, 0.0591, 0.0858, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 16:30:48,062 INFO [train.py:968] (1/2) Epoch 11, batch 11800, giga_loss[loss=0.3233, simple_loss=0.3803, pruned_loss=0.1331, over 28319.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3842, pruned_loss=0.1334, over 5657650.17 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3569, pruned_loss=0.09987, over 5755665.83 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3866, pruned_loss=0.1367, over 5644670.49 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:30:52,215 INFO [optim.py:369] (1/2) 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 16:31:32,641 INFO [train.py:968] (1/2) Epoch 11, batch 11850, giga_loss[loss=0.3109, simple_loss=0.3802, pruned_loss=0.1208, over 28457.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3836, pruned_loss=0.1318, over 5658186.20 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.357, pruned_loss=0.09991, over 5754462.83 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3865, pruned_loss=0.1356, over 5645745.07 frames. ], batch size: 65, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:32:07,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0657, 1.1716, 3.6987, 3.0785], device='cuda:1'), covar=tensor([0.1712, 0.2628, 0.0439, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0660, 0.0591, 0.0859, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 16:32:21,522 INFO [train.py:968] (1/2) Epoch 11, batch 11900, giga_loss[loss=0.2749, simple_loss=0.3464, pruned_loss=0.1017, over 28965.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3828, pruned_loss=0.1311, over 5661669.53 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3567, pruned_loss=0.09977, over 5756091.71 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3855, pruned_loss=0.1345, over 5649534.32 frames. ], batch size: 136, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:32:27,712 INFO [optim.py:369] (1/2) 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 16:33:05,947 INFO [train.py:968] (1/2) Epoch 11, batch 11950, giga_loss[loss=0.3005, simple_loss=0.3656, pruned_loss=0.1177, over 28745.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3817, pruned_loss=0.1303, over 5652986.84 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3572, pruned_loss=0.1001, over 5751942.52 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3843, pruned_loss=0.1337, over 5643355.85 frames. ], batch size: 262, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 16:33:23,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=467509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:33:30,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5268, 1.8090, 1.3919, 1.6855], device='cuda:1'), covar=tensor([0.2579, 0.2461, 0.2837, 0.2265], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.0958, 0.1139, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 16:33:50,993 INFO [train.py:968] (1/2) Epoch 11, batch 12000, giga_loss[loss=0.3338, simple_loss=0.3982, pruned_loss=0.1347, over 28718.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3801, pruned_loss=0.1286, over 5669840.89 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3568, pruned_loss=0.09985, over 5756264.12 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3834, pruned_loss=0.1326, over 5655040.81 frames. ], batch size: 242, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:33:50,994 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 16:33:56,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1611, 1.5154, 1.5535, 1.3735], device='cuda:1'), covar=tensor([0.1454, 0.1311, 0.1647, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0735, 0.0671, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 16:33:59,455 INFO [train.py:1012] (1/2) Epoch 11, validation: loss=0.218, simple_loss=0.3234, pruned_loss=0.05624, over 944034.00 frames. 
+2023-03-05 16:33:59,455 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 16:34:02,061 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-05 16:34:06,302 INFO [optim.py:369] (1/2) 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 16:34:44,946 INFO [train.py:968] (1/2) Epoch 11, batch 12050, libri_loss[loss=0.2383, simple_loss=0.3199, pruned_loss=0.07836, over 28124.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.381, pruned_loss=0.1294, over 5653781.43 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3569, pruned_loss=0.09987, over 5758211.00 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3843, pruned_loss=0.1336, over 5636755.61 frames. ], batch size: 62, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:35:04,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4603, 2.1661, 1.4748, 0.6089], device='cuda:1'), covar=tensor([0.4113, 0.2159, 0.3073, 0.4687], device='cuda:1'), in_proj_covar=tensor([0.1540, 0.1468, 0.1476, 0.1263], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 16:35:09,045 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=467612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:35:32,064 INFO [train.py:968] (1/2) Epoch 11, batch 12100, libri_loss[loss=0.3062, simple_loss=0.3821, pruned_loss=0.1151, over 29536.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3811, pruned_loss=0.1297, over 5667998.96 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3568, pruned_loss=0.09973, over 5761932.24 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3845, pruned_loss=0.1339, over 5648707.40 frames. ], batch size: 84, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:35:37,002 INFO [optim.py:369] (1/2) 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 16:35:48,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=467652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:35:49,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=467655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:36:19,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=467684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:36:22,226 INFO [train.py:968] (1/2) Epoch 11, batch 12150, giga_loss[loss=0.359, simple_loss=0.4011, pruned_loss=0.1584, over 27700.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3799, pruned_loss=0.1292, over 5675639.09 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.357, pruned_loss=0.09983, over 5761811.48 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3826, pruned_loss=0.1329, over 5659235.19 frames. ], batch size: 472, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:37:07,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2619, 1.4400, 1.4598, 1.3846], device='cuda:1'), covar=tensor([0.1084, 0.1103, 0.1473, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0740, 0.0677, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 16:37:08,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4228, 1.7510, 0.9588, 1.4063], device='cuda:1'), covar=tensor([0.1132, 0.0854, 0.1698, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0446, 0.0503, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 16:37:08,495 INFO [train.py:968] (1/2) Epoch 11, batch 12200, giga_loss[loss=0.3168, simple_loss=0.3879, pruned_loss=0.1228, over 28862.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3808, pruned_loss=0.1299, over 5677101.05 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3569, pruned_loss=0.09968, over 5765373.09 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3836, pruned_loss=0.1337, over 5658840.46 frames. ], batch size: 174, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:37:16,164 INFO [optim.py:369] (1/2) 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 16:37:29,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-05 16:37:44,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3984, 1.6647, 1.2464, 1.7317], device='cuda:1'), covar=tensor([0.2381, 0.2377, 0.2629, 0.1992], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.0966, 0.1142, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 16:37:54,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8951, 1.1244, 3.3187, 2.7835], device='cuda:1'), covar=tensor([0.1717, 0.2551, 0.0514, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0662, 0.0592, 0.0861, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 16:37:55,253 INFO [train.py:968] (1/2) Epoch 11, batch 12250, libri_loss[loss=0.2711, simple_loss=0.3491, pruned_loss=0.09649, over 29573.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3816, pruned_loss=0.1303, over 5681223.67 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3568, pruned_loss=0.09958, over 5768709.21 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3849, pruned_loss=0.1344, over 5660396.93 frames. ], batch size: 75, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:38:35,890 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-05 16:38:39,402 INFO [train.py:968] (1/2) Epoch 11, batch 12300, giga_loss[loss=0.295, simple_loss=0.361, pruned_loss=0.1145, over 28669.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3818, pruned_loss=0.1307, over 5675257.98 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3568, pruned_loss=0.09955, over 5767426.17 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3849, pruned_loss=0.1346, over 5658164.81 frames. ], batch size: 85, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:38:46,245 INFO [optim.py:369] (1/2) 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 16:38:57,022 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=467856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:39:31,591 INFO [train.py:968] (1/2) Epoch 11, batch 12350, giga_loss[loss=0.2964, simple_loss=0.3654, pruned_loss=0.1137, over 28915.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3801, pruned_loss=0.1286, over 5679335.65 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3567, pruned_loss=0.09952, over 5769073.44 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3828, pruned_loss=0.132, over 5663863.20 frames. ], batch size: 213, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:39:42,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4458, 1.6930, 1.4194, 1.4165], device='cuda:1'), covar=tensor([0.1826, 0.1744, 0.1718, 0.1573], device='cuda:1'), in_proj_covar=tensor([0.1287, 0.0961, 0.1136, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 16:40:17,472 INFO [train.py:968] (1/2) Epoch 11, batch 12400, giga_loss[loss=0.3101, simple_loss=0.3695, pruned_loss=0.1254, over 28491.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3796, pruned_loss=0.1275, over 5673792.94 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3567, pruned_loss=0.09955, over 5761734.63 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3822, pruned_loss=0.1308, over 5665226.46 frames. ], batch size: 71, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:40:24,177 INFO [optim.py:369] (1/2) 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 16:41:00,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=467987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:41:02,045 INFO [train.py:968] (1/2) Epoch 11, batch 12450, giga_loss[loss=0.2615, simple_loss=0.3389, pruned_loss=0.09201, over 28570.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3786, pruned_loss=0.1268, over 5682698.53 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3561, pruned_loss=0.09928, over 5764301.93 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3816, pruned_loss=0.1301, over 5672091.43 frames. ], batch size: 78, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:41:33,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9027, 1.8641, 1.8237, 1.7698], device='cuda:1'), covar=tensor([0.1383, 0.1990, 0.1971, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0739, 0.0676, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 16:41:41,739 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 16:41:54,310 INFO [train.py:968] (1/2) Epoch 11, batch 12500, giga_loss[loss=0.3371, simple_loss=0.4006, pruned_loss=0.1368, over 28221.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3796, pruned_loss=0.1288, over 5671767.40 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3561, pruned_loss=0.09925, over 5765028.42 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3821, pruned_loss=0.1315, over 5662452.12 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:42:02,155 INFO [optim.py:369] (1/2) 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 16:42:43,459 INFO [train.py:968] (1/2) Epoch 11, batch 12550, giga_loss[loss=0.3332, simple_loss=0.3866, pruned_loss=0.14, over 28708.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.377, pruned_loss=0.1273, over 5671412.03 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3562, pruned_loss=0.09929, over 5766553.66 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.379, pruned_loss=0.1297, over 5661876.93 frames. ], batch size: 262, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:42:44,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3221, 1.7391, 1.3273, 1.5762], device='cuda:1'), covar=tensor([0.2073, 0.1976, 0.2216, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.0961, 0.1138, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 16:42:45,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-05 16:43:22,725 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:43:24,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 12600, giga_loss[loss=0.2893, simple_loss=0.3528, pruned_loss=0.1129, over 28985.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3737, pruned_loss=0.126, over 5680805.65 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3563, pruned_loss=0.09932, over 5769155.80 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3757, pruned_loss=0.1284, over 5669448.53 frames. ], batch size: 213, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:43:40,208 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:44:17,151 INFO [train.py:968] (1/2) Epoch 11, batch 12650, giga_loss[loss=0.3215, simple_loss=0.3788, pruned_loss=0.1322, over 28975.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3714, pruned_loss=0.1246, over 5679170.75 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3568, pruned_loss=0.09965, over 5761577.59 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3727, pruned_loss=0.1268, over 5674511.46 frames. ], batch size: 136, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:44:34,005 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=468203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:44:41,564 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=468213.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 16:44:59,533 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 12700, giga_loss[loss=0.2654, simple_loss=0.3382, pruned_loss=0.09626, over 28580.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3692, pruned_loss=0.1234, over 5686611.67 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3568, pruned_loss=0.09959, over 5762077.66 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3705, pruned_loss=0.1255, over 5681373.54 frames. ], batch size: 307, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:45:13,006 INFO [optim.py:369] (1/2) 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 16:45:53,638 INFO [train.py:968] (1/2) Epoch 11, batch 12750, giga_loss[loss=0.3192, simple_loss=0.3871, pruned_loss=0.1256, over 28189.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.37, pruned_loss=0.1229, over 5675924.92 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3569, pruned_loss=0.0998, over 5754217.72 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3711, pruned_loss=0.1247, over 5677754.51 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:45:59,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5657, 5.3612, 5.0298, 2.6842], device='cuda:1'), covar=tensor([0.0372, 0.0560, 0.0656, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.1075, 0.1006, 0.0881, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 16:46:41,347 INFO [train.py:968] (1/2) Epoch 11, batch 12800, giga_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.09751, over 28859.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3676, pruned_loss=0.1189, over 5681327.48 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3564, pruned_loss=0.09956, over 5758215.01 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3694, pruned_loss=0.1214, over 5676057.03 frames. ], batch size: 145, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:46:47,326 INFO [optim.py:369] (1/2) 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 16:47:17,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:47:20,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:47:24,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.02 vs. limit=2.0
+2023-03-05 16:47:32,525 INFO [train.py:968] (1/2) Epoch 11, batch 12850, libri_loss[loss=0.2537, simple_loss=0.3198, pruned_loss=0.09384, over 29364.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3651, pruned_loss=0.1164, over 5672453.45 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3552, pruned_loss=0.09909, over 5761920.31 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.368, pruned_loss=0.1194, over 5662030.09 frames. ], batch size: 71, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:47:41,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7104, 2.3593, 1.4746, 0.9826], device='cuda:1'), covar=tensor([0.4631, 0.2558, 0.2799, 0.3992], device='cuda:1'), in_proj_covar=tensor([0.1548, 0.1468, 0.1480, 0.1262], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 16:47:46,496 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:48:17,632 INFO [train.py:968] (1/2) Epoch 11, batch 12900, giga_loss[loss=0.246, simple_loss=0.3352, pruned_loss=0.07846, over 28909.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3623, pruned_loss=0.1133, over 5668197.17 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.09904, over 5756996.49 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3654, pruned_loss=0.1165, over 5659998.34 frames. ], batch size: 145, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:48:24,039 INFO [optim.py:369] (1/2) 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 16:48:47,198 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=468468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:49:10,675 INFO [train.py:968] (1/2) Epoch 11, batch 12950, giga_loss[loss=0.2722, simple_loss=0.3587, pruned_loss=0.09292, over 28889.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3589, pruned_loss=0.1105, over 5667507.51 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3541, pruned_loss=0.09883, over 5759430.74 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.362, pruned_loss=0.1134, over 5657723.06 frames. ], batch size: 199, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:49:56,620 INFO [train.py:968] (1/2) Epoch 11, batch 13000, giga_loss[loss=0.2579, simple_loss=0.3479, pruned_loss=0.08392, over 28870.00 frames. ], tot_loss[loss=0.287, simple_loss=0.358, pruned_loss=0.108, over 5664977.98 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3538, pruned_loss=0.09894, over 5754300.34 frames. ], giga_tot_loss[loss=0.2909, simple_loss=0.3608, pruned_loss=0.1105, over 5658531.82 frames. ], batch size: 199, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:50:04,616 INFO [optim.py:369] (1/2) 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 16:50:27,067 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7358, 5.1258, 1.9320, 2.0612], device='cuda:1'), covar=tensor([0.0897, 0.0240, 0.0871, 0.1218], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0512, 0.0338, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 16:50:34,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=468578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:50:36,992 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 16:50:45,719 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 13050, giga_loss[loss=0.3122, simple_loss=0.378, pruned_loss=0.1231, over 28284.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3583, pruned_loss=0.1074, over 5652462.49 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3539, pruned_loss=0.09919, over 5747554.01 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3607, pruned_loss=0.1094, over 5651216.91 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:51:18,052 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-05 16:51:39,169 INFO [train.py:968] (1/2) Epoch 11, batch 13100, giga_loss[loss=0.3198, simple_loss=0.3823, pruned_loss=0.1286, over 28425.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3581, pruned_loss=0.107, over 5659954.04 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3542, pruned_loss=0.09946, over 5749659.01 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3598, pruned_loss=0.1084, over 5655937.03 frames. ], batch size: 368, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:51:41,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4571, 1.7629, 1.3302, 1.7561], device='cuda:1'), covar=tensor([0.2448, 0.2346, 0.2729, 0.2131], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.0958, 0.1142, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 16:51:47,206 INFO [optim.py:369] (1/2) 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 16:52:27,359 INFO [train.py:968] (1/2) Epoch 11, batch 13150, giga_loss[loss=0.2684, simple_loss=0.3426, pruned_loss=0.09709, over 28984.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3559, pruned_loss=0.1058, over 5662136.29 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3539, pruned_loss=0.09935, over 5753001.10 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3576, pruned_loss=0.1071, over 5654421.79 frames. ], batch size: 213, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:52:58,351 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:53:01,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:53:09,919 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468731.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 16:53:11,925 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 13200, giga_loss[loss=0.305, simple_loss=0.3698, pruned_loss=0.1202, over 28714.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3532, pruned_loss=0.1039, over 5674526.10 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3533, pruned_loss=0.09934, over 5757220.01 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.355, pruned_loss=0.1052, over 5661178.68 frames. ], batch size: 262, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:53:23,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-05 16:53:24,634 INFO [optim.py:369] (1/2) 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 16:53:30,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:53:41,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=468763.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 16:54:07,425 INFO [train.py:968] (1/2) Epoch 11, batch 13250, libri_loss[loss=0.253, simple_loss=0.3227, pruned_loss=0.09171, over 29483.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3525, pruned_loss=0.1033, over 5671466.68 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3526, pruned_loss=0.099, over 5759911.74 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3545, pruned_loss=0.1047, over 5656770.82 frames. ], batch size: 70, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:54:56,051 INFO [train.py:968] (1/2) Epoch 11, batch 13300, giga_loss[loss=0.2601, simple_loss=0.3398, pruned_loss=0.09024, over 28532.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3507, pruned_loss=0.1017, over 5669568.83 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3522, pruned_loss=0.09872, over 5762678.16 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3527, pruned_loss=0.1032, over 5653863.58 frames. ], batch size: 336, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 16:55:00,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=468843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:55:03,468 INFO [optim.py:369] (1/2) 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 16:55:48,830 INFO [train.py:968] (1/2) Epoch 11, batch 13350, giga_loss[loss=0.2654, simple_loss=0.3451, pruned_loss=0.09285, over 28963.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3484, pruned_loss=0.0995, over 5673592.65 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3518, pruned_loss=0.09854, over 5764245.18 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3503, pruned_loss=0.1008, over 5659037.29 frames. ], batch size: 213, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:55:51,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-05 16:56:04,156 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-05 16:56:40,025 INFO [train.py:968] (1/2) Epoch 11, batch 13400, giga_loss[loss=0.2592, simple_loss=0.3338, pruned_loss=0.09235, over 28640.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3442, pruned_loss=0.09678, over 5664355.87 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3515, pruned_loss=0.09835, over 5757072.43 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.09796, over 5657814.88 frames. ], batch size: 99, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:56:52,966 INFO [optim.py:369] (1/2) 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 16:57:32,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=468986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:57:34,449 INFO [train.py:968] (1/2) Epoch 11, batch 13450, giga_loss[loss=0.2582, simple_loss=0.3325, pruned_loss=0.09196, over 27965.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3425, pruned_loss=0.09714, over 5651245.94 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3513, pruned_loss=0.09841, over 5756546.18 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3438, pruned_loss=0.09799, over 5644448.66 frames. ], batch size: 412, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:57:34,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=468989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:58:01,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=469018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:58:23,708 INFO [train.py:968] (1/2) Epoch 11, batch 13500, giga_loss[loss=0.2766, simple_loss=0.3528, pruned_loss=0.1002, over 28611.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3426, pruned_loss=0.09807, over 5638630.32 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3514, pruned_loss=0.09866, over 5746443.44 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3433, pruned_loss=0.0985, over 5639515.73 frames. ], batch size: 242, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:58:27,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9567, 4.9420, 2.0579, 2.0137], device='cuda:1'), covar=tensor([0.0820, 0.0247, 0.0804, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0505, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 16:58:34,356 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 16:58:34,757 INFO [optim.py:369] (1/2) 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 16:59:23,607 INFO [train.py:968] (1/2) Epoch 11, batch 13550, giga_loss[loss=0.2978, simple_loss=0.3746, pruned_loss=0.1106, over 28605.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3441, pruned_loss=0.09901, over 5627764.40 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3515, pruned_loss=0.09874, over 5749036.33 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3445, pruned_loss=0.09928, over 5624569.23 frames. ], batch size: 307, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 16:59:25,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9613, 1.9398, 1.4295, 1.5058], device='cuda:1'), covar=tensor([0.0765, 0.0563, 0.0960, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0438, 0.0496, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 17:00:20,378 INFO [train.py:968] (1/2) Epoch 11, batch 13600, giga_loss[loss=0.2619, simple_loss=0.3455, pruned_loss=0.08919, over 28022.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3458, pruned_loss=0.09823, over 5644021.23 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3514, pruned_loss=0.09884, over 5751297.87 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3461, pruned_loss=0.09835, over 5637868.92 frames. ], batch size: 412, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 17:00:29,736 INFO [optim.py:369] (1/2) 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 17:01:19,328 INFO [train.py:968] (1/2) Epoch 11, batch 13650, giga_loss[loss=0.3011, simple_loss=0.3718, pruned_loss=0.1152, over 28033.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3467, pruned_loss=0.09843, over 5640211.31 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3516, pruned_loss=0.09889, over 5752643.28 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3467, pruned_loss=0.09848, over 5633408.23 frames. ], batch size: 412, lr: 2.97e-03, grad_scale: 8.0
+2023-03-05 17:01:42,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 17:02:23,279 INFO [train.py:968] (1/2) Epoch 11, batch 13700, giga_loss[loss=0.3141, simple_loss=0.3737, pruned_loss=0.1273, over 26703.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3461, pruned_loss=0.09821, over 5644297.96 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3516, pruned_loss=0.09895, over 5754680.90 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.346, pruned_loss=0.09819, over 5635457.88 frames. ], batch size: 555, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:02:34,562 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 13750, giga_loss[loss=0.2244, simple_loss=0.3141, pruned_loss=0.0674, over 29176.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3438, pruned_loss=0.09633, over 5651557.58 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.351, pruned_loss=0.09871, over 5758432.46 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3442, pruned_loss=0.09647, over 5638722.35 frames. ], batch size: 113, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:04:18,966 INFO [train.py:968] (1/2) Epoch 11, batch 13800, giga_loss[loss=0.2386, simple_loss=0.3188, pruned_loss=0.07921, over 27626.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3416, pruned_loss=0.0938, over 5653107.82 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3505, pruned_loss=0.09853, over 5759984.61 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3421, pruned_loss=0.094, over 5639253.89 frames. ], batch size: 472, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:04:30,991 INFO [optim.py:369] (1/2) 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 17:05:19,848 INFO [train.py:968] (1/2) Epoch 11, batch 13850, giga_loss[loss=0.2982, simple_loss=0.3606, pruned_loss=0.1179, over 29017.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3388, pruned_loss=0.09308, over 5649385.96 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3499, pruned_loss=0.09819, over 5754669.74 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3395, pruned_loss=0.09337, over 5640336.32 frames. ], batch size: 155, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 17:05:26,012 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:05:33,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9627, 1.3473, 1.1029, 0.1108], device='cuda:1'), covar=tensor([0.2390, 0.2062, 0.3141, 0.4202], device='cuda:1'), in_proj_covar=tensor([0.1540, 0.1457, 0.1470, 0.1261], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 17:05:59,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:06:03,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5192, 1.5706, 1.1866, 1.1840], device='cuda:1'), covar=tensor([0.0727, 0.0451, 0.0900, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0436, 0.0493, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 17:06:08,004 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 13900, giga_loss[loss=0.2708, simple_loss=0.3205, pruned_loss=0.1105, over 24171.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3378, pruned_loss=0.09345, over 5656543.40 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3492, pruned_loss=0.09801, over 5758282.37 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3387, pruned_loss=0.09373, over 5643986.69 frames. ], batch size: 705, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 17:06:31,901 INFO [optim.py:369] (1/2) 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 17:07:16,854 INFO [train.py:968] (1/2) Epoch 11, batch 13950, giga_loss[loss=0.2368, simple_loss=0.3156, pruned_loss=0.07904, over 28808.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3384, pruned_loss=0.09394, over 5667542.60 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3489, pruned_loss=0.09796, over 5761180.82 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3391, pruned_loss=0.09412, over 5653429.00 frames. ], batch size: 243, lr: 2.97e-03, grad_scale: 2.0
+2023-03-05 17:07:26,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2581, 1.6371, 1.2817, 0.6896], device='cuda:1'), covar=tensor([0.3354, 0.1782, 0.2445, 0.3790], device='cuda:1'), in_proj_covar=tensor([0.1532, 0.1452, 0.1467, 0.1257], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 17:08:05,833 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:08:12,099 INFO [train.py:968] (1/2) Epoch 11, batch 14000, giga_loss[loss=0.2512, simple_loss=0.3365, pruned_loss=0.08292, over 28558.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3399, pruned_loss=0.09406, over 5656077.25 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3488, pruned_loss=0.09815, over 5742177.79 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3404, pruned_loss=0.09395, over 5659848.45 frames. ], batch size: 307, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:08:29,252 INFO [optim.py:369] (1/2) 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 17:08:46,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=469566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:08:50,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=469569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:09:15,273 INFO [train.py:968] (1/2) Epoch 11, batch 14050, giga_loss[loss=0.2418, simple_loss=0.3207, pruned_loss=0.08146, over 29128.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3415, pruned_loss=0.09409, over 5667611.65 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3483, pruned_loss=0.09792, over 5743593.57 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3423, pruned_loss=0.09415, over 5668329.20 frames. ], batch size: 200, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:09:19,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4692, 1.5224, 1.1666, 1.0981], device='cuda:1'), covar=tensor([0.0697, 0.0381, 0.0854, 0.1008], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0436, 0.0494, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 17:09:27,778 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:10:21,429 INFO [train.py:968] (1/2) Epoch 11, batch 14100, giga_loss[loss=0.2504, simple_loss=0.3267, pruned_loss=0.08702, over 28969.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3375, pruned_loss=0.09171, over 5673406.96 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3479, pruned_loss=0.09773, over 5746666.14 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3383, pruned_loss=0.09182, over 5669755.89 frames. ], batch size: 120, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:10:41,617 INFO [optim.py:369] (1/2) 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 17:10:42,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 17:11:27,643 INFO [train.py:968] (1/2) Epoch 11, batch 14150, giga_loss[loss=0.3076, simple_loss=0.3642, pruned_loss=0.1255, over 26841.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3387, pruned_loss=0.09282, over 5670852.32 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3475, pruned_loss=0.09755, over 5749985.36 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3395, pruned_loss=0.09299, over 5663592.59 frames. ], batch size: 555, lr: 2.97e-03, grad_scale: 4.0
+2023-03-05 17:11:46,470 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:12:06,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 17:12:20,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-05 17:12:27,888 INFO [train.py:968] (1/2) Epoch 11, batch 14200, giga_loss[loss=0.3014, simple_loss=0.3846, pruned_loss=0.1091, over 28444.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3425, pruned_loss=0.09343, over 5667422.59 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3474, pruned_loss=0.09751, over 5755417.83 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.343, pruned_loss=0.09346, over 5653461.79 frames. ], batch size: 369, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:12:41,791 INFO [optim.py:369] (1/2) 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 17:12:54,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6691, 1.9900, 1.7638, 1.6101], device='cuda:1'), covar=tensor([0.2160, 0.1445, 0.1233, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.1654, 0.1554, 0.1509, 0.1606], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 17:13:03,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:13:28,211 INFO [train.py:968] (1/2) Epoch 11, batch 14250, giga_loss[loss=0.2411, simple_loss=0.3415, pruned_loss=0.07029, over 28931.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3442, pruned_loss=0.09212, over 5662940.24 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3468, pruned_loss=0.09722, over 5749372.66 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.345, pruned_loss=0.09228, over 5654408.63 frames. ], batch size: 155, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:13:45,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3191, 1.7267, 1.6276, 1.1836], device='cuda:1'), covar=tensor([0.1769, 0.2387, 0.1458, 0.1764], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0689, 0.0863, 0.0770], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 17:13:50,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:14:30,734 INFO [train.py:968] (1/2) Epoch 11, batch 14300, giga_loss[loss=0.2698, simple_loss=0.3602, pruned_loss=0.08973, over 28931.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3434, pruned_loss=0.09081, over 5650150.18 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3463, pruned_loss=0.097, over 5750946.66 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3444, pruned_loss=0.09108, over 5641208.51 frames. ], batch size: 227, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:14:42,826 INFO [optim.py:369] (1/2) 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 17:14:47,436 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=469854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:14:55,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8161, 2.0242, 1.8358, 1.8112], device='cuda:1'), covar=tensor([0.1358, 0.2039, 0.1777, 0.1762], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0713, 0.0656, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 17:15:28,663 INFO [train.py:968] (1/2) Epoch 11, batch 14350, giga_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1228, over 26870.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.344, pruned_loss=0.09055, over 5657824.02 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3465, pruned_loss=0.09709, over 5747139.61 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3446, pruned_loss=0.09054, over 5652027.69 frames. ], batch size: 555, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:15:54,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=469907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:15:59,882 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=469911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:16:02,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=469914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:16:34,323 INFO [train.py:968] (1/2) Epoch 11, batch 14400, giga_loss[loss=0.2621, simple_loss=0.338, pruned_loss=0.09309, over 28921.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3455, pruned_loss=0.09272, over 5663096.17 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3465, pruned_loss=0.09709, over 5747015.92 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.346, pruned_loss=0.09269, over 5658285.53 frames. ], batch size: 227, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:16:39,771 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=469943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:16:47,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=469949.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:16:50,560 INFO [optim.py:369] (1/2) 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 17:16:51,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=469952.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:17:26,536 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=469981.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:17:35,775 INFO [train.py:968] (1/2) Epoch 11, batch 14450, giga_loss[loss=0.3056, simple_loss=0.3773, pruned_loss=0.117, over 28735.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3461, pruned_loss=0.09441, over 5665357.35 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3467, pruned_loss=0.09727, over 5752026.10 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3462, pruned_loss=0.09412, over 5654541.95 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:17:55,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:18:31,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-05 17:18:56,057 INFO [train.py:968] (1/2) Epoch 11, batch 14500, giga_loss[loss=0.2316, simple_loss=0.3211, pruned_loss=0.07109, over 28975.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3445, pruned_loss=0.0937, over 5673828.64 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3464, pruned_loss=0.09714, over 5750501.39 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3449, pruned_loss=0.09356, over 5665367.42 frames. ], batch size: 145, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:19:18,210 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:19:19,339 INFO [optim.py:369] (1/2) 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 17:19:21,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:19:49,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2678, 2.1966, 1.6775, 1.8474], device='cuda:1'), covar=tensor([0.0716, 0.0627, 0.0880, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0435, 0.0495, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 17:19:52,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:19:59,571 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 14550, giga_loss[loss=0.2465, simple_loss=0.3268, pruned_loss=0.08311, over 28996.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3397, pruned_loss=0.09093, over 5671712.15 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.0976, over 5754653.90 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3394, pruned_loss=0.09024, over 5658733.28 frames. ], batch size: 199, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:20:19,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1797, 1.5417, 1.3958, 1.0530], device='cuda:1'), covar=tensor([0.2075, 0.1451, 0.0939, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.1666, 0.1560, 0.1509, 0.1616], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 17:20:35,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2204, 1.9091, 1.5360, 0.3474], device='cuda:1'), covar=tensor([0.2858, 0.1968, 0.2808, 0.3845], device='cuda:1'), in_proj_covar=tensor([0.1532, 0.1448, 0.1470, 0.1251], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 17:21:14,694 INFO [train.py:968] (1/2) Epoch 11, batch 14600, giga_loss[loss=0.2212, simple_loss=0.306, pruned_loss=0.0682, over 28878.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3374, pruned_loss=0.08971, over 5674611.53 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3464, pruned_loss=0.09729, over 5758913.34 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3374, pruned_loss=0.08924, over 5658191.42 frames. ], batch size: 174, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:21:20,020 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:21:22,946 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:22:14,810 INFO [train.py:968] (1/2) Epoch 11, batch 14650, giga_loss[loss=0.2684, simple_loss=0.3534, pruned_loss=0.09172, over 28415.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3369, pruned_loss=0.08987, over 5678755.11 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3466, pruned_loss=0.09738, over 5753931.03 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3364, pruned_loss=0.08915, over 5666568.31 frames. ], batch size: 368, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:22:53,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:22:56,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:03,733 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 14700, giga_loss[loss=0.2578, simple_loss=0.324, pruned_loss=0.0958, over 24702.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3417, pruned_loss=0.09247, over 5681472.42 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09717, over 5756289.89 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3415, pruned_loss=0.09202, over 5668786.12 frames. ], batch size: 705, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:23:17,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2376, 1.2368, 1.1661, 0.8881], device='cuda:1'), covar=tensor([0.0818, 0.0508, 0.0985, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0436, 0.0494, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 17:23:22,213 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:22,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3331, 1.5352, 1.3486, 1.3702], device='cuda:1'), covar=tensor([0.1752, 0.1362, 0.1271, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.1658, 0.1549, 0.1505, 0.1607], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 17:23:27,338 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:32,783 INFO [optim.py:369] (1/2) 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 17:23:33,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:23:52,908 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 14750, giga_loss[loss=0.2343, simple_loss=0.3158, pruned_loss=0.07643, over 28363.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3402, pruned_loss=0.09267, over 5686920.84 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3458, pruned_loss=0.09698, over 5759445.76 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3404, pruned_loss=0.09241, over 5672603.91 frames. ], batch size: 71, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:24:37,151 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-05 17:24:54,016 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:25:25,560 INFO [train.py:968] (1/2) Epoch 11, batch 14800, giga_loss[loss=0.2574, simple_loss=0.3381, pruned_loss=0.08834, over 28861.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3424, pruned_loss=0.0956, over 5665638.94 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3459, pruned_loss=0.09704, over 5751018.52 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3425, pruned_loss=0.09532, over 5661979.19 frames. ], batch size: 174, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:25:42,157 INFO [optim.py:369] (1/2) 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 17:26:06,923 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:26:11,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:26:29,148 INFO [train.py:968] (1/2) Epoch 11, batch 14850, libri_loss[loss=0.326, simple_loss=0.3847, pruned_loss=0.1336, over 28498.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3418, pruned_loss=0.09489, over 5663777.09 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3461, pruned_loss=0.09722, over 5750761.43 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3416, pruned_loss=0.09447, over 5660734.33 frames. ], batch size: 106, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:26:29,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3510, 3.0314, 1.5170, 1.5039], device='cuda:1'), covar=tensor([0.0849, 0.0313, 0.0810, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0501, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0029, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 17:26:48,903 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:27:33,213 INFO [train.py:968] (1/2) Epoch 11, batch 14900, giga_loss[loss=0.2805, simple_loss=0.3549, pruned_loss=0.103, over 28751.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3438, pruned_loss=0.09502, over 5659178.36 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.346, pruned_loss=0.09728, over 5742557.56 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3437, pruned_loss=0.09461, over 5661893.13 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:27:53,051 INFO [optim.py:369] (1/2) 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 17:28:45,262 INFO [train.py:968] (1/2) Epoch 11, batch 14950, giga_loss[loss=0.2526, simple_loss=0.3365, pruned_loss=0.08436, over 28929.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3441, pruned_loss=0.09514, over 5667427.13 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3459, pruned_loss=0.09753, over 5747958.57 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.344, pruned_loss=0.09453, over 5661960.12 frames. ], batch size: 106, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:28:59,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:29:58,973 INFO [train.py:968] (1/2) Epoch 11, batch 15000, giga_loss[loss=0.2713, simple_loss=0.3396, pruned_loss=0.1015, over 29030.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3401, pruned_loss=0.09291, over 5683152.63 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3456, pruned_loss=0.09718, over 5752468.21 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3403, pruned_loss=0.09265, over 5672640.95 frames. ], batch size: 200, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:29:58,973 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 17:30:08,327 INFO [train.py:1012] (1/2) Epoch 11, validation: loss=0.2039, simple_loss=0.3039, pruned_loss=0.052, over 944034.00 frames. 
+2023-03-05 17:30:08,327 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 17:30:24,677 INFO [optim.py:369] (1/2) 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 17:31:12,219 INFO [train.py:968] (1/2) Epoch 11, batch 15050, giga_loss[loss=0.2156, simple_loss=0.2929, pruned_loss=0.06911, over 28972.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3352, pruned_loss=0.09114, over 5686031.31 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3457, pruned_loss=0.09728, over 5750147.59 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3351, pruned_loss=0.09074, over 5678719.90 frames. ], batch size: 186, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:31:48,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:31:55,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:32:15,150 INFO [train.py:968] (1/2) Epoch 11, batch 15100, giga_loss[loss=0.268, simple_loss=0.3466, pruned_loss=0.09471, over 28630.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.333, pruned_loss=0.09033, over 5673624.29 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3459, pruned_loss=0.09739, over 5740067.35 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3325, pruned_loss=0.08979, over 5676005.21 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:32:19,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470643.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:32:31,339 INFO [optim.py:369] (1/2) 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 17:32:57,527 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470673.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:33:14,063 INFO [train.py:968] (1/2) Epoch 11, batch 15150, giga_loss[loss=0.2312, simple_loss=0.3201, pruned_loss=0.07114, over 28994.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3334, pruned_loss=0.09089, over 5674215.14 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3455, pruned_loss=0.09719, over 5743450.59 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3332, pruned_loss=0.09052, over 5671716.14 frames. ], batch size: 155, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:33:16,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:34:03,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-05 17:34:09,175 INFO [train.py:968] (1/2) Epoch 11, batch 15200, giga_loss[loss=0.2273, simple_loss=0.311, pruned_loss=0.07185, over 28510.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3323, pruned_loss=0.08974, over 5673219.33 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3453, pruned_loss=0.09703, over 5748620.58 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.332, pruned_loss=0.08941, over 5664993.58 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:34:29,697 INFO [optim.py:369] (1/2) 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 17:34:38,183 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:34:42,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:34:46,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:12,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470786.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:35:15,502 INFO [train.py:968] (1/2) Epoch 11, batch 15250, giga_loss[loss=0.2923, simple_loss=0.3525, pruned_loss=0.116, over 26818.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3303, pruned_loss=0.08791, over 5668938.65 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3449, pruned_loss=0.09694, over 5750570.71 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3301, pruned_loss=0.0876, over 5659626.00 frames. ], batch size: 555, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:35:16,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470789.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:35:18,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3422, 4.1642, 3.9104, 1.8879], device='cuda:1'), covar=tensor([0.0497, 0.0698, 0.0798, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.1017, 0.0947, 0.0824, 0.0640], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 17:35:20,176 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:26,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:30,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=470801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:35:49,474 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=470818.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:36:10,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=470833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:36:14,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=470836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:36:17,223 INFO [train.py:968] (1/2) Epoch 11, batch 15300, giga_loss[loss=0.262, simple_loss=0.3386, pruned_loss=0.09272, over 28496.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3301, pruned_loss=0.08802, over 5656001.38 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3448, pruned_loss=0.09692, over 5744649.01 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3297, pruned_loss=0.08759, over 5652122.11 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:36:20,545 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:36:55,839 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=470873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:37:21,516 INFO [train.py:968] (1/2) Epoch 11, batch 15350, giga_loss[loss=0.2572, simple_loss=0.3425, pruned_loss=0.08596, over 29049.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3301, pruned_loss=0.08803, over 5671066.11 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3445, pruned_loss=0.09683, over 5747019.70 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3296, pruned_loss=0.08748, over 5662867.50 frames. ], batch size: 285, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:37:46,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2914, 1.6113, 1.2491, 1.0366], device='cuda:1'), covar=tensor([0.2490, 0.2370, 0.2740, 0.2103], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.0951, 0.1144, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 17:38:26,167 INFO [train.py:968] (1/2) Epoch 11, batch 15400, giga_loss[loss=0.2929, simple_loss=0.3626, pruned_loss=0.1116, over 28979.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3313, pruned_loss=0.08818, over 5685414.26 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3443, pruned_loss=0.09681, over 5749164.27 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3308, pruned_loss=0.08756, over 5675476.81 frames. ], batch size: 284, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:38:44,241 INFO [optim.py:369] (1/2) 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 17:38:46,587 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 15450, giga_loss[loss=0.2643, simple_loss=0.3418, pruned_loss=0.09337, over 28820.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3324, pruned_loss=0.08933, over 5692132.45 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3439, pruned_loss=0.09653, over 5748366.38 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.332, pruned_loss=0.08881, over 5682905.95 frames. ], batch size: 243, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:40:05,533 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:40:12,302 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 15500, giga_loss[loss=0.246, simple_loss=0.327, pruned_loss=0.0825, over 28469.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3321, pruned_loss=0.08948, over 5690786.85 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3441, pruned_loss=0.09679, over 5750890.95 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3314, pruned_loss=0.08871, over 5680179.30 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:40:43,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3384, 1.5658, 1.5665, 1.3996], device='cuda:1'), covar=tensor([0.1433, 0.1650, 0.1728, 0.1654], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0713, 0.0657, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 17:40:48,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:40:48,434 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:41:38,261 INFO [train.py:968] (1/2) Epoch 11, batch 15550, giga_loss[loss=0.2283, simple_loss=0.3125, pruned_loss=0.07204, over 27633.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3319, pruned_loss=0.08819, over 5675719.67 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3441, pruned_loss=0.09679, over 5750890.95 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3313, pruned_loss=0.08759, over 5667463.67 frames. ], batch size: 472, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:42:43,731 INFO [train.py:968] (1/2) Epoch 11, batch 15600, giga_loss[loss=0.2638, simple_loss=0.3448, pruned_loss=0.09144, over 28708.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3342, pruned_loss=0.08828, over 5668264.96 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.344, pruned_loss=0.09675, over 5751496.49 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3338, pruned_loss=0.08781, over 5660786.41 frames. ], batch size: 262, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:43:03,300 INFO [optim.py:369] (1/2) 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 17:43:31,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:43:40,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4372, 1.4896, 3.5716, 3.1938], device='cuda:1'), covar=tensor([0.1248, 0.2078, 0.0436, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0642, 0.0581, 0.0830, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 17:43:44,408 INFO [train.py:968] (1/2) Epoch 11, batch 15650, giga_loss[loss=0.236, simple_loss=0.3269, pruned_loss=0.07254, over 28904.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3359, pruned_loss=0.08919, over 5654904.65 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3438, pruned_loss=0.09658, over 5744769.92 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3356, pruned_loss=0.08882, over 5654191.07 frames. ], batch size: 227, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:43:47,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471191.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:43:49,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471194.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:44:15,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:44:24,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471223.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 17:44:42,000 INFO [train.py:968] (1/2) Epoch 11, batch 15700, giga_loss[loss=0.2711, simple_loss=0.342, pruned_loss=0.1001, over 28918.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.336, pruned_loss=0.08942, over 5653035.91 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3436, pruned_loss=0.09633, over 5746159.52 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3358, pruned_loss=0.08922, over 5649880.94 frames. ], batch size: 227, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:44:58,420 INFO [optim.py:369] (1/2) 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 17:45:15,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:45:37,754 INFO [train.py:968] (1/2) Epoch 11, batch 15750, giga_loss[loss=0.2585, simple_loss=0.3244, pruned_loss=0.09632, over 26905.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3345, pruned_loss=0.08913, over 5649460.18 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3429, pruned_loss=0.09611, over 5741705.91 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3347, pruned_loss=0.08891, over 5647249.96 frames. ], batch size: 555, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 17:46:14,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:46:18,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:46:32,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:46:43,169 INFO [train.py:968] (1/2) Epoch 11, batch 15800, giga_loss[loss=0.2681, simple_loss=0.3426, pruned_loss=0.09675, over 28064.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3327, pruned_loss=0.08811, over 5649147.57 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.343, pruned_loss=0.09618, over 5740974.15 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3326, pruned_loss=0.08777, over 5646754.03 frames. ], batch size: 412, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:47:00,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:47:03,427 INFO [optim.py:369] (1/2) 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 17:47:09,663 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:47:12,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:47:42,783 INFO [train.py:968] (1/2) Epoch 11, batch 15850, giga_loss[loss=0.2724, simple_loss=0.3381, pruned_loss=0.1034, over 27580.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3317, pruned_loss=0.08842, over 5654977.46 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3429, pruned_loss=0.09621, over 5734974.47 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3316, pruned_loss=0.08794, over 5656588.44 frames. ], batch size: 472, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:47:45,534 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:48:08,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3908, 1.0714, 4.2886, 3.5034], device='cuda:1'), covar=tensor([0.1534, 0.2656, 0.0405, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0584, 0.0838, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 17:48:23,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-05 17:48:37,755 INFO [train.py:968] (1/2) Epoch 11, batch 15900, giga_loss[loss=0.244, simple_loss=0.3305, pruned_loss=0.07873, over 28098.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3322, pruned_loss=0.08853, over 5666077.53 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3428, pruned_loss=0.09628, over 5740967.33 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3317, pruned_loss=0.08777, over 5659109.04 frames. ], batch size: 412, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:48:54,774 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-05 17:48:55,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:48:56,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3846, 1.6523, 1.6800, 1.2654], device='cuda:1'), covar=tensor([0.1611, 0.2341, 0.1366, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0683, 0.0856, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 17:48:57,065 INFO [optim.py:369] (1/2) 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 17:49:22,892 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:49:25,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:49:33,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5461, 3.3041, 2.5802, 2.0634], device='cuda:1'), covar=tensor([0.2056, 0.1016, 0.1264, 0.1740], device='cuda:1'), in_proj_covar=tensor([0.1664, 0.1546, 0.1487, 0.1612], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 17:49:36,260 INFO [train.py:968] (1/2) Epoch 11, batch 15950, giga_loss[loss=0.3228, simple_loss=0.3966, pruned_loss=0.1245, over 28656.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3348, pruned_loss=0.08962, over 5674620.88 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.09611, over 5743707.68 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3344, pruned_loss=0.089, over 5664915.90 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:49:57,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:50:05,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-05 17:50:43,396 INFO [train.py:968] (1/2) Epoch 11, batch 16000, giga_loss[loss=0.2656, simple_loss=0.3397, pruned_loss=0.09573, over 28873.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3352, pruned_loss=0.09068, over 5665690.16 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3421, pruned_loss=0.09591, over 5746376.58 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3351, pruned_loss=0.09022, over 5653954.67 frames. ], batch size: 174, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:50:56,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:51:03,378 INFO [optim.py:369] (1/2) 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 17:51:40,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-05 17:51:41,973 INFO [train.py:968] (1/2) Epoch 11, batch 16050, giga_loss[loss=0.3136, simple_loss=0.3901, pruned_loss=0.1186, over 28430.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3378, pruned_loss=0.09203, over 5667828.04 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3419, pruned_loss=0.09577, over 5749312.62 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09172, over 5654466.07 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:52:38,760 INFO [train.py:968] (1/2) Epoch 11, batch 16100, giga_loss[loss=0.2366, simple_loss=0.3309, pruned_loss=0.07119, over 28795.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3399, pruned_loss=0.09296, over 5658861.93 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3408, pruned_loss=0.09514, over 5752897.92 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3408, pruned_loss=0.0932, over 5642814.21 frames. ], batch size: 174, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:52:44,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:52:56,470 INFO [optim.py:369] (1/2) 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 17:53:38,445 INFO [train.py:968] (1/2) Epoch 11, batch 16150, giga_loss[loss=0.2896, simple_loss=0.3496, pruned_loss=0.1148, over 26983.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3413, pruned_loss=0.09305, over 5658096.99 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3408, pruned_loss=0.09504, over 5756667.32 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.342, pruned_loss=0.09329, over 5639973.27 frames. ], batch size: 555, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:54:46,196 INFO [train.py:968] (1/2) Epoch 11, batch 16200, giga_loss[loss=0.2705, simple_loss=0.3512, pruned_loss=0.09488, over 28402.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3395, pruned_loss=0.092, over 5663025.08 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3403, pruned_loss=0.09472, over 5760360.29 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3406, pruned_loss=0.09241, over 5643092.89 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:54:53,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3688, 1.6033, 1.3275, 1.5391], device='cuda:1'), covar=tensor([0.0723, 0.0303, 0.0321, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0112, 0.0115, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 17:55:08,960 INFO [optim.py:369] (1/2) 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 17:55:47,747 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:55:48,962 INFO [train.py:968] (1/2) Epoch 11, batch 16250, giga_loss[loss=0.2858, simple_loss=0.3597, pruned_loss=0.1059, over 28630.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.338, pruned_loss=0.09183, over 5656988.14 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3402, pruned_loss=0.09462, over 5751421.22 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3389, pruned_loss=0.09218, over 5648461.95 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:55:51,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:56:25,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=471819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:56:37,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:56:49,803 INFO [train.py:968] (1/2) Epoch 11, batch 16300, libri_loss[loss=0.2985, simple_loss=0.3702, pruned_loss=0.1134, over 29541.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3373, pruned_loss=0.09107, over 5677011.43 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.34, pruned_loss=0.09457, over 5759254.28 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.338, pruned_loss=0.09124, over 5659057.34 frames. ], batch size: 89, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:57:03,435 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:57:10,040 INFO [optim.py:369] (1/2) 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 17:57:23,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 17:57:24,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=471867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 17:57:50,672 INFO [train.py:968] (1/2) Epoch 11, batch 16350, giga_loss[loss=0.2308, simple_loss=0.3096, pruned_loss=0.07603, over 28982.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3383, pruned_loss=0.09315, over 5672745.23 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3397, pruned_loss=0.09448, over 5761800.01 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3391, pruned_loss=0.09332, over 5654276.59 frames. ], batch size: 120, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 17:58:05,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6509, 3.4959, 3.2459, 1.9881], device='cuda:1'), covar=tensor([0.0606, 0.0854, 0.0877, 0.1944], device='cuda:1'), in_proj_covar=tensor([0.1015, 0.0945, 0.0830, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 17:58:36,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=471924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:58:56,460 INFO [train.py:968] (1/2) Epoch 11, batch 16400, giga_loss[loss=0.3, simple_loss=0.3565, pruned_loss=0.1218, over 26972.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3358, pruned_loss=0.09235, over 5663925.15 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3397, pruned_loss=0.09452, over 5762942.94 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3364, pruned_loss=0.09243, over 5647498.85 frames. ], batch size: 555, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 17:59:16,644 INFO [optim.py:369] (1/2) 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 17:59:37,194 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=471972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:59:40,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=471975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 17:59:55,087 INFO [train.py:968] (1/2) Epoch 11, batch 16450, giga_loss[loss=0.2588, simple_loss=0.3436, pruned_loss=0.08694, over 28647.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3354, pruned_loss=0.09115, over 5668595.61 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3399, pruned_loss=0.09459, over 5763304.73 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3357, pruned_loss=0.09106, over 5652354.63 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:00:10,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-05 18:00:11,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:00:52,160 INFO [train.py:968] (1/2) Epoch 11, batch 16500, libri_loss[loss=0.3, simple_loss=0.3734, pruned_loss=0.1133, over 29141.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3337, pruned_loss=0.08897, over 5674640.96 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3397, pruned_loss=0.09455, over 5763547.92 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3339, pruned_loss=0.08876, over 5659159.31 frames. ], batch size: 101, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:01:12,310 INFO [optim.py:369] (1/2) 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 18:01:23,679 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 16550, giga_loss[loss=0.2754, simple_loss=0.3578, pruned_loss=0.0965, over 28902.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3349, pruned_loss=0.08698, over 5681581.55 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3394, pruned_loss=0.09446, over 5756302.30 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3352, pruned_loss=0.08679, over 5673517.70 frames. ], batch size: 186, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:02:02,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:02:05,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:02:30,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4867, 2.9071, 1.5639, 1.6758], device='cuda:1'), covar=tensor([0.0725, 0.0296, 0.0769, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0500, 0.0338, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0029, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 18:02:45,292 INFO [train.py:968] (1/2) Epoch 11, batch 16600, giga_loss[loss=0.2725, simple_loss=0.3529, pruned_loss=0.09607, over 28964.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3358, pruned_loss=0.08712, over 5680651.14 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3393, pruned_loss=0.09444, over 5759205.87 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.336, pruned_loss=0.08677, over 5669465.53 frames. ], batch size: 285, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:03:05,099 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:03:05,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 18:03:05,599 INFO [optim.py:369] (1/2) 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 18:03:40,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1450, 1.1129, 1.0499, 1.2851], device='cuda:1'), covar=tensor([0.0819, 0.0325, 0.0309, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 18:03:40,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1448, 1.5598, 1.4550, 1.0714], device='cuda:1'), covar=tensor([0.1484, 0.2221, 0.1224, 0.1472], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0683, 0.0857, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 18:03:48,019 INFO [train.py:968] (1/2) Epoch 11, batch 16650, giga_loss[loss=0.2581, simple_loss=0.3344, pruned_loss=0.09091, over 27586.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3371, pruned_loss=0.08831, over 5672242.27 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3392, pruned_loss=0.09435, over 5761648.89 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3373, pruned_loss=0.08801, over 5660168.65 frames. ], batch size: 472, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:03:56,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:04:27,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4547, 1.4988, 1.1749, 1.0978], device='cuda:1'), covar=tensor([0.0679, 0.0415, 0.0872, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0435, 0.0496, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:04:39,862 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:04:53,071 INFO [train.py:968] (1/2) Epoch 11, batch 16700, giga_loss[loss=0.2354, simple_loss=0.3224, pruned_loss=0.07422, over 28901.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3374, pruned_loss=0.08898, over 5661690.63 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.339, pruned_loss=0.09424, over 5755177.23 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3377, pruned_loss=0.0887, over 5655742.82 frames. ], batch size: 213, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:04:56,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472242.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 18:05:14,923 INFO [optim.py:369] (1/2) 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 18:06:00,312 INFO [train.py:968] (1/2) Epoch 11, batch 16750, giga_loss[loss=0.2383, simple_loss=0.3273, pruned_loss=0.07461, over 28454.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.337, pruned_loss=0.08832, over 5660640.09 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3384, pruned_loss=0.09381, over 5754432.42 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3377, pruned_loss=0.08833, over 5653914.85 frames. ], batch size: 369, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:06:51,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2239, 1.5981, 1.4672, 1.1637], device='cuda:1'), covar=tensor([0.1441, 0.2065, 0.1189, 0.1470], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0680, 0.0856, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 18:07:04,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-03-05 18:07:05,359 INFO [train.py:968] (1/2) Epoch 11, batch 16800, giga_loss[loss=0.2517, simple_loss=0.3378, pruned_loss=0.08278, over 28465.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3378, pruned_loss=0.08853, over 5662434.18 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3387, pruned_loss=0.09408, over 5759770.55 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3381, pruned_loss=0.08803, over 5648368.50 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:07:30,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2218, 4.0466, 3.7962, 2.0122], device='cuda:1'), covar=tensor([0.0507, 0.0660, 0.0709, 0.2275], device='cuda:1'), in_proj_covar=tensor([0.1017, 0.0944, 0.0832, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 18:07:30,762 INFO [optim.py:369] (1/2) 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 18:07:48,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:07:53,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:08:10,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472385.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 18:08:17,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472388.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 18:08:17,513 INFO [train.py:968] (1/2) Epoch 11, batch 16850, giga_loss[loss=0.2796, simple_loss=0.3649, pruned_loss=0.09708, over 28599.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3401, pruned_loss=0.08947, over 5664503.19 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3387, pruned_loss=0.09416, over 5758869.69 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3403, pruned_loss=0.0889, over 5652122.39 frames. ], batch size: 307, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:08:32,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:08:33,753 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:08:54,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 18:09:21,050 INFO [train.py:968] (1/2) Epoch 11, batch 16900, giga_loss[loss=0.2507, simple_loss=0.3394, pruned_loss=0.08098, over 28895.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3417, pruned_loss=0.08976, over 5671013.63 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3389, pruned_loss=0.09414, over 5761283.50 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3418, pruned_loss=0.0892, over 5656346.81 frames. ], batch size: 227, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:09:26,632 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:09:47,742 INFO [optim.py:369] (1/2) 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 18:09:59,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1715, 1.2533, 1.0349, 0.9945], device='cuda:1'), covar=tensor([0.0812, 0.0510, 0.1099, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0430, 0.0491, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:10:10,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:10:27,159 INFO [train.py:968] (1/2) Epoch 11, batch 16950, giga_loss[loss=0.2154, simple_loss=0.3041, pruned_loss=0.06332, over 29033.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3405, pruned_loss=0.08964, over 5681139.42 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3386, pruned_loss=0.09404, over 5763576.59 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3408, pruned_loss=0.08918, over 5665431.49 frames. ], batch size: 128, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:10:43,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-05 18:10:53,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3229, 1.9932, 1.4575, 0.4898], device='cuda:1'), covar=tensor([0.3100, 0.1602, 0.2467, 0.4059], device='cuda:1'), in_proj_covar=tensor([0.1529, 0.1464, 0.1486, 0.1261], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 18:11:21,650 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2184, 1.4037, 1.2098, 1.4008], device='cuda:1'), covar=tensor([0.0735, 0.0295, 0.0323, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0116, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0057, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 18:11:31,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:11:40,121 INFO [train.py:968] (1/2) Epoch 11, batch 17000, giga_loss[loss=0.2614, simple_loss=0.3456, pruned_loss=0.08854, over 28782.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3387, pruned_loss=0.08915, over 5683087.06 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.339, pruned_loss=0.09428, over 5765485.28 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3385, pruned_loss=0.08849, over 5667741.98 frames. ], batch size: 243, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:12:01,152 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:12:04,006 INFO [optim.py:369] (1/2) 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 18:12:13,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-05 18:12:19,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:12:24,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-05 18:12:50,054 INFO [train.py:968] (1/2) Epoch 11, batch 17050, giga_loss[loss=0.2704, simple_loss=0.3334, pruned_loss=0.1037, over 24551.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3362, pruned_loss=0.08701, over 5680141.74 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3388, pruned_loss=0.09415, over 5769502.59 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3363, pruned_loss=0.08642, over 5662166.18 frames. ], batch size: 705, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:13:27,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:13:30,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:13:51,989 INFO [train.py:968] (1/2) Epoch 11, batch 17100, giga_loss[loss=0.2477, simple_loss=0.3383, pruned_loss=0.07859, over 28492.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3362, pruned_loss=0.08711, over 5681317.24 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3383, pruned_loss=0.09394, over 5771641.81 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3367, pruned_loss=0.08674, over 5664050.34 frames. ], batch size: 336, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:14:05,976 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:14:37,706 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:14:40,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:14:52,614 INFO [train.py:968] (1/2) Epoch 11, batch 17150, giga_loss[loss=0.2723, simple_loss=0.3546, pruned_loss=0.09495, over 28881.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3382, pruned_loss=0.08839, over 5680149.49 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3386, pruned_loss=0.09422, over 5774731.11 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3382, pruned_loss=0.08772, over 5662098.66 frames. ], batch size: 284, lr: 2.96e-03, grad_scale: 4.0
+2023-03-05 18:14:59,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=472695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:14,415 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:22,219 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:25,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:15:51,309 INFO [train.py:968] (1/2) Epoch 11, batch 17200, giga_loss[loss=0.2304, simple_loss=0.2975, pruned_loss=0.08161, over 24596.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3404, pruned_loss=0.09002, over 5679455.11 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3383, pruned_loss=0.09409, over 5776456.61 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3408, pruned_loss=0.08955, over 5662394.22 frames. ], batch size: 705, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 18:15:58,211 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:16:13,150 INFO [optim.py:369] (1/2) 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 18:16:38,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:16:52,316 INFO [train.py:968] (1/2) Epoch 11, batch 17250, giga_loss[loss=0.2348, simple_loss=0.3159, pruned_loss=0.07684, over 28734.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.338, pruned_loss=0.09008, over 5671241.64 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3383, pruned_loss=0.09409, over 5776456.61 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3383, pruned_loss=0.08971, over 5657962.94 frames. ], batch size: 243, lr: 2.96e-03, grad_scale: 8.0
+2023-03-05 18:17:25,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:17:37,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1962, 1.1716, 3.8741, 3.1812], device='cuda:1'), covar=tensor([0.1607, 0.2721, 0.0410, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0648, 0.0582, 0.0834, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:17:46,592 INFO [train.py:968] (1/2) Epoch 11, batch 17300, giga_loss[loss=0.2534, simple_loss=0.3274, pruned_loss=0.08966, over 28645.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3376, pruned_loss=0.09092, over 5661181.74 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3379, pruned_loss=0.09402, over 5766197.89 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3382, pruned_loss=0.09057, over 5656957.16 frames. ], batch size: 242, lr: 2.96e-03, grad_scale: 2.0
+2023-03-05 18:18:09,851 INFO [optim.py:369] (1/2) 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 18:18:12,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-05 18:18:42,603 INFO [train.py:968] (1/2) Epoch 11, batch 17350, giga_loss[loss=0.2663, simple_loss=0.3439, pruned_loss=0.09431, over 27603.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3396, pruned_loss=0.09285, over 5647976.17 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3378, pruned_loss=0.09394, over 5761107.85 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3401, pruned_loss=0.09258, over 5646069.02 frames. ], batch size: 472, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:19:19,268 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:21,908 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:27,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=472929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:34,756 INFO [train.py:968] (1/2) Epoch 11, batch 17400, giga_loss[loss=0.3344, simple_loss=0.4062, pruned_loss=0.1313, over 28888.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3481, pruned_loss=0.09775, over 5663365.98 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3374, pruned_loss=0.09369, over 5764450.91 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.349, pruned_loss=0.09779, over 5656022.80 frames. ], batch size: 199, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:19:46,988 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:19:54,244 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=472961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:19:56,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=472964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:20:14,254 INFO [train.py:968] (1/2) Epoch 11, batch 17450, giga_loss[loss=0.2842, simple_loss=0.3622, pruned_loss=0.1031, over 28867.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3555, pruned_loss=0.1024, over 5666425.51 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3376, pruned_loss=0.09371, over 5758041.55 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3568, pruned_loss=0.1027, over 5662755.15 frames. ], batch size: 99, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:20:19,016 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=472993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:20:59,266 INFO [train.py:968] (1/2) Epoch 11, batch 17500, libri_loss[loss=0.2187, simple_loss=0.3041, pruned_loss=0.0666, over 29554.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3561, pruned_loss=0.1035, over 5667980.81 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3374, pruned_loss=0.09352, over 5759402.13 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3575, pruned_loss=0.104, over 5663043.49 frames. ], batch size: 77, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:21:19,847 INFO [optim.py:369] (1/2) 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 18:21:29,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=473070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:21:31,394 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:21:34,769 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 17550, libri_loss[loss=0.2153, simple_loss=0.2987, pruned_loss=0.06599, over 29560.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.35, pruned_loss=0.1008, over 5674993.49 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3373, pruned_loss=0.09346, over 5759282.23 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3516, pruned_loss=0.1015, over 5668178.39 frames. ], batch size: 75, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:21:58,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=473104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:22:29,167 INFO [train.py:968] (1/2) Epoch 11, batch 17600, giga_loss[loss=0.2473, simple_loss=0.319, pruned_loss=0.08778, over 28314.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3429, pruned_loss=0.09743, over 5684362.72 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3373, pruned_loss=0.09335, over 5757434.56 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3444, pruned_loss=0.09822, over 5678740.71 frames. ], batch size: 368, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:22:46,985 INFO [optim.py:369] (1/2) 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 18:22:56,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4020, 1.9815, 1.5173, 0.5682], device='cuda:1'), covar=tensor([0.3605, 0.2149, 0.3197, 0.4345], device='cuda:1'), in_proj_covar=tensor([0.1534, 0.1474, 0.1487, 0.1262], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 18:23:01,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 18:23:12,147 INFO [train.py:968] (1/2) Epoch 11, batch 17650, giga_loss[loss=0.2242, simple_loss=0.2991, pruned_loss=0.07469, over 29039.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.335, pruned_loss=0.09377, over 5695292.81 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3373, pruned_loss=0.09325, over 5761810.78 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3362, pruned_loss=0.09454, over 5685130.97 frames. ], batch size: 136, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:23:33,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:23:37,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=473216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:23:53,738 INFO [train.py:968] (1/2) Epoch 11, batch 17700, giga_loss[loss=0.2062, simple_loss=0.2806, pruned_loss=0.06593, over 29079.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3277, pruned_loss=0.09073, over 5696484.99 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3375, pruned_loss=0.09331, over 5764764.77 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3283, pruned_loss=0.09127, over 5684163.92 frames. ], batch size: 136, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:23:59,597 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 17750, giga_loss[loss=0.2104, simple_loss=0.2865, pruned_loss=0.06714, over 28811.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3214, pruned_loss=0.08787, over 5693643.53 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3374, pruned_loss=0.09322, over 5766408.08 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3219, pruned_loss=0.08832, over 5681865.22 frames. ], batch size: 243, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:24:38,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 18:24:45,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4010, 1.2972, 3.8907, 3.1619], device='cuda:1'), covar=tensor([0.1526, 0.2603, 0.0443, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0647, 0.0582, 0.0835, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:24:54,804 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:25:19,195 INFO [train.py:968] (1/2) Epoch 11, batch 17800, giga_loss[loss=0.2242, simple_loss=0.3006, pruned_loss=0.07386, over 29127.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3194, pruned_loss=0.08697, over 5697053.40 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3383, pruned_loss=0.09347, over 5764302.56 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3185, pruned_loss=0.08695, over 5688030.64 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:25:35,406 INFO [optim.py:369] (1/2) 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 18:25:35,627 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:26:04,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-05 18:26:04,230 INFO [train.py:968] (1/2) Epoch 11, batch 17850, giga_loss[loss=0.2215, simple_loss=0.2937, pruned_loss=0.07468, over 28848.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3163, pruned_loss=0.08555, over 5696021.13 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3383, pruned_loss=0.09347, over 5767407.45 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3152, pruned_loss=0.08542, over 5685131.63 frames. ], batch size: 186, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:26:18,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0877, 1.1477, 3.4265, 2.9253], device='cuda:1'), covar=tensor([0.1601, 0.2566, 0.0478, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0647, 0.0582, 0.0835, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:26:24,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5470, 1.9184, 1.7666, 1.3879], device='cuda:1'), covar=tensor([0.1731, 0.2332, 0.1407, 0.1665], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0689, 0.0865, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 18:26:49,039 INFO [train.py:968] (1/2) Epoch 11, batch 17900, giga_loss[loss=0.2108, simple_loss=0.2838, pruned_loss=0.06885, over 29042.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3125, pruned_loss=0.08379, over 5680815.82 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3384, pruned_loss=0.09347, over 5758610.30 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3112, pruned_loss=0.08356, over 5679743.54 frames. ], batch size: 106, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:27:05,988 INFO [optim.py:369] (1/2) 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 18:27:27,784 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:27:29,488 INFO [train.py:968] (1/2) Epoch 11, batch 17950, giga_loss[loss=0.2418, simple_loss=0.3171, pruned_loss=0.08324, over 29083.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.3096, pruned_loss=0.08237, over 5688224.93 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3382, pruned_loss=0.09339, over 5751227.96 frames. ], giga_tot_loss[loss=0.2361, simple_loss=0.3082, pruned_loss=0.08202, over 5692930.85 frames. ], batch size: 155, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:28:09,988 INFO [train.py:968] (1/2) Epoch 11, batch 18000, giga_loss[loss=0.212, simple_loss=0.2794, pruned_loss=0.07229, over 27673.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.308, pruned_loss=0.08161, over 5687948.65 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3381, pruned_loss=0.09325, over 5754221.25 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3058, pruned_loss=0.08099, over 5686658.38 frames. ], batch size: 472, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:28:09,988 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 18:28:16,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6068, 1.6316, 1.2608, 1.2874], device='cuda:1'), covar=tensor([0.0738, 0.0474, 0.0991, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0432, 0.0498, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:28:18,982 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 18:28:22,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2043, 3.0441, 2.8497, 1.3648], device='cuda:1'), covar=tensor([0.0957, 0.1033, 0.0922, 0.2408], device='cuda:1'), in_proj_covar=tensor([0.1011, 0.0941, 0.0825, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 18:28:38,855 INFO [optim.py:369] (1/2) 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 18:29:03,924 INFO [train.py:968] (1/2) Epoch 11, batch 18050, giga_loss[loss=0.2467, simple_loss=0.312, pruned_loss=0.09069, over 27752.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3052, pruned_loss=0.08011, over 5686045.58 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3381, pruned_loss=0.09307, over 5756161.95 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3025, pruned_loss=0.07943, over 5681248.41 frames. ], batch size: 474, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:29:46,276 INFO [train.py:968] (1/2) Epoch 11, batch 18100, giga_loss[loss=0.2264, simple_loss=0.2967, pruned_loss=0.07804, over 29080.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3031, pruned_loss=0.07856, over 5696051.66 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3384, pruned_loss=0.09308, over 5757626.03 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.2992, pruned_loss=0.07744, over 5688412.30 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:30:09,392 INFO [optim.py:369] (1/2) 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 18:30:16,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2073, 1.4297, 4.2255, 3.2753], device='cuda:1'), covar=tensor([0.1623, 0.2311, 0.0371, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0650, 0.0586, 0.0843, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:30:27,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=473684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:30:29,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3396, 1.5575, 1.2563, 1.4452], device='cuda:1'), covar=tensor([0.0788, 0.0323, 0.0336, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0087], device='cuda:1')
+2023-03-05 18:30:30,105 INFO [train.py:968] (1/2) Epoch 11, batch 18150, libri_loss[loss=0.2558, simple_loss=0.3404, pruned_loss=0.08561, over 29515.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3011, pruned_loss=0.07767, over 5700842.08 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3387, pruned_loss=0.09315, over 5755603.69 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.296, pruned_loss=0.076, over 5693024.26 frames. ], batch size: 84, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:30:50,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5095, 1.7696, 1.4512, 1.3782], device='cuda:1'), covar=tensor([0.2422, 0.2350, 0.2631, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.0959, 0.1151, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 18:31:09,447 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 18200, giga_loss[loss=0.272, simple_loss=0.3465, pruned_loss=0.09878, over 28877.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3033, pruned_loss=0.07956, over 5690069.96 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3389, pruned_loss=0.09327, over 5748476.34 frames. ], giga_tot_loss[loss=0.2273, simple_loss=0.2986, pruned_loss=0.07796, over 5689283.50 frames. ], batch size: 186, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:31:35,703 INFO [optim.py:369] (1/2) 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 18:31:36,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5015, 1.7462, 1.5339, 1.5899], device='cuda:1'), covar=tensor([0.1527, 0.1706, 0.1935, 0.1748], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0725, 0.0664, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 18:31:47,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2804, 1.5197, 1.2239, 1.1210], device='cuda:1'), covar=tensor([0.2275, 0.2309, 0.2531, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.1291, 0.0957, 0.1149, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 18:32:00,925 INFO [train.py:968] (1/2) Epoch 11, batch 18250, giga_loss[loss=0.2904, simple_loss=0.366, pruned_loss=0.1074, over 28673.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3145, pruned_loss=0.0855, over 5698661.23 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3391, pruned_loss=0.09342, over 5752055.42 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3092, pruned_loss=0.08361, over 5692528.16 frames. ], batch size: 242, lr: 2.95e-03, grad_scale: 1.0
+2023-03-05 18:32:16,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7056, 1.6276, 1.2332, 1.2783], device='cuda:1'), covar=tensor([0.0693, 0.0559, 0.1008, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0433, 0.0496, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:32:36,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:32:39,227 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=473836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:32:45,651 INFO [train.py:968] (1/2) Epoch 11, batch 18300, giga_loss[loss=0.279, simple_loss=0.353, pruned_loss=0.1025, over 28945.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3276, pruned_loss=0.09245, over 5698569.26 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3391, pruned_loss=0.09327, over 5754599.55 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.323, pruned_loss=0.09096, over 5690244.54 frames. ], batch size: 106, lr: 2.95e-03, grad_scale: 1.0
+2023-03-05 18:33:01,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=473859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:02,339 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=473861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:04,025 INFO [optim.py:369] (1/2) 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 18:33:16,749 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=473878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:20,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=473881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:33:26,281 INFO [train.py:968] (1/2) Epoch 11, batch 18350, libri_loss[loss=0.2456, simple_loss=0.328, pruned_loss=0.08162, over 29554.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3369, pruned_loss=0.09665, over 5704994.94 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3391, pruned_loss=0.09312, over 5759148.51 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3331, pruned_loss=0.09567, over 5692737.52 frames. ], batch size: 76, lr: 2.95e-03, grad_scale: 1.0
+2023-03-05 18:33:29,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5702, 5.2998, 4.9967, 2.2609], device='cuda:1'), covar=tensor([0.0362, 0.0515, 0.0580, 0.2054], device='cuda:1'), in_proj_covar=tensor([0.1013, 0.0949, 0.0829, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 18:33:43,070 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 18400, giga_loss[loss=0.2713, simple_loss=0.3468, pruned_loss=0.09796, over 28697.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3433, pruned_loss=0.09907, over 5695373.48 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3396, pruned_loss=0.09332, over 5751050.14 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3399, pruned_loss=0.09821, over 5691946.42 frames. ], batch size: 92, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:34:28,390 INFO [optim.py:369] (1/2) 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 18:34:50,925 INFO [train.py:968] (1/2) Epoch 11, batch 18450, giga_loss[loss=0.2727, simple_loss=0.3478, pruned_loss=0.09877, over 28837.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3461, pruned_loss=0.09903, over 5695679.87 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.34, pruned_loss=0.0935, over 5752165.13 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3431, pruned_loss=0.09827, over 5691242.57 frames. ], batch size: 119, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:35:06,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:35:09,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:35:35,381 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 18500, giga_loss[loss=0.271, simple_loss=0.3344, pruned_loss=0.1038, over 23615.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3475, pruned_loss=0.09952, over 5688248.51 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3403, pruned_loss=0.09364, over 5754477.78 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3449, pruned_loss=0.09887, over 5681663.16 frames. ], batch size: 705, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:35:58,661 INFO [optim.py:369] (1/2) 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 18:36:24,019 INFO [train.py:968] (1/2) Epoch 11, batch 18550, giga_loss[loss=0.2831, simple_loss=0.3597, pruned_loss=0.1033, over 28790.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3499, pruned_loss=0.1012, over 5690677.93 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3407, pruned_loss=0.09372, over 5752589.77 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3477, pruned_loss=0.1007, over 5686153.71 frames. ], batch size: 199, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:37:08,129 INFO [train.py:968] (1/2) Epoch 11, batch 18600, giga_loss[loss=0.2828, simple_loss=0.3536, pruned_loss=0.1061, over 29023.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3533, pruned_loss=0.1042, over 5694690.99 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3407, pruned_loss=0.09374, over 5756404.46 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3517, pruned_loss=0.104, over 5686463.78 frames. ], batch size: 155, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:37:26,184 INFO [optim.py:369] (1/2) 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 18:37:37,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9382, 1.9228, 1.3415, 1.5045], device='cuda:1'), covar=tensor([0.0789, 0.0654, 0.1054, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0433, 0.0496, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:37:40,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9419, 1.7355, 1.3512, 1.3313], device='cuda:1'), covar=tensor([0.0734, 0.0645, 0.0972, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0433, 0.0496, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:37:49,214 INFO [train.py:968] (1/2) Epoch 11, batch 18650, giga_loss[loss=0.2875, simple_loss=0.363, pruned_loss=0.106, over 28800.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3564, pruned_loss=0.1058, over 5700267.58 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3409, pruned_loss=0.09378, over 5759282.91 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3553, pruned_loss=0.1058, over 5689922.23 frames. ], batch size: 119, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:37:49,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:38:08,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:38:29,243 INFO [train.py:968] (1/2) Epoch 11, batch 18700, libri_loss[loss=0.3128, simple_loss=0.3783, pruned_loss=0.1237, over 19788.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.359, pruned_loss=0.106, over 5695889.89 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3416, pruned_loss=0.09394, over 5747524.12 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.358, pruned_loss=0.1062, over 5697375.27 frames. ], batch size: 186, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:38:30,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:38:50,659 INFO [optim.py:369] (1/2) 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 18:39:11,674 INFO [train.py:968] (1/2) Epoch 11, batch 18750, giga_loss[loss=0.2834, simple_loss=0.3661, pruned_loss=0.1003, over 28966.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3598, pruned_loss=0.1054, over 5699781.88 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3416, pruned_loss=0.09394, over 5750448.53 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3593, pruned_loss=0.1059, over 5697434.80 frames. ], batch size: 213, lr: 2.95e-03, grad_scale: 2.0
+2023-03-05 18:39:11,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:39:17,092 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:39:50,803 INFO [train.py:968] (1/2) Epoch 11, batch 18800, giga_loss[loss=0.2479, simple_loss=0.3333, pruned_loss=0.08126, over 28826.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3611, pruned_loss=0.1056, over 5691959.48 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3418, pruned_loss=0.09396, over 5744470.37 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.361, pruned_loss=0.1062, over 5694026.37 frames. ], batch size: 112, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:40:03,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:40:05,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:40:09,636 INFO [optim.py:369] (1/2) 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 18:40:23,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4096, 1.5413, 1.2872, 1.5891], device='cuda:1'), covar=tensor([0.2596, 0.2547, 0.2764, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.0953, 0.1143, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 18:40:29,631 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 18850, giga_loss[loss=0.288, simple_loss=0.3645, pruned_loss=0.1058, over 28857.00 frames. ], tot_loss[loss=0.284, simple_loss=0.36, pruned_loss=0.104, over 5693457.80 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3419, pruned_loss=0.09392, over 5746733.32 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.36, pruned_loss=0.1046, over 5692461.99 frames. ], batch size: 199, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:41:11,007 INFO [train.py:968] (1/2) Epoch 11, batch 18900, giga_loss[loss=0.2585, simple_loss=0.3371, pruned_loss=0.08994, over 29041.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3571, pruned_loss=0.1012, over 5705607.26 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3419, pruned_loss=0.09389, over 5749040.38 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3574, pruned_loss=0.1019, over 5701989.83 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:41:28,935 INFO [optim.py:369] (1/2) 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 18:41:40,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5288, 1.6216, 1.3586, 1.6444], device='cuda:1'), covar=tensor([0.2642, 0.2644, 0.2933, 0.2321], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.0953, 0.1142, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 18:41:48,359 INFO [train.py:968] (1/2) Epoch 11, batch 18950, giga_loss[loss=0.2829, simple_loss=0.3526, pruned_loss=0.1066, over 29017.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3585, pruned_loss=0.1024, over 5704007.67 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3432, pruned_loss=0.09461, over 5752896.29 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.358, pruned_loss=0.1025, over 5696459.58 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:41:48,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:42:31,295 INFO [train.py:968] (1/2) Epoch 11, batch 19000, giga_loss[loss=0.284, simple_loss=0.3692, pruned_loss=0.09942, over 28723.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3611, pruned_loss=0.1064, over 5678393.18 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3443, pruned_loss=0.09529, over 5737418.42 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3603, pruned_loss=0.1063, over 5683827.74 frames. ], batch size: 66, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:42:37,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7103, 1.8924, 1.9664, 1.5378], device='cuda:1'), covar=tensor([0.1538, 0.1970, 0.1203, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0693, 0.0866, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 18:42:54,965 INFO [optim.py:369] (1/2) 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 18:42:56,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:43:16,247 INFO [train.py:968] (1/2) Epoch 11, batch 19050, giga_loss[loss=0.3125, simple_loss=0.3693, pruned_loss=0.1278, over 28850.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3626, pruned_loss=0.1096, over 5682656.82 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3445, pruned_loss=0.09531, over 5739811.32 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3622, pruned_loss=0.1096, over 5683778.99 frames. ], batch size: 199, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:43:36,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6667, 4.4113, 4.2162, 2.1807], device='cuda:1'), covar=tensor([0.0486, 0.0627, 0.0611, 0.1935], device='cuda:1'), in_proj_covar=tensor([0.1001, 0.0941, 0.0824, 0.0642], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 18:43:37,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3879, 1.2166, 5.1757, 3.6183], device='cuda:1'), covar=tensor([0.1598, 0.2527, 0.0314, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0582, 0.0839, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:43:37,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:43:55,930 INFO [train.py:968] (1/2) Epoch 11, batch 19100, libri_loss[loss=0.2408, simple_loss=0.3121, pruned_loss=0.08469, over 29402.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3618, pruned_loss=0.1097, over 5692432.89 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3448, pruned_loss=0.09541, over 5743616.09 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3618, pruned_loss=0.1102, over 5687830.22 frames. ], batch size: 67, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:44:03,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4558, 1.4877, 1.4353, 1.2708], device='cuda:1'), covar=tensor([0.1840, 0.1751, 0.1431, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1543, 0.1518, 0.1627], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 18:44:14,200 INFO [optim.py:369] (1/2) 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 18:44:16,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:17,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-05 18:44:22,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=474672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:29,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=474678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:37,347 INFO [train.py:968] (1/2) Epoch 11, batch 19150, giga_loss[loss=0.3227, simple_loss=0.382, pruned_loss=0.1317, over 28878.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3607, pruned_loss=0.1097, over 5692574.88 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3454, pruned_loss=0.09577, over 5738442.55 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3606, pruned_loss=0.1102, over 5691855.55 frames. ], batch size: 174, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:44:53,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:44:55,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:14,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3177, 1.5977, 1.2982, 1.5523], device='cuda:1'), covar=tensor([0.0768, 0.0299, 0.0318, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0112, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0052, 0.0087], device='cuda:1')
+2023-03-05 18:45:21,842 INFO [train.py:968] (1/2) Epoch 11, batch 19200, giga_loss[loss=0.2766, simple_loss=0.3538, pruned_loss=0.0997, over 28771.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3593, pruned_loss=0.1088, over 5684953.49 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3452, pruned_loss=0.09556, over 5733900.69 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3597, pruned_loss=0.1097, over 5687032.74 frames. ], batch size: 242, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:45:22,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:40,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:42,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:45:43,210 INFO [optim.py:369] (1/2) 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 18:45:51,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-05 18:46:02,867 INFO [train.py:968] (1/2) Epoch 11, batch 19250, giga_loss[loss=0.2503, simple_loss=0.3343, pruned_loss=0.08313, over 28984.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3585, pruned_loss=0.1077, over 5687499.50 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3455, pruned_loss=0.09567, over 5736704.19 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3588, pruned_loss=0.1086, over 5685611.75 frames. ], batch size: 106, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:46:03,769 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:18,091 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:21,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:25,525 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=474815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:29,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=474818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:47,928 INFO [train.py:968] (1/2) Epoch 11, batch 19300, giga_loss[loss=0.2773, simple_loss=0.3328, pruned_loss=0.1108, over 23659.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3548, pruned_loss=0.105, over 5679840.53 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3458, pruned_loss=0.09586, over 5731640.65 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3551, pruned_loss=0.1058, over 5681058.12 frames. ], batch size: 705, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:46:48,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=474839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:46:53,276 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:47:08,895 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 19350, giga_loss[loss=0.2314, simple_loss=0.3107, pruned_loss=0.07607, over 28934.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3503, pruned_loss=0.102, over 5682249.46 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3464, pruned_loss=0.09596, over 5733415.73 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3502, pruned_loss=0.1027, over 5679525.68 frames. ], batch size: 174, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:48:11,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5766, 1.7111, 1.5160, 1.3654], device='cuda:1'), covar=tensor([0.1861, 0.1602, 0.1397, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.1681, 0.1555, 0.1520, 0.1642], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 18:48:12,770 INFO [train.py:968] (1/2) Epoch 11, batch 19400, libri_loss[loss=0.2898, simple_loss=0.3595, pruned_loss=0.11, over 29554.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3456, pruned_loss=0.09978, over 5688725.29 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3463, pruned_loss=0.09594, over 5740333.24 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3456, pruned_loss=0.1006, over 5678291.67 frames. ], batch size: 78, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:48:33,723 INFO [optim.py:369] (1/2) 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 18:48:44,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3838, 1.4829, 1.3133, 1.2212], device='cuda:1'), covar=tensor([0.1764, 0.1704, 0.1285, 0.1686], device='cuda:1'), in_proj_covar=tensor([0.1682, 0.1552, 0.1519, 0.1642], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 18:48:57,255 INFO [train.py:968] (1/2) Epoch 11, batch 19450, giga_loss[loss=0.225, simple_loss=0.3037, pruned_loss=0.07315, over 28721.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3409, pruned_loss=0.0973, over 5689203.77 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.347, pruned_loss=0.09618, over 5741482.72 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3402, pruned_loss=0.09782, over 5677589.37 frames. ], batch size: 284, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:49:13,616 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=475007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:49:15,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=475010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:49:23,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1838, 1.4179, 0.9635, 1.0334], device='cuda:1'), covar=tensor([0.1027, 0.0629, 0.1513, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0439, 0.0502, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:49:39,513 INFO [train.py:968] (1/2) Epoch 11, batch 19500, giga_loss[loss=0.2811, simple_loss=0.3527, pruned_loss=0.1047, over 28584.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3401, pruned_loss=0.09649, over 5696788.47 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3472, pruned_loss=0.09627, over 5746074.87 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.339, pruned_loss=0.09685, over 5680882.65 frames. ], batch size: 307, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:49:39,722 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=475039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:49:52,193 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:50:24,210 INFO [train.py:968] (1/2) Epoch 11, batch 19550, giga_loss[loss=0.2681, simple_loss=0.3453, pruned_loss=0.09551, over 28896.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3409, pruned_loss=0.09679, over 5704079.07 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3472, pruned_loss=0.09618, over 5746412.59 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.34, pruned_loss=0.09717, over 5690932.71 frames. ], batch size: 174, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:51:07,329 INFO [train.py:968] (1/2) Epoch 11, batch 19600, giga_loss[loss=0.2445, simple_loss=0.3188, pruned_loss=0.08513, over 28855.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3404, pruned_loss=0.09635, over 5700851.83 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3479, pruned_loss=0.09655, over 5740765.98 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3388, pruned_loss=0.09633, over 5693739.07 frames. ], batch size: 199, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:51:28,673 INFO [optim.py:369] (1/2) 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 18:51:49,848 INFO [train.py:968] (1/2) Epoch 11, batch 19650, giga_loss[loss=0.2467, simple_loss=0.3191, pruned_loss=0.0872, over 28824.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.338, pruned_loss=0.09528, over 5712253.13 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.348, pruned_loss=0.09648, over 5744188.29 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3365, pruned_loss=0.09531, over 5703001.74 frames. ], batch size: 145, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:51:55,020 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=475196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:51:57,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=475199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:52:02,119 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=475205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:52:09,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7211, 1.1316, 2.8185, 2.6836], device='cuda:1'), covar=tensor([0.1753, 0.2402, 0.0550, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0645, 0.0577, 0.0836, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 18:52:20,836 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 19700, giga_loss[loss=0.237, simple_loss=0.3142, pruned_loss=0.07995, over 29037.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3349, pruned_loss=0.09369, over 5716046.26 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3479, pruned_loss=0.09642, over 5740480.21 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3334, pruned_loss=0.09371, over 5710265.69 frames. ], batch size: 164, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:52:47,905 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 19750, giga_loss[loss=0.2358, simple_loss=0.3181, pruned_loss=0.07674, over 28717.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3334, pruned_loss=0.09298, over 5714188.91 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3486, pruned_loss=0.09664, over 5735522.66 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.331, pruned_loss=0.09267, over 5712529.28 frames. ], batch size: 284, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:53:30,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6195, 1.7261, 1.3684, 1.9445], device='cuda:1'), covar=tensor([0.2373, 0.2443, 0.2657, 0.2338], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.0951, 0.1142, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 18:53:44,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4723, 1.6688, 1.3290, 1.6285], device='cuda:1'), covar=tensor([0.2337, 0.2388, 0.2634, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.0953, 0.1143, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 18:53:46,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5430, 1.7529, 1.4210, 1.6197], device='cuda:1'), covar=tensor([0.2373, 0.2393, 0.2602, 0.2233], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.0952, 0.1143, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 18:53:48,991 INFO [train.py:968] (1/2) Epoch 11, batch 19800, giga_loss[loss=0.2454, simple_loss=0.321, pruned_loss=0.0849, over 28959.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.332, pruned_loss=0.09246, over 5714350.42 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.349, pruned_loss=0.0967, over 5733468.64 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3294, pruned_loss=0.09209, over 5714559.81 frames. ], batch size: 164, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:54:07,492 INFO [optim.py:369] (1/2) 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 18:54:28,666 INFO [train.py:968] (1/2) Epoch 11, batch 19850, giga_loss[loss=0.2392, simple_loss=0.3119, pruned_loss=0.08331, over 28929.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3289, pruned_loss=0.09111, over 5711907.01 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3495, pruned_loss=0.09692, over 5728146.98 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3263, pruned_loss=0.09058, over 5716739.28 frames. ], batch size: 106, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:55:02,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-05 18:55:09,238 INFO [train.py:968] (1/2) Epoch 11, batch 19900, giga_loss[loss=0.2508, simple_loss=0.3198, pruned_loss=0.0909, over 28164.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3284, pruned_loss=0.09123, over 5703898.33 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3502, pruned_loss=0.09736, over 5722852.19 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.325, pruned_loss=0.09023, over 5712264.45 frames. ], batch size: 77, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:55:28,356 INFO [optim.py:369] (1/2) 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 18:55:32,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-05 18:55:46,893 INFO [train.py:968] (1/2) Epoch 11, batch 19950, giga_loss[loss=0.2433, simple_loss=0.3064, pruned_loss=0.09009, over 28792.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3269, pruned_loss=0.0905, over 5710015.37 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3512, pruned_loss=0.09775, over 5721915.42 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.323, pruned_loss=0.08925, over 5717017.50 frames. ], batch size: 92, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:56:28,048 INFO [train.py:968] (1/2) Epoch 11, batch 20000, giga_loss[loss=0.2495, simple_loss=0.3248, pruned_loss=0.08711, over 28273.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3241, pruned_loss=0.08873, over 5718567.35 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3513, pruned_loss=0.09774, over 5722495.80 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3206, pruned_loss=0.0877, over 5723461.30 frames. ], batch size: 368, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:56:49,531 INFO [optim.py:369] (1/2) 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 18:57:01,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=475580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:57:08,416 INFO [train.py:968] (1/2) Epoch 11, batch 20050, giga_loss[loss=0.24, simple_loss=0.3117, pruned_loss=0.08418, over 29074.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3241, pruned_loss=0.08895, over 5716362.08 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3523, pruned_loss=0.0982, over 5715327.22 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3202, pruned_loss=0.08763, over 5726307.72 frames. ], batch size: 128, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:57:51,071 INFO [train.py:968] (1/2) Epoch 11, batch 20100, giga_loss[loss=0.2946, simple_loss=0.3462, pruned_loss=0.1215, over 24162.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3262, pruned_loss=0.09051, over 5717906.22 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3524, pruned_loss=0.0982, over 5717197.14 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3227, pruned_loss=0.08938, over 5723962.73 frames. ], batch size: 705, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:58:12,166 INFO [optim.py:369] (1/2) 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 18:58:37,205 INFO [train.py:968] (1/2) Epoch 11, batch 20150, giga_loss[loss=0.3222, simple_loss=0.3861, pruned_loss=0.1291, over 28324.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3327, pruned_loss=0.09464, over 5706517.73 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3528, pruned_loss=0.09839, over 5712846.43 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.329, pruned_loss=0.09343, over 5714634.09 frames. ], batch size: 368, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 18:58:44,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-05 18:58:55,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-05 18:59:10,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=475723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:59:13,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=475726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:59:17,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1520, 1.4585, 1.1789, 0.5503], device='cuda:1'), covar=tensor([0.1612, 0.1179, 0.1579, 0.2865], device='cuda:1'), in_proj_covar=tensor([0.1531, 0.1464, 0.1483, 0.1259], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 18:59:25,779 INFO [train.py:968] (1/2) Epoch 11, batch 20200, giga_loss[loss=0.3912, simple_loss=0.4293, pruned_loss=0.1765, over 27650.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3429, pruned_loss=0.1015, over 5694760.91 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3533, pruned_loss=0.09852, over 5718213.31 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.339, pruned_loss=0.1004, over 5695997.85 frames. ], batch size: 472, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 18:59:39,506 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=475755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 18:59:48,630 INFO [optim.py:369] (1/2) 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 19:00:00,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2981, 1.4759, 1.3690, 1.1351], device='cuda:1'), covar=tensor([0.1733, 0.1594, 0.0964, 0.1488], device='cuda:1'), in_proj_covar=tensor([0.1679, 0.1554, 0.1528, 0.1652], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:00:09,266 INFO [train.py:968] (1/2) Epoch 11, batch 20250, giga_loss[loss=0.3223, simple_loss=0.3925, pruned_loss=0.126, over 28627.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3486, pruned_loss=0.1048, over 5695400.81 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3533, pruned_loss=0.09864, over 5721603.83 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3453, pruned_loss=0.1039, over 5693035.18 frames. ], batch size: 307, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 19:00:16,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8956, 4.8292, 2.0732, 2.0809], device='cuda:1'), covar=tensor([0.0887, 0.0199, 0.0790, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0499, 0.0334, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0029, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 19:00:25,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=475806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:00:38,790 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6175, 1.8936, 1.6664, 1.5083], device='cuda:1'), covar=tensor([0.2051, 0.1630, 0.1651, 0.1848], device='cuda:1'), in_proj_covar=tensor([0.1675, 0.1550, 0.1524, 0.1645], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:00:58,986 INFO [train.py:968] (1/2) Epoch 11, batch 20300, giga_loss[loss=0.3386, simple_loss=0.3991, pruned_loss=0.1391, over 28864.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3524, pruned_loss=0.1059, over 5682122.79 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3534, pruned_loss=0.09867, over 5722556.81 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3498, pruned_loss=0.1052, over 5679394.66 frames. ], batch size: 186, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 19:01:19,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 19:01:22,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7353, 1.9074, 1.6387, 1.7426], device='cuda:1'), covar=tensor([0.1494, 0.2140, 0.1963, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0731, 0.0668, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 19:01:22,491 INFO [optim.py:369] (1/2) 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 19:01:46,324 INFO [train.py:968] (1/2) Epoch 11, batch 20350, giga_loss[loss=0.2739, simple_loss=0.3538, pruned_loss=0.09701, over 28431.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.357, pruned_loss=0.1082, over 5682706.76 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3534, pruned_loss=0.0987, over 5725726.41 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3549, pruned_loss=0.1077, over 5677079.22 frames. ], batch size: 71, lr: 2.95e-03, grad_scale: 4.0
+2023-03-05 19:01:47,386 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=475890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:02:28,330 INFO [train.py:968] (1/2) Epoch 11, batch 20400, libri_loss[loss=0.2162, simple_loss=0.2979, pruned_loss=0.06727, over 28471.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3616, pruned_loss=0.1109, over 5676923.88 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3532, pruned_loss=0.09839, over 5727014.21 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3603, pruned_loss=0.1111, over 5670273.53 frames. ], batch size: 63, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:02:48,240 INFO [optim.py:369] (1/2) 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 19:03:02,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2609, 3.0640, 2.9190, 1.4860], device='cuda:1'), covar=tensor([0.0868, 0.0973, 0.0812, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1019, 0.0952, 0.0833, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 19:03:04,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 19:03:08,858 INFO [train.py:968] (1/2) Epoch 11, batch 20450, giga_loss[loss=0.2216, simple_loss=0.3044, pruned_loss=0.06939, over 28595.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3573, pruned_loss=0.1077, over 5677309.81 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.353, pruned_loss=0.09832, over 5724282.99 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3567, pruned_loss=0.1084, over 5672762.63 frames. ], batch size: 262, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:03:34,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2146, 1.3256, 1.1100, 1.0218], device='cuda:1'), covar=tensor([0.0861, 0.0429, 0.0998, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0439, 0.0505, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 19:03:49,980 INFO [train.py:968] (1/2) Epoch 11, batch 20500, giga_loss[loss=0.2795, simple_loss=0.3558, pruned_loss=0.1016, over 28879.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3545, pruned_loss=0.1052, over 5690063.06 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3529, pruned_loss=0.09824, over 5728867.54 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3542, pruned_loss=0.106, over 5680976.81 frames. ], batch size: 119, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:04:14,048 INFO [optim.py:369] (1/2) 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 19:04:18,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-05 19:04:34,458 INFO [train.py:968] (1/2) Epoch 11, batch 20550, giga_loss[loss=0.2533, simple_loss=0.3345, pruned_loss=0.08606, over 28709.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3529, pruned_loss=0.1035, over 5695648.08 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3527, pruned_loss=0.09812, over 5731241.44 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3527, pruned_loss=0.1043, over 5685960.38 frames. ], batch size: 85, lr: 2.95e-03, grad_scale: 8.0
+2023-03-05 19:04:43,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1525, 3.9769, 3.7448, 1.6795], device='cuda:1'), covar=tensor([0.0563, 0.0678, 0.0649, 0.2141], device='cuda:1'), in_proj_covar=tensor([0.1016, 0.0947, 0.0826, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 19:05:07,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2633, 1.3887, 1.2312, 1.1278], device='cuda:1'), covar=tensor([0.1975, 0.1834, 0.1304, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.1688, 0.1560, 0.1536, 0.1651], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:05:12,199 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7171, 1.8785, 1.5453, 1.9637], device='cuda:1'), covar=tensor([0.2375, 0.2242, 0.2505, 0.2057], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.0957, 0.1146, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 19:05:14,808 INFO [train.py:968] (1/2) Epoch 11, batch 20600, giga_loss[loss=0.2951, simple_loss=0.3626, pruned_loss=0.1138, over 28518.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3544, pruned_loss=0.1039, over 5682234.34 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3532, pruned_loss=0.09835, over 5721934.33 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3539, pruned_loss=0.1045, over 5682214.78 frames. ], batch size: 71, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:05:39,145 INFO [optim.py:369] (1/2) 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 19:05:52,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:05:54,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9470, 1.2075, 3.4227, 2.9533], device='cuda:1'), covar=tensor([0.1692, 0.2490, 0.0473, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0650, 0.0579, 0.0839, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 19:05:59,415 INFO [train.py:968] (1/2) Epoch 11, batch 20650, giga_loss[loss=0.2841, simple_loss=0.3583, pruned_loss=0.1049, over 28777.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3568, pruned_loss=0.1058, over 5679657.17 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3535, pruned_loss=0.09857, over 5714572.39 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3562, pruned_loss=0.1062, over 5686275.95 frames. ], batch size: 99, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:06:35,037 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 20700, giga_loss[loss=0.2565, simple_loss=0.335, pruned_loss=0.08902, over 28355.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3581, pruned_loss=0.1065, over 5697308.77 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3536, pruned_loss=0.09838, over 5721294.92 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3576, pruned_loss=0.1072, over 5695609.63 frames. ], batch size: 77, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:06:44,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1611, 1.4134, 1.4152, 1.0613], device='cuda:1'), covar=tensor([0.1432, 0.2280, 0.1182, 0.1404], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0687, 0.0864, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 19:07:06,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9589, 4.7736, 4.4952, 2.2099], device='cuda:1'), covar=tensor([0.0411, 0.0556, 0.0565, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.1023, 0.0958, 0.0836, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 19:07:09,419 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:07:10,451 INFO [optim.py:369] (1/2) 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 19:07:13,719 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 20750, giga_loss[loss=0.2726, simple_loss=0.3469, pruned_loss=0.09914, over 28355.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3596, pruned_loss=0.1082, over 5680260.24 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3538, pruned_loss=0.09846, over 5721899.44 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3591, pruned_loss=0.1087, over 5678160.50 frames. ], batch size: 78, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:08:05,490 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:08:08,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:08:16,755 INFO [train.py:968] (1/2) Epoch 11, batch 20800, giga_loss[loss=0.2753, simple_loss=0.3457, pruned_loss=0.1025, over 28470.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3605, pruned_loss=0.1094, over 5686068.04 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3537, pruned_loss=0.09834, over 5723861.93 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3602, pruned_loss=0.1101, over 5682248.41 frames. ], batch size: 71, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:08:29,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8535, 1.9571, 1.8218, 1.8348], device='cuda:1'), covar=tensor([0.1333, 0.1609, 0.1830, 0.1495], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0730, 0.0671, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 19:08:31,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=476356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:08:39,923 INFO [optim.py:369] (1/2) 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 19:08:55,015 INFO [train.py:968] (1/2) Epoch 11, batch 20850, giga_loss[loss=0.2887, simple_loss=0.3667, pruned_loss=0.1054, over 28853.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3611, pruned_loss=0.1095, over 5689955.76 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3542, pruned_loss=0.09856, over 5719917.24 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3606, pruned_loss=0.1101, over 5690037.97 frames. ], batch size: 66, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:09:00,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4075, 1.6702, 1.3129, 1.4825], device='cuda:1'), covar=tensor([0.2309, 0.2180, 0.2460, 0.2058], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.0956, 0.1142, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 19:09:10,230 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:09:12,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:09:36,324 INFO [train.py:968] (1/2) Epoch 11, batch 20900, giga_loss[loss=0.2896, simple_loss=0.3662, pruned_loss=0.1065, over 28881.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3613, pruned_loss=0.1089, over 5686343.78 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.355, pruned_loss=0.09904, over 5714147.80 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3604, pruned_loss=0.1091, over 5690334.86 frames. ], batch size: 227, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:09:37,405 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=476440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:09:58,829 INFO [optim.py:369] (1/2) 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 19:10:08,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1839, 3.0111, 2.7992, 1.4442], device='cuda:1'), covar=tensor([0.0865, 0.0997, 0.0879, 0.2498], device='cuda:1'), in_proj_covar=tensor([0.1023, 0.0959, 0.0840, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 19:10:16,042 INFO [train.py:968] (1/2) Epoch 11, batch 20950, libri_loss[loss=0.2565, simple_loss=0.3387, pruned_loss=0.08714, over 29592.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3616, pruned_loss=0.1077, over 5695726.04 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3549, pruned_loss=0.09891, over 5719020.61 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3611, pruned_loss=0.1082, over 5693858.95 frames. ], batch size: 75, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:10:25,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 21000, libri_loss[loss=0.233, simple_loss=0.3109, pruned_loss=0.07755, over 29484.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3612, pruned_loss=0.1071, over 5691717.03 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3554, pruned_loss=0.0993, over 5715443.28 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3606, pruned_loss=0.1075, over 5692742.94 frames. ], batch size: 70, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:10:54,531 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 19:11:02,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3928, 1.7600, 1.3775, 1.3630], device='cuda:1'), covar=tensor([0.2717, 0.2509, 0.2613, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.0958, 0.1142, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 19:11:03,128 INFO [train.py:1012] (1/2) Epoch 11, validation: loss=0.2182, simple_loss=0.3234, pruned_loss=0.05645, over 944034.00 frames. 
+2023-03-05 19:11:03,129 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 19:11:07,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-05 19:11:15,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 19:11:20,016 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-05 19:11:24,398 INFO [optim.py:369] (1/2) 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 19:11:39,943 INFO [train.py:968] (1/2) Epoch 11, batch 21050, giga_loss[loss=0.2662, simple_loss=0.3457, pruned_loss=0.09333, over 28603.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3583, pruned_loss=0.1054, over 5704719.12 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3554, pruned_loss=0.09937, over 5719561.96 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.358, pruned_loss=0.1059, over 5701580.13 frames. ], batch size: 85, lr: 2.94e-03, grad_scale: 2.0
+2023-03-05 19:11:49,816 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 21100, giga_loss[loss=0.2665, simple_loss=0.3431, pruned_loss=0.09492, over 28939.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3557, pruned_loss=0.104, over 5709962.83 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3558, pruned_loss=0.09963, over 5722104.95 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.355, pruned_loss=0.1042, over 5704817.12 frames. ], batch size: 227, lr: 2.94e-03, grad_scale: 2.0
+2023-03-05 19:12:26,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:12:44,511 INFO [optim.py:369] (1/2) 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 19:12:46,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7038, 1.9969, 1.8063, 1.5408], device='cuda:1'), covar=tensor([0.2125, 0.1679, 0.1701, 0.1873], device='cuda:1'), in_proj_covar=tensor([0.1691, 0.1579, 0.1550, 0.1663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:13:00,404 INFO [train.py:968] (1/2) Epoch 11, batch 21150, giga_loss[loss=0.3664, simple_loss=0.4126, pruned_loss=0.1601, over 28692.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3546, pruned_loss=0.1038, over 5711600.93 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3562, pruned_loss=0.09979, over 5725238.78 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3537, pruned_loss=0.1038, over 5704459.29 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 2.0
+2023-03-05 19:13:42,897 INFO [train.py:968] (1/2) Epoch 11, batch 21200, giga_loss[loss=0.2982, simple_loss=0.364, pruned_loss=0.1162, over 28849.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.356, pruned_loss=0.1054, over 5711917.16 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3566, pruned_loss=0.1001, over 5727913.10 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3548, pruned_loss=0.1053, over 5703390.53 frames. ], batch size: 112, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:13:49,225 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:13:51,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:10,650 INFO [optim.py:369] (1/2) 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 19:14:16,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=476777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:25,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=476788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:26,168 INFO [train.py:968] (1/2) Epoch 11, batch 21250, giga_loss[loss=0.2728, simple_loss=0.3526, pruned_loss=0.09643, over 28883.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3564, pruned_loss=0.1054, over 5716239.86 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3569, pruned_loss=0.1003, over 5730562.06 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3552, pruned_loss=0.1051, over 5706859.39 frames. ], batch size: 199, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:14:27,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=476791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:14:40,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6085, 1.6851, 1.8477, 1.4201], device='cuda:1'), covar=tensor([0.1667, 0.2212, 0.1319, 0.1564], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0690, 0.0865, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 19:14:48,776 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 21300, giga_loss[loss=0.2951, simple_loss=0.366, pruned_loss=0.1121, over 28838.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3553, pruned_loss=0.1045, over 5709574.03 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3573, pruned_loss=0.101, over 5730037.56 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3539, pruned_loss=0.1039, over 5701704.90 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:15:28,330 INFO [optim.py:369] (1/2) 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 19:15:34,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=476876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:15:44,292 INFO [train.py:968] (1/2) Epoch 11, batch 21350, giga_loss[loss=0.2698, simple_loss=0.3492, pruned_loss=0.09523, over 28601.00 frames. ], tot_loss[loss=0.28, simple_loss=0.354, pruned_loss=0.103, over 5722361.46 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3568, pruned_loss=0.101, over 5734602.21 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3533, pruned_loss=0.1026, over 5711263.12 frames. ], batch size: 307, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:15:51,764 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-05 19:16:22,597 INFO [train.py:968] (1/2) Epoch 11, batch 21400, giga_loss[loss=0.2706, simple_loss=0.3461, pruned_loss=0.09754, over 28749.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3542, pruned_loss=0.1033, over 5727758.66 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3573, pruned_loss=0.1014, over 5736342.14 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3532, pruned_loss=0.1026, over 5717416.45 frames. ], batch size: 92, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:16:47,070 INFO [optim.py:369] (1/2) 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 19:16:47,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=476970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:17:01,257 INFO [train.py:968] (1/2) Epoch 11, batch 21450, giga_loss[loss=0.273, simple_loss=0.3461, pruned_loss=0.09998, over 28401.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3525, pruned_loss=0.1027, over 5732903.57 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3579, pruned_loss=0.1019, over 5741708.41 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.351, pruned_loss=0.1018, over 5719059.83 frames. ], batch size: 65, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:17:24,987 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=477019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:17:27,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=477022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:17:40,079 INFO [train.py:968] (1/2) Epoch 11, batch 21500, giga_loss[loss=0.2716, simple_loss=0.3372, pruned_loss=0.103, over 28903.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3491, pruned_loss=0.101, over 5728983.13 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3578, pruned_loss=0.1022, over 5744692.34 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3478, pruned_loss=0.09995, over 5714716.71 frames. ], batch size: 99, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:17:50,759 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=477051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:18:03,391 INFO [optim.py:369] (1/2) 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 19:18:21,409 INFO [train.py:968] (1/2) Epoch 11, batch 21550, giga_loss[loss=0.2798, simple_loss=0.3483, pruned_loss=0.1057, over 28836.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.348, pruned_loss=0.1002, over 5730665.42 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3578, pruned_loss=0.1022, over 5745644.88 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3469, pruned_loss=0.09945, over 5718626.13 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:18:54,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-05 19:19:00,163 INFO [train.py:968] (1/2) Epoch 11, batch 21600, giga_loss[loss=0.2391, simple_loss=0.3248, pruned_loss=0.07671, over 28972.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3484, pruned_loss=0.1013, over 5715125.43 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3586, pruned_loss=0.1029, over 5734783.62 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3466, pruned_loss=0.09997, over 5714786.49 frames. ], batch size: 174, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:19:08,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4825, 1.7696, 1.4558, 1.2901], device='cuda:1'), covar=tensor([0.2396, 0.2246, 0.2579, 0.2206], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.0958, 0.1142, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 19:19:24,137 INFO [optim.py:369] (1/2) 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 19:19:36,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477182.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:19:40,778 INFO [train.py:968] (1/2) Epoch 11, batch 21650, giga_loss[loss=0.2763, simple_loss=0.3385, pruned_loss=0.107, over 28458.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3474, pruned_loss=0.1018, over 5706196.11 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3594, pruned_loss=0.1036, over 5724299.73 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3452, pruned_loss=0.1002, over 5715218.61 frames. ], batch size: 85, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:20:20,840 INFO [train.py:968] (1/2) Epoch 11, batch 21700, giga_loss[loss=0.2437, simple_loss=0.3191, pruned_loss=0.08416, over 28952.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3453, pruned_loss=0.1009, over 5709594.70 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3598, pruned_loss=0.1039, over 5726406.11 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3427, pruned_loss=0.09916, over 5714153.77 frames. ], batch size: 106, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:20:45,443 INFO [optim.py:369] (1/2) 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 19:21:00,951 INFO [train.py:968] (1/2) Epoch 11, batch 21750, giga_loss[loss=0.2419, simple_loss=0.3179, pruned_loss=0.08295, over 28934.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3418, pruned_loss=0.09925, over 5701043.70 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3597, pruned_loss=0.1039, over 5719142.54 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3396, pruned_loss=0.09786, over 5710475.30 frames. ], batch size: 199, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:21:02,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-05 19:21:36,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0586, 2.3950, 2.4227, 1.8938], device='cuda:1'), covar=tensor([0.1551, 0.1775, 0.1175, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0694, 0.0865, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 19:21:41,042 INFO [train.py:968] (1/2) Epoch 11, batch 21800, giga_loss[loss=0.2397, simple_loss=0.3203, pruned_loss=0.07957, over 28953.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3402, pruned_loss=0.09867, over 5702233.69 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3601, pruned_loss=0.1043, over 5722467.90 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3377, pruned_loss=0.09709, over 5706310.93 frames. ], batch size: 213, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:21:45,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=477345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:22:05,029 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 21850, giga_loss[loss=0.2875, simple_loss=0.3681, pruned_loss=0.1035, over 28606.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3411, pruned_loss=0.09888, over 5700412.99 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3605, pruned_loss=0.1048, over 5724136.09 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3382, pruned_loss=0.09704, over 5701447.97 frames. ], batch size: 336, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:23:03,064 INFO [train.py:968] (1/2) Epoch 11, batch 21900, giga_loss[loss=0.2645, simple_loss=0.3501, pruned_loss=0.08947, over 28949.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3438, pruned_loss=0.1001, over 5706032.99 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3607, pruned_loss=0.1052, over 5728247.92 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3408, pruned_loss=0.09812, over 5702653.68 frames. ], batch size: 227, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:23:29,595 INFO [optim.py:369] (1/2) 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 19:23:40,359 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-05 19:23:44,436 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=477488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:23:44,847 INFO [train.py:968] (1/2) Epoch 11, batch 21950, giga_loss[loss=0.252, simple_loss=0.3331, pruned_loss=0.08545, over 28278.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3471, pruned_loss=0.1015, over 5711357.54 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3615, pruned_loss=0.1058, over 5729128.21 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3436, pruned_loss=0.09919, over 5707547.20 frames. ], batch size: 77, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:23:47,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=477491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:24:05,997 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:24:12,861 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 22000, giga_loss[loss=0.2863, simple_loss=0.3608, pruned_loss=0.1059, over 28945.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3499, pruned_loss=0.1026, over 5705868.97 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.362, pruned_loss=0.1065, over 5730880.45 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3465, pruned_loss=0.1001, over 5700897.87 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:24:41,979 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:25:05,814 INFO [train.py:968] (1/2) Epoch 11, batch 22050, giga_loss[loss=0.2347, simple_loss=0.3203, pruned_loss=0.07456, over 28840.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3492, pruned_loss=0.1021, over 5702294.78 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3621, pruned_loss=0.107, over 5736425.13 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3459, pruned_loss=0.0994, over 5692186.32 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:25:46,894 INFO [train.py:968] (1/2) Epoch 11, batch 22100, giga_loss[loss=0.2691, simple_loss=0.3376, pruned_loss=0.1003, over 28658.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3481, pruned_loss=0.1012, over 5702475.40 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.362, pruned_loss=0.1073, over 5735757.90 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.345, pruned_loss=0.09852, over 5693567.40 frames. ], batch size: 85, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:26:11,355 INFO [optim.py:369] (1/2) 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 19:26:25,712 INFO [train.py:968] (1/2) Epoch 11, batch 22150, libri_loss[loss=0.2945, simple_loss=0.3721, pruned_loss=0.1085, over 29488.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3499, pruned_loss=0.1025, over 5707362.13 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3626, pruned_loss=0.1079, over 5737749.89 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3464, pruned_loss=0.09961, over 5697243.48 frames. ], batch size: 85, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:26:34,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5649, 1.7354, 1.6091, 1.3595], device='cuda:1'), covar=tensor([0.2636, 0.1971, 0.1546, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1690, 0.1568, 0.1556, 0.1658], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:26:35,233 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=477700.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:26:38,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=477703.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 19:27:02,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=477732.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 19:27:10,154 INFO [train.py:968] (1/2) Epoch 11, batch 22200, giga_loss[loss=0.331, simple_loss=0.3839, pruned_loss=0.1391, over 26585.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3513, pruned_loss=0.1039, over 5708662.82 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3627, pruned_loss=0.1081, over 5740141.71 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3483, pruned_loss=0.1014, over 5698284.22 frames. ], batch size: 555, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:27:36,017 INFO [optim.py:369] (1/2) 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 19:27:51,221 INFO [train.py:968] (1/2) Epoch 11, batch 22250, giga_loss[loss=0.2888, simple_loss=0.3611, pruned_loss=0.1082, over 28830.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3532, pruned_loss=0.1045, over 5711089.71 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3629, pruned_loss=0.1081, over 5744304.78 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3503, pruned_loss=0.1024, over 5698241.23 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:28:32,150 INFO [train.py:968] (1/2) Epoch 11, batch 22300, giga_loss[loss=0.2953, simple_loss=0.3632, pruned_loss=0.1137, over 28920.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3571, pruned_loss=0.1071, over 5714768.89 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3637, pruned_loss=0.1088, over 5745425.95 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.354, pruned_loss=0.1048, over 5702729.42 frames. ], batch size: 106, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:28:44,066 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:29:10,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=477888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:29:11,326 INFO [train.py:968] (1/2) Epoch 11, batch 22350, giga_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.08685, over 28701.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3593, pruned_loss=0.1083, over 5706912.54 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3644, pruned_loss=0.1094, over 5739365.97 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.356, pruned_loss=0.1059, over 5702233.25 frames. ], batch size: 92, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:29:19,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3716, 1.8831, 1.3758, 0.7056], device='cuda:1'), covar=tensor([0.3501, 0.1972, 0.2374, 0.3899], device='cuda:1'), in_proj_covar=tensor([0.1542, 0.1458, 0.1485, 0.1263], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 19:29:38,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=477923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:29:49,255 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 22400, giga_loss[loss=0.2926, simple_loss=0.3551, pruned_loss=0.115, over 23637.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3595, pruned_loss=0.1081, over 5708991.14 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.365, pruned_loss=0.1099, over 5740149.64 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3563, pruned_loss=0.1056, over 5704228.04 frames. ], batch size: 705, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:30:20,111 INFO [optim.py:369] (1/2) 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 19:30:35,748 INFO [train.py:968] (1/2) Epoch 11, batch 22450, giga_loss[loss=0.2577, simple_loss=0.3411, pruned_loss=0.08719, over 28725.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3601, pruned_loss=0.1084, over 5712933.90 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3648, pruned_loss=0.1099, over 5741738.48 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3577, pruned_loss=0.1066, over 5707448.04 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:30:59,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-05 19:31:07,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3368, 1.2731, 1.1376, 1.5228], device='cuda:1'), covar=tensor([0.0703, 0.0317, 0.0335, 0.0768], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0080, 0.0057, 0.0052, 0.0087], device='cuda:1')
+2023-03-05 19:31:10,568 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:31:13,262 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:31:16,447 INFO [train.py:968] (1/2) Epoch 11, batch 22500, giga_loss[loss=0.2671, simple_loss=0.3508, pruned_loss=0.09172, over 28738.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3593, pruned_loss=0.1081, over 5712849.32 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3652, pruned_loss=0.1103, over 5745591.69 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3568, pruned_loss=0.1062, over 5704185.89 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:31:34,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5539, 2.2305, 1.6071, 0.9003], device='cuda:1'), covar=tensor([0.3779, 0.1875, 0.3033, 0.4065], device='cuda:1'), in_proj_covar=tensor([0.1545, 0.1463, 0.1488, 0.1263], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 19:31:35,306 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:32:00,281 INFO [train.py:968] (1/2) Epoch 11, batch 22550, giga_loss[loss=0.2599, simple_loss=0.3328, pruned_loss=0.09346, over 29003.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3568, pruned_loss=0.1069, over 5708710.37 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3657, pruned_loss=0.1109, over 5738425.28 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3543, pruned_loss=0.1048, over 5707732.20 frames. ], batch size: 136, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:32:11,947 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 22600, libri_loss[loss=0.2493, simple_loss=0.3176, pruned_loss=0.09048, over 29666.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.353, pruned_loss=0.1051, over 5712024.28 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3662, pruned_loss=0.1115, over 5740566.70 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3504, pruned_loss=0.1028, over 5708788.97 frames. ], batch size: 69, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:32:59,159 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-05 19:33:04,158 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 22650, giga_loss[loss=0.2347, simple_loss=0.3254, pruned_loss=0.072, over 28854.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3506, pruned_loss=0.1032, over 5707738.75 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3661, pruned_loss=0.1117, over 5736924.25 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3481, pruned_loss=0.1009, over 5707268.13 frames. ], batch size: 174, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:33:51,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:33:53,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 19:34:01,616 INFO [train.py:968] (1/2) Epoch 11, batch 22700, giga_loss[loss=0.2646, simple_loss=0.3566, pruned_loss=0.0863, over 28810.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3507, pruned_loss=0.1013, over 5696993.33 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3663, pruned_loss=0.1119, over 5734055.77 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3484, pruned_loss=0.09915, over 5698761.16 frames. ], batch size: 174, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:34:29,889 INFO [optim.py:369] (1/2) 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 19:34:44,739 INFO [train.py:968] (1/2) Epoch 11, batch 22750, giga_loss[loss=0.2775, simple_loss=0.3567, pruned_loss=0.09913, over 28601.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3539, pruned_loss=0.1028, over 5688494.50 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3664, pruned_loss=0.112, over 5726477.87 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3518, pruned_loss=0.1009, over 5696488.33 frames. ], batch size: 284, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:34:51,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:35:01,453 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 22800, giga_loss[loss=0.2307, simple_loss=0.305, pruned_loss=0.07821, over 28613.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.352, pruned_loss=0.1032, over 5684748.18 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3669, pruned_loss=0.1127, over 5722126.89 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3496, pruned_loss=0.1008, over 5693752.80 frames. ], batch size: 60, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:35:49,099 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:35:52,860 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 22850, giga_loss[loss=0.2473, simple_loss=0.3147, pruned_loss=0.08992, over 28703.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3503, pruned_loss=0.1039, over 5694248.20 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3669, pruned_loss=0.1129, over 5724977.94 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.348, pruned_loss=0.1017, over 5698090.70 frames. ], batch size: 99, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:36:13,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=478402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:42,840 INFO [train.py:968] (1/2) Epoch 11, batch 22900, giga_loss[loss=0.266, simple_loss=0.3365, pruned_loss=0.09774, over 28872.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3498, pruned_loss=0.105, over 5708120.35 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3673, pruned_loss=0.1133, over 5728585.19 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3473, pruned_loss=0.1027, over 5707567.95 frames. ], batch size: 186, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:36:46,343 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:48,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:50,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0230, 1.2262, 1.1106, 0.9887], device='cuda:1'), covar=tensor([0.1437, 0.1518, 0.0870, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1576, 0.1552, 0.1664], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:36:55,540 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:36:58,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=478458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:10,177 INFO [optim.py:369] (1/2) 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 19:37:10,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=478473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:15,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:22,049 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 22950, giga_loss[loss=0.2811, simple_loss=0.3534, pruned_loss=0.1044, over 28658.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3484, pruned_loss=0.1051, over 5698887.47 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3677, pruned_loss=0.1137, over 5721968.81 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3458, pruned_loss=0.1028, over 5703069.66 frames. ], batch size: 262, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:37:24,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:37:49,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5613, 2.3512, 1.6456, 0.7405], device='cuda:1'), covar=tensor([0.4348, 0.2003, 0.3340, 0.4946], device='cuda:1'), in_proj_covar=tensor([0.1544, 0.1461, 0.1494, 0.1267], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 19:38:01,112 INFO [train.py:968] (1/2) Epoch 11, batch 23000, libri_loss[loss=0.2695, simple_loss=0.3381, pruned_loss=0.1005, over 29569.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3483, pruned_loss=0.1049, over 5696964.09 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3682, pruned_loss=0.1142, over 5708993.07 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3451, pruned_loss=0.1022, over 5711879.05 frames. ], batch size: 76, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:38:13,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6867, 1.8482, 1.7425, 1.6663], device='cuda:1'), covar=tensor([0.1330, 0.1669, 0.1845, 0.1651], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0722, 0.0666, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 19:38:26,574 INFO [optim.py:369] (1/2) 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 19:38:39,360 INFO [train.py:968] (1/2) Epoch 11, batch 23050, giga_loss[loss=0.3116, simple_loss=0.3785, pruned_loss=0.1224, over 27877.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3444, pruned_loss=0.1028, over 5691231.93 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3681, pruned_loss=0.1143, over 5704410.81 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3414, pruned_loss=0.1004, over 5707447.90 frames. ], batch size: 412, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:38:53,704 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-05 19:39:05,021 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=478621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:39:08,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 23100, giga_loss[loss=0.2288, simple_loss=0.2994, pruned_loss=0.07914, over 28674.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3407, pruned_loss=0.1013, over 5695422.41 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3689, pruned_loss=0.1151, over 5707089.02 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3373, pruned_loss=0.09849, over 5705745.33 frames. ], batch size: 92, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:39:31,457 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 23150, giga_loss[loss=0.266, simple_loss=0.3504, pruned_loss=0.09085, over 28963.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.339, pruned_loss=0.09998, over 5695884.70 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3695, pruned_loss=0.1156, over 5704044.96 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3354, pruned_loss=0.09707, over 5706572.75 frames. ], batch size: 136, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:40:16,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5649, 3.6266, 1.6316, 1.5429], device='cuda:1'), covar=tensor([0.0895, 0.0316, 0.0908, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0506, 0.0335, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 19:40:40,324 INFO [train.py:968] (1/2) Epoch 11, batch 23200, giga_loss[loss=0.273, simple_loss=0.3518, pruned_loss=0.09706, over 28720.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3395, pruned_loss=0.09939, over 5703377.64 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3693, pruned_loss=0.1157, over 5708163.02 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3361, pruned_loss=0.09678, over 5708167.49 frames. ], batch size: 284, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:40:41,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2581, 2.9070, 1.4315, 1.4143], device='cuda:1'), covar=tensor([0.0925, 0.0385, 0.0881, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0506, 0.0334, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0022, 0.0030, 0.0020, 0.0025], device='cuda:1')
+2023-03-05 19:40:53,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5836, 2.3217, 1.4582, 0.7954], device='cuda:1'), covar=tensor([0.5346, 0.2512, 0.2725, 0.5041], device='cuda:1'), in_proj_covar=tensor([0.1537, 0.1455, 0.1486, 0.1257], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 19:41:08,016 INFO [optim.py:369] (1/2) 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 19:41:22,481 INFO [train.py:968] (1/2) Epoch 11, batch 23250, libri_loss[loss=0.3098, simple_loss=0.3848, pruned_loss=0.1174, over 29241.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3433, pruned_loss=0.1008, over 5709670.66 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3695, pruned_loss=0.1157, over 5712391.96 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3398, pruned_loss=0.09825, over 5709500.57 frames. ], batch size: 94, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:41:53,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4609, 1.8457, 1.5913, 1.3281], device='cuda:1'), covar=tensor([0.1979, 0.1699, 0.1772, 0.1944], device='cuda:1'), in_proj_covar=tensor([0.1694, 0.1583, 0.1555, 0.1659], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:42:02,690 INFO [train.py:968] (1/2) Epoch 11, batch 23300, giga_loss[loss=0.2698, simple_loss=0.3442, pruned_loss=0.09774, over 28802.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3482, pruned_loss=0.1033, over 5709976.71 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3699, pruned_loss=0.1161, over 5715406.57 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3443, pruned_loss=0.1005, over 5706831.01 frames. ], batch size: 119, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:42:26,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=478865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:42:31,572 INFO [optim.py:369] (1/2) 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 19:42:45,657 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:42:46,691 INFO [train.py:968] (1/2) Epoch 11, batch 23350, libri_loss[loss=0.3264, simple_loss=0.389, pruned_loss=0.1319, over 27902.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3513, pruned_loss=0.1044, over 5699152.02 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3704, pruned_loss=0.1164, over 5713256.38 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3475, pruned_loss=0.1017, over 5698595.06 frames. ], batch size: 116, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:43:27,137 INFO [train.py:968] (1/2) Epoch 11, batch 23400, giga_loss[loss=0.2613, simple_loss=0.3393, pruned_loss=0.09162, over 28873.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3535, pruned_loss=0.1055, over 5681377.04 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3699, pruned_loss=0.1162, over 5698116.39 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3505, pruned_loss=0.1033, over 5693687.14 frames. ], batch size: 213, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:43:30,092 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:43:57,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=478970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:43:59,312 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 23450, giga_loss[loss=0.3018, simple_loss=0.3627, pruned_loss=0.1205, over 28919.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3606, pruned_loss=0.1121, over 5685273.65 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3704, pruned_loss=0.1168, over 5704186.91 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3573, pruned_loss=0.1095, over 5689444.79 frames. ], batch size: 112, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:44:33,367 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:44:36,248 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 23500, giga_loss[loss=0.3428, simple_loss=0.3926, pruned_loss=0.1465, over 27527.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3654, pruned_loss=0.1164, over 5681349.20 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.37, pruned_loss=0.1167, over 5706336.90 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.363, pruned_loss=0.1144, over 5681767.45 frames. ], batch size: 472, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:45:03,575 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:45:15,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5281, 1.7234, 1.8109, 1.3400], device='cuda:1'), covar=tensor([0.1582, 0.2280, 0.1248, 0.1520], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0694, 0.0864, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 19:45:38,393 INFO [optim.py:369] (1/2) 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 19:45:51,006 INFO [train.py:968] (1/2) Epoch 11, batch 23550, giga_loss[loss=0.4013, simple_loss=0.4218, pruned_loss=0.1904, over 23936.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3729, pruned_loss=0.122, over 5680853.90 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.37, pruned_loss=0.1169, over 5702162.90 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3709, pruned_loss=0.1202, over 5683552.34 frames. ], batch size: 705, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:45:57,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:46:41,378 INFO [train.py:968] (1/2) Epoch 11, batch 23600, giga_loss[loss=0.3198, simple_loss=0.3794, pruned_loss=0.1301, over 28796.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.379, pruned_loss=0.1274, over 5678910.02 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3698, pruned_loss=0.1168, over 5706330.52 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3777, pruned_loss=0.1262, over 5676914.55 frames. ], batch size: 284, lr: 2.94e-03, grad_scale: 8.0
+2023-03-05 19:47:16,749 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 23650, giga_loss[loss=0.3699, simple_loss=0.4253, pruned_loss=0.1573, over 29023.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3868, pruned_loss=0.1346, over 5665440.65 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.37, pruned_loss=0.1169, over 5708032.34 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3857, pruned_loss=0.1337, over 5662112.80 frames. ], batch size: 136, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:48:24,103 INFO [train.py:968] (1/2) Epoch 11, batch 23700, giga_loss[loss=0.3225, simple_loss=0.3855, pruned_loss=0.1298, over 28961.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3906, pruned_loss=0.1371, over 5658268.26 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3702, pruned_loss=0.117, over 5701790.99 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3898, pruned_loss=0.1366, over 5660578.94 frames. ], batch size: 227, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:48:46,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:48:53,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8595, 1.7192, 1.2335, 1.3474], device='cuda:1'), covar=tensor([0.0723, 0.0584, 0.1008, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0444, 0.0502, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 19:48:59,183 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 23750, giga_loss[loss=0.3448, simple_loss=0.4046, pruned_loss=0.1425, over 28886.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3922, pruned_loss=0.1391, over 5648290.59 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3707, pruned_loss=0.1174, over 5694307.43 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3915, pruned_loss=0.1386, over 5656461.29 frames. ], batch size: 227, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:49:40,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:50:05,566 INFO [train.py:968] (1/2) Epoch 11, batch 23800, giga_loss[loss=0.2961, simple_loss=0.3607, pruned_loss=0.1157, over 28907.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.3943, pruned_loss=0.1421, over 5641063.48 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3706, pruned_loss=0.1175, over 5695655.81 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.394, pruned_loss=0.1419, over 5645699.98 frames. ], batch size: 199, lr: 2.94e-03, grad_scale: 4.0
+2023-03-05 19:50:13,549 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:50:56,479 INFO [train.py:968] (1/2) Epoch 11, batch 23850, libri_loss[loss=0.3332, simple_loss=0.3887, pruned_loss=0.1388, over 29629.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3977, pruned_loss=0.1459, over 5635433.20 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3711, pruned_loss=0.118, over 5693393.20 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3979, pruned_loss=0.1462, over 5638960.87 frames. ], batch size: 91, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:51:03,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6622, 1.0387, 2.8560, 2.6147], device='cuda:1'), covar=tensor([0.1776, 0.2413, 0.0557, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0667, 0.0591, 0.0864, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 19:51:13,402 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:51:16,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:51:50,639 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:51:53,328 INFO [train.py:968] (1/2) Epoch 11, batch 23900, giga_loss[loss=0.3789, simple_loss=0.403, pruned_loss=0.1774, over 23531.00 frames. ], tot_loss[loss=0.3508, simple_loss=0.4017, pruned_loss=0.15, over 5617317.69 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3713, pruned_loss=0.1183, over 5696083.75 frames. ], giga_tot_loss[loss=0.3518, simple_loss=0.4024, pruned_loss=0.1506, over 5616458.89 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:52:16,321 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:19,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:25,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:30,239 INFO [optim.py:369] (1/2) 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 19:52:44,131 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 23950, giga_loss[loss=0.2937, simple_loss=0.3609, pruned_loss=0.1133, over 28948.00 frames. ], tot_loss[loss=0.3501, simple_loss=0.4004, pruned_loss=0.1499, over 5614701.66 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3714, pruned_loss=0.1186, over 5700124.09 frames. ], giga_tot_loss[loss=0.3521, simple_loss=0.4018, pruned_loss=0.1512, over 5608020.30 frames. ], batch size: 145, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:52:47,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:52:48,252 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:53:16,272 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:53:22,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6807, 1.7850, 1.7689, 1.5595], device='cuda:1'), covar=tensor([0.1240, 0.1558, 0.1639, 0.1528], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0733, 0.0673, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 19:53:31,653 INFO [train.py:968] (1/2) Epoch 11, batch 24000, giga_loss[loss=0.3592, simple_loss=0.4171, pruned_loss=0.1506, over 28922.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.3985, pruned_loss=0.1485, over 5627169.02 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3718, pruned_loss=0.1188, over 5700898.86 frames. ], giga_tot_loss[loss=0.3506, simple_loss=0.4003, pruned_loss=0.1505, over 5618355.25 frames. ], batch size: 174, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 19:53:31,654 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 19:53:36,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2594, 1.8708, 1.3441, 0.4315], device='cuda:1'), covar=tensor([0.3359, 0.2493, 0.3789, 0.4412], device='cuda:1'), in_proj_covar=tensor([0.1555, 0.1482, 0.1493, 0.1267], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 19:53:39,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2956, 3.1288, 1.4544, 1.4334], device='cuda:1'), covar=tensor([0.1091, 0.0489, 0.0935, 0.1483], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0509, 0.0336, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 19:53:40,831 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19554MB
+2023-03-05 19:53:47,157 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:54:27,345 INFO [train.py:968] (1/2) Epoch 11, batch 24050, giga_loss[loss=0.4109, simple_loss=0.4321, pruned_loss=0.1948, over 23616.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3981, pruned_loss=0.1481, over 5622618.30 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3721, pruned_loss=0.1193, over 5702028.18 frames. ], giga_tot_loss[loss=0.35, simple_loss=0.3999, pruned_loss=0.15, over 5612429.57 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:54:48,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=479612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:54:51,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=479615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:55:08,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7319, 1.7860, 1.3364, 1.3782], device='cuda:1'), covar=tensor([0.0874, 0.0633, 0.1051, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0445, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 19:55:16,804 INFO [train.py:968] (1/2) Epoch 11, batch 24100, giga_loss[loss=0.3906, simple_loss=0.4344, pruned_loss=0.1734, over 27823.00 frames. ], tot_loss[loss=0.3447, simple_loss=0.3973, pruned_loss=0.146, over 5629347.59 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.372, pruned_loss=0.1193, over 5706196.54 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.3995, pruned_loss=0.1481, over 5616072.85 frames. ], batch size: 412, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:55:21,789 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=479644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:55:50,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479675.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:55:51,342 INFO [optim.py:369] (1/2) 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 19:56:06,632 INFO [train.py:968] (1/2) Epoch 11, batch 24150, giga_loss[loss=0.3033, simple_loss=0.3816, pruned_loss=0.1126, over 28932.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3987, pruned_loss=0.1467, over 5625900.63 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3719, pruned_loss=0.1195, over 5704375.41 frames. ], giga_tot_loss[loss=0.3494, simple_loss=0.4011, pruned_loss=0.1488, over 5615385.25 frames. ], batch size: 174, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:56:15,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2530, 2.7795, 2.1123, 1.7925], device='cuda:1'), covar=tensor([0.1733, 0.1276, 0.1541, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.1726, 0.1616, 0.1587, 0.1699], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 19:56:30,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-05 19:56:51,563 INFO [train.py:968] (1/2) Epoch 11, batch 24200, giga_loss[loss=0.3127, simple_loss=0.3758, pruned_loss=0.1248, over 28542.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.396, pruned_loss=0.1444, over 5627739.31 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.371, pruned_loss=0.1192, over 5700568.63 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.4, pruned_loss=0.1476, over 5619415.97 frames. ], batch size: 336, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:57:12,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:57:16,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7368, 1.7177, 1.3884, 1.4229], device='cuda:1'), covar=tensor([0.0743, 0.0568, 0.0909, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0446, 0.0505, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 19:57:27,980 INFO [optim.py:369] (1/2) 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 19:57:39,175 INFO [train.py:968] (1/2) Epoch 11, batch 24250, libri_loss[loss=0.3154, simple_loss=0.376, pruned_loss=0.1274, over 29537.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3925, pruned_loss=0.141, over 5634602.92 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3709, pruned_loss=0.1196, over 5710708.76 frames. ], giga_tot_loss[loss=0.3428, simple_loss=0.3969, pruned_loss=0.1444, over 5614775.85 frames. ], batch size: 79, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:57:39,577 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-05 19:58:08,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:58:27,093 INFO [train.py:968] (1/2) Epoch 11, batch 24300, giga_loss[loss=0.3504, simple_loss=0.4, pruned_loss=0.1504, over 28022.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3913, pruned_loss=0.1389, over 5642320.13 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3715, pruned_loss=0.1201, over 5712484.94 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3947, pruned_loss=0.1414, over 5623719.43 frames. ], batch size: 412, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:58:34,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:58:53,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2136, 1.5897, 1.1882, 1.4742], device='cuda:1'), covar=tensor([0.2502, 0.2386, 0.2734, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.0959, 0.1145, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 19:58:58,842 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 19:59:00,620 INFO [optim.py:369] (1/2) 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 19:59:13,124 INFO [train.py:968] (1/2) Epoch 11, batch 24350, giga_loss[loss=0.3009, simple_loss=0.3736, pruned_loss=0.1141, over 28270.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3879, pruned_loss=0.1362, over 5631693.18 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.371, pruned_loss=0.12, over 5705898.46 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3914, pruned_loss=0.1387, over 5620504.33 frames. ], batch size: 368, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 19:59:23,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8682, 3.6945, 3.5124, 2.0186], device='cuda:1'), covar=tensor([0.0553, 0.0721, 0.0693, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.1054, 0.0993, 0.0867, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 19:59:37,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.31 vs. limit=5.0
+2023-03-05 19:59:45,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=479922.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 19:59:59,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9944, 3.8191, 3.6094, 1.8641], device='cuda:1'), covar=tensor([0.0629, 0.0746, 0.0762, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0990, 0.0865, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 20:00:03,228 INFO [train.py:968] (1/2) Epoch 11, batch 24400, giga_loss[loss=0.3113, simple_loss=0.3693, pruned_loss=0.1266, over 27599.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3861, pruned_loss=0.1348, over 5638735.57 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3711, pruned_loss=0.12, over 5708936.44 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3891, pruned_loss=0.1371, over 5625890.60 frames. ], batch size: 472, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:00:34,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3085, 1.5632, 1.3451, 1.4801], device='cuda:1'), covar=tensor([0.0743, 0.0328, 0.0314, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 20:00:39,584 INFO [optim.py:369] (1/2) 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 20:00:47,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 20:00:51,084 INFO [train.py:968] (1/2) Epoch 11, batch 24450, giga_loss[loss=0.3271, simple_loss=0.3863, pruned_loss=0.134, over 28177.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.386, pruned_loss=0.1352, over 5638405.69 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.371, pruned_loss=0.12, over 5710776.81 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3885, pruned_loss=0.1371, over 5626160.44 frames. ], batch size: 368, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:00:58,173 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=479996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:01:08,438 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-05 20:01:46,552 INFO [train.py:968] (1/2) Epoch 11, batch 24500, libri_loss[loss=0.3066, simple_loss=0.3673, pruned_loss=0.123, over 29595.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3855, pruned_loss=0.1348, over 5639641.32 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3708, pruned_loss=0.1199, over 5712270.38 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3881, pruned_loss=0.1368, over 5626496.77 frames. ], batch size: 74, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:01:58,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480050.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:02:12,396 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480065.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:02:15,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480068.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:02:16,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3379, 1.3827, 1.2248, 1.4518], device='cuda:1'), covar=tensor([0.0752, 0.0331, 0.0320, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 20:02:23,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:02:24,527 INFO [optim.py:369] (1/2) 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 20:02:30,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-05 20:02:36,027 INFO [train.py:968] (1/2) Epoch 11, batch 24550, giga_loss[loss=0.2967, simple_loss=0.368, pruned_loss=0.1127, over 28906.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3828, pruned_loss=0.1318, over 5658843.49 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3703, pruned_loss=0.1197, over 5717001.83 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3859, pruned_loss=0.1342, over 5641682.46 frames. ], batch size: 213, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:02:43,829 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480097.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:03:18,196 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 24600, giga_loss[loss=0.3188, simple_loss=0.391, pruned_loss=0.1233, over 28887.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3826, pruned_loss=0.1298, over 5662302.87 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3704, pruned_loss=0.1201, over 5720881.16 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3853, pruned_loss=0.1316, over 5643488.73 frames. ], batch size: 186, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:04:05,330 INFO [optim.py:369] (1/2) 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 20:04:17,006 INFO [train.py:968] (1/2) Epoch 11, batch 24650, giga_loss[loss=0.2974, simple_loss=0.3622, pruned_loss=0.1164, over 28382.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3837, pruned_loss=0.1283, over 5670143.47 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3707, pruned_loss=0.1204, over 5721487.86 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3858, pruned_loss=0.1297, over 5653689.89 frames. ], batch size: 65, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:04:22,495 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480193.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:04:23,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:04:25,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480196.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:04:37,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-05 20:04:48,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:04:50,748 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480225.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:05:03,937 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:05,032 INFO [train.py:968] (1/2) Epoch 11, batch 24700, libri_loss[loss=0.3572, simple_loss=0.4069, pruned_loss=0.1538, over 25904.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3849, pruned_loss=0.1301, over 5657837.42 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3708, pruned_loss=0.1207, over 5713704.99 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.387, pruned_loss=0.1311, over 5650002.94 frames. ], batch size: 136, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:05:14,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:37,538 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:39,288 INFO [optim.py:369] (1/2) 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 20:05:42,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:05:50,582 INFO [train.py:968] (1/2) Epoch 11, batch 24750, giga_loss[loss=0.3802, simple_loss=0.4167, pruned_loss=0.1719, over 26595.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.385, pruned_loss=0.1304, over 5668510.82 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3711, pruned_loss=0.1211, over 5708762.16 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3868, pruned_loss=0.1311, over 5665012.23 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:06:08,889 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:06:23,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 20:06:39,229 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:06:39,502 INFO [train.py:968] (1/2) Epoch 11, batch 24800, giga_loss[loss=0.3497, simple_loss=0.3983, pruned_loss=0.1505, over 27600.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3831, pruned_loss=0.1297, over 5680568.05 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3714, pruned_loss=0.1214, over 5712625.87 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3845, pruned_loss=0.1302, over 5673671.06 frames. ], batch size: 472, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:06:41,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:04,410 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:06,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:08,946 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:09,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:15,551 INFO [optim.py:369] (1/2) 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 20:07:25,349 INFO [train.py:968] (1/2) Epoch 11, batch 24850, giga_loss[loss=0.3138, simple_loss=0.3734, pruned_loss=0.1271, over 28982.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3821, pruned_loss=0.1305, over 5676891.25 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3712, pruned_loss=0.1213, over 5712567.74 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3835, pruned_loss=0.131, over 5671384.49 frames. ], batch size: 227, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:07:28,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:30,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:32,649 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:07:56,189 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 24900, libri_loss[loss=0.2974, simple_loss=0.3634, pruned_loss=0.1157, over 29562.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3809, pruned_loss=0.1295, over 5677619.71 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3708, pruned_loss=0.1211, over 5716568.51 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3827, pruned_loss=0.1304, over 5668485.87 frames. ], batch size: 76, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:08:19,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:08:44,834 INFO [optim.py:369] (1/2) 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 20:08:53,170 INFO [train.py:968] (1/2) Epoch 11, batch 24950, giga_loss[loss=0.3169, simple_loss=0.3859, pruned_loss=0.1239, over 28860.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3796, pruned_loss=0.1267, over 5681225.52 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.371, pruned_loss=0.1214, over 5712169.28 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.381, pruned_loss=0.1273, over 5677380.84 frames. ], batch size: 199, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:09:19,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:09:21,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:09:30,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7816, 2.0951, 2.0462, 1.6071], device='cuda:1'), covar=tensor([0.1699, 0.2091, 0.1336, 0.1555], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0701, 0.0867, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 20:09:41,569 INFO [train.py:968] (1/2) Epoch 11, batch 25000, giga_loss[loss=0.4673, simple_loss=0.4689, pruned_loss=0.2328, over 26636.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3815, pruned_loss=0.1284, over 5656889.17 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3718, pruned_loss=0.1222, over 5694400.89 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3822, pruned_loss=0.1283, over 5669106.04 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:09:47,210 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:09:58,770 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-05 20:10:18,257 INFO [optim.py:369] (1/2) 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 20:10:23,116 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 25050, giga_loss[loss=0.2705, simple_loss=0.3507, pruned_loss=0.09513, over 28709.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3793, pruned_loss=0.1266, over 5667941.58 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3716, pruned_loss=0.1221, over 5697385.90 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3802, pruned_loss=0.1267, over 5674287.96 frames. ], batch size: 262, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:10:34,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:10:37,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:10:51,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:11:06,606 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:11:22,077 INFO [train.py:968] (1/2) Epoch 11, batch 25100, giga_loss[loss=0.3651, simple_loss=0.3963, pruned_loss=0.167, over 23461.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3784, pruned_loss=0.1266, over 5661114.38 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.372, pruned_loss=0.1224, over 5690277.56 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3788, pruned_loss=0.1265, over 5671929.71 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:12:00,315 INFO [optim.py:369] (1/2) 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 20:12:09,586 INFO [train.py:968] (1/2) Epoch 11, batch 25150, giga_loss[loss=0.2956, simple_loss=0.3661, pruned_loss=0.1125, over 28965.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3784, pruned_loss=0.1277, over 5657448.99 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3728, pruned_loss=0.1231, over 5694025.91 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3782, pruned_loss=0.127, over 5661976.52 frames. ], batch size: 145, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:12:56,184 INFO [train.py:968] (1/2) Epoch 11, batch 25200, giga_loss[loss=0.3912, simple_loss=0.4187, pruned_loss=0.1819, over 26618.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3782, pruned_loss=0.1283, over 5664229.49 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3726, pruned_loss=0.123, over 5696302.20 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3783, pruned_loss=0.1279, over 5665302.57 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:13:12,428 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=480755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:14,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=480758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:33,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:34,097 INFO [optim.py:369] (1/2) 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 20:13:43,515 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=480787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:13:44,584 INFO [train.py:968] (1/2) Epoch 11, batch 25250, libri_loss[loss=0.4013, simple_loss=0.4412, pruned_loss=0.1807, over 19711.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3777, pruned_loss=0.1283, over 5658069.43 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3734, pruned_loss=0.1237, over 5692317.07 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3772, pruned_loss=0.1276, over 5662387.98 frames. ], batch size: 186, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:14:07,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:14:19,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5541, 1.5536, 1.2039, 1.2560], device='cuda:1'), covar=tensor([0.0734, 0.0532, 0.1026, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0447, 0.0503, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 20:14:30,185 INFO [train.py:968] (1/2) Epoch 11, batch 25300, giga_loss[loss=0.4038, simple_loss=0.4259, pruned_loss=0.1908, over 26746.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3774, pruned_loss=0.1285, over 5655396.28 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3736, pruned_loss=0.1237, over 5685336.64 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1279, over 5664000.07 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:14:50,118 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 20:15:06,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-05 20:15:06,937 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 25350, giga_loss[loss=0.3053, simple_loss=0.3703, pruned_loss=0.1201, over 28898.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3774, pruned_loss=0.1287, over 5660164.69 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5692813.20 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3772, pruned_loss=0.1285, over 5659428.51 frames. ], batch size: 227, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:15:22,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3453, 1.5811, 1.6381, 1.2372], device='cuda:1'), covar=tensor([0.1244, 0.1830, 0.1016, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0700, 0.0865, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 20:16:01,242 INFO [train.py:968] (1/2) Epoch 11, batch 25400, giga_loss[loss=0.2924, simple_loss=0.3716, pruned_loss=0.1066, over 28956.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3775, pruned_loss=0.1277, over 5663281.73 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3735, pruned_loss=0.1237, over 5695167.40 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3775, pruned_loss=0.1276, over 5659500.56 frames. ], batch size: 155, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:16:16,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=480957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:16:18,723 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=480960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:16:35,806 INFO [optim.py:369] (1/2) 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 20:16:44,125 INFO [train.py:968] (1/2) Epoch 11, batch 25450, giga_loss[loss=0.2923, simple_loss=0.3649, pruned_loss=0.1099, over 28813.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3765, pruned_loss=0.1264, over 5663223.15 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1237, over 5690251.56 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3767, pruned_loss=0.1264, over 5663230.75 frames. ], batch size: 284, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:16:55,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3485, 1.6267, 1.2727, 1.6518], device='cuda:1'), covar=tensor([0.2651, 0.2572, 0.2912, 0.2267], device='cuda:1'), in_proj_covar=tensor([0.1294, 0.0963, 0.1143, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 20:17:34,064 INFO [train.py:968] (1/2) Epoch 11, batch 25500, libri_loss[loss=0.3184, simple_loss=0.3851, pruned_loss=0.1258, over 29197.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3768, pruned_loss=0.1266, over 5649659.72 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3735, pruned_loss=0.1238, over 5684578.26 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3768, pruned_loss=0.1265, over 5654026.39 frames. ], batch size: 97, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:17:55,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481063.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:18:09,989 INFO [optim.py:369] (1/2) 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 20:18:10,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1854, 0.9087, 0.9291, 1.3866], device='cuda:1'), covar=tensor([0.0749, 0.0357, 0.0340, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 20:18:17,697 INFO [train.py:968] (1/2) Epoch 11, batch 25550, giga_loss[loss=0.2886, simple_loss=0.3554, pruned_loss=0.1109, over 28797.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3779, pruned_loss=0.1276, over 5654263.51 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3733, pruned_loss=0.1236, over 5684434.30 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3782, pruned_loss=0.1279, over 5656923.32 frames. ], batch size: 66, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:18:29,119 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:18:33,192 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:18:35,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3825, 1.6131, 1.3016, 1.5221], device='cuda:1'), covar=tensor([0.2359, 0.2309, 0.2572, 0.2026], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.0959, 0.1141, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 20:18:39,276 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:18:51,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6495, 1.6308, 1.3660, 1.3717], device='cuda:1'), covar=tensor([0.0647, 0.0436, 0.0904, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0447, 0.0503, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 20:19:00,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:19:05,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-05 20:19:06,241 INFO [train.py:968] (1/2) Epoch 11, batch 25600, giga_loss[loss=0.3074, simple_loss=0.3642, pruned_loss=0.1253, over 28835.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3806, pruned_loss=0.1307, over 5651619.51 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1238, over 5688866.08 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3809, pruned_loss=0.1308, over 5648876.86 frames. ], batch size: 112, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:19:19,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:19:35,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 20:19:47,205 INFO [optim.py:369] (1/2) 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 20:19:57,537 INFO [train.py:968] (1/2) Epoch 11, batch 25650, giga_loss[loss=0.3302, simple_loss=0.3842, pruned_loss=0.1381, over 28627.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3817, pruned_loss=0.1327, over 5658798.20 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3737, pruned_loss=0.1242, over 5689564.39 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3818, pruned_loss=0.1326, over 5655715.07 frames. ], batch size: 85, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:20:00,011 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 25700, giga_loss[loss=0.4197, simple_loss=0.4479, pruned_loss=0.1958, over 24259.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3823, pruned_loss=0.1342, over 5647406.50 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3734, pruned_loss=0.1241, over 5685730.43 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3829, pruned_loss=0.1345, over 5647055.48 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:21:19,089 INFO [optim.py:369] (1/2) 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 20:21:26,113 INFO [train.py:968] (1/2) Epoch 11, batch 25750, giga_loss[loss=0.3125, simple_loss=0.3755, pruned_loss=0.1248, over 28731.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3815, pruned_loss=0.1333, over 5661593.61 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.373, pruned_loss=0.124, over 5695417.98 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3829, pruned_loss=0.1343, over 5650065.25 frames. ], batch size: 92, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:21:33,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:21:36,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:21:46,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-05 20:22:04,798 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:10,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:11,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:12,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:15,324 INFO [train.py:968] (1/2) Epoch 11, batch 25800, giga_loss[loss=0.3847, simple_loss=0.4181, pruned_loss=0.1756, over 24157.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3811, pruned_loss=0.1332, over 5660852.25 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3731, pruned_loss=0.1241, over 5698435.81 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3823, pruned_loss=0.134, over 5648311.20 frames. ], batch size: 705, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:22:25,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2563, 1.3939, 1.3163, 1.2694], device='cuda:1'), covar=tensor([0.1690, 0.1442, 0.1207, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.1706, 0.1608, 0.1580, 0.1693], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 20:22:38,741 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:22:49,326 INFO [optim.py:369] (1/2) 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 20:22:58,695 INFO [train.py:968] (1/2) Epoch 11, batch 25850, giga_loss[loss=0.3187, simple_loss=0.3831, pruned_loss=0.1271, over 28833.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3802, pruned_loss=0.1308, over 5662475.19 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1243, over 5690391.39 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3811, pruned_loss=0.1314, over 5659552.35 frames. ], batch size: 186, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:23:11,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-05 20:23:47,880 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481438.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:23:48,268 INFO [train.py:968] (1/2) Epoch 11, batch 25900, giga_loss[loss=0.3318, simple_loss=0.3864, pruned_loss=0.1387, over 28951.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1283, over 5656514.75 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3734, pruned_loss=0.1244, over 5692280.85 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3779, pruned_loss=0.1287, over 5652119.45 frames. ], batch size: 213, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:24:25,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:24:26,724 INFO [optim.py:369] (1/2) 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 20:24:27,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 20:24:28,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:24:32,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:24:36,769 INFO [train.py:968] (1/2) Epoch 11, batch 25950, giga_loss[loss=0.2936, simple_loss=0.3647, pruned_loss=0.1113, over 29059.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3754, pruned_loss=0.1273, over 5667582.64 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3733, pruned_loss=0.1244, over 5696344.70 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.376, pruned_loss=0.1277, over 5659825.39 frames. ], batch size: 155, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:24:45,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4485, 1.6288, 1.7025, 1.3448], device='cuda:1'), covar=tensor([0.1289, 0.1868, 0.1037, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0702, 0.0869, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 20:24:55,119 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:25:18,366 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 20:25:22,361 INFO [train.py:968] (1/2) Epoch 11, batch 26000, giga_loss[loss=0.307, simple_loss=0.3692, pruned_loss=0.1224, over 28949.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3733, pruned_loss=0.1262, over 5679734.03 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3732, pruned_loss=0.1243, over 5700551.44 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3738, pruned_loss=0.1266, over 5669314.21 frames. ], batch size: 213, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:25:53,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:25:58,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0622, 2.4515, 2.0727, 2.5691], device='cuda:1'), covar=tensor([0.1910, 0.1753, 0.1916, 0.1595], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.0966, 0.1151, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 20:26:00,646 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:04,220 INFO [optim.py:369] (1/2) 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 20:26:04,512 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481581.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:26:06,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481584.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:26:09,586 INFO [train.py:968] (1/2) Epoch 11, batch 26050, libri_loss[loss=0.3074, simple_loss=0.37, pruned_loss=0.1224, over 29532.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3742, pruned_loss=0.127, over 5677208.64 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3737, pruned_loss=0.1246, over 5698933.62 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3742, pruned_loss=0.1271, over 5669401.04 frames. ], batch size: 82, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:26:12,818 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:34,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=481613.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:26:46,601 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=481627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:48,729 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=481630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:26:56,223 INFO [train.py:968] (1/2) Epoch 11, batch 26100, libri_loss[loss=0.3328, simple_loss=0.3817, pruned_loss=0.1419, over 29468.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3768, pruned_loss=0.1282, over 5680706.75 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3732, pruned_loss=0.1245, over 5700346.58 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5672557.00 frames. ], batch size: 70, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:27:15,769 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 26150, giga_loss[loss=0.284, simple_loss=0.3735, pruned_loss=0.09725, over 28864.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3801, pruned_loss=0.1268, over 5678720.82 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3734, pruned_loss=0.1247, over 5698179.75 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3804, pruned_loss=0.1269, over 5673941.94 frames. ], batch size: 145, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:27:56,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9300, 1.8645, 1.3474, 1.5997], device='cuda:1'), covar=tensor([0.0807, 0.0703, 0.1055, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0445, 0.0499, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 20:28:35,889 INFO [train.py:968] (1/2) Epoch 11, batch 26200, giga_loss[loss=0.3343, simple_loss=0.4062, pruned_loss=0.1312, over 28585.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3819, pruned_loss=0.1274, over 5678532.32 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3733, pruned_loss=0.1246, over 5696610.36 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3823, pruned_loss=0.1276, over 5676189.05 frames. ], batch size: 307, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:28:42,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-05 20:29:15,563 INFO [optim.py:369] (1/2) 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 20:29:23,122 INFO [train.py:968] (1/2) Epoch 11, batch 26250, giga_loss[loss=0.3796, simple_loss=0.4275, pruned_loss=0.1659, over 28496.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3837, pruned_loss=0.1297, over 5680358.49 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3724, pruned_loss=0.1242, over 5701919.27 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.385, pruned_loss=0.1303, over 5673466.96 frames. ], batch size: 336, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:29:41,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3448, 1.6199, 1.6107, 1.1884], device='cuda:1'), covar=tensor([0.1561, 0.2270, 0.1260, 0.1499], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0698, 0.0865, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 20:29:46,551 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481818.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:30:06,964 INFO [train.py:968] (1/2) Epoch 11, batch 26300, giga_loss[loss=0.3864, simple_loss=0.4251, pruned_loss=0.1738, over 27499.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3845, pruned_loss=0.1309, over 5680583.20 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3726, pruned_loss=0.1243, over 5704935.21 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3858, pruned_loss=0.1315, over 5672092.51 frames. ], batch size: 472, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:30:15,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3704, 2.7206, 1.5217, 1.3953], device='cuda:1'), covar=tensor([0.0801, 0.0368, 0.0740, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0514, 0.0339, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 20:30:43,324 INFO [optim.py:369] (1/2) 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 20:30:49,383 INFO [train.py:968] (1/2) Epoch 11, batch 26350, giga_loss[loss=0.2743, simple_loss=0.3367, pruned_loss=0.106, over 28656.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3836, pruned_loss=0.1312, over 5686183.65 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1245, over 5708640.22 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.385, pruned_loss=0.1318, over 5675107.20 frames. ], batch size: 92, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:30:52,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6973, 1.6894, 1.2702, 1.3196], device='cuda:1'), covar=tensor([0.0775, 0.0611, 0.1018, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0445, 0.0501, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 20:31:38,058 INFO [train.py:968] (1/2) Epoch 11, batch 26400, giga_loss[loss=0.2968, simple_loss=0.3701, pruned_loss=0.1118, over 28798.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3824, pruned_loss=0.1308, over 5690681.11 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3731, pruned_loss=0.1248, over 5710358.03 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3833, pruned_loss=0.1311, over 5680087.24 frames. ], batch size: 284, lr: 2.93e-03, grad_scale: 8.0
+2023-03-05 20:31:42,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:31:50,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=481953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:31:54,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=481957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:32:04,259 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:32:24,257 INFO [train.py:968] (1/2) Epoch 11, batch 26450, giga_loss[loss=0.3249, simple_loss=0.3785, pruned_loss=0.1357, over 28670.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.381, pruned_loss=0.1307, over 5689493.65 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3739, pruned_loss=0.1253, over 5706459.66 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3813, pruned_loss=0.1306, over 5683241.41 frames. ], batch size: 92, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:32:52,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3769, 3.6270, 1.6078, 1.4433], device='cuda:1'), covar=tensor([0.0895, 0.0329, 0.0796, 0.1278], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0517, 0.0341, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 20:33:14,102 INFO [train.py:968] (1/2) Epoch 11, batch 26500, giga_loss[loss=0.327, simple_loss=0.3904, pruned_loss=0.1318, over 28867.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3792, pruned_loss=0.1297, over 5689027.15 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3739, pruned_loss=0.1253, over 5712313.59 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3796, pruned_loss=0.1299, over 5678257.17 frames. ], batch size: 145, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:33:42,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3865, 1.6778, 1.6850, 1.2495], device='cuda:1'), covar=tensor([0.1409, 0.2103, 0.1159, 0.1373], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0699, 0.0866, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 20:33:54,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4614, 2.1487, 1.9627, 1.9902], device='cuda:1'), covar=tensor([0.1247, 0.2123, 0.1888, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0728, 0.0670, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 20:33:54,762 INFO [optim.py:369] (1/2) 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 20:33:57,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4483, 1.0416, 4.7776, 3.3876], device='cuda:1'), covar=tensor([0.1679, 0.2766, 0.0367, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0669, 0.0594, 0.0863, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 20:33:58,975 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:00,944 INFO [train.py:968] (1/2) Epoch 11, batch 26550, giga_loss[loss=0.2689, simple_loss=0.327, pruned_loss=0.1054, over 28704.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3796, pruned_loss=0.1302, over 5685623.67 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.374, pruned_loss=0.1253, over 5713721.57 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3799, pruned_loss=0.1303, over 5675512.54 frames. ], batch size: 92, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:34:02,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:10,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:11,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:21,358 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:24,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:28,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:34:38,257 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 26600, giga_loss[loss=0.2804, simple_loss=0.3383, pruned_loss=0.1113, over 28518.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3782, pruned_loss=0.1299, over 5683780.42 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1252, over 5716645.43 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3787, pruned_loss=0.1302, over 5672699.37 frames. ], batch size: 71, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:34:50,085 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:35:29,453 INFO [optim.py:369] (1/2) 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 20:35:29,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-05 20:35:34,918 INFO [train.py:968] (1/2) Epoch 11, batch 26650, giga_loss[loss=0.2978, simple_loss=0.3729, pruned_loss=0.1113, over 28916.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.377, pruned_loss=0.1299, over 5664527.12 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.374, pruned_loss=0.1253, over 5719887.62 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3774, pruned_loss=0.1302, over 5652201.94 frames. ], batch size: 145, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:35:39,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=482193.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:36:20,099 INFO [train.py:968] (1/2) Epoch 11, batch 26700, giga_loss[loss=0.2865, simple_loss=0.3566, pruned_loss=0.1083, over 28825.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3769, pruned_loss=0.1291, over 5665317.16 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3741, pruned_loss=0.1254, over 5714665.03 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3772, pruned_loss=0.1293, over 5658671.59 frames. ], batch size: 119, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:36:37,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6402, 1.7277, 1.6588, 1.6448], device='cuda:1'), covar=tensor([0.1282, 0.1790, 0.1810, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0734, 0.0672, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 20:36:42,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-05 20:37:04,243 INFO [optim.py:369] (1/2) 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 20:37:09,593 INFO [train.py:968] (1/2) Epoch 11, batch 26750, giga_loss[loss=0.416, simple_loss=0.4407, pruned_loss=0.1956, over 26691.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3789, pruned_loss=0.1299, over 5665863.13 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3734, pruned_loss=0.125, over 5717451.80 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3797, pruned_loss=0.1305, over 5657574.65 frames. ], batch size: 555, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:37:56,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=482332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:37:59,084 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482336.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:38:00,916 INFO [train.py:968] (1/2) Epoch 11, batch 26800, giga_loss[loss=0.2792, simple_loss=0.3537, pruned_loss=0.1024, over 28690.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3796, pruned_loss=0.1305, over 5662387.21 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3736, pruned_loss=0.1251, over 5718627.56 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3802, pruned_loss=0.131, over 5653782.89 frames. ], batch size: 85, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:38:02,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482339.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 20:38:21,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2422, 1.2938, 1.0993, 1.0629], device='cuda:1'), covar=tensor([0.0779, 0.0513, 0.1043, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0445, 0.0501, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 20:38:29,327 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482368.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 20:38:42,341 INFO [optim.py:369] (1/2) 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 20:38:46,925 INFO [train.py:968] (1/2) Epoch 11, batch 26850, giga_loss[loss=0.2935, simple_loss=0.3813, pruned_loss=0.1029, over 28548.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3809, pruned_loss=0.13, over 5679188.31 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3741, pruned_loss=0.1256, over 5721750.61 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.381, pruned_loss=0.13, over 5668586.57 frames. ], batch size: 78, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:39:33,123 INFO [train.py:968] (1/2) Epoch 11, batch 26900, giga_loss[loss=0.2868, simple_loss=0.3785, pruned_loss=0.09762, over 29069.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3821, pruned_loss=0.1279, over 5667143.72 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3745, pruned_loss=0.1259, over 5704522.67 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.382, pruned_loss=0.1277, over 5673345.01 frames. ], batch size: 136, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:40:06,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:40:08,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=482478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:40:09,612 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:40:16,634 INFO [train.py:968] (1/2) Epoch 11, batch 26950, giga_loss[loss=0.3389, simple_loss=0.3987, pruned_loss=0.1395, over 28571.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.385, pruned_loss=0.1288, over 5674460.55 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3745, pruned_loss=0.126, over 5706696.69 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3852, pruned_loss=0.1286, over 5676363.66 frames. ], batch size: 336, lr: 2.93e-03, grad_scale: 4.0
+2023-03-05 20:40:32,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=482507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:41:00,995 INFO [train.py:968] (1/2) Epoch 11, batch 27000, giga_loss[loss=0.336, simple_loss=0.3939, pruned_loss=0.1391, over 28587.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3881, pruned_loss=0.1316, over 5682222.67 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3748, pruned_loss=0.1263, over 5711814.00 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3883, pruned_loss=0.1313, over 5678249.17 frames. ], batch size: 78, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:41:00,996 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 20:41:09,805 INFO [train.py:1012] (1/2) Epoch 11, validation: loss=0.2156, simple_loss=0.321, pruned_loss=0.05513, over 944034.00 frames. 
+2023-03-05 20:41:09,806 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-05 20:41:12,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7740, 1.7437, 1.6892, 1.5623], device='cuda:1'), covar=tensor([0.1272, 0.1889, 0.1736, 0.1716], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0730, 0.0667, 0.0654], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 20:41:52,331 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 27050, giga_loss[loss=0.4226, simple_loss=0.4506, pruned_loss=0.1973, over 27520.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3898, pruned_loss=0.1341, over 5668587.24 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.375, pruned_loss=0.1264, over 5705978.74 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3903, pruned_loss=0.1339, over 5670167.01 frames. ], batch size: 472, lr: 2.93e-03, grad_scale: 2.0
+2023-03-05 20:42:29,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3495, 1.3298, 1.1631, 1.6079], device='cuda:1'), covar=tensor([0.0697, 0.0311, 0.0308, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 20:42:46,787 INFO [train.py:968] (1/2) Epoch 11, batch 27100, giga_loss[loss=0.2892, simple_loss=0.3574, pruned_loss=0.1105, over 28540.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3919, pruned_loss=0.1374, over 5641254.00 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3748, pruned_loss=0.1263, over 5699706.66 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3927, pruned_loss=0.1375, over 5646647.94 frames. ], batch size: 78, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 20:43:30,904 INFO [optim.py:369] (1/2) 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 20:43:36,840 INFO [train.py:968] (1/2) Epoch 11, batch 27150, giga_loss[loss=0.3803, simple_loss=0.4158, pruned_loss=0.1724, over 26543.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3899, pruned_loss=0.1355, over 5650313.07 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3748, pruned_loss=0.1262, over 5701958.40 frames. ], giga_tot_loss[loss=0.3312, simple_loss=0.3907, pruned_loss=0.1358, over 5651982.81 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 20:44:22,643 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-05 20:44:24,263 INFO [train.py:968] (1/2) Epoch 11, batch 27200, giga_loss[loss=0.3412, simple_loss=0.3819, pruned_loss=0.1502, over 23740.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3896, pruned_loss=0.1343, over 5647035.87 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3748, pruned_loss=0.1263, over 5703782.49 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3904, pruned_loss=0.1346, over 5646261.49 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:45:07,579 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 27250, giga_loss[loss=0.2978, simple_loss=0.3715, pruned_loss=0.1121, over 28853.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3887, pruned_loss=0.132, over 5660945.57 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3747, pruned_loss=0.1263, over 5704979.50 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3896, pruned_loss=0.1323, over 5658535.49 frames. ], batch size: 199, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:45:54,743 INFO [train.py:968] (1/2) Epoch 11, batch 27300, giga_loss[loss=0.3321, simple_loss=0.4001, pruned_loss=0.132, over 28704.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.388, pruned_loss=0.1316, over 5648828.70 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3747, pruned_loss=0.1263, over 5692596.89 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3894, pruned_loss=0.1321, over 5656394.57 frames. ], batch size: 242, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:46:11,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=482855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:46:42,565 INFO [optim.py:369] (1/2) 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 20:46:46,277 INFO [train.py:968] (1/2) Epoch 11, batch 27350, giga_loss[loss=0.3049, simple_loss=0.3782, pruned_loss=0.1158, over 28907.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.3891, pruned_loss=0.133, over 5659155.53 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1266, over 5695954.84 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3902, pruned_loss=0.1332, over 5661521.43 frames. ], batch size: 174, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:47:05,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2478, 2.9388, 1.3428, 1.4156], device='cuda:1'), covar=tensor([0.0967, 0.0375, 0.0887, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0517, 0.0343, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 20:47:31,051 INFO [train.py:968] (1/2) Epoch 11, batch 27400, giga_loss[loss=0.2922, simple_loss=0.3565, pruned_loss=0.114, over 28235.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3878, pruned_loss=0.1324, over 5671010.89 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1268, over 5700594.55 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3885, pruned_loss=0.1326, over 5667821.02 frames. ], batch size: 77, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:48:05,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-05 20:48:16,516 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 27450, giga_loss[loss=0.3809, simple_loss=0.4102, pruned_loss=0.1759, over 26609.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3861, pruned_loss=0.1332, over 5644960.37 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3759, pruned_loss=0.1271, over 5692621.53 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3865, pruned_loss=0.1331, over 5648656.93 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:48:23,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-05 20:48:29,580 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=482998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:48:32,334 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=483001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:49:02,215 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=483030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:49:09,587 INFO [train.py:968] (1/2) Epoch 11, batch 27500, giga_loss[loss=0.319, simple_loss=0.3808, pruned_loss=0.1286, over 28673.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3843, pruned_loss=0.1326, over 5633661.81 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3756, pruned_loss=0.1269, over 5688394.84 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3852, pruned_loss=0.133, over 5639989.23 frames. ], batch size: 262, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:49:53,513 INFO [optim.py:369] (1/2) 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 20:49:56,670 INFO [train.py:968] (1/2) Epoch 11, batch 27550, giga_loss[loss=0.319, simple_loss=0.3782, pruned_loss=0.1299, over 28641.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3828, pruned_loss=0.1324, over 5632139.48 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3756, pruned_loss=0.127, over 5674431.83 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3837, pruned_loss=0.1327, over 5647836.35 frames. ], batch size: 307, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:50:00,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 20:50:31,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4694, 2.2260, 1.6700, 0.6398], device='cuda:1'), covar=tensor([0.4124, 0.2082, 0.3073, 0.4417], device='cuda:1'), in_proj_covar=tensor([0.1574, 0.1496, 0.1504, 0.1282], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 20:50:42,431 INFO [train.py:968] (1/2) Epoch 11, batch 27600, giga_loss[loss=0.2883, simple_loss=0.3606, pruned_loss=0.108, over 28841.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3824, pruned_loss=0.1326, over 5636722.32 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3758, pruned_loss=0.1272, over 5676768.60 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.383, pruned_loss=0.1327, over 5646562.32 frames. ], batch size: 145, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 20:51:23,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 20:51:23,320 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 27650, giga_loss[loss=0.2928, simple_loss=0.364, pruned_loss=0.1108, over 29021.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3801, pruned_loss=0.1307, over 5646330.60 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3758, pruned_loss=0.1272, over 5681035.06 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3806, pruned_loss=0.1309, over 5649402.30 frames. ], batch size: 128, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:52:09,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1551, 1.5009, 1.4410, 1.0365], device='cuda:1'), covar=tensor([0.1655, 0.2439, 0.1356, 0.1521], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0704, 0.0874, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 20:52:15,938 INFO [train.py:968] (1/2) Epoch 11, batch 27700, giga_loss[loss=0.2962, simple_loss=0.37, pruned_loss=0.1113, over 28642.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.375, pruned_loss=0.1251, over 5651639.58 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3759, pruned_loss=0.1272, over 5682541.82 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3754, pruned_loss=0.1252, over 5652361.74 frames. ], batch size: 242, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:52:26,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4625, 1.4043, 1.6072, 1.1603], device='cuda:1'), covar=tensor([0.1832, 0.3159, 0.1529, 0.1780], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0704, 0.0873, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 20:53:03,157 INFO [optim.py:369] (1/2) 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 20:53:06,203 INFO [train.py:968] (1/2) Epoch 11, batch 27750, libri_loss[loss=0.3678, simple_loss=0.4219, pruned_loss=0.1569, over 29647.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3748, pruned_loss=0.1244, over 5658974.15 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1273, over 5686966.39 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3748, pruned_loss=0.1243, over 5654790.53 frames. ], batch size: 91, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:53:21,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2532, 1.5311, 1.2535, 0.9566], device='cuda:1'), covar=tensor([0.2504, 0.2311, 0.2644, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1297, 0.0963, 0.1150, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 20:53:33,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4006, 2.9986, 1.4213, 1.4909], device='cuda:1'), covar=tensor([0.0926, 0.0337, 0.0854, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0516, 0.0342, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 20:53:57,138 INFO [train.py:968] (1/2) Epoch 11, batch 27800, giga_loss[loss=0.2624, simple_loss=0.3341, pruned_loss=0.09535, over 28891.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.1239, over 5652029.82 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3762, pruned_loss=0.1273, over 5690006.01 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.373, pruned_loss=0.1238, over 5645515.89 frames. ], batch size: 186, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:54:33,575 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=483370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 20:54:46,586 INFO [optim.py:369] (1/2) 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 20:54:51,818 INFO [train.py:968] (1/2) Epoch 11, batch 27850, giga_loss[loss=0.3915, simple_loss=0.4219, pruned_loss=0.1806, over 26717.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3697, pruned_loss=0.1223, over 5665848.15 frames. ], libri_tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1275, over 5692618.75 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.122, over 5658042.92 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:55:39,905 INFO [train.py:968] (1/2) Epoch 11, batch 27900, giga_loss[loss=0.3108, simple_loss=0.372, pruned_loss=0.1248, over 28928.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3722, pruned_loss=0.1235, over 5670489.22 frames. ], libri_tot_loss[loss=0.3164, simple_loss=0.377, pruned_loss=0.1279, over 5692440.80 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3712, pruned_loss=0.1228, over 5663681.85 frames. ], batch size: 106, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:56:19,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3338, 3.2398, 1.5811, 1.3968], device='cuda:1'), covar=tensor([0.0940, 0.0280, 0.0860, 0.1398], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0514, 0.0340, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 20:56:20,219 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 27950, giga_loss[loss=0.4244, simple_loss=0.4433, pruned_loss=0.2027, over 26475.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3747, pruned_loss=0.1252, over 5661806.42 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3774, pruned_loss=0.128, over 5698593.01 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3734, pruned_loss=0.1244, over 5650050.42 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:56:29,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7601, 2.0225, 1.2448, 1.5759], device='cuda:1'), covar=tensor([0.0833, 0.0514, 0.1091, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0441, 0.0501, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 20:56:57,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4741, 2.0889, 1.5555, 0.6872], device='cuda:1'), covar=tensor([0.3686, 0.1964, 0.2671, 0.4168], device='cuda:1'), in_proj_covar=tensor([0.1561, 0.1480, 0.1490, 0.1270], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 20:57:12,756 INFO [train.py:968] (1/2) Epoch 11, batch 28000, libri_loss[loss=0.3608, simple_loss=0.4173, pruned_loss=0.1521, over 29303.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.375, pruned_loss=0.1252, over 5663939.34 frames. ], libri_tot_loss[loss=0.3166, simple_loss=0.3774, pruned_loss=0.1279, over 5702467.22 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5650034.49 frames. ], batch size: 94, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 20:57:57,130 INFO [optim.py:369] (1/2) 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 20:57:58,921 INFO [train.py:968] (1/2) Epoch 11, batch 28050, giga_loss[loss=0.3195, simple_loss=0.3792, pruned_loss=0.1299, over 28639.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3762, pruned_loss=0.1264, over 5656174.71 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3778, pruned_loss=0.1282, over 5696713.91 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.375, pruned_loss=0.1256, over 5648481.73 frames. ], batch size: 336, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:58:05,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7196, 1.7823, 1.3169, 1.4062], device='cuda:1'), covar=tensor([0.0754, 0.0525, 0.0987, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0439, 0.0499, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 20:58:22,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-05 20:58:46,460 INFO [train.py:968] (1/2) Epoch 11, batch 28100, giga_loss[loss=0.3171, simple_loss=0.3882, pruned_loss=0.123, over 28659.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3773, pruned_loss=0.1271, over 5663658.66 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5697385.34 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3761, pruned_loss=0.1264, over 5656925.17 frames. ], batch size: 307, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:58:46,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6156, 3.4366, 3.2579, 1.5845], device='cuda:1'), covar=tensor([0.0735, 0.0807, 0.0801, 0.2409], device='cuda:1'), in_proj_covar=tensor([0.1059, 0.0996, 0.0872, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 20:59:33,885 INFO [optim.py:369] (1/2) 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 20:59:37,148 INFO [train.py:968] (1/2) Epoch 11, batch 28150, giga_loss[loss=0.2909, simple_loss=0.3682, pruned_loss=0.1068, over 28898.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3788, pruned_loss=0.1278, over 5662789.96 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5695125.41 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3779, pruned_loss=0.1273, over 5659408.29 frames. ], batch size: 199, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 20:59:40,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3301, 2.8503, 1.5032, 1.3702], device='cuda:1'), covar=tensor([0.0827, 0.0344, 0.0738, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0516, 0.0341, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 21:00:25,834 INFO [train.py:968] (1/2) Epoch 11, batch 28200, giga_loss[loss=0.3373, simple_loss=0.3983, pruned_loss=0.1381, over 28746.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3802, pruned_loss=0.1287, over 5664591.45 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5697301.66 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3794, pruned_loss=0.1283, over 5659781.91 frames. ], batch size: 284, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:00:34,546 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=483745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:00:34,635 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:01:16,463 INFO [train.py:968] (1/2) Epoch 11, batch 28250, libri_loss[loss=0.2991, simple_loss=0.3689, pruned_loss=0.1147, over 29233.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3807, pruned_loss=0.1297, over 5650422.94 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3782, pruned_loss=0.1283, over 5699188.10 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.38, pruned_loss=0.1294, over 5644372.85 frames. ], batch size: 94, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:01:18,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2056, 1.5176, 1.3249, 1.1193], device='cuda:1'), covar=tensor([0.2405, 0.1932, 0.1403, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.1712, 0.1617, 0.1592, 0.1689], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 21:02:09,841 INFO [train.py:968] (1/2) Epoch 11, batch 28300, libri_loss[loss=0.2507, simple_loss=0.3165, pruned_loss=0.09242, over 27379.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3816, pruned_loss=0.1296, over 5654175.67 frames. ], libri_tot_loss[loss=0.3177, simple_loss=0.3783, pruned_loss=0.1285, over 5700329.99 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.381, pruned_loss=0.1292, over 5647827.80 frames. ], batch size: 60, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:02:52,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4752, 1.9490, 1.5733, 0.8834], device='cuda:1'), covar=tensor([0.2774, 0.1993, 0.2286, 0.3066], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1492, 0.1496, 0.1279], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 21:02:55,843 INFO [optim.py:369] (1/2) 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 21:03:00,279 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 28350, giga_loss[loss=0.3083, simple_loss=0.3754, pruned_loss=0.1206, over 28564.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3809, pruned_loss=0.1282, over 5645468.96 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3782, pruned_loss=0.1288, over 5687080.08 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3807, pruned_loss=0.1277, over 5651116.35 frames. ], batch size: 71, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:03:02,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=483891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:03:21,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3547, 1.6721, 1.2422, 1.5370], device='cuda:1'), covar=tensor([0.2316, 0.2255, 0.2587, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.0959, 0.1147, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 21:03:31,044 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=483920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:03:45,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4685, 2.1870, 1.6499, 0.6475], device='cuda:1'), covar=tensor([0.3860, 0.2166, 0.3113, 0.4234], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1496, 0.1498, 0.1279], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 21:03:46,109 INFO [train.py:968] (1/2) Epoch 11, batch 28400, giga_loss[loss=0.3199, simple_loss=0.3794, pruned_loss=0.1302, over 28839.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.38, pruned_loss=0.1281, over 5659267.05 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3776, pruned_loss=0.1283, over 5692159.22 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3805, pruned_loss=0.1281, over 5658056.78 frames. ], batch size: 119, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:04:42,330 INFO [optim.py:369] (1/2) 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 21:04:43,581 INFO [train.py:968] (1/2) Epoch 11, batch 28450, giga_loss[loss=0.2845, simple_loss=0.3534, pruned_loss=0.1078, over 29026.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3782, pruned_loss=0.1276, over 5662629.27 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3773, pruned_loss=0.1281, over 5691841.42 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3788, pruned_loss=0.1277, over 5661843.30 frames. ], batch size: 136, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:04:55,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5114, 3.4646, 1.6399, 1.5085], device='cuda:1'), covar=tensor([0.0916, 0.0322, 0.0837, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0519, 0.0341, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 21:05:40,361 INFO [train.py:968] (1/2) Epoch 11, batch 28500, giga_loss[loss=0.3013, simple_loss=0.3633, pruned_loss=0.1196, over 28593.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3771, pruned_loss=0.1274, over 5670311.72 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3776, pruned_loss=0.1282, over 5694605.94 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3774, pruned_loss=0.1274, over 5666646.20 frames. ], batch size: 242, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:06:27,348 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 28550, giga_loss[loss=0.2999, simple_loss=0.3626, pruned_loss=0.1186, over 28967.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3764, pruned_loss=0.1272, over 5671329.74 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1284, over 5695374.23 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3763, pruned_loss=0.127, over 5667170.56 frames. ], batch size: 112, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:06:29,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4403, 1.6051, 1.5648, 1.4484], device='cuda:1'), covar=tensor([0.1221, 0.1445, 0.1665, 0.1480], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0734, 0.0674, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 21:06:55,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=484120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:07:13,353 INFO [train.py:968] (1/2) Epoch 11, batch 28600, libri_loss[loss=0.298, simple_loss=0.3695, pruned_loss=0.1133, over 29740.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1272, over 5672771.46 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3773, pruned_loss=0.1279, over 5700607.82 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3763, pruned_loss=0.1275, over 5664185.60 frames. ], batch size: 87, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:07:33,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-05 21:07:51,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6795, 1.7890, 1.2317, 1.4099], device='cuda:1'), covar=tensor([0.0786, 0.0637, 0.0968, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0443, 0.0500, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-05 21:07:58,173 INFO [optim.py:369] (1/2) 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 21:07:59,014 INFO [train.py:968] (1/2) Epoch 11, batch 28650, giga_loss[loss=0.2951, simple_loss=0.3582, pruned_loss=0.1159, over 28558.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3765, pruned_loss=0.1281, over 5669058.38 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3771, pruned_loss=0.1277, over 5708265.54 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3771, pruned_loss=0.1285, over 5653977.94 frames. ], batch size: 71, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:08:45,209 INFO [train.py:968] (1/2) Epoch 11, batch 28700, giga_loss[loss=0.3134, simple_loss=0.3794, pruned_loss=0.1237, over 28721.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3771, pruned_loss=0.1289, over 5666959.03 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3768, pruned_loss=0.1277, over 5711889.37 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3778, pruned_loss=0.1292, over 5650527.76 frames. ], batch size: 284, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:09:08,232 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=484263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:09:10,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=484266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:09:30,807 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 28750, giga_loss[loss=0.3632, simple_loss=0.4168, pruned_loss=0.1548, over 28675.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3791, pruned_loss=0.1305, over 5667928.52 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3762, pruned_loss=0.1273, over 5712451.63 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3802, pruned_loss=0.1311, over 5653822.44 frames. ], batch size: 242, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:09:38,410 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=484295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:10:20,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2540, 1.5813, 1.2835, 0.9839], device='cuda:1'), covar=tensor([0.2168, 0.2048, 0.2327, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.0961, 0.1150, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 21:10:21,850 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 11, batch 28800, giga_loss[loss=0.3434, simple_loss=0.3964, pruned_loss=0.1452, over 28530.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3793, pruned_loss=0.131, over 5654895.19 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3762, pruned_loss=0.1273, over 5716101.76 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3802, pruned_loss=0.1316, over 5639421.77 frames. ], batch size: 307, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:11:07,492 INFO [optim.py:369] (1/2) 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 21:11:09,243 INFO [train.py:968] (1/2) Epoch 11, batch 28850, giga_loss[loss=0.3922, simple_loss=0.415, pruned_loss=0.1848, over 23732.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3805, pruned_loss=0.1327, over 5643262.94 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.3763, pruned_loss=0.1273, over 5706376.44 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3812, pruned_loss=0.1333, over 5639459.47 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:11:53,705 INFO [train.py:968] (1/2) Epoch 11, batch 28900, giga_loss[loss=0.3195, simple_loss=0.3865, pruned_loss=0.1263, over 28966.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3796, pruned_loss=0.1318, over 5656305.43 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3763, pruned_loss=0.1272, over 5711622.44 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3802, pruned_loss=0.1325, over 5647114.60 frames. ], batch size: 213, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:12:42,216 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 28950, giga_loss[loss=0.3605, simple_loss=0.4111, pruned_loss=0.1549, over 28825.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3803, pruned_loss=0.132, over 5648822.99 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3759, pruned_loss=0.1267, over 5714565.48 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3813, pruned_loss=0.1331, over 5637121.95 frames. ], batch size: 174, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:13:27,506 INFO [train.py:968] (1/2) Epoch 11, batch 29000, giga_loss[loss=0.3395, simple_loss=0.3926, pruned_loss=0.1432, over 28268.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3806, pruned_loss=0.1311, over 5661557.75 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 5719618.16 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3817, pruned_loss=0.1322, over 5646687.22 frames. ], batch size: 368, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:14:15,100 INFO [optim.py:369] (1/2) 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 21:14:16,519 INFO [train.py:968] (1/2) Epoch 11, batch 29050, giga_loss[loss=0.3088, simple_loss=0.3752, pruned_loss=0.1212, over 28921.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3819, pruned_loss=0.132, over 5665782.67 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3757, pruned_loss=0.1266, over 5720662.07 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3828, pruned_loss=0.1329, over 5653082.05 frames. ], batch size: 227, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:14:20,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8684, 4.7177, 4.4541, 2.4004], device='cuda:1'), covar=tensor([0.0514, 0.0665, 0.0742, 0.1732], device='cuda:1'), in_proj_covar=tensor([0.1060, 0.0999, 0.0878, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 21:15:03,620 INFO [train.py:968] (1/2) Epoch 11, batch 29100, giga_loss[loss=0.413, simple_loss=0.4284, pruned_loss=0.1988, over 26603.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3833, pruned_loss=0.1336, over 5674462.06 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3759, pruned_loss=0.1268, over 5722788.43 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3839, pruned_loss=0.1343, over 5661719.42 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:15:41,313 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5644, 4.3850, 4.1503, 2.0853], device='cuda:1'), covar=tensor([0.0540, 0.0714, 0.0742, 0.1930], device='cuda:1'), in_proj_covar=tensor([0.1065, 0.1005, 0.0881, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 21:15:44,370 INFO [optim.py:369] (1/2) 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 21:15:45,492 INFO [train.py:968] (1/2) Epoch 11, batch 29150, giga_loss[loss=0.2803, simple_loss=0.3511, pruned_loss=0.1048, over 28768.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3823, pruned_loss=0.1326, over 5680604.58 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3754, pruned_loss=0.1264, over 5725882.09 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3834, pruned_loss=0.1336, over 5666330.26 frames. ], batch size: 112, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:15:56,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2915, 1.5559, 1.4395, 1.1739], device='cuda:1'), covar=tensor([0.1771, 0.1469, 0.1133, 0.1608], device='cuda:1'), in_proj_covar=tensor([0.1701, 0.1608, 0.1575, 0.1681], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 21:16:25,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3954, 1.6249, 1.2634, 1.5859], device='cuda:1'), covar=tensor([0.2473, 0.2439, 0.2824, 0.2105], device='cuda:1'), in_proj_covar=tensor([0.1295, 0.0958, 0.1147, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 21:16:38,166 INFO [train.py:968] (1/2) Epoch 11, batch 29200, giga_loss[loss=0.3327, simple_loss=0.4041, pruned_loss=0.1307, over 29013.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3845, pruned_loss=0.1335, over 5673779.84 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3753, pruned_loss=0.1264, over 5726565.46 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3856, pruned_loss=0.1344, over 5661329.23 frames. ], batch size: 155, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:16:45,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4474, 1.6493, 1.5745, 1.4573], device='cuda:1'), covar=tensor([0.1624, 0.1865, 0.1925, 0.1892], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0739, 0.0678, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 21:16:51,119 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-05 21:17:27,485 INFO [optim.py:369] (1/2) 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 21:17:28,182 INFO [train.py:968] (1/2) Epoch 11, batch 29250, giga_loss[loss=0.3008, simple_loss=0.3709, pruned_loss=0.1153, over 29007.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3838, pruned_loss=0.1321, over 5668209.71 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3753, pruned_loss=0.1265, over 5729339.04 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3849, pruned_loss=0.1329, over 5653819.73 frames. ], batch size: 136, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:17:31,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-05 21:18:11,691 INFO [train.py:968] (1/2) Epoch 11, batch 29300, giga_loss[loss=0.3055, simple_loss=0.3662, pruned_loss=0.1225, over 27867.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3807, pruned_loss=0.1297, over 5676288.62 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.375, pruned_loss=0.1264, over 5731566.81 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.382, pruned_loss=0.1306, over 5661569.85 frames. ], batch size: 412, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:18:12,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3720, 1.5583, 1.4633, 1.2758], device='cuda:1'), covar=tensor([0.1991, 0.1838, 0.1366, 0.1853], device='cuda:1'), in_proj_covar=tensor([0.1707, 0.1610, 0.1579, 0.1684], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 21:18:48,184 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:18:55,471 INFO [train.py:968] (1/2) Epoch 11, batch 29350, giga_loss[loss=0.3277, simple_loss=0.3917, pruned_loss=0.1318, over 28921.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3808, pruned_loss=0.1304, over 5668219.67 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1265, over 5734818.34 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3821, pruned_loss=0.1311, over 5651806.25 frames. ], batch size: 174, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:18:56,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=484891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:19:00,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3142, 1.5479, 1.2307, 1.3367], device='cuda:1'), covar=tensor([0.2098, 0.2015, 0.2228, 0.1808], device='cuda:1'), in_proj_covar=tensor([0.1297, 0.0960, 0.1149, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 21:19:21,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-05 21:19:43,631 INFO [train.py:968] (1/2) Epoch 11, batch 29400, giga_loss[loss=0.3352, simple_loss=0.3932, pruned_loss=0.1386, over 28606.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3821, pruned_loss=0.1309, over 5663862.37 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3753, pruned_loss=0.1266, over 5726255.53 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.383, pruned_loss=0.1314, over 5656819.63 frames. ], batch size: 307, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:20:32,506 INFO [optim.py:369] (1/2) 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 21:20:32,518 INFO [train.py:968] (1/2) Epoch 11, batch 29450, giga_loss[loss=0.2696, simple_loss=0.3467, pruned_loss=0.09625, over 28941.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3827, pruned_loss=0.1316, over 5665270.85 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3753, pruned_loss=0.1266, over 5731048.43 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3835, pruned_loss=0.1321, over 5653756.12 frames. ], batch size: 145, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:20:50,139 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:21:09,276 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 29500, giga_loss[loss=0.3381, simple_loss=0.3942, pruned_loss=0.141, over 29059.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3824, pruned_loss=0.1323, over 5668533.32 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1266, over 5732854.40 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3833, pruned_loss=0.1328, over 5656974.03 frames. ], batch size: 136, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:21:41,390 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.03 vs. limit=2.0
+2023-03-05 21:22:03,915 INFO [train.py:968] (1/2) Epoch 11, batch 29550, giga_loss[loss=0.2935, simple_loss=0.3576, pruned_loss=0.1147, over 28551.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3835, pruned_loss=0.1339, over 5637328.32 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1268, over 5710939.75 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3842, pruned_loss=0.1344, over 5645264.68 frames. ], batch size: 78, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:22:04,566 INFO [optim.py:369] (1/2) 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 21:22:35,638 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-05 21:22:37,716 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:22:49,854 INFO [train.py:968] (1/2) Epoch 11, batch 29600, libri_loss[loss=0.2864, simple_loss=0.3508, pruned_loss=0.111, over 29368.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3843, pruned_loss=0.1341, over 5653858.43 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1268, over 5714790.97 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3851, pruned_loss=0.1347, over 5655029.59 frames. ], batch size: 67, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:23:27,933 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 21:23:38,910 INFO [train.py:968] (1/2) Epoch 11, batch 29650, giga_loss[loss=0.355, simple_loss=0.4044, pruned_loss=0.1528, over 27913.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3858, pruned_loss=0.1357, over 5643375.70 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1271, over 5716047.36 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3863, pruned_loss=0.136, over 5641713.99 frames. ], batch size: 412, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:23:39,490 INFO [optim.py:369] (1/2) 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 21:23:58,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.67 vs. limit=5.0
+2023-03-05 21:24:19,373 INFO [train.py:968] (1/2) Epoch 11, batch 29700, libri_loss[loss=0.2789, simple_loss=0.3443, pruned_loss=0.1068, over 29591.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3845, pruned_loss=0.1338, over 5649245.67 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.3761, pruned_loss=0.1272, over 5696925.91 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3852, pruned_loss=0.1343, over 5661491.49 frames. ], batch size: 77, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:24:38,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:24:48,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:25:09,451 INFO [train.py:968] (1/2) Epoch 11, batch 29750, giga_loss[loss=0.3253, simple_loss=0.3949, pruned_loss=0.1279, over 28443.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3839, pruned_loss=0.1328, over 5651253.31 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3758, pruned_loss=0.127, over 5698836.43 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.3848, pruned_loss=0.1335, over 5658815.97 frames. ], batch size: 71, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:25:10,439 INFO [optim.py:369] (1/2) 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 21:25:11,571 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:25:55,716 INFO [train.py:968] (1/2) Epoch 11, batch 29800, giga_loss[loss=0.2967, simple_loss=0.3665, pruned_loss=0.1135, over 28845.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.383, pruned_loss=0.132, over 5654368.96 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.375, pruned_loss=0.1265, over 5700085.65 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3846, pruned_loss=0.1332, over 5658212.69 frames. ], batch size: 285, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:26:05,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3654, 1.6714, 1.6244, 1.2064], device='cuda:1'), covar=tensor([0.1447, 0.2162, 0.1204, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0703, 0.0870, 0.0772], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 21:26:34,413 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:38,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:45,465 INFO [train.py:968] (1/2) Epoch 11, batch 29850, giga_loss[loss=0.3355, simple_loss=0.3965, pruned_loss=0.1373, over 28680.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3806, pruned_loss=0.1304, over 5656267.88 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1266, over 5700511.49 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3818, pruned_loss=0.1312, over 5658562.37 frames. ], batch size: 284, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:26:46,205 INFO [optim.py:369] (1/2) 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 21:26:56,220 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:56,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:26:58,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:04,479 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:06,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:22,142 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:30,713 INFO [train.py:968] (1/2) Epoch 11, batch 29900, giga_loss[loss=0.2754, simple_loss=0.3421, pruned_loss=0.1044, over 28950.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3805, pruned_loss=0.1307, over 5649216.06 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3756, pruned_loss=0.1268, over 5695741.11 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3812, pruned_loss=0.1313, over 5654552.09 frames. ], batch size: 112, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:27:33,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:27:42,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2478, 1.5351, 1.2833, 1.0963], device='cuda:1'), covar=tensor([0.1840, 0.1782, 0.1755, 0.1638], device='cuda:1'), in_proj_covar=tensor([0.1294, 0.0960, 0.1149, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 21:27:59,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8808, 2.4137, 1.9431, 1.7126], device='cuda:1'), covar=tensor([0.2521, 0.1667, 0.1741, 0.1997], device='cuda:1'), in_proj_covar=tensor([0.1719, 0.1620, 0.1585, 0.1684], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 21:28:17,112 INFO [train.py:968] (1/2) Epoch 11, batch 29950, giga_loss[loss=0.3514, simple_loss=0.3957, pruned_loss=0.1535, over 26455.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3778, pruned_loss=0.1292, over 5652584.25 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.3759, pruned_loss=0.1269, over 5699047.69 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3781, pruned_loss=0.1296, over 5653305.25 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 2.0
+2023-03-05 21:28:18,586 INFO [optim.py:369] (1/2) 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 21:28:22,880 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:28:50,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:28:53,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:04,824 INFO [train.py:968] (1/2) Epoch 11, batch 30000, giga_loss[loss=0.2656, simple_loss=0.3384, pruned_loss=0.09639, over 28976.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3736, pruned_loss=0.1268, over 5667681.46 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3758, pruned_loss=0.1269, over 5697742.95 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3739, pruned_loss=0.1271, over 5668576.18 frames. ], batch size: 164, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:29:04,824 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 21:29:13,228 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-05 21:29:16,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:18,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:25,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:42,110 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:49,366 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:29:54,325 INFO [train.py:968] (1/2) Epoch 11, batch 30050, giga_loss[loss=0.3306, simple_loss=0.3803, pruned_loss=0.1404, over 27915.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3725, pruned_loss=0.1265, over 5685107.31 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3761, pruned_loss=0.127, over 5703003.53 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3725, pruned_loss=0.1267, over 5680241.55 frames. ], batch size: 412, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:29:55,714 INFO [optim.py:369] (1/2) 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 21:30:27,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5455, 5.3862, 5.0751, 2.3796], device='cuda:1'), covar=tensor([0.0429, 0.0623, 0.0664, 0.1908], device='cuda:1'), in_proj_covar=tensor([0.1076, 0.1008, 0.0883, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 21:30:41,684 INFO [train.py:968] (1/2) Epoch 11, batch 30100, giga_loss[loss=0.3599, simple_loss=0.4008, pruned_loss=0.1595, over 24007.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3725, pruned_loss=0.1265, over 5684493.92 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5699275.84 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3722, pruned_loss=0.1266, over 5682794.61 frames. ], batch size: 705, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:30:43,769 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:30:45,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:30:56,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3490, 4.2087, 3.9455, 1.8995], device='cuda:1'), covar=tensor([0.0542, 0.0690, 0.0773, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1070, 0.1003, 0.0879, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 21:31:07,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:31:14,206 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:31:26,754 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 21:31:30,144 INFO [train.py:968] (1/2) Epoch 11, batch 30150, giga_loss[loss=0.2848, simple_loss=0.3565, pruned_loss=0.1065, over 28872.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3726, pruned_loss=0.1254, over 5683756.74 frames. ], libri_tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1269, over 5702324.78 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3725, pruned_loss=0.1256, over 5679390.79 frames. ], batch size: 199, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:31:33,019 INFO [optim.py:369] (1/2) 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 21:31:43,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-05 21:32:14,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2396, 1.6054, 1.4824, 1.1285], device='cuda:1'), covar=tensor([0.1674, 0.2442, 0.1427, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0699, 0.0869, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 21:32:20,906 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 30200, giga_loss[loss=0.3071, simple_loss=0.3649, pruned_loss=0.1246, over 26779.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3707, pruned_loss=0.1221, over 5673994.78 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5703325.90 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3703, pruned_loss=0.122, over 5669419.46 frames. ], batch size: 555, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:32:40,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:33:16,188 INFO [train.py:968] (1/2) Epoch 11, batch 30250, giga_loss[loss=0.2661, simple_loss=0.3514, pruned_loss=0.09039, over 28599.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3668, pruned_loss=0.1185, over 5666294.45 frames. ], libri_tot_loss[loss=0.3153, simple_loss=0.376, pruned_loss=0.1273, over 5699126.68 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3666, pruned_loss=0.1182, over 5665004.52 frames. ], batch size: 307, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:33:18,287 INFO [optim.py:369] (1/2) 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 21:33:18,525 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=485791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:33:36,908 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:33:40,664 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 30300, giga_loss[loss=0.2799, simple_loss=0.3515, pruned_loss=0.1041, over 28850.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.363, pruned_loss=0.1147, over 5657132.42 frames. ], libri_tot_loss[loss=0.3152, simple_loss=0.3757, pruned_loss=0.1273, over 5698054.32 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3629, pruned_loss=0.1143, over 5656709.77 frames. ], batch size: 285, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:34:08,722 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:34:54,792 INFO [train.py:968] (1/2) Epoch 11, batch 30350, giga_loss[loss=0.2539, simple_loss=0.3374, pruned_loss=0.08517, over 28808.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3598, pruned_loss=0.1113, over 5659496.40 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3752, pruned_loss=0.1271, over 5702188.34 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3599, pruned_loss=0.1108, over 5654745.95 frames. ], batch size: 186, lr: 2.92e-03, grad_scale: 4.0
+2023-03-05 21:34:56,457 INFO [optim.py:369] (1/2) 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 21:35:01,347 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=485895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:35:03,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3112, 1.8376, 1.3245, 0.6390], device='cuda:1'), covar=tensor([0.3404, 0.1856, 0.2832, 0.4025], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1476, 0.1488, 0.1269], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 21:35:04,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=485898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:35:24,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-05 21:35:36,345 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=485927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:35:49,453 INFO [train.py:968] (1/2) Epoch 11, batch 30400, giga_loss[loss=0.2726, simple_loss=0.3552, pruned_loss=0.09502, over 27984.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3577, pruned_loss=0.1082, over 5635745.64 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3751, pruned_loss=0.1272, over 5694186.88 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3577, pruned_loss=0.1077, over 5638116.75 frames. ], batch size: 412, lr: 2.92e-03, grad_scale: 8.0
+2023-03-05 21:36:09,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=485957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:36:37,194 INFO [train.py:968] (1/2) Epoch 11, batch 30450, giga_loss[loss=0.2677, simple_loss=0.3412, pruned_loss=0.09713, over 27963.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3579, pruned_loss=0.1085, over 5637842.38 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3745, pruned_loss=0.127, over 5696606.83 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3579, pruned_loss=0.1076, over 5635257.38 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:36:39,868 INFO [optim.py:369] (1/2) 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 21:36:46,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3060, 1.5002, 1.2436, 1.5581], device='cuda:1'), covar=tensor([0.0698, 0.0426, 0.0360, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0112, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:1')
+2023-03-05 21:36:51,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4462, 1.6655, 1.8142, 1.3823], device='cuda:1'), covar=tensor([0.1496, 0.2020, 0.1180, 0.1556], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0692, 0.0862, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 21:36:56,170 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:37:07,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8429, 3.6748, 3.4404, 1.9439], device='cuda:1'), covar=tensor([0.0576, 0.0746, 0.0795, 0.2269], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0990, 0.0862, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 21:37:23,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:37:26,168 INFO [train.py:968] (1/2) Epoch 11, batch 30500, giga_loss[loss=0.2514, simple_loss=0.3267, pruned_loss=0.08805, over 28739.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3553, pruned_loss=0.1071, over 5640756.11 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3736, pruned_loss=0.1266, over 5701091.00 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3557, pruned_loss=0.1062, over 5632845.15 frames. ], batch size: 92, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:38:15,096 INFO [train.py:968] (1/2) Epoch 11, batch 30550, libri_loss[loss=0.3157, simple_loss=0.37, pruned_loss=0.1307, over 29658.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3526, pruned_loss=0.1048, over 5645550.51 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3736, pruned_loss=0.1266, over 5702981.39 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3526, pruned_loss=0.1038, over 5636937.68 frames. ], batch size: 88, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:38:17,991 INFO [optim.py:369] (1/2) 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 21:38:26,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:30,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:36,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:53,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6248, 1.7555, 1.2387, 1.3746], device='cuda:1'), covar=tensor([0.0732, 0.0492, 0.0986, 0.0995], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0438, 0.0497, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 21:38:58,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:38:59,008 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-05 21:39:04,068 INFO [train.py:968] (1/2) Epoch 11, batch 30600, giga_loss[loss=0.2403, simple_loss=0.3298, pruned_loss=0.07542, over 28555.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3518, pruned_loss=0.1045, over 5647729.28 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3731, pruned_loss=0.1263, over 5706516.83 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3517, pruned_loss=0.1033, over 5636045.51 frames. ], batch size: 60, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:39:27,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:39:50,895 INFO [train.py:968] (1/2) Epoch 11, batch 30650, giga_loss[loss=0.2866, simple_loss=0.3601, pruned_loss=0.1065, over 28307.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3529, pruned_loss=0.1051, over 5644394.12 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3725, pruned_loss=0.1261, over 5700494.81 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3528, pruned_loss=0.1039, over 5638333.48 frames. ], batch size: 368, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:39:53,088 INFO [optim.py:369] (1/2) 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 21:40:31,722 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:40:39,640 INFO [train.py:968] (1/2) Epoch 11, batch 30700, giga_loss[loss=0.2554, simple_loss=0.3344, pruned_loss=0.08819, over 28958.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3506, pruned_loss=0.1028, over 5651160.50 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3724, pruned_loss=0.1264, over 5699796.67 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3502, pruned_loss=0.1012, over 5646122.86 frames. ], batch size: 186, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:40:54,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:40:56,968 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 30750, giga_loss[loss=0.2642, simple_loss=0.3414, pruned_loss=0.09352, over 28567.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3478, pruned_loss=0.1007, over 5648054.32 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3723, pruned_loss=0.1264, over 5704217.66 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3472, pruned_loss=0.09893, over 5639254.96 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:41:31,539 INFO [optim.py:369] (1/2) 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 21:41:48,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:41:52,557 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 30800, libri_loss[loss=0.2671, simple_loss=0.3359, pruned_loss=0.09916, over 29581.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3435, pruned_loss=0.09849, over 5645798.29 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3717, pruned_loss=0.1262, over 5707363.78 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3428, pruned_loss=0.09653, over 5634197.04 frames. ], batch size: 74, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:42:20,348 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:42:59,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:43:04,718 INFO [train.py:968] (1/2) Epoch 11, batch 30850, giga_loss[loss=0.2121, simple_loss=0.2808, pruned_loss=0.07165, over 23814.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3418, pruned_loss=0.09814, over 5651486.86 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3704, pruned_loss=0.1254, over 5711448.17 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3414, pruned_loss=0.09625, over 5636451.95 frames. ], batch size: 705, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:43:07,186 INFO [optim.py:369] (1/2) 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 21:43:26,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:43:55,684 INFO [train.py:968] (1/2) Epoch 11, batch 30900, giga_loss[loss=0.2851, simple_loss=0.3429, pruned_loss=0.1136, over 26668.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3407, pruned_loss=0.09784, over 5640046.42 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.37, pruned_loss=0.1251, over 5714301.65 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3401, pruned_loss=0.09604, over 5624261.99 frames. ], batch size: 555, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:44:27,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-05 21:44:51,963 INFO [train.py:968] (1/2) Epoch 11, batch 30950, libri_loss[loss=0.2805, simple_loss=0.341, pruned_loss=0.1101, over 29548.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3428, pruned_loss=0.09888, over 5637926.56 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3694, pruned_loss=0.1248, over 5717790.88 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3423, pruned_loss=0.09714, over 5620651.25 frames. ], batch size: 78, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:44:57,738 INFO [optim.py:369] (1/2) 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 21:45:18,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2936, 1.6461, 1.3542, 1.5399], device='cuda:1'), covar=tensor([0.0784, 0.0324, 0.0342, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0057, 0.0052, 0.0089], device='cuda:1')
+2023-03-05 21:45:21,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1490, 1.1835, 3.7240, 2.9770], device='cuda:1'), covar=tensor([0.1588, 0.2553, 0.0419, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0660, 0.0586, 0.0850, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 21:45:34,134 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:45:36,210 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 31000, giga_loss[loss=0.2603, simple_loss=0.3454, pruned_loss=0.08758, over 28545.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3457, pruned_loss=0.09929, over 5651694.74 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3687, pruned_loss=0.1244, over 5721503.61 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3453, pruned_loss=0.0977, over 5632908.72 frames. ], batch size: 336, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:46:07,200 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:46:09,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:46:11,031 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:46:46,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 31050, libri_loss[loss=0.3319, simple_loss=0.3781, pruned_loss=0.1428, over 29257.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3469, pruned_loss=0.09984, over 5665572.72 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3683, pruned_loss=0.1243, over 5722260.06 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3464, pruned_loss=0.098, over 5648383.08 frames. ], batch size: 97, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:46:57,260 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=486606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:47:55,324 INFO [train.py:968] (1/2) Epoch 11, batch 31100, giga_loss[loss=0.2268, simple_loss=0.293, pruned_loss=0.08028, over 24443.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3454, pruned_loss=0.09879, over 5665745.10 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3681, pruned_loss=0.1242, over 5721839.49 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3449, pruned_loss=0.09709, over 5651844.33 frames. ], batch size: 705, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:48:53,244 INFO [train.py:968] (1/2) Epoch 11, batch 31150, giga_loss[loss=0.2542, simple_loss=0.3357, pruned_loss=0.08635, over 28424.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3448, pruned_loss=0.09823, over 5659342.32 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.368, pruned_loss=0.1243, over 5716139.55 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3437, pruned_loss=0.09589, over 5651796.90 frames. ], batch size: 336, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:48:55,966 INFO [optim.py:369] (1/2) 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 21:49:53,320 INFO [train.py:968] (1/2) Epoch 11, batch 31200, giga_loss[loss=0.2747, simple_loss=0.348, pruned_loss=0.1007, over 28666.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3432, pruned_loss=0.09651, over 5656463.99 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.368, pruned_loss=0.1245, over 5709169.27 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3419, pruned_loss=0.09403, over 5654877.51 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:50:07,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=486749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:50:12,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=486752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:50:43,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4156, 1.6192, 1.3193, 1.6202], device='cuda:1'), covar=tensor([0.0732, 0.0310, 0.0343, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0112, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0057, 0.0052, 0.0089], device='cuda:1')
+2023-03-05 21:50:46,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=486781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:50:52,498 INFO [train.py:968] (1/2) Epoch 11, batch 31250, giga_loss[loss=0.2479, simple_loss=0.3251, pruned_loss=0.08533, over 28947.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3398, pruned_loss=0.09565, over 5666246.66 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3675, pruned_loss=0.1243, over 5711281.85 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3386, pruned_loss=0.09315, over 5661777.54 frames. ], batch size: 213, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:51:01,979 INFO [optim.py:369] (1/2) 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 21:51:21,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=486810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:51:32,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3165, 1.5838, 1.4726, 1.2299], device='cuda:1'), covar=tensor([0.1975, 0.1483, 0.1294, 0.1667], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1574, 0.1528, 0.1625], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 21:51:53,164 INFO [train.py:968] (1/2) Epoch 11, batch 31300, giga_loss[loss=0.2697, simple_loss=0.3467, pruned_loss=0.09637, over 28925.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3396, pruned_loss=0.09591, over 5663555.48 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.367, pruned_loss=0.124, over 5714221.38 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3385, pruned_loss=0.09348, over 5656234.90 frames. ], batch size: 186, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:52:49,098 INFO [train.py:968] (1/2) Epoch 11, batch 31350, giga_loss[loss=0.2599, simple_loss=0.3432, pruned_loss=0.08831, over 28950.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3398, pruned_loss=0.096, over 5652192.88 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3671, pruned_loss=0.1243, over 5698377.92 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3382, pruned_loss=0.09321, over 5659550.63 frames. ], batch size: 186, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:52:55,638 INFO [optim.py:369] (1/2) 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 21:53:47,747 INFO [train.py:968] (1/2) Epoch 11, batch 31400, giga_loss[loss=0.2484, simple_loss=0.3348, pruned_loss=0.081, over 28535.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3405, pruned_loss=0.09528, over 5653647.20 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3669, pruned_loss=0.1243, over 5699609.76 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3388, pruned_loss=0.0924, over 5657368.45 frames. ], batch size: 85, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:54:49,778 INFO [train.py:968] (1/2) Epoch 11, batch 31450, giga_loss[loss=0.2416, simple_loss=0.3127, pruned_loss=0.08531, over 27561.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3408, pruned_loss=0.09519, over 5662903.05 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3659, pruned_loss=0.124, over 5706498.56 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3394, pruned_loss=0.09223, over 5658483.92 frames. ], batch size: 472, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:54:50,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4796, 1.8315, 1.7976, 1.3626], device='cuda:1'), covar=tensor([0.1778, 0.2421, 0.1430, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.0828, 0.0687, 0.0866, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 21:54:56,024 INFO [optim.py:369] (1/2) 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 21:54:58,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5885, 2.2731, 1.5752, 0.7031], device='cuda:1'), covar=tensor([0.2968, 0.1636, 0.2794, 0.3595], device='cuda:1'), in_proj_covar=tensor([0.1546, 0.1468, 0.1478, 0.1262], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 21:55:36,684 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 31500, giga_loss[loss=0.2349, simple_loss=0.3169, pruned_loss=0.07644, over 28893.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3391, pruned_loss=0.0946, over 5672084.08 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3657, pruned_loss=0.1238, over 5712926.60 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.337, pruned_loss=0.0911, over 5660880.37 frames. ], batch size: 136, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:56:03,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-05 21:56:07,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7013, 1.7481, 1.1172, 1.3695], device='cuda:1'), covar=tensor([0.0748, 0.0554, 0.1099, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0438, 0.0501, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 21:56:48,468 INFO [train.py:968] (1/2) Epoch 11, batch 31550, libri_loss[loss=0.3054, simple_loss=0.3613, pruned_loss=0.1247, over 19905.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3409, pruned_loss=0.09611, over 5672427.76 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.365, pruned_loss=0.1234, over 5710813.71 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3389, pruned_loss=0.09249, over 5664759.00 frames. ], batch size: 186, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 21:56:53,578 INFO [optim.py:369] (1/2) 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 21:57:10,704 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 31600, libri_loss[loss=0.3266, simple_loss=0.3504, pruned_loss=0.1514, over 29673.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.344, pruned_loss=0.09615, over 5667284.02 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3644, pruned_loss=0.1232, over 5716000.93 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3424, pruned_loss=0.09276, over 5655312.44 frames. ], batch size: 69, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:58:13,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0714, 1.1932, 3.6600, 2.9988], device='cuda:1'), covar=tensor([0.1726, 0.2808, 0.0362, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0662, 0.0593, 0.0856, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 21:58:22,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-05 21:58:44,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 31650, giga_loss[loss=0.2128, simple_loss=0.2851, pruned_loss=0.07029, over 24306.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3453, pruned_loss=0.09454, over 5661786.10 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3639, pruned_loss=0.123, over 5716818.00 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.344, pruned_loss=0.09142, over 5650292.95 frames. ], batch size: 705, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 21:58:55,954 INFO [optim.py:369] (1/2) 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 21:59:26,982 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 21:59:46,228 INFO [train.py:968] (1/2) Epoch 11, batch 31700, giga_loss[loss=0.2416, simple_loss=0.333, pruned_loss=0.07506, over 28347.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3456, pruned_loss=0.09375, over 5654810.50 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.364, pruned_loss=0.1232, over 5709612.52 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09025, over 5650333.99 frames. ], batch size: 369, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:00:44,494 INFO [train.py:968] (1/2) Epoch 11, batch 31750, giga_loss[loss=0.2203, simple_loss=0.3156, pruned_loss=0.06254, over 28613.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3441, pruned_loss=0.09234, over 5648200.55 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3638, pruned_loss=0.123, over 5702141.01 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3425, pruned_loss=0.08897, over 5649966.26 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:00:49,496 INFO [optim.py:369] (1/2) 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 22:01:31,555 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:01:34,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:01:41,262 INFO [train.py:968] (1/2) Epoch 11, batch 31800, giga_loss[loss=0.2352, simple_loss=0.3166, pruned_loss=0.07688, over 28988.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3446, pruned_loss=0.09359, over 5647364.31 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3638, pruned_loss=0.1229, over 5697775.81 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3427, pruned_loss=0.08998, over 5652114.63 frames. ], batch size: 145, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:02:08,543 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:02:16,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3834, 2.1399, 1.5537, 0.5552], device='cuda:1'), covar=tensor([0.3178, 0.1924, 0.2887, 0.3762], device='cuda:1'), in_proj_covar=tensor([0.1558, 0.1482, 0.1489, 0.1271], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 22:02:52,492 INFO [train.py:968] (1/2) Epoch 11, batch 31850, giga_loss[loss=0.2609, simple_loss=0.3411, pruned_loss=0.09035, over 28985.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3439, pruned_loss=0.09424, over 5656135.34 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3635, pruned_loss=0.1229, over 5700155.77 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3423, pruned_loss=0.09093, over 5657010.16 frames. ], batch size: 186, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:02:59,504 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:03:31,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5621, 1.9008, 1.6187, 1.4011], device='cuda:1'), covar=tensor([0.2344, 0.1431, 0.1229, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.1662, 0.1564, 0.1516, 0.1616], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 22:04:11,609 INFO [train.py:968] (1/2) Epoch 11, batch 31900, giga_loss[loss=0.2271, simple_loss=0.3092, pruned_loss=0.07251, over 28957.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3432, pruned_loss=0.09404, over 5663887.34 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.363, pruned_loss=0.1225, over 5699534.27 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3422, pruned_loss=0.09136, over 5664666.53 frames. ], batch size: 186, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:04:31,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:04:57,106 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:05:11,280 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 31950, giga_loss[loss=0.2815, simple_loss=0.3455, pruned_loss=0.1088, over 26849.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3392, pruned_loss=0.09196, over 5670298.76 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3625, pruned_loss=0.1223, over 5704832.02 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3383, pruned_loss=0.08934, over 5665288.08 frames. ], batch size: 555, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:05:28,977 INFO [optim.py:369] (1/2) 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 22:05:36,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4781, 1.7775, 1.4586, 1.5159], device='cuda:1'), covar=tensor([0.2246, 0.2129, 0.2357, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.0956, 0.1154, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 22:06:24,309 INFO [train.py:968] (1/2) Epoch 11, batch 32000, giga_loss[loss=0.2449, simple_loss=0.3237, pruned_loss=0.08302, over 28927.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3378, pruned_loss=0.0914, over 5664923.35 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3622, pruned_loss=0.1223, over 5699438.24 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3368, pruned_loss=0.08876, over 5666022.32 frames. ], batch size: 213, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:06:37,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:06:39,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:07:12,564 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:07:23,948 INFO [train.py:968] (1/2) Epoch 11, batch 32050, giga_loss[loss=0.2119, simple_loss=0.2941, pruned_loss=0.06484, over 28534.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3375, pruned_loss=0.09205, over 5667965.34 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3621, pruned_loss=0.1224, over 5705779.43 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.336, pruned_loss=0.08872, over 5661974.54 frames. ], batch size: 71, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:07:35,582 INFO [optim.py:369] (1/2) 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 22:07:38,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:08:13,893 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:08:16,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:08:27,211 INFO [train.py:968] (1/2) Epoch 11, batch 32100, giga_loss[loss=0.2497, simple_loss=0.3374, pruned_loss=0.081, over 28774.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3423, pruned_loss=0.09442, over 5676299.96 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3614, pruned_loss=0.122, over 5709807.90 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3413, pruned_loss=0.09162, over 5666893.53 frames. ], batch size: 243, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:08:46,876 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:09:24,551 INFO [train.py:968] (1/2) Epoch 11, batch 32150, giga_loss[loss=0.2562, simple_loss=0.3306, pruned_loss=0.09087, over 29007.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3428, pruned_loss=0.09579, over 5676284.07 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3606, pruned_loss=0.1216, over 5714732.80 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3422, pruned_loss=0.09301, over 5663303.82 frames. ], batch size: 112, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:09:34,108 INFO [optim.py:369] (1/2) 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 22:10:26,440 INFO [train.py:968] (1/2) Epoch 11, batch 32200, giga_loss[loss=0.2914, simple_loss=0.3595, pruned_loss=0.1116, over 28917.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3422, pruned_loss=0.09658, over 5675420.64 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3601, pruned_loss=0.1214, over 5717320.97 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3416, pruned_loss=0.09379, over 5661813.62 frames. ], batch size: 213, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:10:28,537 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:10:30,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:11:07,880 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=487772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:11:26,426 INFO [train.py:968] (1/2) Epoch 11, batch 32250, giga_loss[loss=0.281, simple_loss=0.3573, pruned_loss=0.1023, over 28690.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3419, pruned_loss=0.0964, over 5676892.90 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3599, pruned_loss=0.1213, over 5720658.83 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3413, pruned_loss=0.0938, over 5662014.94 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:11:32,932 INFO [optim.py:369] (1/2) 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 22:12:12,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:12:27,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-05 22:12:29,950 INFO [train.py:968] (1/2) Epoch 11, batch 32300, giga_loss[loss=0.2397, simple_loss=0.3265, pruned_loss=0.07642, over 28646.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3424, pruned_loss=0.09607, over 5674558.05 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3598, pruned_loss=0.1214, over 5722327.68 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3415, pruned_loss=0.09307, over 5659591.14 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:12:37,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=487845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:13:39,863 INFO [train.py:968] (1/2) Epoch 11, batch 32350, giga_loss[loss=0.2746, simple_loss=0.3513, pruned_loss=0.09893, over 28760.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3441, pruned_loss=0.0962, over 5677147.04 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3597, pruned_loss=0.1215, over 5718091.52 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3428, pruned_loss=0.09293, over 5667846.69 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:13:51,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2266, 1.5239, 1.5123, 1.4448], device='cuda:1'), covar=tensor([0.1372, 0.1424, 0.1831, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0719, 0.0656, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-05 22:13:51,446 INFO [optim.py:369] (1/2) 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 22:14:56,603 INFO [train.py:968] (1/2) Epoch 11, batch 32400, giga_loss[loss=0.2246, simple_loss=0.3087, pruned_loss=0.07024, over 29014.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3424, pruned_loss=0.09504, over 5664208.88 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3598, pruned_loss=0.1215, over 5711782.77 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3412, pruned_loss=0.09214, over 5661520.36 frames. ], batch size: 155, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:15:24,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=487962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:33,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:37,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:56,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=487988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:15:57,240 INFO [train.py:968] (1/2) Epoch 11, batch 32450, giga_loss[loss=0.2142, simple_loss=0.2923, pruned_loss=0.06808, over 28085.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3405, pruned_loss=0.09603, over 5642688.39 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3601, pruned_loss=0.1221, over 5684392.19 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3385, pruned_loss=0.09216, over 5664364.35 frames. ], batch size: 412, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:15:59,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=487991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:16:05,763 INFO [optim.py:369] (1/2) 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 22:16:10,206 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=488001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:16:38,923 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=488020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:17:02,397 INFO [train.py:968] (1/2) Epoch 11, batch 32500, giga_loss[loss=0.2727, simple_loss=0.3409, pruned_loss=0.1023, over 26781.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.334, pruned_loss=0.09273, over 5643834.71 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.36, pruned_loss=0.122, over 5687376.24 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3323, pruned_loss=0.08933, over 5657715.56 frames. ], batch size: 555, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:18:01,268 INFO [train.py:968] (1/2) Epoch 11, batch 32550, giga_loss[loss=0.2482, simple_loss=0.3279, pruned_loss=0.08429, over 28968.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3339, pruned_loss=0.09266, over 5652231.18 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3592, pruned_loss=0.1216, over 5691267.74 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3326, pruned_loss=0.08966, over 5659038.70 frames. ], batch size: 227, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:18:13,504 INFO [optim.py:369] (1/2) 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 22:19:01,413 INFO [train.py:968] (1/2) Epoch 11, batch 32600, giga_loss[loss=0.2487, simple_loss=0.3322, pruned_loss=0.08258, over 29042.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3349, pruned_loss=0.09347, over 5648122.21 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3591, pruned_loss=0.1217, over 5694192.98 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3336, pruned_loss=0.09066, over 5650174.71 frames. ], batch size: 285, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:20:00,253 INFO [train.py:968] (1/2) Epoch 11, batch 32650, giga_loss[loss=0.2482, simple_loss=0.3396, pruned_loss=0.07834, over 28824.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3329, pruned_loss=0.09147, over 5652588.80 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3584, pruned_loss=0.1212, over 5698516.18 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.332, pruned_loss=0.08901, over 5649686.84 frames. ], batch size: 174, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:20:11,039 INFO [optim.py:369] (1/2) 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 22:21:05,381 INFO [train.py:968] (1/2) Epoch 11, batch 32700, giga_loss[loss=0.242, simple_loss=0.3205, pruned_loss=0.08179, over 28946.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3327, pruned_loss=0.09101, over 5661273.21 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3587, pruned_loss=0.1215, over 5699480.63 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3311, pruned_loss=0.0881, over 5656962.32 frames. ], batch size: 199, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:22:06,000 INFO [train.py:968] (1/2) Epoch 11, batch 32750, giga_loss[loss=0.3049, simple_loss=0.3607, pruned_loss=0.1245, over 26868.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3332, pruned_loss=0.09227, over 5660808.72 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3584, pruned_loss=0.1213, over 5697534.23 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3311, pruned_loss=0.08882, over 5657730.92 frames. ], batch size: 555, lr: 2.91e-03, grad_scale: 2.0
+2023-03-05 22:22:19,357 INFO [optim.py:369] (1/2) 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 22:23:11,430 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 32800, giga_loss[loss=0.2202, simple_loss=0.303, pruned_loss=0.06875, over 28432.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.332, pruned_loss=0.09071, over 5656961.10 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3586, pruned_loss=0.1216, over 5699615.22 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3298, pruned_loss=0.08728, over 5652158.25 frames. ], batch size: 60, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:24:20,542 INFO [train.py:968] (1/2) Epoch 11, batch 32850, giga_loss[loss=0.2549, simple_loss=0.3277, pruned_loss=0.09108, over 28950.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3311, pruned_loss=0.08994, over 5653322.23 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3583, pruned_loss=0.1214, over 5693504.37 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3293, pruned_loss=0.08703, over 5653827.03 frames. ], batch size: 199, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:24:29,722 INFO [optim.py:369] (1/2) 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 22:25:08,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3629, 1.6290, 1.2549, 1.5419], device='cuda:1'), covar=tensor([0.2531, 0.2383, 0.2721, 0.2047], device='cuda:1'), in_proj_covar=tensor([0.1297, 0.0958, 0.1153, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 22:25:16,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1071, 1.3938, 1.1902, 1.1032], device='cuda:1'), covar=tensor([0.1812, 0.1639, 0.1077, 0.1439], device='cuda:1'), in_proj_covar=tensor([0.1687, 0.1564, 0.1527, 0.1638], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 22:25:23,612 INFO [train.py:968] (1/2) Epoch 11, batch 32900, giga_loss[loss=0.2315, simple_loss=0.3142, pruned_loss=0.07436, over 28929.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3333, pruned_loss=0.09193, over 5648668.12 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3588, pruned_loss=0.1219, over 5683468.54 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3311, pruned_loss=0.08892, over 5656636.14 frames. ], batch size: 155, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:26:11,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4839, 1.5451, 1.2618, 1.2066], device='cuda:1'), covar=tensor([0.0647, 0.0307, 0.0763, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0434, 0.0498, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 22:26:11,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=488480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:26:14,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0828, 1.7089, 1.4369, 1.3796], device='cuda:1'), covar=tensor([0.0892, 0.0290, 0.0309, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:1')
+2023-03-05 22:26:15,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=488483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:26:19,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 32950, giga_loss[loss=0.2625, simple_loss=0.3474, pruned_loss=0.08878, over 28547.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3324, pruned_loss=0.09093, over 5642942.25 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3586, pruned_loss=0.1218, over 5679000.20 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3301, pruned_loss=0.08778, over 5652028.22 frames. ], batch size: 307, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:26:33,430 INFO [optim.py:369] (1/2) 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 22:26:49,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=488512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:27:00,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8752, 4.9784, 1.9637, 2.0498], device='cuda:1'), covar=tensor([0.0890, 0.0345, 0.0851, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0507, 0.0342, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 22:27:15,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3521, 3.3139, 1.4754, 1.4780], device='cuda:1'), covar=tensor([0.0950, 0.0269, 0.0897, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0507, 0.0342, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 22:27:18,046 INFO [train.py:968] (1/2) Epoch 11, batch 33000, giga_loss[loss=0.2456, simple_loss=0.3339, pruned_loss=0.07871, over 28901.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3344, pruned_loss=0.09055, over 5645484.66 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3583, pruned_loss=0.1217, over 5672401.38 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3323, pruned_loss=0.08734, over 5657410.05 frames. ], batch size: 284, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:27:18,046 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 22:27:25,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3423, 1.8095, 1.6731, 1.1807], device='cuda:1'), covar=tensor([0.1661, 0.2400, 0.1448, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0683, 0.0864, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 22:27:26,905 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-05 22:28:23,439 INFO [train.py:968] (1/2) Epoch 11, batch 33050, giga_loss[loss=0.2535, simple_loss=0.3397, pruned_loss=0.08364, over 28614.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.337, pruned_loss=0.0918, over 5638712.56 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3582, pruned_loss=0.1218, over 5666014.47 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3348, pruned_loss=0.08824, over 5653891.92 frames. ], batch size: 242, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:28:34,962 INFO [optim.py:369] (1/2) 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 22:28:44,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2078, 0.8511, 0.9454, 1.4205], device='cuda:1'), covar=tensor([0.0747, 0.0403, 0.0350, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0058, 0.0053, 0.0090], device='cuda:1')
+2023-03-05 22:29:08,500 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 33100, giga_loss[loss=0.2633, simple_loss=0.3452, pruned_loss=0.09069, over 28690.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3367, pruned_loss=0.09161, over 5631513.83 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.358, pruned_loss=0.1217, over 5664370.39 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3345, pruned_loss=0.08808, over 5644113.88 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:30:11,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2612, 3.3015, 1.4178, 1.3902], device='cuda:1'), covar=tensor([0.0982, 0.0307, 0.0938, 0.1398], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0508, 0.0343, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 22:30:12,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3996, 3.0846, 1.4315, 1.5118], device='cuda:1'), covar=tensor([0.0910, 0.0441, 0.0910, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0509, 0.0343, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 22:30:27,983 INFO [train.py:968] (1/2) Epoch 11, batch 33150, giga_loss[loss=0.23, simple_loss=0.3157, pruned_loss=0.07216, over 28908.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3369, pruned_loss=0.0917, over 5644724.71 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3584, pruned_loss=0.1218, over 5669632.26 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3344, pruned_loss=0.08819, over 5649369.20 frames. ], batch size: 145, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:30:30,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:30:37,752 INFO [optim.py:369] (1/2) 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 22:30:51,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3299, 1.5032, 1.1830, 1.4697], device='cuda:1'), covar=tensor([0.0761, 0.0333, 0.0345, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:1')
+2023-03-05 22:31:02,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4519, 1.7197, 1.3548, 1.5624], device='cuda:1'), covar=tensor([0.2261, 0.2109, 0.2353, 0.1948], device='cuda:1'), in_proj_covar=tensor([0.1297, 0.0959, 0.1155, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 22:31:13,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5452, 1.7532, 1.8586, 1.3865], device='cuda:1'), covar=tensor([0.1834, 0.2295, 0.1451, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0682, 0.0863, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 22:31:19,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2855, 1.6631, 1.4176, 1.2250], device='cuda:1'), covar=tensor([0.2248, 0.1657, 0.1211, 0.1626], device='cuda:1'), in_proj_covar=tensor([0.1694, 0.1573, 0.1532, 0.1645], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 22:31:23,240 INFO [train.py:968] (1/2) Epoch 11, batch 33200, giga_loss[loss=0.2225, simple_loss=0.3143, pruned_loss=0.06537, over 28690.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3348, pruned_loss=0.08999, over 5656296.71 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3583, pruned_loss=0.1217, over 5674904.72 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3324, pruned_loss=0.08656, over 5654724.28 frames. ], batch size: 262, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:31:44,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5414, 3.3815, 3.1638, 1.7388], device='cuda:1'), covar=tensor([0.0781, 0.0880, 0.0870, 0.2436], device='cuda:1'), in_proj_covar=tensor([0.1029, 0.0966, 0.0839, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 22:32:17,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:32:18,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5004, 1.6371, 1.3713, 1.6280], device='cuda:1'), covar=tensor([0.0760, 0.0290, 0.0330, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0090], device='cuda:1')
+2023-03-05 22:32:20,724 INFO [train.py:968] (1/2) Epoch 11, batch 33250, giga_loss[loss=0.2491, simple_loss=0.333, pruned_loss=0.08253, over 28540.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3343, pruned_loss=0.09015, over 5651509.97 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.358, pruned_loss=0.1215, over 5666558.83 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3319, pruned_loss=0.0866, over 5656845.25 frames. ], batch size: 336, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:32:32,938 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 33300, giga_loss[loss=0.2524, simple_loss=0.3328, pruned_loss=0.08599, over 29024.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3322, pruned_loss=0.08976, over 5660784.44 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3576, pruned_loss=0.1213, over 5671081.36 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3302, pruned_loss=0.08651, over 5660440.24 frames. ], batch size: 285, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:33:41,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=488861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:34:17,528 INFO [train.py:968] (1/2) Epoch 11, batch 33350, giga_loss[loss=0.2773, simple_loss=0.3588, pruned_loss=0.09787, over 29042.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3349, pruned_loss=0.09096, over 5667527.09 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3568, pruned_loss=0.1208, over 5673168.22 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3331, pruned_loss=0.08783, over 5664866.10 frames. ], batch size: 285, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:34:29,178 INFO [optim.py:369] (1/2) 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 22:34:54,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8470, 1.0914, 2.8627, 2.7240], device='cuda:1'), covar=tensor([0.1647, 0.2465, 0.0555, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0663, 0.0590, 0.0854, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 22:34:57,575 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=488919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:35:20,384 INFO [train.py:968] (1/2) Epoch 11, batch 33400, giga_loss[loss=0.2666, simple_loss=0.3455, pruned_loss=0.0939, over 28793.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3361, pruned_loss=0.0918, over 5666556.73 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3563, pruned_loss=0.1205, over 5674130.08 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3348, pruned_loss=0.08907, over 5663274.88 frames. ], batch size: 263, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:36:25,964 INFO [train.py:968] (1/2) Epoch 11, batch 33450, giga_loss[loss=0.2794, simple_loss=0.3569, pruned_loss=0.101, over 28897.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3371, pruned_loss=0.0927, over 5665255.30 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3562, pruned_loss=0.1203, over 5674180.87 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3356, pruned_loss=0.08991, over 5662260.58 frames. ], batch size: 164, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:36:41,423 INFO [optim.py:369] (1/2) 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 22:36:41,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=488999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:36:49,830 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:36:53,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:37:23,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9399, 2.0381, 1.3893, 1.6820], device='cuda:1'), covar=tensor([0.0844, 0.0650, 0.1018, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0434, 0.0497, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 22:37:30,226 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:37:32,038 INFO [train.py:968] (1/2) Epoch 11, batch 33500, libri_loss[loss=0.303, simple_loss=0.3654, pruned_loss=0.1203, over 29501.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3407, pruned_loss=0.09437, over 5669443.36 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3559, pruned_loss=0.12, over 5678435.78 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3394, pruned_loss=0.09189, over 5662888.77 frames. ], batch size: 85, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:37:58,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:38:28,522 INFO [train.py:968] (1/2) Epoch 11, batch 33550, giga_loss[loss=0.2715, simple_loss=0.3529, pruned_loss=0.09502, over 27744.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3422, pruned_loss=0.0944, over 5662080.50 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3557, pruned_loss=0.12, over 5679289.76 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3412, pruned_loss=0.09203, over 5655850.47 frames. ], batch size: 474, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:38:38,488 INFO [optim.py:369] (1/2) 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 22:39:28,926 INFO [train.py:968] (1/2) Epoch 11, batch 33600, libri_loss[loss=0.2717, simple_loss=0.3336, pruned_loss=0.1049, over 29531.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3418, pruned_loss=0.09492, over 5661471.96 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3552, pruned_loss=0.1196, over 5674780.75 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.341, pruned_loss=0.09234, over 5658896.22 frames. ], batch size: 80, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:39:35,992 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:39:41,696 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:39:55,518 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=489156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:40:02,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:40:19,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:40:38,422 INFO [train.py:968] (1/2) Epoch 11, batch 33650, libri_loss[loss=0.3011, simple_loss=0.3617, pruned_loss=0.1202, over 25705.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3394, pruned_loss=0.09377, over 5661380.03 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3548, pruned_loss=0.1194, over 5675819.28 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3388, pruned_loss=0.0913, over 5658236.60 frames. ], batch size: 136, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:40:50,379 INFO [optim.py:369] (1/2) 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 22:41:03,715 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:41:07,915 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:41:37,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-05 22:41:39,421 INFO [train.py:968] (1/2) Epoch 11, batch 33700, giga_loss[loss=0.2291, simple_loss=0.3201, pruned_loss=0.06901, over 28421.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3395, pruned_loss=0.09392, over 5665093.96 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3549, pruned_loss=0.1195, over 5677345.97 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3386, pruned_loss=0.09142, over 5660945.27 frames. ], batch size: 368, lr: 2.91e-03, grad_scale: 8.0
+2023-03-05 22:41:41,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:41:50,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3407, 1.6565, 1.3566, 1.6064], device='cuda:1'), covar=tensor([0.0700, 0.0273, 0.0305, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0117, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:1')
+2023-03-05 22:42:25,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2872, 1.6692, 1.5680, 1.1807], device='cuda:1'), covar=tensor([0.1654, 0.2385, 0.1375, 0.1624], device='cuda:1'), in_proj_covar=tensor([0.0821, 0.0681, 0.0859, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 22:42:45,485 INFO [train.py:968] (1/2) Epoch 11, batch 33750, giga_loss[loss=0.2892, simple_loss=0.3509, pruned_loss=0.1138, over 27602.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3391, pruned_loss=0.09421, over 5654148.98 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3553, pruned_loss=0.1197, over 5682974.30 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3376, pruned_loss=0.09144, over 5645443.82 frames. ], batch size: 472, lr: 2.91e-03, grad_scale: 4.0
+2023-03-05 22:42:51,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:42:59,190 INFO [optim.py:369] (1/2) 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 22:43:04,740 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:43:09,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:43:39,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4398, 3.4191, 1.5282, 1.4823], device='cuda:1'), covar=tensor([0.0857, 0.0306, 0.0875, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0507, 0.0341, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 22:43:45,752 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:43:45,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4296, 2.7620, 1.5671, 1.5942], device='cuda:1'), covar=tensor([0.0737, 0.0328, 0.0738, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0508, 0.0341, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 22:43:47,630 INFO [train.py:968] (1/2) Epoch 11, batch 33800, giga_loss[loss=0.1991, simple_loss=0.2768, pruned_loss=0.06072, over 28464.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3378, pruned_loss=0.09435, over 5658999.53 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3557, pruned_loss=0.1199, over 5689343.01 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3357, pruned_loss=0.09112, over 5645189.14 frames. ], batch size: 71, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:44:22,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3543, 1.6474, 1.4630, 1.2621], device='cuda:1'), covar=tensor([0.2178, 0.1716, 0.1097, 0.1547], device='cuda:1'), in_proj_covar=tensor([0.1704, 0.1567, 0.1531, 0.1646], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 22:44:46,092 INFO [train.py:968] (1/2) Epoch 11, batch 33850, giga_loss[loss=0.3051, simple_loss=0.3652, pruned_loss=0.1225, over 26788.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3368, pruned_loss=0.09377, over 5650736.08 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3554, pruned_loss=0.1198, over 5687323.24 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3349, pruned_loss=0.09075, over 5640770.55 frames. ], batch size: 555, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:44:57,319 INFO [optim.py:369] (1/2) 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 22:45:19,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-05 22:45:25,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0915, 1.4038, 1.1377, 0.3246], device='cuda:1'), covar=tensor([0.1984, 0.2032, 0.2898, 0.3719], device='cuda:1'), in_proj_covar=tensor([0.1544, 0.1472, 0.1481, 0.1268], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 22:45:45,342 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:45:47,114 INFO [train.py:968] (1/2) Epoch 11, batch 33900, giga_loss[loss=0.2094, simple_loss=0.3028, pruned_loss=0.058, over 28873.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.336, pruned_loss=0.09183, over 5666764.83 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3555, pruned_loss=0.1198, over 5691628.39 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3341, pruned_loss=0.08893, over 5654540.35 frames. ], batch size: 145, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:45:49,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:46:22,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=489469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:46:33,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9005, 2.8193, 1.7416, 0.9496], device='cuda:1'), covar=tensor([0.4920, 0.2154, 0.3057, 0.4030], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1480, 0.1487, 0.1274], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 22:46:43,467 INFO [train.py:968] (1/2) Epoch 11, batch 33950, giga_loss[loss=0.3103, simple_loss=0.3771, pruned_loss=0.1218, over 26836.00 frames. ], tot_loss[loss=0.259, simple_loss=0.337, pruned_loss=0.09047, over 5677792.23 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3556, pruned_loss=0.1198, over 5695711.98 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.335, pruned_loss=0.08758, over 5663967.13 frames. ], batch size: 555, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:46:56,797 INFO [optim.py:369] (1/2) 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 22:47:36,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:47:38,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=489533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:47:41,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2365, 1.3869, 3.8575, 3.1440], device='cuda:1'), covar=tensor([0.1929, 0.2578, 0.0616, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0660, 0.0586, 0.0847, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 22:47:44,357 INFO [train.py:968] (1/2) Epoch 11, batch 34000, giga_loss[loss=0.2325, simple_loss=0.3138, pruned_loss=0.07564, over 28669.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3386, pruned_loss=0.08979, over 5672213.65 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3555, pruned_loss=0.1197, over 5696785.42 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3371, pruned_loss=0.08749, over 5660296.41 frames. ], batch size: 92, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:48:44,775 INFO [train.py:968] (1/2) Epoch 11, batch 34050, giga_loss[loss=0.2593, simple_loss=0.3415, pruned_loss=0.08853, over 28990.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3384, pruned_loss=0.08924, over 5662344.22 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3555, pruned_loss=0.1198, over 5689141.98 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3369, pruned_loss=0.08691, over 5660153.05 frames. ], batch size: 145, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:48:45,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-05 22:49:02,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-05 22:49:05,302 INFO [optim.py:369] (1/2) 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 22:49:51,388 INFO [train.py:968] (1/2) Epoch 11, batch 34100, giga_loss[loss=0.2587, simple_loss=0.351, pruned_loss=0.08318, over 28863.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3382, pruned_loss=0.08933, over 5661771.93 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.355, pruned_loss=0.1195, over 5684371.59 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3369, pruned_loss=0.0868, over 5663085.31 frames. ], batch size: 199, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:50:13,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3850, 1.9194, 1.3536, 0.7607], device='cuda:1'), covar=tensor([0.4150, 0.2141, 0.3155, 0.3999], device='cuda:1'), in_proj_covar=tensor([0.1542, 0.1467, 0.1477, 0.1262], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 22:50:40,518 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=489674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:50:44,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7839, 2.2665, 1.8976, 1.6181], device='cuda:1'), covar=tensor([0.2184, 0.1443, 0.1480, 0.1688], device='cuda:1'), in_proj_covar=tensor([0.1701, 0.1565, 0.1533, 0.1645], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 22:50:46,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=489677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:51:03,820 INFO [train.py:968] (1/2) Epoch 11, batch 34150, giga_loss[loss=0.2476, simple_loss=0.3351, pruned_loss=0.08006, over 29011.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3374, pruned_loss=0.08869, over 5661095.70 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3547, pruned_loss=0.1193, over 5686636.11 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3365, pruned_loss=0.08662, over 5659761.52 frames. ], batch size: 128, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:51:20,586 INFO [optim.py:369] (1/2) 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 22:51:25,900 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 34200, giga_loss[loss=0.2605, simple_loss=0.3472, pruned_loss=0.08694, over 28783.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3385, pruned_loss=0.08919, over 5656459.49 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3553, pruned_loss=0.1197, over 5681788.20 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3369, pruned_loss=0.08649, over 5658749.98 frames. ], batch size: 243, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:52:39,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4563, 3.3639, 1.5311, 1.5415], device='cuda:1'), covar=tensor([0.0845, 0.0358, 0.0864, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0506, 0.0343, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 22:53:22,149 INFO [train.py:968] (1/2) Epoch 11, batch 34250, giga_loss[loss=0.2749, simple_loss=0.3597, pruned_loss=0.09504, over 29008.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3391, pruned_loss=0.08941, over 5653761.32 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3555, pruned_loss=0.1199, over 5684904.29 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3375, pruned_loss=0.08672, over 5652513.57 frames. ], batch size: 186, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:53:36,682 INFO [optim.py:369] (1/2) 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 22:54:06,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3693, 3.3133, 1.4254, 1.5298], device='cuda:1'), covar=tensor([0.0934, 0.0324, 0.0924, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0507, 0.0343, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-05 22:54:21,985 INFO [train.py:968] (1/2) Epoch 11, batch 34300, giga_loss[loss=0.2502, simple_loss=0.335, pruned_loss=0.08271, over 29007.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3416, pruned_loss=0.0904, over 5671623.26 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3549, pruned_loss=0.1195, over 5691783.85 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3403, pruned_loss=0.08768, over 5663772.47 frames. ], batch size: 186, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:54:58,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9390, 1.2480, 1.1569, 0.8596], device='cuda:1'), covar=tensor([0.0887, 0.1427, 0.0757, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0680, 0.0859, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 22:55:25,213 INFO [train.py:968] (1/2) Epoch 11, batch 34350, giga_loss[loss=0.219, simple_loss=0.3057, pruned_loss=0.06617, over 29156.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3413, pruned_loss=0.09087, over 5675069.38 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3548, pruned_loss=0.1193, over 5689314.33 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3398, pruned_loss=0.08775, over 5670454.01 frames. ], batch size: 113, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 22:55:41,804 INFO [optim.py:369] (1/2) 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 22:55:48,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=489908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:56:21,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=489933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:56:26,053 INFO [train.py:968] (1/2) Epoch 11, batch 34400, giga_loss[loss=0.2749, simple_loss=0.3455, pruned_loss=0.1022, over 26989.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3399, pruned_loss=0.09119, over 5681871.00 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3543, pruned_loss=0.1189, over 5690625.28 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3384, pruned_loss=0.08745, over 5676570.85 frames. ], batch size: 555, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:57:33,463 INFO [train.py:968] (1/2) Epoch 11, batch 34450, giga_loss[loss=0.2376, simple_loss=0.3224, pruned_loss=0.07645, over 27722.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3388, pruned_loss=0.09047, over 5682810.77 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3545, pruned_loss=0.119, over 5691205.35 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3371, pruned_loss=0.08666, over 5677975.67 frames. ], batch size: 472, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:57:54,650 INFO [optim.py:369] (1/2) 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 22:58:35,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6468, 1.6879, 1.7427, 1.2830], device='cuda:1'), covar=tensor([0.1621, 0.2634, 0.1320, 0.1480], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0679, 0.0860, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 22:58:41,131 INFO [train.py:968] (1/2) Epoch 11, batch 34500, giga_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09415, over 28823.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3369, pruned_loss=0.08834, over 5692907.99 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3546, pruned_loss=0.1191, over 5689677.65 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3353, pruned_loss=0.08499, over 5690565.70 frames. ], batch size: 243, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:58:58,548 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:59:00,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:59:38,116 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=490083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 22:59:43,068 INFO [train.py:968] (1/2) Epoch 11, batch 34550, libri_loss[loss=0.3041, simple_loss=0.353, pruned_loss=0.1277, over 29564.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3382, pruned_loss=0.08969, over 5690414.14 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3545, pruned_loss=0.1191, over 5695104.71 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3365, pruned_loss=0.0862, over 5683721.53 frames. ], batch size: 78, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 22:59:58,395 INFO [optim.py:369] (1/2) 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 23:00:37,510 INFO [train.py:968] (1/2) Epoch 11, batch 34600, giga_loss[loss=0.2484, simple_loss=0.3382, pruned_loss=0.07934, over 28606.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3397, pruned_loss=0.09074, over 5684832.47 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3538, pruned_loss=0.1185, over 5697954.40 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3384, pruned_loss=0.08731, over 5676229.48 frames. ], batch size: 307, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:01:24,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:01:29,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4084, 1.6432, 1.7000, 1.2797], device='cuda:1'), covar=tensor([0.1634, 0.2261, 0.1310, 0.1559], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0678, 0.0860, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 23:01:35,957 INFO [train.py:968] (1/2) Epoch 11, batch 34650, libri_loss[loss=0.2576, simple_loss=0.3252, pruned_loss=0.09503, over 29526.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3389, pruned_loss=0.09063, over 5675877.74 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3531, pruned_loss=0.118, over 5699278.51 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3382, pruned_loss=0.08772, over 5667300.91 frames. ], batch size: 79, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:01:49,818 INFO [optim.py:369] (1/2) 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 23:02:27,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.18 vs. limit=5.0
+2023-03-05 23:02:29,263 INFO [train.py:968] (1/2) Epoch 11, batch 34700, giga_loss[loss=0.231, simple_loss=0.3154, pruned_loss=0.07334, over 28765.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3363, pruned_loss=0.0904, over 5659407.04 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3534, pruned_loss=0.1185, over 5685044.60 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.335, pruned_loss=0.08699, over 5664710.94 frames. ], batch size: 119, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:03:21,515 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:03:21,836 INFO [train.py:968] (1/2) Epoch 11, batch 34750, giga_loss[loss=0.2615, simple_loss=0.3318, pruned_loss=0.09555, over 28679.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3369, pruned_loss=0.09149, over 5668378.77 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3534, pruned_loss=0.1184, over 5690821.97 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3353, pruned_loss=0.08793, over 5666908.23 frames. ], batch size: 242, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:03:39,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-05 23:03:39,810 INFO [optim.py:369] (1/2) 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 23:03:46,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:04:17,280 INFO [train.py:968] (1/2) Epoch 11, batch 34800, giga_loss[loss=0.2747, simple_loss=0.3462, pruned_loss=0.1016, over 28060.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3419, pruned_loss=0.09485, over 5667084.95 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3534, pruned_loss=0.1184, over 5695375.06 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3404, pruned_loss=0.09147, over 5661317.16 frames. ], batch size: 412, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:04:32,519 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-05 23:05:03,933 INFO [train.py:968] (1/2) Epoch 11, batch 34850, giga_loss[loss=0.3083, simple_loss=0.3812, pruned_loss=0.1177, over 28955.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3503, pruned_loss=0.09943, over 5676161.48 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3529, pruned_loss=0.1181, over 5698756.67 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3494, pruned_loss=0.09653, over 5668075.39 frames. ], batch size: 106, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:05:12,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3290, 1.8242, 1.4629, 1.5314], device='cuda:1'), covar=tensor([0.0798, 0.0282, 0.0319, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0117, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0058, 0.0053, 0.0090], device='cuda:1')
+2023-03-05 23:05:14,965 INFO [optim.py:369] (1/2) 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 23:05:47,425 INFO [train.py:968] (1/2) Epoch 11, batch 34900, giga_loss[loss=0.2675, simple_loss=0.3449, pruned_loss=0.09503, over 28861.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3571, pruned_loss=0.1035, over 5675410.99 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3532, pruned_loss=0.1183, over 5692498.41 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.356, pruned_loss=0.1006, over 5673347.76 frames. ], batch size: 112, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:05:58,041 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:00,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:25,891 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=490483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:26,447 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:06:29,756 INFO [train.py:968] (1/2) Epoch 11, batch 34950, libri_loss[loss=0.2883, simple_loss=0.3448, pruned_loss=0.1159, over 29550.00 frames. ], tot_loss[loss=0.281, simple_loss=0.355, pruned_loss=0.1035, over 5679536.33 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3533, pruned_loss=0.1184, over 5696445.33 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3542, pruned_loss=0.1008, over 5673942.07 frames. ], batch size: 79, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:06:40,995 INFO [optim.py:369] (1/2) 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 23:07:04,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3073, 3.1181, 1.3631, 1.4959], device='cuda:1'), covar=tensor([0.0873, 0.0372, 0.0830, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0504, 0.0340, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 23:07:11,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6355, 1.7820, 1.6850, 1.5465], device='cuda:1'), covar=tensor([0.1484, 0.1877, 0.2021, 0.1822], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0727, 0.0665, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 23:07:11,832 INFO [train.py:968] (1/2) Epoch 11, batch 35000, giga_loss[loss=0.2369, simple_loss=0.3134, pruned_loss=0.08022, over 28780.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3484, pruned_loss=0.1008, over 5683903.39 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.354, pruned_loss=0.1187, over 5698589.67 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3471, pruned_loss=0.09813, over 5677151.06 frames. ], batch size: 199, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:07:22,339 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:07:56,777 INFO [train.py:968] (1/2) Epoch 11, batch 35050, giga_loss[loss=0.2439, simple_loss=0.3069, pruned_loss=0.09048, over 26652.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3405, pruned_loss=0.09725, over 5685715.39 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.354, pruned_loss=0.1187, over 5700465.30 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3394, pruned_loss=0.09491, over 5678546.63 frames. ], batch size: 555, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:08:02,908 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:08:07,243 INFO [optim.py:369] (1/2) 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 23:08:36,786 INFO [train.py:968] (1/2) Epoch 11, batch 35100, giga_loss[loss=0.2202, simple_loss=0.293, pruned_loss=0.07367, over 28847.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3334, pruned_loss=0.09422, over 5682978.57 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3538, pruned_loss=0.1184, over 5697878.24 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3321, pruned_loss=0.09193, over 5679294.73 frames. ], batch size: 99, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:08:45,648 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=490652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:08:52,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490663.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:09:15,056 INFO [train.py:968] (1/2) Epoch 11, batch 35150, giga_loss[loss=0.2077, simple_loss=0.2882, pruned_loss=0.06359, over 29037.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3275, pruned_loss=0.092, over 5685304.06 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3544, pruned_loss=0.1187, over 5702198.46 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3252, pruned_loss=0.08903, over 5677978.10 frames. ], batch size: 136, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:09:20,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:09:22,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:09:27,778 INFO [optim.py:369] (1/2) 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 23:09:43,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-05 23:09:44,190 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 35200, giga_loss[loss=0.2353, simple_loss=0.3055, pruned_loss=0.0826, over 28732.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3233, pruned_loss=0.09015, over 5676156.20 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3548, pruned_loss=0.1189, over 5687599.28 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3202, pruned_loss=0.08678, over 5683919.24 frames. ], batch size: 119, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:10:00,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.33 vs. limit=5.0
+2023-03-05 23:10:30,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4954, 1.7327, 1.3969, 1.7755], device='cuda:1'), covar=tensor([0.2050, 0.1927, 0.2026, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.0954, 0.1145, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 23:10:37,248 INFO [train.py:968] (1/2) Epoch 11, batch 35250, giga_loss[loss=0.2287, simple_loss=0.2992, pruned_loss=0.07906, over 27611.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3207, pruned_loss=0.08912, over 5659946.30 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3555, pruned_loss=0.1193, over 5662787.65 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3171, pruned_loss=0.08552, over 5688673.59 frames. ], batch size: 472, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:10:51,310 INFO [optim.py:369] (1/2) 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 23:10:52,755 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=490806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:10:55,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=490809.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:11:18,645 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 35300, giga_loss[loss=0.2276, simple_loss=0.2964, pruned_loss=0.07946, over 28813.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3184, pruned_loss=0.0885, over 5666700.90 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3563, pruned_loss=0.1197, over 5668524.36 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3139, pruned_loss=0.08456, over 5684426.81 frames. ], batch size: 92, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:11:33,824 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 35350, giga_loss[loss=0.2157, simple_loss=0.2914, pruned_loss=0.06996, over 28938.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3159, pruned_loss=0.08754, over 5663437.78 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3568, pruned_loss=0.1198, over 5672940.82 frames. ], giga_tot_loss[loss=0.2394, simple_loss=0.3112, pruned_loss=0.08376, over 5673646.52 frames. ], batch size: 164, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:12:12,590 INFO [optim.py:369] (1/2) 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 23:12:23,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3589, 1.4998, 1.3699, 1.5870], device='cuda:1'), covar=tensor([0.0767, 0.0321, 0.0326, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:1')
+2023-03-05 23:12:34,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4012, 1.7185, 1.4036, 1.3736], device='cuda:1'), covar=tensor([0.2266, 0.1670, 0.1829, 0.1814], device='cuda:1'), in_proj_covar=tensor([0.1715, 0.1577, 0.1551, 0.1663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 23:12:42,006 INFO [train.py:968] (1/2) Epoch 11, batch 35400, giga_loss[loss=0.1955, simple_loss=0.267, pruned_loss=0.06197, over 28856.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3131, pruned_loss=0.08611, over 5674417.17 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3572, pruned_loss=0.1198, over 5678929.48 frames. ], giga_tot_loss[loss=0.2361, simple_loss=0.3078, pruned_loss=0.08215, over 5677195.52 frames. ], batch size: 99, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:13:01,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4181, 3.0090, 1.4579, 1.5210], device='cuda:1'), covar=tensor([0.0925, 0.0330, 0.0870, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0502, 0.0338, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 23:13:05,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-05 23:13:07,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=490973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:21,505 INFO [train.py:968] (1/2) Epoch 11, batch 35450, giga_loss[loss=0.2074, simple_loss=0.2824, pruned_loss=0.0662, over 28947.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3109, pruned_loss=0.08476, over 5686036.00 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3575, pruned_loss=0.12, over 5686213.00 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3047, pruned_loss=0.08016, over 5681609.25 frames. ], batch size: 106, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:13:33,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=491002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:35,018 INFO [optim.py:369] (1/2) 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 23:13:36,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:53,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=491027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:13:59,039 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=491034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:14:03,066 INFO [train.py:968] (1/2) Epoch 11, batch 35500, giga_loss[loss=0.215, simple_loss=0.2711, pruned_loss=0.07943, over 23948.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.3078, pruned_loss=0.08342, over 5689271.33 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.358, pruned_loss=0.1203, over 5689655.19 frames. ], giga_tot_loss[loss=0.2298, simple_loss=0.3017, pruned_loss=0.07895, over 5682504.77 frames. ], batch size: 705, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:14:11,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5331, 1.7517, 1.7874, 1.3929], device='cuda:1'), covar=tensor([0.1764, 0.2242, 0.1397, 0.1568], device='cuda:1'), in_proj_covar=tensor([0.0842, 0.0698, 0.0879, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 23:14:43,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0941, 2.1497, 1.3847, 1.7499], device='cuda:1'), covar=tensor([0.0781, 0.0674, 0.1079, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0434, 0.0497, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:14:44,806 INFO [train.py:968] (1/2) Epoch 11, batch 35550, giga_loss[loss=0.2142, simple_loss=0.2845, pruned_loss=0.07198, over 28577.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3055, pruned_loss=0.08238, over 5689976.94 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3585, pruned_loss=0.1205, over 5692202.73 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.2992, pruned_loss=0.07789, over 5682291.15 frames. ], batch size: 307, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:14:57,468 INFO [optim.py:369] (1/2) 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 23:15:08,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=491116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:09,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:27,671 INFO [train.py:968] (1/2) Epoch 11, batch 35600, giga_loss[loss=0.1903, simple_loss=0.2672, pruned_loss=0.05668, over 29070.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.3049, pruned_loss=0.08292, over 5679536.90 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3596, pruned_loss=0.1213, over 5689981.34 frames. ], giga_tot_loss[loss=0.2257, simple_loss=0.2969, pruned_loss=0.07724, over 5674840.78 frames. ], batch size: 128, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:15:34,634 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=491148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:54,735 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=491170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:15:57,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:16:11,439 INFO [train.py:968] (1/2) Epoch 11, batch 35650, giga_loss[loss=0.2734, simple_loss=0.3371, pruned_loss=0.1049, over 28208.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3118, pruned_loss=0.08665, over 5679875.91 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.36, pruned_loss=0.1215, over 5690786.45 frames. ], giga_tot_loss[loss=0.2333, simple_loss=0.3041, pruned_loss=0.08121, over 5674783.77 frames. ], batch size: 77, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:16:18,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2539, 1.1031, 4.2639, 3.4179], device='cuda:1'), covar=tensor([0.2117, 0.3240, 0.0629, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0654, 0.0584, 0.0852, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:16:24,396 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:16:55,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3230, 1.2010, 1.0868, 1.5462], device='cuda:1'), covar=tensor([0.0705, 0.0326, 0.0331, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:1')
+2023-03-05 23:16:58,762 INFO [train.py:968] (1/2) Epoch 11, batch 35700, giga_loss[loss=0.3311, simple_loss=0.3926, pruned_loss=0.1348, over 27603.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3267, pruned_loss=0.09501, over 5683491.31 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3605, pruned_loss=0.1217, over 5695519.33 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3194, pruned_loss=0.08996, over 5675007.98 frames. ], batch size: 472, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:17:41,998 INFO [train.py:968] (1/2) Epoch 11, batch 35750, giga_loss[loss=0.2957, simple_loss=0.3713, pruned_loss=0.1101, over 28999.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3377, pruned_loss=0.1003, over 5677225.65 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3605, pruned_loss=0.1216, over 5689335.35 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3314, pruned_loss=0.09596, over 5676424.99 frames. ], batch size: 164, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:17:52,479 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 35800, giga_loss[loss=0.2786, simple_loss=0.3617, pruned_loss=0.09776, over 28979.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.346, pruned_loss=0.1039, over 5686298.44 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3608, pruned_loss=0.1215, over 5695455.55 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.34, pruned_loss=0.09979, over 5679916.82 frames. ], batch size: 136, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:18:36,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-05 23:19:02,134 INFO [train.py:968] (1/2) Epoch 11, batch 35850, libri_loss[loss=0.2914, simple_loss=0.3534, pruned_loss=0.1147, over 29575.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3501, pruned_loss=0.1049, over 5696723.69 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3608, pruned_loss=0.1213, over 5704611.83 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3446, pruned_loss=0.101, over 5682708.06 frames. ], batch size: 76, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:19:05,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-05 23:19:14,921 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 35900, giga_loss[loss=0.2623, simple_loss=0.3246, pruned_loss=0.09999, over 23697.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3506, pruned_loss=0.1035, over 5679369.99 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3612, pruned_loss=0.1214, over 5705455.64 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3457, pruned_loss=0.1, over 5666928.98 frames. ], batch size: 705, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:20:10,005 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=491461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:20:31,123 INFO [train.py:968] (1/2) Epoch 11, batch 35950, giga_loss[loss=0.3257, simple_loss=0.3822, pruned_loss=0.1345, over 29041.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3524, pruned_loss=0.1047, over 5683417.57 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3612, pruned_loss=0.1214, over 5708183.87 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3484, pruned_loss=0.1017, over 5670724.87 frames. ], batch size: 155, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:20:47,650 INFO [optim.py:369] (1/2) 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 23:21:13,932 INFO [train.py:968] (1/2) Epoch 11, batch 36000, libri_loss[loss=0.2902, simple_loss=0.3657, pruned_loss=0.1074, over 29251.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3544, pruned_loss=0.106, over 5697534.26 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.362, pruned_loss=0.1216, over 5714017.26 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3501, pruned_loss=0.1026, over 5681107.11 frames. ], batch size: 97, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:21:13,932 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-05 23:21:21,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2560, 3.1476, 1.4203, 1.4321], device='cuda:1'), covar=tensor([0.0967, 0.0432, 0.0915, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.0355, 0.0503, 0.0337, 0.0388], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 23:21:22,667 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-05 23:21:47,680 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 36050, giga_loss[loss=0.2663, simple_loss=0.351, pruned_loss=0.0908, over 28926.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3571, pruned_loss=0.1083, over 5694910.19 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3618, pruned_loss=0.1215, over 5717471.97 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3536, pruned_loss=0.1054, over 5678265.96 frames. ], batch size: 186, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:22:17,239 INFO [optim.py:369] (1/2) 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 23:22:41,613 INFO [train.py:968] (1/2) Epoch 11, batch 36100, giga_loss[loss=0.2921, simple_loss=0.3637, pruned_loss=0.1103, over 28947.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3601, pruned_loss=0.1093, over 5690567.97 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3617, pruned_loss=0.1212, over 5703948.37 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3573, pruned_loss=0.1067, over 5687735.17 frames. ], batch size: 136, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:23:24,531 INFO [train.py:968] (1/2) Epoch 11, batch 36150, giga_loss[loss=0.271, simple_loss=0.3526, pruned_loss=0.09474, over 29016.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.362, pruned_loss=0.1093, over 5691189.53 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3617, pruned_loss=0.1212, over 5704951.63 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3598, pruned_loss=0.1072, over 5688133.86 frames. ], batch size: 66, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:23:28,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:23:44,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9753, 1.2466, 3.5932, 3.0719], device='cuda:1'), covar=tensor([0.1750, 0.2659, 0.0405, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0655, 0.0586, 0.0850, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:24:07,563 INFO [train.py:968] (1/2) Epoch 11, batch 36200, giga_loss[loss=0.2631, simple_loss=0.3514, pruned_loss=0.08744, over 28972.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3639, pruned_loss=0.1096, over 5698932.75 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3623, pruned_loss=0.1215, over 5709911.79 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3617, pruned_loss=0.1074, over 5691613.92 frames. ], batch size: 145, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:24:43,148 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=491785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:24:46,845 INFO [train.py:968] (1/2) Epoch 11, batch 36250, giga_loss[loss=0.2591, simple_loss=0.3449, pruned_loss=0.08659, over 28887.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3645, pruned_loss=0.1094, over 5697489.83 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3629, pruned_loss=0.1219, over 5713008.38 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3622, pruned_loss=0.107, over 5688572.23 frames. ], batch size: 186, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:25:00,823 INFO [optim.py:369] (1/2) 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 23:25:11,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8463, 1.1325, 3.6837, 3.0271], device='cuda:1'), covar=tensor([0.1976, 0.2907, 0.0438, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0653, 0.0584, 0.0848, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:25:23,862 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 36300, giga_loss[loss=0.2836, simple_loss=0.3626, pruned_loss=0.1023, over 28786.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3627, pruned_loss=0.1068, over 5687429.74 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.363, pruned_loss=0.1218, over 5694711.11 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3607, pruned_loss=0.1046, over 5697859.77 frames. ], batch size: 262, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:25:43,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9638, 1.1351, 3.5140, 2.9062], device='cuda:1'), covar=tensor([0.1723, 0.2714, 0.0423, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0654, 0.0585, 0.0848, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:26:05,913 INFO [train.py:968] (1/2) Epoch 11, batch 36350, giga_loss[loss=0.3155, simple_loss=0.3751, pruned_loss=0.128, over 28598.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3604, pruned_loss=0.1049, over 5688807.31 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3633, pruned_loss=0.1218, over 5696686.83 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3584, pruned_loss=0.1027, over 5695177.31 frames. ], batch size: 85, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:26:21,284 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 36400, giga_loss[loss=0.3021, simple_loss=0.3734, pruned_loss=0.1154, over 28378.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3615, pruned_loss=0.1061, over 5676483.54 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.364, pruned_loss=0.1222, over 5689405.95 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3594, pruned_loss=0.1039, over 5688422.43 frames. ], batch size: 65, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:26:51,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=491943.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:27:24,669 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=491982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:27:32,564 INFO [train.py:968] (1/2) Epoch 11, batch 36450, libri_loss[loss=0.3408, simple_loss=0.4023, pruned_loss=0.1396, over 29657.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3643, pruned_loss=0.1107, over 5681019.23 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3641, pruned_loss=0.1222, over 5695100.68 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3624, pruned_loss=0.1085, over 5685261.87 frames. ], batch size: 88, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:27:47,624 INFO [optim.py:369] (1/2) 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 23:27:51,479 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:28:13,907 INFO [train.py:968] (1/2) Epoch 11, batch 36500, giga_loss[loss=0.3157, simple_loss=0.3783, pruned_loss=0.1265, over 28699.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3664, pruned_loss=0.1143, over 5684292.07 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3642, pruned_loss=0.1224, over 5697777.33 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3648, pruned_loss=0.1121, over 5685282.02 frames. ], batch size: 262, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:28:34,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3004, 1.5410, 1.5066, 1.4516], device='cuda:1'), covar=tensor([0.1209, 0.1245, 0.1759, 0.1304], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0735, 0.0670, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 23:28:37,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:28:37,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3834, 1.4399, 1.3111, 1.5392], device='cuda:1'), covar=tensor([0.0648, 0.0280, 0.0289, 0.0670], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0112, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0057, 0.0052, 0.0089], device='cuda:1')
+2023-03-05 23:28:40,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4716, 1.6206, 1.5467, 1.5438], device='cuda:1'), covar=tensor([0.1322, 0.1709, 0.1713, 0.1535], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0735, 0.0670, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 23:28:40,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492070.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:28:53,382 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 36550, giga_loss[loss=0.2918, simple_loss=0.3582, pruned_loss=0.1127, over 29066.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3656, pruned_loss=0.1149, over 5689419.98 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3647, pruned_loss=0.1226, over 5703152.83 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3639, pruned_loss=0.1128, over 5684886.66 frames. ], batch size: 113, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:28:55,339 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492089.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:29:09,368 INFO [optim.py:369] (1/2) 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 23:29:19,520 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492118.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:29:34,429 INFO [train.py:968] (1/2) Epoch 11, batch 36600, giga_loss[loss=0.2729, simple_loss=0.3406, pruned_loss=0.1026, over 27634.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3633, pruned_loss=0.1135, over 5699865.74 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3654, pruned_loss=0.1228, over 5705489.81 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3614, pruned_loss=0.1114, over 5694234.72 frames. ], batch size: 472, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:29:36,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-05 23:29:48,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:29:50,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 36650, giga_loss[loss=0.2854, simple_loss=0.3445, pruned_loss=0.1131, over 28516.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3617, pruned_loss=0.1127, over 5695018.79 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3649, pruned_loss=0.1225, over 5699587.93 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3604, pruned_loss=0.111, over 5695470.93 frames. ], batch size: 85, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:30:31,561 INFO [optim.py:369] (1/2) 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 23:30:31,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5982, 2.2872, 1.6454, 0.7995], device='cuda:1'), covar=tensor([0.4020, 0.1957, 0.3151, 0.4524], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1482, 0.1499, 0.1282], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 23:30:37,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492213.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:30:40,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492216.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:30:57,799 INFO [train.py:968] (1/2) Epoch 11, batch 36700, giga_loss[loss=0.291, simple_loss=0.3548, pruned_loss=0.1136, over 27654.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3605, pruned_loss=0.1112, over 5690285.02 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3653, pruned_loss=0.1227, over 5699618.28 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.359, pruned_loss=0.1094, over 5690534.96 frames. ], batch size: 472, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:31:03,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:31:26,611 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492268.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:31:46,598 INFO [train.py:968] (1/2) Epoch 11, batch 36750, giga_loss[loss=0.2756, simple_loss=0.3443, pruned_loss=0.1035, over 28892.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3566, pruned_loss=0.1083, over 5683240.51 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3654, pruned_loss=0.1227, over 5694188.43 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3552, pruned_loss=0.1066, over 5687349.48 frames. ], batch size: 112, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:31:58,498 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:32:00,864 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:32:21,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:32:28,169 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 36800, giga_loss[loss=0.2389, simple_loss=0.3136, pruned_loss=0.08206, over 28694.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3504, pruned_loss=0.1044, over 5695921.45 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3654, pruned_loss=0.1226, over 5700241.45 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.349, pruned_loss=0.1027, over 5693664.00 frames. ], batch size: 307, lr: 2.90e-03, grad_scale: 8.0
+2023-03-05 23:33:03,954 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-05 23:33:10,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-05 23:33:22,098 INFO [train.py:968] (1/2) Epoch 11, batch 36850, libri_loss[loss=0.2606, simple_loss=0.322, pruned_loss=0.09963, over 28194.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3445, pruned_loss=0.1012, over 5691750.22 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3652, pruned_loss=0.1224, over 5703544.90 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3428, pruned_loss=0.09918, over 5686736.02 frames. ], batch size: 62, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:33:43,804 INFO [optim.py:369] (1/2) 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 23:34:02,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8615, 2.0193, 2.1864, 1.6636], device='cuda:1'), covar=tensor([0.1711, 0.2119, 0.1233, 0.1533], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0697, 0.0873, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-05 23:34:15,845 INFO [train.py:968] (1/2) Epoch 11, batch 36900, giga_loss[loss=0.2779, simple_loss=0.3439, pruned_loss=0.106, over 28851.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3387, pruned_loss=0.09798, over 5677865.13 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3653, pruned_loss=0.1225, over 5704585.17 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3372, pruned_loss=0.09621, over 5672951.73 frames. ], batch size: 112, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:34:19,206 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 36950, giga_loss[loss=0.2809, simple_loss=0.3546, pruned_loss=0.1036, over 28835.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3409, pruned_loss=0.09921, over 5682783.25 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3657, pruned_loss=0.1227, over 5708068.82 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3389, pruned_loss=0.09721, over 5675342.85 frames. ], batch size: 112, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:35:16,957 INFO [optim.py:369] (1/2) 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 23:35:19,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-05 23:35:39,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-05 23:35:40,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:35:44,418 INFO [train.py:968] (1/2) Epoch 11, batch 37000, giga_loss[loss=0.2618, simple_loss=0.3413, pruned_loss=0.0912, over 29047.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3399, pruned_loss=0.09794, over 5688315.01 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3658, pruned_loss=0.1228, over 5700358.52 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3379, pruned_loss=0.09604, over 5688500.34 frames. ], batch size: 164, lr: 2.90e-03, grad_scale: 4.0
+2023-03-05 23:35:50,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-05 23:36:26,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:36:27,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:36:28,080 INFO [train.py:968] (1/2) Epoch 11, batch 37050, giga_loss[loss=0.2518, simple_loss=0.3247, pruned_loss=0.08945, over 28955.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3399, pruned_loss=0.09877, over 5688961.42 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3668, pruned_loss=0.1234, over 5703341.22 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3369, pruned_loss=0.09623, over 5686112.92 frames. ], batch size: 227, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:36:33,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4883, 1.8587, 1.5803, 1.2441], device='cuda:1'), covar=tensor([0.1578, 0.1282, 0.0921, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.1707, 0.1582, 0.1548, 0.1663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-05 23:36:34,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8916, 1.8517, 1.3904, 1.3752], device='cuda:1'), covar=tensor([0.0802, 0.0668, 0.1045, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0440, 0.0503, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:36:45,362 INFO [optim.py:369] (1/2) 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 23:36:51,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:09,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4198, 1.5865, 1.5472, 1.5121], device='cuda:1'), covar=tensor([0.1596, 0.1861, 0.2177, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0738, 0.0672, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-05 23:37:09,894 INFO [train.py:968] (1/2) Epoch 11, batch 37100, giga_loss[loss=0.2729, simple_loss=0.3384, pruned_loss=0.1037, over 28871.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3375, pruned_loss=0.0974, over 5696116.73 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.367, pruned_loss=0.1233, over 5706934.62 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3345, pruned_loss=0.09501, over 5690541.76 frames. ], batch size: 186, lr: 2.90e-03, grad_scale: 2.0
+2023-03-05 23:37:13,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:37:39,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4639, 4.2864, 4.0642, 1.9247], device='cuda:1'), covar=tensor([0.0486, 0.0627, 0.0614, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.1023, 0.0958, 0.0839, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-05 23:37:41,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:43,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:52,339 INFO [train.py:968] (1/2) Epoch 11, batch 37150, giga_loss[loss=0.2543, simple_loss=0.3318, pruned_loss=0.08842, over 28718.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3342, pruned_loss=0.0956, over 5704798.31 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.367, pruned_loss=0.1233, over 5706934.62 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3318, pruned_loss=0.09374, over 5700459.24 frames. ], batch size: 284, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:37:53,047 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:37:53,305 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-05 23:38:04,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=492704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:38:07,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:38:08,377 INFO [optim.py:369] (1/2) 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 23:38:31,778 INFO [train.py:968] (1/2) Epoch 11, batch 37200, giga_loss[loss=0.2312, simple_loss=0.3023, pruned_loss=0.08005, over 28887.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3319, pruned_loss=0.09453, over 5705138.33 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3674, pruned_loss=0.1233, over 5700839.07 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.329, pruned_loss=0.09248, over 5707704.09 frames. ], batch size: 186, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:39:11,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492786.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:39:12,967 INFO [train.py:968] (1/2) Epoch 11, batch 37250, giga_loss[loss=0.2217, simple_loss=0.2997, pruned_loss=0.07189, over 29014.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3307, pruned_loss=0.09435, over 5693843.37 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3677, pruned_loss=0.1234, over 5692913.67 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3279, pruned_loss=0.09245, over 5702498.76 frames. ], batch size: 136, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:39:13,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492789.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:39:28,350 INFO [optim.py:369] (1/2) 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 23:39:32,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=492814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:39:36,961 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492818.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:39:37,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5809, 1.8052, 1.5333, 1.6415], device='cuda:1'), covar=tensor([0.2348, 0.2379, 0.2496, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.1294, 0.0960, 0.1147, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 23:39:51,347 INFO [train.py:968] (1/2) Epoch 11, batch 37300, giga_loss[loss=0.2435, simple_loss=0.3141, pruned_loss=0.08649, over 28782.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3299, pruned_loss=0.09378, over 5706476.82 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3687, pruned_loss=0.1236, over 5700100.91 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3256, pruned_loss=0.09116, over 5707006.47 frames. ], batch size: 119, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:39:58,009 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=492847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:39:59,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4944, 3.9182, 1.5741, 1.6553], device='cuda:1'), covar=tensor([0.0839, 0.0337, 0.0845, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0506, 0.0339, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 23:40:01,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=492850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:40:06,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2245, 1.4132, 4.1143, 3.2408], device='cuda:1'), covar=tensor([0.1746, 0.2501, 0.0408, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0657, 0.0585, 0.0852, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:40:23,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=492879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:40:30,294 INFO [train.py:968] (1/2) Epoch 11, batch 37350, giga_loss[loss=0.2954, simple_loss=0.3595, pruned_loss=0.1156, over 27617.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3283, pruned_loss=0.09283, over 5704756.51 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3694, pruned_loss=0.1237, over 5694678.97 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3228, pruned_loss=0.08959, over 5710399.09 frames. ], batch size: 472, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:40:49,499 INFO [optim.py:369] (1/2) 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 23:41:14,219 INFO [train.py:968] (1/2) Epoch 11, batch 37400, giga_loss[loss=0.2209, simple_loss=0.2944, pruned_loss=0.07377, over 28547.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3266, pruned_loss=0.09184, over 5716966.51 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3704, pruned_loss=0.1242, over 5697562.88 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3203, pruned_loss=0.08812, over 5719016.49 frames. ], batch size: 85, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:41:54,605 INFO [train.py:968] (1/2) Epoch 11, batch 37450, giga_loss[loss=0.2067, simple_loss=0.2869, pruned_loss=0.06322, over 28923.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3249, pruned_loss=0.09034, over 5725210.60 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3707, pruned_loss=0.1242, over 5700736.32 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.319, pruned_loss=0.08694, over 5724251.01 frames. ], batch size: 145, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:42:15,566 INFO [optim.py:369] (1/2) 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 23:42:32,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1869, 2.5437, 1.3342, 1.2826], device='cuda:1'), covar=tensor([0.0992, 0.0363, 0.0884, 0.1398], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0506, 0.0339, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 23:42:36,412 INFO [train.py:968] (1/2) Epoch 11, batch 37500, giga_loss[loss=0.2264, simple_loss=0.3038, pruned_loss=0.0745, over 28379.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3292, pruned_loss=0.09331, over 5721124.73 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3721, pruned_loss=0.1248, over 5701734.58 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3223, pruned_loss=0.08923, over 5719830.72 frames. ], batch size: 65, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:42:45,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:42:47,258 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-05 23:42:48,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4098, 1.4609, 1.2814, 1.5296], device='cuda:1'), covar=tensor([0.0755, 0.0332, 0.0327, 0.0784], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0116, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:1')
+2023-03-05 23:43:01,079 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 37550, giga_loss[loss=0.3537, simple_loss=0.3933, pruned_loss=0.1571, over 23691.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3345, pruned_loss=0.09692, over 5703250.80 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3728, pruned_loss=0.1254, over 5695838.74 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3278, pruned_loss=0.0928, over 5707969.89 frames. ], batch size: 705, lr: 2.89e-03, grad_scale: 1.0
+2023-03-05 23:43:28,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7871, 2.0950, 1.6640, 1.9248], device='cuda:1'), covar=tensor([0.2064, 0.2030, 0.2183, 0.1988], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.0961, 0.1149, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 23:43:40,092 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 37600, giga_loss[loss=0.3719, simple_loss=0.4282, pruned_loss=0.1578, over 28340.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3436, pruned_loss=0.1031, over 5691264.92 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3733, pruned_loss=0.1256, over 5692902.55 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3371, pruned_loss=0.09904, over 5697523.32 frames. ], batch size: 369, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:44:56,924 INFO [train.py:968] (1/2) Epoch 11, batch 37650, libri_loss[loss=0.2638, simple_loss=0.327, pruned_loss=0.1003, over 29505.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3502, pruned_loss=0.1075, over 5684706.09 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3732, pruned_loss=0.1255, over 5693110.68 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3446, pruned_loss=0.104, over 5689445.89 frames. ], batch size: 70, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:44:57,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493189.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:45:18,856 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:45:22,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:45:32,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-05 23:45:45,920 INFO [train.py:968] (1/2) Epoch 11, batch 37700, giga_loss[loss=0.297, simple_loss=0.374, pruned_loss=0.11, over 28720.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3545, pruned_loss=0.109, over 5667884.18 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3736, pruned_loss=0.1257, over 5686554.63 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3495, pruned_loss=0.1058, over 5677726.89 frames. ], batch size: 284, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:45:46,647 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=493240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:46:28,979 INFO [train.py:968] (1/2) Epoch 11, batch 37750, giga_loss[loss=0.3541, simple_loss=0.3999, pruned_loss=0.1541, over 23378.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3606, pruned_loss=0.1126, over 5673864.51 frames. ], libri_tot_loss[loss=0.3127, simple_loss=0.3736, pruned_loss=0.1259, over 5694488.93 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.356, pruned_loss=0.1093, over 5674299.85 frames. ], batch size: 705, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:46:31,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5977, 2.2957, 1.7561, 0.7883], device='cuda:1'), covar=tensor([0.3021, 0.2095, 0.2878, 0.3445], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1473, 0.1493, 0.1266], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 23:46:44,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:46:48,358 INFO [optim.py:369] (1/2) 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 23:47:07,880 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493332.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:47:09,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493335.0, num_to_drop=1, layers_to_drop={1}
+2023-03-05 23:47:12,967 INFO [train.py:968] (1/2) Epoch 11, batch 37800, giga_loss[loss=0.3208, simple_loss=0.3883, pruned_loss=0.1267, over 28572.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3665, pruned_loss=0.1162, over 5678864.81 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3739, pruned_loss=0.1261, over 5697662.55 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3623, pruned_loss=0.1131, over 5675906.44 frames. ], batch size: 307, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:47:29,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4995, 3.2755, 1.6397, 1.6275], device='cuda:1'), covar=tensor([0.0874, 0.0287, 0.0821, 0.1278], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0508, 0.0337, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-05 23:47:33,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=493364.0, num_to_drop=1, layers_to_drop={0}
+2023-03-05 23:47:39,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3044, 1.9154, 1.5038, 1.5435], device='cuda:1'), covar=tensor([0.0680, 0.0247, 0.0271, 0.0663], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-05 23:47:49,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-05 23:47:52,188 INFO [train.py:968] (1/2) Epoch 11, batch 37850, giga_loss[loss=0.2232, simple_loss=0.3086, pruned_loss=0.06891, over 29010.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3624, pruned_loss=0.1133, over 5679829.33 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3736, pruned_loss=0.1261, over 5702234.69 frames. ], giga_tot_loss[loss=0.2897, simple_loss=0.3588, pruned_loss=0.1103, over 5672652.58 frames. ], batch size: 136, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:48:09,229 INFO [optim.py:369] (1/2) 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 23:48:19,375 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 37900, giga_loss[loss=0.2556, simple_loss=0.3407, pruned_loss=0.08528, over 28585.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3587, pruned_loss=0.1096, over 5675737.40 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3741, pruned_loss=0.1265, over 5692809.50 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3552, pruned_loss=0.1068, over 5678344.63 frames. ], batch size: 60, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:49:01,643 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:49:21,184 INFO [train.py:968] (1/2) Epoch 11, batch 37950, giga_loss[loss=0.3378, simple_loss=0.3834, pruned_loss=0.1461, over 26777.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3565, pruned_loss=0.1075, over 5678066.34 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.374, pruned_loss=0.1265, over 5691489.19 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3537, pruned_loss=0.1052, over 5681316.25 frames. ], batch size: 555, lr: 2.89e-03, grad_scale: 2.0
+2023-03-05 23:49:38,086 INFO [optim.py:369] (1/2) 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 23:49:41,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-05 23:49:56,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-05 23:50:01,970 INFO [train.py:968] (1/2) Epoch 11, batch 38000, libri_loss[loss=0.2977, simple_loss=0.3698, pruned_loss=0.1128, over 29490.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3563, pruned_loss=0.1069, over 5685712.75 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3743, pruned_loss=0.1266, over 5695219.88 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3533, pruned_loss=0.1044, over 5684684.29 frames. ], batch size: 85, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:50:23,249 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:50:26,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:50:42,447 INFO [train.py:968] (1/2) Epoch 11, batch 38050, giga_loss[loss=0.2938, simple_loss=0.3646, pruned_loss=0.1115, over 28816.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3589, pruned_loss=0.1084, over 5667426.93 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3747, pruned_loss=0.1269, over 5674415.15 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3556, pruned_loss=0.1055, over 5684713.16 frames. ], batch size: 99, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:50:49,257 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:51:19,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9172, 1.8552, 1.4931, 1.4823], device='cuda:1'), covar=tensor([0.0641, 0.0457, 0.0863, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0437, 0.0501, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:51:24,173 INFO [train.py:968] (1/2) Epoch 11, batch 38100, giga_loss[loss=0.2756, simple_loss=0.3483, pruned_loss=0.1014, over 28852.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3606, pruned_loss=0.1099, over 5674300.06 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3739, pruned_loss=0.1265, over 5679175.97 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3583, pruned_loss=0.1075, over 5683656.97 frames. ], batch size: 227, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:52:05,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:52:12,350 INFO [train.py:968] (1/2) Epoch 11, batch 38150, giga_loss[loss=0.2751, simple_loss=0.3531, pruned_loss=0.09851, over 28559.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3615, pruned_loss=0.1106, over 5682658.30 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3736, pruned_loss=0.1262, over 5680377.96 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3598, pruned_loss=0.1088, over 5688951.27 frames. ], batch size: 78, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:52:31,393 INFO [optim.py:369] (1/2) 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 23:52:54,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5686, 1.7735, 1.4059, 1.6787], device='cuda:1'), covar=tensor([0.2247, 0.2222, 0.2483, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.0961, 0.1146, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 23:52:56,633 INFO [train.py:968] (1/2) Epoch 11, batch 38200, giga_loss[loss=0.2825, simple_loss=0.3622, pruned_loss=0.1014, over 28817.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3621, pruned_loss=0.1111, over 5686223.61 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3739, pruned_loss=0.1262, over 5684517.43 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3603, pruned_loss=0.1094, over 5687533.25 frames. ], batch size: 174, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:53:27,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5332, 2.2133, 1.6171, 0.7379], device='cuda:1'), covar=tensor([0.3345, 0.2092, 0.2784, 0.3692], device='cuda:1'), in_proj_covar=tensor([0.1561, 0.1476, 0.1503, 0.1277], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-05 23:53:39,588 INFO [train.py:968] (1/2) Epoch 11, batch 38250, giga_loss[loss=0.2688, simple_loss=0.3447, pruned_loss=0.09645, over 28771.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3626, pruned_loss=0.1115, over 5698169.83 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3741, pruned_loss=0.1261, over 5688585.83 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3608, pruned_loss=0.11, over 5695487.10 frames. ], batch size: 119, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:53:55,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=493807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:53:57,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4719, 2.1535, 1.5304, 1.7271], device='cuda:1'), covar=tensor([0.0703, 0.0234, 0.0303, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0116, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0089], device='cuda:1')
+2023-03-05 23:53:57,932 INFO [optim.py:369] (1/2) 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 23:54:08,860 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:54:08,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4410, 1.6769, 1.3954, 1.5681], device='cuda:1'), covar=tensor([0.2278, 0.2299, 0.2406, 0.2100], device='cuda:1'), in_proj_covar=tensor([0.1299, 0.0968, 0.1151, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 23:54:11,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:54:21,681 INFO [train.py:968] (1/2) Epoch 11, batch 38300, giga_loss[loss=0.2529, simple_loss=0.3403, pruned_loss=0.08271, over 28748.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3636, pruned_loss=0.1113, over 5704944.41 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3746, pruned_loss=0.1264, over 5693179.43 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3615, pruned_loss=0.1095, over 5699018.67 frames. ], batch size: 60, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:54:23,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-05 23:54:24,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=493842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:54:34,952 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 38350, giga_loss[loss=0.273, simple_loss=0.3268, pruned_loss=0.1096, over 23488.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3632, pruned_loss=0.11, over 5707008.25 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3746, pruned_loss=0.1264, over 5697522.27 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3614, pruned_loss=0.1083, over 5698656.89 frames. ], batch size: 705, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:55:20,212 INFO [optim.py:369] (1/2) 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 23:55:24,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3125, 1.4793, 3.8965, 3.2101], device='cuda:1'), covar=tensor([0.1505, 0.2367, 0.0404, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0655, 0.0577, 0.0848, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-05 23:55:41,919 INFO [train.py:968] (1/2) Epoch 11, batch 38400, giga_loss[loss=0.3044, simple_loss=0.3764, pruned_loss=0.1162, over 29009.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3628, pruned_loss=0.1092, over 5687034.41 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1266, over 5673724.53 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.361, pruned_loss=0.1073, over 5702598.14 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:56:21,426 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=493985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:56:23,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=493988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:56:23,687 INFO [train.py:968] (1/2) Epoch 11, batch 38450, giga_loss[loss=0.2691, simple_loss=0.341, pruned_loss=0.09861, over 28759.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3597, pruned_loss=0.1076, over 5692297.74 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3745, pruned_loss=0.1264, over 5677234.16 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3582, pruned_loss=0.1059, over 5701555.35 frames. ], batch size: 242, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:56:46,191 INFO [optim.py:369] (1/2) 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 23:56:50,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:57:08,700 INFO [train.py:968] (1/2) Epoch 11, batch 38500, giga_loss[loss=0.2768, simple_loss=0.3523, pruned_loss=0.1007, over 28940.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3578, pruned_loss=0.1064, over 5699485.33 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3748, pruned_loss=0.1265, over 5675997.05 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3561, pruned_loss=0.1047, over 5708343.17 frames. ], batch size: 136, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:57:13,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5409, 1.7861, 1.4353, 1.7785], device='cuda:1'), covar=tensor([0.2384, 0.2368, 0.2600, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.0958, 0.1139, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-05 23:57:46,681 INFO [train.py:968] (1/2) Epoch 11, batch 38550, giga_loss[loss=0.2672, simple_loss=0.3451, pruned_loss=0.09462, over 28895.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3562, pruned_loss=0.1059, over 5712116.15 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3749, pruned_loss=0.1265, over 5680901.13 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3545, pruned_loss=0.1041, over 5715399.61 frames. ], batch size: 145, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:58:06,436 INFO [optim.py:369] (1/2) 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 23:58:21,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:58:29,361 INFO [train.py:968] (1/2) Epoch 11, batch 38600, giga_loss[loss=0.2781, simple_loss=0.3506, pruned_loss=0.1028, over 28989.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3564, pruned_loss=0.1066, over 5708882.52 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1266, over 5680590.95 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3546, pruned_loss=0.1048, over 5711893.20 frames. ], batch size: 136, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:59:03,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=494182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-05 23:59:08,916 INFO [train.py:968] (1/2) Epoch 11, batch 38650, giga_loss[loss=0.2636, simple_loss=0.3477, pruned_loss=0.08973, over 29005.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3568, pruned_loss=0.1067, over 5708282.93 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3752, pruned_loss=0.1267, over 5683663.20 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3551, pruned_loss=0.1051, over 5708404.31 frames. ], batch size: 155, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:59:28,192 INFO [optim.py:369] (1/2) 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 23:59:49,590 INFO [train.py:968] (1/2) Epoch 11, batch 38700, giga_loss[loss=0.2827, simple_loss=0.3593, pruned_loss=0.1031, over 29059.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3564, pruned_loss=0.1057, over 5712821.82 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3754, pruned_loss=0.1268, over 5688090.11 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3545, pruned_loss=0.1039, over 5709603.93 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 4.0
+2023-03-05 23:59:54,188 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:00:10,577 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 00:00:28,751 INFO [train.py:968] (1/2) Epoch 11, batch 38750, giga_loss[loss=0.3164, simple_loss=0.3849, pruned_loss=0.1239, over 28668.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3553, pruned_loss=0.1046, over 5715129.19 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3755, pruned_loss=0.1269, over 5691238.53 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3532, pruned_loss=0.1026, over 5710192.41 frames. ], batch size: 242, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:00:46,459 INFO [optim.py:369] (1/2) 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 00:01:00,701 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=494325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:01:03,003 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 38800, giga_loss[loss=0.3081, simple_loss=0.374, pruned_loss=0.1211, over 28573.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3543, pruned_loss=0.1042, over 5713207.05 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3755, pruned_loss=0.1268, over 5692444.95 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3524, pruned_loss=0.1023, over 5708582.87 frames. ], batch size: 336, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:01:25,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:01:37,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4342, 1.7557, 1.3421, 1.4640], device='cuda:1'), covar=tensor([0.2496, 0.2334, 0.2626, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1285, 0.0955, 0.1135, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 00:01:49,501 INFO [train.py:968] (1/2) Epoch 11, batch 38850, giga_loss[loss=0.2663, simple_loss=0.3317, pruned_loss=0.1005, over 28946.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3515, pruned_loss=0.1028, over 5703492.70 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3756, pruned_loss=0.1269, over 5686561.73 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3496, pruned_loss=0.1009, over 5704778.13 frames. ], batch size: 106, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:02:09,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0446, 1.2506, 1.1528, 0.9851], device='cuda:1'), covar=tensor([0.1340, 0.1492, 0.0905, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.1723, 0.1614, 0.1588, 0.1683], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 00:02:10,512 INFO [optim.py:369] (1/2) 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 00:02:30,683 INFO [train.py:968] (1/2) Epoch 11, batch 38900, giga_loss[loss=0.2649, simple_loss=0.3333, pruned_loss=0.09825, over 29072.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.349, pruned_loss=0.1019, over 5706229.28 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.376, pruned_loss=0.1274, over 5688520.80 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3465, pruned_loss=0.09942, over 5705853.03 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:03:12,045 INFO [train.py:968] (1/2) Epoch 11, batch 38950, giga_loss[loss=0.2661, simple_loss=0.3469, pruned_loss=0.09263, over 28662.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.348, pruned_loss=0.1017, over 5704936.70 frames. ], libri_tot_loss[loss=0.3154, simple_loss=0.3761, pruned_loss=0.1274, over 5689348.19 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3459, pruned_loss=0.09966, over 5703990.01 frames. ], batch size: 262, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:03:14,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 00:03:27,149 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=494504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:03:34,568 INFO [optim.py:369] (1/2) 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 00:03:54,261 INFO [train.py:968] (1/2) Epoch 11, batch 39000, giga_loss[loss=0.3832, simple_loss=0.4207, pruned_loss=0.1729, over 26705.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3505, pruned_loss=0.1042, over 5701814.39 frames. ], libri_tot_loss[loss=0.3162, simple_loss=0.3766, pruned_loss=0.1279, over 5695539.43 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3472, pruned_loss=0.1012, over 5695808.35 frames. ], batch size: 555, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:03:54,261 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 00:04:02,711 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 00:04:39,623 INFO [train.py:968] (1/2) Epoch 11, batch 39050, libri_loss[loss=0.3355, simple_loss=0.4036, pruned_loss=0.1337, over 29179.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3483, pruned_loss=0.1033, over 5707083.02 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3772, pruned_loss=0.1281, over 5698419.09 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3446, pruned_loss=0.1001, over 5700017.58 frames. ], batch size: 101, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:04:46,504 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 00:04:58,962 INFO [optim.py:369] (1/2) 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 00:05:05,393 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=494619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:05:20,553 INFO [train.py:968] (1/2) Epoch 11, batch 39100, giga_loss[loss=0.279, simple_loss=0.3398, pruned_loss=0.1091, over 28925.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3452, pruned_loss=0.1018, over 5712871.68 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.377, pruned_loss=0.1278, over 5699345.48 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3417, pruned_loss=0.09887, over 5706859.94 frames. ], batch size: 106, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:05:26,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=494647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:05:28,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=494650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:05:31,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4999, 1.6194, 1.4639, 1.3734], device='cuda:1'), covar=tensor([0.2094, 0.1804, 0.1481, 0.1721], device='cuda:1'), in_proj_covar=tensor([0.1708, 0.1602, 0.1578, 0.1674], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 00:05:32,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6703, 1.6836, 1.7127, 1.5088], device='cuda:1'), covar=tensor([0.1457, 0.1941, 0.1992, 0.1917], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0731, 0.0669, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 00:05:51,878 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:06:00,245 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-06 00:06:00,368 INFO [train.py:968] (1/2) Epoch 11, batch 39150, giga_loss[loss=0.2388, simple_loss=0.3154, pruned_loss=0.08113, over 28928.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3426, pruned_loss=0.1008, over 5702021.22 frames. ], libri_tot_loss[loss=0.3168, simple_loss=0.3773, pruned_loss=0.1282, over 5693289.75 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.339, pruned_loss=0.09771, over 5702468.63 frames. ], batch size: 164, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:06:21,626 INFO [optim.py:369] (1/2) 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 00:06:26,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494719.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:06:45,297 INFO [train.py:968] (1/2) Epoch 11, batch 39200, giga_loss[loss=0.3026, simple_loss=0.3763, pruned_loss=0.1145, over 28737.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3399, pruned_loss=0.09901, over 5708774.66 frames. ], libri_tot_loss[loss=0.3161, simple_loss=0.3767, pruned_loss=0.1278, over 5695678.62 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3369, pruned_loss=0.09644, over 5707174.08 frames. ], batch size: 242, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:07:02,668 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=494762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:07:04,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=494765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:07:26,266 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-06 00:07:27,125 INFO [train.py:968] (1/2) Epoch 11, batch 39250, giga_loss[loss=0.2587, simple_loss=0.3276, pruned_loss=0.09488, over 28681.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3412, pruned_loss=0.09939, over 5693312.96 frames. ], libri_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1278, over 5682978.59 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3379, pruned_loss=0.09673, over 5703286.65 frames. ], batch size: 92, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:07:34,377 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=494794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:07:34,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4093, 1.1828, 4.7447, 3.4755], device='cuda:1'), covar=tensor([0.1635, 0.2742, 0.0312, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0651, 0.0575, 0.0845, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:07:51,746 INFO [optim.py:369] (1/2) 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 00:07:58,626 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494821.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:08:00,726 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=494823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:08:12,082 INFO [train.py:968] (1/2) Epoch 11, batch 39300, giga_loss[loss=0.3578, simple_loss=0.4094, pruned_loss=0.1531, over 28679.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3456, pruned_loss=0.1019, over 5695634.45 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3772, pruned_loss=0.1281, over 5688652.78 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.342, pruned_loss=0.09885, over 5698600.07 frames. ], batch size: 307, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:08:29,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3145, 2.0803, 1.6458, 1.8505], device='cuda:1'), covar=tensor([0.0699, 0.0756, 0.0970, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0437, 0.0500, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:08:57,528 INFO [train.py:968] (1/2) Epoch 11, batch 39350, giga_loss[loss=0.3024, simple_loss=0.3781, pruned_loss=0.1133, over 28905.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3486, pruned_loss=0.103, over 5695369.90 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1281, over 5691434.35 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3452, pruned_loss=0.1001, over 5695304.02 frames. ], batch size: 213, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:09:19,590 INFO [optim.py:369] (1/2) 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 00:09:40,883 INFO [train.py:968] (1/2) Epoch 11, batch 39400, giga_loss[loss=0.2428, simple_loss=0.324, pruned_loss=0.08083, over 28856.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3501, pruned_loss=0.1035, over 5682128.30 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1282, over 5684234.74 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3465, pruned_loss=0.1005, over 5687904.03 frames. ], batch size: 66, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:09:46,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4615, 2.2068, 1.6322, 0.7441], device='cuda:1'), covar=tensor([0.3810, 0.1920, 0.3135, 0.4315], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1465, 0.1485, 0.1266], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 00:09:48,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1624, 2.6385, 1.1970, 1.3553], device='cuda:1'), covar=tensor([0.0880, 0.0380, 0.0873, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0353, 0.0506, 0.0336, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 00:10:24,495 INFO [train.py:968] (1/2) Epoch 11, batch 39450, giga_loss[loss=0.283, simple_loss=0.353, pruned_loss=0.1065, over 28865.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.349, pruned_loss=0.1019, over 5694319.29 frames. ], libri_tot_loss[loss=0.317, simple_loss=0.3774, pruned_loss=0.1283, over 5687350.79 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3457, pruned_loss=0.09915, over 5696258.05 frames. ], batch size: 186, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:10:37,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 00:10:38,305 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 00:10:42,784 INFO [optim.py:369] (1/2) 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 00:11:02,148 INFO [train.py:968] (1/2) Epoch 11, batch 39500, giga_loss[loss=0.2336, simple_loss=0.3227, pruned_loss=0.07228, over 28953.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3506, pruned_loss=0.1036, over 5690048.89 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.3783, pruned_loss=0.129, over 5684110.34 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3462, pruned_loss=0.09975, over 5694649.51 frames. ], batch size: 174, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:11:36,804 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 39550, libri_loss[loss=0.3818, simple_loss=0.4267, pruned_loss=0.1685, over 29240.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3519, pruned_loss=0.1047, over 5702643.45 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1295, over 5687828.74 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3466, pruned_loss=0.09997, over 5703283.58 frames. ], batch size: 94, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:11:46,651 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:12:02,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6274, 1.8270, 1.8506, 1.4071], device='cuda:1'), covar=tensor([0.1625, 0.2062, 0.1296, 0.1454], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0692, 0.0868, 0.0770], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 00:12:25,359 INFO [train.py:968] (1/2) Epoch 11, batch 39600, giga_loss[loss=0.262, simple_loss=0.3404, pruned_loss=0.09181, over 28905.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3524, pruned_loss=0.105, over 5708042.95 frames. ], libri_tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1296, over 5684592.85 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3473, pruned_loss=0.1005, over 5711625.13 frames. ], batch size: 227, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:12:30,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9095, 1.7771, 1.4040, 1.4784], device='cuda:1'), covar=tensor([0.0734, 0.0669, 0.0952, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0437, 0.0498, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:13:05,730 INFO [train.py:968] (1/2) Epoch 11, batch 39650, giga_loss[loss=0.2854, simple_loss=0.362, pruned_loss=0.1044, over 28864.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3555, pruned_loss=0.1065, over 5708314.21 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3788, pruned_loss=0.1295, over 5687293.13 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3511, pruned_loss=0.1025, over 5709138.99 frames. ], batch size: 145, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:13:12,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=495196.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:13:14,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=495198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:13:27,976 INFO [optim.py:369] (1/2) 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 00:13:44,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495237.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:13:45,699 INFO [train.py:968] (1/2) Epoch 11, batch 39700, giga_loss[loss=0.3067, simple_loss=0.3864, pruned_loss=0.1135, over 28712.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3588, pruned_loss=0.1083, over 5703324.03 frames. ], libri_tot_loss[loss=0.3193, simple_loss=0.3791, pruned_loss=0.1297, over 5683789.21 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3542, pruned_loss=0.1043, over 5707869.81 frames. ], batch size: 262, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:13:46,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495240.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:13:53,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9532, 1.1598, 1.0930, 0.8156], device='cuda:1'), covar=tensor([0.1720, 0.1820, 0.0968, 0.1664], device='cuda:1'), in_proj_covar=tensor([0.1728, 0.1625, 0.1592, 0.1690], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 00:14:10,029 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495269.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:14:26,924 INFO [train.py:968] (1/2) Epoch 11, batch 39750, giga_loss[loss=0.3496, simple_loss=0.398, pruned_loss=0.1506, over 26710.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3603, pruned_loss=0.109, over 5696392.58 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3792, pruned_loss=0.1296, over 5677984.47 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3563, pruned_loss=0.1054, over 5705636.77 frames. ], batch size: 555, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:14:50,493 INFO [optim.py:369] (1/2) 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 00:15:11,184 INFO [train.py:968] (1/2) Epoch 11, batch 39800, giga_loss[loss=0.274, simple_loss=0.3461, pruned_loss=0.101, over 28898.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3613, pruned_loss=0.1093, over 5699471.78 frames. ], libri_tot_loss[loss=0.3192, simple_loss=0.3792, pruned_loss=0.1296, over 5679133.55 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3581, pruned_loss=0.1065, over 5705721.39 frames. ], batch size: 119, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:15:11,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495339.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:15:12,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:15:13,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495342.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:15:15,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:15:35,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9281, 3.7089, 3.5432, 1.8154], device='cuda:1'), covar=tensor([0.0624, 0.0833, 0.0776, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.1033, 0.0971, 0.0843, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 00:15:38,609 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495371.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:15:39,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:15:52,775 INFO [train.py:968] (1/2) Epoch 11, batch 39850, giga_loss[loss=0.3535, simple_loss=0.4148, pruned_loss=0.1461, over 29087.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3627, pruned_loss=0.1106, over 5702034.18 frames. ], libri_tot_loss[loss=0.3198, simple_loss=0.3797, pruned_loss=0.13, over 5678076.95 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3593, pruned_loss=0.1076, over 5708272.92 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:16:02,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-06 00:16:04,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3003, 1.2071, 4.4071, 3.3612], device='cuda:1'), covar=tensor([0.1695, 0.2728, 0.0379, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0658, 0.0580, 0.0855, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:16:13,144 INFO [optim.py:369] (1/2) 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 00:16:17,664 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-06 00:16:30,732 INFO [train.py:968] (1/2) Epoch 11, batch 39900, giga_loss[loss=0.3171, simple_loss=0.3785, pruned_loss=0.1278, over 28968.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3617, pruned_loss=0.1104, over 5710809.75 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3801, pruned_loss=0.1304, over 5683823.62 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3582, pruned_loss=0.1072, over 5710990.98 frames. ], batch size: 227, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:16:44,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=495456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:16:46,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9188, 1.7237, 1.3430, 1.5336], device='cuda:1'), covar=tensor([0.0683, 0.0674, 0.0981, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0437, 0.0498, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:16:52,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3388, 2.2392, 2.2324, 1.8269], device='cuda:1'), covar=tensor([0.1581, 0.2374, 0.1914, 0.2295], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0733, 0.0671, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 00:16:59,245 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-06 00:17:10,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3933, 1.7621, 1.5324, 1.5439], device='cuda:1'), covar=tensor([0.0735, 0.0279, 0.0299, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0114, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0057, 0.0052, 0.0088], device='cuda:1')
+2023-03-06 00:17:11,415 INFO [train.py:968] (1/2) Epoch 11, batch 39950, giga_loss[loss=0.2974, simple_loss=0.3661, pruned_loss=0.1143, over 28302.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3591, pruned_loss=0.1092, over 5712684.23 frames. ], libri_tot_loss[loss=0.3208, simple_loss=0.3805, pruned_loss=0.1306, over 5684640.01 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3555, pruned_loss=0.106, over 5712771.96 frames. ], batch size: 368, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:17:30,673 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-06 00:17:31,905 INFO [optim.py:369] (1/2) 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 00:17:32,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-06 00:17:52,820 INFO [train.py:968] (1/2) Epoch 11, batch 40000, giga_loss[loss=0.2999, simple_loss=0.3673, pruned_loss=0.1162, over 28592.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3548, pruned_loss=0.1066, over 5710289.06 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3803, pruned_loss=0.1303, over 5687784.80 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3516, pruned_loss=0.1038, over 5708143.82 frames. ], batch size: 307, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:18:09,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7700, 1.0417, 2.8322, 2.6895], device='cuda:1'), covar=tensor([0.1722, 0.2599, 0.0587, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0658, 0.0580, 0.0855, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:18:31,688 INFO [train.py:968] (1/2) Epoch 11, batch 40050, giga_loss[loss=0.291, simple_loss=0.3689, pruned_loss=0.1065, over 28689.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3532, pruned_loss=0.1053, over 5716772.91 frames. ], libri_tot_loss[loss=0.3196, simple_loss=0.3795, pruned_loss=0.1298, over 5691939.31 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3504, pruned_loss=0.1026, over 5712434.75 frames. ], batch size: 242, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:18:39,962 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=495599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:18:41,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=495602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:18:51,180 INFO [optim.py:369] (1/2) 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 00:19:00,079 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:19:04,045 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=495631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:19:08,771 INFO [train.py:968] (1/2) Epoch 11, batch 40100, libri_loss[loss=0.3885, simple_loss=0.4297, pruned_loss=0.1737, over 28535.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3565, pruned_loss=0.1056, over 5714183.48 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3799, pruned_loss=0.1301, over 5684669.40 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1023, over 5718290.89 frames. ], batch size: 106, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:19:52,426 INFO [train.py:968] (1/2) Epoch 11, batch 40150, giga_loss[loss=0.2948, simple_loss=0.3658, pruned_loss=0.1119, over 28669.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3566, pruned_loss=0.1053, over 5707636.49 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3802, pruned_loss=0.1303, over 5686958.99 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3534, pruned_loss=0.1022, over 5708947.99 frames. ], batch size: 262, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:20:12,920 INFO [optim.py:369] (1/2) 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 00:20:30,751 INFO [train.py:968] (1/2) Epoch 11, batch 40200, giga_loss[loss=0.2858, simple_loss=0.3521, pruned_loss=0.1098, over 28912.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3558, pruned_loss=0.1061, over 5702426.06 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.3801, pruned_loss=0.1303, over 5680945.89 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3527, pruned_loss=0.103, over 5709410.77 frames. ], batch size: 145, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:20:36,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-06 00:21:10,163 INFO [train.py:968] (1/2) Epoch 11, batch 40250, giga_loss[loss=0.2532, simple_loss=0.3305, pruned_loss=0.08796, over 28518.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3553, pruned_loss=0.107, over 5707992.23 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3803, pruned_loss=0.1304, over 5686642.45 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3521, pruned_loss=0.1039, over 5708817.26 frames. ], batch size: 336, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:21:24,194 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495806.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:21:32,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 00:21:32,468 INFO [optim.py:369] (1/2) 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 00:21:43,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4552, 4.2811, 4.0316, 1.7922], device='cuda:1'), covar=tensor([0.0518, 0.0703, 0.0711, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.1037, 0.0971, 0.0847, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 00:21:48,857 INFO [zipformer.py:1188] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.2935, 1.5389, 1.4991, 1.2519], device='cuda:1'), covar=tensor([0.2848, 0.1979, 0.1554, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.1732, 0.1627, 0.1590, 0.1698], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 00:21:51,404 INFO [train.py:968] (1/2) Epoch 11, batch 40300, giga_loss[loss=0.2579, simple_loss=0.3262, pruned_loss=0.09477, over 29003.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.354, pruned_loss=0.1075, over 5700083.55 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3803, pruned_loss=0.1304, over 5683806.85 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3507, pruned_loss=0.1044, over 5703126.51 frames. ], batch size: 128, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:22:35,023 INFO [train.py:968] (1/2) Epoch 11, batch 40350, libri_loss[loss=0.3823, simple_loss=0.4359, pruned_loss=0.1644, over 29528.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3521, pruned_loss=0.1071, over 5711774.22 frames. ], libri_tot_loss[loss=0.3207, simple_loss=0.3805, pruned_loss=0.1305, over 5686032.17 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.349, pruned_loss=0.1043, over 5712244.80 frames. ], batch size: 89, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:22:35,555 INFO [scaling.py:679] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 40400, giga_loss[loss=0.2732, simple_loss=0.3502, pruned_loss=0.09808, over 28998.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3507, pruned_loss=0.1066, over 5691279.56 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3806, pruned_loss=0.1307, over 5661284.57 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3476, pruned_loss=0.1038, over 5714845.33 frames. ], batch size: 164, lr: 2.89e-03, grad_scale: 8.0
+2023-03-06 00:23:45,044 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=495976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:23:55,480 INFO [train.py:968] (1/2) Epoch 11, batch 40450, giga_loss[loss=0.2701, simple_loss=0.3354, pruned_loss=0.1024, over 28783.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3469, pruned_loss=0.1044, over 5688927.13 frames. ], libri_tot_loss[loss=0.3212, simple_loss=0.3807, pruned_loss=0.1308, over 5652428.73 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3439, pruned_loss=0.1018, over 5716519.70 frames. ], batch size: 119, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:24:07,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=496001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:24:19,660 INFO [optim.py:369] (1/2) 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 00:24:35,008 INFO [train.py:968] (1/2) Epoch 11, batch 40500, giga_loss[loss=0.2199, simple_loss=0.2931, pruned_loss=0.07332, over 28874.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3428, pruned_loss=0.1024, over 5691527.63 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1304, over 5648803.12 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3393, pruned_loss=0.09929, over 5719208.76 frames. ], batch size: 112, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:24:54,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-06 00:25:07,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6002, 3.6195, 1.7454, 1.7250], device='cuda:1'), covar=tensor([0.0872, 0.0279, 0.0851, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0512, 0.0340, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 00:25:15,880 INFO [train.py:968] (1/2) Epoch 11, batch 40550, giga_loss[loss=0.2421, simple_loss=0.3251, pruned_loss=0.07953, over 28911.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3391, pruned_loss=0.09995, over 5694513.21 frames. ], libri_tot_loss[loss=0.3203, simple_loss=0.38, pruned_loss=0.1303, over 5650215.03 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3358, pruned_loss=0.09707, over 5716222.06 frames. ], batch size: 164, lr: 2.89e-03, grad_scale: 4.0
+2023-03-06 00:25:24,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 00:25:29,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 00:25:37,801 INFO [optim.py:369] (1/2) 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 00:25:55,777 INFO [train.py:968] (1/2) Epoch 11, batch 40600, giga_loss[loss=0.2587, simple_loss=0.3324, pruned_loss=0.09249, over 28470.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3418, pruned_loss=0.1012, over 5688048.80 frames. ], libri_tot_loss[loss=0.3205, simple_loss=0.3802, pruned_loss=0.1304, over 5648412.24 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3377, pruned_loss=0.09787, over 5707663.65 frames. ], batch size: 71, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:25:59,563 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:26:01,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:26:05,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4013, 4.1926, 3.9759, 1.8502], device='cuda:1'), covar=tensor([0.0495, 0.0633, 0.0613, 0.2034], device='cuda:1'), in_proj_covar=tensor([0.1042, 0.0975, 0.0849, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 00:26:20,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 00:26:26,110 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:26:30,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=496181.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:26:36,381 INFO [train.py:968] (1/2) Epoch 11, batch 40650, giga_loss[loss=0.2793, simple_loss=0.3517, pruned_loss=0.1035, over 28944.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3449, pruned_loss=0.102, over 5698140.54 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.38, pruned_loss=0.1302, over 5653458.89 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3409, pruned_loss=0.0989, over 5710221.39 frames. ], batch size: 213, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:26:55,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2359, 1.3811, 1.5084, 1.3468], device='cuda:1'), covar=tensor([0.1108, 0.0955, 0.1398, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0733, 0.0675, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 00:26:56,250 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 11, batch 40700, giga_loss[loss=0.2831, simple_loss=0.3606, pruned_loss=0.1027, over 28629.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3483, pruned_loss=0.1032, over 5701199.49 frames. ], libri_tot_loss[loss=0.3204, simple_loss=0.3802, pruned_loss=0.1303, over 5655885.34 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3441, pruned_loss=0.1, over 5709483.75 frames. ], batch size: 262, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:27:18,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2790, 1.0523, 4.5309, 3.4771], device='cuda:1'), covar=tensor([0.1730, 0.2923, 0.0381, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0661, 0.0582, 0.0858, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:27:26,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5639, 4.3824, 4.1401, 1.8630], device='cuda:1'), covar=tensor([0.0446, 0.0586, 0.0597, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.1043, 0.0974, 0.0849, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 00:27:41,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1859, 2.1598, 1.9895, 1.9550], device='cuda:1'), covar=tensor([0.1355, 0.1994, 0.1770, 0.1821], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0735, 0.0677, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 00:27:57,857 INFO [train.py:968] (1/2) Epoch 11, batch 40750, giga_loss[loss=0.2785, simple_loss=0.3463, pruned_loss=0.1053, over 28707.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3516, pruned_loss=0.1048, over 5692716.84 frames. ], libri_tot_loss[loss=0.3209, simple_loss=0.3806, pruned_loss=0.1306, over 5639680.42 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.347, pruned_loss=0.1011, over 5716220.29 frames. ], batch size: 119, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:28:21,702 INFO [optim.py:369] (1/2) 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 00:28:29,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496324.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:28:31,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496327.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:28:36,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 00:28:41,788 INFO [train.py:968] (1/2) Epoch 11, batch 40800, libri_loss[loss=0.2728, simple_loss=0.3436, pruned_loss=0.101, over 29565.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3534, pruned_loss=0.1057, over 5697430.75 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.38, pruned_loss=0.1302, over 5644921.47 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3498, pruned_loss=0.1027, over 5712278.67 frames. ], batch size: 77, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:28:50,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=496351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:28:54,490 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:28:57,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496356.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:28:57,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:29:23,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:29:27,614 INFO [train.py:968] (1/2) Epoch 11, batch 40850, giga_loss[loss=0.3128, simple_loss=0.3803, pruned_loss=0.1227, over 28716.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3559, pruned_loss=0.1078, over 5689583.67 frames. ], libri_tot_loss[loss=0.3201, simple_loss=0.3799, pruned_loss=0.1301, over 5640471.38 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3527, pruned_loss=0.1051, over 5706423.84 frames. ], batch size: 262, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:29:54,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-06 00:29:58,869 INFO [optim.py:369] (1/2) 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 00:30:14,332 INFO [train.py:968] (1/2) Epoch 11, batch 40900, libri_loss[loss=0.3264, simple_loss=0.3808, pruned_loss=0.136, over 29539.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1161, over 5680256.71 frames. ], libri_tot_loss[loss=0.3206, simple_loss=0.3801, pruned_loss=0.1305, over 5653689.87 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3609, pruned_loss=0.1126, over 5683547.38 frames. ], batch size: 80, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:30:58,904 INFO [train.py:968] (1/2) Epoch 11, batch 40950, giga_loss[loss=0.3361, simple_loss=0.4012, pruned_loss=0.1355, over 28854.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3703, pruned_loss=0.1199, over 5683079.45 frames. ], libri_tot_loss[loss=0.3202, simple_loss=0.3798, pruned_loss=0.1303, over 5655594.02 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3671, pruned_loss=0.117, over 5684622.78 frames. ], batch size: 119, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:31:06,466 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=496494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:31:09,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=496497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:31:13,754 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-06 00:31:26,110 INFO [optim.py:369] (1/2) 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 00:31:34,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=496526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:31:44,889 INFO [train.py:968] (1/2) Epoch 11, batch 41000, giga_loss[loss=0.336, simple_loss=0.4008, pruned_loss=0.1356, over 28624.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.375, pruned_loss=0.1233, over 5674395.31 frames. ], libri_tot_loss[loss=0.32, simple_loss=0.3798, pruned_loss=0.1301, over 5661524.04 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3723, pruned_loss=0.1209, over 5671018.77 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:31:58,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 00:32:31,113 INFO [train.py:968] (1/2) Epoch 11, batch 41050, giga_loss[loss=0.3337, simple_loss=0.3917, pruned_loss=0.1379, over 28963.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.382, pruned_loss=0.1301, over 5671666.28 frames. ], libri_tot_loss[loss=0.3197, simple_loss=0.3796, pruned_loss=0.13, over 5660863.00 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3801, pruned_loss=0.1283, over 5669622.15 frames. ], batch size: 106, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:32:55,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7080, 1.7915, 1.3611, 1.3245], device='cuda:1'), covar=tensor([0.0761, 0.0557, 0.0902, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0441, 0.0499, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:32:57,382 INFO [optim.py:369] (1/2) 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 00:33:14,311 INFO [train.py:968] (1/2) Epoch 11, batch 41100, giga_loss[loss=0.3238, simple_loss=0.3899, pruned_loss=0.1289, over 28783.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3862, pruned_loss=0.1332, over 5664040.84 frames. ], libri_tot_loss[loss=0.3188, simple_loss=0.379, pruned_loss=0.1294, over 5655547.17 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3854, pruned_loss=0.1324, over 5668317.23 frames. ], batch size: 284, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:33:55,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3440, 1.6448, 1.3593, 1.5528], device='cuda:1'), covar=tensor([0.0782, 0.0311, 0.0321, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 00:34:06,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5389, 1.7262, 1.4482, 1.2874], device='cuda:1'), covar=tensor([0.2254, 0.1805, 0.1601, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.1728, 0.1632, 0.1601, 0.1695], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 00:34:06,855 INFO [train.py:968] (1/2) Epoch 11, batch 41150, giga_loss[loss=0.3108, simple_loss=0.3795, pruned_loss=0.1211, over 28859.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3882, pruned_loss=0.1355, over 5650315.26 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3787, pruned_loss=0.1291, over 5650953.14 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.388, pruned_loss=0.1352, over 5658012.01 frames. ], batch size: 174, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:34:11,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 00:34:14,528 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=496696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:34:41,706 INFO [optim.py:369] (1/2) 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 00:35:06,127 INFO [train.py:968] (1/2) Epoch 11, batch 41200, giga_loss[loss=0.3432, simple_loss=0.3988, pruned_loss=0.1438, over 28726.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3919, pruned_loss=0.14, over 5625164.94 frames. ], libri_tot_loss[loss=0.3185, simple_loss=0.3787, pruned_loss=0.1291, over 5650953.14 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3917, pruned_loss=0.1397, over 5631155.42 frames. ], batch size: 242, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:36:00,754 INFO [train.py:968] (1/2) Epoch 11, batch 41250, giga_loss[loss=0.4805, simple_loss=0.4823, pruned_loss=0.2393, over 26612.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3956, pruned_loss=0.1439, over 5616466.79 frames. ], libri_tot_loss[loss=0.3186, simple_loss=0.3788, pruned_loss=0.1292, over 5651996.86 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3955, pruned_loss=0.1438, over 5619857.78 frames. ], batch size: 555, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:36:31,044 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 41300, giga_loss[loss=0.5121, simple_loss=0.4989, pruned_loss=0.2627, over 26501.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.4004, pruned_loss=0.148, over 5624960.35 frames. ], libri_tot_loss[loss=0.3189, simple_loss=0.3791, pruned_loss=0.1294, over 5654161.76 frames. ], giga_tot_loss[loss=0.3483, simple_loss=0.4005, pruned_loss=0.1481, over 5624675.45 frames. ], batch size: 555, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:37:25,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3257, 1.6714, 1.2849, 0.7895], device='cuda:1'), covar=tensor([0.3263, 0.2134, 0.1851, 0.3381], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1504, 0.1510, 0.1293], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 00:37:43,403 INFO [train.py:968] (1/2) Epoch 11, batch 41350, giga_loss[loss=0.3123, simple_loss=0.3779, pruned_loss=0.1233, over 28712.00 frames. ], tot_loss[loss=0.3486, simple_loss=0.4003, pruned_loss=0.1485, over 5633271.34 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3784, pruned_loss=0.1289, over 5659205.94 frames. ], giga_tot_loss[loss=0.3504, simple_loss=0.4017, pruned_loss=0.1496, over 5627561.28 frames. ], batch size: 262, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:37:56,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0652, 5.8499, 5.5261, 2.8439], device='cuda:1'), covar=tensor([0.0571, 0.0763, 0.0888, 0.1662], device='cuda:1'), in_proj_covar=tensor([0.1059, 0.0992, 0.0868, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 00:38:06,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=496911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:38:12,735 INFO [optim.py:369] (1/2) 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 00:38:31,209 INFO [train.py:968] (1/2) Epoch 11, batch 41400, libri_loss[loss=0.3574, simple_loss=0.4117, pruned_loss=0.1516, over 29669.00 frames. ], tot_loss[loss=0.3461, simple_loss=0.3978, pruned_loss=0.1472, over 5638074.92 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3783, pruned_loss=0.1287, over 5663742.20 frames. ], giga_tot_loss[loss=0.3484, simple_loss=0.3994, pruned_loss=0.1487, over 5629008.57 frames. ], batch size: 88, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:39:04,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4711, 4.4746, 1.6510, 1.7109], device='cuda:1'), covar=tensor([0.0950, 0.0250, 0.0831, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0511, 0.0338, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 00:39:17,673 INFO [train.py:968] (1/2) Epoch 11, batch 41450, giga_loss[loss=0.3561, simple_loss=0.4051, pruned_loss=0.1536, over 27902.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3969, pruned_loss=0.1464, over 5635092.83 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3782, pruned_loss=0.1285, over 5667111.21 frames. ], giga_tot_loss[loss=0.3478, simple_loss=0.399, pruned_loss=0.1484, over 5623726.12 frames. ], batch size: 412, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:39:49,132 INFO [optim.py:369] (1/2) 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 00:39:49,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-06 00:39:58,294 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 00:40:10,019 INFO [train.py:968] (1/2) Epoch 11, batch 41500, giga_loss[loss=0.3349, simple_loss=0.3941, pruned_loss=0.1378, over 28987.00 frames. ], tot_loss[loss=0.3434, simple_loss=0.3963, pruned_loss=0.1453, over 5616631.81 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3784, pruned_loss=0.1287, over 5660232.52 frames. ], giga_tot_loss[loss=0.3459, simple_loss=0.398, pruned_loss=0.1469, over 5614413.14 frames. ], batch size: 213, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:40:43,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:40:43,366 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-06 00:41:00,265 INFO [train.py:968] (1/2) Epoch 11, batch 41550, libri_loss[loss=0.3509, simple_loss=0.4016, pruned_loss=0.1501, over 29663.00 frames. ], tot_loss[loss=0.3471, simple_loss=0.3993, pruned_loss=0.1475, over 5607978.28 frames. ], libri_tot_loss[loss=0.3182, simple_loss=0.3786, pruned_loss=0.1289, over 5654784.20 frames. ], giga_tot_loss[loss=0.3493, simple_loss=0.4008, pruned_loss=0.1489, over 5610100.57 frames. ], batch size: 91, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:41:34,557 INFO [optim.py:369] (1/2) 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 00:41:59,047 INFO [train.py:968] (1/2) Epoch 11, batch 41600, giga_loss[loss=0.3318, simple_loss=0.3904, pruned_loss=0.1366, over 28596.00 frames. ], tot_loss[loss=0.3455, simple_loss=0.3977, pruned_loss=0.1466, over 5600917.40 frames. ], libri_tot_loss[loss=0.318, simple_loss=0.3784, pruned_loss=0.1288, over 5659727.36 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.3994, pruned_loss=0.1482, over 5597056.42 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:42:30,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9888, 1.2314, 1.3159, 1.0955], device='cuda:1'), covar=tensor([0.1328, 0.1092, 0.1879, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0733, 0.0676, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 00:42:31,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2814, 4.0668, 3.8412, 1.8025], device='cuda:1'), covar=tensor([0.0587, 0.0780, 0.0804, 0.2089], device='cuda:1'), in_proj_covar=tensor([0.1062, 0.0997, 0.0869, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 00:42:46,679 INFO [train.py:968] (1/2) Epoch 11, batch 41650, giga_loss[loss=0.3073, simple_loss=0.3809, pruned_loss=0.1169, over 29114.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3936, pruned_loss=0.1415, over 5611537.51 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3782, pruned_loss=0.1286, over 5656498.65 frames. ], giga_tot_loss[loss=0.3412, simple_loss=0.3956, pruned_loss=0.1434, over 5609326.52 frames. ], batch size: 155, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:43:11,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:43:14,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:43:14,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3999, 3.1932, 1.4420, 1.5439], device='cuda:1'), covar=tensor([0.0907, 0.0376, 0.0908, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0512, 0.0340, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 00:43:15,007 INFO [optim.py:369] (1/2) 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 00:43:19,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4790, 1.5699, 1.2300, 1.1588], device='cuda:1'), covar=tensor([0.0850, 0.0536, 0.1020, 0.1055], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0440, 0.0500, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:43:21,742 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 00:43:33,383 INFO [train.py:968] (1/2) Epoch 11, batch 41700, giga_loss[loss=0.2807, simple_loss=0.3536, pruned_loss=0.1039, over 28970.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3903, pruned_loss=0.1377, over 5633215.05 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3777, pruned_loss=0.1283, over 5665234.44 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3928, pruned_loss=0.1398, over 5622235.27 frames. ], batch size: 106, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:43:34,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:43:40,048 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:44:22,648 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497286.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:44:22,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:44:24,522 INFO [train.py:968] (1/2) Epoch 11, batch 41750, giga_loss[loss=0.3242, simple_loss=0.3801, pruned_loss=0.1341, over 28876.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3879, pruned_loss=0.1362, over 5630634.49 frames. ], libri_tot_loss[loss=0.3172, simple_loss=0.3777, pruned_loss=0.1284, over 5665664.54 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3901, pruned_loss=0.1379, over 5620706.09 frames. ], batch size: 112, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 00:44:54,195 INFO [optim.py:369] (1/2) 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 00:44:55,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9428, 1.0910, 3.4170, 3.0630], device='cuda:1'), covar=tensor([0.1745, 0.2567, 0.0450, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0665, 0.0583, 0.0857, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:45:14,064 INFO [train.py:968] (1/2) Epoch 11, batch 41800, giga_loss[loss=0.2618, simple_loss=0.341, pruned_loss=0.09133, over 28832.00 frames. ], tot_loss[loss=0.3272, simple_loss=0.3857, pruned_loss=0.1344, over 5618858.25 frames. ], libri_tot_loss[loss=0.3174, simple_loss=0.3776, pruned_loss=0.1286, over 5650476.94 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3876, pruned_loss=0.1356, over 5623440.48 frames. ], batch size: 145, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:45:30,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:46:04,758 INFO [train.py:968] (1/2) Epoch 11, batch 41850, giga_loss[loss=0.333, simple_loss=0.396, pruned_loss=0.135, over 28718.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3853, pruned_loss=0.1341, over 5635712.76 frames. ], libri_tot_loss[loss=0.3181, simple_loss=0.378, pruned_loss=0.129, over 5657786.70 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3868, pruned_loss=0.1349, over 5631941.75 frames. ], batch size: 242, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:46:34,305 INFO [optim.py:369] (1/2) 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 00:46:43,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:46:45,422 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 41900, giga_loss[loss=0.3842, simple_loss=0.4045, pruned_loss=0.182, over 23754.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3849, pruned_loss=0.1338, over 5641832.39 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3777, pruned_loss=0.1289, over 5659500.65 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3864, pruned_loss=0.1347, over 5637257.64 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:47:15,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-06 00:47:15,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6128, 1.7025, 1.3158, 1.3324], device='cuda:1'), covar=tensor([0.0886, 0.0637, 0.1050, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0442, 0.0502, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 00:47:16,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:47:49,615 INFO [train.py:968] (1/2) Epoch 11, batch 41950, giga_loss[loss=0.2818, simple_loss=0.3544, pruned_loss=0.1046, over 28636.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3818, pruned_loss=0.131, over 5637964.24 frames. ], libri_tot_loss[loss=0.3176, simple_loss=0.3775, pruned_loss=0.1288, over 5661152.86 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3832, pruned_loss=0.1317, over 5632785.43 frames. ], batch size: 242, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:48:22,401 INFO [optim.py:369] (1/2) 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 00:48:39,710 INFO [train.py:968] (1/2) Epoch 11, batch 42000, libri_loss[loss=0.3354, simple_loss=0.3908, pruned_loss=0.14, over 29143.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3812, pruned_loss=0.1277, over 5648118.15 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3769, pruned_loss=0.1284, over 5667992.82 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.383, pruned_loss=0.1287, over 5637098.85 frames. ], batch size: 101, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:48:39,711 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 00:48:48,331 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 00:49:38,859 INFO [train.py:968] (1/2) Epoch 11, batch 42050, giga_loss[loss=0.3192, simple_loss=0.3934, pruned_loss=0.1225, over 28624.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3823, pruned_loss=0.1267, over 5660745.10 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.377, pruned_loss=0.1285, over 5671303.88 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3837, pruned_loss=0.1274, over 5648919.41 frames. ], batch size: 307, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:49:51,479 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:50:08,916 INFO [optim.py:369] (1/2) 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 00:50:28,859 INFO [train.py:968] (1/2) Epoch 11, batch 42100, giga_loss[loss=0.2973, simple_loss=0.3687, pruned_loss=0.1129, over 28688.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3833, pruned_loss=0.1279, over 5669293.92 frames. ], libri_tot_loss[loss=0.3169, simple_loss=0.3769, pruned_loss=0.1285, over 5676228.96 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3847, pruned_loss=0.1284, over 5655388.10 frames. ], batch size: 242, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:50:50,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497661.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:51:11,256 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:51:14,188 INFO [train.py:968] (1/2) Epoch 11, batch 42150, giga_loss[loss=0.3993, simple_loss=0.4339, pruned_loss=0.1823, over 28865.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3841, pruned_loss=0.129, over 5665449.64 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3772, pruned_loss=0.1287, over 5679724.87 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.385, pruned_loss=0.1293, over 5650995.30 frames. ], batch size: 174, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:51:44,832 INFO [optim.py:369] (1/2) 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 00:51:51,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:00,641 INFO [train.py:968] (1/2) Epoch 11, batch 42200, giga_loss[loss=0.3231, simple_loss=0.3801, pruned_loss=0.133, over 28572.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.381, pruned_loss=0.1281, over 5667103.52 frames. ], libri_tot_loss[loss=0.3171, simple_loss=0.3769, pruned_loss=0.1287, over 5673941.83 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3821, pruned_loss=0.1282, over 5661482.82 frames. ], batch size: 336, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:52:19,632 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:23,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:48,059 INFO [train.py:968] (1/2) Epoch 11, batch 42250, giga_loss[loss=0.3329, simple_loss=0.3693, pruned_loss=0.1482, over 23556.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3798, pruned_loss=0.1286, over 5664490.06 frames. ], libri_tot_loss[loss=0.3173, simple_loss=0.3771, pruned_loss=0.1287, over 5678474.35 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3806, pruned_loss=0.1287, over 5655743.25 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:52:48,879 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:52:53,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5669, 1.7221, 1.5369, 1.3791], device='cuda:1'), covar=tensor([0.2139, 0.1793, 0.1555, 0.1903], device='cuda:1'), in_proj_covar=tensor([0.1734, 0.1631, 0.1598, 0.1698], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 00:52:54,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3547, 3.7392, 1.6045, 1.5231], device='cuda:1'), covar=tensor([0.0996, 0.0400, 0.0862, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0513, 0.0341, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 00:53:03,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497804.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 00:53:05,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497807.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:53:22,276 INFO [optim.py:369] (1/2) 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 00:53:37,451 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497836.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 00:53:40,358 INFO [train.py:968] (1/2) Epoch 11, batch 42300, giga_loss[loss=0.3474, simple_loss=0.417, pruned_loss=0.1389, over 28988.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3809, pruned_loss=0.1288, over 5657619.14 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3773, pruned_loss=0.1288, over 5669105.10 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3814, pruned_loss=0.1287, over 5658126.19 frames. ], batch size: 145, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:53:53,188 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=497853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:06,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=497871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:08,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=497874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:22,163 INFO [train.py:968] (1/2) Epoch 11, batch 42350, giga_loss[loss=0.3016, simple_loss=0.3687, pruned_loss=0.1172, over 28555.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3814, pruned_loss=0.1279, over 5665930.45 frames. ], libri_tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.129, over 5665900.60 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3815, pruned_loss=0.1277, over 5669322.36 frames. ], batch size: 85, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:54:36,041 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=497903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:54:51,853 INFO [optim.py:369] (1/2) 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 00:55:09,123 INFO [train.py:968] (1/2) Epoch 11, batch 42400, giga_loss[loss=0.3298, simple_loss=0.3903, pruned_loss=0.1346, over 28370.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3798, pruned_loss=0.1266, over 5669272.85 frames. ], libri_tot_loss[loss=0.3167, simple_loss=0.3768, pruned_loss=0.1283, over 5670927.37 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3808, pruned_loss=0.1271, over 5667647.54 frames. ], batch size: 368, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:55:18,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2764, 1.7129, 1.4323, 1.5157], device='cuda:1'), covar=tensor([0.0730, 0.0286, 0.0289, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 00:55:40,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=497976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:55:54,440 INFO [train.py:968] (1/2) Epoch 11, batch 42450, giga_loss[loss=0.3705, simple_loss=0.4104, pruned_loss=0.1653, over 27684.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3799, pruned_loss=0.1272, over 5669253.80 frames. ], libri_tot_loss[loss=0.3165, simple_loss=0.3767, pruned_loss=0.1281, over 5676325.30 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3809, pruned_loss=0.1277, over 5663085.11 frames. ], batch size: 472, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 00:56:21,845 INFO [optim.py:369] (1/2) 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 00:56:39,067 INFO [train.py:968] (1/2) Epoch 11, batch 42500, giga_loss[loss=0.3001, simple_loss=0.3681, pruned_loss=0.1161, over 28739.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3762, pruned_loss=0.1246, over 5683943.94 frames. ], libri_tot_loss[loss=0.3158, simple_loss=0.3763, pruned_loss=0.1276, over 5682307.71 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3774, pruned_loss=0.1253, over 5673933.70 frames. ], batch size: 119, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:56:59,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=498059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:57:27,604 INFO [train.py:968] (1/2) Epoch 11, batch 42550, giga_loss[loss=0.2794, simple_loss=0.3454, pruned_loss=0.1067, over 28833.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3746, pruned_loss=0.1244, over 5680060.26 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3757, pruned_loss=0.1272, over 5687933.11 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3761, pruned_loss=0.1253, over 5667089.18 frames. ], batch size: 112, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:57:57,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=498119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:58:02,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=498122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:58:02,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7501, 2.5693, 1.5806, 0.8081], device='cuda:1'), covar=tensor([0.5847, 0.2705, 0.3075, 0.5367], device='cuda:1'), in_proj_covar=tensor([0.1568, 0.1495, 0.1495, 0.1288], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 00:58:02,542 INFO [optim.py:369] (1/2) 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 00:58:17,879 INFO [train.py:968] (1/2) Epoch 11, batch 42600, giga_loss[loss=0.2945, simple_loss=0.3626, pruned_loss=0.1132, over 28978.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1242, over 5684080.58 frames. ], libri_tot_loss[loss=0.3151, simple_loss=0.3757, pruned_loss=0.1272, over 5689808.25 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3744, pruned_loss=0.1249, over 5672094.35 frames. ], batch size: 136, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:58:30,350 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=498151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:07,378 INFO [train.py:968] (1/2) Epoch 11, batch 42650, giga_loss[loss=0.3115, simple_loss=0.3772, pruned_loss=0.1229, over 28973.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1243, over 5687617.37 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3758, pruned_loss=0.1271, over 5693890.41 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.1249, over 5674199.49 frames. ], batch size: 164, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:59:20,944 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=498202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:22,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=498205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:38,127 INFO [optim.py:369] (1/2) 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 00:59:44,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=498228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:50,942 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=498234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 00:59:55,600 INFO [train.py:968] (1/2) Epoch 11, batch 42700, giga_loss[loss=0.2844, simple_loss=0.3563, pruned_loss=0.1063, over 28926.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3717, pruned_loss=0.1243, over 5658628.92 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3755, pruned_loss=0.127, over 5680906.94 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3725, pruned_loss=0.1248, over 5659702.64 frames. ], batch size: 174, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 00:59:55,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0019, 1.1343, 1.2051, 1.0534], device='cuda:1'), covar=tensor([0.0929, 0.0891, 0.1381, 0.0965], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0737, 0.0677, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 01:00:25,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1706, 1.3872, 1.1551, 1.0173], device='cuda:1'), covar=tensor([0.1824, 0.1697, 0.1241, 0.1644], device='cuda:1'), in_proj_covar=tensor([0.1716, 0.1630, 0.1591, 0.1688], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 01:00:44,023 INFO [train.py:968] (1/2) Epoch 11, batch 42750, giga_loss[loss=0.2891, simple_loss=0.3591, pruned_loss=0.1096, over 28886.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3706, pruned_loss=0.1235, over 5659497.37 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3748, pruned_loss=0.1265, over 5685853.95 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3717, pruned_loss=0.1243, over 5655055.93 frames. ], batch size: 227, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:00:45,587 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 01:00:46,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9472, 1.1840, 1.0700, 0.8064], device='cuda:1'), covar=tensor([0.1690, 0.1831, 0.1076, 0.1566], device='cuda:1'), in_proj_covar=tensor([0.1722, 0.1638, 0.1601, 0.1695], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 01:01:16,284 INFO [optim.py:369] (1/2) 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 01:01:32,006 INFO [train.py:968] (1/2) Epoch 11, batch 42800, giga_loss[loss=0.3172, simple_loss=0.3787, pruned_loss=0.1279, over 28250.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1235, over 5661887.39 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3746, pruned_loss=0.1263, over 5678619.80 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1243, over 5664173.48 frames. ], batch size: 368, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:01:43,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 01:01:46,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1959, 1.2905, 1.1243, 0.9256], device='cuda:1'), covar=tensor([0.0829, 0.0505, 0.1044, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0442, 0.0501, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 01:02:00,200 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=498371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:02:03,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4450, 1.5145, 1.5595, 1.3587], device='cuda:1'), covar=tensor([0.1368, 0.1855, 0.1912, 0.1820], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0739, 0.0678, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 01:02:03,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=498374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:02:18,629 INFO [train.py:968] (1/2) Epoch 11, batch 42850, libri_loss[loss=0.3226, simple_loss=0.3849, pruned_loss=0.1302, over 29778.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1245, over 5666344.35 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1267, over 5680560.29 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3746, pruned_loss=0.1247, over 5666069.38 frames. ], batch size: 87, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:02:31,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=498403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:02:50,698 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 42900, giga_loss[loss=0.3124, simple_loss=0.3763, pruned_loss=0.1242, over 29046.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3745, pruned_loss=0.1241, over 5674371.25 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3752, pruned_loss=0.1265, over 5683154.59 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1244, over 5671536.48 frames. ], batch size: 136, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:03:43,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 01:03:53,808 INFO [train.py:968] (1/2) Epoch 11, batch 42950, giga_loss[loss=0.3583, simple_loss=0.4083, pruned_loss=0.1542, over 28269.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3773, pruned_loss=0.1268, over 5674047.66 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3747, pruned_loss=0.1264, over 5676906.66 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3779, pruned_loss=0.1271, over 5677419.24 frames. ], batch size: 368, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:04:09,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 01:04:27,622 INFO [optim.py:369] (1/2) 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 01:04:42,436 INFO [train.py:968] (1/2) Epoch 11, batch 43000, giga_loss[loss=0.327, simple_loss=0.3878, pruned_loss=0.1331, over 28782.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3796, pruned_loss=0.1297, over 5669455.83 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3748, pruned_loss=0.1265, over 5671230.16 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3801, pruned_loss=0.1299, over 5676424.38 frames. ], batch size: 284, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:05:09,544 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=498562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:05:12,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-06 01:05:37,133 INFO [train.py:968] (1/2) Epoch 11, batch 43050, giga_loss[loss=0.325, simple_loss=0.3842, pruned_loss=0.1329, over 28738.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3789, pruned_loss=0.1301, over 5671867.86 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3747, pruned_loss=0.1263, over 5678608.57 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3796, pruned_loss=0.1305, over 5670697.18 frames. ], batch size: 284, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:06:12,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6133, 1.8104, 1.6087, 1.4804], device='cuda:1'), covar=tensor([0.1575, 0.1980, 0.1905, 0.2070], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0733, 0.0676, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 01:06:13,755 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 43100, giga_loss[loss=0.4388, simple_loss=0.4602, pruned_loss=0.2087, over 24201.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3816, pruned_loss=0.1329, over 5654303.03 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3748, pruned_loss=0.1263, over 5674349.04 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3822, pruned_loss=0.1334, over 5657676.57 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:06:53,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9984, 1.0036, 3.3957, 2.9947], device='cuda:1'), covar=tensor([0.1640, 0.2729, 0.0535, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0667, 0.0589, 0.0866, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 01:07:10,704 INFO [train.py:968] (1/2) Epoch 11, batch 43150, giga_loss[loss=0.2723, simple_loss=0.3426, pruned_loss=0.101, over 28685.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3812, pruned_loss=0.1325, over 5648066.88 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3757, pruned_loss=0.1269, over 5660724.59 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.381, pruned_loss=0.1325, over 5661931.30 frames. ], batch size: 71, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:07:15,653 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 01:07:41,020 INFO [optim.py:369] (1/2) 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 01:07:52,987 INFO [train.py:968] (1/2) Epoch 11, batch 43200, libri_loss[loss=0.2756, simple_loss=0.3539, pruned_loss=0.09861, over 29509.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3805, pruned_loss=0.132, over 5648789.79 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3759, pruned_loss=0.127, over 5664333.59 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3804, pruned_loss=0.1321, over 5656039.10 frames. ], batch size: 82, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:08:36,768 INFO [train.py:968] (1/2) Epoch 11, batch 43250, giga_loss[loss=0.2762, simple_loss=0.3548, pruned_loss=0.09882, over 28535.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3801, pruned_loss=0.1302, over 5655241.29 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3759, pruned_loss=0.1269, over 5667070.87 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1305, over 5658545.10 frames. ], batch size: 85, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:08:46,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4810, 4.2505, 1.5635, 1.6364], device='cuda:1'), covar=tensor([0.0970, 0.0262, 0.0926, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0515, 0.0341, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 01:09:11,921 INFO [optim.py:369] (1/2) 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 01:09:26,342 INFO [train.py:968] (1/2) Epoch 11, batch 43300, giga_loss[loss=0.2997, simple_loss=0.3638, pruned_loss=0.1178, over 28724.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3779, pruned_loss=0.1285, over 5652314.10 frames. ], libri_tot_loss[loss=0.315, simple_loss=0.3759, pruned_loss=0.1271, over 5668381.51 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3779, pruned_loss=0.1286, over 5653799.86 frames. ], batch size: 262, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:09:28,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3554, 1.5134, 1.3607, 1.5289], device='cuda:1'), covar=tensor([0.0773, 0.0311, 0.0312, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0114, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 01:10:08,542 INFO [train.py:968] (1/2) Epoch 11, batch 43350, giga_loss[loss=0.3104, simple_loss=0.3708, pruned_loss=0.125, over 28996.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3763, pruned_loss=0.1275, over 5663001.72 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3756, pruned_loss=0.1268, over 5666569.84 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3765, pruned_loss=0.1279, over 5665648.28 frames. ], batch size: 136, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:10:43,216 INFO [optim.py:369] (1/2) 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 01:10:58,086 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 43400, giga_loss[loss=0.2915, simple_loss=0.3525, pruned_loss=0.1152, over 28703.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3758, pruned_loss=0.1286, over 5659912.73 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3757, pruned_loss=0.1269, over 5667750.99 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.376, pruned_loss=0.1288, over 5661025.90 frames. ], batch size: 66, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:11:06,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3611, 1.9023, 1.4063, 0.4888], device='cuda:1'), covar=tensor([0.2876, 0.1727, 0.2448, 0.4119], device='cuda:1'), in_proj_covar=tensor([0.1574, 0.1512, 0.1505, 0.1291], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 01:11:43,506 INFO [train.py:968] (1/2) Epoch 11, batch 43450, giga_loss[loss=0.3716, simple_loss=0.4252, pruned_loss=0.159, over 28814.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3781, pruned_loss=0.1297, over 5670399.27 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3757, pruned_loss=0.1267, over 5669118.65 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3782, pruned_loss=0.1301, over 5669954.69 frames. ], batch size: 199, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:11:47,048 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=498993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:12:16,874 INFO [optim.py:369] (1/2) 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 01:12:30,390 INFO [train.py:968] (1/2) Epoch 11, batch 43500, giga_loss[loss=0.34, simple_loss=0.4093, pruned_loss=0.1354, over 28730.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3827, pruned_loss=0.1317, over 5656062.12 frames. ], libri_tot_loss[loss=0.3147, simple_loss=0.3756, pruned_loss=0.1268, over 5663121.79 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.383, pruned_loss=0.132, over 5660907.64 frames. ], batch size: 284, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:13:10,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=499080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:13:12,047 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 43550, giga_loss[loss=0.2834, simple_loss=0.3573, pruned_loss=0.1048, over 28660.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3832, pruned_loss=0.1292, over 5665435.65 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3758, pruned_loss=0.1269, over 5667049.69 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3835, pruned_loss=0.1296, over 5665968.45 frames. ], batch size: 92, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:13:40,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:14:02,412 INFO [train.py:968] (1/2) Epoch 11, batch 43600, giga_loss[loss=0.4044, simple_loss=0.4344, pruned_loss=0.1872, over 27590.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3827, pruned_loss=0.1293, over 5658659.15 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3743, pruned_loss=0.1262, over 5665830.71 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3846, pruned_loss=0.1302, over 5660126.34 frames. ], batch size: 472, lr: 2.88e-03, grad_scale: 8.0
+2023-03-06 01:14:20,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0376, 1.3238, 1.0330, 0.3129], device='cuda:1'), covar=tensor([0.2482, 0.2260, 0.3330, 0.4096], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1500, 0.1491, 0.1278], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 01:14:48,840 INFO [train.py:968] (1/2) Epoch 11, batch 43650, giga_loss[loss=0.3123, simple_loss=0.3753, pruned_loss=0.1247, over 28828.00 frames. ], tot_loss[loss=0.325, simple_loss=0.386, pruned_loss=0.132, over 5660477.96 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3746, pruned_loss=0.1264, over 5670305.51 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3875, pruned_loss=0.1327, over 5657666.79 frames. ], batch size: 186, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:15:09,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5159, 1.7507, 1.5709, 1.4701], device='cuda:1'), covar=tensor([0.2149, 0.1650, 0.1520, 0.1714], device='cuda:1'), in_proj_covar=tensor([0.1726, 0.1632, 0.1607, 0.1692], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 01:15:26,403 INFO [optim.py:369] (1/2) 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 01:15:36,913 INFO [train.py:968] (1/2) Epoch 11, batch 43700, giga_loss[loss=0.2602, simple_loss=0.3425, pruned_loss=0.08892, over 28602.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3857, pruned_loss=0.1325, over 5663082.06 frames. ], libri_tot_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1264, over 5675277.52 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3873, pruned_loss=0.1332, over 5656000.45 frames. ], batch size: 60, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:16:24,364 INFO [train.py:968] (1/2) Epoch 11, batch 43750, giga_loss[loss=0.3579, simple_loss=0.3936, pruned_loss=0.161, over 23320.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3846, pruned_loss=0.1326, over 5666180.37 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3745, pruned_loss=0.1265, over 5675419.96 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.386, pruned_loss=0.1332, over 5660647.01 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:17:02,031 INFO [optim.py:369] (1/2) 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 01:17:14,427 INFO [train.py:968] (1/2) Epoch 11, batch 43800, giga_loss[loss=0.3203, simple_loss=0.3603, pruned_loss=0.1402, over 23485.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3823, pruned_loss=0.1316, over 5659069.27 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1265, over 5677968.85 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3833, pruned_loss=0.1321, over 5652393.41 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:17:42,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=499368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:17:48,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4104, 1.6436, 1.5224, 1.2150], device='cuda:1'), covar=tensor([0.2303, 0.1793, 0.1250, 0.1825], device='cuda:1'), in_proj_covar=tensor([0.1730, 0.1634, 0.1604, 0.1691], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 01:18:03,149 INFO [train.py:968] (1/2) Epoch 11, batch 43850, giga_loss[loss=0.3361, simple_loss=0.373, pruned_loss=0.1496, over 23463.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3803, pruned_loss=0.1304, over 5672298.90 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3748, pruned_loss=0.1265, over 5682345.15 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3812, pruned_loss=0.1308, over 5662969.79 frames. ], batch size: 705, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:18:12,159 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 01:18:15,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8893, 1.8858, 1.7468, 2.0463], device='cuda:1'), covar=tensor([0.0676, 0.0286, 0.0277, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 01:18:41,142 INFO [optim.py:369] (1/2) 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 01:18:46,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=499430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:18:54,429 INFO [train.py:968] (1/2) Epoch 11, batch 43900, giga_loss[loss=0.3776, simple_loss=0.425, pruned_loss=0.1651, over 27926.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3803, pruned_loss=0.1309, over 5674241.67 frames. ], libri_tot_loss[loss=0.3133, simple_loss=0.3743, pruned_loss=0.1261, over 5682797.83 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3815, pruned_loss=0.1317, over 5666469.26 frames. ], batch size: 412, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:19:41,886 INFO [train.py:968] (1/2) Epoch 11, batch 43950, giga_loss[loss=0.3852, simple_loss=0.4236, pruned_loss=0.1734, over 26601.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3827, pruned_loss=0.1333, over 5671657.43 frames. ], libri_tot_loss[loss=0.3137, simple_loss=0.3746, pruned_loss=0.1264, over 5686942.03 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3837, pruned_loss=0.134, over 5661176.58 frames. ], batch size: 555, lr: 2.88e-03, grad_scale: 2.0
+2023-03-06 01:20:05,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=499511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:20:07,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=499514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:20:15,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5220, 4.3375, 4.1163, 1.8234], device='cuda:1'), covar=tensor([0.0567, 0.0754, 0.0763, 0.2093], device='cuda:1'), in_proj_covar=tensor([0.1068, 0.1004, 0.0874, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 01:20:15,422 INFO [optim.py:369] (1/2) 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 01:20:26,844 INFO [train.py:968] (1/2) Epoch 11, batch 44000, giga_loss[loss=0.2793, simple_loss=0.3405, pruned_loss=0.1091, over 28430.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3813, pruned_loss=0.1324, over 5675045.52 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.375, pruned_loss=0.1265, over 5689417.62 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3819, pruned_loss=0.133, over 5664163.03 frames. ], batch size: 60, lr: 2.88e-03, grad_scale: 4.0
+2023-03-06 01:20:30,802 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=499543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:21:11,705 INFO [train.py:968] (1/2) Epoch 11, batch 44050, giga_loss[loss=0.2888, simple_loss=0.3615, pruned_loss=0.1081, over 28920.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3781, pruned_loss=0.13, over 5681527.01 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.375, pruned_loss=0.1263, over 5692698.26 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3788, pruned_loss=0.1307, over 5669672.61 frames. ], batch size: 164, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:21:27,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 01:21:37,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7892, 3.6157, 3.3934, 1.8430], device='cuda:1'), covar=tensor([0.0670, 0.0780, 0.0780, 0.2123], device='cuda:1'), in_proj_covar=tensor([0.1067, 0.1000, 0.0873, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 01:21:46,416 INFO [optim.py:369] (1/2) 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 01:21:55,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9167, 3.0168, 2.0739, 1.0146], device='cuda:1'), covar=tensor([0.5821, 0.2462, 0.2825, 0.5514], device='cuda:1'), in_proj_covar=tensor([0.1572, 0.1503, 0.1497, 0.1288], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 01:21:56,858 INFO [train.py:968] (1/2) Epoch 11, batch 44100, giga_loss[loss=0.3415, simple_loss=0.4019, pruned_loss=0.1405, over 29019.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3791, pruned_loss=0.1302, over 5684338.17 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1265, over 5697392.33 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3796, pruned_loss=0.1308, over 5670105.26 frames. ], batch size: 128, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:22:51,093 INFO [train.py:968] (1/2) Epoch 11, batch 44150, giga_loss[loss=0.4492, simple_loss=0.472, pruned_loss=0.2132, over 26692.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.382, pruned_loss=0.1318, over 5666712.44 frames. ], libri_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1265, over 5689507.92 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3825, pruned_loss=0.1324, over 5662057.49 frames. ], batch size: 555, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:23:01,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1377, 2.6249, 1.1926, 1.3344], device='cuda:1'), covar=tensor([0.0968, 0.0475, 0.0914, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0514, 0.0340, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 01:23:25,833 INFO [optim.py:369] (1/2) 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 01:23:29,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1977, 1.0074, 3.9576, 3.2352], device='cuda:1'), covar=tensor([0.1766, 0.2921, 0.0449, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0590, 0.0868, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 01:23:40,975 INFO [train.py:968] (1/2) Epoch 11, batch 44200, giga_loss[loss=0.311, simple_loss=0.3675, pruned_loss=0.1272, over 28879.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3826, pruned_loss=0.1332, over 5665057.24 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3755, pruned_loss=0.1268, over 5688938.86 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3828, pruned_loss=0.1333, over 5661954.14 frames. ], batch size: 186, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:23:45,947 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=499745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:24:12,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3443, 3.3910, 1.5565, 1.4882], device='cuda:1'), covar=tensor([0.0978, 0.0357, 0.0835, 0.1354], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0515, 0.0339, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 01:24:24,593 INFO [train.py:968] (1/2) Epoch 11, batch 44250, giga_loss[loss=0.3153, simple_loss=0.3859, pruned_loss=0.1223, over 29056.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3826, pruned_loss=0.1317, over 5652259.45 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.3754, pruned_loss=0.1268, over 5675128.10 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.383, pruned_loss=0.132, over 5661036.41 frames. ], batch size: 155, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:24:25,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-06 01:24:36,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5845, 2.3374, 1.5869, 0.8168], device='cuda:1'), covar=tensor([0.4303, 0.2294, 0.3148, 0.4869], device='cuda:1'), in_proj_covar=tensor([0.1567, 0.1498, 0.1490, 0.1280], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 01:24:39,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=499805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:24:56,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2095, 1.1682, 3.7573, 3.0598], device='cuda:1'), covar=tensor([0.1666, 0.2569, 0.0502, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0677, 0.0593, 0.0874, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 01:24:56,784 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 44300, giga_loss[loss=0.3093, simple_loss=0.3829, pruned_loss=0.1178, over 28912.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3828, pruned_loss=0.1289, over 5669037.80 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3749, pruned_loss=0.1264, over 5678513.21 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3837, pruned_loss=0.1295, over 5672653.29 frames. ], batch size: 106, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:25:34,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-06 01:25:53,291 INFO [train.py:968] (1/2) Epoch 11, batch 44350, giga_loss[loss=0.343, simple_loss=0.4055, pruned_loss=0.1403, over 28737.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3852, pruned_loss=0.1296, over 5670019.63 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3752, pruned_loss=0.1269, over 5677823.02 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.386, pruned_loss=0.1299, over 5673801.77 frames. ], batch size: 242, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:26:31,855 INFO [optim.py:369] (1/2) 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 01:26:43,232 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3014, 4.0792, 3.8800, 1.7745], device='cuda:1'), covar=tensor([0.0624, 0.0865, 0.0844, 0.2167], device='cuda:1'), in_proj_covar=tensor([0.1068, 0.1003, 0.0873, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 01:26:47,489 INFO [train.py:968] (1/2) Epoch 11, batch 44400, giga_loss[loss=0.4093, simple_loss=0.4417, pruned_loss=0.1885, over 28227.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3885, pruned_loss=0.1324, over 5663252.99 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3751, pruned_loss=0.1269, over 5670695.43 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3894, pruned_loss=0.1327, over 5671774.98 frames. ], batch size: 368, lr: 2.87e-03, grad_scale: 8.0
+2023-03-06 01:26:55,373 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=499948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:26:58,615 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=499951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:27:25,471 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=499980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:27:26,160 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 44450, giga_loss[loss=0.3557, simple_loss=0.4092, pruned_loss=0.1511, over 27944.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3902, pruned_loss=0.1352, over 5671951.69 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3746, pruned_loss=0.1268, over 5680187.63 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.392, pruned_loss=0.1357, over 5670162.43 frames. ], batch size: 412, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:28:10,543 INFO [optim.py:369] (1/2) 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 01:28:19,526 INFO [train.py:968] (1/2) Epoch 11, batch 44500, giga_loss[loss=0.2977, simple_loss=0.3667, pruned_loss=0.1143, over 28704.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3911, pruned_loss=0.1371, over 5658504.86 frames. ], libri_tot_loss[loss=0.3144, simple_loss=0.3747, pruned_loss=0.127, over 5683850.38 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3926, pruned_loss=0.1375, over 5653609.83 frames. ], batch size: 262, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:28:37,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5088, 1.6551, 1.5134, 1.4190], device='cuda:1'), covar=tensor([0.1663, 0.1992, 0.1920, 0.1916], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0731, 0.0671, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 01:29:07,206 INFO [train.py:968] (1/2) Epoch 11, batch 44550, giga_loss[loss=0.3023, simple_loss=0.3735, pruned_loss=0.1156, over 28370.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3893, pruned_loss=0.1355, over 5664648.40 frames. ], libri_tot_loss[loss=0.3138, simple_loss=0.3742, pruned_loss=0.1267, over 5686651.99 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3912, pruned_loss=0.1363, over 5657911.60 frames. ], batch size: 368, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:29:38,007 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:29:51,005 INFO [train.py:968] (1/2) Epoch 11, batch 44600, giga_loss[loss=0.2888, simple_loss=0.3708, pruned_loss=0.1034, over 28925.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3867, pruned_loss=0.1318, over 5667120.87 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3738, pruned_loss=0.1265, over 5685715.57 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3887, pruned_loss=0.1327, over 5662490.38 frames. ], batch size: 136, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:30:23,429 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 01:30:25,808 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:30:42,036 INFO [train.py:968] (1/2) Epoch 11, batch 44650, giga_loss[loss=0.3519, simple_loss=0.4163, pruned_loss=0.1437, over 27892.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3872, pruned_loss=0.1306, over 5661730.59 frames. ], libri_tot_loss[loss=0.3134, simple_loss=0.3739, pruned_loss=0.1265, over 5685459.67 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3889, pruned_loss=0.1313, over 5657887.97 frames. ], batch size: 412, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:31:19,232 INFO [optim.py:369] (1/2) 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 01:31:29,978 INFO [train.py:968] (1/2) Epoch 11, batch 44700, libri_loss[loss=0.2555, simple_loss=0.334, pruned_loss=0.08844, over 29544.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3883, pruned_loss=0.1312, over 5667518.77 frames. ], libri_tot_loss[loss=0.3132, simple_loss=0.3739, pruned_loss=0.1262, over 5682065.97 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3902, pruned_loss=0.1321, over 5666615.71 frames. ], batch size: 79, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:31:46,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3641, 3.3057, 1.4056, 1.4218], device='cuda:1'), covar=tensor([0.0945, 0.0358, 0.0898, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0519, 0.0342, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 01:31:52,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:31:52,787 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:31:55,363 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 44750, giga_loss[loss=0.3188, simple_loss=0.3888, pruned_loss=0.1244, over 28915.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3878, pruned_loss=0.1316, over 5655841.85 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3741, pruned_loss=0.1265, over 5667716.20 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3894, pruned_loss=0.1322, over 5667700.40 frames. ], batch size: 164, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:32:22,596 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:33:01,497 INFO [train.py:968] (1/2) Epoch 11, batch 44800, giga_loss[loss=0.3662, simple_loss=0.4101, pruned_loss=0.1611, over 28935.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3873, pruned_loss=0.1318, over 5648766.49 frames. ], libri_tot_loss[loss=0.3145, simple_loss=0.375, pruned_loss=0.127, over 5662496.49 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3883, pruned_loss=0.132, over 5661692.61 frames. ], batch size: 213, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:33:05,004 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7314, 2.4199, 1.5625, 0.8192], device='cuda:1'), covar=tensor([0.3875, 0.2214, 0.3249, 0.4462], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1510, 0.1507, 0.1294], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 01:33:15,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:33:47,380 INFO [train.py:968] (1/2) Epoch 11, batch 44850, giga_loss[loss=0.2971, simple_loss=0.367, pruned_loss=0.1136, over 28776.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3856, pruned_loss=0.132, over 5633662.59 frames. ], libri_tot_loss[loss=0.3148, simple_loss=0.3752, pruned_loss=0.1272, over 5659588.37 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3867, pruned_loss=0.1323, over 5646271.92 frames. ], batch size: 243, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:34:19,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4619, 1.6250, 1.4765, 1.2371], device='cuda:1'), covar=tensor([0.2064, 0.1734, 0.1368, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.1725, 0.1627, 0.1611, 0.1697], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 01:34:23,678 INFO [optim.py:369] (1/2) 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 01:34:32,428 INFO [train.py:968] (1/2) Epoch 11, batch 44900, giga_loss[loss=0.3113, simple_loss=0.3774, pruned_loss=0.1226, over 28943.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3833, pruned_loss=0.1311, over 5641657.86 frames. ], libri_tot_loss[loss=0.3146, simple_loss=0.3751, pruned_loss=0.1271, over 5657453.82 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3846, pruned_loss=0.1315, over 5652859.43 frames. ], batch size: 136, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:35:17,382 INFO [train.py:968] (1/2) Epoch 11, batch 44950, giga_loss[loss=0.2796, simple_loss=0.3567, pruned_loss=0.1012, over 29018.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3811, pruned_loss=0.1297, over 5652010.35 frames. ], libri_tot_loss[loss=0.3142, simple_loss=0.3749, pruned_loss=0.1267, over 5664663.27 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3827, pruned_loss=0.1305, over 5654015.70 frames. ], batch size: 155, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:35:25,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:35:28,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=500502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:35:37,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6252, 1.7120, 1.9119, 1.4337], device='cuda:1'), covar=tensor([0.1648, 0.2089, 0.1248, 0.1489], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0699, 0.0867, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 01:35:50,177 INFO [optim.py:369] (1/2) 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 01:35:54,023 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=500531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:36:02,772 INFO [train.py:968] (1/2) Epoch 11, batch 45000, giga_loss[loss=0.4016, simple_loss=0.4367, pruned_loss=0.1833, over 27586.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3805, pruned_loss=0.1303, over 5659107.88 frames. ], libri_tot_loss[loss=0.3141, simple_loss=0.3748, pruned_loss=0.1267, over 5670965.85 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.382, pruned_loss=0.131, over 5655113.25 frames. ], batch size: 472, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:36:02,772 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 01:36:11,324 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 01:36:18,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:36:53,263 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:36:54,304 INFO [train.py:968] (1/2) Epoch 11, batch 45050, giga_loss[loss=0.2999, simple_loss=0.3739, pruned_loss=0.113, over 28966.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3779, pruned_loss=0.1281, over 5672107.97 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3737, pruned_loss=0.1258, over 5680703.36 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3803, pruned_loss=0.1297, over 5659477.72 frames. ], batch size: 164, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:37:09,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-06 01:37:32,934 INFO [optim.py:369] (1/2) 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 01:37:40,876 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:37:41,208 INFO [train.py:968] (1/2) Epoch 11, batch 45100, giga_loss[loss=0.249, simple_loss=0.338, pruned_loss=0.07997, over 28689.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1246, over 5668014.42 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.374, pruned_loss=0.126, over 5685023.39 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3764, pruned_loss=0.1257, over 5653956.35 frames. ], batch size: 284, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:38:13,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 01:38:26,650 INFO [train.py:968] (1/2) Epoch 11, batch 45150, giga_loss[loss=0.3116, simple_loss=0.3905, pruned_loss=0.1163, over 28795.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3732, pruned_loss=0.1229, over 5677415.07 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.126, over 5686414.32 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3746, pruned_loss=0.1237, over 5664927.77 frames. ], batch size: 174, lr: 2.87e-03, grad_scale: 2.0
+2023-03-06 01:38:28,111 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:38:30,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=500693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:39:04,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=500722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:39:11,035 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 11, batch 45200, libri_loss[loss=0.2828, simple_loss=0.3552, pruned_loss=0.1052, over 29580.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3727, pruned_loss=0.1227, over 5666109.23 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3739, pruned_loss=0.1259, over 5687653.24 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3738, pruned_loss=0.1234, over 5654914.28 frames. ], batch size: 77, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:39:33,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5218, 3.8931, 1.6208, 1.6358], device='cuda:1'), covar=tensor([0.0853, 0.0278, 0.0806, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0518, 0.0341, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 01:39:53,512 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:40:01,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=500781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:40:04,903 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 11, batch 45250, giga_loss[loss=0.3145, simple_loss=0.3735, pruned_loss=0.1278, over 28581.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3702, pruned_loss=0.1229, over 5644645.37 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3735, pruned_loss=0.1257, over 5691635.16 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3714, pruned_loss=0.1236, over 5631940.51 frames. ], batch size: 336, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:40:34,102 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=500813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:40:46,093 INFO [optim.py:369] (1/2) 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 01:40:55,491 INFO [train.py:968] (1/2) Epoch 11, batch 45300, giga_loss[loss=0.3453, simple_loss=0.3944, pruned_loss=0.1481, over 27571.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3703, pruned_loss=0.1231, over 5653027.38 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3736, pruned_loss=0.1257, over 5695343.25 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 5638347.61 frames. ], batch size: 472, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:41:32,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6025, 4.4290, 4.1980, 1.9118], device='cuda:1'), covar=tensor([0.0457, 0.0630, 0.0634, 0.2060], device='cuda:1'), in_proj_covar=tensor([0.1076, 0.1011, 0.0882, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 01:41:38,013 INFO [train.py:968] (1/2) Epoch 11, batch 45350, giga_loss[loss=0.3238, simple_loss=0.3858, pruned_loss=0.1309, over 28623.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3724, pruned_loss=0.1236, over 5658109.91 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3735, pruned_loss=0.1258, over 5700311.12 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.1239, over 5641327.60 frames. ], batch size: 307, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:41:58,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0155, 0.9542, 3.5981, 2.9933], device='cuda:1'), covar=tensor([0.1732, 0.2803, 0.0430, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0670, 0.0588, 0.0870, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 01:42:16,468 INFO [optim.py:369] (1/2) 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 01:42:25,416 INFO [train.py:968] (1/2) Epoch 11, batch 45400, giga_loss[loss=0.2755, simple_loss=0.3469, pruned_loss=0.1021, over 29128.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3735, pruned_loss=0.1241, over 5655167.09 frames. ], libri_tot_loss[loss=0.3123, simple_loss=0.3735, pruned_loss=0.1256, over 5705512.01 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.374, pruned_loss=0.1244, over 5636055.96 frames. ], batch size: 128, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:42:46,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=500962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:42:54,863 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:43:04,700 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=500983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:43:09,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3810, 1.7960, 1.4822, 1.5262], device='cuda:1'), covar=tensor([0.0624, 0.0256, 0.0274, 0.0633], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 01:43:10,074 INFO [train.py:968] (1/2) Epoch 11, batch 45450, giga_loss[loss=0.35, simple_loss=0.3874, pruned_loss=0.1563, over 23380.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1244, over 5621556.43 frames. ], libri_tot_loss[loss=0.3139, simple_loss=0.3748, pruned_loss=0.1265, over 5691233.76 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3727, pruned_loss=0.1238, over 5617912.71 frames. ], batch size: 705, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:43:11,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2006, 1.5542, 1.4787, 1.4175], device='cuda:1'), covar=tensor([0.1724, 0.1396, 0.1942, 0.1509], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0735, 0.0671, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 01:43:47,318 INFO [optim.py:369] (1/2) 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 01:43:56,857 INFO [train.py:968] (1/2) Epoch 11, batch 45500, giga_loss[loss=0.3264, simple_loss=0.3878, pruned_loss=0.1325, over 28868.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3737, pruned_loss=0.1249, over 5626589.65 frames. ], libri_tot_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1267, over 5683036.73 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3727, pruned_loss=0.1241, over 5630719.33 frames. ], batch size: 199, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:44:08,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501049.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 01:44:24,134 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 01:44:42,512 INFO [train.py:968] (1/2) Epoch 11, batch 45550, giga_loss[loss=0.3242, simple_loss=0.3811, pruned_loss=0.1337, over 27917.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3771, pruned_loss=0.1277, over 5596234.29 frames. ], libri_tot_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1275, over 5638872.37 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3755, pruned_loss=0.1264, over 5638940.15 frames. ], batch size: 412, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:44:59,966 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:45:01,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:45:07,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2332, 1.6839, 1.2793, 0.6720], device='cuda:1'), covar=tensor([0.2262, 0.1684, 0.1912, 0.2804], device='cuda:1'), in_proj_covar=tensor([0.1587, 0.1511, 0.1506, 0.1297], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 01:45:18,524 INFO [optim.py:369] (1/2) 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 01:45:25,518 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:45:26,622 INFO [train.py:968] (1/2) Epoch 11, batch 45600, giga_loss[loss=0.2974, simple_loss=0.3682, pruned_loss=0.1133, over 28740.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3796, pruned_loss=0.1288, over 5549562.89 frames. ], libri_tot_loss[loss=0.3175, simple_loss=0.3774, pruned_loss=0.1288, over 5562120.73 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3771, pruned_loss=0.1265, over 5652011.71 frames. ], batch size: 262, lr: 2.87e-03, grad_scale: 8.0
+2023-03-06 01:45:38,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:46:12,828 INFO [train.py:968] (1/2) Epoch 11, batch 45650, giga_loss[loss=0.3774, simple_loss=0.4029, pruned_loss=0.1759, over 23568.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.382, pruned_loss=0.1311, over 5543871.28 frames. ], libri_tot_loss[loss=0.3178, simple_loss=0.3776, pruned_loss=0.129, over 5537871.74 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3798, pruned_loss=0.1292, over 5647290.03 frames. ], batch size: 705, lr: 2.87e-03, grad_scale: 4.0
+2023-03-06 01:46:17,991 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-06 01:46:50,988 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 01:47:22,671 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:47:40,811 INFO [train.py:968] (1/2) Epoch 12, batch 50, giga_loss[loss=0.2837, simple_loss=0.3653, pruned_loss=0.101, over 28453.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3751, pruned_loss=0.1113, over 1267979.67 frames. ], libri_tot_loss[loss=0.2423, simple_loss=0.3208, pruned_loss=0.08188, over 146378.78 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3813, pruned_loss=0.1147, over 1150694.68 frames. ], batch size: 71, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:48:29,101 INFO [train.py:968] (1/2) Epoch 12, batch 100, libri_loss[loss=0.2795, simple_loss=0.3532, pruned_loss=0.1028, over 29541.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3703, pruned_loss=0.111, over 2244336.27 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09155, over 371747.70 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3752, pruned_loss=0.114, over 2001648.86 frames. ], batch size: 79, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:48:30,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:48:32,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:48:57,013 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:49:00,322 INFO [optim.py:369] (1/2) 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 01:49:10,940 INFO [train.py:968] (1/2) Epoch 12, batch 150, giga_loss[loss=0.2377, simple_loss=0.3131, pruned_loss=0.08114, over 28871.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.356, pruned_loss=0.1039, over 3018390.06 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3361, pruned_loss=0.08983, over 612822.89 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3602, pruned_loss=0.1067, over 2691417.27 frames. ], batch size: 199, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:49:14,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:49:25,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:49:45,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5497, 1.6605, 1.4620, 1.5822], device='cuda:1'), covar=tensor([0.0743, 0.0311, 0.0305, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 01:49:55,395 INFO [train.py:968] (1/2) Epoch 12, batch 200, giga_loss[loss=0.2631, simple_loss=0.3286, pruned_loss=0.09878, over 28527.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3411, pruned_loss=0.09653, over 3606417.11 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3372, pruned_loss=0.09099, over 661332.37 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3426, pruned_loss=0.09768, over 3330103.60 frames. ], batch size: 307, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:50:14,494 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-06 01:50:26,692 INFO [optim.py:369] (1/2) 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 01:50:36,909 INFO [train.py:968] (1/2) Epoch 12, batch 250, giga_loss[loss=0.264, simple_loss=0.3082, pruned_loss=0.1099, over 23887.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3313, pruned_loss=0.09145, over 4074751.30 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.341, pruned_loss=0.09255, over 816368.13 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3309, pruned_loss=0.09173, over 3798290.62 frames. ], batch size: 705, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:51:15,065 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:16,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9155, 3.7060, 3.5146, 1.7643], device='cuda:1'), covar=tensor([0.0595, 0.0812, 0.0767, 0.2399], device='cuda:1'), in_proj_covar=tensor([0.1065, 0.0998, 0.0870, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 01:51:17,379 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 300, giga_loss[loss=0.2104, simple_loss=0.2864, pruned_loss=0.06715, over 28727.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3234, pruned_loss=0.08805, over 4426586.18 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3429, pruned_loss=0.09414, over 981185.03 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3217, pruned_loss=0.08765, over 4160939.24 frames. ], batch size: 119, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:51:21,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6825, 1.8366, 1.3371, 1.4508], device='cuda:1'), covar=tensor([0.0768, 0.0540, 0.0957, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0440, 0.0500, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 01:51:27,327 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=501501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:27,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6650, 1.8397, 1.5189, 1.9155], device='cuda:1'), covar=tensor([0.2420, 0.2549, 0.2744, 0.2537], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.0971, 0.1159, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 01:51:29,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=501504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:43,543 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=501520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:51:53,454 INFO [optim.py:369] (1/2) 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 01:51:56,298 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 350, giga_loss[loss=0.1745, simple_loss=0.262, pruned_loss=0.04352, over 29031.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3167, pruned_loss=0.08501, over 4699931.63 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3458, pruned_loss=0.0948, over 1077263.39 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.314, pruned_loss=0.08423, over 4466788.65 frames. ], batch size: 164, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:52:21,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8536, 2.2471, 1.9282, 1.6916], device='cuda:1'), covar=tensor([0.2469, 0.1672, 0.1744, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.1735, 0.1625, 0.1611, 0.1695], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 01:52:23,929 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=501566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:52:43,997 INFO [train.py:968] (1/2) Epoch 12, batch 400, giga_loss[loss=0.2072, simple_loss=0.2797, pruned_loss=0.06742, over 28568.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3138, pruned_loss=0.08379, over 4918903.47 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3463, pruned_loss=0.09533, over 1233894.96 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.3102, pruned_loss=0.08259, over 4706161.31 frames. ], batch size: 60, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:53:11,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2308, 1.1617, 3.7035, 3.2986], device='cuda:1'), covar=tensor([0.1588, 0.2712, 0.0429, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0670, 0.0590, 0.0870, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 01:53:15,754 INFO [optim.py:369] (1/2) 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 01:53:24,741 INFO [train.py:968] (1/2) Epoch 12, batch 450, libri_loss[loss=0.2558, simple_loss=0.3452, pruned_loss=0.08318, over 29467.00 frames. ], tot_loss[loss=0.2381, simple_loss=0.3112, pruned_loss=0.08255, over 5091594.99 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3467, pruned_loss=0.09557, over 1321816.78 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3076, pruned_loss=0.08129, over 4910787.43 frames. ], batch size: 85, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:54:06,796 INFO [train.py:968] (1/2) Epoch 12, batch 500, giga_loss[loss=0.2041, simple_loss=0.2912, pruned_loss=0.0585, over 28865.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3091, pruned_loss=0.08149, over 5221348.68 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3475, pruned_loss=0.09528, over 1455189.81 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3048, pruned_loss=0.08011, over 5058149.61 frames. ], batch size: 174, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:54:31,973 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=501720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:54:41,801 INFO [optim.py:369] (1/2) 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 01:54:50,690 INFO [train.py:968] (1/2) Epoch 12, batch 550, giga_loss[loss=0.216, simple_loss=0.2929, pruned_loss=0.06955, over 28855.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3075, pruned_loss=0.08069, over 5324327.60 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3497, pruned_loss=0.09622, over 1561328.39 frames. ], giga_tot_loss[loss=0.2302, simple_loss=0.3024, pruned_loss=0.07897, over 5183722.27 frames. ], batch size: 145, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:55:37,150 INFO [train.py:968] (1/2) Epoch 12, batch 600, giga_loss[loss=0.2091, simple_loss=0.2847, pruned_loss=0.0667, over 28922.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3033, pruned_loss=0.0782, over 5405376.57 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3495, pruned_loss=0.09603, over 1625921.59 frames. ], giga_tot_loss[loss=0.2259, simple_loss=0.2987, pruned_loss=0.0766, over 5287222.04 frames. ], batch size: 213, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:55:56,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6452, 1.8669, 1.4969, 2.0301], device='cuda:1'), covar=tensor([0.2556, 0.2511, 0.2766, 0.2289], device='cuda:1'), in_proj_covar=tensor([0.1315, 0.0972, 0.1161, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 01:56:10,020 INFO [optim.py:369] (1/2) 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 01:56:18,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0629, 4.8423, 4.5896, 2.2261], device='cuda:1'), covar=tensor([0.0391, 0.0609, 0.0635, 0.1977], device='cuda:1'), in_proj_covar=tensor([0.1056, 0.0992, 0.0866, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 01:56:20,480 INFO [train.py:968] (1/2) Epoch 12, batch 650, giga_loss[loss=0.1954, simple_loss=0.2735, pruned_loss=0.05861, over 28770.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3028, pruned_loss=0.0779, over 5476575.15 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.35, pruned_loss=0.09625, over 1814142.43 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.2969, pruned_loss=0.07576, over 5362299.29 frames. ], batch size: 199, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:56:41,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2926, 2.5201, 1.3400, 1.4559], device='cuda:1'), covar=tensor([0.0935, 0.0359, 0.0880, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0514, 0.0341, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 01:56:46,312 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=501867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:57:09,645 INFO [train.py:968] (1/2) Epoch 12, batch 700, giga_loss[loss=0.1914, simple_loss=0.2625, pruned_loss=0.06014, over 28379.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.299, pruned_loss=0.07581, over 5525698.42 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3502, pruned_loss=0.09621, over 1834750.72 frames. ], giga_tot_loss[loss=0.2211, simple_loss=0.2941, pruned_loss=0.07405, over 5434393.19 frames. ], batch size: 65, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:57:35,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2805, 1.3649, 1.0736, 1.2417], device='cuda:1'), covar=tensor([0.1748, 0.1281, 0.1324, 0.1388], device='cuda:1'), in_proj_covar=tensor([0.1727, 0.1619, 0.1597, 0.1684], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 01:57:43,320 INFO [optim.py:369] (1/2) 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 01:57:52,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=501941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 01:57:52,943 INFO [train.py:968] (1/2) Epoch 12, batch 750, giga_loss[loss=0.2118, simple_loss=0.2823, pruned_loss=0.07071, over 28589.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2972, pruned_loss=0.07541, over 5548883.21 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3504, pruned_loss=0.09635, over 1884483.97 frames. ], giga_tot_loss[loss=0.22, simple_loss=0.2926, pruned_loss=0.0737, over 5481778.71 frames. ], batch size: 307, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 01:58:37,793 INFO [train.py:968] (1/2) Epoch 12, batch 800, giga_loss[loss=0.2294, simple_loss=0.2947, pruned_loss=0.08203, over 29102.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2953, pruned_loss=0.07467, over 5573619.78 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3512, pruned_loss=0.0967, over 1973031.94 frames. ], giga_tot_loss[loss=0.2178, simple_loss=0.2901, pruned_loss=0.0727, over 5523031.88 frames. ], batch size: 128, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:59:09,961 INFO [optim.py:369] (1/2) 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 01:59:24,338 INFO [train.py:968] (1/2) Epoch 12, batch 850, giga_loss[loss=0.2903, simple_loss=0.3576, pruned_loss=0.1115, over 29025.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3029, pruned_loss=0.07868, over 5592274.67 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.351, pruned_loss=0.09653, over 2116571.28 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.297, pruned_loss=0.07646, over 5549835.56 frames. ], batch size: 106, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 01:59:30,870 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=502051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:04,278 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:06,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:12,217 INFO [train.py:968] (1/2) Epoch 12, batch 900, giga_loss[loss=0.3006, simple_loss=0.3669, pruned_loss=0.1171, over 28984.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3173, pruned_loss=0.08644, over 5616449.16 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3497, pruned_loss=0.09579, over 2191565.92 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3124, pruned_loss=0.08472, over 5577694.92 frames. ], batch size: 112, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:00:14,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:00:35,182 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:00:53,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0058, 1.1231, 3.3573, 2.8213], device='cuda:1'), covar=tensor([0.1639, 0.2659, 0.0436, 0.1537], device='cuda:1'), in_proj_covar=tensor([0.0665, 0.0587, 0.0865, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:00:54,232 INFO [train.py:968] (1/2) Epoch 12, batch 950, giga_loss[loss=0.3588, simple_loss=0.417, pruned_loss=0.1503, over 27551.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3296, pruned_loss=0.09291, over 5635780.49 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3491, pruned_loss=0.09513, over 2299604.82 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3252, pruned_loss=0.09165, over 5598869.59 frames. ], batch size: 472, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:01:24,304 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=502176.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 02:01:36,050 INFO [train.py:968] (1/2) Epoch 12, batch 1000, giga_loss[loss=0.2856, simple_loss=0.3638, pruned_loss=0.1037, over 28590.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3375, pruned_loss=0.09628, over 5651177.97 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.349, pruned_loss=0.09501, over 2388481.99 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3338, pruned_loss=0.09535, over 5616194.25 frames. ], batch size: 336, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:01:37,565 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 02:02:08,253 INFO [optim.py:369] (1/2) 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 02:02:13,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:02:15,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:02:16,226 INFO [train.py:968] (1/2) Epoch 12, batch 1050, giga_loss[loss=0.2531, simple_loss=0.3384, pruned_loss=0.08394, over 28845.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3411, pruned_loss=0.09646, over 5661388.52 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3498, pruned_loss=0.09541, over 2488445.68 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3375, pruned_loss=0.09561, over 5630395.54 frames. ], batch size: 186, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:02:16,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:02:43,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:03:01,319 INFO [train.py:968] (1/2) Epoch 12, batch 1100, giga_loss[loss=0.2643, simple_loss=0.3427, pruned_loss=0.09298, over 28884.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3416, pruned_loss=0.09551, over 5661835.40 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3491, pruned_loss=0.09497, over 2589082.69 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3388, pruned_loss=0.09502, over 5630839.18 frames. ], batch size: 145, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:03:19,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0373, 2.0391, 1.7739, 1.8799], device='cuda:1'), covar=tensor([0.1545, 0.2307, 0.2028, 0.1932], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0729, 0.0664, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 02:03:29,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4427, 1.6545, 1.3348, 1.5916], device='cuda:1'), covar=tensor([0.2384, 0.2291, 0.2446, 0.2085], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.0971, 0.1156, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 02:03:34,498 INFO [optim.py:369] (1/2) 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 02:03:44,726 INFO [train.py:968] (1/2) Epoch 12, batch 1150, giga_loss[loss=0.2382, simple_loss=0.3152, pruned_loss=0.0806, over 28409.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3445, pruned_loss=0.09779, over 5662495.80 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3482, pruned_loss=0.0946, over 2622023.84 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3427, pruned_loss=0.09759, over 5636119.87 frames. ], batch size: 65, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:03:59,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 02:04:23,035 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:04:24,858 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 1200, giga_loss[loss=0.2938, simple_loss=0.3678, pruned_loss=0.11, over 28306.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3481, pruned_loss=0.1002, over 5664803.42 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3506, pruned_loss=0.09609, over 2721852.88 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3455, pruned_loss=0.09946, over 5648483.71 frames. ], batch size: 65, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:04:49,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:04:59,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:05:03,517 INFO [optim.py:369] (1/2) 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 02:05:11,952 INFO [train.py:968] (1/2) Epoch 12, batch 1250, giga_loss[loss=0.2846, simple_loss=0.3684, pruned_loss=0.1004, over 29002.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3522, pruned_loss=0.1029, over 5671542.27 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3509, pruned_loss=0.0961, over 2783165.60 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.35, pruned_loss=0.1025, over 5656152.16 frames. ], batch size: 155, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:05:18,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9227, 2.2887, 1.8153, 1.6540], device='cuda:1'), covar=tensor([0.1940, 0.1539, 0.1760, 0.1815], device='cuda:1'), in_proj_covar=tensor([0.1722, 0.1617, 0.1596, 0.1688], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 02:05:57,801 INFO [train.py:968] (1/2) Epoch 12, batch 1300, giga_loss[loss=0.2808, simple_loss=0.356, pruned_loss=0.1028, over 28882.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3557, pruned_loss=0.1042, over 5681153.47 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3514, pruned_loss=0.0964, over 2813979.36 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3539, pruned_loss=0.1038, over 5667255.55 frames. ], batch size: 227, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:06:21,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3786, 1.7828, 1.5095, 1.5052], device='cuda:1'), covar=tensor([0.0733, 0.0355, 0.0300, 0.0747], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 02:06:30,377 INFO [optim.py:369] (1/2) 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 02:06:35,909 INFO [train.py:968] (1/2) Epoch 12, batch 1350, libri_loss[loss=0.292, simple_loss=0.3487, pruned_loss=0.1177, over 29317.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3579, pruned_loss=0.1051, over 5683307.76 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3516, pruned_loss=0.09664, over 2901879.95 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3565, pruned_loss=0.1049, over 5669619.17 frames. ], batch size: 71, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:06:43,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=502551.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 02:07:00,521 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:07:03,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:07:21,826 INFO [train.py:968] (1/2) Epoch 12, batch 1400, giga_loss[loss=0.2802, simple_loss=0.3626, pruned_loss=0.09884, over 28746.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3592, pruned_loss=0.105, over 5684298.76 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3519, pruned_loss=0.0968, over 2913395.55 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.358, pruned_loss=0.1048, over 5675480.25 frames. ], batch size: 242, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:07:27,707 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:07:43,331 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-06 02:07:52,803 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8360, 1.8557, 1.6064, 2.0530], device='cuda:1'), covar=tensor([0.2277, 0.2473, 0.2549, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.1307, 0.0970, 0.1153, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 02:07:53,894 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 1450, libri_loss[loss=0.2532, simple_loss=0.3286, pruned_loss=0.08885, over 29506.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3592, pruned_loss=0.1039, over 5695685.65 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3523, pruned_loss=0.09714, over 3042002.87 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3583, pruned_loss=0.1039, over 5682862.96 frames. ], batch size: 70, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:08:03,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-06 02:08:07,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-06 02:08:40,478 INFO [train.py:968] (1/2) Epoch 12, batch 1500, libri_loss[loss=0.2672, simple_loss=0.3458, pruned_loss=0.09431, over 29579.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3575, pruned_loss=0.102, over 5700982.79 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3516, pruned_loss=0.09675, over 3099115.88 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3573, pruned_loss=0.1023, over 5686604.54 frames. ], batch size: 74, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:08:42,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=502694.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 02:08:44,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=502697.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 02:08:57,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5406, 1.6242, 1.4107, 1.2714], device='cuda:1'), covar=tensor([0.2103, 0.1929, 0.1626, 0.1884], device='cuda:1'), in_proj_covar=tensor([0.1712, 0.1611, 0.1593, 0.1678], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 02:09:03,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3218, 1.7529, 1.3538, 1.4264], device='cuda:1'), covar=tensor([0.0759, 0.0315, 0.0314, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 02:09:08,065 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=502726.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 02:09:12,538 INFO [optim.py:369] (1/2) 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 02:09:20,105 INFO [train.py:968] (1/2) Epoch 12, batch 1550, giga_loss[loss=0.2599, simple_loss=0.3447, pruned_loss=0.08749, over 28438.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.356, pruned_loss=0.1003, over 5713334.51 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3511, pruned_loss=0.09671, over 3193122.37 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3562, pruned_loss=0.1007, over 5697789.52 frames. ], batch size: 60, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:09:27,474 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 1600, giga_loss[loss=0.3059, simple_loss=0.3647, pruned_loss=0.1235, over 28931.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3562, pruned_loss=0.1017, over 5698200.04 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3504, pruned_loss=0.09622, over 3246285.63 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3568, pruned_loss=0.1023, over 5683123.69 frames. ], batch size: 136, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:10:20,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6819, 1.9382, 2.0065, 1.5259], device='cuda:1'), covar=tensor([0.1760, 0.2180, 0.1357, 0.1601], device='cuda:1'), in_proj_covar=tensor([0.0842, 0.0697, 0.0882, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 02:10:31,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-06 02:10:39,087 INFO [optim.py:369] (1/2) 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 02:10:43,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=502837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:10:47,585 INFO [train.py:968] (1/2) Epoch 12, batch 1650, giga_loss[loss=0.3073, simple_loss=0.3717, pruned_loss=0.1215, over 28458.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.359, pruned_loss=0.1056, over 5707200.67 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3495, pruned_loss=0.09576, over 3325301.25 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3601, pruned_loss=0.1066, over 5689572.44 frames. ], batch size: 65, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:10:57,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3692, 1.6091, 1.3198, 1.4251], device='cuda:1'), covar=tensor([0.2140, 0.2139, 0.2243, 0.1807], device='cuda:1'), in_proj_covar=tensor([0.1306, 0.0970, 0.1151, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 02:11:31,043 INFO [train.py:968] (1/2) Epoch 12, batch 1700, giga_loss[loss=0.339, simple_loss=0.3898, pruned_loss=0.1441, over 28914.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3607, pruned_loss=0.1085, over 5715985.87 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3498, pruned_loss=0.096, over 3376878.40 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3616, pruned_loss=0.1093, over 5698190.81 frames. ], batch size: 213, lr: 2.75e-03, grad_scale: 8.0
+2023-03-06 02:12:05,752 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 1750, giga_loss[loss=0.289, simple_loss=0.3634, pruned_loss=0.1073, over 28608.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3595, pruned_loss=0.1085, over 5705988.25 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3504, pruned_loss=0.09639, over 3441340.22 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3602, pruned_loss=0.1093, over 5695294.23 frames. ], batch size: 307, lr: 2.75e-03, grad_scale: 4.0
+2023-03-06 02:12:58,003 INFO [train.py:968] (1/2) Epoch 12, batch 1800, giga_loss[loss=0.2719, simple_loss=0.3447, pruned_loss=0.09958, over 28712.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.357, pruned_loss=0.1079, over 5683148.12 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3505, pruned_loss=0.09655, over 3443799.52 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3575, pruned_loss=0.1085, over 5683038.04 frames. ], batch size: 242, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:13:34,141 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 1850, giga_loss[loss=0.2759, simple_loss=0.3395, pruned_loss=0.1062, over 28665.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3563, pruned_loss=0.1071, over 5683660.83 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3506, pruned_loss=0.09641, over 3492578.32 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3568, pruned_loss=0.1078, over 5679964.87 frames. ], batch size: 92, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:14:02,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-06 02:14:25,271 INFO [train.py:968] (1/2) Epoch 12, batch 1900, giga_loss[loss=0.2388, simple_loss=0.3199, pruned_loss=0.07882, over 28545.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3552, pruned_loss=0.1055, over 5692599.14 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3513, pruned_loss=0.09672, over 3564491.88 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3554, pruned_loss=0.1062, over 5683298.70 frames. ], batch size: 78, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:14:34,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9584, 1.1486, 1.0879, 0.8825], device='cuda:1'), covar=tensor([0.1694, 0.2007, 0.1102, 0.1530], device='cuda:1'), in_proj_covar=tensor([0.1718, 0.1622, 0.1598, 0.1681], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 02:14:58,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:15:02,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-06 02:15:03,982 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 1950, giga_loss[loss=0.2795, simple_loss=0.3461, pruned_loss=0.1065, over 28613.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.351, pruned_loss=0.1029, over 5692234.21 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.351, pruned_loss=0.09667, over 3633653.80 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3513, pruned_loss=0.1036, over 5678935.21 frames. ], batch size: 92, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:15:42,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5703, 4.2549, 1.8224, 1.6801], device='cuda:1'), covar=tensor([0.0903, 0.0197, 0.0820, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0508, 0.0341, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 02:15:58,032 INFO [train.py:968] (1/2) Epoch 12, batch 2000, giga_loss[loss=0.2415, simple_loss=0.3213, pruned_loss=0.08087, over 28975.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3455, pruned_loss=0.09995, over 5682249.64 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3508, pruned_loss=0.09651, over 3684234.27 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3458, pruned_loss=0.1008, over 5671826.26 frames. ], batch size: 128, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:16:03,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:16:17,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:16:37,432 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 2050, giga_loss[loss=0.2408, simple_loss=0.3141, pruned_loss=0.08371, over 28601.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3392, pruned_loss=0.09641, over 5680162.21 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3514, pruned_loss=0.09698, over 3705740.79 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3389, pruned_loss=0.09678, over 5670443.05 frames. ], batch size: 336, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:17:12,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:17:15,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7177, 1.7698, 1.3248, 1.3968], device='cuda:1'), covar=tensor([0.0796, 0.0604, 0.1010, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0434, 0.0499, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:17:15,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:17:34,344 INFO [train.py:968] (1/2) Epoch 12, batch 2100, libri_loss[loss=0.2744, simple_loss=0.3538, pruned_loss=0.09752, over 29612.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3372, pruned_loss=0.09561, over 5664969.99 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3517, pruned_loss=0.09708, over 3738523.44 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3366, pruned_loss=0.09581, over 5653955.17 frames. ], batch size: 74, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:17:40,605 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:17:45,918 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 12, batch 2150, giga_loss[loss=0.2864, simple_loss=0.3523, pruned_loss=0.1103, over 29019.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3387, pruned_loss=0.09595, over 5674359.77 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3511, pruned_loss=0.09656, over 3798321.90 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3383, pruned_loss=0.09637, over 5663468.15 frames. ], batch size: 128, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:18:24,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:18:26,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:18:31,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-06 02:18:49,554 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:18:52,669 INFO [train.py:968] (1/2) Epoch 12, batch 2200, giga_loss[loss=0.2517, simple_loss=0.3319, pruned_loss=0.08573, over 28741.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3378, pruned_loss=0.09519, over 5691814.61 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3511, pruned_loss=0.09628, over 3840608.52 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3372, pruned_loss=0.09566, over 5678827.86 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:19:06,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:19:27,615 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 2250, libri_loss[loss=0.3295, simple_loss=0.4031, pruned_loss=0.1279, over 29670.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3361, pruned_loss=0.09423, over 5699081.00 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3517, pruned_loss=0.09631, over 3911881.06 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3347, pruned_loss=0.0945, over 5681661.52 frames. ], batch size: 91, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:20:16,800 INFO [train.py:968] (1/2) Epoch 12, batch 2300, giga_loss[loss=0.2371, simple_loss=0.3053, pruned_loss=0.0844, over 28539.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3328, pruned_loss=0.09259, over 5709051.93 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3514, pruned_loss=0.09613, over 3941780.32 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3316, pruned_loss=0.09287, over 5692376.42 frames. ], batch size: 71, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:20:26,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-06 02:20:51,614 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 2350, giga_loss[loss=0.2546, simple_loss=0.3205, pruned_loss=0.09435, over 28536.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3297, pruned_loss=0.09093, over 5709875.89 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3514, pruned_loss=0.09601, over 3980975.24 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3283, pruned_loss=0.09117, over 5692762.44 frames. ], batch size: 71, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:21:22,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:21:36,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3118, 1.1969, 1.1105, 1.3597], device='cuda:1'), covar=tensor([0.0699, 0.0321, 0.0312, 0.0760], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 02:21:40,422 INFO [train.py:968] (1/2) Epoch 12, batch 2400, giga_loss[loss=0.2026, simple_loss=0.2813, pruned_loss=0.06194, over 28718.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.328, pruned_loss=0.09037, over 5707549.95 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3518, pruned_loss=0.0961, over 4017607.48 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3263, pruned_loss=0.0904, over 5691837.36 frames. ], batch size: 71, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:22:12,989 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 2450, giga_loss[loss=0.2485, simple_loss=0.314, pruned_loss=0.09146, over 28969.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3256, pruned_loss=0.08947, over 5716116.72 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3517, pruned_loss=0.09603, over 4045868.13 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3239, pruned_loss=0.08943, over 5701033.86 frames. ], batch size: 106, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:22:49,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:22:56,073 INFO [train.py:968] (1/2) Epoch 12, batch 2500, giga_loss[loss=0.2406, simple_loss=0.3139, pruned_loss=0.08362, over 28852.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3235, pruned_loss=0.0885, over 5718908.53 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3518, pruned_loss=0.09576, over 4079289.42 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3216, pruned_loss=0.08852, over 5706482.20 frames. ], batch size: 243, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:23:16,379 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:23:18,729 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:23:32,086 INFO [optim.py:369] (1/2) 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 02:23:36,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6946, 0.9411, 2.8377, 2.7567], device='cuda:1'), covar=tensor([0.1747, 0.2615, 0.0568, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0664, 0.0583, 0.0854, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:23:39,409 INFO [train.py:968] (1/2) Epoch 12, batch 2550, giga_loss[loss=0.2325, simple_loss=0.3053, pruned_loss=0.07983, over 29107.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3208, pruned_loss=0.08698, over 5725422.97 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3518, pruned_loss=0.09568, over 4088441.97 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3192, pruned_loss=0.08701, over 5714946.28 frames. ], batch size: 155, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:23:43,578 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:10,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=503782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:18,154 INFO [train.py:968] (1/2) Epoch 12, batch 2600, giga_loss[loss=0.2311, simple_loss=0.2964, pruned_loss=0.08293, over 28365.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.32, pruned_loss=0.08658, over 5724138.16 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3522, pruned_loss=0.09562, over 4130901.48 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3178, pruned_loss=0.08643, over 5713952.20 frames. ], batch size: 65, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:24:45,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:47,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:24:51,730 INFO [optim.py:369] (1/2) 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 02:24:57,426 INFO [train.py:968] (1/2) Epoch 12, batch 2650, giga_loss[loss=0.2612, simple_loss=0.3308, pruned_loss=0.09583, over 29088.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3209, pruned_loss=0.08706, over 5728163.38 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3526, pruned_loss=0.09561, over 4174596.46 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.318, pruned_loss=0.08674, over 5716101.77 frames. ], batch size: 128, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:25:10,560 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=503858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:25:28,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=503879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:25:37,354 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 2700, giga_loss[loss=0.2818, simple_loss=0.3511, pruned_loss=0.1063, over 28671.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3259, pruned_loss=0.08992, over 5725822.13 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.353, pruned_loss=0.09571, over 4223621.80 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3225, pruned_loss=0.0894, over 5713483.88 frames. ], batch size: 284, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:26:11,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=503925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:26:14,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=503928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:26:19,018 INFO [optim.py:369] (1/2) 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 02:26:27,286 INFO [train.py:968] (1/2) Epoch 12, batch 2750, giga_loss[loss=0.2828, simple_loss=0.3628, pruned_loss=0.1014, over 28904.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3321, pruned_loss=0.09385, over 5719435.90 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3533, pruned_loss=0.09561, over 4254817.82 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3288, pruned_loss=0.09343, over 5708527.46 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:26:41,409 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 2800, libri_loss[loss=0.3204, simple_loss=0.387, pruned_loss=0.1269, over 19085.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3408, pruned_loss=0.09972, over 5702403.23 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3536, pruned_loss=0.09583, over 4269051.65 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3378, pruned_loss=0.09924, over 5701673.42 frames. ], batch size: 187, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:27:25,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6055, 1.5793, 1.9188, 1.4619], device='cuda:1'), covar=tensor([0.1189, 0.1694, 0.0939, 0.1258], device='cuda:1'), in_proj_covar=tensor([0.0841, 0.0699, 0.0881, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 02:27:53,189 INFO [optim.py:369] (1/2) 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 02:27:57,980 INFO [train.py:968] (1/2) Epoch 12, batch 2850, giga_loss[loss=0.3393, simple_loss=0.393, pruned_loss=0.1428, over 27612.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3465, pruned_loss=0.1026, over 5690559.35 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3537, pruned_loss=0.09585, over 4332263.53 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3435, pruned_loss=0.1025, over 5696244.66 frames. ], batch size: 472, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:28:49,776 INFO [train.py:968] (1/2) Epoch 12, batch 2900, giga_loss[loss=0.3312, simple_loss=0.3919, pruned_loss=0.1353, over 27639.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3509, pruned_loss=0.1038, over 5696453.55 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3534, pruned_loss=0.09567, over 4348090.95 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3488, pruned_loss=0.1039, over 5698992.59 frames. ], batch size: 472, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:29:06,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1382, 1.4600, 1.4671, 1.0792], device='cuda:1'), covar=tensor([0.1410, 0.2118, 0.1179, 0.1469], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0698, 0.0878, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 02:29:08,790 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 12, batch 2950, giga_loss[loss=0.2802, simple_loss=0.3572, pruned_loss=0.1017, over 28884.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3559, pruned_loss=0.1063, over 5701916.86 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3531, pruned_loss=0.09563, over 4400225.65 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3543, pruned_loss=0.1067, over 5698890.14 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:29:57,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 02:30:09,428 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 3000, giga_loss[loss=0.3047, simple_loss=0.379, pruned_loss=0.1152, over 28865.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3627, pruned_loss=0.1114, over 5682219.50 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3534, pruned_loss=0.09587, over 4422602.25 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3613, pruned_loss=0.1118, over 5676899.88 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:30:17,775 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 02:30:26,066 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 02:31:04,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6595, 1.9471, 1.5603, 1.7084], device='cuda:1'), covar=tensor([0.0774, 0.0270, 0.0300, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0112, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 02:31:05,020 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 3050, giga_loss[loss=0.2745, simple_loss=0.343, pruned_loss=0.103, over 27985.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3575, pruned_loss=0.1075, over 5691203.34 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3529, pruned_loss=0.09565, over 4444836.22 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3569, pruned_loss=0.1081, over 5684055.25 frames. ], batch size: 412, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:31:18,838 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:31:26,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 3100, giga_loss[loss=0.2804, simple_loss=0.3532, pruned_loss=0.1038, over 27683.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.355, pruned_loss=0.1051, over 5698959.05 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3528, pruned_loss=0.09566, over 4459698.05 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3546, pruned_loss=0.1057, over 5691215.51 frames. ], batch size: 472, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:32:29,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6472, 4.5070, 1.7191, 1.8793], device='cuda:1'), covar=tensor([0.0901, 0.0233, 0.0827, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0508, 0.0342, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 02:32:32,013 INFO [optim.py:369] (1/2) 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 02:32:36,608 INFO [train.py:968] (1/2) Epoch 12, batch 3150, giga_loss[loss=0.2522, simple_loss=0.3247, pruned_loss=0.08982, over 23411.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3534, pruned_loss=0.1033, over 5700999.93 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3528, pruned_loss=0.09569, over 4513860.69 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3531, pruned_loss=0.104, over 5695633.75 frames. ], batch size: 705, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:32:52,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4580, 1.8067, 1.3657, 1.5387], device='cuda:1'), covar=tensor([0.2514, 0.2410, 0.2756, 0.2260], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.0970, 0.1150, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 02:33:15,412 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:18,558 INFO [train.py:968] (1/2) Epoch 12, batch 3200, giga_loss[loss=0.2935, simple_loss=0.3588, pruned_loss=0.1141, over 28831.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3543, pruned_loss=0.1031, over 5710535.50 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3524, pruned_loss=0.09558, over 4554256.24 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3543, pruned_loss=0.104, over 5701728.17 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:33:23,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:27,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:31,393 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:34,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:52,152 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:33:57,776 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:34:01,582 INFO [train.py:968] (1/2) Epoch 12, batch 3250, giga_loss[loss=0.2937, simple_loss=0.3648, pruned_loss=0.1113, over 29077.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3566, pruned_loss=0.1045, over 5715113.91 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3526, pruned_loss=0.09559, over 4581124.46 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3566, pruned_loss=0.1053, over 5704473.49 frames. ], batch size: 155, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:34:46,243 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 3300, giga_loss[loss=0.2746, simple_loss=0.3488, pruned_loss=0.1002, over 29058.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3577, pruned_loss=0.1055, over 5708161.77 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3529, pruned_loss=0.09575, over 4600890.27 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3575, pruned_loss=0.1062, over 5697332.24 frames. ], batch size: 128, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:35:25,999 INFO [optim.py:369] (1/2) 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 02:35:29,657 INFO [train.py:968] (1/2) Epoch 12, batch 3350, giga_loss[loss=0.285, simple_loss=0.3579, pruned_loss=0.1061, over 28639.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3584, pruned_loss=0.1065, over 5708930.51 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3527, pruned_loss=0.09556, over 4607451.81 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3585, pruned_loss=0.1072, over 5699631.43 frames. ], batch size: 92, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:35:43,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0281, 1.3826, 1.1236, 0.2183], device='cuda:1'), covar=tensor([0.2010, 0.1871, 0.2895, 0.3403], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1472, 0.1497, 0.1279], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 02:35:45,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:36:16,245 INFO [train.py:968] (1/2) Epoch 12, batch 3400, giga_loss[loss=0.278, simple_loss=0.3533, pruned_loss=0.1013, over 28680.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3589, pruned_loss=0.1071, over 5710578.90 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3531, pruned_loss=0.09588, over 4616847.84 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3587, pruned_loss=0.1076, over 5709584.81 frames. ], batch size: 71, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:36:50,270 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:36:53,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:36:55,232 INFO [optim.py:369] (1/2) 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 02:36:56,735 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:36:57,859 INFO [train.py:968] (1/2) Epoch 12, batch 3450, giga_loss[loss=0.2682, simple_loss=0.3436, pruned_loss=0.0964, over 28949.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3591, pruned_loss=0.1071, over 5719270.21 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3535, pruned_loss=0.09605, over 4649024.44 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3588, pruned_loss=0.1076, over 5713892.57 frames. ], batch size: 106, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:37:18,207 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:37:38,625 INFO [train.py:968] (1/2) Epoch 12, batch 3500, giga_loss[loss=0.2499, simple_loss=0.3378, pruned_loss=0.08105, over 28606.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3589, pruned_loss=0.1061, over 5716490.82 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3534, pruned_loss=0.09593, over 4666972.29 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3589, pruned_loss=0.1068, over 5710488.45 frames. ], batch size: 60, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:37:46,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:37:48,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:38:13,454 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:38:22,382 INFO [train.py:968] (1/2) Epoch 12, batch 3550, giga_loss[loss=0.2685, simple_loss=0.3491, pruned_loss=0.09393, over 29091.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3591, pruned_loss=0.105, over 5719731.01 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3537, pruned_loss=0.09611, over 4696583.72 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3589, pruned_loss=0.1056, over 5711570.68 frames. ], batch size: 155, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:38:42,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=504762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:39:04,387 INFO [train.py:968] (1/2) Epoch 12, batch 3600, giga_loss[loss=0.2406, simple_loss=0.3263, pruned_loss=0.07746, over 28889.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.357, pruned_loss=0.1032, over 5713155.96 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3532, pruned_loss=0.0959, over 4709208.35 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3574, pruned_loss=0.1039, over 5713323.26 frames. ], batch size: 174, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:39:39,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0499, 2.0015, 1.8745, 1.7727], device='cuda:1'), covar=tensor([0.1526, 0.2213, 0.1940, 0.1941], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0730, 0.0676, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 02:39:43,653 INFO [optim.py:369] (1/2) 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 02:39:45,737 INFO [train.py:968] (1/2) Epoch 12, batch 3650, giga_loss[loss=0.2529, simple_loss=0.3307, pruned_loss=0.08758, over 28278.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3556, pruned_loss=0.1028, over 5721577.91 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3531, pruned_loss=0.09594, over 4738675.41 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.356, pruned_loss=0.1035, over 5717462.96 frames. ], batch size: 77, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:39:47,131 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-06 02:40:01,245 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=504862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:23,579 INFO [train.py:968] (1/2) Epoch 12, batch 3700, giga_loss[loss=0.2851, simple_loss=0.362, pruned_loss=0.1041, over 28248.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3531, pruned_loss=0.1015, over 5722010.14 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.353, pruned_loss=0.09597, over 4782373.16 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3535, pruned_loss=0.1023, over 5714155.65 frames. ], batch size: 368, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:40:33,519 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=504905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:36,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=504908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:57,851 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=504937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:40:59,116 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 3750, giga_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.1239, over 28892.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3522, pruned_loss=0.1011, over 5727888.07 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.353, pruned_loss=0.09589, over 4803109.13 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3525, pruned_loss=0.1019, over 5719771.09 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:41:29,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-06 02:41:35,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3329, 3.2454, 1.4812, 1.4558], device='cuda:1'), covar=tensor([0.0910, 0.0260, 0.0836, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0503, 0.0339, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 02:41:47,552 INFO [train.py:968] (1/2) Epoch 12, batch 3800, giga_loss[loss=0.3274, simple_loss=0.3881, pruned_loss=0.1333, over 27573.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3535, pruned_loss=0.1025, over 5727901.20 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3531, pruned_loss=0.09611, over 4824652.28 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.1031, over 5718603.27 frames. ], batch size: 472, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:41:54,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-06 02:42:04,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:42:07,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4713, 1.6770, 1.7345, 1.3119], device='cuda:1'), covar=tensor([0.1733, 0.2414, 0.1369, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.0834, 0.0696, 0.0878, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 02:42:21,795 INFO [optim.py:369] (1/2) 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 02:42:25,250 INFO [train.py:968] (1/2) Epoch 12, batch 3850, giga_loss[loss=0.2949, simple_loss=0.3597, pruned_loss=0.1151, over 28902.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3541, pruned_loss=0.1025, over 5734267.31 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3532, pruned_loss=0.09619, over 4845493.45 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3542, pruned_loss=0.103, over 5724336.99 frames. ], batch size: 112, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:42:38,563 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:43:04,934 INFO [train.py:968] (1/2) Epoch 12, batch 3900, giga_loss[loss=0.243, simple_loss=0.3305, pruned_loss=0.07768, over 29061.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3526, pruned_loss=0.101, over 5729069.57 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3528, pruned_loss=0.09599, over 4870928.39 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.353, pruned_loss=0.1017, over 5717849.70 frames. ], batch size: 128, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:43:20,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4158, 1.6039, 1.6044, 1.5008], device='cuda:1'), covar=tensor([0.1658, 0.1728, 0.2079, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0723, 0.0671, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 02:43:41,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3697, 1.5052, 1.2503, 1.5262], device='cuda:1'), covar=tensor([0.0766, 0.0315, 0.0329, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 02:43:45,636 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 3950, giga_loss[loss=0.2985, simple_loss=0.3552, pruned_loss=0.1209, over 26515.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.352, pruned_loss=0.1008, over 5719610.74 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3525, pruned_loss=0.09601, over 4877296.67 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3525, pruned_loss=0.1014, over 5715631.05 frames. ], batch size: 555, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:43:52,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5092, 4.3033, 4.0813, 1.8949], device='cuda:1'), covar=tensor([0.0493, 0.0679, 0.0711, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.1037, 0.0967, 0.0844, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 02:44:00,499 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:44:10,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2729, 1.3890, 1.4632, 1.3270], device='cuda:1'), covar=tensor([0.1113, 0.0937, 0.1492, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0726, 0.0671, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 02:44:27,226 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=505190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:44:28,363 INFO [train.py:968] (1/2) Epoch 12, batch 4000, giga_loss[loss=0.2575, simple_loss=0.3339, pruned_loss=0.09059, over 28419.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3504, pruned_loss=0.1001, over 5719293.17 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3532, pruned_loss=0.09634, over 4902071.53 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1004, over 5712368.21 frames. ], batch size: 65, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:45:03,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9380, 1.3175, 1.1189, 0.1781], device='cuda:1'), covar=tensor([0.2832, 0.2163, 0.3544, 0.4682], device='cuda:1'), in_proj_covar=tensor([0.1537, 0.1452, 0.1472, 0.1266], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 02:45:04,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:45:06,728 INFO [optim.py:369] (1/2) 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 02:45:09,268 INFO [train.py:968] (1/2) Epoch 12, batch 4050, giga_loss[loss=0.2705, simple_loss=0.3457, pruned_loss=0.09766, over 28644.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3473, pruned_loss=0.09872, over 5714153.18 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3531, pruned_loss=0.09649, over 4917050.33 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.347, pruned_loss=0.09887, over 5706247.08 frames. ], batch size: 307, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:45:47,768 INFO [train.py:968] (1/2) Epoch 12, batch 4100, giga_loss[loss=0.2484, simple_loss=0.3222, pruned_loss=0.08729, over 28671.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3448, pruned_loss=0.09747, over 5718583.24 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3533, pruned_loss=0.09656, over 4946120.01 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3442, pruned_loss=0.09756, over 5707307.07 frames. ], batch size: 99, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:46:18,035 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0941, 3.2627, 2.3517, 1.2227], device='cuda:1'), covar=tensor([0.5601, 0.1882, 0.2841, 0.4877], device='cuda:1'), in_proj_covar=tensor([0.1533, 0.1449, 0.1470, 0.1262], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 02:46:27,877 INFO [optim.py:369] (1/2) 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 02:46:29,266 INFO [train.py:968] (1/2) Epoch 12, batch 4150, giga_loss[loss=0.291, simple_loss=0.3626, pruned_loss=0.1097, over 28698.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.09822, over 5715984.24 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3531, pruned_loss=0.09629, over 4966213.29 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3448, pruned_loss=0.09853, over 5705947.83 frames. ], batch size: 242, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:46:31,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3615, 1.7071, 1.3443, 1.3901], device='cuda:1'), covar=tensor([0.2486, 0.2328, 0.2705, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.0964, 0.1148, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 02:46:59,630 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=505380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:02,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:09,050 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-06 02:47:10,575 INFO [train.py:968] (1/2) Epoch 12, batch 4200, giga_loss[loss=0.25, simple_loss=0.324, pruned_loss=0.08804, over 28654.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3441, pruned_loss=0.09834, over 5713195.82 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3532, pruned_loss=0.0963, over 4985100.54 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3435, pruned_loss=0.09863, over 5701609.06 frames. ], batch size: 92, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:47:26,085 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=505412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:45,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:47:50,439 INFO [optim.py:369] (1/2) 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 02:47:51,662 INFO [train.py:968] (1/2) Epoch 12, batch 4250, giga_loss[loss=0.2749, simple_loss=0.3472, pruned_loss=0.1013, over 28665.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3425, pruned_loss=0.09783, over 5714833.98 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3529, pruned_loss=0.096, over 5012530.27 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.342, pruned_loss=0.09833, over 5700374.11 frames. ], batch size: 307, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:48:06,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:48:10,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-06 02:48:29,157 INFO [train.py:968] (1/2) Epoch 12, batch 4300, giga_loss[loss=0.2428, simple_loss=0.311, pruned_loss=0.08728, over 28650.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3406, pruned_loss=0.09754, over 5725930.29 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3521, pruned_loss=0.09593, over 5056008.42 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3403, pruned_loss=0.09811, over 5706097.42 frames. ], batch size: 92, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:48:43,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2796, 1.6629, 1.6229, 1.4951], device='cuda:1'), covar=tensor([0.0749, 0.0289, 0.0288, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 02:48:49,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4640, 2.0995, 1.8334, 1.6642], device='cuda:1'), covar=tensor([0.0712, 0.0236, 0.0279, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 02:49:06,047 INFO [optim.py:369] (1/2) 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 02:49:08,484 INFO [train.py:968] (1/2) Epoch 12, batch 4350, giga_loss[loss=0.2724, simple_loss=0.3422, pruned_loss=0.1013, over 28870.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3376, pruned_loss=0.09607, over 5720661.94 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3524, pruned_loss=0.09608, over 5067199.87 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.337, pruned_loss=0.09641, over 5703797.14 frames. ], batch size: 186, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:49:10,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7098, 1.6436, 1.2786, 1.3357], device='cuda:1'), covar=tensor([0.0764, 0.0603, 0.1036, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0434, 0.0498, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:49:36,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=505576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:49:39,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:49:48,123 INFO [train.py:968] (1/2) Epoch 12, batch 4400, libri_loss[loss=0.2533, simple_loss=0.3406, pruned_loss=0.08303, over 29537.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3368, pruned_loss=0.09548, over 5714482.55 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3528, pruned_loss=0.09631, over 5083351.79 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3354, pruned_loss=0.09553, over 5705783.42 frames. ], batch size: 84, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:49:55,240 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:49:59,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=505608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:50:28,608 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 4450, libri_loss[loss=0.2829, simple_loss=0.3654, pruned_loss=0.1002, over 29553.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3392, pruned_loss=0.09649, over 5715834.59 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3525, pruned_loss=0.09627, over 5104738.28 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3379, pruned_loss=0.09656, over 5706352.24 frames. ], batch size: 80, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:50:34,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8284, 1.8257, 1.3436, 1.4108], device='cuda:1'), covar=tensor([0.0720, 0.0565, 0.0967, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0436, 0.0500, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 02:51:12,458 INFO [train.py:968] (1/2) Epoch 12, batch 4500, giga_loss[loss=0.3157, simple_loss=0.3852, pruned_loss=0.1231, over 27943.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.342, pruned_loss=0.09767, over 5700243.02 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3523, pruned_loss=0.09611, over 5115521.22 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3409, pruned_loss=0.09786, over 5698578.61 frames. ], batch size: 412, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:51:35,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2347, 1.9553, 1.4785, 0.3998], device='cuda:1'), covar=tensor([0.3382, 0.1756, 0.2876, 0.3815], device='cuda:1'), in_proj_covar=tensor([0.1544, 0.1460, 0.1483, 0.1272], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 02:51:53,485 INFO [optim.py:369] (1/2) 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 02:51:56,849 INFO [train.py:968] (1/2) Epoch 12, batch 4550, libri_loss[loss=0.3014, simple_loss=0.3783, pruned_loss=0.1123, over 29504.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3454, pruned_loss=0.099, over 5704988.97 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.353, pruned_loss=0.09641, over 5134053.12 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3438, pruned_loss=0.09896, over 5699967.64 frames. ], batch size: 85, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:52:41,505 INFO [train.py:968] (1/2) Epoch 12, batch 4600, giga_loss[loss=0.2503, simple_loss=0.3351, pruned_loss=0.08281, over 28904.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.346, pruned_loss=0.09866, over 5700332.48 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3525, pruned_loss=0.09625, over 5149272.57 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.345, pruned_loss=0.0988, over 5693001.74 frames. ], batch size: 174, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:53:21,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:53:23,785 INFO [optim.py:369] (1/2) 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 02:53:25,481 INFO [train.py:968] (1/2) Epoch 12, batch 4650, giga_loss[loss=0.2715, simple_loss=0.3503, pruned_loss=0.0963, over 28974.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3453, pruned_loss=0.09821, over 5699984.60 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3518, pruned_loss=0.09596, over 5160424.29 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3449, pruned_loss=0.09858, over 5691591.51 frames. ], batch size: 213, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:54:04,699 INFO [train.py:968] (1/2) Epoch 12, batch 4700, giga_loss[loss=0.2432, simple_loss=0.3185, pruned_loss=0.08399, over 28487.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3448, pruned_loss=0.09806, over 5706855.56 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.352, pruned_loss=0.09617, over 5193417.79 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.344, pruned_loss=0.09828, over 5696176.87 frames. ], batch size: 60, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:54:43,927 INFO [optim.py:369] (1/2) 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 02:54:45,473 INFO [train.py:968] (1/2) Epoch 12, batch 4750, giga_loss[loss=0.2861, simple_loss=0.353, pruned_loss=0.1096, over 28942.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3468, pruned_loss=0.09993, over 5702620.38 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.352, pruned_loss=0.09616, over 5205429.79 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3461, pruned_loss=0.1002, over 5692523.94 frames. ], batch size: 227, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:54:53,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=505951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:15,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=505977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:16,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=505979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:18,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=505982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:26,602 INFO [train.py:968] (1/2) Epoch 12, batch 4800, giga_loss[loss=0.2747, simple_loss=0.349, pruned_loss=0.1002, over 28582.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3486, pruned_loss=0.1012, over 5694283.83 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3525, pruned_loss=0.09667, over 5206520.93 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3474, pruned_loss=0.1011, over 5691699.49 frames. ], batch size: 85, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:55:31,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2695, 1.8946, 1.4128, 0.5769], device='cuda:1'), covar=tensor([0.4281, 0.2111, 0.2960, 0.4789], device='cuda:1'), in_proj_covar=tensor([0.1535, 0.1449, 0.1473, 0.1260], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 02:55:46,139 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:55:55,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1772, 3.3792, 2.3310, 1.2391], device='cuda:1'), covar=tensor([0.4506, 0.2052, 0.2517, 0.4250], device='cuda:1'), in_proj_covar=tensor([0.1537, 0.1451, 0.1473, 0.1261], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 02:55:56,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9162, 1.4997, 5.4072, 3.8318], device='cuda:1'), covar=tensor([0.1446, 0.2516, 0.0315, 0.0625], device='cuda:1'), in_proj_covar=tensor([0.0661, 0.0584, 0.0850, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:56:05,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9582, 1.8691, 1.4460, 1.5782], device='cuda:1'), covar=tensor([0.0761, 0.0725, 0.0973, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0437, 0.0498, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:56:07,579 INFO [optim.py:369] (1/2) 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 02:56:09,797 INFO [train.py:968] (1/2) Epoch 12, batch 4850, giga_loss[loss=0.2483, simple_loss=0.3366, pruned_loss=0.08003, over 28924.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3509, pruned_loss=0.1021, over 5696845.56 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3522, pruned_loss=0.09656, over 5220543.09 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3503, pruned_loss=0.1021, over 5693074.39 frames. ], batch size: 164, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:56:25,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:56:25,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5806, 1.7885, 1.3826, 1.9182], device='cuda:1'), covar=tensor([0.2317, 0.2385, 0.2670, 0.2415], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.0965, 0.1148, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 02:56:50,518 INFO [train.py:968] (1/2) Epoch 12, batch 4900, giga_loss[loss=0.2597, simple_loss=0.3434, pruned_loss=0.08799, over 28908.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3533, pruned_loss=0.1025, over 5705271.46 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3523, pruned_loss=0.09649, over 5232472.21 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3526, pruned_loss=0.1027, over 5701952.86 frames. ], batch size: 199, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 02:57:15,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:57:17,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:57:32,089 INFO [optim.py:369] (1/2) 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 02:57:32,656 INFO [train.py:968] (1/2) Epoch 12, batch 4950, giga_loss[loss=0.2892, simple_loss=0.3669, pruned_loss=0.1057, over 28995.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3544, pruned_loss=0.1032, over 5709349.58 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3523, pruned_loss=0.09643, over 5243876.28 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.354, pruned_loss=0.1036, over 5704852.85 frames. ], batch size: 174, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:57:35,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5575, 1.5716, 1.2999, 1.2316], device='cuda:1'), covar=tensor([0.0741, 0.0545, 0.0976, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0439, 0.0499, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:57:41,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:57:54,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1688, 2.0454, 1.5739, 1.7290], device='cuda:1'), covar=tensor([0.0726, 0.0734, 0.1003, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0439, 0.0499, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 02:58:15,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 02:58:15,640 INFO [train.py:968] (1/2) Epoch 12, batch 5000, libri_loss[loss=0.2554, simple_loss=0.3292, pruned_loss=0.09082, over 29368.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3548, pruned_loss=0.1032, over 5716819.43 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3525, pruned_loss=0.09658, over 5250167.73 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1034, over 5711842.98 frames. ], batch size: 67, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:58:54,617 INFO [optim.py:369] (1/2) 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 02:58:56,482 INFO [train.py:968] (1/2) Epoch 12, batch 5050, giga_loss[loss=0.2453, simple_loss=0.3284, pruned_loss=0.08109, over 28437.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3544, pruned_loss=0.1029, over 5719135.22 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.353, pruned_loss=0.09674, over 5255772.36 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.1031, over 5716244.24 frames. ], batch size: 71, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 02:58:59,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-06 02:59:07,774 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 02:59:37,449 INFO [train.py:968] (1/2) Epoch 12, batch 5100, giga_loss[loss=0.3191, simple_loss=0.373, pruned_loss=0.1326, over 26748.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3529, pruned_loss=0.1023, over 5713322.68 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3532, pruned_loss=0.09675, over 5261511.37 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3522, pruned_loss=0.1025, over 5711838.51 frames. ], batch size: 555, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 03:00:07,146 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:00:18,226 INFO [train.py:968] (1/2) Epoch 12, batch 5150, giga_loss[loss=0.2787, simple_loss=0.3518, pruned_loss=0.1028, over 28287.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3495, pruned_loss=0.1004, over 5722752.93 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3534, pruned_loss=0.09699, over 5274446.46 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3486, pruned_loss=0.1005, over 5718051.20 frames. ], batch size: 368, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 03:00:30,696 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:00:40,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7693, 3.5604, 3.3860, 1.6564], device='cuda:1'), covar=tensor([0.0695, 0.0790, 0.0784, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.1047, 0.0974, 0.0856, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 03:00:59,062 INFO [train.py:968] (1/2) Epoch 12, batch 5200, giga_loss[loss=0.2615, simple_loss=0.3344, pruned_loss=0.09433, over 28918.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3464, pruned_loss=0.09873, over 5727271.18 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3535, pruned_loss=0.09709, over 5290003.74 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3455, pruned_loss=0.09873, over 5719632.24 frames. ], batch size: 145, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:01:35,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:01:38,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7780, 1.7619, 1.3717, 1.4712], device='cuda:1'), covar=tensor([0.0781, 0.0674, 0.0997, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0438, 0.0499, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 03:01:38,985 INFO [optim.py:369] (1/2) 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 03:01:40,307 INFO [train.py:968] (1/2) Epoch 12, batch 5250, giga_loss[loss=0.2433, simple_loss=0.3322, pruned_loss=0.07719, over 29147.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3476, pruned_loss=0.09889, over 5720081.68 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3537, pruned_loss=0.09727, over 5298754.25 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3467, pruned_loss=0.09878, over 5712069.22 frames. ], batch size: 155, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:01:56,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9454, 2.0266, 1.4772, 1.7815], device='cuda:1'), covar=tensor([0.0788, 0.0653, 0.0943, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0439, 0.0499, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 03:02:04,284 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:02:06,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:02:18,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-06 03:02:23,794 INFO [train.py:968] (1/2) Epoch 12, batch 5300, giga_loss[loss=0.2475, simple_loss=0.3206, pruned_loss=0.08724, over 28497.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.35, pruned_loss=0.09914, over 5714822.58 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3539, pruned_loss=0.09742, over 5308103.00 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09897, over 5705885.31 frames. ], batch size: 85, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:02:31,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:02:47,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6407, 1.0722, 2.8565, 2.6451], device='cuda:1'), covar=tensor([0.1723, 0.2483, 0.0554, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0663, 0.0584, 0.0856, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 03:03:05,385 INFO [optim.py:369] (1/2) 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 03:03:06,085 INFO [train.py:968] (1/2) Epoch 12, batch 5350, giga_loss[loss=0.305, simple_loss=0.3717, pruned_loss=0.1191, over 27611.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3496, pruned_loss=0.09925, over 5711166.72 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.353, pruned_loss=0.09697, over 5322974.45 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3495, pruned_loss=0.09955, over 5699614.69 frames. ], batch size: 472, lr: 2.74e-03, grad_scale: 8.0
+2023-03-06 03:03:38,016 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:03:41,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:03:47,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5711, 1.8913, 1.4622, 1.6937], device='cuda:1'), covar=tensor([0.2232, 0.2208, 0.2532, 0.2159], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.0964, 0.1150, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 03:03:48,531 INFO [train.py:968] (1/2) Epoch 12, batch 5400, giga_loss[loss=0.2701, simple_loss=0.3401, pruned_loss=0.1001, over 28954.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1, over 5710559.91 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3532, pruned_loss=0.09707, over 5328533.46 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3481, pruned_loss=0.1002, over 5700112.52 frames. ], batch size: 136, lr: 2.74e-03, grad_scale: 4.0
+2023-03-06 03:03:49,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4220, 1.3974, 1.2486, 1.5481], device='cuda:1'), covar=tensor([0.0713, 0.0321, 0.0319, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0112, 0.0114, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 03:03:57,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:04:05,479 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:04:22,908 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 5450, giga_loss[loss=0.2664, simple_loss=0.3451, pruned_loss=0.09391, over 28674.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3478, pruned_loss=0.1011, over 5705348.00 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3539, pruned_loss=0.09749, over 5338647.86 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3469, pruned_loss=0.101, over 5697686.57 frames. ], batch size: 262, lr: 2.74e-03, grad_scale: 2.0
+2023-03-06 03:04:31,592 INFO [optim.py:369] (1/2) 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 03:04:49,235 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 5500, giga_loss[loss=0.2489, simple_loss=0.325, pruned_loss=0.08636, over 28971.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3461, pruned_loss=0.101, over 5704340.19 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3543, pruned_loss=0.09771, over 5345612.63 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3448, pruned_loss=0.1007, over 5698753.06 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:05:46,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:05:50,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2457, 1.2189, 1.1791, 1.4436], device='cuda:1'), covar=tensor([0.0691, 0.0396, 0.0334, 0.0771], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0112, 0.0114, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 03:05:55,859 INFO [train.py:968] (1/2) Epoch 12, batch 5550, giga_loss[loss=0.3286, simple_loss=0.3671, pruned_loss=0.1451, over 28624.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3444, pruned_loss=0.1006, over 5708046.53 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3549, pruned_loss=0.09807, over 5353498.08 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3429, pruned_loss=0.1002, over 5701604.33 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:05:58,243 INFO [optim.py:369] (1/2) 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 03:06:24,817 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:06:26,677 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 5600, giga_loss[loss=0.2347, simple_loss=0.3144, pruned_loss=0.0775, over 28917.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3423, pruned_loss=0.09935, over 5716728.34 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3544, pruned_loss=0.09779, over 5366825.69 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3413, pruned_loss=0.09929, over 5707544.12 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:06:51,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:07:21,099 INFO [train.py:968] (1/2) Epoch 12, batch 5650, giga_loss[loss=0.26, simple_loss=0.3345, pruned_loss=0.09278, over 27941.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3375, pruned_loss=0.09693, over 5720391.95 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3545, pruned_loss=0.09783, over 5370668.76 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3364, pruned_loss=0.09688, over 5714340.82 frames. ], batch size: 412, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:07:21,797 INFO [optim.py:369] (1/2) 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 03:07:30,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=506852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:07:44,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1979, 4.0480, 3.7793, 1.7989], device='cuda:1'), covar=tensor([0.0550, 0.0672, 0.0670, 0.2078], device='cuda:1'), in_proj_covar=tensor([0.1061, 0.0988, 0.0865, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 03:07:48,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=506875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:07:50,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=506878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:07:50,206 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-06 03:07:51,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1025, 3.9409, 3.6869, 1.7439], device='cuda:1'), covar=tensor([0.0580, 0.0672, 0.0686, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1060, 0.0986, 0.0863, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 03:08:00,695 INFO [train.py:968] (1/2) Epoch 12, batch 5700, giga_loss[loss=0.2322, simple_loss=0.303, pruned_loss=0.0807, over 28900.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3331, pruned_loss=0.0944, over 5723566.59 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.354, pruned_loss=0.09761, over 5382117.34 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3322, pruned_loss=0.09447, over 5716076.48 frames. ], batch size: 106, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:08:14,774 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=506907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:08:41,711 INFO [train.py:968] (1/2) Epoch 12, batch 5750, giga_loss[loss=0.2923, simple_loss=0.3537, pruned_loss=0.1155, over 28193.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3355, pruned_loss=0.09592, over 5721459.56 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3543, pruned_loss=0.09777, over 5386588.89 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3344, pruned_loss=0.09583, over 5714644.87 frames. ], batch size: 77, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:08:42,242 INFO [optim.py:369] (1/2) 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 03:09:10,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=506978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:09:19,770 INFO [train.py:968] (1/2) Epoch 12, batch 5800, giga_loss[loss=0.2564, simple_loss=0.3396, pruned_loss=0.08657, over 28977.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3398, pruned_loss=0.09812, over 5726547.99 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3544, pruned_loss=0.09799, over 5402041.39 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3384, pruned_loss=0.09787, over 5716068.17 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:09:57,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:09:57,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-06 03:09:59,168 INFO [train.py:968] (1/2) Epoch 12, batch 5850, giga_loss[loss=0.2717, simple_loss=0.3422, pruned_loss=0.1006, over 29139.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3426, pruned_loss=0.09889, over 5727356.69 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3545, pruned_loss=0.09796, over 5411107.36 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3411, pruned_loss=0.09873, over 5718026.50 frames. ], batch size: 113, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:10:00,649 INFO [optim.py:369] (1/2) 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 03:10:12,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5708, 4.3361, 1.7372, 1.7895], device='cuda:1'), covar=tensor([0.0871, 0.0301, 0.0858, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0511, 0.0341, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 03:10:13,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:10:23,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8649, 2.0414, 2.1598, 1.6987], device='cuda:1'), covar=tensor([0.1707, 0.2033, 0.1316, 0.1487], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0691, 0.0868, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 03:10:42,824 INFO [train.py:968] (1/2) Epoch 12, batch 5900, giga_loss[loss=0.3108, simple_loss=0.3814, pruned_loss=0.1201, over 28040.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3457, pruned_loss=0.1, over 5711201.24 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3548, pruned_loss=0.09833, over 5411331.99 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3441, pruned_loss=0.09959, over 5708932.82 frames. ], batch size: 412, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:11:06,798 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:11:09,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:11:23,959 INFO [train.py:968] (1/2) Epoch 12, batch 5950, giga_loss[loss=0.2897, simple_loss=0.3622, pruned_loss=0.1086, over 28949.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3483, pruned_loss=0.1008, over 5717890.94 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3553, pruned_loss=0.09865, over 5430487.11 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3462, pruned_loss=0.1003, over 5710766.34 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:11:24,727 INFO [optim.py:369] (1/2) 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 03:11:35,188 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:12:01,146 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:12:03,874 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 6000, giga_loss[loss=0.34, simple_loss=0.3916, pruned_loss=0.1442, over 28830.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3516, pruned_loss=0.1036, over 5713521.25 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3553, pruned_loss=0.09873, over 5435579.68 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3499, pruned_loss=0.1032, over 5705932.11 frames. ], batch size: 119, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:12:11,491 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 03:12:21,240 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 03:12:41,640 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:12:53,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:13:07,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5495, 1.7392, 1.8148, 1.3768], device='cuda:1'), covar=tensor([0.1469, 0.2063, 0.1206, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0828, 0.0692, 0.0871, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 03:13:08,579 INFO [train.py:968] (1/2) Epoch 12, batch 6050, giga_loss[loss=0.3165, simple_loss=0.3765, pruned_loss=0.1283, over 29088.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3579, pruned_loss=0.1088, over 5710923.52 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3552, pruned_loss=0.0986, over 5442693.27 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3566, pruned_loss=0.1087, over 5702317.94 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:13:10,128 INFO [optim.py:369] (1/2) 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 03:13:40,699 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:13:49,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8884, 1.0569, 1.0719, 0.8319], device='cuda:1'), covar=tensor([0.1326, 0.1530, 0.0899, 0.1258], device='cuda:1'), in_proj_covar=tensor([0.1719, 0.1635, 0.1603, 0.1683], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 03:13:59,859 INFO [train.py:968] (1/2) Epoch 12, batch 6100, giga_loss[loss=0.3777, simple_loss=0.4388, pruned_loss=0.1584, over 28657.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3644, pruned_loss=0.1143, over 5675425.62 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3553, pruned_loss=0.09867, over 5442133.86 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3634, pruned_loss=0.1143, over 5674774.62 frames. ], batch size: 262, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:14:47,786 INFO [train.py:968] (1/2) Epoch 12, batch 6150, giga_loss[loss=0.3017, simple_loss=0.3764, pruned_loss=0.1134, over 28767.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3727, pruned_loss=0.1203, over 5680295.19 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3557, pruned_loss=0.09896, over 5448810.67 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3718, pruned_loss=0.1203, over 5677006.15 frames. ], batch size: 119, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:14:49,657 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:15:19,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:15:37,570 INFO [train.py:968] (1/2) Epoch 12, batch 6200, giga_loss[loss=0.3355, simple_loss=0.3933, pruned_loss=0.1388, over 28915.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3777, pruned_loss=0.1251, over 5679976.97 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3561, pruned_loss=0.09919, over 5459623.87 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.377, pruned_loss=0.1255, over 5672804.45 frames. ], batch size: 106, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:15:45,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:15:52,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 03:16:11,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5575, 1.8277, 1.9173, 1.4193], device='cuda:1'), covar=tensor([0.1548, 0.2066, 0.1209, 0.1497], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0693, 0.0871, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 03:16:13,815 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 6250, giga_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 28703.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.383, pruned_loss=0.1292, over 5666729.03 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3564, pruned_loss=0.09944, over 5452996.18 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3829, pruned_loss=0.13, over 5673761.64 frames. ], batch size: 92, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:16:25,721 INFO [optim.py:369] (1/2) 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 03:16:48,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3213, 1.5211, 1.0427, 1.2135], device='cuda:1'), covar=tensor([0.1054, 0.0785, 0.1475, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0443, 0.0499, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 03:17:17,468 INFO [train.py:968] (1/2) Epoch 12, batch 6300, giga_loss[loss=0.3834, simple_loss=0.4257, pruned_loss=0.1705, over 28624.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3887, pruned_loss=0.1352, over 5647260.54 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3566, pruned_loss=0.09947, over 5455232.20 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3885, pruned_loss=0.1359, over 5651573.14 frames. ], batch size: 307, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:17:53,795 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:18:07,941 INFO [train.py:968] (1/2) Epoch 12, batch 6350, libri_loss[loss=0.292, simple_loss=0.3712, pruned_loss=0.1064, over 28715.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3915, pruned_loss=0.1387, over 5638069.37 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3569, pruned_loss=0.09961, over 5468041.31 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3921, pruned_loss=0.1403, over 5634841.88 frames. ], batch size: 106, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:18:08,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:18:09,276 INFO [optim.py:369] (1/2) 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 03:18:44,625 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:18:49,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:18:58,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6500, 1.0307, 2.8593, 2.6446], device='cuda:1'), covar=tensor([0.1710, 0.2425, 0.0564, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0664, 0.0586, 0.0862, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 03:19:02,180 INFO [train.py:968] (1/2) Epoch 12, batch 6400, giga_loss[loss=0.5353, simple_loss=0.5227, pruned_loss=0.2739, over 26500.00 frames. ], tot_loss[loss=0.3422, simple_loss=0.3959, pruned_loss=0.1442, over 5615417.83 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3567, pruned_loss=0.09956, over 5470647.06 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3969, pruned_loss=0.1459, over 5611863.63 frames. ], batch size: 555, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:19:21,011 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:19:57,635 INFO [train.py:968] (1/2) Epoch 12, batch 6450, giga_loss[loss=0.3054, simple_loss=0.3792, pruned_loss=0.1158, over 28914.00 frames. ], tot_loss[loss=0.3453, simple_loss=0.3983, pruned_loss=0.1462, over 5613074.05 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3566, pruned_loss=0.09959, over 5475399.16 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.3997, pruned_loss=0.1482, over 5607746.87 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:19:59,570 INFO [optim.py:369] (1/2) 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 03:20:01,091 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4599, 4.2737, 4.0350, 1.8031], device='cuda:1'), covar=tensor([0.0588, 0.0748, 0.0782, 0.2127], device='cuda:1'), in_proj_covar=tensor([0.1063, 0.0992, 0.0865, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 03:20:06,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:20:34,574 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=507678.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 03:20:48,752 INFO [train.py:968] (1/2) Epoch 12, batch 6500, giga_loss[loss=0.414, simple_loss=0.4407, pruned_loss=0.1937, over 26620.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3984, pruned_loss=0.1467, over 5624063.04 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3562, pruned_loss=0.09941, over 5479575.06 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4002, pruned_loss=0.1489, over 5617358.91 frames. ], batch size: 555, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:20:49,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1617, 1.4824, 1.4661, 1.0957], device='cuda:1'), covar=tensor([0.1107, 0.1779, 0.0917, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0692, 0.0865, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 03:21:40,299 INFO [train.py:968] (1/2) Epoch 12, batch 6550, giga_loss[loss=0.4327, simple_loss=0.4424, pruned_loss=0.2115, over 23536.00 frames. ], tot_loss[loss=0.3444, simple_loss=0.3965, pruned_loss=0.1461, over 5631827.69 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3563, pruned_loss=0.09949, over 5488043.96 frames. ], giga_tot_loss[loss=0.348, simple_loss=0.3987, pruned_loss=0.1487, over 5621567.58 frames. ], batch size: 705, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:21:42,349 INFO [optim.py:369] (1/2) 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 03:22:27,717 INFO [train.py:968] (1/2) Epoch 12, batch 6600, giga_loss[loss=0.3827, simple_loss=0.4359, pruned_loss=0.1648, over 28829.00 frames. ], tot_loss[loss=0.3411, simple_loss=0.3944, pruned_loss=0.1439, over 5635999.13 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3561, pruned_loss=0.09938, over 5500817.62 frames. ], giga_tot_loss[loss=0.3463, simple_loss=0.3976, pruned_loss=0.1475, over 5620828.34 frames. ], batch size: 186, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:22:27,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=507792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:22:31,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=507795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:23:01,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=507824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:23:17,619 INFO [train.py:968] (1/2) Epoch 12, batch 6650, giga_loss[loss=0.3127, simple_loss=0.3822, pruned_loss=0.1216, over 28715.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3949, pruned_loss=0.143, over 5634409.29 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3562, pruned_loss=0.09947, over 5495542.39 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3979, pruned_loss=0.1464, over 5628898.03 frames. ], batch size: 262, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:23:19,652 INFO [optim.py:369] (1/2) 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 03:23:46,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-06 03:24:08,904 INFO [train.py:968] (1/2) Epoch 12, batch 6700, giga_loss[loss=0.3815, simple_loss=0.4091, pruned_loss=0.1769, over 23531.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3967, pruned_loss=0.1445, over 5605903.57 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3562, pruned_loss=0.09953, over 5491993.23 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.3995, pruned_loss=0.1476, over 5606713.03 frames. ], batch size: 705, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:24:19,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:24:32,063 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-06 03:24:35,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=507918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:24:50,877 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 6750, giga_loss[loss=0.4451, simple_loss=0.46, pruned_loss=0.2151, over 26355.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.395, pruned_loss=0.1425, over 5608810.91 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3567, pruned_loss=0.09979, over 5494749.08 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.3979, pruned_loss=0.146, over 5609366.40 frames. ], batch size: 555, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:24:59,372 INFO [optim.py:369] (1/2) 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 03:25:03,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1716, 1.7464, 1.5584, 1.2724], device='cuda:1'), covar=tensor([0.0713, 0.0274, 0.0250, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0113, 0.0114, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0081, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 03:25:04,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1321, 1.6110, 1.4937, 1.0335], device='cuda:1'), covar=tensor([0.1603, 0.2477, 0.1378, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0695, 0.0866, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 03:25:30,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1407, 2.0639, 1.6165, 1.8874], device='cuda:1'), covar=tensor([0.0772, 0.0726, 0.0966, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0442, 0.0498, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 03:25:48,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 6800, libri_loss[loss=0.3223, simple_loss=0.4054, pruned_loss=0.1196, over 26153.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3925, pruned_loss=0.139, over 5615183.45 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.357, pruned_loss=0.0999, over 5501180.24 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3954, pruned_loss=0.1426, over 5612479.79 frames. ], batch size: 136, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:26:11,366 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 12, batch 6850, giga_loss[loss=0.2572, simple_loss=0.3331, pruned_loss=0.09063, over 28688.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3892, pruned_loss=0.1353, over 5630976.59 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3568, pruned_loss=0.09978, over 5506631.39 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3921, pruned_loss=0.1388, over 5625538.74 frames. ], batch size: 78, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:26:42,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:26:42,516 INFO [optim.py:369] (1/2) 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 03:26:44,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:26:50,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508053.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 03:26:50,352 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2927, 1.4348, 1.4112, 1.3619], device='cuda:1'), covar=tensor([0.1910, 0.1684, 0.1519, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.1730, 0.1644, 0.1609, 0.1700], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 03:26:58,258 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:27:00,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:27:14,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:27:28,068 INFO [train.py:968] (1/2) Epoch 12, batch 6900, giga_loss[loss=0.2852, simple_loss=0.3605, pruned_loss=0.1049, over 28656.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3869, pruned_loss=0.1335, over 5637323.48 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3566, pruned_loss=0.09972, over 5508150.84 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3897, pruned_loss=0.1368, over 5633291.55 frames. ], batch size: 242, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:27:30,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:27:33,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5086, 1.7687, 1.4246, 1.7088], device='cuda:1'), covar=tensor([0.2385, 0.2393, 0.2717, 0.2179], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.0966, 0.1156, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 03:27:41,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-06 03:28:14,916 INFO [train.py:968] (1/2) Epoch 12, batch 6950, libri_loss[loss=0.226, simple_loss=0.312, pruned_loss=0.06998, over 29537.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3833, pruned_loss=0.1305, over 5648465.84 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3564, pruned_loss=0.09959, over 5518771.77 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3869, pruned_loss=0.1345, over 5639883.95 frames. ], batch size: 74, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:28:18,209 INFO [optim.py:369] (1/2) 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 03:28:54,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4864, 1.7991, 1.4289, 1.5662], device='cuda:1'), covar=tensor([0.2370, 0.2365, 0.2680, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.0966, 0.1154, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 03:28:59,195 INFO [train.py:968] (1/2) Epoch 12, batch 7000, giga_loss[loss=0.3058, simple_loss=0.3774, pruned_loss=0.1171, over 29005.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3826, pruned_loss=0.1302, over 5655857.98 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3564, pruned_loss=0.09975, over 5529910.02 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3862, pruned_loss=0.1341, over 5642531.22 frames. ], batch size: 164, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:29:04,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508196.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 03:29:06,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508199.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 03:29:42,567 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508228.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 03:29:56,526 INFO [train.py:968] (1/2) Epoch 12, batch 7050, libri_loss[loss=0.2627, simple_loss=0.3435, pruned_loss=0.09095, over 29540.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3819, pruned_loss=0.1293, over 5662536.21 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3562, pruned_loss=0.09964, over 5537412.51 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3855, pruned_loss=0.133, over 5647390.27 frames. ], batch size: 80, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:30:01,210 INFO [optim.py:369] (1/2) 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 03:30:16,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 03:30:47,683 INFO [train.py:968] (1/2) Epoch 12, batch 7100, giga_loss[loss=0.2737, simple_loss=0.3694, pruned_loss=0.08899, over 28879.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3787, pruned_loss=0.1257, over 5672517.90 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3561, pruned_loss=0.09962, over 5544588.50 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.382, pruned_loss=0.1293, over 5656295.81 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:31:04,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:31:26,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-06 03:31:42,222 INFO [train.py:968] (1/2) Epoch 12, batch 7150, giga_loss[loss=0.3152, simple_loss=0.3913, pruned_loss=0.1195, over 28944.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3786, pruned_loss=0.1234, over 5673295.22 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3563, pruned_loss=0.09966, over 5553050.38 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3818, pruned_loss=0.127, over 5656315.21 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:31:47,109 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=508351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:32:04,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:32:26,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-06 03:32:27,373 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 7200, giga_loss[loss=0.3008, simple_loss=0.3688, pruned_loss=0.1164, over 28919.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3801, pruned_loss=0.1233, over 5678781.98 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.356, pruned_loss=0.09943, over 5562749.39 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3836, pruned_loss=0.127, over 5659595.34 frames. ], batch size: 66, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:32:29,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:32:35,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5220, 1.8753, 1.5905, 1.8132], device='cuda:1'), covar=tensor([0.0729, 0.0276, 0.0288, 0.0765], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 03:33:24,639 INFO [train.py:968] (1/2) Epoch 12, batch 7250, giga_loss[loss=0.3265, simple_loss=0.3832, pruned_loss=0.1348, over 27973.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3811, pruned_loss=0.1251, over 5669674.69 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3559, pruned_loss=0.09938, over 5557697.82 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3844, pruned_loss=0.1286, over 5660723.69 frames. ], batch size: 412, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:33:29,049 INFO [optim.py:369] (1/2) 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 03:33:33,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:33:37,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:04,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:11,703 INFO [train.py:968] (1/2) Epoch 12, batch 7300, giga_loss[loss=0.2951, simple_loss=0.3659, pruned_loss=0.1121, over 28673.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3794, pruned_loss=0.1242, over 5671717.59 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.09933, over 5562027.51 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3825, pruned_loss=0.1273, over 5662222.16 frames. ], batch size: 242, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:34:25,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:26,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-06 03:34:28,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 03:34:30,729 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:55,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:56,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:56,748 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:34:59,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=508538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:35:00,656 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 7350, libri_loss[loss=0.3297, simple_loss=0.4041, pruned_loss=0.1277, over 29375.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3784, pruned_loss=0.125, over 5668038.54 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3561, pruned_loss=0.09952, over 5568707.00 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3811, pruned_loss=0.1279, over 5656577.87 frames. ], batch size: 92, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:35:06,089 INFO [optim.py:369] (1/2) 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 03:35:20,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:35:25,485 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:35:39,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-06 03:35:46,363 INFO [train.py:968] (1/2) Epoch 12, batch 7400, giga_loss[loss=0.2895, simple_loss=0.3639, pruned_loss=0.1076, over 28913.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3766, pruned_loss=0.124, over 5678892.12 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3561, pruned_loss=0.09935, over 5577039.06 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3793, pruned_loss=0.1272, over 5664828.94 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:36:38,441 INFO [train.py:968] (1/2) Epoch 12, batch 7450, giga_loss[loss=0.3232, simple_loss=0.3984, pruned_loss=0.124, over 29043.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3757, pruned_loss=0.1224, over 5689848.33 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3562, pruned_loss=0.09941, over 5584103.88 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3782, pruned_loss=0.1255, over 5674530.45 frames. ], batch size: 155, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:36:41,844 INFO [optim.py:369] (1/2) 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 03:36:45,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3828, 1.6364, 1.5578, 1.4540], device='cuda:1'), covar=tensor([0.1576, 0.1829, 0.2177, 0.2004], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0728, 0.0672, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 03:36:49,344 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 03:37:09,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1064, 1.3648, 1.2982, 0.9997], device='cuda:1'), covar=tensor([0.2518, 0.2059, 0.1334, 0.1955], device='cuda:1'), in_proj_covar=tensor([0.1737, 0.1643, 0.1604, 0.1705], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 03:37:26,866 INFO [train.py:968] (1/2) Epoch 12, batch 7500, giga_loss[loss=0.2968, simple_loss=0.37, pruned_loss=0.1118, over 28652.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3754, pruned_loss=0.121, over 5697752.90 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3563, pruned_loss=0.0995, over 5587607.09 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3776, pruned_loss=0.1237, over 5683923.97 frames. ], batch size: 92, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:37:51,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=508717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:37:59,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=508726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:38:13,077 INFO [train.py:968] (1/2) Epoch 12, batch 7550, giga_loss[loss=0.2707, simple_loss=0.3432, pruned_loss=0.09903, over 28606.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3751, pruned_loss=0.121, over 5696931.51 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3561, pruned_loss=0.09936, over 5592964.73 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3774, pruned_loss=0.1236, over 5682981.77 frames. ], batch size: 60, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:38:16,756 INFO [optim.py:369] (1/2) 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 03:38:20,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-06 03:38:24,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2532, 1.5777, 1.3355, 1.4839], device='cuda:1'), covar=tensor([0.0706, 0.0367, 0.0309, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 03:38:41,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-06 03:38:56,923 INFO [train.py:968] (1/2) Epoch 12, batch 7600, giga_loss[loss=0.2969, simple_loss=0.3655, pruned_loss=0.1142, over 28653.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3737, pruned_loss=0.1204, over 5691837.66 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.356, pruned_loss=0.09937, over 5595685.18 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3763, pruned_loss=0.1233, over 5682236.19 frames. ], batch size: 262, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:39:42,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-06 03:39:43,494 INFO [train.py:968] (1/2) Epoch 12, batch 7650, giga_loss[loss=0.3322, simple_loss=0.3817, pruned_loss=0.1414, over 28773.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3713, pruned_loss=0.1192, over 5697643.75 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3556, pruned_loss=0.0991, over 5603337.48 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.374, pruned_loss=0.1222, over 5685513.93 frames. ], batch size: 92, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:39:52,020 INFO [optim.py:369] (1/2) 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 03:40:01,972 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=508869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:40:16,404 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 7700, giga_loss[loss=0.3509, simple_loss=0.403, pruned_loss=0.1495, over 29043.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3722, pruned_loss=0.1209, over 5694617.84 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3558, pruned_loss=0.0992, over 5609395.27 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3745, pruned_loss=0.1236, over 5680849.54 frames. ], batch size: 155, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:40:43,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=508901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:41:25,360 INFO [train.py:968] (1/2) Epoch 12, batch 7750, giga_loss[loss=0.2655, simple_loss=0.3447, pruned_loss=0.09315, over 28926.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3712, pruned_loss=0.1208, over 5701660.63 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3556, pruned_loss=0.09898, over 5613941.34 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1234, over 5687823.74 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:41:30,024 INFO [optim.py:369] (1/2) 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 03:42:14,533 INFO [train.py:968] (1/2) Epoch 12, batch 7800, giga_loss[loss=0.4009, simple_loss=0.4343, pruned_loss=0.1838, over 28625.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1209, over 5701745.67 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3559, pruned_loss=0.09905, over 5618181.56 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3721, pruned_loss=0.1234, over 5688741.26 frames. ], batch size: 307, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:42:22,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-06 03:42:54,269 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=509035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:43:03,046 INFO [train.py:968] (1/2) Epoch 12, batch 7850, giga_loss[loss=0.3483, simple_loss=0.4056, pruned_loss=0.1455, over 28759.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1213, over 5703307.64 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.356, pruned_loss=0.09916, over 5621037.69 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3712, pruned_loss=0.1233, over 5691281.89 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:43:08,158 INFO [optim.py:369] (1/2) 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 03:43:52,391 INFO [train.py:968] (1/2) Epoch 12, batch 7900, giga_loss[loss=0.2597, simple_loss=0.3399, pruned_loss=0.08978, over 28986.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3704, pruned_loss=0.1217, over 5695401.98 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.356, pruned_loss=0.09919, over 5621534.21 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3717, pruned_loss=0.1236, over 5686642.50 frames. ], batch size: 136, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:43:52,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=509092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:44:36,366 INFO [train.py:968] (1/2) Epoch 12, batch 7950, libri_loss[loss=0.3249, simple_loss=0.3865, pruned_loss=0.1317, over 19503.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3711, pruned_loss=0.1217, over 5687345.81 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3558, pruned_loss=0.0992, over 5623780.72 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3729, pruned_loss=0.124, over 5682276.02 frames. ], batch size: 187, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:44:42,383 INFO [optim.py:369] (1/2) 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 03:45:19,127 INFO [train.py:968] (1/2) Epoch 12, batch 8000, giga_loss[loss=0.3119, simple_loss=0.3797, pruned_loss=0.1221, over 28568.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3704, pruned_loss=0.1201, over 5678469.88 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.09924, over 5624679.82 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3723, pruned_loss=0.1227, over 5675164.44 frames. ], batch size: 85, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:45:59,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=509233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:46:01,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=509235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:46:04,129 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 8050, giga_loss[loss=0.278, simple_loss=0.3489, pruned_loss=0.1035, over 28819.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.372, pruned_loss=0.1214, over 5674254.77 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3562, pruned_loss=0.09951, over 5628056.95 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1236, over 5669229.61 frames. ], batch size: 119, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:46:14,233 INFO [optim.py:369] (1/2) 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 03:46:32,447 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=509267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:46:59,299 INFO [train.py:968] (1/2) Epoch 12, batch 8100, giga_loss[loss=0.2864, simple_loss=0.3575, pruned_loss=0.1077, over 28690.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3727, pruned_loss=0.1221, over 5684121.65 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.356, pruned_loss=0.09944, over 5633161.32 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3743, pruned_loss=0.1243, over 5676417.00 frames. ], batch size: 262, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:47:47,127 INFO [train.py:968] (1/2) Epoch 12, batch 8150, giga_loss[loss=0.3998, simple_loss=0.4429, pruned_loss=0.1783, over 27461.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.1241, over 5683374.83 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3558, pruned_loss=0.09932, over 5641863.42 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3764, pruned_loss=0.1269, over 5671043.15 frames. ], batch size: 472, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:47:53,740 INFO [optim.py:369] (1/2) 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 03:48:13,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8450, 1.6705, 1.3686, 1.4402], device='cuda:1'), covar=tensor([0.0747, 0.0731, 0.0964, 0.1166], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0444, 0.0502, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 03:48:20,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=509376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:48:22,249 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 8200, giga_loss[loss=0.292, simple_loss=0.3636, pruned_loss=0.1102, over 28899.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3743, pruned_loss=0.1247, over 5682451.18 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3559, pruned_loss=0.09969, over 5643738.09 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1278, over 5671574.50 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:48:48,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=509408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:48:52,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=509410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:49:24,310 INFO [train.py:968] (1/2) Epoch 12, batch 8250, giga_loss[loss=0.3874, simple_loss=0.4213, pruned_loss=0.1768, over 29056.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3787, pruned_loss=0.1294, over 5671315.55 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3562, pruned_loss=0.1, over 5648445.53 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.381, pruned_loss=0.1322, over 5659384.04 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:49:29,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5933, 3.7094, 1.6257, 1.6954], device='cuda:1'), covar=tensor([0.0904, 0.0308, 0.0827, 0.1245], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0519, 0.0344, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 03:49:31,022 INFO [optim.py:369] (1/2) 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 03:49:49,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2618, 1.5150, 1.2591, 1.0653], device='cuda:1'), covar=tensor([0.1910, 0.1811, 0.1845, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.1307, 0.0972, 0.1157, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 03:50:11,466 INFO [train.py:968] (1/2) Epoch 12, batch 8300, giga_loss[loss=0.2663, simple_loss=0.3364, pruned_loss=0.09813, over 29103.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3786, pruned_loss=0.1296, over 5665253.49 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3564, pruned_loss=0.1, over 5643392.70 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3806, pruned_loss=0.1324, over 5661005.46 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:50:54,617 INFO [train.py:968] (1/2) Epoch 12, batch 8350, giga_loss[loss=0.3032, simple_loss=0.38, pruned_loss=0.1132, over 28933.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3771, pruned_loss=0.1283, over 5666074.01 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3561, pruned_loss=0.09985, over 5644251.52 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3794, pruned_loss=0.1313, over 5662755.75 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:51:01,319 INFO [optim.py:369] (1/2) 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 03:51:03,484 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=509553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:51:06,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=509556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:51:28,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5946, 2.3341, 1.5838, 0.6688], device='cuda:1'), covar=tensor([0.3654, 0.2467, 0.3595, 0.4553], device='cuda:1'), in_proj_covar=tensor([0.1552, 0.1488, 0.1481, 0.1289], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 03:51:30,949 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=509585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:51:36,713 INFO [train.py:968] (1/2) Epoch 12, batch 8400, giga_loss[loss=0.3155, simple_loss=0.3714, pruned_loss=0.1298, over 28312.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3776, pruned_loss=0.1275, over 5675729.16 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3561, pruned_loss=0.0998, over 5652121.71 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1308, over 5666762.11 frames. ], batch size: 368, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:52:13,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3444, 1.6546, 1.2549, 1.3856], device='cuda:1'), covar=tensor([0.2428, 0.2388, 0.2708, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.0973, 0.1152, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 03:52:19,852 INFO [train.py:968] (1/2) Epoch 12, batch 8450, giga_loss[loss=0.2833, simple_loss=0.3477, pruned_loss=0.1095, over 28278.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3751, pruned_loss=0.1251, over 5667406.84 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3565, pruned_loss=0.1001, over 5649681.67 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3773, pruned_loss=0.1281, over 5662118.57 frames. ], batch size: 369, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:52:27,020 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 8500, giga_loss[loss=0.2738, simple_loss=0.3421, pruned_loss=0.1027, over 28887.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3732, pruned_loss=0.1242, over 5677595.01 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3564, pruned_loss=0.1002, over 5657501.34 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3759, pruned_loss=0.1277, over 5667238.82 frames. ], batch size: 145, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:53:47,729 INFO [train.py:968] (1/2) Epoch 12, batch 8550, giga_loss[loss=0.3159, simple_loss=0.3749, pruned_loss=0.1285, over 28881.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3711, pruned_loss=0.1233, over 5681205.51 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3561, pruned_loss=0.1001, over 5663908.42 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3739, pruned_loss=0.1269, over 5667587.14 frames. ], batch size: 186, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:53:57,299 INFO [optim.py:369] (1/2) 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 03:54:41,358 INFO [train.py:968] (1/2) Epoch 12, batch 8600, giga_loss[loss=0.2926, simple_loss=0.3606, pruned_loss=0.1124, over 29057.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3724, pruned_loss=0.1251, over 5658987.96 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3563, pruned_loss=0.1001, over 5667196.55 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3748, pruned_loss=0.1283, over 5645483.23 frames. ], batch size: 128, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:54:58,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4145, 1.7107, 1.3563, 1.5585], device='cuda:1'), covar=tensor([0.2673, 0.2605, 0.2917, 0.2234], device='cuda:1'), in_proj_covar=tensor([0.1305, 0.0971, 0.1154, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 03:55:15,799 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-06 03:55:27,908 INFO [train.py:968] (1/2) Epoch 12, batch 8650, giga_loss[loss=0.2897, simple_loss=0.3786, pruned_loss=0.1004, over 28976.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3747, pruned_loss=0.1251, over 5673388.49 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3562, pruned_loss=0.09997, over 5675847.72 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3773, pruned_loss=0.1287, over 5654342.55 frames. ], batch size: 155, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:55:34,571 INFO [optim.py:369] (1/2) 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 03:55:39,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3404, 1.7059, 1.3380, 1.4670], device='cuda:1'), covar=tensor([0.0730, 0.0325, 0.0313, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 03:56:14,868 INFO [train.py:968] (1/2) Epoch 12, batch 8700, giga_loss[loss=0.3052, simple_loss=0.3615, pruned_loss=0.1245, over 23804.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3762, pruned_loss=0.1238, over 5670240.06 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3557, pruned_loss=0.09967, over 5676608.02 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3793, pruned_loss=0.1275, over 5654485.95 frames. ], batch size: 705, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:57:01,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3113, 1.5691, 1.4533, 1.2637], device='cuda:1'), covar=tensor([0.2162, 0.1917, 0.1330, 0.1688], device='cuda:1'), in_proj_covar=tensor([0.1733, 0.1634, 0.1603, 0.1695], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 03:57:03,617 INFO [train.py:968] (1/2) Epoch 12, batch 8750, giga_loss[loss=0.3594, simple_loss=0.401, pruned_loss=0.1589, over 23666.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.378, pruned_loss=0.1238, over 5677510.39 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3551, pruned_loss=0.09934, over 5680212.59 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3812, pruned_loss=0.1274, over 5661862.00 frames. ], batch size: 705, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:57:11,479 INFO [optim.py:369] (1/2) 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 03:57:29,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3999, 2.9556, 2.3433, 2.1060], device='cuda:1'), covar=tensor([0.2073, 0.1349, 0.1674, 0.1756], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1640, 0.1608, 0.1701], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 03:57:48,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-06 03:57:48,967 INFO [train.py:968] (1/2) Epoch 12, batch 8800, giga_loss[loss=0.3325, simple_loss=0.3981, pruned_loss=0.1334, over 28664.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3816, pruned_loss=0.127, over 5675047.74 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3551, pruned_loss=0.09934, over 5682902.00 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3846, pruned_loss=0.1303, over 5660144.20 frames. ], batch size: 284, lr: 2.73e-03, grad_scale: 8.0
+2023-03-06 03:58:35,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=510039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 03:58:36,877 INFO [train.py:968] (1/2) Epoch 12, batch 8850, libri_loss[loss=0.3005, simple_loss=0.3744, pruned_loss=0.1133, over 20325.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3816, pruned_loss=0.1278, over 5655591.72 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3546, pruned_loss=0.09903, over 5679441.85 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3853, pruned_loss=0.1316, over 5646954.81 frames. ], batch size: 187, lr: 2.73e-03, grad_scale: 4.0
+2023-03-06 03:58:44,763 INFO [optim.py:369] (1/2) 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 03:59:14,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7242, 3.4589, 1.6767, 1.7940], device='cuda:1'), covar=tensor([0.0810, 0.0343, 0.0796, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0518, 0.0342, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 03:59:22,758 INFO [train.py:968] (1/2) Epoch 12, batch 8900, giga_loss[loss=0.35, simple_loss=0.4051, pruned_loss=0.1475, over 28860.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.38, pruned_loss=0.1273, over 5654208.46 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3547, pruned_loss=0.09915, over 5675389.77 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3834, pruned_loss=0.1308, over 5649881.38 frames. ], batch size: 174, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 03:59:31,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5892, 4.1607, 1.7401, 1.8157], device='cuda:1'), covar=tensor([0.0892, 0.0313, 0.0807, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0518, 0.0342, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 03:59:53,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9273, 5.7056, 5.3693, 3.1663], device='cuda:1'), covar=tensor([0.0480, 0.0720, 0.0843, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.1079, 0.1013, 0.0883, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 04:00:14,638 INFO [train.py:968] (1/2) Epoch 12, batch 8950, giga_loss[loss=0.3406, simple_loss=0.3935, pruned_loss=0.1438, over 27985.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3796, pruned_loss=0.1281, over 5642457.13 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3547, pruned_loss=0.09917, over 5678273.07 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3826, pruned_loss=0.1313, over 5636262.04 frames. ], batch size: 412, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:00:21,932 INFO [optim.py:369] (1/2) 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 04:00:43,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-06 04:00:59,272 INFO [train.py:968] (1/2) Epoch 12, batch 9000, giga_loss[loss=0.3213, simple_loss=0.3809, pruned_loss=0.1309, over 28877.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.377, pruned_loss=0.1262, over 5661086.12 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3546, pruned_loss=0.09898, over 5687555.24 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3805, pruned_loss=0.1302, over 5646484.45 frames. ], batch size: 186, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:00:59,272 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 04:01:07,815 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 04:01:12,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=510199.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:01:54,081 INFO [train.py:968] (1/2) Epoch 12, batch 9050, giga_loss[loss=0.3128, simple_loss=0.376, pruned_loss=0.1248, over 28612.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.377, pruned_loss=0.1271, over 5667341.20 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3543, pruned_loss=0.09899, over 5691437.15 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3806, pruned_loss=0.1309, over 5651729.42 frames. ], batch size: 307, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:02:03,999 INFO [optim.py:369] (1/2) 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 04:02:41,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 04:02:42,898 INFO [train.py:968] (1/2) Epoch 12, batch 9100, giga_loss[loss=0.3696, simple_loss=0.4151, pruned_loss=0.162, over 28786.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3777, pruned_loss=0.1283, over 5657329.23 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3539, pruned_loss=0.0986, over 5697736.51 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3816, pruned_loss=0.1325, over 5638458.14 frames. ], batch size: 243, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:03:32,530 INFO [train.py:968] (1/2) Epoch 12, batch 9150, giga_loss[loss=0.307, simple_loss=0.3746, pruned_loss=0.1197, over 28950.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3771, pruned_loss=0.1289, over 5663072.08 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.354, pruned_loss=0.09858, over 5698388.83 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3802, pruned_loss=0.1323, over 5647577.78 frames. ], batch size: 227, lr: 2.73e-03, grad_scale: 2.0
+2023-03-06 04:03:42,387 INFO [optim.py:369] (1/2) 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 04:03:50,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4431, 1.5938, 1.5067, 1.4463], device='cuda:1'), covar=tensor([0.1404, 0.1757, 0.2060, 0.1739], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0737, 0.0674, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 04:04:02,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2081, 1.8240, 1.3986, 0.3700], device='cuda:1'), covar=tensor([0.3013, 0.1902, 0.3055, 0.4065], device='cuda:1'), in_proj_covar=tensor([0.1567, 0.1498, 0.1487, 0.1298], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 04:04:23,228 INFO [train.py:968] (1/2) Epoch 12, batch 9200, giga_loss[loss=0.3321, simple_loss=0.366, pruned_loss=0.1491, over 23484.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3751, pruned_loss=0.1278, over 5657462.17 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.354, pruned_loss=0.09847, over 5702264.90 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3779, pruned_loss=0.1311, over 5641156.09 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:04:25,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-06 04:04:39,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2677, 3.0829, 2.8898, 1.3647], device='cuda:1'), covar=tensor([0.0913, 0.1025, 0.0978, 0.2354], device='cuda:1'), in_proj_covar=tensor([0.1083, 0.1014, 0.0886, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 04:04:40,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=510414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:05:05,916 INFO [train.py:968] (1/2) Epoch 12, batch 9250, libri_loss[loss=0.2893, simple_loss=0.3687, pruned_loss=0.105, over 29482.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3745, pruned_loss=0.1261, over 5666112.40 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3542, pruned_loss=0.09863, over 5704981.77 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.377, pruned_loss=0.1292, over 5649746.92 frames. ], batch size: 85, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:05:17,942 INFO [optim.py:369] (1/2) 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 04:05:58,194 INFO [train.py:968] (1/2) Epoch 12, batch 9300, giga_loss[loss=0.2966, simple_loss=0.369, pruned_loss=0.1121, over 28901.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3769, pruned_loss=0.1271, over 5668126.13 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3539, pruned_loss=0.09837, over 5708878.98 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3796, pruned_loss=0.1304, over 5650948.51 frames. ], batch size: 145, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:06:41,765 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-06 04:06:42,987 INFO [train.py:968] (1/2) Epoch 12, batch 9350, giga_loss[loss=0.3235, simple_loss=0.3733, pruned_loss=0.1369, over 28837.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3781, pruned_loss=0.1283, over 5652605.42 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3541, pruned_loss=0.09867, over 5702968.46 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3809, pruned_loss=0.1317, over 5642779.50 frames. ], batch size: 119, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:06:51,866 INFO [optim.py:369] (1/2) 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 04:06:56,504 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=510557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:06:58,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=510560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:07:14,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=510574.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:07:29,140 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=510589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:07:31,084 INFO [train.py:968] (1/2) Epoch 12, batch 9400, libri_loss[loss=0.3046, simple_loss=0.3831, pruned_loss=0.1131, over 27902.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.379, pruned_loss=0.1282, over 5656196.85 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3541, pruned_loss=0.0987, over 5704404.38 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3815, pruned_loss=0.1312, over 5646740.84 frames. ], batch size: 116, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:07:46,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4952, 1.7034, 1.3848, 1.6007], device='cuda:1'), covar=tensor([0.2393, 0.2306, 0.2533, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.1306, 0.0972, 0.1156, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 04:08:14,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 04:08:16,926 INFO [train.py:968] (1/2) Epoch 12, batch 9450, giga_loss[loss=0.3448, simple_loss=0.4069, pruned_loss=0.1414, over 28723.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3781, pruned_loss=0.1248, over 5659764.75 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.354, pruned_loss=0.09859, over 5699919.08 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3809, pruned_loss=0.1281, over 5655419.16 frames. ], batch size: 262, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:08:26,398 INFO [optim.py:369] (1/2) 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 04:08:58,364 INFO [train.py:968] (1/2) Epoch 12, batch 9500, giga_loss[loss=0.3148, simple_loss=0.3851, pruned_loss=0.1223, over 28972.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3804, pruned_loss=0.1245, over 5662535.77 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3539, pruned_loss=0.0985, over 5689815.31 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3831, pruned_loss=0.1276, over 5667851.35 frames. ], batch size: 213, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:09:25,846 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=510717.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:09:29,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=510720.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:09:49,651 INFO [train.py:968] (1/2) Epoch 12, batch 9550, giga_loss[loss=0.29, simple_loss=0.3583, pruned_loss=0.1108, over 28820.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3835, pruned_loss=0.1273, over 5656752.66 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3536, pruned_loss=0.09835, over 5690388.86 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3865, pruned_loss=0.1304, over 5660272.65 frames. ], batch size: 119, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:09:57,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=510749.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:09:59,074 INFO [optim.py:369] (1/2) 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 04:10:15,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=510767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:10:33,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0521, 1.4741, 1.3372, 1.2914], device='cuda:1'), covar=tensor([0.1747, 0.1436, 0.1860, 0.1530], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0726, 0.0664, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 04:10:33,415 INFO [train.py:968] (1/2) Epoch 12, batch 9600, libri_loss[loss=0.2961, simple_loss=0.3735, pruned_loss=0.1094, over 29737.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.384, pruned_loss=0.1282, over 5674691.38 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.353, pruned_loss=0.09802, over 5699701.05 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3881, pruned_loss=0.1323, over 5667950.70 frames. ], batch size: 87, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:11:19,984 INFO [train.py:968] (1/2) Epoch 12, batch 9650, giga_loss[loss=0.3371, simple_loss=0.3946, pruned_loss=0.1398, over 28962.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3862, pruned_loss=0.1314, over 5661958.03 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3528, pruned_loss=0.09796, over 5701414.87 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3905, pruned_loss=0.1354, over 5654597.47 frames. ], batch size: 164, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:11:30,259 INFO [optim.py:369] (1/2) 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 04:11:50,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7358, 3.5532, 3.3997, 1.5594], device='cuda:1'), covar=tensor([0.0734, 0.0833, 0.0824, 0.2293], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.1006, 0.0880, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 04:12:10,359 INFO [train.py:968] (1/2) Epoch 12, batch 9700, giga_loss[loss=0.398, simple_loss=0.4279, pruned_loss=0.1841, over 27516.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3863, pruned_loss=0.1317, over 5661198.15 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3529, pruned_loss=0.09797, over 5699916.46 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3898, pruned_loss=0.1351, over 5656209.72 frames. ], batch size: 472, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:12:21,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5970, 1.7376, 1.8674, 1.4052], device='cuda:1'), covar=tensor([0.1689, 0.2355, 0.1363, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0700, 0.0877, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 04:12:54,305 INFO [train.py:968] (1/2) Epoch 12, batch 9750, giga_loss[loss=0.3478, simple_loss=0.3812, pruned_loss=0.1572, over 23485.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3852, pruned_loss=0.1296, over 5670447.49 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3532, pruned_loss=0.09831, over 5705249.06 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3885, pruned_loss=0.1328, over 5660841.54 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:13:02,672 INFO [optim.py:369] (1/2) 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 04:13:19,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6089, 1.6501, 1.3202, 1.2905], device='cuda:1'), covar=tensor([0.0760, 0.0555, 0.0934, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0441, 0.0498, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 04:13:37,783 INFO [train.py:968] (1/2) Epoch 12, batch 9800, giga_loss[loss=0.3719, simple_loss=0.4088, pruned_loss=0.1676, over 26703.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3842, pruned_loss=0.1274, over 5675685.87 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.353, pruned_loss=0.09823, over 5708615.46 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3877, pruned_loss=0.1307, over 5664438.36 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:13:42,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5123, 1.9998, 1.8097, 1.3605], device='cuda:1'), covar=tensor([0.1766, 0.2415, 0.1469, 0.1747], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0699, 0.0879, 0.0779], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 04:13:49,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3999, 2.0571, 1.4769, 0.6475], device='cuda:1'), covar=tensor([0.3994, 0.2055, 0.3107, 0.4559], device='cuda:1'), in_proj_covar=tensor([0.1584, 0.1504, 0.1503, 0.1305], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 04:14:25,548 INFO [train.py:968] (1/2) Epoch 12, batch 9850, giga_loss[loss=0.3282, simple_loss=0.3937, pruned_loss=0.1313, over 28938.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3844, pruned_loss=0.1272, over 5675441.92 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3527, pruned_loss=0.09806, over 5709062.86 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3881, pruned_loss=0.1306, over 5665231.34 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:14:37,428 INFO [optim.py:369] (1/2) 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 04:15:14,169 INFO [train.py:968] (1/2) Epoch 12, batch 9900, giga_loss[loss=0.3397, simple_loss=0.3966, pruned_loss=0.1414, over 28793.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3859, pruned_loss=0.1295, over 5670797.98 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3527, pruned_loss=0.09823, over 5715261.36 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3897, pruned_loss=0.1328, over 5656115.76 frames. ], batch size: 119, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:15:21,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:16:01,142 INFO [train.py:968] (1/2) Epoch 12, batch 9950, giga_loss[loss=0.3087, simple_loss=0.375, pruned_loss=0.1212, over 28920.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3836, pruned_loss=0.1284, over 5661417.98 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3526, pruned_loss=0.0981, over 5712942.19 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3874, pruned_loss=0.1319, over 5650904.65 frames. ], batch size: 213, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:16:01,368 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:16:54,986 INFO [train.py:968] (1/2) Epoch 12, batch 10000, giga_loss[loss=0.3133, simple_loss=0.3718, pruned_loss=0.1274, over 28675.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3823, pruned_loss=0.1291, over 5659317.33 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3523, pruned_loss=0.09791, over 5715591.74 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.386, pruned_loss=0.1324, over 5647891.70 frames. ], batch size: 307, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:17:42,019 INFO [train.py:968] (1/2) Epoch 12, batch 10050, giga_loss[loss=0.3292, simple_loss=0.3784, pruned_loss=0.14, over 28974.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3797, pruned_loss=0.1279, over 5660371.25 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3525, pruned_loss=0.09793, over 5709815.53 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3829, pruned_loss=0.131, over 5654935.19 frames. ], batch size: 200, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:17:46,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2632, 1.3594, 1.2506, 1.2168], device='cuda:1'), covar=tensor([0.1453, 0.1345, 0.1100, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1652, 0.1617, 0.1710], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 04:17:53,973 INFO [optim.py:369] (1/2) 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 04:18:30,997 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=511285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:18:34,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=511288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:18:38,602 INFO [train.py:968] (1/2) Epoch 12, batch 10100, giga_loss[loss=0.282, simple_loss=0.3499, pruned_loss=0.107, over 29018.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.379, pruned_loss=0.1284, over 5646704.45 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3529, pruned_loss=0.0981, over 5708952.49 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3816, pruned_loss=0.1311, over 5642611.33 frames. ], batch size: 128, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:19:01,393 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 10150, giga_loss[loss=0.3207, simple_loss=0.3815, pruned_loss=0.1299, over 28639.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3784, pruned_loss=0.1283, over 5662320.15 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3533, pruned_loss=0.09835, over 5712798.45 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3807, pruned_loss=0.1309, over 5654497.99 frames. ], batch size: 262, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:19:34,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9390, 3.7613, 3.5396, 1.5936], device='cuda:1'), covar=tensor([0.0647, 0.0758, 0.0734, 0.2339], device='cuda:1'), in_proj_covar=tensor([0.1076, 0.1011, 0.0883, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 04:19:36,841 INFO [optim.py:369] (1/2) 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 04:20:11,646 INFO [train.py:968] (1/2) Epoch 12, batch 10200, giga_loss[loss=0.263, simple_loss=0.3476, pruned_loss=0.08918, over 28961.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1251, over 5661911.83 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3531, pruned_loss=0.09813, over 5715865.17 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3777, pruned_loss=0.1279, over 5652062.71 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:20:56,301 INFO [train.py:968] (1/2) Epoch 12, batch 10250, giga_loss[loss=0.3, simple_loss=0.369, pruned_loss=0.1155, over 28888.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3708, pruned_loss=0.1203, over 5659250.69 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3528, pruned_loss=0.09796, over 5718282.04 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3736, pruned_loss=0.1235, over 5647339.18 frames. ], batch size: 213, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:21:07,786 INFO [optim.py:369] (1/2) 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 04:21:14,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6731, 1.5047, 5.1777, 3.6429], device='cuda:1'), covar=tensor([0.1572, 0.2592, 0.0307, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0668, 0.0591, 0.0867, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 04:21:30,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=511475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:21:42,438 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:21:45,213 INFO [train.py:968] (1/2) Epoch 12, batch 10300, giga_loss[loss=0.3547, simple_loss=0.4031, pruned_loss=0.1532, over 27536.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3702, pruned_loss=0.119, over 5667825.99 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3536, pruned_loss=0.09835, over 5722033.39 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1217, over 5653635.24 frames. ], batch size: 472, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:22:11,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 04:22:29,290 INFO [train.py:968] (1/2) Epoch 12, batch 10350, giga_loss[loss=0.2736, simple_loss=0.3346, pruned_loss=0.1063, over 28939.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3688, pruned_loss=0.1183, over 5675009.04 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3528, pruned_loss=0.09782, over 5726900.20 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3717, pruned_loss=0.1218, over 5657278.88 frames. ], batch size: 112, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:22:39,196 INFO [optim.py:369] (1/2) 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 04:22:39,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2432, 3.0457, 2.3638, 1.9010], device='cuda:1'), covar=tensor([0.2047, 0.0972, 0.1256, 0.1626], device='cuda:1'), in_proj_covar=tensor([0.1737, 0.1651, 0.1609, 0.1708], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 04:23:07,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1573, 1.2407, 1.0590, 0.9107], device='cuda:1'), covar=tensor([0.0869, 0.0508, 0.1089, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0440, 0.0499, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 04:23:14,587 INFO [train.py:968] (1/2) Epoch 12, batch 10400, giga_loss[loss=0.3005, simple_loss=0.36, pruned_loss=0.1205, over 27997.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3658, pruned_loss=0.1171, over 5680502.07 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3527, pruned_loss=0.09777, over 5735088.02 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1209, over 5656448.56 frames. ], batch size: 412, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:23:42,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=511618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:23:44,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=511621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:24:04,239 INFO [train.py:968] (1/2) Epoch 12, batch 10450, giga_loss[loss=0.3595, simple_loss=0.4002, pruned_loss=0.1593, over 28840.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3659, pruned_loss=0.118, over 5670391.84 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3529, pruned_loss=0.09805, over 5726679.30 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3683, pruned_loss=0.1209, over 5658418.61 frames. ], batch size: 243, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:24:11,615 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:24:19,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:24:48,803 INFO [train.py:968] (1/2) Epoch 12, batch 10500, giga_loss[loss=0.3867, simple_loss=0.4204, pruned_loss=0.1766, over 26750.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3684, pruned_loss=0.1188, over 5667256.98 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.353, pruned_loss=0.09791, over 5726321.98 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3706, pruned_loss=0.1218, over 5656461.17 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:25:35,027 INFO [train.py:968] (1/2) Epoch 12, batch 10550, giga_loss[loss=0.3059, simple_loss=0.3682, pruned_loss=0.1218, over 28954.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3712, pruned_loss=0.1206, over 5659927.36 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3536, pruned_loss=0.09815, over 5723917.87 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3729, pruned_loss=0.1233, over 5651742.13 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:25:47,781 INFO [optim.py:369] (1/2) 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 04:26:24,293 INFO [train.py:968] (1/2) Epoch 12, batch 10600, giga_loss[loss=0.2968, simple_loss=0.3626, pruned_loss=0.1155, over 28768.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3704, pruned_loss=0.1204, over 5658937.88 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3538, pruned_loss=0.09823, over 5727889.89 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3721, pruned_loss=0.1232, over 5646699.32 frames. ], batch size: 119, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:27:09,149 INFO [train.py:968] (1/2) Epoch 12, batch 10650, giga_loss[loss=0.3416, simple_loss=0.3925, pruned_loss=0.1453, over 28982.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5660370.81 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3538, pruned_loss=0.09812, over 5731749.96 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 5645517.92 frames. ], batch size: 213, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:27:21,010 INFO [optim.py:369] (1/2) 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 04:27:24,254 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=511859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:27:27,135 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 10700, giga_loss[loss=0.3012, simple_loss=0.3654, pruned_loss=0.1184, over 28187.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1231, over 5668237.22 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3545, pruned_loss=0.0986, over 5735054.19 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3747, pruned_loss=0.1258, over 5650854.56 frames. ], batch size: 77, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:28:45,157 INFO [train.py:968] (1/2) Epoch 12, batch 10750, giga_loss[loss=0.3167, simple_loss=0.3775, pruned_loss=0.128, over 28758.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3745, pruned_loss=0.1233, over 5667032.14 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3548, pruned_loss=0.09863, over 5735031.40 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3759, pruned_loss=0.1261, over 5651225.92 frames. ], batch size: 284, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:28:56,672 INFO [optim.py:369] (1/2) 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 04:29:27,921 INFO [train.py:968] (1/2) Epoch 12, batch 10800, giga_loss[loss=0.2827, simple_loss=0.357, pruned_loss=0.1042, over 28800.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.376, pruned_loss=0.1246, over 5675916.46 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3548, pruned_loss=0.09866, over 5739999.22 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3776, pruned_loss=0.1276, over 5656969.28 frames. ], batch size: 99, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:29:42,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:29:44,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:29:53,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2110, 3.0627, 2.4777, 2.0880], device='cuda:1'), covar=tensor([0.2211, 0.1038, 0.1179, 0.1446], device='cuda:1'), in_proj_covar=tensor([0.1746, 0.1662, 0.1615, 0.1717], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 04:29:55,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3932, 1.7105, 1.6800, 1.4672], device='cuda:1'), covar=tensor([0.1505, 0.1451, 0.1723, 0.1622], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0731, 0.0673, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 04:29:59,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3828, 3.1812, 3.0125, 1.8504], device='cuda:1'), covar=tensor([0.0785, 0.0937, 0.0888, 0.1891], device='cuda:1'), in_proj_covar=tensor([0.1074, 0.1006, 0.0879, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 04:30:07,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:30:10,901 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 10850, libri_loss[loss=0.2636, simple_loss=0.3506, pruned_loss=0.08833, over 29537.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3763, pruned_loss=0.125, over 5682277.93 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3546, pruned_loss=0.09825, over 5744544.04 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3787, pruned_loss=0.1288, over 5660370.98 frames. ], batch size: 81, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:30:20,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=512048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:30:29,904 INFO [optim.py:369] (1/2) 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 04:31:00,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5462, 1.4653, 1.1617, 1.1749], device='cuda:1'), covar=tensor([0.0753, 0.0559, 0.1023, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0442, 0.0501, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 04:31:03,032 INFO [train.py:968] (1/2) Epoch 12, batch 10900, giga_loss[loss=0.3659, simple_loss=0.41, pruned_loss=0.1609, over 27630.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3776, pruned_loss=0.1253, over 5675713.16 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3543, pruned_loss=0.09815, over 5743888.30 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3801, pruned_loss=0.1289, over 5657968.69 frames. ], batch size: 472, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:31:52,690 INFO [train.py:968] (1/2) Epoch 12, batch 10950, giga_loss[loss=0.3934, simple_loss=0.4225, pruned_loss=0.1821, over 26639.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3776, pruned_loss=0.1244, over 5671796.55 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3544, pruned_loss=0.0982, over 5744569.81 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3802, pruned_loss=0.128, over 5655341.06 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:32:05,551 INFO [optim.py:369] (1/2) 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 04:32:27,598 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:32:30,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:32:42,764 INFO [train.py:968] (1/2) Epoch 12, batch 11000, libri_loss[loss=0.2358, simple_loss=0.3087, pruned_loss=0.08147, over 29533.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3779, pruned_loss=0.1259, over 5667609.71 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3537, pruned_loss=0.09806, over 5749992.64 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3814, pruned_loss=0.1298, over 5646855.85 frames. ], batch size: 70, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:32:58,336 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:33:18,611 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=512227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:33:25,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:33:35,736 INFO [train.py:968] (1/2) Epoch 12, batch 11050, giga_loss[loss=0.3622, simple_loss=0.4119, pruned_loss=0.1562, over 28030.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3781, pruned_loss=0.1271, over 5650835.21 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3535, pruned_loss=0.0979, over 5750803.79 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3816, pruned_loss=0.1309, over 5632019.74 frames. ], batch size: 77, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:33:51,706 INFO [optim.py:369] (1/2) 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 04:34:31,044 INFO [train.py:968] (1/2) Epoch 12, batch 11100, giga_loss[loss=0.267, simple_loss=0.3414, pruned_loss=0.09635, over 28807.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3763, pruned_loss=0.1262, over 5654824.81 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3538, pruned_loss=0.09816, over 5752499.18 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3789, pruned_loss=0.1292, over 5637775.26 frames. ], batch size: 119, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:35:00,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2247, 1.5663, 1.1911, 0.6911], device='cuda:1'), covar=tensor([0.2061, 0.1426, 0.1688, 0.3069], device='cuda:1'), in_proj_covar=tensor([0.1564, 0.1487, 0.1490, 0.1284], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 04:35:16,532 INFO [train.py:968] (1/2) Epoch 12, batch 11150, giga_loss[loss=0.3638, simple_loss=0.4055, pruned_loss=0.1611, over 28599.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.375, pruned_loss=0.1258, over 5655005.96 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3535, pruned_loss=0.09806, over 5753971.69 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.378, pruned_loss=0.1291, over 5637585.66 frames. ], batch size: 307, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:35:26,894 INFO [optim.py:369] (1/2) 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 04:35:48,828 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:35:51,262 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 11200, giga_loss[loss=0.3456, simple_loss=0.3742, pruned_loss=0.1585, over 23588.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3763, pruned_loss=0.127, over 5664097.17 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3539, pruned_loss=0.09823, over 5755996.33 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3786, pruned_loss=0.1297, over 5647315.36 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:36:18,592 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:36:35,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:36:37,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 04:36:55,527 INFO [train.py:968] (1/2) Epoch 12, batch 11250, giga_loss[loss=0.2988, simple_loss=0.3608, pruned_loss=0.1184, over 29185.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3753, pruned_loss=0.1269, over 5657625.82 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3539, pruned_loss=0.09819, over 5757625.61 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3774, pruned_loss=0.1293, over 5642401.18 frames. ], batch size: 113, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:37:02,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5077, 3.8192, 1.5751, 1.8114], device='cuda:1'), covar=tensor([0.0907, 0.0295, 0.0834, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0516, 0.0344, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 04:37:07,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7331, 4.5849, 4.2949, 1.9203], device='cuda:1'), covar=tensor([0.0511, 0.0635, 0.0734, 0.2076], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.1008, 0.0882, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 04:37:09,682 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 11300, giga_loss[loss=0.2805, simple_loss=0.3484, pruned_loss=0.1063, over 28924.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3745, pruned_loss=0.1264, over 5652587.83 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3534, pruned_loss=0.09795, over 5751102.62 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3771, pruned_loss=0.1294, over 5643226.18 frames. ], batch size: 112, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:38:27,468 INFO [train.py:968] (1/2) Epoch 12, batch 11350, giga_loss[loss=0.3193, simple_loss=0.3798, pruned_loss=0.1294, over 28884.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3776, pruned_loss=0.1287, over 5660573.48 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3535, pruned_loss=0.09801, over 5753237.39 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3799, pruned_loss=0.1314, over 5649917.81 frames. ], batch size: 186, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:38:41,017 INFO [optim.py:369] (1/2) 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 04:38:48,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:38:51,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:39:14,185 INFO [train.py:968] (1/2) Epoch 12, batch 11400, giga_loss[loss=0.2703, simple_loss=0.3453, pruned_loss=0.09762, over 28858.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3775, pruned_loss=0.1291, over 5652627.93 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3537, pruned_loss=0.09812, over 5759689.89 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3802, pruned_loss=0.1325, over 5634240.53 frames. ], batch size: 174, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:39:19,923 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:39:22,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=512602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:40:01,546 INFO [train.py:968] (1/2) Epoch 12, batch 11450, libri_loss[loss=0.2974, simple_loss=0.3711, pruned_loss=0.1119, over 25935.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3773, pruned_loss=0.1289, over 5663730.48 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3538, pruned_loss=0.09827, over 5759624.44 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3802, pruned_loss=0.1327, over 5645557.98 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:40:04,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3014, 4.0929, 3.8475, 1.8404], device='cuda:1'), covar=tensor([0.0638, 0.0826, 0.1011, 0.2081], device='cuda:1'), in_proj_covar=tensor([0.1080, 0.1015, 0.0886, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 04:40:14,879 INFO [optim.py:369] (1/2) 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 04:40:45,525 INFO [train.py:968] (1/2) Epoch 12, batch 11500, giga_loss[loss=0.3089, simple_loss=0.3649, pruned_loss=0.1265, over 28825.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.377, pruned_loss=0.1283, over 5669014.14 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3538, pruned_loss=0.09825, over 5763359.15 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3798, pruned_loss=0.132, over 5649013.47 frames. ], batch size: 119, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:41:21,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5457, 1.6703, 1.7937, 1.3458], device='cuda:1'), covar=tensor([0.1594, 0.2255, 0.1258, 0.1500], device='cuda:1'), in_proj_covar=tensor([0.0839, 0.0706, 0.0881, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 04:41:33,522 INFO [train.py:968] (1/2) Epoch 12, batch 11550, giga_loss[loss=0.344, simple_loss=0.3949, pruned_loss=0.1466, over 28001.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.378, pruned_loss=0.1288, over 5667597.90 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3538, pruned_loss=0.09845, over 5766852.48 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3813, pruned_loss=0.1329, over 5644203.01 frames. ], batch size: 412, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:41:35,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=512745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:41:37,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=512748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:41:45,751 INFO [optim.py:369] (1/2) 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 04:41:48,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1830, 5.9844, 5.6621, 3.2213], device='cuda:1'), covar=tensor([0.0400, 0.0527, 0.0676, 0.1499], device='cuda:1'), in_proj_covar=tensor([0.1079, 0.1012, 0.0882, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 04:42:04,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=512777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:42:15,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7239, 1.6278, 1.3258, 1.3760], device='cuda:1'), covar=tensor([0.0676, 0.0569, 0.0879, 0.0977], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0441, 0.0501, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 04:42:16,871 INFO [train.py:968] (1/2) Epoch 12, batch 11600, libri_loss[loss=0.2848, simple_loss=0.3667, pruned_loss=0.1014, over 25885.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.377, pruned_loss=0.1269, over 5683114.26 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3541, pruned_loss=0.09863, over 5768679.03 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3804, pruned_loss=0.1312, over 5658884.60 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:42:54,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3973, 1.5497, 1.3198, 1.5024], device='cuda:1'), covar=tensor([0.0636, 0.0405, 0.0305, 0.0656], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0089], device='cuda:1')
+2023-03-06 04:43:04,079 INFO [train.py:968] (1/2) Epoch 12, batch 11650, giga_loss[loss=0.3651, simple_loss=0.4078, pruned_loss=0.1612, over 27445.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3796, pruned_loss=0.1292, over 5668458.83 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.0989, over 5770612.55 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3829, pruned_loss=0.1336, over 5643593.55 frames. ], batch size: 472, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:43:18,413 INFO [optim.py:369] (1/2) 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 04:43:52,539 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=512890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:43:53,549 INFO [train.py:968] (1/2) Epoch 12, batch 11700, giga_loss[loss=0.3601, simple_loss=0.401, pruned_loss=0.1596, over 29055.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3827, pruned_loss=0.132, over 5668534.56 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3546, pruned_loss=0.09902, over 5772224.97 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3857, pruned_loss=0.136, over 5645519.44 frames. ], batch size: 128, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:44:40,274 INFO [train.py:968] (1/2) Epoch 12, batch 11750, giga_loss[loss=0.3172, simple_loss=0.3828, pruned_loss=0.1258, over 28701.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3824, pruned_loss=0.1324, over 5667768.17 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3543, pruned_loss=0.09895, over 5776154.96 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3858, pruned_loss=0.1366, over 5643263.60 frames. ], batch size: 92, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:44:54,906 INFO [optim.py:369] (1/2) 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 04:45:26,886 INFO [train.py:968] (1/2) Epoch 12, batch 11800, libri_loss[loss=0.2949, simple_loss=0.3758, pruned_loss=0.107, over 29536.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3835, pruned_loss=0.1322, over 5667862.20 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3544, pruned_loss=0.0989, over 5778527.57 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3869, pruned_loss=0.1364, over 5643427.06 frames. ], batch size: 84, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:45:40,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4576, 2.0438, 1.5684, 1.7027], device='cuda:1'), covar=tensor([0.0738, 0.0260, 0.0298, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0059, 0.0052, 0.0089], device='cuda:1')
+2023-03-06 04:45:47,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-06 04:46:02,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-06 04:46:13,421 INFO [train.py:968] (1/2) Epoch 12, batch 11850, giga_loss[loss=0.2816, simple_loss=0.351, pruned_loss=0.1061, over 28666.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3821, pruned_loss=0.1301, over 5669523.94 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3543, pruned_loss=0.09883, over 5778534.74 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3858, pruned_loss=0.1345, over 5646817.31 frames. ], batch size: 92, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:46:29,135 INFO [optim.py:369] (1/2) 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 04:46:55,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-06 04:46:59,298 INFO [train.py:968] (1/2) Epoch 12, batch 11900, giga_loss[loss=0.3312, simple_loss=0.3929, pruned_loss=0.1348, over 28772.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3815, pruned_loss=0.1295, over 5661029.29 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.0989, over 5769986.90 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.385, pruned_loss=0.1337, over 5647178.01 frames. ], batch size: 119, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:47:45,574 INFO [train.py:968] (1/2) Epoch 12, batch 11950, libri_loss[loss=0.3048, simple_loss=0.3821, pruned_loss=0.1137, over 25781.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3786, pruned_loss=0.1275, over 5656091.23 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3548, pruned_loss=0.09906, over 5765069.43 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3815, pruned_loss=0.1312, over 5647530.31 frames. ], batch size: 136, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:47:59,545 INFO [optim.py:369] (1/2) 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 04:48:34,059 INFO [train.py:968] (1/2) Epoch 12, batch 12000, libri_loss[loss=0.2839, simple_loss=0.3727, pruned_loss=0.09752, over 29276.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3787, pruned_loss=0.1275, over 5663720.50 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3541, pruned_loss=0.09856, over 5767767.01 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3825, pruned_loss=0.1319, over 5650977.74 frames. ], batch size: 94, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:48:34,059 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 04:48:42,400 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 04:48:58,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8100, 4.6104, 4.3431, 2.0193], device='cuda:1'), covar=tensor([0.0441, 0.0613, 0.0630, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.1089, 0.1015, 0.0890, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 04:49:03,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5047, 2.2666, 1.7763, 1.8573], device='cuda:1'), covar=tensor([0.0714, 0.0666, 0.0940, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0446, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 04:49:21,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4472, 1.7462, 1.3861, 1.5622], device='cuda:1'), covar=tensor([0.2276, 0.2233, 0.2455, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.1318, 0.0981, 0.1163, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 04:49:28,441 INFO [train.py:968] (1/2) Epoch 12, batch 12050, giga_loss[loss=0.3318, simple_loss=0.3977, pruned_loss=0.133, over 28925.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3796, pruned_loss=0.1281, over 5658641.56 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3541, pruned_loss=0.0985, over 5767089.50 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3832, pruned_loss=0.1324, over 5646537.06 frames. ], batch size: 174, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:49:42,900 INFO [optim.py:369] (1/2) 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 04:49:49,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=513265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:49:55,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3986, 1.5669, 1.4721, 1.3650], device='cuda:1'), covar=tensor([0.1647, 0.1591, 0.1041, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.1737, 0.1659, 0.1605, 0.1719], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 04:50:04,023 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=513279.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:50:16,576 INFO [train.py:968] (1/2) Epoch 12, batch 12100, giga_loss[loss=0.3148, simple_loss=0.374, pruned_loss=0.1278, over 28317.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3791, pruned_loss=0.1287, over 5660586.53 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3545, pruned_loss=0.09881, over 5759342.72 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.382, pruned_loss=0.1323, over 5656117.44 frames. ], batch size: 368, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:51:03,610 INFO [train.py:968] (1/2) Epoch 12, batch 12150, libri_loss[loss=0.2605, simple_loss=0.3375, pruned_loss=0.09173, over 29570.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3785, pruned_loss=0.1284, over 5660772.76 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3546, pruned_loss=0.0988, over 5753529.13 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3815, pruned_loss=0.1321, over 5658562.52 frames. ], batch size: 77, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:51:15,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2233, 1.2751, 1.2497, 1.3993], device='cuda:1'), covar=tensor([0.0751, 0.0329, 0.0314, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 04:51:20,142 INFO [optim.py:369] (1/2) 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 04:51:41,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=513379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:51:54,233 INFO [train.py:968] (1/2) Epoch 12, batch 12200, giga_loss[loss=0.3242, simple_loss=0.389, pruned_loss=0.1297, over 28919.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3808, pruned_loss=0.1301, over 5664610.95 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3547, pruned_loss=0.09883, over 5755058.66 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3834, pruned_loss=0.1334, over 5660476.57 frames. ], batch size: 164, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:52:10,303 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=513408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:52:12,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=513411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:52:38,882 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=513440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:52:41,691 INFO [train.py:968] (1/2) Epoch 12, batch 12250, giga_loss[loss=0.3901, simple_loss=0.4102, pruned_loss=0.185, over 23493.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3822, pruned_loss=0.1312, over 5653828.53 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3553, pruned_loss=0.09926, over 5750670.87 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3844, pruned_loss=0.1344, over 5651218.73 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:52:56,669 INFO [optim.py:369] (1/2) 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 04:52:57,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4535, 1.6349, 1.5914, 1.4574], device='cuda:1'), covar=tensor([0.1535, 0.1733, 0.2011, 0.1808], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0738, 0.0679, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 04:53:30,685 INFO [train.py:968] (1/2) Epoch 12, batch 12300, libri_loss[loss=0.3067, simple_loss=0.3818, pruned_loss=0.1158, over 29539.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3812, pruned_loss=0.1297, over 5673874.73 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3554, pruned_loss=0.09927, over 5754024.93 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3834, pruned_loss=0.1328, over 5667151.91 frames. ], batch size: 83, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:53:50,287 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-06 04:54:02,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0150, 1.2585, 1.0453, 0.2042], device='cuda:1'), covar=tensor([0.2203, 0.1817, 0.2430, 0.3904], device='cuda:1'), in_proj_covar=tensor([0.1578, 0.1505, 0.1504, 0.1294], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 04:54:14,785 INFO [train.py:968] (1/2) Epoch 12, batch 12350, giga_loss[loss=0.296, simple_loss=0.368, pruned_loss=0.112, over 28866.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3816, pruned_loss=0.1296, over 5666895.66 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3561, pruned_loss=0.09964, over 5758317.38 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3836, pruned_loss=0.1329, over 5654191.26 frames. ], batch size: 174, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:54:33,020 INFO [optim.py:369] (1/2) 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 04:54:37,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6910, 1.0728, 2.8243, 2.6550], device='cuda:1'), covar=tensor([0.1705, 0.2430, 0.0579, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0593, 0.0875, 0.0779], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 04:55:02,201 INFO [train.py:968] (1/2) Epoch 12, batch 12400, giga_loss[loss=0.3446, simple_loss=0.4012, pruned_loss=0.144, over 28988.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3808, pruned_loss=0.1285, over 5669886.95 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3562, pruned_loss=0.09975, over 5751364.28 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3827, pruned_loss=0.1313, over 5665482.78 frames. ], batch size: 213, lr: 2.72e-03, grad_scale: 8.0
+2023-03-06 04:55:48,804 INFO [train.py:968] (1/2) Epoch 12, batch 12450, giga_loss[loss=0.3488, simple_loss=0.3826, pruned_loss=0.1574, over 23782.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3793, pruned_loss=0.127, over 5680645.82 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3565, pruned_loss=0.09982, over 5755716.55 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3813, pruned_loss=0.1303, over 5670380.03 frames. ], batch size: 705, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 04:55:53,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1774, 1.5931, 1.5156, 1.0768], device='cuda:1'), covar=tensor([0.1487, 0.2312, 0.1256, 0.1470], device='cuda:1'), in_proj_covar=tensor([0.0837, 0.0704, 0.0880, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 04:56:01,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=513654.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:56:06,980 INFO [optim.py:369] (1/2) 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 04:56:31,812 INFO [train.py:968] (1/2) Epoch 12, batch 12500, giga_loss[loss=0.2926, simple_loss=0.3615, pruned_loss=0.1118, over 28806.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 5681538.52 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3562, pruned_loss=0.09974, over 5759941.13 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.38, pruned_loss=0.1297, over 5667097.75 frames. ], batch size: 243, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:56:50,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2452, 1.6330, 1.2623, 0.7156], device='cuda:1'), covar=tensor([0.3297, 0.2078, 0.1899, 0.3923], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1512, 0.1509, 0.1303], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 04:57:21,661 INFO [train.py:968] (1/2) Epoch 12, batch 12550, giga_loss[loss=0.2839, simple_loss=0.3492, pruned_loss=0.1093, over 28837.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3754, pruned_loss=0.1259, over 5668442.41 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3563, pruned_loss=0.09992, over 5752327.10 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3776, pruned_loss=0.1288, over 5662538.07 frames. ], batch size: 186, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:57:32,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=513754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:57:40,138 INFO [optim.py:369] (1/2) 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 04:58:09,368 INFO [train.py:968] (1/2) Epoch 12, batch 12600, giga_loss[loss=0.4427, simple_loss=0.4503, pruned_loss=0.2175, over 26597.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1236, over 5682286.20 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3554, pruned_loss=0.09929, over 5757473.34 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3744, pruned_loss=0.1274, over 5670253.28 frames. ], batch size: 555, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:58:12,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=513797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 04:58:14,876 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=513800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:58:39,701 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 04:58:40,789 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=513829.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 04:58:51,618 INFO [train.py:968] (1/2) Epoch 12, batch 12650, giga_loss[loss=0.2808, simple_loss=0.3475, pruned_loss=0.1071, over 28674.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3691, pruned_loss=0.1225, over 5691062.76 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3547, pruned_loss=0.09887, over 5760760.62 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3728, pruned_loss=0.1268, over 5675995.56 frames. ], batch size: 242, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:59:06,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3083, 3.1168, 2.9565, 1.3852], device='cuda:1'), covar=tensor([0.0899, 0.1046, 0.0976, 0.2299], device='cuda:1'), in_proj_covar=tensor([0.1093, 0.1017, 0.0892, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 04:59:11,100 INFO [optim.py:369] (1/2) 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 04:59:11,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1477, 1.4796, 1.4187, 1.1574], device='cuda:1'), covar=tensor([0.2070, 0.1634, 0.1043, 0.1483], device='cuda:1'), in_proj_covar=tensor([0.1744, 0.1663, 0.1615, 0.1734], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 04:59:32,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 04:59:39,118 INFO [train.py:968] (1/2) Epoch 12, batch 12700, libri_loss[loss=0.2492, simple_loss=0.3293, pruned_loss=0.08454, over 28123.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3678, pruned_loss=0.1217, over 5699426.41 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3541, pruned_loss=0.09849, over 5764886.38 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3719, pruned_loss=0.1265, over 5680646.78 frames. ], batch size: 62, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 04:59:42,541 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=513897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 04:59:44,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=513900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:00:15,611 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 12750, giga_loss[loss=0.2952, simple_loss=0.3718, pruned_loss=0.1093, over 28737.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.1219, over 5695496.81 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3545, pruned_loss=0.0988, over 5767253.16 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3724, pruned_loss=0.1258, over 5677749.03 frames. ], batch size: 284, lr: 2.72e-03, grad_scale: 2.0
+2023-03-06 05:00:47,276 INFO [optim.py:369] (1/2) 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 05:01:00,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5044, 2.1754, 1.4753, 0.8187], device='cuda:1'), covar=tensor([0.3927, 0.2089, 0.3318, 0.4166], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1494, 0.1501, 0.1293], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 05:01:12,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-06 05:01:18,990 INFO [train.py:968] (1/2) Epoch 12, batch 12800, giga_loss[loss=0.2702, simple_loss=0.3508, pruned_loss=0.09476, over 28700.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3678, pruned_loss=0.1189, over 5688429.91 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3547, pruned_loss=0.09888, over 5767755.42 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3703, pruned_loss=0.1222, over 5672950.66 frames. ], batch size: 284, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:02:12,222 INFO [train.py:968] (1/2) Epoch 12, batch 12850, giga_loss[loss=0.268, simple_loss=0.3482, pruned_loss=0.09388, over 28657.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3658, pruned_loss=0.1161, over 5678171.37 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3548, pruned_loss=0.09893, over 5769366.73 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3678, pruned_loss=0.1189, over 5663338.21 frames. ], batch size: 242, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:02:12,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7251, 2.2764, 1.7690, 1.9448], device='cuda:1'), covar=tensor([0.0705, 0.0233, 0.0296, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 05:02:32,510 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 12900, giga_loss[loss=0.3049, simple_loss=0.3731, pruned_loss=0.1184, over 28466.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3629, pruned_loss=0.1134, over 5666553.59 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3549, pruned_loss=0.09918, over 5765426.68 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3645, pruned_loss=0.1155, over 5657723.71 frames. ], batch size: 336, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:03:44,208 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:03:56,170 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 12950, giga_loss[loss=0.2873, simple_loss=0.3624, pruned_loss=0.1061, over 28056.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.36, pruned_loss=0.1101, over 5672949.12 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3548, pruned_loss=0.09915, over 5767661.78 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3614, pruned_loss=0.112, over 5662666.31 frames. ], batch size: 412, lr: 2.72e-03, grad_scale: 4.0
+2023-03-06 05:04:16,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4826, 1.6657, 1.8005, 1.3330], device='cuda:1'), covar=tensor([0.1793, 0.2324, 0.1405, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0693, 0.0872, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 05:04:16,963 INFO [optim.py:369] (1/2) 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 05:04:46,035 INFO [train.py:968] (1/2) Epoch 12, batch 13000, giga_loss[loss=0.3239, simple_loss=0.3778, pruned_loss=0.135, over 26735.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3584, pruned_loss=0.1071, over 5665196.90 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3551, pruned_loss=0.09951, over 5761080.51 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3594, pruned_loss=0.1085, over 5659613.66 frames. ], batch size: 555, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:05:06,814 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 13050, giga_loss[loss=0.2775, simple_loss=0.3517, pruned_loss=0.1016, over 28735.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3584, pruned_loss=0.1073, over 5662404.15 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3546, pruned_loss=0.09931, over 5766664.32 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3599, pruned_loss=0.1089, over 5649296.81 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:05:55,528 INFO [optim.py:369] (1/2) 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 05:06:26,622 INFO [train.py:968] (1/2) Epoch 12, batch 13100, libri_loss[loss=0.2968, simple_loss=0.3715, pruned_loss=0.1111, over 29526.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3567, pruned_loss=0.1058, over 5659974.12 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09913, over 5759386.37 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3585, pruned_loss=0.1074, over 5653828.93 frames. ], batch size: 78, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:07:15,024 INFO [train.py:968] (1/2) Epoch 12, batch 13150, giga_loss[loss=0.3007, simple_loss=0.367, pruned_loss=0.1172, over 28907.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3533, pruned_loss=0.1035, over 5665129.16 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3535, pruned_loss=0.09887, over 5761588.02 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3551, pruned_loss=0.1051, over 5656504.54 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:07:20,698 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:07:23,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-06 05:07:30,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2189, 1.2701, 1.1273, 0.9975], device='cuda:1'), covar=tensor([0.0797, 0.0462, 0.0969, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0436, 0.0496, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:07:34,510 INFO [optim.py:369] (1/2) 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 05:07:55,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2086, 2.5909, 1.2173, 1.3218], device='cuda:1'), covar=tensor([0.1002, 0.0484, 0.0965, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0516, 0.0344, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 05:08:05,208 INFO [train.py:968] (1/2) Epoch 12, batch 13200, giga_loss[loss=0.2705, simple_loss=0.3497, pruned_loss=0.09563, over 28659.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3518, pruned_loss=0.1029, over 5667404.86 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3534, pruned_loss=0.09912, over 5763588.20 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3534, pruned_loss=0.104, over 5656952.64 frames. ], batch size: 92, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:08:13,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2799, 1.0489, 4.2320, 3.3658], device='cuda:1'), covar=tensor([0.1663, 0.2835, 0.0409, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0664, 0.0591, 0.0864, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:08:55,448 INFO [train.py:968] (1/2) Epoch 12, batch 13250, libri_loss[loss=0.2584, simple_loss=0.3412, pruned_loss=0.0878, over 29643.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3515, pruned_loss=0.1021, over 5673833.96 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3529, pruned_loss=0.09881, over 5765977.11 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3531, pruned_loss=0.1034, over 5661388.84 frames. ], batch size: 91, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:09:15,072 INFO [optim.py:369] (1/2) 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 05:09:43,382 INFO [train.py:968] (1/2) Epoch 12, batch 13300, giga_loss[loss=0.2267, simple_loss=0.3184, pruned_loss=0.06752, over 28956.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3494, pruned_loss=0.1003, over 5671401.27 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3529, pruned_loss=0.09887, over 5767542.96 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3507, pruned_loss=0.1013, over 5658285.29 frames. ], batch size: 155, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:09:54,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:10:08,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:10:16,501 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:10:17,624 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-06 05:10:33,654 INFO [train.py:968] (1/2) Epoch 12, batch 13350, giga_loss[loss=0.2562, simple_loss=0.3354, pruned_loss=0.08855, over 28071.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3463, pruned_loss=0.09775, over 5666537.56 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3522, pruned_loss=0.09847, over 5761111.58 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3479, pruned_loss=0.09892, over 5658532.84 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:10:53,614 INFO [optim.py:369] (1/2) 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 05:10:57,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:11:12,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=514580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:11:20,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:11:24,872 INFO [train.py:968] (1/2) Epoch 12, batch 13400, giga_loss[loss=0.2745, simple_loss=0.3466, pruned_loss=0.1012, over 28340.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3422, pruned_loss=0.09534, over 5667784.78 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3513, pruned_loss=0.09799, over 5764605.85 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.344, pruned_loss=0.09665, over 5656238.53 frames. ], batch size: 368, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:11:38,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0709, 1.2153, 3.4983, 3.0103], device='cuda:1'), covar=tensor([0.1661, 0.2546, 0.0495, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0666, 0.0592, 0.0866, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:12:21,188 INFO [train.py:968] (1/2) Epoch 12, batch 13450, giga_loss[loss=0.2582, simple_loss=0.3346, pruned_loss=0.09089, over 28772.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3402, pruned_loss=0.09521, over 5651426.94 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3509, pruned_loss=0.09801, over 5767632.32 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3418, pruned_loss=0.09619, over 5637861.38 frames. ], batch size: 199, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:12:25,083 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:12:28,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=514648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:12:37,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:12:40,804 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:12:57,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=514677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:13:10,859 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 13500, libri_loss[loss=0.2143, simple_loss=0.2921, pruned_loss=0.0682, over 28496.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3399, pruned_loss=0.09607, over 5656107.51 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3503, pruned_loss=0.09786, over 5769919.83 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3415, pruned_loss=0.09693, over 5640572.72 frames. ], batch size: 63, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:13:20,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6807, 1.7622, 1.3046, 1.4302], device='cuda:1'), covar=tensor([0.0786, 0.0619, 0.0885, 0.1231], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0438, 0.0500, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:13:44,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:13:58,951 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:14:01,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=514734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:14:12,088 INFO [train.py:968] (1/2) Epoch 12, batch 13550, giga_loss[loss=0.3041, simple_loss=0.3799, pruned_loss=0.1141, over 28333.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.09652, over 5645377.07 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3499, pruned_loss=0.09767, over 5771682.16 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3426, pruned_loss=0.09734, over 5630249.33 frames. ], batch size: 368, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:14:28,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4788, 1.6635, 1.3864, 1.6460], device='cuda:1'), covar=tensor([0.2479, 0.2244, 0.2614, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.0964, 0.1164, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 05:14:34,807 INFO [optim.py:369] (1/2) 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 05:14:35,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=514763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:15:09,811 INFO [train.py:968] (1/2) Epoch 12, batch 13600, giga_loss[loss=0.248, simple_loss=0.3345, pruned_loss=0.08075, over 28947.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3431, pruned_loss=0.09595, over 5651515.51 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3497, pruned_loss=0.09761, over 5773713.67 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3444, pruned_loss=0.09665, over 5636040.32 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:15:17,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2445, 1.3851, 1.2610, 1.3195], device='cuda:1'), covar=tensor([0.1673, 0.1285, 0.1128, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.1686, 0.1596, 0.1549, 0.1657], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 05:16:08,520 INFO [train.py:968] (1/2) Epoch 12, batch 13650, giga_loss[loss=0.267, simple_loss=0.3477, pruned_loss=0.09311, over 28652.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3432, pruned_loss=0.09604, over 5641808.07 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3489, pruned_loss=0.09728, over 5766411.98 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3448, pruned_loss=0.09687, over 5632375.74 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:16:08,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3650, 3.2446, 1.4472, 1.4717], device='cuda:1'), covar=tensor([0.0931, 0.0302, 0.0916, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0514, 0.0344, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 05:16:30,767 INFO [optim.py:369] (1/2) 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 05:16:32,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=514864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:16:37,458 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=514867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:17:07,773 INFO [train.py:968] (1/2) Epoch 12, batch 13700, giga_loss[loss=0.2327, simple_loss=0.3156, pruned_loss=0.07491, over 28914.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.342, pruned_loss=0.09527, over 5650935.45 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3493, pruned_loss=0.09765, over 5768348.89 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3428, pruned_loss=0.09558, over 5639109.69 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:17:12,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=514896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:17:14,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:17:50,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1326, 1.3113, 3.1684, 2.8178], device='cuda:1'), covar=tensor([0.1478, 0.2526, 0.0506, 0.1804], device='cuda:1'), in_proj_covar=tensor([0.0665, 0.0593, 0.0866, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:17:53,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-06 05:18:09,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:18:11,145 INFO [train.py:968] (1/2) Epoch 12, batch 13750, giga_loss[loss=0.3162, simple_loss=0.3722, pruned_loss=0.1301, over 26898.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3405, pruned_loss=0.09364, over 5646945.17 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.349, pruned_loss=0.09746, over 5769738.90 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3414, pruned_loss=0.09403, over 5635455.31 frames. ], batch size: 555, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:18:28,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=514955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:18:39,391 INFO [optim.py:369] (1/2) 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 05:19:12,827 INFO [train.py:968] (1/2) Epoch 12, batch 13800, giga_loss[loss=0.2161, simple_loss=0.2874, pruned_loss=0.0724, over 24454.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3393, pruned_loss=0.09252, over 5635935.18 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3487, pruned_loss=0.09754, over 5761564.19 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.34, pruned_loss=0.09266, over 5633025.98 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:19:48,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5818, 3.3759, 3.1947, 2.0455], device='cuda:1'), covar=tensor([0.0683, 0.0926, 0.0849, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.1044, 0.0971, 0.0842, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 05:19:58,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5440, 1.4185, 4.8444, 3.6410], device='cuda:1'), covar=tensor([0.1631, 0.2661, 0.0331, 0.0728], device='cuda:1'), in_proj_covar=tensor([0.0663, 0.0591, 0.0861, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:20:14,577 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:20:15,916 INFO [train.py:968] (1/2) Epoch 12, batch 13850, giga_loss[loss=0.203, simple_loss=0.2918, pruned_loss=0.05706, over 28980.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3354, pruned_loss=0.0911, over 5648986.02 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3484, pruned_loss=0.09733, over 5763811.42 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3362, pruned_loss=0.09134, over 5643158.64 frames. ], batch size: 136, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:20:17,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:20:41,117 INFO [optim.py:369] (1/2) 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 05:20:52,224 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:05,964 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:09,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:17,652 INFO [train.py:968] (1/2) Epoch 12, batch 13900, giga_loss[loss=0.2343, simple_loss=0.3237, pruned_loss=0.07247, over 28887.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3354, pruned_loss=0.09145, over 5644002.70 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3482, pruned_loss=0.09727, over 5752010.05 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3361, pruned_loss=0.09163, over 5647513.14 frames. ], batch size: 164, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:21:22,997 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:26,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:40,531 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:21:59,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:22:13,255 INFO [train.py:968] (1/2) Epoch 12, batch 13950, libri_loss[loss=0.2353, simple_loss=0.3036, pruned_loss=0.0835, over 29340.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3355, pruned_loss=0.0911, over 5657257.80 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.348, pruned_loss=0.0974, over 5752945.58 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3359, pruned_loss=0.09098, over 5656396.59 frames. ], batch size: 71, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:22:18,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:22:37,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4257, 1.6141, 1.3196, 1.6532], device='cuda:1'), covar=tensor([0.2576, 0.2444, 0.2755, 0.2187], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.0965, 0.1163, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 05:22:38,095 INFO [optim.py:369] (1/2) 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 05:23:18,019 INFO [train.py:968] (1/2) Epoch 12, batch 14000, giga_loss[loss=0.2533, simple_loss=0.3414, pruned_loss=0.08259, over 28716.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3389, pruned_loss=0.09241, over 5665438.77 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3477, pruned_loss=0.09725, over 5754231.17 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3394, pruned_loss=0.09239, over 5662903.53 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:23:42,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 05:24:21,039 INFO [train.py:968] (1/2) Epoch 12, batch 14050, libri_loss[loss=0.2295, simple_loss=0.3068, pruned_loss=0.07613, over 29558.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3383, pruned_loss=0.09171, over 5670946.89 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3475, pruned_loss=0.09717, over 5757681.40 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3386, pruned_loss=0.09164, over 5663367.06 frames. ], batch size: 77, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:24:49,531 INFO [optim.py:369] (1/2) 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 05:25:11,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-06 05:25:27,504 INFO [train.py:968] (1/2) Epoch 12, batch 14100, libri_loss[loss=0.2422, simple_loss=0.3112, pruned_loss=0.08653, over 29649.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3363, pruned_loss=0.09089, over 5673032.98 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3471, pruned_loss=0.09696, over 5750671.26 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3367, pruned_loss=0.09091, over 5670759.06 frames. ], batch size: 73, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:26:24,717 INFO [train.py:968] (1/2) Epoch 12, batch 14150, giga_loss[loss=0.2522, simple_loss=0.3463, pruned_loss=0.07905, over 28487.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3378, pruned_loss=0.092, over 5670802.84 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3467, pruned_loss=0.09686, over 5755075.42 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3381, pruned_loss=0.09189, over 5661653.34 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:26:56,505 INFO [optim.py:369] (1/2) 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 05:27:05,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1380, 1.4134, 1.2579, 1.0570], device='cuda:1'), covar=tensor([0.1732, 0.1652, 0.1149, 0.1524], device='cuda:1'), in_proj_covar=tensor([0.1692, 0.1600, 0.1545, 0.1660], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 05:27:32,827 INFO [train.py:968] (1/2) Epoch 12, batch 14200, giga_loss[loss=0.2675, simple_loss=0.3627, pruned_loss=0.08616, over 28660.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3418, pruned_loss=0.09172, over 5659268.38 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3467, pruned_loss=0.09679, over 5754075.69 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3419, pruned_loss=0.09163, over 5651761.51 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:27:51,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5302, 3.0973, 2.4273, 2.2787], device='cuda:1'), covar=tensor([0.1621, 0.0916, 0.1149, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.1683, 0.1591, 0.1538, 0.1653], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 05:28:16,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-06 05:28:29,573 INFO [train.py:968] (1/2) Epoch 12, batch 14250, giga_loss[loss=0.1846, simple_loss=0.2625, pruned_loss=0.05342, over 24349.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3424, pruned_loss=0.0907, over 5655221.90 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3468, pruned_loss=0.09692, over 5758648.44 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3423, pruned_loss=0.09033, over 5642025.74 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:28:57,400 INFO [optim.py:369] (1/2) 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 05:29:28,258 INFO [train.py:968] (1/2) Epoch 12, batch 14300, giga_loss[loss=0.2648, simple_loss=0.3507, pruned_loss=0.08939, over 28914.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3423, pruned_loss=0.08925, over 5652980.88 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3468, pruned_loss=0.09702, over 5751408.00 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.08869, over 5646063.43 frames. ], batch size: 199, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:30:05,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=515521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:30:31,815 INFO [train.py:968] (1/2) Epoch 12, batch 14350, giga_loss[loss=0.2585, simple_loss=0.3401, pruned_loss=0.08851, over 29044.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3434, pruned_loss=0.09025, over 5658684.08 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3464, pruned_loss=0.09689, over 5751057.36 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3435, pruned_loss=0.08978, over 5651997.50 frames. ], batch size: 155, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:30:49,031 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:30:57,466 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 14400, giga_loss[loss=0.2523, simple_loss=0.3122, pruned_loss=0.0962, over 24841.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3434, pruned_loss=0.09175, over 5664998.24 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3464, pruned_loss=0.09692, over 5753153.66 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3434, pruned_loss=0.09122, over 5655684.94 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:31:46,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3452, 3.0934, 1.3355, 1.5360], device='cuda:1'), covar=tensor([0.0911, 0.0308, 0.0921, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0508, 0.0344, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 05:32:37,839 INFO [train.py:968] (1/2) Epoch 12, batch 14450, giga_loss[loss=0.2581, simple_loss=0.3414, pruned_loss=0.08738, over 28906.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3439, pruned_loss=0.09305, over 5668801.43 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3462, pruned_loss=0.09681, over 5756709.98 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.344, pruned_loss=0.09262, over 5656628.28 frames. ], batch size: 164, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:32:50,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7951, 2.1920, 2.0724, 1.5210], device='cuda:1'), covar=tensor([0.1513, 0.2102, 0.1208, 0.1501], device='cuda:1'), in_proj_covar=tensor([0.0830, 0.0686, 0.0872, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 05:33:12,556 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=515664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:33:12,897 INFO [optim.py:369] (1/2) 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 05:33:15,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=515667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:33:57,288 INFO [train.py:968] (1/2) Epoch 12, batch 14500, giga_loss[loss=0.2314, simple_loss=0.3204, pruned_loss=0.07123, over 28164.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.341, pruned_loss=0.09139, over 5682899.72 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3457, pruned_loss=0.09658, over 5761253.43 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3415, pruned_loss=0.09115, over 5666778.65 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:34:07,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=515696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:34:54,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2668, 1.5482, 1.4560, 1.3710], device='cuda:1'), covar=tensor([0.1349, 0.1570, 0.1787, 0.1559], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0705, 0.0653, 0.0643], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 05:35:07,357 INFO [train.py:968] (1/2) Epoch 12, batch 14550, giga_loss[loss=0.273, simple_loss=0.3549, pruned_loss=0.09551, over 27977.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3375, pruned_loss=0.08952, over 5675302.42 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3455, pruned_loss=0.09634, over 5763054.91 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3379, pruned_loss=0.08933, over 5657679.89 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:35:33,910 INFO [optim.py:369] (1/2) 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 05:35:58,669 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-06 05:36:10,993 INFO [train.py:968] (1/2) Epoch 12, batch 14600, giga_loss[loss=0.2554, simple_loss=0.3349, pruned_loss=0.08789, over 28606.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3355, pruned_loss=0.08891, over 5678408.28 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3447, pruned_loss=0.09609, over 5766383.57 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3363, pruned_loss=0.08884, over 5659529.66 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:36:20,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:36:35,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:37:13,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 05:37:15,089 INFO [train.py:968] (1/2) Epoch 12, batch 14650, giga_loss[loss=0.3352, simple_loss=0.4015, pruned_loss=0.1345, over 28699.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3361, pruned_loss=0.08945, over 5679773.72 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3443, pruned_loss=0.09583, over 5758316.24 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.337, pruned_loss=0.08952, over 5670503.15 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:37:23,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:37:49,271 INFO [optim.py:369] (1/2) 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 05:37:56,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2379, 1.8781, 1.5975, 1.3900], device='cuda:1'), covar=tensor([0.0663, 0.0290, 0.0242, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 05:38:24,388 INFO [train.py:968] (1/2) Epoch 12, batch 14700, giga_loss[loss=0.2616, simple_loss=0.3478, pruned_loss=0.08765, over 28652.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3398, pruned_loss=0.09125, over 5674393.06 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3443, pruned_loss=0.09583, over 5758316.24 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3404, pruned_loss=0.09131, over 5667177.67 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:38:39,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=515904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:39:12,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=515931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:39:26,294 INFO [train.py:968] (1/2) Epoch 12, batch 14750, giga_loss[loss=0.2653, simple_loss=0.3423, pruned_loss=0.09415, over 28476.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3376, pruned_loss=0.09127, over 5682266.41 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.344, pruned_loss=0.09572, over 5762158.22 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09131, over 5671283.07 frames. ], batch size: 336, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:39:58,177 INFO [optim.py:369] (1/2) 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 05:40:30,401 INFO [train.py:968] (1/2) Epoch 12, batch 14800, giga_loss[loss=0.2687, simple_loss=0.3453, pruned_loss=0.09601, over 28898.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3395, pruned_loss=0.09336, over 5674848.99 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3438, pruned_loss=0.09572, over 5761766.07 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3401, pruned_loss=0.09333, over 5665093.10 frames. ], batch size: 284, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:40:33,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8197, 2.0707, 1.9098, 1.7171], device='cuda:1'), covar=tensor([0.1378, 0.1999, 0.1599, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0704, 0.0654, 0.0641], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 05:40:42,077 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:41:28,142 INFO [train.py:968] (1/2) Epoch 12, batch 14850, giga_loss[loss=0.2705, simple_loss=0.3559, pruned_loss=0.09253, over 28998.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3395, pruned_loss=0.09313, over 5673754.86 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.0958, over 5763348.55 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3399, pruned_loss=0.09293, over 5661706.84 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:41:49,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4479, 1.5512, 1.1530, 1.1801], device='cuda:1'), covar=tensor([0.0830, 0.0529, 0.1038, 0.1000], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0434, 0.0495, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:42:01,197 INFO [optim.py:369] (1/2) 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 05:42:04,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5994, 1.6652, 1.1705, 1.2536], device='cuda:1'), covar=tensor([0.0744, 0.0520, 0.0979, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0434, 0.0496, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:42:11,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:42:15,658 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 14900, libri_loss[loss=0.2435, simple_loss=0.3108, pruned_loss=0.08807, over 29353.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3419, pruned_loss=0.09316, over 5676778.95 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3436, pruned_loss=0.09584, over 5765754.23 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3423, pruned_loss=0.09291, over 5663406.23 frames. ], batch size: 67, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:42:56,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:43:53,030 INFO [train.py:968] (1/2) Epoch 12, batch 14950, giga_loss[loss=0.2432, simple_loss=0.325, pruned_loss=0.08065, over 28058.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3416, pruned_loss=0.09284, over 5668423.31 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3431, pruned_loss=0.09562, over 5761014.55 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3423, pruned_loss=0.0928, over 5660104.73 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:44:29,535 INFO [optim.py:369] (1/2) 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 05:44:30,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:44:39,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:45:05,253 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:45:11,461 INFO [train.py:968] (1/2) Epoch 12, batch 15000, giga_loss[loss=0.2808, simple_loss=0.3478, pruned_loss=0.1069, over 28737.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3383, pruned_loss=0.0919, over 5678870.74 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3426, pruned_loss=0.0955, over 5752899.63 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3393, pruned_loss=0.09193, over 5677654.37 frames. ], batch size: 243, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:45:11,461 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 05:45:20,525 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 05:45:46,035 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3114, 1.9782, 1.4918, 1.6278], device='cuda:1'), covar=tensor([0.0751, 0.0315, 0.0313, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 05:46:01,052 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 15050, giga_loss[loss=0.2277, simple_loss=0.3064, pruned_loss=0.07444, over 28705.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3317, pruned_loss=0.08901, over 5680712.80 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.342, pruned_loss=0.09521, over 5753255.93 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3329, pruned_loss=0.08918, over 5677898.92 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:46:34,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9571, 3.7657, 3.5711, 1.5407], device='cuda:1'), covar=tensor([0.0653, 0.0780, 0.0842, 0.2234], device='cuda:1'), in_proj_covar=tensor([0.1034, 0.0961, 0.0842, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 05:46:52,922 INFO [optim.py:369] (1/2) 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 05:47:07,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:47:18,074 INFO [train.py:968] (1/2) Epoch 12, batch 15100, giga_loss[loss=0.2479, simple_loss=0.3257, pruned_loss=0.08503, over 28874.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3321, pruned_loss=0.08955, over 5685583.90 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3421, pruned_loss=0.09525, over 5757644.28 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3324, pruned_loss=0.08934, over 5675753.49 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:47:36,230 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:47:49,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:47:53,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:04,605 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:07,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:15,422 INFO [train.py:968] (1/2) Epoch 12, batch 15150, giga_loss[loss=0.2733, simple_loss=0.3476, pruned_loss=0.0995, over 28591.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3338, pruned_loss=0.09143, over 5678849.52 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3419, pruned_loss=0.09518, over 5758440.75 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3339, pruned_loss=0.09121, over 5668063.50 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 2.0
+2023-03-06 05:48:22,849 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:39,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:44,100 INFO [optim.py:369] (1/2) 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 05:48:44,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:47,280 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:48:52,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:49:09,861 INFO [train.py:968] (1/2) Epoch 12, batch 15200, giga_loss[loss=0.2791, simple_loss=0.3482, pruned_loss=0.105, over 28770.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3331, pruned_loss=0.0908, over 5678054.05 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3415, pruned_loss=0.09494, over 5761834.25 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3334, pruned_loss=0.09071, over 5663940.18 frames. ], batch size: 99, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:49:22,442 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:49:34,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 05:49:50,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:49:55,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:50:14,358 INFO [train.py:968] (1/2) Epoch 12, batch 15250, giga_loss[loss=0.2254, simple_loss=0.3153, pruned_loss=0.06774, over 28624.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3302, pruned_loss=0.08802, over 5674662.12 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09469, over 5763326.21 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3307, pruned_loss=0.08812, over 5661317.37 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:50:29,885 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:50:30,040 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 05:50:45,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2034, 1.2987, 1.1599, 0.9300], device='cuda:1'), covar=tensor([0.0844, 0.0463, 0.0961, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0433, 0.0498, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 05:50:45,377 INFO [optim.py:369] (1/2) 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 05:50:47,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3830, 3.1219, 1.4400, 1.5275], device='cuda:1'), covar=tensor([0.0901, 0.0299, 0.0895, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0505, 0.0343, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 05:50:57,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2178, 1.6203, 1.4983, 1.1767], device='cuda:1'), covar=tensor([0.1352, 0.1928, 0.1108, 0.1466], device='cuda:1'), in_proj_covar=tensor([0.0828, 0.0682, 0.0870, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 05:51:20,916 INFO [train.py:968] (1/2) Epoch 12, batch 15300, giga_loss[loss=0.2376, simple_loss=0.3165, pruned_loss=0.07939, over 28453.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.328, pruned_loss=0.0868, over 5667108.85 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3407, pruned_loss=0.09455, over 5765557.82 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3286, pruned_loss=0.08689, over 5653220.48 frames. ], batch size: 369, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:51:23,421 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:02,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:05,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:14,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=516530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:52:30,395 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 15350, giga_loss[loss=0.2301, simple_loss=0.3187, pruned_loss=0.07077, over 28708.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3286, pruned_loss=0.08691, over 5679982.04 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3412, pruned_loss=0.09504, over 5767000.40 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3283, pruned_loss=0.08637, over 5665619.09 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:52:42,307 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:53:05,137 INFO [optim.py:369] (1/2) 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 05:53:33,881 INFO [train.py:968] (1/2) Epoch 12, batch 15400, libri_loss[loss=0.2704, simple_loss=0.3474, pruned_loss=0.09676, over 29666.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08639, over 5686343.84 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3412, pruned_loss=0.09501, over 5764773.72 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3284, pruned_loss=0.08583, over 5675234.79 frames. ], batch size: 91, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:54:38,873 INFO [train.py:968] (1/2) Epoch 12, batch 15450, giga_loss[loss=0.2494, simple_loss=0.3274, pruned_loss=0.0857, over 28892.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3303, pruned_loss=0.08786, over 5688828.74 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3413, pruned_loss=0.09521, over 5766408.47 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3297, pruned_loss=0.08712, over 5677234.78 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:55:10,305 INFO [optim.py:369] (1/2) 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 05:55:26,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:55:36,210 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:55:38,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=516687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:55:45,389 INFO [train.py:968] (1/2) Epoch 12, batch 15500, giga_loss[loss=0.2312, simple_loss=0.2917, pruned_loss=0.08539, over 24381.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3285, pruned_loss=0.0867, over 5684078.22 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3413, pruned_loss=0.09514, over 5767236.29 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.08611, over 5673670.69 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:56:08,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-06 05:56:11,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:56:39,530 INFO [train.py:968] (1/2) Epoch 12, batch 15550, giga_loss[loss=0.2951, simple_loss=0.3683, pruned_loss=0.111, over 28392.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3293, pruned_loss=0.08619, over 5675006.35 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.341, pruned_loss=0.09495, over 5770679.63 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3287, pruned_loss=0.08561, over 5660776.03 frames. ], batch size: 369, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:57:10,360 INFO [optim.py:369] (1/2) 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 05:57:14,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7420, 2.6191, 1.7652, 1.1178], device='cuda:1'), covar=tensor([0.5441, 0.2601, 0.3002, 0.4405], device='cuda:1'), in_proj_covar=tensor([0.1575, 0.1500, 0.1512, 0.1291], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 05:57:43,740 INFO [train.py:968] (1/2) Epoch 12, batch 15600, giga_loss[loss=0.2318, simple_loss=0.3273, pruned_loss=0.06814, over 28936.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3332, pruned_loss=0.0875, over 5671625.68 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.341, pruned_loss=0.09493, over 5771343.53 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3326, pruned_loss=0.08699, over 5659072.62 frames. ], batch size: 164, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 05:57:46,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-06 05:58:24,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=516822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:58:26,619 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 15650, giga_loss[loss=0.3039, simple_loss=0.3814, pruned_loss=0.1132, over 28981.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.335, pruned_loss=0.08824, over 5670970.28 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.341, pruned_loss=0.09496, over 5773729.82 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3345, pruned_loss=0.08771, over 5657297.54 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 05:58:59,687 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=516854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:59:17,200 INFO [optim.py:369] (1/2) 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 05:59:18,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 05:59:44,853 INFO [train.py:968] (1/2) Epoch 12, batch 15700, giga_loss[loss=0.2765, simple_loss=0.3529, pruned_loss=0.1001, over 28989.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.08914, over 5660084.33 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3407, pruned_loss=0.09483, over 5774349.43 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3355, pruned_loss=0.08874, over 5647209.88 frames. ], batch size: 285, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:00:02,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=516905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:00:44,903 INFO [train.py:968] (1/2) Epoch 12, batch 15750, giga_loss[loss=0.2145, simple_loss=0.298, pruned_loss=0.06551, over 28768.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3334, pruned_loss=0.08811, over 5658809.02 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3406, pruned_loss=0.09482, over 5775856.69 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3332, pruned_loss=0.08772, over 5645726.26 frames. ], batch size: 92, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:01:07,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 06:01:17,582 INFO [optim.py:369] (1/2) 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 06:01:50,953 INFO [train.py:968] (1/2) Epoch 12, batch 15800, giga_loss[loss=0.242, simple_loss=0.3192, pruned_loss=0.08243, over 28603.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3315, pruned_loss=0.08691, over 5661610.75 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3404, pruned_loss=0.0946, over 5777979.65 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3314, pruned_loss=0.08669, over 5647886.29 frames. ], batch size: 242, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:02:15,557 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=517012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:02:17,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=517015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:02:31,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6007, 1.7445, 1.9043, 1.4642], device='cuda:1'), covar=tensor([0.1625, 0.2190, 0.1279, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0681, 0.0867, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 06:02:46,592 INFO [train.py:968] (1/2) Epoch 12, batch 15850, giga_loss[loss=0.2214, simple_loss=0.2978, pruned_loss=0.0725, over 28923.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3303, pruned_loss=0.08688, over 5673398.28 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3399, pruned_loss=0.09431, over 5780468.78 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3305, pruned_loss=0.08678, over 5656994.55 frames. ], batch size: 175, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:02:50,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=517044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:02:56,021 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=517048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:00,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=517051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:19,216 INFO [optim.py:369] (1/2) 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 06:03:19,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:34,566 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=517080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:03:46,992 INFO [train.py:968] (1/2) Epoch 12, batch 15900, giga_loss[loss=0.267, simple_loss=0.349, pruned_loss=0.09248, over 28631.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3308, pruned_loss=0.08666, over 5682021.97 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3393, pruned_loss=0.09398, over 5784317.11 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3312, pruned_loss=0.08668, over 5662340.73 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:04:22,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9585, 2.5392, 2.3712, 1.7568], device='cuda:1'), covar=tensor([0.1693, 0.1937, 0.1284, 0.1557], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0681, 0.0868, 0.0775], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 06:04:53,677 INFO [train.py:968] (1/2) Epoch 12, batch 15950, giga_loss[loss=0.2535, simple_loss=0.3298, pruned_loss=0.08864, over 28964.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3315, pruned_loss=0.08682, over 5675958.88 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3392, pruned_loss=0.09386, over 5785567.67 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3319, pruned_loss=0.08689, over 5658402.56 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:05:26,997 INFO [optim.py:369] (1/2) 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 06:05:40,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-06 06:05:54,970 INFO [train.py:968] (1/2) Epoch 12, batch 16000, libri_loss[loss=0.2352, simple_loss=0.3078, pruned_loss=0.08126, over 29372.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3332, pruned_loss=0.0888, over 5663022.90 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3392, pruned_loss=0.09392, over 5775219.52 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3333, pruned_loss=0.08862, over 5654582.12 frames. ], batch size: 71, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:06:44,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3109, 1.5158, 1.4003, 1.2306], device='cuda:1'), covar=tensor([0.2010, 0.1764, 0.1247, 0.1558], device='cuda:1'), in_proj_covar=tensor([0.1705, 0.1597, 0.1536, 0.1648], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:06:55,617 INFO [train.py:968] (1/2) Epoch 12, batch 16050, giga_loss[loss=0.2376, simple_loss=0.306, pruned_loss=0.08459, over 24314.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3374, pruned_loss=0.09115, over 5655482.64 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3393, pruned_loss=0.094, over 5774712.26 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3374, pruned_loss=0.09091, over 5648201.60 frames. ], batch size: 705, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:07:02,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9655, 1.3665, 1.0905, 0.1915], device='cuda:1'), covar=tensor([0.2822, 0.2242, 0.3773, 0.4658], device='cuda:1'), in_proj_covar=tensor([0.1563, 0.1495, 0.1494, 0.1278], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:07:28,752 INFO [optim.py:369] (1/2) 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 06:07:55,362 INFO [train.py:968] (1/2) Epoch 12, batch 16100, giga_loss[loss=0.323, simple_loss=0.3904, pruned_loss=0.1278, over 28395.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.34, pruned_loss=0.09181, over 5646506.88 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3395, pruned_loss=0.09409, over 5766904.43 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3398, pruned_loss=0.09152, over 5646335.03 frames. ], batch size: 368, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:08:57,064 INFO [train.py:968] (1/2) Epoch 12, batch 16150, giga_loss[loss=0.2627, simple_loss=0.3418, pruned_loss=0.09182, over 28644.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3398, pruned_loss=0.09191, over 5649967.76 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.339, pruned_loss=0.09392, over 5772051.66 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3401, pruned_loss=0.09172, over 5640801.40 frames. ], batch size: 307, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:09:35,691 INFO [optim.py:369] (1/2) 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 06:10:04,070 INFO [train.py:968] (1/2) Epoch 12, batch 16200, giga_loss[loss=0.25, simple_loss=0.3248, pruned_loss=0.08762, over 28146.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3371, pruned_loss=0.09044, over 5652291.97 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3389, pruned_loss=0.09383, over 5766179.41 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3374, pruned_loss=0.0903, over 5646826.25 frames. ], batch size: 412, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:10:29,007 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 06:10:34,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4039, 2.6189, 1.9486, 2.4182], device='cuda:1'), covar=tensor([0.0714, 0.0524, 0.0910, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0436, 0.0502, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 06:10:42,157 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-06 06:11:01,030 INFO [train.py:968] (1/2) Epoch 12, batch 16250, giga_loss[loss=0.2569, simple_loss=0.3506, pruned_loss=0.08154, over 28913.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3357, pruned_loss=0.08953, over 5666909.10 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3384, pruned_loss=0.09365, over 5769395.97 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3362, pruned_loss=0.08942, over 5655235.40 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:11:04,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=517443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:11:34,489 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 16300, giga_loss[loss=0.249, simple_loss=0.3278, pruned_loss=0.0851, over 28923.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3349, pruned_loss=0.08928, over 5655777.76 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3383, pruned_loss=0.0935, over 5752036.79 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3354, pruned_loss=0.08921, over 5659144.96 frames. ], batch size: 186, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:12:22,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2433, 1.8823, 1.3052, 0.4684], device='cuda:1'), covar=tensor([0.3184, 0.1627, 0.2863, 0.3991], device='cuda:1'), in_proj_covar=tensor([0.1573, 0.1506, 0.1508, 0.1292], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:13:06,292 INFO [train.py:968] (1/2) Epoch 12, batch 16350, giga_loss[loss=0.2624, simple_loss=0.3316, pruned_loss=0.09658, over 28888.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3346, pruned_loss=0.09063, over 5647911.24 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3385, pruned_loss=0.09358, over 5753902.96 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3348, pruned_loss=0.09046, over 5647772.56 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:13:34,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 06:13:37,083 INFO [optim.py:369] (1/2) 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 06:13:58,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=517586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:14:00,366 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 16400, giga_loss[loss=0.2227, simple_loss=0.3137, pruned_loss=0.06589, over 28908.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3324, pruned_loss=0.08934, over 5648039.32 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3379, pruned_loss=0.09336, over 5748680.30 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.333, pruned_loss=0.08933, over 5649658.83 frames. ], batch size: 145, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:14:40,680 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=517618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:15:06,207 INFO [train.py:968] (1/2) Epoch 12, batch 16450, libri_loss[loss=0.2688, simple_loss=0.3252, pruned_loss=0.1062, over 29356.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3315, pruned_loss=0.08771, over 5661320.21 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3375, pruned_loss=0.09318, over 5750978.08 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3322, pruned_loss=0.08777, over 5659202.71 frames. ], batch size: 67, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:15:34,377 INFO [optim.py:369] (1/2) 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 06:15:59,484 INFO [train.py:968] (1/2) Epoch 12, batch 16500, giga_loss[loss=0.2455, simple_loss=0.3411, pruned_loss=0.07494, over 28956.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3308, pruned_loss=0.08594, over 5675520.21 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3367, pruned_loss=0.0928, over 5755660.00 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3319, pruned_loss=0.08609, over 5666086.55 frames. ], batch size: 284, lr: 2.71e-03, grad_scale: 8.0
+2023-03-06 06:16:11,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517701.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:16:54,238 INFO [train.py:968] (1/2) Epoch 12, batch 16550, giga_loss[loss=0.3039, simple_loss=0.3834, pruned_loss=0.1122, over 28903.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3338, pruned_loss=0.08603, over 5684089.92 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3362, pruned_loss=0.09241, over 5756775.72 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.335, pruned_loss=0.08632, over 5673682.47 frames. ], batch size: 213, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:16:56,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-06 06:17:16,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3271, 4.1485, 3.9106, 1.9303], device='cuda:1'), covar=tensor([0.0516, 0.0690, 0.0728, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1037, 0.0960, 0.0840, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 06:17:27,273 INFO [optim.py:369] (1/2) 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 06:17:33,322 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-06 06:17:51,462 INFO [train.py:968] (1/2) Epoch 12, batch 16600, giga_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08752, over 28730.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.335, pruned_loss=0.08644, over 5674935.58 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.336, pruned_loss=0.0923, over 5756272.36 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3361, pruned_loss=0.0867, over 5666466.76 frames. ], batch size: 262, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:18:08,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4692, 1.8709, 1.5488, 1.4826], device='cuda:1'), covar=tensor([0.2032, 0.1513, 0.1721, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.1731, 0.1616, 0.1556, 0.1675], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:18:57,308 INFO [train.py:968] (1/2) Epoch 12, batch 16650, giga_loss[loss=0.2574, simple_loss=0.3326, pruned_loss=0.09109, over 27527.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3341, pruned_loss=0.08628, over 5662867.27 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3356, pruned_loss=0.0921, over 5755243.44 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3354, pruned_loss=0.08659, over 5655997.64 frames. ], batch size: 472, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:19:04,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:19:05,619 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=517846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:19:37,515 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 16700, giga_loss[loss=0.2607, simple_loss=0.3423, pruned_loss=0.08957, over 29058.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3347, pruned_loss=0.08667, over 5655691.89 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3357, pruned_loss=0.09221, over 5754913.17 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3355, pruned_loss=0.08676, over 5649898.61 frames. ], batch size: 214, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:21:16,672 INFO [train.py:968] (1/2) Epoch 12, batch 16750, giga_loss[loss=0.2662, simple_loss=0.3584, pruned_loss=0.08701, over 28901.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.334, pruned_loss=0.08529, over 5664107.32 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3354, pruned_loss=0.09204, over 5758081.91 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3349, pruned_loss=0.08539, over 5654989.55 frames. ], batch size: 227, lr: 2.71e-03, grad_scale: 4.0
+2023-03-06 06:21:51,599 INFO [optim.py:369] (1/2) 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 06:22:12,915 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 06:22:18,784 INFO [train.py:968] (1/2) Epoch 12, batch 16800, giga_loss[loss=0.2187, simple_loss=0.2888, pruned_loss=0.0743, over 24368.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3336, pruned_loss=0.08537, over 5658081.63 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3348, pruned_loss=0.09172, over 5756221.23 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3349, pruned_loss=0.08542, over 5646895.77 frames. ], batch size: 705, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:22:48,180 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-06 06:23:25,546 INFO [train.py:968] (1/2) Epoch 12, batch 16850, giga_loss[loss=0.2912, simple_loss=0.3738, pruned_loss=0.1043, over 28417.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3385, pruned_loss=0.08745, over 5669383.18 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3349, pruned_loss=0.09174, over 5756798.52 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3394, pruned_loss=0.08736, over 5657879.00 frames. ], batch size: 369, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:24:03,594 INFO [optim.py:369] (1/2) 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 06:24:11,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518076.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:24:31,396 INFO [train.py:968] (1/2) Epoch 12, batch 16900, giga_loss[loss=0.2239, simple_loss=0.3145, pruned_loss=0.06671, over 28892.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3385, pruned_loss=0.08738, over 5677396.03 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.335, pruned_loss=0.09168, over 5758964.29 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3393, pruned_loss=0.08728, over 5664732.59 frames. ], batch size: 227, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:25:03,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4935, 2.2953, 1.9262, 1.4580], device='cuda:1'), covar=tensor([0.3030, 0.1288, 0.1360, 0.2026], device='cuda:1'), in_proj_covar=tensor([0.1733, 0.1613, 0.1544, 0.1669], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:25:13,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2801, 2.2518, 1.3190, 1.4598], device='cuda:1'), covar=tensor([0.0801, 0.0447, 0.0733, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0504, 0.0344, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 06:25:24,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5657, 1.7839, 1.8769, 1.3867], device='cuda:1'), covar=tensor([0.1851, 0.2271, 0.1482, 0.1645], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0677, 0.0866, 0.0772], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 06:25:31,542 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-06 06:25:40,742 INFO [train.py:968] (1/2) Epoch 12, batch 16950, giga_loss[loss=0.2171, simple_loss=0.3047, pruned_loss=0.06475, over 28930.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3373, pruned_loss=0.08764, over 5675226.85 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3348, pruned_loss=0.09155, over 5759638.89 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3381, pruned_loss=0.08764, over 5663363.93 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:26:24,250 INFO [optim.py:369] (1/2) 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 06:26:46,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6612, 2.4438, 1.6132, 0.7388], device='cuda:1'), covar=tensor([0.6015, 0.3075, 0.3347, 0.5460], device='cuda:1'), in_proj_covar=tensor([0.1559, 0.1486, 0.1492, 0.1278], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:26:51,655 INFO [train.py:968] (1/2) Epoch 12, batch 17000, giga_loss[loss=0.2268, simple_loss=0.3114, pruned_loss=0.07111, over 28605.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3353, pruned_loss=0.08645, over 5669562.23 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3346, pruned_loss=0.09151, over 5748715.30 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3361, pruned_loss=0.08638, over 5668334.94 frames. ], batch size: 78, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:26:54,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5468, 1.7754, 1.4622, 1.7726], device='cuda:1'), covar=tensor([0.2580, 0.2460, 0.2730, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.0962, 0.1154, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 06:27:29,326 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-06 06:27:33,378 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518219.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:27:33,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:27:36,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:27:36,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518222.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:28:00,924 INFO [train.py:968] (1/2) Epoch 12, batch 17050, giga_loss[loss=0.2401, simple_loss=0.3281, pruned_loss=0.07609, over 28483.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3339, pruned_loss=0.08531, over 5665556.45 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3346, pruned_loss=0.09153, over 5750279.43 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3346, pruned_loss=0.08519, over 5662381.55 frames. ], batch size: 336, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:28:01,214 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518242.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:28:09,510 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518251.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:28:20,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5993, 2.0385, 1.2963, 0.9426], device='cuda:1'), covar=tensor([0.5101, 0.3072, 0.3125, 0.4560], device='cuda:1'), in_proj_covar=tensor([0.1559, 0.1486, 0.1488, 0.1277], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:28:32,132 INFO [optim.py:369] (1/2) 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 06:28:55,175 INFO [train.py:968] (1/2) Epoch 12, batch 17100, giga_loss[loss=0.2475, simple_loss=0.3333, pruned_loss=0.0809, over 28651.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3342, pruned_loss=0.08574, over 5663861.58 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3345, pruned_loss=0.09151, over 5743059.43 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3348, pruned_loss=0.08544, over 5665382.32 frames. ], batch size: 307, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:29:04,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-06 06:29:48,452 INFO [train.py:968] (1/2) Epoch 12, batch 17150, giga_loss[loss=0.2463, simple_loss=0.3302, pruned_loss=0.0812, over 28911.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.337, pruned_loss=0.08763, over 5649523.03 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3347, pruned_loss=0.09177, over 5727915.47 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3374, pruned_loss=0.08695, over 5661605.75 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:30:13,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:16,950 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:18,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:19,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:23,924 INFO [optim.py:369] (1/2) 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 06:30:46,505 INFO [train.py:968] (1/2) Epoch 12, batch 17200, giga_loss[loss=0.2516, simple_loss=0.3254, pruned_loss=0.08891, over 27593.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3375, pruned_loss=0.08826, over 5655744.79 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3347, pruned_loss=0.09173, over 5725601.14 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3378, pruned_loss=0.08773, over 5666574.41 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:30:50,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:52,327 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:55,783 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:30:56,683 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:31:32,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1939, 1.7727, 1.3237, 0.3532], device='cuda:1'), covar=tensor([0.3247, 0.2056, 0.3482, 0.4270], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1491, 0.1497, 0.1283], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:31:44,332 INFO [train.py:968] (1/2) Epoch 12, batch 17250, giga_loss[loss=0.2596, simple_loss=0.3358, pruned_loss=0.09165, over 28881.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3346, pruned_loss=0.08812, over 5656318.72 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3344, pruned_loss=0.09154, over 5728398.76 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3352, pruned_loss=0.08782, over 5661485.55 frames. ], batch size: 186, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:31:56,754 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:32:01,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3979, 1.7322, 1.4427, 1.3284], device='cuda:1'), covar=tensor([0.1930, 0.1369, 0.1381, 0.1594], device='cuda:1'), in_proj_covar=tensor([0.1711, 0.1600, 0.1539, 0.1663], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:32:10,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 06:32:19,876 INFO [optim.py:369] (1/2) 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 06:32:22,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0226, 1.2186, 1.3719, 1.0074], device='cuda:1'), covar=tensor([0.1284, 0.1190, 0.1881, 0.1441], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0708, 0.0660, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 06:32:39,472 INFO [train.py:968] (1/2) Epoch 12, batch 17300, giga_loss[loss=0.2298, simple_loss=0.3155, pruned_loss=0.07209, over 28844.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3335, pruned_loss=0.08825, over 5654173.33 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3341, pruned_loss=0.09138, over 5734040.80 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3343, pruned_loss=0.08805, over 5650811.88 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:33:18,633 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:33:32,387 INFO [train.py:968] (1/2) Epoch 12, batch 17350, giga_loss[loss=0.3383, simple_loss=0.4066, pruned_loss=0.135, over 28667.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3374, pruned_loss=0.09072, over 5662252.08 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.334, pruned_loss=0.09138, over 5739982.12 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3381, pruned_loss=0.09049, over 5650942.60 frames. ], batch size: 92, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:34:04,744 INFO [optim.py:369] (1/2) 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 06:34:13,310 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:34:20,089 INFO [train.py:968] (1/2) Epoch 12, batch 17400, giga_loss[loss=0.2903, simple_loss=0.3783, pruned_loss=0.1012, over 28872.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3464, pruned_loss=0.09578, over 5667576.57 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3341, pruned_loss=0.09143, over 5736117.48 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3472, pruned_loss=0.09561, over 5659501.04 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:34:20,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8600, 1.9066, 1.8403, 1.6808], device='cuda:1'), covar=tensor([0.1309, 0.1793, 0.1881, 0.1786], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0708, 0.0659, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 06:34:22,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8709, 3.6674, 3.4716, 2.1367], device='cuda:1'), covar=tensor([0.0554, 0.0752, 0.0745, 0.1768], device='cuda:1'), in_proj_covar=tensor([0.1036, 0.0955, 0.0837, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 06:34:43,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518617.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:35:03,282 INFO [train.py:968] (1/2) Epoch 12, batch 17450, giga_loss[loss=0.2715, simple_loss=0.3494, pruned_loss=0.09676, over 28848.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3521, pruned_loss=0.09885, over 5670289.12 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3342, pruned_loss=0.09145, over 5735836.62 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.353, pruned_loss=0.09884, over 5662748.43 frames. ], batch size: 199, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:35:06,904 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=518647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:35:30,052 INFO [optim.py:369] (1/2) 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 06:35:46,489 INFO [train.py:968] (1/2) Epoch 12, batch 17500, giga_loss[loss=0.2614, simple_loss=0.3372, pruned_loss=0.09284, over 28281.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3517, pruned_loss=0.1, over 5669865.95 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3345, pruned_loss=0.09158, over 5736103.34 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3526, pruned_loss=0.1001, over 5662357.37 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:36:30,125 INFO [train.py:968] (1/2) Epoch 12, batch 17550, giga_loss[loss=0.2708, simple_loss=0.3352, pruned_loss=0.1032, over 28852.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3453, pruned_loss=0.09711, over 5682528.85 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3345, pruned_loss=0.09146, over 5739610.38 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3464, pruned_loss=0.09753, over 5671458.71 frames. ], batch size: 112, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:36:31,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4251, 1.7051, 1.3758, 1.7123], device='cuda:1'), covar=tensor([0.2586, 0.2470, 0.2804, 0.2178], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.0962, 0.1152, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 06:36:45,603 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518760.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:36:48,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518763.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:36:57,878 INFO [optim.py:369] (1/2) 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 06:36:59,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:37:00,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:37:17,552 INFO [train.py:968] (1/2) Epoch 12, batch 17600, giga_loss[loss=0.2221, simple_loss=0.2941, pruned_loss=0.07498, over 28795.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3388, pruned_loss=0.09454, over 5691001.87 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3343, pruned_loss=0.09131, over 5742116.22 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.34, pruned_loss=0.09509, over 5679030.34 frames. ], batch size: 99, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:37:17,811 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518792.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:37:46,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:37:57,812 INFO [train.py:968] (1/2) Epoch 12, batch 17650, giga_loss[loss=0.2195, simple_loss=0.2898, pruned_loss=0.07465, over 28944.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3321, pruned_loss=0.09175, over 5700024.46 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3346, pruned_loss=0.09135, over 5746968.12 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3328, pruned_loss=0.09216, over 5684532.82 frames. ], batch size: 213, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:38:25,017 INFO [optim.py:369] (1/2) 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 06:38:40,341 INFO [train.py:968] (1/2) Epoch 12, batch 17700, giga_loss[loss=0.2439, simple_loss=0.314, pruned_loss=0.08687, over 28540.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3257, pruned_loss=0.08929, over 5694011.63 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.335, pruned_loss=0.09146, over 5740248.03 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3257, pruned_loss=0.08951, over 5685998.87 frames. ], batch size: 336, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:38:49,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:01,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:02,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:04,479 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:05,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 17750, giga_loss[loss=0.2208, simple_loss=0.2956, pruned_loss=0.07299, over 28688.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3203, pruned_loss=0.08679, over 5693873.69 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3354, pruned_loss=0.09155, over 5742554.14 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3198, pruned_loss=0.08684, over 5684949.04 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:39:28,514 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:30,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=518950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:37,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=518959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:44,407 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=518968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:45,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 06:39:48,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=518971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:39:49,023 INFO [optim.py:369] (1/2) 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 06:40:02,289 INFO [train.py:968] (1/2) Epoch 12, batch 17800, giga_loss[loss=0.1975, simple_loss=0.2804, pruned_loss=0.05734, over 28748.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3162, pruned_loss=0.08458, over 5702363.85 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.335, pruned_loss=0.09121, over 5746136.69 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3157, pruned_loss=0.08479, over 5690925.66 frames. ], batch size: 243, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:40:08,705 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:40:29,127 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:40:48,185 INFO [train.py:968] (1/2) Epoch 12, batch 17850, giga_loss[loss=0.2436, simple_loss=0.3145, pruned_loss=0.08633, over 28687.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3129, pruned_loss=0.08301, over 5691729.12 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3352, pruned_loss=0.09131, over 5739123.40 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3121, pruned_loss=0.08299, over 5687615.32 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:40:50,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:40:52,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:41:14,581 INFO [optim.py:369] (1/2) 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 06:41:17,812 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:41:24,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 06:41:29,320 INFO [train.py:968] (1/2) Epoch 12, batch 17900, giga_loss[loss=0.2317, simple_loss=0.3024, pruned_loss=0.08049, over 28612.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3095, pruned_loss=0.08136, over 5688464.44 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3356, pruned_loss=0.09149, over 5740485.33 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3083, pruned_loss=0.08109, over 5683568.89 frames. ], batch size: 307, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:41:40,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:41:43,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:07,337 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:09,470 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 17950, giga_loss[loss=0.2278, simple_loss=0.2986, pruned_loss=0.07857, over 28391.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3073, pruned_loss=0.08041, over 5690288.78 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.336, pruned_loss=0.09165, over 5733980.43 frames. ], giga_tot_loss[loss=0.2324, simple_loss=0.3052, pruned_loss=0.07979, over 5691826.14 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:42:33,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:36,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:42:40,906 INFO [optim.py:369] (1/2) 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 06:42:57,539 INFO [train.py:968] (1/2) Epoch 12, batch 18000, giga_loss[loss=0.1931, simple_loss=0.2736, pruned_loss=0.0563, over 28973.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3046, pruned_loss=0.07948, over 5675776.88 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3364, pruned_loss=0.09181, over 5726043.22 frames. ], giga_tot_loss[loss=0.23, simple_loss=0.3025, pruned_loss=0.07878, over 5683440.10 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:42:57,539 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 06:43:06,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0161, 1.1866, 3.4745, 3.0134], device='cuda:1'), covar=tensor([0.1884, 0.2847, 0.0509, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0664, 0.0582, 0.0856, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 06:43:06,968 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 06:43:07,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:43:10,666 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:43:24,886 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 18050, giga_loss[loss=0.2306, simple_loss=0.2997, pruned_loss=0.08077, over 28608.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.302, pruned_loss=0.07831, over 5683002.00 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3367, pruned_loss=0.09195, over 5727466.29 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.2994, pruned_loss=0.07739, over 5687112.14 frames. ], batch size: 336, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:43:53,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4856, 1.7781, 1.6250, 1.4793], device='cuda:1'), covar=tensor([0.2204, 0.1674, 0.1720, 0.1791], device='cuda:1'), in_proj_covar=tensor([0.1737, 0.1628, 0.1568, 0.1690], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:44:14,253 INFO [optim.py:369] (1/2) 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 06:44:35,159 INFO [train.py:968] (1/2) Epoch 12, batch 18100, giga_loss[loss=0.1973, simple_loss=0.2717, pruned_loss=0.06148, over 28773.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.2991, pruned_loss=0.07677, over 5691900.00 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3368, pruned_loss=0.09197, over 5728706.00 frames. ], giga_tot_loss[loss=0.2238, simple_loss=0.2962, pruned_loss=0.07573, over 5693201.62 frames. ], batch size: 284, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:45:20,187 INFO [train.py:968] (1/2) Epoch 12, batch 18150, giga_loss[loss=0.2243, simple_loss=0.3002, pruned_loss=0.07415, over 28606.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.2982, pruned_loss=0.07706, over 5693314.01 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3371, pruned_loss=0.09204, over 5730728.86 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.2952, pruned_loss=0.07602, over 5692112.16 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:45:51,067 INFO [optim.py:369] (1/2) 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 06:45:51,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4863, 1.7038, 1.4920, 1.3146], device='cuda:1'), covar=tensor([0.2434, 0.1916, 0.1565, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1757, 0.1650, 0.1590, 0.1714], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:46:08,228 INFO [train.py:968] (1/2) Epoch 12, batch 18200, giga_loss[loss=0.284, simple_loss=0.3375, pruned_loss=0.1152, over 23692.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.3083, pruned_loss=0.08242, over 5693684.47 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.337, pruned_loss=0.09185, over 5735194.08 frames. ], giga_tot_loss[loss=0.2341, simple_loss=0.3053, pruned_loss=0.08145, over 5688253.98 frames. ], batch size: 705, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:46:26,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5576, 3.0754, 2.4920, 2.2707], device='cuda:1'), covar=tensor([0.1784, 0.1119, 0.1273, 0.1443], device='cuda:1'), in_proj_covar=tensor([0.1751, 0.1645, 0.1584, 0.1708], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:46:49,382 INFO [train.py:968] (1/2) Epoch 12, batch 18250, giga_loss[loss=0.3392, simple_loss=0.3997, pruned_loss=0.1394, over 29113.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3223, pruned_loss=0.08965, over 5692953.22 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3372, pruned_loss=0.0919, over 5731761.87 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3187, pruned_loss=0.08853, over 5690818.97 frames. ], batch size: 128, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:47:01,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4986, 2.1377, 1.6303, 0.5952], device='cuda:1'), covar=tensor([0.4443, 0.2296, 0.3211, 0.4744], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1477, 0.1481, 0.1269], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:47:12,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7049, 2.4400, 1.6109, 0.8802], device='cuda:1'), covar=tensor([0.5101, 0.2637, 0.2886, 0.4675], device='cuda:1'), in_proj_covar=tensor([0.1555, 0.1479, 0.1483, 0.1269], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:47:14,437 INFO [optim.py:369] (1/2) 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 06:47:18,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 06:47:21,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6989, 1.9623, 1.5389, 1.9288], device='cuda:1'), covar=tensor([0.0731, 0.0264, 0.0296, 0.0758], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 06:47:27,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2308, 1.4834, 1.3242, 1.1644], device='cuda:1'), covar=tensor([0.1790, 0.1853, 0.1209, 0.1539], device='cuda:1'), in_proj_covar=tensor([0.1753, 0.1649, 0.1586, 0.1713], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:47:27,738 INFO [train.py:968] (1/2) Epoch 12, batch 18300, giga_loss[loss=0.2931, simple_loss=0.3655, pruned_loss=0.1103, over 28206.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3337, pruned_loss=0.09564, over 5698003.94 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3372, pruned_loss=0.09179, over 5733095.72 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3305, pruned_loss=0.09487, over 5694369.94 frames. ], batch size: 65, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:47:39,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:47:45,799 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:48:04,875 INFO [train.py:968] (1/2) Epoch 12, batch 18350, giga_loss[loss=0.2936, simple_loss=0.3762, pruned_loss=0.1055, over 28602.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3399, pruned_loss=0.09747, over 5707545.94 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3377, pruned_loss=0.0918, over 5742063.27 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3368, pruned_loss=0.09708, over 5694879.79 frames. ], batch size: 307, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 06:48:26,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:48:32,614 INFO [optim.py:369] (1/2) 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 06:48:40,776 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:48:43,253 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 18400, giga_loss[loss=0.2978, simple_loss=0.3778, pruned_loss=0.1089, over 28914.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3439, pruned_loss=0.09827, over 5702992.29 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.338, pruned_loss=0.09195, over 5741628.93 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3413, pruned_loss=0.09795, over 5692446.60 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:49:29,539 INFO [train.py:968] (1/2) Epoch 12, batch 18450, giga_loss[loss=0.2999, simple_loss=0.3493, pruned_loss=0.1252, over 23644.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3459, pruned_loss=0.09864, over 5693183.93 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3384, pruned_loss=0.09226, over 5744479.15 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3436, pruned_loss=0.0982, over 5681862.69 frames. ], batch size: 705, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:49:39,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:49:41,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:49:58,511 INFO [optim.py:369] (1/2) 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 06:50:02,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4355, 1.9470, 1.3503, 1.6309], device='cuda:1'), covar=tensor([0.2533, 0.2321, 0.2677, 0.2263], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.0967, 0.1157, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 06:50:05,304 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:12,858 INFO [train.py:968] (1/2) Epoch 12, batch 18500, giga_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08768, over 28964.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3477, pruned_loss=0.09983, over 5688940.16 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3385, pruned_loss=0.09233, over 5738764.50 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3459, pruned_loss=0.09957, over 5684035.75 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:50:14,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-06 06:50:29,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:32,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:48,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=519731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:50,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=519734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:56,915 INFO [train.py:968] (1/2) Epoch 12, batch 18550, giga_loss[loss=0.2822, simple_loss=0.3611, pruned_loss=0.1016, over 28663.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3512, pruned_loss=0.1025, over 5689333.96 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3385, pruned_loss=0.09233, over 5738764.50 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3498, pruned_loss=0.1023, over 5685516.80 frames. ], batch size: 284, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:50:58,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:50:58,892 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-06 06:51:16,156 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=519763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:51:25,179 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 18600, giga_loss[loss=0.2741, simple_loss=0.352, pruned_loss=0.09811, over 28765.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3536, pruned_loss=0.104, over 5696525.93 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3391, pruned_loss=0.09259, over 5739091.17 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3523, pruned_loss=0.1038, over 5692182.13 frames. ], batch size: 119, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:52:01,120 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 18650, giga_loss[loss=0.2736, simple_loss=0.3587, pruned_loss=0.09425, over 28483.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.357, pruned_loss=0.1051, over 5701013.01 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3392, pruned_loss=0.09253, over 5741499.44 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3561, pruned_loss=0.1052, over 5695089.89 frames. ], batch size: 60, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:52:45,246 INFO [optim.py:369] (1/2) 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 06:52:59,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:52:59,811 INFO [train.py:968] (1/2) Epoch 12, batch 18700, giga_loss[loss=0.2726, simple_loss=0.3616, pruned_loss=0.09175, over 28984.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3581, pruned_loss=0.1049, over 5701723.18 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3395, pruned_loss=0.09282, over 5738038.95 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3577, pruned_loss=0.1051, over 5698069.43 frames. ], batch size: 164, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:53:01,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=519894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:53:36,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 06:53:37,038 INFO [train.py:968] (1/2) Epoch 12, batch 18750, giga_loss[loss=0.2884, simple_loss=0.3636, pruned_loss=0.1066, over 28915.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3591, pruned_loss=0.1045, over 5699730.60 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3403, pruned_loss=0.09332, over 5731723.76 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3585, pruned_loss=0.1046, over 5702133.68 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:53:54,524 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:54:20,134 INFO [train.py:968] (1/2) Epoch 12, batch 18800, giga_loss[loss=0.3053, simple_loss=0.376, pruned_loss=0.1173, over 28922.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3598, pruned_loss=0.1043, over 5693611.32 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3406, pruned_loss=0.09344, over 5733144.99 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3593, pruned_loss=0.1043, over 5693902.48 frames. ], batch size: 227, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:54:24,333 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=519998.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 06:54:52,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4794, 1.7079, 1.7462, 1.5769], device='cuda:1'), covar=tensor([0.1308, 0.1469, 0.1645, 0.1464], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0725, 0.0671, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 06:54:55,638 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:54:57,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:54:57,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:54:59,502 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 18850, giga_loss[loss=0.2549, simple_loss=0.3389, pruned_loss=0.0854, over 28574.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3584, pruned_loss=0.1025, over 5698574.68 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3411, pruned_loss=0.09365, over 5734016.93 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3578, pruned_loss=0.1025, over 5697681.17 frames. ], batch size: 85, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:55:20,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:55:22,587 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:55:27,365 INFO [optim.py:369] (1/2) 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 06:55:34,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9624, 1.3629, 1.1208, 0.1113], device='cuda:1'), covar=tensor([0.2401, 0.1873, 0.2953, 0.3571], device='cuda:1'), in_proj_covar=tensor([0.1559, 0.1476, 0.1487, 0.1272], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 06:55:41,102 INFO [train.py:968] (1/2) Epoch 12, batch 18900, giga_loss[loss=0.3132, simple_loss=0.3836, pruned_loss=0.1214, over 28893.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.357, pruned_loss=0.1013, over 5701549.03 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3412, pruned_loss=0.09356, over 5734000.73 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3567, pruned_loss=0.1015, over 5699952.95 frames. ], batch size: 186, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:55:47,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4417, 2.8002, 1.9949, 1.7653], device='cuda:1'), covar=tensor([0.2139, 0.1611, 0.2055, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1636, 0.1575, 0.1696], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 06:55:49,886 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:55:51,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:56:08,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-06 06:56:17,991 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:56:23,660 INFO [train.py:968] (1/2) Epoch 12, batch 18950, giga_loss[loss=0.2917, simple_loss=0.3606, pruned_loss=0.1114, over 28718.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3594, pruned_loss=0.1047, over 5694200.38 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3416, pruned_loss=0.09362, over 5736565.25 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3591, pruned_loss=0.105, over 5690114.32 frames. ], batch size: 92, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 06:56:30,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=520149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:56:31,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3598, 1.4333, 1.3612, 1.5398], device='cuda:1'), covar=tensor([0.0741, 0.0338, 0.0305, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 06:56:56,231 INFO [optim.py:369] (1/2) 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 06:56:59,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-06 06:57:10,484 INFO [train.py:968] (1/2) Epoch 12, batch 19000, giga_loss[loss=0.2955, simple_loss=0.364, pruned_loss=0.1135, over 29079.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3612, pruned_loss=0.1082, over 5678891.15 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3419, pruned_loss=0.09371, over 5732212.15 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3614, pruned_loss=0.1088, over 5677650.03 frames. ], batch size: 128, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:57:11,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=520194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:57:48,718 INFO [train.py:968] (1/2) Epoch 12, batch 19050, giga_loss[loss=0.3007, simple_loss=0.3641, pruned_loss=0.1186, over 28547.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3603, pruned_loss=0.1084, over 5689893.63 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3416, pruned_loss=0.09344, over 5733303.37 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.361, pruned_loss=0.1094, over 5687360.51 frames. ], batch size: 336, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 06:58:14,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2052, 1.1806, 4.1434, 3.2479], device='cuda:1'), covar=tensor([0.1715, 0.2741, 0.0381, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0663, 0.0586, 0.0856, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 06:58:16,127 INFO [optim.py:369] (1/2) 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 06:58:25,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3760, 1.4470, 1.4530, 1.3853], device='cuda:1'), covar=tensor([0.1319, 0.1691, 0.1863, 0.1586], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0728, 0.0673, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 06:58:26,664 INFO [train.py:968] (1/2) Epoch 12, batch 19100, libri_loss[loss=0.2601, simple_loss=0.3358, pruned_loss=0.09222, over 29589.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.359, pruned_loss=0.1084, over 5681867.25 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3423, pruned_loss=0.0938, over 5721734.35 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3596, pruned_loss=0.1094, over 5688365.66 frames. ], batch size: 74, lr: 2.70e-03, grad_scale: 1.0
+2023-03-06 06:58:54,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-06 06:59:06,041 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:59:10,783 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 19150, giga_loss[loss=0.2979, simple_loss=0.3631, pruned_loss=0.1164, over 28934.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3575, pruned_loss=0.1079, over 5678785.81 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3424, pruned_loss=0.09376, over 5713718.98 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3581, pruned_loss=0.1089, over 5690231.84 frames. ], batch size: 227, lr: 2.70e-03, grad_scale: 1.0
+2023-03-06 06:59:35,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 06:59:38,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=520373.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 06:59:41,778 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-06 06:59:42,678 INFO [optim.py:369] (1/2) 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 06:59:54,359 INFO [train.py:968] (1/2) Epoch 12, batch 19200, giga_loss[loss=0.247, simple_loss=0.3306, pruned_loss=0.08166, over 28880.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3565, pruned_loss=0.1065, over 5668510.10 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3427, pruned_loss=0.09379, over 5707419.63 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.357, pruned_loss=0.1075, over 5682634.54 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:00:37,132 INFO [train.py:968] (1/2) Epoch 12, batch 19250, giga_loss[loss=0.2697, simple_loss=0.3435, pruned_loss=0.09788, over 28996.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3544, pruned_loss=0.1044, over 5669444.37 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3431, pruned_loss=0.09401, over 5702533.42 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3546, pruned_loss=0.1053, over 5683801.48 frames. ], batch size: 155, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:01:06,015 INFO [optim.py:369] (1/2) 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 07:01:19,714 INFO [train.py:968] (1/2) Epoch 12, batch 19300, giga_loss[loss=0.2531, simple_loss=0.3249, pruned_loss=0.09066, over 28557.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3502, pruned_loss=0.1018, over 5666069.36 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3431, pruned_loss=0.09397, over 5700849.66 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3508, pruned_loss=0.1029, over 5678216.43 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:01:24,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4186, 4.2150, 3.9900, 2.0949], device='cuda:1'), covar=tensor([0.0644, 0.0781, 0.0790, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1028, 0.0961, 0.0843, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 07:01:41,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3633, 1.9976, 1.5875, 0.5208], device='cuda:1'), covar=tensor([0.3812, 0.2037, 0.3018, 0.4610], device='cuda:1'), in_proj_covar=tensor([0.1558, 0.1478, 0.1483, 0.1274], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:01:41,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520516.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 07:01:45,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520519.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:01:49,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=520524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:02:07,072 INFO [train.py:968] (1/2) Epoch 12, batch 19350, giga_loss[loss=0.2362, simple_loss=0.3122, pruned_loss=0.08013, over 29055.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.344, pruned_loss=0.09855, over 5672906.99 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3431, pruned_loss=0.09396, over 5703564.64 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3446, pruned_loss=0.09949, over 5679750.27 frames. ], batch size: 128, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:02:12,946 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520548.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:02:22,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3899, 1.7075, 1.3633, 1.2438], device='cuda:1'), covar=tensor([0.2621, 0.2440, 0.2729, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.1308, 0.0970, 0.1156, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:02:39,842 INFO [optim.py:369] (1/2) 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 07:02:54,452 INFO [train.py:968] (1/2) Epoch 12, batch 19400, giga_loss[loss=0.2324, simple_loss=0.3088, pruned_loss=0.07798, over 28680.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3393, pruned_loss=0.09621, over 5677039.82 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3436, pruned_loss=0.09417, over 5708337.04 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3392, pruned_loss=0.09684, over 5677265.62 frames. ], batch size: 307, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:03:39,337 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 19450, giga_loss[loss=0.2536, simple_loss=0.3321, pruned_loss=0.08755, over 28718.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3382, pruned_loss=0.09543, over 5684298.39 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3438, pruned_loss=0.09424, over 5711256.25 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3379, pruned_loss=0.09588, over 5681536.55 frames. ], batch size: 66, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:04:03,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=520667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:04:06,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=520670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:04:07,047 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 12, batch 19500, giga_loss[loss=0.2531, simple_loss=0.3327, pruned_loss=0.08681, over 28886.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3393, pruned_loss=0.09554, over 5692657.17 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3442, pruned_loss=0.09436, over 5709327.39 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3386, pruned_loss=0.09582, over 5691751.44 frames. ], batch size: 227, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:04:32,452 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=520699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:05:07,017 INFO [train.py:968] (1/2) Epoch 12, batch 19550, giga_loss[loss=0.3288, simple_loss=0.3814, pruned_loss=0.1381, over 26595.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3396, pruned_loss=0.09519, over 5688364.78 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3451, pruned_loss=0.09447, over 5704476.29 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3379, pruned_loss=0.09533, over 5691865.37 frames. ], batch size: 555, lr: 2.70e-03, grad_scale: 2.0
+2023-03-06 07:05:32,858 INFO [optim.py:369] (1/2) 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 07:05:46,915 INFO [train.py:968] (1/2) Epoch 12, batch 19600, giga_loss[loss=0.2274, simple_loss=0.3068, pruned_loss=0.074, over 28824.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3372, pruned_loss=0.0938, over 5703070.11 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3456, pruned_loss=0.09459, over 5707379.43 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3354, pruned_loss=0.09381, over 5703001.74 frames. ], batch size: 145, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:06:06,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4971, 1.7322, 1.7463, 1.3282], device='cuda:1'), covar=tensor([0.1727, 0.2334, 0.1372, 0.1573], device='cuda:1'), in_proj_covar=tensor([0.0834, 0.0689, 0.0874, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:06:26,074 INFO [train.py:968] (1/2) Epoch 12, batch 19650, giga_loss[loss=0.2305, simple_loss=0.3043, pruned_loss=0.07829, over 28958.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3349, pruned_loss=0.09266, over 5706953.09 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3465, pruned_loss=0.09492, over 5703395.95 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3324, pruned_loss=0.09235, over 5710818.16 frames. ], batch size: 213, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:06:51,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4111, 2.1126, 1.5812, 0.7010], device='cuda:1'), covar=tensor([0.3446, 0.1855, 0.2982, 0.4348], device='cuda:1'), in_proj_covar=tensor([0.1555, 0.1472, 0.1486, 0.1269], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:06:55,571 INFO [optim.py:369] (1/2) 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 07:07:06,624 INFO [train.py:968] (1/2) Epoch 12, batch 19700, giga_loss[loss=0.2691, simple_loss=0.3371, pruned_loss=0.1005, over 28869.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3324, pruned_loss=0.09181, over 5711253.12 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3469, pruned_loss=0.09499, over 5705821.98 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3299, pruned_loss=0.09147, over 5712274.56 frames. ], batch size: 174, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:07:47,894 INFO [train.py:968] (1/2) Epoch 12, batch 19750, giga_loss[loss=0.2822, simple_loss=0.3418, pruned_loss=0.1113, over 28712.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3305, pruned_loss=0.09114, over 5719737.36 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3473, pruned_loss=0.09505, over 5712549.79 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3277, pruned_loss=0.0907, over 5714714.62 frames. ], batch size: 262, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:07:59,398 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6528, 1.7865, 1.8214, 1.4417], device='cuda:1'), covar=tensor([0.1560, 0.2122, 0.1213, 0.1403], device='cuda:1'), in_proj_covar=tensor([0.0835, 0.0687, 0.0875, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:08:15,742 INFO [optim.py:369] (1/2) 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 07:08:17,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=520979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:08:26,542 INFO [train.py:968] (1/2) Epoch 12, batch 19800, giga_loss[loss=0.2875, simple_loss=0.3462, pruned_loss=0.1143, over 28885.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3278, pruned_loss=0.08971, over 5725613.34 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3475, pruned_loss=0.09499, over 5717601.43 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.325, pruned_loss=0.08932, over 5717289.63 frames. ], batch size: 199, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:08:42,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:09:06,085 INFO [train.py:968] (1/2) Epoch 12, batch 19850, giga_loss[loss=0.2344, simple_loss=0.3125, pruned_loss=0.07814, over 28821.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3265, pruned_loss=0.08938, over 5723114.40 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3479, pruned_loss=0.09501, over 5722150.76 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3234, pruned_loss=0.08891, over 5712495.99 frames. ], batch size: 199, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:09:08,958 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:09:49,225 INFO [train.py:968] (1/2) Epoch 12, batch 19900, giga_loss[loss=0.2717, simple_loss=0.3363, pruned_loss=0.1035, over 27964.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3239, pruned_loss=0.08807, over 5727216.83 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3482, pruned_loss=0.095, over 5724833.00 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3209, pruned_loss=0.08763, over 5716528.64 frames. ], batch size: 412, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:10:27,040 INFO [train.py:968] (1/2) Epoch 12, batch 19950, giga_loss[loss=0.2555, simple_loss=0.3263, pruned_loss=0.09233, over 28907.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3218, pruned_loss=0.08678, over 5735964.07 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3482, pruned_loss=0.09489, over 5729354.24 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3188, pruned_loss=0.08637, over 5723758.65 frames. ], batch size: 186, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:10:39,002 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=521156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:10:39,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-06 07:10:41,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:10:46,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=521167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:10:51,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6636, 1.8737, 1.5641, 1.7333], device='cuda:1'), covar=tensor([0.2339, 0.2346, 0.2628, 0.2141], device='cuda:1'), in_proj_covar=tensor([0.1320, 0.0975, 0.1162, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:10:54,530 INFO [optim.py:369] (1/2) 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 07:11:02,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:11:02,771 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 20000, giga_loss[loss=0.2393, simple_loss=0.3091, pruned_loss=0.08473, over 29074.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3218, pruned_loss=0.08679, over 5737125.03 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3489, pruned_loss=0.09503, over 5730497.59 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3181, pruned_loss=0.0861, over 5726307.72 frames. ], batch size: 128, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:11:04,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:11:28,328 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:11:42,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1898, 3.7424, 1.4227, 1.3860], device='cuda:1'), covar=tensor([0.1213, 0.0391, 0.0995, 0.1556], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0504, 0.0341, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 07:11:46,899 INFO [train.py:968] (1/2) Epoch 12, batch 20050, giga_loss[loss=0.2951, simple_loss=0.3661, pruned_loss=0.112, over 28997.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3241, pruned_loss=0.0884, over 5736337.33 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3492, pruned_loss=0.09521, over 5729767.40 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3206, pruned_loss=0.08762, over 5728442.95 frames. ], batch size: 155, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:12:18,499 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 20100, giga_loss[loss=0.2751, simple_loss=0.3481, pruned_loss=0.101, over 28324.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3301, pruned_loss=0.09228, over 5722531.86 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3496, pruned_loss=0.09528, over 5731201.15 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3264, pruned_loss=0.09148, over 5714634.09 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:13:22,197 INFO [train.py:968] (1/2) Epoch 12, batch 20150, giga_loss[loss=0.3526, simple_loss=0.4039, pruned_loss=0.1507, over 27650.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3405, pruned_loss=0.09921, over 5703281.81 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3503, pruned_loss=0.09558, over 5732005.90 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3364, pruned_loss=0.09827, over 5695997.85 frames. ], batch size: 472, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:13:31,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:13:43,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3974, 1.6753, 1.2785, 1.5949], device='cuda:1'), covar=tensor([0.2268, 0.2178, 0.2432, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.0969, 0.1159, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:13:52,753 INFO [optim.py:369] (1/2) 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 07:13:54,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9885, 1.2956, 1.0073, 0.2496], device='cuda:1'), covar=tensor([0.2619, 0.2302, 0.3517, 0.4357], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1470, 0.1486, 0.1262], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:14:04,353 INFO [train.py:968] (1/2) Epoch 12, batch 20200, giga_loss[loss=0.2815, simple_loss=0.3519, pruned_loss=0.1055, over 28475.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3459, pruned_loss=0.1022, over 5696939.42 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3503, pruned_loss=0.09568, over 5728787.70 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3424, pruned_loss=0.1016, over 5692068.54 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:14:50,721 INFO [train.py:968] (1/2) Epoch 12, batch 20250, giga_loss[loss=0.3036, simple_loss=0.3852, pruned_loss=0.111, over 28883.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3507, pruned_loss=0.104, over 5683305.23 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3507, pruned_loss=0.09585, over 5731713.65 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3474, pruned_loss=0.1035, over 5676205.12 frames. ], batch size: 186, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:14:54,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 07:15:20,846 INFO [optim.py:369] (1/2) 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 07:15:28,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9161, 1.1886, 3.5784, 3.0515], device='cuda:1'), covar=tensor([0.1786, 0.2640, 0.0439, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0659, 0.0581, 0.0850, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:15:35,470 INFO [train.py:968] (1/2) Epoch 12, batch 20300, giga_loss[loss=0.3202, simple_loss=0.3861, pruned_loss=0.1272, over 28962.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3553, pruned_loss=0.1064, over 5677066.02 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3504, pruned_loss=0.09563, over 5725102.02 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3531, pruned_loss=0.1064, over 5675419.96 frames. ], batch size: 164, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:15:40,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=521497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:15:43,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:16:11,426 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:16:21,738 INFO [train.py:968] (1/2) Epoch 12, batch 20350, giga_loss[loss=0.3545, simple_loss=0.4091, pruned_loss=0.15, over 28158.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3616, pruned_loss=0.1104, over 5673373.14 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3504, pruned_loss=0.09561, over 5727053.81 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3599, pruned_loss=0.1107, over 5669389.69 frames. ], batch size: 368, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:16:21,987 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:16:49,086 INFO [optim.py:369] (1/2) 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 07:16:59,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=521587.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 07:17:03,964 INFO [train.py:968] (1/2) Epoch 12, batch 20400, giga_loss[loss=0.2531, simple_loss=0.3382, pruned_loss=0.08394, over 28426.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.358, pruned_loss=0.1076, over 5682261.43 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3505, pruned_loss=0.09568, over 5729945.43 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3569, pruned_loss=0.1082, over 5674623.04 frames. ], batch size: 60, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:17:45,634 INFO [train.py:968] (1/2) Epoch 12, batch 20450, giga_loss[loss=0.3156, simple_loss=0.386, pruned_loss=0.1226, over 28589.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3543, pruned_loss=0.1048, over 5688188.84 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3503, pruned_loss=0.09574, over 5733017.51 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3538, pruned_loss=0.1055, over 5677907.54 frames. ], batch size: 336, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:17:46,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1325, 1.5495, 1.4462, 1.0877], device='cuda:1'), covar=tensor([0.1567, 0.2153, 0.1316, 0.1553], device='cuda:1'), in_proj_covar=tensor([0.0834, 0.0689, 0.0875, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:17:47,326 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-06 07:17:48,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-06 07:18:15,925 INFO [optim.py:369] (1/2) 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 07:18:21,154 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=521685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:18:24,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=521688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:18:29,892 INFO [train.py:968] (1/2) Epoch 12, batch 20500, giga_loss[loss=0.2528, simple_loss=0.333, pruned_loss=0.08628, over 28869.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3519, pruned_loss=0.1026, over 5687155.18 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3504, pruned_loss=0.09585, over 5722910.02 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3515, pruned_loss=0.1032, over 5687458.63 frames. ], batch size: 112, lr: 2.70e-03, grad_scale: 8.0
+2023-03-06 07:18:51,053 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=521717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:19:11,721 INFO [train.py:968] (1/2) Epoch 12, batch 20550, giga_loss[loss=0.2875, simple_loss=0.3585, pruned_loss=0.1083, over 28003.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3533, pruned_loss=0.1032, over 5673984.37 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3502, pruned_loss=0.09581, over 5716612.12 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3531, pruned_loss=0.1039, over 5679748.56 frames. ], batch size: 77, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:19:43,102 INFO [optim.py:369] (1/2) 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 07:19:54,750 INFO [train.py:968] (1/2) Epoch 12, batch 20600, giga_loss[loss=0.3325, simple_loss=0.4008, pruned_loss=0.1321, over 28646.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3563, pruned_loss=0.1053, over 5682580.26 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3507, pruned_loss=0.09608, over 5719386.73 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3559, pruned_loss=0.1057, over 5683830.73 frames. ], batch size: 92, lr: 2.70e-03, grad_scale: 4.0
+2023-03-06 07:20:36,817 INFO [train.py:968] (1/2) Epoch 12, batch 20650, giga_loss[loss=0.2683, simple_loss=0.3425, pruned_loss=0.09708, over 28849.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3586, pruned_loss=0.107, over 5696667.03 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3513, pruned_loss=0.09636, over 5722615.76 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3578, pruned_loss=0.1074, over 5694096.07 frames. ], batch size: 145, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:20:43,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=521852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:21:08,673 INFO [optim.py:369] (1/2) 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 07:21:09,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3108, 4.1332, 3.9122, 2.0562], device='cuda:1'), covar=tensor([0.0582, 0.0719, 0.0681, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.1037, 0.0967, 0.0845, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 07:21:21,101 INFO [train.py:968] (1/2) Epoch 12, batch 20700, libri_loss[loss=0.2848, simple_loss=0.3716, pruned_loss=0.09902, over 28561.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3599, pruned_loss=0.1081, over 5680823.67 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3514, pruned_loss=0.09631, over 5717552.32 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3593, pruned_loss=0.1088, over 5681910.55 frames. ], batch size: 106, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:22:04,288 INFO [train.py:968] (1/2) Epoch 12, batch 20750, giga_loss[loss=0.3144, simple_loss=0.3833, pruned_loss=0.1227, over 29008.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3612, pruned_loss=0.1094, over 5687064.86 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3515, pruned_loss=0.09616, over 5724208.56 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3611, pruned_loss=0.1105, over 5680541.30 frames. ], batch size: 136, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:22:23,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=521962.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:22:35,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7359, 1.0292, 2.8260, 2.7403], device='cuda:1'), covar=tensor([0.1728, 0.2600, 0.0601, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0668, 0.0588, 0.0861, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:22:36,867 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 20800, giga_loss[loss=0.3317, simple_loss=0.3895, pruned_loss=0.1369, over 28964.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3605, pruned_loss=0.1091, over 5685173.48 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3515, pruned_loss=0.09612, over 5717747.69 frames. ], giga_tot_loss[loss=0.2905, simple_loss=0.3605, pruned_loss=0.1102, over 5685936.53 frames. ], batch size: 227, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:23:25,921 INFO [train.py:968] (1/2) Epoch 12, batch 20850, giga_loss[loss=0.272, simple_loss=0.3455, pruned_loss=0.09919, over 28848.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.36, pruned_loss=0.1079, over 5689375.91 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3515, pruned_loss=0.09607, over 5712279.53 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3602, pruned_loss=0.1091, over 5694750.55 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:23:55,118 INFO [optim.py:369] (1/2) 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 07:23:59,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=522083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:24:05,746 INFO [train.py:968] (1/2) Epoch 12, batch 20900, giga_loss[loss=0.2789, simple_loss=0.3624, pruned_loss=0.09772, over 28996.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3594, pruned_loss=0.1063, over 5692732.16 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3516, pruned_loss=0.09603, over 5715804.99 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3596, pruned_loss=0.1075, over 5693380.93 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:24:17,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=522105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:24:19,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=522108.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 07:24:43,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=522137.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 07:24:46,082 INFO [train.py:968] (1/2) Epoch 12, batch 20950, giga_loss[loss=0.2593, simple_loss=0.3423, pruned_loss=0.08809, over 28272.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3596, pruned_loss=0.1056, over 5694849.53 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3521, pruned_loss=0.0963, over 5717203.13 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3596, pruned_loss=0.1066, over 5693295.46 frames. ], batch size: 77, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:25:00,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8945, 2.0273, 1.6987, 2.0863], device='cuda:1'), covar=tensor([0.2235, 0.2219, 0.2399, 0.2187], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.0971, 0.1157, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:25:14,456 INFO [optim.py:369] (1/2) 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 07:25:26,185 INFO [train.py:968] (1/2) Epoch 12, batch 21000, giga_loss[loss=0.2491, simple_loss=0.3319, pruned_loss=0.08318, over 28987.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3573, pruned_loss=0.1044, over 5704151.61 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3519, pruned_loss=0.09613, over 5720958.21 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3576, pruned_loss=0.1055, over 5699272.47 frames. ], batch size: 128, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:25:26,185 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 07:25:34,736 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 07:25:38,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5340, 1.7564, 1.4019, 1.7682], device='cuda:1'), covar=tensor([0.2415, 0.2254, 0.2534, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.1314, 0.0972, 0.1158, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:26:00,387 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 21050, giga_loss[loss=0.2838, simple_loss=0.3582, pruned_loss=0.1048, over 28806.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3551, pruned_loss=0.1036, over 5709886.27 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3518, pruned_loss=0.09615, over 5725650.32 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3555, pruned_loss=0.1046, over 5701287.98 frames. ], batch size: 262, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:26:37,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-06 07:26:42,761 INFO [optim.py:369] (1/2) 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 07:26:51,020 INFO [train.py:968] (1/2) Epoch 12, batch 21100, giga_loss[loss=0.2529, simple_loss=0.3298, pruned_loss=0.08802, over 28512.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3535, pruned_loss=0.1027, over 5718698.17 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3517, pruned_loss=0.09613, over 5729755.50 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3539, pruned_loss=0.1038, over 5707818.12 frames. ], batch size: 65, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:27:35,309 INFO [train.py:968] (1/2) Epoch 12, batch 21150, giga_loss[loss=0.2636, simple_loss=0.3355, pruned_loss=0.09583, over 28480.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3537, pruned_loss=0.1038, over 5713174.53 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3519, pruned_loss=0.0963, over 5731720.85 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3539, pruned_loss=0.1046, over 5702750.09 frames. ], batch size: 71, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:27:44,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7202, 1.0401, 2.8635, 2.7341], device='cuda:1'), covar=tensor([0.1707, 0.2539, 0.0555, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0663, 0.0585, 0.0854, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:27:58,624 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=522370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:28:01,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=522373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:28:08,314 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 21200, giga_loss[loss=0.2617, simple_loss=0.3413, pruned_loss=0.09104, over 28216.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3547, pruned_loss=0.1041, over 5722498.18 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3523, pruned_loss=0.09653, over 5736255.84 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3545, pruned_loss=0.1046, over 5709695.76 frames. ], batch size: 77, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:28:23,606 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=522402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:28:32,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4819, 1.1389, 4.8041, 3.4302], device='cuda:1'), covar=tensor([0.1715, 0.2890, 0.0335, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0662, 0.0584, 0.0853, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:28:40,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6081, 1.8684, 1.8548, 1.4306], device='cuda:1'), covar=tensor([0.1738, 0.2459, 0.1431, 0.1689], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0694, 0.0876, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:28:53,959 INFO [train.py:968] (1/2) Epoch 12, batch 21250, giga_loss[loss=0.2602, simple_loss=0.345, pruned_loss=0.08771, over 28975.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3543, pruned_loss=0.1035, over 5711934.20 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3523, pruned_loss=0.09658, over 5732675.25 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3543, pruned_loss=0.1042, over 5703471.71 frames. ], batch size: 145, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:29:08,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=522458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:29:25,927 INFO [optim.py:369] (1/2) 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 07:29:34,542 INFO [train.py:968] (1/2) Epoch 12, batch 21300, giga_loss[loss=0.2763, simple_loss=0.3488, pruned_loss=0.1019, over 28503.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3523, pruned_loss=0.1011, over 5716801.57 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3522, pruned_loss=0.09652, over 5735561.19 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3524, pruned_loss=0.1019, over 5707065.83 frames. ], batch size: 71, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:29:58,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4847, 3.6304, 1.6275, 1.5624], device='cuda:1'), covar=tensor([0.0925, 0.0202, 0.0880, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0499, 0.0338, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0029, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 07:30:14,041 INFO [train.py:968] (1/2) Epoch 12, batch 21350, giga_loss[loss=0.278, simple_loss=0.3554, pruned_loss=0.1003, over 28925.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3527, pruned_loss=0.1016, over 5721284.98 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3528, pruned_loss=0.09732, over 5733687.50 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3522, pruned_loss=0.1017, over 5714523.28 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:30:45,325 INFO [optim.py:369] (1/2) 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 07:30:48,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2557, 1.4322, 1.4647, 1.3052], device='cuda:1'), covar=tensor([0.1451, 0.1474, 0.1977, 0.1617], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0723, 0.0668, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 07:30:54,481 INFO [train.py:968] (1/2) Epoch 12, batch 21400, giga_loss[loss=0.2784, simple_loss=0.3489, pruned_loss=0.1039, over 28972.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3509, pruned_loss=0.101, over 5722007.97 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3525, pruned_loss=0.0972, over 5734420.96 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3508, pruned_loss=0.1013, over 5716014.08 frames. ], batch size: 112, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:30:55,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0177, 1.3355, 1.0544, 0.2337], device='cuda:1'), covar=tensor([0.2774, 0.2136, 0.3809, 0.4651], device='cuda:1'), in_proj_covar=tensor([0.1546, 0.1449, 0.1474, 0.1258], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:31:01,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=522601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:31:03,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=522604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:31:26,925 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=522633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:31:33,179 INFO [train.py:968] (1/2) Epoch 12, batch 21450, giga_loss[loss=0.2379, simple_loss=0.3149, pruned_loss=0.08045, over 28544.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3485, pruned_loss=0.09983, over 5718380.20 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3523, pruned_loss=0.09714, over 5737239.99 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3485, pruned_loss=0.1002, over 5710778.67 frames. ], batch size: 60, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:32:05,833 INFO [optim.py:369] (1/2) 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 07:32:12,960 INFO [train.py:968] (1/2) Epoch 12, batch 21500, giga_loss[loss=0.2452, simple_loss=0.3216, pruned_loss=0.08441, over 28925.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3468, pruned_loss=0.09894, over 5727709.27 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3526, pruned_loss=0.09741, over 5741109.89 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3465, pruned_loss=0.09903, over 5717838.80 frames. ], batch size: 145, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:32:51,531 INFO [train.py:968] (1/2) Epoch 12, batch 21550, giga_loss[loss=0.2549, simple_loss=0.3308, pruned_loss=0.08946, over 28932.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3469, pruned_loss=0.09952, over 5732705.36 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3528, pruned_loss=0.0976, over 5745205.44 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3463, pruned_loss=0.09946, over 5720934.77 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:33:25,154 INFO [optim.py:369] (1/2) 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 07:33:34,018 INFO [train.py:968] (1/2) Epoch 12, batch 21600, giga_loss[loss=0.2702, simple_loss=0.342, pruned_loss=0.09923, over 28692.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3461, pruned_loss=0.09995, over 5720796.45 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3522, pruned_loss=0.09737, over 5740633.73 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.346, pruned_loss=0.1002, over 5715064.15 frames. ], batch size: 242, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:34:14,622 INFO [train.py:968] (1/2) Epoch 12, batch 21650, giga_loss[loss=0.2219, simple_loss=0.3055, pruned_loss=0.06914, over 28058.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.346, pruned_loss=0.1008, over 5713887.82 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3524, pruned_loss=0.09751, over 5739031.36 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3457, pruned_loss=0.1009, over 5710189.68 frames. ], batch size: 77, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:34:43,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3697, 1.6672, 1.2690, 1.4807], device='cuda:1'), covar=tensor([0.0715, 0.0293, 0.0332, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0052, 0.0090], device='cuda:1')
+2023-03-06 07:34:44,877 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 21700, giga_loss[loss=0.262, simple_loss=0.3358, pruned_loss=0.09405, over 28848.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3438, pruned_loss=0.1002, over 5714266.69 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3527, pruned_loss=0.09778, over 5740881.80 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3431, pruned_loss=0.1001, over 5709396.98 frames. ], batch size: 285, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:35:30,270 INFO [train.py:968] (1/2) Epoch 12, batch 21750, giga_loss[loss=0.2577, simple_loss=0.3242, pruned_loss=0.0956, over 28505.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3412, pruned_loss=0.09901, over 5714768.44 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3526, pruned_loss=0.09804, over 5742502.71 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3403, pruned_loss=0.09878, over 5707859.64 frames. ], batch size: 78, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:36:01,724 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 21800, giga_loss[loss=0.2531, simple_loss=0.3305, pruned_loss=0.08783, over 28999.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3401, pruned_loss=0.09854, over 5711464.30 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.353, pruned_loss=0.09851, over 5744955.88 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3389, pruned_loss=0.09794, over 5703425.25 frames. ], batch size: 136, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:36:46,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5793, 1.7732, 1.5079, 1.8594], device='cuda:1'), covar=tensor([0.2368, 0.2474, 0.2619, 0.2367], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.0969, 0.1152, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:36:50,743 INFO [train.py:968] (1/2) Epoch 12, batch 21850, libri_loss[loss=0.3268, simple_loss=0.3883, pruned_loss=0.1327, over 28584.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3421, pruned_loss=0.09923, over 5706659.79 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3539, pruned_loss=0.0992, over 5737981.17 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.34, pruned_loss=0.09811, over 5704936.90 frames. ], batch size: 106, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:36:57,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2543, 1.6231, 1.0091, 1.2200], device='cuda:1'), covar=tensor([0.1027, 0.0693, 0.1567, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0436, 0.0498, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:37:25,583 INFO [optim.py:369] (1/2) 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 07:37:28,546 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-06 07:37:32,590 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:37:32,969 INFO [train.py:968] (1/2) Epoch 12, batch 21900, giga_loss[loss=0.2755, simple_loss=0.3537, pruned_loss=0.09865, over 28944.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3446, pruned_loss=0.1003, over 5712461.16 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3541, pruned_loss=0.09961, over 5743219.74 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3425, pruned_loss=0.09909, over 5705792.51 frames. ], batch size: 213, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:38:06,033 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 07:38:15,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3509, 1.9017, 1.3448, 0.6570], device='cuda:1'), covar=tensor([0.4274, 0.2097, 0.3431, 0.4653], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1461, 0.1484, 0.1265], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:38:15,714 INFO [train.py:968] (1/2) Epoch 12, batch 21950, giga_loss[loss=0.268, simple_loss=0.3525, pruned_loss=0.09176, over 28618.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.347, pruned_loss=0.1008, over 5711691.09 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3541, pruned_loss=0.09986, over 5745111.44 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3451, pruned_loss=0.0996, over 5703971.66 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 07:38:34,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9346, 1.3203, 1.0787, 0.1382], device='cuda:1'), covar=tensor([0.2971, 0.2206, 0.3433, 0.4893], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1455, 0.1479, 0.1261], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:38:49,613 INFO [optim.py:369] (1/2) 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 07:38:57,226 INFO [train.py:968] (1/2) Epoch 12, batch 22000, giga_loss[loss=0.2719, simple_loss=0.3441, pruned_loss=0.09981, over 28955.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3477, pruned_loss=0.1004, over 5708999.65 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3543, pruned_loss=0.1001, over 5748664.25 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3459, pruned_loss=0.09928, over 5699046.25 frames. ], batch size: 213, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:39:08,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:39:18,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 07:39:39,209 INFO [train.py:968] (1/2) Epoch 12, batch 22050, giga_loss[loss=0.2826, simple_loss=0.3588, pruned_loss=0.1032, over 28642.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3475, pruned_loss=0.1003, over 5704906.62 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3547, pruned_loss=0.1006, over 5752324.08 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3455, pruned_loss=0.09894, over 5692526.59 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:40:13,892 INFO [optim.py:369] (1/2) 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 07:40:16,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6304, 1.6291, 1.2163, 1.2511], device='cuda:1'), covar=tensor([0.0720, 0.0566, 0.0957, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0437, 0.0497, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:40:21,078 INFO [train.py:968] (1/2) Epoch 12, batch 22100, giga_loss[loss=0.2709, simple_loss=0.3466, pruned_loss=0.09762, over 28972.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3475, pruned_loss=0.1004, over 5710053.89 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3547, pruned_loss=0.1006, over 5753734.29 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.346, pruned_loss=0.0993, over 5698548.58 frames. ], batch size: 136, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:40:48,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4252, 1.3260, 4.6902, 3.3470], device='cuda:1'), covar=tensor([0.1669, 0.2712, 0.0346, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0667, 0.0590, 0.0859, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:41:03,644 INFO [train.py:968] (1/2) Epoch 12, batch 22150, giga_loss[loss=0.2549, simple_loss=0.3355, pruned_loss=0.08718, over 28924.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3471, pruned_loss=0.1003, over 5707242.79 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3547, pruned_loss=0.1006, over 5753734.29 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3459, pruned_loss=0.09948, over 5698288.06 frames. ], batch size: 227, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:41:06,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-06 07:41:37,560 INFO [optim.py:369] (1/2) 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 07:41:44,968 INFO [train.py:968] (1/2) Epoch 12, batch 22200, giga_loss[loss=0.298, simple_loss=0.3759, pruned_loss=0.1101, over 28690.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3503, pruned_loss=0.1022, over 5711139.34 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3549, pruned_loss=0.1008, over 5756561.33 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.349, pruned_loss=0.1014, over 5700613.63 frames. ], batch size: 242, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:42:06,708 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:42:07,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2155, 1.5243, 1.2177, 1.4977], device='cuda:1'), covar=tensor([0.0742, 0.0328, 0.0332, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 07:42:27,796 INFO [train.py:968] (1/2) Epoch 12, batch 22250, giga_loss[loss=0.2914, simple_loss=0.3723, pruned_loss=0.1052, over 28753.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3528, pruned_loss=0.1034, over 5709684.93 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3551, pruned_loss=0.1009, over 5757086.53 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3516, pruned_loss=0.1027, over 5700728.37 frames. ], batch size: 242, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:42:45,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4066, 4.2345, 4.0039, 2.1376], device='cuda:1'), covar=tensor([0.0553, 0.0674, 0.0702, 0.1878], device='cuda:1'), in_proj_covar=tensor([0.1049, 0.0976, 0.0856, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 07:42:49,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=523466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:43:00,497 INFO [optim.py:369] (1/2) 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 07:43:08,418 INFO [train.py:968] (1/2) Epoch 12, batch 22300, giga_loss[loss=0.2885, simple_loss=0.3652, pruned_loss=0.1059, over 28992.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3545, pruned_loss=0.1044, over 5711939.57 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3557, pruned_loss=0.1013, over 5759507.66 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3531, pruned_loss=0.1035, over 5702111.94 frames. ], batch size: 213, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:43:47,064 INFO [train.py:968] (1/2) Epoch 12, batch 22350, giga_loss[loss=0.3148, simple_loss=0.3745, pruned_loss=0.1276, over 28826.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3555, pruned_loss=0.1048, over 5719438.74 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3563, pruned_loss=0.1018, over 5761859.00 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3538, pruned_loss=0.1038, over 5708216.95 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:43:56,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0706, 0.9020, 0.8647, 1.4493], device='cuda:1'), covar=tensor([0.0753, 0.0356, 0.0349, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0058, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 07:44:05,059 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:44:21,007 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 12, batch 22400, giga_loss[loss=0.2797, simple_loss=0.3592, pruned_loss=0.1001, over 28569.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3568, pruned_loss=0.1055, over 5718272.37 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3571, pruned_loss=0.1024, over 5763607.99 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3547, pruned_loss=0.1042, over 5706883.63 frames. ], batch size: 242, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:44:45,434 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=523609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:44:48,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=523612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:44:58,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6146, 1.7213, 1.9068, 1.4008], device='cuda:1'), covar=tensor([0.1772, 0.2152, 0.1424, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0689, 0.0870, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:45:12,072 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=523641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:45:12,410 INFO [train.py:968] (1/2) Epoch 12, batch 22450, giga_loss[loss=0.2958, simple_loss=0.3695, pruned_loss=0.111, over 28583.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3556, pruned_loss=0.1051, over 5715794.91 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3573, pruned_loss=0.1026, over 5765132.09 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3538, pruned_loss=0.104, over 5705127.52 frames. ], batch size: 336, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:45:14,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0102, 5.7905, 5.5313, 2.9084], device='cuda:1'), covar=tensor([0.0392, 0.0559, 0.0601, 0.1610], device='cuda:1'), in_proj_covar=tensor([0.1050, 0.0977, 0.0857, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 07:45:31,013 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:45:41,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-06 07:45:48,401 INFO [optim.py:369] (1/2) 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 07:45:57,008 INFO [train.py:968] (1/2) Epoch 12, batch 22500, giga_loss[loss=0.2687, simple_loss=0.3479, pruned_loss=0.09474, over 28811.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3521, pruned_loss=0.1032, over 5716134.33 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3574, pruned_loss=0.1026, over 5765873.43 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3506, pruned_loss=0.1022, over 5706888.98 frames. ], batch size: 284, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:46:16,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4197, 1.5746, 1.6742, 1.2663], device='cuda:1'), covar=tensor([0.1570, 0.2144, 0.1297, 0.1497], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0689, 0.0871, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:46:22,746 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=523724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:46:24,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=523727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:46:37,138 INFO [train.py:968] (1/2) Epoch 12, batch 22550, giga_loss[loss=0.2754, simple_loss=0.347, pruned_loss=0.1019, over 28311.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3482, pruned_loss=0.1011, over 5717451.27 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3576, pruned_loss=0.1029, over 5767322.14 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3468, pruned_loss=0.1001, over 5708556.02 frames. ], batch size: 368, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:46:48,569 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=523756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:47:00,161 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:47:11,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=523787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:47:15,656 INFO [train.py:968] (1/2) Epoch 12, batch 22600, giga_loss[loss=0.2433, simple_loss=0.3228, pruned_loss=0.08189, over 28878.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3484, pruned_loss=0.1011, over 5712182.46 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3588, pruned_loss=0.1041, over 5767531.77 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3459, pruned_loss=0.09916, over 5703181.34 frames. ], batch size: 145, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:47:15,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=523792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:47:59,212 INFO [train.py:968] (1/2) Epoch 12, batch 22650, giga_loss[loss=0.2696, simple_loss=0.3575, pruned_loss=0.09086, over 28726.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3496, pruned_loss=0.1001, over 5705490.56 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.359, pruned_loss=0.1043, over 5765412.73 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3474, pruned_loss=0.0983, over 5699731.47 frames. ], batch size: 242, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:48:09,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3823, 1.7972, 1.3478, 1.6638], device='cuda:1'), covar=tensor([0.2541, 0.2450, 0.2885, 0.2183], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.0972, 0.1154, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:48:32,803 INFO [optim.py:369] (1/2) 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 07:48:34,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0679, 1.4014, 1.3211, 1.0232], device='cuda:1'), covar=tensor([0.1291, 0.1747, 0.1125, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0689, 0.0871, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:48:39,444 INFO [train.py:968] (1/2) Epoch 12, batch 22700, giga_loss[loss=0.2788, simple_loss=0.3581, pruned_loss=0.09978, over 28718.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3519, pruned_loss=0.1012, over 5702864.12 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3593, pruned_loss=0.1047, over 5765603.71 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3497, pruned_loss=0.09935, over 5696723.89 frames. ], batch size: 262, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:49:13,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=523935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:49:15,585 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=523938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:49:16,187 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 22750, giga_loss[loss=0.2413, simple_loss=0.3096, pruned_loss=0.0865, over 28664.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3501, pruned_loss=0.1015, over 5702612.02 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3593, pruned_loss=0.1049, over 5768068.26 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3481, pruned_loss=0.09988, over 5694415.92 frames. ], batch size: 78, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:49:34,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1066, 1.1761, 3.6560, 3.1285], device='cuda:1'), covar=tensor([0.1621, 0.2584, 0.0432, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0664, 0.0589, 0.0863, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:49:40,382 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=523967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:49:54,351 INFO [optim.py:369] (1/2) 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 07:49:54,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3206, 1.6409, 1.2593, 1.3879], device='cuda:1'), covar=tensor([0.2314, 0.2151, 0.2540, 0.1931], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.0970, 0.1150, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 07:50:00,802 INFO [train.py:968] (1/2) Epoch 12, batch 22800, giga_loss[loss=0.2692, simple_loss=0.3407, pruned_loss=0.09889, over 29016.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3491, pruned_loss=0.1026, over 5706890.55 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3592, pruned_loss=0.1049, over 5768547.65 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3475, pruned_loss=0.1012, over 5699324.25 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:50:29,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7995, 1.7575, 1.3519, 1.4301], device='cuda:1'), covar=tensor([0.0716, 0.0523, 0.0999, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0437, 0.0496, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:50:36,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:50:40,036 INFO [train.py:968] (1/2) Epoch 12, batch 22850, giga_loss[loss=0.2684, simple_loss=0.3366, pruned_loss=0.1001, over 28872.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3485, pruned_loss=0.1035, over 5711396.50 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3599, pruned_loss=0.1056, over 5763464.92 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3462, pruned_loss=0.1017, over 5707567.95 frames. ], batch size: 186, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:50:47,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1487, 1.4705, 1.1861, 0.5056], device='cuda:1'), covar=tensor([0.2176, 0.1375, 0.1765, 0.4151], device='cuda:1'), in_proj_covar=tensor([0.1573, 0.1476, 0.1494, 0.1282], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:51:14,171 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:15,853 INFO [optim.py:369] (1/2) 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 07:51:16,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:17,294 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 22900, giga_loss[loss=0.3078, simple_loss=0.3527, pruned_loss=0.1314, over 28711.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3477, pruned_loss=0.1042, over 5711345.32 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3606, pruned_loss=0.1063, over 5767751.61 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3449, pruned_loss=0.102, over 5702527.93 frames. ], batch size: 92, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:51:37,123 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:54,217 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:51:57,057 INFO [train.py:968] (1/2) Epoch 12, batch 22950, libri_loss[loss=0.2709, simple_loss=0.3457, pruned_loss=0.09799, over 29589.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.348, pruned_loss=0.1039, over 5718584.58 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3612, pruned_loss=0.107, over 5764951.97 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3446, pruned_loss=0.1014, over 5711947.04 frames. ], batch size: 75, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:52:00,556 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:12,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:23,991 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:27,789 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:31,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:52:32,563 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 23000, giga_loss[loss=0.2678, simple_loss=0.3458, pruned_loss=0.0949, over 27877.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3429, pruned_loss=0.1013, over 5713958.99 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.361, pruned_loss=0.107, over 5765726.50 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3402, pruned_loss=0.09925, over 5707447.90 frames. ], batch size: 412, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:52:51,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:53:10,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5433, 2.2149, 1.6639, 0.8144], device='cuda:1'), covar=tensor([0.4403, 0.1995, 0.3169, 0.4770], device='cuda:1'), in_proj_covar=tensor([0.1567, 0.1467, 0.1485, 0.1271], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 07:53:16,420 INFO [train.py:968] (1/2) Epoch 12, batch 23050, giga_loss[loss=0.2293, simple_loss=0.3071, pruned_loss=0.07575, over 28981.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3391, pruned_loss=0.0995, over 5716104.48 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3612, pruned_loss=0.1074, over 5769502.03 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3363, pruned_loss=0.09736, over 5706250.67 frames. ], batch size: 213, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:53:49,924 INFO [optim.py:369] (1/2) 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 07:53:53,687 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 23100, giga_loss[loss=0.3105, simple_loss=0.3735, pruned_loss=0.1237, over 28768.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3368, pruned_loss=0.09783, over 5709528.10 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3615, pruned_loss=0.1077, over 5762430.28 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3339, pruned_loss=0.0957, over 5707944.41 frames. ], batch size: 199, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:53:56,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:05,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:07,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:18,633 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:54:19,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 07:54:31,016 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 23150, libri_loss[loss=0.2425, simple_loss=0.3129, pruned_loss=0.08601, over 27834.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3388, pruned_loss=0.0985, over 5718658.54 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3612, pruned_loss=0.1078, over 5767022.31 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3357, pruned_loss=0.09621, over 5710762.47 frames. ], batch size: 61, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:55:08,470 INFO [optim.py:369] (1/2) 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 07:55:15,480 INFO [train.py:968] (1/2) Epoch 12, batch 23200, giga_loss[loss=0.3187, simple_loss=0.3825, pruned_loss=0.1274, over 28267.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.341, pruned_loss=0.09902, over 5710037.93 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3613, pruned_loss=0.1079, over 5758204.88 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3382, pruned_loss=0.09701, over 5710195.68 frames. ], batch size: 368, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:55:35,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9077, 1.9256, 1.3855, 1.4687], device='cuda:1'), covar=tensor([0.0779, 0.0621, 0.1007, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0438, 0.0496, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:55:57,972 INFO [train.py:968] (1/2) Epoch 12, batch 23250, giga_loss[loss=0.2659, simple_loss=0.346, pruned_loss=0.09288, over 28687.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3448, pruned_loss=0.1003, over 5707105.72 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3616, pruned_loss=0.1082, over 5758731.73 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.342, pruned_loss=0.09842, over 5706219.55 frames. ], batch size: 307, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:56:15,410 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:56:34,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0578, 1.2599, 3.6281, 2.9382], device='cuda:1'), covar=tensor([0.1669, 0.2629, 0.0424, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0665, 0.0589, 0.0863, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:56:38,964 INFO [train.py:968] (1/2) Epoch 12, batch 23300, giga_loss[loss=0.2621, simple_loss=0.3455, pruned_loss=0.08936, over 28968.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3483, pruned_loss=0.1021, over 5707567.31 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3614, pruned_loss=0.1082, over 5761527.59 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.346, pruned_loss=0.1004, over 5703111.26 frames. ], batch size: 174, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 07:56:58,419 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 23350, giga_loss[loss=0.2949, simple_loss=0.358, pruned_loss=0.1159, over 28815.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3516, pruned_loss=0.1041, over 5699480.34 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3618, pruned_loss=0.1085, over 5764066.27 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3492, pruned_loss=0.1023, over 5691972.59 frames. ], batch size: 119, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:57:27,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=524550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:57:30,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4451, 1.5621, 1.3208, 1.6274], device='cuda:1'), covar=tensor([0.0719, 0.0308, 0.0318, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 07:57:30,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6573, 1.8766, 1.9358, 1.4615], device='cuda:1'), covar=tensor([0.1601, 0.2089, 0.1301, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.0834, 0.0693, 0.0875, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 07:57:57,344 INFO [optim.py:369] (1/2) 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 07:58:02,741 INFO [train.py:968] (1/2) Epoch 12, batch 23400, libri_loss[loss=0.307, simple_loss=0.379, pruned_loss=0.1176, over 20447.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.357, pruned_loss=0.1091, over 5678973.10 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3624, pruned_loss=0.1093, over 5744977.94 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.354, pruned_loss=0.1068, over 5687283.92 frames. ], batch size: 187, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:58:14,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:58:17,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:58:47,712 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:58:51,563 INFO [train.py:968] (1/2) Epoch 12, batch 23450, libri_loss[loss=0.2921, simple_loss=0.3575, pruned_loss=0.1134, over 29547.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3621, pruned_loss=0.1133, over 5678625.13 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3627, pruned_loss=0.1095, over 5746983.90 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3594, pruned_loss=0.1113, over 5682097.34 frames. ], batch size: 77, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:59:05,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:09,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:22,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1486, 1.2283, 3.4750, 2.9709], device='cuda:1'), covar=tensor([0.1538, 0.2560, 0.0455, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.0667, 0.0590, 0.0865, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 07:59:37,768 INFO [optim.py:369] (1/2) 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 07:59:39,406 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:42,943 INFO [train.py:968] (1/2) Epoch 12, batch 23500, giga_loss[loss=0.2956, simple_loss=0.3718, pruned_loss=0.1097, over 29037.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3698, pruned_loss=0.1188, over 5687069.01 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3624, pruned_loss=0.1094, over 5748427.98 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3679, pruned_loss=0.1174, over 5688056.63 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 07:59:45,069 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=524693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 07:59:48,912 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=524696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:00:11,185 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:00:17,309 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=524725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:00:33,479 INFO [train.py:968] (1/2) Epoch 12, batch 23550, giga_loss[loss=0.3141, simple_loss=0.3811, pruned_loss=0.1235, over 28706.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3765, pruned_loss=0.1251, over 5679426.43 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3629, pruned_loss=0.11, over 5751479.49 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3748, pruned_loss=0.1237, over 5676501.05 frames. ], batch size: 262, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:00:44,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3614, 1.2824, 4.0439, 3.2590], device='cuda:1'), covar=tensor([0.1558, 0.2516, 0.0396, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0662, 0.0584, 0.0857, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 08:01:17,538 INFO [optim.py:369] (1/2) 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 08:01:20,123 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=524788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:01:23,067 INFO [train.py:968] (1/2) Epoch 12, batch 23600, giga_loss[loss=0.3083, simple_loss=0.3772, pruned_loss=0.1196, over 28928.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3845, pruned_loss=0.1324, over 5667046.66 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3638, pruned_loss=0.1107, over 5754504.56 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3826, pruned_loss=0.131, over 5660844.62 frames. ], batch size: 112, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 08:01:51,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9580, 2.9366, 1.8377, 1.0132], device='cuda:1'), covar=tensor([0.5218, 0.2268, 0.3154, 0.5146], device='cuda:1'), in_proj_covar=tensor([0.1577, 0.1484, 0.1506, 0.1283], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 08:02:13,800 INFO [train.py:968] (1/2) Epoch 12, batch 23650, giga_loss[loss=0.465, simple_loss=0.4735, pruned_loss=0.2283, over 26541.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3901, pruned_loss=0.1367, over 5666694.31 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3642, pruned_loss=0.111, over 5756020.48 frames. ], giga_tot_loss[loss=0.3299, simple_loss=0.3885, pruned_loss=0.1356, over 5659241.74 frames. ], batch size: 555, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:02:56,195 INFO [optim.py:369] (1/2) 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 08:03:00,946 INFO [train.py:968] (1/2) Epoch 12, batch 23700, giga_loss[loss=0.3198, simple_loss=0.3793, pruned_loss=0.1302, over 28668.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3918, pruned_loss=0.1389, over 5657801.01 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3646, pruned_loss=0.1113, over 5751475.11 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.391, pruned_loss=0.1384, over 5654304.09 frames. ], batch size: 262, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:03:26,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1358, 1.1956, 3.8581, 3.1687], device='cuda:1'), covar=tensor([0.1715, 0.2632, 0.0411, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0666, 0.0589, 0.0862, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 08:03:50,962 INFO [train.py:968] (1/2) Epoch 12, batch 23750, giga_loss[loss=0.3984, simple_loss=0.4337, pruned_loss=0.1816, over 27794.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3925, pruned_loss=0.1405, over 5650404.66 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3643, pruned_loss=0.1112, over 5754633.00 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3928, pruned_loss=0.1409, over 5642745.43 frames. ], batch size: 412, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:04:03,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4386, 4.0688, 1.6686, 1.6638], device='cuda:1'), covar=tensor([0.0924, 0.0265, 0.0814, 0.1215], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0510, 0.0341, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 08:04:10,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2211, 0.8336, 0.8848, 1.3450], device='cuda:1'), covar=tensor([0.0727, 0.0375, 0.0331, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 08:04:32,975 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 23800, giga_loss[loss=0.3, simple_loss=0.3683, pruned_loss=0.1159, over 29049.00 frames. ], tot_loss[loss=0.3383, simple_loss=0.3934, pruned_loss=0.1416, over 5651935.62 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3649, pruned_loss=0.1121, over 5757968.51 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3942, pruned_loss=0.1424, over 5638125.32 frames. ], batch size: 128, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:05:33,251 INFO [train.py:968] (1/2) Epoch 12, batch 23850, libri_loss[loss=0.3152, simple_loss=0.3766, pruned_loss=0.1268, over 25826.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3973, pruned_loss=0.1451, over 5642415.51 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3652, pruned_loss=0.1124, over 5757498.75 frames. ], giga_tot_loss[loss=0.3454, simple_loss=0.3984, pruned_loss=0.1462, over 5629575.29 frames. ], batch size: 136, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:05:52,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2889, 1.4819, 3.3537, 3.1066], device='cuda:1'), covar=tensor([0.1284, 0.2131, 0.0433, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0668, 0.0591, 0.0867, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 08:06:23,088 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 23900, giga_loss[loss=0.3591, simple_loss=0.4085, pruned_loss=0.1548, over 28938.00 frames. ], tot_loss[loss=0.3469, simple_loss=0.3984, pruned_loss=0.1476, over 5610985.09 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3654, pruned_loss=0.1127, over 5750751.91 frames. ], giga_tot_loss[loss=0.3489, simple_loss=0.3999, pruned_loss=0.149, over 5603865.12 frames. ], batch size: 227, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:06:29,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:06:31,683 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:06:56,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2372, 4.0446, 3.8788, 1.8699], device='cuda:1'), covar=tensor([0.0596, 0.0769, 0.0745, 0.2146], device='cuda:1'), in_proj_covar=tensor([0.1075, 0.1004, 0.0875, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 08:07:18,753 INFO [train.py:968] (1/2) Epoch 12, batch 23950, giga_loss[loss=0.3227, simple_loss=0.3824, pruned_loss=0.1315, over 29045.00 frames. ], tot_loss[loss=0.3443, simple_loss=0.3962, pruned_loss=0.1462, over 5625048.23 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3657, pruned_loss=0.1129, over 5751230.35 frames. ], giga_tot_loss[loss=0.3464, simple_loss=0.3976, pruned_loss=0.1476, over 5616699.35 frames. ], batch size: 155, lr: 2.69e-03, grad_scale: 2.0
+2023-03-06 08:07:40,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:07:47,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 08:08:02,989 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 24000, giga_loss[loss=0.4133, simple_loss=0.4246, pruned_loss=0.201, over 23517.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.3943, pruned_loss=0.1446, over 5627867.04 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3659, pruned_loss=0.1132, over 5753099.47 frames. ], giga_tot_loss[loss=0.3442, simple_loss=0.3959, pruned_loss=0.1462, over 5616898.06 frames. ], batch size: 705, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:08:07,015 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 08:08:14,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2964, 1.6070, 1.2778, 0.5138], device='cuda:1'), covar=tensor([0.2850, 0.2050, 0.2814, 0.3924], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1495, 0.1503, 0.1283], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 08:08:15,697 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 08:08:25,434 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:08:57,322 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:09:00,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:09:02,689 INFO [train.py:968] (1/2) Epoch 12, batch 24050, giga_loss[loss=0.3101, simple_loss=0.3839, pruned_loss=0.1182, over 28912.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3944, pruned_loss=0.1431, over 5621514.98 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3664, pruned_loss=0.1137, over 5746628.20 frames. ], giga_tot_loss[loss=0.3426, simple_loss=0.3959, pruned_loss=0.1446, over 5616331.51 frames. ], batch size: 174, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:09:29,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:09:50,760 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 24100, giga_loss[loss=0.3005, simple_loss=0.3682, pruned_loss=0.1164, over 28865.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3965, pruned_loss=0.1446, over 5621914.84 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3661, pruned_loss=0.1137, over 5747202.21 frames. ], giga_tot_loss[loss=0.3458, simple_loss=0.3986, pruned_loss=0.1465, over 5614525.88 frames. ], batch size: 66, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:10:08,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:10:10,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:10:22,134 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525319.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 08:10:39,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:10:42,362 INFO [train.py:968] (1/2) Epoch 12, batch 24150, giga_loss[loss=0.3815, simple_loss=0.4196, pruned_loss=0.1717, over 27374.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3952, pruned_loss=0.1432, over 5607696.99 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3661, pruned_loss=0.1136, over 5722872.36 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3978, pruned_loss=0.1457, over 5619285.83 frames. ], batch size: 472, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:10:44,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5740, 1.9091, 1.4302, 1.5457], device='cuda:1'), covar=tensor([0.0739, 0.0275, 0.0318, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0178, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 08:10:59,401 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:11:06,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:11:27,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3009, 3.1489, 1.3336, 1.4652], device='cuda:1'), covar=tensor([0.0958, 0.0360, 0.0933, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0513, 0.0343, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 08:11:29,572 INFO [optim.py:369] (1/2) 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 08:11:32,634 INFO [train.py:968] (1/2) Epoch 12, batch 24200, giga_loss[loss=0.3055, simple_loss=0.3873, pruned_loss=0.1119, over 28866.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3926, pruned_loss=0.14, over 5605413.73 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3661, pruned_loss=0.1138, over 5714629.20 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3949, pruned_loss=0.1421, over 5619847.42 frames. ], batch size: 174, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:12:26,605 INFO [train.py:968] (1/2) Epoch 12, batch 24250, giga_loss[loss=0.3171, simple_loss=0.357, pruned_loss=0.1386, over 23727.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3891, pruned_loss=0.1366, over 5607602.50 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3658, pruned_loss=0.1136, over 5714717.06 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3914, pruned_loss=0.1386, over 5618329.00 frames. ], batch size: 705, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:12:51,615 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:13:05,682 INFO [optim.py:369] (1/2) 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 08:13:10,606 INFO [train.py:968] (1/2) Epoch 12, batch 24300, giga_loss[loss=0.2794, simple_loss=0.3564, pruned_loss=0.1012, over 28998.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.384, pruned_loss=0.1324, over 5624028.94 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3654, pruned_loss=0.1138, over 5719089.57 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.387, pruned_loss=0.1347, over 5624879.68 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:14:00,095 INFO [train.py:968] (1/2) Epoch 12, batch 24350, libri_loss[loss=0.3469, simple_loss=0.4133, pruned_loss=0.1402, over 29117.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3812, pruned_loss=0.1305, over 5632985.94 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3655, pruned_loss=0.114, over 5720807.85 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3838, pruned_loss=0.1324, over 5630418.09 frames. ], batch size: 101, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:14:22,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:14:37,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:14:46,012 INFO [optim.py:369] (1/2) 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 08:14:48,819 INFO [train.py:968] (1/2) Epoch 12, batch 24400, giga_loss[loss=0.3051, simple_loss=0.3737, pruned_loss=0.1182, over 28874.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3819, pruned_loss=0.1315, over 5631775.98 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3655, pruned_loss=0.114, over 5722691.88 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3842, pruned_loss=0.1332, over 5627010.36 frames. ], batch size: 186, lr: 2.69e-03, grad_scale: 8.0
+2023-03-06 08:15:13,844 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:15:17,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:15:20,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5750, 1.6055, 1.8349, 1.4068], device='cuda:1'), covar=tensor([0.1363, 0.2252, 0.1115, 0.1474], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0691, 0.0866, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 08:15:44,908 INFO [train.py:968] (1/2) Epoch 12, batch 24450, giga_loss[loss=0.3033, simple_loss=0.3695, pruned_loss=0.1185, over 28609.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3814, pruned_loss=0.1309, over 5630524.62 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3657, pruned_loss=0.1142, over 5716176.90 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3834, pruned_loss=0.1325, over 5629961.50 frames. ], batch size: 85, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:15:46,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-06 08:15:47,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:16:21,712 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:16:33,740 INFO [train.py:968] (1/2) Epoch 12, batch 24500, giga_loss[loss=0.3389, simple_loss=0.4075, pruned_loss=0.1352, over 28951.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3786, pruned_loss=0.1279, over 5649208.12 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3657, pruned_loss=0.1144, over 5716252.52 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3807, pruned_loss=0.1295, over 5645557.02 frames. ], batch size: 164, lr: 2.69e-03, grad_scale: 4.0
+2023-03-06 08:16:37,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525694.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 08:16:44,816 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=525702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:06,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:09,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:18,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:17:23,322 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 24550, giga_loss[loss=0.3325, simple_loss=0.4045, pruned_loss=0.1303, over 28901.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.379, pruned_loss=0.1257, over 5656186.70 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3655, pruned_loss=0.1144, over 5719753.85 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3812, pruned_loss=0.1273, over 5648398.23 frames. ], batch size: 213, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:17:38,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2123, 1.4267, 1.2033, 1.4467], device='cuda:1'), covar=tensor([0.0740, 0.0328, 0.0320, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 08:17:40,980 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:18:18,059 INFO [train.py:968] (1/2) Epoch 12, batch 24600, giga_loss[loss=0.3027, simple_loss=0.3723, pruned_loss=0.1166, over 28973.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3819, pruned_loss=0.127, over 5662560.47 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3657, pruned_loss=0.1148, over 5724407.87 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3839, pruned_loss=0.1282, over 5650618.20 frames. ], batch size: 164, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:19:02,273 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525837.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 08:19:06,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525840.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 08:19:06,936 INFO [train.py:968] (1/2) Epoch 12, batch 24650, giga_loss[loss=0.3189, simple_loss=0.3875, pruned_loss=0.1252, over 28307.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3833, pruned_loss=0.1287, over 5666657.99 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.366, pruned_loss=0.1151, over 5727096.45 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.385, pruned_loss=0.1297, over 5653504.77 frames. ], batch size: 368, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:19:16,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5331, 1.5913, 1.2668, 1.1848], device='cuda:1'), covar=tensor([0.0807, 0.0548, 0.0992, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0442, 0.0501, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 08:19:31,588 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525869.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 08:19:36,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:39,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:42,702 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=525883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:45,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=525886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:19:48,693 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 24700, giga_loss[loss=0.3115, simple_loss=0.3821, pruned_loss=0.1205, over 29052.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3831, pruned_loss=0.1287, over 5676562.42 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3659, pruned_loss=0.1152, over 5721206.64 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3849, pruned_loss=0.1297, over 5669409.13 frames. ], batch size: 155, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:20:08,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3967, 1.5959, 1.2948, 1.2960], device='cuda:1'), covar=tensor([0.2376, 0.2451, 0.2677, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1322, 0.0982, 0.1165, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 08:20:09,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:20:14,227 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=525915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:20:38,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=525940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:20:40,454 INFO [train.py:968] (1/2) Epoch 12, batch 24750, giga_loss[loss=0.3261, simple_loss=0.3776, pruned_loss=0.1373, over 27891.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3802, pruned_loss=0.128, over 5684632.87 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3657, pruned_loss=0.1155, over 5726718.66 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3822, pruned_loss=0.1289, over 5673047.78 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:20:41,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2359, 1.2847, 1.0785, 1.2165], device='cuda:1'), covar=tensor([0.1359, 0.1427, 0.1105, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.1721, 0.1640, 0.1577, 0.1695], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 08:21:17,052 INFO [optim.py:369] (1/2) 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 08:21:18,376 INFO [train.py:968] (1/2) Epoch 12, batch 24800, giga_loss[loss=0.3341, simple_loss=0.4007, pruned_loss=0.1338, over 28866.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3792, pruned_loss=0.1283, over 5690867.90 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3655, pruned_loss=0.1155, over 5734554.04 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3818, pruned_loss=0.1298, over 5671893.56 frames. ], batch size: 174, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:22:04,260 INFO [train.py:968] (1/2) Epoch 12, batch 24850, libri_loss[loss=0.3264, simple_loss=0.3936, pruned_loss=0.1296, over 29644.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3795, pruned_loss=0.1282, over 5678855.89 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3658, pruned_loss=0.1159, over 5726054.87 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3816, pruned_loss=0.1292, over 5669726.47 frames. ], batch size: 88, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:22:12,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:22:36,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:22:37,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5761, 1.7028, 1.1816, 1.3422], device='cuda:1'), covar=tensor([0.0818, 0.0601, 0.1066, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0442, 0.0502, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 08:22:40,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:22:43,004 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:22:48,122 INFO [train.py:968] (1/2) Epoch 12, batch 24900, giga_loss[loss=0.3122, simple_loss=0.3838, pruned_loss=0.1203, over 28960.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3774, pruned_loss=0.1247, over 5689920.48 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3658, pruned_loss=0.1158, over 5729411.12 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3794, pruned_loss=0.1259, over 5678538.10 frames. ], batch size: 106, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:23:12,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:23:13,107 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 08:23:35,072 INFO [train.py:968] (1/2) Epoch 12, batch 24950, giga_loss[loss=0.2857, simple_loss=0.3563, pruned_loss=0.1076, over 28819.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3779, pruned_loss=0.125, over 5684479.79 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3659, pruned_loss=0.1159, over 5732131.49 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3798, pruned_loss=0.126, over 5671886.85 frames. ], batch size: 119, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:23:37,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3182, 1.5416, 1.4988, 1.4672], device='cuda:1'), covar=tensor([0.0758, 0.0330, 0.0299, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 08:24:17,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:23,443 INFO [optim.py:369] (1/2) 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 08:24:25,384 INFO [train.py:968] (1/2) Epoch 12, batch 25000, giga_loss[loss=0.4051, simple_loss=0.4313, pruned_loss=0.1895, over 26703.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3756, pruned_loss=0.1235, over 5686461.94 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3658, pruned_loss=0.1159, over 5733507.92 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3773, pruned_loss=0.1244, over 5674346.72 frames. ], batch size: 555, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:24:26,726 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:28,483 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:52,212 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:54,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:24:58,758 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:25:01,367 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-06 08:25:10,717 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 25050, giga_loss[loss=0.3523, simple_loss=0.3847, pruned_loss=0.16, over 23461.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3758, pruned_loss=0.1241, over 5685205.80 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3657, pruned_loss=0.1158, over 5735728.95 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3776, pruned_loss=0.1253, over 5671929.71 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:25:23,454 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:25:58,914 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 25100, libri_loss[loss=0.3105, simple_loss=0.3665, pruned_loss=0.1272, over 29590.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1238, over 5673122.73 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.366, pruned_loss=0.116, over 5735164.54 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3756, pruned_loss=0.1247, over 5661976.52 frames. ], batch size: 74, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:26:36,426 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=526329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:26:48,049 INFO [train.py:968] (1/2) Epoch 12, batch 25150, giga_loss[loss=0.3051, simple_loss=0.3701, pruned_loss=0.12, over 28725.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1248, over 5676048.16 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3662, pruned_loss=0.1161, over 5735957.71 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3755, pruned_loss=0.1254, over 5666238.84 frames. ], batch size: 262, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:27:05,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2280, 1.8126, 1.4974, 1.4726], device='cuda:1'), covar=tensor([0.0735, 0.0331, 0.0295, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 08:27:37,567 INFO [optim.py:369] (1/2) 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 08:27:38,950 INFO [train.py:968] (1/2) Epoch 12, batch 25200, giga_loss[loss=0.2839, simple_loss=0.356, pruned_loss=0.106, over 28859.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.1251, over 5666563.81 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3666, pruned_loss=0.1165, over 5727484.99 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 5665395.35 frames. ], batch size: 145, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:28:25,201 INFO [train.py:968] (1/2) Epoch 12, batch 25250, giga_loss[loss=0.2857, simple_loss=0.3494, pruned_loss=0.111, over 28790.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3744, pruned_loss=0.1263, over 5655656.39 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3669, pruned_loss=0.1168, over 5720293.70 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3746, pruned_loss=0.1264, over 5660508.41 frames. ], batch size: 99, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:28:26,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5351, 1.5851, 1.7864, 1.3698], device='cuda:1'), covar=tensor([0.1498, 0.2176, 0.1217, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.0827, 0.0693, 0.0868, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 08:29:12,642 INFO [optim.py:369] (1/2) 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 08:29:13,236 INFO [train.py:968] (1/2) Epoch 12, batch 25300, giga_loss[loss=0.2986, simple_loss=0.3668, pruned_loss=0.1152, over 28699.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3744, pruned_loss=0.1258, over 5660451.02 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3671, pruned_loss=0.117, over 5723034.42 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3746, pruned_loss=0.1259, over 5660701.02 frames. ], batch size: 262, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:30:00,600 INFO [train.py:968] (1/2) Epoch 12, batch 25350, giga_loss[loss=0.2849, simple_loss=0.3641, pruned_loss=0.1028, over 28740.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3746, pruned_loss=0.1252, over 5664074.09 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3671, pruned_loss=0.1172, over 5725719.24 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3748, pruned_loss=0.1252, over 5660732.19 frames. ], batch size: 92, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:30:16,976 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:30:24,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:30:43,009 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-06 08:30:46,708 INFO [optim.py:369] (1/2) 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 08:30:47,416 INFO [train.py:968] (1/2) Epoch 12, batch 25400, giga_loss[loss=0.2517, simple_loss=0.3314, pruned_loss=0.08601, over 28895.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1236, over 5661697.86 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3672, pruned_loss=0.1173, over 5726882.37 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3736, pruned_loss=0.1236, over 5657540.52 frames. ], batch size: 86, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:31:11,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:31:34,669 INFO [train.py:968] (1/2) Epoch 12, batch 25450, giga_loss[loss=0.3116, simple_loss=0.3764, pruned_loss=0.1234, over 28811.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5663908.02 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.367, pruned_loss=0.1171, over 5729281.42 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1235, over 5657599.09 frames. ], batch size: 199, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:31:44,838 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:32:17,611 INFO [train.py:968] (1/2) Epoch 12, batch 25500, libri_loss[loss=0.3122, simple_loss=0.3572, pruned_loss=0.1336, over 29648.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3752, pruned_loss=0.1259, over 5666309.30 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5733912.32 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1257, over 5653604.19 frames. ], batch size: 69, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:32:28,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:30,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=526704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:31,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:38,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:40,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:32:58,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:04,827 INFO [train.py:968] (1/2) Epoch 12, batch 25550, giga_loss[loss=0.2977, simple_loss=0.3593, pruned_loss=0.118, over 28938.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3769, pruned_loss=0.1281, over 5670721.77 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.367, pruned_loss=0.1179, over 5739081.40 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1282, over 5653805.98 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:33:06,854 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:23,923 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:27,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:53,541 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:33:55,869 INFO [optim.py:369] (1/2) 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 08:33:55,882 INFO [train.py:968] (1/2) Epoch 12, batch 25600, giga_loss[loss=0.3824, simple_loss=0.4162, pruned_loss=0.1743, over 28964.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3784, pruned_loss=0.1303, over 5674521.30 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5741198.20 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3791, pruned_loss=0.1306, over 5658143.72 frames. ], batch size: 213, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:34:11,890 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.02 vs. limit=2.0
+2023-03-06 08:34:48,307 INFO [train.py:968] (1/2) Epoch 12, batch 25650, giga_loss[loss=0.3444, simple_loss=0.3948, pruned_loss=0.147, over 28641.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3813, pruned_loss=0.1336, over 5655532.36 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3675, pruned_loss=0.1184, over 5739476.98 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3817, pruned_loss=0.1336, over 5642432.35 frames. ], batch size: 336, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:34:53,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=526847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:34:56,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=526850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:35:15,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2925, 1.4096, 1.3485, 1.5292], device='cuda:1'), covar=tensor([0.0743, 0.0376, 0.0322, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 08:35:22,544 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=526879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:35:34,354 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 25700, giga_loss[loss=0.3023, simple_loss=0.3566, pruned_loss=0.124, over 28832.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3798, pruned_loss=0.1323, over 5662951.71 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3676, pruned_loss=0.1185, over 5740303.60 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3802, pruned_loss=0.1324, over 5651626.51 frames. ], batch size: 99, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:36:05,341 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 08:36:17,507 INFO [train.py:968] (1/2) Epoch 12, batch 25750, libri_loss[loss=0.2981, simple_loss=0.3685, pruned_loss=0.1138, over 29221.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3802, pruned_loss=0.1323, over 5666650.35 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3685, pruned_loss=0.1193, over 5743990.92 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3802, pruned_loss=0.1322, over 5650639.79 frames. ], batch size: 97, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:36:31,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1900, 1.1913, 1.0873, 0.9405], device='cuda:1'), covar=tensor([0.0827, 0.0556, 0.1077, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0442, 0.0502, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 08:36:56,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7843, 1.8436, 1.3287, 1.5402], device='cuda:1'), covar=tensor([0.0688, 0.0484, 0.0930, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0442, 0.0502, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 08:37:00,463 INFO [optim.py:369] (1/2) 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 08:37:00,480 INFO [train.py:968] (1/2) Epoch 12, batch 25800, giga_loss[loss=0.3124, simple_loss=0.3969, pruned_loss=0.114, over 28836.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3789, pruned_loss=0.1295, over 5681056.06 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3688, pruned_loss=0.1195, over 5748400.16 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3789, pruned_loss=0.1295, over 5662573.69 frames. ], batch size: 112, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:37:28,861 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:37:38,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:37:47,775 INFO [train.py:968] (1/2) Epoch 12, batch 25850, giga_loss[loss=0.3418, simple_loss=0.3887, pruned_loss=0.1474, over 27579.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1267, over 5674052.13 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3687, pruned_loss=0.1194, over 5750315.72 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3755, pruned_loss=0.1269, over 5656556.41 frames. ], batch size: 472, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:38:34,375 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 25900, giga_loss[loss=0.3431, simple_loss=0.3901, pruned_loss=0.148, over 28305.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3729, pruned_loss=0.1254, over 5679626.75 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3689, pruned_loss=0.1195, over 5752374.22 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.373, pruned_loss=0.1256, over 5663050.75 frames. ], batch size: 368, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:39:26,651 INFO [train.py:968] (1/2) Epoch 12, batch 25950, giga_loss[loss=0.2776, simple_loss=0.347, pruned_loss=0.1041, over 29086.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3706, pruned_loss=0.124, over 5682813.96 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3688, pruned_loss=0.1195, over 5744892.43 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3707, pruned_loss=0.1242, over 5674998.56 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:39:55,064 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:39:58,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:40:12,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 08:40:15,666 INFO [optim.py:369] (1/2) 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 08:40:15,678 INFO [train.py:968] (1/2) Epoch 12, batch 26000, giga_loss[loss=0.3537, simple_loss=0.4015, pruned_loss=0.1529, over 27517.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3729, pruned_loss=0.125, over 5673646.86 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3688, pruned_loss=0.1195, over 5738142.75 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3731, pruned_loss=0.1253, over 5671413.10 frames. ], batch size: 472, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:40:16,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-06 08:40:23,126 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:40:26,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=527205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:40:41,086 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3509, 1.6546, 1.2834, 1.5079], device='cuda:1'), covar=tensor([0.2560, 0.2497, 0.2891, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.1321, 0.0979, 0.1164, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 08:41:00,572 INFO [train.py:968] (1/2) Epoch 12, batch 26050, giga_loss[loss=0.3197, simple_loss=0.3902, pruned_loss=0.1246, over 28937.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.378, pruned_loss=0.1263, over 5665554.39 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3694, pruned_loss=0.1199, over 5722284.81 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3777, pruned_loss=0.1262, over 5675963.83 frames. ], batch size: 186, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:41:33,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=527275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:41:47,605 INFO [train.py:968] (1/2) Epoch 12, batch 26100, giga_loss[loss=0.3636, simple_loss=0.4144, pruned_loss=0.1564, over 28661.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3797, pruned_loss=0.1253, over 5670773.73 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3693, pruned_loss=0.12, over 5725110.66 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3798, pruned_loss=0.1253, over 5675687.10 frames. ], batch size: 307, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:41:50,115 INFO [optim.py:369] (1/2) 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 08:42:27,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-06 08:42:36,686 INFO [train.py:968] (1/2) Epoch 12, batch 26150, libri_loss[loss=0.2822, simple_loss=0.3413, pruned_loss=0.1115, over 29673.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3811, pruned_loss=0.1271, over 5677452.18 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1198, over 5727725.63 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3821, pruned_loss=0.1274, over 5677388.71 frames. ], batch size: 73, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:43:10,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4694, 4.3157, 4.0447, 1.7962], device='cuda:1'), covar=tensor([0.0540, 0.0687, 0.0696, 0.2237], device='cuda:1'), in_proj_covar=tensor([0.1081, 0.1013, 0.0883, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 08:43:20,650 INFO [train.py:968] (1/2) Epoch 12, batch 26200, giga_loss[loss=0.338, simple_loss=0.3973, pruned_loss=0.1394, over 28896.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3834, pruned_loss=0.1294, over 5668783.84 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3693, pruned_loss=0.1205, over 5716336.19 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3839, pruned_loss=0.1293, over 5677737.75 frames. ], batch size: 199, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:43:21,995 INFO [optim.py:369] (1/2) 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 08:43:23,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:43:40,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3667, 1.4966, 3.2373, 3.1490], device='cuda:1'), covar=tensor([0.1246, 0.2188, 0.0438, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0676, 0.0598, 0.0878, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 08:44:03,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4212, 1.1962, 1.0783, 1.5778], device='cuda:1'), covar=tensor([0.0726, 0.0359, 0.0336, 0.0792], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0209], device='cuda:1'), out_proj_covar=tensor([0.0082, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 08:44:09,704 INFO [train.py:968] (1/2) Epoch 12, batch 26250, giga_loss[loss=0.2863, simple_loss=0.3572, pruned_loss=0.1077, over 28863.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3838, pruned_loss=0.1306, over 5665459.44 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3691, pruned_loss=0.1203, over 5718230.97 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3845, pruned_loss=0.1307, over 5670367.29 frames. ], batch size: 174, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:44:31,843 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 26300, giga_loss[loss=0.407, simple_loss=0.4411, pruned_loss=0.1865, over 27954.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3837, pruned_loss=0.1311, over 5672232.26 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3698, pruned_loss=0.1209, over 5713213.42 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3839, pruned_loss=0.1309, over 5678775.15 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:44:59,443 INFO [optim.py:369] (1/2) 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 08:45:21,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-06 08:45:36,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 08:45:39,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9217, 1.9593, 1.3510, 1.6788], device='cuda:1'), covar=tensor([0.0737, 0.0627, 0.0995, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0443, 0.0502, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 08:45:40,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:45:42,690 INFO [train.py:968] (1/2) Epoch 12, batch 26350, giga_loss[loss=0.2949, simple_loss=0.3641, pruned_loss=0.1128, over 29009.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3803, pruned_loss=0.1292, over 5675973.30 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1208, over 5715389.18 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3808, pruned_loss=0.1293, over 5678674.18 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 08:45:43,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:46:09,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=527571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:46:16,847 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:46:18,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9675, 5.1530, 1.9609, 2.6792], device='cuda:1'), covar=tensor([0.0865, 0.0214, 0.0805, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0518, 0.0345, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 08:46:30,209 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 26400, giga_loss[loss=0.2785, simple_loss=0.3536, pruned_loss=0.1016, over 28990.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3781, pruned_loss=0.1283, over 5672506.56 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.37, pruned_loss=0.1211, over 5706501.34 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3784, pruned_loss=0.1283, over 5682737.62 frames. ], batch size: 106, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:46:36,712 INFO [optim.py:369] (1/2) 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 08:47:22,535 INFO [train.py:968] (1/2) Epoch 12, batch 26450, libri_loss[loss=0.3122, simple_loss=0.3825, pruned_loss=0.1209, over 29231.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3783, pruned_loss=0.1287, over 5667359.52 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3703, pruned_loss=0.1211, over 5707714.59 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3784, pruned_loss=0.1288, over 5673318.36 frames. ], batch size: 97, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:47:28,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:48:02,702 INFO [train.py:968] (1/2) Epoch 12, batch 26500, giga_loss[loss=0.3147, simple_loss=0.3748, pruned_loss=0.1274, over 28234.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1283, over 5676932.95 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5710460.11 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.378, pruned_loss=0.1286, over 5677884.37 frames. ], batch size: 368, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:48:04,047 INFO [optim.py:369] (1/2) 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 08:48:27,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:48:29,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:48:47,329 INFO [train.py:968] (1/2) Epoch 12, batch 26550, giga_loss[loss=0.2751, simple_loss=0.342, pruned_loss=0.1041, over 28237.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3775, pruned_loss=0.1293, over 5668173.17 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3708, pruned_loss=0.1214, over 5712064.28 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3776, pruned_loss=0.1295, over 5666496.58 frames. ], batch size: 60, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:48:58,079 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 26600, giga_loss[loss=0.341, simple_loss=0.3936, pruned_loss=0.1442, over 28684.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3765, pruned_loss=0.1291, over 5649612.46 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 5705666.32 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3766, pruned_loss=0.1293, over 5653673.71 frames. ], batch size: 262, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:49:41,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:49:41,282 INFO [optim.py:369] (1/2) 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 08:49:43,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=527796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:50:10,770 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=527825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:50:25,724 INFO [train.py:968] (1/2) Epoch 12, batch 26650, giga_loss[loss=0.2876, simple_loss=0.3601, pruned_loss=0.1076, over 28863.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3761, pruned_loss=0.1271, over 5660968.54 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3711, pruned_loss=0.1216, over 5708373.47 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3762, pruned_loss=0.1274, over 5660956.89 frames. ], batch size: 186, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:50:26,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:50:36,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 08:51:04,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.09 vs. limit=2.0
+2023-03-06 08:51:15,722 INFO [train.py:968] (1/2) Epoch 12, batch 26700, giga_loss[loss=0.3778, simple_loss=0.3954, pruned_loss=0.1801, over 23480.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3804, pruned_loss=0.1307, over 5650694.38 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3711, pruned_loss=0.1216, over 5708574.31 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3806, pruned_loss=0.131, over 5649924.63 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:51:19,373 INFO [optim.py:369] (1/2) 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 08:52:01,407 INFO [train.py:968] (1/2) Epoch 12, batch 26750, giga_loss[loss=0.2688, simple_loss=0.3366, pruned_loss=0.1005, over 28552.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3791, pruned_loss=0.1299, over 5661809.30 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3709, pruned_loss=0.1215, over 5710399.25 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3796, pruned_loss=0.1305, over 5657958.32 frames. ], batch size: 85, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:52:18,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=527962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:52:37,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=527985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:52:39,722 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 26800, libri_loss[loss=0.2664, simple_loss=0.3322, pruned_loss=0.1003, over 29537.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3787, pruned_loss=0.127, over 5669583.78 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3709, pruned_loss=0.1216, over 5712469.04 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3794, pruned_loss=0.1276, over 5663156.88 frames. ], batch size: 79, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:52:44,857 INFO [optim.py:369] (1/2) 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 08:52:49,797 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 08:53:04,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:53:26,083 INFO [train.py:968] (1/2) Epoch 12, batch 26850, giga_loss[loss=0.2995, simple_loss=0.381, pruned_loss=0.109, over 28853.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3794, pruned_loss=0.1249, over 5690129.12 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3701, pruned_loss=0.121, over 5719727.60 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.381, pruned_loss=0.1261, over 5677215.54 frames. ], batch size: 174, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:53:29,469 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:53:33,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2917, 1.6265, 1.2890, 1.3286], device='cuda:1'), covar=tensor([0.2359, 0.2268, 0.2551, 0.1975], device='cuda:1'), in_proj_covar=tensor([0.1318, 0.0977, 0.1160, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 08:54:10,131 INFO [train.py:968] (1/2) Epoch 12, batch 26900, giga_loss[loss=0.3077, simple_loss=0.3811, pruned_loss=0.1172, over 28600.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3817, pruned_loss=0.1256, over 5695963.97 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1208, over 5723182.27 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3835, pruned_loss=0.1268, over 5682003.35 frames. ], batch size: 60, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:54:11,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 08:54:12,705 INFO [optim.py:369] (1/2) 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 08:54:21,735 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:54:24,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=528108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:54:53,722 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 26950, giga_loss[loss=0.3652, simple_loss=0.4168, pruned_loss=0.1568, over 28179.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3857, pruned_loss=0.1299, over 5687325.47 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3698, pruned_loss=0.1208, over 5723328.51 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3874, pruned_loss=0.131, over 5675579.41 frames. ], batch size: 77, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:55:15,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-06 08:55:16,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0401, 5.8360, 5.4772, 3.1686], device='cuda:1'), covar=tensor([0.0432, 0.0589, 0.0745, 0.1488], device='cuda:1'), in_proj_covar=tensor([0.1093, 0.1019, 0.0889, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 08:55:47,933 INFO [train.py:968] (1/2) Epoch 12, batch 27000, giga_loss[loss=0.3274, simple_loss=0.3913, pruned_loss=0.1317, over 28634.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3879, pruned_loss=0.1331, over 5678252.27 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5725818.80 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3895, pruned_loss=0.1342, over 5666420.95 frames. ], batch size: 262, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:55:47,933 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 08:55:56,464 INFO [train.py:1012] (1/2) Epoch 12, validation: loss=0.2134, simple_loss=0.3197, pruned_loss=0.05357, over 944034.00 frames. 
+2023-03-06 08:55:56,465 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 08:55:58,625 INFO [optim.py:369] (1/2) 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 08:56:36,321 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:56:42,936 INFO [train.py:968] (1/2) Epoch 12, batch 27050, giga_loss[loss=0.3539, simple_loss=0.4074, pruned_loss=0.1503, over 28901.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3884, pruned_loss=0.1346, over 5667296.23 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1207, over 5728240.24 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3901, pruned_loss=0.1357, over 5654232.49 frames. ], batch size: 145, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:57:29,186 INFO [train.py:968] (1/2) Epoch 12, batch 27100, libri_loss[loss=0.3073, simple_loss=0.3606, pruned_loss=0.127, over 29601.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3863, pruned_loss=0.1332, over 5659176.73 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3692, pruned_loss=0.1205, over 5731520.19 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3889, pruned_loss=0.1349, over 5643128.55 frames. ], batch size: 74, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:57:31,870 INFO [optim.py:369] (1/2) 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 08:58:17,029 INFO [train.py:968] (1/2) Epoch 12, batch 27150, giga_loss[loss=0.4006, simple_loss=0.4263, pruned_loss=0.1875, over 26556.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.386, pruned_loss=0.1314, over 5664943.88 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3694, pruned_loss=0.1207, over 5733691.99 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3881, pruned_loss=0.1327, over 5649339.52 frames. ], batch size: 555, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 08:58:31,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2249, 1.4787, 1.4444, 1.2804], device='cuda:1'), covar=tensor([0.1385, 0.1567, 0.1902, 0.1667], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0730, 0.0671, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 08:59:04,584 INFO [train.py:968] (1/2) Epoch 12, batch 27200, giga_loss[loss=0.3038, simple_loss=0.3778, pruned_loss=0.1149, over 28826.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3855, pruned_loss=0.1288, over 5679194.45 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3695, pruned_loss=0.1208, over 5735068.26 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3872, pruned_loss=0.1299, over 5664736.23 frames. ], batch size: 186, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 08:59:07,597 INFO [optim.py:369] (1/2) 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 08:59:12,251 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=528400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:59:33,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 08:59:55,605 INFO [train.py:968] (1/2) Epoch 12, batch 27250, giga_loss[loss=0.3934, simple_loss=0.4428, pruned_loss=0.172, over 28783.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3864, pruned_loss=0.1301, over 5672395.20 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3692, pruned_loss=0.1207, over 5737498.07 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3883, pruned_loss=0.1311, over 5657949.64 frames. ], batch size: 243, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:00:04,402 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-06 09:00:44,101 INFO [train.py:968] (1/2) Epoch 12, batch 27300, giga_loss[loss=0.2663, simple_loss=0.3474, pruned_loss=0.09264, over 28954.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3859, pruned_loss=0.1299, over 5679291.41 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3692, pruned_loss=0.1206, over 5738186.58 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3875, pruned_loss=0.1308, over 5667029.73 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:00:47,960 INFO [optim.py:369] (1/2) 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 09:01:16,585 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 27350, giga_loss[loss=0.2983, simple_loss=0.3577, pruned_loss=0.1195, over 28671.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3831, pruned_loss=0.1293, over 5660526.51 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3694, pruned_loss=0.1207, over 5731626.54 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3845, pruned_loss=0.1302, over 5655492.44 frames. ], batch size: 92, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:01:52,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:01:55,010 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 27400, giga_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 28853.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3827, pruned_loss=0.1306, over 5646331.42 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3702, pruned_loss=0.1211, over 5729545.78 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3835, pruned_loss=0.1311, over 5642221.02 frames. ], batch size: 199, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:02:23,640 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:02:24,813 INFO [optim.py:369] (1/2) 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 09:02:25,401 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 09:02:42,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:03:04,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0996, 1.1753, 3.4088, 2.9630], device='cuda:1'), covar=tensor([0.1642, 0.2539, 0.0511, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0602, 0.0875, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 09:03:13,938 INFO [train.py:968] (1/2) Epoch 12, batch 27450, giga_loss[loss=0.2978, simple_loss=0.3551, pruned_loss=0.1203, over 28921.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3803, pruned_loss=0.1293, over 5654600.89 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3698, pruned_loss=0.1209, over 5731502.62 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3813, pruned_loss=0.13, over 5648702.05 frames. ], batch size: 112, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:03:48,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3619, 1.5761, 1.6380, 1.3007], device='cuda:1'), covar=tensor([0.1079, 0.1626, 0.0902, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0699, 0.0873, 0.0780], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 09:03:49,402 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-06 09:04:00,027 INFO [train.py:968] (1/2) Epoch 12, batch 27500, giga_loss[loss=0.3855, simple_loss=0.4195, pruned_loss=0.1758, over 28624.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3811, pruned_loss=0.1315, over 5642245.95 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3701, pruned_loss=0.1211, over 5724095.69 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3818, pruned_loss=0.1321, over 5643931.55 frames. ], batch size: 336, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:04:04,673 INFO [optim.py:369] (1/2) 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 09:04:07,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8558, 1.8561, 1.3670, 1.4961], device='cuda:1'), covar=tensor([0.0785, 0.0670, 0.1015, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0445, 0.0501, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 09:04:10,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2827, 0.8762, 0.9765, 1.3870], device='cuda:1'), covar=tensor([0.0730, 0.0353, 0.0319, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 09:04:27,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3684, 2.1937, 1.9503, 1.9072], device='cuda:1'), covar=tensor([0.0758, 0.0727, 0.0858, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0445, 0.0501, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 09:04:46,628 INFO [train.py:968] (1/2) Epoch 12, batch 27550, giga_loss[loss=0.2735, simple_loss=0.3498, pruned_loss=0.09864, over 28894.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3805, pruned_loss=0.1312, over 5628339.07 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3706, pruned_loss=0.1216, over 5705873.47 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3807, pruned_loss=0.1313, over 5645210.35 frames. ], batch size: 174, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:04:57,799 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:04:59,501 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:05:26,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1640, 1.5494, 1.4394, 1.0410], device='cuda:1'), covar=tensor([0.1294, 0.2331, 0.1228, 0.1450], device='cuda:1'), in_proj_covar=tensor([0.0834, 0.0700, 0.0874, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 09:05:27,839 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=528786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:05:33,897 INFO [train.py:968] (1/2) Epoch 12, batch 27600, giga_loss[loss=0.2939, simple_loss=0.3654, pruned_loss=0.1112, over 28875.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3769, pruned_loss=0.1268, over 5644608.46 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3712, pruned_loss=0.1221, over 5708725.25 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3766, pruned_loss=0.1265, over 5654307.12 frames. ], batch size: 227, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 09:05:36,561 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 27650, giga_loss[loss=0.2845, simple_loss=0.3496, pruned_loss=0.1097, over 28935.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3723, pruned_loss=0.1224, over 5640584.19 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.1221, over 5697372.50 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3724, pruned_loss=0.1222, over 5657765.26 frames. ], batch size: 106, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:06:28,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2567, 1.5116, 1.3326, 1.4962], device='cuda:1'), covar=tensor([0.0742, 0.0336, 0.0316, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 09:07:10,522 INFO [train.py:968] (1/2) Epoch 12, batch 27700, giga_loss[loss=0.3323, simple_loss=0.386, pruned_loss=0.1393, over 27866.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3724, pruned_loss=0.1227, over 5640604.25 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1219, over 5699968.73 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3727, pruned_loss=0.1227, over 5651240.30 frames. ], batch size: 412, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:07:15,247 INFO [optim.py:369] (1/2) 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 09:07:17,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=528899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:07:36,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=528918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:07:38,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=528921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:07:45,907 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 27750, giga_loss[loss=0.2706, simple_loss=0.3411, pruned_loss=0.1, over 28968.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5646978.68 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3706, pruned_loss=0.1219, over 5697436.04 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3688, pruned_loss=0.1208, over 5655807.26 frames. ], batch size: 136, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:08:13,199 INFO [zipformer.py:1188] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.2859, 1.6812, 1.2976, 1.4725], device='cuda:1'), covar=tensor([0.0748, 0.0335, 0.0320, 0.0787], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0053, 0.0090], device='cuda:1')
+2023-03-06 09:08:53,979 INFO [train.py:968] (1/2) Epoch 12, batch 27800, giga_loss[loss=0.2873, simple_loss=0.3543, pruned_loss=0.1102, over 28927.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3677, pruned_loss=0.1209, over 5641911.99 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3706, pruned_loss=0.122, over 5693123.81 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3678, pruned_loss=0.1208, over 5652232.88 frames. ], batch size: 227, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:08:58,709 INFO [optim.py:369] (1/2) 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 09:09:30,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5943, 1.7086, 1.5829, 1.4025], device='cuda:1'), covar=tensor([0.2021, 0.1960, 0.1488, 0.1851], device='cuda:1'), in_proj_covar=tensor([0.1721, 0.1630, 0.1579, 0.1693], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 09:09:40,060 INFO [train.py:968] (1/2) Epoch 12, batch 27850, libri_loss[loss=0.3128, simple_loss=0.38, pruned_loss=0.1228, over 29668.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.121, over 5655974.67 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3705, pruned_loss=0.1219, over 5693776.81 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3694, pruned_loss=0.121, over 5662819.02 frames. ], batch size: 88, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:09:40,309 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:09:42,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:10:10,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:10:18,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1576, 2.4131, 1.2172, 1.2872], device='cuda:1'), covar=tensor([0.0896, 0.0323, 0.0847, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0515, 0.0344, 0.0393], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 09:10:26,571 INFO [train.py:968] (1/2) Epoch 12, batch 27900, giga_loss[loss=0.471, simple_loss=0.4731, pruned_loss=0.2345, over 26483.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3714, pruned_loss=0.1227, over 5647309.07 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3706, pruned_loss=0.1221, over 5694922.83 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1225, over 5650246.46 frames. ], batch size: 555, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:10:31,037 INFO [optim.py:369] (1/2) 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 09:11:11,624 INFO [train.py:968] (1/2) Epoch 12, batch 27950, giga_loss[loss=0.2714, simple_loss=0.3399, pruned_loss=0.1014, over 28100.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3727, pruned_loss=0.1239, over 5643113.26 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3711, pruned_loss=0.1226, over 5689660.19 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3723, pruned_loss=0.1233, over 5648358.03 frames. ], batch size: 77, lr: 2.68e-03, grad_scale: 2.0
+2023-03-06 09:11:55,042 INFO [train.py:968] (1/2) Epoch 12, batch 28000, giga_loss[loss=0.3702, simple_loss=0.4131, pruned_loss=0.1636, over 28858.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3741, pruned_loss=0.1256, over 5650470.27 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1225, over 5698974.98 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.374, pruned_loss=0.1252, over 5643461.88 frames. ], batch size: 186, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:11:59,865 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 12, batch 28050, giga_loss[loss=0.3419, simple_loss=0.4031, pruned_loss=0.1404, over 28901.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3766, pruned_loss=0.1272, over 5669478.69 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3708, pruned_loss=0.1223, over 5702783.50 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3768, pruned_loss=0.1272, over 5659512.08 frames. ], batch size: 174, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:13:21,749 INFO [train.py:968] (1/2) Epoch 12, batch 28100, giga_loss[loss=0.3094, simple_loss=0.3751, pruned_loss=0.1218, over 28530.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3789, pruned_loss=0.1286, over 5654398.23 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 5692091.20 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3786, pruned_loss=0.1281, over 5654827.40 frames. ], batch size: 336, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:13:28,309 INFO [optim.py:369] (1/2) 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 09:13:31,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:13:35,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3330, 3.1300, 2.9631, 1.6725], device='cuda:1'), covar=tensor([0.0912, 0.1059, 0.1003, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.1085, 0.1014, 0.0884, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 09:13:56,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6256, 1.6266, 1.2370, 1.2591], device='cuda:1'), covar=tensor([0.0769, 0.0632, 0.0952, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0445, 0.0502, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 09:14:08,007 INFO [train.py:968] (1/2) Epoch 12, batch 28150, giga_loss[loss=0.3185, simple_loss=0.3859, pruned_loss=0.1256, over 28979.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3795, pruned_loss=0.1287, over 5663578.72 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3718, pruned_loss=0.1232, over 5696567.05 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3791, pruned_loss=0.1282, over 5659024.69 frames. ], batch size: 164, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:14:58,062 INFO [train.py:968] (1/2) Epoch 12, batch 28200, libri_loss[loss=0.3325, simple_loss=0.3928, pruned_loss=0.1361, over 29357.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3802, pruned_loss=0.1296, over 5655976.96 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3721, pruned_loss=0.1234, over 5701343.68 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3798, pruned_loss=0.1292, over 5646725.09 frames. ], batch size: 92, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:15:01,970 INFO [optim.py:369] (1/2) 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 09:15:39,255 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 28250, giga_loss[loss=0.3104, simple_loss=0.3855, pruned_loss=0.1177, over 28926.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3815, pruned_loss=0.1304, over 5658390.64 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3717, pruned_loss=0.1232, over 5708566.01 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3819, pruned_loss=0.1305, over 5643058.95 frames. ], batch size: 213, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:15:50,782 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:15:55,671 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=529452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:16:24,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:16:32,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1090, 1.4318, 1.3966, 1.0264], device='cuda:1'), covar=tensor([0.1471, 0.2322, 0.1245, 0.1454], device='cuda:1'), in_proj_covar=tensor([0.0835, 0.0699, 0.0876, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 09:16:38,629 INFO [train.py:968] (1/2) Epoch 12, batch 28300, giga_loss[loss=0.3168, simple_loss=0.3592, pruned_loss=0.1372, over 23615.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3799, pruned_loss=0.1273, over 5661999.41 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3712, pruned_loss=0.1229, over 5707967.24 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3808, pruned_loss=0.1278, over 5649343.80 frames. ], batch size: 705, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:16:43,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0242, 1.2193, 3.3962, 2.9772], device='cuda:1'), covar=tensor([0.1676, 0.2498, 0.0503, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0674, 0.0597, 0.0871, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 09:16:45,411 INFO [optim.py:369] (1/2) 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 09:17:22,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7287, 1.7876, 1.5345, 1.8157], device='cuda:1'), covar=tensor([0.2438, 0.2543, 0.2709, 0.2297], device='cuda:1'), in_proj_covar=tensor([0.1333, 0.0986, 0.1172, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 09:17:28,420 INFO [train.py:968] (1/2) Epoch 12, batch 28350, giga_loss[loss=0.2851, simple_loss=0.3522, pruned_loss=0.109, over 28839.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3798, pruned_loss=0.1278, over 5670002.13 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1227, over 5709724.53 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3808, pruned_loss=0.1284, over 5658056.78 frames. ], batch size: 119, lr: 2.68e-03, grad_scale: 4.0
+2023-03-06 09:18:08,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:18:19,289 INFO [train.py:968] (1/2) Epoch 12, batch 28400, libri_loss[loss=0.3511, simple_loss=0.4055, pruned_loss=0.1483, over 29256.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3792, pruned_loss=0.1283, over 5675051.41 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3713, pruned_loss=0.1229, over 5715709.25 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3799, pruned_loss=0.1287, over 5658530.11 frames. ], batch size: 97, lr: 2.68e-03, grad_scale: 8.0
+2023-03-06 09:18:22,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1373, 1.3736, 1.2756, 1.0453], device='cuda:1'), covar=tensor([0.1808, 0.1694, 0.1172, 0.1550], device='cuda:1'), in_proj_covar=tensor([0.1723, 0.1633, 0.1596, 0.1694], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 09:18:25,915 INFO [optim.py:369] (1/2) 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 09:18:27,911 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-06 09:19:17,393 INFO [train.py:968] (1/2) Epoch 12, batch 28450, giga_loss[loss=0.3097, simple_loss=0.382, pruned_loss=0.1186, over 28775.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3774, pruned_loss=0.1275, over 5678693.54 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.371, pruned_loss=0.1228, over 5716205.89 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3784, pruned_loss=0.1281, over 5664169.02 frames. ], batch size: 119, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:19:58,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5513, 2.1097, 1.5628, 0.7489], device='cuda:1'), covar=tensor([0.4067, 0.2089, 0.3096, 0.4686], device='cuda:1'), in_proj_covar=tensor([0.1568, 0.1489, 0.1490, 0.1274], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 09:20:09,857 INFO [train.py:968] (1/2) Epoch 12, batch 28500, giga_loss[loss=0.3135, simple_loss=0.3752, pruned_loss=0.126, over 29038.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3758, pruned_loss=0.1267, over 5680119.71 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1225, over 5718885.78 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3771, pruned_loss=0.1275, over 5665374.26 frames. ], batch size: 128, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:20:14,432 INFO [optim.py:369] (1/2) 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 09:20:39,505 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:20:41,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:20:54,489 INFO [train.py:968] (1/2) Epoch 12, batch 28550, giga_loss[loss=0.3733, simple_loss=0.3961, pruned_loss=0.1752, over 23552.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5679121.57 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5722395.70 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3777, pruned_loss=0.1284, over 5663329.21 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:21:06,659 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:21:37,760 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=529788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:21:42,604 INFO [train.py:968] (1/2) Epoch 12, batch 28600, giga_loss[loss=0.2966, simple_loss=0.368, pruned_loss=0.1126, over 28906.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3764, pruned_loss=0.1279, over 5671999.25 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5727432.32 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3773, pruned_loss=0.1287, over 5653436.57 frames. ], batch size: 145, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:21:49,882 INFO [optim.py:369] (1/2) 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 09:22:00,112 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:22:00,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4604, 1.5760, 1.6926, 1.3203], device='cuda:1'), covar=tensor([0.1445, 0.2155, 0.1196, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0699, 0.0873, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 09:22:09,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4659, 1.3943, 0.9916, 1.1298], device='cuda:1'), covar=tensor([0.1063, 0.0976, 0.1505, 0.1464], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0443, 0.0497, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 09:22:16,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=529827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:22:28,635 INFO [train.py:968] (1/2) Epoch 12, batch 28650, giga_loss[loss=0.345, simple_loss=0.3971, pruned_loss=0.1465, over 28721.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3774, pruned_loss=0.1289, over 5641219.00 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3711, pruned_loss=0.1227, over 5698435.33 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3779, pruned_loss=0.1293, over 5650527.76 frames. ], batch size: 284, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:23:15,885 INFO [train.py:968] (1/2) Epoch 12, batch 28700, giga_loss[loss=0.2838, simple_loss=0.3531, pruned_loss=0.1072, over 28972.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3799, pruned_loss=0.131, over 5638620.03 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1232, over 5689463.88 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3799, pruned_loss=0.1311, over 5653003.53 frames. ], batch size: 112, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:23:22,240 INFO [optim.py:369] (1/2) 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 09:24:05,381 INFO [train.py:968] (1/2) Epoch 12, batch 28750, giga_loss[loss=0.3126, simple_loss=0.3711, pruned_loss=0.1271, over 27882.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3805, pruned_loss=0.1321, over 5631752.92 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1237, over 5694264.88 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3802, pruned_loss=0.1319, over 5637667.15 frames. ], batch size: 412, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:24:16,525 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:18,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:31,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=529970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:34,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=529973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:45,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=529986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:24:50,508 INFO [train.py:968] (1/2) Epoch 12, batch 28800, giga_loss[loss=0.2998, simple_loss=0.367, pruned_loss=0.1163, over 28690.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3798, pruned_loss=0.132, over 5641644.68 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5696495.28 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3794, pruned_loss=0.1316, over 5643947.20 frames. ], batch size: 262, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:24:58,723 INFO [optim.py:369] (1/2) 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 09:25:01,014 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:25:12,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5297, 1.6701, 1.5576, 1.4400], device='cuda:1'), covar=tensor([0.2234, 0.1975, 0.1704, 0.1902], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1648, 0.1605, 0.1709], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 09:25:14,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:25:33,615 INFO [train.py:968] (1/2) Epoch 12, batch 28850, giga_loss[loss=0.2881, simple_loss=0.3615, pruned_loss=0.1074, over 28973.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3789, pruned_loss=0.1313, over 5647847.08 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3727, pruned_loss=0.1241, over 5698267.17 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3786, pruned_loss=0.1311, over 5646380.51 frames. ], batch size: 136, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:25:47,654 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:26:22,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:26:23,231 INFO [train.py:968] (1/2) Epoch 12, batch 28900, giga_loss[loss=0.2922, simple_loss=0.3546, pruned_loss=0.1149, over 28325.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3791, pruned_loss=0.131, over 5639741.59 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.1239, over 5699436.67 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3792, pruned_loss=0.1312, over 5636479.35 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:26:33,347 INFO [optim.py:369] (1/2) 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 09:27:10,126 INFO [train.py:968] (1/2) Epoch 12, batch 28950, giga_loss[loss=0.348, simple_loss=0.4012, pruned_loss=0.1474, over 28268.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3807, pruned_loss=0.1316, over 5653534.83 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1242, over 5703166.44 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3806, pruned_loss=0.1316, over 5646687.22 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:27:28,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:27:54,640 INFO [train.py:968] (1/2) Epoch 12, batch 29000, giga_loss[loss=0.2818, simple_loss=0.352, pruned_loss=0.1058, over 28835.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3822, pruned_loss=0.1324, over 5657205.21 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1245, over 5704028.36 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3818, pruned_loss=0.1324, over 5649764.13 frames. ], batch size: 112, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:28:02,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8503, 4.6691, 4.4417, 2.0529], device='cuda:1'), covar=tensor([0.0513, 0.0753, 0.0741, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.1020, 0.0888, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 09:28:04,068 INFO [optim.py:369] (1/2) 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 09:28:38,438 INFO [train.py:968] (1/2) Epoch 12, batch 29050, giga_loss[loss=0.2703, simple_loss=0.3553, pruned_loss=0.0927, over 29002.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3817, pruned_loss=0.1318, over 5675626.20 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1242, over 5708885.83 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3818, pruned_loss=0.1322, over 5664097.85 frames. ], batch size: 164, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:29:23,648 INFO [train.py:968] (1/2) Epoch 12, batch 29100, giga_loss[loss=0.3766, simple_loss=0.4223, pruned_loss=0.1655, over 28276.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.382, pruned_loss=0.1321, over 5677477.26 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3731, pruned_loss=0.1241, over 5711752.78 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3823, pruned_loss=0.1328, over 5665444.20 frames. ], batch size: 369, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:29:31,729 INFO [optim.py:369] (1/2) 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 09:29:37,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:29:39,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:30:08,736 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 29150, giga_loss[loss=0.296, simple_loss=0.3444, pruned_loss=0.1238, over 23445.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3829, pruned_loss=0.1322, over 5666514.70 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3727, pruned_loss=0.1238, over 5705290.97 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3839, pruned_loss=0.1332, over 5661547.50 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 09:30:39,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-06 09:31:02,521 INFO [train.py:968] (1/2) Epoch 12, batch 29200, giga_loss[loss=0.3386, simple_loss=0.3897, pruned_loss=0.1437, over 28565.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3816, pruned_loss=0.1304, over 5660129.33 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.372, pruned_loss=0.1234, over 5710896.09 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3832, pruned_loss=0.1317, over 5650147.30 frames. ], batch size: 85, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:31:02,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:31:09,291 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:31:39,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:31:48,811 INFO [train.py:968] (1/2) Epoch 12, batch 29250, giga_loss[loss=0.3327, simple_loss=0.3893, pruned_loss=0.138, over 29014.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3801, pruned_loss=0.1291, over 5669782.05 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.372, pruned_loss=0.1234, over 5710896.09 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3814, pruned_loss=0.1302, over 5662012.91 frames. ], batch size: 128, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:31:57,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4416, 1.8175, 1.4238, 1.7439], device='cuda:1'), covar=tensor([0.2473, 0.2357, 0.2667, 0.2153], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.0980, 0.1166, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 09:32:08,814 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 29300, giga_loss[loss=0.2814, simple_loss=0.3597, pruned_loss=0.1016, over 28921.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3798, pruned_loss=0.1295, over 5662598.99 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3717, pruned_loss=0.1231, over 5715460.28 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3812, pruned_loss=0.1307, over 5651806.25 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:32:42,275 INFO [optim.py:369] (1/2) 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 09:33:13,857 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:18,376 INFO [train.py:968] (1/2) Epoch 12, batch 29350, giga_loss[loss=0.3085, simple_loss=0.3777, pruned_loss=0.1197, over 28751.00 frames. ], tot_loss[loss=0.319, simple_loss=0.38, pruned_loss=0.129, over 5676030.25 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.1231, over 5721070.78 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3814, pruned_loss=0.1302, over 5660734.02 frames. ], batch size: 284, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:33:46,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:52,804 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:55,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:33:58,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:09,569 INFO [train.py:968] (1/2) Epoch 12, batch 29400, giga_loss[loss=0.2812, simple_loss=0.3553, pruned_loss=0.1036, over 28941.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3809, pruned_loss=0.1305, over 5666574.87 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1229, over 5721531.27 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3826, pruned_loss=0.1317, over 5653756.12 frames. ], batch size: 145, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:34:19,990 INFO [optim.py:369] (1/2) 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 09:34:25,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:26,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=530609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:29,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:56,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:34:57,240 INFO [train.py:968] (1/2) Epoch 12, batch 29450, giga_loss[loss=0.3182, simple_loss=0.3771, pruned_loss=0.1296, over 29035.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3807, pruned_loss=0.1313, over 5661582.92 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3715, pruned_loss=0.1231, over 5715767.51 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3821, pruned_loss=0.1323, over 5654693.36 frames. ], batch size: 164, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:35:43,195 INFO [train.py:968] (1/2) Epoch 12, batch 29500, giga_loss[loss=0.3042, simple_loss=0.3732, pruned_loss=0.1177, over 28707.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3808, pruned_loss=0.132, over 5655484.45 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3709, pruned_loss=0.1228, over 5719310.13 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3826, pruned_loss=0.1333, over 5645745.35 frames. ], batch size: 242, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:35:51,396 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 29550, libri_loss[loss=0.3571, simple_loss=0.407, pruned_loss=0.1536, over 29513.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3812, pruned_loss=0.1318, over 5668445.71 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3708, pruned_loss=0.1228, over 5723981.61 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.383, pruned_loss=0.133, over 5655134.44 frames. ], batch size: 81, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:37:19,558 INFO [train.py:968] (1/2) Epoch 12, batch 29600, giga_loss[loss=0.3077, simple_loss=0.3746, pruned_loss=0.1204, over 28753.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3821, pruned_loss=0.1326, over 5654093.76 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5723687.92 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3839, pruned_loss=0.1338, over 5643173.10 frames. ], batch size: 284, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:37:24,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:37:26,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8248, 1.1283, 2.8834, 2.7120], device='cuda:1'), covar=tensor([0.1584, 0.2357, 0.0528, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0677, 0.0595, 0.0869, 0.0776], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 09:37:27,231 INFO [optim.py:369] (1/2) 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 09:37:59,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-06 09:38:03,436 INFO [train.py:968] (1/2) Epoch 12, batch 29650, giga_loss[loss=0.3072, simple_loss=0.3778, pruned_loss=0.1183, over 28668.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3802, pruned_loss=0.1299, over 5676619.22 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1225, over 5727937.62 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.382, pruned_loss=0.1313, over 5661894.53 frames. ], batch size: 262, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:38:15,288 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=530853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:38:18,222 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-06 09:38:51,248 INFO [train.py:968] (1/2) Epoch 12, batch 29700, giga_loss[loss=0.3061, simple_loss=0.3762, pruned_loss=0.118, over 28942.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3801, pruned_loss=0.1294, over 5666033.20 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3707, pruned_loss=0.1226, over 5721729.27 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3816, pruned_loss=0.1306, over 5658636.55 frames. ], batch size: 186, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:39:02,039 INFO [optim.py:369] (1/2) 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 09:39:38,497 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 29750, libri_loss[loss=0.4183, simple_loss=0.4411, pruned_loss=0.1977, over 29513.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3798, pruned_loss=0.1291, over 5671747.79 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3708, pruned_loss=0.1228, over 5725095.50 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3809, pruned_loss=0.1299, over 5661658.99 frames. ], batch size: 85, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:39:40,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=530943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:39:44,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=530947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:40:08,966 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=530972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:40:10,978 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4530, 1.5587, 4.1850, 3.3565], device='cuda:1'), covar=tensor([0.1890, 0.2599, 0.0715, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0597, 0.0869, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 09:40:25,420 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 29800, giga_loss[loss=0.2861, simple_loss=0.3584, pruned_loss=0.107, over 28901.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3781, pruned_loss=0.1283, over 5670022.70 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.123, over 5725720.73 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.379, pruned_loss=0.1288, over 5660792.80 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:40:37,711 INFO [optim.py:369] (1/2) 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 09:41:14,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-06 09:41:15,846 INFO [train.py:968] (1/2) Epoch 12, batch 29850, giga_loss[loss=0.3705, simple_loss=0.4069, pruned_loss=0.167, over 27625.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3773, pruned_loss=0.1282, over 5666748.88 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3711, pruned_loss=0.1231, over 5728412.55 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.378, pruned_loss=0.1287, over 5656262.26 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:41:55,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4820, 1.7640, 1.4368, 1.4986], device='cuda:1'), covar=tensor([0.2347, 0.2343, 0.2607, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1328, 0.0982, 0.1167, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 09:41:57,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:41:58,788 INFO [train.py:968] (1/2) Epoch 12, batch 29900, giga_loss[loss=0.3348, simple_loss=0.3843, pruned_loss=0.1426, over 29029.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3748, pruned_loss=0.1268, over 5667340.73 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.123, over 5729968.57 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3756, pruned_loss=0.1274, over 5655306.45 frames. ], batch size: 128, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:41:59,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:42:00,038 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 09:42:01,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4563, 1.7981, 1.4153, 1.4933], device='cuda:1'), covar=tensor([0.2428, 0.2345, 0.2693, 0.2188], device='cuda:1'), in_proj_covar=tensor([0.1327, 0.0982, 0.1167, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 09:42:09,903 INFO [optim.py:369] (1/2) 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 09:42:29,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:42:32,643 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:42:46,337 INFO [train.py:968] (1/2) Epoch 12, batch 29950, giga_loss[loss=0.3049, simple_loss=0.3628, pruned_loss=0.1235, over 28744.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3706, pruned_loss=0.1242, over 5683746.88 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1232, over 5733932.85 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3712, pruned_loss=0.1247, over 5669290.84 frames. ], batch size: 262, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:43:27,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6309, 1.6620, 1.8775, 1.4385], device='cuda:1'), covar=tensor([0.1422, 0.2124, 0.1138, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0700, 0.0876, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 09:43:31,125 INFO [train.py:968] (1/2) Epoch 12, batch 30000, giga_loss[loss=0.3123, simple_loss=0.3633, pruned_loss=0.1306, over 28599.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3704, pruned_loss=0.1247, over 5688189.45 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1232, over 5727546.18 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3705, pruned_loss=0.125, over 5680901.81 frames. ], batch size: 78, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:43:31,126 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 09:43:38,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3399, 3.1503, 1.4248, 1.5509], device='cuda:1'), covar=tensor([0.1048, 0.0353, 0.1030, 0.1503], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0520, 0.0348, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 09:43:39,671 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 09:43:48,774 INFO [optim.py:369] (1/2) 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 09:44:15,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:44:29,482 INFO [train.py:968] (1/2) Epoch 12, batch 30050, libri_loss[loss=0.3138, simple_loss=0.3805, pruned_loss=0.1235, over 29271.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3702, pruned_loss=0.1247, over 5693952.30 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5732440.80 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3708, pruned_loss=0.1253, over 5682575.67 frames. ], batch size: 94, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:45:16,630 INFO [train.py:968] (1/2) Epoch 12, batch 30100, giga_loss[loss=0.3028, simple_loss=0.3754, pruned_loss=0.1151, over 28646.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3696, pruned_loss=0.1229, over 5692904.95 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5736003.48 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.37, pruned_loss=0.1234, over 5679639.84 frames. ], batch size: 262, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:45:28,613 INFO [optim.py:369] (1/2) 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 09:45:48,501 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:46:14,528 INFO [train.py:968] (1/2) Epoch 12, batch 30150, giga_loss[loss=0.2921, simple_loss=0.371, pruned_loss=0.1066, over 28735.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.367, pruned_loss=0.1192, over 5677438.28 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3707, pruned_loss=0.1228, over 5736810.46 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3675, pruned_loss=0.1197, over 5666216.20 frames. ], batch size: 262, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:46:25,628 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.03 vs. limit=2.0
+2023-03-06 09:46:36,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:46:42,100 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:46:45,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:46:55,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3796, 4.1923, 3.9331, 1.9497], device='cuda:1'), covar=tensor([0.0524, 0.0747, 0.0815, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.1080, 0.1012, 0.0878, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 09:47:00,087 INFO [train.py:968] (1/2) Epoch 12, batch 30200, giga_loss[loss=0.2531, simple_loss=0.3349, pruned_loss=0.08565, over 28599.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3633, pruned_loss=0.1157, over 5677603.44 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3696, pruned_loss=0.1222, over 5738073.74 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3646, pruned_loss=0.1164, over 5665004.52 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:47:11,946 INFO [optim.py:369] (1/2) 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 09:47:12,167 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:47:52,810 INFO [train.py:968] (1/2) Epoch 12, batch 30250, giga_loss[loss=0.2443, simple_loss=0.3269, pruned_loss=0.08086, over 28942.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1124, over 5670390.93 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3688, pruned_loss=0.122, over 5741370.57 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3612, pruned_loss=0.1131, over 5656140.48 frames. ], batch size: 106, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:47:53,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9444, 3.7287, 3.5383, 1.7969], device='cuda:1'), covar=tensor([0.0651, 0.0852, 0.0926, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.1075, 0.1009, 0.0876, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 09:48:41,714 INFO [train.py:968] (1/2) Epoch 12, batch 30300, giga_loss[loss=0.2749, simple_loss=0.36, pruned_loss=0.09495, over 28668.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3568, pruned_loss=0.1088, over 5667626.58 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3687, pruned_loss=0.1219, over 5742147.09 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3581, pruned_loss=0.1092, over 5655140.22 frames. ], batch size: 242, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:48:52,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:48:53,815 INFO [optim.py:369] (1/2) 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 09:48:57,546 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:49:01,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:49:29,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:49:32,772 INFO [train.py:968] (1/2) Epoch 12, batch 30350, giga_loss[loss=0.2783, simple_loss=0.3323, pruned_loss=0.1122, over 23978.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3552, pruned_loss=0.1064, over 5654159.22 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3684, pruned_loss=0.122, over 5745741.31 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3561, pruned_loss=0.1063, over 5638902.89 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:50:23,009 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-06 09:50:25,769 INFO [train.py:968] (1/2) Epoch 12, batch 30400, giga_loss[loss=0.2595, simple_loss=0.3415, pruned_loss=0.08874, over 28641.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3561, pruned_loss=0.1069, over 5652049.07 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3677, pruned_loss=0.1215, over 5748096.23 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3572, pruned_loss=0.1069, over 5635722.10 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:50:34,752 INFO [optim.py:369] (1/2) 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 09:51:14,760 INFO [train.py:968] (1/2) Epoch 12, batch 30450, giga_loss[loss=0.2878, simple_loss=0.3629, pruned_loss=0.1064, over 28819.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3545, pruned_loss=0.1058, over 5648890.56 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3679, pruned_loss=0.1219, over 5747189.22 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3551, pruned_loss=0.1053, over 5635527.49 frames. ], batch size: 284, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:51:16,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:51:20,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:51:47,462 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:01,091 INFO [train.py:968] (1/2) Epoch 12, batch 30500, giga_loss[loss=0.2406, simple_loss=0.3298, pruned_loss=0.07568, over 28862.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3507, pruned_loss=0.1032, over 5653885.72 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3668, pruned_loss=0.1213, over 5751416.97 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3518, pruned_loss=0.1028, over 5636937.68 frames. ], batch size: 174, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:52:06,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=531696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:09,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:14,557 INFO [optim.py:369] (1/2) 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 09:52:30,618 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:52:53,749 INFO [train.py:968] (1/2) Epoch 12, batch 30550, giga_loss[loss=0.2819, simple_loss=0.3555, pruned_loss=0.1041, over 28325.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3506, pruned_loss=0.1034, over 5643028.11 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.367, pruned_loss=0.1216, over 5743139.98 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1027, over 5635625.52 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:53:28,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2632, 2.6502, 1.4224, 1.3892], device='cuda:1'), covar=tensor([0.0887, 0.0361, 0.0853, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0515, 0.0347, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 09:53:43,220 INFO [train.py:968] (1/2) Epoch 12, batch 30600, giga_loss[loss=0.2646, simple_loss=0.3426, pruned_loss=0.09333, over 28264.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3512, pruned_loss=0.1034, over 5634711.08 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3669, pruned_loss=0.1216, over 5732006.08 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3514, pruned_loss=0.1025, over 5638005.48 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:53:55,956 INFO [optim.py:369] (1/2) 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 09:53:56,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1870, 1.7199, 1.4676, 1.4244], device='cuda:1'), covar=tensor([0.0830, 0.0312, 0.0314, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 09:53:59,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4600, 1.7376, 1.7843, 1.3283], device='cuda:1'), covar=tensor([0.1816, 0.2424, 0.1450, 0.1688], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0692, 0.0873, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 09:54:31,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5141, 1.6240, 1.2694, 1.2045], device='cuda:1'), covar=tensor([0.0834, 0.0478, 0.0959, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0436, 0.0495, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 09:54:31,959 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=531839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:54:33,884 INFO [train.py:968] (1/2) Epoch 12, batch 30650, giga_loss[loss=0.241, simple_loss=0.3223, pruned_loss=0.07989, over 28527.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3478, pruned_loss=0.09999, over 5646194.09 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3667, pruned_loss=0.1215, over 5732902.05 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3481, pruned_loss=0.09928, over 5647487.75 frames. ], batch size: 336, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:54:35,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=531842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:55:04,211 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=531871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:55:25,203 INFO [train.py:968] (1/2) Epoch 12, batch 30700, giga_loss[loss=0.2369, simple_loss=0.3217, pruned_loss=0.07602, over 28824.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.344, pruned_loss=0.09752, over 5640455.59 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3661, pruned_loss=0.1213, over 5734711.08 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3444, pruned_loss=0.09675, over 5638325.00 frames. ], batch size: 199, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:55:34,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=531900.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 09:55:38,014 INFO [optim.py:369] (1/2) 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 09:56:07,716 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 09:56:16,482 INFO [train.py:968] (1/2) Epoch 12, batch 30750, giga_loss[loss=0.2741, simple_loss=0.3488, pruned_loss=0.09973, over 28306.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.34, pruned_loss=0.09549, over 5644063.80 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3657, pruned_loss=0.121, over 5736520.36 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3403, pruned_loss=0.09467, over 5639204.07 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:57:05,347 INFO [train.py:968] (1/2) Epoch 12, batch 30800, giga_loss[loss=0.3172, simple_loss=0.3749, pruned_loss=0.1298, over 27662.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3397, pruned_loss=0.09604, over 5646039.29 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3657, pruned_loss=0.121, over 5736452.73 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3394, pruned_loss=0.09491, over 5640493.84 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:57:17,191 INFO [optim.py:369] (1/2) 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 09:57:17,404 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 30850, giga_loss[loss=0.2595, simple_loss=0.3409, pruned_loss=0.08907, over 27977.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3383, pruned_loss=0.09533, over 5629773.06 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3652, pruned_loss=0.1208, over 5739197.09 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3379, pruned_loss=0.09406, over 5620730.52 frames. ], batch size: 412, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 09:58:34,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 09:58:43,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7526, 2.0434, 1.5955, 2.2431], device='cuda:1'), covar=tensor([0.2321, 0.2206, 0.2479, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1330, 0.0973, 0.1172, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 09:58:55,841 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 30900, giga_loss[loss=0.2901, simple_loss=0.3642, pruned_loss=0.108, over 28252.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3407, pruned_loss=0.09577, over 5635384.70 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3649, pruned_loss=0.1207, over 5741323.75 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3403, pruned_loss=0.0945, over 5624728.49 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 09:59:09,835 INFO [optim.py:369] (1/2) 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 09:59:10,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4584, 1.6303, 1.1796, 1.1908], device='cuda:1'), covar=tensor([0.0821, 0.0486, 0.0975, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0438, 0.0500, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 09:59:30,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0079, 2.1076, 2.0434, 1.7969], device='cuda:1'), covar=tensor([0.1492, 0.2224, 0.1719, 0.1967], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0725, 0.0666, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 09:59:37,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6727, 2.2510, 1.3502, 0.8802], device='cuda:1'), covar=tensor([0.5291, 0.2867, 0.3454, 0.4787], device='cuda:1'), in_proj_covar=tensor([0.1564, 0.1481, 0.1483, 0.1273], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 09:59:52,045 INFO [train.py:968] (1/2) Epoch 12, batch 30950, giga_loss[loss=0.3325, simple_loss=0.3762, pruned_loss=0.1444, over 26898.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3427, pruned_loss=0.09601, over 5636707.59 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3641, pruned_loss=0.1204, over 5730795.75 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3426, pruned_loss=0.09478, over 5635980.18 frames. ], batch size: 555, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:00:04,059 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 31000, giga_loss[loss=0.2531, simple_loss=0.3227, pruned_loss=0.09174, over 26849.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3431, pruned_loss=0.09609, over 5638562.41 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3635, pruned_loss=0.1203, over 5711904.68 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.343, pruned_loss=0.0946, over 5653993.58 frames. ], batch size: 555, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:01:11,226 INFO [optim.py:369] (1/2) 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 10:01:25,413 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:29,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:47,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:49,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:01:53,954 INFO [train.py:968] (1/2) Epoch 12, batch 31050, libri_loss[loss=0.2707, simple_loss=0.3269, pruned_loss=0.1073, over 29641.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.34, pruned_loss=0.09451, over 5634175.56 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3629, pruned_loss=0.12, over 5705357.72 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.34, pruned_loss=0.0929, over 5649715.01 frames. ], batch size: 73, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:01:59,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1286, 1.5021, 1.2607, 1.3240], device='cuda:1'), covar=tensor([0.1514, 0.1539, 0.1816, 0.1601], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0724, 0.0667, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 10:02:01,731 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:02:20,749 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:02:33,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532275.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 10:02:57,068 INFO [train.py:968] (1/2) Epoch 12, batch 31100, libri_loss[loss=0.3065, simple_loss=0.3663, pruned_loss=0.1234, over 28645.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3401, pruned_loss=0.09327, over 5647333.79 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.363, pruned_loss=0.12, over 5709195.62 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3394, pruned_loss=0.09138, over 5654890.71 frames. ], batch size: 106, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:03:15,027 INFO [optim.py:369] (1/2) 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 10:03:58,473 INFO [train.py:968] (1/2) Epoch 12, batch 31150, giga_loss[loss=0.2433, simple_loss=0.3231, pruned_loss=0.08177, over 28474.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3374, pruned_loss=0.0918, over 5651756.35 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3624, pruned_loss=0.1197, over 5709405.72 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.09003, over 5656381.64 frames. ], batch size: 369, lr: 2.67e-03, grad_scale: 2.0
+2023-03-06 10:04:02,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5729, 2.2908, 1.6360, 0.7198], device='cuda:1'), covar=tensor([0.3705, 0.1951, 0.3174, 0.4440], device='cuda:1'), in_proj_covar=tensor([0.1579, 0.1493, 0.1497, 0.1289], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 10:04:40,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:05:00,944 INFO [train.py:968] (1/2) Epoch 12, batch 31200, giga_loss[loss=0.2545, simple_loss=0.3198, pruned_loss=0.09461, over 24379.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3347, pruned_loss=0.09144, over 5649889.91 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3621, pruned_loss=0.1198, over 5712899.94 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3341, pruned_loss=0.08937, over 5649450.57 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:05:18,975 INFO [optim.py:369] (1/2) 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 10:05:32,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532418.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:05:34,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532421.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:05:51,020 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:05:56,850 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-06 10:05:57,036 INFO [train.py:968] (1/2) Epoch 12, batch 31250, libri_loss[loss=0.2903, simple_loss=0.3457, pruned_loss=0.1174, over 29560.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3348, pruned_loss=0.09175, over 5664837.63 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3617, pruned_loss=0.1196, over 5716592.16 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08954, over 5659773.16 frames. ], batch size: 78, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:06:07,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532450.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 10:06:20,749 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 10:06:37,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1806, 1.2233, 3.4903, 3.0595], device='cuda:1'), covar=tensor([0.1569, 0.2590, 0.0457, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0671, 0.0596, 0.0864, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 10:06:54,580 INFO [train.py:968] (1/2) Epoch 12, batch 31300, giga_loss[loss=0.268, simple_loss=0.3505, pruned_loss=0.09273, over 28307.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3376, pruned_loss=0.09363, over 5663537.96 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3614, pruned_loss=0.1196, over 5717628.18 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3369, pruned_loss=0.09154, over 5658008.99 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:07:12,053 INFO [optim.py:369] (1/2) 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 10:07:32,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:07:37,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:07:40,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:07:58,756 INFO [train.py:968] (1/2) Epoch 12, batch 31350, giga_loss[loss=0.3044, simple_loss=0.3772, pruned_loss=0.1158, over 28130.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3395, pruned_loss=0.09339, over 5663591.77 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3613, pruned_loss=0.1195, over 5718583.04 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.339, pruned_loss=0.09171, over 5658209.27 frames. ], batch size: 412, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:08:18,667 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:08:28,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3831, 3.1422, 1.5891, 1.4895], device='cuda:1'), covar=tensor([0.0905, 0.0243, 0.0863, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0510, 0.0348, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 10:08:46,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 10:09:04,872 INFO [train.py:968] (1/2) Epoch 12, batch 31400, giga_loss[loss=0.2003, simple_loss=0.268, pruned_loss=0.06633, over 24700.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3354, pruned_loss=0.0906, over 5665623.14 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3613, pruned_loss=0.1197, over 5720704.10 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3347, pruned_loss=0.0889, over 5659071.28 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:09:23,123 INFO [optim.py:369] (1/2) 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 10:10:08,119 INFO [train.py:968] (1/2) Epoch 12, batch 31450, giga_loss[loss=0.2526, simple_loss=0.3385, pruned_loss=0.08335, over 28882.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3369, pruned_loss=0.09209, over 5672003.21 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3608, pruned_loss=0.1195, over 5717695.42 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.336, pruned_loss=0.08983, over 5666975.36 frames. ], batch size: 164, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:10:46,512 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:10:50,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:11:11,504 INFO [train.py:968] (1/2) Epoch 12, batch 31500, giga_loss[loss=0.3281, simple_loss=0.4053, pruned_loss=0.1255, over 28847.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3403, pruned_loss=0.09304, over 5665700.40 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3608, pruned_loss=0.1195, over 5715691.90 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3391, pruned_loss=0.09083, over 5662707.78 frames. ], batch size: 227, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:11:23,549 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:11:23,563 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:12:14,799 INFO [train.py:968] (1/2) Epoch 12, batch 31550, libri_loss[loss=0.251, simple_loss=0.3139, pruned_loss=0.09405, over 29368.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3431, pruned_loss=0.09254, over 5657262.19 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3603, pruned_loss=0.1194, over 5711712.10 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3421, pruned_loss=0.09013, over 5657106.98 frames. ], batch size: 71, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:13:18,210 INFO [train.py:968] (1/2) Epoch 12, batch 31600, giga_loss[loss=0.2473, simple_loss=0.3379, pruned_loss=0.07833, over 28472.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3431, pruned_loss=0.09092, over 5652105.99 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3605, pruned_loss=0.1195, over 5713835.31 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08869, over 5649827.97 frames. ], batch size: 336, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 10:13:33,845 INFO [optim.py:369] (1/2) 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 10:13:39,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=532811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:14:17,370 INFO [train.py:968] (1/2) Epoch 12, batch 31650, giga_loss[loss=0.3298, simple_loss=0.4041, pruned_loss=0.1278, over 29050.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3434, pruned_loss=0.0908, over 5645863.93 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3605, pruned_loss=0.1195, over 5708178.82 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3422, pruned_loss=0.08838, over 5648550.50 frames. ], batch size: 199, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:14:35,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:15:19,612 INFO [train.py:968] (1/2) Epoch 12, batch 31700, giga_loss[loss=0.2722, simple_loss=0.3456, pruned_loss=0.09936, over 28370.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3447, pruned_loss=0.09241, over 5641709.64 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3603, pruned_loss=0.1195, over 5700500.16 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3437, pruned_loss=0.09022, over 5649087.14 frames. ], batch size: 368, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:15:22,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7091, 1.8055, 1.3354, 1.4629], device='cuda:1'), covar=tensor([0.0753, 0.0508, 0.0973, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0358, 0.0430, 0.0495, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 10:15:40,424 INFO [optim.py:369] (1/2) 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 10:16:11,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=532931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:16:21,955 INFO [train.py:968] (1/2) Epoch 12, batch 31750, giga_loss[loss=0.2463, simple_loss=0.3357, pruned_loss=0.07845, over 28675.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3425, pruned_loss=0.0924, over 5657440.09 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3595, pruned_loss=0.1189, over 5705768.50 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3418, pruned_loss=0.09001, over 5656177.42 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:16:41,632 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=532954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:16:48,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=532957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:17:34,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6321, 2.2347, 1.6769, 0.8273], device='cuda:1'), covar=tensor([0.4286, 0.2363, 0.3222, 0.4574], device='cuda:1'), in_proj_covar=tensor([0.1584, 0.1493, 0.1503, 0.1291], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 10:17:37,995 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=532986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:17:45,471 INFO [train.py:968] (1/2) Epoch 12, batch 31800, giga_loss[loss=0.2803, simple_loss=0.3501, pruned_loss=0.1052, over 27482.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3445, pruned_loss=0.09449, over 5664328.16 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3594, pruned_loss=0.1189, over 5706728.66 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3439, pruned_loss=0.09247, over 5662308.70 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:18:06,838 INFO [optim.py:369] (1/2) 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 10:18:56,512 INFO [train.py:968] (1/2) Epoch 12, batch 31850, libri_loss[loss=0.2901, simple_loss=0.3531, pruned_loss=0.1136, over 29758.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3403, pruned_loss=0.09233, over 5671448.29 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3589, pruned_loss=0.1188, over 5712917.77 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3397, pruned_loss=0.09013, over 5663065.88 frames. ], batch size: 87, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:19:42,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=533077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:20:01,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5815, 2.5108, 2.1793, 2.1340], device='cuda:1'), covar=tensor([0.0712, 0.0627, 0.0778, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0438, 0.0502, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 10:20:01,895 INFO [train.py:968] (1/2) Epoch 12, batch 31900, giga_loss[loss=0.2169, simple_loss=0.3025, pruned_loss=0.06561, over 28402.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3377, pruned_loss=0.09076, over 5672851.39 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3584, pruned_loss=0.1184, over 5714070.95 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3376, pruned_loss=0.08895, over 5664769.98 frames. ], batch size: 71, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:20:20,732 INFO [optim.py:369] (1/2) 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 10:20:53,790 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-06 10:21:05,802 INFO [train.py:968] (1/2) Epoch 12, batch 31950, giga_loss[loss=0.2216, simple_loss=0.2913, pruned_loss=0.07591, over 24352.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3363, pruned_loss=0.09072, over 5657742.71 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3582, pruned_loss=0.1186, over 5702002.90 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3358, pruned_loss=0.08861, over 5661786.78 frames. ], batch size: 705, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:22:12,823 INFO [train.py:968] (1/2) Epoch 12, batch 32000, giga_loss[loss=0.2463, simple_loss=0.3333, pruned_loss=0.07963, over 29033.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3403, pruned_loss=0.09239, over 5663711.90 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3583, pruned_loss=0.1186, over 5704042.79 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3397, pruned_loss=0.09047, over 5664491.57 frames. ], batch size: 136, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 10:22:29,096 INFO [optim.py:369] (1/2) 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 10:22:33,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3065, 1.5574, 1.4557, 1.2004], device='cuda:1'), covar=tensor([0.1560, 0.1440, 0.0945, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1609, 0.1538, 0.1650], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:22:46,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=533220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:22:49,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=533223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:22:56,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=533231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:23:08,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4042, 4.2569, 4.0025, 1.9376], device='cuda:1'), covar=tensor([0.0484, 0.0592, 0.0740, 0.2279], device='cuda:1'), in_proj_covar=tensor([0.1040, 0.0971, 0.0848, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 10:23:08,475 INFO [train.py:968] (1/2) Epoch 12, batch 32050, giga_loss[loss=0.3031, simple_loss=0.368, pruned_loss=0.1191, over 27628.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.09394, over 5660506.75 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3582, pruned_loss=0.1186, over 5698200.44 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3404, pruned_loss=0.09152, over 5665565.44 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:23:27,462 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=533252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:23:38,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-03-06 10:24:15,293 INFO [train.py:968] (1/2) Epoch 12, batch 32100, giga_loss[loss=0.2469, simple_loss=0.3245, pruned_loss=0.08462, over 28903.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3398, pruned_loss=0.09402, over 5659752.08 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3582, pruned_loss=0.1187, over 5700409.74 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3387, pruned_loss=0.09155, over 5661084.05 frames. ], batch size: 112, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:24:31,884 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:25:04,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2273, 1.5388, 1.4356, 1.1895], device='cuda:1'), covar=tensor([0.2498, 0.1726, 0.1241, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.1702, 0.1620, 0.1545, 0.1657], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:25:11,647 INFO [train.py:968] (1/2) Epoch 12, batch 32150, giga_loss[loss=0.261, simple_loss=0.3446, pruned_loss=0.08865, over 28548.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3393, pruned_loss=0.09397, over 5663826.38 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3578, pruned_loss=0.1185, over 5704965.65 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3382, pruned_loss=0.09152, over 5659814.57 frames. ], batch size: 336, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:25:59,778 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=533374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:26:05,180 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 32200, giga_loss[loss=0.2822, simple_loss=0.3557, pruned_loss=0.1044, over 28646.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3416, pruned_loss=0.09492, over 5656887.30 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3577, pruned_loss=0.1186, over 5697927.38 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3407, pruned_loss=0.09257, over 5659591.14 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:26:36,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5013, 1.6321, 1.8415, 1.4149], device='cuda:1'), covar=tensor([0.1665, 0.2117, 0.1301, 0.1642], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0684, 0.0870, 0.0779], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 10:26:45,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=533406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:26:47,882 INFO [optim.py:369] (1/2) 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 10:27:43,133 INFO [train.py:968] (1/2) Epoch 12, batch 32250, giga_loss[loss=0.2569, simple_loss=0.3403, pruned_loss=0.08672, over 28786.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3432, pruned_loss=0.09437, over 5670308.33 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3576, pruned_loss=0.1185, over 5699163.25 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3425, pruned_loss=0.09242, over 5670968.72 frames. ], batch size: 99, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:27:56,205 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=533449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:28:01,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=533452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:28:10,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8799, 1.0378, 1.0213, 0.8892], device='cuda:1'), covar=tensor([0.1382, 0.1604, 0.0962, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.1702, 0.1615, 0.1542, 0.1654], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:28:33,391 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 10:28:42,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=533481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:28:56,972 INFO [train.py:968] (1/2) Epoch 12, batch 32300, giga_loss[loss=0.3073, simple_loss=0.3554, pruned_loss=0.1297, over 27575.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3409, pruned_loss=0.09333, over 5665581.01 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3573, pruned_loss=0.1184, over 5702434.02 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3403, pruned_loss=0.09146, over 5662633.36 frames. ], batch size: 472, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:29:15,486 INFO [optim.py:369] (1/2) 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 10:29:59,674 INFO [train.py:968] (1/2) Epoch 12, batch 32350, giga_loss[loss=0.2416, simple_loss=0.3149, pruned_loss=0.08415, over 28900.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3369, pruned_loss=0.09248, over 5665135.75 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3577, pruned_loss=0.1187, over 5696383.65 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3356, pruned_loss=0.09003, over 5666916.11 frames. ], batch size: 136, lr: 2.67e-03, grad_scale: 4.0
+2023-03-06 10:31:06,173 INFO [train.py:968] (1/2) Epoch 12, batch 32400, giga_loss[loss=0.2375, simple_loss=0.315, pruned_loss=0.08003, over 28594.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3316, pruned_loss=0.09001, over 5661411.60 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3576, pruned_loss=0.1187, over 5699867.07 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3301, pruned_loss=0.08744, over 5659051.34 frames. ], batch size: 307, lr: 2.67e-03, grad_scale: 8.0
+2023-03-06 10:31:22,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=533604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:31:25,676 INFO [optim.py:369] (1/2) 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 10:31:47,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8819, 2.2294, 2.1348, 1.6069], device='cuda:1'), covar=tensor([0.1526, 0.2126, 0.1257, 0.1606], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0689, 0.0877, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 10:32:05,168 INFO [train.py:968] (1/2) Epoch 12, batch 32450, giga_loss[loss=0.2791, simple_loss=0.3542, pruned_loss=0.102, over 28311.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3334, pruned_loss=0.09164, over 5658841.46 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3572, pruned_loss=0.1185, over 5701630.28 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.332, pruned_loss=0.08915, over 5654241.21 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:32:25,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2469, 1.5417, 1.4669, 1.2443], device='cuda:1'), covar=tensor([0.2225, 0.1706, 0.1384, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.1707, 0.1619, 0.1543, 0.1655], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:33:02,278 INFO [train.py:968] (1/2) Epoch 12, batch 32500, giga_loss[loss=0.2346, simple_loss=0.3193, pruned_loss=0.07491, over 29126.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.335, pruned_loss=0.0931, over 5660904.83 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.357, pruned_loss=0.1186, over 5706596.30 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3336, pruned_loss=0.09047, over 5651939.17 frames. ], batch size: 200, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:33:23,120 INFO [optim.py:369] (1/2) 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 10:34:02,056 INFO [train.py:968] (1/2) Epoch 12, batch 32550, giga_loss[loss=0.2598, simple_loss=0.333, pruned_loss=0.09329, over 28667.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3326, pruned_loss=0.09113, over 5659723.61 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3566, pruned_loss=0.1183, over 5707484.50 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.331, pruned_loss=0.0884, over 5650105.41 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:34:59,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5434, 1.7019, 1.4054, 1.5876], device='cuda:1'), covar=tensor([0.2662, 0.2387, 0.2764, 0.2032], device='cuda:1'), in_proj_covar=tensor([0.1323, 0.0971, 0.1167, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 10:35:04,999 INFO [train.py:968] (1/2) Epoch 12, batch 32600, giga_loss[loss=0.2284, simple_loss=0.3091, pruned_loss=0.07387, over 28979.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3316, pruned_loss=0.09007, over 5660945.24 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3567, pruned_loss=0.1183, over 5701625.38 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3298, pruned_loss=0.08728, over 5657656.51 frames. ], batch size: 136, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:35:25,152 INFO [optim.py:369] (1/2) 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 10:35:27,354 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=533812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:36:10,048 INFO [train.py:968] (1/2) Epoch 12, batch 32650, giga_loss[loss=0.1975, simple_loss=0.2669, pruned_loss=0.06404, over 24210.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3304, pruned_loss=0.09022, over 5652755.40 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3564, pruned_loss=0.1183, over 5700462.24 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3283, pruned_loss=0.087, over 5649675.59 frames. ], batch size: 705, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:36:34,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3400, 1.6486, 1.3950, 1.2377], device='cuda:1'), covar=tensor([0.2248, 0.1731, 0.1330, 0.1712], device='cuda:1'), in_proj_covar=tensor([0.1703, 0.1615, 0.1542, 0.1653], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:36:40,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2037, 1.2041, 1.0094, 0.8995], device='cuda:1'), covar=tensor([0.0708, 0.0391, 0.0837, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0434, 0.0500, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 10:36:53,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4935, 3.3296, 3.0931, 1.9253], device='cuda:1'), covar=tensor([0.0720, 0.0872, 0.0954, 0.1903], device='cuda:1'), in_proj_covar=tensor([0.1042, 0.0974, 0.0844, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 10:37:10,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3358, 1.6978, 1.3363, 1.3504], device='cuda:1'), covar=tensor([0.2574, 0.2524, 0.2810, 0.2191], device='cuda:1'), in_proj_covar=tensor([0.1321, 0.0972, 0.1165, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 10:37:12,251 INFO [train.py:968] (1/2) Epoch 12, batch 32700, giga_loss[loss=0.2373, simple_loss=0.3211, pruned_loss=0.07676, over 28443.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3311, pruned_loss=0.08988, over 5656958.45 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3566, pruned_loss=0.1186, over 5702206.48 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3285, pruned_loss=0.08627, over 5651986.19 frames. ], batch size: 369, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:37:35,220 INFO [optim.py:369] (1/2) 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 10:38:16,303 INFO [train.py:968] (1/2) Epoch 12, batch 32750, giga_loss[loss=0.2478, simple_loss=0.3273, pruned_loss=0.08417, over 28332.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3315, pruned_loss=0.09046, over 5641903.31 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3566, pruned_loss=0.1186, over 5685414.81 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.08685, over 5651274.69 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:38:19,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 10:38:59,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=533979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:39:11,575 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 32800, giga_loss[loss=0.2406, simple_loss=0.3205, pruned_loss=0.08032, over 28368.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.333, pruned_loss=0.09214, over 5634729.13 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3567, pruned_loss=0.1189, over 5669103.55 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3301, pruned_loss=0.08824, over 5654544.78 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:39:36,496 INFO [optim.py:369] (1/2) 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 10:40:01,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8074, 1.8466, 1.8435, 1.6697], device='cuda:1'), covar=tensor([0.1092, 0.1433, 0.1606, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0711, 0.0656, 0.0646], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 10:40:19,366 INFO [train.py:968] (1/2) Epoch 12, batch 32850, giga_loss[loss=0.2531, simple_loss=0.3183, pruned_loss=0.09391, over 26838.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3303, pruned_loss=0.09009, over 5638225.63 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3561, pruned_loss=0.1185, over 5673854.20 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3279, pruned_loss=0.08669, over 5649061.11 frames. ], batch size: 555, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:40:23,305 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:41:18,065 INFO [train.py:968] (1/2) Epoch 12, batch 32900, giga_loss[loss=0.2576, simple_loss=0.3427, pruned_loss=0.08623, over 28913.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3326, pruned_loss=0.0894, over 5649648.56 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3559, pruned_loss=0.1183, over 5676847.43 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3306, pruned_loss=0.08646, over 5655063.01 frames. ], batch size: 112, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:41:39,509 INFO [optim.py:369] (1/2) 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 10:41:58,024 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:42:00,057 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:42:19,249 INFO [train.py:968] (1/2) Epoch 12, batch 32950, giga_loss[loss=0.2603, simple_loss=0.3454, pruned_loss=0.08758, over 28614.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3349, pruned_loss=0.08984, over 5650902.53 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3554, pruned_loss=0.118, over 5677836.28 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3335, pruned_loss=0.0875, over 5653891.92 frames. ], batch size: 242, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:42:25,571 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5469, 1.5878, 1.1692, 1.2114], device='cuda:1'), covar=tensor([0.0671, 0.0389, 0.0879, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0429, 0.0495, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 10:42:32,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-06 10:42:33,892 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:43:15,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:43:19,618 INFO [train.py:968] (1/2) Epoch 12, batch 33000, giga_loss[loss=0.2532, simple_loss=0.341, pruned_loss=0.08267, over 28525.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3354, pruned_loss=0.09041, over 5647085.36 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3551, pruned_loss=0.1179, over 5683508.05 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3339, pruned_loss=0.08777, over 5643642.09 frames. ], batch size: 336, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:43:19,618 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 10:43:28,201 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 10:43:28,477 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 10:43:30,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9750, 1.4536, 1.3407, 1.2201], device='cuda:1'), covar=tensor([0.1755, 0.1572, 0.2021, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0712, 0.0655, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 10:43:46,320 INFO [optim.py:369] (1/2) 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 10:44:24,287 INFO [train.py:968] (1/2) Epoch 12, batch 33050, giga_loss[loss=0.262, simple_loss=0.3537, pruned_loss=0.08511, over 28663.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3359, pruned_loss=0.09103, over 5658884.73 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3542, pruned_loss=0.1172, over 5687724.14 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3348, pruned_loss=0.08851, over 5650952.05 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:45:01,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:45:20,310 INFO [train.py:968] (1/2) Epoch 12, batch 33100, giga_loss[loss=0.2149, simple_loss=0.3011, pruned_loss=0.06434, over 28660.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3346, pruned_loss=0.09031, over 5656437.73 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3543, pruned_loss=0.1173, over 5681475.04 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.333, pruned_loss=0.08732, over 5655252.68 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:45:43,911 INFO [optim.py:369] (1/2) 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 10:46:09,994 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:46:12,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:46:23,838 INFO [train.py:968] (1/2) Epoch 12, batch 33150, giga_loss[loss=0.2738, simple_loss=0.3482, pruned_loss=0.0997, over 28957.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.333, pruned_loss=0.08915, over 5660251.93 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3544, pruned_loss=0.1174, over 5685077.80 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3313, pruned_loss=0.08621, over 5655700.52 frames. ], batch size: 186, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:46:34,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-06 10:46:46,102 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:46:49,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:47:21,813 INFO [train.py:968] (1/2) Epoch 12, batch 33200, giga_loss[loss=0.2853, simple_loss=0.3561, pruned_loss=0.1073, over 28850.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3305, pruned_loss=0.08867, over 5669510.78 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3538, pruned_loss=0.1173, over 5691765.06 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.329, pruned_loss=0.0856, over 5659304.68 frames. ], batch size: 213, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 10:47:44,171 INFO [optim.py:369] (1/2) 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 10:47:55,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:48:21,965 INFO [train.py:968] (1/2) Epoch 12, batch 33250, libri_loss[loss=0.2429, simple_loss=0.3113, pruned_loss=0.08724, over 29521.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3338, pruned_loss=0.09031, over 5674658.97 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3531, pruned_loss=0.1168, over 5695584.58 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3328, pruned_loss=0.08765, over 5662563.22 frames. ], batch size: 80, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 10:48:37,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 10:49:24,883 INFO [train.py:968] (1/2) Epoch 12, batch 33300, giga_loss[loss=0.2449, simple_loss=0.3238, pruned_loss=0.08305, over 28946.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3355, pruned_loss=0.09131, over 5672576.48 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3527, pruned_loss=0.1167, over 5696922.35 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3346, pruned_loss=0.08877, over 5661580.57 frames. ], batch size: 186, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:49:45,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:49:50,533 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:50:23,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5657, 1.7176, 1.3917, 1.8972], device='cuda:1'), covar=tensor([0.2451, 0.2510, 0.2776, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.0970, 0.1162, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 10:50:23,864 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:50:28,480 INFO [train.py:968] (1/2) Epoch 12, batch 33350, giga_loss[loss=0.2741, simple_loss=0.3592, pruned_loss=0.09449, over 28721.00 frames. ], tot_loss[loss=0.261, simple_loss=0.337, pruned_loss=0.09256, over 5676602.48 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.353, pruned_loss=0.1167, over 5703444.43 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3354, pruned_loss=0.0896, over 5660763.52 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:50:57,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:51:00,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:51:03,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534567.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:51:33,475 INFO [train.py:968] (1/2) Epoch 12, batch 33400, giga_loss[loss=0.2606, simple_loss=0.3475, pruned_loss=0.08689, over 28894.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3408, pruned_loss=0.09444, over 5670648.73 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3531, pruned_loss=0.117, over 5694864.56 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3393, pruned_loss=0.0915, over 5665204.74 frames. ], batch size: 227, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:51:37,535 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:51:48,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4285, 1.7144, 1.4946, 1.4033], device='cuda:1'), covar=tensor([0.2089, 0.1639, 0.1697, 0.1712], device='cuda:1'), in_proj_covar=tensor([0.1705, 0.1605, 0.1541, 0.1644], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:51:50,570 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:52:29,611 INFO [train.py:968] (1/2) Epoch 12, batch 33450, giga_loss[loss=0.2407, simple_loss=0.3377, pruned_loss=0.07178, over 29036.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3426, pruned_loss=0.0945, over 5658340.23 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3532, pruned_loss=0.117, over 5689137.37 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3412, pruned_loss=0.09175, over 5658064.37 frames. ], batch size: 128, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:52:30,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6016, 1.7667, 1.7465, 1.4689], device='cuda:1'), covar=tensor([0.2564, 0.1816, 0.1453, 0.1876], device='cuda:1'), in_proj_covar=tensor([0.1709, 0.1608, 0.1544, 0.1648], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:52:39,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=534650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:53:02,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9144, 1.0899, 1.1013, 0.9188], device='cuda:1'), covar=tensor([0.1641, 0.1821, 0.0956, 0.1292], device='cuda:1'), in_proj_covar=tensor([0.1712, 0.1612, 0.1547, 0.1650], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 10:53:33,397 INFO [train.py:968] (1/2) Epoch 12, batch 33500, libri_loss[loss=0.2375, simple_loss=0.3106, pruned_loss=0.08223, over 29591.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3421, pruned_loss=0.09417, over 5650991.06 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3531, pruned_loss=0.1168, over 5680738.92 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3408, pruned_loss=0.09154, over 5656892.10 frames. ], batch size: 74, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:53:43,480 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-06 10:53:45,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.08 vs. limit=5.0
+2023-03-06 10:53:45,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5106, 1.7073, 1.8061, 1.3497], device='cuda:1'), covar=tensor([0.1479, 0.2192, 0.1233, 0.1545], device='cuda:1'), in_proj_covar=tensor([0.0831, 0.0683, 0.0870, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 10:54:00,446 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534710.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:54:02,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3937, 1.6667, 1.4051, 1.5432], device='cuda:1'), covar=tensor([0.0725, 0.0298, 0.0304, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0116, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 10:54:03,233 INFO [optim.py:369] (1/2) 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 10:54:04,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534713.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:54:33,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-06 10:54:44,282 INFO [train.py:968] (1/2) Epoch 12, batch 33550, libri_loss[loss=0.3297, simple_loss=0.3846, pruned_loss=0.1374, over 29082.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3398, pruned_loss=0.09353, over 5650969.45 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3528, pruned_loss=0.1168, over 5677635.45 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3386, pruned_loss=0.09069, over 5657682.27 frames. ], batch size: 101, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 10:54:44,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 10:55:43,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2876, 1.8328, 1.3495, 0.4385], device='cuda:1'), covar=tensor([0.3845, 0.1992, 0.3229, 0.4741], device='cuda:1'), in_proj_covar=tensor([0.1575, 0.1500, 0.1500, 0.1295], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 10:55:46,371 INFO [train.py:968] (1/2) Epoch 12, batch 33600, giga_loss[loss=0.2297, simple_loss=0.3158, pruned_loss=0.07179, over 28961.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3387, pruned_loss=0.09302, over 5648465.27 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3527, pruned_loss=0.1168, over 5670735.17 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3376, pruned_loss=0.09048, over 5659696.73 frames. ], batch size: 120, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:55:49,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=534793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:55:53,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=534796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:56:17,037 INFO [optim.py:369] (1/2) 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 10:56:36,156 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=534825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:57:00,227 INFO [train.py:968] (1/2) Epoch 12, batch 33650, giga_loss[loss=0.247, simple_loss=0.3213, pruned_loss=0.08635, over 27602.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.338, pruned_loss=0.09308, over 5637583.78 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3528, pruned_loss=0.1169, over 5671599.49 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.337, pruned_loss=0.0909, over 5645443.82 frames. ], batch size: 472, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:57:05,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=534845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 10:57:42,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-06 10:58:06,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2911, 1.2210, 4.2896, 3.3165], device='cuda:1'), covar=tensor([0.1731, 0.2835, 0.0392, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0667, 0.0595, 0.0855, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 10:58:06,506 INFO [train.py:968] (1/2) Epoch 12, batch 33700, giga_loss[loss=0.2326, simple_loss=0.3171, pruned_loss=0.07401, over 28620.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3353, pruned_loss=0.09206, over 5642987.75 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3525, pruned_loss=0.1168, over 5674443.01 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3345, pruned_loss=0.09015, over 5646178.87 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:58:28,798 INFO [optim.py:369] (1/2) 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 10:58:46,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5975, 1.7501, 1.2403, 1.3322], device='cuda:1'), covar=tensor([0.0871, 0.0528, 0.1062, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0433, 0.0500, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 10:59:08,211 INFO [train.py:968] (1/2) Epoch 12, batch 33750, giga_loss[loss=0.28, simple_loss=0.3584, pruned_loss=0.1008, over 28660.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3355, pruned_loss=0.09225, over 5642790.69 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3528, pruned_loss=0.1168, over 5678136.92 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3344, pruned_loss=0.09029, over 5641706.81 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 10:59:58,181 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 33800, giga_loss[loss=0.2217, simple_loss=0.307, pruned_loss=0.06819, over 28142.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3342, pruned_loss=0.09033, over 5653144.26 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3529, pruned_loss=0.117, over 5674018.34 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3328, pruned_loss=0.08797, over 5655351.25 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:00:34,945 INFO [optim.py:369] (1/2) 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 11:00:59,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5545, 1.6860, 1.4100, 1.6438], device='cuda:1'), covar=tensor([0.2511, 0.2469, 0.2814, 0.2258], device='cuda:1'), in_proj_covar=tensor([0.1326, 0.0975, 0.1172, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 11:01:06,300 INFO [train.py:968] (1/2) Epoch 12, batch 33850, giga_loss[loss=0.3008, simple_loss=0.3803, pruned_loss=0.1107, over 28646.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3348, pruned_loss=0.08873, over 5668655.87 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3523, pruned_loss=0.1166, over 5679562.68 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3338, pruned_loss=0.08654, over 5665040.70 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:01:22,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5877, 1.9438, 1.8905, 1.4140], device='cuda:1'), covar=tensor([0.1935, 0.2346, 0.1523, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0685, 0.0875, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 11:01:43,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-06 11:02:02,046 INFO [train.py:968] (1/2) Epoch 12, batch 33900, giga_loss[loss=0.2329, simple_loss=0.3244, pruned_loss=0.07072, over 28076.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3369, pruned_loss=0.08896, over 5669992.61 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3521, pruned_loss=0.1166, over 5687330.02 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3357, pruned_loss=0.08634, over 5660164.41 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:02:11,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1709, 1.2386, 1.0304, 0.9452], device='cuda:1'), covar=tensor([0.0891, 0.0505, 0.1093, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0434, 0.0502, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:02:25,664 INFO [optim.py:369] (1/2) 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 11:02:39,987 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:02:42,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:02:50,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3243, 1.5457, 1.4330, 1.2836], device='cuda:1'), covar=tensor([0.2091, 0.1640, 0.1400, 0.1676], device='cuda:1'), in_proj_covar=tensor([0.1706, 0.1588, 0.1530, 0.1638], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 11:02:56,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3569, 1.7054, 1.4899, 1.5277], device='cuda:1'), covar=tensor([0.0695, 0.0388, 0.0310, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0116, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 11:02:57,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5424, 1.7247, 1.4062, 1.6808], device='cuda:1'), covar=tensor([0.2420, 0.2279, 0.2561, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.1325, 0.0974, 0.1167, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 11:02:59,425 INFO [train.py:968] (1/2) Epoch 12, batch 33950, giga_loss[loss=0.2339, simple_loss=0.3011, pruned_loss=0.08331, over 24280.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.337, pruned_loss=0.089, over 5660026.94 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.352, pruned_loss=0.1166, over 5681583.83 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3356, pruned_loss=0.08605, over 5656850.20 frames. ], batch size: 705, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:03:22,598 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:03:45,042 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-06 11:04:02,601 INFO [train.py:968] (1/2) Epoch 12, batch 34000, giga_loss[loss=0.2503, simple_loss=0.3288, pruned_loss=0.08592, over 28649.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3365, pruned_loss=0.08915, over 5663865.71 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3511, pruned_loss=0.116, over 5678769.20 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3357, pruned_loss=0.08626, over 5663226.44 frames. ], batch size: 242, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:04:34,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1528, 1.4631, 1.3181, 1.0766], device='cuda:1'), covar=tensor([0.2053, 0.1724, 0.1147, 0.1565], device='cuda:1'), in_proj_covar=tensor([0.1709, 0.1588, 0.1531, 0.1639], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 11:04:35,076 INFO [optim.py:369] (1/2) 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 11:04:46,246 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 34050, giga_loss[loss=0.2723, simple_loss=0.3436, pruned_loss=0.1005, over 28946.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3353, pruned_loss=0.088, over 5666632.70 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3506, pruned_loss=0.1158, over 5682983.52 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3349, pruned_loss=0.08537, over 5662026.98 frames. ], batch size: 199, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:05:36,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 11:05:51,530 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 34100, giga_loss[loss=0.257, simple_loss=0.3458, pruned_loss=0.08412, over 28368.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3365, pruned_loss=0.08845, over 5670796.34 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.351, pruned_loss=0.1162, over 5689417.38 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3354, pruned_loss=0.08525, over 5660970.70 frames. ], batch size: 369, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:06:53,750 INFO [optim.py:369] (1/2) 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 11:07:14,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3264, 1.5676, 1.4707, 1.3418], device='cuda:1'), covar=tensor([0.1767, 0.1439, 0.1193, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.1706, 0.1584, 0.1524, 0.1635], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 11:07:31,117 INFO [train.py:968] (1/2) Epoch 12, batch 34150, giga_loss[loss=0.3239, simple_loss=0.389, pruned_loss=0.1294, over 28987.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3373, pruned_loss=0.08936, over 5663916.19 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3507, pruned_loss=0.1162, over 5692371.54 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3363, pruned_loss=0.08601, over 5652553.23 frames. ], batch size: 136, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:07:55,947 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:07:58,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:08:14,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5821, 1.7213, 1.8521, 1.3389], device='cuda:1'), covar=tensor([0.1711, 0.2293, 0.1385, 0.1609], device='cuda:1'), in_proj_covar=tensor([0.0833, 0.0683, 0.0872, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 11:08:29,396 INFO [train.py:968] (1/2) Epoch 12, batch 34200, giga_loss[loss=0.2503, simple_loss=0.3394, pruned_loss=0.0806, over 28386.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3406, pruned_loss=0.09099, over 5667797.41 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3504, pruned_loss=0.1159, over 5689887.46 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3396, pruned_loss=0.08751, over 5660203.32 frames. ], batch size: 368, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:08:35,637 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:08:37,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-06 11:08:42,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.28 vs. limit=5.0
+2023-03-06 11:08:56,142 INFO [optim.py:369] (1/2) 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 11:09:17,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:09:27,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1498, 1.4952, 1.3104, 1.1658], device='cuda:1'), covar=tensor([0.2451, 0.1397, 0.1241, 0.1562], device='cuda:1'), in_proj_covar=tensor([0.1715, 0.1591, 0.1533, 0.1639], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 11:09:34,347 INFO [train.py:968] (1/2) Epoch 12, batch 34250, giga_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.0977, over 27856.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3418, pruned_loss=0.09137, over 5679698.04 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3506, pruned_loss=0.116, over 5693037.36 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08782, over 5670337.48 frames. ], batch size: 476, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:10:41,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2296, 1.4776, 1.5126, 1.2655], device='cuda:1'), covar=tensor([0.1321, 0.1478, 0.1982, 0.1766], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0715, 0.0660, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 11:10:51,530 INFO [train.py:968] (1/2) Epoch 12, batch 34300, giga_loss[loss=0.209, simple_loss=0.2958, pruned_loss=0.06109, over 28533.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.34, pruned_loss=0.0908, over 5682984.52 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3506, pruned_loss=0.116, over 5692574.56 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3388, pruned_loss=0.08766, over 5675938.41 frames. ], batch size: 85, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:11:18,787 INFO [optim.py:369] (1/2) 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 11:11:39,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3538, 1.1084, 4.4799, 3.3413], device='cuda:1'), covar=tensor([0.1721, 0.2838, 0.0364, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0669, 0.0598, 0.0860, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:12:00,028 INFO [train.py:968] (1/2) Epoch 12, batch 34350, giga_loss[loss=0.2414, simple_loss=0.3319, pruned_loss=0.07544, over 28749.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3374, pruned_loss=0.08942, over 5692169.88 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3498, pruned_loss=0.1154, over 5698462.73 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3368, pruned_loss=0.08667, over 5680927.49 frames. ], batch size: 243, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:13:08,134 INFO [train.py:968] (1/2) Epoch 12, batch 34400, giga_loss[loss=0.2591, simple_loss=0.3432, pruned_loss=0.08747, over 28688.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3357, pruned_loss=0.08737, over 5699018.48 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3502, pruned_loss=0.1157, over 5700474.04 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3348, pruned_loss=0.08455, over 5688378.08 frames. ], batch size: 307, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:13:36,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6066, 2.2237, 1.5165, 0.8048], device='cuda:1'), covar=tensor([0.4175, 0.2277, 0.3606, 0.4770], device='cuda:1'), in_proj_covar=tensor([0.1562, 0.1484, 0.1491, 0.1281], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 11:13:41,983 INFO [optim.py:369] (1/2) 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 11:13:46,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6166, 2.1665, 1.9184, 1.6980], device='cuda:1'), covar=tensor([0.0745, 0.0234, 0.0273, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0116, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 11:13:47,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:14:10,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4746, 1.5983, 1.7098, 1.3222], device='cuda:1'), covar=tensor([0.1664, 0.2235, 0.1326, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0683, 0.0874, 0.0784], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 11:14:16,705 INFO [train.py:968] (1/2) Epoch 12, batch 34450, giga_loss[loss=0.2506, simple_loss=0.3324, pruned_loss=0.08437, over 28960.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3361, pruned_loss=0.08773, over 5690136.35 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3501, pruned_loss=0.1156, over 5698697.70 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3353, pruned_loss=0.08525, over 5683507.98 frames. ], batch size: 199, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:14:20,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=535645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:15:01,949 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:15:15,578 INFO [train.py:968] (1/2) Epoch 12, batch 34500, giga_loss[loss=0.2625, simple_loss=0.3496, pruned_loss=0.08777, over 28727.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3387, pruned_loss=0.08944, over 5669289.96 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3503, pruned_loss=0.1158, over 5683955.03 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3376, pruned_loss=0.08674, over 5676003.50 frames. ], batch size: 243, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:15:40,157 INFO [optim.py:369] (1/2) 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 11:15:59,712 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-06 11:16:12,751 INFO [train.py:968] (1/2) Epoch 12, batch 34550, giga_loss[loss=0.2263, simple_loss=0.2957, pruned_loss=0.07846, over 24203.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3386, pruned_loss=0.08979, over 5667222.42 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3503, pruned_loss=0.1158, over 5689897.32 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3374, pruned_loss=0.08692, over 5666740.32 frames. ], batch size: 705, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:17:07,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:17:09,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:17:09,995 INFO [train.py:968] (1/2) Epoch 12, batch 34600, giga_loss[loss=0.2666, simple_loss=0.3217, pruned_loss=0.1058, over 24449.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3348, pruned_loss=0.08883, over 5661033.20 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3498, pruned_loss=0.1156, over 5686287.36 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3339, pruned_loss=0.086, over 5663678.67 frames. ], batch size: 705, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:17:23,962 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:17:44,995 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:18:06,070 INFO [train.py:968] (1/2) Epoch 12, batch 34650, giga_loss[loss=0.2643, simple_loss=0.3333, pruned_loss=0.09764, over 28679.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3349, pruned_loss=0.08968, over 5656042.70 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3499, pruned_loss=0.1157, over 5675909.58 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3338, pruned_loss=0.08677, over 5666908.23 frames. ], batch size: 242, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:18:07,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3837, 1.6694, 1.3226, 1.5300], device='cuda:1'), covar=tensor([0.0777, 0.0308, 0.0338, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0116, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0059, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 11:18:57,860 INFO [train.py:968] (1/2) Epoch 12, batch 34700, giga_loss[loss=0.2918, simple_loss=0.3572, pruned_loss=0.1132, over 28045.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3402, pruned_loss=0.09348, over 5656842.96 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3496, pruned_loss=0.1154, over 5680559.98 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3391, pruned_loss=0.09066, over 5660841.19 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:19:13,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=535910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:19:17,681 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 34750, libri_loss[loss=0.2194, simple_loss=0.2878, pruned_loss=0.07548, over 29636.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3496, pruned_loss=0.09918, over 5672922.35 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3492, pruned_loss=0.115, over 5685223.16 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.349, pruned_loss=0.09668, over 5671454.56 frames. ], batch size: 69, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:19:50,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=535947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:19:52,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=535950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:20:20,763 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=535979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:20:33,208 INFO [train.py:968] (1/2) Epoch 12, batch 34800, libri_loss[loss=0.3407, simple_loss=0.3908, pruned_loss=0.1454, over 29494.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3572, pruned_loss=0.1037, over 5675370.23 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3495, pruned_loss=0.1152, over 5686477.00 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3565, pruned_loss=0.1014, over 5672851.01 frames. ], batch size: 85, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:20:34,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=535993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:20:42,019 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 11:20:50,667 INFO [optim.py:369] (1/2) 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 11:21:04,467 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3306, 1.6402, 1.5070, 1.4610], device='cuda:1'), covar=tensor([0.1637, 0.1687, 0.2037, 0.1820], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0722, 0.0665, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 11:21:11,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1632, 1.1814, 3.7598, 3.1141], device='cuda:1'), covar=tensor([0.1645, 0.2656, 0.0409, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0667, 0.0597, 0.0864, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:21:12,161 INFO [train.py:968] (1/2) Epoch 12, batch 34850, giga_loss[loss=0.2436, simple_loss=0.3184, pruned_loss=0.08436, over 28908.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3552, pruned_loss=0.1036, over 5678158.60 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3494, pruned_loss=0.1148, over 5690523.43 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3552, pruned_loss=0.1015, over 5671239.09 frames. ], batch size: 199, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:21:21,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5710, 3.0634, 1.5923, 1.7047], device='cuda:1'), covar=tensor([0.0863, 0.0283, 0.0851, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0506, 0.0344, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 11:21:24,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=536057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:21:39,650 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2943, 1.6144, 1.5658, 1.1777], device='cuda:1'), covar=tensor([0.1602, 0.2195, 0.1294, 0.1455], device='cuda:1'), in_proj_covar=tensor([0.0837, 0.0686, 0.0875, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 11:21:49,017 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 11:21:52,141 INFO [train.py:968] (1/2) Epoch 12, batch 34900, giga_loss[loss=0.2371, simple_loss=0.3094, pruned_loss=0.08236, over 28924.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3487, pruned_loss=0.1008, over 5676167.61 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3493, pruned_loss=0.1147, over 5685795.76 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3487, pruned_loss=0.0989, over 5674307.61 frames. ], batch size: 112, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:22:12,457 INFO [optim.py:369] (1/2) 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 11:22:29,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-06 11:22:32,199 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=536136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:22:34,075 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 34950, giga_loss[loss=0.2175, simple_loss=0.2905, pruned_loss=0.07222, over 28482.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3411, pruned_loss=0.09729, over 5679214.03 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.349, pruned_loss=0.1143, over 5686912.13 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3414, pruned_loss=0.09584, over 5676442.74 frames. ], batch size: 85, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:22:59,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=536168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:23:11,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7461, 1.8117, 1.3924, 1.2657], device='cuda:1'), covar=tensor([0.0845, 0.0605, 0.1010, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0435, 0.0503, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:23:19,915 INFO [train.py:968] (1/2) Epoch 12, batch 35000, giga_loss[loss=0.2283, simple_loss=0.2985, pruned_loss=0.07902, over 27879.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3326, pruned_loss=0.09337, over 5679829.27 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3492, pruned_loss=0.1143, over 5688418.58 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3325, pruned_loss=0.09201, over 5676242.35 frames. ], batch size: 412, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:23:27,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=536200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:23:29,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:23:53,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=536232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:23:59,907 INFO [train.py:968] (1/2) Epoch 12, batch 35050, giga_loss[loss=0.2264, simple_loss=0.3004, pruned_loss=0.07618, over 28745.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3255, pruned_loss=0.08994, over 5688578.97 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3497, pruned_loss=0.1146, over 5692947.66 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3245, pruned_loss=0.08823, over 5681549.77 frames. ], batch size: 199, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:24:41,995 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 35100, giga_loss[loss=0.2199, simple_loss=0.2926, pruned_loss=0.07356, over 28732.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3206, pruned_loss=0.0879, over 5692215.22 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.35, pruned_loss=0.1147, over 5695787.20 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3192, pruned_loss=0.08606, over 5683919.24 frames. ], batch size: 119, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:25:06,013 INFO [optim.py:369] (1/2) 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 11:25:13,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 11:25:30,331 INFO [train.py:968] (1/2) Epoch 12, batch 35150, giga_loss[loss=0.2275, simple_loss=0.3007, pruned_loss=0.07716, over 27611.00 frames. ], tot_loss[loss=0.246, simple_loss=0.318, pruned_loss=0.08702, over 5682939.37 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.35, pruned_loss=0.1147, over 5681542.78 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3163, pruned_loss=0.08502, over 5688673.59 frames. ], batch size: 472, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:25:44,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4818, 3.6545, 2.5587, 1.2008], device='cuda:1'), covar=tensor([0.4609, 0.1498, 0.2538, 0.4973], device='cuda:1'), in_proj_covar=tensor([0.1561, 0.1489, 0.1489, 0.1273], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 11:26:11,164 INFO [train.py:968] (1/2) Epoch 12, batch 35200, giga_loss[loss=0.2408, simple_loss=0.3053, pruned_loss=0.08812, over 28452.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3159, pruned_loss=0.08615, over 5685093.09 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3505, pruned_loss=0.1149, over 5687881.83 frames. ], giga_tot_loss[loss=0.2401, simple_loss=0.3131, pruned_loss=0.08356, over 5684033.98 frames. ], batch size: 85, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:26:19,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2991, 1.7661, 1.3103, 0.7176], device='cuda:1'), covar=tensor([0.4518, 0.2155, 0.2426, 0.4613], device='cuda:1'), in_proj_covar=tensor([0.1560, 0.1488, 0.1490, 0.1273], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 11:26:28,222 INFO [optim.py:369] (1/2) 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 11:26:38,964 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=536428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:26:43,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=536431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:26:54,182 INFO [train.py:968] (1/2) Epoch 12, batch 35250, giga_loss[loss=0.2202, simple_loss=0.2999, pruned_loss=0.07027, over 29060.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3133, pruned_loss=0.08535, over 5664818.41 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3508, pruned_loss=0.1151, over 5677340.23 frames. ], giga_tot_loss[loss=0.2378, simple_loss=0.3103, pruned_loss=0.08267, over 5673073.89 frames. ], batch size: 155, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:27:12,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=536460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:27:39,344 INFO [train.py:968] (1/2) Epoch 12, batch 35300, giga_loss[loss=0.2217, simple_loss=0.2939, pruned_loss=0.07475, over 28960.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3097, pruned_loss=0.08343, over 5662780.35 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3514, pruned_loss=0.1153, over 5667457.12 frames. ], giga_tot_loss[loss=0.2336, simple_loss=0.3061, pruned_loss=0.08058, over 5677769.54 frames. ], batch size: 145, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:27:59,791 INFO [optim.py:369] (1/2) 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 11:28:19,287 INFO [train.py:968] (1/2) Epoch 12, batch 35350, giga_loss[loss=0.2378, simple_loss=0.3014, pruned_loss=0.08707, over 27661.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3092, pruned_loss=0.08325, over 5676021.88 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3519, pruned_loss=0.1155, over 5675359.54 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3042, pruned_loss=0.07957, over 5680862.20 frames. ], batch size: 472, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:28:49,810 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 35400, giga_loss[loss=0.1972, simple_loss=0.2745, pruned_loss=0.05992, over 28665.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3055, pruned_loss=0.08153, over 5664000.95 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3522, pruned_loss=0.1158, over 5659086.65 frames. ], giga_tot_loss[loss=0.2284, simple_loss=0.3007, pruned_loss=0.07799, over 5682757.24 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:29:24,761 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 11:29:24,938 INFO [optim.py:369] (1/2) 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 11:29:42,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5294, 2.1891, 1.6036, 0.7283], device='cuda:1'), covar=tensor([0.5046, 0.2470, 0.3653, 0.5274], device='cuda:1'), in_proj_covar=tensor([0.1584, 0.1509, 0.1508, 0.1291], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 11:29:47,221 INFO [train.py:968] (1/2) Epoch 12, batch 35450, giga_loss[loss=0.2334, simple_loss=0.2955, pruned_loss=0.08565, over 28722.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3028, pruned_loss=0.08027, over 5670825.51 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3525, pruned_loss=0.1159, over 5662491.54 frames. ], giga_tot_loss[loss=0.2259, simple_loss=0.298, pruned_loss=0.07687, over 5682561.44 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:29:57,945 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=536653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:30:30,832 INFO [train.py:968] (1/2) Epoch 12, batch 35500, giga_loss[loss=0.3128, simple_loss=0.3581, pruned_loss=0.1337, over 26659.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3013, pruned_loss=0.08006, over 5665160.64 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.353, pruned_loss=0.1162, over 5662315.39 frames. ], giga_tot_loss[loss=0.2245, simple_loss=0.2961, pruned_loss=0.07644, over 5674933.36 frames. ], batch size: 555, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:30:52,772 INFO [optim.py:369] (1/2) 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 11:31:18,054 INFO [train.py:968] (1/2) Epoch 12, batch 35550, giga_loss[loss=0.2751, simple_loss=0.3538, pruned_loss=0.09817, over 28623.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3095, pruned_loss=0.08417, over 5673419.80 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3534, pruned_loss=0.1161, over 5668464.17 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3039, pruned_loss=0.08051, over 5675993.34 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 2.0
+2023-03-06 11:32:06,822 INFO [train.py:968] (1/2) Epoch 12, batch 35600, giga_loss[loss=0.3504, simple_loss=0.4063, pruned_loss=0.1472, over 27588.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3239, pruned_loss=0.09202, over 5676615.63 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.353, pruned_loss=0.1158, over 5672004.06 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3192, pruned_loss=0.08886, over 5675697.28 frames. ], batch size: 472, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:32:31,817 INFO [optim.py:369] (1/2) 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 11:32:52,563 INFO [train.py:968] (1/2) Epoch 12, batch 35650, giga_loss[loss=0.2785, simple_loss=0.3589, pruned_loss=0.09906, over 28726.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3364, pruned_loss=0.09839, over 5679900.33 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3532, pruned_loss=0.1159, over 5672794.04 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3324, pruned_loss=0.09575, over 5678456.66 frames. ], batch size: 242, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:33:21,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1990, 1.2949, 3.4659, 2.8768], device='cuda:1'), covar=tensor([0.1545, 0.2606, 0.0426, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0662, 0.0590, 0.0855, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:33:35,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1860, 1.2570, 3.3588, 2.8313], device='cuda:1'), covar=tensor([0.1540, 0.2604, 0.0425, 0.1714], device='cuda:1'), in_proj_covar=tensor([0.0663, 0.0591, 0.0855, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:33:36,419 INFO [train.py:968] (1/2) Epoch 12, batch 35700, giga_loss[loss=0.2978, simple_loss=0.3778, pruned_loss=0.1089, over 27537.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.344, pruned_loss=0.1017, over 5678139.35 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3536, pruned_loss=0.116, over 5668499.03 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3401, pruned_loss=0.09901, over 5680216.90 frames. ], batch size: 472, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:33:55,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 11:33:59,152 INFO [optim.py:369] (1/2) 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 11:34:12,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=536932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:34:21,732 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6147, 1.6695, 1.8730, 1.4363], device='cuda:1'), covar=tensor([0.1641, 0.2044, 0.1276, 0.1542], device='cuda:1'), in_proj_covar=tensor([0.0844, 0.0690, 0.0883, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 11:34:21,960 INFO [train.py:968] (1/2) Epoch 12, batch 35750, giga_loss[loss=0.2841, simple_loss=0.3548, pruned_loss=0.1067, over 26598.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3467, pruned_loss=0.1017, over 5677141.33 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3534, pruned_loss=0.1157, over 5672164.81 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3436, pruned_loss=0.09968, over 5675652.38 frames. ], batch size: 555, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:34:34,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=536954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:34:37,258 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=536958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:35:08,526 INFO [train.py:968] (1/2) Epoch 12, batch 35800, libri_loss[loss=0.4083, simple_loss=0.4473, pruned_loss=0.1846, over 26192.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3481, pruned_loss=0.1015, over 5663339.41 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3543, pruned_loss=0.1163, over 5662854.69 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3447, pruned_loss=0.09903, over 5670570.44 frames. ], batch size: 136, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:35:22,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-06 11:35:32,573 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 12, batch 35850, giga_loss[loss=0.2677, simple_loss=0.3393, pruned_loss=0.09804, over 28609.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3506, pruned_loss=0.1031, over 5670973.70 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3542, pruned_loss=0.1161, over 5666108.30 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3479, pruned_loss=0.101, over 5673786.64 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:36:37,996 INFO [train.py:968] (1/2) Epoch 12, batch 35900, libri_loss[loss=0.3218, simple_loss=0.396, pruned_loss=0.1238, over 29403.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3529, pruned_loss=0.1047, over 5680507.32 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3545, pruned_loss=0.1162, over 5671027.30 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3504, pruned_loss=0.1028, over 5678411.21 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:36:42,362 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:36:45,354 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:37:07,248 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:17,506 INFO [train.py:968] (1/2) Epoch 12, batch 35950, libri_loss[loss=0.3831, simple_loss=0.412, pruned_loss=0.1771, over 29550.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3556, pruned_loss=0.1067, over 5687930.71 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.355, pruned_loss=0.1163, over 5679758.76 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3531, pruned_loss=0.1046, over 5678709.11 frames. ], batch size: 78, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:37:43,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:45,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:52,391 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=537185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:37:56,783 INFO [train.py:968] (1/2) Epoch 12, batch 36000, giga_loss[loss=0.3293, simple_loss=0.3936, pruned_loss=0.1325, over 28609.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3592, pruned_loss=0.1076, over 5695287.60 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.356, pruned_loss=0.1167, over 5675724.87 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3564, pruned_loss=0.1053, over 5692456.66 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:37:56,783 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 11:38:03,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2468, 1.8106, 1.3865, 0.4061], device='cuda:1'), covar=tensor([0.3284, 0.2219, 0.3643, 0.3961], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1491, 0.1495, 0.1283], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 11:38:05,159 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 11:38:14,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:38:25,033 INFO [optim.py:369] (1/2) 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 11:38:47,398 INFO [train.py:968] (1/2) Epoch 12, batch 36050, giga_loss[loss=0.2928, simple_loss=0.3658, pruned_loss=0.1099, over 28665.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3601, pruned_loss=0.1072, over 5699678.24 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.356, pruned_loss=0.1166, over 5681460.47 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3579, pruned_loss=0.1052, over 5692887.07 frames. ], batch size: 92, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:38:55,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9334, 1.1542, 3.3239, 2.8229], device='cuda:1'), covar=tensor([0.1666, 0.2607, 0.0430, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0663, 0.0591, 0.0854, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:39:29,488 INFO [train.py:968] (1/2) Epoch 12, batch 36100, libri_loss[loss=0.2916, simple_loss=0.3653, pruned_loss=0.1089, over 29544.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3603, pruned_loss=0.1062, over 5701171.23 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.356, pruned_loss=0.1164, over 5680523.85 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3586, pruned_loss=0.1046, over 5696726.33 frames. ], batch size: 83, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:39:42,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:39:49,699 INFO [optim.py:369] (1/2) 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 11:40:05,385 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 36150, giga_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08954, over 28666.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3601, pruned_loss=0.1051, over 5694271.30 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3564, pruned_loss=0.1166, over 5677435.94 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3585, pruned_loss=0.1035, over 5693631.59 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:40:20,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3008, 4.1174, 3.9207, 1.8244], device='cuda:1'), covar=tensor([0.0498, 0.0651, 0.0647, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.1026, 0.0956, 0.0829, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 11:40:54,518 INFO [train.py:968] (1/2) Epoch 12, batch 36200, giga_loss[loss=0.3059, simple_loss=0.3763, pruned_loss=0.1178, over 28667.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3584, pruned_loss=0.1031, over 5696435.25 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3567, pruned_loss=0.1167, over 5677352.40 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3569, pruned_loss=0.1016, over 5696309.87 frames. ], batch size: 262, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:41:15,720 INFO [optim.py:369] (1/2) 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 11:41:36,539 INFO [train.py:968] (1/2) Epoch 12, batch 36250, giga_loss[loss=0.2765, simple_loss=0.3562, pruned_loss=0.09843, over 28770.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3585, pruned_loss=0.1034, over 5689897.46 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.358, pruned_loss=0.1175, over 5679137.29 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3562, pruned_loss=0.1011, over 5688722.10 frames. ], batch size: 99, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:41:43,711 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:41:45,452 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:03,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:06,575 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:09,364 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 36300, giga_loss[loss=0.2889, simple_loss=0.3647, pruned_loss=0.1066, over 28968.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3604, pruned_loss=0.1059, over 5693240.46 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3585, pruned_loss=0.1176, over 5685419.27 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3581, pruned_loss=0.1035, over 5686880.28 frames. ], batch size: 164, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:42:20,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 11:42:30,419 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:42:38,449 INFO [optim.py:369] (1/2) 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 11:42:44,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6449, 1.7052, 1.7349, 1.5865], device='cuda:1'), covar=tensor([0.1237, 0.1580, 0.1677, 0.1468], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0730, 0.0671, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 11:42:58,290 INFO [train.py:968] (1/2) Epoch 12, batch 36350, giga_loss[loss=0.3569, simple_loss=0.4046, pruned_loss=0.1546, over 28350.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3643, pruned_loss=0.111, over 5697382.46 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3592, pruned_loss=0.1181, over 5689188.61 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3619, pruned_loss=0.1084, over 5688906.88 frames. ], batch size: 65, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:43:13,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:43:24,245 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 36400, giga_loss[loss=0.2419, simple_loss=0.3205, pruned_loss=0.0816, over 28893.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3646, pruned_loss=0.113, over 5694699.86 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3592, pruned_loss=0.1181, over 5690281.67 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3628, pruned_loss=0.1109, over 5687203.54 frames. ], batch size: 174, lr: 2.66e-03, grad_scale: 8.0
+2023-03-06 11:44:08,485 INFO [optim.py:369] (1/2) 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 11:44:30,039 INFO [train.py:968] (1/2) Epoch 12, batch 36450, giga_loss[loss=0.2624, simple_loss=0.3339, pruned_loss=0.09546, over 28596.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.363, pruned_loss=0.1129, over 5688401.88 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.36, pruned_loss=0.1186, over 5684829.07 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.361, pruned_loss=0.1108, over 5687376.49 frames. ], batch size: 85, lr: 2.66e-03, grad_scale: 4.0
+2023-03-06 11:44:46,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1845, 2.2906, 1.5295, 1.7532], device='cuda:1'), covar=tensor([0.0812, 0.0598, 0.0983, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0432, 0.0498, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:45:08,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0042, 2.0266, 1.4137, 1.5504], device='cuda:1'), covar=tensor([0.0835, 0.0680, 0.1021, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0433, 0.0498, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:45:09,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5580, 3.5054, 2.6360, 1.1525], device='cuda:1'), covar=tensor([0.2591, 0.1469, 0.1614, 0.3337], device='cuda:1'), in_proj_covar=tensor([0.1567, 0.1490, 0.1488, 0.1279], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 11:45:10,655 INFO [train.py:968] (1/2) Epoch 12, batch 36500, giga_loss[loss=0.2886, simple_loss=0.3614, pruned_loss=0.1079, over 28651.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3611, pruned_loss=0.1121, over 5699490.22 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3603, pruned_loss=0.1188, over 5689286.33 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3592, pruned_loss=0.11, over 5694746.67 frames. ], batch size: 262, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:45:19,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=537703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:45:21,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=537706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:45:22,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-06 11:45:30,205 INFO [optim.py:369] (1/2) 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 11:45:42,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=537735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:45:48,629 INFO [train.py:968] (1/2) Epoch 12, batch 36550, giga_loss[loss=0.2681, simple_loss=0.3506, pruned_loss=0.09281, over 28323.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3595, pruned_loss=0.1109, over 5688322.78 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3601, pruned_loss=0.1186, over 5678846.20 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3581, pruned_loss=0.1089, over 5694578.39 frames. ], batch size: 368, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:46:28,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7068, 1.9791, 1.5347, 2.3904], device='cuda:1'), covar=tensor([0.2400, 0.2392, 0.2712, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.1320, 0.0975, 0.1163, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 11:46:36,576 INFO [train.py:968] (1/2) Epoch 12, batch 36600, giga_loss[loss=0.2501, simple_loss=0.3312, pruned_loss=0.0845, over 28575.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3576, pruned_loss=0.1085, over 5687766.11 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3605, pruned_loss=0.1188, over 5677212.78 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3561, pruned_loss=0.1066, over 5694025.62 frames. ], batch size: 307, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:47:01,835 INFO [optim.py:369] (1/2) 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 11:47:11,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6850, 1.7368, 1.7551, 1.6059], device='cuda:1'), covar=tensor([0.1676, 0.2379, 0.2115, 0.2085], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0738, 0.0676, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 11:47:24,236 INFO [train.py:968] (1/2) Epoch 12, batch 36650, giga_loss[loss=0.2253, simple_loss=0.3127, pruned_loss=0.06892, over 28799.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3523, pruned_loss=0.1052, over 5686366.95 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3603, pruned_loss=0.1187, over 5679775.48 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3513, pruned_loss=0.1037, over 5689155.62 frames. ], batch size: 242, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:47:33,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5454, 1.6584, 1.2513, 1.2929], device='cuda:1'), covar=tensor([0.0860, 0.0528, 0.1064, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0433, 0.0501, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:48:09,164 INFO [train.py:968] (1/2) Epoch 12, batch 36700, giga_loss[loss=0.2477, simple_loss=0.3188, pruned_loss=0.08827, over 28325.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3473, pruned_loss=0.1022, over 5696135.73 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3609, pruned_loss=0.1189, over 5681326.75 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3457, pruned_loss=0.1004, over 5697459.40 frames. ], batch size: 368, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:48:39,528 INFO [optim.py:369] (1/2) 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 11:49:03,077 INFO [train.py:968] (1/2) Epoch 12, batch 36750, giga_loss[loss=0.2477, simple_loss=0.3208, pruned_loss=0.0873, over 28860.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.342, pruned_loss=0.09961, over 5673358.47 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.361, pruned_loss=0.1188, over 5677218.59 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3399, pruned_loss=0.09761, over 5677956.63 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:49:06,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=537946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:49:09,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2085, 4.0509, 3.8009, 1.7732], device='cuda:1'), covar=tensor([0.0542, 0.0664, 0.0651, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.1038, 0.0970, 0.0844, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 11:49:52,680 INFO [train.py:968] (1/2) Epoch 12, batch 36800, giga_loss[loss=0.2941, simple_loss=0.3699, pruned_loss=0.1092, over 28860.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3385, pruned_loss=0.09699, over 5673637.25 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3612, pruned_loss=0.1189, over 5677997.49 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3367, pruned_loss=0.09527, over 5676603.43 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:50:15,578 INFO [optim.py:369] (1/2) 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 11:50:33,919 INFO [train.py:968] (1/2) Epoch 12, batch 36850, giga_loss[loss=0.2916, simple_loss=0.3546, pruned_loss=0.1143, over 28855.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3401, pruned_loss=0.09741, over 5681413.97 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3621, pruned_loss=0.119, over 5685723.17 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3369, pruned_loss=0.0951, over 5676714.36 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:50:59,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2967, 1.3416, 1.1626, 1.5425], device='cuda:1'), covar=tensor([0.0790, 0.0351, 0.0344, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 11:51:11,569 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=538089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:51:13,154 INFO [train.py:968] (1/2) Epoch 12, batch 36900, giga_loss[loss=0.3275, simple_loss=0.3842, pruned_loss=0.1354, over 28011.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3403, pruned_loss=0.09738, over 5694127.04 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3625, pruned_loss=0.1191, over 5687521.89 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3371, pruned_loss=0.09502, over 5688820.95 frames. ], batch size: 412, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:51:13,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=538092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:51:38,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-06 11:51:39,022 INFO [optim.py:369] (1/2) 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 11:51:39,909 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 36950, giga_loss[loss=0.276, simple_loss=0.3505, pruned_loss=0.1007, over 28761.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3401, pruned_loss=0.0981, over 5688741.46 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3635, pruned_loss=0.1198, over 5690157.43 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3361, pruned_loss=0.09517, over 5682145.11 frames. ], batch size: 284, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:52:37,288 INFO [train.py:968] (1/2) Epoch 12, batch 37000, libri_loss[loss=0.3676, simple_loss=0.4293, pruned_loss=0.153, over 29203.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3386, pruned_loss=0.09747, over 5702936.69 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3646, pruned_loss=0.1202, over 5695945.85 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3337, pruned_loss=0.09409, over 5692487.64 frames. ], batch size: 97, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:52:59,772 INFO [optim.py:369] (1/2) 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 11:53:06,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3607, 2.2300, 2.1320, 2.0393], device='cuda:1'), covar=tensor([0.1385, 0.2223, 0.1869, 0.1963], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0742, 0.0681, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 11:53:15,236 INFO [train.py:968] (1/2) Epoch 12, batch 37050, giga_loss[loss=0.26, simple_loss=0.3276, pruned_loss=0.0962, over 28622.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3373, pruned_loss=0.09693, over 5709892.43 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3654, pruned_loss=0.1207, over 5696828.07 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3318, pruned_loss=0.0931, over 5701023.95 frames. ], batch size: 78, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:53:34,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8006, 1.9071, 1.4421, 1.4971], device='cuda:1'), covar=tensor([0.0832, 0.0571, 0.0952, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0435, 0.0503, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 11:53:48,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3425, 3.3426, 1.5178, 1.4988], device='cuda:1'), covar=tensor([0.1010, 0.0313, 0.0904, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0504, 0.0342, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 11:53:52,748 INFO [train.py:968] (1/2) Epoch 12, batch 37100, giga_loss[loss=0.2694, simple_loss=0.3386, pruned_loss=0.1001, over 29013.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3358, pruned_loss=0.09632, over 5711552.25 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3659, pruned_loss=0.1208, over 5693574.73 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3301, pruned_loss=0.09245, over 5708178.57 frames. ], batch size: 164, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:54:16,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538320.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 11:54:17,026 INFO [optim.py:369] (1/2) 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 11:54:33,475 INFO [train.py:968] (1/2) Epoch 12, batch 37150, giga_loss[loss=0.2371, simple_loss=0.3097, pruned_loss=0.0822, over 28199.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3335, pruned_loss=0.09551, over 5706685.89 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3658, pruned_loss=0.1205, over 5696076.02 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3285, pruned_loss=0.09223, over 5701994.74 frames. ], batch size: 77, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:55:15,000 INFO [train.py:968] (1/2) Epoch 12, batch 37200, giga_loss[loss=0.2539, simple_loss=0.3278, pruned_loss=0.09003, over 28844.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3303, pruned_loss=0.09365, over 5713347.37 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.366, pruned_loss=0.1206, over 5698371.58 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3257, pruned_loss=0.09066, over 5707617.17 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:55:38,637 INFO [optim.py:369] (1/2) 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 11:55:55,826 INFO [train.py:968] (1/2) Epoch 12, batch 37250, giga_loss[loss=0.2537, simple_loss=0.3269, pruned_loss=0.09028, over 28923.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3283, pruned_loss=0.09273, over 5708390.00 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3665, pruned_loss=0.1207, over 5690382.38 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3233, pruned_loss=0.08955, over 5711613.86 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:56:05,972 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:56:12,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-06 11:56:35,760 INFO [train.py:968] (1/2) Epoch 12, batch 37300, giga_loss[loss=0.212, simple_loss=0.2944, pruned_loss=0.06479, over 29116.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3255, pruned_loss=0.09079, over 5714916.48 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.367, pruned_loss=0.1209, over 5688751.19 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3207, pruned_loss=0.08783, over 5719035.33 frames. ], batch size: 155, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:57:00,196 INFO [optim.py:369] (1/2) 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 11:57:07,395 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:57:18,045 INFO [train.py:968] (1/2) Epoch 12, batch 37350, libri_loss[loss=0.3632, simple_loss=0.4232, pruned_loss=0.1516, over 19404.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3258, pruned_loss=0.09091, over 5705098.51 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3681, pruned_loss=0.1214, over 5675919.69 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3197, pruned_loss=0.08715, over 5722539.67 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 11:57:19,588 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=538544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 11:57:54,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3594, 1.6445, 1.3121, 1.4241], device='cuda:1'), covar=tensor([0.2316, 0.2214, 0.2411, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1331, 0.0979, 0.1172, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 11:57:54,998 INFO [train.py:968] (1/2) Epoch 12, batch 37400, giga_loss[loss=0.272, simple_loss=0.3481, pruned_loss=0.09796, over 28977.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3281, pruned_loss=0.09226, over 5703690.87 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3684, pruned_loss=0.1214, over 5674331.65 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3212, pruned_loss=0.08795, over 5719765.81 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:58:02,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-06 11:58:23,541 INFO [optim.py:369] (1/2) 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 11:58:40,465 INFO [train.py:968] (1/2) Epoch 12, batch 37450, giga_loss[loss=0.3148, simple_loss=0.3779, pruned_loss=0.1259, over 28557.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3331, pruned_loss=0.09542, over 5703260.19 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3689, pruned_loss=0.1218, over 5677431.57 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.326, pruned_loss=0.09084, over 5714272.97 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:59:23,534 INFO [train.py:968] (1/2) Epoch 12, batch 37500, libri_loss[loss=0.2612, simple_loss=0.3197, pruned_loss=0.1014, over 29640.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3403, pruned_loss=0.1002, over 5695410.24 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3695, pruned_loss=0.1223, over 5680691.55 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3321, pruned_loss=0.0946, over 5702684.77 frames. ], batch size: 69, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 11:59:25,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538695.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 11:59:36,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-06 11:59:55,188 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 37550, giga_loss[loss=0.2956, simple_loss=0.3658, pruned_loss=0.1127, over 29115.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.35, pruned_loss=0.1071, over 5690934.93 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3703, pruned_loss=0.1229, over 5684769.35 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3422, pruned_loss=0.1018, over 5693136.99 frames. ], batch size: 128, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:00:12,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2793, 1.6011, 1.4131, 1.5188], device='cuda:1'), covar=tensor([0.0772, 0.0311, 0.0301, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 12:01:02,510 INFO [train.py:968] (1/2) Epoch 12, batch 37600, giga_loss[loss=0.2953, simple_loss=0.3717, pruned_loss=0.1095, over 28727.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3542, pruned_loss=0.109, over 5678704.60 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1227, over 5689330.58 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3474, pruned_loss=0.1044, over 5676323.58 frames. ], batch size: 242, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:01:27,580 INFO [optim.py:369] (1/2) 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 12:01:36,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:01:43,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=538838.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:01:45,659 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=538841.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:01:45,981 INFO [train.py:968] (1/2) Epoch 12, batch 37650, giga_loss[loss=0.3357, simple_loss=0.4036, pruned_loss=0.1339, over 28862.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3583, pruned_loss=0.1104, over 5676031.86 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3701, pruned_loss=0.1225, over 5685377.50 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3528, pruned_loss=0.1066, over 5678071.46 frames. ], batch size: 174, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:02:09,960 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=538870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 12:02:28,760 INFO [train.py:968] (1/2) Epoch 12, batch 37700, giga_loss[loss=0.303, simple_loss=0.3781, pruned_loss=0.1139, over 28948.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3636, pruned_loss=0.1139, over 5656359.15 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.123, over 5672163.14 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3583, pruned_loss=0.11, over 5671112.50 frames. ], batch size: 213, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:02:39,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:02:50,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=538919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:02:52,592 INFO [optim.py:369] (1/2) 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 12:03:07,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9702, 1.1719, 1.0736, 0.9586], device='cuda:1'), covar=tensor([0.1583, 0.2008, 0.0949, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.1726, 0.1615, 0.1584, 0.1692], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 12:03:08,224 INFO [train.py:968] (1/2) Epoch 12, batch 37750, giga_loss[loss=0.2517, simple_loss=0.3305, pruned_loss=0.08641, over 29054.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3649, pruned_loss=0.1146, over 5656346.07 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3707, pruned_loss=0.1231, over 5668695.88 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3603, pruned_loss=0.1109, over 5670645.69 frames. ], batch size: 155, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:03:34,982 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=538974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:03:37,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=538977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:03:49,781 INFO [train.py:968] (1/2) Epoch 12, batch 37800, giga_loss[loss=0.2522, simple_loss=0.334, pruned_loss=0.08523, over 28795.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3584, pruned_loss=0.1095, over 5658661.45 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3707, pruned_loss=0.1233, over 5659954.61 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3547, pruned_loss=0.1064, over 5676698.09 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:04:02,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:18,407 INFO [optim.py:369] (1/2) 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 12:04:36,543 INFO [train.py:968] (1/2) Epoch 12, batch 37850, giga_loss[loss=0.2702, simple_loss=0.3454, pruned_loss=0.09752, over 28694.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3563, pruned_loss=0.1071, over 5669777.84 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3707, pruned_loss=0.1233, over 5659954.61 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3534, pruned_loss=0.1047, over 5683815.98 frames. ], batch size: 262, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:04:42,702 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:44,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:54,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:04:56,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:05:08,630 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 37900, giga_loss[loss=0.2956, simple_loss=0.3652, pruned_loss=0.113, over 28764.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3548, pruned_loss=0.1059, over 5673617.75 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3706, pruned_loss=0.1233, over 5666832.62 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.352, pruned_loss=0.1032, over 5678833.25 frames. ], batch size: 119, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:05:19,970 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:05:24,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3672, 1.7004, 1.6297, 1.2204], device='cuda:1'), covar=tensor([0.1625, 0.2146, 0.1299, 0.1485], device='cuda:1'), in_proj_covar=tensor([0.0844, 0.0690, 0.0882, 0.0786], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 12:05:32,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=539109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:05:40,341 INFO [optim.py:369] (1/2) 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 12:05:58,596 INFO [train.py:968] (1/2) Epoch 12, batch 37950, giga_loss[loss=0.2796, simple_loss=0.3406, pruned_loss=0.1093, over 23747.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3573, pruned_loss=0.1074, over 5680216.76 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3718, pruned_loss=0.1243, over 5673083.31 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3533, pruned_loss=0.1036, over 5679421.23 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:06:41,210 INFO [train.py:968] (1/2) Epoch 12, batch 38000, giga_loss[loss=0.3174, simple_loss=0.382, pruned_loss=0.1263, over 28607.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3605, pruned_loss=0.1097, over 5686978.54 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3721, pruned_loss=0.1244, over 5677326.58 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3569, pruned_loss=0.1063, over 5682913.77 frames. ], batch size: 60, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:07:12,305 INFO [optim.py:369] (1/2) 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 12:07:20,024 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 38050, giga_loss[loss=0.2525, simple_loss=0.3309, pruned_loss=0.08705, over 28465.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3621, pruned_loss=0.1109, over 5684793.47 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3726, pruned_loss=0.1247, over 5671055.00 frames. ], giga_tot_loss[loss=0.2872, simple_loss=0.3587, pruned_loss=0.1078, over 5687492.93 frames. ], batch size: 71, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:07:46,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9271, 0.9514, 3.4108, 2.9250], device='cuda:1'), covar=tensor([0.1747, 0.2744, 0.0475, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0666, 0.0590, 0.0858, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 12:08:11,469 INFO [train.py:968] (1/2) Epoch 12, batch 38100, giga_loss[loss=0.2925, simple_loss=0.3594, pruned_loss=0.1128, over 28928.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3622, pruned_loss=0.111, over 5692490.32 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1246, over 5676504.54 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3589, pruned_loss=0.1081, over 5690177.38 frames. ], batch size: 227, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:08:39,431 INFO [optim.py:369] (1/2) 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 12:08:54,442 INFO [train.py:968] (1/2) Epoch 12, batch 38150, giga_loss[loss=0.3026, simple_loss=0.3691, pruned_loss=0.1181, over 28875.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3625, pruned_loss=0.1116, over 5692561.20 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1244, over 5677931.08 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3598, pruned_loss=0.1093, over 5689627.66 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:09:23,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5773, 2.0131, 1.7244, 1.5486], device='cuda:1'), covar=tensor([0.2506, 0.1764, 0.1886, 0.2078], device='cuda:1'), in_proj_covar=tensor([0.1726, 0.1611, 0.1577, 0.1673], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 12:09:39,188 INFO [train.py:968] (1/2) Epoch 12, batch 38200, giga_loss[loss=0.2701, simple_loss=0.3406, pruned_loss=0.0998, over 28944.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3614, pruned_loss=0.1102, over 5699088.61 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1241, over 5683242.34 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3593, pruned_loss=0.1083, over 5692520.28 frames. ], batch size: 227, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:09:50,642 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 12:10:02,033 INFO [optim.py:369] (1/2) 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 12:10:16,304 INFO [train.py:968] (1/2) Epoch 12, batch 38250, giga_loss[loss=0.2679, simple_loss=0.3475, pruned_loss=0.09412, over 28945.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.361, pruned_loss=0.1087, over 5699165.96 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1241, over 5676614.35 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3591, pruned_loss=0.107, over 5699912.45 frames. ], batch size: 106, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:10:52,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0842, 2.9397, 1.8276, 1.0712], device='cuda:1'), covar=tensor([0.5287, 0.2200, 0.3285, 0.5173], device='cuda:1'), in_proj_covar=tensor([0.1564, 0.1482, 0.1490, 0.1281], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 12:10:52,817 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=539484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:10:58,776 INFO [train.py:968] (1/2) Epoch 12, batch 38300, giga_loss[loss=0.2734, simple_loss=0.3511, pruned_loss=0.09788, over 28617.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3607, pruned_loss=0.1076, over 5702552.27 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3719, pruned_loss=0.1237, over 5681263.42 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3594, pruned_loss=0.1062, over 5699455.19 frames. ], batch size: 78, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:11:01,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1113, 1.2078, 3.6500, 3.0529], device='cuda:1'), covar=tensor([0.1720, 0.2645, 0.0446, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0666, 0.0592, 0.0855, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 12:11:23,011 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 38350, giga_loss[loss=0.2834, simple_loss=0.3568, pruned_loss=0.105, over 28769.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3593, pruned_loss=0.1065, over 5707251.01 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3725, pruned_loss=0.1241, over 5681563.86 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3575, pruned_loss=0.1046, over 5704952.16 frames. ], batch size: 119, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:12:15,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9752, 2.0068, 1.8130, 1.8603], device='cuda:1'), covar=tensor([0.1533, 0.2171, 0.2015, 0.1870], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0730, 0.0674, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 12:12:21,313 INFO [train.py:968] (1/2) Epoch 12, batch 38400, giga_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09677, over 28516.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3584, pruned_loss=0.1069, over 5692161.23 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5667478.26 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3563, pruned_loss=0.1047, over 5702761.09 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:12:31,867 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=539607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:12:45,385 INFO [optim.py:369] (1/2) 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 12:12:48,420 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:12:51,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:13:01,161 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5010, 1.7687, 1.7660, 1.3100], device='cuda:1'), covar=tensor([0.1675, 0.2460, 0.1338, 0.1619], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0696, 0.0883, 0.0786], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 12:13:02,176 INFO [train.py:968] (1/2) Epoch 12, batch 38450, giga_loss[loss=0.2677, simple_loss=0.3336, pruned_loss=0.1009, over 28692.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1054, over 5702608.90 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3726, pruned_loss=0.1243, over 5669620.61 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3541, pruned_loss=0.1034, over 5710070.44 frames. ], batch size: 85, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:13:15,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:13:17,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=539662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:13:40,896 INFO [train.py:968] (1/2) Epoch 12, batch 38500, giga_loss[loss=0.3255, simple_loss=0.3629, pruned_loss=0.1441, over 23560.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3557, pruned_loss=0.1057, over 5706395.26 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1242, over 5676185.56 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3535, pruned_loss=0.1036, over 5707279.77 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:13:54,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4637, 1.8103, 1.4263, 1.6530], device='cuda:1'), covar=tensor([0.2441, 0.2295, 0.2588, 0.2168], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.0972, 0.1154, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 12:14:05,451 INFO [optim.py:369] (1/2) 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 12:14:19,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5763, 1.7529, 1.7752, 1.3875], device='cuda:1'), covar=tensor([0.1678, 0.2445, 0.1337, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0696, 0.0883, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 12:14:20,067 INFO [train.py:968] (1/2) Epoch 12, batch 38550, giga_loss[loss=0.2748, simple_loss=0.3455, pruned_loss=0.102, over 28259.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3566, pruned_loss=0.1069, over 5714025.59 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5682904.91 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3552, pruned_loss=0.1052, over 5709577.33 frames. ], batch size: 77, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:14:28,154 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=539750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:14:30,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=539753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:14:36,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2071, 1.2105, 3.4913, 3.1117], device='cuda:1'), covar=tensor([0.1573, 0.2778, 0.0415, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0665, 0.0591, 0.0857, 0.0770], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 12:14:52,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=539782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:15:00,110 INFO [train.py:968] (1/2) Epoch 12, batch 38600, giga_loss[loss=0.2801, simple_loss=0.3591, pruned_loss=0.1006, over 28822.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3565, pruned_loss=0.1062, over 5715759.93 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5683419.58 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.355, pruned_loss=0.1044, over 5712104.44 frames. ], batch size: 199, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:15:01,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7780, 2.0632, 1.6099, 2.1642], device='cuda:1'), covar=tensor([0.2276, 0.2189, 0.2460, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.1319, 0.0970, 0.1156, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 12:15:11,288 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=539807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:15:19,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5754, 4.2225, 1.7970, 1.8076], device='cuda:1'), covar=tensor([0.0952, 0.0206, 0.0908, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0357, 0.0499, 0.0339, 0.0385], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0025], device='cuda:1')
+2023-03-06 12:15:24,509 INFO [optim.py:369] (1/2) 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 12:15:38,984 INFO [train.py:968] (1/2) Epoch 12, batch 38650, libri_loss[loss=0.2754, simple_loss=0.3434, pruned_loss=0.1037, over 29470.00 frames. ], tot_loss[loss=0.282, simple_loss=0.355, pruned_loss=0.1045, over 5701600.44 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5675912.09 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3534, pruned_loss=0.1026, over 5705922.26 frames. ], batch size: 70, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:16:18,850 INFO [train.py:968] (1/2) Epoch 12, batch 38700, giga_loss[loss=0.2951, simple_loss=0.3675, pruned_loss=0.1113, over 28716.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3536, pruned_loss=0.1031, over 5707820.98 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3721, pruned_loss=0.1238, over 5674621.25 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3521, pruned_loss=0.1014, over 5712678.96 frames. ], batch size: 262, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:16:45,992 INFO [optim.py:369] (1/2) 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 12:16:58,465 INFO [train.py:968] (1/2) Epoch 12, batch 38750, libri_loss[loss=0.302, simple_loss=0.3662, pruned_loss=0.119, over 29530.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3521, pruned_loss=0.1026, over 5702004.93 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1237, over 5670615.18 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3507, pruned_loss=0.1008, over 5709333.80 frames. ], batch size: 84, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:17:24,789 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-06 12:17:37,867 INFO [train.py:968] (1/2) Epoch 12, batch 38800, giga_loss[loss=0.2738, simple_loss=0.3485, pruned_loss=0.09958, over 27635.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3495, pruned_loss=0.1016, over 5700311.31 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3716, pruned_loss=0.1235, over 5672394.95 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3482, pruned_loss=0.1, over 5705000.35 frames. ], batch size: 472, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:17:51,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4605, 1.7085, 1.7092, 1.3056], device='cuda:1'), covar=tensor([0.1716, 0.2379, 0.1388, 0.1607], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0697, 0.0886, 0.0790], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 12:18:02,580 INFO [optim.py:369] (1/2) 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 12:18:12,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:18:15,534 INFO [train.py:968] (1/2) Epoch 12, batch 38850, giga_loss[loss=0.2844, simple_loss=0.3509, pruned_loss=0.109, over 29020.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.347, pruned_loss=0.1006, over 5696376.75 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3714, pruned_loss=0.1232, over 5666557.09 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3454, pruned_loss=0.09885, over 5706462.91 frames. ], batch size: 164, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:18:33,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 12:18:43,024 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:18:53,292 INFO [train.py:968] (1/2) Epoch 12, batch 38900, giga_loss[loss=0.2518, simple_loss=0.3299, pruned_loss=0.08684, over 28972.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3467, pruned_loss=0.1011, over 5704862.67 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.1231, over 5676598.76 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.345, pruned_loss=0.09881, over 5705101.17 frames. ], batch size: 136, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:19:00,686 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:19:24,142 INFO [optim.py:369] (1/2) 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 12:19:36,782 INFO [train.py:968] (1/2) Epoch 12, batch 38950, giga_loss[loss=0.2376, simple_loss=0.3176, pruned_loss=0.07881, over 28897.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.347, pruned_loss=0.102, over 5682798.71 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3706, pruned_loss=0.1231, over 5660155.06 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3455, pruned_loss=0.1, over 5698888.06 frames. ], batch size: 174, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:20:06,026 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:07,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:09,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:15,668 INFO [train.py:968] (1/2) Epoch 12, batch 39000, giga_loss[loss=0.2551, simple_loss=0.331, pruned_loss=0.08958, over 29059.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3443, pruned_loss=0.1005, over 5693259.41 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.123, over 5663976.13 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3428, pruned_loss=0.09866, over 5702870.45 frames. ], batch size: 155, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:20:15,669 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 12:20:24,144 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 12:20:39,512 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:20:49,193 INFO [optim.py:369] (1/2) 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 12:20:57,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:21:01,662 INFO [train.py:968] (1/2) Epoch 12, batch 39050, giga_loss[loss=0.3019, simple_loss=0.3646, pruned_loss=0.1196, over 28613.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.342, pruned_loss=0.09947, over 5700940.81 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3709, pruned_loss=0.1232, over 5666051.91 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3399, pruned_loss=0.09744, over 5707218.73 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:21:43,914 INFO [train.py:968] (1/2) Epoch 12, batch 39100, giga_loss[loss=0.2409, simple_loss=0.3169, pruned_loss=0.08248, over 28688.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3395, pruned_loss=0.09847, over 5698577.79 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.371, pruned_loss=0.1234, over 5666416.81 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3375, pruned_loss=0.09644, over 5703446.58 frames. ], batch size: 242, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:22:12,255 INFO [optim.py:369] (1/2) 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 12:22:13,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:22:16,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:22:28,384 INFO [train.py:968] (1/2) Epoch 12, batch 39150, giga_loss[loss=0.2571, simple_loss=0.3364, pruned_loss=0.08891, over 28916.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3384, pruned_loss=0.0973, over 5702827.22 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3711, pruned_loss=0.1233, over 5668482.24 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3364, pruned_loss=0.09554, over 5705189.23 frames. ], batch size: 199, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:22:36,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4388, 1.7885, 1.2243, 0.8787], device='cuda:1'), covar=tensor([0.4277, 0.2697, 0.2595, 0.4116], device='cuda:1'), in_proj_covar=tensor([0.1575, 0.1487, 0.1500, 0.1286], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 12:22:42,640 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:23:10,768 INFO [train.py:968] (1/2) Epoch 12, batch 39200, giga_loss[loss=0.2937, simple_loss=0.369, pruned_loss=0.1092, over 28536.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3405, pruned_loss=0.09797, over 5702355.50 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1234, over 5671903.73 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3384, pruned_loss=0.09598, over 5701534.62 frames. ], batch size: 336, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:23:32,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9532, 1.0053, 3.5905, 2.8674], device='cuda:1'), covar=tensor([0.1828, 0.2899, 0.0412, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0658, 0.0588, 0.0854, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 12:23:38,969 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 39250, libri_loss[loss=0.3107, simple_loss=0.37, pruned_loss=0.1257, over 29590.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3455, pruned_loss=0.101, over 5688005.43 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3718, pruned_loss=0.1238, over 5670446.35 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3423, pruned_loss=0.09827, over 5689001.82 frames. ], batch size: 76, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:24:05,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:24:25,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:24:38,933 INFO [train.py:968] (1/2) Epoch 12, batch 39300, libri_loss[loss=0.2491, simple_loss=0.32, pruned_loss=0.08907, over 29338.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3477, pruned_loss=0.1019, over 5681568.34 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3718, pruned_loss=0.124, over 5660915.61 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3445, pruned_loss=0.09899, over 5692139.00 frames. ], batch size: 71, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:24:50,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-06 12:25:08,334 INFO [optim.py:369] (1/2) 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 12:25:20,113 INFO [train.py:968] (1/2) Epoch 12, batch 39350, giga_loss[loss=0.2965, simple_loss=0.3584, pruned_loss=0.1173, over 26552.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3482, pruned_loss=0.1014, over 5688493.96 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3716, pruned_loss=0.1237, over 5669664.13 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3448, pruned_loss=0.09846, over 5689861.22 frames. ], batch size: 555, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:26:02,463 INFO [train.py:968] (1/2) Epoch 12, batch 39400, giga_loss[loss=0.2878, simple_loss=0.3578, pruned_loss=0.1088, over 28972.00 frames. ], tot_loss[loss=0.275, simple_loss=0.348, pruned_loss=0.101, over 5701541.85 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5676711.31 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3443, pruned_loss=0.09795, over 5696967.30 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:26:05,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:07,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:16,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=540611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:21,868 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:24,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:26,060 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-06 12:26:27,616 INFO [optim.py:369] (1/2) 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 12:26:28,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6896, 4.4883, 4.2747, 1.9380], device='cuda:1'), covar=tensor([0.0586, 0.0744, 0.0744, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0980, 0.0850, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 12:26:30,163 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:26:34,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4942, 1.6638, 1.4920, 1.3976], device='cuda:1'), covar=tensor([0.2643, 0.2056, 0.1518, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.1721, 0.1618, 0.1581, 0.1674], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 12:26:40,406 INFO [train.py:968] (1/2) Epoch 12, batch 39450, giga_loss[loss=0.2683, simple_loss=0.3354, pruned_loss=0.1006, over 28895.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3485, pruned_loss=0.1018, over 5706003.55 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3721, pruned_loss=0.124, over 5684567.36 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3446, pruned_loss=0.09846, over 5695833.68 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:26:48,989 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=540649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:27:11,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1157, 1.2260, 3.7906, 3.1324], device='cuda:1'), covar=tensor([0.1716, 0.2621, 0.0431, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0660, 0.0589, 0.0856, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 12:27:15,199 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5645, 1.6856, 1.7917, 1.3617], device='cuda:1'), covar=tensor([0.1683, 0.2219, 0.1409, 0.1526], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0691, 0.0882, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 12:27:22,840 INFO [train.py:968] (1/2) Epoch 12, batch 39500, giga_loss[loss=0.2629, simple_loss=0.3432, pruned_loss=0.09133, over 28657.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3501, pruned_loss=0.1027, over 5704894.33 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3729, pruned_loss=0.1243, over 5675812.10 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3453, pruned_loss=0.09891, over 5706329.45 frames. ], batch size: 242, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:27:45,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:27:53,549 INFO [optim.py:369] (1/2) 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 12:28:02,479 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 39550, giga_loss[loss=0.288, simple_loss=0.3467, pruned_loss=0.1147, over 23951.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3505, pruned_loss=0.103, over 5702621.84 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3726, pruned_loss=0.1241, over 5671730.97 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3464, pruned_loss=0.09956, over 5709080.68 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:28:12,417 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 12:28:14,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=540754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:28:17,163 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=540757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:28:43,709 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 39600, giga_loss[loss=0.2997, simple_loss=0.3702, pruned_loss=0.1146, over 28880.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3542, pruned_loss=0.1047, over 5701468.79 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3724, pruned_loss=0.124, over 5674004.88 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3509, pruned_loss=0.1018, over 5704984.17 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:29:16,201 INFO [optim.py:369] (1/2) 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 12:29:16,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=540826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:29:28,526 INFO [train.py:968] (1/2) Epoch 12, batch 39650, giga_loss[loss=0.2978, simple_loss=0.3684, pruned_loss=0.1136, over 28815.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3565, pruned_loss=0.1057, over 5710391.16 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5678312.00 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.353, pruned_loss=0.103, over 5710516.35 frames. ], batch size: 119, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:29:31,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8686, 3.6711, 3.4318, 2.3571], device='cuda:1'), covar=tensor([0.0531, 0.0701, 0.0680, 0.1499], device='cuda:1'), in_proj_covar=tensor([0.1058, 0.0981, 0.0855, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 12:29:53,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6060, 1.7186, 1.6949, 1.6047], device='cuda:1'), covar=tensor([0.1571, 0.2093, 0.1906, 0.1951], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0726, 0.0670, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 12:30:09,646 INFO [train.py:968] (1/2) Epoch 12, batch 39700, giga_loss[loss=0.2823, simple_loss=0.3552, pruned_loss=0.1046, over 29034.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.357, pruned_loss=0.106, over 5710993.96 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3727, pruned_loss=0.1237, over 5682938.51 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3538, pruned_loss=0.1034, over 5707690.79 frames. ], batch size: 128, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:30:40,434 INFO [optim.py:369] (1/2) 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 12:30:49,580 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 39750, giga_loss[loss=0.2717, simple_loss=0.3432, pruned_loss=0.1001, over 28918.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3579, pruned_loss=0.1062, over 5712764.55 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3725, pruned_loss=0.1236, over 5685354.89 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3554, pruned_loss=0.1041, over 5708311.35 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:31:25,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3297, 1.5064, 1.1983, 1.4790], device='cuda:1'), covar=tensor([0.0663, 0.0372, 0.0337, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 12:31:31,679 INFO [train.py:968] (1/2) Epoch 12, batch 39800, giga_loss[loss=0.2681, simple_loss=0.349, pruned_loss=0.09366, over 28926.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3588, pruned_loss=0.1071, over 5700753.97 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3727, pruned_loss=0.1238, over 5677585.17 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3565, pruned_loss=0.105, over 5704888.57 frames. ], batch size: 174, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:31:55,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3343, 4.2535, 1.5782, 1.5654], device='cuda:1'), covar=tensor([0.0982, 0.0278, 0.0939, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0507, 0.0343, 0.0389], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 12:31:59,346 INFO [optim.py:369] (1/2) 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 12:32:03,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4537, 1.6789, 1.6257, 1.5299], device='cuda:1'), covar=tensor([0.1468, 0.1712, 0.1856, 0.1658], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0723, 0.0670, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 12:32:11,253 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-06 12:32:11,490 INFO [train.py:968] (1/2) Epoch 12, batch 39850, libri_loss[loss=0.3402, simple_loss=0.4045, pruned_loss=0.1379, over 29147.00 frames. ], tot_loss[loss=0.286, simple_loss=0.358, pruned_loss=0.107, over 5709547.93 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3725, pruned_loss=0.1236, over 5681655.58 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3557, pruned_loss=0.105, over 5710105.44 frames. ], batch size: 101, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:32:50,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:32:51,110 INFO [train.py:968] (1/2) Epoch 12, batch 39900, giga_loss[loss=0.2358, simple_loss=0.3159, pruned_loss=0.07781, over 28926.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3547, pruned_loss=0.1054, over 5716533.70 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5684362.95 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3524, pruned_loss=0.1033, over 5715017.97 frames. ], batch size: 145, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:33:09,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:33:20,027 INFO [optim.py:369] (1/2) 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 12:33:27,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3449, 1.5904, 1.3085, 1.4510], device='cuda:1'), covar=tensor([0.0728, 0.0319, 0.0334, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 12:33:34,111 INFO [train.py:968] (1/2) Epoch 12, batch 39950, giga_loss[loss=0.2564, simple_loss=0.3358, pruned_loss=0.08847, over 29006.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3523, pruned_loss=0.1044, over 5705779.87 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3736, pruned_loss=0.1244, over 5679607.74 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3493, pruned_loss=0.1018, over 5709160.21 frames. ], batch size: 213, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:33:48,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9722, 1.2652, 0.9866, 0.1617], device='cuda:1'), covar=tensor([0.2203, 0.1708, 0.2615, 0.4300], device='cuda:1'), in_proj_covar=tensor([0.1569, 0.1483, 0.1498, 0.1283], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 12:34:11,276 INFO [train.py:968] (1/2) Epoch 12, batch 40000, giga_loss[loss=0.291, simple_loss=0.3705, pruned_loss=0.1057, over 27615.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.352, pruned_loss=0.1034, over 5704310.43 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3738, pruned_loss=0.1244, over 5673786.98 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3488, pruned_loss=0.1006, over 5712941.06 frames. ], batch size: 472, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:34:18,777 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:34:47,205 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:34:49,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:34:54,664 INFO [train.py:968] (1/2) Epoch 12, batch 40050, giga_loss[loss=0.2623, simple_loss=0.3344, pruned_loss=0.09512, over 28793.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3542, pruned_loss=0.103, over 5699979.18 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1248, over 5676457.11 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.351, pruned_loss=0.1002, over 5704935.23 frames. ], batch size: 99, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:35:06,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:09,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:16,244 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:34,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:37,496 INFO [train.py:968] (1/2) Epoch 12, batch 40100, giga_loss[loss=0.2779, simple_loss=0.3544, pruned_loss=0.1007, over 28573.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1024, over 5705728.37 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3745, pruned_loss=0.125, over 5678971.63 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.35, pruned_loss=0.09981, over 5707512.67 frames. ], batch size: 307, lr: 2.65e-03, grad_scale: 8.0
+2023-03-06 12:35:52,143 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=541312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:35:52,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541312.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:36:05,245 INFO [optim.py:369] (1/2) 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 12:36:16,873 INFO [train.py:968] (1/2) Epoch 12, batch 40150, giga_loss[loss=0.2763, simple_loss=0.3475, pruned_loss=0.1025, over 28815.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3523, pruned_loss=0.1032, over 5707225.81 frames. ], libri_tot_loss[loss=0.3129, simple_loss=0.3751, pruned_loss=0.1253, over 5683330.80 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.349, pruned_loss=0.1005, over 5705437.27 frames. ], batch size: 186, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:36:18,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9617, 1.2583, 1.0107, 0.1241], device='cuda:1'), covar=tensor([0.2846, 0.2111, 0.3077, 0.5367], device='cuda:1'), in_proj_covar=tensor([0.1578, 0.1489, 0.1501, 0.1293], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 12:36:20,694 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:36:22,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:36:43,439 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 40200, giga_loss[loss=0.3125, simple_loss=0.3662, pruned_loss=0.1294, over 24131.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.351, pruned_loss=0.1039, over 5706347.39 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.375, pruned_loss=0.1251, over 5687851.96 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3478, pruned_loss=0.1012, over 5701491.04 frames. ], batch size: 705, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:37:29,950 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 40250, libri_loss[loss=0.3277, simple_loss=0.364, pruned_loss=0.1457, over 27253.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3504, pruned_loss=0.105, over 5712917.52 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3744, pruned_loss=0.1247, over 5691634.46 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3474, pruned_loss=0.1024, over 5706644.02 frames. ], batch size: 60, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:37:46,318 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-06 12:37:49,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541455.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:37:51,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541458.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 12:38:01,330 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-06 12:38:16,891 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541487.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 12:38:21,390 INFO [train.py:968] (1/2) Epoch 12, batch 40300, giga_loss[loss=0.31, simple_loss=0.3821, pruned_loss=0.119, over 28624.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3487, pruned_loss=0.1044, over 5721598.68 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3744, pruned_loss=0.1247, over 5696736.42 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3458, pruned_loss=0.1019, over 5712883.24 frames. ], batch size: 307, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:38:51,079 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 40350, giga_loss[loss=0.2635, simple_loss=0.3402, pruned_loss=0.09345, over 28566.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.347, pruned_loss=0.1034, over 5700477.82 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5671664.77 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3437, pruned_loss=0.1006, over 5715785.35 frames. ], batch size: 307, lr: 2.65e-03, grad_scale: 2.0
+2023-03-06 12:39:05,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-06 12:39:38,672 INFO [train.py:968] (1/2) Epoch 12, batch 40400, giga_loss[loss=0.2179, simple_loss=0.3015, pruned_loss=0.06721, over 29074.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.343, pruned_loss=0.1012, over 5710061.55 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3748, pruned_loss=0.1252, over 5677228.47 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3393, pruned_loss=0.09813, over 5718447.63 frames. ], batch size: 155, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:39:52,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2146, 1.1859, 3.8372, 3.1497], device='cuda:1'), covar=tensor([0.1584, 0.2669, 0.0400, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0665, 0.0596, 0.0863, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 12:39:58,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=541613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:40:11,354 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 40450, giga_loss[loss=0.2701, simple_loss=0.3275, pruned_loss=0.1064, over 28569.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3384, pruned_loss=0.09875, over 5710276.36 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3749, pruned_loss=0.1251, over 5681046.81 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3348, pruned_loss=0.09585, over 5714098.01 frames. ], batch size: 85, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:40:39,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 12:40:55,340 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 40500, giga_loss[loss=0.2568, simple_loss=0.3285, pruned_loss=0.09253, over 28615.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3376, pruned_loss=0.0983, over 5717472.54 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1246, over 5691384.85 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3336, pruned_loss=0.09515, over 5712551.51 frames. ], batch size: 71, lr: 2.65e-03, grad_scale: 4.0
+2023-03-06 12:41:18,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 12:41:27,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5631, 1.6746, 1.2586, 1.3948], device='cuda:1'), covar=tensor([0.0812, 0.0620, 0.1068, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0439, 0.0503, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 12:41:29,360 INFO [optim.py:369] (1/2) 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 12:41:40,009 INFO [train.py:968] (1/2) Epoch 12, batch 40550, giga_loss[loss=0.2694, simple_loss=0.3479, pruned_loss=0.09549, over 28853.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3395, pruned_loss=0.09867, over 5717336.34 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1245, over 5695384.14 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3357, pruned_loss=0.0956, over 5710226.03 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:42:15,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2450, 1.2685, 1.0903, 1.0585], device='cuda:1'), covar=tensor([0.0748, 0.0486, 0.1026, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0437, 0.0501, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 12:42:21,735 INFO [train.py:968] (1/2) Epoch 12, batch 40600, giga_loss[loss=0.28, simple_loss=0.3438, pruned_loss=0.1081, over 28458.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3445, pruned_loss=0.1012, over 5711546.54 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3742, pruned_loss=0.1247, over 5689964.65 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3407, pruned_loss=0.09818, over 5710897.53 frames. ], batch size: 71, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:42:51,025 INFO [optim.py:369] (1/2) 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 12:42:52,213 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=541830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:42:56,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=541833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:42:59,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.86 vs. limit=2.0
+2023-03-06 12:43:01,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-06 12:43:01,996 INFO [train.py:968] (1/2) Epoch 12, batch 40650, giga_loss[loss=0.2498, simple_loss=0.3275, pruned_loss=0.08612, over 28837.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3484, pruned_loss=0.103, over 5710145.81 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3745, pruned_loss=0.125, over 5689319.92 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.344, pruned_loss=0.09944, over 5710609.59 frames. ], batch size: 99, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:43:17,761 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=541862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:43:44,895 INFO [train.py:968] (1/2) Epoch 12, batch 40700, giga_loss[loss=0.3015, simple_loss=0.3725, pruned_loss=0.1153, over 28574.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3502, pruned_loss=0.1033, over 5718860.95 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3743, pruned_loss=0.1249, over 5691029.50 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3465, pruned_loss=0.1003, over 5718034.36 frames. ], batch size: 307, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:44:14,202 INFO [optim.py:369] (1/2) 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 12:44:23,279 INFO [train.py:968] (1/2) Epoch 12, batch 40750, giga_loss[loss=0.27, simple_loss=0.3516, pruned_loss=0.09421, over 28917.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3533, pruned_loss=0.1052, over 5718659.79 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3744, pruned_loss=0.1248, over 5697378.08 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3493, pruned_loss=0.1019, over 5713124.21 frames. ], batch size: 164, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:45:08,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=541988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:45:13,291 INFO [train.py:968] (1/2) Epoch 12, batch 40800, giga_loss[loss=0.3245, simple_loss=0.3827, pruned_loss=0.1332, over 28904.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3559, pruned_loss=0.1075, over 5704162.32 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3745, pruned_loss=0.1249, over 5689364.81 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3525, pruned_loss=0.1048, over 5706119.18 frames. ], batch size: 213, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:45:53,137 INFO [optim.py:369] (1/2) 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 12:46:04,700 INFO [train.py:968] (1/2) Epoch 12, batch 40850, giga_loss[loss=0.3102, simple_loss=0.3808, pruned_loss=0.1198, over 28692.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3641, pruned_loss=0.1149, over 5673002.25 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1247, over 5683815.99 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3612, pruned_loss=0.1126, over 5679788.63 frames. ], batch size: 242, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:46:50,335 INFO [train.py:968] (1/2) Epoch 12, batch 40900, giga_loss[loss=0.3345, simple_loss=0.3938, pruned_loss=0.1376, over 28927.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3705, pruned_loss=0.1198, over 5681308.23 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3741, pruned_loss=0.1247, over 5689623.51 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3682, pruned_loss=0.1177, over 5681553.12 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:47:26,213 INFO [optim.py:369] (1/2) 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 12:47:27,511 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=542131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:47:29,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=542134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:47:35,413 INFO [train.py:968] (1/2) Epoch 12, batch 40950, giga_loss[loss=0.3143, simple_loss=0.3779, pruned_loss=0.1254, over 28731.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3751, pruned_loss=0.1239, over 5664372.87 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1243, over 5687021.39 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3735, pruned_loss=0.1225, over 5666680.90 frames. ], batch size: 78, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:47:55,658 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=542163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:48:19,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3069, 4.1290, 3.9256, 1.8331], device='cuda:1'), covar=tensor([0.0561, 0.0686, 0.0684, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.1071, 0.0990, 0.0864, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 12:48:21,777 INFO [train.py:968] (1/2) Epoch 12, batch 41000, giga_loss[loss=0.3885, simple_loss=0.4313, pruned_loss=0.1728, over 28867.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3809, pruned_loss=0.1287, over 5663322.41 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.374, pruned_loss=0.1247, over 5677566.84 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3795, pruned_loss=0.1274, over 5673914.89 frames. ], batch size: 285, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:48:55,509 INFO [optim.py:369] (1/2) 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 12:49:07,081 INFO [train.py:968] (1/2) Epoch 12, batch 41050, giga_loss[loss=0.3283, simple_loss=0.3921, pruned_loss=0.1322, over 28861.00 frames. ], tot_loss[loss=0.3258, simple_loss=0.3858, pruned_loss=0.1329, over 5650509.86 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3737, pruned_loss=0.1244, over 5672862.17 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3852, pruned_loss=0.1322, over 5662375.10 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:49:59,798 INFO [train.py:968] (1/2) Epoch 12, batch 41100, giga_loss[loss=0.2968, simple_loss=0.3697, pruned_loss=0.112, over 28891.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3871, pruned_loss=0.1345, over 5656533.89 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3738, pruned_loss=0.1246, over 5677719.66 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3869, pruned_loss=0.134, over 5661051.69 frames. ], batch size: 112, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:50:42,609 INFO [optim.py:369] (1/2) 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 12:50:52,747 INFO [train.py:968] (1/2) Epoch 12, batch 41150, giga_loss[loss=0.3598, simple_loss=0.4035, pruned_loss=0.158, over 28547.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3901, pruned_loss=0.1387, over 5637502.08 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3729, pruned_loss=0.124, over 5685591.31 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3912, pruned_loss=0.1392, over 5633002.75 frames. ], batch size: 85, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:51:45,313 INFO [train.py:968] (1/2) Epoch 12, batch 41200, giga_loss[loss=0.39, simple_loss=0.424, pruned_loss=0.178, over 28290.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.393, pruned_loss=0.1419, over 5631254.98 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3727, pruned_loss=0.1239, over 5691035.87 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3947, pruned_loss=0.143, over 5620905.07 frames. ], batch size: 368, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:52:05,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6821, 1.6573, 1.2546, 1.3381], device='cuda:1'), covar=tensor([0.0658, 0.0497, 0.0888, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0440, 0.0501, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 12:52:09,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5943, 1.6792, 1.4115, 1.6078], device='cuda:1'), covar=tensor([0.2198, 0.2231, 0.2433, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.1320, 0.0974, 0.1160, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 12:52:24,583 INFO [optim.py:369] (1/2) 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 12:52:39,578 INFO [train.py:968] (1/2) Epoch 12, batch 41250, giga_loss[loss=0.3736, simple_loss=0.431, pruned_loss=0.1581, over 28955.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3979, pruned_loss=0.1458, over 5637808.12 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3726, pruned_loss=0.1239, over 5693478.83 frames. ], giga_tot_loss[loss=0.347, simple_loss=0.3998, pruned_loss=0.147, over 5626305.98 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:53:13,799 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-06 12:53:35,882 INFO [train.py:968] (1/2) Epoch 12, batch 41300, giga_loss[loss=0.3662, simple_loss=0.417, pruned_loss=0.1577, over 28282.00 frames. ], tot_loss[loss=0.3478, simple_loss=0.3997, pruned_loss=0.148, over 5638151.29 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5692968.25 frames. ], giga_tot_loss[loss=0.3499, simple_loss=0.4015, pruned_loss=0.1492, over 5629196.03 frames. ], batch size: 369, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:53:45,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5674, 1.5482, 1.2378, 1.1615], device='cuda:1'), covar=tensor([0.0698, 0.0541, 0.0900, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0440, 0.0501, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 12:53:55,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.48 vs. limit=5.0
+2023-03-06 12:54:16,813 INFO [optim.py:369] (1/2) 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 12:54:27,206 INFO [train.py:968] (1/2) Epoch 12, batch 41350, giga_loss[loss=0.3619, simple_loss=0.4136, pruned_loss=0.1551, over 28878.00 frames. ], tot_loss[loss=0.3452, simple_loss=0.397, pruned_loss=0.1467, over 5638185.51 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5695263.02 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.3986, pruned_loss=0.1478, over 5628624.41 frames. ], batch size: 227, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:55:16,802 INFO [train.py:968] (1/2) Epoch 12, batch 41400, giga_loss[loss=0.3487, simple_loss=0.4097, pruned_loss=0.1439, over 28991.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3977, pruned_loss=0.1471, over 5632430.79 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3732, pruned_loss=0.1242, over 5696633.86 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.399, pruned_loss=0.1484, over 5622458.49 frames. ], batch size: 213, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:55:53,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6115, 2.6301, 1.6946, 0.8715], device='cuda:1'), covar=tensor([0.6425, 0.2751, 0.3237, 0.5661], device='cuda:1'), in_proj_covar=tensor([0.1609, 0.1527, 0.1521, 0.1326], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 12:55:59,535 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 41450, giga_loss[loss=0.3247, simple_loss=0.3849, pruned_loss=0.1322, over 28884.00 frames. ], tot_loss[loss=0.344, simple_loss=0.3969, pruned_loss=0.1456, over 5626044.10 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3735, pruned_loss=0.1244, over 5699272.14 frames. ], giga_tot_loss[loss=0.3457, simple_loss=0.398, pruned_loss=0.1467, over 5614597.24 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:57:00,852 INFO [train.py:968] (1/2) Epoch 12, batch 41500, giga_loss[loss=0.3538, simple_loss=0.3845, pruned_loss=0.1616, over 23302.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3985, pruned_loss=0.1467, over 5604288.84 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3739, pruned_loss=0.1247, over 5693989.87 frames. ], giga_tot_loss[loss=0.3479, simple_loss=0.3997, pruned_loss=0.148, over 5597342.60 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:57:41,509 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 41550, giga_loss[loss=0.2716, simple_loss=0.3541, pruned_loss=0.09454, over 28863.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3966, pruned_loss=0.145, over 5587102.91 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3743, pruned_loss=0.1251, over 5671963.21 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.3981, pruned_loss=0.1465, over 5597934.14 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 12:58:05,755 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=542757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 12:58:40,909 INFO [train.py:968] (1/2) Epoch 12, batch 41600, libri_loss[loss=0.3123, simple_loss=0.3783, pruned_loss=0.1231, over 29533.00 frames. ], tot_loss[loss=0.3382, simple_loss=0.3938, pruned_loss=0.1412, over 5607013.58 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.374, pruned_loss=0.1249, over 5676897.55 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3956, pruned_loss=0.1429, over 5609829.96 frames. ], batch size: 84, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:59:21,452 INFO [optim.py:369] (1/2) 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 12:59:29,086 INFO [train.py:968] (1/2) Epoch 12, batch 41650, giga_loss[loss=0.2731, simple_loss=0.3523, pruned_loss=0.09694, over 28529.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3898, pruned_loss=0.1367, over 5623292.32 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1248, over 5677840.82 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3917, pruned_loss=0.1384, over 5623549.31 frames. ], batch size: 71, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 12:59:53,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2055, 1.2157, 3.7096, 3.1334], device='cuda:1'), covar=tensor([0.1576, 0.2697, 0.0416, 0.1826], device='cuda:1'), in_proj_covar=tensor([0.0668, 0.0595, 0.0866, 0.0773], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 12:59:54,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0524, 3.3845, 2.2101, 1.0659], device='cuda:1'), covar=tensor([0.4786, 0.1621, 0.2357, 0.4106], device='cuda:1'), in_proj_covar=tensor([0.1594, 0.1516, 0.1509, 0.1314], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 12:59:57,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6950, 2.1843, 2.0210, 1.5266], device='cuda:1'), covar=tensor([0.1734, 0.2158, 0.1428, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.0832, 0.0688, 0.0873, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:00:20,777 INFO [train.py:968] (1/2) Epoch 12, batch 41700, giga_loss[loss=0.3085, simple_loss=0.3701, pruned_loss=0.1234, over 28818.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.387, pruned_loss=0.1348, over 5627767.12 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3737, pruned_loss=0.1246, over 5682437.67 frames. ], giga_tot_loss[loss=0.3311, simple_loss=0.3891, pruned_loss=0.1366, over 5622484.37 frames. ], batch size: 99, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:00:57,234 INFO [optim.py:369] (1/2) 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 13:01:04,600 INFO [train.py:968] (1/2) Epoch 12, batch 41750, giga_loss[loss=0.3338, simple_loss=0.391, pruned_loss=0.1384, over 27940.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3834, pruned_loss=0.1321, over 5637290.14 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3725, pruned_loss=0.124, over 5691817.14 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3868, pruned_loss=0.1345, over 5621957.88 frames. ], batch size: 412, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:01:58,743 INFO [train.py:968] (1/2) Epoch 12, batch 41800, libri_loss[loss=0.3204, simple_loss=0.3816, pruned_loss=0.1296, over 29546.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3828, pruned_loss=0.1318, over 5643757.39 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3725, pruned_loss=0.124, over 5694677.54 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3856, pruned_loss=0.1338, over 5628304.18 frames. ], batch size: 83, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:02:35,946 INFO [optim.py:369] (1/2) 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 13:02:46,986 INFO [train.py:968] (1/2) Epoch 12, batch 41850, giga_loss[loss=0.3223, simple_loss=0.393, pruned_loss=0.1258, over 28595.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3819, pruned_loss=0.1308, over 5652154.40 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3725, pruned_loss=0.1239, over 5694876.38 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3842, pruned_loss=0.1325, over 5639157.72 frames. ], batch size: 307, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:03:13,081 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 41900, giga_loss[loss=0.2656, simple_loss=0.3486, pruned_loss=0.09137, over 28799.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3802, pruned_loss=0.1295, over 5639848.64 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3722, pruned_loss=0.1238, over 5691219.83 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3826, pruned_loss=0.1312, over 5630925.65 frames. ], batch size: 119, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:03:38,252 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:04:19,254 INFO [optim.py:369] (1/2) 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 13:04:21,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:04:23,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6797, 2.6168, 2.0702, 2.4810], device='cuda:1'), covar=tensor([0.0629, 0.0492, 0.0777, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0441, 0.0504, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 13:04:29,472 INFO [train.py:968] (1/2) Epoch 12, batch 41950, giga_loss[loss=0.2694, simple_loss=0.354, pruned_loss=0.09243, over 28608.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3789, pruned_loss=0.1261, over 5651061.58 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3716, pruned_loss=0.1234, over 5698344.57 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3817, pruned_loss=0.128, over 5635849.25 frames. ], batch size: 60, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:05:20,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6282, 3.5270, 1.6298, 1.7525], device='cuda:1'), covar=tensor([0.0863, 0.0445, 0.0869, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0517, 0.0347, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 13:05:21,354 INFO [train.py:968] (1/2) Epoch 12, batch 42000, giga_loss[loss=0.3164, simple_loss=0.3862, pruned_loss=0.1232, over 28958.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3812, pruned_loss=0.126, over 5657286.79 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5696068.39 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3832, pruned_loss=0.1273, over 5646792.75 frames. ], batch size: 145, lr: 2.64e-03, grad_scale: 8.0
+2023-03-06 13:05:21,354 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 13:05:29,512 INFO [train.py:1012] (1/2) Epoch 12, validation: loss=0.2102, simple_loss=0.3152, pruned_loss=0.05262, over 944034.00 frames. 
+2023-03-06 13:05:29,513 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 13:06:07,889 INFO [optim.py:369] (1/2) 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 13:06:17,136 INFO [train.py:968] (1/2) Epoch 12, batch 42050, giga_loss[loss=0.4511, simple_loss=0.4591, pruned_loss=0.2216, over 26783.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3835, pruned_loss=0.1282, over 5665288.32 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5699153.25 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3853, pruned_loss=0.1293, over 5653880.30 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 8.0
+2023-03-06 13:06:47,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:06:50,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:07:02,075 INFO [train.py:968] (1/2) Epoch 12, batch 42100, giga_loss[loss=0.2845, simple_loss=0.3557, pruned_loss=0.1067, over 28823.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3825, pruned_loss=0.1281, over 5643319.43 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3722, pruned_loss=0.124, over 5676981.01 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.384, pruned_loss=0.1289, over 5652678.37 frames. ], batch size: 99, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:07:14,866 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:07:37,717 INFO [optim.py:369] (1/2) 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 13:07:46,218 INFO [train.py:968] (1/2) Epoch 12, batch 42150, giga_loss[loss=0.3325, simple_loss=0.381, pruned_loss=0.142, over 28345.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3813, pruned_loss=0.1285, over 5659057.99 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3722, pruned_loss=0.1241, over 5681292.18 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3828, pruned_loss=0.1291, over 5661996.93 frames. ], batch size: 368, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:08:00,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4943, 1.7475, 1.7723, 1.3193], device='cuda:1'), covar=tensor([0.1714, 0.2285, 0.1370, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.0837, 0.0697, 0.0879, 0.0784], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:08:35,189 INFO [train.py:968] (1/2) Epoch 12, batch 42200, libri_loss[loss=0.3483, simple_loss=0.3987, pruned_loss=0.149, over 19770.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3797, pruned_loss=0.1288, over 5649417.70 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3721, pruned_loss=0.124, over 5676046.39 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3811, pruned_loss=0.1294, over 5657007.57 frames. ], batch size: 187, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:08:50,781 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:09:14,767 INFO [optim.py:369] (1/2) 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 13:09:23,663 INFO [train.py:968] (1/2) Epoch 12, batch 42250, giga_loss[loss=0.3259, simple_loss=0.4005, pruned_loss=0.1257, over 28778.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3786, pruned_loss=0.1276, over 5638054.56 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3719, pruned_loss=0.1239, over 5662913.38 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3801, pruned_loss=0.1283, over 5655570.54 frames. ], batch size: 284, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:09:25,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543444.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:09:48,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:10:11,070 INFO [train.py:968] (1/2) Epoch 12, batch 42300, giga_loss[loss=0.345, simple_loss=0.406, pruned_loss=0.142, over 28686.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3785, pruned_loss=0.1258, over 5642914.19 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1238, over 5653406.23 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3798, pruned_loss=0.1264, over 5666017.81 frames. ], batch size: 242, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:10:50,396 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 42350, giga_loss[loss=0.3586, simple_loss=0.4095, pruned_loss=0.1538, over 28629.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3797, pruned_loss=0.1268, over 5641588.46 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3726, pruned_loss=0.1243, over 5649092.35 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3803, pruned_loss=0.127, over 5664186.08 frames. ], batch size: 336, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:11:07,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3118, 1.8478, 1.3711, 0.4637], device='cuda:1'), covar=tensor([0.3281, 0.1832, 0.2691, 0.4557], device='cuda:1'), in_proj_covar=tensor([0.1596, 0.1519, 0.1512, 0.1311], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 13:11:13,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-06 13:11:43,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543587.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:11:46,280 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543590.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:11:46,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 13:11:47,304 INFO [train.py:968] (1/2) Epoch 12, batch 42400, giga_loss[loss=0.2994, simple_loss=0.3709, pruned_loss=0.114, over 28593.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.38, pruned_loss=0.1272, over 5641651.94 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.373, pruned_loss=0.1247, over 5643652.14 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3802, pruned_loss=0.127, over 5664659.45 frames. ], batch size: 307, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:12:03,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:12:06,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:12:08,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5249, 1.5736, 1.2925, 1.2366], device='cuda:1'), covar=tensor([0.0731, 0.0492, 0.0936, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0442, 0.0505, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 13:12:09,307 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3862, 1.7096, 1.4209, 1.5423], device='cuda:1'), covar=tensor([0.0761, 0.0292, 0.0307, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0083, 0.0059, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 13:12:10,606 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543619.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:12:25,370 INFO [optim.py:369] (1/2) 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 13:12:34,691 INFO [train.py:968] (1/2) Epoch 12, batch 42450, libri_loss[loss=0.3222, simple_loss=0.383, pruned_loss=0.1307, over 29540.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3771, pruned_loss=0.1253, over 5660544.91 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1245, over 5650595.34 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3775, pruned_loss=0.1255, over 5672918.48 frames. ], batch size: 76, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:12:34,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:13:25,179 INFO [train.py:968] (1/2) Epoch 12, batch 42500, giga_loss[loss=0.3086, simple_loss=0.3704, pruned_loss=0.1234, over 28578.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.125, over 5656612.63 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3734, pruned_loss=0.1248, over 5652120.67 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3758, pruned_loss=0.1248, over 5665149.26 frames. ], batch size: 336, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:14:07,634 INFO [optim.py:369] (1/2) 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 13:14:17,499 INFO [train.py:968] (1/2) Epoch 12, batch 42550, giga_loss[loss=0.3526, simple_loss=0.3843, pruned_loss=0.1605, over 23555.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1249, over 5667449.96 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3733, pruned_loss=0.1246, over 5654810.08 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.375, pruned_loss=0.1249, over 5671800.09 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:14:34,517 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:14:59,582 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=543784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:15:05,967 INFO [train.py:968] (1/2) Epoch 12, batch 42600, giga_loss[loss=0.4597, simple_loss=0.4689, pruned_loss=0.2253, over 26746.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3742, pruned_loss=0.1252, over 5672383.61 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.373, pruned_loss=0.1243, over 5659176.47 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3746, pruned_loss=0.1255, over 5672216.42 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:15:42,928 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 13:15:47,165 INFO [optim.py:369] (1/2) 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 13:15:54,238 INFO [train.py:968] (1/2) Epoch 12, batch 42650, giga_loss[loss=0.347, simple_loss=0.3972, pruned_loss=0.1485, over 28577.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3738, pruned_loss=0.1259, over 5654353.53 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3726, pruned_loss=0.1241, over 5656289.36 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3745, pruned_loss=0.1264, over 5657026.18 frames. ], batch size: 307, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:16:07,910 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=543856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:16:17,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5526, 1.6189, 1.8262, 1.4137], device='cuda:1'), covar=tensor([0.1437, 0.1985, 0.1129, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0837, 0.0697, 0.0878, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:16:45,598 INFO [train.py:968] (1/2) Epoch 12, batch 42700, giga_loss[loss=0.2893, simple_loss=0.3576, pruned_loss=0.1105, over 28886.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3732, pruned_loss=0.1256, over 5646227.80 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3727, pruned_loss=0.1241, over 5647758.91 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3737, pruned_loss=0.126, over 5655055.93 frames. ], batch size: 227, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:17:15,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4060, 2.6605, 1.4662, 1.4577], device='cuda:1'), covar=tensor([0.0821, 0.0360, 0.0798, 0.1136], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0518, 0.0347, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 13:17:19,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=543927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:17:21,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=543930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:17:26,060 INFO [optim.py:369] (1/2) 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 13:17:31,395 INFO [train.py:968] (1/2) Epoch 12, batch 42750, libri_loss[loss=0.2874, simple_loss=0.3649, pruned_loss=0.1049, over 29244.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5651028.17 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3733, pruned_loss=0.1245, over 5643512.70 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3741, pruned_loss=0.1253, over 5662725.10 frames. ], batch size: 94, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:17:47,324 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=543959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:17:55,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2989, 3.4151, 1.4591, 1.4403], device='cuda:1'), covar=tensor([0.0988, 0.0401, 0.0914, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0517, 0.0347, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 13:18:16,258 INFO [train.py:968] (1/2) Epoch 12, batch 42800, giga_loss[loss=0.298, simple_loss=0.3698, pruned_loss=0.1131, over 28948.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5662150.72 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3729, pruned_loss=0.1242, over 5649584.43 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3736, pruned_loss=0.1239, over 5666443.19 frames. ], batch size: 186, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:18:57,454 INFO [optim.py:369] (1/2) 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 13:19:05,388 INFO [train.py:968] (1/2) Epoch 12, batch 42850, giga_loss[loss=0.3056, simple_loss=0.3736, pruned_loss=0.1189, over 29124.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1238, over 5657496.28 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3733, pruned_loss=0.1244, over 5641649.41 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3742, pruned_loss=0.1238, over 5668034.40 frames. ], batch size: 155, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:19:54,762 INFO [train.py:968] (1/2) Epoch 12, batch 42900, giga_loss[loss=0.3373, simple_loss=0.3972, pruned_loss=0.1387, over 28792.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3773, pruned_loss=0.1263, over 5666218.27 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1249, over 5639104.18 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3768, pruned_loss=0.1259, over 5676960.41 frames. ], batch size: 284, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:20:14,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-06 13:20:38,965 INFO [optim.py:369] (1/2) 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 13:20:39,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=544135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:20:47,331 INFO [train.py:968] (1/2) Epoch 12, batch 42950, giga_loss[loss=0.3216, simple_loss=0.3769, pruned_loss=0.1332, over 28806.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3797, pruned_loss=0.1291, over 5661308.65 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.1249, over 5634253.53 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3793, pruned_loss=0.1288, over 5674901.74 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:21:21,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3387, 1.6056, 1.4019, 1.5262], device='cuda:1'), covar=tensor([0.0782, 0.0314, 0.0312, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 13:21:42,255 INFO [train.py:968] (1/2) Epoch 12, batch 43000, giga_loss[loss=0.3035, simple_loss=0.3719, pruned_loss=0.1176, over 29050.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3803, pruned_loss=0.1308, over 5654624.81 frames. ], libri_tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.1251, over 5628835.52 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3798, pruned_loss=0.1305, over 5670310.73 frames. ], batch size: 164, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:22:14,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8347, 2.6925, 2.6778, 2.3227], device='cuda:1'), covar=tensor([0.1216, 0.1753, 0.1433, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0732, 0.0677, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 13:22:21,501 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:22:32,539 INFO [train.py:968] (1/2) Epoch 12, batch 43050, giga_loss[loss=0.4488, simple_loss=0.4694, pruned_loss=0.2141, over 24096.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3817, pruned_loss=0.1329, over 5651143.71 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1248, over 5631593.51 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3817, pruned_loss=0.133, over 5661784.49 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:23:06,184 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=544278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:23:09,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=544281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:23:18,502 INFO [train.py:968] (1/2) Epoch 12, batch 43100, giga_loss[loss=0.287, simple_loss=0.3577, pruned_loss=0.1081, over 28827.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3803, pruned_loss=0.132, over 5652585.22 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1245, over 5635217.23 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3808, pruned_loss=0.1325, over 5658175.46 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:23:34,447 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=544310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:23:57,947 INFO [optim.py:369] (1/2) 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 13:24:04,185 INFO [train.py:968] (1/2) Epoch 12, batch 43150, giga_loss[loss=0.2596, simple_loss=0.3364, pruned_loss=0.09141, over 28904.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3783, pruned_loss=0.13, over 5660340.86 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1241, over 5642854.45 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3792, pruned_loss=0.131, over 5658536.61 frames. ], batch size: 199, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:24:31,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=544374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:24:33,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=544377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:24:45,249 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 43200, giga_loss[loss=0.2671, simple_loss=0.3482, pruned_loss=0.09301, over 28835.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1284, over 5656012.39 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3735, pruned_loss=0.1247, over 5638327.64 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3786, pruned_loss=0.1289, over 5659087.37 frames. ], batch size: 119, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:24:49,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-06 13:24:59,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=544406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:25:06,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 13:25:25,118 INFO [optim.py:369] (1/2) 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 13:25:31,749 INFO [train.py:968] (1/2) Epoch 12, batch 43250, giga_loss[loss=0.3233, simple_loss=0.3817, pruned_loss=0.1324, over 28616.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3759, pruned_loss=0.1265, over 5655332.64 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3739, pruned_loss=0.1251, over 5640393.85 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3762, pruned_loss=0.1266, over 5656097.70 frames. ], batch size: 307, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:25:44,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6582, 1.8226, 1.7660, 1.6509], device='cuda:1'), covar=tensor([0.1438, 0.1744, 0.2005, 0.1812], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0732, 0.0678, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 13:26:16,715 INFO [train.py:968] (1/2) Epoch 12, batch 43300, giga_loss[loss=0.3168, simple_loss=0.362, pruned_loss=0.1358, over 26534.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3742, pruned_loss=0.1257, over 5656986.93 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3741, pruned_loss=0.1253, over 5633385.89 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1256, over 5663599.62 frames. ], batch size: 555, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:26:31,635 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:26:58,217 INFO [optim.py:369] (1/2) 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 13:27:04,094 INFO [train.py:968] (1/2) Epoch 12, batch 43350, giga_loss[loss=0.2607, simple_loss=0.3371, pruned_loss=0.09219, over 28447.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3735, pruned_loss=0.1262, over 5649802.16 frames. ], libri_tot_loss[loss=0.3131, simple_loss=0.3747, pruned_loss=0.1258, over 5628171.64 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.373, pruned_loss=0.1257, over 5660554.14 frames. ], batch size: 71, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:27:49,649 INFO [train.py:968] (1/2) Epoch 12, batch 43400, giga_loss[loss=0.3249, simple_loss=0.3786, pruned_loss=0.1356, over 28705.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3744, pruned_loss=0.1268, over 5661864.50 frames. ], libri_tot_loss[loss=0.313, simple_loss=0.3747, pruned_loss=0.1256, over 5633598.28 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.374, pruned_loss=0.1266, over 5666227.01 frames. ], batch size: 99, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:28:30,641 INFO [optim.py:369] (1/2) 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 13:28:35,502 INFO [train.py:968] (1/2) Epoch 12, batch 43450, libri_loss[loss=0.325, simple_loss=0.387, pruned_loss=0.1315, over 25803.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.378, pruned_loss=0.1286, over 5650472.14 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3744, pruned_loss=0.1253, over 5628097.88 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3781, pruned_loss=0.1288, over 5660530.58 frames. ], batch size: 136, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:29:24,777 INFO [train.py:968] (1/2) Epoch 12, batch 43500, giga_loss[loss=0.3558, simple_loss=0.4127, pruned_loss=0.1494, over 27576.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3811, pruned_loss=0.1273, over 5663656.60 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3744, pruned_loss=0.1253, over 5632228.96 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3812, pruned_loss=0.1275, over 5668250.47 frames. ], batch size: 472, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:29:37,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:29:46,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5216, 1.5969, 1.1891, 1.2124], device='cuda:1'), covar=tensor([0.0729, 0.0471, 0.0965, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0444, 0.0503, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 13:30:08,601 INFO [optim.py:369] (1/2) 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 13:30:14,162 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4585, 2.8796, 1.5989, 1.5838], device='cuda:1'), covar=tensor([0.0742, 0.0322, 0.0683, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0517, 0.0348, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 13:30:14,456 INFO [train.py:968] (1/2) Epoch 12, batch 43550, giga_loss[loss=0.3547, simple_loss=0.3867, pruned_loss=0.1613, over 23810.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3809, pruned_loss=0.1269, over 5662952.14 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3741, pruned_loss=0.125, over 5639997.23 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3814, pruned_loss=0.1273, over 5660529.91 frames. ], batch size: 705, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:30:36,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=544765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:30:54,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544783.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:31:00,807 INFO [train.py:968] (1/2) Epoch 12, batch 43600, giga_loss[loss=0.322, simple_loss=0.3874, pruned_loss=0.1283, over 28980.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3835, pruned_loss=0.1291, over 5667398.26 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3736, pruned_loss=0.1246, over 5645126.00 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3845, pruned_loss=0.1298, over 5661396.96 frames. ], batch size: 213, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:31:30,184 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=544823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:31:38,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3131, 1.6427, 1.5467, 1.4552], device='cuda:1'), covar=tensor([0.1816, 0.1713, 0.2155, 0.1851], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0733, 0.0679, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 13:31:41,495 INFO [optim.py:369] (1/2) 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 13:31:45,787 INFO [train.py:968] (1/2) Epoch 12, batch 43650, libri_loss[loss=0.3193, simple_loss=0.3804, pruned_loss=0.1291, over 29530.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3837, pruned_loss=0.1302, over 5668693.32 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1244, over 5654731.59 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3857, pruned_loss=0.1314, over 5655562.95 frames. ], batch size: 84, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:32:20,234 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 43700, giga_loss[loss=0.2961, simple_loss=0.3659, pruned_loss=0.1132, over 29034.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.383, pruned_loss=0.1309, over 5677505.50 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 5658901.89 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3854, pruned_loss=0.1323, over 5663720.64 frames. ], batch size: 136, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:32:42,632 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=544908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:32:45,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=544911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:33:08,727 INFO [optim.py:369] (1/2) 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 13:33:11,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=544940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:33:13,050 INFO [train.py:968] (1/2) Epoch 12, batch 43750, giga_loss[loss=0.3027, simple_loss=0.3589, pruned_loss=0.1233, over 28759.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.381, pruned_loss=0.1302, over 5668838.92 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5656608.42 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3837, pruned_loss=0.1319, over 5659819.48 frames. ], batch size: 99, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:33:50,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1649, 1.1496, 3.6552, 2.9604], device='cuda:1'), covar=tensor([0.1667, 0.2617, 0.0460, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0604, 0.0878, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 13:34:01,625 INFO [train.py:968] (1/2) Epoch 12, batch 43800, giga_loss[loss=0.2708, simple_loss=0.3479, pruned_loss=0.09686, over 28740.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3784, pruned_loss=0.1289, over 5667328.79 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5649644.04 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3807, pruned_loss=0.1303, over 5666599.93 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:34:37,036 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:34:41,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:34:49,641 INFO [optim.py:369] (1/2) 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 13:34:54,028 INFO [train.py:968] (1/2) Epoch 12, batch 43850, giga_loss[loss=0.3089, simple_loss=0.3756, pruned_loss=0.1211, over 29005.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3783, pruned_loss=0.1293, over 5668996.22 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1237, over 5653476.04 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3801, pruned_loss=0.1305, over 5665037.07 frames. ], batch size: 136, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:35:11,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-06 13:35:11,804 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:35:32,652 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 43900, giga_loss[loss=0.3284, simple_loss=0.3868, pruned_loss=0.135, over 28890.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3796, pruned_loss=0.1307, over 5655800.94 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3719, pruned_loss=0.1237, over 5638092.10 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3813, pruned_loss=0.1318, over 5665972.97 frames. ], batch size: 112, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:36:28,578 INFO [optim.py:369] (1/2) 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 13:36:31,102 INFO [train.py:968] (1/2) Epoch 12, batch 43950, giga_loss[loss=0.2731, simple_loss=0.3415, pruned_loss=0.1023, over 28532.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3783, pruned_loss=0.13, over 5663558.95 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3715, pruned_loss=0.1234, over 5644342.32 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3802, pruned_loss=0.1313, over 5666435.82 frames. ], batch size: 85, lr: 2.64e-03, grad_scale: 2.0
+2023-03-06 13:36:35,760 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545146.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:36:47,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=545158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:36:58,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 13:37:17,022 INFO [train.py:968] (1/2) Epoch 12, batch 44000, giga_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.113, over 28874.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3762, pruned_loss=0.1287, over 5662766.84 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.372, pruned_loss=0.1238, over 5638467.86 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1296, over 5669843.24 frames. ], batch size: 112, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:37:24,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=545198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:37:24,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5427, 1.7357, 1.4254, 1.9838], device='cuda:1'), covar=tensor([0.2398, 0.2494, 0.2683, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.1330, 0.0983, 0.1171, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 13:37:47,361 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:37:51,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:38:00,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6587, 1.9105, 1.9351, 1.4376], device='cuda:1'), covar=tensor([0.1726, 0.2187, 0.1313, 0.1556], device='cuda:1'), in_proj_covar=tensor([0.0839, 0.0698, 0.0881, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:38:02,168 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 44050, giga_loss[loss=0.3424, simple_loss=0.4018, pruned_loss=0.1415, over 28684.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3763, pruned_loss=0.1284, over 5663928.78 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3718, pruned_loss=0.1237, over 5639842.52 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3774, pruned_loss=0.1292, over 5668403.65 frames. ], batch size: 284, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:38:19,089 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:38:52,027 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-06 13:38:59,479 INFO [train.py:968] (1/2) Epoch 12, batch 44100, giga_loss[loss=0.3251, simple_loss=0.397, pruned_loss=0.1266, over 28756.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5653850.48 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.372, pruned_loss=0.1237, over 5638330.56 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3791, pruned_loss=0.1298, over 5658806.82 frames. ], batch size: 119, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:39:04,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2077, 1.0951, 4.0700, 3.4250], device='cuda:1'), covar=tensor([0.1788, 0.2851, 0.0422, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0604, 0.0877, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 13:39:07,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545301.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:39:10,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:39:38,972 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:39:41,996 INFO [optim.py:369] (1/2) 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 13:39:45,391 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:39:45,768 INFO [train.py:968] (1/2) Epoch 12, batch 44150, giga_loss[loss=0.3122, simple_loss=0.3655, pruned_loss=0.1294, over 28786.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3782, pruned_loss=0.1287, over 5670144.76 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3721, pruned_loss=0.1236, over 5643424.38 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1295, over 5669893.45 frames. ], batch size: 99, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:39:49,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:40:12,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:40:28,642 INFO [train.py:968] (1/2) Epoch 12, batch 44200, giga_loss[loss=0.2922, simple_loss=0.392, pruned_loss=0.09621, over 28997.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1282, over 5670770.16 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5653817.93 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3794, pruned_loss=0.1297, over 5662396.80 frames. ], batch size: 136, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:41:09,355 INFO [optim.py:369] (1/2) 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 13:41:12,291 INFO [train.py:968] (1/2) Epoch 12, batch 44250, giga_loss[loss=0.3511, simple_loss=0.4122, pruned_loss=0.145, over 28895.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3808, pruned_loss=0.1282, over 5669782.29 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3708, pruned_loss=0.1228, over 5651621.05 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3828, pruned_loss=0.1298, over 5665045.34 frames. ], batch size: 243, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:41:27,479 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 13:41:56,698 INFO [train.py:968] (1/2) Epoch 12, batch 44300, giga_loss[loss=0.3232, simple_loss=0.3961, pruned_loss=0.1251, over 28657.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3821, pruned_loss=0.1271, over 5683382.99 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3707, pruned_loss=0.1229, over 5657400.86 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.384, pruned_loss=0.1283, over 5674937.18 frames. ], batch size: 92, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:42:00,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5723, 1.7538, 1.4760, 1.7489], device='cuda:1'), covar=tensor([0.1838, 0.1707, 0.1756, 0.1631], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.0981, 0.1171, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 13:42:12,503 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-06 13:42:25,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=545521.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:42:40,273 INFO [optim.py:369] (1/2) 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 13:42:44,783 INFO [train.py:968] (1/2) Epoch 12, batch 44350, giga_loss[loss=0.2791, simple_loss=0.3595, pruned_loss=0.09931, over 28831.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3845, pruned_loss=0.1284, over 5689988.92 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3707, pruned_loss=0.1229, over 5660647.77 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3865, pruned_loss=0.1295, over 5680999.28 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:43:15,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1739, 1.3628, 1.0731, 0.9856], device='cuda:1'), covar=tensor([0.0824, 0.0460, 0.1064, 0.0940], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0441, 0.0502, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 13:43:36,190 INFO [train.py:968] (1/2) Epoch 12, batch 44400, giga_loss[loss=0.3308, simple_loss=0.3957, pruned_loss=0.1329, over 28917.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.388, pruned_loss=0.1323, over 5675585.81 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3708, pruned_loss=0.123, over 5658308.53 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3896, pruned_loss=0.1331, over 5670768.57 frames. ], batch size: 145, lr: 2.64e-03, grad_scale: 8.0
+2023-03-06 13:44:14,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:44:20,413 INFO [optim.py:369] (1/2) 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 13:44:23,987 INFO [train.py:968] (1/2) Epoch 12, batch 44450, giga_loss[loss=0.3076, simple_loss=0.3792, pruned_loss=0.1181, over 28899.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3885, pruned_loss=0.1342, over 5662644.28 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3706, pruned_loss=0.1231, over 5663535.36 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3905, pruned_loss=0.1352, over 5654216.77 frames. ], batch size: 174, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:44:42,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:44:47,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=545664.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 13:44:50,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=545667.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:45:00,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7359, 1.9116, 2.0155, 1.5253], device='cuda:1'), covar=tensor([0.1711, 0.2172, 0.1334, 0.1600], device='cuda:1'), in_proj_covar=tensor([0.0836, 0.0697, 0.0880, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:45:12,260 INFO [train.py:968] (1/2) Epoch 12, batch 44500, giga_loss[loss=0.3114, simple_loss=0.3798, pruned_loss=0.1215, over 29030.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3873, pruned_loss=0.1332, over 5667746.53 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.1231, over 5664694.68 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3889, pruned_loss=0.1341, over 5660173.51 frames. ], batch size: 136, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:45:17,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=545696.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 13:45:46,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5620, 1.5875, 1.8402, 1.3849], device='cuda:1'), covar=tensor([0.1730, 0.2420, 0.1396, 0.1646], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0700, 0.0883, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:45:57,651 INFO [optim.py:369] (1/2) 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 13:45:59,839 INFO [train.py:968] (1/2) Epoch 12, batch 44550, giga_loss[loss=0.2847, simple_loss=0.364, pruned_loss=0.1027, over 28401.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3852, pruned_loss=0.1308, over 5655948.19 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.371, pruned_loss=0.1233, over 5658188.02 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3864, pruned_loss=0.1314, over 5656349.37 frames. ], batch size: 65, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:46:47,310 INFO [train.py:968] (1/2) Epoch 12, batch 44600, giga_loss[loss=0.4091, simple_loss=0.4517, pruned_loss=0.1832, over 28680.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3875, pruned_loss=0.1305, over 5661990.79 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1236, over 5659382.08 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3883, pruned_loss=0.1307, over 5661186.28 frames. ], batch size: 336, lr: 2.64e-03, grad_scale: 4.0
+2023-03-06 13:47:25,754 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:47:35,211 INFO [optim.py:369] (1/2) 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 13:47:37,282 INFO [train.py:968] (1/2) Epoch 12, batch 44650, giga_loss[loss=0.3201, simple_loss=0.3936, pruned_loss=0.1233, over 28860.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3886, pruned_loss=0.1311, over 5668550.94 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3714, pruned_loss=0.1236, over 5662916.75 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3893, pruned_loss=0.1313, over 5664701.47 frames. ], batch size: 174, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:48:20,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5862, 1.6123, 1.8353, 1.3861], device='cuda:1'), covar=tensor([0.1549, 0.2184, 0.1219, 0.1492], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0697, 0.0881, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:48:29,583 INFO [train.py:968] (1/2) Epoch 12, batch 44700, giga_loss[loss=0.2936, simple_loss=0.3682, pruned_loss=0.1095, over 28725.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3883, pruned_loss=0.1314, over 5664901.55 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3718, pruned_loss=0.1239, over 5657432.02 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3887, pruned_loss=0.1315, over 5666558.80 frames. ], batch size: 262, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:48:44,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8601, 1.8188, 1.2528, 1.4374], device='cuda:1'), covar=tensor([0.0731, 0.0608, 0.1026, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0442, 0.0503, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 13:49:10,102 INFO [optim.py:369] (1/2) 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 13:49:13,987 INFO [train.py:968] (1/2) Epoch 12, batch 44750, giga_loss[loss=0.2972, simple_loss=0.367, pruned_loss=0.1137, over 28242.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3874, pruned_loss=0.1316, over 5665100.06 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3724, pruned_loss=0.1244, over 5663966.35 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3876, pruned_loss=0.1313, over 5661062.92 frames. ], batch size: 65, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:49:15,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9443, 3.7085, 3.5486, 1.6778], device='cuda:1'), covar=tensor([0.0843, 0.1096, 0.1093, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1087, 0.1020, 0.0883, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 13:49:16,590 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-06 13:49:38,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-06 13:49:40,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=545972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:50:02,031 INFO [train.py:968] (1/2) Epoch 12, batch 44800, giga_loss[loss=0.3345, simple_loss=0.3656, pruned_loss=0.1517, over 23639.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3857, pruned_loss=0.1319, over 5653876.34 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3725, pruned_loss=0.1243, over 5669473.32 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3862, pruned_loss=0.1321, over 5645724.54 frames. ], batch size: 705, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 13:50:14,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:50:39,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:50:43,796 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 12, batch 44850, giga_loss[loss=0.285, simple_loss=0.357, pruned_loss=0.1065, over 28854.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3828, pruned_loss=0.1304, over 5667179.00 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3726, pruned_loss=0.1243, over 5677320.99 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3835, pruned_loss=0.1307, over 5653449.13 frames. ], batch size: 199, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 13:50:56,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:51:05,420 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-06 13:51:05,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:51:35,587 INFO [train.py:968] (1/2) Epoch 12, batch 44900, giga_loss[loss=0.3104, simple_loss=0.3759, pruned_loss=0.1224, over 28845.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3798, pruned_loss=0.1287, over 5672276.00 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3725, pruned_loss=0.1242, over 5682253.36 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3806, pruned_loss=0.1292, over 5656941.47 frames. ], batch size: 186, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:51:56,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2105, 1.2081, 3.9131, 3.3168], device='cuda:1'), covar=tensor([0.1571, 0.2569, 0.0416, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0677, 0.0602, 0.0877, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 13:52:04,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:18,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:23,012 INFO [optim.py:369] (1/2) 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 13:52:25,803 INFO [train.py:968] (1/2) Epoch 12, batch 44950, giga_loss[loss=0.3139, simple_loss=0.3805, pruned_loss=0.1236, over 28937.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3781, pruned_loss=0.128, over 5664025.92 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3725, pruned_loss=0.1241, over 5674140.74 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3788, pruned_loss=0.1284, over 5659822.24 frames. ], batch size: 227, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:52:33,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:33,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5012, 1.6635, 1.8048, 1.3483], device='cuda:1'), covar=tensor([0.1681, 0.2297, 0.1322, 0.1568], device='cuda:1'), in_proj_covar=tensor([0.0839, 0.0696, 0.0881, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:52:35,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:52:55,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3385, 1.7457, 1.2905, 0.6476], device='cuda:1'), covar=tensor([0.3493, 0.1818, 0.2148, 0.4494], device='cuda:1'), in_proj_covar=tensor([0.1590, 0.1518, 0.1501, 0.1310], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 13:52:55,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2897, 1.5759, 1.5731, 1.1613], device='cuda:1'), covar=tensor([0.1618, 0.2342, 0.1318, 0.1539], device='cuda:1'), in_proj_covar=tensor([0.0838, 0.0695, 0.0881, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 13:53:02,322 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:05,130 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:05,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:13,656 INFO [train.py:968] (1/2) Epoch 12, batch 45000, giga_loss[loss=0.3047, simple_loss=0.3713, pruned_loss=0.1191, over 27947.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3763, pruned_loss=0.1263, over 5666644.48 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1242, over 5673705.04 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3768, pruned_loss=0.1266, over 5663730.24 frames. ], batch size: 412, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:53:13,656 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 13:53:22,189 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 13:53:38,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:41,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:53:53,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:54:05,699 INFO [optim.py:369] (1/2) 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 13:54:06,951 INFO [train.py:968] (1/2) Epoch 12, batch 45050, giga_loss[loss=0.2627, simple_loss=0.343, pruned_loss=0.09121, over 28931.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3733, pruned_loss=0.123, over 5646007.45 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3731, pruned_loss=0.1244, over 5656192.36 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3732, pruned_loss=0.123, over 5658762.05 frames. ], batch size: 136, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:54:42,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9688, 3.7595, 3.5999, 1.7053], device='cuda:1'), covar=tensor([0.0719, 0.0943, 0.0955, 0.2058], device='cuda:1'), in_proj_covar=tensor([0.1083, 0.1014, 0.0879, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 13:54:53,764 INFO [train.py:968] (1/2) Epoch 12, batch 45100, giga_loss[loss=0.2654, simple_loss=0.3516, pruned_loss=0.08962, over 28973.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3726, pruned_loss=0.1219, over 5640638.65 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3733, pruned_loss=0.1246, over 5643384.53 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3723, pruned_loss=0.1217, over 5662074.46 frames. ], batch size: 164, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:55:38,089 INFO [optim.py:369] (1/2) 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 13:55:40,532 INFO [train.py:968] (1/2) Epoch 12, batch 45150, giga_loss[loss=0.2944, simple_loss=0.3656, pruned_loss=0.1116, over 28877.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3707, pruned_loss=0.1212, over 5634238.66 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3731, pruned_loss=0.1245, over 5647843.20 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3707, pruned_loss=0.1211, over 5647343.33 frames. ], batch size: 174, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:55:46,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:55:49,304 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:55:52,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:56:23,036 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:56:31,059 INFO [train.py:968] (1/2) Epoch 12, batch 45200, giga_loss[loss=0.2813, simple_loss=0.3491, pruned_loss=0.1068, over 28556.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3682, pruned_loss=0.1207, over 5629093.57 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1244, over 5653865.55 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1206, over 5633865.06 frames. ], batch size: 336, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 13:57:03,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:57:11,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:57:14,647 INFO [optim.py:369] (1/2) 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 13:57:16,476 INFO [train.py:968] (1/2) Epoch 12, batch 45250, giga_loss[loss=0.3506, simple_loss=0.4065, pruned_loss=0.1474, over 27976.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.369, pruned_loss=0.1213, over 5635135.15 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3729, pruned_loss=0.1242, over 5648925.84 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.369, pruned_loss=0.1213, over 5642789.64 frames. ], batch size: 412, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:57:29,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3466, 1.5053, 1.2463, 1.5284], device='cuda:1'), covar=tensor([0.0716, 0.0366, 0.0330, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 13:58:02,145 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:03,111 INFO [train.py:968] (1/2) Epoch 12, batch 45300, giga_loss[loss=0.3371, simple_loss=0.398, pruned_loss=0.1381, over 28726.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3716, pruned_loss=0.1222, over 5634810.68 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1243, over 5644495.63 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5644333.35 frames. ], batch size: 242, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:58:05,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:06,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:09,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:12,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2193, 1.5171, 1.2398, 1.0333], device='cuda:1'), covar=tensor([0.2056, 0.1891, 0.2034, 0.1778], device='cuda:1'), in_proj_covar=tensor([0.1339, 0.0986, 0.1181, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 13:58:19,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:31,523 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:58:37,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4304, 1.4983, 1.5048, 1.2708], device='cuda:1'), covar=tensor([0.2133, 0.2032, 0.1757, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.1747, 0.1647, 0.1610, 0.1703], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 13:58:51,596 INFO [optim.py:369] (1/2) 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 13:58:52,184 INFO [train.py:968] (1/2) Epoch 12, batch 45350, giga_loss[loss=0.3352, simple_loss=0.3925, pruned_loss=0.1389, over 28860.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3739, pruned_loss=0.1242, over 5627666.05 frames. ], libri_tot_loss[loss=0.3115, simple_loss=0.3737, pruned_loss=0.1246, over 5647592.05 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3731, pruned_loss=0.1238, over 5632367.50 frames. ], batch size: 186, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:59:17,982 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:20,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:26,811 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:28,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:32,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1996, 1.1918, 4.0332, 3.0828], device='cuda:1'), covar=tensor([0.1704, 0.2655, 0.0416, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0600, 0.0871, 0.0784], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 13:59:37,745 INFO [train.py:968] (1/2) Epoch 12, batch 45400, giga_loss[loss=0.3037, simple_loss=0.3725, pruned_loss=0.1174, over 29031.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.124, over 5617998.57 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1248, over 5644814.88 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3726, pruned_loss=0.1235, over 5623481.66 frames. ], batch size: 128, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 13:59:42,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3748, 1.6439, 1.2717, 1.4445], device='cuda:1'), covar=tensor([0.2471, 0.2425, 0.2665, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.0984, 0.1177, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 13:59:44,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:47,136 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 13:59:54,253 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:22,986 INFO [optim.py:369] (1/2) 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 14:00:26,195 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 45450, giga_loss[loss=0.301, simple_loss=0.373, pruned_loss=0.1145, over 28977.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3738, pruned_loss=0.1242, over 5631887.65 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3744, pruned_loss=0.1249, over 5647933.73 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3728, pruned_loss=0.1237, over 5632937.40 frames. ], batch size: 136, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:00:28,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:32,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=546650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:35,100 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:37,147 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:41,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2806, 1.5219, 1.4535, 1.3978], device='cuda:1'), covar=tensor([0.1122, 0.1117, 0.1618, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0746, 0.0689, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 14:00:53,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:00:59,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1662, 1.1794, 3.6837, 3.1148], device='cuda:1'), covar=tensor([0.1651, 0.2652, 0.0467, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0603, 0.0877, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 14:01:05,376 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:01:14,391 INFO [train.py:968] (1/2) Epoch 12, batch 45500, giga_loss[loss=0.2742, simple_loss=0.3478, pruned_loss=0.1003, over 28829.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3775, pruned_loss=0.1274, over 5634132.79 frames. ], libri_tot_loss[loss=0.3126, simple_loss=0.3749, pruned_loss=0.1251, over 5641925.55 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3763, pruned_loss=0.1268, over 5640116.06 frames. ], batch size: 66, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:02:00,517 INFO [optim.py:369] (1/2) 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 14:02:00,530 INFO [train.py:968] (1/2) Epoch 12, batch 45550, giga_loss[loss=0.2858, simple_loss=0.3517, pruned_loss=0.1099, over 28584.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3791, pruned_loss=0.1283, over 5637485.48 frames. ], libri_tot_loss[loss=0.3124, simple_loss=0.3746, pruned_loss=0.1251, over 5636660.45 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3785, pruned_loss=0.1279, over 5646756.02 frames. ], batch size: 85, lr: 2.63e-03, grad_scale: 2.0
+2023-03-06 14:02:02,081 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=546743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:02:04,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=546746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:02:16,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-06 14:02:32,975 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=546775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:02:41,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-06 14:02:49,433 INFO [train.py:968] (1/2) Epoch 12, batch 45600, giga_loss[loss=0.3013, simple_loss=0.3742, pruned_loss=0.1142, over 29004.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3805, pruned_loss=0.1296, over 5634913.83 frames. ], libri_tot_loss[loss=0.3125, simple_loss=0.3747, pruned_loss=0.1251, over 5630126.77 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3799, pruned_loss=0.1294, over 5647263.79 frames. ], batch size: 227, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:03:08,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1075, 1.2653, 3.4894, 3.0149], device='cuda:1'), covar=tensor([0.1623, 0.2541, 0.0482, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0679, 0.0604, 0.0878, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 14:03:41,564 INFO [optim.py:369] (1/2) 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 14:03:41,577 INFO [train.py:968] (1/2) Epoch 12, batch 45650, giga_loss[loss=0.3217, simple_loss=0.39, pruned_loss=0.1267, over 28741.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3807, pruned_loss=0.1304, over 5640042.88 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1248, over 5631945.41 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3808, pruned_loss=0.1306, over 5648234.37 frames. ], batch size: 284, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:04:13,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=546870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:04:40,433 INFO [train.py:968] (1/2) Epoch 12, batch 45700, giga_loss[loss=0.3292, simple_loss=0.3852, pruned_loss=0.1366, over 28962.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3809, pruned_loss=0.1286, over 5643562.70 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3744, pruned_loss=0.1249, over 5633368.69 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3809, pruned_loss=0.1286, over 5648730.49 frames. ], batch size: 106, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:04:50,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3566, 2.1435, 1.5145, 0.4668], device='cuda:1'), covar=tensor([0.3747, 0.2399, 0.3068, 0.4499], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1543, 0.1522, 0.1328], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 14:05:27,722 INFO [optim.py:369] (1/2) 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 14:05:27,740 INFO [train.py:968] (1/2) Epoch 12, batch 45750, giga_loss[loss=0.3788, simple_loss=0.4032, pruned_loss=0.1772, over 23357.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3798, pruned_loss=0.1278, over 5641293.40 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3742, pruned_loss=0.1249, over 5632786.18 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3803, pruned_loss=0.128, over 5645263.54 frames. ], batch size: 705, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:06:14,630 INFO [train.py:968] (1/2) Epoch 12, batch 45800, giga_loss[loss=0.3026, simple_loss=0.3709, pruned_loss=0.1171, over 28848.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3781, pruned_loss=0.1271, over 5639658.38 frames. ], libri_tot_loss[loss=0.3117, simple_loss=0.3741, pruned_loss=0.1247, over 5626773.06 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3786, pruned_loss=0.1275, over 5648681.06 frames. ], batch size: 199, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:06:41,758 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=547013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:06:44,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=547016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:06:56,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=547025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:07:15,809 INFO [optim.py:369] (1/2) 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 14:07:15,822 INFO [train.py:968] (1/2) Epoch 12, batch 45850, giga_loss[loss=0.2942, simple_loss=0.3621, pruned_loss=0.1131, over 29014.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3778, pruned_loss=0.1278, over 5629540.52 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1248, over 5628281.20 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3781, pruned_loss=0.1281, over 5635154.15 frames. ], batch size: 213, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:07:20,615 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=547045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:07:39,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4892, 4.0413, 1.6599, 1.5471], device='cuda:1'), covar=tensor([0.0985, 0.0314, 0.0889, 0.1374], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0523, 0.0351, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 14:08:06,141 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 14:08:07,708 INFO [train.py:968] (1/2) Epoch 12, batch 45900, giga_loss[loss=0.3436, simple_loss=0.3875, pruned_loss=0.1498, over 28674.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3773, pruned_loss=0.1282, over 5633261.34 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1245, over 5636179.01 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1287, over 5630293.69 frames. ], batch size: 85, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:08:56,697 INFO [optim.py:369] (1/2) 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 14:08:56,711 INFO [train.py:968] (1/2) Epoch 12, batch 45950, giga_loss[loss=0.2971, simple_loss=0.3682, pruned_loss=0.113, over 28848.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3777, pruned_loss=0.1293, over 5634660.63 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3744, pruned_loss=0.1247, over 5641161.13 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3779, pruned_loss=0.1296, over 5627611.18 frames. ], batch size: 145, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:09:13,057 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=547161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:09:17,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2508, 1.4463, 1.4141, 1.1349], device='cuda:1'), covar=tensor([0.1989, 0.1951, 0.1127, 0.1662], device='cuda:1'), in_proj_covar=tensor([0.1769, 0.1660, 0.1622, 0.1716], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 14:09:19,220 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=547168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:09:22,308 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 12, batch 46000, giga_loss[loss=0.3308, simple_loss=0.3814, pruned_loss=0.1401, over 28531.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.376, pruned_loss=0.1278, over 5645270.51 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1243, over 5642888.89 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3768, pruned_loss=0.1285, over 5637834.40 frames. ], batch size: 336, lr: 2.63e-03, grad_scale: 8.0
+2023-03-06 14:09:47,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=547200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:10:23,264 INFO [train.py:968] (1/2) Epoch 12, batch 46050, giga_loss[loss=0.2898, simple_loss=0.3647, pruned_loss=0.1075, over 28910.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3778, pruned_loss=0.1292, over 5648388.45 frames. ], libri_tot_loss[loss=0.312, simple_loss=0.3745, pruned_loss=0.1248, over 5642370.80 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3779, pruned_loss=0.1295, over 5641870.16 frames. ], batch size: 164, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:10:25,527 INFO [optim.py:369] (1/2) 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 14:10:59,086 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0067, 1.2115, 1.2590, 1.0746], device='cuda:1'), covar=tensor([0.1265, 0.1172, 0.1793, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0745, 0.0687, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 14:11:19,811 INFO [train.py:968] (1/2) Epoch 12, batch 46100, giga_loss[loss=0.3229, simple_loss=0.3795, pruned_loss=0.1331, over 28862.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3793, pruned_loss=0.1307, over 5647092.63 frames. ], libri_tot_loss[loss=0.3118, simple_loss=0.3744, pruned_loss=0.1247, over 5643696.94 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3794, pruned_loss=0.131, over 5640930.29 frames. ], batch size: 186, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:12:04,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8187, 1.9395, 1.3242, 1.4737], device='cuda:1'), covar=tensor([0.0799, 0.0572, 0.1028, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0444, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 14:12:04,571 INFO [train.py:968] (1/2) Epoch 12, batch 46150, giga_loss[loss=0.3019, simple_loss=0.3624, pruned_loss=0.1207, over 29075.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3794, pruned_loss=0.1307, over 5645168.11 frames. ], libri_tot_loss[loss=0.3119, simple_loss=0.3744, pruned_loss=0.1247, over 5641168.43 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3795, pruned_loss=0.131, over 5642496.80 frames. ], batch size: 113, lr: 2.63e-03, grad_scale: 4.0
+2023-03-06 14:12:05,379 INFO [optim.py:369] (1/2) 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 14:12:42,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2968, 1.5879, 1.4428, 1.4706], device='cuda:1'), covar=tensor([0.0740, 0.0325, 0.0303, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 14:12:46,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8203, 1.0869, 2.8496, 2.8264], device='cuda:1'), covar=tensor([0.1634, 0.2425, 0.0591, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0681, 0.0603, 0.0879, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 14:12:49,539 INFO [train.py:968] (1/2) Epoch 12, batch 46200, libri_loss[loss=0.379, simple_loss=0.4126, pruned_loss=0.1727, over 20257.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3811, pruned_loss=0.1324, over 5602090.37 frames. ], libri_tot_loss[loss=0.3128, simple_loss=0.3751, pruned_loss=0.1252, over 5586968.28 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3807, pruned_loss=0.1324, over 5649158.91 frames. ], batch size: 186, lr: 2.63e-03, grad_scale: 2.0
+2023-03-06 14:13:15,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3832, 2.9170, 1.5560, 1.4637], device='cuda:1'), covar=tensor([0.0828, 0.0354, 0.0784, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0518, 0.0348, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 14:13:29,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-06 14:13:34,459 INFO [train.py:968] (1/2) Epoch 12, batch 46250, giga_loss[loss=0.364, simple_loss=0.3872, pruned_loss=0.1704, over 23579.00 frames. ], tot_loss[loss=0.323, simple_loss=0.3808, pruned_loss=0.1326, over 5561613.34 frames. ], libri_tot_loss[loss=0.3135, simple_loss=0.3757, pruned_loss=0.1257, over 5537716.98 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3801, pruned_loss=0.1322, over 5645809.41 frames. ], batch size: 705, lr: 2.63e-03, grad_scale: 2.0
+2023-03-06 14:13:36,850 INFO [optim.py:369] (1/2) 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 14:13:55,229 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-06 14:14:58,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-06 14:15:15,477 INFO [train.py:968] (1/2) Epoch 13, batch 50, giga_loss[loss=0.3366, simple_loss=0.3981, pruned_loss=0.1376, over 28969.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3778, pruned_loss=0.1127, over 1267775.04 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3598, pruned_loss=0.101, over 117533.14 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3794, pruned_loss=0.1137, over 1173910.21 frames. ], batch size: 106, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:15:29,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6554, 1.8002, 1.7734, 1.5818], device='cuda:1'), covar=tensor([0.1554, 0.2077, 0.1975, 0.1973], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0743, 0.0685, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 14:15:36,868 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=547536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:15:44,432 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 100, giga_loss[loss=0.2657, simple_loss=0.3359, pruned_loss=0.09771, over 28479.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3689, pruned_loss=0.1092, over 2243387.64 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09364, over 261028.18 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3714, pruned_loss=0.111, over 2075987.45 frames. ], batch size: 71, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:16:50,983 INFO [train.py:968] (1/2) Epoch 13, batch 150, giga_loss[loss=0.2302, simple_loss=0.3076, pruned_loss=0.07635, over 28675.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3523, pruned_loss=0.101, over 3009007.11 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.345, pruned_loss=0.09257, over 317400.70 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3534, pruned_loss=0.102, over 2848519.12 frames. ], batch size: 242, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:16:58,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=547624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:17:14,209 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 200, giga_loss[loss=0.264, simple_loss=0.325, pruned_loss=0.1015, over 26633.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3391, pruned_loss=0.09478, over 3608015.92 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3496, pruned_loss=0.09403, over 428007.22 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3387, pruned_loss=0.09511, over 3434845.23 frames. ], batch size: 555, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:17:43,126 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=547679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:17:45,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=547682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:17:53,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 14:18:11,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=547711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:18:15,218 INFO [train.py:968] (1/2) Epoch 13, batch 250, giga_loss[loss=0.2371, simple_loss=0.2925, pruned_loss=0.09086, over 23869.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3285, pruned_loss=0.08982, over 4063346.06 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3483, pruned_loss=0.09353, over 536628.47 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3276, pruned_loss=0.08987, over 3889195.90 frames. ], batch size: 705, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:18:40,221 INFO [optim.py:369] (1/2) 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 14:18:58,920 INFO [train.py:968] (1/2) Epoch 13, batch 300, giga_loss[loss=0.232, simple_loss=0.3074, pruned_loss=0.07835, over 28771.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3196, pruned_loss=0.08571, over 4428116.46 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3462, pruned_loss=0.09342, over 640615.50 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3183, pruned_loss=0.0855, over 4262847.56 frames. ], batch size: 119, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:19:42,614 INFO [train.py:968] (1/2) Epoch 13, batch 350, giga_loss[loss=0.2118, simple_loss=0.2832, pruned_loss=0.07019, over 28672.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3133, pruned_loss=0.08254, over 4708851.13 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3451, pruned_loss=0.09341, over 845107.31 frames. ], giga_tot_loss[loss=0.2373, simple_loss=0.3109, pruned_loss=0.08181, over 4529499.84 frames. ], batch size: 92, lr: 2.53e-03, grad_scale: 2.0
+2023-03-06 14:19:58,285 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:20:14,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=547854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:20:24,070 INFO [train.py:968] (1/2) Epoch 13, batch 400, giga_loss[loss=0.2364, simple_loss=0.3095, pruned_loss=0.08171, over 28751.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3088, pruned_loss=0.08017, over 4928555.05 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3453, pruned_loss=0.09322, over 917575.37 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3063, pruned_loss=0.07939, over 4775490.76 frames. ], batch size: 284, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:21:01,045 INFO [train.py:968] (1/2) Epoch 13, batch 450, giga_loss[loss=0.2083, simple_loss=0.2866, pruned_loss=0.06503, over 28290.00 frames. ], tot_loss[loss=0.2357, simple_loss=0.3095, pruned_loss=0.08092, over 5098518.89 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3484, pruned_loss=0.09502, over 1150403.06 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3051, pruned_loss=0.07932, over 4943676.11 frames. ], batch size: 368, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:21:13,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2139, 1.2515, 1.1294, 0.8929], device='cuda:1'), covar=tensor([0.0845, 0.0536, 0.1070, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0442, 0.0503, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 14:21:22,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 14:21:23,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6950, 2.7289, 2.3317, 1.9897], device='cuda:1'), covar=tensor([0.0778, 0.0202, 0.0239, 0.0926], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 14:21:28,539 INFO [optim.py:369] (1/2) 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 14:21:47,846 INFO [train.py:968] (1/2) Epoch 13, batch 500, giga_loss[loss=0.251, simple_loss=0.3115, pruned_loss=0.09525, over 27728.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.3074, pruned_loss=0.08039, over 5220211.02 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3486, pruned_loss=0.09499, over 1198109.88 frames. ], giga_tot_loss[loss=0.2308, simple_loss=0.3035, pruned_loss=0.07901, over 5091621.98 frames. ], batch size: 472, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:22:17,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=547999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:22:31,532 INFO [train.py:968] (1/2) Epoch 13, batch 550, giga_loss[loss=0.2126, simple_loss=0.2855, pruned_loss=0.06987, over 28927.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3049, pruned_loss=0.07901, over 5332918.45 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3486, pruned_loss=0.09487, over 1313939.39 frames. ], giga_tot_loss[loss=0.2278, simple_loss=0.3006, pruned_loss=0.07752, over 5216633.29 frames. ], batch size: 227, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:22:41,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 14:22:45,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1588, 2.9645, 2.8169, 1.4972], device='cuda:1'), covar=tensor([0.0974, 0.1087, 0.0958, 0.2496], device='cuda:1'), in_proj_covar=tensor([0.1063, 0.0987, 0.0862, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 14:22:58,439 INFO [optim.py:369] (1/2) 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 14:23:10,402 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 14:23:18,032 INFO [train.py:968] (1/2) Epoch 13, batch 600, giga_loss[loss=0.2131, simple_loss=0.2855, pruned_loss=0.07034, over 28819.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3015, pruned_loss=0.07719, over 5414349.45 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3481, pruned_loss=0.09444, over 1360017.03 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.2979, pruned_loss=0.07596, over 5318110.07 frames. ], batch size: 186, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:24:04,560 INFO [train.py:968] (1/2) Epoch 13, batch 650, giga_loss[loss=0.2111, simple_loss=0.285, pruned_loss=0.06867, over 28858.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2995, pruned_loss=0.07613, over 5477800.79 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3483, pruned_loss=0.09442, over 1467847.61 frames. ], giga_tot_loss[loss=0.2224, simple_loss=0.2954, pruned_loss=0.07469, over 5395115.92 frames. ], batch size: 186, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:24:05,666 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.02 vs. limit=2.0
+2023-03-06 14:24:26,350 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:24:27,328 INFO [optim.py:369] (1/2) 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 14:24:28,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=548145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:24:33,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0679, 1.1061, 3.6038, 3.1092], device='cuda:1'), covar=tensor([0.1734, 0.2767, 0.0461, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0672, 0.0598, 0.0871, 0.0781], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 14:24:46,193 INFO [train.py:968] (1/2) Epoch 13, batch 700, giga_loss[loss=0.2014, simple_loss=0.279, pruned_loss=0.0619, over 28720.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.2987, pruned_loss=0.07586, over 5517428.59 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3493, pruned_loss=0.0956, over 1626736.46 frames. ], giga_tot_loss[loss=0.2204, simple_loss=0.2933, pruned_loss=0.07372, over 5449108.53 frames. ], batch size: 92, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:24:54,321 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:25:19,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7412, 2.3938, 1.5986, 0.8268], device='cuda:1'), covar=tensor([0.6630, 0.3389, 0.3712, 0.6350], device='cuda:1'), in_proj_covar=tensor([0.1591, 0.1515, 0.1500, 0.1308], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 14:25:26,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:25:29,929 INFO [train.py:968] (1/2) Epoch 13, batch 750, libri_loss[loss=0.2734, simple_loss=0.3431, pruned_loss=0.1019, over 29562.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.296, pruned_loss=0.07462, over 5549681.10 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3496, pruned_loss=0.09576, over 1711819.13 frames. ], giga_tot_loss[loss=0.2178, simple_loss=0.2906, pruned_loss=0.07247, over 5488650.74 frames. ], batch size: 76, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:25:41,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:25:54,044 INFO [optim.py:369] (1/2) 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 14:26:06,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.68 vs. limit=5.0
+2023-03-06 14:26:13,091 INFO [train.py:968] (1/2) Epoch 13, batch 800, giga_loss[loss=0.2188, simple_loss=0.2999, pruned_loss=0.06883, over 29007.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.2956, pruned_loss=0.07474, over 5582337.21 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3501, pruned_loss=0.09603, over 1794959.91 frames. ], giga_tot_loss[loss=0.2176, simple_loss=0.2902, pruned_loss=0.07255, over 5527835.32 frames. ], batch size: 164, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:26:43,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4665, 1.7104, 1.4059, 1.5793], device='cuda:1'), covar=tensor([0.2434, 0.2415, 0.2697, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.0983, 0.1178, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 14:26:51,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2980, 1.9347, 1.4560, 0.4882], device='cuda:1'), covar=tensor([0.4061, 0.2273, 0.3244, 0.4934], device='cuda:1'), in_proj_covar=tensor([0.1591, 0.1512, 0.1500, 0.1309], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 14:26:55,300 INFO [train.py:968] (1/2) Epoch 13, batch 850, giga_loss[loss=0.3104, simple_loss=0.3822, pruned_loss=0.1192, over 28297.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.305, pruned_loss=0.07943, over 5614609.70 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3488, pruned_loss=0.09512, over 1975969.96 frames. ], giga_tot_loss[loss=0.2266, simple_loss=0.2989, pruned_loss=0.07716, over 5558041.13 frames. ], batch size: 65, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:27:08,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6776, 1.8200, 1.9699, 1.4809], device='cuda:1'), covar=tensor([0.1807, 0.2314, 0.1407, 0.1699], device='cuda:1'), in_proj_covar=tensor([0.0853, 0.0703, 0.0895, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-06 14:27:19,761 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-06 14:27:24,820 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=548353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:31,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:35,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=548356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:42,077 INFO [train.py:968] (1/2) Epoch 13, batch 900, giga_loss[loss=0.2815, simple_loss=0.3579, pruned_loss=0.1025, over 28888.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3187, pruned_loss=0.0866, over 5634760.93 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3479, pruned_loss=0.09433, over 2091159.71 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3133, pruned_loss=0.08477, over 5582981.56 frames. ], batch size: 186, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:27:49,351 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:27:51,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=548375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:28:00,594 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:28:15,078 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:28:24,155 INFO [train.py:968] (1/2) Epoch 13, batch 950, giga_loss[loss=0.2865, simple_loss=0.365, pruned_loss=0.104, over 28557.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3302, pruned_loss=0.09258, over 5647588.72 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3471, pruned_loss=0.09389, over 2240211.34 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3253, pruned_loss=0.09114, over 5596969.16 frames. ], batch size: 336, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:28:48,342 INFO [optim.py:369] (1/2) 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 14:29:06,858 INFO [train.py:968] (1/2) Epoch 13, batch 1000, libri_loss[loss=0.2916, simple_loss=0.3617, pruned_loss=0.1108, over 29502.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3373, pruned_loss=0.09542, over 5657525.99 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3462, pruned_loss=0.09362, over 2330766.99 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3335, pruned_loss=0.0944, over 5611642.69 frames. ], batch size: 82, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:29:38,048 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=548504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:29:46,505 INFO [train.py:968] (1/2) Epoch 13, batch 1050, giga_loss[loss=0.2562, simple_loss=0.3443, pruned_loss=0.08399, over 28609.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3402, pruned_loss=0.0952, over 5675937.50 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3456, pruned_loss=0.09332, over 2419263.47 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3372, pruned_loss=0.09456, over 5633922.46 frames. ], batch size: 307, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:30:16,972 INFO [optim.py:369] (1/2) 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 14:30:32,011 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 1100, giga_loss[loss=0.2972, simple_loss=0.3723, pruned_loss=0.1111, over 28791.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3426, pruned_loss=0.09618, over 5670527.72 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3461, pruned_loss=0.09354, over 2436441.16 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3402, pruned_loss=0.0956, over 5636739.00 frames. ], batch size: 186, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:31:16,804 INFO [train.py:968] (1/2) Epoch 13, batch 1150, giga_loss[loss=0.2628, simple_loss=0.3329, pruned_loss=0.09641, over 28498.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.345, pruned_loss=0.09801, over 5671665.89 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3458, pruned_loss=0.09343, over 2488493.36 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3432, pruned_loss=0.09766, over 5641776.05 frames. ], batch size: 85, lr: 2.53e-03, grad_scale: 4.0
+2023-03-06 14:31:18,883 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:31:47,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6930, 1.7467, 1.6213, 1.4665], device='cuda:1'), covar=tensor([0.1884, 0.1938, 0.1529, 0.1852], device='cuda:1'), in_proj_covar=tensor([0.1747, 0.1656, 0.1621, 0.1712], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 14:31:53,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 14:32:02,585 INFO [train.py:968] (1/2) Epoch 13, batch 1200, libri_loss[loss=0.2961, simple_loss=0.3804, pruned_loss=0.1059, over 29149.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3483, pruned_loss=0.1003, over 5679642.12 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3461, pruned_loss=0.09323, over 2555483.81 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3467, pruned_loss=0.1002, over 5652618.71 frames. ], batch size: 101, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:32:07,663 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-06 14:32:35,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5820, 1.7577, 1.8158, 1.6515], device='cuda:1'), covar=tensor([0.1517, 0.1797, 0.1729, 0.1616], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0735, 0.0678, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 14:32:43,838 INFO [train.py:968] (1/2) Epoch 13, batch 1250, giga_loss[loss=0.2762, simple_loss=0.361, pruned_loss=0.09571, over 28872.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3512, pruned_loss=0.1019, over 5687763.60 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3465, pruned_loss=0.0933, over 2638234.70 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3499, pruned_loss=0.102, over 5661498.94 frames. ], batch size: 174, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:32:54,153 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:33:26,186 INFO [train.py:968] (1/2) Epoch 13, batch 1300, libri_loss[loss=0.2726, simple_loss=0.3559, pruned_loss=0.09469, over 29516.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3549, pruned_loss=0.1032, over 5691887.37 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3473, pruned_loss=0.09351, over 2761592.48 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1035, over 5667190.73 frames. ], batch size: 84, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:34:05,660 INFO [train.py:968] (1/2) Epoch 13, batch 1350, giga_loss[loss=0.2805, simple_loss=0.3641, pruned_loss=0.09847, over 28919.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3565, pruned_loss=0.1036, over 5689045.46 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3474, pruned_loss=0.09345, over 2807867.50 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3557, pruned_loss=0.104, over 5667097.82 frames. ], batch size: 186, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:34:12,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 14:34:33,776 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 1400, giga_loss[loss=0.2827, simple_loss=0.3666, pruned_loss=0.09935, over 28255.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3578, pruned_loss=0.1034, over 5688763.42 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3481, pruned_loss=0.09364, over 2871585.75 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.357, pruned_loss=0.1038, over 5677206.55 frames. ], batch size: 368, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:34:52,635 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=548871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:34:54,529 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:35:18,860 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=548903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:35:30,003 INFO [train.py:968] (1/2) Epoch 13, batch 1450, giga_loss[loss=0.2555, simple_loss=0.3398, pruned_loss=0.08558, over 28695.00 frames. ], tot_loss[loss=0.279, simple_loss=0.356, pruned_loss=0.101, over 5692780.62 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3486, pruned_loss=0.09386, over 2960784.36 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3554, pruned_loss=0.1015, over 5678353.40 frames. ], batch size: 242, lr: 2.53e-03, grad_scale: 8.0
+2023-03-06 14:35:47,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:35:53,341 INFO [optim.py:369] (1/2) 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 14:36:09,641 INFO [train.py:968] (1/2) Epoch 13, batch 1500, giga_loss[loss=0.2543, simple_loss=0.344, pruned_loss=0.08229, over 28923.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3546, pruned_loss=0.09933, over 5704280.03 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3482, pruned_loss=0.09365, over 3033619.89 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3545, pruned_loss=0.1, over 5688111.65 frames. ], batch size: 164, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:36:28,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=548990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:36:38,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-06 14:36:46,405 INFO [train.py:968] (1/2) Epoch 13, batch 1550, giga_loss[loss=0.3426, simple_loss=0.3836, pruned_loss=0.1507, over 26697.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3535, pruned_loss=0.0981, over 5714911.03 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.349, pruned_loss=0.09432, over 3156585.18 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3533, pruned_loss=0.0985, over 5697102.84 frames. ], batch size: 555, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:36:52,893 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:36:56,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:37:13,201 INFO [optim.py:369] (1/2) 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 14:37:21,716 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:37:29,192 INFO [train.py:968] (1/2) Epoch 13, batch 1600, giga_loss[loss=0.284, simple_loss=0.3605, pruned_loss=0.1038, over 29079.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3547, pruned_loss=0.1, over 5695660.87 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3489, pruned_loss=0.09423, over 3228828.18 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3546, pruned_loss=0.1005, over 5684099.97 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:37:30,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9264, 1.2680, 1.0375, 0.1271], device='cuda:1'), covar=tensor([0.2417, 0.2002, 0.2719, 0.4246], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1490, 0.1493, 0.1295], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 14:37:40,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:37:43,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549082.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:38:05,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:38:08,111 INFO [train.py:968] (1/2) Epoch 13, batch 1650, giga_loss[loss=0.3143, simple_loss=0.3778, pruned_loss=0.1254, over 28796.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3585, pruned_loss=0.1048, over 5695126.37 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3491, pruned_loss=0.09443, over 3309427.16 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3586, pruned_loss=0.1053, over 5690256.03 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:38:24,355 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:38:27,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:38:36,857 INFO [optim.py:369] (1/2) 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 14:38:44,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4628, 1.6402, 1.7180, 1.2978], device='cuda:1'), covar=tensor([0.1722, 0.2373, 0.1344, 0.1585], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0696, 0.0888, 0.0792], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 14:38:53,124 INFO [train.py:968] (1/2) Epoch 13, batch 1700, giga_loss[loss=0.2954, simple_loss=0.3537, pruned_loss=0.1186, over 28548.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3595, pruned_loss=0.1072, over 5705041.17 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3488, pruned_loss=0.09417, over 3335599.37 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3598, pruned_loss=0.1079, over 5699331.10 frames. ], batch size: 60, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:38:53,373 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:39:27,282 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:39:38,850 INFO [train.py:968] (1/2) Epoch 13, batch 1750, libri_loss[loss=0.254, simple_loss=0.3279, pruned_loss=0.08998, over 29545.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.358, pruned_loss=0.1074, over 5698317.76 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3482, pruned_loss=0.09378, over 3386885.70 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3588, pruned_loss=0.1084, over 5689598.63 frames. ], batch size: 76, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:40:06,040 INFO [optim.py:369] (1/2) 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 14:40:21,417 INFO [train.py:968] (1/2) Epoch 13, batch 1800, giga_loss[loss=0.2837, simple_loss=0.3564, pruned_loss=0.1056, over 28311.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.356, pruned_loss=0.1071, over 5695241.73 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3481, pruned_loss=0.09374, over 3449107.86 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.3569, pruned_loss=0.1082, over 5683525.69 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:40:23,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0455, 1.1296, 3.7379, 3.1116], device='cuda:1'), covar=tensor([0.1759, 0.2785, 0.0420, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0666, 0.0593, 0.0861, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 14:41:04,379 INFO [train.py:968] (1/2) Epoch 13, batch 1850, libri_loss[loss=0.2728, simple_loss=0.3579, pruned_loss=0.09383, over 25613.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3544, pruned_loss=0.1056, over 5691119.35 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3478, pruned_loss=0.09347, over 3517932.07 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3555, pruned_loss=0.107, over 5679964.87 frames. ], batch size: 136, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:41:04,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6299, 1.6542, 1.2661, 1.2485], device='cuda:1'), covar=tensor([0.0766, 0.0533, 0.0923, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0442, 0.0507, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 14:41:19,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2988, 1.2647, 1.1337, 1.4608], device='cuda:1'), covar=tensor([0.0750, 0.0351, 0.0333, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 14:41:29,660 INFO [optim.py:369] (1/2) 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 14:41:31,269 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549349.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:41:48,569 INFO [train.py:968] (1/2) Epoch 13, batch 1900, giga_loss[loss=0.2762, simple_loss=0.3437, pruned_loss=0.1043, over 28637.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.353, pruned_loss=0.1042, over 5698805.41 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3479, pruned_loss=0.09395, over 3600117.32 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.354, pruned_loss=0.1054, over 5683169.55 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:42:35,461 INFO [train.py:968] (1/2) Epoch 13, batch 1950, giga_loss[loss=0.2331, simple_loss=0.313, pruned_loss=0.07667, over 28874.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3481, pruned_loss=0.1009, over 5695628.11 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09379, over 3645984.87 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3491, pruned_loss=0.1021, over 5679414.53 frames. ], batch size: 199, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:43:07,374 INFO [optim.py:369] (1/2) 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 14:43:22,684 INFO [train.py:968] (1/2) Epoch 13, batch 2000, giga_loss[loss=0.2341, simple_loss=0.3118, pruned_loss=0.07825, over 28748.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3421, pruned_loss=0.09764, over 5687660.35 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3476, pruned_loss=0.09367, over 3698868.76 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3429, pruned_loss=0.0988, over 5672215.13 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:44:06,047 INFO [train.py:968] (1/2) Epoch 13, batch 2050, giga_loss[loss=0.2693, simple_loss=0.3348, pruned_loss=0.1019, over 28638.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3375, pruned_loss=0.09493, over 5687899.54 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3469, pruned_loss=0.09334, over 3794376.67 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3382, pruned_loss=0.09619, over 5668190.45 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:44:37,367 INFO [optim.py:369] (1/2) 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 14:44:56,605 INFO [train.py:968] (1/2) Epoch 13, batch 2100, giga_loss[loss=0.2717, simple_loss=0.343, pruned_loss=0.1002, over 28567.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3344, pruned_loss=0.09393, over 5668243.14 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3465, pruned_loss=0.09308, over 3815583.44 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3351, pruned_loss=0.09507, over 5650574.25 frames. ], batch size: 307, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:44:57,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-06 14:45:05,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=549578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:45:29,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-06 14:45:34,728 INFO [train.py:968] (1/2) Epoch 13, batch 2150, giga_loss[loss=0.2631, simple_loss=0.3374, pruned_loss=0.0944, over 28735.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3364, pruned_loss=0.09405, over 5682231.98 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3464, pruned_loss=0.09277, over 3903486.53 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3365, pruned_loss=0.09521, over 5662651.18 frames. ], batch size: 99, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:45:59,955 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 2200, giga_loss[loss=0.2433, simple_loss=0.3236, pruned_loss=0.08152, over 29038.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3361, pruned_loss=0.09366, over 5693513.54 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3464, pruned_loss=0.09265, over 3949107.47 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.336, pruned_loss=0.09468, over 5676917.56 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:46:57,665 INFO [train.py:968] (1/2) Epoch 13, batch 2250, libri_loss[loss=0.2322, simple_loss=0.321, pruned_loss=0.07167, over 29588.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3341, pruned_loss=0.09224, over 5700382.05 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.347, pruned_loss=0.09265, over 4015869.28 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3331, pruned_loss=0.09307, over 5680521.38 frames. ], batch size: 75, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:47:02,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:47:04,495 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=549724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:47:04,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:47:24,081 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 13, batch 2300, giga_loss[loss=0.2611, simple_loss=0.3261, pruned_loss=0.09805, over 28896.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3338, pruned_loss=0.09215, over 5699492.83 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3479, pruned_loss=0.09295, over 4069910.57 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.332, pruned_loss=0.09259, over 5687891.20 frames. ], batch size: 112, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:47:54,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:48:18,378 INFO [train.py:968] (1/2) Epoch 13, batch 2350, giga_loss[loss=0.2117, simple_loss=0.2863, pruned_loss=0.06859, over 28938.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3312, pruned_loss=0.09084, over 5706574.85 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3489, pruned_loss=0.0934, over 4124563.13 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3288, pruned_loss=0.09089, over 5691668.32 frames. ], batch size: 213, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:48:47,411 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 2400, giga_loss[loss=0.242, simple_loss=0.3153, pruned_loss=0.08435, over 28902.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3288, pruned_loss=0.08983, over 5705773.24 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3496, pruned_loss=0.09374, over 4141991.88 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3262, pruned_loss=0.08959, over 5692487.24 frames. ], batch size: 186, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:49:02,919 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=549867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:49:07,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=549870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 14:49:28,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=549899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:49:43,281 INFO [train.py:968] (1/2) Epoch 13, batch 2450, libri_loss[loss=0.2967, simple_loss=0.3796, pruned_loss=0.1069, over 27782.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3268, pruned_loss=0.08862, over 5709959.24 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3501, pruned_loss=0.09387, over 4201195.48 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3235, pruned_loss=0.08816, over 5694471.35 frames. ], batch size: 116, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:49:53,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=549929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:50:06,787 INFO [optim.py:369] (1/2) 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 14:50:20,002 INFO [train.py:968] (1/2) Epoch 13, batch 2500, giga_loss[loss=0.2382, simple_loss=0.3132, pruned_loss=0.08162, over 28973.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3245, pruned_loss=0.08802, over 5714201.18 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3507, pruned_loss=0.09411, over 4214572.87 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3213, pruned_loss=0.08745, over 5703633.28 frames. ], batch size: 227, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:50:47,087 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-06 14:51:03,526 INFO [train.py:968] (1/2) Epoch 13, batch 2550, giga_loss[loss=0.2565, simple_loss=0.3162, pruned_loss=0.09833, over 28518.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3215, pruned_loss=0.08679, over 5719708.93 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.351, pruned_loss=0.09423, over 4222092.00 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3187, pruned_loss=0.08624, over 5711431.45 frames. ], batch size: 78, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:51:04,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9291, 3.7231, 3.5859, 1.7940], device='cuda:1'), covar=tensor([0.0642, 0.0816, 0.0829, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.1046, 0.0973, 0.0853, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 14:51:30,794 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 2600, giga_loss[loss=0.2644, simple_loss=0.3376, pruned_loss=0.09563, over 27983.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3203, pruned_loss=0.08578, over 5704606.60 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3518, pruned_loss=0.09452, over 4255182.87 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3167, pruned_loss=0.08494, over 5713444.33 frames. ], batch size: 412, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:51:57,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:52:16,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3759, 1.5652, 1.4658, 1.3180], device='cuda:1'), covar=tensor([0.2293, 0.1985, 0.1305, 0.1786], device='cuda:1'), in_proj_covar=tensor([0.1739, 0.1642, 0.1611, 0.1712], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 14:52:22,439 INFO [train.py:968] (1/2) Epoch 13, batch 2650, giga_loss[loss=0.28, simple_loss=0.3423, pruned_loss=0.1088, over 28037.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3199, pruned_loss=0.08586, over 5714293.09 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3517, pruned_loss=0.09429, over 4303886.81 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3163, pruned_loss=0.08511, over 5716516.65 frames. ], batch size: 412, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:52:52,939 INFO [optim.py:369] (1/2) 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 14:53:01,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:53:06,400 INFO [train.py:968] (1/2) Epoch 13, batch 2700, giga_loss[loss=0.2428, simple_loss=0.3213, pruned_loss=0.08217, over 28929.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3244, pruned_loss=0.08874, over 5719158.76 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3525, pruned_loss=0.09463, over 4334219.29 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3204, pruned_loss=0.08778, over 5718931.13 frames. ], batch size: 227, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:53:08,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5286, 3.5199, 1.5873, 1.5851], device='cuda:1'), covar=tensor([0.0963, 0.0343, 0.0886, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0503, 0.0344, 0.0392], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 14:53:17,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7085, 1.7250, 1.3458, 1.3003], device='cuda:1'), covar=tensor([0.0811, 0.0650, 0.0924, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0439, 0.0503, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 14:53:28,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6195, 1.7856, 1.7436, 1.6264], device='cuda:1'), covar=tensor([0.1750, 0.1857, 0.2000, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0735, 0.0680, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 14:53:52,286 INFO [train.py:968] (1/2) Epoch 13, batch 2750, giga_loss[loss=0.2725, simple_loss=0.3413, pruned_loss=0.1019, over 29035.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3288, pruned_loss=0.0917, over 5711594.21 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3525, pruned_loss=0.09458, over 4346380.57 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3254, pruned_loss=0.09094, over 5712988.94 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:54:17,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:54:24,486 INFO [optim.py:369] (1/2) 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 14:54:38,873 INFO [train.py:968] (1/2) Epoch 13, batch 2800, giga_loss[loss=0.2535, simple_loss=0.3237, pruned_loss=0.0916, over 28515.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.336, pruned_loss=0.09581, over 5704275.36 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3526, pruned_loss=0.09457, over 4382119.33 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3326, pruned_loss=0.09519, over 5710215.10 frames. ], batch size: 71, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:54:42,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8455, 5.6497, 5.3010, 2.7415], device='cuda:1'), covar=tensor([0.0377, 0.0522, 0.0635, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.1049, 0.0972, 0.0851, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 14:55:07,967 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550293.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 14:55:16,532 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:55:17,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:55:18,459 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 2850, giga_loss[loss=0.2561, simple_loss=0.3363, pruned_loss=0.08792, over 28223.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3436, pruned_loss=0.1012, over 5692915.71 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3524, pruned_loss=0.09438, over 4404589.52 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3409, pruned_loss=0.1009, over 5694960.86 frames. ], batch size: 65, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:55:45,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:55:57,104 INFO [optim.py:369] (1/2) 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 14:55:57,406 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:56:12,906 INFO [train.py:968] (1/2) Epoch 13, batch 2900, giga_loss[loss=0.3263, simple_loss=0.3956, pruned_loss=0.1284, over 28680.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3471, pruned_loss=0.1013, over 5706077.01 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3522, pruned_loss=0.09413, over 4441487.41 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.345, pruned_loss=0.1014, over 5703294.36 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:56:38,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6295, 1.6782, 1.3317, 1.3197], device='cuda:1'), covar=tensor([0.0722, 0.0454, 0.0966, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0438, 0.0502, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 14:56:58,325 INFO [train.py:968] (1/2) Epoch 13, batch 2950, giga_loss[loss=0.2937, simple_loss=0.3634, pruned_loss=0.112, over 28663.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3521, pruned_loss=0.1037, over 5704278.32 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3522, pruned_loss=0.09416, over 4476069.66 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3504, pruned_loss=0.104, over 5698940.12 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 14:57:03,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:27,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:28,229 INFO [optim.py:369] (1/2) 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 14:57:30,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:33,028 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:37,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:40,247 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:57:43,683 INFO [train.py:968] (1/2) Epoch 13, batch 3000, giga_loss[loss=0.3632, simple_loss=0.4087, pruned_loss=0.1589, over 27520.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3577, pruned_loss=0.1077, over 5691163.12 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3514, pruned_loss=0.09381, over 4524582.67 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3569, pruned_loss=0.1085, over 5681301.62 frames. ], batch size: 472, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:57:43,684 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 14:57:48,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5855, 1.6698, 1.2910, 1.3537], device='cuda:1'), covar=tensor([0.0785, 0.0468, 0.0974, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0441, 0.0505, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 14:57:52,533 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 14:58:03,814 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:58:35,505 INFO [train.py:968] (1/2) Epoch 13, batch 3050, giga_loss[loss=0.2652, simple_loss=0.3382, pruned_loss=0.09614, over 28550.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3561, pruned_loss=0.106, over 5695234.80 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3516, pruned_loss=0.09394, over 4558584.68 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3554, pruned_loss=0.1068, over 5682788.55 frames. ], batch size: 336, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:59:04,794 INFO [optim.py:369] (1/2) 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 14:59:18,037 INFO [train.py:968] (1/2) Epoch 13, batch 3100, giga_loss[loss=0.2396, simple_loss=0.3243, pruned_loss=0.0775, over 28827.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3525, pruned_loss=0.1032, over 5703154.52 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.352, pruned_loss=0.09447, over 4591838.70 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3518, pruned_loss=0.1037, over 5688784.47 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 14:59:50,283 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 14:59:54,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:00:03,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:00:04,059 INFO [train.py:968] (1/2) Epoch 13, batch 3150, giga_loss[loss=0.2559, simple_loss=0.3271, pruned_loss=0.0923, over 28899.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3511, pruned_loss=0.1016, over 5711288.44 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3523, pruned_loss=0.09461, over 4611598.09 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3503, pruned_loss=0.102, over 5697271.65 frames. ], batch size: 112, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:00:19,893 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:00:23,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-06 15:00:36,040 INFO [optim.py:369] (1/2) 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 15:00:49,831 INFO [train.py:968] (1/2) Epoch 13, batch 3200, giga_loss[loss=0.2769, simple_loss=0.3494, pruned_loss=0.1022, over 28613.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3515, pruned_loss=0.1011, over 5711164.92 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3523, pruned_loss=0.09461, over 4611598.09 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3509, pruned_loss=0.1014, over 5700255.48 frames. ], batch size: 85, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:00:52,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550668.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:01:08,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-06 15:01:30,019 INFO [train.py:968] (1/2) Epoch 13, batch 3250, giga_loss[loss=0.3177, simple_loss=0.3784, pruned_loss=0.1285, over 27653.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3538, pruned_loss=0.1023, over 5712412.36 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3527, pruned_loss=0.09498, over 4640464.47 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3531, pruned_loss=0.1025, over 5702396.34 frames. ], batch size: 472, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:01:31,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3707, 1.7376, 1.5422, 1.4297], device='cuda:1'), covar=tensor([0.0767, 0.0302, 0.0294, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 15:01:39,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:02,540 INFO [optim.py:369] (1/2) 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 15:02:08,868 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:11,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:14,976 INFO [train.py:968] (1/2) Epoch 13, batch 3300, libri_loss[loss=0.265, simple_loss=0.3474, pruned_loss=0.0913, over 29545.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3548, pruned_loss=0.1032, over 5708265.85 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3527, pruned_loss=0.09491, over 4672293.04 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1036, over 5695390.38 frames. ], batch size: 78, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:02:36,508 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:41,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:02:53,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550811.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:02:55,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550814.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:02:55,505 INFO [train.py:968] (1/2) Epoch 13, batch 3350, giga_loss[loss=0.3058, simple_loss=0.3705, pruned_loss=0.1205, over 28485.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3564, pruned_loss=0.1043, over 5714576.09 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3535, pruned_loss=0.09526, over 4723124.09 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3555, pruned_loss=0.1047, over 5699511.51 frames. ], batch size: 78, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:03:08,093 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:03:13,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=550836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:03:20,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550843.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:03:28,256 INFO [optim.py:369] (1/2) 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 15:03:42,894 INFO [train.py:968] (1/2) Epoch 13, batch 3400, giga_loss[loss=0.2655, simple_loss=0.3407, pruned_loss=0.09518, over 28823.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3574, pruned_loss=0.1055, over 5719435.92 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3533, pruned_loss=0.09516, over 4734496.67 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3568, pruned_loss=0.106, over 5706234.84 frames. ], batch size: 199, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:03:44,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:03:46,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:13,409 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:16,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0337, 5.8594, 5.5576, 2.9751], device='cuda:1'), covar=tensor([0.0442, 0.0579, 0.0785, 0.1488], device='cuda:1'), in_proj_covar=tensor([0.1051, 0.0980, 0.0857, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 15:04:23,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2594, 2.5243, 1.2768, 1.3090], device='cuda:1'), covar=tensor([0.0970, 0.0308, 0.0879, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0508, 0.0344, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 15:04:26,671 INFO [train.py:968] (1/2) Epoch 13, batch 3450, giga_loss[loss=0.2741, simple_loss=0.3518, pruned_loss=0.09817, over 28902.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3563, pruned_loss=0.1047, over 5725399.89 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3531, pruned_loss=0.09503, over 4746230.28 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.356, pruned_loss=0.1052, over 5713375.75 frames. ], batch size: 227, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:04:47,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:48,860 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=550941.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:04:49,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:04:53,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6029, 1.7490, 1.5713, 1.6143], device='cuda:1'), covar=tensor([0.1553, 0.2304, 0.2095, 0.1933], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0734, 0.0678, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 15:04:56,750 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 3500, giga_loss[loss=0.2458, simple_loss=0.3303, pruned_loss=0.0806, over 28955.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3569, pruned_loss=0.1048, over 5723790.90 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3532, pruned_loss=0.09521, over 4774667.24 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3567, pruned_loss=0.1053, over 5710446.64 frames. ], batch size: 112, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:05:13,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=550971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:14,319 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:16,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:20,708 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=550979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:22,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=550982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:39,164 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=551001.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:05:42,213 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:48,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:05:51,491 INFO [train.py:968] (1/2) Epoch 13, batch 3550, giga_loss[loss=0.2846, simple_loss=0.3646, pruned_loss=0.1023, over 29052.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3568, pruned_loss=0.1037, over 5726215.95 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3526, pruned_loss=0.09495, over 4802634.97 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3572, pruned_loss=0.1045, over 5711590.97 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:05:59,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-06 15:06:08,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4242, 1.6234, 1.6170, 1.1830], device='cuda:1'), covar=tensor([0.1742, 0.2605, 0.1502, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0702, 0.0890, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 15:06:24,639 INFO [optim.py:369] (1/2) 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 15:06:37,358 INFO [train.py:968] (1/2) Epoch 13, batch 3600, giga_loss[loss=0.2922, simple_loss=0.3635, pruned_loss=0.1104, over 28719.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3577, pruned_loss=0.1038, over 5715480.42 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3526, pruned_loss=0.09501, over 4808444.92 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.358, pruned_loss=0.1045, over 5710788.92 frames. ], batch size: 242, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:07:16,055 INFO [train.py:968] (1/2) Epoch 13, batch 3650, giga_loss[loss=0.2575, simple_loss=0.3412, pruned_loss=0.08688, over 28859.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3564, pruned_loss=0.1031, over 5722872.54 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3528, pruned_loss=0.095, over 4847560.62 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3568, pruned_loss=0.1039, over 5716214.69 frames. ], batch size: 145, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:07:48,491 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 3700, giga_loss[loss=0.2683, simple_loss=0.3422, pruned_loss=0.09722, over 27960.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3545, pruned_loss=0.1026, over 5717161.03 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3531, pruned_loss=0.09515, over 4864477.15 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3545, pruned_loss=0.1033, over 5711992.46 frames. ], batch size: 412, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:08:39,441 INFO [train.py:968] (1/2) Epoch 13, batch 3750, giga_loss[loss=0.2475, simple_loss=0.329, pruned_loss=0.08302, over 28723.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3512, pruned_loss=0.1006, over 5724264.97 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.353, pruned_loss=0.09545, over 4897616.78 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3513, pruned_loss=0.1011, over 5716697.36 frames. ], batch size: 119, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:08:56,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2492, 1.2394, 1.2520, 1.4480], device='cuda:1'), covar=tensor([0.0757, 0.0354, 0.0312, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0053, 0.0091], device='cuda:1')
+2023-03-06 15:09:13,444 INFO [optim.py:369] (1/2) 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 15:09:23,637 INFO [train.py:968] (1/2) Epoch 13, batch 3800, giga_loss[loss=0.2715, simple_loss=0.3516, pruned_loss=0.09567, over 29085.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.352, pruned_loss=0.1011, over 5733630.19 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3532, pruned_loss=0.09554, over 4912803.33 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3519, pruned_loss=0.1015, over 5725258.56 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:10:06,158 INFO [train.py:968] (1/2) Epoch 13, batch 3850, libri_loss[loss=0.2682, simple_loss=0.3504, pruned_loss=0.09299, over 25709.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3525, pruned_loss=0.1017, over 5728645.52 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3532, pruned_loss=0.09565, over 4928481.73 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3524, pruned_loss=0.1021, over 5722776.24 frames. ], batch size: 136, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:10:08,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=551316.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:10:34,985 INFO [optim.py:369] (1/2) 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 15:10:45,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2286, 1.7831, 1.4208, 0.4111], device='cuda:1'), covar=tensor([0.3543, 0.2172, 0.3364, 0.4614], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1476, 0.1480, 0.1284], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:10:46,112 INFO [train.py:968] (1/2) Epoch 13, batch 3900, giga_loss[loss=0.2595, simple_loss=0.3427, pruned_loss=0.08811, over 28443.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3528, pruned_loss=0.1015, over 5720967.38 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3525, pruned_loss=0.09546, over 4960428.41 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3532, pruned_loss=0.1021, over 5719583.50 frames. ], batch size: 65, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:10:55,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=551376.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:11:27,915 INFO [train.py:968] (1/2) Epoch 13, batch 3950, giga_loss[loss=0.2567, simple_loss=0.3404, pruned_loss=0.08654, over 28889.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.351, pruned_loss=0.09979, over 5718122.20 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3524, pruned_loss=0.09547, over 4979825.95 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3515, pruned_loss=0.1004, over 5716371.22 frames. ], batch size: 174, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:11:32,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-06 15:11:42,341 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=551433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:11:57,824 INFO [optim.py:369] (1/2) 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 15:12:02,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.85 vs. limit=2.0
+2023-03-06 15:12:03,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=551459.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:12:05,590 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 15:12:07,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=551462.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:12:09,318 INFO [train.py:968] (1/2) Epoch 13, batch 4000, giga_loss[loss=0.2852, simple_loss=0.3546, pruned_loss=0.1079, over 28799.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3506, pruned_loss=0.09973, over 5713765.21 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3524, pruned_loss=0.09541, over 4988472.42 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3509, pruned_loss=0.1003, over 5718668.39 frames. ], batch size: 284, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:12:17,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4975, 3.5279, 1.5526, 1.5698], device='cuda:1'), covar=tensor([0.0940, 0.0269, 0.0884, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0502, 0.0342, 0.0390], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 15:12:28,806 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551491.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:12:49,081 INFO [train.py:968] (1/2) Epoch 13, batch 4050, giga_loss[loss=0.2875, simple_loss=0.3559, pruned_loss=0.1095, over 27600.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3488, pruned_loss=0.09948, over 5705998.83 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3523, pruned_loss=0.0955, over 4995888.33 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3491, pruned_loss=0.09993, over 5709679.16 frames. ], batch size: 472, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:12:52,357 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=551519.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:12:54,112 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=551522.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:13:11,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8942, 3.6906, 3.4782, 1.6831], device='cuda:1'), covar=tensor([0.0658, 0.0833, 0.0845, 0.2152], device='cuda:1'), in_proj_covar=tensor([0.1059, 0.0980, 0.0858, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 15:13:17,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551551.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:13:17,779 INFO [optim.py:369] (1/2) 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 15:13:25,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-06 15:13:28,219 INFO [train.py:968] (1/2) Epoch 13, batch 4100, giga_loss[loss=0.2674, simple_loss=0.3399, pruned_loss=0.09741, over 28793.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3463, pruned_loss=0.09839, over 5707500.18 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3523, pruned_loss=0.09548, over 5018606.58 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3465, pruned_loss=0.09886, over 5705745.78 frames. ], batch size: 99, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:14:08,982 INFO [train.py:968] (1/2) Epoch 13, batch 4150, giga_loss[loss=0.266, simple_loss=0.3389, pruned_loss=0.09652, over 29075.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3455, pruned_loss=0.09847, over 5705447.00 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3515, pruned_loss=0.09506, over 5037413.89 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3461, pruned_loss=0.09926, over 5702371.89 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:14:22,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5913, 1.3343, 4.8005, 3.5284], device='cuda:1'), covar=tensor([0.1506, 0.2603, 0.0306, 0.0773], device='cuda:1'), in_proj_covar=tensor([0.0668, 0.0593, 0.0857, 0.0779], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 15:14:40,986 INFO [optim.py:369] (1/2) 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 15:14:49,798 INFO [train.py:968] (1/2) Epoch 13, batch 4200, giga_loss[loss=0.2519, simple_loss=0.3285, pruned_loss=0.08762, over 28957.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3442, pruned_loss=0.09851, over 5704128.00 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3513, pruned_loss=0.09511, over 5046506.35 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3447, pruned_loss=0.09914, over 5702500.43 frames. ], batch size: 227, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:15:11,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7606, 1.7426, 1.2996, 1.4272], device='cuda:1'), covar=tensor([0.0787, 0.0726, 0.1027, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0440, 0.0504, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 15:15:29,908 INFO [train.py:968] (1/2) Epoch 13, batch 4250, giga_loss[loss=0.2682, simple_loss=0.3451, pruned_loss=0.09567, over 28659.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.343, pruned_loss=0.09799, over 5705101.00 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3514, pruned_loss=0.0951, over 5073531.26 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3431, pruned_loss=0.0986, over 5699859.55 frames. ], batch size: 336, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:16:03,414 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 4300, giga_loss[loss=0.2781, simple_loss=0.3493, pruned_loss=0.1034, over 28247.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.34, pruned_loss=0.09668, over 5714901.20 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3505, pruned_loss=0.09452, over 5094091.85 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3407, pruned_loss=0.0977, over 5706385.71 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:16:27,431 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 15:16:47,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=551808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:16:52,378 INFO [train.py:968] (1/2) Epoch 13, batch 4350, giga_loss[loss=0.2575, simple_loss=0.3311, pruned_loss=0.09196, over 28886.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3391, pruned_loss=0.09693, over 5710959.06 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3501, pruned_loss=0.09431, over 5112920.12 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3397, pruned_loss=0.09796, over 5700833.18 frames. ], batch size: 199, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:16:54,749 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-06 15:17:24,399 INFO [optim.py:369] (1/2) 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 15:17:33,721 INFO [train.py:968] (1/2) Epoch 13, batch 4400, giga_loss[loss=0.2521, simple_loss=0.3277, pruned_loss=0.08829, over 28710.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3377, pruned_loss=0.09619, over 5720579.34 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3505, pruned_loss=0.09453, over 5131452.24 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3375, pruned_loss=0.09686, over 5709220.57 frames. ], batch size: 262, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:17:39,233 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=551872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:18:15,339 INFO [train.py:968] (1/2) Epoch 13, batch 4450, giga_loss[loss=0.2582, simple_loss=0.3293, pruned_loss=0.09348, over 28431.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3385, pruned_loss=0.0965, over 5719814.01 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3508, pruned_loss=0.09486, over 5146937.00 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3379, pruned_loss=0.09678, over 5707272.73 frames. ], batch size: 60, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:18:29,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7873, 3.5911, 3.3786, 1.6444], device='cuda:1'), covar=tensor([0.0676, 0.0812, 0.0778, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.1059, 0.0981, 0.0857, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 15:18:35,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7598, 1.1379, 1.1166, 0.9366], device='cuda:1'), covar=tensor([0.1670, 0.1274, 0.2019, 0.1548], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0735, 0.0677, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 15:18:49,785 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=551951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:18:51,642 INFO [optim.py:369] (1/2) 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 15:18:52,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=551954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:18:57,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-06 15:19:00,973 INFO [train.py:968] (1/2) Epoch 13, batch 4500, giga_loss[loss=0.3009, simple_loss=0.3726, pruned_loss=0.1145, over 27667.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3426, pruned_loss=0.09879, over 5699373.01 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3511, pruned_loss=0.09528, over 5151854.69 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3415, pruned_loss=0.09872, over 5697699.86 frames. ], batch size: 472, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:19:16,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=551983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:19:41,451 INFO [train.py:968] (1/2) Epoch 13, batch 4550, giga_loss[loss=0.2436, simple_loss=0.3187, pruned_loss=0.0842, over 28652.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3454, pruned_loss=0.09957, over 5706304.53 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3514, pruned_loss=0.09554, over 5172075.05 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3442, pruned_loss=0.09942, over 5701437.67 frames. ], batch size: 85, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:20:16,917 INFO [optim.py:369] (1/2) 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 15:20:26,586 INFO [train.py:968] (1/2) Epoch 13, batch 4600, giga_loss[loss=0.238, simple_loss=0.3256, pruned_loss=0.07525, over 28672.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.348, pruned_loss=0.1006, over 5695930.92 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3519, pruned_loss=0.0961, over 5182890.22 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3465, pruned_loss=0.1001, over 5691921.63 frames. ], batch size: 284, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:20:46,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4829, 5.2800, 5.0021, 2.3770], device='cuda:1'), covar=tensor([0.0349, 0.0508, 0.0566, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.1053, 0.0978, 0.0855, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 15:20:53,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2466, 1.8186, 1.3804, 0.4577], device='cuda:1'), covar=tensor([0.3323, 0.2074, 0.3262, 0.4633], device='cuda:1'), in_proj_covar=tensor([0.1562, 0.1473, 0.1480, 0.1283], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:21:09,413 INFO [train.py:968] (1/2) Epoch 13, batch 4650, giga_loss[loss=0.2555, simple_loss=0.3334, pruned_loss=0.08883, over 29044.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3475, pruned_loss=0.09973, over 5696342.33 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3518, pruned_loss=0.09612, over 5206876.74 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3463, pruned_loss=0.09947, over 5689249.49 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:21:34,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1765, 1.8126, 1.4189, 0.3606], device='cuda:1'), covar=tensor([0.3584, 0.2270, 0.3641, 0.4592], device='cuda:1'), in_proj_covar=tensor([0.1561, 0.1476, 0.1481, 0.1282], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:21:44,220 INFO [optim.py:369] (1/2) 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 15:21:46,360 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 4700, giga_loss[loss=0.2466, simple_loss=0.3306, pruned_loss=0.08136, over 28870.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3465, pruned_loss=0.09951, over 5703812.95 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3516, pruned_loss=0.09606, over 5211636.77 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3457, pruned_loss=0.09938, over 5698285.54 frames. ], batch size: 174, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:22:33,183 INFO [train.py:968] (1/2) Epoch 13, batch 4750, giga_loss[loss=0.2646, simple_loss=0.3379, pruned_loss=0.09568, over 29089.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3459, pruned_loss=0.09935, over 5705289.46 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3509, pruned_loss=0.09569, over 5234678.99 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3456, pruned_loss=0.09966, over 5695253.94 frames. ], batch size: 128, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:22:44,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4524, 1.6385, 1.5387, 1.4317], device='cuda:1'), covar=tensor([0.2450, 0.1820, 0.1549, 0.1926], device='cuda:1'), in_proj_covar=tensor([0.1723, 0.1642, 0.1608, 0.1697], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 15:22:49,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3986, 1.6243, 1.3296, 1.3252], device='cuda:1'), covar=tensor([0.2405, 0.2380, 0.2744, 0.2251], device='cuda:1'), in_proj_covar=tensor([0.1340, 0.0982, 0.1178, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 15:23:00,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=552247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:23:05,110 INFO [optim.py:369] (1/2) 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 15:23:14,913 INFO [train.py:968] (1/2) Epoch 13, batch 4800, giga_loss[loss=0.2671, simple_loss=0.3467, pruned_loss=0.09375, over 28246.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.347, pruned_loss=0.09994, over 5706039.63 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3512, pruned_loss=0.09583, over 5250294.48 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3464, pruned_loss=0.1001, over 5694468.33 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:23:23,973 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-06 15:23:56,779 INFO [train.py:968] (1/2) Epoch 13, batch 4850, giga_loss[loss=0.2833, simple_loss=0.3458, pruned_loss=0.1104, over 28710.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3503, pruned_loss=0.1017, over 5707591.68 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.351, pruned_loss=0.09573, over 5269872.70 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3498, pruned_loss=0.1021, over 5692795.54 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:24:30,519 INFO [optim.py:369] (1/2) 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 15:24:40,242 INFO [train.py:968] (1/2) Epoch 13, batch 4900, giga_loss[loss=0.2851, simple_loss=0.3575, pruned_loss=0.1063, over 28907.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3533, pruned_loss=0.1033, over 5715185.77 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3509, pruned_loss=0.09567, over 5278988.14 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3531, pruned_loss=0.1038, over 5701072.60 frames. ], batch size: 112, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:24:40,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-06 15:25:00,537 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=552390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:25:04,638 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 4950, giga_loss[loss=0.2491, simple_loss=0.3262, pruned_loss=0.08601, over 28783.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3563, pruned_loss=0.105, over 5717461.88 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3516, pruned_loss=0.09603, over 5291040.16 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3557, pruned_loss=0.1052, over 5703223.74 frames. ], batch size: 66, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:25:28,392 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=552422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:25:50,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5542, 1.7501, 1.8030, 1.3570], device='cuda:1'), covar=tensor([0.1682, 0.2288, 0.1367, 0.1566], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0694, 0.0882, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 15:25:54,941 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 5000, giga_loss[loss=0.2676, simple_loss=0.3346, pruned_loss=0.1003, over 28765.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3564, pruned_loss=0.1047, over 5718867.03 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3517, pruned_loss=0.09615, over 5293840.80 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3558, pruned_loss=0.1049, over 5709040.93 frames. ], batch size: 99, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:26:02,963 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=552465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:26:09,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4543, 2.1487, 1.6189, 0.6763], device='cuda:1'), covar=tensor([0.5445, 0.2457, 0.3200, 0.5305], device='cuda:1'), in_proj_covar=tensor([0.1572, 0.1479, 0.1486, 0.1291], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:26:46,233 INFO [train.py:968] (1/2) Epoch 13, batch 5050, giga_loss[loss=0.2913, simple_loss=0.3585, pruned_loss=0.1121, over 28272.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3557, pruned_loss=0.1039, over 5724938.47 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3518, pruned_loss=0.09614, over 5303712.76 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3552, pruned_loss=0.1042, over 5716060.82 frames. ], batch size: 71, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:26:48,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1226, 1.3313, 1.0074, 0.9869], device='cuda:1'), covar=tensor([0.0776, 0.0429, 0.1118, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0440, 0.0501, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 15:26:56,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2721, 1.2704, 1.2802, 1.5441], device='cuda:1'), covar=tensor([0.0741, 0.0324, 0.0318, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 15:27:00,767 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:27:27,227 INFO [train.py:968] (1/2) Epoch 13, batch 5100, giga_loss[loss=0.265, simple_loss=0.339, pruned_loss=0.09555, over 28627.00 frames. ], tot_loss[loss=0.281, simple_loss=0.355, pruned_loss=0.1035, over 5723495.73 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3519, pruned_loss=0.09614, over 5315757.93 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3546, pruned_loss=0.1039, over 5713162.98 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:28:03,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=552604.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:28:12,268 INFO [train.py:968] (1/2) Epoch 13, batch 5150, giga_loss[loss=0.2463, simple_loss=0.3289, pruned_loss=0.08182, over 28890.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3517, pruned_loss=0.1017, over 5727280.23 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3522, pruned_loss=0.09634, over 5324946.05 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3511, pruned_loss=0.1019, over 5716240.40 frames. ], batch size: 227, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:28:42,983 INFO [optim.py:369] (1/2) 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 15:28:51,006 INFO [train.py:968] (1/2) Epoch 13, batch 5200, giga_loss[loss=0.2142, simple_loss=0.3021, pruned_loss=0.06315, over 29021.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09931, over 5731720.09 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3518, pruned_loss=0.09615, over 5341183.51 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.347, pruned_loss=0.09974, over 5718962.41 frames. ], batch size: 164, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:28:59,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3510, 2.0428, 1.4592, 0.5979], device='cuda:1'), covar=tensor([0.4485, 0.2244, 0.3634, 0.5138], device='cuda:1'), in_proj_covar=tensor([0.1569, 0.1474, 0.1482, 0.1284], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:29:00,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=552675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:29:02,849 INFO [zipformer.py:1188] (1/2) 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] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 15:29:26,539 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=552707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:29:32,224 INFO [train.py:968] (1/2) Epoch 13, batch 5250, giga_loss[loss=0.304, simple_loss=0.3705, pruned_loss=0.1187, over 26531.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3471, pruned_loss=0.09914, over 5723726.37 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3522, pruned_loss=0.09647, over 5346173.79 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3465, pruned_loss=0.09922, over 5714515.92 frames. ], batch size: 555, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:30:06,807 INFO [optim.py:369] (1/2) 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 15:30:14,951 INFO [train.py:968] (1/2) Epoch 13, batch 5300, giga_loss[loss=0.2542, simple_loss=0.3448, pruned_loss=0.08178, over 29023.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.349, pruned_loss=0.09863, over 5704294.91 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3527, pruned_loss=0.09675, over 5345751.61 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3481, pruned_loss=0.09849, over 5706516.69 frames. ], batch size: 155, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:30:58,395 INFO [train.py:968] (1/2) Epoch 13, batch 5350, giga_loss[loss=0.2742, simple_loss=0.3602, pruned_loss=0.09411, over 28788.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3484, pruned_loss=0.09796, over 5701480.03 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3527, pruned_loss=0.09677, over 5353089.82 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3475, pruned_loss=0.09786, over 5701222.61 frames. ], batch size: 243, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:31:19,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=552840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:31:31,277 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 15:31:33,682 INFO [optim.py:369] (1/2) 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 15:31:40,086 INFO [train.py:968] (1/2) Epoch 13, batch 5400, giga_loss[loss=0.2697, simple_loss=0.342, pruned_loss=0.09873, over 28877.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3482, pruned_loss=0.099, over 5704673.63 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.353, pruned_loss=0.09701, over 5362312.03 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3472, pruned_loss=0.09876, over 5702281.05 frames. ], batch size: 174, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:32:01,794 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 15:32:21,282 INFO [train.py:968] (1/2) Epoch 13, batch 5450, libri_loss[loss=0.2981, simple_loss=0.3697, pruned_loss=0.1132, over 19930.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3483, pruned_loss=0.1008, over 5680207.63 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.354, pruned_loss=0.09767, over 5352570.90 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3466, pruned_loss=0.1001, over 5698276.95 frames. ], batch size: 187, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:32:39,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4630, 1.8119, 1.4511, 1.4241], device='cuda:1'), covar=tensor([0.2297, 0.2286, 0.2605, 0.2231], device='cuda:1'), in_proj_covar=tensor([0.1331, 0.0979, 0.1173, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 15:32:50,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0229, 1.3031, 1.0526, 0.1887], device='cuda:1'), covar=tensor([0.2754, 0.2329, 0.3527, 0.5308], device='cuda:1'), in_proj_covar=tensor([0.1571, 0.1474, 0.1486, 0.1290], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:32:53,821 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 5500, giga_loss[loss=0.2608, simple_loss=0.33, pruned_loss=0.09581, over 28895.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3469, pruned_loss=0.1009, over 5694480.07 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3535, pruned_loss=0.09742, over 5368235.16 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3457, pruned_loss=0.1007, over 5703699.59 frames. ], batch size: 145, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:33:13,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=552979.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:33:15,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=552983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:33:19,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=552986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:33:29,570 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 5550, giga_loss[loss=0.273, simple_loss=0.3341, pruned_loss=0.106, over 28759.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3456, pruned_loss=0.1013, over 5698020.89 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3535, pruned_loss=0.09751, over 5380399.19 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3444, pruned_loss=0.1011, over 5701670.50 frames. ], batch size: 92, lr: 2.52e-03, grad_scale: 4.0
+2023-03-06 15:33:42,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:33:59,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2301, 3.2750, 1.4086, 1.4092], device='cuda:1'), covar=tensor([0.0946, 0.0302, 0.0910, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0512, 0.0346, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 15:34:19,738 INFO [optim.py:369] (1/2) 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 15:34:28,026 INFO [train.py:968] (1/2) Epoch 13, batch 5600, giga_loss[loss=0.2741, simple_loss=0.349, pruned_loss=0.09964, over 28326.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3444, pruned_loss=0.1004, over 5705609.52 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3542, pruned_loss=0.09779, over 5388264.41 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3428, pruned_loss=0.1001, over 5705947.61 frames. ], batch size: 368, lr: 2.52e-03, grad_scale: 8.0
+2023-03-06 15:34:42,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1073, 1.2153, 3.4704, 3.0409], device='cuda:1'), covar=tensor([0.1595, 0.2607, 0.0435, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0671, 0.0598, 0.0865, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 15:35:08,265 INFO [train.py:968] (1/2) Epoch 13, batch 5650, giga_loss[loss=0.2365, simple_loss=0.3053, pruned_loss=0.0839, over 28434.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3404, pruned_loss=0.09815, over 5716291.39 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3541, pruned_loss=0.0977, over 5405071.98 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3388, pruned_loss=0.09802, over 5711311.13 frames. ], batch size: 85, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:35:15,643 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553122.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:35:17,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553125.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 15:35:18,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3316, 1.3417, 4.0241, 3.3379], device='cuda:1'), covar=tensor([0.1567, 0.2586, 0.0395, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0670, 0.0596, 0.0863, 0.0784], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 15:35:43,202 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553154.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 15:35:45,391 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 5700, giga_loss[loss=0.2288, simple_loss=0.2992, pruned_loss=0.07925, over 28372.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3352, pruned_loss=0.09564, over 5720027.33 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3537, pruned_loss=0.09753, over 5414985.32 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.334, pruned_loss=0.09565, over 5712728.74 frames. ], batch size: 65, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:36:29,479 INFO [train.py:968] (1/2) Epoch 13, batch 5750, giga_loss[loss=0.2946, simple_loss=0.3619, pruned_loss=0.1136, over 28588.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3345, pruned_loss=0.09516, over 5721559.10 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3535, pruned_loss=0.09745, over 5427935.66 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3332, pruned_loss=0.09517, over 5712149.44 frames. ], batch size: 307, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:37:04,573 INFO [optim.py:369] (1/2) 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 15:37:09,144 INFO [train.py:968] (1/2) Epoch 13, batch 5800, giga_loss[loss=0.2347, simple_loss=0.3127, pruned_loss=0.07833, over 28985.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3358, pruned_loss=0.09532, over 5722945.78 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.09749, over 5429611.18 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3344, pruned_loss=0.09528, over 5716792.34 frames. ], batch size: 106, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:37:21,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 15:37:25,581 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=553286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:37:43,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=553309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:37:46,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1982, 1.1480, 0.9763, 1.4489], device='cuda:1'), covar=tensor([0.0780, 0.0354, 0.0377, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0115, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0084, 0.0059, 0.0054, 0.0091], device='cuda:1')
+2023-03-06 15:37:47,532 INFO [train.py:968] (1/2) Epoch 13, batch 5850, libri_loss[loss=0.319, simple_loss=0.3836, pruned_loss=0.1272, over 29504.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3399, pruned_loss=0.09705, over 5726950.37 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3536, pruned_loss=0.09753, over 5447107.81 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3382, pruned_loss=0.09692, over 5717753.84 frames. ], batch size: 81, lr: 2.52e-03, grad_scale: 2.0
+2023-03-06 15:38:24,779 INFO [optim.py:369] (1/2) 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 15:38:30,099 INFO [train.py:968] (1/2) Epoch 13, batch 5900, giga_loss[loss=0.2739, simple_loss=0.3557, pruned_loss=0.09604, over 28972.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3439, pruned_loss=0.09883, over 5717881.03 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3542, pruned_loss=0.09797, over 5448572.44 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3417, pruned_loss=0.09834, over 5715918.00 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:38:36,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5031, 1.6115, 1.7222, 1.3906], device='cuda:1'), covar=tensor([0.1419, 0.1863, 0.1687, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0735, 0.0678, 0.0656], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 15:38:37,353 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=553373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:39:11,704 INFO [train.py:968] (1/2) Epoch 13, batch 5950, giga_loss[loss=0.2789, simple_loss=0.3603, pruned_loss=0.09871, over 28829.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3469, pruned_loss=0.09989, over 5711293.63 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3543, pruned_loss=0.09794, over 5457644.46 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3448, pruned_loss=0.09957, over 5712008.60 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:39:53,677 INFO [optim.py:369] (1/2) 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 15:39:59,846 INFO [train.py:968] (1/2) Epoch 13, batch 6000, giga_loss[loss=0.3103, simple_loss=0.3775, pruned_loss=0.1215, over 28986.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3489, pruned_loss=0.1013, over 5709374.37 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3541, pruned_loss=0.09797, over 5461822.53 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3473, pruned_loss=0.1011, over 5708495.34 frames. ], batch size: 128, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:39:59,846 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 15:40:08,375 INFO [train.py:1012] (1/2) Epoch 13, validation: loss=0.2192, simple_loss=0.3255, pruned_loss=0.05648, over 944034.00 frames. 
+2023-03-06 15:40:08,375 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 15:40:40,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6816, 2.5123, 1.6679, 0.9139], device='cuda:1'), covar=tensor([0.6279, 0.2776, 0.3124, 0.5307], device='cuda:1'), in_proj_covar=tensor([0.1582, 0.1492, 0.1494, 0.1301], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:40:49,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 15:40:52,490 INFO [train.py:968] (1/2) Epoch 13, batch 6050, giga_loss[loss=0.3523, simple_loss=0.4099, pruned_loss=0.1474, over 28988.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.355, pruned_loss=0.1061, over 5711035.70 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3548, pruned_loss=0.09842, over 5473454.72 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.353, pruned_loss=0.1056, over 5707507.13 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:40:54,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:40:56,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:41:21,753 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:41:30,588 INFO [optim.py:369] (1/2) 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 15:41:36,775 INFO [train.py:968] (1/2) Epoch 13, batch 6100, giga_loss[loss=0.3075, simple_loss=0.3734, pruned_loss=0.1208, over 28916.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.36, pruned_loss=0.1099, over 5714025.98 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3551, pruned_loss=0.09858, over 5491937.91 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3583, pruned_loss=0.1098, over 5703674.39 frames. ], batch size: 213, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:42:25,863 INFO [train.py:968] (1/2) Epoch 13, batch 6150, giga_loss[loss=0.315, simple_loss=0.3876, pruned_loss=0.1212, over 28956.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3667, pruned_loss=0.1154, over 5674481.37 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3551, pruned_loss=0.09863, over 5485675.82 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3655, pruned_loss=0.1156, over 5675557.27 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:42:40,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9089, 1.0731, 3.3030, 2.9138], device='cuda:1'), covar=tensor([0.1750, 0.2734, 0.0505, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0673, 0.0600, 0.0869, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 15:43:08,715 INFO [optim.py:369] (1/2) 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 15:43:15,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=553661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:43:19,561 INFO [train.py:968] (1/2) Epoch 13, batch 6200, giga_loss[loss=0.3654, simple_loss=0.4165, pruned_loss=0.1571, over 28778.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3744, pruned_loss=0.1219, over 5672519.44 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3552, pruned_loss=0.09875, over 5487764.44 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3734, pruned_loss=0.122, over 5672366.44 frames. ], batch size: 284, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:43:37,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=553684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:43:42,142 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-06 15:44:06,903 INFO [train.py:968] (1/2) Epoch 13, batch 6250, giga_loss[loss=0.3268, simple_loss=0.3859, pruned_loss=0.1338, over 28826.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3802, pruned_loss=0.1271, over 5670991.33 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3557, pruned_loss=0.0991, over 5486689.66 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3794, pruned_loss=0.1273, over 5674763.70 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:44:17,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2962, 1.4942, 1.2400, 1.0755], device='cuda:1'), covar=tensor([0.1731, 0.1596, 0.1679, 0.1530], device='cuda:1'), in_proj_covar=tensor([0.1327, 0.0979, 0.1172, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 15:44:23,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5005, 1.7042, 1.4102, 1.6213], device='cuda:1'), covar=tensor([0.2068, 0.2038, 0.2173, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.1326, 0.0978, 0.1170, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 15:44:48,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3735, 3.0572, 1.4924, 1.5017], device='cuda:1'), covar=tensor([0.0934, 0.0364, 0.0837, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0514, 0.0346, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 15:44:48,706 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 6300, giga_loss[loss=0.3325, simple_loss=0.3915, pruned_loss=0.1367, over 28948.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3842, pruned_loss=0.1304, over 5662766.12 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3559, pruned_loss=0.09929, over 5491889.94 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3841, pruned_loss=0.1313, over 5665980.27 frames. ], batch size: 106, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:45:29,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7221, 1.7762, 1.2876, 1.3787], device='cuda:1'), covar=tensor([0.0777, 0.0612, 0.1014, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0445, 0.0505, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 15:45:38,808 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:45:40,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:45:48,808 INFO [train.py:968] (1/2) Epoch 13, batch 6350, giga_loss[loss=0.3881, simple_loss=0.4333, pruned_loss=0.1715, over 28613.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3867, pruned_loss=0.1337, over 5645118.97 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3557, pruned_loss=0.0992, over 5498113.37 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3872, pruned_loss=0.1349, over 5644200.18 frames. ], batch size: 336, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:45:51,182 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=553817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:03,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=553827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:05,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=553830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:10,811 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:32,813 INFO [optim.py:369] (1/2) 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 15:46:35,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=553859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:46:43,303 INFO [train.py:968] (1/2) Epoch 13, batch 6400, giga_loss[loss=0.4619, simple_loss=0.4612, pruned_loss=0.2313, over 23510.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3894, pruned_loss=0.1375, over 5631907.54 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3554, pruned_loss=0.09906, over 5508381.50 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3908, pruned_loss=0.1393, over 5625473.34 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 15:47:04,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2317, 1.5029, 1.4126, 1.1635], device='cuda:1'), covar=tensor([0.1802, 0.1790, 0.1115, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.1753, 0.1667, 0.1639, 0.1727], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 15:47:35,254 INFO [train.py:968] (1/2) Epoch 13, batch 6450, giga_loss[loss=0.3218, simple_loss=0.3776, pruned_loss=0.133, over 28917.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3932, pruned_loss=0.1419, over 5624857.49 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.355, pruned_loss=0.09881, over 5516145.87 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3954, pruned_loss=0.1444, over 5615006.93 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:47:54,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9821, 1.3422, 1.0877, 0.2355], device='cuda:1'), covar=tensor([0.3061, 0.2577, 0.3684, 0.4871], device='cuda:1'), in_proj_covar=tensor([0.1601, 0.1510, 0.1506, 0.1313], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:48:16,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-06 15:48:23,965 INFO [optim.py:369] (1/2) 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 15:48:26,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.10 vs. limit=5.0
+2023-03-06 15:48:30,506 INFO [train.py:968] (1/2) Epoch 13, batch 6500, giga_loss[loss=0.3918, simple_loss=0.43, pruned_loss=0.1768, over 28892.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3969, pruned_loss=0.1449, over 5618521.84 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3546, pruned_loss=0.09873, over 5522134.29 frames. ], giga_tot_loss[loss=0.3474, simple_loss=0.3995, pruned_loss=0.1476, over 5606548.94 frames. ], batch size: 285, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:49:19,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9439, 1.1931, 2.8208, 2.6214], device='cuda:1'), covar=tensor([0.1413, 0.2181, 0.0574, 0.1798], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0603, 0.0873, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 15:49:19,839 INFO [train.py:968] (1/2) Epoch 13, batch 6550, giga_loss[loss=0.5163, simple_loss=0.5047, pruned_loss=0.2639, over 26528.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3975, pruned_loss=0.1461, over 5629388.66 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3547, pruned_loss=0.09874, over 5525878.00 frames. ], giga_tot_loss[loss=0.349, simple_loss=0.4001, pruned_loss=0.1489, over 5617743.88 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:50:06,573 INFO [optim.py:369] (1/2) 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 15:50:11,394 INFO [train.py:968] (1/2) Epoch 13, batch 6600, giga_loss[loss=0.3002, simple_loss=0.3704, pruned_loss=0.115, over 28843.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3934, pruned_loss=0.143, over 5640437.04 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3549, pruned_loss=0.0989, over 5536737.52 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3965, pruned_loss=0.1464, over 5624403.62 frames. ], batch size: 174, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:51:03,623 INFO [train.py:968] (1/2) Epoch 13, batch 6650, giga_loss[loss=0.3644, simple_loss=0.4123, pruned_loss=0.1583, over 28550.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3924, pruned_loss=0.1414, over 5633664.76 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3549, pruned_loss=0.09894, over 5539193.59 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.396, pruned_loss=0.1454, over 5622092.69 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:51:45,743 INFO [optim.py:369] (1/2) 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 15:51:51,259 INFO [train.py:968] (1/2) Epoch 13, batch 6700, giga_loss[loss=0.3071, simple_loss=0.3767, pruned_loss=0.1187, over 28834.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3916, pruned_loss=0.1397, over 5645887.56 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3543, pruned_loss=0.09868, over 5548256.59 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3959, pruned_loss=0.1441, over 5630755.13 frames. ], batch size: 284, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:52:19,539 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 6750, giga_loss[loss=0.3371, simple_loss=0.4001, pruned_loss=0.1371, over 28734.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3932, pruned_loss=0.1409, over 5612970.74 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3542, pruned_loss=0.09861, over 5543126.80 frames. ], giga_tot_loss[loss=0.3439, simple_loss=0.3974, pruned_loss=0.1452, over 5607795.93 frames. ], batch size: 262, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:53:25,472 INFO [optim.py:369] (1/2) 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 15:53:28,901 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:53:30,033 INFO [train.py:968] (1/2) Epoch 13, batch 6800, giga_loss[loss=0.4384, simple_loss=0.4566, pruned_loss=0.21, over 26355.00 frames. ], tot_loss[loss=0.3348, simple_loss=0.3914, pruned_loss=0.1391, over 5608571.55 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3544, pruned_loss=0.09886, over 5541514.75 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3961, pruned_loss=0.144, over 5609366.40 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:54:19,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-06 15:54:25,723 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:54:26,829 INFO [train.py:968] (1/2) Epoch 13, batch 6850, libri_loss[loss=0.2834, simple_loss=0.3637, pruned_loss=0.1016, over 29525.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3885, pruned_loss=0.1358, over 5610223.96 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3545, pruned_loss=0.09887, over 5543332.18 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3923, pruned_loss=0.1398, over 5609364.42 frames. ], batch size: 81, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:54:44,609 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:54:46,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:55:06,147 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 6900, giga_loss[loss=0.2834, simple_loss=0.3594, pruned_loss=0.1037, over 28919.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.385, pruned_loss=0.1317, over 5632661.17 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3548, pruned_loss=0.09924, over 5553946.83 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.389, pruned_loss=0.1359, over 5624975.62 frames. ], batch size: 106, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:55:14,293 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=554367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:56:01,904 INFO [train.py:968] (1/2) Epoch 13, batch 6950, giga_loss[loss=0.2904, simple_loss=0.3606, pruned_loss=0.1101, over 28895.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3823, pruned_loss=0.1294, over 5642140.21 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3548, pruned_loss=0.0992, over 5558912.17 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3859, pruned_loss=0.1332, over 5632799.55 frames. ], batch size: 145, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 15:56:48,952 INFO [optim.py:369] (1/2) 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 15:56:52,493 INFO [train.py:968] (1/2) Epoch 13, batch 7000, giga_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1226, over 27882.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3796, pruned_loss=0.1274, over 5647372.72 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3545, pruned_loss=0.099, over 5564243.29 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.383, pruned_loss=0.131, over 5636325.16 frames. ], batch size: 412, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:57:00,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 15:57:03,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8451, 1.8604, 1.3755, 1.6125], device='cuda:1'), covar=tensor([0.0839, 0.0716, 0.1014, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0445, 0.0505, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 15:57:41,696 INFO [train.py:968] (1/2) Epoch 13, batch 7050, giga_loss[loss=0.3336, simple_loss=0.3913, pruned_loss=0.1379, over 28969.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3801, pruned_loss=0.1278, over 5653323.79 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3548, pruned_loss=0.09917, over 5564627.08 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3829, pruned_loss=0.1309, over 5645602.57 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:58:16,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0999, 1.1984, 3.5988, 3.1350], device='cuda:1'), covar=tensor([0.1700, 0.2655, 0.0465, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0603, 0.0873, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 15:58:36,096 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 7100, giga_loss[loss=0.3103, simple_loss=0.3793, pruned_loss=0.1206, over 28889.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3794, pruned_loss=0.127, over 5662040.58 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3543, pruned_loss=0.0989, over 5569845.83 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3824, pruned_loss=0.1301, over 5652240.83 frames. ], batch size: 145, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:58:47,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8231, 2.8768, 1.6849, 0.9786], device='cuda:1'), covar=tensor([0.6557, 0.2510, 0.3883, 0.5636], device='cuda:1'), in_proj_covar=tensor([0.1589, 0.1501, 0.1496, 0.1303], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 15:59:34,349 INFO [train.py:968] (1/2) Epoch 13, batch 7150, giga_loss[loss=0.2887, simple_loss=0.3643, pruned_loss=0.1066, over 28640.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3774, pruned_loss=0.125, over 5668511.36 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3538, pruned_loss=0.09862, over 5574538.77 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3806, pruned_loss=0.1281, over 5657838.84 frames. ], batch size: 85, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 15:59:37,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:00:03,729 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:00:29,683 INFO [optim.py:369] (1/2) 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 16:00:36,391 INFO [train.py:968] (1/2) Epoch 13, batch 7200, giga_loss[loss=0.2932, simple_loss=0.3675, pruned_loss=0.1094, over 28835.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3787, pruned_loss=0.1237, over 5661456.48 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3539, pruned_loss=0.09867, over 5576206.08 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3813, pruned_loss=0.1262, over 5652047.26 frames. ], batch size: 99, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:00:59,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:01:24,744 INFO [train.py:968] (1/2) Epoch 13, batch 7250, giga_loss[loss=0.4163, simple_loss=0.4455, pruned_loss=0.1935, over 26689.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3804, pruned_loss=0.1246, over 5660819.65 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3538, pruned_loss=0.09859, over 5583180.83 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3831, pruned_loss=0.1273, over 5648831.03 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:01:39,746 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 13, batch 7300, giga_loss[loss=0.3209, simple_loss=0.3817, pruned_loss=0.1301, over 28918.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3796, pruned_loss=0.1243, over 5679960.69 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3534, pruned_loss=0.09846, over 5590826.81 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3828, pruned_loss=0.1272, over 5665788.71 frames. ], batch size: 213, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:02:29,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:02:32,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:02:47,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1035, 3.9282, 3.7399, 1.7267], device='cuda:1'), covar=tensor([0.0644, 0.0759, 0.0727, 0.2232], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.1014, 0.0882, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 16:02:58,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=554813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:02:59,988 INFO [train.py:968] (1/2) Epoch 13, batch 7350, giga_loss[loss=0.309, simple_loss=0.3657, pruned_loss=0.1261, over 27565.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3793, pruned_loss=0.1248, over 5676666.80 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3538, pruned_loss=0.09867, over 5594044.26 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.382, pruned_loss=0.1274, over 5664348.83 frames. ], batch size: 472, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:03:22,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:26,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:32,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:35,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:50,059 INFO [optim.py:369] (1/2) 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 16:03:51,593 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=554863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:03:52,865 INFO [train.py:968] (1/2) Epoch 13, batch 7400, giga_loss[loss=0.3215, simple_loss=0.3771, pruned_loss=0.133, over 27958.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3781, pruned_loss=0.1255, over 5666238.91 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3538, pruned_loss=0.0986, over 5597945.92 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3808, pruned_loss=0.1282, over 5654597.41 frames. ], batch size: 412, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:04:15,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-06 16:04:39,539 INFO [train.py:968] (1/2) Epoch 13, batch 7450, giga_loss[loss=0.3491, simple_loss=0.4012, pruned_loss=0.1485, over 28568.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3763, pruned_loss=0.1243, over 5680267.97 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3539, pruned_loss=0.09876, over 5600638.44 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3785, pruned_loss=0.1266, over 5669360.83 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:04:53,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=554927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:05:16,231 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 16:05:28,974 INFO [optim.py:369] (1/2) 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 16:05:34,388 INFO [train.py:968] (1/2) Epoch 13, batch 7500, giga_loss[loss=0.3305, simple_loss=0.3958, pruned_loss=0.1326, over 28950.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3761, pruned_loss=0.1229, over 5693368.91 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3538, pruned_loss=0.0987, over 5606495.37 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3784, pruned_loss=0.1252, over 5681037.77 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:05:58,686 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=554989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:06:00,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=554992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:06:01,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=554993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:06:22,420 INFO [train.py:968] (1/2) Epoch 13, batch 7550, giga_loss[loss=0.3035, simple_loss=0.3758, pruned_loss=0.1156, over 28739.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3766, pruned_loss=0.1224, over 5697861.98 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.354, pruned_loss=0.09887, over 5608885.91 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3786, pruned_loss=0.1245, over 5687446.64 frames. ], batch size: 262, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:06:30,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:07:05,484 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 7600, giga_loss[loss=0.2818, simple_loss=0.3566, pruned_loss=0.1035, over 28224.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3756, pruned_loss=0.122, over 5694178.80 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3545, pruned_loss=0.09924, over 5613737.85 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3772, pruned_loss=0.1237, over 5683140.45 frames. ], batch size: 77, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:07:39,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-06 16:07:46,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=555103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:07:56,817 INFO [train.py:968] (1/2) Epoch 13, batch 7650, giga_loss[loss=0.3435, simple_loss=0.3914, pruned_loss=0.1478, over 28265.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.374, pruned_loss=0.1211, over 5700397.34 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.099, over 5622305.66 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.376, pruned_loss=0.1234, over 5686180.28 frames. ], batch size: 65, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:08:17,093 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:08:19,860 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 13, batch 7700, giga_loss[loss=0.2934, simple_loss=0.3642, pruned_loss=0.1113, over 28870.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3744, pruned_loss=0.1223, over 5689350.86 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3546, pruned_loss=0.09901, over 5626196.68 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3761, pruned_loss=0.1245, over 5675868.34 frames. ], batch size: 174, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:08:54,442 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:08:55,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4223, 1.7467, 1.3600, 1.5998], device='cuda:1'), covar=tensor([0.2458, 0.2439, 0.2739, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1337, 0.0989, 0.1176, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 16:09:41,734 INFO [train.py:968] (1/2) Epoch 13, batch 7750, giga_loss[loss=0.2715, simple_loss=0.3402, pruned_loss=0.1014, over 29166.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3737, pruned_loss=0.1226, over 5698137.94 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3547, pruned_loss=0.09907, over 5631886.36 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3753, pruned_loss=0.1247, over 5683313.60 frames. ], batch size: 113, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:09:43,365 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=555216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:14,136 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:18,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=555249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:29,004 INFO [optim.py:369] (1/2) 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 16:10:33,790 INFO [train.py:968] (1/2) Epoch 13, batch 7800, giga_loss[loss=0.4634, simple_loss=0.4656, pruned_loss=0.2306, over 26716.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3716, pruned_loss=0.1215, over 5701441.78 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3547, pruned_loss=0.09895, over 5636112.73 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1236, over 5686802.42 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:10:44,904 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:10:50,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-06 16:11:09,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=555302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:11:20,961 INFO [train.py:968] (1/2) Epoch 13, batch 7850, giga_loss[loss=0.2839, simple_loss=0.3571, pruned_loss=0.1053, over 28602.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3698, pruned_loss=0.121, over 5705430.60 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3543, pruned_loss=0.09878, over 5641087.51 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1233, over 5690573.08 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:11:23,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 16:12:04,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:12:07,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=555362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:12:07,918 INFO [optim.py:369] (1/2) 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 16:12:09,318 INFO [train.py:968] (1/2) Epoch 13, batch 7900, libri_loss[loss=0.2634, simple_loss=0.3418, pruned_loss=0.09247, over 29545.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.121, over 5707283.80 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3544, pruned_loss=0.0988, over 5645341.69 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5692469.63 frames. ], batch size: 76, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:12:36,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:12:57,152 INFO [train.py:968] (1/2) Epoch 13, batch 7950, giga_loss[loss=0.3274, simple_loss=0.3948, pruned_loss=0.13, over 28915.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3695, pruned_loss=0.1208, over 5690373.60 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3541, pruned_loss=0.09863, over 5644315.39 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3716, pruned_loss=0.1234, over 5681283.94 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:13:12,794 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-06 16:13:27,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=555445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:13:31,501 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=555448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:13:45,441 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 8000, giga_loss[loss=0.2941, simple_loss=0.367, pruned_loss=0.1106, over 28954.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3699, pruned_loss=0.1202, over 5684796.68 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3541, pruned_loss=0.09863, over 5646942.64 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1224, over 5675724.70 frames. ], batch size: 106, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:13:58,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=555477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:14:34,342 INFO [train.py:968] (1/2) Epoch 13, batch 8050, giga_loss[loss=0.4089, simple_loss=0.447, pruned_loss=0.1854, over 28756.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3709, pruned_loss=0.1207, over 5676376.90 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3535, pruned_loss=0.09835, over 5649374.29 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.373, pruned_loss=0.123, over 5667437.69 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:15:00,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 16:15:19,743 INFO [optim.py:369] (1/2) 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 16:15:21,334 INFO [train.py:968] (1/2) Epoch 13, batch 8100, libri_loss[loss=0.2646, simple_loss=0.3379, pruned_loss=0.09569, over 29557.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3706, pruned_loss=0.1206, over 5687780.33 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3534, pruned_loss=0.09854, over 5657926.75 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3731, pruned_loss=0.1232, over 5673657.59 frames. ], batch size: 78, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:16:12,736 INFO [train.py:968] (1/2) Epoch 13, batch 8150, giga_loss[loss=0.3221, simple_loss=0.3763, pruned_loss=0.134, over 28865.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3742, pruned_loss=0.1238, over 5677957.74 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3536, pruned_loss=0.09879, over 5652423.63 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3764, pruned_loss=0.126, over 5672858.17 frames. ], batch size: 112, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:17:03,126 INFO [optim.py:369] (1/2) 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 16:17:05,868 INFO [train.py:968] (1/2) Epoch 13, batch 8200, giga_loss[loss=0.4284, simple_loss=0.4423, pruned_loss=0.2073, over 26558.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1258, over 5676547.26 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.354, pruned_loss=0.09907, over 5649023.49 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3774, pruned_loss=0.1279, over 5676202.15 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:17:37,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-06 16:17:45,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5468, 1.9528, 1.5764, 1.5537], device='cuda:1'), covar=tensor([0.0738, 0.0277, 0.0298, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 16:17:49,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4394, 1.9946, 1.4535, 0.6462], device='cuda:1'), covar=tensor([0.3599, 0.1850, 0.2456, 0.4360], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1514, 0.1503, 0.1306], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 16:17:54,056 INFO [train.py:968] (1/2) Epoch 13, batch 8250, giga_loss[loss=0.2688, simple_loss=0.3413, pruned_loss=0.09813, over 28457.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3768, pruned_loss=0.1275, over 5672728.33 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3537, pruned_loss=0.09883, over 5658492.61 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3794, pruned_loss=0.1306, over 5664413.98 frames. ], batch size: 65, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:18:34,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2303, 2.3758, 1.3220, 1.2897], device='cuda:1'), covar=tensor([0.0912, 0.0414, 0.0816, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0521, 0.0348, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 16:18:41,333 INFO [optim.py:369] (1/2) 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 16:18:42,668 INFO [train.py:968] (1/2) Epoch 13, batch 8300, giga_loss[loss=0.3078, simple_loss=0.3612, pruned_loss=0.1272, over 28760.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3775, pruned_loss=0.1287, over 5663914.45 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3536, pruned_loss=0.09871, over 5655298.14 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3802, pruned_loss=0.1319, over 5660375.04 frames. ], batch size: 85, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:19:30,541 INFO [train.py:968] (1/2) Epoch 13, batch 8350, giga_loss[loss=0.3813, simple_loss=0.4131, pruned_loss=0.1747, over 28704.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3765, pruned_loss=0.1284, over 5664318.71 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3535, pruned_loss=0.09861, over 5658679.67 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3794, pruned_loss=0.1318, over 5658492.82 frames. ], batch size: 284, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:20:02,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3071, 1.0871, 3.8532, 3.2541], device='cuda:1'), covar=tensor([0.1677, 0.2801, 0.0493, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0683, 0.0605, 0.0878, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 16:20:12,817 INFO [optim.py:369] (1/2) 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 16:20:14,146 INFO [train.py:968] (1/2) Epoch 13, batch 8400, giga_loss[loss=0.3038, simple_loss=0.3744, pruned_loss=0.1166, over 28990.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3758, pruned_loss=0.1265, over 5672790.25 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3532, pruned_loss=0.09833, over 5663124.46 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3788, pruned_loss=0.1301, over 5664285.61 frames. ], batch size: 128, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:20:48,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-06 16:20:52,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 16:21:00,703 INFO [train.py:968] (1/2) Epoch 13, batch 8450, giga_loss[loss=0.2588, simple_loss=0.3073, pruned_loss=0.1051, over 23843.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3729, pruned_loss=0.1239, over 5666864.97 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3528, pruned_loss=0.09809, over 5669539.96 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3764, pruned_loss=0.1277, over 5654649.43 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:21:23,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 16:21:24,311 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=555942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:21:42,488 INFO [optim.py:369] (1/2) 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 16:21:43,467 INFO [train.py:968] (1/2) Epoch 13, batch 8500, giga_loss[loss=0.277, simple_loss=0.3489, pruned_loss=0.1025, over 29094.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3723, pruned_loss=0.1234, over 5681102.56 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3529, pruned_loss=0.09811, over 5674655.96 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3753, pruned_loss=0.127, over 5666944.69 frames. ], batch size: 155, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:22:12,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-06 16:22:25,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-06 16:22:33,847 INFO [train.py:968] (1/2) Epoch 13, batch 8550, giga_loss[loss=0.3286, simple_loss=0.3811, pruned_loss=0.1381, over 28958.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3701, pruned_loss=0.1226, over 5683855.63 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.353, pruned_loss=0.09817, over 5678907.68 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3727, pruned_loss=0.1259, over 5668856.05 frames. ], batch size: 106, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:22:36,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3777, 1.5034, 1.4080, 1.3312], device='cuda:1'), covar=tensor([0.1971, 0.1738, 0.1345, 0.1566], device='cuda:1'), in_proj_covar=tensor([0.1754, 0.1672, 0.1642, 0.1733], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 16:23:21,172 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 8600, giga_loss[loss=0.3562, simple_loss=0.3861, pruned_loss=0.1632, over 23715.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3695, pruned_loss=0.1224, over 5670414.08 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3532, pruned_loss=0.09824, over 5680651.45 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.372, pruned_loss=0.1257, over 5656186.28 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:24:07,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-06 16:24:17,041 INFO [train.py:968] (1/2) Epoch 13, batch 8650, giga_loss[loss=0.424, simple_loss=0.4595, pruned_loss=0.1943, over 28898.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.372, pruned_loss=0.1235, over 5670383.11 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3529, pruned_loss=0.09788, over 5685442.70 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3746, pruned_loss=0.127, over 5654374.21 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:24:20,214 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:25:02,601 INFO [train.py:968] (1/2) Epoch 13, batch 8700, giga_loss[loss=0.3069, simple_loss=0.3909, pruned_loss=0.1114, over 28966.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3758, pruned_loss=0.1237, over 5648832.73 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3538, pruned_loss=0.09852, over 5660732.53 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3777, pruned_loss=0.1266, over 5656911.26 frames. ], batch size: 164, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:25:51,033 INFO [train.py:968] (1/2) Epoch 13, batch 8750, giga_loss[loss=0.3292, simple_loss=0.3937, pruned_loss=0.1323, over 28733.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3771, pruned_loss=0.1229, over 5650234.50 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3536, pruned_loss=0.09854, over 5647615.52 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3792, pruned_loss=0.1257, over 5668029.58 frames. ], batch size: 284, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:26:36,993 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 8800, giga_loss[loss=0.31, simple_loss=0.3783, pruned_loss=0.1208, over 28944.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3799, pruned_loss=0.1255, over 5648342.56 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3536, pruned_loss=0.0986, over 5643445.94 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3819, pruned_loss=0.1281, over 5666068.65 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:27:24,125 INFO [train.py:968] (1/2) Epoch 13, batch 8850, giga_loss[loss=0.2889, simple_loss=0.3579, pruned_loss=0.1099, over 28908.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3809, pruned_loss=0.1268, over 5648615.33 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3533, pruned_loss=0.09836, over 5649910.09 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3835, pruned_loss=0.1298, over 5656706.81 frames. ], batch size: 227, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:27:25,672 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:28:11,108 INFO [train.py:968] (1/2) Epoch 13, batch 8900, giga_loss[loss=0.3263, simple_loss=0.3871, pruned_loss=0.1327, over 28600.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3805, pruned_loss=0.1276, over 5648253.33 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3529, pruned_loss=0.09818, over 5655340.23 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3834, pruned_loss=0.1306, over 5650040.35 frames. ], batch size: 307, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:29:06,681 INFO [train.py:968] (1/2) Epoch 13, batch 8950, giga_loss[loss=0.2865, simple_loss=0.355, pruned_loss=0.109, over 28960.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3796, pruned_loss=0.1282, over 5625116.41 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.353, pruned_loss=0.09824, over 5645637.15 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3822, pruned_loss=0.1309, over 5634683.32 frames. ], batch size: 213, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:29:31,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8026, 2.3382, 2.0994, 1.5906], device='cuda:1'), covar=tensor([0.1757, 0.2201, 0.1462, 0.1718], device='cuda:1'), in_proj_covar=tensor([0.0844, 0.0699, 0.0885, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 16:29:43,395 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=556453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:29:51,184 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=556460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:29:53,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=556463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:29:55,869 INFO [optim.py:369] (1/2) 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 16:29:55,882 INFO [train.py:968] (1/2) Epoch 13, batch 9000, giga_loss[loss=0.3213, simple_loss=0.3725, pruned_loss=0.1351, over 28671.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3776, pruned_loss=0.1274, over 5643006.68 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.353, pruned_loss=0.09826, over 5649517.73 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.38, pruned_loss=0.1299, over 5647095.00 frames. ], batch size: 92, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:29:55,882 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 16:30:04,609 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 16:30:10,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3012, 2.9920, 1.4249, 1.3489], device='cuda:1'), covar=tensor([0.0929, 0.0359, 0.0830, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0523, 0.0348, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 16:30:30,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=556492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:30:30,647 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 9050, giga_loss[loss=0.2852, simple_loss=0.3548, pruned_loss=0.1078, over 28825.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3774, pruned_loss=0.1279, over 5649520.12 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3532, pruned_loss=0.09841, over 5654510.63 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3795, pruned_loss=0.1303, over 5648268.97 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:31:07,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0984, 1.2485, 3.3869, 3.0070], device='cuda:1'), covar=tensor([0.1583, 0.2417, 0.0480, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0682, 0.0602, 0.0875, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 16:31:36,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 16:31:45,863 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 9100, giga_loss[loss=0.3335, simple_loss=0.3671, pruned_loss=0.1499, over 23416.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5638398.78 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3537, pruned_loss=0.09876, over 5655718.46 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3798, pruned_loss=0.131, over 5636468.41 frames. ], batch size: 705, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:32:35,676 INFO [train.py:968] (1/2) Epoch 13, batch 9150, giga_loss[loss=0.2875, simple_loss=0.3429, pruned_loss=0.1161, over 28764.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3761, pruned_loss=0.1276, over 5653728.29 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3535, pruned_loss=0.09851, over 5663194.66 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3783, pruned_loss=0.1302, over 5645057.33 frames. ], batch size: 92, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:32:56,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=556636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:33:01,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=556639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:33:26,944 INFO [optim.py:369] (1/2) 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 16:33:26,957 INFO [train.py:968] (1/2) Epoch 13, batch 9200, giga_loss[loss=0.3318, simple_loss=0.3909, pruned_loss=0.1363, over 28668.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3736, pruned_loss=0.1265, over 5657162.15 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3534, pruned_loss=0.09844, over 5664477.69 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3754, pruned_loss=0.1288, over 5649329.60 frames. ], batch size: 262, lr: 2.51e-03, grad_scale: 8.0
+2023-03-06 16:33:31,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=556668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:33:36,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3880, 1.4648, 1.1475, 1.5331], device='cuda:1'), covar=tensor([0.0734, 0.0340, 0.0336, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0115, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 16:33:44,228 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-06 16:34:10,568 INFO [train.py:968] (1/2) Epoch 13, batch 9250, giga_loss[loss=0.2595, simple_loss=0.3381, pruned_loss=0.09046, over 28402.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.373, pruned_loss=0.1252, over 5653592.87 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3536, pruned_loss=0.09864, over 5667109.69 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3752, pruned_loss=0.128, over 5644560.31 frames. ], batch size: 71, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:34:26,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=556731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:35:00,394 INFO [train.py:968] (1/2) Epoch 13, batch 9300, giga_loss[loss=0.2941, simple_loss=0.3627, pruned_loss=0.1128, over 28808.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3752, pruned_loss=0.1256, over 5663193.41 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3541, pruned_loss=0.09882, over 5672523.95 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3771, pruned_loss=0.1284, over 5650888.64 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:35:01,099 INFO [optim.py:369] (1/2) 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 16:35:29,893 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=556796.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 16:35:48,486 INFO [train.py:968] (1/2) Epoch 13, batch 9350, giga_loss[loss=0.3821, simple_loss=0.4169, pruned_loss=0.1737, over 26941.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3758, pruned_loss=0.1262, over 5658653.04 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3539, pruned_loss=0.09869, over 5680906.66 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3782, pruned_loss=0.1295, over 5640885.51 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:35:59,192 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=556828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:36:34,867 INFO [train.py:968] (1/2) Epoch 13, batch 9400, giga_loss[loss=0.3082, simple_loss=0.3743, pruned_loss=0.1211, over 28961.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3767, pruned_loss=0.1266, over 5668525.09 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3542, pruned_loss=0.0988, over 5686798.70 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3789, pruned_loss=0.1298, over 5648592.26 frames. ], batch size: 112, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:36:35,845 INFO [optim.py:369] (1/2) 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 16:37:21,491 INFO [train.py:968] (1/2) Epoch 13, batch 9450, giga_loss[loss=0.3095, simple_loss=0.3786, pruned_loss=0.1203, over 28848.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3772, pruned_loss=0.1242, over 5668255.29 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3543, pruned_loss=0.0989, over 5688588.93 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3793, pruned_loss=0.1272, over 5650366.99 frames. ], batch size: 119, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:37:38,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5135, 4.0357, 1.5639, 1.6932], device='cuda:1'), covar=tensor([0.0900, 0.0411, 0.0933, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0521, 0.0348, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 16:38:06,019 INFO [train.py:968] (1/2) Epoch 13, batch 9500, giga_loss[loss=0.3295, simple_loss=0.3965, pruned_loss=0.1312, over 28883.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3788, pruned_loss=0.1236, over 5681518.80 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.354, pruned_loss=0.09877, over 5694692.82 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3815, pruned_loss=0.1268, over 5661383.30 frames. ], batch size: 99, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:38:07,432 INFO [optim.py:369] (1/2) 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 16:38:11,194 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=556971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:38:13,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=556974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:38:42,880 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 9550, giga_loss[loss=0.3818, simple_loss=0.4245, pruned_loss=0.1696, over 27908.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3824, pruned_loss=0.1255, over 5686349.50 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3541, pruned_loss=0.0988, over 5698517.74 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3849, pruned_loss=0.1285, over 5666579.06 frames. ], batch size: 412, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:39:33,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-06 16:39:41,914 INFO [train.py:968] (1/2) Epoch 13, batch 9600, giga_loss[loss=0.3274, simple_loss=0.3991, pruned_loss=0.1279, over 28860.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3846, pruned_loss=0.1281, over 5679824.34 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3541, pruned_loss=0.09872, over 5691964.90 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3872, pruned_loss=0.1311, over 5668937.77 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:39:44,106 INFO [optim.py:369] (1/2) 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 16:39:47,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8994, 2.8133, 1.7417, 0.9170], device='cuda:1'), covar=tensor([0.5372, 0.2528, 0.3036, 0.5267], device='cuda:1'), in_proj_covar=tensor([0.1587, 0.1510, 0.1494, 0.1307], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 16:40:22,050 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 9650, giga_loss[loss=0.2825, simple_loss=0.3636, pruned_loss=0.1007, over 29007.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3871, pruned_loss=0.1315, over 5663449.98 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3543, pruned_loss=0.09893, over 5693904.82 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3897, pruned_loss=0.1345, over 5652621.30 frames. ], batch size: 164, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:41:16,003 INFO [train.py:968] (1/2) Epoch 13, batch 9700, giga_loss[loss=0.2973, simple_loss=0.3766, pruned_loss=0.109, over 28944.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3858, pruned_loss=0.1302, over 5673847.69 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3543, pruned_loss=0.0987, over 5700424.87 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.389, pruned_loss=0.1339, over 5658137.01 frames. ], batch size: 136, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:41:18,945 INFO [optim.py:369] (1/2) 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 16:41:20,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=557171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 16:41:49,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2121, 1.2653, 1.0705, 0.9273], device='cuda:1'), covar=tensor([0.0883, 0.0550, 0.1112, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0446, 0.0509, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 16:42:00,517 INFO [train.py:968] (1/2) Epoch 13, batch 9750, giga_loss[loss=0.3605, simple_loss=0.4043, pruned_loss=0.1583, over 26594.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3839, pruned_loss=0.1281, over 5678390.14 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3544, pruned_loss=0.09869, over 5705110.91 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.387, pruned_loss=0.1317, over 5661330.95 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:42:15,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0834, 2.1165, 1.9732, 1.7422], device='cuda:1'), covar=tensor([0.1543, 0.2215, 0.2029, 0.2362], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0732, 0.0675, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 16:42:31,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=557249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:42:33,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=557252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:42:44,365 INFO [train.py:968] (1/2) Epoch 13, batch 9800, giga_loss[loss=0.2742, simple_loss=0.3608, pruned_loss=0.0938, over 28398.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3838, pruned_loss=0.1262, over 5683151.61 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3542, pruned_loss=0.09869, over 5708946.40 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3872, pruned_loss=0.1299, over 5665465.01 frames. ], batch size: 60, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:42:46,391 INFO [optim.py:369] (1/2) 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 16:42:55,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4792, 1.6508, 1.3529, 1.6017], device='cuda:1'), covar=tensor([0.2491, 0.2569, 0.2797, 0.2312], device='cuda:1'), in_proj_covar=tensor([0.1343, 0.0996, 0.1186, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 16:43:01,506 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=557281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:43:30,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=557314.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 16:43:30,409 INFO [train.py:968] (1/2) Epoch 13, batch 9850, giga_loss[loss=0.3015, simple_loss=0.3794, pruned_loss=0.1118, over 28867.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3834, pruned_loss=0.1255, over 5687001.61 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3541, pruned_loss=0.09858, over 5713761.40 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.387, pruned_loss=0.1292, over 5667796.46 frames. ], batch size: 145, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:43:31,949 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=557317.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 16:43:56,671 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=557338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:44:04,118 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=557346.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 16:44:22,133 INFO [train.py:968] (1/2) Epoch 13, batch 9900, giga_loss[loss=0.3809, simple_loss=0.4178, pruned_loss=0.172, over 26600.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3852, pruned_loss=0.1277, over 5679488.47 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3539, pruned_loss=0.09846, over 5716920.45 frames. ], giga_tot_loss[loss=0.3259, simple_loss=0.389, pruned_loss=0.1314, over 5660783.56 frames. ], batch size: 555, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:44:26,134 INFO [optim.py:369] (1/2) 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 16:44:37,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.78 vs. limit=5.0
+2023-03-06 16:45:14,707 INFO [train.py:968] (1/2) Epoch 13, batch 9950, libri_loss[loss=0.2583, simple_loss=0.3409, pruned_loss=0.08779, over 29541.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3839, pruned_loss=0.1277, over 5669155.02 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3536, pruned_loss=0.09826, over 5719555.46 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3876, pruned_loss=0.1312, over 5651368.15 frames. ], batch size: 79, lr: 2.51e-03, grad_scale: 2.0
+2023-03-06 16:46:07,219 INFO [train.py:968] (1/2) Epoch 13, batch 10000, giga_loss[loss=0.409, simple_loss=0.4423, pruned_loss=0.1878, over 27581.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3831, pruned_loss=0.1291, over 5658091.69 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3534, pruned_loss=0.09807, over 5721284.11 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3865, pruned_loss=0.1324, over 5642094.32 frames. ], batch size: 472, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:46:09,278 INFO [optim.py:369] (1/2) 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 16:46:52,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-06 16:46:54,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 16:46:56,946 INFO [train.py:968] (1/2) Epoch 13, batch 10050, giga_loss[loss=0.277, simple_loss=0.3541, pruned_loss=0.09996, over 28917.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3799, pruned_loss=0.1272, over 5671175.28 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3535, pruned_loss=0.09814, over 5724288.45 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3829, pruned_loss=0.1302, over 5654853.69 frames. ], batch size: 174, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:47:37,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4786, 1.6900, 1.3882, 1.7911], device='cuda:1'), covar=tensor([0.2191, 0.2211, 0.2308, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.1338, 0.0993, 0.1181, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 16:47:47,100 INFO [train.py:968] (1/2) Epoch 13, batch 10100, giga_loss[loss=0.294, simple_loss=0.3569, pruned_loss=0.1156, over 28704.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3767, pruned_loss=0.1258, over 5662805.33 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3533, pruned_loss=0.09795, over 5727670.82 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3798, pruned_loss=0.129, over 5645505.95 frames. ], batch size: 262, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:47:51,745 INFO [optim.py:369] (1/2) 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 16:48:03,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2116, 4.0414, 3.8335, 1.7325], device='cuda:1'), covar=tensor([0.0526, 0.0662, 0.0672, 0.2179], device='cuda:1'), in_proj_covar=tensor([0.1097, 0.1022, 0.0889, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 16:48:34,249 INFO [train.py:968] (1/2) Epoch 13, batch 10150, giga_loss[loss=0.318, simple_loss=0.3847, pruned_loss=0.1256, over 28927.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3763, pruned_loss=0.1264, over 5668217.11 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.353, pruned_loss=0.09778, over 5731646.76 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3797, pruned_loss=0.1298, over 5649488.68 frames. ], batch size: 199, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:49:23,130 INFO [train.py:968] (1/2) Epoch 13, batch 10200, giga_loss[loss=0.3147, simple_loss=0.3758, pruned_loss=0.1268, over 27605.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3754, pruned_loss=0.1256, over 5669768.00 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.353, pruned_loss=0.09784, over 5735024.10 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3784, pruned_loss=0.1288, over 5650857.14 frames. ], batch size: 472, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:49:25,488 INFO [optim.py:369] (1/2) 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 16:50:11,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=557713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:50:13,061 INFO [train.py:968] (1/2) Epoch 13, batch 10250, giga_loss[loss=0.2485, simple_loss=0.3298, pruned_loss=0.08357, over 28819.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3709, pruned_loss=0.1206, over 5673117.25 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3529, pruned_loss=0.09778, over 5737116.82 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1236, over 5655081.14 frames. ], batch size: 60, lr: 2.51e-03, grad_scale: 4.0
+2023-03-06 16:50:33,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-06 16:50:59,106 INFO [train.py:968] (1/2) Epoch 13, batch 10300, giga_loss[loss=0.2711, simple_loss=0.353, pruned_loss=0.09459, over 29085.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3679, pruned_loss=0.1174, over 5661184.30 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3533, pruned_loss=0.0981, over 5725444.01 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3707, pruned_loss=0.1206, over 5653969.17 frames. ], batch size: 128, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 16:51:02,946 INFO [optim.py:369] (1/2) 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 16:51:42,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-06 16:51:49,026 INFO [train.py:968] (1/2) Epoch 13, batch 10350, giga_loss[loss=0.3468, simple_loss=0.3956, pruned_loss=0.149, over 27540.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3683, pruned_loss=0.1176, over 5669436.57 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3534, pruned_loss=0.09822, over 5728934.68 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3707, pruned_loss=0.1203, over 5659520.48 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 16:52:33,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=557856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:52:37,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=557859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:52:38,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4025, 1.7537, 1.4168, 1.3627], device='cuda:1'), covar=tensor([0.2537, 0.2484, 0.2719, 0.2153], device='cuda:1'), in_proj_covar=tensor([0.1347, 0.0999, 0.1190, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 16:52:43,155 INFO [train.py:968] (1/2) Epoch 13, batch 10400, giga_loss[loss=0.274, simple_loss=0.3461, pruned_loss=0.101, over 28642.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.366, pruned_loss=0.1174, over 5663901.33 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3536, pruned_loss=0.09835, over 5728881.93 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1197, over 5655616.72 frames. ], batch size: 242, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:52:46,699 INFO [optim.py:369] (1/2) 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 16:53:09,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=557888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:53:31,247 INFO [train.py:968] (1/2) Epoch 13, batch 10450, giga_loss[loss=0.2813, simple_loss=0.3494, pruned_loss=0.1066, over 28562.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3649, pruned_loss=0.117, over 5657740.57 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3539, pruned_loss=0.09847, over 5722545.77 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3661, pruned_loss=0.119, over 5655803.68 frames. ], batch size: 85, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:54:17,419 INFO [train.py:968] (1/2) Epoch 13, batch 10500, giga_loss[loss=0.2754, simple_loss=0.3507, pruned_loss=0.1001, over 28653.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3676, pruned_loss=0.1181, over 5662232.87 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3541, pruned_loss=0.09842, over 5725611.10 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3687, pruned_loss=0.12, over 5656724.30 frames. ], batch size: 85, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:54:21,105 INFO [optim.py:369] (1/2) 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 16:54:48,730 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1590, 1.1779, 3.4933, 3.0613], device='cuda:1'), covar=tensor([0.1834, 0.2806, 0.0724, 0.1549], device='cuda:1'), in_proj_covar=tensor([0.0681, 0.0601, 0.0872, 0.0792], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 16:55:07,555 INFO [train.py:968] (1/2) Epoch 13, batch 10550, giga_loss[loss=0.3161, simple_loss=0.3759, pruned_loss=0.1282, over 28954.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3699, pruned_loss=0.1196, over 5658046.74 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3541, pruned_loss=0.09844, over 5726478.10 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.371, pruned_loss=0.1215, over 5651742.13 frames. ], batch size: 136, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:55:29,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9064, 1.0577, 0.8826, 0.3003], device='cuda:1'), covar=tensor([0.2186, 0.1850, 0.2112, 0.3785], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1509, 0.1495, 0.1303], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 16:55:57,772 INFO [train.py:968] (1/2) Epoch 13, batch 10600, giga_loss[loss=0.3093, simple_loss=0.3738, pruned_loss=0.1224, over 28943.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3698, pruned_loss=0.1198, over 5649406.69 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3544, pruned_loss=0.09843, over 5720852.16 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3707, pruned_loss=0.1218, over 5647408.82 frames. ], batch size: 136, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:56:03,314 INFO [optim.py:369] (1/2) 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 16:56:10,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 16:56:14,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4949, 1.7393, 1.7092, 1.4243], device='cuda:1'), covar=tensor([0.1945, 0.1620, 0.1093, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.1763, 0.1675, 0.1652, 0.1739], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 16:56:43,814 INFO [train.py:968] (1/2) Epoch 13, batch 10650, giga_loss[loss=0.3192, simple_loss=0.3866, pruned_loss=0.1259, over 28982.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3698, pruned_loss=0.1201, over 5649512.75 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3547, pruned_loss=0.09843, over 5721718.00 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3707, pruned_loss=0.1222, over 5645517.92 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:57:35,853 INFO [train.py:968] (1/2) Epoch 13, batch 10700, giga_loss[loss=0.2772, simple_loss=0.3584, pruned_loss=0.09795, over 28886.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3728, pruned_loss=0.1224, over 5654383.47 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3548, pruned_loss=0.09843, over 5720487.26 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1243, over 5651719.80 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:57:41,940 INFO [optim.py:369] (1/2) 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 16:58:26,003 INFO [train.py:968] (1/2) Epoch 13, batch 10750, giga_loss[loss=0.3188, simple_loss=0.3804, pruned_loss=0.1286, over 28904.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3756, pruned_loss=0.1239, over 5660929.39 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3548, pruned_loss=0.09847, over 5723998.40 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3766, pruned_loss=0.1259, over 5654315.24 frames. ], batch size: 145, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 16:58:26,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3407, 2.6166, 1.7800, 2.4923], device='cuda:1'), covar=tensor([0.0741, 0.0551, 0.0951, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0444, 0.0507, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 16:59:03,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-06 16:59:13,898 INFO [train.py:968] (1/2) Epoch 13, batch 10800, giga_loss[loss=0.3026, simple_loss=0.3713, pruned_loss=0.117, over 28983.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3774, pruned_loss=0.1255, over 5657537.45 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3549, pruned_loss=0.09849, over 5715495.96 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3784, pruned_loss=0.1275, over 5658774.35 frames. ], batch size: 136, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 16:59:18,646 INFO [optim.py:369] (1/2) 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 17:00:06,200 INFO [train.py:968] (1/2) Epoch 13, batch 10850, giga_loss[loss=0.3085, simple_loss=0.3814, pruned_loss=0.1178, over 29048.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3772, pruned_loss=0.1256, over 5667151.73 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3552, pruned_loss=0.09857, over 5718954.53 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3781, pruned_loss=0.1276, over 5664026.22 frames. ], batch size: 128, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:00:14,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4546, 1.5321, 1.1926, 1.1519], device='cuda:1'), covar=tensor([0.0764, 0.0514, 0.0972, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0441, 0.0503, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 17:00:57,681 INFO [train.py:968] (1/2) Epoch 13, batch 10900, giga_loss[loss=0.2936, simple_loss=0.3711, pruned_loss=0.108, over 28549.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3792, pruned_loss=0.1258, over 5660639.54 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3553, pruned_loss=0.09864, over 5721782.55 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3802, pruned_loss=0.1276, over 5655258.29 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:01:03,863 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.3729, 1.4705, 1.2766, 1.3114], device='cuda:1'), covar=tensor([0.1581, 0.1477, 0.1373, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.1767, 0.1679, 0.1658, 0.1744], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 17:01:50,169 INFO [train.py:968] (1/2) Epoch 13, batch 10950, giga_loss[loss=0.3336, simple_loss=0.393, pruned_loss=0.1371, over 28657.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3778, pruned_loss=0.1246, over 5659558.65 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3553, pruned_loss=0.09877, over 5720235.73 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.379, pruned_loss=0.1264, over 5655574.46 frames. ], batch size: 242, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:02:29,530 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 11000, giga_loss[loss=0.2624, simple_loss=0.3404, pruned_loss=0.09225, over 28947.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3763, pruned_loss=0.1247, over 5646535.02 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3547, pruned_loss=0.09844, over 5714137.30 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3782, pruned_loss=0.1271, over 5646697.45 frames. ], batch size: 164, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:02:48,862 INFO [optim.py:369] (1/2) 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 17:03:35,255 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 11050, libri_loss[loss=0.2955, simple_loss=0.3643, pruned_loss=0.1134, over 29515.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3758, pruned_loss=0.1251, over 5643405.20 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3549, pruned_loss=0.09857, over 5716649.19 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3775, pruned_loss=0.1272, over 5640443.93 frames. ], batch size: 81, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:04:19,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2618, 1.5825, 1.2589, 1.0501], device='cuda:1'), covar=tensor([0.2405, 0.2349, 0.2604, 0.2013], device='cuda:1'), in_proj_covar=tensor([0.1334, 0.0988, 0.1176, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 17:04:20,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4046, 1.4935, 1.3231, 1.6652], device='cuda:1'), covar=tensor([0.0755, 0.0324, 0.0317, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 17:04:30,968 INFO [train.py:968] (1/2) Epoch 13, batch 11100, giga_loss[loss=0.3527, simple_loss=0.3854, pruned_loss=0.16, over 28529.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3746, pruned_loss=0.125, over 5648131.32 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3551, pruned_loss=0.09864, over 5719918.29 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3762, pruned_loss=0.1272, over 5641237.93 frames. ], batch size: 78, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:04:37,894 INFO [optim.py:369] (1/2) 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 17:05:00,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=558596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:05:02,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=558599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:05:16,801 INFO [train.py:968] (1/2) Epoch 13, batch 11150, giga_loss[loss=0.2796, simple_loss=0.3427, pruned_loss=0.1082, over 28758.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3744, pruned_loss=0.1254, over 5653524.25 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3557, pruned_loss=0.09902, over 5722831.47 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3757, pruned_loss=0.1274, over 5643699.52 frames. ], batch size: 99, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:05:29,441 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=558628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:05:52,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4920, 1.7302, 1.6716, 1.6417], device='cuda:1'), covar=tensor([0.1661, 0.1730, 0.2092, 0.1629], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0738, 0.0680, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 17:06:03,163 INFO [train.py:968] (1/2) Epoch 13, batch 11200, giga_loss[loss=0.369, simple_loss=0.4058, pruned_loss=0.1661, over 27629.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3728, pruned_loss=0.1243, over 5661704.10 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3558, pruned_loss=0.09911, over 5727180.64 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3742, pruned_loss=0.1265, over 5648399.25 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:06:08,503 INFO [optim.py:369] (1/2) 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 17:06:54,560 INFO [train.py:968] (1/2) Epoch 13, batch 11250, giga_loss[loss=0.2811, simple_loss=0.3549, pruned_loss=0.1037, over 28981.00 frames. ], tot_loss[loss=0.311, simple_loss=0.373, pruned_loss=0.1245, over 5661095.70 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3558, pruned_loss=0.09912, over 5730532.94 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3743, pruned_loss=0.1266, over 5646311.20 frames. ], batch size: 106, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:07:39,158 INFO [train.py:968] (1/2) Epoch 13, batch 11300, giga_loss[loss=0.3826, simple_loss=0.4248, pruned_loss=0.1702, over 28722.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3739, pruned_loss=0.1254, over 5663149.46 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3553, pruned_loss=0.09872, over 5736819.05 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3761, pruned_loss=0.1285, over 5642889.15 frames. ], batch size: 284, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:07:45,025 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:07:45,363 INFO [optim.py:369] (1/2) 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 17:07:49,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6824, 1.7158, 1.2985, 1.2961], device='cuda:1'), covar=tensor([0.0796, 0.0643, 0.1016, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0445, 0.0504, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 17:08:17,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4001, 1.6421, 1.4523, 1.4974], device='cuda:1'), covar=tensor([0.0769, 0.0316, 0.0301, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0115, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 17:08:26,698 INFO [train.py:968] (1/2) Epoch 13, batch 11350, giga_loss[loss=0.3582, simple_loss=0.3922, pruned_loss=0.1621, over 23532.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3763, pruned_loss=0.1275, over 5657462.80 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3552, pruned_loss=0.09865, over 5740240.44 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3784, pruned_loss=0.1305, over 5637300.72 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:08:29,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3177, 1.1284, 4.3330, 3.3631], device='cuda:1'), covar=tensor([0.1690, 0.2789, 0.0382, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0686, 0.0606, 0.0879, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 17:09:15,355 INFO [train.py:968] (1/2) Epoch 13, batch 11400, giga_loss[loss=0.3324, simple_loss=0.3685, pruned_loss=0.1482, over 23618.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3758, pruned_loss=0.1274, over 5658101.56 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3551, pruned_loss=0.09848, over 5744951.92 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3781, pruned_loss=0.1305, over 5636197.61 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:09:22,355 INFO [optim.py:369] (1/2) 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 17:09:33,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:09:34,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=558882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:09:34,856 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=558883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:10:05,121 INFO [train.py:968] (1/2) Epoch 13, batch 11450, giga_loss[loss=0.388, simple_loss=0.4057, pruned_loss=0.1852, over 23596.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5659613.80 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.355, pruned_loss=0.09842, over 5742438.15 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3778, pruned_loss=0.1302, over 5643092.41 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:10:47,946 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-06 17:10:53,807 INFO [train.py:968] (1/2) Epoch 13, batch 11500, giga_loss[loss=0.2864, simple_loss=0.3591, pruned_loss=0.1068, over 28916.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3764, pruned_loss=0.1278, over 5654858.16 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3545, pruned_loss=0.09813, over 5744171.34 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3791, pruned_loss=0.1311, over 5638255.65 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:10:59,374 INFO [optim.py:369] (1/2) 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 17:11:22,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5444, 4.3665, 4.1314, 2.0842], device='cuda:1'), covar=tensor([0.0458, 0.0591, 0.0657, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1104, 0.1029, 0.0894, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 17:11:44,546 INFO [train.py:968] (1/2) Epoch 13, batch 11550, giga_loss[loss=0.3241, simple_loss=0.3842, pruned_loss=0.132, over 28821.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.377, pruned_loss=0.1275, over 5666819.71 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3546, pruned_loss=0.09819, over 5743735.59 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3792, pruned_loss=0.1303, over 5653518.31 frames. ], batch size: 199, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:11:55,241 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:11:57,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:12:25,791 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:12:34,003 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:12:34,429 INFO [train.py:968] (1/2) Epoch 13, batch 11600, giga_loss[loss=0.4159, simple_loss=0.432, pruned_loss=0.1999, over 26462.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3772, pruned_loss=0.1272, over 5666290.45 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3546, pruned_loss=0.09801, over 5745469.62 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3795, pruned_loss=0.1303, over 5652599.99 frames. ], batch size: 555, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:12:43,755 INFO [optim.py:369] (1/2) 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 17:13:29,144 INFO [train.py:968] (1/2) Epoch 13, batch 11650, giga_loss[loss=0.3117, simple_loss=0.3811, pruned_loss=0.1212, over 28989.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.381, pruned_loss=0.1309, over 5654503.58 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3547, pruned_loss=0.09799, over 5746270.28 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3828, pruned_loss=0.1335, over 5642782.22 frames. ], batch size: 164, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:13:59,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:14:09,628 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 17:14:16,889 INFO [train.py:968] (1/2) Epoch 13, batch 11700, giga_loss[loss=0.3383, simple_loss=0.3892, pruned_loss=0.1437, over 28899.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3806, pruned_loss=0.1311, over 5659693.34 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3548, pruned_loss=0.0981, over 5749403.03 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3825, pruned_loss=0.1337, over 5646022.95 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:14:22,918 INFO [optim.py:369] (1/2) 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 17:14:44,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2546, 1.6116, 1.3027, 1.4325], device='cuda:1'), covar=tensor([0.0777, 0.0312, 0.0328, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 17:14:48,064 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 17:15:03,040 INFO [train.py:968] (1/2) Epoch 13, batch 11750, giga_loss[loss=0.2898, simple_loss=0.3621, pruned_loss=0.1088, over 28959.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3809, pruned_loss=0.1301, over 5658860.82 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3549, pruned_loss=0.09815, over 5750504.18 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3831, pruned_loss=0.133, over 5644687.27 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:15:23,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0412, 1.4071, 5.0883, 4.0755], device='cuda:1'), covar=tensor([0.1733, 0.3169, 0.0720, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0685, 0.0605, 0.0879, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 17:15:42,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:15:44,750 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 11800, giga_loss[loss=0.2812, simple_loss=0.3568, pruned_loss=0.1028, over 28757.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3812, pruned_loss=0.1294, over 5656819.94 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3552, pruned_loss=0.0984, over 5751372.56 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3829, pruned_loss=0.1319, over 5643586.28 frames. ], batch size: 119, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:15:59,478 INFO [optim.py:369] (1/2) 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 17:16:16,395 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:16:18,273 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:16:42,758 INFO [train.py:968] (1/2) Epoch 13, batch 11850, giga_loss[loss=0.3005, simple_loss=0.3682, pruned_loss=0.1164, over 29022.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3808, pruned_loss=0.1291, over 5658078.54 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3552, pruned_loss=0.09834, over 5752143.70 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3823, pruned_loss=0.1313, over 5646639.20 frames. ], batch size: 213, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:16:49,752 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:17:04,741 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 17:17:18,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5870, 1.6368, 1.2071, 1.2765], device='cuda:1'), covar=tensor([0.0709, 0.0518, 0.0933, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0448, 0.0507, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 17:17:27,204 INFO [train.py:968] (1/2) Epoch 13, batch 11900, giga_loss[loss=0.2885, simple_loss=0.3577, pruned_loss=0.1097, over 28749.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3786, pruned_loss=0.1277, over 5662321.18 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3551, pruned_loss=0.09827, over 5757870.08 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3809, pruned_loss=0.1307, over 5644469.12 frames. ], batch size: 284, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:17:32,591 INFO [optim.py:369] (1/2) 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 17:18:02,171 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:04,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:04,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:06,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:14,126 INFO [train.py:968] (1/2) Epoch 13, batch 11950, giga_loss[loss=0.3153, simple_loss=0.3852, pruned_loss=0.1227, over 28696.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3787, pruned_loss=0.1275, over 5666067.97 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.355, pruned_loss=0.09826, over 5757565.71 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3814, pruned_loss=0.1307, over 5649377.30 frames. ], batch size: 307, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:18:27,672 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-06 17:18:31,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:34,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:18:41,143 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:19:06,053 INFO [train.py:968] (1/2) Epoch 13, batch 12000, giga_loss[loss=0.3792, simple_loss=0.4124, pruned_loss=0.1731, over 27516.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3794, pruned_loss=0.128, over 5658115.71 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3549, pruned_loss=0.09821, over 5755366.92 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3817, pruned_loss=0.1309, over 5645841.27 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:19:06,053 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 17:19:14,775 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 17:19:20,310 INFO [optim.py:369] (1/2) 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 17:19:38,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-06 17:20:01,611 INFO [train.py:968] (1/2) Epoch 13, batch 12050, giga_loss[loss=0.3505, simple_loss=0.3828, pruned_loss=0.1592, over 23672.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3778, pruned_loss=0.1275, over 5669540.20 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3546, pruned_loss=0.09796, over 5757575.56 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3804, pruned_loss=0.1305, over 5656080.84 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:20:49,263 INFO [train.py:968] (1/2) Epoch 13, batch 12100, giga_loss[loss=0.3655, simple_loss=0.408, pruned_loss=0.1615, over 28307.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3784, pruned_loss=0.1281, over 5671490.76 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3545, pruned_loss=0.0979, over 5756702.71 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3812, pruned_loss=0.1314, over 5658996.51 frames. ], batch size: 368, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:20:52,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7820, 1.1926, 5.2090, 3.6563], device='cuda:1'), covar=tensor([0.1562, 0.2826, 0.0389, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0689, 0.0606, 0.0884, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 17:20:54,567 INFO [optim.py:369] (1/2) 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 17:21:04,200 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=559582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:09,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:36,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=559614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:37,081 INFO [train.py:968] (1/2) Epoch 13, batch 12150, giga_loss[loss=0.2946, simple_loss=0.3701, pruned_loss=0.1095, over 28914.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.379, pruned_loss=0.1283, over 5665610.07 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3543, pruned_loss=0.09774, over 5750466.98 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3821, pruned_loss=0.1321, over 5659173.71 frames. ], batch size: 186, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:21:43,860 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:21:46,354 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:22:24,601 INFO [train.py:968] (1/2) Epoch 13, batch 12200, giga_loss[loss=0.3726, simple_loss=0.4112, pruned_loss=0.167, over 28429.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3792, pruned_loss=0.1286, over 5662427.51 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.354, pruned_loss=0.09747, over 5750338.61 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3827, pruned_loss=0.1325, over 5655262.19 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:22:31,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:22:31,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8748, 1.8115, 1.3597, 1.4116], device='cuda:1'), covar=tensor([0.0690, 0.0572, 0.0917, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0449, 0.0508, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 17:22:32,984 INFO [optim.py:369] (1/2) 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 17:22:48,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3673, 1.7499, 1.6077, 1.2027], device='cuda:1'), covar=tensor([0.1862, 0.2654, 0.1604, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.0842, 0.0699, 0.0884, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 17:23:11,592 INFO [train.py:968] (1/2) Epoch 13, batch 12250, giga_loss[loss=0.3303, simple_loss=0.3912, pruned_loss=0.1347, over 28746.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3783, pruned_loss=0.1269, over 5675269.72 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3542, pruned_loss=0.09736, over 5752369.20 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3816, pruned_loss=0.131, over 5665417.97 frames. ], batch size: 243, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:24:04,844 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=559763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:24:05,830 INFO [train.py:968] (1/2) Epoch 13, batch 12300, giga_loss[loss=0.293, simple_loss=0.3625, pruned_loss=0.1117, over 28866.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3787, pruned_loss=0.1272, over 5663849.28 frames. ], libri_tot_loss[loss=0.2749, simple_loss=0.3545, pruned_loss=0.09764, over 5753848.06 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3812, pruned_loss=0.1304, over 5654191.26 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:24:12,116 INFO [optim.py:369] (1/2) 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 17:24:46,531 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 12350, libri_loss[loss=0.2812, simple_loss=0.367, pruned_loss=0.09772, over 29688.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3777, pruned_loss=0.1252, over 5671909.25 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3547, pruned_loss=0.09776, over 5750546.70 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3804, pruned_loss=0.1288, over 5663255.23 frames. ], batch size: 88, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:24:48,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=559817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:25:16,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 12400, libri_loss[loss=0.2374, simple_loss=0.3188, pruned_loss=0.07804, over 29656.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3765, pruned_loss=0.1242, over 5687680.27 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3544, pruned_loss=0.09751, over 5754261.08 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3795, pruned_loss=0.1279, over 5675789.84 frames. ], batch size: 73, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:25:43,117 INFO [optim.py:369] (1/2) 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 17:26:07,754 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-06 17:26:20,989 INFO [train.py:968] (1/2) Epoch 13, batch 12450, giga_loss[loss=0.2759, simple_loss=0.3517, pruned_loss=0.1001, over 28983.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3758, pruned_loss=0.1245, over 5678729.50 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3545, pruned_loss=0.09756, over 5755140.74 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3789, pruned_loss=0.1283, over 5665531.52 frames. ], batch size: 128, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:26:40,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.91 vs. limit=2.0
+2023-03-06 17:27:08,487 INFO [train.py:968] (1/2) Epoch 13, batch 12500, libri_loss[loss=0.2363, simple_loss=0.3131, pruned_loss=0.07975, over 29640.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3726, pruned_loss=0.1227, over 5673814.08 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.354, pruned_loss=0.09726, over 5751866.00 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5663174.25 frames. ], batch size: 73, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:27:08,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-06 17:27:17,053 INFO [optim.py:369] (1/2) 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 17:27:37,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:27:40,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=559999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:27:55,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 17:27:55,899 INFO [train.py:968] (1/2) Epoch 13, batch 12550, giga_loss[loss=0.2838, simple_loss=0.3526, pruned_loss=0.1074, over 29082.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5682324.10 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.354, pruned_loss=0.09724, over 5754165.84 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3719, pruned_loss=0.1247, over 5670400.98 frames. ], batch size: 155, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:28:44,005 INFO [train.py:968] (1/2) Epoch 13, batch 12600, giga_loss[loss=0.2719, simple_loss=0.3421, pruned_loss=0.1008, over 28674.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3669, pruned_loss=0.1205, over 5689469.45 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3536, pruned_loss=0.09698, over 5757149.94 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3701, pruned_loss=0.1242, over 5675995.56 frames. ], batch size: 242, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:28:53,734 INFO [optim.py:369] (1/2) 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 17:29:33,249 INFO [train.py:968] (1/2) Epoch 13, batch 12650, giga_loss[loss=0.2781, simple_loss=0.3447, pruned_loss=0.1058, over 28903.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3657, pruned_loss=0.1195, over 5697709.32 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3541, pruned_loss=0.0971, over 5760588.01 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3681, pruned_loss=0.123, over 5681869.68 frames. ], batch size: 136, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:29:55,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:29:56,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:29:58,098 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 17:29:59,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:30:03,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:30:21,355 INFO [train.py:968] (1/2) Epoch 13, batch 12700, giga_loss[loss=0.2576, simple_loss=0.3286, pruned_loss=0.09337, over 28595.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.366, pruned_loss=0.1188, over 5694251.90 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3536, pruned_loss=0.09685, over 5762932.02 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3686, pruned_loss=0.1224, over 5677955.29 frames. ], batch size: 78, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:30:30,195 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:30:31,730 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:31:12,702 INFO [train.py:968] (1/2) Epoch 13, batch 12750, giga_loss[loss=0.2891, simple_loss=0.3652, pruned_loss=0.1065, over 28700.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1162, over 5689829.27 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3533, pruned_loss=0.09667, over 5765967.94 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 5672950.66 frames. ], batch size: 284, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:31:46,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-06 17:32:09,023 INFO [train.py:968] (1/2) Epoch 13, batch 12800, giga_loss[loss=0.2786, simple_loss=0.3432, pruned_loss=0.107, over 28676.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.363, pruned_loss=0.1137, over 5674253.61 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3532, pruned_loss=0.09657, over 5762014.04 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3652, pruned_loss=0.1166, over 5663697.51 frames. ], batch size: 85, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:32:17,321 INFO [optim.py:369] (1/2) 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 17:32:26,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:32:28,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:32:43,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5601, 1.9356, 1.7541, 1.4874], device='cuda:1'), covar=tensor([0.2261, 0.1555, 0.1680, 0.1924], device='cuda:1'), in_proj_covar=tensor([0.1737, 0.1649, 0.1607, 0.1703], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 17:32:46,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:32:58,965 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 12850, giga_loss[loss=0.2848, simple_loss=0.3532, pruned_loss=0.1083, over 27631.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3595, pruned_loss=0.1106, over 5660434.32 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3528, pruned_loss=0.09649, over 5754686.96 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3617, pruned_loss=0.1132, over 5657545.44 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:33:21,760 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 12900, libri_loss[loss=0.2949, simple_loss=0.366, pruned_loss=0.112, over 29239.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3563, pruned_loss=0.1076, over 5650769.49 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3524, pruned_loss=0.09661, over 5738636.29 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3587, pruned_loss=0.1102, over 5659984.66 frames. ], batch size: 97, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:33:58,584 INFO [optim.py:369] (1/2) 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 17:34:09,801 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-06 17:34:14,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-06 17:34:39,164 INFO [train.py:968] (1/2) Epoch 13, batch 12950, giga_loss[loss=0.2349, simple_loss=0.3327, pruned_loss=0.06858, over 28878.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3546, pruned_loss=0.1043, over 5655588.37 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3519, pruned_loss=0.09642, over 5739267.23 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.357, pruned_loss=0.1066, over 5660905.81 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:35:33,432 INFO [train.py:968] (1/2) Epoch 13, batch 13000, giga_loss[loss=0.2382, simple_loss=0.3078, pruned_loss=0.08431, over 24052.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3539, pruned_loss=0.1036, over 5649425.98 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3514, pruned_loss=0.09625, over 5742888.39 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3563, pruned_loss=0.1058, over 5648805.84 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:35:40,448 INFO [optim.py:369] (1/2) 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 17:35:42,128 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 17:35:46,368 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 13050, giga_loss[loss=0.2404, simple_loss=0.3275, pruned_loss=0.07667, over 28905.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3535, pruned_loss=0.1028, over 5665119.17 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3515, pruned_loss=0.09636, over 5746736.36 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3555, pruned_loss=0.1047, over 5658140.17 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:36:25,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560518.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 17:36:32,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6114, 1.9567, 1.7020, 1.4601], device='cuda:1'), covar=tensor([0.2261, 0.1492, 0.1416, 0.1769], device='cuda:1'), in_proj_covar=tensor([0.1727, 0.1635, 0.1593, 0.1688], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 17:37:12,202 INFO [train.py:968] (1/2) Epoch 13, batch 13100, giga_loss[loss=0.2893, simple_loss=0.3513, pruned_loss=0.1136, over 27629.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.351, pruned_loss=0.1015, over 5664310.26 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3505, pruned_loss=0.09603, over 5748381.80 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3536, pruned_loss=0.1036, over 5654439.53 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:37:18,732 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2198, 1.1227, 3.6766, 3.1286], device='cuda:1'), covar=tensor([0.1625, 0.2861, 0.0436, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0686, 0.0604, 0.0877, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 17:37:20,615 INFO [optim.py:369] (1/2) 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 17:38:02,043 INFO [train.py:968] (1/2) Epoch 13, batch 13150, giga_loss[loss=0.2516, simple_loss=0.3369, pruned_loss=0.08313, over 28915.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3483, pruned_loss=0.09994, over 5662036.39 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3495, pruned_loss=0.09556, over 5742069.32 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3511, pruned_loss=0.1021, over 5657083.57 frames. ], batch size: 145, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:38:18,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1767, 1.3036, 3.7923, 3.1370], device='cuda:1'), covar=tensor([0.1752, 0.2658, 0.0472, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0684, 0.0602, 0.0877, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 17:38:38,414 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=560651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:38:47,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560661.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 17:38:52,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560664.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 17:38:52,903 INFO [train.py:968] (1/2) Epoch 13, batch 13200, giga_loss[loss=0.293, simple_loss=0.3681, pruned_loss=0.1089, over 28806.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3486, pruned_loss=0.09986, over 5666717.75 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3492, pruned_loss=0.09543, over 5744312.01 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3511, pruned_loss=0.1017, over 5659946.79 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:39:01,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:39:02,038 INFO [optim.py:369] (1/2) 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 17:39:22,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560693.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 17:39:39,506 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 13250, giga_loss[loss=0.2645, simple_loss=0.345, pruned_loss=0.09197, over 28822.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3475, pruned_loss=0.09891, over 5665343.14 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3489, pruned_loss=0.09531, over 5744983.45 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3497, pruned_loss=0.1005, over 5659141.76 frames. ], batch size: 199, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:40:38,567 INFO [train.py:968] (1/2) Epoch 13, batch 13300, giga_loss[loss=0.2941, simple_loss=0.3597, pruned_loss=0.1143, over 28071.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3448, pruned_loss=0.0966, over 5666529.67 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3486, pruned_loss=0.09507, over 5747415.45 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3468, pruned_loss=0.09809, over 5658532.84 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:40:47,331 INFO [optim.py:369] (1/2) 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 17:41:03,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6517, 2.4217, 1.7199, 0.9040], device='cuda:1'), covar=tensor([0.4428, 0.2557, 0.3363, 0.4459], device='cuda:1'), in_proj_covar=tensor([0.1595, 0.1511, 0.1503, 0.1302], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 17:41:29,120 INFO [train.py:968] (1/2) Epoch 13, batch 13350, giga_loss[loss=0.2411, simple_loss=0.3212, pruned_loss=0.08051, over 28746.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3413, pruned_loss=0.09448, over 5670701.22 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3482, pruned_loss=0.09489, over 5752341.42 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3432, pruned_loss=0.09588, over 5657703.95 frames. ], batch size: 119, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:41:32,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:41:32,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6810, 2.0227, 1.9905, 1.4827], device='cuda:1'), covar=tensor([0.1810, 0.2386, 0.1454, 0.1744], device='cuda:1'), in_proj_covar=tensor([0.0842, 0.0688, 0.0881, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 17:41:35,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:06,419 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:09,205 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:11,147 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=560853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:12,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:42:20,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4409, 1.9957, 1.4433, 0.7031], device='cuda:1'), covar=tensor([0.3883, 0.2258, 0.2628, 0.4315], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1515, 0.1505, 0.1305], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 17:42:21,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4298, 1.7513, 1.7103, 1.2613], device='cuda:1'), covar=tensor([0.1667, 0.2501, 0.1370, 0.1703], device='cuda:1'), in_proj_covar=tensor([0.0841, 0.0686, 0.0880, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 17:42:21,843 INFO [train.py:968] (1/2) Epoch 13, batch 13400, libri_loss[loss=0.2762, simple_loss=0.3498, pruned_loss=0.1013, over 29174.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3394, pruned_loss=0.09417, over 5661665.76 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3475, pruned_loss=0.09464, over 5755355.58 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3413, pruned_loss=0.09551, over 5645826.74 frames. ], batch size: 97, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:42:36,493 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=560884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:43:16,153 INFO [train.py:968] (1/2) Epoch 13, batch 13450, giga_loss[loss=0.2796, simple_loss=0.3434, pruned_loss=0.1079, over 26710.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3391, pruned_loss=0.0949, over 5657007.96 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3476, pruned_loss=0.09474, over 5757570.70 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3404, pruned_loss=0.09589, over 5640941.63 frames. ], batch size: 555, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:43:43,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7312, 1.1684, 1.1673, 0.9987], device='cuda:1'), covar=tensor([0.1817, 0.1216, 0.2002, 0.1574], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0719, 0.0664, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 17:44:14,699 INFO [train.py:968] (1/2) Epoch 13, batch 13500, libri_loss[loss=0.2231, simple_loss=0.2985, pruned_loss=0.07388, over 29585.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.34, pruned_loss=0.09566, over 5645757.86 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3467, pruned_loss=0.09443, over 5759744.22 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3417, pruned_loss=0.09679, over 5627927.58 frames. ], batch size: 75, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:44:24,222 INFO [optim.py:369] (1/2) 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 17:44:48,778 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=560996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:44:52,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=560999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:45:07,279 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:45:09,210 INFO [train.py:968] (1/2) Epoch 13, batch 13550, giga_loss[loss=0.2714, simple_loss=0.3522, pruned_loss=0.09527, over 27994.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3415, pruned_loss=0.09475, over 5650937.82 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3461, pruned_loss=0.09413, over 5758880.81 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3433, pruned_loss=0.09594, over 5635912.80 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 17:45:23,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=561026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:45:25,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=561028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:46:09,700 INFO [train.py:968] (1/2) Epoch 13, batch 13600, giga_loss[loss=0.2861, simple_loss=0.3579, pruned_loss=0.1072, over 27529.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3425, pruned_loss=0.09517, over 5637285.67 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3463, pruned_loss=0.09431, over 5747697.41 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3437, pruned_loss=0.09597, over 5633899.72 frames. ], batch size: 472, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:46:14,542 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-06 17:46:23,424 INFO [optim.py:369] (1/2) 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 17:47:10,841 INFO [train.py:968] (1/2) Epoch 13, batch 13650, giga_loss[loss=0.3693, simple_loss=0.4146, pruned_loss=0.1621, over 28638.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3426, pruned_loss=0.09576, over 5635371.54 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.346, pruned_loss=0.09428, over 5741879.55 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3438, pruned_loss=0.09645, over 5635918.59 frames. ], batch size: 307, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:47:48,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4596, 1.6835, 1.3431, 1.6226], device='cuda:1'), covar=tensor([0.2554, 0.2342, 0.2637, 0.2083], device='cuda:1'), in_proj_covar=tensor([0.1339, 0.0983, 0.1184, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 17:48:09,275 INFO [train.py:968] (1/2) Epoch 13, batch 13700, libri_loss[loss=0.2452, simple_loss=0.335, pruned_loss=0.0777, over 28656.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3408, pruned_loss=0.09407, over 5636475.08 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3458, pruned_loss=0.09424, over 5735990.18 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3418, pruned_loss=0.09467, over 5638722.35 frames. ], batch size: 106, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:48:14,713 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=561169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:48:17,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=561172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:48:21,284 INFO [optim.py:369] (1/2) 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 17:48:48,098 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=561201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:49:00,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-06 17:49:03,910 INFO [train.py:968] (1/2) Epoch 13, batch 13750, giga_loss[loss=0.2924, simple_loss=0.3484, pruned_loss=0.1182, over 26783.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3397, pruned_loss=0.0924, over 5640160.47 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3457, pruned_loss=0.0944, over 5735913.53 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3404, pruned_loss=0.09271, over 5638731.57 frames. ], batch size: 555, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:49:22,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9905, 2.8194, 2.6544, 1.5773], device='cuda:1'), covar=tensor([0.1029, 0.1163, 0.1047, 0.2137], device='cuda:1'), in_proj_covar=tensor([0.1069, 0.0992, 0.0855, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 17:50:03,897 INFO [train.py:968] (1/2) Epoch 13, batch 13800, giga_loss[loss=0.2564, simple_loss=0.3288, pruned_loss=0.09201, over 28859.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3365, pruned_loss=0.09089, over 5651698.99 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.345, pruned_loss=0.09408, over 5742833.37 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3374, pruned_loss=0.09134, over 5640718.37 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:50:18,222 INFO [optim.py:369] (1/2) 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 17:50:59,103 INFO [train.py:968] (1/2) Epoch 13, batch 13850, giga_loss[loss=0.3051, simple_loss=0.3773, pruned_loss=0.1165, over 28430.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3362, pruned_loss=0.09162, over 5661600.26 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.345, pruned_loss=0.09421, over 5746807.33 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3366, pruned_loss=0.09175, over 5645917.36 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:51:23,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4329, 1.6294, 1.5968, 1.3865], device='cuda:1'), covar=tensor([0.1975, 0.1577, 0.1114, 0.1527], device='cuda:1'), in_proj_covar=tensor([0.1739, 0.1633, 0.1584, 0.1691], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 17:51:53,111 INFO [train.py:968] (1/2) Epoch 13, batch 13900, libri_loss[loss=0.2574, simple_loss=0.3367, pruned_loss=0.08908, over 29669.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3368, pruned_loss=0.0926, over 5644144.08 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3454, pruned_loss=0.09472, over 5725244.50 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3363, pruned_loss=0.09212, over 5646950.28 frames. ], batch size: 88, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:52:07,271 INFO [optim.py:369] (1/2) 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 17:52:19,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=561387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:52:51,335 INFO [train.py:968] (1/2) Epoch 13, batch 13950, giga_loss[loss=0.2745, simple_loss=0.3538, pruned_loss=0.09763, over 28859.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.338, pruned_loss=0.09234, over 5649477.18 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3453, pruned_loss=0.09468, over 5717161.36 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3376, pruned_loss=0.09194, over 5658429.21 frames. ], batch size: 119, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:53:49,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2086, 1.5395, 1.2138, 1.0442], device='cuda:1'), covar=tensor([0.2402, 0.2231, 0.2449, 0.1951], device='cuda:1'), in_proj_covar=tensor([0.1343, 0.0984, 0.1187, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 17:53:55,359 INFO [train.py:968] (1/2) Epoch 13, batch 14000, giga_loss[loss=0.2378, simple_loss=0.3261, pruned_loss=0.07474, over 29036.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3418, pruned_loss=0.09362, over 5658467.49 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3457, pruned_loss=0.09509, over 5715173.15 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.341, pruned_loss=0.09292, over 5666949.61 frames. ], batch size: 155, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 17:54:10,794 INFO [optim.py:369] (1/2) 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 17:54:23,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:54:58,176 INFO [train.py:968] (1/2) Epoch 13, batch 14050, giga_loss[loss=0.2191, simple_loss=0.303, pruned_loss=0.06756, over 28482.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.338, pruned_loss=0.09124, over 5666998.17 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3456, pruned_loss=0.09513, over 5721767.08 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3372, pruned_loss=0.09053, over 5665540.65 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:55:15,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=561530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:55:17,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5554, 1.7779, 1.7679, 1.5482], device='cuda:1'), covar=tensor([0.1484, 0.2009, 0.1819, 0.1866], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0719, 0.0665, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 17:55:18,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=561533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:55:53,389 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 14100, giga_loss[loss=0.2716, simple_loss=0.3464, pruned_loss=0.09835, over 28916.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3378, pruned_loss=0.09132, over 5681115.12 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3448, pruned_loss=0.09466, over 5726392.83 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3375, pruned_loss=0.09096, over 5673201.12 frames. ], batch size: 199, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:56:10,578 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 14150, giga_loss[loss=0.2693, simple_loss=0.3642, pruned_loss=0.08725, over 28917.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3387, pruned_loss=0.09193, over 5664427.03 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3441, pruned_loss=0.09454, over 5731697.22 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3389, pruned_loss=0.09167, over 5651511.16 frames. ], batch size: 227, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:58:03,207 INFO [train.py:968] (1/2) Epoch 13, batch 14200, giga_loss[loss=0.2681, simple_loss=0.3536, pruned_loss=0.09133, over 28129.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3422, pruned_loss=0.09155, over 5663716.79 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3439, pruned_loss=0.09461, over 5735336.66 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3425, pruned_loss=0.09124, over 5648910.62 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:58:04,903 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-06 17:58:19,695 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 14250, giga_loss[loss=0.2246, simple_loss=0.3256, pruned_loss=0.06174, over 28964.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3413, pruned_loss=0.08977, over 5650947.69 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3437, pruned_loss=0.09445, over 5739000.18 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3417, pruned_loss=0.08952, over 5633533.54 frames. ], batch size: 155, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 17:59:12,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 17:59:28,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3995, 2.0402, 1.4714, 0.6003], device='cuda:1'), covar=tensor([0.4843, 0.2070, 0.3221, 0.5012], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1516, 0.1506, 0.1301], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 17:59:51,250 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=561754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:00:05,002 INFO [train.py:968] (1/2) Epoch 13, batch 14300, giga_loss[loss=0.2341, simple_loss=0.3275, pruned_loss=0.07038, over 28833.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3411, pruned_loss=0.08813, over 5665686.99 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3438, pruned_loss=0.09458, over 5741244.90 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3413, pruned_loss=0.08774, over 5648763.86 frames. ], batch size: 174, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:00:18,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2605, 1.6088, 1.6729, 1.3927], device='cuda:1'), covar=tensor([0.1424, 0.1295, 0.1732, 0.1422], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0721, 0.0667, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 18:00:19,460 INFO [optim.py:369] (1/2) 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 18:01:02,670 INFO [train.py:968] (1/2) Epoch 13, batch 14350, giga_loss[loss=0.2551, simple_loss=0.3386, pruned_loss=0.08583, over 28752.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3426, pruned_loss=0.08978, over 5673335.03 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3438, pruned_loss=0.0947, over 5744261.09 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08908, over 5653773.80 frames. ], batch size: 243, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:02:00,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=561862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:02:02,483 INFO [train.py:968] (1/2) Epoch 13, batch 14400, libri_loss[loss=0.1992, simple_loss=0.2777, pruned_loss=0.06033, over 29348.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3406, pruned_loss=0.08988, over 5671101.62 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3432, pruned_loss=0.09445, over 5740448.11 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3412, pruned_loss=0.0894, over 5655627.36 frames. ], batch size: 71, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 18:02:06,258 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-06 18:02:19,509 INFO [optim.py:369] (1/2) 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 18:03:10,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3009, 1.3335, 4.1264, 3.1266], device='cuda:1'), covar=tensor([0.1618, 0.2567, 0.0397, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0674, 0.0597, 0.0861, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 18:03:18,830 INFO [train.py:968] (1/2) Epoch 13, batch 14450, giga_loss[loss=0.271, simple_loss=0.3256, pruned_loss=0.1082, over 24701.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3421, pruned_loss=0.09198, over 5667849.40 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3428, pruned_loss=0.0943, over 5741996.30 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3428, pruned_loss=0.09172, over 5653531.55 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 8.0
+2023-03-06 18:03:55,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7210, 2.2068, 2.1261, 1.5239], device='cuda:1'), covar=tensor([0.1869, 0.2186, 0.1423, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0840, 0.0683, 0.0880, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 18:04:38,914 INFO [train.py:968] (1/2) Epoch 13, batch 14500, libri_loss[loss=0.2793, simple_loss=0.3551, pruned_loss=0.1018, over 28665.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3387, pruned_loss=0.08961, over 5677407.06 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3427, pruned_loss=0.09422, over 5742416.95 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3394, pruned_loss=0.08941, over 5664336.61 frames. ], batch size: 106, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:04:58,829 INFO [optim.py:369] (1/2) 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 18:05:28,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3359, 1.6479, 1.3755, 1.5190], device='cuda:1'), covar=tensor([0.0705, 0.0345, 0.0320, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0060, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 18:05:37,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:05:42,523 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 14550, giga_loss[loss=0.2415, simple_loss=0.3256, pruned_loss=0.07871, over 28372.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3356, pruned_loss=0.08809, over 5671962.82 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3423, pruned_loss=0.09399, over 5743560.72 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3364, pruned_loss=0.08805, over 5659832.52 frames. ], batch size: 368, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:06:21,164 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:06:59,979 INFO [train.py:968] (1/2) Epoch 13, batch 14600, giga_loss[loss=0.2608, simple_loss=0.3406, pruned_loss=0.09055, over 28422.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3324, pruned_loss=0.08659, over 5674803.04 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3418, pruned_loss=0.09376, over 5745122.87 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3333, pruned_loss=0.08666, over 5662828.19 frames. ], batch size: 336, lr: 2.50e-03, grad_scale: 4.0
+2023-03-06 18:07:21,270 INFO [optim.py:369] (1/2) 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 18:07:23,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0676, 3.8968, 3.6714, 1.9690], device='cuda:1'), covar=tensor([0.0629, 0.0774, 0.0842, 0.2159], device='cuda:1'), in_proj_covar=tensor([0.1070, 0.0993, 0.0860, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 18:07:41,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:07:54,586 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:08:02,376 INFO [train.py:968] (1/2) Epoch 13, batch 14650, giga_loss[loss=0.2953, simple_loss=0.371, pruned_loss=0.1097, over 28084.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3372, pruned_loss=0.08962, over 5679807.33 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3412, pruned_loss=0.09356, over 5740130.26 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3383, pruned_loss=0.08971, over 5672058.73 frames. ], batch size: 412, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 18:08:21,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:09:03,615 INFO [train.py:968] (1/2) Epoch 13, batch 14700, giga_loss[loss=0.215, simple_loss=0.281, pruned_loss=0.07455, over 24416.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3384, pruned_loss=0.09084, over 5678569.65 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3403, pruned_loss=0.09312, over 5744396.93 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3401, pruned_loss=0.09124, over 5666915.44 frames. ], batch size: 705, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 18:09:23,430 INFO [optim.py:369] (1/2) 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 18:10:03,946 INFO [train.py:968] (1/2) Epoch 13, batch 14750, giga_loss[loss=0.2748, simple_loss=0.3503, pruned_loss=0.09967, over 28727.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3375, pruned_loss=0.09149, over 5684617.24 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3399, pruned_loss=0.09298, over 5745357.70 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3391, pruned_loss=0.09189, over 5672665.34 frames. ], batch size: 307, lr: 2.50e-03, grad_scale: 2.0
+2023-03-06 18:10:41,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:10:45,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:10,329 INFO [train.py:968] (1/2) Epoch 13, batch 14800, giga_loss[loss=0.2545, simple_loss=0.3257, pruned_loss=0.0917, over 28398.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3399, pruned_loss=0.09393, over 5681646.73 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3399, pruned_loss=0.09307, over 5747893.47 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3411, pruned_loss=0.09418, over 5668990.12 frames. ], batch size: 78, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:11:19,709 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:19,749 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:23,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:11:29,117 INFO [optim.py:369] (1/2) 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 18:11:59,130 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:12:11,920 INFO [train.py:968] (1/2) Epoch 13, batch 14850, giga_loss[loss=0.2434, simple_loss=0.3325, pruned_loss=0.07719, over 29019.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3397, pruned_loss=0.09313, over 5679571.38 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3395, pruned_loss=0.0928, over 5752578.95 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3411, pruned_loss=0.0936, over 5662948.82 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:13:17,482 INFO [train.py:968] (1/2) Epoch 13, batch 14900, giga_loss[loss=0.2582, simple_loss=0.3371, pruned_loss=0.08959, over 28131.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3407, pruned_loss=0.09237, over 5681734.05 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3389, pruned_loss=0.09241, over 5755635.26 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3425, pruned_loss=0.09309, over 5663920.35 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:13:44,695 INFO [optim.py:369] (1/2) 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 18:14:45,141 INFO [train.py:968] (1/2) Epoch 13, batch 14950, giga_loss[loss=0.2615, simple_loss=0.3399, pruned_loss=0.09157, over 28989.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3402, pruned_loss=0.09183, over 5680038.85 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3389, pruned_loss=0.09241, over 5755635.26 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3415, pruned_loss=0.09239, over 5666174.22 frames. ], batch size: 136, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:15:21,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4929, 3.3028, 1.5312, 1.5804], device='cuda:1'), covar=tensor([0.0912, 0.0415, 0.0874, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0513, 0.0348, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 18:15:30,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:15:58,693 INFO [train.py:968] (1/2) Epoch 13, batch 15000, giga_loss[loss=0.2336, simple_loss=0.3062, pruned_loss=0.08052, over 28779.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.336, pruned_loss=0.0904, over 5692052.98 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3388, pruned_loss=0.09246, over 5755342.38 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.337, pruned_loss=0.09076, over 5680238.00 frames. ], batch size: 99, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:15:58,694 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 18:16:07,817 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 18:16:14,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2221, 2.6050, 1.2864, 1.3619], device='cuda:1'), covar=tensor([0.0945, 0.0349, 0.0871, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0514, 0.0348, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 18:16:20,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 18:16:25,121 INFO [optim.py:369] (1/2) 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 18:16:30,600 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 15050, giga_loss[loss=0.2465, simple_loss=0.3187, pruned_loss=0.08721, over 28746.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3297, pruned_loss=0.0873, over 5687952.25 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3384, pruned_loss=0.09219, over 5758686.38 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3308, pruned_loss=0.08778, over 5674531.09 frames. ], batch size: 92, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:17:15,360 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:18:12,586 INFO [train.py:968] (1/2) Epoch 13, batch 15100, giga_loss[loss=0.2624, simple_loss=0.3396, pruned_loss=0.09265, over 28946.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3309, pruned_loss=0.08834, over 5684863.60 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3381, pruned_loss=0.09206, over 5759920.93 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3318, pruned_loss=0.08875, over 5671739.25 frames. ], batch size: 199, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:18:32,896 INFO [optim.py:369] (1/2) 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 18:19:09,689 INFO [train.py:968] (1/2) Epoch 13, batch 15150, giga_loss[loss=0.2456, simple_loss=0.3284, pruned_loss=0.08141, over 28899.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3318, pruned_loss=0.08939, over 5685103.23 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.338, pruned_loss=0.09206, over 5764497.65 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3325, pruned_loss=0.08966, over 5668186.08 frames. ], batch size: 199, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:19:21,225 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:19:24,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:20:02,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:20:08,540 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-06 18:20:09,372 INFO [train.py:968] (1/2) Epoch 13, batch 15200, giga_loss[loss=0.2199, simple_loss=0.3034, pruned_loss=0.06817, over 28873.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3295, pruned_loss=0.08768, over 5674355.69 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3377, pruned_loss=0.09188, over 5766689.88 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3303, pruned_loss=0.08801, over 5657827.98 frames. ], batch size: 227, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:20:29,969 INFO [optim.py:369] (1/2) 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 18:20:37,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9438, 2.1838, 2.0059, 2.0807], device='cuda:1'), covar=tensor([0.0585, 0.0256, 0.0235, 0.0624], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0116, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-06 18:21:10,776 INFO [train.py:968] (1/2) Epoch 13, batch 15250, giga_loss[loss=0.2504, simple_loss=0.3272, pruned_loss=0.0868, over 28167.00 frames. ], tot_loss[loss=0.251, simple_loss=0.329, pruned_loss=0.08647, over 5668088.83 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.338, pruned_loss=0.09204, over 5759864.99 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3292, pruned_loss=0.08646, over 5658269.53 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:22:10,309 INFO [train.py:968] (1/2) Epoch 13, batch 15300, libri_loss[loss=0.2596, simple_loss=0.3442, pruned_loss=0.08753, over 29477.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3289, pruned_loss=0.08712, over 5675604.78 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3374, pruned_loss=0.0918, over 5765242.66 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.329, pruned_loss=0.08706, over 5657776.65 frames. ], batch size: 85, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:22:33,442 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 15350, giga_loss[loss=0.3018, simple_loss=0.373, pruned_loss=0.1153, over 28924.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3302, pruned_loss=0.08763, over 5680503.33 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3378, pruned_loss=0.09216, over 5759198.33 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3299, pruned_loss=0.08718, over 5670867.17 frames. ], batch size: 227, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:23:17,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6746, 1.9857, 1.7508, 1.7689], device='cuda:1'), covar=tensor([0.1465, 0.2093, 0.1754, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0718, 0.0663, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 18:23:19,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:24:16,165 INFO [train.py:968] (1/2) Epoch 13, batch 15400, giga_loss[loss=0.2171, simple_loss=0.3042, pruned_loss=0.06501, over 28965.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3292, pruned_loss=0.08665, over 5689221.71 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3372, pruned_loss=0.09185, over 5759511.50 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3292, pruned_loss=0.08638, over 5678744.57 frames. ], batch size: 164, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:24:42,394 INFO [optim.py:369] (1/2) 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 18:24:44,663 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-06 18:24:57,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=562893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:25:01,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5932, 3.4319, 3.2149, 1.8468], device='cuda:1'), covar=tensor([0.0690, 0.0816, 0.0857, 0.2043], device='cuda:1'), in_proj_covar=tensor([0.1060, 0.0983, 0.0851, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 18:25:11,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 18:25:25,464 INFO [train.py:968] (1/2) Epoch 13, batch 15450, giga_loss[loss=0.2382, simple_loss=0.3187, pruned_loss=0.0789, over 28725.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3312, pruned_loss=0.08852, over 5691291.03 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3373, pruned_loss=0.09192, over 5760744.95 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.331, pruned_loss=0.08818, over 5680758.58 frames. ], batch size: 307, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:25:29,263 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=562917.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 18:26:20,933 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=562960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:26:25,947 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:26:26,912 INFO [train.py:968] (1/2) Epoch 13, batch 15500, giga_loss[loss=0.2541, simple_loss=0.3256, pruned_loss=0.09133, over 27536.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3291, pruned_loss=0.08731, over 5685593.12 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3369, pruned_loss=0.09171, over 5763237.58 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3291, pruned_loss=0.08713, over 5673090.77 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:26:37,296 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:26:56,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=562992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:27:20,958 INFO [train.py:968] (1/2) Epoch 13, batch 15550, giga_loss[loss=0.2176, simple_loss=0.3198, pruned_loss=0.05771, over 28912.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3298, pruned_loss=0.08633, over 5676077.87 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3371, pruned_loss=0.09196, over 5764447.55 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3294, pruned_loss=0.08578, over 5662208.88 frames. ], batch size: 164, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 18:27:49,983 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:27:55,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:28:25,919 INFO [train.py:968] (1/2) Epoch 13, batch 15600, giga_loss[loss=0.246, simple_loss=0.3079, pruned_loss=0.09205, over 24288.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3328, pruned_loss=0.08728, over 5667994.02 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3371, pruned_loss=0.09196, over 5765483.77 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3323, pruned_loss=0.08675, over 5654249.78 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:28:29,483 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:28:44,831 INFO [optim.py:369] (1/2) 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 18:29:25,468 INFO [train.py:968] (1/2) Epoch 13, batch 15650, giga_loss[loss=0.2614, simple_loss=0.3433, pruned_loss=0.08976, over 27979.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3347, pruned_loss=0.0878, over 5669833.52 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3372, pruned_loss=0.09194, over 5767246.15 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3342, pruned_loss=0.08733, over 5655978.02 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:29:41,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5000, 1.5452, 1.1868, 1.2455], device='cuda:1'), covar=tensor([0.0633, 0.0292, 0.0801, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0434, 0.0499, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 18:30:28,854 INFO [train.py:968] (1/2) Epoch 13, batch 15700, giga_loss[loss=0.279, simple_loss=0.3444, pruned_loss=0.1068, over 26721.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3345, pruned_loss=0.08833, over 5658273.96 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3368, pruned_loss=0.09174, over 5769151.73 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3344, pruned_loss=0.08808, over 5644353.06 frames. ], batch size: 555, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:30:47,147 INFO [optim.py:369] (1/2) 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 18:30:55,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4808, 1.8460, 1.7625, 1.3270], device='cuda:1'), covar=tensor([0.1809, 0.2323, 0.1488, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0683, 0.0881, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 18:31:26,130 INFO [train.py:968] (1/2) Epoch 13, batch 15750, giga_loss[loss=0.2019, simple_loss=0.2727, pruned_loss=0.06551, over 24588.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3328, pruned_loss=0.08761, over 5644611.34 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3368, pruned_loss=0.09174, over 5752119.98 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3327, pruned_loss=0.08729, over 5644538.88 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:32:13,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:32:30,936 INFO [train.py:968] (1/2) Epoch 13, batch 15800, giga_loss[loss=0.2886, simple_loss=0.3601, pruned_loss=0.1085, over 28497.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3311, pruned_loss=0.08646, over 5651698.57 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3367, pruned_loss=0.09165, over 5754394.93 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.331, pruned_loss=0.0862, over 5648156.67 frames. ], batch size: 336, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:32:41,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1557, 1.5295, 1.4376, 1.0904], device='cuda:1'), covar=tensor([0.1491, 0.2287, 0.1270, 0.1500], device='cuda:1'), in_proj_covar=tensor([0.0845, 0.0685, 0.0883, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 18:32:51,785 INFO [optim.py:369] (1/2) 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 18:33:05,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563292.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 18:33:31,499 INFO [train.py:968] (1/2) Epoch 13, batch 15850, libri_loss[loss=0.2164, simple_loss=0.2902, pruned_loss=0.07124, over 29642.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3291, pruned_loss=0.08582, over 5664704.57 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3364, pruned_loss=0.09154, over 5757702.02 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3291, pruned_loss=0.08561, over 5656994.55 frames. ], batch size: 73, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:33:47,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:34:13,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4164, 2.1788, 1.5131, 0.5806], device='cuda:1'), covar=tensor([0.4291, 0.2303, 0.3638, 0.5101], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1539, 0.1515, 0.1322], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 18:34:15,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:34:24,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:34:32,469 INFO [train.py:968] (1/2) Epoch 13, batch 15900, giga_loss[loss=0.285, simple_loss=0.3562, pruned_loss=0.1069, over 27648.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3306, pruned_loss=0.08639, over 5672890.81 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3361, pruned_loss=0.09135, over 5760982.12 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3308, pruned_loss=0.0863, over 5662082.29 frames. ], batch size: 474, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:34:55,759 INFO [optim.py:369] (1/2) 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 18:35:31,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3011, 4.1292, 3.8738, 1.8635], device='cuda:1'), covar=tensor([0.0550, 0.0734, 0.0769, 0.2142], device='cuda:1'), in_proj_covar=tensor([0.1065, 0.0983, 0.0859, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 18:35:38,621 INFO [train.py:968] (1/2) Epoch 13, batch 15950, giga_loss[loss=0.2546, simple_loss=0.3291, pruned_loss=0.09003, over 28951.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3322, pruned_loss=0.08728, over 5668175.81 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3364, pruned_loss=0.09155, over 5760420.64 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08698, over 5659061.54 frames. ], batch size: 186, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:36:05,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563435.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 18:36:09,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 18:36:44,921 INFO [train.py:968] (1/2) Epoch 13, batch 16000, giga_loss[loss=0.2542, simple_loss=0.34, pruned_loss=0.08422, over 28974.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3345, pruned_loss=0.08955, over 5664116.18 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3364, pruned_loss=0.09159, over 5761155.23 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3343, pruned_loss=0.08921, over 5654546.62 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:36:49,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563467.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 18:36:51,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6769, 2.3816, 1.9271, 1.4993], device='cuda:1'), covar=tensor([0.2825, 0.1476, 0.1772, 0.2147], device='cuda:1'), in_proj_covar=tensor([0.1754, 0.1629, 0.1585, 0.1688], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 18:37:05,340 INFO [optim.py:369] (1/2) 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 18:37:19,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:37:21,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:37:41,871 INFO [train.py:968] (1/2) Epoch 13, batch 16050, giga_loss[loss=0.2761, simple_loss=0.357, pruned_loss=0.09762, over 28026.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3386, pruned_loss=0.09172, over 5660214.93 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3361, pruned_loss=0.09152, over 5763715.40 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3388, pruned_loss=0.09149, over 5647986.59 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:37:52,386 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 16100, giga_loss[loss=0.2655, simple_loss=0.3492, pruned_loss=0.0909, over 28395.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3396, pruned_loss=0.09138, over 5659158.40 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3359, pruned_loss=0.09144, over 5765922.57 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.34, pruned_loss=0.09127, over 5646335.03 frames. ], batch size: 368, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:39:02,561 INFO [optim.py:369] (1/2) 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 18:39:05,978 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-06 18:39:43,609 INFO [train.py:968] (1/2) Epoch 13, batch 16150, giga_loss[loss=0.2156, simple_loss=0.3033, pruned_loss=0.06397, over 28937.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3397, pruned_loss=0.09152, over 5649395.93 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3359, pruned_loss=0.09143, over 5760276.48 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.34, pruned_loss=0.09144, over 5641534.39 frames. ], batch size: 145, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:40:02,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:40:11,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 18:40:41,323 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 18:40:53,859 INFO [train.py:968] (1/2) Epoch 13, batch 16200, giga_loss[loss=0.2519, simple_loss=0.3278, pruned_loss=0.08797, over 28468.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.337, pruned_loss=0.09012, over 5658266.39 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3355, pruned_loss=0.09117, over 5763249.80 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3377, pruned_loss=0.09027, over 5647215.21 frames. ], batch size: 336, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:41:13,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-06 18:41:17,066 INFO [optim.py:369] (1/2) 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 18:41:44,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=563701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:41:51,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3005, 1.2149, 3.9236, 3.1918], device='cuda:1'), covar=tensor([0.1643, 0.2709, 0.0468, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0676, 0.0597, 0.0862, 0.0777], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 18:42:03,448 INFO [train.py:968] (1/2) Epoch 13, batch 16250, giga_loss[loss=0.2468, simple_loss=0.3285, pruned_loss=0.08258, over 28921.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.336, pruned_loss=0.08922, over 5667124.37 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3357, pruned_loss=0.09127, over 5764661.00 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3363, pruned_loss=0.08923, over 5656196.53 frames. ], batch size: 112, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:42:26,153 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 16300, giga_loss[loss=0.2544, simple_loss=0.3334, pruned_loss=0.08773, over 28934.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3351, pruned_loss=0.08942, over 5669208.34 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3355, pruned_loss=0.09114, over 5766609.84 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3356, pruned_loss=0.08951, over 5657604.58 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:43:10,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:43:13,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:43:29,349 INFO [optim.py:369] (1/2) 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 18:43:51,525 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:44:05,829 INFO [train.py:968] (1/2) Epoch 13, batch 16350, giga_loss[loss=0.3041, simple_loss=0.3704, pruned_loss=0.1189, over 28641.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.333, pruned_loss=0.08935, over 5655336.71 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3353, pruned_loss=0.09108, over 5759751.73 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3335, pruned_loss=0.08942, over 5650375.23 frames. ], batch size: 307, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:44:42,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:44:44,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1165, 1.5046, 1.4080, 1.0451], device='cuda:1'), covar=tensor([0.1522, 0.2275, 0.1276, 0.1537], device='cuda:1'), in_proj_covar=tensor([0.0840, 0.0681, 0.0879, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 18:44:46,437 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 16400, giga_loss[loss=0.2481, simple_loss=0.3359, pruned_loss=0.08021, over 28889.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3316, pruned_loss=0.08848, over 5654632.20 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3353, pruned_loss=0.09113, over 5761842.89 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.332, pruned_loss=0.08847, over 5647733.51 frames. ], batch size: 145, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:45:22,487 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=563875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:23,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:27,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=563878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:31,440 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:45:55,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:46:03,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 18:46:04,060 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=563907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:46:05,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-06 18:46:12,037 INFO [train.py:968] (1/2) Epoch 13, batch 16450, giga_loss[loss=0.2495, simple_loss=0.3286, pruned_loss=0.08522, over 28956.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3308, pruned_loss=0.08665, over 5664533.37 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3353, pruned_loss=0.09116, over 5754941.31 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3311, pruned_loss=0.08659, over 5663342.15 frames. ], batch size: 175, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:46:36,001 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:46:40,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0377, 1.2337, 3.4518, 2.9460], device='cuda:1'), covar=tensor([0.1638, 0.2440, 0.0402, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0598, 0.0861, 0.0779], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 18:47:09,579 INFO [train.py:968] (1/2) Epoch 13, batch 16500, giga_loss[loss=0.2448, simple_loss=0.3457, pruned_loss=0.07196, over 29090.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3317, pruned_loss=0.08479, over 5674148.55 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.335, pruned_loss=0.09099, over 5757713.77 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3321, pruned_loss=0.08479, over 5669236.56 frames. ], batch size: 155, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:47:29,764 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=563989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:48:12,220 INFO [train.py:968] (1/2) Epoch 13, batch 16550, giga_loss[loss=0.2318, simple_loss=0.3228, pruned_loss=0.07039, over 28801.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.333, pruned_loss=0.08434, over 5671484.32 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3348, pruned_loss=0.09094, over 5756798.72 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3334, pruned_loss=0.08427, over 5666983.50 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:49:13,812 INFO [train.py:968] (1/2) Epoch 13, batch 16600, giga_loss[loss=0.2386, simple_loss=0.3249, pruned_loss=0.07614, over 29024.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3329, pruned_loss=0.08428, over 5670516.54 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.335, pruned_loss=0.09113, over 5758831.19 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.333, pruned_loss=0.08398, over 5664126.37 frames. ], batch size: 214, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:49:34,658 INFO [optim.py:369] (1/2) 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 18:50:18,641 INFO [train.py:968] (1/2) Epoch 13, batch 16650, giga_loss[loss=0.258, simple_loss=0.3337, pruned_loss=0.09118, over 28168.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3329, pruned_loss=0.08498, over 5667182.33 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3344, pruned_loss=0.09084, over 5761425.52 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3335, pruned_loss=0.08481, over 5657318.22 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:51:16,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-06 18:51:26,447 INFO [train.py:968] (1/2) Epoch 13, batch 16700, giga_loss[loss=0.2454, simple_loss=0.3313, pruned_loss=0.07974, over 28588.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3328, pruned_loss=0.085, over 5663174.67 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3344, pruned_loss=0.09092, over 5763632.43 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3332, pruned_loss=0.08465, over 5651055.60 frames. ], batch size: 242, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:51:50,681 INFO [optim.py:369] (1/2) 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 18:52:29,981 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:52:34,439 INFO [train.py:968] (1/2) Epoch 13, batch 16750, giga_loss[loss=0.2166, simple_loss=0.2888, pruned_loss=0.0722, over 24941.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3337, pruned_loss=0.0849, over 5665974.86 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3344, pruned_loss=0.09091, over 5766145.32 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.334, pruned_loss=0.08448, over 5651305.64 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:53:45,188 INFO [train.py:968] (1/2) Epoch 13, batch 16800, giga_loss[loss=0.2885, simple_loss=0.3666, pruned_loss=0.1052, over 28889.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3357, pruned_loss=0.08639, over 5665576.10 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3343, pruned_loss=0.09097, over 5769577.73 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3361, pruned_loss=0.0859, over 5648499.13 frames. ], batch size: 227, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:53:47,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:54:00,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:54:04,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-06 18:54:07,056 INFO [optim.py:369] (1/2) 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 18:54:25,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-06 18:54:33,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3696, 1.9917, 1.4028, 0.5805], device='cuda:1'), covar=tensor([0.3583, 0.1863, 0.2852, 0.4311], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1516, 0.1507, 0.1311], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 18:54:47,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:54:53,252 INFO [train.py:968] (1/2) Epoch 13, batch 16850, giga_loss[loss=0.2493, simple_loss=0.336, pruned_loss=0.08136, over 28182.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3377, pruned_loss=0.08662, over 5672204.17 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3337, pruned_loss=0.09058, over 5772826.42 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3386, pruned_loss=0.08648, over 5653671.07 frames. ], batch size: 77, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:55:06,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3067, 4.1050, 3.8905, 1.9463], device='cuda:1'), covar=tensor([0.0535, 0.0689, 0.0742, 0.2005], device='cuda:1'), in_proj_covar=tensor([0.1054, 0.0980, 0.0852, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 18:55:24,589 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:55:54,420 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 16900, giga_loss[loss=0.2179, simple_loss=0.3041, pruned_loss=0.06582, over 28765.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3385, pruned_loss=0.08789, over 5672849.73 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3341, pruned_loss=0.09088, over 5758422.50 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.339, pruned_loss=0.0874, over 5667333.11 frames. ], batch size: 99, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 18:56:07,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:56:23,474 INFO [optim.py:369] (1/2) 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 18:56:58,249 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:02,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:05,360 INFO [train.py:968] (1/2) Epoch 13, batch 16950, giga_loss[loss=0.2576, simple_loss=0.3304, pruned_loss=0.09235, over 29089.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3374, pruned_loss=0.08834, over 5672993.04 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3341, pruned_loss=0.09084, over 5760810.87 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3379, pruned_loss=0.0879, over 5664205.55 frames. ], batch size: 200, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:57:13,376 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:18,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:46,948 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:57:47,233 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 18:58:00,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:58:02,319 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:58:06,916 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 17000, giga_loss[loss=0.2321, simple_loss=0.3046, pruned_loss=0.07977, over 24919.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3357, pruned_loss=0.0869, over 5674586.92 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3342, pruned_loss=0.09097, over 5762529.40 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3361, pruned_loss=0.08641, over 5664977.79 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:58:47,968 INFO [optim.py:369] (1/2) 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 18:58:48,782 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:58:53,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 18:59:09,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:59:16,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:59:19,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 18:59:27,307 INFO [train.py:968] (1/2) Epoch 13, batch 17050, giga_loss[loss=0.248, simple_loss=0.3291, pruned_loss=0.08345, over 27685.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3349, pruned_loss=0.0863, over 5668909.62 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3347, pruned_loss=0.09125, over 5755290.75 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3348, pruned_loss=0.08552, over 5664758.87 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 18:59:44,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2950, 3.1846, 1.4516, 1.4826], device='cuda:1'), covar=tensor([0.0959, 0.0299, 0.0895, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0514, 0.0349, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 18:59:56,078 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 17100, giga_loss[loss=0.2306, simple_loss=0.3107, pruned_loss=0.07525, over 28668.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3359, pruned_loss=0.08695, over 5671110.62 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3348, pruned_loss=0.09144, over 5757048.85 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3357, pruned_loss=0.08611, over 5665335.70 frames. ], batch size: 92, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:00:55,120 INFO [optim.py:369] (1/2) 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 19:00:57,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:01:07,389 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564592.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 19:01:36,462 INFO [train.py:968] (1/2) Epoch 13, batch 17150, giga_loss[loss=0.2348, simple_loss=0.3249, pruned_loss=0.07237, over 28956.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3386, pruned_loss=0.08822, over 5666889.30 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3349, pruned_loss=0.0915, over 5755160.36 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3383, pruned_loss=0.08748, over 5663156.53 frames. ], batch size: 164, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:02:26,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9305, 2.0150, 1.4840, 1.6100], device='cuda:1'), covar=tensor([0.0803, 0.0588, 0.0971, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0436, 0.0504, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 19:02:34,399 INFO [train.py:968] (1/2) Epoch 13, batch 17200, giga_loss[loss=0.2367, simple_loss=0.3149, pruned_loss=0.07921, over 28888.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3367, pruned_loss=0.0884, over 5671466.97 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3347, pruned_loss=0.09136, over 5757474.83 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3368, pruned_loss=0.08788, over 5664443.21 frames. ], batch size: 186, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:02:39,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-06 19:03:00,229 INFO [optim.py:369] (1/2) 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 19:03:34,966 INFO [train.py:968] (1/2) Epoch 13, batch 17250, giga_loss[loss=0.2473, simple_loss=0.3305, pruned_loss=0.08211, over 28808.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3342, pruned_loss=0.08817, over 5669002.68 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3341, pruned_loss=0.09101, over 5760667.32 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3349, pruned_loss=0.088, over 5658162.24 frames. ], batch size: 119, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:03:36,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:03:53,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:03:57,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:04:18,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:04:34,227 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:04:39,526 INFO [train.py:968] (1/2) Epoch 13, batch 17300, giga_loss[loss=0.309, simple_loss=0.3656, pruned_loss=0.1262, over 26837.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3348, pruned_loss=0.08916, over 5661894.13 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3339, pruned_loss=0.09083, over 5764191.42 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3356, pruned_loss=0.08914, over 5647842.26 frames. ], batch size: 555, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:05:00,343 INFO [optim.py:369] (1/2) 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 19:05:39,443 INFO [train.py:968] (1/2) Epoch 13, batch 17350, giga_loss[loss=0.2995, simple_loss=0.38, pruned_loss=0.1095, over 28955.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3437, pruned_loss=0.09433, over 5668354.20 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3337, pruned_loss=0.09075, over 5765644.98 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3445, pruned_loss=0.0944, over 5655229.33 frames. ], batch size: 145, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:06:23,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:25,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:29,009 INFO [train.py:968] (1/2) Epoch 13, batch 17400, giga_loss[loss=0.3214, simple_loss=0.3876, pruned_loss=0.1276, over 28867.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3533, pruned_loss=0.09971, over 5671681.21 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3337, pruned_loss=0.0907, over 5764368.75 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.354, pruned_loss=0.09986, over 5662199.15 frames. ], batch size: 112, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:06:40,721 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:06:51,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:53,898 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=564892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:06:57,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=564895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:07:18,407 INFO [train.py:968] (1/2) Epoch 13, batch 17450, giga_loss[loss=0.3056, simple_loss=0.3713, pruned_loss=0.12, over 28629.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3555, pruned_loss=0.1021, over 5672502.75 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3337, pruned_loss=0.09075, over 5766112.27 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3566, pruned_loss=0.1024, over 5661402.88 frames. ], batch size: 92, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:07:26,967 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=564924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:08:01,164 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=564956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:08:07,898 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-06 19:08:08,104 INFO [train.py:968] (1/2) Epoch 13, batch 17500, giga_loss[loss=0.2352, simple_loss=0.3099, pruned_loss=0.08022, over 28908.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09975, over 5678913.41 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3337, pruned_loss=0.09072, over 5767473.02 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3505, pruned_loss=0.1001, over 5668178.39 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:08:10,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=564967.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 19:08:28,008 INFO [optim.py:369] (1/2) 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 19:08:42,518 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:08:55,398 INFO [train.py:968] (1/2) Epoch 13, batch 17550, giga_loss[loss=0.2452, simple_loss=0.3175, pruned_loss=0.08646, over 28314.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.342, pruned_loss=0.09627, over 5674984.13 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3338, pruned_loss=0.0907, over 5751456.43 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3429, pruned_loss=0.09673, over 5678740.71 frames. ], batch size: 368, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:09:01,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:03,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:13,527 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:29,047 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:09:42,267 INFO [train.py:968] (1/2) Epoch 13, batch 17600, giga_loss[loss=0.2367, simple_loss=0.3075, pruned_loss=0.08297, over 28923.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3346, pruned_loss=0.09331, over 5684210.62 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3336, pruned_loss=0.09062, over 5753573.25 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3356, pruned_loss=0.0938, over 5684295.78 frames. ], batch size: 227, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:10:01,013 INFO [optim.py:369] (1/2) 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 19:10:21,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565110.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 19:10:24,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 19:10:25,131 INFO [train.py:968] (1/2) Epoch 13, batch 17650, giga_loss[loss=0.1995, simple_loss=0.2755, pruned_loss=0.06178, over 29078.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3261, pruned_loss=0.08959, over 5687004.00 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3336, pruned_loss=0.09056, over 5754605.97 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3269, pruned_loss=0.09005, over 5685111.77 frames. ], batch size: 128, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:10:47,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-06 19:10:50,134 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 19:10:58,238 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 17700, giga_loss[loss=0.1877, simple_loss=0.2619, pruned_loss=0.05676, over 28502.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3195, pruned_loss=0.08636, over 5685542.84 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3338, pruned_loss=0.09061, over 5755093.38 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3198, pruned_loss=0.08662, over 5682784.71 frames. ], batch size: 60, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:11:26,139 INFO [optim.py:369] (1/2) 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 19:11:49,562 INFO [train.py:968] (1/2) Epoch 13, batch 17750, giga_loss[loss=0.212, simple_loss=0.2919, pruned_loss=0.06602, over 28928.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3164, pruned_loss=0.08491, over 5685859.03 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3341, pruned_loss=0.09076, over 5748588.07 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3161, pruned_loss=0.08489, over 5689018.83 frames. ], batch size: 227, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:12:18,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4935, 1.7014, 1.5894, 1.5400], device='cuda:1'), covar=tensor([0.1493, 0.1814, 0.2011, 0.1852], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0723, 0.0669, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 19:12:33,167 INFO [train.py:968] (1/2) Epoch 13, batch 17800, giga_loss[loss=0.2313, simple_loss=0.3078, pruned_loss=0.07737, over 28834.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3137, pruned_loss=0.08338, over 5681369.61 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3339, pruned_loss=0.09044, over 5744692.11 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.313, pruned_loss=0.08346, over 5684675.65 frames. ], batch size: 186, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:12:49,789 INFO [optim.py:369] (1/2) 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 19:12:50,665 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:13:11,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 19:13:14,537 INFO [train.py:968] (1/2) Epoch 13, batch 17850, giga_loss[loss=0.1979, simple_loss=0.2666, pruned_loss=0.06457, over 28343.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3119, pruned_loss=0.0828, over 5672389.78 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3346, pruned_loss=0.09067, over 5739722.78 frames. ], giga_tot_loss[loss=0.2373, simple_loss=0.3099, pruned_loss=0.08237, over 5678196.84 frames. ], batch size: 71, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:13:27,729 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:13:55,777 INFO [train.py:968] (1/2) Epoch 13, batch 17900, giga_loss[loss=0.2049, simple_loss=0.2833, pruned_loss=0.06326, over 28709.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.308, pruned_loss=0.08054, over 5692300.36 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3351, pruned_loss=0.09092, over 5743230.01 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3054, pruned_loss=0.07977, over 5692482.18 frames. ], batch size: 262, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:14:08,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:14:17,194 INFO [optim.py:369] (1/2) 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 19:14:36,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:14:40,741 INFO [train.py:968] (1/2) Epoch 13, batch 17950, giga_loss[loss=0.1868, simple_loss=0.2625, pruned_loss=0.0556, over 28645.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3049, pruned_loss=0.07933, over 5681068.10 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3351, pruned_loss=0.09094, over 5736106.00 frames. ], giga_tot_loss[loss=0.2298, simple_loss=0.3024, pruned_loss=0.07855, over 5687445.28 frames. ], batch size: 71, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:15:25,159 INFO [train.py:968] (1/2) Epoch 13, batch 18000, giga_loss[loss=0.2015, simple_loss=0.274, pruned_loss=0.06452, over 28890.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3024, pruned_loss=0.07854, over 5681411.11 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3353, pruned_loss=0.09099, over 5738632.59 frames. ], giga_tot_loss[loss=0.2277, simple_loss=0.2999, pruned_loss=0.07773, over 5683609.26 frames. ], batch size: 112, lr: 2.49e-03, grad_scale: 8.0
+2023-03-06 19:15:25,160 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 19:15:33,540 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 19:15:41,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:15:43,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:15:51,927 INFO [optim.py:369] (1/2) 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 19:16:07,419 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:11,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3056, 1.4299, 1.3508, 1.2854], device='cuda:1'), covar=tensor([0.2263, 0.2096, 0.1439, 0.1799], device='cuda:1'), in_proj_covar=tensor([0.1779, 0.1647, 0.1601, 0.1709], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 19:16:14,369 INFO [train.py:968] (1/2) Epoch 13, batch 18050, giga_loss[loss=0.2327, simple_loss=0.3022, pruned_loss=0.08154, over 28528.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.3003, pruned_loss=0.07696, over 5695111.36 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3356, pruned_loss=0.09102, over 5743191.88 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.2968, pruned_loss=0.07584, over 5690967.90 frames. ], batch size: 336, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:16:19,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:21,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:25,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=565527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:47,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:47,911 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:49,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:16:59,865 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-06 19:17:00,017 INFO [train.py:968] (1/2) Epoch 13, batch 18100, giga_loss[loss=0.2365, simple_loss=0.3013, pruned_loss=0.08587, over 27620.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.2977, pruned_loss=0.07588, over 5688934.71 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3362, pruned_loss=0.09111, over 5737034.40 frames. ], giga_tot_loss[loss=0.2215, simple_loss=0.2937, pruned_loss=0.07463, over 5689871.69 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:17:10,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-06 19:17:13,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:17:15,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7592, 1.8961, 1.4386, 1.4819], device='cuda:1'), covar=tensor([0.0741, 0.0466, 0.0912, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0435, 0.0503, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 19:17:16,133 INFO [optim.py:369] (1/2) 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 19:17:31,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7162, 1.8624, 1.3448, 1.4187], device='cuda:1'), covar=tensor([0.0861, 0.0635, 0.1011, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0434, 0.0503, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 19:17:46,161 INFO [train.py:968] (1/2) Epoch 13, batch 18150, giga_loss[loss=0.2836, simple_loss=0.3585, pruned_loss=0.1043, over 28596.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3032, pruned_loss=0.07907, over 5696050.49 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3364, pruned_loss=0.0912, over 5742544.55 frames. ], giga_tot_loss[loss=0.2268, simple_loss=0.2986, pruned_loss=0.07753, over 5690863.75 frames. ], batch size: 307, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:18:13,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5500, 2.2447, 1.6489, 0.6050], device='cuda:1'), covar=tensor([0.4722, 0.2304, 0.3205, 0.5170], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1510, 0.1487, 0.1296], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 19:18:31,839 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 18200, giga_loss[loss=0.3182, simple_loss=0.3848, pruned_loss=0.1258, over 29027.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3155, pruned_loss=0.08555, over 5699256.67 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.09156, over 5746270.21 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3107, pruned_loss=0.08378, over 5690276.09 frames. ], batch size: 128, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:18:37,863 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:18:39,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:18:52,241 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 13, batch 18250, giga_loss[loss=0.3533, simple_loss=0.4046, pruned_loss=0.151, over 27808.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.329, pruned_loss=0.09271, over 5702590.35 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3371, pruned_loss=0.0915, over 5750513.96 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3246, pruned_loss=0.09128, over 5690370.57 frames. ], batch size: 412, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:19:45,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:19:54,038 INFO [train.py:968] (1/2) Epoch 13, batch 18300, giga_loss[loss=0.3189, simple_loss=0.3924, pruned_loss=0.1227, over 28819.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.337, pruned_loss=0.09601, over 5696422.97 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3372, pruned_loss=0.09154, over 5741341.37 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3333, pruned_loss=0.09491, over 5693635.48 frames. ], batch size: 284, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:20:00,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5884, 2.1587, 1.5127, 0.9132], device='cuda:1'), covar=tensor([0.3931, 0.2274, 0.3554, 0.4080], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1519, 0.1500, 0.1301], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 19:20:15,089 INFO [optim.py:369] (1/2) 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 19:20:31,214 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=565805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:20:33,943 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=565808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:20:41,257 INFO [train.py:968] (1/2) Epoch 13, batch 18350, giga_loss[loss=0.3519, simple_loss=0.4004, pruned_loss=0.1517, over 26673.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3423, pruned_loss=0.09747, over 5693462.14 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3372, pruned_loss=0.09154, over 5741341.37 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3394, pruned_loss=0.09661, over 5691292.60 frames. ], batch size: 555, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:20:58,337 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=565837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:21:17,877 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=565861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:21:21,384 INFO [train.py:968] (1/2) Epoch 13, batch 18400, giga_loss[loss=0.2625, simple_loss=0.3305, pruned_loss=0.0972, over 23815.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.345, pruned_loss=0.09792, over 5690943.02 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3377, pruned_loss=0.09177, over 5745745.09 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3424, pruned_loss=0.09719, over 5683952.62 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:21:43,013 INFO [optim.py:369] (1/2) 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 19:21:55,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2757, 1.4084, 1.3128, 1.1924], device='cuda:1'), covar=tensor([0.1963, 0.2040, 0.1272, 0.1610], device='cuda:1'), in_proj_covar=tensor([0.1780, 0.1655, 0.1613, 0.1716], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 19:22:06,562 INFO [train.py:968] (1/2) Epoch 13, batch 18450, giga_loss[loss=0.2997, simple_loss=0.3701, pruned_loss=0.1146, over 28908.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3467, pruned_loss=0.09874, over 5682230.79 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3381, pruned_loss=0.09201, over 5738570.56 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3444, pruned_loss=0.09804, over 5682889.05 frames. ], batch size: 174, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:22:16,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1393, 1.1562, 3.4803, 2.9139], device='cuda:1'), covar=tensor([0.1655, 0.2729, 0.0434, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0668, 0.0591, 0.0851, 0.0771], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 19:22:21,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2838, 4.0750, 3.8744, 1.8633], device='cuda:1'), covar=tensor([0.0570, 0.0687, 0.0697, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.1047, 0.0978, 0.0851, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 19:22:47,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9086, 3.6960, 3.5105, 1.7142], device='cuda:1'), covar=tensor([0.0661, 0.0789, 0.0766, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1049, 0.0980, 0.0852, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 19:22:49,181 INFO [train.py:968] (1/2) Epoch 13, batch 18500, giga_loss[loss=0.3653, simple_loss=0.4089, pruned_loss=0.1608, over 26503.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3493, pruned_loss=0.1008, over 5683226.44 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3383, pruned_loss=0.09207, over 5734136.62 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3475, pruned_loss=0.1004, over 5685983.89 frames. ], batch size: 555, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:23:12,125 INFO [optim.py:369] (1/2) 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 19:23:12,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-06 19:23:34,887 INFO [train.py:968] (1/2) Epoch 13, batch 18550, giga_loss[loss=0.2751, simple_loss=0.3479, pruned_loss=0.1011, over 28768.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3526, pruned_loss=0.1032, over 5689622.42 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3386, pruned_loss=0.09212, over 5736467.95 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.351, pruned_loss=0.1029, over 5689105.49 frames. ], batch size: 112, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:24:15,592 INFO [train.py:968] (1/2) Epoch 13, batch 18600, giga_loss[loss=0.2772, simple_loss=0.3563, pruned_loss=0.09901, over 28721.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3558, pruned_loss=0.1044, over 5688049.97 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3396, pruned_loss=0.09236, over 5730698.67 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3542, pruned_loss=0.1044, over 5691168.07 frames. ], batch size: 284, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:24:28,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4393, 2.7579, 2.0546, 1.8870], device='cuda:1'), covar=tensor([0.2068, 0.1675, 0.1952, 0.2168], device='cuda:1'), in_proj_covar=tensor([0.1770, 0.1645, 0.1604, 0.1708], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 19:24:36,337 INFO [optim.py:369] (1/2) 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 19:24:55,882 INFO [train.py:968] (1/2) Epoch 13, batch 18650, giga_loss[loss=0.29, simple_loss=0.3692, pruned_loss=0.1054, over 28914.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3575, pruned_loss=0.1043, over 5699905.73 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3401, pruned_loss=0.09255, over 5734372.17 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3563, pruned_loss=0.1045, over 5698425.66 frames. ], batch size: 199, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:25:05,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=566128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:25:15,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 19:25:37,191 INFO [train.py:968] (1/2) Epoch 13, batch 18700, giga_loss[loss=0.2772, simple_loss=0.3587, pruned_loss=0.09786, over 28872.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3586, pruned_loss=0.1042, over 5696527.33 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3404, pruned_loss=0.09266, over 5730311.99 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3578, pruned_loss=0.1046, over 5698604.26 frames. ], batch size: 186, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:25:48,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8136, 1.1691, 3.3273, 2.8065], device='cuda:1'), covar=tensor([0.1816, 0.2651, 0.0460, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0672, 0.0595, 0.0855, 0.0778], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 19:25:56,024 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 18750, giga_loss[loss=0.317, simple_loss=0.3844, pruned_loss=0.1249, over 28936.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3597, pruned_loss=0.1042, over 5689349.72 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3409, pruned_loss=0.09292, over 5726739.29 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3588, pruned_loss=0.1044, over 5693899.46 frames. ], batch size: 106, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:26:36,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=566236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:26:52,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3444, 1.6316, 1.3005, 1.3537], device='cuda:1'), covar=tensor([0.2182, 0.2144, 0.2286, 0.2086], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.0978, 0.1176, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 19:27:02,975 INFO [train.py:968] (1/2) Epoch 13, batch 18800, giga_loss[loss=0.3044, simple_loss=0.374, pruned_loss=0.1174, over 29008.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3586, pruned_loss=0.1027, over 5689347.53 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3409, pruned_loss=0.09292, over 5726739.29 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3579, pruned_loss=0.1028, over 5692888.65 frames. ], batch size: 106, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:27:07,618 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=566271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:27:09,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=566274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:27:21,112 INFO [optim.py:369] (1/2) 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 19:27:31,800 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=566303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:27:40,602 INFO [train.py:968] (1/2) Epoch 13, batch 18850, giga_loss[loss=0.3095, simple_loss=0.3767, pruned_loss=0.1211, over 27581.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3569, pruned_loss=0.1006, over 5697269.14 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3418, pruned_loss=0.09322, over 5722562.33 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.356, pruned_loss=0.1007, over 5702417.03 frames. ], batch size: 472, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:28:11,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1871, 1.3204, 1.1461, 1.0924], device='cuda:1'), covar=tensor([0.1710, 0.1519, 0.1313, 0.1556], device='cuda:1'), in_proj_covar=tensor([0.1771, 0.1646, 0.1615, 0.1712], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 19:28:21,606 INFO [train.py:968] (1/2) Epoch 13, batch 18900, giga_loss[loss=0.2476, simple_loss=0.3353, pruned_loss=0.07992, over 28394.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3577, pruned_loss=0.1021, over 5685244.32 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3417, pruned_loss=0.09325, over 5714832.15 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3573, pruned_loss=0.1023, over 5695730.99 frames. ], batch size: 60, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:28:36,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=566379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:28:38,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=566382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:28:38,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3578, 1.5805, 1.3419, 1.0102], device='cuda:1'), covar=tensor([0.2046, 0.2029, 0.2160, 0.1867], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.0979, 0.1178, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 19:28:43,848 INFO [optim.py:369] (1/2) 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 19:28:45,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3759, 3.6445, 1.3926, 1.5396], device='cuda:1'), covar=tensor([0.0907, 0.0230, 0.0879, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0507, 0.0346, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 19:29:01,568 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 18950, giga_loss[loss=0.299, simple_loss=0.3639, pruned_loss=0.1171, over 28864.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3589, pruned_loss=0.1054, over 5680921.21 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3418, pruned_loss=0.09323, over 5718537.93 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3592, pruned_loss=0.106, over 5684749.12 frames. ], batch size: 93, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:29:42,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=566459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:29:48,348 INFO [train.py:968] (1/2) Epoch 13, batch 19000, giga_loss[loss=0.2787, simple_loss=0.3326, pruned_loss=0.1124, over 23881.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3601, pruned_loss=0.108, over 5681041.83 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3419, pruned_loss=0.09315, over 5721470.98 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3605, pruned_loss=0.1087, over 5680930.81 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:30:05,785 INFO [optim.py:369] (1/2) 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 19:30:29,657 INFO [train.py:968] (1/2) Epoch 13, batch 19050, libri_loss[loss=0.2622, simple_loss=0.3397, pruned_loss=0.09239, over 29539.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.359, pruned_loss=0.1082, over 5690500.46 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3417, pruned_loss=0.09311, over 5724086.81 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3597, pruned_loss=0.109, over 5687557.35 frames. ], batch size: 78, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:31:10,953 INFO [train.py:968] (1/2) Epoch 13, batch 19100, giga_loss[loss=0.2854, simple_loss=0.3538, pruned_loss=0.1085, over 28945.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.357, pruned_loss=0.1073, over 5692768.67 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3426, pruned_loss=0.09358, over 5718510.39 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3571, pruned_loss=0.1078, over 5694190.98 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:31:27,163 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-06 19:31:30,162 INFO [optim.py:369] (1/2) 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 19:31:48,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-06 19:31:53,250 INFO [train.py:968] (1/2) Epoch 13, batch 19150, giga_loss[loss=0.2811, simple_loss=0.3549, pruned_loss=0.1037, over 28933.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3565, pruned_loss=0.1069, over 5685660.95 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3423, pruned_loss=0.09327, over 5722414.11 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3574, pruned_loss=0.1081, over 5682250.61 frames. ], batch size: 213, lr: 2.49e-03, grad_scale: 2.0
+2023-03-06 19:32:36,248 INFO [train.py:968] (1/2) Epoch 13, batch 19200, giga_loss[loss=0.32, simple_loss=0.3711, pruned_loss=0.1344, over 23814.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3547, pruned_loss=0.1049, over 5686934.88 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3423, pruned_loss=0.09327, over 5722414.11 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3554, pruned_loss=0.1059, over 5684280.57 frames. ], batch size: 705, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:33:00,473 INFO [optim.py:369] (1/2) 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 19:33:21,885 INFO [train.py:968] (1/2) Epoch 13, batch 19250, giga_loss[loss=0.2355, simple_loss=0.3219, pruned_loss=0.07454, over 28974.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3504, pruned_loss=0.1019, over 5688398.62 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3425, pruned_loss=0.09339, over 5726259.46 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.351, pruned_loss=0.1027, over 5682306.68 frames. ], batch size: 164, lr: 2.49e-03, grad_scale: 4.0
+2023-03-06 19:34:08,652 INFO [train.py:968] (1/2) Epoch 13, batch 19300, giga_loss[loss=0.2537, simple_loss=0.3255, pruned_loss=0.09092, over 28277.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.344, pruned_loss=0.09827, over 5676946.21 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3425, pruned_loss=0.0934, over 5719372.10 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3446, pruned_loss=0.09904, over 5677974.98 frames. ], batch size: 368, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:34:30,927 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 19350, giga_loss[loss=0.2383, simple_loss=0.311, pruned_loss=0.08281, over 28537.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3388, pruned_loss=0.09539, over 5666612.70 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3426, pruned_loss=0.09332, over 5715061.25 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3391, pruned_loss=0.0962, over 5670254.93 frames. ], batch size: 307, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:35:12,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=566834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:35:41,636 INFO [train.py:968] (1/2) Epoch 13, batch 19400, giga_loss[loss=0.2613, simple_loss=0.3415, pruned_loss=0.09055, over 28695.00 frames. ], tot_loss[loss=0.262, simple_loss=0.336, pruned_loss=0.09399, over 5676114.52 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3426, pruned_loss=0.09331, over 5711603.92 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3362, pruned_loss=0.09467, over 5681405.17 frames. ], batch size: 242, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:36:04,076 INFO [optim.py:369] (1/2) 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 19:36:24,194 INFO [train.py:968] (1/2) Epoch 13, batch 19450, giga_loss[loss=0.2185, simple_loss=0.3097, pruned_loss=0.06363, over 28738.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3375, pruned_loss=0.09431, over 5685288.62 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3425, pruned_loss=0.09325, over 5715564.23 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3375, pruned_loss=0.09492, over 5685300.28 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:37:03,074 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=566956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:37:08,883 INFO [train.py:968] (1/2) Epoch 13, batch 19500, giga_loss[loss=0.2724, simple_loss=0.3471, pruned_loss=0.09882, over 28712.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3368, pruned_loss=0.09365, over 5684625.04 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3426, pruned_loss=0.09338, over 5708651.59 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3367, pruned_loss=0.09401, over 5690768.95 frames. ], batch size: 284, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:37:19,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=566977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:37:21,322 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:37:43,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:37:48,301 INFO [train.py:968] (1/2) Epoch 13, batch 19550, libri_loss[loss=0.2823, simple_loss=0.3715, pruned_loss=0.09652, over 29766.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3353, pruned_loss=0.09276, over 5699825.59 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.343, pruned_loss=0.09339, over 5713746.60 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3345, pruned_loss=0.09304, over 5699609.96 frames. ], batch size: 87, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:38:28,023 INFO [train.py:968] (1/2) Epoch 13, batch 19600, giga_loss[loss=0.2307, simple_loss=0.3065, pruned_loss=0.07747, over 29049.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3336, pruned_loss=0.09206, over 5702507.56 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3437, pruned_loss=0.09342, over 5708004.45 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3322, pruned_loss=0.0922, over 5707678.98 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:38:36,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-06 19:38:48,439 INFO [optim.py:369] (1/2) 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 19:39:08,365 INFO [train.py:968] (1/2) Epoch 13, batch 19650, giga_loss[loss=0.2341, simple_loss=0.3051, pruned_loss=0.08156, over 28876.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3312, pruned_loss=0.09092, over 5707424.85 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3445, pruned_loss=0.09367, over 5708002.68 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3293, pruned_loss=0.0908, over 5711345.45 frames. ], batch size: 99, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:39:24,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8711, 1.8216, 1.8404, 1.6419], device='cuda:1'), covar=tensor([0.1593, 0.2228, 0.2143, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0731, 0.0676, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 19:39:48,748 INFO [train.py:968] (1/2) Epoch 13, batch 19700, giga_loss[loss=0.2509, simple_loss=0.3239, pruned_loss=0.08897, over 28785.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3297, pruned_loss=0.09046, over 5713690.65 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3447, pruned_loss=0.09384, over 5713808.69 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3275, pruned_loss=0.09014, over 5711733.20 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:39:51,841 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:40:08,645 INFO [optim.py:369] (1/2) 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 19:40:27,261 INFO [train.py:968] (1/2) Epoch 13, batch 19750, giga_loss[loss=0.3264, simple_loss=0.3714, pruned_loss=0.1407, over 24333.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3295, pruned_loss=0.09065, over 5709751.30 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3459, pruned_loss=0.09441, over 5709515.16 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3261, pruned_loss=0.08973, over 5712163.65 frames. ], batch size: 705, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:41:05,398 INFO [train.py:968] (1/2) Epoch 13, batch 19800, libri_loss[loss=0.3247, simple_loss=0.4057, pruned_loss=0.1219, over 19315.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3275, pruned_loss=0.08949, over 5711285.86 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3465, pruned_loss=0.09463, over 5706727.26 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3235, pruned_loss=0.08838, over 5716840.56 frames. ], batch size: 187, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:41:26,089 INFO [optim.py:369] (1/2) 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 19:41:26,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:41:45,311 INFO [train.py:968] (1/2) Epoch 13, batch 19850, giga_loss[loss=0.2396, simple_loss=0.3204, pruned_loss=0.07941, over 27853.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3261, pruned_loss=0.08881, over 5711800.28 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3472, pruned_loss=0.09468, over 5711119.63 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3218, pruned_loss=0.08771, over 5712440.05 frames. ], batch size: 412, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:41:57,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:42:16,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2881, 1.6011, 1.4495, 1.6003], device='cuda:1'), covar=tensor([0.0812, 0.0323, 0.0315, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0092], device='cuda:1')
+2023-03-06 19:42:25,504 INFO [train.py:968] (1/2) Epoch 13, batch 19900, giga_loss[loss=0.2215, simple_loss=0.2942, pruned_loss=0.07441, over 28974.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3243, pruned_loss=0.08745, over 5725078.68 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3471, pruned_loss=0.0945, over 5717222.67 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.32, pruned_loss=0.08649, over 5720403.05 frames. ], batch size: 106, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:42:46,508 INFO [optim.py:369] (1/2) 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 19:43:06,444 INFO [train.py:968] (1/2) Epoch 13, batch 19950, libri_loss[loss=0.2326, simple_loss=0.3237, pruned_loss=0.07076, over 29557.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3221, pruned_loss=0.08684, over 5726577.54 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.347, pruned_loss=0.09438, over 5718193.56 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3188, pruned_loss=0.08617, over 5721991.40 frames. ], batch size: 79, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 19:43:26,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1739, 2.3662, 2.1474, 1.8374], device='cuda:1'), covar=tensor([0.2097, 0.1741, 0.1534, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.1766, 0.1640, 0.1612, 0.1717], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 19:43:45,325 INFO [train.py:968] (1/2) Epoch 13, batch 20000, giga_loss[loss=0.2448, simple_loss=0.3169, pruned_loss=0.08631, over 29014.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3218, pruned_loss=0.08642, over 5726366.38 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3481, pruned_loss=0.09476, over 5710058.68 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3175, pruned_loss=0.08537, over 5730168.16 frames. ], batch size: 106, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:43:50,142 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:43:53,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:43:55,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=567477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:44:07,501 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:44:26,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9783, 1.2020, 2.7153, 2.6380], device='cuda:1'), covar=tensor([0.1345, 0.2168, 0.0531, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0672, 0.0592, 0.0854, 0.0774], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 19:44:30,151 INFO [train.py:968] (1/2) Epoch 13, batch 20050, giga_loss[loss=0.2505, simple_loss=0.3211, pruned_loss=0.0899, over 28682.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3262, pruned_loss=0.08961, over 5714589.19 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3481, pruned_loss=0.09469, over 5710322.48 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3226, pruned_loss=0.08877, over 5717525.78 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:44:58,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:45:13,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-06 19:45:20,191 INFO [train.py:968] (1/2) Epoch 13, batch 20100, giga_loss[loss=0.3325, simple_loss=0.3916, pruned_loss=0.1366, over 28761.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3339, pruned_loss=0.09456, over 5714801.04 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3487, pruned_loss=0.09513, over 5712314.50 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3302, pruned_loss=0.09345, over 5715318.57 frames. ], batch size: 199, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:45:48,770 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 20150, libri_loss[loss=0.2789, simple_loss=0.3591, pruned_loss=0.09938, over 29570.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3442, pruned_loss=0.1019, over 5699633.88 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3492, pruned_loss=0.09533, over 5716109.25 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3406, pruned_loss=0.1009, over 5695961.39 frames. ], batch size: 78, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:46:50,939 INFO [train.py:968] (1/2) Epoch 13, batch 20200, giga_loss[loss=0.281, simple_loss=0.3587, pruned_loss=0.1017, over 28720.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.348, pruned_loss=0.1028, over 5702343.87 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3498, pruned_loss=0.09554, over 5722771.88 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3444, pruned_loss=0.1021, over 5692767.19 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:46:54,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:46:55,350 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:47:13,158 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:47:15,257 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:47:35,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7147, 1.8056, 1.7556, 1.6541], device='cuda:1'), covar=tensor([0.1676, 0.1502, 0.1159, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.1758, 0.1640, 0.1616, 0.1719], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 19:47:35,833 INFO [train.py:968] (1/2) Epoch 13, batch 20250, libri_loss[loss=0.2791, simple_loss=0.3654, pruned_loss=0.09638, over 26038.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.353, pruned_loss=0.1049, over 5686797.36 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3502, pruned_loss=0.09558, over 5723843.86 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3498, pruned_loss=0.1046, over 5677199.04 frames. ], batch size: 136, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:47:38,862 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:10,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5383, 2.0704, 1.5047, 0.7446], device='cuda:1'), covar=tensor([0.3629, 0.2139, 0.2845, 0.4252], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1505, 0.1497, 0.1297], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 19:48:19,586 INFO [train.py:968] (1/2) Epoch 13, batch 20300, giga_loss[loss=0.3305, simple_loss=0.3974, pruned_loss=0.1318, over 28663.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3576, pruned_loss=0.1073, over 5672298.04 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3505, pruned_loss=0.09557, over 5716593.79 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3549, pruned_loss=0.1074, over 5669325.71 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:48:23,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:32,188 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:40,115 INFO [optim.py:369] (1/2) 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 19:48:44,481 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:56,796 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:58,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=567813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:48:59,783 INFO [train.py:968] (1/2) Epoch 13, batch 20350, giga_loss[loss=0.2397, simple_loss=0.3238, pruned_loss=0.07783, over 28965.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3611, pruned_loss=0.1097, over 5672822.19 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3505, pruned_loss=0.09555, over 5718679.32 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3592, pruned_loss=0.11, over 5667907.52 frames. ], batch size: 227, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:49:24,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=567842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:49:27,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=567846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:49:29,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4349, 1.7408, 1.3581, 1.3698], device='cuda:1'), covar=tensor([0.2499, 0.2303, 0.2501, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.1330, 0.0977, 0.1174, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 19:49:44,632 INFO [train.py:968] (1/2) Epoch 13, batch 20400, giga_loss[loss=0.262, simple_loss=0.3494, pruned_loss=0.08736, over 28961.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3547, pruned_loss=0.1047, over 5680033.99 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3506, pruned_loss=0.09564, over 5723358.55 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3532, pruned_loss=0.1052, over 5671191.96 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:50:03,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=567888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:50:06,246 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 20450, giga_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08759, over 29105.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3531, pruned_loss=0.1029, over 5698398.44 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3505, pruned_loss=0.09564, over 5727424.17 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.352, pruned_loss=0.1035, over 5686630.35 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:50:38,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5754, 1.7718, 1.5267, 1.2796], device='cuda:1'), covar=tensor([0.2329, 0.1890, 0.1709, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1747, 0.1631, 0.1613, 0.1707], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 19:51:08,341 INFO [train.py:968] (1/2) Epoch 13, batch 20500, giga_loss[loss=0.2587, simple_loss=0.3384, pruned_loss=0.08953, over 28776.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3518, pruned_loss=0.1016, over 5690773.53 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3504, pruned_loss=0.09556, over 5722508.66 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.351, pruned_loss=0.1022, over 5685292.65 frames. ], batch size: 119, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:51:29,772 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=567989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:51:31,515 INFO [optim.py:369] (1/2) 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 19:51:32,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=567992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:51:48,457 INFO [train.py:968] (1/2) Epoch 13, batch 20550, giga_loss[loss=0.2947, simple_loss=0.3635, pruned_loss=0.1129, over 28939.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.353, pruned_loss=0.1017, over 5699006.63 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3503, pruned_loss=0.09555, over 5729563.08 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3526, pruned_loss=0.1024, over 5686978.31 frames. ], batch size: 213, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:51:55,098 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:52:13,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:52:29,845 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:52:31,082 INFO [train.py:968] (1/2) Epoch 13, batch 20600, giga_loss[loss=0.2863, simple_loss=0.3614, pruned_loss=0.1056, over 28521.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3543, pruned_loss=0.1031, over 5701815.93 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3498, pruned_loss=0.09536, over 5732586.23 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3546, pruned_loss=0.1041, over 5688610.45 frames. ], batch size: 85, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:52:55,507 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 20650, giga_loss[loss=0.2898, simple_loss=0.3626, pruned_loss=0.1085, over 28915.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3561, pruned_loss=0.1044, over 5706697.13 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3499, pruned_loss=0.09537, over 5734650.19 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3563, pruned_loss=0.1054, over 5693479.36 frames. ], batch size: 186, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:53:20,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 19:53:40,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:53:48,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:53:56,613 INFO [train.py:968] (1/2) Epoch 13, batch 20700, giga_loss[loss=0.2691, simple_loss=0.3447, pruned_loss=0.09672, over 28581.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3584, pruned_loss=0.1066, over 5697652.02 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3502, pruned_loss=0.09534, over 5740990.98 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3587, pruned_loss=0.1078, over 5679828.80 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:54:03,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:54:15,368 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:54:18,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:54:20,803 INFO [optim.py:369] (1/2) 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 19:54:40,566 INFO [train.py:968] (1/2) Epoch 13, batch 20750, giga_loss[loss=0.2759, simple_loss=0.347, pruned_loss=0.1024, over 28874.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3587, pruned_loss=0.1076, over 5692320.95 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3504, pruned_loss=0.09554, over 5734612.00 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3589, pruned_loss=0.1085, over 5683145.75 frames. ], batch size: 112, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:54:44,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:19,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:20,155 INFO [train.py:968] (1/2) Epoch 13, batch 20800, giga_loss[loss=0.2957, simple_loss=0.3679, pruned_loss=0.1118, over 28868.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3582, pruned_loss=0.1067, over 5700495.56 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3507, pruned_loss=0.09555, over 5735896.54 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3581, pruned_loss=0.1076, over 5691799.83 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:55:33,337 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:38,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:41,814 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:55:48,142 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:50,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:56,568 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:55:59,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:00,272 INFO [train.py:968] (1/2) Epoch 13, batch 20850, giga_loss[loss=0.2595, simple_loss=0.3473, pruned_loss=0.08586, over 28969.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3583, pruned_loss=0.1057, over 5703575.34 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3513, pruned_loss=0.09597, over 5739878.80 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3579, pruned_loss=0.1063, over 5692132.15 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 19:56:02,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:05,620 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:13,073 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:18,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4273, 2.4654, 1.7676, 1.9552], device='cuda:1'), covar=tensor([0.0769, 0.0590, 0.0906, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0429, 0.0499, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 19:56:25,489 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:56:39,228 INFO [train.py:968] (1/2) Epoch 13, batch 20900, libri_loss[loss=0.2913, simple_loss=0.3618, pruned_loss=0.1104, over 29583.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3586, pruned_loss=0.1046, over 5706330.72 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3517, pruned_loss=0.09608, over 5740857.46 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3581, pruned_loss=0.1052, over 5694672.35 frames. ], batch size: 79, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:57:02,783 INFO [optim.py:369] (1/2) 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 19:57:12,763 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:14,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:18,753 INFO [train.py:968] (1/2) Epoch 13, batch 20950, libri_loss[loss=0.2748, simple_loss=0.3428, pruned_loss=0.1034, over 29668.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3569, pruned_loss=0.1034, over 5701398.67 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3516, pruned_loss=0.0961, over 5738166.15 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3567, pruned_loss=0.1042, over 5692966.49 frames. ], batch size: 69, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:57:35,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:35,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:57:56,926 INFO [train.py:968] (1/2) Epoch 13, batch 21000, giga_loss[loss=0.277, simple_loss=0.3483, pruned_loss=0.1028, over 28906.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3543, pruned_loss=0.102, over 5712420.74 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3515, pruned_loss=0.09596, over 5740630.85 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3544, pruned_loss=0.1029, over 5702906.81 frames. ], batch size: 285, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:57:56,926 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 19:58:04,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2996, 1.7676, 1.5971, 1.1482], device='cuda:1'), covar=tensor([0.1766, 0.2700, 0.1575, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.0844, 0.0688, 0.0885, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 19:58:05,934 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 19:58:27,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6731, 3.7186, 1.8049, 1.8421], device='cuda:1'), covar=tensor([0.0829, 0.0234, 0.0781, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0504, 0.0343, 0.0391], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0030, 0.0021, 0.0026], device='cuda:1')
+2023-03-06 19:58:28,703 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 21050, libri_loss[loss=0.2631, simple_loss=0.349, pruned_loss=0.08867, over 29524.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3519, pruned_loss=0.1008, over 5710261.54 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3516, pruned_loss=0.09592, over 5736907.69 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3519, pruned_loss=0.1017, over 5705423.15 frames. ], batch size: 84, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:59:15,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8446, 5.6015, 5.2983, 2.7075], device='cuda:1'), covar=tensor([0.0391, 0.0618, 0.0692, 0.1756], device='cuda:1'), in_proj_covar=tensor([0.1069, 0.0994, 0.0864, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 19:59:21,733 INFO [train.py:968] (1/2) Epoch 13, batch 21100, giga_loss[loss=0.2668, simple_loss=0.3364, pruned_loss=0.0986, over 28868.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3501, pruned_loss=0.1004, over 5711178.90 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3511, pruned_loss=0.09568, over 5739512.29 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3505, pruned_loss=0.1013, over 5704693.26 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 19:59:35,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:59:38,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 19:59:46,097 INFO [optim.py:369] (1/2) 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 19:59:47,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1615, 1.2369, 3.8797, 3.1316], device='cuda:1'), covar=tensor([0.1720, 0.2694, 0.0403, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0677, 0.0593, 0.0859, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 20:00:00,934 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 21150, giga_loss[loss=0.3058, simple_loss=0.377, pruned_loss=0.1173, over 29132.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3516, pruned_loss=0.1018, over 5711174.27 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3509, pruned_loss=0.09553, over 5741098.45 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3521, pruned_loss=0.1028, over 5704234.48 frames. ], batch size: 128, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:00:35,816 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:00:44,242 INFO [train.py:968] (1/2) Epoch 13, batch 21200, giga_loss[loss=0.2663, simple_loss=0.3413, pruned_loss=0.09562, over 28726.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3521, pruned_loss=0.1017, over 5709033.07 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3511, pruned_loss=0.09564, over 5735993.03 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3524, pruned_loss=0.1026, over 5706508.94 frames. ], batch size: 119, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:01:02,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=568684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:01:09,371 INFO [optim.py:369] (1/2) 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 20:01:26,157 INFO [train.py:968] (1/2) Epoch 13, batch 21250, giga_loss[loss=0.2858, simple_loss=0.3642, pruned_loss=0.1037, over 28751.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3505, pruned_loss=0.1, over 5707093.58 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.351, pruned_loss=0.0956, over 5737671.76 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3508, pruned_loss=0.1008, over 5703274.54 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:02:10,900 INFO [train.py:968] (1/2) Epoch 13, batch 21300, giga_loss[loss=0.2851, simple_loss=0.3556, pruned_loss=0.1073, over 28921.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3502, pruned_loss=0.09956, over 5718252.16 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3509, pruned_loss=0.09565, over 5740463.64 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3504, pruned_loss=0.1002, over 5712120.27 frames. ], batch size: 136, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:02:35,093 INFO [optim.py:369] (1/2) 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 20:02:37,855 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:02:40,013 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 21350, giga_loss[loss=0.2456, simple_loss=0.3175, pruned_loss=0.08687, over 28433.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3493, pruned_loss=0.09915, over 5726021.83 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3513, pruned_loss=0.09592, over 5742349.83 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3492, pruned_loss=0.09947, over 5719462.65 frames. ], batch size: 71, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:03:03,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=568827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:04,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:05,649 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=568831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:27,770 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 21400, giga_loss[loss=0.2762, simple_loss=0.3508, pruned_loss=0.1009, over 28542.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3467, pruned_loss=0.09823, over 5722825.42 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.351, pruned_loss=0.09579, over 5744001.93 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3469, pruned_loss=0.09861, over 5716048.60 frames. ], batch size: 336, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:03:40,099 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:03:50,295 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=568887.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 20:03:55,953 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:04:11,393 INFO [train.py:968] (1/2) Epoch 13, batch 21450, giga_loss[loss=0.3226, simple_loss=0.3947, pruned_loss=0.1253, over 28759.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3454, pruned_loss=0.0979, over 5716264.96 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3512, pruned_loss=0.09604, over 5738193.71 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3452, pruned_loss=0.09805, over 5714619.44 frames. ], batch size: 284, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:04:49,267 INFO [train.py:968] (1/2) Epoch 13, batch 21500, giga_loss[loss=0.2943, simple_loss=0.3633, pruned_loss=0.1127, over 28599.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3454, pruned_loss=0.09799, over 5725526.56 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3513, pruned_loss=0.09601, over 5743372.34 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.345, pruned_loss=0.09818, over 5719054.64 frames. ], batch size: 262, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:05:13,961 INFO [optim.py:369] (1/2) 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 20:05:20,131 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 20:05:30,752 INFO [train.py:968] (1/2) Epoch 13, batch 21550, giga_loss[loss=0.3157, simple_loss=0.3727, pruned_loss=0.1293, over 27877.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3445, pruned_loss=0.09821, over 5721974.71 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3512, pruned_loss=0.09597, over 5744783.39 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3442, pruned_loss=0.09841, over 5715583.82 frames. ], batch size: 412, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:05:37,036 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:05:50,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:06:11,025 INFO [train.py:968] (1/2) Epoch 13, batch 21600, giga_loss[loss=0.2603, simple_loss=0.3374, pruned_loss=0.09165, over 28334.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3426, pruned_loss=0.09762, over 5719047.84 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3514, pruned_loss=0.09601, over 5747264.11 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.342, pruned_loss=0.09779, over 5711201.27 frames. ], batch size: 368, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:06:34,376 INFO [optim.py:369] (1/2) 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 20:06:49,204 INFO [train.py:968] (1/2) Epoch 13, batch 21650, giga_loss[loss=0.2563, simple_loss=0.3315, pruned_loss=0.09057, over 28633.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3425, pruned_loss=0.09849, over 5721430.49 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3519, pruned_loss=0.0965, over 5750582.45 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3413, pruned_loss=0.09826, over 5711169.86 frames. ], batch size: 336, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:07:12,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 21700, giga_loss[loss=0.2395, simple_loss=0.3047, pruned_loss=0.08713, over 28166.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.341, pruned_loss=0.09811, over 5702241.40 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3529, pruned_loss=0.0972, over 5735500.88 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3386, pruned_loss=0.09733, over 5706050.27 frames. ], batch size: 77, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:07:37,541 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:07:39,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:07:47,566 INFO [optim.py:369] (1/2) 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 20:07:57,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 20:07:57,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2316, 1.4219, 1.4807, 1.2845], device='cuda:1'), covar=tensor([0.1501, 0.1512, 0.1906, 0.1664], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0724, 0.0672, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 20:08:01,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:08:01,916 INFO [train.py:968] (1/2) Epoch 13, batch 21750, giga_loss[loss=0.3082, simple_loss=0.3793, pruned_loss=0.1186, over 28807.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3385, pruned_loss=0.09637, over 5709908.33 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3525, pruned_loss=0.09705, over 5740486.23 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3367, pruned_loss=0.09589, over 5707936.89 frames. ], batch size: 199, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:08:32,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:08:42,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569262.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:08:44,700 INFO [train.py:968] (1/2) Epoch 13, batch 21800, giga_loss[loss=0.2624, simple_loss=0.3434, pruned_loss=0.09072, over 29048.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3394, pruned_loss=0.09703, over 5699747.20 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3524, pruned_loss=0.09714, over 5732886.11 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3378, pruned_loss=0.09657, over 5704300.49 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:08:48,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569268.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 20:09:14,490 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 21850, giga_loss[loss=0.2462, simple_loss=0.3325, pruned_loss=0.07998, over 29034.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3424, pruned_loss=0.09848, over 5702022.85 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3526, pruned_loss=0.09737, over 5733740.68 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.341, pruned_loss=0.09793, over 5704725.11 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:09:32,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6167, 1.8231, 1.8235, 1.3855], device='cuda:1'), covar=tensor([0.1727, 0.2342, 0.1436, 0.1659], device='cuda:1'), in_proj_covar=tensor([0.0842, 0.0686, 0.0883, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 20:10:10,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1957, 1.6223, 1.3396, 1.3617], device='cuda:1'), covar=tensor([0.0738, 0.0372, 0.0340, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0093], device='cuda:1')
+2023-03-06 20:10:12,108 INFO [train.py:968] (1/2) Epoch 13, batch 21900, giga_loss[loss=0.2525, simple_loss=0.3232, pruned_loss=0.09088, over 28594.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3451, pruned_loss=0.09942, over 5688249.57 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3537, pruned_loss=0.09814, over 5718065.73 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3426, pruned_loss=0.0983, over 5704405.17 frames. ], batch size: 85, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:10:34,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:10:37,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:10:39,060 INFO [optim.py:369] (1/2) 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 20:10:39,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:10:46,348 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569405.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:10:48,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569408.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:10:50,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569411.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:10:51,925 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 21950, giga_loss[loss=0.2775, simple_loss=0.3427, pruned_loss=0.1062, over 28714.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.347, pruned_loss=0.1001, over 5690215.58 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3543, pruned_loss=0.0991, over 5723383.77 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3443, pruned_loss=0.09827, over 5697091.42 frames. ], batch size: 92, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:10:58,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:11:10,069 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 20:11:14,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4703, 1.6757, 1.3675, 1.5970], device='cuda:1'), covar=tensor([0.0713, 0.0291, 0.0314, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0054, 0.0093], device='cuda:1')
+2023-03-06 20:11:15,228 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569443.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 20:11:35,775 INFO [train.py:968] (1/2) Epoch 13, batch 22000, giga_loss[loss=0.2689, simple_loss=0.345, pruned_loss=0.0964, over 28719.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3467, pruned_loss=0.09944, over 5691191.61 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3548, pruned_loss=0.0995, over 5724284.83 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3439, pruned_loss=0.09767, over 5695204.89 frames. ], batch size: 242, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:12:00,613 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 22050, giga_loss[loss=0.274, simple_loss=0.3382, pruned_loss=0.1048, over 28567.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3463, pruned_loss=0.09923, over 5691465.32 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.355, pruned_loss=0.09978, over 5720094.76 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3436, pruned_loss=0.09757, over 5698117.79 frames. ], batch size: 78, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:12:22,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=569522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:12:37,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:12:39,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:12:56,719 INFO [train.py:968] (1/2) Epoch 13, batch 22100, giga_loss[loss=0.2704, simple_loss=0.3362, pruned_loss=0.1023, over 28795.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3467, pruned_loss=0.09961, over 5696028.41 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.355, pruned_loss=0.0998, over 5722868.83 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3445, pruned_loss=0.09828, over 5698500.06 frames. ], batch size: 99, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:13:01,805 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:13:10,114 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:13:23,600 INFO [optim.py:369] (1/2) 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 20:13:38,688 INFO [train.py:968] (1/2) Epoch 13, batch 22150, giga_loss[loss=0.2853, simple_loss=0.3607, pruned_loss=0.105, over 28668.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3487, pruned_loss=0.1009, over 5699973.29 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3552, pruned_loss=0.1, over 5722403.49 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3466, pruned_loss=0.09972, over 5701574.66 frames. ], batch size: 307, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:14:19,208 INFO [train.py:968] (1/2) Epoch 13, batch 22200, libri_loss[loss=0.2587, simple_loss=0.342, pruned_loss=0.08773, over 29577.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3511, pruned_loss=0.1025, over 5696760.29 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3548, pruned_loss=0.09969, over 5725280.06 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3498, pruned_loss=0.1018, over 5694939.34 frames. ], batch size: 76, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:14:19,563 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=569665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:14:21,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=569668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:14:38,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:14:45,328 INFO [optim.py:369] (1/2) 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 20:14:45,523 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=569697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:15:00,512 INFO [train.py:968] (1/2) Epoch 13, batch 22250, giga_loss[loss=0.3451, simple_loss=0.4069, pruned_loss=0.1417, over 28848.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3539, pruned_loss=0.1038, over 5698720.08 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3554, pruned_loss=0.1001, over 5721311.09 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.1031, over 5700421.26 frames. ], batch size: 186, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:15:38,363 INFO [train.py:968] (1/2) Epoch 13, batch 22300, giga_loss[loss=0.257, simple_loss=0.3448, pruned_loss=0.08465, over 28888.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.355, pruned_loss=0.1042, over 5707867.76 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3561, pruned_loss=0.1008, over 5725698.09 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3529, pruned_loss=0.103, over 5704619.55 frames. ], batch size: 174, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:15:56,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:16:14,463 INFO [train.py:968] (1/2) Epoch 13, batch 22350, giga_loss[loss=0.2769, simple_loss=0.3444, pruned_loss=0.1047, over 28774.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.357, pruned_loss=0.1053, over 5711582.52 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3568, pruned_loss=0.1013, over 5727264.39 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3547, pruned_loss=0.104, over 5707206.66 frames. ], batch size: 99, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:16:58,704 INFO [train.py:968] (1/2) Epoch 13, batch 22400, giga_loss[loss=0.2745, simple_loss=0.3548, pruned_loss=0.09706, over 28913.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3576, pruned_loss=0.106, over 5716868.80 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3567, pruned_loss=0.1013, over 5730729.85 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3559, pruned_loss=0.105, over 5710050.62 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:17:12,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2387, 1.2508, 1.1551, 0.8819], device='cuda:1'), covar=tensor([0.0804, 0.0522, 0.1027, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0438, 0.0503, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 20:17:25,334 INFO [optim.py:369] (1/2) 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 20:17:34,619 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-06 20:17:36,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 20:17:40,728 INFO [train.py:968] (1/2) Epoch 13, batch 22450, giga_loss[loss=0.2907, simple_loss=0.3566, pruned_loss=0.1124, over 29015.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.356, pruned_loss=0.1053, over 5710552.30 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3565, pruned_loss=0.1013, over 5728132.49 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3547, pruned_loss=0.1046, over 5706877.51 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:18:15,628 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 22500, giga_loss[loss=0.255, simple_loss=0.3325, pruned_loss=0.08871, over 29107.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3531, pruned_loss=0.1038, over 5710005.96 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3566, pruned_loss=0.1014, over 5730334.87 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.352, pruned_loss=0.1032, over 5705028.49 frames. ], batch size: 155, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:18:25,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-06 20:18:30,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=569975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:18:48,592 INFO [optim.py:369] (1/2) 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 20:19:02,070 INFO [train.py:968] (1/2) Epoch 13, batch 22550, giga_loss[loss=0.2244, simple_loss=0.3036, pruned_loss=0.07264, over 28564.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3481, pruned_loss=0.1009, over 5710250.41 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3563, pruned_loss=0.1012, over 5733998.07 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3474, pruned_loss=0.1007, over 5702557.80 frames. ], batch size: 85, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:19:11,023 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:19:40,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:19:41,820 INFO [train.py:968] (1/2) Epoch 13, batch 22600, giga_loss[loss=0.2922, simple_loss=0.3681, pruned_loss=0.1082, over 28044.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3474, pruned_loss=0.09981, over 5709727.37 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3565, pruned_loss=0.1017, over 5732948.10 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3464, pruned_loss=0.09916, over 5704714.73 frames. ], batch size: 412, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:20:02,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-06 20:20:12,011 INFO [optim.py:369] (1/2) 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 20:20:13,537 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:20:16,558 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 22650, giga_loss[loss=0.2384, simple_loss=0.3272, pruned_loss=0.07484, over 28868.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3491, pruned_loss=0.09908, over 5705150.64 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3567, pruned_loss=0.1018, over 5732150.97 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3481, pruned_loss=0.09842, over 5701703.75 frames. ], batch size: 174, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:20:38,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:20:52,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2708, 4.0806, 3.8587, 2.0167], device='cuda:1'), covar=tensor([0.0518, 0.0691, 0.0694, 0.1977], device='cuda:1'), in_proj_covar=tensor([0.1069, 0.0988, 0.0864, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 20:21:05,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:21:05,677 INFO [train.py:968] (1/2) Epoch 13, batch 22700, giga_loss[loss=0.3534, simple_loss=0.3913, pruned_loss=0.1578, over 24102.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3501, pruned_loss=0.1003, over 5701646.43 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3567, pruned_loss=0.1019, over 5736390.11 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3493, pruned_loss=0.0996, over 5694304.01 frames. ], batch size: 705, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:21:25,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5282, 1.7755, 1.3921, 1.6716], device='cuda:1'), covar=tensor([0.2483, 0.2465, 0.2848, 0.2377], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.0984, 0.1179, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 20:21:32,836 INFO [optim.py:369] (1/2) 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 20:21:41,314 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:21:43,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:21:48,397 INFO [train.py:968] (1/2) Epoch 13, batch 22750, giga_loss[loss=0.3092, simple_loss=0.3662, pruned_loss=0.126, over 26727.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3479, pruned_loss=0.1005, over 5702345.34 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3565, pruned_loss=0.102, over 5737988.43 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3473, pruned_loss=0.09987, over 5694628.79 frames. ], batch size: 555, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:22:06,593 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:22:07,896 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 22800, giga_loss[loss=0.28, simple_loss=0.3504, pruned_loss=0.1048, over 27863.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3457, pruned_loss=0.1005, over 5707582.29 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3566, pruned_loss=0.1021, over 5738806.43 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3449, pruned_loss=0.09978, over 5700337.74 frames. ], batch size: 412, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:22:42,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6264, 1.7598, 1.6311, 1.5647], device='cuda:1'), covar=tensor([0.2598, 0.2046, 0.1722, 0.1883], device='cuda:1'), in_proj_covar=tensor([0.1780, 0.1687, 0.1648, 0.1732], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 20:22:55,540 INFO [optim.py:369] (1/2) 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 20:23:02,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:23:04,890 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 22850, giga_loss[loss=0.253, simple_loss=0.3243, pruned_loss=0.09085, over 29033.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3436, pruned_loss=0.1001, over 5716987.78 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3564, pruned_loss=0.1022, over 5741473.96 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3428, pruned_loss=0.09943, over 5707907.71 frames. ], batch size: 128, lr: 2.48e-03, grad_scale: 8.0
+2023-03-06 20:23:10,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3711, 1.3214, 4.5692, 3.4348], device='cuda:1'), covar=tensor([0.1689, 0.2645, 0.0358, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0600, 0.0873, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 20:23:30,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:23:40,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:23:50,595 INFO [train.py:968] (1/2) Epoch 13, batch 22900, libri_loss[loss=0.2941, simple_loss=0.3704, pruned_loss=0.1089, over 29641.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3439, pruned_loss=0.101, over 5715615.27 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3568, pruned_loss=0.1025, over 5743341.77 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3426, pruned_loss=0.1001, over 5706010.96 frames. ], batch size: 91, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:23:55,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-06 20:24:17,159 INFO [optim.py:369] (1/2) 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 20:24:18,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:24:31,517 INFO [train.py:968] (1/2) Epoch 13, batch 22950, giga_loss[loss=0.2386, simple_loss=0.3113, pruned_loss=0.08293, over 28808.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3423, pruned_loss=0.1002, over 5719791.22 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3575, pruned_loss=0.1032, over 5744361.14 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3406, pruned_loss=0.09899, over 5711008.17 frames. ], batch size: 119, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:24:40,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2227, 1.5746, 1.2372, 1.0080], device='cuda:1'), covar=tensor([0.2280, 0.2200, 0.2596, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.1328, 0.0977, 0.1172, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 20:25:06,888 INFO [train.py:968] (1/2) Epoch 13, batch 23000, giga_loss[loss=0.2301, simple_loss=0.3062, pruned_loss=0.077, over 29037.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3398, pruned_loss=0.09928, over 5724503.65 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.358, pruned_loss=0.1039, over 5750410.17 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3372, pruned_loss=0.09742, over 5710655.99 frames. ], batch size: 136, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:25:19,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:29,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:32,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:34,717 INFO [optim.py:369] (1/2) 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 20:25:46,590 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:25:46,987 INFO [train.py:968] (1/2) Epoch 13, batch 23050, giga_loss[loss=0.2699, simple_loss=0.3424, pruned_loss=0.09871, over 28643.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.336, pruned_loss=0.09717, over 5710296.41 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.358, pruned_loss=0.104, over 5743922.56 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3335, pruned_loss=0.09544, over 5704933.58 frames. ], batch size: 307, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:25:53,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9968, 1.3038, 1.0881, 0.2024], device='cuda:1'), covar=tensor([0.3254, 0.2528, 0.3963, 0.5357], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1503, 0.1497, 0.1305], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 20:25:54,622 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:10,474 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:12,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=570547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:18,712 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-06 20:26:26,189 INFO [train.py:968] (1/2) Epoch 13, batch 23100, giga_loss[loss=0.2584, simple_loss=0.3382, pruned_loss=0.08934, over 28960.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3366, pruned_loss=0.09709, over 5710436.05 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3578, pruned_loss=0.1039, over 5737690.25 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3344, pruned_loss=0.09572, over 5710721.96 frames. ], batch size: 174, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:26:30,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-06 20:26:36,553 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:39,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:26:45,034 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:27:07,239 INFO [train.py:968] (1/2) Epoch 13, batch 23150, libri_loss[loss=0.2818, simple_loss=0.3563, pruned_loss=0.1036, over 29527.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3386, pruned_loss=0.09789, over 5710417.68 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3578, pruned_loss=0.1042, over 5741532.60 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3361, pruned_loss=0.09629, over 5706236.08 frames. ], batch size: 82, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:27:07,452 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:27:48,043 INFO [train.py:968] (1/2) Epoch 13, batch 23200, giga_loss[loss=0.3585, simple_loss=0.4051, pruned_loss=0.156, over 26630.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3423, pruned_loss=0.09949, over 5711509.93 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.358, pruned_loss=0.1047, over 5744430.92 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3397, pruned_loss=0.09763, over 5704856.34 frames. ], batch size: 555, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:28:12,803 INFO [optim.py:369] (1/2) 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 20:28:23,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3173, 5.1266, 4.8411, 2.0960], device='cuda:1'), covar=tensor([0.0360, 0.0533, 0.0602, 0.2078], device='cuda:1'), in_proj_covar=tensor([0.1082, 0.1003, 0.0871, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 20:28:25,464 INFO [train.py:968] (1/2) Epoch 13, batch 23250, giga_loss[loss=0.2666, simple_loss=0.3457, pruned_loss=0.09379, over 28848.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3464, pruned_loss=0.1016, over 5713861.73 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3584, pruned_loss=0.1055, over 5742033.98 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3434, pruned_loss=0.09919, over 5708979.82 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:28:31,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-06 20:28:33,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:28:49,446 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:28:59,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=570758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:29:02,144 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 23300, giga_loss[loss=0.2408, simple_loss=0.3198, pruned_loss=0.08087, over 28633.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3495, pruned_loss=0.1028, over 5696798.58 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3587, pruned_loss=0.1058, over 5734275.13 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3465, pruned_loss=0.1005, over 5699223.89 frames. ], batch size: 78, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:29:27,850 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=570790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:29:35,710 INFO [optim.py:369] (1/2) 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 20:29:45,136 INFO [train.py:968] (1/2) Epoch 13, batch 23350, giga_loss[loss=0.3291, simple_loss=0.3885, pruned_loss=0.1349, over 28872.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3529, pruned_loss=0.1049, over 5694847.65 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3594, pruned_loss=0.1065, over 5733862.68 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3497, pruned_loss=0.1024, over 5695978.75 frames. ], batch size: 186, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:30:24,773 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 23400, giga_loss[loss=0.3083, simple_loss=0.3736, pruned_loss=0.1215, over 28776.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3597, pruned_loss=0.1112, over 5691388.06 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3594, pruned_loss=0.1066, over 5736458.64 frames. ], giga_tot_loss[loss=0.2876, simple_loss=0.3571, pruned_loss=0.1091, over 5689342.08 frames. ], batch size: 243, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:30:52,474 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=570884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:30:57,369 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:31:23,072 INFO [train.py:968] (1/2) Epoch 13, batch 23450, libri_loss[loss=0.3047, simple_loss=0.3744, pruned_loss=0.1175, over 29219.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3648, pruned_loss=0.115, over 5687428.09 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3601, pruned_loss=0.1072, over 5733674.11 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3621, pruned_loss=0.1129, over 5687034.14 frames. ], batch size: 97, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:31:50,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8728, 2.2911, 1.8968, 1.4616], device='cuda:1'), covar=tensor([0.2098, 0.1409, 0.1565, 0.2011], device='cuda:1'), in_proj_covar=tensor([0.1792, 0.1695, 0.1660, 0.1742], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 20:32:07,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:32:13,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=570961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:32:15,543 INFO [train.py:968] (1/2) Epoch 13, batch 23500, giga_loss[loss=0.299, simple_loss=0.3654, pruned_loss=0.1163, over 28670.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3732, pruned_loss=0.1212, over 5683923.09 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3601, pruned_loss=0.1074, over 5735497.66 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3712, pruned_loss=0.1196, over 5681409.82 frames. ], batch size: 119, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:32:48,064 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:32:50,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:02,175 INFO [train.py:968] (1/2) Epoch 13, batch 23550, giga_loss[loss=0.3053, simple_loss=0.3747, pruned_loss=0.1179, over 28999.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.379, pruned_loss=0.1267, over 5682754.52 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3611, pruned_loss=0.1083, over 5739346.28 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.377, pruned_loss=0.1251, over 5675249.78 frames. ], batch size: 174, lr: 2.48e-03, grad_scale: 2.0
+2023-03-06 20:33:15,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6528, 1.7696, 1.7394, 1.3583], device='cuda:1'), covar=tensor([0.1784, 0.1708, 0.1278, 0.1668], device='cuda:1'), in_proj_covar=tensor([0.1791, 0.1694, 0.1656, 0.1740], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 20:33:16,627 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:17,235 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:19,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:49,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:33:49,663 INFO [train.py:968] (1/2) Epoch 13, batch 23600, giga_loss[loss=0.3185, simple_loss=0.3896, pruned_loss=0.1237, over 28884.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3859, pruned_loss=0.1329, over 5670607.23 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3614, pruned_loss=0.1086, over 5742920.26 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3845, pruned_loss=0.1319, over 5659967.51 frames. ], batch size: 145, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:34:22,076 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:23,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:23,635 INFO [optim.py:369] (1/2) 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 20:34:23,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:26,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:29,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:33,183 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:33,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3216, 1.4780, 1.5477, 1.3270], device='cuda:1'), covar=tensor([0.1657, 0.1649, 0.2126, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0738, 0.0685, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 20:34:34,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-06 20:34:35,144 INFO [train.py:968] (1/2) Epoch 13, batch 23650, libri_loss[loss=0.2784, simple_loss=0.3396, pruned_loss=0.1086, over 29662.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3872, pruned_loss=0.1337, over 5677590.61 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3615, pruned_loss=0.1089, over 5745363.63 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3873, pruned_loss=0.1338, over 5663050.95 frames. ], batch size: 73, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:34:39,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:48,093 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:34:54,467 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 23700, giga_loss[loss=0.3102, simple_loss=0.3786, pruned_loss=0.1208, over 28943.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3887, pruned_loss=0.1355, over 5675048.71 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3616, pruned_loss=0.1091, over 5748984.35 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3893, pruned_loss=0.136, over 5658433.92 frames. ], batch size: 136, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:35:56,688 INFO [optim.py:369] (1/2) 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 20:36:11,949 INFO [train.py:968] (1/2) Epoch 13, batch 23750, libri_loss[loss=0.3203, simple_loss=0.3838, pruned_loss=0.1284, over 28669.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3906, pruned_loss=0.1382, over 5644986.39 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3621, pruned_loss=0.1096, over 5731167.31 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3913, pruned_loss=0.1388, over 5645699.98 frames. ], batch size: 106, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:36:39,749 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:36:42,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:36:58,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:03,999 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:04,369 INFO [train.py:968] (1/2) Epoch 13, batch 23800, libri_loss[loss=0.2697, simple_loss=0.3409, pruned_loss=0.09929, over 29551.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3939, pruned_loss=0.1417, over 5645313.78 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3624, pruned_loss=0.1099, over 5734199.94 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3948, pruned_loss=0.1424, over 5641593.87 frames. ], batch size: 80, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:37:05,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:05,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:12,437 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:14,908 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-06 20:37:32,914 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:37:39,239 INFO [optim.py:369] (1/2) 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 20:37:53,994 INFO [train.py:968] (1/2) Epoch 13, batch 23850, giga_loss[loss=0.4038, simple_loss=0.4173, pruned_loss=0.1952, over 23188.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3961, pruned_loss=0.1444, over 5625559.76 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3621, pruned_loss=0.11, over 5731460.14 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.3985, pruned_loss=0.1463, over 5621033.05 frames. ], batch size: 705, lr: 2.48e-03, grad_scale: 4.0
+2023-03-06 20:38:44,819 INFO [train.py:968] (1/2) Epoch 13, batch 23900, giga_loss[loss=0.4099, simple_loss=0.4382, pruned_loss=0.1908, over 28328.00 frames. ], tot_loss[loss=0.3439, simple_loss=0.3963, pruned_loss=0.1457, over 5615295.92 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3624, pruned_loss=0.1104, over 5732029.65 frames. ], giga_tot_loss[loss=0.3476, simple_loss=0.399, pruned_loss=0.1481, over 5607107.84 frames. ], batch size: 368, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:39:20,694 INFO [optim.py:369] (1/2) 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 20:39:22,484 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:39:27,571 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 23950, giga_loss[loss=0.2882, simple_loss=0.3559, pruned_loss=0.1103, over 28471.00 frames. ], tot_loss[loss=0.3421, simple_loss=0.3945, pruned_loss=0.1449, over 5629210.22 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3624, pruned_loss=0.1105, over 5732940.50 frames. ], giga_tot_loss[loss=0.3452, simple_loss=0.3967, pruned_loss=0.1468, over 5621548.29 frames. ], batch size: 78, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:39:44,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2099, 1.1392, 4.0084, 3.2999], device='cuda:1'), covar=tensor([0.1707, 0.2738, 0.0435, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0605, 0.0885, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 20:39:52,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:40:11,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1879, 1.2215, 3.4659, 3.0960], device='cuda:1'), covar=tensor([0.1562, 0.2662, 0.0452, 0.1700], device='cuda:1'), in_proj_covar=tensor([0.0689, 0.0605, 0.0883, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 20:40:21,753 INFO [train.py:968] (1/2) Epoch 13, batch 24000, giga_loss[loss=0.3011, simple_loss=0.3695, pruned_loss=0.1164, over 28973.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3934, pruned_loss=0.1436, over 5621286.40 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3629, pruned_loss=0.1111, over 5727904.24 frames. ], giga_tot_loss[loss=0.3432, simple_loss=0.3955, pruned_loss=0.1455, over 5617315.01 frames. ], batch size: 128, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 20:40:21,753 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 20:40:30,434 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 20:40:49,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:41:04,935 INFO [optim.py:369] (1/2) 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 20:41:19,215 INFO [train.py:968] (1/2) Epoch 13, batch 24050, giga_loss[loss=0.4196, simple_loss=0.4382, pruned_loss=0.2005, over 26487.00 frames. ], tot_loss[loss=0.3407, simple_loss=0.3944, pruned_loss=0.1435, over 5621892.30 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3636, pruned_loss=0.1118, over 5732614.57 frames. ], giga_tot_loss[loss=0.3435, simple_loss=0.3964, pruned_loss=0.1453, over 5612039.78 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:41:56,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6028, 1.6733, 1.6356, 1.5157], device='cuda:1'), covar=tensor([0.1549, 0.2016, 0.2075, 0.1809], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0741, 0.0685, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 20:42:10,581 INFO [train.py:968] (1/2) Epoch 13, batch 24100, giga_loss[loss=0.3129, simple_loss=0.3818, pruned_loss=0.1219, over 28632.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3961, pruned_loss=0.1443, over 5622284.79 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.364, pruned_loss=0.1122, over 5735234.78 frames. ], giga_tot_loss[loss=0.3446, simple_loss=0.3977, pruned_loss=0.1457, over 5610995.90 frames. ], batch size: 242, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:42:25,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4870, 3.3006, 1.5367, 1.4712], device='cuda:1'), covar=tensor([0.0917, 0.0270, 0.0823, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0517, 0.0348, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 20:42:48,808 INFO [optim.py:369] (1/2) 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 20:42:51,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5421, 2.2249, 1.5428, 0.7785], device='cuda:1'), covar=tensor([0.4836, 0.2478, 0.3587, 0.5338], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1518, 0.1510, 0.1315], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 20:43:01,651 INFO [train.py:968] (1/2) Epoch 13, batch 24150, giga_loss[loss=0.3351, simple_loss=0.3968, pruned_loss=0.1367, over 28573.00 frames. ], tot_loss[loss=0.3399, simple_loss=0.3945, pruned_loss=0.1426, over 5624236.41 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3647, pruned_loss=0.1128, over 5734513.87 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3955, pruned_loss=0.1436, over 5614472.64 frames. ], batch size: 336, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:43:18,823 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:21,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:21,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1690, 2.6505, 1.1627, 1.3341], device='cuda:1'), covar=tensor([0.0997, 0.0357, 0.0953, 0.1422], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0516, 0.0348, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 20:43:29,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=571642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:48,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:43:50,590 INFO [train.py:968] (1/2) Epoch 13, batch 24200, giga_loss[loss=0.3082, simple_loss=0.3806, pruned_loss=0.1179, over 28940.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3915, pruned_loss=0.1394, over 5632155.41 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3649, pruned_loss=0.1132, over 5738160.72 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3927, pruned_loss=0.1404, over 5618811.22 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:43:57,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2452, 1.3472, 3.3695, 3.0514], device='cuda:1'), covar=tensor([0.1525, 0.2431, 0.0505, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0683, 0.0599, 0.0875, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 20:44:00,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1463, 5.9532, 5.6718, 2.9582], device='cuda:1'), covar=tensor([0.0473, 0.0647, 0.0777, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.1089, 0.1015, 0.0885, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 20:44:03,826 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-06 20:44:24,573 INFO [optim.py:369] (1/2) 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 20:44:32,223 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:44:39,504 INFO [train.py:968] (1/2) Epoch 13, batch 24250, giga_loss[loss=0.3119, simple_loss=0.3872, pruned_loss=0.1183, over 28556.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3877, pruned_loss=0.1357, over 5634928.66 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3651, pruned_loss=0.1137, over 5739153.44 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3892, pruned_loss=0.1367, over 5620538.66 frames. ], batch size: 71, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:45:26,107 INFO [train.py:968] (1/2) Epoch 13, batch 24300, giga_loss[loss=0.3818, simple_loss=0.4156, pruned_loss=0.1739, over 26597.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.384, pruned_loss=0.1326, over 5637208.22 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3649, pruned_loss=0.1137, over 5740925.85 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3858, pruned_loss=0.1338, over 5622394.86 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:45:47,131 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=571785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:45:49,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=571788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:46:01,139 INFO [optim.py:369] (1/2) 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 20:46:14,894 INFO [train.py:968] (1/2) Epoch 13, batch 24350, giga_loss[loss=0.3767, simple_loss=0.4116, pruned_loss=0.171, over 26687.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3819, pruned_loss=0.1314, over 5640818.87 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3651, pruned_loss=0.1141, over 5743198.40 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3833, pruned_loss=0.1322, over 5625574.13 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:46:16,363 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=571817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:46:19,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-06 20:46:55,409 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-06 20:46:58,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:46:59,109 INFO [train.py:968] (1/2) Epoch 13, batch 24400, giga_loss[loss=0.275, simple_loss=0.3426, pruned_loss=0.1037, over 28553.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3818, pruned_loss=0.1316, over 5645334.57 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.365, pruned_loss=0.1143, over 5746615.12 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3836, pruned_loss=0.1326, over 5626671.65 frames. ], batch size: 71, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:47:35,743 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:47:52,241 INFO [train.py:968] (1/2) Epoch 13, batch 24450, giga_loss[loss=0.2976, simple_loss=0.3696, pruned_loss=0.1128, over 29006.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3825, pruned_loss=0.1319, over 5650239.94 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3659, pruned_loss=0.1153, over 5747470.74 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3833, pruned_loss=0.1321, over 5632886.68 frames. ], batch size: 164, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:48:47,861 INFO [train.py:968] (1/2) Epoch 13, batch 24500, giga_loss[loss=0.3502, simple_loss=0.4262, pruned_loss=0.1371, over 28126.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3802, pruned_loss=0.1285, over 5659669.96 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3662, pruned_loss=0.1155, over 5746498.39 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3807, pruned_loss=0.1285, over 5646532.27 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:49:14,926 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=571988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:49:27,867 INFO [optim.py:369] (1/2) 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 20:49:42,217 INFO [train.py:968] (1/2) Epoch 13, batch 24550, giga_loss[loss=0.3147, simple_loss=0.3936, pruned_loss=0.1179, over 29039.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3816, pruned_loss=0.1268, over 5660151.64 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3665, pruned_loss=0.1158, over 5738626.91 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3819, pruned_loss=0.1268, over 5654803.68 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:50:32,470 INFO [train.py:968] (1/2) Epoch 13, batch 24600, giga_loss[loss=0.3011, simple_loss=0.3762, pruned_loss=0.113, over 29059.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3826, pruned_loss=0.1275, over 5651803.47 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3664, pruned_loss=0.1159, over 5738776.81 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3833, pruned_loss=0.1275, over 5645570.06 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:50:41,847 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-06 20:50:52,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:51:13,417 INFO [optim.py:369] (1/2) 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 20:51:23,547 INFO [train.py:968] (1/2) Epoch 13, batch 24650, giga_loss[loss=0.2917, simple_loss=0.3571, pruned_loss=0.1132, over 28936.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3827, pruned_loss=0.1276, over 5669981.74 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3669, pruned_loss=0.1162, over 5741154.21 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.383, pruned_loss=0.1275, over 5661793.49 frames. ], batch size: 112, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:51:34,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-06 20:51:56,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-06 20:52:13,425 INFO [train.py:968] (1/2) Epoch 13, batch 24700, giga_loss[loss=0.2889, simple_loss=0.3552, pruned_loss=0.1113, over 28870.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3809, pruned_loss=0.1269, over 5685953.63 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3665, pruned_loss=0.1162, over 5745490.28 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3817, pruned_loss=0.1271, over 5674204.50 frames. ], batch size: 99, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:52:51,073 INFO [optim.py:369] (1/2) 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 20:53:00,460 INFO [train.py:968] (1/2) Epoch 13, batch 24750, giga_loss[loss=0.3103, simple_loss=0.3688, pruned_loss=0.1259, over 28685.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3794, pruned_loss=0.1273, over 5674378.67 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3666, pruned_loss=0.1164, over 5735871.02 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3802, pruned_loss=0.1275, over 5671957.51 frames. ], batch size: 242, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:53:10,915 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=572225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:13,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:15,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5868, 1.1587, 4.9591, 3.4704], device='cuda:1'), covar=tensor([0.1694, 0.2853, 0.0372, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0688, 0.0606, 0.0882, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 20:53:23,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:41,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:53:48,131 INFO [train.py:968] (1/2) Epoch 13, batch 24800, giga_loss[loss=0.3101, simple_loss=0.3769, pruned_loss=0.1216, over 28695.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3768, pruned_loss=0.126, over 5674476.84 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3662, pruned_loss=0.1162, over 5738619.96 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.378, pruned_loss=0.1265, over 5669240.06 frames. ], batch size: 242, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 20:53:53,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=572271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:54:19,863 INFO [optim.py:369] (1/2) 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 20:54:28,693 INFO [train.py:968] (1/2) Epoch 13, batch 24850, giga_loss[loss=0.3134, simple_loss=0.3818, pruned_loss=0.1225, over 28915.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3767, pruned_loss=0.1243, over 5682898.13 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3667, pruned_loss=0.1166, over 5740380.13 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3774, pruned_loss=0.1246, over 5675695.61 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:55:04,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2246, 1.1763, 4.0838, 3.1770], device='cuda:1'), covar=tensor([0.1666, 0.2727, 0.0436, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0605, 0.0884, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 20:55:17,201 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 24900, giga_loss[loss=0.2753, simple_loss=0.3515, pruned_loss=0.09956, over 28259.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3764, pruned_loss=0.1232, over 5678480.84 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3669, pruned_loss=0.1168, over 5734844.85 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3771, pruned_loss=0.1233, over 5675739.94 frames. ], batch size: 77, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:55:36,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=572382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:55:38,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:55:55,485 INFO [optim.py:369] (1/2) 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 20:56:05,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:56:05,231 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 24950, giga_loss[loss=0.2564, simple_loss=0.3363, pruned_loss=0.08827, over 29080.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3752, pruned_loss=0.1228, over 5673902.65 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3666, pruned_loss=0.1168, over 5735787.88 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3762, pruned_loss=0.1231, over 5669417.97 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:56:08,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:56:35,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:56:51,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3984, 1.5469, 1.4906, 1.4450], device='cuda:1'), covar=tensor([0.1235, 0.1599, 0.1728, 0.1437], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0742, 0.0683, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 20:56:52,189 INFO [train.py:968] (1/2) Epoch 13, batch 25000, giga_loss[loss=0.2808, simple_loss=0.351, pruned_loss=0.1053, over 28843.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3746, pruned_loss=0.1228, over 5683825.32 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3673, pruned_loss=0.1174, over 5741174.28 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.375, pruned_loss=0.1227, over 5673797.88 frames. ], batch size: 199, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:57:02,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-06 20:57:28,260 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6220, 1.8021, 1.5870, 1.5706], device='cuda:1'), covar=tensor([0.0717, 0.0294, 0.0280, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0054, 0.0093], device='cuda:1')
+2023-03-06 20:57:33,046 INFO [optim.py:369] (1/2) 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 20:57:36,794 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=572506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:57:39,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=572509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:57:44,166 INFO [train.py:968] (1/2) Epoch 13, batch 25050, giga_loss[loss=0.3062, simple_loss=0.3708, pruned_loss=0.1208, over 29072.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5671110.84 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3677, pruned_loss=0.1176, over 5743319.18 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.375, pruned_loss=0.124, over 5660460.03 frames. ], batch size: 128, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:58:05,923 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=572538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 20:58:31,800 INFO [train.py:968] (1/2) Epoch 13, batch 25100, giga_loss[loss=0.3654, simple_loss=0.4009, pruned_loss=0.165, over 26661.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3739, pruned_loss=0.1242, over 5673512.68 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3676, pruned_loss=0.1176, over 5746933.10 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3742, pruned_loss=0.1242, over 5660256.83 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:59:07,609 INFO [optim.py:369] (1/2) 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 20:59:15,815 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 25150, giga_loss[loss=0.3186, simple_loss=0.3583, pruned_loss=0.1395, over 23662.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3736, pruned_loss=0.1245, over 5674475.83 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3678, pruned_loss=0.1179, over 5747966.87 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1243, over 5661979.15 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 20:59:50,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3442, 1.9117, 1.4184, 0.7105], device='cuda:1'), covar=tensor([0.4196, 0.2091, 0.2861, 0.4812], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1532, 0.1518, 0.1324], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 21:00:05,983 INFO [train.py:968] (1/2) Epoch 13, batch 25200, giga_loss[loss=0.2857, simple_loss=0.353, pruned_loss=0.1091, over 28562.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3717, pruned_loss=0.1235, over 5678369.01 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3678, pruned_loss=0.1179, over 5750393.56 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.372, pruned_loss=0.1234, over 5665375.81 frames. ], batch size: 85, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:00:08,316 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:00:55,713 INFO [train.py:968] (1/2) Epoch 13, batch 25250, libri_loss[loss=0.3052, simple_loss=0.3763, pruned_loss=0.117, over 29367.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3727, pruned_loss=0.1246, over 5672069.81 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5754075.23 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3727, pruned_loss=0.1246, over 5657036.70 frames. ], batch size: 92, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:01:07,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9692, 1.1947, 1.2544, 1.0467], device='cuda:1'), covar=tensor([0.1492, 0.1303, 0.2131, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0745, 0.0687, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 21:01:42,503 INFO [train.py:968] (1/2) Epoch 13, batch 25300, giga_loss[loss=0.2576, simple_loss=0.3432, pruned_loss=0.08603, over 28827.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5663979.00 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3681, pruned_loss=0.1182, over 5747704.30 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3738, pruned_loss=0.1249, over 5655108.24 frames. ], batch size: 174, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:02:10,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.23 vs. limit=5.0
+2023-03-06 21:02:16,416 INFO [optim.py:369] (1/2) 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 21:02:19,300 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 25350, giga_loss[loss=0.3227, simple_loss=0.3926, pruned_loss=0.1263, over 29020.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1237, over 5677277.86 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3679, pruned_loss=0.1184, over 5750388.01 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3738, pruned_loss=0.1237, over 5665379.33 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:02:40,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 21:02:50,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3947, 1.5502, 1.4553, 1.3223], device='cuda:1'), covar=tensor([0.2075, 0.1900, 0.1332, 0.1662], device='cuda:1'), in_proj_covar=tensor([0.1776, 0.1680, 0.1648, 0.1750], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 21:02:54,434 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-06 21:03:13,279 INFO [train.py:968] (1/2) Epoch 13, batch 25400, giga_loss[loss=0.298, simple_loss=0.3708, pruned_loss=0.1126, over 28625.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.124, over 5670696.94 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3679, pruned_loss=0.1185, over 5752316.50 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.124, over 5658456.45 frames. ], batch size: 307, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:03:16,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3013, 1.4666, 1.4039, 1.5034], device='cuda:1'), covar=tensor([0.0705, 0.0402, 0.0308, 0.0730], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0115, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 21:03:50,569 INFO [optim.py:369] (1/2) 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 21:04:00,388 INFO [train.py:968] (1/2) Epoch 13, batch 25450, giga_loss[loss=0.3058, simple_loss=0.3693, pruned_loss=0.1212, over 28433.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3744, pruned_loss=0.1248, over 5668743.49 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3681, pruned_loss=0.1185, over 5753258.87 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3746, pruned_loss=0.1249, over 5657223.26 frames. ], batch size: 368, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:04:08,254 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=572923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:04:48,148 INFO [train.py:968] (1/2) Epoch 13, batch 25500, giga_loss[loss=0.2854, simple_loss=0.3541, pruned_loss=0.1083, over 28690.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1275, over 5660924.15 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3685, pruned_loss=0.1188, over 5754598.81 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3772, pruned_loss=0.1275, over 5648084.26 frames. ], batch size: 242, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:05:10,166 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:05:37,656 INFO [train.py:968] (1/2) Epoch 13, batch 25550, giga_loss[loss=0.3469, simple_loss=0.3952, pruned_loss=0.1493, over 28726.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3774, pruned_loss=0.1287, over 5667280.33 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3683, pruned_loss=0.1187, over 5756597.49 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3778, pruned_loss=0.129, over 5653766.99 frames. ], batch size: 262, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:05:39,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6622, 1.7751, 1.8733, 1.4327], device='cuda:1'), covar=tensor([0.1758, 0.2463, 0.1404, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0844, 0.0694, 0.0883, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 21:06:05,530 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 25600, giga_loss[loss=0.3294, simple_loss=0.3823, pruned_loss=0.1382, over 28994.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3778, pruned_loss=0.1297, over 5660138.52 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3688, pruned_loss=0.1191, over 5745840.85 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3779, pruned_loss=0.1298, over 5657049.26 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:06:42,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3990, 1.5605, 1.3601, 1.4301], device='cuda:1'), covar=tensor([0.1549, 0.1830, 0.1549, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.1777, 0.1687, 0.1651, 0.1754], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 21:07:05,459 INFO [optim.py:369] (1/2) 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 21:07:08,309 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:07:11,303 INFO [train.py:968] (1/2) Epoch 13, batch 25650, giga_loss[loss=0.284, simple_loss=0.3469, pruned_loss=0.1106, over 28670.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3781, pruned_loss=0.1301, over 5653794.17 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3683, pruned_loss=0.1188, over 5747502.29 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3788, pruned_loss=0.1307, over 5647141.14 frames. ], batch size: 92, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:07:27,956 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:07:30,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:07:36,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5505, 2.2348, 1.5727, 0.7456], device='cuda:1'), covar=tensor([0.4800, 0.2215, 0.3325, 0.5062], device='cuda:1'), in_proj_covar=tensor([0.1611, 0.1536, 0.1518, 0.1324], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 21:07:55,417 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 25700, libri_loss[loss=0.3044, simple_loss=0.3695, pruned_loss=0.1197, over 29524.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3767, pruned_loss=0.1289, over 5657521.00 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3686, pruned_loss=0.1191, over 5741857.23 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3773, pruned_loss=0.1294, over 5654613.98 frames. ], batch size: 82, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:08:12,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:08:17,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:08:18,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4559, 1.6385, 1.6028, 1.4364], device='cuda:1'), covar=tensor([0.1489, 0.1699, 0.1939, 0.1797], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0741, 0.0684, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 21:08:19,447 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:08:38,809 INFO [train.py:968] (1/2) Epoch 13, batch 25750, giga_loss[loss=0.2988, simple_loss=0.3723, pruned_loss=0.1126, over 28953.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3765, pruned_loss=0.1281, over 5657100.35 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3687, pruned_loss=0.1193, over 5738685.55 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3773, pruned_loss=0.1288, over 5653885.98 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:08:41,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:09:04,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4808, 2.3318, 2.4080, 2.0328], device='cuda:1'), covar=tensor([0.1340, 0.2140, 0.1663, 0.2104], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0737, 0.0681, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 21:09:23,633 INFO [train.py:968] (1/2) Epoch 13, batch 25800, giga_loss[loss=0.41, simple_loss=0.4193, pruned_loss=0.2003, over 26500.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3749, pruned_loss=0.1258, over 5653568.84 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3689, pruned_loss=0.1197, over 5731571.56 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3755, pruned_loss=0.1262, over 5655862.65 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:09:53,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:09:56,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2713, 1.4035, 3.2953, 3.1295], device='cuda:1'), covar=tensor([0.1349, 0.2346, 0.0455, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0605, 0.0884, 0.0805], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:10:00,524 INFO [optim.py:369] (1/2) 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 21:10:05,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-06 21:10:06,760 INFO [train.py:968] (1/2) Epoch 13, batch 25850, libri_loss[loss=0.2821, simple_loss=0.3472, pruned_loss=0.1085, over 29378.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3731, pruned_loss=0.125, over 5658301.50 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3683, pruned_loss=0.1195, over 5737588.56 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3743, pruned_loss=0.1257, over 5651651.06 frames. ], batch size: 71, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:10:17,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:10:20,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:10:47,202 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 25900, giga_loss[loss=0.3529, simple_loss=0.399, pruned_loss=0.1534, over 28524.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3725, pruned_loss=0.1254, over 5668603.19 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.369, pruned_loss=0.12, over 5741612.64 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3729, pruned_loss=0.1256, over 5657770.73 frames. ], batch size: 336, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:11:32,769 INFO [optim.py:369] (1/2) 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 21:11:40,288 INFO [train.py:968] (1/2) Epoch 13, batch 25950, giga_loss[loss=0.2657, simple_loss=0.3413, pruned_loss=0.09502, over 28930.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3715, pruned_loss=0.1247, over 5682223.56 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3691, pruned_loss=0.1201, over 5741252.36 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3719, pruned_loss=0.125, over 5671597.30 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:11:58,465 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:06,147 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:08,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:28,368 INFO [train.py:968] (1/2) Epoch 13, batch 26000, giga_loss[loss=0.3162, simple_loss=0.3799, pruned_loss=0.1263, over 28893.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.374, pruned_loss=0.1261, over 5664682.97 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3696, pruned_loss=0.1205, over 5721998.95 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.126, over 5672830.70 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:12:35,243 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:12:46,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:13:02,496 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:13:10,334 INFO [train.py:968] (1/2) Epoch 13, batch 26050, libri_loss[loss=0.2988, simple_loss=0.3702, pruned_loss=0.1137, over 29544.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3768, pruned_loss=0.1255, over 5677938.55 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3693, pruned_loss=0.1205, over 5729497.15 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3773, pruned_loss=0.1257, over 5675403.85 frames. ], batch size: 82, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:13:56,427 INFO [train.py:968] (1/2) Epoch 13, batch 26100, giga_loss[loss=0.3387, simple_loss=0.4072, pruned_loss=0.1351, over 28917.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3773, pruned_loss=0.1238, over 5680164.99 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1202, over 5731163.62 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3785, pruned_loss=0.1243, over 5675347.02 frames. ], batch size: 199, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:14:35,345 INFO [optim.py:369] (1/2) 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 21:14:43,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.15 vs. limit=2.0
+2023-03-06 21:14:43,740 INFO [train.py:968] (1/2) Epoch 13, batch 26150, giga_loss[loss=0.3483, simple_loss=0.3798, pruned_loss=0.1584, over 23664.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3781, pruned_loss=0.1246, over 5684956.94 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.1199, over 5734467.15 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3797, pruned_loss=0.1253, over 5677182.63 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:14:58,007 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:01,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:04,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3797, 1.6552, 1.3151, 1.3913], device='cuda:1'), covar=tensor([0.2246, 0.2204, 0.2427, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1341, 0.0988, 0.1183, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 21:15:19,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:28,649 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:15:31,528 INFO [train.py:968] (1/2) Epoch 13, batch 26200, giga_loss[loss=0.3103, simple_loss=0.3799, pruned_loss=0.1204, over 29064.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3797, pruned_loss=0.126, over 5687270.78 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5736102.84 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3812, pruned_loss=0.1267, over 5679289.96 frames. ], batch size: 155, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:16:03,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-06 21:16:04,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2564, 1.5329, 1.2537, 1.0428], device='cuda:1'), covar=tensor([0.2283, 0.2288, 0.2545, 0.2018], device='cuda:1'), in_proj_covar=tensor([0.1342, 0.0989, 0.1183, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 21:16:11,480 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 26250, giga_loss[loss=0.4332, simple_loss=0.4518, pruned_loss=0.2073, over 26648.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3811, pruned_loss=0.1283, over 5678825.34 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3682, pruned_loss=0.1199, over 5739453.89 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3823, pruned_loss=0.1289, over 5668663.46 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:17:07,558 INFO [train.py:968] (1/2) Epoch 13, batch 26300, giga_loss[loss=0.3101, simple_loss=0.374, pruned_loss=0.1231, over 28573.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3796, pruned_loss=0.1276, over 5691308.49 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1198, over 5742411.71 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3809, pruned_loss=0.1284, over 5679666.82 frames. ], batch size: 336, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:17:16,446 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:17:47,384 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:17:55,017 INFO [train.py:968] (1/2) Epoch 13, batch 26350, giga_loss[loss=0.3041, simple_loss=0.3649, pruned_loss=0.1217, over 28777.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3775, pruned_loss=0.1268, over 5684903.85 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1197, over 5736442.70 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3788, pruned_loss=0.1277, over 5679625.58 frames. ], batch size: 99, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:18:46,427 INFO [train.py:968] (1/2) Epoch 13, batch 26400, giga_loss[loss=0.3387, simple_loss=0.3934, pruned_loss=0.142, over 28879.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3771, pruned_loss=0.1272, over 5683039.53 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3681, pruned_loss=0.1197, over 5730021.28 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3782, pruned_loss=0.128, over 5683202.93 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:18:58,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=573879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:11,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:22,859 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:23,524 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=573907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:19:23,858 INFO [optim.py:369] (1/2) 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 21:19:31,499 INFO [train.py:968] (1/2) Epoch 13, batch 26450, giga_loss[loss=0.3522, simple_loss=0.3866, pruned_loss=0.1589, over 23620.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3773, pruned_loss=0.1278, over 5676036.89 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1196, over 5731962.74 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3785, pruned_loss=0.1288, over 5672550.05 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:20:01,573 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=573949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:20:04,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=573952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:20:07,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3154, 3.1388, 2.9457, 1.4524], device='cuda:1'), covar=tensor([0.0889, 0.1049, 0.0959, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1113, 0.1033, 0.0901, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-06 21:20:15,090 INFO [train.py:968] (1/2) Epoch 13, batch 26500, giga_loss[loss=0.296, simple_loss=0.3696, pruned_loss=0.1112, over 29002.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3775, pruned_loss=0.1284, over 5680035.11 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1195, over 5730869.15 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3788, pruned_loss=0.1293, over 5677290.38 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:20:29,784 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=573981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:20:47,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3003, 1.2440, 1.1697, 1.4721], device='cuda:1'), covar=tensor([0.0708, 0.0384, 0.0319, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 21:20:53,908 INFO [optim.py:369] (1/2) 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 21:20:59,535 INFO [train.py:968] (1/2) Epoch 13, batch 26550, libri_loss[loss=0.3363, simple_loss=0.3942, pruned_loss=0.1392, over 29766.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3751, pruned_loss=0.1278, over 5657318.05 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3677, pruned_loss=0.1194, over 5724682.31 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3765, pruned_loss=0.1288, over 5658775.13 frames. ], batch size: 87, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:21:06,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:09,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:10,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:22,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2154, 1.3285, 3.6361, 3.0955], device='cuda:1'), covar=tensor([0.1575, 0.2477, 0.0479, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0685, 0.0602, 0.0882, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:21:40,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:21:50,257 INFO [train.py:968] (1/2) Epoch 13, batch 26600, giga_loss[loss=0.2708, simple_loss=0.3396, pruned_loss=0.1011, over 28735.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3739, pruned_loss=0.127, over 5655808.02 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3678, pruned_loss=0.1195, over 5726376.88 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.375, pruned_loss=0.1278, over 5654961.60 frames. ], batch size: 92, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:22:04,027 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6029, 4.7646, 1.8778, 1.8491], device='cuda:1'), covar=tensor([0.0945, 0.0206, 0.0791, 0.1215], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0520, 0.0348, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0023, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 21:22:08,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7534, 2.0469, 1.6167, 1.9705], device='cuda:1'), covar=tensor([0.2489, 0.2435, 0.2655, 0.2271], device='cuda:1'), in_proj_covar=tensor([0.1345, 0.0992, 0.1186, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 21:22:27,074 INFO [optim.py:369] (1/2) 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 21:22:34,574 INFO [train.py:968] (1/2) Epoch 13, batch 26650, giga_loss[loss=0.3266, simple_loss=0.3933, pruned_loss=0.1299, over 28947.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3744, pruned_loss=0.1258, over 5668809.27 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5730654.79 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3756, pruned_loss=0.1268, over 5662592.76 frames. ], batch size: 164, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:22:51,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5924, 1.7416, 1.8605, 1.3931], device='cuda:1'), covar=tensor([0.1602, 0.2220, 0.1306, 0.1557], device='cuda:1'), in_proj_covar=tensor([0.0844, 0.0695, 0.0884, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 21:23:06,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1242, 1.1601, 3.7246, 3.1739], device='cuda:1'), covar=tensor([0.1710, 0.2711, 0.0454, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0686, 0.0603, 0.0881, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:23:09,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:23:09,354 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-06 21:23:29,193 INFO [train.py:968] (1/2) Epoch 13, batch 26700, giga_loss[loss=0.3891, simple_loss=0.4296, pruned_loss=0.1744, over 28193.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3778, pruned_loss=0.1287, over 5649207.26 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3678, pruned_loss=0.1195, over 5721754.51 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3787, pruned_loss=0.1294, over 5651681.20 frames. ], batch size: 368, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:23:34,084 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:23:35,970 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:23:54,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6157, 1.9424, 1.4802, 1.9267], device='cuda:1'), covar=tensor([0.2425, 0.2344, 0.2652, 0.2306], device='cuda:1'), in_proj_covar=tensor([0.1348, 0.0992, 0.1190, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 21:24:04,230 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:24:09,719 INFO [optim.py:369] (1/2) 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 21:24:14,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:24:16,570 INFO [train.py:968] (1/2) Epoch 13, batch 26750, giga_loss[loss=0.2943, simple_loss=0.3585, pruned_loss=0.115, over 28945.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3768, pruned_loss=0.1277, over 5661611.16 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1194, over 5722630.74 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3775, pruned_loss=0.1283, over 5662550.47 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:24:28,247 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:24:30,451 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-06 21:25:00,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 21:25:00,467 INFO [train.py:968] (1/2) Epoch 13, batch 26800, giga_loss[loss=0.3497, simple_loss=0.4044, pruned_loss=0.1475, over 28293.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3782, pruned_loss=0.126, over 5659970.73 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5718194.63 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3789, pruned_loss=0.1265, over 5664357.21 frames. ], batch size: 368, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:25:01,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:15,334 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:26,101 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:28,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:25:35,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-06 21:25:42,884 INFO [optim.py:369] (1/2) 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 21:25:47,198 INFO [train.py:968] (1/2) Epoch 13, batch 26850, giga_loss[loss=0.4297, simple_loss=0.4476, pruned_loss=0.2059, over 26480.00 frames. ], tot_loss[loss=0.315, simple_loss=0.38, pruned_loss=0.125, over 5667153.47 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.368, pruned_loss=0.1197, over 5712427.11 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3806, pruned_loss=0.1253, over 5674472.68 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:25:56,788 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:26:34,683 INFO [train.py:968] (1/2) Epoch 13, batch 26900, giga_loss[loss=0.2773, simple_loss=0.3529, pruned_loss=0.1008, over 28783.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3839, pruned_loss=0.1278, over 5669070.85 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5713462.98 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3845, pruned_loss=0.128, over 5673686.55 frames. ], batch size: 119, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:27:11,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2858, 2.8791, 1.3432, 1.4253], device='cuda:1'), covar=tensor([0.0899, 0.0405, 0.0860, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0524, 0.0351, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 21:27:15,276 INFO [optim.py:369] (1/2) 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 21:27:17,814 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:19,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:21,148 INFO [train.py:968] (1/2) Epoch 13, batch 26950, giga_loss[loss=0.3175, simple_loss=0.3831, pruned_loss=0.126, over 28826.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3861, pruned_loss=0.1305, over 5667098.64 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3681, pruned_loss=0.1198, over 5708895.84 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3869, pruned_loss=0.131, over 5673202.27 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:27:32,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:33,417 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:35,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:36,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:27:47,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:28:01,702 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:28:02,256 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:28:11,038 INFO [train.py:968] (1/2) Epoch 13, batch 27000, giga_loss[loss=0.3254, simple_loss=0.3924, pruned_loss=0.1292, over 28851.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3885, pruned_loss=0.1341, over 5636334.36 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3678, pruned_loss=0.1198, over 5700812.20 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3899, pruned_loss=0.1348, over 5647616.91 frames. ], batch size: 145, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:28:11,038 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 21:28:19,820 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 21:28:41,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:29:00,760 INFO [optim.py:369] (1/2) 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 21:29:08,434 INFO [train.py:968] (1/2) Epoch 13, batch 27050, giga_loss[loss=0.3091, simple_loss=0.3754, pruned_loss=0.1214, over 28225.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.387, pruned_loss=0.1333, over 5649498.24 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.1201, over 5700715.47 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3882, pruned_loss=0.134, over 5657334.89 frames. ], batch size: 368, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:29:31,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1999, 1.2654, 1.0771, 0.8756], device='cuda:1'), covar=tensor([0.0864, 0.0531, 0.1087, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0443, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:29:54,495 INFO [train.py:968] (1/2) Epoch 13, batch 27100, giga_loss[loss=0.3192, simple_loss=0.3833, pruned_loss=0.1275, over 28753.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3865, pruned_loss=0.1328, over 5638708.67 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3686, pruned_loss=0.1202, over 5700386.66 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3876, pruned_loss=0.1335, over 5644085.30 frames. ], batch size: 284, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:29:55,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7031, 4.7084, 1.6698, 1.8256], device='cuda:1'), covar=tensor([0.0962, 0.0305, 0.0958, 0.1359], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0522, 0.0350, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 21:30:00,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:30:03,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3058, 1.5480, 1.2624, 1.0566], device='cuda:1'), covar=tensor([0.2550, 0.2503, 0.2921, 0.2114], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.0989, 0.1188, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 21:30:08,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-06 21:30:15,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:30:30,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:30:35,796 INFO [optim.py:369] (1/2) 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 21:30:40,786 INFO [train.py:968] (1/2) Epoch 13, batch 27150, giga_loss[loss=0.3341, simple_loss=0.405, pruned_loss=0.1316, over 28874.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3847, pruned_loss=0.1292, over 5653368.42 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1201, over 5700378.47 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3859, pruned_loss=0.1301, over 5656798.62 frames. ], batch size: 199, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:31:27,245 INFO [train.py:968] (1/2) Epoch 13, batch 27200, giga_loss[loss=0.3501, simple_loss=0.385, pruned_loss=0.1576, over 23573.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3836, pruned_loss=0.1276, over 5659534.29 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3682, pruned_loss=0.1199, over 5704988.47 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3854, pruned_loss=0.1287, over 5656895.15 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:31:38,616 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:31:46,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4929, 1.7596, 1.4236, 1.5760], device='cuda:1'), covar=tensor([0.2297, 0.2212, 0.2503, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1349, 0.0991, 0.1189, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 21:32:12,022 INFO [optim.py:369] (1/2) 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 21:32:17,082 INFO [train.py:968] (1/2) Epoch 13, batch 27250, giga_loss[loss=0.3026, simple_loss=0.3694, pruned_loss=0.1179, over 28951.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3842, pruned_loss=0.1285, over 5668247.90 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1194, over 5710025.69 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3867, pruned_loss=0.13, over 5660181.30 frames. ], batch size: 200, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:32:24,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6380, 1.7867, 1.5682, 1.8437], device='cuda:1'), covar=tensor([0.1886, 0.1852, 0.1842, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.1350, 0.0993, 0.1191, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 21:32:32,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:36,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:46,818 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=574746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:50,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=574749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:32:53,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6623, 1.7342, 1.3175, 1.3279], device='cuda:1'), covar=tensor([0.0775, 0.0543, 0.0951, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0444, 0.0502, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:33:01,419 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:33:03,188 INFO [train.py:968] (1/2) Epoch 13, batch 27300, giga_loss[loss=0.2946, simple_loss=0.3651, pruned_loss=0.1121, over 28556.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3826, pruned_loss=0.1275, over 5677799.77 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3674, pruned_loss=0.1195, over 5713239.54 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3852, pruned_loss=0.129, over 5667250.58 frames. ], batch size: 92, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:33:15,502 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=574778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:33:32,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9612, 1.8323, 1.3940, 1.5422], device='cuda:1'), covar=tensor([0.0696, 0.0579, 0.0970, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0445, 0.0504, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:33:46,879 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 27350, libri_loss[loss=0.276, simple_loss=0.3324, pruned_loss=0.1098, over 29353.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3805, pruned_loss=0.1279, over 5663136.72 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3672, pruned_loss=0.1197, over 5711339.17 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3834, pruned_loss=0.1292, over 5655096.93 frames. ], batch size: 71, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:34:11,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0319, 3.4147, 2.2657, 1.2155], device='cuda:1'), covar=tensor([0.5131, 0.2201, 0.2696, 0.4462], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1533, 0.1510, 0.1324], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 21:34:31,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:34:37,862 INFO [train.py:968] (1/2) Epoch 13, batch 27400, giga_loss[loss=0.3094, simple_loss=0.3778, pruned_loss=0.1205, over 28794.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3787, pruned_loss=0.1278, over 5639877.19 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3669, pruned_loss=0.1196, over 5701678.32 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3816, pruned_loss=0.1292, over 5640889.35 frames. ], batch size: 243, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:34:41,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=574867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:35:02,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1785, 1.1188, 3.6988, 3.2619], device='cuda:1'), covar=tensor([0.1687, 0.2750, 0.0464, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0686, 0.0606, 0.0883, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:35:26,167 INFO [optim.py:369] (1/2) 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 21:35:31,652 INFO [train.py:968] (1/2) Epoch 13, batch 27450, giga_loss[loss=0.3003, simple_loss=0.3643, pruned_loss=0.1182, over 28732.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3765, pruned_loss=0.1267, over 5650522.87 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3668, pruned_loss=0.1195, over 5704872.99 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.379, pruned_loss=0.128, over 5647971.69 frames. ], batch size: 119, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:35:59,546 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 27500, giga_loss[loss=0.3173, simple_loss=0.3772, pruned_loss=0.1287, over 28303.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3767, pruned_loss=0.1279, over 5649227.41 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3671, pruned_loss=0.1196, over 5709786.14 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3786, pruned_loss=0.1291, over 5641449.48 frames. ], batch size: 368, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:36:37,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=574989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:36:58,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:36:58,355 INFO [optim.py:369] (1/2) 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 21:36:59,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:37:02,231 INFO [train.py:968] (1/2) Epoch 13, batch 27550, giga_loss[loss=0.3063, simple_loss=0.3751, pruned_loss=0.1188, over 29040.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3764, pruned_loss=0.1276, over 5645701.92 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3678, pruned_loss=0.1199, over 5704115.15 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1283, over 5643792.46 frames. ], batch size: 164, lr: 2.47e-03, grad_scale: 2.0
+2023-03-06 21:37:12,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2559, 2.0065, 1.4279, 0.5096], device='cuda:1'), covar=tensor([0.4370, 0.2500, 0.3939, 0.5392], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1533, 0.1512, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 21:37:24,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2026, 1.4804, 1.5282, 1.3162], device='cuda:1'), covar=tensor([0.1661, 0.1617, 0.2097, 0.1739], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0744, 0.0689, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 21:37:25,166 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:37:34,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:37:46,189 INFO [train.py:968] (1/2) Epoch 13, batch 27600, giga_loss[loss=0.2905, simple_loss=0.3696, pruned_loss=0.1057, over 28558.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3726, pruned_loss=0.1236, over 5653967.60 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5705392.87 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3738, pruned_loss=0.1243, over 5649862.21 frames. ], batch size: 336, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:38:08,711 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:38:12,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:38:30,408 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 27650, giga_loss[loss=0.3686, simple_loss=0.401, pruned_loss=0.1681, over 26573.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.37, pruned_loss=0.1207, over 5651482.58 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3679, pruned_loss=0.1204, over 5696968.26 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3707, pruned_loss=0.1209, over 5656012.53 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:38:41,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:38:53,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6574, 1.8763, 1.7447, 1.7729], device='cuda:1'), covar=tensor([0.1608, 0.2027, 0.2071, 0.1777], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0742, 0.0686, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 21:39:26,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3752, 1.6636, 1.3646, 1.5685], device='cuda:1'), covar=tensor([0.0738, 0.0298, 0.0308, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 21:39:28,804 INFO [train.py:968] (1/2) Epoch 13, batch 27700, giga_loss[loss=0.3182, simple_loss=0.3932, pruned_loss=0.1216, over 28933.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1213, over 5648409.39 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3679, pruned_loss=0.1204, over 5697976.42 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.371, pruned_loss=0.1214, over 5650966.29 frames. ], batch size: 164, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:39:43,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6067, 1.7376, 1.4923, 1.5819], device='cuda:1'), covar=tensor([0.1130, 0.1736, 0.1719, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0739, 0.0684, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 21:39:59,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:39:59,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:40:02,979 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:40:21,724 INFO [train.py:968] (1/2) Epoch 13, batch 27750, giga_loss[loss=0.2801, simple_loss=0.3497, pruned_loss=0.1052, over 28968.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3676, pruned_loss=0.12, over 5655451.57 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5698300.04 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3673, pruned_loss=0.1197, over 5655807.26 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:40:35,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:40:42,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:41:03,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0276, 2.5266, 2.0760, 1.5393], device='cuda:1'), covar=tensor([0.2236, 0.1631, 0.1646, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.1786, 0.1691, 0.1666, 0.1760], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 21:41:09,432 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-06 21:41:12,681 INFO [train.py:968] (1/2) Epoch 13, batch 27800, giga_loss[loss=0.2842, simple_loss=0.3534, pruned_loss=0.1075, over 28927.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3672, pruned_loss=0.1205, over 5657437.62 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1206, over 5703242.32 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3672, pruned_loss=0.1205, over 5652232.88 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:41:27,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3617, 1.6108, 1.6489, 1.3199], device='cuda:1'), covar=tensor([0.1119, 0.1059, 0.1453, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0739, 0.0685, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 21:41:53,299 INFO [optim.py:369] (1/2) 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 21:41:53,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3485, 1.6409, 1.3648, 1.0198], device='cuda:1'), covar=tensor([0.2419, 0.2429, 0.2666, 0.2064], device='cuda:1'), in_proj_covar=tensor([0.1345, 0.0988, 0.1188, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 21:41:56,799 INFO [train.py:968] (1/2) Epoch 13, batch 27850, giga_loss[loss=0.2713, simple_loss=0.355, pruned_loss=0.09374, over 28640.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1212, over 5669278.82 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5706389.17 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3692, pruned_loss=0.121, over 5661105.81 frames. ], batch size: 71, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:42:17,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2923, 1.1884, 3.9042, 3.2995], device='cuda:1'), covar=tensor([0.1540, 0.2670, 0.0385, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0685, 0.0603, 0.0878, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:42:46,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:42:47,202 INFO [train.py:968] (1/2) Epoch 13, batch 27900, giga_loss[loss=0.2692, simple_loss=0.3427, pruned_loss=0.09778, over 28965.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3716, pruned_loss=0.1225, over 5642379.73 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1213, over 5689080.22 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.122, over 5651247.55 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:42:58,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:43:01,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:43:30,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:43:32,000 INFO [optim.py:369] (1/2) 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 21:43:35,248 INFO [train.py:968] (1/2) Epoch 13, batch 27950, giga_loss[loss=0.3815, simple_loss=0.4114, pruned_loss=0.1758, over 26475.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3721, pruned_loss=0.1228, over 5644709.94 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1212, over 5692168.29 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1224, over 5648376.22 frames. ], batch size: 555, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:44:23,045 INFO [train.py:968] (1/2) Epoch 13, batch 28000, giga_loss[loss=0.3063, simple_loss=0.3693, pruned_loss=0.1217, over 28910.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1248, over 5643603.48 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3702, pruned_loss=0.1219, over 5693163.97 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3729, pruned_loss=0.124, over 5644634.92 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:44:47,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-06 21:44:49,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:01,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=575507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:02,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1037, 1.0433, 1.0597, 1.2668], device='cuda:1'), covar=tensor([0.0708, 0.0449, 0.0294, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 21:45:04,147 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:45:09,058 INFO [train.py:968] (1/2) Epoch 13, batch 28050, giga_loss[loss=0.2608, simple_loss=0.3394, pruned_loss=0.09111, over 28483.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3764, pruned_loss=0.1264, over 5651170.57 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.371, pruned_loss=0.1226, over 5688580.44 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5654893.36 frames. ], batch size: 60, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:45:09,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:13,301 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:26,367 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-06 21:45:29,878 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=575539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:45:51,660 INFO [train.py:968] (1/2) Epoch 13, batch 28100, giga_loss[loss=0.313, simple_loss=0.3832, pruned_loss=0.1214, over 28951.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3786, pruned_loss=0.1275, over 5645311.71 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5677822.26 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3769, pruned_loss=0.1261, over 5657031.81 frames. ], batch size: 227, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:45:57,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:46:14,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7185, 2.4631, 1.5172, 0.6734], device='cuda:1'), covar=tensor([0.5663, 0.2924, 0.3133, 0.5959], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1529, 0.1512, 0.1332], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 21:46:36,336 INFO [optim.py:369] (1/2) 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 21:46:40,109 INFO [train.py:968] (1/2) Epoch 13, batch 28150, giga_loss[loss=0.3703, simple_loss=0.3978, pruned_loss=0.1714, over 23462.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3785, pruned_loss=0.1279, over 5646262.81 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3713, pruned_loss=0.123, over 5683983.93 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3777, pruned_loss=0.127, over 5649352.50 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:47:23,811 INFO [train.py:968] (1/2) Epoch 13, batch 28200, giga_loss[loss=0.321, simple_loss=0.3821, pruned_loss=0.13, over 28886.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3793, pruned_loss=0.1287, over 5638148.56 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3714, pruned_loss=0.1231, over 5676670.55 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3787, pruned_loss=0.128, over 5646991.41 frames. ], batch size: 186, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:47:38,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-06 21:48:14,314 INFO [optim.py:369] (1/2) 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 21:48:16,262 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 28250, giga_loss[loss=0.2957, simple_loss=0.3601, pruned_loss=0.1157, over 28914.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3808, pruned_loss=0.1288, over 5644480.39 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3714, pruned_loss=0.1232, over 5679760.79 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3805, pruned_loss=0.1283, over 5648073.70 frames. ], batch size: 106, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:48:19,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=575716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:48:50,323 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 28300, giga_loss[loss=0.2887, simple_loss=0.3624, pruned_loss=0.1075, over 28695.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3803, pruned_loss=0.1272, over 5645910.01 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3715, pruned_loss=0.1232, over 5670858.99 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.38, pruned_loss=0.1268, over 5655943.89 frames. ], batch size: 262, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:49:59,624 INFO [optim.py:369] (1/2) 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 21:50:03,101 INFO [train.py:968] (1/2) Epoch 13, batch 28350, giga_loss[loss=0.3323, simple_loss=0.3767, pruned_loss=0.1439, over 28497.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3807, pruned_loss=0.1286, over 5654953.17 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3718, pruned_loss=0.1235, over 5671358.28 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3804, pruned_loss=0.1281, over 5662038.18 frames. ], batch size: 85, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:50:37,039 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=575848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:50:55,681 INFO [train.py:968] (1/2) Epoch 13, batch 28400, giga_loss[loss=0.324, simple_loss=0.3835, pruned_loss=0.1322, over 28950.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3798, pruned_loss=0.1292, over 5662974.12 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3714, pruned_loss=0.1234, over 5675208.96 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3801, pruned_loss=0.1291, over 5664827.09 frames. ], batch size: 136, lr: 2.47e-03, grad_scale: 8.0
+2023-03-06 21:51:03,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:51:29,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=575891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:51:32,700 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:51:53,923 INFO [train.py:968] (1/2) Epoch 13, batch 28450, giga_loss[loss=0.3107, simple_loss=0.3541, pruned_loss=0.1337, over 23524.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.1271, over 5667290.92 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3712, pruned_loss=0.1231, over 5679589.25 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3771, pruned_loss=0.1274, over 5664522.31 frames. ], batch size: 705, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:52:38,723 INFO [train.py:968] (1/2) Epoch 13, batch 28500, giga_loss[loss=0.3098, simple_loss=0.3712, pruned_loss=0.1242, over 28037.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1275, over 5678171.14 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5687655.23 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3769, pruned_loss=0.1275, over 5668356.18 frames. ], batch size: 412, lr: 2.47e-03, grad_scale: 4.0
+2023-03-06 21:52:42,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7853, 1.9276, 2.0752, 1.5756], device='cuda:1'), covar=tensor([0.1804, 0.2290, 0.1376, 0.1596], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0699, 0.0887, 0.0792], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 21:53:22,061 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:24,416 INFO [optim.py:369] (1/2) 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 21:53:25,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:25,650 INFO [train.py:968] (1/2) Epoch 13, batch 28550, giga_loss[loss=0.31, simple_loss=0.38, pruned_loss=0.12, over 28903.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3763, pruned_loss=0.128, over 5664362.42 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3714, pruned_loss=0.1233, over 5688945.14 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3768, pruned_loss=0.1281, over 5655454.92 frames. ], batch size: 145, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:53:28,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:44,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:48,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:49,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:51,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:53:57,852 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:54:12,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-06 21:54:13,065 INFO [train.py:968] (1/2) Epoch 13, batch 28600, giga_loss[loss=0.3069, simple_loss=0.3692, pruned_loss=0.1223, over 28654.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1266, over 5651220.89 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1234, over 5672880.43 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3754, pruned_loss=0.1268, over 5659584.20 frames. ], batch size: 92, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:54:14,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:54:16,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:54:37,505 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 13, batch 28650, giga_loss[loss=0.3213, simple_loss=0.3791, pruned_loss=0.1317, over 28607.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3769, pruned_loss=0.1283, over 5651855.79 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.1231, over 5678423.47 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3776, pruned_loss=0.1289, over 5652446.48 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:55:19,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5578, 2.2128, 1.5820, 0.6641], device='cuda:1'), covar=tensor([0.4262, 0.2313, 0.3306, 0.4973], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1529, 0.1518, 0.1333], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 21:55:39,514 INFO [train.py:968] (1/2) Epoch 13, batch 28700, giga_loss[loss=0.3387, simple_loss=0.392, pruned_loss=0.1427, over 29085.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.378, pruned_loss=0.1296, over 5659879.68 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.1229, over 5686478.72 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3793, pruned_loss=0.1305, over 5651867.14 frames. ], batch size: 128, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:56:02,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-06 21:56:02,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4043, 1.4723, 1.2113, 1.5157], device='cuda:1'), covar=tensor([0.0752, 0.0328, 0.0336, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 21:56:26,446 INFO [optim.py:369] (1/2) 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 21:56:27,218 INFO [train.py:968] (1/2) Epoch 13, batch 28750, giga_loss[loss=0.3015, simple_loss=0.3653, pruned_loss=0.1189, over 29011.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3785, pruned_loss=0.1306, over 5651015.50 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1227, over 5688957.18 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3801, pruned_loss=0.1318, over 5641294.00 frames. ], batch size: 106, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 21:56:36,146 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 28800, giga_loss[loss=0.3476, simple_loss=0.394, pruned_loss=0.1506, over 27631.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3795, pruned_loss=0.1321, over 5649242.85 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1228, over 5690243.26 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3811, pruned_loss=0.1332, over 5638943.62 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:57:55,845 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 28850, giga_loss[loss=0.4093, simple_loss=0.434, pruned_loss=0.1923, over 26612.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3787, pruned_loss=0.1314, over 5651604.38 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3704, pruned_loss=0.1228, over 5691377.18 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3801, pruned_loss=0.1324, over 5641462.54 frames. ], batch size: 555, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:58:07,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1740, 1.2255, 1.0536, 0.8516], device='cuda:1'), covar=tensor([0.0787, 0.0475, 0.0949, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0446, 0.0505, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 21:58:13,644 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 28900, giga_loss[loss=0.2873, simple_loss=0.3559, pruned_loss=0.1093, over 28556.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3788, pruned_loss=0.1307, over 5648967.50 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3702, pruned_loss=0.1226, over 5693251.24 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3802, pruned_loss=0.1317, over 5639203.37 frames. ], batch size: 85, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:58:51,365 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:58:54,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:59:06,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=576384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:59:20,270 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 21:59:33,982 INFO [optim.py:369] (1/2) 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 21:59:34,742 INFO [train.py:968] (1/2) Epoch 13, batch 28950, giga_loss[loss=0.2986, simple_loss=0.3614, pruned_loss=0.1179, over 28828.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3795, pruned_loss=0.1305, over 5657650.08 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3702, pruned_loss=0.1226, over 5698740.47 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3808, pruned_loss=0.1316, over 5643700.21 frames. ], batch size: 92, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 21:59:48,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2719, 5.1278, 4.8310, 2.5606], device='cuda:1'), covar=tensor([0.0416, 0.0559, 0.0649, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.1038, 0.0905, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-06 22:00:17,784 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-06 22:00:19,951 INFO [train.py:968] (1/2) Epoch 13, batch 29000, giga_loss[loss=0.3321, simple_loss=0.3907, pruned_loss=0.1368, over 28736.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3804, pruned_loss=0.131, over 5660093.27 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.37, pruned_loss=0.1225, over 5691793.02 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3818, pruned_loss=0.1321, over 5653552.64 frames. ], batch size: 284, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:00:22,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=576468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:00:58,794 INFO [optim.py:369] (1/2) 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 22:00:59,418 INFO [train.py:968] (1/2) Epoch 13, batch 29050, giga_loss[loss=0.3264, simple_loss=0.3807, pruned_loss=0.1361, over 27927.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1308, over 5679822.09 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.1219, over 5702338.28 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3824, pruned_loss=0.1327, over 5663704.36 frames. ], batch size: 412, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:01:10,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:01:12,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:01:37,769 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:01:43,974 INFO [train.py:968] (1/2) Epoch 13, batch 29100, giga_loss[loss=0.2802, simple_loss=0.3587, pruned_loss=0.1009, over 28937.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3811, pruned_loss=0.1322, over 5683099.37 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3692, pruned_loss=0.122, over 5708019.89 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3834, pruned_loss=0.1339, over 5664687.12 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:02:21,019 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-06 22:02:23,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4920, 1.6833, 1.6379, 1.4729], device='cuda:1'), covar=tensor([0.1486, 0.1922, 0.1815, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0742, 0.0687, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 22:02:28,494 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:02:30,165 INFO [optim.py:369] (1/2) 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 22:02:30,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:02:30,842 INFO [train.py:968] (1/2) Epoch 13, batch 29150, giga_loss[loss=0.3241, simple_loss=0.3938, pruned_loss=0.1272, over 28690.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3817, pruned_loss=0.1313, over 5684500.23 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3692, pruned_loss=0.1219, over 5711021.65 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3839, pruned_loss=0.1329, over 5666434.67 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:03:00,285 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:03:23,063 INFO [train.py:968] (1/2) Epoch 13, batch 29200, giga_loss[loss=0.3091, simple_loss=0.3812, pruned_loss=0.1185, over 28981.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3825, pruned_loss=0.1313, over 5659283.13 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3693, pruned_loss=0.1221, over 5704375.01 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3842, pruned_loss=0.1326, over 5650877.57 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:03:24,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2312, 1.5101, 1.4421, 1.4384], device='cuda:1'), covar=tensor([0.1698, 0.1562, 0.2106, 0.1620], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0746, 0.0690, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 22:03:40,279 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:03:59,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=576707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:04:01,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:04:06,149 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 29250, giga_loss[loss=0.3033, simple_loss=0.3696, pruned_loss=0.1185, over 28929.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3789, pruned_loss=0.1278, over 5675552.52 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3689, pruned_loss=0.1219, over 5707456.35 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3808, pruned_loss=0.1291, over 5665749.19 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:04:39,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3872, 1.7179, 1.5914, 1.5871], device='cuda:1'), covar=tensor([0.0751, 0.0300, 0.0286, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0115, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 22:04:52,699 INFO [train.py:968] (1/2) Epoch 13, batch 29300, giga_loss[loss=0.2883, simple_loss=0.3637, pruned_loss=0.1064, over 29053.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3784, pruned_loss=0.1284, over 5664411.45 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3685, pruned_loss=0.1215, over 5712870.96 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3805, pruned_loss=0.1299, over 5650779.43 frames. ], batch size: 155, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:05:25,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2746, 0.9931, 4.2183, 3.3298], device='cuda:1'), covar=tensor([0.1697, 0.2915, 0.0382, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0603, 0.0881, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 22:05:31,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5295, 1.7203, 1.4085, 1.7526], device='cuda:1'), covar=tensor([0.2374, 0.2441, 0.2724, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.1345, 0.0989, 0.1188, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 22:05:31,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1369, 1.2318, 3.4907, 3.0738], device='cuda:1'), covar=tensor([0.1534, 0.2433, 0.0461, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0603, 0.0881, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 22:05:38,152 INFO [optim.py:369] (1/2) 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 22:05:39,010 INFO [train.py:968] (1/2) Epoch 13, batch 29350, giga_loss[loss=0.3874, simple_loss=0.4178, pruned_loss=0.1785, over 27580.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3795, pruned_loss=0.1284, over 5673852.30 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 5714895.68 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3813, pruned_loss=0.1296, over 5660284.45 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:05:43,814 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:13,901 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=576850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:16,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=576853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:26,542 INFO [train.py:968] (1/2) Epoch 13, batch 29400, giga_loss[loss=0.3852, simple_loss=0.4215, pruned_loss=0.1744, over 28254.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3811, pruned_loss=0.1303, over 5667856.43 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3687, pruned_loss=0.1218, over 5720540.61 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3829, pruned_loss=0.1315, over 5649820.20 frames. ], batch size: 368, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:06:36,753 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:06:44,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=576882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:07:16,641 INFO [optim.py:369] (1/2) 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 22:07:16,654 INFO [train.py:968] (1/2) Epoch 13, batch 29450, libri_loss[loss=0.3272, simple_loss=0.3853, pruned_loss=0.1345, over 29517.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3804, pruned_loss=0.1308, over 5668401.00 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3686, pruned_loss=0.1218, over 5721788.47 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3821, pruned_loss=0.1318, over 5652290.75 frames. ], batch size: 81, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:07:39,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-06 22:08:03,217 INFO [train.py:968] (1/2) Epoch 13, batch 29500, giga_loss[loss=0.3184, simple_loss=0.3746, pruned_loss=0.1311, over 27878.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3818, pruned_loss=0.1325, over 5657852.91 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3692, pruned_loss=0.1221, over 5720802.98 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3829, pruned_loss=0.1333, over 5644523.88 frames. ], batch size: 412, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:08:23,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=576986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:08:24,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7918, 2.0523, 1.7772, 2.0026], device='cuda:1'), covar=tensor([0.1507, 0.1704, 0.1958, 0.1474], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0749, 0.0693, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 22:08:51,522 INFO [optim.py:369] (1/2) 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 22:08:51,540 INFO [train.py:968] (1/2) Epoch 13, batch 29550, giga_loss[loss=0.3221, simple_loss=0.3787, pruned_loss=0.1328, over 28895.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3819, pruned_loss=0.1324, over 5667222.66 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3689, pruned_loss=0.1219, over 5722736.12 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3832, pruned_loss=0.1333, over 5654579.49 frames. ], batch size: 112, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:09:34,955 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 29600, giga_loss[loss=0.3361, simple_loss=0.3863, pruned_loss=0.143, over 28823.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3823, pruned_loss=0.1329, over 5657664.67 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3686, pruned_loss=0.1216, over 5727287.84 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3838, pruned_loss=0.1341, over 5642011.05 frames. ], batch size: 112, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:09:56,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:10:08,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:10:24,047 INFO [train.py:968] (1/2) Epoch 13, batch 29650, giga_loss[loss=0.3026, simple_loss=0.3706, pruned_loss=0.1173, over 28939.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3815, pruned_loss=0.1317, over 5679375.59 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3688, pruned_loss=0.122, over 5732141.33 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3829, pruned_loss=0.1326, over 5660926.83 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:10:25,551 INFO [optim.py:369] (1/2) 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 22:10:30,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9967, 1.2931, 1.0085, 0.2648], device='cuda:1'), covar=tensor([0.2458, 0.1903, 0.3034, 0.4284], device='cuda:1'), in_proj_covar=tensor([0.1598, 0.1515, 0.1505, 0.1320], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 22:10:59,603 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577150.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:11:12,357 INFO [train.py:968] (1/2) Epoch 13, batch 29700, libri_loss[loss=0.2559, simple_loss=0.3282, pruned_loss=0.09182, over 29551.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.38, pruned_loss=0.1302, over 5676864.72 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3682, pruned_loss=0.1215, over 5736468.43 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.382, pruned_loss=0.1315, over 5657016.77 frames. ], batch size: 77, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:11:44,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:11:51,881 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:11:53,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:02,105 INFO [train.py:968] (1/2) Epoch 13, batch 29750, giga_loss[loss=0.2933, simple_loss=0.3672, pruned_loss=0.1097, over 28915.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.378, pruned_loss=0.1279, over 5678038.40 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3679, pruned_loss=0.1212, over 5737484.62 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.38, pruned_loss=0.1293, over 5660631.49 frames. ], batch size: 136, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:12:04,600 INFO [optim.py:369] (1/2) 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 22:12:09,275 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:11,520 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:16,384 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=577227.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:12:18,562 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:20,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:24,555 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:38,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:47,019 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:12:51,623 INFO [train.py:968] (1/2) Epoch 13, batch 29800, giga_loss[loss=0.2725, simple_loss=0.3414, pruned_loss=0.1018, over 28631.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3762, pruned_loss=0.1269, over 5677595.39 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3679, pruned_loss=0.1212, over 5741092.43 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.378, pruned_loss=0.1282, over 5658900.55 frames. ], batch size: 71, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:12:57,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2792, 2.9435, 1.5061, 1.3931], device='cuda:1'), covar=tensor([0.0886, 0.0359, 0.0814, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0522, 0.0352, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 22:13:34,944 INFO [train.py:968] (1/2) Epoch 13, batch 29850, libri_loss[loss=0.3112, simple_loss=0.3623, pruned_loss=0.1301, over 29569.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3758, pruned_loss=0.1267, over 5683941.99 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3682, pruned_loss=0.1212, over 5747167.72 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1281, over 5660274.42 frames. ], batch size: 74, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:13:35,548 INFO [optim.py:369] (1/2) 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 22:13:43,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0758, 5.3881, 2.4179, 2.4079], device='cuda:1'), covar=tensor([0.0892, 0.0315, 0.0763, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0523, 0.0352, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 22:13:56,267 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:16,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:19,734 INFO [train.py:968] (1/2) Epoch 13, batch 29900, giga_loss[loss=0.2461, simple_loss=0.3197, pruned_loss=0.08628, over 29038.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3735, pruned_loss=0.1256, over 5681024.80 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3686, pruned_loss=0.1216, over 5748978.96 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1264, over 5659180.91 frames. ], batch size: 155, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:14:25,826 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:26,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2486, 1.1654, 1.1519, 1.5147], device='cuda:1'), covar=tensor([0.0759, 0.0356, 0.0336, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 22:14:49,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3674, 1.5451, 1.3870, 1.3396], device='cuda:1'), covar=tensor([0.1971, 0.1716, 0.1665, 0.1709], device='cuda:1'), in_proj_covar=tensor([0.1780, 0.1691, 0.1659, 0.1757], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 22:14:49,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:14:52,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:15:06,708 INFO [train.py:968] (1/2) Epoch 13, batch 29950, giga_loss[loss=0.3253, simple_loss=0.3811, pruned_loss=0.1348, over 28597.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3705, pruned_loss=0.1241, over 5680760.87 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3689, pruned_loss=0.1218, over 5744154.09 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3711, pruned_loss=0.1246, over 5666226.08 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 22:15:07,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4956, 1.5473, 1.1937, 1.1816], device='cuda:1'), covar=tensor([0.0783, 0.0528, 0.1012, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0443, 0.0503, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 22:15:08,863 INFO [optim.py:369] (1/2) 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 22:15:19,094 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 30000, giga_loss[loss=0.2966, simple_loss=0.365, pruned_loss=0.1141, over 28565.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3692, pruned_loss=0.1237, over 5687541.28 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.122, over 5739738.91 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3693, pruned_loss=0.124, over 5679173.22 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:15:53,636 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 22:15:58,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3614, 1.7567, 1.3414, 1.3723], device='cuda:1'), covar=tensor([0.2900, 0.2625, 0.3145, 0.2321], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.0986, 0.1184, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 22:16:02,156 INFO [train.py:1012] (1/2) Epoch 13, validation: loss=0.2139, simple_loss=0.3215, pruned_loss=0.05312, over 944034.00 frames. 
+2023-03-06 22:16:02,156 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 22:16:08,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:16:38,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:16:42,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:16:48,158 INFO [train.py:968] (1/2) Epoch 13, batch 30050, giga_loss[loss=0.3703, simple_loss=0.4024, pruned_loss=0.1691, over 27563.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.368, pruned_loss=0.1228, over 5698803.82 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.369, pruned_loss=0.1217, over 5744423.18 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3682, pruned_loss=0.1233, over 5686501.14 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:16:49,588 INFO [optim.py:369] (1/2) 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 22:16:58,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577525.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:17:09,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:17:32,069 INFO [train.py:968] (1/2) Epoch 13, batch 30100, giga_loss[loss=0.3193, simple_loss=0.3822, pruned_loss=0.1282, over 27983.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3674, pruned_loss=0.1209, over 5696970.10 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3687, pruned_loss=0.1214, over 5749729.78 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3679, pruned_loss=0.1216, over 5680270.19 frames. ], batch size: 412, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:17:51,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9193, 2.1280, 1.9811, 1.9182], device='cuda:1'), covar=tensor([0.1673, 0.2180, 0.1764, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0742, 0.0686, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 22:18:05,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:07,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:13,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=577602.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:18:27,077 INFO [train.py:968] (1/2) Epoch 13, batch 30150, giga_loss[loss=0.2881, simple_loss=0.3703, pruned_loss=0.1029, over 28514.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3653, pruned_loss=0.1172, over 5695100.38 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3684, pruned_loss=0.1212, over 5752038.64 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3659, pruned_loss=0.1179, over 5678921.95 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:18:28,601 INFO [optim.py:369] (1/2) 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 22:18:30,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:34,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:18:51,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3102, 1.4059, 1.4612, 1.0850], device='cuda:1'), covar=tensor([0.1657, 0.3028, 0.1408, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0698, 0.0889, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 22:19:04,206 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:19:17,190 INFO [train.py:968] (1/2) Epoch 13, batch 30200, giga_loss[loss=0.2687, simple_loss=0.3443, pruned_loss=0.09652, over 28925.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3627, pruned_loss=0.115, over 5678943.35 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3684, pruned_loss=0.1215, over 5752851.03 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3631, pruned_loss=0.1151, over 5663833.19 frames. ], batch size: 213, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:19:21,585 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577668.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:19:25,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577671.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:19:53,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577700.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:20:04,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1533, 1.1236, 3.5120, 3.1214], device='cuda:1'), covar=tensor([0.1588, 0.2759, 0.0471, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0689, 0.0608, 0.0891, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 22:20:04,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3737, 1.9216, 1.4058, 0.6317], device='cuda:1'), covar=tensor([0.3105, 0.1919, 0.2824, 0.4197], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1541, 0.1521, 0.1338], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 22:20:09,305 INFO [train.py:968] (1/2) Epoch 13, batch 30250, giga_loss[loss=0.3037, simple_loss=0.3614, pruned_loss=0.123, over 28565.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3588, pruned_loss=0.1115, over 5658514.88 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3681, pruned_loss=0.1215, over 5743289.91 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3593, pruned_loss=0.1115, over 5653214.89 frames. ], batch size: 85, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:20:11,183 INFO [optim.py:369] (1/2) 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 22:20:31,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:33,195 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:35,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:37,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:20:38,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=577745.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:20:41,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=577748.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:20:48,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4310, 2.9297, 1.5262, 1.5661], device='cuda:1'), covar=tensor([0.0845, 0.0318, 0.0893, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0521, 0.0350, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 22:20:58,294 INFO [train.py:968] (1/2) Epoch 13, batch 30300, giga_loss[loss=0.2477, simple_loss=0.3364, pruned_loss=0.07949, over 28948.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.356, pruned_loss=0.1084, over 5659192.75 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3679, pruned_loss=0.1216, over 5743044.66 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3563, pruned_loss=0.1081, over 5653685.33 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:21:02,149 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:21:04,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:21:08,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=577777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:21:46,305 INFO [train.py:968] (1/2) Epoch 13, batch 30350, giga_loss[loss=0.2597, simple_loss=0.3482, pruned_loss=0.08557, over 28987.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3552, pruned_loss=0.1063, over 5633399.73 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3683, pruned_loss=0.1223, over 5725221.07 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3548, pruned_loss=0.1052, over 5642753.85 frames. ], batch size: 164, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:21:48,757 INFO [optim.py:369] (1/2) 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 22:22:36,473 INFO [train.py:968] (1/2) Epoch 13, batch 30400, giga_loss[loss=0.257, simple_loss=0.3404, pruned_loss=0.0868, over 28744.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3562, pruned_loss=0.1071, over 5639367.75 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.3679, pruned_loss=0.1222, over 5729240.28 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3559, pruned_loss=0.1058, over 5640743.67 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:23:27,406 INFO [train.py:968] (1/2) Epoch 13, batch 30450, libri_loss[loss=0.286, simple_loss=0.347, pruned_loss=0.1125, over 25698.00 frames. ], tot_loss[loss=0.284, simple_loss=0.355, pruned_loss=0.1065, over 5639230.03 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3672, pruned_loss=0.122, over 5730224.26 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3551, pruned_loss=0.1053, over 5637688.87 frames. ], batch size: 136, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:23:31,228 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 30500, libri_loss[loss=0.2677, simple_loss=0.3287, pruned_loss=0.1033, over 29557.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3522, pruned_loss=0.1048, over 5638441.35 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3669, pruned_loss=0.1221, over 5727099.02 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.1031, over 5635462.16 frames. ], batch size: 80, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:25:05,435 INFO [train.py:968] (1/2) Epoch 13, batch 30550, giga_loss[loss=0.2854, simple_loss=0.3602, pruned_loss=0.1053, over 29040.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3493, pruned_loss=0.1029, over 5644053.93 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3658, pruned_loss=0.1215, over 5731809.15 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3498, pruned_loss=0.1015, over 5634743.40 frames. ], batch size: 155, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:25:07,527 INFO [optim.py:369] (1/2) 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 22:25:48,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-06 22:25:52,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8197, 1.9838, 1.3134, 1.5719], device='cuda:1'), covar=tensor([0.0818, 0.0623, 0.0982, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0441, 0.0501, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 22:25:53,901 INFO [train.py:968] (1/2) Epoch 13, batch 30600, libri_loss[loss=0.3325, simple_loss=0.3908, pruned_loss=0.1372, over 29212.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3493, pruned_loss=0.1025, over 5643230.71 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3655, pruned_loss=0.1215, over 5726558.54 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3495, pruned_loss=0.1008, over 5637939.87 frames. ], batch size: 97, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:26:40,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2982, 1.6292, 1.5705, 1.1563], device='cuda:1'), covar=tensor([0.1712, 0.2601, 0.1472, 0.1716], device='cuda:1'), in_proj_covar=tensor([0.0845, 0.0689, 0.0884, 0.0790], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 22:26:43,188 INFO [train.py:968] (1/2) Epoch 13, batch 30650, giga_loss[loss=0.2725, simple_loss=0.3532, pruned_loss=0.09589, over 28912.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3473, pruned_loss=0.1004, over 5639599.36 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.365, pruned_loss=0.1212, over 5719338.60 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3476, pruned_loss=0.09898, over 5640105.94 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:26:45,017 INFO [optim.py:369] (1/2) 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 22:26:58,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2691, 3.0575, 1.3913, 1.4945], device='cuda:1'), covar=tensor([0.0986, 0.0287, 0.0930, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0518, 0.0350, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-06 22:27:30,412 INFO [train.py:968] (1/2) Epoch 13, batch 30700, giga_loss[loss=0.2422, simple_loss=0.3097, pruned_loss=0.08739, over 24163.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3448, pruned_loss=0.09831, over 5646231.70 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3647, pruned_loss=0.1213, over 5721059.07 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3448, pruned_loss=0.0965, over 5642871.14 frames. ], batch size: 705, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:28:19,620 INFO [train.py:968] (1/2) Epoch 13, batch 30750, giga_loss[loss=0.2275, simple_loss=0.3108, pruned_loss=0.07212, over 28653.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3416, pruned_loss=0.09623, over 5634582.94 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.365, pruned_loss=0.1216, over 5711560.58 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3409, pruned_loss=0.09412, over 5638479.35 frames. ], batch size: 242, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 22:28:22,437 INFO [optim.py:369] (1/2) 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 22:28:32,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5938, 1.8230, 1.4403, 1.6895], device='cuda:1'), covar=tensor([0.2766, 0.2425, 0.2901, 0.2132], device='cuda:1'), in_proj_covar=tensor([0.1348, 0.0985, 0.1192, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 22:28:53,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2000, 2.9619, 1.3655, 1.3688], device='cuda:1'), covar=tensor([0.1014, 0.0350, 0.0975, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0521, 0.0352, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 22:29:06,885 INFO [train.py:968] (1/2) Epoch 13, batch 30800, giga_loss[loss=0.3222, simple_loss=0.3798, pruned_loss=0.1323, over 28667.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3409, pruned_loss=0.09681, over 5622886.43 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3649, pruned_loss=0.1217, over 5694402.44 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3399, pruned_loss=0.09437, over 5638750.16 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:29:33,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:29:57,009 INFO [train.py:968] (1/2) Epoch 13, batch 30850, libri_loss[loss=0.3105, simple_loss=0.3713, pruned_loss=0.1248, over 29517.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3394, pruned_loss=0.09678, over 5621634.52 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3647, pruned_loss=0.1217, over 5698136.26 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3382, pruned_loss=0.0942, over 5629237.12 frames. ], batch size: 82, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:30:01,136 INFO [optim.py:369] (1/2) 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 22:30:01,543 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-06 22:30:33,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5449, 1.7157, 1.3724, 1.8302], device='cuda:1'), covar=tensor([0.2730, 0.2553, 0.2851, 0.2386], device='cuda:1'), in_proj_covar=tensor([0.1349, 0.0985, 0.1191, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 22:30:44,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3148, 1.2604, 3.7435, 3.2180], device='cuda:1'), covar=tensor([0.1496, 0.2735, 0.0418, 0.1897], device='cuda:1'), in_proj_covar=tensor([0.0681, 0.0602, 0.0874, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 22:30:51,641 INFO [train.py:968] (1/2) Epoch 13, batch 30900, giga_loss[loss=0.3057, simple_loss=0.364, pruned_loss=0.1237, over 26512.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3404, pruned_loss=0.09721, over 5615985.47 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3646, pruned_loss=0.1217, over 5700080.38 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3392, pruned_loss=0.09502, over 5619601.72 frames. ], batch size: 555, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:31:48,394 INFO [train.py:968] (1/2) Epoch 13, batch 30950, giga_loss[loss=0.2662, simple_loss=0.3499, pruned_loss=0.09131, over 28936.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3431, pruned_loss=0.09743, over 5622270.14 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3643, pruned_loss=0.1215, over 5694267.03 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3421, pruned_loss=0.09532, over 5629337.94 frames. ], batch size: 145, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:31:52,027 INFO [optim.py:369] (1/2) 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 22:32:32,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3864, 2.0354, 1.4401, 0.6174], device='cuda:1'), covar=tensor([0.3782, 0.2146, 0.3480, 0.4742], device='cuda:1'), in_proj_covar=tensor([0.1592, 0.1516, 0.1495, 0.1307], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-06 22:32:49,519 INFO [train.py:968] (1/2) Epoch 13, batch 31000, libri_loss[loss=0.263, simple_loss=0.3184, pruned_loss=0.1038, over 29643.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3439, pruned_loss=0.09689, over 5643090.24 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.364, pruned_loss=0.1213, over 5697360.48 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3431, pruned_loss=0.09507, over 5644980.05 frames. ], batch size: 69, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:33:14,159 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:33:42,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-06 22:33:54,325 INFO [train.py:968] (1/2) Epoch 13, batch 31050, giga_loss[loss=0.2774, simple_loss=0.3554, pruned_loss=0.0997, over 29004.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3419, pruned_loss=0.09547, over 5654953.93 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3634, pruned_loss=0.121, over 5699379.39 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3414, pruned_loss=0.09379, over 5653949.54 frames. ], batch size: 213, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:33:57,619 INFO [optim.py:369] (1/2) 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 22:34:04,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3530, 1.3016, 1.1465, 1.4826], device='cuda:1'), covar=tensor([0.0755, 0.0328, 0.0345, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 22:34:52,060 INFO [train.py:968] (1/2) Epoch 13, batch 31100, giga_loss[loss=0.2522, simple_loss=0.3326, pruned_loss=0.08589, over 28958.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09366, over 5650372.28 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3631, pruned_loss=0.1208, over 5695186.97 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3389, pruned_loss=0.09183, over 5651837.42 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:35:12,311 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:35:36,611 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:35:53,478 INFO [train.py:968] (1/2) Epoch 13, batch 31150, giga_loss[loss=0.3088, simple_loss=0.3546, pruned_loss=0.1315, over 26863.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3393, pruned_loss=0.09288, over 5655270.98 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3629, pruned_loss=0.1211, over 5698204.96 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3383, pruned_loss=0.09058, over 5652873.10 frames. ], batch size: 555, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:35:57,567 INFO [optim.py:369] (1/2) 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 22:36:40,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4483, 4.2542, 4.0460, 1.9346], device='cuda:1'), covar=tensor([0.0551, 0.0774, 0.0772, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1091, 0.1012, 0.0876, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-06 22:36:51,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8490, 2.0209, 1.4299, 1.5817], device='cuda:1'), covar=tensor([0.0835, 0.0505, 0.0966, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0436, 0.0500, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 22:36:51,642 INFO [train.py:968] (1/2) Epoch 13, batch 31200, giga_loss[loss=0.2501, simple_loss=0.3172, pruned_loss=0.09149, over 28729.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.336, pruned_loss=0.09184, over 5662416.06 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3625, pruned_loss=0.1211, over 5701527.38 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3348, pruned_loss=0.08909, over 5656408.41 frames. ], batch size: 99, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:36:53,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:37:22,829 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-06 22:37:49,828 INFO [train.py:968] (1/2) Epoch 13, batch 31250, giga_loss[loss=0.2708, simple_loss=0.3551, pruned_loss=0.09328, over 28457.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3349, pruned_loss=0.0919, over 5656628.94 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3618, pruned_loss=0.1207, over 5696410.67 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3339, pruned_loss=0.08922, over 5655299.24 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:37:55,167 INFO [optim.py:369] (1/2) 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 22:37:55,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4183, 1.5774, 1.3839, 1.6112], device='cuda:1'), covar=tensor([0.0751, 0.0307, 0.0325, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-06 22:38:46,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:38:47,849 INFO [train.py:968] (1/2) Epoch 13, batch 31300, giga_loss[loss=0.294, simple_loss=0.3635, pruned_loss=0.1123, over 28912.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3352, pruned_loss=0.09247, over 5665205.22 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3614, pruned_loss=0.1205, over 5700505.11 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3342, pruned_loss=0.09009, over 5660094.85 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:39:09,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:39:36,707 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-06 22:39:39,445 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=578809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:39:42,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=578812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:39:45,397 INFO [train.py:968] (1/2) Epoch 13, batch 31350, giga_loss[loss=0.2734, simple_loss=0.3505, pruned_loss=0.0981, over 28985.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3366, pruned_loss=0.09233, over 5665198.86 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3603, pruned_loss=0.1198, over 5704623.14 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3363, pruned_loss=0.09049, over 5657064.76 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:39:53,338 INFO [optim.py:369] (1/2) 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 22:40:14,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1835, 2.6249, 1.2176, 1.4968], device='cuda:1'), covar=tensor([0.0976, 0.0318, 0.0930, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0520, 0.0353, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 22:40:19,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=578841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:40:33,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5039, 3.1987, 1.6151, 1.5014], device='cuda:1'), covar=tensor([0.0850, 0.0283, 0.0841, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0519, 0.0352, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 22:40:40,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:40:47,704 INFO [train.py:968] (1/2) Epoch 13, batch 31400, giga_loss[loss=0.2493, simple_loss=0.3325, pruned_loss=0.08306, over 28704.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3392, pruned_loss=0.09352, over 5672438.49 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3602, pruned_loss=0.1199, over 5708617.98 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3382, pruned_loss=0.09097, over 5660899.93 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:41:16,158 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 31450, giga_loss[loss=0.2602, simple_loss=0.3321, pruned_loss=0.0942, over 28603.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3359, pruned_loss=0.09164, over 5663853.92 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3603, pruned_loss=0.1199, over 5702153.84 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3344, pruned_loss=0.0889, over 5658617.47 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:41:49,539 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:41:52,950 INFO [optim.py:369] (1/2) 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 22:42:40,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:42:45,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1731, 1.1269, 3.4265, 2.9868], device='cuda:1'), covar=tensor([0.1559, 0.2800, 0.0423, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.0677, 0.0602, 0.0870, 0.0787], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 22:42:49,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2249, 1.4734, 1.4093, 1.1342], device='cuda:1'), covar=tensor([0.2124, 0.1618, 0.1227, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.1738, 0.1644, 0.1596, 0.1697], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 22:42:53,640 INFO [train.py:968] (1/2) Epoch 13, batch 31500, giga_loss[loss=0.2492, simple_loss=0.3407, pruned_loss=0.07889, over 29001.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3372, pruned_loss=0.09255, over 5673111.37 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3592, pruned_loss=0.1191, over 5707154.55 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3365, pruned_loss=0.09043, over 5663731.30 frames. ], batch size: 200, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:43:05,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=578974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:09,381 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=578978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:35,206 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:38,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:43:51,910 INFO [train.py:968] (1/2) Epoch 13, batch 31550, giga_loss[loss=0.2478, simple_loss=0.3495, pruned_loss=0.07308, over 28476.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.34, pruned_loss=0.09311, over 5655969.78 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3588, pruned_loss=0.1191, over 5699180.72 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3391, pruned_loss=0.09054, over 5654987.70 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:43:58,619 INFO [optim.py:369] (1/2) 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 22:44:15,661 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:44:54,764 INFO [train.py:968] (1/2) Epoch 13, batch 31600, giga_loss[loss=0.2645, simple_loss=0.3525, pruned_loss=0.08823, over 28437.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3429, pruned_loss=0.09213, over 5663472.69 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.359, pruned_loss=0.1192, over 5702863.54 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3418, pruned_loss=0.0895, over 5658710.26 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:45:33,598 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:45:38,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:45:53,802 INFO [train.py:968] (1/2) Epoch 13, batch 31650, giga_loss[loss=0.2806, simple_loss=0.349, pruned_loss=0.1061, over 26940.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3424, pruned_loss=0.0909, over 5654464.50 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3584, pruned_loss=0.119, over 5704408.55 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.0884, over 5648224.82 frames. ], batch size: 555, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:45:56,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:45:59,228 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-06 22:45:59,360 INFO [optim.py:369] (1/2) 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 22:46:00,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:02,842 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579122.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:46:12,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:20,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:36,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:48,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:46:52,216 INFO [train.py:968] (1/2) Epoch 13, batch 31700, giga_loss[loss=0.2285, simple_loss=0.3266, pruned_loss=0.06525, over 28411.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3435, pruned_loss=0.09133, over 5655496.79 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3584, pruned_loss=0.1191, over 5702633.01 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3425, pruned_loss=0.0886, over 5650634.40 frames. ], batch size: 60, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:47:54,615 INFO [train.py:968] (1/2) Epoch 13, batch 31750, giga_loss[loss=0.2363, simple_loss=0.3201, pruned_loss=0.0763, over 28916.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3442, pruned_loss=0.09249, over 5650379.58 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3587, pruned_loss=0.1194, over 5695781.80 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3428, pruned_loss=0.08957, over 5652002.56 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:47:55,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-06 22:48:03,895 INFO [optim.py:369] (1/2) 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 22:48:21,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:48:37,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:48:38,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2441, 1.2555, 1.1070, 1.5143], device='cuda:1'), covar=tensor([0.0776, 0.0329, 0.0352, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-06 22:49:02,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:02,630 INFO [train.py:968] (1/2) Epoch 13, batch 31800, giga_loss[loss=0.266, simple_loss=0.3379, pruned_loss=0.09707, over 28924.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3433, pruned_loss=0.09312, over 5659372.03 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3581, pruned_loss=0.1189, over 5699802.22 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3425, pruned_loss=0.09064, over 5655950.41 frames. ], batch size: 106, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:49:28,992 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:32,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:39,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:49:59,563 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:50:18,244 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 31850, giga_loss[loss=0.3218, simple_loss=0.3825, pruned_loss=0.1306, over 28193.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3442, pruned_loss=0.09452, over 5663169.62 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3579, pruned_loss=0.1188, over 5695610.21 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3432, pruned_loss=0.09184, over 5662589.60 frames. ], batch size: 412, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:50:27,947 INFO [optim.py:369] (1/2) 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 22:50:47,094 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:51:15,165 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 31900, giga_loss[loss=0.2307, simple_loss=0.3128, pruned_loss=0.07428, over 28842.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3389, pruned_loss=0.09144, over 5661430.94 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3573, pruned_loss=0.1185, over 5690343.70 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3384, pruned_loss=0.08899, over 5664660.07 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:52:19,402 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:52:23,625 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 31950, giga_loss[loss=0.2429, simple_loss=0.3186, pruned_loss=0.08356, over 28912.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3372, pruned_loss=0.09074, over 5669725.02 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3563, pruned_loss=0.1178, over 5697920.01 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3369, pruned_loss=0.08834, over 5664458.73 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:52:38,413 INFO [optim.py:369] (1/2) 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 22:52:58,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:53:01,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:53:03,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:53:33,723 INFO [train.py:968] (1/2) Epoch 13, batch 32000, giga_loss[loss=0.299, simple_loss=0.3699, pruned_loss=0.114, over 28982.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3361, pruned_loss=0.0907, over 5674254.62 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3561, pruned_loss=0.1179, over 5702096.67 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3356, pruned_loss=0.08823, over 5665763.60 frames. ], batch size: 199, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 22:53:37,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:54:13,537 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:54:14,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 22:54:15,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:54:37,233 INFO [train.py:968] (1/2) Epoch 13, batch 32050, giga_loss[loss=0.2536, simple_loss=0.3424, pruned_loss=0.08236, over 28700.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3403, pruned_loss=0.09305, over 5673444.59 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3561, pruned_loss=0.118, over 5705626.64 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3395, pruned_loss=0.09036, over 5662840.21 frames. ], batch size: 262, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:54:44,464 INFO [optim.py:369] (1/2) 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 22:54:52,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:55:39,746 INFO [train.py:968] (1/2) Epoch 13, batch 32100, giga_loss[loss=0.2438, simple_loss=0.3209, pruned_loss=0.08336, over 28958.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3404, pruned_loss=0.09305, over 5676311.21 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3559, pruned_loss=0.1179, over 5707332.98 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3399, pruned_loss=0.09084, over 5666195.61 frames. ], batch size: 119, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:56:38,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:56:46,392 INFO [train.py:968] (1/2) Epoch 13, batch 32150, giga_loss[loss=0.278, simple_loss=0.3486, pruned_loss=0.1037, over 28917.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3397, pruned_loss=0.09397, over 5672471.95 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.356, pruned_loss=0.118, over 5709595.18 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.339, pruned_loss=0.09176, over 5661813.62 frames. ], batch size: 213, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:56:54,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=579621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:56:54,788 INFO [optim.py:369] (1/2) 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 22:57:18,167 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:57:47,696 INFO [train.py:968] (1/2) Epoch 13, batch 32200, giga_loss[loss=0.2558, simple_loss=0.3308, pruned_loss=0.09039, over 27661.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3397, pruned_loss=0.09401, over 5669816.95 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3555, pruned_loss=0.1178, over 5709264.32 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3394, pruned_loss=0.09216, over 5661365.86 frames. ], batch size: 474, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:57:56,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579672.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 22:58:53,922 INFO [train.py:968] (1/2) Epoch 13, batch 32250, libri_loss[loss=0.342, simple_loss=0.3983, pruned_loss=0.1429, over 29772.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3419, pruned_loss=0.09478, over 5670995.33 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3559, pruned_loss=0.1182, over 5711746.88 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3409, pruned_loss=0.09223, over 5660632.40 frames. ], batch size: 87, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 22:59:06,605 INFO [optim.py:369] (1/2) 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 22:59:43,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7285, 4.9152, 1.9335, 1.9368], device='cuda:1'), covar=tensor([0.0888, 0.0267, 0.0838, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0516, 0.0351, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 22:59:48,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 22:59:52,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:04,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=579764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:05,002 INFO [train.py:968] (1/2) Epoch 13, batch 32300, giga_loss[loss=0.2572, simple_loss=0.3436, pruned_loss=0.08545, over 28970.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3438, pruned_loss=0.09501, over 5682428.47 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.356, pruned_loss=0.1182, over 5715299.95 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3426, pruned_loss=0.09234, over 5669855.67 frames. ], batch size: 128, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:00:11,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=579767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:20,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5374, 2.1793, 1.3902, 0.8312], device='cuda:1'), covar=tensor([0.4834, 0.2547, 0.2649, 0.4708], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1521, 0.1501, 0.1321], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 23:00:25,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4916, 1.8402, 1.6039, 1.6300], device='cuda:1'), covar=tensor([0.1598, 0.1861, 0.2011, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0717, 0.0663, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 23:00:37,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:56,303 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=579796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:00:57,549 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=579797.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:01:18,877 INFO [train.py:968] (1/2) Epoch 13, batch 32350, giga_loss[loss=0.2478, simple_loss=0.3243, pruned_loss=0.08567, over 28923.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3417, pruned_loss=0.09382, over 5672408.54 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3558, pruned_loss=0.1181, over 5714263.91 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3406, pruned_loss=0.09136, over 5662658.53 frames. ], batch size: 106, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:01:27,328 INFO [optim.py:369] (1/2) 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 23:02:00,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3332, 1.9108, 1.4452, 0.4216], device='cuda:1'), covar=tensor([0.3041, 0.2020, 0.3461, 0.4359], device='cuda:1'), in_proj_covar=tensor([0.1600, 0.1524, 0.1504, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 23:02:15,201 INFO [train.py:968] (1/2) Epoch 13, batch 32400, giga_loss[loss=0.2129, simple_loss=0.2965, pruned_loss=0.06463, over 28085.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3377, pruned_loss=0.09334, over 5674281.97 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3548, pruned_loss=0.1176, over 5715071.02 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3369, pruned_loss=0.0905, over 5664364.35 frames. ], batch size: 412, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:03:15,368 INFO [train.py:968] (1/2) Epoch 13, batch 32450, giga_loss[loss=0.2647, simple_loss=0.3329, pruned_loss=0.09821, over 27863.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3322, pruned_loss=0.0908, over 5680275.94 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3545, pruned_loss=0.1173, over 5719489.13 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3312, pruned_loss=0.08793, over 5667178.44 frames. ], batch size: 476, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:03:22,224 INFO [optim.py:369] (1/2) 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 23:03:52,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3705, 1.3662, 1.2887, 1.5457], device='cuda:1'), covar=tensor([0.0752, 0.0319, 0.0321, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0116, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-06 23:03:58,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8412, 1.9030, 1.4238, 1.5047], device='cuda:1'), covar=tensor([0.0687, 0.0468, 0.0927, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0436, 0.0503, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 23:04:11,455 INFO [train.py:968] (1/2) Epoch 13, batch 32500, giga_loss[loss=0.3135, simple_loss=0.3693, pruned_loss=0.1288, over 28505.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3337, pruned_loss=0.09261, over 5664667.61 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3541, pruned_loss=0.1173, over 5715478.83 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3322, pruned_loss=0.08914, over 5656456.53 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:04:16,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1742, 1.2431, 3.6111, 3.0822], device='cuda:1'), covar=tensor([0.1617, 0.2680, 0.0437, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0602, 0.0872, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 23:05:03,007 INFO [train.py:968] (1/2) Epoch 13, batch 32550, giga_loss[loss=0.2667, simple_loss=0.3452, pruned_loss=0.09412, over 28933.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3363, pruned_loss=0.09453, over 5643016.50 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3544, pruned_loss=0.1175, over 5691739.21 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3341, pruned_loss=0.09065, over 5654176.51 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:05:09,888 INFO [optim.py:369] (1/2) 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 23:05:45,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2612, 1.5880, 1.2288, 0.9752], device='cuda:1'), covar=tensor([0.2507, 0.2356, 0.2652, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1351, 0.0988, 0.1193, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:06:01,665 INFO [train.py:968] (1/2) Epoch 13, batch 32600, giga_loss[loss=0.217, simple_loss=0.3058, pruned_loss=0.06409, over 28861.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3339, pruned_loss=0.09277, over 5643388.48 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.354, pruned_loss=0.1173, over 5688276.41 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.332, pruned_loss=0.08915, over 5653538.35 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:06:11,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3505, 3.1288, 1.5890, 1.4323], device='cuda:1'), covar=tensor([0.0884, 0.0309, 0.0849, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0516, 0.0351, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 23:06:38,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4038, 1.8047, 1.4399, 1.5481], device='cuda:1'), covar=tensor([0.0790, 0.0279, 0.0331, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0116, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-06 23:06:54,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3664, 4.2028, 3.9469, 2.1619], device='cuda:1'), covar=tensor([0.0566, 0.0766, 0.0909, 0.1938], device='cuda:1'), in_proj_covar=tensor([0.1081, 0.1003, 0.0871, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 23:07:03,326 INFO [train.py:968] (1/2) Epoch 13, batch 32650, giga_loss[loss=0.2342, simple_loss=0.3141, pruned_loss=0.07712, over 27590.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3321, pruned_loss=0.09048, over 5644436.85 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3535, pruned_loss=0.117, over 5683674.79 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3306, pruned_loss=0.08732, over 5655088.37 frames. ], batch size: 472, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:07:06,035 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-06 23:07:12,851 INFO [optim.py:369] (1/2) 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 23:07:26,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-06 23:07:28,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:08:09,745 INFO [train.py:968] (1/2) Epoch 13, batch 32700, giga_loss[loss=0.2053, simple_loss=0.2951, pruned_loss=0.0578, over 28384.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3302, pruned_loss=0.08943, over 5651094.67 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3533, pruned_loss=0.1168, over 5685416.01 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08669, over 5657463.55 frames. ], batch size: 65, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:08:18,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580172.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:09:20,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9922, 1.3281, 1.0636, 0.2116], device='cuda:1'), covar=tensor([0.2765, 0.2350, 0.3714, 0.4806], device='cuda:1'), in_proj_covar=tensor([0.1604, 0.1532, 0.1509, 0.1332], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 23:09:20,785 INFO [train.py:968] (1/2) Epoch 13, batch 32750, giga_loss[loss=0.2772, simple_loss=0.3529, pruned_loss=0.1007, over 28887.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3291, pruned_loss=0.0882, over 5637303.59 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3538, pruned_loss=0.1171, over 5677923.14 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3274, pruned_loss=0.08547, over 5647834.38 frames. ], batch size: 227, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:09:30,678 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 32800, giga_loss[loss=0.2364, simple_loss=0.2985, pruned_loss=0.08714, over 24432.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3288, pruned_loss=0.08788, over 5642799.10 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3532, pruned_loss=0.1168, over 5681091.93 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3276, pruned_loss=0.08561, over 5648035.63 frames. ], batch size: 705, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:11:25,116 INFO [train.py:968] (1/2) Epoch 13, batch 32850, giga_loss[loss=0.2437, simple_loss=0.3221, pruned_loss=0.08264, over 28906.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3307, pruned_loss=0.08981, over 5654392.78 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3529, pruned_loss=0.1167, over 5686559.24 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3294, pruned_loss=0.08745, over 5652852.42 frames. ], batch size: 186, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:11:25,753 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=580315.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:11:30,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=580318.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:11:36,638 INFO [optim.py:369] (1/2) 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 23:12:09,487 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 32900, libri_loss[loss=0.2689, simple_loss=0.3343, pruned_loss=0.1018, over 29560.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3294, pruned_loss=0.08872, over 5657486.05 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3527, pruned_loss=0.1166, over 5687386.88 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3281, pruned_loss=0.08627, over 5654486.01 frames. ], batch size: 77, lr: 2.46e-03, grad_scale: 8.0
+2023-03-06 23:12:50,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5116, 1.6863, 1.3823, 1.6861], device='cuda:1'), covar=tensor([0.2558, 0.2436, 0.2769, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.1345, 0.0982, 0.1190, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:13:02,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.34 vs. limit=5.0
+2023-03-06 23:13:28,383 INFO [train.py:968] (1/2) Epoch 13, batch 32950, giga_loss[loss=0.3014, simple_loss=0.3737, pruned_loss=0.1146, over 28480.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3303, pruned_loss=0.0874, over 5653851.78 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3527, pruned_loss=0.1166, over 5685941.40 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3291, pruned_loss=0.0852, over 5652589.02 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:13:35,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4477, 1.6429, 1.6869, 1.2409], device='cuda:1'), covar=tensor([0.1695, 0.2525, 0.1470, 0.1813], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0683, 0.0886, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 23:13:35,774 INFO [optim.py:369] (1/2) 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 23:14:24,126 INFO [train.py:968] (1/2) Epoch 13, batch 33000, libri_loss[loss=0.2874, simple_loss=0.3568, pruned_loss=0.109, over 29529.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3336, pruned_loss=0.08859, over 5660889.50 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3526, pruned_loss=0.1164, over 5692672.12 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3321, pruned_loss=0.08619, over 5652812.88 frames. ], batch size: 89, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:14:24,127 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-06 23:14:31,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2291, 1.6749, 1.2639, 0.4020], device='cuda:1'), covar=tensor([0.3421, 0.2411, 0.3895, 0.5114], device='cuda:1'), in_proj_covar=tensor([0.1602, 0.1531, 0.1510, 0.1324], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 23:14:32,735 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-06 23:14:44,549 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:15:12,219 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:15:24,546 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 33050, giga_loss[loss=0.2974, simple_loss=0.3688, pruned_loss=0.113, over 28495.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3344, pruned_loss=0.08912, over 5650490.19 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3525, pruned_loss=0.1163, over 5695820.47 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.333, pruned_loss=0.08678, over 5640848.74 frames. ], batch size: 336, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:15:46,961 INFO [optim.py:369] (1/2) 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 23:16:39,129 INFO [train.py:968] (1/2) Epoch 13, batch 33100, giga_loss[loss=0.3009, simple_loss=0.3783, pruned_loss=0.1117, over 28663.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3351, pruned_loss=0.08961, over 5653342.23 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3525, pruned_loss=0.1163, over 5689366.25 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3337, pruned_loss=0.08741, over 5650952.05 frames. ], batch size: 307, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 23:17:17,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:17:37,949 INFO [train.py:968] (1/2) Epoch 13, batch 33150, giga_loss[loss=0.234, simple_loss=0.3076, pruned_loss=0.08021, over 24672.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3333, pruned_loss=0.08842, over 5647491.06 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3524, pruned_loss=0.1163, over 5681384.55 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3318, pruned_loss=0.08613, over 5651481.00 frames. ], batch size: 705, lr: 2.46e-03, grad_scale: 2.0
+2023-03-06 23:17:49,948 INFO [optim.py:369] (1/2) 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 23:18:18,808 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=580651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:18:22,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=580654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:18:22,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-06 23:18:33,915 INFO [train.py:968] (1/2) Epoch 13, batch 33200, giga_loss[loss=0.2352, simple_loss=0.3206, pruned_loss=0.07491, over 28943.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3327, pruned_loss=0.08864, over 5659510.22 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3519, pruned_loss=0.116, over 5689383.12 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3314, pruned_loss=0.08604, over 5654282.03 frames. ], batch size: 284, lr: 2.46e-03, grad_scale: 4.0
+2023-03-06 23:18:43,850 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-06 23:18:53,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=580683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:19:30,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4295, 4.2372, 4.0085, 1.9584], device='cuda:1'), covar=tensor([0.0568, 0.0719, 0.0761, 0.2044], device='cuda:1'), in_proj_covar=tensor([0.1076, 0.0994, 0.0863, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 23:19:30,531 INFO [train.py:968] (1/2) Epoch 13, batch 33250, giga_loss[loss=0.2151, simple_loss=0.3037, pruned_loss=0.06326, over 28874.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.33, pruned_loss=0.08789, over 5659811.11 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3516, pruned_loss=0.1158, over 5684830.28 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3286, pruned_loss=0.08528, over 5658270.78 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:19:45,681 INFO [optim.py:369] (1/2) 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 23:19:49,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3251, 3.2918, 1.4819, 1.4828], device='cuda:1'), covar=tensor([0.0979, 0.0258, 0.0953, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0517, 0.0351, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 23:20:31,587 INFO [train.py:968] (1/2) Epoch 13, batch 33300, giga_loss[loss=0.3001, simple_loss=0.3723, pruned_loss=0.114, over 27647.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3322, pruned_loss=0.08869, over 5667137.81 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3513, pruned_loss=0.1156, over 5688107.95 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3311, pruned_loss=0.08636, over 5662563.22 frames. ], batch size: 474, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:21:33,716 INFO [train.py:968] (1/2) Epoch 13, batch 33350, giga_loss[loss=0.1973, simple_loss=0.2779, pruned_loss=0.05835, over 28455.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3343, pruned_loss=0.0901, over 5670039.76 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3511, pruned_loss=0.1155, over 5693351.02 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.0876, over 5660939.26 frames. ], batch size: 71, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:21:44,693 INFO [optim.py:369] (1/2) 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 23:22:08,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3550, 1.2202, 1.1626, 1.5462], device='cuda:1'), covar=tensor([0.0735, 0.0367, 0.0342, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-06 23:22:18,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:22:40,518 INFO [train.py:968] (1/2) Epoch 13, batch 33400, giga_loss[loss=0.2552, simple_loss=0.3392, pruned_loss=0.08557, over 28913.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.336, pruned_loss=0.09151, over 5660119.59 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3514, pruned_loss=0.1157, over 5685067.34 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3346, pruned_loss=0.08902, over 5660434.09 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:22:51,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=580874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:23:18,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4098, 1.6339, 1.2428, 1.5714], device='cuda:1'), covar=tensor([0.0760, 0.0307, 0.0349, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-06 23:23:41,670 INFO [train.py:968] (1/2) Epoch 13, batch 33450, giga_loss[loss=0.2535, simple_loss=0.3376, pruned_loss=0.08467, over 29055.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3396, pruned_loss=0.09334, over 5669984.27 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3512, pruned_loss=0.1156, over 5689255.51 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3383, pruned_loss=0.09087, over 5665933.72 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:23:54,683 INFO [optim.py:369] (1/2) 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 23:24:39,010 INFO [train.py:968] (1/2) Epoch 13, batch 33500, giga_loss[loss=0.281, simple_loss=0.3592, pruned_loss=0.1014, over 28401.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3415, pruned_loss=0.09347, over 5655951.85 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3512, pruned_loss=0.1156, over 5681663.62 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3403, pruned_loss=0.09117, over 5659038.39 frames. ], batch size: 368, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:24:46,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4861, 3.7529, 1.5165, 1.6999], device='cuda:1'), covar=tensor([0.0905, 0.0325, 0.0938, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0515, 0.0350, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 23:24:46,106 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=580994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:25:18,674 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=580995.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:25:21,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=580997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:25:36,277 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-06 23:25:44,337 INFO [train.py:968] (1/2) Epoch 13, batch 33550, giga_loss[loss=0.2521, simple_loss=0.3444, pruned_loss=0.07993, over 28707.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3411, pruned_loss=0.09342, over 5661165.51 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.351, pruned_loss=0.1156, over 5686288.58 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3402, pruned_loss=0.09119, over 5659308.74 frames. ], batch size: 307, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:25:50,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:25:57,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:26:02,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5119, 3.3465, 3.1287, 1.8992], device='cuda:1'), covar=tensor([0.0694, 0.0897, 0.0907, 0.1861], device='cuda:1'), in_proj_covar=tensor([0.1071, 0.0988, 0.0860, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 23:26:05,065 INFO [optim.py:369] (1/2) 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 23:26:05,731 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:26:26,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:26:48,284 INFO [train.py:968] (1/2) Epoch 13, batch 33600, giga_loss[loss=0.2396, simple_loss=0.3222, pruned_loss=0.07852, over 28834.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3388, pruned_loss=0.09298, over 5656406.68 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3512, pruned_loss=0.1159, over 5681651.78 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3373, pruned_loss=0.08974, over 5657682.27 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:27:47,831 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:27:48,093 INFO [train.py:968] (1/2) Epoch 13, batch 33650, giga_loss[loss=0.3167, simple_loss=0.3796, pruned_loss=0.1269, over 28515.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3385, pruned_loss=0.09316, over 5654108.76 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3511, pruned_loss=0.1158, over 5678170.25 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.09008, over 5657854.42 frames. ], batch size: 368, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:27:51,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:28:04,932 INFO [optim.py:369] (1/2) 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 23:28:30,524 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:28:31,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:28:47,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4695, 1.8106, 1.4708, 1.5258], device='cuda:1'), covar=tensor([0.2409, 0.2098, 0.2409, 0.1859], device='cuda:1'), in_proj_covar=tensor([0.1347, 0.0983, 0.1189, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:28:54,194 INFO [train.py:968] (1/2) Epoch 13, batch 33700, giga_loss[loss=0.3026, simple_loss=0.3631, pruned_loss=0.1211, over 28015.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3375, pruned_loss=0.09284, over 5644939.50 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3509, pruned_loss=0.1158, over 5673557.50 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3363, pruned_loss=0.08997, over 5652103.41 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:29:03,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5352, 1.8936, 1.4852, 1.5819], device='cuda:1'), covar=tensor([0.2477, 0.2251, 0.2702, 0.2108], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.0983, 0.1188, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:30:01,499 INFO [train.py:968] (1/2) Epoch 13, batch 33750, giga_loss[loss=0.2314, simple_loss=0.2917, pruned_loss=0.08553, over 24616.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3368, pruned_loss=0.09326, over 5641793.73 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.351, pruned_loss=0.1159, over 5672752.45 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3355, pruned_loss=0.09061, over 5647980.87 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:30:13,434 INFO [optim.py:369] (1/2) 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 23:30:38,971 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 33800, giga_loss[loss=0.2038, simple_loss=0.2951, pruned_loss=0.0562, over 29102.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3352, pruned_loss=0.09239, over 5641821.58 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3509, pruned_loss=0.1158, over 5678402.03 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3339, pruned_loss=0.08978, over 5640861.85 frames. ], batch size: 113, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:31:13,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3777, 1.9325, 1.4087, 1.5049], device='cuda:1'), covar=tensor([0.0754, 0.0259, 0.0322, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0056, 0.0095], device='cuda:1')
+2023-03-06 23:31:32,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9150, 4.6967, 2.0234, 2.0163], device='cuda:1'), covar=tensor([0.0868, 0.0237, 0.0845, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0517, 0.0352, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-06 23:32:03,341 INFO [train.py:968] (1/2) Epoch 13, batch 33850, giga_loss[loss=0.2556, simple_loss=0.3362, pruned_loss=0.08749, over 28923.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3332, pruned_loss=0.08954, over 5657109.05 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3506, pruned_loss=0.1156, over 5680786.64 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3323, pruned_loss=0.08742, over 5653937.04 frames. ], batch size: 227, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:32:18,191 INFO [optim.py:369] (1/2) 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 23:32:58,575 INFO [train.py:968] (1/2) Epoch 13, batch 33900, giga_loss[loss=0.2255, simple_loss=0.3188, pruned_loss=0.06613, over 28955.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3347, pruned_loss=0.08875, over 5663739.29 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3502, pruned_loss=0.1153, over 5675672.74 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3338, pruned_loss=0.08661, over 5665081.27 frames. ], batch size: 199, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:33:05,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581370.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:33:58,082 INFO [train.py:968] (1/2) Epoch 13, batch 33950, giga_loss[loss=0.223, simple_loss=0.3169, pruned_loss=0.06454, over 28059.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3376, pruned_loss=0.08894, over 5653134.44 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3508, pruned_loss=0.1158, over 5668538.24 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3363, pruned_loss=0.08648, over 5660390.36 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:34:10,144 INFO [optim.py:369] (1/2) 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 23:34:21,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2230, 0.8314, 0.8326, 1.3906], device='cuda:1'), covar=tensor([0.0790, 0.0378, 0.0384, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0113, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0095], device='cuda:1')
+2023-03-06 23:34:45,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3845, 4.2025, 3.9796, 1.8129], device='cuda:1'), covar=tensor([0.0538, 0.0709, 0.0749, 0.2237], device='cuda:1'), in_proj_covar=tensor([0.1069, 0.0990, 0.0860, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 23:34:47,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:34:57,285 INFO [train.py:968] (1/2) Epoch 13, batch 34000, giga_loss[loss=0.2488, simple_loss=0.329, pruned_loss=0.08431, over 28596.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3377, pruned_loss=0.08906, over 5656655.93 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3507, pruned_loss=0.1157, over 5673717.08 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3364, pruned_loss=0.08657, over 5657161.95 frames. ], batch size: 92, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:35:00,635 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:35:50,730 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-06 23:36:02,825 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581513.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:36:04,227 INFO [train.py:968] (1/2) Epoch 13, batch 34050, giga_loss[loss=0.2845, simple_loss=0.3622, pruned_loss=0.1034, over 28638.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3376, pruned_loss=0.08922, over 5664764.03 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3505, pruned_loss=0.1156, over 5673387.43 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3365, pruned_loss=0.08671, over 5664999.91 frames. ], batch size: 307, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:36:05,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2493, 1.6209, 1.4264, 1.4481], device='cuda:1'), covar=tensor([0.1541, 0.1731, 0.2093, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0716, 0.0665, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 23:36:05,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581516.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:36:12,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:36:14,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-06 23:36:21,576 INFO [optim.py:369] (1/2) 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 23:36:42,075 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581545.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:37:06,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-06 23:37:07,049 INFO [train.py:968] (1/2) Epoch 13, batch 34100, giga_loss[loss=0.2559, simple_loss=0.3392, pruned_loss=0.0863, over 29066.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3378, pruned_loss=0.08947, over 5668631.71 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3503, pruned_loss=0.1153, over 5679796.25 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3367, pruned_loss=0.08684, over 5662782.85 frames. ], batch size: 285, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:38:19,634 INFO [train.py:968] (1/2) Epoch 13, batch 34150, giga_loss[loss=0.2573, simple_loss=0.3435, pruned_loss=0.08552, over 29046.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3372, pruned_loss=0.08884, over 5658553.93 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3502, pruned_loss=0.1154, over 5674967.17 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3362, pruned_loss=0.08628, over 5658670.34 frames. ], batch size: 285, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:38:27,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581620.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:38:39,811 INFO [optim.py:369] (1/2) 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 23:39:29,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:39:31,064 INFO [train.py:968] (1/2) Epoch 13, batch 34200, giga_loss[loss=0.2798, simple_loss=0.3648, pruned_loss=0.09743, over 28510.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3375, pruned_loss=0.08858, over 5655887.05 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3501, pruned_loss=0.1152, over 5676895.65 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3364, pruned_loss=0.08603, over 5654211.41 frames. ], batch size: 336, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:39:33,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:40:09,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:40:16,632 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9403, 1.3040, 1.0635, 0.1676], device='cuda:1'), covar=tensor([0.3110, 0.2409, 0.3848, 0.5268], device='cuda:1'), in_proj_covar=tensor([0.1591, 0.1525, 0.1501, 0.1319], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 23:40:23,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-06 23:40:35,225 INFO [train.py:968] (1/2) Epoch 13, batch 34250, giga_loss[loss=0.2541, simple_loss=0.3435, pruned_loss=0.08232, over 28911.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3401, pruned_loss=0.08976, over 5663700.16 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3491, pruned_loss=0.1147, over 5681219.86 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08752, over 5657981.78 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:40:47,640 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:41:20,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6006, 1.8209, 1.5120, 1.5623], device='cuda:1'), covar=tensor([0.1300, 0.1928, 0.1910, 0.1919], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0709, 0.0657, 0.0645], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0008], device='cuda:1')
+2023-03-06 23:41:33,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:41:37,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581763.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:41:38,311 INFO [train.py:968] (1/2) Epoch 13, batch 34300, libri_loss[loss=0.3236, simple_loss=0.3798, pruned_loss=0.1338, over 29535.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3408, pruned_loss=0.09006, over 5680431.83 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3495, pruned_loss=0.1148, over 5686333.38 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.34, pruned_loss=0.08717, over 5670780.10 frames. ], batch size: 84, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:41:41,360 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:42:20,970 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=581795.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:42:48,793 INFO [train.py:968] (1/2) Epoch 13, batch 34350, giga_loss[loss=0.2496, simple_loss=0.3354, pruned_loss=0.08193, over 29046.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3393, pruned_loss=0.08957, over 5685123.59 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3499, pruned_loss=0.1149, over 5687398.04 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.338, pruned_loss=0.08664, over 5676712.21 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:43:03,994 INFO [optim.py:369] (1/2) 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 23:43:11,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:43:29,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=581842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:43:57,066 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-06 23:44:01,770 INFO [train.py:968] (1/2) Epoch 13, batch 34400, giga_loss[loss=0.2309, simple_loss=0.321, pruned_loss=0.07039, over 27722.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3366, pruned_loss=0.08756, over 5678742.15 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3499, pruned_loss=0.115, over 5680005.04 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3355, pruned_loss=0.08486, over 5677975.67 frames. ], batch size: 472, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:44:38,749 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=581889.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:44:53,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-06 23:45:09,605 INFO [train.py:968] (1/2) Epoch 13, batch 34450, giga_loss[loss=0.232, simple_loss=0.3221, pruned_loss=0.07094, over 28506.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.335, pruned_loss=0.08594, over 5685646.57 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3501, pruned_loss=0.1152, over 5674581.15 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3337, pruned_loss=0.08318, over 5689648.91 frames. ], batch size: 336, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:45:28,640 INFO [optim.py:369] (1/2) 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 23:46:15,737 INFO [train.py:968] (1/2) Epoch 13, batch 34500, giga_loss[loss=0.2684, simple_loss=0.3299, pruned_loss=0.1034, over 24514.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3364, pruned_loss=0.08729, over 5678780.23 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3507, pruned_loss=0.1156, over 5676453.21 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3347, pruned_loss=0.08444, over 5680289.94 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:46:26,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:46:30,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:46:39,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=581985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:46:43,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=581988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:47:05,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:47:05,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5316, 1.7382, 1.3476, 1.9960], device='cuda:1'), covar=tensor([0.2463, 0.2460, 0.2715, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.1340, 0.0987, 0.1189, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:47:16,396 INFO [train.py:968] (1/2) Epoch 13, batch 34550, giga_loss[loss=0.2612, simple_loss=0.3474, pruned_loss=0.08751, over 28667.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.339, pruned_loss=0.08877, over 5667343.61 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.351, pruned_loss=0.1158, over 5669944.59 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3373, pruned_loss=0.086, over 5674201.13 frames. ], batch size: 307, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:47:20,578 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:47:31,801 INFO [optim.py:369] (1/2) 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 23:47:45,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-06 23:47:48,185 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-06 23:48:14,150 INFO [train.py:968] (1/2) Epoch 13, batch 34600, giga_loss[loss=0.2566, simple_loss=0.3293, pruned_loss=0.09193, over 27972.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3391, pruned_loss=0.08977, over 5669415.72 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3507, pruned_loss=0.1158, over 5675376.26 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3377, pruned_loss=0.08688, over 5669717.58 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:48:45,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2579, 1.3587, 1.2853, 1.2610], device='cuda:1'), covar=tensor([0.1725, 0.1539, 0.1073, 0.1404], device='cuda:1'), in_proj_covar=tensor([0.1753, 0.1634, 0.1591, 0.1692], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 23:48:54,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:48:58,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-06 23:48:59,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2781, 1.5427, 1.5631, 1.4719], device='cuda:1'), covar=tensor([0.1376, 0.1236, 0.1496, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0713, 0.0662, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-06 23:48:59,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5856, 1.8567, 1.6606, 1.4330], device='cuda:1'), covar=tensor([0.2008, 0.1617, 0.1296, 0.1583], device='cuda:1'), in_proj_covar=tensor([0.1753, 0.1632, 0.1590, 0.1691], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-06 23:49:10,088 INFO [train.py:968] (1/2) Epoch 13, batch 34650, giga_loss[loss=0.2548, simple_loss=0.333, pruned_loss=0.08836, over 28453.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3359, pruned_loss=0.08908, over 5670243.48 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3507, pruned_loss=0.1157, over 5678760.12 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3344, pruned_loss=0.08617, over 5667003.41 frames. ], batch size: 60, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:49:22,922 INFO [optim.py:369] (1/2) 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 23:49:24,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4607, 1.5915, 1.7097, 1.3025], device='cuda:1'), covar=tensor([0.1785, 0.2468, 0.1459, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.0842, 0.0681, 0.0882, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-06 23:49:31,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:49:44,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1675, 1.1668, 3.4873, 3.0835], device='cuda:1'), covar=tensor([0.1563, 0.2762, 0.0444, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0683, 0.0602, 0.0875, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-06 23:49:54,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:50:04,992 INFO [train.py:968] (1/2) Epoch 13, batch 34700, giga_loss[loss=0.2831, simple_loss=0.3485, pruned_loss=0.1089, over 26847.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3359, pruned_loss=0.09005, over 5672439.82 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3501, pruned_loss=0.1153, over 5684130.44 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3348, pruned_loss=0.08738, over 5665177.41 frames. ], batch size: 555, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:50:18,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5285, 1.7970, 1.4178, 1.8903], device='cuda:1'), covar=tensor([0.2517, 0.2458, 0.2688, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.1338, 0.0982, 0.1185, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:50:33,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3463, 3.1597, 3.0168, 1.3114], device='cuda:1'), covar=tensor([0.0843, 0.0996, 0.0873, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1071, 0.0983, 0.0859, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-06 23:50:56,429 INFO [train.py:968] (1/2) Epoch 13, batch 34750, giga_loss[loss=0.2907, simple_loss=0.3751, pruned_loss=0.1031, over 28854.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3417, pruned_loss=0.09409, over 5657692.22 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3502, pruned_loss=0.1154, over 5676664.49 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3404, pruned_loss=0.09115, over 5658349.94 frames. ], batch size: 174, lr: 2.45e-03, grad_scale: 2.0
+2023-03-06 23:51:09,198 INFO [optim.py:369] (1/2) 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 23:51:19,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:20,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:20,951 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6513, 1.9913, 1.6226, 1.9547], device='cuda:1'), covar=tensor([0.2415, 0.2335, 0.2621, 0.2299], device='cuda:1'), in_proj_covar=tensor([0.1336, 0.0984, 0.1183, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-06 23:51:39,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582264.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:51:39,905 INFO [train.py:968] (1/2) Epoch 13, batch 34800, giga_loss[loss=0.3429, simple_loss=0.4191, pruned_loss=0.1334, over 29107.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3501, pruned_loss=0.09904, over 5676855.83 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3498, pruned_loss=0.1152, over 5685021.81 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3492, pruned_loss=0.09613, over 5669462.52 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:51:47,880 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:52,440 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:51:55,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:06,189 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=582293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:10,695 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:12,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3318, 1.8703, 1.4083, 0.5094], device='cuda:1'), covar=tensor([0.3963, 0.2122, 0.3064, 0.5002], device='cuda:1'), in_proj_covar=tensor([0.1604, 0.1533, 0.1513, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-06 23:52:12,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:52:20,748 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 34850, giga_loss[loss=0.287, simple_loss=0.3592, pruned_loss=0.1074, over 28939.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3556, pruned_loss=0.1025, over 5685189.22 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3504, pruned_loss=0.1156, over 5690244.32 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3545, pruned_loss=0.09943, over 5674448.62 frames. ], batch size: 106, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:52:36,704 INFO [optim.py:369] (1/2) 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 23:52:38,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:53:02,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7774, 1.8699, 1.3083, 1.5221], device='cuda:1'), covar=tensor([0.0782, 0.0609, 0.0960, 0.1226], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0430, 0.0499, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-06 23:53:07,967 INFO [train.py:968] (1/2) Epoch 13, batch 34900, giga_loss[loss=0.2486, simple_loss=0.3211, pruned_loss=0.08803, over 28760.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3526, pruned_loss=0.1016, over 5686549.69 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3504, pruned_loss=0.1155, over 5693862.22 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3517, pruned_loss=0.09901, over 5674620.20 frames. ], batch size: 284, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:53:42,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582407.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:53:45,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582410.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:53:48,154 INFO [train.py:968] (1/2) Epoch 13, batch 34950, giga_loss[loss=0.2124, simple_loss=0.2966, pruned_loss=0.06405, over 29005.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3457, pruned_loss=0.09873, over 5690114.61 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3506, pruned_loss=0.1156, over 5697098.91 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3449, pruned_loss=0.09624, over 5677770.30 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:54:02,436 INFO [optim.py:369] (1/2) 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 23:54:09,970 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582439.0, num_to_drop=1, layers_to_drop={0}
+2023-03-06 23:54:31,693 INFO [train.py:968] (1/2) Epoch 13, batch 35000, giga_loss[loss=0.2806, simple_loss=0.3316, pruned_loss=0.1148, over 26652.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3382, pruned_loss=0.09539, over 5673752.79 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3508, pruned_loss=0.1157, over 5680082.63 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3372, pruned_loss=0.09304, over 5678546.63 frames. ], batch size: 555, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:55:13,863 INFO [train.py:968] (1/2) Epoch 13, batch 35050, libri_loss[loss=0.3871, simple_loss=0.4261, pruned_loss=0.174, over 25710.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3318, pruned_loss=0.09289, over 5669945.72 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3514, pruned_loss=0.116, over 5679492.56 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.33, pruned_loss=0.09017, over 5674751.25 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:55:23,654 INFO [optim.py:369] (1/2) 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 23:55:42,839 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=582554.0, num_to_drop=1, layers_to_drop={1}
+2023-03-06 23:55:52,053 INFO [train.py:968] (1/2) Epoch 13, batch 35100, giga_loss[loss=0.2004, simple_loss=0.2797, pruned_loss=0.06053, over 28905.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3255, pruned_loss=0.09014, over 5679648.57 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3518, pruned_loss=0.1162, over 5684466.82 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3231, pruned_loss=0.0871, over 5678863.46 frames. ], batch size: 66, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:56:09,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6365, 1.9871, 1.9011, 1.4453], device='cuda:1'), covar=tensor([0.1502, 0.2303, 0.1347, 0.1604], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0687, 0.0889, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-06 23:56:21,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6864, 1.8981, 1.5713, 1.7901], device='cuda:1'), covar=tensor([0.2500, 0.2538, 0.2756, 0.2536], device='cuda:1'), in_proj_covar=tensor([0.1344, 0.0989, 0.1188, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:56:35,348 INFO [train.py:968] (1/2) Epoch 13, batch 35150, giga_loss[loss=0.2263, simple_loss=0.3014, pruned_loss=0.07553, over 29039.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3215, pruned_loss=0.08839, over 5686702.49 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3519, pruned_loss=0.1162, over 5685611.64 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3187, pruned_loss=0.08523, over 5685267.68 frames. ], batch size: 106, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:56:46,196 INFO [optim.py:369] (1/2) 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 23:56:51,061 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-06 23:57:17,163 INFO [train.py:968] (1/2) Epoch 13, batch 35200, giga_loss[loss=0.2535, simple_loss=0.3025, pruned_loss=0.1022, over 23763.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3186, pruned_loss=0.08727, over 5689343.61 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3519, pruned_loss=0.116, over 5688552.02 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3159, pruned_loss=0.08457, over 5685568.87 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 8.0
+2023-03-06 23:57:20,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:57:58,986 INFO [train.py:968] (1/2) Epoch 13, batch 35250, giga_loss[loss=0.2044, simple_loss=0.2786, pruned_loss=0.06506, over 28494.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3169, pruned_loss=0.08674, over 5680718.21 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3533, pruned_loss=0.1168, over 5686321.67 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3123, pruned_loss=0.08286, over 5679877.06 frames. ], batch size: 60, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:58:11,542 INFO [optim.py:369] (1/2) 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 23:58:12,882 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-06 23:58:42,378 INFO [train.py:968] (1/2) Epoch 13, batch 35300, giga_loss[loss=0.258, simple_loss=0.3269, pruned_loss=0.09451, over 27952.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.315, pruned_loss=0.08628, over 5669273.31 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3533, pruned_loss=0.1167, over 5682312.88 frames. ], giga_tot_loss[loss=0.2366, simple_loss=0.3096, pruned_loss=0.08184, over 5672060.07 frames. ], batch size: 412, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:59:09,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3933, 1.6113, 1.6584, 1.2576], device='cuda:1'), covar=tensor([0.1606, 0.2199, 0.1272, 0.1496], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0690, 0.0893, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-06 23:59:20,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=582811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:59:22,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=582814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-06 23:59:22,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4166, 1.7127, 1.3950, 1.2817], device='cuda:1'), covar=tensor([0.2139, 0.2020, 0.2169, 0.2013], device='cuda:1'), in_proj_covar=tensor([0.1340, 0.0990, 0.1187, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-06 23:59:22,391 INFO [train.py:968] (1/2) Epoch 13, batch 35350, giga_loss[loss=0.3064, simple_loss=0.3472, pruned_loss=0.1328, over 26603.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3131, pruned_loss=0.0855, over 5679993.30 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3533, pruned_loss=0.1164, over 5689099.89 frames. ], giga_tot_loss[loss=0.2346, simple_loss=0.3073, pruned_loss=0.081, over 5675761.00 frames. ], batch size: 555, lr: 2.45e-03, grad_scale: 4.0
+2023-03-06 23:59:33,081 INFO [optim.py:369] (1/2) 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 23:59:46,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=582843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:00:00,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=582861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:00:02,916 INFO [train.py:968] (1/2) Epoch 13, batch 35400, giga_loss[loss=0.2053, simple_loss=0.281, pruned_loss=0.06478, over 28904.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3116, pruned_loss=0.08481, over 5680898.15 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.354, pruned_loss=0.1167, over 5682667.33 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.305, pruned_loss=0.08, over 5682346.90 frames. ], batch size: 145, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:00:15,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-07 00:00:35,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2807, 1.3269, 1.3775, 1.0404], device='cuda:1'), covar=tensor([0.1444, 0.2593, 0.1234, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0852, 0.0688, 0.0891, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 00:00:43,269 INFO [train.py:968] (1/2) Epoch 13, batch 35450, giga_loss[loss=0.2032, simple_loss=0.2834, pruned_loss=0.06146, over 28769.00 frames. ], tot_loss[loss=0.238, simple_loss=0.309, pruned_loss=0.08353, over 5688627.30 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3547, pruned_loss=0.1169, over 5687428.35 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.302, pruned_loss=0.07872, over 5685471.19 frames. ], batch size: 284, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:00:56,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=582929.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:00:56,850 INFO [optim.py:369] (1/2) 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 00:01:24,236 INFO [train.py:968] (1/2) Epoch 13, batch 35500, giga_loss[loss=0.2005, simple_loss=0.2742, pruned_loss=0.06342, over 28959.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.307, pruned_loss=0.08245, over 5692858.43 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.355, pruned_loss=0.1168, over 5694195.97 frames. ], giga_tot_loss[loss=0.227, simple_loss=0.2993, pruned_loss=0.07733, over 5684065.06 frames. ], batch size: 106, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:02:10,001 INFO [train.py:968] (1/2) Epoch 13, batch 35550, giga_loss[loss=0.2458, simple_loss=0.3258, pruned_loss=0.08288, over 28668.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3042, pruned_loss=0.0814, over 5675248.36 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3552, pruned_loss=0.1169, over 5686968.01 frames. ], giga_tot_loss[loss=0.2246, simple_loss=0.2966, pruned_loss=0.0763, over 5674482.18 frames. ], batch size: 242, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:02:22,382 INFO [optim.py:369] (1/2) 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 00:02:37,055 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 00:02:54,179 INFO [train.py:968] (1/2) Epoch 13, batch 35600, giga_loss[loss=0.2733, simple_loss=0.3525, pruned_loss=0.09707, over 28886.00 frames. ], tot_loss[loss=0.238, simple_loss=0.3084, pruned_loss=0.08383, over 5674315.92 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3554, pruned_loss=0.117, over 5687601.98 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.3012, pruned_loss=0.07914, over 5672753.93 frames. ], batch size: 145, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:03:00,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583072.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:03:03,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583075.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:03:30,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=583104.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:03:43,494 INFO [train.py:968] (1/2) Epoch 13, batch 35650, giga_loss[loss=0.3012, simple_loss=0.3759, pruned_loss=0.1132, over 29081.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3224, pruned_loss=0.09145, over 5686060.18 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.356, pruned_loss=0.1174, over 5692129.45 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3154, pruned_loss=0.08681, over 5680755.17 frames. ], batch size: 128, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:03:58,526 INFO [optim.py:369] (1/2) 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 00:04:20,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:04:28,541 INFO [train.py:968] (1/2) Epoch 13, batch 35700, giga_loss[loss=0.3889, simple_loss=0.4277, pruned_loss=0.175, over 27639.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3349, pruned_loss=0.09833, over 5680178.87 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3557, pruned_loss=0.1171, over 5695452.92 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3293, pruned_loss=0.09463, over 5672884.38 frames. ], batch size: 472, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:04:49,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-07 00:05:13,117 INFO [train.py:968] (1/2) Epoch 13, batch 35750, giga_loss[loss=0.2756, simple_loss=0.3534, pruned_loss=0.09886, over 28805.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3427, pruned_loss=0.1012, over 5679979.39 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3559, pruned_loss=0.1171, over 5689762.67 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3378, pruned_loss=0.09798, over 5678443.72 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:05:27,358 INFO [optim.py:369] (1/2) 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 00:05:32,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=583236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:05:32,990 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:05:43,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1992, 1.1999, 3.7972, 2.9595], device='cuda:1'), covar=tensor([0.1638, 0.2742, 0.0416, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0594, 0.0868, 0.0782], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 00:05:57,568 INFO [train.py:968] (1/2) Epoch 13, batch 35800, giga_loss[loss=0.2384, simple_loss=0.3285, pruned_loss=0.07412, over 29050.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3458, pruned_loss=0.101, over 5685692.93 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.356, pruned_loss=0.1171, over 5691076.85 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3418, pruned_loss=0.09838, over 5683269.16 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:06:42,330 INFO [train.py:968] (1/2) Epoch 13, batch 35850, giga_loss[loss=0.2433, simple_loss=0.3219, pruned_loss=0.08238, over 28456.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3466, pruned_loss=0.1004, over 5668628.74 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3562, pruned_loss=0.1171, over 5689201.44 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3429, pruned_loss=0.09771, over 5667595.95 frames. ], batch size: 65, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:06:58,467 INFO [optim.py:369] (1/2) 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 00:07:24,567 INFO [train.py:968] (1/2) Epoch 13, batch 35900, giga_loss[loss=0.2982, simple_loss=0.3653, pruned_loss=0.1156, over 29041.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3494, pruned_loss=0.1021, over 5674619.11 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.357, pruned_loss=0.1177, over 5692242.67 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3455, pruned_loss=0.09907, over 5670724.87 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:07:36,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:07:39,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:07:54,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3671, 3.2041, 1.5735, 1.4434], device='cuda:1'), covar=tensor([0.1001, 0.0275, 0.0851, 0.1408], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0513, 0.0349, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 00:08:06,125 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 35950, giga_loss[loss=0.2898, simple_loss=0.3623, pruned_loss=0.1087, over 28331.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3508, pruned_loss=0.1034, over 5690397.90 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3568, pruned_loss=0.1176, over 5697690.71 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3478, pruned_loss=0.1007, over 5682050.91 frames. ], batch size: 65, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:08:09,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1941, 5.0104, 4.7279, 2.3638], device='cuda:1'), covar=tensor([0.0423, 0.0536, 0.0595, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.1064, 0.0992, 0.0863, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 00:08:23,940 INFO [optim.py:369] (1/2) 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 00:08:24,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3257, 1.4416, 3.3712, 3.0638], device='cuda:1'), covar=tensor([0.1655, 0.2759, 0.0779, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0675, 0.0593, 0.0868, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 00:08:27,692 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-07 00:08:51,862 INFO [train.py:968] (1/2) Epoch 13, batch 36000, giga_loss[loss=0.2663, simple_loss=0.3491, pruned_loss=0.09171, over 28988.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3541, pruned_loss=0.1056, over 5686490.42 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3571, pruned_loss=0.1176, over 5701625.72 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3514, pruned_loss=0.1033, over 5676086.40 frames. ], batch size: 164, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:08:51,862 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 00:09:00,167 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 00:09:38,296 INFO [train.py:968] (1/2) Epoch 13, batch 36050, giga_loss[loss=0.2948, simple_loss=0.3675, pruned_loss=0.111, over 28706.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3571, pruned_loss=0.106, over 5702835.45 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3578, pruned_loss=0.1179, over 5704814.13 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3542, pruned_loss=0.1035, over 5691320.68 frames. ], batch size: 92, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:09:52,265 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:09:56,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4377, 4.2164, 3.9729, 2.1191], device='cuda:1'), covar=tensor([0.0500, 0.0648, 0.0660, 0.1918], device='cuda:1'), in_proj_covar=tensor([0.1065, 0.0991, 0.0864, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 00:10:17,744 INFO [train.py:968] (1/2) Epoch 13, batch 36100, giga_loss[loss=0.3184, simple_loss=0.387, pruned_loss=0.1249, over 28280.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.36, pruned_loss=0.1073, over 5694026.59 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3591, pruned_loss=0.1187, over 5702081.85 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3565, pruned_loss=0.104, over 5687555.64 frames. ], batch size: 368, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:10:40,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3174, 2.9051, 1.3797, 1.4598], device='cuda:1'), covar=tensor([0.0983, 0.0287, 0.0898, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0512, 0.0348, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 00:10:56,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=583612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:10:58,282 INFO [train.py:968] (1/2) Epoch 13, batch 36150, giga_loss[loss=0.2703, simple_loss=0.3534, pruned_loss=0.09354, over 29050.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3626, pruned_loss=0.1084, over 5690971.66 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3601, pruned_loss=0.1194, over 5697676.81 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3589, pruned_loss=0.1048, over 5690633.57 frames. ], batch size: 128, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:11:10,144 INFO [optim.py:369] (1/2) 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 00:11:34,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:11:39,399 INFO [train.py:968] (1/2) Epoch 13, batch 36200, giga_loss[loss=0.2361, simple_loss=0.3301, pruned_loss=0.07103, over 29040.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.362, pruned_loss=0.1068, over 5690786.75 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3598, pruned_loss=0.1192, over 5698799.43 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3593, pruned_loss=0.1041, over 5689570.49 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:11:46,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:11:48,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:12:11,205 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 36250, giga_loss[loss=0.2883, simple_loss=0.3718, pruned_loss=0.1024, over 28819.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3596, pruned_loss=0.104, over 5703713.51 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3593, pruned_loss=0.1187, over 5703740.26 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.358, pruned_loss=0.102, over 5698188.30 frames. ], batch size: 174, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:12:32,445 INFO [optim.py:369] (1/2) 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 00:12:51,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=583755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:12:53,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=583758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:12:58,385 INFO [train.py:968] (1/2) Epoch 13, batch 36300, giga_loss[loss=0.24, simple_loss=0.3303, pruned_loss=0.07485, over 28277.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3568, pruned_loss=0.1019, over 5693525.27 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3593, pruned_loss=0.1185, over 5697800.88 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3555, pruned_loss=0.1001, over 5695092.80 frames. ], batch size: 77, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:13:16,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=583787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:13:26,284 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:13:31,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4119, 2.5326, 1.7699, 2.0007], device='cuda:1'), covar=tensor([0.0829, 0.0645, 0.1056, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0433, 0.0505, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 00:13:37,521 INFO [train.py:968] (1/2) Epoch 13, batch 36350, giga_loss[loss=0.2705, simple_loss=0.3521, pruned_loss=0.09452, over 28378.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3575, pruned_loss=0.1029, over 5688703.72 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3598, pruned_loss=0.1187, over 5699323.33 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3559, pruned_loss=0.1009, over 5688422.43 frames. ], batch size: 65, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:13:49,002 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=583827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:13:50,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 00:13:52,344 INFO [optim.py:369] (1/2) 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 00:14:18,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5372, 1.6853, 1.3649, 1.7301], device='cuda:1'), covar=tensor([0.2472, 0.2490, 0.2680, 0.2143], device='cuda:1'), in_proj_covar=tensor([0.1339, 0.0987, 0.1181, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0010, 0.0008], device='cuda:1')
+2023-03-07 00:14:23,339 INFO [train.py:968] (1/2) Epoch 13, batch 36400, giga_loss[loss=0.3307, simple_loss=0.3855, pruned_loss=0.138, over 28799.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3597, pruned_loss=0.1071, over 5680893.11 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3593, pruned_loss=0.1184, over 5693892.77 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3588, pruned_loss=0.1053, over 5685057.16 frames. ], batch size: 112, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:15:06,776 INFO [train.py:968] (1/2) Epoch 13, batch 36450, libri_loss[loss=0.3551, simple_loss=0.3969, pruned_loss=0.1567, over 18782.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.3612, pruned_loss=0.1099, over 5675135.60 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3595, pruned_loss=0.1186, over 5687181.97 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3603, pruned_loss=0.1082, over 5685282.02 frames. ], batch size: 187, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:15:21,136 INFO [optim.py:369] (1/2) 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 00:15:50,700 INFO [train.py:968] (1/2) Epoch 13, batch 36500, giga_loss[loss=0.2957, simple_loss=0.3566, pruned_loss=0.1175, over 28621.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3608, pruned_loss=0.1108, over 5677216.01 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3598, pruned_loss=0.1187, over 5687293.42 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3598, pruned_loss=0.1091, over 5685083.23 frames. ], batch size: 85, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:16:30,238 INFO [train.py:968] (1/2) Epoch 13, batch 36550, giga_loss[loss=0.2742, simple_loss=0.3412, pruned_loss=0.1036, over 28868.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3598, pruned_loss=0.1107, over 5683648.21 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3611, pruned_loss=0.1195, over 5681400.30 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3579, pruned_loss=0.1085, over 5694631.54 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:16:46,297 INFO [optim.py:369] (1/2) 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 00:16:47,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:17:11,814 INFO [train.py:968] (1/2) Epoch 13, batch 36600, libri_loss[loss=0.3515, simple_loss=0.4058, pruned_loss=0.1486, over 29654.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3585, pruned_loss=0.11, over 5693077.49 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3616, pruned_loss=0.1198, over 5687418.71 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3565, pruned_loss=0.1077, over 5696495.60 frames. ], batch size: 91, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:17:55,693 INFO [train.py:968] (1/2) Epoch 13, batch 36650, giga_loss[loss=0.2592, simple_loss=0.3436, pruned_loss=0.08737, over 28794.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3576, pruned_loss=0.1086, over 5693355.43 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3616, pruned_loss=0.1197, over 5692094.34 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3558, pruned_loss=0.1066, over 5692057.92 frames. ], batch size: 243, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:18:13,636 INFO [optim.py:369] (1/2) 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 00:18:38,582 INFO [train.py:968] (1/2) Epoch 13, batch 36700, giga_loss[loss=0.2614, simple_loss=0.3168, pruned_loss=0.103, over 23398.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3551, pruned_loss=0.1068, over 5680438.20 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3619, pruned_loss=0.1195, over 5681099.63 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3533, pruned_loss=0.1048, over 5688607.92 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:18:48,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:18:49,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:18:51,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:18:56,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 00:19:09,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:19:17,365 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:19:21,877 INFO [train.py:968] (1/2) Epoch 13, batch 36750, giga_loss[loss=0.2448, simple_loss=0.3198, pruned_loss=0.08489, over 28716.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3486, pruned_loss=0.1025, over 5691938.68 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3618, pruned_loss=0.1192, over 5687126.97 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3469, pruned_loss=0.1009, over 5693043.40 frames. ], batch size: 284, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:19:38,013 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 36800, giga_loss[loss=0.2371, simple_loss=0.3142, pruned_loss=0.08005, over 28803.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3432, pruned_loss=0.09948, over 5687011.97 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3615, pruned_loss=0.1188, over 5686455.52 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3415, pruned_loss=0.09775, over 5688229.31 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:20:20,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2971, 1.6614, 1.4465, 1.5389], device='cuda:1'), covar=tensor([0.0805, 0.0299, 0.0315, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 00:20:58,110 INFO [train.py:968] (1/2) Epoch 13, batch 36850, giga_loss[loss=0.276, simple_loss=0.3543, pruned_loss=0.09885, over 28968.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3387, pruned_loss=0.09755, over 5680108.07 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3617, pruned_loss=0.1187, over 5693323.15 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3366, pruned_loss=0.09564, over 5674646.27 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:21:03,966 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:06,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:09,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6247, 4.4843, 4.2267, 2.0598], device='cuda:1'), covar=tensor([0.0463, 0.0555, 0.0557, 0.1945], device='cuda:1'), in_proj_covar=tensor([0.1066, 0.0993, 0.0864, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 00:21:15,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4747, 1.5208, 1.4691, 1.3113], device='cuda:1'), covar=tensor([0.2177, 0.1999, 0.1524, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.1749, 0.1646, 0.1614, 0.1712], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 00:21:16,019 INFO [optim.py:369] (1/2) 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 00:21:27,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:29,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:32,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:40,556 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:43,932 INFO [train.py:968] (1/2) Epoch 13, batch 36900, giga_loss[loss=0.25, simple_loss=0.3279, pruned_loss=0.08602, over 27742.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3394, pruned_loss=0.09751, over 5682859.00 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.362, pruned_loss=0.1188, over 5699833.67 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3368, pruned_loss=0.09525, over 5672110.68 frames. ], batch size: 472, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:21:49,917 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:21:53,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:22:13,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-07 00:22:25,325 INFO [train.py:968] (1/2) Epoch 13, batch 36950, libri_loss[loss=0.2628, simple_loss=0.3347, pruned_loss=0.09547, over 29590.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3394, pruned_loss=0.09708, over 5696604.94 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3618, pruned_loss=0.1186, over 5703768.10 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.337, pruned_loss=0.09508, over 5684211.55 frames. ], batch size: 74, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:22:30,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 00:22:40,430 INFO [optim.py:369] (1/2) 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 00:22:44,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6287, 2.0107, 1.6155, 1.9446], device='cuda:1'), covar=tensor([0.2401, 0.2261, 0.2537, 0.1955], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.0989, 0.1188, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 00:23:08,146 INFO [train.py:968] (1/2) Epoch 13, batch 37000, giga_loss[loss=0.2461, simple_loss=0.3182, pruned_loss=0.087, over 28943.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3384, pruned_loss=0.09679, over 5696008.42 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.362, pruned_loss=0.1186, over 5703071.44 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3361, pruned_loss=0.09493, over 5686760.63 frames. ], batch size: 106, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:23:30,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1190, 1.0253, 0.9886, 1.3044], device='cuda:1'), covar=tensor([0.0810, 0.0372, 0.0349, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 00:23:48,062 INFO [train.py:968] (1/2) Epoch 13, batch 37050, giga_loss[loss=0.2414, simple_loss=0.3197, pruned_loss=0.08156, over 29086.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3367, pruned_loss=0.09618, over 5693526.77 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3624, pruned_loss=0.1187, over 5698121.85 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3339, pruned_loss=0.09411, over 5690085.73 frames. ], batch size: 128, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:24:00,260 INFO [optim.py:369] (1/2) 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 00:24:28,288 INFO [train.py:968] (1/2) Epoch 13, batch 37100, giga_loss[loss=0.2348, simple_loss=0.3113, pruned_loss=0.07915, over 28358.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3353, pruned_loss=0.09573, over 5702135.91 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3632, pruned_loss=0.119, over 5698682.13 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.332, pruned_loss=0.09336, over 5698743.83 frames. ], batch size: 77, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:24:37,141 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-07 00:24:53,698 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:25:07,670 INFO [train.py:968] (1/2) Epoch 13, batch 37150, giga_loss[loss=0.2794, simple_loss=0.347, pruned_loss=0.1059, over 28961.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3325, pruned_loss=0.09416, over 5710522.01 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3634, pruned_loss=0.119, over 5701103.42 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3293, pruned_loss=0.09198, over 5705759.25 frames. ], batch size: 145, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:25:14,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-07 00:25:23,512 INFO [optim.py:369] (1/2) 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 00:25:26,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-07 00:25:48,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7250, 1.9041, 1.7259, 1.7846], device='cuda:1'), covar=tensor([0.1667, 0.2059, 0.2335, 0.1952], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0737, 0.0683, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 00:25:50,626 INFO [train.py:968] (1/2) Epoch 13, batch 37200, giga_loss[loss=0.2328, simple_loss=0.3071, pruned_loss=0.0792, over 29032.00 frames. ], tot_loss[loss=0.258, simple_loss=0.33, pruned_loss=0.09305, over 5707978.49 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3636, pruned_loss=0.1191, over 5702125.90 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3271, pruned_loss=0.09117, over 5703483.90 frames. ], batch size: 106, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:26:01,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3453, 1.8247, 1.3271, 0.6843], device='cuda:1'), covar=tensor([0.3397, 0.1695, 0.2606, 0.4717], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1507, 0.1498, 0.1312], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 00:26:30,847 INFO [train.py:968] (1/2) Epoch 13, batch 37250, giga_loss[loss=0.2711, simple_loss=0.3388, pruned_loss=0.1017, over 28859.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3276, pruned_loss=0.09178, over 5708493.03 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3642, pruned_loss=0.1193, over 5702248.77 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3243, pruned_loss=0.08967, over 5704676.06 frames. ], batch size: 186, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:26:37,811 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:26:46,222 INFO [optim.py:369] (1/2) 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 00:26:49,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:26:56,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584747.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:26:59,085 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 37300, giga_loss[loss=0.2245, simple_loss=0.3037, pruned_loss=0.07265, over 28906.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3265, pruned_loss=0.0913, over 5715996.33 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3658, pruned_loss=0.1204, over 5704040.43 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3214, pruned_loss=0.08792, over 5711761.88 frames. ], batch size: 145, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:27:49,113 INFO [train.py:968] (1/2) Epoch 13, batch 37350, giga_loss[loss=0.2522, simple_loss=0.3204, pruned_loss=0.09199, over 28762.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3242, pruned_loss=0.09003, over 5723531.38 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3657, pruned_loss=0.1203, over 5707004.06 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3197, pruned_loss=0.08706, over 5717915.12 frames. ], batch size: 119, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:27:54,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:05,084 INFO [optim.py:369] (1/2) 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 00:28:27,885 INFO [train.py:968] (1/2) Epoch 13, batch 37400, libri_loss[loss=0.4012, simple_loss=0.4456, pruned_loss=0.1784, over 29147.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3237, pruned_loss=0.0898, over 5718164.53 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3668, pruned_loss=0.1209, over 5697731.45 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3182, pruned_loss=0.08619, over 5722420.51 frames. ], batch size: 101, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:28:40,601 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:43,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:48,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584890.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:28:50,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584893.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:28:51,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=584895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:28:53,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=584898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:29:05,686 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:29:07,270 INFO [train.py:968] (1/2) Epoch 13, batch 37450, giga_loss[loss=0.2793, simple_loss=0.3416, pruned_loss=0.1084, over 28895.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3261, pruned_loss=0.09123, over 5719953.03 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5704468.32 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3197, pruned_loss=0.08746, over 5718122.95 frames. ], batch size: 106, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:29:12,605 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=584922.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:29:16,711 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:29:42,520 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=584955.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:29:49,737 INFO [train.py:968] (1/2) Epoch 13, batch 37500, libri_loss[loss=0.3183, simple_loss=0.3874, pruned_loss=0.1246, over 26128.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3313, pruned_loss=0.09449, over 5713567.78 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3676, pruned_loss=0.1207, over 5703859.69 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3257, pruned_loss=0.09126, over 5713240.63 frames. ], batch size: 136, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:29:55,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=584971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:30:19,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 00:30:33,023 INFO [train.py:968] (1/2) Epoch 13, batch 37550, giga_loss[loss=0.3162, simple_loss=0.3768, pruned_loss=0.1278, over 29068.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3389, pruned_loss=0.0993, over 5696669.79 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.369, pruned_loss=0.1215, over 5693382.00 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3318, pruned_loss=0.095, over 5707230.65 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:30:43,973 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:30:53,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 00:31:19,968 INFO [train.py:968] (1/2) Epoch 13, batch 37600, libri_loss[loss=0.3303, simple_loss=0.3953, pruned_loss=0.1326, over 29224.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3483, pruned_loss=0.1059, over 5692988.51 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.122, over 5697533.87 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3412, pruned_loss=0.1014, over 5697749.30 frames. ], batch size: 94, lr: 2.45e-03, grad_scale: 8.0
+2023-03-07 00:31:31,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3984, 1.5202, 1.2683, 1.3848], device='cuda:1'), covar=tensor([0.1532, 0.1445, 0.1559, 0.1449], device='cuda:1'), in_proj_covar=tensor([0.1756, 0.1660, 0.1639, 0.1732], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 00:31:52,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:10,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:11,268 INFO [train.py:968] (1/2) Epoch 13, batch 37650, libri_loss[loss=0.2854, simple_loss=0.359, pruned_loss=0.1059, over 29663.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3517, pruned_loss=0.107, over 5676520.08 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3698, pruned_loss=0.122, over 5699701.31 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3456, pruned_loss=0.1032, over 5677911.70 frames. ], batch size: 88, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:32:12,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-07 00:32:12,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:28,745 INFO [optim.py:369] (1/2) 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 00:32:37,106 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:32:54,756 INFO [train.py:968] (1/2) Epoch 13, batch 37700, giga_loss[loss=0.3937, simple_loss=0.4258, pruned_loss=0.1808, over 23647.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.356, pruned_loss=0.1086, over 5667881.24 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1218, over 5691028.58 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3508, pruned_loss=0.1053, over 5676783.85 frames. ], batch size: 705, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:33:22,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:33:39,120 INFO [train.py:968] (1/2) Epoch 13, batch 37750, giga_loss[loss=0.3771, simple_loss=0.4333, pruned_loss=0.1605, over 28644.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3608, pruned_loss=0.1114, over 5660446.20 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3695, pruned_loss=0.1217, over 5686731.87 frames. ], giga_tot_loss[loss=0.2869, simple_loss=0.3566, pruned_loss=0.1086, over 5670303.60 frames. ], batch size: 242, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:33:52,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2873, 1.4841, 1.5210, 1.3175], device='cuda:1'), covar=tensor([0.1578, 0.1595, 0.2049, 0.1710], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0731, 0.0678, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 00:33:56,701 INFO [optim.py:369] (1/2) 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 00:34:02,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:34:04,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5761, 2.0915, 1.6252, 1.8391], device='cuda:1'), covar=tensor([0.0786, 0.0268, 0.0295, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 00:34:05,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:34:06,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5158, 1.7446, 1.7728, 1.3652], device='cuda:1'), covar=tensor([0.1737, 0.2372, 0.1441, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.0850, 0.0689, 0.0890, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 00:34:22,006 INFO [train.py:968] (1/2) Epoch 13, batch 37800, giga_loss[loss=0.2796, simple_loss=0.348, pruned_loss=0.1056, over 28929.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3619, pruned_loss=0.1116, over 5662674.06 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3698, pruned_loss=0.1219, over 5688884.34 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3582, pruned_loss=0.1091, over 5668161.82 frames. ], batch size: 145, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:34:29,060 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:34:38,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4543, 1.6448, 1.5121, 1.3976], device='cuda:1'), covar=tensor([0.2249, 0.1814, 0.1401, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.1765, 0.1666, 0.1648, 0.1738], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 00:35:04,025 INFO [train.py:968] (1/2) Epoch 13, batch 37850, giga_loss[loss=0.2376, simple_loss=0.3277, pruned_loss=0.07374, over 29056.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3563, pruned_loss=0.1071, over 5676099.78 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3699, pruned_loss=0.1219, over 5691300.93 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3532, pruned_loss=0.1049, over 5678154.59 frames. ], batch size: 155, lr: 2.45e-03, grad_scale: 4.0
+2023-03-07 00:35:17,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6360, 4.4476, 4.2059, 1.7476], device='cuda:1'), covar=tensor([0.0597, 0.0756, 0.0778, 0.2377], device='cuda:1'), in_proj_covar=tensor([0.1073, 0.0997, 0.0864, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 00:35:17,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585330.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:35:24,004 INFO [optim.py:369] (1/2) 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 00:35:27,130 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:35:28,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:35:45,926 INFO [train.py:968] (1/2) Epoch 13, batch 37900, giga_loss[loss=0.2405, simple_loss=0.3234, pruned_loss=0.07878, over 28583.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3555, pruned_loss=0.1057, over 5674301.21 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5683017.90 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3522, pruned_loss=0.1033, over 5683539.67 frames. ], batch size: 336, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:35:51,392 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:36:03,546 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 00:36:16,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=585400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:36:27,549 INFO [train.py:968] (1/2) Epoch 13, batch 37950, giga_loss[loss=0.306, simple_loss=0.3724, pruned_loss=0.1199, over 28612.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3549, pruned_loss=0.1049, over 5676158.22 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5686552.41 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3518, pruned_loss=0.1026, over 5680000.97 frames. ], batch size: 85, lr: 2.45e-03, grad_scale: 2.0
+2023-03-07 00:36:46,818 INFO [optim.py:369] (1/2) 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 00:37:13,708 INFO [train.py:968] (1/2) Epoch 13, batch 38000, giga_loss[loss=0.2825, simple_loss=0.3595, pruned_loss=0.1028, over 28996.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3568, pruned_loss=0.1057, over 5680109.21 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5686552.41 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3544, pruned_loss=0.104, over 5683100.07 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:37:20,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585473.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:37:22,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585476.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 00:37:23,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-07 00:37:47,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585505.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 00:37:55,417 INFO [train.py:968] (1/2) Epoch 13, batch 38050, giga_loss[loss=0.2688, simple_loss=0.3456, pruned_loss=0.09594, over 28558.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3601, pruned_loss=0.1084, over 5676683.96 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5681624.27 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3573, pruned_loss=0.1063, over 5683923.04 frames. ], batch size: 71, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:38:15,919 INFO [optim.py:369] (1/2) 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 00:38:21,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=585543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:38:24,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=585546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:38:40,135 INFO [train.py:968] (1/2) Epoch 13, batch 38100, giga_loss[loss=0.3058, simple_loss=0.3714, pruned_loss=0.1202, over 28547.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3604, pruned_loss=0.109, over 5684400.90 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3705, pruned_loss=0.1221, over 5682959.72 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3583, pruned_loss=0.1072, over 5688948.68 frames. ], batch size: 85, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:38:49,469 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=585575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:39:23,439 INFO [train.py:968] (1/2) Epoch 13, batch 38150, giga_loss[loss=0.2853, simple_loss=0.3548, pruned_loss=0.1079, over 28718.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3612, pruned_loss=0.11, over 5681377.75 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3708, pruned_loss=0.1223, over 5683741.78 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3591, pruned_loss=0.1083, over 5684345.76 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:39:43,589 INFO [optim.py:369] (1/2) 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 00:40:05,486 INFO [train.py:968] (1/2) Epoch 13, batch 38200, giga_loss[loss=0.2845, simple_loss=0.3544, pruned_loss=0.1073, over 28597.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3618, pruned_loss=0.1108, over 5693169.50 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3707, pruned_loss=0.1223, over 5686867.62 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3601, pruned_loss=0.1093, over 5692722.09 frames. ], batch size: 307, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:40:46,597 INFO [train.py:968] (1/2) Epoch 13, batch 38250, giga_loss[loss=0.2848, simple_loss=0.3598, pruned_loss=0.1049, over 28811.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3612, pruned_loss=0.1096, over 5699988.97 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3709, pruned_loss=0.1224, over 5691338.86 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3595, pruned_loss=0.1081, over 5695769.94 frames. ], batch size: 199, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:41:01,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5673, 1.3082, 4.8179, 3.4389], device='cuda:1'), covar=tensor([0.1690, 0.2801, 0.0347, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0597, 0.0871, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 00:41:03,452 INFO [optim.py:369] (1/2) 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 00:41:11,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4010, 3.3164, 1.4885, 1.5191], device='cuda:1'), covar=tensor([0.0996, 0.0236, 0.0897, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0511, 0.0346, 0.0396], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 00:41:26,137 INFO [train.py:968] (1/2) Epoch 13, batch 38300, giga_loss[loss=0.2563, simple_loss=0.3429, pruned_loss=0.08489, over 29040.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3592, pruned_loss=0.1069, over 5707159.18 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.122, over 5694621.84 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3581, pruned_loss=0.1057, over 5701201.89 frames. ], batch size: 155, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:41:43,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-07 00:42:05,586 INFO [train.py:968] (1/2) Epoch 13, batch 38350, giga_loss[loss=0.2644, simple_loss=0.3427, pruned_loss=0.09309, over 29004.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3598, pruned_loss=0.1068, over 5697192.01 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3706, pruned_loss=0.1221, over 5686021.02 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3586, pruned_loss=0.1054, over 5700130.31 frames. ], batch size: 155, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 00:42:05,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8210, 1.8575, 1.3201, 1.4611], device='cuda:1'), covar=tensor([0.0803, 0.0640, 0.1031, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0429, 0.0500, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 00:42:24,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9243, 3.7472, 3.5155, 1.7835], device='cuda:1'), covar=tensor([0.0609, 0.0734, 0.0737, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.1072, 0.0996, 0.0866, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 00:42:24,478 INFO [optim.py:369] (1/2) 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 00:42:30,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-07 00:42:46,204 INFO [train.py:968] (1/2) Epoch 13, batch 38400, giga_loss[loss=0.2924, simple_loss=0.3599, pruned_loss=0.1124, over 28614.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3583, pruned_loss=0.1058, over 5694502.50 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3709, pruned_loss=0.1224, over 5683570.35 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3566, pruned_loss=0.104, over 5700414.91 frames. ], batch size: 92, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:43:26,217 INFO [train.py:968] (1/2) Epoch 13, batch 38450, giga_loss[loss=0.2484, simple_loss=0.3307, pruned_loss=0.08302, over 28407.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3565, pruned_loss=0.1049, over 5706556.75 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3706, pruned_loss=0.1221, over 5690515.63 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.355, pruned_loss=0.1033, over 5705528.58 frames. ], batch size: 60, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:43:44,957 INFO [optim.py:369] (1/2) 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 00:44:06,286 INFO [train.py:968] (1/2) Epoch 13, batch 38500, libri_loss[loss=0.3525, simple_loss=0.4081, pruned_loss=0.1485, over 25789.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3543, pruned_loss=0.1041, over 5704699.03 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3707, pruned_loss=0.1223, over 5683082.97 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3525, pruned_loss=0.102, over 5711305.53 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:44:11,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0011, 2.1816, 2.3185, 1.8042], device='cuda:1'), covar=tensor([0.1697, 0.2203, 0.1281, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0850, 0.0691, 0.0890, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 00:44:28,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3931, 1.2836, 4.6551, 3.4266], device='cuda:1'), covar=tensor([0.1757, 0.2806, 0.0366, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0676, 0.0595, 0.0869, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 00:44:47,209 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:44:48,275 INFO [train.py:968] (1/2) Epoch 13, batch 38550, giga_loss[loss=0.3574, simple_loss=0.3975, pruned_loss=0.1587, over 26562.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3541, pruned_loss=0.1046, over 5704671.40 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3708, pruned_loss=0.1223, over 5686213.91 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3523, pruned_loss=0.1027, over 5707434.23 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:45:02,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:45:06,998 INFO [optim.py:369] (1/2) 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 00:45:08,327 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.71 vs. limit=5.0
+2023-03-07 00:45:20,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5469, 4.3542, 4.1205, 2.0246], device='cuda:1'), covar=tensor([0.0508, 0.0681, 0.0746, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.0998, 0.0873, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 00:45:24,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 00:45:26,487 INFO [train.py:968] (1/2) Epoch 13, batch 38600, giga_loss[loss=0.2827, simple_loss=0.3544, pruned_loss=0.1055, over 28855.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.356, pruned_loss=0.1063, over 5696290.35 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1227, over 5672169.43 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3538, pruned_loss=0.104, over 5711977.57 frames. ], batch size: 112, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:45:47,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:45:50,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5221, 1.6615, 1.4020, 1.7113], device='cuda:1'), covar=tensor([0.2471, 0.2429, 0.2568, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.1346, 0.0994, 0.1189, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 00:46:04,946 INFO [train.py:968] (1/2) Epoch 13, batch 38650, giga_loss[loss=0.2902, simple_loss=0.3583, pruned_loss=0.1111, over 28623.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3557, pruned_loss=0.1053, over 5703113.45 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3716, pruned_loss=0.1228, over 5674249.17 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3535, pruned_loss=0.1032, over 5713883.04 frames. ], batch size: 85, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:46:22,837 INFO [optim.py:369] (1/2) 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 00:46:38,155 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:46:43,129 INFO [train.py:968] (1/2) Epoch 13, batch 38700, giga_loss[loss=0.2977, simple_loss=0.3575, pruned_loss=0.1189, over 26473.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3555, pruned_loss=0.1048, over 5699316.69 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3721, pruned_loss=0.1231, over 5675268.41 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3531, pruned_loss=0.1026, over 5707426.19 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:47:02,909 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-07 00:47:21,223 INFO [train.py:968] (1/2) Epoch 13, batch 38750, giga_loss[loss=0.2564, simple_loss=0.3354, pruned_loss=0.08864, over 28500.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3543, pruned_loss=0.1038, over 5700437.12 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5665770.10 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3522, pruned_loss=0.1019, over 5715039.79 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:47:42,432 INFO [optim.py:369] (1/2) 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 00:48:06,180 INFO [train.py:968] (1/2) Epoch 13, batch 38800, giga_loss[loss=0.2306, simple_loss=0.308, pruned_loss=0.07655, over 28881.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3518, pruned_loss=0.1026, over 5699421.61 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3721, pruned_loss=0.1231, over 5666924.25 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.35, pruned_loss=0.1009, over 5710026.33 frames. ], batch size: 112, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:48:47,100 INFO [train.py:968] (1/2) Epoch 13, batch 38850, giga_loss[loss=0.2586, simple_loss=0.3344, pruned_loss=0.09145, over 28444.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3485, pruned_loss=0.1009, over 5696160.36 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5670752.92 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3468, pruned_loss=0.09944, over 5701704.71 frames. ], batch size: 65, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:49:05,513 INFO [optim.py:369] (1/2) 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 00:49:24,765 INFO [train.py:968] (1/2) Epoch 13, batch 38900, libri_loss[loss=0.2698, simple_loss=0.3327, pruned_loss=0.1034, over 29666.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3458, pruned_loss=0.0998, over 5707313.06 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3719, pruned_loss=0.1229, over 5674400.02 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3442, pruned_loss=0.09831, over 5708718.22 frames. ], batch size: 73, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:49:43,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:49:59,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:50:04,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-07 00:50:07,132 INFO [train.py:968] (1/2) Epoch 13, batch 38950, giga_loss[loss=0.2762, simple_loss=0.3475, pruned_loss=0.1025, over 28927.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3462, pruned_loss=0.1002, over 5684973.82 frames. ], libri_tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1234, over 5657630.33 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.344, pruned_loss=0.09822, over 5701786.98 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:50:25,810 INFO [optim.py:369] (1/2) 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 00:50:36,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3736, 1.4331, 1.2165, 1.5714], device='cuda:1'), covar=tensor([0.0724, 0.0317, 0.0335, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 00:50:46,454 INFO [train.py:968] (1/2) Epoch 13, batch 39000, giga_loss[loss=0.2459, simple_loss=0.3214, pruned_loss=0.0852, over 28621.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3443, pruned_loss=0.09965, over 5684267.00 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3727, pruned_loss=0.1235, over 5657006.26 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3419, pruned_loss=0.09753, over 5698832.71 frames. ], batch size: 71, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:50:46,455 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 00:50:54,792 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 00:50:55,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:34,701 INFO [train.py:968] (1/2) Epoch 13, batch 39050, giga_loss[loss=0.2527, simple_loss=0.325, pruned_loss=0.09019, over 28965.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3419, pruned_loss=0.09858, over 5696467.18 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3722, pruned_loss=0.1232, over 5661642.86 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3399, pruned_loss=0.09673, over 5704678.86 frames. ], batch size: 164, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:51:47,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=586531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:49,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=586534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:49,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:51:52,893 INFO [optim.py:369] (1/2) 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 00:52:00,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=586549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:02,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:10,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:12,183 INFO [train.py:968] (1/2) Epoch 13, batch 39100, giga_loss[loss=0.2492, simple_loss=0.3261, pruned_loss=0.08609, over 29018.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.341, pruned_loss=0.09892, over 5690960.66 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3726, pruned_loss=0.1237, over 5646557.31 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3386, pruned_loss=0.09671, over 5712337.42 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:52:27,167 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:49,667 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=586609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:53,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:52:54,711 INFO [train.py:968] (1/2) Epoch 13, batch 39150, giga_loss[loss=0.2511, simple_loss=0.3246, pruned_loss=0.0888, over 28906.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3407, pruned_loss=0.09916, over 5693145.75 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5654800.60 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3376, pruned_loss=0.09644, over 5703942.45 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:53:14,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8029, 2.3673, 2.5402, 2.1485], device='cuda:1'), covar=tensor([0.1225, 0.1902, 0.1433, 0.1595], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0736, 0.0683, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 00:53:15,775 INFO [optim.py:369] (1/2) 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 00:53:17,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:53:18,835 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 00:53:37,765 INFO [train.py:968] (1/2) Epoch 13, batch 39200, giga_loss[loss=0.2462, simple_loss=0.3338, pruned_loss=0.07935, over 28597.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3412, pruned_loss=0.099, over 5693142.34 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3727, pruned_loss=0.1237, over 5656447.79 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3385, pruned_loss=0.09671, over 5700567.39 frames. ], batch size: 307, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:53:48,026 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=586680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:54:14,075 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=586707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:54:15,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=586709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:54:20,266 INFO [train.py:968] (1/2) Epoch 13, batch 39250, giga_loss[loss=0.2743, simple_loss=0.3573, pruned_loss=0.09566, over 28754.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3447, pruned_loss=0.1008, over 5693816.19 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3728, pruned_loss=0.1239, over 5661122.19 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3419, pruned_loss=0.09844, over 5696082.16 frames. ], batch size: 242, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:54:42,304 INFO [optim.py:369] (1/2) 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 00:55:06,363 INFO [train.py:968] (1/2) Epoch 13, batch 39300, giga_loss[loss=0.3006, simple_loss=0.38, pruned_loss=0.1106, over 28429.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3491, pruned_loss=0.103, over 5687585.91 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5658087.90 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3459, pruned_loss=0.1003, over 5692643.34 frames. ], batch size: 368, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:55:45,518 INFO [train.py:968] (1/2) Epoch 13, batch 39350, giga_loss[loss=0.255, simple_loss=0.3352, pruned_loss=0.08745, over 28884.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3504, pruned_loss=0.1032, over 5680092.01 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1243, over 5647779.30 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3468, pruned_loss=0.1001, over 5694969.57 frames. ], batch size: 145, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:56:07,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2843, 1.3175, 4.0056, 3.3260], device='cuda:1'), covar=tensor([0.1888, 0.2766, 0.0645, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0674, 0.0593, 0.0865, 0.0783], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 00:56:07,845 INFO [optim.py:369] (1/2) 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 00:56:29,687 INFO [train.py:968] (1/2) Epoch 13, batch 39400, giga_loss[loss=0.235, simple_loss=0.3259, pruned_loss=0.07201, over 28609.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3494, pruned_loss=0.1019, over 5686531.49 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3731, pruned_loss=0.1243, over 5652392.28 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3462, pruned_loss=0.09897, over 5694543.75 frames. ], batch size: 307, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:56:30,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4843, 2.2196, 1.6648, 0.6140], device='cuda:1'), covar=tensor([0.4885, 0.2254, 0.3467, 0.5750], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1491, 0.1490, 0.1308], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 00:57:10,720 INFO [train.py:968] (1/2) Epoch 13, batch 39450, giga_loss[loss=0.3182, simple_loss=0.3819, pruned_loss=0.1273, over 28659.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3486, pruned_loss=0.1015, over 5693315.85 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3733, pruned_loss=0.1243, over 5653425.56 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.0986, over 5699850.56 frames. ], batch size: 99, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:57:30,151 INFO [optim.py:369] (1/2) 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 00:57:52,200 INFO [train.py:968] (1/2) Epoch 13, batch 39500, giga_loss[loss=0.2773, simple_loss=0.3536, pruned_loss=0.1005, over 28893.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3486, pruned_loss=0.1017, over 5694997.21 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1241, over 5654288.59 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3456, pruned_loss=0.09907, over 5700414.90 frames. ], batch size: 227, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:58:35,832 INFO [train.py:968] (1/2) Epoch 13, batch 39550, giga_loss[loss=0.2785, simple_loss=0.3557, pruned_loss=0.1007, over 28869.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3481, pruned_loss=0.1012, over 5709183.40 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3729, pruned_loss=0.1239, over 5656810.54 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3457, pruned_loss=0.09907, over 5711633.53 frames. ], batch size: 145, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 00:58:47,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5876, 1.7781, 1.8469, 1.3891], device='cuda:1'), covar=tensor([0.1742, 0.2280, 0.1418, 0.1650], device='cuda:1'), in_proj_covar=tensor([0.0846, 0.0688, 0.0885, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 00:58:49,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8217, 1.8005, 1.4829, 1.5712], device='cuda:1'), covar=tensor([0.0645, 0.0432, 0.0844, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0432, 0.0500, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 00:58:56,931 INFO [optim.py:369] (1/2) 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 00:58:57,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8054, 1.0720, 2.8096, 2.7786], device='cuda:1'), covar=tensor([0.1645, 0.2597, 0.0539, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0676, 0.0594, 0.0866, 0.0785], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 00:59:16,730 INFO [train.py:968] (1/2) Epoch 13, batch 39600, giga_loss[loss=0.3722, simple_loss=0.4097, pruned_loss=0.1673, over 26698.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3521, pruned_loss=0.1033, over 5700796.97 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5651725.61 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3499, pruned_loss=0.1014, over 5707879.64 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 00:59:31,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 00:59:57,067 INFO [train.py:968] (1/2) Epoch 13, batch 39650, giga_loss[loss=0.2632, simple_loss=0.3419, pruned_loss=0.09221, over 28790.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3549, pruned_loss=0.1046, over 5702427.85 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3725, pruned_loss=0.1235, over 5654232.44 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.353, pruned_loss=0.1029, over 5707143.01 frames. ], batch size: 106, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:00:00,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:00:18,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6624, 5.4488, 5.1527, 2.4518], device='cuda:1'), covar=tensor([0.0405, 0.0574, 0.0591, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.0997, 0.0872, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 01:00:19,460 INFO [optim.py:369] (1/2) 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 01:00:26,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4065, 1.7216, 1.4858, 1.5529], device='cuda:1'), covar=tensor([0.0605, 0.0262, 0.0286, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 01:00:38,000 INFO [train.py:968] (1/2) Epoch 13, batch 39700, giga_loss[loss=0.3469, simple_loss=0.4092, pruned_loss=0.1423, over 28048.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3562, pruned_loss=0.1051, over 5705204.18 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3726, pruned_loss=0.1235, over 5656666.41 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3544, pruned_loss=0.1035, over 5707464.09 frames. ], batch size: 412, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:01:21,473 INFO [train.py:968] (1/2) Epoch 13, batch 39750, giga_loss[loss=0.2845, simple_loss=0.351, pruned_loss=0.109, over 28370.00 frames. ], tot_loss[loss=0.2865, simple_loss=0.3588, pruned_loss=0.1071, over 5705707.24 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3729, pruned_loss=0.1238, over 5660523.48 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3568, pruned_loss=0.1053, over 5704527.72 frames. ], batch size: 65, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:01:30,002 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=587225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:01:31,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=587228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:01:42,394 INFO [optim.py:369] (1/2) 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 01:01:52,987 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 39800, giga_loss[loss=0.3137, simple_loss=0.3739, pruned_loss=0.1267, over 28920.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3602, pruned_loss=0.1085, over 5712570.86 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.124, over 5667097.88 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3578, pruned_loss=0.1064, over 5706949.63 frames. ], batch size: 112, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:02:37,571 INFO [train.py:968] (1/2) Epoch 13, batch 39850, giga_loss[loss=0.2687, simple_loss=0.3454, pruned_loss=0.096, over 28906.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.359, pruned_loss=0.1077, over 5710440.06 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3731, pruned_loss=0.1239, over 5664665.94 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3569, pruned_loss=0.1057, over 5709115.61 frames. ], batch size: 227, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:02:57,600 INFO [optim.py:369] (1/2) 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 01:03:18,605 INFO [train.py:968] (1/2) Epoch 13, batch 39900, giga_loss[loss=0.2353, simple_loss=0.308, pruned_loss=0.08127, over 28846.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.356, pruned_loss=0.1063, over 5713095.64 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3731, pruned_loss=0.1239, over 5664424.05 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3542, pruned_loss=0.1046, over 5712796.12 frames. ], batch size: 112, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:03:43,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3854, 1.2978, 1.1981, 1.6286], device='cuda:1'), covar=tensor([0.0714, 0.0323, 0.0331, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0112, 0.0113, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0060, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 01:03:47,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4642, 4.2745, 4.0512, 1.8664], device='cuda:1'), covar=tensor([0.0578, 0.0711, 0.0697, 0.2160], device='cuda:1'), in_proj_covar=tensor([0.1075, 0.0996, 0.0869, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 01:03:57,549 INFO [train.py:968] (1/2) Epoch 13, batch 39950, giga_loss[loss=0.2357, simple_loss=0.3186, pruned_loss=0.07638, over 28691.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3527, pruned_loss=0.1048, over 5717776.02 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3735, pruned_loss=0.1243, over 5671394.18 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3505, pruned_loss=0.1026, over 5712694.15 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:04:22,554 INFO [optim.py:369] (1/2) 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 01:04:39,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7807, 1.8026, 1.3218, 1.4909], device='cuda:1'), covar=tensor([0.0765, 0.0592, 0.1004, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0436, 0.0503, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 01:04:40,735 INFO [train.py:968] (1/2) Epoch 13, batch 40000, giga_loss[loss=0.3029, simple_loss=0.378, pruned_loss=0.1139, over 28572.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3509, pruned_loss=0.1031, over 5720644.75 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.124, over 5677992.91 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3486, pruned_loss=0.1008, over 5711901.54 frames. ], batch size: 336, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:05:02,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:05:14,202 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:05:19,146 INFO [train.py:968] (1/2) Epoch 13, batch 40050, giga_loss[loss=0.2883, simple_loss=0.3706, pruned_loss=0.103, over 28915.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3535, pruned_loss=0.103, over 5709478.57 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3735, pruned_loss=0.1243, over 5661945.58 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1006, over 5718290.89 frames. ], batch size: 227, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:05:20,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9349, 3.7355, 3.4978, 1.7896], device='cuda:1'), covar=tensor([0.0556, 0.0723, 0.0725, 0.2463], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.1000, 0.0872, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 01:05:24,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6745, 2.4533, 1.8174, 0.7760], device='cuda:1'), covar=tensor([0.3352, 0.1904, 0.2526, 0.3860], device='cuda:1'), in_proj_covar=tensor([0.1601, 0.1508, 0.1506, 0.1320], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 01:05:42,533 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 40100, giga_loss[loss=0.2685, simple_loss=0.3383, pruned_loss=0.09936, over 28851.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3538, pruned_loss=0.103, over 5705878.70 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3734, pruned_loss=0.1243, over 5668577.10 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3515, pruned_loss=0.1005, over 5708151.36 frames. ], batch size: 106, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:06:31,466 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:06:34,263 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 40150, giga_loss[loss=0.2734, simple_loss=0.3381, pruned_loss=0.1044, over 29107.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3521, pruned_loss=0.1028, over 5709507.81 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3735, pruned_loss=0.1243, over 5671909.14 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3499, pruned_loss=0.1006, over 5709043.99 frames. ], batch size: 128, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:07:00,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=587636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:07:02,126 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:07:20,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 01:07:20,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-07 01:07:22,406 INFO [train.py:968] (1/2) Epoch 13, batch 40200, giga_loss[loss=0.26, simple_loss=0.3385, pruned_loss=0.09077, over 28737.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3522, pruned_loss=0.1044, over 5711136.18 frames. ], libri_tot_loss[loss=0.3114, simple_loss=0.3738, pruned_loss=0.1245, over 5674093.54 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3501, pruned_loss=0.1023, over 5709109.59 frames. ], batch size: 242, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:07:24,562 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=587668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:07:24,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4564, 1.7190, 1.7301, 1.2903], device='cuda:1'), covar=tensor([0.1584, 0.2171, 0.1311, 0.1481], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0683, 0.0879, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 01:07:36,530 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:08:04,760 INFO [train.py:968] (1/2) Epoch 13, batch 40250, giga_loss[loss=0.2334, simple_loss=0.3155, pruned_loss=0.07563, over 28807.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3509, pruned_loss=0.1048, over 5709342.11 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1244, over 5678700.87 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3487, pruned_loss=0.1027, over 5704111.13 frames. ], batch size: 174, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:08:19,912 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:08:25,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-07 01:08:26,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-07 01:08:27,491 INFO [optim.py:369] (1/2) 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 01:08:48,027 INFO [train.py:968] (1/2) Epoch 13, batch 40300, giga_loss[loss=0.2495, simple_loss=0.3349, pruned_loss=0.08202, over 28877.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3491, pruned_loss=0.1042, over 5720675.43 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3736, pruned_loss=0.1242, over 5684445.43 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3469, pruned_loss=0.1023, over 5712244.80 frames. ], batch size: 285, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:09:27,402 INFO [train.py:968] (1/2) Epoch 13, batch 40350, giga_loss[loss=0.2603, simple_loss=0.3357, pruned_loss=0.09246, over 29057.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3479, pruned_loss=0.1035, over 5712886.86 frames. ], libri_tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1245, over 5675374.62 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.345, pruned_loss=0.1012, over 5714419.42 frames. ], batch size: 128, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:09:39,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4066, 1.2639, 4.6942, 3.5657], device='cuda:1'), covar=tensor([0.1655, 0.2800, 0.0364, 0.0786], device='cuda:1'), in_proj_covar=tensor([0.0678, 0.0599, 0.0871, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 01:09:47,392 INFO [optim.py:369] (1/2) 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 01:09:50,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5348, 1.9460, 1.5148, 1.7312], device='cuda:1'), covar=tensor([0.2340, 0.2201, 0.2500, 0.2014], device='cuda:1'), in_proj_covar=tensor([0.1347, 0.0993, 0.1188, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 01:09:55,334 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=587851.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:10:04,829 INFO [train.py:968] (1/2) Epoch 13, batch 40400, giga_loss[loss=0.2363, simple_loss=0.3119, pruned_loss=0.08034, over 28872.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3449, pruned_loss=0.1021, over 5709107.38 frames. ], libri_tot_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1248, over 5669607.86 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3415, pruned_loss=0.09941, over 5716060.97 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:10:15,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8246, 5.1356, 2.0278, 2.4104], device='cuda:1'), covar=tensor([0.0810, 0.0332, 0.0810, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0516, 0.0348, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 01:10:19,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:10:36,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4329, 4.2849, 4.0311, 1.8905], device='cuda:1'), covar=tensor([0.0508, 0.0602, 0.0611, 0.2025], device='cuda:1'), in_proj_covar=tensor([0.1083, 0.1007, 0.0877, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 01:10:44,791 INFO [train.py:968] (1/2) Epoch 13, batch 40450, giga_loss[loss=0.2285, simple_loss=0.3021, pruned_loss=0.07748, over 28874.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3394, pruned_loss=0.09907, over 5717440.93 frames. ], libri_tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1243, over 5675072.11 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3366, pruned_loss=0.09671, over 5719208.76 frames. ], batch size: 112, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:11:07,294 INFO [optim.py:369] (1/2) 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 01:11:08,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4345, 1.6143, 1.6826, 1.5153], device='cuda:1'), covar=tensor([0.1636, 0.2030, 0.2044, 0.1991], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0736, 0.0681, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 01:11:24,921 INFO [train.py:968] (1/2) Epoch 13, batch 40500, giga_loss[loss=0.2689, simple_loss=0.3426, pruned_loss=0.09759, over 28933.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3368, pruned_loss=0.09768, over 5719148.11 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3732, pruned_loss=0.1241, over 5682037.37 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3341, pruned_loss=0.09532, over 5715853.78 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:11:29,362 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-07 01:11:35,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:11:36,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4482, 2.1412, 1.5824, 0.6605], device='cuda:1'), covar=tensor([0.4789, 0.2343, 0.3462, 0.5473], device='cuda:1'), in_proj_covar=tensor([0.1591, 0.1502, 0.1501, 0.1317], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 01:11:38,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=587982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:07,350 INFO [train.py:968] (1/2) Epoch 13, batch 40550, giga_loss[loss=0.2567, simple_loss=0.3409, pruned_loss=0.0862, over 28886.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3383, pruned_loss=0.09784, over 5705089.11 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5674633.80 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3359, pruned_loss=0.09579, over 5708921.08 frames. ], batch size: 174, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:12:14,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:16,645 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:12:40,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:41,690 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:12:48,273 INFO [train.py:968] (1/2) Epoch 13, batch 40600, giga_loss[loss=0.2898, simple_loss=0.3637, pruned_loss=0.1079, over 29025.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3432, pruned_loss=0.1001, over 5710127.81 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3733, pruned_loss=0.1241, over 5679713.58 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3403, pruned_loss=0.0978, over 5709746.94 frames. ], batch size: 155, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:13:20,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:27,394 INFO [train.py:968] (1/2) Epoch 13, batch 40650, giga_loss[loss=0.2801, simple_loss=0.3554, pruned_loss=0.1024, over 28797.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3471, pruned_loss=0.1019, over 5693096.93 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3735, pruned_loss=0.1242, over 5663336.16 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3435, pruned_loss=0.09908, over 5709055.04 frames. ], batch size: 284, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:13:32,775 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:34,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:35,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:37,166 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:13:49,504 INFO [optim.py:369] (1/2) 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 01:13:57,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:02,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:07,805 INFO [train.py:968] (1/2) Epoch 13, batch 40700, giga_loss[loss=0.2815, simple_loss=0.3475, pruned_loss=0.1077, over 28707.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3488, pruned_loss=0.1021, over 5706580.78 frames. ], libri_tot_loss[loss=0.3105, simple_loss=0.3732, pruned_loss=0.1239, over 5667763.10 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3456, pruned_loss=0.09961, over 5716220.29 frames. ], batch size: 119, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:14:36,426 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:37,669 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=588200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:37,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-07 01:14:39,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:14:40,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4699, 4.2571, 4.0905, 1.8380], device='cuda:1'), covar=tensor([0.0608, 0.0867, 0.0910, 0.2076], device='cuda:1'), in_proj_covar=tensor([0.1086, 0.1009, 0.0878, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 01:14:49,886 INFO [train.py:968] (1/2) Epoch 13, batch 40750, giga_loss[loss=0.2801, simple_loss=0.3579, pruned_loss=0.1011, over 28725.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3515, pruned_loss=0.1036, over 5697602.60 frames. ], libri_tot_loss[loss=0.3108, simple_loss=0.3735, pruned_loss=0.124, over 5660872.15 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3485, pruned_loss=0.1012, over 5712442.28 frames. ], batch size: 242, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:14:58,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588226.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:15:03,106 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:15:13,660 INFO [optim.py:369] (1/2) 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 01:15:18,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:15:21,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:15:33,610 INFO [train.py:968] (1/2) Epoch 13, batch 40800, giga_loss[loss=0.2662, simple_loss=0.3407, pruned_loss=0.09585, over 28944.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3531, pruned_loss=0.1052, over 5697119.99 frames. ], libri_tot_loss[loss=0.3104, simple_loss=0.3732, pruned_loss=0.1238, over 5664878.48 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3503, pruned_loss=0.103, over 5706239.04 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:15:45,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5287, 1.6795, 1.4975, 1.4397], device='cuda:1'), covar=tensor([0.2230, 0.1834, 0.1805, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.1773, 0.1674, 0.1642, 0.1726], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 01:15:51,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:16:22,245 INFO [train.py:968] (1/2) Epoch 13, batch 40850, giga_loss[loss=0.4333, simple_loss=0.4509, pruned_loss=0.2078, over 26654.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3621, pruned_loss=0.1133, over 5661258.47 frames. ], libri_tot_loss[loss=0.3109, simple_loss=0.3736, pruned_loss=0.1241, over 5650535.51 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3591, pruned_loss=0.1109, over 5682306.08 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:16:48,588 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-07 01:16:48,658 INFO [optim.py:369] (1/2) 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 01:16:50,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2910, 1.9010, 1.3983, 0.5376], device='cuda:1'), covar=tensor([0.3216, 0.1833, 0.2745, 0.3528], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1510, 0.1508, 0.1319], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 01:17:06,669 INFO [train.py:968] (1/2) Epoch 13, batch 40900, giga_loss[loss=0.4519, simple_loss=0.4677, pruned_loss=0.2181, over 26716.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3684, pruned_loss=0.1178, over 5658551.11 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3739, pruned_loss=0.1243, over 5648126.72 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3655, pruned_loss=0.1155, over 5678733.99 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:17:11,650 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588369.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:17:14,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:17:40,390 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=588401.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:17:43,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 01:17:52,946 INFO [train.py:968] (1/2) Epoch 13, batch 40950, giga_loss[loss=0.4276, simple_loss=0.4342, pruned_loss=0.2105, over 23388.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3748, pruned_loss=0.123, over 5652897.56 frames. ], libri_tot_loss[loss=0.311, simple_loss=0.3738, pruned_loss=0.1241, over 5650903.09 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3726, pruned_loss=0.1212, over 5666841.88 frames. ], batch size: 705, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:18:17,802 INFO [optim.py:369] (1/2) 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 01:18:35,641 INFO [train.py:968] (1/2) Epoch 13, batch 41000, giga_loss[loss=0.376, simple_loss=0.4318, pruned_loss=0.1601, over 29011.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3807, pruned_loss=0.1282, over 5657480.02 frames. ], libri_tot_loss[loss=0.3113, simple_loss=0.3741, pruned_loss=0.1243, over 5647751.63 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3787, pruned_loss=0.1266, over 5670733.67 frames. ], batch size: 128, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:18:55,673 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=588488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:19:17,352 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5737, 5.3753, 5.0653, 2.5262], device='cuda:1'), covar=tensor([0.0413, 0.0604, 0.0681, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.1013, 0.0881, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 01:19:20,979 INFO [train.py:968] (1/2) Epoch 13, batch 41050, giga_loss[loss=0.286, simple_loss=0.3662, pruned_loss=0.1029, over 28999.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3837, pruned_loss=0.1305, over 5666763.06 frames. ], libri_tot_loss[loss=0.3106, simple_loss=0.3735, pruned_loss=0.1238, over 5656208.26 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3829, pruned_loss=0.1298, over 5670352.95 frames. ], batch size: 155, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:19:52,220 INFO [optim.py:369] (1/2) 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 01:20:11,977 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4493, 1.8941, 1.4165, 1.2636], device='cuda:1'), covar=tensor([0.2317, 0.2193, 0.2567, 0.1979], device='cuda:1'), in_proj_covar=tensor([0.1339, 0.0986, 0.1186, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 01:20:13,015 INFO [train.py:968] (1/2) Epoch 13, batch 41100, libri_loss[loss=0.3052, simple_loss=0.3774, pruned_loss=0.1165, over 29275.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3851, pruned_loss=0.1327, over 5658627.11 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3731, pruned_loss=0.1237, over 5657854.43 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3851, pruned_loss=0.1325, over 5659573.37 frames. ], batch size: 97, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:20:23,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:20:44,262 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3560, 1.5093, 1.3789, 1.2262], device='cuda:1'), covar=tensor([0.2500, 0.2212, 0.1616, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.1775, 0.1684, 0.1654, 0.1745], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 01:21:11,009 INFO [train.py:968] (1/2) Epoch 13, batch 41150, giga_loss[loss=0.3472, simple_loss=0.4095, pruned_loss=0.1424, over 28819.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3888, pruned_loss=0.1373, over 5633033.12 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3731, pruned_loss=0.1237, over 5659334.16 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3888, pruned_loss=0.1372, over 5632618.84 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:21:40,211 INFO [optim.py:369] (1/2) 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 01:22:02,182 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=588663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:22:03,795 INFO [train.py:968] (1/2) Epoch 13, batch 41200, giga_loss[loss=0.4561, simple_loss=0.4678, pruned_loss=0.2222, over 26612.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3924, pruned_loss=0.1412, over 5624655.92 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3731, pruned_loss=0.1236, over 5663915.53 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3928, pruned_loss=0.1415, over 5619857.78 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:22:53,989 INFO [train.py:968] (1/2) Epoch 13, batch 41250, giga_loss[loss=0.4927, simple_loss=0.4859, pruned_loss=0.2498, over 26501.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.3964, pruned_loss=0.1442, over 5625906.29 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3731, pruned_loss=0.1236, over 5659370.21 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3971, pruned_loss=0.1449, over 5624675.45 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:22:56,548 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=588718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:22:59,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=588721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:23:04,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-07 01:23:23,268 INFO [optim.py:369] (1/2) 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 01:23:28,131 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 41300, giga_loss[loss=0.3684, simple_loss=0.4164, pruned_loss=0.1602, over 28282.00 frames. ], tot_loss[loss=0.3424, simple_loss=0.396, pruned_loss=0.1444, over 5634576.45 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3725, pruned_loss=0.1232, over 5663910.96 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.3975, pruned_loss=0.1456, over 5629196.03 frames. ], batch size: 369, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:24:24,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4045, 1.5898, 1.4720, 1.5657], device='cuda:1'), covar=tensor([0.0626, 0.0281, 0.0265, 0.0620], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 01:24:31,581 INFO [train.py:968] (1/2) Epoch 13, batch 41350, giga_loss[loss=0.361, simple_loss=0.4109, pruned_loss=0.1556, over 28932.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3936, pruned_loss=0.1433, over 5640130.68 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3721, pruned_loss=0.1228, over 5671413.99 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.3963, pruned_loss=0.1456, over 5627638.02 frames. ], batch size: 174, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:24:36,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5356, 5.3726, 5.1214, 2.5239], device='cuda:1'), covar=tensor([0.0464, 0.0597, 0.0670, 0.1826], device='cuda:1'), in_proj_covar=tensor([0.1103, 0.1026, 0.0896, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 01:25:01,647 INFO [optim.py:369] (1/2) 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 01:25:20,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=588863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:25:21,470 INFO [train.py:968] (1/2) Epoch 13, batch 41400, giga_loss[loss=0.3182, simple_loss=0.3808, pruned_loss=0.1278, over 28909.00 frames. ], tot_loss[loss=0.3414, simple_loss=0.3945, pruned_loss=0.1442, over 5637146.75 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3721, pruned_loss=0.1227, over 5673705.99 frames. ], giga_tot_loss[loss=0.345, simple_loss=0.397, pruned_loss=0.1465, over 5624516.49 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:26:11,181 INFO [train.py:968] (1/2) Epoch 13, batch 41450, giga_loss[loss=0.2998, simple_loss=0.3721, pruned_loss=0.1138, over 28895.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3933, pruned_loss=0.1422, over 5629372.31 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3722, pruned_loss=0.1226, over 5679107.53 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3957, pruned_loss=0.1446, over 5613493.61 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:26:42,643 INFO [optim.py:369] (1/2) 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 01:26:58,573 INFO [train.py:968] (1/2) Epoch 13, batch 41500, giga_loss[loss=0.3583, simple_loss=0.3859, pruned_loss=0.1653, over 23603.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3971, pruned_loss=0.1452, over 5611286.82 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3727, pruned_loss=0.1232, over 5665650.51 frames. ], giga_tot_loss[loss=0.3468, simple_loss=0.3992, pruned_loss=0.1472, over 5609787.66 frames. ], batch size: 705, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:27:41,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:27:46,057 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 41550, giga_loss[loss=0.3029, simple_loss=0.3668, pruned_loss=0.1195, over 29058.00 frames. ], tot_loss[loss=0.3412, simple_loss=0.3949, pruned_loss=0.1438, over 5604919.22 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1226, over 5671292.21 frames. ], giga_tot_loss[loss=0.3455, simple_loss=0.398, pruned_loss=0.1465, over 5596778.22 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:28:16,667 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:28:16,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=589038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:28:23,388 INFO [optim.py:369] (1/2) 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 01:28:43,192 INFO [train.py:968] (1/2) Epoch 13, batch 41600, giga_loss[loss=0.3265, simple_loss=0.3878, pruned_loss=0.1326, over 27953.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.391, pruned_loss=0.1391, over 5617729.77 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3716, pruned_loss=0.1224, over 5674756.99 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3941, pruned_loss=0.1417, over 5607318.38 frames. ], batch size: 412, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:29:33,317 INFO [train.py:968] (1/2) Epoch 13, batch 41650, giga_loss[loss=0.4172, simple_loss=0.444, pruned_loss=0.1952, over 26554.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3882, pruned_loss=0.1357, over 5630568.23 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3718, pruned_loss=0.1226, over 5674691.08 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3907, pruned_loss=0.1378, over 5621522.67 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:30:01,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5488, 2.2021, 1.6218, 0.7494], device='cuda:1'), covar=tensor([0.5125, 0.2459, 0.3372, 0.5608], device='cuda:1'), in_proj_covar=tensor([0.1623, 0.1535, 0.1519, 0.1334], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 01:30:01,325 INFO [optim.py:369] (1/2) 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 01:30:16,641 INFO [train.py:968] (1/2) Epoch 13, batch 41700, giga_loss[loss=0.321, simple_loss=0.3813, pruned_loss=0.1304, over 28615.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3836, pruned_loss=0.1326, over 5634797.51 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.1221, over 5684214.43 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3874, pruned_loss=0.1353, over 5616782.87 frames. ], batch size: 307, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:30:20,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=589171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:30:29,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:30:32,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:30:58,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:31:01,762 INFO [train.py:968] (1/2) Epoch 13, batch 41750, giga_loss[loss=0.2675, simple_loss=0.3389, pruned_loss=0.09805, over 28484.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3817, pruned_loss=0.1309, over 5636906.32 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3704, pruned_loss=0.122, over 5684521.96 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3854, pruned_loss=0.1335, over 5620330.91 frames. ], batch size: 78, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:31:02,761 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=589216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:31:31,649 INFO [optim.py:369] (1/2) 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 01:31:34,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=589247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:31:48,886 INFO [train.py:968] (1/2) Epoch 13, batch 41800, libri_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1209, over 28590.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3803, pruned_loss=0.1296, over 5651401.81 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3705, pruned_loss=0.122, over 5688886.27 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3835, pruned_loss=0.132, over 5632873.54 frames. ], batch size: 106, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:32:09,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 01:32:34,317 INFO [train.py:968] (1/2) Epoch 13, batch 41850, giga_loss[loss=0.2824, simple_loss=0.3596, pruned_loss=0.1026, over 28994.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3789, pruned_loss=0.1285, over 5657574.52 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3705, pruned_loss=0.1222, over 5692786.88 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3816, pruned_loss=0.1303, over 5638718.29 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:33:06,300 INFO [optim.py:369] (1/2) 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 01:33:27,965 INFO [train.py:968] (1/2) Epoch 13, batch 41900, giga_loss[loss=0.3223, simple_loss=0.3747, pruned_loss=0.135, over 27544.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3792, pruned_loss=0.1288, over 5645846.50 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5696051.21 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3814, pruned_loss=0.1302, over 5627424.54 frames. ], batch size: 472, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:34:23,095 INFO [train.py:968] (1/2) Epoch 13, batch 41950, giga_loss[loss=0.2802, simple_loss=0.3697, pruned_loss=0.09533, over 28962.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3789, pruned_loss=0.126, over 5650920.94 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3707, pruned_loss=0.1224, over 5696931.95 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3805, pruned_loss=0.127, over 5635418.34 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:34:58,450 INFO [optim.py:369] (1/2) 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 01:35:14,864 INFO [train.py:968] (1/2) Epoch 13, batch 42000, giga_loss[loss=0.3051, simple_loss=0.3782, pruned_loss=0.116, over 28958.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3805, pruned_loss=0.1256, over 5652682.20 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1227, over 5688834.02 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3818, pruned_loss=0.1262, over 5646792.75 frames. ], batch size: 145, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:35:14,864 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 01:35:20,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3475, 1.3623, 1.1986, 1.5204], device='cuda:1'), covar=tensor([0.0789, 0.0329, 0.0333, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 01:35:23,302 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 01:35:44,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3395, 1.6148, 1.4241, 1.5113], device='cuda:1'), covar=tensor([0.0774, 0.0305, 0.0295, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 01:36:08,394 INFO [train.py:968] (1/2) Epoch 13, batch 42050, giga_loss[loss=0.2955, simple_loss=0.3687, pruned_loss=0.1111, over 28956.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3812, pruned_loss=0.126, over 5664021.58 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3705, pruned_loss=0.1225, over 5693051.10 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3828, pruned_loss=0.1267, over 5654591.72 frames. ], batch size: 164, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:36:36,087 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:36:52,857 INFO [train.py:968] (1/2) Epoch 13, batch 42100, giga_loss[loss=0.2902, simple_loss=0.3608, pruned_loss=0.1098, over 29015.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.381, pruned_loss=0.1264, over 5665816.77 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1225, over 5699414.44 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3825, pruned_loss=0.1271, over 5652034.12 frames. ], batch size: 164, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:36:59,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5159, 2.1637, 1.8738, 1.5855], device='cuda:1'), covar=tensor([0.0731, 0.0244, 0.0251, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 01:37:16,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=589591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:37:38,110 INFO [train.py:968] (1/2) Epoch 13, batch 42150, giga_loss[loss=0.3137, simple_loss=0.3732, pruned_loss=0.1271, over 28922.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3786, pruned_loss=0.1256, over 5668220.93 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5692552.68 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3799, pruned_loss=0.1261, over 5663076.61 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:37:44,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=589622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:38:06,708 INFO [optim.py:369] (1/2) 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 01:38:20,058 INFO [train.py:968] (1/2) Epoch 13, batch 42200, giga_loss[loss=0.3112, simple_loss=0.3746, pruned_loss=0.1239, over 28267.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3778, pruned_loss=0.1266, over 5659972.99 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3707, pruned_loss=0.1225, over 5690512.56 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3791, pruned_loss=0.1273, over 5655571.46 frames. ], batch size: 368, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:38:44,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:38:46,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:07,735 INFO [train.py:968] (1/2) Epoch 13, batch 42250, giga_loss[loss=0.3782, simple_loss=0.4172, pruned_loss=0.1696, over 26615.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3779, pruned_loss=0.1269, over 5656544.71 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1232, over 5687248.52 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3785, pruned_loss=0.1269, over 5655316.67 frames. ], batch size: 555, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:39:13,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:27,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:29,035 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:39:48,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5195, 1.5656, 1.2225, 1.2063], device='cuda:1'), covar=tensor([0.0745, 0.0479, 0.0963, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0437, 0.0501, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 01:39:52,736 INFO [train.py:968] (1/2) Epoch 13, batch 42300, giga_loss[loss=0.3492, simple_loss=0.4067, pruned_loss=0.1459, over 28908.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3782, pruned_loss=0.1256, over 5672034.46 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3713, pruned_loss=0.1231, over 5692936.32 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3789, pruned_loss=0.1258, over 5665039.15 frames. ], batch size: 174, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:39:53,049 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=589765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:53,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:39:56,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=589768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:40:03,898 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4147, 1.5949, 1.6536, 1.2292], device='cuda:1'), covar=tensor([0.1701, 0.2344, 0.1384, 0.1650], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0696, 0.0885, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 01:40:19,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2861, 1.2050, 3.7792, 3.2098], device='cuda:1'), covar=tensor([0.1686, 0.2779, 0.0496, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0685, 0.0605, 0.0883, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 01:40:23,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=589797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:40:29,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3457, 1.5924, 1.2331, 1.5840], device='cuda:1'), covar=tensor([0.2670, 0.2601, 0.2992, 0.2167], device='cuda:1'), in_proj_covar=tensor([0.1351, 0.0992, 0.1197, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 01:40:32,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-07 01:40:41,865 INFO [train.py:968] (1/2) Epoch 13, batch 42350, giga_loss[loss=0.3374, simple_loss=0.3763, pruned_loss=0.1493, over 23611.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3777, pruned_loss=0.1248, over 5677716.96 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3711, pruned_loss=0.1229, over 5696018.97 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3786, pruned_loss=0.1252, over 5669186.48 frames. ], batch size: 705, lr: 2.44e-03, grad_scale: 2.0
+2023-03-07 01:41:13,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6215, 1.6232, 1.2125, 1.2531], device='cuda:1'), covar=tensor([0.0723, 0.0559, 0.0933, 0.1203], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0438, 0.0502, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 01:41:14,997 INFO [optim.py:369] (1/2) 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 01:41:30,102 INFO [train.py:968] (1/2) Epoch 13, batch 42400, giga_loss[loss=0.2921, simple_loss=0.3585, pruned_loss=0.1128, over 28817.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3771, pruned_loss=0.1249, over 5674390.80 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3703, pruned_loss=0.1224, over 5701055.80 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3787, pruned_loss=0.1257, over 5662511.96 frames. ], batch size: 186, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:41:41,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1749, 1.3730, 1.2203, 0.9876], device='cuda:1'), covar=tensor([0.2145, 0.2007, 0.1418, 0.1915], device='cuda:1'), in_proj_covar=tensor([0.1790, 0.1698, 0.1662, 0.1756], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 01:42:17,204 INFO [train.py:968] (1/2) Epoch 13, batch 42450, libri_loss[loss=0.3425, simple_loss=0.3905, pruned_loss=0.1473, over 29565.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3751, pruned_loss=0.1237, over 5684851.26 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3701, pruned_loss=0.1223, over 5703954.65 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3766, pruned_loss=0.1244, over 5672279.70 frames. ], batch size: 76, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:42:24,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5409, 1.7687, 1.7952, 1.3421], device='cuda:1'), covar=tensor([0.1654, 0.2325, 0.1347, 0.1586], device='cuda:1'), in_proj_covar=tensor([0.0850, 0.0697, 0.0889, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 01:42:47,948 INFO [optim.py:369] (1/2) 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 01:43:04,517 INFO [train.py:968] (1/2) Epoch 13, batch 42500, giga_loss[loss=0.279, simple_loss=0.3547, pruned_loss=0.1017, over 28984.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3743, pruned_loss=0.1239, over 5675410.38 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5705542.28 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3757, pruned_loss=0.1246, over 5664010.30 frames. ], batch size: 164, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:43:14,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-07 01:43:55,784 INFO [train.py:968] (1/2) Epoch 13, batch 42550, giga_loss[loss=0.2595, simple_loss=0.3431, pruned_loss=0.08797, over 28829.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1236, over 5687578.51 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1221, over 5708529.00 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3743, pruned_loss=0.1242, over 5675439.35 frames. ], batch size: 174, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:44:27,729 INFO [optim.py:369] (1/2) 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 01:44:43,432 INFO [train.py:968] (1/2) Epoch 13, batch 42600, libri_loss[loss=0.3663, simple_loss=0.4179, pruned_loss=0.1574, over 25955.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3715, pruned_loss=0.123, over 5679406.37 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3699, pruned_loss=0.122, over 5703019.99 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3725, pruned_loss=0.1236, over 5674006.84 frames. ], batch size: 136, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:45:00,277 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:45:01,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-07 01:45:28,085 INFO [train.py:968] (1/2) Epoch 13, batch 42650, giga_loss[loss=0.267, simple_loss=0.3394, pruned_loss=0.09724, over 28619.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3719, pruned_loss=0.1241, over 5672425.80 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.1219, over 5707260.31 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3729, pruned_loss=0.1248, over 5663887.07 frames. ], batch size: 85, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:45:38,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 01:45:59,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:45:59,663 INFO [optim.py:369] (1/2) 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 01:46:16,462 INFO [train.py:968] (1/2) Epoch 13, batch 42700, giga_loss[loss=0.3067, simple_loss=0.3716, pruned_loss=0.1209, over 28685.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3712, pruned_loss=0.1241, over 5644844.93 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3699, pruned_loss=0.122, over 5689556.85 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3719, pruned_loss=0.1246, over 5653534.42 frames. ], batch size: 262, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:46:25,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3362, 1.6592, 1.3137, 1.5441], device='cuda:1'), covar=tensor([0.0766, 0.0307, 0.0325, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 01:47:00,873 INFO [train.py:968] (1/2) Epoch 13, batch 42750, giga_loss[loss=0.3581, simple_loss=0.4119, pruned_loss=0.1522, over 28509.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1244, over 5651875.95 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3704, pruned_loss=0.1225, over 5690454.48 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3728, pruned_loss=0.1244, over 5657154.73 frames. ], batch size: 336, lr: 2.44e-03, grad_scale: 4.0
+2023-03-07 01:47:02,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9105, 0.9942, 3.4107, 2.9554], device='cuda:1'), covar=tensor([0.1777, 0.2741, 0.0508, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0692, 0.0612, 0.0891, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 01:47:16,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 01:47:31,719 INFO [optim.py:369] (1/2) 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 01:47:46,681 INFO [train.py:968] (1/2) Epoch 13, batch 42800, giga_loss[loss=0.2824, simple_loss=0.3568, pruned_loss=0.104, over 28911.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.124, over 5660626.24 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1225, over 5692538.65 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.124, over 5662581.55 frames. ], batch size: 213, lr: 2.44e-03, grad_scale: 8.0
+2023-03-07 01:48:00,399 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.60 vs. limit=5.0
+2023-03-07 01:48:31,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:48:32,153 INFO [train.py:968] (1/2) Epoch 13, batch 42850, giga_loss[loss=0.3242, simple_loss=0.3807, pruned_loss=0.1338, over 28679.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3737, pruned_loss=0.1232, over 5674074.60 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3707, pruned_loss=0.1226, over 5695450.36 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3736, pruned_loss=0.1231, over 5672455.16 frames. ], batch size: 92, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:48:56,462 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-07 01:49:08,692 INFO [optim.py:369] (1/2) 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 01:49:24,060 INFO [train.py:968] (1/2) Epoch 13, batch 42900, giga_loss[loss=0.3577, simple_loss=0.4059, pruned_loss=0.1547, over 28505.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3739, pruned_loss=0.1235, over 5672449.74 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1224, over 5691514.31 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3741, pruned_loss=0.1236, over 5673981.96 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:49:39,831 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590383.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:49:54,115 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 01:50:10,090 INFO [train.py:968] (1/2) Epoch 13, batch 42950, giga_loss[loss=0.305, simple_loss=0.3666, pruned_loss=0.1217, over 28670.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3764, pruned_loss=0.126, over 5683111.51 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5695393.66 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3767, pruned_loss=0.1262, over 5680619.19 frames. ], batch size: 242, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:50:43,572 INFO [optim.py:369] (1/2) 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 01:50:55,147 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:50:58,404 INFO [train.py:968] (1/2) Epoch 13, batch 43000, giga_loss[loss=0.4004, simple_loss=0.4079, pruned_loss=0.1964, over 23490.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5678010.80 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3703, pruned_loss=0.1223, over 5692076.75 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3774, pruned_loss=0.128, over 5677701.01 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:51:52,221 INFO [train.py:968] (1/2) Epoch 13, batch 43050, giga_loss[loss=0.4072, simple_loss=0.4375, pruned_loss=0.1884, over 27632.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3793, pruned_loss=0.1305, over 5673034.13 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5693827.55 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3799, pruned_loss=0.1309, over 5671097.18 frames. ], batch size: 472, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:51:58,636 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:52:08,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7667, 5.6315, 5.3250, 2.7210], device='cuda:1'), covar=tensor([0.0383, 0.0518, 0.0602, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.1116, 0.1033, 0.0902, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 01:52:25,089 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 43100, giga_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1243, over 28561.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3801, pruned_loss=0.1317, over 5662208.77 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5694146.49 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3804, pruned_loss=0.1319, over 5659777.94 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:53:11,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:53:13,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:53:21,866 INFO [train.py:968] (1/2) Epoch 13, batch 43150, giga_loss[loss=0.3887, simple_loss=0.4233, pruned_loss=0.1771, over 28483.00 frames. ], tot_loss[loss=0.319, simple_loss=0.378, pruned_loss=0.13, over 5658783.47 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5691395.82 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.379, pruned_loss=0.1308, over 5658248.57 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 01:53:42,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=590635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:53:49,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5144, 1.7304, 1.5728, 1.5597], device='cuda:1'), covar=tensor([0.1729, 0.2121, 0.2280, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0739, 0.0686, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 01:53:55,784 INFO [optim.py:369] (1/2) 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 01:54:06,892 INFO [train.py:968] (1/2) Epoch 13, batch 43200, libri_loss[loss=0.2632, simple_loss=0.3393, pruned_loss=0.09354, over 29574.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3785, pruned_loss=0.1289, over 5668895.24 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5694114.35 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3793, pruned_loss=0.1297, over 5665302.35 frames. ], batch size: 77, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:54:07,280 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:54:12,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:54:18,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-07 01:54:30,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:54:36,068 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 43250, giga_loss[loss=0.2855, simple_loss=0.3288, pruned_loss=0.1211, over 23932.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3766, pruned_loss=0.1269, over 5664466.16 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1225, over 5699111.65 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3772, pruned_loss=0.1273, over 5656266.17 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:55:24,265 INFO [optim.py:369] (1/2) 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 01:55:32,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=590758.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:55:37,946 INFO [train.py:968] (1/2) Epoch 13, batch 43300, giga_loss[loss=0.2758, simple_loss=0.3454, pruned_loss=0.1031, over 28864.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3747, pruned_loss=0.1259, over 5669067.63 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5696252.41 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3751, pruned_loss=0.1262, over 5664430.46 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:56:16,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:56:22,284 INFO [train.py:968] (1/2) Epoch 13, batch 43350, libri_loss[loss=0.312, simple_loss=0.3837, pruned_loss=0.1201, over 29497.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3736, pruned_loss=0.1261, over 5657697.01 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.1229, over 5691324.79 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3735, pruned_loss=0.1261, over 5658419.77 frames. ], batch size: 85, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:56:34,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-07 01:56:38,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:56:40,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:56:51,787 INFO [optim.py:369] (1/2) 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 01:57:04,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=590864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 01:57:05,217 INFO [train.py:968] (1/2) Epoch 13, batch 43400, giga_loss[loss=0.2833, simple_loss=0.3607, pruned_loss=0.103, over 28881.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3738, pruned_loss=0.1264, over 5651264.43 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3706, pruned_loss=0.1228, over 5677510.23 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3742, pruned_loss=0.1266, over 5663189.19 frames. ], batch size: 227, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:57:38,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=590901.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:57:42,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590904.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 01:57:50,486 INFO [train.py:968] (1/2) Epoch 13, batch 43450, giga_loss[loss=0.306, simple_loss=0.3751, pruned_loss=0.1184, over 29044.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.377, pruned_loss=0.1279, over 5650508.13 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3705, pruned_loss=0.1227, over 5675535.81 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1283, over 5661780.04 frames. ], batch size: 155, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:57:57,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2817, 4.0570, 3.8919, 1.7002], device='cuda:1'), covar=tensor([0.0740, 0.1009, 0.1038, 0.2265], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.1040, 0.0907, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 01:58:07,785 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=590933.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:58:15,678 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=590942.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 01:58:22,829 INFO [optim.py:369] (1/2) 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 01:58:38,336 INFO [train.py:968] (1/2) Epoch 13, batch 43500, giga_loss[loss=0.272, simple_loss=0.3611, pruned_loss=0.0915, over 28749.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3782, pruned_loss=0.1254, over 5662437.38 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3701, pruned_loss=0.1225, over 5681461.06 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3792, pruned_loss=0.126, over 5665431.90 frames. ], batch size: 242, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 01:59:28,234 INFO [train.py:968] (1/2) Epoch 13, batch 43550, giga_loss[loss=0.2931, simple_loss=0.3652, pruned_loss=0.1105, over 29014.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3798, pruned_loss=0.126, over 5655783.10 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3703, pruned_loss=0.1228, over 5674721.80 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3806, pruned_loss=0.1262, over 5663726.40 frames. ], batch size: 136, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:00:02,271 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 43600, giga_loss[loss=0.363, simple_loss=0.4132, pruned_loss=0.1564, over 28664.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3825, pruned_loss=0.1284, over 5652327.74 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.1231, over 5670655.60 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3834, pruned_loss=0.1286, over 5661851.79 frames. ], batch size: 262, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:01:03,113 INFO [train.py:968] (1/2) Epoch 13, batch 43650, giga_loss[loss=0.3248, simple_loss=0.386, pruned_loss=0.1318, over 28874.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3829, pruned_loss=0.1294, over 5643396.23 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3709, pruned_loss=0.1233, over 5659997.11 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3834, pruned_loss=0.1294, over 5659166.99 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:01:19,808 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=591134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:01:34,715 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 43700, giga_loss[loss=0.3826, simple_loss=0.4014, pruned_loss=0.1819, over 23715.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1304, over 5652803.67 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.1231, over 5664121.95 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3841, pruned_loss=0.1308, over 5661409.49 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:02:01,739 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 43750, giga_loss[loss=0.3376, simple_loss=0.3947, pruned_loss=0.1403, over 28742.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3817, pruned_loss=0.1304, over 5657136.56 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3701, pruned_loss=0.1228, over 5668791.19 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3831, pruned_loss=0.1311, over 5659357.27 frames. ], batch size: 284, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:03:05,569 INFO [optim.py:369] (1/2) 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 02:03:17,897 INFO [train.py:968] (1/2) Epoch 13, batch 43800, giga_loss[loss=0.2815, simple_loss=0.3502, pruned_loss=0.1064, over 29006.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3787, pruned_loss=0.1285, over 5660372.19 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3705, pruned_loss=0.1229, over 5666446.42 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3799, pruned_loss=0.1292, over 5664272.63 frames. ], batch size: 128, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:04:06,919 INFO [train.py:968] (1/2) Epoch 13, batch 43850, giga_loss[loss=0.2771, simple_loss=0.3476, pruned_loss=0.1033, over 29029.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3768, pruned_loss=0.1278, over 5662678.80 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1227, over 5669870.78 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3781, pruned_loss=0.1287, over 5662542.35 frames. ], batch size: 128, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:04:10,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=591317.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 02:04:16,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=591324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:04:19,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=591327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:04:41,145 INFO [optim.py:369] (1/2) 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 02:04:47,202 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=591356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:04:55,326 INFO [train.py:968] (1/2) Epoch 13, batch 43900, giga_loss[loss=0.2971, simple_loss=0.3656, pruned_loss=0.1143, over 28840.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3779, pruned_loss=0.1286, over 5671141.45 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3706, pruned_loss=0.123, over 5668748.55 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3789, pruned_loss=0.1293, over 5672141.91 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:05:27,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3510, 3.1520, 3.0075, 1.2978], device='cuda:1'), covar=tensor([0.0915, 0.1083, 0.0940, 0.2288], device='cuda:1'), in_proj_covar=tensor([0.1124, 0.1042, 0.0909, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 02:05:44,223 INFO [train.py:968] (1/2) Epoch 13, batch 43950, giga_loss[loss=0.3134, simple_loss=0.3808, pruned_loss=0.1231, over 28985.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3772, pruned_loss=0.1287, over 5666524.75 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3702, pruned_loss=0.1227, over 5672964.23 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3784, pruned_loss=0.1296, over 5663488.78 frames. ], batch size: 213, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:06:16,352 INFO [optim.py:369] (1/2) 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 02:06:20,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3091, 2.7805, 1.4930, 1.4422], device='cuda:1'), covar=tensor([0.0891, 0.0348, 0.0807, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0525, 0.0352, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 02:06:25,708 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=591460.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 02:06:27,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=591463.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 02:06:29,278 INFO [train.py:968] (1/2) Epoch 13, batch 44000, giga_loss[loss=0.303, simple_loss=0.3682, pruned_loss=0.1189, over 28873.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3754, pruned_loss=0.1279, over 5662264.04 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3706, pruned_loss=0.123, over 5663412.57 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3762, pruned_loss=0.1284, over 5667633.32 frames. ], batch size: 227, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:06:54,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=591492.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 02:07:10,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=591509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:07:13,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2989, 1.4590, 1.4372, 1.2930], device='cuda:1'), covar=tensor([0.1402, 0.1522, 0.1904, 0.1612], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0737, 0.0683, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 02:07:15,188 INFO [train.py:968] (1/2) Epoch 13, batch 44050, giga_loss[loss=0.3085, simple_loss=0.3809, pruned_loss=0.1181, over 28966.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3751, pruned_loss=0.1273, over 5659114.66 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.1231, over 5656863.22 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3757, pruned_loss=0.1277, over 5668900.14 frames. ], batch size: 186, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:07:55,024 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 44100, giga_loss[loss=0.2918, simple_loss=0.3661, pruned_loss=0.1088, over 28717.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3773, pruned_loss=0.1281, over 5658970.73 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.371, pruned_loss=0.1234, over 5657286.54 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3775, pruned_loss=0.1282, over 5666378.98 frames. ], batch size: 119, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:08:40,180 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=591604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:08:44,451 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-07 02:08:48,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4685, 1.5926, 1.5810, 1.3314], device='cuda:1'), covar=tensor([0.2168, 0.1977, 0.1463, 0.1853], device='cuda:1'), in_proj_covar=tensor([0.1786, 0.1685, 0.1659, 0.1756], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 02:08:51,917 INFO [train.py:968] (1/2) Epoch 13, batch 44150, giga_loss[loss=0.3864, simple_loss=0.4223, pruned_loss=0.1752, over 28254.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3776, pruned_loss=0.1281, over 5671813.82 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3708, pruned_loss=0.1231, over 5664340.00 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3781, pruned_loss=0.1286, over 5671387.86 frames. ], batch size: 368, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:09:14,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2569, 4.0991, 3.8854, 1.8799], device='cuda:1'), covar=tensor([0.0590, 0.0723, 0.0766, 0.2123], device='cuda:1'), in_proj_covar=tensor([0.1121, 0.1042, 0.0909, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 02:09:25,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=591648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:09:29,019 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=591652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:09:29,331 INFO [optim.py:369] (1/2) 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 02:09:31,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=591655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:09:37,971 INFO [train.py:968] (1/2) Epoch 13, batch 44200, giga_loss[loss=0.3125, simple_loss=0.3812, pruned_loss=0.1219, over 28606.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3774, pruned_loss=0.128, over 5664795.81 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3712, pruned_loss=0.1234, over 5666262.89 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3777, pruned_loss=0.1283, over 5662690.82 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:09:46,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3833, 1.5380, 1.3618, 1.5887], device='cuda:1'), covar=tensor([0.0672, 0.0386, 0.0312, 0.0697], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:09:56,779 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=591684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:10:23,317 INFO [train.py:968] (1/2) Epoch 13, batch 44250, giga_loss[loss=0.2777, simple_loss=0.3615, pruned_loss=0.0969, over 28519.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3783, pruned_loss=0.1261, over 5662279.91 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3708, pruned_loss=0.1232, over 5660066.82 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3789, pruned_loss=0.1265, over 5666237.39 frames. ], batch size: 85, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:10:53,851 INFO [optim.py:369] (1/2) 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 02:11:04,194 INFO [train.py:968] (1/2) Epoch 13, batch 44300, giga_loss[loss=0.3205, simple_loss=0.3974, pruned_loss=0.1218, over 28843.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3795, pruned_loss=0.1246, over 5678684.90 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.371, pruned_loss=0.1233, over 5666834.08 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3801, pruned_loss=0.1249, over 5676034.65 frames. ], batch size: 174, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:11:28,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5518, 1.7665, 1.6625, 1.3912], device='cuda:1'), covar=tensor([0.2396, 0.1703, 0.1311, 0.1792], device='cuda:1'), in_proj_covar=tensor([0.1779, 0.1682, 0.1651, 0.1751], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 02:11:54,522 INFO [train.py:968] (1/2) Epoch 13, batch 44350, giga_loss[loss=0.3283, simple_loss=0.3887, pruned_loss=0.134, over 28696.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3815, pruned_loss=0.1258, over 5682250.10 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3704, pruned_loss=0.1229, over 5665924.54 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3829, pruned_loss=0.1264, over 5681290.29 frames. ], batch size: 262, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:12:31,255 INFO [optim.py:369] (1/2) 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 02:12:34,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4469, 1.8371, 1.4042, 1.6719], device='cuda:1'), covar=tensor([0.2577, 0.2485, 0.2792, 0.2244], device='cuda:1'), in_proj_covar=tensor([0.1351, 0.0990, 0.1190, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 02:12:41,568 INFO [train.py:968] (1/2) Epoch 13, batch 44400, giga_loss[loss=0.3391, simple_loss=0.393, pruned_loss=0.1426, over 28744.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3845, pruned_loss=0.1292, over 5672595.00 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3701, pruned_loss=0.1226, over 5662672.74 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3863, pruned_loss=0.1302, over 5675214.69 frames. ], batch size: 99, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:13:20,872 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-07 02:13:34,555 INFO [train.py:968] (1/2) Epoch 13, batch 44450, giga_loss[loss=0.334, simple_loss=0.3988, pruned_loss=0.1345, over 28609.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3861, pruned_loss=0.1316, over 5653344.17 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3703, pruned_loss=0.1227, over 5663536.38 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3874, pruned_loss=0.1323, over 5654554.69 frames. ], batch size: 262, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:13:34,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5208, 1.9693, 1.5080, 1.4710], device='cuda:1'), covar=tensor([0.0739, 0.0259, 0.0303, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:14:06,977 INFO [optim.py:369] (1/2) 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 02:14:15,896 INFO [train.py:968] (1/2) Epoch 13, batch 44500, giga_loss[loss=0.2945, simple_loss=0.3661, pruned_loss=0.1115, over 29015.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3852, pruned_loss=0.1316, over 5664959.62 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3701, pruned_loss=0.1227, over 5672334.04 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3872, pruned_loss=0.1326, over 5657488.03 frames. ], batch size: 128, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:14:27,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=591979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:14:41,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-07 02:15:01,080 INFO [train.py:968] (1/2) Epoch 13, batch 44550, giga_loss[loss=0.2916, simple_loss=0.3609, pruned_loss=0.1111, over 28969.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3823, pruned_loss=0.1286, over 5669583.09 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.37, pruned_loss=0.1226, over 5677083.32 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3843, pruned_loss=0.1296, over 5659526.05 frames. ], batch size: 106, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:15:09,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:15:24,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3073, 3.3179, 1.5861, 1.4282], device='cuda:1'), covar=tensor([0.0875, 0.0356, 0.0818, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0524, 0.0351, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 02:15:34,507 INFO [optim.py:369] (1/2) 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 02:15:47,038 INFO [train.py:968] (1/2) Epoch 13, batch 44600, giga_loss[loss=0.2903, simple_loss=0.3719, pruned_loss=0.1044, over 28886.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3824, pruned_loss=0.1263, over 5671924.44 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.37, pruned_loss=0.1225, over 5675889.70 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.384, pruned_loss=0.1272, over 5665094.03 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:16:24,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 02:16:33,519 INFO [train.py:968] (1/2) Epoch 13, batch 44650, giga_loss[loss=0.2974, simple_loss=0.3755, pruned_loss=0.1097, over 28307.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3841, pruned_loss=0.1269, over 5664801.32 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3704, pruned_loss=0.1228, over 5674039.08 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3852, pruned_loss=0.1275, over 5661049.60 frames. ], batch size: 65, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:16:40,538 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:16:43,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:16:50,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:08,553 INFO [optim.py:369] (1/2) 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 02:17:11,103 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:21,200 INFO [train.py:968] (1/2) Epoch 13, batch 44700, giga_loss[loss=0.326, simple_loss=0.3861, pruned_loss=0.133, over 28835.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3853, pruned_loss=0.1286, over 5658022.97 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3712, pruned_loss=0.1234, over 5663222.19 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3859, pruned_loss=0.1286, over 5663829.29 frames. ], batch size: 186, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:17:22,775 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:23,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6030, 1.5996, 1.2021, 1.2802], device='cuda:1'), covar=tensor([0.0916, 0.0669, 0.1134, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0443, 0.0508, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 02:17:26,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:52,939 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:17:59,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 02:18:06,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9503, 3.7612, 3.5656, 1.7395], device='cuda:1'), covar=tensor([0.0654, 0.0806, 0.0743, 0.2234], device='cuda:1'), in_proj_covar=tensor([0.1115, 0.1039, 0.0901, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 02:18:07,188 INFO [train.py:968] (1/2) Epoch 13, batch 44750, giga_loss[loss=0.3542, simple_loss=0.4076, pruned_loss=0.1504, over 28195.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3842, pruned_loss=0.1282, over 5671615.97 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3713, pruned_loss=0.1234, over 5666755.36 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.385, pruned_loss=0.1285, over 5673257.08 frames. ], batch size: 368, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:18:09,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4877, 1.6604, 1.6933, 1.3033], device='cuda:1'), covar=tensor([0.1422, 0.2138, 0.1178, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0701, 0.0888, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 02:18:38,271 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 44800, giga_loss[loss=0.3465, simple_loss=0.3765, pruned_loss=0.1583, over 23241.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.383, pruned_loss=0.129, over 5626504.00 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3715, pruned_loss=0.1236, over 5640303.50 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3838, pruned_loss=0.1291, over 5650276.14 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 8.0
+2023-03-07 02:19:19,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:19:35,094 INFO [train.py:968] (1/2) Epoch 13, batch 44850, giga_loss[loss=0.3247, simple_loss=0.3591, pruned_loss=0.1452, over 23536.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3801, pruned_loss=0.1276, over 5631495.97 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3711, pruned_loss=0.1233, over 5637961.88 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3813, pruned_loss=0.1282, over 5651699.74 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 8.0
+2023-03-07 02:20:10,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:20:12,063 INFO [optim.py:369] (1/2) 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 02:20:16,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5636, 2.2213, 1.6361, 0.8171], device='cuda:1'), covar=tensor([0.5051, 0.2435, 0.3510, 0.5142], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1539, 0.1518, 0.1335], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 02:20:20,937 INFO [train.py:968] (1/2) Epoch 13, batch 44900, giga_loss[loss=0.2659, simple_loss=0.3381, pruned_loss=0.09679, over 29082.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3781, pruned_loss=0.1268, over 5634081.63 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3716, pruned_loss=0.1236, over 5633013.75 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3788, pruned_loss=0.127, over 5654312.92 frames. ], batch size: 128, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:21:07,443 INFO [train.py:968] (1/2) Epoch 13, batch 44950, giga_loss[loss=0.3825, simple_loss=0.4018, pruned_loss=0.1816, over 23607.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3767, pruned_loss=0.1271, over 5640264.99 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3714, pruned_loss=0.1235, over 5636920.96 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3776, pruned_loss=0.1274, over 5652793.07 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:21:14,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5739, 1.5447, 1.2725, 1.1969], device='cuda:1'), covar=tensor([0.0724, 0.0500, 0.0906, 0.1140], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0443, 0.0507, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 02:21:42,124 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 13, batch 45000, giga_loss[loss=0.2452, simple_loss=0.3262, pruned_loss=0.08213, over 29058.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3774, pruned_loss=0.1275, over 5656038.09 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1238, over 5640307.25 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3779, pruned_loss=0.1277, over 5663708.84 frames. ], batch size: 155, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:21:51,573 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 02:22:00,265 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 02:22:36,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:22:45,280 INFO [train.py:968] (1/2) Epoch 13, batch 45050, giga_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1137, over 28847.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3751, pruned_loss=0.1251, over 5631387.92 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1241, over 5625535.12 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3752, pruned_loss=0.125, over 5650438.26 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:23:18,579 INFO [optim.py:369] (1/2) 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 02:23:26,873 INFO [train.py:968] (1/2) Epoch 13, batch 45100, libri_loss[loss=0.2999, simple_loss=0.3735, pruned_loss=0.1132, over 29469.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.372, pruned_loss=0.1219, over 5653566.55 frames. ], libri_tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.124, over 5633034.50 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3721, pruned_loss=0.1219, over 5662691.43 frames. ], batch size: 85, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:23:52,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-07 02:24:16,195 INFO [train.py:968] (1/2) Epoch 13, batch 45150, giga_loss[loss=0.2832, simple_loss=0.355, pruned_loss=0.1057, over 28911.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3708, pruned_loss=0.1213, over 5651083.43 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3721, pruned_loss=0.124, over 5638620.09 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3711, pruned_loss=0.1212, over 5653721.83 frames. ], batch size: 213, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:24:36,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-07 02:24:43,873 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:24:45,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3751, 1.6205, 1.3769, 1.5814], device='cuda:1'), covar=tensor([0.0747, 0.0307, 0.0309, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:24:46,400 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:24:56,853 INFO [train.py:968] (1/2) Epoch 13, batch 45200, giga_loss[loss=0.3174, simple_loss=0.375, pruned_loss=0.1299, over 27948.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3687, pruned_loss=0.1202, over 5638248.23 frames. ], libri_tot_loss[loss=0.31, simple_loss=0.3722, pruned_loss=0.1239, over 5628952.74 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3687, pruned_loss=0.1201, over 5650063.65 frames. ], batch size: 412, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:25:03,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:25:12,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:25:44,841 INFO [train.py:968] (1/2) Epoch 13, batch 45250, giga_loss[loss=0.326, simple_loss=0.384, pruned_loss=0.1339, over 28380.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3665, pruned_loss=0.1197, over 5629972.75 frames. ], libri_tot_loss[loss=0.3095, simple_loss=0.3718, pruned_loss=0.1236, over 5626243.17 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3667, pruned_loss=0.1198, over 5641951.20 frames. ], batch size: 65, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:25:56,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=592726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:26:22,267 INFO [optim.py:369] (1/2) 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 02:26:30,553 INFO [train.py:968] (1/2) Epoch 13, batch 45300, giga_loss[loss=0.2775, simple_loss=0.3631, pruned_loss=0.09597, over 28936.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3676, pruned_loss=0.1198, over 5633789.58 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3717, pruned_loss=0.1235, over 5630238.88 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3677, pruned_loss=0.1199, over 5639572.70 frames. ], batch size: 164, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:26:31,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8923, 1.0178, 0.9537, 0.8596], device='cuda:1'), covar=tensor([0.1410, 0.1623, 0.1101, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.1793, 0.1700, 0.1659, 0.1763], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 02:26:41,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:26:45,703 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:14,247 INFO [train.py:968] (1/2) Epoch 13, batch 45350, giga_loss[loss=0.3245, simple_loss=0.3847, pruned_loss=0.1322, over 28719.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 5623481.74 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3715, pruned_loss=0.1235, over 5615947.42 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3703, pruned_loss=0.1213, over 5640965.60 frames. ], batch size: 284, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:27:14,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4569, 1.7625, 1.5522, 1.5517], device='cuda:1'), covar=tensor([0.0781, 0.0302, 0.0298, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:27:14,661 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:17,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:43,410 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:27:51,620 INFO [optim.py:369] (1/2) 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 02:27:57,567 INFO [train.py:968] (1/2) Epoch 13, batch 45400, giga_loss[loss=0.3254, simple_loss=0.3693, pruned_loss=0.1408, over 23413.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3712, pruned_loss=0.1223, over 5622145.05 frames. ], libri_tot_loss[loss=0.3101, simple_loss=0.3723, pruned_loss=0.1239, over 5626219.06 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1218, over 5627181.26 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:28:00,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=592869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:28:02,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=592872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:28:27,764 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=592901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:28:40,218 INFO [train.py:968] (1/2) Epoch 13, batch 45450, giga_loss[loss=0.2843, simple_loss=0.3512, pruned_loss=0.1087, over 28672.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3701, pruned_loss=0.1215, over 5633004.60 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3715, pruned_loss=0.1232, over 5634997.42 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3704, pruned_loss=0.1216, over 5629261.39 frames. ], batch size: 242, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:28:56,730 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:28:58,480 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 02:29:16,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 02:29:17,322 INFO [optim.py:369] (1/2) 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 02:29:25,122 INFO [train.py:968] (1/2) Epoch 13, batch 45500, giga_loss[loss=0.3183, simple_loss=0.3847, pruned_loss=0.126, over 28991.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3723, pruned_loss=0.1229, over 5637536.64 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3714, pruned_loss=0.1231, over 5633423.32 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1231, over 5635828.40 frames. ], batch size: 136, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:29:37,249 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=592978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:30:12,059 INFO [train.py:968] (1/2) Epoch 13, batch 45550, giga_loss[loss=0.3291, simple_loss=0.3925, pruned_loss=0.1329, over 28875.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3755, pruned_loss=0.1252, over 5652387.56 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3714, pruned_loss=0.123, over 5640291.63 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3758, pruned_loss=0.1255, over 5645041.61 frames. ], batch size: 199, lr: 2.43e-03, grad_scale: 2.0
+2023-03-07 02:30:35,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-07 02:30:45,268 INFO [optim.py:369] (1/2) 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 02:30:52,666 INFO [train.py:968] (1/2) Epoch 13, batch 45600, giga_loss[loss=0.3644, simple_loss=0.4134, pruned_loss=0.1577, over 28516.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.377, pruned_loss=0.1262, over 5664694.37 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3716, pruned_loss=0.1233, over 5652467.09 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3773, pruned_loss=0.1263, over 5648435.40 frames. ], batch size: 336, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:31:37,497 INFO [train.py:968] (1/2) Epoch 13, batch 45650, libri_loss[loss=0.3425, simple_loss=0.3918, pruned_loss=0.1466, over 19084.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3795, pruned_loss=0.1288, over 5656934.42 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1234, over 5647625.99 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3798, pruned_loss=0.1289, over 5649423.63 frames. ], batch size: 186, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:32:00,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3773, 3.3428, 1.4770, 1.5063], device='cuda:1'), covar=tensor([0.0920, 0.0315, 0.0825, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0527, 0.0354, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 02:32:15,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:32:17,979 INFO [optim.py:369] (1/2) 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 02:32:18,841 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 13, batch 45700, giga_loss[loss=0.3511, simple_loss=0.3782, pruned_loss=0.162, over 23739.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3802, pruned_loss=0.1298, over 5618554.79 frames. ], libri_tot_loss[loss=0.3099, simple_loss=0.3722, pruned_loss=0.1238, over 5611775.51 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3801, pruned_loss=0.1296, over 5645739.79 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:33:20,420 INFO [train.py:968] (1/2) Epoch 13, batch 45750, giga_loss[loss=0.331, simple_loss=0.3933, pruned_loss=0.1343, over 28860.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3808, pruned_loss=0.1284, over 5597104.30 frames. ], libri_tot_loss[loss=0.3103, simple_loss=0.3725, pruned_loss=0.1241, over 5578734.42 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3805, pruned_loss=0.128, over 5649346.84 frames. ], batch size: 119, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:33:22,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1652, 1.5047, 1.4728, 1.0743], device='cuda:1'), covar=tensor([0.1546, 0.2329, 0.1315, 0.1531], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0698, 0.0888, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 02:33:56,787 INFO [optim.py:369] (1/2) 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 02:34:05,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7496, 1.8973, 1.6190, 1.8257], device='cuda:1'), covar=tensor([0.1993, 0.1896, 0.1902, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.0996, 0.1201, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 02:34:07,235 INFO [train.py:968] (1/2) Epoch 13, batch 45800, giga_loss[loss=0.3556, simple_loss=0.388, pruned_loss=0.1616, over 24000.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3802, pruned_loss=0.1281, over 5548413.10 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.373, pruned_loss=0.1246, over 5521749.73 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3797, pruned_loss=0.1274, over 5641037.22 frames. ], batch size: 705, lr: 2.43e-03, grad_scale: 4.0
+2023-03-07 02:34:37,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:34:40,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:34:42,172 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-07 02:35:13,679 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:16,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:21,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:37,936 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:40,842 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:35:46,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:36:00,117 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:36:01,623 INFO [train.py:968] (1/2) Epoch 14, batch 50, giga_loss[loss=0.3009, simple_loss=0.3806, pruned_loss=0.1106, over 28871.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3694, pruned_loss=0.1069, over 1269647.66 frames. ], libri_tot_loss[loss=0.2399, simple_loss=0.3219, pruned_loss=0.07898, over 259298.46 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3802, pruned_loss=0.1133, over 1057786.75 frames. ], batch size: 112, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:36:04,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:36:47,191 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:36:50,503 INFO [train.py:968] (1/2) Epoch 14, batch 100, giga_loss[loss=0.2933, simple_loss=0.3608, pruned_loss=0.1129, over 28665.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3662, pruned_loss=0.1069, over 2235352.69 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3289, pruned_loss=0.08438, over 414937.73 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3727, pruned_loss=0.1107, over 1965170.65 frames. ], batch size: 262, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:37:13,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0047, 2.3536, 2.2939, 1.8329], device='cuda:1'), covar=tensor([0.1794, 0.2083, 0.1338, 0.1539], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0697, 0.0890, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 02:37:36,283 INFO [train.py:968] (1/2) Epoch 14, batch 150, giga_loss[loss=0.2447, simple_loss=0.3205, pruned_loss=0.08449, over 28891.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3522, pruned_loss=0.1005, over 3000276.18 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3288, pruned_loss=0.085, over 539846.68 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3561, pruned_loss=0.1029, over 2721003.98 frames. ], batch size: 227, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:37:36,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:37:39,636 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:37:42,834 INFO [optim.py:369] (1/2) 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 02:38:05,543 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:14,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3886, 2.8540, 1.5224, 1.4708], device='cuda:1'), covar=tensor([0.0872, 0.0319, 0.0832, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0523, 0.0352, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 02:38:16,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:18,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:18,858 INFO [train.py:968] (1/2) Epoch 14, batch 200, giga_loss[loss=0.2138, simple_loss=0.2965, pruned_loss=0.06551, over 28591.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3378, pruned_loss=0.09383, over 3596251.00 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3273, pruned_loss=0.08445, over 566486.45 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3396, pruned_loss=0.09508, over 3366548.99 frames. ], batch size: 336, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:38:42,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:38:47,186 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 02:39:02,158 INFO [train.py:968] (1/2) Epoch 14, batch 250, giga_loss[loss=0.2102, simple_loss=0.2838, pruned_loss=0.06832, over 28590.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3272, pruned_loss=0.0887, over 4059629.67 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3311, pruned_loss=0.08655, over 645701.57 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3276, pruned_loss=0.08916, over 3851184.53 frames. ], batch size: 71, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:39:06,573 INFO [optim.py:369] (1/2) 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 02:39:41,359 INFO [train.py:968] (1/2) Epoch 14, batch 300, libri_loss[loss=0.265, simple_loss=0.3464, pruned_loss=0.09178, over 29561.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3194, pruned_loss=0.08513, over 4423188.51 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.331, pruned_loss=0.0865, over 851349.04 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3189, pruned_loss=0.08533, over 4195734.63 frames. ], batch size: 78, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:40:01,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4791, 3.5659, 1.6092, 1.5731], device='cuda:1'), covar=tensor([0.0997, 0.0311, 0.0909, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0523, 0.0353, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 02:40:27,263 INFO [train.py:968] (1/2) Epoch 14, batch 350, giga_loss[loss=0.2254, simple_loss=0.2952, pruned_loss=0.0778, over 29012.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3116, pruned_loss=0.0817, over 4701990.05 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3305, pruned_loss=0.08614, over 926954.46 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3108, pruned_loss=0.08172, over 4504530.06 frames. ], batch size: 227, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 02:40:31,910 INFO [optim.py:369] (1/2) 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 02:41:03,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3032, 0.8270, 0.8987, 1.4633], device='cuda:1'), covar=tensor([0.0774, 0.0369, 0.0358, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:41:05,429 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 400, giga_loss[loss=0.2137, simple_loss=0.2783, pruned_loss=0.07454, over 28622.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3078, pruned_loss=0.07981, over 4929275.23 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3318, pruned_loss=0.08664, over 1049948.62 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3062, pruned_loss=0.07948, over 4747656.57 frames. ], batch size: 85, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:41:28,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:41:34,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=593732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:41:48,825 INFO [train.py:968] (1/2) Epoch 14, batch 450, giga_loss[loss=0.2598, simple_loss=0.3254, pruned_loss=0.09709, over 28903.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3076, pruned_loss=0.08013, over 5105154.28 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3371, pruned_loss=0.08967, over 1170471.45 frames. ], giga_tot_loss[loss=0.2313, simple_loss=0.3045, pruned_loss=0.07904, over 4940086.54 frames. ], batch size: 213, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:41:55,866 INFO [optim.py:369] (1/2) 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 02:42:06,908 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 02:42:07,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=593770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:42:28,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4209, 1.5980, 1.3245, 1.5310], device='cuda:1'), covar=tensor([0.0767, 0.0322, 0.0333, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:42:31,456 INFO [train.py:968] (1/2) Epoch 14, batch 500, giga_loss[loss=0.2168, simple_loss=0.2939, pruned_loss=0.06987, over 28203.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.3046, pruned_loss=0.07856, over 5231787.66 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3372, pruned_loss=0.08975, over 1286698.96 frames. ], giga_tot_loss[loss=0.2281, simple_loss=0.3013, pruned_loss=0.0774, over 5084423.37 frames. ], batch size: 368, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:43:06,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:10,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:14,746 INFO [train.py:968] (1/2) Epoch 14, batch 550, giga_loss[loss=0.1908, simple_loss=0.2709, pruned_loss=0.0554, over 28849.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3023, pruned_loss=0.07758, over 5335334.97 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3381, pruned_loss=0.0901, over 1376524.15 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.2989, pruned_loss=0.07634, over 5208827.44 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:43:19,750 INFO [optim.py:369] (1/2) 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 02:43:30,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5756, 1.6328, 1.2353, 1.3421], device='cuda:1'), covar=tensor([0.0784, 0.0556, 0.0967, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0441, 0.0504, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 02:43:30,868 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:32,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:36,626 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:57,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=593898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:43:58,722 INFO [train.py:968] (1/2) Epoch 14, batch 600, giga_loss[loss=0.2003, simple_loss=0.2763, pruned_loss=0.0621, over 28834.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3013, pruned_loss=0.07675, over 5410315.98 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3396, pruned_loss=0.09058, over 1523323.34 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.2969, pruned_loss=0.07519, over 5299079.27 frames. ], batch size: 119, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:44:08,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5966, 2.5947, 1.8878, 2.2150], device='cuda:1'), covar=tensor([0.0752, 0.0667, 0.1006, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0442, 0.0506, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 02:44:10,882 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=593913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:44:12,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=593916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:44:37,372 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 650, giga_loss[loss=0.2112, simple_loss=0.2887, pruned_loss=0.06689, over 28911.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.2993, pruned_loss=0.07564, over 5478921.79 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3394, pruned_loss=0.09043, over 1652831.31 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2947, pruned_loss=0.07402, over 5377806.11 frames. ], batch size: 174, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:44:49,495 INFO [optim.py:369] (1/2) 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 02:45:18,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5834, 1.8422, 1.4924, 1.7398], device='cuda:1'), covar=tensor([0.2551, 0.2508, 0.2814, 0.2404], device='cuda:1'), in_proj_covar=tensor([0.1357, 0.0991, 0.1201, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 02:45:29,042 INFO [train.py:968] (1/2) Epoch 14, batch 700, giga_loss[loss=0.1688, simple_loss=0.2484, pruned_loss=0.0446, over 28394.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2958, pruned_loss=0.0741, over 5526715.61 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3389, pruned_loss=0.09014, over 1674069.15 frames. ], giga_tot_loss[loss=0.2189, simple_loss=0.2921, pruned_loss=0.07282, over 5445943.38 frames. ], batch size: 60, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:46:07,294 INFO [train.py:968] (1/2) Epoch 14, batch 750, giga_loss[loss=0.1965, simple_loss=0.271, pruned_loss=0.06094, over 28589.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2962, pruned_loss=0.07424, over 5558212.43 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3396, pruned_loss=0.09019, over 1892277.86 frames. ], giga_tot_loss[loss=0.2177, simple_loss=0.2906, pruned_loss=0.07234, over 5481778.71 frames. ], batch size: 307, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:46:15,747 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 800, libri_loss[loss=0.3032, simple_loss=0.3717, pruned_loss=0.1173, over 19576.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.2962, pruned_loss=0.07452, over 5579116.03 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3416, pruned_loss=0.09096, over 2072491.82 frames. ], giga_tot_loss[loss=0.2164, simple_loss=0.2889, pruned_loss=0.07195, over 5517026.15 frames. ], batch size: 187, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:46:54,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:47:35,232 INFO [train.py:968] (1/2) Epoch 14, batch 850, giga_loss[loss=0.2816, simple_loss=0.3508, pruned_loss=0.1062, over 28802.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.3024, pruned_loss=0.0779, over 5603088.23 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3427, pruned_loss=0.09196, over 2184788.35 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.295, pruned_loss=0.07507, over 5546047.59 frames. ], batch size: 284, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:47:41,105 INFO [optim.py:369] (1/2) 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 02:48:15,350 INFO [train.py:968] (1/2) Epoch 14, batch 900, giga_loss[loss=0.3151, simple_loss=0.387, pruned_loss=0.1217, over 28931.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3149, pruned_loss=0.08418, over 5627483.96 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3416, pruned_loss=0.09146, over 2408269.37 frames. ], giga_tot_loss[loss=0.235, simple_loss=0.3072, pruned_loss=0.08141, over 5573207.84 frames. ], batch size: 186, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:49:00,763 INFO [train.py:968] (1/2) Epoch 14, batch 950, giga_loss[loss=0.2816, simple_loss=0.3633, pruned_loss=0.09997, over 28956.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3272, pruned_loss=0.09103, over 5641489.90 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3415, pruned_loss=0.09158, over 2477821.77 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3208, pruned_loss=0.08875, over 5593587.19 frames. ], batch size: 227, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:49:01,043 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=594250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:04,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=594253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:04,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7102, 1.8057, 1.5178, 1.9271], device='cuda:1'), covar=tensor([0.2658, 0.2701, 0.2971, 0.2484], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.0990, 0.1200, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 02:49:07,009 INFO [optim.py:369] (1/2) 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 02:49:21,445 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:29,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=594282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:49:43,198 INFO [train.py:968] (1/2) Epoch 14, batch 1000, giga_loss[loss=0.3015, simple_loss=0.3774, pruned_loss=0.1128, over 28894.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3367, pruned_loss=0.09538, over 5653780.57 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3423, pruned_loss=0.0919, over 2545926.22 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3312, pruned_loss=0.0935, over 5611108.59 frames. ], batch size: 227, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:49:55,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3490, 1.5349, 1.3014, 1.3458], device='cuda:1'), covar=tensor([0.1811, 0.1555, 0.1658, 0.1520], device='cuda:1'), in_proj_covar=tensor([0.1803, 0.1711, 0.1666, 0.1763], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 02:50:02,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=5.05 vs. limit=5.0
+2023-03-07 02:50:21,225 INFO [train.py:968] (1/2) Epoch 14, batch 1050, giga_loss[loss=0.2456, simple_loss=0.3296, pruned_loss=0.08081, over 29075.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3402, pruned_loss=0.09574, over 5663332.43 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3421, pruned_loss=0.09167, over 2656442.83 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3358, pruned_loss=0.09446, over 5625564.33 frames. ], batch size: 155, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:50:27,215 INFO [optim.py:369] (1/2) 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 02:50:29,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4648, 1.6708, 1.7754, 1.3127], device='cuda:1'), covar=tensor([0.1759, 0.2513, 0.1408, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.0858, 0.0700, 0.0900, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 02:51:06,416 INFO [train.py:968] (1/2) Epoch 14, batch 1100, giga_loss[loss=0.2626, simple_loss=0.3429, pruned_loss=0.09116, over 28956.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3431, pruned_loss=0.09616, over 5666186.45 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3434, pruned_loss=0.09243, over 2721334.98 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3391, pruned_loss=0.09492, over 5631571.32 frames. ], batch size: 155, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:51:49,896 INFO [train.py:968] (1/2) Epoch 14, batch 1150, giga_loss[loss=0.3048, simple_loss=0.3769, pruned_loss=0.1163, over 27625.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3452, pruned_loss=0.09733, over 5669179.69 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3437, pruned_loss=0.09271, over 2752756.14 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.342, pruned_loss=0.09627, over 5640088.82 frames. ], batch size: 472, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:51:57,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-07 02:51:57,769 INFO [optim.py:369] (1/2) 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 02:52:34,570 INFO [train.py:968] (1/2) Epoch 14, batch 1200, giga_loss[loss=0.2779, simple_loss=0.3553, pruned_loss=0.1002, over 28795.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3465, pruned_loss=0.09836, over 5675081.09 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3449, pruned_loss=0.093, over 2844978.43 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3434, pruned_loss=0.09754, over 5646127.00 frames. ], batch size: 119, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:52:59,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3011, 1.3162, 3.4207, 3.1617], device='cuda:1'), covar=tensor([0.1341, 0.2570, 0.0430, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0679, 0.0600, 0.0877, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 02:53:16,892 INFO [train.py:968] (1/2) Epoch 14, batch 1250, giga_loss[loss=0.2611, simple_loss=0.3427, pruned_loss=0.08971, over 28936.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.35, pruned_loss=0.1006, over 5683719.37 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3452, pruned_loss=0.09311, over 2935093.32 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3474, pruned_loss=0.1001, over 5654985.13 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:53:24,763 INFO [optim.py:369] (1/2) 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 02:53:25,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:54:00,398 INFO [train.py:968] (1/2) Epoch 14, batch 1300, giga_loss[loss=0.2741, simple_loss=0.356, pruned_loss=0.09615, over 28663.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3526, pruned_loss=0.1015, over 5682674.89 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3452, pruned_loss=0.09314, over 2980792.54 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3507, pruned_loss=0.1012, over 5668487.62 frames. ], batch size: 92, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:54:11,522 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:54:37,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:54:38,544 INFO [train.py:968] (1/2) Epoch 14, batch 1350, libri_loss[loss=0.2186, simple_loss=0.3051, pruned_loss=0.06607, over 29503.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3551, pruned_loss=0.1027, over 5685925.50 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3454, pruned_loss=0.09339, over 3080183.58 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3538, pruned_loss=0.1027, over 5668405.76 frames. ], batch size: 70, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:54:43,691 INFO [optim.py:369] (1/2) 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 02:54:50,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 02:55:18,687 INFO [train.py:968] (1/2) Epoch 14, batch 1400, giga_loss[loss=0.2573, simple_loss=0.3433, pruned_loss=0.08571, over 28642.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3565, pruned_loss=0.1029, over 5689633.97 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.345, pruned_loss=0.09309, over 3122369.95 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3558, pruned_loss=0.1031, over 5672905.89 frames. ], batch size: 85, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:55:40,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4989, 1.5788, 1.2412, 1.1672], device='cuda:1'), covar=tensor([0.0948, 0.0597, 0.1080, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0444, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 02:55:45,024 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:55:50,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5869, 1.6269, 1.9036, 1.4387], device='cuda:1'), covar=tensor([0.1547, 0.1910, 0.1225, 0.1514], device='cuda:1'), in_proj_covar=tensor([0.0855, 0.0697, 0.0897, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 02:55:59,064 INFO [train.py:968] (1/2) Epoch 14, batch 1450, giga_loss[loss=0.3004, simple_loss=0.3709, pruned_loss=0.115, over 27609.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3564, pruned_loss=0.1016, over 5693837.80 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3449, pruned_loss=0.09309, over 3212231.40 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3562, pruned_loss=0.1021, over 5679063.62 frames. ], batch size: 472, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:56:05,410 INFO [optim.py:369] (1/2) 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 02:56:21,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2932, 1.8879, 1.5048, 1.5665], device='cuda:1'), covar=tensor([0.0799, 0.0299, 0.0310, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:56:32,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=594791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:56:35,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=594794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:56:38,836 INFO [train.py:968] (1/2) Epoch 14, batch 1500, libri_loss[loss=0.3039, simple_loss=0.3751, pruned_loss=0.1163, over 19684.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3554, pruned_loss=0.1001, over 5692603.87 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.346, pruned_loss=0.09378, over 3269393.79 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3549, pruned_loss=0.1003, over 5685488.29 frames. ], batch size: 187, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:56:56,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=594823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:56:56,619 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=594823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:57:11,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 02:57:15,739 INFO [train.py:968] (1/2) Epoch 14, batch 1550, giga_loss[loss=0.2595, simple_loss=0.3532, pruned_loss=0.08291, over 28932.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3525, pruned_loss=0.09731, over 5706690.44 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3448, pruned_loss=0.09296, over 3370394.03 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3531, pruned_loss=0.0981, over 5695911.64 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 02:57:19,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3513, 1.7497, 1.4709, 1.5137], device='cuda:1'), covar=tensor([0.0814, 0.0307, 0.0315, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 02:57:23,599 INFO [optim.py:369] (1/2) 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 02:57:42,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5713, 1.6634, 1.8773, 1.4339], device='cuda:1'), covar=tensor([0.1389, 0.1845, 0.1125, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.0861, 0.0701, 0.0903, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 02:57:55,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 02:57:58,098 INFO [train.py:968] (1/2) Epoch 14, batch 1600, giga_loss[loss=0.2637, simple_loss=0.341, pruned_loss=0.0932, over 28783.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3528, pruned_loss=0.09783, over 5702318.35 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3453, pruned_loss=0.0932, over 3431202.01 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3532, pruned_loss=0.09844, over 5690952.21 frames. ], batch size: 284, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:58:27,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:58:42,579 INFO [train.py:968] (1/2) Epoch 14, batch 1650, giga_loss[loss=0.3513, simple_loss=0.4066, pruned_loss=0.1481, over 28264.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3548, pruned_loss=0.1009, over 5701728.68 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3459, pruned_loss=0.09343, over 3468118.91 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3549, pruned_loss=0.1014, over 5689990.27 frames. ], batch size: 368, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 02:58:52,850 INFO [optim.py:369] (1/2) 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 02:59:19,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=594989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 02:59:28,737 INFO [train.py:968] (1/2) Epoch 14, batch 1700, giga_loss[loss=0.2608, simple_loss=0.3304, pruned_loss=0.09558, over 28891.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3567, pruned_loss=0.1043, over 5709870.98 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3459, pruned_loss=0.09347, over 3516541.61 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.357, pruned_loss=0.1048, over 5696879.95 frames. ], batch size: 99, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:00:09,897 INFO [train.py:968] (1/2) Epoch 14, batch 1750, libri_loss[loss=0.2408, simple_loss=0.3235, pruned_loss=0.0791, over 29578.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3559, pruned_loss=0.1048, over 5711203.81 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3457, pruned_loss=0.09302, over 3585253.75 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3565, pruned_loss=0.1057, over 5697321.72 frames. ], batch size: 74, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:00:12,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8090, 3.1968, 2.1826, 0.8240], device='cuda:1'), covar=tensor([0.5994, 0.2408, 0.3155, 0.5833], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1518, 0.1509, 0.1322], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 03:00:17,848 INFO [optim.py:369] (1/2) 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 03:00:24,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:32,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:34,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:51,826 INFO [train.py:968] (1/2) Epoch 14, batch 1800, giga_loss[loss=0.2943, simple_loss=0.3613, pruned_loss=0.1136, over 28638.00 frames. ], tot_loss[loss=0.279, simple_loss=0.352, pruned_loss=0.103, over 5691537.37 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3448, pruned_loss=0.09269, over 3633123.75 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3532, pruned_loss=0.1042, over 5685326.60 frames. ], batch size: 307, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:00:57,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:57,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:00:59,776 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6528, 1.7167, 1.9403, 1.4667], device='cuda:1'), covar=tensor([0.1790, 0.2303, 0.1365, 0.1654], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0697, 0.0896, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 03:01:16,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:01:18,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:01:25,655 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:01:29,390 INFO [train.py:968] (1/2) Epoch 14, batch 1850, giga_loss[loss=0.2819, simple_loss=0.3502, pruned_loss=0.1068, over 27561.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3518, pruned_loss=0.103, over 5695107.29 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3449, pruned_loss=0.09252, over 3739786.30 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.353, pruned_loss=0.1046, over 5684285.57 frames. ], batch size: 472, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 03:01:37,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6875, 1.0153, 2.9717, 2.8003], device='cuda:1'), covar=tensor([0.1756, 0.2519, 0.0554, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0683, 0.0604, 0.0883, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:01:38,366 INFO [optim.py:369] (1/2) 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 03:01:40,631 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:02:08,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:02:09,523 INFO [train.py:968] (1/2) Epoch 14, batch 1900, giga_loss[loss=0.2571, simple_loss=0.3401, pruned_loss=0.08706, over 29055.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3505, pruned_loss=0.1017, over 5674828.56 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3449, pruned_loss=0.09275, over 3784761.15 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3516, pruned_loss=0.1031, over 5677865.87 frames. ], batch size: 136, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 03:02:38,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1427, 3.1956, 1.9839, 0.9930], device='cuda:1'), covar=tensor([0.6369, 0.2344, 0.3476, 0.5751], device='cuda:1'), in_proj_covar=tensor([0.1599, 0.1510, 0.1499, 0.1313], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 03:02:52,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:02:58,758 INFO [train.py:968] (1/2) Epoch 14, batch 1950, giga_loss[loss=0.2261, simple_loss=0.3121, pruned_loss=0.07, over 28771.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3479, pruned_loss=0.09972, over 5674023.80 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3446, pruned_loss=0.0925, over 3812308.02 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3489, pruned_loss=0.101, over 5677293.43 frames. ], batch size: 284, lr: 2.34e-03, grad_scale: 2.0
+2023-03-07 03:03:00,713 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:03:02,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:03:08,492 INFO [optim.py:369] (1/2) 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 03:03:09,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6278, 0.9534, 2.8822, 2.6544], device='cuda:1'), covar=tensor([0.1783, 0.2531, 0.0560, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0680, 0.0603, 0.0879, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:03:27,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:03:27,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-07 03:03:41,513 INFO [train.py:968] (1/2) Epoch 14, batch 2000, giga_loss[loss=0.246, simple_loss=0.3241, pruned_loss=0.084, over 28957.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3429, pruned_loss=0.0969, over 5668892.24 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3446, pruned_loss=0.09229, over 3863656.40 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3438, pruned_loss=0.09824, over 5675315.91 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:04:21,021 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:04:23,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:04:28,523 INFO [train.py:968] (1/2) Epoch 14, batch 2050, giga_loss[loss=0.2636, simple_loss=0.3365, pruned_loss=0.09535, over 28836.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3372, pruned_loss=0.09423, over 5659912.57 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3442, pruned_loss=0.09211, over 3894109.04 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3381, pruned_loss=0.09544, over 5662183.95 frames. ], batch size: 174, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:04:38,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2529, 0.9412, 0.9600, 1.3373], device='cuda:1'), covar=tensor([0.0774, 0.0349, 0.0334, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0113, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 03:04:39,279 INFO [optim.py:369] (1/2) 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 03:04:48,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:05:14,197 INFO [train.py:968] (1/2) Epoch 14, batch 2100, giga_loss[loss=0.2459, simple_loss=0.3217, pruned_loss=0.085, over 27884.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3328, pruned_loss=0.09188, over 5660480.92 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3443, pruned_loss=0.09191, over 3962279.57 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3331, pruned_loss=0.09301, over 5655835.84 frames. ], batch size: 412, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:05:41,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5584, 1.6560, 1.3805, 1.8391], device='cuda:1'), covar=tensor([0.2341, 0.2415, 0.2490, 0.2259], device='cuda:1'), in_proj_covar=tensor([0.1355, 0.0994, 0.1197, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:05:48,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5533, 1.6964, 1.4134, 1.6800], device='cuda:1'), covar=tensor([0.2408, 0.2579, 0.2742, 0.2275], device='cuda:1'), in_proj_covar=tensor([0.1357, 0.0996, 0.1199, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:05:49,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:05:57,078 INFO [train.py:968] (1/2) Epoch 14, batch 2150, giga_loss[loss=0.2581, simple_loss=0.3324, pruned_loss=0.09192, over 29076.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3334, pruned_loss=0.09203, over 5663694.67 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3443, pruned_loss=0.09188, over 3971976.18 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3336, pruned_loss=0.09293, over 5658922.25 frames. ], batch size: 128, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:06:06,320 INFO [optim.py:369] (1/2) 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 03:06:38,063 INFO [train.py:968] (1/2) Epoch 14, batch 2200, giga_loss[loss=0.2266, simple_loss=0.3029, pruned_loss=0.07519, over 28921.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3339, pruned_loss=0.09214, over 5680177.83 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3444, pruned_loss=0.09202, over 3991204.99 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3338, pruned_loss=0.09276, over 5674365.89 frames. ], batch size: 112, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:06:54,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=595519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:07:00,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0771, 2.4041, 2.3753, 1.8707], device='cuda:1'), covar=tensor([0.1740, 0.1881, 0.1284, 0.1487], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0697, 0.0897, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 03:07:17,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.32 vs. limit=5.0
+2023-03-07 03:07:20,483 INFO [train.py:968] (1/2) Epoch 14, batch 2250, giga_loss[loss=0.219, simple_loss=0.3013, pruned_loss=0.06832, over 28866.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3315, pruned_loss=0.09092, over 5686055.92 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3447, pruned_loss=0.09201, over 4018019.34 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3312, pruned_loss=0.09141, over 5680166.90 frames. ], batch size: 174, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:07:29,950 INFO [optim.py:369] (1/2) 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 03:07:48,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:07:51,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:08:00,553 INFO [train.py:968] (1/2) Epoch 14, batch 2300, libri_loss[loss=0.2557, simple_loss=0.3435, pruned_loss=0.08393, over 29552.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3294, pruned_loss=0.08969, over 5694847.03 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3452, pruned_loss=0.09199, over 4054432.13 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3285, pruned_loss=0.09006, over 5687176.25 frames. ], batch size: 78, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:08:16,690 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:08:19,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 2350, giga_loss[loss=0.2311, simple_loss=0.3074, pruned_loss=0.07742, over 28734.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3262, pruned_loss=0.08828, over 5698991.24 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3452, pruned_loss=0.09199, over 4054432.13 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3255, pruned_loss=0.08857, over 5693020.97 frames. ], batch size: 262, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:08:53,017 INFO [optim.py:369] (1/2) 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 03:08:54,145 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:08:56,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:09:22,070 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:09:25,746 INFO [train.py:968] (1/2) Epoch 14, batch 2400, giga_loss[loss=0.2304, simple_loss=0.3027, pruned_loss=0.0791, over 28902.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3252, pruned_loss=0.08804, over 5694730.88 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3463, pruned_loss=0.09239, over 4099550.26 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3234, pruned_loss=0.08791, over 5692487.24 frames. ], batch size: 186, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 03:09:32,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6011, 5.4292, 5.1248, 2.2644], device='cuda:1'), covar=tensor([0.0352, 0.0470, 0.0501, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.0993, 0.0870, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 03:09:55,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3814, 1.7983, 1.2647, 0.8329], device='cuda:1'), covar=tensor([0.4925, 0.2320, 0.2313, 0.4456], device='cuda:1'), in_proj_covar=tensor([0.1596, 0.1493, 0.1488, 0.1305], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 03:10:03,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5862, 1.6919, 1.5916, 1.4900], device='cuda:1'), covar=tensor([0.2886, 0.2348, 0.1767, 0.2404], device='cuda:1'), in_proj_covar=tensor([0.1771, 0.1671, 0.1640, 0.1746], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 03:10:04,844 INFO [train.py:968] (1/2) Epoch 14, batch 2450, giga_loss[loss=0.2437, simple_loss=0.3158, pruned_loss=0.08581, over 28900.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3224, pruned_loss=0.08704, over 5699011.46 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3464, pruned_loss=0.09244, over 4116575.69 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3208, pruned_loss=0.08686, over 5696245.61 frames. ], batch size: 145, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 03:10:05,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1580, 1.7483, 1.4133, 0.3141], device='cuda:1'), covar=tensor([0.3956, 0.2280, 0.4036, 0.5261], device='cuda:1'), in_proj_covar=tensor([0.1596, 0.1493, 0.1489, 0.1305], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 03:10:12,585 INFO [optim.py:369] (1/2) 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 03:10:14,184 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=595763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:10:16,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=595766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:10:19,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 03:10:37,065 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=595795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:10:37,277 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-07 03:10:40,232 INFO [train.py:968] (1/2) Epoch 14, batch 2500, giga_loss[loss=0.1999, simple_loss=0.2789, pruned_loss=0.06042, over 28760.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3213, pruned_loss=0.08633, over 5710052.56 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3473, pruned_loss=0.09261, over 4168019.13 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3187, pruned_loss=0.08592, over 5703875.11 frames. ], batch size: 60, lr: 2.34e-03, grad_scale: 8.0
+2023-03-07 03:11:19,194 INFO [train.py:968] (1/2) Epoch 14, batch 2550, giga_loss[loss=0.3627, simple_loss=0.396, pruned_loss=0.1647, over 26669.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3209, pruned_loss=0.0864, over 5713076.53 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3481, pruned_loss=0.09305, over 4217036.57 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3175, pruned_loss=0.08555, over 5711576.84 frames. ], batch size: 555, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:11:22,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3517, 1.5932, 1.3324, 1.0640], device='cuda:1'), covar=tensor([0.2423, 0.2470, 0.2743, 0.2177], device='cuda:1'), in_proj_covar=tensor([0.1361, 0.0993, 0.1200, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:11:27,480 INFO [optim.py:369] (1/2) 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 03:11:56,534 INFO [train.py:968] (1/2) Epoch 14, batch 2600, giga_loss[loss=0.2492, simple_loss=0.3161, pruned_loss=0.09116, over 28422.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3205, pruned_loss=0.08603, over 5708537.79 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3491, pruned_loss=0.09339, over 4254040.04 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3164, pruned_loss=0.08495, over 5713278.21 frames. ], batch size: 78, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:12:08,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-07 03:12:34,175 INFO [train.py:968] (1/2) Epoch 14, batch 2650, giga_loss[loss=0.2283, simple_loss=0.3026, pruned_loss=0.07704, over 28455.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3211, pruned_loss=0.08619, over 5710183.21 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3497, pruned_loss=0.09344, over 4315681.21 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3161, pruned_loss=0.08495, over 5715824.07 frames. ], batch size: 71, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:12:34,399 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=595950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:12:43,022 INFO [optim.py:369] (1/2) 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 03:13:15,013 INFO [train.py:968] (1/2) Epoch 14, batch 2700, giga_loss[loss=0.2771, simple_loss=0.3467, pruned_loss=0.1038, over 28905.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3234, pruned_loss=0.08777, over 5713166.80 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3505, pruned_loss=0.09371, over 4342789.76 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3184, pruned_loss=0.0865, over 5718139.02 frames. ], batch size: 106, lr: 2.34e-03, grad_scale: 4.0
+2023-03-07 03:13:52,740 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:14:00,097 INFO [train.py:968] (1/2) Epoch 14, batch 2750, giga_loss[loss=0.2656, simple_loss=0.3388, pruned_loss=0.09622, over 28851.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3287, pruned_loss=0.09117, over 5710034.46 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3504, pruned_loss=0.09354, over 4365864.94 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3245, pruned_loss=0.09023, over 5711499.54 frames. ], batch size: 106, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:14:11,281 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 2800, giga_loss[loss=0.3244, simple_loss=0.3858, pruned_loss=0.1315, over 27954.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3356, pruned_loss=0.09568, over 5708898.02 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3506, pruned_loss=0.09366, over 4371658.62 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3322, pruned_loss=0.09486, over 5710961.05 frames. ], batch size: 412, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:14:53,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:15:33,693 INFO [train.py:968] (1/2) Epoch 14, batch 2850, giga_loss[loss=0.2646, simple_loss=0.3488, pruned_loss=0.09019, over 28930.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.343, pruned_loss=0.1003, over 5694653.37 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3505, pruned_loss=0.09359, over 4413426.17 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3399, pruned_loss=0.09979, over 5693936.34 frames. ], batch size: 227, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:15:43,389 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:16:23,339 INFO [train.py:968] (1/2) Epoch 14, batch 2900, giga_loss[loss=0.2507, simple_loss=0.3406, pruned_loss=0.08041, over 28981.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3468, pruned_loss=0.1015, over 5704755.94 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3505, pruned_loss=0.09376, over 4421012.04 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3444, pruned_loss=0.101, over 5703075.38 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:16:57,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:17:03,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-07 03:17:07,858 INFO [train.py:968] (1/2) Epoch 14, batch 2950, giga_loss[loss=0.3055, simple_loss=0.3789, pruned_loss=0.1161, over 29035.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3522, pruned_loss=0.104, over 5701205.44 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3506, pruned_loss=0.09379, over 4447593.74 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3501, pruned_loss=0.1037, over 5698770.98 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:17:21,601 INFO [optim.py:369] (1/2) 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 03:17:25,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3127, 3.0925, 2.9435, 1.3688], device='cuda:1'), covar=tensor([0.0864, 0.1068, 0.0897, 0.2462], device='cuda:1'), in_proj_covar=tensor([0.1072, 0.0996, 0.0867, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 03:17:57,053 INFO [train.py:968] (1/2) Epoch 14, batch 3000, giga_loss[loss=0.3004, simple_loss=0.3706, pruned_loss=0.1151, over 28839.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3586, pruned_loss=0.1085, over 5681103.58 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3504, pruned_loss=0.0936, over 4462163.41 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3573, pruned_loss=0.1086, over 5677188.13 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:17:57,053 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 03:18:01,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2420, 1.6291, 1.5496, 1.1119], device='cuda:1'), covar=tensor([0.1949, 0.2660, 0.1644, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.0853, 0.0697, 0.0896, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 03:18:05,436 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 03:18:27,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:18:48,560 INFO [train.py:968] (1/2) Epoch 14, batch 3050, giga_loss[loss=0.2908, simple_loss=0.3485, pruned_loss=0.1166, over 26821.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3561, pruned_loss=0.1062, over 5689631.00 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3504, pruned_loss=0.09386, over 4483864.79 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3552, pruned_loss=0.1063, over 5683441.59 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:19:00,452 INFO [optim.py:369] (1/2) 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 03:19:17,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0797, 3.8810, 3.6447, 1.6725], device='cuda:1'), covar=tensor([0.0610, 0.0709, 0.0693, 0.2308], device='cuda:1'), in_proj_covar=tensor([0.1078, 0.0999, 0.0869, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 03:19:32,978 INFO [train.py:968] (1/2) Epoch 14, batch 3100, giga_loss[loss=0.2958, simple_loss=0.3671, pruned_loss=0.1122, over 28877.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3528, pruned_loss=0.103, over 5696468.99 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3505, pruned_loss=0.09398, over 4497839.99 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.352, pruned_loss=0.1031, over 5689835.81 frames. ], batch size: 199, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:19:45,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:14,519 INFO [train.py:968] (1/2) Epoch 14, batch 3150, giga_loss[loss=0.2695, simple_loss=0.348, pruned_loss=0.09543, over 28998.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3513, pruned_loss=0.1016, over 5708787.90 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3512, pruned_loss=0.09461, over 4537762.92 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3502, pruned_loss=0.1014, over 5699809.79 frames. ], batch size: 145, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:20:23,833 INFO [optim.py:369] (1/2) 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 03:20:29,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:31,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:40,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:20:54,290 INFO [train.py:968] (1/2) Epoch 14, batch 3200, giga_loss[loss=0.2214, simple_loss=0.3086, pruned_loss=0.06704, over 28540.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3517, pruned_loss=0.1011, over 5713313.03 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3507, pruned_loss=0.09434, over 4590379.64 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3512, pruned_loss=0.1014, over 5700014.27 frames. ], batch size: 60, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:20:54,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:21:22,127 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 03:21:26,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:21:35,726 INFO [train.py:968] (1/2) Epoch 14, batch 3250, giga_loss[loss=0.336, simple_loss=0.39, pruned_loss=0.141, over 27653.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3544, pruned_loss=0.1029, over 5707694.50 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3506, pruned_loss=0.09443, over 4600613.81 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3541, pruned_loss=0.1032, over 5702396.34 frames. ], batch size: 472, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:21:43,959 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 03:21:44,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:21:46,138 INFO [optim.py:369] (1/2) 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 03:21:46,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:08,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-07 03:22:09,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:15,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5302, 1.5318, 1.1983, 1.1457], device='cuda:1'), covar=tensor([0.0747, 0.0518, 0.1007, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0440, 0.0508, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:22:15,688 INFO [train.py:968] (1/2) Epoch 14, batch 3300, giga_loss[loss=0.302, simple_loss=0.3714, pruned_loss=0.1163, over 28638.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3548, pruned_loss=0.1033, over 5709501.89 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3502, pruned_loss=0.09424, over 4649580.68 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.355, pruned_loss=0.104, over 5700121.59 frames. ], batch size: 336, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:22:26,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=596612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:40,552 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:22:42,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:00,282 INFO [train.py:968] (1/2) Epoch 14, batch 3350, giga_loss[loss=0.2967, simple_loss=0.3648, pruned_loss=0.1143, over 28812.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3553, pruned_loss=0.1042, over 5700013.34 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3502, pruned_loss=0.0943, over 4652706.34 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3555, pruned_loss=0.1048, over 5699996.67 frames. ], batch size: 199, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:23:07,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:11,584 INFO [optim.py:369] (1/2) 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 03:23:28,937 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:31,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=596684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:23:45,003 INFO [train.py:968] (1/2) Epoch 14, batch 3400, giga_loss[loss=0.3039, simple_loss=0.3675, pruned_loss=0.1201, over 28558.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3553, pruned_loss=0.1042, over 5708035.90 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3501, pruned_loss=0.09405, over 4675362.65 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3557, pruned_loss=0.105, over 5706432.63 frames. ], batch size: 307, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:23:56,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:24:25,816 INFO [train.py:968] (1/2) Epoch 14, batch 3450, libri_loss[loss=0.2885, simple_loss=0.3681, pruned_loss=0.1045, over 29536.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3553, pruned_loss=0.104, over 5714582.77 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3499, pruned_loss=0.09392, over 4695942.55 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3558, pruned_loss=0.105, over 5713375.75 frames. ], batch size: 82, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:24:29,349 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=596755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:24:29,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-07 03:24:31,568 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:24:38,036 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:24:54,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=596787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:25:05,360 INFO [train.py:968] (1/2) Epoch 14, batch 3500, giga_loss[loss=0.4003, simple_loss=0.4525, pruned_loss=0.1741, over 28891.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3563, pruned_loss=0.1038, over 5715989.50 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3507, pruned_loss=0.09406, over 4736190.08 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3563, pruned_loss=0.1048, over 5710249.09 frames. ], batch size: 199, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:25:44,904 INFO [train.py:968] (1/2) Epoch 14, batch 3550, giga_loss[loss=0.2517, simple_loss=0.3291, pruned_loss=0.08716, over 28628.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.356, pruned_loss=0.1024, over 5717319.64 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3509, pruned_loss=0.09414, over 4757530.34 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.356, pruned_loss=0.1033, over 5711094.44 frames. ], batch size: 85, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:25:56,275 INFO [optim.py:369] (1/2) 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 03:26:28,681 INFO [train.py:968] (1/2) Epoch 14, batch 3600, giga_loss[loss=0.2578, simple_loss=0.3155, pruned_loss=0.1001, over 23534.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3563, pruned_loss=0.1024, over 5717215.39 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3506, pruned_loss=0.09396, over 4773834.63 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3564, pruned_loss=0.1033, over 5710623.04 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:27:05,823 INFO [train.py:968] (1/2) Epoch 14, batch 3650, giga_loss[loss=0.2584, simple_loss=0.3317, pruned_loss=0.09253, over 28495.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3538, pruned_loss=0.101, over 5722257.26 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3496, pruned_loss=0.09355, over 4798078.45 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3547, pruned_loss=0.1022, over 5716367.34 frames. ], batch size: 71, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:27:19,044 INFO [optim.py:369] (1/2) 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 03:27:40,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=596989.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:27:50,708 INFO [train.py:968] (1/2) Epoch 14, batch 3700, giga_loss[loss=0.262, simple_loss=0.3338, pruned_loss=0.09515, over 28686.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3527, pruned_loss=0.1013, over 5709320.39 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3495, pruned_loss=0.09346, over 4798623.15 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3535, pruned_loss=0.1024, over 5713248.84 frames. ], batch size: 99, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:28:27,650 INFO [train.py:968] (1/2) Epoch 14, batch 3750, giga_loss[loss=0.2652, simple_loss=0.3424, pruned_loss=0.094, over 28723.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3502, pruned_loss=0.09975, over 5723287.82 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3488, pruned_loss=0.09318, over 4857990.95 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3515, pruned_loss=0.1012, over 5716697.36 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:28:37,927 INFO [optim.py:369] (1/2) 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 03:29:11,414 INFO [train.py:968] (1/2) Epoch 14, batch 3800, giga_loss[loss=0.257, simple_loss=0.3313, pruned_loss=0.09133, over 28626.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3502, pruned_loss=0.09964, over 5730409.88 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3482, pruned_loss=0.09279, over 4884692.36 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3518, pruned_loss=0.1013, over 5724379.59 frames. ], batch size: 78, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:29:34,004 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3171, 1.5938, 1.5919, 1.1805], device='cuda:1'), covar=tensor([0.1634, 0.2559, 0.1394, 0.1687], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0698, 0.0896, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 03:29:44,346 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:29:44,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:29:50,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7910, 1.9392, 1.6144, 2.0154], device='cuda:1'), covar=tensor([0.2306, 0.2298, 0.2488, 0.2233], device='cuda:1'), in_proj_covar=tensor([0.1351, 0.0991, 0.1191, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:29:51,430 INFO [train.py:968] (1/2) Epoch 14, batch 3850, giga_loss[loss=0.2664, simple_loss=0.3525, pruned_loss=0.09015, over 28971.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3507, pruned_loss=0.1001, over 5733822.62 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3477, pruned_loss=0.09261, over 4915208.60 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3525, pruned_loss=0.1018, over 5723974.99 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:30:00,876 INFO [optim.py:369] (1/2) 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 03:30:27,794 INFO [train.py:968] (1/2) Epoch 14, batch 3900, giga_loss[loss=0.2578, simple_loss=0.3403, pruned_loss=0.08762, over 28590.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3501, pruned_loss=0.09956, over 5713499.82 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3469, pruned_loss=0.09232, over 4936971.81 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3523, pruned_loss=0.1014, over 5718308.09 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:30:46,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 03:31:07,166 INFO [train.py:968] (1/2) Epoch 14, batch 3950, giga_loss[loss=0.3126, simple_loss=0.3944, pruned_loss=0.1154, over 28965.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3497, pruned_loss=0.09817, over 5716621.77 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.347, pruned_loss=0.09235, over 4981506.74 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3515, pruned_loss=0.09999, over 5714680.09 frames. ], batch size: 145, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:31:17,990 INFO [optim.py:369] (1/2) 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 03:31:33,417 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=597282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:31:35,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=597285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:31:46,303 INFO [train.py:968] (1/2) Epoch 14, batch 4000, giga_loss[loss=0.2856, simple_loss=0.358, pruned_loss=0.1066, over 29088.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3495, pruned_loss=0.09865, over 5712548.50 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3474, pruned_loss=0.09282, over 4985716.90 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3507, pruned_loss=0.09978, over 5717789.01 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:31:53,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7836, 1.9194, 1.3349, 1.4581], device='cuda:1'), covar=tensor([0.0848, 0.0653, 0.1012, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0437, 0.0505, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:31:58,600 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=597314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:32:00,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-07 03:32:02,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5248, 3.4630, 1.6563, 1.5799], device='cuda:1'), covar=tensor([0.0909, 0.0248, 0.0822, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0509, 0.0347, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 03:32:10,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1793, 3.9695, 3.7715, 1.7491], device='cuda:1'), covar=tensor([0.0578, 0.0777, 0.0755, 0.2201], device='cuda:1'), in_proj_covar=tensor([0.1081, 0.1004, 0.0876, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 03:32:27,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 03:32:27,259 INFO [train.py:968] (1/2) Epoch 14, batch 4050, giga_loss[loss=0.252, simple_loss=0.3311, pruned_loss=0.08647, over 28969.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.09773, over 5705376.99 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3474, pruned_loss=0.09275, over 4997657.55 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.348, pruned_loss=0.09876, over 5708562.84 frames. ], batch size: 136, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:32:34,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6863, 1.9414, 1.8910, 1.6545], device='cuda:1'), covar=tensor([0.1491, 0.1626, 0.1708, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0730, 0.0683, 0.0658], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 03:32:38,313 INFO [optim.py:369] (1/2) 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 03:32:38,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597364.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:33:06,300 INFO [train.py:968] (1/2) Epoch 14, batch 4100, giga_loss[loss=0.2547, simple_loss=0.3317, pruned_loss=0.08884, over 28773.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.345, pruned_loss=0.09672, over 5708433.88 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3475, pruned_loss=0.09284, over 5016363.21 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3456, pruned_loss=0.09757, over 5709850.90 frames. ], batch size: 242, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:33:45,592 INFO [train.py:968] (1/2) Epoch 14, batch 4150, libri_loss[loss=0.2483, simple_loss=0.323, pruned_loss=0.08677, over 29640.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3437, pruned_loss=0.09658, over 5703455.57 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3475, pruned_loss=0.09278, over 5034347.51 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3441, pruned_loss=0.09737, over 5700649.99 frames. ], batch size: 73, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:33:55,746 INFO [optim.py:369] (1/2) 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 03:34:11,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5171, 3.5917, 1.6429, 1.6099], device='cuda:1'), covar=tensor([0.0883, 0.0355, 0.0849, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0512, 0.0348, 0.0394], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 03:34:24,701 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597498.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:34:25,672 INFO [train.py:968] (1/2) Epoch 14, batch 4200, giga_loss[loss=0.2408, simple_loss=0.3207, pruned_loss=0.08047, over 29028.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3437, pruned_loss=0.09738, over 5704501.62 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3478, pruned_loss=0.09295, over 5042774.58 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3439, pruned_loss=0.09791, over 5700891.68 frames. ], batch size: 128, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:34:30,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=597507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:34:32,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0229, 1.9789, 1.7318, 1.6659], device='cuda:1'), covar=tensor([0.1591, 0.2416, 0.2228, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0735, 0.0688, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 03:34:32,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=597510.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:34:33,369 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:34:36,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:34:46,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 03:34:56,288 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=597539.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:35:04,087 INFO [train.py:968] (1/2) Epoch 14, batch 4250, libri_loss[loss=0.2695, simple_loss=0.3561, pruned_loss=0.0914, over 29659.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3414, pruned_loss=0.09644, over 5704929.36 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09302, over 5061681.75 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3413, pruned_loss=0.09689, over 5699642.98 frames. ], batch size: 91, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:35:14,071 INFO [optim.py:369] (1/2) 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 03:35:23,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4344, 1.4645, 1.2012, 1.1394], device='cuda:1'), covar=tensor([0.0784, 0.0564, 0.0999, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0437, 0.0504, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:35:35,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2393, 3.7946, 1.4884, 1.4517], device='cuda:1'), covar=tensor([0.0957, 0.0250, 0.0902, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0512, 0.0348, 0.0395], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 03:35:43,275 INFO [train.py:968] (1/2) Epoch 14, batch 4300, giga_loss[loss=0.3085, simple_loss=0.3678, pruned_loss=0.1246, over 28885.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3403, pruned_loss=0.09616, over 5705301.55 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09317, over 5076481.88 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3398, pruned_loss=0.09648, over 5705097.13 frames. ], batch size: 213, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:36:23,248 INFO [train.py:968] (1/2) Epoch 14, batch 4350, giga_loss[loss=0.2634, simple_loss=0.3415, pruned_loss=0.09265, over 28620.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3385, pruned_loss=0.09616, over 5706588.47 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09316, over 5084747.50 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3381, pruned_loss=0.09643, over 5704663.66 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:36:29,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=597657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:36:31,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=597660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:36:34,336 INFO [optim.py:369] (1/2) 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 03:36:54,380 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=597689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:37:03,031 INFO [train.py:968] (1/2) Epoch 14, batch 4400, giga_loss[loss=0.2231, simple_loss=0.3011, pruned_loss=0.07257, over 28433.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3353, pruned_loss=0.09441, over 5713429.18 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3475, pruned_loss=0.09289, over 5107608.49 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3351, pruned_loss=0.0949, over 5707983.36 frames. ], batch size: 71, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:37:28,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6646, 2.2657, 1.6076, 2.2990], device='cuda:1'), covar=tensor([0.2470, 0.2315, 0.2662, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.0994, 0.1194, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:37:39,537 INFO [train.py:968] (1/2) Epoch 14, batch 4450, giga_loss[loss=0.2509, simple_loss=0.3322, pruned_loss=0.08482, over 28735.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3354, pruned_loss=0.09412, over 5719522.50 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.348, pruned_loss=0.09331, over 5127088.09 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3345, pruned_loss=0.0942, over 5711244.37 frames. ], batch size: 242, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:37:53,281 INFO [optim.py:369] (1/2) 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 03:38:21,219 INFO [train.py:968] (1/2) Epoch 14, batch 4500, giga_loss[loss=0.2581, simple_loss=0.3305, pruned_loss=0.09283, over 28490.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3389, pruned_loss=0.09552, over 5706268.50 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09345, over 5146623.18 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3376, pruned_loss=0.09553, over 5702490.15 frames. ], batch size: 85, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:38:24,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2646, 1.7481, 1.3148, 0.5608], device='cuda:1'), covar=tensor([0.3382, 0.1809, 0.2596, 0.4313], device='cuda:1'), in_proj_covar=tensor([0.1601, 0.1499, 0.1505, 0.1309], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 03:38:30,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-07 03:39:02,569 INFO [train.py:968] (1/2) Epoch 14, batch 4550, giga_loss[loss=0.2682, simple_loss=0.348, pruned_loss=0.09423, over 28851.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3404, pruned_loss=0.09573, over 5698143.52 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.348, pruned_loss=0.09342, over 5152531.13 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3394, pruned_loss=0.0958, over 5699789.23 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:39:04,515 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:39:12,681 INFO [optim.py:369] (1/2) 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 03:39:20,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 03:39:26,665 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=597879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:39:31,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=597886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:39:45,556 INFO [train.py:968] (1/2) Epoch 14, batch 4600, giga_loss[loss=0.2644, simple_loss=0.3513, pruned_loss=0.08872, over 28622.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3418, pruned_loss=0.09556, over 5698033.10 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3476, pruned_loss=0.09322, over 5164003.57 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3412, pruned_loss=0.0958, over 5696557.63 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:40:05,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4452, 1.6974, 1.3273, 1.4587], device='cuda:1'), covar=tensor([0.2494, 0.2390, 0.2768, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1363, 0.0996, 0.1199, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:40:29,612 INFO [train.py:968] (1/2) Epoch 14, batch 4650, giga_loss[loss=0.2345, simple_loss=0.3127, pruned_loss=0.07815, over 28863.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3426, pruned_loss=0.09551, over 5692077.55 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3479, pruned_loss=0.09331, over 5177616.95 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3419, pruned_loss=0.09566, over 5688102.24 frames. ], batch size: 112, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:40:40,428 INFO [optim.py:369] (1/2) 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 03:40:53,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 03:41:06,655 INFO [train.py:968] (1/2) Epoch 14, batch 4700, giga_loss[loss=0.2734, simple_loss=0.3468, pruned_loss=0.09999, over 28863.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3418, pruned_loss=0.09505, over 5707178.14 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3474, pruned_loss=0.09311, over 5211487.45 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3413, pruned_loss=0.09543, over 5696189.90 frames. ], batch size: 199, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:41:20,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:41:22,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:41:30,126 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:41:32,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:41:45,577 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 03:41:46,594 INFO [train.py:968] (1/2) Epoch 14, batch 4750, libri_loss[loss=0.27, simple_loss=0.3562, pruned_loss=0.09188, over 25817.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3433, pruned_loss=0.09647, over 5691989.80 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3483, pruned_loss=0.09363, over 5214994.04 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.342, pruned_loss=0.09641, over 5694186.70 frames. ], batch size: 136, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:41:55,306 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:41:58,716 INFO [optim.py:369] (1/2) 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 03:42:29,028 INFO [train.py:968] (1/2) Epoch 14, batch 4800, giga_loss[loss=0.2791, simple_loss=0.3582, pruned_loss=0.1, over 28942.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3442, pruned_loss=0.09753, over 5693111.37 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3481, pruned_loss=0.09356, over 5221951.58 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3433, pruned_loss=0.09757, over 5693041.39 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:43:12,668 INFO [train.py:968] (1/2) Epoch 14, batch 4850, giga_loss[loss=0.3161, simple_loss=0.3829, pruned_loss=0.1247, over 28702.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3469, pruned_loss=0.09917, over 5695749.80 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.348, pruned_loss=0.09363, over 5235919.06 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3462, pruned_loss=0.09924, over 5691440.46 frames. ], batch size: 242, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:43:23,929 INFO [optim.py:369] (1/2) 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 03:43:52,861 INFO [train.py:968] (1/2) Epoch 14, batch 4900, giga_loss[loss=0.2988, simple_loss=0.3772, pruned_loss=0.1102, over 28844.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3501, pruned_loss=0.1007, over 5696360.27 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3483, pruned_loss=0.0938, over 5236304.52 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3493, pruned_loss=0.1007, over 5699635.48 frames. ], batch size: 227, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:44:03,792 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=598212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:44:16,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:44:35,274 INFO [train.py:968] (1/2) Epoch 14, batch 4950, giga_loss[loss=0.2487, simple_loss=0.3404, pruned_loss=0.07854, over 29026.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3526, pruned_loss=0.1021, over 5701884.79 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3485, pruned_loss=0.09384, over 5243094.43 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3519, pruned_loss=0.1021, over 5702586.45 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:44:41,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:44:49,600 INFO [optim.py:369] (1/2) 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 03:44:57,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-07 03:44:59,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4947, 1.7387, 1.4274, 1.4864], device='cuda:1'), covar=tensor([0.2415, 0.2354, 0.2663, 0.2297], device='cuda:1'), in_proj_covar=tensor([0.1361, 0.0995, 0.1200, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:45:16,499 INFO [train.py:968] (1/2) Epoch 14, batch 5000, giga_loss[loss=0.256, simple_loss=0.3393, pruned_loss=0.08633, over 28808.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1017, over 5710863.38 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3482, pruned_loss=0.09367, over 5255989.91 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3523, pruned_loss=0.102, over 5708053.00 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:45:33,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4587, 1.5701, 1.5632, 1.4374], device='cuda:1'), covar=tensor([0.1591, 0.1928, 0.2202, 0.1913], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0732, 0.0684, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 03:45:57,513 INFO [train.py:968] (1/2) Epoch 14, batch 5050, libri_loss[loss=0.2409, simple_loss=0.3162, pruned_loss=0.08275, over 29491.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3524, pruned_loss=0.1017, over 5719294.20 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3484, pruned_loss=0.09382, over 5271258.19 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3522, pruned_loss=0.1021, over 5715446.09 frames. ], batch size: 70, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:46:04,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2778, 3.0976, 2.9507, 1.4839], device='cuda:1'), covar=tensor([0.0901, 0.0998, 0.0859, 0.2240], device='cuda:1'), in_proj_covar=tensor([0.1074, 0.0998, 0.0871, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 03:46:09,004 INFO [optim.py:369] (1/2) 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 03:46:12,711 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:15,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:34,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:35,797 INFO [train.py:968] (1/2) Epoch 14, batch 5100, libri_loss[loss=0.2734, simple_loss=0.3625, pruned_loss=0.09217, over 28673.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3516, pruned_loss=0.1012, over 5723058.58 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09389, over 5286101.37 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3511, pruned_loss=0.1016, over 5716615.07 frames. ], batch size: 106, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:46:36,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:46:37,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=598402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:47:00,294 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 5150, libri_loss[loss=0.2596, simple_loss=0.3416, pruned_loss=0.08879, over 29597.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3493, pruned_loss=0.09997, over 5722549.39 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.349, pruned_loss=0.09408, over 5304105.92 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3488, pruned_loss=0.1004, over 5714588.89 frames. ], batch size: 74, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:47:18,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=598450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:47:32,817 INFO [optim.py:369] (1/2) 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 03:47:51,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8871, 2.1081, 2.2043, 1.6725], device='cuda:1'), covar=tensor([0.1709, 0.2207, 0.1368, 0.1600], device='cuda:1'), in_proj_covar=tensor([0.0846, 0.0690, 0.0889, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 03:48:00,510 INFO [train.py:968] (1/2) Epoch 14, batch 5200, giga_loss[loss=0.2682, simple_loss=0.3448, pruned_loss=0.09579, over 28768.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3449, pruned_loss=0.09754, over 5716597.02 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3492, pruned_loss=0.09421, over 5298003.39 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3443, pruned_loss=0.09776, over 5717909.94 frames. ], batch size: 284, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:48:40,594 INFO [train.py:968] (1/2) Epoch 14, batch 5250, giga_loss[loss=0.2274, simple_loss=0.3068, pruned_loss=0.07402, over 28579.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3429, pruned_loss=0.09617, over 5715690.23 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3489, pruned_loss=0.09409, over 5304068.51 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3427, pruned_loss=0.09645, over 5715206.16 frames. ], batch size: 85, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:48:47,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8332, 1.7525, 1.4343, 1.4084], device='cuda:1'), covar=tensor([0.0806, 0.0695, 0.0975, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0436, 0.0498, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:48:53,844 INFO [optim.py:369] (1/2) 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 03:49:11,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:49:22,290 INFO [train.py:968] (1/2) Epoch 14, batch 5300, giga_loss[loss=0.3004, simple_loss=0.3779, pruned_loss=0.1114, over 28529.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3449, pruned_loss=0.0957, over 5710119.41 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3495, pruned_loss=0.0943, over 5326560.13 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.344, pruned_loss=0.09583, over 5705757.73 frames. ], batch size: 336, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:50:04,984 INFO [train.py:968] (1/2) Epoch 14, batch 5350, giga_loss[loss=0.3351, simple_loss=0.3833, pruned_loss=0.1435, over 24087.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.346, pruned_loss=0.09631, over 5705345.81 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3495, pruned_loss=0.09438, over 5332388.83 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3452, pruned_loss=0.09634, over 5700197.50 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:50:15,806 INFO [optim.py:369] (1/2) 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 03:50:19,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1541, 5.3694, 2.2503, 2.4143], device='cuda:1'), covar=tensor([0.0792, 0.0281, 0.0766, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0366, 0.0518, 0.0351, 0.0397], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 03:50:42,076 INFO [train.py:968] (1/2) Epoch 14, batch 5400, giga_loss[loss=0.2553, simple_loss=0.3278, pruned_loss=0.09139, over 29071.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3465, pruned_loss=0.09742, over 5713527.62 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09457, over 5355040.87 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3454, pruned_loss=0.09741, over 5703789.39 frames. ], batch size: 136, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:51:08,238 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:51:11,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:51:15,982 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=598739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:51:16,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-07 03:51:25,962 INFO [train.py:968] (1/2) Epoch 14, batch 5450, giga_loss[loss=0.244, simple_loss=0.3133, pruned_loss=0.08733, over 28545.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3452, pruned_loss=0.09799, over 5711714.12 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3501, pruned_loss=0.09467, over 5357797.66 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3442, pruned_loss=0.09792, over 5703157.94 frames. ], batch size: 85, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:51:37,660 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:52:00,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9994, 1.3840, 1.2184, 1.1698], device='cuda:1'), covar=tensor([0.1581, 0.1362, 0.1901, 0.1597], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0735, 0.0689, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 03:52:08,348 INFO [train.py:968] (1/2) Epoch 14, batch 5500, giga_loss[loss=0.302, simple_loss=0.368, pruned_loss=0.118, over 28708.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3446, pruned_loss=0.0992, over 5711341.16 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09466, over 5362852.91 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3437, pruned_loss=0.09919, over 5703151.85 frames. ], batch size: 284, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:52:27,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=598825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:52:49,039 INFO [train.py:968] (1/2) Epoch 14, batch 5550, giga_loss[loss=0.2427, simple_loss=0.3203, pruned_loss=0.08254, over 28710.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3432, pruned_loss=0.0993, over 5711395.50 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09458, over 5376295.25 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3423, pruned_loss=0.09948, over 5700680.37 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:52:57,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5960, 1.5054, 1.3124, 1.1834], device='cuda:1'), covar=tensor([0.0595, 0.0407, 0.0770, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0439, 0.0502, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:53:03,014 INFO [optim.py:369] (1/2) 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 03:53:33,186 INFO [train.py:968] (1/2) Epoch 14, batch 5600, giga_loss[loss=0.3105, simple_loss=0.3681, pruned_loss=0.1264, over 27673.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.342, pruned_loss=0.09853, over 5718842.29 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3506, pruned_loss=0.09476, over 5390645.84 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3407, pruned_loss=0.09863, over 5706178.73 frames. ], batch size: 472, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:53:38,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4232, 1.6550, 1.3364, 1.4761], device='cuda:1'), covar=tensor([0.2334, 0.2333, 0.2591, 0.2160], device='cuda:1'), in_proj_covar=tensor([0.1354, 0.0989, 0.1194, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 03:53:52,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7681, 1.1293, 2.9589, 2.8669], device='cuda:1'), covar=tensor([0.1611, 0.2349, 0.0536, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0684, 0.0603, 0.0877, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 03:53:54,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 03:54:15,656 INFO [train.py:968] (1/2) Epoch 14, batch 5650, giga_loss[loss=0.2293, simple_loss=0.3093, pruned_loss=0.07467, over 28895.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3389, pruned_loss=0.09681, over 5717136.91 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3506, pruned_loss=0.09475, over 5395858.81 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3377, pruned_loss=0.09699, over 5710548.05 frames. ], batch size: 213, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 03:54:19,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3189, 1.2139, 1.2145, 1.5234], device='cuda:1'), covar=tensor([0.0727, 0.0340, 0.0328, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 03:54:28,297 INFO [optim.py:369] (1/2) 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 03:54:30,141 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=598968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:54:32,176 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=598971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:54:54,874 INFO [train.py:968] (1/2) Epoch 14, batch 5700, giga_loss[loss=0.2776, simple_loss=0.3357, pruned_loss=0.1097, over 23970.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3336, pruned_loss=0.0941, over 5721266.31 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09451, over 5406066.06 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3328, pruned_loss=0.09447, over 5712444.80 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:54:55,125 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=599000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:55:00,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3880, 1.4950, 1.4102, 1.5834], device='cuda:1'), covar=tensor([0.0723, 0.0311, 0.0313, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0210], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 03:55:35,198 INFO [train.py:968] (1/2) Epoch 14, batch 5750, giga_loss[loss=0.2529, simple_loss=0.3327, pruned_loss=0.08658, over 29015.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3324, pruned_loss=0.09359, over 5721534.14 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09452, over 5419713.48 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3313, pruned_loss=0.09384, over 5710605.89 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:55:48,968 INFO [optim.py:369] (1/2) 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 03:56:12,190 INFO [train.py:968] (1/2) Epoch 14, batch 5800, giga_loss[loss=0.2936, simple_loss=0.3588, pruned_loss=0.1142, over 28995.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3341, pruned_loss=0.09418, over 5730574.61 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3498, pruned_loss=0.09443, over 5433069.51 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3329, pruned_loss=0.09443, over 5717868.48 frames. ], batch size: 227, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:56:23,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=599114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:56:28,143 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3748, 1.9278, 1.4304, 0.5742], device='cuda:1'), covar=tensor([0.4292, 0.2134, 0.2868, 0.5235], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1520, 0.1512, 0.1319], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 03:56:33,619 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-07 03:56:49,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-07 03:56:52,155 INFO [train.py:968] (1/2) Epoch 14, batch 5850, giga_loss[loss=0.3198, simple_loss=0.3888, pruned_loss=0.1254, over 28732.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3381, pruned_loss=0.09594, over 5729368.02 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3496, pruned_loss=0.09441, over 5443628.83 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3369, pruned_loss=0.09615, over 5716562.69 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:56:58,362 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-07 03:57:06,078 INFO [optim.py:369] (1/2) 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 03:57:32,781 INFO [train.py:968] (1/2) Epoch 14, batch 5900, giga_loss[loss=0.2518, simple_loss=0.3227, pruned_loss=0.09045, over 28565.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3415, pruned_loss=0.09737, over 5728291.07 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3494, pruned_loss=0.09434, over 5454684.24 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3406, pruned_loss=0.09766, over 5714611.69 frames. ], batch size: 85, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:57:42,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-07 03:57:46,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6348, 4.4397, 4.2263, 2.1355], device='cuda:1'), covar=tensor([0.0468, 0.0624, 0.0611, 0.1781], device='cuda:1'), in_proj_covar=tensor([0.1087, 0.1005, 0.0879, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 03:57:48,730 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3592, 2.0261, 1.5295, 0.5051], device='cuda:1'), covar=tensor([0.3820, 0.2137, 0.3274, 0.5087], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1513, 0.1512, 0.1314], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 03:57:56,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2623, 1.4129, 1.4539, 1.3021], device='cuda:1'), covar=tensor([0.1408, 0.1475, 0.1918, 0.1508], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0741, 0.0691, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 03:58:09,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2397, 3.6030, 1.4063, 1.4943], device='cuda:1'), covar=tensor([0.0975, 0.0331, 0.0927, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0521, 0.0351, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 03:58:17,976 INFO [train.py:968] (1/2) Epoch 14, batch 5950, giga_loss[loss=0.2798, simple_loss=0.3485, pruned_loss=0.1055, over 28694.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3447, pruned_loss=0.09859, over 5726631.46 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3495, pruned_loss=0.09435, over 5465635.92 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3436, pruned_loss=0.09889, over 5711738.29 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 2.0
+2023-03-07 03:58:24,797 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=599257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:58:26,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=599260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:58:33,122 INFO [optim.py:369] (1/2) 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 03:58:52,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 03:58:53,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=599289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 03:59:01,856 INFO [train.py:968] (1/2) Epoch 14, batch 6000, giga_loss[loss=0.2666, simple_loss=0.3431, pruned_loss=0.09503, over 28626.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3467, pruned_loss=0.09975, over 5714228.50 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3498, pruned_loss=0.0946, over 5462983.18 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3456, pruned_loss=0.09984, over 5708853.86 frames. ], batch size: 242, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 03:59:01,857 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 03:59:09,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3324, 1.3118, 1.1144, 1.5410], device='cuda:1'), covar=tensor([0.0749, 0.0328, 0.0339, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0085, 0.0061, 0.0055, 0.0093], device='cuda:1')
+2023-03-07 03:59:10,162 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 03:59:56,782 INFO [train.py:968] (1/2) Epoch 14, batch 6050, giga_loss[loss=0.3157, simple_loss=0.3824, pruned_loss=0.1245, over 28869.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3533, pruned_loss=0.1055, over 5706259.90 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3501, pruned_loss=0.09467, over 5464937.27 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3522, pruned_loss=0.1055, over 5701431.14 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:00:08,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-07 04:00:11,848 INFO [optim.py:369] (1/2) 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 04:00:14,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1586, 1.2921, 1.1646, 1.1322], device='cuda:1'), covar=tensor([0.1812, 0.1655, 0.1408, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.1776, 0.1700, 0.1665, 0.1763], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 04:00:42,683 INFO [train.py:968] (1/2) Epoch 14, batch 6100, giga_loss[loss=0.3324, simple_loss=0.3986, pruned_loss=0.1331, over 28265.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3591, pruned_loss=0.1102, over 5692635.53 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3504, pruned_loss=0.09487, over 5469562.34 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3581, pruned_loss=0.1105, over 5692751.38 frames. ], batch size: 368, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:00:55,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2532, 1.4761, 1.2718, 1.0451], device='cuda:1'), covar=tensor([0.1873, 0.1956, 0.1377, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.1775, 0.1695, 0.1661, 0.1761], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 04:01:31,534 INFO [train.py:968] (1/2) Epoch 14, batch 6150, giga_loss[loss=0.4143, simple_loss=0.4568, pruned_loss=0.1859, over 28321.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.366, pruned_loss=0.1152, over 5677984.48 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3504, pruned_loss=0.09489, over 5473955.87 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3653, pruned_loss=0.1155, over 5676144.03 frames. ], batch size: 368, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:01:47,638 INFO [optim.py:369] (1/2) 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 04:02:20,052 INFO [train.py:968] (1/2) Epoch 14, batch 6200, giga_loss[loss=0.3201, simple_loss=0.3814, pruned_loss=0.1294, over 28746.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3724, pruned_loss=0.1207, over 5678301.60 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3505, pruned_loss=0.09512, over 5485516.57 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3723, pruned_loss=0.1215, over 5671891.10 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:02:46,594 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-07 04:03:03,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 04:03:08,119 INFO [train.py:968] (1/2) Epoch 14, batch 6250, giga_loss[loss=0.3197, simple_loss=0.3898, pruned_loss=0.1248, over 29032.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3793, pruned_loss=0.1268, over 5681841.84 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3507, pruned_loss=0.09521, over 5491625.82 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3794, pruned_loss=0.1276, over 5674104.72 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:03:22,374 INFO [optim.py:369] (1/2) 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 04:03:52,592 INFO [train.py:968] (1/2) Epoch 14, batch 6300, giga_loss[loss=0.3174, simple_loss=0.3761, pruned_loss=0.1294, over 28520.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3831, pruned_loss=0.1295, over 5681824.63 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3509, pruned_loss=0.09528, over 5511890.70 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3843, pruned_loss=0.1317, over 5665359.06 frames. ], batch size: 71, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:04:33,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1376, 4.9513, 4.7156, 2.3593], device='cuda:1'), covar=tensor([0.0452, 0.0581, 0.0647, 0.1765], device='cuda:1'), in_proj_covar=tensor([0.1095, 0.1014, 0.0888, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 04:04:41,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3484, 3.5152, 1.5903, 1.5081], device='cuda:1'), covar=tensor([0.0879, 0.0316, 0.0760, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0522, 0.0351, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0026], device='cuda:1')
+2023-03-07 04:04:44,598 INFO [train.py:968] (1/2) Epoch 14, batch 6350, giga_loss[loss=0.3252, simple_loss=0.3905, pruned_loss=0.1299, over 28982.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3873, pruned_loss=0.1337, over 5664452.76 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3513, pruned_loss=0.0954, over 5519865.19 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3887, pruned_loss=0.1362, over 5647650.36 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:05:01,758 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 6400, libri_loss[loss=0.2411, simple_loss=0.3273, pruned_loss=0.0774, over 29566.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3898, pruned_loss=0.1373, over 5647245.63 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3513, pruned_loss=0.09539, over 5527541.12 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3917, pruned_loss=0.1401, over 5629488.35 frames. ], batch size: 75, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:05:56,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4354, 4.2455, 4.0859, 1.8589], device='cuda:1'), covar=tensor([0.0572, 0.0720, 0.0744, 0.2053], device='cuda:1'), in_proj_covar=tensor([0.1101, 0.1019, 0.0893, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 04:06:10,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6173, 2.0768, 1.3149, 0.8491], device='cuda:1'), covar=tensor([0.5059, 0.3209, 0.2443, 0.4786], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1531, 0.1521, 0.1324], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 04:06:30,643 INFO [train.py:968] (1/2) Epoch 14, batch 6450, giga_loss[loss=0.4297, simple_loss=0.4441, pruned_loss=0.2076, over 27884.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3941, pruned_loss=0.1424, over 5628584.06 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3513, pruned_loss=0.09543, over 5529285.40 frames. ], giga_tot_loss[loss=0.3433, simple_loss=0.3962, pruned_loss=0.1452, over 5614160.73 frames. ], batch size: 412, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:06:36,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4355, 1.4314, 1.2725, 1.6047], device='cuda:1'), covar=tensor([0.0740, 0.0324, 0.0318, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0113, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 04:06:51,229 INFO [optim.py:369] (1/2) 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 04:07:14,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0119, 2.1719, 2.2931, 1.7710], device='cuda:1'), covar=tensor([0.1665, 0.2047, 0.1241, 0.1572], device='cuda:1'), in_proj_covar=tensor([0.0841, 0.0690, 0.0885, 0.0789], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 04:07:22,373 INFO [train.py:968] (1/2) Epoch 14, batch 6500, giga_loss[loss=0.3293, simple_loss=0.3878, pruned_loss=0.1354, over 28681.00 frames. ], tot_loss[loss=0.3433, simple_loss=0.3973, pruned_loss=0.1446, over 5618606.23 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3517, pruned_loss=0.0958, over 5526888.04 frames. ], giga_tot_loss[loss=0.3471, simple_loss=0.3994, pruned_loss=0.1474, over 5610201.37 frames. ], batch size: 262, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:08:13,055 INFO [train.py:968] (1/2) Epoch 14, batch 6550, giga_loss[loss=0.2997, simple_loss=0.3614, pruned_loss=0.119, over 28508.00 frames. ], tot_loss[loss=0.3445, simple_loss=0.3976, pruned_loss=0.1457, over 5625042.92 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3518, pruned_loss=0.09584, over 5527910.76 frames. ], giga_tot_loss[loss=0.3485, simple_loss=0.3999, pruned_loss=0.1485, over 5618741.15 frames. ], batch size: 71, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:08:27,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-07 04:08:31,810 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 6600, giga_loss[loss=0.3591, simple_loss=0.4077, pruned_loss=0.1553, over 28885.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3942, pruned_loss=0.1433, over 5634189.27 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3514, pruned_loss=0.09545, over 5538212.58 frames. ], giga_tot_loss[loss=0.3461, simple_loss=0.3976, pruned_loss=0.1473, over 5622778.91 frames. ], batch size: 285, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:09:33,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4869, 1.7619, 1.6835, 1.5156], device='cuda:1'), covar=tensor([0.1361, 0.1396, 0.1809, 0.1573], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0745, 0.0692, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 04:09:51,894 INFO [train.py:968] (1/2) Epoch 14, batch 6650, giga_loss[loss=0.3573, simple_loss=0.4048, pruned_loss=0.1549, over 27845.00 frames. ], tot_loss[loss=0.3379, simple_loss=0.3926, pruned_loss=0.1416, over 5622605.36 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3514, pruned_loss=0.09552, over 5532295.32 frames. ], giga_tot_loss[loss=0.3445, simple_loss=0.3966, pruned_loss=0.1462, over 5621013.00 frames. ], batch size: 412, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:10:08,439 INFO [optim.py:369] (1/2) 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 04:10:12,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 04:10:38,662 INFO [train.py:968] (1/2) Epoch 14, batch 6700, giga_loss[loss=0.3444, simple_loss=0.4043, pruned_loss=0.1422, over 28604.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3921, pruned_loss=0.1398, over 5632782.40 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3518, pruned_loss=0.096, over 5537271.78 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3959, pruned_loss=0.144, over 5629535.16 frames. ], batch size: 336, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:11:09,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5637, 1.8666, 1.4652, 1.3537], device='cuda:1'), covar=tensor([0.2920, 0.2752, 0.3172, 0.2504], device='cuda:1'), in_proj_covar=tensor([0.1351, 0.0992, 0.1194, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 04:11:17,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3472, 3.0968, 1.3693, 1.5675], device='cuda:1'), covar=tensor([0.0947, 0.0277, 0.0930, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0521, 0.0352, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 04:11:22,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1512, 1.2875, 1.0296, 1.0569], device='cuda:1'), covar=tensor([0.1067, 0.1128, 0.0913, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.1785, 0.1703, 0.1661, 0.1771], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 04:11:27,725 INFO [train.py:968] (1/2) Epoch 14, batch 6750, giga_loss[loss=0.3635, simple_loss=0.3905, pruned_loss=0.1682, over 23371.00 frames. ], tot_loss[loss=0.3367, simple_loss=0.393, pruned_loss=0.1402, over 5610824.39 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3518, pruned_loss=0.09604, over 5537149.53 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3971, pruned_loss=0.1449, over 5611238.23 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:11:45,402 INFO [optim.py:369] (1/2) 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 04:12:06,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:12:15,554 INFO [train.py:968] (1/2) Epoch 14, batch 6800, giga_loss[loss=0.4489, simple_loss=0.4595, pruned_loss=0.2191, over 26412.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3923, pruned_loss=0.1397, over 5609266.32 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3522, pruned_loss=0.09626, over 5540717.30 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.396, pruned_loss=0.144, over 5607731.57 frames. ], batch size: 555, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:13:08,548 INFO [train.py:968] (1/2) Epoch 14, batch 6850, giga_loss[loss=0.3761, simple_loss=0.4041, pruned_loss=0.1741, over 24016.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3888, pruned_loss=0.1355, over 5614492.54 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3524, pruned_loss=0.09646, over 5548138.73 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3923, pruned_loss=0.1395, over 5607870.82 frames. ], batch size: 705, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:13:13,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:13:26,461 INFO [optim.py:369] (1/2) 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 04:13:29,152 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:13:55,222 INFO [train.py:968] (1/2) Epoch 14, batch 6900, giga_loss[loss=0.2784, simple_loss=0.3473, pruned_loss=0.1047, over 28919.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3848, pruned_loss=0.131, over 5634448.31 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3528, pruned_loss=0.09669, over 5555228.84 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3882, pruned_loss=0.135, over 5624975.62 frames. ], batch size: 106, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:14:43,928 INFO [train.py:968] (1/2) Epoch 14, batch 6950, giga_loss[loss=0.317, simple_loss=0.3717, pruned_loss=0.1311, over 28595.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3814, pruned_loss=0.1283, over 5639867.33 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3525, pruned_loss=0.09662, over 5556425.29 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.385, pruned_loss=0.1323, over 5633619.98 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:14:47,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3809, 1.8012, 1.5740, 1.5702], device='cuda:1'), covar=tensor([0.0646, 0.0259, 0.0274, 0.0674], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 04:15:01,914 INFO [optim.py:369] (1/2) 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 04:15:14,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3496, 1.5930, 1.2672, 1.2459], device='cuda:1'), covar=tensor([0.2472, 0.2498, 0.2801, 0.2132], device='cuda:1'), in_proj_covar=tensor([0.1349, 0.0987, 0.1191, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 04:15:30,585 INFO [train.py:968] (1/2) Epoch 14, batch 7000, giga_loss[loss=0.2683, simple_loss=0.3444, pruned_loss=0.09612, over 29029.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3789, pruned_loss=0.1265, over 5649685.75 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3523, pruned_loss=0.09652, over 5564825.52 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3828, pruned_loss=0.1306, over 5639264.27 frames. ], batch size: 128, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:15:58,663 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600327.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 04:16:17,737 INFO [train.py:968] (1/2) Epoch 14, batch 7050, giga_loss[loss=0.3184, simple_loss=0.3919, pruned_loss=0.1225, over 28885.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3782, pruned_loss=0.1257, over 5655887.17 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3521, pruned_loss=0.09625, over 5574686.06 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3824, pruned_loss=0.1302, over 5640999.90 frames. ], batch size: 199, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:16:20,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:16:38,954 INFO [optim.py:369] (1/2) 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 04:16:46,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2675, 1.4383, 1.4161, 1.2944], device='cuda:1'), covar=tensor([0.1241, 0.1373, 0.1855, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0743, 0.0688, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 04:17:10,776 INFO [train.py:968] (1/2) Epoch 14, batch 7100, giga_loss[loss=0.2688, simple_loss=0.3458, pruned_loss=0.09592, over 28497.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3781, pruned_loss=0.1254, over 5667334.19 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3521, pruned_loss=0.09619, over 5580941.64 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3819, pruned_loss=0.1296, over 5651555.95 frames. ], batch size: 65, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:17:53,307 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:17:53,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:18:00,737 INFO [train.py:968] (1/2) Epoch 14, batch 7150, giga_loss[loss=0.3306, simple_loss=0.3845, pruned_loss=0.1384, over 28586.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.374, pruned_loss=0.1216, over 5676817.82 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3514, pruned_loss=0.09574, over 5592581.35 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3788, pruned_loss=0.1267, over 5657089.47 frames. ], batch size: 307, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:18:03,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2750, 2.0296, 1.8856, 1.7245], device='cuda:1'), covar=tensor([0.0836, 0.0764, 0.0880, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0363, 0.0438, 0.0499, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 04:18:15,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:18:19,483 INFO [optim.py:369] (1/2) 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 04:18:36,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:18:53,728 INFO [train.py:968] (1/2) Epoch 14, batch 7200, giga_loss[loss=0.2932, simple_loss=0.3772, pruned_loss=0.1046, over 29007.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.375, pruned_loss=0.1209, over 5664290.10 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3513, pruned_loss=0.09565, over 5589188.19 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3793, pruned_loss=0.1254, over 5653599.14 frames. ], batch size: 155, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:19:28,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:19:36,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2306, 1.2450, 3.3879, 3.0382], device='cuda:1'), covar=tensor([0.1499, 0.2624, 0.0462, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0607, 0.0884, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 04:19:43,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:19:44,192 INFO [train.py:968] (1/2) Epoch 14, batch 7250, giga_loss[loss=0.289, simple_loss=0.3638, pruned_loss=0.1071, over 28899.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3776, pruned_loss=0.1215, over 5661456.62 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3516, pruned_loss=0.09597, over 5584222.15 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3811, pruned_loss=0.1251, over 5658743.91 frames. ], batch size: 119, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:20:00,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=600565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:20:08,009 INFO [optim.py:369] (1/2) 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 04:20:27,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 04:20:40,654 INFO [train.py:968] (1/2) Epoch 14, batch 7300, giga_loss[loss=0.3974, simple_loss=0.4321, pruned_loss=0.1813, over 27973.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3784, pruned_loss=0.123, over 5664661.03 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3518, pruned_loss=0.09614, over 5587042.89 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3812, pruned_loss=0.1259, over 5660723.69 frames. ], batch size: 412, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:20:46,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:20:48,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:21:15,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:21:16,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3608, 3.1536, 2.9763, 1.3628], device='cuda:1'), covar=tensor([0.1063, 0.1169, 0.1254, 0.2391], device='cuda:1'), in_proj_covar=tensor([0.1105, 0.1031, 0.0898, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 04:21:25,520 INFO [train.py:968] (1/2) Epoch 14, batch 7350, giga_loss[loss=0.3234, simple_loss=0.3856, pruned_loss=0.1306, over 28729.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.377, pruned_loss=0.1221, over 5664464.46 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3522, pruned_loss=0.09637, over 5590657.95 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3797, pruned_loss=0.1251, over 5660466.20 frames. ], batch size: 284, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:21:45,093 INFO [optim.py:369] (1/2) 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 04:21:49,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0817, 1.0436, 3.6627, 3.1424], device='cuda:1'), covar=tensor([0.2259, 0.3308, 0.0873, 0.1292], device='cuda:1'), in_proj_covar=tensor([0.0689, 0.0607, 0.0882, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 04:21:49,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:21:51,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:05,799 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:06,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-07 04:22:08,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:12,763 INFO [train.py:968] (1/2) Epoch 14, batch 7400, giga_loss[loss=0.3267, simple_loss=0.3895, pruned_loss=0.1319, over 28363.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3754, pruned_loss=0.1224, over 5663354.89 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3522, pruned_loss=0.09633, over 5597841.55 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3782, pruned_loss=0.1254, over 5655210.04 frames. ], batch size: 369, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:22:15,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600702.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 04:22:18,423 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:34,510 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:39,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:22:59,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-07 04:22:59,641 INFO [train.py:968] (1/2) Epoch 14, batch 7450, giga_loss[loss=0.3361, simple_loss=0.3881, pruned_loss=0.142, over 28881.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3738, pruned_loss=0.1224, over 5674304.28 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3515, pruned_loss=0.0959, over 5604114.03 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.377, pruned_loss=0.1257, over 5663584.05 frames. ], batch size: 186, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:23:18,258 INFO [optim.py:369] (1/2) 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 04:23:37,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-07 04:23:48,461 INFO [train.py:968] (1/2) Epoch 14, batch 7500, giga_loss[loss=0.3056, simple_loss=0.3745, pruned_loss=0.1183, over 28945.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3735, pruned_loss=0.1213, over 5680074.22 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3518, pruned_loss=0.09628, over 5603793.18 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3764, pruned_loss=0.1243, over 5673228.99 frames. ], batch size: 86, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:23:53,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7553, 4.5325, 4.2986, 2.1646], device='cuda:1'), covar=tensor([0.0541, 0.0731, 0.0721, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.1106, 0.1029, 0.0899, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 04:24:03,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:04,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:04,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7044, 1.7425, 1.9043, 1.4265], device='cuda:1'), covar=tensor([0.1952, 0.2299, 0.1486, 0.1736], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0695, 0.0890, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 04:24:29,677 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600845.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 04:24:34,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600848.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 04:24:36,327 INFO [train.py:968] (1/2) Epoch 14, batch 7550, giga_loss[loss=0.2947, simple_loss=0.3719, pruned_loss=0.1087, over 28982.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3731, pruned_loss=0.1199, over 5690300.14 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3514, pruned_loss=0.09622, over 5607132.92 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3762, pruned_loss=0.1229, over 5683766.59 frames. ], batch size: 164, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:24:45,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:52,852 INFO [optim.py:369] (1/2) 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 04:24:53,146 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:55,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:24:58,813 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600877.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 04:25:19,431 INFO [train.py:968] (1/2) Epoch 14, batch 7600, giga_loss[loss=0.2891, simple_loss=0.3562, pruned_loss=0.111, over 28768.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3728, pruned_loss=0.1192, over 5700029.10 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3517, pruned_loss=0.09636, over 5619229.84 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3761, pruned_loss=0.1225, over 5687176.90 frames. ], batch size: 92, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:25:22,834 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:25:52,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:02,818 INFO [train.py:968] (1/2) Epoch 14, batch 7650, giga_loss[loss=0.2558, simple_loss=0.3406, pruned_loss=0.08554, over 28854.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3724, pruned_loss=0.1194, over 5701421.09 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.352, pruned_loss=0.09652, over 5626088.32 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3752, pruned_loss=0.1224, over 5686721.44 frames. ], batch size: 106, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:26:14,069 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:14,853 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=600960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:16,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:17,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=600963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:24,672 INFO [optim.py:369] (1/2) 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 04:26:43,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:44,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=600992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:51,212 INFO [train.py:968] (1/2) Epoch 14, batch 7700, giga_loss[loss=0.3789, simple_loss=0.4214, pruned_loss=0.1682, over 28312.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1192, over 5698822.86 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3514, pruned_loss=0.09629, over 5632289.26 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3743, pruned_loss=0.1226, over 5683753.14 frames. ], batch size: 368, lr: 2.33e-03, grad_scale: 8.0
+2023-03-07 04:26:52,785 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=601002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:26:55,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=601005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:27:26,126 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=601034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:27:39,194 INFO [train.py:968] (1/2) Epoch 14, batch 7750, giga_loss[loss=0.3199, simple_loss=0.3782, pruned_loss=0.1307, over 27564.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3722, pruned_loss=0.121, over 5695283.50 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3514, pruned_loss=0.09632, over 5639752.00 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3754, pruned_loss=0.1244, over 5678043.74 frames. ], batch size: 472, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:27:43,756 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=601055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:27:58,927 INFO [optim.py:369] (1/2) 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 04:28:12,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=601083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:28:14,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=601086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:28:27,020 INFO [train.py:968] (1/2) Epoch 14, batch 7800, giga_loss[loss=0.2812, simple_loss=0.3474, pruned_loss=0.1075, over 28223.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3709, pruned_loss=0.1204, over 5702766.37 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3514, pruned_loss=0.09635, over 5644772.45 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5685674.64 frames. ], batch size: 77, lr: 2.33e-03, grad_scale: 4.0
+2023-03-07 04:28:43,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=601115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:28:56,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 04:29:16,488 INFO [train.py:968] (1/2) Epoch 14, batch 7850, libri_loss[loss=0.2481, simple_loss=0.3238, pruned_loss=0.08621, over 29601.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3685, pruned_loss=0.1195, over 5708668.81 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3511, pruned_loss=0.09618, over 5650004.47 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1225, over 5691433.03 frames. ], batch size: 69, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:29:24,638 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-07 04:29:38,399 INFO [optim.py:369] (1/2) 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 04:30:03,998 INFO [train.py:968] (1/2) Epoch 14, batch 7900, giga_loss[loss=0.3131, simple_loss=0.3717, pruned_loss=0.1273, over 28569.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5709747.21 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3512, pruned_loss=0.09624, over 5653675.17 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1229, over 5693778.50 frames. ], batch size: 307, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:30:20,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 04:30:50,730 INFO [train.py:968] (1/2) Epoch 14, batch 7950, giga_loss[loss=0.2972, simple_loss=0.3612, pruned_loss=0.1166, over 28426.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3699, pruned_loss=0.1212, over 5699957.21 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3514, pruned_loss=0.09637, over 5659581.38 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1238, over 5683061.64 frames. ], batch size: 71, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:31:13,687 INFO [optim.py:369] (1/2) 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 04:31:37,975 INFO [train.py:968] (1/2) Epoch 14, batch 8000, libri_loss[loss=0.2856, simple_loss=0.3668, pruned_loss=0.1022, over 29539.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.37, pruned_loss=0.1206, over 5698864.06 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3511, pruned_loss=0.09617, over 5665763.63 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1234, over 5680452.68 frames. ], batch size: 81, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 04:32:23,620 INFO [train.py:968] (1/2) Epoch 14, batch 8050, giga_loss[loss=0.2915, simple_loss=0.3672, pruned_loss=0.1079, over 28831.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3705, pruned_loss=0.1202, over 5688036.70 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3516, pruned_loss=0.09651, over 5665575.96 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3724, pruned_loss=0.1226, over 5674058.01 frames. ], batch size: 174, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:32:46,630 INFO [optim.py:369] (1/2) 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 04:33:12,454 INFO [train.py:968] (1/2) Epoch 14, batch 8100, libri_loss[loss=0.2556, simple_loss=0.3334, pruned_loss=0.08888, over 29558.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3718, pruned_loss=0.1211, over 5667037.68 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3516, pruned_loss=0.09659, over 5653463.22 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3737, pruned_loss=0.1234, over 5666893.49 frames. ], batch size: 75, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:33:40,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=601430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:33:59,165 INFO [train.py:968] (1/2) Epoch 14, batch 8150, giga_loss[loss=0.2891, simple_loss=0.3653, pruned_loss=0.1065, over 28987.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3735, pruned_loss=0.1225, over 5669515.53 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3525, pruned_loss=0.09716, over 5644401.03 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3747, pruned_loss=0.1244, over 5676842.47 frames. ], batch size: 213, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:34:25,918 INFO [optim.py:369] (1/2) 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 04:34:26,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 04:34:55,521 INFO [train.py:968] (1/2) Epoch 14, batch 8200, giga_loss[loss=0.3686, simple_loss=0.409, pruned_loss=0.1641, over 28836.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3756, pruned_loss=0.125, over 5666371.36 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3522, pruned_loss=0.09699, over 5645808.03 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3769, pruned_loss=0.1268, over 5670829.74 frames. ], batch size: 186, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:35:37,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8321, 0.9539, 0.8297, 0.7950], device='cuda:1'), covar=tensor([0.1325, 0.1462, 0.1045, 0.1404], device='cuda:1'), in_proj_covar=tensor([0.1807, 0.1725, 0.1683, 0.1791], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 04:35:48,328 INFO [train.py:968] (1/2) Epoch 14, batch 8250, giga_loss[loss=0.2969, simple_loss=0.3636, pruned_loss=0.1151, over 29110.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3775, pruned_loss=0.1278, over 5670462.90 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3523, pruned_loss=0.09707, over 5649438.06 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3788, pruned_loss=0.1295, over 5671172.32 frames. ], batch size: 128, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:36:12,537 INFO [optim.py:369] (1/2) 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 04:36:12,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=601573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:36:15,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=601576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:36:36,859 INFO [train.py:968] (1/2) Epoch 14, batch 8300, giga_loss[loss=0.3072, simple_loss=0.3694, pruned_loss=0.1225, over 28996.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3792, pruned_loss=0.1297, over 5663210.60 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3527, pruned_loss=0.0973, over 5652153.25 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3804, pruned_loss=0.1316, over 5661907.06 frames. ], batch size: 164, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:36:41,183 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=601605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:37:27,775 INFO [train.py:968] (1/2) Epoch 14, batch 8350, giga_loss[loss=0.2495, simple_loss=0.3307, pruned_loss=0.08413, over 29002.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3783, pruned_loss=0.1294, over 5658640.97 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.353, pruned_loss=0.0975, over 5645940.14 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3794, pruned_loss=0.1311, over 5663143.68 frames. ], batch size: 136, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:37:47,416 INFO [optim.py:369] (1/2) 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 04:38:08,632 INFO [train.py:968] (1/2) Epoch 14, batch 8400, giga_loss[loss=0.3998, simple_loss=0.4359, pruned_loss=0.1818, over 26301.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3765, pruned_loss=0.1274, over 5665402.22 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3529, pruned_loss=0.09741, over 5655255.30 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3785, pruned_loss=0.1302, over 5660742.98 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 04:38:29,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5570, 4.4403, 1.7964, 1.6652], device='cuda:1'), covar=tensor([0.0962, 0.0305, 0.0868, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0526, 0.0354, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 04:38:45,342 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 8450, giga_loss[loss=0.3839, simple_loss=0.4158, pruned_loss=0.176, over 23789.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.377, pruned_loss=0.1264, over 5669287.57 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3528, pruned_loss=0.09737, over 5657599.87 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3789, pruned_loss=0.129, over 5663640.45 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:39:17,696 INFO [optim.py:369] (1/2) 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 04:39:38,521 INFO [train.py:968] (1/2) Epoch 14, batch 8500, libri_loss[loss=0.2603, simple_loss=0.3315, pruned_loss=0.09457, over 29403.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3737, pruned_loss=0.1237, over 5671433.34 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3529, pruned_loss=0.09767, over 5661737.98 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3758, pruned_loss=0.1264, over 5663453.55 frames. ], batch size: 67, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:40:23,977 INFO [train.py:968] (1/2) Epoch 14, batch 8550, giga_loss[loss=0.2879, simple_loss=0.3643, pruned_loss=0.1057, over 29068.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1231, over 5671391.14 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3534, pruned_loss=0.09788, over 5655849.99 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1254, over 5670261.09 frames. ], batch size: 128, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:40:44,634 INFO [optim.py:369] (1/2) 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 04:41:09,979 INFO [train.py:968] (1/2) Epoch 14, batch 8600, libri_loss[loss=0.2227, simple_loss=0.3064, pruned_loss=0.0695, over 29359.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3703, pruned_loss=0.1223, over 5672459.67 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3533, pruned_loss=0.09785, over 5664747.30 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3723, pruned_loss=0.1251, over 5663963.84 frames. ], batch size: 67, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:41:10,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3170, 1.4471, 3.0764, 2.9345], device='cuda:1'), covar=tensor([0.1314, 0.2264, 0.0497, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0692, 0.0609, 0.0886, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 04:41:36,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3774, 1.6800, 1.3715, 1.3030], device='cuda:1'), covar=tensor([0.2012, 0.1967, 0.2056, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.1355, 0.0995, 0.1193, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 04:41:57,922 INFO [train.py:968] (1/2) Epoch 14, batch 8650, giga_loss[loss=0.2985, simple_loss=0.3612, pruned_loss=0.1179, over 28932.00 frames. ], tot_loss[loss=0.309, simple_loss=0.371, pruned_loss=0.1235, over 5651251.53 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3532, pruned_loss=0.09781, over 5659945.76 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.373, pruned_loss=0.1261, over 5648322.74 frames. ], batch size: 112, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:42:03,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=601953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:42:23,619 INFO [optim.py:369] (1/2) 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 04:42:49,693 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 04:42:49,791 INFO [train.py:968] (1/2) Epoch 14, batch 8700, giga_loss[loss=0.3757, simple_loss=0.4168, pruned_loss=0.1673, over 28261.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.125, over 5659183.70 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3532, pruned_loss=0.09781, over 5659945.76 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3773, pruned_loss=0.127, over 5656904.18 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:43:41,377 INFO [train.py:968] (1/2) Epoch 14, batch 8750, giga_loss[loss=0.3735, simple_loss=0.4297, pruned_loss=0.1586, over 28613.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3774, pruned_loss=0.1235, over 5666059.17 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3532, pruned_loss=0.09792, over 5662526.78 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3788, pruned_loss=0.1254, over 5661935.11 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:44:04,336 INFO [optim.py:369] (1/2) 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 04:44:28,435 INFO [train.py:968] (1/2) Epoch 14, batch 8800, giga_loss[loss=0.3547, simple_loss=0.4035, pruned_loss=0.1529, over 27906.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3784, pruned_loss=0.1242, over 5668936.78 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3527, pruned_loss=0.09761, over 5668780.43 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3807, pruned_loss=0.1267, over 5659808.65 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:44:32,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 04:44:40,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=602114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:44:56,633 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 04:45:11,922 INFO [train.py:968] (1/2) Epoch 14, batch 8850, giga_loss[loss=0.3954, simple_loss=0.4364, pruned_loss=0.1772, over 27665.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3809, pruned_loss=0.1268, over 5656988.30 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3529, pruned_loss=0.09779, over 5663647.17 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3833, pruned_loss=0.1293, over 5654290.77 frames. ], batch size: 474, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:45:35,240 INFO [optim.py:369] (1/2) 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 04:45:37,033 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-07 04:45:55,792 INFO [train.py:968] (1/2) Epoch 14, batch 8900, libri_loss[loss=0.2615, simple_loss=0.3467, pruned_loss=0.08817, over 29544.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3808, pruned_loss=0.1266, over 5654684.54 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3534, pruned_loss=0.09794, over 5663546.20 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3835, pruned_loss=0.1298, over 5652364.71 frames. ], batch size: 83, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:46:10,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3684, 2.0824, 1.6176, 0.5204], device='cuda:1'), covar=tensor([0.4391, 0.2295, 0.3092, 0.5044], device='cuda:1'), in_proj_covar=tensor([0.1609, 0.1539, 0.1511, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 04:46:45,817 INFO [train.py:968] (1/2) Epoch 14, batch 8950, giga_loss[loss=0.2737, simple_loss=0.3493, pruned_loss=0.09904, over 29066.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3801, pruned_loss=0.1277, over 5644256.07 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3534, pruned_loss=0.098, over 5662554.05 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3824, pruned_loss=0.1304, over 5643526.26 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:46:52,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=602257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:46:55,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=602260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:46:55,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=602260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:47:10,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6979, 1.7592, 1.5324, 1.5144], device='cuda:1'), covar=tensor([0.1466, 0.2348, 0.2165, 0.2132], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0738, 0.0685, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 04:47:12,270 INFO [optim.py:369] (1/2) 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 04:47:24,141 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=602289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:47:34,890 INFO [train.py:968] (1/2) Epoch 14, batch 9000, giga_loss[loss=0.3418, simple_loss=0.3892, pruned_loss=0.1472, over 28837.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3777, pruned_loss=0.1265, over 5645488.31 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3536, pruned_loss=0.09818, over 5662427.19 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3797, pruned_loss=0.1288, over 5644591.30 frames. ], batch size: 112, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:47:34,891 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 04:47:43,431 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 04:48:09,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=602328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:48:27,456 INFO [train.py:968] (1/2) Epoch 14, batch 9050, libri_loss[loss=0.2374, simple_loss=0.3153, pruned_loss=0.07977, over 29476.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.375, pruned_loss=0.1245, over 5660846.24 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3537, pruned_loss=0.0982, over 5672858.18 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3775, pruned_loss=0.1275, over 5650088.83 frames. ], batch size: 70, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:48:52,196 INFO [optim.py:369] (1/2) 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 04:49:16,605 INFO [train.py:968] (1/2) Epoch 14, batch 9100, giga_loss[loss=0.272, simple_loss=0.3384, pruned_loss=0.1028, over 28531.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3753, pruned_loss=0.1256, over 5657851.43 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.354, pruned_loss=0.09835, over 5676443.70 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3776, pruned_loss=0.1285, over 5645610.96 frames. ], batch size: 78, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:49:18,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5315, 1.7043, 1.3697, 1.5549], device='cuda:1'), covar=tensor([0.2208, 0.2301, 0.2495, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.1356, 0.0998, 0.1196, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 04:49:24,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-07 04:50:07,162 INFO [train.py:968] (1/2) Epoch 14, batch 9150, giga_loss[loss=0.3262, simple_loss=0.3829, pruned_loss=0.1347, over 28672.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3764, pruned_loss=0.1268, over 5648281.36 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.354, pruned_loss=0.09816, over 5678639.47 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3787, pruned_loss=0.1299, over 5636102.63 frames. ], batch size: 92, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 04:50:29,916 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=602471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:50:32,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:50:54,418 INFO [train.py:968] (1/2) Epoch 14, batch 9200, giga_loss[loss=0.4021, simple_loss=0.4317, pruned_loss=0.1863, over 26579.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3731, pruned_loss=0.1252, over 5661948.43 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3539, pruned_loss=0.09811, over 5681717.52 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3753, pruned_loss=0.1281, over 5649217.69 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:50:58,759 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=602503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:51:44,574 INFO [train.py:968] (1/2) Epoch 14, batch 9250, giga_loss[loss=0.3525, simple_loss=0.3853, pruned_loss=0.1598, over 23486.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3715, pruned_loss=0.1242, over 5656640.74 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3535, pruned_loss=0.09788, over 5686110.24 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3739, pruned_loss=0.1273, over 5641919.28 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:52:07,339 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=602591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:52:32,327 INFO [train.py:968] (1/2) Epoch 14, batch 9300, giga_loss[loss=0.3983, simple_loss=0.4271, pruned_loss=0.1848, over 26542.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1232, over 5662764.50 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3534, pruned_loss=0.0979, over 5689726.20 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3743, pruned_loss=0.1261, over 5647289.84 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:53:02,406 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 9350, giga_loss[loss=0.28, simple_loss=0.3506, pruned_loss=0.1047, over 28837.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1248, over 5661848.46 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3533, pruned_loss=0.09779, over 5686133.09 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3771, pruned_loss=0.1278, over 5651656.07 frames. ], batch size: 99, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:53:43,949 INFO [optim.py:369] (1/2) 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 04:54:07,316 INFO [train.py:968] (1/2) Epoch 14, batch 9400, giga_loss[loss=0.2816, simple_loss=0.3569, pruned_loss=0.1032, over 28835.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3757, pruned_loss=0.1267, over 5655762.94 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3533, pruned_loss=0.09779, over 5686133.09 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3776, pruned_loss=0.1291, over 5647830.07 frames. ], batch size: 174, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:54:47,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 04:54:54,848 INFO [train.py:968] (1/2) Epoch 14, batch 9450, giga_loss[loss=0.2595, simple_loss=0.3556, pruned_loss=0.08172, over 29010.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.376, pruned_loss=0.1242, over 5652141.15 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3531, pruned_loss=0.0978, over 5675192.93 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3784, pruned_loss=0.127, over 5655128.13 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:55:14,914 INFO [optim.py:369] (1/2) 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 04:55:16,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=602778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:55:18,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=602781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:55:25,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=602788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:55:35,089 INFO [train.py:968] (1/2) Epoch 14, batch 9500, giga_loss[loss=0.3027, simple_loss=0.3847, pruned_loss=0.1104, over 28500.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3753, pruned_loss=0.1219, over 5661383.26 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3526, pruned_loss=0.09754, over 5684885.32 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3789, pruned_loss=0.1257, over 5653977.81 frames. ], batch size: 85, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:55:44,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=602810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:55:55,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1681, 1.5248, 1.4363, 1.0615], device='cuda:1'), covar=tensor([0.1641, 0.2389, 0.1393, 0.1583], device='cuda:1'), in_proj_covar=tensor([0.0846, 0.0696, 0.0888, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 04:56:23,461 INFO [train.py:968] (1/2) Epoch 14, batch 9550, giga_loss[loss=0.3357, simple_loss=0.3948, pruned_loss=0.1383, over 28724.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3791, pruned_loss=0.1236, over 5668686.50 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3525, pruned_loss=0.09745, over 5687086.94 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3823, pruned_loss=0.1269, over 5660800.28 frames. ], batch size: 242, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:56:46,190 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:57:13,271 INFO [train.py:968] (1/2) Epoch 14, batch 9600, giga_loss[loss=0.3223, simple_loss=0.3883, pruned_loss=0.1281, over 28869.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3816, pruned_loss=0.1258, over 5669071.08 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3525, pruned_loss=0.09749, over 5686886.91 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3846, pruned_loss=0.129, over 5662718.37 frames. ], batch size: 136, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 04:57:58,771 INFO [train.py:968] (1/2) Epoch 14, batch 9650, giga_loss[loss=0.475, simple_loss=0.4739, pruned_loss=0.2381, over 26698.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3844, pruned_loss=0.129, over 5678863.32 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3527, pruned_loss=0.09764, over 5691500.64 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3873, pruned_loss=0.132, over 5669242.48 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:58:17,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=602966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 04:58:21,116 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-07 04:58:26,431 INFO [optim.py:369] (1/2) 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 04:58:47,677 INFO [train.py:968] (1/2) Epoch 14, batch 9700, giga_loss[loss=0.3757, simple_loss=0.4209, pruned_loss=0.1652, over 27442.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3845, pruned_loss=0.1304, over 5663269.23 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3525, pruned_loss=0.09754, over 5695533.69 frames. ], giga_tot_loss[loss=0.3273, simple_loss=0.3877, pruned_loss=0.1335, over 5651648.64 frames. ], batch size: 472, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:58:59,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3167, 3.1085, 2.9726, 1.3921], device='cuda:1'), covar=tensor([0.0891, 0.1010, 0.0937, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1118, 0.1044, 0.0912, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 04:59:33,967 INFO [train.py:968] (1/2) Epoch 14, batch 9750, giga_loss[loss=0.3105, simple_loss=0.3706, pruned_loss=0.1252, over 28484.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.383, pruned_loss=0.1289, over 5668832.39 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3521, pruned_loss=0.0972, over 5696430.73 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3866, pruned_loss=0.1324, over 5657982.13 frames. ], batch size: 85, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 04:59:57,261 INFO [optim.py:369] (1/2) 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 05:00:19,158 INFO [train.py:968] (1/2) Epoch 14, batch 9800, giga_loss[loss=0.2776, simple_loss=0.3669, pruned_loss=0.09417, over 29021.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3822, pruned_loss=0.1263, over 5668713.33 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3521, pruned_loss=0.09717, over 5693044.76 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3852, pruned_loss=0.1294, over 5662937.18 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:00:28,686 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:00:30,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:00:58,405 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:01:05,983 INFO [train.py:968] (1/2) Epoch 14, batch 9850, giga_loss[loss=0.311, simple_loss=0.3737, pruned_loss=0.1241, over 28686.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3826, pruned_loss=0.1257, over 5672299.48 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3521, pruned_loss=0.09721, over 5695222.83 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3853, pruned_loss=0.1283, over 5665506.09 frames. ], batch size: 242, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:01:15,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:01:29,150 INFO [optim.py:369] (1/2) 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 05:01:32,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:01:53,048 INFO [train.py:968] (1/2) Epoch 14, batch 9900, giga_loss[loss=0.4373, simple_loss=0.454, pruned_loss=0.2103, over 26437.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3848, pruned_loss=0.1279, over 5669469.01 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3521, pruned_loss=0.09727, over 5697475.35 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3878, pruned_loss=0.1307, over 5661320.31 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:02:16,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7051, 1.8226, 1.6670, 1.4729], device='cuda:1'), covar=tensor([0.2451, 0.2140, 0.1926, 0.2239], device='cuda:1'), in_proj_covar=tensor([0.1782, 0.1693, 0.1658, 0.1764], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 05:02:18,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2427, 1.5537, 1.2908, 1.4227], device='cuda:1'), covar=tensor([0.0689, 0.0418, 0.0329, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 05:02:40,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:02:42,901 INFO [train.py:968] (1/2) Epoch 14, batch 9950, giga_loss[loss=0.3499, simple_loss=0.3866, pruned_loss=0.1566, over 23552.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3832, pruned_loss=0.1275, over 5665232.20 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3519, pruned_loss=0.09722, over 5702898.50 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3869, pruned_loss=0.1308, over 5652782.32 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:03:04,263 INFO [optim.py:369] (1/2) 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 05:03:07,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3322, 1.5085, 1.3847, 1.5190], device='cuda:1'), covar=tensor([0.0734, 0.0348, 0.0310, 0.0762], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0094], device='cuda:1')
+2023-03-07 05:03:26,493 INFO [train.py:968] (1/2) Epoch 14, batch 10000, giga_loss[loss=0.3318, simple_loss=0.3906, pruned_loss=0.1365, over 28863.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3822, pruned_loss=0.128, over 5657007.58 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3522, pruned_loss=0.09755, over 5699286.87 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3858, pruned_loss=0.1312, over 5649212.22 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:03:34,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:03:37,082 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:04:13,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-07 05:04:17,763 INFO [train.py:968] (1/2) Epoch 14, batch 10050, giga_loss[loss=0.285, simple_loss=0.36, pruned_loss=0.105, over 29039.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3813, pruned_loss=0.1284, over 5664929.06 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3521, pruned_loss=0.09746, over 5701173.61 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3845, pruned_loss=0.1314, over 5656587.81 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:04:27,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-07 05:04:44,872 INFO [optim.py:369] (1/2) 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 05:05:00,881 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:05:03,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:05:08,424 INFO [train.py:968] (1/2) Epoch 14, batch 10100, giga_loss[loss=0.2583, simple_loss=0.3337, pruned_loss=0.09147, over 28732.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3789, pruned_loss=0.1278, over 5658061.12 frames. ], libri_tot_loss[loss=0.2734, simple_loss=0.3521, pruned_loss=0.09739, over 5703006.84 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.382, pruned_loss=0.1308, over 5649069.46 frames. ], batch size: 99, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:05:33,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:05:56,588 INFO [train.py:968] (1/2) Epoch 14, batch 10150, giga_loss[loss=0.2837, simple_loss=0.3563, pruned_loss=0.1056, over 28897.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3773, pruned_loss=0.1271, over 5659438.63 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3518, pruned_loss=0.09718, over 5704228.20 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3807, pruned_loss=0.1305, over 5649970.83 frames. ], batch size: 186, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:06:10,116 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:06:43,946 INFO [train.py:968] (1/2) Epoch 14, batch 10200, giga_loss[loss=0.323, simple_loss=0.3814, pruned_loss=0.1323, over 28874.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3762, pruned_loss=0.1267, over 5661948.34 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3515, pruned_loss=0.09706, over 5707407.23 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3797, pruned_loss=0.1301, over 5650774.41 frames. ], batch size: 186, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:06:46,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5168, 1.7599, 1.4761, 1.6022], device='cuda:1'), covar=tensor([0.2114, 0.2033, 0.2154, 0.1898], device='cuda:1'), in_proj_covar=tensor([0.1365, 0.1005, 0.1202, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 05:07:30,494 INFO [train.py:968] (1/2) Epoch 14, batch 10250, giga_loss[loss=0.251, simple_loss=0.3392, pruned_loss=0.08136, over 28986.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3725, pruned_loss=0.1223, over 5671575.53 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3516, pruned_loss=0.09707, over 5709608.92 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3759, pruned_loss=0.1259, over 5658752.93 frames. ], batch size: 136, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:07:36,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:07:54,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:07:55,030 INFO [optim.py:369] (1/2) 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 05:08:13,116 INFO [train.py:968] (1/2) Epoch 14, batch 10300, giga_loss[loss=0.3018, simple_loss=0.3696, pruned_loss=0.117, over 28749.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3687, pruned_loss=0.1184, over 5657454.56 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3517, pruned_loss=0.0971, over 5707699.85 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.372, pruned_loss=0.1221, over 5647623.93 frames. ], batch size: 262, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:08:24,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-07 05:09:02,211 INFO [train.py:968] (1/2) Epoch 14, batch 10350, giga_loss[loss=0.2675, simple_loss=0.3424, pruned_loss=0.09626, over 28917.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3683, pruned_loss=0.1175, over 5666228.68 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3515, pruned_loss=0.09705, over 5711675.82 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5653493.92 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:09:14,343 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:09:28,255 INFO [optim.py:369] (1/2) 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 05:09:49,818 INFO [train.py:968] (1/2) Epoch 14, batch 10400, giga_loss[loss=0.3069, simple_loss=0.3582, pruned_loss=0.1277, over 28452.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3668, pruned_loss=0.1173, over 5659716.94 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3515, pruned_loss=0.0971, over 5706347.75 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3698, pruned_loss=0.1205, over 5654155.81 frames. ], batch size: 78, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:09:50,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:09:53,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:10:19,668 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=603732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:10:34,295 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:10:38,375 INFO [train.py:968] (1/2) Epoch 14, batch 10450, giga_loss[loss=0.3547, simple_loss=0.382, pruned_loss=0.1637, over 23524.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3644, pruned_loss=0.1165, over 5654517.68 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3516, pruned_loss=0.09725, over 5702545.76 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.367, pruned_loss=0.1195, over 5652125.97 frames. ], batch size: 705, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:11:06,170 INFO [optim.py:369] (1/2) 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 05:11:24,255 INFO [train.py:968] (1/2) Epoch 14, batch 10500, giga_loss[loss=0.2928, simple_loss=0.3612, pruned_loss=0.1122, over 28319.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3662, pruned_loss=0.1175, over 5665407.86 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3517, pruned_loss=0.09715, over 5707378.17 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1205, over 5658119.88 frames. ], batch size: 77, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:11:56,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 10550, giga_loss[loss=0.2919, simple_loss=0.3633, pruned_loss=0.1102, over 29003.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3675, pruned_loss=0.1177, over 5671833.68 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3514, pruned_loss=0.09694, over 5716079.32 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3704, pruned_loss=0.1213, over 5656135.43 frames. ], batch size: 128, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:12:37,001 INFO [optim.py:369] (1/2) 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 05:12:54,349 INFO [train.py:968] (1/2) Epoch 14, batch 10600, giga_loss[loss=0.2864, simple_loss=0.3584, pruned_loss=0.1072, over 28756.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3694, pruned_loss=0.1192, over 5667066.52 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09708, over 5720415.10 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3719, pruned_loss=0.1225, over 5649383.74 frames. ], batch size: 242, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:12:59,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=603903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:13:26,797 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 10650, giga_loss[loss=0.2851, simple_loss=0.3557, pruned_loss=0.1073, over 28963.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3683, pruned_loss=0.1182, over 5658349.90 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3521, pruned_loss=0.09724, over 5713608.49 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3705, pruned_loss=0.1215, over 5647736.55 frames. ], batch size: 145, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:13:44,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=603952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:13:49,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6503, 1.4298, 1.8407, 1.3125], device='cuda:1'), covar=tensor([0.2080, 0.3072, 0.1511, 0.1746], device='cuda:1'), in_proj_covar=tensor([0.0855, 0.0703, 0.0899, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 05:14:10,744 INFO [optim.py:369] (1/2) 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 05:14:13,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=603982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:14:15,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=603985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:14:27,830 INFO [train.py:968] (1/2) Epoch 14, batch 10700, libri_loss[loss=0.269, simple_loss=0.3381, pruned_loss=0.09992, over 29388.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3693, pruned_loss=0.1196, over 5651796.79 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.3525, pruned_loss=0.09743, over 5706901.86 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.371, pruned_loss=0.1224, over 5647478.90 frames. ], batch size: 67, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:14:28,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-07 05:14:43,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:15:05,639 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:15:19,975 INFO [train.py:968] (1/2) Epoch 14, batch 10750, giga_loss[loss=0.2945, simple_loss=0.3666, pruned_loss=0.1112, over 28822.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5648321.03 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3524, pruned_loss=0.09731, over 5707522.66 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3735, pruned_loss=0.1242, over 5643445.50 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:15:23,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6068, 1.7409, 1.6634, 1.5986], device='cuda:1'), covar=tensor([0.1508, 0.1904, 0.1951, 0.1804], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0744, 0.0689, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:15:49,901 INFO [optim.py:369] (1/2) 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 05:16:02,479 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:03,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5693, 1.6813, 1.8427, 1.3716], device='cuda:1'), covar=tensor([0.1604, 0.2106, 0.1275, 0.1506], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0699, 0.0895, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 05:16:04,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:05,587 INFO [train.py:968] (1/2) Epoch 14, batch 10800, libri_loss[loss=0.2666, simple_loss=0.3473, pruned_loss=0.09298, over 29684.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1203, over 5665263.58 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3515, pruned_loss=0.09678, over 5713566.25 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3741, pruned_loss=0.1239, over 5653971.50 frames. ], batch size: 73, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:16:21,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:28,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1350, 1.1652, 3.8237, 3.0998], device='cuda:1'), covar=tensor([0.1793, 0.2718, 0.0483, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0693, 0.0611, 0.0891, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 05:16:28,783 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:16:50,743 INFO [train.py:968] (1/2) Epoch 14, batch 10850, giga_loss[loss=0.2778, simple_loss=0.351, pruned_loss=0.1023, over 28380.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.1219, over 5667787.88 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3519, pruned_loss=0.09694, over 5709008.74 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3758, pruned_loss=0.1252, over 5661438.55 frames. ], batch size: 77, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:17:19,836 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:17:21,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:17:40,797 INFO [train.py:968] (1/2) Epoch 14, batch 10900, giga_loss[loss=0.2743, simple_loss=0.3434, pruned_loss=0.1026, over 28830.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3743, pruned_loss=0.1233, over 5672134.64 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3522, pruned_loss=0.09702, over 5710676.83 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3764, pruned_loss=0.1262, over 5664949.18 frames. ], batch size: 112, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:17:52,153 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=604211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:17:52,167 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:17:53,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3407, 1.5113, 1.4862, 1.3566], device='cuda:1'), covar=tensor([0.1384, 0.1761, 0.1821, 0.1690], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0740, 0.0687, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:18:03,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1119, 3.9400, 3.6784, 1.8890], device='cuda:1'), covar=tensor([0.0609, 0.0779, 0.0880, 0.2207], device='cuda:1'), in_proj_covar=tensor([0.1124, 0.1049, 0.0915, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 05:18:31,182 INFO [train.py:968] (1/2) Epoch 14, batch 10950, giga_loss[loss=0.2849, simple_loss=0.3578, pruned_loss=0.106, over 28816.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3755, pruned_loss=0.1229, over 5656304.89 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3522, pruned_loss=0.09704, over 5704384.98 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3776, pruned_loss=0.1257, over 5656300.69 frames. ], batch size: 119, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:18:47,749 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:18:51,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:19:02,186 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:19:08,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5661, 4.8156, 1.8037, 1.6611], device='cuda:1'), covar=tensor([0.0968, 0.0301, 0.0864, 0.1378], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0529, 0.0355, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 05:19:19,754 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:19:24,660 INFO [train.py:968] (1/2) Epoch 14, batch 11000, giga_loss[loss=0.3811, simple_loss=0.4286, pruned_loss=0.1668, over 28840.00 frames. ], tot_loss[loss=0.31, simple_loss=0.375, pruned_loss=0.1225, over 5652654.72 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3519, pruned_loss=0.09689, over 5708542.27 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3773, pruned_loss=0.1253, over 5647993.31 frames. ], batch size: 174, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:19:32,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4245, 1.7115, 1.4096, 1.3239], device='cuda:1'), covar=tensor([0.2250, 0.2114, 0.2345, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.1362, 0.1004, 0.1199, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 05:19:33,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:20:12,734 INFO [train.py:968] (1/2) Epoch 14, batch 11050, giga_loss[loss=0.2934, simple_loss=0.3681, pruned_loss=0.1094, over 29040.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3752, pruned_loss=0.1234, over 5640831.30 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3521, pruned_loss=0.09698, over 5699325.81 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 5644931.85 frames. ], batch size: 155, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:20:45,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-07 05:20:51,863 INFO [optim.py:369] (1/2) 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 05:21:03,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7288, 1.0737, 2.8660, 2.6986], device='cuda:1'), covar=tensor([0.1683, 0.2505, 0.0600, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0613, 0.0894, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 05:21:15,336 INFO [train.py:968] (1/2) Epoch 14, batch 11100, libri_loss[loss=0.2712, simple_loss=0.3553, pruned_loss=0.09355, over 29512.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.375, pruned_loss=0.1243, over 5636461.42 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3523, pruned_loss=0.09696, over 5701367.39 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3767, pruned_loss=0.1267, over 5637060.13 frames. ], batch size: 81, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:21:32,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=604421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:21:32,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:21:36,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:21:59,420 INFO [train.py:968] (1/2) Epoch 14, batch 11150, giga_loss[loss=0.341, simple_loss=0.3932, pruned_loss=0.1444, over 27560.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3724, pruned_loss=0.1228, over 5642892.95 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3517, pruned_loss=0.0965, over 5706199.04 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1261, over 5637106.44 frames. ], batch size: 472, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:22:02,530 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:22:02,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:22:05,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:22:29,549 INFO [optim.py:369] (1/2) 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 05:22:33,496 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 11200, giga_loss[loss=0.3467, simple_loss=0.4071, pruned_loss=0.1431, over 28823.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3724, pruned_loss=0.1238, over 5647282.47 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3517, pruned_loss=0.0965, over 5706199.04 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3745, pruned_loss=0.1263, over 5642778.76 frames. ], batch size: 174, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:23:35,652 INFO [train.py:968] (1/2) Epoch 14, batch 11250, giga_loss[loss=0.4158, simple_loss=0.4302, pruned_loss=0.2007, over 26712.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3721, pruned_loss=0.124, over 5652737.33 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3519, pruned_loss=0.09663, over 5709152.87 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3739, pruned_loss=0.1264, over 5645706.85 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:24:05,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=604578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:24:07,466 INFO [optim.py:369] (1/2) 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 05:24:12,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:24:24,243 INFO [train.py:968] (1/2) Epoch 14, batch 11300, giga_loss[loss=0.3078, simple_loss=0.3748, pruned_loss=0.1204, over 28857.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3717, pruned_loss=0.1237, over 5656481.26 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3517, pruned_loss=0.09647, over 5711547.59 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3739, pruned_loss=0.1265, over 5647522.45 frames. ], batch size: 174, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:25:12,562 INFO [train.py:968] (1/2) Epoch 14, batch 11350, giga_loss[loss=0.3925, simple_loss=0.4231, pruned_loss=0.1809, over 26600.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3741, pruned_loss=0.1259, over 5656275.93 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3516, pruned_loss=0.09626, over 5715231.62 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3763, pruned_loss=0.1288, over 5644821.38 frames. ], batch size: 555, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:25:15,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3739, 1.7240, 1.3551, 1.3589], device='cuda:1'), covar=tensor([0.2446, 0.2448, 0.2753, 0.2134], device='cuda:1'), in_proj_covar=tensor([0.1361, 0.1003, 0.1198, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 05:25:21,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-07 05:25:38,421 INFO [optim.py:369] (1/2) 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 05:25:51,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6115, 1.5673, 1.2251, 1.1938], device='cuda:1'), covar=tensor([0.0771, 0.0593, 0.0987, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0444, 0.0503, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 05:25:58,751 INFO [train.py:968] (1/2) Epoch 14, batch 11400, libri_loss[loss=0.3047, simple_loss=0.3768, pruned_loss=0.1163, over 29525.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3749, pruned_loss=0.1263, over 5653548.07 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3516, pruned_loss=0.09623, over 5721067.42 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3774, pruned_loss=0.1296, over 5637468.63 frames. ], batch size: 84, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:26:01,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3106, 1.5916, 1.2745, 1.1357], device='cuda:1'), covar=tensor([0.2634, 0.2544, 0.2878, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1365, 0.1008, 0.1202, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 05:26:27,400 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:26:30,114 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 11450, giga_loss[loss=0.2959, simple_loss=0.3569, pruned_loss=0.1175, over 28778.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3753, pruned_loss=0.1274, over 5651668.36 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3516, pruned_loss=0.09631, over 5722765.55 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3775, pruned_loss=0.1302, over 5636945.09 frames. ], batch size: 119, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:26:53,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7001, 1.7928, 1.6736, 1.6235], device='cuda:1'), covar=tensor([0.1695, 0.2321, 0.2272, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0744, 0.0690, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:27:00,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:27:07,435 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 05:27:13,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4096, 1.5605, 1.5943, 1.4882], device='cuda:1'), covar=tensor([0.1279, 0.1468, 0.1477, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0744, 0.0690, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:27:17,368 INFO [optim.py:369] (1/2) 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 05:27:29,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:27:32,854 INFO [train.py:968] (1/2) Epoch 14, batch 11500, giga_loss[loss=0.3317, simple_loss=0.3829, pruned_loss=0.1402, over 28978.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3743, pruned_loss=0.1261, over 5667566.19 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3515, pruned_loss=0.09619, over 5729279.20 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.377, pruned_loss=0.1296, over 5647607.14 frames. ], batch size: 106, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:28:23,308 INFO [train.py:968] (1/2) Epoch 14, batch 11550, giga_loss[loss=0.3163, simple_loss=0.3757, pruned_loss=0.1285, over 28554.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3752, pruned_loss=0.1265, over 5660614.75 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3516, pruned_loss=0.09635, over 5731762.17 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3778, pruned_loss=0.1298, over 5640842.43 frames. ], batch size: 92, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:28:41,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3502, 1.9144, 1.4194, 1.4707], device='cuda:1'), covar=tensor([0.0766, 0.0283, 0.0312, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0094], device='cuda:1')
+2023-03-07 05:28:54,511 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 11600, giga_loss[loss=0.299, simple_loss=0.3679, pruned_loss=0.115, over 28555.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3752, pruned_loss=0.126, over 5674683.85 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3515, pruned_loss=0.09626, over 5733675.89 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3777, pruned_loss=0.1292, over 5656312.78 frames. ], batch size: 85, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:29:50,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=604939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:29:54,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=604942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:30:04,109 INFO [train.py:968] (1/2) Epoch 14, batch 11650, giga_loss[loss=0.3957, simple_loss=0.427, pruned_loss=0.1822, over 27567.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3779, pruned_loss=0.1286, over 5651610.46 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3513, pruned_loss=0.09625, over 5726866.20 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3804, pruned_loss=0.1316, over 5641821.65 frames. ], batch size: 472, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:30:06,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=604953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:30:13,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4866, 3.6967, 1.5913, 1.7484], device='cuda:1'), covar=tensor([0.0951, 0.0332, 0.0833, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0530, 0.0354, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 05:30:22,970 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=604971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:30:31,753 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 05:30:33,436 INFO [optim.py:369] (1/2) 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 05:30:52,202 INFO [train.py:968] (1/2) Epoch 14, batch 11700, giga_loss[loss=0.3155, simple_loss=0.3752, pruned_loss=0.1279, over 28658.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3789, pruned_loss=0.1291, over 5649439.27 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3511, pruned_loss=0.09605, over 5721844.04 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3819, pruned_loss=0.1326, over 5643960.21 frames. ], batch size: 307, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:30:54,143 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:31:38,327 INFO [train.py:968] (1/2) Epoch 14, batch 11750, giga_loss[loss=0.3104, simple_loss=0.3744, pruned_loss=0.1233, over 28843.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3788, pruned_loss=0.1295, over 5650809.39 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3512, pruned_loss=0.09613, over 5721701.17 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3816, pruned_loss=0.1329, over 5645416.93 frames. ], batch size: 213, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:31:50,615 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:32:07,479 INFO [optim.py:369] (1/2) 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 05:32:21,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:32:24,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:32:24,328 INFO [train.py:968] (1/2) Epoch 14, batch 11800, giga_loss[loss=0.2999, simple_loss=0.3704, pruned_loss=0.1147, over 28290.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3795, pruned_loss=0.1289, over 5645484.26 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3519, pruned_loss=0.09673, over 5716278.29 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3818, pruned_loss=0.1318, over 5644510.71 frames. ], batch size: 368, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:32:25,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 05:32:50,979 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:33:12,815 INFO [train.py:968] (1/2) Epoch 14, batch 11850, giga_loss[loss=0.3224, simple_loss=0.3887, pruned_loss=0.1281, over 28683.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3795, pruned_loss=0.128, over 5649824.39 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3518, pruned_loss=0.0966, over 5719941.43 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3819, pruned_loss=0.1311, over 5644709.11 frames. ], batch size: 262, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:33:37,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0803, 1.0007, 3.9100, 3.2548], device='cuda:1'), covar=tensor([0.2326, 0.3292, 0.0854, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0611, 0.0894, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 05:33:42,699 INFO [optim.py:369] (1/2) 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 05:34:01,216 INFO [train.py:968] (1/2) Epoch 14, batch 11900, giga_loss[loss=0.2895, simple_loss=0.3625, pruned_loss=0.1083, over 28730.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3781, pruned_loss=0.127, over 5654513.33 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3517, pruned_loss=0.09649, over 5723739.67 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3807, pruned_loss=0.13, over 5645846.44 frames. ], batch size: 262, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:34:24,997 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605227.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:34:28,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 05:34:46,200 INFO [train.py:968] (1/2) Epoch 14, batch 11950, giga_loss[loss=0.264, simple_loss=0.3431, pruned_loss=0.0924, over 28972.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3766, pruned_loss=0.1262, over 5656561.36 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3519, pruned_loss=0.09662, over 5726209.59 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3788, pruned_loss=0.129, over 5646847.55 frames. ], batch size: 164, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:35:20,751 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 12000, giga_loss[loss=0.3308, simple_loss=0.3858, pruned_loss=0.1378, over 28909.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3779, pruned_loss=0.1273, over 5653028.58 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3518, pruned_loss=0.09658, over 5724476.94 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.38, pruned_loss=0.1298, over 5646224.43 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:35:39,160 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 05:35:47,545 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 05:35:51,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2026, 1.7420, 1.3212, 0.3328], device='cuda:1'), covar=tensor([0.2962, 0.1932, 0.2928, 0.4408], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1540, 0.1515, 0.1330], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 05:36:34,012 INFO [train.py:968] (1/2) Epoch 14, batch 12050, giga_loss[loss=0.3134, simple_loss=0.3764, pruned_loss=0.1252, over 28836.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3786, pruned_loss=0.1281, over 5653498.81 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.352, pruned_loss=0.09673, over 5728148.95 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3807, pruned_loss=0.1306, over 5643378.01 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:36:47,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 05:36:50,056 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:37:00,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:37:06,216 INFO [optim.py:369] (1/2) 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 05:37:22,837 INFO [train.py:968] (1/2) Epoch 14, batch 12100, libri_loss[loss=0.2605, simple_loss=0.3422, pruned_loss=0.08936, over 29535.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3763, pruned_loss=0.1266, over 5668296.99 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3519, pruned_loss=0.09669, over 5729723.30 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3786, pruned_loss=0.1293, over 5657154.95 frames. ], batch size: 75, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:37:25,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-07 05:37:49,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4026, 1.5999, 1.6635, 1.4226], device='cuda:1'), covar=tensor([0.1499, 0.1699, 0.1847, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0739, 0.0691, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:37:56,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:38:06,694 INFO [train.py:968] (1/2) Epoch 14, batch 12150, giga_loss[loss=0.3346, simple_loss=0.3866, pruned_loss=0.1413, over 27861.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3768, pruned_loss=0.127, over 5674370.68 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3523, pruned_loss=0.09701, over 5734104.63 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3793, pruned_loss=0.1302, over 5658562.52 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:38:39,716 INFO [optim.py:369] (1/2) 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 05:38:55,674 INFO [train.py:968] (1/2) Epoch 14, batch 12200, giga_loss[loss=0.3333, simple_loss=0.3959, pruned_loss=0.1354, over 28534.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3778, pruned_loss=0.1275, over 5676473.93 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3523, pruned_loss=0.09695, over 5737198.35 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3804, pruned_loss=0.1308, over 5659856.85 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:39:05,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 05:39:15,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:39:19,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:39:45,135 INFO [train.py:968] (1/2) Epoch 14, batch 12250, giga_loss[loss=0.2818, simple_loss=0.3556, pruned_loss=0.104, over 28937.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3783, pruned_loss=0.1284, over 5668150.96 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3521, pruned_loss=0.09682, over 5738928.06 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3807, pruned_loss=0.1313, over 5653188.97 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:39:47,866 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:40:15,894 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:40:16,192 INFO [optim.py:369] (1/2) 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 05:40:18,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:40:32,443 INFO [train.py:968] (1/2) Epoch 14, batch 12300, giga_loss[loss=0.2985, simple_loss=0.3657, pruned_loss=0.1157, over 28838.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3768, pruned_loss=0.1262, over 5684429.10 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3519, pruned_loss=0.09653, over 5742540.49 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3796, pruned_loss=0.1296, over 5667654.15 frames. ], batch size: 186, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:40:35,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605602.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:40:45,534 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:41:22,185 INFO [train.py:968] (1/2) Epoch 14, batch 12350, giga_loss[loss=0.3109, simple_loss=0.3769, pruned_loss=0.1224, over 28979.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3759, pruned_loss=0.1251, over 5673738.82 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3521, pruned_loss=0.09653, over 5744814.34 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3783, pruned_loss=0.1282, over 5657913.32 frames. ], batch size: 164, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:41:50,781 INFO [optim.py:369] (1/2) 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 05:42:03,601 INFO [train.py:968] (1/2) Epoch 14, batch 12400, giga_loss[loss=0.3258, simple_loss=0.3924, pruned_loss=0.1297, over 28546.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3745, pruned_loss=0.1232, over 5686978.23 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3511, pruned_loss=0.09589, over 5751485.69 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3782, pruned_loss=0.1273, over 5665701.37 frames. ], batch size: 307, lr: 2.32e-03, grad_scale: 8.0
+2023-03-07 05:42:45,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=605742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:42:49,748 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605745.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:42:53,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=605748.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:42:54,854 INFO [train.py:968] (1/2) Epoch 14, batch 12450, giga_loss[loss=0.3059, simple_loss=0.3683, pruned_loss=0.1218, over 28880.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3741, pruned_loss=0.1231, over 5680299.37 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3514, pruned_loss=0.0962, over 5744712.04 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3771, pruned_loss=0.1265, over 5667723.95 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:43:21,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:43:29,202 INFO [optim.py:369] (1/2) 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 05:43:43,582 INFO [train.py:968] (1/2) Epoch 14, batch 12500, giga_loss[loss=0.2844, simple_loss=0.356, pruned_loss=0.1064, over 28914.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3736, pruned_loss=0.1234, over 5682282.96 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3512, pruned_loss=0.09606, over 5747252.95 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3764, pruned_loss=0.1266, over 5669174.49 frames. ], batch size: 186, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:44:08,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7048, 1.8835, 1.9661, 1.4818], device='cuda:1'), covar=tensor([0.1673, 0.2300, 0.1296, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0852, 0.0700, 0.0897, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 05:44:26,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3241, 1.7842, 1.4589, 1.5954], device='cuda:1'), covar=tensor([0.0761, 0.0290, 0.0298, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 05:44:31,335 INFO [train.py:968] (1/2) Epoch 14, batch 12550, giga_loss[loss=0.3819, simple_loss=0.4135, pruned_loss=0.1752, over 27490.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3723, pruned_loss=0.1234, over 5678712.24 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3513, pruned_loss=0.09612, over 5751312.91 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1266, over 5662807.45 frames. ], batch size: 472, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:45:06,092 INFO [optim.py:369] (1/2) 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 05:45:07,432 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=605885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:45:09,643 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 12600, giga_loss[loss=0.3104, simple_loss=0.3737, pruned_loss=0.1235, over 28617.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3696, pruned_loss=0.1227, over 5689757.35 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3513, pruned_loss=0.09628, over 5753471.25 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.372, pruned_loss=0.1254, over 5674409.83 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:45:39,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1105, 1.9943, 1.5413, 1.7157], device='cuda:1'), covar=tensor([0.0838, 0.0747, 0.1034, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0444, 0.0504, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 05:45:39,895 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=605917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:45:50,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=605930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:46:09,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5685, 4.8683, 1.9332, 1.7539], device='cuda:1'), covar=tensor([0.0963, 0.0272, 0.0795, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0527, 0.0353, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 05:46:12,024 INFO [train.py:968] (1/2) Epoch 14, batch 12650, giga_loss[loss=0.3255, simple_loss=0.381, pruned_loss=0.135, over 28863.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.368, pruned_loss=0.1221, over 5694626.54 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3513, pruned_loss=0.09625, over 5754275.89 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.37, pruned_loss=0.1244, over 5681520.51 frames. ], batch size: 199, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:46:45,244 INFO [optim.py:369] (1/2) 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 05:47:01,092 INFO [train.py:968] (1/2) Epoch 14, batch 12700, giga_loss[loss=0.2884, simple_loss=0.3659, pruned_loss=0.1055, over 27864.00 frames. ], tot_loss[loss=0.306, simple_loss=0.368, pruned_loss=0.122, over 5683525.18 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3517, pruned_loss=0.09649, over 5748465.73 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3696, pruned_loss=0.1242, over 5677564.06 frames. ], batch size: 412, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:47:13,205 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-07 05:47:51,561 INFO [train.py:968] (1/2) Epoch 14, batch 12750, giga_loss[loss=0.282, simple_loss=0.3608, pruned_loss=0.1016, over 28516.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3666, pruned_loss=0.119, over 5679961.81 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3515, pruned_loss=0.09648, over 5749285.48 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3682, pruned_loss=0.1211, over 5673648.63 frames. ], batch size: 336, lr: 2.32e-03, grad_scale: 2.0
+2023-03-07 05:47:52,337 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:48:38,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 05:48:39,087 INFO [train.py:968] (1/2) Epoch 14, batch 12800, giga_loss[loss=0.2413, simple_loss=0.3247, pruned_loss=0.07898, over 28953.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3638, pruned_loss=0.1151, over 5672139.01 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3516, pruned_loss=0.09671, over 5750266.76 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3655, pruned_loss=0.1172, over 5663890.45 frames. ], batch size: 164, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:48:51,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-07 05:49:03,287 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606123.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:49:27,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3094, 1.3128, 3.8820, 3.1329], device='cuda:1'), covar=tensor([0.1655, 0.2677, 0.0462, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0701, 0.0616, 0.0901, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 05:49:29,031 INFO [train.py:968] (1/2) Epoch 14, batch 12850, giga_loss[loss=0.256, simple_loss=0.3365, pruned_loss=0.08777, over 29037.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3609, pruned_loss=0.112, over 5672916.35 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3516, pruned_loss=0.09692, over 5752525.33 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3624, pruned_loss=0.1139, over 5662551.28 frames. ], batch size: 128, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:49:49,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2813, 3.2047, 1.5028, 1.4694], device='cuda:1'), covar=tensor([0.0996, 0.0313, 0.0944, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0526, 0.0352, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 05:50:00,129 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:50:19,931 INFO [train.py:968] (1/2) Epoch 14, batch 12900, giga_loss[loss=0.3149, simple_loss=0.3758, pruned_loss=0.127, over 28729.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3579, pruned_loss=0.1091, over 5662780.03 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3511, pruned_loss=0.09669, over 5747271.52 frames. ], giga_tot_loss[loss=0.2911, simple_loss=0.3598, pruned_loss=0.1111, over 5657721.68 frames. ], batch size: 284, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:50:44,406 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 12950, giga_loss[loss=0.2743, simple_loss=0.355, pruned_loss=0.09679, over 28943.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3545, pruned_loss=0.1056, over 5662130.49 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3506, pruned_loss=0.09668, over 5739935.46 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3566, pruned_loss=0.1075, over 5663028.54 frames. ], batch size: 227, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:51:37,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0992, 1.5167, 1.5222, 1.3289], device='cuda:1'), covar=tensor([0.1640, 0.1423, 0.1748, 0.1550], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0724, 0.0680, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:51:42,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4933, 1.7699, 1.7518, 1.2896], device='cuda:1'), covar=tensor([0.1821, 0.2663, 0.1521, 0.1823], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0692, 0.0891, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 05:51:43,597 INFO [optim.py:369] (1/2) 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 05:52:00,267 INFO [train.py:968] (1/2) Epoch 14, batch 13000, giga_loss[loss=0.3344, simple_loss=0.3902, pruned_loss=0.1394, over 28616.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3535, pruned_loss=0.1032, over 5663209.00 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3497, pruned_loss=0.09635, over 5744720.26 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.356, pruned_loss=0.1052, over 5657589.61 frames. ], batch size: 307, lr: 2.32e-03, grad_scale: 4.0
+2023-03-07 05:52:04,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:52:38,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-07 05:52:40,182 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606341.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:52:47,123 INFO [train.py:968] (1/2) Epoch 14, batch 13050, giga_loss[loss=0.2583, simple_loss=0.3372, pruned_loss=0.08972, over 28923.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.354, pruned_loss=0.1036, over 5664386.23 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3489, pruned_loss=0.0959, over 5749120.54 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5653086.58 frames. ], batch size: 213, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:53:04,497 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8903, 1.3391, 1.3550, 1.1421], device='cuda:1'), covar=tensor([0.1616, 0.1033, 0.1890, 0.1475], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0722, 0.0679, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:53:20,925 INFO [optim.py:369] (1/2) 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 05:53:34,397 INFO [train.py:968] (1/2) Epoch 14, batch 13100, giga_loss[loss=0.2598, simple_loss=0.3408, pruned_loss=0.08937, over 28848.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3522, pruned_loss=0.1024, over 5671537.56 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3482, pruned_loss=0.09573, over 5754391.48 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3553, pruned_loss=0.1046, over 5655411.19 frames. ], batch size: 285, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:54:01,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606425.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:54:24,407 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:54:25,370 INFO [train.py:968] (1/2) Epoch 14, batch 13150, giga_loss[loss=0.252, simple_loss=0.334, pruned_loss=0.08501, over 29016.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3491, pruned_loss=0.1001, over 5675061.82 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3476, pruned_loss=0.09531, over 5757315.45 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3522, pruned_loss=0.1024, over 5657585.92 frames. ], batch size: 128, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:54:26,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:54:30,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2879, 3.1371, 1.4749, 1.3863], device='cuda:1'), covar=tensor([0.0965, 0.0350, 0.0942, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0525, 0.0353, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 05:54:32,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1647, 1.3392, 3.3500, 2.9427], device='cuda:1'), covar=tensor([0.1577, 0.2512, 0.0502, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.0692, 0.0609, 0.0889, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 05:54:41,774 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-07 05:54:57,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:55:02,129 INFO [optim.py:369] (1/2) 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 05:55:09,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 05:55:14,236 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 13200, giga_loss[loss=0.2691, simple_loss=0.3452, pruned_loss=0.09652, over 28555.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3475, pruned_loss=0.09917, over 5675285.57 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3474, pruned_loss=0.09527, over 5759075.49 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3501, pruned_loss=0.1011, over 5658622.69 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:55:35,490 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:55:53,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606537.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:55:59,810 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 05:56:05,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8062, 2.1939, 1.9384, 1.7805], device='cuda:1'), covar=tensor([0.1549, 0.1936, 0.1733, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0722, 0.0681, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 05:56:05,789 INFO [train.py:968] (1/2) Epoch 14, batch 13250, giga_loss[loss=0.261, simple_loss=0.3426, pruned_loss=0.08966, over 28909.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3471, pruned_loss=0.09846, over 5667908.29 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3474, pruned_loss=0.09544, over 5751940.21 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3492, pruned_loss=0.09987, over 5659327.73 frames. ], batch size: 227, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:56:14,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:56:25,625 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606568.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:56:27,639 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606571.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:56:41,396 INFO [optim.py:369] (1/2) 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 05:56:49,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:56:53,655 INFO [train.py:968] (1/2) Epoch 14, batch 13300, giga_loss[loss=0.2743, simple_loss=0.3531, pruned_loss=0.0978, over 28759.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3458, pruned_loss=0.09725, over 5666856.14 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.347, pruned_loss=0.09527, over 5747844.87 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3479, pruned_loss=0.09862, over 5660167.17 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:56:53,976 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:57:34,675 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606641.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 05:57:38,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606644.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:57:43,726 INFO [train.py:968] (1/2) Epoch 14, batch 13350, giga_loss[loss=0.2323, simple_loss=0.3175, pruned_loss=0.07349, over 28956.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3423, pruned_loss=0.09441, over 5672965.54 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3465, pruned_loss=0.09504, over 5752446.27 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3443, pruned_loss=0.09574, over 5661358.44 frames. ], batch size: 145, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:58:08,828 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606673.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:58:20,987 INFO [optim.py:369] (1/2) 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 05:58:32,430 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=606695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:58:35,831 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 13400, giga_loss[loss=0.233, simple_loss=0.3153, pruned_loss=0.07531, over 28609.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3383, pruned_loss=0.09237, over 5666891.62 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3456, pruned_loss=0.09467, over 5754427.72 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3405, pruned_loss=0.09373, over 5654059.53 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 05:58:37,976 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:58:49,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4822, 1.7865, 1.7519, 1.3168], device='cuda:1'), covar=tensor([0.1842, 0.2468, 0.1517, 0.1823], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0687, 0.0892, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 05:58:52,556 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606716.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 05:59:06,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:59:16,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:59:19,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:59:26,718 INFO [train.py:968] (1/2) Epoch 14, batch 13450, giga_loss[loss=0.2642, simple_loss=0.35, pruned_loss=0.08918, over 28772.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.338, pruned_loss=0.09307, over 5648759.59 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3459, pruned_loss=0.09505, over 5750162.34 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3392, pruned_loss=0.09372, over 5637861.38 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 05:59:30,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3766, 1.9290, 1.5268, 1.5064], device='cuda:1'), covar=tensor([0.0737, 0.0313, 0.0324, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 05:59:47,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 05:59:48,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 06:00:03,149 INFO [optim.py:369] (1/2) 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 06:00:17,919 INFO [train.py:968] (1/2) Epoch 14, batch 13500, giga_loss[loss=0.2912, simple_loss=0.3736, pruned_loss=0.1043, over 28618.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3375, pruned_loss=0.09359, over 5644409.98 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3456, pruned_loss=0.09498, over 5742251.20 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3387, pruned_loss=0.09413, over 5641690.46 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:00:39,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5600, 4.3522, 4.1350, 1.9796], device='cuda:1'), covar=tensor([0.0538, 0.0727, 0.0810, 0.2311], device='cuda:1'), in_proj_covar=tensor([0.1087, 0.1012, 0.0880, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 06:01:18,599 INFO [train.py:968] (1/2) Epoch 14, batch 13550, giga_loss[loss=0.2948, simple_loss=0.376, pruned_loss=0.1068, over 28333.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3395, pruned_loss=0.09476, over 5636554.56 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.345, pruned_loss=0.09482, over 5743937.34 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3408, pruned_loss=0.09535, over 5630249.33 frames. ], batch size: 368, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:01:26,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=606859.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:01:31,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=606862.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:02:01,877 INFO [optim.py:369] (1/2) 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 06:02:06,622 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=606891.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:02:07,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:02:14,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 06:02:15,788 INFO [train.py:968] (1/2) Epoch 14, batch 13600, giga_loss[loss=0.2537, simple_loss=0.3374, pruned_loss=0.08498, over 28947.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3421, pruned_loss=0.09518, over 5644354.25 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3441, pruned_loss=0.09437, over 5746123.89 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3439, pruned_loss=0.09606, over 5636040.32 frames. ], batch size: 213, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:02:33,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=606912.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:02:54,880 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-07 06:03:19,067 INFO [train.py:968] (1/2) Epoch 14, batch 13650, giga_loss[loss=0.2414, simple_loss=0.3311, pruned_loss=0.0759, over 29011.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3438, pruned_loss=0.09645, over 5642915.73 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3441, pruned_loss=0.09442, over 5747907.02 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3453, pruned_loss=0.09713, over 5633824.43 frames. ], batch size: 155, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:04:06,471 INFO [optim.py:369] (1/2) 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 06:04:20,200 INFO [train.py:968] (1/2) Epoch 14, batch 13700, giga_loss[loss=0.21, simple_loss=0.3019, pruned_loss=0.05906, over 28760.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3415, pruned_loss=0.09493, over 5650373.39 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3437, pruned_loss=0.09426, over 5749969.90 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3429, pruned_loss=0.09563, over 5639406.96 frames. ], batch size: 119, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:04:32,797 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 06:05:01,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=607032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:07,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:10,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:23,302 INFO [train.py:968] (1/2) Epoch 14, batch 13750, giga_loss[loss=0.2419, simple_loss=0.3281, pruned_loss=0.07781, over 28956.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3408, pruned_loss=0.09364, over 5647070.75 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3437, pruned_loss=0.09423, over 5749782.35 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3419, pruned_loss=0.09423, over 5637557.85 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:05:30,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607055.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:05:33,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607058.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:05:45,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:05:47,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:06:04,252 INFO [optim.py:369] (1/2) 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 06:06:04,948 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 13800, giga_loss[loss=0.2032, simple_loss=0.2754, pruned_loss=0.06549, over 24425.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3387, pruned_loss=0.09176, over 5650614.06 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.343, pruned_loss=0.09392, over 5751646.72 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3402, pruned_loss=0.09247, over 5637121.36 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:07:23,744 INFO [train.py:968] (1/2) Epoch 14, batch 13850, giga_loss[loss=0.235, simple_loss=0.31, pruned_loss=0.07998, over 28965.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3349, pruned_loss=0.09057, over 5655514.36 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3425, pruned_loss=0.09373, over 5753613.41 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3364, pruned_loss=0.09125, over 5641802.61 frames. ], batch size: 284, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:08:06,359 INFO [optim.py:369] (1/2) 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 06:08:21,866 INFO [train.py:968] (1/2) Epoch 14, batch 13900, libri_loss[loss=0.2769, simple_loss=0.3484, pruned_loss=0.1027, over 29136.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3345, pruned_loss=0.0904, over 5663025.98 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3421, pruned_loss=0.09351, over 5756739.91 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3359, pruned_loss=0.09105, over 5647285.02 frames. ], batch size: 101, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:08:23,902 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=607202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:08:36,679 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:08:37,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1933, 2.6218, 1.2584, 1.3403], device='cuda:1'), covar=tensor([0.0976, 0.0313, 0.0939, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0524, 0.0355, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 06:08:39,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:09:09,588 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 13950, giga_loss[loss=0.2114, simple_loss=0.2859, pruned_loss=0.06844, over 24343.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3335, pruned_loss=0.08953, over 5667258.96 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3412, pruned_loss=0.09301, over 5753476.99 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3351, pruned_loss=0.09041, over 5653442.42 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:10:00,811 INFO [optim.py:369] (1/2) 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 06:10:17,039 INFO [train.py:968] (1/2) Epoch 14, batch 14000, giga_loss[loss=0.245, simple_loss=0.3344, pruned_loss=0.07785, over 28694.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3358, pruned_loss=0.09006, over 5672329.16 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.341, pruned_loss=0.09293, over 5752640.45 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3372, pruned_loss=0.09076, over 5660597.69 frames. ], batch size: 243, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:11:20,326 INFO [train.py:968] (1/2) Epoch 14, batch 14050, giga_loss[loss=0.2364, simple_loss=0.3196, pruned_loss=0.07655, over 28170.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3365, pruned_loss=0.08982, over 5676798.76 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3407, pruned_loss=0.09276, over 5752646.49 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3378, pruned_loss=0.09047, over 5665683.82 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:12:10,795 INFO [optim.py:369] (1/2) 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 06:12:21,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4419, 1.6035, 1.5099, 1.4017], device='cuda:1'), covar=tensor([0.1863, 0.1517, 0.1251, 0.1492], device='cuda:1'), in_proj_covar=tensor([0.1743, 0.1653, 0.1606, 0.1727], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 06:12:27,332 INFO [train.py:968] (1/2) Epoch 14, batch 14100, giga_loss[loss=0.1992, simple_loss=0.2683, pruned_loss=0.06504, over 24828.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3343, pruned_loss=0.08888, over 5671844.80 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3406, pruned_loss=0.09281, over 5745608.50 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3353, pruned_loss=0.08927, over 5667320.46 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:12:36,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:13:01,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5248, 1.7188, 1.6278, 1.5561], device='cuda:1'), covar=tensor([0.1577, 0.2172, 0.1765, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0710, 0.0668, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 06:13:01,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 06:13:04,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=607432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:13:20,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5522, 3.3932, 3.2024, 1.6801], device='cuda:1'), covar=tensor([0.0719, 0.0816, 0.0774, 0.2484], device='cuda:1'), in_proj_covar=tensor([0.1095, 0.1010, 0.0878, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 06:13:24,473 INFO [train.py:968] (1/2) Epoch 14, batch 14150, giga_loss[loss=0.2679, simple_loss=0.3363, pruned_loss=0.09971, over 24444.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09032, over 5672286.44 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3403, pruned_loss=0.09266, over 5752641.92 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3371, pruned_loss=0.09064, over 5659489.81 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:13:45,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1395, 1.4525, 1.4169, 1.2426], device='cuda:1'), covar=tensor([0.1486, 0.1531, 0.1875, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0710, 0.0667, 0.0650], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 06:14:12,815 INFO [optim.py:369] (1/2) 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 06:14:28,159 INFO [train.py:968] (1/2) Epoch 14, batch 14200, giga_loss[loss=0.271, simple_loss=0.3614, pruned_loss=0.09027, over 28704.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3398, pruned_loss=0.09071, over 5662580.75 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3399, pruned_loss=0.09242, over 5751853.93 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3408, pruned_loss=0.09115, over 5650160.09 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:15:28,745 INFO [train.py:968] (1/2) Epoch 14, batch 14250, libri_loss[loss=0.2989, simple_loss=0.3671, pruned_loss=0.1153, over 19521.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3424, pruned_loss=0.08997, over 5654378.91 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3399, pruned_loss=0.09245, over 5744181.76 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3432, pruned_loss=0.09025, over 5650401.82 frames. ], batch size: 187, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:15:29,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:15:31,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:16:02,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:16:09,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:16:15,740 INFO [optim.py:369] (1/2) 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 06:16:30,637 INFO [train.py:968] (1/2) Epoch 14, batch 14300, giga_loss[loss=0.2262, simple_loss=0.3244, pruned_loss=0.06401, over 28978.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3424, pruned_loss=0.08921, over 5650166.03 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3398, pruned_loss=0.0924, over 5747469.71 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08942, over 5642614.25 frames. ], batch size: 112, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:17:28,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-07 06:17:30,126 INFO [train.py:968] (1/2) Epoch 14, batch 14350, giga_loss[loss=0.2771, simple_loss=0.3557, pruned_loss=0.09926, over 28921.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3419, pruned_loss=0.08868, over 5663757.97 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3392, pruned_loss=0.0921, over 5751177.70 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.343, pruned_loss=0.08904, over 5652604.83 frames. ], batch size: 186, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:18:16,973 INFO [optim.py:369] (1/2) 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 06:18:34,295 INFO [train.py:968] (1/2) Epoch 14, batch 14400, giga_loss[loss=0.249, simple_loss=0.326, pruned_loss=0.08604, over 28930.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3426, pruned_loss=0.09066, over 5666885.68 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3392, pruned_loss=0.09215, over 5749819.65 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3437, pruned_loss=0.09088, over 5657652.79 frames. ], batch size: 112, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:19:00,456 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:19:03,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0626, 2.1092, 1.4717, 1.7363], device='cuda:1'), covar=tensor([0.0798, 0.0644, 0.0983, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0433, 0.0497, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 06:19:03,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:19:33,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4920, 1.8441, 1.6670, 1.4692], device='cuda:1'), covar=tensor([0.2410, 0.1551, 0.1719, 0.1931], device='cuda:1'), in_proj_covar=tensor([0.1744, 0.1652, 0.1600, 0.1726], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 06:19:36,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4733, 1.7543, 1.7114, 1.3060], device='cuda:1'), covar=tensor([0.1412, 0.1846, 0.1150, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0683, 0.0889, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 06:19:39,603 INFO [train.py:968] (1/2) Epoch 14, batch 14450, giga_loss[loss=0.2518, simple_loss=0.3339, pruned_loss=0.08483, over 28734.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3419, pruned_loss=0.0915, over 5658124.55 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3386, pruned_loss=0.09182, over 5744256.48 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3434, pruned_loss=0.09193, over 5653889.78 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:19:43,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:19:49,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3274, 1.6955, 1.5805, 1.4508], device='cuda:1'), covar=tensor([0.1551, 0.1772, 0.1895, 0.1821], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0710, 0.0666, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 06:20:28,908 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-07 06:20:42,155 INFO [optim.py:369] (1/2) 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 06:20:58,764 INFO [train.py:968] (1/2) Epoch 14, batch 14500, giga_loss[loss=0.2261, simple_loss=0.3095, pruned_loss=0.07139, over 28228.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3403, pruned_loss=0.09051, over 5670529.30 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3382, pruned_loss=0.09165, over 5746580.99 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3419, pruned_loss=0.09101, over 5663853.27 frames. ], batch size: 368, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:21:12,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=607807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:21:25,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-07 06:22:14,313 INFO [train.py:968] (1/2) Epoch 14, batch 14550, giga_loss[loss=0.2391, simple_loss=0.322, pruned_loss=0.07814, over 28996.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3356, pruned_loss=0.0879, over 5667647.57 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3382, pruned_loss=0.09173, over 5748790.41 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3369, pruned_loss=0.08818, over 5658733.28 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:22:18,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1747, 1.5806, 1.1729, 0.3918], device='cuda:1'), covar=tensor([0.2935, 0.1619, 0.2546, 0.4557], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1530, 0.1514, 0.1332], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 06:23:05,843 INFO [optim.py:369] (1/2) 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 06:23:16,767 INFO [train.py:968] (1/2) Epoch 14, batch 14600, giga_loss[loss=0.3086, simple_loss=0.365, pruned_loss=0.1262, over 26852.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3348, pruned_loss=0.088, over 5653143.06 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.338, pruned_loss=0.09169, over 5733764.54 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3359, pruned_loss=0.08814, over 5657601.43 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:24:23,937 INFO [train.py:968] (1/2) Epoch 14, batch 14650, giga_loss[loss=0.2427, simple_loss=0.3183, pruned_loss=0.08349, over 28628.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08819, over 5658306.08 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3379, pruned_loss=0.09165, over 5726083.11 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3354, pruned_loss=0.08828, over 5667669.91 frames. ], batch size: 92, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:24:24,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=607950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:24:29,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=607953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:25:03,047 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=607982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:25:05,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-07 06:25:12,247 INFO [optim.py:369] (1/2) 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 06:25:25,771 INFO [train.py:968] (1/2) Epoch 14, batch 14700, libri_loss[loss=0.2579, simple_loss=0.3288, pruned_loss=0.09354, over 29506.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3386, pruned_loss=0.09008, over 5667061.70 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09166, over 5729922.82 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3394, pruned_loss=0.09007, over 5669406.19 frames. ], batch size: 84, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:26:26,922 INFO [train.py:968] (1/2) Epoch 14, batch 14750, giga_loss[loss=0.316, simple_loss=0.365, pruned_loss=0.1335, over 26883.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3379, pruned_loss=0.09096, over 5673680.67 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09166, over 5733615.52 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3386, pruned_loss=0.09094, over 5671123.89 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:26:55,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1215, 1.5395, 1.4360, 1.0319], device='cuda:1'), covar=tensor([0.1464, 0.2522, 0.1335, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0683, 0.0887, 0.0790], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 06:27:18,585 INFO [optim.py:369] (1/2) 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 06:27:33,533 INFO [train.py:968] (1/2) Epoch 14, batch 14800, giga_loss[loss=0.316, simple_loss=0.3771, pruned_loss=0.1275, over 27042.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3399, pruned_loss=0.09351, over 5660656.35 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3379, pruned_loss=0.09177, over 5728695.24 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3403, pruned_loss=0.09338, over 5660891.77 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:28:02,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6002, 1.8655, 1.8346, 1.6486], device='cuda:1'), covar=tensor([0.1589, 0.1804, 0.1923, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0712, 0.0667, 0.0649], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 06:28:21,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3333, 1.7830, 1.4181, 1.5085], device='cuda:1'), covar=tensor([0.0761, 0.0282, 0.0308, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 06:28:28,916 INFO [train.py:968] (1/2) Epoch 14, batch 14850, giga_loss[loss=0.2603, simple_loss=0.3351, pruned_loss=0.09272, over 28608.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.339, pruned_loss=0.09268, over 5663890.31 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3377, pruned_loss=0.09164, over 5728212.72 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3396, pruned_loss=0.09277, over 5662052.71 frames. ], batch size: 65, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:29:22,441 INFO [optim.py:369] (1/2) 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 06:29:35,112 INFO [train.py:968] (1/2) Epoch 14, batch 14900, giga_loss[loss=0.2772, simple_loss=0.3387, pruned_loss=0.1078, over 24460.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3397, pruned_loss=0.09193, over 5668332.69 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3375, pruned_loss=0.09165, over 5731720.66 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3404, pruned_loss=0.09201, over 5662147.29 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:29:56,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4007, 2.0265, 1.3377, 0.6622], device='cuda:1'), covar=tensor([0.4954, 0.2449, 0.3894, 0.5236], device='cuda:1'), in_proj_covar=tensor([0.1611, 0.1531, 0.1509, 0.1328], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 06:30:09,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4119, 1.4565, 1.1637, 1.5511], device='cuda:1'), covar=tensor([0.0756, 0.0312, 0.0347, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 06:30:10,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:30:27,270 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:30:49,244 INFO [train.py:968] (1/2) Epoch 14, batch 14950, giga_loss[loss=0.2629, simple_loss=0.3416, pruned_loss=0.09206, over 28931.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3406, pruned_loss=0.09189, over 5674153.10 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3371, pruned_loss=0.09143, over 5735391.71 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3416, pruned_loss=0.09217, over 5664826.76 frames. ], batch size: 145, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:31:45,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8816, 2.3171, 2.3242, 1.6823], device='cuda:1'), covar=tensor([0.1651, 0.2202, 0.1283, 0.1650], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0681, 0.0887, 0.0790], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 06:31:51,458 INFO [optim.py:369] (1/2) 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 06:32:06,096 INFO [train.py:968] (1/2) Epoch 14, batch 15000, giga_loss[loss=0.2286, simple_loss=0.3076, pruned_loss=0.07475, over 28748.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3371, pruned_loss=0.0901, over 5682591.69 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3367, pruned_loss=0.09127, over 5739513.19 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3383, pruned_loss=0.09045, over 5670119.79 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:32:06,096 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 06:32:14,461 INFO [train.py:1012] (1/2) Epoch 14, validation: loss=0.2017, simple_loss=0.3019, pruned_loss=0.05081, over 944034.00 frames. 
+2023-03-07 06:32:14,462 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 06:33:17,680 INFO [train.py:968] (1/2) Epoch 14, batch 15050, giga_loss[loss=0.2305, simple_loss=0.3098, pruned_loss=0.07563, over 28685.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3328, pruned_loss=0.08863, over 5696300.23 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3365, pruned_loss=0.09115, over 5744507.47 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3339, pruned_loss=0.08897, over 5680234.98 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:34:09,851 INFO [optim.py:369] (1/2) 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 06:34:20,631 INFO [train.py:968] (1/2) Epoch 14, batch 15100, giga_loss[loss=0.254, simple_loss=0.3347, pruned_loss=0.08671, over 28884.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.329, pruned_loss=0.08703, over 5688502.12 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3362, pruned_loss=0.09102, over 5744619.11 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.33, pruned_loss=0.08735, over 5674477.99 frames. ], batch size: 284, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:35:20,511 INFO [train.py:968] (1/2) Epoch 14, batch 15150, giga_loss[loss=0.2132, simple_loss=0.2829, pruned_loss=0.07172, over 24360.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3311, pruned_loss=0.08905, over 5683890.12 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3359, pruned_loss=0.09087, over 5748321.92 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3321, pruned_loss=0.08937, over 5668209.79 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:36:02,627 INFO [optim.py:369] (1/2) 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 06:36:10,144 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:36:14,409 INFO [train.py:968] (1/2) Epoch 14, batch 15200, libri_loss[loss=0.2591, simple_loss=0.3392, pruned_loss=0.08948, over 29304.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3313, pruned_loss=0.08899, over 5682453.02 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3358, pruned_loss=0.09086, over 5752018.58 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.332, pruned_loss=0.08921, over 5665021.25 frames. ], batch size: 94, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:36:28,881 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:37:14,299 INFO [train.py:968] (1/2) Epoch 14, batch 15250, giga_loss[loss=0.2321, simple_loss=0.3168, pruned_loss=0.07369, over 28149.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08647, over 5676199.71 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3359, pruned_loss=0.09084, over 5749952.75 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08659, over 5661942.39 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:37:44,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4659, 1.8754, 1.7979, 1.5985], device='cuda:1'), covar=tensor([0.1587, 0.1648, 0.1887, 0.1853], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0710, 0.0663, 0.0648], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 06:38:02,269 INFO [optim.py:369] (1/2) 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 06:38:11,393 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:38:11,709 INFO [train.py:968] (1/2) Epoch 14, batch 15300, giga_loss[loss=0.209, simple_loss=0.2764, pruned_loss=0.07081, over 24205.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3276, pruned_loss=0.08556, over 5676376.60 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.336, pruned_loss=0.09093, over 5754471.84 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3275, pruned_loss=0.08541, over 5658639.03 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:38:15,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:38:29,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:38:49,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 06:39:23,062 INFO [train.py:968] (1/2) Epoch 14, batch 15350, giga_loss[loss=0.2478, simple_loss=0.3285, pruned_loss=0.08356, over 28141.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3261, pruned_loss=0.08509, over 5682506.61 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3351, pruned_loss=0.09061, over 5759470.24 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3266, pruned_loss=0.08511, over 5661618.69 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:40:15,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6168, 2.2414, 1.5759, 0.7381], device='cuda:1'), covar=tensor([0.4430, 0.2632, 0.3723, 0.5368], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1540, 0.1515, 0.1331], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 06:40:16,715 INFO [optim.py:369] (1/2) 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 06:40:30,266 INFO [train.py:968] (1/2) Epoch 14, batch 15400, giga_loss[loss=0.2402, simple_loss=0.3239, pruned_loss=0.07825, over 28967.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3271, pruned_loss=0.08494, over 5696449.97 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.335, pruned_loss=0.0907, over 5762659.31 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3274, pruned_loss=0.08473, over 5675758.55 frames. ], batch size: 199, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:41:24,206 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=608744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:41:27,964 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 15450, giga_loss[loss=0.2122, simple_loss=0.2887, pruned_loss=0.06785, over 28703.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3274, pruned_loss=0.08544, over 5694924.81 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3346, pruned_loss=0.09054, over 5755519.98 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3277, pruned_loss=0.08528, over 5682426.79 frames. ], batch size: 92, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:41:32,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=608752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:41:36,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=608755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:42:02,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608776.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:42:02,328 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=608776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:42:13,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=608784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:42:24,333 INFO [optim.py:369] (1/2) 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 06:42:33,830 INFO [train.py:968] (1/2) Epoch 14, batch 15500, giga_loss[loss=0.2106, simple_loss=0.3012, pruned_loss=0.06, over 28604.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3271, pruned_loss=0.08575, over 5692251.35 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3338, pruned_loss=0.0901, over 5754836.39 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.328, pruned_loss=0.08594, over 5681324.05 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:43:03,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0204, 4.8761, 4.5359, 2.1689], device='cuda:1'), covar=tensor([0.0386, 0.0505, 0.0634, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.1080, 0.1002, 0.0871, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 06:43:37,078 INFO [train.py:968] (1/2) Epoch 14, batch 15550, giga_loss[loss=0.2418, simple_loss=0.3405, pruned_loss=0.07155, over 28936.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.08463, over 5686894.57 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3338, pruned_loss=0.09003, over 5757371.50 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3272, pruned_loss=0.08474, over 5674549.48 frames. ], batch size: 145, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 06:44:02,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:44:04,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:44:20,330 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:44:36,178 INFO [train.py:968] (1/2) Epoch 14, batch 15600, giga_loss[loss=0.2404, simple_loss=0.3341, pruned_loss=0.0734, over 28576.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.329, pruned_loss=0.08492, over 5669533.12 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3339, pruned_loss=0.09004, over 5760756.73 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3292, pruned_loss=0.08489, over 5655357.60 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:45:18,570 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-07 06:45:39,234 INFO [train.py:968] (1/2) Epoch 14, batch 15650, giga_loss[loss=0.2494, simple_loss=0.338, pruned_loss=0.08035, over 28650.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3308, pruned_loss=0.0855, over 5669956.08 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3333, pruned_loss=0.08974, over 5763361.79 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3314, pruned_loss=0.08567, over 5654789.44 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:46:04,119 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=608971.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:46:06,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=608974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:46:13,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5421, 4.3508, 1.7481, 1.6267], device='cuda:1'), covar=tensor([0.0955, 0.0283, 0.0895, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0515, 0.0352, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 06:46:27,815 INFO [optim.py:369] (1/2) 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 06:46:36,427 INFO [train.py:968] (1/2) Epoch 14, batch 15700, giga_loss[loss=0.2246, simple_loss=0.2941, pruned_loss=0.07758, over 24606.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3324, pruned_loss=0.08648, over 5666947.19 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3338, pruned_loss=0.0901, over 5766317.85 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3325, pruned_loss=0.08621, over 5650493.58 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:46:54,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:46:57,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:15,461 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:18,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:33,230 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:35,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:36,341 INFO [train.py:968] (1/2) Epoch 14, batch 15750, giga_loss[loss=0.2493, simple_loss=0.336, pruned_loss=0.08128, over 28508.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3324, pruned_loss=0.08684, over 5655988.82 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3335, pruned_loss=0.09004, over 5761215.24 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3327, pruned_loss=0.08659, over 5643476.57 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:47:50,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:47:59,485 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4479, 1.6693, 1.3782, 1.5903], device='cuda:1'), covar=tensor([0.0770, 0.0300, 0.0335, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 06:48:27,447 INFO [optim.py:369] (1/2) 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 06:48:32,993 INFO [train.py:968] (1/2) Epoch 14, batch 15800, giga_loss[loss=0.2403, simple_loss=0.3254, pruned_loss=0.07759, over 28899.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3289, pruned_loss=0.08472, over 5663795.91 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3332, pruned_loss=0.09004, over 5765142.02 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3293, pruned_loss=0.0844, over 5647349.61 frames. ], batch size: 213, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:48:55,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:48:59,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:49:26,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 06:49:33,577 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 15850, giga_loss[loss=0.2209, simple_loss=0.2975, pruned_loss=0.07217, over 28972.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3281, pruned_loss=0.08472, over 5672555.76 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3331, pruned_loss=0.08998, over 5767698.31 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3285, pruned_loss=0.08443, over 5655541.36 frames. ], batch size: 213, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:49:38,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609151.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:49:44,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-07 06:49:59,809 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:50:07,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-07 06:50:27,166 INFO [optim.py:369] (1/2) 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 06:50:35,533 INFO [train.py:968] (1/2) Epoch 14, batch 15900, giga_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09026, over 28686.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3275, pruned_loss=0.08463, over 5666164.17 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3331, pruned_loss=0.09001, over 5760055.79 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3277, pruned_loss=0.08429, over 5658643.05 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:51:36,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:51:37,517 INFO [train.py:968] (1/2) Epoch 14, batch 15950, libri_loss[loss=0.2389, simple_loss=0.3131, pruned_loss=0.0823, over 29630.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3305, pruned_loss=0.08614, over 5673073.43 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3332, pruned_loss=0.0901, over 5761624.77 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3305, pruned_loss=0.08568, over 5662818.26 frames. ], batch size: 73, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:52:23,642 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:52:34,293 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609294.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:52:36,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609297.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:52:41,159 INFO [train.py:968] (1/2) Epoch 14, batch 16000, giga_loss[loss=0.2748, simple_loss=0.3401, pruned_loss=0.1047, over 26904.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3316, pruned_loss=0.08738, over 5663482.09 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3331, pruned_loss=0.09006, over 5765439.80 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3316, pruned_loss=0.08697, over 5649915.50 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:53:13,540 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:53:37,935 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609346.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:53:42,792 INFO [train.py:968] (1/2) Epoch 14, batch 16050, giga_loss[loss=0.3002, simple_loss=0.3775, pruned_loss=0.1114, over 28724.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3325, pruned_loss=0.08791, over 5672628.48 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3324, pruned_loss=0.08972, over 5766935.01 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.333, pruned_loss=0.08782, over 5658850.60 frames. ], batch size: 262, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:54:14,022 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:54:30,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:54:30,876 INFO [optim.py:369] (1/2) 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 06:54:34,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:54:38,373 INFO [train.py:968] (1/2) Epoch 14, batch 16100, giga_loss[loss=0.233, simple_loss=0.3101, pruned_loss=0.078, over 28474.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3369, pruned_loss=0.09027, over 5653058.41 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3327, pruned_loss=0.08986, over 5758029.62 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3372, pruned_loss=0.09007, over 5646229.42 frames. ], batch size: 78, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:54:59,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1708, 1.0713, 3.7101, 3.1170], device='cuda:1'), covar=tensor([0.1721, 0.2873, 0.0433, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0682, 0.0604, 0.0875, 0.0788], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0009], device='cuda:1')
+2023-03-07 06:55:03,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:55:03,748 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 16150, giga_loss[loss=0.2686, simple_loss=0.3421, pruned_loss=0.09754, over 28755.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.338, pruned_loss=0.08987, over 5659830.01 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3328, pruned_loss=0.08991, over 5762230.09 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3383, pruned_loss=0.08967, over 5647834.51 frames. ], batch size: 99, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 06:56:29,231 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609489.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:56:34,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609492.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 06:56:35,032 INFO [optim.py:369] (1/2) 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 06:56:44,028 INFO [train.py:968] (1/2) Epoch 14, batch 16200, giga_loss[loss=0.32, simple_loss=0.364, pruned_loss=0.138, over 26879.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3387, pruned_loss=0.09118, over 5645601.06 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3329, pruned_loss=0.09011, over 5754386.50 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3389, pruned_loss=0.09085, over 5640101.88 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:57:10,220 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609521.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 06:57:39,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:57:44,891 INFO [train.py:968] (1/2) Epoch 14, batch 16250, giga_loss[loss=0.2893, simple_loss=0.3531, pruned_loss=0.1128, over 28036.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3364, pruned_loss=0.09006, over 5652683.17 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3327, pruned_loss=0.08997, over 5748183.85 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.337, pruned_loss=0.08994, over 5650978.29 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:58:07,419 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:58:13,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:58:43,150 INFO [optim.py:369] (1/2) 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 06:58:53,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 06:58:54,602 INFO [train.py:968] (1/2) Epoch 14, batch 16300, giga_loss[loss=0.2556, simple_loss=0.3342, pruned_loss=0.08855, over 28480.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3355, pruned_loss=0.08917, over 5662878.23 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3326, pruned_loss=0.08984, over 5751332.53 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.336, pruned_loss=0.0892, over 5656888.37 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 06:59:58,123 INFO [train.py:968] (1/2) Epoch 14, batch 16350, giga_loss[loss=0.2361, simple_loss=0.3192, pruned_loss=0.07652, over 29015.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3356, pruned_loss=0.08994, over 5665238.06 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3328, pruned_loss=0.08991, over 5752423.01 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3359, pruned_loss=0.0899, over 5658767.44 frames. ], batch size: 136, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:00:11,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:00:42,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:00:45,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:00:47,366 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 16400, giga_loss[loss=0.2443, simple_loss=0.3238, pruned_loss=0.08242, over 28962.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3336, pruned_loss=0.08978, over 5663060.63 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3328, pruned_loss=0.08999, over 5755709.36 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3338, pruned_loss=0.08967, over 5651692.86 frames. ], batch size: 186, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:01:19,338 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:01:42,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-07 07:01:52,912 INFO [train.py:968] (1/2) Epoch 14, batch 16450, giga_loss[loss=0.234, simple_loss=0.3255, pruned_loss=0.07128, over 28940.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3318, pruned_loss=0.08848, over 5656143.30 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3327, pruned_loss=0.08989, over 5750566.56 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3321, pruned_loss=0.08844, over 5648962.52 frames. ], batch size: 213, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:01:55,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=609752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:02:43,778 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-07 07:02:43,980 INFO [optim.py:369] (1/2) 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 07:02:51,314 INFO [train.py:968] (1/2) Epoch 14, batch 16500, giga_loss[loss=0.2596, simple_loss=0.3238, pruned_loss=0.09771, over 26763.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3312, pruned_loss=0.08723, over 5675139.50 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3324, pruned_loss=0.09002, over 5755232.01 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3316, pruned_loss=0.08704, over 5662546.38 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:02:55,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:02:57,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:03:04,331 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.80 vs. limit=5.0
+2023-03-07 07:03:32,536 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=609835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:03:34,150 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:03:49,429 INFO [train.py:968] (1/2) Epoch 14, batch 16550, giga_loss[loss=0.261, simple_loss=0.3493, pruned_loss=0.08634, over 28744.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3318, pruned_loss=0.08536, over 5682196.57 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3324, pruned_loss=0.08999, over 5758160.73 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3322, pruned_loss=0.08515, over 5668054.28 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:03:53,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2969, 0.7913, 0.8349, 1.4323], device='cuda:1'), covar=tensor([0.0747, 0.0366, 0.0371, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0061, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 07:04:11,092 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=609870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:04:41,372 INFO [optim.py:369] (1/2) 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 07:04:44,260 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=609895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:04:47,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=609898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:04:49,002 INFO [train.py:968] (1/2) Epoch 14, batch 16600, giga_loss[loss=0.2404, simple_loss=0.3119, pruned_loss=0.08449, over 24652.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3335, pruned_loss=0.08501, over 5674384.23 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3324, pruned_loss=0.08997, over 5750960.36 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3338, pruned_loss=0.08477, over 5667992.03 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:05:19,781 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 16650, giga_loss[loss=0.2773, simple_loss=0.3581, pruned_loss=0.09831, over 28968.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3334, pruned_loss=0.08479, over 5667803.83 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3323, pruned_loss=0.08993, over 5749265.60 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3337, pruned_loss=0.0846, over 5663419.47 frames. ], batch size: 186, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:06:46,650 INFO [optim.py:369] (1/2) 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 07:06:54,320 INFO [train.py:968] (1/2) Epoch 14, batch 16700, giga_loss[loss=0.2416, simple_loss=0.3221, pruned_loss=0.0805, over 28168.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3332, pruned_loss=0.08499, over 5665794.72 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3324, pruned_loss=0.08997, over 5752664.82 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3335, pruned_loss=0.08469, over 5657318.22 frames. ], batch size: 412, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:07:58,966 INFO [train.py:968] (1/2) Epoch 14, batch 16750, giga_loss[loss=0.3034, simple_loss=0.3706, pruned_loss=0.1181, over 28588.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3329, pruned_loss=0.08482, over 5661749.24 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3325, pruned_loss=0.09001, over 5754919.99 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.333, pruned_loss=0.08443, over 5651055.60 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:08:14,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-07 07:08:54,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2728, 2.8053, 1.4525, 1.3969], device='cuda:1'), covar=tensor([0.0938, 0.0342, 0.0903, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0516, 0.0353, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 07:09:00,561 INFO [optim.py:369] (1/2) 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 07:09:06,750 INFO [train.py:968] (1/2) Epoch 14, batch 16800, libri_loss[loss=0.2562, simple_loss=0.3188, pruned_loss=0.09673, over 29388.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3334, pruned_loss=0.08456, over 5660499.41 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.332, pruned_loss=0.08987, over 5748121.17 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.334, pruned_loss=0.08418, over 5655144.55 frames. ], batch size: 67, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:09:15,778 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 07:09:21,592 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=610120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:09:32,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6271, 1.8618, 1.6690, 1.6480], device='cuda:1'), covar=tensor([0.1513, 0.1945, 0.1939, 0.1765], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0711, 0.0662, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 07:10:10,573 INFO [train.py:968] (1/2) Epoch 14, batch 16850, giga_loss[loss=0.2812, simple_loss=0.3581, pruned_loss=0.1021, over 29014.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3357, pruned_loss=0.08632, over 5656453.19 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3323, pruned_loss=0.09006, over 5750687.00 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.336, pruned_loss=0.08568, over 5647049.56 frames. ], batch size: 128, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:11:15,323 INFO [optim.py:369] (1/2) 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 07:11:23,849 INFO [train.py:968] (1/2) Epoch 14, batch 16900, giga_loss[loss=0.2966, simple_loss=0.3592, pruned_loss=0.117, over 26849.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3389, pruned_loss=0.08744, over 5663882.10 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.332, pruned_loss=0.08992, over 5753293.02 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3395, pruned_loss=0.08703, over 5652855.13 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:11:37,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:12:20,967 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:12:25,463 INFO [train.py:968] (1/2) Epoch 14, batch 16950, libri_loss[loss=0.2764, simple_loss=0.3502, pruned_loss=0.1013, over 28624.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3379, pruned_loss=0.08709, over 5679823.51 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3325, pruned_loss=0.09013, over 5755141.68 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3381, pruned_loss=0.0865, over 5666902.62 frames. ], batch size: 106, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:12:33,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 07:13:31,401 INFO [optim.py:369] (1/2) 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 07:13:36,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-07 07:13:38,884 INFO [train.py:968] (1/2) Epoch 14, batch 17000, giga_loss[loss=0.2308, simple_loss=0.3144, pruned_loss=0.07358, over 28908.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3369, pruned_loss=0.08764, over 5677866.87 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3322, pruned_loss=0.08994, over 5756655.21 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3374, pruned_loss=0.08733, over 5665603.22 frames. ], batch size: 213, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:13:55,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-07 07:14:50,502 INFO [train.py:968] (1/2) Epoch 14, batch 17050, giga_loss[loss=0.2095, simple_loss=0.3077, pruned_loss=0.05563, over 28890.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3348, pruned_loss=0.08575, over 5680276.94 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3318, pruned_loss=0.08962, over 5760587.66 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3356, pruned_loss=0.08571, over 5665542.90 frames. ], batch size: 164, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:14:54,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8382, 1.0775, 2.8218, 2.6359], device='cuda:1'), covar=tensor([0.1670, 0.2674, 0.0560, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0610, 0.0884, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 07:14:55,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:15:02,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:15:39,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:15:40,717 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610388.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:15:43,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610391.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:15:49,446 INFO [optim.py:369] (1/2) 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 07:15:55,829 INFO [train.py:968] (1/2) Epoch 14, batch 17100, giga_loss[loss=0.2829, simple_loss=0.3529, pruned_loss=0.1065, over 26972.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3341, pruned_loss=0.08537, over 5676700.87 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3319, pruned_loss=0.08965, over 5761543.72 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3347, pruned_loss=0.08529, over 5664108.94 frames. ], batch size: 555, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:16:22,406 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610420.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 07:16:44,285 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=610442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:16:49,426 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-07 07:16:54,401 INFO [train.py:968] (1/2) Epoch 14, batch 17150, giga_loss[loss=0.2344, simple_loss=0.3099, pruned_loss=0.07943, over 24566.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3348, pruned_loss=0.08572, over 5678945.60 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3319, pruned_loss=0.08955, over 5765695.87 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3353, pruned_loss=0.08562, over 5662865.61 frames. ], batch size: 705, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:17:38,726 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:17:49,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:17:53,306 INFO [train.py:968] (1/2) Epoch 14, batch 17200, libri_loss[loss=0.2645, simple_loss=0.3463, pruned_loss=0.09136, over 29508.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3375, pruned_loss=0.08716, over 5678695.43 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3318, pruned_loss=0.08949, over 5767423.02 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3379, pruned_loss=0.08709, over 5663264.14 frames. ], batch size: 85, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:18:07,781 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=610513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:18:43,037 INFO [train.py:968] (1/2) Epoch 14, batch 17250, giga_loss[loss=0.2435, simple_loss=0.3224, pruned_loss=0.08235, over 28930.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3357, pruned_loss=0.08717, over 5685751.85 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3321, pruned_loss=0.08964, over 5771515.56 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3361, pruned_loss=0.08685, over 5665305.11 frames. ], batch size: 285, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:19:34,415 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 07:19:34,555 INFO [optim.py:369] (1/2) 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 07:19:38,347 INFO [train.py:968] (1/2) Epoch 14, batch 17300, giga_loss[loss=0.2439, simple_loss=0.3264, pruned_loss=0.08064, over 28944.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.333, pruned_loss=0.08692, over 5670981.18 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3317, pruned_loss=0.08946, over 5765581.70 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3337, pruned_loss=0.08678, over 5657567.29 frames. ], batch size: 164, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:20:15,250 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:18,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:23,268 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:27,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:37,810 INFO [train.py:968] (1/2) Epoch 14, batch 17350, giga_loss[loss=0.2995, simple_loss=0.3738, pruned_loss=0.1126, over 28639.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3336, pruned_loss=0.08792, over 5660247.20 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3315, pruned_loss=0.0894, over 5766191.69 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3343, pruned_loss=0.08784, over 5648242.05 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:20:42,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5411, 1.7887, 1.3680, 1.8643], device='cuda:1'), covar=tensor([0.2371, 0.2314, 0.2628, 0.2201], device='cuda:1'), in_proj_covar=tensor([0.1349, 0.0991, 0.1198, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 07:20:51,080 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:20:57,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=610670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:21:29,847 INFO [optim.py:369] (1/2) 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 07:21:33,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2973, 1.2556, 1.1752, 1.4942], device='cuda:1'), covar=tensor([0.0795, 0.0332, 0.0339, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 07:21:34,360 INFO [train.py:968] (1/2) Epoch 14, batch 17400, giga_loss[loss=0.3493, simple_loss=0.4057, pruned_loss=0.1465, over 27598.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3427, pruned_loss=0.09351, over 5665394.82 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3315, pruned_loss=0.08948, over 5765563.36 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3434, pruned_loss=0.09339, over 5654547.06 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:22:17,556 INFO [train.py:968] (1/2) Epoch 14, batch 17450, giga_loss[loss=0.3075, simple_loss=0.3874, pruned_loss=0.1138, over 28881.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3523, pruned_loss=0.09927, over 5675791.96 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3313, pruned_loss=0.08931, over 5768982.93 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3534, pruned_loss=0.09948, over 5662088.35 frames. ], batch size: 112, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:22:57,936 INFO [optim.py:369] (1/2) 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 07:23:00,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4389, 1.7495, 1.4503, 1.2303], device='cuda:1'), covar=tensor([0.2067, 0.1891, 0.2132, 0.1983], device='cuda:1'), in_proj_covar=tensor([0.1350, 0.0993, 0.1196, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 07:23:02,027 INFO [train.py:968] (1/2) Epoch 14, batch 17500, giga_loss[loss=0.2325, simple_loss=0.3048, pruned_loss=0.08005, over 28680.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3536, pruned_loss=0.1009, over 5669648.66 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3318, pruned_loss=0.08966, over 5766570.01 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3543, pruned_loss=0.1009, over 5659138.05 frames. ], batch size: 71, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:23:18,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=610817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:23:27,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4241, 1.5605, 1.5775, 1.4424], device='cuda:1'), covar=tensor([0.1593, 0.1889, 0.2181, 0.1884], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0721, 0.0673, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 07:23:48,348 INFO [train.py:968] (1/2) Epoch 14, batch 17550, giga_loss[loss=0.2408, simple_loss=0.3219, pruned_loss=0.07982, over 28558.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3478, pruned_loss=0.09834, over 5676208.08 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3318, pruned_loss=0.08966, over 5767577.89 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3486, pruned_loss=0.09847, over 5666282.79 frames. ], batch size: 336, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:24:00,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5663, 1.8375, 1.8291, 1.3561], device='cuda:1'), covar=tensor([0.1737, 0.2412, 0.1473, 0.1740], device='cuda:1'), in_proj_covar=tensor([0.0846, 0.0677, 0.0889, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 07:24:19,024 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:24:30,553 INFO [train.py:968] (1/2) Epoch 14, batch 17600, giga_loss[loss=0.2266, simple_loss=0.3058, pruned_loss=0.07372, over 28471.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3423, pruned_loss=0.09586, over 5690938.25 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3323, pruned_loss=0.08963, over 5771908.48 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.343, pruned_loss=0.09629, over 5675370.37 frames. ], batch size: 65, lr: 2.31e-03, grad_scale: 8.0
+2023-03-07 07:24:43,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3091, 1.6052, 1.5769, 1.1868], device='cuda:1'), covar=tensor([0.1564, 0.2272, 0.1277, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0847, 0.0678, 0.0889, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 07:25:12,079 INFO [train.py:968] (1/2) Epoch 14, batch 17650, giga_loss[loss=0.2049, simple_loss=0.2818, pruned_loss=0.06398, over 28421.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3343, pruned_loss=0.09212, over 5689613.00 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3324, pruned_loss=0.08961, over 5764287.05 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3348, pruned_loss=0.09254, over 5682522.50 frames. ], batch size: 78, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:25:22,026 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=610960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:25:24,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=610963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:25:51,116 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:25:58,551 INFO [train.py:968] (1/2) Epoch 14, batch 17700, giga_loss[loss=0.1987, simple_loss=0.2655, pruned_loss=0.06589, over 28522.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3263, pruned_loss=0.08881, over 5688884.04 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3324, pruned_loss=0.08949, over 5763248.46 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3267, pruned_loss=0.08927, over 5682856.76 frames. ], batch size: 60, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:26:23,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5896, 1.7266, 1.7035, 1.6460], device='cuda:1'), covar=tensor([0.1661, 0.1923, 0.2119, 0.1837], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0727, 0.0674, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 07:26:24,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=611031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:26:26,487 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 17750, giga_loss[loss=0.2082, simple_loss=0.2878, pruned_loss=0.0643, over 28685.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3205, pruned_loss=0.08605, over 5672335.77 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3325, pruned_loss=0.08937, over 5745034.62 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3204, pruned_loss=0.08644, over 5683433.74 frames. ], batch size: 242, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:26:48,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=611063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:27:15,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 07:27:17,156 INFO [optim.py:369] (1/2) 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 07:27:18,498 INFO [train.py:968] (1/2) Epoch 14, batch 17800, libri_loss[loss=0.2873, simple_loss=0.365, pruned_loss=0.1048, over 29486.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3167, pruned_loss=0.08431, over 5680087.22 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3329, pruned_loss=0.08954, over 5748186.66 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3159, pruned_loss=0.08438, over 5684781.85 frames. ], batch size: 85, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:27:59,842 INFO [train.py:968] (1/2) Epoch 14, batch 17850, libri_loss[loss=0.2628, simple_loss=0.3511, pruned_loss=0.08725, over 29522.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3149, pruned_loss=0.08333, over 5680275.17 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3331, pruned_loss=0.08953, over 5741541.97 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.3135, pruned_loss=0.08322, over 5687131.35 frames. ], batch size: 81, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:28:19,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.41 vs. limit=5.0
+2023-03-07 07:28:38,862 INFO [optim.py:369] (1/2) 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 07:28:40,537 INFO [train.py:968] (1/2) Epoch 14, batch 17900, giga_loss[loss=0.2093, simple_loss=0.2942, pruned_loss=0.06218, over 28325.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3121, pruned_loss=0.08234, over 5684985.72 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.333, pruned_loss=0.08938, over 5744715.86 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3106, pruned_loss=0.08223, over 5686181.93 frames. ], batch size: 368, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:28:50,517 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=611209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:29:22,364 INFO [train.py:968] (1/2) Epoch 14, batch 17950, giga_loss[loss=0.2212, simple_loss=0.2901, pruned_loss=0.07618, over 27676.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3096, pruned_loss=0.08101, over 5693195.71 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3335, pruned_loss=0.08945, over 5748870.74 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3073, pruned_loss=0.08061, over 5688402.44 frames. ], batch size: 472, lr: 2.31e-03, grad_scale: 2.0
+2023-03-07 07:30:04,861 INFO [optim.py:369] (1/2) 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 07:30:06,024 INFO [train.py:968] (1/2) Epoch 14, batch 18000, giga_loss[loss=0.2178, simple_loss=0.2923, pruned_loss=0.07166, over 28596.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3071, pruned_loss=0.07981, over 5699494.16 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3338, pruned_loss=0.08954, over 5751685.36 frames. ], giga_tot_loss[loss=0.2314, simple_loss=0.3044, pruned_loss=0.07921, over 5691915.12 frames. ], batch size: 307, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:30:06,024 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 07:30:14,508 INFO [train.py:1012] (1/2) Epoch 14, validation: loss=0.2099, simple_loss=0.3147, pruned_loss=0.05255, over 944034.00 frames. 
+2023-03-07 07:30:14,508 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 07:30:55,350 INFO [train.py:968] (1/2) Epoch 14, batch 18050, giga_loss[loss=0.211, simple_loss=0.2931, pruned_loss=0.06447, over 28934.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3054, pruned_loss=0.07912, over 5688362.13 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.334, pruned_loss=0.08957, over 5747543.70 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.3017, pruned_loss=0.07815, over 5683178.65 frames. ], batch size: 174, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:31:32,923 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=611389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:31:40,651 INFO [optim.py:369] (1/2) 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 07:31:42,025 INFO [train.py:968] (1/2) Epoch 14, batch 18100, giga_loss[loss=0.1749, simple_loss=0.2523, pruned_loss=0.04868, over 28314.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3019, pruned_loss=0.07761, over 5690693.98 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3339, pruned_loss=0.0895, over 5750072.97 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.2986, pruned_loss=0.07674, over 5683480.08 frames. ], batch size: 65, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:32:31,057 INFO [train.py:968] (1/2) Epoch 14, batch 18150, giga_loss[loss=0.1869, simple_loss=0.2601, pruned_loss=0.05685, over 28848.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.2988, pruned_loss=0.07615, over 5695796.51 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3344, pruned_loss=0.08982, over 5743261.62 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2952, pruned_loss=0.07497, over 5694726.53 frames. ], batch size: 119, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:33:13,726 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 18200, giga_loss[loss=0.2455, simple_loss=0.3139, pruned_loss=0.08857, over 28801.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.2985, pruned_loss=0.07631, over 5688939.45 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3344, pruned_loss=0.08976, over 5736909.01 frames. ], giga_tot_loss[loss=0.2227, simple_loss=0.295, pruned_loss=0.07518, over 5693201.35 frames. ], batch size: 99, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:34:06,179 INFO [train.py:968] (1/2) Epoch 14, batch 18250, giga_loss[loss=0.338, simple_loss=0.3961, pruned_loss=0.1399, over 28957.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3103, pruned_loss=0.08235, over 5691713.64 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3347, pruned_loss=0.08971, over 5739475.28 frames. ], giga_tot_loss[loss=0.2348, simple_loss=0.3069, pruned_loss=0.08135, over 5692316.74 frames. ], batch size: 136, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:34:38,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=611584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:34:47,909 INFO [optim.py:369] (1/2) 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 07:34:49,189 INFO [train.py:968] (1/2) Epoch 14, batch 18300, libri_loss[loss=0.2866, simple_loss=0.3568, pruned_loss=0.1082, over 29540.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3243, pruned_loss=0.0896, over 5691342.07 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3352, pruned_loss=0.08987, over 5741643.81 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3206, pruned_loss=0.08853, over 5688539.78 frames. ], batch size: 80, lr: 2.31e-03, grad_scale: 4.0
+2023-03-07 07:35:01,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7120, 1.8576, 1.7012, 1.7334], device='cuda:1'), covar=tensor([0.1508, 0.1824, 0.1979, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0725, 0.0674, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 07:35:26,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3027, 3.1261, 1.4952, 1.5177], device='cuda:1'), covar=tensor([0.1022, 0.0303, 0.0907, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0517, 0.0353, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 07:35:32,034 INFO [train.py:968] (1/2) Epoch 14, batch 18350, giga_loss[loss=0.3137, simple_loss=0.3652, pruned_loss=0.1311, over 23604.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3346, pruned_loss=0.09467, over 5698800.63 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3351, pruned_loss=0.08969, over 5744686.36 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3316, pruned_loss=0.09406, over 5692906.05 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:35:59,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=611685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:36:02,341 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 07:36:10,391 INFO [optim.py:369] (1/2) 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 07:36:11,734 INFO [train.py:968] (1/2) Epoch 14, batch 18400, giga_loss[loss=0.2867, simple_loss=0.3645, pruned_loss=0.1045, over 29114.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3399, pruned_loss=0.09607, over 5702010.12 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3349, pruned_loss=0.08945, over 5750393.67 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3377, pruned_loss=0.096, over 5690302.36 frames. ], batch size: 113, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:36:32,529 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=611727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:36:34,509 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 18450, giga_loss[loss=0.2911, simple_loss=0.3729, pruned_loss=0.1047, over 29066.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3429, pruned_loss=0.09657, over 5704857.66 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.335, pruned_loss=0.08933, over 5755551.72 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3413, pruned_loss=0.09687, over 5688799.74 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:36:55,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1997, 3.9941, 3.7726, 1.6819], device='cuda:1'), covar=tensor([0.0548, 0.0718, 0.0715, 0.2291], device='cuda:1'), in_proj_covar=tensor([0.1085, 0.1004, 0.0876, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 07:36:59,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=611759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:37:03,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=611764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:37:03,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6206, 1.7344, 1.6559, 1.5680], device='cuda:1'), covar=tensor([0.1610, 0.2419, 0.2015, 0.2086], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0726, 0.0676, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 07:37:37,960 INFO [optim.py:369] (1/2) 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 07:37:38,617 INFO [train.py:968] (1/2) Epoch 14, batch 18500, giga_loss[loss=0.2731, simple_loss=0.343, pruned_loss=0.1016, over 28751.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3446, pruned_loss=0.09705, over 5691581.52 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3354, pruned_loss=0.08949, over 5749092.85 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3431, pruned_loss=0.09724, over 5683532.83 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:38:23,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4236, 1.1741, 4.6042, 3.2179], device='cuda:1'), covar=tensor([0.1679, 0.2944, 0.0340, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0684, 0.0604, 0.0880, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 07:38:25,228 INFO [train.py:968] (1/2) Epoch 14, batch 18550, giga_loss[loss=0.3306, simple_loss=0.3956, pruned_loss=0.1328, over 28676.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3472, pruned_loss=0.09928, over 5692262.83 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3355, pruned_loss=0.08954, over 5750543.77 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.346, pruned_loss=0.09945, over 5684329.09 frames. ], batch size: 242, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:38:58,404 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5814, 1.6793, 1.7893, 1.3640], device='cuda:1'), covar=tensor([0.1647, 0.2369, 0.1318, 0.1586], device='cuda:1'), in_proj_covar=tensor([0.0852, 0.0683, 0.0893, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 07:39:07,923 INFO [optim.py:369] (1/2) 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 07:39:07,935 INFO [train.py:968] (1/2) Epoch 14, batch 18600, giga_loss[loss=0.2833, simple_loss=0.3498, pruned_loss=0.1085, over 28980.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3501, pruned_loss=0.1012, over 5702334.65 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3357, pruned_loss=0.08938, over 5755493.21 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3494, pruned_loss=0.1019, over 5689579.22 frames. ], batch size: 106, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:39:14,106 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=611907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:39:16,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=611910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:39:41,673 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=611939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:39:46,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 07:39:49,152 INFO [train.py:968] (1/2) Epoch 14, batch 18650, giga_loss[loss=0.2996, simple_loss=0.3536, pruned_loss=0.1228, over 23323.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3526, pruned_loss=0.1023, over 5696213.27 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3357, pruned_loss=0.08929, over 5750313.34 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3525, pruned_loss=0.1033, over 5688295.65 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:40:04,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 07:40:27,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6774, 1.9645, 1.7663, 1.5306], device='cuda:1'), covar=tensor([0.2044, 0.1844, 0.1973, 0.2078], device='cuda:1'), in_proj_covar=tensor([0.1783, 0.1679, 0.1625, 0.1769], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 07:40:30,008 INFO [optim.py:369] (1/2) 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 07:40:30,021 INFO [train.py:968] (1/2) Epoch 14, batch 18700, giga_loss[loss=0.2521, simple_loss=0.3429, pruned_loss=0.08065, over 28913.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3542, pruned_loss=0.1019, over 5709205.62 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3362, pruned_loss=0.08954, over 5755042.19 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3542, pruned_loss=0.1028, over 5697315.06 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:41:09,942 INFO [train.py:968] (1/2) Epoch 14, batch 18750, libri_loss[loss=0.2571, simple_loss=0.3359, pruned_loss=0.08914, over 29596.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3554, pruned_loss=0.1023, over 5715523.53 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3361, pruned_loss=0.08951, over 5759339.54 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3563, pruned_loss=0.1036, over 5699890.92 frames. ], batch size: 75, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:41:16,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 07:41:17,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:41:48,889 INFO [optim.py:369] (1/2) 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 07:41:48,901 INFO [train.py:968] (1/2) Epoch 14, batch 18800, giga_loss[loss=0.3057, simple_loss=0.3579, pruned_loss=0.1267, over 23413.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.357, pruned_loss=0.1027, over 5713554.21 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3363, pruned_loss=0.08969, over 5762193.14 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.358, pruned_loss=0.104, over 5697493.52 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:42:29,873 INFO [train.py:968] (1/2) Epoch 14, batch 18850, giga_loss[loss=0.2683, simple_loss=0.3495, pruned_loss=0.09356, over 28787.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3572, pruned_loss=0.1022, over 5710672.09 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3372, pruned_loss=0.09027, over 5763816.85 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3577, pruned_loss=0.1029, over 5695527.27 frames. ], batch size: 262, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:42:32,030 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:43:11,227 INFO [train.py:968] (1/2) Epoch 14, batch 18900, giga_loss[loss=0.2695, simple_loss=0.3471, pruned_loss=0.09595, over 28531.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3555, pruned_loss=0.09999, over 5712315.71 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.09061, over 5765254.96 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3556, pruned_loss=0.1004, over 5698342.18 frames. ], batch size: 65, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:43:13,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:14,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 07:43:16,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:22,387 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:39,548 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:43:43,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-07 07:43:50,549 INFO [train.py:968] (1/2) Epoch 14, batch 18950, giga_loss[loss=0.2957, simple_loss=0.3689, pruned_loss=0.1112, over 28774.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3539, pruned_loss=0.09858, over 5715221.81 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.09057, over 5767879.62 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3545, pruned_loss=0.09915, over 5700218.46 frames. ], batch size: 262, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:44:00,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6115, 1.5803, 1.8933, 1.4843], device='cuda:1'), covar=tensor([0.1223, 0.1706, 0.0998, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0684, 0.0896, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 07:44:14,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 07:44:32,081 INFO [optim.py:369] (1/2) 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 07:44:32,093 INFO [train.py:968] (1/2) Epoch 14, batch 19000, giga_loss[loss=0.2724, simple_loss=0.3439, pruned_loss=0.1004, over 29075.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3561, pruned_loss=0.1021, over 5706765.22 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3379, pruned_loss=0.09071, over 5770592.79 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3568, pruned_loss=0.1026, over 5691233.69 frames. ], batch size: 128, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:45:16,137 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-07 07:45:21,462 INFO [train.py:968] (1/2) Epoch 14, batch 19050, giga_loss[loss=0.2996, simple_loss=0.3626, pruned_loss=0.1182, over 28916.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3598, pruned_loss=0.1075, over 5693604.23 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.338, pruned_loss=0.09078, over 5770591.68 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3604, pruned_loss=0.108, over 5680779.09 frames. ], batch size: 112, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:45:59,139 INFO [optim.py:369] (1/2) 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 07:45:59,152 INFO [train.py:968] (1/2) Epoch 14, batch 19100, giga_loss[loss=0.2618, simple_loss=0.3401, pruned_loss=0.09174, over 29037.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3586, pruned_loss=0.1071, over 5704875.63 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3384, pruned_loss=0.09098, over 5774285.57 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3592, pruned_loss=0.1078, over 5689259.88 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:46:01,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3693, 2.9602, 2.5346, 1.9612], device='cuda:1'), covar=tensor([0.2222, 0.1302, 0.1418, 0.1806], device='cuda:1'), in_proj_covar=tensor([0.1785, 0.1688, 0.1629, 0.1774], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 07:46:20,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.07 vs. limit=2.0
+2023-03-07 07:46:26,593 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612435.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:46:40,627 INFO [train.py:968] (1/2) Epoch 14, batch 19150, giga_loss[loss=0.306, simple_loss=0.3726, pruned_loss=0.1197, over 29017.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3571, pruned_loss=0.1071, over 5708569.43 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3387, pruned_loss=0.09094, over 5777851.86 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3579, pruned_loss=0.1081, over 5691111.32 frames. ], batch size: 164, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:46:45,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2070, 3.0092, 2.8534, 1.3848], device='cuda:1'), covar=tensor([0.0959, 0.1011, 0.0886, 0.2447], device='cuda:1'), in_proj_covar=tensor([0.1076, 0.0996, 0.0871, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 07:47:22,389 INFO [train.py:968] (1/2) Epoch 14, batch 19200, giga_loss[loss=0.2919, simple_loss=0.3612, pruned_loss=0.1112, over 28775.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3557, pruned_loss=0.1061, over 5709513.26 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3389, pruned_loss=0.09096, over 5781476.70 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3565, pruned_loss=0.1072, over 5690902.03 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:47:23,917 INFO [optim.py:369] (1/2) 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 07:47:47,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-07 07:47:50,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:47:58,251 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:48:06,281 INFO [train.py:968] (1/2) Epoch 14, batch 19250, giga_loss[loss=0.2328, simple_loss=0.3245, pruned_loss=0.07051, over 29084.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.355, pruned_loss=0.1051, over 5702752.92 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3391, pruned_loss=0.09107, over 5783488.89 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3558, pruned_loss=0.1062, over 5684011.52 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:48:22,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5140, 1.8468, 1.4548, 1.5017], device='cuda:1'), covar=tensor([0.2432, 0.2304, 0.2622, 0.2195], device='cuda:1'), in_proj_covar=tensor([0.1351, 0.0995, 0.1192, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 07:48:40,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:48:49,731 INFO [train.py:968] (1/2) Epoch 14, batch 19300, giga_loss[loss=0.2357, simple_loss=0.3218, pruned_loss=0.07477, over 28921.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3522, pruned_loss=0.1027, over 5704616.69 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3386, pruned_loss=0.09076, over 5786347.01 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3536, pruned_loss=0.1042, over 5685266.49 frames. ], batch size: 174, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:48:52,123 INFO [optim.py:369] (1/2) 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 07:49:16,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3994, 1.6328, 1.3443, 1.3615], device='cuda:1'), covar=tensor([0.2894, 0.2630, 0.2949, 0.2234], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.1001, 0.1198, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 07:49:34,452 INFO [train.py:968] (1/2) Epoch 14, batch 19350, giga_loss[loss=0.2394, simple_loss=0.3191, pruned_loss=0.07989, over 28337.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3483, pruned_loss=0.1005, over 5694841.50 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3391, pruned_loss=0.0909, over 5784726.53 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3492, pruned_loss=0.1018, over 5679325.31 frames. ], batch size: 65, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:49:54,662 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:49:56,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:20,401 INFO [train.py:968] (1/2) Epoch 14, batch 19400, giga_loss[loss=0.2343, simple_loss=0.3134, pruned_loss=0.07757, over 28245.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3427, pruned_loss=0.09757, over 5690704.31 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3393, pruned_loss=0.09093, over 5783054.16 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3434, pruned_loss=0.0987, over 5677718.27 frames. ], batch size: 368, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:50:21,022 INFO [optim.py:369] (1/2) 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 07:50:21,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:25,855 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:49,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:53,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:50:54,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6837, 1.7099, 1.3343, 1.2415], device='cuda:1'), covar=tensor([0.0759, 0.0501, 0.0895, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0435, 0.0502, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 07:51:09,999 INFO [train.py:968] (1/2) Epoch 14, batch 19450, giga_loss[loss=0.2239, simple_loss=0.3026, pruned_loss=0.07254, over 28927.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3363, pruned_loss=0.09409, over 5691526.39 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3392, pruned_loss=0.09085, over 5784289.64 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3369, pruned_loss=0.09509, over 5679606.95 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:51:24,266 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:51:57,775 INFO [train.py:968] (1/2) Epoch 14, batch 19500, giga_loss[loss=0.2409, simple_loss=0.3272, pruned_loss=0.07736, over 28632.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3356, pruned_loss=0.09308, over 5692976.34 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3395, pruned_loss=0.09097, over 5786298.84 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3358, pruned_loss=0.09379, over 5680613.76 frames. ], batch size: 242, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:51:58,441 INFO [optim.py:369] (1/2) 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 07:51:59,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:52:07,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:52:41,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5690, 5.3766, 5.0844, 2.6712], device='cuda:1'), covar=tensor([0.0377, 0.0500, 0.0538, 0.1691], device='cuda:1'), in_proj_covar=tensor([0.1061, 0.0983, 0.0862, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 07:52:42,082 INFO [train.py:968] (1/2) Epoch 14, batch 19550, giga_loss[loss=0.225, simple_loss=0.3067, pruned_loss=0.0717, over 28808.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3365, pruned_loss=0.09307, over 5706678.37 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3401, pruned_loss=0.09125, over 5787312.14 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.336, pruned_loss=0.0934, over 5694848.27 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:53:04,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5908, 1.7376, 1.4648, 1.6274], device='cuda:1'), covar=tensor([0.2458, 0.2513, 0.2754, 0.2259], device='cuda:1'), in_proj_covar=tensor([0.1355, 0.0999, 0.1198, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 07:53:04,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-07 07:53:05,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4839, 1.7516, 1.3823, 1.7294], device='cuda:1'), covar=tensor([0.2501, 0.2527, 0.2851, 0.2333], device='cuda:1'), in_proj_covar=tensor([0.1355, 0.0999, 0.1198, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 07:53:25,200 INFO [train.py:968] (1/2) Epoch 14, batch 19600, giga_loss[loss=0.2459, simple_loss=0.3231, pruned_loss=0.08441, over 28873.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3364, pruned_loss=0.09342, over 5707395.97 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3403, pruned_loss=0.09115, over 5787487.04 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3357, pruned_loss=0.09379, over 5696774.88 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:53:25,862 INFO [optim.py:369] (1/2) 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 07:53:35,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=612913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:54:05,536 INFO [train.py:968] (1/2) Epoch 14, batch 19650, giga_loss[loss=0.2354, simple_loss=0.3068, pruned_loss=0.08198, over 28407.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3339, pruned_loss=0.09196, over 5715292.27 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.341, pruned_loss=0.09121, over 5787040.19 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3326, pruned_loss=0.09223, over 5705780.69 frames. ], batch size: 71, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 07:54:09,247 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=612953.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:54:11,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=612956.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:54:14,646 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:54:33,188 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=612985.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 07:54:37,470 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=612990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:54:44,003 INFO [train.py:968] (1/2) Epoch 14, batch 19700, giga_loss[loss=0.24, simple_loss=0.3168, pruned_loss=0.08158, over 28831.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3326, pruned_loss=0.09189, over 5713766.96 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3421, pruned_loss=0.09186, over 5777992.91 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3305, pruned_loss=0.09154, over 5712316.44 frames. ], batch size: 112, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:54:45,290 INFO [optim.py:369] (1/2) 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 07:55:27,280 INFO [train.py:968] (1/2) Epoch 14, batch 19750, giga_loss[loss=0.2481, simple_loss=0.3214, pruned_loss=0.08738, over 28950.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3299, pruned_loss=0.0914, over 5713695.65 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3423, pruned_loss=0.09205, over 5778467.95 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3281, pruned_loss=0.09096, over 5711888.63 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:55:32,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:55:35,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:55:50,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-07 07:55:54,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:55:58,860 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 19800, libri_loss[loss=0.3253, simple_loss=0.4083, pruned_loss=0.1212, over 29463.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3281, pruned_loss=0.09059, over 5720952.98 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3425, pruned_loss=0.09219, over 5780091.36 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3262, pruned_loss=0.0901, over 5717245.56 frames. ], batch size: 85, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:56:10,136 INFO [optim.py:369] (1/2) 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 07:56:46,341 INFO [train.py:968] (1/2) Epoch 14, batch 19850, giga_loss[loss=0.2269, simple_loss=0.2915, pruned_loss=0.08116, over 28621.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3268, pruned_loss=0.08988, over 5721262.27 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3439, pruned_loss=0.09292, over 5780605.58 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3235, pruned_loss=0.08876, over 5716253.40 frames. ], batch size: 85, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:57:07,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:57:27,431 INFO [train.py:968] (1/2) Epoch 14, batch 19900, libri_loss[loss=0.2526, simple_loss=0.3429, pruned_loss=0.08108, over 29542.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3245, pruned_loss=0.08832, over 5712483.88 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3445, pruned_loss=0.09304, over 5773274.67 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.321, pruned_loss=0.08725, over 5713152.82 frames. ], batch size: 75, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:57:29,560 INFO [optim.py:369] (1/2) 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 07:57:47,303 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:57:52,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:58:08,176 INFO [train.py:968] (1/2) Epoch 14, batch 19950, giga_loss[loss=0.2289, simple_loss=0.3127, pruned_loss=0.07249, over 28731.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3227, pruned_loss=0.08703, over 5719606.41 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.345, pruned_loss=0.09322, over 5771206.77 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3189, pruned_loss=0.08586, over 5720532.03 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 07:58:14,277 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 20000, giga_loss[loss=0.2474, simple_loss=0.3174, pruned_loss=0.08868, over 28808.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3204, pruned_loss=0.08584, over 5722875.95 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3452, pruned_loss=0.0932, over 5772569.38 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.317, pruned_loss=0.08488, over 5721991.40 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:58:49,445 INFO [optim.py:369] (1/2) 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 07:59:01,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:59:04,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:59:16,604 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 20050, giga_loss[loss=0.2441, simple_loss=0.3182, pruned_loss=0.08504, over 29014.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3198, pruned_loss=0.08521, over 5725868.93 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3461, pruned_loss=0.0936, over 5766547.37 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3155, pruned_loss=0.08383, over 5730168.16 frames. ], batch size: 106, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 07:59:27,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 07:59:39,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:00:08,734 INFO [train.py:968] (1/2) Epoch 14, batch 20100, giga_loss[loss=0.2486, simple_loss=0.3209, pruned_loss=0.0882, over 28682.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3246, pruned_loss=0.08877, over 5716032.93 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3461, pruned_loss=0.09358, over 5768180.84 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.321, pruned_loss=0.08764, over 5717525.78 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:00:10,871 INFO [optim.py:369] (1/2) 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 08:00:53,806 INFO [train.py:968] (1/2) Epoch 14, batch 20150, giga_loss[loss=0.298, simple_loss=0.3666, pruned_loss=0.1148, over 28247.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3312, pruned_loss=0.09248, over 5720174.08 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.346, pruned_loss=0.09343, over 5773766.06 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3278, pruned_loss=0.09162, over 5714342.74 frames. ], batch size: 368, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:01:24,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:26,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:32,246 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-07 08:01:41,315 INFO [train.py:968] (1/2) Epoch 14, batch 20200, giga_loss[loss=0.2718, simple_loss=0.3491, pruned_loss=0.09729, over 28957.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3407, pruned_loss=0.09911, over 5705667.87 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3459, pruned_loss=0.09329, over 5777208.95 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3377, pruned_loss=0.09861, over 5696453.00 frames. ], batch size: 128, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:01:44,347 INFO [optim.py:369] (1/2) 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 08:01:44,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:45,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-07 08:01:50,466 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=613508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:52,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=613511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:01:53,975 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:02:08,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3402, 1.5421, 1.5976, 1.3130], device='cuda:1'), covar=tensor([0.1592, 0.1861, 0.1947, 0.1854], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0733, 0.0684, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 08:02:19,146 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=613540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:02:28,665 INFO [train.py:968] (1/2) Epoch 14, batch 20250, giga_loss[loss=0.3089, simple_loss=0.3844, pruned_loss=0.1166, over 28638.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3451, pruned_loss=0.1007, over 5704628.53 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3459, pruned_loss=0.09322, over 5780554.88 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3427, pruned_loss=0.1006, over 5692509.73 frames. ], batch size: 71, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:02:49,282 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:03:14,761 INFO [train.py:968] (1/2) Epoch 14, batch 20300, giga_loss[loss=0.2967, simple_loss=0.3733, pruned_loss=0.1101, over 29057.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3508, pruned_loss=0.1036, over 5690763.78 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.346, pruned_loss=0.09324, over 5782499.93 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3489, pruned_loss=0.1036, over 5678104.87 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:03:17,977 INFO [optim.py:369] (1/2) 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 08:04:02,445 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:04:04,326 INFO [train.py:968] (1/2) Epoch 14, batch 20350, giga_loss[loss=0.3547, simple_loss=0.4192, pruned_loss=0.1451, over 28949.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3574, pruned_loss=0.108, over 5681943.40 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.346, pruned_loss=0.09322, over 5783103.43 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3559, pruned_loss=0.108, over 5671023.28 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:04:44,163 INFO [train.py:968] (1/2) Epoch 14, batch 20400, libri_loss[loss=0.2501, simple_loss=0.324, pruned_loss=0.08807, over 29662.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3585, pruned_loss=0.1084, over 5685103.43 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3463, pruned_loss=0.09339, over 5786508.30 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3574, pruned_loss=0.1087, over 5670549.63 frames. ], batch size: 69, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:04:46,288 INFO [optim.py:369] (1/2) 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 08:05:27,669 INFO [train.py:968] (1/2) Epoch 14, batch 20450, giga_loss[loss=0.2466, simple_loss=0.3301, pruned_loss=0.08153, over 28884.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3526, pruned_loss=0.1037, over 5685744.71 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3458, pruned_loss=0.09311, over 5787460.37 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3524, pruned_loss=0.1045, over 5671118.45 frames. ], batch size: 112, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:05:46,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=613773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:06:08,917 INFO [train.py:968] (1/2) Epoch 14, batch 20500, giga_loss[loss=0.2591, simple_loss=0.3336, pruned_loss=0.09228, over 28916.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3512, pruned_loss=0.102, over 5702428.56 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3461, pruned_loss=0.09335, over 5787315.50 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3508, pruned_loss=0.1026, over 5688683.65 frames. ], batch size: 106, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:06:11,385 INFO [optim.py:369] (1/2) 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 08:06:18,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5472, 2.0304, 1.5687, 1.8901], device='cuda:1'), covar=tensor([0.0749, 0.0261, 0.0291, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0061, 0.0055, 0.0095], device='cuda:1')
+2023-03-07 08:06:49,248 INFO [train.py:968] (1/2) Epoch 14, batch 20550, libri_loss[loss=0.2851, simple_loss=0.3695, pruned_loss=0.1003, over 29760.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3502, pruned_loss=0.1008, over 5701501.22 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3462, pruned_loss=0.09336, over 5788677.51 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.35, pruned_loss=0.1016, over 5686021.85 frames. ], batch size: 87, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:07:13,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:07:32,537 INFO [train.py:968] (1/2) Epoch 14, batch 20600, giga_loss[loss=0.3256, simple_loss=0.3743, pruned_loss=0.1385, over 23835.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.354, pruned_loss=0.1029, over 5695278.85 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3464, pruned_loss=0.09333, over 5788992.78 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3538, pruned_loss=0.1036, over 5682081.98 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:07:37,101 INFO [optim.py:369] (1/2) 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 08:08:12,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=613944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:08:18,703 INFO [train.py:968] (1/2) Epoch 14, batch 20650, giga_loss[loss=0.2681, simple_loss=0.3481, pruned_loss=0.094, over 28950.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3556, pruned_loss=0.1041, over 5699382.89 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3465, pruned_loss=0.09335, over 5782411.29 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3555, pruned_loss=0.1049, over 5692965.36 frames. ], batch size: 174, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:09:02,709 INFO [train.py:968] (1/2) Epoch 14, batch 20700, giga_loss[loss=0.2585, simple_loss=0.3368, pruned_loss=0.09006, over 28623.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3572, pruned_loss=0.1058, over 5684885.65 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3465, pruned_loss=0.09332, over 5776011.59 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3574, pruned_loss=0.1068, over 5683311.29 frames. ], batch size: 78, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:09:06,472 INFO [optim.py:369] (1/2) 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 08:09:20,877 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:09:21,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:09:23,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:09:47,603 INFO [train.py:968] (1/2) Epoch 14, batch 20750, giga_loss[loss=0.3655, simple_loss=0.4059, pruned_loss=0.1626, over 26521.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3595, pruned_loss=0.1082, over 5678681.13 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3468, pruned_loss=0.09349, over 5773951.92 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3596, pruned_loss=0.109, over 5677911.54 frames. ], batch size: 555, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:09:50,518 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:10:21,412 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:10:23,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:10:31,268 INFO [train.py:968] (1/2) Epoch 14, batch 20800, giga_loss[loss=0.3207, simple_loss=0.3814, pruned_loss=0.13, over 28584.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3583, pruned_loss=0.1076, over 5679490.33 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3468, pruned_loss=0.09351, over 5765221.80 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3584, pruned_loss=0.1083, over 5685268.73 frames. ], batch size: 78, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:10:36,243 INFO [optim.py:369] (1/2) 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 08:10:45,870 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:09,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:10,091 INFO [train.py:968] (1/2) Epoch 14, batch 20850, giga_loss[loss=0.2715, simple_loss=0.3499, pruned_loss=0.09655, over 28992.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3585, pruned_loss=0.107, over 5690968.32 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3469, pruned_loss=0.09343, over 5766601.15 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3586, pruned_loss=0.1077, over 5693663.68 frames. ], batch size: 106, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:11:20,253 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:23,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:25,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:48,296 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:11:50,087 INFO [train.py:968] (1/2) Epoch 14, batch 20900, giga_loss[loss=0.2653, simple_loss=0.3517, pruned_loss=0.08942, over 28917.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3584, pruned_loss=0.1057, over 5686928.19 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3473, pruned_loss=0.09367, over 5762398.97 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3584, pruned_loss=0.1065, over 5690495.33 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:11:53,990 INFO [optim.py:369] (1/2) 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 08:12:28,299 INFO [train.py:968] (1/2) Epoch 14, batch 20950, giga_loss[loss=0.3148, simple_loss=0.3839, pruned_loss=0.1228, over 28653.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.358, pruned_loss=0.1039, over 5696553.64 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3477, pruned_loss=0.09355, over 5765649.83 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3581, pruned_loss=0.1051, over 5693621.85 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:13:00,449 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:13:02,364 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 21000, giga_loss[loss=0.2399, simple_loss=0.33, pruned_loss=0.07488, over 28622.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3562, pruned_loss=0.103, over 5704244.42 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3481, pruned_loss=0.09376, over 5768840.52 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3564, pruned_loss=0.1041, over 5696848.74 frames. ], batch size: 78, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:13:06,033 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 08:13:14,937 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 08:13:18,308 INFO [optim.py:369] (1/2) 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 08:13:33,674 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:13:46,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:13:52,286 INFO [train.py:968] (1/2) Epoch 14, batch 21050, giga_loss[loss=0.2947, simple_loss=0.363, pruned_loss=0.1132, over 28823.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.354, pruned_loss=0.1019, over 5716779.87 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3478, pruned_loss=0.0935, over 5772668.52 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3544, pruned_loss=0.1032, over 5706061.82 frames. ], batch size: 243, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:14:32,358 INFO [train.py:968] (1/2) Epoch 14, batch 21100, giga_loss[loss=0.2677, simple_loss=0.345, pruned_loss=0.09521, over 28896.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3518, pruned_loss=0.1011, over 5717129.58 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3477, pruned_loss=0.09352, over 5775161.47 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3524, pruned_loss=0.1023, over 5705360.35 frames. ], batch size: 227, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:14:35,543 INFO [optim.py:369] (1/2) 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 08:15:08,203 INFO [train.py:968] (1/2) Epoch 14, batch 21150, giga_loss[loss=0.3058, simple_loss=0.3744, pruned_loss=0.1187, over 28865.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3506, pruned_loss=0.1006, over 5722724.21 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3478, pruned_loss=0.09375, over 5779130.40 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3511, pruned_loss=0.1016, over 5707124.90 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:15:11,273 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:15:43,693 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=614494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:15:48,228 INFO [train.py:968] (1/2) Epoch 14, batch 21200, giga_loss[loss=0.2987, simple_loss=0.368, pruned_loss=0.1147, over 27933.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3521, pruned_loss=0.1021, over 5718650.36 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3476, pruned_loss=0.09368, over 5778742.56 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3528, pruned_loss=0.1031, over 5705212.81 frames. ], batch size: 412, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:15:54,555 INFO [optim.py:369] (1/2) 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 08:16:20,817 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 21250, giga_loss[loss=0.3064, simple_loss=0.3752, pruned_loss=0.1188, over 28706.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3521, pruned_loss=0.1016, over 5717537.44 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09376, over 5779504.85 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3525, pruned_loss=0.1024, over 5706135.57 frames. ], batch size: 242, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:16:33,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-07 08:17:11,442 INFO [train.py:968] (1/2) Epoch 14, batch 21300, giga_loss[loss=0.258, simple_loss=0.3439, pruned_loss=0.08605, over 29069.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3502, pruned_loss=0.09991, over 5717445.64 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3476, pruned_loss=0.09391, over 5781157.39 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3507, pruned_loss=0.1006, over 5704987.48 frames. ], batch size: 128, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:17:16,799 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 21350, giga_loss[loss=0.2774, simple_loss=0.3542, pruned_loss=0.1003, over 28474.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3498, pruned_loss=0.09948, over 5725139.34 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3479, pruned_loss=0.09418, over 5782271.48 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.35, pruned_loss=0.09978, over 5713792.92 frames. ], batch size: 65, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:18:03,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3686, 1.7377, 1.3586, 1.3170], device='cuda:1'), covar=tensor([0.2472, 0.2296, 0.2648, 0.2147], device='cuda:1'), in_proj_covar=tensor([0.1352, 0.0995, 0.1195, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 08:18:18,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:18:20,544 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:18:34,829 INFO [train.py:968] (1/2) Epoch 14, batch 21400, giga_loss[loss=0.2737, simple_loss=0.3472, pruned_loss=0.1001, over 29005.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3484, pruned_loss=0.09879, over 5726879.14 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3479, pruned_loss=0.09439, over 5776949.50 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3486, pruned_loss=0.09907, over 5719469.69 frames. ], batch size: 106, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:18:40,415 INFO [optim.py:369] (1/2) 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 08:18:44,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:18:47,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:18:48,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-07 08:19:14,031 INFO [train.py:968] (1/2) Epoch 14, batch 21450, giga_loss[loss=0.2602, simple_loss=0.3279, pruned_loss=0.09619, over 28685.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3465, pruned_loss=0.09805, over 5723764.92 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3483, pruned_loss=0.09484, over 5777207.29 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3462, pruned_loss=0.09802, over 5715964.86 frames. ], batch size: 92, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:19:26,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6839, 1.9388, 1.5633, 1.7343], device='cuda:1'), covar=tensor([0.2378, 0.2290, 0.2635, 0.2241], device='cuda:1'), in_proj_covar=tensor([0.1352, 0.0994, 0.1195, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 08:19:55,303 INFO [train.py:968] (1/2) Epoch 14, batch 21500, giga_loss[loss=0.2679, simple_loss=0.3446, pruned_loss=0.09564, over 28885.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3437, pruned_loss=0.09673, over 5727010.39 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3481, pruned_loss=0.09486, over 5781839.40 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3436, pruned_loss=0.09677, over 5714657.39 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:20:00,328 INFO [optim.py:369] (1/2) 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 08:20:09,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4506, 1.5153, 1.3189, 1.6205], device='cuda:1'), covar=tensor([0.0753, 0.0328, 0.0314, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0112, 0.0113, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0095], device='cuda:1')
+2023-03-07 08:20:16,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2697, 1.6331, 0.9747, 1.2574], device='cuda:1'), covar=tensor([0.1062, 0.0606, 0.1464, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0431, 0.0501, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 08:20:18,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:20:33,015 INFO [train.py:968] (1/2) Epoch 14, batch 21550, giga_loss[loss=0.2512, simple_loss=0.3254, pruned_loss=0.08848, over 28422.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09736, over 5732201.08 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3481, pruned_loss=0.09524, over 5784480.58 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3438, pruned_loss=0.09713, over 5718077.03 frames. ], batch size: 65, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:20:39,247 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:20:41,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:20:47,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=614869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:21:02,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3379, 1.5216, 1.2586, 1.4851], device='cuda:1'), covar=tensor([0.2805, 0.2614, 0.3101, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.1358, 0.1000, 0.1199, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 08:21:03,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=614890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:21:10,067 INFO [train.py:968] (1/2) Epoch 14, batch 21600, giga_loss[loss=0.2853, simple_loss=0.3573, pruned_loss=0.1067, over 28640.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3444, pruned_loss=0.0981, over 5732962.90 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3485, pruned_loss=0.09561, over 5788933.16 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3437, pruned_loss=0.09768, over 5715710.84 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:21:14,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9815, 4.7811, 4.5523, 2.1298], device='cuda:1'), covar=tensor([0.0382, 0.0554, 0.0588, 0.1989], device='cuda:1'), in_proj_covar=tensor([0.1084, 0.1004, 0.0873, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 08:21:15,899 INFO [optim.py:369] (1/2) 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 08:21:51,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6786, 1.7973, 1.2676, 1.4437], device='cuda:1'), covar=tensor([0.0787, 0.0596, 0.1059, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0434, 0.0505, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 08:21:54,379 INFO [train.py:968] (1/2) Epoch 14, batch 21650, giga_loss[loss=0.2207, simple_loss=0.3067, pruned_loss=0.06733, over 29004.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3431, pruned_loss=0.09832, over 5730198.98 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3487, pruned_loss=0.0958, over 5790334.10 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3424, pruned_loss=0.09782, over 5714581.11 frames. ], batch size: 164, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:21:59,276 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1764, 3.9990, 3.8014, 1.7744], device='cuda:1'), covar=tensor([0.0568, 0.0697, 0.0730, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.1079, 0.1002, 0.0871, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 08:22:11,086 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=614971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:13,160 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=614974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:32,959 INFO [train.py:968] (1/2) Epoch 14, batch 21700, giga_loss[loss=0.2325, simple_loss=0.2993, pruned_loss=0.0828, over 28660.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3416, pruned_loss=0.09806, over 5721692.85 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3496, pruned_loss=0.09648, over 5783747.99 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.34, pruned_loss=0.09712, over 5714263.40 frames. ], batch size: 71, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:22:35,065 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=615003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:39,022 INFO [optim.py:369] (1/2) 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 08:22:41,873 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=615012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:44,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=615015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:22:58,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7970, 1.6095, 1.9267, 1.4754], device='cuda:1'), covar=tensor([0.2226, 0.3105, 0.1674, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0687, 0.0894, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 08:23:08,032 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 21750, giga_loss[loss=0.3062, simple_loss=0.3585, pruned_loss=0.127, over 23684.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3389, pruned_loss=0.09711, over 5706996.92 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3497, pruned_loss=0.09668, over 5775845.56 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3374, pruned_loss=0.09622, over 5706340.35 frames. ], batch size: 705, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:23:52,608 INFO [train.py:968] (1/2) Epoch 14, batch 21800, giga_loss[loss=0.2777, simple_loss=0.3552, pruned_loss=0.1001, over 28861.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.337, pruned_loss=0.09583, over 5710084.20 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3499, pruned_loss=0.09683, over 5776706.80 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3356, pruned_loss=0.095, over 5708258.20 frames. ], batch size: 186, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:23:58,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 08:23:59,377 INFO [optim.py:369] (1/2) 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 08:24:32,751 INFO [train.py:968] (1/2) Epoch 14, batch 21850, libri_loss[loss=0.2889, simple_loss=0.3675, pruned_loss=0.1052, over 29234.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3401, pruned_loss=0.09799, over 5712671.33 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3517, pruned_loss=0.09831, over 5782694.64 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3367, pruned_loss=0.09595, over 5702642.67 frames. ], batch size: 97, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:25:13,708 INFO [train.py:968] (1/2) Epoch 14, batch 21900, giga_loss[loss=0.2618, simple_loss=0.3447, pruned_loss=0.08946, over 28599.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3419, pruned_loss=0.09827, over 5715323.20 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3513, pruned_loss=0.09819, over 5784840.10 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3393, pruned_loss=0.09673, over 5703240.50 frames. ], batch size: 307, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:25:20,316 INFO [optim.py:369] (1/2) 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 08:25:35,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=615223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:25:43,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1560, 1.5319, 1.4391, 1.0644], device='cuda:1'), covar=tensor([0.1638, 0.2304, 0.1410, 0.1635], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0687, 0.0893, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 08:25:58,080 INFO [train.py:968] (1/2) Epoch 14, batch 21950, giga_loss[loss=0.2827, simple_loss=0.3635, pruned_loss=0.101, over 28338.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3435, pruned_loss=0.09822, over 5717626.91 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3519, pruned_loss=0.09873, over 5782826.58 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3408, pruned_loss=0.09652, over 5708327.82 frames. ], batch size: 368, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:26:36,668 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-07 08:26:38,783 INFO [train.py:968] (1/2) Epoch 14, batch 22000, giga_loss[loss=0.2785, simple_loss=0.3532, pruned_loss=0.1019, over 28859.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.345, pruned_loss=0.09821, over 5714298.58 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3519, pruned_loss=0.099, over 5786597.45 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3425, pruned_loss=0.09657, over 5701252.38 frames. ], batch size: 199, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:26:44,889 INFO [optim.py:369] (1/2) 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 08:27:17,793 INFO [train.py:968] (1/2) Epoch 14, batch 22050, giga_loss[loss=0.2428, simple_loss=0.3276, pruned_loss=0.07901, over 28902.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3451, pruned_loss=0.09796, over 5707224.83 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.352, pruned_loss=0.09927, over 5788007.18 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09632, over 5692627.39 frames. ], batch size: 174, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:27:57,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4238, 1.9201, 1.3645, 0.8466], device='cuda:1'), covar=tensor([0.4749, 0.2424, 0.3101, 0.5079], device='cuda:1'), in_proj_covar=tensor([0.1596, 0.1516, 0.1509, 0.1317], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 08:28:02,790 INFO [train.py:968] (1/2) Epoch 14, batch 22100, giga_loss[loss=0.2228, simple_loss=0.3091, pruned_loss=0.06826, over 28954.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3445, pruned_loss=0.09769, over 5707923.21 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3521, pruned_loss=0.09954, over 5788397.15 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.09619, over 5695883.76 frames. ], batch size: 128, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:28:09,418 INFO [optim.py:369] (1/2) 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 08:28:35,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.32 vs. limit=5.0
+2023-03-07 08:28:45,887 INFO [train.py:968] (1/2) Epoch 14, batch 22150, libri_loss[loss=0.2805, simple_loss=0.3421, pruned_loss=0.1095, over 28568.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.345, pruned_loss=0.09841, over 5707549.12 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3521, pruned_loss=0.09963, over 5787547.46 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3433, pruned_loss=0.09711, over 5697810.76 frames. ], batch size: 63, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:29:09,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-07 08:29:27,609 INFO [train.py:968] (1/2) Epoch 14, batch 22200, libri_loss[loss=0.3648, simple_loss=0.4155, pruned_loss=0.157, over 29532.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3462, pruned_loss=0.09925, over 5712628.24 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3524, pruned_loss=0.1, over 5789799.86 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3444, pruned_loss=0.09779, over 5701530.47 frames. ], batch size: 82, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:29:33,775 INFO [optim.py:369] (1/2) 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 08:30:08,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3464, 2.0054, 1.4839, 0.5451], device='cuda:1'), covar=tensor([0.3946, 0.2133, 0.3445, 0.4933], device='cuda:1'), in_proj_covar=tensor([0.1595, 0.1515, 0.1512, 0.1314], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 08:30:08,825 INFO [train.py:968] (1/2) Epoch 14, batch 22250, giga_loss[loss=0.2969, simple_loss=0.3712, pruned_loss=0.1114, over 28863.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3495, pruned_loss=0.1014, over 5702478.52 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3532, pruned_loss=0.1006, over 5787197.34 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3473, pruned_loss=0.09969, over 5694981.31 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:30:39,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2888, 1.5812, 1.3076, 0.9627], device='cuda:1'), covar=tensor([0.2313, 0.2355, 0.2675, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.1354, 0.0995, 0.1198, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 08:30:47,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=615598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:30:48,899 INFO [train.py:968] (1/2) Epoch 14, batch 22300, giga_loss[loss=0.2648, simple_loss=0.3447, pruned_loss=0.09246, over 28986.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3505, pruned_loss=0.1016, over 5710895.57 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3531, pruned_loss=0.1009, over 5791316.43 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3487, pruned_loss=0.1, over 5698741.08 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:30:54,499 INFO [optim.py:369] (1/2) 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 08:31:18,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5362, 1.6692, 1.8083, 1.3419], device='cuda:1'), covar=tensor([0.1736, 0.2196, 0.1435, 0.1611], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0686, 0.0894, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 08:31:28,697 INFO [train.py:968] (1/2) Epoch 14, batch 22350, giga_loss[loss=0.2879, simple_loss=0.3631, pruned_loss=0.1063, over 28888.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3524, pruned_loss=0.1026, over 5716171.03 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3535, pruned_loss=0.1013, over 5792335.99 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3506, pruned_loss=0.101, over 5704619.55 frames. ], batch size: 174, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:31:49,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1102, 3.9218, 3.7057, 1.9800], device='cuda:1'), covar=tensor([0.0619, 0.0762, 0.0734, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1082, 0.0999, 0.0870, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 08:31:52,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2180, 1.4440, 1.4250, 1.3037], device='cuda:1'), covar=tensor([0.1719, 0.1747, 0.2207, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0732, 0.0684, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 08:32:03,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5903, 1.4382, 1.5676, 1.1770], device='cuda:1'), covar=tensor([0.1600, 0.2557, 0.1365, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0850, 0.0685, 0.0892, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 08:32:07,115 INFO [train.py:968] (1/2) Epoch 14, batch 22400, giga_loss[loss=0.27, simple_loss=0.3468, pruned_loss=0.09662, over 28866.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3549, pruned_loss=0.1042, over 5703760.63 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3547, pruned_loss=0.1022, over 5775574.22 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3524, pruned_loss=0.1021, over 5707536.63 frames. ], batch size: 186, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:32:14,306 INFO [optim.py:369] (1/2) 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 08:32:42,881 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=615741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:32:44,882 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 22450, giga_loss[loss=0.2525, simple_loss=0.3347, pruned_loss=0.08513, over 29045.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3555, pruned_loss=0.1046, over 5707830.24 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3551, pruned_loss=0.1025, over 5774626.04 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3532, pruned_loss=0.1027, over 5710545.36 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:33:08,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=615773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:33:25,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7025, 1.7993, 1.7095, 1.5338], device='cuda:1'), covar=tensor([0.1504, 0.2047, 0.2026, 0.2044], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0729, 0.0682, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 08:33:30,368 INFO [train.py:968] (1/2) Epoch 14, batch 22500, giga_loss[loss=0.2671, simple_loss=0.3366, pruned_loss=0.09877, over 28735.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1035, over 5704740.54 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3553, pruned_loss=0.1028, over 5772939.58 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3514, pruned_loss=0.1017, over 5707078.12 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:33:35,581 INFO [optim.py:369] (1/2) 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 08:33:51,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=615826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:34:11,961 INFO [train.py:968] (1/2) Epoch 14, batch 22550, giga_loss[loss=0.2429, simple_loss=0.3174, pruned_loss=0.08425, over 28883.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3507, pruned_loss=0.1021, over 5696093.64 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3558, pruned_loss=0.1032, over 5763908.71 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3485, pruned_loss=0.1003, over 5705386.35 frames. ], batch size: 186, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:34:33,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3499, 1.6139, 1.3531, 1.4722], device='cuda:1'), covar=tensor([0.0731, 0.0301, 0.0323, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0113, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0095], device='cuda:1')
+2023-03-07 08:34:43,622 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-07 08:34:53,676 INFO [train.py:968] (1/2) Epoch 14, batch 22600, giga_loss[loss=0.2556, simple_loss=0.3295, pruned_loss=0.09088, over 28801.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3476, pruned_loss=0.1008, over 5697204.86 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3557, pruned_loss=0.1031, over 5764541.68 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3459, pruned_loss=0.0994, over 5702846.01 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:34:59,033 INFO [optim.py:369] (1/2) 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 08:35:28,996 INFO [train.py:968] (1/2) Epoch 14, batch 22650, giga_loss[loss=0.2398, simple_loss=0.3258, pruned_loss=0.07691, over 29026.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3479, pruned_loss=0.1004, over 5694919.58 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3568, pruned_loss=0.1042, over 5755858.24 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.345, pruned_loss=0.09811, over 5704079.42 frames. ], batch size: 128, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:36:14,229 INFO [train.py:968] (1/2) Epoch 14, batch 22700, giga_loss[loss=0.2688, simple_loss=0.3451, pruned_loss=0.09626, over 28796.00 frames. ], tot_loss[loss=0.275, simple_loss=0.35, pruned_loss=0.1, over 5695689.96 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3571, pruned_loss=0.1045, over 5757466.68 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3474, pruned_loss=0.09791, over 5700974.98 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:36:16,151 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 08:36:21,101 INFO [optim.py:369] (1/2) 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 08:36:52,660 INFO [train.py:968] (1/2) Epoch 14, batch 22750, giga_loss[loss=0.2666, simple_loss=0.335, pruned_loss=0.09911, over 28816.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3517, pruned_loss=0.101, over 5697131.32 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3577, pruned_loss=0.1051, over 5760982.23 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3489, pruned_loss=0.09864, over 5696220.10 frames. ], batch size: 112, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:37:24,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3862, 2.0367, 1.6137, 0.5943], device='cuda:1'), covar=tensor([0.4406, 0.2259, 0.3091, 0.5759], device='cuda:1'), in_proj_covar=tensor([0.1619, 0.1538, 0.1524, 0.1335], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 08:37:33,052 INFO [train.py:968] (1/2) Epoch 14, batch 22800, giga_loss[loss=0.2584, simple_loss=0.3346, pruned_loss=0.09112, over 29077.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3493, pruned_loss=0.1009, over 5698366.31 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3577, pruned_loss=0.1052, over 5763034.78 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3469, pruned_loss=0.09894, over 5695081.58 frames. ], batch size: 128, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:37:37,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5566, 1.7797, 1.7840, 1.3490], device='cuda:1'), covar=tensor([0.1543, 0.2172, 0.1353, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0845, 0.0682, 0.0887, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 08:37:43,821 INFO [optim.py:369] (1/2) 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 08:37:53,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2246, 0.7990, 0.7432, 1.4489], device='cuda:1'), covar=tensor([0.0738, 0.0363, 0.0375, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0113, 0.0113, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0086, 0.0061, 0.0055, 0.0095], device='cuda:1')
+2023-03-07 08:37:55,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5934, 1.5861, 1.2881, 1.2376], device='cuda:1'), covar=tensor([0.0757, 0.0529, 0.0984, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0436, 0.0503, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 08:38:02,473 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=616131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:38:15,425 INFO [train.py:968] (1/2) Epoch 14, batch 22850, giga_loss[loss=0.2773, simple_loss=0.3438, pruned_loss=0.1054, over 29062.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3482, pruned_loss=0.1019, over 5708131.87 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3578, pruned_loss=0.1053, over 5766545.33 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.346, pruned_loss=0.1001, over 5700500.94 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:38:20,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6833, 1.7962, 1.6086, 1.5521], device='cuda:1'), covar=tensor([0.1353, 0.2072, 0.1933, 0.2071], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0733, 0.0684, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 08:38:36,089 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=616176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:38:57,194 INFO [train.py:968] (1/2) Epoch 14, batch 22900, giga_loss[loss=0.2626, simple_loss=0.3326, pruned_loss=0.09635, over 28898.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3465, pruned_loss=0.102, over 5708099.89 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3581, pruned_loss=0.1055, over 5757352.40 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3443, pruned_loss=0.1003, over 5709123.30 frames. ], batch size: 199, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:38:58,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=616201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:39:01,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-07 08:39:04,000 INFO [optim.py:369] (1/2) 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 08:39:19,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7535, 1.5764, 1.2847, 1.2495], device='cuda:1'), covar=tensor([0.0729, 0.0639, 0.0960, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0437, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 08:39:36,049 INFO [train.py:968] (1/2) Epoch 14, batch 22950, giga_loss[loss=0.2606, simple_loss=0.3371, pruned_loss=0.09203, over 29063.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3463, pruned_loss=0.1028, over 5697409.54 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3584, pruned_loss=0.1059, over 5749447.01 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.344, pruned_loss=0.101, over 5703689.02 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:39:56,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8932, 2.4049, 1.9821, 1.6505], device='cuda:1'), covar=tensor([0.3257, 0.1925, 0.2363, 0.2624], device='cuda:1'), in_proj_covar=tensor([0.1810, 0.1720, 0.1668, 0.1777], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 08:40:14,917 INFO [train.py:968] (1/2) Epoch 14, batch 23000, giga_loss[loss=0.2653, simple_loss=0.3417, pruned_loss=0.09444, over 28538.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.345, pruned_loss=0.1018, over 5709841.90 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3586, pruned_loss=0.1063, over 5751927.35 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3427, pruned_loss=0.1, over 5711478.09 frames. ], batch size: 336, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:40:23,450 INFO [optim.py:369] (1/2) 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 08:40:25,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8462, 3.6247, 3.4288, 1.7488], device='cuda:1'), covar=tensor([0.0720, 0.0901, 0.0789, 0.2230], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.1011, 0.0880, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 08:40:25,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3644, 1.5942, 1.5982, 1.1905], device='cuda:1'), covar=tensor([0.1682, 0.2327, 0.1462, 0.1644], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0687, 0.0892, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 08:40:51,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=616344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:40:53,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=616347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:40:53,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-07 08:40:54,731 INFO [train.py:968] (1/2) Epoch 14, batch 23050, giga_loss[loss=0.2199, simple_loss=0.2946, pruned_loss=0.07257, over 28885.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3407, pruned_loss=0.09959, over 5710272.29 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3591, pruned_loss=0.1066, over 5754269.45 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3382, pruned_loss=0.09773, over 5708760.65 frames. ], batch size: 112, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:41:16,485 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=616376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:41:36,496 INFO [train.py:968] (1/2) Epoch 14, batch 23100, giga_loss[loss=0.2406, simple_loss=0.3164, pruned_loss=0.0824, over 28768.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3376, pruned_loss=0.09842, over 5702455.40 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3599, pruned_loss=0.1073, over 5752292.29 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3344, pruned_loss=0.09619, over 5701784.72 frames. ], batch size: 284, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:41:42,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-07 08:41:42,824 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 23150, giga_loss[loss=0.2353, simple_loss=0.3038, pruned_loss=0.08344, over 28826.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3378, pruned_loss=0.09829, over 5713783.94 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3601, pruned_loss=0.1075, over 5756950.51 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3344, pruned_loss=0.0961, over 5708061.29 frames. ], batch size: 145, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:42:42,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8576, 1.0596, 2.7933, 2.6867], device='cuda:1'), covar=tensor([0.1574, 0.2594, 0.0587, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0684, 0.0602, 0.0879, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 08:42:56,490 INFO [train.py:968] (1/2) Epoch 14, batch 23200, giga_loss[loss=0.2733, simple_loss=0.3534, pruned_loss=0.09654, over 28661.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3391, pruned_loss=0.09815, over 5714682.45 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3602, pruned_loss=0.1077, over 5757177.18 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3361, pruned_loss=0.09612, over 5709363.55 frames. ], batch size: 242, lr: 2.30e-03, grad_scale: 8.0
+2023-03-07 08:43:01,647 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:43:39,386 INFO [train.py:968] (1/2) Epoch 14, batch 23250, giga_loss[loss=0.2592, simple_loss=0.3395, pruned_loss=0.08945, over 29000.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3426, pruned_loss=0.09963, over 5715860.73 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3607, pruned_loss=0.1082, over 5761520.65 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.339, pruned_loss=0.09721, over 5706288.02 frames. ], batch size: 213, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:43:40,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=616551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:44:20,867 INFO [train.py:968] (1/2) Epoch 14, batch 23300, giga_loss[loss=0.2403, simple_loss=0.3312, pruned_loss=0.07476, over 28869.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3465, pruned_loss=0.1012, over 5716953.40 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3609, pruned_loss=0.1083, over 5762110.05 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3435, pruned_loss=0.0992, over 5708675.20 frames. ], batch size: 174, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:44:30,477 INFO [optim.py:369] (1/2) 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 08:45:03,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=616649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:03,582 INFO [train.py:968] (1/2) Epoch 14, batch 23350, giga_loss[loss=0.2885, simple_loss=0.357, pruned_loss=0.1099, over 28863.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3502, pruned_loss=0.103, over 5707661.98 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3619, pruned_loss=0.1092, over 5762304.14 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3466, pruned_loss=0.1004, over 5699590.78 frames. ], batch size: 119, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:45:05,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=616652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:18,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6299, 1.6813, 1.6679, 1.5141], device='cuda:1'), covar=tensor([0.1579, 0.2216, 0.2169, 0.2105], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0730, 0.0683, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 08:45:30,282 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=616681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:39,699 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=616694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:41,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=616697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:45:44,428 INFO [train.py:968] (1/2) Epoch 14, batch 23400, giga_loss[loss=0.2831, simple_loss=0.3501, pruned_loss=0.1081, over 28502.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3525, pruned_loss=0.1042, over 5705729.80 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3624, pruned_loss=0.1096, over 5764315.29 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3489, pruned_loss=0.1016, over 5696000.85 frames. ], batch size: 85, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:45:52,621 INFO [optim.py:369] (1/2) 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 08:46:06,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=616726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:46:27,579 INFO [train.py:968] (1/2) Epoch 14, batch 23450, giga_loss[loss=0.3244, simple_loss=0.3854, pruned_loss=0.1317, over 28987.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3581, pruned_loss=0.1098, over 5705238.40 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3617, pruned_loss=0.1093, over 5769449.02 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3555, pruned_loss=0.1078, over 5691136.28 frames. ], batch size: 155, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:46:32,984 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-07 08:46:48,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-07 08:47:19,726 INFO [train.py:968] (1/2) Epoch 14, batch 23500, giga_loss[loss=0.2681, simple_loss=0.3516, pruned_loss=0.09235, over 28938.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3629, pruned_loss=0.1136, over 5699148.67 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3617, pruned_loss=0.1095, over 5770396.10 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3609, pruned_loss=0.112, over 5686201.77 frames. ], batch size: 136, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:47:32,696 INFO [optim.py:369] (1/2) 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 08:48:07,271 INFO [train.py:968] (1/2) Epoch 14, batch 23550, giga_loss[loss=0.3459, simple_loss=0.4116, pruned_loss=0.1401, over 28751.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3696, pruned_loss=0.1188, over 5677572.18 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.362, pruned_loss=0.1098, over 5755321.33 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3679, pruned_loss=0.1174, over 5678753.32 frames. ], batch size: 242, lr: 2.30e-03, grad_scale: 2.0
+2023-03-07 08:48:29,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4618, 3.8040, 1.6078, 1.6097], device='cuda:1'), covar=tensor([0.0968, 0.0296, 0.0850, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0520, 0.0350, 0.0398], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 08:48:40,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5766, 1.7464, 1.4606, 1.7433], device='cuda:1'), covar=tensor([0.1944, 0.1899, 0.2003, 0.1781], device='cuda:1'), in_proj_covar=tensor([0.1355, 0.0999, 0.1202, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 08:48:55,823 INFO [train.py:968] (1/2) Epoch 14, batch 23600, giga_loss[loss=0.3735, simple_loss=0.4104, pruned_loss=0.1683, over 27628.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3743, pruned_loss=0.1229, over 5672488.39 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3625, pruned_loss=0.1103, over 5750436.48 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3726, pruned_loss=0.1216, over 5675795.55 frames. ], batch size: 472, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:49:05,417 INFO [optim.py:369] (1/2) 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 08:49:42,571 INFO [train.py:968] (1/2) Epoch 14, batch 23650, libri_loss[loss=0.3017, simple_loss=0.3696, pruned_loss=0.1169, over 19753.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3807, pruned_loss=0.1287, over 5653952.09 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3631, pruned_loss=0.1109, over 5740793.09 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3794, pruned_loss=0.1276, over 5663105.57 frames. ], batch size: 186, lr: 2.30e-03, grad_scale: 4.0
+2023-03-07 08:49:58,477 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-07 08:50:29,716 INFO [train.py:968] (1/2) Epoch 14, batch 23700, giga_loss[loss=0.2976, simple_loss=0.3754, pruned_loss=0.1099, over 28956.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3846, pruned_loss=0.1317, over 5653210.42 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3628, pruned_loss=0.1111, over 5736863.26 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3845, pruned_loss=0.1315, over 5661005.77 frames. ], batch size: 155, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:50:36,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2367, 1.1832, 1.1212, 0.8136], device='cuda:1'), covar=tensor([0.0778, 0.0473, 0.0972, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0437, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 08:50:39,710 INFO [optim.py:369] (1/2) 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 08:50:47,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6073, 2.2234, 1.6394, 1.8038], device='cuda:1'), covar=tensor([0.0679, 0.0239, 0.0283, 0.0746], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 08:51:16,153 INFO [train.py:968] (1/2) Epoch 14, batch 23750, giga_loss[loss=0.3165, simple_loss=0.3797, pruned_loss=0.1267, over 28897.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3861, pruned_loss=0.1339, over 5652037.54 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3625, pruned_loss=0.111, over 5738956.92 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3867, pruned_loss=0.1342, over 5654929.57 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:52:09,221 INFO [train.py:968] (1/2) Epoch 14, batch 23800, libri_loss[loss=0.3548, simple_loss=0.4036, pruned_loss=0.153, over 27991.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3878, pruned_loss=0.1366, over 5643064.52 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3627, pruned_loss=0.1113, over 5738705.84 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3888, pruned_loss=0.1372, over 5643489.70 frames. ], batch size: 116, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:52:19,427 INFO [optim.py:369] (1/2) 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 08:52:54,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-07 08:52:54,451 INFO [train.py:968] (1/2) Epoch 14, batch 23850, giga_loss[loss=0.3104, simple_loss=0.3702, pruned_loss=0.1253, over 28993.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3895, pruned_loss=0.1386, over 5647350.01 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3626, pruned_loss=0.1115, over 5745092.55 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3912, pruned_loss=0.1398, over 5638685.29 frames. ], batch size: 106, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:53:08,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 08:53:38,509 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:53:51,019 INFO [train.py:968] (1/2) Epoch 14, batch 23900, giga_loss[loss=0.323, simple_loss=0.3834, pruned_loss=0.1313, over 28383.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3939, pruned_loss=0.1427, over 5642251.87 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3626, pruned_loss=0.1118, over 5748412.32 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3962, pruned_loss=0.1442, over 5629810.42 frames. ], batch size: 85, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:54:06,425 INFO [optim.py:369] (1/2) 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 08:54:36,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3804, 1.4547, 1.4376, 1.4021], device='cuda:1'), covar=tensor([0.1894, 0.1796, 0.1486, 0.1609], device='cuda:1'), in_proj_covar=tensor([0.1797, 0.1719, 0.1656, 0.1776], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 08:54:45,745 INFO [train.py:968] (1/2) Epoch 14, batch 23950, giga_loss[loss=0.2744, simple_loss=0.3427, pruned_loss=0.1031, over 28528.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3937, pruned_loss=0.144, over 5611124.29 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3624, pruned_loss=0.1118, over 5741740.54 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3964, pruned_loss=0.1459, over 5604373.79 frames. ], batch size: 71, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:55:17,959 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 08:55:32,368 INFO [train.py:968] (1/2) Epoch 14, batch 24000, giga_loss[loss=0.4784, simple_loss=0.473, pruned_loss=0.2419, over 26600.00 frames. ], tot_loss[loss=0.34, simple_loss=0.3926, pruned_loss=0.1437, over 5625870.91 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3622, pruned_loss=0.1117, over 5743392.64 frames. ], giga_tot_loss[loss=0.3441, simple_loss=0.3958, pruned_loss=0.1462, over 5615733.01 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 08:55:32,368 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 08:55:38,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3012, 1.5680, 1.4510, 1.2197], device='cuda:1'), covar=tensor([0.2189, 0.1896, 0.1241, 0.1611], device='cuda:1'), in_proj_covar=tensor([0.1794, 0.1716, 0.1652, 0.1774], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 08:55:41,146 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 08:55:51,771 INFO [optim.py:369] (1/2) 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 08:56:29,828 INFO [train.py:968] (1/2) Epoch 14, batch 24050, giga_loss[loss=0.4113, simple_loss=0.4337, pruned_loss=0.1944, over 23616.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3909, pruned_loss=0.1421, over 5622231.50 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3622, pruned_loss=0.1117, over 5743918.96 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3939, pruned_loss=0.1445, over 5612429.57 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:57:16,888 INFO [train.py:968] (1/2) Epoch 14, batch 24100, giga_loss[loss=0.3531, simple_loss=0.4093, pruned_loss=0.1484, over 27823.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3911, pruned_loss=0.141, over 5621478.14 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3624, pruned_loss=0.112, over 5737781.89 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.3938, pruned_loss=0.1432, over 5616072.85 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:57:26,161 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3863, 1.6530, 1.6935, 1.2232], device='cuda:1'), covar=tensor([0.1652, 0.2312, 0.1330, 0.1593], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0686, 0.0887, 0.0790], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 08:57:30,130 INFO [optim.py:369] (1/2) 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 08:57:48,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-07 08:57:51,115 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 08:58:07,289 INFO [train.py:968] (1/2) Epoch 14, batch 24150, giga_loss[loss=0.3591, simple_loss=0.4201, pruned_loss=0.149, over 28932.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3934, pruned_loss=0.1423, over 5624798.75 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3627, pruned_loss=0.1124, over 5740474.93 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.396, pruned_loss=0.1445, over 5615385.25 frames. ], batch size: 174, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 08:58:57,265 INFO [train.py:968] (1/2) Epoch 14, batch 24200, libri_loss[loss=0.322, simple_loss=0.3839, pruned_loss=0.1301, over 29538.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3928, pruned_loss=0.1417, over 5635629.42 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.363, pruned_loss=0.1126, over 5745347.40 frames. ], giga_tot_loss[loss=0.3421, simple_loss=0.3957, pruned_loss=0.1442, over 5619845.40 frames. ], batch size: 80, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 08:59:12,834 INFO [optim.py:369] (1/2) 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 08:59:36,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3657, 1.6762, 1.3440, 1.3554], device='cuda:1'), covar=tensor([0.2133, 0.2050, 0.2245, 0.1860], device='cuda:1'), in_proj_covar=tensor([0.1357, 0.1000, 0.1201, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 08:59:45,206 INFO [train.py:968] (1/2) Epoch 14, batch 24250, giga_loss[loss=0.3246, simple_loss=0.3912, pruned_loss=0.129, over 28772.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3891, pruned_loss=0.1378, over 5640523.39 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3625, pruned_loss=0.1125, over 5751458.37 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3929, pruned_loss=0.1409, over 5618060.53 frames. ], batch size: 284, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 08:59:59,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=617562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:00:07,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5942, 2.3626, 1.7837, 0.9163], device='cuda:1'), covar=tensor([0.4359, 0.2531, 0.3491, 0.4669], device='cuda:1'), in_proj_covar=tensor([0.1605, 0.1534, 0.1523, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 09:00:34,426 INFO [train.py:968] (1/2) Epoch 14, batch 24300, libri_loss[loss=0.3323, simple_loss=0.3912, pruned_loss=0.1367, over 26032.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3875, pruned_loss=0.1356, over 5643752.95 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3629, pruned_loss=0.1128, over 5751297.72 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3907, pruned_loss=0.1383, over 5624386.04 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 09:00:35,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-07 09:00:50,626 INFO [optim.py:369] (1/2) 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 09:01:21,327 INFO [train.py:968] (1/2) Epoch 14, batch 24350, giga_loss[loss=0.3551, simple_loss=0.3968, pruned_loss=0.1567, over 26546.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.383, pruned_loss=0.1321, over 5642538.87 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3628, pruned_loss=0.1131, over 5753739.24 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3863, pruned_loss=0.1346, over 5621474.99 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 09:01:29,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5804, 1.8268, 1.8273, 1.3463], device='cuda:1'), covar=tensor([0.1590, 0.2550, 0.1405, 0.1756], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0685, 0.0887, 0.0790], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0012], device='cuda:1')
+2023-03-07 09:01:39,510 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:01:51,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8296, 2.6408, 1.7021, 0.8879], device='cuda:1'), covar=tensor([0.5913, 0.2847, 0.3320, 0.5919], device='cuda:1'), in_proj_covar=tensor([0.1598, 0.1528, 0.1516, 0.1323], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 09:02:09,115 INFO [train.py:968] (1/2) Epoch 14, batch 24400, giga_loss[loss=0.3023, simple_loss=0.3683, pruned_loss=0.1182, over 28661.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.38, pruned_loss=0.1297, over 5649056.15 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3629, pruned_loss=0.1132, over 5756425.03 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3829, pruned_loss=0.132, over 5627537.27 frames. ], batch size: 307, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:02:15,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=617705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:02:17,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=617708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:02:25,038 INFO [optim.py:369] (1/2) 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 09:02:46,878 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=617737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:02:57,729 INFO [train.py:968] (1/2) Epoch 14, batch 24450, giga_loss[loss=0.375, simple_loss=0.4155, pruned_loss=0.1673, over 27580.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3806, pruned_loss=0.1305, over 5636780.69 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3635, pruned_loss=0.1138, over 5747584.36 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3826, pruned_loss=0.132, over 5625609.63 frames. ], batch size: 472, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:03:42,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:03:49,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-07 09:03:51,674 INFO [train.py:968] (1/2) Epoch 14, batch 24500, giga_loss[loss=0.3, simple_loss=0.3723, pruned_loss=0.1138, over 28848.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3785, pruned_loss=0.1287, over 5624922.12 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.363, pruned_loss=0.1137, over 5731070.46 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.381, pruned_loss=0.1305, over 5626496.77 frames. ], batch size: 99, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:04:01,013 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 14, batch 24550, giga_loss[loss=0.2724, simple_loss=0.3472, pruned_loss=0.09884, over 28865.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3769, pruned_loss=0.1268, over 5646543.18 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3631, pruned_loss=0.1138, over 5734353.58 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3793, pruned_loss=0.1286, over 5642339.05 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:05:32,252 INFO [train.py:968] (1/2) Epoch 14, batch 24600, giga_loss[loss=0.2923, simple_loss=0.3749, pruned_loss=0.1048, over 28880.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3773, pruned_loss=0.1247, over 5652586.02 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3629, pruned_loss=0.1137, over 5737110.47 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3795, pruned_loss=0.1263, over 5645638.82 frames. ], batch size: 199, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:05:44,306 INFO [optim.py:369] (1/2) 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 09:05:49,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-07 09:06:01,024 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=617929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:06:12,231 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 09:06:22,934 INFO [train.py:968] (1/2) Epoch 14, batch 24650, giga_loss[loss=0.324, simple_loss=0.3912, pruned_loss=0.1284, over 28024.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3792, pruned_loss=0.1246, over 5651510.37 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3632, pruned_loss=0.1141, over 5730774.05 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3812, pruned_loss=0.1259, over 5649315.34 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:06:24,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=617951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:06:26,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=617954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:06:49,038 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-07 09:06:54,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=617983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:07:08,834 INFO [train.py:968] (1/2) Epoch 14, batch 24700, giga_loss[loss=0.3064, simple_loss=0.3686, pruned_loss=0.1221, over 28935.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3809, pruned_loss=0.1262, over 5650664.99 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3639, pruned_loss=0.1146, over 5725414.01 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3822, pruned_loss=0.127, over 5651663.49 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:07:23,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4025, 1.6697, 1.3086, 1.5948], device='cuda:1'), covar=tensor([0.2564, 0.2519, 0.2766, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.1360, 0.1001, 0.1206, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 09:07:23,444 INFO [optim.py:369] (1/2) 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 09:07:47,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:07:54,886 INFO [train.py:968] (1/2) Epoch 14, batch 24750, giga_loss[loss=0.3297, simple_loss=0.3911, pruned_loss=0.1341, over 28591.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3809, pruned_loss=0.1264, over 5671738.95 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3639, pruned_loss=0.1148, over 5730472.01 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3824, pruned_loss=0.1272, over 5666436.84 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:08:12,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6094, 1.8497, 1.4988, 1.8157], device='cuda:1'), covar=tensor([0.2357, 0.2465, 0.2702, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.1362, 0.1003, 0.1207, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 09:08:22,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-07 09:08:46,759 INFO [train.py:968] (1/2) Epoch 14, batch 24800, giga_loss[loss=0.2815, simple_loss=0.3468, pruned_loss=0.1081, over 28676.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3786, pruned_loss=0.1257, over 5678045.44 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3641, pruned_loss=0.1149, over 5728790.19 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3797, pruned_loss=0.1263, over 5674870.75 frames. ], batch size: 307, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:09:01,411 INFO [optim.py:369] (1/2) 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 09:09:30,698 INFO [train.py:968] (1/2) Epoch 14, batch 24850, giga_loss[loss=0.2912, simple_loss=0.3577, pruned_loss=0.1123, over 28577.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.378, pruned_loss=0.1265, over 5669981.58 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3644, pruned_loss=0.1151, over 5721499.47 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3789, pruned_loss=0.127, over 5672463.54 frames. ], batch size: 78, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:09:47,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1132, 1.1190, 3.6957, 3.1240], device='cuda:1'), covar=tensor([0.1681, 0.2695, 0.0446, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0693, 0.0608, 0.0886, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:09:47,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:09:56,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7559, 2.3750, 1.5684, 0.9469], device='cuda:1'), covar=tensor([0.5425, 0.3054, 0.2980, 0.5256], device='cuda:1'), in_proj_covar=tensor([0.1601, 0.1527, 0.1519, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 09:10:03,998 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:07,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:18,029 INFO [train.py:968] (1/2) Epoch 14, batch 24900, giga_loss[loss=0.2985, simple_loss=0.3621, pruned_loss=0.1174, over 28292.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3776, pruned_loss=0.1254, over 5672964.22 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3642, pruned_loss=0.1149, over 5724316.87 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3786, pruned_loss=0.1261, over 5671807.08 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:10:20,903 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:10:31,557 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:10:54,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-07 09:11:04,598 INFO [train.py:968] (1/2) Epoch 14, batch 24950, giga_loss[loss=0.3658, simple_loss=0.4163, pruned_loss=0.1577, over 28361.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3768, pruned_loss=0.1234, over 5683226.33 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.364, pruned_loss=0.1151, over 5726488.35 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3779, pruned_loss=0.1239, over 5679804.23 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:11:11,789 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:11:20,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6046, 2.2089, 1.6629, 0.7131], device='cuda:1'), covar=tensor([0.4134, 0.2381, 0.3410, 0.5385], device='cuda:1'), in_proj_covar=tensor([0.1600, 0.1526, 0.1519, 0.1326], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 09:11:28,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:11:48,697 INFO [train.py:968] (1/2) Epoch 14, batch 25000, giga_loss[loss=0.298, simple_loss=0.3669, pruned_loss=0.1146, over 28821.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3774, pruned_loss=0.1243, over 5664354.34 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3643, pruned_loss=0.1154, over 5709400.47 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3784, pruned_loss=0.1247, over 5674282.36 frames. ], batch size: 186, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:11:55,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:11:58,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:12:02,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:12:03,942 INFO [optim.py:369] (1/2) 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 09:12:30,379 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:12:37,996 INFO [train.py:968] (1/2) Epoch 14, batch 25050, giga_loss[loss=0.2862, simple_loss=0.349, pruned_loss=0.1117, over 28182.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3756, pruned_loss=0.1237, over 5672555.90 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3642, pruned_loss=0.1155, over 5714124.53 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3768, pruned_loss=0.1241, over 5675161.88 frames. ], batch size: 77, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:13:29,709 INFO [train.py:968] (1/2) Epoch 14, batch 25100, giga_loss[loss=0.3358, simple_loss=0.3765, pruned_loss=0.1476, over 23563.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3759, pruned_loss=0.1252, over 5660845.38 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3644, pruned_loss=0.1157, over 5716212.80 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3767, pruned_loss=0.1254, over 5660481.63 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:13:44,524 INFO [optim.py:369] (1/2) 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 09:13:51,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 09:14:13,553 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:14:14,106 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:14:17,129 INFO [train.py:968] (1/2) Epoch 14, batch 25150, giga_loss[loss=0.2555, simple_loss=0.3333, pruned_loss=0.0888, over 28975.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5664618.25 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3645, pruned_loss=0.1158, over 5715771.90 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3755, pruned_loss=0.1251, over 5663911.49 frames. ], batch size: 164, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:14:17,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:14:22,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3428, 1.2218, 3.5385, 3.1697], device='cuda:1'), covar=tensor([0.1429, 0.2606, 0.0438, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0612, 0.0895, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:14:42,575 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:15:01,315 INFO [train.py:968] (1/2) Epoch 14, batch 25200, giga_loss[loss=0.3391, simple_loss=0.3923, pruned_loss=0.143, over 27530.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1256, over 5670003.35 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3641, pruned_loss=0.1157, over 5722368.23 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3757, pruned_loss=0.1262, over 5661782.63 frames. ], batch size: 472, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:15:16,271 INFO [optim.py:369] (1/2) 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 09:15:36,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3713, 2.9993, 1.4291, 1.4941], device='cuda:1'), covar=tensor([0.0942, 0.0385, 0.0845, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0527, 0.0353, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 09:15:47,510 INFO [train.py:968] (1/2) Epoch 14, batch 25250, giga_loss[loss=0.2598, simple_loss=0.3274, pruned_loss=0.09607, over 28582.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3721, pruned_loss=0.1243, over 5669557.39 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3642, pruned_loss=0.1158, over 5718718.82 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3734, pruned_loss=0.125, over 5664260.96 frames. ], batch size: 85, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:16:13,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:16:24,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7089, 1.8154, 1.8086, 1.6571], device='cuda:1'), covar=tensor([0.1635, 0.2132, 0.2128, 0.2112], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0732, 0.0686, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 09:16:32,977 INFO [train.py:968] (1/2) Epoch 14, batch 25300, giga_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1145, over 28233.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.371, pruned_loss=0.1238, over 5658912.35 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3636, pruned_loss=0.1155, over 5712177.57 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3727, pruned_loss=0.1248, over 5660018.50 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:16:44,193 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:16:46,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4343, 1.6202, 1.4769, 1.2752], device='cuda:1'), covar=tensor([0.2554, 0.2122, 0.1543, 0.1972], device='cuda:1'), in_proj_covar=tensor([0.1811, 0.1738, 0.1668, 0.1793], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 09:16:48,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3725, 1.6262, 1.4507, 1.4142], device='cuda:1'), covar=tensor([0.1809, 0.1890, 0.2073, 0.1988], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0735, 0.0688, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 09:16:51,298 INFO [optim.py:369] (1/2) 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 09:17:06,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:17:20,112 INFO [train.py:968] (1/2) Epoch 14, batch 25350, giga_loss[loss=0.281, simple_loss=0.3554, pruned_loss=0.1033, over 28201.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1233, over 5661576.99 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.364, pruned_loss=0.1159, over 5712674.05 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3718, pruned_loss=0.1239, over 5660986.47 frames. ], batch size: 77, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:17:21,813 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 25400, giga_loss[loss=0.3282, simple_loss=0.3948, pruned_loss=0.1308, over 28868.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3714, pruned_loss=0.123, over 5665491.01 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3636, pruned_loss=0.1158, over 5717497.48 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3728, pruned_loss=0.1237, over 5659602.22 frames. ], batch size: 186, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:18:23,945 INFO [optim.py:369] (1/2) 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 09:18:26,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:18:31,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:18:51,661 INFO [train.py:968] (1/2) Epoch 14, batch 25450, giga_loss[loss=0.4022, simple_loss=0.4246, pruned_loss=0.1899, over 26480.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3709, pruned_loss=0.1216, over 5666344.42 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3638, pruned_loss=0.116, over 5713943.87 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1223, over 5663392.09 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:18:56,978 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:18:57,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2922, 1.8975, 1.3696, 0.4607], device='cuda:1'), covar=tensor([0.3858, 0.2193, 0.3292, 0.5164], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1535, 0.1520, 0.1327], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 09:19:14,415 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:18,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4979, 1.5539, 1.1858, 1.1153], device='cuda:1'), covar=tensor([0.0718, 0.0443, 0.0924, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0440, 0.0504, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:19:18,575 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:32,447 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:34,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:37,448 INFO [train.py:968] (1/2) Epoch 14, batch 25500, giga_loss[loss=0.3097, simple_loss=0.3777, pruned_loss=0.1209, over 28982.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3722, pruned_loss=0.1227, over 5658503.85 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3648, pruned_loss=0.1168, over 5713646.65 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1226, over 5654738.25 frames. ], batch size: 213, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:19:42,293 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:48,076 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:19:51,101 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:19:57,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:20:18,256 INFO [train.py:968] (1/2) Epoch 14, batch 25550, giga_loss[loss=0.293, simple_loss=0.3583, pruned_loss=0.1138, over 28797.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3735, pruned_loss=0.1243, over 5656088.89 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1172, over 5707103.57 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3737, pruned_loss=0.1241, over 5656923.32 frames. ], batch size: 66, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:20:22,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7159, 1.8234, 1.6009, 1.7344], device='cuda:1'), covar=tensor([0.1349, 0.1993, 0.1893, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0731, 0.0685, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 09:20:54,317 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=618888.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 09:21:02,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1578, 1.0243, 3.7318, 3.1987], device='cuda:1'), covar=tensor([0.2042, 0.3057, 0.0916, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0610, 0.0894, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:21:03,374 INFO [train.py:968] (1/2) Epoch 14, batch 25600, giga_loss[loss=0.3044, simple_loss=0.3733, pruned_loss=0.1178, over 29039.00 frames. ], tot_loss[loss=0.319, simple_loss=0.379, pruned_loss=0.1295, over 5651246.57 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3658, pruned_loss=0.1178, over 5710899.42 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3789, pruned_loss=0.1291, over 5646558.19 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:21:13,505 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:21:28,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1197, 1.7368, 1.3129, 0.3403], device='cuda:1'), covar=tensor([0.3118, 0.2034, 0.2688, 0.4451], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1547, 0.1529, 0.1332], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 09:21:51,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-07 09:21:54,908 INFO [train.py:968] (1/2) Epoch 14, batch 25650, giga_loss[loss=0.3145, simple_loss=0.3763, pruned_loss=0.1264, over 28608.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3787, pruned_loss=0.13, over 5664248.11 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3659, pruned_loss=0.1181, over 5714124.13 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3786, pruned_loss=0.1295, over 5656808.48 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:22:09,093 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=618964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:14,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=618967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:31,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=618985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:44,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=618996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:22:47,834 INFO [train.py:968] (1/2) Epoch 14, batch 25700, giga_loss[loss=0.472, simple_loss=0.4879, pruned_loss=0.228, over 24230.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3803, pruned_loss=0.1322, over 5652361.95 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.1179, over 5717241.73 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3806, pruned_loss=0.1322, over 5643050.20 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:23:02,889 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 25750, giga_loss[loss=0.3665, simple_loss=0.4056, pruned_loss=0.1637, over 27912.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3809, pruned_loss=0.1328, over 5660756.68 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3661, pruned_loss=0.1183, over 5723397.82 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3813, pruned_loss=0.133, over 5645891.20 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:23:57,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9010, 1.1141, 1.0560, 0.7862], device='cuda:1'), covar=tensor([0.2090, 0.2299, 0.1444, 0.1886], device='cuda:1'), in_proj_covar=tensor([0.1803, 0.1730, 0.1667, 0.1790], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 09:24:11,107 INFO [train.py:968] (1/2) Epoch 14, batch 25800, giga_loss[loss=0.3299, simple_loss=0.3836, pruned_loss=0.1381, over 27923.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3794, pruned_loss=0.1315, over 5676144.63 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.366, pruned_loss=0.1184, over 5733055.73 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3806, pruned_loss=0.1323, over 5652054.55 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:24:27,227 INFO [optim.py:369] (1/2) 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 09:24:36,484 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:24:38,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:24:47,792 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:24:53,500 INFO [train.py:968] (1/2) Epoch 14, batch 25850, giga_loss[loss=0.3007, simple_loss=0.3701, pruned_loss=0.1156, over 28907.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3786, pruned_loss=0.1298, over 5671721.19 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3658, pruned_loss=0.1185, over 5724725.51 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3801, pruned_loss=0.1308, over 5657840.56 frames. ], batch size: 112, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:25:03,638 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:25:16,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4333, 1.6876, 1.6115, 1.3746], device='cuda:1'), covar=tensor([0.2624, 0.2084, 0.1672, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.1798, 0.1724, 0.1662, 0.1782], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 09:25:32,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:25:33,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4785, 1.6464, 1.3053, 1.2161], device='cuda:1'), covar=tensor([0.0816, 0.0455, 0.0918, 0.1000], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0440, 0.0502, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:25:35,050 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:25:43,067 INFO [train.py:968] (1/2) Epoch 14, batch 25900, giga_loss[loss=0.3184, simple_loss=0.3715, pruned_loss=0.1327, over 27957.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3749, pruned_loss=0.1262, over 5667750.37 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3658, pruned_loss=0.1184, over 5726906.12 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3763, pruned_loss=0.1272, over 5653963.81 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:25:57,857 INFO [optim.py:369] (1/2) 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 09:26:20,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:26:28,951 INFO [train.py:968] (1/2) Epoch 14, batch 25950, giga_loss[loss=0.2803, simple_loss=0.3465, pruned_loss=0.1071, over 28946.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3724, pruned_loss=0.1245, over 5662276.69 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1183, over 5719535.27 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3739, pruned_loss=0.1256, over 5655538.83 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:26:42,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619263.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 09:26:42,974 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-07 09:27:00,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:13,710 INFO [train.py:968] (1/2) Epoch 14, batch 26000, giga_loss[loss=0.2778, simple_loss=0.3447, pruned_loss=0.1055, over 28883.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3711, pruned_loss=0.1243, over 5676193.67 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3659, pruned_loss=0.1186, over 5724292.01 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3723, pruned_loss=0.1251, over 5664625.42 frames. ], batch size: 119, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:27:30,287 INFO [optim.py:369] (1/2) 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 09:27:48,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:49,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:50,933 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:27:53,314 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 26050, giga_loss[loss=0.2861, simple_loss=0.3497, pruned_loss=0.1113, over 28336.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.371, pruned_loss=0.1242, over 5669411.09 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3661, pruned_loss=0.119, over 5714556.60 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3718, pruned_loss=0.1246, over 5666939.48 frames. ], batch size: 65, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:28:17,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:28:20,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:28:50,209 INFO [train.py:968] (1/2) Epoch 14, batch 26100, libri_loss[loss=0.35, simple_loss=0.3964, pruned_loss=0.1518, over 19389.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3738, pruned_loss=0.1251, over 5668820.09 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3665, pruned_loss=0.1192, over 5708471.70 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5672318.50 frames. ], batch size: 187, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:28:56,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619406.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 09:28:58,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619409.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 09:29:07,122 INFO [optim.py:369] (1/2) 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 09:29:17,203 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:29:19,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:29:26,352 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619438.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 09:29:34,458 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-07 09:29:37,565 INFO [train.py:968] (1/2) Epoch 14, batch 26150, giga_loss[loss=0.2881, simple_loss=0.3651, pruned_loss=0.1056, over 28524.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3758, pruned_loss=0.123, over 5678639.41 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3661, pruned_loss=0.1189, over 5713604.10 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3766, pruned_loss=0.1235, over 5676336.07 frames. ], batch size: 60, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:29:48,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:30:24,814 INFO [train.py:968] (1/2) Epoch 14, batch 26200, giga_loss[loss=0.3028, simple_loss=0.3744, pruned_loss=0.1156, over 28943.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.377, pruned_loss=0.1224, over 5685584.92 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3663, pruned_loss=0.119, over 5718981.37 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3777, pruned_loss=0.1229, over 5677881.16 frames. ], batch size: 227, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:30:30,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1775, 1.4131, 1.4194, 1.0837], device='cuda:1'), covar=tensor([0.1279, 0.2059, 0.1061, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0693, 0.0894, 0.0795], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 09:30:40,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=619518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:30:40,712 INFO [optim.py:369] (1/2) 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 09:31:12,148 INFO [train.py:968] (1/2) Epoch 14, batch 26250, libri_loss[loss=0.3654, simple_loss=0.3924, pruned_loss=0.1693, over 29647.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3802, pruned_loss=0.1261, over 5687212.14 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3663, pruned_loss=0.1193, over 5723508.69 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3812, pruned_loss=0.1263, over 5676205.50 frames. ], batch size: 69, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:31:43,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2914, 2.0971, 2.1136, 1.8640], device='cuda:1'), covar=tensor([0.1662, 0.2621, 0.2252, 0.2349], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0734, 0.0687, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 09:31:55,472 INFO [train.py:968] (1/2) Epoch 14, batch 26300, giga_loss[loss=0.296, simple_loss=0.3666, pruned_loss=0.1127, over 29028.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3811, pruned_loss=0.1268, over 5694487.92 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3665, pruned_loss=0.1193, over 5727173.48 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3819, pruned_loss=0.1271, over 5682091.60 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:32:01,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5764, 1.6853, 1.4556, 1.5858], device='cuda:1'), covar=tensor([0.2170, 0.2043, 0.2060, 0.1870], device='cuda:1'), in_proj_covar=tensor([0.1360, 0.1001, 0.1204, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 09:32:15,024 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:32:24,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 09:32:45,067 INFO [train.py:968] (1/2) Epoch 14, batch 26350, giga_loss[loss=0.3652, simple_loss=0.4206, pruned_loss=0.155, over 28630.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.382, pruned_loss=0.1291, over 5686126.36 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3664, pruned_loss=0.1194, over 5729015.28 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.383, pruned_loss=0.1295, over 5673729.58 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:32:55,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4794, 1.6041, 1.4970, 1.4206], device='cuda:1'), covar=tensor([0.1556, 0.2100, 0.2187, 0.1980], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0738, 0.0690, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 09:32:55,787 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:32:58,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=619664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:33:00,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:33:27,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:33:31,895 INFO [train.py:968] (1/2) Epoch 14, batch 26400, giga_loss[loss=0.2791, simple_loss=0.3519, pruned_loss=0.1031, over 28915.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3807, pruned_loss=0.1289, over 5688539.81 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3659, pruned_loss=0.1189, over 5723455.18 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3823, pruned_loss=0.1297, over 5682881.48 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:33:49,059 INFO [optim.py:369] (1/2) 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 09:34:20,486 INFO [train.py:968] (1/2) Epoch 14, batch 26450, giga_loss[loss=0.3996, simple_loss=0.44, pruned_loss=0.1796, over 28609.00 frames. ], tot_loss[loss=0.318, simple_loss=0.379, pruned_loss=0.1285, over 5676960.97 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5714114.86 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3804, pruned_loss=0.1293, over 5680373.23 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:34:28,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7014, 3.5263, 3.3611, 1.6772], device='cuda:1'), covar=tensor([0.0740, 0.0834, 0.0808, 0.2177], device='cuda:1'), in_proj_covar=tensor([0.1129, 0.1048, 0.0912, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 09:34:28,599 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=619760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:34:32,758 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=619772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:35:00,214 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=619792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:35:08,273 INFO [train.py:968] (1/2) Epoch 14, batch 26500, giga_loss[loss=0.2948, simple_loss=0.3675, pruned_loss=0.111, over 28657.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3778, pruned_loss=0.1282, over 5674113.75 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3666, pruned_loss=0.1194, over 5707710.49 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3787, pruned_loss=0.1287, over 5681264.53 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:35:24,973 INFO [optim.py:369] (1/2) 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 09:35:53,852 INFO [train.py:968] (1/2) Epoch 14, batch 26550, giga_loss[loss=0.2914, simple_loss=0.3592, pruned_loss=0.1118, over 28790.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3785, pruned_loss=0.1291, over 5675472.44 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3662, pruned_loss=0.1191, over 5711609.03 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3799, pruned_loss=0.13, over 5676912.73 frames. ], batch size: 119, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:35:58,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2903, 1.5327, 1.5338, 1.2043], device='cuda:1'), covar=tensor([0.2079, 0.1762, 0.1265, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.1798, 0.1719, 0.1657, 0.1781], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 09:36:06,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4025, 2.9517, 1.5246, 1.4757], device='cuda:1'), covar=tensor([0.0876, 0.0326, 0.0815, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0528, 0.0355, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 09:36:40,827 INFO [train.py:968] (1/2) Epoch 14, batch 26600, giga_loss[loss=0.3543, simple_loss=0.3846, pruned_loss=0.1619, over 23727.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3782, pruned_loss=0.1296, over 5674911.15 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3659, pruned_loss=0.1189, over 5714203.40 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3796, pruned_loss=0.1305, over 5673338.71 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:36:57,707 INFO [optim.py:369] (1/2) 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 09:37:21,889 INFO [train.py:968] (1/2) Epoch 14, batch 26650, giga_loss[loss=0.3051, simple_loss=0.3716, pruned_loss=0.1193, over 28725.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3751, pruned_loss=0.1281, over 5665332.15 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5718180.53 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3766, pruned_loss=0.1292, over 5658611.06 frames. ], batch size: 262, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:38:12,051 INFO [train.py:968] (1/2) Epoch 14, batch 26700, giga_loss[loss=0.261, simple_loss=0.3366, pruned_loss=0.0927, over 28939.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3742, pruned_loss=0.1271, over 5651778.18 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5709088.15 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3757, pruned_loss=0.1283, over 5652963.33 frames. ], batch size: 106, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:38:29,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 09:38:29,532 INFO [optim.py:369] (1/2) 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 09:38:40,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:38:50,833 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 26750, giga_loss[loss=0.2779, simple_loss=0.3533, pruned_loss=0.1012, over 28965.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3741, pruned_loss=0.1255, over 5662658.38 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3656, pruned_loss=0.1186, over 5712875.90 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3757, pruned_loss=0.1269, over 5659224.95 frames. ], batch size: 213, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:39:41,898 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:39:51,300 INFO [train.py:968] (1/2) Epoch 14, batch 26800, giga_loss[loss=0.3538, simple_loss=0.4045, pruned_loss=0.1516, over 28184.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5659251.40 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1187, over 5715794.91 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.378, pruned_loss=0.1287, over 5652898.59 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:40:00,841 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:40:02,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3993, 1.7007, 1.6417, 1.2133], device='cuda:1'), covar=tensor([0.1525, 0.2285, 0.1290, 0.1523], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0695, 0.0892, 0.0793], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 09:40:09,410 INFO [optim.py:369] (1/2) 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 09:40:26,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9621, 1.2367, 1.3421, 1.0481], device='cuda:1'), covar=tensor([0.1620, 0.1263, 0.2097, 0.1636], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0742, 0.0692, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 09:40:36,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:40:38,135 INFO [train.py:968] (1/2) Epoch 14, batch 26850, giga_loss[loss=0.3276, simple_loss=0.4014, pruned_loss=0.1269, over 29000.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.377, pruned_loss=0.1274, over 5671257.25 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3658, pruned_loss=0.1187, over 5717420.06 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3782, pruned_loss=0.1284, over 5664409.39 frames. ], batch size: 213, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:40:44,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-07 09:40:54,099 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:06,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:09,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:20,644 INFO [train.py:968] (1/2) Epoch 14, batch 26900, giga_loss[loss=0.3474, simple_loss=0.4075, pruned_loss=0.1437, over 28643.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3783, pruned_loss=0.1255, over 5677955.56 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3661, pruned_loss=0.1188, over 5724021.27 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3794, pruned_loss=0.1264, over 5665236.83 frames. ], batch size: 307, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:41:36,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:38,804 INFO [optim.py:369] (1/2) 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 09:41:49,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:41:55,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:42:04,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2319, 0.8195, 0.8712, 1.3997], device='cuda:1'), covar=tensor([0.0785, 0.0377, 0.0377, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 09:42:06,582 INFO [train.py:968] (1/2) Epoch 14, batch 26950, giga_loss[loss=0.4703, simple_loss=0.472, pruned_loss=0.2343, over 26480.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3811, pruned_loss=0.1255, over 5678914.00 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3663, pruned_loss=0.119, over 5716491.09 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3821, pruned_loss=0.1262, over 5674472.68 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:42:29,933 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-07 09:42:41,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:42:43,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:42:49,864 INFO [train.py:968] (1/2) Epoch 14, batch 27000, giga_loss[loss=0.3822, simple_loss=0.4293, pruned_loss=0.1675, over 28537.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3839, pruned_loss=0.1274, over 5686601.25 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3659, pruned_loss=0.1188, over 5721338.26 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3855, pruned_loss=0.1283, over 5677844.20 frames. ], batch size: 336, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:42:49,865 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 09:42:57,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4663, 1.8474, 1.4395, 1.3351], device='cuda:1'), covar=tensor([0.2762, 0.2578, 0.2762, 0.2276], device='cuda:1'), in_proj_covar=tensor([0.1366, 0.1005, 0.1210, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 09:42:58,437 INFO [train.py:1012] (1/2) Epoch 14, validation: loss=0.2106, simple_loss=0.3165, pruned_loss=0.05238, over 944034.00 frames. 
+2023-03-07 09:42:58,437 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 09:43:15,408 INFO [optim.py:369] (1/2) 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 09:43:16,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:43:17,217 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:43:36,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6536, 1.8955, 1.1375, 1.6472], device='cuda:1'), covar=tensor([0.1070, 0.0840, 0.1559, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0444, 0.0505, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:43:44,900 INFO [train.py:968] (1/2) Epoch 14, batch 27050, giga_loss[loss=0.3342, simple_loss=0.3911, pruned_loss=0.1387, over 28915.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3856, pruned_loss=0.1299, over 5684313.16 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3657, pruned_loss=0.1187, over 5726069.97 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3875, pruned_loss=0.1309, over 5672383.89 frames. ], batch size: 174, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:44:12,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6424, 1.8413, 1.4842, 1.8304], device='cuda:1'), covar=tensor([0.2384, 0.2436, 0.2678, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.1364, 0.1005, 0.1207, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 09:44:14,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4098, 2.8289, 1.5166, 1.4441], device='cuda:1'), covar=tensor([0.0801, 0.0345, 0.0760, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0529, 0.0356, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 09:44:37,319 INFO [train.py:968] (1/2) Epoch 14, batch 27100, giga_loss[loss=0.3062, simple_loss=0.3723, pruned_loss=0.12, over 28864.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3884, pruned_loss=0.1335, over 5661525.79 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1188, over 5728919.26 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3901, pruned_loss=0.1345, over 5648676.57 frames. ], batch size: 112, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:44:41,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:44:55,038 INFO [optim.py:369] (1/2) 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 09:45:13,326 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 27150, giga_loss[loss=0.2511, simple_loss=0.342, pruned_loss=0.08012, over 28889.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3858, pruned_loss=0.1315, over 5670380.83 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1188, over 5729752.51 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3873, pruned_loss=0.1324, over 5659055.15 frames. ], batch size: 174, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:45:28,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:45:44,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:45:48,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4260, 2.6035, 1.5091, 1.5832], device='cuda:1'), covar=tensor([0.0754, 0.0311, 0.0693, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0527, 0.0355, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 09:45:58,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:10,145 INFO [train.py:968] (1/2) Epoch 14, batch 27200, giga_loss[loss=0.4314, simple_loss=0.4404, pruned_loss=0.2112, over 26760.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3848, pruned_loss=0.1311, over 5658244.31 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3659, pruned_loss=0.119, over 5733506.34 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3868, pruned_loss=0.1322, over 5642624.87 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:46:30,378 INFO [optim.py:369] (1/2) 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 09:46:40,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:49,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:51,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:46:54,942 INFO [train.py:968] (1/2) Epoch 14, batch 27250, giga_loss[loss=0.3434, simple_loss=0.4124, pruned_loss=0.1372, over 28874.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.383, pruned_loss=0.1277, over 5674161.36 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3659, pruned_loss=0.1192, over 5737084.83 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.385, pruned_loss=0.1287, over 5656798.62 frames. ], batch size: 199, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:46:55,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:22,558 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:39,711 INFO [train.py:968] (1/2) Epoch 14, batch 27300, giga_loss[loss=0.3043, simple_loss=0.3744, pruned_loss=0.1171, over 28682.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3825, pruned_loss=0.1262, over 5675529.68 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1195, over 5729292.55 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3842, pruned_loss=0.127, over 5666354.08 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:47:47,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:50,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:51,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:47:53,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:01,789 INFO [optim.py:369] (1/2) 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 09:48:07,208 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:09,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:19,833 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:48:28,927 INFO [train.py:968] (1/2) Epoch 14, batch 27350, giga_loss[loss=0.3541, simple_loss=0.4028, pruned_loss=0.1527, over 28276.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3826, pruned_loss=0.1267, over 5671747.67 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1194, over 5732771.05 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3846, pruned_loss=0.1276, over 5658666.88 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:48:38,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:05,474 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:08,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:12,010 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 27400, giga_loss[loss=0.2813, simple_loss=0.3543, pruned_loss=0.1041, over 28556.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3828, pruned_loss=0.1274, over 5673441.08 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1195, over 5726499.73 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3847, pruned_loss=0.1282, over 5667250.58 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:49:16,338 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:34,585 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:49:42,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:49:45,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 09:50:04,315 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:04,577 INFO [train.py:968] (1/2) Epoch 14, batch 27450, giga_loss[loss=0.4052, simple_loss=0.4099, pruned_loss=0.2003, over 23358.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.38, pruned_loss=0.1271, over 5657264.92 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3665, pruned_loss=0.1195, over 5726207.23 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3819, pruned_loss=0.1279, over 5651270.02 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:50:06,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:07,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:12,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:34,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:41,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:50:55,328 INFO [train.py:968] (1/2) Epoch 14, batch 27500, giga_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 28271.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3784, pruned_loss=0.127, over 5649901.75 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1194, over 5726882.47 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.38, pruned_loss=0.1278, over 5644036.68 frames. ], batch size: 368, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:51:11,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:51:17,898 INFO [optim.py:369] (1/2) 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 09:51:23,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:51:34,929 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:51:40,097 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 27550, giga_loss[loss=0.3447, simple_loss=0.3888, pruned_loss=0.1503, over 26668.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3765, pruned_loss=0.1261, over 5655977.45 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3669, pruned_loss=0.1196, over 5729266.75 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3776, pruned_loss=0.1268, over 5647470.30 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:51:46,004 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4552, 4.2048, 1.6632, 1.5230], device='cuda:1'), covar=tensor([0.0989, 0.0281, 0.0888, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0531, 0.0356, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 09:52:02,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2979, 2.1106, 1.7331, 1.4562], device='cuda:1'), covar=tensor([0.0821, 0.0261, 0.0272, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 09:52:07,182 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 27600, libri_loss[loss=0.2799, simple_loss=0.3504, pruned_loss=0.1047, over 29547.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3767, pruned_loss=0.1279, over 5648088.77 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3674, pruned_loss=0.1199, over 5724744.17 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3775, pruned_loss=0.1283, over 5643525.18 frames. ], batch size: 79, lr: 2.29e-03, grad_scale: 8.0
+2023-03-07 09:52:35,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=620906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:52:47,876 INFO [optim.py:369] (1/2) 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 09:53:14,897 INFO [train.py:968] (1/2) Epoch 14, batch 27650, giga_loss[loss=0.3073, simple_loss=0.3683, pruned_loss=0.1232, over 29037.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3757, pruned_loss=0.1273, over 5652696.54 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3672, pruned_loss=0.1196, over 5728299.06 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3767, pruned_loss=0.1281, over 5644217.20 frames. ], batch size: 213, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:53:19,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:22,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:22,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=620959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:31,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=620969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:33,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=620972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:50,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=620987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:53:59,798 INFO [train.py:968] (1/2) Epoch 14, batch 27700, giga_loss[loss=0.2454, simple_loss=0.3267, pruned_loss=0.08206, over 29052.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3716, pruned_loss=0.1226, over 5662470.29 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.367, pruned_loss=0.1195, over 5730413.25 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3727, pruned_loss=0.1234, over 5652821.29 frames. ], batch size: 164, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:54:02,417 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:54:47,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:54:47,960 INFO [train.py:968] (1/2) Epoch 14, batch 27750, giga_loss[loss=0.3967, simple_loss=0.4432, pruned_loss=0.1751, over 27984.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5650155.56 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3672, pruned_loss=0.1198, over 5711451.14 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5657115.84 frames. ], batch size: 412, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:54:49,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:55:13,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:55:19,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:55:30,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3173, 3.3897, 1.4213, 1.4899], device='cuda:1'), covar=tensor([0.0956, 0.0349, 0.0876, 0.1314], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0528, 0.0354, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 09:55:33,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4161, 2.1275, 1.6158, 0.5942], device='cuda:1'), covar=tensor([0.4657, 0.2459, 0.3431, 0.5331], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1555, 0.1524, 0.1331], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 09:55:36,267 INFO [train.py:968] (1/2) Epoch 14, batch 27800, giga_loss[loss=0.2872, simple_loss=0.3555, pruned_loss=0.1095, over 28921.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3685, pruned_loss=0.1195, over 5640901.47 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3675, pruned_loss=0.12, over 5704546.71 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3689, pruned_loss=0.1196, over 5651632.46 frames. ], batch size: 145, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:55:39,733 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:55:42,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3307, 1.4296, 3.3153, 3.1494], device='cuda:1'), covar=tensor([0.1284, 0.2338, 0.0437, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0613, 0.0895, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:55:55,793 INFO [optim.py:369] (1/2) 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 09:56:22,943 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=621148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:56:24,428 INFO [train.py:968] (1/2) Epoch 14, batch 27850, giga_loss[loss=0.2823, simple_loss=0.3467, pruned_loss=0.1089, over 28914.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5638181.40 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3672, pruned_loss=0.1199, over 5692443.90 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3654, pruned_loss=0.118, over 5654712.00 frames. ], batch size: 112, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:57:16,114 INFO [train.py:968] (1/2) Epoch 14, batch 27900, giga_loss[loss=0.3428, simple_loss=0.3917, pruned_loss=0.1469, over 27511.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3646, pruned_loss=0.1183, over 5641586.78 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3678, pruned_loss=0.1203, over 5695490.18 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3645, pruned_loss=0.118, over 5650976.06 frames. ], batch size: 472, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:57:25,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5411, 1.2105, 4.8779, 3.4577], device='cuda:1'), covar=tensor([0.1738, 0.2893, 0.0396, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0697, 0.0613, 0.0895, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 09:57:33,585 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:35,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:35,905 INFO [optim.py:369] (1/2) 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 09:57:54,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:56,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:57:57,912 INFO [train.py:968] (1/2) Epoch 14, batch 27950, libri_loss[loss=0.3294, simple_loss=0.3878, pruned_loss=0.1355, over 27548.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.367, pruned_loss=0.119, over 5659065.69 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3676, pruned_loss=0.1202, over 5694908.85 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3669, pruned_loss=0.1188, over 5665632.63 frames. ], batch size: 116, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 09:57:58,949 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:58:45,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9077, 2.0817, 1.7299, 2.2905], device='cuda:1'), covar=tensor([0.2244, 0.2210, 0.2428, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.1364, 0.1005, 0.1207, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 09:58:45,956 INFO [train.py:968] (1/2) Epoch 14, batch 28000, giga_loss[loss=0.2861, simple_loss=0.3578, pruned_loss=0.1073, over 28603.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3696, pruned_loss=0.1208, over 5655055.85 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3672, pruned_loss=0.1201, over 5700153.86 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3699, pruned_loss=0.1207, over 5654336.44 frames. ], batch size: 78, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 09:59:09,714 INFO [optim.py:369] (1/2) 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 09:59:20,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=621334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 09:59:35,809 INFO [train.py:968] (1/2) Epoch 14, batch 28050, giga_loss[loss=0.307, simple_loss=0.3766, pruned_loss=0.1187, over 28812.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3714, pruned_loss=0.1223, over 5649953.93 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3672, pruned_loss=0.12, over 5702646.31 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1223, over 5646214.27 frames. ], batch size: 186, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:00:24,125 INFO [train.py:968] (1/2) Epoch 14, batch 28100, giga_loss[loss=0.2735, simple_loss=0.3525, pruned_loss=0.09725, over 28935.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1234, over 5648218.02 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.367, pruned_loss=0.1198, over 5704540.78 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3728, pruned_loss=0.1237, over 5643040.30 frames. ], batch size: 155, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:00:44,055 INFO [optim.py:369] (1/2) 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 10:00:48,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9492, 2.9419, 1.1148, 1.2492], device='cuda:1'), covar=tensor([0.1320, 0.0490, 0.1024, 0.1647], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0527, 0.0354, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 10:01:06,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=621449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:01:06,571 INFO [train.py:968] (1/2) Epoch 14, batch 28150, giga_loss[loss=0.264, simple_loss=0.3365, pruned_loss=0.09571, over 28814.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1246, over 5670993.35 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3672, pruned_loss=0.1199, over 5709971.68 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1248, over 5660470.74 frames. ], batch size: 66, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:01:32,345 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:01:34,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:01:54,946 INFO [train.py:968] (1/2) Epoch 14, batch 28200, giga_loss[loss=0.2977, simple_loss=0.3706, pruned_loss=0.1125, over 28893.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3775, pruned_loss=0.1269, over 5656654.67 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3676, pruned_loss=0.1204, over 5702031.82 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3775, pruned_loss=0.1268, over 5654771.49 frames. ], batch size: 174, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:02:02,824 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:02:15,292 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:02:43,673 INFO [train.py:968] (1/2) Epoch 14, batch 28250, giga_loss[loss=0.4024, simple_loss=0.44, pruned_loss=0.1824, over 27564.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3777, pruned_loss=0.1269, over 5659764.23 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3678, pruned_loss=0.1204, over 5706036.68 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3778, pruned_loss=0.1269, over 5653571.16 frames. ], batch size: 472, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:03:13,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4386, 1.9323, 1.4076, 1.6444], device='cuda:1'), covar=tensor([0.2599, 0.2411, 0.2737, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1369, 0.1009, 0.1212, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 10:03:29,024 INFO [train.py:968] (1/2) Epoch 14, batch 28300, giga_loss[loss=0.3041, simple_loss=0.3733, pruned_loss=0.1174, over 28886.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3768, pruned_loss=0.1266, over 5656185.24 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3674, pruned_loss=0.1202, over 5709699.01 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3774, pruned_loss=0.1269, over 5646991.41 frames. ], batch size: 186, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:03:54,719 INFO [optim.py:369] (1/2) 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 10:04:19,162 INFO [train.py:968] (1/2) Epoch 14, batch 28350, giga_loss[loss=0.301, simple_loss=0.3805, pruned_loss=0.1107, over 28979.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3775, pruned_loss=0.1262, over 5650931.74 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3675, pruned_loss=0.1203, over 5701753.62 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3782, pruned_loss=0.1265, over 5648567.67 frames. ], batch size: 136, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:04:37,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:04:39,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:04:41,786 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-07 10:05:02,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-07 10:05:10,588 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 28400, giga_loss[loss=0.2687, simple_loss=0.3502, pruned_loss=0.09367, over 28898.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3768, pruned_loss=0.1242, over 5655681.26 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.367, pruned_loss=0.1201, over 5701801.25 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3779, pruned_loss=0.1247, over 5652862.92 frames. ], batch size: 145, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 10:05:30,071 INFO [optim.py:369] (1/2) 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 10:05:53,832 INFO [train.py:968] (1/2) Epoch 14, batch 28450, giga_loss[loss=0.3244, simple_loss=0.3787, pruned_loss=0.135, over 28707.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3766, pruned_loss=0.1253, over 5669304.58 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3667, pruned_loss=0.12, over 5709462.15 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3783, pruned_loss=0.1261, over 5658473.49 frames. ], batch size: 92, lr: 2.29e-03, grad_scale: 4.0
+2023-03-07 10:06:07,605 WARNING [optim.py:389] (1/2) Scaling gradients by 0.09737467765808105, model_norm_threshold=4090.94189453125
+2023-03-07 10:06:07,685 INFO [optim.py:451] (1/2) 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 10:06:33,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9629, 1.9166, 1.8330, 1.7156], device='cuda:1'), covar=tensor([0.1727, 0.2515, 0.2124, 0.2219], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0739, 0.0691, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 10:06:34,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6473, 1.7227, 1.8796, 1.4411], device='cuda:1'), covar=tensor([0.1850, 0.2384, 0.1439, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0697, 0.0893, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 10:06:43,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-07 10:06:43,439 INFO [train.py:968] (1/2) Epoch 14, batch 28500, giga_loss[loss=0.3789, simple_loss=0.4126, pruned_loss=0.1726, over 26648.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3749, pruned_loss=0.125, over 5655593.05 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3665, pruned_loss=0.1201, over 5695854.09 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1257, over 5657769.90 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:06:43,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7641, 3.5832, 3.3901, 1.7164], device='cuda:1'), covar=tensor([0.0732, 0.0838, 0.0834, 0.2211], device='cuda:1'), in_proj_covar=tensor([0.1133, 0.1051, 0.0914, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 10:07:13,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:07:41,372 INFO [train.py:968] (1/2) Epoch 14, batch 28550, giga_loss[loss=0.3612, simple_loss=0.3903, pruned_loss=0.166, over 23632.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3748, pruned_loss=0.1252, over 5662474.42 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3663, pruned_loss=0.1199, over 5695835.92 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3766, pruned_loss=0.126, over 5663712.58 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:07:49,918 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=621859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:08:20,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2765, 1.5463, 1.2416, 1.5479], device='cuda:1'), covar=tensor([0.0778, 0.0321, 0.0326, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 10:08:30,436 INFO [train.py:968] (1/2) Epoch 14, batch 28600, giga_loss[loss=0.2865, simple_loss=0.3558, pruned_loss=0.1086, over 28678.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.1249, over 5668172.16 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3666, pruned_loss=0.1201, over 5698512.53 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3753, pruned_loss=0.1256, over 5665915.74 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:08:52,458 INFO [optim.py:369] (1/2) 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 10:09:13,124 INFO [train.py:968] (1/2) Epoch 14, batch 28650, giga_loss[loss=0.3084, simple_loss=0.3781, pruned_loss=0.1194, over 28682.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 5662286.94 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3671, pruned_loss=0.1204, over 5687022.82 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.1259, over 5670866.74 frames. ], batch size: 242, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:09:31,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=621967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:09:35,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=621970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:10:01,155 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=621999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:10:01,532 INFO [train.py:968] (1/2) Epoch 14, batch 28700, giga_loss[loss=0.3092, simple_loss=0.3767, pruned_loss=0.1209, over 28954.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1264, over 5650673.38 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.367, pruned_loss=0.1206, over 5691633.25 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3754, pruned_loss=0.1267, over 5652392.17 frames. ], batch size: 155, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:10:26,008 INFO [optim.py:369] (1/2) 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 10:10:45,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9293, 1.8233, 1.0196, 1.0118], device='cuda:1'), covar=tensor([0.0678, 0.0406, 0.0648, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0528, 0.0353, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 10:10:49,690 INFO [train.py:968] (1/2) Epoch 14, batch 28750, giga_loss[loss=0.3276, simple_loss=0.372, pruned_loss=0.1416, over 28878.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3763, pruned_loss=0.1283, over 5642734.12 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3677, pruned_loss=0.1212, over 5686697.39 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5648545.51 frames. ], batch size: 99, lr: 2.29e-03, grad_scale: 1.0
+2023-03-07 10:11:34,635 INFO [train.py:968] (1/2) Epoch 14, batch 28800, giga_loss[loss=0.291, simple_loss=0.3649, pruned_loss=0.1086, over 28891.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3787, pruned_loss=0.1303, over 5650896.15 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3681, pruned_loss=0.1213, over 5692264.96 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3789, pruned_loss=0.1302, over 5649610.58 frames. ], batch size: 164, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:11:37,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-07 10:11:57,951 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 28850, giga_loss[loss=0.3586, simple_loss=0.3908, pruned_loss=0.1632, over 23412.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.378, pruned_loss=0.1298, over 5647692.94 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3674, pruned_loss=0.1207, over 5697624.34 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.379, pruned_loss=0.1306, over 5640783.78 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:12:31,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0577, 1.3423, 1.0755, 0.2240], device='cuda:1'), covar=tensor([0.2570, 0.2300, 0.3430, 0.4869], device='cuda:1'), in_proj_covar=tensor([0.1626, 0.1559, 0.1530, 0.1330], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 10:13:08,574 INFO [train.py:968] (1/2) Epoch 14, batch 28900, giga_loss[loss=0.397, simple_loss=0.4186, pruned_loss=0.1877, over 26549.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.377, pruned_loss=0.1296, over 5644296.70 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3672, pruned_loss=0.1206, over 5691837.81 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3782, pruned_loss=0.1304, over 5642608.48 frames. ], batch size: 555, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:13:29,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 10:13:33,084 INFO [optim.py:369] (1/2) 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 10:13:33,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=622226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:13:38,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=622234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:13:53,589 INFO [train.py:968] (1/2) Epoch 14, batch 28950, giga_loss[loss=0.2583, simple_loss=0.3352, pruned_loss=0.09074, over 28635.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3773, pruned_loss=0.1301, over 5651724.13 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3672, pruned_loss=0.1206, over 5697257.64 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3785, pruned_loss=0.131, over 5644495.86 frames. ], batch size: 71, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:14:41,965 INFO [train.py:968] (1/2) Epoch 14, batch 29000, giga_loss[loss=0.3845, simple_loss=0.4001, pruned_loss=0.1844, over 23560.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.378, pruned_loss=0.1304, over 5642630.48 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3671, pruned_loss=0.1206, over 5700236.50 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3792, pruned_loss=0.1312, over 5633503.44 frames. ], batch size: 705, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:15:08,924 INFO [optim.py:369] (1/2) 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 10:15:09,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7847, 1.9448, 1.7349, 1.7355], device='cuda:1'), covar=tensor([0.1711, 0.2109, 0.2213, 0.2011], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0743, 0.0691, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 10:15:31,172 INFO [train.py:968] (1/2) Epoch 14, batch 29050, giga_loss[loss=0.337, simple_loss=0.4, pruned_loss=0.1369, over 28821.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3779, pruned_loss=0.1292, over 5652321.08 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.367, pruned_loss=0.1205, over 5699785.09 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.379, pruned_loss=0.13, over 5644843.31 frames. ], batch size: 119, lr: 2.29e-03, grad_scale: 2.0
+2023-03-07 10:15:38,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-07 10:15:50,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-07 10:15:57,417 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=622377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:15:59,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=622380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:16:18,749 INFO [train.py:968] (1/2) Epoch 14, batch 29100, giga_loss[loss=0.3378, simple_loss=0.401, pruned_loss=0.1373, over 28839.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3792, pruned_loss=0.1299, over 5657577.71 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.367, pruned_loss=0.1205, over 5702464.81 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3803, pruned_loss=0.1307, over 5648654.28 frames. ], batch size: 145, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:16:27,742 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=622409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:16:41,573 INFO [optim.py:369] (1/2) 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 10:17:06,067 INFO [train.py:968] (1/2) Epoch 14, batch 29150, giga_loss[loss=0.3546, simple_loss=0.4014, pruned_loss=0.1539, over 28680.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3802, pruned_loss=0.1307, over 5673125.49 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3667, pruned_loss=0.1203, over 5704555.92 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3813, pruned_loss=0.1315, over 5664065.84 frames. ], batch size: 284, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:17:52,568 INFO [train.py:968] (1/2) Epoch 14, batch 29200, libri_loss[loss=0.2335, simple_loss=0.3089, pruned_loss=0.07906, over 29383.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3811, pruned_loss=0.1315, over 5676738.97 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3667, pruned_loss=0.1202, over 5707552.10 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3822, pruned_loss=0.1325, over 5666276.10 frames. ], batch size: 67, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:18:18,542 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 29250, giga_loss[loss=0.326, simple_loss=0.3991, pruned_loss=0.1264, over 29013.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3835, pruned_loss=0.1325, over 5669690.64 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3672, pruned_loss=0.1205, over 5707599.34 frames. ], giga_tot_loss[loss=0.3251, simple_loss=0.3841, pruned_loss=0.1331, over 5661329.23 frames. ], batch size: 155, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:19:10,124 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=622573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:19:36,706 INFO [train.py:968] (1/2) Epoch 14, batch 29300, giga_loss[loss=0.3005, simple_loss=0.3677, pruned_loss=0.1167, over 29007.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3824, pruned_loss=0.1307, over 5653952.90 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3673, pruned_loss=0.1206, over 5699483.76 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3833, pruned_loss=0.1314, over 5653819.73 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:19:37,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=622601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:20:00,649 INFO [optim.py:369] (1/2) 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 10:20:22,770 INFO [train.py:968] (1/2) Epoch 14, batch 29350, giga_loss[loss=0.2833, simple_loss=0.3551, pruned_loss=0.1058, over 28692.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3797, pruned_loss=0.1286, over 5653733.01 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3675, pruned_loss=0.1208, over 5691996.84 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3803, pruned_loss=0.1291, over 5660206.93 frames. ], batch size: 284, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:21:09,369 INFO [train.py:968] (1/2) Epoch 14, batch 29400, giga_loss[loss=0.3284, simple_loss=0.395, pruned_loss=0.131, over 29042.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3788, pruned_loss=0.1284, over 5652624.91 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3671, pruned_loss=0.1205, over 5694991.66 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3799, pruned_loss=0.1292, over 5654605.62 frames. ], batch size: 155, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:21:35,544 INFO [optim.py:369] (1/2) 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 10:21:41,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 10:21:53,342 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=622744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:21:55,917 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=622747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:21:57,588 INFO [train.py:968] (1/2) Epoch 14, batch 29450, libri_loss[loss=0.2414, simple_loss=0.3075, pruned_loss=0.08762, over 29484.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3797, pruned_loss=0.1287, over 5651428.16 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3666, pruned_loss=0.1201, over 5689083.25 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3813, pruned_loss=0.1298, over 5657921.39 frames. ], batch size: 70, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:22:26,223 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=622776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:22:46,831 INFO [train.py:968] (1/2) Epoch 14, batch 29500, giga_loss[loss=0.2653, simple_loss=0.3356, pruned_loss=0.09753, over 28765.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3809, pruned_loss=0.1302, over 5644238.85 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3665, pruned_loss=0.12, over 5682431.65 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3826, pruned_loss=0.1313, over 5654221.49 frames. ], batch size: 199, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:23:07,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-07 10:23:15,046 INFO [optim.py:369] (1/2) 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 10:23:39,764 INFO [train.py:968] (1/2) Epoch 14, batch 29550, giga_loss[loss=0.3301, simple_loss=0.3874, pruned_loss=0.1364, over 28722.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3808, pruned_loss=0.131, over 5646460.73 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3665, pruned_loss=0.12, over 5682431.65 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3821, pruned_loss=0.132, over 5654230.34 frames. ], batch size: 262, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:24:22,430 INFO [train.py:968] (1/2) Epoch 14, batch 29600, giga_loss[loss=0.3648, simple_loss=0.4108, pruned_loss=0.1594, over 28658.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3826, pruned_loss=0.1325, over 5639920.15 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3675, pruned_loss=0.1204, over 5677778.43 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3833, pruned_loss=0.1334, over 5648782.23 frames. ], batch size: 92, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:24:48,392 INFO [optim.py:369] (1/2) 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 10:24:49,273 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=622927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:25:10,428 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 29650, giga_loss[loss=0.3533, simple_loss=0.4015, pruned_loss=0.1526, over 27649.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3826, pruned_loss=0.1324, over 5639627.25 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3677, pruned_loss=0.1206, over 5671283.94 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3831, pruned_loss=0.1331, over 5652853.20 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:26:00,996 INFO [train.py:968] (1/2) Epoch 14, batch 29700, giga_loss[loss=0.3296, simple_loss=0.3835, pruned_loss=0.1378, over 29002.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3817, pruned_loss=0.1316, over 5637097.11 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1209, over 5671480.90 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3821, pruned_loss=0.1322, over 5646967.15 frames. ], batch size: 106, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:26:25,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4483, 1.8030, 1.3998, 1.6001], device='cuda:1'), covar=tensor([0.2423, 0.2313, 0.2583, 0.2074], device='cuda:1'), in_proj_covar=tensor([0.1365, 0.1004, 0.1209, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 10:26:26,991 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 29750, giga_loss[loss=0.3122, simple_loss=0.3755, pruned_loss=0.1245, over 29241.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3803, pruned_loss=0.1295, over 5653830.64 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3682, pruned_loss=0.1209, over 5667568.53 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3808, pruned_loss=0.1301, over 5665210.00 frames. ], batch size: 113, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:26:53,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2825, 1.6148, 1.2088, 1.3014], device='cuda:1'), covar=tensor([0.2334, 0.2258, 0.2555, 0.1978], device='cuda:1'), in_proj_covar=tensor([0.1367, 0.1006, 0.1211, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 10:27:27,055 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:27:29,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=623094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:27:36,562 INFO [train.py:968] (1/2) Epoch 14, batch 29800, giga_loss[loss=0.3071, simple_loss=0.3803, pruned_loss=0.117, over 29006.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3795, pruned_loss=0.1286, over 5654958.44 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3678, pruned_loss=0.1206, over 5673586.86 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3805, pruned_loss=0.1295, over 5658054.14 frames. ], batch size: 128, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:28:00,361 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=623123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:28:04,169 INFO [optim.py:369] (1/2) 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 10:28:25,904 INFO [train.py:968] (1/2) Epoch 14, batch 29850, giga_loss[loss=0.3441, simple_loss=0.3795, pruned_loss=0.1543, over 23894.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3788, pruned_loss=0.1278, over 5656002.36 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3683, pruned_loss=0.1209, over 5671615.26 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3794, pruned_loss=0.1284, over 5659722.67 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:28:26,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2713, 4.1166, 3.8837, 1.8256], device='cuda:1'), covar=tensor([0.0610, 0.0723, 0.0777, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.1131, 0.1052, 0.0911, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 10:28:57,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=623181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:29:15,101 INFO [train.py:968] (1/2) Epoch 14, batch 29900, giga_loss[loss=0.3553, simple_loss=0.4102, pruned_loss=0.1502, over 27836.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3772, pruned_loss=0.1273, over 5654913.15 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3683, pruned_loss=0.1209, over 5672732.55 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3777, pruned_loss=0.1279, over 5656519.52 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:29:20,035 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5861, 1.6879, 1.2020, 1.2845], device='cuda:1'), covar=tensor([0.0838, 0.0594, 0.0958, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0438, 0.0504, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 10:29:22,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0537, 1.0644, 3.7920, 3.0785], device='cuda:1'), covar=tensor([0.1747, 0.2783, 0.0439, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0618, 0.0900, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 10:29:39,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-07 10:29:40,193 INFO [optim.py:369] (1/2) 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 10:29:59,154 INFO [train.py:968] (1/2) Epoch 14, batch 29950, giga_loss[loss=0.3316, simple_loss=0.3688, pruned_loss=0.1472, over 23592.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3758, pruned_loss=0.1268, over 5658587.03 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3682, pruned_loss=0.1206, over 5678407.52 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3766, pruned_loss=0.1277, over 5654169.27 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:29:59,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5322, 1.7229, 1.5218, 1.3842], device='cuda:1'), covar=tensor([0.2190, 0.1858, 0.1713, 0.1919], device='cuda:1'), in_proj_covar=tensor([0.1806, 0.1733, 0.1658, 0.1793], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 10:30:48,305 INFO [train.py:968] (1/2) Epoch 14, batch 30000, giga_loss[loss=0.3077, simple_loss=0.363, pruned_loss=0.1262, over 28605.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3725, pruned_loss=0.125, over 5668021.19 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.368, pruned_loss=0.1203, over 5684075.96 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3734, pruned_loss=0.1261, over 5658864.68 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 10:30:48,305 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 10:30:56,783 INFO [train.py:1012] (1/2) Epoch 14, validation: loss=0.2118, simple_loss=0.3195, pruned_loss=0.05203, over 944034.00 frames. 
+2023-03-07 10:30:56,783 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 10:30:59,372 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:31:40,292 INFO [train.py:968] (1/2) Epoch 14, batch 30050, giga_loss[loss=0.3081, simple_loss=0.3656, pruned_loss=0.1253, over 29001.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3699, pruned_loss=0.1239, over 5665346.89 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3682, pruned_loss=0.1205, over 5668394.76 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3705, pruned_loss=0.1247, over 5672042.29 frames. ], batch size: 128, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:31:52,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7020, 0.9246, 0.9572, 0.7924], device='cuda:1'), covar=tensor([0.0981, 0.0798, 0.1283, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0739, 0.0691, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 10:32:26,362 INFO [train.py:968] (1/2) Epoch 14, batch 30100, giga_loss[loss=0.2562, simple_loss=0.3335, pruned_loss=0.08939, over 28848.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3695, pruned_loss=0.1243, over 5671949.68 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1209, over 5662516.13 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3695, pruned_loss=0.1246, over 5682459.71 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:32:56,575 INFO [optim.py:369] (1/2) 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 10:32:59,455 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-07 10:33:15,988 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:33:17,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=623448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:33:18,774 INFO [train.py:968] (1/2) Epoch 14, batch 30150, giga_loss[loss=0.2781, simple_loss=0.3431, pruned_loss=0.1066, over 28749.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.369, pruned_loss=0.1234, over 5674954.69 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.369, pruned_loss=0.121, over 5666142.45 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3689, pruned_loss=0.1237, over 5680037.85 frames. ], batch size: 99, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:33:43,006 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=623477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:34:06,288 INFO [train.py:968] (1/2) Epoch 14, batch 30200, giga_loss[loss=0.2277, simple_loss=0.2967, pruned_loss=0.07941, over 24082.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3676, pruned_loss=0.1207, over 5676162.69 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.121, over 5673196.12 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5674033.00 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:34:32,647 INFO [optim.py:369] (1/2) 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 10:34:53,135 INFO [train.py:968] (1/2) Epoch 14, batch 30250, libri_loss[loss=0.2521, simple_loss=0.3087, pruned_loss=0.09779, over 29347.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3657, pruned_loss=0.1182, over 5675725.88 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.368, pruned_loss=0.1208, over 5681301.01 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3664, pruned_loss=0.1185, over 5666402.79 frames. ], batch size: 67, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:34:59,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=623556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:35:34,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6220, 1.7177, 1.6683, 1.4017], device='cuda:1'), covar=tensor([0.1171, 0.1433, 0.1755, 0.1535], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0735, 0.0689, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 10:35:39,320 INFO [train.py:968] (1/2) Epoch 14, batch 30300, giga_loss[loss=0.2721, simple_loss=0.3453, pruned_loss=0.09947, over 28508.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3627, pruned_loss=0.1153, over 5670819.82 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3685, pruned_loss=0.1215, over 5684158.06 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3627, pruned_loss=0.1148, over 5660560.06 frames. ], batch size: 71, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:36:11,424 INFO [optim.py:369] (1/2) 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 10:36:22,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 10:36:30,548 INFO [train.py:968] (1/2) Epoch 14, batch 30350, giga_loss[loss=0.2575, simple_loss=0.3415, pruned_loss=0.08672, over 28904.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1115, over 5664659.32 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3681, pruned_loss=0.1212, over 5683188.29 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3592, pruned_loss=0.1112, over 5657343.84 frames. ], batch size: 164, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:37:17,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:37:17,861 INFO [train.py:968] (1/2) Epoch 14, batch 30400, giga_loss[loss=0.2622, simple_loss=0.3476, pruned_loss=0.08847, over 29077.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3562, pruned_loss=0.1081, over 5665378.97 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3678, pruned_loss=0.1213, over 5685037.25 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3565, pruned_loss=0.1075, over 5657229.01 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:37:20,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=623702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:37:27,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2316, 1.5501, 1.2728, 1.1301], device='cuda:1'), covar=tensor([0.2199, 0.2034, 0.2099, 0.1859], device='cuda:1'), in_proj_covar=tensor([0.1375, 0.1006, 0.1219, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 10:37:48,854 INFO [optim.py:369] (1/2) 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 10:37:50,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=623731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:38:04,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=623745.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 10:38:10,728 INFO [train.py:968] (1/2) Epoch 14, batch 30450, giga_loss[loss=0.2438, simple_loss=0.3302, pruned_loss=0.07874, over 28332.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3543, pruned_loss=0.1054, over 5649121.82 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3677, pruned_loss=0.1213, over 5689319.44 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3542, pruned_loss=0.1046, over 5638258.17 frames. ], batch size: 368, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:39:01,384 INFO [train.py:968] (1/2) Epoch 14, batch 30500, giga_loss[loss=0.2825, simple_loss=0.3584, pruned_loss=0.1033, over 28864.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3557, pruned_loss=0.1069, over 5653161.77 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.367, pruned_loss=0.121, over 5697052.01 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3559, pruned_loss=0.106, over 5636407.49 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:39:30,858 INFO [optim.py:369] (1/2) 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 10:39:52,766 INFO [train.py:968] (1/2) Epoch 14, batch 30550, giga_loss[loss=0.2519, simple_loss=0.3143, pruned_loss=0.09473, over 24120.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3531, pruned_loss=0.1049, over 5645101.28 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3668, pruned_loss=0.121, over 5697265.72 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3534, pruned_loss=0.1041, over 5631307.03 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:39:57,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=623852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:40:32,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4254, 1.9549, 1.4013, 0.7096], device='cuda:1'), covar=tensor([0.5150, 0.2673, 0.3713, 0.4992], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1549, 0.1525, 0.1323], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 10:40:42,509 INFO [train.py:968] (1/2) Epoch 14, batch 30600, giga_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.0889, over 28860.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3498, pruned_loss=0.1025, over 5648998.28 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3661, pruned_loss=0.1207, over 5697921.61 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3503, pruned_loss=0.1018, over 5636157.22 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:41:04,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1975, 1.2157, 3.7933, 3.1880], device='cuda:1'), covar=tensor([0.1684, 0.2648, 0.0415, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0609, 0.0890, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 10:41:11,345 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 30650, giga_loss[loss=0.3014, simple_loss=0.3606, pruned_loss=0.1211, over 26809.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3493, pruned_loss=0.1025, over 5646043.14 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3663, pruned_loss=0.1211, over 5696420.51 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3489, pruned_loss=0.1008, over 5634674.28 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:41:38,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3026, 1.6400, 1.3030, 1.4940], device='cuda:1'), covar=tensor([0.0728, 0.0363, 0.0350, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 10:42:17,207 INFO [train.py:968] (1/2) Epoch 14, batch 30700, giga_loss[loss=0.29, simple_loss=0.3652, pruned_loss=0.1074, over 27970.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3496, pruned_loss=0.1021, over 5648905.36 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3664, pruned_loss=0.1213, over 5698458.55 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3489, pruned_loss=0.1002, over 5637157.92 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:42:36,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-07 10:42:45,661 INFO [optim.py:369] (1/2) 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 10:43:04,760 INFO [train.py:968] (1/2) Epoch 14, batch 30750, giga_loss[loss=0.242, simple_loss=0.333, pruned_loss=0.07548, over 28571.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3467, pruned_loss=0.09936, over 5662725.92 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3653, pruned_loss=0.1207, over 5704915.77 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3466, pruned_loss=0.09786, over 5646463.25 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:43:09,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4281, 3.6840, 1.6787, 1.5606], device='cuda:1'), covar=tensor([0.0977, 0.0269, 0.0896, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0523, 0.0351, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 10:43:56,231 INFO [train.py:968] (1/2) Epoch 14, batch 30800, giga_loss[loss=0.2489, simple_loss=0.3302, pruned_loss=0.08384, over 28567.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3438, pruned_loss=0.09718, over 5638263.44 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3651, pruned_loss=0.1208, over 5688135.96 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3435, pruned_loss=0.09552, over 5639254.96 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:44:17,935 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=624120.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 10:44:25,687 INFO [optim.py:369] (1/2) 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 10:44:41,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4903, 1.7107, 1.7553, 1.3153], device='cuda:1'), covar=tensor([0.1794, 0.2491, 0.1487, 0.1804], device='cuda:1'), in_proj_covar=tensor([0.0852, 0.0686, 0.0892, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 10:44:45,696 INFO [train.py:968] (1/2) Epoch 14, batch 30850, giga_loss[loss=0.2347, simple_loss=0.3154, pruned_loss=0.07698, over 28799.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3396, pruned_loss=0.09516, over 5637523.48 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3643, pruned_loss=0.1205, over 5690503.83 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3395, pruned_loss=0.09352, over 5634825.05 frames. ], batch size: 262, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:45:00,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5888, 1.8350, 1.8841, 1.3830], device='cuda:1'), covar=tensor([0.1905, 0.2490, 0.1533, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.0852, 0.0686, 0.0892, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 10:45:01,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.88 vs. limit=5.0
+2023-03-07 10:45:31,411 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-07 10:45:32,855 INFO [train.py:968] (1/2) Epoch 14, batch 30900, giga_loss[loss=0.2635, simple_loss=0.345, pruned_loss=0.09098, over 28241.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3394, pruned_loss=0.09562, over 5645526.56 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3638, pruned_loss=0.1201, over 5695356.63 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3392, pruned_loss=0.09393, over 5637903.03 frames. ], batch size: 368, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:46:03,311 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:46:29,183 INFO [train.py:968] (1/2) Epoch 14, batch 30950, giga_loss[loss=0.2428, simple_loss=0.3, pruned_loss=0.09277, over 24138.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3379, pruned_loss=0.09506, over 5628636.58 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3638, pruned_loss=0.1201, over 5695356.63 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3377, pruned_loss=0.09374, over 5622703.09 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:46:41,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=624263.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 10:46:45,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=624266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 10:47:15,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=624295.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 10:47:21,371 INFO [train.py:968] (1/2) Epoch 14, batch 31000, giga_loss[loss=0.2683, simple_loss=0.3376, pruned_loss=0.09951, over 28655.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.34, pruned_loss=0.09588, over 5624318.11 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3631, pruned_loss=0.1199, over 5688422.27 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3397, pruned_loss=0.09422, over 5622962.61 frames. ], batch size: 92, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:47:58,249 INFO [optim.py:369] (1/2) 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 10:48:18,982 INFO [train.py:968] (1/2) Epoch 14, batch 31050, giga_loss[loss=0.314, simple_loss=0.3775, pruned_loss=0.1253, over 28904.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3425, pruned_loss=0.09602, over 5638422.73 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.362, pruned_loss=0.1193, over 5689260.98 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3427, pruned_loss=0.09461, over 5634965.39 frames. ], batch size: 199, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:48:33,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4749, 1.8271, 1.6505, 1.5159], device='cuda:1'), covar=tensor([0.1527, 0.1804, 0.1748, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0719, 0.0675, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 10:48:43,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=624370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:48:45,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=624373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:49:21,875 INFO [train.py:968] (1/2) Epoch 14, batch 31100, libri_loss[loss=0.3264, simple_loss=0.3772, pruned_loss=0.1377, over 29517.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3439, pruned_loss=0.09675, over 5658094.95 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.362, pruned_loss=0.1194, over 5693451.77 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3436, pruned_loss=0.09499, over 5650592.04 frames. ], batch size: 89, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:49:24,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=624402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:49:56,826 INFO [optim.py:369] (1/2) 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 10:50:21,796 INFO [train.py:968] (1/2) Epoch 14, batch 31150, giga_loss[loss=0.2229, simple_loss=0.2915, pruned_loss=0.07716, over 24443.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3428, pruned_loss=0.09663, over 5662628.12 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3615, pruned_loss=0.1192, over 5697735.98 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3424, pruned_loss=0.09469, over 5651844.33 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 10:51:21,040 INFO [train.py:968] (1/2) Epoch 14, batch 31200, libri_loss[loss=0.3396, simple_loss=0.3865, pruned_loss=0.1463, over 19526.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3421, pruned_loss=0.09574, over 5637361.86 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3615, pruned_loss=0.1193, over 5672623.38 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3413, pruned_loss=0.09349, over 5652017.92 frames. ], batch size: 188, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:51:56,221 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 31250, giga_loss[loss=0.2263, simple_loss=0.313, pruned_loss=0.06983, over 29057.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.34, pruned_loss=0.09387, over 5651387.18 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3606, pruned_loss=0.1188, over 5678948.06 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3392, pruned_loss=0.09123, over 5655911.96 frames. ], batch size: 175, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:52:38,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4274, 1.4269, 1.2809, 1.7322], device='cuda:1'), covar=tensor([0.0751, 0.0317, 0.0357, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 10:53:04,175 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.10 vs. limit=5.0
+2023-03-07 10:53:13,674 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=624596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:53:16,650 INFO [train.py:968] (1/2) Epoch 14, batch 31300, giga_loss[loss=0.263, simple_loss=0.3319, pruned_loss=0.09705, over 29028.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3371, pruned_loss=0.09307, over 5660711.67 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3603, pruned_loss=0.1187, over 5682070.08 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3363, pruned_loss=0.09074, over 5661118.92 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:53:23,901 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 10:53:56,300 INFO [optim.py:369] (1/2) 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 10:54:16,667 INFO [train.py:968] (1/2) Epoch 14, batch 31350, giga_loss[loss=0.2349, simple_loss=0.3202, pruned_loss=0.07481, over 28654.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3367, pruned_loss=0.09294, over 5665805.92 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3601, pruned_loss=0.1186, over 5688244.07 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3357, pruned_loss=0.09046, over 5659969.86 frames. ], batch size: 262, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:55:02,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2073, 2.5925, 1.1787, 1.3829], device='cuda:1'), covar=tensor([0.0989, 0.0383, 0.0979, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0523, 0.0353, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 10:55:09,831 INFO [train.py:968] (1/2) Epoch 14, batch 31400, giga_loss[loss=0.2736, simple_loss=0.3562, pruned_loss=0.09545, over 28952.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3365, pruned_loss=0.09333, over 5647193.25 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3599, pruned_loss=0.1187, over 5665012.88 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3349, pruned_loss=0.09019, over 5662833.71 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:55:46,150 INFO [optim.py:369] (1/2) 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 10:56:08,618 INFO [train.py:968] (1/2) Epoch 14, batch 31450, giga_loss[loss=0.2633, simple_loss=0.3513, pruned_loss=0.08766, over 28902.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3386, pruned_loss=0.09372, over 5650171.23 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.36, pruned_loss=0.1189, over 5667800.06 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3368, pruned_loss=0.09063, over 5659696.99 frames. ], batch size: 164, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:57:16,096 INFO [train.py:968] (1/2) Epoch 14, batch 31500, giga_loss[loss=0.2538, simple_loss=0.335, pruned_loss=0.08636, over 28832.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3392, pruned_loss=0.09327, over 5653913.73 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3598, pruned_loss=0.1188, over 5670400.14 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3376, pruned_loss=0.09053, over 5658760.10 frames. ], batch size: 243, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:57:25,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6885, 2.3290, 1.5047, 0.9639], device='cuda:1'), covar=tensor([0.6811, 0.3363, 0.3924, 0.5842], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1544, 0.1523, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 10:57:49,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6624, 2.0033, 1.8946, 1.5275], device='cuda:1'), covar=tensor([0.2364, 0.1683, 0.1628, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.1773, 0.1678, 0.1605, 0.1739], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 10:57:49,543 INFO [optim.py:369] (1/2) 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 10:57:51,638 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=624831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 10:58:19,232 INFO [train.py:968] (1/2) Epoch 14, batch 31550, giga_loss[loss=0.2597, simple_loss=0.3351, pruned_loss=0.09217, over 28893.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3346, pruned_loss=0.09058, over 5663141.47 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3588, pruned_loss=0.1182, over 5677137.04 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3337, pruned_loss=0.08818, over 5660880.37 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 10:59:27,226 INFO [train.py:968] (1/2) Epoch 14, batch 31600, giga_loss[loss=0.2568, simple_loss=0.3415, pruned_loss=0.08606, over 28462.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3374, pruned_loss=0.09271, over 5662506.13 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3587, pruned_loss=0.1182, over 5671757.77 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3364, pruned_loss=0.09038, over 5664974.06 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:00:07,849 INFO [optim.py:369] (1/2) 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 11:00:13,157 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-07 11:00:27,983 INFO [train.py:968] (1/2) Epoch 14, batch 31650, giga_loss[loss=0.1977, simple_loss=0.2788, pruned_loss=0.05835, over 24628.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.341, pruned_loss=0.09288, over 5648225.64 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3588, pruned_loss=0.1183, over 5668700.78 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3396, pruned_loss=0.09013, over 5652587.29 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:00:52,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=624971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:01:28,998 INFO [train.py:968] (1/2) Epoch 14, batch 31700, giga_loss[loss=0.2696, simple_loss=0.3509, pruned_loss=0.09413, over 27819.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3435, pruned_loss=0.0924, over 5652326.66 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3585, pruned_loss=0.1183, over 5672780.42 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3422, pruned_loss=0.08971, over 5651847.86 frames. ], batch size: 474, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:02:02,322 INFO [optim.py:369] (1/2) 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 11:02:24,704 INFO [train.py:968] (1/2) Epoch 14, batch 31750, giga_loss[loss=0.2538, simple_loss=0.3414, pruned_loss=0.08314, over 28882.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3441, pruned_loss=0.09194, over 5655083.23 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3583, pruned_loss=0.1181, over 5675838.96 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3427, pruned_loss=0.08895, over 5651577.57 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:02:38,437 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=625062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:02:42,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6243, 1.8605, 1.9032, 1.4371], device='cuda:1'), covar=tensor([0.1958, 0.2479, 0.1525, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0851, 0.0682, 0.0893, 0.0797], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 11:03:21,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2880, 3.5780, 1.5370, 1.5167], device='cuda:1'), covar=tensor([0.1011, 0.0314, 0.0890, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0521, 0.0353, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 11:03:26,660 INFO [train.py:968] (1/2) Epoch 14, batch 31800, giga_loss[loss=0.2689, simple_loss=0.3425, pruned_loss=0.09767, over 27608.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3424, pruned_loss=0.09026, over 5656433.68 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3583, pruned_loss=0.1181, over 5678250.76 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3412, pruned_loss=0.08772, over 5651565.58 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:03:44,379 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=625114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:03:49,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=625117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:04:07,228 INFO [optim.py:369] (1/2) 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 11:04:26,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=625146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:04:30,009 INFO [train.py:968] (1/2) Epoch 14, batch 31850, giga_loss[loss=0.2436, simple_loss=0.3267, pruned_loss=0.08022, over 28880.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3435, pruned_loss=0.09239, over 5655480.47 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3583, pruned_loss=0.1181, over 5680246.80 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3422, pruned_loss=0.0897, over 5649626.39 frames. ], batch size: 174, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:05:37,774 INFO [train.py:968] (1/2) Epoch 14, batch 31900, giga_loss[loss=0.2591, simple_loss=0.3276, pruned_loss=0.09532, over 24753.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3433, pruned_loss=0.09337, over 5665957.63 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3579, pruned_loss=0.1178, over 5687913.37 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3419, pruned_loss=0.0904, over 5653443.46 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:05:51,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=625206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:06:23,510 INFO [optim.py:369] (1/2) 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 11:06:57,103 INFO [train.py:968] (1/2) Epoch 14, batch 31950, giga_loss[loss=0.2689, simple_loss=0.3481, pruned_loss=0.09481, over 29003.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3431, pruned_loss=0.09347, over 5662834.62 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3578, pruned_loss=0.1179, over 5673075.55 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3418, pruned_loss=0.09057, over 5665346.20 frames. ], batch size: 128, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:08:06,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2392, 1.3785, 1.3467, 1.2426], device='cuda:1'), covar=tensor([0.1855, 0.1409, 0.1220, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.1773, 0.1676, 0.1604, 0.1742], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 11:08:09,134 INFO [train.py:968] (1/2) Epoch 14, batch 32000, giga_loss[loss=0.2145, simple_loss=0.3064, pruned_loss=0.0613, over 28091.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3385, pruned_loss=0.09072, over 5667241.92 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3576, pruned_loss=0.1178, over 5676628.64 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3374, pruned_loss=0.08821, over 5666010.72 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:08:52,698 INFO [optim.py:369] (1/2) 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 11:09:16,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=625349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:09:16,465 INFO [train.py:968] (1/2) Epoch 14, batch 32050, giga_loss[loss=0.2244, simple_loss=0.2848, pruned_loss=0.08204, over 24108.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3366, pruned_loss=0.08997, over 5665878.07 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3577, pruned_loss=0.1177, over 5680647.04 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3353, pruned_loss=0.08748, over 5661007.05 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:09:19,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=625352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:09:33,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3357, 1.1605, 4.1237, 3.3386], device='cuda:1'), covar=tensor([0.1611, 0.2797, 0.0397, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0605, 0.0880, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 11:09:41,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4957, 1.6813, 1.5286, 1.4749], device='cuda:1'), covar=tensor([0.2415, 0.1780, 0.1627, 0.1793], device='cuda:1'), in_proj_covar=tensor([0.1775, 0.1673, 0.1604, 0.1743], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 11:09:58,265 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=625381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:09:59,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7781, 1.8959, 1.2129, 1.5948], device='cuda:1'), covar=tensor([0.0915, 0.0716, 0.1195, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0439, 0.0503, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 11:10:20,853 INFO [train.py:968] (1/2) Epoch 14, batch 32100, giga_loss[loss=0.2812, simple_loss=0.3581, pruned_loss=0.1022, over 28078.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3372, pruned_loss=0.09103, over 5666378.42 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3576, pruned_loss=0.1177, over 5683364.17 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3357, pruned_loss=0.08837, over 5659576.54 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:10:58,749 INFO [optim.py:369] (1/2) 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 11:11:03,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=625437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:11:17,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5480, 1.7340, 1.4584, 1.6966], device='cuda:1'), covar=tensor([0.2466, 0.2277, 0.2495, 0.2116], device='cuda:1'), in_proj_covar=tensor([0.1369, 0.1003, 0.1214, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 11:11:17,667 INFO [train.py:968] (1/2) Epoch 14, batch 32150, giga_loss[loss=0.2536, simple_loss=0.3367, pruned_loss=0.08532, over 27623.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3404, pruned_loss=0.09264, over 5671192.31 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3569, pruned_loss=0.1176, over 5681148.66 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3393, pruned_loss=0.08985, over 5666456.80 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:12:15,763 INFO [train.py:968] (1/2) Epoch 14, batch 32200, giga_loss[loss=0.2488, simple_loss=0.3269, pruned_loss=0.08532, over 28903.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3394, pruned_loss=0.09279, over 5665020.13 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3563, pruned_loss=0.1173, over 5676230.16 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3387, pruned_loss=0.09016, over 5665547.23 frames. ], batch size: 145, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:12:37,547 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=625515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:12:59,100 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 32250, giga_loss[loss=0.2473, simple_loss=0.3285, pruned_loss=0.08305, over 28976.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3391, pruned_loss=0.09331, over 5668056.53 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3563, pruned_loss=0.1172, over 5679572.95 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3383, pruned_loss=0.09097, over 5665586.47 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:13:36,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-07 11:13:57,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=625580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:14:00,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=625583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:14:20,684 INFO [train.py:968] (1/2) Epoch 14, batch 32300, giga_loss[loss=0.2618, simple_loss=0.3445, pruned_loss=0.08959, over 28672.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3402, pruned_loss=0.09425, over 5668087.55 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3566, pruned_loss=0.1176, over 5681931.66 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.339, pruned_loss=0.09151, over 5663584.05 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:14:37,219 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:15:35,078 INFO [train.py:968] (1/2) Epoch 14, batch 32350, giga_loss[loss=0.3036, simple_loss=0.3577, pruned_loss=0.1247, over 26854.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3425, pruned_loss=0.09486, over 5663378.78 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3563, pruned_loss=0.1175, over 5684104.99 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3416, pruned_loss=0.0926, over 5657789.58 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:16:32,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0006, 1.3760, 1.0789, 0.1663], device='cuda:1'), covar=tensor([0.2584, 0.2144, 0.3464, 0.4483], device='cuda:1'), in_proj_covar=tensor([0.1631, 0.1558, 0.1535, 0.1338], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 11:16:47,094 INFO [train.py:968] (1/2) Epoch 14, batch 32400, giga_loss[loss=0.2377, simple_loss=0.3061, pruned_loss=0.08469, over 24307.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3425, pruned_loss=0.09421, over 5669097.78 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.356, pruned_loss=0.1175, over 5687757.79 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3417, pruned_loss=0.09172, over 5660719.96 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:17:10,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5717, 4.3222, 1.7134, 1.7394], device='cuda:1'), covar=tensor([0.0923, 0.0335, 0.0897, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0520, 0.0354, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 11:17:20,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2844, 1.2331, 3.8445, 3.1225], device='cuda:1'), covar=tensor([0.1579, 0.2707, 0.0398, 0.0954], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0610, 0.0883, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 11:17:34,597 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 32450, giga_loss[loss=0.2385, simple_loss=0.3104, pruned_loss=0.08332, over 28903.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3396, pruned_loss=0.09285, over 5667800.44 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3553, pruned_loss=0.117, over 5683246.39 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3392, pruned_loss=0.09066, over 5664901.84 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:18:49,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 11:18:55,560 INFO [train.py:968] (1/2) Epoch 14, batch 32500, libri_loss[loss=0.2213, simple_loss=0.2939, pruned_loss=0.07433, over 28551.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09136, over 5665018.03 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3544, pruned_loss=0.1164, over 5677943.22 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3348, pruned_loss=0.08928, over 5667700.82 frames. ], batch size: 63, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:19:39,789 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 32550, libri_loss[loss=0.26, simple_loss=0.3256, pruned_loss=0.0972, over 29546.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3305, pruned_loss=0.08991, over 5663586.00 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3539, pruned_loss=0.1161, over 5684539.06 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3301, pruned_loss=0.08757, over 5659051.34 frames. ], batch size: 78, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:20:21,062 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-07 11:20:47,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=625890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:20:55,985 INFO [train.py:968] (1/2) Epoch 14, batch 32600, giga_loss[loss=0.3625, simple_loss=0.3984, pruned_loss=0.1633, over 27537.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.333, pruned_loss=0.09203, over 5648774.51 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.354, pruned_loss=0.1164, over 5673303.27 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3319, pruned_loss=0.08914, over 5654461.53 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:21:36,573 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 32650, giga_loss[loss=0.2433, simple_loss=0.3267, pruned_loss=0.07994, over 29126.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09271, over 5649033.48 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3543, pruned_loss=0.1165, over 5674973.18 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.333, pruned_loss=0.09002, over 5651939.17 frames. ], batch size: 200, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:22:56,368 INFO [train.py:968] (1/2) Epoch 14, batch 32700, giga_loss[loss=0.2538, simple_loss=0.3286, pruned_loss=0.08947, over 26761.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.332, pruned_loss=0.09054, over 5643691.85 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3545, pruned_loss=0.1168, over 5669383.16 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3304, pruned_loss=0.08775, over 5649726.13 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:23:25,941 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:23:38,915 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:23:39,218 INFO [optim.py:369] (1/2) 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 11:23:43,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:23:59,872 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 11:24:00,190 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 32750, libri_loss[loss=0.2881, simple_loss=0.3305, pruned_loss=0.1229, over 29630.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3299, pruned_loss=0.08869, over 5640365.23 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3546, pruned_loss=0.117, over 5653824.34 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3282, pruned_loss=0.08585, over 5657564.82 frames. ], batch size: 69, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:24:20,752 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:25:03,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 11:25:06,632 INFO [train.py:968] (1/2) Epoch 14, batch 32800, giga_loss[loss=0.3327, simple_loss=0.3999, pruned_loss=0.1327, over 28082.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3298, pruned_loss=0.08936, over 5645746.71 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3544, pruned_loss=0.1171, over 5661187.41 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.328, pruned_loss=0.08622, over 5652674.30 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:25:53,608 INFO [optim.py:369] (1/2) 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 11:25:55,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5438, 1.8247, 1.4693, 1.7597], device='cuda:1'), covar=tensor([0.2629, 0.2587, 0.2997, 0.2288], device='cuda:1'), in_proj_covar=tensor([0.1370, 0.1001, 0.1216, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 11:26:16,829 INFO [train.py:968] (1/2) Epoch 14, batch 32850, giga_loss[loss=0.2784, simple_loss=0.3425, pruned_loss=0.1071, over 26804.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3293, pruned_loss=0.0882, over 5648270.34 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3542, pruned_loss=0.117, over 5665525.18 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3276, pruned_loss=0.08533, over 5649938.55 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:26:21,598 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626154.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 11:27:03,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5699, 1.7458, 1.8736, 1.4024], device='cuda:1'), covar=tensor([0.1835, 0.2422, 0.1429, 0.1752], device='cuda:1'), in_proj_covar=tensor([0.0848, 0.0679, 0.0890, 0.0796], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 11:27:03,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-07 11:27:18,992 INFO [train.py:968] (1/2) Epoch 14, batch 32900, giga_loss[loss=0.2651, simple_loss=0.3433, pruned_loss=0.0934, over 28879.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3297, pruned_loss=0.08863, over 5658662.33 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3544, pruned_loss=0.1172, over 5670911.08 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3277, pruned_loss=0.08569, over 5655151.72 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:28:00,960 INFO [optim.py:369] (1/2) 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 11:28:21,105 INFO [train.py:968] (1/2) Epoch 14, batch 32950, giga_loss[loss=0.2221, simple_loss=0.307, pruned_loss=0.06863, over 28641.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3305, pruned_loss=0.08952, over 5662212.72 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.354, pruned_loss=0.1168, over 5671952.78 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3288, pruned_loss=0.08696, over 5658340.19 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:29:16,167 INFO [train.py:968] (1/2) Epoch 14, batch 33000, giga_loss[loss=0.2107, simple_loss=0.3035, pruned_loss=0.05893, over 28373.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3299, pruned_loss=0.08888, over 5646251.19 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3536, pruned_loss=0.1167, over 5660275.20 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3281, pruned_loss=0.08585, over 5652822.51 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:29:16,167 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 11:29:24,643 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 11:30:05,537 INFO [optim.py:369] (1/2) 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 11:30:20,628 INFO [train.py:968] (1/2) Epoch 14, batch 33050, giga_loss[loss=0.2209, simple_loss=0.2946, pruned_loss=0.07363, over 24516.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3325, pruned_loss=0.08929, over 5657019.22 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3538, pruned_loss=0.117, over 5668964.87 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3299, pruned_loss=0.08555, over 5653639.00 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:30:31,176 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:31:16,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=626398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:31:18,250 INFO [train.py:968] (1/2) Epoch 14, batch 33100, giga_loss[loss=0.2919, simple_loss=0.352, pruned_loss=0.1158, over 26723.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3361, pruned_loss=0.09108, over 5660153.98 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.354, pruned_loss=0.1173, over 5673745.99 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3336, pruned_loss=0.08739, over 5653190.14 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:31:51,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=626424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:31:57,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-07 11:32:04,176 INFO [optim.py:369] (1/2) 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 11:32:08,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0758, 1.1440, 3.3177, 2.9059], device='cuda:1'), covar=tensor([0.1624, 0.2692, 0.0479, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0607, 0.0878, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 11:32:21,636 INFO [train.py:968] (1/2) Epoch 14, batch 33150, giga_loss[loss=0.2623, simple_loss=0.3345, pruned_loss=0.09506, over 28958.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3365, pruned_loss=0.09123, over 5657728.81 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3537, pruned_loss=0.1169, over 5679690.03 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3344, pruned_loss=0.08798, over 5646458.80 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:33:24,160 INFO [train.py:968] (1/2) Epoch 14, batch 33200, libri_loss[loss=0.2749, simple_loss=0.3516, pruned_loss=0.09912, over 29299.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3357, pruned_loss=0.09071, over 5660822.38 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3531, pruned_loss=0.1165, over 5680141.64 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3342, pruned_loss=0.08793, over 5650831.74 frames. ], batch size: 94, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:33:59,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=626529.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 11:34:06,961 INFO [optim.py:369] (1/2) 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 11:34:13,289 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:16,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:22,834 INFO [train.py:968] (1/2) Epoch 14, batch 33250, giga_loss[loss=0.2512, simple_loss=0.3297, pruned_loss=0.08636, over 28886.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3333, pruned_loss=0.08896, over 5661229.91 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3532, pruned_loss=0.1167, over 5674458.25 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3315, pruned_loss=0.0861, over 5656962.95 frames. ], batch size: 199, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:34:43,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:46,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:34:51,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:35:25,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:35:25,934 INFO [train.py:968] (1/2) Epoch 14, batch 33300, giga_loss[loss=0.2482, simple_loss=0.3162, pruned_loss=0.09007, over 28640.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3316, pruned_loss=0.08835, over 5659882.48 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3534, pruned_loss=0.1168, over 5676475.01 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3298, pruned_loss=0.08563, over 5654518.35 frames. ], batch size: 85, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:36:01,159 INFO [optim.py:369] (1/2) 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 11:36:16,755 INFO [train.py:968] (1/2) Epoch 14, batch 33350, giga_loss[loss=0.357, simple_loss=0.4039, pruned_loss=0.155, over 27719.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3312, pruned_loss=0.08907, over 5666107.61 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3531, pruned_loss=0.1167, over 5677697.87 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3289, pruned_loss=0.08557, over 5659857.04 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:36:44,665 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626672.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 11:36:48,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626675.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 11:37:17,576 INFO [train.py:968] (1/2) Epoch 14, batch 33400, giga_loss[loss=0.2784, simple_loss=0.3594, pruned_loss=0.09874, over 28644.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3344, pruned_loss=0.09022, over 5670823.70 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3531, pruned_loss=0.1167, over 5681536.80 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08692, over 5662084.73 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:37:20,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4099, 1.6511, 1.4598, 1.6286], device='cuda:1'), covar=tensor([0.0697, 0.0276, 0.0309, 0.0699], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 11:37:23,527 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 11:38:00,519 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:38:16,612 INFO [train.py:968] (1/2) Epoch 14, batch 33450, giga_loss[loss=0.2705, simple_loss=0.3504, pruned_loss=0.0953, over 28793.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.335, pruned_loss=0.0909, over 5676284.64 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3525, pruned_loss=0.1164, over 5687659.43 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.08751, over 5663274.88 frames. ], batch size: 263, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:38:27,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:39:21,152 INFO [train.py:968] (1/2) Epoch 14, batch 33500, giga_loss[loss=0.2631, simple_loss=0.3383, pruned_loss=0.09392, over 29064.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3358, pruned_loss=0.09174, over 5670153.81 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3516, pruned_loss=0.1159, over 5685553.75 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3345, pruned_loss=0.08861, over 5661195.92 frames. ], batch size: 187, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:40:10,299 INFO [optim.py:369] (1/2) 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 11:40:12,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3809, 1.3765, 1.2354, 1.0750], device='cuda:1'), covar=tensor([0.0711, 0.0408, 0.0964, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0436, 0.0502, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 11:40:26,339 INFO [train.py:968] (1/2) Epoch 14, batch 33550, libri_loss[loss=0.2886, simple_loss=0.3535, pruned_loss=0.1119, over 27825.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3394, pruned_loss=0.09347, over 5663523.34 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3514, pruned_loss=0.1157, over 5679680.20 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3382, pruned_loss=0.09047, over 5662016.05 frames. ], batch size: 116, lr: 2.28e-03, grad_scale: 2.0
+2023-03-07 11:40:31,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2691, 3.0571, 1.4986, 1.4904], device='cuda:1'), covar=tensor([0.1027, 0.0425, 0.0902, 0.1388], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0519, 0.0357, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 11:40:35,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4462, 4.2910, 4.0662, 1.8649], device='cuda:1'), covar=tensor([0.0537, 0.0686, 0.0818, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.1005, 0.0869, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 11:40:39,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:40:43,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4453, 4.2817, 4.0044, 1.9509], device='cuda:1'), covar=tensor([0.0550, 0.0714, 0.0880, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.1091, 0.1005, 0.0869, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 11:40:45,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3121, 1.9128, 1.6118, 1.5376], device='cuda:1'), covar=tensor([0.0793, 0.0297, 0.0314, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 11:40:45,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4975, 1.7597, 1.7647, 1.3400], device='cuda:1'), covar=tensor([0.1909, 0.2380, 0.1506, 0.1802], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0673, 0.0887, 0.0792], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 11:40:54,632 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:40:54,686 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=626877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:40:59,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=626880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:41:19,111 INFO [train.py:968] (1/2) Epoch 14, batch 33600, libri_loss[loss=0.2726, simple_loss=0.3269, pruned_loss=0.1091, over 29570.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3415, pruned_loss=0.09393, over 5643818.81 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3512, pruned_loss=0.1157, over 5663562.78 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3404, pruned_loss=0.09078, over 5657314.95 frames. ], batch size: 76, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:41:30,119 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=626909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:42:10,689 INFO [optim.py:369] (1/2) 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 11:42:24,252 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=626948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:42:25,372 INFO [train.py:968] (1/2) Epoch 14, batch 33650, giga_loss[loss=0.2551, simple_loss=0.3326, pruned_loss=0.08881, over 28954.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3415, pruned_loss=0.09415, over 5649355.44 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3511, pruned_loss=0.1158, over 5667219.35 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3404, pruned_loss=0.09093, over 5656563.92 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:43:35,920 INFO [train.py:968] (1/2) Epoch 14, batch 33700, libri_loss[loss=0.3159, simple_loss=0.3756, pruned_loss=0.1281, over 29530.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3385, pruned_loss=0.09262, over 5655343.91 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.351, pruned_loss=0.1157, over 5672067.73 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3375, pruned_loss=0.08965, over 5656361.86 frames. ], batch size: 82, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:44:21,594 INFO [optim.py:369] (1/2) 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 11:44:35,105 INFO [train.py:968] (1/2) Epoch 14, batch 33750, giga_loss[loss=0.2356, simple_loss=0.3288, pruned_loss=0.07123, over 28886.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.338, pruned_loss=0.0926, over 5661951.84 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3506, pruned_loss=0.1153, over 5676592.50 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3373, pruned_loss=0.08993, over 5658451.15 frames. ], batch size: 136, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:45:40,492 INFO [train.py:968] (1/2) Epoch 14, batch 33800, giga_loss[loss=0.3487, simple_loss=0.3949, pruned_loss=0.1512, over 28015.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3376, pruned_loss=0.09287, over 5651189.15 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3508, pruned_loss=0.1155, over 5672398.93 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3365, pruned_loss=0.09004, over 5652103.41 frames. ], batch size: 412, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:45:41,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3376, 1.5907, 1.5808, 1.4029], device='cuda:1'), covar=tensor([0.1395, 0.1534, 0.1680, 0.1596], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0713, 0.0668, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 11:46:07,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5438, 1.7949, 1.8570, 1.3964], device='cuda:1'), covar=tensor([0.1741, 0.2402, 0.1429, 0.1694], device='cuda:1'), in_proj_covar=tensor([0.0843, 0.0674, 0.0886, 0.0791], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0014, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 11:46:17,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:46:20,348 INFO [optim.py:369] (1/2) 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 11:46:39,276 INFO [train.py:968] (1/2) Epoch 14, batch 33850, giga_loss[loss=0.2281, simple_loss=0.3071, pruned_loss=0.0746, over 28466.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3364, pruned_loss=0.09304, over 5656079.55 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3504, pruned_loss=0.1154, over 5668490.98 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3351, pruned_loss=0.08974, over 5659388.29 frames. ], batch size: 78, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:47:45,738 INFO [train.py:968] (1/2) Epoch 14, batch 33900, giga_loss[loss=0.2511, simple_loss=0.3366, pruned_loss=0.08285, over 28626.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3345, pruned_loss=0.09225, over 5632282.87 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3503, pruned_loss=0.1154, over 5662291.31 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3335, pruned_loss=0.08938, over 5640997.28 frames. ], batch size: 307, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:48:25,725 INFO [optim.py:369] (1/2) 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 11:48:27,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:48:43,291 INFO [train.py:968] (1/2) Epoch 14, batch 33950, giga_loss[loss=0.2343, simple_loss=0.3219, pruned_loss=0.07335, over 28591.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3335, pruned_loss=0.09057, over 5649034.71 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3494, pruned_loss=0.1149, over 5668213.12 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.333, pruned_loss=0.08812, over 5650091.88 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:48:48,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:49:18,340 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:49:21,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:49:22,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0014, 1.4947, 1.4366, 1.2067], device='cuda:1'), covar=tensor([0.1876, 0.1490, 0.2095, 0.1816], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0713, 0.0668, 0.0651], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 11:49:42,264 INFO [train.py:968] (1/2) Epoch 14, batch 34000, giga_loss[loss=0.2488, simple_loss=0.3504, pruned_loss=0.07365, over 28972.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3336, pruned_loss=0.089, over 5662896.76 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3496, pruned_loss=0.115, over 5670296.07 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3327, pruned_loss=0.08629, over 5661768.04 frames. ], batch size: 145, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:49:52,387 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:50:04,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:50:23,487 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 34050, giga_loss[loss=0.3082, simple_loss=0.3787, pruned_loss=0.1189, over 26779.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3367, pruned_loss=0.08953, over 5652062.17 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3497, pruned_loss=0.1151, over 5658079.89 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3354, pruned_loss=0.08642, over 5662547.15 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:51:06,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=627374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:16,452 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:19,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:30,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:34,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:51:35,405 INFO [train.py:968] (1/2) Epoch 14, batch 34100, giga_loss[loss=0.263, simple_loss=0.3426, pruned_loss=0.09167, over 28892.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3361, pruned_loss=0.08836, over 5652733.57 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3489, pruned_loss=0.1146, over 5660651.35 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3355, pruned_loss=0.08585, over 5658517.27 frames. ], batch size: 227, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:51:54,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:52:15,891 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:52:31,883 INFO [optim.py:369] (1/2) 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 11:52:34,803 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3182, 3.1537, 2.9842, 1.3795], device='cuda:1'), covar=tensor([0.0895, 0.1024, 0.0935, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.1084, 0.1001, 0.0866, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 11:52:49,817 INFO [train.py:968] (1/2) Epoch 14, batch 34150, giga_loss[loss=0.2465, simple_loss=0.33, pruned_loss=0.08156, over 28915.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3357, pruned_loss=0.0877, over 5659084.10 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3488, pruned_loss=0.1144, over 5664428.84 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3351, pruned_loss=0.0855, over 5660265.04 frames. ], batch size: 186, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:53:01,708 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-07 11:53:11,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:53:16,793 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:53:36,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 11:53:54,135 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 34200, giga_loss[loss=0.2558, simple_loss=0.3393, pruned_loss=0.08619, over 28386.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3372, pruned_loss=0.08875, over 5670376.14 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3492, pruned_loss=0.1146, over 5671096.05 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.336, pruned_loss=0.08605, over 5665224.32 frames. ], batch size: 336, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:54:50,277 INFO [optim.py:369] (1/2) 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 11:55:08,110 INFO [train.py:968] (1/2) Epoch 14, batch 34250, giga_loss[loss=0.2502, simple_loss=0.3268, pruned_loss=0.08677, over 26937.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3365, pruned_loss=0.08789, over 5662222.04 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3492, pruned_loss=0.1146, over 5669951.00 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3353, pruned_loss=0.08546, over 5659224.95 frames. ], batch size: 555, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:56:16,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3754, 4.1937, 3.9834, 1.9066], device='cuda:1'), covar=tensor([0.0568, 0.0679, 0.0732, 0.1996], device='cuda:1'), in_proj_covar=tensor([0.1085, 0.1004, 0.0866, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 11:56:21,317 INFO [train.py:968] (1/2) Epoch 14, batch 34300, giga_loss[loss=0.2208, simple_loss=0.2869, pruned_loss=0.07734, over 24685.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3362, pruned_loss=0.08739, over 5662381.07 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3486, pruned_loss=0.1141, over 5674838.22 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3356, pruned_loss=0.08517, over 5655556.79 frames. ], batch size: 705, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:57:12,099 INFO [optim.py:369] (1/2) 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 11:57:28,031 INFO [train.py:968] (1/2) Epoch 14, batch 34350, giga_loss[loss=0.3193, simple_loss=0.3814, pruned_loss=0.1287, over 27678.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3399, pruned_loss=0.08894, over 5665361.50 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3485, pruned_loss=0.1139, over 5675202.04 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3392, pruned_loss=0.08672, over 5659518.50 frames. ], batch size: 472, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 11:58:34,800 INFO [train.py:968] (1/2) Epoch 14, batch 34400, giga_loss[loss=0.2554, simple_loss=0.3393, pruned_loss=0.0857, over 28784.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3408, pruned_loss=0.08904, over 5668442.05 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3488, pruned_loss=0.1142, over 5668081.50 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3398, pruned_loss=0.08647, over 5670835.66 frames. ], batch size: 262, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 11:59:30,913 INFO [optim.py:369] (1/2) 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 11:59:45,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=627749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 11:59:46,121 INFO [train.py:968] (1/2) Epoch 14, batch 34450, giga_loss[loss=0.2196, simple_loss=0.3022, pruned_loss=0.06847, over 28362.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3401, pruned_loss=0.08987, over 5679050.59 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3489, pruned_loss=0.1142, over 5673254.45 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.339, pruned_loss=0.08713, over 5676271.04 frames. ], batch size: 71, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 12:00:54,534 INFO [train.py:968] (1/2) Epoch 14, batch 34500, giga_loss[loss=0.2123, simple_loss=0.3006, pruned_loss=0.062, over 28806.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3382, pruned_loss=0.08949, over 5687312.63 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.348, pruned_loss=0.1136, over 5679664.84 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3378, pruned_loss=0.08714, over 5679780.85 frames. ], batch size: 119, lr: 2.28e-03, grad_scale: 8.0
+2023-03-07 12:01:07,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=627807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:01:50,065 INFO [optim.py:369] (1/2) 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 12:02:07,026 INFO [train.py:968] (1/2) Epoch 14, batch 34550, giga_loss[loss=0.2311, simple_loss=0.316, pruned_loss=0.07308, over 28952.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3363, pruned_loss=0.08719, over 5695171.15 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3482, pruned_loss=0.1137, over 5682061.86 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3358, pruned_loss=0.08503, over 5687117.53 frames. ], batch size: 213, lr: 2.28e-03, grad_scale: 4.0
+2023-03-07 12:03:02,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=627892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:03:06,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=627895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:03:13,086 INFO [train.py:968] (1/2) Epoch 14, batch 34600, libri_loss[loss=0.2843, simple_loss=0.35, pruned_loss=0.1093, over 29528.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3363, pruned_loss=0.08726, over 5695881.78 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3482, pruned_loss=0.1137, over 5685955.62 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3355, pruned_loss=0.08493, over 5686228.18 frames. ], batch size: 83, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:03:42,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=627924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:03:58,659 INFO [optim.py:369] (1/2) 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 12:04:12,814 INFO [train.py:968] (1/2) Epoch 14, batch 34650, giga_loss[loss=0.2504, simple_loss=0.3383, pruned_loss=0.08122, over 28591.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3378, pruned_loss=0.08828, over 5690519.57 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3483, pruned_loss=0.1136, over 5690845.66 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3369, pruned_loss=0.08585, over 5678599.98 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:04:42,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2062, 1.4821, 1.4963, 1.2685], device='cuda:1'), covar=tensor([0.1523, 0.1511, 0.1892, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0710, 0.0667, 0.0647], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 12:05:11,165 INFO [train.py:968] (1/2) Epoch 14, batch 34700, giga_loss[loss=0.2349, simple_loss=0.3204, pruned_loss=0.07467, over 28795.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.339, pruned_loss=0.08937, over 5674966.54 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3485, pruned_loss=0.1139, over 5686297.38 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3378, pruned_loss=0.08652, over 5668837.39 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:05:27,977 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:06:02,645 INFO [optim.py:369] (1/2) 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 12:06:14,503 INFO [train.py:968] (1/2) Epoch 14, batch 34750, giga_loss[loss=0.2405, simple_loss=0.3237, pruned_loss=0.07861, over 28890.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3355, pruned_loss=0.08843, over 5674494.12 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3486, pruned_loss=0.114, over 5688660.09 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3343, pruned_loss=0.08589, over 5667567.51 frames. ], batch size: 186, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:07:13,571 INFO [train.py:968] (1/2) Epoch 14, batch 34800, giga_loss[loss=0.2256, simple_loss=0.3094, pruned_loss=0.07096, over 28679.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.0892, over 5673203.33 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3485, pruned_loss=0.1139, over 5689215.40 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3346, pruned_loss=0.08691, over 5666908.23 frames. ], batch size: 242, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:07:57,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 12:07:59,488 INFO [optim.py:369] (1/2) 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 12:08:09,412 INFO [train.py:968] (1/2) Epoch 14, batch 34850, giga_loss[loss=0.3057, simple_loss=0.3862, pruned_loss=0.1126, over 29028.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3396, pruned_loss=0.09194, over 5670393.52 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3486, pruned_loss=0.1139, over 5692833.18 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3384, pruned_loss=0.08948, over 5661564.98 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:08:39,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:08:58,140 INFO [train.py:968] (1/2) Epoch 14, batch 34900, giga_loss[loss=0.2973, simple_loss=0.3741, pruned_loss=0.1103, over 28955.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3479, pruned_loss=0.09646, over 5679479.55 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3481, pruned_loss=0.1136, over 5697359.74 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3473, pruned_loss=0.09443, over 5668075.39 frames. ], batch size: 106, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:09:37,776 INFO [optim.py:369] (1/2) 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 12:09:41,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-07 12:09:46,130 INFO [train.py:968] (1/2) Epoch 14, batch 34950, giga_loss[loss=0.2884, simple_loss=0.3605, pruned_loss=0.1081, over 28939.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3537, pruned_loss=0.09976, over 5682929.88 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3485, pruned_loss=0.1138, over 5696627.95 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.353, pruned_loss=0.09776, over 5674448.62 frames. ], batch size: 106, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:10:07,419 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:10:12,071 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:10:30,817 INFO [train.py:968] (1/2) Epoch 14, batch 35000, giga_loss[loss=0.2778, simple_loss=0.346, pruned_loss=0.1048, over 27675.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3507, pruned_loss=0.09877, over 5683137.52 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3488, pruned_loss=0.114, over 5699237.24 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3498, pruned_loss=0.09687, over 5673922.10 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:10:50,637 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:10:52,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:11:02,124 INFO [optim.py:369] (1/2) 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 12:11:11,802 INFO [train.py:968] (1/2) Epoch 14, batch 35050, giga_loss[loss=0.2492, simple_loss=0.3223, pruned_loss=0.08806, over 29005.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3446, pruned_loss=0.09684, over 5687479.02 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3491, pruned_loss=0.1142, over 5701739.55 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3436, pruned_loss=0.09461, over 5677770.30 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:11:18,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=628357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:11:43,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:11:46,419 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 12:11:52,649 INFO [train.py:968] (1/2) Epoch 14, batch 35100, libri_loss[loss=0.2671, simple_loss=0.3293, pruned_loss=0.1024, over 29481.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3385, pruned_loss=0.09482, over 5673804.60 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3494, pruned_loss=0.1144, over 5686111.20 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3371, pruned_loss=0.09209, over 5679067.57 frames. ], batch size: 70, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:12:28,652 INFO [optim.py:369] (1/2) 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 12:12:35,886 INFO [train.py:968] (1/2) Epoch 14, batch 35150, giga_loss[loss=0.2172, simple_loss=0.2909, pruned_loss=0.07172, over 28999.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3309, pruned_loss=0.09157, over 5677484.14 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3497, pruned_loss=0.1145, over 5689723.41 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3292, pruned_loss=0.08887, over 5678269.44 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:13:04,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2249, 4.0404, 3.8272, 1.9185], device='cuda:1'), covar=tensor([0.0591, 0.0753, 0.0715, 0.2097], device='cuda:1'), in_proj_covar=tensor([0.1090, 0.1009, 0.0871, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 12:13:17,197 INFO [train.py:968] (1/2) Epoch 14, batch 35200, giga_loss[loss=0.2026, simple_loss=0.2816, pruned_loss=0.06177, over 29095.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3241, pruned_loss=0.08855, over 5683710.84 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3503, pruned_loss=0.1149, over 5692391.29 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3218, pruned_loss=0.08551, over 5681772.60 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:13:23,977 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-07 12:13:35,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5934, 2.2411, 1.6942, 0.7581], device='cuda:1'), covar=tensor([0.4788, 0.2498, 0.3503, 0.5339], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1541, 0.1518, 0.1319], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 12:13:49,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:13:51,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:13:55,145 INFO [optim.py:369] (1/2) 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 12:14:04,362 INFO [train.py:968] (1/2) Epoch 14, batch 35250, giga_loss[loss=0.2656, simple_loss=0.3289, pruned_loss=0.1011, over 28732.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3192, pruned_loss=0.0865, over 5687954.66 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3503, pruned_loss=0.1149, over 5694365.88 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.317, pruned_loss=0.08375, over 5684651.94 frames. ], batch size: 243, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:14:16,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=628564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:14:44,571 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5854, 1.2828, 4.5493, 3.5110], device='cuda:1'), covar=tensor([0.1588, 0.2729, 0.0339, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0609, 0.0886, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 12:14:45,667 INFO [train.py:968] (1/2) Epoch 14, batch 35300, giga_loss[loss=0.2419, simple_loss=0.3141, pruned_loss=0.08485, over 28813.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3169, pruned_loss=0.08586, over 5694194.09 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3508, pruned_loss=0.115, over 5695841.65 frames. ], giga_tot_loss[loss=0.2404, simple_loss=0.3144, pruned_loss=0.0832, over 5690257.15 frames. ], batch size: 186, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:15:21,764 INFO [optim.py:369] (1/2) 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 12:15:28,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:15:29,294 INFO [train.py:968] (1/2) Epoch 14, batch 35350, giga_loss[loss=0.2107, simple_loss=0.2717, pruned_loss=0.07481, over 23936.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3141, pruned_loss=0.08481, over 5664470.79 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3514, pruned_loss=0.1154, over 5674626.22 frames. ], giga_tot_loss[loss=0.2366, simple_loss=0.3103, pruned_loss=0.08139, over 5679940.67 frames. ], batch size: 705, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:15:32,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=628654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:15:58,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7441, 4.5884, 4.3173, 2.2361], device='cuda:1'), covar=tensor([0.0480, 0.0658, 0.0690, 0.1791], device='cuda:1'), in_proj_covar=tensor([0.1082, 0.1001, 0.0864, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 12:16:13,748 INFO [train.py:968] (1/2) Epoch 14, batch 35400, giga_loss[loss=0.2143, simple_loss=0.2836, pruned_loss=0.07248, over 28733.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3106, pruned_loss=0.08322, over 5664934.89 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3512, pruned_loss=0.1152, over 5677758.26 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3073, pruned_loss=0.08026, over 5674437.91 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:16:33,435 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-07 12:16:46,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3689, 1.4629, 1.3422, 1.6172], device='cuda:1'), covar=tensor([0.0767, 0.0335, 0.0336, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0057, 0.0096], device='cuda:1')
+2023-03-07 12:16:50,654 INFO [optim.py:369] (1/2) 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 12:16:54,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-07 12:16:56,252 INFO [train.py:968] (1/2) Epoch 14, batch 35450, giga_loss[loss=0.1918, simple_loss=0.2696, pruned_loss=0.05704, over 28404.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3078, pruned_loss=0.08193, over 5667307.10 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3515, pruned_loss=0.1153, over 5674117.42 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3039, pruned_loss=0.0787, over 5678116.18 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:17:04,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4904, 1.7643, 1.4144, 1.6813], device='cuda:1'), covar=tensor([0.2469, 0.2328, 0.2577, 0.2174], device='cuda:1'), in_proj_covar=tensor([0.1363, 0.0997, 0.1210, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 12:17:07,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5344, 4.3457, 4.1623, 2.0837], device='cuda:1'), covar=tensor([0.0706, 0.0883, 0.0922, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1078, 0.0998, 0.0860, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 12:17:16,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6854, 1.8436, 1.6960, 1.5480], device='cuda:1'), covar=tensor([0.1797, 0.2445, 0.2113, 0.2372], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0724, 0.0674, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 12:17:32,630 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:34,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:36,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=628797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:38,148 INFO [train.py:968] (1/2) Epoch 14, batch 35500, giga_loss[loss=0.1934, simple_loss=0.2722, pruned_loss=0.05732, over 28558.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.3068, pruned_loss=0.08173, over 5670100.59 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3521, pruned_loss=0.1157, over 5670644.14 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.302, pruned_loss=0.07794, over 5682063.51 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:17:38,458 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=628800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:17:58,658 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=628829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:18:11,609 INFO [optim.py:369] (1/2) 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 12:18:18,662 INFO [train.py:968] (1/2) Epoch 14, batch 35550, giga_loss[loss=0.2207, simple_loss=0.2832, pruned_loss=0.07917, over 23948.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.304, pruned_loss=0.08028, over 5676341.17 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3521, pruned_loss=0.1154, over 5674920.95 frames. ], giga_tot_loss[loss=0.2254, simple_loss=0.2984, pruned_loss=0.07618, over 5682504.77 frames. ], batch size: 705, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:18:47,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0738, 1.2032, 3.4241, 2.9382], device='cuda:1'), covar=tensor([0.1631, 0.2627, 0.0503, 0.0962], device='cuda:1'), in_proj_covar=tensor([0.0691, 0.0610, 0.0887, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 12:18:53,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5754, 1.4413, 5.1803, 3.6061], device='cuda:1'), covar=tensor([0.1776, 0.2850, 0.0327, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0610, 0.0887, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 12:19:01,078 INFO [train.py:968] (1/2) Epoch 14, batch 35600, giga_loss[loss=0.1858, simple_loss=0.2671, pruned_loss=0.0522, over 28907.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3035, pruned_loss=0.08037, over 5679696.13 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.353, pruned_loss=0.1159, over 5677633.99 frames. ], giga_tot_loss[loss=0.2238, simple_loss=0.2966, pruned_loss=0.07547, over 5682306.43 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:19:36,867 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 35650, giga_loss[loss=0.2159, simple_loss=0.2881, pruned_loss=0.07181, over 28590.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3024, pruned_loss=0.08029, over 5676911.16 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.353, pruned_loss=0.1159, over 5681863.20 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.2939, pruned_loss=0.07431, over 5674699.77 frames. ], batch size: 336, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:20:24,518 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=628998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:20:25,671 INFO [train.py:968] (1/2) Epoch 14, batch 35700, giga_loss[loss=0.2906, simple_loss=0.3671, pruned_loss=0.107, over 28886.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3074, pruned_loss=0.08341, over 5659525.69 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3535, pruned_loss=0.1162, over 5664646.84 frames. ], giga_tot_loss[loss=0.2273, simple_loss=0.2993, pruned_loss=0.07766, over 5672753.93 frames. ], batch size: 145, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:20:41,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2663, 1.2156, 3.9965, 3.1745], device='cuda:1'), covar=tensor([0.1720, 0.2766, 0.0400, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0692, 0.0611, 0.0890, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 12:21:02,303 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 35750, giga_loss[loss=0.268, simple_loss=0.3469, pruned_loss=0.09456, over 28866.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.32, pruned_loss=0.09013, over 5677598.91 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3534, pruned_loss=0.1161, over 5672590.69 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3119, pruned_loss=0.0845, over 5681461.44 frames. ], batch size: 199, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:21:57,061 INFO [train.py:968] (1/2) Epoch 14, batch 35800, giga_loss[loss=0.3824, simple_loss=0.431, pruned_loss=0.1669, over 27611.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.333, pruned_loss=0.09697, over 5671159.21 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3534, pruned_loss=0.116, over 5675067.31 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3261, pruned_loss=0.09215, over 5672131.80 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:22:04,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:22:32,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629139.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 12:22:34,490 INFO [optim.py:369] (1/2) 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 12:22:41,256 INFO [train.py:968] (1/2) Epoch 14, batch 35850, libri_loss[loss=0.3527, simple_loss=0.398, pruned_loss=0.1537, over 19262.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3418, pruned_loss=0.1008, over 5666774.10 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3539, pruned_loss=0.1162, over 5665831.19 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3356, pruned_loss=0.09651, over 5676460.40 frames. ], batch size: 187, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:22:49,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-07 12:23:20,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2613, 2.6426, 1.2621, 1.3921], device='cuda:1'), covar=tensor([0.0867, 0.0406, 0.0858, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0513, 0.0351, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 12:23:23,411 INFO [train.py:968] (1/2) Epoch 14, batch 35900, giga_loss[loss=0.253, simple_loss=0.3308, pruned_loss=0.08757, over 28771.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3458, pruned_loss=0.1017, over 5680440.20 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3544, pruned_loss=0.1166, over 5672905.90 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3401, pruned_loss=0.09753, over 5681753.93 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:23:59,775 INFO [optim.py:369] (1/2) 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 12:24:05,689 INFO [train.py:968] (1/2) Epoch 14, batch 35950, giga_loss[loss=0.2534, simple_loss=0.3388, pruned_loss=0.08401, over 28963.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.348, pruned_loss=0.1021, over 5663171.71 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3552, pruned_loss=0.117, over 5666781.86 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3424, pruned_loss=0.09771, over 5669813.41 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:24:53,898 INFO [train.py:968] (1/2) Epoch 14, batch 36000, giga_loss[loss=0.2551, simple_loss=0.3335, pruned_loss=0.08833, over 28444.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3489, pruned_loss=0.1019, over 5663484.49 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3554, pruned_loss=0.1171, over 5669255.37 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3443, pruned_loss=0.0983, over 5666619.81 frames. ], batch size: 71, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:24:53,899 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 12:25:03,381 INFO [train.py:1012] (1/2) Epoch 14, validation: loss=0.2087, simple_loss=0.3148, pruned_loss=0.0513, over 944034.00 frames. 
+2023-03-07 12:25:03,382 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 12:25:38,858 INFO [optim.py:369] (1/2) 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 12:25:45,989 INFO [train.py:968] (1/2) Epoch 14, batch 36050, giga_loss[loss=0.2989, simple_loss=0.379, pruned_loss=0.1094, over 28674.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3524, pruned_loss=0.1044, over 5679838.59 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3556, pruned_loss=0.1171, over 5672774.11 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3483, pruned_loss=0.1011, over 5679445.08 frames. ], batch size: 262, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:26:04,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:26:25,413 INFO [train.py:968] (1/2) Epoch 14, batch 36100, giga_loss[loss=0.271, simple_loss=0.3479, pruned_loss=0.0971, over 28899.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3546, pruned_loss=0.1064, over 5684787.86 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3553, pruned_loss=0.1168, over 5680560.43 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3515, pruned_loss=0.1035, over 5678112.15 frames. ], batch size: 99, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:26:43,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4441, 1.6513, 1.4701, 1.4751], device='cuda:1'), covar=tensor([0.0821, 0.0310, 0.0321, 0.0917], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 12:26:44,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:27:01,452 INFO [optim.py:369] (1/2) 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 12:27:08,413 INFO [train.py:968] (1/2) Epoch 14, batch 36150, giga_loss[loss=0.28, simple_loss=0.3638, pruned_loss=0.09808, over 29040.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3573, pruned_loss=0.1072, over 5682046.83 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.356, pruned_loss=0.117, over 5672833.05 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3542, pruned_loss=0.1043, over 5684360.86 frames. ], batch size: 155, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:27:10,196 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:27:34,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:27:47,531 INFO [train.py:968] (1/2) Epoch 14, batch 36200, giga_loss[loss=0.3187, simple_loss=0.3906, pruned_loss=0.1234, over 28373.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.359, pruned_loss=0.1066, over 5687749.22 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3562, pruned_loss=0.1171, over 5665163.80 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3563, pruned_loss=0.1041, over 5697369.77 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:28:00,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629514.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 12:28:02,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:28:04,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:28:25,718 INFO [optim.py:369] (1/2) 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 12:28:33,214 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:28:34,172 INFO [train.py:968] (1/2) Epoch 14, batch 36250, libri_loss[loss=0.2994, simple_loss=0.3739, pruned_loss=0.1124, over 29528.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3607, pruned_loss=0.1072, over 5683715.50 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3566, pruned_loss=0.1172, over 5670416.19 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3583, pruned_loss=0.1048, over 5686889.24 frames. ], batch size: 82, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:28:36,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:29:00,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8072, 4.6283, 4.4628, 1.9510], device='cuda:1'), covar=tensor([0.0581, 0.0712, 0.0794, 0.1785], device='cuda:1'), in_proj_covar=tensor([0.1077, 0.0998, 0.0862, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 12:29:13,775 INFO [train.py:968] (1/2) Epoch 14, batch 36300, giga_loss[loss=0.2561, simple_loss=0.319, pruned_loss=0.09659, over 23637.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3618, pruned_loss=0.1068, over 5688540.67 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3567, pruned_loss=0.1172, over 5672791.09 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3598, pruned_loss=0.1049, over 5689058.36 frames. ], batch size: 705, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:29:35,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:29:37,024 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:29:54,509 INFO [train.py:968] (1/2) Epoch 14, batch 36350, giga_loss[loss=0.2701, simple_loss=0.35, pruned_loss=0.09511, over 28276.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3601, pruned_loss=0.1044, over 5694017.77 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3572, pruned_loss=0.1176, over 5671006.75 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3582, pruned_loss=0.1024, over 5696517.14 frames. ], batch size: 368, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:30:01,542 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:30:01,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629657.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 12:30:03,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5288, 1.6785, 1.3660, 1.6756], device='cuda:1'), covar=tensor([0.2561, 0.2602, 0.2867, 0.2349], device='cuda:1'), in_proj_covar=tensor([0.1360, 0.0997, 0.1205, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 12:30:04,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629660.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 12:30:29,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629689.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 12:30:37,875 INFO [train.py:968] (1/2) Epoch 14, batch 36400, giga_loss[loss=0.2873, simple_loss=0.3573, pruned_loss=0.1086, over 28598.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3581, pruned_loss=0.1026, over 5694893.92 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3573, pruned_loss=0.1176, over 5673292.04 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3565, pruned_loss=0.1009, over 5695177.31 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:31:13,602 INFO [optim.py:369] (1/2) 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 12:31:17,892 INFO [train.py:968] (1/2) Epoch 14, batch 36450, giga_loss[loss=0.2614, simple_loss=0.3481, pruned_loss=0.08734, over 28378.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3583, pruned_loss=0.1032, over 5682477.47 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3573, pruned_loss=0.1176, over 5666768.89 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3571, pruned_loss=0.1017, over 5688422.43 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:32:01,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:32:04,775 INFO [train.py:968] (1/2) Epoch 14, batch 36500, giga_loss[loss=0.3005, simple_loss=0.3687, pruned_loss=0.1161, over 28630.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3617, pruned_loss=0.108, over 5685049.27 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3581, pruned_loss=0.1179, over 5671926.89 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3601, pruned_loss=0.1062, over 5685261.87 frames. ], batch size: 71, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:32:06,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6940, 1.7610, 1.7106, 1.6310], device='cuda:1'), covar=tensor([0.1717, 0.2087, 0.2225, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0728, 0.0683, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 12:32:22,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-07 12:32:28,354 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 14, batch 36550, giga_loss[loss=0.2933, simple_loss=0.3551, pruned_loss=0.1158, over 28765.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3626, pruned_loss=0.1105, over 5688482.54 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.358, pruned_loss=0.1177, over 5674273.55 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3614, pruned_loss=0.1091, over 5686684.64 frames. ], batch size: 99, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:33:34,559 INFO [train.py:968] (1/2) Epoch 14, batch 36600, giga_loss[loss=0.281, simple_loss=0.3523, pruned_loss=0.1049, over 28562.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3619, pruned_loss=0.1113, over 5689112.99 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3584, pruned_loss=0.1179, over 5676281.91 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3608, pruned_loss=0.11, over 5686081.93 frames. ], batch size: 336, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:33:49,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 12:33:55,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=629923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:33:58,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=629928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:07,362 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:09,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:10,543 INFO [optim.py:369] (1/2) 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 12:34:15,983 INFO [train.py:968] (1/2) Epoch 14, batch 36650, giga_loss[loss=0.2842, simple_loss=0.3518, pruned_loss=0.1083, over 28901.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3618, pruned_loss=0.1123, over 5696396.72 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3588, pruned_loss=0.1182, over 5674515.68 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3605, pruned_loss=0.1109, over 5695720.11 frames. ], batch size: 112, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:34:31,828 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:33,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=629971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:34,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=629972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:34:58,850 INFO [train.py:968] (1/2) Epoch 14, batch 36700, libri_loss[loss=0.3019, simple_loss=0.3732, pruned_loss=0.1153, over 29512.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3593, pruned_loss=0.1106, over 5693929.13 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3593, pruned_loss=0.1183, over 5672231.80 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.358, pruned_loss=0.109, over 5695589.52 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:35:00,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=630001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:35:34,904 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 36750, giga_loss[loss=0.2634, simple_loss=0.3395, pruned_loss=0.09371, over 29059.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3574, pruned_loss=0.1083, over 5695556.89 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3594, pruned_loss=0.1184, over 5677406.15 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3561, pruned_loss=0.1068, over 5692896.14 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:36:00,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=630071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:36:04,052 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=630074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:36:27,826 INFO [train.py:968] (1/2) Epoch 14, batch 36800, giga_loss[loss=0.2461, simple_loss=0.3235, pruned_loss=0.08434, over 28855.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3539, pruned_loss=0.1059, over 5677465.12 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3597, pruned_loss=0.1186, over 5663594.59 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3526, pruned_loss=0.1043, over 5687493.70 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:36:31,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=630103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:36:44,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-07 12:37:10,625 INFO [optim.py:369] (1/2) 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 12:37:17,360 INFO [train.py:968] (1/2) Epoch 14, batch 36850, giga_loss[loss=0.235, simple_loss=0.3138, pruned_loss=0.07811, over 28587.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.348, pruned_loss=0.1023, over 5680361.29 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3601, pruned_loss=0.1187, over 5656518.31 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3465, pruned_loss=0.1007, over 5694657.61 frames. ], batch size: 336, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:38:09,436 INFO [train.py:968] (1/2) Epoch 14, batch 36900, giga_loss[loss=0.234, simple_loss=0.3115, pruned_loss=0.07824, over 28900.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.341, pruned_loss=0.09868, over 5664421.91 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3604, pruned_loss=0.119, over 5656237.50 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3393, pruned_loss=0.09692, over 5675891.52 frames. ], batch size: 186, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:38:54,684 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 36950, giga_loss[loss=0.2562, simple_loss=0.3404, pruned_loss=0.08603, over 29036.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3378, pruned_loss=0.09669, over 5658661.92 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3611, pruned_loss=0.1195, over 5648427.78 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3354, pruned_loss=0.09443, over 5674745.04 frames. ], batch size: 155, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:39:06,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 12:39:11,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-07 12:39:33,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9211, 1.0817, 1.0521, 0.8575], device='cuda:1'), covar=tensor([0.2236, 0.2346, 0.1218, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.1809, 0.1695, 0.1639, 0.1784], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 12:39:45,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=630298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:39:46,174 INFO [train.py:968] (1/2) Epoch 14, batch 37000, giga_loss[loss=0.28, simple_loss=0.3565, pruned_loss=0.1018, over 28942.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3392, pruned_loss=0.09712, over 5662195.48 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3613, pruned_loss=0.1196, over 5647605.64 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3371, pruned_loss=0.09517, over 5675471.34 frames. ], batch size: 213, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:39:58,934 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-07 12:40:13,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6770, 1.7385, 1.8210, 1.4013], device='cuda:1'), covar=tensor([0.1696, 0.2280, 0.1348, 0.1615], device='cuda:1'), in_proj_covar=tensor([0.0853, 0.0688, 0.0897, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 12:40:24,282 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 37050, giga_loss[loss=0.2391, simple_loss=0.3097, pruned_loss=0.08423, over 28613.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3389, pruned_loss=0.09646, over 5679901.89 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3619, pruned_loss=0.1197, over 5649376.09 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3364, pruned_loss=0.09448, over 5689256.23 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:41:01,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7840, 1.9358, 2.0570, 1.5680], device='cuda:1'), covar=tensor([0.1939, 0.2146, 0.1440, 0.1633], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0688, 0.0896, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 12:41:12,076 INFO [train.py:968] (1/2) Epoch 14, batch 37100, giga_loss[loss=0.2826, simple_loss=0.3591, pruned_loss=0.103, over 28761.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3387, pruned_loss=0.09709, over 5677047.36 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3626, pruned_loss=0.12, over 5652263.92 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3357, pruned_loss=0.09493, over 5682145.11 frames. ], batch size: 284, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:41:31,763 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=630425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:41:41,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-07 12:41:45,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=630441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:41:47,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=630444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:41:48,136 INFO [optim.py:369] (1/2) 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 12:41:52,056 INFO [train.py:968] (1/2) Epoch 14, batch 37150, giga_loss[loss=0.2172, simple_loss=0.2913, pruned_loss=0.07157, over 28306.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3362, pruned_loss=0.0957, over 5691673.53 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3627, pruned_loss=0.1199, over 5654942.78 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3335, pruned_loss=0.09384, over 5693577.02 frames. ], batch size: 77, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:42:04,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4624, 1.6020, 1.4758, 1.3039], device='cuda:1'), covar=tensor([0.2553, 0.2205, 0.1657, 0.2344], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1694, 0.1639, 0.1792], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 12:42:10,429 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 37200, libri_loss[loss=0.2827, simple_loss=0.3529, pruned_loss=0.1063, over 29557.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3352, pruned_loss=0.09528, over 5698930.69 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3634, pruned_loss=0.12, over 5656732.30 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3309, pruned_loss=0.09247, over 5701023.95 frames. ], batch size: 79, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:43:05,524 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 37250, giga_loss[loss=0.2375, simple_loss=0.3155, pruned_loss=0.0798, over 29013.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3331, pruned_loss=0.09386, over 5712082.75 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3636, pruned_loss=0.1196, over 5664314.70 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3287, pruned_loss=0.09119, over 5708178.57 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:43:22,487 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5897, 4.5358, 1.7798, 1.6607], device='cuda:1'), covar=tensor([0.0951, 0.0268, 0.0843, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0513, 0.0352, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 12:43:39,056 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=630588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:43:48,244 INFO [train.py:968] (1/2) Epoch 14, batch 37300, giga_loss[loss=0.298, simple_loss=0.3599, pruned_loss=0.1181, over 27536.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3322, pruned_loss=0.0938, over 5705063.61 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3643, pruned_loss=0.1199, over 5665004.36 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3273, pruned_loss=0.09084, over 5702307.27 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:44:24,600 INFO [optim.py:369] (1/2) 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 12:44:26,679 INFO [train.py:968] (1/2) Epoch 14, batch 37350, giga_loss[loss=0.2423, simple_loss=0.317, pruned_loss=0.08377, over 28978.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3308, pruned_loss=0.09317, over 5702592.62 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3658, pruned_loss=0.1208, over 5658089.98 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3245, pruned_loss=0.08914, over 5707361.06 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:45:05,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1421, 3.2272, 2.3079, 1.1105], device='cuda:1'), covar=tensor([0.4785, 0.2294, 0.2585, 0.4563], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1527, 0.1513, 0.1306], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 12:45:07,847 INFO [train.py:968] (1/2) Epoch 14, batch 37400, giga_loss[loss=0.2228, simple_loss=0.2969, pruned_loss=0.07435, over 28919.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3284, pruned_loss=0.09187, over 5699561.60 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3669, pruned_loss=0.1213, over 5650135.37 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3218, pruned_loss=0.0877, over 5710766.09 frames. ], batch size: 227, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:45:17,660 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-07 12:45:37,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5123, 1.6996, 1.3668, 1.7215], device='cuda:1'), covar=tensor([0.2541, 0.2610, 0.2878, 0.2323], device='cuda:1'), in_proj_covar=tensor([0.1368, 0.1004, 0.1208, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 12:45:43,583 INFO [optim.py:369] (1/2) 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 12:45:45,634 INFO [train.py:968] (1/2) Epoch 14, batch 37450, giga_loss[loss=0.2465, simple_loss=0.3201, pruned_loss=0.08638, over 28547.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3278, pruned_loss=0.0919, over 5712662.36 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3678, pruned_loss=0.1219, over 5654514.92 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3207, pruned_loss=0.08734, over 5719016.49 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:46:13,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4146, 1.5862, 1.2743, 1.5071], device='cuda:1'), covar=tensor([0.0775, 0.0322, 0.0329, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 12:46:27,407 INFO [train.py:968] (1/2) Epoch 14, batch 37500, giga_loss[loss=0.2576, simple_loss=0.3223, pruned_loss=0.09642, over 28874.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3256, pruned_loss=0.09045, over 5721999.98 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.368, pruned_loss=0.1218, over 5659238.89 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.319, pruned_loss=0.08628, over 5723947.75 frames. ], batch size: 136, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:46:27,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=630800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:47:05,589 INFO [optim.py:369] (1/2) 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 12:47:08,451 INFO [train.py:968] (1/2) Epoch 14, batch 37550, giga_loss[loss=0.2638, simple_loss=0.3425, pruned_loss=0.09254, over 28977.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3283, pruned_loss=0.09216, over 5713977.14 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3693, pruned_loss=0.1225, over 5655823.16 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3207, pruned_loss=0.08735, over 5719765.81 frames. ], batch size: 145, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 12:47:17,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9413, 3.7472, 3.6080, 1.6302], device='cuda:1'), covar=tensor([0.0729, 0.0882, 0.0930, 0.1940], device='cuda:1'), in_proj_covar=tensor([0.1082, 0.1004, 0.0867, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 12:47:53,111 INFO [train.py:968] (1/2) Epoch 14, batch 37600, giga_loss[loss=0.2684, simple_loss=0.3367, pruned_loss=0.1001, over 28181.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3326, pruned_loss=0.09462, over 5705082.88 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3698, pruned_loss=0.1226, over 5651785.01 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3254, pruned_loss=0.09007, over 5713882.61 frames. ], batch size: 77, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:48:06,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8435, 1.9589, 1.7164, 1.7814], device='cuda:1'), covar=tensor([0.1770, 0.2132, 0.2226, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0736, 0.0691, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 12:48:11,887 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=630921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:48:34,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=630943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:48:38,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=630946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:48:38,569 INFO [optim.py:369] (1/2) 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 12:48:40,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3824, 1.9222, 1.3563, 0.6214], device='cuda:1'), covar=tensor([0.4380, 0.2214, 0.2942, 0.5104], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1526, 0.1513, 0.1307], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1')
+2023-03-07 12:48:41,794 INFO [train.py:968] (1/2) Epoch 14, batch 37650, giga_loss[loss=0.3609, simple_loss=0.4172, pruned_loss=0.1523, over 27560.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3394, pruned_loss=0.09907, over 5692480.42 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3698, pruned_loss=0.1227, over 5654250.98 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3333, pruned_loss=0.09523, over 5697671.68 frames. ], batch size: 472, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:48:55,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=630963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:49:05,773 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=630975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:49:28,386 INFO [train.py:968] (1/2) Epoch 14, batch 37700, libri_loss[loss=0.337, simple_loss=0.3883, pruned_loss=0.1429, over 29525.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3466, pruned_loss=0.1041, over 5691494.61 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3693, pruned_loss=0.1224, over 5663365.60 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3413, pruned_loss=0.1006, over 5688501.62 frames. ], batch size: 80, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:49:31,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2112, 3.0567, 1.4332, 1.3369], device='cuda:1'), covar=tensor([0.1011, 0.0334, 0.0896, 0.1388], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0518, 0.0354, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 12:49:33,950 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5463, 1.7390, 1.6100, 1.5141], device='cuda:1'), covar=tensor([0.1727, 0.2003, 0.2125, 0.2069], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0733, 0.0686, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 12:50:18,874 INFO [optim.py:369] (1/2) 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 12:50:21,850 INFO [train.py:968] (1/2) Epoch 14, batch 37750, giga_loss[loss=0.2729, simple_loss=0.3489, pruned_loss=0.09843, over 28777.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3503, pruned_loss=0.1054, over 5681892.67 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1223, over 5667235.85 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3458, pruned_loss=0.1024, over 5676441.42 frames. ], batch size: 119, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:51:08,586 INFO [train.py:968] (1/2) Epoch 14, batch 37800, giga_loss[loss=0.2796, simple_loss=0.3558, pruned_loss=0.1017, over 28992.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3555, pruned_loss=0.1075, over 5680736.05 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1222, over 5668155.67 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.352, pruned_loss=0.1052, over 5675761.27 frames. ], batch size: 155, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:51:14,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:51:17,733 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:51:34,167 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:51:42,766 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 14, batch 37850, giga_loss[loss=0.315, simple_loss=0.3755, pruned_loss=0.1273, over 28867.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.361, pruned_loss=0.111, over 5678115.05 frames. ], libri_tot_loss[loss=0.307, simple_loss=0.3694, pruned_loss=0.1223, over 5667593.75 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3577, pruned_loss=0.1086, over 5675502.52 frames. ], batch size: 106, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:52:25,754 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:52:33,179 INFO [train.py:968] (1/2) Epoch 14, batch 37900, libri_loss[loss=0.3606, simple_loss=0.4121, pruned_loss=0.1545, over 29041.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3597, pruned_loss=0.1101, over 5670008.99 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3695, pruned_loss=0.1224, over 5662168.77 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3565, pruned_loss=0.1076, over 5672002.59 frames. ], batch size: 101, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:53:10,704 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 37950, giga_loss[loss=0.2762, simple_loss=0.3497, pruned_loss=0.1014, over 28991.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3551, pruned_loss=0.1061, over 5685282.71 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.3694, pruned_loss=0.1225, over 5668385.08 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3523, pruned_loss=0.1036, over 5681641.58 frames. ], batch size: 106, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:53:36,465 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-07 12:53:55,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:53:57,487 INFO [train.py:968] (1/2) Epoch 14, batch 38000, giga_loss[loss=0.2561, simple_loss=0.3319, pruned_loss=0.09015, over 28744.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3533, pruned_loss=0.1043, over 5687890.66 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3693, pruned_loss=0.1224, over 5671933.31 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3511, pruned_loss=0.1022, over 5682066.95 frames. ], batch size: 99, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:54:09,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4303, 1.6839, 1.3512, 1.6335], device='cuda:1'), covar=tensor([0.0793, 0.0298, 0.0316, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0112, 0.0113, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 12:54:17,220 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:54:38,379 INFO [optim.py:369] (1/2) 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 12:54:39,795 INFO [train.py:968] (1/2) Epoch 14, batch 38050, giga_loss[loss=0.2539, simple_loss=0.3375, pruned_loss=0.08513, over 28652.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3532, pruned_loss=0.1039, over 5693699.89 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3697, pruned_loss=0.1227, over 5677348.82 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3506, pruned_loss=0.1015, over 5684249.11 frames. ], batch size: 60, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:55:08,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0245, 3.2972, 2.1910, 1.0578], device='cuda:1'), covar=tensor([0.5573, 0.2538, 0.2709, 0.5235], device='cuda:1'), in_proj_covar=tensor([0.1616, 0.1529, 0.1523, 0.1318], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 12:55:08,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7888, 2.1810, 2.1034, 1.6280], device='cuda:1'), covar=tensor([0.1697, 0.2248, 0.1436, 0.1737], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0688, 0.0898, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 12:55:23,345 INFO [train.py:968] (1/2) Epoch 14, batch 38100, giga_loss[loss=0.3052, simple_loss=0.3816, pruned_loss=0.1144, over 28980.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3553, pruned_loss=0.1052, over 5697014.41 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3694, pruned_loss=0.1224, over 5683271.50 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3531, pruned_loss=0.103, over 5684511.31 frames. ], batch size: 213, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:55:57,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:55:59,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:56:03,674 INFO [optim.py:369] (1/2) 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 12:56:05,964 INFO [train.py:968] (1/2) Epoch 14, batch 38150, giga_loss[loss=0.2828, simple_loss=0.3596, pruned_loss=0.103, over 29093.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3578, pruned_loss=0.1071, over 5695777.68 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3703, pruned_loss=0.123, over 5687160.47 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3549, pruned_loss=0.1044, over 5682610.12 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:56:27,124 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:56:32,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6663, 1.7764, 1.5817, 1.3633], device='cuda:1'), covar=tensor([0.2172, 0.1851, 0.1795, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.1801, 0.1691, 0.1648, 0.1783], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 12:56:52,510 INFO [train.py:968] (1/2) Epoch 14, batch 38200, giga_loss[loss=0.3151, simple_loss=0.3779, pruned_loss=0.1261, over 27917.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3589, pruned_loss=0.1079, over 5701655.74 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3702, pruned_loss=0.1228, over 5689005.45 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3563, pruned_loss=0.1057, over 5689659.00 frames. ], batch size: 412, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:56:55,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:56:59,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3998, 1.2970, 4.1201, 3.1965], device='cuda:1'), covar=tensor([0.1471, 0.2446, 0.0423, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0689, 0.0604, 0.0881, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 12:56:59,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5212, 2.3586, 1.6877, 0.6324], device='cuda:1'), covar=tensor([0.4265, 0.2351, 0.3130, 0.4729], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1525, 0.1517, 0.1314], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 12:57:35,236 INFO [optim.py:369] (1/2) 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 12:57:36,600 INFO [train.py:968] (1/2) Epoch 14, batch 38250, giga_loss[loss=0.2999, simple_loss=0.3595, pruned_loss=0.1202, over 28633.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3605, pruned_loss=0.1095, over 5698143.51 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3706, pruned_loss=0.1228, over 5692742.35 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3578, pruned_loss=0.1073, over 5685731.39 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:57:51,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:58:18,850 INFO [train.py:968] (1/2) Epoch 14, batch 38300, libri_loss[loss=0.3297, simple_loss=0.4015, pruned_loss=0.129, over 29475.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3607, pruned_loss=0.1097, over 5706707.54 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5695901.11 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3581, pruned_loss=0.1076, over 5694024.72 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:58:36,809 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:58:55,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:58:57,562 INFO [optim.py:369] (1/2) 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 12:58:57,792 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 38350, giga_loss[loss=0.3027, simple_loss=0.3777, pruned_loss=0.1138, over 28881.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3609, pruned_loss=0.1092, over 5712300.72 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5702179.57 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3584, pruned_loss=0.1071, over 5696823.72 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 12:59:22,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:59:33,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8140, 1.9700, 1.8893, 1.7988], device='cuda:1'), covar=tensor([0.1625, 0.1866, 0.2016, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0728, 0.0681, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 12:59:41,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:59:43,874 INFO [train.py:968] (1/2) Epoch 14, batch 38400, giga_loss[loss=0.2647, simple_loss=0.3434, pruned_loss=0.09298, over 28674.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3591, pruned_loss=0.1065, over 5715280.29 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5702179.57 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3571, pruned_loss=0.1049, over 5703234.34 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 12:59:52,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 12:59:55,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:00:06,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-07 13:00:18,276 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:00:23,907 INFO [optim.py:369] (1/2) 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 13:00:26,587 INFO [train.py:968] (1/2) Epoch 14, batch 38450, giga_loss[loss=0.2704, simple_loss=0.3477, pruned_loss=0.09658, over 27905.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.359, pruned_loss=0.1057, over 5712976.03 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3712, pruned_loss=0.1231, over 5703903.17 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3573, pruned_loss=0.1042, over 5701990.15 frames. ], batch size: 412, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:00:54,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0810, 1.0463, 3.6628, 2.9630], device='cuda:1'), covar=tensor([0.1743, 0.2874, 0.0433, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0688, 0.0603, 0.0881, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:01:09,799 INFO [train.py:968] (1/2) Epoch 14, batch 38500, giga_loss[loss=0.2897, simple_loss=0.3611, pruned_loss=0.1092, over 29057.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3566, pruned_loss=0.1046, over 5711102.53 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3713, pruned_loss=0.1231, over 5704106.21 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.355, pruned_loss=0.1033, over 5702298.82 frames. ], batch size: 155, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:01:44,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=631839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:01:46,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=631842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:01:52,380 INFO [optim.py:369] (1/2) 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 13:01:53,637 INFO [train.py:968] (1/2) Epoch 14, batch 38550, giga_loss[loss=0.2852, simple_loss=0.3577, pruned_loss=0.1064, over 29089.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3543, pruned_loss=0.1032, over 5712916.95 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3711, pruned_loss=0.1229, over 5701859.64 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1022, over 5707834.75 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:02:11,879 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=631871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:02:34,275 INFO [train.py:968] (1/2) Epoch 14, batch 38600, giga_loss[loss=0.2488, simple_loss=0.3275, pruned_loss=0.08502, over 28496.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1023, over 5716193.14 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3713, pruned_loss=0.123, over 5703805.62 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3511, pruned_loss=0.1012, over 5710639.01 frames. ], batch size: 60, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:03:00,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 13:03:16,973 INFO [optim.py:369] (1/2) 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 13:03:17,664 INFO [train.py:968] (1/2) Epoch 14, batch 38650, giga_loss[loss=0.2811, simple_loss=0.3602, pruned_loss=0.101, over 28944.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3548, pruned_loss=0.1047, over 5716131.67 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1233, over 5705386.53 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3533, pruned_loss=0.1034, over 5710459.99 frames. ], batch size: 106, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:03:42,247 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=631981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:03:53,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=631996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:03:57,016 INFO [train.py:968] (1/2) Epoch 14, batch 38700, giga_loss[loss=0.2895, simple_loss=0.358, pruned_loss=0.1105, over 28944.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3557, pruned_loss=0.1053, over 5710993.76 frames. ], libri_tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1236, over 5702304.26 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3535, pruned_loss=0.1034, over 5709695.75 frames. ], batch size: 66, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:04:08,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-07 13:04:18,467 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-07 13:04:34,983 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 38750, giga_loss[loss=0.2571, simple_loss=0.341, pruned_loss=0.08657, over 29036.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3555, pruned_loss=0.1045, over 5714136.67 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 5707078.29 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3538, pruned_loss=0.1029, over 5708786.58 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:05:12,720 INFO [train.py:968] (1/2) Epoch 14, batch 38800, libri_loss[loss=0.2895, simple_loss=0.343, pruned_loss=0.118, over 29474.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.355, pruned_loss=0.1041, over 5718714.34 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1233, over 5712506.73 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1022, over 5709852.45 frames. ], batch size: 70, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:05:46,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:05:48,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:05:54,672 INFO [optim.py:369] (1/2) 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 13:05:55,355 INFO [train.py:968] (1/2) Epoch 14, batch 38850, giga_loss[loss=0.2861, simple_loss=0.3508, pruned_loss=0.1107, over 28728.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3537, pruned_loss=0.1037, over 5717670.77 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3718, pruned_loss=0.1234, over 5714672.93 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3519, pruned_loss=0.1018, over 5708767.96 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:05:57,671 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:06:13,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:06:33,264 INFO [train.py:968] (1/2) Epoch 14, batch 38900, giga_loss[loss=0.2685, simple_loss=0.3484, pruned_loss=0.09432, over 28609.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3516, pruned_loss=0.1027, over 5714809.83 frames. ], libri_tot_loss[loss=0.3093, simple_loss=0.3719, pruned_loss=0.1234, over 5718150.59 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3494, pruned_loss=0.1005, over 5704383.82 frames. ], batch size: 85, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:06:45,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6973, 0.9687, 2.9024, 2.6237], device='cuda:1'), covar=tensor([0.1757, 0.2652, 0.0538, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0602, 0.0876, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:07:15,712 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 38950, giga_loss[loss=0.3004, simple_loss=0.3664, pruned_loss=0.1172, over 28798.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3484, pruned_loss=0.1015, over 5717780.27 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5722991.88 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.347, pruned_loss=0.09973, over 5705044.79 frames. ], batch size: 199, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:07:31,455 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:07:53,205 INFO [train.py:968] (1/2) Epoch 14, batch 39000, giga_loss[loss=0.2565, simple_loss=0.3318, pruned_loss=0.09062, over 28989.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3464, pruned_loss=0.1004, over 5715996.46 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3709, pruned_loss=0.1228, over 5725939.70 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3448, pruned_loss=0.09859, over 5703099.14 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:07:53,206 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 13:08:01,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3489, 2.9717, 1.4513, 1.5042], device='cuda:1'), covar=tensor([0.0949, 0.0348, 0.0920, 0.1399], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0509, 0.0351, 0.0399], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0030, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 13:08:02,125 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 13:08:11,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2179, 1.4658, 1.4884, 1.2808], device='cuda:1'), covar=tensor([0.1616, 0.1554, 0.2176, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0729, 0.0687, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 13:08:30,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-07 13:08:43,074 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 39050, libri_loss[loss=0.2784, simple_loss=0.3389, pruned_loss=0.1089, over 29420.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3463, pruned_loss=0.1006, over 5697299.01 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3708, pruned_loss=0.1227, over 5712660.23 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3444, pruned_loss=0.09848, over 5698057.27 frames. ], batch size: 67, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:08:48,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:09:22,103 INFO [train.py:968] (1/2) Epoch 14, batch 39100, giga_loss[loss=0.2503, simple_loss=0.3229, pruned_loss=0.08882, over 28967.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3444, pruned_loss=0.09983, over 5702208.99 frames. ], libri_tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1226, over 5715767.97 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3424, pruned_loss=0.09789, over 5699834.62 frames. ], batch size: 145, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:09:59,015 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:10:05,241 INFO [train.py:968] (1/2) Epoch 14, batch 39150, giga_loss[loss=0.2544, simple_loss=0.3288, pruned_loss=0.09006, over 28863.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3424, pruned_loss=0.0992, over 5708262.43 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5715664.63 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.34, pruned_loss=0.09699, over 5706001.28 frames. ], batch size: 199, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:10:43,666 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 39200, giga_loss[loss=0.2252, simple_loss=0.3079, pruned_loss=0.07121, over 28949.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3399, pruned_loss=0.0981, over 5710220.23 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3711, pruned_loss=0.1229, over 5717553.54 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3376, pruned_loss=0.09603, over 5706753.66 frames. ], batch size: 174, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:10:47,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:11:09,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:11:15,036 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:11:30,777 INFO [optim.py:369] (1/2) 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 13:11:30,790 INFO [train.py:968] (1/2) Epoch 14, batch 39250, giga_loss[loss=0.241, simple_loss=0.3064, pruned_loss=0.08778, over 28455.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3374, pruned_loss=0.09651, over 5710649.07 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.371, pruned_loss=0.1227, over 5718522.77 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3356, pruned_loss=0.0949, over 5707093.21 frames. ], batch size: 71, lr: 2.27e-03, grad_scale: 8.0
+2023-03-07 13:12:07,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632593.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:12:08,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-07 13:12:11,842 INFO [train.py:968] (1/2) Epoch 14, batch 39300, giga_loss[loss=0.2304, simple_loss=0.3181, pruned_loss=0.07137, over 29010.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.34, pruned_loss=0.09778, over 5701013.71 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1229, over 5712476.50 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3377, pruned_loss=0.09582, over 5703780.21 frames. ], batch size: 155, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:12:48,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1240, 3.9361, 3.7261, 1.9713], device='cuda:1'), covar=tensor([0.0574, 0.0702, 0.0728, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.1092, 0.1013, 0.0874, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 13:12:52,947 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:12:55,470 INFO [train.py:968] (1/2) Epoch 14, batch 39350, giga_loss[loss=0.257, simple_loss=0.3431, pruned_loss=0.08542, over 28679.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3439, pruned_loss=0.09955, over 5690589.20 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.123, over 5706397.96 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3408, pruned_loss=0.09719, over 5698600.07 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:12:56,089 INFO [optim.py:369] (1/2) 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 13:13:15,563 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:13:18,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:13:30,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6059, 3.6563, 1.6303, 1.6175], device='cuda:1'), covar=tensor([0.0848, 0.0251, 0.0883, 0.1215], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0516, 0.0353, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 13:13:41,799 INFO [train.py:968] (1/2) Epoch 14, batch 39400, giga_loss[loss=0.2965, simple_loss=0.3686, pruned_loss=0.1122, over 28650.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3464, pruned_loss=0.1001, over 5690801.05 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 5707875.77 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3436, pruned_loss=0.09796, over 5695477.50 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:13:46,365 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:02,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2541, 1.5760, 1.4700, 1.4281], device='cuda:1'), covar=tensor([0.1709, 0.1610, 0.2002, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0728, 0.0686, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 13:14:04,276 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:08,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:25,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:26,302 INFO [train.py:968] (1/2) Epoch 14, batch 39450, giga_loss[loss=0.2732, simple_loss=0.3546, pruned_loss=0.09592, over 28989.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3481, pruned_loss=0.1008, over 5677349.70 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.123, over 5698213.92 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3454, pruned_loss=0.09868, over 5689560.90 frames. ], batch size: 164, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:14:26,934 INFO [optim.py:369] (1/2) 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 13:14:37,794 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:56,286 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=632787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:14:59,767 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:15:03,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:15:08,927 INFO [train.py:968] (1/2) Epoch 14, batch 39500, giga_loss[loss=0.2658, simple_loss=0.3475, pruned_loss=0.09209, over 28718.00 frames. ], tot_loss[loss=0.272, simple_loss=0.346, pruned_loss=0.09898, over 5688680.50 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3716, pruned_loss=0.123, over 5699836.40 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3435, pruned_loss=0.09681, over 5696494.76 frames. ], batch size: 284, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 13:15:19,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-07 13:15:23,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632816.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:15:26,690 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:15:52,823 INFO [train.py:968] (1/2) Epoch 14, batch 39550, giga_loss[loss=0.237, simple_loss=0.3168, pruned_loss=0.07855, over 28456.00 frames. ], tot_loss[loss=0.271, simple_loss=0.345, pruned_loss=0.09846, over 5691230.83 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1228, over 5702719.56 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3428, pruned_loss=0.09655, over 5694825.73 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 13:15:54,127 INFO [optim.py:369] (1/2) 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 13:16:22,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 13:16:37,114 INFO [train.py:968] (1/2) Epoch 14, batch 39600, giga_loss[loss=0.2869, simple_loss=0.3536, pruned_loss=0.1101, over 28331.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3458, pruned_loss=0.09896, over 5706242.47 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3719, pruned_loss=0.123, over 5706017.43 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3432, pruned_loss=0.09692, over 5706128.80 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:17:21,182 INFO [train.py:968] (1/2) Epoch 14, batch 39650, giga_loss[loss=0.2757, simple_loss=0.3515, pruned_loss=0.09988, over 28476.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3469, pruned_loss=0.09969, over 5710407.81 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1227, over 5707172.19 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3447, pruned_loss=0.09787, over 5709475.66 frames. ], batch size: 65, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:17:23,612 INFO [optim.py:369] (1/2) 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 13:17:29,636 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=632959.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:17:31,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=632962.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:17:37,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=632968.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:17:42,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3499, 1.5315, 1.4255, 1.1812], device='cuda:1'), covar=tensor([0.2041, 0.1885, 0.1473, 0.1941], device='cuda:1'), in_proj_covar=tensor([0.1806, 0.1699, 0.1663, 0.1786], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 13:17:51,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3281, 2.6950, 1.4098, 1.3863], device='cuda:1'), covar=tensor([0.0849, 0.0344, 0.0851, 0.1215], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0517, 0.0353, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 13:17:56,763 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=632991.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:18:04,587 INFO [train.py:968] (1/2) Epoch 14, batch 39700, giga_loss[loss=0.2745, simple_loss=0.3583, pruned_loss=0.09532, over 28803.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3526, pruned_loss=0.103, over 5704256.45 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3718, pruned_loss=0.123, over 5705312.21 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3501, pruned_loss=0.101, over 5705262.25 frames. ], batch size: 284, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:18:23,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3310, 4.1468, 3.9637, 2.0076], device='cuda:1'), covar=tensor([0.0616, 0.0779, 0.0855, 0.1794], device='cuda:1'), in_proj_covar=tensor([0.1095, 0.1014, 0.0877, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 13:18:38,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=633039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:18:45,488 INFO [train.py:968] (1/2) Epoch 14, batch 39750, giga_loss[loss=0.3002, simple_loss=0.3631, pruned_loss=0.1187, over 28705.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3546, pruned_loss=0.1044, over 5698592.02 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3717, pruned_loss=0.1228, over 5697475.95 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3526, pruned_loss=0.1026, over 5707263.35 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:18:46,836 INFO [optim.py:369] (1/2) 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 13:18:48,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-07 13:18:56,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5062, 1.5794, 1.7023, 1.2797], device='cuda:1'), covar=tensor([0.1677, 0.2341, 0.1406, 0.1629], device='cuda:1'), in_proj_covar=tensor([0.0855, 0.0688, 0.0897, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 13:18:59,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3640, 1.9583, 1.3943, 0.5353], device='cuda:1'), covar=tensor([0.4743, 0.2293, 0.3556, 0.5549], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1525, 0.1518, 0.1314], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 13:19:12,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9194, 1.1562, 1.0443, 0.8031], device='cuda:1'), covar=tensor([0.1885, 0.1992, 0.1193, 0.1829], device='cuda:1'), in_proj_covar=tensor([0.1813, 0.1709, 0.1667, 0.1787], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 13:19:25,500 INFO [train.py:968] (1/2) Epoch 14, batch 39800, giga_loss[loss=0.2819, simple_loss=0.3631, pruned_loss=0.1003, over 28893.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3552, pruned_loss=0.1045, over 5705665.37 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3716, pruned_loss=0.1229, over 5701168.27 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3533, pruned_loss=0.1026, over 5709384.37 frames. ], batch size: 227, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:19:25,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:19:29,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:19:37,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:19:38,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633114.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:19:44,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:19:51,861 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-07 13:20:00,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:20:02,665 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:20:07,044 INFO [train.py:968] (1/2) Epoch 14, batch 39850, giga_loss[loss=0.2408, simple_loss=0.3221, pruned_loss=0.0797, over 28451.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3568, pruned_loss=0.1051, over 5709124.90 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3719, pruned_loss=0.1228, over 5704790.14 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3546, pruned_loss=0.1032, over 5708662.02 frames. ], batch size: 71, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:20:08,263 INFO [optim.py:369] (1/2) 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 13:20:18,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:20:46,354 INFO [train.py:968] (1/2) Epoch 14, batch 39900, giga_loss[loss=0.2868, simple_loss=0.3651, pruned_loss=0.1043, over 28627.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3575, pruned_loss=0.1059, over 5705094.09 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3719, pruned_loss=0.1228, over 5704530.29 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3554, pruned_loss=0.104, over 5704991.12 frames. ], batch size: 307, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:21:19,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:21,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:22,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:24,575 INFO [train.py:968] (1/2) Epoch 14, batch 39950, giga_loss[loss=0.2575, simple_loss=0.3271, pruned_loss=0.09396, over 28779.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3555, pruned_loss=0.1048, over 5704674.49 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3718, pruned_loss=0.1227, over 5698741.46 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3536, pruned_loss=0.1031, over 5709883.52 frames. ], batch size: 78, lr: 2.27e-03, grad_scale: 2.0
+2023-03-07 13:21:24,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:27,362 INFO [optim.py:369] (1/2) 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 13:21:31,061 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-07 13:21:36,517 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:38,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:43,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:48,222 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:50,519 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:21:53,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:03,074 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 40000, giga_loss[loss=0.3249, simple_loss=0.3804, pruned_loss=0.1347, over 26573.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3525, pruned_loss=0.1035, over 5712426.38 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3721, pruned_loss=0.1228, over 5703115.50 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3504, pruned_loss=0.1016, over 5713015.88 frames. ], batch size: 555, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:22:09,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:12,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:16,250 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:35,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:22:46,503 INFO [train.py:968] (1/2) Epoch 14, batch 40050, giga_loss[loss=0.2302, simple_loss=0.3011, pruned_loss=0.0797, over 28648.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3481, pruned_loss=0.1008, over 5712861.93 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1224, over 5707292.80 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3463, pruned_loss=0.09923, over 5709723.29 frames. ], batch size: 92, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:22:49,292 INFO [optim.py:369] (1/2) 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 13:23:24,112 INFO [train.py:968] (1/2) Epoch 14, batch 40100, giga_loss[loss=0.2585, simple_loss=0.3348, pruned_loss=0.09104, over 29011.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3479, pruned_loss=0.0996, over 5719051.98 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3715, pruned_loss=0.1223, over 5710133.25 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.346, pruned_loss=0.09791, over 5714151.41 frames. ], batch size: 128, lr: 2.27e-03, grad_scale: 4.0
+2023-03-07 13:23:36,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=633414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:24:06,456 INFO [train.py:968] (1/2) Epoch 14, batch 40150, giga_loss[loss=0.303, simple_loss=0.3682, pruned_loss=0.1189, over 26673.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3492, pruned_loss=0.09898, over 5710506.98 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3712, pruned_loss=0.1221, over 5712903.32 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09745, over 5704111.69 frames. ], batch size: 555, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:24:06,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5387, 1.6748, 1.7806, 1.3011], device='cuda:1'), covar=tensor([0.1800, 0.2483, 0.1518, 0.1736], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0689, 0.0900, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 13:24:08,339 INFO [optim.py:369] (1/2) 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 13:24:45,897 INFO [train.py:968] (1/2) Epoch 14, batch 40200, libri_loss[loss=0.3287, simple_loss=0.3943, pruned_loss=0.1316, over 28629.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3487, pruned_loss=0.09919, over 5707226.77 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3714, pruned_loss=0.1223, over 5705547.61 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.347, pruned_loss=0.09746, over 5707847.92 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:25:24,409 INFO [train.py:968] (1/2) Epoch 14, batch 40250, giga_loss[loss=0.3325, simple_loss=0.3806, pruned_loss=0.1422, over 26722.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3487, pruned_loss=0.1007, over 5703331.78 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3712, pruned_loss=0.1221, over 5706111.81 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.347, pruned_loss=0.09908, over 5703465.73 frames. ], batch size: 555, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:25:26,783 INFO [optim.py:369] (1/2) 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 13:25:31,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=633557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:25:32,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=633560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:25:45,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9404, 1.5684, 5.1842, 3.8725], device='cuda:1'), covar=tensor([0.1459, 0.2469, 0.0347, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0689, 0.0604, 0.0881, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:25:53,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=633589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:26:03,321 INFO [train.py:968] (1/2) Epoch 14, batch 40300, giga_loss[loss=0.2679, simple_loss=0.3348, pruned_loss=0.1005, over 28934.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.348, pruned_loss=0.1018, over 5692939.28 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1225, over 5697869.60 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3457, pruned_loss=0.09968, over 5701302.76 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:26:41,841 INFO [train.py:968] (1/2) Epoch 14, batch 40350, giga_loss[loss=0.294, simple_loss=0.3587, pruned_loss=0.1146, over 27675.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3473, pruned_loss=0.1023, over 5706506.18 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1223, over 5705454.93 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3446, pruned_loss=0.09998, over 5706550.81 frames. ], batch size: 472, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:26:43,671 INFO [optim.py:369] (1/2) 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 13:27:11,783 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 13:27:15,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3797, 1.5986, 1.4832, 1.2564], device='cuda:1'), covar=tensor([0.2481, 0.2117, 0.1476, 0.2076], device='cuda:1'), in_proj_covar=tensor([0.1809, 0.1710, 0.1659, 0.1782], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 13:27:20,450 INFO [train.py:968] (1/2) Epoch 14, batch 40400, giga_loss[loss=0.2543, simple_loss=0.3125, pruned_loss=0.09804, over 28507.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3452, pruned_loss=0.1013, over 5715807.10 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3716, pruned_loss=0.1222, over 5708207.79 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3428, pruned_loss=0.09936, over 5713411.79 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:27:42,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2047, 2.4998, 1.2768, 1.3153], device='cuda:1'), covar=tensor([0.0956, 0.0386, 0.0928, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0521, 0.0354, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 13:28:00,354 INFO [train.py:968] (1/2) Epoch 14, batch 40450, giga_loss[loss=0.2303, simple_loss=0.2997, pruned_loss=0.08042, over 28614.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.343, pruned_loss=0.1002, over 5722679.68 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1225, over 5712291.71 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3403, pruned_loss=0.09794, over 5717312.16 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:28:02,209 INFO [optim.py:369] (1/2) 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 13:28:39,864 INFO [train.py:968] (1/2) Epoch 14, batch 40500, giga_loss[loss=0.2278, simple_loss=0.3071, pruned_loss=0.07419, over 28675.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3388, pruned_loss=0.09823, over 5713582.51 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3722, pruned_loss=0.1227, over 5705675.83 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.336, pruned_loss=0.09589, over 5714831.82 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:29:02,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1630, 1.2526, 3.4344, 3.0172], device='cuda:1'), covar=tensor([0.1576, 0.2698, 0.0421, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0687, 0.0600, 0.0877, 0.0800], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:29:12,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=633838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:29:20,347 INFO [train.py:968] (1/2) Epoch 14, batch 40550, giga_loss[loss=0.2194, simple_loss=0.2921, pruned_loss=0.07336, over 29033.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3344, pruned_loss=0.09577, over 5713379.30 frames. ], libri_tot_loss[loss=0.3094, simple_loss=0.3727, pruned_loss=0.123, over 5703847.58 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3312, pruned_loss=0.09327, over 5716461.68 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:29:22,873 INFO [optim.py:369] (1/2) 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 13:29:57,083 INFO [train.py:968] (1/2) Epoch 14, batch 40600, giga_loss[loss=0.2965, simple_loss=0.3656, pruned_loss=0.1137, over 28895.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3349, pruned_loss=0.09595, over 5718413.64 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5710841.95 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3312, pruned_loss=0.09289, over 5714890.27 frames. ], batch size: 227, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:30:02,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5036, 1.6147, 1.3456, 1.6758], device='cuda:1'), covar=tensor([0.2712, 0.2774, 0.2992, 0.2684], device='cuda:1'), in_proj_covar=tensor([0.1365, 0.1001, 0.1207, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 13:30:40,947 INFO [train.py:968] (1/2) Epoch 14, batch 40650, giga_loss[loss=0.3216, simple_loss=0.3841, pruned_loss=0.1295, over 28295.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3376, pruned_loss=0.09659, over 5707339.64 frames. ], libri_tot_loss[loss=0.3091, simple_loss=0.3723, pruned_loss=0.1229, over 5703720.17 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3339, pruned_loss=0.09361, over 5711650.60 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:30:43,723 INFO [optim.py:369] (1/2) 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 13:30:45,828 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 13:31:22,198 INFO [train.py:968] (1/2) Epoch 14, batch 40700, giga_loss[loss=0.2933, simple_loss=0.3672, pruned_loss=0.1097, over 29083.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3405, pruned_loss=0.09751, over 5704771.26 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3721, pruned_loss=0.1229, over 5705602.91 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3375, pruned_loss=0.095, over 5706397.42 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:31:33,280 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:31:44,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3867, 3.3236, 1.5058, 1.5102], device='cuda:1'), covar=tensor([0.0949, 0.0318, 0.0917, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0520, 0.0354, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 13:31:45,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3299, 4.1359, 3.9061, 1.9058], device='cuda:1'), covar=tensor([0.0532, 0.0666, 0.0656, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.1102, 0.1018, 0.0880, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 13:32:01,977 INFO [train.py:968] (1/2) Epoch 14, batch 40750, giga_loss[loss=0.2571, simple_loss=0.3352, pruned_loss=0.08952, over 28614.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3436, pruned_loss=0.09851, over 5713995.79 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1225, over 5708914.34 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3412, pruned_loss=0.0964, over 5712481.09 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:32:04,730 INFO [optim.py:369] (1/2) 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 13:32:24,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0222, 3.8347, 3.6408, 1.8644], device='cuda:1'), covar=tensor([0.0598, 0.0729, 0.0692, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1102, 0.1019, 0.0880, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 13:32:45,975 INFO [train.py:968] (1/2) Epoch 14, batch 40800, giga_loss[loss=0.3021, simple_loss=0.375, pruned_loss=0.1146, over 27888.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3467, pruned_loss=0.1002, over 5715241.88 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3716, pruned_loss=0.1225, over 5709897.77 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3446, pruned_loss=0.09838, over 5713132.17 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:33:05,477 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:33:18,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-07 13:33:28,903 INFO [train.py:968] (1/2) Epoch 14, batch 40850, giga_loss[loss=0.3169, simple_loss=0.3761, pruned_loss=0.1289, over 28710.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3494, pruned_loss=0.1023, over 5700893.73 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1227, over 5701241.06 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3472, pruned_loss=0.1004, over 5707718.07 frames. ], batch size: 99, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:33:31,524 INFO [optim.py:369] (1/2) 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 13:33:34,118 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:34:20,229 INFO [train.py:968] (1/2) Epoch 14, batch 40900, giga_loss[loss=0.36, simple_loss=0.4193, pruned_loss=0.1503, over 28184.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3572, pruned_loss=0.109, over 5683887.93 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5702386.72 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3549, pruned_loss=0.1071, over 5688310.60 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:34:31,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:35:02,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0184, 1.3420, 3.3129, 2.8030], device='cuda:1'), covar=tensor([0.1657, 0.2386, 0.0505, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0692, 0.0602, 0.0885, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:35:04,059 INFO [train.py:968] (1/2) Epoch 14, batch 40950, libri_loss[loss=0.3441, simple_loss=0.4074, pruned_loss=0.1404, over 29645.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3643, pruned_loss=0.1146, over 5676561.96 frames. ], libri_tot_loss[loss=0.3092, simple_loss=0.3724, pruned_loss=0.123, over 5697165.89 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.362, pruned_loss=0.1127, over 5683494.84 frames. ], batch size: 91, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:35:08,308 INFO [optim.py:369] (1/2) 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 13:35:50,668 INFO [train.py:968] (1/2) Epoch 14, batch 41000, giga_loss[loss=0.3158, simple_loss=0.3765, pruned_loss=0.1275, over 28626.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3705, pruned_loss=0.1194, over 5665516.13 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3722, pruned_loss=0.1228, over 5697870.46 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3687, pruned_loss=0.1179, over 5669975.79 frames. ], batch size: 78, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:35:53,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 13:36:03,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5884, 1.6776, 1.5537, 1.5151], device='cuda:1'), covar=tensor([0.1423, 0.1685, 0.2040, 0.1602], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0732, 0.0689, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 13:36:21,309 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:36:30,441 INFO [train.py:968] (1/2) Epoch 14, batch 41050, giga_loss[loss=0.3338, simple_loss=0.3966, pruned_loss=0.1355, over 28921.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3768, pruned_loss=0.1248, over 5671729.77 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1225, over 5703132.20 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3753, pruned_loss=0.1238, over 5669459.81 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:36:36,073 INFO [optim.py:369] (1/2) 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 13:36:36,948 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:36:37,862 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:36:39,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:37:05,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8999, 3.7473, 3.5877, 1.7140], device='cuda:1'), covar=tensor([0.0640, 0.0761, 0.0742, 0.2259], device='cuda:1'), in_proj_covar=tensor([0.1111, 0.1028, 0.0885, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 13:37:05,136 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:37:05,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:37:15,619 INFO [train.py:968] (1/2) Epoch 14, batch 41100, giga_loss[loss=0.3184, simple_loss=0.3845, pruned_loss=0.1261, over 28488.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3828, pruned_loss=0.1299, over 5664778.81 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1227, over 5697181.95 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3817, pruned_loss=0.1292, over 5667293.07 frames. ], batch size: 71, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:38:08,787 INFO [train.py:968] (1/2) Epoch 14, batch 41150, giga_loss[loss=0.3124, simple_loss=0.3779, pruned_loss=0.1235, over 28859.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3846, pruned_loss=0.1322, over 5656554.55 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3718, pruned_loss=0.1224, over 5697342.61 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3844, pruned_loss=0.132, over 5658012.01 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:38:14,785 INFO [optim.py:369] (1/2) 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 13:38:54,267 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:38:58,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:00,488 INFO [train.py:968] (1/2) Epoch 14, batch 41200, giga_loss[loss=0.3093, simple_loss=0.3729, pruned_loss=0.1229, over 29062.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3872, pruned_loss=0.1356, over 5642569.76 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3721, pruned_loss=0.1225, over 5702242.21 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3871, pruned_loss=0.1355, over 5638300.18 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:39:35,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:35,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:37,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:39:50,284 INFO [train.py:968] (1/2) Epoch 14, batch 41250, giga_loss[loss=0.3275, simple_loss=0.3852, pruned_loss=0.1349, over 29048.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3904, pruned_loss=0.139, over 5631939.00 frames. ], libri_tot_loss[loss=0.3086, simple_loss=0.3722, pruned_loss=0.1225, over 5705665.22 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3908, pruned_loss=0.1395, over 5623257.59 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:39:56,153 INFO [optim.py:369] (1/2) 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 13:40:05,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:40:24,290 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 13:40:41,684 INFO [train.py:968] (1/2) Epoch 14, batch 41300, giga_loss[loss=0.3102, simple_loss=0.3847, pruned_loss=0.1179, over 28550.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3946, pruned_loss=0.1431, over 5633656.82 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.122, over 5710752.99 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3959, pruned_loss=0.1443, over 5620465.32 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:40:49,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3489, 1.6539, 1.3758, 1.6387], device='cuda:1'), covar=tensor([0.0768, 0.0310, 0.0312, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 13:40:51,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:20,713 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:23,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:28,064 INFO [train.py:968] (1/2) Epoch 14, batch 41350, giga_loss[loss=0.453, simple_loss=0.4646, pruned_loss=0.2207, over 26645.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3967, pruned_loss=0.1449, over 5634026.78 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.372, pruned_loss=0.1225, over 5705555.09 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3982, pruned_loss=0.1462, over 5626051.62 frames. ], batch size: 555, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:41:32,788 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-07 13:41:38,677 INFO [optim.py:369] (1/2) 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 13:41:48,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1445, 4.9558, 4.7654, 2.3057], device='cuda:1'), covar=tensor([0.0492, 0.0631, 0.0714, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1118, 0.1036, 0.0895, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 13:41:53,700 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:55,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:41:58,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:42:18,541 INFO [train.py:968] (1/2) Epoch 14, batch 41400, giga_loss[loss=0.3477, simple_loss=0.3991, pruned_loss=0.1481, over 28610.00 frames. ], tot_loss[loss=0.3417, simple_loss=0.395, pruned_loss=0.1442, over 5641704.02 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1228, over 5708712.55 frames. ], giga_tot_loss[loss=0.3443, simple_loss=0.397, pruned_loss=0.1458, over 5629928.65 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:42:26,011 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:42:51,386 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 41450, giga_loss[loss=0.3064, simple_loss=0.3756, pruned_loss=0.1186, over 29132.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3934, pruned_loss=0.1434, over 5634466.28 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3723, pruned_loss=0.1228, over 5710630.55 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3951, pruned_loss=0.1449, over 5622800.99 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:43:16,162 INFO [optim.py:369] (1/2) 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 13:43:16,406 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:43:42,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2445, 4.0232, 3.8420, 1.8073], device='cuda:1'), covar=tensor([0.0743, 0.0955, 0.1067, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1122, 0.1039, 0.0899, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 13:43:54,846 INFO [train.py:968] (1/2) Epoch 14, batch 41500, giga_loss[loss=0.2962, simple_loss=0.3708, pruned_loss=0.1108, over 29062.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3922, pruned_loss=0.1417, over 5618602.14 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1224, over 5698215.95 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3949, pruned_loss=0.1441, over 5616944.11 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:44:42,391 INFO [train.py:968] (1/2) Epoch 14, batch 41550, giga_loss[loss=0.3321, simple_loss=0.3953, pruned_loss=0.1345, over 28624.00 frames. ], tot_loss[loss=0.3388, simple_loss=0.3932, pruned_loss=0.1422, over 5607339.43 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3715, pruned_loss=0.1222, over 5692740.94 frames. ], giga_tot_loss[loss=0.3429, simple_loss=0.3962, pruned_loss=0.1448, over 5609227.69 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 13:44:48,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:44:50,036 INFO [optim.py:369] (1/2) 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 13:44:50,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:44:53,386 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=634861.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:44:59,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:10,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=634875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:14,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=634878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:22,631 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:45:33,172 INFO [train.py:968] (1/2) Epoch 14, batch 41600, giga_loss[loss=0.3373, simple_loss=0.4085, pruned_loss=0.1331, over 28861.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3945, pruned_loss=0.1432, over 5578716.18 frames. ], libri_tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5678086.88 frames. ], giga_tot_loss[loss=0.3434, simple_loss=0.3966, pruned_loss=0.1451, over 5591335.57 frames. ], batch size: 199, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:45:42,454 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=634907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:46:13,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5991, 4.4317, 4.2192, 2.0833], device='cuda:1'), covar=tensor([0.0534, 0.0681, 0.0757, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.1122, 0.1040, 0.0898, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 13:46:22,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6838, 1.8306, 1.4984, 2.0719], device='cuda:1'), covar=tensor([0.2798, 0.2710, 0.3137, 0.2226], device='cuda:1'), in_proj_covar=tensor([0.1366, 0.1004, 0.1210, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 13:46:22,822 INFO [train.py:968] (1/2) Epoch 14, batch 41650, giga_loss[loss=0.2889, simple_loss=0.3699, pruned_loss=0.1039, over 29007.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1391, over 5600143.39 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.372, pruned_loss=0.1228, over 5680912.89 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3933, pruned_loss=0.1413, over 5605133.61 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:46:26,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 13:46:29,839 INFO [optim.py:369] (1/2) 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 13:46:48,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 13:46:55,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=634983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:09,704 INFO [train.py:968] (1/2) Epoch 14, batch 41700, giga_loss[loss=0.3157, simple_loss=0.3811, pruned_loss=0.1252, over 28550.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3893, pruned_loss=0.1364, over 5617432.97 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.372, pruned_loss=0.1228, over 5681610.61 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3916, pruned_loss=0.1383, over 5619227.58 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:47:21,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:23,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:47,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8201, 4.6432, 4.4464, 2.1622], device='cuda:1'), covar=tensor([0.0508, 0.0665, 0.0771, 0.1893], device='cuda:1'), in_proj_covar=tensor([0.1125, 0.1047, 0.0903, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 13:47:49,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:47:55,962 INFO [train.py:968] (1/2) Epoch 14, batch 41750, giga_loss[loss=0.2845, simple_loss=0.3537, pruned_loss=0.1076, over 28407.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3849, pruned_loss=0.1335, over 5616152.82 frames. ], libri_tot_loss[loss=0.3084, simple_loss=0.3712, pruned_loss=0.1228, over 5681773.67 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3881, pruned_loss=0.1356, over 5614198.83 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:47:56,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:48:01,372 INFO [optim.py:369] (1/2) 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 13:48:23,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4490, 1.6323, 1.5897, 1.4506], device='cuda:1'), covar=tensor([0.1540, 0.1845, 0.1959, 0.1847], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0731, 0.0684, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 13:48:44,731 INFO [train.py:968] (1/2) Epoch 14, batch 41800, giga_loss[loss=0.3226, simple_loss=0.3854, pruned_loss=0.1299, over 28383.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3817, pruned_loss=0.1304, over 5618316.29 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3712, pruned_loss=0.1227, over 5685413.11 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3844, pruned_loss=0.1322, over 5612536.30 frames. ], batch size: 71, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:49:06,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1572, 1.3050, 3.5168, 3.0860], device='cuda:1'), covar=tensor([0.1597, 0.2497, 0.0444, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0699, 0.0610, 0.0895, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:49:11,231 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:49:13,339 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:49:15,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=635132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:49:32,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 13:49:33,909 INFO [train.py:968] (1/2) Epoch 14, batch 41850, giga_loss[loss=0.3198, simple_loss=0.3793, pruned_loss=0.1302, over 28263.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3807, pruned_loss=0.1295, over 5633378.68 frames. ], libri_tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 5677427.43 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3826, pruned_loss=0.1307, over 5634406.58 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:49:41,559 INFO [optim.py:369] (1/2) 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 13:49:42,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:50:00,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6758, 1.8329, 1.9803, 1.5098], device='cuda:1'), covar=tensor([0.1820, 0.2313, 0.1402, 0.1671], device='cuda:1'), in_proj_covar=tensor([0.0849, 0.0688, 0.0891, 0.0794], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0014, 0.0013], device='cuda:1')
+2023-03-07 13:50:17,642 INFO [train.py:968] (1/2) Epoch 14, batch 41900, giga_loss[loss=0.3199, simple_loss=0.3815, pruned_loss=0.1291, over 28971.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3806, pruned_loss=0.1296, over 5625918.52 frames. ], libri_tot_loss[loss=0.3097, simple_loss=0.3723, pruned_loss=0.1236, over 5662643.72 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3817, pruned_loss=0.1303, over 5639646.87 frames. ], batch size: 213, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:50:28,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 13:50:38,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8230, 3.6234, 3.4299, 1.7215], device='cuda:1'), covar=tensor([0.0705, 0.0857, 0.0819, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.1130, 0.1050, 0.0907, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 13:50:40,172 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:50:42,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1575, 1.2487, 1.1019, 0.9112], device='cuda:1'), covar=tensor([0.0923, 0.0556, 0.1078, 0.1086], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0443, 0.0507, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:50:56,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=635236.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:51:07,871 INFO [train.py:968] (1/2) Epoch 14, batch 41950, giga_loss[loss=0.2762, simple_loss=0.3484, pruned_loss=0.102, over 28599.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3787, pruned_loss=0.1281, over 5631628.81 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3714, pruned_loss=0.123, over 5668512.92 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3806, pruned_loss=0.1293, over 5636452.47 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:51:14,588 INFO [optim.py:369] (1/2) 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 13:51:35,560 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:51:38,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:51:59,265 INFO [train.py:968] (1/2) Epoch 14, batch 42000, giga_loss[loss=0.2817, simple_loss=0.3635, pruned_loss=0.09995, over 28932.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.377, pruned_loss=0.1256, over 5632700.46 frames. ], libri_tot_loss[loss=0.3087, simple_loss=0.3714, pruned_loss=0.123, over 5674155.69 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3786, pruned_loss=0.1266, over 5630875.52 frames. ], batch size: 199, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:51:59,265 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 13:52:08,189 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 13:52:15,446 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:52:55,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 13:52:55,744 INFO [train.py:968] (1/2) Epoch 14, batch 42050, libri_loss[loss=0.4039, simple_loss=0.4483, pruned_loss=0.1797, over 29540.00 frames. ], tot_loss[loss=0.313, simple_loss=0.378, pruned_loss=0.124, over 5649217.85 frames. ], libri_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.1229, over 5679396.89 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3799, pruned_loss=0.125, over 5641224.49 frames. ], batch size: 82, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:53:03,358 INFO [optim.py:369] (1/2) 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 13:53:23,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635379.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:53:26,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635382.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 13:53:30,186 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.08 vs. limit=5.0
+2023-03-07 13:53:40,404 INFO [train.py:968] (1/2) Epoch 14, batch 42100, libri_loss[loss=0.3161, simple_loss=0.378, pruned_loss=0.1271, over 29079.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3793, pruned_loss=0.1245, over 5662992.75 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5684737.29 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3815, pruned_loss=0.1256, over 5650921.44 frames. ], batch size: 101, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:53:51,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635411.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 13:54:01,790 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-07 13:54:06,771 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 42150, libri_loss[loss=0.3196, simple_loss=0.3825, pruned_loss=0.1283, over 29539.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3805, pruned_loss=0.126, over 5649584.21 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3705, pruned_loss=0.1226, over 5672019.97 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3826, pruned_loss=0.1269, over 5650399.14 frames. ], batch size: 83, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:54:34,987 INFO [optim.py:369] (1/2) 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 13:54:39,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.58 vs. limit=5.0
+2023-03-07 13:55:09,233 INFO [train.py:968] (1/2) Epoch 14, batch 42200, libri_loss[loss=0.3164, simple_loss=0.3764, pruned_loss=0.1281, over 29256.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3785, pruned_loss=0.125, over 5650728.58 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3704, pruned_loss=0.1226, over 5668621.22 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3808, pruned_loss=0.1259, over 5653173.23 frames. ], batch size: 94, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:55:46,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 13:55:54,047 INFO [train.py:968] (1/2) Epoch 14, batch 42250, giga_loss[loss=0.269, simple_loss=0.3431, pruned_loss=0.09745, over 29099.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3765, pruned_loss=0.1249, over 5659447.55 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.37, pruned_loss=0.1223, over 5669475.43 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3789, pruned_loss=0.126, over 5660699.87 frames. ], batch size: 128, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:56:01,001 INFO [optim.py:369] (1/2) 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 13:56:13,044 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:56:16,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:56:38,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=635596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:56:42,644 INFO [train.py:968] (1/2) Epoch 14, batch 42300, giga_loss[loss=0.3322, simple_loss=0.3855, pruned_loss=0.1395, over 27888.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.377, pruned_loss=0.1264, over 5660462.56 frames. ], libri_tot_loss[loss=0.3073, simple_loss=0.3701, pruned_loss=0.1223, over 5672566.47 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3788, pruned_loss=0.1273, over 5658521.16 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:56:46,554 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:56:48,002 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-07 13:57:32,361 INFO [train.py:968] (1/2) Epoch 14, batch 42350, giga_loss[loss=0.3179, simple_loss=0.3832, pruned_loss=0.1263, over 28629.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3765, pruned_loss=0.1248, over 5666029.19 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3702, pruned_loss=0.1224, over 5675993.57 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.378, pruned_loss=0.1255, over 5661148.99 frames. ], batch size: 262, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:57:32,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6632, 1.8089, 1.3247, 1.4130], device='cuda:1'), covar=tensor([0.0873, 0.0633, 0.1032, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0442, 0.0504, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 13:57:39,843 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 14, batch 42400, giga_loss[loss=0.2847, simple_loss=0.3567, pruned_loss=0.1063, over 28925.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3759, pruned_loss=0.1235, over 5675485.39 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3702, pruned_loss=0.1224, over 5677600.03 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3772, pruned_loss=0.124, over 5670182.56 frames. ], batch size: 227, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 13:58:23,376 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:58:57,083 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=635739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:59:00,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=635742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:59:06,540 INFO [train.py:968] (1/2) Epoch 14, batch 42450, giga_loss[loss=0.3249, simple_loss=0.3837, pruned_loss=0.133, over 27923.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3757, pruned_loss=0.123, over 5664607.71 frames. ], libri_tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5671435.14 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3771, pruned_loss=0.1236, over 5665143.14 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 13:59:14,820 INFO [optim.py:369] (1/2) 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 13:59:25,771 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=635771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 13:59:49,905 INFO [train.py:968] (1/2) Epoch 14, batch 42500, giga_loss[loss=0.326, simple_loss=0.3858, pruned_loss=0.1331, over 28973.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.375, pruned_loss=0.1233, over 5655664.55 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3703, pruned_loss=0.1225, over 5654265.88 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3761, pruned_loss=0.1236, over 5670260.01 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:00:17,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2579, 1.4515, 1.4063, 1.4027], device='cuda:1'), covar=tensor([0.0814, 0.0323, 0.0281, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 14:00:33,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3577, 1.5595, 1.4939, 1.1937], device='cuda:1'), covar=tensor([0.2386, 0.2080, 0.1432, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.1816, 0.1710, 0.1669, 0.1789], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 14:00:37,320 INFO [train.py:968] (1/2) Epoch 14, batch 42550, giga_loss[loss=0.3162, simple_loss=0.3636, pruned_loss=0.1344, over 23368.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3739, pruned_loss=0.1234, over 5654338.02 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.1229, over 5659294.07 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3745, pruned_loss=0.1233, over 5661357.20 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:00:40,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=635854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:00:42,946 INFO [optim.py:369] (1/2) 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 14:00:47,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4299, 1.7443, 1.4966, 1.6065], device='cuda:1'), covar=tensor([0.0739, 0.0299, 0.0291, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 14:01:22,308 INFO [train.py:968] (1/2) Epoch 14, batch 42600, giga_loss[loss=0.28, simple_loss=0.3418, pruned_loss=0.1091, over 28606.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3735, pruned_loss=0.1239, over 5661431.39 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3704, pruned_loss=0.1226, over 5660119.56 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.1241, over 5666151.29 frames. ], batch size: 71, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:02:10,109 INFO [train.py:968] (1/2) Epoch 14, batch 42650, giga_loss[loss=0.288, simple_loss=0.3574, pruned_loss=0.1092, over 28874.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3728, pruned_loss=0.1236, over 5672904.14 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5664790.69 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3733, pruned_loss=0.1238, over 5672727.84 frames. ], batch size: 227, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:02:16,905 INFO [optim.py:369] (1/2) 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 14:02:57,893 INFO [train.py:968] (1/2) Epoch 14, batch 42700, giga_loss[loss=0.3005, simple_loss=0.3679, pruned_loss=0.1166, over 28067.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1231, over 5655791.83 frames. ], libri_tot_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1226, over 5649324.79 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1233, over 5669147.42 frames. ], batch size: 77, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:03:42,124 INFO [train.py:968] (1/2) Epoch 14, batch 42750, giga_loss[loss=0.2805, simple_loss=0.3535, pruned_loss=0.1038, over 28934.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3709, pruned_loss=0.1235, over 5645943.32 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3705, pruned_loss=0.1224, over 5648449.94 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3714, pruned_loss=0.1238, over 5657391.41 frames. ], batch size: 164, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:03:51,468 INFO [optim.py:369] (1/2) 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 14:04:09,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:04:27,038 INFO [train.py:968] (1/2) Epoch 14, batch 42800, giga_loss[loss=0.3042, simple_loss=0.3869, pruned_loss=0.1108, over 28812.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5653967.36 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5652961.65 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3716, pruned_loss=0.1234, over 5659133.61 frames. ], batch size: 199, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:05:10,880 INFO [train.py:968] (1/2) Epoch 14, batch 42850, giga_loss[loss=0.269, simple_loss=0.351, pruned_loss=0.09355, over 28983.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.122, over 5652767.03 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3702, pruned_loss=0.1224, over 5641140.10 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3717, pruned_loss=0.1223, over 5667944.29 frames. ], batch size: 112, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:05:20,283 INFO [optim.py:369] (1/2) 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 14:05:28,166 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-07 14:05:39,985 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636180.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:05:57,275 INFO [train.py:968] (1/2) Epoch 14, batch 42900, giga_loss[loss=0.3087, simple_loss=0.3816, pruned_loss=0.1179, over 28928.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3725, pruned_loss=0.1224, over 5659507.82 frames. ], libri_tot_loss[loss=0.3076, simple_loss=0.3703, pruned_loss=0.1224, over 5642691.40 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3728, pruned_loss=0.1226, over 5670098.20 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:06:15,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:06:19,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:06:26,417 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:06:46,779 INFO [train.py:968] (1/2) Epoch 14, batch 42950, giga_loss[loss=0.3328, simple_loss=0.3863, pruned_loss=0.1396, over 27867.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3716, pruned_loss=0.1215, over 5667954.35 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.122, over 5646493.64 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3724, pruned_loss=0.122, over 5673129.24 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:06:51,312 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:07:37,297 INFO [train.py:968] (1/2) Epoch 14, batch 43000, giga_loss[loss=0.2946, simple_loss=0.3661, pruned_loss=0.1115, over 28498.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3747, pruned_loss=0.1242, over 5672206.44 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.122, over 5647480.17 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3753, pruned_loss=0.1246, over 5675555.88 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:07:47,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7209, 5.5291, 5.2580, 2.8616], device='cuda:1'), covar=tensor([0.0396, 0.0557, 0.0694, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.1131, 0.1052, 0.0909, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 14:08:27,849 INFO [train.py:968] (1/2) Epoch 14, batch 43050, giga_loss[loss=0.3101, simple_loss=0.3729, pruned_loss=0.1237, over 28674.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3753, pruned_loss=0.1259, over 5682038.03 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3694, pruned_loss=0.1218, over 5651367.13 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3761, pruned_loss=0.1265, over 5681632.61 frames. ], batch size: 119, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:08:39,738 INFO [optim.py:369] (1/2) 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 14:08:48,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2827, 1.3176, 3.6673, 3.2464], device='cuda:1'), covar=tensor([0.1523, 0.2467, 0.0459, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0699, 0.0609, 0.0896, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:08:49,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3655, 1.5143, 1.4161, 1.3012], device='cuda:1'), covar=tensor([0.2332, 0.2099, 0.1516, 0.1772], device='cuda:1'), in_proj_covar=tensor([0.1833, 0.1733, 0.1695, 0.1812], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 14:08:52,966 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:09:18,450 INFO [train.py:968] (1/2) Epoch 14, batch 43100, libri_loss[loss=0.3122, simple_loss=0.3825, pruned_loss=0.121, over 29360.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.376, pruned_loss=0.1276, over 5678677.91 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3695, pruned_loss=0.1219, over 5655986.14 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3768, pruned_loss=0.1281, over 5674560.18 frames. ], batch size: 92, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:09:21,879 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=636404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:09:26,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0224, 1.3072, 1.0694, 0.1869], device='cuda:1'), covar=tensor([0.2552, 0.2200, 0.3013, 0.4563], device='cuda:1'), in_proj_covar=tensor([0.1641, 0.1566, 0.1537, 0.1340], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 14:09:29,388 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 43150, giga_loss[loss=0.324, simple_loss=0.3848, pruned_loss=0.1315, over 28936.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3801, pruned_loss=0.1315, over 5663253.09 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3697, pruned_loss=0.1219, over 5658828.75 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3808, pruned_loss=0.132, over 5657963.33 frames. ], batch size: 227, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:10:15,520 INFO [optim.py:369] (1/2) 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 14:10:30,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636477.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:10:47,663 INFO [train.py:968] (1/2) Epoch 14, batch 43200, libri_loss[loss=0.2665, simple_loss=0.3319, pruned_loss=0.1006, over 29559.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3775, pruned_loss=0.1298, over 5670661.03 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3692, pruned_loss=0.1218, over 5666205.01 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3787, pruned_loss=0.1307, over 5659402.85 frames. ], batch size: 75, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:10:51,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 14:11:07,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5779, 1.7729, 1.4393, 1.9124], device='cuda:1'), covar=tensor([0.2413, 0.2405, 0.2532, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.1370, 0.1008, 0.1214, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 14:11:10,807 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:11:10,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3728, 2.3865, 1.6690, 2.0274], device='cuda:1'), covar=tensor([0.0835, 0.0692, 0.1035, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0448, 0.0508, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 14:11:32,593 INFO [train.py:968] (1/2) Epoch 14, batch 43250, giga_loss[loss=0.4503, simple_loss=0.4597, pruned_loss=0.2204, over 26443.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3757, pruned_loss=0.1278, over 5674585.27 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1213, over 5670062.26 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3775, pruned_loss=0.1291, over 5662232.56 frames. ], batch size: 555, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:11:36,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636555.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:11:39,793 INFO [optim.py:369] (1/2) 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 14:11:40,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7511, 1.8763, 1.3416, 1.3878], device='cuda:1'), covar=tensor([0.0917, 0.0645, 0.1076, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0449, 0.0508, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 14:12:15,851 INFO [train.py:968] (1/2) Epoch 14, batch 43300, giga_loss[loss=0.2858, simple_loss=0.3645, pruned_loss=0.1036, over 28985.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3754, pruned_loss=0.1261, over 5661847.72 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3691, pruned_loss=0.1217, over 5663390.48 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3764, pruned_loss=0.1268, over 5657838.20 frames. ], batch size: 164, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:12:22,540 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 14:13:00,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3153, 1.7146, 1.4402, 1.4161], device='cuda:1'), covar=tensor([0.1958, 0.1755, 0.2102, 0.1925], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0729, 0.0683, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 14:13:03,018 INFO [train.py:968] (1/2) Epoch 14, batch 43350, giga_loss[loss=0.3232, simple_loss=0.3814, pruned_loss=0.1326, over 28764.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3735, pruned_loss=0.125, over 5650943.50 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3693, pruned_loss=0.1218, over 5653648.40 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3742, pruned_loss=0.1255, over 5656113.71 frames. ], batch size: 119, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:13:12,302 INFO [optim.py:369] (1/2) 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 14:13:48,008 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636698.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:13:48,925 INFO [train.py:968] (1/2) Epoch 14, batch 43400, giga_loss[loss=0.2877, simple_loss=0.3519, pruned_loss=0.1118, over 28453.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.372, pruned_loss=0.1242, over 5655836.70 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.1219, over 5649579.91 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3725, pruned_loss=0.1246, over 5663560.58 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:13:49,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636701.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:13:54,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4320, 2.2277, 2.2254, 2.0830], device='cuda:1'), covar=tensor([0.1721, 0.2566, 0.2098, 0.2180], device='cuda:1'), in_proj_covar=tensor([0.0443, 0.0730, 0.0684, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 14:14:17,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=636730.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:14:25,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5288, 1.6684, 1.5561, 1.4684], device='cuda:1'), covar=tensor([0.2237, 0.1957, 0.1750, 0.1813], device='cuda:1'), in_proj_covar=tensor([0.1832, 0.1736, 0.1689, 0.1810], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 14:14:34,599 INFO [train.py:968] (1/2) Epoch 14, batch 43450, giga_loss[loss=0.2955, simple_loss=0.3647, pruned_loss=0.1131, over 28639.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3712, pruned_loss=0.1242, over 5661949.18 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3694, pruned_loss=0.122, over 5651312.70 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3716, pruned_loss=0.1244, over 5666678.10 frames. ], batch size: 262, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:14:36,850 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-07 14:14:45,306 INFO [optim.py:369] (1/2) 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 14:14:46,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5645, 1.6413, 1.7714, 1.3910], device='cuda:1'), covar=tensor([0.1601, 0.2302, 0.1307, 0.1545], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0693, 0.0897, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 14:14:51,554 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:15:08,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:15:20,332 INFO [train.py:968] (1/2) Epoch 14, batch 43500, giga_loss[loss=0.3167, simple_loss=0.3863, pruned_loss=0.1236, over 28224.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3741, pruned_loss=0.126, over 5656069.47 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.1219, over 5649286.74 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3746, pruned_loss=0.1264, over 5661823.31 frames. ], batch size: 368, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:15:35,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:16:05,864 INFO [train.py:968] (1/2) Epoch 14, batch 43550, giga_loss[loss=0.3188, simple_loss=0.3771, pruned_loss=0.1302, over 26616.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3777, pruned_loss=0.126, over 5658566.74 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3687, pruned_loss=0.1214, over 5652549.21 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3788, pruned_loss=0.1268, over 5660084.21 frames. ], batch size: 555, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:16:09,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=636852.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:16:19,031 INFO [optim.py:369] (1/2) 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 14:16:58,471 INFO [train.py:968] (1/2) Epoch 14, batch 43600, giga_loss[loss=0.2932, simple_loss=0.3698, pruned_loss=0.1083, over 28640.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3788, pruned_loss=0.1251, over 5663623.85 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.369, pruned_loss=0.1216, over 5653387.95 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3796, pruned_loss=0.1257, over 5663930.05 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:16:58,746 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:17:31,495 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:17:44,375 INFO [train.py:968] (1/2) Epoch 14, batch 43650, giga_loss[loss=0.3226, simple_loss=0.3962, pruned_loss=0.1245, over 28592.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3814, pruned_loss=0.1274, over 5665186.52 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3686, pruned_loss=0.1214, over 5659858.51 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3826, pruned_loss=0.1281, over 5659755.34 frames. ], batch size: 307, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:17:54,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=636959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:17:55,776 INFO [optim.py:369] (1/2) 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 14:17:56,579 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=636962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:18:30,958 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=636995.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:18:34,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=636998.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:18:35,052 INFO [train.py:968] (1/2) Epoch 14, batch 43700, giga_loss[loss=0.3312, simple_loss=0.3917, pruned_loss=0.1353, over 28610.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3833, pruned_loss=0.1293, over 5668041.87 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1216, over 5661147.21 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3842, pruned_loss=0.1298, over 5662544.33 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:18:43,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1795, 1.1134, 3.7572, 3.1808], device='cuda:1'), covar=tensor([0.1691, 0.2730, 0.0470, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0613, 0.0902, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:18:59,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637027.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:19:13,041 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:19:16,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:19:18,953 INFO [train.py:968] (1/2) Epoch 14, batch 43750, giga_loss[loss=0.354, simple_loss=0.3846, pruned_loss=0.1617, over 23715.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3821, pruned_loss=0.1294, over 5664856.84 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1216, over 5659904.33 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.383, pruned_loss=0.1299, over 5661409.49 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:19:30,957 INFO [optim.py:369] (1/2) 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 14:19:43,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2992, 3.0334, 1.4608, 1.4737], device='cuda:1'), covar=tensor([0.0968, 0.0332, 0.0853, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0531, 0.0357, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 14:19:44,280 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:20:08,591 INFO [train.py:968] (1/2) Epoch 14, batch 43800, giga_loss[loss=0.2993, simple_loss=0.3558, pruned_loss=0.1214, over 28759.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3795, pruned_loss=0.1285, over 5666678.11 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1214, over 5664587.70 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3806, pruned_loss=0.1291, over 5659819.48 frames. ], batch size: 99, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:20:47,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:20:52,952 INFO [train.py:968] (1/2) Epoch 14, batch 43850, giga_loss[loss=0.321, simple_loss=0.3822, pruned_loss=0.1299, over 28751.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3785, pruned_loss=0.1285, over 5673149.95 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.369, pruned_loss=0.1218, over 5666677.69 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3791, pruned_loss=0.1288, over 5665717.47 frames. ], batch size: 284, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:21:04,358 INFO [optim.py:369] (1/2) 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 14:21:30,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:21:41,994 INFO [train.py:968] (1/2) Epoch 14, batch 43900, giga_loss[loss=0.2724, simple_loss=0.3453, pruned_loss=0.09976, over 28920.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3762, pruned_loss=0.1274, over 5658754.66 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3693, pruned_loss=0.1221, over 5654380.12 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3766, pruned_loss=0.1275, over 5663724.04 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:22:27,546 INFO [train.py:968] (1/2) Epoch 14, batch 43950, giga_loss[loss=0.3588, simple_loss=0.3909, pruned_loss=0.1634, over 23515.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3768, pruned_loss=0.1278, over 5664266.63 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3692, pruned_loss=0.122, over 5660511.28 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3774, pruned_loss=0.1281, over 5662744.83 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:22:35,276 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:22:41,851 INFO [optim.py:369] (1/2) 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 14:23:03,452 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:05,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:14,191 INFO [train.py:968] (1/2) Epoch 14, batch 44000, giga_loss[loss=0.2968, simple_loss=0.3656, pruned_loss=0.114, over 28831.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.377, pruned_loss=0.1284, over 5661812.22 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.369, pruned_loss=0.122, over 5655954.45 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3779, pruned_loss=0.1289, over 5665001.19 frames. ], batch size: 186, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:23:26,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4827, 1.7458, 1.4180, 1.5618], device='cuda:1'), covar=tensor([0.2413, 0.2402, 0.2778, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.1375, 0.1014, 0.1219, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 14:23:28,586 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:45,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:45,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:48,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:23:58,990 INFO [train.py:968] (1/2) Epoch 14, batch 44050, giga_loss[loss=0.2982, simple_loss=0.3655, pruned_loss=0.1154, over 28854.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3757, pruned_loss=0.1279, over 5666471.82 frames. ], libri_tot_loss[loss=0.3068, simple_loss=0.3694, pruned_loss=0.1221, over 5658229.08 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3762, pruned_loss=0.1283, over 5667378.86 frames. ], batch size: 66, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:24:09,693 INFO [optim.py:369] (1/2) 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 14:24:12,814 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:24:31,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:24:40,962 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:24:41,345 INFO [train.py:968] (1/2) Epoch 14, batch 44100, giga_loss[loss=0.297, simple_loss=0.3689, pruned_loss=0.1125, over 28966.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3739, pruned_loss=0.1261, over 5674220.70 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.369, pruned_loss=0.1217, over 5664819.17 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3748, pruned_loss=0.1268, over 5668900.14 frames. ], batch size: 186, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:25:30,108 INFO [train.py:968] (1/2) Epoch 14, batch 44150, giga_loss[loss=0.2761, simple_loss=0.3449, pruned_loss=0.1037, over 28542.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3753, pruned_loss=0.1264, over 5673265.38 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.369, pruned_loss=0.1218, over 5668943.69 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5665525.56 frames. ], batch size: 85, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:25:41,740 INFO [optim.py:369] (1/2) 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 14:25:53,821 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:25:56,556 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:26:15,200 INFO [train.py:968] (1/2) Epoch 14, batch 44200, giga_loss[loss=0.3171, simple_loss=0.3781, pruned_loss=0.128, over 28700.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3766, pruned_loss=0.1272, over 5679028.59 frames. ], libri_tot_loss[loss=0.3067, simple_loss=0.3693, pruned_loss=0.1221, over 5670973.06 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3772, pruned_loss=0.1276, over 5671242.53 frames. ], batch size: 99, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:26:15,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3422, 1.1648, 1.1453, 1.4551], device='cuda:1'), covar=tensor([0.0721, 0.0379, 0.0336, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 14:26:23,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:27:02,752 INFO [train.py:968] (1/2) Epoch 14, batch 44250, giga_loss[loss=0.3192, simple_loss=0.4089, pruned_loss=0.1148, over 28997.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3765, pruned_loss=0.1273, over 5671575.42 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3689, pruned_loss=0.1217, over 5674091.03 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3775, pruned_loss=0.1281, over 5662396.80 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:27:14,890 INFO [optim.py:369] (1/2) 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 14:27:46,797 INFO [train.py:968] (1/2) Epoch 14, batch 44300, giga_loss[loss=0.2848, simple_loss=0.3582, pruned_loss=0.1057, over 28681.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3782, pruned_loss=0.1261, over 5675451.14 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3683, pruned_loss=0.1214, over 5676117.28 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3798, pruned_loss=0.1272, over 5666048.64 frames. ], batch size: 92, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:27:49,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637603.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:27:58,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0643, 1.4303, 1.5320, 1.1925], device='cuda:1'), covar=tensor([0.1985, 0.1775, 0.2149, 0.2208], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0739, 0.0692, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 14:28:14,668 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 44350, giga_loss[loss=0.3198, simple_loss=0.3926, pruned_loss=0.1235, over 28976.00 frames. ], tot_loss[loss=0.315, simple_loss=0.38, pruned_loss=0.125, over 5681679.29 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3684, pruned_loss=0.1214, over 5678205.14 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3813, pruned_loss=0.1258, over 5672429.55 frames. ], batch size: 106, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:28:42,655 INFO [optim.py:369] (1/2) 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 14:29:16,934 INFO [train.py:968] (1/2) Epoch 14, batch 44400, giga_loss[loss=0.3167, simple_loss=0.3876, pruned_loss=0.1228, over 28904.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3819, pruned_loss=0.1258, over 5694753.08 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3685, pruned_loss=0.1214, over 5683903.85 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3833, pruned_loss=0.1266, over 5682528.58 frames. ], batch size: 186, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:29:21,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3139, 1.2492, 1.2150, 1.4534], device='cuda:1'), covar=tensor([0.0743, 0.0348, 0.0322, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 14:29:55,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7414, 2.4420, 1.5061, 0.8007], device='cuda:1'), covar=tensor([0.6202, 0.3028, 0.3324, 0.5798], device='cuda:1'), in_proj_covar=tensor([0.1636, 0.1560, 0.1525, 0.1337], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 14:29:55,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2828, 1.5690, 1.3041, 1.0630], device='cuda:1'), covar=tensor([0.2469, 0.2512, 0.2831, 0.2114], device='cuda:1'), in_proj_covar=tensor([0.1367, 0.1006, 0.1211, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 14:29:59,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0146, 2.2111, 1.3502, 1.9194], device='cuda:1'), covar=tensor([0.0736, 0.0467, 0.1014, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0445, 0.0505, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:30:03,644 INFO [train.py:968] (1/2) Epoch 14, batch 44450, giga_loss[loss=0.3092, simple_loss=0.3745, pruned_loss=0.122, over 28877.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3849, pruned_loss=0.1293, over 5684504.48 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3687, pruned_loss=0.1214, over 5687347.98 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3861, pruned_loss=0.13, over 5671887.89 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:30:15,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:18,302 INFO [optim.py:369] (1/2) 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 14:30:19,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=637766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:27,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:27,887 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:32,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:36,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-07 14:30:43,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2452, 1.3742, 1.2060, 1.4474], device='cuda:1'), covar=tensor([0.0747, 0.0371, 0.0334, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 14:30:50,667 INFO [train.py:968] (1/2) Epoch 14, batch 44500, giga_loss[loss=0.3221, simple_loss=0.3902, pruned_loss=0.127, over 28609.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3859, pruned_loss=0.1314, over 5667879.41 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1214, over 5690015.88 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3876, pruned_loss=0.1325, over 5654554.69 frames. ], batch size: 262, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:30:56,047 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:30:56,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-07 14:31:33,106 INFO [train.py:968] (1/2) Epoch 14, batch 44550, giga_loss[loss=0.2684, simple_loss=0.3497, pruned_loss=0.09359, over 28937.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3851, pruned_loss=0.1313, over 5655837.88 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3688, pruned_loss=0.1216, over 5674438.47 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3867, pruned_loss=0.1322, over 5658373.17 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:31:44,327 INFO [optim.py:369] (1/2) 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 14:32:16,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-07 14:32:17,227 INFO [train.py:968] (1/2) Epoch 14, batch 44600, giga_loss[loss=0.3069, simple_loss=0.3778, pruned_loss=0.1179, over 28994.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.383, pruned_loss=0.1291, over 5659102.33 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3684, pruned_loss=0.1214, over 5675888.76 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3848, pruned_loss=0.1301, over 5659692.78 frames. ], batch size: 136, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:32:20,466 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:23,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:30,924 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=637917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:34,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=637920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:32:48,801 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:33:01,351 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=637949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:33:01,800 INFO [train.py:968] (1/2) Epoch 14, batch 44650, libri_loss[loss=0.3849, simple_loss=0.4191, pruned_loss=0.1754, over 19589.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3837, pruned_loss=0.1275, over 5657953.13 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1215, over 5669905.31 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3851, pruned_loss=0.1282, over 5664367.53 frames. ], batch size: 187, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:33:17,203 INFO [optim.py:369] (1/2) 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 14:33:18,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 14:33:27,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=637978.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:33:48,614 INFO [train.py:968] (1/2) Epoch 14, batch 44700, giga_loss[loss=0.331, simple_loss=0.3957, pruned_loss=0.1331, over 28592.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3852, pruned_loss=0.1281, over 5651355.09 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3691, pruned_loss=0.122, over 5661762.13 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.386, pruned_loss=0.1283, over 5663328.40 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:34:41,435 INFO [train.py:968] (1/2) Epoch 14, batch 44750, giga_loss[loss=0.304, simple_loss=0.3773, pruned_loss=0.1153, over 28891.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3848, pruned_loss=0.1285, over 5657862.79 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.369, pruned_loss=0.1219, over 5664320.44 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3857, pruned_loss=0.1289, over 5664846.59 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:34:53,341 INFO [optim.py:369] (1/2) 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 14:35:10,935 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 44800, giga_loss[loss=0.3039, simple_loss=0.3791, pruned_loss=0.1143, over 28939.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3834, pruned_loss=0.1283, over 5662688.66 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3691, pruned_loss=0.1219, over 5666756.05 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3841, pruned_loss=0.1285, over 5666102.07 frames. ], batch size: 164, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:35:46,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8434, 1.8042, 1.3592, 1.4637], device='cuda:1'), covar=tensor([0.0786, 0.0681, 0.0924, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0445, 0.0505, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:35:46,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638121.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:35:48,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638124.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 14:36:03,149 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 14, batch 44850, giga_loss[loss=0.2823, simple_loss=0.3574, pruned_loss=0.1036, over 28904.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3824, pruned_loss=0.1291, over 5656313.64 frames. ], libri_tot_loss[loss=0.3069, simple_loss=0.3695, pruned_loss=0.1221, over 5671123.33 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3832, pruned_loss=0.1294, over 5654381.29 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:36:14,556 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638153.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 14:36:26,297 INFO [optim.py:369] (1/2) 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 14:36:37,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2234, 1.3204, 3.7832, 3.1549], device='cuda:1'), covar=tensor([0.1597, 0.2450, 0.0461, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0703, 0.0613, 0.0902, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:36:54,504 INFO [train.py:968] (1/2) Epoch 14, batch 44900, libri_loss[loss=0.3433, simple_loss=0.3821, pruned_loss=0.1522, over 28104.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3803, pruned_loss=0.1285, over 5654617.09 frames. ], libri_tot_loss[loss=0.3078, simple_loss=0.3702, pruned_loss=0.1227, over 5671737.38 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3806, pruned_loss=0.1284, over 5652177.32 frames. ], batch size: 62, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:37:39,923 INFO [train.py:968] (1/2) Epoch 14, batch 44950, giga_loss[loss=0.3534, simple_loss=0.3977, pruned_loss=0.1545, over 28768.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3775, pruned_loss=0.127, over 5662552.28 frames. ], libri_tot_loss[loss=0.3071, simple_loss=0.3697, pruned_loss=0.1222, over 5677179.82 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3784, pruned_loss=0.1274, over 5655279.01 frames. ], batch size: 284, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:37:43,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8496, 2.9249, 1.8349, 0.9285], device='cuda:1'), covar=tensor([0.6504, 0.2508, 0.3595, 0.6169], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1553, 0.1521, 0.1327], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 14:37:53,999 INFO [optim.py:369] (1/2) 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 14:37:57,904 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-07 14:38:03,543 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 14:38:07,222 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:38:09,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:38:11,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=638289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:38:14,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7436, 2.1746, 1.8778, 1.5330], device='cuda:1'), covar=tensor([0.2366, 0.1756, 0.1925, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1804, 0.1718, 0.1678, 0.1794], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 14:38:23,196 INFO [train.py:968] (1/2) Epoch 14, batch 45000, giga_loss[loss=0.2869, simple_loss=0.3563, pruned_loss=0.1088, over 28933.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3763, pruned_loss=0.1268, over 5665910.76 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.37, pruned_loss=0.1225, over 5680913.08 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3769, pruned_loss=0.1271, over 5656286.53 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:38:23,196 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 14:38:32,169 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 14:38:46,604 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:39:07,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 14:39:14,700 INFO [train.py:968] (1/2) Epoch 14, batch 45050, giga_loss[loss=0.2977, simple_loss=0.3714, pruned_loss=0.112, over 28616.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3754, pruned_loss=0.1258, over 5664202.62 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.3701, pruned_loss=0.1226, over 5677123.49 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.376, pruned_loss=0.126, over 5659196.86 frames. ], batch size: 336, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:39:26,428 INFO [optim.py:369] (1/2) 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 14:39:56,813 INFO [train.py:968] (1/2) Epoch 14, batch 45100, giga_loss[loss=0.2678, simple_loss=0.3482, pruned_loss=0.09373, over 28689.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3722, pruned_loss=0.1225, over 5655700.37 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3699, pruned_loss=0.1226, over 5673828.82 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.373, pruned_loss=0.1228, over 5653956.35 frames. ], batch size: 284, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:40:09,982 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-07 14:40:11,530 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=638419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:40:36,676 INFO [train.py:968] (1/2) Epoch 14, batch 45150, giga_loss[loss=0.3127, simple_loss=0.3787, pruned_loss=0.1234, over 28944.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3704, pruned_loss=0.1207, over 5672932.27 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.37, pruned_loss=0.1228, over 5681410.45 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.371, pruned_loss=0.1206, over 5664209.05 frames. ], batch size: 227, lr: 2.26e-03, grad_scale: 2.0
+2023-03-07 14:40:45,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=638459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:40:53,261 INFO [optim.py:369] (1/2) 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 14:41:25,931 INFO [train.py:968] (1/2) Epoch 14, batch 45200, giga_loss[loss=0.2674, simple_loss=0.3454, pruned_loss=0.0947, over 28873.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3699, pruned_loss=0.1205, over 5651627.42 frames. ], libri_tot_loss[loss=0.3079, simple_loss=0.3701, pruned_loss=0.1228, over 5674045.24 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3703, pruned_loss=0.1203, over 5651498.20 frames. ], batch size: 174, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:42:15,647 INFO [train.py:968] (1/2) Epoch 14, batch 45250, giga_loss[loss=0.2901, simple_loss=0.3539, pruned_loss=0.1132, over 27989.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3668, pruned_loss=0.1195, over 5642587.49 frames. ], libri_tot_loss[loss=0.3074, simple_loss=0.3697, pruned_loss=0.1225, over 5676561.93 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1196, over 5639775.54 frames. ], batch size: 412, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:42:27,277 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 14:42:30,683 INFO [optim.py:369] (1/2) 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 14:43:02,589 INFO [train.py:968] (1/2) Epoch 14, batch 45300, giga_loss[loss=0.3167, simple_loss=0.358, pruned_loss=0.1377, over 23261.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3662, pruned_loss=0.1195, over 5640968.23 frames. ], libri_tot_loss[loss=0.3077, simple_loss=0.37, pruned_loss=0.1227, over 5677378.77 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1195, over 5637638.23 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:43:05,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:07,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:32,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:43,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 14:43:45,356 INFO [train.py:968] (1/2) Epoch 14, batch 45350, giga_loss[loss=0.3515, simple_loss=0.3824, pruned_loss=0.1603, over 23541.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5636120.54 frames. ], libri_tot_loss[loss=0.3075, simple_loss=0.3698, pruned_loss=0.1226, over 5670516.82 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5638563.86 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:43:59,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=638664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:43:59,996 INFO [optim.py:369] (1/2) 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 14:44:11,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1768, 1.3401, 3.4422, 3.0287], device='cuda:1'), covar=tensor([0.1511, 0.2468, 0.0425, 0.1777], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0613, 0.0901, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:44:29,266 INFO [train.py:968] (1/2) Epoch 14, batch 45400, giga_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 29093.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 5626418.41 frames. ], libri_tot_loss[loss=0.3082, simple_loss=0.3705, pruned_loss=0.123, over 5657855.89 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5639721.26 frames. ], batch size: 155, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:45:15,563 INFO [train.py:968] (1/2) Epoch 14, batch 45450, giga_loss[loss=0.3079, simple_loss=0.3721, pruned_loss=0.1218, over 28643.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3707, pruned_loss=0.1217, over 5619788.86 frames. ], libri_tot_loss[loss=0.3083, simple_loss=0.3707, pruned_loss=0.1229, over 5655517.16 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5631961.34 frames. ], batch size: 262, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:45:27,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=638764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:45:29,105 INFO [optim.py:369] (1/2) 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 14:45:55,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=638794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:00,439 INFO [train.py:968] (1/2) Epoch 14, batch 45500, giga_loss[loss=0.3132, simple_loss=0.3661, pruned_loss=0.1302, over 28675.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3716, pruned_loss=0.1228, over 5605583.82 frames. ], libri_tot_loss[loss=0.3089, simple_loss=0.3712, pruned_loss=0.1233, over 5639145.88 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3708, pruned_loss=0.1222, over 5629672.10 frames. ], batch size: 92, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:46:07,399 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:10,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:32,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:46:41,286 INFO [train.py:968] (1/2) Epoch 14, batch 45550, giga_loss[loss=0.3216, simple_loss=0.3828, pruned_loss=0.1302, over 28932.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3751, pruned_loss=0.1256, over 5561292.08 frames. ], libri_tot_loss[loss=0.3107, simple_loss=0.3725, pruned_loss=0.1244, over 5571487.09 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5640260.39 frames. ], batch size: 145, lr: 2.26e-03, grad_scale: 4.0
+2023-03-07 14:46:59,950 INFO [optim.py:369] (1/2) 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 14:47:10,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4334, 1.3655, 4.1057, 3.4212], device='cuda:1'), covar=tensor([0.1542, 0.2624, 0.0409, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0703, 0.0613, 0.0900, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:47:33,175 INFO [train.py:968] (1/2) Epoch 14, batch 45600, giga_loss[loss=0.344, simple_loss=0.3814, pruned_loss=0.1533, over 23290.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3763, pruned_loss=0.1259, over 5569495.96 frames. ], libri_tot_loss[loss=0.3112, simple_loss=0.3729, pruned_loss=0.1248, over 5554935.38 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3745, pruned_loss=0.1244, over 5647041.65 frames. ], batch size: 705, lr: 2.26e-03, grad_scale: 8.0
+2023-03-07 14:47:49,938 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-07 14:48:29,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5198, 1.9679, 1.4251, 1.6175], device='cuda:1'), covar=tensor([0.2676, 0.2548, 0.2958, 0.2418], device='cuda:1'), in_proj_covar=tensor([0.1379, 0.1014, 0.1219, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 14:48:33,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 14:48:38,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=638937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:48:40,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=638940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:49:02,706 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 50, giga_loss[loss=0.2905, simple_loss=0.3747, pruned_loss=0.1031, over 28933.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3796, pruned_loss=0.1146, over 1270455.61 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3552, pruned_loss=0.1002, over 232636.42 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3843, pruned_loss=0.1174, over 1081430.82 frames. ], batch size: 213, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:49:07,326 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=638969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:49:46,723 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-07 14:49:54,882 INFO [train.py:968] (1/2) Epoch 15, batch 100, giga_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1128, over 27581.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3716, pruned_loss=0.1109, over 2243694.91 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3487, pruned_loss=0.0959, over 372184.96 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3753, pruned_loss=0.1133, over 2001648.86 frames. ], batch size: 472, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:49:56,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2354, 1.4661, 1.3405, 1.1299], device='cuda:1'), covar=tensor([0.2019, 0.1882, 0.1209, 0.1652], device='cuda:1'), in_proj_covar=tensor([0.1806, 0.1720, 0.1686, 0.1796], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 14:49:58,287 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 14:50:36,174 INFO [optim.py:369] (1/2) 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 14:50:38,197 INFO [train.py:968] (1/2) Epoch 15, batch 150, giga_loss[loss=0.2837, simple_loss=0.3537, pruned_loss=0.1069, over 28301.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3541, pruned_loss=0.1016, over 3017203.87 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3444, pruned_loss=0.09328, over 535830.73 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3563, pruned_loss=0.1033, over 2735699.96 frames. ], batch size: 369, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:51:08,265 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:51:14,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5799, 2.3271, 1.7907, 0.7524], device='cuda:1'), covar=tensor([0.5844, 0.2854, 0.3715, 0.5903], device='cuda:1'), in_proj_covar=tensor([0.1639, 0.1559, 0.1527, 0.1335], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 14:51:20,786 INFO [train.py:968] (1/2) Epoch 15, batch 200, giga_loss[loss=0.2377, simple_loss=0.32, pruned_loss=0.07771, over 28958.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3384, pruned_loss=0.09377, over 3611391.70 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3435, pruned_loss=0.09304, over 615787.64 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3388, pruned_loss=0.09438, over 3354731.65 frames. ], batch size: 164, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:51:21,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2453, 3.0350, 2.8800, 1.3855], device='cuda:1'), covar=tensor([0.0905, 0.1100, 0.0934, 0.2420], device='cuda:1'), in_proj_covar=tensor([0.1122, 0.1046, 0.0903, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 14:51:26,086 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:51:36,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:51:43,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-07 14:52:02,461 INFO [optim.py:369] (1/2) 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 14:52:02,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8742, 2.3773, 2.0256, 1.5415], device='cuda:1'), covar=tensor([0.3172, 0.2010, 0.2006, 0.2729], device='cuda:1'), in_proj_covar=tensor([0.1815, 0.1727, 0.1690, 0.1800], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 14:52:03,948 INFO [train.py:968] (1/2) Epoch 15, batch 250, giga_loss[loss=0.2246, simple_loss=0.2928, pruned_loss=0.07818, over 28590.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3276, pruned_loss=0.089, over 4070220.68 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3428, pruned_loss=0.09248, over 667623.39 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3273, pruned_loss=0.08922, over 3851184.53 frames. ], batch size: 71, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:52:42,400 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:52:45,632 INFO [train.py:968] (1/2) Epoch 15, batch 300, giga_loss[loss=0.2476, simple_loss=0.3048, pruned_loss=0.09518, over 26604.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3203, pruned_loss=0.08598, over 4427319.75 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3471, pruned_loss=0.09384, over 795643.99 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3184, pruned_loss=0.08561, over 4217050.29 frames. ], batch size: 555, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:53:01,028 INFO [scaling.py:679] (1/2) 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] (1/2) 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] (1/2) Epoch 15, batch 350, giga_loss[loss=0.1954, simple_loss=0.2677, pruned_loss=0.06152, over 28934.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3145, pruned_loss=0.08331, over 4706274.20 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3471, pruned_loss=0.09428, over 968284.77 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3115, pruned_loss=0.08244, over 4498008.10 frames. ], batch size: 106, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:53:39,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:53:42,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:54:01,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 14:54:06,496 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 400, giga_loss[loss=0.213, simple_loss=0.289, pruned_loss=0.06851, over 28439.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3092, pruned_loss=0.08066, over 4931056.20 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3456, pruned_loss=0.09361, over 1017677.87 frames. ], giga_tot_loss[loss=0.2333, simple_loss=0.3067, pruned_loss=0.0799, over 4757442.45 frames. ], batch size: 78, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:54:50,634 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 450, giga_loss[loss=0.2027, simple_loss=0.2632, pruned_loss=0.07117, over 23833.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3078, pruned_loss=0.08068, over 5092276.61 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3471, pruned_loss=0.09463, over 1066203.07 frames. ], giga_tot_loss[loss=0.2324, simple_loss=0.3052, pruned_loss=0.07977, over 4948065.40 frames. ], batch size: 705, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:54:57,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-07 14:55:11,935 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-07 14:55:21,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2149, 1.3854, 1.3455, 1.1880], device='cuda:1'), covar=tensor([0.2175, 0.1918, 0.1216, 0.1732], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1717, 0.1680, 0.1795], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 14:55:37,286 INFO [train.py:968] (1/2) Epoch 15, batch 500, giga_loss[loss=0.2333, simple_loss=0.2882, pruned_loss=0.08924, over 23613.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3068, pruned_loss=0.08009, over 5216788.71 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3477, pruned_loss=0.09471, over 1220701.73 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3033, pruned_loss=0.07887, over 5089340.68 frames. ], batch size: 705, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 14:56:19,513 INFO [optim.py:369] (1/2) 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 14:56:21,710 INFO [train.py:968] (1/2) Epoch 15, batch 550, libri_loss[loss=0.2719, simple_loss=0.3556, pruned_loss=0.09405, over 29140.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.303, pruned_loss=0.07804, over 5322272.89 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3472, pruned_loss=0.09454, over 1267013.75 frames. ], giga_tot_loss[loss=0.2269, simple_loss=0.2999, pruned_loss=0.07692, over 5216659.89 frames. ], batch size: 101, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:56:28,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:56:47,718 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 600, giga_loss[loss=0.2041, simple_loss=0.2787, pruned_loss=0.06478, over 29099.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3017, pruned_loss=0.07705, over 5411636.37 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3492, pruned_loss=0.09541, over 1424918.28 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.2972, pruned_loss=0.07538, over 5311282.97 frames. ], batch size: 128, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:57:03,674 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=639519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:57:44,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6614, 1.8649, 1.4592, 1.3920], device='cuda:1'), covar=tensor([0.0916, 0.0572, 0.1029, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0443, 0.0504, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:57:48,553 INFO [optim.py:369] (1/2) 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 14:57:49,207 INFO [train.py:968] (1/2) Epoch 15, batch 650, giga_loss[loss=0.1968, simple_loss=0.2699, pruned_loss=0.06188, over 29069.00 frames. ], tot_loss[loss=0.225, simple_loss=0.2989, pruned_loss=0.07555, over 5479338.12 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3484, pruned_loss=0.09477, over 1514000.85 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.2946, pruned_loss=0.07402, over 5391362.86 frames. ], batch size: 136, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:58:09,964 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 700, libri_loss[loss=0.2611, simple_loss=0.3392, pruned_loss=0.09149, over 29569.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2962, pruned_loss=0.07429, over 5529580.92 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.348, pruned_loss=0.09464, over 1601297.46 frames. ], giga_tot_loss[loss=0.2187, simple_loss=0.2919, pruned_loss=0.07272, over 5452404.80 frames. ], batch size: 78, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:58:35,995 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:58:39,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:58:55,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:58:58,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:59:05,397 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=639652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:59:20,717 INFO [optim.py:369] (1/2) 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 14:59:21,291 INFO [train.py:968] (1/2) Epoch 15, batch 750, giga_loss[loss=0.2872, simple_loss=0.3316, pruned_loss=0.1214, over 26620.00 frames. ], tot_loss[loss=0.2197, simple_loss=0.2934, pruned_loss=0.073, over 5555596.23 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3474, pruned_loss=0.09415, over 1686939.63 frames. ], giga_tot_loss[loss=0.2161, simple_loss=0.2892, pruned_loss=0.0715, over 5488024.25 frames. ], batch size: 555, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 14:59:24,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4043, 1.2935, 4.1685, 3.3024], device='cuda:1'), covar=tensor([0.1621, 0.2748, 0.0399, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0699, 0.0610, 0.0899, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 14:59:26,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=639675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 14:59:55,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5397, 1.6833, 1.4805, 1.6119], device='cuda:1'), covar=tensor([0.0747, 0.0320, 0.0321, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0057, 0.0096], device='cuda:1')
+2023-03-07 14:59:58,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-07 15:00:03,512 INFO [train.py:968] (1/2) Epoch 15, batch 800, giga_loss[loss=0.2335, simple_loss=0.3101, pruned_loss=0.07845, over 29058.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2929, pruned_loss=0.07295, over 5587702.05 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3467, pruned_loss=0.09367, over 1729264.67 frames. ], giga_tot_loss[loss=0.2164, simple_loss=0.2893, pruned_loss=0.07171, over 5531462.91 frames. ], batch size: 128, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 15:00:16,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=639733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:00:19,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=639736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:00:41,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8287, 2.0657, 1.9557, 1.7960], device='cuda:1'), covar=tensor([0.2198, 0.1634, 0.1448, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.1811, 0.1716, 0.1678, 0.1794], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 15:00:47,648 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=639765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:00:50,476 INFO [optim.py:369] (1/2) 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 15:00:50,491 INFO [train.py:968] (1/2) Epoch 15, batch 850, giga_loss[loss=0.3187, simple_loss=0.3889, pruned_loss=0.1243, over 28316.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3046, pruned_loss=0.07924, over 5606796.28 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3473, pruned_loss=0.09398, over 1849419.36 frames. ], giga_tot_loss[loss=0.2279, simple_loss=0.3003, pruned_loss=0.07771, over 5556739.30 frames. ], batch size: 368, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:00:51,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-07 15:01:35,678 INFO [train.py:968] (1/2) Epoch 15, batch 900, giga_loss[loss=0.2985, simple_loss=0.3716, pruned_loss=0.1127, over 28849.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3182, pruned_loss=0.08583, over 5629247.22 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3479, pruned_loss=0.0942, over 1908491.16 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3143, pruned_loss=0.08446, over 5587116.28 frames. ], batch size: 186, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:01:50,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6784, 1.9156, 1.4796, 2.0660], device='cuda:1'), covar=tensor([0.2447, 0.2565, 0.2795, 0.2276], device='cuda:1'), in_proj_covar=tensor([0.1383, 0.1015, 0.1224, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 15:02:14,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-07 15:02:18,763 INFO [optim.py:369] (1/2) 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 15:02:18,775 INFO [train.py:968] (1/2) Epoch 15, batch 950, giga_loss[loss=0.2826, simple_loss=0.3625, pruned_loss=0.1013, over 28898.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3295, pruned_loss=0.09181, over 5630094.66 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3477, pruned_loss=0.09437, over 2014296.91 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3259, pruned_loss=0.09059, over 5599815.06 frames. ], batch size: 227, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:02:39,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=639894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:03:00,057 INFO [train.py:968] (1/2) Epoch 15, batch 1000, giga_loss[loss=0.2799, simple_loss=0.3611, pruned_loss=0.09939, over 28907.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3363, pruned_loss=0.09445, over 5634021.42 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3475, pruned_loss=0.09434, over 2062990.79 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3333, pruned_loss=0.09348, over 5614844.24 frames. ], batch size: 174, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:03:38,363 INFO [optim.py:369] (1/2) 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 15:03:38,376 INFO [train.py:968] (1/2) Epoch 15, batch 1050, giga_loss[loss=0.2538, simple_loss=0.3352, pruned_loss=0.08624, over 28848.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.339, pruned_loss=0.09434, over 5659676.99 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3469, pruned_loss=0.09412, over 2214548.74 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3365, pruned_loss=0.09365, over 5634105.02 frames. ], batch size: 119, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:04:23,757 INFO [train.py:968] (1/2) Epoch 15, batch 1100, giga_loss[loss=0.2804, simple_loss=0.3575, pruned_loss=0.1016, over 28985.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3404, pruned_loss=0.09453, over 5657813.37 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3464, pruned_loss=0.09376, over 2251171.94 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3386, pruned_loss=0.09412, over 5635619.47 frames. ], batch size: 145, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:04:31,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3306, 1.1711, 1.0828, 1.4394], device='cuda:1'), covar=tensor([0.0749, 0.0380, 0.0353, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 15:04:40,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=640037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:04:42,422 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:05:06,818 INFO [train.py:968] (1/2) Epoch 15, batch 1150, giga_loss[loss=0.2469, simple_loss=0.3235, pruned_loss=0.08516, over 28385.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3433, pruned_loss=0.09666, over 5665973.25 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3466, pruned_loss=0.09389, over 2341907.32 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3416, pruned_loss=0.09634, over 5642770.08 frames. ], batch size: 65, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:05:07,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=640069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:05:18,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5935, 1.8023, 1.8519, 1.4051], device='cuda:1'), covar=tensor([0.1782, 0.2498, 0.1427, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0695, 0.0906, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 15:05:21,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 15:05:43,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2598, 2.5091, 2.0755, 1.8254], device='cuda:1'), covar=tensor([0.1843, 0.1726, 0.1963, 0.1951], device='cuda:1'), in_proj_covar=tensor([0.1790, 0.1705, 0.1659, 0.1776], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 15:05:49,332 INFO [train.py:968] (1/2) Epoch 15, batch 1200, libri_loss[loss=0.2112, simple_loss=0.292, pruned_loss=0.06522, over 28105.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3456, pruned_loss=0.09821, over 5670689.22 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3463, pruned_loss=0.09371, over 2461684.26 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3444, pruned_loss=0.09815, over 5647614.27 frames. ], batch size: 62, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 15:06:29,407 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 1250, giga_loss[loss=0.3226, simple_loss=0.3844, pruned_loss=0.1304, over 27581.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3486, pruned_loss=0.09941, over 5677425.63 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3475, pruned_loss=0.09414, over 2542808.45 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09931, over 5657613.60 frames. ], batch size: 472, lr: 2.18e-03, grad_scale: 8.0
+2023-03-07 15:06:32,161 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:07:09,811 INFO [train.py:968] (1/2) Epoch 15, batch 1300, giga_loss[loss=0.2784, simple_loss=0.3592, pruned_loss=0.09877, over 28987.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3519, pruned_loss=0.1006, over 5678169.41 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3478, pruned_loss=0.09428, over 2588689.56 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3507, pruned_loss=0.1006, over 5663387.68 frames. ], batch size: 136, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:07:48,427 INFO [train.py:968] (1/2) Epoch 15, batch 1350, giga_loss[loss=0.2934, simple_loss=0.3797, pruned_loss=0.1036, over 29009.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3533, pruned_loss=0.1009, over 5682219.37 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3467, pruned_loss=0.09374, over 2683963.17 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.353, pruned_loss=0.1013, over 5667107.32 frames. ], batch size: 155, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:07:48,932 INFO [optim.py:369] (1/2) 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 15:08:27,748 INFO [train.py:968] (1/2) Epoch 15, batch 1400, giga_loss[loss=0.2894, simple_loss=0.3622, pruned_loss=0.1083, over 28254.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3548, pruned_loss=0.1009, over 5693906.35 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3463, pruned_loss=0.09344, over 2715879.44 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3547, pruned_loss=0.1014, over 5680360.26 frames. ], batch size: 368, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:09:11,706 INFO [train.py:968] (1/2) Epoch 15, batch 1450, libri_loss[loss=0.2061, simple_loss=0.2941, pruned_loss=0.05909, over 29653.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3535, pruned_loss=0.09915, over 5695483.38 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3451, pruned_loss=0.09259, over 2779690.98 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3541, pruned_loss=0.1, over 5680613.69 frames. ], batch size: 73, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:09:12,361 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 1500, giga_loss[loss=0.2527, simple_loss=0.3472, pruned_loss=0.07904, over 29066.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3522, pruned_loss=0.0971, over 5703794.54 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3454, pruned_loss=0.09272, over 2807180.84 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3526, pruned_loss=0.09777, over 5693721.26 frames. ], batch size: 155, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:10:34,138 INFO [train.py:968] (1/2) Epoch 15, batch 1550, giga_loss[loss=0.2795, simple_loss=0.3527, pruned_loss=0.1031, over 28856.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3523, pruned_loss=0.09729, over 5699684.28 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3463, pruned_loss=0.09289, over 2868726.97 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3524, pruned_loss=0.09783, over 5688289.72 frames. ], batch size: 199, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:10:34,713 INFO [optim.py:369] (1/2) 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 15:10:41,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8036, 1.8583, 1.3213, 1.4724], device='cuda:1'), covar=tensor([0.0814, 0.0591, 0.0971, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0440, 0.0504, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 15:11:15,999 INFO [train.py:968] (1/2) Epoch 15, batch 1600, giga_loss[loss=0.3026, simple_loss=0.3689, pruned_loss=0.1181, over 29003.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3543, pruned_loss=0.1005, over 5702110.68 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3464, pruned_loss=0.09318, over 2984619.61 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3546, pruned_loss=0.101, over 5688742.71 frames. ], batch size: 106, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:11:34,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-07 15:11:38,742 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-07 15:11:42,270 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 1650, giga_loss[loss=0.3017, simple_loss=0.3663, pruned_loss=0.1186, over 27940.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3562, pruned_loss=0.1038, over 5710370.99 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3458, pruned_loss=0.09301, over 3098084.14 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.357, pruned_loss=0.1046, over 5693553.09 frames. ], batch size: 412, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:12:01,820 INFO [optim.py:369] (1/2) 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 15:12:28,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:12:32,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-07 15:12:43,195 INFO [train.py:968] (1/2) Epoch 15, batch 1700, giga_loss[loss=0.2598, simple_loss=0.337, pruned_loss=0.0913, over 28923.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3567, pruned_loss=0.1053, over 5719210.71 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.346, pruned_loss=0.09316, over 3167874.66 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3575, pruned_loss=0.1062, over 5701132.15 frames. ], batch size: 66, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:13:07,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-07 15:13:14,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:13:16,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 15:13:25,082 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 1750, giga_loss[loss=0.2279, simple_loss=0.3079, pruned_loss=0.0739, over 28570.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3547, pruned_loss=0.1048, over 5696114.00 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3468, pruned_loss=0.09366, over 3225047.20 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3551, pruned_loss=0.1055, over 5686610.08 frames. ], batch size: 60, lr: 2.18e-03, grad_scale: 4.0
+2023-03-07 15:13:28,882 INFO [optim.py:369] (1/2) 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 15:13:42,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4565, 1.6113, 1.5394, 1.4550], device='cuda:1'), covar=tensor([0.1534, 0.2037, 0.2037, 0.1869], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0733, 0.0686, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 15:13:44,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=640691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:13:46,639 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=640694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:14:08,249 INFO [train.py:968] (1/2) Epoch 15, batch 1800, giga_loss[loss=0.2899, simple_loss=0.3499, pruned_loss=0.1149, over 23797.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3535, pruned_loss=0.1045, over 5694143.09 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3475, pruned_loss=0.09407, over 3291712.75 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3537, pruned_loss=0.1051, over 5681844.67 frames. ], batch size: 705, lr: 2.18e-03, grad_scale: 2.0
+2023-03-07 15:14:10,934 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=640723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:14:41,264 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:14:48,287 INFO [train.py:968] (1/2) Epoch 15, batch 1850, giga_loss[loss=0.2691, simple_loss=0.3456, pruned_loss=0.09629, over 28705.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3524, pruned_loss=0.1036, over 5692008.82 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.347, pruned_loss=0.09371, over 3330133.79 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3528, pruned_loss=0.1044, over 5680251.04 frames. ], batch size: 242, lr: 2.18e-03, grad_scale: 2.0
+2023-03-07 15:14:51,255 INFO [optim.py:369] (1/2) 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 15:15:19,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-07 15:15:33,442 INFO [train.py:968] (1/2) Epoch 15, batch 1900, giga_loss[loss=0.2813, simple_loss=0.3494, pruned_loss=0.1066, over 28830.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3493, pruned_loss=0.1008, over 5686376.26 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3472, pruned_loss=0.09355, over 3383105.83 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3497, pruned_loss=0.1017, over 5682134.92 frames. ], batch size: 99, lr: 2.18e-03, grad_scale: 2.0
+2023-03-07 15:15:35,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=640821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:16:15,969 INFO [train.py:968] (1/2) Epoch 15, batch 1950, giga_loss[loss=0.2605, simple_loss=0.3344, pruned_loss=0.09326, over 28709.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3453, pruned_loss=0.09871, over 5672707.38 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3473, pruned_loss=0.09376, over 3460858.57 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3454, pruned_loss=0.0995, over 5671172.08 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 15:16:18,428 INFO [optim.py:369] (1/2) 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 15:16:30,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6232, 1.7072, 1.5709, 1.4898], device='cuda:1'), covar=tensor([0.2228, 0.2093, 0.1852, 0.1957], device='cuda:1'), in_proj_covar=tensor([0.1791, 0.1708, 0.1665, 0.1776], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 15:17:04,546 INFO [train.py:968] (1/2) Epoch 15, batch 2000, giga_loss[loss=0.2237, simple_loss=0.3074, pruned_loss=0.06998, over 29040.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3393, pruned_loss=0.09586, over 5668712.05 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3477, pruned_loss=0.09404, over 3485210.13 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3392, pruned_loss=0.09635, over 5665378.11 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:17:17,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9722, 1.2980, 1.0725, 0.1794], device='cuda:1'), covar=tensor([0.3347, 0.2585, 0.3901, 0.5241], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1540, 0.1511, 0.1319], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 15:17:25,614 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.44 vs. limit=5.0
+2023-03-07 15:17:36,709 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-07 15:17:48,577 INFO [train.py:968] (1/2) Epoch 15, batch 2050, giga_loss[loss=0.2888, simple_loss=0.3325, pruned_loss=0.1226, over 23543.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3339, pruned_loss=0.09285, over 5665049.43 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3465, pruned_loss=0.09331, over 3576784.26 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3341, pruned_loss=0.09367, over 5657228.39 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:17:52,185 INFO [optim.py:369] (1/2) 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 15:17:56,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=640975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:18:33,301 INFO [train.py:968] (1/2) Epoch 15, batch 2100, giga_loss[loss=0.2272, simple_loss=0.3139, pruned_loss=0.07029, over 29025.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3341, pruned_loss=0.09261, over 5662281.41 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3469, pruned_loss=0.09366, over 3607738.85 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3337, pruned_loss=0.09304, over 5656111.19 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:18:35,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 15:18:40,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:18:44,612 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-07 15:18:50,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-07 15:18:53,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:19:15,298 INFO [train.py:968] (1/2) Epoch 15, batch 2150, giga_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1053, over 28356.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3355, pruned_loss=0.09267, over 5678095.48 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3476, pruned_loss=0.09365, over 3672551.45 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3345, pruned_loss=0.09299, over 5669939.57 frames. ], batch size: 368, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:19:17,213 INFO [optim.py:369] (1/2) 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 15:19:53,344 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:19:53,758 INFO [train.py:968] (1/2) Epoch 15, batch 2200, libri_loss[loss=0.282, simple_loss=0.3682, pruned_loss=0.09792, over 29263.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3355, pruned_loss=0.09259, over 5684142.14 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3484, pruned_loss=0.09358, over 3748399.00 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3337, pruned_loss=0.09284, over 5671504.33 frames. ], batch size: 94, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:19:55,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:08,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:19,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:34,868 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 15:20:35,065 INFO [train.py:968] (1/2) Epoch 15, batch 2250, giga_loss[loss=0.2147, simple_loss=0.2941, pruned_loss=0.06763, over 28904.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3324, pruned_loss=0.0911, over 5695203.22 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.09366, over 3770091.80 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3307, pruned_loss=0.09123, over 5683177.63 frames. ], batch size: 112, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:20:37,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:37,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-07 15:20:37,669 INFO [optim.py:369] (1/2) 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 15:20:40,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:48,270 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:50,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:20:59,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:21:05,563 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:21:05,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6579, 1.7552, 1.7090, 1.5578], device='cuda:1'), covar=tensor([0.1667, 0.2373, 0.2350, 0.2345], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0736, 0.0689, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 15:21:17,908 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=641217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:21:19,073 INFO [train.py:968] (1/2) Epoch 15, batch 2300, giga_loss[loss=0.2269, simple_loss=0.3099, pruned_loss=0.07196, over 28532.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3297, pruned_loss=0.08962, over 5704354.85 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3491, pruned_loss=0.09364, over 3812237.09 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3279, pruned_loss=0.08968, over 5691536.93 frames. ], batch size: 336, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:21:58,446 INFO [train.py:968] (1/2) Epoch 15, batch 2350, giga_loss[loss=0.2449, simple_loss=0.3157, pruned_loss=0.08706, over 28316.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.328, pruned_loss=0.08867, over 5694295.40 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3502, pruned_loss=0.09381, over 3866492.84 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3252, pruned_loss=0.08848, over 5693653.95 frames. ], batch size: 65, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:21:59,707 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 15:22:00,927 INFO [optim.py:369] (1/2) 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 15:22:06,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:22:09,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:22:31,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3014, 1.7535, 1.4034, 1.4202], device='cuda:1'), covar=tensor([0.0791, 0.0332, 0.0329, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 15:22:32,083 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 2400, giga_loss[loss=0.3666, simple_loss=0.4072, pruned_loss=0.163, over 27558.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3266, pruned_loss=0.08842, over 5699376.12 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3501, pruned_loss=0.09354, over 3916420.40 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.324, pruned_loss=0.08833, over 5694739.60 frames. ], batch size: 472, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:22:54,918 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:22:57,442 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 2450, giga_loss[loss=0.2314, simple_loss=0.3027, pruned_loss=0.07999, over 28222.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3242, pruned_loss=0.08726, over 5706328.04 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.35, pruned_loss=0.09323, over 3963315.63 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3216, pruned_loss=0.08726, over 5700272.65 frames. ], batch size: 77, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:23:17,528 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:23:53,184 INFO [train.py:968] (1/2) Epoch 15, batch 2500, giga_loss[loss=0.2368, simple_loss=0.314, pruned_loss=0.0798, over 28425.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3215, pruned_loss=0.08573, over 5716261.94 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3509, pruned_loss=0.09364, over 4001929.62 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3183, pruned_loss=0.08534, over 5707883.12 frames. ], batch size: 71, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:24:27,945 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:24:32,980 INFO [train.py:968] (1/2) Epoch 15, batch 2550, giga_loss[loss=0.2261, simple_loss=0.3068, pruned_loss=0.07265, over 28747.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3203, pruned_loss=0.08528, over 5723560.56 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3516, pruned_loss=0.09405, over 4030365.51 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3171, pruned_loss=0.08463, over 5714572.39 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:24:36,899 INFO [optim.py:369] (1/2) 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 15:25:06,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-07 15:25:11,683 INFO [train.py:968] (1/2) Epoch 15, batch 2600, giga_loss[loss=0.2376, simple_loss=0.3119, pruned_loss=0.08165, over 29048.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3199, pruned_loss=0.085, over 5722268.37 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3526, pruned_loss=0.09436, over 4080317.27 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3157, pruned_loss=0.08404, over 5715212.24 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:25:17,238 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:25:20,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5946, 1.7836, 1.8221, 1.3746], device='cuda:1'), covar=tensor([0.1825, 0.2416, 0.1475, 0.1696], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0698, 0.0909, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 15:25:50,915 INFO [train.py:968] (1/2) Epoch 15, batch 2650, giga_loss[loss=0.2248, simple_loss=0.299, pruned_loss=0.07529, over 28843.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3203, pruned_loss=0.08574, over 5722690.67 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3527, pruned_loss=0.09427, over 4106212.15 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3165, pruned_loss=0.08492, over 5715937.71 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:25:53,564 INFO [optim.py:369] (1/2) 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 15:26:32,733 INFO [train.py:968] (1/2) Epoch 15, batch 2700, giga_loss[loss=0.274, simple_loss=0.3398, pruned_loss=0.1042, over 28886.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.325, pruned_loss=0.08843, over 5722870.84 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3533, pruned_loss=0.09447, over 4157162.58 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3207, pruned_loss=0.08744, over 5714459.42 frames. ], batch size: 106, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:27:17,523 INFO [train.py:968] (1/2) Epoch 15, batch 2750, giga_loss[loss=0.3335, simple_loss=0.3945, pruned_loss=0.1362, over 28797.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3309, pruned_loss=0.09191, over 5716482.93 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3533, pruned_loss=0.09431, over 4195375.76 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3269, pruned_loss=0.09113, over 5710290.03 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:27:21,450 INFO [optim.py:369] (1/2) 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 15:28:01,817 INFO [train.py:968] (1/2) Epoch 15, batch 2800, giga_loss[loss=0.3271, simple_loss=0.3926, pruned_loss=0.1307, over 28629.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3384, pruned_loss=0.09687, over 5711421.40 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3531, pruned_loss=0.09425, over 4237274.71 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3349, pruned_loss=0.09627, over 5702774.96 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:28:03,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-07 15:28:43,441 INFO [train.py:968] (1/2) Epoch 15, batch 2850, giga_loss[loss=0.3012, simple_loss=0.373, pruned_loss=0.1147, over 27815.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3432, pruned_loss=0.09926, over 5709119.59 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3523, pruned_loss=0.09374, over 4300205.25 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3405, pruned_loss=0.09929, over 5697194.76 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:28:46,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4791, 1.6830, 1.4109, 1.3382], device='cuda:1'), covar=tensor([0.2559, 0.2598, 0.2858, 0.2339], device='cuda:1'), in_proj_covar=tensor([0.1378, 0.1011, 0.1220, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 15:28:47,666 INFO [optim.py:369] (1/2) 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 15:29:25,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9930, 2.3260, 2.0161, 1.6270], device='cuda:1'), covar=tensor([0.2177, 0.1815, 0.1936, 0.2226], device='cuda:1'), in_proj_covar=tensor([0.1783, 0.1698, 0.1662, 0.1777], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 15:29:30,191 INFO [train.py:968] (1/2) Epoch 15, batch 2900, giga_loss[loss=0.2793, simple_loss=0.356, pruned_loss=0.1013, over 28493.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3484, pruned_loss=0.1012, over 5712258.07 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3527, pruned_loss=0.09412, over 4345545.46 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3457, pruned_loss=0.1012, over 5699406.57 frames. ], batch size: 60, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:29:47,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:30:12,564 INFO [train.py:968] (1/2) Epoch 15, batch 2950, giga_loss[loss=0.3577, simple_loss=0.4176, pruned_loss=0.1488, over 27905.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3532, pruned_loss=0.1037, over 5712043.73 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3525, pruned_loss=0.09396, over 4391295.60 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3512, pruned_loss=0.104, over 5696972.84 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:30:15,793 INFO [optim.py:369] (1/2) 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 15:30:27,001 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641885.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:30:31,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1661, 1.5642, 1.4860, 1.0492], device='cuda:1'), covar=tensor([0.1644, 0.2575, 0.1482, 0.1660], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0694, 0.0904, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 15:30:41,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=641900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:30:55,079 INFO [train.py:968] (1/2) Epoch 15, batch 3000, giga_loss[loss=0.2633, simple_loss=0.3474, pruned_loss=0.08957, over 29022.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3572, pruned_loss=0.1066, over 5693905.55 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3521, pruned_loss=0.09391, over 4462878.87 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3559, pruned_loss=0.1074, over 5674872.29 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:30:55,079 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 15:31:03,255 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 15:31:33,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2387, 1.4960, 1.3178, 1.1548], device='cuda:1'), covar=tensor([0.2076, 0.1934, 0.1199, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.1788, 0.1700, 0.1663, 0.1780], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 15:31:46,304 INFO [train.py:968] (1/2) Epoch 15, batch 3050, giga_loss[loss=0.3187, simple_loss=0.3716, pruned_loss=0.133, over 26761.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3535, pruned_loss=0.1037, over 5700186.43 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3518, pruned_loss=0.09382, over 4477437.34 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3528, pruned_loss=0.1045, over 5683282.03 frames. ], batch size: 555, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:31:48,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=641972.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 15:31:50,395 INFO [optim.py:369] (1/2) 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 15:31:54,791 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=641980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:31:56,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=641983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:32:13,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-07 15:32:18,571 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1517, 2.5577, 1.2331, 1.3725], device='cuda:1'), covar=tensor([0.1061, 0.0332, 0.0927, 0.1452], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0518, 0.0354, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 15:32:19,746 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 3100, giga_loss[loss=0.2605, simple_loss=0.3417, pruned_loss=0.08965, over 28877.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3515, pruned_loss=0.1017, over 5712328.38 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3511, pruned_loss=0.09359, over 4546771.50 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3514, pruned_loss=0.1029, over 5689835.81 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:32:25,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9515, 1.2718, 0.9971, 0.1618], device='cuda:1'), covar=tensor([0.2971, 0.2286, 0.3723, 0.5363], device='cuda:1'), in_proj_covar=tensor([0.1626, 0.1546, 0.1529, 0.1333], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 15:32:46,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0904, 1.2472, 3.6728, 3.0795], device='cuda:1'), covar=tensor([0.1852, 0.2821, 0.0445, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0692, 0.0604, 0.0879, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0008, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 15:32:46,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:32:49,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:32:58,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3257, 1.0984, 4.4557, 3.3640], device='cuda:1'), covar=tensor([0.1824, 0.3042, 0.0380, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0693, 0.0605, 0.0880, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 15:33:05,687 INFO [train.py:968] (1/2) Epoch 15, batch 3150, libri_loss[loss=0.2814, simple_loss=0.3612, pruned_loss=0.1008, over 26045.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3497, pruned_loss=0.1, over 5711519.51 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3507, pruned_loss=0.09363, over 4587182.16 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3499, pruned_loss=0.1012, over 5698801.34 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:33:10,753 INFO [optim.py:369] (1/2) 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 15:33:11,637 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:33:45,772 INFO [train.py:968] (1/2) Epoch 15, batch 3200, giga_loss[loss=0.2612, simple_loss=0.3458, pruned_loss=0.08829, over 29061.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3497, pruned_loss=0.09947, over 5712469.36 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3505, pruned_loss=0.09357, over 4628337.37 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3501, pruned_loss=0.1007, over 5699974.14 frames. ], batch size: 128, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:34:24,640 INFO [train.py:968] (1/2) Epoch 15, batch 3250, giga_loss[loss=0.2714, simple_loss=0.349, pruned_loss=0.09683, over 28435.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3526, pruned_loss=0.101, over 5714261.30 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3507, pruned_loss=0.09373, over 4662211.66 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3527, pruned_loss=0.102, over 5702526.55 frames. ], batch size: 71, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:34:29,936 INFO [optim.py:369] (1/2) 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 15:34:47,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:35:04,351 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 3300, giga_loss[loss=0.2906, simple_loss=0.3476, pruned_loss=0.1168, over 23674.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.354, pruned_loss=0.1022, over 5705217.15 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3508, pruned_loss=0.09374, over 4668151.60 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3541, pruned_loss=0.1031, over 5695390.38 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:35:13,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4515, 2.1047, 1.4794, 0.5666], device='cuda:1'), covar=tensor([0.4653, 0.2330, 0.3726, 0.5339], device='cuda:1'), in_proj_covar=tensor([0.1622, 0.1539, 0.1523, 0.1326], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 15:35:41,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5252, 1.6162, 1.5958, 1.3252], device='cuda:1'), covar=tensor([0.1964, 0.1955, 0.1449, 0.1837], device='cuda:1'), in_proj_covar=tensor([0.1783, 0.1700, 0.1661, 0.1774], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 15:35:44,249 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 3350, libri_loss[loss=0.2328, simple_loss=0.3232, pruned_loss=0.07118, over 29553.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3543, pruned_loss=0.1029, over 5712336.84 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3504, pruned_loss=0.09336, over 4698777.94 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3547, pruned_loss=0.1041, over 5699996.67 frames. ], batch size: 79, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:35:56,554 INFO [optim.py:369] (1/2) 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 15:36:23,792 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:36:34,611 INFO [train.py:968] (1/2) Epoch 15, batch 3400, giga_loss[loss=0.2861, simple_loss=0.3539, pruned_loss=0.1091, over 28558.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.356, pruned_loss=0.1043, over 5719575.06 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3507, pruned_loss=0.09349, over 4722715.37 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3562, pruned_loss=0.1054, over 5706432.63 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:36:58,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642347.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 15:37:08,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8375, 2.0845, 1.6754, 2.1822], device='cuda:1'), covar=tensor([0.2541, 0.2518, 0.2863, 0.2431], device='cuda:1'), in_proj_covar=tensor([0.1374, 0.1009, 0.1216, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 15:37:14,098 INFO [train.py:968] (1/2) Epoch 15, batch 3450, giga_loss[loss=0.2833, simple_loss=0.3582, pruned_loss=0.1041, over 28945.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.355, pruned_loss=0.1037, over 5718220.53 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3503, pruned_loss=0.09329, over 4751086.69 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3556, pruned_loss=0.105, over 5712817.83 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:37:18,774 INFO [optim.py:369] (1/2) 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 15:37:22,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5902, 1.7904, 1.8476, 1.4018], device='cuda:1'), covar=tensor([0.1519, 0.2465, 0.1331, 0.1665], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0694, 0.0903, 0.0805], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 15:37:31,809 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:37:38,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642403.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 15:37:40,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:37:51,634 INFO [train.py:968] (1/2) Epoch 15, batch 3500, giga_loss[loss=0.2779, simple_loss=0.3352, pruned_loss=0.1103, over 23936.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3556, pruned_loss=0.1037, over 5718325.71 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3501, pruned_loss=0.09332, over 4804554.34 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3564, pruned_loss=0.1052, over 5708415.13 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:38:02,721 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642435.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 15:38:22,169 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-07 15:38:28,251 INFO [train.py:968] (1/2) Epoch 15, batch 3550, giga_loss[loss=0.2811, simple_loss=0.3639, pruned_loss=0.09915, over 29025.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3557, pruned_loss=0.1028, over 5722715.78 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3502, pruned_loss=0.09353, over 4841774.62 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3565, pruned_loss=0.1043, over 5709503.56 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:38:33,752 INFO [optim.py:369] (1/2) 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 15:38:46,734 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642490.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:38:49,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642493.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:39:10,165 INFO [train.py:968] (1/2) Epoch 15, batch 3600, giga_loss[loss=0.2755, simple_loss=0.3552, pruned_loss=0.09792, over 28591.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3561, pruned_loss=0.1022, over 5717617.01 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.35, pruned_loss=0.0934, over 4846095.45 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3569, pruned_loss=0.1035, over 5714486.07 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:39:11,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 15:39:12,280 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642522.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 15:39:26,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4755, 3.3123, 1.5379, 1.6546], device='cuda:1'), covar=tensor([0.0932, 0.0266, 0.0859, 0.1279], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0517, 0.0354, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 15:39:31,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4447, 1.6444, 1.6989, 1.2703], device='cuda:1'), covar=tensor([0.1665, 0.2410, 0.1365, 0.1651], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0695, 0.0904, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 15:39:47,174 INFO [train.py:968] (1/2) Epoch 15, batch 3650, giga_loss[loss=0.2796, simple_loss=0.3531, pruned_loss=0.103, over 28620.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3544, pruned_loss=0.1014, over 5716814.04 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3495, pruned_loss=0.09313, over 4858229.11 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3555, pruned_loss=0.1027, over 5715129.87 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:39:48,563 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:40:05,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:40:17,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5501, 3.6050, 1.6627, 1.5742], device='cuda:1'), covar=tensor([0.0940, 0.0276, 0.0861, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0516, 0.0353, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 15:40:28,438 INFO [train.py:968] (1/2) Epoch 15, batch 3700, giga_loss[loss=0.2704, simple_loss=0.3505, pruned_loss=0.0951, over 28906.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3523, pruned_loss=0.1007, over 5713893.43 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3496, pruned_loss=0.0931, over 4865008.48 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3531, pruned_loss=0.1018, over 5714187.52 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:40:32,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1711, 2.3744, 1.2982, 1.3242], device='cuda:1'), covar=tensor([0.0994, 0.0342, 0.0853, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0516, 0.0353, 0.0401], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 15:40:43,758 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:40:49,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1824, 1.3668, 4.0215, 3.1984], device='cuda:1'), covar=tensor([0.1697, 0.2605, 0.0386, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0696, 0.0606, 0.0886, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 15:41:06,744 INFO [train.py:968] (1/2) Epoch 15, batch 3750, giga_loss[loss=0.265, simple_loss=0.3409, pruned_loss=0.09455, over 28627.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3504, pruned_loss=0.09959, over 5722443.73 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3491, pruned_loss=0.09274, over 4905648.67 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3516, pruned_loss=0.1011, over 5716168.40 frames. ], batch size: 336, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:41:10,538 INFO [optim.py:369] (1/2) 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 15:41:18,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=642683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:41:31,447 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:41:47,396 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:41:49,723 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 3800, giga_loss[loss=0.288, simple_loss=0.3587, pruned_loss=0.1087, over 28672.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3499, pruned_loss=0.09921, over 5728693.68 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3491, pruned_loss=0.09268, over 4908908.43 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3508, pruned_loss=0.1004, over 5724429.69 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:42:01,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5820, 2.0607, 1.3523, 0.7805], device='cuda:1'), covar=tensor([0.4827, 0.2598, 0.2393, 0.4893], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1536, 0.1521, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 15:42:01,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:03,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:13,930 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:16,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-07 15:42:27,296 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:42:30,523 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 3850, giga_loss[loss=0.2831, simple_loss=0.3556, pruned_loss=0.1053, over 28982.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1004, over 5725907.99 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3487, pruned_loss=0.09251, over 4929185.88 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3522, pruned_loss=0.1017, over 5722460.67 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:42:35,539 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 3900, giga_loss[loss=0.2877, simple_loss=0.3604, pruned_loss=0.1076, over 28955.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3519, pruned_loss=0.1005, over 5725826.14 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3487, pruned_loss=0.09248, over 4948396.96 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3528, pruned_loss=0.1017, over 5719867.03 frames. ], batch size: 186, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:43:16,519 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:18,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=642829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:43,170 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=642857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:43,953 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:43:46,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-07 15:43:52,538 INFO [train.py:968] (1/2) Epoch 15, batch 3950, giga_loss[loss=0.3259, simple_loss=0.3816, pruned_loss=0.1351, over 28908.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3497, pruned_loss=0.0985, over 5724359.96 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3478, pruned_loss=0.0921, over 4970481.18 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3512, pruned_loss=0.0999, over 5716873.62 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:43:58,108 INFO [optim.py:369] (1/2) 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 15:44:18,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2313, 0.8115, 0.9101, 1.3875], device='cuda:1'), covar=tensor([0.0780, 0.0366, 0.0359, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 15:44:25,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=642911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:44:28,500 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 4000, giga_loss[loss=0.266, simple_loss=0.343, pruned_loss=0.09455, over 28913.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.09928, over 5717113.84 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3483, pruned_loss=0.09238, over 4980760.81 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3511, pruned_loss=0.1003, over 5718988.05 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:44:47,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8770, 1.9274, 1.5316, 1.5980], device='cuda:1'), covar=tensor([0.0925, 0.0749, 0.0957, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0439, 0.0506, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 15:44:50,991 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=642943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:45:12,992 INFO [train.py:968] (1/2) Epoch 15, batch 4050, giga_loss[loss=0.2565, simple_loss=0.3304, pruned_loss=0.09129, over 28562.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.09843, over 5709197.94 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3482, pruned_loss=0.09228, over 4985428.00 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3485, pruned_loss=0.09936, over 5709661.53 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:45:17,375 INFO [optim.py:369] (1/2) 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 15:45:31,440 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-07 15:45:40,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:45:45,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:45:47,405 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-07 15:45:50,294 INFO [train.py:968] (1/2) Epoch 15, batch 4100, giga_loss[loss=0.3151, simple_loss=0.3889, pruned_loss=0.1207, over 28348.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3449, pruned_loss=0.09687, over 5706834.03 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3484, pruned_loss=0.09248, over 5008464.97 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.0976, over 5705565.06 frames. ], batch size: 368, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:46:16,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2308, 0.8319, 0.8828, 1.3041], device='cuda:1'), covar=tensor([0.0756, 0.0376, 0.0355, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 15:46:20,715 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 4150, giga_loss[loss=0.2884, simple_loss=0.3546, pruned_loss=0.1111, over 28699.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3443, pruned_loss=0.0971, over 5706959.66 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3486, pruned_loss=0.09257, over 5025887.16 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3444, pruned_loss=0.09768, over 5702926.32 frames. ], batch size: 284, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:46:31,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:46:33,792 INFO [optim.py:369] (1/2) 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 15:47:06,795 INFO [train.py:968] (1/2) Epoch 15, batch 4200, giga_loss[loss=0.2506, simple_loss=0.329, pruned_loss=0.08609, over 28722.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3424, pruned_loss=0.09619, over 5708005.30 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3485, pruned_loss=0.0925, over 5056375.71 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3424, pruned_loss=0.09687, over 5701567.54 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:47:27,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3019, 4.1228, 3.9288, 1.9815], device='cuda:1'), covar=tensor([0.0566, 0.0713, 0.0747, 0.2087], device='cuda:1'), in_proj_covar=tensor([0.1101, 0.1026, 0.0888, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 15:47:35,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2818, 1.5147, 1.5737, 1.4471], device='cuda:1'), covar=tensor([0.1702, 0.1582, 0.2152, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0733, 0.0690, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 15:47:36,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:47:38,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:47:41,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3364, 1.5214, 1.5754, 1.2488], device='cuda:1'), covar=tensor([0.1292, 0.1826, 0.1078, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0696, 0.0904, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 15:47:48,720 INFO [train.py:968] (1/2) Epoch 15, batch 4250, giga_loss[loss=0.2608, simple_loss=0.3326, pruned_loss=0.09444, over 29027.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3408, pruned_loss=0.096, over 5707698.64 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3485, pruned_loss=0.09261, over 5068491.58 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3407, pruned_loss=0.0965, over 5700387.25 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:47:57,754 INFO [optim.py:369] (1/2) 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 15:48:05,206 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:27,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:29,407 INFO [train.py:968] (1/2) Epoch 15, batch 4300, giga_loss[loss=0.2967, simple_loss=0.3619, pruned_loss=0.1157, over 28247.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3381, pruned_loss=0.09492, over 5712581.87 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3478, pruned_loss=0.09225, over 5088222.26 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3382, pruned_loss=0.09567, over 5706385.71 frames. ], batch size: 368, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:48:29,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:29,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4419, 2.1086, 1.5544, 0.6698], device='cuda:1'), covar=tensor([0.6011, 0.2705, 0.3634, 0.5992], device='cuda:1'), in_proj_covar=tensor([0.1625, 0.1531, 0.1520, 0.1323], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 15:48:39,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:48:52,547 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:49:08,133 INFO [train.py:968] (1/2) Epoch 15, batch 4350, giga_loss[loss=0.3084, simple_loss=0.3656, pruned_loss=0.1256, over 23642.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3365, pruned_loss=0.09469, over 5692294.56 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3478, pruned_loss=0.09221, over 5090661.25 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3364, pruned_loss=0.09538, over 5700449.43 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:49:14,362 INFO [optim.py:369] (1/2) 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 15:49:28,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-07 15:49:47,139 INFO [train.py:968] (1/2) Epoch 15, batch 4400, giga_loss[loss=0.3006, simple_loss=0.3583, pruned_loss=0.1214, over 29021.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3349, pruned_loss=0.09352, over 5707034.13 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3482, pruned_loss=0.09241, over 5110772.66 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3342, pruned_loss=0.09394, over 5709055.84 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:50:05,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5529, 4.0590, 1.6762, 1.6455], device='cuda:1'), covar=tensor([0.0872, 0.0291, 0.0880, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0516, 0.0352, 0.0400], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 15:50:29,129 INFO [train.py:968] (1/2) Epoch 15, batch 4450, giga_loss[loss=0.2612, simple_loss=0.3418, pruned_loss=0.09026, over 28919.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3356, pruned_loss=0.09361, over 5707259.38 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3483, pruned_loss=0.09245, over 5118145.89 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3348, pruned_loss=0.09392, over 5707655.36 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:50:34,677 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:50:35,678 INFO [optim.py:369] (1/2) 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 15:50:36,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:50:40,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:50:51,246 INFO [scaling.py:679] (1/2) 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] (1/2) 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] (1/2) Epoch 15, batch 4500, giga_loss[loss=0.286, simple_loss=0.3604, pruned_loss=0.1058, over 28986.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3396, pruned_loss=0.09584, over 5701048.95 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3485, pruned_loss=0.09247, over 5133171.66 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3386, pruned_loss=0.09611, over 5698520.24 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:51:25,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:51:52,894 INFO [train.py:968] (1/2) Epoch 15, batch 4550, giga_loss[loss=0.2754, simple_loss=0.3523, pruned_loss=0.09923, over 28836.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.343, pruned_loss=0.09714, over 5704484.37 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.349, pruned_loss=0.09281, over 5139621.07 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3418, pruned_loss=0.0971, over 5701664.56 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:51:59,373 INFO [optim.py:369] (1/2) 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 15:52:07,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-07 15:52:10,299 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:52:32,922 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 4600, giga_loss[loss=0.3331, simple_loss=0.3862, pruned_loss=0.14, over 26680.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3458, pruned_loss=0.09839, over 5697256.61 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3494, pruned_loss=0.09321, over 5155184.95 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3444, pruned_loss=0.09812, over 5690932.05 frames. ], batch size: 555, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:52:42,914 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:52:44,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:08,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:17,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-07 15:53:18,610 INFO [train.py:968] (1/2) Epoch 15, batch 4650, giga_loss[loss=0.2706, simple_loss=0.3539, pruned_loss=0.09366, over 27958.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3452, pruned_loss=0.09751, over 5698854.53 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3498, pruned_loss=0.0936, over 5175171.00 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.0971, over 5690449.11 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:53:25,642 INFO [optim.py:369] (1/2) 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 15:53:26,056 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=643577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:28,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=643580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:53:36,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6303, 1.8252, 1.9171, 1.4203], device='cuda:1'), covar=tensor([0.1964, 0.2282, 0.1571, 0.1769], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0692, 0.0900, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 15:53:53,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=643609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:54:00,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-07 15:54:01,136 INFO [train.py:968] (1/2) Epoch 15, batch 4700, giga_loss[loss=0.2417, simple_loss=0.3179, pruned_loss=0.08279, over 28487.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3433, pruned_loss=0.09653, over 5701908.25 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3498, pruned_loss=0.09354, over 5176850.14 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.342, pruned_loss=0.09627, over 5696176.87 frames. ], batch size: 60, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:54:39,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5023, 1.7039, 1.5226, 1.2644], device='cuda:1'), covar=tensor([0.2989, 0.2076, 0.1933, 0.2605], device='cuda:1'), in_proj_covar=tensor([0.1790, 0.1709, 0.1665, 0.1771], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 15:54:44,486 INFO [train.py:968] (1/2) Epoch 15, batch 4750, libri_loss[loss=0.308, simple_loss=0.3895, pruned_loss=0.1132, over 29243.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3439, pruned_loss=0.09692, over 5699884.92 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3496, pruned_loss=0.09326, over 5190217.86 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3429, pruned_loss=0.09701, over 5692523.94 frames. ], batch size: 94, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:54:50,236 INFO [optim.py:369] (1/2) 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 15:55:02,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0128, 3.1771, 2.0009, 0.8373], device='cuda:1'), covar=tensor([0.6077, 0.2398, 0.3796, 0.6683], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1529, 0.1523, 0.1322], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 15:55:21,051 INFO [train.py:968] (1/2) Epoch 15, batch 4800, giga_loss[loss=0.2686, simple_loss=0.3448, pruned_loss=0.09622, over 28888.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3454, pruned_loss=0.09803, over 5697591.20 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3499, pruned_loss=0.09344, over 5203671.05 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3443, pruned_loss=0.09807, over 5691111.92 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:55:52,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2032, 1.5533, 1.1771, 1.3168], device='cuda:1'), covar=tensor([0.2637, 0.2558, 0.3067, 0.2267], device='cuda:1'), in_proj_covar=tensor([0.1374, 0.1010, 0.1218, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 15:56:03,252 INFO [train.py:968] (1/2) Epoch 15, batch 4850, giga_loss[loss=0.3171, simple_loss=0.3914, pruned_loss=0.1214, over 28853.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3488, pruned_loss=0.09993, over 5701580.90 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3498, pruned_loss=0.09346, over 5223167.19 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.348, pruned_loss=0.1001, over 5691855.09 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:56:08,855 INFO [optim.py:369] (1/2) 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 15:56:42,162 INFO [train.py:968] (1/2) Epoch 15, batch 4900, giga_loss[loss=0.3123, simple_loss=0.3789, pruned_loss=0.1229, over 27654.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3517, pruned_loss=0.1011, over 5704855.65 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3502, pruned_loss=0.09373, over 5230104.11 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3507, pruned_loss=0.1012, over 5701394.34 frames. ], batch size: 472, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:57:19,080 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 4950, giga_loss[loss=0.2683, simple_loss=0.3441, pruned_loss=0.09624, over 28504.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3517, pruned_loss=0.1007, over 5708579.72 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3499, pruned_loss=0.09354, over 5241548.26 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3511, pruned_loss=0.1011, over 5704052.37 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 15:57:30,479 INFO [optim.py:369] (1/2) 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 15:57:36,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=643886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:58:03,594 INFO [train.py:968] (1/2) Epoch 15, batch 5000, giga_loss[loss=0.2965, simple_loss=0.3637, pruned_loss=0.1146, over 28874.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3528, pruned_loss=0.1012, over 5716764.02 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3501, pruned_loss=0.09367, over 5248304.10 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3522, pruned_loss=0.1015, over 5711398.77 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:58:07,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:58:38,779 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=643959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:58:45,055 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-07 15:58:45,843 INFO [train.py:968] (1/2) Epoch 15, batch 5050, giga_loss[loss=0.2585, simple_loss=0.3355, pruned_loss=0.09072, over 28921.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3532, pruned_loss=0.1018, over 5724405.19 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3505, pruned_loss=0.09383, over 5264532.87 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3525, pruned_loss=0.102, over 5715877.50 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:58:53,729 INFO [optim.py:369] (1/2) 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 15:59:14,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2710, 1.4243, 1.2564, 1.5080], device='cuda:1'), covar=tensor([0.0714, 0.0341, 0.0340, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0062, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 15:59:16,460 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:19,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:26,793 INFO [train.py:968] (1/2) Epoch 15, batch 5100, giga_loss[loss=0.312, simple_loss=0.3869, pruned_loss=0.1185, over 28789.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3513, pruned_loss=0.1008, over 5722545.18 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3507, pruned_loss=0.09384, over 5276806.52 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3506, pruned_loss=0.1011, over 5712917.23 frames. ], batch size: 243, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 15:59:35,817 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:37,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 15:59:41,776 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:00:08,084 INFO [train.py:968] (1/2) Epoch 15, batch 5150, giga_loss[loss=0.2648, simple_loss=0.3236, pruned_loss=0.103, over 28797.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09973, over 5730150.86 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3506, pruned_loss=0.09383, over 5295518.03 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3481, pruned_loss=0.1002, over 5717478.98 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:00:14,928 INFO [optim.py:369] (1/2) 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 16:00:19,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4789, 2.1967, 1.6638, 0.7707], device='cuda:1'), covar=tensor([0.5203, 0.2600, 0.3780, 0.5600], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1529, 0.1517, 0.1320], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 16:00:47,633 INFO [train.py:968] (1/2) Epoch 15, batch 5200, giga_loss[loss=0.2616, simple_loss=0.3297, pruned_loss=0.0968, over 28564.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3447, pruned_loss=0.09752, over 5732717.55 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3503, pruned_loss=0.09373, over 5307477.54 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3445, pruned_loss=0.09807, over 5719701.47 frames. ], batch size: 78, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:01:17,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2504, 1.8554, 1.3280, 0.4908], device='cuda:1'), covar=tensor([0.3838, 0.1915, 0.2969, 0.4889], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1525, 0.1514, 0.1317], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 16:01:27,564 INFO [train.py:968] (1/2) Epoch 15, batch 5250, giga_loss[loss=0.2597, simple_loss=0.3454, pruned_loss=0.08704, over 28924.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3458, pruned_loss=0.09808, over 5731646.22 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3506, pruned_loss=0.09393, over 5325446.87 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3453, pruned_loss=0.09848, over 5716168.06 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:01:36,212 INFO [optim.py:369] (1/2) 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 16:02:01,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3867, 1.5988, 1.4519, 1.2496], device='cuda:1'), covar=tensor([0.2785, 0.2123, 0.1968, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.1794, 0.1722, 0.1675, 0.1785], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 16:02:07,400 INFO [train.py:968] (1/2) Epoch 15, batch 5300, giga_loss[loss=0.2414, simple_loss=0.3173, pruned_loss=0.08275, over 28498.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3474, pruned_loss=0.0977, over 5721552.85 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3506, pruned_loss=0.09409, over 5344155.63 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3468, pruned_loss=0.09802, over 5706315.36 frames. ], batch size: 60, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:02:49,981 INFO [train.py:968] (1/2) Epoch 15, batch 5350, giga_loss[loss=0.3457, simple_loss=0.3954, pruned_loss=0.1481, over 26884.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.348, pruned_loss=0.09793, over 5705609.35 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3515, pruned_loss=0.09474, over 5344177.01 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3468, pruned_loss=0.09771, over 5699600.50 frames. ], batch size: 555, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:02:58,746 INFO [optim.py:369] (1/2) 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 16:03:13,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:03:25,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644312.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:03:31,226 INFO [train.py:968] (1/2) Epoch 15, batch 5400, giga_loss[loss=0.2783, simple_loss=0.3516, pruned_loss=0.1025, over 28728.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09872, over 5710418.12 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3511, pruned_loss=0.09462, over 5352677.00 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3467, pruned_loss=0.0987, over 5702869.38 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:03:43,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3172, 1.5019, 1.4225, 1.3027], device='cuda:1'), covar=tensor([0.2195, 0.1873, 0.1898, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.1800, 0.1722, 0.1674, 0.1784], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 16:03:45,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:04:14,804 INFO [train.py:968] (1/2) Epoch 15, batch 5450, giga_loss[loss=0.2645, simple_loss=0.3298, pruned_loss=0.09956, over 28879.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3457, pruned_loss=0.09913, over 5709622.67 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3513, pruned_loss=0.09482, over 5366513.74 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3448, pruned_loss=0.09906, over 5698914.95 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:04:23,123 INFO [optim.py:369] (1/2) 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 16:04:57,573 INFO [train.py:968] (1/2) Epoch 15, batch 5500, giga_loss[loss=0.2851, simple_loss=0.3448, pruned_loss=0.1128, over 24125.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3441, pruned_loss=0.09898, over 5706999.96 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3513, pruned_loss=0.09474, over 5374487.72 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3434, pruned_loss=0.09904, over 5696096.66 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:05:12,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0074, 1.1284, 3.3715, 2.9907], device='cuda:1'), covar=tensor([0.1682, 0.2659, 0.0484, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0605, 0.0884, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 16:05:16,397 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:20,441 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 5550, giga_loss[loss=0.2875, simple_loss=0.3543, pruned_loss=0.1103, over 28676.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3421, pruned_loss=0.09863, over 5708861.95 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3514, pruned_loss=0.09479, over 5376870.29 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3414, pruned_loss=0.09866, over 5699740.22 frames. ], batch size: 242, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:05:42,645 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:44,763 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:05:48,347 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:05:50,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:06:16,924 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=644509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:06:24,146 INFO [train.py:968] (1/2) Epoch 15, batch 5600, giga_loss[loss=0.2529, simple_loss=0.329, pruned_loss=0.08838, over 28676.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3403, pruned_loss=0.09721, over 5717523.54 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3515, pruned_loss=0.09473, over 5389793.67 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3395, pruned_loss=0.09735, over 5706334.89 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:06:29,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1846, 1.1824, 0.9995, 1.4540], device='cuda:1'), covar=tensor([0.0787, 0.0369, 0.0366, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0063, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 16:06:55,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4276, 3.2721, 1.4629, 1.5364], device='cuda:1'), covar=tensor([0.0920, 0.0324, 0.0935, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0523, 0.0355, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 16:07:04,645 INFO [train.py:968] (1/2) Epoch 15, batch 5650, libri_loss[loss=0.2817, simple_loss=0.3618, pruned_loss=0.1008, over 29517.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3371, pruned_loss=0.09537, over 5726956.39 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.352, pruned_loss=0.09509, over 5410333.48 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3354, pruned_loss=0.09522, over 5712678.03 frames. ], batch size: 82, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:07:12,793 INFO [optim.py:369] (1/2) 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 16:07:43,269 INFO [train.py:968] (1/2) Epoch 15, batch 5700, giga_loss[loss=0.2288, simple_loss=0.3043, pruned_loss=0.07661, over 29054.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3326, pruned_loss=0.09335, over 5718813.23 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.352, pruned_loss=0.09528, over 5409989.41 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.331, pruned_loss=0.09306, over 5713543.31 frames. ], batch size: 136, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:08:10,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 16:08:24,525 INFO [train.py:968] (1/2) Epoch 15, batch 5750, giga_loss[loss=0.2561, simple_loss=0.3304, pruned_loss=0.09083, over 28967.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3325, pruned_loss=0.09346, over 5719184.40 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3518, pruned_loss=0.09522, over 5417347.83 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3311, pruned_loss=0.09325, over 5712706.29 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:08:32,275 INFO [optim.py:369] (1/2) 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 16:08:36,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644687.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:08:52,000 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-07 16:09:03,109 INFO [train.py:968] (1/2) Epoch 15, batch 5800, giga_loss[loss=0.2527, simple_loss=0.3344, pruned_loss=0.08552, over 28615.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3357, pruned_loss=0.09475, over 5724400.73 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3521, pruned_loss=0.09551, over 5425498.37 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3341, pruned_loss=0.09433, over 5717333.87 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:09:43,401 INFO [train.py:968] (1/2) Epoch 15, batch 5850, giga_loss[loss=0.2588, simple_loss=0.328, pruned_loss=0.09476, over 28619.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3394, pruned_loss=0.09588, over 5721898.43 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3526, pruned_loss=0.09576, over 5431467.96 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3372, pruned_loss=0.09532, over 5720315.03 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:09:51,984 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 5900, giga_loss[loss=0.2881, simple_loss=0.3575, pruned_loss=0.1094, over 28902.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3429, pruned_loss=0.09739, over 5715508.01 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3526, pruned_loss=0.09579, over 5443610.84 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3408, pruned_loss=0.09692, over 5711952.79 frames. ], batch size: 186, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:10:33,815 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:10:38,974 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644833.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:10:47,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=644844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:11:02,475 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 5950, giga_loss[loss=0.2786, simple_loss=0.3603, pruned_loss=0.09846, over 28804.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3461, pruned_loss=0.09882, over 5720481.52 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3529, pruned_loss=0.09601, over 5457763.08 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3438, pruned_loss=0.09833, over 5713108.19 frames. ], batch size: 284, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:11:14,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-07 16:11:17,106 INFO [optim.py:369] (1/2) 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 16:11:47,266 INFO [train.py:968] (1/2) Epoch 15, batch 6000, giga_loss[loss=0.2678, simple_loss=0.3384, pruned_loss=0.09862, over 28430.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3478, pruned_loss=0.09997, over 5703689.37 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3527, pruned_loss=0.09618, over 5456204.33 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3459, pruned_loss=0.09955, over 5709187.93 frames. ], batch size: 71, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:11:47,266 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 16:11:55,558 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 16:12:18,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:12:21,282 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=644947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:12:40,078 INFO [train.py:968] (1/2) Epoch 15, batch 6050, giga_loss[loss=0.2866, simple_loss=0.362, pruned_loss=0.1057, over 28888.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3535, pruned_loss=0.105, over 5696976.11 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3525, pruned_loss=0.09612, over 5461914.47 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.352, pruned_loss=0.1049, over 5701811.98 frames. ], batch size: 174, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:12:45,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 16:12:47,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3176, 5.1087, 4.8830, 2.6701], device='cuda:1'), covar=tensor([0.0440, 0.0658, 0.0652, 0.1610], device='cuda:1'), in_proj_covar=tensor([0.1114, 0.1037, 0.0896, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 16:12:52,617 INFO [optim.py:369] (1/2) 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 16:12:58,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=644987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:13:02,262 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=644990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:13:12,884 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645003.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:13:27,790 INFO [train.py:968] (1/2) Epoch 15, batch 6100, giga_loss[loss=0.3349, simple_loss=0.3982, pruned_loss=0.1357, over 28652.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3582, pruned_loss=0.1091, over 5691728.18 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3526, pruned_loss=0.09625, over 5470898.33 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3571, pruned_loss=0.1091, over 5693839.18 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:13:28,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:13:34,612 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 16:13:53,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8320, 1.7539, 1.3383, 1.4159], device='cuda:1'), covar=tensor([0.0765, 0.0639, 0.0983, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0438, 0.0500, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 16:14:13,781 INFO [train.py:968] (1/2) Epoch 15, batch 6150, libri_loss[loss=0.2922, simple_loss=0.3528, pruned_loss=0.1158, over 29564.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.364, pruned_loss=0.1134, over 5678688.69 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3523, pruned_loss=0.09627, over 5481953.20 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3637, pruned_loss=0.1139, over 5676676.43 frames. ], batch size: 76, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:14:23,671 INFO [optim.py:369] (1/2) 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 16:14:59,146 INFO [train.py:968] (1/2) Epoch 15, batch 6200, libri_loss[loss=0.2707, simple_loss=0.3599, pruned_loss=0.09075, over 29532.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3699, pruned_loss=0.1177, over 5679198.09 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3526, pruned_loss=0.09633, over 5498448.90 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.37, pruned_loss=0.119, over 5671311.55 frames. ], batch size: 83, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:15:26,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 16:15:35,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 16:15:41,590 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:15:43,554 INFO [train.py:968] (1/2) Epoch 15, batch 6250, giga_loss[loss=0.3396, simple_loss=0.3967, pruned_loss=0.1413, over 28814.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3751, pruned_loss=0.1226, over 5686411.25 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3526, pruned_loss=0.09634, over 5507986.88 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3756, pruned_loss=0.1242, over 5675620.58 frames. ], batch size: 66, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:15:57,786 INFO [optim.py:369] (1/2) 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 16:16:07,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 16:16:31,034 INFO [train.py:968] (1/2) Epoch 15, batch 6300, giga_loss[loss=0.4781, simple_loss=0.4866, pruned_loss=0.2348, over 26686.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3804, pruned_loss=0.1268, over 5676935.16 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3529, pruned_loss=0.09625, over 5517887.48 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3814, pruned_loss=0.129, over 5663907.03 frames. ], batch size: 555, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:16:37,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5544, 4.2973, 1.6477, 1.7479], device='cuda:1'), covar=tensor([0.0911, 0.0323, 0.0820, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0523, 0.0354, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 16:16:44,257 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-07 16:17:21,463 INFO [train.py:968] (1/2) Epoch 15, batch 6350, giga_loss[loss=0.3753, simple_loss=0.4179, pruned_loss=0.1663, over 27922.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3819, pruned_loss=0.1286, over 5660007.78 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3528, pruned_loss=0.09615, over 5519959.02 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3836, pruned_loss=0.1315, over 5651237.27 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 2.0
+2023-03-07 16:17:34,610 INFO [optim.py:369] (1/2) 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 16:18:05,628 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.09 vs. limit=2.0
+2023-03-07 16:18:15,857 INFO [train.py:968] (1/2) Epoch 15, batch 6400, giga_loss[loss=0.4198, simple_loss=0.4502, pruned_loss=0.1947, over 27544.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3846, pruned_loss=0.1323, over 5643707.52 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.353, pruned_loss=0.09629, over 5523918.04 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.3861, pruned_loss=0.1348, over 5634081.03 frames. ], batch size: 472, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:18:18,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:18:19,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:18:33,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5165, 1.7087, 1.5885, 1.5382], device='cuda:1'), covar=tensor([0.1389, 0.1497, 0.1726, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0734, 0.0693, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 16:19:08,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645368.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:19:08,643 INFO [train.py:968] (1/2) Epoch 15, batch 6450, giga_loss[loss=0.4251, simple_loss=0.4391, pruned_loss=0.2056, over 28257.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.388, pruned_loss=0.1365, over 5627081.46 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3526, pruned_loss=0.09612, over 5535260.03 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3906, pruned_loss=0.1397, over 5612061.31 frames. ], batch size: 368, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:19:18,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645378.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:19:23,984 INFO [optim.py:369] (1/2) 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 16:19:29,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3660, 1.6807, 1.3900, 1.6593], device='cuda:1'), covar=tensor([0.0743, 0.0298, 0.0318, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0114, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0063, 0.0056, 0.0095], device='cuda:1')
+2023-03-07 16:20:01,595 INFO [train.py:968] (1/2) Epoch 15, batch 6500, giga_loss[loss=0.3731, simple_loss=0.4219, pruned_loss=0.1621, over 28914.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.392, pruned_loss=0.14, over 5622709.34 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3522, pruned_loss=0.09595, over 5540612.96 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.395, pruned_loss=0.1435, over 5607746.87 frames. ], batch size: 174, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:20:03,910 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5063, 1.7422, 1.4408, 1.5469], device='cuda:1'), covar=tensor([0.2225, 0.2101, 0.2214, 0.1780], device='cuda:1'), in_proj_covar=tensor([0.1371, 0.1007, 0.1216, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 16:20:16,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-07 16:20:49,396 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:51,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:52,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:53,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:20:54,269 INFO [train.py:968] (1/2) Epoch 15, batch 6550, giga_loss[loss=0.3326, simple_loss=0.3788, pruned_loss=0.1432, over 28758.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3929, pruned_loss=0.141, over 5624867.33 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3523, pruned_loss=0.09603, over 5535154.13 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3956, pruned_loss=0.1442, over 5618833.08 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:21:08,881 INFO [optim.py:369] (1/2) 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 16:21:20,510 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:21:21,936 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:21:28,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5297, 1.7905, 1.6086, 1.3376], device='cuda:1'), covar=tensor([0.2596, 0.2007, 0.1712, 0.2249], device='cuda:1'), in_proj_covar=tensor([0.1824, 0.1749, 0.1698, 0.1803], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 16:21:43,795 INFO [train.py:968] (1/2) Epoch 15, batch 6600, giga_loss[loss=0.3241, simple_loss=0.3917, pruned_loss=0.1283, over 28915.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.391, pruned_loss=0.1403, over 5627095.15 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3523, pruned_loss=0.09593, over 5534707.04 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3945, pruned_loss=0.1443, over 5625004.82 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:21:47,330 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645521.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:21:49,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645524.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:22:07,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:22:21,468 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645553.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:22:36,188 INFO [train.py:968] (1/2) Epoch 15, batch 6650, giga_loss[loss=0.3132, simple_loss=0.3826, pruned_loss=0.122, over 28788.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.39, pruned_loss=0.1396, over 5626711.73 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3519, pruned_loss=0.09591, over 5542039.00 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3937, pruned_loss=0.1437, over 5620099.84 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:22:50,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-07 16:22:51,016 INFO [optim.py:369] (1/2) 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 16:23:27,044 INFO [train.py:968] (1/2) Epoch 15, batch 6700, giga_loss[loss=0.3406, simple_loss=0.4027, pruned_loss=0.1392, over 28715.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3902, pruned_loss=0.1385, over 5636535.97 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3519, pruned_loss=0.0958, over 5545548.30 frames. ], giga_tot_loss[loss=0.339, simple_loss=0.3936, pruned_loss=0.1422, over 5628898.03 frames. ], batch size: 262, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:23:54,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:24:18,714 INFO [train.py:968] (1/2) Epoch 15, batch 6750, giga_loss[loss=0.3521, simple_loss=0.385, pruned_loss=0.1596, over 23531.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3912, pruned_loss=0.1391, over 5616423.13 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3519, pruned_loss=0.09587, over 5550944.02 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3944, pruned_loss=0.1426, over 5606713.03 frames. ], batch size: 705, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:24:33,123 INFO [optim.py:369] (1/2) 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 16:24:34,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:24:37,868 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:24:46,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4153, 1.6181, 1.4757, 1.2648], device='cuda:1'), covar=tensor([0.2668, 0.2064, 0.1861, 0.2208], device='cuda:1'), in_proj_covar=tensor([0.1824, 0.1748, 0.1694, 0.1797], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 16:25:05,288 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:25:07,122 INFO [train.py:968] (1/2) Epoch 15, batch 6800, libri_loss[loss=0.2718, simple_loss=0.3512, pruned_loss=0.09621, over 29535.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3895, pruned_loss=0.1379, over 5624381.22 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.351, pruned_loss=0.09544, over 5560692.56 frames. ], giga_tot_loss[loss=0.3394, simple_loss=0.3941, pruned_loss=0.1423, over 5609366.40 frames. ], batch size: 81, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:25:36,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=645743.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:25:58,916 INFO [train.py:968] (1/2) Epoch 15, batch 6850, giga_loss[loss=0.2885, simple_loss=0.3663, pruned_loss=0.1053, over 28739.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3866, pruned_loss=0.1343, over 5631385.71 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3508, pruned_loss=0.09532, over 5571023.55 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3915, pruned_loss=0.1391, over 5611678.18 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:26:14,575 INFO [optim.py:369] (1/2) 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 16:26:48,542 INFO [train.py:968] (1/2) Epoch 15, batch 6900, giga_loss[loss=0.2561, simple_loss=0.3388, pruned_loss=0.08672, over 28694.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.383, pruned_loss=0.13, over 5647402.95 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3507, pruned_loss=0.09537, over 5579393.53 frames. ], giga_tot_loss[loss=0.3285, simple_loss=0.3877, pruned_loss=0.1347, over 5625705.03 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:27:38,608 INFO [train.py:968] (1/2) Epoch 15, batch 6950, giga_loss[loss=0.272, simple_loss=0.3489, pruned_loss=0.09756, over 28803.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3802, pruned_loss=0.1277, over 5653901.37 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3506, pruned_loss=0.09537, over 5584863.03 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3846, pruned_loss=0.1319, over 5633190.93 frames. ], batch size: 86, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:27:43,222 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=645873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:27:53,293 INFO [optim.py:369] (1/2) 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 16:27:54,947 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=645886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 16:27:55,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-07 16:27:57,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=645889.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:28:20,885 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=645918.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 16:28:21,289 INFO [train.py:968] (1/2) Epoch 15, batch 7000, giga_loss[loss=0.2818, simple_loss=0.3531, pruned_loss=0.1053, over 29000.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3771, pruned_loss=0.1252, over 5661648.80 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3506, pruned_loss=0.09553, over 5595938.18 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3818, pruned_loss=0.1298, over 5637457.92 frames. ], batch size: 145, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:28:29,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8392, 5.2208, 2.0173, 2.1557], device='cuda:1'), covar=tensor([0.0922, 0.0213, 0.0846, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0528, 0.0356, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 16:28:41,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2276, 1.3730, 1.3338, 1.2165], device='cuda:1'), covar=tensor([0.2523, 0.1937, 0.1881, 0.2048], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1748, 0.1691, 0.1799], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 16:29:11,859 INFO [train.py:968] (1/2) Epoch 15, batch 7050, giga_loss[loss=0.3361, simple_loss=0.395, pruned_loss=0.1386, over 28885.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.377, pruned_loss=0.1253, over 5663025.88 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3501, pruned_loss=0.09525, over 5600317.89 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3815, pruned_loss=0.1296, over 5640999.90 frames. ], batch size: 199, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:29:26,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 16:29:26,660 INFO [optim.py:369] (1/2) 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 16:29:33,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8487, 2.8657, 1.8138, 0.9694], device='cuda:1'), covar=tensor([0.5424, 0.2417, 0.3202, 0.5602], device='cuda:1'), in_proj_covar=tensor([0.1636, 0.1552, 0.1534, 0.1340], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 16:30:04,544 INFO [train.py:968] (1/2) Epoch 15, batch 7100, giga_loss[loss=0.3145, simple_loss=0.3755, pruned_loss=0.1268, over 28701.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3765, pruned_loss=0.1246, over 5672650.13 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.35, pruned_loss=0.09527, over 5607469.46 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.381, pruned_loss=0.1288, over 5650506.06 frames. ], batch size: 307, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:30:06,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:31:00,062 INFO [train.py:968] (1/2) Epoch 15, batch 7150, giga_loss[loss=0.2944, simple_loss=0.3628, pruned_loss=0.113, over 28922.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3744, pruned_loss=0.1225, over 5675762.47 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3499, pruned_loss=0.09522, over 5613389.81 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3785, pruned_loss=0.1265, over 5654072.11 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:31:15,690 INFO [optim.py:369] (1/2) 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 16:31:33,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1664, 1.2722, 1.0504, 0.8959], device='cuda:1'), covar=tensor([0.0783, 0.0417, 0.0950, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0441, 0.0503, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 16:31:56,991 INFO [train.py:968] (1/2) Epoch 15, batch 7200, giga_loss[loss=0.2949, simple_loss=0.3789, pruned_loss=0.1054, over 28906.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3758, pruned_loss=0.1215, over 5675574.84 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3498, pruned_loss=0.09516, over 5618430.09 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3798, pruned_loss=0.1254, over 5655022.57 frames. ], batch size: 174, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:32:30,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-07 16:32:38,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:32:41,878 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 7250, libri_loss[loss=0.2871, simple_loss=0.3682, pruned_loss=0.103, over 29458.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3762, pruned_loss=0.1204, over 5681999.96 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3495, pruned_loss=0.0951, over 5628885.31 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3807, pruned_loss=0.1245, over 5657653.86 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:33:00,602 INFO [optim.py:369] (1/2) 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 16:33:12,977 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 7300, giga_loss[loss=0.3184, simple_loss=0.3783, pruned_loss=0.1293, over 27973.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3775, pruned_loss=0.1221, over 5679672.57 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3495, pruned_loss=0.0951, over 5628885.31 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.381, pruned_loss=0.1253, over 5660723.69 frames. ], batch size: 412, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:33:46,270 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:34:11,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:34:27,588 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:34:28,845 INFO [train.py:968] (1/2) Epoch 15, batch 7350, giga_loss[loss=0.3049, simple_loss=0.3744, pruned_loss=0.1177, over 28565.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3766, pruned_loss=0.1219, over 5680452.60 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3494, pruned_loss=0.09515, over 5635206.13 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3803, pruned_loss=0.1252, over 5660728.87 frames. ], batch size: 65, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:34:32,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-07 16:34:42,016 INFO [optim.py:369] (1/2) 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 16:35:20,549 INFO [train.py:968] (1/2) Epoch 15, batch 7400, giga_loss[loss=0.2665, simple_loss=0.3312, pruned_loss=0.1009, over 28760.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3746, pruned_loss=0.1221, over 5673871.97 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3494, pruned_loss=0.09513, over 5637859.48 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3778, pruned_loss=0.125, over 5656577.87 frames. ], batch size: 99, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:36:05,950 INFO [train.py:968] (1/2) Epoch 15, batch 7450, giga_loss[loss=0.3162, simple_loss=0.3737, pruned_loss=0.1293, over 28511.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3737, pruned_loss=0.1218, over 5683083.39 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3495, pruned_loss=0.09515, over 5643637.76 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3767, pruned_loss=0.1247, over 5665015.80 frames. ], batch size: 85, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:36:14,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-07 16:36:20,933 INFO [optim.py:369] (1/2) 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 16:36:21,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-07 16:36:28,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:36:30,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=646394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:37:00,326 INFO [train.py:968] (1/2) Epoch 15, batch 7500, giga_loss[loss=0.3148, simple_loss=0.3916, pruned_loss=0.119, over 28923.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3731, pruned_loss=0.1206, over 5693384.00 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3491, pruned_loss=0.09493, over 5647754.38 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3761, pruned_loss=0.1234, over 5676061.68 frames. ], batch size: 174, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:37:03,194 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=646423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:37:50,093 INFO [train.py:968] (1/2) Epoch 15, batch 7550, giga_loss[loss=0.3544, simple_loss=0.399, pruned_loss=0.1548, over 27514.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3719, pruned_loss=0.1187, over 5700199.30 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3486, pruned_loss=0.09452, over 5652623.21 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3752, pruned_loss=0.1218, over 5682987.88 frames. ], batch size: 472, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:38:00,965 INFO [optim.py:369] (1/2) 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 16:38:02,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4837, 4.1254, 1.6244, 1.6504], device='cuda:1'), covar=tensor([0.0966, 0.0376, 0.0920, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0530, 0.0358, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 16:38:24,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3085, 3.1033, 2.9389, 1.4230], device='cuda:1'), covar=tensor([0.0898, 0.1041, 0.0960, 0.2290], device='cuda:1'), in_proj_covar=tensor([0.1126, 0.1047, 0.0905, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 16:38:34,792 INFO [train.py:968] (1/2) Epoch 15, batch 7600, giga_loss[loss=0.2935, simple_loss=0.3792, pruned_loss=0.1039, over 28918.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3713, pruned_loss=0.1183, over 5700414.91 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3481, pruned_loss=0.09433, over 5657345.86 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.375, pruned_loss=0.1214, over 5683484.86 frames. ], batch size: 227, lr: 2.17e-03, grad_scale: 8.0
+2023-03-07 16:39:00,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3273, 1.4706, 1.4420, 1.2350], device='cuda:1'), covar=tensor([0.2116, 0.1866, 0.1372, 0.1820], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1742, 0.1678, 0.1794], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 16:39:15,678 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 7650, giga_loss[loss=0.3088, simple_loss=0.3729, pruned_loss=0.1224, over 28213.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.371, pruned_loss=0.1183, over 5700582.16 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3485, pruned_loss=0.0944, over 5664301.39 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3743, pruned_loss=0.1215, over 5682090.57 frames. ], batch size: 368, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:39:32,659 INFO [optim.py:369] (1/2) 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 16:39:40,894 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:39:46,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:40:05,175 INFO [train.py:968] (1/2) Epoch 15, batch 7700, giga_loss[loss=0.2956, simple_loss=0.3583, pruned_loss=0.1165, over 28773.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3699, pruned_loss=0.118, over 5705483.96 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3491, pruned_loss=0.09479, over 5671570.79 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3727, pruned_loss=0.121, over 5685513.93 frames. ], batch size: 92, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:40:23,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-07 16:40:28,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:40:56,360 INFO [train.py:968] (1/2) Epoch 15, batch 7750, giga_loss[loss=0.2964, simple_loss=0.3628, pruned_loss=0.115, over 29060.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3703, pruned_loss=0.1191, over 5699239.25 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3496, pruned_loss=0.09505, over 5675826.33 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3726, pruned_loss=0.1219, over 5680207.58 frames. ], batch size: 155, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:41:06,135 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-07 16:41:15,881 INFO [optim.py:369] (1/2) 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 16:41:29,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0965, 1.8662, 1.5108, 1.5065], device='cuda:1'), covar=tensor([0.0760, 0.0744, 0.0918, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0368, 0.0439, 0.0501, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 16:41:46,532 INFO [train.py:968] (1/2) Epoch 15, batch 7800, giga_loss[loss=0.3532, simple_loss=0.4013, pruned_loss=0.1526, over 28979.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3692, pruned_loss=0.1187, over 5697626.65 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3495, pruned_loss=0.09499, over 5670992.74 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3717, pruned_loss=0.1216, over 5686728.01 frames. ], batch size: 213, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:42:13,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:15,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=646745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:37,252 INFO [train.py:968] (1/2) Epoch 15, batch 7850, giga_loss[loss=0.2534, simple_loss=0.3294, pruned_loss=0.08874, over 28832.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3673, pruned_loss=0.1181, over 5696399.27 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3492, pruned_loss=0.0948, over 5670897.19 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3699, pruned_loss=0.1209, over 5688306.59 frames. ], batch size: 119, lr: 2.17e-03, grad_scale: 4.0
+2023-03-07 16:42:41,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=646774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:51,321 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=646785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:42:51,664 INFO [optim.py:369] (1/2) 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 16:42:54,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=646788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:43:15,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2711, 3.4294, 1.3244, 1.5829], device='cuda:1'), covar=tensor([0.1222, 0.0529, 0.0976, 0.1546], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0530, 0.0357, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 16:43:19,578 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 7900, giga_loss[loss=0.2807, simple_loss=0.3534, pruned_loss=0.104, over 28818.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3666, pruned_loss=0.1181, over 5705613.74 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3492, pruned_loss=0.09493, over 5676621.43 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1208, over 5694705.88 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:44:13,311 INFO [train.py:968] (1/2) Epoch 15, batch 7950, giga_loss[loss=0.3117, simple_loss=0.3821, pruned_loss=0.1207, over 28812.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3676, pruned_loss=0.119, over 5695428.17 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3488, pruned_loss=0.09459, over 5679785.74 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.37, pruned_loss=0.1217, over 5684439.41 frames. ], batch size: 243, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:44:18,636 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:44:32,086 INFO [optim.py:369] (1/2) 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 16:44:52,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2554, 1.4475, 1.4427, 1.2386], device='cuda:1'), covar=tensor([0.1553, 0.1516, 0.2028, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0737, 0.0693, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 16:45:00,974 INFO [train.py:968] (1/2) Epoch 15, batch 8000, giga_loss[loss=0.2797, simple_loss=0.352, pruned_loss=0.1037, over 28617.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3692, pruned_loss=0.1194, over 5694941.16 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09484, over 5684411.48 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3711, pruned_loss=0.1218, over 5682276.02 frames. ], batch size: 60, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:45:22,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:45:47,329 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=646967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:45:47,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=646967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:45:49,738 INFO [train.py:968] (1/2) Epoch 15, batch 8050, giga_loss[loss=0.3559, simple_loss=0.3978, pruned_loss=0.1569, over 26506.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3692, pruned_loss=0.1185, over 5686726.96 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3495, pruned_loss=0.09489, over 5686389.07 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3707, pruned_loss=0.1206, over 5674995.01 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:46:09,013 INFO [optim.py:369] (1/2) 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 16:46:36,512 INFO [train.py:968] (1/2) Epoch 15, batch 8100, giga_loss[loss=0.2903, simple_loss=0.3582, pruned_loss=0.1112, over 28819.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3702, pruned_loss=0.1193, over 5686360.04 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3495, pruned_loss=0.09498, over 5694272.26 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3722, pruned_loss=0.1218, over 5669229.61 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:47:30,561 INFO [train.py:968] (1/2) Epoch 15, batch 8150, giga_loss[loss=0.3542, simple_loss=0.4043, pruned_loss=0.1521, over 28842.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3714, pruned_loss=0.1205, over 5691929.44 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3496, pruned_loss=0.09513, over 5695372.90 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.373, pruned_loss=0.1224, over 5677598.57 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:47:44,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:47:46,774 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:48:14,136 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:48:16,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:48:20,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:48:25,443 INFO [train.py:968] (1/2) Epoch 15, batch 8200, giga_loss[loss=0.3605, simple_loss=0.4084, pruned_loss=0.1563, over 28996.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3737, pruned_loss=0.1233, over 5686382.55 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3498, pruned_loss=0.09521, over 5699456.65 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3751, pruned_loss=0.1253, over 5671360.47 frames. ], batch size: 155, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:48:47,178 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:49:14,370 INFO [train.py:968] (1/2) Epoch 15, batch 8250, libri_loss[loss=0.268, simple_loss=0.3536, pruned_loss=0.09121, over 29469.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3741, pruned_loss=0.1245, over 5690688.02 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3501, pruned_loss=0.09535, over 5704462.75 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 5673997.69 frames. ], batch size: 85, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:49:29,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-07 16:49:33,654 INFO [optim.py:369] (1/2) 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 16:50:03,376 INFO [train.py:968] (1/2) Epoch 15, batch 8300, giga_loss[loss=0.3361, simple_loss=0.3833, pruned_loss=0.1444, over 27973.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3757, pruned_loss=0.1268, over 5682615.23 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3499, pruned_loss=0.09519, over 5709561.89 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3777, pruned_loss=0.1294, over 5663965.92 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:50:34,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=647248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:50:50,068 INFO [train.py:968] (1/2) Epoch 15, batch 8350, giga_loss[loss=0.403, simple_loss=0.4275, pruned_loss=0.1892, over 23836.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3741, pruned_loss=0.1258, over 5672777.86 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3497, pruned_loss=0.09511, over 5706315.40 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3767, pruned_loss=0.129, over 5660443.02 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:51:08,621 INFO [optim.py:369] (1/2) 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 16:51:31,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:51:34,801 INFO [train.py:968] (1/2) Epoch 15, batch 8400, giga_loss[loss=0.329, simple_loss=0.3927, pruned_loss=0.1326, over 27986.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3738, pruned_loss=0.1253, over 5674319.12 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3496, pruned_loss=0.09503, over 5709437.85 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3764, pruned_loss=0.1286, over 5660914.94 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:51:53,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=647342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:52:01,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7092, 2.1094, 2.0101, 1.4976], device='cuda:1'), covar=tensor([0.1907, 0.2375, 0.1567, 0.1837], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0693, 0.0898, 0.0798], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 16:52:17,345 INFO [train.py:968] (1/2) Epoch 15, batch 8450, giga_loss[loss=0.3076, simple_loss=0.3785, pruned_loss=0.1184, over 28671.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3731, pruned_loss=0.1231, over 5682901.69 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3497, pruned_loss=0.09504, over 5716355.03 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3759, pruned_loss=0.1267, over 5665115.26 frames. ], batch size: 85, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:52:36,213 INFO [optim.py:369] (1/2) 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 16:52:38,503 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:52:41,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:53:02,242 INFO [train.py:968] (1/2) Epoch 15, batch 8500, giga_loss[loss=0.3657, simple_loss=0.4027, pruned_loss=0.1644, over 28637.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.371, pruned_loss=0.1217, over 5681223.09 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3496, pruned_loss=0.09505, over 5717588.66 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3738, pruned_loss=0.1251, over 5664838.26 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:53:06,095 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:53:26,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4767, 1.6564, 1.6008, 1.4335], device='cuda:1'), covar=tensor([0.1643, 0.1842, 0.2169, 0.1986], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0734, 0.0690, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 16:53:49,643 INFO [train.py:968] (1/2) Epoch 15, batch 8550, giga_loss[loss=0.3248, simple_loss=0.3811, pruned_loss=0.1342, over 28676.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3691, pruned_loss=0.1211, over 5686518.98 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3497, pruned_loss=0.0951, over 5718799.16 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3715, pruned_loss=0.1241, over 5672192.48 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:53:53,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6714, 1.8686, 1.4962, 2.0642], device='cuda:1'), covar=tensor([0.2390, 0.2560, 0.2763, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.1377, 0.1011, 0.1219, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 16:54:06,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:54:08,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:54:09,232 INFO [optim.py:369] (1/2) 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 16:54:36,848 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:54:38,161 INFO [train.py:968] (1/2) Epoch 15, batch 8600, giga_loss[loss=0.301, simple_loss=0.369, pruned_loss=0.1164, over 28681.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1213, over 5678267.21 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3496, pruned_loss=0.09515, over 5722434.42 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3708, pruned_loss=0.1243, over 5662670.46 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:55:23,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-07 16:55:31,067 INFO [train.py:968] (1/2) Epoch 15, batch 8650, giga_loss[loss=0.3051, simple_loss=0.375, pruned_loss=0.1176, over 28859.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3695, pruned_loss=0.1222, over 5667055.35 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3495, pruned_loss=0.09499, over 5722471.83 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3718, pruned_loss=0.1252, over 5653785.55 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:55:50,306 INFO [optim.py:369] (1/2) 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 16:56:19,151 INFO [train.py:968] (1/2) Epoch 15, batch 8700, giga_loss[loss=0.3864, simple_loss=0.439, pruned_loss=0.1669, over 28916.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3731, pruned_loss=0.1225, over 5664381.03 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3497, pruned_loss=0.09514, over 5716711.24 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3753, pruned_loss=0.1253, over 5657248.69 frames. ], batch size: 136, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:56:20,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4610, 1.6840, 1.6307, 1.4237], device='cuda:1'), covar=tensor([0.2261, 0.1801, 0.1894, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.1817, 0.1748, 0.1681, 0.1797], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 16:57:09,520 INFO [train.py:968] (1/2) Epoch 15, batch 8750, giga_loss[loss=0.3585, simple_loss=0.4116, pruned_loss=0.1527, over 28840.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3754, pruned_loss=0.1218, over 5672672.01 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.35, pruned_loss=0.09535, over 5719760.14 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3772, pruned_loss=0.1242, over 5663447.52 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 16:57:24,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5356, 2.2118, 1.7302, 0.7440], device='cuda:1'), covar=tensor([0.4898, 0.2551, 0.2870, 0.5575], device='cuda:1'), in_proj_covar=tensor([0.1637, 0.1560, 0.1533, 0.1337], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 16:57:29,395 INFO [optim.py:369] (1/2) 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 16:57:30,188 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 8800, giga_loss[loss=0.3541, simple_loss=0.4102, pruned_loss=0.1489, over 28489.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3773, pruned_loss=0.1233, over 5671969.94 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3498, pruned_loss=0.09537, over 5723501.04 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3794, pruned_loss=0.1257, over 5660614.35 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:58:30,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-07 16:58:45,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=647767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:58:46,731 INFO [train.py:968] (1/2) Epoch 15, batch 8850, giga_loss[loss=0.2883, simple_loss=0.363, pruned_loss=0.1068, over 29109.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3792, pruned_loss=0.1251, over 5662231.27 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3501, pruned_loss=0.09567, over 5723633.15 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.381, pruned_loss=0.1272, over 5652023.75 frames. ], batch size: 136, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:58:47,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9701, 1.2059, 1.2405, 1.0836], device='cuda:1'), covar=tensor([0.1411, 0.1171, 0.1877, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0731, 0.0688, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 16:59:02,541 INFO [optim.py:369] (1/2) 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 16:59:02,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:59:20,116 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:59:31,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3791, 1.5081, 1.2977, 1.4823], device='cuda:1'), covar=tensor([0.0735, 0.0352, 0.0318, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0113, 0.0114, 0.0211], device='cuda:1'), out_proj_covar=tensor([0.0087, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 16:59:32,715 INFO [train.py:968] (1/2) Epoch 15, batch 8900, giga_loss[loss=0.275, simple_loss=0.3453, pruned_loss=0.1023, over 28888.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.379, pruned_loss=0.1256, over 5666650.10 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3501, pruned_loss=0.09566, over 5726069.31 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3809, pruned_loss=0.1276, over 5655815.48 frames. ], batch size: 213, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 16:59:47,603 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=647833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 16:59:49,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:02,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:17,874 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:21,313 INFO [train.py:968] (1/2) Epoch 15, batch 8950, giga_loss[loss=0.3792, simple_loss=0.4059, pruned_loss=0.1762, over 23405.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3769, pruned_loss=0.1254, over 5650676.82 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3499, pruned_loss=0.09553, over 5729010.15 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3794, pruned_loss=0.128, over 5637511.74 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:00:42,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 17:00:42,844 INFO [optim.py:369] (1/2) 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 17:00:51,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=647899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:00:59,694 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=647913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:01:07,631 INFO [train.py:968] (1/2) Epoch 15, batch 9000, giga_loss[loss=0.2847, simple_loss=0.3567, pruned_loss=0.1064, over 28911.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3732, pruned_loss=0.1224, over 5639251.14 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3499, pruned_loss=0.09565, over 5705958.19 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1255, over 5645933.10 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:01:07,631 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 17:01:15,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5893, 1.6483, 1.2671, 1.2949], device='cuda:1'), covar=tensor([0.0675, 0.0416, 0.0908, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0441, 0.0500, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 17:01:15,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1339, 1.5592, 1.5686, 1.2900], device='cuda:1'), covar=tensor([0.1641, 0.1607, 0.2031, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0733, 0.0689, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 17:01:16,528 INFO [train.py:1012] (1/2) Epoch 15, validation: loss=0.2116, simple_loss=0.3182, pruned_loss=0.05252, over 944034.00 frames. 
+2023-03-07 17:01:16,529 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 17:01:38,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=647942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:01:53,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6098, 1.7558, 1.8855, 1.4024], device='cuda:1'), covar=tensor([0.1682, 0.2337, 0.1353, 0.1607], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0695, 0.0901, 0.0801], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 17:02:02,275 INFO [train.py:968] (1/2) Epoch 15, batch 9050, giga_loss[loss=0.3656, simple_loss=0.3934, pruned_loss=0.1689, over 23715.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3721, pruned_loss=0.1226, over 5649161.09 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3494, pruned_loss=0.0953, over 5711968.93 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3758, pruned_loss=0.1264, over 5646977.60 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:02:18,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 17:02:25,274 INFO [optim.py:369] (1/2) 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 17:02:55,605 INFO [train.py:968] (1/2) Epoch 15, batch 9100, giga_loss[loss=0.3325, simple_loss=0.3897, pruned_loss=0.1377, over 27968.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3734, pruned_loss=0.1242, over 5651476.38 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3494, pruned_loss=0.0953, over 5714860.97 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3766, pruned_loss=0.1276, over 5646375.92 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:03:19,445 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-07 17:03:44,115 INFO [train.py:968] (1/2) Epoch 15, batch 9150, giga_loss[loss=0.3198, simple_loss=0.3813, pruned_loss=0.1292, over 28603.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5641582.90 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3496, pruned_loss=0.09545, over 5711742.21 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5637627.27 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:04:01,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8560, 2.8365, 1.8162, 0.8772], device='cuda:1'), covar=tensor([0.6611, 0.2613, 0.3489, 0.6079], device='cuda:1'), in_proj_covar=tensor([0.1642, 0.1562, 0.1534, 0.1343], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 17:04:02,979 INFO [optim.py:369] (1/2) 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 17:04:28,661 INFO [train.py:968] (1/2) Epoch 15, batch 9200, giga_loss[loss=0.2766, simple_loss=0.34, pruned_loss=0.1066, over 28976.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.1231, over 5651908.07 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3495, pruned_loss=0.09528, over 5708305.10 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3742, pruned_loss=0.1271, over 5650177.04 frames. ], batch size: 106, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:05:14,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:05:17,486 INFO [train.py:968] (1/2) Epoch 15, batch 9250, libri_loss[loss=0.2463, simple_loss=0.3251, pruned_loss=0.08382, over 29648.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3698, pruned_loss=0.1226, over 5651645.92 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3496, pruned_loss=0.09531, over 5711349.93 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3728, pruned_loss=0.1262, over 5646666.61 frames. ], batch size: 73, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:05:27,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:05:30,421 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:05:43,998 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=648197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:05:51,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-07 17:05:58,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-07 17:06:06,515 INFO [train.py:968] (1/2) Epoch 15, batch 9300, libri_loss[loss=0.2335, simple_loss=0.3187, pruned_loss=0.07416, over 29589.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1212, over 5660257.60 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3495, pruned_loss=0.09529, over 5718325.28 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3729, pruned_loss=0.125, over 5647866.73 frames. ], batch size: 74, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:06:13,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 17:06:16,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:06:56,525 INFO [train.py:968] (1/2) Epoch 15, batch 9350, giga_loss[loss=0.3846, simple_loss=0.4243, pruned_loss=0.1725, over 28196.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.372, pruned_loss=0.1227, over 5655538.10 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3492, pruned_loss=0.09508, over 5712537.96 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3752, pruned_loss=0.1264, over 5650376.32 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:07:01,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:07:18,185 INFO [optim.py:369] (1/2) 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 17:07:26,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5229, 1.2408, 4.1533, 3.3147], device='cuda:1'), covar=tensor([0.1877, 0.2808, 0.0771, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0615, 0.0907, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 17:07:31,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:07:34,201 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 9400, giga_loss[loss=0.272, simple_loss=0.3535, pruned_loss=0.09531, over 29040.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3721, pruned_loss=0.1232, over 5652537.17 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3496, pruned_loss=0.09535, over 5713669.48 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3749, pruned_loss=0.1266, over 5645736.29 frames. ], batch size: 155, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:07:46,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:07:49,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:02,321 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=648346.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:08:16,514 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 9450, giga_loss[loss=0.2795, simple_loss=0.373, pruned_loss=0.09304, over 28556.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3732, pruned_loss=0.1219, over 5663986.51 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3496, pruned_loss=0.09529, over 5719691.89 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3762, pruned_loss=0.1256, over 5651058.42 frames. ], batch size: 60, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:08:30,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:33,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:08:37,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5810, 1.6507, 1.2920, 1.3254], device='cuda:1'), covar=tensor([0.0779, 0.0494, 0.0943, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0445, 0.0504, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 17:08:46,704 INFO [optim.py:369] (1/2) 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 17:08:57,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:05,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=648411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:09,736 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 9500, giga_loss[loss=0.2935, simple_loss=0.3706, pruned_loss=0.1082, over 28833.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3727, pruned_loss=0.1194, over 5677688.65 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3492, pruned_loss=0.09509, over 5726487.68 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3765, pruned_loss=0.1236, over 5659329.91 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:09:14,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:17,385 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-07 17:09:42,116 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:09:51,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5806, 1.8324, 1.8270, 1.3771], device='cuda:1'), covar=tensor([0.1683, 0.2469, 0.1449, 0.1735], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0697, 0.0901, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 17:09:55,848 INFO [train.py:968] (1/2) Epoch 15, batch 9550, giga_loss[loss=0.2973, simple_loss=0.3719, pruned_loss=0.1113, over 28940.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3755, pruned_loss=0.1203, over 5686423.82 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3488, pruned_loss=0.09505, over 5728408.58 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3797, pruned_loss=0.1244, over 5668623.48 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:10:19,637 INFO [optim.py:369] (1/2) 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 17:10:49,259 INFO [train.py:968] (1/2) Epoch 15, batch 9600, giga_loss[loss=0.3709, simple_loss=0.417, pruned_loss=0.1624, over 28750.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3799, pruned_loss=0.1245, over 5675371.48 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3488, pruned_loss=0.095, over 5729038.54 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3833, pruned_loss=0.1279, over 5660595.43 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:10:54,116 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-07 17:11:21,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:11:32,881 INFO [train.py:968] (1/2) Epoch 15, batch 9650, giga_loss[loss=0.3428, simple_loss=0.3997, pruned_loss=0.1429, over 28312.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3816, pruned_loss=0.1265, over 5685104.06 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3485, pruned_loss=0.09486, over 5734324.87 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3858, pruned_loss=0.1304, over 5666837.10 frames. ], batch size: 369, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:11:36,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:11:44,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 17:11:56,534 INFO [optim.py:369] (1/2) 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 17:12:22,925 INFO [train.py:968] (1/2) Epoch 15, batch 9700, giga_loss[loss=0.3485, simple_loss=0.4014, pruned_loss=0.1478, over 27947.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3818, pruned_loss=0.1277, over 5670731.94 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3484, pruned_loss=0.09472, over 5736045.08 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3859, pruned_loss=0.1316, over 5653758.19 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:13:10,043 INFO [train.py:968] (1/2) Epoch 15, batch 9750, giga_loss[loss=0.3386, simple_loss=0.4047, pruned_loss=0.1363, over 29078.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3805, pruned_loss=0.1265, over 5675778.92 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3479, pruned_loss=0.09447, over 5739874.26 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.3853, pruned_loss=0.1309, over 5656565.60 frames. ], batch size: 155, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:13:28,045 INFO [optim.py:369] (1/2) 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 17:13:37,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:39,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:47,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:50,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:13:50,431 INFO [train.py:968] (1/2) Epoch 15, batch 9800, giga_loss[loss=0.2646, simple_loss=0.3561, pruned_loss=0.0865, over 28958.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3781, pruned_loss=0.1232, over 5682509.39 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.348, pruned_loss=0.09444, over 5742703.21 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3835, pruned_loss=0.1283, over 5661495.33 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:13:55,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648721.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:14:03,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:14:17,109 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:14:36,652 INFO [train.py:968] (1/2) Epoch 15, batch 9850, giga_loss[loss=0.2985, simple_loss=0.3748, pruned_loss=0.1111, over 28766.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3778, pruned_loss=0.1216, over 5686044.99 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3475, pruned_loss=0.09417, over 5746397.65 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.383, pruned_loss=0.1265, over 5664882.17 frames. ], batch size: 99, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:14:49,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3544, 5.1580, 4.8953, 2.2199], device='cuda:1'), covar=tensor([0.0477, 0.0670, 0.0732, 0.1966], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.1065, 0.0915, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 17:14:54,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=648786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:14:57,698 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 9900, giga_loss[loss=0.3508, simple_loss=0.3871, pruned_loss=0.1573, over 23470.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3781, pruned_loss=0.1214, over 5687306.60 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3477, pruned_loss=0.09422, over 5751514.29 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3833, pruned_loss=0.1263, over 5663522.55 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:16:11,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648864.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:16:15,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648867.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:16:16,852 INFO [train.py:968] (1/2) Epoch 15, batch 9950, giga_loss[loss=0.3735, simple_loss=0.4093, pruned_loss=0.1689, over 26512.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3806, pruned_loss=0.1242, over 5674929.14 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3478, pruned_loss=0.09416, over 5750925.99 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3853, pruned_loss=0.1288, over 5655600.76 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 1.0
+2023-03-07 17:16:43,541 INFO [optim.py:369] (1/2) 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 17:16:46,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648896.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:17:05,742 INFO [train.py:968] (1/2) Epoch 15, batch 10000, giga_loss[loss=0.3002, simple_loss=0.366, pruned_loss=0.1172, over 28813.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3791, pruned_loss=0.124, over 5669872.42 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3478, pruned_loss=0.09404, over 5753205.17 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3832, pruned_loss=0.1281, over 5651463.13 frames. ], batch size: 112, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:17:14,317 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=648929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:17:18,594 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=648932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:17:54,660 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=648961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:18:00,676 INFO [train.py:968] (1/2) Epoch 15, batch 10050, giga_loss[loss=0.2918, simple_loss=0.3559, pruned_loss=0.1139, over 28896.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3787, pruned_loss=0.1252, over 5666316.83 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.348, pruned_loss=0.09412, over 5755618.79 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3822, pruned_loss=0.1288, over 5648548.24 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:18:10,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2160, 2.1805, 1.6339, 1.7325], device='cuda:1'), covar=tensor([0.0837, 0.0709, 0.0985, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0449, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 17:18:23,527 INFO [optim.py:369] (1/2) 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 17:18:47,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 17:18:49,211 INFO [train.py:968] (1/2) Epoch 15, batch 10100, giga_loss[loss=0.3062, simple_loss=0.3516, pruned_loss=0.1304, over 23696.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3754, pruned_loss=0.1235, over 5668061.95 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3482, pruned_loss=0.09407, over 5757894.45 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3784, pruned_loss=0.1269, over 5650737.92 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:19:44,869 INFO [train.py:968] (1/2) Epoch 15, batch 10150, giga_loss[loss=0.3381, simple_loss=0.389, pruned_loss=0.1436, over 28295.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3745, pruned_loss=0.1241, over 5650731.03 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3487, pruned_loss=0.09441, over 5749908.50 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3772, pruned_loss=0.1272, over 5641802.34 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:20:05,096 INFO [optim.py:369] (1/2) 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 17:20:31,922 INFO [train.py:968] (1/2) Epoch 15, batch 10200, libri_loss[loss=0.2331, simple_loss=0.3163, pruned_loss=0.07497, over 29562.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3742, pruned_loss=0.1242, over 5660540.26 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3487, pruned_loss=0.09446, over 5746162.29 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3771, pruned_loss=0.1276, over 5653921.72 frames. ], batch size: 78, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:21:17,740 INFO [train.py:968] (1/2) Epoch 15, batch 10250, giga_loss[loss=0.2762, simple_loss=0.3636, pruned_loss=0.09444, over 28848.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3717, pruned_loss=0.122, over 5659899.95 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3486, pruned_loss=0.09438, over 5748772.83 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3745, pruned_loss=0.1252, over 5650901.01 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:21:40,975 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 10300, giga_loss[loss=0.3253, simple_loss=0.3783, pruned_loss=0.1361, over 28434.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3683, pruned_loss=0.1184, over 5653013.63 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3486, pruned_loss=0.0945, over 5750457.54 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3706, pruned_loss=0.1211, over 5643800.33 frames. ], batch size: 85, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:22:11,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3003, 2.9514, 1.4145, 1.5061], device='cuda:1'), covar=tensor([0.0946, 0.0358, 0.0877, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0532, 0.0359, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 17:22:11,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-07 17:22:59,533 INFO [train.py:968] (1/2) Epoch 15, batch 10350, giga_loss[loss=0.2754, simple_loss=0.359, pruned_loss=0.0959, over 28516.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3663, pruned_loss=0.1162, over 5665999.27 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3486, pruned_loss=0.09454, over 5751827.78 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3684, pruned_loss=0.1185, over 5656247.90 frames. ], batch size: 71, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:23:08,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 17:23:19,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3741, 3.3314, 1.5646, 1.4981], device='cuda:1'), covar=tensor([0.0896, 0.0399, 0.0836, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0533, 0.0359, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 17:23:23,865 INFO [optim.py:369] (1/2) 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 17:23:50,624 INFO [train.py:968] (1/2) Epoch 15, batch 10400, giga_loss[loss=0.3336, simple_loss=0.3842, pruned_loss=0.1415, over 28775.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3656, pruned_loss=0.1163, over 5668427.99 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3489, pruned_loss=0.09465, over 5753450.83 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3671, pruned_loss=0.1182, over 5658585.40 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:24:41,972 INFO [train.py:968] (1/2) Epoch 15, batch 10450, libri_loss[loss=0.2802, simple_loss=0.3685, pruned_loss=0.09598, over 29542.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5665947.87 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3492, pruned_loss=0.09479, over 5752798.73 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3643, pruned_loss=0.1175, over 5656886.40 frames. ], batch size: 89, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:25:06,788 INFO [optim.py:369] (1/2) 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 17:25:31,822 INFO [train.py:968] (1/2) Epoch 15, batch 10500, giga_loss[loss=0.2653, simple_loss=0.3578, pruned_loss=0.08635, over 28905.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3636, pruned_loss=0.1158, over 5667570.56 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3489, pruned_loss=0.09461, over 5754093.88 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.365, pruned_loss=0.1176, over 5658393.40 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:26:16,952 INFO [train.py:968] (1/2) Epoch 15, batch 10550, giga_loss[loss=0.32, simple_loss=0.3863, pruned_loss=0.1269, over 28640.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3653, pruned_loss=0.1161, over 5664847.06 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3491, pruned_loss=0.09465, over 5749185.86 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3667, pruned_loss=0.1181, over 5658778.72 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:26:39,036 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 10600, giga_loss[loss=0.3396, simple_loss=0.3941, pruned_loss=0.1425, over 28756.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3667, pruned_loss=0.1167, over 5664700.07 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3499, pruned_loss=0.09491, over 5755889.10 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.368, pruned_loss=0.1192, over 5649383.74 frames. ], batch size: 242, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:27:18,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3344, 1.5440, 1.4592, 1.5318], device='cuda:1'), covar=tensor([0.0750, 0.0333, 0.0297, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0212], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 17:27:26,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8922, 4.7252, 4.4679, 2.2123], device='cuda:1'), covar=tensor([0.0453, 0.0571, 0.0625, 0.1975], device='cuda:1'), in_proj_covar=tensor([0.1138, 0.1060, 0.0916, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 17:27:28,464 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 17:27:37,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8298, 1.9030, 1.3030, 1.5261], device='cuda:1'), covar=tensor([0.0810, 0.0617, 0.1026, 0.1022], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0444, 0.0504, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 17:27:46,672 INFO [train.py:968] (1/2) Epoch 15, batch 10650, giga_loss[loss=0.2855, simple_loss=0.3544, pruned_loss=0.1083, over 28963.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3666, pruned_loss=0.1166, over 5668356.88 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3508, pruned_loss=0.09541, over 5761132.76 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3675, pruned_loss=0.1191, over 5647736.55 frames. ], batch size: 145, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:28:07,622 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 10700, giga_loss[loss=0.2755, simple_loss=0.3493, pruned_loss=0.1008, over 28961.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3672, pruned_loss=0.1176, over 5667926.23 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3509, pruned_loss=0.09545, over 5764260.77 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.1199, over 5646961.70 frames. ], batch size: 136, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:29:12,514 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649659.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:29:24,456 INFO [train.py:968] (1/2) Epoch 15, batch 10750, giga_loss[loss=0.3023, simple_loss=0.3716, pruned_loss=0.1165, over 28956.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.37, pruned_loss=0.1195, over 5664201.12 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3511, pruned_loss=0.09546, over 5766791.88 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.1221, over 5642295.01 frames. ], batch size: 213, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:29:44,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2277, 1.2518, 4.0514, 3.2403], device='cuda:1'), covar=tensor([0.1715, 0.2668, 0.0434, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0711, 0.0618, 0.0911, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 17:29:48,883 INFO [optim.py:369] (1/2) 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 17:29:51,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2976, 1.6310, 1.4032, 1.5496], device='cuda:1'), covar=tensor([0.0745, 0.0316, 0.0309, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0057, 0.0096], device='cuda:1')
+2023-03-07 17:30:16,609 INFO [train.py:968] (1/2) Epoch 15, batch 10800, libri_loss[loss=0.2868, simple_loss=0.3526, pruned_loss=0.1105, over 29601.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3715, pruned_loss=0.1203, over 5673351.67 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3509, pruned_loss=0.09541, over 5768070.39 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3728, pruned_loss=0.1227, over 5654022.53 frames. ], batch size: 74, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:30:38,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4660, 1.6707, 1.3486, 1.5998], device='cuda:1'), covar=tensor([0.2440, 0.2300, 0.2550, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.1379, 0.1010, 0.1224, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 17:30:53,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-07 17:31:02,112 INFO [train.py:968] (1/2) Epoch 15, batch 10850, giga_loss[loss=0.435, simple_loss=0.4471, pruned_loss=0.2115, over 28926.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3732, pruned_loss=0.1218, over 5666968.00 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3511, pruned_loss=0.09564, over 5752720.60 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3746, pruned_loss=0.1241, over 5662057.35 frames. ], batch size: 100, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:31:28,974 INFO [optim.py:369] (1/2) 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 17:31:40,500 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:31:52,654 INFO [train.py:968] (1/2) Epoch 15, batch 10900, giga_loss[loss=0.2986, simple_loss=0.3767, pruned_loss=0.1102, over 28881.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3749, pruned_loss=0.1239, over 5666139.00 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3518, pruned_loss=0.09602, over 5747314.94 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3758, pruned_loss=0.1258, over 5665505.43 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:32:42,353 INFO [train.py:968] (1/2) Epoch 15, batch 10950, giga_loss[loss=0.2801, simple_loss=0.3468, pruned_loss=0.1067, over 28519.00 frames. ], tot_loss[loss=0.311, simple_loss=0.376, pruned_loss=0.123, over 5661859.80 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3519, pruned_loss=0.09605, over 5751141.46 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3772, pruned_loss=0.1254, over 5655763.51 frames. ], batch size: 78, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:32:57,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3404, 1.2215, 1.1262, 1.6329], device='cuda:1'), covar=tensor([0.0743, 0.0343, 0.0352, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0056, 0.0096], device='cuda:1')
+2023-03-07 17:33:07,835 INFO [optim.py:369] (1/2) 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 17:33:22,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:33:25,059 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=649912.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:33:32,148 INFO [train.py:968] (1/2) Epoch 15, batch 11000, giga_loss[loss=0.3803, simple_loss=0.3993, pruned_loss=0.1806, over 23585.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3761, pruned_loss=0.1236, over 5638173.33 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3523, pruned_loss=0.09662, over 5728056.02 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3774, pruned_loss=0.1256, over 5650881.59 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:34:18,688 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 11050, giga_loss[loss=0.3011, simple_loss=0.3699, pruned_loss=0.1162, over 28745.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3749, pruned_loss=0.1234, over 5641781.66 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3524, pruned_loss=0.09671, over 5731211.45 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3764, pruned_loss=0.1256, over 5647208.96 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:34:51,651 INFO [optim.py:369] (1/2) 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 17:35:23,073 INFO [train.py:968] (1/2) Epoch 15, batch 11100, giga_loss[loss=0.2529, simple_loss=0.3249, pruned_loss=0.09046, over 28338.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3738, pruned_loss=0.1232, over 5638286.56 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3525, pruned_loss=0.09676, over 5733091.32 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3751, pruned_loss=0.1252, over 5639880.00 frames. ], batch size: 71, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:35:38,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650034.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:36:10,185 INFO [train.py:968] (1/2) Epoch 15, batch 11150, libri_loss[loss=0.3157, simple_loss=0.3752, pruned_loss=0.1281, over 29675.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3713, pruned_loss=0.1222, over 5644295.93 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.352, pruned_loss=0.09655, over 5734032.08 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3733, pruned_loss=0.1245, over 5642392.08 frames. ], batch size: 73, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:36:28,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5049, 4.2898, 4.0441, 1.9876], device='cuda:1'), covar=tensor([0.0675, 0.0916, 0.1004, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.1063, 0.0921, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 17:36:37,459 INFO [optim.py:369] (1/2) 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 17:37:00,645 INFO [train.py:968] (1/2) Epoch 15, batch 11200, giga_loss[loss=0.3336, simple_loss=0.3865, pruned_loss=0.1403, over 28892.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1232, over 5651424.72 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.352, pruned_loss=0.09656, over 5736582.27 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1255, over 5646352.32 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:37:39,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6691, 1.6882, 1.6390, 1.4528], device='cuda:1'), covar=tensor([0.1536, 0.1990, 0.2259, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0740, 0.0694, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 17:37:50,346 INFO [train.py:968] (1/2) Epoch 15, batch 11250, giga_loss[loss=0.4025, simple_loss=0.4213, pruned_loss=0.1919, over 27535.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3709, pruned_loss=0.1232, over 5651470.00 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3518, pruned_loss=0.09642, over 5737335.36 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3726, pruned_loss=0.1252, over 5646453.27 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:37:59,598 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650177.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:38:01,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650180.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:38:02,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:38:18,832 INFO [optim.py:369] (1/2) 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 17:38:31,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650209.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:38:40,853 INFO [train.py:968] (1/2) Epoch 15, batch 11300, giga_loss[loss=0.3309, simple_loss=0.3907, pruned_loss=0.1356, over 28665.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3712, pruned_loss=0.1234, over 5656882.48 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3517, pruned_loss=0.09625, over 5741818.02 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3731, pruned_loss=0.1258, over 5646899.95 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:38:58,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5024, 1.7833, 1.3406, 1.6348], device='cuda:1'), covar=tensor([0.2401, 0.2401, 0.2788, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.1385, 0.1016, 0.1231, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 17:39:26,298 INFO [train.py:968] (1/2) Epoch 15, batch 11350, giga_loss[loss=0.3282, simple_loss=0.3885, pruned_loss=0.1339, over 28901.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3732, pruned_loss=0.1252, over 5655800.77 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3514, pruned_loss=0.09619, over 5742438.30 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3757, pruned_loss=0.1282, over 5643876.27 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:39:33,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0718, 2.0188, 1.5499, 1.6611], device='cuda:1'), covar=tensor([0.0828, 0.0722, 0.0957, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0445, 0.0504, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 17:39:42,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:39:44,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650287.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:39:52,179 INFO [optim.py:369] (1/2) 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 17:40:16,582 INFO [train.py:968] (1/2) Epoch 15, batch 11400, giga_loss[loss=0.388, simple_loss=0.4055, pruned_loss=0.1853, over 23403.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3754, pruned_loss=0.1268, over 5649751.10 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3515, pruned_loss=0.09622, over 5744906.09 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3778, pruned_loss=0.1298, over 5636367.06 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:40:20,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:40:24,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:40:41,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:40:53,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:41:09,346 INFO [train.py:968] (1/2) Epoch 15, batch 11450, giga_loss[loss=0.4331, simple_loss=0.4568, pruned_loss=0.2047, over 27898.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3771, pruned_loss=0.1291, over 5647743.83 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3517, pruned_loss=0.09632, over 5745443.55 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3789, pruned_loss=0.1314, over 5636348.83 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:41:35,998 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 11500, giga_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 29092.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3754, pruned_loss=0.1272, over 5656500.97 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.352, pruned_loss=0.09649, over 5741441.53 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3771, pruned_loss=0.1296, over 5649106.83 frames. ], batch size: 128, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:42:06,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:42:08,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650430.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:42:08,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:42:11,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650433.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:42:39,460 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:42:40,675 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650462.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:42:45,230 INFO [train.py:968] (1/2) Epoch 15, batch 11550, giga_loss[loss=0.3607, simple_loss=0.4081, pruned_loss=0.1567, over 28257.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3769, pruned_loss=0.1283, over 5644405.52 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.352, pruned_loss=0.09643, over 5736354.84 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3789, pruned_loss=0.131, over 5641555.69 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:43:01,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=650484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:03,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=650487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:06,465 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:08,846 INFO [optim.py:369] (1/2) 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 17:43:30,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=650516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:43:32,258 INFO [train.py:968] (1/2) Epoch 15, batch 11600, giga_loss[loss=0.3217, simple_loss=0.3888, pruned_loss=0.1273, over 28669.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3764, pruned_loss=0.127, over 5663051.74 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3524, pruned_loss=0.0969, over 5738741.12 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3785, pruned_loss=0.1299, over 5655528.86 frames. ], batch size: 284, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 17:43:39,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3814, 1.7471, 1.3850, 1.4295], device='cuda:1'), covar=tensor([0.2328, 0.2226, 0.2422, 0.2001], device='cuda:1'), in_proj_covar=tensor([0.1387, 0.1018, 0.1230, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 17:43:57,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2816, 1.7127, 1.2252, 0.5666], device='cuda:1'), covar=tensor([0.3669, 0.1913, 0.2311, 0.4811], device='cuda:1'), in_proj_covar=tensor([0.1638, 0.1565, 0.1541, 0.1337], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 17:43:58,445 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:44:01,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8155, 2.1964, 1.9204, 1.6248], device='cuda:1'), covar=tensor([0.2829, 0.1973, 0.2197, 0.2504], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1753, 0.1696, 0.1808], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 17:44:05,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 11650, giga_loss[loss=0.4275, simple_loss=0.4461, pruned_loss=0.2044, over 26642.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5653032.60 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3523, pruned_loss=0.09686, over 5740900.38 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3802, pruned_loss=0.1312, over 5643580.12 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:44:50,269 INFO [optim.py:369] (1/2) 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 17:45:12,870 INFO [train.py:968] (1/2) Epoch 15, batch 11700, giga_loss[loss=0.4085, simple_loss=0.4394, pruned_loss=0.1888, over 28332.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3802, pruned_loss=0.13, over 5655072.87 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3525, pruned_loss=0.09689, over 5744003.84 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3825, pruned_loss=0.1331, over 5642900.31 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:45:47,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4481, 1.7308, 1.3664, 1.5271], device='cuda:1'), covar=tensor([0.2559, 0.2560, 0.2849, 0.2275], device='cuda:1'), in_proj_covar=tensor([0.1382, 0.1015, 0.1225, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 17:45:59,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5614, 4.3822, 4.1443, 2.0077], device='cuda:1'), covar=tensor([0.0542, 0.0708, 0.0732, 0.2057], device='cuda:1'), in_proj_covar=tensor([0.1146, 0.1064, 0.0917, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 17:46:00,399 INFO [train.py:968] (1/2) Epoch 15, batch 11750, giga_loss[loss=0.2788, simple_loss=0.3467, pruned_loss=0.1054, over 28899.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3806, pruned_loss=0.1306, over 5661472.82 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3527, pruned_loss=0.09695, over 5748117.93 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3828, pruned_loss=0.1338, over 5646022.95 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:46:25,950 INFO [optim.py:369] (1/2) 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 17:46:48,466 INFO [train.py:968] (1/2) Epoch 15, batch 11800, giga_loss[loss=0.3986, simple_loss=0.4378, pruned_loss=0.1797, over 27486.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3807, pruned_loss=0.1301, over 5661014.30 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3521, pruned_loss=0.09663, over 5751301.61 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3836, pruned_loss=0.1336, over 5644442.92 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:47:00,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8260, 1.9547, 1.6313, 2.0529], device='cuda:1'), covar=tensor([0.2395, 0.2525, 0.2808, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.1383, 0.1016, 0.1229, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 17:47:16,228 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:47:35,375 INFO [train.py:968] (1/2) Epoch 15, batch 11850, giga_loss[loss=0.2895, simple_loss=0.3717, pruned_loss=0.1036, over 28879.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3801, pruned_loss=0.1285, over 5655288.53 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3524, pruned_loss=0.09675, over 5747112.32 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3826, pruned_loss=0.1317, over 5644247.35 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:48:02,939 INFO [optim.py:369] (1/2) 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 17:48:12,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:48:24,093 INFO [train.py:968] (1/2) Epoch 15, batch 11900, giga_loss[loss=0.354, simple_loss=0.4103, pruned_loss=0.1488, over 28004.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3794, pruned_loss=0.1276, over 5661773.92 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3526, pruned_loss=0.097, over 5750743.41 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3822, pruned_loss=0.1311, over 5646367.35 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:49:11,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 11950, giga_loss[loss=0.2869, simple_loss=0.3576, pruned_loss=0.1081, over 28832.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3763, pruned_loss=0.126, over 5659819.46 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3518, pruned_loss=0.09666, over 5753795.18 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3797, pruned_loss=0.1297, over 5643179.72 frames. ], batch size: 119, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:49:24,610 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-07 17:49:42,685 INFO [optim.py:369] (1/2) 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 17:50:01,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4208, 3.5204, 1.6357, 1.5174], device='cuda:1'), covar=tensor([0.0982, 0.0384, 0.0844, 0.1369], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0532, 0.0359, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 17:50:02,688 INFO [train.py:968] (1/2) Epoch 15, batch 12000, giga_loss[loss=0.3016, simple_loss=0.372, pruned_loss=0.1156, over 29005.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3769, pruned_loss=0.126, over 5667106.00 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3522, pruned_loss=0.09687, over 5757006.05 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3799, pruned_loss=0.1294, over 5648925.93 frames. ], batch size: 106, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:50:02,688 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 17:50:11,132 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 17:50:12,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:50:19,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=650928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:50:24,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4261, 1.6160, 1.5506, 1.4825], device='cuda:1'), covar=tensor([0.1667, 0.1720, 0.2086, 0.1801], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0737, 0.0690, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 17:50:43,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:50:45,805 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=650955.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:50:57,308 INFO [train.py:968] (1/2) Epoch 15, batch 12050, giga_loss[loss=0.3053, simple_loss=0.3708, pruned_loss=0.1199, over 29113.00 frames. ], tot_loss[loss=0.316, simple_loss=0.378, pruned_loss=0.1269, over 5654345.27 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3523, pruned_loss=0.09692, over 5748978.52 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3807, pruned_loss=0.13, over 5646005.08 frames. ], batch size: 128, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:51:25,645 INFO [optim.py:369] (1/2) 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 17:51:38,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:51:43,006 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 12100, giga_loss[loss=0.2619, simple_loss=0.3326, pruned_loss=0.09565, over 28579.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3765, pruned_loss=0.1265, over 5662801.51 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3519, pruned_loss=0.09667, over 5742986.34 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3794, pruned_loss=0.1296, over 5660265.73 frames. ], batch size: 78, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:52:13,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:22,180 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:33,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:36,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:38,039 INFO [train.py:968] (1/2) Epoch 15, batch 12150, giga_loss[loss=0.3224, simple_loss=0.3811, pruned_loss=0.1318, over 28307.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3774, pruned_loss=0.1276, over 5665246.95 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3521, pruned_loss=0.09676, over 5745908.55 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3799, pruned_loss=0.1306, over 5658996.51 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:52:39,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:42,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:52:49,535 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:53:04,825 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:53:13,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:53:15,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651106.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:53:28,370 INFO [train.py:968] (1/2) Epoch 15, batch 12200, giga_loss[loss=0.3007, simple_loss=0.371, pruned_loss=0.1152, over 28534.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3798, pruned_loss=0.1295, over 5659347.44 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3524, pruned_loss=0.09688, over 5738145.82 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3818, pruned_loss=0.1322, over 5659856.85 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:53:33,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:54:18,997 INFO [train.py:968] (1/2) Epoch 15, batch 12250, giga_loss[loss=0.2781, simple_loss=0.3575, pruned_loss=0.09937, over 28857.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3806, pruned_loss=0.1302, over 5655655.55 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3529, pruned_loss=0.09712, over 5741333.25 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3824, pruned_loss=0.1327, over 5651819.64 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:54:29,941 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:55:07,041 INFO [train.py:968] (1/2) Epoch 15, batch 12300, giga_loss[loss=0.2766, simple_loss=0.3548, pruned_loss=0.09923, over 28838.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3786, pruned_loss=0.1278, over 5666593.19 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3522, pruned_loss=0.09677, over 5735054.96 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.381, pruned_loss=0.1306, over 5667654.15 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:55:18,287 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:55:58,076 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:00,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3257, 1.8594, 1.4741, 0.5156], device='cuda:1'), covar=tensor([0.3780, 0.2115, 0.2798, 0.4807], device='cuda:1'), in_proj_covar=tensor([0.1637, 0.1566, 0.1541, 0.1337], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 17:56:00,353 INFO [train.py:968] (1/2) Epoch 15, batch 12350, giga_loss[loss=0.3171, simple_loss=0.3865, pruned_loss=0.1239, over 28615.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3775, pruned_loss=0.1266, over 5657979.66 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3522, pruned_loss=0.09672, over 5735630.69 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3795, pruned_loss=0.129, over 5658017.30 frames. ], batch size: 336, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 17:56:00,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:08,791 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651277.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:56:26,916 INFO [optim.py:369] (1/2) 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 17:56:27,196 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:46,003 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 12400, giga_loss[loss=0.3003, simple_loss=0.3698, pruned_loss=0.1154, over 28933.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.377, pruned_loss=0.1256, over 5668954.76 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3525, pruned_loss=0.09687, over 5736514.90 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3789, pruned_loss=0.1281, over 5666662.96 frames. ], batch size: 174, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:56:53,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:55,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:56,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:56,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:56:56,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651330.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 17:57:25,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:57:37,984 INFO [train.py:968] (1/2) Epoch 15, batch 12450, giga_loss[loss=0.2996, simple_loss=0.3639, pruned_loss=0.1176, over 28413.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3756, pruned_loss=0.125, over 5663620.21 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3523, pruned_loss=0.09672, over 5738760.31 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3776, pruned_loss=0.1275, over 5659104.11 frames. ], batch size: 78, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:58:04,329 INFO [optim.py:369] (1/2) 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 17:58:26,668 INFO [train.py:968] (1/2) Epoch 15, batch 12500, giga_loss[loss=0.3642, simple_loss=0.3864, pruned_loss=0.1709, over 23616.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1247, over 5668743.58 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3523, pruned_loss=0.09654, over 5739727.35 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3767, pruned_loss=0.1273, over 5663141.22 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:58:35,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:58:57,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:12,290 INFO [train.py:968] (1/2) Epoch 15, batch 12550, giga_loss[loss=0.2792, simple_loss=0.3386, pruned_loss=0.1099, over 28200.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5671227.52 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.352, pruned_loss=0.09637, over 5741587.75 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3731, pruned_loss=0.1253, over 5663423.06 frames. ], batch size: 77, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 17:59:14,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:16,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651473.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:59:17,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:19,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651476.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:59:24,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651481.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 17:59:40,898 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 17:59:50,017 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651505.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:00:05,360 INFO [train.py:968] (1/2) Epoch 15, batch 12600, giga_loss[loss=0.2955, simple_loss=0.3595, pruned_loss=0.1157, over 28895.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3682, pruned_loss=0.1215, over 5675931.02 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3521, pruned_loss=0.09659, over 5733698.00 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3701, pruned_loss=0.1239, over 5675945.19 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:00:32,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 18:00:52,565 INFO [train.py:968] (1/2) Epoch 15, batch 12650, giga_loss[loss=0.3122, simple_loss=0.3716, pruned_loss=0.1264, over 28285.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3669, pruned_loss=0.1209, over 5686262.24 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3522, pruned_loss=0.09659, over 5737791.24 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3688, pruned_loss=0.1234, over 5681397.91 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:00:53,268 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:23,309 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:01:26,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:26,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:28,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:01:44,424 INFO [train.py:968] (1/2) Epoch 15, batch 12700, giga_loss[loss=0.3137, simple_loss=0.361, pruned_loss=0.1332, over 24051.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3676, pruned_loss=0.1215, over 5678630.93 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3522, pruned_loss=0.09664, over 5739306.74 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3692, pruned_loss=0.1237, over 5672797.22 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:01:50,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651624.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:01:52,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651627.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:01:53,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:02:14,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651652.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:02:18,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651656.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:02:28,956 INFO [train.py:968] (1/2) Epoch 15, batch 12750, giga_loss[loss=0.3174, simple_loss=0.3828, pruned_loss=0.126, over 28630.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3659, pruned_loss=0.1184, over 5687121.51 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3516, pruned_loss=0.09638, over 5745417.45 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3686, pruned_loss=0.1218, over 5673761.61 frames. ], batch size: 307, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:02:53,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:02:57,670 INFO [optim.py:369] (1/2) 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 18:03:04,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=651705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:03:21,543 INFO [train.py:968] (1/2) Epoch 15, batch 12800, giga_loss[loss=0.2716, simple_loss=0.3558, pruned_loss=0.09371, over 28677.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3634, pruned_loss=0.1147, over 5676176.82 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.351, pruned_loss=0.09615, over 5747946.85 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3662, pruned_loss=0.1179, over 5662758.51 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:03:51,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5311, 5.3100, 5.0559, 2.4833], device='cuda:1'), covar=tensor([0.0439, 0.0666, 0.0770, 0.1862], device='cuda:1'), in_proj_covar=tensor([0.1146, 0.1065, 0.0916, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 18:03:53,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:03:55,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:04:04,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5634, 1.8406, 1.6710, 1.4835], device='cuda:1'), covar=tensor([0.2097, 0.1552, 0.1445, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.1818, 0.1746, 0.1680, 0.1798], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 18:04:14,085 INFO [train.py:968] (1/2) Epoch 15, batch 12850, giga_loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08814, over 29037.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3606, pruned_loss=0.1119, over 5676357.11 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3508, pruned_loss=0.09614, over 5750149.70 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3632, pruned_loss=0.1147, over 5662551.28 frames. ], batch size: 128, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:04:16,496 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:04:24,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:04:32,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1636, 1.3018, 1.1350, 1.1398], device='cuda:1'), covar=tensor([0.1512, 0.1176, 0.0882, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1743, 0.1676, 0.1794], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 18:04:42,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651795.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:04:45,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651798.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:04:45,894 INFO [optim.py:369] (1/2) 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 18:05:07,296 INFO [train.py:968] (1/2) Epoch 15, batch 12900, giga_loss[loss=0.2771, simple_loss=0.3538, pruned_loss=0.1002, over 28723.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3578, pruned_loss=0.109, over 5670110.90 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3506, pruned_loss=0.09622, over 5749420.34 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3602, pruned_loss=0.1113, over 5658575.36 frames. ], batch size: 243, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:05:15,458 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651827.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:05:27,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:05:30,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:05:32,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3270, 1.7267, 1.7072, 1.3889], device='cuda:1'), covar=tensor([0.1636, 0.1565, 0.1713, 0.1779], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0729, 0.0685, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 18:05:41,441 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:05:43,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=651851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:06:00,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=651867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:06:01,823 INFO [train.py:968] (1/2) Epoch 15, batch 12950, giga_loss[loss=0.2463, simple_loss=0.3281, pruned_loss=0.08226, over 28916.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3547, pruned_loss=0.1053, over 5674334.40 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3507, pruned_loss=0.09637, over 5750764.93 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3565, pruned_loss=0.1072, over 5663101.83 frames. ], batch size: 106, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:06:14,576 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 15, batch 13000, giga_loss[loss=0.2857, simple_loss=0.3705, pruned_loss=0.1005, over 28699.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3547, pruned_loss=0.1035, over 5656826.25 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.351, pruned_loss=0.09665, over 5734915.76 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3561, pruned_loss=0.105, over 5659300.85 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:07:26,029 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=651951.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:07:39,518 INFO [train.py:968] (1/2) Epoch 15, batch 13050, giga_loss[loss=0.2838, simple_loss=0.3602, pruned_loss=0.1037, over 28853.00 frames. ], tot_loss[loss=0.283, simple_loss=0.356, pruned_loss=0.105, over 5653822.90 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.351, pruned_loss=0.09686, over 5737735.09 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3571, pruned_loss=0.1061, over 5651338.99 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:08:09,175 INFO [optim.py:369] (1/2) 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 18:08:24,915 INFO [train.py:968] (1/2) Epoch 15, batch 13100, giga_loss[loss=0.2206, simple_loss=0.3051, pruned_loss=0.06811, over 29211.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3529, pruned_loss=0.1027, over 5662738.33 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3504, pruned_loss=0.09647, over 5742884.90 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3545, pruned_loss=0.1042, over 5653421.78 frames. ], batch size: 113, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:08:42,604 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:08:47,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8724, 1.0620, 2.8666, 2.7191], device='cuda:1'), covar=tensor([0.1635, 0.2588, 0.0552, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0702, 0.0612, 0.0898, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 18:09:15,077 INFO [train.py:968] (1/2) Epoch 15, batch 13150, giga_loss[loss=0.2366, simple_loss=0.3193, pruned_loss=0.07697, over 28945.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3498, pruned_loss=0.1005, over 5667262.87 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3502, pruned_loss=0.09637, over 5744539.94 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3512, pruned_loss=0.1018, over 5657727.70 frames. ], batch size: 186, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:09:34,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3758, 1.5456, 1.4379, 1.4195], device='cuda:1'), covar=tensor([0.1814, 0.1470, 0.1439, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.1789, 0.1724, 0.1649, 0.1769], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 18:09:43,592 INFO [optim.py:369] (1/2) 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 18:10:02,057 INFO [train.py:968] (1/2) Epoch 15, batch 13200, giga_loss[loss=0.2574, simple_loss=0.3355, pruned_loss=0.08962, over 27687.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.348, pruned_loss=0.09935, over 5675422.87 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3489, pruned_loss=0.09581, over 5750270.28 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3503, pruned_loss=0.101, over 5660280.10 frames. ], batch size: 472, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 18:10:06,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5272, 4.3728, 4.1316, 1.8969], device='cuda:1'), covar=tensor([0.0567, 0.0730, 0.0848, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.1125, 0.1046, 0.0900, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 18:10:26,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:10:48,782 INFO [train.py:968] (1/2) Epoch 15, batch 13250, giga_loss[loss=0.2815, simple_loss=0.3496, pruned_loss=0.1067, over 28256.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3484, pruned_loss=0.09929, over 5677953.54 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.349, pruned_loss=0.09604, over 5754042.98 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3501, pruned_loss=0.1005, over 5660991.33 frames. ], batch size: 368, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 18:11:18,761 INFO [optim.py:369] (1/2) 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 18:11:29,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:11:39,453 INFO [train.py:968] (1/2) Epoch 15, batch 13300, giga_loss[loss=0.2371, simple_loss=0.3259, pruned_loss=0.07414, over 28834.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3462, pruned_loss=0.09746, over 5667641.03 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3488, pruned_loss=0.0959, over 5746571.79 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3478, pruned_loss=0.09858, over 5660270.29 frames. ], batch size: 285, lr: 2.16e-03, grad_scale: 8.0
+2023-03-07 18:12:33,433 INFO [train.py:968] (1/2) Epoch 15, batch 13350, giga_loss[loss=0.2839, simple_loss=0.3577, pruned_loss=0.1051, over 28666.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3431, pruned_loss=0.09503, over 5667822.34 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3485, pruned_loss=0.09575, over 5746404.93 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3446, pruned_loss=0.09604, over 5661848.39 frames. ], batch size: 262, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:12:53,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:12:56,061 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 15, batch 13400, giga_loss[loss=0.2752, simple_loss=0.3324, pruned_loss=0.109, over 26524.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3394, pruned_loss=0.09345, over 5658825.99 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3477, pruned_loss=0.09538, over 5748790.54 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3411, pruned_loss=0.09456, over 5650495.37 frames. ], batch size: 555, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:13:28,293 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:13:32,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:13:47,490 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 13450, giga_loss[loss=0.2597, simple_loss=0.3358, pruned_loss=0.09185, over 28971.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3376, pruned_loss=0.09313, over 5641844.27 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3471, pruned_loss=0.0952, over 5740156.72 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3392, pruned_loss=0.09413, over 5640196.64 frames. ], batch size: 213, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:14:35,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3949, 4.2268, 1.5857, 1.6525], device='cuda:1'), covar=tensor([0.0977, 0.0296, 0.0922, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0526, 0.0357, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 18:14:47,425 INFO [optim.py:369] (1/2) 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 18:14:56,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:15:05,176 INFO [train.py:968] (1/2) Epoch 15, batch 13500, giga_loss[loss=0.2836, simple_loss=0.3565, pruned_loss=0.1053, over 27943.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3381, pruned_loss=0.09397, over 5645257.59 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.347, pruned_loss=0.09524, over 5743875.82 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3394, pruned_loss=0.09469, over 5638415.86 frames. ], batch size: 412, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:15:57,721 INFO [train.py:968] (1/2) Epoch 15, batch 13550, giga_loss[loss=0.2737, simple_loss=0.3297, pruned_loss=0.1089, over 24519.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3393, pruned_loss=0.09392, over 5645802.48 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.346, pruned_loss=0.09469, over 5747654.09 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.341, pruned_loss=0.09498, over 5632107.77 frames. ], batch size: 705, lr: 2.16e-03, grad_scale: 2.0
+2023-03-07 18:15:58,077 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652469.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:16:01,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652472.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:16:25,571 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:16:28,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4870, 1.7927, 1.7346, 1.2693], device='cuda:1'), covar=tensor([0.1828, 0.2607, 0.1525, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.0858, 0.0688, 0.0902, 0.0805], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 18:16:34,280 INFO [optim.py:369] (1/2) 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 18:16:35,577 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652501.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:16:54,889 INFO [train.py:968] (1/2) Epoch 15, batch 13600, giga_loss[loss=0.2435, simple_loss=0.3268, pruned_loss=0.08011, over 28938.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3417, pruned_loss=0.09426, over 5641113.65 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3454, pruned_loss=0.09437, over 5741191.20 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3435, pruned_loss=0.0954, over 5633976.96 frames. ], batch size: 199, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:17:19,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-07 18:17:29,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1320, 1.2188, 3.2839, 2.9240], device='cuda:1'), covar=tensor([0.1581, 0.2654, 0.0506, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0698, 0.0610, 0.0896, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 18:17:35,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:17:37,639 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 13650, giga_loss[loss=0.2622, simple_loss=0.3406, pruned_loss=0.09192, over 28936.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3421, pruned_loss=0.09463, over 5632709.66 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3452, pruned_loss=0.09427, over 5734551.15 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3437, pruned_loss=0.09564, over 5631321.72 frames. ], batch size: 227, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:18:13,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:18:14,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:18:33,287 INFO [optim.py:369] (1/2) 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 18:18:52,238 INFO [train.py:968] (1/2) Epoch 15, batch 13700, libri_loss[loss=0.3109, simple_loss=0.39, pruned_loss=0.1159, over 29291.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3401, pruned_loss=0.09328, over 5647707.05 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3449, pruned_loss=0.09414, over 5738726.33 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3415, pruned_loss=0.09419, over 5640541.06 frames. ], batch size: 94, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:19:29,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-07 18:19:39,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4939, 3.1200, 1.5794, 1.7474], device='cuda:1'), covar=tensor([0.0768, 0.0286, 0.0773, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0527, 0.0358, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 18:19:48,674 INFO [train.py:968] (1/2) Epoch 15, batch 13750, libri_loss[loss=0.2617, simple_loss=0.3196, pruned_loss=0.1019, over 29364.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3384, pruned_loss=0.09171, over 5648728.19 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3441, pruned_loss=0.09397, over 5741860.38 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3401, pruned_loss=0.09255, over 5637018.06 frames. ], batch size: 67, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:20:03,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:20:25,054 INFO [optim.py:369] (1/2) 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 18:20:45,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:20:48,376 INFO [train.py:968] (1/2) Epoch 15, batch 13800, giga_loss[loss=0.2392, simple_loss=0.3266, pruned_loss=0.07594, over 28936.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3368, pruned_loss=0.09005, over 5648499.04 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3441, pruned_loss=0.09394, over 5743393.95 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3381, pruned_loss=0.0907, over 5637103.79 frames. ], batch size: 106, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:20:59,939 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=652728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:20:59,997 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=652728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:21:03,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:21:38,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:21:48,332 INFO [train.py:968] (1/2) Epoch 15, batch 13850, giga_loss[loss=0.2269, simple_loss=0.305, pruned_loss=0.0744, over 28874.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3338, pruned_loss=0.08942, over 5657134.21 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3437, pruned_loss=0.09377, over 5746240.67 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3349, pruned_loss=0.09, over 5644042.15 frames. ], batch size: 120, lr: 2.16e-03, grad_scale: 4.0
+2023-03-07 18:22:24,765 INFO [optim.py:369] (1/2) 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 18:22:46,461 INFO [train.py:968] (1/2) Epoch 15, batch 13900, giga_loss[loss=0.2617, simple_loss=0.3244, pruned_loss=0.09947, over 28490.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3328, pruned_loss=0.08906, over 5651365.81 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3438, pruned_loss=0.09384, over 5739178.01 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3336, pruned_loss=0.0894, over 5645510.92 frames. ], batch size: 78, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:23:32,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-07 18:23:32,790 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=652861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:23:42,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=652867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:23:43,727 INFO [train.py:968] (1/2) Epoch 15, batch 13950, giga_loss[loss=0.2688, simple_loss=0.3533, pruned_loss=0.09214, over 28649.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3336, pruned_loss=0.08893, over 5664397.77 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3434, pruned_loss=0.09369, over 5741748.89 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3343, pruned_loss=0.08926, over 5656349.87 frames. ], batch size: 262, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:24:06,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0721, 2.4093, 2.1300, 2.1139], device='cuda:1'), covar=tensor([0.1825, 0.2172, 0.1931, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0722, 0.0677, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 18:24:11,980 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:24:26,776 INFO [optim.py:369] (1/2) 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 18:24:45,987 INFO [train.py:968] (1/2) Epoch 15, batch 14000, libri_loss[loss=0.2884, simple_loss=0.3672, pruned_loss=0.1048, over 27852.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3372, pruned_loss=0.09002, over 5661394.13 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3438, pruned_loss=0.09399, over 5730495.15 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3373, pruned_loss=0.08991, over 5664457.09 frames. ], batch size: 116, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 18:25:45,747 INFO [train.py:968] (1/2) Epoch 15, batch 14050, giga_loss[loss=0.2568, simple_loss=0.3352, pruned_loss=0.08914, over 28418.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3362, pruned_loss=0.08978, over 5660944.24 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3443, pruned_loss=0.09464, over 5728329.71 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3356, pruned_loss=0.089, over 5663877.22 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:26:34,237 INFO [optim.py:369] (1/2) 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 18:26:43,081 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:26:45,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:26:51,766 INFO [train.py:968] (1/2) Epoch 15, batch 14100, giga_loss[loss=0.2945, simple_loss=0.362, pruned_loss=0.1135, over 28502.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3355, pruned_loss=0.08962, over 5672814.39 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3443, pruned_loss=0.09469, over 5731183.81 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3349, pruned_loss=0.08883, over 5671270.93 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:26:52,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6025, 1.8789, 1.4322, 1.9491], device='cuda:1'), covar=tensor([0.2653, 0.2614, 0.3010, 0.2446], device='cuda:1'), in_proj_covar=tensor([0.1386, 0.1017, 0.1236, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 18:27:16,947 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:27:34,040 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-07 18:27:37,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:27:54,883 INFO [train.py:968] (1/2) Epoch 15, batch 14150, giga_loss[loss=0.2331, simple_loss=0.2945, pruned_loss=0.08586, over 24470.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.338, pruned_loss=0.09132, over 5652627.04 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3446, pruned_loss=0.09502, over 5726723.52 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.337, pruned_loss=0.09027, over 5653097.22 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:28:30,616 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=653096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:28:37,014 INFO [optim.py:369] (1/2) 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 18:28:39,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:28:58,721 INFO [train.py:968] (1/2) Epoch 15, batch 14200, libri_loss[loss=0.264, simple_loss=0.3482, pruned_loss=0.08984, over 25969.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3412, pruned_loss=0.09063, over 5649229.67 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3448, pruned_loss=0.09523, over 5726712.37 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3402, pruned_loss=0.08953, over 5648540.80 frames. ], batch size: 136, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:29:12,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-07 18:29:53,883 INFO [train.py:968] (1/2) Epoch 15, batch 14250, giga_loss[loss=0.2831, simple_loss=0.3548, pruned_loss=0.1057, over 26932.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3408, pruned_loss=0.0888, over 5648495.10 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3444, pruned_loss=0.09493, over 5732367.20 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3402, pruned_loss=0.08802, over 5640603.61 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:30:32,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:30:33,763 INFO [optim.py:369] (1/2) 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 18:30:34,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:30:52,288 INFO [train.py:968] (1/2) Epoch 15, batch 14300, giga_loss[loss=0.2594, simple_loss=0.3463, pruned_loss=0.08627, over 28789.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3399, pruned_loss=0.087, over 5653931.52 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3442, pruned_loss=0.09487, over 5732440.13 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3396, pruned_loss=0.08628, over 5646276.28 frames. ], batch size: 263, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:31:07,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:31:23,597 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=653244.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:31:24,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:31:28,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:31:50,258 INFO [train.py:968] (1/2) Epoch 15, batch 14350, giga_loss[loss=0.3765, simple_loss=0.4156, pruned_loss=0.1686, over 26948.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3418, pruned_loss=0.08916, over 5664868.10 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3438, pruned_loss=0.09464, over 5737099.35 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08858, over 5651905.36 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:31:59,705 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:32:13,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-07 18:32:28,627 INFO [optim.py:369] (1/2) 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 18:32:47,432 INFO [train.py:968] (1/2) Epoch 15, batch 14400, giga_loss[loss=0.2592, simple_loss=0.3386, pruned_loss=0.08989, over 28649.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3412, pruned_loss=0.09042, over 5672071.32 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3431, pruned_loss=0.09433, over 5742566.74 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3418, pruned_loss=0.09005, over 5654824.89 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 18:33:05,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 18:33:53,869 INFO [train.py:968] (1/2) Epoch 15, batch 14450, libri_loss[loss=0.2133, simple_loss=0.2916, pruned_loss=0.06751, over 29653.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3412, pruned_loss=0.09108, over 5678297.79 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3422, pruned_loss=0.09387, over 5748252.07 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3424, pruned_loss=0.09113, over 5656691.92 frames. ], batch size: 73, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:34:43,203 INFO [optim.py:369] (1/2) 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 18:35:16,773 INFO [train.py:968] (1/2) Epoch 15, batch 14500, giga_loss[loss=0.2364, simple_loss=0.3174, pruned_loss=0.07765, over 28366.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3381, pruned_loss=0.08922, over 5688536.33 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3418, pruned_loss=0.09363, over 5749976.66 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3394, pruned_loss=0.08944, over 5669099.15 frames. ], batch size: 369, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:36:24,180 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-07 18:36:25,977 INFO [train.py:968] (1/2) Epoch 15, batch 14550, libri_loss[loss=0.329, simple_loss=0.385, pruned_loss=0.1365, over 19860.00 frames. ], tot_loss[loss=0.255, simple_loss=0.335, pruned_loss=0.08747, over 5668508.41 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3417, pruned_loss=0.09358, over 5741981.35 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.336, pruned_loss=0.08755, over 5659842.82 frames. ], batch size: 187, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:36:28,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:37:12,219 INFO [optim.py:369] (1/2) 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 18:37:31,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-07 18:37:32,977 INFO [train.py:968] (1/2) Epoch 15, batch 14600, giga_loss[loss=0.2681, simple_loss=0.3405, pruned_loss=0.0979, over 28689.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3321, pruned_loss=0.08612, over 5665890.57 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3417, pruned_loss=0.09358, over 5739356.45 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3329, pruned_loss=0.08617, over 5660878.05 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:38:09,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7336, 2.4838, 1.5336, 0.9704], device='cuda:1'), covar=tensor([0.6668, 0.3578, 0.3664, 0.5642], device='cuda:1'), in_proj_covar=tensor([0.1626, 0.1550, 0.1531, 0.1334], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 18:38:22,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2136, 1.2131, 4.0945, 3.3436], device='cuda:1'), covar=tensor([0.1720, 0.2692, 0.0466, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0690, 0.0605, 0.0886, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 18:38:31,612 INFO [train.py:968] (1/2) Epoch 15, batch 14650, giga_loss[loss=0.2896, simple_loss=0.3557, pruned_loss=0.1118, over 26909.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3355, pruned_loss=0.0881, over 5669510.43 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3414, pruned_loss=0.09348, over 5733395.98 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3361, pruned_loss=0.08802, over 5669211.41 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:39:13,860 INFO [optim.py:369] (1/2) 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 18:39:27,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:39:30,791 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 14700, giga_loss[loss=0.3369, simple_loss=0.4065, pruned_loss=0.1336, over 28704.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.338, pruned_loss=0.08953, over 5673294.12 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.341, pruned_loss=0.09322, over 5736538.12 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3388, pruned_loss=0.08956, over 5668721.40 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:39:32,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=653619.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:40:05,784 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:40:35,019 INFO [train.py:968] (1/2) Epoch 15, batch 14750, giga_loss[loss=0.2813, simple_loss=0.349, pruned_loss=0.1068, over 27669.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3362, pruned_loss=0.0899, over 5674627.17 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3406, pruned_loss=0.09308, over 5734245.64 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.08999, over 5671626.76 frames. ], batch size: 474, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:41:16,176 INFO [optim.py:369] (1/2) 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 18:41:32,980 INFO [train.py:968] (1/2) Epoch 15, batch 14800, giga_loss[loss=0.2664, simple_loss=0.3458, pruned_loss=0.0935, over 28683.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3372, pruned_loss=0.09113, over 5663160.06 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3398, pruned_loss=0.09268, over 5727086.26 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3386, pruned_loss=0.09148, over 5665450.63 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 18:42:07,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1919, 1.1747, 3.6582, 3.1682], device='cuda:1'), covar=tensor([0.1627, 0.2687, 0.0440, 0.0965], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0608, 0.0889, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 18:42:27,380 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=653762.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 18:42:32,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=653765.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:42:37,249 INFO [train.py:968] (1/2) Epoch 15, batch 14850, giga_loss[loss=0.27, simple_loss=0.3599, pruned_loss=0.09003, over 28823.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3375, pruned_loss=0.09098, over 5662871.97 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3396, pruned_loss=0.0926, over 5726571.83 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3387, pruned_loss=0.09133, over 5664832.97 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:42:43,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3112, 1.6961, 1.2960, 1.2836], device='cuda:1'), covar=tensor([0.2552, 0.2337, 0.2789, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1367, 0.1000, 0.1219, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 18:43:11,557 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=653794.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 18:43:27,349 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 14900, giga_loss[loss=0.2968, simple_loss=0.3883, pruned_loss=0.1026, over 28981.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3395, pruned_loss=0.09093, over 5666656.40 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3392, pruned_loss=0.09238, over 5729647.58 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3408, pruned_loss=0.09139, over 5663483.64 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:45:01,815 INFO [train.py:968] (1/2) Epoch 15, batch 14950, giga_loss[loss=0.2616, simple_loss=0.3388, pruned_loss=0.09225, over 28964.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.339, pruned_loss=0.09067, over 5666997.52 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3384, pruned_loss=0.092, over 5730270.74 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3408, pruned_loss=0.09133, over 5662077.15 frames. ], batch size: 284, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:45:20,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3763, 1.5330, 1.4046, 1.3901], device='cuda:1'), covar=tensor([0.2255, 0.1730, 0.1649, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1699, 0.1625, 0.1752], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 18:45:59,903 INFO [optim.py:369] (1/2) 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 18:46:18,597 INFO [train.py:968] (1/2) Epoch 15, batch 15000, giga_loss[loss=0.2304, simple_loss=0.3096, pruned_loss=0.07557, over 28770.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3363, pruned_loss=0.09002, over 5678876.41 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3386, pruned_loss=0.09225, over 5724869.33 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3375, pruned_loss=0.09028, over 5678516.41 frames. ], batch size: 119, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:46:18,597 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 18:46:28,732 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 18:46:40,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3611, 1.5642, 1.3132, 1.5586], device='cuda:1'), covar=tensor([0.0760, 0.0329, 0.0333, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 18:47:07,306 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=653947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:47:38,552 INFO [train.py:968] (1/2) Epoch 15, batch 15050, giga_loss[loss=0.2575, simple_loss=0.3321, pruned_loss=0.0915, over 28377.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3302, pruned_loss=0.08749, over 5684129.52 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3383, pruned_loss=0.09208, over 5730097.45 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3313, pruned_loss=0.08776, over 5677886.42 frames. ], batch size: 368, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:48:18,586 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 15100, giga_loss[loss=0.2372, simple_loss=0.3249, pruned_loss=0.0747, over 28732.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3302, pruned_loss=0.08798, over 5678199.96 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3377, pruned_loss=0.09175, over 5726075.69 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3314, pruned_loss=0.08834, over 5675608.41 frames. ], batch size: 262, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:48:47,214 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=654027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:48:54,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5818, 2.1402, 1.8418, 1.5098], device='cuda:1'), covar=tensor([0.2443, 0.1675, 0.1786, 0.2089], device='cuda:1'), in_proj_covar=tensor([0.1778, 0.1686, 0.1617, 0.1747], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 18:49:36,357 INFO [train.py:968] (1/2) Epoch 15, batch 15150, giga_loss[loss=0.2649, simple_loss=0.3435, pruned_loss=0.0931, over 28915.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3312, pruned_loss=0.08923, over 5675592.88 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3373, pruned_loss=0.09158, over 5730019.05 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3323, pruned_loss=0.08962, over 5669091.54 frames. ], batch size: 284, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:50:16,010 INFO [optim.py:369] (1/2) 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 18:50:37,452 INFO [train.py:968] (1/2) Epoch 15, batch 15200, giga_loss[loss=0.2826, simple_loss=0.3527, pruned_loss=0.1062, over 28696.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3294, pruned_loss=0.08802, over 5662056.47 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3373, pruned_loss=0.09161, over 5733404.58 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3302, pruned_loss=0.08824, over 5652684.53 frames. ], batch size: 262, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:51:09,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3052, 1.2376, 4.0141, 3.2378], device='cuda:1'), covar=tensor([0.1639, 0.2786, 0.0444, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0696, 0.0609, 0.0886, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 18:51:34,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2838, 1.2355, 3.7089, 3.0109], device='cuda:1'), covar=tensor([0.1557, 0.2659, 0.0467, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0608, 0.0885, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 18:51:38,002 INFO [train.py:968] (1/2) Epoch 15, batch 15250, giga_loss[loss=0.2194, simple_loss=0.3091, pruned_loss=0.06481, over 28824.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3287, pruned_loss=0.08638, over 5665994.09 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3374, pruned_loss=0.09175, over 5727948.24 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08628, over 5662512.04 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:52:07,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4636, 1.5485, 1.3049, 1.6193], device='cuda:1'), covar=tensor([0.0762, 0.0295, 0.0337, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-07 18:52:27,972 INFO [optim.py:369] (1/2) 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 18:52:47,407 INFO [train.py:968] (1/2) Epoch 15, batch 15300, giga_loss[loss=0.2493, simple_loss=0.3279, pruned_loss=0.08528, over 28641.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3262, pruned_loss=0.08511, over 5660838.11 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3369, pruned_loss=0.09149, over 5730460.54 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3267, pruned_loss=0.08519, over 5655124.64 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:53:35,934 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 18:54:00,459 INFO [train.py:968] (1/2) Epoch 15, batch 15350, giga_loss[loss=0.2425, simple_loss=0.3236, pruned_loss=0.08071, over 28926.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3266, pruned_loss=0.08485, over 5673774.70 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.09152, over 5730958.24 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3268, pruned_loss=0.08485, over 5668528.31 frames. ], batch size: 213, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:54:49,294 INFO [optim.py:369] (1/2) 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 18:54:51,667 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 15400, giga_loss[loss=0.2806, simple_loss=0.3508, pruned_loss=0.1052, over 28475.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3281, pruned_loss=0.08542, over 5686403.63 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.337, pruned_loss=0.09148, over 5734741.83 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3281, pruned_loss=0.08534, over 5678170.42 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:55:13,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:55:20,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9858, 1.8166, 4.4674, 3.8487], device='cuda:1'), covar=tensor([0.1284, 0.2484, 0.0395, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0693, 0.0608, 0.0885, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 18:55:35,842 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=654339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:56:11,919 INFO [train.py:968] (1/2) Epoch 15, batch 15450, libri_loss[loss=0.2071, simple_loss=0.2852, pruned_loss=0.06449, over 29375.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3283, pruned_loss=0.0861, over 5683781.54 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3361, pruned_loss=0.09105, over 5729307.21 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.329, pruned_loss=0.08634, over 5680301.02 frames. ], batch size: 67, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:56:18,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2234, 3.1285, 1.4927, 1.4061], device='cuda:1'), covar=tensor([0.1015, 0.0501, 0.0896, 0.1402], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0522, 0.0357, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 18:57:00,836 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:57:05,663 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 15500, giga_loss[loss=0.2507, simple_loss=0.3389, pruned_loss=0.0813, over 28948.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3269, pruned_loss=0.08539, over 5678402.54 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.336, pruned_loss=0.09097, over 5731106.45 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3274, pruned_loss=0.0856, over 5673673.32 frames. ], batch size: 145, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:58:18,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:58:22,147 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:58:22,428 INFO [train.py:968] (1/2) Epoch 15, batch 15550, giga_loss[loss=0.2057, simple_loss=0.2821, pruned_loss=0.06466, over 24146.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3284, pruned_loss=0.08487, over 5665249.77 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3362, pruned_loss=0.091, over 5732480.85 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3286, pruned_loss=0.08491, over 5659164.33 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 18:58:43,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-07 18:58:58,682 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 18:59:11,043 INFO [optim.py:369] (1/2) 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 18:59:24,016 INFO [train.py:968] (1/2) Epoch 15, batch 15600, giga_loss[loss=0.2664, simple_loss=0.3509, pruned_loss=0.09092, over 28947.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3307, pruned_loss=0.08537, over 5662396.64 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3358, pruned_loss=0.09073, over 5732837.89 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.331, pruned_loss=0.0855, over 5655327.27 frames. ], batch size: 285, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 18:59:24,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-07 18:59:56,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:00:01,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:00:24,412 INFO [train.py:968] (1/2) Epoch 15, batch 15650, giga_loss[loss=0.2522, simple_loss=0.3329, pruned_loss=0.08569, over 28950.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3334, pruned_loss=0.08685, over 5664291.95 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3359, pruned_loss=0.09084, over 5733413.79 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3336, pruned_loss=0.08681, over 5657282.05 frames. ], batch size: 213, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:00:35,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:01:09,969 INFO [optim.py:369] (1/2) 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 19:01:19,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3222, 1.6022, 1.3540, 1.5294], device='cuda:1'), covar=tensor([0.0754, 0.0318, 0.0327, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-07 19:01:27,689 INFO [train.py:968] (1/2) Epoch 15, batch 15700, giga_loss[loss=0.2462, simple_loss=0.322, pruned_loss=0.08523, over 28722.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3331, pruned_loss=0.0874, over 5654255.51 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3356, pruned_loss=0.09072, over 5735889.89 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3334, pruned_loss=0.08742, over 5645364.95 frames. ], batch size: 92, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:02:26,431 INFO [train.py:968] (1/2) Epoch 15, batch 15750, giga_loss[loss=0.3215, simple_loss=0.3869, pruned_loss=0.1281, over 28040.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3309, pruned_loss=0.08618, over 5658581.45 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3354, pruned_loss=0.09058, over 5739075.67 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3313, pruned_loss=0.08621, over 5646286.24 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:02:42,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:03:14,108 INFO [optim.py:369] (1/2) 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 19:03:26,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:03:30,820 INFO [train.py:968] (1/2) Epoch 15, batch 15800, giga_loss[loss=0.2336, simple_loss=0.3183, pruned_loss=0.07447, over 28063.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3305, pruned_loss=0.08639, over 5662390.84 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3353, pruned_loss=0.09063, over 5742824.13 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3308, pruned_loss=0.08629, over 5647709.85 frames. ], batch size: 412, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:04:14,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-07 19:04:16,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-07 19:04:28,600 INFO [train.py:968] (1/2) Epoch 15, batch 15850, giga_loss[loss=0.2424, simple_loss=0.3193, pruned_loss=0.08272, over 28561.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3288, pruned_loss=0.08605, over 5674005.17 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3351, pruned_loss=0.09063, over 5746039.28 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3291, pruned_loss=0.08584, over 5657270.58 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:04:43,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=654781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:05:13,827 INFO [optim.py:369] (1/2) 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 19:05:30,251 INFO [train.py:968] (1/2) Epoch 15, batch 15900, giga_loss[loss=0.2682, simple_loss=0.3459, pruned_loss=0.09528, over 27648.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3299, pruned_loss=0.08596, over 5668789.93 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3352, pruned_loss=0.09073, over 5736083.90 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3299, pruned_loss=0.0856, over 5662082.29 frames. ], batch size: 474, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:05:40,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:05:44,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:20,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:20,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:25,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:06:37,051 INFO [train.py:968] (1/2) Epoch 15, batch 15950, giga_loss[loss=0.2808, simple_loss=0.349, pruned_loss=0.1062, over 28939.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3306, pruned_loss=0.0864, over 5668629.91 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3345, pruned_loss=0.09039, over 5740634.98 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3311, pruned_loss=0.08633, over 5657727.19 frames. ], batch size: 213, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:07:03,714 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:07:39,522 INFO [train.py:968] (1/2) Epoch 15, batch 16000, libri_loss[loss=0.2467, simple_loss=0.3342, pruned_loss=0.07958, over 29646.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3314, pruned_loss=0.08757, over 5661984.49 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.334, pruned_loss=0.09021, over 5737101.00 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3322, pruned_loss=0.0876, over 5653810.17 frames. ], batch size: 91, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:07:47,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=654924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:07:49,559 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=654927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:08:20,598 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=654956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:08:36,406 INFO [train.py:968] (1/2) Epoch 15, batch 16050, giga_loss[loss=0.2913, simple_loss=0.3772, pruned_loss=0.1027, over 28799.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3359, pruned_loss=0.08999, over 5654565.31 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3338, pruned_loss=0.09008, over 5731004.67 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3367, pruned_loss=0.09009, over 5652046.42 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:08:56,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 19:09:18,430 INFO [optim.py:369] (1/2) 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 19:09:30,538 INFO [train.py:968] (1/2) Epoch 15, batch 16100, giga_loss[loss=0.2694, simple_loss=0.3554, pruned_loss=0.09167, over 28546.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3367, pruned_loss=0.08981, over 5658188.44 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3335, pruned_loss=0.09009, over 5735748.47 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3378, pruned_loss=0.08989, over 5649096.44 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:10:33,810 INFO [train.py:968] (1/2) Epoch 15, batch 16150, giga_loss[loss=0.225, simple_loss=0.3164, pruned_loss=0.06684, over 28896.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3381, pruned_loss=0.09039, over 5654934.08 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3333, pruned_loss=0.08987, over 5737942.11 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3392, pruned_loss=0.09065, over 5644805.91 frames. ], batch size: 164, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:10:53,494 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:11:42,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3218, 1.1713, 1.1200, 1.4775], device='cuda:1'), covar=tensor([0.0741, 0.0345, 0.0334, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-07 19:11:45,936 INFO [train.py:968] (1/2) Epoch 15, batch 16200, giga_loss[loss=0.1971, simple_loss=0.2884, pruned_loss=0.05288, over 29023.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.336, pruned_loss=0.08921, over 5663547.26 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3331, pruned_loss=0.0898, over 5741919.47 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3372, pruned_loss=0.08948, over 5650016.82 frames. ], batch size: 155, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:12:18,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3926, 1.7126, 1.6584, 1.2145], device='cuda:1'), covar=tensor([0.1754, 0.2500, 0.1468, 0.1754], device='cuda:1'), in_proj_covar=tensor([0.0856, 0.0683, 0.0900, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 19:12:49,202 INFO [train.py:968] (1/2) Epoch 15, batch 16250, giga_loss[loss=0.2531, simple_loss=0.3381, pruned_loss=0.08406, over 28913.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3335, pruned_loss=0.08771, over 5667468.24 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3326, pruned_loss=0.08954, over 5741384.99 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3349, pruned_loss=0.08814, over 5655235.40 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:13:16,829 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655190.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:13:35,427 INFO [optim.py:369] (1/2) 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 19:13:50,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3086, 1.2573, 1.1927, 1.5762], device='cuda:1'), covar=tensor([0.0743, 0.0326, 0.0331, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-07 19:13:51,544 INFO [train.py:968] (1/2) Epoch 15, batch 16300, libri_loss[loss=0.2807, simple_loss=0.3557, pruned_loss=0.1028, over 29749.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3342, pruned_loss=0.08841, over 5676338.12 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.333, pruned_loss=0.08968, over 5747483.03 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3349, pruned_loss=0.08855, over 5658514.54 frames. ], batch size: 87, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:14:54,307 INFO [train.py:968] (1/2) Epoch 15, batch 16350, giga_loss[loss=0.2418, simple_loss=0.3215, pruned_loss=0.08109, over 28583.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3328, pruned_loss=0.08902, over 5664654.13 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.333, pruned_loss=0.08971, over 5750119.02 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3334, pruned_loss=0.08908, over 5647120.16 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:15:17,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-07 19:15:38,565 INFO [optim.py:369] (1/2) 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 19:15:44,009 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655311.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:15:53,217 INFO [train.py:968] (1/2) Epoch 15, batch 16400, giga_loss[loss=0.2611, simple_loss=0.344, pruned_loss=0.08904, over 28663.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3309, pruned_loss=0.08798, over 5666654.86 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3322, pruned_loss=0.08929, over 5751559.29 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3321, pruned_loss=0.0884, over 5650073.53 frames. ], batch size: 262, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:16:46,211 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 16450, giga_loss[loss=0.2142, simple_loss=0.3068, pruned_loss=0.06085, over 28955.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.33, pruned_loss=0.08631, over 5677691.77 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3319, pruned_loss=0.08913, over 5755809.74 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3312, pruned_loss=0.08673, over 5658606.75 frames. ], batch size: 155, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:17:25,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4256, 2.0285, 1.5323, 0.5977], device='cuda:1'), covar=tensor([0.3834, 0.2531, 0.3308, 0.4987], device='cuda:1'), in_proj_covar=tensor([0.1601, 0.1530, 0.1516, 0.1325], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 19:17:39,886 INFO [optim.py:369] (1/2) 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 19:17:51,535 INFO [train.py:968] (1/2) Epoch 15, batch 16500, libri_loss[loss=0.2462, simple_loss=0.3335, pruned_loss=0.07943, over 29537.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3303, pruned_loss=0.08484, over 5685622.13 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3319, pruned_loss=0.08901, over 5756892.24 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3313, pruned_loss=0.08519, over 5667017.24 frames. ], batch size: 84, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:18:33,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:18:50,499 INFO [train.py:968] (1/2) Epoch 15, batch 16550, giga_loss[loss=0.2419, simple_loss=0.3315, pruned_loss=0.07611, over 28377.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3328, pruned_loss=0.08452, over 5691583.89 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3317, pruned_loss=0.08884, over 5759886.85 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3336, pruned_loss=0.08486, over 5672949.68 frames. ], batch size: 368, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:19:26,659 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:19:30,511 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 16600, giga_loss[loss=0.3254, simple_loss=0.3666, pruned_loss=0.1421, over 26732.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3336, pruned_loss=0.0852, over 5678198.44 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3317, pruned_loss=0.08875, over 5753356.83 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3344, pruned_loss=0.08542, over 5665758.89 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:20:40,808 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:20:43,993 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 16650, giga_loss[loss=0.2281, simple_loss=0.3118, pruned_loss=0.07222, over 28442.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3328, pruned_loss=0.08464, over 5668374.99 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3314, pruned_loss=0.0886, over 5752449.74 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3337, pruned_loss=0.0849, over 5658165.24 frames. ], batch size: 85, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:21:10,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2421, 1.5831, 1.2583, 0.9557], device='cuda:1'), covar=tensor([0.2365, 0.2217, 0.2551, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.1369, 0.1000, 0.1222, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 19:21:28,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:21:30,341 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:21:30,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-07 19:21:38,847 INFO [optim.py:369] (1/2) 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 19:21:43,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1908, 1.3403, 1.2508, 1.2348], device='cuda:1'), covar=tensor([0.1862, 0.1507, 0.1262, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.1786, 0.1679, 0.1606, 0.1744], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 19:21:53,915 INFO [train.py:968] (1/2) Epoch 15, batch 16700, giga_loss[loss=0.2824, simple_loss=0.3455, pruned_loss=0.1096, over 26977.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3318, pruned_loss=0.08425, over 5662127.04 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3311, pruned_loss=0.08839, over 5757335.66 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3328, pruned_loss=0.08452, over 5646938.98 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:22:04,828 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:22:11,489 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:22:25,183 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 16750, giga_loss[loss=0.2678, simple_loss=0.3513, pruned_loss=0.09216, over 28686.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3311, pruned_loss=0.08338, over 5669420.72 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3305, pruned_loss=0.08815, over 5760601.14 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3325, pruned_loss=0.08374, over 5652583.64 frames. ], batch size: 307, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:23:26,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:23:27,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655686.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:23:58,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655708.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:23:58,357 INFO [optim.py:369] (1/2) 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 19:24:01,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655711.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:24:12,617 INFO [train.py:968] (1/2) Epoch 15, batch 16800, libri_loss[loss=0.2676, simple_loss=0.3443, pruned_loss=0.09548, over 29522.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.332, pruned_loss=0.08373, over 5665856.49 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3308, pruned_loss=0.0883, over 5761011.80 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3329, pruned_loss=0.08372, over 5649252.22 frames. ], batch size: 89, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:24:38,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:24:42,438 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655740.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:25:18,209 INFO [train.py:968] (1/2) Epoch 15, batch 16850, giga_loss[loss=0.2676, simple_loss=0.3552, pruned_loss=0.09003, over 28865.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3359, pruned_loss=0.08562, over 5674975.90 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3309, pruned_loss=0.08838, over 5763555.32 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3367, pruned_loss=0.08546, over 5656662.77 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:25:20,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2356, 1.3476, 1.2650, 1.3606], device='cuda:1'), covar=tensor([0.0806, 0.0330, 0.0334, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-07 19:26:08,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=655806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:26:10,716 INFO [optim.py:369] (1/2) 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 19:26:27,093 INFO [train.py:968] (1/2) Epoch 15, batch 16900, giga_loss[loss=0.2685, simple_loss=0.3494, pruned_loss=0.09381, over 28939.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3355, pruned_loss=0.08521, over 5671996.06 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3305, pruned_loss=0.08818, over 5756372.28 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3366, pruned_loss=0.08521, over 5662017.62 frames. ], batch size: 112, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:26:42,192 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655829.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:26:45,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655832.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:27:04,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-03-07 19:27:28,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655861.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:27:37,727 INFO [train.py:968] (1/2) Epoch 15, batch 16950, libri_loss[loss=0.2189, simple_loss=0.3028, pruned_loss=0.06755, over 29566.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3347, pruned_loss=0.08562, over 5666227.96 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3301, pruned_loss=0.0879, over 5750801.18 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.336, pruned_loss=0.08583, over 5660638.69 frames. ], batch size: 79, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:27:53,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:27:53,720 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=655880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:27:57,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=655883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:28:35,046 INFO [optim.py:369] (1/2) 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 19:28:38,309 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=655912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:28:45,895 INFO [train.py:968] (1/2) Epoch 15, batch 17000, giga_loss[loss=0.2167, simple_loss=0.3072, pruned_loss=0.06315, over 29117.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3327, pruned_loss=0.08428, over 5676165.46 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3302, pruned_loss=0.08796, over 5752618.22 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3337, pruned_loss=0.08431, over 5667990.13 frames. ], batch size: 200, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:29:14,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=655937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:29:26,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6611, 1.7339, 1.2471, 1.4337], device='cuda:1'), covar=tensor([0.0912, 0.0695, 0.1056, 0.1233], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0431, 0.0500, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 19:29:54,806 INFO [train.py:968] (1/2) Epoch 15, batch 17050, giga_loss[loss=0.2379, simple_loss=0.3187, pruned_loss=0.07852, over 29031.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3312, pruned_loss=0.083, over 5671586.26 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3303, pruned_loss=0.08788, over 5753291.50 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.332, pruned_loss=0.08299, over 5663085.32 frames. ], batch size: 186, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:30:32,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:30:42,683 INFO [optim.py:369] (1/2) 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 19:30:49,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656014.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:30:55,628 INFO [train.py:968] (1/2) Epoch 15, batch 17100, giga_loss[loss=0.2217, simple_loss=0.3115, pruned_loss=0.06598, over 28965.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3312, pruned_loss=0.08295, over 5668748.19 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3299, pruned_loss=0.08773, over 5747448.27 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3322, pruned_loss=0.08297, over 5665023.72 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:30:59,221 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:01,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:37,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:43,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:31:53,351 INFO [train.py:968] (1/2) Epoch 15, batch 17150, giga_loss[loss=0.336, simple_loss=0.3835, pruned_loss=0.1443, over 24623.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3351, pruned_loss=0.08589, over 5668133.71 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3297, pruned_loss=0.08767, over 5750927.16 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.336, pruned_loss=0.08587, over 5660558.24 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:32:09,435 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:32:40,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1742, 3.0207, 2.8426, 1.3866], device='cuda:1'), covar=tensor([0.0945, 0.1003, 0.0912, 0.2472], device='cuda:1'), in_proj_covar=tensor([0.1086, 0.0996, 0.0865, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 19:32:40,579 INFO [optim.py:369] (1/2) 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 19:32:43,257 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 17200, libri_loss[loss=0.21, simple_loss=0.2947, pruned_loss=0.06261, over 29567.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3354, pruned_loss=0.0864, over 5667548.65 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3294, pruned_loss=0.08748, over 5743327.01 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3367, pruned_loss=0.08655, over 5665069.46 frames. ], batch size: 76, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:33:17,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:33:20,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:33:34,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656157.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:33:36,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656160.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:33:45,696 INFO [train.py:968] (1/2) Epoch 15, batch 17250, giga_loss[loss=0.2632, simple_loss=0.3365, pruned_loss=0.09491, over 29047.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3317, pruned_loss=0.08541, over 5662646.74 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3291, pruned_loss=0.08732, over 5743200.52 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.333, pruned_loss=0.08565, over 5660268.82 frames. ], batch size: 128, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:33:55,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:34:00,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=656181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:34:12,826 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656189.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:34:28,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:34:32,339 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:34:48,143 INFO [train.py:968] (1/2) Epoch 15, batch 17300, giga_loss[loss=0.236, simple_loss=0.2973, pruned_loss=0.08732, over 24670.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3327, pruned_loss=0.08733, over 5644528.08 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3293, pruned_loss=0.08752, over 5734790.49 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3336, pruned_loss=0.08734, over 5648938.54 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:35:05,842 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:35:34,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7851, 4.5758, 4.2957, 2.0648], device='cuda:1'), covar=tensor([0.0475, 0.0650, 0.0702, 0.1977], device='cuda:1'), in_proj_covar=tensor([0.1092, 0.0997, 0.0865, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 19:35:34,753 INFO [train.py:968] (1/2) Epoch 15, batch 17350, giga_loss[loss=0.3094, simple_loss=0.3791, pruned_loss=0.1199, over 28806.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3347, pruned_loss=0.0888, over 5655957.41 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3287, pruned_loss=0.08712, over 5741466.64 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3363, pruned_loss=0.08923, over 5649158.11 frames. ], batch size: 99, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:35:38,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3549, 1.1928, 3.6074, 3.0342], device='cuda:1'), covar=tensor([0.1551, 0.2729, 0.0475, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0697, 0.0609, 0.0888, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 19:35:39,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1694, 1.4535, 1.3490, 1.3035], device='cuda:1'), covar=tensor([0.1572, 0.1532, 0.2018, 0.1513], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0721, 0.0673, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 19:36:20,059 INFO [optim.py:369] (1/2) 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 19:36:26,487 INFO [train.py:968] (1/2) Epoch 15, batch 17400, giga_loss[loss=0.3812, simple_loss=0.4398, pruned_loss=0.1613, over 29015.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3449, pruned_loss=0.09455, over 5670552.69 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3285, pruned_loss=0.08713, over 5745964.41 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3466, pruned_loss=0.09501, over 5658976.02 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:36:31,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=656324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:36:35,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=656327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:37:01,843 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=656356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:37:12,667 INFO [train.py:968] (1/2) Epoch 15, batch 17450, giga_loss[loss=0.2931, simple_loss=0.366, pruned_loss=0.1101, over 28704.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3522, pruned_loss=0.09899, over 5672345.35 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3287, pruned_loss=0.0873, over 5747570.66 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3536, pruned_loss=0.09931, over 5661015.39 frames. ], batch size: 262, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:37:47,848 INFO [optim.py:369] (1/2) 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 19:37:56,799 INFO [train.py:968] (1/2) Epoch 15, batch 17500, giga_loss[loss=0.2195, simple_loss=0.3084, pruned_loss=0.06534, over 28861.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3504, pruned_loss=0.09873, over 5674417.98 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3286, pruned_loss=0.08703, over 5750271.68 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3522, pruned_loss=0.0995, over 5661296.08 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:37:57,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-07 19:38:38,982 INFO [train.py:968] (1/2) Epoch 15, batch 17550, giga_loss[loss=0.2274, simple_loss=0.3064, pruned_loss=0.07416, over 28499.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3445, pruned_loss=0.09629, over 5676472.63 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3283, pruned_loss=0.08687, over 5743441.18 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3467, pruned_loss=0.09731, over 5670569.37 frames. ], batch size: 65, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:39:15,448 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 17600, giga_loss[loss=0.2359, simple_loss=0.3106, pruned_loss=0.08058, over 28738.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3378, pruned_loss=0.09351, over 5685606.72 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3284, pruned_loss=0.08686, over 5744474.88 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3398, pruned_loss=0.09451, over 5678492.57 frames. ], batch size: 284, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:39:38,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5075, 1.6760, 1.6100, 1.5360], device='cuda:1'), covar=tensor([0.1588, 0.1923, 0.2010, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0722, 0.0674, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 19:40:06,483 INFO [train.py:968] (1/2) Epoch 15, batch 17650, giga_loss[loss=0.326, simple_loss=0.3672, pruned_loss=0.1424, over 26619.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3302, pruned_loss=0.08998, over 5691250.80 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3287, pruned_loss=0.08691, over 5744547.86 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3315, pruned_loss=0.09081, over 5683772.73 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:40:42,595 INFO [optim.py:369] (1/2) 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 19:40:49,247 INFO [train.py:968] (1/2) Epoch 15, batch 17700, libri_loss[loss=0.3084, simple_loss=0.3787, pruned_loss=0.119, over 29529.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3237, pruned_loss=0.08736, over 5696014.29 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3295, pruned_loss=0.0873, over 5747922.15 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3239, pruned_loss=0.08769, over 5685998.87 frames. ], batch size: 81, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:41:29,271 INFO [train.py:968] (1/2) Epoch 15, batch 17750, giga_loss[loss=0.2166, simple_loss=0.2871, pruned_loss=0.07303, over 28982.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3186, pruned_loss=0.08509, over 5687883.81 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3299, pruned_loss=0.0874, over 5741887.58 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3182, pruned_loss=0.08523, over 5684280.64 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:42:01,867 INFO [optim.py:369] (1/2) 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 19:42:07,366 INFO [train.py:968] (1/2) Epoch 15, batch 17800, giga_loss[loss=0.3143, simple_loss=0.3625, pruned_loss=0.133, over 28638.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3164, pruned_loss=0.08406, over 5698789.23 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3306, pruned_loss=0.08752, over 5748019.76 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.315, pruned_loss=0.08394, over 5688533.50 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:42:25,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5665, 1.5828, 4.3664, 3.4431], device='cuda:1'), covar=tensor([0.1818, 0.2728, 0.0601, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0696, 0.0608, 0.0890, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 19:42:42,413 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 17850, giga_loss[loss=0.2308, simple_loss=0.302, pruned_loss=0.07976, over 28848.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3143, pruned_loss=0.08318, over 5700809.57 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.331, pruned_loss=0.08772, over 5754122.55 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3122, pruned_loss=0.08275, over 5685131.63 frames. ], batch size: 186, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:43:24,126 INFO [optim.py:369] (1/2) 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 19:43:31,323 INFO [train.py:968] (1/2) Epoch 15, batch 17900, giga_loss[loss=0.2534, simple_loss=0.3148, pruned_loss=0.09597, over 26572.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3115, pruned_loss=0.08177, over 5695656.57 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3312, pruned_loss=0.08774, over 5757520.19 frames. ], giga_tot_loss[loss=0.2357, simple_loss=0.309, pruned_loss=0.08121, over 5678245.06 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:43:44,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-07 19:44:12,350 INFO [train.py:968] (1/2) Epoch 15, batch 17950, giga_loss[loss=0.2309, simple_loss=0.2981, pruned_loss=0.0818, over 28853.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3079, pruned_loss=0.08019, over 5708503.40 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3314, pruned_loss=0.08777, over 5758756.07 frames. ], giga_tot_loss[loss=0.2323, simple_loss=0.3054, pruned_loss=0.07961, over 5692872.76 frames. ], batch size: 186, lr: 2.15e-03, grad_scale: 2.0
+2023-03-07 19:44:15,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-07 19:44:34,085 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=656891.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:44:46,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9483, 1.2938, 1.0703, 0.1929], device='cuda:1'), covar=tensor([0.3596, 0.3015, 0.4523, 0.5797], device='cuda:1'), in_proj_covar=tensor([0.1619, 0.1552, 0.1526, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 19:44:49,134 INFO [optim.py:369] (1/2) 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 19:44:54,716 INFO [train.py:968] (1/2) Epoch 15, batch 18000, giga_loss[loss=0.2341, simple_loss=0.3047, pruned_loss=0.0818, over 28635.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3056, pruned_loss=0.07891, over 5704573.40 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3314, pruned_loss=0.08773, over 5761286.80 frames. ], giga_tot_loss[loss=0.2295, simple_loss=0.3026, pruned_loss=0.07817, over 5687422.49 frames. ], batch size: 262, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:44:54,716 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 19:45:03,168 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 19:45:11,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5793, 3.5101, 1.6926, 1.6778], device='cuda:1'), covar=tensor([0.0908, 0.0335, 0.0832, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0520, 0.0356, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 19:45:36,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=656958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:45:45,539 INFO [train.py:968] (1/2) Epoch 15, batch 18050, giga_loss[loss=0.2384, simple_loss=0.2884, pruned_loss=0.09418, over 23799.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3038, pruned_loss=0.07818, over 5700508.73 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3321, pruned_loss=0.08783, over 5765037.80 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.2999, pruned_loss=0.0772, over 5681340.65 frames. ], batch size: 705, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:45:57,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6447, 1.8220, 1.8923, 1.4857], device='cuda:1'), covar=tensor([0.1725, 0.2365, 0.1373, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0687, 0.0910, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 19:46:22,151 INFO [optim.py:369] (1/2) 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 19:46:27,304 INFO [train.py:968] (1/2) Epoch 15, batch 18100, giga_loss[loss=0.2104, simple_loss=0.284, pruned_loss=0.06842, over 28995.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3019, pruned_loss=0.07723, over 5709577.53 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3322, pruned_loss=0.08793, over 5769769.57 frames. ], giga_tot_loss[loss=0.2248, simple_loss=0.2976, pruned_loss=0.07598, over 5687771.16 frames. ], batch size: 106, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:46:36,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0289, 2.4739, 2.2522, 1.8811], device='cuda:1'), covar=tensor([0.2810, 0.1788, 0.1777, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.1808, 0.1697, 0.1629, 0.1773], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 19:47:10,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-07 19:47:13,171 INFO [train.py:968] (1/2) Epoch 15, batch 18150, giga_loss[loss=0.1977, simple_loss=0.2799, pruned_loss=0.05776, over 29019.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.2985, pruned_loss=0.0758, over 5714456.73 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3324, pruned_loss=0.08805, over 5770608.06 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.294, pruned_loss=0.07433, over 5694494.24 frames. ], batch size: 155, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:47:49,388 INFO [optim.py:369] (1/2) 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 19:47:54,916 INFO [train.py:968] (1/2) Epoch 15, batch 18200, giga_loss[loss=0.234, simple_loss=0.3074, pruned_loss=0.08028, over 28568.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3012, pruned_loss=0.07757, over 5709912.09 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3328, pruned_loss=0.08821, over 5773632.70 frames. ], giga_tot_loss[loss=0.2238, simple_loss=0.296, pruned_loss=0.07584, over 5688850.76 frames. ], batch size: 85, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:48:11,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:48:45,894 INFO [train.py:968] (1/2) Epoch 15, batch 18250, giga_loss[loss=0.297, simple_loss=0.3708, pruned_loss=0.1116, over 28832.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3131, pruned_loss=0.08366, over 5710543.30 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3333, pruned_loss=0.08839, over 5772467.02 frames. ], giga_tot_loss[loss=0.2359, simple_loss=0.308, pruned_loss=0.08195, over 5693294.68 frames. ], batch size: 199, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:49:21,748 INFO [optim.py:369] (1/2) 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 19:49:26,214 INFO [train.py:968] (1/2) Epoch 15, batch 18300, giga_loss[loss=0.2767, simple_loss=0.3481, pruned_loss=0.1027, over 28564.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3265, pruned_loss=0.09089, over 5694065.35 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.334, pruned_loss=0.08871, over 5760398.66 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3209, pruned_loss=0.08913, over 5688539.78 frames. ], batch size: 78, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:49:27,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8060, 1.9535, 2.0067, 1.5787], device='cuda:1'), covar=tensor([0.1703, 0.2319, 0.1369, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0689, 0.0909, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 19:49:57,548 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:03,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657266.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:50:04,871 INFO [train.py:968] (1/2) Epoch 15, batch 18350, giga_loss[loss=0.3265, simple_loss=0.387, pruned_loss=0.1331, over 28807.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3358, pruned_loss=0.0951, over 5707887.43 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3338, pruned_loss=0.08841, over 5765283.09 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3314, pruned_loss=0.0941, over 5697138.51 frames. ], batch size: 112, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:50:05,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7771, 2.6231, 1.7593, 1.0177], device='cuda:1'), covar=tensor([0.6479, 0.2930, 0.3401, 0.5573], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1544, 0.1524, 0.1327], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 19:50:08,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-07 19:50:10,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:12,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:37,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:50:40,137 INFO [optim.py:369] (1/2) 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 19:50:46,686 INFO [train.py:968] (1/2) Epoch 15, batch 18400, giga_loss[loss=0.3086, simple_loss=0.3811, pruned_loss=0.118, over 29114.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3405, pruned_loss=0.0964, over 5702034.83 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3335, pruned_loss=0.08821, over 5767717.96 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3373, pruned_loss=0.09592, over 5690302.36 frames. ], batch size: 113, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:50:59,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:51:26,546 INFO [train.py:968] (1/2) Epoch 15, batch 18450, giga_loss[loss=0.2462, simple_loss=0.3407, pruned_loss=0.07584, over 28585.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3429, pruned_loss=0.09623, over 5702263.36 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3337, pruned_loss=0.08821, over 5769355.14 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3405, pruned_loss=0.09609, over 5689468.09 frames. ], batch size: 336, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:51:53,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657396.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:51:56,376 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:52:00,959 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657407.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:52:02,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657409.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:52:04,497 INFO [optim.py:369] (1/2) 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 19:52:05,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657412.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:52:11,712 INFO [train.py:968] (1/2) Epoch 15, batch 18500, giga_loss[loss=0.2862, simple_loss=0.3501, pruned_loss=0.1112, over 28751.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3446, pruned_loss=0.09672, over 5698636.48 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3337, pruned_loss=0.08815, over 5772836.67 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3429, pruned_loss=0.09685, over 5683532.83 frames. ], batch size: 92, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:52:28,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.13 vs. limit=5.0
+2023-03-07 19:52:34,238 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657441.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:52:55,392 INFO [train.py:968] (1/2) Epoch 15, batch 18550, giga_loss[loss=0.2512, simple_loss=0.3283, pruned_loss=0.08707, over 28473.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3461, pruned_loss=0.09781, over 5700282.85 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3337, pruned_loss=0.08809, over 5776385.30 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.345, pruned_loss=0.09825, over 5682843.47 frames. ], batch size: 71, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:53:01,221 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:53:02,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:53:28,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:53:31,622 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 18600, giga_loss[loss=0.3251, simple_loss=0.3887, pruned_loss=0.1308, over 28948.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3497, pruned_loss=0.1008, over 5707235.31 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3336, pruned_loss=0.08793, over 5779390.93 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3494, pruned_loss=0.1016, over 5688815.52 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:54:21,584 INFO [train.py:968] (1/2) Epoch 15, batch 18650, giga_loss[loss=0.2804, simple_loss=0.3501, pruned_loss=0.1053, over 28743.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3536, pruned_loss=0.1032, over 5705851.62 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3343, pruned_loss=0.08829, over 5781529.22 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.353, pruned_loss=0.1037, over 5688597.17 frames. ], batch size: 99, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:54:58,352 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 18700, giga_loss[loss=0.2709, simple_loss=0.3504, pruned_loss=0.09572, over 28706.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3562, pruned_loss=0.1037, over 5710432.63 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3345, pruned_loss=0.08832, over 5781244.62 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3557, pruned_loss=0.1041, over 5696471.66 frames. ], batch size: 92, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:55:16,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 18750, giga_loss[loss=0.3275, simple_loss=0.3964, pruned_loss=0.1293, over 28548.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3567, pruned_loss=0.1032, over 5710813.66 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3344, pruned_loss=0.08806, over 5782704.80 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.357, pruned_loss=0.1042, over 5696294.75 frames. ], batch size: 85, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:55:57,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 19:56:19,888 INFO [optim.py:369] (1/2) 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 19:56:25,299 INFO [train.py:968] (1/2) Epoch 15, batch 18800, giga_loss[loss=0.254, simple_loss=0.3399, pruned_loss=0.08402, over 29038.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3579, pruned_loss=0.1032, over 5713243.12 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3346, pruned_loss=0.08818, over 5786112.69 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3586, pruned_loss=0.1043, over 5697189.97 frames. ], batch size: 128, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:57:02,828 INFO [train.py:968] (1/2) Epoch 15, batch 18850, giga_loss[loss=0.2771, simple_loss=0.3559, pruned_loss=0.0991, over 28844.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3575, pruned_loss=0.102, over 5712662.30 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.335, pruned_loss=0.08838, over 5787916.72 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3582, pruned_loss=0.103, over 5696543.73 frames. ], batch size: 112, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:57:04,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657771.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:57:11,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:57:13,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:57:15,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:57:16,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=657782.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:57:38,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7075, 2.0289, 1.5016, 2.1228], device='cuda:1'), covar=tensor([0.2437, 0.2374, 0.2803, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1374, 0.1010, 0.1224, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 19:57:39,022 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:57:44,870 INFO [train.py:968] (1/2) Epoch 15, batch 18900, giga_loss[loss=0.269, simple_loss=0.3517, pruned_loss=0.09317, over 28952.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.356, pruned_loss=0.1002, over 5712996.80 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3356, pruned_loss=0.0887, over 5787419.70 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3563, pruned_loss=0.1009, over 5699538.00 frames. ], batch size: 213, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:57:51,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:58:24,457 INFO [train.py:968] (1/2) Epoch 15, batch 18950, giga_loss[loss=0.2463, simple_loss=0.3329, pruned_loss=0.07985, over 28244.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3548, pruned_loss=0.0993, over 5708750.43 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3361, pruned_loss=0.08898, over 5780722.81 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3551, pruned_loss=0.09987, over 5701157.06 frames. ], batch size: 77, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 19:58:27,357 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=657873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:58:38,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-07 19:58:45,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-07 19:59:00,381 INFO [optim.py:369] (1/2) 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 19:59:02,158 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657914.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:59:05,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657917.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:59:06,174 INFO [train.py:968] (1/2) Epoch 15, batch 19000, giga_loss[loss=0.2876, simple_loss=0.3579, pruned_loss=0.1087, over 28875.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3575, pruned_loss=0.1036, over 5694160.06 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3364, pruned_loss=0.08911, over 5779588.03 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.358, pruned_loss=0.1043, over 5687029.52 frames. ], batch size: 119, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 19:59:06,473 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:59:08,329 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=657922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:59:13,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=657925.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:59:15,360 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657946.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 19:59:35,390 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 19:59:40,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=657957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 19:59:48,817 INFO [train.py:968] (1/2) Epoch 15, batch 19050, giga_loss[loss=0.2874, simple_loss=0.35, pruned_loss=0.1124, over 28892.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3597, pruned_loss=0.1069, over 5693520.05 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3369, pruned_loss=0.08926, over 5784348.62 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3604, pruned_loss=0.1079, over 5680264.41 frames. ], batch size: 112, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:00:22,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5451, 1.7833, 1.4364, 1.5884], device='cuda:1'), covar=tensor([0.2578, 0.2578, 0.2946, 0.2456], device='cuda:1'), in_proj_covar=tensor([0.1368, 0.1005, 0.1222, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 20:00:22,522 INFO [optim.py:369] (1/2) 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 20:00:27,280 INFO [train.py:968] (1/2) Epoch 15, batch 19100, giga_loss[loss=0.2661, simple_loss=0.3431, pruned_loss=0.09459, over 28663.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3583, pruned_loss=0.1066, over 5701666.23 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3368, pruned_loss=0.08912, over 5784439.22 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3593, pruned_loss=0.1078, over 5689476.58 frames. ], batch size: 262, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:01:11,622 INFO [train.py:968] (1/2) Epoch 15, batch 19150, giga_loss[loss=0.2729, simple_loss=0.3418, pruned_loss=0.102, over 28735.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3576, pruned_loss=0.1072, over 5706255.85 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3373, pruned_loss=0.08919, over 5787718.13 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3586, pruned_loss=0.1085, over 5691552.40 frames. ], batch size: 119, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:01:49,099 INFO [optim.py:369] (1/2) 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 20:01:55,594 INFO [train.py:968] (1/2) Epoch 15, batch 19200, giga_loss[loss=0.2753, simple_loss=0.3492, pruned_loss=0.1007, over 28670.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3563, pruned_loss=0.1065, over 5699440.12 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3376, pruned_loss=0.0893, over 5788724.06 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3569, pruned_loss=0.1076, over 5686232.38 frames. ], batch size: 92, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:02:06,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2573, 1.6098, 1.5815, 1.1649], device='cuda:1'), covar=tensor([0.1549, 0.2261, 0.1271, 0.1542], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0687, 0.0904, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 20:02:34,816 INFO [train.py:968] (1/2) Epoch 15, batch 19250, giga_loss[loss=0.2728, simple_loss=0.3485, pruned_loss=0.09854, over 28952.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3543, pruned_loss=0.1047, over 5699334.63 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3372, pruned_loss=0.08913, over 5791742.45 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3557, pruned_loss=0.1062, over 5683290.12 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:02:55,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3782, 1.7915, 1.4474, 1.5988], device='cuda:1'), covar=tensor([0.0809, 0.0287, 0.0323, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 20:03:01,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:03:09,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2259, 1.6640, 1.2537, 0.8117], device='cuda:1'), covar=tensor([0.4464, 0.2265, 0.2534, 0.4516], device='cuda:1'), in_proj_covar=tensor([0.1602, 0.1536, 0.1516, 0.1321], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 20:03:12,057 INFO [optim.py:369] (1/2) 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 20:03:16,775 INFO [train.py:968] (1/2) Epoch 15, batch 19300, giga_loss[loss=0.3257, simple_loss=0.3712, pruned_loss=0.1401, over 26639.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3528, pruned_loss=0.1031, over 5699445.35 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3379, pruned_loss=0.08935, over 5791848.01 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3538, pruned_loss=0.1046, over 5683479.15 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:03:41,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:03:59,572 INFO [train.py:968] (1/2) Epoch 15, batch 19350, giga_loss[loss=0.2604, simple_loss=0.3322, pruned_loss=0.09432, over 28330.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3487, pruned_loss=0.1006, over 5692082.30 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3387, pruned_loss=0.08968, over 5789172.76 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3492, pruned_loss=0.1019, over 5679385.23 frames. ], batch size: 77, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:04:41,187 INFO [optim.py:369] (1/2) 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 20:04:44,386 INFO [train.py:968] (1/2) Epoch 15, batch 19400, libri_loss[loss=0.2354, simple_loss=0.3073, pruned_loss=0.08175, over 29485.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3434, pruned_loss=0.09802, over 5692927.59 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.339, pruned_loss=0.08987, over 5791860.40 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3436, pruned_loss=0.09905, over 5677862.58 frames. ], batch size: 70, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:05:08,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:12,339 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:31,845 INFO [train.py:968] (1/2) Epoch 15, batch 19450, giga_loss[loss=0.2169, simple_loss=0.3021, pruned_loss=0.06582, over 28887.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.338, pruned_loss=0.09522, over 5693930.12 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3397, pruned_loss=0.09041, over 5794011.65 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3375, pruned_loss=0.09565, over 5678170.07 frames. ], batch size: 199, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:05:36,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-07 20:05:40,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:54,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:05:56,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:06:12,663 INFO [optim.py:369] (1/2) 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 20:06:17,936 INFO [train.py:968] (1/2) Epoch 15, batch 19500, giga_loss[loss=0.3137, simple_loss=0.3641, pruned_loss=0.1316, over 26448.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3367, pruned_loss=0.09381, over 5693838.51 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.34, pruned_loss=0.09033, over 5793347.08 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3359, pruned_loss=0.09431, over 5679665.59 frames. ], batch size: 555, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:06:20,933 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=658423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:06:20,960 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:06:21,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=658424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:07:00,390 INFO [train.py:968] (1/2) Epoch 15, batch 19550, giga_loss[loss=0.2619, simple_loss=0.3266, pruned_loss=0.09854, over 28557.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3376, pruned_loss=0.09368, over 5705558.39 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3404, pruned_loss=0.0905, over 5793043.47 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3366, pruned_loss=0.09395, over 5693202.79 frames. ], batch size: 71, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:07:40,650 INFO [optim.py:369] (1/2) 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 20:07:43,977 INFO [train.py:968] (1/2) Epoch 15, batch 19600, giga_loss[loss=0.244, simple_loss=0.3269, pruned_loss=0.08055, over 28924.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3364, pruned_loss=0.09296, over 5708230.51 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3404, pruned_loss=0.09032, over 5795730.63 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3355, pruned_loss=0.0934, over 5693876.45 frames. ], batch size: 174, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:08:04,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7254, 1.7625, 1.9823, 1.5326], device='cuda:1'), covar=tensor([0.1836, 0.2227, 0.1394, 0.1605], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0690, 0.0906, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 20:08:13,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9336, 2.4056, 2.3084, 1.7339], device='cuda:1'), covar=tensor([0.2539, 0.1536, 0.1470, 0.1952], device='cuda:1'), in_proj_covar=tensor([0.1803, 0.1708, 0.1650, 0.1783], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 20:08:26,529 INFO [train.py:968] (1/2) Epoch 15, batch 19650, giga_loss[loss=0.2249, simple_loss=0.3022, pruned_loss=0.0738, over 28350.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3338, pruned_loss=0.09157, over 5714364.91 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3409, pruned_loss=0.09042, over 5792623.19 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3325, pruned_loss=0.09188, over 5704232.66 frames. ], batch size: 78, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:09:00,012 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 19700, giga_loss[loss=0.2665, simple_loss=0.3382, pruned_loss=0.09737, over 29069.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3322, pruned_loss=0.09061, over 5717754.99 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3415, pruned_loss=0.09047, over 5786969.88 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3303, pruned_loss=0.09084, over 5712545.11 frames. ], batch size: 155, lr: 2.15e-03, grad_scale: 8.0
+2023-03-07 20:09:07,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1821, 1.7309, 1.3477, 0.3836], device='cuda:1'), covar=tensor([0.3769, 0.2399, 0.3849, 0.5079], device='cuda:1'), in_proj_covar=tensor([0.1603, 0.1535, 0.1514, 0.1321], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 20:09:46,374 INFO [train.py:968] (1/2) Epoch 15, batch 19750, giga_loss[loss=0.233, simple_loss=0.3088, pruned_loss=0.07856, over 28921.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3305, pruned_loss=0.0903, over 5711082.05 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3422, pruned_loss=0.09075, over 5779074.64 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3282, pruned_loss=0.09022, over 5713174.36 frames. ], batch size: 227, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:10:25,552 INFO [optim.py:369] (1/2) 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 20:10:28,109 INFO [train.py:968] (1/2) Epoch 15, batch 19800, giga_loss[loss=0.2125, simple_loss=0.2912, pruned_loss=0.06687, over 28907.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3281, pruned_loss=0.08937, over 5716997.03 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3427, pruned_loss=0.09089, over 5780515.53 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3258, pruned_loss=0.08917, over 5716385.97 frames. ], batch size: 112, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:10:49,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-07 20:11:09,467 INFO [train.py:968] (1/2) Epoch 15, batch 19850, giga_loss[loss=0.2333, simple_loss=0.3063, pruned_loss=0.0802, over 28799.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3255, pruned_loss=0.08838, over 5715438.06 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.343, pruned_loss=0.09094, over 5780812.95 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3233, pruned_loss=0.08817, over 5714359.27 frames. ], batch size: 199, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:11:32,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:11:33,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=658799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:11:44,724 INFO [optim.py:369] (1/2) 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 20:11:48,204 INFO [train.py:968] (1/2) Epoch 15, batch 19900, giga_loss[loss=0.2516, simple_loss=0.3223, pruned_loss=0.09048, over 28863.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3251, pruned_loss=0.0881, over 5715938.71 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3443, pruned_loss=0.09156, over 5780821.31 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3215, pruned_loss=0.08726, over 5713085.63 frames. ], batch size: 119, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:12:21,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4242, 1.5794, 1.6195, 1.5055], device='cuda:1'), covar=tensor([0.1936, 0.2118, 0.2349, 0.2014], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0733, 0.0687, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 20:12:30,687 INFO [train.py:968] (1/2) Epoch 15, batch 19950, giga_loss[loss=0.2381, simple_loss=0.3153, pruned_loss=0.08042, over 28987.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3237, pruned_loss=0.08732, over 5696088.81 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3452, pruned_loss=0.09214, over 5752779.00 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3195, pruned_loss=0.08601, over 5718203.14 frames. ], batch size: 155, lr: 2.15e-03, grad_scale: 4.0
+2023-03-07 20:13:08,377 INFO [optim.py:369] (1/2) 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 20:13:11,722 INFO [train.py:968] (1/2) Epoch 15, batch 20000, giga_loss[loss=0.2144, simple_loss=0.297, pruned_loss=0.06591, over 28973.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3211, pruned_loss=0.08578, over 5706462.11 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3457, pruned_loss=0.0922, over 5754875.33 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3171, pruned_loss=0.08464, over 5721564.40 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:13:21,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1029, 4.9398, 4.6668, 2.0705], device='cuda:1'), covar=tensor([0.0381, 0.0490, 0.0469, 0.2117], device='cuda:1'), in_proj_covar=tensor([0.1103, 0.1013, 0.0877, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 20:13:21,352 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9033, 2.0151, 1.8884, 1.7742], device='cuda:1'), covar=tensor([0.1818, 0.2423, 0.2198, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0731, 0.0686, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 20:13:28,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:28,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=658942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:29,838 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:30,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=658945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:31,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2808, 3.0669, 1.4727, 1.4305], device='cuda:1'), covar=tensor([0.1033, 0.0330, 0.0865, 0.1359], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0520, 0.0355, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 20:13:48,624 INFO [train.py:968] (1/2) Epoch 15, batch 20050, giga_loss[loss=0.2105, simple_loss=0.2936, pruned_loss=0.06367, over 29040.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3205, pruned_loss=0.08529, over 5717726.86 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3463, pruned_loss=0.09229, over 5755087.32 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3162, pruned_loss=0.08414, over 5728455.75 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:13:51,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:13:52,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=658974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:14:24,392 INFO [optim.py:369] (1/2) 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 20:14:29,350 INFO [train.py:968] (1/2) Epoch 15, batch 20100, giga_loss[loss=0.2526, simple_loss=0.3257, pruned_loss=0.08971, over 28790.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3234, pruned_loss=0.08714, over 5711445.84 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3461, pruned_loss=0.09219, over 5748967.07 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3193, pruned_loss=0.0861, over 5724899.48 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:14:45,093 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:15:16,029 INFO [train.py:968] (1/2) Epoch 15, batch 20150, giga_loss[loss=0.2902, simple_loss=0.3701, pruned_loss=0.1051, over 29016.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.329, pruned_loss=0.09063, over 5700469.92 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.346, pruned_loss=0.09216, over 5742471.49 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3253, pruned_loss=0.08973, over 5715850.17 frames. ], batch size: 128, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:15:56,545 INFO [optim.py:369] (1/2) 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 20:15:59,473 INFO [train.py:968] (1/2) Epoch 15, batch 20200, giga_loss[loss=0.3736, simple_loss=0.4168, pruned_loss=0.1653, over 27650.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3393, pruned_loss=0.09731, over 5692143.38 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3466, pruned_loss=0.09217, over 5749351.11 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.335, pruned_loss=0.09665, over 5695997.85 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:16:36,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:16:43,789 INFO [train.py:968] (1/2) Epoch 15, batch 20250, giga_loss[loss=0.3741, simple_loss=0.4304, pruned_loss=0.1588, over 28627.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3456, pruned_loss=0.1011, over 5674356.35 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3469, pruned_loss=0.09231, over 5730994.58 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3419, pruned_loss=0.1005, over 5693035.18 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:17:28,101 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 20300, giga_loss[loss=0.2661, simple_loss=0.3533, pruned_loss=0.08945, over 28923.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3491, pruned_loss=0.1022, over 5665373.95 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3468, pruned_loss=0.09229, over 5732393.28 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3462, pruned_loss=0.102, over 5677450.04 frames. ], batch size: 213, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:17:54,947 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-07 20:18:16,450 INFO [train.py:968] (1/2) Epoch 15, batch 20350, giga_loss[loss=0.2893, simple_loss=0.3598, pruned_loss=0.1094, over 28781.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.354, pruned_loss=0.1046, over 5667299.69 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3472, pruned_loss=0.09263, over 5729985.78 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3515, pruned_loss=0.1043, over 5677795.40 frames. ], batch size: 99, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:18:55,324 INFO [optim.py:369] (1/2) 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 20:18:58,603 INFO [train.py:968] (1/2) Epoch 15, batch 20400, giga_loss[loss=0.2907, simple_loss=0.373, pruned_loss=0.1042, over 28801.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3597, pruned_loss=0.1085, over 5669878.04 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3471, pruned_loss=0.09254, over 5736063.10 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3581, pruned_loss=0.1088, over 5670679.47 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:19:41,870 INFO [train.py:968] (1/2) Epoch 15, batch 20450, libri_loss[loss=0.3202, simple_loss=0.3956, pruned_loss=0.1224, over 27864.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3559, pruned_loss=0.1056, over 5678357.87 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3475, pruned_loss=0.09287, over 5736434.43 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3544, pruned_loss=0.1058, over 5677194.25 frames. ], batch size: 116, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:19:50,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1536, 0.8031, 0.8814, 1.3401], device='cuda:1'), covar=tensor([0.0816, 0.0390, 0.0353, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 20:20:15,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1854, 1.3924, 1.3239, 1.0943], device='cuda:1'), covar=tensor([0.2274, 0.2098, 0.1243, 0.1896], device='cuda:1'), in_proj_covar=tensor([0.1819, 0.1722, 0.1672, 0.1796], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 20:20:21,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=659413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:20:24,295 INFO [optim.py:369] (1/2) 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 20:20:26,162 INFO [train.py:968] (1/2) Epoch 15, batch 20500, giga_loss[loss=0.2598, simple_loss=0.3412, pruned_loss=0.08915, over 28917.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3524, pruned_loss=0.1026, over 5684429.59 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.09288, over 5737292.26 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3513, pruned_loss=0.1029, over 5682561.31 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:21:10,315 INFO [train.py:968] (1/2) Epoch 15, batch 20550, giga_loss[loss=0.2535, simple_loss=0.3385, pruned_loss=0.08424, over 29027.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.351, pruned_loss=0.1014, over 5688785.90 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3474, pruned_loss=0.09288, over 5739509.39 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3503, pruned_loss=0.1017, over 5684711.29 frames. ], batch size: 155, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:21:52,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 20:21:53,189 INFO [optim.py:369] (1/2) 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 20:21:54,621 INFO [train.py:968] (1/2) Epoch 15, batch 20600, libri_loss[loss=0.2952, simple_loss=0.3808, pruned_loss=0.1048, over 29523.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3523, pruned_loss=0.1017, over 5682131.57 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.0931, over 5732985.50 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3514, pruned_loss=0.1019, over 5683481.42 frames. ], batch size: 83, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:21:59,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9251, 1.2739, 1.0671, 0.1788], device='cuda:1'), covar=tensor([0.3256, 0.2630, 0.4153, 0.5395], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1539, 0.1526, 0.1333], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 20:22:08,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=659535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:22:26,782 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=659556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:22:30,584 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=659559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:22:38,382 INFO [train.py:968] (1/2) Epoch 15, batch 20650, giga_loss[loss=0.2478, simple_loss=0.3305, pruned_loss=0.08256, over 28562.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3543, pruned_loss=0.1032, over 5691068.29 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3479, pruned_loss=0.0932, over 5737492.25 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3537, pruned_loss=0.1035, over 5686994.59 frames. ], batch size: 60, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:22:38,736 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-07 20:22:55,541 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=659588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:23:07,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6642, 1.7502, 1.9349, 1.4576], device='cuda:1'), covar=tensor([0.1661, 0.2358, 0.1327, 0.1572], device='cuda:1'), in_proj_covar=tensor([0.0861, 0.0689, 0.0906, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 20:23:17,652 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 20700, giga_loss[loss=0.2686, simple_loss=0.3457, pruned_loss=0.09575, over 28887.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3554, pruned_loss=0.1039, over 5702295.28 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3479, pruned_loss=0.09314, over 5738442.80 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3551, pruned_loss=0.1045, over 5697003.87 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:23:26,775 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:23:41,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-07 20:23:56,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7874, 2.6903, 1.7198, 0.7653], device='cuda:1'), covar=tensor([0.6141, 0.2612, 0.3352, 0.5830], device='cuda:1'), in_proj_covar=tensor([0.1606, 0.1538, 0.1520, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 20:24:07,106 INFO [train.py:968] (1/2) Epoch 15, batch 20750, giga_loss[loss=0.266, simple_loss=0.3435, pruned_loss=0.09422, over 28720.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.357, pruned_loss=0.1057, over 5684808.90 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3483, pruned_loss=0.09333, over 5739869.22 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3566, pruned_loss=0.1062, over 5678489.70 frames. ], batch size: 66, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:24:12,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-07 20:24:16,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=659678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:24:18,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=659681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:24:44,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=659710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:24:50,100 INFO [optim.py:369] (1/2) 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 20:24:51,546 INFO [train.py:968] (1/2) Epoch 15, batch 20800, giga_loss[loss=0.2734, simple_loss=0.3427, pruned_loss=0.102, over 28470.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3577, pruned_loss=0.1067, over 5690368.33 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09316, over 5742404.15 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3578, pruned_loss=0.1075, over 5682248.41 frames. ], batch size: 71, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:25:13,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3400, 3.0378, 1.5224, 1.4303], device='cuda:1'), covar=tensor([0.0899, 0.0309, 0.0796, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0522, 0.0355, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 20:25:26,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-07 20:25:29,210 INFO [train.py:968] (1/2) Epoch 15, batch 20850, libri_loss[loss=0.2411, simple_loss=0.3324, pruned_loss=0.0749, over 29535.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3571, pruned_loss=0.1058, over 5703342.02 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3481, pruned_loss=0.09318, over 5748320.58 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3575, pruned_loss=0.1069, over 5689633.14 frames. ], batch size: 80, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:26:07,323 INFO [optim.py:369] (1/2) 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 20:26:08,671 INFO [train.py:968] (1/2) Epoch 15, batch 20900, giga_loss[loss=0.2725, simple_loss=0.3463, pruned_loss=0.09937, over 27634.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.357, pruned_loss=0.1052, over 5704942.04 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3483, pruned_loss=0.09334, over 5751841.79 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3573, pruned_loss=0.1063, over 5689916.96 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:26:24,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659839.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:26:38,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3172, 1.7651, 0.8980, 1.3583], device='cuda:1'), covar=tensor([0.1241, 0.0704, 0.1659, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0434, 0.0505, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 20:26:47,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5985, 1.7584, 1.8490, 1.4097], device='cuda:1'), covar=tensor([0.1746, 0.2385, 0.1396, 0.1668], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0691, 0.0907, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 20:26:48,256 INFO [train.py:968] (1/2) Epoch 15, batch 20950, giga_loss[loss=0.272, simple_loss=0.3542, pruned_loss=0.09488, over 28877.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3569, pruned_loss=0.1036, over 5701004.48 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3486, pruned_loss=0.09351, over 5745932.06 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3572, pruned_loss=0.1045, over 5692970.61 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:27:21,998 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=659910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:27:27,177 INFO [optim.py:369] (1/2) 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 20:27:27,917 INFO [train.py:968] (1/2) Epoch 15, batch 21000, giga_loss[loss=0.2646, simple_loss=0.339, pruned_loss=0.09508, over 28662.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3562, pruned_loss=0.1031, over 5701494.51 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09356, over 5747631.79 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3566, pruned_loss=0.1041, over 5692115.52 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:27:27,917 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 20:27:36,926 INFO [train.py:1012] (1/2) Epoch 15, validation: loss=0.2119, simple_loss=0.3186, pruned_loss=0.0526, over 944034.00 frames. 
+2023-03-07 20:27:36,926 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 20:28:16,350 INFO [train.py:968] (1/2) Epoch 15, batch 21050, giga_loss[loss=0.257, simple_loss=0.3294, pruned_loss=0.09229, over 28905.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3532, pruned_loss=0.1015, over 5710341.30 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09345, over 5748104.78 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1025, over 5702083.79 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:28:43,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:28:45,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 20:28:54,405 INFO [optim.py:369] (1/2) 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 20:28:55,114 INFO [train.py:968] (1/2) Epoch 15, batch 21100, giga_loss[loss=0.2428, simple_loss=0.3198, pruned_loss=0.08285, over 28435.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3518, pruned_loss=0.101, over 5713893.23 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3483, pruned_loss=0.09351, over 5750350.14 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3523, pruned_loss=0.1018, over 5704895.63 frames. ], batch size: 65, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:28:55,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:29:19,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3666, 1.7275, 1.3924, 1.2168], device='cuda:1'), covar=tensor([0.2511, 0.2444, 0.2654, 0.2308], device='cuda:1'), in_proj_covar=tensor([0.1375, 0.1012, 0.1222, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 20:29:34,919 INFO [train.py:968] (1/2) Epoch 15, batch 21150, giga_loss[loss=0.2743, simple_loss=0.347, pruned_loss=0.1008, over 28692.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1002, over 5716357.59 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3484, pruned_loss=0.0936, over 5754369.55 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3503, pruned_loss=0.1009, over 5704459.29 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:29:50,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4615, 3.5373, 1.7042, 1.5389], device='cuda:1'), covar=tensor([0.0990, 0.0275, 0.0826, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0521, 0.0354, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 20:30:17,433 INFO [optim.py:369] (1/2) 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 20:30:17,934 INFO [train.py:968] (1/2) Epoch 15, batch 21200, giga_loss[loss=0.3196, simple_loss=0.3833, pruned_loss=0.1279, over 28915.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3515, pruned_loss=0.1019, over 5713760.71 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09361, over 5754599.28 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3517, pruned_loss=0.1026, over 5703056.82 frames. ], batch size: 199, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:30:39,183 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:30:39,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:30:41,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:30:59,741 INFO [train.py:968] (1/2) Epoch 15, batch 21250, giga_loss[loss=0.3042, simple_loss=0.3758, pruned_loss=0.1163, over 28675.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3523, pruned_loss=0.1021, over 5716970.27 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3485, pruned_loss=0.09377, over 5756209.46 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3525, pruned_loss=0.1027, over 5706314.51 frames. ], batch size: 242, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:31:04,371 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:31:04,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 20:31:32,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4819, 1.8829, 1.4475, 1.6639], device='cuda:1'), covar=tensor([0.2606, 0.2477, 0.2852, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1372, 0.1008, 0.1220, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 20:31:35,733 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 15, batch 21300, giga_loss[loss=0.2698, simple_loss=0.3515, pruned_loss=0.09398, over 28838.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3513, pruned_loss=0.1012, over 5706231.91 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3488, pruned_loss=0.09418, over 5750975.87 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3514, pruned_loss=0.1016, over 5701704.90 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:31:41,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6630, 1.7993, 1.9206, 1.4786], device='cuda:1'), covar=tensor([0.1940, 0.2552, 0.1531, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0694, 0.0908, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 20:32:19,205 INFO [train.py:968] (1/2) Epoch 15, batch 21350, giga_loss[loss=0.2402, simple_loss=0.3172, pruned_loss=0.0816, over 28919.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3505, pruned_loss=0.1002, over 5707874.20 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3491, pruned_loss=0.09448, over 5740904.65 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3503, pruned_loss=0.1003, over 5711625.80 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:32:31,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:33:01,150 INFO [optim.py:369] (1/2) 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 20:33:01,162 INFO [train.py:968] (1/2) Epoch 15, batch 21400, giga_loss[loss=0.3091, simple_loss=0.3704, pruned_loss=0.1239, over 28812.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3493, pruned_loss=0.09934, over 5716849.81 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3489, pruned_loss=0.09442, over 5742563.55 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3493, pruned_loss=0.09958, over 5718080.08 frames. ], batch size: 199, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:33:24,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2001, 1.3106, 1.1303, 0.9605], device='cuda:1'), covar=tensor([0.0895, 0.0510, 0.1051, 0.0990], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0434, 0.0505, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 20:33:29,870 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660357.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:33:31,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660360.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:33:38,130 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:33:38,545 INFO [train.py:968] (1/2) Epoch 15, batch 21450, giga_loss[loss=0.2659, simple_loss=0.3426, pruned_loss=0.09467, over 28987.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3473, pruned_loss=0.09848, over 5723598.03 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3496, pruned_loss=0.09486, over 5746686.64 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3467, pruned_loss=0.09837, over 5720367.87 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:33:55,894 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660389.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:33:59,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:34:18,699 INFO [optim.py:369] (1/2) 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 20:34:18,712 INFO [train.py:968] (1/2) Epoch 15, batch 21500, libri_loss[loss=0.3133, simple_loss=0.3846, pruned_loss=0.121, over 25894.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3447, pruned_loss=0.09752, over 5716432.68 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3491, pruned_loss=0.09483, over 5746087.01 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3445, pruned_loss=0.09752, over 5713925.37 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:34:25,948 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:34:51,530 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:34:54,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1166, 1.3237, 1.0146, 0.9690], device='cuda:1'), covar=tensor([0.0936, 0.0461, 0.1191, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0433, 0.0504, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 20:34:58,163 INFO [train.py:968] (1/2) Epoch 15, batch 21550, libri_loss[loss=0.2749, simple_loss=0.3485, pruned_loss=0.1007, over 29572.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3455, pruned_loss=0.09829, over 5722480.47 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3496, pruned_loss=0.09521, over 5747430.02 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3448, pruned_loss=0.09805, over 5718271.00 frames. ], batch size: 75, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:35:01,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:35:37,943 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660518.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:35:38,303 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 21600, libri_loss[loss=0.3215, simple_loss=0.3884, pruned_loss=0.1273, over 29271.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3456, pruned_loss=0.09927, over 5722450.68 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3498, pruned_loss=0.09556, over 5751059.49 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3448, pruned_loss=0.0988, over 5715293.55 frames. ], batch size: 94, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:35:53,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:35:56,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:36:07,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0005, 2.8889, 1.9182, 1.0276], device='cuda:1'), covar=tensor([0.6524, 0.2690, 0.3644, 0.6017], device='cuda:1'), in_proj_covar=tensor([0.1600, 0.1524, 0.1512, 0.1321], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 20:36:19,792 INFO [train.py:968] (1/2) Epoch 15, batch 21650, giga_loss[loss=0.2551, simple_loss=0.3318, pruned_loss=0.08919, over 28809.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3438, pruned_loss=0.09899, over 5721585.23 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3497, pruned_loss=0.0956, over 5752711.04 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3432, pruned_loss=0.09863, over 5714147.85 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:36:19,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:36:22,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-07 20:36:45,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4817, 3.5155, 1.6797, 1.6258], device='cuda:1'), covar=tensor([0.0881, 0.0367, 0.0842, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0519, 0.0353, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 20:36:58,597 INFO [train.py:968] (1/2) Epoch 15, batch 21700, giga_loss[loss=0.2876, simple_loss=0.3514, pruned_loss=0.1119, over 27648.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3424, pruned_loss=0.09841, over 5724018.31 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3507, pruned_loss=0.09632, over 5756101.61 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3407, pruned_loss=0.0975, over 5713758.58 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:36:59,161 INFO [optim.py:369] (1/2) 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 20:37:33,051 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660661.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:37:34,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660664.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:37:39,028 INFO [train.py:968] (1/2) Epoch 15, batch 21750, giga_loss[loss=0.2569, simple_loss=0.3237, pruned_loss=0.09506, over 28773.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3398, pruned_loss=0.09761, over 5720946.77 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3508, pruned_loss=0.09642, over 5757655.73 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3383, pruned_loss=0.09683, over 5711212.41 frames. ], batch size: 99, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:37:56,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660693.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:38:15,998 INFO [train.py:968] (1/2) Epoch 15, batch 21800, giga_loss[loss=0.303, simple_loss=0.3706, pruned_loss=0.1177, over 28763.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3382, pruned_loss=0.09677, over 5719436.76 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3512, pruned_loss=0.09697, over 5761147.74 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3363, pruned_loss=0.09565, over 5707640.17 frames. ], batch size: 284, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:38:16,539 INFO [optim.py:369] (1/2) 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 20:38:38,804 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 21850, giga_loss[loss=0.2853, simple_loss=0.3606, pruned_loss=0.105, over 28869.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3386, pruned_loss=0.09685, over 5705386.22 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3516, pruned_loss=0.09727, over 5752889.92 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3367, pruned_loss=0.09569, over 5703586.28 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 20:39:20,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3500, 1.2800, 3.8422, 3.1671], device='cuda:1'), covar=tensor([0.1600, 0.2642, 0.0407, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0696, 0.0606, 0.0882, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 20:39:43,072 INFO [train.py:968] (1/2) Epoch 15, batch 21900, giga_loss[loss=0.2496, simple_loss=0.3397, pruned_loss=0.07977, over 29018.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3416, pruned_loss=0.09782, over 5707303.94 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3518, pruned_loss=0.0976, over 5755348.81 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3397, pruned_loss=0.09663, over 5703109.88 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 20:39:45,624 INFO [optim.py:369] (1/2) 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 20:40:11,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=660849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:40:14,890 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:40:22,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 20:40:25,939 INFO [train.py:968] (1/2) Epoch 15, batch 21950, libri_loss[loss=0.3599, simple_loss=0.4198, pruned_loss=0.15, over 27909.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3435, pruned_loss=0.09813, over 5710402.63 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3527, pruned_loss=0.09836, over 5756253.00 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.341, pruned_loss=0.09642, over 5705219.94 frames. ], batch size: 116, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 20:40:38,636 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=660889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:40:46,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7296, 1.9128, 1.5003, 2.2445], device='cuda:1'), covar=tensor([0.2269, 0.2415, 0.2693, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.1376, 0.1012, 0.1223, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 20:41:05,626 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=660918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:41:05,984 INFO [train.py:968] (1/2) Epoch 15, batch 22000, giga_loss[loss=0.255, simple_loss=0.3348, pruned_loss=0.08759, over 29037.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3453, pruned_loss=0.09866, over 5705613.85 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3523, pruned_loss=0.09851, over 5760094.57 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3433, pruned_loss=0.09713, over 5696861.73 frames. ], batch size: 155, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:41:08,626 INFO [optim.py:369] (1/2) 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 20:41:19,574 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=660934.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:41:51,145 INFO [train.py:968] (1/2) Epoch 15, batch 22050, libri_loss[loss=0.3305, simple_loss=0.3937, pruned_loss=0.1337, over 29252.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3439, pruned_loss=0.09714, over 5705282.63 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.353, pruned_loss=0.09909, over 5762165.59 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3415, pruned_loss=0.09535, over 5695606.86 frames. ], batch size: 97, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:42:09,943 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=660992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:42:11,878 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 22100, giga_loss[loss=0.258, simple_loss=0.329, pruned_loss=0.09351, over 28892.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3439, pruned_loss=0.09733, over 5707157.61 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3533, pruned_loss=0.09944, over 5761207.58 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3416, pruned_loss=0.09555, over 5698849.95 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:42:33,381 INFO [optim.py:369] (1/2) 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 20:42:36,467 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:43:02,086 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 22150, giga_loss[loss=0.2557, simple_loss=0.3321, pruned_loss=0.08962, over 28459.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3452, pruned_loss=0.09838, over 5712329.69 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3534, pruned_loss=0.09977, over 5767220.93 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3429, pruned_loss=0.09658, over 5698277.85 frames. ], batch size: 78, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:43:53,598 INFO [train.py:968] (1/2) Epoch 15, batch 22200, giga_loss[loss=0.3336, simple_loss=0.3946, pruned_loss=0.1363, over 27616.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3468, pruned_loss=0.09954, over 5707767.51 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3539, pruned_loss=0.1002, over 5760161.44 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3444, pruned_loss=0.09772, over 5701413.73 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:43:57,863 INFO [optim.py:369] (1/2) 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 20:44:15,641 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:44:36,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-07 20:44:37,558 INFO [train.py:968] (1/2) Epoch 15, batch 22250, giga_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08934, over 28685.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3492, pruned_loss=0.1011, over 5701124.76 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3541, pruned_loss=0.1004, over 5760813.51 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3471, pruned_loss=0.09948, over 5694939.34 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:44:46,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7211, 1.8238, 1.7302, 1.5493], device='cuda:1'), covar=tensor([0.1477, 0.1867, 0.1938, 0.1868], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0732, 0.0687, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 20:45:18,496 INFO [train.py:968] (1/2) Epoch 15, batch 22300, giga_loss[loss=0.2841, simple_loss=0.3623, pruned_loss=0.103, over 28872.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1015, over 5708655.22 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.354, pruned_loss=0.1004, over 5761684.44 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3493, pruned_loss=0.1003, over 5702731.02 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:45:19,861 INFO [optim.py:369] (1/2) 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 20:45:23,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:45:27,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:45:57,278 INFO [train.py:968] (1/2) Epoch 15, batch 22350, giga_loss[loss=0.2429, simple_loss=0.3259, pruned_loss=0.07993, over 28844.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3519, pruned_loss=0.1019, over 5711488.97 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3542, pruned_loss=0.1006, over 5761052.12 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3504, pruned_loss=0.1008, over 5706489.05 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:46:32,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661309.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 20:46:41,076 INFO [train.py:968] (1/2) Epoch 15, batch 22400, giga_loss[loss=0.257, simple_loss=0.3255, pruned_loss=0.09419, over 28575.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3519, pruned_loss=0.1016, over 5714096.13 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3542, pruned_loss=0.1006, over 5761782.86 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3507, pruned_loss=0.1007, over 5709389.49 frames. ], batch size: 78, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:46:42,303 INFO [optim.py:369] (1/2) 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 20:46:53,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3659, 1.3403, 1.0589, 1.4689], device='cuda:1'), covar=tensor([0.0714, 0.0339, 0.0360, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0088, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 20:47:19,627 INFO [train.py:968] (1/2) Epoch 15, batch 22450, giga_loss[loss=0.3134, simple_loss=0.375, pruned_loss=0.1259, over 27914.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3533, pruned_loss=0.1029, over 5713687.49 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3549, pruned_loss=0.1011, over 5763871.87 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3517, pruned_loss=0.1017, over 5706856.71 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:47:22,794 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:47:24,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:47:39,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5495, 1.4903, 1.3593, 1.2275], device='cuda:1'), covar=tensor([0.0720, 0.0479, 0.0879, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0367, 0.0434, 0.0502, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 20:47:47,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:47:58,272 INFO [train.py:968] (1/2) Epoch 15, batch 22500, libri_loss[loss=0.3477, simple_loss=0.4002, pruned_loss=0.1476, over 19650.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3519, pruned_loss=0.1023, over 5710809.35 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3554, pruned_loss=0.1016, over 5758696.25 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3501, pruned_loss=0.1009, over 5709235.79 frames. ], batch size: 187, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:48:00,135 INFO [optim.py:369] (1/2) 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 20:48:10,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:48:23,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:48:27,894 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661452.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:48:29,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:48:29,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661455.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:48:40,543 INFO [train.py:968] (1/2) Epoch 15, batch 22550, libri_loss[loss=0.2773, simple_loss=0.3501, pruned_loss=0.1023, over 29572.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3502, pruned_loss=0.1019, over 5704154.06 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3562, pruned_loss=0.1024, over 5758172.41 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3479, pruned_loss=0.1, over 5701907.67 frames. ], batch size: 76, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:48:52,240 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661484.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 20:48:54,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-07 20:49:05,128 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:49:21,667 INFO [train.py:968] (1/2) Epoch 15, batch 22600, giga_loss[loss=0.2387, simple_loss=0.318, pruned_loss=0.07972, over 28992.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3472, pruned_loss=0.1003, over 5706395.52 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3566, pruned_loss=0.1028, over 5758242.33 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3448, pruned_loss=0.09846, over 5703905.62 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:49:23,643 INFO [optim.py:369] (1/2) 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 20:49:44,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:02,445 INFO [train.py:968] (1/2) Epoch 15, batch 22650, giga_loss[loss=0.272, simple_loss=0.3517, pruned_loss=0.09617, over 28737.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3476, pruned_loss=0.09973, over 5702702.24 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3571, pruned_loss=0.1033, over 5758094.04 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3451, pruned_loss=0.09767, over 5699791.28 frames. ], batch size: 119, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:50:06,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:11,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:26,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:32,354 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:50:40,632 INFO [train.py:968] (1/2) Epoch 15, batch 22700, giga_loss[loss=0.3, simple_loss=0.3794, pruned_loss=0.1103, over 28574.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3503, pruned_loss=0.1001, over 5702251.89 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3576, pruned_loss=0.104, over 5754045.98 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09762, over 5701769.24 frames. ], batch size: 336, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:50:44,218 INFO [optim.py:369] (1/2) 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 20:51:15,420 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:51:18,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:51:19,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-07 20:51:21,209 INFO [train.py:968] (1/2) Epoch 15, batch 22750, giga_loss[loss=0.2933, simple_loss=0.3581, pruned_loss=0.1142, over 28853.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09979, over 5694686.19 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3577, pruned_loss=0.1043, over 5752234.77 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.347, pruned_loss=0.09756, over 5694933.72 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:51:42,504 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:51:52,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7184, 4.5106, 4.3577, 2.0057], device='cuda:1'), covar=tensor([0.0627, 0.0832, 0.0891, 0.1903], device='cuda:1'), in_proj_covar=tensor([0.1109, 0.1019, 0.0881, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 20:52:02,615 INFO [train.py:968] (1/2) Epoch 15, batch 22800, giga_loss[loss=0.3166, simple_loss=0.3697, pruned_loss=0.1317, over 26727.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3481, pruned_loss=0.1006, over 5698870.63 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3578, pruned_loss=0.1046, over 5755704.03 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3458, pruned_loss=0.09841, over 5694628.79 frames. ], batch size: 555, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:52:05,074 INFO [optim.py:369] (1/2) 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 20:52:22,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:52:24,658 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 22850, giga_loss[loss=0.2336, simple_loss=0.3087, pruned_loss=0.07931, over 28919.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3461, pruned_loss=0.1007, over 5703705.36 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.358, pruned_loss=0.1048, over 5754023.12 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3438, pruned_loss=0.09865, over 5700979.64 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:52:47,824 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:53:14,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:53:22,911 INFO [train.py:968] (1/2) Epoch 15, batch 22900, libri_loss[loss=0.2709, simple_loss=0.3414, pruned_loss=0.1002, over 29565.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3443, pruned_loss=0.1006, over 5714522.17 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3579, pruned_loss=0.1049, over 5758283.70 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3423, pruned_loss=0.09877, over 5707059.06 frames. ], batch size: 76, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:53:23,875 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:53:31,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:53:59,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7066, 1.8375, 1.6764, 1.4828], device='cuda:1'), covar=tensor([0.2787, 0.2260, 0.1977, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1765, 0.1694, 0.1820], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 20:54:03,922 INFO [train.py:968] (1/2) Epoch 15, batch 22950, giga_loss[loss=0.2555, simple_loss=0.3229, pruned_loss=0.09402, over 28883.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3438, pruned_loss=0.101, over 5713243.42 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3588, pruned_loss=0.1058, over 5758690.89 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3409, pruned_loss=0.09856, over 5704939.92 frames. ], batch size: 106, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:54:06,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7451, 4.7367, 1.8275, 2.0823], device='cuda:1'), covar=tensor([0.0915, 0.0385, 0.0874, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0524, 0.0355, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 20:54:06,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:54:07,875 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=661875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:54:10,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6784, 1.7765, 1.3272, 1.5380], device='cuda:1'), covar=tensor([0.0830, 0.0604, 0.1030, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0369, 0.0434, 0.0502, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 20:54:42,507 INFO [train.py:968] (1/2) Epoch 15, batch 23000, libri_loss[loss=0.3084, simple_loss=0.3662, pruned_loss=0.1253, over 29551.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3428, pruned_loss=0.1003, over 5720862.60 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3594, pruned_loss=0.1065, over 5763551.79 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3393, pruned_loss=0.0976, over 5708579.64 frames. ], batch size: 76, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:54:44,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=661922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:54:45,078 INFO [optim.py:369] (1/2) 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 20:55:15,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:18,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:19,992 INFO [train.py:968] (1/2) Epoch 15, batch 23050, giga_loss[loss=0.2028, simple_loss=0.2881, pruned_loss=0.05871, over 28907.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3386, pruned_loss=0.09828, over 5714430.67 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3588, pruned_loss=0.1063, over 5756815.43 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3359, pruned_loss=0.09609, over 5709016.62 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:55:22,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=661972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:24,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=661975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:41,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=661995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:46,141 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-07 20:55:50,316 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:52,356 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:55:52,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4481, 1.6341, 1.3196, 1.4965], device='cuda:1'), covar=tensor([0.2674, 0.2663, 0.3149, 0.2344], device='cuda:1'), in_proj_covar=tensor([0.1376, 0.1009, 0.1219, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 20:55:59,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:01,526 INFO [train.py:968] (1/2) Epoch 15, batch 23100, giga_loss[loss=0.266, simple_loss=0.3441, pruned_loss=0.09389, over 29060.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3352, pruned_loss=0.09666, over 5707079.37 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3591, pruned_loss=0.1067, over 5755482.08 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3326, pruned_loss=0.09447, over 5703749.27 frames. ], batch size: 164, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:56:01,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:04,440 INFO [optim.py:369] (1/2) 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 20:56:11,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3537, 4.1689, 3.9485, 1.8311], device='cuda:1'), covar=tensor([0.0553, 0.0709, 0.0650, 0.2239], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.1026, 0.0887, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 20:56:24,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:37,529 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:39,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:56:39,850 INFO [train.py:968] (1/2) Epoch 15, batch 23150, giga_loss[loss=0.2336, simple_loss=0.3145, pruned_loss=0.07633, over 29084.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3349, pruned_loss=0.09603, over 5712548.14 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3592, pruned_loss=0.1068, over 5754445.99 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3324, pruned_loss=0.09402, over 5710253.80 frames. ], batch size: 128, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:57:01,472 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:57:07,860 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-07 20:57:17,344 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 23200, giga_loss[loss=0.3189, simple_loss=0.3875, pruned_loss=0.1252, over 28868.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3381, pruned_loss=0.09742, over 5718000.40 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3592, pruned_loss=0.1071, over 5759150.89 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3353, pruned_loss=0.09519, over 5710408.59 frames. ], batch size: 66, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 20:57:19,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-07 20:57:22,668 INFO [optim.py:369] (1/2) 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 20:57:41,436 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:57:41,667 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-07 20:57:58,389 INFO [train.py:968] (1/2) Epoch 15, batch 23250, giga_loss[loss=0.2399, simple_loss=0.3254, pruned_loss=0.07725, over 28870.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3426, pruned_loss=0.09985, over 5719166.69 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3595, pruned_loss=0.1075, over 5764189.10 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3392, pruned_loss=0.09729, over 5706626.58 frames. ], batch size: 174, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:58:07,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:58:37,820 INFO [train.py:968] (1/2) Epoch 15, batch 23300, giga_loss[loss=0.2675, simple_loss=0.3423, pruned_loss=0.09635, over 28778.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3456, pruned_loss=0.1006, over 5719211.48 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3597, pruned_loss=0.1077, over 5764612.36 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3424, pruned_loss=0.09828, over 5707988.88 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 20:58:41,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4290, 1.5753, 1.6301, 1.4680], device='cuda:1'), covar=tensor([0.1427, 0.1667, 0.1560, 0.1631], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0736, 0.0689, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 20:58:41,877 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:58:42,292 INFO [optim.py:369] (1/2) 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 20:59:02,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 20:59:18,748 INFO [train.py:968] (1/2) Epoch 15, batch 23350, giga_loss[loss=0.2812, simple_loss=0.3536, pruned_loss=0.1044, over 28912.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3489, pruned_loss=0.1026, over 5692838.66 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3602, pruned_loss=0.1086, over 5744873.95 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3454, pruned_loss=0.0996, over 5699014.09 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:00:01,552 INFO [train.py:968] (1/2) Epoch 15, batch 23400, giga_loss[loss=0.2681, simple_loss=0.3462, pruned_loss=0.09497, over 28941.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3506, pruned_loss=0.1032, over 5694122.74 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3601, pruned_loss=0.1087, over 5748219.63 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3477, pruned_loss=0.1006, over 5694735.37 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:00:04,863 INFO [optim.py:369] (1/2) 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 21:00:06,208 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:09,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:36,760 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:40,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3655, 1.5942, 1.6300, 1.2115], device='cuda:1'), covar=tensor([0.1542, 0.2311, 0.1314, 0.1530], device='cuda:1'), in_proj_covar=tensor([0.0862, 0.0693, 0.0908, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 21:00:46,796 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:00:46,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9267, 1.3367, 1.1488, 0.2297], device='cuda:1'), covar=tensor([0.3274, 0.2643, 0.3786, 0.5222], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1534, 0.1529, 0.1335], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 21:00:47,897 INFO [train.py:968] (1/2) Epoch 15, batch 23450, giga_loss[loss=0.3176, simple_loss=0.3791, pruned_loss=0.128, over 28891.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3562, pruned_loss=0.1085, over 5685238.24 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3604, pruned_loss=0.1091, over 5741875.06 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3534, pruned_loss=0.106, over 5690183.86 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:01:00,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:01:09,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:01:12,378 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:01:14,639 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 23500, giga_loss[loss=0.3292, simple_loss=0.3932, pruned_loss=0.1326, over 29048.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3617, pruned_loss=0.1129, over 5683279.02 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3608, pruned_loss=0.1095, over 5744352.41 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3591, pruned_loss=0.1105, over 5683844.86 frames. ], batch size: 128, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:01:42,187 INFO [optim.py:369] (1/2) 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 21:01:43,059 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:02:03,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-07 21:02:12,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3773, 3.0926, 1.4390, 1.4870], device='cuda:1'), covar=tensor([0.0975, 0.0384, 0.0882, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0528, 0.0358, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:02:26,117 INFO [train.py:968] (1/2) Epoch 15, batch 23550, giga_loss[loss=0.3092, simple_loss=0.3792, pruned_loss=0.1196, over 28857.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3683, pruned_loss=0.1175, over 5681942.93 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3609, pruned_loss=0.1096, over 5746089.64 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3662, pruned_loss=0.1157, over 5679289.44 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:02:34,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:02:39,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:02:48,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:02:55,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-07 21:03:13,034 INFO [train.py:968] (1/2) Epoch 15, batch 23600, giga_loss[loss=0.3432, simple_loss=0.3997, pruned_loss=0.1434, over 28352.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3738, pruned_loss=0.1222, over 5679665.48 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3609, pruned_loss=0.1096, over 5745516.01 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3723, pruned_loss=0.1209, over 5677156.08 frames. ], batch size: 368, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:03:17,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:03:19,099 INFO [optim.py:369] (1/2) 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 21:03:20,332 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:03:23,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:03:28,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4802, 1.6654, 1.4296, 1.3637], device='cuda:1'), covar=tensor([0.1931, 0.1816, 0.1859, 0.1700], device='cuda:1'), in_proj_covar=tensor([0.1375, 0.1010, 0.1219, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 21:03:53,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:04:02,404 INFO [train.py:968] (1/2) Epoch 15, batch 23650, giga_loss[loss=0.3392, simple_loss=0.3992, pruned_loss=0.1397, over 28686.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3812, pruned_loss=0.1291, over 5666803.23 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3611, pruned_loss=0.1098, over 5745586.29 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3802, pruned_loss=0.1282, over 5663105.57 frames. ], batch size: 242, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:04:33,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:04:53,722 INFO [train.py:968] (1/2) Epoch 15, batch 23700, giga_loss[loss=0.2681, simple_loss=0.3396, pruned_loss=0.09826, over 28489.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.3847, pruned_loss=0.1318, over 5658836.20 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3609, pruned_loss=0.1098, over 5737984.96 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3843, pruned_loss=0.1313, over 5662624.07 frames. ], batch size: 78, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:04:58,447 INFO [optim.py:369] (1/2) 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 21:05:08,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:11,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:37,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:39,004 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:05:39,984 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 23750, giga_loss[loss=0.2985, simple_loss=0.3649, pruned_loss=0.116, over 28943.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3867, pruned_loss=0.1344, over 5658177.81 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3605, pruned_loss=0.1097, over 5739390.24 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3871, pruned_loss=0.1344, over 5658433.92 frames. ], batch size: 136, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:06:10,801 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 23800, giga_loss[loss=0.3129, simple_loss=0.3696, pruned_loss=0.128, over 28500.00 frames. ], tot_loss[loss=0.332, simple_loss=0.389, pruned_loss=0.1375, over 5647334.56 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3611, pruned_loss=0.1103, over 5739155.31 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3893, pruned_loss=0.1375, over 5646020.74 frames. ], batch size: 78, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:06:38,999 INFO [optim.py:369] (1/2) 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 21:06:49,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 21:06:55,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:06:55,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:06:57,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:07:07,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4102, 3.7240, 1.5127, 1.6922], device='cuda:1'), covar=tensor([0.0981, 0.0316, 0.0843, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0528, 0.0356, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 21:07:21,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:07:23,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8947, 1.1248, 1.0584, 0.8006], device='cuda:1'), covar=tensor([0.1948, 0.2238, 0.1306, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1759, 0.1687, 0.1809], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 21:07:24,035 INFO [train.py:968] (1/2) Epoch 15, batch 23850, giga_loss[loss=0.3604, simple_loss=0.4208, pruned_loss=0.15, over 28231.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3913, pruned_loss=0.1396, over 5648332.61 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3611, pruned_loss=0.1103, over 5742888.62 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.392, pruned_loss=0.1401, over 5642333.33 frames. ], batch size: 77, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:07:29,899 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:07:59,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3046, 3.0450, 1.4176, 1.4768], device='cuda:1'), covar=tensor([0.0999, 0.0345, 0.0840, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0528, 0.0356, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0022, 0.0027], device='cuda:1')
+2023-03-07 21:08:03,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3744, 1.5040, 1.4496, 1.3217], device='cuda:1'), covar=tensor([0.2004, 0.1881, 0.1446, 0.1765], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1758, 0.1685, 0.1808], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 21:08:22,583 INFO [train.py:968] (1/2) Epoch 15, batch 23900, giga_loss[loss=0.336, simple_loss=0.3882, pruned_loss=0.1419, over 28894.00 frames. ], tot_loss[loss=0.3408, simple_loss=0.3948, pruned_loss=0.1434, over 5622251.62 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3616, pruned_loss=0.1107, over 5742987.33 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3957, pruned_loss=0.1442, over 5615436.87 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:08:28,395 INFO [optim.py:369] (1/2) 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 21:08:34,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=662831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:08:56,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=662851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:05,630 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 23950, giga_loss[loss=0.2851, simple_loss=0.3416, pruned_loss=0.1143, over 28650.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3931, pruned_loss=0.1429, over 5615189.62 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3619, pruned_loss=0.111, over 5737799.80 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3942, pruned_loss=0.1439, over 5611007.18 frames. ], batch size: 85, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:09:30,520 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:33,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:50,360 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:53,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:09:59,165 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:10:00,202 INFO [train.py:968] (1/2) Epoch 15, batch 24000, giga_loss[loss=0.3044, simple_loss=0.3691, pruned_loss=0.1198, over 28693.00 frames. ], tot_loss[loss=0.3372, simple_loss=0.3908, pruned_loss=0.1418, over 5626129.48 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.362, pruned_loss=0.1113, over 5732816.13 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3926, pruned_loss=0.1433, over 5623220.89 frames. ], batch size: 71, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:10:00,203 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 21:10:09,290 INFO [train.py:1012] (1/2) Epoch 15, validation: loss=0.2103, simple_loss=0.3175, pruned_loss=0.05152, over 944034.00 frames. 
+2023-03-07 21:10:09,291 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 21:10:14,571 INFO [optim.py:369] (1/2) 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 21:10:30,307 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=662940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:10:54,846 INFO [train.py:968] (1/2) Epoch 15, batch 24050, giga_loss[loss=0.3032, simple_loss=0.3747, pruned_loss=0.1158, over 28723.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3909, pruned_loss=0.1414, over 5627012.02 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3621, pruned_loss=0.1114, over 5735846.44 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3928, pruned_loss=0.1431, over 5620203.78 frames. ], batch size: 284, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:11:03,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5022, 1.7567, 1.3963, 1.8375], device='cuda:1'), covar=tensor([0.2440, 0.2482, 0.2729, 0.2187], device='cuda:1'), in_proj_covar=tensor([0.1381, 0.1015, 0.1225, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 21:11:13,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6961, 5.0655, 1.7207, 2.0854], device='cuda:1'), covar=tensor([0.0883, 0.0336, 0.0852, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0531, 0.0357, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:11:22,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=662994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:11:24,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=662997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:11:27,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6614, 4.4560, 4.2637, 2.1896], device='cuda:1'), covar=tensor([0.0479, 0.0641, 0.0644, 0.2005], device='cuda:1'), in_proj_covar=tensor([0.1128, 0.1040, 0.0894, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 21:11:29,740 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:11:33,441 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 24100, libri_loss[loss=0.3098, simple_loss=0.3734, pruned_loss=0.1231, over 28592.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.39, pruned_loss=0.1397, over 5619022.11 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3618, pruned_loss=0.1113, over 5738979.45 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3924, pruned_loss=0.1419, over 5608107.23 frames. ], batch size: 106, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:11:53,390 INFO [optim.py:369] (1/2) 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 21:11:53,753 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:12:00,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:12:17,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7793, 3.6097, 3.4230, 1.8655], device='cuda:1'), covar=tensor([0.0613, 0.0763, 0.0727, 0.2279], device='cuda:1'), in_proj_covar=tensor([0.1129, 0.1042, 0.0896, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 21:12:31,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:12:37,096 INFO [train.py:968] (1/2) Epoch 15, batch 24150, giga_loss[loss=0.3371, simple_loss=0.3947, pruned_loss=0.1398, over 28864.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3909, pruned_loss=0.1394, over 5623181.86 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3617, pruned_loss=0.1112, over 5738798.84 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3932, pruned_loss=0.1414, over 5613712.57 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:12:45,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3080, 3.1181, 2.9875, 1.5885], device='cuda:1'), covar=tensor([0.0918, 0.1071, 0.0931, 0.2124], device='cuda:1'), in_proj_covar=tensor([0.1130, 0.1044, 0.0897, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 21:13:08,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8247, 2.1739, 1.6303, 2.1404], device='cuda:1'), covar=tensor([0.2437, 0.2441, 0.2799, 0.2207], device='cuda:1'), in_proj_covar=tensor([0.1381, 0.1018, 0.1226, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 21:13:28,941 INFO [train.py:968] (1/2) Epoch 15, batch 24200, giga_loss[loss=0.3728, simple_loss=0.413, pruned_loss=0.1663, over 27909.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3892, pruned_loss=0.1377, over 5616885.05 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3612, pruned_loss=0.111, over 5733234.69 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3921, pruned_loss=0.1402, over 5611829.38 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:13:35,632 INFO [optim.py:369] (1/2) 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 21:13:42,659 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:13:53,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4419, 1.5385, 1.1742, 1.1843], device='cuda:1'), covar=tensor([0.0781, 0.0455, 0.0993, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0439, 0.0506, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 21:14:11,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1169, 2.5677, 1.2110, 1.3098], device='cuda:1'), covar=tensor([0.1036, 0.0392, 0.0931, 0.1409], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0529, 0.0357, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:14:19,165 INFO [train.py:968] (1/2) Epoch 15, batch 24250, giga_loss[loss=0.2824, simple_loss=0.3615, pruned_loss=0.1016, over 29123.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3872, pruned_loss=0.135, over 5618250.37 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3619, pruned_loss=0.1116, over 5722241.68 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3893, pruned_loss=0.1368, over 5621519.39 frames. ], batch size: 155, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:14:20,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-07 21:14:55,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:15:07,196 INFO [train.py:968] (1/2) Epoch 15, batch 24300, giga_loss[loss=0.3075, simple_loss=0.3703, pruned_loss=0.1224, over 27960.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3842, pruned_loss=0.1323, over 5617445.46 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3616, pruned_loss=0.1116, over 5723206.43 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3867, pruned_loss=0.1343, over 5617160.34 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:15:12,681 INFO [optim.py:369] (1/2) 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 21:15:47,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9831, 1.3154, 1.4903, 1.0139], device='cuda:1'), covar=tensor([0.1448, 0.1095, 0.1776, 0.1473], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0736, 0.0688, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 21:15:52,138 INFO [train.py:968] (1/2) Epoch 15, batch 24350, giga_loss[loss=0.2658, simple_loss=0.3345, pruned_loss=0.09855, over 28502.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.38, pruned_loss=0.1293, over 5630987.15 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3613, pruned_loss=0.1116, over 5727333.67 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.383, pruned_loss=0.1315, over 5624352.45 frames. ], batch size: 85, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:16:40,700 INFO [train.py:968] (1/2) Epoch 15, batch 24400, giga_loss[loss=0.2552, simple_loss=0.3311, pruned_loss=0.08962, over 28562.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3774, pruned_loss=0.1276, over 5632367.02 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3612, pruned_loss=0.1118, over 5730901.54 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3801, pruned_loss=0.1295, over 5622366.16 frames. ], batch size: 60, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:16:44,735 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:16:47,824 INFO [optim.py:369] (1/2) 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 21:17:12,840 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:17:15,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=663352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:17:30,586 INFO [train.py:968] (1/2) Epoch 15, batch 24450, giga_loss[loss=0.35, simple_loss=0.3996, pruned_loss=0.1502, over 28769.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3784, pruned_loss=0.1285, over 5639619.68 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3615, pruned_loss=0.112, over 5731566.90 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3804, pruned_loss=0.1299, over 5629898.12 frames. ], batch size: 284, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:17:45,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:17:51,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8804, 1.1080, 3.3099, 2.8972], device='cuda:1'), covar=tensor([0.1857, 0.2803, 0.0537, 0.1055], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0616, 0.0901, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 21:18:24,680 INFO [train.py:968] (1/2) Epoch 15, batch 24500, libri_loss[loss=0.2466, simple_loss=0.3151, pruned_loss=0.08903, over 27750.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3775, pruned_loss=0.1275, over 5645551.79 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3613, pruned_loss=0.1121, over 5732831.35 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3797, pruned_loss=0.1289, over 5635124.44 frames. ], batch size: 61, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:18:31,770 INFO [optim.py:369] (1/2) 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 21:18:43,967 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:19:12,492 INFO [train.py:968] (1/2) Epoch 15, batch 24550, giga_loss[loss=0.3329, simple_loss=0.4048, pruned_loss=0.1305, over 28839.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3753, pruned_loss=0.1245, over 5660984.78 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3616, pruned_loss=0.1124, over 5735840.34 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3774, pruned_loss=0.1259, over 5646639.47 frames. ], batch size: 284, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:19:28,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3446, 1.5039, 1.6129, 1.4086], device='cuda:1'), covar=tensor([0.1519, 0.1541, 0.1629, 0.1567], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0733, 0.0685, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 21:19:51,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:20:03,834 INFO [train.py:968] (1/2) Epoch 15, batch 24600, libri_loss[loss=0.3144, simple_loss=0.379, pruned_loss=0.1249, over 18869.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3771, pruned_loss=0.1234, over 5657000.18 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.362, pruned_loss=0.1128, over 5724287.45 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3787, pruned_loss=0.1244, over 5655237.30 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:20:10,415 INFO [optim.py:369] (1/2) 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 21:20:20,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.54 vs. limit=5.0
+2023-03-07 21:20:27,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3915, 3.0940, 1.4991, 1.5018], device='cuda:1'), covar=tensor([0.0933, 0.0377, 0.0893, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0533, 0.0359, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:20:56,051 INFO [train.py:968] (1/2) Epoch 15, batch 24650, giga_loss[loss=0.2804, simple_loss=0.3594, pruned_loss=0.1007, over 28725.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3784, pruned_loss=0.1243, over 5639795.72 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3621, pruned_loss=0.1129, over 5716926.23 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3798, pruned_loss=0.1251, over 5644168.58 frames. ], batch size: 284, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:21:10,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:21:13,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=663584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:21:42,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:21:46,703 INFO [train.py:968] (1/2) Epoch 15, batch 24700, giga_loss[loss=0.322, simple_loss=0.3887, pruned_loss=0.1277, over 28633.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3788, pruned_loss=0.1245, over 5659967.14 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3622, pruned_loss=0.113, over 5718795.49 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.38, pruned_loss=0.1252, over 5661163.31 frames. ], batch size: 262, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:21:56,309 INFO [optim.py:369] (1/2) 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 21:22:13,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2274, 1.5570, 1.5344, 1.1147], device='cuda:1'), covar=tensor([0.1576, 0.2365, 0.1305, 0.1528], device='cuda:1'), in_proj_covar=tensor([0.0857, 0.0693, 0.0902, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 21:22:21,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:22:23,488 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 24750, giga_loss[loss=0.2996, simple_loss=0.3769, pruned_loss=0.1111, over 28685.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3776, pruned_loss=0.1242, over 5674594.35 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3623, pruned_loss=0.1131, over 5719668.66 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3789, pruned_loss=0.1249, over 5673703.01 frames. ], batch size: 85, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:22:38,093 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=663669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:22:50,030 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:23:04,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2616, 1.2529, 3.9895, 3.3649], device='cuda:1'), covar=tensor([0.1635, 0.2775, 0.0445, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0615, 0.0900, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 21:23:04,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=663698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:23:22,079 INFO [train.py:968] (1/2) Epoch 15, batch 24800, giga_loss[loss=0.3058, simple_loss=0.3758, pruned_loss=0.1179, over 29085.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3758, pruned_loss=0.1243, over 5672764.69 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3626, pruned_loss=0.1136, over 5719244.07 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3767, pruned_loss=0.1246, over 5671630.67 frames. ], batch size: 155, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:23:29,792 INFO [optim.py:369] (1/2) 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 21:23:53,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5539, 1.6745, 1.8035, 1.3728], device='cuda:1'), covar=tensor([0.1779, 0.2427, 0.1438, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0696, 0.0905, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 21:23:57,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 21:24:02,706 INFO [train.py:968] (1/2) Epoch 15, batch 24850, giga_loss[loss=0.321, simple_loss=0.3781, pruned_loss=0.132, over 28246.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3756, pruned_loss=0.1248, over 5675299.96 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3635, pruned_loss=0.1144, over 5726291.86 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3761, pruned_loss=0.1247, over 5666133.22 frames. ], batch size: 368, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:24:38,004 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6309, 1.7204, 1.2958, 1.3688], device='cuda:1'), covar=tensor([0.0823, 0.0589, 0.1001, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0370, 0.0437, 0.0503, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 21:24:46,386 INFO [train.py:968] (1/2) Epoch 15, batch 24900, giga_loss[loss=0.3227, simple_loss=0.3968, pruned_loss=0.1243, over 28456.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3753, pruned_loss=0.1231, over 5673493.34 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3639, pruned_loss=0.1147, over 5717312.84 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3754, pruned_loss=0.123, over 5672805.45 frames. ], batch size: 78, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:24:56,228 INFO [optim.py:369] (1/2) 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 21:25:07,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5013, 1.7365, 1.4035, 1.5826], device='cuda:1'), covar=tensor([0.2635, 0.2704, 0.3047, 0.2330], device='cuda:1'), in_proj_covar=tensor([0.1387, 0.1019, 0.1231, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 21:25:08,964 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=663841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:25:13,148 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 15, batch 24950, giga_loss[loss=0.3023, simple_loss=0.3707, pruned_loss=0.117, over 28863.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3746, pruned_loss=0.1218, over 5677482.36 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3636, pruned_loss=0.1145, over 5718882.79 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3751, pruned_loss=0.1219, over 5675249.49 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:25:42,687 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=663873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:26:01,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4082, 3.6753, 1.4659, 1.6029], device='cuda:1'), covar=tensor([0.1012, 0.0346, 0.0946, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0531, 0.0358, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:26:21,591 INFO [train.py:968] (1/2) Epoch 15, batch 25000, giga_loss[loss=0.3121, simple_loss=0.3865, pruned_loss=0.1188, over 28854.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3754, pruned_loss=0.1228, over 5663653.13 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3641, pruned_loss=0.1149, over 5708027.75 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3757, pruned_loss=0.1228, over 5669130.50 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:26:29,375 INFO [optim.py:369] (1/2) 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 21:27:08,137 INFO [train.py:968] (1/2) Epoch 15, batch 25050, giga_loss[loss=0.2925, simple_loss=0.363, pruned_loss=0.111, over 28952.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3734, pruned_loss=0.122, over 5675091.26 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3639, pruned_loss=0.1151, over 5705865.50 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3741, pruned_loss=0.122, over 5679543.45 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:27:42,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-07 21:27:57,996 INFO [train.py:968] (1/2) Epoch 15, batch 25100, giga_loss[loss=0.3527, simple_loss=0.4004, pruned_loss=0.1525, over 28523.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3735, pruned_loss=0.1234, over 5661925.12 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3642, pruned_loss=0.1155, over 5707096.89 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3741, pruned_loss=0.1233, over 5663281.06 frames. ], batch size: 307, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:28:05,245 INFO [optim.py:369] (1/2) 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 21:28:18,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:28:41,046 INFO [train.py:968] (1/2) Epoch 15, batch 25150, giga_loss[loss=0.3203, simple_loss=0.3775, pruned_loss=0.1316, over 27857.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3739, pruned_loss=0.1244, over 5666057.70 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3647, pruned_loss=0.1159, over 5709354.68 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3743, pruned_loss=0.1242, over 5663870.22 frames. ], batch size: 412, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:28:54,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8600, 1.8615, 1.7857, 1.6113], device='cuda:1'), covar=tensor([0.1607, 0.2235, 0.1994, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0735, 0.0688, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 21:29:29,162 INFO [train.py:968] (1/2) Epoch 15, batch 25200, giga_loss[loss=0.263, simple_loss=0.3334, pruned_loss=0.09631, over 28758.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3732, pruned_loss=0.1244, over 5669110.48 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3644, pruned_loss=0.1158, over 5713455.02 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1245, over 5662952.73 frames. ], batch size: 99, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:29:36,577 INFO [optim.py:369] (1/2) 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 21:29:49,930 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8430, 2.3143, 2.0126, 1.4993], device='cuda:1'), covar=tensor([0.3080, 0.1850, 0.2076, 0.2820], device='cuda:1'), in_proj_covar=tensor([0.1837, 0.1760, 0.1696, 0.1822], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 21:30:15,057 INFO [train.py:968] (1/2) Epoch 15, batch 25250, giga_loss[loss=0.3096, simple_loss=0.3741, pruned_loss=0.1226, over 28953.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3707, pruned_loss=0.1228, over 5673839.50 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3643, pruned_loss=0.1157, over 5713944.50 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5667424.98 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:30:17,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4953, 1.5682, 1.2443, 1.1361], device='cuda:1'), covar=tensor([0.0933, 0.0603, 0.1108, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0438, 0.0503, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 21:30:27,292 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-07 21:30:33,345 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:30:35,982 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 25300, giga_loss[loss=0.3003, simple_loss=0.3584, pruned_loss=0.1211, over 28636.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3714, pruned_loss=0.1241, over 5668557.28 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3645, pruned_loss=0.1159, over 5717626.50 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.372, pruned_loss=0.1244, over 5658910.44 frames. ], batch size: 85, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:31:05,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=664219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:31:15,529 INFO [optim.py:369] (1/2) 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 21:31:28,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:31:45,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 21:31:57,359 INFO [train.py:968] (1/2) Epoch 15, batch 25350, giga_loss[loss=0.3675, simple_loss=0.4157, pruned_loss=0.1596, over 28240.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.373, pruned_loss=0.1247, over 5669213.25 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3649, pruned_loss=0.1162, over 5719559.44 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3732, pruned_loss=0.1248, over 5659070.39 frames. ], batch size: 368, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:31:59,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-07 21:32:01,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664275.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:32:29,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:32:40,082 INFO [train.py:968] (1/2) Epoch 15, batch 25400, giga_loss[loss=0.2986, simple_loss=0.3621, pruned_loss=0.1175, over 28850.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.373, pruned_loss=0.1234, over 5671170.09 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3648, pruned_loss=0.1161, over 5719594.82 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3735, pruned_loss=0.1237, over 5662163.87 frames. ], batch size: 112, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:32:47,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0345, 0.9792, 3.3898, 3.0741], device='cuda:1'), covar=tensor([0.1728, 0.2870, 0.0553, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0703, 0.0613, 0.0902, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 21:32:51,748 INFO [optim.py:369] (1/2) 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 21:33:03,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4820, 1.7380, 1.3589, 1.6335], device='cuda:1'), covar=tensor([0.2536, 0.2579, 0.2966, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.1386, 0.1018, 0.1230, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 21:33:28,481 INFO [train.py:968] (1/2) Epoch 15, batch 25450, giga_loss[loss=0.2823, simple_loss=0.3551, pruned_loss=0.1048, over 28879.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1222, over 5666793.27 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3647, pruned_loss=0.116, over 5721521.11 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3726, pruned_loss=0.1226, over 5657464.73 frames. ], batch size: 227, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:33:44,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:33:47,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=664385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:34:12,320 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 25500, giga_loss[loss=0.2791, simple_loss=0.3433, pruned_loss=0.1075, over 28304.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1228, over 5673466.94 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3642, pruned_loss=0.1157, over 5725708.86 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1237, over 5660739.77 frames. ], batch size: 77, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:34:25,518 INFO [optim.py:369] (1/2) 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 21:34:31,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-07 21:34:42,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3640, 1.5584, 1.3380, 1.4702], device='cuda:1'), covar=tensor([0.0768, 0.0325, 0.0319, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 21:35:08,090 INFO [train.py:968] (1/2) Epoch 15, batch 25550, giga_loss[loss=0.3709, simple_loss=0.4173, pruned_loss=0.1623, over 28239.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3772, pruned_loss=0.1276, over 5659220.91 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3644, pruned_loss=0.1157, over 5727346.17 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.378, pruned_loss=0.1283, over 5647029.69 frames. ], batch size: 368, lr: 2.14e-03, grad_scale: 2.0
+2023-03-07 21:35:57,752 INFO [train.py:968] (1/2) Epoch 15, batch 25600, giga_loss[loss=0.2944, simple_loss=0.3564, pruned_loss=0.1162, over 28934.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3776, pruned_loss=0.1287, over 5663661.80 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3645, pruned_loss=0.1159, over 5729156.23 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3782, pruned_loss=0.1293, over 5652071.03 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:36:08,902 INFO [optim.py:369] (1/2) 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 21:36:27,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8198, 1.1081, 2.8937, 2.7750], device='cuda:1'), covar=tensor([0.1674, 0.2503, 0.0620, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0701, 0.0610, 0.0898, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 21:36:38,902 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 25650, giga_loss[loss=0.2625, simple_loss=0.3367, pruned_loss=0.09415, over 28628.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3794, pruned_loss=0.1309, over 5662363.44 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3649, pruned_loss=0.116, over 5724894.05 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3801, pruned_loss=0.1318, over 5655328.24 frames. ], batch size: 71, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:36:47,862 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:37:30,466 INFO [train.py:968] (1/2) Epoch 15, batch 25700, giga_loss[loss=0.3653, simple_loss=0.4067, pruned_loss=0.1619, over 27581.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3808, pruned_loss=0.1322, over 5652034.26 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3652, pruned_loss=0.1163, over 5724440.21 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3815, pruned_loss=0.1331, over 5644962.63 frames. ], batch size: 472, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:37:32,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:37:39,493 INFO [optim.py:369] (1/2) 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 21:37:41,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8250, 2.2632, 1.8170, 1.2977], device='cuda:1'), covar=tensor([0.2835, 0.2323, 0.2256, 0.3345], device='cuda:1'), in_proj_covar=tensor([0.1625, 0.1555, 0.1529, 0.1337], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 21:37:57,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:38:10,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5041, 1.6175, 1.5935, 1.3823], device='cuda:1'), covar=tensor([0.1590, 0.2080, 0.2002, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0738, 0.0689, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 21:38:13,625 INFO [train.py:968] (1/2) Epoch 15, batch 25750, giga_loss[loss=0.2937, simple_loss=0.368, pruned_loss=0.1097, over 28980.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3799, pruned_loss=0.1316, over 5658952.00 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3654, pruned_loss=0.1166, over 5719893.24 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3808, pruned_loss=0.1325, over 5655244.57 frames. ], batch size: 128, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:38:28,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:38:52,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4710, 1.7608, 1.6239, 1.3490], device='cuda:1'), covar=tensor([0.3082, 0.2255, 0.2059, 0.2680], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1767, 0.1691, 0.1825], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 21:38:57,336 INFO [train.py:968] (1/2) Epoch 15, batch 25800, giga_loss[loss=0.2818, simple_loss=0.3577, pruned_loss=0.1029, over 28924.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3804, pruned_loss=0.1314, over 5660613.71 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3661, pruned_loss=0.1171, over 5722139.35 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3809, pruned_loss=0.1321, over 5653198.98 frames. ], batch size: 199, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:39:07,178 INFO [optim.py:369] (1/2) 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 21:39:13,506 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=664737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:39:43,359 INFO [train.py:968] (1/2) Epoch 15, batch 25850, giga_loss[loss=0.3223, simple_loss=0.3818, pruned_loss=0.1314, over 28815.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3789, pruned_loss=0.1288, over 5654795.79 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3667, pruned_loss=0.1176, over 5712203.71 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.379, pruned_loss=0.1292, over 5656398.34 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:40:07,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664793.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:40:09,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=664796.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 21:40:30,812 INFO [train.py:968] (1/2) Epoch 15, batch 25900, giga_loss[loss=0.3019, simple_loss=0.3714, pruned_loss=0.1162, over 28707.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5651147.95 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.1181, over 5711210.48 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3759, pruned_loss=0.1271, over 5652430.46 frames. ], batch size: 284, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:40:36,066 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=664824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:40:36,792 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=664825.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:40:38,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=664827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:40:41,505 INFO [optim.py:369] (1/2) 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 21:40:52,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2804, 2.5822, 1.3563, 1.3912], device='cuda:1'), covar=tensor([0.0976, 0.0351, 0.0859, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0533, 0.0359, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:41:06,898 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=664856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:41:19,549 INFO [train.py:968] (1/2) Epoch 15, batch 25950, giga_loss[loss=0.2868, simple_loss=0.3544, pruned_loss=0.1096, over 28868.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3733, pruned_loss=0.1258, over 5663237.36 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3675, pruned_loss=0.1183, over 5711805.43 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3729, pruned_loss=0.1256, over 5663420.83 frames. ], batch size: 145, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:42:13,194 INFO [train.py:968] (1/2) Epoch 15, batch 26000, giga_loss[loss=0.319, simple_loss=0.38, pruned_loss=0.129, over 28971.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3721, pruned_loss=0.1247, over 5668641.65 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3674, pruned_loss=0.1182, over 5712773.79 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3719, pruned_loss=0.1247, over 5667573.10 frames. ], batch size: 155, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:42:20,987 INFO [optim.py:369] (1/2) 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 21:42:24,651 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:42:55,048 INFO [train.py:968] (1/2) Epoch 15, batch 26050, giga_loss[loss=0.3248, simple_loss=0.3911, pruned_loss=0.1293, over 28921.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3751, pruned_loss=0.126, over 5670321.51 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3679, pruned_loss=0.1186, over 5708874.20 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3747, pruned_loss=0.1259, over 5672318.50 frames. ], batch size: 186, lr: 2.14e-03, grad_scale: 8.0
+2023-03-07 21:43:16,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-07 21:43:23,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=664997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:43:43,280 INFO [train.py:968] (1/2) Epoch 15, batch 26100, giga_loss[loss=0.2815, simple_loss=0.3649, pruned_loss=0.09907, over 28704.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3774, pruned_loss=0.1244, over 5678790.18 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3675, pruned_loss=0.1183, over 5711741.06 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3776, pruned_loss=0.1246, over 5677104.47 frames. ], batch size: 92, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:43:53,813 INFO [optim.py:369] (1/2) 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 21:44:05,395 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:12,332 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:25,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7860, 2.1515, 1.7724, 1.8877], device='cuda:1'), covar=tensor([0.0741, 0.0261, 0.0303, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 21:44:29,727 INFO [train.py:968] (1/2) Epoch 15, batch 26150, giga_loss[loss=0.3234, simple_loss=0.394, pruned_loss=0.1264, over 28989.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3782, pruned_loss=0.1241, over 5682777.25 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3679, pruned_loss=0.1188, over 5714165.93 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3783, pruned_loss=0.124, over 5678480.76 frames. ], batch size: 128, lr: 2.14e-03, grad_scale: 4.0
+2023-03-07 21:44:38,330 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:40,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:50,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:44:52,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:11,527 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:12,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:19,618 INFO [train.py:968] (1/2) Epoch 15, batch 26200, giga_loss[loss=0.3162, simple_loss=0.3811, pruned_loss=0.1257, over 28901.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3799, pruned_loss=0.1259, over 5682664.81 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3678, pruned_loss=0.1187, over 5716000.79 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3802, pruned_loss=0.1259, over 5677519.85 frames. ], batch size: 227, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:45:23,672 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:30,513 INFO [optim.py:369] (1/2) 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 21:45:35,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5542, 2.2751, 1.6424, 0.7200], device='cuda:1'), covar=tensor([0.5130, 0.2436, 0.3655, 0.5758], device='cuda:1'), in_proj_covar=tensor([0.1622, 0.1554, 0.1525, 0.1335], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 21:45:38,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:45:41,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:46:00,556 INFO [train.py:968] (1/2) Epoch 15, batch 26250, giga_loss[loss=0.2895, simple_loss=0.367, pruned_loss=0.106, over 29040.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3803, pruned_loss=0.1262, over 5689507.04 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3676, pruned_loss=0.1188, over 5717299.01 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3809, pruned_loss=0.1264, over 5683634.87 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:46:05,820 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:46:16,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-07 21:46:49,297 INFO [train.py:968] (1/2) Epoch 15, batch 26300, giga_loss[loss=0.2767, simple_loss=0.3505, pruned_loss=0.1015, over 28476.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3812, pruned_loss=0.1284, over 5683392.35 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3677, pruned_loss=0.1188, over 5722101.54 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3819, pruned_loss=0.1287, over 5673703.84 frames. ], batch size: 78, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:47:00,249 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:47:26,095 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:47:28,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:47:36,548 INFO [train.py:968] (1/2) Epoch 15, batch 26350, giga_loss[loss=0.2883, simple_loss=0.3539, pruned_loss=0.1113, over 28978.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3785, pruned_loss=0.1269, over 5685447.39 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3674, pruned_loss=0.1187, over 5713073.90 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3796, pruned_loss=0.1275, over 5684922.80 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:47:40,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665273.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:47:52,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:48:21,219 INFO [train.py:968] (1/2) Epoch 15, batch 26400, giga_loss[loss=0.314, simple_loss=0.3734, pruned_loss=0.1273, over 28852.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.377, pruned_loss=0.1269, over 5673501.96 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3677, pruned_loss=0.1191, over 5703274.43 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.378, pruned_loss=0.1273, over 5680823.37 frames. ], batch size: 112, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 21:48:34,136 INFO [optim.py:369] (1/2) 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 21:49:01,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-07 21:49:12,913 INFO [train.py:968] (1/2) Epoch 15, batch 26450, giga_loss[loss=0.2886, simple_loss=0.3548, pruned_loss=0.1112, over 28867.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.375, pruned_loss=0.1261, over 5672897.51 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3675, pruned_loss=0.1191, over 5703863.51 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3761, pruned_loss=0.1265, over 5677489.97 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 21:49:55,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:49:56,152 INFO [train.py:968] (1/2) Epoch 15, batch 26500, giga_loss[loss=0.3135, simple_loss=0.3858, pruned_loss=0.1206, over 28686.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3773, pruned_loss=0.1279, over 5678522.58 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.368, pruned_loss=0.1194, over 5708442.67 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3779, pruned_loss=0.1282, over 5677214.17 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:50:02,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:50:05,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2851, 1.9296, 1.4128, 0.5808], device='cuda:1'), covar=tensor([0.3957, 0.2168, 0.3207, 0.4742], device='cuda:1'), in_proj_covar=tensor([0.1634, 0.1565, 0.1532, 0.1341], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 21:50:08,473 INFO [optim.py:369] (1/2) 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 21:50:19,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6160, 4.4986, 1.7957, 1.7457], device='cuda:1'), covar=tensor([0.0944, 0.0313, 0.0842, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0534, 0.0358, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:50:41,808 INFO [train.py:968] (1/2) Epoch 15, batch 26550, giga_loss[loss=0.3077, simple_loss=0.3695, pruned_loss=0.1229, over 28789.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3774, pruned_loss=0.1286, over 5682400.82 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3681, pruned_loss=0.1194, over 5712103.82 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3781, pruned_loss=0.1291, over 5677409.77 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:50:54,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9584, 3.7813, 3.6213, 1.8451], device='cuda:1'), covar=tensor([0.0682, 0.0784, 0.0790, 0.1884], device='cuda:1'), in_proj_covar=tensor([0.1149, 0.1061, 0.0915, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 21:51:00,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5092, 1.7032, 1.6187, 1.5489], device='cuda:1'), covar=tensor([0.1301, 0.1567, 0.1720, 0.1504], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0742, 0.0692, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 21:51:24,542 INFO [train.py:968] (1/2) Epoch 15, batch 26600, giga_loss[loss=0.2833, simple_loss=0.3572, pruned_loss=0.1047, over 28725.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3751, pruned_loss=0.1281, over 5658164.04 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1198, over 5706581.73 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3755, pruned_loss=0.1283, over 5658611.06 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:51:38,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-07 21:51:39,767 INFO [optim.py:369] (1/2) 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 21:52:08,613 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:11,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:14,270 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:14,596 INFO [train.py:968] (1/2) Epoch 15, batch 26650, giga_loss[loss=0.3249, simple_loss=0.3859, pruned_loss=0.1319, over 28763.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3747, pruned_loss=0.1273, over 5656926.63 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3688, pruned_loss=0.1201, over 5708635.32 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3748, pruned_loss=0.1273, over 5655010.70 frames. ], batch size: 285, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:52:16,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:38,384 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:52:44,952 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:53:00,084 INFO [train.py:968] (1/2) Epoch 15, batch 26700, giga_loss[loss=0.2902, simple_loss=0.3642, pruned_loss=0.1081, over 28683.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3752, pruned_loss=0.1262, over 5666396.03 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3691, pruned_loss=0.1204, over 5711553.24 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1262, over 5661060.90 frames. ], batch size: 242, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:53:02,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:53:12,792 INFO [optim.py:369] (1/2) 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 21:53:24,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 21:53:28,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=665648.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:53:51,800 INFO [train.py:968] (1/2) Epoch 15, batch 26750, libri_loss[loss=0.3357, simple_loss=0.3973, pruned_loss=0.1371, over 29539.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.377, pruned_loss=0.1279, over 5662518.50 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5714556.68 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3773, pruned_loss=0.1281, over 5654982.32 frames. ], batch size: 89, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:54:01,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5664, 1.2724, 4.2112, 3.3325], device='cuda:1'), covar=tensor([0.1514, 0.2643, 0.0427, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0617, 0.0909, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 21:54:24,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4217, 1.6033, 1.5197, 1.4357], device='cuda:1'), covar=tensor([0.1533, 0.1692, 0.1946, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0744, 0.0693, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 21:54:36,639 INFO [train.py:968] (1/2) Epoch 15, batch 26800, giga_loss[loss=0.3499, simple_loss=0.4015, pruned_loss=0.1492, over 28277.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3762, pruned_loss=0.1268, over 5668860.53 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1199, over 5712850.98 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3769, pruned_loss=0.1273, over 5663793.04 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 21:54:48,294 INFO [optim.py:369] (1/2) 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 21:55:15,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:55:18,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:55:20,419 INFO [train.py:968] (1/2) Epoch 15, batch 26850, giga_loss[loss=0.326, simple_loss=0.3917, pruned_loss=0.1301, over 28659.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3784, pruned_loss=0.1252, over 5671897.41 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3688, pruned_loss=0.12, over 5711808.80 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3788, pruned_loss=0.1256, over 5668314.34 frames. ], batch size: 242, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:55:41,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=665791.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:55:44,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=665794.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:55:45,625 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 26900, giga_loss[loss=0.3049, simple_loss=0.377, pruned_loss=0.1164, over 28748.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3814, pruned_loss=0.1261, over 5687311.52 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1198, over 5719400.96 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3824, pruned_loss=0.1268, over 5676431.57 frames. ], batch size: 99, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:56:07,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-07 21:56:10,717 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=665823.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:56:19,573 INFO [optim.py:369] (1/2) 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 21:56:51,161 INFO [train.py:968] (1/2) Epoch 15, batch 26950, giga_loss[loss=0.4033, simple_loss=0.4344, pruned_loss=0.1861, over 28995.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3845, pruned_loss=0.1282, over 5685573.02 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3683, pruned_loss=0.1197, over 5720627.60 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3857, pruned_loss=0.129, over 5675440.97 frames. ], batch size: 128, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:57:22,391 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 21:57:27,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4003, 3.8024, 1.7352, 1.5923], device='cuda:1'), covar=tensor([0.0892, 0.0361, 0.0846, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0532, 0.0358, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 21:57:41,430 INFO [train.py:968] (1/2) Epoch 15, batch 27000, giga_loss[loss=0.4638, simple_loss=0.469, pruned_loss=0.2293, over 23563.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3873, pruned_loss=0.1316, over 5677865.53 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.1199, over 5720527.46 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3884, pruned_loss=0.1322, over 5669145.38 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:57:41,430 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 21:57:49,831 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 21:58:03,104 INFO [optim.py:369] (1/2) 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 21:58:39,112 INFO [train.py:968] (1/2) Epoch 15, batch 27050, giga_loss[loss=0.3306, simple_loss=0.3904, pruned_loss=0.1354, over 28040.00 frames. ], tot_loss[loss=0.3275, simple_loss=0.388, pruned_loss=0.1335, over 5654108.87 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3684, pruned_loss=0.12, over 5714614.44 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3893, pruned_loss=0.1342, over 5651344.42 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:58:41,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=665971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 21:59:09,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5423, 1.1865, 4.8862, 3.5849], device='cuda:1'), covar=tensor([0.1745, 0.2896, 0.0390, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0706, 0.0613, 0.0909, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 21:59:26,157 INFO [train.py:968] (1/2) Epoch 15, batch 27100, giga_loss[loss=0.3749, simple_loss=0.4053, pruned_loss=0.1723, over 23485.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3858, pruned_loss=0.1319, over 5659589.25 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.12, over 5715832.34 frames. ], giga_tot_loss[loss=0.3266, simple_loss=0.3874, pruned_loss=0.1329, over 5654244.88 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 21:59:40,937 INFO [optim.py:369] (1/2) 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 22:00:14,787 INFO [train.py:968] (1/2) Epoch 15, batch 27150, giga_loss[loss=0.3255, simple_loss=0.3925, pruned_loss=0.1292, over 28928.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3845, pruned_loss=0.1305, over 5641614.52 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3687, pruned_loss=0.1203, over 5707685.91 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3857, pruned_loss=0.1311, over 5644619.87 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:00:33,003 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=666088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:01:04,223 INFO [train.py:968] (1/2) Epoch 15, batch 27200, giga_loss[loss=0.3159, simple_loss=0.3848, pruned_loss=0.1235, over 28607.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3833, pruned_loss=0.1275, over 5643325.64 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5699761.53 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3843, pruned_loss=0.128, over 5652188.51 frames. ], batch size: 92, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:01:17,489 INFO [optim.py:369] (1/2) 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 22:01:29,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2887, 1.1778, 3.9941, 3.3929], device='cuda:1'), covar=tensor([0.1630, 0.2741, 0.0438, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0617, 0.0912, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-07 22:01:49,578 INFO [train.py:968] (1/2) Epoch 15, batch 27250, giga_loss[loss=0.4001, simple_loss=0.4278, pruned_loss=0.1862, over 26553.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3832, pruned_loss=0.127, over 5650861.67 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1205, over 5701900.77 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3842, pruned_loss=0.1274, over 5654714.56 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:02:37,091 INFO [train.py:968] (1/2) Epoch 15, batch 27300, giga_loss[loss=0.2966, simple_loss=0.3641, pruned_loss=0.1146, over 28923.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3832, pruned_loss=0.1276, over 5661431.56 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3694, pruned_loss=0.1209, over 5708584.66 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3841, pruned_loss=0.1278, over 5657485.33 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:02:53,886 INFO [optim.py:369] (1/2) 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 22:03:23,217 INFO [train.py:968] (1/2) Epoch 15, batch 27350, libri_loss[loss=0.2676, simple_loss=0.3324, pruned_loss=0.1014, over 29652.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3824, pruned_loss=0.1277, over 5674717.38 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3695, pruned_loss=0.121, over 5714584.05 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3834, pruned_loss=0.128, over 5664757.32 frames. ], batch size: 69, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:03:26,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=666274.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:04:08,941 INFO [train.py:968] (1/2) Epoch 15, batch 27400, giga_loss[loss=0.4454, simple_loss=0.4515, pruned_loss=0.2196, over 26693.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3796, pruned_loss=0.1273, over 5664230.65 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3688, pruned_loss=0.1205, over 5716745.73 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3815, pruned_loss=0.1282, over 5652681.16 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:04:22,105 INFO [optim.py:369] (1/2) 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 22:04:35,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=666346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:04:57,331 INFO [train.py:968] (1/2) Epoch 15, batch 27450, giga_loss[loss=0.277, simple_loss=0.3479, pruned_loss=0.103, over 28964.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3789, pruned_loss=0.1278, over 5643744.26 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.369, pruned_loss=0.1207, over 5708637.15 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3803, pruned_loss=0.1285, over 5640878.76 frames. ], batch size: 227, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:05:20,790 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9470, 1.3429, 1.1480, 0.2272], device='cuda:1'), covar=tensor([0.3026, 0.2479, 0.3264, 0.4209], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1553, 0.1527, 0.1331], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 22:05:28,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-07 22:05:48,331 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=666417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:05:49,563 INFO [train.py:968] (1/2) Epoch 15, batch 27500, giga_loss[loss=0.357, simple_loss=0.4067, pruned_loss=0.1536, over 28519.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3777, pruned_loss=0.128, over 5650010.29 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3688, pruned_loss=0.1207, over 5711509.09 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3792, pruned_loss=0.1286, over 5644118.55 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:05:50,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=666420.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:05:51,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5379, 1.7469, 1.6746, 1.5860], device='cuda:1'), covar=tensor([0.1710, 0.2046, 0.2169, 0.1978], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0744, 0.0694, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 22:06:01,966 INFO [optim.py:369] (1/2) 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 22:06:16,516 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=666449.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:06:30,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=666463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:06:34,138 INFO [train.py:968] (1/2) Epoch 15, batch 27550, giga_loss[loss=0.3002, simple_loss=0.368, pruned_loss=0.1162, over 28723.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3774, pruned_loss=0.1286, over 5643805.24 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3693, pruned_loss=0.1211, over 5704414.35 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3784, pruned_loss=0.129, over 5642312.90 frames. ], batch size: 92, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:06:51,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=666489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:06:53,742 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 27600, giga_loss[loss=0.2994, simple_loss=0.3484, pruned_loss=0.1252, over 23707.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3762, pruned_loss=0.1279, over 5649225.19 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3692, pruned_loss=0.121, over 5709751.81 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3774, pruned_loss=0.1284, over 5641649.57 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:07:19,138 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=666521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:07:27,599 INFO [optim.py:369] (1/2) 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 22:07:55,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4120, 1.4640, 1.3279, 1.5700], device='cuda:1'), covar=tensor([0.0791, 0.0341, 0.0337, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 22:08:01,790 INFO [train.py:968] (1/2) Epoch 15, batch 27650, giga_loss[loss=0.2722, simple_loss=0.3497, pruned_loss=0.09738, over 28875.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3717, pruned_loss=0.1227, over 5662678.12 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3685, pruned_loss=0.1205, over 5710492.87 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1238, over 5654307.12 frames. ], batch size: 227, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:08:33,171 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=666606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:08:36,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=666609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:08:45,579 INFO [train.py:968] (1/2) Epoch 15, batch 27700, libri_loss[loss=0.3711, simple_loss=0.4194, pruned_loss=0.1614, over 29192.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.369, pruned_loss=0.1197, over 5673184.44 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1205, over 5717928.66 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3702, pruned_loss=0.1205, over 5657765.26 frames. ], batch size: 101, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:08:50,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6590, 1.8238, 1.5699, 1.6128], device='cuda:1'), covar=tensor([0.1664, 0.2434, 0.2237, 0.2145], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0743, 0.0693, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 22:09:02,232 INFO [optim.py:369] (1/2) 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 22:09:04,874 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=666638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:09:31,699 INFO [train.py:968] (1/2) Epoch 15, batch 27750, giga_loss[loss=0.307, simple_loss=0.3689, pruned_loss=0.1226, over 28350.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1199, over 5660918.08 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3684, pruned_loss=0.1206, over 5715355.50 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3699, pruned_loss=0.1204, over 5648248.13 frames. ], batch size: 65, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:10:13,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=666708.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:10:23,644 INFO [train.py:968] (1/2) Epoch 15, batch 27800, giga_loss[loss=0.306, simple_loss=0.3684, pruned_loss=0.1218, over 28950.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1184, over 5669375.83 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3678, pruned_loss=0.1202, over 5719683.19 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3671, pruned_loss=0.1192, over 5654068.34 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:10:43,134 INFO [optim.py:369] (1/2) 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 22:10:53,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-07 22:11:08,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7203, 4.5358, 4.3127, 2.4072], device='cuda:1'), covar=tensor([0.0548, 0.0687, 0.0706, 0.1642], device='cuda:1'), in_proj_covar=tensor([0.1147, 0.1061, 0.0913, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 22:11:15,545 INFO [train.py:968] (1/2) Epoch 15, batch 27850, giga_loss[loss=0.3697, simple_loss=0.4129, pruned_loss=0.1632, over 27511.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3654, pruned_loss=0.119, over 5648971.36 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1207, over 5702420.49 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.366, pruned_loss=0.1192, over 5650976.06 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:11:20,827 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=666773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:12:00,614 INFO [train.py:968] (1/2) Epoch 15, batch 27900, giga_loss[loss=0.3311, simple_loss=0.3691, pruned_loss=0.1465, over 23573.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5663446.30 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3679, pruned_loss=0.1203, over 5704304.01 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1202, over 5661936.14 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:12:17,027 INFO [optim.py:369] (1/2) 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 22:12:34,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-07 22:12:47,114 INFO [train.py:968] (1/2) Epoch 15, batch 27950, giga_loss[loss=0.3474, simple_loss=0.4035, pruned_loss=0.1457, over 28258.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.369, pruned_loss=0.1203, over 5663734.93 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5711103.02 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3701, pruned_loss=0.121, over 5654990.71 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:13:18,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1861, 0.8007, 1.0596, 1.3706], device='cuda:1'), covar=tensor([0.0839, 0.0379, 0.0341, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 22:13:30,140 INFO [train.py:968] (1/2) Epoch 15, batch 28000, giga_loss[loss=0.3856, simple_loss=0.4125, pruned_loss=0.1794, over 26445.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3695, pruned_loss=0.1205, over 5662899.80 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3674, pruned_loss=0.1196, over 5713683.89 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3706, pruned_loss=0.1214, over 5650045.98 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:13:45,107 INFO [optim.py:369] (1/2) 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 22:14:00,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 22:14:17,792 INFO [train.py:968] (1/2) Epoch 15, batch 28050, giga_loss[loss=0.329, simple_loss=0.3673, pruned_loss=0.1454, over 23540.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1213, over 5654323.71 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1196, over 5715838.71 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.122, over 5641422.97 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:14:55,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 22:14:58,980 INFO [train.py:968] (1/2) Epoch 15, batch 28100, giga_loss[loss=0.3119, simple_loss=0.3779, pruned_loss=0.123, over 28404.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3726, pruned_loss=0.1228, over 5671958.14 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.368, pruned_loss=0.1199, over 5719443.29 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1232, over 5656934.62 frames. ], batch size: 369, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:15:01,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2422, 3.3663, 1.4838, 1.4836], device='cuda:1'), covar=tensor([0.1007, 0.0303, 0.0859, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0531, 0.0357, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 22:15:15,764 INFO [optim.py:369] (1/2) 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 22:15:18,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:15:47,807 INFO [train.py:968] (1/2) Epoch 15, batch 28150, giga_loss[loss=0.3937, simple_loss=0.4196, pruned_loss=0.1839, over 23459.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3739, pruned_loss=0.1238, over 5669650.68 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5722509.51 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3747, pruned_loss=0.1243, over 5654570.25 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:16:01,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667083.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:16:28,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6563, 1.9557, 1.4982, 1.9799], device='cuda:1'), covar=tensor([0.2470, 0.2522, 0.2854, 0.2424], device='cuda:1'), in_proj_covar=tensor([0.1393, 0.1025, 0.1236, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 22:16:35,792 INFO [train.py:968] (1/2) Epoch 15, batch 28200, giga_loss[loss=0.3132, simple_loss=0.3855, pruned_loss=0.1205, over 28979.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3748, pruned_loss=0.1245, over 5675607.49 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1198, over 5726276.72 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.125, over 5659024.69 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:16:52,457 INFO [optim.py:369] (1/2) 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 22:17:04,213 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 28250, giga_loss[loss=0.3002, simple_loss=0.367, pruned_loss=0.1167, over 28753.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3771, pruned_loss=0.127, over 5646624.24 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3682, pruned_loss=0.1204, over 5709004.46 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3773, pruned_loss=0.1269, over 5646725.09 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:18:04,976 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:18:14,922 INFO [train.py:968] (1/2) Epoch 15, batch 28300, giga_loss[loss=0.2885, simple_loss=0.3719, pruned_loss=0.1025, over 28824.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3787, pruned_loss=0.1278, over 5648121.60 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1209, over 5712003.90 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3787, pruned_loss=0.1276, over 5643667.65 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:18:22,524 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:18:26,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667229.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:18:31,791 INFO [optim.py:369] (1/2) 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 22:18:55,627 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667258.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 22:19:07,615 INFO [train.py:968] (1/2) Epoch 15, batch 28350, giga_loss[loss=0.2746, simple_loss=0.3493, pruned_loss=0.09995, over 28947.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3797, pruned_loss=0.1271, over 5651879.09 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1212, over 5710232.60 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3795, pruned_loss=0.1266, over 5649446.92 frames. ], batch size: 145, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:19:22,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3670, 5.1741, 4.9432, 2.5772], device='cuda:1'), covar=tensor([0.0487, 0.0652, 0.0769, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.1151, 0.1064, 0.0919, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 22:19:28,249 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:19:30,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:19:51,282 INFO [train.py:968] (1/2) Epoch 15, batch 28400, giga_loss[loss=0.3281, simple_loss=0.3855, pruned_loss=0.1353, over 28516.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3781, pruned_loss=0.1263, over 5666629.19 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3685, pruned_loss=0.121, over 5716583.83 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3789, pruned_loss=0.1264, over 5656866.67 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:19:55,398 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:20:11,038 INFO [optim.py:369] (1/2) 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 22:20:38,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3708, 1.6074, 1.6181, 1.2133], device='cuda:1'), covar=tensor([0.1603, 0.2345, 0.1292, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0702, 0.0909, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 22:20:43,248 INFO [train.py:968] (1/2) Epoch 15, batch 28450, giga_loss[loss=0.3572, simple_loss=0.399, pruned_loss=0.1578, over 26648.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3769, pruned_loss=0.1262, over 5666316.28 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3688, pruned_loss=0.1211, over 5716188.57 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3774, pruned_loss=0.1263, over 5657769.90 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:20:52,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:21:15,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-07 22:21:32,175 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 28500, giga_loss[loss=0.2927, simple_loss=0.3593, pruned_loss=0.1131, over 29010.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3749, pruned_loss=0.125, over 5680661.55 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1206, over 5721496.63 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3762, pruned_loss=0.1256, over 5667659.26 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:21:59,492 INFO [optim.py:369] (1/2) 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 22:22:31,467 INFO [train.py:968] (1/2) Epoch 15, batch 28550, giga_loss[loss=0.2778, simple_loss=0.3516, pruned_loss=0.1021, over 28859.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.373, pruned_loss=0.1243, over 5668873.84 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1207, over 5713774.61 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1248, over 5664659.56 frames. ], batch size: 199, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:22:36,576 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-07 22:23:13,405 INFO [train.py:968] (1/2) Epoch 15, batch 28600, giga_loss[loss=0.256, simple_loss=0.34, pruned_loss=0.08594, over 28885.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.1251, over 5674383.45 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1209, over 5710134.77 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3746, pruned_loss=0.1255, over 5672359.67 frames. ], batch size: 174, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:23:34,278 INFO [optim.py:369] (1/2) 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 22:23:50,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:23:54,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:23:55,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 22:24:03,216 INFO [train.py:968] (1/2) Epoch 15, batch 28650, giga_loss[loss=0.3293, simple_loss=0.3901, pruned_loss=0.1342, over 28906.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3738, pruned_loss=0.1254, over 5657538.18 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.369, pruned_loss=0.1209, over 5711986.61 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3741, pruned_loss=0.1257, over 5653436.57 frames. ], batch size: 145, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:24:14,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:24:20,405 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:24:48,099 INFO [train.py:968] (1/2) Epoch 15, batch 28700, libri_loss[loss=0.2746, simple_loss=0.3467, pruned_loss=0.1013, over 29506.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.374, pruned_loss=0.1259, over 5660590.77 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3686, pruned_loss=0.1206, over 5718882.30 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3746, pruned_loss=0.1265, over 5649403.06 frames. ], batch size: 81, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:25:06,544 INFO [optim.py:369] (1/2) 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 22:25:08,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5143, 4.3463, 1.6763, 1.7225], device='cuda:1'), covar=tensor([0.0972, 0.0306, 0.0868, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0534, 0.0360, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 22:25:31,187 INFO [train.py:968] (1/2) Epoch 15, batch 28750, giga_loss[loss=0.3273, simple_loss=0.3901, pruned_loss=0.1323, over 28750.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3771, pruned_loss=0.1284, over 5665252.04 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1206, over 5722650.69 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3779, pruned_loss=0.1291, over 5651547.74 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:26:18,593 INFO [train.py:968] (1/2) Epoch 15, batch 28800, giga_loss[loss=0.3008, simple_loss=0.3663, pruned_loss=0.1177, over 28786.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3787, pruned_loss=0.13, over 5650697.22 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3687, pruned_loss=0.1208, over 5718364.06 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3794, pruned_loss=0.1307, over 5642240.71 frames. ], batch size: 112, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:26:26,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:26:28,588 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:26:50,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=667751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:26:54,907 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=667758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:27:05,513 INFO [train.py:968] (1/2) Epoch 15, batch 28850, giga_loss[loss=0.2843, simple_loss=0.352, pruned_loss=0.1083, over 29072.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3783, pruned_loss=0.1303, over 5652577.22 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3687, pruned_loss=0.1208, over 5721197.30 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.379, pruned_loss=0.1309, over 5642589.46 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:27:49,176 INFO [train.py:968] (1/2) Epoch 15, batch 28900, giga_loss[loss=0.3043, simple_loss=0.3679, pruned_loss=0.1204, over 28644.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3782, pruned_loss=0.1302, over 5657283.98 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1214, over 5725294.82 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3782, pruned_loss=0.1304, over 5643891.19 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:28:05,823 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 28950, libri_loss[loss=0.3219, simple_loss=0.3795, pruned_loss=0.1321, over 19963.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3788, pruned_loss=0.1304, over 5636273.31 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.3697, pruned_loss=0.1216, over 5709399.36 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.379, pruned_loss=0.1307, over 5638134.11 frames. ], batch size: 187, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:28:57,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=667894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:28:59,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=667897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:29:22,819 INFO [train.py:968] (1/2) Epoch 15, batch 29000, giga_loss[loss=0.3058, simple_loss=0.3749, pruned_loss=0.1184, over 28855.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3775, pruned_loss=0.1287, over 5643084.16 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3693, pruned_loss=0.1213, over 5707435.09 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3781, pruned_loss=0.1293, over 5644916.11 frames. ], batch size: 227, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:29:23,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=667920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:29:28,294 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:30:04,564 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 22:30:05,484 INFO [train.py:968] (1/2) Epoch 15, batch 29050, giga_loss[loss=0.3041, simple_loss=0.3783, pruned_loss=0.115, over 28971.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3771, pruned_loss=0.1278, over 5651926.25 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3692, pruned_loss=0.121, over 5712320.57 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3781, pruned_loss=0.1289, over 5646878.90 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:30:42,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3044, 1.4184, 1.3117, 1.5597], device='cuda:1'), covar=tensor([0.0768, 0.0335, 0.0323, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 22:30:50,032 INFO [train.py:968] (1/2) Epoch 15, batch 29100, giga_loss[loss=0.3135, simple_loss=0.3741, pruned_loss=0.1264, over 28952.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3788, pruned_loss=0.1292, over 5670164.44 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3689, pruned_loss=0.1208, over 5716344.93 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.38, pruned_loss=0.1304, over 5661485.20 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:30:53,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:31:07,008 INFO [optim.py:369] (1/2) 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 22:31:17,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2385, 3.5459, 1.3459, 1.4441], device='cuda:1'), covar=tensor([0.1137, 0.0497, 0.1018, 0.1519], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0536, 0.0361, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 22:31:27,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 22:31:32,874 INFO [train.py:968] (1/2) Epoch 15, batch 29150, giga_loss[loss=0.2823, simple_loss=0.3593, pruned_loss=0.1026, over 28904.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3786, pruned_loss=0.1291, over 5676669.00 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3687, pruned_loss=0.1207, over 5718938.81 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.38, pruned_loss=0.1304, over 5666439.49 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:31:33,982 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 22:31:58,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5789, 3.3911, 3.2283, 2.0072], device='cuda:1'), covar=tensor([0.0717, 0.0925, 0.0861, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.1151, 0.1066, 0.0919, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 22:32:06,197 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:32:18,505 INFO [train.py:968] (1/2) Epoch 15, batch 29200, giga_loss[loss=0.288, simple_loss=0.3667, pruned_loss=0.1046, over 28881.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3798, pruned_loss=0.1296, over 5664520.98 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3688, pruned_loss=0.1209, over 5710397.89 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3812, pruned_loss=0.1307, over 5662690.40 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:32:40,274 INFO [optim.py:369] (1/2) 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 22:32:51,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-07 22:33:12,059 INFO [train.py:968] (1/2) Epoch 15, batch 29250, giga_loss[loss=0.3354, simple_loss=0.3736, pruned_loss=0.1486, over 23493.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3801, pruned_loss=0.129, over 5652603.51 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3689, pruned_loss=0.121, over 5712150.84 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3812, pruned_loss=0.1299, over 5649013.15 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:33:43,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 22:33:55,164 INFO [train.py:968] (1/2) Epoch 15, batch 29300, giga_loss[loss=0.2902, simple_loss=0.3536, pruned_loss=0.1134, over 28584.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3767, pruned_loss=0.1258, over 5672471.49 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3685, pruned_loss=0.1208, over 5716294.15 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3781, pruned_loss=0.1268, over 5664546.51 frames. ], batch size: 85, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:34:10,804 INFO [optim.py:369] (1/2) 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 22:34:36,494 INFO [train.py:968] (1/2) Epoch 15, batch 29350, libri_loss[loss=0.3262, simple_loss=0.3925, pruned_loss=0.1299, over 29138.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3743, pruned_loss=0.1242, over 5673543.56 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3677, pruned_loss=0.1201, over 5722331.04 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3765, pruned_loss=0.1258, over 5659933.23 frames. ], batch size: 101, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:35:00,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=668295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:35:20,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3333, 1.2612, 1.1629, 1.5280], device='cuda:1'), covar=tensor([0.0677, 0.0329, 0.0318, 0.0732], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-07 22:35:24,334 INFO [train.py:968] (1/2) Epoch 15, batch 29400, giga_loss[loss=0.3121, simple_loss=0.3812, pruned_loss=0.1214, over 28682.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3766, pruned_loss=0.1262, over 5667044.37 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.368, pruned_loss=0.1203, over 5720638.52 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3782, pruned_loss=0.1273, over 5657597.31 frames. ], batch size: 242, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:35:42,179 INFO [optim.py:369] (1/2) 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 22:36:14,486 INFO [train.py:968] (1/2) Epoch 15, batch 29450, giga_loss[loss=0.3258, simple_loss=0.39, pruned_loss=0.1308, over 28843.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3779, pruned_loss=0.1274, over 5656436.78 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3678, pruned_loss=0.1201, over 5714621.27 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3795, pruned_loss=0.1286, over 5653631.18 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:36:41,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=668398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:02,390 INFO [train.py:968] (1/2) Epoch 15, batch 29500, giga_loss[loss=0.2983, simple_loss=0.3678, pruned_loss=0.1144, over 28578.00 frames. ], tot_loss[loss=0.317, simple_loss=0.378, pruned_loss=0.128, over 5655852.80 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.368, pruned_loss=0.1201, over 5708771.84 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3793, pruned_loss=0.1292, over 5657518.74 frames. ], batch size: 71, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:37:16,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0127, 1.1813, 1.1355, 0.9145], device='cuda:1'), covar=tensor([0.1680, 0.2158, 0.1167, 0.1645], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1775, 0.1697, 0.1830], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 22:37:20,818 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=668438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:21,733 INFO [optim.py:369] (1/2) 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 22:37:23,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=668441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:47,201 INFO [train.py:968] (1/2) Epoch 15, batch 29550, giga_loss[loss=0.3079, simple_loss=0.3753, pruned_loss=0.1202, over 28966.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3783, pruned_loss=0.129, over 5644112.97 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3681, pruned_loss=0.1202, over 5703073.23 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3795, pruned_loss=0.1301, over 5649409.83 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 2.0
+2023-03-07 22:37:50,282 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=668470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:37:58,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=668482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:38:05,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:38:31,208 INFO [train.py:968] (1/2) Epoch 15, batch 29600, giga_loss[loss=0.3151, simple_loss=0.3885, pruned_loss=0.1209, over 28639.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1284, over 5658921.89 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3677, pruned_loss=0.1197, over 5709863.27 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3796, pruned_loss=0.1299, over 5655144.09 frames. ], batch size: 307, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:38:45,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5927, 1.6512, 1.2706, 1.1859], device='cuda:1'), covar=tensor([0.0842, 0.0572, 0.0944, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0444, 0.0506, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 22:38:50,948 INFO [optim.py:369] (1/2) 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 22:38:51,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=668541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:38:54,870 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 29650, libri_loss[loss=0.3173, simple_loss=0.3765, pruned_loss=0.1291, over 29556.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3779, pruned_loss=0.1283, over 5647827.64 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1193, over 5704979.49 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3802, pruned_loss=0.1302, over 5647977.83 frames. ], batch size: 89, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:39:20,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=668573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:40:03,541 INFO [train.py:968] (1/2) Epoch 15, batch 29700, giga_loss[loss=0.2867, simple_loss=0.3543, pruned_loss=0.1096, over 28588.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3773, pruned_loss=0.1275, over 5658901.09 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1192, over 5705100.32 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3795, pruned_loss=0.1293, over 5657584.87 frames. ], batch size: 85, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:40:09,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=668625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:40:12,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=668628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:40:25,019 INFO [optim.py:369] (1/2) 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 22:40:42,483 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=668657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:40:43,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6406, 1.9542, 1.4334, 1.9426], device='cuda:1'), covar=tensor([0.2853, 0.2792, 0.3118, 0.2576], device='cuda:1'), in_proj_covar=tensor([0.1399, 0.1024, 0.1236, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 22:40:52,818 INFO [train.py:968] (1/2) Epoch 15, batch 29750, giga_loss[loss=0.328, simple_loss=0.3686, pruned_loss=0.1437, over 23843.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3759, pruned_loss=0.1257, over 5667605.09 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3669, pruned_loss=0.1191, over 5710276.64 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3779, pruned_loss=0.1274, over 5660851.37 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:40:56,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1717, 1.5782, 0.9674, 1.1270], device='cuda:1'), covar=tensor([0.1130, 0.0586, 0.1550, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0444, 0.0506, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 22:41:41,218 INFO [train.py:968] (1/2) Epoch 15, batch 29800, giga_loss[loss=0.2919, simple_loss=0.3678, pruned_loss=0.108, over 28948.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3776, pruned_loss=0.1269, over 5662112.60 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3672, pruned_loss=0.1193, over 5712255.86 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.379, pruned_loss=0.1281, over 5654498.14 frames. ], batch size: 145, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:41:53,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=668731.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:42:00,673 INFO [optim.py:369] (1/2) 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 22:42:29,509 INFO [train.py:968] (1/2) Epoch 15, batch 29850, libri_loss[loss=0.2633, simple_loss=0.3255, pruned_loss=0.1006, over 29626.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1269, over 5660277.14 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3676, pruned_loss=0.1197, over 5710783.22 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.378, pruned_loss=0.1277, over 5654304.49 frames. ], batch size: 69, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:42:44,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3304, 2.5973, 2.4525, 2.0468], device='cuda:1'), covar=tensor([0.1561, 0.1771, 0.1585, 0.1933], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0745, 0.0694, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 22:42:51,924 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 29900, giga_loss[loss=0.2759, simple_loss=0.3443, pruned_loss=0.1038, over 28918.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3756, pruned_loss=0.1259, over 5670864.34 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3678, pruned_loss=0.1197, over 5714228.77 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3765, pruned_loss=0.1268, over 5662034.10 frames. ], batch size: 199, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:43:32,583 INFO [optim.py:369] (1/2) 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 22:43:54,944 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 29950, giga_loss[loss=0.3566, simple_loss=0.3956, pruned_loss=0.1588, over 27567.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.375, pruned_loss=0.1262, over 5657447.94 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.368, pruned_loss=0.1199, over 5709171.82 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3756, pruned_loss=0.1268, over 5653641.05 frames. ], batch size: 472, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:44:05,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0892, 1.1877, 3.4732, 3.0295], device='cuda:1'), covar=tensor([0.1635, 0.2590, 0.0526, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0616, 0.0906, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 22:44:33,250 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-07 22:44:33,729 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 22:44:44,466 INFO [train.py:968] (1/2) Epoch 15, batch 30000, giga_loss[loss=0.2861, simple_loss=0.3555, pruned_loss=0.1084, over 28953.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3713, pruned_loss=0.1244, over 5670166.81 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3685, pruned_loss=0.1205, over 5711602.02 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3716, pruned_loss=0.1246, over 5663359.42 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:44:44,466 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 22:44:51,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8335, 3.6202, 3.4546, 1.7807], device='cuda:1'), covar=tensor([0.0778, 0.0932, 0.0913, 0.2246], device='cuda:1'), in_proj_covar=tensor([0.1155, 0.1071, 0.0921, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 22:44:52,857 INFO [train.py:1012] (1/2) Epoch 15, validation: loss=0.2104, simple_loss=0.3176, pruned_loss=0.05162, over 944034.00 frames. 
+2023-03-07 22:44:52,857 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 22:45:09,514 INFO [optim.py:369] (1/2) 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 22:45:22,306 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-07 22:45:33,012 INFO [train.py:968] (1/2) Epoch 15, batch 30050, giga_loss[loss=0.3009, simple_loss=0.3627, pruned_loss=0.1196, over 28595.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3702, pruned_loss=0.1244, over 5687798.82 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3687, pruned_loss=0.1207, over 5715532.16 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3702, pruned_loss=0.1244, over 5677762.27 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:46:14,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:46:17,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:46:19,910 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8927, 2.3987, 2.0026, 1.6867], device='cuda:1'), covar=tensor([0.2773, 0.1926, 0.2045, 0.2515], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1773, 0.1692, 0.1829], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 22:46:22,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2570, 1.4541, 1.4800, 1.2796], device='cuda:1'), covar=tensor([0.1629, 0.1617, 0.2029, 0.1735], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0744, 0.0692, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 22:46:23,008 INFO [train.py:968] (1/2) Epoch 15, batch 30100, giga_loss[loss=0.3108, simple_loss=0.3718, pruned_loss=0.1249, over 28961.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3692, pruned_loss=0.124, over 5694841.04 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3685, pruned_loss=0.1206, over 5715519.50 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3695, pruned_loss=0.1242, over 5686891.36 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:46:47,993 INFO [optim.py:369] (1/2) 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 22:46:48,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:47:03,779 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=669059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:47:12,639 INFO [train.py:968] (1/2) Epoch 15, batch 30150, giga_loss[loss=0.2696, simple_loss=0.3488, pruned_loss=0.09522, over 28273.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3687, pruned_loss=0.1226, over 5690218.84 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1206, over 5717445.49 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3689, pruned_loss=0.1227, over 5681782.32 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:47:31,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=669086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:47:53,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=669106.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:48:08,841 INFO [train.py:968] (1/2) Epoch 15, batch 30200, giga_loss[loss=0.2958, simple_loss=0.3703, pruned_loss=0.1107, over 28689.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3665, pruned_loss=0.1188, over 5686227.31 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3683, pruned_loss=0.1205, over 5719155.26 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3669, pruned_loss=0.119, over 5677941.81 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:48:09,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2187, 1.2430, 1.1244, 0.9293], device='cuda:1'), covar=tensor([0.0864, 0.0505, 0.1029, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0443, 0.0504, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 22:48:26,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3487, 1.9373, 1.3918, 0.7309], device='cuda:1'), covar=tensor([0.2691, 0.1572, 0.2028, 0.3595], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1546, 0.1528, 0.1339], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 22:48:32,717 INFO [optim.py:369] (1/2) 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 22:49:01,306 INFO [train.py:968] (1/2) Epoch 15, batch 30250, giga_loss[loss=0.2441, simple_loss=0.3258, pruned_loss=0.08119, over 28722.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3636, pruned_loss=0.1156, over 5671484.29 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3681, pruned_loss=0.1205, over 5720434.32 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.3639, pruned_loss=0.1158, over 5663224.31 frames. ], batch size: 119, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:49:03,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=669170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:49:37,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-07 22:49:50,769 INFO [train.py:968] (1/2) Epoch 15, batch 30300, libri_loss[loss=0.3239, simple_loss=0.3849, pruned_loss=0.1314, over 25782.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3604, pruned_loss=0.1129, over 5662936.83 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3683, pruned_loss=0.121, over 5724199.96 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3602, pruned_loss=0.1123, over 5651050.02 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:50:09,555 INFO [optim.py:369] (1/2) 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 22:50:20,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669249.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:50:22,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669252.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:50:37,470 INFO [train.py:968] (1/2) Epoch 15, batch 30350, giga_loss[loss=0.2676, simple_loss=0.3473, pruned_loss=0.09395, over 28542.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3574, pruned_loss=0.1101, over 5658457.93 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3684, pruned_loss=0.1214, over 5718787.67 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.357, pruned_loss=0.109, over 5652035.61 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:50:50,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669281.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 22:51:20,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:51:24,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:51:26,342 INFO [train.py:968] (1/2) Epoch 15, batch 30400, giga_loss[loss=0.2352, simple_loss=0.3286, pruned_loss=0.07089, over 29012.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3554, pruned_loss=0.1069, over 5637914.77 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3677, pruned_loss=0.1212, over 5704023.03 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3555, pruned_loss=0.1059, over 5644438.81 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:51:29,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9905, 4.8348, 4.5417, 2.3056], device='cuda:1'), covar=tensor([0.0437, 0.0588, 0.0756, 0.2011], device='cuda:1'), in_proj_covar=tensor([0.1147, 0.1064, 0.0914, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 22:51:51,699 INFO [optim.py:369] (1/2) 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 22:51:55,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:52:17,271 INFO [train.py:968] (1/2) Epoch 15, batch 30450, giga_loss[loss=0.2755, simple_loss=0.3482, pruned_loss=0.1014, over 28636.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3551, pruned_loss=0.1063, over 5640662.75 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3669, pruned_loss=0.1208, over 5709724.98 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3556, pruned_loss=0.1055, over 5638813.36 frames. ], batch size: 85, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:53:09,185 INFO [train.py:968] (1/2) Epoch 15, batch 30500, giga_loss[loss=0.2241, simple_loss=0.3089, pruned_loss=0.06961, over 28972.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3531, pruned_loss=0.1048, over 5641276.80 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3667, pruned_loss=0.1208, over 5712589.18 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3535, pruned_loss=0.1039, over 5636303.65 frames. ], batch size: 155, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:53:26,076 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:53:54,577 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 30550, giga_loss[loss=0.2907, simple_loss=0.3618, pruned_loss=0.1097, over 28775.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3507, pruned_loss=0.1029, over 5637794.30 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3668, pruned_loss=0.1208, over 5713341.23 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3509, pruned_loss=0.1022, over 5633023.23 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 22:54:46,752 INFO [train.py:968] (1/2) Epoch 15, batch 30600, giga_loss[loss=0.2403, simple_loss=0.3239, pruned_loss=0.07833, over 28961.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3499, pruned_loss=0.1031, over 5640120.20 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3661, pruned_loss=0.1209, over 5712321.10 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3498, pruned_loss=0.1015, over 5633187.70 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:55:06,559 INFO [optim.py:369] (1/2) 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 22:55:33,045 INFO [train.py:968] (1/2) Epoch 15, batch 30650, giga_loss[loss=0.2548, simple_loss=0.3363, pruned_loss=0.08659, over 28243.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3493, pruned_loss=0.1023, over 5647746.18 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3654, pruned_loss=0.1206, over 5715099.27 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3493, pruned_loss=0.1006, over 5637163.65 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:55:33,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0437, 3.8738, 3.6406, 1.8286], device='cuda:1'), covar=tensor([0.0635, 0.0813, 0.0894, 0.2395], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.1050, 0.0899, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-07 22:55:40,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:55:43,880 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:07,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-07 22:56:09,040 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=669604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:11,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=669607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:13,692 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:25,167 INFO [train.py:968] (1/2) Epoch 15, batch 30700, giga_loss[loss=0.2486, simple_loss=0.3372, pruned_loss=0.08006, over 28817.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3475, pruned_loss=0.1001, over 5649337.69 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3654, pruned_loss=0.1206, over 5716313.48 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3473, pruned_loss=0.0986, over 5639390.89 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:56:41,034 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=669636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 22:56:46,317 INFO [optim.py:369] (1/2) 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 22:57:15,043 INFO [train.py:968] (1/2) Epoch 15, batch 30750, giga_loss[loss=0.2582, simple_loss=0.3479, pruned_loss=0.08423, over 28990.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3451, pruned_loss=0.09782, over 5638166.21 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3652, pruned_loss=0.1206, over 5698772.91 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3449, pruned_loss=0.09626, over 5644495.17 frames. ], batch size: 145, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:57:19,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4751, 1.9063, 1.6775, 1.6645], device='cuda:1'), covar=tensor([0.1666, 0.1938, 0.1702, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0724, 0.0674, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 22:57:25,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3134, 1.5921, 1.6028, 1.4498], device='cuda:1'), covar=tensor([0.1404, 0.1356, 0.1558, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0725, 0.0675, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-07 22:58:02,769 INFO [train.py:968] (1/2) Epoch 15, batch 30800, giga_loss[loss=0.2645, simple_loss=0.3424, pruned_loss=0.0933, over 28892.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3423, pruned_loss=0.09621, over 5639876.08 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3653, pruned_loss=0.1208, over 5703735.98 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3415, pruned_loss=0.09429, over 5639178.95 frames. ], batch size: 199, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:58:08,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-07 22:58:30,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4907, 4.2174, 1.6237, 1.5395], device='cuda:1'), covar=tensor([0.0975, 0.0311, 0.0946, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0531, 0.0359, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 22:58:30,672 INFO [optim.py:369] (1/2) 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 22:58:56,031 INFO [train.py:968] (1/2) Epoch 15, batch 30850, giga_loss[loss=0.2768, simple_loss=0.3465, pruned_loss=0.1035, over 28739.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3403, pruned_loss=0.09551, over 5642964.39 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3652, pruned_loss=0.1207, over 5705775.57 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3396, pruned_loss=0.09384, over 5640203.11 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:59:49,332 INFO [train.py:968] (1/2) Epoch 15, batch 30900, giga_loss[loss=0.2373, simple_loss=0.3139, pruned_loss=0.08031, over 28748.00 frames. ], tot_loss[loss=0.265, simple_loss=0.339, pruned_loss=0.09545, over 5637805.78 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3651, pruned_loss=0.1206, over 5708140.17 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3383, pruned_loss=0.09403, over 5632866.26 frames. ], batch size: 99, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 22:59:54,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6645, 2.3851, 1.5149, 0.7902], device='cuda:1'), covar=tensor([0.5820, 0.3179, 0.3134, 0.5344], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1545, 0.1529, 0.1340], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 23:00:16,013 INFO [optim.py:369] (1/2) 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 23:00:28,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0954, 3.2367, 2.3569, 1.3087], device='cuda:1'), covar=tensor([0.5506, 0.2495, 0.2491, 0.4631], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1546, 0.1529, 0.1341], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 23:00:32,191 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-07 23:00:40,192 INFO [train.py:968] (1/2) Epoch 15, batch 30950, giga_loss[loss=0.2626, simple_loss=0.3464, pruned_loss=0.08942, over 28673.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3406, pruned_loss=0.09685, over 5634324.66 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3648, pruned_loss=0.1207, over 5714577.40 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3395, pruned_loss=0.09493, over 5622392.12 frames. ], batch size: 242, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:01:11,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-07 23:01:32,675 INFO [train.py:968] (1/2) Epoch 15, batch 31000, giga_loss[loss=0.2611, simple_loss=0.3477, pruned_loss=0.08725, over 28428.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3427, pruned_loss=0.09674, over 5634781.49 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3643, pruned_loss=0.1205, over 5706530.91 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3415, pruned_loss=0.09449, over 5629619.25 frames. ], batch size: 369, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:02:00,752 INFO [optim.py:369] (1/2) 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 23:02:31,648 INFO [train.py:968] (1/2) Epoch 15, batch 31050, giga_loss[loss=0.2872, simple_loss=0.3594, pruned_loss=0.1075, over 28134.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3434, pruned_loss=0.09626, over 5650528.74 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3644, pruned_loss=0.1206, over 5707355.33 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3421, pruned_loss=0.09409, over 5644889.15 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:03:24,415 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:03:33,381 INFO [train.py:968] (1/2) Epoch 15, batch 31100, giga_loss[loss=0.2233, simple_loss=0.3113, pruned_loss=0.0677, over 29077.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3425, pruned_loss=0.09582, over 5665443.78 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3641, pruned_loss=0.1206, over 5714253.11 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3411, pruned_loss=0.0933, over 5653084.02 frames. ], batch size: 128, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:04:02,516 INFO [optim.py:369] (1/2) 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 23:04:35,236 INFO [train.py:968] (1/2) Epoch 15, batch 31150, giga_loss[loss=0.2441, simple_loss=0.3275, pruned_loss=0.0804, over 28958.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09473, over 5656576.19 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.364, pruned_loss=0.1207, over 5707714.37 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3397, pruned_loss=0.09221, over 5651837.42 frames. ], batch size: 186, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:05:09,312 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:05:35,843 INFO [train.py:968] (1/2) Epoch 15, batch 31200, giga_loss[loss=0.2636, simple_loss=0.3456, pruned_loss=0.0908, over 28443.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3392, pruned_loss=0.09288, over 5664195.39 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.363, pruned_loss=0.1202, over 5712904.68 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3383, pruned_loss=0.09053, over 5654281.51 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:06:02,545 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 31250, giga_loss[loss=0.2362, simple_loss=0.3102, pruned_loss=0.08114, over 29018.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3376, pruned_loss=0.0929, over 5668318.13 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3628, pruned_loss=0.1201, over 5716024.22 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3364, pruned_loss=0.09024, over 5655959.21 frames. ], batch size: 136, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:07:28,011 INFO [train.py:968] (1/2) Epoch 15, batch 31300, giga_loss[loss=0.2609, simple_loss=0.342, pruned_loss=0.08985, over 28067.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3368, pruned_loss=0.09352, over 5663134.58 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3623, pruned_loss=0.1202, over 5712527.39 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3352, pruned_loss=0.09022, over 5654689.54 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:07:33,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-07 23:07:50,910 INFO [optim.py:369] (1/2) 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 23:08:19,849 INFO [train.py:968] (1/2) Epoch 15, batch 31350, libri_loss[loss=0.2276, simple_loss=0.295, pruned_loss=0.08006, over 29657.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3354, pruned_loss=0.09336, over 5659877.43 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3613, pruned_loss=0.1199, over 5703542.92 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3339, pruned_loss=0.08981, over 5658411.57 frames. ], batch size: 73, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:09:13,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5927, 1.6008, 1.2663, 1.2648], device='cuda:1'), covar=tensor([0.0681, 0.0322, 0.0777, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0438, 0.0503, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 23:09:14,131 INFO [train.py:968] (1/2) Epoch 15, batch 31400, giga_loss[loss=0.1986, simple_loss=0.2768, pruned_loss=0.06017, over 24208.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3371, pruned_loss=0.09379, over 5659241.93 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.361, pruned_loss=0.1198, over 5705824.58 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3355, pruned_loss=0.09028, over 5654827.24 frames. ], batch size: 705, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:09:19,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3349, 1.4446, 1.2813, 1.5882], device='cuda:1'), covar=tensor([0.0761, 0.0336, 0.0338, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:1')
+2023-03-07 23:09:42,715 INFO [optim.py:369] (1/2) 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 23:10:13,927 INFO [train.py:968] (1/2) Epoch 15, batch 31450, giga_loss[loss=0.2842, simple_loss=0.3596, pruned_loss=0.1044, over 28130.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3391, pruned_loss=0.09402, over 5666377.47 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3606, pruned_loss=0.1197, over 5709595.37 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.09058, over 5658209.27 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:10:22,653 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:10:36,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:11:17,923 INFO [train.py:968] (1/2) Epoch 15, batch 31500, giga_loss[loss=0.2216, simple_loss=0.3082, pruned_loss=0.06756, over 28648.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3352, pruned_loss=0.09099, over 5662905.32 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3605, pruned_loss=0.1197, over 5700173.98 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3338, pruned_loss=0.08785, over 5664241.91 frames. ], batch size: 307, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:11:49,049 INFO [optim.py:369] (1/2) 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 23:12:22,188 INFO [train.py:968] (1/2) Epoch 15, batch 31550, giga_loss[loss=0.2095, simple_loss=0.2947, pruned_loss=0.06212, over 28826.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3349, pruned_loss=0.09109, over 5671925.57 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3595, pruned_loss=0.119, over 5705904.29 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3335, pruned_loss=0.08786, over 5666016.86 frames. ], batch size: 92, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:12:24,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:12:41,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4278, 1.7339, 1.4947, 1.2432], device='cuda:1'), covar=tensor([0.2333, 0.1801, 0.1581, 0.2066], device='cuda:1'), in_proj_covar=tensor([0.1804, 0.1736, 0.1646, 0.1791], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 23:12:50,444 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:12:50,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-03-07 23:13:22,108 INFO [train.py:968] (1/2) Epoch 15, batch 31600, giga_loss[loss=0.2851, simple_loss=0.3772, pruned_loss=0.09657, over 28936.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3381, pruned_loss=0.09238, over 5667829.18 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3593, pruned_loss=0.1188, over 5707202.23 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3365, pruned_loss=0.08921, over 5660748.92 frames. ], batch size: 284, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:13:32,084 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=670528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:13:37,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=670531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:13:54,493 INFO [optim.py:369] (1/2) 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 23:13:56,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3607, 1.5893, 1.6281, 1.2253], device='cuda:1'), covar=tensor([0.1877, 0.2608, 0.1532, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0688, 0.0906, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 23:14:05,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-07 23:14:15,459 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 31650, giga_loss[loss=0.2599, simple_loss=0.3513, pruned_loss=0.0842, over 28745.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3416, pruned_loss=0.09161, over 5664596.42 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3589, pruned_loss=0.1187, over 5710348.86 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3405, pruned_loss=0.08888, over 5655575.36 frames. ], batch size: 262, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:15:20,083 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=670613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:15:23,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=670616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:15:25,566 INFO [train.py:968] (1/2) Epoch 15, batch 31700, giga_loss[loss=0.2668, simple_loss=0.3367, pruned_loss=0.09842, over 26873.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3437, pruned_loss=0.09189, over 5661253.16 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3584, pruned_loss=0.1185, over 5706734.13 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08921, over 5656175.46 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:15:56,406 INFO [optim.py:369] (1/2) 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 23:15:56,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=670645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:16:22,055 INFO [train.py:968] (1/2) Epoch 15, batch 31750, libri_loss[loss=0.3252, simple_loss=0.378, pruned_loss=0.1362, over 27588.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.343, pruned_loss=0.091, over 5660618.52 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3579, pruned_loss=0.1182, over 5711337.31 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3422, pruned_loss=0.08819, over 5650469.69 frames. ], batch size: 115, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:16:25,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8278, 1.8442, 1.3618, 1.4826], device='cuda:1'), covar=tensor([0.0882, 0.0669, 0.1081, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0439, 0.0505, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 23:16:45,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5043, 1.6163, 1.7959, 1.3439], device='cuda:1'), covar=tensor([0.1876, 0.2447, 0.1451, 0.1736], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0688, 0.0906, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 23:17:17,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:17:19,145 INFO [train.py:968] (1/2) Epoch 15, batch 31800, giga_loss[loss=0.3226, simple_loss=0.3864, pruned_loss=0.1294, over 28054.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3437, pruned_loss=0.09207, over 5666949.64 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3579, pruned_loss=0.1181, over 5713363.27 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3426, pruned_loss=0.08883, over 5655017.50 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:17:52,623 INFO [optim.py:369] (1/2) 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 23:17:58,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:18:21,884 INFO [train.py:968] (1/2) Epoch 15, batch 31850, giga_loss[loss=0.234, simple_loss=0.3175, pruned_loss=0.07522, over 28987.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3432, pruned_loss=0.09315, over 5653918.56 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3576, pruned_loss=0.1181, over 5705265.76 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3422, pruned_loss=0.08994, over 5650900.35 frames. ], batch size: 213, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:19:42,009 INFO [train.py:968] (1/2) Epoch 15, batch 31900, libri_loss[loss=0.345, simple_loss=0.3936, pruned_loss=0.1482, over 28497.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3445, pruned_loss=0.09449, over 5656115.57 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3581, pruned_loss=0.1185, over 5699011.51 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.343, pruned_loss=0.09098, over 5658534.94 frames. ], batch size: 106, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:20:29,566 INFO [optim.py:369] (1/2) 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 23:20:48,068 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=670865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:21:08,891 INFO [train.py:968] (1/2) Epoch 15, batch 31950, giga_loss[loss=0.2661, simple_loss=0.3411, pruned_loss=0.09561, over 28368.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3404, pruned_loss=0.09206, over 5665502.57 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3581, pruned_loss=0.1185, over 5699011.51 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3392, pruned_loss=0.08933, over 5667385.60 frames. ], batch size: 369, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:21:26,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3047, 1.2686, 3.8733, 3.1801], device='cuda:1'), covar=tensor([0.1525, 0.2571, 0.0412, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0691, 0.0606, 0.0887, 0.0799], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 23:21:29,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=670883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:21:42,865 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=670892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:21:47,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=670895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:22:01,066 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 32000, giga_loss[loss=0.2478, simple_loss=0.3299, pruned_loss=0.0829, over 28817.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.337, pruned_loss=0.09025, over 5658611.22 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3576, pruned_loss=0.1184, over 5695538.91 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3359, pruned_loss=0.08742, over 5661985.83 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:22:25,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=670924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:22:48,869 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 32050, giga_loss[loss=0.2585, simple_loss=0.3464, pruned_loss=0.08535, over 28087.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3336, pruned_loss=0.08842, over 5665611.02 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3577, pruned_loss=0.1187, over 5699997.86 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3323, pruned_loss=0.0854, over 5663919.86 frames. ], batch size: 412, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:23:32,409 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-07 23:23:56,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0647, 1.4497, 1.2925, 1.0248], device='cuda:1'), covar=tensor([0.2570, 0.1842, 0.1390, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.1804, 0.1731, 0.1644, 0.1791], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 23:24:16,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-07 23:24:19,699 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:24:22,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:24:34,474 INFO [train.py:968] (1/2) Epoch 15, batch 32100, giga_loss[loss=0.2907, simple_loss=0.3751, pruned_loss=0.1031, over 28322.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3374, pruned_loss=0.09072, over 5669388.84 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3572, pruned_loss=0.1186, over 5703201.56 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3364, pruned_loss=0.0879, over 5664596.03 frames. ], batch size: 368, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:24:59,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:25:04,291 INFO [optim.py:369] (1/2) 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 23:25:04,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2247, 4.0619, 3.8340, 1.8140], device='cuda:1'), covar=tensor([0.0524, 0.0682, 0.0780, 0.2146], device='cuda:1'), in_proj_covar=tensor([0.1108, 0.1022, 0.0879, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-07 23:25:33,021 INFO [train.py:968] (1/2) Epoch 15, batch 32150, giga_loss[loss=0.2544, simple_loss=0.3303, pruned_loss=0.08925, over 28920.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.34, pruned_loss=0.09254, over 5666170.65 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.357, pruned_loss=0.1184, over 5700774.16 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3387, pruned_loss=0.08947, over 5662540.85 frames. ], batch size: 164, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:25:38,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1896, 3.3889, 1.4151, 1.3628], device='cuda:1'), covar=tensor([0.1071, 0.0388, 0.0986, 0.1439], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0526, 0.0359, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 23:26:03,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4946, 1.7677, 1.4525, 1.7191], device='cuda:1'), covar=tensor([0.0734, 0.0293, 0.0321, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:1')
+2023-03-07 23:26:07,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:26:42,306 INFO [train.py:968] (1/2) Epoch 15, batch 32200, giga_loss[loss=0.2419, simple_loss=0.3211, pruned_loss=0.08134, over 28860.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3381, pruned_loss=0.0926, over 5666066.41 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3566, pruned_loss=0.1182, over 5703840.65 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3372, pruned_loss=0.09005, over 5660035.24 frames. ], batch size: 227, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:26:56,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4484, 3.3795, 1.5404, 1.6153], device='cuda:1'), covar=tensor([0.0925, 0.0331, 0.0940, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0524, 0.0357, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 23:27:02,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1966, 1.3021, 1.1542, 0.9142], device='cuda:1'), covar=tensor([0.0905, 0.0482, 0.1018, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0435, 0.0501, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 23:27:14,089 INFO [optim.py:369] (1/2) 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 23:27:34,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-07 23:27:42,691 INFO [train.py:968] (1/2) Epoch 15, batch 32250, giga_loss[loss=0.2618, simple_loss=0.3412, pruned_loss=0.09125, over 28575.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3381, pruned_loss=0.09326, over 5670862.00 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3557, pruned_loss=0.1177, over 5708182.44 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3376, pruned_loss=0.09081, over 5660972.19 frames. ], batch size: 336, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:28:54,210 INFO [train.py:968] (1/2) Epoch 15, batch 32300, giga_loss[loss=0.2869, simple_loss=0.3497, pruned_loss=0.112, over 26942.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3393, pruned_loss=0.09324, over 5672670.11 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3557, pruned_loss=0.1176, over 5711461.26 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3387, pruned_loss=0.09103, over 5661364.73 frames. ], batch size: 555, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:29:01,064 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-07 23:29:10,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:29:13,730 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671233.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:29:16,308 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:29:20,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:29:33,530 INFO [optim.py:369] (1/2) 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 23:29:51,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:30:05,519 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 32350, libri_loss[loss=0.2708, simple_loss=0.3319, pruned_loss=0.1048, over 29551.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3417, pruned_loss=0.0935, over 5672129.83 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3557, pruned_loss=0.1177, over 5706036.78 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.341, pruned_loss=0.09125, over 5666849.22 frames. ], batch size: 78, lr: 2.13e-03, grad_scale: 4.0
+2023-03-07 23:30:17,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6063, 1.7913, 1.4434, 1.8828], device='cuda:1'), covar=tensor([0.2589, 0.2553, 0.2826, 0.2377], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1018, 0.1238, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 23:30:29,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:31:29,208 INFO [train.py:968] (1/2) Epoch 15, batch 32400, giga_loss[loss=0.239, simple_loss=0.3233, pruned_loss=0.07732, over 28611.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3405, pruned_loss=0.0923, over 5673759.10 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3554, pruned_loss=0.1175, over 5709960.01 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.34, pruned_loss=0.09031, over 5665710.72 frames. ], batch size: 307, lr: 2.13e-03, grad_scale: 8.0
+2023-03-07 23:32:10,741 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 32450, giga_loss[loss=0.2586, simple_loss=0.3297, pruned_loss=0.09373, over 28464.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3374, pruned_loss=0.09216, over 5675123.43 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3551, pruned_loss=0.1173, over 5712209.44 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3369, pruned_loss=0.09024, over 5665904.68 frames. ], batch size: 336, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:32:48,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671374.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:32:52,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:32:55,105 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:22,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:25,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:27,789 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:33:44,680 INFO [train.py:968] (1/2) Epoch 15, batch 32500, giga_loss[loss=0.2233, simple_loss=0.3067, pruned_loss=0.06992, over 28950.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3313, pruned_loss=0.08948, over 5679835.33 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3548, pruned_loss=0.1172, over 5714025.35 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3304, pruned_loss=0.08709, over 5669268.25 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:33:50,802 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:33:55,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:34:00,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2177, 1.6012, 1.4904, 1.1690], device='cuda:1'), covar=tensor([0.1406, 0.1965, 0.1204, 0.1583], device='cuda:1'), in_proj_covar=tensor([0.0861, 0.0689, 0.0909, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-07 23:34:06,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:34:24,205 INFO [optim.py:369] (1/2) 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 23:34:36,632 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:34:50,183 INFO [train.py:968] (1/2) Epoch 15, batch 32550, giga_loss[loss=0.2449, simple_loss=0.3278, pruned_loss=0.08104, over 28307.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3312, pruned_loss=0.08992, over 5656014.24 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3548, pruned_loss=0.1173, over 5697869.12 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.33, pruned_loss=0.08723, over 5660012.33 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:35:51,448 INFO [train.py:968] (1/2) Epoch 15, batch 32600, giga_loss[loss=0.2592, simple_loss=0.3398, pruned_loss=0.08936, over 28750.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3326, pruned_loss=0.09102, over 5655670.22 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3541, pruned_loss=0.1169, over 5701736.31 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3318, pruned_loss=0.08876, over 5654655.63 frames. ], batch size: 262, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:36:29,100 INFO [optim.py:369] (1/2) 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 23:36:41,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2524, 1.5085, 1.4916, 1.2486], device='cuda:1'), covar=tensor([0.2864, 0.2159, 0.1710, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.1795, 0.1716, 0.1627, 0.1782], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 23:36:57,528 INFO [train.py:968] (1/2) Epoch 15, batch 32650, giga_loss[loss=0.2338, simple_loss=0.3178, pruned_loss=0.07489, over 28951.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.33, pruned_loss=0.08885, over 5652562.92 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3542, pruned_loss=0.1169, over 5700535.60 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.329, pruned_loss=0.08665, over 5651942.37 frames. ], batch size: 199, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:37:48,076 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 32700, giga_loss[loss=0.2066, simple_loss=0.2955, pruned_loss=0.05885, over 28532.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3286, pruned_loss=0.08689, over 5660538.38 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3541, pruned_loss=0.117, over 5703459.43 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3274, pruned_loss=0.08457, over 5656517.05 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:38:26,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6130, 1.8024, 1.5087, 1.7837], device='cuda:1'), covar=tensor([0.2439, 0.2177, 0.2375, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.1389, 0.1015, 0.1236, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-07 23:38:43,279 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 32750, giga_loss[loss=0.2443, simple_loss=0.3207, pruned_loss=0.08396, over 28121.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3284, pruned_loss=0.08733, over 5664292.47 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.354, pruned_loss=0.1169, over 5705478.80 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3273, pruned_loss=0.08533, over 5658911.93 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:40:26,656 INFO [train.py:968] (1/2) Epoch 15, batch 32800, giga_loss[loss=0.2329, simple_loss=0.3237, pruned_loss=0.07103, over 29018.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3281, pruned_loss=0.08669, over 5642542.09 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3541, pruned_loss=0.117, over 5689864.76 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3267, pruned_loss=0.08448, over 5650845.92 frames. ], batch size: 155, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:41:06,166 INFO [optim.py:369] (1/2) 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 23:41:11,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671751.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:41:16,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671754.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:41:18,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=671756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:41:30,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6346, 1.5831, 1.2824, 1.2296], device='cuda:1'), covar=tensor([0.0677, 0.0429, 0.0821, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0438, 0.0506, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 23:41:35,797 INFO [train.py:968] (1/2) Epoch 15, batch 32850, giga_loss[loss=0.2413, simple_loss=0.3184, pruned_loss=0.08205, over 29012.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3283, pruned_loss=0.08713, over 5633124.24 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3543, pruned_loss=0.1173, over 5676054.93 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3265, pruned_loss=0.08445, over 5650096.80 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:41:56,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671783.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:41:57,925 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:42:40,098 INFO [train.py:968] (1/2) Epoch 15, batch 32900, giga_loss[loss=0.2477, simple_loss=0.3252, pruned_loss=0.08516, over 28602.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3306, pruned_loss=0.08908, over 5643439.28 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3543, pruned_loss=0.1172, over 5677988.43 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3287, pruned_loss=0.08645, over 5654449.02 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:43:00,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:43:21,631 INFO [optim.py:369] (1/2) 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 23:43:44,963 INFO [train.py:968] (1/2) Epoch 15, batch 32950, giga_loss[loss=0.238, simple_loss=0.3052, pruned_loss=0.08535, over 24337.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3298, pruned_loss=0.08843, over 5646104.27 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3546, pruned_loss=0.1174, over 5681851.52 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3273, pruned_loss=0.08554, over 5650475.49 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:44:27,965 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=671899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:44:32,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=671902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:44:52,299 INFO [train.py:968] (1/2) Epoch 15, batch 33000, giga_loss[loss=0.2611, simple_loss=0.3502, pruned_loss=0.08603, over 28905.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3318, pruned_loss=0.08776, over 5650094.31 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3548, pruned_loss=0.1175, over 5679453.27 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3297, pruned_loss=0.08523, over 5655692.69 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:44:52,299 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-07 23:45:01,953 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-07 23:45:18,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=671931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:45:31,656 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671944.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:45:35,522 INFO [optim.py:369] (1/2) 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 23:45:45,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2817, 2.0089, 1.5210, 0.5566], device='cuda:1'), covar=tensor([0.4292, 0.2460, 0.3424, 0.5048], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1554, 0.1527, 0.1341], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-07 23:45:59,173 INFO [train.py:968] (1/2) Epoch 15, batch 33050, giga_loss[loss=0.247, simple_loss=0.3329, pruned_loss=0.08052, over 28844.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3343, pruned_loss=0.08928, over 5655183.26 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3544, pruned_loss=0.1173, over 5685062.82 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3318, pruned_loss=0.08598, over 5653392.81 frames. ], batch size: 112, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:46:22,741 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=671988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:47:06,708 INFO [train.py:968] (1/2) Epoch 15, batch 33100, giga_loss[loss=0.248, simple_loss=0.3302, pruned_loss=0.08292, over 29052.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3346, pruned_loss=0.08941, over 5644929.50 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3544, pruned_loss=0.1174, over 5686588.83 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3325, pruned_loss=0.08639, over 5641864.41 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:47:45,938 INFO [optim.py:369] (1/2) 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 23:48:11,594 INFO [train.py:968] (1/2) Epoch 15, batch 33150, giga_loss[loss=0.2403, simple_loss=0.3224, pruned_loss=0.07912, over 28984.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3348, pruned_loss=0.08956, over 5657841.86 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.354, pruned_loss=0.1171, over 5690078.49 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3331, pruned_loss=0.08688, over 5651681.29 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:49:07,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=672115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:49:12,474 INFO [train.py:968] (1/2) Epoch 15, batch 33200, libri_loss[loss=0.3101, simple_loss=0.3681, pruned_loss=0.1261, over 27478.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3332, pruned_loss=0.08894, over 5659263.83 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3539, pruned_loss=0.1173, over 5691762.50 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3314, pruned_loss=0.08589, over 5652147.59 frames. ], batch size: 115, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:49:43,945 INFO [optim.py:369] (1/2) 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 23:50:02,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:50:13,279 INFO [train.py:968] (1/2) Epoch 15, batch 33250, giga_loss[loss=0.252, simple_loss=0.3306, pruned_loss=0.08672, over 29008.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3314, pruned_loss=0.08791, over 5660133.11 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3536, pruned_loss=0.1171, over 5690368.24 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3294, pruned_loss=0.08468, over 5655018.62 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:50:50,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=672201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:51:00,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:51:12,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7022, 1.8953, 1.4996, 2.2300], device='cuda:1'), covar=tensor([0.2599, 0.2647, 0.2975, 0.2211], device='cuda:1'), in_proj_covar=tensor([0.1381, 0.1009, 0.1228, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-07 23:51:16,216 INFO [train.py:968] (1/2) Epoch 15, batch 33300, libri_loss[loss=0.2762, simple_loss=0.3284, pruned_loss=0.112, over 28625.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.329, pruned_loss=0.08719, over 5657483.55 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3534, pruned_loss=0.117, over 5682720.22 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3271, pruned_loss=0.08397, over 5658748.42 frames. ], batch size: 63, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:51:38,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6259, 4.8400, 1.9033, 1.8887], device='cuda:1'), covar=tensor([0.0942, 0.0236, 0.0863, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0524, 0.0358, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-07 23:51:48,507 INFO [optim.py:369] (1/2) 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 23:52:16,485 INFO [train.py:968] (1/2) Epoch 15, batch 33350, giga_loss[loss=0.2671, simple_loss=0.35, pruned_loss=0.09217, over 28396.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3316, pruned_loss=0.08848, over 5657585.67 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3527, pruned_loss=0.1167, over 5675164.94 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.33, pruned_loss=0.08539, over 5664574.25 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:52:22,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5225, 1.8685, 1.5232, 1.6629], device='cuda:1'), covar=tensor([0.0776, 0.0276, 0.0320, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0063, 0.0058, 0.0098], device='cuda:1')
+2023-03-07 23:53:05,673 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:53:10,935 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:53:27,627 INFO [train.py:968] (1/2) Epoch 15, batch 33400, giga_loss[loss=0.1904, simple_loss=0.2728, pruned_loss=0.05396, over 28455.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3331, pruned_loss=0.08913, over 5658768.53 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3527, pruned_loss=0.1166, over 5678825.84 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3315, pruned_loss=0.08626, over 5660939.26 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:53:27,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672319.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:53:49,286 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:54:08,565 INFO [optim.py:369] (1/2) 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 23:54:13,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:54:14,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:54:25,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:54:37,621 INFO [train.py:968] (1/2) Epoch 15, batch 33450, giga_loss[loss=0.2627, simple_loss=0.3311, pruned_loss=0.09718, over 26642.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3322, pruned_loss=0.08869, over 5662212.57 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3518, pruned_loss=0.1161, over 5683625.74 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3313, pruned_loss=0.08628, over 5659101.70 frames. ], batch size: 555, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:54:54,457 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:55:30,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8083, 2.6579, 2.1485, 1.6978], device='cuda:1'), covar=tensor([0.2502, 0.1199, 0.1562, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.1797, 0.1716, 0.1621, 0.1784], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-07 23:55:32,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-07 23:55:41,876 INFO [train.py:968] (1/2) Epoch 15, batch 33500, giga_loss[loss=0.253, simple_loss=0.3377, pruned_loss=0.0841, over 28943.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3366, pruned_loss=0.09078, over 5671595.77 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3519, pruned_loss=0.1161, over 5688001.53 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3353, pruned_loss=0.08805, over 5664633.91 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:56:12,188 INFO [optim.py:369] (1/2) 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 23:56:29,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672462.0, num_to_drop=1, layers_to_drop={1}
+2023-03-07 23:56:31,868 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672465.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:56:35,775 INFO [train.py:968] (1/2) Epoch 15, batch 33550, giga_loss[loss=0.2385, simple_loss=0.3343, pruned_loss=0.0714, over 28838.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3381, pruned_loss=0.09103, over 5664025.70 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3513, pruned_loss=0.1158, over 5686116.88 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.337, pruned_loss=0.08815, over 5659304.39 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 4.0
+2023-03-07 23:57:00,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:57:04,547 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672494.0, num_to_drop=1, layers_to_drop={0}
+2023-03-07 23:57:18,007 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-07 23:57:23,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:57:27,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:57:42,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0784, 1.0913, 3.7181, 3.0717], device='cuda:1'), covar=tensor([0.1799, 0.2921, 0.0474, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0697, 0.0613, 0.0894, 0.0806], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-07 23:57:42,657 INFO [train.py:968] (1/2) Epoch 15, batch 33600, giga_loss[loss=0.2615, simple_loss=0.3384, pruned_loss=0.09235, over 28979.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3379, pruned_loss=0.09121, over 5664840.71 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3514, pruned_loss=0.1159, over 5693008.13 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3364, pruned_loss=0.08786, over 5654244.10 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:58:11,343 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:58:25,023 INFO [optim.py:369] (1/2) 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 23:58:37,498 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-07 23:58:53,638 INFO [train.py:968] (1/2) Epoch 15, batch 33650, libri_loss[loss=0.2939, simple_loss=0.3467, pruned_loss=0.1205, over 20354.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3358, pruned_loss=0.09032, over 5659063.26 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3511, pruned_loss=0.1159, over 5682511.51 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3344, pruned_loss=0.08673, over 5659084.64 frames. ], batch size: 187, lr: 2.12e-03, grad_scale: 8.0
+2023-03-07 23:59:00,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=672576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-07 23:59:46,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3993, 1.7081, 1.0270, 1.2775], device='cuda:1'), covar=tensor([0.1293, 0.0897, 0.1689, 0.1645], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0437, 0.0505, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:00:00,139 INFO [train.py:968] (1/2) Epoch 15, batch 33700, giga_loss[loss=0.2858, simple_loss=0.3557, pruned_loss=0.108, over 28036.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3352, pruned_loss=0.09053, over 5652068.04 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3513, pruned_loss=0.116, over 5675230.75 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3336, pruned_loss=0.08707, over 5657370.92 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:00:12,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-08 00:00:19,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:00:24,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:00:44,854 INFO [optim.py:369] (1/2) 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 00:01:04,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:01:05,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-08 00:01:07,904 INFO [train.py:968] (1/2) Epoch 15, batch 33750, giga_loss[loss=0.2395, simple_loss=0.3209, pruned_loss=0.07901, over 29006.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3345, pruned_loss=0.09013, over 5649259.32 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3509, pruned_loss=0.1158, over 5676238.49 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3331, pruned_loss=0.08675, over 5651814.71 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:02:05,135 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 00:02:11,280 INFO [train.py:968] (1/2) Epoch 15, batch 33800, giga_loss[loss=0.221, simple_loss=0.3027, pruned_loss=0.06965, over 28881.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3343, pruned_loss=0.09077, over 5648846.56 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.351, pruned_loss=0.1158, over 5669098.06 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3326, pruned_loss=0.08728, over 5656679.00 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:02:13,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=672719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:02:16,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=672722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:02:22,391 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:02:55,895 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=672751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:03:19,428 INFO [train.py:968] (1/2) Epoch 15, batch 33850, giga_loss[loss=0.2963, simple_loss=0.3555, pruned_loss=0.1186, over 26768.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3333, pruned_loss=0.09098, over 5636368.98 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3512, pruned_loss=0.116, over 5670741.01 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3315, pruned_loss=0.08771, over 5640746.97 frames. ], batch size: 555, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:04:19,552 INFO [train.py:968] (1/2) Epoch 15, batch 33900, giga_loss[loss=0.2331, simple_loss=0.3184, pruned_loss=0.07387, over 28422.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3331, pruned_loss=0.08999, over 5641578.11 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.351, pruned_loss=0.116, over 5667622.34 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3314, pruned_loss=0.08671, over 5647390.02 frames. ], batch size: 368, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:04:22,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4240, 1.5892, 1.1255, 1.1588], device='cuda:1'), covar=tensor([0.0884, 0.0488, 0.1054, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0435, 0.0505, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:04:29,044 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-08 00:05:00,566 INFO [optim.py:369] (1/2) 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 00:05:14,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5265, 1.7703, 1.6781, 1.5438], device='cuda:1'), covar=tensor([0.2644, 0.1848, 0.1635, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.1799, 0.1707, 0.1623, 0.1778], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 00:05:22,879 INFO [train.py:968] (1/2) Epoch 15, batch 33950, giga_loss[loss=0.2232, simple_loss=0.3186, pruned_loss=0.06391, over 28968.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3327, pruned_loss=0.08813, over 5650780.68 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3512, pruned_loss=0.1161, over 5661611.34 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3309, pruned_loss=0.085, over 5659988.66 frames. ], batch size: 284, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:05:38,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6195, 1.8280, 1.9373, 1.4229], device='cuda:1'), covar=tensor([0.2012, 0.2545, 0.1631, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.0856, 0.0682, 0.0901, 0.0802], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 00:06:07,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=672905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:06:21,651 INFO [train.py:968] (1/2) Epoch 15, batch 34000, giga_loss[loss=0.2692, simple_loss=0.3562, pruned_loss=0.09103, over 28166.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3342, pruned_loss=0.08708, over 5662139.18 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3513, pruned_loss=0.1163, over 5667563.74 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3323, pruned_loss=0.08372, over 5663994.57 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:07:01,083 INFO [optim.py:369] (1/2) 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 00:07:08,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1425, 2.2716, 1.5967, 1.7663], device='cuda:1'), covar=tensor([0.0852, 0.0580, 0.0962, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0434, 0.0504, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:07:24,494 INFO [train.py:968] (1/2) Epoch 15, batch 34050, libri_loss[loss=0.2971, simple_loss=0.3468, pruned_loss=0.1237, over 29595.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3344, pruned_loss=0.08655, over 5659430.00 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3511, pruned_loss=0.1162, over 5670023.06 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3329, pruned_loss=0.08368, over 5658517.27 frames. ], batch size: 76, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:08:33,336 INFO [train.py:968] (1/2) Epoch 15, batch 34100, giga_loss[loss=0.2631, simple_loss=0.3183, pruned_loss=0.1039, over 24607.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3348, pruned_loss=0.08728, over 5661174.64 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3506, pruned_loss=0.1158, over 5672189.43 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3333, pruned_loss=0.08408, over 5658294.81 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:09:15,943 INFO [optim.py:369] (1/2) 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 00:09:17,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 00:09:38,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2392, 1.5905, 1.5041, 1.4229], device='cuda:1'), covar=tensor([0.1673, 0.1698, 0.1994, 0.1702], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0714, 0.0671, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 00:09:41,139 INFO [train.py:968] (1/2) Epoch 15, batch 34150, giga_loss[loss=0.2596, simple_loss=0.3453, pruned_loss=0.087, over 28672.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3351, pruned_loss=0.08721, over 5656543.52 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3504, pruned_loss=0.1156, over 5658805.33 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3338, pruned_loss=0.08421, over 5664871.74 frames. ], batch size: 262, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:10:21,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=673101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:10:43,918 INFO [train.py:968] (1/2) Epoch 15, batch 34200, libri_loss[loss=0.2938, simple_loss=0.3589, pruned_loss=0.1143, over 29754.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3348, pruned_loss=0.08735, over 5668908.72 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3501, pruned_loss=0.1153, over 5670616.20 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3332, pruned_loss=0.08367, over 5664912.63 frames. ], batch size: 87, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:11:28,773 INFO [optim.py:369] (1/2) 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 00:12:00,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6187, 1.6930, 1.2620, 1.3680], device='cuda:1'), covar=tensor([0.0796, 0.0493, 0.0855, 0.1086], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0436, 0.0504, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:12:00,663 INFO [train.py:968] (1/2) Epoch 15, batch 34250, giga_loss[loss=0.1851, simple_loss=0.2583, pruned_loss=0.05593, over 24601.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3348, pruned_loss=0.08686, over 5649164.86 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3501, pruned_loss=0.1154, over 5655905.23 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3332, pruned_loss=0.08328, over 5658675.58 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:13:09,101 INFO [train.py:968] (1/2) Epoch 15, batch 34300, giga_loss[loss=0.2619, simple_loss=0.3513, pruned_loss=0.08623, over 28677.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3375, pruned_loss=0.0882, over 5644017.53 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3498, pruned_loss=0.1153, over 5649626.79 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3362, pruned_loss=0.08501, over 5656229.12 frames. ], batch size: 262, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:13:42,450 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=673244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:13:46,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=673247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:13:47,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-08 00:13:50,803 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 34350, giga_loss[loss=0.2126, simple_loss=0.307, pruned_loss=0.05912, over 29060.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3392, pruned_loss=0.08849, over 5661929.86 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3498, pruned_loss=0.1152, over 5652875.89 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.338, pruned_loss=0.08557, over 5668757.24 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:14:30,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=673276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:14:35,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=673280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:15:35,635 INFO [train.py:968] (1/2) Epoch 15, batch 34400, giga_loss[loss=0.2848, simple_loss=0.3568, pruned_loss=0.1064, over 28648.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3383, pruned_loss=0.0885, over 5674664.55 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3498, pruned_loss=0.1152, over 5655301.46 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3373, pruned_loss=0.08599, over 5677993.37 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:15:53,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2494, 1.5625, 1.5263, 1.0960], device='cuda:1'), covar=tensor([0.1434, 0.2347, 0.1249, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.0855, 0.0682, 0.0900, 0.0804], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 00:16:16,509 INFO [optim.py:369] (1/2) 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 00:16:42,657 INFO [train.py:968] (1/2) Epoch 15, batch 34450, giga_loss[loss=0.2302, simple_loss=0.3173, pruned_loss=0.07151, over 28946.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3367, pruned_loss=0.08867, over 5677854.71 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.35, pruned_loss=0.1153, over 5661809.53 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3351, pruned_loss=0.08526, over 5675277.10 frames. ], batch size: 93, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:17:50,869 INFO [train.py:968] (1/2) Epoch 15, batch 34500, giga_loss[loss=0.2149, simple_loss=0.3082, pruned_loss=0.06078, over 28157.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3354, pruned_loss=0.08731, over 5679947.19 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3501, pruned_loss=0.1151, over 5657544.86 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3334, pruned_loss=0.08362, over 5682351.42 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:17:57,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=673423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:18:02,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=673426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:18:39,657 INFO [optim.py:369] (1/2) 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 00:18:40,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7435, 4.5977, 4.3349, 1.8897], device='cuda:1'), covar=tensor([0.0476, 0.0584, 0.0707, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1108, 0.1015, 0.0875, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0013, 0.0011, 0.0011], device='cuda:1')
+2023-03-08 00:18:42,936 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 34550, giga_loss[loss=0.2488, simple_loss=0.3294, pruned_loss=0.08417, over 28915.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3351, pruned_loss=0.08747, over 5682224.24 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3504, pruned_loss=0.1155, over 5653626.31 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3329, pruned_loss=0.08358, over 5687615.77 frames. ], batch size: 112, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:20:04,412 INFO [train.py:968] (1/2) Epoch 15, batch 34600, giga_loss[loss=0.273, simple_loss=0.3449, pruned_loss=0.1006, over 28703.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3363, pruned_loss=0.08821, over 5678279.28 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3504, pruned_loss=0.1154, over 5658339.35 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3343, pruned_loss=0.08467, over 5678721.84 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:20:45,979 INFO [optim.py:369] (1/2) 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 00:20:52,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7700, 1.0136, 2.8237, 2.6157], device='cuda:1'), covar=tensor([0.1677, 0.2557, 0.0594, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0694, 0.0607, 0.0887, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:21:07,300 INFO [train.py:968] (1/2) Epoch 15, batch 34650, giga_loss[loss=0.2679, simple_loss=0.3429, pruned_loss=0.09643, over 28902.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3393, pruned_loss=0.09031, over 5667849.80 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3504, pruned_loss=0.1153, over 5658696.00 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3375, pruned_loss=0.08694, over 5668464.73 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:21:13,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3732, 3.4439, 1.4668, 1.5553], device='cuda:1'), covar=tensor([0.1003, 0.0329, 0.0961, 0.1371], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0528, 0.0362, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 00:21:36,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-08 00:22:10,401 INFO [train.py:968] (1/2) Epoch 15, batch 34700, giga_loss[loss=0.2527, simple_loss=0.3266, pruned_loss=0.0894, over 28857.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3357, pruned_loss=0.08859, over 5674586.00 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3498, pruned_loss=0.115, over 5663546.87 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3345, pruned_loss=0.0857, over 5670515.06 frames. ], batch size: 227, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:22:32,153 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-08 00:22:47,542 INFO [optim.py:369] (1/2) 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 00:22:50,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=673654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:23:05,678 INFO [train.py:968] (1/2) Epoch 15, batch 34750, giga_loss[loss=0.2724, simple_loss=0.3491, pruned_loss=0.09784, over 28880.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3339, pruned_loss=0.08844, over 5678716.09 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.349, pruned_loss=0.1144, over 5669944.70 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.333, pruned_loss=0.08552, over 5669656.05 frames. ], batch size: 199, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:23:06,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-08 00:23:37,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2617, 2.6666, 1.3140, 1.4431], device='cuda:1'), covar=tensor([0.0944, 0.0461, 0.0931, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0527, 0.0361, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 00:24:01,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-08 00:24:06,526 INFO [train.py:968] (1/2) Epoch 15, batch 34800, giga_loss[loss=0.2422, simple_loss=0.3118, pruned_loss=0.08629, over 24200.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3358, pruned_loss=0.08996, over 5673415.52 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3489, pruned_loss=0.1143, over 5673619.37 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.335, pruned_loss=0.0874, over 5663022.55 frames. ], batch size: 705, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:24:37,948 INFO [optim.py:369] (1/2) 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 00:24:53,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0807, 2.0528, 1.4570, 1.6753], device='cuda:1'), covar=tensor([0.0826, 0.0664, 0.0989, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0436, 0.0504, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:24:54,696 INFO [train.py:968] (1/2) Epoch 15, batch 34850, giga_loss[loss=0.3431, simple_loss=0.4132, pruned_loss=0.1364, over 28602.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3449, pruned_loss=0.09545, over 5666700.53 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3493, pruned_loss=0.1145, over 5667577.46 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3436, pruned_loss=0.09261, over 5663372.60 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:25:12,587 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=673789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:25:43,539 INFO [train.py:968] (1/2) Epoch 15, batch 34900, giga_loss[loss=0.2894, simple_loss=0.3742, pruned_loss=0.1023, over 28923.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.353, pruned_loss=0.1002, over 5678410.77 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3493, pruned_loss=0.1146, over 5670488.76 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3519, pruned_loss=0.0974, over 5673411.33 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:26:09,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1933, 1.3988, 1.2828, 1.0631], device='cuda:1'), covar=tensor([0.2317, 0.2144, 0.1481, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.1801, 0.1711, 0.1622, 0.1785], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 00:26:12,603 INFO [optim.py:369] (1/2) 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 00:26:26,814 INFO [train.py:968] (1/2) Epoch 15, batch 34950, giga_loss[loss=0.2975, simple_loss=0.3695, pruned_loss=0.1128, over 28899.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3552, pruned_loss=0.1022, over 5665608.60 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3492, pruned_loss=0.1145, over 5656530.97 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3545, pruned_loss=0.09967, over 5674037.38 frames. ], batch size: 213, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:26:41,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6575, 1.7646, 1.8564, 1.4396], device='cuda:1'), covar=tensor([0.1795, 0.2345, 0.1428, 0.1647], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0686, 0.0904, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 00:27:10,290 INFO [train.py:968] (1/2) Epoch 15, batch 35000, giga_loss[loss=0.2672, simple_loss=0.3391, pruned_loss=0.09763, over 28117.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3521, pruned_loss=0.1021, over 5668285.11 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3496, pruned_loss=0.1149, over 5659124.51 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3513, pruned_loss=0.09936, over 5673320.42 frames. ], batch size: 77, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:27:20,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4419, 3.4308, 1.5145, 1.5952], device='cuda:1'), covar=tensor([0.1011, 0.0279, 0.0931, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0526, 0.0360, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 00:27:38,641 INFO [optim.py:369] (1/2) 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 00:27:54,983 INFO [train.py:968] (1/2) Epoch 15, batch 35050, giga_loss[loss=0.2314, simple_loss=0.3078, pruned_loss=0.0775, over 28935.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3444, pruned_loss=0.09849, over 5673090.55 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3494, pruned_loss=0.1146, over 5664256.11 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.344, pruned_loss=0.09634, over 5672801.86 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:28:37,746 INFO [train.py:968] (1/2) Epoch 15, batch 35100, giga_loss[loss=0.2087, simple_loss=0.2773, pruned_loss=0.06999, over 28995.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3365, pruned_loss=0.09483, over 5686839.76 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3496, pruned_loss=0.1146, over 5668722.25 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3359, pruned_loss=0.09276, over 5682850.88 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:28:47,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:29:06,873 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 35150, giga_loss[loss=0.2359, simple_loss=0.309, pruned_loss=0.08142, over 29044.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3288, pruned_loss=0.09125, over 5687801.18 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3496, pruned_loss=0.1144, over 5673405.91 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3279, pruned_loss=0.08934, over 5680786.31 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:29:20,738 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:29:20,934 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-08 00:29:56,086 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:30:05,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9417, 1.8892, 1.4930, 1.5930], device='cuda:1'), covar=tensor([0.0653, 0.0517, 0.0875, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0436, 0.0504, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:30:05,677 INFO [train.py:968] (1/2) Epoch 15, batch 35200, libri_loss[loss=0.3264, simple_loss=0.3862, pruned_loss=0.1333, over 29113.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3223, pruned_loss=0.08819, over 5675567.91 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3502, pruned_loss=0.1148, over 5664521.85 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3204, pruned_loss=0.08584, over 5679098.34 frames. ], batch size: 101, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:30:09,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1552, 1.2853, 3.6349, 2.9931], device='cuda:1'), covar=tensor([0.1665, 0.2734, 0.0448, 0.0937], device='cuda:1'), in_proj_covar=tensor([0.0696, 0.0611, 0.0895, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 00:30:34,979 INFO [optim.py:369] (1/2) 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 00:30:44,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:30:48,186 INFO [train.py:968] (1/2) Epoch 15, batch 35250, giga_loss[loss=0.2199, simple_loss=0.2967, pruned_loss=0.07156, over 28982.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3185, pruned_loss=0.08637, over 5690775.66 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3503, pruned_loss=0.1146, over 5667863.48 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3166, pruned_loss=0.08429, over 5690850.33 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:30:50,830 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:30:56,558 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:31:21,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:31:22,276 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1711, 3.9810, 3.7579, 1.8654], device='cuda:1'), covar=tensor([0.0585, 0.0760, 0.0706, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.1110, 0.1025, 0.0878, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-08 00:31:31,544 INFO [train.py:968] (1/2) Epoch 15, batch 35300, libri_loss[loss=0.2774, simple_loss=0.3463, pruned_loss=0.1042, over 29499.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3167, pruned_loss=0.08587, over 5695984.14 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3509, pruned_loss=0.1147, over 5677183.11 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3133, pruned_loss=0.08297, over 5688050.50 frames. ], batch size: 84, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:32:00,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-08 00:32:00,401 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 35350, giga_loss[loss=0.2657, simple_loss=0.3223, pruned_loss=0.1045, over 26611.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3153, pruned_loss=0.08554, over 5693949.05 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3514, pruned_loss=0.1148, over 5683954.23 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3109, pruned_loss=0.08217, over 5681657.22 frames. ], batch size: 555, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:32:27,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:32:51,782 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:32:53,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:33:00,918 INFO [train.py:968] (1/2) Epoch 15, batch 35400, giga_loss[loss=0.208, simple_loss=0.2778, pruned_loss=0.06911, over 28885.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3108, pruned_loss=0.0833, over 5685546.30 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3513, pruned_loss=0.1146, over 5683146.00 frames. ], giga_tot_loss[loss=0.2338, simple_loss=0.3069, pruned_loss=0.08037, over 5676187.45 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:33:19,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:33:19,502 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:33:43,722 INFO [train.py:968] (1/2) Epoch 15, batch 35450, giga_loss[loss=0.2106, simple_loss=0.2891, pruned_loss=0.06607, over 29044.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3082, pruned_loss=0.08187, over 5693442.69 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3512, pruned_loss=0.1144, over 5687931.67 frames. ], giga_tot_loss[loss=0.231, simple_loss=0.3041, pruned_loss=0.07898, over 5681872.73 frames. ], batch size: 128, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:33:56,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 00:34:27,862 INFO [train.py:968] (1/2) Epoch 15, batch 35500, giga_loss[loss=0.1983, simple_loss=0.2685, pruned_loss=0.06406, over 28672.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3058, pruned_loss=0.081, over 5691793.40 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3512, pruned_loss=0.1144, over 5690215.05 frames. ], giga_tot_loss[loss=0.2296, simple_loss=0.3022, pruned_loss=0.07845, over 5680708.98 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:34:50,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3706, 1.7145, 1.5446, 1.3086], device='cuda:1'), covar=tensor([0.3249, 0.2256, 0.2395, 0.2721], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1730, 0.1642, 0.1804], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 00:34:51,935 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:34:56,615 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 35550, giga_loss[loss=0.2082, simple_loss=0.2752, pruned_loss=0.07064, over 28564.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.305, pruned_loss=0.0809, over 5692630.27 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3517, pruned_loss=0.1145, over 5692928.12 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.3, pruned_loss=0.07754, over 5681382.32 frames. ], batch size: 78, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:35:19,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:35:46,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6377, 4.4775, 4.1976, 1.8344], device='cuda:1'), covar=tensor([0.0474, 0.0649, 0.0666, 0.2069], device='cuda:1'), in_proj_covar=tensor([0.1108, 0.1024, 0.0877, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-08 00:35:54,320 INFO [train.py:968] (1/2) Epoch 15, batch 35600, giga_loss[loss=0.1941, simple_loss=0.27, pruned_loss=0.05904, over 28668.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3035, pruned_loss=0.08083, over 5690122.83 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3523, pruned_loss=0.1149, over 5698642.18 frames. ], giga_tot_loss[loss=0.2254, simple_loss=0.2974, pruned_loss=0.07666, over 5675895.08 frames. ], batch size: 242, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:36:21,673 INFO [optim.py:369] (1/2) 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 00:36:28,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:36:37,283 INFO [train.py:968] (1/2) Epoch 15, batch 35650, giga_loss[loss=0.24, simple_loss=0.3138, pruned_loss=0.08309, over 27902.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3052, pruned_loss=0.08208, over 5676239.85 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3531, pruned_loss=0.1152, over 5689772.00 frames. ], giga_tot_loss[loss=0.2265, simple_loss=0.2981, pruned_loss=0.07746, over 5671725.11 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:36:55,608 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:36:58,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:37:22,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5791, 1.5872, 1.7834, 1.4069], device='cuda:1'), covar=tensor([0.1330, 0.1964, 0.1093, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0693, 0.0912, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 00:37:23,234 INFO [train.py:968] (1/2) Epoch 15, batch 35700, giga_loss[loss=0.3208, simple_loss=0.3913, pruned_loss=0.1252, over 28632.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3163, pruned_loss=0.08753, over 5682179.38 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3536, pruned_loss=0.1156, over 5686852.42 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3091, pruned_loss=0.08269, over 5680157.81 frames. ], batch size: 336, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:37:26,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:37:28,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:37:31,000 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:38:01,524 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:05,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:13,651 INFO [train.py:968] (1/2) Epoch 15, batch 35750, giga_loss[loss=0.2855, simple_loss=0.3557, pruned_loss=0.1076, over 28668.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3293, pruned_loss=0.09434, over 5683479.45 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3538, pruned_loss=0.1159, over 5688950.10 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3231, pruned_loss=0.09002, over 5679941.98 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:38:41,604 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:43,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:43,632 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=674714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:38:55,971 INFO [train.py:968] (1/2) Epoch 15, batch 35800, giga_loss[loss=0.3002, simple_loss=0.3751, pruned_loss=0.1127, over 28874.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.338, pruned_loss=0.09839, over 5688046.30 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.354, pruned_loss=0.1158, over 5694591.77 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3322, pruned_loss=0.09437, over 5679799.51 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:39:10,361 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:39:14,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4150, 2.0420, 1.4942, 0.5509], device='cuda:1'), covar=tensor([0.4987, 0.2665, 0.3709, 0.5550], device='cuda:1'), in_proj_covar=tensor([0.1635, 0.1566, 0.1540, 0.1335], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 00:39:27,395 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 35850, giga_loss[loss=0.3073, simple_loss=0.3752, pruned_loss=0.1197, over 28784.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3442, pruned_loss=0.1007, over 5677155.49 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3542, pruned_loss=0.1159, over 5687568.56 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3394, pruned_loss=0.09722, over 5677266.33 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:40:14,204 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:40:16,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:40:27,724 INFO [train.py:968] (1/2) Epoch 15, batch 35900, giga_loss[loss=0.2398, simple_loss=0.3288, pruned_loss=0.07534, over 29041.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3461, pruned_loss=0.1006, over 5675294.98 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3547, pruned_loss=0.116, over 5694008.09 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3414, pruned_loss=0.09711, over 5669211.88 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:40:29,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674821.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:40:43,166 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:40:51,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4121, 1.5899, 1.3939, 1.5961], device='cuda:1'), covar=tensor([0.0791, 0.0321, 0.0328, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:1')
+2023-03-08 00:40:57,755 INFO [optim.py:369] (1/2) 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 00:41:02,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=674857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:41:04,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=674860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:41:14,763 INFO [train.py:968] (1/2) Epoch 15, batch 35950, giga_loss[loss=0.2917, simple_loss=0.3628, pruned_loss=0.1103, over 28781.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3484, pruned_loss=0.1015, over 5664972.76 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3559, pruned_loss=0.1169, over 5681578.34 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3432, pruned_loss=0.0972, over 5671239.46 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:41:21,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=674878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:41:32,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=674889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:42:01,736 INFO [train.py:968] (1/2) Epoch 15, batch 36000, giga_loss[loss=0.2555, simple_loss=0.3347, pruned_loss=0.08809, over 29056.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3499, pruned_loss=0.1024, over 5672324.83 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3558, pruned_loss=0.1169, over 5682820.45 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3459, pruned_loss=0.09899, over 5675939.24 frames. ], batch size: 155, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:42:01,737 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 00:42:10,748 INFO [train.py:1012] (1/2) Epoch 15, validation: loss=0.2079, simple_loss=0.3149, pruned_loss=0.0505, over 944034.00 frames. 
+2023-03-08 00:42:10,749 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 00:42:42,101 INFO [optim.py:369] (1/2) 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 00:42:56,186 INFO [train.py:968] (1/2) Epoch 15, batch 36050, giga_loss[loss=0.251, simple_loss=0.3317, pruned_loss=0.08508, over 28497.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3525, pruned_loss=0.1043, over 5667014.51 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3558, pruned_loss=0.1168, over 5676385.11 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3493, pruned_loss=0.1016, over 5674899.06 frames. ], batch size: 78, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:43:40,060 INFO [train.py:968] (1/2) Epoch 15, batch 36100, giga_loss[loss=0.2832, simple_loss=0.3723, pruned_loss=0.09699, over 28945.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3555, pruned_loss=0.1056, over 5681135.67 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3563, pruned_loss=0.1168, over 5681677.57 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3525, pruned_loss=0.1031, over 5682355.23 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:43:50,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 00:44:07,333 INFO [optim.py:369] (1/2) 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 00:44:17,600 INFO [train.py:968] (1/2) Epoch 15, batch 36150, giga_loss[loss=0.2838, simple_loss=0.3686, pruned_loss=0.09954, over 28608.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3586, pruned_loss=0.1066, over 5687688.54 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3569, pruned_loss=0.117, over 5674720.72 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3555, pruned_loss=0.1039, over 5694553.60 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:44:27,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:45:03,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4031, 1.5850, 1.5613, 1.4107], device='cuda:1'), covar=tensor([0.1639, 0.1849, 0.2100, 0.1807], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0732, 0.0686, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 00:45:05,195 INFO [train.py:968] (1/2) Epoch 15, batch 36200, giga_loss[loss=0.2673, simple_loss=0.3511, pruned_loss=0.0918, over 28928.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3594, pruned_loss=0.1063, over 5686043.13 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3568, pruned_loss=0.117, over 5675904.11 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3572, pruned_loss=0.1042, over 5690395.97 frames. ], batch size: 112, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:45:14,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4428, 1.7676, 1.3023, 1.9324], device='cuda:1'), covar=tensor([0.2662, 0.2616, 0.2983, 0.2327], device='cuda:1'), in_proj_covar=tensor([0.1386, 0.1016, 0.1231, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-08 00:45:34,198 INFO [optim.py:369] (1/2) 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 00:45:43,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:45:47,476 INFO [train.py:968] (1/2) Epoch 15, batch 36250, giga_loss[loss=0.2619, simple_loss=0.3463, pruned_loss=0.08873, over 28774.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3598, pruned_loss=0.1054, over 5684913.50 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3573, pruned_loss=0.1172, over 5671522.24 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3576, pruned_loss=0.1034, over 5692754.09 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:46:10,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675196.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:46:12,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4855, 1.5681, 1.5070, 1.3086], device='cuda:1'), covar=tensor([0.2559, 0.2253, 0.1807, 0.2345], device='cuda:1'), in_proj_covar=tensor([0.1821, 0.1739, 0.1654, 0.1815], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 00:46:28,461 INFO [train.py:968] (1/2) Epoch 15, batch 36300, giga_loss[loss=0.3026, simple_loss=0.3805, pruned_loss=0.1124, over 28728.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3593, pruned_loss=0.1043, over 5682653.46 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3574, pruned_loss=0.1172, over 5667174.28 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3574, pruned_loss=0.1023, over 5693685.85 frames. ], batch size: 262, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:46:31,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:46:34,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:46:57,672 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:46:58,991 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:47:03,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7716, 1.9036, 2.0149, 1.5915], device='cuda:1'), covar=tensor([0.1858, 0.2288, 0.1417, 0.1658], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0689, 0.0910, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 00:47:11,304 INFO [train.py:968] (1/2) Epoch 15, batch 36350, giga_loss[loss=0.2584, simple_loss=0.3387, pruned_loss=0.08907, over 28843.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3573, pruned_loss=0.1025, over 5683214.68 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.358, pruned_loss=0.1175, over 5665648.89 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3553, pruned_loss=0.1003, over 5693923.92 frames. ], batch size: 199, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:47:52,018 INFO [train.py:968] (1/2) Epoch 15, batch 36400, giga_loss[loss=0.3817, simple_loss=0.4256, pruned_loss=0.1689, over 27941.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3575, pruned_loss=0.103, over 5673294.07 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3583, pruned_loss=0.1177, over 5661424.81 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3556, pruned_loss=0.1006, over 5685580.62 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:48:04,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3212, 2.9541, 1.3865, 1.5105], device='cuda:1'), covar=tensor([0.1003, 0.0292, 0.0898, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0523, 0.0357, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 00:48:09,996 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:48:12,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675342.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:48:17,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 00:48:24,158 INFO [optim.py:369] (1/2) 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 00:48:37,342 INFO [train.py:968] (1/2) Epoch 15, batch 36450, giga_loss[loss=0.3182, simple_loss=0.3853, pruned_loss=0.1255, over 28144.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3599, pruned_loss=0.1065, over 5675844.98 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3587, pruned_loss=0.1177, over 5666561.22 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.358, pruned_loss=0.1043, over 5680990.64 frames. ], batch size: 77, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 00:48:40,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675371.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:49:04,977 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675396.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:49:08,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675399.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:49:22,760 INFO [train.py:968] (1/2) Epoch 15, batch 36500, giga_loss[loss=0.2679, simple_loss=0.3375, pruned_loss=0.09921, over 28830.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3629, pruned_loss=0.1107, over 5683102.49 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3592, pruned_loss=0.1181, over 5669456.51 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.361, pruned_loss=0.1083, over 5685098.43 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:49:30,205 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675428.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:49:56,419 INFO [optim.py:369] (1/2) 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 00:50:06,975 INFO [train.py:968] (1/2) Epoch 15, batch 36550, giga_loss[loss=0.274, simple_loss=0.3474, pruned_loss=0.1003, over 28825.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3624, pruned_loss=0.1114, over 5687977.65 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3594, pruned_loss=0.1181, over 5673232.21 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.3608, pruned_loss=0.1094, over 5686566.69 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:50:42,653 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:50:55,218 INFO [train.py:968] (1/2) Epoch 15, batch 36600, giga_loss[loss=0.2631, simple_loss=0.3404, pruned_loss=0.0929, over 28836.00 frames. ], tot_loss[loss=0.2905, simple_loss=0.36, pruned_loss=0.1104, over 5696376.86 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3591, pruned_loss=0.1177, over 5680246.10 frames. ], giga_tot_loss[loss=0.2885, simple_loss=0.3591, pruned_loss=0.1089, over 5689128.71 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:50:59,340 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:51:11,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:51:14,993 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=675543.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:51:26,964 INFO [optim.py:369] (1/2) 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 00:51:38,330 INFO [train.py:968] (1/2) Epoch 15, batch 36650, giga_loss[loss=0.3321, simple_loss=0.3865, pruned_loss=0.1389, over 28743.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3586, pruned_loss=0.1101, over 5691835.72 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3594, pruned_loss=0.1179, over 5675219.08 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3576, pruned_loss=0.1086, over 5690646.86 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:52:21,190 INFO [train.py:968] (1/2) Epoch 15, batch 36700, libri_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1222, over 29488.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3578, pruned_loss=0.1093, over 5697565.88 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3597, pruned_loss=0.1182, over 5683230.64 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3566, pruned_loss=0.1075, over 5689612.28 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:52:55,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3721, 4.1949, 3.9590, 1.9952], device='cuda:1'), covar=tensor([0.0474, 0.0604, 0.0602, 0.2290], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.1031, 0.0886, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 00:52:55,768 INFO [optim.py:369] (1/2) 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 00:52:57,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5599, 1.4559, 1.6452, 1.2547], device='cuda:1'), covar=tensor([0.1889, 0.3095, 0.1499, 0.1650], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0691, 0.0906, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 00:53:07,026 INFO [train.py:968] (1/2) Epoch 15, batch 36750, giga_loss[loss=0.2875, simple_loss=0.3474, pruned_loss=0.1138, over 27648.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3558, pruned_loss=0.107, over 5698939.45 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3603, pruned_loss=0.1185, over 5680957.42 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3543, pruned_loss=0.1051, over 5695122.16 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:53:18,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3492, 4.1928, 3.9462, 1.8242], device='cuda:1'), covar=tensor([0.0526, 0.0661, 0.0696, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.1115, 0.1029, 0.0884, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-08 00:53:18,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:23,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:51,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:51,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:53:53,574 INFO [train.py:968] (1/2) Epoch 15, batch 36800, giga_loss[loss=0.2443, simple_loss=0.3259, pruned_loss=0.08132, over 28930.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3505, pruned_loss=0.1035, over 5693641.30 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3604, pruned_loss=0.1183, over 5682767.51 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3491, pruned_loss=0.1019, over 5689058.27 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 00:54:13,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2600, 1.6114, 1.3184, 1.3757], device='cuda:1'), covar=tensor([0.2330, 0.2219, 0.2521, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0739, 0.0692, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 00:54:32,439 INFO [optim.py:369] (1/2) 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 00:54:35,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3857, 1.6783, 1.5124, 1.4766], device='cuda:1'), covar=tensor([0.0789, 0.0309, 0.0315, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 00:54:45,478 INFO [train.py:968] (1/2) Epoch 15, batch 36850, giga_loss[loss=0.2347, simple_loss=0.3131, pruned_loss=0.07814, over 28864.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3439, pruned_loss=0.09957, over 5699895.11 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3607, pruned_loss=0.1184, over 5684098.54 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3423, pruned_loss=0.09796, over 5695420.30 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:55:41,086 INFO [train.py:968] (1/2) Epoch 15, batch 36900, giga_loss[loss=0.2521, simple_loss=0.3256, pruned_loss=0.08937, over 29060.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3384, pruned_loss=0.09718, over 5675965.37 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3605, pruned_loss=0.1183, over 5677698.49 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3369, pruned_loss=0.09553, over 5677672.73 frames. ], batch size: 136, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:56:16,341 INFO [optim.py:369] (1/2) 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 00:56:16,655 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=675858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:20,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=675861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:26,536 INFO [train.py:968] (1/2) Epoch 15, batch 36950, giga_loss[loss=0.2375, simple_loss=0.3206, pruned_loss=0.07716, over 28918.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3383, pruned_loss=0.09702, over 5675065.67 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3605, pruned_loss=0.1183, over 5678893.18 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3365, pruned_loss=0.09517, over 5675067.52 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:56:36,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:45,408 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=675890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:56:53,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:57:08,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=675918.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:57:09,454 INFO [train.py:968] (1/2) Epoch 15, batch 37000, giga_loss[loss=0.2564, simple_loss=0.3303, pruned_loss=0.09132, over 28584.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3392, pruned_loss=0.09706, over 5680208.17 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3608, pruned_loss=0.1183, over 5679045.38 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3368, pruned_loss=0.09489, over 5679801.78 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:57:42,490 INFO [optim.py:369] (1/2) 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 00:57:53,314 INFO [train.py:968] (1/2) Epoch 15, batch 37050, giga_loss[loss=0.2396, simple_loss=0.3194, pruned_loss=0.07993, over 28921.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3389, pruned_loss=0.09653, over 5679536.03 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3614, pruned_loss=0.1186, over 5672366.97 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3363, pruned_loss=0.09431, over 5685432.18 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:58:36,908 INFO [train.py:968] (1/2) Epoch 15, batch 37100, giga_loss[loss=0.2354, simple_loss=0.3105, pruned_loss=0.08012, over 28547.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3358, pruned_loss=0.09492, over 5684593.40 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3614, pruned_loss=0.1185, over 5673563.14 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3336, pruned_loss=0.09313, over 5688207.58 frames. ], batch size: 78, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:58:40,859 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:58:43,847 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:58:56,868 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:58:59,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:59:08,123 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:59:09,939 INFO [optim.py:369] (1/2) 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 00:59:13,566 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676061.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 00:59:16,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676064.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 00:59:20,238 INFO [train.py:968] (1/2) Epoch 15, batch 37150, giga_loss[loss=0.243, simple_loss=0.307, pruned_loss=0.0895, over 28557.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3334, pruned_loss=0.09368, over 5696795.31 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3618, pruned_loss=0.1186, over 5675598.05 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3311, pruned_loss=0.09199, over 5698002.23 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 00:59:25,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 00:59:39,540 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676093.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:00:01,591 INFO [train.py:968] (1/2) Epoch 15, batch 37200, giga_loss[loss=0.2445, simple_loss=0.3172, pruned_loss=0.08588, over 28881.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3314, pruned_loss=0.0927, over 5705212.85 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3627, pruned_loss=0.119, over 5677893.84 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3282, pruned_loss=0.09048, over 5704634.08 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:00:32,705 INFO [optim.py:369] (1/2) 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 01:00:44,678 INFO [train.py:968] (1/2) Epoch 15, batch 37250, giga_loss[loss=0.245, simple_loss=0.3144, pruned_loss=0.08776, over 28801.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3296, pruned_loss=0.09218, over 5702374.74 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3631, pruned_loss=0.1192, over 5678269.21 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3265, pruned_loss=0.0901, over 5701912.47 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:01:28,087 INFO [train.py:968] (1/2) Epoch 15, batch 37300, giga_loss[loss=0.1955, simple_loss=0.2777, pruned_loss=0.05671, over 28605.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3268, pruned_loss=0.0906, over 5697636.08 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3638, pruned_loss=0.1195, over 5671778.15 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3233, pruned_loss=0.08834, over 5702679.73 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:01:53,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:02:10,306 INFO [train.py:968] (1/2) Epoch 15, batch 37350, giga_loss[loss=0.2435, simple_loss=0.3161, pruned_loss=0.0854, over 28906.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3233, pruned_loss=0.08879, over 5708716.33 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3641, pruned_loss=0.1197, over 5672951.26 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3202, pruned_loss=0.08667, over 5711761.88 frames. ], batch size: 145, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:02:23,993 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=676283.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:02:53,233 INFO [train.py:968] (1/2) Epoch 15, batch 37400, giga_loss[loss=0.2439, simple_loss=0.3195, pruned_loss=0.08412, over 28907.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3215, pruned_loss=0.08782, over 5717594.07 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3642, pruned_loss=0.1197, over 5675271.97 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3184, pruned_loss=0.08591, over 5718232.55 frames. ], batch size: 112, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:03:24,947 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 37450, giga_loss[loss=0.2288, simple_loss=0.3062, pruned_loss=0.07568, over 28701.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3207, pruned_loss=0.08728, over 5723538.45 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3647, pruned_loss=0.1198, over 5677640.89 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3173, pruned_loss=0.08529, over 5722580.50 frames. ], batch size: 284, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:03:41,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-08 01:04:17,137 INFO [train.py:968] (1/2) Epoch 15, batch 37500, giga_loss[loss=0.2867, simple_loss=0.3605, pruned_loss=0.1064, over 28569.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.324, pruned_loss=0.08961, over 5717497.89 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3655, pruned_loss=0.1201, over 5676826.30 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3195, pruned_loss=0.08694, over 5718827.88 frames. ], batch size: 307, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:04:37,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4893, 1.7553, 1.4294, 1.5548], device='cuda:1'), covar=tensor([0.2372, 0.2307, 0.2522, 0.2181], device='cuda:1'), in_proj_covar=tensor([0.1389, 0.1016, 0.1231, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 01:04:51,998 INFO [optim.py:369] (1/2) 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 01:05:00,319 INFO [train.py:968] (1/2) Epoch 15, batch 37550, libri_loss[loss=0.3008, simple_loss=0.3747, pruned_loss=0.1135, over 29368.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3296, pruned_loss=0.09278, over 5713485.75 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3662, pruned_loss=0.1201, over 5679590.85 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.324, pruned_loss=0.08959, over 5713240.63 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:05:01,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-08 01:05:07,509 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=676476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:05:13,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=676482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:05:44,566 INFO [train.py:968] (1/2) Epoch 15, batch 37600, libri_loss[loss=0.2893, simple_loss=0.3659, pruned_loss=0.1064, over 29389.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3356, pruned_loss=0.0965, over 5703001.19 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3667, pruned_loss=0.1203, over 5677132.14 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3291, pruned_loss=0.09257, over 5706696.13 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 01:06:18,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6767, 1.6350, 1.2031, 1.2431], device='cuda:1'), covar=tensor([0.0694, 0.0475, 0.0905, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0438, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 01:06:23,528 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 37650, giga_loss[loss=0.2972, simple_loss=0.3569, pruned_loss=0.1187, over 28749.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3447, pruned_loss=0.1028, over 5693791.00 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3665, pruned_loss=0.1201, over 5676513.97 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3389, pruned_loss=0.09935, over 5697597.29 frames. ], batch size: 99, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:07:25,213 INFO [train.py:968] (1/2) Epoch 15, batch 37700, giga_loss[loss=0.2701, simple_loss=0.3422, pruned_loss=0.09902, over 28817.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3487, pruned_loss=0.1047, over 5679291.59 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3663, pruned_loss=0.1198, over 5681983.33 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3438, pruned_loss=0.1018, over 5677197.93 frames. ], batch size: 112, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:07:29,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:07:59,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676658.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:07:59,962 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 37750, giga_loss[loss=0.2849, simple_loss=0.3515, pruned_loss=0.1091, over 28778.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3537, pruned_loss=0.107, over 5686441.46 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3666, pruned_loss=0.12, over 5685536.89 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3492, pruned_loss=0.1042, over 5681544.57 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:08:58,804 INFO [train.py:968] (1/2) Epoch 15, batch 37800, giga_loss[loss=0.3186, simple_loss=0.3842, pruned_loss=0.1265, over 28696.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3582, pruned_loss=0.1095, over 5678077.67 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3664, pruned_loss=0.1199, over 5689951.79 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3545, pruned_loss=0.107, over 5670009.65 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:09:02,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2073, 1.4091, 3.7420, 3.0340], device='cuda:1'), covar=tensor([0.1638, 0.2493, 0.0419, 0.0976], device='cuda:1'), in_proj_covar=tensor([0.0695, 0.0607, 0.0888, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 01:09:33,821 INFO [optim.py:369] (1/2) 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 01:09:40,622 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 37850, libri_loss[loss=0.3007, simple_loss=0.3619, pruned_loss=0.1198, over 29556.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3599, pruned_loss=0.1099, over 5682701.84 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3664, pruned_loss=0.1198, over 5695279.57 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3568, pruned_loss=0.1078, over 5670921.46 frames. ], batch size: 75, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:09:42,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:05,652 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:06,281 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1925, 2.0592, 1.6958, 1.4371], device='cuda:1'), covar=tensor([0.0874, 0.0276, 0.0276, 0.1076], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 01:10:07,560 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676801.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:10:09,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676804.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:10:23,730 INFO [train.py:968] (1/2) Epoch 15, batch 37900, giga_loss[loss=0.2599, simple_loss=0.3396, pruned_loss=0.09015, over 28805.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3548, pruned_loss=0.1063, over 5680337.02 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3661, pruned_loss=0.1198, over 5690278.96 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3524, pruned_loss=0.1044, over 5676081.92 frames. ], batch size: 119, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:10:33,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=676833.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:10:45,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9170, 1.2900, 1.0950, 0.1965], device='cuda:1'), covar=tensor([0.3822, 0.2814, 0.4316, 0.5739], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1559, 0.1538, 0.1332], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 01:10:52,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:56,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=676857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:10:57,703 INFO [optim.py:369] (1/2) 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 01:11:09,187 INFO [train.py:968] (1/2) Epoch 15, batch 37950, giga_loss[loss=0.3223, simple_loss=0.3866, pruned_loss=0.129, over 27649.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3534, pruned_loss=0.1045, over 5690234.44 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3662, pruned_loss=0.1197, over 5695579.29 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3511, pruned_loss=0.1026, over 5681960.26 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:11:18,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4881, 1.7218, 1.4519, 1.4682], device='cuda:1'), covar=tensor([0.2014, 0.1914, 0.1956, 0.1742], device='cuda:1'), in_proj_covar=tensor([0.1389, 0.1016, 0.1232, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 01:11:51,876 INFO [train.py:968] (1/2) Epoch 15, batch 38000, giga_loss[loss=0.281, simple_loss=0.3534, pruned_loss=0.1043, over 27882.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3526, pruned_loss=0.1036, over 5688542.80 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3663, pruned_loss=0.1198, over 5698789.42 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3504, pruned_loss=0.1016, over 5678650.75 frames. ], batch size: 412, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 01:12:29,162 INFO [optim.py:369] (1/2) 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 01:12:37,259 INFO [train.py:968] (1/2) Epoch 15, batch 38050, giga_loss[loss=0.2628, simple_loss=0.3485, pruned_loss=0.08852, over 28978.00 frames. ], tot_loss[loss=0.281, simple_loss=0.354, pruned_loss=0.104, over 5664379.60 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3665, pruned_loss=0.12, over 5671696.55 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3518, pruned_loss=0.1021, over 5680002.13 frames. ], batch size: 164, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:12:59,568 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=676994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:01,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=676997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:03,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:05,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=677003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:20,727 INFO [train.py:968] (1/2) Epoch 15, batch 38100, giga_loss[loss=0.3272, simple_loss=0.3867, pruned_loss=0.1339, over 27570.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3563, pruned_loss=0.1057, over 5671988.06 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3667, pruned_loss=0.12, over 5674356.61 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3541, pruned_loss=0.1036, over 5682306.91 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:13:26,347 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:31,389 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:52,869 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:13:57,840 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 38150, giga_loss[loss=0.2709, simple_loss=0.348, pruned_loss=0.09687, over 28192.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3576, pruned_loss=0.107, over 5674714.49 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3663, pruned_loss=0.1199, over 5671236.70 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3558, pruned_loss=0.105, over 5686266.28 frames. ], batch size: 77, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:14:44,877 INFO [train.py:968] (1/2) Epoch 15, batch 38200, libri_loss[loss=0.2583, simple_loss=0.3246, pruned_loss=0.096, over 29328.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3584, pruned_loss=0.1076, over 5679138.39 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3664, pruned_loss=0.1199, over 5668924.98 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3567, pruned_loss=0.1056, over 5691143.31 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:15:20,033 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:15:21,612 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-08 01:15:24,366 INFO [optim.py:369] (1/2) 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 01:15:30,982 INFO [train.py:968] (1/2) Epoch 15, batch 38250, giga_loss[loss=0.2543, simple_loss=0.335, pruned_loss=0.08677, over 28997.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3605, pruned_loss=0.1097, over 5680438.83 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3665, pruned_loss=0.1197, over 5672804.98 frames. ], giga_tot_loss[loss=0.2875, simple_loss=0.3588, pruned_loss=0.1081, over 5686356.63 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:16:12,216 INFO [train.py:968] (1/2) Epoch 15, batch 38300, giga_loss[loss=0.3152, simple_loss=0.382, pruned_loss=0.1242, over 28999.00 frames. ], tot_loss[loss=0.2895, simple_loss=0.3603, pruned_loss=0.1093, over 5694059.06 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3667, pruned_loss=0.1198, over 5677435.03 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3587, pruned_loss=0.1078, over 5694804.94 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:16:25,064 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.00 vs. limit=5.0
+2023-03-08 01:16:47,835 INFO [optim.py:369] (1/2) 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 01:16:55,519 INFO [train.py:968] (1/2) Epoch 15, batch 38350, giga_loss[loss=0.2894, simple_loss=0.3664, pruned_loss=0.1062, over 28752.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3585, pruned_loss=0.107, over 5700522.67 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3669, pruned_loss=0.12, over 5680891.14 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.357, pruned_loss=0.1054, over 5698099.96 frames. ], batch size: 92, lr: 2.12e-03, grad_scale: 2.0
+2023-03-08 01:17:38,274 INFO [train.py:968] (1/2) Epoch 15, batch 38400, giga_loss[loss=0.277, simple_loss=0.3524, pruned_loss=0.1008, over 28964.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3585, pruned_loss=0.1059, over 5705939.41 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3668, pruned_loss=0.1197, over 5687236.57 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3571, pruned_loss=0.1045, over 5698794.79 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:18:00,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4392, 2.0530, 1.4528, 0.5975], device='cuda:1'), covar=tensor([0.5009, 0.2359, 0.3450, 0.5205], device='cuda:1'), in_proj_covar=tensor([0.1613, 0.1537, 0.1514, 0.1314], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 01:18:01,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5066, 1.6686, 1.4082, 1.6593], device='cuda:1'), covar=tensor([0.0546, 0.0257, 0.0256, 0.0574], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0063, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 01:18:10,861 INFO [optim.py:369] (1/2) 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 01:18:17,289 INFO [train.py:968] (1/2) Epoch 15, batch 38450, giga_loss[loss=0.3143, simple_loss=0.3714, pruned_loss=0.1286, over 27582.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3582, pruned_loss=0.1055, over 5702634.88 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3669, pruned_loss=0.1197, over 5680610.05 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3568, pruned_loss=0.1039, over 5703513.77 frames. ], batch size: 472, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:18:58,502 INFO [train.py:968] (1/2) Epoch 15, batch 38500, giga_loss[loss=0.2422, simple_loss=0.3242, pruned_loss=0.08008, over 28872.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3562, pruned_loss=0.1049, over 5698648.47 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.367, pruned_loss=0.1198, over 5676323.70 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3547, pruned_loss=0.1031, over 5703127.12 frames. ], batch size: 174, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:19:10,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:19:35,365 INFO [optim.py:369] (1/2) 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 01:19:41,584 INFO [train.py:968] (1/2) Epoch 15, batch 38550, giga_loss[loss=0.251, simple_loss=0.3291, pruned_loss=0.08648, over 28938.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3543, pruned_loss=0.104, over 5710278.98 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3669, pruned_loss=0.1198, over 5682459.11 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3527, pruned_loss=0.1021, over 5709009.44 frames. ], batch size: 186, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:19:58,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1346, 3.1263, 1.9788, 1.1365], device='cuda:1'), covar=tensor([0.4802, 0.2522, 0.2786, 0.4623], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1537, 0.1516, 0.1317], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 01:19:58,662 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:20:21,350 INFO [train.py:968] (1/2) Epoch 15, batch 38600, giga_loss[loss=0.2713, simple_loss=0.3479, pruned_loss=0.09734, over 29034.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3515, pruned_loss=0.1023, over 5706047.41 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3668, pruned_loss=0.1197, over 5677889.93 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3499, pruned_loss=0.1004, over 5710464.91 frames. ], batch size: 155, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:20:32,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:20:57,819 INFO [optim.py:369] (1/2) 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 01:20:58,005 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 38650, giga_loss[loss=0.255, simple_loss=0.3316, pruned_loss=0.08917, over 28588.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.353, pruned_loss=0.1037, over 5707650.93 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3664, pruned_loss=0.1194, over 5679046.13 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3519, pruned_loss=0.1023, over 5710495.69 frames. ], batch size: 85, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:21:08,186 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:21:10,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=677577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:21:35,452 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:21:45,208 INFO [train.py:968] (1/2) Epoch 15, batch 38700, giga_loss[loss=0.2696, simple_loss=0.3463, pruned_loss=0.09652, over 28887.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3534, pruned_loss=0.1038, over 5711448.81 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3664, pruned_loss=0.1193, over 5682909.97 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1024, over 5710768.82 frames. ], batch size: 106, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:22:18,651 INFO [optim.py:369] (1/2) 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 01:22:24,257 INFO [train.py:968] (1/2) Epoch 15, batch 38750, giga_loss[loss=0.2699, simple_loss=0.3439, pruned_loss=0.09798, over 28549.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3531, pruned_loss=0.1027, over 5712539.39 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3665, pruned_loss=0.1194, over 5685820.23 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 5709791.77 frames. ], batch size: 71, lr: 2.12e-03, grad_scale: 4.0
+2023-03-08 01:22:28,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=677674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:22:32,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=677677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:22:38,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7087, 1.8579, 1.9403, 1.5049], device='cuda:1'), covar=tensor([0.1966, 0.2467, 0.1532, 0.1753], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0696, 0.0913, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 01:22:49,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4480, 3.5099, 1.4995, 1.6309], device='cuda:1'), covar=tensor([0.0967, 0.0242, 0.0936, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0519, 0.0355, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-08 01:22:52,820 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=677706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:22:53,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5109, 4.2744, 1.7105, 1.6356], device='cuda:1'), covar=tensor([0.0998, 0.0197, 0.0936, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0519, 0.0356, 0.0402], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0022, 0.0027], device='cuda:1')
+2023-03-08 01:22:56,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8384, 2.1362, 1.9230, 1.8234], device='cuda:1'), covar=tensor([0.2075, 0.2362, 0.2320, 0.2352], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0729, 0.0682, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 01:23:04,633 INFO [train.py:968] (1/2) Epoch 15, batch 38800, giga_loss[loss=0.2733, simple_loss=0.3453, pruned_loss=0.1006, over 28620.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.353, pruned_loss=0.1024, over 5707831.73 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3664, pruned_loss=0.1192, over 5681938.84 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3517, pruned_loss=0.1008, over 5709593.66 frames. ], batch size: 78, lr: 2.12e-03, grad_scale: 8.0
+2023-03-08 01:23:40,127 INFO [optim.py:369] (1/2) 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 01:23:44,456 INFO [train.py:968] (1/2) Epoch 15, batch 38850, giga_loss[loss=0.3014, simple_loss=0.3661, pruned_loss=0.1183, over 28660.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3525, pruned_loss=0.1025, over 5711006.17 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3665, pruned_loss=0.1193, over 5686446.93 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.351, pruned_loss=0.1007, over 5708884.12 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:23:50,869 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:24:08,401 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:24:24,937 INFO [train.py:968] (1/2) Epoch 15, batch 38900, giga_loss[loss=0.2803, simple_loss=0.3473, pruned_loss=0.1066, over 28922.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.35, pruned_loss=0.1012, over 5711288.44 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3666, pruned_loss=0.1193, over 5692237.18 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3485, pruned_loss=0.0995, over 5705176.24 frames. ], batch size: 106, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:25:00,569 INFO [optim.py:369] (1/2) 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 01:25:02,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:25:03,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677865.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:25:06,032 INFO [train.py:968] (1/2) Epoch 15, batch 38950, giga_loss[loss=0.2152, simple_loss=0.2977, pruned_loss=0.06634, over 28439.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3484, pruned_loss=0.1011, over 5714721.34 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3671, pruned_loss=0.1196, over 5696722.05 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3464, pruned_loss=0.09911, over 5706246.07 frames. ], batch size: 60, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:25:14,360 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=677878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:25:47,450 INFO [train.py:968] (1/2) Epoch 15, batch 39000, giga_loss[loss=0.252, simple_loss=0.3322, pruned_loss=0.0859, over 28750.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3472, pruned_loss=0.1006, over 5710560.34 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3668, pruned_loss=0.1194, over 5696503.82 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3457, pruned_loss=0.09896, over 5704220.06 frames. ], batch size: 284, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:25:47,450 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 01:25:56,539 INFO [train.py:1012] (1/2) Epoch 15, validation: loss=0.2145, simple_loss=0.3218, pruned_loss=0.05359, over 944034.00 frames. 
+2023-03-08 01:25:56,540 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 01:26:12,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=677936.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:26:35,572 INFO [optim.py:369] (1/2) 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 01:26:40,803 INFO [train.py:968] (1/2) Epoch 15, batch 39050, giga_loss[loss=0.267, simple_loss=0.3352, pruned_loss=0.09934, over 28702.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.347, pruned_loss=0.1011, over 5705099.23 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3667, pruned_loss=0.1193, over 5699484.95 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3456, pruned_loss=0.09968, over 5697639.03 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:27:11,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:13,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:17,844 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:20,813 INFO [train.py:968] (1/2) Epoch 15, batch 39100, giga_loss[loss=0.2409, simple_loss=0.3217, pruned_loss=0.0801, over 28984.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3448, pruned_loss=0.1001, over 5711541.76 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3671, pruned_loss=0.1196, over 5705251.60 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3427, pruned_loss=0.09816, over 5700639.61 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:27:40,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:27:58,984 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 39150, libri_loss[loss=0.3371, simple_loss=0.3891, pruned_loss=0.1426, over 29355.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3423, pruned_loss=0.09916, over 5714267.80 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1199, over 5704775.97 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3398, pruned_loss=0.0969, over 5706157.07 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:28:11,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:28:14,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678082.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:28:35,666 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:28:42,230 INFO [train.py:968] (1/2) Epoch 15, batch 39200, giga_loss[loss=0.2396, simple_loss=0.324, pruned_loss=0.07759, over 28612.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3402, pruned_loss=0.09823, over 5705146.53 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3674, pruned_loss=0.12, over 5699354.84 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3375, pruned_loss=0.09579, over 5703238.40 frames. ], batch size: 307, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:28:47,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4047, 2.1385, 1.6885, 1.7107], device='cuda:1'), covar=tensor([0.0753, 0.0241, 0.0312, 0.0851], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-08 01:29:13,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:29:21,645 INFO [optim.py:369] (1/2) 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 01:29:29,399 INFO [train.py:968] (1/2) Epoch 15, batch 39250, giga_loss[loss=0.2474, simple_loss=0.3202, pruned_loss=0.08734, over 28988.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3377, pruned_loss=0.09651, over 5710060.35 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3676, pruned_loss=0.12, over 5700047.41 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.335, pruned_loss=0.09424, over 5707996.85 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:29:33,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:30:08,370 INFO [train.py:968] (1/2) Epoch 15, batch 39300, giga_loss[loss=0.2813, simple_loss=0.3558, pruned_loss=0.1034, over 28718.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3414, pruned_loss=0.09866, over 5705410.78 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3683, pruned_loss=0.1206, over 5701614.26 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3373, pruned_loss=0.09524, over 5703121.25 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:30:15,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1593, 1.2620, 1.1367, 0.8895], device='cuda:1'), covar=tensor([0.0842, 0.0500, 0.1032, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0435, 0.0505, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 01:30:27,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678240.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:30:29,367 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:30:39,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:30:47,336 INFO [optim.py:369] (1/2) 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 01:30:49,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8969, 4.7163, 4.4170, 2.1188], device='cuda:1'), covar=tensor([0.0387, 0.0536, 0.0584, 0.2088], device='cuda:1'), in_proj_covar=tensor([0.1125, 0.1035, 0.0888, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 01:30:52,190 INFO [train.py:968] (1/2) Epoch 15, batch 39350, giga_loss[loss=0.2815, simple_loss=0.3525, pruned_loss=0.1052, over 27980.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3444, pruned_loss=0.1001, over 5686306.41 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3685, pruned_loss=0.1208, over 5687230.17 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3403, pruned_loss=0.09667, over 5697876.51 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:31:18,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:21,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:35,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:36,027 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 01:31:38,622 INFO [train.py:968] (1/2) Epoch 15, batch 39400, giga_loss[loss=0.268, simple_loss=0.3581, pruned_loss=0.0889, over 28906.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3467, pruned_loss=0.1005, over 5688187.54 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.368, pruned_loss=0.1203, over 5690443.81 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3432, pruned_loss=0.09757, over 5694594.53 frames. ], batch size: 174, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:31:39,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:43,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-08 01:31:48,005 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:31:52,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:05,628 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:18,899 INFO [optim.py:369] (1/2) 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 01:32:21,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5586, 1.7258, 1.7903, 1.5420], device='cuda:1'), covar=tensor([0.1782, 0.1943, 0.1969, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0731, 0.0684, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 01:32:23,921 INFO [train.py:968] (1/2) Epoch 15, batch 39450, giga_loss[loss=0.2802, simple_loss=0.3669, pruned_loss=0.09674, over 28716.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3481, pruned_loss=0.1011, over 5671695.15 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1207, over 5675836.95 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3445, pruned_loss=0.09784, over 5689513.13 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:32:36,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678383.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:32:39,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678386.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:32:42,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:47,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:32:51,143 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:33:03,207 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678415.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:33:06,670 INFO [train.py:968] (1/2) Epoch 15, batch 39500, giga_loss[loss=0.2462, simple_loss=0.3167, pruned_loss=0.08786, over 28782.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3476, pruned_loss=0.1004, over 5675455.88 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3682, pruned_loss=0.1207, over 5671805.25 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3442, pruned_loss=0.09726, over 5694121.98 frames. ], batch size: 66, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:33:12,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3782, 2.7137, 2.3683, 1.9613], device='cuda:1'), covar=tensor([0.2686, 0.1724, 0.1854, 0.2429], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1751, 0.1686, 0.1829], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 01:33:16,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:33:45,677 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 39550, giga_loss[loss=0.2674, simple_loss=0.3411, pruned_loss=0.09688, over 29010.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3463, pruned_loss=0.0996, over 5686536.20 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1206, over 5675480.12 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3433, pruned_loss=0.0968, over 5698392.01 frames. ], batch size: 128, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:34:13,851 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:34:24,167 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:34:33,305 INFO [train.py:968] (1/2) Epoch 15, batch 39600, giga_loss[loss=0.243, simple_loss=0.3268, pruned_loss=0.07957, over 29075.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3471, pruned_loss=0.1004, over 5697430.98 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3687, pruned_loss=0.121, over 5679938.02 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3437, pruned_loss=0.09738, over 5703283.58 frames. ], batch size: 155, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 01:34:45,539 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:34:47,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:11,270 INFO [optim.py:369] (1/2) 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 01:35:12,122 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:15,599 INFO [train.py:968] (1/2) Epoch 15, batch 39650, giga_loss[loss=0.2757, simple_loss=0.3519, pruned_loss=0.09976, over 28644.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3474, pruned_loss=0.1005, over 5703203.48 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3691, pruned_loss=0.1211, over 5676155.85 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3437, pruned_loss=0.09736, over 5711711.00 frames. ], batch size: 307, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:35:30,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0812, 1.1024, 3.6688, 3.0528], device='cuda:1'), covar=tensor([0.1734, 0.2859, 0.0425, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0694, 0.0607, 0.0889, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 01:35:47,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=678604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:57,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:35:59,970 INFO [train.py:968] (1/2) Epoch 15, batch 39700, giga_loss[loss=0.3615, simple_loss=0.4214, pruned_loss=0.1508, over 28574.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.1029, over 5700289.90 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3692, pruned_loss=0.1211, over 5676429.00 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3489, pruned_loss=0.1003, over 5707236.31 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:36:05,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5269, 1.8940, 1.7797, 1.4783], device='cuda:1'), covar=tensor([0.2110, 0.1479, 0.1280, 0.1665], device='cuda:1'), in_proj_covar=tensor([0.1838, 0.1759, 0.1694, 0.1836], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 01:36:38,154 INFO [optim.py:369] (1/2) 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 01:36:42,885 INFO [train.py:968] (1/2) Epoch 15, batch 39750, giga_loss[loss=0.2751, simple_loss=0.3526, pruned_loss=0.0988, over 29006.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.354, pruned_loss=0.1037, over 5706520.22 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1212, over 5680006.18 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3511, pruned_loss=0.1013, over 5709127.88 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:37:07,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4868, 1.7043, 1.6183, 1.5445], device='cuda:1'), covar=tensor([0.1540, 0.1763, 0.1950, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0735, 0.0688, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 01:37:14,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:37:20,671 INFO [train.py:968] (1/2) Epoch 15, batch 39800, giga_loss[loss=0.3073, simple_loss=0.3774, pruned_loss=0.1186, over 28048.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3549, pruned_loss=0.1044, over 5712026.01 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3688, pruned_loss=0.1208, over 5689037.49 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3522, pruned_loss=0.1019, over 5707464.09 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:37:35,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7420, 4.7958, 1.7184, 1.8705], device='cuda:1'), covar=tensor([0.0838, 0.0265, 0.0862, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0522, 0.0357, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 01:37:55,618 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:37:57,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:37:58,506 INFO [optim.py:369] (1/2) 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 01:38:05,498 INFO [train.py:968] (1/2) Epoch 15, batch 39850, giga_loss[loss=0.3043, simple_loss=0.3733, pruned_loss=0.1177, over 27778.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3561, pruned_loss=0.1052, over 5709853.77 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3693, pruned_loss=0.1214, over 5690618.64 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1023, over 5705243.34 frames. ], batch size: 474, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:38:24,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:38:48,536 INFO [train.py:968] (1/2) Epoch 15, batch 39900, giga_loss[loss=0.2583, simple_loss=0.3365, pruned_loss=0.09009, over 28920.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3565, pruned_loss=0.1055, over 5713278.51 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5693994.58 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3542, pruned_loss=0.1031, over 5706949.63 frames. ], batch size: 112, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:39:14,093 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=678852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:39:16,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=678855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:39:18,202 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 01:39:23,253 INFO [optim.py:369] (1/2) 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 01:39:28,733 INFO [train.py:968] (1/2) Epoch 15, batch 39950, giga_loss[loss=0.2647, simple_loss=0.3418, pruned_loss=0.09377, over 28672.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3557, pruned_loss=0.105, over 5717841.89 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3691, pruned_loss=0.1211, over 5698417.31 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5709242.03 frames. ], batch size: 242, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:39:32,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678873.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:39:40,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=678884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:39:41,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:40:07,949 INFO [train.py:968] (1/2) Epoch 15, batch 40000, giga_loss[loss=0.252, simple_loss=0.3194, pruned_loss=0.09233, over 28876.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3537, pruned_loss=0.1045, over 5727613.82 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3695, pruned_loss=0.1214, over 5708545.70 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.351, pruned_loss=0.1016, over 5712102.79 frames. ], batch size: 99, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:40:13,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5978, 1.9122, 1.7343, 1.6929], device='cuda:1'), covar=tensor([0.1683, 0.1826, 0.2112, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0733, 0.0686, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 01:40:44,602 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 40050, giga_loss[loss=0.2325, simple_loss=0.3197, pruned_loss=0.07265, over 29022.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3506, pruned_loss=0.103, over 5728155.66 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1209, over 5714273.97 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3483, pruned_loss=0.1005, over 5711209.78 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:40:57,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=678979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:41:28,796 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:41:30,352 INFO [train.py:968] (1/2) Epoch 15, batch 40100, libri_loss[loss=0.3431, simple_loss=0.4134, pruned_loss=0.1364, over 28690.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3491, pruned_loss=0.1014, over 5729569.92 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3694, pruned_loss=0.121, over 5718589.99 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3463, pruned_loss=0.09875, over 5712399.90 frames. ], batch size: 106, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:41:31,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:41:39,475 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:41:41,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:41:55,332 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679048.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:41:57,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3465, 1.0977, 4.3622, 3.4293], device='cuda:1'), covar=tensor([0.1669, 0.2954, 0.0392, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0700, 0.0610, 0.0897, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 01:42:00,003 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679054.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:42:04,141 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:42:07,280 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 40150, giga_loss[loss=0.2634, simple_loss=0.3445, pruned_loss=0.09113, over 28854.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3504, pruned_loss=0.101, over 5731865.92 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3689, pruned_loss=0.1208, over 5721447.83 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3481, pruned_loss=0.09863, over 5715766.43 frames. ], batch size: 112, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:42:49,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4502, 1.7757, 1.5330, 1.6653], device='cuda:1'), covar=tensor([0.0713, 0.0270, 0.0298, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0113, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0063, 0.0057, 0.0097], device='cuda:1')
+2023-03-08 01:42:53,056 INFO [train.py:968] (1/2) Epoch 15, batch 40200, giga_loss[loss=0.2668, simple_loss=0.3462, pruned_loss=0.09368, over 28922.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3507, pruned_loss=0.1005, over 5725801.96 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3686, pruned_loss=0.1206, over 5727423.07 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3484, pruned_loss=0.09798, over 5707158.83 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:42:55,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:42:57,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:43:00,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8738, 2.8961, 1.7832, 0.8736], device='cuda:1'), covar=tensor([0.6595, 0.2933, 0.3833, 0.6590], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1553, 0.1534, 0.1333], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 01:43:22,902 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:43:31,829 INFO [optim.py:369] (1/2) 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 01:43:35,996 INFO [train.py:968] (1/2) Epoch 15, batch 40250, giga_loss[loss=0.2875, simple_loss=0.3595, pruned_loss=0.1077, over 28278.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.35, pruned_loss=0.1007, over 5725643.17 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3681, pruned_loss=0.1203, over 5730238.49 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3484, pruned_loss=0.09868, over 5708468.83 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:43:44,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1749, 1.2203, 3.4459, 3.0049], device='cuda:1'), covar=tensor([0.1532, 0.2691, 0.0442, 0.1351], device='cuda:1'), in_proj_covar=tensor([0.0696, 0.0609, 0.0894, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 01:43:46,945 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-08 01:44:16,398 INFO [train.py:968] (1/2) Epoch 15, batch 40300, giga_loss[loss=0.2639, simple_loss=0.3343, pruned_loss=0.09674, over 28824.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3492, pruned_loss=0.1016, over 5723346.67 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3681, pruned_loss=0.1203, over 5732351.25 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3476, pruned_loss=0.0998, over 5707744.53 frames. ], batch size: 112, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:44:58,612 INFO [optim.py:369] (1/2) 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 01:45:02,535 INFO [train.py:968] (1/2) Epoch 15, batch 40350, giga_loss[loss=0.2247, simple_loss=0.2918, pruned_loss=0.07883, over 28650.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3478, pruned_loss=0.1021, over 5716246.68 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3681, pruned_loss=0.1202, over 5734269.64 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3464, pruned_loss=0.1006, over 5702227.78 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:45:10,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-08 01:45:19,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0105, 1.3362, 1.1498, 0.1889], device='cuda:1'), covar=tensor([0.3010, 0.2307, 0.3819, 0.5518], device='cuda:1'), in_proj_covar=tensor([0.1631, 0.1546, 0.1529, 0.1334], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 01:45:32,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7101, 2.1211, 1.9937, 1.5010], device='cuda:1'), covar=tensor([0.1745, 0.2333, 0.1452, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.0861, 0.0693, 0.0906, 0.0807], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 01:45:43,783 INFO [train.py:968] (1/2) Epoch 15, batch 40400, libri_loss[loss=0.375, simple_loss=0.4131, pruned_loss=0.1685, over 19303.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3472, pruned_loss=0.1027, over 5700776.63 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 5711153.51 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3451, pruned_loss=0.1008, over 5710338.89 frames. ], batch size: 187, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 01:46:12,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8375, 1.8404, 1.9384, 1.7006], device='cuda:1'), covar=tensor([0.1630, 0.2041, 0.1878, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0734, 0.0686, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 01:46:23,338 INFO [optim.py:369] (1/2) 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 01:46:27,151 INFO [train.py:968] (1/2) Epoch 15, batch 40450, libri_loss[loss=0.3284, simple_loss=0.3856, pruned_loss=0.1356, over 29516.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3459, pruned_loss=0.1022, over 5712762.45 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3683, pruned_loss=0.1203, over 5715192.19 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3437, pruned_loss=0.1003, over 5716369.60 frames. ], batch size: 81, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:47:08,379 INFO [train.py:968] (1/2) Epoch 15, batch 40500, giga_loss[loss=0.2299, simple_loss=0.3071, pruned_loss=0.07638, over 28474.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3422, pruned_loss=0.09989, over 5706031.42 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3685, pruned_loss=0.1203, over 5707691.25 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.34, pruned_loss=0.09805, over 5715583.55 frames. ], batch size: 85, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:47:16,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:47:23,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:47:46,617 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:47:50,951 INFO [train.py:968] (1/2) Epoch 15, batch 40550, giga_loss[loss=0.2484, simple_loss=0.3222, pruned_loss=0.08731, over 28912.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3374, pruned_loss=0.0973, over 5706361.12 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3683, pruned_loss=0.1202, over 5702927.07 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3353, pruned_loss=0.0955, over 5717576.11 frames. ], batch size: 227, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:48:33,702 INFO [train.py:968] (1/2) Epoch 15, batch 40600, giga_loss[loss=0.2055, simple_loss=0.2793, pruned_loss=0.06584, over 28164.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.334, pruned_loss=0.09529, over 5705643.42 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1201, over 5703466.14 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3316, pruned_loss=0.09353, over 5714143.31 frames. ], batch size: 77, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:48:35,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2994, 1.4319, 1.3622, 1.5368], device='cuda:1'), covar=tensor([0.0780, 0.0331, 0.0332, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 01:48:46,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:49:12,436 INFO [optim.py:369] (1/2) 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 01:49:15,881 INFO [train.py:968] (1/2) Epoch 15, batch 40650, libri_loss[loss=0.3632, simple_loss=0.3977, pruned_loss=0.1644, over 29544.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3363, pruned_loss=0.0964, over 5706823.88 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3682, pruned_loss=0.1201, over 5709173.48 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3336, pruned_loss=0.09428, over 5708433.31 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:49:19,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679572.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:49:21,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679575.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:49:37,714 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679604.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:50:00,180 INFO [train.py:968] (1/2) Epoch 15, batch 40700, giga_loss[loss=0.2624, simple_loss=0.3409, pruned_loss=0.09199, over 29020.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3392, pruned_loss=0.09715, over 5708958.99 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.368, pruned_loss=0.1199, over 5710207.62 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3372, pruned_loss=0.09552, over 5709364.34 frames. ], batch size: 128, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:50:38,394 INFO [optim.py:369] (1/2) 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 01:50:41,466 INFO [train.py:968] (1/2) Epoch 15, batch 40750, giga_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08951, over 28892.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3428, pruned_loss=0.09848, over 5700090.46 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1202, over 5700128.62 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3404, pruned_loss=0.09659, over 5709401.76 frames. ], batch size: 66, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:51:15,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:51:22,730 INFO [train.py:968] (1/2) Epoch 15, batch 40800, giga_loss[loss=0.2532, simple_loss=0.327, pruned_loss=0.08974, over 28707.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3461, pruned_loss=0.09986, over 5703066.76 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3686, pruned_loss=0.1202, over 5692812.66 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3433, pruned_loss=0.09775, over 5716220.29 frames. ], batch size: 119, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 01:51:47,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4153, 2.0937, 1.6183, 0.6240], device='cuda:1'), covar=tensor([0.5366, 0.2569, 0.3338, 0.5717], device='cuda:1'), in_proj_covar=tensor([0.1624, 0.1545, 0.1521, 0.1330], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 01:52:06,682 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 40850, giga_loss[loss=0.2437, simple_loss=0.3186, pruned_loss=0.08441, over 28587.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3492, pruned_loss=0.1015, over 5703807.45 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3688, pruned_loss=0.1201, over 5695478.92 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3463, pruned_loss=0.0994, over 5712167.51 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:52:31,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5023, 1.7156, 1.7826, 1.5004], device='cuda:1'), covar=tensor([0.1403, 0.1480, 0.1768, 0.1527], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0735, 0.0686, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 01:52:45,084 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:52:45,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:52:52,530 INFO [train.py:968] (1/2) Epoch 15, batch 40900, giga_loss[loss=0.2894, simple_loss=0.3616, pruned_loss=0.1086, over 28835.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.352, pruned_loss=0.104, over 5694163.55 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.369, pruned_loss=0.1204, over 5690417.65 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3491, pruned_loss=0.1017, over 5705824.16 frames. ], batch size: 174, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:53:19,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679840.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:53:25,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4339, 1.6469, 1.6489, 1.4171], device='cuda:1'), covar=tensor([0.2466, 0.2122, 0.1585, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.1858, 0.1771, 0.1698, 0.1833], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 01:53:44,729 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 40950, libri_loss[loss=0.3254, simple_loss=0.3899, pruned_loss=0.1304, over 29374.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3604, pruned_loss=0.1114, over 5677172.62 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3691, pruned_loss=0.1203, over 5694537.77 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3577, pruned_loss=0.1094, over 5682306.08 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:54:09,565 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=679890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:54:28,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=679909.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:54:35,387 INFO [train.py:968] (1/2) Epoch 15, batch 41000, giga_loss[loss=0.3696, simple_loss=0.4251, pruned_loss=0.1571, over 28230.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3668, pruned_loss=0.1162, over 5677602.53 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3688, pruned_loss=0.1201, over 5697085.36 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3649, pruned_loss=0.1147, over 5678910.93 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:55:13,192 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:55:15,343 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:55:25,210 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 41050, giga_loss[loss=0.3159, simple_loss=0.3796, pruned_loss=0.1261, over 28731.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3737, pruned_loss=0.1218, over 5655789.36 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3692, pruned_loss=0.1204, over 5686190.91 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3717, pruned_loss=0.1202, over 5665764.18 frames. ], batch size: 242, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:55:35,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=679983.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:55:37,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=679986.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:55:38,237 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=679987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:55:42,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4426, 1.5300, 1.4960, 1.3245], device='cuda:1'), covar=tensor([0.1903, 0.1998, 0.1603, 0.1874], device='cuda:1'), in_proj_covar=tensor([0.1864, 0.1779, 0.1700, 0.1841], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 01:56:02,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3945, 1.5195, 1.1943, 1.5311], device='cuda:1'), covar=tensor([0.0755, 0.0315, 0.0330, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 01:56:06,206 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:56:10,165 INFO [train.py:968] (1/2) Epoch 15, batch 41100, giga_loss[loss=0.321, simple_loss=0.3917, pruned_loss=0.1252, over 29011.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3794, pruned_loss=0.1267, over 5653743.15 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3698, pruned_loss=0.1207, over 5674528.11 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3775, pruned_loss=0.1252, over 5670733.67 frames. ], batch size: 128, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:56:29,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5329, 1.7552, 1.3544, 1.5366], device='cuda:1'), covar=tensor([0.2349, 0.2393, 0.2795, 0.2250], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1020, 0.1237, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 01:56:42,773 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680052.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:56:44,469 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-08 01:56:45,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680055.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 01:56:58,682 INFO [optim.py:369] (1/2) 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 01:57:00,863 INFO [train.py:968] (1/2) Epoch 15, batch 41150, giga_loss[loss=0.3941, simple_loss=0.4151, pruned_loss=0.1865, over 23761.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3849, pruned_loss=0.1313, over 5642267.76 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 5667718.97 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3833, pruned_loss=0.13, over 5661168.05 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:57:17,079 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680084.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 01:57:17,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:57:29,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8283, 1.9126, 2.0704, 1.6295], device='cuda:1'), covar=tensor([0.1757, 0.2134, 0.1364, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.0854, 0.0686, 0.0900, 0.0803], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 01:57:42,116 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-08 01:57:43,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:57:48,571 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:57:52,848 INFO [train.py:968] (1/2) Epoch 15, batch 41200, giga_loss[loss=0.3786, simple_loss=0.442, pruned_loss=0.1576, over 28867.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3864, pruned_loss=0.1331, over 5641106.24 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3704, pruned_loss=0.1212, over 5662050.23 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3851, pruned_loss=0.1321, over 5659957.79 frames. ], batch size: 145, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 01:58:19,376 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:58:48,297 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 41250, giga_loss[loss=0.3485, simple_loss=0.4048, pruned_loss=0.1461, over 28819.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3886, pruned_loss=0.1363, over 5621132.70 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3702, pruned_loss=0.1211, over 5665609.28 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3882, pruned_loss=0.1359, over 5632618.84 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:58:48,528 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:59:42,237 INFO [train.py:968] (1/2) Epoch 15, batch 41300, giga_loss[loss=0.3407, simple_loss=0.3912, pruned_loss=0.1451, over 29013.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.39, pruned_loss=0.1384, over 5617394.69 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3694, pruned_loss=0.1204, over 5670169.98 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3911, pruned_loss=0.1393, over 5620718.67 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 01:59:51,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 01:59:54,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:00:21,560 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:00:31,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:00:34,418 INFO [optim.py:369] (1/2) 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 02:00:34,430 INFO [train.py:968] (1/2) Epoch 15, batch 41350, giga_loss[loss=0.3051, simple_loss=0.3815, pruned_loss=0.1144, over 28686.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3949, pruned_loss=0.1428, over 5618267.69 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1208, over 5664252.13 frames. ], giga_tot_loss[loss=0.3416, simple_loss=0.3961, pruned_loss=0.1436, over 5624913.94 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:01:27,527 INFO [train.py:968] (1/2) Epoch 15, batch 41400, giga_loss[loss=0.4019, simple_loss=0.4183, pruned_loss=0.1928, over 23595.00 frames. ], tot_loss[loss=0.3435, simple_loss=0.3969, pruned_loss=0.1451, over 5625413.80 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3699, pruned_loss=0.121, over 5666877.26 frames. ], giga_tot_loss[loss=0.3453, simple_loss=0.3983, pruned_loss=0.1461, over 5626984.20 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:01:39,654 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:01:42,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:01:59,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4143, 2.1464, 1.5649, 0.5680], device='cuda:1'), covar=tensor([0.4113, 0.2537, 0.3647, 0.5049], device='cuda:1'), in_proj_covar=tensor([0.1638, 0.1563, 0.1538, 0.1336], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:02:02,880 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-08 02:02:10,336 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:02:18,315 INFO [optim.py:369] (1/2) 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 02:02:18,328 INFO [train.py:968] (1/2) Epoch 15, batch 41450, giga_loss[loss=0.3252, simple_loss=0.3854, pruned_loss=0.1325, over 28699.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.394, pruned_loss=0.1436, over 5632261.67 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3699, pruned_loss=0.121, over 5671176.07 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3955, pruned_loss=0.1449, over 5629008.57 frames. ], batch size: 242, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:02:21,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=680372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:02:54,949 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=680403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:02:58,514 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:03:00,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:03:08,242 INFO [train.py:968] (1/2) Epoch 15, batch 41500, giga_loss[loss=0.3251, simple_loss=0.3871, pruned_loss=0.1315, over 28909.00 frames. ], tot_loss[loss=0.3402, simple_loss=0.3936, pruned_loss=0.1434, over 5613765.86 frames. ], libri_tot_loss[loss=0.3066, simple_loss=0.3704, pruned_loss=0.1214, over 5653919.87 frames. ], giga_tot_loss[loss=0.3422, simple_loss=0.3951, pruned_loss=0.1447, over 5624516.49 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:03:24,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5193, 1.5558, 1.2099, 1.1264], device='cuda:1'), covar=tensor([0.0729, 0.0460, 0.0895, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0446, 0.0509, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 02:03:27,465 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:03:56,552 INFO [optim.py:369] (1/2) 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 02:03:56,565 INFO [train.py:968] (1/2) Epoch 15, batch 41550, giga_loss[loss=0.3534, simple_loss=0.4031, pruned_loss=0.1518, over 27491.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3925, pruned_loss=0.1419, over 5619678.72 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3701, pruned_loss=0.1213, over 5665633.75 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.395, pruned_loss=0.144, over 5615894.36 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:04:01,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 02:04:46,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3166, 3.1351, 2.9602, 1.4448], device='cuda:1'), covar=tensor([0.0907, 0.1039, 0.0907, 0.2332], device='cuda:1'), in_proj_covar=tensor([0.1141, 0.1054, 0.0906, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 02:04:48,919 INFO [train.py:968] (1/2) Epoch 15, batch 41600, giga_loss[loss=0.3263, simple_loss=0.3863, pruned_loss=0.1332, over 28896.00 frames. ], tot_loss[loss=0.3406, simple_loss=0.3946, pruned_loss=0.1433, over 5617738.72 frames. ], libri_tot_loss[loss=0.3064, simple_loss=0.3701, pruned_loss=0.1214, over 5665981.11 frames. ], giga_tot_loss[loss=0.3437, simple_loss=0.397, pruned_loss=0.1452, over 5613674.31 frames. ], batch size: 112, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:04:55,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2029, 1.2212, 1.0295, 0.8895], device='cuda:1'), covar=tensor([0.0685, 0.0356, 0.0816, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0444, 0.0508, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 02:05:20,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:05:43,055 INFO [optim.py:369] (1/2) 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 02:05:43,067 INFO [train.py:968] (1/2) Epoch 15, batch 41650, giga_loss[loss=0.2831, simple_loss=0.3497, pruned_loss=0.1082, over 28827.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3915, pruned_loss=0.1408, over 5597826.50 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3696, pruned_loss=0.121, over 5663152.92 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3946, pruned_loss=0.1433, over 5595306.24 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:06:29,195 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-08 02:06:35,625 INFO [train.py:968] (1/2) Epoch 15, batch 41700, giga_loss[loss=0.3717, simple_loss=0.4015, pruned_loss=0.171, over 23856.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3882, pruned_loss=0.1371, over 5613773.04 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3694, pruned_loss=0.121, over 5668734.30 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3914, pruned_loss=0.1395, over 5605291.65 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:06:49,265 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 15, batch 41750, giga_loss[loss=0.3528, simple_loss=0.3879, pruned_loss=0.1589, over 24156.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3865, pruned_loss=0.1345, over 5629915.39 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 5671468.39 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3896, pruned_loss=0.1369, over 5619804.19 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:07:28,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-08 02:07:41,149 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:07:43,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:08:12,823 INFO [train.py:968] (1/2) Epoch 15, batch 41800, giga_loss[loss=0.3521, simple_loss=0.3994, pruned_loss=0.1524, over 26562.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3822, pruned_loss=0.1311, over 5635368.36 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3683, pruned_loss=0.1204, over 5677984.51 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.386, pruned_loss=0.1339, over 5620503.16 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:08:15,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:08:46,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:08:47,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3679, 1.7845, 1.5691, 1.5189], device='cuda:1'), covar=tensor([0.1925, 0.1778, 0.2196, 0.1991], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0734, 0.0687, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 02:08:51,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3734, 2.0139, 1.5692, 0.5794], device='cuda:1'), covar=tensor([0.4273, 0.2517, 0.3368, 0.5220], device='cuda:1'), in_proj_covar=tensor([0.1639, 0.1559, 0.1532, 0.1339], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:09:02,281 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8328, 1.8493, 1.3334, 1.4933], device='cuda:1'), covar=tensor([0.0781, 0.0641, 0.1010, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0444, 0.0506, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 02:09:05,688 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 41850, giga_loss[loss=0.2598, simple_loss=0.3331, pruned_loss=0.09321, over 28484.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.38, pruned_loss=0.1292, over 5638375.03 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3683, pruned_loss=0.1204, over 5683612.50 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3833, pruned_loss=0.1316, over 5620330.91 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:09:13,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=680778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:09:56,594 INFO [train.py:968] (1/2) Epoch 15, batch 41900, giga_loss[loss=0.2987, simple_loss=0.377, pruned_loss=0.1102, over 28865.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3793, pruned_loss=0.1288, over 5648791.22 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3677, pruned_loss=0.1201, over 5688699.00 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3827, pruned_loss=0.1312, over 5628970.94 frames. ], batch size: 145, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:10:11,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 02:10:47,356 INFO [optim.py:369] (1/2) 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 02:10:47,368 INFO [train.py:968] (1/2) Epoch 15, batch 41950, giga_loss[loss=0.2657, simple_loss=0.352, pruned_loss=0.08973, over 29024.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3799, pruned_loss=0.1291, over 5648736.61 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3679, pruned_loss=0.1202, over 5680513.48 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3825, pruned_loss=0.1309, over 5639496.69 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:11:11,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:13,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:41,981 INFO [train.py:968] (1/2) Epoch 15, batch 42000, giga_loss[loss=0.2891, simple_loss=0.3567, pruned_loss=0.1107, over 28623.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3776, pruned_loss=0.1271, over 5640459.54 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3679, pruned_loss=0.1201, over 5684444.66 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3799, pruned_loss=0.1288, over 5628993.19 frames. ], batch size: 307, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 02:11:41,982 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 02:11:45,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1016, 1.5317, 1.6043, 1.3389], device='cuda:1'), covar=tensor([0.1781, 0.1486, 0.2002, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0736, 0.0689, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 02:11:50,697 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 02:11:53,056 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=680921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:53,742 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:11:56,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=680924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:12:30,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=680953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:12:46,249 INFO [train.py:968] (1/2) Epoch 15, batch 42050, giga_loss[loss=0.3108, simple_loss=0.3898, pruned_loss=0.1159, over 28608.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3783, pruned_loss=0.1253, over 5648379.00 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3678, pruned_loss=0.1202, over 5687547.56 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3803, pruned_loss=0.1266, over 5635849.25 frames. ], batch size: 60, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:12:46,902 INFO [optim.py:369] (1/2) 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 02:13:27,468 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 42100, giga_loss[loss=0.3067, simple_loss=0.3571, pruned_loss=0.1281, over 23700.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3804, pruned_loss=0.1254, over 5656453.88 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3678, pruned_loss=0.1202, over 5687403.00 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3823, pruned_loss=0.1265, over 5645374.95 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:14:06,056 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-08 02:14:23,908 INFO [train.py:968] (1/2) Epoch 15, batch 42150, giga_loss[loss=0.2725, simple_loss=0.3535, pruned_loss=0.09572, over 28777.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3805, pruned_loss=0.1256, over 5656895.34 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3675, pruned_loss=0.1201, over 5681440.62 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3827, pruned_loss=0.1268, over 5653109.79 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:14:24,584 INFO [optim.py:369] (1/2) 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 02:14:53,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:15:01,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:15:12,085 INFO [train.py:968] (1/2) Epoch 15, batch 42200, giga_loss[loss=0.2884, simple_loss=0.3609, pruned_loss=0.108, over 29015.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3794, pruned_loss=0.1252, over 5651145.39 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3676, pruned_loss=0.1201, over 5675800.37 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3813, pruned_loss=0.1263, over 5652034.12 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:15:31,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4226, 1.7592, 1.3569, 1.5135], device='cuda:1'), covar=tensor([0.2406, 0.2375, 0.2662, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1390, 0.1019, 0.1235, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 02:15:43,349 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:15:46,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:15:59,003 INFO [train.py:968] (1/2) Epoch 15, batch 42250, giga_loss[loss=0.3255, simple_loss=0.3808, pruned_loss=0.1351, over 28345.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3775, pruned_loss=0.1245, over 5666952.28 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.12, over 5681719.13 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3792, pruned_loss=0.1256, over 5661996.93 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:15:59,411 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-08 02:15:59,611 INFO [optim.py:369] (1/2) 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 02:16:14,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:16:51,456 INFO [train.py:968] (1/2) Epoch 15, batch 42300, giga_loss[loss=0.3661, simple_loss=0.4225, pruned_loss=0.1548, over 28685.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3771, pruned_loss=0.1257, over 5657250.30 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3679, pruned_loss=0.1202, over 5676129.47 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3783, pruned_loss=0.1265, over 5658140.81 frames. ], batch size: 242, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:17:05,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8827, 1.9111, 1.3990, 1.5544], device='cuda:1'), covar=tensor([0.0856, 0.0712, 0.1003, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0443, 0.0505, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 02:17:44,961 INFO [train.py:968] (1/2) Epoch 15, batch 42350, giga_loss[loss=0.2906, simple_loss=0.3777, pruned_loss=0.1018, over 29055.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3752, pruned_loss=0.1237, over 5659788.81 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3677, pruned_loss=0.12, over 5678195.68 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3764, pruned_loss=0.1245, over 5658355.99 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:17:45,878 INFO [optim.py:369] (1/2) 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 02:18:19,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2644, 1.5584, 1.2502, 0.9272], device='cuda:1'), covar=tensor([0.2656, 0.2555, 0.2965, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.1392, 0.1018, 0.1235, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 02:18:29,847 INFO [train.py:968] (1/2) Epoch 15, batch 42400, giga_loss[loss=0.2906, simple_loss=0.3734, pruned_loss=0.1039, over 28863.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3744, pruned_loss=0.1216, over 5674591.05 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3673, pruned_loss=0.1196, over 5684845.94 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.376, pruned_loss=0.1227, over 5667309.12 frames. ], batch size: 112, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 02:19:23,815 INFO [train.py:968] (1/2) Epoch 15, batch 42450, giga_loss[loss=0.2665, simple_loss=0.3374, pruned_loss=0.09785, over 28658.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3763, pruned_loss=0.1229, over 5673548.87 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5685075.71 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3775, pruned_loss=0.1237, over 5667615.62 frames. ], batch size: 60, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:19:25,091 INFO [optim.py:369] (1/2) 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 02:19:25,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 02:19:48,287 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681397.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:20:06,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3540, 2.2089, 1.6821, 1.8674], device='cuda:1'), covar=tensor([0.0829, 0.0744, 0.1013, 0.1143], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0445, 0.0508, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 02:20:12,242 INFO [train.py:968] (1/2) Epoch 15, batch 42500, giga_loss[loss=0.3, simple_loss=0.3647, pruned_loss=0.1177, over 28859.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.377, pruned_loss=0.1243, over 5669559.09 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3676, pruned_loss=0.1198, over 5685895.75 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.378, pruned_loss=0.1249, over 5663628.69 frames. ], batch size: 199, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:20:12,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1926, 1.1396, 3.5978, 3.1689], device='cuda:1'), covar=tensor([0.1558, 0.2666, 0.0457, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0612, 0.0901, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 02:20:45,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5276, 2.2068, 1.6336, 0.6660], device='cuda:1'), covar=tensor([0.5177, 0.2664, 0.3470, 0.5780], device='cuda:1'), in_proj_covar=tensor([0.1639, 0.1562, 0.1529, 0.1339], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:20:59,085 INFO [train.py:968] (1/2) Epoch 15, batch 42550, giga_loss[loss=0.3372, simple_loss=0.3899, pruned_loss=0.1423, over 27578.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3737, pruned_loss=0.122, over 5682065.39 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1196, over 5689696.07 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3746, pruned_loss=0.1227, over 5673695.58 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:21:01,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3898, 2.1633, 1.6530, 0.5036], device='cuda:1'), covar=tensor([0.3849, 0.2433, 0.3116, 0.4950], device='cuda:1'), in_proj_covar=tensor([0.1639, 0.1562, 0.1528, 0.1339], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:21:01,847 INFO [optim.py:369] (1/2) 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 02:21:04,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:21:10,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:21:46,529 INFO [train.py:968] (1/2) Epoch 15, batch 42600, giga_loss[loss=0.2843, simple_loss=0.3574, pruned_loss=0.1056, over 29041.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3731, pruned_loss=0.1223, over 5668763.26 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3685, pruned_loss=0.1204, over 5683718.51 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3732, pruned_loss=0.1223, over 5666589.13 frames. ], batch size: 155, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:21:48,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-08 02:22:29,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3851, 3.4218, 1.5146, 1.5211], device='cuda:1'), covar=tensor([0.0964, 0.0339, 0.0849, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0531, 0.0359, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 02:22:35,617 INFO [train.py:968] (1/2) Epoch 15, batch 42650, giga_loss[loss=0.2797, simple_loss=0.3401, pruned_loss=0.1097, over 28575.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.372, pruned_loss=0.1221, over 5683576.14 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3686, pruned_loss=0.1204, over 5689180.63 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3722, pruned_loss=0.1221, over 5676628.23 frames. ], batch size: 85, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:22:37,585 INFO [optim.py:369] (1/2) 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 02:22:41,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5493, 2.2174, 1.6063, 0.8070], device='cuda:1'), covar=tensor([0.4503, 0.2434, 0.3710, 0.5141], device='cuda:1'), in_proj_covar=tensor([0.1648, 0.1570, 0.1538, 0.1346], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:22:46,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5062, 1.7022, 1.4081, 1.5716], device='cuda:1'), covar=tensor([0.2713, 0.2735, 0.3099, 0.2456], device='cuda:1'), in_proj_covar=tensor([0.1397, 0.1022, 0.1238, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 02:22:58,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:21,958 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:23,461 INFO [train.py:968] (1/2) Epoch 15, batch 42700, giga_loss[loss=0.3523, simple_loss=0.4008, pruned_loss=0.1519, over 28539.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.371, pruned_loss=0.1223, over 5680378.65 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1202, over 5690688.15 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3714, pruned_loss=0.1226, over 5673412.45 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:23:24,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:25,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3281, 3.0969, 1.3850, 1.4730], device='cuda:1'), covar=tensor([0.0970, 0.0391, 0.0889, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0531, 0.0359, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 02:23:30,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:32,785 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:32,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:23:53,641 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:24:00,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:24:11,695 INFO [train.py:968] (1/2) Epoch 15, batch 42750, giga_loss[loss=0.3441, simple_loss=0.3833, pruned_loss=0.1524, over 23707.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3709, pruned_loss=0.1233, over 5654058.64 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3683, pruned_loss=0.1202, over 5678467.20 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3714, pruned_loss=0.1235, over 5659274.63 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:24:15,167 INFO [optim.py:369] (1/2) 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 02:24:30,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=681686.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:24:50,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5702, 1.8108, 1.4709, 1.5327], device='cuda:1'), covar=tensor([0.2543, 0.2534, 0.2930, 0.2241], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1020, 0.1235, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 02:25:00,892 INFO [train.py:968] (1/2) Epoch 15, batch 42800, giga_loss[loss=0.2974, simple_loss=0.3645, pruned_loss=0.1151, over 28685.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3698, pruned_loss=0.1228, over 5644142.05 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3685, pruned_loss=0.1204, over 5672074.70 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3701, pruned_loss=0.123, over 5653534.42 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:25:08,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5084, 2.3017, 2.3803, 1.9296], device='cuda:1'), covar=tensor([0.1596, 0.2333, 0.1837, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0736, 0.0688, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 02:25:27,802 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 42850, giga_loss[loss=0.2716, simple_loss=0.3445, pruned_loss=0.09937, over 28660.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3696, pruned_loss=0.1218, over 5647066.51 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3687, pruned_loss=0.1206, over 5667187.71 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3696, pruned_loss=0.1218, over 5658751.79 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:25:52,852 INFO [optim.py:369] (1/2) 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 02:25:53,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681772.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:25:57,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5595, 1.7683, 1.4516, 1.7134], device='cuda:1'), covar=tensor([0.0767, 0.0293, 0.0328, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 02:26:18,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3682, 1.6378, 1.5636, 1.3023], device='cuda:1'), covar=tensor([0.2113, 0.1760, 0.1368, 0.1776], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1779, 0.1714, 0.1858], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 02:26:35,829 INFO [train.py:968] (1/2) Epoch 15, batch 42900, giga_loss[loss=0.3093, simple_loss=0.3779, pruned_loss=0.1203, over 28911.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3704, pruned_loss=0.1214, over 5638105.17 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3691, pruned_loss=0.1209, over 5649590.75 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3702, pruned_loss=0.1212, over 5662581.55 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:26:56,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4296, 2.0321, 1.7384, 1.7043], device='cuda:1'), covar=tensor([0.0767, 0.0268, 0.0307, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0115, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 02:27:25,446 INFO [train.py:968] (1/2) Epoch 15, batch 42950, giga_loss[loss=0.3154, simple_loss=0.3789, pruned_loss=0.126, over 28679.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3703, pruned_loss=0.1204, over 5647700.48 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.369, pruned_loss=0.1208, over 5644228.19 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3702, pruned_loss=0.1203, over 5672455.16 frames. ], batch size: 92, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:27:29,202 INFO [optim.py:369] (1/2) 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 02:28:13,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4183, 1.3533, 1.2524, 1.5879], device='cuda:1'), covar=tensor([0.0772, 0.0349, 0.0329, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 02:28:14,656 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=681915.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:28:16,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=681918.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:28:17,034 INFO [train.py:968] (1/2) Epoch 15, batch 43000, giga_loss[loss=0.4062, simple_loss=0.4455, pruned_loss=0.1835, over 27590.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3719, pruned_loss=0.1218, over 5658147.12 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.369, pruned_loss=0.1207, over 5650214.71 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3719, pruned_loss=0.1218, over 5673212.26 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:28:43,916 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=681947.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:28:47,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 02:29:04,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=681968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:29:05,152 INFO [train.py:968] (1/2) Epoch 15, batch 43050, giga_loss[loss=0.2599, simple_loss=0.3369, pruned_loss=0.09149, over 28969.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3747, pruned_loss=0.1244, over 5672991.56 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.369, pruned_loss=0.1207, over 5656319.26 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3748, pruned_loss=0.1245, over 5680010.65 frames. ], batch size: 155, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:29:05,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4438, 1.6234, 1.5163, 1.4097], device='cuda:1'), covar=tensor([0.1455, 0.1710, 0.2033, 0.1752], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0735, 0.0685, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 02:29:08,707 INFO [optim.py:369] (1/2) 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 02:29:35,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5626, 1.7678, 1.7643, 1.3684], device='cuda:1'), covar=tensor([0.1645, 0.2200, 0.1378, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.0862, 0.0698, 0.0908, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 02:29:40,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:29:56,718 INFO [train.py:968] (1/2) Epoch 15, batch 43100, giga_loss[loss=0.3252, simple_loss=0.3819, pruned_loss=0.1343, over 28981.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3759, pruned_loss=0.1269, over 5680533.95 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3683, pruned_loss=0.1202, over 5660720.98 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5682624.13 frames. ], batch size: 136, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:30:42,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682061.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:30:45,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5067, 2.3220, 1.7016, 0.7405], device='cuda:1'), covar=tensor([0.5452, 0.2646, 0.3653, 0.5667], device='cuda:1'), in_proj_covar=tensor([0.1651, 0.1576, 0.1541, 0.1351], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:30:49,150 INFO [train.py:968] (1/2) Epoch 15, batch 43150, giga_loss[loss=0.3455, simple_loss=0.3961, pruned_loss=0.1475, over 28608.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3757, pruned_loss=0.1276, over 5680225.09 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.12, over 5665613.97 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3769, pruned_loss=0.1285, over 5677596.04 frames. ], batch size: 307, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:30:54,585 INFO [optim.py:369] (1/2) 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 02:31:32,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:34,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:37,504 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:37,887 INFO [train.py:968] (1/2) Epoch 15, batch 43200, giga_loss[loss=0.2887, simple_loss=0.358, pruned_loss=0.1097, over 28922.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3772, pruned_loss=0.1292, over 5666734.88 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3675, pruned_loss=0.1196, over 5670538.28 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3792, pruned_loss=0.1308, over 5659932.82 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:31:43,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:31:44,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682125.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:32:01,619 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:05,246 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:07,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:25,612 INFO [train.py:968] (1/2) Epoch 15, batch 43250, giga_loss[loss=0.4229, simple_loss=0.456, pruned_loss=0.1949, over 26427.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.377, pruned_loss=0.1294, over 5669765.60 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3673, pruned_loss=0.1195, over 5674106.27 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3788, pruned_loss=0.1309, over 5660850.02 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:32:29,048 INFO [optim.py:369] (1/2) 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 02:32:34,402 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:32:58,912 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682204.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:33:01,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682207.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:33:09,769 INFO [train.py:968] (1/2) Epoch 15, batch 43300, giga_loss[loss=0.2669, simple_loss=0.3519, pruned_loss=0.09093, over 28965.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.376, pruned_loss=0.1271, over 5667367.26 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1192, over 5669021.49 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3781, pruned_loss=0.1288, over 5664155.31 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:33:28,302 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682236.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:33:49,790 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9151, 3.7588, 3.5447, 1.8288], device='cuda:1'), covar=tensor([0.0669, 0.0778, 0.0721, 0.2177], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.1059, 0.0910, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 02:33:55,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1570, 1.1927, 3.2549, 2.9196], device='cuda:1'), covar=tensor([0.1530, 0.2541, 0.0555, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0711, 0.0619, 0.0909, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 02:33:59,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:34:00,266 INFO [train.py:968] (1/2) Epoch 15, batch 43350, giga_loss[loss=0.2568, simple_loss=0.3394, pruned_loss=0.08708, over 28923.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1248, over 5662262.54 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5666667.97 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3759, pruned_loss=0.126, over 5661928.82 frames. ], batch size: 128, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:34:03,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:34:05,097 INFO [optim.py:369] (1/2) 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 02:34:27,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:34:45,107 INFO [train.py:968] (1/2) Epoch 15, batch 43400, giga_loss[loss=0.2962, simple_loss=0.3647, pruned_loss=0.1138, over 28732.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1243, over 5660617.84 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3671, pruned_loss=0.1194, over 5669176.59 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1256, over 5657670.43 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:35:10,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1786, 3.9899, 3.8103, 1.7690], device='cuda:1'), covar=tensor([0.0663, 0.0768, 0.0778, 0.2177], device='cuda:1'), in_proj_covar=tensor([0.1152, 0.1064, 0.0916, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 02:35:25,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4823, 2.4767, 1.9239, 2.0225], device='cuda:1'), covar=tensor([0.0805, 0.0664, 0.0924, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0445, 0.0503, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 02:35:31,076 INFO [train.py:968] (1/2) Epoch 15, batch 43450, giga_loss[loss=0.3188, simple_loss=0.3828, pruned_loss=0.1274, over 28453.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3716, pruned_loss=0.1242, over 5662425.37 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.367, pruned_loss=0.1195, over 5665283.10 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3731, pruned_loss=0.1252, over 5663560.58 frames. ], batch size: 85, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:35:32,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9639, 1.1953, 1.1260, 0.8720], device='cuda:1'), covar=tensor([0.1860, 0.2051, 0.1159, 0.1740], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1775, 0.1703, 0.1849], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 02:35:33,745 INFO [optim.py:369] (1/2) 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 02:36:19,657 INFO [train.py:968] (1/2) Epoch 15, batch 43500, giga_loss[loss=0.2995, simple_loss=0.369, pruned_loss=0.115, over 28639.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3702, pruned_loss=0.1233, over 5660553.21 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3669, pruned_loss=0.1194, over 5660315.89 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3716, pruned_loss=0.1243, over 5666678.10 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:37:09,122 INFO [train.py:968] (1/2) Epoch 15, batch 43550, giga_loss[loss=0.3853, simple_loss=0.426, pruned_loss=0.1723, over 27547.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3735, pruned_loss=0.1254, over 5660630.59 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3669, pruned_loss=0.1194, over 5665277.70 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3747, pruned_loss=0.1263, over 5661061.19 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:37:14,232 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 02:37:14,396 INFO [optim.py:369] (1/2) 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 02:37:17,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-08 02:37:33,793 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:37:39,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=682500.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:37:56,923 INFO [train.py:968] (1/2) Epoch 15, batch 43600, giga_loss[loss=0.3825, simple_loss=0.4181, pruned_loss=0.1734, over 26616.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3771, pruned_loss=0.1253, over 5655477.41 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 5659927.01 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3787, pruned_loss=0.1265, over 5660084.21 frames. ], batch size: 555, lr: 2.11e-03, grad_scale: 8.0
+2023-03-08 02:38:09,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5591, 4.3796, 4.1249, 2.0844], device='cuda:1'), covar=tensor([0.0660, 0.0834, 0.0900, 0.1925], device='cuda:1'), in_proj_covar=tensor([0.1153, 0.1066, 0.0916, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 02:38:56,145 INFO [train.py:968] (1/2) Epoch 15, batch 43650, giga_loss[loss=0.2648, simple_loss=0.3486, pruned_loss=0.09048, over 28991.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3773, pruned_loss=0.1239, over 5662954.10 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3662, pruned_loss=0.1189, over 5661283.37 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3788, pruned_loss=0.125, over 5665221.64 frames. ], batch size: 155, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:38:59,345 INFO [optim.py:369] (1/2) 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 02:39:44,025 INFO [train.py:968] (1/2) Epoch 15, batch 43700, giga_loss[loss=0.3845, simple_loss=0.4263, pruned_loss=0.1713, over 27904.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3803, pruned_loss=0.1263, over 5661677.66 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3668, pruned_loss=0.1192, over 5664061.75 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3812, pruned_loss=0.127, over 5660808.29 frames. ], batch size: 412, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:40:02,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:05,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:10,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=682643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:40:13,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=682646.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 02:40:13,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:35,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:40:35,832 INFO [train.py:968] (1/2) Epoch 15, batch 43750, giga_loss[loss=0.3125, simple_loss=0.3766, pruned_loss=0.1242, over 28610.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3815, pruned_loss=0.1276, over 5666252.27 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1191, over 5667384.61 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3824, pruned_loss=0.1284, over 5662544.33 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:40:40,624 INFO [optim.py:369] (1/2) 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 02:40:43,661 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=682675.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 02:41:13,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4955, 1.8422, 1.4288, 1.3748], device='cuda:1'), covar=tensor([0.2563, 0.2579, 0.2880, 0.2295], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1022, 0.1236, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 02:41:18,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-08 02:41:20,434 INFO [train.py:968] (1/2) Epoch 15, batch 43800, giga_loss[loss=0.2769, simple_loss=0.3471, pruned_loss=0.1034, over 28579.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3815, pruned_loss=0.1289, over 5664236.56 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3671, pruned_loss=0.1194, over 5662339.06 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3825, pruned_loss=0.1296, over 5665052.42 frames. ], batch size: 78, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:41:38,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4045, 1.9788, 1.4863, 0.6781], device='cuda:1'), covar=tensor([0.4267, 0.2142, 0.3157, 0.5137], device='cuda:1'), in_proj_covar=tensor([0.1645, 0.1578, 0.1539, 0.1350], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:42:11,536 INFO [train.py:968] (1/2) Epoch 15, batch 43850, giga_loss[loss=0.2989, simple_loss=0.3706, pruned_loss=0.1136, over 28660.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3786, pruned_loss=0.1275, over 5662238.84 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3671, pruned_loss=0.1194, over 5665265.32 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3795, pruned_loss=0.1281, over 5660284.37 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:42:15,292 INFO [optim.py:369] (1/2) 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 02:42:24,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4822, 1.7781, 1.5102, 1.6243], device='cuda:1'), covar=tensor([0.0773, 0.0314, 0.0307, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 02:42:35,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:43:01,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1236, 1.2802, 3.4843, 2.9443], device='cuda:1'), covar=tensor([0.1682, 0.2566, 0.0492, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0714, 0.0619, 0.0913, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 02:43:01,435 INFO [train.py:968] (1/2) Epoch 15, batch 43900, giga_loss[loss=0.3205, simple_loss=0.3776, pruned_loss=0.1317, over 28660.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3755, pruned_loss=0.1258, over 5668708.49 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3668, pruned_loss=0.1191, over 5667664.45 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3767, pruned_loss=0.1267, over 5665000.94 frames. ], batch size: 336, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:43:31,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-08 02:43:43,323 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-08 02:43:48,575 INFO [train.py:968] (1/2) Epoch 15, batch 43950, giga_loss[loss=0.2958, simple_loss=0.3651, pruned_loss=0.1133, over 28741.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3744, pruned_loss=0.1257, over 5665832.60 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3664, pruned_loss=0.1187, over 5667493.23 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5662389.24 frames. ], batch size: 60, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:43:58,473 INFO [optim.py:369] (1/2) 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 02:44:10,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 02:44:41,080 INFO [train.py:968] (1/2) Epoch 15, batch 44000, giga_loss[loss=0.2885, simple_loss=0.3576, pruned_loss=0.1097, over 29047.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3757, pruned_loss=0.1267, over 5676721.09 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1192, over 5670460.74 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3768, pruned_loss=0.1276, over 5671660.21 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:44:55,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=682931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:45:12,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4832, 4.3263, 4.1022, 2.0869], device='cuda:1'), covar=tensor([0.0565, 0.0676, 0.0695, 0.1894], device='cuda:1'), in_proj_covar=tensor([0.1149, 0.1062, 0.0913, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 02:45:32,865 INFO [train.py:968] (1/2) Epoch 15, batch 44050, giga_loss[loss=0.322, simple_loss=0.3732, pruned_loss=0.1354, over 28928.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5668311.01 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3671, pruned_loss=0.1191, over 5671806.40 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3775, pruned_loss=0.1291, over 5662817.87 frames. ], batch size: 106, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:45:37,694 INFO [optim.py:369] (1/2) 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 02:46:20,138 INFO [train.py:968] (1/2) Epoch 15, batch 44100, giga_loss[loss=0.3087, simple_loss=0.3718, pruned_loss=0.1229, over 28873.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3745, pruned_loss=0.1274, over 5673931.66 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3668, pruned_loss=0.119, over 5674015.31 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3756, pruned_loss=0.1283, over 5667633.32 frames. ], batch size: 227, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:46:25,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:46:43,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3947, 1.3988, 3.8533, 3.1067], device='cuda:1'), covar=tensor([0.1579, 0.2575, 0.0448, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0712, 0.0618, 0.0912, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 02:46:49,863 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 44150, giga_loss[loss=0.2944, simple_loss=0.3647, pruned_loss=0.1121, over 28864.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3735, pruned_loss=0.1257, over 5676537.47 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3668, pruned_loss=0.1188, over 5677462.35 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3745, pruned_loss=0.1268, over 5668275.52 frames. ], batch size: 174, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:47:13,098 INFO [optim.py:369] (1/2) 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 02:48:00,384 INFO [train.py:968] (1/2) Epoch 15, batch 44200, giga_loss[loss=0.3165, simple_loss=0.3838, pruned_loss=0.1246, over 28794.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.376, pruned_loss=0.1269, over 5667491.03 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.367, pruned_loss=0.119, over 5671020.79 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3767, pruned_loss=0.1277, over 5665991.94 frames. ], batch size: 243, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:48:08,911 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 02:48:45,994 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:48:48,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:48:49,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:48:49,359 INFO [train.py:968] (1/2) Epoch 15, batch 44250, giga_loss[loss=0.3004, simple_loss=0.3667, pruned_loss=0.117, over 28254.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3763, pruned_loss=0.1271, over 5678296.19 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3671, pruned_loss=0.1191, over 5677155.87 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3771, pruned_loss=0.1278, over 5671387.86 frames. ], batch size: 368, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:48:54,138 INFO [optim.py:369] (1/2) 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 02:48:56,652 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:49:16,026 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 44300, giga_loss[loss=0.2942, simple_loss=0.3673, pruned_loss=0.1106, over 28716.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3762, pruned_loss=0.1271, over 5666557.97 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3667, pruned_loss=0.1186, over 5677026.02 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3776, pruned_loss=0.1285, over 5660913.42 frames. ], batch size: 284, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:50:13,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-08 02:50:20,273 INFO [train.py:968] (1/2) Epoch 15, batch 44350, giga_loss[loss=0.2794, simple_loss=0.3585, pruned_loss=0.1002, over 28923.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3779, pruned_loss=0.126, over 5665151.24 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3667, pruned_loss=0.1187, over 5673944.12 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3793, pruned_loss=0.1272, over 5664084.86 frames. ], batch size: 227, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:50:23,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4779, 3.4811, 1.5896, 1.6474], device='cuda:1'), covar=tensor([0.0912, 0.0402, 0.0899, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0532, 0.0358, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 02:50:25,338 INFO [optim.py:369] (1/2) 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 02:50:25,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3465, 1.7684, 1.3810, 1.5872], device='cuda:1'), covar=tensor([0.0825, 0.0305, 0.0354, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0097], device='cuda:1')
+2023-03-08 02:50:49,147 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:50:53,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:50:57,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:51:02,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9999, 1.3438, 1.2563, 1.1240], device='cuda:1'), covar=tensor([0.1182, 0.0902, 0.1581, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0735, 0.0686, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 02:51:02,922 INFO [train.py:968] (1/2) Epoch 15, batch 44400, giga_loss[loss=0.3675, simple_loss=0.4214, pruned_loss=0.1568, over 28710.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.379, pruned_loss=0.1244, over 5676248.72 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3668, pruned_loss=0.1188, over 5671488.31 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3804, pruned_loss=0.1255, over 5677720.82 frames. ], batch size: 262, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:51:24,756 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:51:51,931 INFO [train.py:968] (1/2) Epoch 15, batch 44450, giga_loss[loss=0.4025, simple_loss=0.432, pruned_loss=0.1865, over 27555.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.381, pruned_loss=0.1249, over 5671866.40 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1192, over 5666135.68 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3822, pruned_loss=0.1256, over 5678136.97 frames. ], batch size: 472, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:51:58,969 INFO [optim.py:369] (1/2) 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 02:52:09,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2630, 1.3787, 1.3944, 1.2448], device='cuda:1'), covar=tensor([0.2412, 0.2053, 0.1661, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.1837, 0.1758, 0.1687, 0.1833], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 02:52:36,428 INFO [train.py:968] (1/2) Epoch 15, batch 44500, giga_loss[loss=0.2897, simple_loss=0.3638, pruned_loss=0.1078, over 28999.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3827, pruned_loss=0.1274, over 5675811.23 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3668, pruned_loss=0.1193, over 5671443.23 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3846, pruned_loss=0.1281, over 5676545.33 frames. ], batch size: 164, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:52:40,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:53:07,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 02:53:09,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:53:11,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:53:20,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6770, 1.8208, 1.5945, 1.8348], device='cuda:1'), covar=tensor([0.1957, 0.1879, 0.1815, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.1396, 0.1021, 0.1240, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 02:53:31,164 INFO [train.py:968] (1/2) Epoch 15, batch 44550, giga_loss[loss=0.3263, simple_loss=0.382, pruned_loss=0.1353, over 28150.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3851, pruned_loss=0.1309, over 5646266.27 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3667, pruned_loss=0.1194, over 5663953.94 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3868, pruned_loss=0.1314, over 5653004.78 frames. ], batch size: 77, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:53:31,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3786, 1.9644, 1.4123, 0.5398], device='cuda:1'), covar=tensor([0.4130, 0.2470, 0.3572, 0.5305], device='cuda:1'), in_proj_covar=tensor([0.1648, 0.1580, 0.1538, 0.1351], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 02:53:39,958 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:54:18,804 INFO [train.py:968] (1/2) Epoch 15, batch 44600, giga_loss[loss=0.3303, simple_loss=0.3876, pruned_loss=0.1365, over 28856.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.386, pruned_loss=0.1323, over 5653724.88 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3669, pruned_loss=0.1195, over 5668532.37 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3876, pruned_loss=0.1329, over 5654391.61 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:54:28,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-08 02:54:51,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:54:59,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:55:01,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:55:02,182 INFO [train.py:968] (1/2) Epoch 15, batch 44650, giga_loss[loss=0.3259, simple_loss=0.3871, pruned_loss=0.1323, over 28987.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3838, pruned_loss=0.13, over 5664710.22 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3667, pruned_loss=0.1193, over 5677217.73 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.386, pruned_loss=0.1312, over 5656776.64 frames. ], batch size: 227, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:55:09,889 INFO [optim.py:369] (1/2) 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 02:55:27,742 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=683597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:55:48,964 INFO [train.py:968] (1/2) Epoch 15, batch 44700, libri_loss[loss=0.3541, simple_loss=0.4119, pruned_loss=0.1482, over 28620.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3828, pruned_loss=0.1272, over 5661213.38 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3669, pruned_loss=0.1192, over 5671842.50 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3848, pruned_loss=0.1283, over 5659378.69 frames. ], batch size: 106, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:55:53,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:56:02,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3811, 1.4640, 1.4657, 1.3428], device='cuda:1'), covar=tensor([0.2000, 0.1755, 0.1726, 0.1710], device='cuda:1'), in_proj_covar=tensor([0.1847, 0.1768, 0.1695, 0.1841], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 02:56:13,504 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:56:30,528 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=683660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:56:37,390 INFO [train.py:968] (1/2) Epoch 15, batch 44750, giga_loss[loss=0.3441, simple_loss=0.405, pruned_loss=0.1416, over 28948.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3832, pruned_loss=0.126, over 5666231.59 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.367, pruned_loss=0.1192, over 5674500.88 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3849, pruned_loss=0.127, over 5662179.98 frames. ], batch size: 213, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:56:44,797 INFO [optim.py:369] (1/2) 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 02:56:46,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 02:57:06,851 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=683697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:57:10,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=683700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 02:57:26,285 INFO [train.py:968] (1/2) Epoch 15, batch 44800, giga_loss[loss=0.3034, simple_loss=0.3706, pruned_loss=0.1181, over 28597.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3829, pruned_loss=0.1262, over 5671008.26 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3667, pruned_loss=0.119, over 5676951.51 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.385, pruned_loss=0.1275, over 5665386.50 frames. ], batch size: 307, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:57:37,031 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 44850, giga_loss[loss=0.3138, simple_loss=0.3817, pruned_loss=0.1229, over 28774.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3812, pruned_loss=0.1257, over 5681318.94 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.366, pruned_loss=0.1186, over 5682637.62 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3839, pruned_loss=0.1273, over 5671525.03 frames. ], batch size: 284, lr: 2.11e-03, grad_scale: 4.0
+2023-03-08 02:58:16,849 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-08 02:58:21,456 INFO [optim.py:369] (1/2) 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 02:58:27,185 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 15, batch 44900, giga_loss[loss=0.3192, simple_loss=0.3602, pruned_loss=0.1391, over 23391.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3808, pruned_loss=0.1267, over 5616705.00 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3665, pruned_loss=0.1191, over 5630357.86 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3829, pruned_loss=0.1277, over 5656729.92 frames. ], batch size: 705, lr: 2.11e-03, grad_scale: 2.0
+2023-03-08 02:59:51,025 INFO [train.py:968] (1/2) Epoch 15, batch 44950, giga_loss[loss=0.3028, simple_loss=0.3658, pruned_loss=0.1199, over 28824.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3797, pruned_loss=0.1274, over 5582232.29 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3675, pruned_loss=0.1201, over 5588238.84 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3807, pruned_loss=0.1274, over 5652000.40 frames. ], batch size: 119, lr: 2.11e-03, grad_scale: 1.0
+2023-03-08 02:59:58,720 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 15, batch 45000, giga_loss[loss=0.275, simple_loss=0.3526, pruned_loss=0.09867, over 28869.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3771, pruned_loss=0.1261, over 5556885.93 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3682, pruned_loss=0.1206, over 5546348.59 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3775, pruned_loss=0.1257, over 5650287.12 frames. ], batch size: 186, lr: 2.11e-03, grad_scale: 1.0
+2023-03-08 03:00:37,611 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 03:00:46,434 INFO [train.py:1012] (1/2) Epoch 15, validation: loss=0.2109, simple_loss=0.3196, pruned_loss=0.05111, over 944034.00 frames. 
+2023-03-08 03:00:46,435 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 03:01:19,676 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-08 03:02:07,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1184, 3.8992, 3.7206, 1.8567], device='cuda:1'), covar=tensor([0.0618, 0.0801, 0.0750, 0.2316], device='cuda:1'), in_proj_covar=tensor([0.1163, 0.1076, 0.0922, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 03:02:19,216 INFO [optim.py:369] (1/2) 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 03:02:39,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3151, 2.5751, 2.3989, 2.0481], device='cuda:1'), covar=tensor([0.2757, 0.2129, 0.2069, 0.2500], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1782, 0.1712, 0.1861], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 03:02:41,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=683998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:02:43,357 INFO [train.py:968] (1/2) Epoch 16, batch 50, giga_loss[loss=0.314, simple_loss=0.3891, pruned_loss=0.1194, over 28692.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3735, pruned_loss=0.1094, over 1257523.53 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3481, pruned_loss=0.09454, over 189585.13 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3774, pruned_loss=0.1117, over 1104715.67 frames. ], batch size: 262, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:03:02,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:03:03,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1460, 1.3418, 1.2257, 1.0664], device='cuda:1'), covar=tensor([0.1917, 0.1862, 0.1363, 0.1772], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1783, 0.1714, 0.1863], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 03:03:19,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:03:25,701 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:03:34,322 INFO [train.py:968] (1/2) Epoch 16, batch 100, giga_loss[loss=0.2416, simple_loss=0.3279, pruned_loss=0.0777, over 28593.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3669, pruned_loss=0.1071, over 2235149.48 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3511, pruned_loss=0.09496, over 331274.45 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3692, pruned_loss=0.1088, over 2020233.61 frames. ], batch size: 307, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:03:39,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5497, 1.6593, 1.6660, 1.4239], device='cuda:1'), covar=tensor([0.2607, 0.2346, 0.1875, 0.2240], device='cuda:1'), in_proj_covar=tensor([0.1861, 0.1787, 0.1718, 0.1865], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 03:04:00,637 INFO [optim.py:369] (1/2) 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 03:04:22,243 INFO [train.py:968] (1/2) Epoch 16, batch 150, giga_loss[loss=0.2441, simple_loss=0.3229, pruned_loss=0.08269, over 28640.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3508, pruned_loss=0.09905, over 3004022.27 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3489, pruned_loss=0.09399, over 414551.46 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3516, pruned_loss=0.09989, over 2790548.87 frames. ], batch size: 242, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:04:50,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3970, 1.1095, 4.3749, 3.4738], device='cuda:1'), covar=tensor([0.1667, 0.2986, 0.0380, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0617, 0.0914, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 03:04:56,395 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 200, libri_loss[loss=0.2337, simple_loss=0.3083, pruned_loss=0.07957, over 29503.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3372, pruned_loss=0.0925, over 3610159.39 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3395, pruned_loss=0.08919, over 683049.66 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.338, pruned_loss=0.09364, over 3318167.43 frames. ], batch size: 70, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:05:08,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:11,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:13,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:21,078 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:22,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4318, 1.5279, 1.3727, 1.5403], device='cuda:1'), covar=tensor([0.0749, 0.0343, 0.0325, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 03:05:24,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:25,255 INFO [optim.py:369] (1/2) 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 03:05:28,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:38,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:05:46,532 INFO [train.py:968] (1/2) Epoch 16, batch 250, giga_loss[loss=0.2109, simple_loss=0.2752, pruned_loss=0.07331, over 23613.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3255, pruned_loss=0.08678, over 4071146.51 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3383, pruned_loss=0.08811, over 761233.84 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3254, pruned_loss=0.08741, over 3812955.24 frames. ], batch size: 705, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:05:55,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-08 03:05:55,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:06:30,789 INFO [train.py:968] (1/2) Epoch 16, batch 300, giga_loss[loss=0.2257, simple_loss=0.2983, pruned_loss=0.0766, over 28790.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3172, pruned_loss=0.08306, over 4429257.54 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3382, pruned_loss=0.08783, over 911424.58 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.316, pruned_loss=0.08323, over 4179882.28 frames. ], batch size: 262, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:06:53,688 INFO [optim.py:369] (1/2) 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 03:07:16,894 INFO [train.py:968] (1/2) Epoch 16, batch 350, giga_loss[loss=0.2277, simple_loss=0.2898, pruned_loss=0.08277, over 27660.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3111, pruned_loss=0.08041, over 4707743.23 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3374, pruned_loss=0.08709, over 1059421.45 frames. ], giga_tot_loss[loss=0.2349, simple_loss=0.3092, pruned_loss=0.08034, over 4472114.70 frames. ], batch size: 472, lr: 2.04e-03, grad_scale: 1.0
+2023-03-08 03:07:18,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:07:19,455 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 03:07:20,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:07:43,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:07:57,767 INFO [train.py:968] (1/2) Epoch 16, batch 400, giga_loss[loss=0.2136, simple_loss=0.2881, pruned_loss=0.06954, over 29014.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3067, pruned_loss=0.07823, over 4929474.23 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.339, pruned_loss=0.08797, over 1174015.39 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.304, pruned_loss=0.07769, over 4722052.69 frames. ], batch size: 106, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:08:01,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:08:20,129 INFO [optim.py:369] (1/2) 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 03:08:35,725 INFO [train.py:968] (1/2) Epoch 16, batch 450, giga_loss[loss=0.2486, simple_loss=0.3208, pruned_loss=0.0882, over 28813.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.308, pruned_loss=0.07924, over 5108538.52 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3402, pruned_loss=0.08919, over 1470168.40 frames. ], giga_tot_loss[loss=0.2296, simple_loss=0.3034, pruned_loss=0.07788, over 4886133.38 frames. ], batch size: 186, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:08:37,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5586, 1.7445, 1.4952, 1.7272], device='cuda:1'), covar=tensor([0.2609, 0.2569, 0.2757, 0.2327], device='cuda:1'), in_proj_covar=tensor([0.1408, 0.1031, 0.1250, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 03:08:48,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:09:17,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8182, 4.6378, 4.4008, 1.9953], device='cuda:1'), covar=tensor([0.0581, 0.0720, 0.0854, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.1137, 0.1052, 0.0900, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 03:09:18,052 INFO [train.py:968] (1/2) Epoch 16, batch 500, libri_loss[loss=0.2517, simple_loss=0.3359, pruned_loss=0.08376, over 29531.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3064, pruned_loss=0.07854, over 5233028.05 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3413, pruned_loss=0.08945, over 1640725.10 frames. ], giga_tot_loss[loss=0.2274, simple_loss=0.301, pruned_loss=0.07691, over 5030043.73 frames. ], batch size: 82, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:09:43,883 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 550, giga_loss[loss=0.2207, simple_loss=0.302, pruned_loss=0.06969, over 28213.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3033, pruned_loss=0.07702, over 5336717.92 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3402, pruned_loss=0.08868, over 1746316.68 frames. ], giga_tot_loss[loss=0.2248, simple_loss=0.2984, pruned_loss=0.07564, over 5160852.92 frames. ], batch size: 368, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:10:07,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 03:10:50,492 INFO [train.py:968] (1/2) Epoch 16, batch 600, giga_loss[loss=0.2128, simple_loss=0.288, pruned_loss=0.0688, over 28532.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3011, pruned_loss=0.07605, over 5414066.28 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.341, pruned_loss=0.08898, over 1808245.64 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2964, pruned_loss=0.07468, over 5268222.94 frames. ], batch size: 307, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:10:58,164 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:11:00,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:11:19,487 INFO [optim.py:369] (1/2) 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 03:11:28,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:11:30,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4296, 1.6057, 1.6361, 1.5275], device='cuda:1'), covar=tensor([0.1597, 0.1866, 0.1699, 0.1650], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0738, 0.0688, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 03:11:37,105 INFO [train.py:968] (1/2) Epoch 16, batch 650, giga_loss[loss=0.1957, simple_loss=0.2672, pruned_loss=0.06206, over 28207.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2983, pruned_loss=0.0746, over 5478072.29 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3391, pruned_loss=0.08813, over 1925773.82 frames. ], giga_tot_loss[loss=0.2203, simple_loss=0.2939, pruned_loss=0.07336, over 5352583.61 frames. ], batch size: 77, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:11:48,017 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-08 03:12:21,229 INFO [train.py:968] (1/2) Epoch 16, batch 700, giga_loss[loss=0.2332, simple_loss=0.2891, pruned_loss=0.08868, over 23858.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2962, pruned_loss=0.07342, over 5528691.89 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3378, pruned_loss=0.08756, over 2058704.30 frames. ], giga_tot_loss[loss=0.218, simple_loss=0.2917, pruned_loss=0.07214, over 5418894.63 frames. ], batch size: 705, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:12:49,120 INFO [optim.py:369] (1/2) 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 03:12:49,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684679.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 03:13:06,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1375, 3.9333, 3.7054, 1.9401], device='cuda:1'), covar=tensor([0.0593, 0.0806, 0.0783, 0.2180], device='cuda:1'), in_proj_covar=tensor([0.1130, 0.1047, 0.0896, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 03:13:06,709 INFO [train.py:968] (1/2) Epoch 16, batch 750, giga_loss[loss=0.1957, simple_loss=0.2723, pruned_loss=0.05952, over 28635.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2939, pruned_loss=0.07245, over 5567007.35 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3374, pruned_loss=0.0875, over 2135041.15 frames. ], giga_tot_loss[loss=0.216, simple_loss=0.2897, pruned_loss=0.07117, over 5473514.91 frames. ], batch size: 307, lr: 2.04e-03, grad_scale: 2.0
+2023-03-08 03:13:37,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=684730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:13:55,451 INFO [train.py:968] (1/2) Epoch 16, batch 800, giga_loss[loss=0.1946, simple_loss=0.2694, pruned_loss=0.05996, over 28199.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2912, pruned_loss=0.07186, over 5589587.63 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3379, pruned_loss=0.08775, over 2172314.42 frames. ], giga_tot_loss[loss=0.2142, simple_loss=0.2872, pruned_loss=0.07061, over 5513411.68 frames. ], batch size: 77, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:14:18,564 INFO [optim.py:369] (1/2) 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 03:14:18,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9548, 2.1335, 2.2618, 1.7521], device='cuda:1'), covar=tensor([0.1875, 0.2186, 0.1425, 0.1612], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0701, 0.0922, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 03:14:41,535 INFO [train.py:968] (1/2) Epoch 16, batch 850, giga_loss[loss=0.2399, simple_loss=0.3182, pruned_loss=0.08082, over 28892.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2962, pruned_loss=0.07426, over 5615987.19 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3391, pruned_loss=0.08832, over 2282911.54 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2914, pruned_loss=0.07264, over 5546810.07 frames. ], batch size: 186, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:14:52,751 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:15:06,709 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684822.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 03:15:08,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684825.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 03:15:32,833 INFO [train.py:968] (1/2) Epoch 16, batch 900, giga_loss[loss=0.3563, simple_loss=0.4045, pruned_loss=0.154, over 26687.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.31, pruned_loss=0.08136, over 5630145.69 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3384, pruned_loss=0.08796, over 2371176.43 frames. ], giga_tot_loss[loss=0.2329, simple_loss=0.3057, pruned_loss=0.08004, over 5569864.95 frames. ], batch size: 555, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:15:33,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6433, 1.8405, 1.3495, 1.3945], device='cuda:1'), covar=tensor([0.0806, 0.0488, 0.0959, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0440, 0.0502, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:15:36,256 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684854.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 03:15:36,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-08 03:15:53,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=684873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:15:56,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=684876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:15:58,375 INFO [optim.py:369] (1/2) 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 03:15:58,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:16:11,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3090, 2.6789, 1.4219, 1.4162], device='cuda:1'), covar=tensor([0.0915, 0.0333, 0.0867, 0.1304], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0527, 0.0357, 0.0404], device='cuda:1'), out_proj_covar=tensor([0.0024, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 03:16:16,871 INFO [train.py:968] (1/2) Epoch 16, batch 950, giga_loss[loss=0.2913, simple_loss=0.3649, pruned_loss=0.1088, over 28399.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3218, pruned_loss=0.08727, over 5650025.17 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3375, pruned_loss=0.08755, over 2491888.55 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3181, pruned_loss=0.08632, over 5594018.25 frames. ], batch size: 65, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:16:20,033 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=684905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:16:49,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 03:16:56,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2942, 1.5596, 1.2263, 0.9814], device='cuda:1'), covar=tensor([0.2634, 0.2654, 0.2983, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.1403, 0.1026, 0.1242, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 03:16:58,888 INFO [train.py:968] (1/2) Epoch 16, batch 1000, giga_loss[loss=0.2666, simple_loss=0.3461, pruned_loss=0.09354, over 29061.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.331, pruned_loss=0.09164, over 5661068.82 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3371, pruned_loss=0.08738, over 2559754.30 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3281, pruned_loss=0.09103, over 5612114.05 frames. ], batch size: 155, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:17:24,943 INFO [optim.py:369] (1/2) 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 03:17:35,586 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=684992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:17:43,265 INFO [train.py:968] (1/2) Epoch 16, batch 1050, giga_loss[loss=0.2562, simple_loss=0.34, pruned_loss=0.08621, over 28958.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3351, pruned_loss=0.09216, over 5668132.66 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3369, pruned_loss=0.08728, over 2576504.52 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.333, pruned_loss=0.09175, over 5628709.75 frames. ], batch size: 145, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:18:30,735 INFO [train.py:968] (1/2) Epoch 16, batch 1100, giga_loss[loss=0.2275, simple_loss=0.3072, pruned_loss=0.07391, over 28314.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3373, pruned_loss=0.09261, over 5663950.39 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3368, pruned_loss=0.08719, over 2656308.42 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3357, pruned_loss=0.09245, over 5629301.21 frames. ], batch size: 65, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:18:56,298 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 1150, giga_loss[loss=0.2496, simple_loss=0.3281, pruned_loss=0.08554, over 28971.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3412, pruned_loss=0.09498, over 5663867.72 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.337, pruned_loss=0.0873, over 2672289.88 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3398, pruned_loss=0.09484, over 5635890.32 frames. ], batch size: 200, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:19:20,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5405, 1.6114, 1.5583, 1.4144], device='cuda:1'), covar=tensor([0.2577, 0.2461, 0.1856, 0.2269], device='cuda:1'), in_proj_covar=tensor([0.1836, 0.1762, 0.1699, 0.1845], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 03:20:03,380 INFO [train.py:968] (1/2) Epoch 16, batch 1200, giga_loss[loss=0.2793, simple_loss=0.3588, pruned_loss=0.09986, over 28770.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.344, pruned_loss=0.09724, over 5676716.67 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3371, pruned_loss=0.08721, over 2752531.28 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.343, pruned_loss=0.09738, over 5649011.30 frames. ], batch size: 99, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:20:27,540 INFO [optim.py:369] (1/2) 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 03:20:33,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=685184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:20:46,753 INFO [train.py:968] (1/2) Epoch 16, batch 1250, libri_loss[loss=0.2663, simple_loss=0.3531, pruned_loss=0.08973, over 29668.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3474, pruned_loss=0.0994, over 5684419.01 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3371, pruned_loss=0.08715, over 2860792.58 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3469, pruned_loss=0.09995, over 5655171.15 frames. ], batch size: 88, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:21:09,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1559, 3.0196, 1.3662, 1.2939], device='cuda:1'), covar=tensor([0.1087, 0.0320, 0.0967, 0.1535], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0525, 0.0357, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 03:21:30,025 INFO [train.py:968] (1/2) Epoch 16, batch 1300, giga_loss[loss=0.2715, simple_loss=0.3556, pruned_loss=0.09368, over 28743.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3502, pruned_loss=0.09966, over 5696617.53 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3378, pruned_loss=0.08745, over 2934435.23 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3497, pruned_loss=0.1002, over 5669904.77 frames. ], batch size: 262, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:21:34,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=685254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:21:34,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4658, 3.2728, 3.0965, 1.8226], device='cuda:1'), covar=tensor([0.0727, 0.0909, 0.0826, 0.1936], device='cuda:1'), in_proj_covar=tensor([0.1127, 0.1036, 0.0890, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 03:21:52,758 INFO [optim.py:369] (1/2) 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 03:22:10,130 INFO [train.py:968] (1/2) Epoch 16, batch 1350, giga_loss[loss=0.2881, simple_loss=0.366, pruned_loss=0.1051, over 28886.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3528, pruned_loss=0.1007, over 5693284.35 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3389, pruned_loss=0.08777, over 3049557.46 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3524, pruned_loss=0.1015, over 5665576.82 frames. ], batch size: 112, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:22:33,603 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=685327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:22:35,607 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 1400, giga_loss[loss=0.2748, simple_loss=0.3533, pruned_loss=0.09818, over 28423.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3542, pruned_loss=0.1007, over 5696921.52 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3388, pruned_loss=0.08775, over 3104122.44 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3542, pruned_loss=0.1016, over 5673003.98 frames. ], batch size: 71, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:23:01,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=685359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:08,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=685367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:17,082 INFO [optim.py:369] (1/2) 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 03:23:34,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=685397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:36,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=685400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:23:36,963 INFO [train.py:968] (1/2) Epoch 16, batch 1450, giga_loss[loss=0.2636, simple_loss=0.3442, pruned_loss=0.09152, over 28221.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3536, pruned_loss=0.09906, over 5700141.63 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3393, pruned_loss=0.08807, over 3142217.66 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3536, pruned_loss=0.09975, over 5681898.17 frames. ], batch size: 77, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:24:00,326 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=685429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:24:12,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5204, 1.5810, 1.1665, 1.1969], device='cuda:1'), covar=tensor([0.0805, 0.0511, 0.0956, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0444, 0.0509, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 03:24:18,461 INFO [train.py:968] (1/2) Epoch 16, batch 1500, giga_loss[loss=0.2657, simple_loss=0.3494, pruned_loss=0.09101, over 28921.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3527, pruned_loss=0.09734, over 5703533.27 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3398, pruned_loss=0.08823, over 3183109.49 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3526, pruned_loss=0.09793, over 5686628.46 frames. ], batch size: 112, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:24:32,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6610, 1.8776, 1.7046, 1.5391], device='cuda:1'), covar=tensor([0.1696, 0.1878, 0.2181, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0735, 0.0687, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 03:24:43,253 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.6848, 2.2998, 1.7614, 0.8525], device='cuda:1'), covar=tensor([0.5111, 0.2469, 0.3763, 0.5823], device='cuda:1'), in_proj_covar=tensor([0.1636, 0.1557, 0.1536, 0.1338], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 03:25:01,470 INFO [train.py:968] (1/2) Epoch 16, batch 1550, giga_loss[loss=0.2688, simple_loss=0.3531, pruned_loss=0.09229, over 28585.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3517, pruned_loss=0.09616, over 5713052.31 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3403, pruned_loss=0.08849, over 3222890.47 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3516, pruned_loss=0.0966, over 5697885.57 frames. ], batch size: 307, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:25:09,421 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=685510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:25:12,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=685513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:25:37,973 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=685542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:25:47,021 INFO [train.py:968] (1/2) Epoch 16, batch 1600, giga_loss[loss=0.2975, simple_loss=0.3588, pruned_loss=0.1181, over 28931.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.353, pruned_loss=0.09826, over 5697506.02 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3405, pruned_loss=0.08855, over 3287180.80 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.353, pruned_loss=0.09876, over 5683123.69 frames. ], batch size: 136, lr: 2.04e-03, grad_scale: 8.0
+2023-03-08 03:25:49,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3334, 1.9948, 1.4801, 0.5872], device='cuda:1'), covar=tensor([0.4240, 0.2213, 0.2981, 0.4853], device='cuda:1'), in_proj_covar=tensor([0.1641, 0.1561, 0.1541, 0.1340], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 03:26:11,181 INFO [optim.py:369] (1/2) 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 03:26:16,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-08 03:26:21,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4084, 1.5661, 1.5955, 1.4206], device='cuda:1'), covar=tensor([0.1563, 0.1669, 0.1977, 0.1714], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0730, 0.0682, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 03:26:30,863 INFO [train.py:968] (1/2) Epoch 16, batch 1650, giga_loss[loss=0.3505, simple_loss=0.4063, pruned_loss=0.1474, over 28796.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3555, pruned_loss=0.102, over 5705371.27 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3404, pruned_loss=0.08841, over 3339310.53 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3558, pruned_loss=0.1026, over 5690256.03 frames. ], batch size: 119, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:27:19,634 INFO [train.py:968] (1/2) Epoch 16, batch 1700, giga_loss[loss=0.3169, simple_loss=0.3791, pruned_loss=0.1273, over 28976.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3575, pruned_loss=0.105, over 5704254.68 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3404, pruned_loss=0.0884, over 3367660.74 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3579, pruned_loss=0.1058, over 5699279.50 frames. ], batch size: 227, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:27:19,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4271, 2.6694, 2.5276, 1.8521], device='cuda:1'), covar=tensor([0.0834, 0.0203, 0.0211, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 03:27:46,251 INFO [optim.py:369] (1/2) 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 03:27:52,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8655, 1.6436, 5.2853, 3.7170], device='cuda:1'), covar=tensor([0.1634, 0.2658, 0.0331, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0702, 0.0612, 0.0902, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:28:04,726 INFO [train.py:968] (1/2) Epoch 16, batch 1750, giga_loss[loss=0.3487, simple_loss=0.3963, pruned_loss=0.1506, over 26609.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3565, pruned_loss=0.1058, over 5698801.57 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.34, pruned_loss=0.08814, over 3443213.16 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3576, pruned_loss=0.107, over 5688803.84 frames. ], batch size: 555, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:28:13,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5437, 1.7529, 1.5062, 1.7550], device='cuda:1'), covar=tensor([0.2603, 0.2490, 0.2722, 0.2154], device='cuda:1'), in_proj_covar=tensor([0.1402, 0.1026, 0.1239, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 03:28:46,582 INFO [train.py:968] (1/2) Epoch 16, batch 1800, giga_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09462, over 28712.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.354, pruned_loss=0.105, over 5694622.72 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3401, pruned_loss=0.08817, over 3491980.88 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.355, pruned_loss=0.1063, over 5683038.04 frames. ], batch size: 242, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:29:10,288 INFO [optim.py:369] (1/2) 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 03:29:30,741 INFO [train.py:968] (1/2) Epoch 16, batch 1850, giga_loss[loss=0.2509, simple_loss=0.3213, pruned_loss=0.0903, over 28665.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3517, pruned_loss=0.1031, over 5691250.44 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3396, pruned_loss=0.08785, over 3538162.67 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.353, pruned_loss=0.1046, over 5679964.87 frames. ], batch size: 92, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:29:47,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 03:30:18,189 INFO [train.py:968] (1/2) Epoch 16, batch 1900, giga_loss[loss=0.2405, simple_loss=0.3238, pruned_loss=0.07857, over 28910.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3491, pruned_loss=0.1007, over 5696064.08 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3396, pruned_loss=0.08773, over 3573237.21 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3503, pruned_loss=0.1021, over 5684489.97 frames. ], batch size: 199, lr: 2.04e-03, grad_scale: 4.0
+2023-03-08 03:30:49,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=685879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:30:49,576 INFO [optim.py:369] (1/2) 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 03:30:58,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=685889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:30:58,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 03:31:08,331 INFO [train.py:968] (1/2) Epoch 16, batch 1950, giga_loss[loss=0.2577, simple_loss=0.3231, pruned_loss=0.09617, over 27865.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3447, pruned_loss=0.09816, over 5690782.20 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3399, pruned_loss=0.08782, over 3606193.38 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3456, pruned_loss=0.09938, over 5680291.52 frames. ], batch size: 412, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:31:21,891 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-08 03:31:57,743 INFO [train.py:968] (1/2) Epoch 16, batch 2000, giga_loss[loss=0.2306, simple_loss=0.3058, pruned_loss=0.07773, over 28771.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3398, pruned_loss=0.09548, over 5686207.74 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3409, pruned_loss=0.08838, over 3693116.17 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3399, pruned_loss=0.09645, over 5672736.42 frames. ], batch size: 99, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:32:07,754 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=685960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:32:21,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3545, 3.4703, 1.5745, 1.4995], device='cuda:1'), covar=tensor([0.1019, 0.0247, 0.0912, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0524, 0.0356, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 03:32:23,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 03:32:25,466 INFO [optim.py:369] (1/2) 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 03:32:29,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3560, 1.4693, 1.3385, 1.2361], device='cuda:1'), covar=tensor([0.1952, 0.2009, 0.1561, 0.1950], device='cuda:1'), in_proj_covar=tensor([0.1829, 0.1746, 0.1688, 0.1835], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 03:32:45,866 INFO [train.py:968] (1/2) Epoch 16, batch 2050, giga_loss[loss=0.2246, simple_loss=0.3029, pruned_loss=0.07318, over 28857.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.334, pruned_loss=0.09214, over 5684747.51 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3415, pruned_loss=0.08884, over 3736046.09 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3337, pruned_loss=0.09273, over 5670947.83 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:33:39,497 INFO [train.py:968] (1/2) Epoch 16, batch 2100, giga_loss[loss=0.2468, simple_loss=0.3068, pruned_loss=0.09339, over 23568.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3321, pruned_loss=0.09164, over 5663389.75 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08919, over 3768586.60 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3314, pruned_loss=0.09194, over 5649253.39 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:34:03,865 INFO [optim.py:369] (1/2) 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 03:34:21,715 INFO [train.py:968] (1/2) Epoch 16, batch 2150, giga_loss[loss=0.234, simple_loss=0.3209, pruned_loss=0.07359, over 28875.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3333, pruned_loss=0.09141, over 5680248.89 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3416, pruned_loss=0.08894, over 3810372.49 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3328, pruned_loss=0.09184, over 5665434.92 frames. ], batch size: 174, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:34:44,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4793, 1.6838, 1.4482, 1.5497], device='cuda:1'), covar=tensor([0.0765, 0.0319, 0.0323, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 03:35:03,732 INFO [train.py:968] (1/2) Epoch 16, batch 2200, giga_loss[loss=0.2728, simple_loss=0.3426, pruned_loss=0.1015, over 28039.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3328, pruned_loss=0.09098, over 5690198.70 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.342, pruned_loss=0.08893, over 3852027.01 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.332, pruned_loss=0.09137, over 5674349.40 frames. ], batch size: 412, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:35:28,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5885, 1.8145, 1.4524, 1.6908], device='cuda:1'), covar=tensor([0.2470, 0.2546, 0.2836, 0.2409], device='cuda:1'), in_proj_covar=tensor([0.1396, 0.1021, 0.1233, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 03:35:29,608 INFO [optim.py:369] (1/2) 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 03:35:41,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-08 03:35:45,283 INFO [train.py:968] (1/2) Epoch 16, batch 2250, giga_loss[loss=0.3299, simple_loss=0.38, pruned_loss=0.1399, over 26768.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3315, pruned_loss=0.0905, over 5697516.83 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3414, pruned_loss=0.08863, over 3929255.03 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3309, pruned_loss=0.09107, over 5680351.25 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:36:07,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6344, 1.7904, 1.6573, 1.5820], device='cuda:1'), covar=tensor([0.1855, 0.2256, 0.2377, 0.2303], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0737, 0.0689, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 03:36:32,669 INFO [train.py:968] (1/2) Epoch 16, batch 2300, giga_loss[loss=0.2301, simple_loss=0.3073, pruned_loss=0.07645, over 28994.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3302, pruned_loss=0.09042, over 5704115.23 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08942, over 3949116.48 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3289, pruned_loss=0.0904, over 5688740.58 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:36:34,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:36:42,346 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 2350, giga_loss[loss=0.234, simple_loss=0.3118, pruned_loss=0.07812, over 28224.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3279, pruned_loss=0.08883, over 5709614.06 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3435, pruned_loss=0.08959, over 4006406.23 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3258, pruned_loss=0.08871, over 5692522.63 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:37:41,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:37:54,454 INFO [train.py:968] (1/2) Epoch 16, batch 2400, giga_loss[loss=0.2267, simple_loss=0.3077, pruned_loss=0.07283, over 28655.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3257, pruned_loss=0.08802, over 5708281.69 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3439, pruned_loss=0.08969, over 4043272.37 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3235, pruned_loss=0.08783, over 5692033.17 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:38:08,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:18,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5750, 1.7012, 1.6124, 1.4354], device='cuda:1'), covar=tensor([0.2928, 0.2378, 0.2047, 0.2451], device='cuda:1'), in_proj_covar=tensor([0.1821, 0.1738, 0.1684, 0.1828], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 03:38:18,908 INFO [optim.py:369] (1/2) 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 03:38:27,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9796, 1.1682, 3.2712, 2.8087], device='cuda:1'), covar=tensor([0.1718, 0.2726, 0.0479, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0702, 0.0612, 0.0896, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:38:30,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-08 03:38:30,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:33,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:33,623 INFO [train.py:968] (1/2) Epoch 16, batch 2450, giga_loss[loss=0.2181, simple_loss=0.2911, pruned_loss=0.07258, over 28839.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3242, pruned_loss=0.08718, over 5715107.91 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3444, pruned_loss=0.08968, over 4085867.17 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3216, pruned_loss=0.087, over 5700567.70 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:38:37,872 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:40,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:53,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:38:55,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:03,222 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:13,705 INFO [train.py:968] (1/2) Epoch 16, batch 2500, libri_loss[loss=0.2277, simple_loss=0.3164, pruned_loss=0.06945, over 29509.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3219, pruned_loss=0.08616, over 5722386.39 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3447, pruned_loss=0.08968, over 4131137.50 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3192, pruned_loss=0.08596, over 5706465.33 frames. ], batch size: 70, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:39:25,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-08 03:39:36,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:38,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:39:39,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5875, 1.3600, 4.9364, 3.4756], device='cuda:1'), covar=tensor([0.1655, 0.2780, 0.0335, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0615, 0.0901, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:39:40,183 INFO [optim.py:369] (1/2) 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 03:39:54,080 INFO [train.py:968] (1/2) Epoch 16, batch 2550, giga_loss[loss=0.228, simple_loss=0.3028, pruned_loss=0.07665, over 28809.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3203, pruned_loss=0.08527, over 5728993.70 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3454, pruned_loss=0.09018, over 4173825.33 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3172, pruned_loss=0.08467, over 5713338.24 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:40:02,845 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:40:35,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8890, 1.1171, 3.3833, 2.9086], device='cuda:1'), covar=tensor([0.1862, 0.2816, 0.0470, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0615, 0.0901, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:40:36,715 INFO [train.py:968] (1/2) Epoch 16, batch 2600, giga_loss[loss=0.2586, simple_loss=0.3304, pruned_loss=0.09341, over 28890.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.319, pruned_loss=0.0846, over 5728330.59 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3459, pruned_loss=0.09035, over 4207225.39 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3156, pruned_loss=0.08389, over 5713619.12 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:40:54,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9672, 1.1115, 3.6228, 3.0180], device='cuda:1'), covar=tensor([0.1939, 0.2947, 0.0452, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0615, 0.0901, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:40:56,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 03:41:01,798 INFO [optim.py:369] (1/2) 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 03:41:16,441 INFO [train.py:968] (1/2) Epoch 16, batch 2650, giga_loss[loss=0.2385, simple_loss=0.3092, pruned_loss=0.0839, over 28816.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3185, pruned_loss=0.08417, over 5730724.01 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3464, pruned_loss=0.09026, over 4262447.44 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3145, pruned_loss=0.08347, over 5716187.92 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:41:28,312 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:41:36,713 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:42:02,468 INFO [train.py:968] (1/2) Epoch 16, batch 2700, giga_loss[loss=0.303, simple_loss=0.3674, pruned_loss=0.1193, over 28613.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3225, pruned_loss=0.08675, over 5721523.40 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3466, pruned_loss=0.09024, over 4273775.73 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.319, pruned_loss=0.08618, over 5713218.63 frames. ], batch size: 336, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:42:28,883 INFO [optim.py:369] (1/2) 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 03:42:37,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4001, 1.5304, 1.4880, 1.4117], device='cuda:1'), covar=tensor([0.1471, 0.1763, 0.1856, 0.1634], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0734, 0.0689, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 03:42:46,757 INFO [train.py:968] (1/2) Epoch 16, batch 2750, giga_loss[loss=0.2594, simple_loss=0.3371, pruned_loss=0.09078, over 28977.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3273, pruned_loss=0.08932, over 5715458.12 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3465, pruned_loss=0.09006, over 4310580.02 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3241, pruned_loss=0.08892, over 5707975.69 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 03:43:28,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:43:35,180 INFO [train.py:968] (1/2) Epoch 16, batch 2800, giga_loss[loss=0.2907, simple_loss=0.3602, pruned_loss=0.1106, over 28260.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3356, pruned_loss=0.09483, over 5712518.92 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3466, pruned_loss=0.0903, over 4355919.39 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3325, pruned_loss=0.09437, over 5703254.68 frames. ], batch size: 65, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:43:41,207 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=686758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:43:56,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5138, 1.7997, 1.4206, 1.7144], device='cuda:1'), covar=tensor([0.2505, 0.2444, 0.2765, 0.2200], device='cuda:1'), in_proj_covar=tensor([0.1402, 0.1026, 0.1240, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 03:44:04,544 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 2850, giga_loss[loss=0.2476, simple_loss=0.3347, pruned_loss=0.08029, over 29153.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.342, pruned_loss=0.09862, over 5708773.54 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3473, pruned_loss=0.09071, over 4401139.32 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3388, pruned_loss=0.09815, over 5696963.18 frames. ], batch size: 155, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:44:25,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:45:09,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 03:45:14,423 INFO [train.py:968] (1/2) Epoch 16, batch 2900, giga_loss[loss=0.2901, simple_loss=0.3696, pruned_loss=0.1053, over 29031.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3464, pruned_loss=0.1, over 5711127.95 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3476, pruned_loss=0.09094, over 4416126.18 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3437, pruned_loss=0.09958, over 5700261.87 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:45:39,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5902, 4.3960, 4.1865, 2.0150], device='cuda:1'), covar=tensor([0.0529, 0.0689, 0.0681, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.1125, 0.1038, 0.0894, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 03:45:43,127 INFO [optim.py:369] (1/2) 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 03:45:47,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=686887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:45:48,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=686890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:45:58,583 INFO [train.py:968] (1/2) Epoch 16, batch 2950, giga_loss[loss=0.3238, simple_loss=0.3896, pruned_loss=0.129, over 28884.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3517, pruned_loss=0.103, over 5711012.93 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3474, pruned_loss=0.09091, over 4445828.51 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3497, pruned_loss=0.1028, over 5698890.14 frames. ], batch size: 112, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:46:20,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:46:46,850 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 3000, giga_loss[loss=0.2692, simple_loss=0.3509, pruned_loss=0.09373, over 28847.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3568, pruned_loss=0.1065, over 5689004.80 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.347, pruned_loss=0.09071, over 4487350.72 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3557, pruned_loss=0.1069, over 5675344.93 frames. ], batch size: 174, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:46:48,726 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 03:46:58,140 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 03:46:59,013 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=686981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:47:26,418 INFO [optim.py:369] (1/2) 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 03:47:33,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=686991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:47:41,940 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 3050, libri_loss[loss=0.2845, simple_loss=0.3692, pruned_loss=0.09992, over 29662.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3532, pruned_loss=0.1034, over 5696813.64 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3472, pruned_loss=0.09081, over 4501633.75 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3522, pruned_loss=0.1038, over 5683982.62 frames. ], batch size: 91, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:48:26,005 INFO [train.py:968] (1/2) Epoch 16, batch 3100, giga_loss[loss=0.2608, simple_loss=0.3541, pruned_loss=0.08372, over 28881.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3502, pruned_loss=0.1006, over 5705059.23 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3468, pruned_loss=0.09076, over 4522719.03 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3497, pruned_loss=0.1011, over 5692120.27 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:48:55,665 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 3150, giga_loss[loss=0.249, simple_loss=0.3301, pruned_loss=0.08399, over 29063.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3494, pruned_loss=0.09984, over 5709212.63 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3474, pruned_loss=0.09107, over 4548952.13 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3487, pruned_loss=0.1002, over 5696436.32 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:49:41,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=687133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:42,420 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=687134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:44,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=687137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:47,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=687142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:50,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=687145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:49:56,290 INFO [train.py:968] (1/2) Epoch 16, batch 3200, giga_loss[loss=0.3262, simple_loss=0.3815, pruned_loss=0.1355, over 26486.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3506, pruned_loss=0.09989, over 5713151.62 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3474, pruned_loss=0.09096, over 4568837.58 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3501, pruned_loss=0.1004, over 5700775.35 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:50:10,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=687166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:50:19,535 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=687174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:50:26,185 INFO [optim.py:369] (1/2) 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 03:50:41,409 INFO [train.py:968] (1/2) Epoch 16, batch 3250, giga_loss[loss=0.2912, simple_loss=0.3673, pruned_loss=0.1076, over 28284.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3518, pruned_loss=0.1005, over 5712835.02 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3472, pruned_loss=0.09089, over 4588486.17 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3516, pruned_loss=0.1011, over 5700616.94 frames. ], batch size: 368, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:51:11,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-08 03:51:26,409 INFO [train.py:968] (1/2) Epoch 16, batch 3300, giga_loss[loss=0.2985, simple_loss=0.371, pruned_loss=0.113, over 28896.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3527, pruned_loss=0.1017, over 5711510.29 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3467, pruned_loss=0.09069, over 4614307.31 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.353, pruned_loss=0.1025, over 5698615.12 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:51:47,306 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=687276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:51:49,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=687279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:51:53,022 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 3350, giga_loss[loss=0.2971, simple_loss=0.364, pruned_loss=0.1151, over 28639.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3541, pruned_loss=0.1029, over 5713216.47 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3469, pruned_loss=0.09077, over 4650354.53 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3544, pruned_loss=0.1038, over 5699631.43 frames. ], batch size: 92, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:52:13,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4714, 2.0946, 1.5662, 0.6655], device='cuda:1'), covar=tensor([0.5198, 0.2432, 0.3381, 0.5575], device='cuda:1'), in_proj_covar=tensor([0.1635, 0.1556, 0.1529, 0.1337], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 03:52:14,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=687308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:52:42,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2888, 2.6170, 1.3205, 1.4234], device='cuda:1'), covar=tensor([0.1008, 0.0376, 0.0852, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0526, 0.0357, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 03:52:53,626 INFO [train.py:968] (1/2) Epoch 16, batch 3400, giga_loss[loss=0.2665, simple_loss=0.3479, pruned_loss=0.09255, over 28997.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.355, pruned_loss=0.1037, over 5723724.17 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3469, pruned_loss=0.09088, over 4685407.95 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3554, pruned_loss=0.1047, over 5709680.44 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:53:21,965 INFO [optim.py:369] (1/2) 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 03:53:37,760 INFO [train.py:968] (1/2) Epoch 16, batch 3450, giga_loss[loss=0.2838, simple_loss=0.3622, pruned_loss=0.1027, over 28845.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3548, pruned_loss=0.1038, over 5727136.18 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3468, pruned_loss=0.09094, over 4703722.14 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3553, pruned_loss=0.1047, over 5713511.13 frames. ], batch size: 174, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:53:42,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9086, 1.2549, 1.3231, 1.1631], device='cuda:1'), covar=tensor([0.2019, 0.1364, 0.2246, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0734, 0.0687, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 03:54:17,486 INFO [train.py:968] (1/2) Epoch 16, batch 3500, giga_loss[loss=0.2741, simple_loss=0.3583, pruned_loss=0.09491, over 28609.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3551, pruned_loss=0.1031, over 5724183.84 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3472, pruned_loss=0.09114, over 4732406.14 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3555, pruned_loss=0.104, over 5710117.79 frames. ], batch size: 71, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:54:45,823 INFO [optim.py:369] (1/2) 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 03:55:02,876 INFO [train.py:968] (1/2) Epoch 16, batch 3550, giga_loss[loss=0.253, simple_loss=0.3386, pruned_loss=0.0837, over 29091.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3559, pruned_loss=0.1026, over 5723169.61 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3474, pruned_loss=0.09121, over 4742955.03 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3561, pruned_loss=0.1033, over 5711570.68 frames. ], batch size: 155, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:55:17,390 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=687517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:55:45,031 INFO [train.py:968] (1/2) Epoch 16, batch 3600, giga_loss[loss=0.2752, simple_loss=0.3606, pruned_loss=0.09494, over 28927.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3546, pruned_loss=0.1012, over 5722621.82 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3475, pruned_loss=0.09129, over 4752540.63 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3547, pruned_loss=0.1018, over 5713682.03 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 03:56:12,527 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 3650, giga_loss[loss=0.2619, simple_loss=0.3415, pruned_loss=0.09119, over 28278.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3527, pruned_loss=0.1004, over 5717658.70 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3472, pruned_loss=0.0912, over 4778759.12 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3532, pruned_loss=0.1013, over 5717462.96 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:56:51,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4044, 5.2128, 4.9448, 2.5130], device='cuda:1'), covar=tensor([0.0415, 0.0587, 0.0638, 0.1778], device='cuda:1'), in_proj_covar=tensor([0.1123, 0.1039, 0.0893, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 03:57:01,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=687642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:57:03,584 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=687645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 03:57:08,068 INFO [train.py:968] (1/2) Epoch 16, batch 3700, giga_loss[loss=0.2646, simple_loss=0.3361, pruned_loss=0.09653, over 28966.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3517, pruned_loss=0.1003, over 5719342.20 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3475, pruned_loss=0.09137, over 4811319.31 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.352, pruned_loss=0.1011, over 5714803.83 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:57:33,425 INFO [optim.py:369] (1/2) 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 03:57:35,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3765, 1.8940, 1.3509, 0.6025], device='cuda:1'), covar=tensor([0.4426, 0.2134, 0.2928, 0.5364], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1536, 0.1515, 0.1327], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 03:57:43,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8776, 2.0100, 1.4382, 1.4657], device='cuda:1'), covar=tensor([0.0858, 0.0614, 0.1037, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0439, 0.0505, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:57:47,173 INFO [train.py:968] (1/2) Epoch 16, batch 3750, giga_loss[loss=0.2796, simple_loss=0.3534, pruned_loss=0.1029, over 28767.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3502, pruned_loss=0.0996, over 5726468.19 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3474, pruned_loss=0.09135, over 4832275.32 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5720299.57 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:58:13,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5877, 1.2679, 5.0204, 3.6706], device='cuda:1'), covar=tensor([0.1704, 0.2863, 0.0358, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0699, 0.0606, 0.0891, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:58:25,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2335, 1.6096, 1.5459, 1.5952], device='cuda:1'), covar=tensor([0.0812, 0.0317, 0.0300, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 03:58:33,165 INFO [train.py:968] (1/2) Epoch 16, batch 3800, giga_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.09646, over 29069.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3517, pruned_loss=0.1007, over 5725994.55 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3478, pruned_loss=0.09145, over 4856652.79 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3518, pruned_loss=0.1015, over 5719079.26 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:58:59,690 INFO [optim.py:369] (1/2) 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 03:59:12,877 INFO [train.py:968] (1/2) Epoch 16, batch 3850, giga_loss[loss=0.261, simple_loss=0.3401, pruned_loss=0.09099, over 28437.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.353, pruned_loss=0.1013, over 5732611.02 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3475, pruned_loss=0.09133, over 4886022.20 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3533, pruned_loss=0.1023, over 5724055.26 frames. ], batch size: 71, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 03:59:39,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2490, 1.0619, 3.8110, 3.0980], device='cuda:1'), covar=tensor([0.1693, 0.2921, 0.0412, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0699, 0.0606, 0.0890, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 03:59:56,709 INFO [train.py:968] (1/2) Epoch 16, batch 3900, giga_loss[loss=0.2779, simple_loss=0.3664, pruned_loss=0.09463, over 28610.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.352, pruned_loss=0.1001, over 5720599.59 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3475, pruned_loss=0.09125, over 4891096.09 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3524, pruned_loss=0.1009, over 5713045.35 frames. ], batch size: 60, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:00:25,153 INFO [optim.py:369] (1/2) 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 04:00:30,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=687892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:00:35,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6412, 1.7628, 1.8975, 1.4172], device='cuda:1'), covar=tensor([0.1779, 0.2626, 0.1491, 0.1781], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0697, 0.0916, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 04:00:38,147 INFO [train.py:968] (1/2) Epoch 16, batch 3950, giga_loss[loss=0.3507, simple_loss=0.3891, pruned_loss=0.1561, over 26515.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3517, pruned_loss=0.09985, over 5716984.85 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3475, pruned_loss=0.09125, over 4911249.32 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3521, pruned_loss=0.1007, over 5715631.05 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:01:19,577 INFO [train.py:968] (1/2) Epoch 16, batch 4000, giga_loss[loss=0.2447, simple_loss=0.3228, pruned_loss=0.08328, over 28854.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3499, pruned_loss=0.09929, over 5713001.47 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3477, pruned_loss=0.09142, over 4943472.55 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3502, pruned_loss=0.1002, over 5711636.01 frames. ], batch size: 119, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:01:48,111 INFO [optim.py:369] (1/2) 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 04:02:00,272 INFO [train.py:968] (1/2) Epoch 16, batch 4050, giga_loss[loss=0.3058, simple_loss=0.3759, pruned_loss=0.1178, over 28645.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3477, pruned_loss=0.09821, over 5705363.28 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.348, pruned_loss=0.09183, over 4956356.81 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3476, pruned_loss=0.09877, over 5710289.93 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:02:12,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:15,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:26,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8875, 3.7065, 3.4943, 1.8319], device='cuda:1'), covar=tensor([0.0635, 0.0779, 0.0747, 0.2355], device='cuda:1'), in_proj_covar=tensor([0.1129, 0.1039, 0.0894, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 04:02:27,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:29,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:02:38,470 INFO [train.py:968] (1/2) Epoch 16, batch 4100, giga_loss[loss=0.2769, simple_loss=0.3479, pruned_loss=0.1029, over 28740.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3445, pruned_loss=0.09656, over 5705652.19 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3482, pruned_loss=0.09211, over 4983855.77 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3443, pruned_loss=0.09696, over 5707965.08 frames. ], batch size: 242, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:02:54,197 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:03:07,109 INFO [optim.py:369] (1/2) 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 04:03:20,549 INFO [train.py:968] (1/2) Epoch 16, batch 4150, giga_loss[loss=0.2511, simple_loss=0.3322, pruned_loss=0.08496, over 28932.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3431, pruned_loss=0.0964, over 5705815.00 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3481, pruned_loss=0.09208, over 4993239.39 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.343, pruned_loss=0.09679, over 5705658.27 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:03:48,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9528, 1.2624, 1.1014, 0.1611], device='cuda:1'), covar=tensor([0.3561, 0.2755, 0.3970, 0.5616], device='cuda:1'), in_proj_covar=tensor([0.1634, 0.1539, 0.1523, 0.1330], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 04:04:01,418 INFO [train.py:968] (1/2) Epoch 16, batch 4200, giga_loss[loss=0.2709, simple_loss=0.3433, pruned_loss=0.09929, over 28996.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3419, pruned_loss=0.09638, over 5703140.15 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3475, pruned_loss=0.09177, over 5016606.63 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3421, pruned_loss=0.09705, over 5700714.12 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:04:08,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:09,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1798, 1.4687, 1.3087, 1.2601], device='cuda:1'), covar=tensor([0.1613, 0.1511, 0.1858, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0736, 0.0687, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 04:04:11,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:11,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:13,989 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:04:34,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:38,570 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:04:43,187 INFO [train.py:968] (1/2) Epoch 16, batch 4250, giga_loss[loss=0.2344, simple_loss=0.3095, pruned_loss=0.07963, over 28874.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3406, pruned_loss=0.09597, over 5697709.56 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3479, pruned_loss=0.09201, over 5037090.90 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3402, pruned_loss=0.09645, over 5698894.25 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:05:25,128 INFO [train.py:968] (1/2) Epoch 16, batch 4300, giga_loss[loss=0.2851, simple_loss=0.3466, pruned_loss=0.1118, over 28376.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3385, pruned_loss=0.09573, over 5706606.96 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3477, pruned_loss=0.09181, over 5045933.49 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3383, pruned_loss=0.09629, over 5705566.34 frames. ], batch size: 65, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:05:45,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3028, 2.2054, 2.4289, 1.9381], device='cuda:1'), covar=tensor([0.1506, 0.2118, 0.1576, 0.1955], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0734, 0.0687, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 04:05:51,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3124, 1.5177, 1.5941, 1.3544], device='cuda:1'), covar=tensor([0.2751, 0.2164, 0.1419, 0.2047], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1741, 0.1685, 0.1826], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 04:05:53,263 INFO [optim.py:369] (1/2) 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 04:06:04,638 INFO [train.py:968] (1/2) Epoch 16, batch 4350, giga_loss[loss=0.236, simple_loss=0.3146, pruned_loss=0.07871, over 28870.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3369, pruned_loss=0.09514, over 5696720.28 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3482, pruned_loss=0.09208, over 5052379.32 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3361, pruned_loss=0.09546, over 5703797.14 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:06:33,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=688335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:06:44,307 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6172, 1.8277, 1.7871, 1.6312], device='cuda:1'), covar=tensor([0.1799, 0.2249, 0.2168, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0735, 0.0686, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 04:06:44,630 INFO [train.py:968] (1/2) Epoch 16, batch 4400, giga_loss[loss=0.2123, simple_loss=0.2934, pruned_loss=0.06563, over 28903.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3351, pruned_loss=0.09389, over 5704955.99 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3482, pruned_loss=0.09201, over 5072690.42 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3342, pruned_loss=0.09426, over 5706608.32 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:06:45,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=688352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:06:52,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 04:07:15,894 INFO [optim.py:369] (1/2) 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 04:07:16,119 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=688386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:07:21,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-08 04:07:29,378 INFO [train.py:968] (1/2) Epoch 16, batch 4450, giga_loss[loss=0.2459, simple_loss=0.3294, pruned_loss=0.08122, over 28919.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3377, pruned_loss=0.09478, over 5706362.18 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3486, pruned_loss=0.0923, over 5081736.32 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3365, pruned_loss=0.09487, over 5707740.30 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:08:14,885 INFO [train.py:968] (1/2) Epoch 16, batch 4500, libri_loss[loss=0.2297, simple_loss=0.3067, pruned_loss=0.0763, over 29638.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3402, pruned_loss=0.09601, over 5702568.82 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.348, pruned_loss=0.09209, over 5098147.32 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3395, pruned_loss=0.09632, over 5699710.55 frames. ], batch size: 69, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:08:44,223 INFO [optim.py:369] (1/2) 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 04:08:59,084 INFO [train.py:968] (1/2) Epoch 16, batch 4550, giga_loss[loss=0.2547, simple_loss=0.3377, pruned_loss=0.0858, over 28930.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3418, pruned_loss=0.09586, over 5706362.49 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3479, pruned_loss=0.09205, over 5106445.55 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3413, pruned_loss=0.09617, over 5702126.79 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:09:20,136 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-08 04:09:41,320 INFO [train.py:968] (1/2) Epoch 16, batch 4600, giga_loss[loss=0.2729, simple_loss=0.3343, pruned_loss=0.1058, over 28404.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3435, pruned_loss=0.09625, over 5699051.31 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3476, pruned_loss=0.09187, over 5134107.32 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3432, pruned_loss=0.09683, over 5693751.25 frames. ], batch size: 71, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:10:06,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4276, 2.1916, 1.7234, 0.6242], device='cuda:1'), covar=tensor([0.4826, 0.2247, 0.3296, 0.5352], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1535, 0.1526, 0.1329], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 04:10:14,157 INFO [optim.py:369] (1/2) 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 04:10:28,416 INFO [train.py:968] (1/2) Epoch 16, batch 4650, giga_loss[loss=0.2638, simple_loss=0.3387, pruned_loss=0.09446, over 28942.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3428, pruned_loss=0.09563, over 5701087.33 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3477, pruned_loss=0.09204, over 5149384.82 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3424, pruned_loss=0.09602, over 5693491.97 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:11:08,829 INFO [train.py:968] (1/2) Epoch 16, batch 4700, giga_loss[loss=0.2695, simple_loss=0.3463, pruned_loss=0.09634, over 28611.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3415, pruned_loss=0.09513, over 5706404.16 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3481, pruned_loss=0.09223, over 5163777.59 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3407, pruned_loss=0.09536, over 5697483.64 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:11:41,451 INFO [optim.py:369] (1/2) 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 04:11:46,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7915, 1.0817, 2.9752, 2.8730], device='cuda:1'), covar=tensor([0.1612, 0.2411, 0.0549, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0702, 0.0612, 0.0894, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:11:52,756 INFO [train.py:968] (1/2) Epoch 16, batch 4750, giga_loss[loss=0.2473, simple_loss=0.3278, pruned_loss=0.08346, over 28859.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3432, pruned_loss=0.09686, over 5703655.04 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3484, pruned_loss=0.09241, over 5174742.71 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3424, pruned_loss=0.09695, over 5693795.55 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:12:01,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:12:14,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:12:36,887 INFO [train.py:968] (1/2) Epoch 16, batch 4800, giga_loss[loss=0.322, simple_loss=0.386, pruned_loss=0.129, over 28873.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.346, pruned_loss=0.09866, over 5704673.08 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3483, pruned_loss=0.09247, over 5192628.71 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3453, pruned_loss=0.0988, over 5692616.72 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:12:47,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=688761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:13:06,956 INFO [optim.py:369] (1/2) 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 04:13:08,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8566, 2.2354, 1.9356, 1.6172], device='cuda:1'), covar=tensor([0.2404, 0.1682, 0.1882, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.1829, 0.1754, 0.1703, 0.1835], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 04:13:20,648 INFO [train.py:968] (1/2) Epoch 16, batch 4850, giga_loss[loss=0.2823, simple_loss=0.3606, pruned_loss=0.102, over 28802.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.09988, over 5707205.62 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3483, pruned_loss=0.09245, over 5206843.55 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3481, pruned_loss=0.1001, over 5693919.91 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:13:59,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6192, 1.4606, 1.6107, 1.2356], device='cuda:1'), covar=tensor([0.1999, 0.3020, 0.1580, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0688, 0.0911, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 04:14:00,714 INFO [train.py:968] (1/2) Epoch 16, batch 4900, giga_loss[loss=0.2904, simple_loss=0.3575, pruned_loss=0.1116, over 28882.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3513, pruned_loss=0.1008, over 5717508.63 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.348, pruned_loss=0.09233, over 5221556.35 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3511, pruned_loss=0.1013, over 5704908.82 frames. ], batch size: 112, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:14:02,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:04,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:16,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:18,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:28,264 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:28,638 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 4950, libri_loss[loss=0.265, simple_loss=0.3395, pruned_loss=0.09524, over 29661.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3529, pruned_loss=0.1017, over 5720304.58 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3488, pruned_loss=0.09276, over 5244795.04 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3523, pruned_loss=0.1021, over 5706728.04 frames. ], batch size: 69, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:14:41,586 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=688902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:42,880 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=688904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:14:45,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=688907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:15:10,603 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 5000, giga_loss[loss=0.2663, simple_loss=0.3341, pruned_loss=0.09926, over 28753.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3533, pruned_loss=0.1014, over 5726604.24 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3495, pruned_loss=0.09307, over 5260485.67 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3523, pruned_loss=0.1016, over 5712176.08 frames. ], batch size: 99, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:15:46,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3484, 1.7934, 1.6871, 1.5292], device='cuda:1'), covar=tensor([0.1648, 0.1445, 0.1854, 0.1608], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0733, 0.0685, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 04:15:46,797 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-08 04:15:51,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6220, 1.5752, 4.0079, 3.3749], device='cuda:1'), covar=tensor([0.1420, 0.2453, 0.0448, 0.0954], device='cuda:1'), in_proj_covar=tensor([0.0706, 0.0614, 0.0902, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:15:52,735 INFO [optim.py:369] (1/2) 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 04:16:04,277 INFO [train.py:968] (1/2) Epoch 16, batch 5050, giga_loss[loss=0.2575, simple_loss=0.3338, pruned_loss=0.09057, over 28225.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3529, pruned_loss=0.1013, over 5728117.19 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3501, pruned_loss=0.09349, over 5272677.16 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3517, pruned_loss=0.1013, over 5716157.44 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:16:45,113 INFO [train.py:968] (1/2) Epoch 16, batch 5100, giga_loss[loss=0.2342, simple_loss=0.3135, pruned_loss=0.07742, over 28328.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 5708776.95 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3505, pruned_loss=0.09379, over 5271003.84 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3498, pruned_loss=0.1004, over 5713658.81 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:17:16,050 INFO [optim.py:369] (1/2) 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 04:17:27,261 INFO [train.py:968] (1/2) Epoch 16, batch 5150, giga_loss[loss=0.2472, simple_loss=0.3276, pruned_loss=0.08339, over 28713.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3479, pruned_loss=0.09894, over 5718389.21 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3504, pruned_loss=0.0937, over 5286677.65 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.347, pruned_loss=0.09913, over 5717950.74 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:18:10,004 INFO [train.py:968] (1/2) Epoch 16, batch 5200, giga_loss[loss=0.2577, simple_loss=0.3346, pruned_loss=0.09037, over 28883.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3447, pruned_loss=0.0971, over 5721952.16 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3507, pruned_loss=0.09388, over 5297887.04 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3436, pruned_loss=0.09716, over 5719310.79 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:18:39,716 INFO [optim.py:369] (1/2) 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 04:18:50,956 INFO [train.py:968] (1/2) Epoch 16, batch 5250, giga_loss[loss=0.2679, simple_loss=0.357, pruned_loss=0.08943, over 28666.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3439, pruned_loss=0.09585, over 5699418.47 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3506, pruned_loss=0.09387, over 5295886.90 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.343, pruned_loss=0.09599, over 5712086.91 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:18:57,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-08 04:19:30,447 INFO [train.py:968] (1/2) Epoch 16, batch 5300, giga_loss[loss=0.2478, simple_loss=0.3236, pruned_loss=0.08594, over 28720.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3466, pruned_loss=0.09597, over 5701920.11 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3504, pruned_loss=0.09375, over 5316850.51 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.09625, over 5705875.60 frames. ], batch size: 99, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:19:33,122 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 5350, giga_loss[loss=0.3197, simple_loss=0.3827, pruned_loss=0.1284, over 27578.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.348, pruned_loss=0.09755, over 5696356.77 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3506, pruned_loss=0.09393, over 5324816.25 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3471, pruned_loss=0.09767, over 5699074.06 frames. ], batch size: 472, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:20:56,171 INFO [train.py:968] (1/2) Epoch 16, batch 5400, giga_loss[loss=0.2555, simple_loss=0.3321, pruned_loss=0.0894, over 28907.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3465, pruned_loss=0.09759, over 5700602.24 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3504, pruned_loss=0.09377, over 5334380.78 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3459, pruned_loss=0.09795, over 5703834.01 frames. ], batch size: 227, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:21:30,761 INFO [optim.py:369] (1/2) 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 04:21:42,361 INFO [train.py:968] (1/2) Epoch 16, batch 5450, giga_loss[loss=0.2557, simple_loss=0.327, pruned_loss=0.09223, over 28863.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09856, over 5696465.32 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3507, pruned_loss=0.09395, over 5340230.13 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3447, pruned_loss=0.09875, over 5697084.05 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:22:26,171 INFO [train.py:968] (1/2) Epoch 16, batch 5500, giga_loss[loss=0.2488, simple_loss=0.3143, pruned_loss=0.09167, over 28258.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3449, pruned_loss=0.09934, over 5700724.56 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3511, pruned_loss=0.09416, over 5356470.42 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3438, pruned_loss=0.09943, over 5695952.28 frames. ], batch size: 65, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:22:56,385 INFO [optim.py:369] (1/2) 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 04:23:04,766 INFO [train.py:968] (1/2) Epoch 16, batch 5550, giga_loss[loss=0.2112, simple_loss=0.2891, pruned_loss=0.06666, over 28162.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3435, pruned_loss=0.09922, over 5707800.38 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3514, pruned_loss=0.09441, over 5374239.04 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3421, pruned_loss=0.09925, over 5698714.65 frames. ], batch size: 77, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:23:07,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3995, 1.6541, 1.3061, 1.5505], device='cuda:1'), covar=tensor([0.0729, 0.0316, 0.0340, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0089, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 04:23:53,180 INFO [train.py:968] (1/2) Epoch 16, batch 5600, giga_loss[loss=0.263, simple_loss=0.3369, pruned_loss=0.09459, over 28676.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.342, pruned_loss=0.09839, over 5715302.91 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.3515, pruned_loss=0.09446, over 5379661.41 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3408, pruned_loss=0.0984, over 5706334.89 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:24:25,443 INFO [optim.py:369] (1/2) 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 04:24:35,111 INFO [train.py:968] (1/2) Epoch 16, batch 5650, giga_loss[loss=0.214, simple_loss=0.2973, pruned_loss=0.06537, over 28884.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3375, pruned_loss=0.09598, over 5722699.28 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3522, pruned_loss=0.09492, over 5388216.59 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3357, pruned_loss=0.09563, over 5713786.80 frames. ], batch size: 174, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:24:36,287 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:24:37,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:24:54,927 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:24:56,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=689627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:25:08,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8225, 1.4378, 5.3361, 3.7625], device='cuda:1'), covar=tensor([0.1578, 0.2703, 0.0321, 0.0734], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0613, 0.0899, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:25:15,479 INFO [train.py:968] (1/2) Epoch 16, batch 5700, giga_loss[loss=0.244, simple_loss=0.3206, pruned_loss=0.08372, over 29096.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3326, pruned_loss=0.09312, over 5724724.38 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3522, pruned_loss=0.09496, over 5398663.70 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3308, pruned_loss=0.0928, over 5714233.04 frames. ], batch size: 155, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:25:46,440 INFO [optim.py:369] (1/2) 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 04:25:56,569 INFO [train.py:968] (1/2) Epoch 16, batch 5750, giga_loss[loss=0.2762, simple_loss=0.355, pruned_loss=0.09873, over 28626.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3322, pruned_loss=0.09283, over 5717919.19 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3524, pruned_loss=0.09512, over 5404462.21 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3301, pruned_loss=0.09239, over 5712894.60 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:26:36,763 INFO [train.py:968] (1/2) Epoch 16, batch 5800, giga_loss[loss=0.247, simple_loss=0.3292, pruned_loss=0.08237, over 29048.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3345, pruned_loss=0.09345, over 5723987.84 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3521, pruned_loss=0.09509, over 5409636.30 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3329, pruned_loss=0.09312, over 5718195.06 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:26:50,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=689770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:26:53,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=689773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:27:08,472 INFO [optim.py:369] (1/2) 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 04:27:10,552 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4356, 2.1189, 1.5473, 0.6372], device='cuda:1'), covar=tensor([0.5312, 0.2508, 0.3725, 0.5867], device='cuda:1'), in_proj_covar=tensor([0.1635, 0.1548, 0.1538, 0.1342], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 04:27:18,779 INFO [train.py:968] (1/2) Epoch 16, batch 5850, giga_loss[loss=0.2591, simple_loss=0.3289, pruned_loss=0.09459, over 28902.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3381, pruned_loss=0.09518, over 5716990.03 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.352, pruned_loss=0.09518, over 5410486.89 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3367, pruned_loss=0.09483, over 5717013.16 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:27:19,566 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=689802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:28:02,462 INFO [train.py:968] (1/2) Epoch 16, batch 5900, giga_loss[loss=0.3005, simple_loss=0.3727, pruned_loss=0.1141, over 27925.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3424, pruned_loss=0.09739, over 5712615.49 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3519, pruned_loss=0.09509, over 5411293.45 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3413, pruned_loss=0.09718, over 5713246.63 frames. ], batch size: 412, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:28:02,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3963, 1.6736, 1.3065, 1.3606], device='cuda:1'), covar=tensor([0.2642, 0.2544, 0.2959, 0.2206], device='cuda:1'), in_proj_covar=tensor([0.1397, 0.1020, 0.1236, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 04:28:13,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3379, 1.4776, 0.9767, 1.1750], device='cuda:1'), covar=tensor([0.1046, 0.0756, 0.1529, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0371, 0.0439, 0.0500, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:28:13,962 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=689865.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 04:28:36,899 INFO [optim.py:369] (1/2) 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 04:28:47,126 INFO [train.py:968] (1/2) Epoch 16, batch 5950, giga_loss[loss=0.2913, simple_loss=0.3628, pruned_loss=0.1099, over 28794.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3453, pruned_loss=0.09877, over 5713392.24 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3522, pruned_loss=0.09523, over 5424235.24 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.344, pruned_loss=0.0986, over 5710264.19 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:28:47,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4644, 1.6178, 1.5801, 1.5023], device='cuda:1'), covar=tensor([0.1340, 0.1780, 0.1816, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0735, 0.0689, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 04:29:18,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 04:29:35,681 INFO [train.py:968] (1/2) Epoch 16, batch 6000, giga_loss[loss=0.2562, simple_loss=0.3297, pruned_loss=0.09133, over 28750.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3486, pruned_loss=0.1015, over 5707983.70 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3523, pruned_loss=0.0953, over 5426607.06 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3475, pruned_loss=0.1013, over 5704788.78 frames. ], batch size: 99, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:29:35,681 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 04:29:44,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2568, 1.5853, 1.4812, 1.2512], device='cuda:1'), covar=tensor([0.1885, 0.1762, 0.1134, 0.1514], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1772, 0.1715, 0.1828], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 04:29:45,709 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 04:30:08,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=689977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:30:09,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=689979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:30:19,564 INFO [optim.py:369] (1/2) 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 04:30:22,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3968, 1.6523, 1.3108, 1.5965], device='cuda:1'), covar=tensor([0.2378, 0.2312, 0.2699, 0.2008], device='cuda:1'), in_proj_covar=tensor([0.1396, 0.1020, 0.1236, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 04:30:26,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6643, 1.5879, 1.2862, 1.2397], device='cuda:1'), covar=tensor([0.0737, 0.0564, 0.0896, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0439, 0.0502, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:30:27,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4798, 2.5949, 1.5896, 1.6080], device='cuda:1'), covar=tensor([0.0738, 0.0369, 0.0678, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0530, 0.0359, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 04:30:33,514 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 6050, giga_loss[loss=0.2894, simple_loss=0.3577, pruned_loss=0.1105, over 29017.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.354, pruned_loss=0.1058, over 5707241.62 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3522, pruned_loss=0.09537, over 5436201.89 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.353, pruned_loss=0.1058, over 5701404.04 frames. ], batch size: 66, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:31:27,605 INFO [train.py:968] (1/2) Epoch 16, batch 6100, giga_loss[loss=0.3993, simple_loss=0.4248, pruned_loss=0.1869, over 23696.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3613, pruned_loss=0.112, over 5680783.27 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3524, pruned_loss=0.09545, over 5440897.17 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3605, pruned_loss=0.112, over 5674102.59 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:32:02,411 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 6150, giga_loss[loss=0.4927, simple_loss=0.4856, pruned_loss=0.2499, over 26558.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3681, pruned_loss=0.1175, over 5681790.68 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3515, pruned_loss=0.095, over 5450375.45 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3688, pruned_loss=0.1185, over 5675122.02 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:32:31,624 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:32,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:33,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:35,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:55,328 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:32:57,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:00,386 INFO [train.py:968] (1/2) Epoch 16, batch 6200, giga_loss[loss=0.2809, simple_loss=0.3497, pruned_loss=0.106, over 28924.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3723, pruned_loss=0.1215, over 5679325.41 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3517, pruned_loss=0.09543, over 5465735.38 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3733, pruned_loss=0.123, over 5669286.60 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:33:02,233 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:04,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:26,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:33:40,078 INFO [optim.py:369] (1/2) 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 04:33:47,081 INFO [train.py:968] (1/2) Epoch 16, batch 6250, libri_loss[loss=0.2448, simple_loss=0.3326, pruned_loss=0.07855, over 29577.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3782, pruned_loss=0.1258, over 5688050.46 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3519, pruned_loss=0.09555, over 5475369.76 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3794, pruned_loss=0.1276, over 5675965.25 frames. ], batch size: 76, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:34:11,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9174, 2.2383, 2.0925, 1.6871], device='cuda:1'), covar=tensor([0.2622, 0.2098, 0.2173, 0.2357], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1764, 0.1705, 0.1823], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 04:34:32,023 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 6300, giga_loss[loss=0.317, simple_loss=0.3817, pruned_loss=0.1261, over 28789.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3817, pruned_loss=0.129, over 5664605.45 frames. ], libri_tot_loss[loss=0.2715, simple_loss=0.3519, pruned_loss=0.09553, over 5482541.53 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3834, pruned_loss=0.1312, over 5652370.54 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:34:40,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5124, 4.3215, 4.0476, 1.8513], device='cuda:1'), covar=tensor([0.0678, 0.0909, 0.1037, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.1137, 0.1045, 0.0901, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 04:35:17,601 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 6350, giga_loss[loss=0.3177, simple_loss=0.3798, pruned_loss=0.1278, over 28956.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3818, pruned_loss=0.1297, over 5668153.55 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3515, pruned_loss=0.09526, over 5500645.63 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3848, pruned_loss=0.1332, over 5648968.69 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:35:41,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7112, 1.7444, 1.9248, 1.4530], device='cuda:1'), covar=tensor([0.1598, 0.2249, 0.1303, 0.1625], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0690, 0.0907, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 04:35:49,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-08 04:36:21,981 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-08 04:36:22,065 INFO [train.py:968] (1/2) Epoch 16, batch 6400, giga_loss[loss=0.3639, simple_loss=0.4095, pruned_loss=0.1591, over 28301.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.386, pruned_loss=0.1346, over 5646767.93 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3514, pruned_loss=0.09516, over 5511697.49 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3896, pruned_loss=0.1387, over 5625921.68 frames. ], batch size: 368, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:36:25,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4851, 2.2398, 1.5663, 0.5779], device='cuda:1'), covar=tensor([0.5020, 0.2945, 0.4114, 0.5548], device='cuda:1'), in_proj_covar=tensor([0.1646, 0.1560, 0.1542, 0.1346], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 04:36:59,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=690383.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 04:37:04,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=690386.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 04:37:08,094 INFO [optim.py:369] (1/2) 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 04:37:20,308 INFO [train.py:968] (1/2) Epoch 16, batch 6450, libri_loss[loss=0.2466, simple_loss=0.333, pruned_loss=0.08013, over 29537.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3906, pruned_loss=0.139, over 5634363.18 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3514, pruned_loss=0.09513, over 5520751.87 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3944, pruned_loss=0.1434, over 5612090.59 frames. ], batch size: 83, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:37:35,126 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=690415.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 04:37:43,385 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-08 04:38:12,358 INFO [train.py:968] (1/2) Epoch 16, batch 6500, giga_loss[loss=0.3412, simple_loss=0.3919, pruned_loss=0.1453, over 27925.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3926, pruned_loss=0.1405, over 5627624.05 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3513, pruned_loss=0.09505, over 5522883.83 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3961, pruned_loss=0.1445, over 5609105.70 frames. ], batch size: 412, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:38:31,420 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 04:38:55,172 INFO [optim.py:369] (1/2) 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 04:38:55,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5193, 2.2226, 1.6554, 0.6191], device='cuda:1'), covar=tensor([0.3690, 0.1972, 0.2750, 0.4505], device='cuda:1'), in_proj_covar=tensor([0.1652, 0.1564, 0.1547, 0.1347], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 04:39:02,766 INFO [train.py:968] (1/2) Epoch 16, batch 6550, giga_loss[loss=0.3338, simple_loss=0.3869, pruned_loss=0.1403, over 28989.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3914, pruned_loss=0.1401, over 5635552.64 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3514, pruned_loss=0.0952, over 5519341.08 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3951, pruned_loss=0.1441, over 5626923.48 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:39:45,159 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2036, 0.9505, 1.0976, 1.4013], device='cuda:1'), covar=tensor([0.0675, 0.0437, 0.0313, 0.0686], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 04:39:57,751 INFO [train.py:968] (1/2) Epoch 16, batch 6600, giga_loss[loss=0.328, simple_loss=0.3855, pruned_loss=0.1353, over 28499.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3911, pruned_loss=0.1404, over 5636463.40 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3515, pruned_loss=0.09513, over 5524992.06 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3944, pruned_loss=0.1443, over 5625955.85 frames. ], batch size: 71, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:40:00,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6199, 1.7881, 1.2667, 1.4358], device='cuda:1'), covar=tensor([0.0878, 0.0595, 0.1092, 0.1014], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0444, 0.0506, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:40:39,980 INFO [optim.py:369] (1/2) 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 04:40:46,692 INFO [train.py:968] (1/2) Epoch 16, batch 6650, giga_loss[loss=0.368, simple_loss=0.4239, pruned_loss=0.1561, over 28222.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3905, pruned_loss=0.1387, over 5631174.15 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3517, pruned_loss=0.09523, over 5522958.29 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.3938, pruned_loss=0.1427, over 5627711.21 frames. ], batch size: 368, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:41:38,102 INFO [train.py:968] (1/2) Epoch 16, batch 6700, libri_loss[loss=0.2665, simple_loss=0.3483, pruned_loss=0.09231, over 29531.00 frames. ], tot_loss[loss=0.3354, simple_loss=0.3921, pruned_loss=0.1393, over 5638123.81 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3514, pruned_loss=0.0951, over 5529191.94 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3958, pruned_loss=0.1435, over 5631585.84 frames. ], batch size: 81, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:42:21,755 INFO [optim.py:369] (1/2) 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 04:42:29,521 INFO [train.py:968] (1/2) Epoch 16, batch 6750, libri_loss[loss=0.2761, simple_loss=0.3585, pruned_loss=0.09684, over 29532.00 frames. ], tot_loss[loss=0.335, simple_loss=0.392, pruned_loss=0.139, over 5619433.11 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3515, pruned_loss=0.09498, over 5537124.54 frames. ], giga_tot_loss[loss=0.3415, simple_loss=0.3959, pruned_loss=0.1435, over 5608718.31 frames. ], batch size: 84, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:43:23,868 INFO [train.py:968] (1/2) Epoch 16, batch 6800, giga_loss[loss=0.296, simple_loss=0.3769, pruned_loss=0.1076, over 29027.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3877, pruned_loss=0.1351, over 5622170.54 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3513, pruned_loss=0.09504, over 5545301.28 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.392, pruned_loss=0.1399, over 5608521.29 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:43:44,571 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6151, 3.0969, 1.6543, 1.6821], device='cuda:1'), covar=tensor([0.0702, 0.0301, 0.0668, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0533, 0.0361, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 04:44:04,654 INFO [optim.py:369] (1/2) 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 04:44:12,156 INFO [train.py:968] (1/2) Epoch 16, batch 6850, giga_loss[loss=0.3764, simple_loss=0.4304, pruned_loss=0.1612, over 28891.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3852, pruned_loss=0.1319, over 5634363.07 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3513, pruned_loss=0.09504, over 5552859.93 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3892, pruned_loss=0.1364, over 5618232.96 frames. ], batch size: 284, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:44:49,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3270, 1.5952, 1.7369, 1.4078], device='cuda:1'), covar=tensor([0.2473, 0.1894, 0.1229, 0.1815], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1774, 0.1721, 0.1834], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 04:45:02,373 INFO [train.py:968] (1/2) Epoch 16, batch 6900, giga_loss[loss=0.2605, simple_loss=0.341, pruned_loss=0.09003, over 28783.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3807, pruned_loss=0.1278, over 5647940.84 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.351, pruned_loss=0.09491, over 5560732.33 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3849, pruned_loss=0.1322, over 5629796.23 frames. ], batch size: 284, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:45:40,935 INFO [optim.py:369] (1/2) 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 04:45:50,615 INFO [train.py:968] (1/2) Epoch 16, batch 6950, giga_loss[loss=0.2973, simple_loss=0.3744, pruned_loss=0.1101, over 29120.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3778, pruned_loss=0.1253, over 5652281.13 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3507, pruned_loss=0.0948, over 5569774.18 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3824, pruned_loss=0.13, over 5632261.17 frames. ], batch size: 113, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:46:34,088 INFO [train.py:968] (1/2) Epoch 16, batch 7000, libri_loss[loss=0.2788, simple_loss=0.3573, pruned_loss=0.1002, over 29283.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3752, pruned_loss=0.1236, over 5643363.76 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3504, pruned_loss=0.09479, over 5562698.68 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3803, pruned_loss=0.1288, over 5636655.14 frames. ], batch size: 94, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:46:45,908 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=690962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:46:58,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0329, 1.1003, 3.5800, 3.0542], device='cuda:1'), covar=tensor([0.2211, 0.3139, 0.0829, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0709, 0.0619, 0.0910, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:47:15,813 INFO [optim.py:369] (1/2) 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 04:47:24,969 INFO [train.py:968] (1/2) Epoch 16, batch 7050, giga_loss[loss=0.4086, simple_loss=0.4199, pruned_loss=0.1987, over 23816.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3749, pruned_loss=0.1233, over 5651314.40 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3502, pruned_loss=0.09467, over 5566264.89 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3794, pruned_loss=0.1279, over 5644115.72 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:48:18,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4004, 3.2280, 3.0714, 2.0018], device='cuda:1'), covar=tensor([0.0720, 0.0808, 0.0739, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.1149, 0.1060, 0.0913, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 04:48:24,101 INFO [train.py:968] (1/2) Epoch 16, batch 7100, giga_loss[loss=0.2634, simple_loss=0.3456, pruned_loss=0.09058, over 28656.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3747, pruned_loss=0.1228, over 5659162.48 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.35, pruned_loss=0.09457, over 5572944.14 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.379, pruned_loss=0.127, over 5649272.53 frames. ], batch size: 71, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:49:08,467 INFO [optim.py:369] (1/2) 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 04:49:17,697 INFO [train.py:968] (1/2) Epoch 16, batch 7150, giga_loss[loss=0.3419, simple_loss=0.4073, pruned_loss=0.1382, over 28932.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3742, pruned_loss=0.1209, over 5670281.72 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3505, pruned_loss=0.09487, over 5575265.86 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3775, pruned_loss=0.1244, over 5661846.66 frames. ], batch size: 136, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 04:50:09,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6159, 1.8216, 1.4961, 1.9072], device='cuda:1'), covar=tensor([0.2621, 0.2715, 0.3098, 0.2214], device='cuda:1'), in_proj_covar=tensor([0.1390, 0.1017, 0.1234, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0008], device='cuda:1')
+2023-03-08 04:50:14,914 INFO [train.py:968] (1/2) Epoch 16, batch 7200, giga_loss[loss=0.2821, simple_loss=0.3683, pruned_loss=0.09797, over 28520.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3732, pruned_loss=0.1184, over 5664963.80 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3495, pruned_loss=0.0945, over 5583180.25 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3773, pruned_loss=0.1223, over 5653618.81 frames. ], batch size: 336, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:50:50,774 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:50:58,974 INFO [optim.py:369] (1/2) 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 04:51:07,153 INFO [train.py:968] (1/2) Epoch 16, batch 7250, giga_loss[loss=0.3325, simple_loss=0.3913, pruned_loss=0.1368, over 28673.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3759, pruned_loss=0.1202, over 5663341.98 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3497, pruned_loss=0.09456, over 5585687.34 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3793, pruned_loss=0.1234, over 5652880.26 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:51:56,208 INFO [train.py:968] (1/2) Epoch 16, batch 7300, giga_loss[loss=0.3017, simple_loss=0.3761, pruned_loss=0.1137, over 28813.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3745, pruned_loss=0.1192, over 5676327.24 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3494, pruned_loss=0.09432, over 5595908.67 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3785, pruned_loss=0.1231, over 5661988.19 frames. ], batch size: 174, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:52:03,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0490, 1.3881, 1.2344, 1.4827], device='cuda:1'), covar=tensor([0.0821, 0.0323, 0.0313, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 04:52:33,304 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 7350, giga_loss[loss=0.2982, simple_loss=0.359, pruned_loss=0.1188, over 28996.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.372, pruned_loss=0.1177, over 5682662.63 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.349, pruned_loss=0.09407, over 5608339.42 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.377, pruned_loss=0.1225, over 5663921.39 frames. ], batch size: 128, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:53:16,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=691337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:53:30,162 INFO [train.py:968] (1/2) Epoch 16, batch 7400, giga_loss[loss=0.3591, simple_loss=0.3858, pruned_loss=0.1662, over 23640.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.372, pruned_loss=0.1196, over 5664008.35 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3491, pruned_loss=0.09419, over 5606854.54 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3766, pruned_loss=0.124, over 5652625.09 frames. ], batch size: 705, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:53:58,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 04:54:09,130 INFO [optim.py:369] (1/2) 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 04:54:14,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8906, 1.1800, 2.8590, 2.6157], device='cuda:1'), covar=tensor([0.1564, 0.2372, 0.0600, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0707, 0.0616, 0.0906, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 04:54:15,615 INFO [train.py:968] (1/2) Epoch 16, batch 7450, giga_loss[loss=0.2651, simple_loss=0.3461, pruned_loss=0.09203, over 28957.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3712, pruned_loss=0.1193, over 5671216.87 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3493, pruned_loss=0.09433, over 5601673.69 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.375, pruned_loss=0.1232, over 5668500.26 frames. ], batch size: 213, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:55:06,553 INFO [train.py:968] (1/2) Epoch 16, batch 7500, giga_loss[loss=0.3218, simple_loss=0.3811, pruned_loss=0.1313, over 28693.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3709, pruned_loss=0.1182, over 5675426.46 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3495, pruned_loss=0.09443, over 5598973.04 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3744, pruned_loss=0.1218, over 5677835.92 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:55:36,035 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=691480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:55:40,881 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 7550, libri_loss[loss=0.2121, simple_loss=0.2989, pruned_loss=0.06266, over 29662.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3703, pruned_loss=0.1168, over 5686642.32 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3494, pruned_loss=0.09436, over 5603938.04 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3735, pruned_loss=0.12, over 5685091.37 frames. ], batch size: 73, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 04:56:07,654 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=691512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:56:20,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 04:56:41,855 INFO [train.py:968] (1/2) Epoch 16, batch 7600, giga_loss[loss=0.2999, simple_loss=0.3693, pruned_loss=0.1153, over 28846.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3709, pruned_loss=0.1176, over 5685404.79 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3496, pruned_loss=0.09471, over 5608891.70 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3739, pruned_loss=0.1205, over 5682163.93 frames. ], batch size: 186, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:56:49,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=691560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:57:01,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-08 04:57:19,931 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 7650, giga_loss[loss=0.3008, simple_loss=0.3661, pruned_loss=0.1177, over 28601.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3686, pruned_loss=0.1164, over 5693538.95 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3488, pruned_loss=0.09423, over 5619171.44 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3724, pruned_loss=0.1199, over 5683911.53 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:57:34,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4590, 1.7904, 1.7422, 1.2649], device='cuda:1'), covar=tensor([0.1832, 0.2761, 0.1583, 0.1859], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0697, 0.0911, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 04:58:14,520 INFO [train.py:968] (1/2) Epoch 16, batch 7700, giga_loss[loss=0.3742, simple_loss=0.4237, pruned_loss=0.1623, over 28733.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3677, pruned_loss=0.1165, over 5689475.60 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3486, pruned_loss=0.09421, over 5616975.05 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3715, pruned_loss=0.12, over 5686294.71 frames. ], batch size: 262, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:58:52,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4973, 1.8112, 1.4413, 1.5404], device='cuda:1'), covar=tensor([0.2418, 0.2443, 0.2796, 0.2336], device='cuda:1'), in_proj_covar=tensor([0.1395, 0.1021, 0.1238, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 04:58:58,115 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 7750, giga_loss[loss=0.3079, simple_loss=0.3688, pruned_loss=0.1235, over 28909.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3674, pruned_loss=0.1173, over 5680798.79 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3483, pruned_loss=0.09406, over 5618861.82 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.371, pruned_loss=0.1207, over 5677602.93 frames. ], batch size: 199, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 04:59:09,840 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=691703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:12,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=691706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:17,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 04:59:31,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:40,548 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=691735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:42,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3457, 3.1715, 3.0038, 1.4860], device='cuda:1'), covar=tensor([0.0926, 0.1058, 0.0966, 0.2184], device='cuda:1'), in_proj_covar=tensor([0.1147, 0.1056, 0.0911, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 04:59:56,206 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 04:59:58,353 INFO [train.py:968] (1/2) Epoch 16, batch 7800, giga_loss[loss=0.3019, simple_loss=0.3677, pruned_loss=0.1181, over 28921.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3657, pruned_loss=0.1164, over 5696496.19 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.348, pruned_loss=0.09398, over 5626155.38 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3692, pruned_loss=0.1197, over 5689111.93 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:00:45,074 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:00:46,262 INFO [optim.py:369] (1/2) 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 05:00:51,724 INFO [train.py:968] (1/2) Epoch 16, batch 7850, libri_loss[loss=0.2485, simple_loss=0.3397, pruned_loss=0.07863, over 29293.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5696950.00 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.348, pruned_loss=0.0939, over 5627317.60 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3665, pruned_loss=0.1184, over 5690361.55 frames. ], batch size: 94, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:01:40,540 INFO [train.py:968] (1/2) Epoch 16, batch 7900, giga_loss[loss=0.292, simple_loss=0.3578, pruned_loss=0.1131, over 28975.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5702948.97 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3478, pruned_loss=0.09367, over 5633031.51 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.366, pruned_loss=0.1186, over 5693787.34 frames. ], batch size: 164, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:02:24,558 INFO [optim.py:369] (1/2) 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 05:02:32,642 INFO [train.py:968] (1/2) Epoch 16, batch 7950, giga_loss[loss=0.3088, simple_loss=0.3672, pruned_loss=0.1252, over 27685.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3652, pruned_loss=0.1172, over 5695223.62 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3479, pruned_loss=0.09371, over 5637044.47 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3675, pruned_loss=0.1196, over 5685286.93 frames. ], batch size: 472, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:03:18,936 INFO [train.py:968] (1/2) Epoch 16, batch 8000, giga_loss[loss=0.2772, simple_loss=0.3541, pruned_loss=0.1002, over 28771.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1172, over 5692849.18 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3485, pruned_loss=0.09406, over 5643148.80 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3681, pruned_loss=0.1195, over 5680674.09 frames. ], batch size: 99, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 05:03:51,753 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=691990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:03:56,270 INFO [optim.py:369] (1/2) 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 05:04:05,026 INFO [train.py:968] (1/2) Epoch 16, batch 8050, libri_loss[loss=0.2639, simple_loss=0.351, pruned_loss=0.08843, over 29547.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3664, pruned_loss=0.1164, over 5685819.39 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3482, pruned_loss=0.09377, over 5650293.90 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3686, pruned_loss=0.1192, over 5671010.51 frames. ], batch size: 84, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:04:55,803 INFO [train.py:968] (1/2) Epoch 16, batch 8100, giga_loss[loss=0.2824, simple_loss=0.352, pruned_loss=0.1064, over 28476.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3671, pruned_loss=0.1168, over 5687311.35 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3483, pruned_loss=0.09376, over 5654269.47 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3692, pruned_loss=0.1195, over 5672576.21 frames. ], batch size: 85, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:05:23,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692079.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:05:41,881 INFO [optim.py:369] (1/2) 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 05:05:46,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:05:47,260 INFO [train.py:968] (1/2) Epoch 16, batch 8150, giga_loss[loss=0.4055, simple_loss=0.4477, pruned_loss=0.1816, over 27545.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3692, pruned_loss=0.1187, over 5690494.13 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3481, pruned_loss=0.09368, over 5657933.60 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3713, pruned_loss=0.1212, over 5676044.18 frames. ], batch size: 472, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:05:50,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8975, 1.2528, 2.8071, 2.7528], device='cuda:1'), covar=tensor([0.1574, 0.2365, 0.0598, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0714, 0.0620, 0.0915, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 05:06:10,332 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:06:28,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5663, 1.8224, 1.5285, 1.4605], device='cuda:1'), covar=tensor([0.2684, 0.2724, 0.3029, 0.2412], device='cuda:1'), in_proj_covar=tensor([0.1401, 0.1026, 0.1244, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 05:06:31,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.45 vs. limit=5.0
+2023-03-08 05:06:39,976 INFO [train.py:968] (1/2) Epoch 16, batch 8200, giga_loss[loss=0.2699, simple_loss=0.3497, pruned_loss=0.09503, over 28996.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3704, pruned_loss=0.1203, over 5680841.52 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3484, pruned_loss=0.09376, over 5655584.76 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3727, pruned_loss=0.1234, over 5671360.47 frames. ], batch size: 155, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:06:55,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:07:24,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4631, 1.5840, 1.4797, 1.3543], device='cuda:1'), covar=tensor([0.2281, 0.2085, 0.1647, 0.1971], device='cuda:1'), in_proj_covar=tensor([0.1851, 0.1787, 0.1727, 0.1844], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 05:07:25,063 INFO [optim.py:369] (1/2) 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 05:07:31,557 INFO [train.py:968] (1/2) Epoch 16, batch 8250, giga_loss[loss=0.3138, simple_loss=0.3717, pruned_loss=0.1279, over 28653.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.122, over 5677219.50 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3484, pruned_loss=0.09383, over 5655627.28 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3732, pruned_loss=0.1247, over 5669980.32 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:08:18,524 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:24,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:28,940 INFO [train.py:968] (1/2) Epoch 16, batch 8300, giga_loss[loss=0.3079, simple_loss=0.3742, pruned_loss=0.1208, over 28915.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3739, pruned_loss=0.1256, over 5667734.45 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3484, pruned_loss=0.09383, over 5655627.28 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3756, pruned_loss=0.1277, over 5662100.10 frames. ], batch size: 106, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:08:45,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:48,610 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:49,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:08:56,022 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:09:16,199 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:17,990 INFO [train.py:968] (1/2) Epoch 16, batch 8350, libri_loss[loss=0.2722, simple_loss=0.3528, pruned_loss=0.09579, over 29610.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3732, pruned_loss=0.1254, over 5665993.37 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3485, pruned_loss=0.09382, over 5657930.13 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3747, pruned_loss=0.1274, over 5659393.23 frames. ], batch size: 75, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:09:27,408 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:54,835 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:09:55,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2071, 2.5256, 1.2725, 1.3365], device='cuda:1'), covar=tensor([0.0988, 0.0387, 0.0911, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0534, 0.0359, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 05:10:03,377 INFO [train.py:968] (1/2) Epoch 16, batch 8400, giga_loss[loss=0.3423, simple_loss=0.3871, pruned_loss=0.1488, over 26455.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3731, pruned_loss=0.1244, over 5670777.72 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3484, pruned_loss=0.09377, over 5660807.29 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3749, pruned_loss=0.1268, over 5663154.98 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 8.0
+2023-03-08 05:10:14,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:10:23,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6204, 1.6608, 1.3074, 1.2324], device='cuda:1'), covar=tensor([0.0818, 0.0527, 0.0900, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0446, 0.0507, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 05:10:44,722 INFO [optim.py:369] (1/2) 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 05:10:48,335 INFO [train.py:968] (1/2) Epoch 16, batch 8450, giga_loss[loss=0.2775, simple_loss=0.3478, pruned_loss=0.1036, over 28697.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5668177.40 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3486, pruned_loss=0.0937, over 5665719.68 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3732, pruned_loss=0.1247, over 5657960.98 frames. ], batch size: 242, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:11:29,388 INFO [train.py:968] (1/2) Epoch 16, batch 8500, giga_loss[loss=0.3349, simple_loss=0.3799, pruned_loss=0.1449, over 26668.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3693, pruned_loss=0.1199, over 5674572.90 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.349, pruned_loss=0.09386, over 5666875.14 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3713, pruned_loss=0.123, over 5665330.78 frames. ], batch size: 555, lr: 2.03e-03, grad_scale: 4.0
+2023-03-08 05:11:33,207 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:12:16,678 INFO [optim.py:369] (1/2) 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 05:12:20,326 INFO [train.py:968] (1/2) Epoch 16, batch 8550, giga_loss[loss=0.3212, simple_loss=0.3833, pruned_loss=0.1295, over 28595.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1191, over 5678454.27 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3491, pruned_loss=0.09393, over 5669330.24 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3686, pruned_loss=0.1217, over 5668971.05 frames. ], batch size: 307, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 05:12:27,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:12:30,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:12:58,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2596, 1.5130, 1.5347, 1.2915], device='cuda:1'), covar=tensor([0.1660, 0.1382, 0.1020, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.1850, 0.1783, 0.1724, 0.1846], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 05:13:05,260 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:13:15,277 INFO [train.py:968] (1/2) Epoch 16, batch 8600, giga_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.0983, over 28935.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3668, pruned_loss=0.12, over 5665163.18 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.349, pruned_loss=0.0939, over 5670532.59 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3682, pruned_loss=0.1222, over 5656840.35 frames. ], batch size: 145, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 05:13:24,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-08 05:14:03,815 INFO [optim.py:369] (1/2) 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 05:14:04,178 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692597.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:14:09,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:14:09,829 INFO [train.py:968] (1/2) Epoch 16, batch 8650, giga_loss[loss=0.3356, simple_loss=0.3946, pruned_loss=0.1383, over 28316.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1206, over 5665364.45 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3487, pruned_loss=0.0937, over 5675199.63 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3704, pruned_loss=0.123, over 5654418.34 frames. ], batch size: 368, lr: 2.03e-03, grad_scale: 2.0
+2023-03-08 05:14:10,904 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 05:14:37,502 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:14:56,932 INFO [train.py:968] (1/2) Epoch 16, batch 8700, giga_loss[loss=0.36, simple_loss=0.4084, pruned_loss=0.1558, over 26683.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.372, pruned_loss=0.1205, over 5667907.29 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09341, over 5678475.76 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3745, pruned_loss=0.1235, over 5655547.95 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:15:22,795 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:15:30,326 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:15:45,386 INFO [optim.py:369] (1/2) 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 05:15:47,932 INFO [train.py:968] (1/2) Epoch 16, batch 8750, giga_loss[loss=0.2928, simple_loss=0.3691, pruned_loss=0.1082, over 28953.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3738, pruned_loss=0.12, over 5681727.44 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09343, over 5682021.24 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.376, pruned_loss=0.1227, over 5668745.74 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:16:27,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 05:16:35,344 INFO [train.py:968] (1/2) Epoch 16, batch 8800, giga_loss[loss=0.3406, simple_loss=0.3936, pruned_loss=0.1438, over 27963.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3756, pruned_loss=0.1217, over 5682340.44 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3482, pruned_loss=0.0934, over 5687845.20 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3782, pruned_loss=0.1246, over 5666358.81 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:17:10,870 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=692786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:17:13,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=692789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:17:18,446 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 8850, giga_loss[loss=0.297, simple_loss=0.3631, pruned_loss=0.1155, over 28742.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.378, pruned_loss=0.1242, over 5668310.09 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.09368, over 5690167.57 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3801, pruned_loss=0.1267, over 5653359.51 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:17:40,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=692818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:18:11,892 INFO [train.py:968] (1/2) Epoch 16, batch 8900, giga_loss[loss=0.2433, simple_loss=0.3198, pruned_loss=0.08336, over 28278.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3765, pruned_loss=0.1241, over 5664353.50 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3482, pruned_loss=0.09348, over 5691433.20 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3795, pruned_loss=0.1272, over 5650722.65 frames. ], batch size: 77, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:18:16,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3778, 3.3498, 1.5303, 1.5139], device='cuda:1'), covar=tensor([0.0992, 0.0372, 0.0856, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0533, 0.0358, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 05:18:23,063 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=692863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:18:45,504 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 05:18:59,352 INFO [optim.py:369] (1/2) 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 05:19:02,243 INFO [train.py:968] (1/2) Epoch 16, batch 8950, libri_loss[loss=0.2865, simple_loss=0.3659, pruned_loss=0.1036, over 29550.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3746, pruned_loss=0.1235, over 5645570.98 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3484, pruned_loss=0.09361, over 5690579.65 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3779, pruned_loss=0.1271, over 5634337.25 frames. ], batch size: 82, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:19:52,384 INFO [train.py:968] (1/2) Epoch 16, batch 9000, giga_loss[loss=0.2729, simple_loss=0.3365, pruned_loss=0.1047, over 28883.00 frames. ], tot_loss[loss=0.311, simple_loss=0.374, pruned_loss=0.124, over 5654557.42 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3486, pruned_loss=0.09372, over 5690573.01 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3766, pruned_loss=0.1269, over 5645319.42 frames. ], batch size: 106, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:19:52,384 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 05:20:00,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2279, 2.8248, 1.3121, 1.3960], device='cuda:1'), covar=tensor([0.0970, 0.0342, 0.0928, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0535, 0.0359, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 05:20:01,271 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 05:20:19,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0261, 3.8701, 3.6966, 1.8915], device='cuda:1'), covar=tensor([0.0602, 0.0735, 0.0720, 0.2188], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.1064, 0.0916, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 05:20:44,019 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=692993.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:20:47,115 INFO [optim.py:369] (1/2) 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 05:20:50,854 INFO [train.py:968] (1/2) Epoch 16, batch 9050, giga_loss[loss=0.3301, simple_loss=0.382, pruned_loss=0.1391, over 28719.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3741, pruned_loss=0.1248, over 5651546.96 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3488, pruned_loss=0.09373, over 5685654.47 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3765, pruned_loss=0.1277, over 5648556.63 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:20:51,783 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693002.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:20:55,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:20:58,336 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:21:29,106 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:21:41,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:21:42,093 INFO [train.py:968] (1/2) Epoch 16, batch 9100, giga_loss[loss=0.3029, simple_loss=0.3681, pruned_loss=0.1189, over 28556.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3745, pruned_loss=0.1255, over 5645842.36 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09379, over 5689947.68 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3767, pruned_loss=0.1284, over 5639371.02 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:21:52,192 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:22:34,130 INFO [optim.py:369] (1/2) 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 05:22:36,135 INFO [train.py:968] (1/2) Epoch 16, batch 9150, giga_loss[loss=0.2906, simple_loss=0.3521, pruned_loss=0.1146, over 28859.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.373, pruned_loss=0.1252, over 5643236.33 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3491, pruned_loss=0.09393, over 5681926.94 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3746, pruned_loss=0.1273, over 5645691.05 frames. ], batch size: 112, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:23:25,018 INFO [train.py:968] (1/2) Epoch 16, batch 9200, giga_loss[loss=0.2651, simple_loss=0.3447, pruned_loss=0.09281, over 28811.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3699, pruned_loss=0.1234, over 5649625.44 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3491, pruned_loss=0.09399, over 5684631.58 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3716, pruned_loss=0.1256, over 5648501.63 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:23:35,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:03,578 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:05,393 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:07,285 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 9250, giga_loss[loss=0.2399, simple_loss=0.3179, pruned_loss=0.08098, over 28492.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3699, pruned_loss=0.1229, over 5647396.20 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.0939, over 5685315.31 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.372, pruned_loss=0.1254, over 5645038.65 frames. ], batch size: 78, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:24:10,015 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:13,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:32,171 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:24:42,063 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 9300, giga_loss[loss=0.2817, simple_loss=0.3547, pruned_loss=0.1043, over 28899.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3706, pruned_loss=0.1218, over 5660735.36 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3486, pruned_loss=0.09371, over 5692346.08 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3732, pruned_loss=0.125, over 5651533.20 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:25:48,985 INFO [optim.py:369] (1/2) 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 05:25:50,890 INFO [train.py:968] (1/2) Epoch 16, batch 9350, giga_loss[loss=0.4014, simple_loss=0.4333, pruned_loss=0.1848, over 27888.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3728, pruned_loss=0.1237, over 5643519.90 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3489, pruned_loss=0.0938, over 5685880.97 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1266, over 5641453.70 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:26:39,607 INFO [train.py:968] (1/2) Epoch 16, batch 9400, giga_loss[loss=0.4068, simple_loss=0.4336, pruned_loss=0.19, over 28674.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3733, pruned_loss=0.1236, over 5657623.38 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3485, pruned_loss=0.09361, over 5690487.78 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.376, pruned_loss=0.1268, over 5651084.10 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:26:55,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693368.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:27:08,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:27:26,530 INFO [optim.py:369] (1/2) 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 05:27:30,949 INFO [train.py:968] (1/2) Epoch 16, batch 9450, giga_loss[loss=0.2927, simple_loss=0.3693, pruned_loss=0.1081, over 28908.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3753, pruned_loss=0.1225, over 5660777.24 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3488, pruned_loss=0.09373, over 5692258.50 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3773, pruned_loss=0.1252, over 5653870.21 frames. ], batch size: 227, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:28:00,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1543, 1.2971, 1.0455, 0.8699], device='cuda:1'), covar=tensor([0.0677, 0.0347, 0.0830, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0445, 0.0504, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 05:28:13,216 INFO [train.py:968] (1/2) Epoch 16, batch 9500, giga_loss[loss=0.3427, simple_loss=0.3943, pruned_loss=0.1455, over 28034.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3765, pruned_loss=0.1217, over 5669103.26 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.09362, over 5692063.48 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3796, pruned_loss=0.1252, over 5662224.01 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:28:58,369 INFO [optim.py:369] (1/2) 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 05:29:02,645 INFO [train.py:968] (1/2) Epoch 16, batch 9550, giga_loss[loss=0.297, simple_loss=0.3716, pruned_loss=0.1112, over 28421.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3793, pruned_loss=0.1228, over 5670648.23 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3488, pruned_loss=0.09364, over 5695280.78 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3823, pruned_loss=0.1263, over 5662031.56 frames. ], batch size: 65, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:29:07,364 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693507.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:11,284 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693511.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:14,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693514.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:29:21,126 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693520.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:29:24,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:38,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:29:43,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:29:50,604 INFO [train.py:968] (1/2) Epoch 16, batch 9600, giga_loss[loss=0.4365, simple_loss=0.456, pruned_loss=0.2085, over 28924.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.383, pruned_loss=0.1267, over 5676821.32 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.0936, over 5694569.28 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3857, pruned_loss=0.1298, over 5670460.70 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 05:29:51,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693552.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:30:01,096 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:30:10,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 05:30:36,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3835, 1.8277, 1.2365, 0.8641], device='cuda:1'), covar=tensor([0.4939, 0.2724, 0.2455, 0.4513], device='cuda:1'), in_proj_covar=tensor([0.1661, 0.1576, 0.1544, 0.1356], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 05:30:40,704 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 9650, giga_loss[loss=0.3674, simple_loss=0.4111, pruned_loss=0.1619, over 28928.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3837, pruned_loss=0.1286, over 5663652.37 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3484, pruned_loss=0.09328, over 5698279.38 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3868, pruned_loss=0.1319, over 5655158.05 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:31:00,852 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:31:31,289 INFO [train.py:968] (1/2) Epoch 16, batch 9700, giga_loss[loss=0.3047, simple_loss=0.3791, pruned_loss=0.1152, over 28854.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3824, pruned_loss=0.1278, over 5666064.98 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.09355, over 5701422.96 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3853, pruned_loss=0.1308, over 5655931.79 frames. ], batch size: 99, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:31:56,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=693679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:31:58,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=693682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:32:13,870 INFO [optim.py:369] (1/2) 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 05:32:14,970 INFO [train.py:968] (1/2) Epoch 16, batch 9750, giga_loss[loss=0.2985, simple_loss=0.384, pruned_loss=0.1065, over 28559.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3814, pruned_loss=0.1258, over 5678612.74 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3492, pruned_loss=0.09384, over 5707831.78 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3843, pruned_loss=0.1291, over 5663709.43 frames. ], batch size: 71, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:32:17,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-08 05:32:26,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=693711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:32:43,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6599, 2.1802, 1.9770, 1.5063], device='cuda:1'), covar=tensor([0.3050, 0.2064, 0.2174, 0.2641], device='cuda:1'), in_proj_covar=tensor([0.1842, 0.1780, 0.1720, 0.1835], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 05:33:03,034 INFO [train.py:968] (1/2) Epoch 16, batch 9800, giga_loss[loss=0.2844, simple_loss=0.3613, pruned_loss=0.1038, over 28761.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3809, pruned_loss=0.124, over 5678127.71 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3491, pruned_loss=0.09385, over 5709912.92 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3836, pruned_loss=0.1269, over 5664291.94 frames. ], batch size: 242, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:33:30,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:33:39,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4317, 1.2477, 4.0931, 3.3031], device='cuda:1'), covar=tensor([0.1698, 0.2882, 0.0469, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0718, 0.0625, 0.0923, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 05:33:47,525 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 9850, giga_loss[loss=0.281, simple_loss=0.3651, pruned_loss=0.09841, over 28971.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3807, pruned_loss=0.1233, over 5676219.02 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3493, pruned_loss=0.09402, over 5706994.31 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3835, pruned_loss=0.1262, over 5666683.26 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:34:06,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=693814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:34:10,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4832, 1.6268, 1.7300, 1.3409], device='cuda:1'), covar=tensor([0.1289, 0.1928, 0.1070, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0862, 0.0697, 0.0910, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 05:34:13,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9994, 1.3061, 1.0880, 0.1903], device='cuda:1'), covar=tensor([0.2941, 0.2239, 0.2966, 0.5145], device='cuda:1'), in_proj_covar=tensor([0.1657, 0.1567, 0.1539, 0.1348], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 05:34:33,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4757, 3.3530, 1.4769, 1.5382], device='cuda:1'), covar=tensor([0.0951, 0.0348, 0.0875, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0537, 0.0362, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 05:34:41,429 INFO [train.py:968] (1/2) Epoch 16, batch 9900, giga_loss[loss=0.3835, simple_loss=0.4289, pruned_loss=0.169, over 28726.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3832, pruned_loss=0.1266, over 5670049.21 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3491, pruned_loss=0.09403, over 5710205.89 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3861, pruned_loss=0.1294, over 5659170.45 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:35:00,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5976, 5.4136, 5.0928, 2.9463], device='cuda:1'), covar=tensor([0.0445, 0.0602, 0.0662, 0.1516], device='cuda:1'), in_proj_covar=tensor([0.1159, 0.1070, 0.0923, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 05:35:08,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.91 vs. limit=2.0
+2023-03-08 05:35:14,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693882.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:35:31,220 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 9950, giga_loss[loss=0.3239, simple_loss=0.3861, pruned_loss=0.1308, over 28618.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.381, pruned_loss=0.1258, over 5668028.30 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3487, pruned_loss=0.09377, over 5714815.67 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3845, pruned_loss=0.1291, over 5654286.10 frames. ], batch size: 242, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:36:09,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:36:20,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3812, 3.5270, 1.5370, 1.5619], device='cuda:1'), covar=tensor([0.0959, 0.0340, 0.0865, 0.1278], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0537, 0.0361, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 05:36:22,899 INFO [train.py:968] (1/2) Epoch 16, batch 10000, giga_loss[loss=0.2981, simple_loss=0.3678, pruned_loss=0.1142, over 28755.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3806, pruned_loss=0.127, over 5659188.72 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3491, pruned_loss=0.09402, over 5717690.01 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3836, pruned_loss=0.13, over 5644789.02 frames. ], batch size: 242, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:37:04,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=693993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:37:11,205 INFO [optim.py:369] (1/2) 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 05:37:12,451 INFO [train.py:968] (1/2) Epoch 16, batch 10050, giga_loss[loss=0.2604, simple_loss=0.3304, pruned_loss=0.09521, over 28633.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3784, pruned_loss=0.126, over 5671236.89 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3494, pruned_loss=0.09408, over 5719817.05 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3813, pruned_loss=0.129, over 5656744.61 frames. ], batch size: 85, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:37:32,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694025.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:37:34,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694028.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:37:44,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:37:59,055 INFO [train.py:968] (1/2) Epoch 16, batch 10100, giga_loss[loss=0.3658, simple_loss=0.4021, pruned_loss=0.1648, over 26723.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3751, pruned_loss=0.1242, over 5665590.44 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.349, pruned_loss=0.09403, over 5727410.86 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.379, pruned_loss=0.1281, over 5645027.09 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:38:08,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694057.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:38:19,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9893, 1.2926, 1.2987, 1.0523], device='cuda:1'), covar=tensor([0.1792, 0.1401, 0.2331, 0.1698], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0741, 0.0695, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 05:38:31,232 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:38:31,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3762, 1.5434, 1.2687, 1.6043], device='cuda:1'), covar=tensor([0.0758, 0.0316, 0.0326, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:1')
+2023-03-08 05:38:34,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:38:49,889 INFO [optim.py:369] (1/2) 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 05:38:52,669 INFO [train.py:968] (1/2) Epoch 16, batch 10150, giga_loss[loss=0.3222, simple_loss=0.3857, pruned_loss=0.1293, over 28740.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3751, pruned_loss=0.1251, over 5660239.41 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3493, pruned_loss=0.09425, over 5720201.01 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3782, pruned_loss=0.1283, over 5649981.23 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:39:06,165 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:20,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2202, 0.8637, 0.9156, 1.3742], device='cuda:1'), covar=tensor([0.0757, 0.0366, 0.0353, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:1')
+2023-03-08 05:39:26,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:30,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:41,102 INFO [train.py:968] (1/2) Epoch 16, batch 10200, giga_loss[loss=0.2734, simple_loss=0.3468, pruned_loss=0.1, over 29017.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3727, pruned_loss=0.1228, over 5656230.65 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3499, pruned_loss=0.0946, over 5714667.68 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3753, pruned_loss=0.1259, over 5651509.69 frames. ], batch size: 128, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:39:43,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:39:57,691 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:40:26,090 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 10250, giga_loss[loss=0.3532, simple_loss=0.3994, pruned_loss=0.1535, over 26761.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3695, pruned_loss=0.1189, over 5663277.08 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3501, pruned_loss=0.09463, over 5719709.56 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1222, over 5653296.72 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:40:55,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-08 05:41:19,161 INFO [train.py:968] (1/2) Epoch 16, batch 10300, giga_loss[loss=0.3089, simple_loss=0.3707, pruned_loss=0.1236, over 28923.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3661, pruned_loss=0.1159, over 5665276.53 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3503, pruned_loss=0.09478, over 5720374.15 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3685, pruned_loss=0.1189, over 5655068.23 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:42:03,130 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:42:06,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:42:06,410 INFO [train.py:968] (1/2) Epoch 16, batch 10350, giga_loss[loss=0.2938, simple_loss=0.3665, pruned_loss=0.1106, over 28860.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3657, pruned_loss=0.1151, over 5674720.67 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09451, over 5725425.44 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3685, pruned_loss=0.1185, over 5660281.89 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:42:36,558 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:42:39,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:42:43,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:43:01,737 INFO [train.py:968] (1/2) Epoch 16, batch 10400, giga_loss[loss=0.2741, simple_loss=0.3446, pruned_loss=0.1017, over 28690.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3636, pruned_loss=0.1149, over 5670919.04 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3496, pruned_loss=0.09427, over 5729916.98 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3664, pruned_loss=0.1183, over 5654040.92 frames. ], batch size: 66, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 05:43:05,490 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:43:11,616 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:43:21,372 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:43:51,527 INFO [train.py:968] (1/2) Epoch 16, batch 10450, giga_loss[loss=0.2717, simple_loss=0.3473, pruned_loss=0.09808, over 28949.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3619, pruned_loss=0.1146, over 5667226.72 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09434, over 5728517.55 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3642, pruned_loss=0.1174, over 5654332.79 frames. ], batch size: 145, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:44:03,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:44:06,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-08 05:44:15,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 05:44:19,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6565, 1.4739, 1.8349, 1.3281], device='cuda:1'), covar=tensor([0.1948, 0.2922, 0.1471, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0700, 0.0913, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 05:44:24,636 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:44:35,602 INFO [train.py:968] (1/2) Epoch 16, batch 10500, giga_loss[loss=0.3233, simple_loss=0.3795, pruned_loss=0.1336, over 28586.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3654, pruned_loss=0.1163, over 5674334.19 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3498, pruned_loss=0.09434, over 5733489.78 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3676, pruned_loss=0.1192, over 5657897.22 frames. ], batch size: 92, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:45:12,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-08 05:45:26,021 INFO [optim.py:369] (1/2) 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 05:45:26,033 INFO [train.py:968] (1/2) Epoch 16, batch 10550, giga_loss[loss=0.3019, simple_loss=0.3706, pruned_loss=0.1166, over 28670.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3677, pruned_loss=0.1176, over 5667943.94 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3492, pruned_loss=0.09401, over 5737036.84 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3704, pruned_loss=0.1208, over 5649947.93 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:46:15,792 INFO [train.py:968] (1/2) Epoch 16, batch 10600, giga_loss[loss=0.391, simple_loss=0.4204, pruned_loss=0.1808, over 26594.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3661, pruned_loss=0.1165, over 5654308.71 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3495, pruned_loss=0.09419, over 5730852.08 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3683, pruned_loss=0.1194, over 5643973.74 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:46:22,448 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:46:25,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:46:28,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9931, 1.2013, 3.3567, 2.8760], device='cuda:1'), covar=tensor([0.1714, 0.2660, 0.0507, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0718, 0.0625, 0.0918, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 05:46:53,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:46:55,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5936, 2.4996, 1.9683, 2.2153], device='cuda:1'), covar=tensor([0.0741, 0.0626, 0.0895, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0445, 0.0507, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 05:47:04,362 INFO [optim.py:369] (1/2) 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 05:47:04,378 INFO [train.py:968] (1/2) Epoch 16, batch 10650, giga_loss[loss=0.3621, simple_loss=0.3965, pruned_loss=0.1638, over 23661.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3661, pruned_loss=0.1169, over 5650770.88 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3498, pruned_loss=0.09439, over 5726885.50 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1197, over 5644384.40 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:47:39,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:47:54,383 INFO [train.py:968] (1/2) Epoch 16, batch 10700, giga_loss[loss=0.2718, simple_loss=0.3501, pruned_loss=0.09675, over 28998.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3684, pruned_loss=0.1187, over 5658336.03 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3499, pruned_loss=0.09428, over 5724641.48 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3706, pruned_loss=0.1218, over 5653035.23 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:48:02,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0289, 1.3274, 1.0843, 0.2433], device='cuda:1'), covar=tensor([0.2906, 0.2546, 0.3807, 0.5006], device='cuda:1'), in_proj_covar=tensor([0.1664, 0.1574, 0.1544, 0.1359], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 05:48:42,961 INFO [optim.py:369] (1/2) 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 05:48:42,974 INFO [train.py:968] (1/2) Epoch 16, batch 10750, giga_loss[loss=0.3399, simple_loss=0.396, pruned_loss=0.1419, over 28880.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.37, pruned_loss=0.1192, over 5658628.70 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3501, pruned_loss=0.09445, over 5724889.29 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3722, pruned_loss=0.1224, over 5652023.16 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:49:12,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:49:26,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-08 05:49:27,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:49:29,163 INFO [train.py:968] (1/2) Epoch 16, batch 10800, giga_loss[loss=0.3996, simple_loss=0.4319, pruned_loss=0.1836, over 27661.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3716, pruned_loss=0.1205, over 5666020.27 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3498, pruned_loss=0.09422, over 5731359.39 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1243, over 5652306.55 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 05:50:17,956 INFO [train.py:968] (1/2) Epoch 16, batch 10850, giga_loss[loss=0.2957, simple_loss=0.3598, pruned_loss=0.1159, over 28707.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3729, pruned_loss=0.1218, over 5672811.59 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3497, pruned_loss=0.09412, over 5732090.25 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3758, pruned_loss=0.1255, over 5659896.07 frames. ], batch size: 92, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:50:18,693 INFO [optim.py:369] (1/2) 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 05:50:28,501 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=694812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:08,325 INFO [train.py:968] (1/2) Epoch 16, batch 10900, giga_loss[loss=0.2958, simple_loss=0.3718, pruned_loss=0.1099, over 28081.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3748, pruned_loss=0.1228, over 5680142.85 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3502, pruned_loss=0.0945, over 5734879.22 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3771, pruned_loss=0.126, over 5666224.49 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:51:09,404 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 05:51:25,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8064, 4.5891, 4.3691, 2.3263], device='cuda:1'), covar=tensor([0.0621, 0.0848, 0.1029, 0.1896], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.1070, 0.0926, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 05:51:32,741 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:35,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:43,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=694884.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 05:51:53,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:51:56,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:52:03,039 INFO [train.py:968] (1/2) Epoch 16, batch 10950, giga_loss[loss=0.3089, simple_loss=0.3591, pruned_loss=0.1293, over 23860.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3744, pruned_loss=0.1217, over 5669797.65 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3505, pruned_loss=0.09466, over 5738476.10 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3764, pruned_loss=0.1246, over 5654501.93 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:52:04,932 INFO [optim.py:369] (1/2) 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 05:52:07,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:52:29,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=694924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:52:57,390 INFO [train.py:968] (1/2) Epoch 16, batch 11000, giga_loss[loss=0.3177, simple_loss=0.379, pruned_loss=0.1282, over 28636.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3752, pruned_loss=0.123, over 5661681.55 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3507, pruned_loss=0.09476, over 5740382.34 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3771, pruned_loss=0.1258, over 5646226.84 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:53:01,358 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=694955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:53:04,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=694958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:53:36,197 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 11050, giga_loss[loss=0.3505, simple_loss=0.3951, pruned_loss=0.153, over 28324.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3747, pruned_loss=0.1234, over 5642421.87 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3509, pruned_loss=0.09491, over 5733813.18 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3765, pruned_loss=0.126, over 5635281.27 frames. ], batch size: 368, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:53:51,247 INFO [optim.py:369] (1/2) 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 05:54:01,652 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 11100, giga_loss[loss=0.2562, simple_loss=0.3319, pruned_loss=0.09025, over 28368.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3719, pruned_loss=0.1219, over 5643079.02 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3504, pruned_loss=0.09467, over 5730198.90 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3745, pruned_loss=0.1251, over 5638430.95 frames. ], batch size: 71, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:54:48,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 05:55:24,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5647, 1.8032, 1.4528, 1.5124], device='cuda:1'), covar=tensor([0.2365, 0.2419, 0.2714, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1400, 0.1022, 0.1238, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 05:55:32,513 INFO [train.py:968] (1/2) Epoch 16, batch 11150, giga_loss[loss=0.2818, simple_loss=0.3518, pruned_loss=0.1059, over 28993.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5644979.12 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3502, pruned_loss=0.09453, over 5734922.77 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3724, pruned_loss=0.1242, over 5634678.78 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 05:55:34,945 INFO [optim.py:369] (1/2) 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 05:56:21,209 INFO [train.py:968] (1/2) Epoch 16, batch 11200, giga_loss[loss=0.3276, simple_loss=0.3845, pruned_loss=0.1353, over 28569.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5661735.73 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09447, over 5738231.98 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3731, pruned_loss=0.1247, over 5649032.98 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:56:23,359 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=695153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:56:27,639 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=695156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:56:53,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=695185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:56:56,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0046, 2.2986, 2.0554, 1.7576], device='cuda:1'), covar=tensor([0.2393, 0.1841, 0.2080, 0.2201], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1791, 0.1740, 0.1855], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 05:57:10,843 INFO [train.py:968] (1/2) Epoch 16, batch 11250, giga_loss[loss=0.3197, simple_loss=0.3742, pruned_loss=0.1326, over 28894.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3697, pruned_loss=0.1215, over 5662363.66 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3497, pruned_loss=0.09419, over 5743038.67 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3728, pruned_loss=0.1252, over 5645882.06 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:57:12,238 INFO [optim.py:369] (1/2) 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 05:57:23,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3513, 1.5454, 1.6203, 1.2570], device='cuda:1'), covar=tensor([0.1167, 0.1841, 0.0990, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0702, 0.0915, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 05:57:39,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-08 05:57:59,350 INFO [train.py:968] (1/2) Epoch 16, batch 11300, giga_loss[loss=0.2824, simple_loss=0.3549, pruned_loss=0.105, over 28565.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3695, pruned_loss=0.1214, over 5665875.92 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3491, pruned_loss=0.0937, over 5747129.18 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3732, pruned_loss=0.1256, over 5646681.56 frames. ], batch size: 85, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:58:08,268 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=695257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 05:58:09,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=695259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 05:58:48,974 INFO [train.py:968] (1/2) Epoch 16, batch 11350, giga_loss[loss=0.3462, simple_loss=0.4099, pruned_loss=0.1413, over 28309.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3728, pruned_loss=0.124, over 5669600.84 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3492, pruned_loss=0.09379, over 5750550.04 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3761, pruned_loss=0.1278, over 5649623.35 frames. ], batch size: 368, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:58:50,462 INFO [optim.py:369] (1/2) 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 05:58:56,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2127, 1.2834, 3.5246, 3.1032], device='cuda:1'), covar=tensor([0.1559, 0.2701, 0.0470, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0715, 0.0624, 0.0917, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 05:59:34,693 INFO [train.py:968] (1/2) Epoch 16, batch 11400, giga_loss[loss=0.3485, simple_loss=0.4048, pruned_loss=0.1461, over 28582.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1253, over 5652852.68 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3492, pruned_loss=0.09392, over 5745706.16 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3774, pruned_loss=0.1295, over 5636597.12 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 05:59:58,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5858, 1.8650, 1.4771, 1.7006], device='cuda:1'), covar=tensor([0.2331, 0.2389, 0.2721, 0.2178], device='cuda:1'), in_proj_covar=tensor([0.1409, 0.1031, 0.1246, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 06:00:25,320 INFO [train.py:968] (1/2) Epoch 16, batch 11450, giga_loss[loss=0.2883, simple_loss=0.3615, pruned_loss=0.1076, over 28982.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3749, pruned_loss=0.1269, over 5652181.81 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09382, over 5742159.18 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3788, pruned_loss=0.1313, over 5639882.77 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:00:26,355 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=695402.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:00:26,607 INFO [optim.py:369] (1/2) 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 06:00:29,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=695405.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 06:00:34,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3595, 1.5971, 1.6600, 1.2243], device='cuda:1'), covar=tensor([0.1526, 0.2320, 0.1280, 0.1528], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0701, 0.0913, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 06:00:35,594 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 06:00:58,425 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 11500, giga_loss[loss=0.3374, simple_loss=0.3926, pruned_loss=0.1412, over 29042.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3742, pruned_loss=0.1262, over 5658730.63 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.349, pruned_loss=0.09392, over 5742694.38 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3773, pruned_loss=0.1297, over 5648249.26 frames. ], batch size: 106, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:02:06,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 06:02:06,363 INFO [train.py:968] (1/2) Epoch 16, batch 11550, giga_loss[loss=0.2948, simple_loss=0.3669, pruned_loss=0.1114, over 28453.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3748, pruned_loss=0.1262, over 5656455.00 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.349, pruned_loss=0.09386, over 5745714.62 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3777, pruned_loss=0.1296, over 5643791.88 frames. ], batch size: 71, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:02:08,149 INFO [optim.py:369] (1/2) 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 06:02:48,965 INFO [train.py:968] (1/2) Epoch 16, batch 11600, giga_loss[loss=0.2744, simple_loss=0.3517, pruned_loss=0.09855, over 28840.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3746, pruned_loss=0.125, over 5675451.72 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09385, over 5750469.35 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3778, pruned_loss=0.1288, over 5658458.90 frames. ], batch size: 99, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:03:41,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3801, 2.0231, 1.5331, 0.6518], device='cuda:1'), covar=tensor([0.4333, 0.2431, 0.3245, 0.5048], device='cuda:1'), in_proj_covar=tensor([0.1660, 0.1577, 0.1547, 0.1361], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 06:03:45,639 INFO [train.py:968] (1/2) Epoch 16, batch 11650, giga_loss[loss=0.298, simple_loss=0.3633, pruned_loss=0.1164, over 28673.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3768, pruned_loss=0.1274, over 5658172.73 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.349, pruned_loss=0.09383, over 5747564.00 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3794, pruned_loss=0.1305, over 5647060.08 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:03:46,949 INFO [optim.py:369] (1/2) 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 06:03:49,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3546, 3.3093, 1.5620, 1.5118], device='cuda:1'), covar=tensor([0.0949, 0.0363, 0.0836, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0536, 0.0360, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 06:04:12,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5387, 1.2970, 4.7894, 3.5334], device='cuda:1'), covar=tensor([0.1657, 0.2808, 0.0432, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0712, 0.0619, 0.0912, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 06:04:19,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=695632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:04:26,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4120, 1.7150, 1.3727, 1.6347], device='cuda:1'), covar=tensor([0.0650, 0.0402, 0.0320, 0.0662], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:1')
+2023-03-08 06:04:38,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9410, 1.1731, 1.3578, 0.9587], device='cuda:1'), covar=tensor([0.1676, 0.1341, 0.2047, 0.1643], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0745, 0.0700, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 06:04:38,474 INFO [train.py:968] (1/2) Epoch 16, batch 11700, giga_loss[loss=0.3131, simple_loss=0.3755, pruned_loss=0.1253, over 29034.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3779, pruned_loss=0.1282, over 5658882.69 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3493, pruned_loss=0.09392, over 5748878.39 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3798, pruned_loss=0.1309, over 5648194.19 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:04:47,798 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 11750, giga_loss[loss=0.3743, simple_loss=0.3984, pruned_loss=0.1751, over 23740.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3778, pruned_loss=0.1286, over 5648579.59 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3492, pruned_loss=0.09396, over 5742450.46 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3806, pruned_loss=0.132, over 5641977.40 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:05:26,859 INFO [optim.py:369] (1/2) 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 06:05:45,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=695723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:05:58,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8119, 3.6227, 3.4286, 1.9290], device='cuda:1'), covar=tensor([0.0713, 0.0910, 0.0905, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.1161, 0.1071, 0.0923, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 06:06:15,410 INFO [train.py:968] (1/2) Epoch 16, batch 11800, libri_loss[loss=0.2513, simple_loss=0.3336, pruned_loss=0.08457, over 29466.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3779, pruned_loss=0.1272, over 5650079.17 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09382, over 5744649.18 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3808, pruned_loss=0.1305, over 5641614.65 frames. ], batch size: 70, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:06:37,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=695775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:06:40,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=695778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:07:05,733 INFO [train.py:968] (1/2) Epoch 16, batch 11850, giga_loss[loss=0.3239, simple_loss=0.3867, pruned_loss=0.1306, over 28721.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3772, pruned_loss=0.1261, over 5653400.40 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3488, pruned_loss=0.09374, over 5748018.86 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3801, pruned_loss=0.1294, over 5642021.15 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:07:08,529 INFO [optim.py:369] (1/2) 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 06:07:10,765 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 11900, giga_loss[loss=0.3547, simple_loss=0.3869, pruned_loss=0.1613, over 23706.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3763, pruned_loss=0.1255, over 5649819.76 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.349, pruned_loss=0.09376, over 5749940.67 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3789, pruned_loss=0.1286, over 5637835.06 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:08:33,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0141, 1.3915, 1.1581, 0.1544], device='cuda:1'), covar=tensor([0.3305, 0.2887, 0.3961, 0.5711], device='cuda:1'), in_proj_covar=tensor([0.1656, 0.1573, 0.1543, 0.1358], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 06:08:42,409 INFO [train.py:968] (1/2) Epoch 16, batch 11950, giga_loss[loss=0.2927, simple_loss=0.3588, pruned_loss=0.1133, over 28699.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3748, pruned_loss=0.1246, over 5665621.27 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3489, pruned_loss=0.09396, over 5753344.19 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3775, pruned_loss=0.1276, over 5651069.05 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 2.0
+2023-03-08 06:08:48,111 INFO [optim.py:369] (1/2) 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 06:09:35,574 INFO [train.py:968] (1/2) Epoch 16, batch 12000, giga_loss[loss=0.3731, simple_loss=0.3954, pruned_loss=0.1754, over 23600.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3761, pruned_loss=0.1257, over 5645803.70 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3489, pruned_loss=0.09399, over 5752006.66 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3785, pruned_loss=0.1285, over 5634035.35 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:09:35,574 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 06:09:43,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7458, 3.4967, 3.3792, 1.6921], device='cuda:1'), covar=tensor([0.0770, 0.0909, 0.0851, 0.2212], device='cuda:1'), in_proj_covar=tensor([0.1165, 0.1074, 0.0926, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 06:09:44,945 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 06:10:00,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2089, 1.5460, 1.5086, 1.1001], device='cuda:1'), covar=tensor([0.1485, 0.2193, 0.1202, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.0868, 0.0702, 0.0914, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 06:10:29,227 INFO [train.py:968] (1/2) Epoch 16, batch 12050, giga_loss[loss=0.3973, simple_loss=0.4202, pruned_loss=0.1871, over 26694.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3762, pruned_loss=0.1256, over 5661782.38 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3492, pruned_loss=0.09397, over 5755947.50 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3788, pruned_loss=0.1289, over 5645434.11 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:10:31,969 INFO [optim.py:369] (1/2) 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 06:11:00,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:11:11,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 06:11:11,632 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:11:19,049 INFO [train.py:968] (1/2) Epoch 16, batch 12100, giga_loss[loss=0.3064, simple_loss=0.3659, pruned_loss=0.1235, over 28655.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.1239, over 5679157.69 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.09354, over 5759450.95 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3776, pruned_loss=0.1283, over 5659432.47 frames. ], batch size: 242, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:11:46,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-08 06:12:09,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:12:10,899 INFO [train.py:968] (1/2) Epoch 16, batch 12150, libri_loss[loss=0.2649, simple_loss=0.3508, pruned_loss=0.08947, over 29651.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3754, pruned_loss=0.1256, over 5674675.21 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09345, over 5761892.79 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3788, pruned_loss=0.1296, over 5655730.66 frames. ], batch size: 88, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:12:13,938 INFO [optim.py:369] (1/2) 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 06:13:00,538 INFO [train.py:968] (1/2) Epoch 16, batch 12200, giga_loss[loss=0.3306, simple_loss=0.3704, pruned_loss=0.1454, over 23521.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.377, pruned_loss=0.1266, over 5678261.30 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3488, pruned_loss=0.09369, over 5765122.31 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3802, pruned_loss=0.1304, over 5657947.86 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:13:17,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:13:25,229 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:13:28,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:13:34,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4994, 1.7771, 1.3771, 1.6001], device='cuda:1'), covar=tensor([0.2481, 0.2424, 0.2834, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1029, 0.1248, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 06:13:47,934 INFO [train.py:968] (1/2) Epoch 16, batch 12250, giga_loss[loss=0.3904, simple_loss=0.4188, pruned_loss=0.181, over 26677.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3785, pruned_loss=0.1283, over 5669154.81 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09355, over 5764204.48 frames. ], giga_tot_loss[loss=0.323, simple_loss=0.3818, pruned_loss=0.1321, over 5652166.50 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:13:51,725 INFO [optim.py:369] (1/2) 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 06:13:56,637 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:12,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:28,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:33,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:14:39,198 INFO [train.py:968] (1/2) Epoch 16, batch 12300, libri_loss[loss=0.2137, simple_loss=0.3079, pruned_loss=0.05976, over 29553.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3774, pruned_loss=0.1266, over 5685824.81 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3484, pruned_loss=0.09335, over 5767489.41 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3809, pruned_loss=0.1306, over 5667308.28 frames. ], batch size: 79, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:15:00,243 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:15:25,817 INFO [train.py:968] (1/2) Epoch 16, batch 12350, giga_loss[loss=0.2943, simple_loss=0.3649, pruned_loss=0.1119, over 28713.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3761, pruned_loss=0.125, over 5681229.49 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.0934, over 5771900.83 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3796, pruned_loss=0.1291, over 5659931.81 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:15:29,332 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 12400, giga_loss[loss=0.3876, simple_loss=0.4184, pruned_loss=0.1784, over 26688.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3753, pruned_loss=0.124, over 5679108.90 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3484, pruned_loss=0.09333, over 5764612.24 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3787, pruned_loss=0.1278, over 5667237.60 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:16:26,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-08 06:17:00,933 INFO [train.py:968] (1/2) Epoch 16, batch 12450, giga_loss[loss=0.2811, simple_loss=0.3438, pruned_loss=0.1092, over 28237.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3742, pruned_loss=0.1234, over 5673895.05 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.09361, over 5767486.54 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3772, pruned_loss=0.127, over 5659624.17 frames. ], batch size: 77, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:17:05,995 INFO [optim.py:369] (1/2) 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 06:17:17,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=696418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:17:39,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6517, 1.7292, 1.7334, 1.5711], device='cuda:1'), covar=tensor([0.2576, 0.2100, 0.1778, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1833, 0.1763, 0.1705, 0.1827], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 06:17:48,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4350, 1.8650, 1.3971, 1.7545], device='cuda:1'), covar=tensor([0.2741, 0.2695, 0.3108, 0.2354], device='cuda:1'), in_proj_covar=tensor([0.1408, 0.1032, 0.1249, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 06:17:51,747 INFO [train.py:968] (1/2) Epoch 16, batch 12500, giga_loss[loss=0.2956, simple_loss=0.3413, pruned_loss=0.1249, over 23277.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3738, pruned_loss=0.1237, over 5668191.91 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.349, pruned_loss=0.09367, over 5769819.42 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.1271, over 5653102.93 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:18:06,659 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 06:18:34,263 INFO [train.py:968] (1/2) Epoch 16, batch 12550, giga_loss[loss=0.2815, simple_loss=0.3452, pruned_loss=0.1089, over 28418.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3704, pruned_loss=0.1215, over 5680907.12 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3492, pruned_loss=0.0939, over 5771596.81 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3734, pruned_loss=0.1253, over 5663135.44 frames. ], batch size: 60, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:18:39,065 INFO [optim.py:369] (1/2) 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 06:19:02,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-08 06:19:18,376 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 12600, giga_loss[loss=0.3492, simple_loss=0.3942, pruned_loss=0.1521, over 27819.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3673, pruned_loss=0.1202, over 5691713.45 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3492, pruned_loss=0.09395, over 5773465.16 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3699, pruned_loss=0.1235, over 5674789.27 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:19:37,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:19:40,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:19:53,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-08 06:20:10,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:20:14,186 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 12650, libri_loss[loss=0.2808, simple_loss=0.3598, pruned_loss=0.1009, over 29742.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3676, pruned_loss=0.1211, over 5689323.60 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3498, pruned_loss=0.09442, over 5765431.10 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3693, pruned_loss=0.1236, over 5681397.91 frames. ], batch size: 87, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:20:18,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 06:20:21,389 INFO [optim.py:369] (1/2) 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 06:21:01,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:21:07,071 INFO [train.py:968] (1/2) Epoch 16, batch 12700, giga_loss[loss=0.3236, simple_loss=0.3869, pruned_loss=0.1301, over 27649.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3664, pruned_loss=0.12, over 5684247.54 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09451, over 5764909.55 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3679, pruned_loss=0.1224, over 5677131.88 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:21:24,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6008, 1.8167, 1.2857, 1.4225], device='cuda:1'), covar=tensor([0.0870, 0.0526, 0.0981, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0449, 0.0507, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 06:21:41,093 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:21:43,353 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:21:51,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4363, 1.9077, 1.3574, 0.7300], device='cuda:1'), covar=tensor([0.4502, 0.2543, 0.3073, 0.4892], device='cuda:1'), in_proj_covar=tensor([0.1656, 0.1576, 0.1544, 0.1360], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 06:21:54,267 INFO [train.py:968] (1/2) Epoch 16, batch 12750, giga_loss[loss=0.282, simple_loss=0.3636, pruned_loss=0.1002, over 28909.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5683772.94 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3497, pruned_loss=0.0944, over 5764868.21 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3668, pruned_loss=0.12, over 5674301.81 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:21:58,517 INFO [optim.py:369] (1/2) 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 06:22:09,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:27,702 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:31,482 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=696741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:33,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3692, 3.2687, 1.4678, 1.5446], device='cuda:1'), covar=tensor([0.0960, 0.0321, 0.0986, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0540, 0.0363, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 06:22:35,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=696744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:22:42,932 INFO [train.py:968] (1/2) Epoch 16, batch 12800, libri_loss[loss=0.2719, simple_loss=0.3526, pruned_loss=0.09563, over 29365.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3626, pruned_loss=0.1137, over 5668910.75 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3493, pruned_loss=0.0943, over 5758508.87 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3652, pruned_loss=0.117, over 5664400.06 frames. ], batch size: 92, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:23:04,088 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=696773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:23:07,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2469, 1.1818, 3.7351, 3.2498], device='cuda:1'), covar=tensor([0.1661, 0.2976, 0.0440, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.0714, 0.0623, 0.0914, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 06:23:31,483 INFO [train.py:968] (1/2) Epoch 16, batch 12850, giga_loss[loss=0.2784, simple_loss=0.3478, pruned_loss=0.1045, over 28637.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3601, pruned_loss=0.1108, over 5674117.36 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3487, pruned_loss=0.09407, over 5761355.79 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3632, pruned_loss=0.1145, over 5664231.56 frames. ], batch size: 92, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:23:39,813 INFO [optim.py:369] (1/2) 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 06:24:09,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4462, 1.6042, 1.2288, 1.2239], device='cuda:1'), covar=tensor([0.0887, 0.0460, 0.0983, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0445, 0.0507, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 06:24:26,191 INFO [train.py:968] (1/2) Epoch 16, batch 12900, giga_loss[loss=0.3021, simple_loss=0.3679, pruned_loss=0.1181, over 28938.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3568, pruned_loss=0.1077, over 5670000.60 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3486, pruned_loss=0.09401, over 5763145.29 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3596, pruned_loss=0.1108, over 5659283.58 frames. ], batch size: 199, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:24:45,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2255, 1.3186, 1.1856, 0.9823], device='cuda:1'), covar=tensor([0.0920, 0.0472, 0.1022, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0444, 0.0505, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 06:25:18,595 INFO [train.py:968] (1/2) Epoch 16, batch 12950, giga_loss[loss=0.2265, simple_loss=0.3123, pruned_loss=0.07032, over 28322.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.354, pruned_loss=0.1045, over 5666185.59 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3485, pruned_loss=0.09405, over 5755404.93 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3563, pruned_loss=0.1071, over 5663118.79 frames. ], batch size: 77, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:25:24,009 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.31 vs. limit=5.0
+2023-03-08 06:25:24,136 INFO [optim.py:369] (1/2) 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 06:25:42,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=696924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:25:51,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3700, 1.5557, 1.6283, 1.2285], device='cuda:1'), covar=tensor([0.1829, 0.2671, 0.1555, 0.1814], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0695, 0.0910, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 06:26:07,676 INFO [train.py:968] (1/2) Epoch 16, batch 13000, libri_loss[loss=0.2579, simple_loss=0.3496, pruned_loss=0.08311, over 29462.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3528, pruned_loss=0.1021, over 5664534.86 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3482, pruned_loss=0.09398, over 5754463.74 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3552, pruned_loss=0.1046, over 5659670.07 frames. ], batch size: 85, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:27:03,009 INFO [train.py:968] (1/2) Epoch 16, batch 13050, giga_loss[loss=0.2696, simple_loss=0.344, pruned_loss=0.09757, over 28923.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3531, pruned_loss=0.1022, over 5658721.15 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3478, pruned_loss=0.09384, over 5755856.18 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3554, pruned_loss=0.1044, over 5653086.58 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:27:10,202 INFO [optim.py:369] (1/2) 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 06:27:22,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:27:52,851 INFO [train.py:968] (1/2) Epoch 16, batch 13100, giga_loss[loss=0.2535, simple_loss=0.3326, pruned_loss=0.08716, over 28848.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3502, pruned_loss=0.1003, over 5655709.60 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3469, pruned_loss=0.09362, over 5747744.27 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3529, pruned_loss=0.1025, over 5655411.19 frames. ], batch size: 285, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:28:14,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7717, 2.8361, 1.6865, 0.7582], device='cuda:1'), covar=tensor([0.7050, 0.3281, 0.4243, 0.6696], device='cuda:1'), in_proj_covar=tensor([0.1651, 0.1571, 0.1540, 0.1361], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 06:28:40,506 INFO [train.py:968] (1/2) Epoch 16, batch 13150, libri_loss[loss=0.1836, simple_loss=0.2625, pruned_loss=0.05238, over 29664.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3465, pruned_loss=0.09787, over 5668522.69 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3455, pruned_loss=0.09308, over 5755268.62 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1005, over 5656504.54 frames. ], batch size: 69, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:28:45,835 INFO [optim.py:369] (1/2) 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 06:28:51,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:29:30,312 INFO [train.py:968] (1/2) Epoch 16, batch 13200, libri_loss[loss=0.2751, simple_loss=0.351, pruned_loss=0.09963, over 26483.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3448, pruned_loss=0.09694, over 5669622.65 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3449, pruned_loss=0.093, over 5756954.55 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3482, pruned_loss=0.09916, over 5656264.85 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:29:44,823 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:29:48,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:30:14,167 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:30:14,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:30:19,533 INFO [train.py:968] (1/2) Epoch 16, batch 13250, giga_loss[loss=0.2415, simple_loss=0.3284, pruned_loss=0.07728, over 28822.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3443, pruned_loss=0.0961, over 5676685.61 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3443, pruned_loss=0.09272, over 5760862.80 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3476, pruned_loss=0.09821, over 5660561.40 frames. ], batch size: 145, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:30:21,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4284, 1.6651, 1.5546, 1.4609], device='cuda:1'), covar=tensor([0.2225, 0.1754, 0.1286, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1747, 0.1688, 0.1813], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 06:30:23,616 INFO [optim.py:369] (1/2) 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 06:31:06,686 INFO [train.py:968] (1/2) Epoch 16, batch 13300, giga_loss[loss=0.2365, simple_loss=0.3253, pruned_loss=0.07384, over 28932.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3435, pruned_loss=0.09506, over 5674086.90 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3443, pruned_loss=0.09282, over 5761251.22 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09673, over 5658612.83 frames. ], batch size: 145, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:31:12,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:31:16,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:31:51,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:32:03,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:32:04,026 INFO [train.py:968] (1/2) Epoch 16, batch 13350, giga_loss[loss=0.2331, simple_loss=0.3192, pruned_loss=0.07347, over 28071.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3412, pruned_loss=0.09311, over 5672633.64 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3444, pruned_loss=0.09284, over 5762540.06 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3432, pruned_loss=0.09441, over 5658532.84 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:32:10,187 INFO [optim.py:369] (1/2) 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 06:32:22,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2872, 2.8925, 1.5431, 1.4492], device='cuda:1'), covar=tensor([0.0917, 0.0333, 0.0877, 0.1307], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0533, 0.0360, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 06:33:02,159 INFO [train.py:968] (1/2) Epoch 16, batch 13400, giga_loss[loss=0.2449, simple_loss=0.3226, pruned_loss=0.08359, over 28639.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3376, pruned_loss=0.09124, over 5666148.48 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3439, pruned_loss=0.09264, over 5763809.18 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3395, pruned_loss=0.09245, over 5652447.99 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:33:02,929 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:34:01,579 INFO [train.py:968] (1/2) Epoch 16, batch 13450, giga_loss[loss=0.275, simple_loss=0.3272, pruned_loss=0.1114, over 23862.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.336, pruned_loss=0.09132, over 5645275.51 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3439, pruned_loss=0.09264, over 5763809.18 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3375, pruned_loss=0.09227, over 5634612.24 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:34:07,972 INFO [optim.py:369] (1/2) 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 06:34:38,287 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-08 06:34:46,646 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:34:50,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:34:55,302 INFO [train.py:968] (1/2) Epoch 16, batch 13500, libri_loss[loss=0.2795, simple_loss=0.3577, pruned_loss=0.1007, over 29121.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3358, pruned_loss=0.09176, over 5651419.51 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3438, pruned_loss=0.09261, over 5766296.99 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3369, pruned_loss=0.09251, over 5638796.82 frames. ], batch size: 101, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:35:25,428 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:35:42,419 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-08 06:35:56,538 INFO [train.py:968] (1/2) Epoch 16, batch 13550, giga_loss[loss=0.2505, simple_loss=0.3398, pruned_loss=0.08063, over 28811.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3387, pruned_loss=0.09324, over 5648337.73 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3443, pruned_loss=0.09302, over 5768023.01 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.339, pruned_loss=0.09344, over 5635157.52 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:36:04,781 INFO [optim.py:369] (1/2) 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 06:36:15,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5017, 1.6556, 1.7948, 1.3402], device='cuda:1'), covar=tensor([0.1974, 0.2574, 0.1553, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0688, 0.0906, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 06:36:26,138 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 13600, giga_loss[loss=0.271, simple_loss=0.3528, pruned_loss=0.09457, over 28967.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3409, pruned_loss=0.09337, over 5653980.81 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3441, pruned_loss=0.09307, over 5771169.01 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3414, pruned_loss=0.09348, over 5637810.00 frames. ], batch size: 285, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:37:00,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4670, 2.7929, 1.5824, 1.5870], device='cuda:1'), covar=tensor([0.0778, 0.0268, 0.0800, 0.1166], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0532, 0.0360, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 06:37:23,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:37:58,427 INFO [train.py:968] (1/2) Epoch 16, batch 13650, giga_loss[loss=0.257, simple_loss=0.3328, pruned_loss=0.09054, over 27684.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3415, pruned_loss=0.09433, over 5646127.90 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3437, pruned_loss=0.09302, over 5771908.54 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3421, pruned_loss=0.09451, over 5630464.77 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:38:08,106 INFO [optim.py:369] (1/2) 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 06:38:26,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697621.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 06:39:01,527 INFO [train.py:968] (1/2) Epoch 16, batch 13700, giga_loss[loss=0.2435, simple_loss=0.3247, pruned_loss=0.08116, over 28907.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3402, pruned_loss=0.09358, over 5645823.61 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3436, pruned_loss=0.09321, over 5761174.00 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3407, pruned_loss=0.09358, over 5638841.16 frames. ], batch size: 284, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:39:20,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5634, 1.7536, 1.4344, 1.5870], device='cuda:1'), covar=tensor([0.2829, 0.2668, 0.3073, 0.2498], device='cuda:1'), in_proj_covar=tensor([0.1404, 0.1022, 0.1245, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 06:40:05,102 INFO [train.py:968] (1/2) Epoch 16, batch 13750, giga_loss[loss=0.2636, simple_loss=0.3467, pruned_loss=0.09027, over 28811.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3381, pruned_loss=0.09125, over 5645000.64 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3434, pruned_loss=0.09303, over 5762620.67 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3386, pruned_loss=0.09139, over 5636089.03 frames. ], batch size: 243, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:40:16,093 INFO [optim.py:369] (1/2) 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 06:40:23,412 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:40:26,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:40:36,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:40:59,442 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=697746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:41:05,194 INFO [train.py:968] (1/2) Epoch 16, batch 13800, giga_loss[loss=0.1974, simple_loss=0.2734, pruned_loss=0.06071, over 24425.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.08916, over 5651363.68 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3421, pruned_loss=0.09249, over 5767074.05 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3371, pruned_loss=0.0897, over 5637121.36 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:42:07,801 INFO [train.py:968] (1/2) Epoch 16, batch 13850, giga_loss[loss=0.2615, simple_loss=0.3398, pruned_loss=0.09163, over 28387.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3331, pruned_loss=0.0889, over 5650112.38 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3415, pruned_loss=0.09221, over 5760841.94 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3346, pruned_loss=0.08949, over 5640042.99 frames. ], batch size: 368, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:42:17,906 INFO [optim.py:369] (1/2) 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 06:42:26,480 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=697814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:43:04,109 INFO [train.py:968] (1/2) Epoch 16, batch 13900, libri_loss[loss=0.2993, simple_loss=0.3589, pruned_loss=0.1198, over 18599.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3337, pruned_loss=0.08976, over 5650094.61 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3411, pruned_loss=0.0922, over 5752516.56 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.335, pruned_loss=0.09013, over 5646028.22 frames. ], batch size: 186, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:43:26,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=697870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:43:29,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4705, 3.5706, 1.5995, 1.8058], device='cuda:1'), covar=tensor([0.0962, 0.0230, 0.0934, 0.1221], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0532, 0.0362, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 06:43:29,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=697873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:44:01,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:44:03,553 INFO [train.py:968] (1/2) Epoch 16, batch 13950, giga_loss[loss=0.2408, simple_loss=0.3285, pruned_loss=0.07651, over 29073.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3337, pruned_loss=0.08972, over 5654888.05 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3406, pruned_loss=0.09212, over 5746094.10 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3349, pruned_loss=0.09004, over 5656106.16 frames. ], batch size: 136, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:44:05,980 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:45:02,812 INFO [train.py:968] (1/2) Epoch 16, batch 14000, giga_loss[loss=0.2262, simple_loss=0.3117, pruned_loss=0.0703, over 28378.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3354, pruned_loss=0.0899, over 5664338.67 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3411, pruned_loss=0.09249, over 5750200.55 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3358, pruned_loss=0.08975, over 5658918.66 frames. ], batch size: 78, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:46:00,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=697996.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:46:06,624 INFO [train.py:968] (1/2) Epoch 16, batch 14050, giga_loss[loss=0.24, simple_loss=0.3088, pruned_loss=0.08558, over 24764.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3374, pruned_loss=0.09031, over 5673726.78 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3411, pruned_loss=0.09251, over 5753185.54 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09014, over 5665414.58 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:46:17,075 INFO [optim.py:369] (1/2) 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 06:47:03,337 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:47:07,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=698045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:47:12,963 INFO [train.py:968] (1/2) Epoch 16, batch 14100, giga_loss[loss=0.2675, simple_loss=0.3447, pruned_loss=0.09515, over 28980.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3341, pruned_loss=0.08854, over 5681594.22 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3407, pruned_loss=0.09223, over 5756430.71 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3346, pruned_loss=0.0886, over 5670387.11 frames. ], batch size: 285, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:47:43,832 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:48:05,546 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=698090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:48:18,249 INFO [train.py:968] (1/2) Epoch 16, batch 14150, giga_loss[loss=0.2797, simple_loss=0.3549, pruned_loss=0.1023, over 28790.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3358, pruned_loss=0.08994, over 5678731.46 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3404, pruned_loss=0.09222, over 5757143.80 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3363, pruned_loss=0.0899, over 5666534.64 frames. ], batch size: 174, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:48:30,911 INFO [optim.py:369] (1/2) 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 06:49:15,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 06:49:17,792 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 14200, giga_loss[loss=0.2774, simple_loss=0.3725, pruned_loss=0.09121, over 28968.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3394, pruned_loss=0.09103, over 5657195.55 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.34, pruned_loss=0.09208, over 5748166.22 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3401, pruned_loss=0.09109, over 5653284.21 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:49:34,751 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 06:49:56,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 06:50:15,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=698189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:50:28,183 INFO [train.py:968] (1/2) Epoch 16, batch 14250, giga_loss[loss=0.2524, simple_loss=0.3451, pruned_loss=0.07981, over 28973.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3409, pruned_loss=0.08935, over 5662564.80 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3397, pruned_loss=0.09188, over 5751633.17 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3418, pruned_loss=0.08953, over 5653746.36 frames. ], batch size: 186, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:50:38,432 INFO [optim.py:369] (1/2) 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 06:50:49,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5331, 1.7437, 1.3505, 1.6584], device='cuda:1'), covar=tensor([0.2651, 0.2706, 0.3108, 0.2491], device='cuda:1'), in_proj_covar=tensor([0.1409, 0.1028, 0.1249, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 06:51:31,974 INFO [train.py:968] (1/2) Epoch 16, batch 14300, giga_loss[loss=0.2701, simple_loss=0.3526, pruned_loss=0.09382, over 28996.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3401, pruned_loss=0.08783, over 5650244.33 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3395, pruned_loss=0.09177, over 5754095.30 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.341, pruned_loss=0.08803, over 5639875.29 frames. ], batch size: 128, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:51:40,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7341, 1.0880, 2.8591, 2.5887], device='cuda:1'), covar=tensor([0.1796, 0.2667, 0.0530, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0699, 0.0611, 0.0892, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 06:51:41,540 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-08 06:52:36,319 INFO [train.py:968] (1/2) Epoch 16, batch 14350, libri_loss[loss=0.2763, simple_loss=0.342, pruned_loss=0.1053, over 19454.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08833, over 5653498.00 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3394, pruned_loss=0.09179, over 5746056.61 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3422, pruned_loss=0.08842, over 5652027.69 frames. ], batch size: 187, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:52:45,399 INFO [optim.py:369] (1/2) 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 06:53:15,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:53:17,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=698335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:53:29,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5556, 1.7427, 1.6046, 1.5263], device='cuda:1'), covar=tensor([0.1637, 0.2353, 0.2027, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0722, 0.0679, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 06:53:34,889 INFO [train.py:968] (1/2) Epoch 16, batch 14400, giga_loss[loss=0.2648, simple_loss=0.3232, pruned_loss=0.1032, over 24683.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3418, pruned_loss=0.08979, over 5657350.20 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3394, pruned_loss=0.09191, over 5742549.86 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.08966, over 5656242.55 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 06:53:49,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:54:41,079 INFO [train.py:968] (1/2) Epoch 16, batch 14450, giga_loss[loss=0.2649, simple_loss=0.3513, pruned_loss=0.08928, over 28414.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3407, pruned_loss=0.09033, over 5660018.24 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3389, pruned_loss=0.09157, over 5743669.43 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3418, pruned_loss=0.09046, over 5656306.12 frames. ], batch size: 336, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:54:54,715 INFO [optim.py:369] (1/2) 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 06:55:58,953 INFO [train.py:968] (1/2) Epoch 16, batch 14500, giga_loss[loss=0.2135, simple_loss=0.2999, pruned_loss=0.06361, over 28949.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3399, pruned_loss=0.09006, over 5659255.09 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3389, pruned_loss=0.09178, over 5736431.74 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3407, pruned_loss=0.08995, over 5661389.09 frames. ], batch size: 186, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:56:05,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4879, 2.0732, 1.4972, 0.8184], device='cuda:1'), covar=tensor([0.5273, 0.2573, 0.3639, 0.5249], device='cuda:1'), in_proj_covar=tensor([0.1642, 0.1555, 0.1538, 0.1351], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 06:56:29,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=698465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:56:41,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3975, 1.5409, 1.1474, 1.1058], device='cuda:1'), covar=tensor([0.0913, 0.0528, 0.1082, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0442, 0.0506, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 06:57:09,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6400, 1.8794, 1.9212, 1.4150], device='cuda:1'), covar=tensor([0.2044, 0.2568, 0.1633, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0687, 0.0908, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 06:57:24,482 INFO [train.py:968] (1/2) Epoch 16, batch 14550, giga_loss[loss=0.2478, simple_loss=0.3226, pruned_loss=0.08645, over 27654.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3343, pruned_loss=0.08684, over 5664479.24 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3384, pruned_loss=0.09147, over 5739934.67 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3354, pruned_loss=0.08697, over 5661818.16 frames. ], batch size: 472, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:57:35,948 INFO [optim.py:369] (1/2) 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 06:57:53,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 06:58:23,735 INFO [train.py:968] (1/2) Epoch 16, batch 14600, giga_loss[loss=0.2328, simple_loss=0.3196, pruned_loss=0.07294, over 28771.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3328, pruned_loss=0.08643, over 5657927.49 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3372, pruned_loss=0.0908, over 5735277.46 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3346, pruned_loss=0.08691, over 5655570.42 frames. ], batch size: 243, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:58:41,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5755, 1.6842, 1.8879, 1.4197], device='cuda:1'), covar=tensor([0.1773, 0.2341, 0.1421, 0.1748], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0687, 0.0908, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 06:59:33,379 INFO [train.py:968] (1/2) Epoch 16, batch 14650, giga_loss[loss=0.2737, simple_loss=0.3462, pruned_loss=0.1006, over 28530.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3319, pruned_loss=0.08647, over 5668966.41 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.337, pruned_loss=0.09075, over 5736519.47 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3334, pruned_loss=0.08686, over 5665463.17 frames. ], batch size: 370, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 06:59:43,736 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=698608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 06:59:48,229 INFO [optim.py:369] (1/2) 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 06:59:48,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=698611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:00:24,283 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=698640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:00:29,021 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=698643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:00:36,104 INFO [train.py:968] (1/2) Epoch 16, batch 14700, giga_loss[loss=0.2812, simple_loss=0.3603, pruned_loss=0.1011, over 28739.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3385, pruned_loss=0.09022, over 5669945.49 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3371, pruned_loss=0.09092, over 5730146.85 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3396, pruned_loss=0.09035, over 5671586.92 frames. ], batch size: 262, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:01:39,742 INFO [train.py:968] (1/2) Epoch 16, batch 14750, giga_loss[loss=0.276, simple_loss=0.349, pruned_loss=0.1015, over 28796.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3373, pruned_loss=0.09032, over 5674537.87 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3366, pruned_loss=0.09068, over 5734383.24 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3387, pruned_loss=0.09065, over 5670678.64 frames. ], batch size: 99, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:01:53,808 INFO [optim.py:369] (1/2) 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 07:01:54,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3128, 1.0835, 4.0786, 3.2569], device='cuda:1'), covar=tensor([0.1725, 0.2900, 0.0444, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0706, 0.0617, 0.0901, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 07:02:04,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-08 07:02:43,231 INFO [train.py:968] (1/2) Epoch 16, batch 14800, giga_loss[loss=0.2986, simple_loss=0.3596, pruned_loss=0.1188, over 26809.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3375, pruned_loss=0.09137, over 5669060.49 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3365, pruned_loss=0.09055, over 5736425.00 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3387, pruned_loss=0.09174, over 5662630.03 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:03:39,094 INFO [train.py:968] (1/2) Epoch 16, batch 14850, giga_loss[loss=0.2245, simple_loss=0.288, pruned_loss=0.08044, over 24361.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3373, pruned_loss=0.09117, over 5676864.97 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.337, pruned_loss=0.09084, over 5741914.25 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.338, pruned_loss=0.09127, over 5663284.52 frames. ], batch size: 705, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:03:51,746 INFO [optim.py:369] (1/2) 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 07:04:45,627 INFO [train.py:968] (1/2) Epoch 16, batch 14900, giga_loss[loss=0.2546, simple_loss=0.3446, pruned_loss=0.08229, over 28776.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3389, pruned_loss=0.09122, over 5679943.05 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.337, pruned_loss=0.09091, over 5745043.08 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3394, pruned_loss=0.09125, over 5665377.53 frames. ], batch size: 119, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:05:58,011 INFO [train.py:968] (1/2) Epoch 16, batch 14950, giga_loss[loss=0.3073, simple_loss=0.3824, pruned_loss=0.116, over 28693.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3406, pruned_loss=0.09139, over 5681872.49 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3368, pruned_loss=0.09085, over 5748676.45 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3412, pruned_loss=0.09147, over 5665110.36 frames. ], batch size: 242, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:06:13,649 INFO [optim.py:369] (1/2) 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 07:06:28,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4015, 1.4990, 1.2347, 1.5208], device='cuda:1'), covar=tensor([0.0769, 0.0336, 0.0354, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 07:07:18,046 INFO [train.py:968] (1/2) Epoch 16, batch 15000, giga_loss[loss=0.2493, simple_loss=0.3307, pruned_loss=0.08399, over 28627.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.339, pruned_loss=0.09058, over 5671200.44 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3367, pruned_loss=0.09078, over 5741063.66 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3396, pruned_loss=0.09071, over 5663850.87 frames. ], batch size: 92, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:07:18,046 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 07:07:27,434 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 07:07:41,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4496, 1.6838, 1.3362, 1.6479], device='cuda:1'), covar=tensor([0.0754, 0.0301, 0.0338, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 07:08:34,635 INFO [train.py:968] (1/2) Epoch 16, batch 15050, giga_loss[loss=0.205, simple_loss=0.2877, pruned_loss=0.06116, over 29082.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3352, pruned_loss=0.08958, over 5679798.97 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3367, pruned_loss=0.09082, over 5735813.51 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3358, pruned_loss=0.08963, over 5677323.35 frames. ], batch size: 128, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:08:51,473 INFO [optim.py:369] (1/2) 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 07:09:03,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=699018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:09:44,469 INFO [train.py:968] (1/2) Epoch 16, batch 15100, giga_loss[loss=0.2569, simple_loss=0.3217, pruned_loss=0.09607, over 26948.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3297, pruned_loss=0.08717, over 5677361.82 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3364, pruned_loss=0.09067, over 5737550.32 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3303, pruned_loss=0.0873, over 5673104.49 frames. ], batch size: 555, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:09:46,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0605, 1.2108, 3.3628, 3.0230], device='cuda:1'), covar=tensor([0.1568, 0.2529, 0.0450, 0.0940], device='cuda:1'), in_proj_covar=tensor([0.0699, 0.0612, 0.0890, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 07:10:46,709 INFO [train.py:968] (1/2) Epoch 16, batch 15150, giga_loss[loss=0.2265, simple_loss=0.3103, pruned_loss=0.07138, over 28957.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3308, pruned_loss=0.08818, over 5680190.52 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3362, pruned_loss=0.09068, over 5740501.77 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3313, pruned_loss=0.0882, over 5673044.08 frames. ], batch size: 155, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:10:58,747 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 15200, giga_loss[loss=0.2639, simple_loss=0.3407, pruned_loss=0.09355, over 28896.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3322, pruned_loss=0.08928, over 5670436.44 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3361, pruned_loss=0.09065, over 5734688.22 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3327, pruned_loss=0.0893, over 5668825.00 frames. ], batch size: 164, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:11:52,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=699161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:11:56,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=699164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:12:30,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=699193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:12:39,713 INFO [train.py:968] (1/2) Epoch 16, batch 15250, libri_loss[loss=0.2635, simple_loss=0.3337, pruned_loss=0.09662, over 29567.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3293, pruned_loss=0.08724, over 5668069.88 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3361, pruned_loss=0.09094, over 5740760.26 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3295, pruned_loss=0.0869, over 5658758.51 frames. ], batch size: 78, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:12:53,635 INFO [optim.py:369] (1/2) 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 07:13:41,429 INFO [train.py:968] (1/2) Epoch 16, batch 15300, libri_loss[loss=0.2528, simple_loss=0.332, pruned_loss=0.08678, over 27948.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3278, pruned_loss=0.08555, over 5670907.46 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3359, pruned_loss=0.09084, over 5742731.09 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.328, pruned_loss=0.08528, over 5660711.35 frames. ], batch size: 116, lr: 2.02e-03, grad_scale: 8.0
+2023-03-08 07:13:49,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 07:14:14,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4956, 1.8535, 1.6020, 1.6188], device='cuda:1'), covar=tensor([0.0751, 0.0265, 0.0319, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 07:14:54,172 INFO [train.py:968] (1/2) Epoch 16, batch 15350, giga_loss[loss=0.2315, simple_loss=0.3159, pruned_loss=0.07351, over 28658.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3262, pruned_loss=0.08515, over 5671140.05 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3356, pruned_loss=0.09076, over 5744171.71 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3265, pruned_loss=0.08492, over 5660355.43 frames. ], batch size: 307, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:15:11,744 INFO [optim.py:369] (1/2) 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 07:16:04,497 INFO [train.py:968] (1/2) Epoch 16, batch 15400, giga_loss[loss=0.2722, simple_loss=0.3495, pruned_loss=0.09748, over 28138.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3265, pruned_loss=0.08443, over 5683119.08 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3357, pruned_loss=0.09095, over 5743599.00 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3264, pruned_loss=0.08394, over 5673974.38 frames. ], batch size: 412, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:16:07,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 07:16:22,195 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.38 vs. limit=5.0
+2023-03-08 07:16:54,867 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-08 07:16:56,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9081, 3.7409, 3.5175, 1.7577], device='cuda:1'), covar=tensor([0.0660, 0.0770, 0.0812, 0.2385], device='cuda:1'), in_proj_covar=tensor([0.1122, 0.1032, 0.0888, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 07:17:06,996 INFO [train.py:968] (1/2) Epoch 16, batch 15450, libri_loss[loss=0.2261, simple_loss=0.3015, pruned_loss=0.07535, over 29584.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3268, pruned_loss=0.08477, over 5693957.94 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3351, pruned_loss=0.09066, over 5747881.15 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3271, pruned_loss=0.0845, over 5681780.58 frames. ], batch size: 75, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:17:21,358 INFO [optim.py:369] (1/2) 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 07:17:44,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4929, 2.0815, 1.5336, 0.6293], device='cuda:1'), covar=tensor([0.4969, 0.2430, 0.3613, 0.5405], device='cuda:1'), in_proj_covar=tensor([0.1649, 0.1565, 0.1544, 0.1360], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 07:18:07,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5538, 1.7985, 1.8057, 1.3670], device='cuda:1'), covar=tensor([0.1810, 0.2409, 0.1476, 0.1776], device='cuda:1'), in_proj_covar=tensor([0.0859, 0.0681, 0.0904, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 07:18:13,033 INFO [train.py:968] (1/2) Epoch 16, batch 15500, giga_loss[loss=0.2642, simple_loss=0.344, pruned_loss=0.09219, over 28385.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3273, pruned_loss=0.0856, over 5697355.26 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3348, pruned_loss=0.09054, over 5752875.14 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08534, over 5681493.75 frames. ], batch size: 368, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:18:45,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2710, 1.6271, 1.5715, 1.3939], device='cuda:1'), covar=tensor([0.1696, 0.1708, 0.1895, 0.1772], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0719, 0.0676, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 07:19:10,636 INFO [train.py:968] (1/2) Epoch 16, batch 15550, giga_loss[loss=0.28, simple_loss=0.3633, pruned_loss=0.09832, over 28982.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3256, pruned_loss=0.08406, over 5679059.03 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3347, pruned_loss=0.09054, over 5744754.85 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3257, pruned_loss=0.08371, over 5672087.27 frames. ], batch size: 213, lr: 2.02e-03, grad_scale: 4.0
+2023-03-08 07:19:15,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 07:19:18,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0039, 2.2205, 1.5541, 1.8627], device='cuda:1'), covar=tensor([0.0887, 0.0642, 0.0992, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0436, 0.0503, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 07:19:22,987 INFO [optim.py:369] (1/2) 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 07:20:09,665 INFO [train.py:968] (1/2) Epoch 16, batch 15600, giga_loss[loss=0.2433, simple_loss=0.3385, pruned_loss=0.07406, over 28956.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3271, pruned_loss=0.08381, over 5665437.89 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3345, pruned_loss=0.09043, over 5746110.67 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3272, pruned_loss=0.08344, over 5656862.07 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:21:14,422 INFO [train.py:968] (1/2) Epoch 16, batch 15650, giga_loss[loss=0.2714, simple_loss=0.3596, pruned_loss=0.09163, over 28667.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3313, pruned_loss=0.0856, over 5665559.88 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3346, pruned_loss=0.09049, over 5746237.33 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3312, pruned_loss=0.08519, over 5657544.29 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:21:29,221 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 15700, libri_loss[loss=0.2823, simple_loss=0.3594, pruned_loss=0.1025, over 28601.00 frames. ], tot_loss[loss=0.253, simple_loss=0.333, pruned_loss=0.08649, over 5657217.44 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3345, pruned_loss=0.09038, over 5740048.97 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3329, pruned_loss=0.08613, over 5653742.71 frames. ], batch size: 106, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:23:16,008 INFO [train.py:968] (1/2) Epoch 16, batch 15750, giga_loss[loss=0.232, simple_loss=0.3222, pruned_loss=0.07088, over 29000.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3328, pruned_loss=0.08667, over 5651703.78 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3347, pruned_loss=0.09045, over 5742442.20 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3325, pruned_loss=0.08626, over 5645456.71 frames. ], batch size: 285, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:23:30,692 INFO [optim.py:369] (1/2) 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 07:23:59,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9045, 1.1643, 1.1900, 0.8768], device='cuda:1'), covar=tensor([0.2445, 0.2190, 0.1262, 0.2032], device='cuda:1'), in_proj_covar=tensor([0.1816, 0.1730, 0.1658, 0.1795], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 07:24:19,180 INFO [train.py:968] (1/2) Epoch 16, batch 15800, giga_loss[loss=0.2482, simple_loss=0.3361, pruned_loss=0.08015, over 28844.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3302, pruned_loss=0.08519, over 5652897.07 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3341, pruned_loss=0.09007, over 5744160.47 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3305, pruned_loss=0.08512, over 5644822.18 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:24:53,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2686, 3.0392, 1.4176, 1.3770], device='cuda:1'), covar=tensor([0.1009, 0.0321, 0.0965, 0.1425], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0529, 0.0362, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 07:25:16,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3939, 1.2173, 1.1445, 1.5067], device='cuda:1'), covar=tensor([0.0741, 0.0328, 0.0349, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 07:25:23,542 INFO [train.py:968] (1/2) Epoch 16, batch 15850, giga_loss[loss=0.2126, simple_loss=0.2952, pruned_loss=0.06504, over 28906.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3295, pruned_loss=0.0851, over 5663776.81 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3338, pruned_loss=0.08991, over 5746544.46 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.33, pruned_loss=0.08514, over 5653978.10 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:25:41,830 INFO [optim.py:369] (1/2) 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 07:26:24,813 INFO [train.py:968] (1/2) Epoch 16, batch 15900, giga_loss[loss=0.2404, simple_loss=0.3159, pruned_loss=0.08247, over 29061.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3283, pruned_loss=0.08508, over 5674439.31 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3336, pruned_loss=0.08984, over 5749528.26 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3286, pruned_loss=0.08502, over 5661151.73 frames. ], batch size: 128, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:27:20,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7646, 5.5403, 5.2714, 2.8245], device='cuda:1'), covar=tensor([0.0422, 0.0650, 0.0744, 0.1613], device='cuda:1'), in_proj_covar=tensor([0.1121, 0.1026, 0.0886, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 07:27:28,518 INFO [train.py:968] (1/2) Epoch 16, batch 15950, libri_loss[loss=0.256, simple_loss=0.339, pruned_loss=0.08651, over 29701.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3303, pruned_loss=0.08585, over 5679173.12 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.333, pruned_loss=0.08954, over 5752571.27 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3311, pruned_loss=0.08599, over 5663900.96 frames. ], batch size: 91, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:27:43,765 INFO [optim.py:369] (1/2) 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 07:28:22,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2993, 1.9218, 1.6628, 1.5520], device='cuda:1'), covar=tensor([0.2128, 0.1800, 0.2170, 0.2025], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0723, 0.0680, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 07:28:30,826 INFO [train.py:968] (1/2) Epoch 16, batch 16000, giga_loss[loss=0.2543, simple_loss=0.3381, pruned_loss=0.08529, over 28483.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3318, pruned_loss=0.08692, over 5671746.38 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3331, pruned_loss=0.0895, over 5754397.19 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3323, pruned_loss=0.08697, over 5655559.03 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:29:04,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4393, 2.0265, 1.3921, 0.7647], device='cuda:1'), covar=tensor([0.4705, 0.2341, 0.3615, 0.4900], device='cuda:1'), in_proj_covar=tensor([0.1643, 0.1558, 0.1536, 0.1355], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 07:29:22,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-08 07:29:38,226 INFO [train.py:968] (1/2) Epoch 16, batch 16050, giga_loss[loss=0.258, simple_loss=0.3342, pruned_loss=0.09091, over 27760.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3337, pruned_loss=0.08855, over 5671122.40 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3331, pruned_loss=0.08943, over 5755540.44 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3341, pruned_loss=0.08864, over 5656644.56 frames. ], batch size: 474, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:29:51,780 INFO [optim.py:369] (1/2) 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 07:30:28,372 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 16100, libri_loss[loss=0.2361, simple_loss=0.3114, pruned_loss=0.08038, over 29564.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3363, pruned_loss=0.08962, over 5669600.83 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3324, pruned_loss=0.08905, over 5761675.87 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3373, pruned_loss=0.09005, over 5648902.76 frames. ], batch size: 74, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:31:20,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6292, 1.6247, 1.2397, 1.2363], device='cuda:1'), covar=tensor([0.0701, 0.0443, 0.0861, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0437, 0.0504, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 07:31:24,929 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:31:29,727 INFO [train.py:968] (1/2) Epoch 16, batch 16150, giga_loss[loss=0.26, simple_loss=0.3482, pruned_loss=0.0859, over 28762.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3379, pruned_loss=0.08985, over 5661496.31 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.332, pruned_loss=0.08879, over 5755685.84 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3392, pruned_loss=0.09043, over 5647622.87 frames. ], batch size: 307, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:31:46,825 INFO [optim.py:369] (1/2) 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 07:32:39,727 INFO [train.py:968] (1/2) Epoch 16, batch 16200, giga_loss[loss=0.2645, simple_loss=0.3369, pruned_loss=0.09608, over 27791.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.338, pruned_loss=0.09021, over 5656868.88 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.332, pruned_loss=0.08873, over 5759381.33 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3391, pruned_loss=0.09077, over 5640403.37 frames. ], batch size: 476, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:33:42,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6401, 1.7174, 1.2895, 1.3311], device='cuda:1'), covar=tensor([0.0908, 0.0587, 0.1068, 0.1045], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0434, 0.0502, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 07:33:46,435 INFO [train.py:968] (1/2) Epoch 16, batch 16250, giga_loss[loss=0.2846, simple_loss=0.3528, pruned_loss=0.1082, over 28562.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3354, pruned_loss=0.08883, over 5667153.58 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.332, pruned_loss=0.08874, over 5763414.99 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3365, pruned_loss=0.08927, over 5648083.93 frames. ], batch size: 242, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:34:03,341 INFO [optim.py:369] (1/2) 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 07:34:35,015 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700239.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:34:51,982 INFO [train.py:968] (1/2) Epoch 16, batch 16300, giga_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.09752, over 28934.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3346, pruned_loss=0.0882, over 5668421.20 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3318, pruned_loss=0.08868, over 5756607.35 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3356, pruned_loss=0.08862, over 5656849.55 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:34:53,097 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:35:18,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3866, 1.7510, 1.3768, 1.5955], device='cuda:1'), covar=tensor([0.0781, 0.0280, 0.0332, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 07:35:57,375 INFO [train.py:968] (1/2) Epoch 16, batch 16350, giga_loss[loss=0.2558, simple_loss=0.3392, pruned_loss=0.08619, over 28390.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3338, pruned_loss=0.08854, over 5669780.20 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3313, pruned_loss=0.08838, over 5758752.00 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3351, pruned_loss=0.08913, over 5657706.55 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:36:14,789 INFO [optim.py:369] (1/2) 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 07:36:18,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-08 07:36:36,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-08 07:37:02,256 INFO [train.py:968] (1/2) Epoch 16, batch 16400, giga_loss[loss=0.2052, simple_loss=0.2831, pruned_loss=0.06358, over 29012.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3317, pruned_loss=0.08835, over 5659590.08 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3313, pruned_loss=0.08831, over 5754466.05 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3328, pruned_loss=0.08888, over 5651912.57 frames. ], batch size: 106, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:37:53,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:38:05,565 INFO [train.py:968] (1/2) Epoch 16, batch 16450, giga_loss[loss=0.2541, simple_loss=0.3332, pruned_loss=0.08752, over 28769.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3303, pruned_loss=0.08733, over 5655607.79 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3313, pruned_loss=0.08834, over 5754048.67 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3311, pruned_loss=0.08771, over 5648263.84 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:38:22,458 INFO [optim.py:369] (1/2) 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 07:38:27,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:39:05,241 INFO [train.py:968] (1/2) Epoch 16, batch 16500, giga_loss[loss=0.2453, simple_loss=0.3298, pruned_loss=0.0804, over 28924.00 frames. ], tot_loss[loss=0.25, simple_loss=0.329, pruned_loss=0.0855, over 5677510.47 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3307, pruned_loss=0.08817, over 5759151.70 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3302, pruned_loss=0.0859, over 5664103.90 frames. ], batch size: 284, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:39:27,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:40:02,127 INFO [train.py:968] (1/2) Epoch 16, batch 16550, giga_loss[loss=0.2571, simple_loss=0.3435, pruned_loss=0.08537, over 28677.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3299, pruned_loss=0.08387, over 5682948.38 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3305, pruned_loss=0.08795, over 5763389.72 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.331, pruned_loss=0.08428, over 5666440.33 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:40:17,103 INFO [optim.py:369] (1/2) 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 07:40:59,167 INFO [train.py:968] (1/2) Epoch 16, batch 16600, giga_loss[loss=0.247, simple_loss=0.3355, pruned_loss=0.07927, over 28904.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3318, pruned_loss=0.08371, over 5693499.12 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3303, pruned_loss=0.08798, over 5767041.28 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3328, pruned_loss=0.08393, over 5675121.26 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:41:12,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:14,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:44,741 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:47,238 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:41:54,819 INFO [train.py:968] (1/2) Epoch 16, batch 16650, giga_loss[loss=0.2459, simple_loss=0.3394, pruned_loss=0.07615, over 28857.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3326, pruned_loss=0.08445, over 5677540.43 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3302, pruned_loss=0.08792, over 5761531.90 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3336, pruned_loss=0.08457, over 5665935.29 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:42:13,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700614.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 07:42:13,536 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:42:17,783 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:42:30,157 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:43:03,550 INFO [train.py:968] (1/2) Epoch 16, batch 16700, giga_loss[loss=0.3035, simple_loss=0.3644, pruned_loss=0.1213, over 26759.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3316, pruned_loss=0.0841, over 5671055.05 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3295, pruned_loss=0.08756, over 5764889.85 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3331, pruned_loss=0.08442, over 5656411.54 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:44:10,600 INFO [train.py:968] (1/2) Epoch 16, batch 16750, giga_loss[loss=0.2701, simple_loss=0.3499, pruned_loss=0.09511, over 28897.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3313, pruned_loss=0.08392, over 5663657.80 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3297, pruned_loss=0.08765, over 5764093.41 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3324, pruned_loss=0.08401, over 5650250.42 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 07:44:33,685 INFO [optim.py:369] (1/2) 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 07:44:56,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4348, 2.0292, 1.4695, 0.6234], device='cuda:1'), covar=tensor([0.4867, 0.2661, 0.4046, 0.5589], device='cuda:1'), in_proj_covar=tensor([0.1622, 0.1539, 0.1522, 0.1341], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 07:45:23,571 INFO [train.py:968] (1/2) Epoch 16, batch 16800, giga_loss[loss=0.2869, simple_loss=0.3458, pruned_loss=0.114, over 27122.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3323, pruned_loss=0.08372, over 5667522.88 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3295, pruned_loss=0.08747, over 5764969.89 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3334, pruned_loss=0.08388, over 5654386.86 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 07:45:32,077 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700757.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:45:36,410 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700760.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 07:45:40,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-08 07:45:45,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:45:50,502 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:45:53,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:45:55,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6526, 1.7316, 1.6723, 1.5488], device='cuda:1'), covar=tensor([0.2339, 0.2098, 0.1798, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1710, 0.1635, 0.1779], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 07:46:20,845 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700789.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 07:46:37,264 INFO [train.py:968] (1/2) Epoch 16, batch 16850, giga_loss[loss=0.2899, simple_loss=0.3573, pruned_loss=0.1112, over 28157.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.334, pruned_loss=0.08488, over 5663191.97 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3295, pruned_loss=0.08746, over 5767420.55 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3349, pruned_loss=0.08496, over 5648823.23 frames. ], batch size: 412, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:46:39,745 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:47:05,229 INFO [optim.py:369] (1/2) 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 07:47:52,017 INFO [train.py:968] (1/2) Epoch 16, batch 16900, giga_loss[loss=0.2406, simple_loss=0.33, pruned_loss=0.07558, over 28808.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3374, pruned_loss=0.08631, over 5667397.58 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3292, pruned_loss=0.08727, over 5769988.95 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3385, pruned_loss=0.08649, over 5652409.41 frames. ], batch size: 243, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:48:58,704 INFO [train.py:968] (1/2) Epoch 16, batch 16950, giga_loss[loss=0.2692, simple_loss=0.3392, pruned_loss=0.09966, over 27056.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.338, pruned_loss=0.0871, over 5681073.26 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3296, pruned_loss=0.08751, over 5771109.87 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3389, pruned_loss=0.08703, over 5664163.44 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:49:08,084 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=700908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:49:10,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1836, 1.4710, 1.4513, 1.0496], device='cuda:1'), covar=tensor([0.1569, 0.2455, 0.1369, 0.1531], device='cuda:1'), in_proj_covar=tensor([0.0860, 0.0679, 0.0904, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 07:49:11,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=700911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:49:23,875 INFO [optim.py:369] (1/2) 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 07:49:30,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 07:49:32,654 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:49:56,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=700940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:50:15,506 INFO [train.py:968] (1/2) Epoch 16, batch 17000, giga_loss[loss=0.2115, simple_loss=0.2947, pruned_loss=0.06418, over 28962.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3373, pruned_loss=0.08762, over 5676615.11 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3297, pruned_loss=0.08758, over 5771510.32 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3378, pruned_loss=0.0875, over 5662666.05 frames. ], batch size: 93, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:50:27,153 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=700957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:50:42,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=700968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:51:24,134 INFO [train.py:968] (1/2) Epoch 16, batch 17050, giga_loss[loss=0.2411, simple_loss=0.3113, pruned_loss=0.0855, over 24579.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3349, pruned_loss=0.08575, over 5682596.07 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3299, pruned_loss=0.08757, over 5772116.98 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3354, pruned_loss=0.08564, over 5667889.05 frames. ], batch size: 705, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:51:51,349 INFO [optim.py:369] (1/2) 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 07:52:29,070 INFO [train.py:968] (1/2) Epoch 16, batch 17100, giga_loss[loss=0.2827, simple_loss=0.3551, pruned_loss=0.1051, over 28928.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3333, pruned_loss=0.08462, over 5681099.45 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3297, pruned_loss=0.08741, over 5775030.61 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3341, pruned_loss=0.08458, over 5663265.39 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:53:33,194 INFO [train.py:968] (1/2) Epoch 16, batch 17150, giga_loss[loss=0.2531, simple_loss=0.3353, pruned_loss=0.08547, over 28971.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3341, pruned_loss=0.08505, over 5683096.99 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3294, pruned_loss=0.08725, over 5776036.52 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.335, pruned_loss=0.08514, over 5667216.34 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 1.0
+2023-03-08 07:53:45,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:53:49,091 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:54:21,577 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 17200, giga_loss[loss=0.2538, simple_loss=0.3435, pruned_loss=0.08206, over 28094.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3364, pruned_loss=0.08622, over 5678086.26 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3299, pruned_loss=0.08751, over 5773294.17 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3369, pruned_loss=0.08603, over 5664402.47 frames. ], batch size: 412, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:55:23,949 INFO [train.py:968] (1/2) Epoch 16, batch 17250, giga_loss[loss=0.243, simple_loss=0.3245, pruned_loss=0.08078, over 29057.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3354, pruned_loss=0.08636, over 5685515.71 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3296, pruned_loss=0.08731, over 5775476.29 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3363, pruned_loss=0.08637, over 5668882.09 frames. ], batch size: 120, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:55:42,366 INFO [optim.py:369] (1/2) 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 07:55:45,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701221.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 07:56:19,403 INFO [train.py:968] (1/2) Epoch 16, batch 17300, giga_loss[loss=0.2367, simple_loss=0.3208, pruned_loss=0.07633, over 28266.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3321, pruned_loss=0.08599, over 5672096.10 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3292, pruned_loss=0.08713, over 5768980.05 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3334, pruned_loss=0.08614, over 5661444.12 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:57:18,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:57:21,007 INFO [train.py:968] (1/2) Epoch 16, batch 17350, giga_loss[loss=0.2761, simple_loss=0.3415, pruned_loss=0.1054, over 26946.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3324, pruned_loss=0.08707, over 5663358.25 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3292, pruned_loss=0.08712, over 5771551.99 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3335, pruned_loss=0.08718, over 5651054.85 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:57:41,739 INFO [optim.py:369] (1/2) 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 07:57:57,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:58:18,877 INFO [train.py:968] (1/2) Epoch 16, batch 17400, giga_loss[loss=0.3188, simple_loss=0.3937, pruned_loss=0.1219, over 28669.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3389, pruned_loss=0.09113, over 5663704.13 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3292, pruned_loss=0.08716, over 5773452.39 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3398, pruned_loss=0.09122, over 5650483.21 frames. ], batch size: 242, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:58:31,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2369, 1.5018, 1.5053, 1.2454], device='cuda:1'), covar=tensor([0.1654, 0.1566, 0.2087, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0716, 0.0674, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 07:59:06,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2443, 4.0477, 3.8456, 1.9280], device='cuda:1'), covar=tensor([0.0476, 0.0650, 0.0646, 0.2351], device='cuda:1'), in_proj_covar=tensor([0.1109, 0.1017, 0.0877, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-08 07:59:07,103 INFO [train.py:968] (1/2) Epoch 16, batch 17450, giga_loss[loss=0.3336, simple_loss=0.3944, pruned_loss=0.1364, over 26729.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3482, pruned_loss=0.0966, over 5674039.53 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3292, pruned_loss=0.08711, over 5777285.59 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3495, pruned_loss=0.09696, over 5656701.35 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 07:59:22,003 INFO [optim.py:369] (1/2) 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 07:59:43,625 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:59:45,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=701444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 07:59:50,496 INFO [train.py:968] (1/2) Epoch 16, batch 17500, giga_loss[loss=0.2648, simple_loss=0.3472, pruned_loss=0.09116, over 29016.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.352, pruned_loss=0.09891, over 5683212.42 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.329, pruned_loss=0.08709, over 5777939.58 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3538, pruned_loss=0.0995, over 5666154.17 frames. ], batch size: 128, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:00:14,633 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=701473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:15,991 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:17,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=701478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:40,336 INFO [train.py:968] (1/2) Epoch 16, batch 17550, giga_loss[loss=0.2504, simple_loss=0.3269, pruned_loss=0.0869, over 28866.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.349, pruned_loss=0.09816, over 5681221.39 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3289, pruned_loss=0.08694, over 5779012.10 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3507, pruned_loss=0.09886, over 5666046.52 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:00:48,864 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=701507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:00:58,852 INFO [optim.py:369] (1/2) 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 08:01:20,829 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 17600, giga_loss[loss=0.2404, simple_loss=0.3134, pruned_loss=0.0837, over 28920.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3419, pruned_loss=0.09515, over 5687422.20 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.08689, over 5780993.65 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3436, pruned_loss=0.09588, over 5672150.46 frames. ], batch size: 112, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:01:42,360 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:01:46,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4462, 2.2024, 1.6637, 0.6616], device='cuda:1'), covar=tensor([0.4802, 0.2193, 0.2732, 0.4503], device='cuda:1'), in_proj_covar=tensor([0.1640, 0.1563, 0.1534, 0.1349], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 08:02:01,614 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-08 08:02:09,106 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 17650, giga_loss[loss=0.2387, simple_loss=0.3067, pruned_loss=0.08533, over 28918.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3348, pruned_loss=0.09195, over 5687945.42 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3292, pruned_loss=0.08709, over 5774255.80 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3359, pruned_loss=0.09246, over 5679540.86 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:02:14,019 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 08:02:28,616 INFO [optim.py:369] (1/2) 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 08:02:32,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-08 08:02:41,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:02:57,744 INFO [train.py:968] (1/2) Epoch 16, batch 17700, giga_loss[loss=0.2015, simple_loss=0.2756, pruned_loss=0.06369, over 28555.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3271, pruned_loss=0.08858, over 5690896.24 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3294, pruned_loss=0.0872, over 5776681.41 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3278, pruned_loss=0.08899, over 5679411.55 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:03:38,153 INFO [train.py:968] (1/2) Epoch 16, batch 17750, giga_loss[loss=0.2128, simple_loss=0.2862, pruned_loss=0.06973, over 29082.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3205, pruned_loss=0.08549, over 5687333.30 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3299, pruned_loss=0.0874, over 5771065.84 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3204, pruned_loss=0.08558, over 5681113.66 frames. ], batch size: 128, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:03:52,365 INFO [optim.py:369] (1/2) 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 08:04:08,615 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=701739.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:04:10,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=701742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:04:19,078 INFO [train.py:968] (1/2) Epoch 16, batch 17800, giga_loss[loss=0.2529, simple_loss=0.3183, pruned_loss=0.09373, over 28879.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3161, pruned_loss=0.0835, over 5690543.67 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3301, pruned_loss=0.08725, over 5773947.51 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3156, pruned_loss=0.08364, over 5681257.84 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:04:35,755 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=701771.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:04:47,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:04:49,766 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701788.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:05:00,200 INFO [train.py:968] (1/2) Epoch 16, batch 17850, giga_loss[loss=0.2279, simple_loss=0.3039, pruned_loss=0.07598, over 28717.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3128, pruned_loss=0.0818, over 5692712.07 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3302, pruned_loss=0.08732, over 5766526.67 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.312, pruned_loss=0.08175, over 5690634.49 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:05:17,762 INFO [optim.py:369] (1/2) 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 08:05:31,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8881, 3.7074, 3.4447, 1.7101], device='cuda:1'), covar=tensor([0.0652, 0.0819, 0.0799, 0.2125], device='cuda:1'), in_proj_covar=tensor([0.1118, 0.1030, 0.0886, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0011, 0.0011], device='cuda:1')
+2023-03-08 08:05:44,215 INFO [train.py:968] (1/2) Epoch 16, batch 17900, giga_loss[loss=0.2186, simple_loss=0.297, pruned_loss=0.07009, over 28864.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3115, pruned_loss=0.0812, over 5688291.08 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3314, pruned_loss=0.08779, over 5760992.08 frames. ], giga_tot_loss[loss=0.2351, simple_loss=0.3091, pruned_loss=0.08054, over 5689364.61 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:05:54,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3951, 1.2862, 3.9853, 3.3280], device='cuda:1'), covar=tensor([0.1621, 0.2775, 0.0414, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0616, 0.0902, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 08:06:25,666 INFO [train.py:968] (1/2) Epoch 16, batch 17950, giga_loss[loss=0.2254, simple_loss=0.2984, pruned_loss=0.07619, over 28860.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3082, pruned_loss=0.07983, over 5688658.53 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3311, pruned_loss=0.08756, over 5764116.49 frames. ], giga_tot_loss[loss=0.2324, simple_loss=0.3061, pruned_loss=0.07938, over 5684995.65 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:06:40,711 INFO [optim.py:369] (1/2) 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 08:06:41,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=701919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:06:44,286 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:06:44,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1072, 1.7321, 5.2684, 3.7115], device='cuda:1'), covar=tensor([0.1478, 0.2551, 0.0334, 0.0787], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0617, 0.0904, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 08:07:02,968 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 18000, giga_loss[loss=0.2295, simple_loss=0.3069, pruned_loss=0.076, over 28718.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3049, pruned_loss=0.07835, over 5693089.88 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3313, pruned_loss=0.08753, over 5764496.98 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3029, pruned_loss=0.07792, over 5688967.05 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:07:11,777 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 08:07:21,359 INFO [train.py:1012] (1/2) Epoch 16, validation: loss=0.2096, simple_loss=0.3154, pruned_loss=0.0519, over 944034.00 frames. 
+2023-03-08 08:07:21,360 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 08:07:40,088 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:07:43,468 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=701977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:07:47,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-08 08:08:04,434 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-08 08:08:05,645 INFO [train.py:968] (1/2) Epoch 16, batch 18050, giga_loss[loss=0.2247, simple_loss=0.2921, pruned_loss=0.07859, over 28746.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3025, pruned_loss=0.07719, over 5691992.07 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3314, pruned_loss=0.08747, over 5767812.44 frames. ], giga_tot_loss[loss=0.2267, simple_loss=0.3001, pruned_loss=0.07668, over 5684166.00 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:08:14,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:08:20,324 INFO [optim.py:369] (1/2) 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 08:08:45,890 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=702050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:08:46,396 INFO [train.py:968] (1/2) Epoch 16, batch 18100, giga_loss[loss=0.2215, simple_loss=0.296, pruned_loss=0.0735, over 28834.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.2994, pruned_loss=0.07548, over 5696368.74 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3313, pruned_loss=0.08732, over 5768403.98 frames. ], giga_tot_loss[loss=0.2233, simple_loss=0.2968, pruned_loss=0.07493, over 5687510.58 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:08:58,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:08:59,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:18,206 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:22,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:31,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:36,470 INFO [train.py:968] (1/2) Epoch 16, batch 18150, giga_loss[loss=0.191, simple_loss=0.2635, pruned_loss=0.05921, over 28545.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2968, pruned_loss=0.07455, over 5702872.15 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3313, pruned_loss=0.08726, over 5769816.28 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.2945, pruned_loss=0.07406, over 5694047.24 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:09:48,635 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:09:52,095 INFO [optim.py:369] (1/2) 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 08:10:04,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 08:10:19,880 INFO [train.py:968] (1/2) Epoch 16, batch 18200, giga_loss[loss=0.1858, simple_loss=0.2651, pruned_loss=0.05321, over 28810.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2963, pruned_loss=0.07497, over 5701625.40 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3317, pruned_loss=0.08745, over 5770307.16 frames. ], giga_tot_loss[loss=0.2209, simple_loss=0.2935, pruned_loss=0.07419, over 5692879.94 frames. ], batch size: 285, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:10:21,763 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:10:23,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:10:27,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:10:30,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702163.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:10:54,483 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 18250, libri_loss[loss=0.2194, simple_loss=0.3116, pruned_loss=0.06357, over 29548.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3069, pruned_loss=0.08072, over 5698454.64 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3316, pruned_loss=0.08731, over 5771868.49 frames. ], giga_tot_loss[loss=0.2322, simple_loss=0.3043, pruned_loss=0.08007, over 5688838.98 frames. ], batch size: 79, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:11:27,707 INFO [optim.py:369] (1/2) 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 08:11:57,461 INFO [train.py:968] (1/2) Epoch 16, batch 18300, giga_loss[loss=0.3271, simple_loss=0.3907, pruned_loss=0.1317, over 28911.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3208, pruned_loss=0.08805, over 5694273.61 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3318, pruned_loss=0.0876, over 5770580.62 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.318, pruned_loss=0.08717, over 5686162.72 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:12:17,378 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=702273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:12:37,079 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 18350, giga_loss[loss=0.3104, simple_loss=0.3841, pruned_loss=0.1183, over 28545.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3318, pruned_loss=0.09339, over 5688159.47 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3322, pruned_loss=0.08789, over 5753898.70 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3291, pruned_loss=0.0925, over 5695933.88 frames. ], batch size: 60, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:12:40,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:12:43,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:12:43,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702306.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:12:45,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702309.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:12:52,065 INFO [optim.py:369] (1/2) 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 08:12:54,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4233, 1.7178, 1.3581, 1.3768], device='cuda:1'), covar=tensor([0.2553, 0.2560, 0.2951, 0.2204], device='cuda:1'), in_proj_covar=tensor([0.1400, 0.1022, 0.1246, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 08:12:59,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-08 08:13:05,653 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:13:09,018 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702338.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:13:15,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:13:17,821 INFO [train.py:968] (1/2) Epoch 16, batch 18400, giga_loss[loss=0.2819, simple_loss=0.3637, pruned_loss=0.1, over 28871.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3395, pruned_loss=0.09666, over 5678203.46 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3327, pruned_loss=0.08819, over 5747463.62 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.337, pruned_loss=0.09583, over 5688260.99 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:13:19,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:13:22,849 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-08 08:13:33,167 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=702368.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:13:46,340 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 08:13:59,242 INFO [train.py:968] (1/2) Epoch 16, batch 18450, giga_loss[loss=0.2561, simple_loss=0.3461, pruned_loss=0.08302, over 28788.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3437, pruned_loss=0.09773, over 5685812.45 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3337, pruned_loss=0.08886, over 5752279.59 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.341, pruned_loss=0.09677, over 5687071.76 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:14:16,891 INFO [optim.py:369] (1/2) 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 08:14:21,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=702425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:14:33,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6359, 1.7875, 1.5140, 1.6309], device='cuda:1'), covar=tensor([0.2349, 0.2267, 0.2271, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.1398, 0.1020, 0.1244, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 08:14:38,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:14:40,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:14:43,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6486, 1.7482, 1.2610, 1.3663], device='cuda:1'), covar=tensor([0.0893, 0.0666, 0.1040, 0.1252], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0439, 0.0508, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 08:14:45,278 INFO [train.py:968] (1/2) Epoch 16, batch 18500, giga_loss[loss=0.2819, simple_loss=0.3559, pruned_loss=0.1039, over 28871.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3449, pruned_loss=0.09759, over 5683124.66 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3338, pruned_loss=0.08888, over 5753812.93 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3428, pruned_loss=0.09689, over 5682104.03 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:15:03,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-08 08:15:09,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:23,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:25,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:26,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:28,311 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:29,980 INFO [train.py:968] (1/2) Epoch 16, batch 18550, giga_loss[loss=0.2636, simple_loss=0.336, pruned_loss=0.09562, over 28811.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3474, pruned_loss=0.09926, over 5687019.33 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3344, pruned_loss=0.08928, over 5750689.48 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3457, pruned_loss=0.09874, over 5686450.84 frames. ], batch size: 112, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:15:50,000 INFO [optim.py:369] (1/2) 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 08:15:50,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:15:53,851 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:16:16,482 INFO [train.py:968] (1/2) Epoch 16, batch 18600, giga_loss[loss=0.3072, simple_loss=0.3749, pruned_loss=0.1197, over 28798.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3503, pruned_loss=0.1012, over 5683953.65 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3347, pruned_loss=0.08934, over 5744359.84 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3489, pruned_loss=0.1009, over 5687442.60 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:16:34,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:16:36,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:17:01,632 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 18650, giga_loss[loss=0.2922, simple_loss=0.3685, pruned_loss=0.1079, over 29015.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 5692674.83 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3345, pruned_loss=0.0892, over 5747012.25 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3519, pruned_loss=0.1026, over 5692158.08 frames. ], batch size: 106, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:17:20,625 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 18700, giga_loss[loss=0.2853, simple_loss=0.3644, pruned_loss=0.103, over 28937.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3563, pruned_loss=0.1038, over 5699097.78 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3349, pruned_loss=0.08944, over 5748600.17 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3557, pruned_loss=0.1038, over 5696886.50 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:18:26,658 INFO [train.py:968] (1/2) Epoch 16, batch 18750, giga_loss[loss=0.3003, simple_loss=0.3693, pruned_loss=0.1156, over 29117.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3579, pruned_loss=0.1039, over 5704366.72 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3356, pruned_loss=0.08971, over 5749127.28 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3572, pruned_loss=0.1039, over 5701186.76 frames. ], batch size: 128, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 08:18:44,473 INFO [optim.py:369] (1/2) 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 08:19:01,438 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 18800, giga_loss[loss=0.2483, simple_loss=0.3339, pruned_loss=0.08135, over 28495.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3592, pruned_loss=0.104, over 5685353.55 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3364, pruned_loss=0.0902, over 5734043.17 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3585, pruned_loss=0.104, over 5695478.17 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:19:43,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:19:45,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:19:49,667 INFO [train.py:968] (1/2) Epoch 16, batch 18850, giga_loss[loss=0.2647, simple_loss=0.3453, pruned_loss=0.09206, over 28787.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3585, pruned_loss=0.1026, over 5683217.71 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3369, pruned_loss=0.09038, over 5727085.76 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3581, pruned_loss=0.1028, over 5695527.27 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:20:07,728 INFO [optim.py:369] (1/2) 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 08:20:08,659 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:20:31,076 INFO [train.py:968] (1/2) Epoch 16, batch 18900, giga_loss[loss=0.2802, simple_loss=0.362, pruned_loss=0.09921, over 29072.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3573, pruned_loss=0.101, over 5676798.39 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3377, pruned_loss=0.09081, over 5712550.67 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3571, pruned_loss=0.1013, over 5697358.94 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:21:00,413 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=702886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:21:02,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=702889.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:21:12,186 INFO [train.py:968] (1/2) Epoch 16, batch 18950, giga_loss[loss=0.2435, simple_loss=0.334, pruned_loss=0.07648, over 28592.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3558, pruned_loss=0.09974, over 5677735.42 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.338, pruned_loss=0.09086, over 5706093.31 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3557, pruned_loss=0.1001, over 5699501.39 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:21:25,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=702918.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:21:28,934 INFO [optim.py:369] (1/2) 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 08:21:35,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6681, 1.9673, 1.8389, 1.4909], device='cuda:1'), covar=tensor([0.2478, 0.1722, 0.1447, 0.2094], device='cuda:1'), in_proj_covar=tensor([0.1849, 0.1757, 0.1676, 0.1833], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 08:21:48,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5137, 1.6066, 1.6934, 1.5176], device='cuda:1'), covar=tensor([0.1583, 0.1828, 0.1955, 0.1742], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0727, 0.0684, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 08:21:55,538 INFO [train.py:968] (1/2) Epoch 16, batch 19000, giga_loss[loss=0.3018, simple_loss=0.3664, pruned_loss=0.1186, over 28915.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3575, pruned_loss=0.1021, over 5666516.22 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3384, pruned_loss=0.09088, over 5699945.01 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3575, pruned_loss=0.1026, over 5689515.77 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:22:35,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-08 08:22:39,892 INFO [train.py:968] (1/2) Epoch 16, batch 19050, giga_loss[loss=0.2804, simple_loss=0.3558, pruned_loss=0.1024, over 28925.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.36, pruned_loss=0.1063, over 5662715.74 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3387, pruned_loss=0.09085, over 5702009.23 frames. ], giga_tot_loss[loss=0.2873, simple_loss=0.3604, pruned_loss=0.1071, over 5678330.20 frames. ], batch size: 145, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:22:55,321 INFO [optim.py:369] (1/2) 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 08:23:08,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 08:23:18,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 08:23:19,062 INFO [train.py:968] (1/2) Epoch 16, batch 19100, giga_loss[loss=0.2711, simple_loss=0.347, pruned_loss=0.09761, over 28947.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3605, pruned_loss=0.1077, over 5677050.05 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3391, pruned_loss=0.09104, over 5705963.51 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3608, pruned_loss=0.1084, over 5685289.78 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:23:23,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5376, 3.1867, 1.6132, 1.5936], device='cuda:1'), covar=tensor([0.0934, 0.0314, 0.0818, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0528, 0.0362, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 08:23:36,388 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:23:58,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5873, 1.6402, 1.8165, 1.3974], device='cuda:1'), covar=tensor([0.1700, 0.2334, 0.1342, 0.1573], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0689, 0.0912, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 08:23:59,559 INFO [train.py:968] (1/2) Epoch 16, batch 19150, giga_loss[loss=0.2887, simple_loss=0.353, pruned_loss=0.1122, over 28923.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3576, pruned_loss=0.1069, over 5684584.13 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3394, pruned_loss=0.09118, over 5708219.24 frames. ], giga_tot_loss[loss=0.2867, simple_loss=0.358, pruned_loss=0.1077, over 5688545.39 frames. ], batch size: 106, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:24:18,873 INFO [optim.py:369] (1/2) 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 08:24:34,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8938, 1.1045, 1.0712, 0.8218], device='cuda:1'), covar=tensor([0.2176, 0.2465, 0.1438, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1758, 0.1682, 0.1834], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 08:24:35,403 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:24:41,410 INFO [train.py:968] (1/2) Epoch 16, batch 19200, giga_loss[loss=0.2615, simple_loss=0.3403, pruned_loss=0.09134, over 29051.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3559, pruned_loss=0.106, over 5692422.49 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3395, pruned_loss=0.09114, over 5712057.05 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3566, pruned_loss=0.1071, over 5691652.85 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:25:14,862 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 19250, giga_loss[loss=0.2596, simple_loss=0.3314, pruned_loss=0.09389, over 28621.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3551, pruned_loss=0.1054, over 5686386.80 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3392, pruned_loss=0.09093, over 5715871.41 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3564, pruned_loss=0.1069, over 5681525.13 frames. ], batch size: 78, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:25:42,149 INFO [optim.py:369] (1/2) 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 08:25:48,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:25:59,419 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:26:01,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703243.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:26:06,876 INFO [train.py:968] (1/2) Epoch 16, batch 19300, giga_loss[loss=0.242, simple_loss=0.3212, pruned_loss=0.0814, over 28875.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3533, pruned_loss=0.1032, over 5696355.98 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3402, pruned_loss=0.09133, over 5721516.54 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.354, pruned_loss=0.1045, over 5686362.21 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:26:54,508 INFO [train.py:968] (1/2) Epoch 16, batch 19350, libri_loss[loss=0.2314, simple_loss=0.3205, pruned_loss=0.07111, over 29514.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3476, pruned_loss=0.0995, over 5690266.05 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3399, pruned_loss=0.09112, over 5724368.41 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3485, pruned_loss=0.1009, over 5679122.66 frames. ], batch size: 80, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:27:10,663 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 19400, giga_loss[loss=0.2634, simple_loss=0.3331, pruned_loss=0.09683, over 28313.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3423, pruned_loss=0.09652, over 5690682.00 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.34, pruned_loss=0.09091, over 5729309.86 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3431, pruned_loss=0.09799, over 5676447.99 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:28:29,116 INFO [train.py:968] (1/2) Epoch 16, batch 19450, giga_loss[loss=0.2304, simple_loss=0.3056, pruned_loss=0.07759, over 28677.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3365, pruned_loss=0.09387, over 5686949.03 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.34, pruned_loss=0.09091, over 5729309.86 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3371, pruned_loss=0.09501, over 5675870.52 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:28:49,793 INFO [optim.py:369] (1/2) 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 08:28:58,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6470, 1.7192, 1.8981, 1.4449], device='cuda:1'), covar=tensor([0.1689, 0.2131, 0.1282, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0685, 0.0910, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 08:29:10,433 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 19500, giga_loss[loss=0.2582, simple_loss=0.351, pruned_loss=0.08264, over 28923.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3344, pruned_loss=0.09205, over 5695929.80 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3405, pruned_loss=0.09105, over 5730530.85 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3343, pruned_loss=0.09293, over 5684753.65 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:29:59,762 INFO [train.py:968] (1/2) Epoch 16, batch 19550, giga_loss[loss=0.2703, simple_loss=0.34, pruned_loss=0.1003, over 29001.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3355, pruned_loss=0.09197, over 5698374.70 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3408, pruned_loss=0.09104, over 5730322.33 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.335, pruned_loss=0.0927, over 5689044.49 frames. ], batch size: 213, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:30:14,148 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:30:46,150 INFO [train.py:968] (1/2) Epoch 16, batch 19600, giga_loss[loss=0.3579, simple_loss=0.398, pruned_loss=0.1589, over 26595.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3352, pruned_loss=0.09206, over 5700017.97 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3411, pruned_loss=0.09118, over 5731190.72 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3346, pruned_loss=0.09251, over 5691865.37 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:30:54,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:02,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 08:31:19,646 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:21,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:29,269 INFO [train.py:968] (1/2) Epoch 16, batch 19650, giga_loss[loss=0.2281, simple_loss=0.3023, pruned_loss=0.07695, over 28286.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3333, pruned_loss=0.09153, over 5711151.35 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3416, pruned_loss=0.09141, over 5732874.71 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3323, pruned_loss=0.0917, over 5703027.71 frames. ], batch size: 65, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:31:31,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:39,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:40,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=703618.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:31:43,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:31:45,709 INFO [optim.py:369] (1/2) 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 08:32:00,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3009, 3.1363, 2.9622, 1.4162], device='cuda:1'), covar=tensor([0.0885, 0.0999, 0.0848, 0.2445], device='cuda:1'), in_proj_covar=tensor([0.1117, 0.1026, 0.0885, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 08:32:07,751 INFO [train.py:968] (1/2) Epoch 16, batch 19700, giga_loss[loss=0.255, simple_loss=0.3247, pruned_loss=0.09269, over 28665.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.331, pruned_loss=0.0904, over 5713964.26 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3417, pruned_loss=0.09138, over 5726631.89 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3299, pruned_loss=0.09058, over 5711776.71 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:32:16,945 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:19,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:39,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:43,008 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:43,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 08:32:48,682 INFO [train.py:968] (1/2) Epoch 16, batch 19750, giga_loss[loss=0.2297, simple_loss=0.3034, pruned_loss=0.07799, over 28500.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3285, pruned_loss=0.08951, over 5716167.80 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3418, pruned_loss=0.09132, over 5728210.75 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3275, pruned_loss=0.08969, over 5712817.45 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:32:51,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:55,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:32:55,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-08 08:33:09,201 INFO [optim.py:369] (1/2) 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 08:33:19,075 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:22,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-08 08:33:29,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:31,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:31,849 INFO [train.py:968] (1/2) Epoch 16, batch 19800, libri_loss[loss=0.2382, simple_loss=0.3284, pruned_loss=0.07402, over 29578.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3274, pruned_loss=0.08882, over 5721318.60 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3425, pruned_loss=0.09127, over 5731487.31 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3255, pruned_loss=0.08894, over 5715242.67 frames. ], batch size: 77, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:33:37,532 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:38,939 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=703761.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:33:39,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=703762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:33:40,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9193, 1.8655, 1.4433, 1.6058], device='cuda:1'), covar=tensor([0.0976, 0.0756, 0.1056, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0440, 0.0507, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 08:33:40,892 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:34:03,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:34:04,375 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=703793.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:34:09,972 INFO [train.py:968] (1/2) Epoch 16, batch 19850, libri_loss[loss=0.269, simple_loss=0.3596, pruned_loss=0.0892, over 29767.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3259, pruned_loss=0.0879, over 5723157.31 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3426, pruned_loss=0.09108, over 5737115.03 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3236, pruned_loss=0.08807, over 5712822.61 frames. ], batch size: 87, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:34:28,553 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 19900, libri_loss[loss=0.2789, simple_loss=0.3734, pruned_loss=0.09216, over 29465.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3247, pruned_loss=0.08757, over 5724340.50 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3431, pruned_loss=0.09103, over 5740364.49 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3221, pruned_loss=0.08767, over 5712979.09 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:35:30,030 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 19950, giga_loss[loss=0.2369, simple_loss=0.3126, pruned_loss=0.08055, over 28753.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3228, pruned_loss=0.08696, over 5725671.64 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3436, pruned_loss=0.09133, over 5739487.92 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3202, pruned_loss=0.08675, over 5717084.78 frames. ], batch size: 284, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:35:48,638 INFO [optim.py:369] (1/2) 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 08:36:11,289 INFO [train.py:968] (1/2) Epoch 16, batch 20000, giga_loss[loss=0.2415, simple_loss=0.3129, pruned_loss=0.08501, over 28741.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3213, pruned_loss=0.08586, over 5732849.60 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3445, pruned_loss=0.09156, over 5742734.59 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.318, pruned_loss=0.08539, over 5723026.88 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:36:12,699 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=703953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:36:47,809 INFO [train.py:968] (1/2) Epoch 16, batch 20050, giga_loss[loss=0.2383, simple_loss=0.3153, pruned_loss=0.08068, over 28951.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3202, pruned_loss=0.08518, over 5727807.41 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3451, pruned_loss=0.09179, over 5733908.68 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3166, pruned_loss=0.08448, over 5728032.69 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:37:07,672 INFO [optim.py:369] (1/2) 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 08:37:22,104 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:24,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=704044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:31,226 INFO [train.py:968] (1/2) Epoch 16, batch 20100, giga_loss[loss=0.266, simple_loss=0.341, pruned_loss=0.09555, over 28804.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3231, pruned_loss=0.08703, over 5726933.49 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3453, pruned_loss=0.09183, over 5736447.68 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3195, pruned_loss=0.08633, over 5724733.15 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:37:44,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:53,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=704073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:37:56,521 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:38:11,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 08:38:20,877 INFO [train.py:968] (1/2) Epoch 16, batch 20150, giga_loss[loss=0.3407, simple_loss=0.3954, pruned_loss=0.143, over 27592.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3294, pruned_loss=0.09127, over 5719043.20 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3454, pruned_loss=0.09187, over 5738183.82 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3264, pruned_loss=0.09068, over 5715537.03 frames. ], batch size: 472, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:38:42,413 INFO [optim.py:369] (1/2) 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 08:38:46,904 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 08:38:54,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-08 08:39:10,110 INFO [train.py:968] (1/2) Epoch 16, batch 20200, giga_loss[loss=0.308, simple_loss=0.3708, pruned_loss=0.1226, over 28676.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3398, pruned_loss=0.09846, over 5700880.44 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3453, pruned_loss=0.09169, over 5742782.79 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3373, pruned_loss=0.09825, over 5693186.40 frames. ], batch size: 242, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:39:56,484 INFO [train.py:968] (1/2) Epoch 16, batch 20250, giga_loss[loss=0.2729, simple_loss=0.3556, pruned_loss=0.09511, over 28966.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3445, pruned_loss=0.1005, over 5699785.39 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3452, pruned_loss=0.09166, over 5746075.09 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3426, pruned_loss=0.1005, over 5689802.97 frames. ], batch size: 164, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:40:02,735 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:40:06,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=704210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:40:07,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3418, 1.5116, 1.3723, 1.2504], device='cuda:1'), covar=tensor([0.2247, 0.2087, 0.1558, 0.2018], device='cuda:1'), in_proj_covar=tensor([0.1840, 0.1753, 0.1693, 0.1842], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 08:40:20,284 INFO [optim.py:369] (1/2) 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 08:40:32,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=704239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:40:42,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3540, 1.5670, 1.3512, 1.5054], device='cuda:1'), covar=tensor([0.0813, 0.0337, 0.0337, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 08:40:43,169 INFO [train.py:968] (1/2) Epoch 16, batch 20300, giga_loss[loss=0.2851, simple_loss=0.3683, pruned_loss=0.101, over 28875.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3487, pruned_loss=0.1021, over 5685227.27 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3451, pruned_loss=0.09165, over 5742839.88 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3473, pruned_loss=0.1024, over 5678362.43 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:41:03,091 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1873, 1.2309, 1.1549, 0.8849], device='cuda:1'), covar=tensor([0.0995, 0.0538, 0.1059, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0442, 0.0510, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 08:41:23,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-08 08:41:23,910 INFO [train.py:968] (1/2) Epoch 16, batch 20350, giga_loss[loss=0.4611, simple_loss=0.473, pruned_loss=0.2246, over 26545.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.353, pruned_loss=0.1042, over 5681875.90 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3452, pruned_loss=0.09179, over 5739595.17 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.352, pruned_loss=0.1048, over 5676479.79 frames. ], batch size: 555, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:41:47,023 INFO [optim.py:369] (1/2) 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 08:41:49,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:41:52,581 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-08 08:42:08,570 INFO [train.py:968] (1/2) Epoch 16, batch 20400, giga_loss[loss=0.2657, simple_loss=0.3458, pruned_loss=0.09283, over 28662.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3594, pruned_loss=0.1085, over 5678199.32 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3453, pruned_loss=0.09183, over 5742318.16 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3588, pruned_loss=0.1093, over 5670273.53 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:42:50,928 INFO [train.py:968] (1/2) Epoch 16, batch 20450, giga_loss[loss=0.2683, simple_loss=0.3217, pruned_loss=0.1075, over 23722.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3553, pruned_loss=0.1052, over 5685040.65 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3461, pruned_loss=0.09237, over 5746345.27 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3546, pruned_loss=0.106, over 5672530.28 frames. ], batch size: 705, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:43:10,732 INFO [optim.py:369] (1/2) 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 08:43:31,377 INFO [train.py:968] (1/2) Epoch 16, batch 20500, giga_loss[loss=0.2524, simple_loss=0.3369, pruned_loss=0.08392, over 28935.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.352, pruned_loss=0.1023, over 5697799.88 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.346, pruned_loss=0.09239, over 5752384.66 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3518, pruned_loss=0.1034, over 5679157.25 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:43:33,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:43:50,433 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:43:52,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=704474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:44:15,550 INFO [train.py:968] (1/2) Epoch 16, batch 20550, giga_loss[loss=0.26, simple_loss=0.3377, pruned_loss=0.09112, over 28575.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3511, pruned_loss=0.1015, over 5701449.49 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3458, pruned_loss=0.09228, over 5753240.70 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3512, pruned_loss=0.1026, over 5685224.81 frames. ], batch size: 78, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:44:17,037 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 20600, libri_loss[loss=0.2606, simple_loss=0.3341, pruned_loss=0.09358, over 29669.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3513, pruned_loss=0.1009, over 5700389.58 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3458, pruned_loss=0.09231, over 5757419.12 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3516, pruned_loss=0.1022, over 5680638.49 frames. ], batch size: 73, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:45:24,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0379, 1.1105, 3.3573, 3.0263], device='cuda:1'), covar=tensor([0.1688, 0.2781, 0.0468, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0707, 0.0615, 0.0902, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 08:45:38,467 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=704596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:45:39,015 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:45:40,389 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 20650, giga_loss[loss=0.3139, simple_loss=0.3823, pruned_loss=0.1227, over 28948.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.353, pruned_loss=0.102, over 5693730.23 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3457, pruned_loss=0.09229, over 5749257.81 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3534, pruned_loss=0.1031, over 5684359.58 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:45:45,110 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704605.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:46:01,464 INFO [optim.py:369] (1/2) 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 08:46:07,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=704628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:46:26,428 INFO [train.py:968] (1/2) Epoch 16, batch 20700, giga_loss[loss=0.2647, simple_loss=0.3432, pruned_loss=0.09313, over 28898.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3538, pruned_loss=0.1029, over 5705276.08 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3455, pruned_loss=0.09225, over 5750780.69 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3544, pruned_loss=0.1039, over 5696136.94 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:46:58,540 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-08 08:47:13,536 INFO [train.py:968] (1/2) Epoch 16, batch 20750, giga_loss[loss=0.3504, simple_loss=0.4, pruned_loss=0.1504, over 27943.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3555, pruned_loss=0.1045, over 5682763.93 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3458, pruned_loss=0.09239, over 5741816.14 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1054, over 5681998.43 frames. ], batch size: 412, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:47:33,468 INFO [optim.py:369] (1/2) 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 08:47:47,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8382, 2.9085, 2.0768, 0.9358], device='cuda:1'), covar=tensor([0.5936, 0.2434, 0.2891, 0.5592], device='cuda:1'), in_proj_covar=tensor([0.1633, 0.1553, 0.1535, 0.1346], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 08:47:57,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6014, 1.7152, 1.5342, 1.4721], device='cuda:1'), covar=tensor([0.2304, 0.2135, 0.1787, 0.2048], device='cuda:1'), in_proj_covar=tensor([0.1852, 0.1766, 0.1699, 0.1849], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 08:47:58,736 INFO [train.py:968] (1/2) Epoch 16, batch 20800, giga_loss[loss=0.2703, simple_loss=0.3467, pruned_loss=0.09695, over 29037.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5681054.54 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3461, pruned_loss=0.09265, over 5736118.85 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3572, pruned_loss=0.1067, over 5683927.95 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:48:23,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2496, 2.5523, 1.2617, 1.3011], device='cuda:1'), covar=tensor([0.1004, 0.0338, 0.0902, 0.1459], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0528, 0.0362, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 08:48:38,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2830, 3.0934, 2.9285, 1.3321], device='cuda:1'), covar=tensor([0.0889, 0.1072, 0.0950, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.1124, 0.1041, 0.0893, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 08:48:38,712 INFO [train.py:968] (1/2) Epoch 16, batch 20850, giga_loss[loss=0.28, simple_loss=0.3629, pruned_loss=0.09856, over 28571.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.357, pruned_loss=0.1059, over 5675981.31 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3464, pruned_loss=0.0929, over 5724537.55 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3571, pruned_loss=0.1066, over 5687358.23 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:48:39,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-08 08:48:57,332 INFO [optim.py:369] (1/2) 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 08:49:05,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=704835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:49:17,489 INFO [train.py:968] (1/2) Epoch 16, batch 20900, giga_loss[loss=0.3679, simple_loss=0.4212, pruned_loss=0.1573, over 28556.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3584, pruned_loss=0.1063, over 5682716.69 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3472, pruned_loss=0.09334, over 5728063.12 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3581, pruned_loss=0.1068, over 5687649.29 frames. ], batch size: 336, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:49:55,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3806, 2.7549, 2.0124, 2.5161], device='cuda:1'), covar=tensor([0.0766, 0.0477, 0.0834, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0441, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 08:49:58,425 INFO [train.py:968] (1/2) Epoch 16, batch 20950, giga_loss[loss=0.2623, simple_loss=0.3461, pruned_loss=0.08924, over 29038.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3577, pruned_loss=0.1044, over 5688378.40 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3469, pruned_loss=0.09323, over 5721179.59 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3578, pruned_loss=0.1051, over 5696997.92 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:50:19,397 INFO [optim.py:369] (1/2) 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 08:50:21,180 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-08 08:50:30,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7287, 4.5586, 4.3172, 2.0749], device='cuda:1'), covar=tensor([0.0468, 0.0660, 0.0644, 0.2103], device='cuda:1'), in_proj_covar=tensor([0.1127, 0.1041, 0.0894, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 08:50:39,217 INFO [train.py:968] (1/2) Epoch 16, batch 21000, giga_loss[loss=0.2642, simple_loss=0.341, pruned_loss=0.09372, over 28947.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3577, pruned_loss=0.1042, over 5689495.98 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3474, pruned_loss=0.09347, over 5726575.33 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3576, pruned_loss=0.1048, over 5690599.60 frames. ], batch size: 227, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:50:39,217 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 08:50:48,174 INFO [train.py:1012] (1/2) Epoch 16, validation: loss=0.2109, simple_loss=0.3179, pruned_loss=0.05197, over 944034.00 frames. 
+2023-03-08 08:50:48,175 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 08:51:04,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:51:10,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=704980.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:51:27,974 INFO [train.py:968] (1/2) Epoch 16, batch 21050, giga_loss[loss=0.2307, simple_loss=0.3169, pruned_loss=0.07218, over 28429.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3552, pruned_loss=0.1029, over 5696190.48 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3475, pruned_loss=0.09356, over 5721487.45 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3553, pruned_loss=0.1035, over 5700508.34 frames. ], batch size: 60, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:51:47,594 INFO [optim.py:369] (1/2) 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 08:52:09,130 INFO [train.py:968] (1/2) Epoch 16, batch 21100, giga_loss[loss=0.2617, simple_loss=0.3362, pruned_loss=0.09361, over 28888.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3524, pruned_loss=0.1016, over 5702798.83 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3471, pruned_loss=0.09339, over 5724441.35 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3529, pruned_loss=0.1024, over 5703147.61 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:52:48,805 INFO [train.py:968] (1/2) Epoch 16, batch 21150, giga_loss[loss=0.2661, simple_loss=0.3397, pruned_loss=0.0962, over 28829.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3507, pruned_loss=0.1007, over 5712663.90 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3473, pruned_loss=0.09353, over 5728177.14 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3511, pruned_loss=0.1014, over 5709162.33 frames. ], batch size: 119, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:52:52,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.14 vs. limit=5.0
+2023-03-08 08:52:58,978 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2020, 0.8780, 0.9608, 1.3555], device='cuda:1'), covar=tensor([0.0791, 0.0342, 0.0341, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 08:52:59,683 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:53:02,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:53:06,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705123.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:53:09,265 INFO [optim.py:369] (1/2) 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 08:53:10,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705126.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 08:53:27,440 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:53:29,989 INFO [train.py:968] (1/2) Epoch 16, batch 21200, libri_loss[loss=0.282, simple_loss=0.3582, pruned_loss=0.103, over 20085.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3514, pruned_loss=0.1017, over 5692206.36 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3482, pruned_loss=0.09428, over 5713077.46 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.351, pruned_loss=0.1019, over 5703300.64 frames. ], batch size: 187, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:53:33,082 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705155.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 08:53:36,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5199, 1.6635, 1.3789, 1.5518], device='cuda:1'), covar=tensor([0.2790, 0.2750, 0.3073, 0.2427], device='cuda:1'), in_proj_covar=tensor([0.1398, 0.1024, 0.1241, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 08:54:09,467 INFO [train.py:968] (1/2) Epoch 16, batch 21250, giga_loss[loss=0.2666, simple_loss=0.3415, pruned_loss=0.09578, over 28666.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3523, pruned_loss=0.102, over 5694878.61 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3487, pruned_loss=0.09467, over 5706010.36 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3517, pruned_loss=0.102, over 5709813.28 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:54:17,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:54:29,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:54:31,358 INFO [optim.py:369] (1/2) 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 08:54:54,504 INFO [train.py:968] (1/2) Epoch 16, batch 21300, giga_loss[loss=0.2792, simple_loss=0.3344, pruned_loss=0.112, over 23580.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1012, over 5682299.07 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3491, pruned_loss=0.09505, over 5699830.89 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3506, pruned_loss=0.1009, over 5699872.78 frames. ], batch size: 705, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:55:07,427 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-08 08:55:19,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:55:35,778 INFO [train.py:968] (1/2) Epoch 16, batch 21350, giga_loss[loss=0.2773, simple_loss=0.3522, pruned_loss=0.1012, over 28885.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3499, pruned_loss=0.09909, over 5696109.62 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3492, pruned_loss=0.09511, over 5699023.73 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3491, pruned_loss=0.09888, over 5710566.75 frames. ], batch size: 199, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:55:47,374 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:55:56,337 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:55:58,987 INFO [optim.py:369] (1/2) 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 08:56:18,842 INFO [train.py:968] (1/2) Epoch 16, batch 21400, giga_loss[loss=0.3321, simple_loss=0.3893, pruned_loss=0.1375, over 28538.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3504, pruned_loss=0.0997, over 5707052.47 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3498, pruned_loss=0.09561, over 5701020.75 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3493, pruned_loss=0.09912, over 5716618.10 frames. ], batch size: 85, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:56:20,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:56:23,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:56:47,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:56:58,685 INFO [train.py:968] (1/2) Epoch 16, batch 21450, giga_loss[loss=0.228, simple_loss=0.3131, pruned_loss=0.07149, over 28838.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3486, pruned_loss=0.09934, over 5711849.19 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3506, pruned_loss=0.09636, over 5701493.71 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.347, pruned_loss=0.09828, over 5719302.53 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:57:19,415 INFO [optim.py:369] (1/2) 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 08:57:38,698 INFO [train.py:968] (1/2) Epoch 16, batch 21500, giga_loss[loss=0.2446, simple_loss=0.3219, pruned_loss=0.08367, over 28514.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3453, pruned_loss=0.0977, over 5705662.11 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3505, pruned_loss=0.09644, over 5698690.40 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.344, pruned_loss=0.09684, over 5714657.12 frames. ], batch size: 65, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:58:19,964 INFO [train.py:968] (1/2) Epoch 16, batch 21550, giga_loss[loss=0.3252, simple_loss=0.3939, pruned_loss=0.1283, over 28661.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3451, pruned_loss=0.09754, over 5714501.53 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3506, pruned_loss=0.09657, over 5702574.72 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3438, pruned_loss=0.09677, over 5718382.96 frames. ], batch size: 242, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 08:58:29,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2336, 1.4436, 1.2981, 1.1260], device='cuda:1'), covar=tensor([0.2328, 0.2295, 0.1475, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1767, 0.1691, 0.1838], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 08:58:31,619 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-08 08:58:41,442 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 21600, libri_loss[loss=0.3225, simple_loss=0.3912, pruned_loss=0.127, over 28713.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3455, pruned_loss=0.09868, over 5712716.43 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.351, pruned_loss=0.09697, over 5704988.42 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.344, pruned_loss=0.0977, over 5714040.85 frames. ], batch size: 106, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 08:59:43,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3735, 1.7902, 1.4797, 1.5170], device='cuda:1'), covar=tensor([0.0736, 0.0280, 0.0316, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0057, 0.0098], device='cuda:1')
+2023-03-08 08:59:44,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 08:59:45,543 INFO [train.py:968] (1/2) Epoch 16, batch 21650, giga_loss[loss=0.2344, simple_loss=0.3025, pruned_loss=0.08315, over 28630.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3426, pruned_loss=0.09763, over 5715539.46 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.351, pruned_loss=0.09705, over 5706061.96 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3413, pruned_loss=0.0968, over 5715604.32 frames. ], batch size: 92, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:00:06,541 INFO [optim.py:369] (1/2) 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 09:00:19,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3585, 3.0643, 1.3881, 1.5053], device='cuda:1'), covar=tensor([0.0921, 0.0407, 0.0898, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0524, 0.0359, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 09:00:25,193 INFO [train.py:968] (1/2) Epoch 16, batch 21700, giga_loss[loss=0.3193, simple_loss=0.3785, pruned_loss=0.13, over 28075.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3416, pruned_loss=0.09773, over 5716029.40 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3512, pruned_loss=0.09724, over 5708807.21 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3402, pruned_loss=0.09691, over 5713770.62 frames. ], batch size: 412, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:00:27,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:00:48,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3094, 1.1809, 3.7614, 3.2948], device='cuda:1'), covar=tensor([0.1463, 0.2704, 0.0368, 0.1728], device='cuda:1'), in_proj_covar=tensor([0.0698, 0.0608, 0.0890, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 09:00:53,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=705687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:01:01,795 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 21750, giga_loss[loss=0.2634, simple_loss=0.3333, pruned_loss=0.09673, over 29063.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3391, pruned_loss=0.09695, over 5710028.63 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3516, pruned_loss=0.09772, over 5707516.44 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3373, pruned_loss=0.09585, over 5709572.62 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:01:04,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2643, 2.0353, 1.4930, 0.5240], device='cuda:1'), covar=tensor([0.3991, 0.1959, 0.2881, 0.4052], device='cuda:1'), in_proj_covar=tensor([0.1633, 0.1548, 0.1534, 0.1348], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 09:01:25,407 INFO [optim.py:369] (1/2) 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 09:01:26,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4034, 1.3328, 4.2186, 3.4574], device='cuda:1'), covar=tensor([0.1606, 0.2678, 0.0350, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0700, 0.0609, 0.0892, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 09:01:28,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3400, 1.5980, 1.6154, 1.3878], device='cuda:1'), covar=tensor([0.2599, 0.2046, 0.1460, 0.2036], device='cuda:1'), in_proj_covar=tensor([0.1834, 0.1764, 0.1689, 0.1833], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 09:01:30,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-08 09:01:35,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:01:38,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:01:44,169 INFO [train.py:968] (1/2) Epoch 16, batch 21800, giga_loss[loss=0.2504, simple_loss=0.3215, pruned_loss=0.08962, over 28338.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3368, pruned_loss=0.0957, over 5708737.74 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.352, pruned_loss=0.0981, over 5710645.22 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3347, pruned_loss=0.09443, over 5705625.49 frames. ], batch size: 77, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:02:05,716 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:26,454 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:28,673 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 21850, giga_loss[loss=0.2748, simple_loss=0.3524, pruned_loss=0.09857, over 28704.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3369, pruned_loss=0.09554, over 5706772.23 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3522, pruned_loss=0.09842, over 5713619.64 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3349, pruned_loss=0.0942, over 5701644.73 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:02:31,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4219, 1.6310, 1.2999, 1.5913], device='cuda:1'), covar=tensor([0.0724, 0.0292, 0.0335, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0090, 0.0064, 0.0058, 0.0098], device='cuda:1')
+2023-03-08 09:02:41,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9358, 2.2753, 2.2073, 1.8604], device='cuda:1'), covar=tensor([0.3004, 0.2067, 0.1868, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.1834, 0.1764, 0.1688, 0.1833], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 09:02:51,894 INFO [optim.py:369] (1/2) 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 09:02:52,658 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:53,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:55,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=705833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:02:55,911 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-08 09:03:03,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=705841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:03:08,152 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 21900, giga_loss[loss=0.2681, simple_loss=0.3477, pruned_loss=0.09426, over 28847.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3398, pruned_loss=0.0966, over 5709457.24 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3526, pruned_loss=0.09881, over 5715994.88 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3375, pruned_loss=0.09516, over 5703157.29 frames. ], batch size: 186, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:03:20,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=705862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:03:31,819 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 21950, giga_loss[loss=0.2562, simple_loss=0.3371, pruned_loss=0.08764, over 29059.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3423, pruned_loss=0.09717, over 5708906.60 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3531, pruned_loss=0.09909, over 5710217.33 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3398, pruned_loss=0.09571, over 5708457.77 frames. ], batch size: 128, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:04:06,858 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=705915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:04:17,240 INFO [optim.py:369] (1/2) 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 09:04:39,795 INFO [train.py:968] (1/2) Epoch 16, batch 22000, giga_loss[loss=0.3042, simple_loss=0.3649, pruned_loss=0.1217, over 23785.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3446, pruned_loss=0.0978, over 5699035.79 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3536, pruned_loss=0.09954, over 5712682.52 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.342, pruned_loss=0.09625, over 5696531.12 frames. ], batch size: 705, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:05:15,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 09:05:20,742 INFO [train.py:968] (1/2) Epoch 16, batch 22050, giga_loss[loss=0.2446, simple_loss=0.331, pruned_loss=0.0791, over 28656.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3437, pruned_loss=0.09677, over 5697639.84 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3546, pruned_loss=0.1005, over 5714546.94 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3404, pruned_loss=0.09458, over 5693379.06 frames. ], batch size: 262, lr: 2.01e-03, grad_scale: 8.0
+2023-03-08 09:05:45,674 INFO [optim.py:369] (1/2) 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 09:06:04,910 INFO [train.py:968] (1/2) Epoch 16, batch 22100, giga_loss[loss=0.2539, simple_loss=0.333, pruned_loss=0.08734, over 28945.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3436, pruned_loss=0.0968, over 5694077.01 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3548, pruned_loss=0.101, over 5709447.77 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3403, pruned_loss=0.09443, over 5693962.31 frames. ], batch size: 174, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:06:44,213 INFO [train.py:968] (1/2) Epoch 16, batch 22150, libri_loss[loss=0.2417, simple_loss=0.3175, pruned_loss=0.08298, over 29587.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.344, pruned_loss=0.09735, over 5701136.10 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3546, pruned_loss=0.1011, over 5712089.47 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3412, pruned_loss=0.09523, over 5698109.76 frames. ], batch size: 75, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:07:07,832 INFO [optim.py:369] (1/2) 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 09:07:25,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:07:26,131 INFO [train.py:968] (1/2) Epoch 16, batch 22200, giga_loss[loss=0.304, simple_loss=0.3757, pruned_loss=0.1162, over 28930.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3457, pruned_loss=0.09867, over 5702286.03 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3548, pruned_loss=0.1012, over 5712361.02 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3433, pruned_loss=0.0969, over 5699615.11 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:08:10,699 INFO [train.py:968] (1/2) Epoch 16, batch 22250, giga_loss[loss=0.2512, simple_loss=0.3323, pruned_loss=0.08505, over 28530.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3489, pruned_loss=0.1006, over 5702941.50 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.355, pruned_loss=0.1012, over 5714690.83 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3468, pruned_loss=0.09917, over 5698680.17 frames. ], batch size: 71, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:08:33,064 INFO [optim.py:369] (1/2) 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 09:08:52,910 INFO [train.py:968] (1/2) Epoch 16, batch 22300, giga_loss[loss=0.2661, simple_loss=0.347, pruned_loss=0.09255, over 28986.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3517, pruned_loss=0.1023, over 5703667.95 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3553, pruned_loss=0.1014, over 5716361.01 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3497, pruned_loss=0.1009, over 5698741.08 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:08:59,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4274, 3.6373, 1.5169, 1.6786], device='cuda:1'), covar=tensor([0.0952, 0.0476, 0.0862, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0529, 0.0361, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 09:08:59,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5928, 1.8945, 1.5241, 1.7557], device='cuda:1'), covar=tensor([0.2276, 0.2329, 0.2593, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.1393, 0.1015, 0.1235, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0010, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 09:09:18,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-08 09:09:23,873 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 22350, giga_loss[loss=0.286, simple_loss=0.3644, pruned_loss=0.1038, over 28191.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3525, pruned_loss=0.1023, over 5698246.06 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3561, pruned_loss=0.1021, over 5705477.72 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3501, pruned_loss=0.1006, over 5704252.82 frames. ], batch size: 368, lr: 2.01e-03, grad_scale: 2.0
+2023-03-08 09:09:43,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-08 09:09:54,491 INFO [optim.py:369] (1/2) 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 09:10:11,636 INFO [train.py:968] (1/2) Epoch 16, batch 22400, giga_loss[loss=0.2398, simple_loss=0.3179, pruned_loss=0.08084, over 28774.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3535, pruned_loss=0.1024, over 5706191.88 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3565, pruned_loss=0.1023, over 5709457.08 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3511, pruned_loss=0.1009, over 5707206.66 frames. ], batch size: 99, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:10:15,639 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:10:58,603 INFO [train.py:968] (1/2) Epoch 16, batch 22450, giga_loss[loss=0.3001, simple_loss=0.3693, pruned_loss=0.1154, over 29045.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3543, pruned_loss=0.1031, over 5711950.06 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3568, pruned_loss=0.1026, over 5712061.55 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3521, pruned_loss=0.1017, over 5710545.36 frames. ], batch size: 136, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:11:18,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 09:11:24,282 INFO [optim.py:369] (1/2) 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 09:11:26,568 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=706433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:11:29,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=706436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:11:40,902 INFO [train.py:968] (1/2) Epoch 16, batch 22500, giga_loss[loss=0.25, simple_loss=0.3247, pruned_loss=0.08763, over 28735.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.353, pruned_loss=0.1028, over 5711208.18 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3571, pruned_loss=0.1028, over 5715081.71 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3509, pruned_loss=0.1014, over 5707078.12 frames. ], batch size: 119, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:11:51,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=706465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:12:22,245 INFO [train.py:968] (1/2) Epoch 16, batch 22550, giga_loss[loss=0.2214, simple_loss=0.3092, pruned_loss=0.06681, over 29062.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3509, pruned_loss=0.102, over 5717882.17 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3571, pruned_loss=0.1031, over 5721547.00 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.349, pruned_loss=0.1006, over 5708593.10 frames. ], batch size: 155, lr: 2.01e-03, grad_scale: 4.0
+2023-03-08 09:12:27,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:12:32,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9645, 1.3196, 1.4092, 1.0920], device='cuda:1'), covar=tensor([0.1593, 0.1080, 0.1917, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0735, 0.0691, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 09:12:41,101 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:12:56,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 09:13:01,522 INFO [train.py:968] (1/2) Epoch 16, batch 22600, giga_loss[loss=0.227, simple_loss=0.3051, pruned_loss=0.07447, over 29073.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3479, pruned_loss=0.1008, over 5710968.89 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3577, pruned_loss=0.1039, over 5717725.53 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3456, pruned_loss=0.09885, over 5707094.51 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:13:03,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3219, 1.4981, 1.3407, 1.2511], device='cuda:1'), covar=tensor([0.2716, 0.2162, 0.1884, 0.2238], device='cuda:1'), in_proj_covar=tensor([0.1850, 0.1786, 0.1721, 0.1854], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 09:13:10,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-08 09:13:38,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3787, 3.4866, 1.4633, 1.5051], device='cuda:1'), covar=tensor([0.0951, 0.0265, 0.0953, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0529, 0.0360, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 09:13:41,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 09:13:44,021 INFO [train.py:968] (1/2) Epoch 16, batch 22650, giga_loss[loss=0.3172, simple_loss=0.3888, pruned_loss=0.1229, over 28634.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3461, pruned_loss=0.09898, over 5708804.80 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3577, pruned_loss=0.1039, over 5719640.62 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3442, pruned_loss=0.09742, over 5704193.02 frames. ], batch size: 307, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:14:09,926 INFO [optim.py:369] (1/2) 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 09:14:28,328 INFO [train.py:968] (1/2) Epoch 16, batch 22700, giga_loss[loss=0.3151, simple_loss=0.378, pruned_loss=0.1261, over 28990.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.09829, over 5697376.22 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3574, pruned_loss=0.1039, over 5710293.91 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3463, pruned_loss=0.09701, over 5701260.70 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:14:44,095 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=706668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:14:47,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=706671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:15:09,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1190, 1.1485, 3.3975, 3.0035], device='cuda:1'), covar=tensor([0.1613, 0.2664, 0.0463, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0702, 0.0613, 0.0896, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 09:15:10,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3157, 1.1822, 4.1033, 3.3556], device='cuda:1'), covar=tensor([0.1640, 0.2780, 0.0397, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0702, 0.0613, 0.0896, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 09:15:11,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=706700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:15:12,240 INFO [train.py:968] (1/2) Epoch 16, batch 22750, libri_loss[loss=0.3336, simple_loss=0.3925, pruned_loss=0.1374, over 28645.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3494, pruned_loss=0.09906, over 5684473.95 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3579, pruned_loss=0.1043, over 5701003.54 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3476, pruned_loss=0.09754, over 5696287.00 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:15:32,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=706728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:15:34,125 INFO [optim.py:369] (1/2) 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 09:15:39,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=706735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:15:53,504 INFO [train.py:968] (1/2) Epoch 16, batch 22800, giga_loss[loss=0.2592, simple_loss=0.3355, pruned_loss=0.09145, over 28905.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3489, pruned_loss=0.1002, over 5691773.70 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3584, pruned_loss=0.1048, over 5707472.58 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3466, pruned_loss=0.09832, over 5694848.84 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:15:55,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5422, 1.7202, 1.6440, 1.5487], device='cuda:1'), covar=tensor([0.1670, 0.1899, 0.2168, 0.1973], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0734, 0.0688, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 09:16:34,067 INFO [train.py:968] (1/2) Epoch 16, batch 22850, giga_loss[loss=0.2858, simple_loss=0.3497, pruned_loss=0.111, over 28996.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.349, pruned_loss=0.1021, over 5701731.91 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3586, pruned_loss=0.1053, over 5713659.13 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3467, pruned_loss=0.1, over 5698596.25 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:16:37,285 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-08 09:16:54,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-08 09:16:59,634 INFO [optim.py:369] (1/2) 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 09:17:16,174 INFO [train.py:968] (1/2) Epoch 16, batch 22900, giga_loss[loss=0.2545, simple_loss=0.3274, pruned_loss=0.09083, over 28872.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3468, pruned_loss=0.1017, over 5714444.92 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.359, pruned_loss=0.1056, over 5718322.99 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3444, pruned_loss=0.09978, over 5707567.95 frames. ], batch size: 186, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:17:22,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4118, 3.3047, 1.5146, 1.5737], device='cuda:1'), covar=tensor([0.0932, 0.0404, 0.0923, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0529, 0.0360, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 09:17:32,370 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=706871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:17:35,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=706874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:17:41,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=706882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:17:58,267 INFO [train.py:968] (1/2) Epoch 16, batch 22950, giga_loss[loss=0.2745, simple_loss=0.3473, pruned_loss=0.1008, over 28658.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.345, pruned_loss=0.1016, over 5711162.15 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3593, pruned_loss=0.1058, over 5721075.90 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3426, pruned_loss=0.0998, over 5703069.66 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:17:59,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=706903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:18:07,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2946, 1.2928, 3.8452, 3.1846], device='cuda:1'), covar=tensor([0.1612, 0.2674, 0.0404, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0703, 0.0614, 0.0899, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 09:18:20,370 INFO [optim.py:369] (1/2) 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 09:18:37,307 INFO [train.py:968] (1/2) Epoch 16, batch 23000, giga_loss[loss=0.2632, simple_loss=0.3339, pruned_loss=0.0962, over 29003.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.344, pruned_loss=0.1006, over 5722315.16 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3591, pruned_loss=0.1058, over 5724535.58 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3419, pruned_loss=0.09899, over 5712559.07 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:18:47,121 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 09:19:12,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2634, 1.4418, 1.4061, 1.2949], device='cuda:1'), covar=tensor([0.2366, 0.1952, 0.1536, 0.1925], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1789, 0.1722, 0.1862], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 09:19:16,943 INFO [train.py:968] (1/2) Epoch 16, batch 23050, giga_loss[loss=0.2882, simple_loss=0.3491, pruned_loss=0.1137, over 28555.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.342, pruned_loss=0.1004, over 5716303.96 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3601, pruned_loss=0.1067, over 5726560.17 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3391, pruned_loss=0.09821, over 5706519.18 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:19:35,868 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:19:38,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:19:40,274 INFO [optim.py:369] (1/2) 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 09:19:46,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4063, 1.5908, 1.6635, 1.2373], device='cuda:1'), covar=tensor([0.1630, 0.2389, 0.1384, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0690, 0.0909, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 09:19:58,328 INFO [train.py:968] (1/2) Epoch 16, batch 23100, libri_loss[loss=0.3013, simple_loss=0.3702, pruned_loss=0.1161, over 29537.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3379, pruned_loss=0.09803, over 5720370.49 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3608, pruned_loss=0.1073, over 5732648.12 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3342, pruned_loss=0.09545, over 5706417.41 frames. ], batch size: 83, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:19:58,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:20:02,785 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:20:07,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6884, 1.6741, 1.2616, 1.3469], device='cuda:1'), covar=tensor([0.0819, 0.0644, 0.1119, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0372, 0.0441, 0.0506, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 09:20:22,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1493, 1.6218, 1.2059, 0.4081], device='cuda:1'), covar=tensor([0.3710, 0.2007, 0.3006, 0.5611], device='cuda:1'), in_proj_covar=tensor([0.1642, 0.1550, 0.1534, 0.1352], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 09:20:35,416 INFO [train.py:968] (1/2) Epoch 16, batch 23150, giga_loss[loss=0.2744, simple_loss=0.3462, pruned_loss=0.1013, over 28768.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3356, pruned_loss=0.09652, over 5716560.93 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3609, pruned_loss=0.1075, over 5729063.91 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.332, pruned_loss=0.09403, over 5707944.41 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:20:37,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9253, 5.7157, 5.4473, 2.8528], device='cuda:1'), covar=tensor([0.0371, 0.0534, 0.0604, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.1126, 0.1042, 0.0897, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 09:20:41,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:20:59,417 INFO [optim.py:369] (1/2) 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 09:21:16,628 INFO [train.py:968] (1/2) Epoch 16, batch 23200, libri_loss[loss=0.2956, simple_loss=0.3575, pruned_loss=0.1168, over 29332.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3377, pruned_loss=0.09727, over 5718594.28 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.361, pruned_loss=0.1078, over 5734373.03 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3339, pruned_loss=0.0946, over 5706294.72 frames. ], batch size: 71, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:21:59,167 INFO [train.py:968] (1/2) Epoch 16, batch 23250, libri_loss[loss=0.2815, simple_loss=0.3408, pruned_loss=0.1111, over 29652.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3409, pruned_loss=0.09881, over 5720437.41 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3609, pruned_loss=0.1079, over 5736209.00 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3377, pruned_loss=0.0965, over 5708801.16 frames. ], batch size: 69, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:22:24,977 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 23300, giga_loss[loss=0.2918, simple_loss=0.364, pruned_loss=0.1098, over 28567.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3447, pruned_loss=0.1003, over 5717761.85 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3612, pruned_loss=0.1081, over 5738161.54 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3417, pruned_loss=0.09819, over 5706561.82 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:22:43,177 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707253.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:22:45,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:23:08,010 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:23:19,652 INFO [train.py:968] (1/2) Epoch 16, batch 23350, giga_loss[loss=0.3026, simple_loss=0.3766, pruned_loss=0.1143, over 28974.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3482, pruned_loss=0.1017, over 5707999.21 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3611, pruned_loss=0.1081, over 5733906.31 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3453, pruned_loss=0.09967, over 5702162.36 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:23:38,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8728, 3.6782, 3.4619, 1.7598], device='cuda:1'), covar=tensor([0.0630, 0.0766, 0.0710, 0.2201], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.1044, 0.0899, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 09:23:47,235 INFO [optim.py:369] (1/2) 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 09:24:03,612 INFO [train.py:968] (1/2) Epoch 16, batch 23400, giga_loss[loss=0.3687, simple_loss=0.4159, pruned_loss=0.1607, over 27951.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3512, pruned_loss=0.1033, over 5702386.46 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3616, pruned_loss=0.1086, over 5734927.80 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.348, pruned_loss=0.101, over 5696075.13 frames. ], batch size: 412, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:24:07,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:24:51,550 INFO [train.py:968] (1/2) Epoch 16, batch 23450, giga_loss[loss=0.3168, simple_loss=0.3884, pruned_loss=0.1226, over 28916.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.355, pruned_loss=0.1072, over 5697604.50 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3611, pruned_loss=0.1085, over 5738620.93 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3528, pruned_loss=0.1054, over 5688806.83 frames. ], batch size: 174, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:25:12,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2330, 1.0932, 3.8883, 3.1747], device='cuda:1'), covar=tensor([0.1722, 0.2882, 0.0451, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0615, 0.0904, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 09:25:15,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:25:21,464 INFO [optim.py:369] (1/2) 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 09:25:37,278 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:25:39,360 INFO [train.py:968] (1/2) Epoch 16, batch 23500, giga_loss[loss=0.3457, simple_loss=0.3982, pruned_loss=0.1466, over 27591.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3597, pruned_loss=0.1113, over 5684143.54 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3607, pruned_loss=0.1085, over 5731993.85 frames. ], giga_tot_loss[loss=0.2891, simple_loss=0.3582, pruned_loss=0.11, over 5682653.72 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:26:31,964 INFO [train.py:968] (1/2) Epoch 16, batch 23550, giga_loss[loss=0.2849, simple_loss=0.3622, pruned_loss=0.1038, over 28842.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3666, pruned_loss=0.116, over 5688473.57 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3606, pruned_loss=0.1085, over 5734864.89 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3656, pruned_loss=0.1151, over 5683976.58 frames. ], batch size: 112, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:26:37,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8091, 1.8168, 1.3907, 1.4649], device='cuda:1'), covar=tensor([0.0800, 0.0566, 0.0969, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0441, 0.0507, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 09:26:45,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7992, 1.8183, 1.7101, 1.6349], device='cuda:1'), covar=tensor([0.1641, 0.2052, 0.2158, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0743, 0.0696, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 09:27:02,693 INFO [optim.py:369] (1/2) 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 09:27:20,633 INFO [train.py:968] (1/2) Epoch 16, batch 23600, giga_loss[loss=0.3149, simple_loss=0.3848, pruned_loss=0.1225, over 28706.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3722, pruned_loss=0.1207, over 5686174.00 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.361, pruned_loss=0.1088, over 5740235.62 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3713, pruned_loss=0.1199, over 5676501.05 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:27:29,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-08 09:27:38,003 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:27:41,657 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:27:43,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:28:09,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5958, 1.6828, 1.2255, 1.2936], device='cuda:1'), covar=tensor([0.0796, 0.0528, 0.0935, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0443, 0.0509, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 09:28:10,665 INFO [train.py:968] (1/2) Epoch 16, batch 23650, giga_loss[loss=0.3265, simple_loss=0.3909, pruned_loss=0.131, over 28928.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.379, pruned_loss=0.1269, over 5672618.45 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3611, pruned_loss=0.1089, over 5743347.49 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3785, pruned_loss=0.1265, over 5660844.62 frames. ], batch size: 112, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:28:10,878 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:28:40,010 INFO [optim.py:369] (1/2) 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 09:28:51,949 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=707644.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:28:57,289 INFO [train.py:968] (1/2) Epoch 16, batch 23700, giga_loss[loss=0.307, simple_loss=0.3769, pruned_loss=0.1185, over 28902.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3824, pruned_loss=0.1294, over 5664939.55 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3618, pruned_loss=0.1097, over 5737353.88 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3821, pruned_loss=0.1291, over 5658550.39 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:29:20,977 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 23750, giga_loss[loss=0.4844, simple_loss=0.4838, pruned_loss=0.2425, over 26397.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3845, pruned_loss=0.1317, over 5666268.06 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3618, pruned_loss=0.1099, over 5738336.36 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3849, pruned_loss=0.1319, over 5658474.06 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:29:49,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6672, 3.5130, 3.3583, 1.6279], device='cuda:1'), covar=tensor([0.0753, 0.0825, 0.0759, 0.2357], device='cuda:1'), in_proj_covar=tensor([0.1146, 0.1058, 0.0911, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 09:30:11,048 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:30:21,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4837, 1.6954, 1.5434, 1.2818], device='cuda:1'), covar=tensor([0.2299, 0.2014, 0.1564, 0.2066], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1792, 0.1723, 0.1860], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 09:30:30,369 INFO [train.py:968] (1/2) Epoch 16, batch 23800, giga_loss[loss=0.3319, simple_loss=0.3826, pruned_loss=0.1406, over 28857.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3869, pruned_loss=0.1347, over 5665708.19 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3622, pruned_loss=0.1105, over 5734026.10 frames. ], giga_tot_loss[loss=0.3288, simple_loss=0.3875, pruned_loss=0.135, over 5661519.57 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:30:45,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4112, 1.6532, 1.3243, 1.3916], device='cuda:1'), covar=tensor([0.2402, 0.2491, 0.2765, 0.2063], device='cuda:1'), in_proj_covar=tensor([0.1400, 0.1026, 0.1241, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 09:30:57,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-08 09:31:24,352 INFO [train.py:968] (1/2) Epoch 16, batch 23850, giga_loss[loss=0.3133, simple_loss=0.3722, pruned_loss=0.1271, over 28907.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3894, pruned_loss=0.138, over 5642930.56 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3622, pruned_loss=0.1104, over 5735833.97 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3901, pruned_loss=0.1385, over 5637262.64 frames. ], batch size: 213, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:31:52,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:32:05,678 INFO [optim.py:369] (1/2) 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 09:32:25,217 INFO [train.py:968] (1/2) Epoch 16, batch 23900, giga_loss[loss=0.3848, simple_loss=0.4261, pruned_loss=0.1717, over 28270.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.394, pruned_loss=0.1424, over 5637172.01 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3619, pruned_loss=0.1103, over 5737669.26 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3952, pruned_loss=0.1432, over 5629931.37 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:32:53,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:32:58,115 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:33:13,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6112, 2.1738, 1.4819, 0.7787], device='cuda:1'), covar=tensor([0.5915, 0.3451, 0.2712, 0.5625], device='cuda:1'), in_proj_covar=tensor([0.1653, 0.1572, 0.1548, 0.1363], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 09:33:22,281 INFO [train.py:968] (1/2) Epoch 16, batch 23950, giga_loss[loss=0.3239, simple_loss=0.3843, pruned_loss=0.1317, over 29032.00 frames. ], tot_loss[loss=0.3401, simple_loss=0.3936, pruned_loss=0.1433, over 5616200.78 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.362, pruned_loss=0.1104, over 5739865.11 frames. ], giga_tot_loss[loss=0.3418, simple_loss=0.3949, pruned_loss=0.1444, over 5606813.91 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:33:27,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:33:53,490 INFO [optim.py:369] (1/2) 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 09:34:00,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=707942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:34:09,041 INFO [train.py:968] (1/2) Epoch 16, batch 24000, giga_loss[loss=0.3267, simple_loss=0.3868, pruned_loss=0.1333, over 28905.00 frames. ], tot_loss[loss=0.3385, simple_loss=0.3919, pruned_loss=0.1425, over 5633383.94 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3621, pruned_loss=0.1105, over 5745255.64 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3938, pruned_loss=0.1444, over 5617520.85 frames. ], batch size: 285, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:34:09,042 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 09:34:17,808 INFO [train.py:1012] (1/2) Epoch 16, validation: loss=0.2099, simple_loss=0.3175, pruned_loss=0.05122, over 944034.00 frames. 
+2023-03-08 09:34:17,808 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 09:34:33,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=707967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:34:36,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=707970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:35:03,292 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=707999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:35:05,630 INFO [train.py:968] (1/2) Epoch 16, batch 24050, giga_loss[loss=0.3582, simple_loss=0.3905, pruned_loss=0.1629, over 23517.00 frames. ], tot_loss[loss=0.3365, simple_loss=0.3905, pruned_loss=0.1413, over 5628694.87 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3619, pruned_loss=0.1104, over 5740629.99 frames. ], giga_tot_loss[loss=0.3401, simple_loss=0.393, pruned_loss=0.1436, over 5616898.06 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:35:21,599 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:35:42,634 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.39 vs. limit=5.0
+2023-03-08 09:35:56,476 INFO [train.py:968] (1/2) Epoch 16, batch 24100, giga_loss[loss=0.3918, simple_loss=0.4378, pruned_loss=0.1729, over 27823.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3905, pruned_loss=0.1397, over 5628191.97 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3622, pruned_loss=0.1107, over 5741618.82 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3928, pruned_loss=0.1418, over 5616072.85 frames. ], batch size: 412, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:35:59,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=708053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:25,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=708081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:29,670 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:31,675 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:36:45,611 INFO [train.py:968] (1/2) Epoch 16, batch 24150, giga_loss[loss=0.2717, simple_loss=0.3463, pruned_loss=0.09852, over 28865.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.391, pruned_loss=0.1397, over 5619181.59 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3617, pruned_loss=0.1106, over 5731881.20 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.394, pruned_loss=0.1422, over 5614525.88 frames. ], batch size: 66, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:37:04,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:37:22,778 INFO [optim.py:369] (1/2) 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 09:37:38,654 INFO [train.py:968] (1/2) Epoch 16, batch 24200, giga_loss[loss=0.3109, simple_loss=0.3815, pruned_loss=0.1202, over 28838.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3896, pruned_loss=0.1378, over 5628990.09 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.362, pruned_loss=0.1108, over 5733493.84 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.392, pruned_loss=0.14, over 5622911.19 frames. ], batch size: 243, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:37:54,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708162.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 09:37:57,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708165.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 09:38:25,404 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 09:38:28,196 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:38:32,106 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:38:33,063 INFO [train.py:968] (1/2) Epoch 16, batch 24250, giga_loss[loss=0.353, simple_loss=0.4066, pruned_loss=0.1497, over 28317.00 frames. ], tot_loss[loss=0.3294, simple_loss=0.3879, pruned_loss=0.1354, over 5628814.64 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3621, pruned_loss=0.111, over 5735296.89 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.39, pruned_loss=0.1372, over 5621293.57 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:38:44,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7240, 2.4588, 2.1124, 1.5368], device='cuda:1'), covar=tensor([0.3449, 0.1912, 0.2004, 0.2708], device='cuda:1'), in_proj_covar=tensor([0.1865, 0.1798, 0.1732, 0.1866], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 09:39:02,882 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 24300, giga_loss[loss=0.2942, simple_loss=0.3633, pruned_loss=0.1125, over 28296.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3854, pruned_loss=0.1329, over 5629115.29 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3625, pruned_loss=0.1113, over 5737310.40 frames. ], giga_tot_loss[loss=0.3278, simple_loss=0.387, pruned_loss=0.1343, over 5620083.07 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:40:11,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2151, 1.8022, 1.3689, 0.4506], device='cuda:1'), covar=tensor([0.3385, 0.2087, 0.2987, 0.4310], device='cuda:1'), in_proj_covar=tensor([0.1649, 0.1565, 0.1544, 0.1360], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 09:40:16,255 INFO [train.py:968] (1/2) Epoch 16, batch 24350, giga_loss[loss=0.3731, simple_loss=0.4126, pruned_loss=0.1668, over 26597.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3815, pruned_loss=0.1297, over 5631858.30 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3625, pruned_loss=0.1115, over 5738302.72 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.383, pruned_loss=0.1309, over 5622394.86 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:40:48,262 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 24400, giga_loss[loss=0.3749, simple_loss=0.4125, pruned_loss=0.1686, over 26687.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3801, pruned_loss=0.1293, over 5632832.81 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3625, pruned_loss=0.1115, over 5736998.78 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3816, pruned_loss=0.1305, over 5625574.13 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:41:24,973 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=708369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:41:28,613 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-08 09:41:55,436 INFO [train.py:968] (1/2) Epoch 16, batch 24450, giga_loss[loss=0.3666, simple_loss=0.4164, pruned_loss=0.1584, over 28723.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3812, pruned_loss=0.1308, over 5637138.70 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3624, pruned_loss=0.1117, over 5739898.18 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3827, pruned_loss=0.1319, over 5626955.24 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:42:35,728 INFO [optim.py:369] (1/2) 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 09:42:50,759 INFO [train.py:968] (1/2) Epoch 16, batch 24500, giga_loss[loss=0.355, simple_loss=0.4063, pruned_loss=0.1519, over 27620.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3786, pruned_loss=0.1288, over 5645975.05 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3615, pruned_loss=0.1115, over 5742889.03 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3811, pruned_loss=0.1304, over 5632342.25 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:42:55,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=708456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:43:39,862 INFO [train.py:968] (1/2) Epoch 16, batch 24550, giga_loss[loss=0.2744, simple_loss=0.3599, pruned_loss=0.09448, over 28949.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3746, pruned_loss=0.1243, over 5662317.89 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3616, pruned_loss=0.1117, over 5747032.79 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 5643376.90 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:43:45,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-08 09:44:17,815 INFO [optim.py:369] (1/2) 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 09:44:30,768 INFO [train.py:968] (1/2) Epoch 16, batch 24600, giga_loss[loss=0.3322, simple_loss=0.3986, pruned_loss=0.1329, over 28655.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.375, pruned_loss=0.1221, over 5649275.12 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3615, pruned_loss=0.1118, over 5727356.37 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3775, pruned_loss=0.1236, over 5651407.55 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:45:22,916 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 09:45:26,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:45:27,662 INFO [train.py:968] (1/2) Epoch 16, batch 24650, giga_loss[loss=0.3001, simple_loss=0.3663, pruned_loss=0.117, over 28710.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3767, pruned_loss=0.1225, over 5650936.25 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.361, pruned_loss=0.1115, over 5729809.23 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3794, pruned_loss=0.1241, over 5649064.54 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:45:29,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:45:42,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-08 09:45:58,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:46:01,150 INFO [optim.py:369] (1/2) 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 09:46:07,138 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=708639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:46:16,957 INFO [train.py:968] (1/2) Epoch 16, batch 24700, giga_loss[loss=0.369, simple_loss=0.4158, pruned_loss=0.1611, over 28630.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3777, pruned_loss=0.1238, over 5649545.54 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3616, pruned_loss=0.1121, over 5719288.46 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3797, pruned_loss=0.125, over 5656287.69 frames. ], batch size: 307, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:47:03,102 INFO [train.py:968] (1/2) Epoch 16, batch 24750, giga_loss[loss=0.2688, simple_loss=0.3511, pruned_loss=0.09321, over 28945.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3762, pruned_loss=0.1229, over 5670629.67 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3618, pruned_loss=0.1124, over 5723800.64 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3781, pruned_loss=0.1238, over 5670437.29 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:47:37,042 INFO [optim.py:369] (1/2) 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 09:47:47,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=708744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:47:52,213 INFO [train.py:968] (1/2) Epoch 16, batch 24800, giga_loss[loss=0.415, simple_loss=0.4364, pruned_loss=0.1968, over 26664.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3741, pruned_loss=0.1225, over 5670431.38 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3615, pruned_loss=0.1124, over 5724885.17 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3763, pruned_loss=0.1236, over 5667629.24 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:48:37,424 INFO [train.py:968] (1/2) Epoch 16, batch 24850, giga_loss[loss=0.3008, simple_loss=0.3718, pruned_loss=0.1149, over 28961.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3742, pruned_loss=0.1237, over 5671490.04 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3617, pruned_loss=0.1125, over 5725204.41 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3759, pruned_loss=0.1245, over 5668354.90 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:49:09,545 INFO [optim.py:369] (1/2) 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 09:49:11,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-08 09:49:20,867 INFO [train.py:968] (1/2) Epoch 16, batch 24900, giga_loss[loss=0.2642, simple_loss=0.3495, pruned_loss=0.08944, over 28860.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3729, pruned_loss=0.1215, over 5681205.11 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3614, pruned_loss=0.1123, over 5729033.35 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3748, pruned_loss=0.1226, over 5674159.14 frames. ], batch size: 174, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:49:50,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5983, 1.7982, 1.8324, 1.3902], device='cuda:1'), covar=tensor([0.1958, 0.2458, 0.1585, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.0861, 0.0690, 0.0907, 0.0808], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 09:49:56,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=708887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:49:59,458 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=708890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:50:10,198 INFO [train.py:968] (1/2) Epoch 16, batch 24950, giga_loss[loss=0.3402, simple_loss=0.377, pruned_loss=0.1517, over 23609.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3733, pruned_loss=0.1208, over 5684120.64 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3615, pruned_loss=0.1125, over 5730750.92 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3748, pruned_loss=0.1216, over 5676739.20 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:50:27,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=708919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:50:42,985 INFO [optim.py:369] (1/2) 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 09:50:49,969 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 09:50:57,282 INFO [train.py:968] (1/2) Epoch 16, batch 25000, giga_loss[loss=0.3176, simple_loss=0.3856, pruned_loss=0.1248, over 28624.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3742, pruned_loss=0.122, over 5680890.02 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3615, pruned_loss=0.1126, over 5733643.55 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3757, pruned_loss=0.1228, over 5671370.41 frames. ], batch size: 307, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:51:46,475 INFO [train.py:968] (1/2) Epoch 16, batch 25050, giga_loss[loss=0.2488, simple_loss=0.3298, pruned_loss=0.08394, over 28875.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3723, pruned_loss=0.1207, over 5685697.95 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3617, pruned_loss=0.1127, over 5733117.89 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3735, pruned_loss=0.1214, over 5677787.09 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:52:00,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:52:25,648 INFO [optim.py:369] (1/2) 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 09:52:30,598 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:52:38,869 INFO [train.py:968] (1/2) Epoch 16, batch 25100, libri_loss[loss=0.2528, simple_loss=0.3194, pruned_loss=0.09315, over 29357.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3698, pruned_loss=0.1199, over 5675123.96 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3609, pruned_loss=0.1123, over 5737188.71 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3717, pruned_loss=0.121, over 5663798.80 frames. ], batch size: 71, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:53:22,452 INFO [train.py:968] (1/2) Epoch 16, batch 25150, giga_loss[loss=0.2974, simple_loss=0.3608, pruned_loss=0.117, over 28975.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3693, pruned_loss=0.1201, over 5677668.90 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.361, pruned_loss=0.1126, over 5739075.83 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3712, pruned_loss=0.1211, over 5663911.49 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:53:54,973 INFO [optim.py:369] (1/2) 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 09:54:11,440 INFO [train.py:968] (1/2) Epoch 16, batch 25200, giga_loss[loss=0.3196, simple_loss=0.3788, pruned_loss=0.1303, over 27530.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1221, over 5680000.16 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3614, pruned_loss=0.1127, over 5744740.03 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3725, pruned_loss=0.123, over 5661782.63 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:54:17,715 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:54:18,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5097, 1.6086, 1.7849, 1.3177], device='cuda:1'), covar=tensor([0.1633, 0.2385, 0.1359, 0.1552], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0693, 0.0910, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 09:54:19,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:54:44,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4539, 1.8522, 1.4002, 1.5791], device='cuda:1'), covar=tensor([0.2670, 0.2603, 0.2968, 0.2286], device='cuda:1'), in_proj_covar=tensor([0.1405, 0.1028, 0.1245, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 09:54:49,516 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 25250, giga_loss[loss=0.3115, simple_loss=0.3757, pruned_loss=0.1237, over 28859.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3694, pruned_loss=0.1212, over 5678197.69 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3613, pruned_loss=0.1129, over 5740348.80 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3709, pruned_loss=0.1221, over 5665395.35 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 09:55:03,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2223, 1.2798, 1.1096, 0.9324], device='cuda:1'), covar=tensor([0.0828, 0.0472, 0.1029, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0445, 0.0510, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 09:55:30,966 INFO [optim.py:369] (1/2) 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 09:55:46,725 INFO [train.py:968] (1/2) Epoch 16, batch 25300, giga_loss[loss=0.3268, simple_loss=0.3813, pruned_loss=0.1362, over 28233.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5672219.47 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3623, pruned_loss=0.1137, over 5740252.94 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3705, pruned_loss=0.1224, over 5660018.50 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:56:32,354 INFO [train.py:968] (1/2) Epoch 16, batch 25350, libri_loss[loss=0.3192, simple_loss=0.3805, pruned_loss=0.1289, over 29293.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3692, pruned_loss=0.1214, over 5677363.28 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3624, pruned_loss=0.1138, over 5745305.13 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1218, over 5659428.51 frames. ], batch size: 94, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 09:57:08,627 INFO [optim.py:369] (1/2) 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 09:57:19,989 INFO [train.py:968] (1/2) Epoch 16, batch 25400, giga_loss[loss=0.3061, simple_loss=0.3744, pruned_loss=0.1189, over 28835.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5670128.87 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3624, pruned_loss=0.114, over 5738114.70 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 5660672.36 frames. ], batch size: 243, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:57:38,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 09:58:01,208 INFO [train.py:968] (1/2) Epoch 16, batch 25450, giga_loss[loss=0.2728, simple_loss=0.3522, pruned_loss=0.09669, over 28853.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3696, pruned_loss=0.1205, over 5660829.20 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3623, pruned_loss=0.1143, over 5724015.36 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3706, pruned_loss=0.1208, over 5662682.21 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:58:17,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:58:37,909 INFO [optim.py:369] (1/2) 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 09:58:45,073 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:58:50,965 INFO [train.py:968] (1/2) Epoch 16, batch 25500, giga_loss[loss=0.3161, simple_loss=0.3735, pruned_loss=0.1294, over 27504.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3702, pruned_loss=0.1209, over 5643421.99 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3626, pruned_loss=0.1146, over 5713408.99 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3709, pruned_loss=0.121, over 5652841.57 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:58:51,425 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 09:59:10,393 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 09:59:37,947 INFO [train.py:968] (1/2) Epoch 16, batch 25550, giga_loss[loss=0.3158, simple_loss=0.3772, pruned_loss=0.1272, over 28787.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3714, pruned_loss=0.122, over 5652444.80 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.363, pruned_loss=0.1148, over 5716222.27 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3717, pruned_loss=0.122, over 5656408.36 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 09:59:49,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7431, 1.8209, 1.2629, 1.4286], device='cuda:1'), covar=tensor([0.0806, 0.0571, 0.0998, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0444, 0.0509, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:00:12,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2943, 1.9091, 1.5734, 1.5369], device='cuda:1'), covar=tensor([0.0753, 0.0308, 0.0293, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 10:00:15,007 INFO [optim.py:369] (1/2) 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 10:00:15,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6855, 1.8536, 1.1627, 1.5385], device='cuda:1'), covar=tensor([0.0843, 0.0587, 0.1097, 0.1056], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0444, 0.0509, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:00:26,670 INFO [train.py:968] (1/2) Epoch 16, batch 25600, giga_loss[loss=0.2843, simple_loss=0.3485, pruned_loss=0.1101, over 28669.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3742, pruned_loss=0.125, over 5649678.13 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3628, pruned_loss=0.1145, over 5720714.12 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3748, pruned_loss=0.1254, over 5647528.92 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:00:35,354 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:00:38,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:00:51,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4162, 1.5529, 1.4863, 1.3601], device='cuda:1'), covar=tensor([0.2684, 0.2254, 0.1722, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.1855, 0.1793, 0.1718, 0.1855], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 10:01:09,807 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=709591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:01:19,952 INFO [train.py:968] (1/2) Epoch 16, batch 25650, giga_loss[loss=0.3611, simple_loss=0.3971, pruned_loss=0.1625, over 26543.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3737, pruned_loss=0.1255, over 5659397.85 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3625, pruned_loss=0.1145, over 5722192.23 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3746, pruned_loss=0.126, over 5655364.90 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:02:00,257 INFO [optim.py:369] (1/2) 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 10:02:16,566 INFO [train.py:968] (1/2) Epoch 16, batch 25700, giga_loss[loss=0.4379, simple_loss=0.4652, pruned_loss=0.2053, over 24230.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3753, pruned_loss=0.1278, over 5649241.65 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3622, pruned_loss=0.1144, over 5724361.68 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3765, pruned_loss=0.1286, over 5643050.20 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:02:49,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0147, 1.3259, 1.4299, 1.1013], device='cuda:1'), covar=tensor([0.1452, 0.1114, 0.1892, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0743, 0.0698, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 10:02:57,755 INFO [train.py:968] (1/2) Epoch 16, batch 25750, giga_loss[loss=0.4243, simple_loss=0.4408, pruned_loss=0.2039, over 26442.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3765, pruned_loss=0.1291, over 5645981.13 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3622, pruned_loss=0.1146, over 5710946.18 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3783, pruned_loss=0.1303, over 5648493.41 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:03:01,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4002, 1.8829, 1.4980, 1.6069], device='cuda:1'), covar=tensor([0.0665, 0.0259, 0.0285, 0.0666], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 10:03:31,852 INFO [optim.py:369] (1/2) 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 10:03:32,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3355, 1.6167, 1.3710, 1.5379], device='cuda:1'), covar=tensor([0.0660, 0.0400, 0.0313, 0.0685], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 10:03:46,183 INFO [train.py:968] (1/2) Epoch 16, batch 25800, giga_loss[loss=0.3646, simple_loss=0.4069, pruned_loss=0.1612, over 24157.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3753, pruned_loss=0.1282, over 5651851.73 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3624, pruned_loss=0.1146, over 5715475.38 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3769, pruned_loss=0.1294, over 5648311.20 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:04:30,657 INFO [train.py:968] (1/2) Epoch 16, batch 25850, giga_loss[loss=0.2848, simple_loss=0.3608, pruned_loss=0.1044, over 28833.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3748, pruned_loss=0.1264, over 5666577.49 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.362, pruned_loss=0.1145, over 5719215.29 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3766, pruned_loss=0.1278, over 5659552.35 frames. ], batch size: 186, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:04:47,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:04:53,143 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=709826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:05:06,747 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 25900, giga_loss[loss=0.2905, simple_loss=0.3714, pruned_loss=0.1048, over 28460.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3728, pruned_loss=0.1241, over 5659508.43 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3624, pruned_loss=0.1146, over 5722021.19 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3741, pruned_loss=0.1253, over 5650716.16 frames. ], batch size: 65, lr: 2.00e-03, grad_scale: 1.0
+2023-03-08 10:06:08,529 INFO [train.py:968] (1/2) Epoch 16, batch 25950, giga_loss[loss=0.3014, simple_loss=0.3618, pruned_loss=0.1205, over 28856.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.371, pruned_loss=0.1233, over 5669295.91 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1146, over 5725321.65 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3722, pruned_loss=0.1244, over 5658299.20 frames. ], batch size: 199, lr: 2.00e-03, grad_scale: 1.0
+2023-03-08 10:06:28,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-08 10:06:41,056 INFO [optim.py:369] (1/2) 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 10:06:52,396 INFO [train.py:968] (1/2) Epoch 16, batch 26000, giga_loss[loss=0.2827, simple_loss=0.3521, pruned_loss=0.1066, over 28813.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3691, pruned_loss=0.1223, over 5683007.82 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3629, pruned_loss=0.1149, over 5731948.11 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3698, pruned_loss=0.1231, over 5665955.42 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:07:00,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=709959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:05,283 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:08,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:13,029 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=709969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:16,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=709972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:39,624 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=709995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:07:44,836 INFO [train.py:968] (1/2) Epoch 16, batch 26050, giga_loss[loss=0.2807, simple_loss=0.3575, pruned_loss=0.1019, over 28502.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3692, pruned_loss=0.1226, over 5686554.20 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3627, pruned_loss=0.1149, over 5735250.10 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.37, pruned_loss=0.1234, over 5669181.95 frames. ], batch size: 71, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:07:45,035 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:08:07,850 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-08 10:08:16,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5449, 4.3603, 4.1257, 2.0605], device='cuda:1'), covar=tensor([0.0504, 0.0631, 0.0705, 0.2137], device='cuda:1'), in_proj_covar=tensor([0.1167, 0.1077, 0.0927, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 10:08:20,392 INFO [optim.py:369] (1/2) 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 10:08:22,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-08 10:08:28,780 INFO [train.py:968] (1/2) Epoch 16, batch 26100, giga_loss[loss=0.3633, simple_loss=0.4131, pruned_loss=0.1568, over 27566.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3723, pruned_loss=0.1242, over 5677208.90 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1154, over 5721961.64 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.373, pruned_loss=0.1248, over 5673176.51 frames. ], batch size: 472, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:08:42,208 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:08:57,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7690, 2.1748, 1.9996, 1.6795], device='cuda:1'), covar=tensor([0.2759, 0.2012, 0.1973, 0.2349], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1793, 0.1719, 0.1857], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 10:08:59,362 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710083.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:09:16,186 INFO [train.py:968] (1/2) Epoch 16, batch 26150, giga_loss[loss=0.287, simple_loss=0.3717, pruned_loss=0.1012, over 28704.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3749, pruned_loss=0.1227, over 5686866.32 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1153, over 5727787.44 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3759, pruned_loss=0.1235, over 5677104.47 frames. ], batch size: 92, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:09:20,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3158, 1.5486, 1.5594, 1.5608], device='cuda:1'), covar=tensor([0.0677, 0.0301, 0.0277, 0.0670], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0213], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 10:09:50,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-08 10:09:54,074 INFO [optim.py:369] (1/2) 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 10:10:05,896 INFO [train.py:968] (1/2) Epoch 16, batch 26200, giga_loss[loss=0.4522, simple_loss=0.4672, pruned_loss=0.2186, over 26772.00 frames. ], tot_loss[loss=0.312, simple_loss=0.377, pruned_loss=0.1235, over 5675693.64 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5721636.36 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3784, pruned_loss=0.1244, over 5672777.42 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:10:09,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-08 10:10:10,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2828, 3.8588, 1.4541, 1.5125], device='cuda:1'), covar=tensor([0.0985, 0.0368, 0.0902, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0537, 0.0365, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 10:10:12,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6652, 1.7840, 1.4382, 1.8330], device='cuda:1'), covar=tensor([0.2777, 0.2821, 0.3202, 0.2406], device='cuda:1'), in_proj_covar=tensor([0.1404, 0.1025, 0.1245, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 10:10:53,466 INFO [train.py:968] (1/2) Epoch 16, batch 26250, giga_loss[loss=0.3151, simple_loss=0.3855, pruned_loss=0.1223, over 29030.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3787, pruned_loss=0.1251, over 5684585.92 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3624, pruned_loss=0.1152, over 5725981.04 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3803, pruned_loss=0.126, over 5677003.87 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:11:09,162 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6264, 2.2594, 1.5074, 0.7922], device='cuda:1'), covar=tensor([0.6446, 0.3287, 0.2964, 0.6051], device='cuda:1'), in_proj_covar=tensor([0.1653, 0.1574, 0.1543, 0.1358], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 10:11:25,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6277, 1.7396, 1.2919, 1.2640], device='cuda:1'), covar=tensor([0.0906, 0.0618, 0.1046, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0446, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:11:27,856 INFO [optim.py:369] (1/2) 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 10:11:28,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5860, 1.6360, 1.8221, 1.3739], device='cuda:1'), covar=tensor([0.1800, 0.2458, 0.1394, 0.1683], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0696, 0.0913, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 10:11:36,700 INFO [train.py:968] (1/2) Epoch 16, batch 26300, giga_loss[loss=0.3107, simple_loss=0.3785, pruned_loss=0.1214, over 29028.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3789, pruned_loss=0.1256, over 5682647.23 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3626, pruned_loss=0.1156, over 5718212.29 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3806, pruned_loss=0.1263, over 5682091.60 frames. ], batch size: 136, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:11:44,505 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-08 10:12:25,752 INFO [train.py:968] (1/2) Epoch 16, batch 26350, giga_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1238, over 28662.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3813, pruned_loss=0.1285, over 5675327.70 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3631, pruned_loss=0.116, over 5718689.38 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3824, pruned_loss=0.1289, over 5674022.93 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:12:30,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1274, 1.4126, 1.4241, 1.0737], device='cuda:1'), covar=tensor([0.1348, 0.2099, 0.1085, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0868, 0.0697, 0.0914, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 10:13:01,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=710334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:13:06,625 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 26400, giga_loss[loss=0.2802, simple_loss=0.3454, pruned_loss=0.1074, over 28978.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3791, pruned_loss=0.1274, over 5685784.71 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3631, pruned_loss=0.116, over 5717938.84 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3802, pruned_loss=0.1279, over 5684922.80 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:13:56,142 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 10:14:02,771 INFO [train.py:968] (1/2) Epoch 16, batch 26450, giga_loss[loss=0.2885, simple_loss=0.3581, pruned_loss=0.1094, over 28852.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3772, pruned_loss=0.1269, over 5688584.34 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1156, over 5724806.24 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3788, pruned_loss=0.1279, over 5680823.37 frames. ], batch size: 112, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:14:43,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=710439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:14:44,137 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 26500, giga_loss[loss=0.332, simple_loss=0.3661, pruned_loss=0.1489, over 23403.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3755, pruned_loss=0.1262, over 5674679.20 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3629, pruned_loss=0.1159, over 5709939.35 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3772, pruned_loss=0.1271, over 5681012.60 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:15:00,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=710458.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:15:02,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9094, 1.0617, 1.0341, 0.8704], device='cuda:1'), covar=tensor([0.1527, 0.1884, 0.1208, 0.1550], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1795, 0.1718, 0.1856], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 10:15:19,887 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=710477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:15:22,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=710480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:15:38,623 INFO [train.py:968] (1/2) Epoch 16, batch 26550, giga_loss[loss=0.3351, simple_loss=0.3983, pruned_loss=0.136, over 28648.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3766, pruned_loss=0.1271, over 5672634.30 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3633, pruned_loss=0.1161, over 5711027.58 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3781, pruned_loss=0.128, over 5675253.06 frames. ], batch size: 307, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:15:44,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:15:46,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0937, 2.0012, 1.7025, 1.5673], device='cuda:1'), covar=tensor([0.0855, 0.0720, 0.0934, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0446, 0.0511, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:16:12,968 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 10:16:14,840 INFO [optim.py:369] (1/2) 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 10:16:23,374 INFO [train.py:968] (1/2) Epoch 16, batch 26600, giga_loss[loss=0.2782, simple_loss=0.3468, pruned_loss=0.1048, over 28445.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3763, pruned_loss=0.127, over 5671021.78 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5704897.16 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3775, pruned_loss=0.1279, over 5677177.54 frames. ], batch size: 71, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:16:35,034 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:16:36,482 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 10:16:52,994 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=710582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:16:55,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=710585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:17:10,866 INFO [train.py:968] (1/2) Epoch 16, batch 26650, giga_loss[loss=0.3659, simple_loss=0.4044, pruned_loss=0.1637, over 26681.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3744, pruned_loss=0.1269, over 5654059.76 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3633, pruned_loss=0.1161, over 5706947.03 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3756, pruned_loss=0.1278, over 5656999.01 frames. ], batch size: 555, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:17:12,825 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=710601.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:17:14,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=710604.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:17:24,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:17:45,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=710633.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:17:51,382 INFO [optim.py:369] (1/2) 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 10:17:58,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5121, 2.3964, 1.9068, 1.9461], device='cuda:1'), covar=tensor([0.0808, 0.0676, 0.0937, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0445, 0.0510, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:18:00,753 INFO [train.py:968] (1/2) Epoch 16, batch 26700, giga_loss[loss=0.2962, simple_loss=0.3669, pruned_loss=0.1127, over 28763.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.374, pruned_loss=0.1264, over 5644647.81 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3636, pruned_loss=0.1164, over 5695810.03 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3748, pruned_loss=0.127, over 5655010.70 frames. ], batch size: 285, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:18:21,347 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:18:47,454 INFO [train.py:968] (1/2) Epoch 16, batch 26750, libri_loss[loss=0.3065, simple_loss=0.3797, pruned_loss=0.1167, over 29373.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.125, over 5637955.11 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3639, pruned_loss=0.1167, over 5679108.84 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3745, pruned_loss=0.1254, over 5659904.83 frames. ], batch size: 92, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:18:57,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3330, 1.4802, 1.3521, 1.3018], device='cuda:1'), covar=tensor([0.1900, 0.1874, 0.1566, 0.1780], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1789, 0.1714, 0.1853], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 10:19:29,523 INFO [optim.py:369] (1/2) 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 10:19:39,904 INFO [train.py:968] (1/2) Epoch 16, batch 26800, giga_loss[loss=0.2549, simple_loss=0.3276, pruned_loss=0.09114, over 28631.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3766, pruned_loss=0.1275, over 5632594.39 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3642, pruned_loss=0.1169, over 5673236.92 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3771, pruned_loss=0.1278, over 5653907.30 frames. ], batch size: 60, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:20:11,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2271, 1.8396, 1.4310, 0.3821], device='cuda:1'), covar=tensor([0.4013, 0.2586, 0.4008, 0.5491], device='cuda:1'), in_proj_covar=tensor([0.1656, 0.1580, 0.1547, 0.1364], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 10:20:24,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4401, 1.5496, 1.2092, 1.1564], device='cuda:1'), covar=tensor([0.0912, 0.0536, 0.1070, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0445, 0.0510, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:20:27,995 INFO [train.py:968] (1/2) Epoch 16, batch 26850, giga_loss[loss=0.3231, simple_loss=0.3975, pruned_loss=0.1244, over 28528.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3763, pruned_loss=0.1268, over 5649246.88 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3644, pruned_loss=0.1171, over 5675006.95 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3765, pruned_loss=0.127, over 5663835.21 frames. ], batch size: 71, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:21:04,429 INFO [optim.py:369] (1/2) 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 10:21:13,652 INFO [train.py:968] (1/2) Epoch 16, batch 26900, giga_loss[loss=0.2882, simple_loss=0.3706, pruned_loss=0.1029, over 28936.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3771, pruned_loss=0.1245, over 5660278.52 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3642, pruned_loss=0.1169, over 5679725.18 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3778, pruned_loss=0.1249, over 5667327.96 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:21:27,783 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=710864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:21:49,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2980, 1.1561, 3.6762, 3.0977], device='cuda:1'), covar=tensor([0.1612, 0.2768, 0.0503, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0713, 0.0619, 0.0915, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:22:04,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3326, 1.5317, 1.4610, 1.3601], device='cuda:1'), covar=tensor([0.1395, 0.1459, 0.1724, 0.1506], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0738, 0.0691, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 10:22:05,936 INFO [train.py:968] (1/2) Epoch 16, batch 26950, giga_loss[loss=0.2998, simple_loss=0.3767, pruned_loss=0.1114, over 28187.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3794, pruned_loss=0.1243, over 5672867.45 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3637, pruned_loss=0.1166, over 5683968.63 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3806, pruned_loss=0.1251, over 5674484.06 frames. ], batch size: 77, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:22:38,127 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:22:49,552 INFO [train.py:968] (1/2) Epoch 16, batch 27000, giga_loss[loss=0.3018, simple_loss=0.3665, pruned_loss=0.1186, over 28995.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3808, pruned_loss=0.1252, over 5679142.11 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3634, pruned_loss=0.1164, over 5689552.67 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3824, pruned_loss=0.1262, over 5675440.97 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:22:49,552 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 10:22:57,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2321, 1.6608, 1.5752, 1.0814], device='cuda:1'), covar=tensor([0.2161, 0.3041, 0.1701, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0696, 0.0913, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 10:22:59,223 INFO [train.py:1012] (1/2) Epoch 16, validation: loss=0.2093, simple_loss=0.3155, pruned_loss=0.05157, over 944034.00 frames. 
+2023-03-08 10:22:59,224 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 10:23:45,078 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 10:23:46,448 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 10:23:47,477 INFO [train.py:968] (1/2) Epoch 16, batch 27050, libri_loss[loss=0.2613, simple_loss=0.3275, pruned_loss=0.09748, over 29513.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3828, pruned_loss=0.1279, over 5656014.26 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3635, pruned_loss=0.1166, over 5667103.25 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3845, pruned_loss=0.1289, over 5673257.65 frames. ], batch size: 70, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:23:51,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6091, 3.4298, 3.2871, 2.0396], device='cuda:1'), covar=tensor([0.0688, 0.0839, 0.0764, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.1084, 0.0931, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 10:23:58,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=711010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:24:35,969 INFO [optim.py:369] (1/2) 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 10:24:40,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:24:41,991 INFO [train.py:968] (1/2) Epoch 16, batch 27100, giga_loss[loss=0.3256, simple_loss=0.3793, pruned_loss=0.136, over 29056.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3852, pruned_loss=0.1314, over 5641125.94 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3638, pruned_loss=0.1169, over 5670571.87 frames. ], giga_tot_loss[loss=0.3253, simple_loss=0.3866, pruned_loss=0.132, over 5650882.64 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:25:13,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:25:16,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:25:31,533 INFO [train.py:968] (1/2) Epoch 16, batch 27150, giga_loss[loss=0.3086, simple_loss=0.379, pruned_loss=0.1191, over 28999.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3829, pruned_loss=0.1298, over 5650703.35 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3636, pruned_loss=0.1167, over 5673871.71 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3847, pruned_loss=0.1308, over 5654972.65 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:25:47,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:26:14,819 INFO [optim.py:369] (1/2) 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 10:26:23,686 INFO [train.py:968] (1/2) Epoch 16, batch 27200, giga_loss[loss=0.2602, simple_loss=0.3496, pruned_loss=0.08539, over 28821.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3819, pruned_loss=0.1284, over 5643925.19 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3634, pruned_loss=0.1166, over 5674874.35 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3836, pruned_loss=0.1293, over 5646389.84 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:27:04,210 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:06,191 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5145, 3.3508, 3.1908, 2.0766], device='cuda:1'), covar=tensor([0.0758, 0.0881, 0.0846, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.1167, 0.1078, 0.0926, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 10:27:06,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:13,839 INFO [train.py:968] (1/2) Epoch 16, batch 27250, giga_loss[loss=0.283, simple_loss=0.3665, pruned_loss=0.09975, over 28971.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3821, pruned_loss=0.1266, over 5654291.93 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3633, pruned_loss=0.1166, over 5677425.28 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3836, pruned_loss=0.1275, over 5653807.94 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:27:36,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:54,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:27:55,893 INFO [optim.py:369] (1/2) 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 10:27:59,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7388, 2.5918, 2.6609, 2.3233], device='cuda:1'), covar=tensor([0.1509, 0.1932, 0.1480, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0739, 0.0694, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 10:28:04,034 INFO [train.py:968] (1/2) Epoch 16, batch 27300, giga_loss[loss=0.2842, simple_loss=0.3621, pruned_loss=0.1031, over 28626.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3823, pruned_loss=0.1265, over 5647956.54 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3638, pruned_loss=0.1171, over 5667313.38 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3834, pruned_loss=0.1269, over 5655970.42 frames. ], batch size: 92, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:28:09,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2192, 1.4910, 1.4845, 1.0845], device='cuda:1'), covar=tensor([0.1549, 0.2425, 0.1310, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0699, 0.0916, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 10:28:59,764 INFO [train.py:968] (1/2) Epoch 16, batch 27350, giga_loss[loss=0.2959, simple_loss=0.3664, pruned_loss=0.1127, over 28710.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3822, pruned_loss=0.1265, over 5656819.78 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.364, pruned_loss=0.1172, over 5668529.81 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.383, pruned_loss=0.1268, over 5661923.82 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:29:28,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4810, 1.6981, 1.5471, 1.6000], device='cuda:1'), covar=tensor([0.1512, 0.1855, 0.2055, 0.1709], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0745, 0.0699, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 10:29:36,379 INFO [optim.py:369] (1/2) 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 10:29:47,029 INFO [train.py:968] (1/2) Epoch 16, batch 27400, giga_loss[loss=0.3198, simple_loss=0.3744, pruned_loss=0.1326, over 29020.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.381, pruned_loss=0.1263, over 5669013.62 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3639, pruned_loss=0.1171, over 5671961.98 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3819, pruned_loss=0.1267, over 5669770.37 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:30:12,323 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=711373.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:30:21,478 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:30:25,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:30:25,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:30:37,175 INFO [train.py:968] (1/2) Epoch 16, batch 27450, giga_loss[loss=0.3811, simple_loss=0.4332, pruned_loss=0.1645, over 28682.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3796, pruned_loss=0.1273, over 5645320.94 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3644, pruned_loss=0.1173, over 5667688.12 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3803, pruned_loss=0.1277, over 5649095.64 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:30:53,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:30:55,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9297, 1.8158, 1.3829, 1.4690], device='cuda:1'), covar=tensor([0.0691, 0.0553, 0.0929, 0.1012], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0449, 0.0513, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:31:25,047 INFO [optim.py:369] (1/2) 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 10:31:36,616 INFO [train.py:968] (1/2) Epoch 16, batch 27500, giga_loss[loss=0.2888, simple_loss=0.3604, pruned_loss=0.1086, over 28814.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3783, pruned_loss=0.1273, over 5642528.90 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3643, pruned_loss=0.1173, over 5668882.68 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3789, pruned_loss=0.1277, over 5644209.06 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:32:06,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8996, 1.1498, 1.1044, 0.8122], device='cuda:1'), covar=tensor([0.0947, 0.1635, 0.0846, 0.1055], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0699, 0.0915, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 10:32:23,228 INFO [train.py:968] (1/2) Epoch 16, batch 27550, libri_loss[loss=0.3503, simple_loss=0.4087, pruned_loss=0.146, over 27678.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3762, pruned_loss=0.127, over 5645676.18 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3641, pruned_loss=0.1172, over 5666069.91 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3775, pruned_loss=0.1277, over 5648879.22 frames. ], batch size: 116, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:32:46,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0632, 2.5761, 1.1480, 1.3188], device='cuda:1'), covar=tensor([0.1114, 0.0469, 0.0927, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0538, 0.0365, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 10:32:52,236 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:32:54,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=711531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:32:55,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-08 10:33:03,424 INFO [optim.py:369] (1/2) 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 10:33:11,387 INFO [train.py:968] (1/2) Epoch 16, batch 27600, giga_loss[loss=0.3031, simple_loss=0.3709, pruned_loss=0.1177, over 28952.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3763, pruned_loss=0.1277, over 5643724.98 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3643, pruned_loss=0.1173, over 5668524.20 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3774, pruned_loss=0.1285, over 5643509.91 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:33:20,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:33:46,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 27650, giga_loss[loss=0.2789, simple_loss=0.3541, pruned_loss=0.1019, over 29021.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3736, pruned_loss=0.1252, over 5654226.63 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.364, pruned_loss=0.1169, over 5672894.53 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3751, pruned_loss=0.1265, over 5649402.30 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:34:32,927 INFO [optim.py:369] (1/2) 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 10:34:41,758 INFO [train.py:968] (1/2) Epoch 16, batch 27700, giga_loss[loss=0.2759, simple_loss=0.3476, pruned_loss=0.1022, over 28302.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5649110.04 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.1171, over 5664971.09 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1213, over 5651979.64 frames. ], batch size: 368, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:35:10,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5216, 1.7522, 1.6930, 1.5552], device='cuda:1'), covar=tensor([0.1983, 0.2201, 0.2354, 0.2245], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0735, 0.0691, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 10:35:33,442 INFO [train.py:968] (1/2) Epoch 16, batch 27750, giga_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 28906.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3687, pruned_loss=0.1195, over 5646771.06 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3645, pruned_loss=0.1172, over 5660773.10 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3695, pruned_loss=0.1202, over 5653488.32 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:36:17,206 INFO [optim.py:369] (1/2) 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 10:36:22,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711748.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:36:26,014 INFO [train.py:968] (1/2) Epoch 16, batch 27800, libri_loss[loss=0.3451, simple_loss=0.4004, pruned_loss=0.1449, over 29745.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3668, pruned_loss=0.1187, over 5642880.44 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3645, pruned_loss=0.1172, over 5665629.30 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3674, pruned_loss=0.1192, over 5643317.48 frames. ], batch size: 87, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:36:26,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4004, 1.5841, 1.4553, 1.5363], device='cuda:1'), covar=tensor([0.0781, 0.0325, 0.0309, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 10:36:43,008 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 10:36:55,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1266, 1.0519, 1.1120, 1.3736], device='cuda:1'), covar=tensor([0.0779, 0.0332, 0.0281, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 10:37:21,321 INFO [train.py:968] (1/2) Epoch 16, batch 27850, giga_loss[loss=0.275, simple_loss=0.346, pruned_loss=0.102, over 28583.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3627, pruned_loss=0.1164, over 5661756.79 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.1171, over 5669115.41 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3634, pruned_loss=0.1169, over 5658846.50 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:38:06,583 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 27900, giga_loss[loss=0.2931, simple_loss=0.368, pruned_loss=0.1091, over 28786.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3654, pruned_loss=0.118, over 5667432.30 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.365, pruned_loss=0.1176, over 5672324.52 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3653, pruned_loss=0.118, over 5661904.09 frames. ], batch size: 174, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:38:54,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=711891.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:38:58,543 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 27950, giga_loss[loss=0.2714, simple_loss=0.356, pruned_loss=0.09339, over 28892.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3689, pruned_loss=0.1203, over 5656651.96 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5673058.90 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3686, pruned_loss=0.1201, over 5651708.22 frames. ], batch size: 186, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:39:07,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=711902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:39:25,739 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=711923.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:39:47,085 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 28000, giga_loss[loss=0.2993, simple_loss=0.3689, pruned_loss=0.1148, over 29056.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3708, pruned_loss=0.1215, over 5655224.15 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3663, pruned_loss=0.1186, over 5673411.29 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5650223.61 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:40:07,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=711966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:40:43,237 INFO [train.py:968] (1/2) Epoch 16, batch 28050, giga_loss[loss=0.3161, simple_loss=0.3753, pruned_loss=0.1284, over 28639.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3722, pruned_loss=0.1228, over 5655368.32 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.367, pruned_loss=0.1191, over 5676195.93 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1218, over 5648481.73 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:40:50,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-03-08 10:40:52,257 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:41:20,328 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 28100, giga_loss[loss=0.313, simple_loss=0.3824, pruned_loss=0.1217, over 28558.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1224, over 5663857.29 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3669, pruned_loss=0.1188, over 5680357.79 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.1219, over 5653604.70 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:42:17,638 INFO [train.py:968] (1/2) Epoch 16, batch 28150, giga_loss[loss=0.2998, simple_loss=0.3703, pruned_loss=0.1147, over 28888.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3749, pruned_loss=0.1243, over 5666056.81 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3669, pruned_loss=0.1188, over 5682504.33 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3743, pruned_loss=0.1241, over 5655719.59 frames. ], batch size: 112, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:42:27,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:30,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=712112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:43,962 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:57,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=712141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:42:58,891 INFO [optim.py:369] (1/2) 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 10:43:05,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6433, 1.1926, 2.8740, 2.8395], device='cuda:1'), covar=tensor([0.2216, 0.2632, 0.1090, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0624, 0.0919, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:43:05,659 INFO [train.py:968] (1/2) Epoch 16, batch 28200, giga_loss[loss=0.3688, simple_loss=0.4117, pruned_loss=0.1629, over 28851.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3763, pruned_loss=0.1251, over 5674542.32 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3675, pruned_loss=0.1192, over 5687513.00 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3755, pruned_loss=0.1246, over 5661403.03 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:43:31,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9490, 1.2451, 1.2887, 1.0589], device='cuda:1'), covar=tensor([0.1704, 0.1332, 0.2178, 0.1649], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0741, 0.0697, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 10:43:45,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8872, 2.0852, 1.7225, 2.4127], device='cuda:1'), covar=tensor([0.2443, 0.2561, 0.2758, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.1413, 0.1036, 0.1252, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 10:43:58,259 INFO [train.py:968] (1/2) Epoch 16, batch 28250, giga_loss[loss=0.3609, simple_loss=0.3936, pruned_loss=0.1641, over 23594.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3775, pruned_loss=0.1268, over 5659038.37 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3674, pruned_loss=0.1191, over 5692702.52 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3772, pruned_loss=0.1267, over 5643376.17 frames. ], batch size: 705, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:44:35,662 INFO [optim.py:369] (1/2) 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 10:44:43,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 10:44:43,810 INFO [train.py:968] (1/2) Epoch 16, batch 28300, giga_loss[loss=0.3389, simple_loss=0.4009, pruned_loss=0.1385, over 28906.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1268, over 5659872.96 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3671, pruned_loss=0.119, over 5693118.82 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3776, pruned_loss=0.1272, over 5644568.86 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:44:49,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3390, 3.1588, 3.0637, 1.5045], device='cuda:1'), covar=tensor([0.0883, 0.1058, 0.0949, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.1168, 0.1085, 0.0935, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 10:45:13,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:45:26,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 10:45:37,100 INFO [train.py:968] (1/2) Epoch 16, batch 28350, giga_loss[loss=0.3252, simple_loss=0.384, pruned_loss=0.1332, over 28771.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3777, pruned_loss=0.1251, over 5662902.12 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5689754.89 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3779, pruned_loss=0.1254, over 5652747.81 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:46:22,501 INFO [optim.py:369] (1/2) 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 10:46:28,787 INFO [train.py:968] (1/2) Epoch 16, batch 28400, giga_loss[loss=0.2958, simple_loss=0.3507, pruned_loss=0.1204, over 28889.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3774, pruned_loss=0.125, over 5672481.97 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1196, over 5694703.18 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3777, pruned_loss=0.125, over 5659227.56 frames. ], batch size: 99, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 10:46:46,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2472, 1.5806, 1.5640, 1.1274], device='cuda:1'), covar=tensor([0.1602, 0.2368, 0.1327, 0.1552], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0697, 0.0915, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 10:47:07,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:47:20,651 INFO [train.py:968] (1/2) Epoch 16, batch 28450, giga_loss[loss=0.2901, simple_loss=0.3614, pruned_loss=0.1094, over 29025.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3754, pruned_loss=0.1245, over 5676209.53 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1195, over 5698675.30 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.376, pruned_loss=0.1247, over 5661552.31 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:47:28,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6213, 1.7681, 1.2331, 1.3539], device='cuda:1'), covar=tensor([0.0843, 0.0522, 0.1006, 0.1014], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0446, 0.0510, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:47:45,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:47:46,968 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:48:14,610 INFO [train.py:968] (1/2) Epoch 16, batch 28500, giga_loss[loss=0.2807, simple_loss=0.3457, pruned_loss=0.1078, over 28910.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3753, pruned_loss=0.1248, over 5668577.14 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1195, over 5689486.40 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1251, over 5665154.20 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:48:15,420 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=712452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:48:24,112 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:49:09,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 10:49:16,031 INFO [train.py:968] (1/2) Epoch 16, batch 28550, giga_loss[loss=0.2791, simple_loss=0.34, pruned_loss=0.1091, over 28968.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3724, pruned_loss=0.1231, over 5672160.53 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3676, pruned_loss=0.1194, over 5690601.97 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3731, pruned_loss=0.1235, over 5668433.02 frames. ], batch size: 106, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:49:18,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:49:26,998 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712512.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:49:31,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2591, 2.6204, 1.3110, 1.4203], device='cuda:1'), covar=tensor([0.1008, 0.0391, 0.0882, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0540, 0.0365, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 10:49:40,637 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:49:41,390 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-08 10:49:43,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=712532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:49:55,244 INFO [optim.py:369] (1/2) 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 10:50:01,866 INFO [train.py:968] (1/2) Epoch 16, batch 28600, giga_loss[loss=0.2813, simple_loss=0.3493, pruned_loss=0.1066, over 28708.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1243, over 5675204.64 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1201, over 5692500.03 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3733, pruned_loss=0.1242, over 5669827.21 frames. ], batch size: 262, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:50:11,435 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=712561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:50:52,713 INFO [train.py:968] (1/2) Epoch 16, batch 28650, giga_loss[loss=0.2741, simple_loss=0.346, pruned_loss=0.1011, over 28255.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3727, pruned_loss=0.1243, over 5664589.59 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3685, pruned_loss=0.1202, over 5695748.19 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3727, pruned_loss=0.1242, over 5656898.15 frames. ], batch size: 65, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:51:35,188 INFO [optim.py:369] (1/2) 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 10:51:37,734 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:51:40,720 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 28700, giga_loss[loss=0.2576, simple_loss=0.3338, pruned_loss=0.09068, over 28926.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.373, pruned_loss=0.1248, over 5661356.96 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5697462.96 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5652945.29 frames. ], batch size: 112, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:52:08,979 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 28750, giga_loss[loss=0.3176, simple_loss=0.3795, pruned_loss=0.1278, over 29078.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3749, pruned_loss=0.1268, over 5658893.96 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3683, pruned_loss=0.1202, over 5702102.93 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3753, pruned_loss=0.127, over 5647358.09 frames. ], batch size: 155, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:53:17,880 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 28800, giga_loss[loss=0.3403, simple_loss=0.3906, pruned_loss=0.1449, over 27798.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3763, pruned_loss=0.128, over 5650265.32 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.12, over 5698734.67 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3771, pruned_loss=0.1285, over 5642996.51 frames. ], batch size: 412, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:53:55,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=712780.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:54:13,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-08 10:54:13,861 INFO [train.py:968] (1/2) Epoch 16, batch 28850, giga_loss[loss=0.2814, simple_loss=0.3585, pruned_loss=0.1021, over 28837.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3748, pruned_loss=0.1274, over 5651126.43 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.368, pruned_loss=0.1199, over 5701777.29 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3757, pruned_loss=0.128, over 5641847.62 frames. ], batch size: 174, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:54:38,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9497, 1.0749, 1.0433, 0.8957], device='cuda:1'), covar=tensor([0.1541, 0.1915, 0.1206, 0.1579], device='cuda:1'), in_proj_covar=tensor([0.1864, 0.1797, 0.1727, 0.1865], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 10:54:41,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-08 10:54:44,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:54:58,481 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 28900, giga_loss[loss=0.2877, simple_loss=0.3515, pruned_loss=0.112, over 28797.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3741, pruned_loss=0.1266, over 5657242.81 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.12, over 5702809.40 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3747, pruned_loss=0.1271, over 5648804.29 frames. ], batch size: 66, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 10:55:40,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=712887.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:55:51,271 INFO [train.py:968] (1/2) Epoch 16, batch 28950, giga_loss[loss=0.2774, simple_loss=0.3572, pruned_loss=0.09884, over 28965.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3748, pruned_loss=0.1271, over 5651350.66 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1201, over 5704541.33 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3752, pruned_loss=0.1275, over 5641483.89 frames. ], batch size: 145, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:56:36,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7677, 1.0568, 2.9043, 2.7077], device='cuda:1'), covar=tensor([0.1736, 0.2596, 0.0598, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0624, 0.0920, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:56:41,746 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 29000, giga_loss[loss=0.3299, simple_loss=0.3849, pruned_loss=0.1374, over 28652.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3752, pruned_loss=0.1268, over 5654116.76 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 5706207.77 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3755, pruned_loss=0.1271, over 5644282.52 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:56:44,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9182, 1.1300, 1.1062, 0.8131], device='cuda:1'), covar=tensor([0.1880, 0.2210, 0.1338, 0.1964], device='cuda:1'), in_proj_covar=tensor([0.1869, 0.1804, 0.1733, 0.1869], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 10:57:10,283 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=712975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:57:12,210 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 29050, libri_loss[loss=0.2444, simple_loss=0.3225, pruned_loss=0.08318, over 29569.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3768, pruned_loss=0.1277, over 5656992.54 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3685, pruned_loss=0.1202, over 5708543.73 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1281, over 5646056.31 frames. ], batch size: 75, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:57:35,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4375, 1.2924, 1.2785, 1.5413], device='cuda:1'), covar=tensor([0.0767, 0.0343, 0.0322, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 10:57:38,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=713007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 10:58:01,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=713030.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 10:58:03,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=713033.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:58:16,652 INFO [optim.py:369] (1/2) 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 10:58:19,491 INFO [train.py:968] (1/2) Epoch 16, batch 29100, giga_loss[loss=0.3012, simple_loss=0.3732, pruned_loss=0.1147, over 28490.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3778, pruned_loss=0.128, over 5671400.32 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1203, over 5707835.30 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1284, over 5662801.88 frames. ], batch size: 60, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:58:22,454 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-08 10:58:30,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=713062.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 10:58:32,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0582, 3.4751, 2.2742, 1.1990], device='cuda:1'), covar=tensor([0.5732, 0.2331, 0.3042, 0.5503], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1591, 0.1555, 0.1370], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 10:58:39,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7910, 1.1804, 2.8333, 2.7530], device='cuda:1'), covar=tensor([0.1683, 0.2555, 0.0634, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0626, 0.0925, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:59:07,487 INFO [train.py:968] (1/2) Epoch 16, batch 29150, giga_loss[loss=0.2854, simple_loss=0.3563, pruned_loss=0.1073, over 28766.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3789, pruned_loss=0.1292, over 5664815.94 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5700757.38 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3791, pruned_loss=0.1295, over 5664070.47 frames. ], batch size: 284, lr: 2.00e-03, grad_scale: 2.0
+2023-03-08 10:59:08,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8265, 1.8826, 1.5130, 1.5550], device='cuda:1'), covar=tensor([0.0872, 0.0684, 0.0953, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0445, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 10:59:51,089 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 29200, giga_loss[loss=0.3035, simple_loss=0.3804, pruned_loss=0.1133, over 29032.00 frames. ], tot_loss[loss=0.319, simple_loss=0.38, pruned_loss=0.129, over 5666513.76 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3694, pruned_loss=0.1208, over 5700217.42 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3799, pruned_loss=0.1292, over 5665442.26 frames. ], batch size: 128, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:00:01,966 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 29250, giga_loss[loss=0.343, simple_loss=0.3989, pruned_loss=0.1436, over 28981.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3788, pruned_loss=0.1275, over 5650829.81 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3691, pruned_loss=0.1205, over 5698721.19 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3791, pruned_loss=0.1279, over 5650877.57 frames. ], batch size: 164, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:01:30,202 INFO [optim.py:369] (1/2) 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 11:01:33,860 INFO [train.py:968] (1/2) Epoch 16, batch 29300, giga_loss[loss=0.2585, simple_loss=0.329, pruned_loss=0.09399, over 28426.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3768, pruned_loss=0.1256, over 5668210.43 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5702099.45 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3771, pruned_loss=0.1258, over 5664290.86 frames. ], batch size: 65, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:01:45,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-08 11:01:50,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:02:20,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=713298.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:02:21,813 INFO [train.py:968] (1/2) Epoch 16, batch 29350, libri_loss[loss=0.2797, simple_loss=0.3359, pruned_loss=0.1118, over 29501.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3761, pruned_loss=0.1261, over 5657731.05 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3685, pruned_loss=0.1204, over 5706144.08 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1268, over 5649976.31 frames. ], batch size: 70, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:02:22,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=713301.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:02:47,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=713330.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:03:07,595 INFO [optim.py:369] (1/2) 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 11:03:11,397 INFO [train.py:968] (1/2) Epoch 16, batch 29400, giga_loss[loss=0.3279, simple_loss=0.3914, pruned_loss=0.1322, over 28930.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3769, pruned_loss=0.126, over 5668309.72 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3687, pruned_loss=0.1203, over 5706325.99 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3777, pruned_loss=0.1267, over 5661009.49 frames. ], batch size: 213, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:03:32,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5279, 1.7431, 1.4222, 1.5293], device='cuda:1'), covar=tensor([0.2422, 0.2506, 0.2774, 0.2293], device='cuda:1'), in_proj_covar=tensor([0.1414, 0.1037, 0.1251, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 11:03:41,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0341, 5.8543, 5.5480, 3.4326], device='cuda:1'), covar=tensor([0.0390, 0.0537, 0.0669, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.1091, 0.0940, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 11:04:05,313 INFO [train.py:968] (1/2) Epoch 16, batch 29450, giga_loss[loss=0.3114, simple_loss=0.3724, pruned_loss=0.1252, over 28949.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.378, pruned_loss=0.1271, over 5660411.16 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3685, pruned_loss=0.1202, over 5707373.00 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.379, pruned_loss=0.1278, over 5653064.91 frames. ], batch size: 213, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:04:07,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-08 11:04:47,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2899, 1.6835, 1.1967, 0.7991], device='cuda:1'), covar=tensor([0.3716, 0.2360, 0.2446, 0.4341], device='cuda:1'), in_proj_covar=tensor([0.1665, 0.1585, 0.1554, 0.1362], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 11:04:52,980 INFO [optim.py:369] (1/2) 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 11:04:55,527 INFO [train.py:968] (1/2) Epoch 16, batch 29500, giga_loss[loss=0.3224, simple_loss=0.3764, pruned_loss=0.1342, over 28527.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3767, pruned_loss=0.1273, over 5663393.40 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3682, pruned_loss=0.12, over 5709171.90 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3778, pruned_loss=0.1281, over 5655513.58 frames. ], batch size: 336, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:05:25,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3265, 2.8269, 1.5642, 1.3965], device='cuda:1'), covar=tensor([0.0880, 0.0357, 0.0793, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0541, 0.0365, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 11:05:42,786 INFO [train.py:968] (1/2) Epoch 16, batch 29550, giga_loss[loss=0.3814, simple_loss=0.4298, pruned_loss=0.1665, over 28707.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3784, pruned_loss=0.129, over 5657181.43 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3685, pruned_loss=0.12, over 5713641.06 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3793, pruned_loss=0.1299, over 5645745.35 frames. ], batch size: 242, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:06:26,598 INFO [optim.py:369] (1/2) 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 11:06:29,463 INFO [train.py:968] (1/2) Epoch 16, batch 29600, giga_loss[loss=0.3162, simple_loss=0.383, pruned_loss=0.1247, over 28380.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3789, pruned_loss=0.1288, over 5669501.70 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3685, pruned_loss=0.1199, over 5718279.65 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3798, pruned_loss=0.1298, over 5655134.44 frames. ], batch size: 369, lr: 2.00e-03, grad_scale: 8.0
+2023-03-08 11:07:21,782 INFO [train.py:968] (1/2) Epoch 16, batch 29650, giga_loss[loss=0.2972, simple_loss=0.3659, pruned_loss=0.1143, over 28964.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3796, pruned_loss=0.1296, over 5649555.95 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3686, pruned_loss=0.12, over 5710507.48 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3804, pruned_loss=0.1304, over 5643921.24 frames. ], batch size: 227, lr: 2.00e-03, grad_scale: 4.0
+2023-03-08 11:07:55,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=713639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:08:03,475 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 29700, giga_loss[loss=0.4014, simple_loss=0.4288, pruned_loss=0.187, over 27643.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3774, pruned_loss=0.127, over 5659921.99 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3686, pruned_loss=0.1199, over 5701635.59 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3782, pruned_loss=0.1279, over 5662254.23 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:08:16,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0336, 1.3648, 1.1138, 0.2607], device='cuda:1'), covar=tensor([0.3251, 0.2735, 0.4214, 0.5237], device='cuda:1'), in_proj_covar=tensor([0.1662, 0.1583, 0.1553, 0.1361], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 11:08:20,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3178, 1.5540, 1.6046, 1.1758], device='cuda:1'), covar=tensor([0.1644, 0.2509, 0.1371, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0697, 0.0913, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 11:08:55,610 INFO [train.py:968] (1/2) Epoch 16, batch 29750, giga_loss[loss=0.2818, simple_loss=0.3544, pruned_loss=0.1046, over 28921.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3788, pruned_loss=0.1278, over 5662015.73 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1202, over 5705471.19 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3795, pruned_loss=0.1285, over 5659264.37 frames. ], batch size: 164, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:09:02,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5387, 1.7198, 1.1235, 1.3372], device='cuda:1'), covar=tensor([0.0897, 0.0589, 0.1131, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0444, 0.0507, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 11:09:23,854 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713727.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:09:31,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1610, 3.9940, 3.7889, 1.9366], device='cuda:1'), covar=tensor([0.0592, 0.0679, 0.0712, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.1168, 0.1084, 0.0936, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 11:09:39,855 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 29800, giga_loss[loss=0.3195, simple_loss=0.3868, pruned_loss=0.1261, over 28915.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3794, pruned_loss=0.1281, over 5668600.19 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3693, pruned_loss=0.1205, over 5709899.52 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3799, pruned_loss=0.1286, over 5660631.49 frames. ], batch size: 136, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:10:15,080 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=713782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:10:17,332 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=713785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:10:30,938 INFO [train.py:968] (1/2) Epoch 16, batch 29850, giga_loss[loss=0.2423, simple_loss=0.3176, pruned_loss=0.08347, over 28631.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3769, pruned_loss=0.1264, over 5667022.30 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3695, pruned_loss=0.1206, over 5711057.81 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3773, pruned_loss=0.1269, over 5658900.55 frames. ], batch size: 71, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:10:43,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:11:19,730 INFO [train.py:968] (1/2) Epoch 16, batch 29900, giga_loss[loss=0.3668, simple_loss=0.4131, pruned_loss=0.1602, over 27935.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1268, over 5647961.79 frames. ], libri_tot_loss[loss=0.3061, simple_loss=0.37, pruned_loss=0.1211, over 5695629.89 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3766, pruned_loss=0.1269, over 5654214.33 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:11:22,019 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:12:06,073 INFO [train.py:968] (1/2) Epoch 16, batch 29950, giga_loss[loss=0.352, simple_loss=0.392, pruned_loss=0.156, over 26455.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3736, pruned_loss=0.125, over 5650955.03 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3699, pruned_loss=0.1209, over 5697744.64 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3738, pruned_loss=0.1253, over 5653305.25 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:12:11,115 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:12:54,395 INFO [optim.py:369] (1/2) 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 11:12:57,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4275, 1.7904, 1.3863, 1.4076], device='cuda:1'), covar=tensor([0.2339, 0.2284, 0.2625, 0.2036], device='cuda:1'), in_proj_covar=tensor([0.1414, 0.1036, 0.1252, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 11:12:57,349 INFO [train.py:968] (1/2) Epoch 16, batch 30000, giga_loss[loss=0.309, simple_loss=0.3707, pruned_loss=0.1236, over 28744.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3698, pruned_loss=0.1232, over 5668848.32 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3698, pruned_loss=0.1209, over 5698990.69 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.37, pruned_loss=0.1235, over 5669290.84 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:12:57,349 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 11:13:05,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3665, 1.7802, 1.3480, 1.3247], device='cuda:1'), covar=tensor([0.2789, 0.2605, 0.2963, 0.2391], device='cuda:1'), in_proj_covar=tensor([0.1414, 0.1036, 0.1252, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 11:13:05,725 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 11:13:25,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=713976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:13:28,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2813, 1.4894, 1.5425, 1.2853], device='cuda:1'), covar=tensor([0.1738, 0.1744, 0.2262, 0.1926], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0738, 0.0695, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 11:13:50,823 INFO [train.py:968] (1/2) Epoch 16, batch 30050, giga_loss[loss=0.2965, simple_loss=0.36, pruned_loss=0.1165, over 28028.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3686, pruned_loss=0.1232, over 5683165.33 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3698, pruned_loss=0.1209, over 5702027.59 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3688, pruned_loss=0.1234, over 5680573.28 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:14:39,820 INFO [optim.py:369] (1/2) 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 11:14:44,380 INFO [train.py:968] (1/2) Epoch 16, batch 30100, giga_loss[loss=0.2659, simple_loss=0.3387, pruned_loss=0.09654, over 29105.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3672, pruned_loss=0.1217, over 5684871.48 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3699, pruned_loss=0.121, over 5705087.10 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3672, pruned_loss=0.1218, over 5679821.73 frames. ], batch size: 128, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:14:50,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2093, 1.2071, 3.4242, 2.9875], device='cuda:1'), covar=tensor([0.1636, 0.2851, 0.0480, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0627, 0.0924, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 11:15:33,757 INFO [train.py:968] (1/2) Epoch 16, batch 30150, giga_loss[loss=0.2957, simple_loss=0.3681, pruned_loss=0.1116, over 28869.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3663, pruned_loss=0.1193, over 5678411.98 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.1209, over 5700385.09 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3665, pruned_loss=0.1195, over 5677701.72 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:15:35,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714102.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:16:25,378 INFO [optim.py:369] (1/2) 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 11:16:28,515 INFO [train.py:968] (1/2) Epoch 16, batch 30200, giga_loss[loss=0.2432, simple_loss=0.3286, pruned_loss=0.0789, over 29071.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1163, over 5666901.80 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5700434.70 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1162, over 5665666.18 frames. ], batch size: 128, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:16:52,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3639, 1.1828, 3.5263, 3.2106], device='cuda:1'), covar=tensor([0.1527, 0.2821, 0.0480, 0.1556], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0627, 0.0923, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 11:17:19,086 INFO [train.py:968] (1/2) Epoch 16, batch 30250, giga_loss[loss=0.2482, simple_loss=0.3077, pruned_loss=0.09431, over 24438.00 frames. ], tot_loss[loss=0.292, simple_loss=0.36, pruned_loss=0.1121, over 5661732.03 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5702691.33 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3599, pruned_loss=0.1119, over 5658078.25 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:17:48,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0658, 1.3755, 1.1055, 0.2671], device='cuda:1'), covar=tensor([0.2976, 0.2796, 0.4240, 0.5314], device='cuda:1'), in_proj_covar=tensor([0.1660, 0.1575, 0.1551, 0.1362], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 11:17:50,639 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:18:06,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:18:09,665 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:18:11,464 INFO [train.py:968] (1/2) Epoch 16, batch 30300, giga_loss[loss=0.2388, simple_loss=0.3216, pruned_loss=0.07799, over 28893.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3559, pruned_loss=0.1082, over 5661160.71 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.369, pruned_loss=0.1207, over 5705907.22 frames. ], giga_tot_loss[loss=0.2863, simple_loss=0.3562, pruned_loss=0.1082, over 5654736.14 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:18:38,588 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714277.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:18:38,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3399, 3.4326, 1.5027, 1.5002], device='cuda:1'), covar=tensor([0.1000, 0.0306, 0.0987, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0535, 0.0364, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 11:18:43,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714282.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:19:01,751 INFO [train.py:968] (1/2) Epoch 16, batch 30350, giga_loss[loss=0.2699, simple_loss=0.3462, pruned_loss=0.09682, over 27714.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.354, pruned_loss=0.1053, over 5664195.98 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3687, pruned_loss=0.1206, over 5710266.62 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3543, pruned_loss=0.1052, over 5654401.41 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:19:54,218 INFO [optim.py:369] (1/2) 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 11:19:55,786 INFO [train.py:968] (1/2) Epoch 16, batch 30400, giga_loss[loss=0.2471, simple_loss=0.3341, pruned_loss=0.08009, over 28867.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3524, pruned_loss=0.1035, over 5647840.26 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3681, pruned_loss=0.1205, over 5712636.05 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3527, pruned_loss=0.103, over 5635658.63 frames. ], batch size: 145, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:19:56,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:12,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3317, 1.1502, 4.1269, 3.2683], device='cuda:1'), covar=tensor([0.1676, 0.2918, 0.0420, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0625, 0.0920, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 11:20:16,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:18,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:19,696 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=714376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:20:33,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 11:20:44,984 INFO [train.py:968] (1/2) Epoch 16, batch 30450, giga_loss[loss=0.2967, simple_loss=0.3659, pruned_loss=0.1138, over 28301.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3541, pruned_loss=0.105, over 5650547.54 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3676, pruned_loss=0.1204, over 5713032.99 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.1041, over 5637355.25 frames. ], batch size: 369, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:20:47,756 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:21:08,753 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714425.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:21:11,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714428.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 11:21:33,197 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 30500, giga_loss[loss=0.2399, simple_loss=0.3063, pruned_loss=0.08675, over 24120.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3516, pruned_loss=0.1033, over 5643123.71 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3675, pruned_loss=0.1206, over 5713120.66 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3516, pruned_loss=0.1021, over 5631307.03 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:21:43,103 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714457.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 11:21:55,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4987, 1.8876, 1.7070, 1.5913], device='cuda:1'), covar=tensor([0.1662, 0.2014, 0.1758, 0.1960], device='cuda:1'), in_proj_covar=tensor([0.0448, 0.0729, 0.0687, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 11:22:19,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:22:21,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:22:25,574 INFO [train.py:968] (1/2) Epoch 16, batch 30550, giga_loss[loss=0.2421, simple_loss=0.3256, pruned_loss=0.07929, over 28470.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3489, pruned_loss=0.1012, over 5649693.81 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.367, pruned_loss=0.1203, over 5715916.77 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3491, pruned_loss=0.1002, over 5636911.31 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:22:54,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=714526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:22:58,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-08 11:23:18,151 INFO [optim.py:369] (1/2) 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 11:23:18,911 INFO [train.py:968] (1/2) Epoch 16, batch 30600, giga_loss[loss=0.2311, simple_loss=0.2989, pruned_loss=0.08163, over 24063.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3473, pruned_loss=0.1, over 5636865.47 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3668, pruned_loss=0.1204, over 5708829.65 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3474, pruned_loss=0.09898, over 5631875.11 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:23:47,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4309, 1.6747, 1.6878, 1.2530], device='cuda:1'), covar=tensor([0.1862, 0.2667, 0.1526, 0.1867], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0690, 0.0909, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 11:24:09,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=714600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:24:10,169 INFO [train.py:968] (1/2) Epoch 16, batch 30650, libri_loss[loss=0.2661, simple_loss=0.3359, pruned_loss=0.09814, over 29513.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3469, pruned_loss=0.09904, over 5637251.21 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3664, pruned_loss=0.1202, over 5699154.90 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.347, pruned_loss=0.09805, over 5641202.38 frames. ], batch size: 82, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:25:04,909 INFO [optim.py:369] (1/2) 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 11:25:06,714 INFO [train.py:968] (1/2) Epoch 16, batch 30700, giga_loss[loss=0.2858, simple_loss=0.3651, pruned_loss=0.1032, over 28292.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3447, pruned_loss=0.09695, over 5639618.85 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3663, pruned_loss=0.1202, over 5699992.12 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3448, pruned_loss=0.09608, over 5641709.48 frames. ], batch size: 368, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:25:10,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9301, 1.4576, 1.4405, 1.1942], device='cuda:1'), covar=tensor([0.2178, 0.1435, 0.1986, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0727, 0.0686, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 11:25:56,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5554, 1.7530, 1.3083, 1.3391], device='cuda:1'), covar=tensor([0.0821, 0.0412, 0.0853, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0438, 0.0504, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 11:26:01,884 INFO [train.py:968] (1/2) Epoch 16, batch 30750, giga_loss[loss=0.2509, simple_loss=0.3296, pruned_loss=0.08612, over 28524.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3416, pruned_loss=0.09487, over 5636748.52 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3663, pruned_loss=0.1203, over 5702265.18 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3413, pruned_loss=0.09368, over 5635385.95 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:26:50,449 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 30800, giga_loss[loss=0.3093, simple_loss=0.3679, pruned_loss=0.1254, over 27688.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3388, pruned_loss=0.09412, over 5644710.74 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3656, pruned_loss=0.1201, over 5708033.74 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3385, pruned_loss=0.0926, over 5636870.24 frames. ], batch size: 474, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:26:52,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:27:43,327 INFO [train.py:968] (1/2) Epoch 16, batch 30850, libri_loss[loss=0.2475, simple_loss=0.3102, pruned_loss=0.09241, over 29497.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.337, pruned_loss=0.0937, over 5653293.01 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3645, pruned_loss=0.1197, over 5713814.29 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.337, pruned_loss=0.09208, over 5639847.37 frames. ], batch size: 70, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:28:35,036 INFO [optim.py:369] (1/2) 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 11:28:35,051 INFO [train.py:968] (1/2) Epoch 16, batch 30900, giga_loss[loss=0.2561, simple_loss=0.3391, pruned_loss=0.08657, over 28678.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3365, pruned_loss=0.09426, over 5642485.35 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3632, pruned_loss=0.1192, over 5718459.06 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3364, pruned_loss=0.09235, over 5624229.08 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:28:39,046 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=714855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:29:16,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3475, 3.3658, 1.5389, 1.4419], device='cuda:1'), covar=tensor([0.0958, 0.0359, 0.0967, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0534, 0.0364, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 11:29:24,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=714894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:29:26,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=714897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:29:30,299 INFO [train.py:968] (1/2) Epoch 16, batch 30950, giga_loss[loss=0.2366, simple_loss=0.3295, pruned_loss=0.07185, over 28897.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.34, pruned_loss=0.09576, over 5645265.97 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3634, pruned_loss=0.1194, over 5720622.62 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.339, pruned_loss=0.09329, over 5626092.69 frames. ], batch size: 112, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:29:57,922 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:30:30,928 INFO [train.py:968] (1/2) Epoch 16, batch 31000, giga_loss[loss=0.2453, simple_loss=0.3289, pruned_loss=0.08087, over 28537.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3422, pruned_loss=0.09588, over 5654773.66 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3633, pruned_loss=0.1194, over 5723796.36 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3412, pruned_loss=0.09346, over 5635535.82 frames. ], batch size: 92, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:31:02,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=714975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:31:08,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4052, 1.1382, 4.1990, 3.3999], device='cuda:1'), covar=tensor([0.1597, 0.2886, 0.0416, 0.0755], device='cuda:1'), in_proj_covar=tensor([0.0710, 0.0619, 0.0907, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 11:31:21,994 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-08 11:31:41,128 INFO [train.py:968] (1/2) Epoch 16, batch 31050, giga_loss[loss=0.2349, simple_loss=0.3172, pruned_loss=0.07626, over 28467.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3429, pruned_loss=0.09574, over 5664598.00 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3633, pruned_loss=0.1195, over 5715399.18 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.342, pruned_loss=0.09363, over 5655421.69 frames. ], batch size: 78, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:32:20,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3215, 4.1280, 3.9320, 1.8960], device='cuda:1'), covar=tensor([0.0696, 0.0958, 0.1034, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.1137, 0.1055, 0.0905, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 11:32:26,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:32:31,116 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:32:33,756 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 31100, giga_loss[loss=0.3268, simple_loss=0.3865, pruned_loss=0.1336, over 27628.00 frames. ], tot_loss[loss=0.264, simple_loss=0.34, pruned_loss=0.09404, over 5650440.26 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3633, pruned_loss=0.1196, over 5708031.11 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.339, pruned_loss=0.09196, over 5649736.64 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:33:56,054 INFO [train.py:968] (1/2) Epoch 16, batch 31150, giga_loss[loss=0.1964, simple_loss=0.2739, pruned_loss=0.05944, over 24262.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3388, pruned_loss=0.09231, over 5642174.66 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3635, pruned_loss=0.1199, over 5697352.99 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09014, over 5649514.32 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:33:56,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:34:05,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2125, 4.0550, 3.8087, 2.1450], device='cuda:1'), covar=tensor([0.0543, 0.0697, 0.0743, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.1135, 0.1052, 0.0903, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 11:34:17,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:34:20,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6140, 2.0906, 1.9295, 1.7038], device='cuda:1'), covar=tensor([0.1856, 0.1922, 0.1925, 0.2004], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0729, 0.0690, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 11:34:21,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7143, 1.8420, 1.2847, 1.4568], device='cuda:1'), covar=tensor([0.0878, 0.0578, 0.1027, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0439, 0.0506, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 11:34:21,896 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:34:59,422 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 31200, giga_loss[loss=0.2294, simple_loss=0.3119, pruned_loss=0.07344, over 28167.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3352, pruned_loss=0.08976, over 5651857.75 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3631, pruned_loss=0.1197, over 5695738.47 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3342, pruned_loss=0.08782, over 5658516.74 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:35:04,898 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:35:24,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4280, 1.3542, 3.7581, 3.1665], device='cuda:1'), covar=tensor([0.1490, 0.2712, 0.0435, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0707, 0.0616, 0.0904, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 11:36:06,089 INFO [train.py:968] (1/2) Epoch 16, batch 31250, giga_loss[loss=0.2303, simple_loss=0.317, pruned_loss=0.07178, over 28862.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3336, pruned_loss=0.09015, over 5652884.83 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3627, pruned_loss=0.1196, over 5700657.49 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3326, pruned_loss=0.08799, over 5652493.06 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:36:39,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:37:04,042 INFO [train.py:968] (1/2) Epoch 16, batch 31300, giga_loss[loss=0.2449, simple_loss=0.3249, pruned_loss=0.08243, over 28519.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3334, pruned_loss=0.09084, over 5664510.56 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.362, pruned_loss=0.1194, over 5706647.84 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3324, pruned_loss=0.08847, over 5657584.52 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:37:04,905 INFO [optim.py:369] (1/2) 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 11:37:37,245 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:37:43,155 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 31350, giga_loss[loss=0.2833, simple_loss=0.3586, pruned_loss=0.104, over 28939.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3342, pruned_loss=0.09058, over 5668267.41 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3613, pruned_loss=0.119, over 5709539.50 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3334, pruned_loss=0.0884, over 5659279.77 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:38:55,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4554, 1.7548, 1.4575, 1.3642], device='cuda:1'), covar=tensor([0.2258, 0.2070, 0.2210, 0.1892], device='cuda:1'), in_proj_covar=tensor([0.1417, 0.1031, 0.1257, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 11:39:05,831 INFO [train.py:968] (1/2) Epoch 16, batch 31400, giga_loss[loss=0.2517, simple_loss=0.3335, pruned_loss=0.08499, over 28754.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3369, pruned_loss=0.09121, over 5661108.27 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3618, pruned_loss=0.1196, over 5704144.33 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3352, pruned_loss=0.08837, over 5657621.23 frames. ], batch size: 243, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:39:07,989 INFO [optim.py:369] (1/2) 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 11:39:21,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 11:39:36,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:39:40,659 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 31450, giga_loss[loss=0.21, simple_loss=0.2957, pruned_loss=0.0621, over 28730.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.334, pruned_loss=0.0891, over 5669720.03 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3617, pruned_loss=0.1196, over 5707090.90 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3325, pruned_loss=0.08646, over 5663913.48 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:40:25,215 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:28,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:33,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:40:35,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:41:27,558 INFO [train.py:968] (1/2) Epoch 16, batch 31500, giga_loss[loss=0.2659, simple_loss=0.3427, pruned_loss=0.09454, over 29020.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3356, pruned_loss=0.09017, over 5674934.73 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3614, pruned_loss=0.1195, over 5710188.28 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3342, pruned_loss=0.08782, over 5667033.76 frames. ], batch size: 120, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:41:29,692 INFO [optim.py:369] (1/2) 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 11:41:49,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1512, 1.1301, 3.6033, 3.1243], device='cuda:1'), covar=tensor([0.1708, 0.2928, 0.0405, 0.1372], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0618, 0.0904, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 11:42:01,934 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 31550, giga_loss[loss=0.2994, simple_loss=0.3868, pruned_loss=0.106, over 28640.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3391, pruned_loss=0.09121, over 5668511.70 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3613, pruned_loss=0.1195, over 5709580.21 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3379, pruned_loss=0.08891, over 5662270.60 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:42:36,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:12,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:41,518 INFO [train.py:968] (1/2) Epoch 16, batch 31600, giga_loss[loss=0.2814, simple_loss=0.363, pruned_loss=0.09987, over 28612.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3418, pruned_loss=0.0902, over 5664826.18 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3609, pruned_loss=0.1193, over 5710740.52 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3411, pruned_loss=0.08837, over 5658566.09 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:43:42,074 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:42,356 INFO [optim.py:369] (1/2) 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 11:43:46,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:47,583 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:51,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:53,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:43:55,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-08 11:43:56,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:44:25,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:44:29,508 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:44:31,321 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 31650, giga_loss[loss=0.3601, simple_loss=0.3959, pruned_loss=0.1621, over 26940.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3427, pruned_loss=0.09018, over 5657250.18 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3606, pruned_loss=0.119, over 5713970.22 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3421, pruned_loss=0.08838, over 5648224.82 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:45:06,096 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:09,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:32,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:43,240 INFO [train.py:968] (1/2) Epoch 16, batch 31700, libri_loss[loss=0.2223, simple_loss=0.3021, pruned_loss=0.07125, over 29558.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3421, pruned_loss=0.08985, over 5659814.29 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.36, pruned_loss=0.1188, over 5714085.17 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3415, pruned_loss=0.0876, over 5649973.27 frames. ], batch size: 78, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:45:43,718 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:44,012 INFO [optim.py:369] (1/2) 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 11:45:46,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:45:53,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:46:11,258 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:46:14,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:46:50,939 INFO [train.py:968] (1/2) Epoch 16, batch 31750, giga_loss[loss=0.2475, simple_loss=0.3272, pruned_loss=0.08387, over 28916.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3422, pruned_loss=0.09025, over 5659165.31 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.36, pruned_loss=0.1188, over 5713190.74 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08836, over 5652002.56 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:46:58,111 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:48:06,473 INFO [train.py:968] (1/2) Epoch 16, batch 31800, giga_loss[loss=0.2731, simple_loss=0.3546, pruned_loss=0.09587, over 28747.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3413, pruned_loss=0.09105, over 5667066.79 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3597, pruned_loss=0.1187, over 5718074.09 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3406, pruned_loss=0.08879, over 5655302.93 frames. ], batch size: 119, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:48:09,879 INFO [optim.py:369] (1/2) 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 11:49:29,202 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:49:33,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:49:34,188 INFO [train.py:968] (1/2) Epoch 16, batch 31850, giga_loss[loss=0.2424, simple_loss=0.3197, pruned_loss=0.08255, over 28390.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3418, pruned_loss=0.09166, over 5674621.15 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3594, pruned_loss=0.1185, over 5718883.71 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3414, pruned_loss=0.08991, over 5664500.18 frames. ], batch size: 369, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:49:36,487 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=715802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:49:41,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=715805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:50:18,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:50:26,974 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=715834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:50:48,996 INFO [train.py:968] (1/2) Epoch 16, batch 31900, giga_loss[loss=0.2669, simple_loss=0.3371, pruned_loss=0.09837, over 26849.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3367, pruned_loss=0.08871, over 5666631.33 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3594, pruned_loss=0.1186, over 5711716.87 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3361, pruned_loss=0.08687, over 5665288.08 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:50:51,700 INFO [optim.py:369] (1/2) 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 11:51:06,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=715867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:51:17,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=715878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:51:45,763 INFO [train.py:968] (1/2) Epoch 16, batch 31950, giga_loss[loss=0.2351, simple_loss=0.3163, pruned_loss=0.07698, over 27727.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3358, pruned_loss=0.08917, over 5657609.70 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3594, pruned_loss=0.119, over 5700701.65 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3343, pruned_loss=0.08596, over 5664548.39 frames. ], batch size: 474, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:52:54,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9947, 1.1730, 3.2662, 2.8744], device='cuda:1'), covar=tensor([0.1663, 0.2715, 0.0513, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0616, 0.0903, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 11:52:56,181 INFO [train.py:968] (1/2) Epoch 16, batch 32000, giga_loss[loss=0.3123, simple_loss=0.3807, pruned_loss=0.1219, over 28683.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3341, pruned_loss=0.0885, over 5656923.52 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3589, pruned_loss=0.1187, over 5700932.85 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3331, pruned_loss=0.0859, over 5661749.28 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:52:57,933 INFO [optim.py:369] (1/2) 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 11:54:00,960 INFO [train.py:968] (1/2) Epoch 16, batch 32050, giga_loss[loss=0.2881, simple_loss=0.3635, pruned_loss=0.1063, over 28774.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3382, pruned_loss=0.0903, over 5654837.60 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.359, pruned_loss=0.1188, over 5690863.57 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.337, pruned_loss=0.08786, over 5666893.53 frames. ], batch size: 243, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 11:54:20,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:54:26,475 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:54:29,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 32100, giga_loss[loss=0.3191, simple_loss=0.3715, pruned_loss=0.1334, over 27723.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3385, pruned_loss=0.09134, over 5654514.66 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3591, pruned_loss=0.1189, over 5689387.46 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3374, pruned_loss=0.08909, over 5664942.49 frames. ], batch size: 474, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:55:13,910 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 16, batch 32150, giga_loss[loss=0.2452, simple_loss=0.3269, pruned_loss=0.08181, over 28915.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.338, pruned_loss=0.09258, over 5653312.73 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3587, pruned_loss=0.1189, over 5685210.25 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3369, pruned_loss=0.09001, over 5664330.80 frames. ], batch size: 164, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:56:14,320 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=716103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:57:15,044 INFO [train.py:968] (1/2) Epoch 16, batch 32200, giga_loss[loss=0.2714, simple_loss=0.3496, pruned_loss=0.09659, over 28672.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09235, over 5648034.49 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3587, pruned_loss=0.1189, over 5676091.23 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.09003, over 5663584.05 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:57:21,102 INFO [optim.py:369] (1/2) 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 11:57:28,630 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:57:31,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:57:34,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3195, 1.6792, 1.2821, 1.2643], device='cuda:1'), covar=tensor([0.2628, 0.2484, 0.2975, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1027, 0.1254, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 11:57:46,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3411, 1.4957, 1.4920, 1.3529], device='cuda:1'), covar=tensor([0.2401, 0.1904, 0.1471, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.1837, 0.1750, 0.1665, 0.1812], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 11:58:09,162 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 32250, giga_loss[loss=0.2981, simple_loss=0.3697, pruned_loss=0.1132, over 27764.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3412, pruned_loss=0.09371, over 5654795.47 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3581, pruned_loss=0.1185, over 5681969.45 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3402, pruned_loss=0.09142, over 5661371.60 frames. ], batch size: 474, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:59:29,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 11:59:45,396 INFO [train.py:968] (1/2) Epoch 16, batch 32300, giga_loss[loss=0.2584, simple_loss=0.3385, pruned_loss=0.08919, over 29016.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3412, pruned_loss=0.09282, over 5662634.76 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3577, pruned_loss=0.1183, over 5685246.46 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3406, pruned_loss=0.09082, over 5664736.65 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 11:59:47,618 INFO [optim.py:369] (1/2) 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 12:00:57,229 INFO [train.py:968] (1/2) Epoch 16, batch 32350, giga_loss[loss=0.2249, simple_loss=0.3074, pruned_loss=0.07122, over 28117.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3381, pruned_loss=0.09121, over 5667064.51 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3576, pruned_loss=0.1183, over 5689700.83 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3374, pruned_loss=0.08918, over 5664694.33 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:02:09,455 INFO [train.py:968] (1/2) Epoch 16, batch 32400, giga_loss[loss=0.2404, simple_loss=0.3159, pruned_loss=0.08249, over 28464.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3331, pruned_loss=0.08942, over 5668613.02 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3578, pruned_loss=0.1185, over 5688020.68 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3322, pruned_loss=0.08743, over 5668241.15 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 12:02:15,337 INFO [optim.py:369] (1/2) 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 12:02:58,968 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:03:02,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:03:15,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1924, 1.3284, 1.1278, 0.9565], device='cuda:1'), covar=tensor([0.0906, 0.0468, 0.1050, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0439, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 12:03:18,927 INFO [train.py:968] (1/2) Epoch 16, batch 32450, giga_loss[loss=0.2664, simple_loss=0.3433, pruned_loss=0.09476, over 28629.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3299, pruned_loss=0.08852, over 5657053.32 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3569, pruned_loss=0.1181, over 5684139.78 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3292, pruned_loss=0.08648, over 5658764.82 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:03:35,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:03:45,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6644, 1.8665, 1.5597, 1.8002], device='cuda:1'), covar=tensor([0.2347, 0.2159, 0.2262, 0.2103], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1027, 0.1254, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 12:04:00,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3202, 1.4538, 1.3000, 1.6442], device='cuda:1'), covar=tensor([0.0751, 0.0333, 0.0338, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 12:04:12,950 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6156, 1.9024, 1.8118, 1.5657], device='cuda:1'), covar=tensor([0.2401, 0.1806, 0.1544, 0.1829], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1743, 0.1659, 0.1807], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 12:04:18,761 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=716449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:04:19,880 INFO [train.py:968] (1/2) Epoch 16, batch 32500, giga_loss[loss=0.2784, simple_loss=0.3517, pruned_loss=0.1025, over 28362.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3313, pruned_loss=0.08989, over 5653198.29 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3561, pruned_loss=0.1177, over 5687370.97 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3309, pruned_loss=0.088, over 5651262.61 frames. ], batch size: 368, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:04:23,777 INFO [optim.py:369] (1/2) 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 12:04:56,017 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 32550, libri_loss[loss=0.2576, simple_loss=0.3202, pruned_loss=0.09753, over 29570.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3319, pruned_loss=0.08995, over 5652803.42 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.356, pruned_loss=0.1176, over 5682629.53 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3312, pruned_loss=0.08799, over 5655080.49 frames. ], batch size: 78, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:05:56,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4551, 2.2737, 1.5957, 0.6044], device='cuda:1'), covar=tensor([0.3871, 0.2204, 0.3320, 0.4900], device='cuda:1'), in_proj_covar=tensor([0.1659, 0.1569, 0.1553, 0.1366], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 12:06:25,870 INFO [train.py:968] (1/2) Epoch 16, batch 32600, giga_loss[loss=0.2477, simple_loss=0.3284, pruned_loss=0.08351, over 28956.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3307, pruned_loss=0.08876, over 5648875.87 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3561, pruned_loss=0.1177, over 5684693.42 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3296, pruned_loss=0.08651, over 5648357.90 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:06:29,325 INFO [optim.py:369] (1/2) 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 12:07:02,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=716581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:07:30,148 INFO [train.py:968] (1/2) Epoch 16, batch 32650, giga_loss[loss=0.2271, simple_loss=0.29, pruned_loss=0.08208, over 24286.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3299, pruned_loss=0.08844, over 5660216.55 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3558, pruned_loss=0.1176, over 5690286.79 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3285, pruned_loss=0.08585, over 5654024.07 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:07:58,673 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:08:04,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:08:40,221 INFO [train.py:968] (1/2) Epoch 16, batch 32700, giga_loss[loss=0.2446, simple_loss=0.3276, pruned_loss=0.08082, over 28870.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3295, pruned_loss=0.08833, over 5661993.00 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3558, pruned_loss=0.1177, over 5692527.49 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3278, pruned_loss=0.08561, over 5654726.85 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:08:42,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:08:45,535 INFO [optim.py:369] (1/2) 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 12:09:51,820 INFO [train.py:968] (1/2) Epoch 16, batch 32750, giga_loss[loss=0.2705, simple_loss=0.3468, pruned_loss=0.09711, over 28346.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3291, pruned_loss=0.08761, over 5653398.50 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3558, pruned_loss=0.1177, over 5691255.07 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3274, pruned_loss=0.08507, over 5648191.91 frames. ], batch size: 368, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:10:51,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2768, 1.1682, 3.8567, 3.2661], device='cuda:1'), covar=tensor([0.1614, 0.2919, 0.0401, 0.1505], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0617, 0.0900, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 12:10:57,162 INFO [train.py:968] (1/2) Epoch 16, batch 32800, giga_loss[loss=0.3029, simple_loss=0.3551, pruned_loss=0.1253, over 26762.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3298, pruned_loss=0.08851, over 5658105.42 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3555, pruned_loss=0.1176, over 5694614.92 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3283, pruned_loss=0.08608, over 5650617.07 frames. ], batch size: 555, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 12:11:01,664 INFO [optim.py:369] (1/2) 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 12:12:01,284 INFO [train.py:968] (1/2) Epoch 16, batch 32850, giga_loss[loss=0.2799, simple_loss=0.3547, pruned_loss=0.1026, over 28633.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3291, pruned_loss=0.08821, over 5661596.79 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3552, pruned_loss=0.1174, over 5694571.59 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3276, pruned_loss=0.08578, over 5654877.93 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:12:31,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:13:03,367 INFO [train.py:968] (1/2) Epoch 16, batch 32900, giga_loss[loss=0.2337, simple_loss=0.332, pruned_loss=0.06773, over 28664.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3302, pruned_loss=0.08765, over 5667815.48 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3556, pruned_loss=0.1176, over 5699387.85 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.328, pruned_loss=0.08491, over 5657368.64 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:13:10,125 INFO [optim.py:369] (1/2) 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 12:13:12,679 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 32950, giga_loss[loss=0.25, simple_loss=0.3367, pruned_loss=0.08162, over 29039.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3325, pruned_loss=0.08758, over 5648194.68 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3557, pruned_loss=0.1177, over 5681870.05 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3305, pruned_loss=0.08498, over 5654057.15 frames. ], batch size: 285, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:15:05,970 INFO [train.py:968] (1/2) Epoch 16, batch 33000, giga_loss[loss=0.2264, simple_loss=0.3144, pruned_loss=0.06927, over 28743.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3337, pruned_loss=0.08806, over 5649182.78 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3552, pruned_loss=0.1175, over 5686327.83 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3321, pruned_loss=0.08569, over 5649221.92 frames. ], batch size: 99, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:15:05,970 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 12:15:10,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0392, 1.1987, 3.3876, 3.0368], device='cuda:1'), covar=tensor([0.1962, 0.3043, 0.0585, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0617, 0.0900, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 12:15:13,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5764, 1.7449, 1.3542, 1.3569], device='cuda:1'), covar=tensor([0.0800, 0.0411, 0.0909, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0435, 0.0505, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 12:15:14,318 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 12:15:19,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=716956.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:15:20,157 INFO [optim.py:369] (1/2) 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 12:15:34,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=716967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:15:37,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=716970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:16:00,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2533, 1.2326, 3.7605, 3.2150], device='cuda:1'), covar=tensor([0.1684, 0.2792, 0.0429, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.0702, 0.0616, 0.0899, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 12:16:01,477 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-08 12:16:10,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=716999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:16:11,691 INFO [train.py:968] (1/2) Epoch 16, batch 33050, giga_loss[loss=0.2531, simple_loss=0.3382, pruned_loss=0.084, over 28886.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.334, pruned_loss=0.08891, over 5648235.25 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3542, pruned_loss=0.117, over 5680810.20 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3326, pruned_loss=0.08601, over 5651366.61 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:17:03,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:17:11,167 INFO [train.py:968] (1/2) Epoch 16, batch 33100, giga_loss[loss=0.1975, simple_loss=0.2925, pruned_loss=0.05128, over 28905.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3334, pruned_loss=0.08903, over 5655251.54 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3532, pruned_loss=0.1164, over 5686178.92 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3326, pruned_loss=0.08646, over 5652122.02 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:17:20,327 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:17:45,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0247, 1.2886, 5.3980, 4.0618], device='cuda:1'), covar=tensor([0.1763, 0.3087, 0.0724, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0619, 0.0904, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 12:18:11,863 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717099.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:18:13,613 INFO [train.py:968] (1/2) Epoch 16, batch 33150, giga_loss[loss=0.2409, simple_loss=0.3288, pruned_loss=0.0765, over 28985.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3323, pruned_loss=0.08825, over 5664100.87 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3534, pruned_loss=0.1167, over 5691488.90 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.331, pruned_loss=0.08523, over 5656140.52 frames. ], batch size: 285, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:18:15,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717102.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:18:43,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-08 12:18:54,283 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717131.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:18:55,150 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717132.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:19:19,425 INFO [train.py:968] (1/2) Epoch 16, batch 33200, giga_loss[loss=0.2436, simple_loss=0.3195, pruned_loss=0.08382, over 28939.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3296, pruned_loss=0.08724, over 5662206.46 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.353, pruned_loss=0.1165, over 5693741.97 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3287, pruned_loss=0.08481, over 5653660.54 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:19:26,875 INFO [optim.py:369] (1/2) 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 12:20:07,615 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:20:08,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6154, 1.7773, 1.6817, 1.5858], device='cuda:1'), covar=tensor([0.1911, 0.1618, 0.1257, 0.1371], device='cuda:1'), in_proj_covar=tensor([0.1824, 0.1740, 0.1654, 0.1801], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 12:20:20,172 INFO [train.py:968] (1/2) Epoch 16, batch 33250, giga_loss[loss=0.2372, simple_loss=0.3172, pruned_loss=0.07856, over 28495.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3301, pruned_loss=0.08751, over 5674179.85 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3526, pruned_loss=0.1162, over 5698160.32 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3292, pruned_loss=0.0851, over 5662766.93 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:21:02,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-08 12:21:09,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:21:30,795 INFO [train.py:968] (1/2) Epoch 16, batch 33300, giga_loss[loss=0.2108, simple_loss=0.2961, pruned_loss=0.06278, over 28093.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3337, pruned_loss=0.08926, over 5659579.40 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3527, pruned_loss=0.1164, over 5688088.37 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3327, pruned_loss=0.08681, over 5660100.30 frames. ], batch size: 77, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:21:34,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6444, 1.8531, 1.2750, 1.3563], device='cuda:1'), covar=tensor([0.0905, 0.0504, 0.1033, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0438, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 12:21:38,216 INFO [optim.py:369] (1/2) 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 12:22:07,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3889, 4.1817, 1.5566, 1.6629], device='cuda:1'), covar=tensor([0.1012, 0.0361, 0.0943, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0527, 0.0364, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 12:22:37,207 INFO [train.py:968] (1/2) Epoch 16, batch 33350, giga_loss[loss=0.2803, simple_loss=0.3544, pruned_loss=0.1032, over 28739.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3347, pruned_loss=0.09036, over 5661574.93 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3522, pruned_loss=0.1162, over 5693127.01 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3338, pruned_loss=0.08804, over 5657114.12 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:23:42,163 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 12:23:42,252 INFO [train.py:968] (1/2) Epoch 16, batch 33400, giga_loss[loss=0.2994, simple_loss=0.3787, pruned_loss=0.1101, over 28861.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3369, pruned_loss=0.0912, over 5678386.71 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3515, pruned_loss=0.1157, over 5699875.33 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3365, pruned_loss=0.0891, over 5668093.54 frames. ], batch size: 119, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:23:49,857 INFO [optim.py:369] (1/2) 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 12:24:13,601 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:24:18,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:24:38,487 INFO [train.py:968] (1/2) Epoch 16, batch 33450, giga_loss[loss=0.2565, simple_loss=0.3444, pruned_loss=0.08427, over 28905.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3398, pruned_loss=0.09271, over 5668742.60 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3513, pruned_loss=0.1157, over 5695107.84 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3393, pruned_loss=0.09036, over 5663383.56 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:24:52,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:25:02,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:25:19,418 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 33500, libri_loss[loss=0.3132, simple_loss=0.3627, pruned_loss=0.1319, over 29383.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3405, pruned_loss=0.09287, over 5660860.42 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3511, pruned_loss=0.1157, over 5688746.58 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3398, pruned_loss=0.08998, over 5661072.89 frames. ], batch size: 92, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:25:46,154 INFO [optim.py:369] (1/2) 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 12:26:05,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4265, 1.7443, 1.4221, 1.3609], device='cuda:1'), covar=tensor([0.2163, 0.1954, 0.2182, 0.1807], device='cuda:1'), in_proj_covar=tensor([0.1406, 0.1026, 0.1245, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 12:26:54,654 INFO [train.py:968] (1/2) Epoch 16, batch 33550, giga_loss[loss=0.2405, simple_loss=0.3273, pruned_loss=0.0768, over 28860.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3389, pruned_loss=0.09201, over 5648790.68 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3512, pruned_loss=0.1158, over 5680006.84 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3382, pruned_loss=0.08954, over 5655757.12 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:27:02,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717507.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:27:48,443 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=717538.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:28:06,252 INFO [train.py:968] (1/2) Epoch 16, batch 33600, giga_loss[loss=0.2546, simple_loss=0.3371, pruned_loss=0.08606, over 28454.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3365, pruned_loss=0.09101, over 5652369.96 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3509, pruned_loss=0.1155, over 5684209.41 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.336, pruned_loss=0.08885, over 5653592.08 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 12:28:12,012 INFO [optim.py:369] (1/2) 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 12:28:12,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-08 12:28:19,072 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:22,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:22,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:32,903 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:37,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:28:58,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:29:08,459 INFO [train.py:968] (1/2) Epoch 16, batch 33650, giga_loss[loss=0.2403, simple_loss=0.3207, pruned_loss=0.07998, over 28738.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3369, pruned_loss=0.09119, over 5656447.51 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3511, pruned_loss=0.1156, over 5691068.68 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3358, pruned_loss=0.08861, over 5650472.50 frames. ], batch size: 119, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:29:17,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:29:29,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 12:30:14,233 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:30:14,505 INFO [train.py:968] (1/2) Epoch 16, batch 33700, giga_loss[loss=0.2866, simple_loss=0.3623, pruned_loss=0.1055, over 28587.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3374, pruned_loss=0.09247, over 5661601.82 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3511, pruned_loss=0.1156, over 5696212.82 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3362, pruned_loss=0.08988, over 5651399.50 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:30:17,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:30:22,181 INFO [optim.py:369] (1/2) 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 12:30:54,771 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717682.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:31:07,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1851, 1.2428, 3.3389, 3.0361], device='cuda:1'), covar=tensor([0.1544, 0.2562, 0.0501, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0706, 0.0618, 0.0904, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 12:31:07,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4161, 1.5750, 1.1820, 1.1318], device='cuda:1'), covar=tensor([0.0893, 0.0492, 0.1011, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0438, 0.0508, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 12:31:15,909 INFO [train.py:968] (1/2) Epoch 16, batch 33750, giga_loss[loss=0.2757, simple_loss=0.3494, pruned_loss=0.101, over 28403.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3348, pruned_loss=0.09184, over 5653876.08 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3508, pruned_loss=0.1154, over 5696064.02 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3337, pruned_loss=0.08903, over 5644672.95 frames. ], batch size: 369, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:31:24,915 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=717706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:31:27,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=717709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:32:00,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=717738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:32:17,129 INFO [train.py:968] (1/2) Epoch 16, batch 33800, libri_loss[loss=0.3256, simple_loss=0.3701, pruned_loss=0.1405, over 19484.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3354, pruned_loss=0.0922, over 5645946.13 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.351, pruned_loss=0.1157, over 5693896.98 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3336, pruned_loss=0.08884, over 5639748.26 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:32:25,277 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 33850, giga_loss[loss=0.2638, simple_loss=0.3467, pruned_loss=0.09047, over 28663.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.334, pruned_loss=0.09013, over 5662407.48 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.351, pruned_loss=0.1156, over 5691671.82 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3321, pruned_loss=0.08672, over 5658006.77 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:34:04,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7353, 2.1954, 2.0286, 1.6790], device='cuda:1'), covar=tensor([0.2759, 0.1786, 0.1797, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1731, 0.1645, 0.1791], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 12:34:12,456 INFO [train.py:968] (1/2) Epoch 16, batch 33900, giga_loss[loss=0.2251, simple_loss=0.3204, pruned_loss=0.06486, over 28941.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3356, pruned_loss=0.0886, over 5666918.84 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3509, pruned_loss=0.1155, over 5686141.08 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3337, pruned_loss=0.08541, over 5667053.04 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:34:21,361 INFO [optim.py:369] (1/2) 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 12:35:12,707 INFO [train.py:968] (1/2) Epoch 16, batch 33950, giga_loss[loss=0.2639, simple_loss=0.339, pruned_loss=0.09445, over 27533.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3364, pruned_loss=0.08796, over 5663007.94 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3506, pruned_loss=0.1154, over 5687950.84 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3351, pruned_loss=0.08516, over 5661065.33 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:35:28,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=717913.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 12:35:56,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3474, 2.8351, 1.5743, 1.4575], device='cuda:1'), covar=tensor([0.0891, 0.0313, 0.0857, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0528, 0.0364, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 12:36:17,505 INFO [train.py:968] (1/2) Epoch 16, batch 34000, giga_loss[loss=0.2313, simple_loss=0.323, pruned_loss=0.06976, over 28966.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3358, pruned_loss=0.08741, over 5666243.79 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3501, pruned_loss=0.1151, over 5690134.59 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3348, pruned_loss=0.08473, over 5662035.39 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:36:31,135 INFO [optim.py:369] (1/2) 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 12:37:31,308 INFO [train.py:968] (1/2) Epoch 16, batch 34050, giga_loss[loss=0.2441, simple_loss=0.3288, pruned_loss=0.07966, over 28980.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3354, pruned_loss=0.08711, over 5670517.51 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.35, pruned_loss=0.1149, over 5693402.49 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3344, pruned_loss=0.08459, over 5663847.24 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:37:45,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 12:38:40,460 INFO [train.py:968] (1/2) Epoch 16, batch 34100, giga_loss[loss=0.2519, simple_loss=0.3411, pruned_loss=0.08129, over 28766.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.335, pruned_loss=0.08672, over 5668215.49 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.35, pruned_loss=0.1149, over 5692964.22 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3341, pruned_loss=0.08442, over 5662858.02 frames. ], batch size: 263, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:38:49,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=718056.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 12:38:53,692 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:38:55,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2770, 1.5485, 1.3617, 1.1381], device='cuda:1'), covar=tensor([0.2084, 0.1746, 0.1267, 0.1733], device='cuda:1'), in_proj_covar=tensor([0.1826, 0.1729, 0.1650, 0.1795], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 12:39:39,423 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 34150, giga_loss[loss=0.2205, simple_loss=0.3162, pruned_loss=0.06247, over 28543.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3353, pruned_loss=0.08613, over 5658957.05 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3501, pruned_loss=0.115, over 5685268.19 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3342, pruned_loss=0.08382, over 5660436.89 frames. ], batch size: 336, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:41:04,243 INFO [train.py:968] (1/2) Epoch 16, batch 34200, libri_loss[loss=0.2965, simple_loss=0.3577, pruned_loss=0.1176, over 29548.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3364, pruned_loss=0.08687, over 5657579.81 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3506, pruned_loss=0.1153, over 5688485.63 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3347, pruned_loss=0.08384, over 5654680.33 frames. ], batch size: 83, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:41:14,916 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 34250, libri_loss[loss=0.229, simple_loss=0.2936, pruned_loss=0.08217, over 28684.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3398, pruned_loss=0.08846, over 5669649.61 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.351, pruned_loss=0.1157, over 5690199.80 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3379, pruned_loss=0.0852, over 5665301.05 frames. ], batch size: 63, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:43:24,586 INFO [train.py:968] (1/2) Epoch 16, batch 34300, giga_loss[loss=0.2628, simple_loss=0.3383, pruned_loss=0.09364, over 28979.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3397, pruned_loss=0.08896, over 5682546.84 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.351, pruned_loss=0.1157, over 5694440.86 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.338, pruned_loss=0.0859, over 5675069.96 frames. ], batch size: 145, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:43:37,827 INFO [optim.py:369] (1/2) 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 12:44:07,244 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.00 vs. limit=5.0
+2023-03-08 12:44:32,347 INFO [train.py:968] (1/2) Epoch 16, batch 34350, giga_loss[loss=0.2903, simple_loss=0.3542, pruned_loss=0.1132, over 27626.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3376, pruned_loss=0.08843, over 5677047.10 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3513, pruned_loss=0.1159, over 5688581.07 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3355, pruned_loss=0.08494, over 5675392.16 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:45:28,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-08 12:45:50,637 INFO [train.py:968] (1/2) Epoch 16, batch 34400, giga_loss[loss=0.2891, simple_loss=0.3599, pruned_loss=0.1091, over 28157.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3353, pruned_loss=0.08589, over 5684700.28 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3514, pruned_loss=0.1159, over 5689658.17 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3336, pruned_loss=0.08305, over 5682351.42 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:46:01,854 INFO [optim.py:369] (1/2) 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 12:46:59,304 INFO [train.py:968] (1/2) Epoch 16, batch 34450, giga_loss[loss=0.2456, simple_loss=0.3277, pruned_loss=0.0818, over 28858.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3347, pruned_loss=0.08585, over 5693087.65 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3512, pruned_loss=0.1157, over 5693048.72 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3331, pruned_loss=0.08319, over 5688035.70 frames. ], batch size: 227, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:47:37,080 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 34500, giga_loss[loss=0.2236, simple_loss=0.3112, pruned_loss=0.068, over 28959.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3358, pruned_loss=0.08682, over 5687125.66 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.351, pruned_loss=0.1156, over 5697319.56 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3345, pruned_loss=0.08429, over 5679287.24 frames. ], batch size: 136, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:48:10,323 INFO [optim.py:369] (1/2) 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 12:48:51,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3386, 1.7080, 1.6080, 1.5753], device='cuda:1'), covar=tensor([0.1564, 0.1460, 0.1490, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0722, 0.0679, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 12:49:06,774 INFO [train.py:968] (1/2) Epoch 16, batch 34550, giga_loss[loss=0.2438, simple_loss=0.3276, pruned_loss=0.07994, over 28746.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3382, pruned_loss=0.0882, over 5677689.45 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.351, pruned_loss=0.1156, over 5699682.00 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.337, pruned_loss=0.08598, over 5669389.32 frames. ], batch size: 243, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:50:04,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1869, 1.5510, 1.4543, 1.0725], device='cuda:1'), covar=tensor([0.1448, 0.2405, 0.1266, 0.1576], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0685, 0.0912, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 12:50:07,716 INFO [train.py:968] (1/2) Epoch 16, batch 34600, giga_loss[loss=0.2384, simple_loss=0.3184, pruned_loss=0.07917, over 28716.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3353, pruned_loss=0.08753, over 5679322.22 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3504, pruned_loss=0.1152, over 5700713.09 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3345, pruned_loss=0.08536, over 5671458.00 frames. ], batch size: 243, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:50:18,106 INFO [optim.py:369] (1/2) 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 12:51:06,073 INFO [train.py:968] (1/2) Epoch 16, batch 34650, giga_loss[loss=0.287, simple_loss=0.3649, pruned_loss=0.1045, over 28603.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3347, pruned_loss=0.08811, over 5675743.06 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3503, pruned_loss=0.1151, over 5703863.39 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3339, pruned_loss=0.08591, over 5666130.59 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:52:06,635 INFO [train.py:968] (1/2) Epoch 16, batch 34700, giga_loss[loss=0.2848, simple_loss=0.3607, pruned_loss=0.1044, over 27619.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3365, pruned_loss=0.09008, over 5660831.22 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3502, pruned_loss=0.1151, over 5699069.62 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3356, pruned_loss=0.08778, over 5657203.13 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:52:14,069 INFO [optim.py:369] (1/2) 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 12:52:23,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=718669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:52:45,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5579, 1.6203, 1.6626, 1.5421], device='cuda:1'), covar=tensor([0.2342, 0.2277, 0.1652, 0.1875], device='cuda:1'), in_proj_covar=tensor([0.1825, 0.1725, 0.1643, 0.1805], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 12:52:49,930 INFO [train.py:968] (1/2) Epoch 16, batch 34750, giga_loss[loss=0.3001, simple_loss=0.3762, pruned_loss=0.112, over 28864.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3435, pruned_loss=0.0946, over 5676624.65 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3492, pruned_loss=0.1144, over 5708041.85 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3433, pruned_loss=0.09233, over 5663919.73 frames. ], batch size: 106, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:53:27,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-08 12:53:37,940 INFO [train.py:968] (1/2) Epoch 16, batch 34800, giga_loss[loss=0.3599, simple_loss=0.397, pruned_loss=0.1615, over 23746.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3524, pruned_loss=0.09968, over 5670059.07 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3491, pruned_loss=0.1143, over 5697588.59 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3523, pruned_loss=0.09767, over 5668708.62 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:53:41,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0284, 1.1835, 1.1929, 1.0074], device='cuda:1'), covar=tensor([0.1673, 0.1974, 0.1095, 0.1349], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1730, 0.1649, 0.1811], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 12:53:45,486 INFO [optim.py:369] (1/2) 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 12:54:14,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5365, 1.7787, 1.4771, 1.4920], device='cuda:1'), covar=tensor([0.2523, 0.2491, 0.2860, 0.2258], device='cuda:1'), in_proj_covar=tensor([0.1408, 0.1026, 0.1250, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 12:54:22,044 INFO [train.py:968] (1/2) Epoch 16, batch 34850, giga_loss[loss=0.3122, simple_loss=0.3604, pruned_loss=0.1321, over 23811.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3542, pruned_loss=0.1015, over 5659629.97 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3496, pruned_loss=0.1147, over 5682699.95 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3538, pruned_loss=0.09931, over 5671391.55 frames. ], batch size: 705, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:54:25,908 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=718806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:54:58,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-08 12:54:59,430 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 12:55:02,974 INFO [train.py:968] (1/2) Epoch 16, batch 34900, giga_loss[loss=0.2669, simple_loss=0.3358, pruned_loss=0.09903, over 28608.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3496, pruned_loss=0.09987, over 5670679.77 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.35, pruned_loss=0.1147, over 5687689.61 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.349, pruned_loss=0.09768, over 5675076.06 frames. ], batch size: 78, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:55:12,513 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 34950, giga_loss[loss=0.2468, simple_loss=0.3126, pruned_loss=0.09054, over 28435.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3428, pruned_loss=0.09701, over 5675205.29 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3503, pruned_loss=0.1147, over 5693729.73 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.342, pruned_loss=0.09473, over 5672781.30 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:55:57,028 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-08 12:56:02,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3295, 4.1264, 3.9305, 1.7238], device='cuda:1'), covar=tensor([0.0707, 0.0851, 0.0905, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.1049, 0.0903, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 12:56:26,382 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=718949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:56:27,259 INFO [train.py:968] (1/2) Epoch 16, batch 35000, giga_loss[loss=0.2219, simple_loss=0.2946, pruned_loss=0.07459, over 28995.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3352, pruned_loss=0.09329, over 5681277.57 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3505, pruned_loss=0.1148, over 5688961.82 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.334, pruned_loss=0.09091, over 5682850.88 frames. ], batch size: 106, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:56:29,610 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:56:55,470 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=718981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:57:10,461 INFO [train.py:968] (1/2) Epoch 16, batch 35050, giga_loss[loss=0.2033, simple_loss=0.2817, pruned_loss=0.06245, over 28950.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3278, pruned_loss=0.09027, over 5679750.79 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3501, pruned_loss=0.1143, over 5689406.71 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3268, pruned_loss=0.08825, over 5680143.03 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:57:48,724 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 35100, giga_loss[loss=0.2326, simple_loss=0.3029, pruned_loss=0.08112, over 28872.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3212, pruned_loss=0.08712, over 5679228.72 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3504, pruned_loss=0.1145, over 5690010.94 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3197, pruned_loss=0.08492, over 5678883.76 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:58:01,259 INFO [optim.py:369] (1/2) 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 12:58:34,343 INFO [train.py:968] (1/2) Epoch 16, batch 35150, giga_loss[loss=0.2214, simple_loss=0.2973, pruned_loss=0.07273, over 29161.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3188, pruned_loss=0.08619, over 5686427.78 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3504, pruned_loss=0.1143, over 5685937.33 frames. ], giga_tot_loss[loss=0.2418, simple_loss=0.3164, pruned_loss=0.08355, over 5689013.60 frames. ], batch size: 128, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 12:59:18,869 INFO [train.py:968] (1/2) Epoch 16, batch 35200, giga_loss[loss=0.1894, simple_loss=0.2653, pruned_loss=0.05673, over 28400.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3146, pruned_loss=0.08417, over 5690265.71 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3503, pruned_loss=0.1141, over 5691801.97 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.312, pruned_loss=0.08157, over 5686875.39 frames. ], batch size: 60, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 12:59:27,329 INFO [optim.py:369] (1/2) 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 12:59:47,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 12:59:50,361 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 35250, giga_loss[loss=0.2606, simple_loss=0.3293, pruned_loss=0.09597, over 29005.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3127, pruned_loss=0.08353, over 5686942.25 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3505, pruned_loss=0.114, over 5693772.03 frames. ], giga_tot_loss[loss=0.2349, simple_loss=0.3091, pruned_loss=0.0804, over 5682142.34 frames. ], batch size: 155, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:00:04,314 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:00:06,054 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719207.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:00:16,698 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:00:37,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719240.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:00:39,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 13:00:45,465 INFO [train.py:968] (1/2) Epoch 16, batch 35300, giga_loss[loss=0.2038, simple_loss=0.2799, pruned_loss=0.06386, over 28725.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3107, pruned_loss=0.08295, over 5669903.38 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3513, pruned_loss=0.1145, over 5684820.34 frames. ], giga_tot_loss[loss=0.2324, simple_loss=0.3061, pruned_loss=0.07933, over 5673991.73 frames. ], batch size: 92, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:00:54,557 INFO [optim.py:369] (1/2) 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 13:00:57,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7002, 1.8081, 1.7556, 1.6139], device='cuda:1'), covar=tensor([0.1657, 0.2157, 0.2216, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0735, 0.0691, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 13:01:15,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0283, 1.2973, 1.0125, 1.0324], device='cuda:1'), covar=tensor([0.1124, 0.0573, 0.1570, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0435, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 13:01:28,460 INFO [train.py:968] (1/2) Epoch 16, batch 35350, giga_loss[loss=0.2233, simple_loss=0.2995, pruned_loss=0.0736, over 28320.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3085, pruned_loss=0.08203, over 5674202.22 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3516, pruned_loss=0.1146, over 5681750.65 frames. ], giga_tot_loss[loss=0.23, simple_loss=0.3035, pruned_loss=0.07824, over 5679547.12 frames. ], batch size: 368, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:01:35,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7958, 2.0451, 1.3435, 1.5614], device='cuda:1'), covar=tensor([0.0887, 0.0574, 0.1162, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0436, 0.0508, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 13:01:56,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:02:01,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3141, 1.4690, 1.4691, 1.2288], device='cuda:1'), covar=tensor([0.2508, 0.2214, 0.1220, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1754, 0.1673, 0.1835], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 13:02:12,761 INFO [train.py:968] (1/2) Epoch 16, batch 35400, giga_loss[loss=0.2035, simple_loss=0.2785, pruned_loss=0.06428, over 28596.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3055, pruned_loss=0.08043, over 5674987.33 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3515, pruned_loss=0.1144, over 5676063.16 frames. ], giga_tot_loss[loss=0.2274, simple_loss=0.3008, pruned_loss=0.07697, over 5684805.72 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:02:21,648 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 35450, giga_loss[loss=0.2087, simple_loss=0.2821, pruned_loss=0.06767, over 28591.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3026, pruned_loss=0.07887, over 5681056.73 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3518, pruned_loss=0.1144, over 5678169.00 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.2976, pruned_loss=0.07535, over 5687048.10 frames. ], batch size: 307, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:02:54,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5766, 3.3912, 3.1794, 2.0440], device='cuda:1'), covar=tensor([0.0683, 0.0890, 0.0786, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.1125, 0.1050, 0.0899, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 13:03:36,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5671, 2.1817, 1.7955, 1.8207], device='cuda:1'), covar=tensor([0.0750, 0.0262, 0.0291, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 13:03:39,846 INFO [train.py:968] (1/2) Epoch 16, batch 35500, giga_loss[loss=0.2127, simple_loss=0.2883, pruned_loss=0.0686, over 28841.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.3, pruned_loss=0.07785, over 5672362.68 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3516, pruned_loss=0.1142, over 5681081.93 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2949, pruned_loss=0.07436, over 5674466.24 frames. ], batch size: 199, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:03:49,190 INFO [optim.py:369] (1/2) 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 13:04:26,342 INFO [train.py:968] (1/2) Epoch 16, batch 35550, giga_loss[loss=0.2547, simple_loss=0.3284, pruned_loss=0.09049, over 28949.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.2997, pruned_loss=0.07803, over 5664414.44 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3517, pruned_loss=0.1141, over 5673385.04 frames. ], giga_tot_loss[loss=0.2221, simple_loss=0.2947, pruned_loss=0.07474, over 5672184.03 frames. ], batch size: 145, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:05:07,678 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-08 13:05:10,998 INFO [train.py:968] (1/2) Epoch 16, batch 35600, libri_loss[loss=0.269, simple_loss=0.3321, pruned_loss=0.103, over 29644.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3122, pruned_loss=0.08441, over 5677382.81 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3517, pruned_loss=0.114, over 5677058.89 frames. ], giga_tot_loss[loss=0.2344, simple_loss=0.307, pruned_loss=0.08092, over 5679925.43 frames. ], batch size: 73, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:05:20,598 INFO [optim.py:369] (1/2) 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 13:05:43,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:05:44,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719582.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:06:00,239 INFO [train.py:968] (1/2) Epoch 16, batch 35650, giga_loss[loss=0.2772, simple_loss=0.3587, pruned_loss=0.09787, over 28896.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3267, pruned_loss=0.09252, over 5669957.07 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3525, pruned_loss=0.1146, over 5670516.08 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3213, pruned_loss=0.08885, over 5678761.23 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:06:14,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719615.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:06:24,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-08 13:06:46,221 INFO [train.py:968] (1/2) Epoch 16, batch 35700, giga_loss[loss=0.2968, simple_loss=0.3636, pruned_loss=0.115, over 28777.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.337, pruned_loss=0.09746, over 5675296.14 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.353, pruned_loss=0.115, over 5672568.29 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3321, pruned_loss=0.09407, over 5680491.94 frames. ], batch size: 92, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:06:57,764 INFO [optim.py:369] (1/2) 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 13:07:32,660 INFO [train.py:968] (1/2) Epoch 16, batch 35750, giga_loss[loss=0.2532, simple_loss=0.3359, pruned_loss=0.08527, over 28918.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3425, pruned_loss=0.09887, over 5669544.06 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3531, pruned_loss=0.115, over 5666401.48 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3384, pruned_loss=0.09599, over 5678401.63 frames. ], batch size: 186, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:07:34,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5794, 2.2597, 1.7411, 0.6960], device='cuda:1'), covar=tensor([0.5866, 0.2731, 0.3906, 0.6036], device='cuda:1'), in_proj_covar=tensor([0.1665, 0.1578, 0.1554, 0.1357], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:07:37,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=719707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:07:51,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:07:52,618 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:07:53,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=719727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:07:54,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=719728.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:08:16,815 INFO [train.py:968] (1/2) Epoch 16, batch 35800, libri_loss[loss=0.3063, simple_loss=0.3738, pruned_loss=0.1194, over 29653.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3443, pruned_loss=0.09908, over 5668621.12 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3532, pruned_loss=0.1149, over 5671982.03 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3406, pruned_loss=0.09625, over 5670373.91 frames. ], batch size: 88, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:08:20,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:08:22,637 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719757.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:08:23,315 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719758.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:08:25,192 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:08:34,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=719770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:08:53,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719790.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:08:54,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3283, 1.5957, 1.4901, 1.4530], device='cuda:1'), covar=tensor([0.1647, 0.1813, 0.2089, 0.1798], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0732, 0.0688, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 13:09:04,282 INFO [train.py:968] (1/2) Epoch 16, batch 35850, giga_loss[loss=0.2966, simple_loss=0.374, pruned_loss=0.1096, over 28602.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3463, pruned_loss=0.09971, over 5666508.65 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3535, pruned_loss=0.115, over 5673884.80 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3428, pruned_loss=0.09685, over 5665889.95 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:09:45,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=719850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:09:46,343 INFO [train.py:968] (1/2) Epoch 16, batch 35900, giga_loss[loss=0.2692, simple_loss=0.3424, pruned_loss=0.09798, over 28946.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3476, pruned_loss=0.1007, over 5673255.73 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3536, pruned_loss=0.1151, over 5669899.10 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3446, pruned_loss=0.09793, over 5676505.55 frames. ], batch size: 164, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 13:09:49,561 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:10:14,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=719882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:10:28,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4218, 1.5659, 1.6523, 1.2244], device='cuda:1'), covar=tensor([0.1608, 0.2486, 0.1318, 0.1617], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0692, 0.0920, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 13:10:30,333 INFO [train.py:968] (1/2) Epoch 16, batch 35950, giga_loss[loss=0.3047, simple_loss=0.3721, pruned_loss=0.1187, over 28622.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3504, pruned_loss=0.1027, over 5674961.82 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3533, pruned_loss=0.115, over 5671187.60 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3482, pruned_loss=0.1006, over 5676440.48 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 2.0
+2023-03-08 13:10:57,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1465, 3.3697, 2.2834, 1.1193], device='cuda:1'), covar=tensor([0.6993, 0.2727, 0.3365, 0.5946], device='cuda:1'), in_proj_covar=tensor([0.1654, 0.1565, 0.1545, 0.1348], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:11:14,244 INFO [train.py:968] (1/2) Epoch 16, batch 36000, giga_loss[loss=0.2745, simple_loss=0.3616, pruned_loss=0.09368, over 29040.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3536, pruned_loss=0.1042, over 5677495.76 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.354, pruned_loss=0.1155, over 5664926.80 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3513, pruned_loss=0.1019, over 5684360.86 frames. ], batch size: 155, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:11:14,244 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 13:11:22,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3143, 1.2641, 1.1158, 1.4541], device='cuda:1'), covar=tensor([0.0767, 0.0340, 0.0343, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 13:11:23,452 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 13:11:32,317 INFO [optim.py:369] (1/2) 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 13:12:03,010 INFO [train.py:968] (1/2) Epoch 16, batch 36050, giga_loss[loss=0.2866, simple_loss=0.3688, pruned_loss=0.1022, over 28894.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3555, pruned_loss=0.1036, over 5695078.91 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3541, pruned_loss=0.1154, over 5668277.20 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3536, pruned_loss=0.1017, over 5697776.92 frames. ], batch size: 145, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:12:48,172 INFO [train.py:968] (1/2) Epoch 16, batch 36100, giga_loss[loss=0.2594, simple_loss=0.3456, pruned_loss=0.08654, over 28999.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3558, pruned_loss=0.1031, over 5688473.07 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3538, pruned_loss=0.1151, over 5671872.27 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3546, pruned_loss=0.1017, over 5687747.87 frames. ], batch size: 213, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:12:57,664 INFO [optim.py:369] (1/2) 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 13:13:29,132 INFO [train.py:968] (1/2) Epoch 16, batch 36150, giga_loss[loss=0.2414, simple_loss=0.3337, pruned_loss=0.07457, over 28717.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3559, pruned_loss=0.1021, over 5692877.14 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3538, pruned_loss=0.1151, over 5675450.49 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.355, pruned_loss=0.1008, over 5689330.07 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:13:36,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3895, 1.6395, 1.4454, 1.5692], device='cuda:1'), covar=tensor([0.0822, 0.0329, 0.0335, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 13:13:55,866 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-08 13:14:05,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=720145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:14:06,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 13:14:09,279 INFO [train.py:968] (1/2) Epoch 16, batch 36200, giga_loss[loss=0.2636, simple_loss=0.3528, pruned_loss=0.08726, over 28446.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3553, pruned_loss=0.1007, over 5702603.20 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3545, pruned_loss=0.1155, over 5679522.54 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3539, pruned_loss=0.09894, over 5696419.86 frames. ], batch size: 60, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:14:13,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3928, 1.1084, 4.6761, 3.5693], device='cuda:1'), covar=tensor([0.1883, 0.3159, 0.0362, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0703, 0.0615, 0.0901, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 13:14:19,366 INFO [optim.py:369] (1/2) 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 13:14:52,338 INFO [train.py:968] (1/2) Epoch 16, batch 36250, giga_loss[loss=0.2574, simple_loss=0.3406, pruned_loss=0.08706, over 28607.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3536, pruned_loss=0.09899, over 5698903.81 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3553, pruned_loss=0.1159, over 5679200.94 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3517, pruned_loss=0.0969, over 5694569.03 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:15:35,376 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-08 13:15:36,187 INFO [train.py:968] (1/2) Epoch 16, batch 36300, giga_loss[loss=0.3108, simple_loss=0.3725, pruned_loss=0.1246, over 28920.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3548, pruned_loss=0.1005, over 5693126.86 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3555, pruned_loss=0.116, over 5682191.23 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3531, pruned_loss=0.09852, over 5687272.49 frames. ], batch size: 106, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:15:48,381 INFO [optim.py:369] (1/2) 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 13:16:12,475 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=720288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:16:16,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=720291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:16:26,136 INFO [train.py:968] (1/2) Epoch 16, batch 36350, giga_loss[loss=0.3342, simple_loss=0.3814, pruned_loss=0.1435, over 27635.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3569, pruned_loss=0.1043, over 5691866.34 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3552, pruned_loss=0.1158, over 5685534.04 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3558, pruned_loss=0.1027, over 5684420.67 frames. ], batch size: 472, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:16:43,179 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=720320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:17:10,723 INFO [train.py:968] (1/2) Epoch 16, batch 36400, giga_loss[loss=0.2802, simple_loss=0.352, pruned_loss=0.1043, over 28699.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3589, pruned_loss=0.1074, over 5692888.84 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3554, pruned_loss=0.1159, over 5687660.11 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3579, pruned_loss=0.1061, over 5685282.02 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:17:22,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=720362.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:17:23,401 INFO [optim.py:369] (1/2) 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 13:17:48,362 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.36 vs. limit=5.0
+2023-03-08 13:17:57,646 INFO [train.py:968] (1/2) Epoch 16, batch 36450, giga_loss[loss=0.309, simple_loss=0.3677, pruned_loss=0.1252, over 28621.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.358, pruned_loss=0.1083, over 5696109.95 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3553, pruned_loss=0.1156, over 5693142.99 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3574, pruned_loss=0.1072, over 5685083.23 frames. ], batch size: 85, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:18:32,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-08 13:18:40,988 INFO [train.py:968] (1/2) Epoch 16, batch 36500, giga_loss[loss=0.2823, simple_loss=0.3594, pruned_loss=0.1026, over 28769.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3566, pruned_loss=0.1079, over 5707964.94 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3558, pruned_loss=0.1159, over 5697729.58 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3556, pruned_loss=0.1066, over 5695052.60 frames. ], batch size: 242, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:18:48,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4290, 4.2091, 4.0476, 2.0751], device='cuda:1'), covar=tensor([0.0712, 0.0870, 0.0793, 0.1990], device='cuda:1'), in_proj_covar=tensor([0.1121, 0.1041, 0.0891, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 13:18:52,857 INFO [optim.py:369] (1/2) 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 13:19:05,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=720479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:19:24,544 INFO [train.py:968] (1/2) Epoch 16, batch 36550, giga_loss[loss=0.2612, simple_loss=0.3427, pruned_loss=0.08988, over 28747.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3553, pruned_loss=0.107, over 5705647.45 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3566, pruned_loss=0.1164, over 5695421.97 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3539, pruned_loss=0.1054, over 5697647.76 frames. ], batch size: 262, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:19:43,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-08 13:20:09,217 INFO [train.py:968] (1/2) Epoch 16, batch 36600, giga_loss[loss=0.26, simple_loss=0.3384, pruned_loss=0.09075, over 29059.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3537, pruned_loss=0.1051, over 5700564.67 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3566, pruned_loss=0.1163, over 5697034.13 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3525, pruned_loss=0.1037, over 5692896.14 frames. ], batch size: 128, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:20:22,780 INFO [optim.py:369] (1/2) 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 13:20:55,271 INFO [train.py:968] (1/2) Epoch 16, batch 36650, giga_loss[loss=0.2249, simple_loss=0.3102, pruned_loss=0.06977, over 28855.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3514, pruned_loss=0.1037, over 5687060.08 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3572, pruned_loss=0.1166, over 5690091.49 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3497, pruned_loss=0.1019, over 5687493.70 frames. ], batch size: 174, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:20:57,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-08 13:21:14,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9016, 1.1572, 2.8499, 2.6877], device='cuda:1'), covar=tensor([0.1551, 0.2528, 0.0543, 0.1435], device='cuda:1'), in_proj_covar=tensor([0.0704, 0.0614, 0.0902, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 13:21:34,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5497, 1.8950, 1.4989, 1.6303], device='cuda:1'), covar=tensor([0.2425, 0.2497, 0.2722, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.1405, 0.1024, 0.1245, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 13:21:42,274 INFO [train.py:968] (1/2) Epoch 16, batch 36700, libri_loss[loss=0.2413, simple_loss=0.3148, pruned_loss=0.0839, over 29695.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3463, pruned_loss=0.1006, over 5699683.68 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3572, pruned_loss=0.1164, over 5695321.15 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3447, pruned_loss=0.09894, over 5695066.66 frames. ], batch size: 73, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:21:52,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5465, 1.7040, 1.4427, 1.4959], device='cuda:1'), covar=tensor([0.2553, 0.2606, 0.2886, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.1407, 0.1025, 0.1246, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 13:21:54,053 INFO [optim.py:369] (1/2) 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 13:22:31,259 INFO [train.py:968] (1/2) Epoch 16, batch 36750, giga_loss[loss=0.2429, simple_loss=0.3161, pruned_loss=0.08486, over 29002.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3403, pruned_loss=0.09771, over 5685933.38 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3577, pruned_loss=0.1168, over 5699316.59 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3382, pruned_loss=0.09551, over 5678553.52 frames. ], batch size: 136, lr: 1.99e-03, grad_scale: 4.0
+2023-03-08 13:23:07,691 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=720734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:23:12,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=720737.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:23:25,124 INFO [train.py:968] (1/2) Epoch 16, batch 36800, giga_loss[loss=0.2184, simple_loss=0.3018, pruned_loss=0.06753, over 28750.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3369, pruned_loss=0.09593, over 5673358.61 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3582, pruned_loss=0.1172, over 5691676.19 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3347, pruned_loss=0.09375, over 5673683.86 frames. ], batch size: 284, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:23:38,799 INFO [optim.py:369] (1/2) 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 13:23:59,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 13:24:13,166 INFO [train.py:968] (1/2) Epoch 16, batch 36850, giga_loss[loss=0.2907, simple_loss=0.3572, pruned_loss=0.1121, over 27913.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3387, pruned_loss=0.09637, over 5677594.34 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3584, pruned_loss=0.1173, over 5692763.81 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3366, pruned_loss=0.09443, over 5676705.39 frames. ], batch size: 412, lr: 1.99e-03, grad_scale: 8.0
+2023-03-08 13:24:16,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-08 13:24:36,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-08 13:24:52,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 13:24:55,421 INFO [train.py:968] (1/2) Epoch 16, batch 36900, giga_loss[loss=0.2465, simple_loss=0.3229, pruned_loss=0.08507, over 28955.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3378, pruned_loss=0.09541, over 5690053.62 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3589, pruned_loss=0.1175, over 5693808.52 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3353, pruned_loss=0.09331, over 5688065.91 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:24:59,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=720854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:25:08,577 INFO [optim.py:369] (1/2) 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 13:25:22,472 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=720880.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:25:24,369 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=720883.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:25:37,933 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-08 13:25:38,087 INFO [train.py:968] (1/2) Epoch 16, batch 36950, giga_loss[loss=0.2151, simple_loss=0.2967, pruned_loss=0.06673, over 29060.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3382, pruned_loss=0.09619, over 5692781.18 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3601, pruned_loss=0.1179, over 5697613.34 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3345, pruned_loss=0.09346, over 5687334.59 frames. ], batch size: 128, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:25:47,544 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=720912.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:26:03,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-08 13:26:19,575 INFO [train.py:968] (1/2) Epoch 16, batch 37000, giga_loss[loss=0.2496, simple_loss=0.3263, pruned_loss=0.08644, over 28742.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3365, pruned_loss=0.09556, over 5690906.04 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3606, pruned_loss=0.1181, over 5691111.19 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3327, pruned_loss=0.0928, over 5692135.78 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:26:23,191 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-08 13:26:33,925 INFO [optim.py:369] (1/2) 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 13:27:00,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=720997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:27:02,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:27:02,364 INFO [train.py:968] (1/2) Epoch 16, batch 37050, giga_loss[loss=0.2243, simple_loss=0.3019, pruned_loss=0.07334, over 28936.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3337, pruned_loss=0.09374, over 5700129.68 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3615, pruned_loss=0.1185, over 5691444.82 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3294, pruned_loss=0.09072, over 5700892.94 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:27:23,200 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 37100, giga_loss[loss=0.2565, simple_loss=0.3231, pruned_loss=0.09498, over 28887.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3315, pruned_loss=0.09278, over 5700481.05 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3623, pruned_loss=0.1187, over 5684188.42 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3268, pruned_loss=0.08972, over 5707704.09 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:27:48,027 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5730, 1.7358, 1.4103, 1.7420], device='cuda:1'), covar=tensor([0.2486, 0.2575, 0.2811, 0.2521], device='cuda:1'), in_proj_covar=tensor([0.1413, 0.1029, 0.1253, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 13:27:54,666 INFO [optim.py:369] (1/2) 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 13:28:23,054 INFO [train.py:968] (1/2) Epoch 16, batch 37150, giga_loss[loss=0.243, simple_loss=0.3122, pruned_loss=0.08693, over 28199.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3306, pruned_loss=0.0928, over 5686288.24 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3629, pruned_loss=0.1189, over 5672866.23 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.326, pruned_loss=0.08989, over 5701994.74 frames. ], batch size: 77, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:28:30,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:28:38,034 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:28:49,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2493, 1.7435, 1.3648, 0.4527], device='cuda:1'), covar=tensor([0.3698, 0.2148, 0.3946, 0.5228], device='cuda:1'), in_proj_covar=tensor([0.1644, 0.1557, 0.1536, 0.1347], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:29:03,244 INFO [train.py:968] (1/2) Epoch 16, batch 37200, giga_loss[loss=0.2187, simple_loss=0.2959, pruned_loss=0.07079, over 28472.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.328, pruned_loss=0.0913, over 5698324.63 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3633, pruned_loss=0.119, over 5677673.85 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3231, pruned_loss=0.08826, over 5706943.44 frames. ], batch size: 78, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:29:07,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-08 13:29:16,668 INFO [optim.py:369] (1/2) 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 13:29:43,699 INFO [train.py:968] (1/2) Epoch 16, batch 37250, giga_loss[loss=0.2182, simple_loss=0.298, pruned_loss=0.06917, over 28923.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3262, pruned_loss=0.09038, over 5709055.05 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3636, pruned_loss=0.1188, over 5684086.94 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3209, pruned_loss=0.08727, over 5711135.26 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:29:59,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4914, 2.2504, 1.6502, 0.7611], device='cuda:1'), covar=tensor([0.4752, 0.2034, 0.3756, 0.5350], device='cuda:1'), in_proj_covar=tensor([0.1646, 0.1558, 0.1538, 0.1348], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:30:02,492 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:21,830 INFO [train.py:968] (1/2) Epoch 16, batch 37300, giga_loss[loss=0.2325, simple_loss=0.3142, pruned_loss=0.07545, over 28328.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3249, pruned_loss=0.08946, over 5720690.73 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3638, pruned_loss=0.1186, over 5688824.77 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3195, pruned_loss=0.08643, over 5718749.41 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:30:23,220 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:25,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:35,643 INFO [optim.py:369] (1/2) 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 13:30:43,707 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:30:47,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:31:02,547 INFO [train.py:968] (1/2) Epoch 16, batch 37350, giga_loss[loss=0.212, simple_loss=0.2922, pruned_loss=0.06586, over 28923.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3234, pruned_loss=0.08847, over 5727382.52 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.364, pruned_loss=0.1185, over 5691175.85 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3183, pruned_loss=0.08568, over 5724251.01 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:31:05,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3406, 1.9477, 1.4807, 0.4959], device='cuda:1'), covar=tensor([0.4510, 0.2256, 0.3940, 0.5607], device='cuda:1'), in_proj_covar=tensor([0.1648, 0.1559, 0.1540, 0.1351], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:31:16,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8445, 4.4998, 1.9037, 1.9648], device='cuda:1'), covar=tensor([0.0906, 0.0279, 0.0854, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0523, 0.0360, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 13:31:44,120 INFO [train.py:968] (1/2) Epoch 16, batch 37400, giga_loss[loss=0.3699, simple_loss=0.4067, pruned_loss=0.1666, over 23879.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3258, pruned_loss=0.09017, over 5720728.49 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3642, pruned_loss=0.1183, over 5695361.04 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3208, pruned_loss=0.08753, over 5715110.56 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:31:57,967 INFO [optim.py:369] (1/2) 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 13:32:09,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 13:32:28,709 INFO [train.py:968] (1/2) Epoch 16, batch 37450, giga_loss[loss=0.2622, simple_loss=0.3378, pruned_loss=0.09329, over 28380.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3297, pruned_loss=0.09232, over 5711277.08 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3643, pruned_loss=0.1183, over 5694843.14 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3251, pruned_loss=0.08988, over 5707810.04 frames. ], batch size: 65, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:33:22,013 INFO [train.py:968] (1/2) Epoch 16, batch 37500, giga_loss[loss=0.3737, simple_loss=0.4177, pruned_loss=0.1648, over 27621.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.338, pruned_loss=0.0977, over 5701070.08 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3643, pruned_loss=0.1183, over 5695935.92 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3343, pruned_loss=0.09571, over 5697404.13 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:33:35,407 INFO [optim.py:369] (1/2) 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 13:33:48,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-08 13:34:01,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:34:08,976 INFO [train.py:968] (1/2) Epoch 16, batch 37550, giga_loss[loss=0.276, simple_loss=0.3403, pruned_loss=0.1059, over 23712.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3455, pruned_loss=0.1027, over 5683673.85 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3646, pruned_loss=0.1183, over 5696796.36 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3419, pruned_loss=0.1008, over 5679969.99 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:34:56,376 INFO [train.py:968] (1/2) Epoch 16, batch 37600, giga_loss[loss=0.3229, simple_loss=0.3865, pruned_loss=0.1297, over 29106.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3493, pruned_loss=0.1039, over 5677241.11 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3651, pruned_loss=0.1186, over 5691371.15 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3457, pruned_loss=0.1018, over 5679419.64 frames. ], batch size: 128, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:35:11,008 INFO [optim.py:369] (1/2) 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 13:35:20,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6460, 1.8956, 1.5364, 1.6632], device='cuda:1'), covar=tensor([0.0758, 0.0287, 0.0326, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 13:35:31,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7502, 2.0184, 1.9191, 1.6392], device='cuda:1'), covar=tensor([0.2387, 0.2184, 0.2063, 0.2381], device='cuda:1'), in_proj_covar=tensor([0.1833, 0.1764, 0.1680, 0.1845], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 13:35:43,045 INFO [train.py:968] (1/2) Epoch 16, batch 37650, giga_loss[loss=0.3281, simple_loss=0.3958, pruned_loss=0.1302, over 28711.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3548, pruned_loss=0.1066, over 5673390.24 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3651, pruned_loss=0.1185, over 5693533.35 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3518, pruned_loss=0.1049, over 5672934.60 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:35:43,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:14,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:16,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:27,001 INFO [train.py:968] (1/2) Epoch 16, batch 37700, giga_loss[loss=0.2779, simple_loss=0.3548, pruned_loss=0.1005, over 28519.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.361, pruned_loss=0.1108, over 5668899.65 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3652, pruned_loss=0.1184, over 5697773.93 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.3584, pruned_loss=0.1092, over 5664636.69 frames. ], batch size: 71, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:36:28,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4155, 3.6498, 1.7070, 1.5866], device='cuda:1'), covar=tensor([0.1020, 0.0328, 0.0876, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0523, 0.0360, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 13:36:30,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=721654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:36:30,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 13:36:39,274 INFO [optim.py:369] (1/2) 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 13:36:40,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:06,814 INFO [train.py:968] (1/2) Epoch 16, batch 37750, giga_loss[loss=0.2474, simple_loss=0.3298, pruned_loss=0.08253, over 28607.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3565, pruned_loss=0.1073, over 5675082.56 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3654, pruned_loss=0.1186, over 5690952.46 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3541, pruned_loss=0.1057, over 5676872.08 frames. ], batch size: 307, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:37:33,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:41,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:43,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:37:48,346 INFO [train.py:968] (1/2) Epoch 16, batch 37800, giga_loss[loss=0.2351, simple_loss=0.3266, pruned_loss=0.07176, over 28907.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3539, pruned_loss=0.1049, over 5682002.01 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3652, pruned_loss=0.1188, over 5694816.89 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3519, pruned_loss=0.1032, over 5679710.58 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:37:55,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:00,562 INFO [optim.py:369] (1/2) 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 13:38:04,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-08 13:38:09,226 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:27,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=721797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:30,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=721800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:30,668 INFO [train.py:968] (1/2) Epoch 16, batch 37850, giga_loss[loss=0.2584, simple_loss=0.3402, pruned_loss=0.08833, over 28985.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3523, pruned_loss=0.1036, over 5668299.67 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3654, pruned_loss=0.1189, over 5680918.79 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3503, pruned_loss=0.1018, over 5677175.50 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:38:33,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1408, 1.5693, 1.3401, 1.4236], device='cuda:1'), covar=tensor([0.2020, 0.1715, 0.2160, 0.1773], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0736, 0.0692, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 13:38:47,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:38:54,830 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=721829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:39:12,725 INFO [train.py:968] (1/2) Epoch 16, batch 37900, giga_loss[loss=0.2565, simple_loss=0.3389, pruned_loss=0.08706, over 28784.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3528, pruned_loss=0.1034, over 5682437.79 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3651, pruned_loss=0.1189, over 5685045.29 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3512, pruned_loss=0.1017, over 5685743.48 frames. ], batch size: 284, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:39:25,020 INFO [optim.py:369] (1/2) 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 13:39:43,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6464, 1.7564, 1.2328, 1.2419], device='cuda:1'), covar=tensor([0.0882, 0.0609, 0.1047, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0441, 0.0513, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 13:39:56,025 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:39:56,413 INFO [train.py:968] (1/2) Epoch 16, batch 37950, giga_loss[loss=0.2697, simple_loss=0.3553, pruned_loss=0.09209, over 28909.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3548, pruned_loss=0.1044, over 5682844.49 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3655, pruned_loss=0.1191, over 5686256.07 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3532, pruned_loss=0.1027, over 5684424.50 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:40:10,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2688, 3.2691, 1.5698, 1.4083], device='cuda:1'), covar=tensor([0.1071, 0.0352, 0.0822, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0523, 0.0359, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 13:40:32,691 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0097, 1.1866, 3.4210, 3.0609], device='cuda:1'), covar=tensor([0.1690, 0.2703, 0.0476, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0706, 0.0615, 0.0900, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 13:40:41,170 INFO [train.py:968] (1/2) Epoch 16, batch 38000, giga_loss[loss=0.3432, simple_loss=0.4002, pruned_loss=0.1431, over 28638.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.357, pruned_loss=0.106, over 5685596.83 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3658, pruned_loss=0.1191, over 5689635.83 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3552, pruned_loss=0.1044, over 5683693.04 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:40:41,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=721951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:40:56,477 INFO [optim.py:369] (1/2) 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 13:41:18,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3226, 1.0907, 4.0388, 3.4246], device='cuda:1'), covar=tensor([0.1962, 0.3182, 0.0726, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0705, 0.0615, 0.0899, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 13:41:25,344 INFO [train.py:968] (1/2) Epoch 16, batch 38050, giga_loss[loss=0.3446, simple_loss=0.4008, pruned_loss=0.1442, over 28928.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3583, pruned_loss=0.1074, over 5686048.23 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3661, pruned_loss=0.1195, over 5686128.13 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3563, pruned_loss=0.1054, over 5688445.79 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:42:04,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2460, 1.8075, 1.4386, 0.4467], device='cuda:1'), covar=tensor([0.3707, 0.2302, 0.3728, 0.4759], device='cuda:1'), in_proj_covar=tensor([0.1644, 0.1557, 0.1539, 0.1345], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:42:09,242 INFO [train.py:968] (1/2) Epoch 16, batch 38100, giga_loss[loss=0.2623, simple_loss=0.3404, pruned_loss=0.09206, over 29028.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.359, pruned_loss=0.1083, over 5679440.72 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3663, pruned_loss=0.1195, over 5679701.76 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3571, pruned_loss=0.1066, over 5686303.61 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:42:23,402 INFO [optim.py:369] (1/2) 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 13:42:50,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8285, 2.1166, 1.6483, 2.2484], device='cuda:1'), covar=tensor([0.2443, 0.2423, 0.2829, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1407, 0.1028, 0.1246, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 13:42:52,056 INFO [train.py:968] (1/2) Epoch 16, batch 38150, giga_loss[loss=0.3, simple_loss=0.3724, pruned_loss=0.1139, over 28721.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3591, pruned_loss=0.1081, over 5690944.00 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3667, pruned_loss=0.1196, over 5682653.40 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3572, pruned_loss=0.1064, over 5693797.25 frames. ], batch size: 284, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:42:55,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4021, 1.6875, 1.3472, 1.2972], device='cuda:1'), covar=tensor([0.2570, 0.2623, 0.2966, 0.2307], device='cuda:1'), in_proj_covar=tensor([0.1408, 0.1028, 0.1246, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 13:42:58,234 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:43:10,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-08 13:43:18,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:43:32,631 INFO [train.py:968] (1/2) Epoch 16, batch 38200, giga_loss[loss=0.2683, simple_loss=0.3506, pruned_loss=0.09295, over 29091.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3584, pruned_loss=0.1067, over 5703411.58 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3669, pruned_loss=0.1198, over 5688222.15 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3564, pruned_loss=0.1049, over 5701071.02 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:43:46,047 INFO [optim.py:369] (1/2) 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 13:44:04,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9572, 1.2790, 1.0238, 0.2048], device='cuda:1'), covar=tensor([0.2665, 0.2299, 0.3030, 0.4229], device='cuda:1'), in_proj_covar=tensor([0.1642, 0.1552, 0.1531, 0.1342], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:44:08,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:44:12,494 INFO [train.py:968] (1/2) Epoch 16, batch 38250, giga_loss[loss=0.2956, simple_loss=0.3721, pruned_loss=0.1095, over 28556.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3589, pruned_loss=0.1061, over 5701307.77 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3672, pruned_loss=0.12, over 5688777.53 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3568, pruned_loss=0.1041, over 5699821.81 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:44:22,375 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:44:55,650 INFO [train.py:968] (1/2) Epoch 16, batch 38300, giga_loss[loss=0.2449, simple_loss=0.328, pruned_loss=0.08088, over 28986.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3572, pruned_loss=0.1045, over 5696853.57 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3671, pruned_loss=0.12, over 5680378.47 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3556, pruned_loss=0.1027, over 5704060.82 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:44:56,033 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:44:58,389 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:45:13,128 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722269.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:45:16,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-08 13:45:17,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:18,658 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:20,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:23,336 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:45:37,761 INFO [train.py:968] (1/2) Epoch 16, batch 38350, giga_loss[loss=0.2637, simple_loss=0.3416, pruned_loss=0.09294, over 28798.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3554, pruned_loss=0.1041, over 5699920.22 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3673, pruned_loss=0.1201, over 5683770.17 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3538, pruned_loss=0.1024, over 5702768.97 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:45:46,177 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:00,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:10,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:12,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:20,071 INFO [train.py:968] (1/2) Epoch 16, batch 38400, giga_loss[loss=0.2818, simple_loss=0.3584, pruned_loss=0.1025, over 28939.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3539, pruned_loss=0.1034, over 5708830.90 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3673, pruned_loss=0.1202, over 5687056.57 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3522, pruned_loss=0.1015, over 5708614.51 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:46:34,671 INFO [optim.py:369] (1/2) 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 13:46:34,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:35,673 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:46:59,325 INFO [train.py:968] (1/2) Epoch 16, batch 38450, giga_loss[loss=0.2405, simple_loss=0.3194, pruned_loss=0.08083, over 28496.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3519, pruned_loss=0.1024, over 5710236.29 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3672, pruned_loss=0.1201, over 5686903.00 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3502, pruned_loss=0.1005, over 5710639.01 frames. ], batch size: 60, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:47:14,809 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:47:17,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:47:42,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:47:42,644 INFO [train.py:968] (1/2) Epoch 16, batch 38500, giga_loss[loss=0.2362, simple_loss=0.3256, pruned_loss=0.07339, over 28697.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3528, pruned_loss=0.1033, over 5712628.16 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3676, pruned_loss=0.1202, over 5689970.19 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3508, pruned_loss=0.1014, over 5710736.00 frames. ], batch size: 60, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:47:57,807 INFO [optim.py:369] (1/2) 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 13:47:58,120 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:48:00,310 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 38550, giga_loss[loss=0.2662, simple_loss=0.3476, pruned_loss=0.09241, over 28908.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3533, pruned_loss=0.1036, over 5713326.33 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3679, pruned_loss=0.1203, over 5692387.73 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3513, pruned_loss=0.1017, over 5710055.27 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:48:23,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:48:38,828 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:48:51,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2145, 1.3880, 1.3410, 1.0963], device='cuda:1'), covar=tensor([0.2566, 0.2572, 0.1612, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1774, 0.1689, 0.1847], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 13:49:00,326 INFO [train.py:968] (1/2) Epoch 16, batch 38600, giga_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09166, over 29109.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.103, over 5713857.99 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3685, pruned_loss=0.1207, over 5694691.23 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3513, pruned_loss=0.1008, over 5709699.96 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:49:15,263 INFO [optim.py:369] (1/2) 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 13:49:24,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1547, 1.1504, 3.7563, 3.0549], device='cuda:1'), covar=tensor([0.1762, 0.2984, 0.0445, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0703, 0.0612, 0.0896, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 13:49:27,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:49:40,167 INFO [train.py:968] (1/2) Epoch 16, batch 38650, giga_loss[loss=0.2557, simple_loss=0.3298, pruned_loss=0.09078, over 28785.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3518, pruned_loss=0.1016, over 5716126.62 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1207, over 5698027.15 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3496, pruned_loss=0.09952, over 5710207.82 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:50:17,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722644.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:50:21,364 INFO [train.py:968] (1/2) Epoch 16, batch 38700, giga_loss[loss=0.2679, simple_loss=0.3407, pruned_loss=0.09749, over 28660.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3498, pruned_loss=0.1007, over 5719059.44 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3677, pruned_loss=0.1202, over 5704907.33 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3482, pruned_loss=0.09882, over 5708884.12 frames. ], batch size: 78, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:50:36,939 INFO [optim.py:369] (1/2) 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 13:50:55,440 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 38750, giga_loss[loss=0.2338, simple_loss=0.3126, pruned_loss=0.0775, over 28644.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09938, over 5716429.79 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3678, pruned_loss=0.1204, over 5708133.88 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3454, pruned_loss=0.09751, over 5705548.20 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:51:01,945 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:17,561 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:24,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:26,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:33,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:39,624 INFO [train.py:968] (1/2) Epoch 16, batch 38800, giga_loss[loss=0.2739, simple_loss=0.3365, pruned_loss=0.1057, over 28788.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3446, pruned_loss=0.09813, over 5718538.40 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1199, over 5713062.58 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3428, pruned_loss=0.09616, over 5705307.57 frames. ], batch size: 92, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:51:47,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-08 13:51:49,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:51:54,260 INFO [optim.py:369] (1/2) 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 13:52:07,214 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722787.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:52:09,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722790.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:52:12,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2704, 1.5392, 1.2442, 1.0377], device='cuda:1'), covar=tensor([0.2577, 0.2593, 0.2935, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.1410, 0.1032, 0.1251, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 13:52:17,226 INFO [train.py:968] (1/2) Epoch 16, batch 38850, giga_loss[loss=0.2741, simple_loss=0.3421, pruned_loss=0.103, over 28295.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3441, pruned_loss=0.09836, over 5720647.15 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1198, over 5721200.04 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3414, pruned_loss=0.09564, over 5702658.25 frames. ], batch size: 77, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:52:19,078 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:52:25,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:52:32,450 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722819.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 13:52:51,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3830, 3.3891, 1.4539, 1.5018], device='cuda:1'), covar=tensor([0.0952, 0.0382, 0.0973, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0521, 0.0358, 0.0403], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 13:52:54,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5661, 2.1283, 1.4848, 0.7944], device='cuda:1'), covar=tensor([0.5307, 0.2453, 0.3895, 0.5541], device='cuda:1'), in_proj_covar=tensor([0.1639, 0.1543, 0.1529, 0.1339], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 13:53:00,541 INFO [train.py:968] (1/2) Epoch 16, batch 38900, giga_loss[loss=0.2753, simple_loss=0.3499, pruned_loss=0.1004, over 27841.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3453, pruned_loss=0.09965, over 5713923.52 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3672, pruned_loss=0.1197, over 5723049.45 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.343, pruned_loss=0.09739, over 5698057.27 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 13:53:14,588 INFO [optim.py:369] (1/2) 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 13:53:26,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-08 13:53:30,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=722887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:53:32,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=722890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:53:38,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=722897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:53:40,376 INFO [train.py:968] (1/2) Epoch 16, batch 38950, giga_loss[loss=0.2873, simple_loss=0.3543, pruned_loss=0.1102, over 27691.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3441, pruned_loss=0.09961, over 5715142.49 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3672, pruned_loss=0.1197, over 5726331.48 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3419, pruned_loss=0.09751, over 5699364.44 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:53:56,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=722919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:54:20,217 INFO [train.py:968] (1/2) Epoch 16, batch 39000, giga_loss[loss=0.284, simple_loss=0.358, pruned_loss=0.105, over 28333.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3422, pruned_loss=0.09882, over 5720529.14 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.367, pruned_loss=0.1195, over 5729089.23 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.34, pruned_loss=0.09678, over 5705112.18 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 13:54:20,217 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 13:54:28,571 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 13:54:40,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=722968.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:54:44,102 INFO [optim.py:369] (1/2) 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 13:55:06,743 INFO [train.py:968] (1/2) Epoch 16, batch 39050, giga_loss[loss=0.2441, simple_loss=0.3216, pruned_loss=0.08329, over 28605.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3399, pruned_loss=0.09812, over 5711492.80 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3668, pruned_loss=0.1195, over 5720426.04 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3381, pruned_loss=0.09634, over 5706336.34 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:55:12,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-08 13:55:24,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:55:30,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7398, 3.5534, 3.3336, 1.8909], device='cuda:1'), covar=tensor([0.0654, 0.0808, 0.0791, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.1126, 0.1046, 0.0898, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 13:55:40,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3227, 1.1292, 4.5634, 3.4453], device='cuda:1'), covar=tensor([0.1741, 0.2970, 0.0316, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0703, 0.0612, 0.0896, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 13:55:42,293 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:55:44,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:55:50,087 INFO [train.py:968] (1/2) Epoch 16, batch 39100, giga_loss[loss=0.2405, simple_loss=0.3185, pruned_loss=0.08121, over 28945.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3386, pruned_loss=0.09756, over 5704479.37 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3672, pruned_loss=0.1198, over 5713207.85 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3363, pruned_loss=0.09538, over 5706024.84 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 13:55:56,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4042, 1.6335, 1.5682, 1.4728], device='cuda:1'), covar=tensor([0.1701, 0.1927, 0.2189, 0.2045], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0738, 0.0697, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 13:56:04,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:08,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:08,919 INFO [optim.py:369] (1/2) 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 13:56:11,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:27,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:30,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:56:31,334 INFO [train.py:968] (1/2) Epoch 16, batch 39150, giga_loss[loss=0.2294, simple_loss=0.3126, pruned_loss=0.0731, over 28906.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3406, pruned_loss=0.09876, over 5703989.48 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3674, pruned_loss=0.12, over 5716848.59 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3375, pruned_loss=0.0961, over 5701551.81 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 13:56:39,230 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 13:56:39,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:57:15,714 INFO [train.py:968] (1/2) Epoch 16, batch 39200, giga_loss[loss=0.2961, simple_loss=0.3699, pruned_loss=0.1112, over 28366.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3429, pruned_loss=0.09925, over 5703454.31 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3672, pruned_loss=0.1197, over 5721524.27 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.34, pruned_loss=0.09692, over 5697004.87 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:57:37,667 INFO [optim.py:369] (1/2) 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 13:57:40,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:57:49,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:02,304 INFO [train.py:968] (1/2) Epoch 16, batch 39250, giga_loss[loss=0.301, simple_loss=0.3747, pruned_loss=0.1136, over 28298.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3456, pruned_loss=0.09994, over 5701129.65 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3673, pruned_loss=0.1197, over 5723401.24 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3431, pruned_loss=0.09791, over 5694159.33 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:58:11,950 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:14,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:18,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5945, 1.8368, 1.5014, 1.5890], device='cuda:1'), covar=tensor([0.2598, 0.2509, 0.2878, 0.2297], device='cuda:1'), in_proj_covar=tensor([0.1407, 0.1028, 0.1246, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 13:58:19,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:22,560 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:35,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:39,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:39,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:40,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:41,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:58:48,188 INFO [train.py:968] (1/2) Epoch 16, batch 39300, giga_loss[loss=0.2269, simple_loss=0.3049, pruned_loss=0.07446, over 28856.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3462, pruned_loss=0.0995, over 5694295.78 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3673, pruned_loss=0.1196, over 5724301.24 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3441, pruned_loss=0.09787, over 5687904.03 frames. ], batch size: 66, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:58:49,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:02,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:04,409 INFO [optim.py:369] (1/2) 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 13:59:05,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:29,989 INFO [train.py:968] (1/2) Epoch 16, batch 39350, giga_loss[loss=0.2858, simple_loss=0.3539, pruned_loss=0.1088, over 28749.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.345, pruned_loss=0.09822, over 5694360.11 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5715874.43 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3429, pruned_loss=0.09653, over 5695872.41 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 13:59:37,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:45,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:47,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:48,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2029, 4.0185, 3.8006, 1.8512], device='cuda:1'), covar=tensor([0.0593, 0.0740, 0.0696, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1127, 0.1046, 0.0895, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 13:59:50,655 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:52,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 13:59:56,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6504, 1.7920, 1.8776, 1.6326], device='cuda:1'), covar=tensor([0.2237, 0.1786, 0.1335, 0.1727], device='cuda:1'), in_proj_covar=tensor([0.1849, 0.1777, 0.1689, 0.1842], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 14:00:01,707 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 39400, libri_loss[loss=0.279, simple_loss=0.3411, pruned_loss=0.1084, over 29624.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3452, pruned_loss=0.09905, over 5698499.59 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3673, pruned_loss=0.1197, over 5720836.62 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3429, pruned_loss=0.097, over 5694649.51 frames. ], batch size: 73, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:00:08,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:10,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3949, 3.7778, 1.4852, 1.7174], device='cuda:1'), covar=tensor([0.0895, 0.0355, 0.0825, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0525, 0.0360, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 14:00:14,646 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:21,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1874, 0.8356, 1.0239, 1.3720], device='cuda:1'), covar=tensor([0.0748, 0.0378, 0.0345, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0113, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 14:00:24,663 INFO [optim.py:369] (1/2) 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 14:00:34,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:00:40,922 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 39450, giga_loss[loss=0.2681, simple_loss=0.3402, pruned_loss=0.09798, over 29060.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3477, pruned_loss=0.1009, over 5710131.73 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3682, pruned_loss=0.1202, over 5723326.41 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3443, pruned_loss=0.09805, over 5703827.16 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:00:55,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723414.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:01:06,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 14:01:27,008 INFO [train.py:968] (1/2) Epoch 16, batch 39500, giga_loss[loss=0.2563, simple_loss=0.3348, pruned_loss=0.08892, over 28318.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3485, pruned_loss=0.1014, over 5718111.12 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3684, pruned_loss=0.1205, over 5723695.14 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3453, pruned_loss=0.09864, over 5712462.41 frames. ], batch size: 78, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:01:27,215 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:01:34,159 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 14:01:47,681 INFO [optim.py:369] (1/2) 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 14:01:58,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723486.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:01:58,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:01:59,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:02:00,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723489.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:02:11,421 INFO [train.py:968] (1/2) Epoch 16, batch 39550, giga_loss[loss=0.2732, simple_loss=0.3523, pruned_loss=0.09704, over 28528.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3538, pruned_loss=0.1043, over 5708182.79 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1208, over 5718226.22 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3507, pruned_loss=0.1016, over 5708881.05 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:02:25,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723518.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:02:31,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6224, 1.6037, 1.8822, 1.4199], device='cuda:1'), covar=tensor([0.1661, 0.2222, 0.1328, 0.1590], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0690, 0.0912, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 14:02:32,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2850, 3.0503, 1.4664, 1.3898], device='cuda:1'), covar=tensor([0.0978, 0.0339, 0.0923, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0523, 0.0360, 0.0405], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 14:02:41,400 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:02:44,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:02:52,620 INFO [train.py:968] (1/2) Epoch 16, batch 39600, libri_loss[loss=0.3157, simple_loss=0.3872, pruned_loss=0.1221, over 29185.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3553, pruned_loss=0.1049, over 5712546.16 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3689, pruned_loss=0.1207, over 5721741.31 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3525, pruned_loss=0.1026, over 5709779.51 frames. ], batch size: 97, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:03:08,193 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:03:09,991 INFO [optim.py:369] (1/2) 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 14:03:30,866 INFO [train.py:968] (1/2) Epoch 16, batch 39650, giga_loss[loss=0.2432, simple_loss=0.3207, pruned_loss=0.08288, over 28806.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3553, pruned_loss=0.1046, over 5710366.25 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3689, pruned_loss=0.1208, over 5722325.17 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3528, pruned_loss=0.1023, over 5707110.31 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:03:38,445 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:03:52,247 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723629.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:03:52,977 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:03:54,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723632.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:03:54,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:04:09,082 INFO [train.py:968] (1/2) Epoch 16, batch 39700, giga_loss[loss=0.2925, simple_loss=0.3679, pruned_loss=0.1085, over 28635.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3568, pruned_loss=0.1056, over 5714870.07 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1207, over 5728499.45 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3545, pruned_loss=0.1033, over 5706133.64 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:04:15,627 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723661.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:04:17,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:04:26,201 INFO [optim.py:369] (1/2) 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 14:04:34,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:04:48,333 INFO [train.py:968] (1/2) Epoch 16, batch 39750, giga_loss[loss=0.2869, simple_loss=0.3481, pruned_loss=0.1129, over 23716.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3574, pruned_loss=0.1058, over 5706945.85 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 5725458.75 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.355, pruned_loss=0.1035, over 5702679.45 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:05:26,156 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 39800, giga_loss[loss=0.2606, simple_loss=0.338, pruned_loss=0.09158, over 29034.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3556, pruned_loss=0.1045, over 5713501.23 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3694, pruned_loss=0.1211, over 5726063.46 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3533, pruned_loss=0.1025, over 5709536.12 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:05:38,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:05:47,251 INFO [optim.py:369] (1/2) 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 14:05:48,225 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:05:59,480 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=723785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:02,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723789.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:06:07,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3985, 3.1105, 1.4022, 1.4738], device='cuda:1'), covar=tensor([0.0913, 0.0358, 0.0953, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0525, 0.0362, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 14:06:11,704 INFO [train.py:968] (1/2) Epoch 16, batch 39850, giga_loss[loss=0.2228, simple_loss=0.3044, pruned_loss=0.07057, over 28926.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3514, pruned_loss=0.1024, over 5719704.73 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.369, pruned_loss=0.1209, over 5727836.60 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3499, pruned_loss=0.1008, over 5715017.97 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:06:32,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:33,805 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:35,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:06:53,548 INFO [train.py:968] (1/2) Epoch 16, batch 39900, giga_loss[loss=0.2872, simple_loss=0.3599, pruned_loss=0.1072, over 28862.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3487, pruned_loss=0.1011, over 5715062.80 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3685, pruned_loss=0.1205, over 5730423.08 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3474, pruned_loss=0.0997, over 5708697.70 frames. ], batch size: 199, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:06:59,062 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:07:09,364 INFO [optim.py:369] (1/2) 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 14:07:30,191 INFO [train.py:968] (1/2) Epoch 16, batch 39950, giga_loss[loss=0.282, simple_loss=0.348, pruned_loss=0.108, over 28685.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3496, pruned_loss=0.1016, over 5709174.60 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3683, pruned_loss=0.1203, over 5718773.25 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3477, pruned_loss=0.09964, over 5713801.37 frames. ], batch size: 119, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:07:33,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:07:36,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:07:55,625 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723932.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:07:57,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723935.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:07:58,663 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:11,781 INFO [train.py:968] (1/2) Epoch 16, batch 40000, giga_loss[loss=0.298, simple_loss=0.3561, pruned_loss=0.1199, over 23834.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3523, pruned_loss=0.1011, over 5703603.95 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3693, pruned_loss=0.121, over 5717476.59 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3495, pruned_loss=0.09851, over 5708502.98 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:08:21,847 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=723964.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 14:08:25,209 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=723969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:27,793 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=723972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:28,117 INFO [optim.py:369] (1/2) 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 14:08:39,580 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=723986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:08:50,032 INFO [train.py:968] (1/2) Epoch 16, batch 40050, giga_loss[loss=0.2508, simple_loss=0.3277, pruned_loss=0.08696, over 28908.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 5712820.25 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3687, pruned_loss=0.1207, over 5725020.21 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3497, pruned_loss=0.0988, over 5709520.74 frames. ], batch size: 145, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:08:52,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:09:17,664 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-08 14:09:30,396 INFO [train.py:968] (1/2) Epoch 16, batch 40100, giga_loss[loss=0.3186, simple_loss=0.3811, pruned_loss=0.128, over 27988.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3504, pruned_loss=0.1014, over 5713581.87 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1207, over 5727344.36 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3483, pruned_loss=0.09906, over 5708641.03 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:09:30,579 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=724051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:09:46,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3872, 1.1916, 1.0310, 1.6002], device='cuda:1'), covar=tensor([0.0716, 0.0332, 0.0353, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 14:09:47,317 INFO [optim.py:369] (1/2) 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 14:10:08,445 INFO [train.py:968] (1/2) Epoch 16, batch 40150, giga_loss[loss=0.2679, simple_loss=0.3433, pruned_loss=0.09626, over 28737.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.35, pruned_loss=0.1028, over 5697629.62 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.121, over 5709758.40 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3476, pruned_loss=0.09998, over 5709109.59 frames. ], batch size: 242, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:10:26,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:32,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:36,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:51,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:10:52,818 INFO [train.py:968] (1/2) Epoch 16, batch 40200, libri_loss[loss=0.2342, simple_loss=0.3038, pruned_loss=0.08228, over 29649.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3477, pruned_loss=0.1026, over 5698560.17 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1207, over 5712689.21 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.346, pruned_loss=0.1004, over 5704693.90 frames. ], batch size: 69, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:11:01,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:11:01,715 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:11:10,830 INFO [optim.py:369] (1/2) 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 14:11:35,968 INFO [train.py:968] (1/2) Epoch 16, batch 40250, giga_loss[loss=0.2904, simple_loss=0.3669, pruned_loss=0.1069, over 28295.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3453, pruned_loss=0.1016, over 5708141.59 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1207, over 5713786.77 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3438, pruned_loss=0.0998, over 5711970.28 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:11:56,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8397, 2.0008, 1.8153, 1.9871], device='cuda:1'), covar=tensor([0.0679, 0.0268, 0.0291, 0.0760], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 14:12:16,538 INFO [train.py:968] (1/2) Epoch 16, batch 40300, giga_loss[loss=0.2808, simple_loss=0.3488, pruned_loss=0.1065, over 27579.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3432, pruned_loss=0.1002, over 5711631.86 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.1209, over 5713664.14 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.341, pruned_loss=0.09804, over 5714979.08 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:12:28,033 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:30,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:34,643 INFO [optim.py:369] (1/2) 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 14:12:39,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 14:12:48,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:50,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:12:51,776 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 40350, giga_loss[loss=0.2232, simple_loss=0.3065, pruned_loss=0.06998, over 29063.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3389, pruned_loss=0.09781, over 5718506.59 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1207, over 5717214.19 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3369, pruned_loss=0.09589, over 5717963.45 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:12:56,851 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:00,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:15,442 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:23,290 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:13:31,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.47 vs. limit=5.0
+2023-03-08 14:13:35,879 INFO [train.py:968] (1/2) Epoch 16, batch 40400, giga_loss[loss=0.2262, simple_loss=0.297, pruned_loss=0.07769, over 28585.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3341, pruned_loss=0.09506, over 5717510.24 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3682, pruned_loss=0.1205, over 5719923.84 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3319, pruned_loss=0.09314, over 5714716.09 frames. ], batch size: 60, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:13:54,476 INFO [optim.py:369] (1/2) 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 14:14:02,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5080, 1.3655, 4.1887, 3.3064], device='cuda:1'), covar=tensor([0.1485, 0.2481, 0.0440, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0706, 0.0614, 0.0904, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 14:14:08,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2041, 4.0258, 3.7764, 1.9523], device='cuda:1'), covar=tensor([0.0620, 0.0728, 0.0710, 0.2023], device='cuda:1'), in_proj_covar=tensor([0.1141, 0.1056, 0.0908, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 14:14:16,425 INFO [train.py:968] (1/2) Epoch 16, batch 40450, giga_loss[loss=0.2348, simple_loss=0.3163, pruned_loss=0.0766, over 29030.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3327, pruned_loss=0.09415, over 5718064.04 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3681, pruned_loss=0.1204, over 5722935.23 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3305, pruned_loss=0.09223, over 5713098.18 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:14:27,271 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-08 14:14:38,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:14:39,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4410, 1.4568, 1.1940, 1.0935], device='cuda:1'), covar=tensor([0.0771, 0.0546, 0.0995, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0374, 0.0441, 0.0507, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:14:52,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6290, 1.9972, 1.8652, 1.8194], device='cuda:1'), covar=tensor([0.0732, 0.0269, 0.0301, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0113, 0.0114, 0.0214], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 14:15:01,436 INFO [train.py:968] (1/2) Epoch 16, batch 40500, giga_loss[loss=0.2256, simple_loss=0.3008, pruned_loss=0.07523, over 28430.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3358, pruned_loss=0.09555, over 5716210.20 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3682, pruned_loss=0.1204, over 5723628.55 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3339, pruned_loss=0.09392, over 5711654.16 frames. ], batch size: 78, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:15:19,217 INFO [optim.py:369] (1/2) 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 14:15:19,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3508, 1.0410, 3.8439, 3.1420], device='cuda:1'), covar=tensor([0.1603, 0.2973, 0.0430, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0708, 0.0615, 0.0906, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 14:15:40,665 INFO [train.py:968] (1/2) Epoch 16, batch 40550, giga_loss[loss=0.2441, simple_loss=0.3288, pruned_loss=0.07974, over 28976.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3394, pruned_loss=0.09698, over 5698995.04 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.368, pruned_loss=0.1204, over 5704542.18 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.337, pruned_loss=0.09494, over 5712256.07 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:15:47,982 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=724509.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:16:21,382 INFO [train.py:968] (1/2) Epoch 16, batch 40600, giga_loss[loss=0.2456, simple_loss=0.3264, pruned_loss=0.08237, over 28781.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3427, pruned_loss=0.09801, over 5696584.95 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3683, pruned_loss=0.1205, over 5696635.27 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3403, pruned_loss=0.09607, over 5713435.68 frames. ], batch size: 119, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:16:38,765 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=724569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:16:43,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=724572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:16:44,990 INFO [optim.py:369] (1/2) 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 14:16:55,772 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 14:17:06,346 INFO [train.py:968] (1/2) Epoch 16, batch 40650, giga_loss[loss=0.3626, simple_loss=0.4073, pruned_loss=0.159, over 26481.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3447, pruned_loss=0.09874, over 5702931.68 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3682, pruned_loss=0.1205, over 5699444.80 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3425, pruned_loss=0.09688, over 5713620.50 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:17:06,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=724601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:17:28,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7719, 1.7880, 1.5628, 2.0039], device='cuda:1'), covar=tensor([0.2465, 0.2679, 0.2888, 0.2462], device='cuda:1'), in_proj_covar=tensor([0.1407, 0.1024, 0.1241, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 14:17:49,976 INFO [train.py:968] (1/2) Epoch 16, batch 40700, giga_loss[loss=0.2617, simple_loss=0.3376, pruned_loss=0.09294, over 28936.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3479, pruned_loss=0.101, over 5697309.21 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3685, pruned_loss=0.1207, over 5698242.57 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3457, pruned_loss=0.09921, over 5706639.70 frames. ], batch size: 106, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:18:12,024 INFO [optim.py:369] (1/2) 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 14:18:41,886 INFO [train.py:968] (1/2) Epoch 16, batch 40750, giga_loss[loss=0.2837, simple_loss=0.3533, pruned_loss=0.107, over 28737.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3558, pruned_loss=0.1081, over 5672510.76 frames. ], libri_tot_loss[loss=0.3057, simple_loss=0.3691, pruned_loss=0.1211, over 5690212.91 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3534, pruned_loss=0.106, over 5686071.12 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:18:42,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4043, 1.5680, 1.3901, 1.3069], device='cuda:1'), covar=tensor([0.0747, 0.0321, 0.0302, 0.1008], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0114, 0.0114, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 14:19:29,301 INFO [train.py:968] (1/2) Epoch 16, batch 40800, giga_loss[loss=0.3127, simple_loss=0.3863, pruned_loss=0.1196, over 28763.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3612, pruned_loss=0.1122, over 5677041.14 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5694951.07 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3596, pruned_loss=0.1107, over 5683201.48 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:19:55,019 INFO [optim.py:369] (1/2) 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 14:20:14,514 INFO [train.py:968] (1/2) Epoch 16, batch 40850, giga_loss[loss=0.3507, simple_loss=0.413, pruned_loss=0.1442, over 28691.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3674, pruned_loss=0.1166, over 5670554.15 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1207, over 5687665.30 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3661, pruned_loss=0.1153, over 5681010.00 frames. ], batch size: 242, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:20:39,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6898, 1.7551, 1.4836, 1.8637], device='cuda:1'), covar=tensor([0.2272, 0.2467, 0.2688, 0.2324], device='cuda:1'), in_proj_covar=tensor([0.1411, 0.1028, 0.1248, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 14:20:55,589 INFO [train.py:968] (1/2) Epoch 16, batch 40900, giga_loss[loss=0.3225, simple_loss=0.3877, pruned_loss=0.1286, over 28731.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3734, pruned_loss=0.1221, over 5666052.42 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.121, over 5691011.07 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3721, pruned_loss=0.1207, over 5670388.08 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:21:22,127 INFO [optim.py:369] (1/2) 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 14:21:27,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2884, 1.7683, 1.3973, 1.4832], device='cuda:1'), covar=tensor([0.0752, 0.0318, 0.0319, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0064, 0.0058, 0.0099], device='cuda:1')
+2023-03-08 14:21:28,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=724884.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:21:43,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8862, 3.6924, 3.5174, 1.6668], device='cuda:1'), covar=tensor([0.0688, 0.0860, 0.0833, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.1147, 0.1061, 0.0912, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 14:21:44,216 INFO [train.py:968] (1/2) Epoch 16, batch 40950, giga_loss[loss=0.3284, simple_loss=0.3909, pruned_loss=0.133, over 28873.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.38, pruned_loss=0.1275, over 5669525.02 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5693292.60 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.379, pruned_loss=0.1265, over 5670674.94 frames. ], batch size: 199, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:21:57,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=724913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:22:38,530 INFO [train.py:968] (1/2) Epoch 16, batch 41000, giga_loss[loss=0.3265, simple_loss=0.3893, pruned_loss=0.1318, over 29082.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3837, pruned_loss=0.1318, over 5653176.18 frames. ], libri_tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1207, over 5697397.48 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3838, pruned_loss=0.1314, over 5649630.90 frames. ], batch size: 113, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:23:05,397 INFO [optim.py:369] (1/2) 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 14:23:22,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7294, 1.8017, 1.6211, 1.5710], device='cuda:1'), covar=tensor([0.1667, 0.2408, 0.2189, 0.2151], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0743, 0.0702, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 14:23:32,911 INFO [train.py:968] (1/2) Epoch 16, batch 41050, giga_loss[loss=0.3248, simple_loss=0.384, pruned_loss=0.1328, over 28763.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3858, pruned_loss=0.1338, over 5652297.07 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1207, over 5697863.61 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.386, pruned_loss=0.1337, over 5648176.10 frames. ], batch size: 284, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:23:34,435 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:23:57,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0554, 2.1161, 1.5428, 1.8109], device='cuda:1'), covar=tensor([0.0812, 0.0645, 0.0983, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0445, 0.0512, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:24:00,142 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725027.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:24:03,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=725030.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:24:26,008 INFO [train.py:968] (1/2) Epoch 16, batch 41100, giga_loss[loss=0.3155, simple_loss=0.381, pruned_loss=0.125, over 28662.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3879, pruned_loss=0.1363, over 5630809.77 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3683, pruned_loss=0.1205, over 5701997.05 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3884, pruned_loss=0.1366, over 5623195.15 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:24:31,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8909, 1.9430, 1.4351, 1.5508], device='cuda:1'), covar=tensor([0.0830, 0.0641, 0.1030, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0446, 0.0513, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:24:33,733 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725059.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 14:24:40,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 14:24:51,968 INFO [optim.py:369] (1/2) 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 14:25:04,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3976, 2.3022, 1.7724, 2.0098], device='cuda:1'), covar=tensor([0.0666, 0.0548, 0.0868, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0446, 0.0514, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:25:16,800 INFO [train.py:968] (1/2) Epoch 16, batch 41150, giga_loss[loss=0.3931, simple_loss=0.4293, pruned_loss=0.1785, over 28997.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3916, pruned_loss=0.1405, over 5627525.80 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3682, pruned_loss=0.1204, over 5703794.10 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3926, pruned_loss=0.1413, over 5618334.70 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:25:36,752 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:26:08,791 INFO [train.py:968] (1/2) Epoch 16, batch 41200, giga_loss[loss=0.3395, simple_loss=0.3951, pruned_loss=0.142, over 28725.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3944, pruned_loss=0.1422, over 5639401.69 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3681, pruned_loss=0.1203, over 5703834.82 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3958, pruned_loss=0.1434, over 5630709.61 frames. ], batch size: 307, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:26:38,887 INFO [optim.py:369] (1/2) 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 14:27:02,921 INFO [train.py:968] (1/2) Epoch 16, batch 41250, giga_loss[loss=0.3421, simple_loss=0.3963, pruned_loss=0.144, over 28982.00 frames. ], tot_loss[loss=0.3394, simple_loss=0.3937, pruned_loss=0.1426, over 5637892.51 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.368, pruned_loss=0.1203, over 5707263.90 frames. ], giga_tot_loss[loss=0.3423, simple_loss=0.3959, pruned_loss=0.1444, over 5625357.51 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:27:21,376 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=725220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:27:29,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6921, 1.7731, 1.8414, 1.4359], device='cuda:1'), covar=tensor([0.1483, 0.2281, 0.1271, 0.1537], device='cuda:1'), in_proj_covar=tensor([0.0861, 0.0694, 0.0909, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 14:27:52,706 INFO [train.py:968] (1/2) Epoch 16, batch 41300, giga_loss[loss=0.4066, simple_loss=0.4302, pruned_loss=0.1915, over 23379.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3927, pruned_loss=0.1427, over 5632835.69 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3679, pruned_loss=0.1202, over 5709501.19 frames. ], giga_tot_loss[loss=0.3417, simple_loss=0.3948, pruned_loss=0.1443, over 5620420.35 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:28:20,365 INFO [optim.py:369] (1/2) 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 14:28:29,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=725288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:28:41,782 INFO [train.py:968] (1/2) Epoch 16, batch 41350, giga_loss[loss=0.2599, simple_loss=0.3474, pruned_loss=0.08623, over 28426.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3918, pruned_loss=0.141, over 5640962.68 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3678, pruned_loss=0.1201, over 5714193.80 frames. ], giga_tot_loss[loss=0.3402, simple_loss=0.3943, pruned_loss=0.1431, over 5624726.19 frames. ], batch size: 65, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:29:29,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4869, 3.9468, 1.6587, 1.8061], device='cuda:1'), covar=tensor([0.0911, 0.0354, 0.0912, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0534, 0.0364, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 14:29:35,225 INFO [train.py:968] (1/2) Epoch 16, batch 41400, libri_loss[loss=0.3137, simple_loss=0.3796, pruned_loss=0.1239, over 29537.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.3926, pruned_loss=0.1414, over 5625257.05 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3675, pruned_loss=0.1199, over 5717971.51 frames. ], giga_tot_loss[loss=0.3413, simple_loss=0.3952, pruned_loss=0.1437, over 5607001.98 frames. ], batch size: 83, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:29:44,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5903, 1.7149, 1.2151, 1.2611], device='cuda:1'), covar=tensor([0.0897, 0.0565, 0.1077, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0445, 0.0513, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:29:57,498 INFO [optim.py:369] (1/2) 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 14:30:00,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=725378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:30:23,873 INFO [train.py:968] (1/2) Epoch 16, batch 41450, giga_loss[loss=0.3198, simple_loss=0.3785, pruned_loss=0.1306, over 28740.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3927, pruned_loss=0.1411, over 5614717.32 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5723559.23 frames. ], giga_tot_loss[loss=0.3414, simple_loss=0.3956, pruned_loss=0.1436, over 5592517.60 frames. ], batch size: 242, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:30:58,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:31:01,725 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 41500, giga_loss[loss=0.2754, simple_loss=0.3572, pruned_loss=0.09675, over 28767.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3897, pruned_loss=0.1377, over 5625268.90 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3679, pruned_loss=0.1201, over 5725870.72 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3919, pruned_loss=0.1397, over 5604147.35 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:31:32,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:32:03,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=725493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:32:08,131 INFO [train.py:968] (1/2) Epoch 16, batch 41550, giga_loss[loss=0.3317, simple_loss=0.3991, pruned_loss=0.1321, over 28945.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3882, pruned_loss=0.135, over 5640740.20 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3678, pruned_loss=0.1199, over 5729565.39 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3905, pruned_loss=0.1371, over 5618771.44 frames. ], batch size: 227, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:32:28,737 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:32:32,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=725524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:32:49,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 14:33:00,130 INFO [train.py:968] (1/2) Epoch 16, batch 41600, giga_loss[loss=0.3025, simple_loss=0.3792, pruned_loss=0.1129, over 28971.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3846, pruned_loss=0.1323, over 5638618.55 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3678, pruned_loss=0.1199, over 5732236.38 frames. ], giga_tot_loss[loss=0.3275, simple_loss=0.3867, pruned_loss=0.1342, over 5617790.21 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:33:00,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4516, 1.7991, 1.7693, 1.5991], device='cuda:1'), covar=tensor([0.1561, 0.1412, 0.1731, 0.1449], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0742, 0.0701, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 14:33:01,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:33:26,032 INFO [optim.py:369] (1/2) 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 14:33:43,259 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 41650, giga_loss[loss=0.2891, simple_loss=0.3569, pruned_loss=0.1106, over 29100.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3822, pruned_loss=0.1303, over 5635717.73 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3682, pruned_loss=0.1202, over 5734876.97 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3842, pruned_loss=0.1319, over 5612214.37 frames. ], batch size: 113, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:33:49,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-08 14:33:59,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5330, 1.8136, 1.4414, 1.5250], device='cuda:1'), covar=tensor([0.2651, 0.2679, 0.3082, 0.2321], device='cuda:1'), in_proj_covar=tensor([0.1413, 0.1029, 0.1249, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 14:34:21,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:34:26,203 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 41700, giga_loss[loss=0.2797, simple_loss=0.3587, pruned_loss=0.1003, over 28982.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3815, pruned_loss=0.1299, over 5648870.04 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1207, over 5729735.25 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.383, pruned_loss=0.131, over 5633056.03 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:34:53,084 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=725668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:35:03,773 INFO [optim.py:369] (1/2) 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 14:35:18,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 14:35:18,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-08 14:35:25,215 INFO [train.py:968] (1/2) Epoch 16, batch 41750, giga_loss[loss=0.2896, simple_loss=0.3598, pruned_loss=0.1097, over 28971.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3805, pruned_loss=0.1292, over 5654000.50 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1207, over 5731197.52 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3818, pruned_loss=0.1302, over 5639646.87 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:36:07,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=725738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:36:10,021 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 41800, giga_loss[loss=0.2904, simple_loss=0.365, pruned_loss=0.1079, over 28649.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3791, pruned_loss=0.128, over 5650987.03 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1208, over 5733462.63 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3802, pruned_loss=0.1287, over 5636713.98 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:36:38,278 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:37:08,527 INFO [train.py:968] (1/2) Epoch 16, batch 41850, giga_loss[loss=0.3111, simple_loss=0.3604, pruned_loss=0.1309, over 23940.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3772, pruned_loss=0.1254, over 5635195.72 frames. ], libri_tot_loss[loss=0.3059, simple_loss=0.3691, pruned_loss=0.1213, over 5723665.16 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3781, pruned_loss=0.1258, over 5630093.99 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 14:37:33,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1633, 1.3497, 1.2943, 1.0674], device='cuda:1'), covar=tensor([0.2010, 0.2163, 0.1448, 0.1929], device='cuda:1'), in_proj_covar=tensor([0.1858, 0.1803, 0.1720, 0.1861], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 14:38:05,681 INFO [train.py:968] (1/2) Epoch 16, batch 41900, giga_loss[loss=0.3444, simple_loss=0.4079, pruned_loss=0.1404, over 28832.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3787, pruned_loss=0.1237, over 5651503.01 frames. ], libri_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1212, over 5727422.75 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3795, pruned_loss=0.1241, over 5642862.44 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 14:38:17,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2560, 0.7638, 0.8252, 1.3688], device='cuda:1'), covar=tensor([0.0768, 0.0384, 0.0362, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0100], device='cuda:1')
+2023-03-08 14:38:32,862 INFO [optim.py:369] (1/2) 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 14:38:38,278 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 41950, giga_loss[loss=0.2976, simple_loss=0.3649, pruned_loss=0.1152, over 28703.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3788, pruned_loss=0.1233, over 5645827.44 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3692, pruned_loss=0.1213, over 5711447.57 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3795, pruned_loss=0.1236, over 5652086.82 frames. ], batch size: 92, lr: 1.98e-03, grad_scale: 1.0
+2023-03-08 14:39:09,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-08 14:39:09,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4397, 1.7323, 1.7311, 1.4528], device='cuda:1'), covar=tensor([0.1565, 0.1532, 0.1887, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0740, 0.0699, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 14:39:10,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 14:39:36,719 INFO [train.py:968] (1/2) Epoch 16, batch 42000, giga_loss[loss=0.27, simple_loss=0.3485, pruned_loss=0.09575, over 28900.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3786, pruned_loss=0.1242, over 5649775.47 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3688, pruned_loss=0.1212, over 5714157.22 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3799, pruned_loss=0.1247, over 5650478.18 frames. ], batch size: 112, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:39:36,719 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 14:39:45,356 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 14:39:45,611 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:40:31,897 INFO [train.py:968] (1/2) Epoch 16, batch 42050, giga_loss[loss=0.2946, simple_loss=0.3633, pruned_loss=0.1129, over 28868.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3769, pruned_loss=0.1236, over 5659664.99 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3683, pruned_loss=0.1209, over 5715709.44 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3785, pruned_loss=0.1244, over 5657968.00 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:40:59,936 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 42100, giga_loss[loss=0.4147, simple_loss=0.4333, pruned_loss=0.198, over 26631.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3757, pruned_loss=0.1244, over 5649466.84 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3685, pruned_loss=0.121, over 5709047.07 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.377, pruned_loss=0.1249, over 5653011.02 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:41:48,842 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 42150, giga_loss[loss=0.3256, simple_loss=0.3868, pruned_loss=0.1322, over 27920.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1238, over 5656737.55 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1208, over 5709849.84 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3758, pruned_loss=0.1244, over 5657773.97 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:42:20,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0687, 1.1691, 3.3747, 3.0147], device='cuda:1'), covar=tensor([0.1775, 0.2719, 0.0542, 0.0898], device='cuda:1'), in_proj_covar=tensor([0.0717, 0.0621, 0.0914, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:42:59,863 INFO [train.py:968] (1/2) Epoch 16, batch 42200, giga_loss[loss=0.3322, simple_loss=0.3902, pruned_loss=0.1371, over 28060.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3743, pruned_loss=0.1222, over 5646500.30 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3685, pruned_loss=0.121, over 5691897.26 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3752, pruned_loss=0.1226, over 5662786.83 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:43:24,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-08 14:43:27,665 INFO [optim.py:369] (1/2) 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 14:43:47,674 INFO [train.py:968] (1/2) Epoch 16, batch 42250, libri_loss[loss=0.3151, simple_loss=0.3826, pruned_loss=0.1238, over 27413.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3748, pruned_loss=0.1221, over 5662898.33 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3684, pruned_loss=0.1209, over 5695071.04 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3758, pruned_loss=0.1225, over 5672256.99 frames. ], batch size: 115, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:44:18,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5119, 1.5765, 1.7380, 1.2806], device='cuda:1'), covar=tensor([0.1642, 0.2534, 0.1364, 0.1653], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0698, 0.0914, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 14:44:35,824 INFO [train.py:968] (1/2) Epoch 16, batch 42300, giga_loss[loss=0.2751, simple_loss=0.3438, pruned_loss=0.1032, over 28800.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3744, pruned_loss=0.1219, over 5649192.08 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3686, pruned_loss=0.1211, over 5689165.92 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3751, pruned_loss=0.1221, over 5661727.50 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:44:44,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:44:44,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9849, 1.3298, 1.0598, 0.1933], device='cuda:1'), covar=tensor([0.3330, 0.2618, 0.3924, 0.5573], device='cuda:1'), in_proj_covar=tensor([0.1656, 0.1559, 0.1544, 0.1354], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 14:44:47,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-08 14:45:04,307 INFO [optim.py:369] (1/2) 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 14:45:17,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4390, 2.7908, 1.6360, 1.6393], device='cuda:1'), covar=tensor([0.0790, 0.0299, 0.0687, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0537, 0.0366, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 14:45:24,951 INFO [train.py:968] (1/2) Epoch 16, batch 42350, giga_loss[loss=0.3719, simple_loss=0.4068, pruned_loss=0.1685, over 26658.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.372, pruned_loss=0.1208, over 5660394.67 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3686, pruned_loss=0.121, over 5690443.36 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3727, pruned_loss=0.121, over 5668786.24 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:45:48,403 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 42400, giga_loss[loss=0.2827, simple_loss=0.3423, pruned_loss=0.1116, over 28564.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3702, pruned_loss=0.1202, over 5655073.51 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3683, pruned_loss=0.1209, over 5684548.00 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3711, pruned_loss=0.1204, over 5665353.08 frames. ], batch size: 85, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:46:20,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 14:46:38,102 INFO [optim.py:369] (1/2) 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 14:46:46,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3198, 3.0093, 1.4847, 1.4392], device='cuda:1'), covar=tensor([0.0953, 0.0351, 0.0859, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0538, 0.0366, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 14:46:57,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:46:59,930 INFO [train.py:968] (1/2) Epoch 16, batch 42450, giga_loss[loss=0.2764, simple_loss=0.3481, pruned_loss=0.1024, over 28830.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3681, pruned_loss=0.1194, over 5664827.91 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3677, pruned_loss=0.1206, over 5687789.78 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3693, pruned_loss=0.1199, over 5669967.99 frames. ], batch size: 174, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:47:00,270 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:05,063 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:06,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:30,353 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:47:47,090 INFO [train.py:968] (1/2) Epoch 16, batch 42500, giga_loss[loss=0.2434, simple_loss=0.3161, pruned_loss=0.08532, over 28888.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.367, pruned_loss=0.1189, over 5675388.58 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3683, pruned_loss=0.1209, over 5690915.17 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.1189, over 5676017.73 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:48:04,718 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:48:08,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:48:13,993 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:48:14,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2735, 1.4117, 1.3441, 1.4715], device='cuda:1'), covar=tensor([0.0729, 0.0376, 0.0318, 0.0781], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0100], device='cuda:1')
+2023-03-08 14:48:14,355 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 42550, libri_loss[loss=0.2572, simple_loss=0.3308, pruned_loss=0.09179, over 29570.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3667, pruned_loss=0.1194, over 5675091.36 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3683, pruned_loss=0.1208, over 5694921.74 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3671, pruned_loss=0.1195, over 5671456.38 frames. ], batch size: 75, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:48:35,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:49:04,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-08 14:49:24,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 42600, giga_loss[loss=0.3074, simple_loss=0.3525, pruned_loss=0.1312, over 23600.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3682, pruned_loss=0.1215, over 5651986.75 frames. ], libri_tot_loss[loss=0.3056, simple_loss=0.3688, pruned_loss=0.1212, over 5687680.01 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.368, pruned_loss=0.1212, over 5654348.29 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:49:27,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:49:53,064 INFO [optim.py:369] (1/2) 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 14:49:54,698 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:50:12,554 INFO [train.py:968] (1/2) Epoch 16, batch 42650, giga_loss[loss=0.2976, simple_loss=0.3663, pruned_loss=0.1145, over 28988.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3682, pruned_loss=0.1206, over 5661530.88 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3687, pruned_loss=0.121, over 5692268.10 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3682, pruned_loss=0.1206, over 5658401.18 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:50:17,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4799, 1.6628, 1.2212, 1.1633], device='cuda:1'), covar=tensor([0.0888, 0.0555, 0.1056, 0.1076], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0442, 0.0508, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:50:59,516 INFO [train.py:968] (1/2) Epoch 16, batch 42700, libri_loss[loss=0.2347, simple_loss=0.3106, pruned_loss=0.07941, over 29606.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3683, pruned_loss=0.12, over 5671060.03 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3681, pruned_loss=0.1206, over 5696378.52 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3688, pruned_loss=0.1203, over 5664325.11 frames. ], batch size: 74, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:51:11,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7782, 2.4447, 1.6962, 0.9315], device='cuda:1'), covar=tensor([0.3860, 0.2488, 0.3149, 0.4676], device='cuda:1'), in_proj_covar=tensor([0.1651, 0.1562, 0.1539, 0.1350], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 14:51:26,396 INFO [optim.py:369] (1/2) 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 14:51:32,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2836, 3.0098, 1.4245, 1.4905], device='cuda:1'), covar=tensor([0.0993, 0.0408, 0.0917, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0537, 0.0365, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 14:51:44,566 INFO [train.py:968] (1/2) Epoch 16, batch 42750, giga_loss[loss=0.3042, simple_loss=0.3633, pruned_loss=0.1226, over 27603.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3685, pruned_loss=0.1192, over 5676985.68 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1206, over 5701444.93 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1194, over 5666444.47 frames. ], batch size: 472, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:52:12,310 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:52:25,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3753, 1.1211, 3.9656, 3.2834], device='cuda:1'), covar=tensor([0.1628, 0.2849, 0.0483, 0.0851], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0622, 0.0917, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 14:52:36,965 INFO [train.py:968] (1/2) Epoch 16, batch 42800, giga_loss[loss=0.274, simple_loss=0.3501, pruned_loss=0.09902, over 28784.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3681, pruned_loss=0.1186, over 5687659.65 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3679, pruned_loss=0.1203, over 5704719.03 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3687, pruned_loss=0.1191, over 5675874.76 frames. ], batch size: 119, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 14:52:52,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-08 14:52:59,540 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:53:06,029 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:53:21,153 INFO [train.py:968] (1/2) Epoch 16, batch 42850, giga_loss[loss=0.2929, simple_loss=0.3677, pruned_loss=0.109, over 28940.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3726, pruned_loss=0.1225, over 5675855.54 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3684, pruned_loss=0.1206, over 5693922.40 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1226, over 5676230.60 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:54:10,164 INFO [train.py:968] (1/2) Epoch 16, batch 42900, giga_loss[loss=0.3097, simple_loss=0.3723, pruned_loss=0.1235, over 28674.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3738, pruned_loss=0.125, over 5674794.91 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3682, pruned_loss=0.1204, over 5686235.68 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5681632.61 frames. ], batch size: 119, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:54:14,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=726854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:54:48,466 INFO [optim.py:369] (1/2) 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 14:55:07,546 INFO [train.py:968] (1/2) Epoch 16, batch 42950, giga_loss[loss=0.2876, simple_loss=0.3567, pruned_loss=0.1093, over 28693.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3744, pruned_loss=0.1264, over 5663313.78 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3684, pruned_loss=0.1206, over 5677376.71 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3746, pruned_loss=0.1265, over 5676034.52 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:55:23,664 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:26,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=726920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:35,406 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=726928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:51,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-08 14:55:54,827 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=726949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:55:55,977 INFO [train.py:968] (1/2) Epoch 16, batch 43000, libri_loss[loss=0.2795, simple_loss=0.3586, pruned_loss=0.1002, over 29629.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3753, pruned_loss=0.1272, over 5646636.26 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3676, pruned_loss=0.1199, over 5675971.84 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3764, pruned_loss=0.1282, over 5658014.52 frames. ], batch size: 91, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:56:23,135 INFO [optim.py:369] (1/2) 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 14:56:37,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=726997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:56:40,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:56:41,280 INFO [train.py:968] (1/2) Epoch 16, batch 43050, giga_loss[loss=0.2543, simple_loss=0.3362, pruned_loss=0.08619, over 28988.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3746, pruned_loss=0.1269, over 5647004.78 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.12, over 5669142.48 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3756, pruned_loss=0.1278, over 5661489.67 frames. ], batch size: 155, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:57:06,432 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 43100, giga_loss[loss=0.391, simple_loss=0.4231, pruned_loss=0.1795, over 26364.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3745, pruned_loss=0.126, over 5656364.00 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3675, pruned_loss=0.1199, over 5676292.25 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3755, pruned_loss=0.127, over 5661004.53 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:57:52,988 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 43150, giga_loss[loss=0.2904, simple_loss=0.3624, pruned_loss=0.1092, over 28995.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.373, pruned_loss=0.1236, over 5649524.35 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.3677, pruned_loss=0.1199, over 5669827.58 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3738, pruned_loss=0.1244, over 5658351.01 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 2.0
+2023-03-08 14:58:13,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:58:14,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:58:58,292 INFO [train.py:968] (1/2) Epoch 16, batch 43200, giga_loss[loss=0.2929, simple_loss=0.3614, pruned_loss=0.1122, over 28732.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3715, pruned_loss=0.123, over 5645407.41 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3682, pruned_loss=0.1203, over 5663401.45 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3718, pruned_loss=0.1234, over 5657670.43 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 14:59:05,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 14:59:05,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0854, 1.9563, 1.8919, 1.7173], device='cuda:1'), covar=tensor([0.1836, 0.2822, 0.2422, 0.2497], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0741, 0.0700, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 14:59:24,716 INFO [optim.py:369] (1/2) 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 14:59:27,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 14:59:32,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2753, 1.5214, 1.4066, 1.4657], device='cuda:1'), covar=tensor([0.0715, 0.0322, 0.0310, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0100], device='cuda:1')
+2023-03-08 14:59:44,066 INFO [train.py:968] (1/2) Epoch 16, batch 43250, giga_loss[loss=0.3308, simple_loss=0.3692, pruned_loss=0.1462, over 28407.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3705, pruned_loss=0.1229, over 5656729.60 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3683, pruned_loss=0.1204, over 5665904.72 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.1231, over 5664032.78 frames. ], batch size: 71, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:00:23,839 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:00:26,102 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 43300, giga_loss[loss=0.4558, simple_loss=0.4639, pruned_loss=0.2239, over 26622.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3691, pruned_loss=0.122, over 5655494.61 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3684, pruned_loss=0.1203, over 5659408.56 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3691, pruned_loss=0.1223, over 5666370.96 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:00:53,469 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:00:54,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3804, 2.0184, 1.6392, 1.6214], device='cuda:1'), covar=tensor([0.0806, 0.0280, 0.0319, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 15:00:56,236 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 43350, giga_loss[loss=0.4191, simple_loss=0.4428, pruned_loss=0.1977, over 26621.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3734, pruned_loss=0.125, over 5645600.94 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3687, pruned_loss=0.1205, over 5652817.57 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3732, pruned_loss=0.1252, over 5659376.89 frames. ], batch size: 555, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:01:18,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:20,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:22,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:46,478 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:01:46,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8321, 2.0504, 1.6632, 2.2413], device='cuda:1'), covar=tensor([0.2560, 0.2620, 0.2991, 0.2221], device='cuda:1'), in_proj_covar=tensor([0.1417, 0.1033, 0.1254, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 15:02:03,076 INFO [train.py:968] (1/2) Epoch 16, batch 43400, giga_loss[loss=0.2978, simple_loss=0.3801, pruned_loss=0.1077, over 28905.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3759, pruned_loss=0.1242, over 5657296.80 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3678, pruned_loss=0.12, over 5660158.41 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3768, pruned_loss=0.1249, over 5661777.09 frames. ], batch size: 186, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:02:12,342 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:02:56,761 INFO [train.py:968] (1/2) Epoch 16, batch 43450, giga_loss[loss=0.2853, simple_loss=0.3627, pruned_loss=0.1039, over 29044.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.378, pruned_loss=0.1243, over 5665404.77 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3678, pruned_loss=0.12, over 5664618.60 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3789, pruned_loss=0.1249, over 5664907.74 frames. ], batch size: 128, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:03:23,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4903, 2.0310, 1.4208, 0.7647], device='cuda:1'), covar=tensor([0.4500, 0.2416, 0.3388, 0.5104], device='cuda:1'), in_proj_covar=tensor([0.1661, 0.1572, 0.1540, 0.1353], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 15:03:36,959 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:03:39,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:03:40,057 INFO [train.py:968] (1/2) Epoch 16, batch 43500, giga_loss[loss=0.3461, simple_loss=0.4012, pruned_loss=0.1455, over 28232.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3795, pruned_loss=0.126, over 5664440.05 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3673, pruned_loss=0.1199, over 5668986.12 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3811, pruned_loss=0.1268, over 5660294.24 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:04:02,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:06,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6271, 1.7514, 1.8568, 1.4305], device='cuda:1'), covar=tensor([0.1732, 0.2357, 0.1414, 0.1657], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0699, 0.0916, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 15:04:09,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:10,280 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:04:30,512 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:32,383 INFO [train.py:968] (1/2) Epoch 16, batch 43550, giga_loss[loss=0.3022, simple_loss=0.3696, pruned_loss=0.1174, over 28826.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.38, pruned_loss=0.1264, over 5668090.88 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3674, pruned_loss=0.12, over 5670169.19 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3813, pruned_loss=0.1271, over 5663679.58 frames. ], batch size: 112, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:04:38,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6884, 1.6858, 1.9796, 1.5013], device='cuda:1'), covar=tensor([0.1713, 0.2255, 0.1325, 0.1604], device='cuda:1'), in_proj_covar=tensor([0.0870, 0.0698, 0.0916, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 15:04:39,742 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:04:45,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1458, 3.3973, 2.2954, 1.1411], device='cuda:1'), covar=tensor([0.6311, 0.2844, 0.3168, 0.5673], device='cuda:1'), in_proj_covar=tensor([0.1661, 0.1574, 0.1541, 0.1352], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 15:05:18,309 INFO [train.py:968] (1/2) Epoch 16, batch 43600, giga_loss[loss=0.3146, simple_loss=0.3775, pruned_loss=0.1259, over 28287.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3797, pruned_loss=0.1274, over 5668083.54 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3672, pruned_loss=0.1198, over 5671584.81 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3813, pruned_loss=0.1282, over 5663012.80 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:05:30,455 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727564.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:05:35,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4122, 1.6539, 1.5667, 1.3927], device='cuda:1'), covar=tensor([0.2084, 0.1849, 0.2089, 0.1867], device='cuda:1'), in_proj_covar=tensor([0.1847, 0.1796, 0.1710, 0.1853], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:05:50,793 INFO [optim.py:369] (1/2) 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 15:06:08,125 INFO [train.py:968] (1/2) Epoch 16, batch 43650, giga_loss[loss=0.3686, simple_loss=0.4119, pruned_loss=0.1626, over 27958.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3778, pruned_loss=0.1266, over 5664433.96 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3673, pruned_loss=0.1197, over 5673898.28 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3791, pruned_loss=0.1274, over 5658131.16 frames. ], batch size: 412, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:06:28,855 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:06:31,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:06:55,878 INFO [train.py:968] (1/2) Epoch 16, batch 43700, giga_loss[loss=0.2709, simple_loss=0.3431, pruned_loss=0.09933, over 28863.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3746, pruned_loss=0.1249, over 5669783.42 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3667, pruned_loss=0.1193, over 5676891.09 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3763, pruned_loss=0.1261, over 5661725.33 frames. ], batch size: 199, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:06:59,112 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:07:27,454 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 43750, giga_loss[loss=0.3101, simple_loss=0.369, pruned_loss=0.1256, over 28504.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3729, pruned_loss=0.1244, over 5661709.00 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3664, pruned_loss=0.1192, over 5664067.97 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3748, pruned_loss=0.1256, over 5666255.55 frames. ], batch size: 336, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:08:17,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:08:31,433 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-08 15:08:32,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5295, 1.6390, 1.6525, 1.4825], device='cuda:1'), covar=tensor([0.2602, 0.2309, 0.1820, 0.2130], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1799, 0.1717, 0.1861], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:08:36,125 INFO [train.py:968] (1/2) Epoch 16, batch 43800, giga_loss[loss=0.3868, simple_loss=0.4283, pruned_loss=0.1726, over 28677.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3739, pruned_loss=0.1252, over 5668521.21 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3663, pruned_loss=0.119, over 5666027.55 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3756, pruned_loss=0.1264, over 5670720.66 frames. ], batch size: 262, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:09:06,682 INFO [optim.py:369] (1/2) 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 15:09:24,663 INFO [train.py:968] (1/2) Epoch 16, batch 43850, giga_loss[loss=0.3727, simple_loss=0.423, pruned_loss=0.1612, over 28149.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3741, pruned_loss=0.1259, over 5665086.24 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3658, pruned_loss=0.1187, over 5672087.93 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3761, pruned_loss=0.1274, over 5661203.00 frames. ], batch size: 368, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:09:36,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2117, 1.5044, 1.4999, 1.1012], device='cuda:1'), covar=tensor([0.1553, 0.2449, 0.1335, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0700, 0.0917, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 15:10:06,327 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 43900, giga_loss[loss=0.3034, simple_loss=0.362, pruned_loss=0.1224, over 28788.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3722, pruned_loss=0.1248, over 5674868.45 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3663, pruned_loss=0.119, over 5680524.66 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3736, pruned_loss=0.1259, over 5663822.72 frames. ], batch size: 99, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:10:12,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-08 15:10:27,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:30,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:32,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:37,712 INFO [optim.py:369] (1/2) 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 15:10:37,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=727883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:10:49,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2450, 1.5224, 1.4772, 1.4130], device='cuda:1'), covar=tensor([0.1697, 0.1565, 0.2200, 0.1625], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0747, 0.0701, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 15:10:52,762 INFO [train.py:968] (1/2) Epoch 16, batch 43950, giga_loss[loss=0.3407, simple_loss=0.3744, pruned_loss=0.1535, over 23490.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3708, pruned_loss=0.1235, over 5668622.96 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3663, pruned_loss=0.1189, over 5674745.62 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.372, pruned_loss=0.1246, over 5665854.80 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:10:59,157 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=727907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:11:10,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3137, 1.7625, 1.4881, 1.5187], device='cuda:1'), covar=tensor([0.0708, 0.0390, 0.0319, 0.0756], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 15:11:18,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4643, 4.2999, 4.0614, 1.7964], device='cuda:1'), covar=tensor([0.0579, 0.0695, 0.0745, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.1174, 0.1093, 0.0940, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 15:11:32,955 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 44000, giga_loss[loss=0.2898, simple_loss=0.3548, pruned_loss=0.1124, over 28615.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3725, pruned_loss=0.1237, over 5668674.75 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3662, pruned_loss=0.1187, over 5680631.86 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5660972.52 frames. ], batch size: 78, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 15:11:54,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=727961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:15,118 INFO [optim.py:369] (1/2) 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 15:12:15,570 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 15:12:22,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=727991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:24,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=727994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:32,260 INFO [train.py:968] (1/2) Epoch 16, batch 44050, giga_loss[loss=0.3076, simple_loss=0.3718, pruned_loss=0.1217, over 28924.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3746, pruned_loss=0.1249, over 5676752.48 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.1188, over 5684134.02 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3755, pruned_loss=0.1257, over 5667693.45 frames. ], batch size: 213, lr: 1.98e-03, grad_scale: 8.0
+2023-03-08 15:12:32,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-08 15:12:46,389 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:49,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:52,032 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:12:57,859 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:01,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:20,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:21,879 INFO [train.py:968] (1/2) Epoch 16, batch 44100, giga_loss[loss=0.3927, simple_loss=0.4104, pruned_loss=0.1875, over 23811.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3738, pruned_loss=0.1251, over 5668424.89 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3658, pruned_loss=0.1183, over 5687764.79 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3752, pruned_loss=0.1263, over 5658141.98 frames. ], batch size: 705, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:13:27,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=728057.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:13:28,463 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:13:51,102 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:13:53,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728085.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:13:59,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-08 15:14:04,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5495, 2.2017, 1.5960, 0.7178], device='cuda:1'), covar=tensor([0.5719, 0.2962, 0.3975, 0.6432], device='cuda:1'), in_proj_covar=tensor([0.1670, 0.1580, 0.1549, 0.1360], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 15:14:05,284 INFO [train.py:968] (1/2) Epoch 16, batch 44150, giga_loss[loss=0.3163, simple_loss=0.3935, pruned_loss=0.1195, over 28983.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3757, pruned_loss=0.1241, over 5671523.18 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5688683.16 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3772, pruned_loss=0.1254, over 5662165.58 frames. ], batch size: 136, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:14:15,765 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.21 vs. limit=5.0
+2023-03-08 15:14:18,062 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728114.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:14:52,432 INFO [train.py:968] (1/2) Epoch 16, batch 44200, giga_loss[loss=0.2579, simple_loss=0.3501, pruned_loss=0.0828, over 28963.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3769, pruned_loss=0.1223, over 5686670.12 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.1181, over 5689790.70 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3784, pruned_loss=0.1235, over 5678295.21 frames. ], batch size: 164, lr: 1.98e-03, grad_scale: 4.0
+2023-03-08 15:15:28,764 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 44250, giga_loss[loss=0.3129, simple_loss=0.3896, pruned_loss=0.1181, over 28851.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3801, pruned_loss=0.1237, over 5691708.70 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3656, pruned_loss=0.118, over 5695291.69 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3815, pruned_loss=0.1247, over 5679762.76 frames. ], batch size: 174, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:16:28,833 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 15:16:31,829 INFO [train.py:968] (1/2) Epoch 16, batch 44300, giga_loss[loss=0.3451, simple_loss=0.4084, pruned_loss=0.141, over 28605.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3825, pruned_loss=0.1266, over 5688155.15 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5699574.45 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3837, pruned_loss=0.1274, over 5674844.91 frames. ], batch size: 307, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:17:04,357 INFO [optim.py:369] (1/2) 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 15:17:22,340 INFO [train.py:968] (1/2) Epoch 16, batch 44350, giga_loss[loss=0.32, simple_loss=0.3865, pruned_loss=0.1267, over 28884.00 frames. ], tot_loss[loss=0.3211, simple_loss=0.3837, pruned_loss=0.1292, over 5668838.85 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3653, pruned_loss=0.1178, over 5704315.61 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3855, pruned_loss=0.1305, over 5653384.48 frames. ], batch size: 174, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:17:24,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2656, 1.3783, 3.3975, 2.9796], device='cuda:1'), covar=tensor([0.1451, 0.2388, 0.0465, 0.1526], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0621, 0.0920, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 15:17:26,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 15:17:26,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9810, 4.8032, 4.5857, 2.0585], device='cuda:1'), covar=tensor([0.0456, 0.0580, 0.0695, 0.2054], device='cuda:1'), in_proj_covar=tensor([0.1175, 0.1097, 0.0940, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 15:17:55,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=728336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:18:08,533 INFO [train.py:968] (1/2) Epoch 16, batch 44400, libri_loss[loss=0.3209, simple_loss=0.3681, pruned_loss=0.1369, over 29660.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3829, pruned_loss=0.1292, over 5673394.15 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5707392.14 frames. ], giga_tot_loss[loss=0.3227, simple_loss=0.3846, pruned_loss=0.1304, over 5657758.59 frames. ], batch size: 73, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:18:10,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 15:18:36,839 INFO [optim.py:369] (1/2) 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 15:18:53,255 INFO [train.py:968] (1/2) Epoch 16, batch 44450, giga_loss[loss=0.2677, simple_loss=0.3542, pruned_loss=0.09058, over 28617.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3798, pruned_loss=0.1265, over 5678029.52 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3646, pruned_loss=0.1174, over 5713596.21 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3824, pruned_loss=0.1282, over 5658851.31 frames. ], batch size: 85, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:19:01,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1748, 1.6308, 1.4061, 1.3840], device='cuda:1'), covar=tensor([0.2104, 0.1965, 0.2350, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0742, 0.0698, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 15:19:11,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.33 vs. limit=2.0
+2023-03-08 15:19:19,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=728432.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:19:37,932 INFO [train.py:968] (1/2) Epoch 16, batch 44500, libri_loss[loss=0.2953, simple_loss=0.3652, pruned_loss=0.1127, over 29510.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3798, pruned_loss=0.1246, over 5672711.70 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3648, pruned_loss=0.1175, over 5707036.14 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3823, pruned_loss=0.1262, over 5661677.12 frames. ], batch size: 81, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:19:41,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4580, 1.5916, 1.2029, 1.2170], device='cuda:1'), covar=tensor([0.0862, 0.0523, 0.1019, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0448, 0.0510, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 15:19:48,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8835, 2.3640, 2.1552, 1.7214], device='cuda:1'), covar=tensor([0.2850, 0.1933, 0.1998, 0.2368], device='cuda:1'), in_proj_covar=tensor([0.1864, 0.1806, 0.1729, 0.1868], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:19:58,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5981, 1.9333, 1.4955, 1.7578], device='cuda:1'), covar=tensor([0.0672, 0.0265, 0.0305, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 15:20:05,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:20:07,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:20:09,151 INFO [optim.py:369] (1/2) 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 15:20:22,202 INFO [train.py:968] (1/2) Epoch 16, batch 44550, giga_loss[loss=0.3052, simple_loss=0.3755, pruned_loss=0.1175, over 27948.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3808, pruned_loss=0.1247, over 5676422.68 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3647, pruned_loss=0.1176, over 5711501.26 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3833, pruned_loss=0.126, over 5662514.68 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:20:31,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=728511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:21:02,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4416, 3.5356, 1.5247, 1.4803], device='cuda:1'), covar=tensor([0.0996, 0.0458, 0.0923, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0541, 0.0367, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 15:21:12,883 INFO [train.py:968] (1/2) Epoch 16, batch 44600, giga_loss[loss=0.3165, simple_loss=0.3977, pruned_loss=0.1176, over 28640.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3817, pruned_loss=0.1259, over 5678791.85 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3649, pruned_loss=0.1178, over 5711691.85 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3838, pruned_loss=0.1269, over 5667092.55 frames. ], batch size: 85, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:21:39,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=728575.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:21:42,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=728578.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:21:47,197 INFO [optim.py:369] (1/2) 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 15:22:01,670 INFO [train.py:968] (1/2) Epoch 16, batch 44650, giga_loss[loss=0.2611, simple_loss=0.3377, pruned_loss=0.09226, over 28459.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3816, pruned_loss=0.1263, over 5686002.34 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1179, over 5715578.65 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3834, pruned_loss=0.1272, over 5672638.15 frames. ], batch size: 60, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:22:07,437 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 44700, giga_loss[loss=0.3417, simple_loss=0.3937, pruned_loss=0.1448, over 27864.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3802, pruned_loss=0.1269, over 5668203.86 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3647, pruned_loss=0.1177, over 5718386.62 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3822, pruned_loss=0.1279, over 5654562.51 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:23:22,693 INFO [optim.py:369] (1/2) 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 15:23:36,482 INFO [train.py:968] (1/2) Epoch 16, batch 44750, giga_loss[loss=0.2883, simple_loss=0.362, pruned_loss=0.1073, over 29066.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3779, pruned_loss=0.1261, over 5668481.88 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3648, pruned_loss=0.1179, over 5718540.26 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3798, pruned_loss=0.127, over 5655973.84 frames. ], batch size: 155, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:24:02,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=728727.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:24:06,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6935, 1.8565, 1.7984, 1.5124], device='cuda:1'), covar=tensor([0.2883, 0.2260, 0.1873, 0.2470], device='cuda:1'), in_proj_covar=tensor([0.1849, 0.1789, 0.1716, 0.1857], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:24:26,140 INFO [train.py:968] (1/2) Epoch 16, batch 44800, giga_loss[loss=0.327, simple_loss=0.3596, pruned_loss=0.1472, over 23429.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1254, over 5661371.36 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.118, over 5720291.60 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3773, pruned_loss=0.1261, over 5649470.36 frames. ], batch size: 705, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:24:39,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4531, 4.2965, 4.0746, 2.0591], device='cuda:1'), covar=tensor([0.0620, 0.0724, 0.0757, 0.1992], device='cuda:1'), in_proj_covar=tensor([0.1171, 0.1097, 0.0941, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 15:24:54,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=728782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:24:56,563 INFO [optim.py:369] (1/2) 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 15:25:09,287 INFO [train.py:968] (1/2) Epoch 16, batch 44850, libri_loss[loss=0.3497, simple_loss=0.4017, pruned_loss=0.1489, over 19540.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5654351.75 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1183, over 5704282.27 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3756, pruned_loss=0.1256, over 5658219.53 frames. ], batch size: 186, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:25:57,706 INFO [train.py:968] (1/2) Epoch 16, batch 44900, giga_loss[loss=0.2692, simple_loss=0.3443, pruned_loss=0.09702, over 29053.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.374, pruned_loss=0.1247, over 5658575.78 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1183, over 5705758.28 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5659548.11 frames. ], batch size: 128, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:26:27,392 INFO [optim.py:369] (1/2) 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 15:26:43,637 INFO [train.py:968] (1/2) Epoch 16, batch 44950, giga_loss[loss=0.2842, simple_loss=0.356, pruned_loss=0.1062, over 28703.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.1219, over 5645531.82 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3652, pruned_loss=0.118, over 5699503.77 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3723, pruned_loss=0.1227, over 5650265.01 frames. ], batch size: 119, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:27:26,140 INFO [train.py:968] (1/2) Epoch 16, batch 45000, giga_loss[loss=0.2555, simple_loss=0.3295, pruned_loss=0.09081, over 28915.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 5646978.82 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1184, over 5684276.65 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3693, pruned_loss=0.1193, over 5663292.98 frames. ], batch size: 106, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:27:26,140 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 15:27:32,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2357, 3.1283, 1.4120, 1.4726], device='cuda:1'), covar=tensor([0.1175, 0.0480, 0.1034, 0.1533], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0538, 0.0365, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 15:27:34,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2772, 1.7124, 1.5965, 1.1564], device='cuda:1'), covar=tensor([0.1894, 0.2815, 0.1695, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0701, 0.0919, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 15:27:34,847 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 15:27:48,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2088, 1.1862, 3.5999, 3.1446], device='cuda:1'), covar=tensor([0.1688, 0.2825, 0.0461, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0622, 0.0918, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 15:28:11,454 INFO [optim.py:369] (1/2) 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 15:28:29,337 INFO [train.py:968] (1/2) Epoch 16, batch 45050, giga_loss[loss=0.3056, simple_loss=0.3703, pruned_loss=0.1204, over 28765.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3677, pruned_loss=0.1186, over 5638503.53 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1184, over 5685143.26 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3683, pruned_loss=0.1189, over 5650301.53 frames. ], batch size: 284, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:29:18,461 INFO [train.py:968] (1/2) Epoch 16, batch 45100, giga_loss[loss=0.2898, simple_loss=0.3539, pruned_loss=0.1129, over 28936.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3656, pruned_loss=0.1184, over 5638991.98 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3653, pruned_loss=0.1184, over 5688524.99 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3662, pruned_loss=0.1186, over 5644821.81 frames. ], batch size: 213, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:29:54,129 INFO [optim.py:369] (1/2) 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 15:30:06,122 INFO [train.py:968] (1/2) Epoch 16, batch 45150, giga_loss[loss=0.315, simple_loss=0.3761, pruned_loss=0.1269, over 27919.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3658, pruned_loss=0.1192, over 5640862.49 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3657, pruned_loss=0.1186, over 5690984.02 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.366, pruned_loss=0.1191, over 5642367.63 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:30:08,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=729102.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:30:52,971 INFO [train.py:968] (1/2) Epoch 16, batch 45200, giga_loss[loss=0.3293, simple_loss=0.365, pruned_loss=0.1468, over 23541.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3678, pruned_loss=0.1197, over 5637832.31 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.366, pruned_loss=0.1187, over 5690487.48 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1196, over 5638563.86 frames. ], batch size: 705, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:31:00,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=729157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:31:29,186 INFO [optim.py:369] (1/2) 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 15:31:42,111 INFO [train.py:968] (1/2) Epoch 16, batch 45250, giga_loss[loss=0.2975, simple_loss=0.3655, pruned_loss=0.1148, over 28919.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5634090.11 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3663, pruned_loss=0.119, over 5684048.52 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3692, pruned_loss=0.1203, over 5640104.00 frames. ], batch size: 227, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:32:04,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=729223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:32:25,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=729245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:32:28,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=729248.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 15:32:30,935 INFO [train.py:968] (1/2) Epoch 16, batch 45300, giga_loss[loss=0.3017, simple_loss=0.3631, pruned_loss=0.1202, over 28628.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3694, pruned_loss=0.1209, over 5618928.22 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 5688281.03 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3691, pruned_loss=0.1207, over 5619009.78 frames. ], batch size: 78, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:32:40,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6692, 1.9618, 1.8628, 1.6125], device='cuda:1'), covar=tensor([0.2155, 0.1695, 0.1430, 0.1696], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1794, 0.1719, 0.1857], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:32:55,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=729277.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 15:33:03,052 INFO [optim.py:369] (1/2) 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 15:33:16,846 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=729300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:33:17,284 INFO [train.py:968] (1/2) Epoch 16, batch 45350, giga_loss[loss=0.2577, simple_loss=0.3385, pruned_loss=0.08848, over 28469.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3693, pruned_loss=0.121, over 5637074.35 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.1189, over 5694400.20 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3692, pruned_loss=0.121, over 5630001.33 frames. ], batch size: 60, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:33:19,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=729303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:33:36,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7389, 1.8986, 1.3204, 1.4859], device='cuda:1'), covar=tensor([0.0937, 0.0677, 0.1087, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0449, 0.0510, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 15:33:45,078 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 45400, giga_loss[loss=0.2901, simple_loss=0.3632, pruned_loss=0.1085, over 28761.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3721, pruned_loss=0.1233, over 5648746.19 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3665, pruned_loss=0.1189, over 5697613.11 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.372, pruned_loss=0.1233, over 5639506.45 frames. ], batch size: 119, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:34:29,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1344, 1.3490, 1.2391, 1.0566], device='cuda:1'), covar=tensor([0.2034, 0.1928, 0.1396, 0.1825], device='cuda:1'), in_proj_covar=tensor([0.1866, 0.1808, 0.1733, 0.1872], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:34:41,412 INFO [optim.py:369] (1/2) 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 15:34:52,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4993, 3.3287, 3.1755, 1.9899], device='cuda:1'), covar=tensor([0.0730, 0.0870, 0.0866, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.1092, 0.0938, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 15:34:54,757 INFO [train.py:968] (1/2) Epoch 16, batch 45450, libri_loss[loss=0.3269, simple_loss=0.3926, pruned_loss=0.1307, over 29290.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1236, over 5654260.29 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3666, pruned_loss=0.119, over 5700831.08 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3732, pruned_loss=0.1236, over 5642779.10 frames. ], batch size: 97, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:34:59,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 15:35:25,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-08 15:35:40,034 INFO [train.py:968] (1/2) Epoch 16, batch 45500, giga_loss[loss=0.2912, simple_loss=0.3581, pruned_loss=0.1121, over 28973.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3758, pruned_loss=0.1259, over 5654571.76 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1193, over 5696388.91 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3755, pruned_loss=0.1258, over 5648551.87 frames. ], batch size: 213, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:36:09,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1981, 1.3078, 3.9805, 3.3229], device='cuda:1'), covar=tensor([0.2032, 0.2841, 0.0795, 0.1032], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0622, 0.0920, 0.0845], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 15:36:22,274 INFO [optim.py:369] (1/2) 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 15:36:35,370 INFO [train.py:968] (1/2) Epoch 16, batch 45550, giga_loss[loss=0.3233, simple_loss=0.3825, pruned_loss=0.132, over 28920.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3778, pruned_loss=0.1278, over 5655587.79 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1193, over 5697065.96 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3776, pruned_loss=0.1278, over 5650016.40 frames. ], batch size: 213, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:37:25,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-08 15:37:27,812 INFO [train.py:968] (1/2) Epoch 16, batch 45600, giga_loss[loss=0.341, simple_loss=0.3996, pruned_loss=0.1412, over 27585.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3783, pruned_loss=0.1262, over 5658553.60 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3675, pruned_loss=0.1195, over 5699655.52 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3779, pruned_loss=0.1261, over 5651056.57 frames. ], batch size: 472, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:38:05,152 INFO [optim.py:369] (1/2) 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 15:38:17,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=729598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:38:19,564 INFO [train.py:968] (1/2) Epoch 16, batch 45650, giga_loss[loss=0.296, simple_loss=0.368, pruned_loss=0.112, over 28940.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.377, pruned_loss=0.1253, over 5653711.75 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1193, over 5701499.42 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3773, pruned_loss=0.1256, over 5645087.47 frames. ], batch size: 136, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:39:09,221 INFO [train.py:968] (1/2) Epoch 16, batch 45700, giga_loss[loss=0.2839, simple_loss=0.3597, pruned_loss=0.104, over 28793.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3764, pruned_loss=0.1259, over 5644831.60 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3674, pruned_loss=0.1195, over 5693260.49 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3765, pruned_loss=0.1261, over 5644879.36 frames. ], batch size: 262, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:39:49,218 INFO [optim.py:369] (1/2) 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 15:40:03,788 INFO [train.py:968] (1/2) Epoch 16, batch 45750, giga_loss[loss=0.3029, simple_loss=0.3653, pruned_loss=0.1202, over 28922.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3763, pruned_loss=0.1268, over 5639283.43 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1193, over 5696058.94 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3767, pruned_loss=0.1272, over 5636364.77 frames. ], batch size: 213, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:40:16,873 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 15:40:40,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6759, 1.9071, 1.8432, 1.5678], device='cuda:1'), covar=tensor([0.1914, 0.2190, 0.2290, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0743, 0.0698, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 15:40:42,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=729741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:40:44,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=729744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:40:52,444 INFO [train.py:968] (1/2) Epoch 16, batch 45800, giga_loss[loss=0.3343, simple_loss=0.3919, pruned_loss=0.1384, over 27959.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3753, pruned_loss=0.1265, over 5641595.82 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5698233.39 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3757, pruned_loss=0.1268, over 5635801.12 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:41:16,726 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=729773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:41:30,148 INFO [optim.py:369] (1/2) 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 15:41:42,835 INFO [train.py:968] (1/2) Epoch 16, batch 45850, giga_loss[loss=0.383, simple_loss=0.4249, pruned_loss=0.1705, over 27848.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3756, pruned_loss=0.1278, over 5628225.97 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3676, pruned_loss=0.1195, over 5699621.85 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.376, pruned_loss=0.1282, over 5620776.22 frames. ], batch size: 412, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:42:26,346 INFO [train.py:968] (1/2) Epoch 16, batch 45900, giga_loss[loss=0.3127, simple_loss=0.3763, pruned_loss=0.1245, over 28648.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3748, pruned_loss=0.127, over 5644426.64 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1196, over 5702867.21 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3752, pruned_loss=0.1274, over 5634379.19 frames. ], batch size: 242, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:42:43,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8231, 2.1383, 2.0238, 1.6832], device='cuda:1'), covar=tensor([0.2841, 0.2158, 0.2101, 0.2434], device='cuda:1'), in_proj_covar=tensor([0.1876, 0.1818, 0.1744, 0.1874], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:42:59,854 INFO [optim.py:369] (1/2) 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 15:43:10,284 INFO [train.py:968] (1/2) Epoch 16, batch 45950, giga_loss[loss=0.3015, simple_loss=0.3655, pruned_loss=0.1187, over 29051.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3754, pruned_loss=0.1275, over 5650650.03 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.368, pruned_loss=0.1197, over 5706656.49 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3756, pruned_loss=0.128, over 5637519.04 frames. ], batch size: 128, lr: 1.97e-03, grad_scale: 2.0
+2023-03-08 15:43:37,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=729926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:44:04,048 INFO [train.py:968] (1/2) Epoch 16, batch 46000, giga_loss[loss=0.29, simple_loss=0.363, pruned_loss=0.1085, over 28911.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3768, pruned_loss=0.1288, over 5648919.58 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1202, over 5704675.03 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3766, pruned_loss=0.1288, over 5639954.89 frames. ], batch size: 186, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:44:35,473 INFO [optim.py:369] (1/2) 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 15:44:47,052 INFO [train.py:968] (1/2) Epoch 16, batch 46050, giga_loss[loss=0.2905, simple_loss=0.3571, pruned_loss=0.1119, over 28403.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3777, pruned_loss=0.1291, over 5651181.74 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3682, pruned_loss=0.1199, over 5709864.71 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3781, pruned_loss=0.1297, over 5637532.31 frames. ], batch size: 77, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:45:20,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5219, 3.3671, 1.6089, 1.5904], device='cuda:1'), covar=tensor([0.0918, 0.0355, 0.0846, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0538, 0.0364, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 15:45:30,608 INFO [train.py:968] (1/2) Epoch 16, batch 46100, giga_loss[loss=0.3209, simple_loss=0.3814, pruned_loss=0.1302, over 28747.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3781, pruned_loss=0.1298, over 5646284.76 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1201, over 5691538.63 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3784, pruned_loss=0.1303, over 5650441.32 frames. ], batch size: 284, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:45:44,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3171, 2.9834, 1.4117, 1.5590], device='cuda:1'), covar=tensor([0.0959, 0.0381, 0.0889, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0538, 0.0364, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 15:46:09,442 INFO [optim.py:369] (1/2) 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 15:46:20,594 INFO [train.py:968] (1/2) Epoch 16, batch 46150, giga_loss[loss=0.2713, simple_loss=0.3412, pruned_loss=0.1007, over 28819.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3775, pruned_loss=0.1295, over 5649155.06 frames. ], libri_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5695695.91 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3783, pruned_loss=0.1304, over 5647587.34 frames. ], batch size: 99, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:46:28,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8195, 1.9539, 1.7915, 1.6597], device='cuda:1'), covar=tensor([0.1608, 0.2159, 0.2014, 0.2085], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0743, 0.0696, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 15:46:44,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2927, 1.9261, 1.5920, 1.4121], device='cuda:1'), covar=tensor([0.0758, 0.0309, 0.0301, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 15:47:07,394 INFO [train.py:968] (1/2) Epoch 16, batch 46200, giga_loss[loss=0.2975, simple_loss=0.3623, pruned_loss=0.1163, over 28988.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3767, pruned_loss=0.1286, over 5652476.42 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3678, pruned_loss=0.1195, over 5698404.28 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3777, pruned_loss=0.1297, over 5648090.14 frames. ], batch size: 136, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:47:21,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-08 15:47:39,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5624, 1.8507, 1.5310, 1.4562], device='cuda:1'), covar=tensor([0.2193, 0.2183, 0.2372, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1421, 0.1036, 0.1259, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 15:47:41,767 INFO [optim.py:369] (1/2) 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 15:47:44,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2144, 1.1264, 3.4188, 3.1299], device='cuda:1'), covar=tensor([0.1565, 0.2732, 0.0566, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0618, 0.0915, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 15:47:51,506 INFO [train.py:968] (1/2) Epoch 16, batch 46250, giga_loss[loss=0.297, simple_loss=0.3679, pruned_loss=0.1131, over 28640.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.378, pruned_loss=0.1291, over 5653905.44 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3683, pruned_loss=0.1198, over 5700082.93 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3786, pruned_loss=0.13, over 5647370.05 frames. ], batch size: 92, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:47:57,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8176, 1.9475, 1.9769, 1.6777], device='cuda:1'), covar=tensor([0.1590, 0.1903, 0.1786, 0.1796], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0748, 0.0700, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 15:48:41,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:48:42,711 INFO [train.py:968] (1/2) Epoch 16, batch 46300, giga_loss[loss=0.4673, simple_loss=0.4684, pruned_loss=0.2331, over 27497.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3787, pruned_loss=0.1292, over 5663055.68 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1197, over 5704536.08 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3796, pruned_loss=0.1303, over 5652693.61 frames. ], batch size: 472, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:49:07,861 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:49:31,817 INFO [train.py:968] (1/2) Epoch 16, batch 46350, libri_loss[loss=0.2906, simple_loss=0.3468, pruned_loss=0.1171, over 29631.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3795, pruned_loss=0.1297, over 5673934.84 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1196, over 5701038.02 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.381, pruned_loss=0.1311, over 5666853.76 frames. ], batch size: 69, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:49:32,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:50:13,393 INFO [train.py:968] (1/2) Epoch 16, batch 46400, giga_loss[loss=0.2573, simple_loss=0.3382, pruned_loss=0.08822, over 28836.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3798, pruned_loss=0.1296, over 5687853.76 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.12, over 5702833.27 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3812, pruned_loss=0.1308, over 5679983.29 frames. ], batch size: 174, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:50:48,277 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 46450, giga_loss[loss=0.293, simple_loss=0.3658, pruned_loss=0.1102, over 28843.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.38, pruned_loss=0.1297, over 5687113.23 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1197, over 5703855.05 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3818, pruned_loss=0.1314, over 5679423.84 frames. ], batch size: 174, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:51:30,228 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:51:39,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=730444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:51:41,974 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 46500, giga_loss[loss=0.2958, simple_loss=0.3724, pruned_loss=0.1096, over 28977.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3797, pruned_loss=0.1283, over 5677053.25 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3678, pruned_loss=0.1195, over 5698938.98 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3816, pruned_loss=0.1299, over 5675665.88 frames. ], batch size: 155, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:52:11,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3432, 1.5868, 1.4910, 1.4450], device='cuda:1'), covar=tensor([0.0661, 0.0285, 0.0267, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 15:52:12,062 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:52:25,023 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 46550, giga_loss[loss=0.3029, simple_loss=0.3792, pruned_loss=0.1133, over 28912.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3784, pruned_loss=0.1267, over 5673755.03 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3676, pruned_loss=0.1195, over 5696814.78 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3806, pruned_loss=0.1285, over 5673073.52 frames. ], batch size: 164, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:53:16,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6432, 1.5697, 1.2993, 1.1910], device='cuda:1'), covar=tensor([0.0669, 0.0440, 0.0804, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0451, 0.0512, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 15:53:25,022 INFO [train.py:968] (1/2) Epoch 16, batch 46600, giga_loss[loss=0.3024, simple_loss=0.372, pruned_loss=0.1164, over 28835.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3767, pruned_loss=0.1258, over 5663378.60 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3673, pruned_loss=0.1192, over 5697338.03 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.379, pruned_loss=0.1277, over 5661943.09 frames. ], batch size: 99, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:54:04,862 INFO [optim.py:369] (1/2) 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 15:54:11,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6112, 2.4402, 1.7071, 0.7324], device='cuda:1'), covar=tensor([0.5017, 0.2647, 0.3656, 0.5756], device='cuda:1'), in_proj_covar=tensor([0.1685, 0.1592, 0.1554, 0.1374], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 15:54:12,917 INFO [train.py:968] (1/2) Epoch 16, batch 46650, giga_loss[loss=0.3887, simple_loss=0.422, pruned_loss=0.1777, over 26511.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3768, pruned_loss=0.1262, over 5675996.45 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3674, pruned_loss=0.1191, over 5703664.59 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3788, pruned_loss=0.1279, over 5668286.49 frames. ], batch size: 555, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:54:34,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:55:00,327 INFO [train.py:968] (1/2) Epoch 16, batch 46700, giga_loss[loss=0.3968, simple_loss=0.4127, pruned_loss=0.1905, over 23722.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3785, pruned_loss=0.1274, over 5662320.98 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3676, pruned_loss=0.1191, over 5698793.91 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3802, pruned_loss=0.1288, over 5660546.39 frames. ], batch size: 705, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:55:02,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:55:03,443 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:55:38,342 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 16, batch 46750, giga_loss[loss=0.2708, simple_loss=0.3391, pruned_loss=0.1013, over 28902.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3793, pruned_loss=0.128, over 5659987.32 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1192, over 5692088.51 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3808, pruned_loss=0.1293, over 5664594.33 frames. ], batch size: 145, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:56:38,808 INFO [train.py:968] (1/2) Epoch 16, batch 46800, giga_loss[loss=0.3247, simple_loss=0.3828, pruned_loss=0.1333, over 28298.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3792, pruned_loss=0.1293, over 5631892.17 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1197, over 5655769.90 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3801, pruned_loss=0.13, over 5668643.71 frames. ], batch size: 368, lr: 1.97e-03, grad_scale: 8.0
+2023-03-08 15:56:58,167 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=730770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:01,828 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=730772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:07,895 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-08 15:57:18,045 INFO [optim.py:369] (1/2) 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 15:57:24,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=730796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:27,467 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:27,534 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 16, batch 46850, giga_loss[loss=0.3444, simple_loss=0.3992, pruned_loss=0.1449, over 28597.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3808, pruned_loss=0.1313, over 5583663.00 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1202, over 5612588.93 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3811, pruned_loss=0.1315, over 5649650.77 frames. ], batch size: 262, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:57:37,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=730808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:57:55,248 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 15:58:04,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2178, 1.4577, 1.3262, 1.0506], device='cuda:1'), covar=tensor([0.2209, 0.2174, 0.1382, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1809, 0.1729, 0.1868], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 15:58:17,788 INFO [train.py:968] (1/2) Epoch 16, batch 46900, giga_loss[loss=0.3492, simple_loss=0.4116, pruned_loss=0.1434, over 28810.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3823, pruned_loss=0.1321, over 5554675.21 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3692, pruned_loss=0.1207, over 5570616.78 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3822, pruned_loss=0.132, over 5645257.54 frames. ], batch size: 284, lr: 1.97e-03, grad_scale: 4.0
+2023-03-08 15:58:39,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-08 15:58:45,430 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-08 15:59:33,269 INFO [optim.py:369] (1/2) 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 16:00:07,031 INFO [train.py:968] (1/2) Epoch 17, batch 50, giga_loss[loss=0.3133, simple_loss=0.3895, pruned_loss=0.1185, over 28692.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3767, pruned_loss=0.1119, over 1260431.74 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.343, pruned_loss=0.09294, over 193393.77 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3824, pruned_loss=0.1151, over 1104715.67 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:00:25,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=730951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:00:29,018 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 100, giga_loss[loss=0.2809, simple_loss=0.3512, pruned_loss=0.1053, over 27581.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3683, pruned_loss=0.1087, over 2237611.50 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3424, pruned_loss=0.0928, over 362717.62 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3725, pruned_loss=0.1113, over 2001648.86 frames. ], batch size: 472, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:00:57,426 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=730983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:01:04,461 INFO [optim.py:369] (1/2) 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 16:01:14,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6164, 1.6963, 1.5909, 1.4303], device='cuda:1'), covar=tensor([0.2460, 0.2326, 0.1901, 0.2225], device='cuda:1'), in_proj_covar=tensor([0.1870, 0.1806, 0.1727, 0.1865], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 16:01:31,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 150, giga_loss[loss=0.2451, simple_loss=0.3192, pruned_loss=0.08546, over 28871.00 frames. ], tot_loss[loss=0.276, simple_loss=0.352, pruned_loss=0.09999, over 3000001.70 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3362, pruned_loss=0.08848, over 589232.99 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3557, pruned_loss=0.1025, over 2691417.27 frames. ], batch size: 199, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:01:37,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:01:47,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-08 16:02:16,370 INFO [train.py:968] (1/2) Epoch 17, batch 200, giga_loss[loss=0.2417, simple_loss=0.3224, pruned_loss=0.08046, over 28917.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3405, pruned_loss=0.09425, over 3596451.07 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3401, pruned_loss=0.08923, over 841603.70 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3416, pruned_loss=0.0959, over 3233543.91 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:02:27,647 INFO [optim.py:369] (1/2) 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 16:02:38,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6069, 1.8374, 1.8759, 1.6216], device='cuda:1'), covar=tensor([0.1968, 0.2074, 0.2260, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0744, 0.0703, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 16:02:57,241 INFO [train.py:968] (1/2) Epoch 17, batch 250, giga_loss[loss=0.2057, simple_loss=0.2886, pruned_loss=0.06136, over 28765.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.331, pruned_loss=0.08966, over 4069752.90 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3405, pruned_loss=0.08956, over 966774.68 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3307, pruned_loss=0.09038, over 3734082.83 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:03:14,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:03:14,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4967, 1.5074, 1.2281, 1.1477], device='cuda:1'), covar=tensor([0.0700, 0.0336, 0.0854, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0446, 0.0509, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 16:03:20,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=731154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:03:25,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8037, 4.6537, 4.3720, 2.1112], device='cuda:1'), covar=tensor([0.0544, 0.0753, 0.0772, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.1158, 0.1078, 0.0929, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 16:03:35,332 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 17, batch 300, giga_loss[loss=0.2172, simple_loss=0.2885, pruned_loss=0.07292, over 28993.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3229, pruned_loss=0.08623, over 4416269.60 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3387, pruned_loss=0.08858, over 1124758.83 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3219, pruned_loss=0.08667, over 4105325.56 frames. ], batch size: 106, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:03:51,726 INFO [optim.py:369] (1/2) 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 16:04:01,462 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 350, giga_loss[loss=0.2576, simple_loss=0.3235, pruned_loss=0.09584, over 28524.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.3149, pruned_loss=0.08265, over 4687550.81 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3368, pruned_loss=0.0872, over 1277620.51 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3135, pruned_loss=0.08297, over 4406073.83 frames. ], batch size: 336, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:05:09,509 INFO [train.py:968] (1/2) Epoch 17, batch 400, giga_loss[loss=0.2306, simple_loss=0.307, pruned_loss=0.07711, over 29020.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3105, pruned_loss=0.08075, over 4911737.58 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3367, pruned_loss=0.08694, over 1346707.07 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.3088, pruned_loss=0.08086, over 4674510.14 frames. ], batch size: 155, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:05:10,766 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=731280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:05:18,839 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:05:20,146 INFO [optim.py:369] (1/2) 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 16:05:21,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=731293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:05:47,084 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 450, giga_loss[loss=0.2348, simple_loss=0.3093, pruned_loss=0.08012, over 28813.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3087, pruned_loss=0.07994, over 5090741.33 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3376, pruned_loss=0.08741, over 1437408.66 frames. ], giga_tot_loss[loss=0.233, simple_loss=0.3065, pruned_loss=0.07972, over 4886133.38 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:06:25,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 16:06:28,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-08 16:06:33,682 INFO [train.py:968] (1/2) Epoch 17, batch 500, giga_loss[loss=0.2548, simple_loss=0.3208, pruned_loss=0.09438, over 27957.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3068, pruned_loss=0.07939, over 5218056.89 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3377, pruned_loss=0.08785, over 1588214.17 frames. ], giga_tot_loss[loss=0.2308, simple_loss=0.3039, pruned_loss=0.0788, over 5032850.51 frames. ], batch size: 412, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:06:38,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-08 16:06:42,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5744, 1.7362, 1.8372, 1.3952], device='cuda:1'), covar=tensor([0.1919, 0.2643, 0.1587, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0700, 0.0925, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 16:06:43,423 INFO [optim.py:369] (1/2) 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 16:06:50,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:07:11,204 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-08 16:07:18,492 INFO [train.py:968] (1/2) Epoch 17, batch 550, libri_loss[loss=0.2563, simple_loss=0.325, pruned_loss=0.09382, over 28999.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3047, pruned_loss=0.07809, over 5323280.90 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3378, pruned_loss=0.08753, over 1732430.32 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.3013, pruned_loss=0.07738, over 5158432.08 frames. ], batch size: 64, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:07:50,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8042, 2.2768, 2.0356, 1.5825], device='cuda:1'), covar=tensor([0.3170, 0.1979, 0.2121, 0.2706], device='cuda:1'), in_proj_covar=tensor([0.1860, 0.1794, 0.1713, 0.1850], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 16:07:56,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3309, 1.6122, 1.6110, 1.2052], device='cuda:1'), covar=tensor([0.1768, 0.2353, 0.1438, 0.1716], device='cuda:1'), in_proj_covar=tensor([0.0881, 0.0702, 0.0926, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 16:08:03,822 INFO [train.py:968] (1/2) Epoch 17, batch 600, libri_loss[loss=0.2991, simple_loss=0.3718, pruned_loss=0.1132, over 20137.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3021, pruned_loss=0.07656, over 5394793.15 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3383, pruned_loss=0.08774, over 1826090.17 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.2984, pruned_loss=0.07565, over 5260845.50 frames. ], batch size: 187, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:08:15,088 INFO [optim.py:369] (1/2) 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 16:08:52,488 INFO [train.py:968] (1/2) Epoch 17, batch 650, giga_loss[loss=0.1852, simple_loss=0.2621, pruned_loss=0.05413, over 28207.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2997, pruned_loss=0.07585, over 5463845.37 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3389, pruned_loss=0.08817, over 1886060.35 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2961, pruned_loss=0.07484, over 5352583.61 frames. ], batch size: 77, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:08:53,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:09:06,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:09:09,054 INFO [zipformer.py:1188] (1/2) 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] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-08 16:09:35,760 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 700, giga_loss[loss=0.193, simple_loss=0.2766, pruned_loss=0.05471, over 29033.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2968, pruned_loss=0.07437, over 5517010.79 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3397, pruned_loss=0.08877, over 1906176.05 frames. ], giga_tot_loss[loss=0.2201, simple_loss=0.2935, pruned_loss=0.07332, over 5428180.75 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:09:53,582 INFO [optim.py:369] (1/2) 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 16:09:55,836 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 17, batch 750, giga_loss[loss=0.1995, simple_loss=0.2736, pruned_loss=0.06269, over 27631.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2945, pruned_loss=0.07301, over 5554152.04 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3405, pruned_loss=0.08897, over 2001368.91 frames. ], giga_tot_loss[loss=0.217, simple_loss=0.2906, pruned_loss=0.07175, over 5478830.25 frames. ], batch size: 472, lr: 1.91e-03, grad_scale: 2.0
+2023-03-08 16:10:51,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=731655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:11:03,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=731672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:11:06,907 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 800, giga_loss[loss=0.1902, simple_loss=0.2584, pruned_loss=0.06099, over 28631.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2924, pruned_loss=0.07244, over 5575133.33 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3406, pruned_loss=0.08912, over 2087921.15 frames. ], giga_tot_loss[loss=0.2151, simple_loss=0.2882, pruned_loss=0.07101, over 5515967.07 frames. ], batch size: 85, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:11:21,658 INFO [optim.py:369] (1/2) 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 16:11:35,044 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=731704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:11:57,820 INFO [train.py:968] (1/2) Epoch 17, batch 850, giga_loss[loss=0.304, simple_loss=0.3767, pruned_loss=0.1156, over 28720.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.2986, pruned_loss=0.07578, over 5597567.04 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3416, pruned_loss=0.08956, over 2180410.95 frames. ], giga_tot_loss[loss=0.2211, simple_loss=0.2939, pruned_loss=0.07415, over 5544970.44 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:12:50,589 INFO [train.py:968] (1/2) Epoch 17, batch 900, giga_loss[loss=0.3239, simple_loss=0.3931, pruned_loss=0.1273, over 28952.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3133, pruned_loss=0.08369, over 5618386.08 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3411, pruned_loss=0.08931, over 2216278.90 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3096, pruned_loss=0.08247, over 5575448.13 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:13:02,560 INFO [optim.py:369] (1/2) 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 16:13:06,589 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=731801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:13:34,036 INFO [train.py:968] (1/2) Epoch 17, batch 950, giga_loss[loss=0.3416, simple_loss=0.3963, pruned_loss=0.1434, over 27551.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.325, pruned_loss=0.08926, over 5637334.17 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3417, pruned_loss=0.08954, over 2289106.57 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3215, pruned_loss=0.08818, over 5598869.59 frames. ], batch size: 472, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:13:36,963 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=731830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:13:48,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9597, 1.3019, 1.0931, 0.1587], device='cuda:1'), covar=tensor([0.3608, 0.2815, 0.4294, 0.5631], device='cuda:1'), in_proj_covar=tensor([0.1667, 0.1573, 0.1544, 0.1358], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 16:14:17,017 INFO [train.py:968] (1/2) Epoch 17, batch 1000, giga_loss[loss=0.2576, simple_loss=0.3354, pruned_loss=0.08986, over 28678.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3326, pruned_loss=0.09231, over 5642441.55 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3415, pruned_loss=0.08972, over 2384515.34 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3296, pruned_loss=0.09144, over 5615682.66 frames. ], batch size: 85, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:14:18,413 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:14:48,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-08 16:14:55,696 INFO [train.py:968] (1/2) Epoch 17, batch 1050, giga_loss[loss=0.2821, simple_loss=0.3636, pruned_loss=0.1003, over 28651.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3376, pruned_loss=0.09356, over 5659088.43 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3421, pruned_loss=0.09045, over 2538931.26 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3346, pruned_loss=0.0927, over 5627891.21 frames. ], batch size: 242, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:15:32,799 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 16:15:44,715 INFO [train.py:968] (1/2) Epoch 17, batch 1100, libri_loss[loss=0.2213, simple_loss=0.3031, pruned_loss=0.06971, over 29681.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3387, pruned_loss=0.09298, over 5656843.67 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3418, pruned_loss=0.09032, over 2603711.72 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3365, pruned_loss=0.09241, over 5629301.21 frames. ], batch size: 73, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:15:54,682 INFO [optim.py:369] (1/2) 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 16:16:24,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2965, 1.6938, 1.4425, 1.4684], device='cuda:1'), covar=tensor([0.0743, 0.0405, 0.0326, 0.0753], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 16:16:27,114 INFO [train.py:968] (1/2) Epoch 17, batch 1150, giga_loss[loss=0.3166, simple_loss=0.3598, pruned_loss=0.1367, over 23674.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3414, pruned_loss=0.09451, over 5656049.31 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.342, pruned_loss=0.09039, over 2647516.07 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3395, pruned_loss=0.09408, over 5635630.95 frames. ], batch size: 705, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:17:10,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6808, 1.7522, 1.3242, 1.3138], device='cuda:1'), covar=tensor([0.0899, 0.0568, 0.0989, 0.1122], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0445, 0.0511, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 16:17:13,245 INFO [train.py:968] (1/2) Epoch 17, batch 1200, giga_loss[loss=0.2818, simple_loss=0.3608, pruned_loss=0.1014, over 28770.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3443, pruned_loss=0.09687, over 5668592.94 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3421, pruned_loss=0.09059, over 2696146.56 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3428, pruned_loss=0.09654, over 5649011.30 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:17:24,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7378, 1.8540, 1.7856, 1.7135], device='cuda:1'), covar=tensor([0.1882, 0.2215, 0.2428, 0.2016], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0741, 0.0699, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 16:17:25,009 INFO [optim.py:369] (1/2) 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 16:17:56,216 INFO [train.py:968] (1/2) Epoch 17, batch 1250, libri_loss[loss=0.3042, simple_loss=0.3764, pruned_loss=0.116, over 19440.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3492, pruned_loss=0.1001, over 5666156.58 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.344, pruned_loss=0.09172, over 2794301.58 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3473, pruned_loss=0.0996, over 5655171.15 frames. ], batch size: 187, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:18:33,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1363, 0.7807, 0.8917, 1.2977], device='cuda:1'), covar=tensor([0.0845, 0.0395, 0.0380, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 16:18:34,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 16:18:40,024 INFO [train.py:968] (1/2) Epoch 17, batch 1300, giga_loss[loss=0.3082, simple_loss=0.3864, pruned_loss=0.115, over 28743.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3518, pruned_loss=0.1005, over 5675269.12 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3446, pruned_loss=0.09204, over 2861320.23 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3501, pruned_loss=0.1001, over 5669904.77 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:18:52,319 INFO [optim.py:369] (1/2) 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 16:19:22,151 INFO [train.py:968] (1/2) Epoch 17, batch 1350, libri_loss[loss=0.2532, simple_loss=0.3271, pruned_loss=0.08962, over 29580.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3544, pruned_loss=0.1019, over 5667033.63 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3447, pruned_loss=0.09224, over 2881883.45 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3531, pruned_loss=0.1016, over 5669299.66 frames. ], batch size: 74, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:19:41,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6683, 4.7672, 1.6103, 2.0871], device='cuda:1'), covar=tensor([0.0895, 0.0203, 0.0890, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0532, 0.0363, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 16:19:44,378 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 1400, giga_loss[loss=0.2944, simple_loss=0.3788, pruned_loss=0.1051, over 28884.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3562, pruned_loss=0.1024, over 5679197.31 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3448, pruned_loss=0.09223, over 2971418.10 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3554, pruned_loss=0.1024, over 5674681.05 frames. ], batch size: 145, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:20:15,107 INFO [optim.py:369] (1/2) 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 16:20:35,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 16:20:45,541 INFO [train.py:968] (1/2) Epoch 17, batch 1450, giga_loss[loss=0.2697, simple_loss=0.3562, pruned_loss=0.09165, over 28682.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3546, pruned_loss=0.1001, over 5690340.00 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3444, pruned_loss=0.09191, over 3029804.97 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3544, pruned_loss=0.1004, over 5682512.52 frames. ], batch size: 242, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:21:08,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:21:29,474 INFO [train.py:968] (1/2) Epoch 17, batch 1500, giga_loss[loss=0.2768, simple_loss=0.3552, pruned_loss=0.09923, over 28244.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.353, pruned_loss=0.09828, over 5694461.57 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3435, pruned_loss=0.09137, over 3058622.87 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3533, pruned_loss=0.09887, over 5686415.51 frames. ], batch size: 368, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:21:42,601 INFO [optim.py:369] (1/2) 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 16:21:46,134 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=732399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:21:47,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=732402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:21:52,073 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:22:08,971 INFO [train.py:968] (1/2) Epoch 17, batch 1550, libri_loss[loss=0.2719, simple_loss=0.3584, pruned_loss=0.09272, over 28612.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3511, pruned_loss=0.09589, over 5709740.18 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.344, pruned_loss=0.09111, over 3155614.45 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3514, pruned_loss=0.09666, over 5697885.57 frames. ], batch size: 106, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:22:10,759 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=732431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:22:52,608 INFO [train.py:968] (1/2) Epoch 17, batch 1600, giga_loss[loss=0.2981, simple_loss=0.3597, pruned_loss=0.1182, over 28931.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3529, pruned_loss=0.09851, over 5688702.72 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3444, pruned_loss=0.09137, over 3213568.31 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3531, pruned_loss=0.09916, over 5683123.69 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:23:06,078 INFO [optim.py:369] (1/2) 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 16:23:13,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9102, 1.1676, 1.3443, 1.0198], device='cuda:1'), covar=tensor([0.1965, 0.1510, 0.2294, 0.1700], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0734, 0.0692, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 16:23:36,016 INFO [train.py:968] (1/2) Epoch 17, batch 1650, giga_loss[loss=0.2924, simple_loss=0.364, pruned_loss=0.1104, over 28613.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1012, over 5696659.64 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.344, pruned_loss=0.09121, over 3314240.50 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.355, pruned_loss=0.1021, over 5689106.78 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:24:23,846 INFO [train.py:968] (1/2) Epoch 17, batch 1700, giga_loss[loss=0.273, simple_loss=0.3444, pruned_loss=0.1008, over 28975.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.356, pruned_loss=0.1041, over 5703928.66 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3442, pruned_loss=0.09127, over 3325468.29 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3565, pruned_loss=0.1048, over 5698674.86 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:24:37,546 INFO [optim.py:369] (1/2) 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 16:25:07,688 INFO [train.py:968] (1/2) Epoch 17, batch 1750, giga_loss[loss=0.2649, simple_loss=0.3383, pruned_loss=0.0957, over 28608.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3548, pruned_loss=0.104, over 5700671.52 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3448, pruned_loss=0.09169, over 3408620.35 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3552, pruned_loss=0.1048, over 5695294.23 frames. ], batch size: 307, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:25:49,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2596, 1.3508, 1.2009, 1.2362], device='cuda:1'), covar=tensor([0.2164, 0.2023, 0.1960, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.1842, 0.1789, 0.1714, 0.1854], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 16:25:51,748 INFO [train.py:968] (1/2) Epoch 17, batch 1800, giga_loss[loss=0.2679, simple_loss=0.3447, pruned_loss=0.09558, over 28764.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3526, pruned_loss=0.1036, over 5689954.41 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3453, pruned_loss=0.09202, over 3469554.16 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3528, pruned_loss=0.1043, over 5681671.60 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:26:03,834 INFO [optim.py:369] (1/2) 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 16:26:18,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7642, 1.0481, 2.8881, 2.6468], device='cuda:1'), covar=tensor([0.1722, 0.2571, 0.0569, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0714, 0.0616, 0.0905, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 16:26:32,316 INFO [train.py:968] (1/2) Epoch 17, batch 1850, giga_loss[loss=0.2933, simple_loss=0.372, pruned_loss=0.1073, over 28854.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3517, pruned_loss=0.1025, over 5694878.57 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3455, pruned_loss=0.09202, over 3530070.01 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3519, pruned_loss=0.1033, over 5683587.59 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:26:35,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=732731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:27:16,787 INFO [train.py:968] (1/2) Epoch 17, batch 1900, giga_loss[loss=0.2342, simple_loss=0.3211, pruned_loss=0.07365, over 28677.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.351, pruned_loss=0.1014, over 5691771.02 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3459, pruned_loss=0.09206, over 3583819.27 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3511, pruned_loss=0.1022, over 5682508.19 frames. ], batch size: 242, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:27:22,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=732783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:27:30,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4086, 1.8720, 1.6180, 1.5704], device='cuda:1'), covar=tensor([0.2277, 0.2084, 0.2402, 0.2194], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0738, 0.0694, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 16:27:31,669 INFO [optim.py:369] (1/2) 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 16:27:56,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7687, 1.8363, 1.4308, 1.4134], device='cuda:1'), covar=tensor([0.0888, 0.0584, 0.0988, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0444, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 16:28:04,034 INFO [train.py:968] (1/2) Epoch 17, batch 1950, giga_loss[loss=0.2344, simple_loss=0.3141, pruned_loss=0.07738, over 28591.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.347, pruned_loss=0.09901, over 5691677.92 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3457, pruned_loss=0.09195, over 3641638.00 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3473, pruned_loss=0.09992, over 5679072.57 frames. ], batch size: 307, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:28:09,934 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:28:48,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3721, 1.6404, 1.6670, 1.2667], device='cuda:1'), covar=tensor([0.1451, 0.1975, 0.1129, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0698, 0.0924, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 16:28:49,208 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=732874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:28:51,887 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 2000, giga_loss[loss=0.2166, simple_loss=0.3043, pruned_loss=0.06444, over 28975.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3407, pruned_loss=0.09566, over 5683608.28 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3456, pruned_loss=0.09188, over 3664273.77 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3409, pruned_loss=0.09646, over 5671826.26 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:28:57,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-08 16:29:07,480 INFO [optim.py:369] (1/2) 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 16:29:17,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=732906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:29:36,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=732926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:29:38,346 INFO [train.py:968] (1/2) Epoch 17, batch 2050, giga_loss[loss=0.2326, simple_loss=0.3062, pruned_loss=0.07951, over 28904.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3356, pruned_loss=0.09304, over 5685233.77 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3458, pruned_loss=0.09212, over 3731070.94 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3355, pruned_loss=0.09361, over 5669455.02 frames. ], batch size: 106, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:29:39,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=732929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:29:44,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=732934.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 16:29:51,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8661, 3.7036, 3.4847, 1.5803], device='cuda:1'), covar=tensor([0.0673, 0.0772, 0.0712, 0.2348], device='cuda:1'), in_proj_covar=tensor([0.1135, 0.1054, 0.0906, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 16:30:05,339 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 2100, giga_loss[loss=0.2326, simple_loss=0.3122, pruned_loss=0.07651, over 28837.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.333, pruned_loss=0.09221, over 5667593.43 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3457, pruned_loss=0.09197, over 3794756.76 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3327, pruned_loss=0.09276, over 5649630.70 frames. ], batch size: 186, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:30:38,914 INFO [optim.py:369] (1/2) 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 16:31:06,664 INFO [train.py:968] (1/2) Epoch 17, batch 2150, giga_loss[loss=0.262, simple_loss=0.3407, pruned_loss=0.09168, over 27879.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3348, pruned_loss=0.09241, over 5672816.72 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.346, pruned_loss=0.092, over 3836720.50 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3341, pruned_loss=0.09285, over 5661613.70 frames. ], batch size: 412, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:31:46,779 INFO [train.py:968] (1/2) Epoch 17, batch 2200, giga_loss[loss=0.381, simple_loss=0.4164, pruned_loss=0.1728, over 26689.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3342, pruned_loss=0.09162, over 5689101.05 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3462, pruned_loss=0.09198, over 3887137.62 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3332, pruned_loss=0.09198, over 5675762.67 frames. ], batch size: 555, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:31:50,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-08 16:32:02,189 INFO [optim.py:369] (1/2) 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 16:32:28,952 INFO [train.py:968] (1/2) Epoch 17, batch 2250, giga_loss[loss=0.2455, simple_loss=0.3213, pruned_loss=0.08487, over 28126.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3314, pruned_loss=0.09009, over 5693318.46 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3463, pruned_loss=0.09187, over 3925500.44 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3303, pruned_loss=0.09042, over 5680480.09 frames. ], batch size: 77, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:33:11,315 INFO [train.py:968] (1/2) Epoch 17, batch 2300, libri_loss[loss=0.2322, simple_loss=0.3159, pruned_loss=0.07422, over 29423.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3289, pruned_loss=0.0886, over 5704028.42 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3459, pruned_loss=0.09141, over 3983003.71 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3278, pruned_loss=0.08909, over 5689263.20 frames. ], batch size: 67, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:33:25,665 INFO [optim.py:369] (1/2) 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 16:33:35,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 2350, giga_loss[loss=0.2656, simple_loss=0.3446, pruned_loss=0.09332, over 28268.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3267, pruned_loss=0.08748, over 5696222.56 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.346, pruned_loss=0.09119, over 4026776.61 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3253, pruned_loss=0.08792, over 5690934.26 frames. ], batch size: 368, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:33:53,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6248, 1.7908, 1.3261, 1.3787], device='cuda:1'), covar=tensor([0.0940, 0.0569, 0.1020, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0441, 0.0506, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 16:34:15,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-08 16:34:33,796 INFO [train.py:968] (1/2) Epoch 17, batch 2400, giga_loss[loss=0.2345, simple_loss=0.3089, pruned_loss=0.08001, over 28638.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3244, pruned_loss=0.08651, over 5693978.04 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.346, pruned_loss=0.091, over 4063804.64 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3229, pruned_loss=0.08689, over 5693432.06 frames. ], batch size: 78, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:34:48,211 INFO [optim.py:369] (1/2) 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 16:34:49,099 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=733297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:34:58,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=733309.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:35:00,833 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-08 16:35:14,321 INFO [train.py:968] (1/2) Epoch 17, batch 2450, giga_loss[loss=0.2414, simple_loss=0.3081, pruned_loss=0.08737, over 28929.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3237, pruned_loss=0.08647, over 5700711.78 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3469, pruned_loss=0.09135, over 4125221.27 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3211, pruned_loss=0.08639, over 5695364.56 frames. ], batch size: 106, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:35:16,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7902, 1.9622, 1.3698, 1.5175], device='cuda:1'), covar=tensor([0.0934, 0.0617, 0.1046, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0440, 0.0505, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 16:35:19,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8715, 1.1429, 2.7864, 2.4950], device='cuda:1'), covar=tensor([0.1230, 0.2052, 0.0437, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0715, 0.0615, 0.0903, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 16:35:31,299 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=733353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:35:33,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=733356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:35:49,922 INFO [train.py:968] (1/2) Epoch 17, batch 2500, giga_loss[loss=0.2426, simple_loss=0.3106, pruned_loss=0.08734, over 28927.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3227, pruned_loss=0.08634, over 5704491.38 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3476, pruned_loss=0.09166, over 4176213.85 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3194, pruned_loss=0.08591, over 5702666.49 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:35:55,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=733385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:36:04,407 INFO [optim.py:369] (1/2) 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 16:36:30,277 INFO [train.py:968] (1/2) Epoch 17, batch 2550, giga_loss[loss=0.2188, simple_loss=0.2977, pruned_loss=0.06993, over 29133.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3209, pruned_loss=0.08502, over 5697714.33 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3482, pruned_loss=0.09184, over 4226790.42 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3168, pruned_loss=0.08429, over 5712680.87 frames. ], batch size: 128, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:36:47,958 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=733452.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:36:51,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=733455.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:36:57,159 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2660, 1.6392, 1.5731, 1.0886], device='cuda:1'), covar=tensor([0.1802, 0.2822, 0.1638, 0.1908], device='cuda:1'), in_proj_covar=tensor([0.0878, 0.0699, 0.0926, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 16:37:09,076 INFO [train.py:968] (1/2) Epoch 17, batch 2600, giga_loss[loss=0.2363, simple_loss=0.305, pruned_loss=0.08378, over 28776.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3206, pruned_loss=0.08498, over 5695922.96 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3487, pruned_loss=0.09189, over 4272821.45 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3161, pruned_loss=0.08417, over 5711725.54 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:37:12,653 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=733484.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:37:23,641 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 2650, giga_loss[loss=0.2296, simple_loss=0.3015, pruned_loss=0.07889, over 28831.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3197, pruned_loss=0.08463, over 5707084.66 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3492, pruned_loss=0.09204, over 4312792.74 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.3151, pruned_loss=0.08371, over 5715569.67 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:38:30,014 INFO [train.py:968] (1/2) Epoch 17, batch 2700, giga_loss[loss=0.2912, simple_loss=0.3538, pruned_loss=0.1143, over 28984.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3203, pruned_loss=0.08524, over 5713491.82 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3491, pruned_loss=0.09194, over 4336312.92 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3164, pruned_loss=0.08448, over 5717823.13 frames. ], batch size: 155, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:38:47,926 INFO [optim.py:369] (1/2) 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 16:38:48,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=733597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:39:16,165 INFO [train.py:968] (1/2) Epoch 17, batch 2750, giga_loss[loss=0.332, simple_loss=0.3925, pruned_loss=0.1357, over 27956.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3255, pruned_loss=0.08848, over 5701585.58 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3495, pruned_loss=0.09204, over 4349383.19 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3218, pruned_loss=0.08774, over 5711204.57 frames. ], batch size: 412, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:39:24,064 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 16:40:00,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=733672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:40:06,941 INFO [train.py:968] (1/2) Epoch 17, batch 2800, giga_loss[loss=0.3236, simple_loss=0.3887, pruned_loss=0.1292, over 28478.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3314, pruned_loss=0.09211, over 5704631.64 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3497, pruned_loss=0.09217, over 4356809.28 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3282, pruned_loss=0.09143, over 5711564.87 frames. ], batch size: 336, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:40:19,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-08 16:40:23,009 INFO [optim.py:369] (1/2) 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 16:40:54,053 INFO [train.py:968] (1/2) Epoch 17, batch 2850, libri_loss[loss=0.2323, simple_loss=0.3106, pruned_loss=0.07697, over 29317.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3383, pruned_loss=0.09678, over 5693183.45 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.349, pruned_loss=0.09183, over 4402165.52 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3359, pruned_loss=0.09655, over 5693473.71 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:41:41,757 INFO [train.py:968] (1/2) Epoch 17, batch 2900, giga_loss[loss=0.2838, simple_loss=0.3695, pruned_loss=0.09909, over 28916.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3424, pruned_loss=0.09795, over 5704865.61 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3494, pruned_loss=0.09212, over 4425075.04 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3401, pruned_loss=0.09763, over 5701965.16 frames. ], batch size: 174, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:42:03,769 INFO [optim.py:369] (1/2) 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 16:42:19,267 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=733815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:42:21,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=733818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:42:30,385 INFO [train.py:968] (1/2) Epoch 17, batch 2950, giga_loss[loss=0.2801, simple_loss=0.3522, pruned_loss=0.104, over 28942.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3479, pruned_loss=0.1005, over 5702940.76 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3494, pruned_loss=0.09203, over 4447248.52 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3461, pruned_loss=0.1004, over 5697814.19 frames. ], batch size: 112, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:42:47,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=733847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:43:19,321 INFO [train.py:968] (1/2) Epoch 17, batch 3000, giga_loss[loss=0.3523, simple_loss=0.4018, pruned_loss=0.1514, over 27520.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3549, pruned_loss=0.1051, over 5688061.31 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3493, pruned_loss=0.09195, over 4475359.39 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3535, pruned_loss=0.1053, over 5681301.62 frames. ], batch size: 472, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:43:19,322 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 16:43:28,116 INFO [train.py:1012] (1/2) Epoch 17, validation: loss=0.2169, simple_loss=0.323, pruned_loss=0.05543, over 944034.00 frames. 
+2023-03-08 16:43:28,117 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 16:43:43,543 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 3050, giga_loss[loss=0.2414, simple_loss=0.3228, pruned_loss=0.08004, over 28550.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3537, pruned_loss=0.1033, over 5692142.68 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3498, pruned_loss=0.0922, over 4509918.22 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3523, pruned_loss=0.1036, over 5682788.55 frames. ], batch size: 336, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:44:48,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=733972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:44:53,892 INFO [train.py:968] (1/2) Epoch 17, batch 3100, giga_loss[loss=0.2589, simple_loss=0.3409, pruned_loss=0.08846, over 28386.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3502, pruned_loss=0.1006, over 5701956.66 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3498, pruned_loss=0.09247, over 4551655.84 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3492, pruned_loss=0.1009, over 5688399.46 frames. ], batch size: 65, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:45:09,637 INFO [optim.py:369] (1/2) 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 16:45:18,130 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:45:35,281 INFO [train.py:968] (1/2) Epoch 17, batch 3150, giga_loss[loss=0.324, simple_loss=0.3837, pruned_loss=0.1321, over 28792.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.349, pruned_loss=0.09918, over 5702026.06 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3496, pruned_loss=0.09259, over 4591494.48 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3482, pruned_loss=0.09954, over 5696033.94 frames. ], batch size: 92, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:45:38,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5961, 1.7573, 1.7559, 1.6113], device='cuda:1'), covar=tensor([0.1916, 0.2306, 0.2174, 0.2123], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0738, 0.0694, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 16:45:55,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6002, 1.6509, 1.5683, 1.4706], device='cuda:1'), covar=tensor([0.2687, 0.2431, 0.2025, 0.2351], device='cuda:1'), in_proj_covar=tensor([0.1840, 0.1778, 0.1713, 0.1860], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 16:46:18,589 INFO [train.py:968] (1/2) Epoch 17, batch 3200, giga_loss[loss=0.3091, simple_loss=0.3807, pruned_loss=0.1187, over 28247.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3486, pruned_loss=0.09848, over 5706911.35 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3494, pruned_loss=0.09261, over 4623970.63 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3482, pruned_loss=0.09891, over 5697724.69 frames. ], batch size: 368, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:46:28,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7688, 1.8621, 1.8406, 1.6930], device='cuda:1'), covar=tensor([0.1759, 0.2139, 0.2115, 0.2024], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0739, 0.0693, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 16:46:34,173 INFO [optim.py:369] (1/2) 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 16:46:50,725 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:46:52,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:47:01,371 INFO [train.py:968] (1/2) Epoch 17, batch 3250, giga_loss[loss=0.315, simple_loss=0.3867, pruned_loss=0.1216, over 28997.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3518, pruned_loss=0.1002, over 5701077.13 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3496, pruned_loss=0.0927, over 4632620.71 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3513, pruned_loss=0.1005, over 5701248.61 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:47:16,716 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:47:44,500 INFO [train.py:968] (1/2) Epoch 17, batch 3300, giga_loss[loss=0.3014, simple_loss=0.3673, pruned_loss=0.1177, over 28772.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3527, pruned_loss=0.1009, over 5701673.95 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3499, pruned_loss=0.09299, over 4656285.59 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3521, pruned_loss=0.1011, over 5699519.77 frames. ], batch size: 284, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:48:00,938 INFO [optim.py:369] (1/2) 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 16:48:27,425 INFO [train.py:968] (1/2) Epoch 17, batch 3350, giga_loss[loss=0.2913, simple_loss=0.3551, pruned_loss=0.1138, over 28403.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3535, pruned_loss=0.1017, over 5706780.25 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3503, pruned_loss=0.09307, over 4710198.29 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3529, pruned_loss=0.1022, over 5697778.05 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:49:11,321 INFO [train.py:968] (1/2) Epoch 17, batch 3400, libri_loss[loss=0.2373, simple_loss=0.3253, pruned_loss=0.07467, over 29551.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3542, pruned_loss=0.1025, over 5714724.06 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3499, pruned_loss=0.09291, over 4739967.99 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.354, pruned_loss=0.1032, over 5703152.25 frames. ], batch size: 78, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:49:28,704 INFO [optim.py:369] (1/2) 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 16:49:40,266 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:49:54,009 INFO [train.py:968] (1/2) Epoch 17, batch 3450, giga_loss[loss=0.2873, simple_loss=0.3593, pruned_loss=0.1076, over 28687.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3549, pruned_loss=0.1031, over 5716415.29 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3507, pruned_loss=0.09348, over 4765051.66 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3544, pruned_loss=0.1035, over 5710421.01 frames. ], batch size: 336, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:50:07,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:50:22,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0620, 0.9855, 3.3538, 2.9472], device='cuda:1'), covar=tensor([0.1678, 0.2797, 0.0482, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0713, 0.0610, 0.0902, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 16:50:37,857 INFO [train.py:968] (1/2) Epoch 17, batch 3500, giga_loss[loss=0.2693, simple_loss=0.3506, pruned_loss=0.09399, over 28514.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3558, pruned_loss=0.1036, over 5718569.53 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3507, pruned_loss=0.09345, over 4770880.40 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3554, pruned_loss=0.104, over 5712862.64 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:50:42,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:50:43,570 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:50:48,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0570, 1.1350, 3.3495, 2.9478], device='cuda:1'), covar=tensor([0.1672, 0.2781, 0.0484, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0714, 0.0611, 0.0902, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 16:50:54,688 INFO [optim.py:369] (1/2) 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 16:51:18,780 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-08 16:51:18,974 INFO [train.py:968] (1/2) Epoch 17, batch 3550, giga_loss[loss=0.2763, simple_loss=0.3599, pruned_loss=0.09635, over 29025.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3549, pruned_loss=0.1021, over 5716859.19 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3501, pruned_loss=0.09304, over 4792682.00 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3551, pruned_loss=0.1029, over 5709503.56 frames. ], batch size: 155, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:51:21,093 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:51:21,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4126, 1.3949, 4.0709, 3.3473], device='cuda:1'), covar=tensor([0.2002, 0.3041, 0.0726, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0711, 0.0610, 0.0899, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 16:51:36,696 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:51:59,930 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5325, 2.5155, 2.5188, 2.3512], device='cuda:1'), covar=tensor([0.1663, 0.2257, 0.1783, 0.1853], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0739, 0.0694, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 16:52:02,620 INFO [train.py:968] (1/2) Epoch 17, batch 3600, giga_loss[loss=0.2442, simple_loss=0.327, pruned_loss=0.08075, over 28642.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3551, pruned_loss=0.1015, over 5723661.33 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3496, pruned_loss=0.09269, over 4823909.86 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3559, pruned_loss=0.1026, over 5714117.21 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:52:21,192 INFO [optim.py:369] (1/2) 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 16:52:43,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:52:44,897 INFO [train.py:968] (1/2) Epoch 17, batch 3650, giga_loss[loss=0.2409, simple_loss=0.3262, pruned_loss=0.07779, over 28889.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3536, pruned_loss=0.1005, over 5724661.92 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3494, pruned_loss=0.09257, over 4839561.71 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3544, pruned_loss=0.1016, over 5714983.54 frames. ], batch size: 174, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:52:45,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:53:10,829 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:53:21,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8457, 2.0465, 1.6619, 2.2052], device='cuda:1'), covar=tensor([0.2447, 0.2482, 0.2754, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.1426, 0.1039, 0.1262, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 16:53:28,334 INFO [train.py:968] (1/2) Epoch 17, batch 3700, giga_loss[loss=0.298, simple_loss=0.361, pruned_loss=0.1175, over 28845.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3513, pruned_loss=0.09964, over 5724158.75 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3492, pruned_loss=0.09231, over 4860444.39 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3522, pruned_loss=0.1009, over 5713720.69 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:53:47,258 INFO [optim.py:369] (1/2) 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 16:53:56,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 16:54:07,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6112, 1.9541, 1.5236, 1.6770], device='cuda:1'), covar=tensor([0.0708, 0.0261, 0.0303, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 16:54:08,175 INFO [train.py:968] (1/2) Epoch 17, batch 3750, giga_loss[loss=0.242, simple_loss=0.3204, pruned_loss=0.08177, over 28643.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3492, pruned_loss=0.09852, over 5726720.17 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3488, pruned_loss=0.09214, over 4889269.66 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3502, pruned_loss=0.09983, over 5715400.09 frames. ], batch size: 60, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:54:28,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9289, 3.7341, 3.5383, 1.8383], device='cuda:1'), covar=tensor([0.0691, 0.0916, 0.0941, 0.2097], device='cuda:1'), in_proj_covar=tensor([0.1135, 0.1048, 0.0903, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 16:54:52,580 INFO [train.py:968] (1/2) Epoch 17, batch 3800, giga_loss[loss=0.2381, simple_loss=0.3201, pruned_loss=0.07803, over 28248.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3489, pruned_loss=0.09854, over 5728816.59 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3484, pruned_loss=0.09207, over 4910771.36 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3501, pruned_loss=0.09984, over 5723524.55 frames. ], batch size: 77, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:54:58,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:55:07,327 INFO [optim.py:369] (1/2) 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 16:55:07,511 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:55:24,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:55:33,061 INFO [train.py:968] (1/2) Epoch 17, batch 3850, giga_loss[loss=0.3354, simple_loss=0.3913, pruned_loss=0.1397, over 28889.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3505, pruned_loss=0.09956, over 5731124.08 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3487, pruned_loss=0.09213, over 4943643.55 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1008, over 5722675.52 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:55:54,682 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 3900, giga_loss[loss=0.2523, simple_loss=0.3319, pruned_loss=0.08638, over 28889.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3501, pruned_loss=0.09895, over 5729937.04 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3478, pruned_loss=0.09155, over 4971402.26 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3515, pruned_loss=0.1006, over 5719172.15 frames. ], batch size: 112, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:56:28,432 INFO [optim.py:369] (1/2) 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 16:56:33,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=734806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:56:47,870 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 3950, giga_loss[loss=0.3002, simple_loss=0.3663, pruned_loss=0.1171, over 27634.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3486, pruned_loss=0.09734, over 5727137.24 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3472, pruned_loss=0.09129, over 5002725.72 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3502, pruned_loss=0.09906, over 5713845.75 frames. ], batch size: 472, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 16:56:55,422 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:56:58,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:00,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4931, 4.3143, 4.1600, 1.8054], device='cuda:1'), covar=tensor([0.0592, 0.0710, 0.0822, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.1135, 0.1046, 0.0902, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 16:57:15,103 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=734856.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 16:57:18,560 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:19,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:20,658 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 4000, giga_loss[loss=0.2712, simple_loss=0.3515, pruned_loss=0.09544, over 28584.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3487, pruned_loss=0.09755, over 5731067.78 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3469, pruned_loss=0.09117, over 5028501.33 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3502, pruned_loss=0.09919, over 5716667.74 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:57:40,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3856, 3.4446, 1.5015, 1.6054], device='cuda:1'), covar=tensor([0.0982, 0.0253, 0.0892, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0525, 0.0361, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 16:57:46,022 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:57:50,599 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:57:53,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:12,391 INFO [train.py:968] (1/2) Epoch 17, batch 4050, giga_loss[loss=0.2595, simple_loss=0.3377, pruned_loss=0.0907, over 28947.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3475, pruned_loss=0.09738, over 5727690.51 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.347, pruned_loss=0.09115, over 5052213.02 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3487, pruned_loss=0.09889, over 5713323.57 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:58:16,237 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:24,740 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 16:58:31,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:32,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:42,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=734965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:44,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=734968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:58:52,505 INFO [train.py:968] (1/2) Epoch 17, batch 4100, giga_loss[loss=0.2362, simple_loss=0.3169, pruned_loss=0.07775, over 29036.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3442, pruned_loss=0.09581, over 5720308.53 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3465, pruned_loss=0.09083, over 5066557.12 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3455, pruned_loss=0.09736, over 5708063.90 frames. ], batch size: 145, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 16:58:54,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=734981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:59:07,427 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:59:14,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1474, 3.9774, 3.7827, 1.6947], device='cuda:1'), covar=tensor([0.0578, 0.0709, 0.0701, 0.2343], device='cuda:1'), in_proj_covar=tensor([0.1132, 0.1044, 0.0899, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 16:59:33,323 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 16:59:33,759 INFO [train.py:968] (1/2) Epoch 17, batch 4150, giga_loss[loss=0.2414, simple_loss=0.3184, pruned_loss=0.08224, over 28837.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3424, pruned_loss=0.09528, over 5712778.21 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3463, pruned_loss=0.09066, over 5078321.38 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3436, pruned_loss=0.09669, over 5701460.12 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:00:08,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:00:11,759 INFO [train.py:968] (1/2) Epoch 17, batch 4200, giga_loss[loss=0.273, simple_loss=0.3388, pruned_loss=0.1036, over 28755.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.342, pruned_loss=0.09572, over 5716375.61 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3462, pruned_loss=0.09072, over 5102777.48 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3429, pruned_loss=0.09696, over 5702432.51 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:00:31,284 INFO [optim.py:369] (1/2) 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 17:00:36,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3472, 1.6951, 1.6187, 1.4164], device='cuda:1'), covar=tensor([0.2059, 0.2149, 0.2228, 0.2327], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0734, 0.0689, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 17:00:55,544 INFO [train.py:968] (1/2) Epoch 17, batch 4250, giga_loss[loss=0.221, simple_loss=0.3037, pruned_loss=0.06909, over 28889.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3409, pruned_loss=0.09591, over 5717095.70 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3466, pruned_loss=0.09095, over 5114918.08 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3413, pruned_loss=0.09679, over 5703280.21 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:01:05,755 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:01:26,638 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 4300, giga_loss[loss=0.2551, simple_loss=0.3271, pruned_loss=0.0915, over 28763.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3388, pruned_loss=0.09552, over 5714744.94 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3467, pruned_loss=0.09102, over 5120929.47 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.339, pruned_loss=0.09619, over 5703788.43 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:01:58,341 INFO [optim.py:369] (1/2) 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 17:02:12,142 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:02:14,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:02:19,451 INFO [train.py:968] (1/2) Epoch 17, batch 4350, libri_loss[loss=0.2733, simple_loss=0.3494, pruned_loss=0.0986, over 29536.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3364, pruned_loss=0.09471, over 5715465.57 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3468, pruned_loss=0.0911, over 5136457.04 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3362, pruned_loss=0.09525, over 5703133.60 frames. ], batch size: 79, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:02:21,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:02:35,129 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 4400, giga_loss[loss=0.2468, simple_loss=0.3245, pruned_loss=0.08461, over 28884.00 frames. ], tot_loss[loss=0.26, simple_loss=0.334, pruned_loss=0.09295, over 5724390.66 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3464, pruned_loss=0.09094, over 5181380.61 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3335, pruned_loss=0.0937, over 5706181.06 frames. ], batch size: 174, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 17:03:13,643 INFO [optim.py:369] (1/2) 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 17:03:35,150 INFO [train.py:968] (1/2) Epoch 17, batch 4450, giga_loss[loss=0.2657, simple_loss=0.3476, pruned_loss=0.09186, over 28989.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3325, pruned_loss=0.09171, over 5725789.75 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3459, pruned_loss=0.09068, over 5191840.20 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3322, pruned_loss=0.09253, over 5709339.29 frames. ], batch size: 164, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:03:38,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2845, 1.2446, 1.2364, 1.4833], device='cuda:1'), covar=tensor([0.0768, 0.0338, 0.0334, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0091, 0.0065, 0.0058, 0.0100], device='cuda:1')
+2023-03-08 17:03:57,923 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:04:04,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:04:15,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735374.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:04:17,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735377.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:04:18,458 INFO [train.py:968] (1/2) Epoch 17, batch 4500, giga_loss[loss=0.276, simple_loss=0.3608, pruned_loss=0.09559, over 28763.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3364, pruned_loss=0.09333, over 5726228.27 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3461, pruned_loss=0.09079, over 5216412.19 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3357, pruned_loss=0.09395, over 5707111.93 frames. ], batch size: 262, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:04:34,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4580, 4.2722, 4.0656, 2.0348], device='cuda:1'), covar=tensor([0.0554, 0.0732, 0.0742, 0.1984], device='cuda:1'), in_proj_covar=tensor([0.1143, 0.1053, 0.0907, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 17:04:36,594 INFO [optim.py:369] (1/2) 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 17:04:39,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:04:41,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:04:58,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4386, 1.6374, 1.6078, 1.4200], device='cuda:1'), covar=tensor([0.3046, 0.2571, 0.1760, 0.2439], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1781, 0.1706, 0.1852], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 17:05:00,611 INFO [train.py:968] (1/2) Epoch 17, batch 4550, giga_loss[loss=0.2258, simple_loss=0.3006, pruned_loss=0.07548, over 28460.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3394, pruned_loss=0.09501, over 5716835.31 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3463, pruned_loss=0.09105, over 5232280.99 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3385, pruned_loss=0.09538, over 5698748.96 frames. ], batch size: 78, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:05:31,150 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-08 17:05:41,981 INFO [train.py:968] (1/2) Epoch 17, batch 4600, giga_loss[loss=0.3119, simple_loss=0.3795, pruned_loss=0.1222, over 27997.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3422, pruned_loss=0.0956, over 5721362.94 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3467, pruned_loss=0.09114, over 5249286.19 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3411, pruned_loss=0.09593, over 5702559.54 frames. ], batch size: 412, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:06:03,878 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:27,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:28,091 INFO [train.py:968] (1/2) Epoch 17, batch 4650, giga_loss[loss=0.2718, simple_loss=0.3447, pruned_loss=0.09947, over 28800.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3435, pruned_loss=0.09599, over 5704902.66 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3467, pruned_loss=0.09136, over 5265150.88 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.0962, over 5687974.53 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:06:38,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:40,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5213, 1.9791, 1.8665, 1.4718], device='cuda:1'), covar=tensor([0.3265, 0.2151, 0.2133, 0.2769], device='cuda:1'), in_proj_covar=tensor([0.1837, 0.1777, 0.1702, 0.1846], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 17:06:45,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:47,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:06:50,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 17:07:11,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:07:11,569 INFO [train.py:968] (1/2) Epoch 17, batch 4700, giga_loss[loss=0.2639, simple_loss=0.3432, pruned_loss=0.09235, over 28847.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3431, pruned_loss=0.09556, over 5704256.65 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.347, pruned_loss=0.09161, over 5271589.31 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3419, pruned_loss=0.09559, over 5694908.84 frames. ], batch size: 227, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:07:28,695 INFO [optim.py:369] (1/2) 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 17:07:50,731 INFO [train.py:968] (1/2) Epoch 17, batch 4750, giga_loss[loss=0.2633, simple_loss=0.3378, pruned_loss=0.09439, over 28995.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3425, pruned_loss=0.0957, over 5699771.35 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3464, pruned_loss=0.09123, over 5277147.30 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.342, pruned_loss=0.09615, over 5698294.48 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:08:09,839 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=735651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:17,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:19,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:36,314 INFO [train.py:968] (1/2) Epoch 17, batch 4800, giga_loss[loss=0.2907, simple_loss=0.3558, pruned_loss=0.1128, over 28968.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3434, pruned_loss=0.09659, over 5697995.23 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3462, pruned_loss=0.09124, over 5285673.18 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3431, pruned_loss=0.09702, over 5694888.69 frames. ], batch size: 136, lr: 1.91e-03, grad_scale: 8.0
+2023-03-08 17:08:41,008 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:42,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:46,470 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:08:55,118 INFO [optim.py:369] (1/2) 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 17:09:02,232 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.89 vs. limit=5.0
+2023-03-08 17:09:06,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:09:20,982 INFO [train.py:968] (1/2) Epoch 17, batch 4850, giga_loss[loss=0.2795, simple_loss=0.3467, pruned_loss=0.1062, over 28654.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3458, pruned_loss=0.09824, over 5693481.87 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3464, pruned_loss=0.09138, over 5296476.10 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09858, over 5688820.09 frames. ], batch size: 85, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:09:23,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:09:27,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:10:02,552 INFO [train.py:968] (1/2) Epoch 17, batch 4900, giga_loss[loss=0.3855, simple_loss=0.4474, pruned_loss=0.1618, over 28747.00 frames. ], tot_loss[loss=0.274, simple_loss=0.349, pruned_loss=0.09954, over 5704463.84 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3468, pruned_loss=0.09172, over 5317025.17 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3483, pruned_loss=0.09977, over 5694418.60 frames. ], batch size: 284, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:10:22,386 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 4950, giga_loss[loss=0.2455, simple_loss=0.3264, pruned_loss=0.0823, over 28600.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3507, pruned_loss=0.1002, over 5716346.69 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3466, pruned_loss=0.09154, over 5325154.40 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3505, pruned_loss=0.1006, over 5706379.22 frames. ], batch size: 60, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:10:58,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 17:11:21,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:24,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:26,929 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=735878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:27,239 INFO [train.py:968] (1/2) Epoch 17, batch 5000, giga_loss[loss=0.2964, simple_loss=0.358, pruned_loss=0.1174, over 28765.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3508, pruned_loss=0.1001, over 5714880.46 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3465, pruned_loss=0.09152, over 5338491.43 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 5703710.16 frames. ], batch size: 119, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:11:29,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=735881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:43,594 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=735902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:44,023 INFO [optim.py:369] (1/2) 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 17:11:47,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:11:53,518 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=735910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:12:07,455 INFO [train.py:968] (1/2) Epoch 17, batch 5050, giga_loss[loss=0.2587, simple_loss=0.3286, pruned_loss=0.09444, over 28814.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3515, pruned_loss=0.1004, over 5722765.38 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3466, pruned_loss=0.09162, over 5354790.85 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3515, pruned_loss=0.101, over 5711102.73 frames. ], batch size: 99, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:12:42,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-08 17:12:44,699 INFO [train.py:968] (1/2) Epoch 17, batch 5100, giga_loss[loss=0.2404, simple_loss=0.3201, pruned_loss=0.08034, over 28436.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3507, pruned_loss=0.09954, over 5730462.17 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3469, pruned_loss=0.09154, over 5376602.15 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3507, pruned_loss=0.1005, over 5716012.66 frames. ], batch size: 71, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:13:00,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6421, 1.6463, 1.9286, 1.4438], device='cuda:1'), covar=tensor([0.1795, 0.2399, 0.1459, 0.1659], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0694, 0.0918, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 17:13:02,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-08 17:13:04,614 INFO [optim.py:369] (1/2) 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 17:13:24,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=736026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:13:26,112 INFO [train.py:968] (1/2) Epoch 17, batch 5150, giga_loss[loss=0.3557, simple_loss=0.4011, pruned_loss=0.1552, over 26748.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3481, pruned_loss=0.09814, over 5726676.79 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3462, pruned_loss=0.09119, over 5390570.43 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3487, pruned_loss=0.09941, over 5711838.51 frames. ], batch size: 555, lr: 1.91e-03, grad_scale: 4.0
+2023-03-08 17:13:38,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:13:41,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:14:07,394 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:14:09,128 INFO [train.py:968] (1/2) Epoch 17, batch 5200, giga_loss[loss=0.3319, simple_loss=0.385, pruned_loss=0.1393, over 28287.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3452, pruned_loss=0.0967, over 5731225.68 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3464, pruned_loss=0.09137, over 5398820.80 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.09766, over 5718051.20 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:14:30,178 INFO [optim.py:369] (1/2) 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 17:14:45,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 17:14:51,628 INFO [train.py:968] (1/2) Epoch 17, batch 5250, giga_loss[loss=0.2405, simple_loss=0.3207, pruned_loss=0.08014, over 28684.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.342, pruned_loss=0.09482, over 5729285.43 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3467, pruned_loss=0.0915, over 5402937.02 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3421, pruned_loss=0.09554, over 5719866.17 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:15:25,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:15:29,061 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:15:33,986 INFO [train.py:968] (1/2) Epoch 17, batch 5300, giga_loss[loss=0.2829, simple_loss=0.3618, pruned_loss=0.102, over 28830.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3427, pruned_loss=0.0946, over 5712802.22 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3468, pruned_loss=0.09168, over 5400754.61 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3426, pruned_loss=0.09503, over 5712607.18 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:15:53,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:15:55,403 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 5350, giga_loss[loss=0.2492, simple_loss=0.3177, pruned_loss=0.09036, over 28556.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3442, pruned_loss=0.09453, over 5701729.57 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3469, pruned_loss=0.09187, over 5401154.37 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3439, pruned_loss=0.09477, over 5705911.88 frames. ], batch size: 78, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:16:30,636 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 5400, giga_loss[loss=0.2516, simple_loss=0.3262, pruned_loss=0.08846, over 28292.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3455, pruned_loss=0.09559, over 5687525.79 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3475, pruned_loss=0.09224, over 5401250.54 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3447, pruned_loss=0.09554, over 5700506.22 frames. ], batch size: 65, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:17:17,152 INFO [optim.py:369] (1/2) 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 17:17:38,940 INFO [train.py:968] (1/2) Epoch 17, batch 5450, giga_loss[loss=0.2561, simple_loss=0.3255, pruned_loss=0.0933, over 28782.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3449, pruned_loss=0.09686, over 5690143.91 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3473, pruned_loss=0.09219, over 5404999.42 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3444, pruned_loss=0.0969, over 5699656.80 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:18:22,714 INFO [train.py:968] (1/2) Epoch 17, batch 5500, giga_loss[loss=0.2708, simple_loss=0.345, pruned_loss=0.0983, over 28674.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.344, pruned_loss=0.09757, over 5685219.13 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3476, pruned_loss=0.09229, over 5406576.05 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3433, pruned_loss=0.09763, over 5697686.57 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:18:41,299 INFO [optim.py:369] (1/2) 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 17:18:53,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4719, 1.6821, 1.4667, 1.3944], device='cuda:1'), covar=tensor([0.2070, 0.2058, 0.2155, 0.2003], device='cuda:1'), in_proj_covar=tensor([0.1418, 0.1030, 0.1253, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 17:19:02,418 INFO [train.py:968] (1/2) Epoch 17, batch 5550, giga_loss[loss=0.2676, simple_loss=0.3338, pruned_loss=0.1007, over 28557.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3429, pruned_loss=0.09782, over 5689465.05 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3478, pruned_loss=0.09239, over 5416321.35 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.342, pruned_loss=0.09792, over 5698010.48 frames. ], batch size: 60, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:19:15,852 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 5600, giga_loss[loss=0.2989, simple_loss=0.3485, pruned_loss=0.1247, over 28678.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3422, pruned_loss=0.09848, over 5689148.51 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3481, pruned_loss=0.09271, over 5412065.50 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3412, pruned_loss=0.09833, over 5701487.76 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:20:12,338 INFO [optim.py:369] (1/2) 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 17:20:31,225 INFO [train.py:968] (1/2) Epoch 17, batch 5650, giga_loss[loss=0.2462, simple_loss=0.3223, pruned_loss=0.0851, over 28685.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3407, pruned_loss=0.09766, over 5685879.38 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.349, pruned_loss=0.09342, over 5406404.88 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.339, pruned_loss=0.09703, over 5707162.93 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:20:59,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6443, 1.7975, 1.8634, 1.4505], device='cuda:1'), covar=tensor([0.1894, 0.2444, 0.1552, 0.1718], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0694, 0.0917, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 17:21:12,827 INFO [train.py:968] (1/2) Epoch 17, batch 5700, giga_loss[loss=0.2335, simple_loss=0.3118, pruned_loss=0.07757, over 28628.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3348, pruned_loss=0.09444, over 5701625.12 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.09349, over 5416032.53 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3332, pruned_loss=0.09391, over 5714972.55 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:21:34,745 INFO [optim.py:369] (1/2) 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 17:21:48,405 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=736621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:21:53,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 17:21:53,545 INFO [train.py:968] (1/2) Epoch 17, batch 5750, giga_loss[loss=0.2547, simple_loss=0.3257, pruned_loss=0.09184, over 28727.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3314, pruned_loss=0.09242, over 5711752.09 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3488, pruned_loss=0.09336, over 5433032.23 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3298, pruned_loss=0.0921, over 5716009.43 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:22:34,926 INFO [train.py:968] (1/2) Epoch 17, batch 5800, giga_loss[loss=0.2755, simple_loss=0.3523, pruned_loss=0.09937, over 28839.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3312, pruned_loss=0.09188, over 5708690.16 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3486, pruned_loss=0.09326, over 5437422.72 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3296, pruned_loss=0.09167, over 5715362.30 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:22:55,273 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 5850, giga_loss[loss=0.2731, simple_loss=0.3441, pruned_loss=0.1011, over 28688.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3344, pruned_loss=0.0931, over 5711782.25 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3488, pruned_loss=0.09336, over 5444544.91 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3327, pruned_loss=0.09285, over 5714394.13 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:23:31,946 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-08 17:23:44,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:23:46,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:23:57,877 INFO [train.py:968] (1/2) Epoch 17, batch 5900, giga_loss[loss=0.291, simple_loss=0.3622, pruned_loss=0.1099, over 27655.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3371, pruned_loss=0.09386, over 5717409.82 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3488, pruned_loss=0.09335, over 5452869.90 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3355, pruned_loss=0.09368, over 5716948.12 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:23:59,666 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-08 17:24:00,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3582, 1.5268, 1.4392, 1.3120], device='cuda:1'), covar=tensor([0.2281, 0.2025, 0.1510, 0.1831], device='cuda:1'), in_proj_covar=tensor([0.1850, 0.1791, 0.1718, 0.1858], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 17:24:01,974 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:24:18,575 INFO [optim.py:369] (1/2) 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 17:24:24,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-08 17:24:29,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=736817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:24:40,062 INFO [train.py:968] (1/2) Epoch 17, batch 5950, giga_loss[loss=0.2853, simple_loss=0.3564, pruned_loss=0.1071, over 28886.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3404, pruned_loss=0.09522, over 5712753.11 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3486, pruned_loss=0.0933, over 5461804.92 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3391, pruned_loss=0.09514, over 5709168.82 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:25:22,283 INFO [train.py:968] (1/2) Epoch 17, batch 6000, giga_loss[loss=0.3205, simple_loss=0.3822, pruned_loss=0.1294, over 28863.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3432, pruned_loss=0.09662, over 5716553.59 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3486, pruned_loss=0.09322, over 5472971.01 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.342, pruned_loss=0.09668, over 5710369.18 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:25:22,283 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 17:25:30,917 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 17:25:53,830 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 6050, giga_loss[loss=0.2427, simple_loss=0.3288, pruned_loss=0.07836, over 29041.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3462, pruned_loss=0.09902, over 5711827.96 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3487, pruned_loss=0.09327, over 5476991.57 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3452, pruned_loss=0.09906, over 5705630.26 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:26:25,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5117, 1.3576, 4.8552, 3.6327], device='cuda:1'), covar=tensor([0.1642, 0.2754, 0.0352, 0.0751], device='cuda:1'), in_proj_covar=tensor([0.0715, 0.0614, 0.0904, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 17:26:48,108 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=736960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:26:50,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=736963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:27:07,946 INFO [train.py:968] (1/2) Epoch 17, batch 6100, giga_loss[loss=0.3195, simple_loss=0.3818, pruned_loss=0.1286, over 28621.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3523, pruned_loss=0.1041, over 5707976.01 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3486, pruned_loss=0.09323, over 5481980.35 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3515, pruned_loss=0.1043, over 5701738.63 frames. ], batch size: 242, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:27:18,235 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=736992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:27:31,085 INFO [optim.py:369] (1/2) 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 17:27:53,371 INFO [train.py:968] (1/2) Epoch 17, batch 6150, giga_loss[loss=0.2859, simple_loss=0.3626, pruned_loss=0.1046, over 29032.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.357, pruned_loss=0.1079, over 5694914.70 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3476, pruned_loss=0.09275, over 5496654.07 frames. ], giga_tot_loss[loss=0.2877, simple_loss=0.3576, pruned_loss=0.1089, over 5683499.09 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:27:55,628 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 6200, giga_loss[loss=0.4131, simple_loss=0.4444, pruned_loss=0.1909, over 26558.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3653, pruned_loss=0.1143, over 5688554.44 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3475, pruned_loss=0.09266, over 5506780.10 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.366, pruned_loss=0.1155, over 5675122.02 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:29:04,536 INFO [optim.py:369] (1/2) 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 17:29:08,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:29:21,358 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 17:29:27,039 INFO [train.py:968] (1/2) Epoch 17, batch 6250, giga_loss[loss=0.2801, simple_loss=0.3493, pruned_loss=0.1054, over 28924.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3687, pruned_loss=0.1175, over 5678240.74 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3478, pruned_loss=0.09293, over 5512931.43 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3698, pruned_loss=0.1191, over 5669286.60 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:29:55,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:30:11,125 INFO [train.py:968] (1/2) Epoch 17, batch 6300, giga_loss[loss=0.3111, simple_loss=0.379, pruned_loss=0.1216, over 29025.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3742, pruned_loss=0.1215, over 5689192.63 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09293, over 5524520.68 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3754, pruned_loss=0.1236, over 5676080.65 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:30:38,529 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 6350, giga_loss[loss=0.4387, simple_loss=0.4502, pruned_loss=0.2136, over 26669.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3791, pruned_loss=0.1263, over 5663229.96 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3482, pruned_loss=0.09305, over 5529472.29 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3804, pruned_loss=0.1283, over 5650777.69 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:31:14,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3978, 3.3376, 1.5786, 1.4852], device='cuda:1'), covar=tensor([0.0874, 0.0371, 0.0798, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0534, 0.0364, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 17:31:42,368 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-08 17:31:59,249 INFO [train.py:968] (1/2) Epoch 17, batch 6400, giga_loss[loss=0.4159, simple_loss=0.4301, pruned_loss=0.2008, over 23374.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3809, pruned_loss=0.1291, over 5648194.34 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3483, pruned_loss=0.0932, over 5529219.36 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3826, pruned_loss=0.1313, over 5640890.04 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:32:02,746 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:32:20,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4598, 4.4539, 1.6671, 1.6767], device='cuda:1'), covar=tensor([0.0972, 0.0289, 0.0872, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0535, 0.0364, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 17:32:28,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:32:32,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:32:32,294 INFO [optim.py:369] (1/2) 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 17:32:55,288 INFO [train.py:968] (1/2) Epoch 17, batch 6450, giga_loss[loss=0.4221, simple_loss=0.4358, pruned_loss=0.2042, over 23496.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.385, pruned_loss=0.1338, over 5628402.75 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3484, pruned_loss=0.09324, over 5531108.27 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3864, pruned_loss=0.1357, over 5621396.80 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:32:59,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4757, 1.7443, 1.4224, 1.3476], device='cuda:1'), covar=tensor([0.2388, 0.2299, 0.2502, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.1416, 0.1032, 0.1255, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 17:33:04,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:33:48,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4118, 1.6812, 1.7054, 1.2596], device='cuda:1'), covar=tensor([0.1572, 0.2292, 0.1295, 0.1552], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0692, 0.0912, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 17:33:55,328 INFO [train.py:968] (1/2) Epoch 17, batch 6500, giga_loss[loss=0.3573, simple_loss=0.429, pruned_loss=0.1428, over 28350.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3911, pruned_loss=0.1397, over 5613019.14 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3482, pruned_loss=0.09314, over 5533037.73 frames. ], giga_tot_loss[loss=0.3378, simple_loss=0.3926, pruned_loss=0.1415, over 5606218.09 frames. ], batch size: 77, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:34:25,617 INFO [optim.py:369] (1/2) 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 17:34:25,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:34:38,735 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:34:48,394 INFO [train.py:968] (1/2) Epoch 17, batch 6550, giga_loss[loss=0.3504, simple_loss=0.4004, pruned_loss=0.1503, over 28775.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3915, pruned_loss=0.14, over 5624772.00 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3483, pruned_loss=0.09317, over 5538515.22 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3932, pruned_loss=0.142, over 5615500.91 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:35:39,366 INFO [train.py:968] (1/2) Epoch 17, batch 6600, giga_loss[loss=0.2989, simple_loss=0.3694, pruned_loss=0.1142, over 28807.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.39, pruned_loss=0.1393, over 5639534.24 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3483, pruned_loss=0.09311, over 5542617.82 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3919, pruned_loss=0.1416, over 5629662.33 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:35:46,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:35:55,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6605, 1.7219, 1.2624, 1.2319], device='cuda:1'), covar=tensor([0.0941, 0.0651, 0.1069, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0442, 0.0509, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 17:36:05,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1977, 1.3117, 3.3689, 3.0095], device='cuda:1'), covar=tensor([0.1476, 0.2481, 0.0450, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.0718, 0.0618, 0.0908, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 17:36:08,223 INFO [optim.py:369] (1/2) 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 17:36:31,451 INFO [train.py:968] (1/2) Epoch 17, batch 6650, libri_loss[loss=0.3469, simple_loss=0.4112, pruned_loss=0.1414, over 28678.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3885, pruned_loss=0.1381, over 5639826.55 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3485, pruned_loss=0.09336, over 5553169.34 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3912, pruned_loss=0.1412, over 5625381.29 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:36:53,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:36:57,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:37:06,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6377, 1.9003, 1.8453, 1.5915], device='cuda:1'), covar=tensor([0.2192, 0.1984, 0.1954, 0.1972], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1799, 0.1718, 0.1855], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 17:37:22,506 INFO [train.py:968] (1/2) Epoch 17, batch 6700, giga_loss[loss=0.3565, simple_loss=0.4159, pruned_loss=0.1485, over 28565.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3887, pruned_loss=0.1369, over 5646256.68 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.0935, over 5558309.60 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3912, pruned_loss=0.1399, over 5631351.64 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:37:26,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:37:50,458 INFO [optim.py:369] (1/2) 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 17:38:11,214 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737625.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:38:14,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:14,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:14,647 INFO [train.py:968] (1/2) Epoch 17, batch 6750, giga_loss[loss=0.3269, simple_loss=0.3991, pruned_loss=0.1273, over 28883.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3893, pruned_loss=0.137, over 5638794.02 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3484, pruned_loss=0.09336, over 5564126.44 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3924, pruned_loss=0.1403, over 5623192.66 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:38:16,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:33,950 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4269, 1.5765, 1.5655, 1.4405], device='cuda:1'), covar=tensor([0.1609, 0.1785, 0.1886, 0.1676], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0741, 0.0697, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 17:38:41,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:44,588 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:38:58,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5658, 1.8436, 1.5914, 1.6058], device='cuda:1'), covar=tensor([0.1562, 0.1900, 0.1933, 0.1902], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0739, 0.0695, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 17:39:04,611 INFO [train.py:968] (1/2) Epoch 17, batch 6800, giga_loss[loss=0.2911, simple_loss=0.3676, pruned_loss=0.1074, over 28895.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3877, pruned_loss=0.1354, over 5621657.44 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3488, pruned_loss=0.09346, over 5568966.57 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3907, pruned_loss=0.1389, over 5606384.69 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:39:05,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:39:30,350 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 6850, giga_loss[loss=0.3379, simple_loss=0.3916, pruned_loss=0.1421, over 27665.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3844, pruned_loss=0.1318, over 5616633.03 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.09365, over 5564631.15 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3879, pruned_loss=0.1357, over 5610170.43 frames. ], batch size: 474, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:40:10,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6197, 4.4284, 4.2030, 1.7642], device='cuda:1'), covar=tensor([0.0557, 0.0749, 0.0813, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1168, 0.1077, 0.0929, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 17:40:29,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5450, 1.6827, 1.7643, 1.3415], device='cuda:1'), covar=tensor([0.1789, 0.2477, 0.1495, 0.1722], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0692, 0.0911, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 17:40:48,612 INFO [train.py:968] (1/2) Epoch 17, batch 6900, giga_loss[loss=0.2843, simple_loss=0.3615, pruned_loss=0.1035, over 29059.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3811, pruned_loss=0.1279, over 5629801.11 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3486, pruned_loss=0.09358, over 5566394.99 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3841, pruned_loss=0.1311, over 5623457.74 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:41:02,990 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:03,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=737794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:10,982 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:16,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:19,820 INFO [optim.py:369] (1/2) 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 17:41:26,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4303, 1.6712, 1.7012, 1.3798], device='cuda:1'), covar=tensor([0.3002, 0.2418, 0.1833, 0.2453], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1792, 0.1717, 0.1854], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 17:41:40,999 INFO [train.py:968] (1/2) Epoch 17, batch 6950, giga_loss[loss=0.2714, simple_loss=0.343, pruned_loss=0.09989, over 29005.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.379, pruned_loss=0.1266, over 5640340.60 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3485, pruned_loss=0.09366, over 5571131.86 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3819, pruned_loss=0.1295, over 5631988.74 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:41:43,900 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=737832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:41:50,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737839.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:42:26,012 INFO [train.py:968] (1/2) Epoch 17, batch 7000, giga_loss[loss=0.3218, simple_loss=0.3698, pruned_loss=0.1369, over 28542.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3772, pruned_loss=0.125, over 5637758.58 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3492, pruned_loss=0.09433, over 5570012.37 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3801, pruned_loss=0.1281, over 5634652.66 frames. ], batch size: 78, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:42:31,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3718, 1.5380, 1.4338, 1.3616], device='cuda:1'), covar=tensor([0.2281, 0.1892, 0.1906, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1794, 0.1718, 0.1854], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 17:42:56,293 INFO [optim.py:369] (1/2) 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 17:43:07,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2256, 3.0568, 2.9056, 1.4602], device='cuda:1'), covar=tensor([0.0975, 0.1032, 0.0882, 0.2335], device='cuda:1'), in_proj_covar=tensor([0.1167, 0.1078, 0.0927, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 17:43:15,841 INFO [train.py:968] (1/2) Epoch 17, batch 7050, giga_loss[loss=0.2871, simple_loss=0.3668, pruned_loss=0.1037, over 28944.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.376, pruned_loss=0.1244, over 5644684.37 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3489, pruned_loss=0.09418, over 5574856.60 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.379, pruned_loss=0.1275, over 5639071.32 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:43:22,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=737937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:43:27,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=737940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:43:50,487 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=737963.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:43:59,176 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 7100, libri_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.0885, over 29502.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3747, pruned_loss=0.1231, over 5658108.95 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3489, pruned_loss=0.0941, over 5580408.77 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.378, pruned_loss=0.1265, over 5650319.99 frames. ], batch size: 81, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 17:44:29,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738000.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:44:31,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:44:34,587 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:44:59,632 INFO [train.py:968] (1/2) Epoch 17, batch 7150, giga_loss[loss=0.2625, simple_loss=0.3388, pruned_loss=0.09304, over 28677.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3725, pruned_loss=0.121, over 5668925.22 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3482, pruned_loss=0.09379, over 5591425.12 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3766, pruned_loss=0.1251, over 5655463.39 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 17:45:17,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0626, 3.8833, 3.7137, 1.6446], device='cuda:1'), covar=tensor([0.0660, 0.0776, 0.0817, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.1084, 0.0933, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 17:45:26,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:45:49,073 INFO [train.py:968] (1/2) Epoch 17, batch 7200, libri_loss[loss=0.2602, simple_loss=0.3325, pruned_loss=0.09392, over 29566.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3705, pruned_loss=0.1181, over 5678151.24 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3471, pruned_loss=0.09321, over 5607090.97 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3761, pruned_loss=0.1231, over 5656038.56 frames. ], batch size: 77, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:46:21,083 INFO [optim.py:369] (1/2) 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 17:46:41,621 INFO [train.py:968] (1/2) Epoch 17, batch 7250, giga_loss[loss=0.3361, simple_loss=0.4039, pruned_loss=0.1342, over 28680.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3723, pruned_loss=0.1172, over 5679970.39 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3468, pruned_loss=0.09292, over 5614564.50 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3775, pruned_loss=0.1221, over 5657262.83 frames. ], batch size: 242, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:46:53,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2307, 1.5133, 1.5480, 1.3376], device='cuda:1'), covar=tensor([0.1747, 0.1519, 0.2097, 0.1814], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0735, 0.0693, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 17:46:55,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738143.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:46:59,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738146.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:46:59,105 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:03,594 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:24,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:33,312 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738175.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:47:36,362 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:36,718 INFO [train.py:968] (1/2) Epoch 17, batch 7300, giga_loss[loss=0.2953, simple_loss=0.3673, pruned_loss=0.1117, over 28902.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3743, pruned_loss=0.1192, over 5677314.86 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.347, pruned_loss=0.09321, over 5617425.29 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3787, pruned_loss=0.1232, over 5657616.46 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:47:44,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-08 17:47:55,149 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:47:57,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:48:04,183 INFO [optim.py:369] (1/2) 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 17:48:10,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738214.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:48:27,178 INFO [train.py:968] (1/2) Epoch 17, batch 7350, giga_loss[loss=0.2951, simple_loss=0.3623, pruned_loss=0.1139, over 28921.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3739, pruned_loss=0.1196, over 5681384.10 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.347, pruned_loss=0.09325, over 5622168.42 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3778, pruned_loss=0.1232, over 5662758.59 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:48:27,542 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:48:31,046 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=738233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:49:17,078 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-08 17:49:17,292 INFO [train.py:968] (1/2) Epoch 17, batch 7400, giga_loss[loss=0.3269, simple_loss=0.3804, pruned_loss=0.1366, over 28585.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3728, pruned_loss=0.1201, over 5661788.57 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3472, pruned_loss=0.0933, over 5614250.63 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3761, pruned_loss=0.1231, over 5655068.77 frames. ], batch size: 78, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:49:46,349 INFO [optim.py:369] (1/2) 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 17:49:48,300 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:49:50,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:50:06,217 INFO [train.py:968] (1/2) Epoch 17, batch 7450, giga_loss[loss=0.3103, simple_loss=0.3782, pruned_loss=0.1212, over 28857.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3723, pruned_loss=0.1206, over 5674530.54 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3472, pruned_loss=0.09326, over 5619954.74 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3753, pruned_loss=0.1236, over 5664969.94 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:50:12,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738338.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:50:17,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:50:19,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-08 17:50:29,976 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738357.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:50:32,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738360.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:50:50,645 INFO [train.py:968] (1/2) Epoch 17, batch 7500, giga_loss[loss=0.3293, simple_loss=0.3975, pruned_loss=0.1305, over 28827.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3718, pruned_loss=0.1198, over 5677568.29 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3483, pruned_loss=0.09392, over 5625020.06 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3743, pruned_loss=0.1229, over 5666946.61 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:50:51,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738379.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:50:59,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738389.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:51:18,554 INFO [optim.py:369] (1/2) 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 17:51:37,423 INFO [train.py:968] (1/2) Epoch 17, batch 7550, giga_loss[loss=0.3948, simple_loss=0.4303, pruned_loss=0.1796, over 28549.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3714, pruned_loss=0.1181, over 5690156.43 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3488, pruned_loss=0.0942, over 5628046.81 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3736, pruned_loss=0.1209, over 5680062.06 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:51:51,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-08 17:52:21,973 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-08 17:52:24,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3806, 1.5802, 1.3663, 1.3014], device='cuda:1'), covar=tensor([0.2102, 0.1907, 0.1992, 0.1956], device='cuda:1'), in_proj_covar=tensor([0.1861, 0.1803, 0.1721, 0.1864], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 17:52:26,878 INFO [train.py:968] (1/2) Epoch 17, batch 7600, giga_loss[loss=0.2752, simple_loss=0.3514, pruned_loss=0.09947, over 28933.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3709, pruned_loss=0.1172, over 5694304.85 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3485, pruned_loss=0.09408, over 5630339.41 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3731, pruned_loss=0.1198, over 5685265.29 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:52:29,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738481.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:52:31,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738484.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:52:47,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-08 17:52:54,128 INFO [optim.py:369] (1/2) 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 17:52:57,479 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738513.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:53:04,083 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738525.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 17:53:09,234 INFO [train.py:968] (1/2) Epoch 17, batch 7650, giga_loss[loss=0.3226, simple_loss=0.3747, pruned_loss=0.1353, over 29036.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.371, pruned_loss=0.1178, over 5690072.50 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3484, pruned_loss=0.09405, over 5624485.87 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3731, pruned_loss=0.1201, over 5688139.21 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:53:37,652 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=738554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:53:37,671 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=738554.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 17:54:01,573 INFO [train.py:968] (1/2) Epoch 17, batch 7700, giga_loss[loss=0.2943, simple_loss=0.3614, pruned_loss=0.1136, over 28886.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3687, pruned_loss=0.117, over 5685034.43 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3482, pruned_loss=0.09398, over 5620504.36 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.371, pruned_loss=0.1194, over 5687967.88 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:54:32,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:54:34,218 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 7750, giga_loss[loss=0.3256, simple_loss=0.3905, pruned_loss=0.1303, over 28641.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3689, pruned_loss=0.1179, over 5674027.35 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3485, pruned_loss=0.09424, over 5618629.69 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3709, pruned_loss=0.12, over 5678874.11 frames. ], batch size: 242, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:55:43,912 INFO [train.py:968] (1/2) Epoch 17, batch 7800, libri_loss[loss=0.2695, simple_loss=0.3537, pruned_loss=0.09268, over 29541.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3691, pruned_loss=0.1187, over 5684463.12 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3487, pruned_loss=0.09431, over 5621492.28 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3707, pruned_loss=0.1206, over 5686143.26 frames. ], batch size: 80, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:56:16,110 INFO [optim.py:369] (1/2) 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 17:56:36,192 INFO [train.py:968] (1/2) Epoch 17, batch 7850, giga_loss[loss=0.3241, simple_loss=0.3825, pruned_loss=0.1328, over 28797.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.368, pruned_loss=0.1187, over 5692421.65 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3488, pruned_loss=0.09441, over 5625726.35 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3695, pruned_loss=0.1204, over 5690896.11 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:56:58,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=738751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:57:00,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=738754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 17:57:22,897 INFO [train.py:968] (1/2) Epoch 17, batch 7900, libri_loss[loss=0.2618, simple_loss=0.3408, pruned_loss=0.09144, over 29573.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3666, pruned_loss=0.1183, over 5697719.29 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3487, pruned_loss=0.09437, over 5628598.91 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.368, pruned_loss=0.1199, over 5694357.36 frames. ], batch size: 76, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:57:27,406 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:58:12,020 INFO [train.py:968] (1/2) Epoch 17, batch 7950, giga_loss[loss=0.2952, simple_loss=0.3608, pruned_loss=0.1148, over 28993.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3667, pruned_loss=0.1182, over 5690237.98 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3485, pruned_loss=0.09417, over 5633655.21 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 5684501.50 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 17:59:00,100 INFO [train.py:968] (1/2) Epoch 17, batch 8000, giga_loss[loss=0.3606, simple_loss=0.3946, pruned_loss=0.1633, over 26565.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3671, pruned_loss=0.118, over 5685869.77 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3488, pruned_loss=0.09454, over 5635952.68 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3685, pruned_loss=0.1197, over 5680081.97 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:59:28,125 INFO [optim.py:369] (1/2) 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 17:59:45,591 INFO [train.py:968] (1/2) Epoch 17, batch 8050, giga_loss[loss=0.2797, simple_loss=0.3559, pruned_loss=0.1018, over 28920.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3663, pruned_loss=0.1165, over 5680111.82 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3488, pruned_loss=0.09437, over 5637886.01 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3678, pruned_loss=0.1184, over 5674765.85 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 17:59:45,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=738929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:00:12,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=738955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:00:26,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3302, 1.1447, 1.0786, 1.6195], device='cuda:1'), covar=tensor([0.0773, 0.0373, 0.0355, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0116, 0.0116, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 18:00:34,659 INFO [train.py:968] (1/2) Epoch 17, batch 8100, giga_loss[loss=0.3585, simple_loss=0.421, pruned_loss=0.148, over 28652.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3677, pruned_loss=0.1173, over 5677975.88 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3488, pruned_loss=0.09428, over 5645552.23 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3694, pruned_loss=0.1196, over 5667584.03 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:00:49,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9060, 2.0878, 1.8762, 1.7134], device='cuda:1'), covar=tensor([0.1679, 0.2061, 0.1994, 0.2152], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0740, 0.0699, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 18:01:02,522 INFO [optim.py:369] (1/2) 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 18:01:18,681 INFO [train.py:968] (1/2) Epoch 17, batch 8150, giga_loss[loss=0.2974, simple_loss=0.3758, pruned_loss=0.1095, over 28989.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3683, pruned_loss=0.1181, over 5690880.30 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3487, pruned_loss=0.09439, over 5652646.54 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3704, pruned_loss=0.1206, over 5677447.25 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:02:05,888 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=739072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:02:10,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=739075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:02:10,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5608, 2.2252, 1.6743, 0.7203], device='cuda:1'), covar=tensor([0.4790, 0.2649, 0.3423, 0.5598], device='cuda:1'), in_proj_covar=tensor([0.1675, 0.1589, 0.1558, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 18:02:14,742 INFO [train.py:968] (1/2) Epoch 17, batch 8200, giga_loss[loss=0.2977, simple_loss=0.3648, pruned_loss=0.1153, over 28852.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3712, pruned_loss=0.1211, over 5684011.13 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3489, pruned_loss=0.09446, over 5656498.84 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3729, pruned_loss=0.1234, over 5670345.47 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:02:40,148 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=739104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:02:47,370 INFO [optim.py:369] (1/2) 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 18:03:08,488 INFO [train.py:968] (1/2) Epoch 17, batch 8250, giga_loss[loss=0.322, simple_loss=0.3761, pruned_loss=0.1339, over 28675.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3718, pruned_loss=0.1225, over 5681897.13 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3489, pruned_loss=0.09446, over 5657663.76 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3735, pruned_loss=0.1247, over 5670414.39 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:03:29,881 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1500, 3.9686, 3.7746, 1.9736], device='cuda:1'), covar=tensor([0.0650, 0.0786, 0.0809, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.1082, 0.0931, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 18:03:56,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4171, 3.7057, 1.5057, 1.6647], device='cuda:1'), covar=tensor([0.1003, 0.0326, 0.0888, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0540, 0.0366, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 18:03:57,929 INFO [train.py:968] (1/2) Epoch 17, batch 8300, giga_loss[loss=0.2927, simple_loss=0.3561, pruned_loss=0.1147, over 28924.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1249, over 5675513.77 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3494, pruned_loss=0.09461, over 5664291.77 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3755, pruned_loss=0.1273, over 5660609.19 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:04:28,612 INFO [optim.py:369] (1/2) 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 18:04:45,119 INFO [train.py:968] (1/2) Epoch 17, batch 8350, giga_loss[loss=0.2742, simple_loss=0.3462, pruned_loss=0.1011, over 28866.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5677558.07 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3497, pruned_loss=0.09471, over 5670505.71 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.376, pruned_loss=0.1282, over 5660203.48 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:05:14,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6201, 2.9100, 2.6377, 2.2393], device='cuda:1'), covar=tensor([0.2587, 0.1798, 0.1879, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.1867, 0.1818, 0.1738, 0.1875], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 18:05:27,081 INFO [train.py:968] (1/2) Epoch 17, batch 8400, giga_loss[loss=0.2578, simple_loss=0.3341, pruned_loss=0.09079, over 28830.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3713, pruned_loss=0.1232, over 5675104.12 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3495, pruned_loss=0.09475, over 5670038.33 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3737, pruned_loss=0.1265, over 5661208.52 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:05:54,726 INFO [optim.py:369] (1/2) 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 18:06:07,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 18:06:11,946 INFO [train.py:968] (1/2) Epoch 17, batch 8450, giga_loss[loss=0.2517, simple_loss=0.3384, pruned_loss=0.08256, over 28932.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.1209, over 5683967.00 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3491, pruned_loss=0.09451, over 5676628.57 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3733, pruned_loss=0.1245, over 5667003.34 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:06:12,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:06:39,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4828, 2.6149, 1.5508, 1.6434], device='cuda:1'), covar=tensor([0.0813, 0.0309, 0.0739, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0544, 0.0368, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 18:06:57,341 INFO [train.py:968] (1/2) Epoch 17, batch 8500, giga_loss[loss=0.322, simple_loss=0.3763, pruned_loss=0.1338, over 28521.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3683, pruned_loss=0.119, over 5675891.19 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3493, pruned_loss=0.09455, over 5676844.44 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3708, pruned_loss=0.1223, over 5662151.33 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:07:19,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-08 18:07:22,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-08 18:07:29,613 INFO [optim.py:369] (1/2) 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 18:07:32,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3197, 1.4649, 1.6518, 1.2895], device='cuda:1'), covar=tensor([0.1711, 0.1654, 0.2013, 0.1803], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0740, 0.0698, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 18:07:44,610 INFO [train.py:968] (1/2) Epoch 17, batch 8550, giga_loss[loss=0.3338, simple_loss=0.3876, pruned_loss=0.14, over 28002.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.367, pruned_loss=0.1183, over 5684442.22 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.35, pruned_loss=0.09487, over 5683311.35 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.369, pruned_loss=0.1215, over 5667331.29 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:07:46,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9799, 1.1265, 3.2849, 2.8437], device='cuda:1'), covar=tensor([0.1704, 0.2739, 0.0512, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0624, 0.0920, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:08:07,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-08 18:08:07,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 18:08:27,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=739473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:08:30,225 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 8600, giga_loss[loss=0.3102, simple_loss=0.3669, pruned_loss=0.1267, over 29012.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3665, pruned_loss=0.1188, over 5682647.23 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3503, pruned_loss=0.09494, over 5684660.27 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3681, pruned_loss=0.1215, over 5668038.51 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:09:00,754 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=739505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:09:04,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6486, 2.3242, 1.7285, 0.7830], device='cuda:1'), covar=tensor([0.5402, 0.2807, 0.3796, 0.5922], device='cuda:1'), in_proj_covar=tensor([0.1669, 0.1591, 0.1554, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 18:09:07,385 INFO [optim.py:369] (1/2) 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 18:09:25,178 INFO [train.py:968] (1/2) Epoch 17, batch 8650, giga_loss[loss=0.2764, simple_loss=0.3468, pruned_loss=0.103, over 28900.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1206, over 5664573.21 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3506, pruned_loss=0.09517, over 5689574.02 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3694, pruned_loss=0.1232, over 5648332.50 frames. ], batch size: 227, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:09:38,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7436, 5.2570, 1.8613, 2.1998], device='cuda:1'), covar=tensor([0.0954, 0.0220, 0.0933, 0.1218], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0541, 0.0367, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 18:09:44,447 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=739547.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:10:16,263 INFO [train.py:968] (1/2) Epoch 17, batch 8700, giga_loss[loss=0.2986, simple_loss=0.3798, pruned_loss=0.1087, over 28745.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3711, pruned_loss=0.1212, over 5673185.59 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.35, pruned_loss=0.09488, over 5694234.98 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.373, pruned_loss=0.124, over 5655626.04 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:10:19,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6073, 1.6587, 1.8501, 1.4109], device='cuda:1'), covar=tensor([0.1603, 0.2167, 0.1291, 0.1595], device='cuda:1'), in_proj_covar=tensor([0.0870, 0.0698, 0.0916, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 18:10:35,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-08 18:10:45,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8576, 2.0354, 1.9417, 1.8194], device='cuda:1'), covar=tensor([0.1853, 0.1572, 0.1350, 0.1492], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1815, 0.1737, 0.1874], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 18:10:49,099 INFO [optim.py:369] (1/2) 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 18:11:04,572 INFO [train.py:968] (1/2) Epoch 17, batch 8750, libri_loss[loss=0.2952, simple_loss=0.3629, pruned_loss=0.1138, over 29684.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3725, pruned_loss=0.1197, over 5668558.00 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.35, pruned_loss=0.09491, over 5691100.82 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3747, pruned_loss=0.1226, over 5656497.16 frames. ], batch size: 91, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:11:14,596 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-08 18:11:43,016 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=739672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:11:51,617 INFO [train.py:968] (1/2) Epoch 17, batch 8800, giga_loss[loss=0.3358, simple_loss=0.3965, pruned_loss=0.1376, over 28534.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3745, pruned_loss=0.1209, over 5671727.57 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3497, pruned_loss=0.09485, over 5694345.57 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3769, pruned_loss=0.1237, over 5659071.40 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:12:18,302 INFO [optim.py:369] (1/2) 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 18:12:31,992 INFO [train.py:968] (1/2) Epoch 17, batch 8850, giga_loss[loss=0.297, simple_loss=0.367, pruned_loss=0.1135, over 29066.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3766, pruned_loss=0.1229, over 5673787.94 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3495, pruned_loss=0.09503, over 5696640.57 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3795, pruned_loss=0.1258, over 5661261.99 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:12:38,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=739736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:12:38,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1638, 1.4862, 1.4466, 1.1057], device='cuda:1'), covar=tensor([0.1172, 0.1779, 0.0980, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0870, 0.0700, 0.0917, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 18:13:20,178 INFO [train.py:968] (1/2) Epoch 17, batch 8900, giga_loss[loss=0.4273, simple_loss=0.4447, pruned_loss=0.2049, over 26625.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3781, pruned_loss=0.1249, over 5663016.60 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3493, pruned_loss=0.09496, over 5702937.77 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3815, pruned_loss=0.1281, over 5646363.65 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:13:28,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2260, 1.4525, 1.4874, 1.2921], device='cuda:1'), covar=tensor([0.1861, 0.1695, 0.2254, 0.1873], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0742, 0.0700, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 18:13:52,713 INFO [optim.py:369] (1/2) 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 18:14:08,253 INFO [train.py:968] (1/2) Epoch 17, batch 8950, giga_loss[loss=0.406, simple_loss=0.4345, pruned_loss=0.1888, over 26651.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3761, pruned_loss=0.1243, over 5659293.48 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09487, over 5704586.39 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3791, pruned_loss=0.1271, over 5644558.30 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:14:19,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5777, 1.7345, 1.1701, 1.2798], device='cuda:1'), covar=tensor([0.0878, 0.0583, 0.1066, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0443, 0.0509, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:14:59,536 INFO [train.py:968] (1/2) Epoch 17, batch 9000, libri_loss[loss=0.2264, simple_loss=0.3034, pruned_loss=0.07466, over 29652.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.373, pruned_loss=0.1227, over 5655328.87 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3488, pruned_loss=0.09465, over 5707832.58 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3761, pruned_loss=0.1256, over 5639899.89 frames. ], batch size: 73, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:14:59,537 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 18:15:08,262 INFO [train.py:1012] (1/2) Epoch 17, validation: loss=0.2118, simple_loss=0.3196, pruned_loss=0.05194, over 944034.00 frames. 
+2023-03-08 18:15:08,263 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 18:15:17,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9663, 1.1929, 1.3142, 0.9706], device='cuda:1'), covar=tensor([0.1776, 0.1460, 0.2266, 0.1773], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0737, 0.0695, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 18:15:42,371 INFO [optim.py:369] (1/2) 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 18:15:49,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739922.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:15:54,870 INFO [train.py:968] (1/2) Epoch 17, batch 9050, giga_loss[loss=0.3182, simple_loss=0.3761, pruned_loss=0.1302, over 28827.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3717, pruned_loss=0.1223, over 5659038.77 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3488, pruned_loss=0.09472, over 5704225.46 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3748, pruned_loss=0.1253, over 5649011.96 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:16:18,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1979, 1.5192, 1.5183, 1.1100], device='cuda:1'), covar=tensor([0.1463, 0.2190, 0.1179, 0.1414], device='cuda:1'), in_proj_covar=tensor([0.0868, 0.0696, 0.0915, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 18:16:41,592 INFO [train.py:968] (1/2) Epoch 17, batch 9100, giga_loss[loss=0.2823, simple_loss=0.3429, pruned_loss=0.1108, over 29065.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.371, pruned_loss=0.1222, over 5671759.93 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3487, pruned_loss=0.09466, over 5711449.25 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 5655923.86 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:17:16,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4002, 1.5673, 1.4606, 1.3078], device='cuda:1'), covar=tensor([0.2287, 0.2173, 0.1870, 0.2103], device='cuda:1'), in_proj_covar=tensor([0.1865, 0.1814, 0.1742, 0.1877], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 18:17:16,290 INFO [optim.py:369] (1/2) 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 18:17:32,310 INFO [train.py:968] (1/2) Epoch 17, batch 9150, giga_loss[loss=0.3304, simple_loss=0.3834, pruned_loss=0.1387, over 28384.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1237, over 5656259.88 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3485, pruned_loss=0.09463, over 5716522.13 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3757, pruned_loss=0.1273, over 5637312.67 frames. ], batch size: 65, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:17:44,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-08 18:17:49,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:18:06,764 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740065.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:18:09,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740068.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:18:17,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2192, 2.5978, 1.3539, 1.3699], device='cuda:1'), covar=tensor([0.0966, 0.0384, 0.0809, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0541, 0.0367, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 18:18:20,129 INFO [train.py:968] (1/2) Epoch 17, batch 9200, giga_loss[loss=0.2594, simple_loss=0.3209, pruned_loss=0.09892, over 28672.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3703, pruned_loss=0.123, over 5665426.98 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3484, pruned_loss=0.09458, over 5715545.09 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3735, pruned_loss=0.1263, over 5650758.62 frames. ], batch size: 85, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:18:24,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:18:36,527 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740097.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:18:51,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:18:53,236 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 9250, libri_loss[loss=0.2623, simple_loss=0.3493, pruned_loss=0.08769, over 29508.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3689, pruned_loss=0.1222, over 5655767.86 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3488, pruned_loss=0.09476, over 5715697.55 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3717, pruned_loss=0.1254, over 5642326.69 frames. ], batch size: 82, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:19:51,733 INFO [train.py:968] (1/2) Epoch 17, batch 9300, giga_loss[loss=0.3328, simple_loss=0.3721, pruned_loss=0.1467, over 23375.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3686, pruned_loss=0.1211, over 5659630.89 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3486, pruned_loss=0.09468, over 5719945.20 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3713, pruned_loss=0.1242, over 5644150.15 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:20:08,709 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:20:12,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:20:18,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3718, 1.6219, 1.5107, 1.2976], device='cuda:1'), covar=tensor([0.2468, 0.2295, 0.1751, 0.2159], device='cuda:1'), in_proj_covar=tensor([0.1866, 0.1816, 0.1743, 0.1878], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 18:20:20,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5470, 1.6541, 1.7896, 1.3499], device='cuda:1'), covar=tensor([0.1682, 0.2335, 0.1382, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0700, 0.0919, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 18:20:32,261 INFO [optim.py:369] (1/2) 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 18:20:41,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:20:46,632 INFO [train.py:968] (1/2) Epoch 17, batch 9350, giga_loss[loss=0.3803, simple_loss=0.4057, pruned_loss=0.1774, over 23512.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3722, pruned_loss=0.1231, over 5654442.41 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3489, pruned_loss=0.09491, over 5713119.03 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1257, over 5648103.11 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:20:58,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-08 18:21:14,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:21:17,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:21:38,191 INFO [train.py:968] (1/2) Epoch 17, batch 9400, giga_loss[loss=0.2516, simple_loss=0.3253, pruned_loss=0.08891, over 28847.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3723, pruned_loss=0.1236, over 5656613.09 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.349, pruned_loss=0.09491, over 5714761.62 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3741, pruned_loss=0.1259, over 5649424.56 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:21:45,019 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:22:11,521 INFO [optim.py:369] (1/2) 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 18:22:22,877 INFO [train.py:968] (1/2) Epoch 17, batch 9450, libri_loss[loss=0.2866, simple_loss=0.3653, pruned_loss=0.1039, over 29747.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3728, pruned_loss=0.1219, over 5652312.84 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3487, pruned_loss=0.09464, over 5709474.17 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3753, pruned_loss=0.1248, over 5650418.29 frames. ], batch size: 87, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:22:48,788 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 18:23:10,095 INFO [train.py:968] (1/2) Epoch 17, batch 9500, giga_loss[loss=0.3672, simple_loss=0.4133, pruned_loss=0.1606, over 27587.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3721, pruned_loss=0.119, over 5665750.95 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3482, pruned_loss=0.09431, over 5714396.71 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3751, pruned_loss=0.1222, over 5658620.26 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:23:39,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:23:56,967 INFO [train.py:968] (1/2) Epoch 17, batch 9550, giga_loss[loss=0.2688, simple_loss=0.354, pruned_loss=0.09183, over 29016.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3748, pruned_loss=0.1193, over 5674875.19 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3479, pruned_loss=0.09418, over 5718225.22 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3779, pruned_loss=0.1224, over 5664931.85 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:24:04,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4908, 2.8329, 2.0546, 2.4271], device='cuda:1'), covar=tensor([0.0853, 0.0600, 0.0930, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0444, 0.0509, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:24:21,264 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:24:27,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:24:27,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2481, 1.1499, 3.8831, 3.3855], device='cuda:1'), covar=tensor([0.1584, 0.2777, 0.0451, 0.1620], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0619, 0.0916, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:24:47,739 INFO [train.py:968] (1/2) Epoch 17, batch 9600, giga_loss[loss=0.3566, simple_loss=0.4037, pruned_loss=0.1548, over 28595.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3792, pruned_loss=0.1234, over 5671370.02 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3477, pruned_loss=0.09403, over 5720688.73 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3823, pruned_loss=0.1263, over 5660887.45 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:24:54,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 18:25:21,579 INFO [optim.py:369] (1/2) 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 18:25:24,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3703, 1.6636, 1.3547, 1.5440], device='cuda:1'), covar=tensor([0.0671, 0.0398, 0.0322, 0.0735], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 18:25:32,871 INFO [train.py:968] (1/2) Epoch 17, batch 9650, giga_loss[loss=0.2959, simple_loss=0.3716, pruned_loss=0.1101, over 29076.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3815, pruned_loss=0.1259, over 5672033.05 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3479, pruned_loss=0.09425, over 5713473.59 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3846, pruned_loss=0.1289, over 5668679.88 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:26:23,343 INFO [train.py:968] (1/2) Epoch 17, batch 9700, giga_loss[loss=0.4705, simple_loss=0.455, pruned_loss=0.243, over 23293.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3816, pruned_loss=0.1271, over 5652231.23 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3479, pruned_loss=0.09424, over 5712457.11 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3847, pruned_loss=0.1302, over 5649406.48 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:26:27,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4052, 1.7518, 1.3599, 1.6701], device='cuda:1'), covar=tensor([0.2606, 0.2610, 0.2994, 0.2340], device='cuda:1'), in_proj_covar=tensor([0.1420, 0.1033, 0.1262, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 18:26:45,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:26:45,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-08 18:26:48,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:26:57,094 INFO [optim.py:369] (1/2) 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 18:27:09,505 INFO [train.py:968] (1/2) Epoch 17, batch 9750, libri_loss[loss=0.2222, simple_loss=0.304, pruned_loss=0.07021, over 29355.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3804, pruned_loss=0.126, over 5663612.98 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3476, pruned_loss=0.09397, over 5716781.27 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.384, pruned_loss=0.1295, over 5655930.90 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:27:12,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:27:52,957 INFO [train.py:968] (1/2) Epoch 17, batch 9800, libri_loss[loss=0.2708, simple_loss=0.3546, pruned_loss=0.09349, over 29278.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.379, pruned_loss=0.1236, over 5676511.31 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.348, pruned_loss=0.09423, over 5724022.25 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3828, pruned_loss=0.1273, over 5661718.87 frames. ], batch size: 94, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:28:17,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3461, 3.4134, 1.5653, 1.4282], device='cuda:1'), covar=tensor([0.0984, 0.0373, 0.0924, 0.1402], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0539, 0.0366, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 18:28:24,263 INFO [optim.py:369] (1/2) 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 18:28:38,751 INFO [train.py:968] (1/2) Epoch 17, batch 9850, giga_loss[loss=0.3109, simple_loss=0.3895, pruned_loss=0.1162, over 28856.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3794, pruned_loss=0.1225, over 5676572.64 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.348, pruned_loss=0.09422, over 5726690.78 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3828, pruned_loss=0.1259, over 5661806.83 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:29:26,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740775.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:29:28,789 INFO [train.py:968] (1/2) Epoch 17, batch 9900, giga_loss[loss=0.331, simple_loss=0.3907, pruned_loss=0.1356, over 29045.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3795, pruned_loss=0.1225, over 5681318.36 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3475, pruned_loss=0.09398, over 5729942.38 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3833, pruned_loss=0.1259, over 5665981.02 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:29:32,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:30:03,335 INFO [optim.py:369] (1/2) 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 18:30:15,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=740828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:30:15,720 INFO [train.py:968] (1/2) Epoch 17, batch 9950, giga_loss[loss=0.2312, simple_loss=0.318, pruned_loss=0.07217, over 28691.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3792, pruned_loss=0.1234, over 5669803.31 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3473, pruned_loss=0.09393, over 5726504.24 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3835, pruned_loss=0.1272, over 5657945.51 frames. ], batch size: 60, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:30:22,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5178, 1.6086, 1.3894, 1.6355], device='cuda:1'), covar=tensor([0.0741, 0.0325, 0.0317, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0114, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0058, 0.0100], device='cuda:1')
+2023-03-08 18:30:30,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=740839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:30:30,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8832, 3.7207, 3.5355, 2.0447], device='cuda:1'), covar=tensor([0.0548, 0.0710, 0.0691, 0.1748], device='cuda:1'), in_proj_covar=tensor([0.1178, 0.1091, 0.0937, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 18:31:07,772 INFO [train.py:968] (1/2) Epoch 17, batch 10000, giga_loss[loss=0.3455, simple_loss=0.3771, pruned_loss=0.1569, over 23653.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3773, pruned_loss=0.123, over 5659912.96 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3473, pruned_loss=0.09392, over 5728326.50 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3809, pruned_loss=0.1262, over 5648365.33 frames. ], batch size: 710, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:31:39,981 INFO [optim.py:369] (1/2) 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 18:31:52,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:31:53,741 INFO [train.py:968] (1/2) Epoch 17, batch 10050, giga_loss[loss=0.3007, simple_loss=0.3732, pruned_loss=0.1142, over 28766.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3748, pruned_loss=0.1218, over 5667009.87 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3472, pruned_loss=0.0937, over 5733032.47 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3792, pruned_loss=0.126, over 5650437.57 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 18:31:55,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:23,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=740959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:35,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=740971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:39,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=740974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:32:46,615 INFO [train.py:968] (1/2) Epoch 17, batch 10100, giga_loss[loss=0.2901, simple_loss=0.3585, pruned_loss=0.1109, over 28944.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.372, pruned_loss=0.1207, over 5667105.80 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3474, pruned_loss=0.09376, over 5736434.80 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3758, pruned_loss=0.1245, over 5649837.54 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:33:11,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:33:27,594 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 10150, giga_loss[loss=0.2865, simple_loss=0.3513, pruned_loss=0.1108, over 28742.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3708, pruned_loss=0.1211, over 5661462.58 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3475, pruned_loss=0.09372, over 5738298.13 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.374, pruned_loss=0.1243, over 5645565.52 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:34:03,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0137, 1.7890, 1.8524, 1.5554], device='cuda:1'), covar=tensor([0.1539, 0.2664, 0.2025, 0.2265], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0741, 0.0696, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 18:34:15,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6019, 3.7704, 1.6595, 1.5767], device='cuda:1'), covar=tensor([0.0976, 0.0291, 0.0882, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0541, 0.0367, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 18:34:18,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-08 18:34:28,822 INFO [train.py:968] (1/2) Epoch 17, batch 10200, giga_loss[loss=0.286, simple_loss=0.3565, pruned_loss=0.1078, over 29084.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3709, pruned_loss=0.1212, over 5676192.79 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3471, pruned_loss=0.09338, over 5743258.69 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3744, pruned_loss=0.1249, over 5656899.56 frames. ], batch size: 155, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:34:59,872 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 10250, giga_loss[loss=0.2888, simple_loss=0.3643, pruned_loss=0.1066, over 28658.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3688, pruned_loss=0.119, over 5675528.25 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3474, pruned_loss=0.09364, over 5749120.42 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1226, over 5652460.40 frames. ], batch size: 262, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:35:32,514 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 10300, giga_loss[loss=0.2589, simple_loss=0.3413, pruned_loss=0.08829, over 28508.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3654, pruned_loss=0.1156, over 5661927.65 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3475, pruned_loss=0.09369, over 5747367.43 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3682, pruned_loss=0.1187, over 5644372.91 frames. ], batch size: 65, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:36:35,420 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:36:49,371 INFO [train.py:968] (1/2) Epoch 17, batch 10350, giga_loss[loss=0.2857, simple_loss=0.3613, pruned_loss=0.1051, over 28946.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3648, pruned_loss=0.1142, over 5672880.07 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3473, pruned_loss=0.09349, over 5748273.29 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3677, pruned_loss=0.1175, over 5655499.45 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:37:33,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741273.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:37:38,911 INFO [train.py:968] (1/2) Epoch 17, batch 10400, giga_loss[loss=0.2658, simple_loss=0.3381, pruned_loss=0.09675, over 28643.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3643, pruned_loss=0.1148, over 5664660.67 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3477, pruned_loss=0.09382, over 5740141.75 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3666, pruned_loss=0.1175, over 5656402.22 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:37:53,670 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741293.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:37:58,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741296.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:38:16,952 INFO [optim.py:369] (1/2) 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 18:38:19,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5814, 1.6558, 1.2569, 1.2428], device='cuda:1'), covar=tensor([0.0813, 0.0530, 0.0884, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0443, 0.0508, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:38:23,267 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 10450, giga_loss[loss=0.3005, simple_loss=0.361, pruned_loss=0.12, over 28997.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3611, pruned_loss=0.1135, over 5668493.15 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3477, pruned_loss=0.09368, over 5739862.95 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3632, pruned_loss=0.1162, over 5660775.62 frames. ], batch size: 213, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:38:51,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:38:54,989 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:38:56,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:39:13,408 INFO [train.py:968] (1/2) Epoch 17, batch 10500, giga_loss[loss=0.2879, simple_loss=0.3644, pruned_loss=0.1057, over 28828.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1143, over 5667642.84 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09367, over 5744141.88 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.364, pruned_loss=0.1169, over 5656004.07 frames. ], batch size: 174, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:39:23,326 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:39:49,027 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 10550, giga_loss[loss=0.3098, simple_loss=0.3751, pruned_loss=0.1222, over 28862.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3641, pruned_loss=0.1148, over 5668712.79 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3482, pruned_loss=0.09373, over 5746297.38 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3655, pruned_loss=0.1171, over 5656716.76 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:40:48,594 INFO [train.py:968] (1/2) Epoch 17, batch 10600, giga_loss[loss=0.281, simple_loss=0.354, pruned_loss=0.104, over 28756.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.366, pruned_loss=0.1161, over 5654262.85 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3485, pruned_loss=0.09387, over 5737974.20 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3672, pruned_loss=0.1184, over 5649383.74 frames. ], batch size: 242, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:41:01,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-08 18:41:09,833 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-08 18:41:13,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1749, 1.4783, 1.1965, 0.9253], device='cuda:1'), covar=tensor([0.2793, 0.2672, 0.3045, 0.2345], device='cuda:1'), in_proj_covar=tensor([0.1429, 0.1035, 0.1265, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 18:41:26,693 INFO [optim.py:369] (1/2) 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 18:41:39,193 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:41:39,656 INFO [train.py:968] (1/2) Epoch 17, batch 10650, giga_loss[loss=0.3143, simple_loss=0.3683, pruned_loss=0.1302, over 28438.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3661, pruned_loss=0.1166, over 5656879.63 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3484, pruned_loss=0.0937, over 5740754.31 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3675, pruned_loss=0.1191, over 5648926.57 frames. ], batch size: 85, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:42:23,398 INFO [train.py:968] (1/2) Epoch 17, batch 10700, libri_loss[loss=0.289, simple_loss=0.3709, pruned_loss=0.1036, over 29550.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3683, pruned_loss=0.1187, over 5658058.78 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3492, pruned_loss=0.09416, over 5741397.31 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.121, over 5648047.50 frames. ], batch size: 89, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:42:40,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6126, 1.7456, 1.5123, 1.9263], device='cuda:1'), covar=tensor([0.2540, 0.2724, 0.2805, 0.2386], device='cuda:1'), in_proj_covar=tensor([0.1428, 0.1035, 0.1265, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 18:43:04,773 INFO [optim.py:369] (1/2) 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 18:43:11,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7191, 1.9133, 1.3293, 1.4592], device='cuda:1'), covar=tensor([0.0867, 0.0571, 0.1048, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0445, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:43:15,798 INFO [train.py:968] (1/2) Epoch 17, batch 10750, giga_loss[loss=0.2899, simple_loss=0.3628, pruned_loss=0.1085, over 28725.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3709, pruned_loss=0.1205, over 5656777.34 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3493, pruned_loss=0.09414, over 5744670.61 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3721, pruned_loss=0.1229, over 5643953.27 frames. ], batch size: 92, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:43:35,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741648.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:44:05,541 INFO [train.py:968] (1/2) Epoch 17, batch 10800, giga_loss[loss=0.3616, simple_loss=0.3862, pruned_loss=0.1685, over 23528.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.374, pruned_loss=0.1226, over 5651579.87 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09441, over 5733596.79 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3749, pruned_loss=0.1247, over 5650595.62 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:44:40,603 INFO [optim.py:369] (1/2) 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 18:44:53,160 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:44:53,485 INFO [train.py:968] (1/2) Epoch 17, batch 10850, giga_loss[loss=0.2679, simple_loss=0.3426, pruned_loss=0.09662, over 28808.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3755, pruned_loss=0.1237, over 5668747.33 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09449, over 5736444.51 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3764, pruned_loss=0.1257, over 5664303.09 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:45:40,454 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 18:45:42,040 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:45:43,920 INFO [train.py:968] (1/2) Epoch 17, batch 10900, libri_loss[loss=0.2394, simple_loss=0.3239, pruned_loss=0.07748, over 29377.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3752, pruned_loss=0.1237, over 5675623.81 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3498, pruned_loss=0.0944, over 5740034.53 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3766, pruned_loss=0.1261, over 5667549.54 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:45:57,361 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741791.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 18:46:00,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741794.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:46:03,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7776, 1.8346, 2.0233, 1.5283], device='cuda:1'), covar=tensor([0.1695, 0.2431, 0.1383, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.0870, 0.0699, 0.0917, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 18:46:27,167 INFO [optim.py:369] (1/2) 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 18:46:30,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741823.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 18:46:36,607 INFO [train.py:968] (1/2) Epoch 17, batch 10950, giga_loss[loss=0.3135, simple_loss=0.3595, pruned_loss=0.1338, over 23864.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3767, pruned_loss=0.1239, over 5663200.79 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3497, pruned_loss=0.09443, over 5742797.37 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3783, pruned_loss=0.1261, over 5653358.56 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:47:14,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5962, 1.5745, 1.2844, 1.1999], device='cuda:1'), covar=tensor([0.0635, 0.0352, 0.0788, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0446, 0.0511, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:47:20,971 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=741871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:47:22,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6171, 1.8161, 1.2673, 1.3320], device='cuda:1'), covar=tensor([0.0877, 0.0578, 0.1042, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0446, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:47:24,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=741874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:47:28,496 INFO [train.py:968] (1/2) Epoch 17, batch 11000, giga_loss[loss=0.3729, simple_loss=0.4238, pruned_loss=0.161, over 28308.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3757, pruned_loss=0.1234, over 5661808.46 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3496, pruned_loss=0.0944, over 5745594.32 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3775, pruned_loss=0.1257, over 5650018.80 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:47:54,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=741903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:47:54,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=741903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:48:07,974 INFO [optim.py:369] (1/2) 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 18:48:18,428 INFO [train.py:968] (1/2) Epoch 17, batch 11050, giga_loss[loss=0.2833, simple_loss=0.3579, pruned_loss=0.1043, over 28967.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3751, pruned_loss=0.1237, over 5644248.03 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3499, pruned_loss=0.09461, over 5736197.31 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3769, pruned_loss=0.126, over 5641432.14 frames. ], batch size: 164, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:48:49,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=741953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:49:17,519 INFO [train.py:968] (1/2) Epoch 17, batch 11100, libri_loss[loss=0.3238, simple_loss=0.3955, pruned_loss=0.1261, over 29290.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3734, pruned_loss=0.1229, over 5642019.17 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.35, pruned_loss=0.09466, over 5737581.89 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3754, pruned_loss=0.1255, over 5635701.41 frames. ], batch size: 94, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:49:52,488 INFO [optim.py:369] (1/2) 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 18:50:01,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4116, 1.3962, 4.1067, 3.3964], device='cuda:1'), covar=tensor([0.1581, 0.2641, 0.0432, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0618, 0.0914, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:50:02,640 INFO [train.py:968] (1/2) Epoch 17, batch 11150, giga_loss[loss=0.2655, simple_loss=0.347, pruned_loss=0.09198, over 28960.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3708, pruned_loss=0.1213, over 5647375.60 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09455, over 5739684.04 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3728, pruned_loss=0.1241, over 5638516.70 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:50:21,448 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:50:23,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:50:45,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:50:50,298 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 11200, giga_loss[loss=0.3282, simple_loss=0.3833, pruned_loss=0.1366, over 28303.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3711, pruned_loss=0.1221, over 5653061.11 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09451, over 5742076.73 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3729, pruned_loss=0.1247, over 5642867.87 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:50:53,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5929, 1.6954, 1.4140, 1.5388], device='cuda:1'), covar=tensor([0.2510, 0.2551, 0.2842, 0.2265], device='cuda:1'), in_proj_covar=tensor([0.1425, 0.1034, 0.1264, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 18:51:29,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3551, 3.0865, 1.4310, 1.4943], device='cuda:1'), covar=tensor([0.1009, 0.0365, 0.0885, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0540, 0.0366, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 18:51:30,774 INFO [optim.py:369] (1/2) 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 18:51:40,106 INFO [train.py:968] (1/2) Epoch 17, batch 11250, giga_loss[loss=0.3085, simple_loss=0.371, pruned_loss=0.123, over 29050.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1215, over 5658230.15 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09448, over 5744283.97 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.372, pruned_loss=0.1241, over 5646528.31 frames. ], batch size: 106, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:52:03,485 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 11300, giga_loss[loss=0.3635, simple_loss=0.4184, pruned_loss=0.1543, over 28665.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3708, pruned_loss=0.1221, over 5664915.06 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09461, over 5750332.07 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.373, pruned_loss=0.1252, over 5646899.95 frames. ], batch size: 307, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:52:47,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3467, 1.5699, 1.3478, 1.5489], device='cuda:1'), covar=tensor([0.0755, 0.0330, 0.0317, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 18:53:04,639 INFO [optim.py:369] (1/2) 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 18:53:13,133 INFO [train.py:968] (1/2) Epoch 17, batch 11350, giga_loss[loss=0.3289, simple_loss=0.389, pruned_loss=0.1344, over 28832.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3732, pruned_loss=0.1245, over 5650006.72 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09454, over 5737024.82 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3759, pruned_loss=0.1279, over 5644488.89 frames. ], batch size: 199, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:54:03,675 INFO [train.py:968] (1/2) Epoch 17, batch 11400, giga_loss[loss=0.3528, simple_loss=0.397, pruned_loss=0.1543, over 28730.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3747, pruned_loss=0.1259, over 5644468.02 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3497, pruned_loss=0.09445, over 5738673.36 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3772, pruned_loss=0.1289, over 5638011.29 frames. ], batch size: 284, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:54:20,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:24,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:34,019 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:43,858 INFO [optim.py:369] (1/2) 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 18:54:53,654 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:55,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:54:55,699 INFO [train.py:968] (1/2) Epoch 17, batch 11450, giga_loss[loss=0.3819, simple_loss=0.4236, pruned_loss=0.1701, over 27898.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3748, pruned_loss=0.1266, over 5642337.43 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3495, pruned_loss=0.0944, over 5737852.96 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3773, pruned_loss=0.1296, over 5636348.83 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:55:07,127 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:55:07,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0917, 1.2726, 3.7668, 3.0958], device='cuda:1'), covar=tensor([0.1790, 0.2634, 0.0457, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0619, 0.0916, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 18:55:44,584 INFO [train.py:968] (1/2) Epoch 17, batch 11500, giga_loss[loss=0.3096, simple_loss=0.3695, pruned_loss=0.1248, over 28242.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.374, pruned_loss=0.1257, over 5658336.31 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3495, pruned_loss=0.09432, over 5741191.17 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3766, pruned_loss=0.1287, over 5649103.29 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:56:25,347 INFO [optim.py:369] (1/2) 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 18:56:34,820 INFO [train.py:968] (1/2) Epoch 17, batch 11550, giga_loss[loss=0.4114, simple_loss=0.4285, pruned_loss=0.1972, over 26476.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3747, pruned_loss=0.1263, over 5643068.73 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3495, pruned_loss=0.09437, over 5734528.37 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3772, pruned_loss=0.1292, over 5640450.98 frames. ], batch size: 555, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 18:56:53,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:15,967 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:17,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:22,602 INFO [train.py:968] (1/2) Epoch 17, batch 11600, giga_loss[loss=0.3305, simple_loss=0.388, pruned_loss=0.1365, over 27831.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3748, pruned_loss=0.1253, over 5662119.43 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3499, pruned_loss=0.09464, over 5736801.41 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3766, pruned_loss=0.1278, over 5656885.08 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:57:47,987 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:57:59,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3196, 1.4377, 1.3492, 1.2366], device='cuda:1'), covar=tensor([0.1884, 0.1718, 0.1617, 0.1821], device='cuda:1'), in_proj_covar=tensor([0.1872, 0.1832, 0.1744, 0.1884], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 18:57:59,660 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-08 18:58:08,812 INFO [optim.py:369] (1/2) 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 18:58:16,881 INFO [train.py:968] (1/2) Epoch 17, batch 11650, giga_loss[loss=0.3641, simple_loss=0.409, pruned_loss=0.1596, over 27445.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3775, pruned_loss=0.128, over 5650928.66 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3501, pruned_loss=0.09472, over 5739216.07 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3791, pruned_loss=0.1303, over 5643593.55 frames. ], batch size: 472, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:58:39,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2957, 1.4455, 1.4241, 1.2344], device='cuda:1'), covar=tensor([0.2174, 0.1906, 0.1470, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.1865, 0.1828, 0.1738, 0.1879], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 18:59:05,905 INFO [train.py:968] (1/2) Epoch 17, batch 11700, giga_loss[loss=0.3228, simple_loss=0.395, pruned_loss=0.1253, over 28964.00 frames. ], tot_loss[loss=0.317, simple_loss=0.378, pruned_loss=0.128, over 5654915.89 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3502, pruned_loss=0.09477, over 5741916.80 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.38, pruned_loss=0.1308, over 5644206.92 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 18:59:17,097 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:59:18,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-08 18:59:18,907 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:59:45,977 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=742623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 18:59:45,992 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 11750, giga_loss[loss=0.2802, simple_loss=0.3556, pruned_loss=0.1025, over 28971.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1279, over 5651875.75 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3495, pruned_loss=0.09444, over 5741064.88 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.38, pruned_loss=0.1314, over 5641431.52 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:00:02,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2412, 3.1631, 1.4571, 1.4419], device='cuda:1'), covar=tensor([0.1015, 0.0393, 0.0891, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0542, 0.0367, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 19:00:36,931 INFO [train.py:968] (1/2) Epoch 17, batch 11800, giga_loss[loss=0.2511, simple_loss=0.3389, pruned_loss=0.08163, over 29159.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3766, pruned_loss=0.1261, over 5659458.95 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.349, pruned_loss=0.09406, over 5744550.07 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3801, pruned_loss=0.13, over 5646242.29 frames. ], batch size: 128, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:00:41,141 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-08 19:00:44,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:01:01,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4888, 1.6618, 1.6168, 1.4267], device='cuda:1'), covar=tensor([0.1834, 0.2132, 0.2232, 0.2203], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0748, 0.0703, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 19:01:13,454 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 17, batch 11850, giga_loss[loss=0.2819, simple_loss=0.3543, pruned_loss=0.1047, over 28942.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3762, pruned_loss=0.1248, over 5660409.48 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3488, pruned_loss=0.09388, over 5746564.56 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3797, pruned_loss=0.1287, over 5646383.22 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:01:56,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5711, 1.5409, 1.2217, 1.1690], device='cuda:1'), covar=tensor([0.0654, 0.0355, 0.0826, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0447, 0.0512, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:02:16,290 INFO [train.py:968] (1/2) Epoch 17, batch 11900, giga_loss[loss=0.3168, simple_loss=0.3601, pruned_loss=0.1368, over 23619.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3759, pruned_loss=0.1247, over 5648722.40 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.349, pruned_loss=0.09398, over 5739922.73 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.379, pruned_loss=0.1282, over 5642561.12 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:02:53,324 INFO [optim.py:369] (1/2) 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 19:02:58,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:02:59,404 INFO [train.py:968] (1/2) Epoch 17, batch 11950, giga_loss[loss=0.3274, simple_loss=0.3907, pruned_loss=0.1321, over 28877.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3734, pruned_loss=0.1227, over 5654217.64 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.349, pruned_loss=0.09384, over 5740178.85 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3768, pruned_loss=0.1267, over 5646106.08 frames. ], batch size: 145, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:03:00,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=742830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:03:27,644 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=742857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:03:28,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:03:31,519 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 12000, giga_loss[loss=0.3268, simple_loss=0.3853, pruned_loss=0.1341, over 28877.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3746, pruned_loss=0.1234, over 5656412.32 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3491, pruned_loss=0.09391, over 5736802.26 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3777, pruned_loss=0.1272, over 5650977.74 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:03:48,687 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 19:03:54,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5242, 1.7583, 1.2878, 1.3418], device='cuda:1'), covar=tensor([0.0916, 0.0468, 0.0989, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0445, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:03:57,324 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 19:03:59,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6484, 3.4661, 3.3088, 1.9278], device='cuda:1'), covar=tensor([0.0640, 0.0848, 0.0792, 0.1851], device='cuda:1'), in_proj_covar=tensor([0.1179, 0.1091, 0.0938, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 19:03:59,997 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-08 19:04:07,109 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=742889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:04:35,769 INFO [optim.py:369] (1/2) 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 19:04:44,189 INFO [train.py:968] (1/2) Epoch 17, batch 12050, giga_loss[loss=0.4065, simple_loss=0.4171, pruned_loss=0.1979, over 23640.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3747, pruned_loss=0.1238, over 5642414.74 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3491, pruned_loss=0.09393, over 5730869.90 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3777, pruned_loss=0.1273, over 5641944.38 frames. ], batch size: 705, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:04:48,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5113, 1.6482, 1.1776, 1.2271], device='cuda:1'), covar=tensor([0.0831, 0.0519, 0.0979, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0447, 0.0511, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:05:32,090 INFO [train.py:968] (1/2) Epoch 17, batch 12100, giga_loss[loss=0.3452, simple_loss=0.3934, pruned_loss=0.1485, over 28317.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3732, pruned_loss=0.1234, over 5661741.12 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3493, pruned_loss=0.09418, over 5734797.21 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.376, pruned_loss=0.1266, over 5656117.44 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:05:49,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0125, 2.1575, 1.4600, 1.7139], device='cuda:1'), covar=tensor([0.0962, 0.0688, 0.1070, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0446, 0.0510, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:05:51,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=742998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:06:04,572 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-08 19:06:13,190 INFO [optim.py:369] (1/2) 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 19:06:20,103 INFO [train.py:968] (1/2) Epoch 17, batch 12150, giga_loss[loss=0.3229, simple_loss=0.3804, pruned_loss=0.1327, over 27861.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1246, over 5668424.46 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3496, pruned_loss=0.09434, over 5739036.77 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1278, over 5658562.52 frames. ], batch size: 412, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:07:06,746 INFO [train.py:968] (1/2) Epoch 17, batch 12200, giga_loss[loss=0.3185, simple_loss=0.3844, pruned_loss=0.1263, over 28914.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3757, pruned_loss=0.1252, over 5672108.02 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09435, over 5741870.08 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3786, pruned_loss=0.129, over 5659173.71 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:07:30,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3761, 1.6000, 1.4471, 1.5712], device='cuda:1'), covar=tensor([0.0799, 0.0327, 0.0314, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 19:07:45,785 INFO [optim.py:369] (1/2) 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 19:07:52,938 INFO [train.py:968] (1/2) Epoch 17, batch 12250, libri_loss[loss=0.2484, simple_loss=0.3258, pruned_loss=0.08548, over 29556.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1252, over 5662010.34 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3497, pruned_loss=0.09442, over 5734072.07 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3795, pruned_loss=0.1297, over 5655544.67 frames. ], batch size: 77, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:07:59,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=743136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:08:04,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=743141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:08:06,713 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 17, batch 12300, giga_loss[loss=0.3342, simple_loss=0.3932, pruned_loss=0.1376, over 28553.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3751, pruned_loss=0.1245, over 5675187.18 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3496, pruned_loss=0.09446, over 5739022.94 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3789, pruned_loss=0.129, over 5663512.74 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:09:00,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4476, 1.6247, 1.6991, 1.2659], device='cuda:1'), covar=tensor([0.1656, 0.2380, 0.1352, 0.1584], device='cuda:1'), in_proj_covar=tensor([0.0869, 0.0698, 0.0917, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 19:09:21,961 INFO [optim.py:369] (1/2) 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 19:09:28,726 INFO [train.py:968] (1/2) Epoch 17, batch 12350, giga_loss[loss=0.3106, simple_loss=0.3763, pruned_loss=0.1224, over 28282.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.374, pruned_loss=0.1232, over 5664092.26 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3498, pruned_loss=0.09453, over 5742477.75 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3774, pruned_loss=0.1273, over 5650176.33 frames. ], batch size: 368, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:10:15,904 INFO [train.py:968] (1/2) Epoch 17, batch 12400, giga_loss[loss=0.297, simple_loss=0.3709, pruned_loss=0.1116, over 28713.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3747, pruned_loss=0.1231, over 5674478.15 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09452, over 5743302.36 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3773, pruned_loss=0.1264, over 5662610.28 frames. ], batch size: 243, lr: 1.90e-03, grad_scale: 8.0
+2023-03-08 19:10:20,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3830, 1.0905, 4.4756, 3.4832], device='cuda:1'), covar=tensor([0.1763, 0.2983, 0.0434, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0624, 0.0923, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:10:58,313 INFO [optim.py:369] (1/2) 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 19:11:05,024 INFO [train.py:968] (1/2) Epoch 17, batch 12450, libri_loss[loss=0.2969, simple_loss=0.3727, pruned_loss=0.1106, over 27789.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3744, pruned_loss=0.1225, over 5687681.88 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3503, pruned_loss=0.09466, over 5745047.65 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3767, pruned_loss=0.1257, over 5675295.32 frames. ], batch size: 116, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:11:18,799 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=743342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:11:46,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7708, 1.9745, 1.4220, 1.5154], device='cuda:1'), covar=tensor([0.0837, 0.0527, 0.0975, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0444, 0.0507, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:11:50,313 INFO [train.py:968] (1/2) Epoch 17, batch 12500, giga_loss[loss=0.2765, simple_loss=0.3376, pruned_loss=0.1076, over 28880.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3737, pruned_loss=0.1228, over 5678576.62 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3504, pruned_loss=0.09469, over 5746643.49 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3762, pruned_loss=0.1263, over 5664872.89 frames. ], batch size: 112, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:12:26,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9804, 1.0088, 3.7052, 3.0969], device='cuda:1'), covar=tensor([0.2181, 0.3128, 0.0833, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0625, 0.0923, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:12:28,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6625, 1.7445, 1.3099, 1.2670], device='cuda:1'), covar=tensor([0.0865, 0.0564, 0.1001, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0445, 0.0509, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:12:31,323 INFO [optim.py:369] (1/2) 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 19:12:33,107 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-08 19:12:37,656 INFO [train.py:968] (1/2) Epoch 17, batch 12550, giga_loss[loss=0.2535, simple_loss=0.325, pruned_loss=0.09099, over 28475.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3715, pruned_loss=0.1217, over 5666288.38 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3504, pruned_loss=0.09475, over 5735999.77 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3741, pruned_loss=0.1251, over 5663174.25 frames. ], batch size: 71, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:13:26,300 INFO [train.py:968] (1/2) Epoch 17, batch 12600, giga_loss[loss=0.3504, simple_loss=0.3913, pruned_loss=0.1547, over 29018.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3688, pruned_loss=0.1212, over 5668794.22 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.35, pruned_loss=0.09446, over 5729773.49 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3717, pruned_loss=0.1248, over 5669667.32 frames. ], batch size: 136, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:13:58,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=743511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:14:09,034 INFO [optim.py:369] (1/2) 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 19:14:13,801 INFO [train.py:968] (1/2) Epoch 17, batch 12650, giga_loss[loss=0.3632, simple_loss=0.4086, pruned_loss=0.1589, over 28884.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3678, pruned_loss=0.1211, over 5678995.45 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09448, over 5733091.11 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3703, pruned_loss=0.1244, over 5675700.06 frames. ], batch size: 186, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:14:29,923 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-08 19:14:31,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 19:15:05,619 INFO [train.py:968] (1/2) Epoch 17, batch 12700, libri_loss[loss=0.2552, simple_loss=0.341, pruned_loss=0.08474, over 29559.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3661, pruned_loss=0.1199, over 5688512.82 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3503, pruned_loss=0.09449, over 5736040.81 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3682, pruned_loss=0.123, over 5682272.33 frames. ], batch size: 79, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:15:50,672 INFO [optim.py:369] (1/2) 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 19:15:55,799 INFO [train.py:968] (1/2) Epoch 17, batch 12750, libri_loss[loss=0.2606, simple_loss=0.3311, pruned_loss=0.09505, over 29644.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3663, pruned_loss=0.119, over 5686965.07 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3497, pruned_loss=0.09424, over 5738684.34 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3689, pruned_loss=0.1222, over 5678409.74 frames. ], batch size: 73, lr: 1.90e-03, grad_scale: 2.0
+2023-03-08 19:16:00,726 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 19:16:03,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3478, 1.8474, 1.4080, 0.6602], device='cuda:1'), covar=tensor([0.4781, 0.2660, 0.3003, 0.5316], device='cuda:1'), in_proj_covar=tensor([0.1673, 0.1590, 0.1555, 0.1376], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 19:16:21,655 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=743654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:16:23,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=743657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:16:26,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2045, 0.8292, 0.8503, 1.4491], device='cuda:1'), covar=tensor([0.0816, 0.0378, 0.0378, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 19:16:31,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4451, 1.9929, 1.4061, 0.8154], device='cuda:1'), covar=tensor([0.4609, 0.2675, 0.3626, 0.5085], device='cuda:1'), in_proj_covar=tensor([0.1669, 0.1586, 0.1553, 0.1374], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 19:16:45,369 INFO [train.py:968] (1/2) Epoch 17, batch 12800, giga_loss[loss=0.308, simple_loss=0.3851, pruned_loss=0.1155, over 28917.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.116, over 5685659.29 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3494, pruned_loss=0.09402, over 5743075.05 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3681, pruned_loss=0.1193, over 5673502.91 frames. ], batch size: 164, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:16:52,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:17:09,067 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7097, 1.2920, 4.8783, 3.5408], device='cuda:1'), covar=tensor([0.1651, 0.3005, 0.0394, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0625, 0.0923, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:17:26,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=743717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:17:31,516 INFO [optim.py:369] (1/2) 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 19:17:38,352 INFO [train.py:968] (1/2) Epoch 17, batch 12850, giga_loss[loss=0.2786, simple_loss=0.355, pruned_loss=0.1011, over 28529.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3618, pruned_loss=0.1125, over 5677385.23 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09385, over 5747011.74 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3647, pruned_loss=0.1158, over 5662996.19 frames. ], batch size: 336, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:18:30,064 INFO [train.py:968] (1/2) Epoch 17, batch 12900, giga_loss[loss=0.25, simple_loss=0.329, pruned_loss=0.08552, over 28736.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3587, pruned_loss=0.1093, over 5677438.91 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3485, pruned_loss=0.09379, over 5749746.30 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3616, pruned_loss=0.1123, over 5662400.47 frames. ], batch size: 99, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:19:19,061 INFO [optim.py:369] (1/2) 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 19:19:24,945 INFO [train.py:968] (1/2) Epoch 17, batch 12950, giga_loss[loss=0.2457, simple_loss=0.3393, pruned_loss=0.07606, over 28763.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3554, pruned_loss=0.1062, over 5677443.39 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09364, over 5752396.77 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3583, pruned_loss=0.1089, over 5661927.73 frames. ], batch size: 119, lr: 1.90e-03, grad_scale: 4.0
+2023-03-08 19:19:54,971 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=743860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:19:58,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=743863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:19:58,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6654, 2.7203, 1.7089, 0.6382], device='cuda:1'), covar=tensor([0.7695, 0.3252, 0.3858, 0.6852], device='cuda:1'), in_proj_covar=tensor([0.1670, 0.1584, 0.1553, 0.1376], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 19:20:13,502 INFO [train.py:968] (1/2) Epoch 17, batch 13000, giga_loss[loss=0.269, simple_loss=0.3571, pruned_loss=0.09045, over 28924.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3541, pruned_loss=0.1034, over 5678503.77 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3478, pruned_loss=0.09365, over 5755408.27 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3568, pruned_loss=0.1058, over 5661525.28 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:20:26,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2219, 1.0127, 1.0789, 1.4405], device='cuda:1'), covar=tensor([0.0723, 0.0361, 0.0328, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 19:20:29,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:21:03,838 INFO [optim.py:369] (1/2) 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 19:21:08,910 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4071, 1.6258, 1.4053, 1.3754], device='cuda:1'), covar=tensor([0.1651, 0.1447, 0.1550, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.1850, 0.1805, 0.1722, 0.1861], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 19:21:09,841 INFO [train.py:968] (1/2) Epoch 17, batch 13050, giga_loss[loss=0.264, simple_loss=0.3444, pruned_loss=0.0918, over 28735.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3543, pruned_loss=0.1036, over 5665755.53 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3474, pruned_loss=0.0935, over 5757092.45 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3569, pruned_loss=0.1059, over 5649296.81 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:21:36,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=743955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:21:57,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9372, 1.2663, 1.2836, 1.1080], device='cuda:1'), covar=tensor([0.1284, 0.0776, 0.1440, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0738, 0.0694, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 19:22:02,127 INFO [train.py:968] (1/2) Epoch 17, batch 13100, giga_loss[loss=0.2513, simple_loss=0.3321, pruned_loss=0.08522, over 28796.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3537, pruned_loss=0.103, over 5669808.37 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3472, pruned_loss=0.09349, over 5758269.30 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.356, pruned_loss=0.1049, over 5654595.43 frames. ], batch size: 284, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:22:45,922 INFO [optim.py:369] (1/2) 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 19:22:47,840 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 13150, giga_loss[loss=0.3021, simple_loss=0.3524, pruned_loss=0.1259, over 26696.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3506, pruned_loss=0.101, over 5671679.68 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3467, pruned_loss=0.09338, over 5761509.55 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3529, pruned_loss=0.1028, over 5655203.06 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:23:42,350 INFO [train.py:968] (1/2) Epoch 17, batch 13200, giga_loss[loss=0.3013, simple_loss=0.3711, pruned_loss=0.1158, over 28844.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09977, over 5672146.94 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3464, pruned_loss=0.09329, over 5762944.62 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3507, pruned_loss=0.1013, over 5657165.83 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:24:28,601 INFO [optim.py:369] (1/2) 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 19:24:31,947 INFO [train.py:968] (1/2) Epoch 17, batch 13250, giga_loss[loss=0.2592, simple_loss=0.3383, pruned_loss=0.09003, over 28581.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3486, pruned_loss=0.09918, over 5674722.92 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3466, pruned_loss=0.0934, over 5761073.04 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3502, pruned_loss=0.1005, over 5662200.47 frames. ], batch size: 242, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:25:21,234 INFO [train.py:968] (1/2) Epoch 17, batch 13300, giga_loss[loss=0.2541, simple_loss=0.3372, pruned_loss=0.08543, over 28902.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3468, pruned_loss=0.09779, over 5669913.18 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.346, pruned_loss=0.09307, over 5759645.70 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3487, pruned_loss=0.0992, over 5659706.70 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:25:38,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6910, 4.4841, 4.2880, 1.9632], device='cuda:1'), covar=tensor([0.0674, 0.0875, 0.0930, 0.2180], device='cuda:1'), in_proj_covar=tensor([0.1148, 0.1064, 0.0911, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 19:25:38,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3190, 1.6657, 1.4284, 1.4858], device='cuda:1'), covar=tensor([0.0772, 0.0305, 0.0328, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 19:26:08,330 INFO [optim.py:369] (1/2) 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 19:26:14,127 INFO [train.py:968] (1/2) Epoch 17, batch 13350, giga_loss[loss=0.2338, simple_loss=0.3197, pruned_loss=0.07393, over 28743.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.344, pruned_loss=0.09564, over 5666058.70 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3454, pruned_loss=0.09286, over 5754216.62 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3459, pruned_loss=0.09707, over 5660350.72 frames. ], batch size: 119, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:26:49,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9425, 2.3735, 1.4493, 1.9308], device='cuda:1'), covar=tensor([0.0964, 0.0567, 0.0996, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0440, 0.0506, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 19:26:50,849 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:27:08,223 INFO [train.py:968] (1/2) Epoch 17, batch 13400, giga_loss[loss=0.245, simple_loss=0.3255, pruned_loss=0.08229, over 28639.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3394, pruned_loss=0.09297, over 5662701.82 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3448, pruned_loss=0.09266, over 5757512.92 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3415, pruned_loss=0.09436, over 5652447.99 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:28:01,231 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 13450, giga_loss[loss=0.2434, simple_loss=0.325, pruned_loss=0.08088, over 28896.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3377, pruned_loss=0.09283, over 5651429.47 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3443, pruned_loss=0.09243, over 5760970.97 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3397, pruned_loss=0.09416, over 5638006.59 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 19:28:05,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:28:12,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 19:28:52,018 INFO [train.py:968] (1/2) Epoch 17, batch 13500, giga_loss[loss=0.2404, simple_loss=0.3147, pruned_loss=0.08306, over 28597.00 frames. ], tot_loss[loss=0.263, simple_loss=0.338, pruned_loss=0.09398, over 5643456.34 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3441, pruned_loss=0.09261, over 5747089.77 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3395, pruned_loss=0.09492, over 5641333.93 frames. ], batch size: 85, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 19:29:10,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:29:13,191 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:29:37,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2393, 1.3247, 3.0684, 2.8787], device='cuda:1'), covar=tensor([0.1339, 0.2520, 0.0484, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0619, 0.0912, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:29:42,371 INFO [optim.py:369] (1/2) 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 19:29:45,129 INFO [train.py:968] (1/2) Epoch 17, batch 13550, giga_loss[loss=0.2719, simple_loss=0.3532, pruned_loss=0.09528, over 28522.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3386, pruned_loss=0.09432, over 5618974.85 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3436, pruned_loss=0.09269, over 5733707.09 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.34, pruned_loss=0.09507, over 5624968.38 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 19:30:37,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:30:40,498 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 13600, giga_loss[loss=0.2451, simple_loss=0.3316, pruned_loss=0.07928, over 28897.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3407, pruned_loss=0.09392, over 5639485.31 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3434, pruned_loss=0.0927, over 5738086.15 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.09453, over 5638037.11 frames. ], batch size: 112, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:30:58,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2366, 0.7940, 0.9330, 1.4174], device='cuda:1'), covar=tensor([0.0768, 0.0374, 0.0364, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0114, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 19:31:16,227 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:31:39,829 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 13650, giga_loss[loss=0.2409, simple_loss=0.3219, pruned_loss=0.07999, over 28673.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.09449, over 5638438.54 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3437, pruned_loss=0.09293, over 5739464.50 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3427, pruned_loss=0.09476, over 5634837.13 frames. ], batch size: 85, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:31:58,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:32:01,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=744544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:32:35,866 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:32:42,248 INFO [train.py:968] (1/2) Epoch 17, batch 13700, giga_loss[loss=0.2649, simple_loss=0.3348, pruned_loss=0.09753, over 28714.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3409, pruned_loss=0.09405, over 5647768.55 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3423, pruned_loss=0.09229, over 5747835.03 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3426, pruned_loss=0.09495, over 5632456.72 frames. ], batch size: 242, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:33:35,565 INFO [optim.py:369] (1/2) 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 19:33:42,651 INFO [train.py:968] (1/2) Epoch 17, batch 13750, giga_loss[loss=0.2451, simple_loss=0.3322, pruned_loss=0.079, over 28649.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.339, pruned_loss=0.09237, over 5657251.30 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3425, pruned_loss=0.09241, over 5750713.64 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3401, pruned_loss=0.09299, over 5640950.93 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:33:49,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:33:56,452 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:34:43,044 INFO [train.py:968] (1/2) Epoch 17, batch 13800, giga_loss[loss=0.258, simple_loss=0.3491, pruned_loss=0.08346, over 28649.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3385, pruned_loss=0.09088, over 5652810.58 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3426, pruned_loss=0.09252, over 5752134.96 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3393, pruned_loss=0.09126, over 5637530.16 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:35:39,826 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 19:35:43,973 INFO [optim.py:369] (1/2) 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 19:35:50,503 INFO [train.py:968] (1/2) Epoch 17, batch 13850, giga_loss[loss=0.2475, simple_loss=0.3231, pruned_loss=0.08594, over 28859.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3349, pruned_loss=0.08923, over 5656131.26 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3424, pruned_loss=0.09248, over 5754351.95 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3356, pruned_loss=0.08952, over 5640718.37 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:36:49,106 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=744776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:36:51,594 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:36:51,896 INFO [train.py:968] (1/2) Epoch 17, batch 13900, giga_loss[loss=0.2586, simple_loss=0.3367, pruned_loss=0.0903, over 28913.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08954, over 5662998.02 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3421, pruned_loss=0.09242, over 5756529.71 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3347, pruned_loss=0.08977, over 5647591.93 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:36:55,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=744781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:37:24,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 19:37:29,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:37:49,072 INFO [optim.py:369] (1/2) 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 19:37:52,133 INFO [train.py:968] (1/2) Epoch 17, batch 13950, giga_loss[loss=0.2844, simple_loss=0.358, pruned_loss=0.1055, over 27771.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3337, pruned_loss=0.08949, over 5667448.81 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3421, pruned_loss=0.09249, over 5757991.38 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08953, over 5652259.72 frames. ], batch size: 474, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:37:54,882 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-08 19:38:27,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2158, 1.4913, 0.8580, 1.1332], device='cuda:1'), covar=tensor([0.1206, 0.0617, 0.1681, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0439, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:38:28,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5111, 1.7391, 1.7846, 1.3199], device='cuda:1'), covar=tensor([0.1834, 0.2545, 0.1504, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0691, 0.0918, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 19:38:40,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3855, 1.6361, 1.2968, 1.3360], device='cuda:1'), covar=tensor([0.2541, 0.2381, 0.2754, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1029, 0.1264, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 19:38:52,236 INFO [train.py:968] (1/2) Epoch 17, batch 14000, giga_loss[loss=0.2495, simple_loss=0.339, pruned_loss=0.08002, over 28558.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3352, pruned_loss=0.08935, over 5666505.22 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3417, pruned_loss=0.0923, over 5750990.00 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3357, pruned_loss=0.0895, over 5659848.45 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:38:59,448 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:00,018 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=744884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:42,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=744919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:46,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=744922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:39:50,635 INFO [optim.py:369] (1/2) 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 19:39:55,326 INFO [train.py:968] (1/2) Epoch 17, batch 14050, giga_loss[loss=0.2662, simple_loss=0.3486, pruned_loss=0.09195, over 28509.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3375, pruned_loss=0.08978, over 5676772.51 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3418, pruned_loss=0.0924, over 5754102.88 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3377, pruned_loss=0.08977, over 5666936.58 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:40:28,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=744951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:41:04,084 INFO [train.py:968] (1/2) Epoch 17, batch 14100, giga_loss[loss=0.213, simple_loss=0.2995, pruned_loss=0.06318, over 28643.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.08716, over 5680313.86 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3415, pruned_loss=0.09221, over 5757232.32 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3334, pruned_loss=0.08725, over 5668507.08 frames. ], batch size: 85, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:41:15,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7812, 1.3839, 5.2957, 3.7324], device='cuda:1'), covar=tensor([0.1560, 0.2727, 0.0343, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0622, 0.0912, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 19:41:32,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-08 19:41:54,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745015.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:42:06,892 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 14150, giga_loss[loss=0.3054, simple_loss=0.3595, pruned_loss=0.1257, over 26875.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3354, pruned_loss=0.08924, over 5677385.14 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3416, pruned_loss=0.09249, over 5759437.59 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3354, pruned_loss=0.08899, over 5665076.97 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:42:17,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1561, 4.9658, 4.7542, 2.2930], device='cuda:1'), covar=tensor([0.0390, 0.0566, 0.0581, 0.1872], device='cuda:1'), in_proj_covar=tensor([0.1146, 0.1056, 0.0904, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 19:43:24,213 INFO [train.py:968] (1/2) Epoch 17, batch 14200, giga_loss[loss=0.2341, simple_loss=0.3392, pruned_loss=0.06448, over 28846.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3387, pruned_loss=0.0893, over 5663333.56 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3416, pruned_loss=0.09253, over 5761127.60 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3387, pruned_loss=0.08904, over 5651444.03 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:44:25,067 INFO [optim.py:369] (1/2) 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 19:44:30,845 INFO [train.py:968] (1/2) Epoch 17, batch 14250, giga_loss[loss=0.35, simple_loss=0.3871, pruned_loss=0.1564, over 26972.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3412, pruned_loss=0.08878, over 5655160.04 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3416, pruned_loss=0.09253, over 5761127.60 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3412, pruned_loss=0.08858, over 5645906.27 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:44:39,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0099, 2.8524, 2.7237, 1.6296], device='cuda:1'), covar=tensor([0.1011, 0.1114, 0.0945, 0.1975], device='cuda:1'), in_proj_covar=tensor([0.1145, 0.1054, 0.0902, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 19:45:06,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:45:10,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745161.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 19:45:30,815 INFO [train.py:968] (1/2) Epoch 17, batch 14300, giga_loss[loss=0.2889, simple_loss=0.3584, pruned_loss=0.1097, over 27601.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3419, pruned_loss=0.08839, over 5657811.54 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3417, pruned_loss=0.09251, over 5763957.91 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3418, pruned_loss=0.08816, over 5645376.31 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:45:46,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745190.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 19:45:57,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8573, 2.0294, 1.4595, 1.6696], device='cuda:1'), covar=tensor([0.0976, 0.0712, 0.1096, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0438, 0.0506, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 19:46:35,249 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 14350, giga_loss[loss=0.2495, simple_loss=0.3258, pruned_loss=0.08662, over 28546.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.342, pruned_loss=0.0884, over 5661797.74 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3416, pruned_loss=0.09248, over 5764700.12 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.342, pruned_loss=0.08823, over 5650946.24 frames. ], batch size: 85, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:47:14,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:47:16,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:47:40,461 INFO [train.py:968] (1/2) Epoch 17, batch 14400, libri_loss[loss=0.2711, simple_loss=0.3444, pruned_loss=0.09891, over 29527.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3402, pruned_loss=0.08859, over 5670631.53 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3408, pruned_loss=0.09211, over 5768421.00 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3409, pruned_loss=0.08868, over 5655848.51 frames. ], batch size: 81, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:47:48,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7008, 1.9402, 1.8874, 1.5969], device='cuda:1'), covar=tensor([0.1627, 0.1682, 0.1815, 0.1816], device='cuda:1'), in_proj_covar=tensor([0.0445, 0.0727, 0.0685, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 19:48:37,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3652, 1.7632, 1.6792, 1.5042], device='cuda:1'), covar=tensor([0.1781, 0.1755, 0.1990, 0.1894], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0725, 0.0684, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 19:48:42,595 INFO [optim.py:369] (1/2) 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 19:48:45,886 INFO [train.py:968] (1/2) Epoch 17, batch 14450, giga_loss[loss=0.2559, simple_loss=0.3413, pruned_loss=0.08528, over 28906.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3406, pruned_loss=0.09, over 5670142.59 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3407, pruned_loss=0.09223, over 5767773.24 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3413, pruned_loss=0.08987, over 5656628.28 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:50:07,044 INFO [train.py:968] (1/2) Epoch 17, batch 14500, giga_loss[loss=0.2342, simple_loss=0.3107, pruned_loss=0.07887, over 28164.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3392, pruned_loss=0.08968, over 5680856.51 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3403, pruned_loss=0.09198, over 5768420.63 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3401, pruned_loss=0.08976, over 5666778.65 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:50:46,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:50:46,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:50:49,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:50:51,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:51:25,466 INFO [optim.py:369] (1/2) 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 19:51:26,227 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 14550, giga_loss[loss=0.2647, simple_loss=0.3449, pruned_loss=0.09226, over 28659.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3339, pruned_loss=0.08658, over 5670177.84 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.34, pruned_loss=0.09183, over 5769603.01 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3349, pruned_loss=0.08674, over 5657237.39 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:51:33,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:51:34,260 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:52:35,770 INFO [train.py:968] (1/2) Epoch 17, batch 14600, libri_loss[loss=0.241, simple_loss=0.3261, pruned_loss=0.07799, over 29658.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3311, pruned_loss=0.08512, over 5677877.33 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3397, pruned_loss=0.09164, over 5772721.89 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.332, pruned_loss=0.08527, over 5662370.07 frames. ], batch size: 91, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:53:22,100 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=745516.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 19:53:34,905 INFO [optim.py:369] (1/2) 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 19:53:37,045 INFO [train.py:968] (1/2) Epoch 17, batch 14650, giga_loss[loss=0.2703, simple_loss=0.3394, pruned_loss=0.1006, over 26813.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3335, pruned_loss=0.08671, over 5684718.64 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3395, pruned_loss=0.09162, over 5775259.94 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3341, pruned_loss=0.08674, over 5668647.68 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:54:03,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5268, 1.6901, 1.4193, 1.5830], device='cuda:1'), covar=tensor([0.3020, 0.2634, 0.3104, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1026, 0.1263, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 19:54:38,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4119, 1.7005, 1.5303, 1.6263], device='cuda:1'), covar=tensor([0.0672, 0.0281, 0.0292, 0.0693], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 19:54:41,777 INFO [train.py:968] (1/2) Epoch 17, batch 14700, giga_loss[loss=0.2797, simple_loss=0.3613, pruned_loss=0.09904, over 28933.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3378, pruned_loss=0.0893, over 5684351.88 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3386, pruned_loss=0.09117, over 5778280.22 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3391, pruned_loss=0.08965, over 5666859.96 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:55:20,598 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=745611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 19:55:39,597 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 14750, libri_loss[loss=0.2095, simple_loss=0.2871, pruned_loss=0.06591, over 28558.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3354, pruned_loss=0.08905, over 5688729.12 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3383, pruned_loss=0.09118, over 5780153.44 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3367, pruned_loss=0.0893, over 5671162.89 frames. ], batch size: 63, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:55:57,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6617, 1.7878, 1.9493, 1.4786], device='cuda:1'), covar=tensor([0.1604, 0.2384, 0.1338, 0.1721], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0689, 0.0912, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 19:56:46,226 INFO [train.py:968] (1/2) Epoch 17, batch 14800, giga_loss[loss=0.2691, simple_loss=0.346, pruned_loss=0.09614, over 28753.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3374, pruned_loss=0.09116, over 5683435.24 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3378, pruned_loss=0.09095, over 5783368.32 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09155, over 5664477.07 frames. ], batch size: 263, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 19:57:46,312 INFO [optim.py:369] (1/2) 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 19:57:47,261 INFO [train.py:968] (1/2) Epoch 17, batch 14850, giga_loss[loss=0.3436, simple_loss=0.4049, pruned_loss=0.1411, over 29069.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3385, pruned_loss=0.09187, over 5676275.85 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3377, pruned_loss=0.09095, over 5780185.99 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3397, pruned_loss=0.0922, over 5660958.70 frames. ], batch size: 128, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:58:59,546 INFO [train.py:968] (1/2) Epoch 17, batch 14900, giga_loss[loss=0.2802, simple_loss=0.3567, pruned_loss=0.1018, over 29033.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3402, pruned_loss=0.09153, over 5676133.51 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3377, pruned_loss=0.09091, over 5779739.50 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3412, pruned_loss=0.09184, over 5663406.23 frames. ], batch size: 285, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 19:59:31,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9067, 3.7650, 3.4905, 1.6615], device='cuda:1'), covar=tensor([0.0659, 0.0781, 0.0812, 0.2406], device='cuda:1'), in_proj_covar=tensor([0.1142, 0.1054, 0.0903, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 19:59:34,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:00:14,520 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=745826.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:00:17,229 INFO [optim.py:369] (1/2) 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 20:00:18,015 INFO [train.py:968] (1/2) Epoch 17, batch 14950, giga_loss[loss=0.2656, simple_loss=0.3424, pruned_loss=0.09442, over 28058.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3402, pruned_loss=0.09128, over 5670852.78 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3374, pruned_loss=0.09076, over 5778418.57 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3413, pruned_loss=0.09166, over 5660104.73 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:00:59,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 20:01:32,773 INFO [train.py:968] (1/2) Epoch 17, batch 15000, giga_loss[loss=0.2493, simple_loss=0.3341, pruned_loss=0.08223, over 28904.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3362, pruned_loss=0.08962, over 5690830.92 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3366, pruned_loss=0.09047, over 5781556.92 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3378, pruned_loss=0.09019, over 5676618.98 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:01:32,774 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 20:01:41,315 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 20:01:55,876 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745891.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:02:48,198 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 15050, giga_loss[loss=0.2363, simple_loss=0.3085, pruned_loss=0.08211, over 29000.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3311, pruned_loss=0.08769, over 5690278.89 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3367, pruned_loss=0.09058, over 5781324.33 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3322, pruned_loss=0.08799, over 5677303.35 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:03:10,358 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=745945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:03:13,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=745948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:03:50,963 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=745977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:03:52,108 INFO [train.py:968] (1/2) Epoch 17, batch 15100, giga_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1236, over 27564.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3307, pruned_loss=0.08789, over 5688087.60 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3367, pruned_loss=0.09066, over 5783136.89 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3315, pruned_loss=0.088, over 5674794.37 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:04:02,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=745986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:04:26,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3848, 1.2355, 3.8646, 3.1608], device='cuda:1'), covar=tensor([0.1601, 0.2826, 0.0455, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0626, 0.0911, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 20:04:48,066 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-08 20:04:48,123 INFO [optim.py:369] (1/2) 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 20:04:49,016 INFO [train.py:968] (1/2) Epoch 17, batch 15150, giga_loss[loss=0.2412, simple_loss=0.3286, pruned_loss=0.07686, over 28907.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3311, pruned_loss=0.08864, over 5681251.01 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3363, pruned_loss=0.09044, over 5782495.87 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3319, pruned_loss=0.08885, over 5668160.94 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:04:53,863 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=746034.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:04:58,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=746037.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:05:32,537 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=746066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:05:32,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6424, 1.8850, 1.2043, 1.4523], device='cuda:1'), covar=tensor([0.1028, 0.0605, 0.1163, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0437, 0.0506, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 20:05:37,108 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=746069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:05:47,034 INFO [train.py:968] (1/2) Epoch 17, batch 15200, libri_loss[loss=0.2645, simple_loss=0.3426, pruned_loss=0.09323, over 28700.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3296, pruned_loss=0.0875, over 5672150.16 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3362, pruned_loss=0.09041, over 5784937.12 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3301, pruned_loss=0.08763, over 5656407.11 frames. ], batch size: 106, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:06:27,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6489, 4.4954, 4.2213, 1.9721], device='cuda:1'), covar=tensor([0.0528, 0.0758, 0.0940, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.1139, 0.1048, 0.0898, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 20:06:45,014 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 15250, libri_loss[loss=0.2361, simple_loss=0.3127, pruned_loss=0.07977, over 29564.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3277, pruned_loss=0.08551, over 5677623.07 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3357, pruned_loss=0.09022, over 5785690.37 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3283, pruned_loss=0.08563, over 5659626.00 frames. ], batch size: 78, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:06:46,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=746129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:06:49,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=746132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:06:51,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-08 20:06:58,954 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.59 vs. limit=5.0
+2023-03-08 20:07:23,300 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=746161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:07:25,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8974, 1.2109, 1.2918, 1.0345], device='cuda:1'), covar=tensor([0.1639, 0.1189, 0.1998, 0.1491], device='cuda:1'), in_proj_covar=tensor([0.0439, 0.0720, 0.0680, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 20:07:48,989 INFO [train.py:968] (1/2) Epoch 17, batch 15300, giga_loss[loss=0.2306, simple_loss=0.3134, pruned_loss=0.07391, over 29065.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3259, pruned_loss=0.08457, over 5672413.16 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3348, pruned_loss=0.08979, over 5789231.71 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.327, pruned_loss=0.0849, over 5651885.54 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:08:22,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=746201.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:08:32,523 INFO [scaling.py:679] (1/2) 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] (1/2) 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] (1/2) Epoch 17, batch 15350, libri_loss[loss=0.2172, simple_loss=0.2909, pruned_loss=0.07173, over 29567.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3259, pruned_loss=0.08455, over 5684876.40 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3346, pruned_loss=0.0898, over 5792134.93 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3267, pruned_loss=0.08469, over 5663261.16 frames. ], batch size: 75, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:09:17,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4135, 1.7360, 1.6432, 1.2265], device='cuda:1'), covar=tensor([0.1836, 0.2648, 0.1581, 0.1882], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0684, 0.0911, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 20:09:19,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1092, 4.9364, 4.6395, 2.5425], device='cuda:1'), covar=tensor([0.0456, 0.0571, 0.0683, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.1134, 0.1047, 0.0897, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 20:09:55,452 INFO [train.py:968] (1/2) Epoch 17, batch 15400, giga_loss[loss=0.2828, simple_loss=0.3538, pruned_loss=0.1059, over 29047.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3269, pruned_loss=0.08443, over 5703973.38 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3339, pruned_loss=0.08938, over 5795307.59 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3278, pruned_loss=0.08464, over 5677663.21 frames. ], batch size: 285, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:10:59,531 INFO [optim.py:369] (1/2) 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 20:11:00,159 INFO [train.py:968] (1/2) Epoch 17, batch 15450, giga_loss[loss=0.3061, simple_loss=0.3646, pruned_loss=0.1238, over 28527.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3281, pruned_loss=0.08574, over 5706241.43 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3337, pruned_loss=0.08942, over 5797764.73 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3288, pruned_loss=0.08575, over 5680952.46 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:11:19,174 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=746347.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:12:01,304 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 15500, giga_loss[loss=0.2659, simple_loss=0.344, pruned_loss=0.09385, over 28818.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3284, pruned_loss=0.08615, over 5704748.83 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3338, pruned_loss=0.08942, over 5797172.06 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3287, pruned_loss=0.08607, over 5682265.49 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:13:01,577 INFO [optim.py:369] (1/2) 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 20:13:01,590 INFO [train.py:968] (1/2) Epoch 17, batch 15550, giga_loss[loss=0.2291, simple_loss=0.3323, pruned_loss=0.06301, over 28936.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3273, pruned_loss=0.08511, over 5689958.02 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3337, pruned_loss=0.08951, over 5790089.96 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3274, pruned_loss=0.08483, over 5674549.48 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:13:19,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=746444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:14:01,451 INFO [train.py:968] (1/2) Epoch 17, batch 15600, giga_loss[loss=0.2542, simple_loss=0.3387, pruned_loss=0.08489, over 27656.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.329, pruned_loss=0.08496, over 5672768.84 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3333, pruned_loss=0.08925, over 5793080.54 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3294, pruned_loss=0.08487, over 5655056.88 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:14:02,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5480, 4.1881, 1.7009, 1.6528], device='cuda:1'), covar=tensor([0.0948, 0.0251, 0.0895, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0532, 0.0366, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 20:14:09,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1482, 1.2057, 1.0784, 0.9356], device='cuda:1'), covar=tensor([0.0951, 0.0517, 0.1052, 0.1045], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0439, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 20:14:33,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-08 20:15:03,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4576, 1.5614, 1.1601, 1.2171], device='cuda:1'), covar=tensor([0.0836, 0.0394, 0.0901, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0437, 0.0505, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 20:15:04,520 INFO [optim.py:369] (1/2) 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 20:15:04,532 INFO [train.py:968] (1/2) Epoch 17, batch 15650, giga_loss[loss=0.2696, simple_loss=0.3496, pruned_loss=0.09479, over 28491.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3319, pruned_loss=0.08634, over 5670046.66 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3329, pruned_loss=0.08903, over 5792087.61 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3326, pruned_loss=0.08643, over 5654411.50 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:15:22,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5343, 1.6584, 1.8037, 1.3475], device='cuda:1'), covar=tensor([0.1898, 0.2549, 0.1517, 0.1903], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0685, 0.0912, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 20:15:30,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0355, 1.1769, 3.4247, 2.9294], device='cuda:1'), covar=tensor([0.1768, 0.2780, 0.0483, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0713, 0.0617, 0.0898, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 20:15:56,243 INFO [train.py:968] (1/2) Epoch 17, batch 15700, libri_loss[loss=0.2756, simple_loss=0.3524, pruned_loss=0.09939, over 29751.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3331, pruned_loss=0.08681, over 5665363.31 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3331, pruned_loss=0.08922, over 5788149.41 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3334, pruned_loss=0.08659, over 5650754.14 frames. ], batch size: 87, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:16:07,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=746587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:16:10,746 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=746619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:17:00,594 INFO [train.py:968] (1/2) Epoch 17, batch 15750, giga_loss[loss=0.2209, simple_loss=0.311, pruned_loss=0.06539, over 28880.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3329, pruned_loss=0.0872, over 5656510.82 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3326, pruned_loss=0.0889, over 5789196.88 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3335, pruned_loss=0.08729, over 5642916.08 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:17:01,983 INFO [optim.py:369] (1/2) 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 20:17:40,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7516, 4.5866, 4.3380, 2.0550], device='cuda:1'), covar=tensor([0.0505, 0.0635, 0.0774, 0.1939], device='cuda:1'), in_proj_covar=tensor([0.1138, 0.1047, 0.0898, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 20:17:57,446 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 15800, giga_loss[loss=0.2067, simple_loss=0.2918, pruned_loss=0.06084, over 28940.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.329, pruned_loss=0.0849, over 5662340.14 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3321, pruned_loss=0.08867, over 5791281.23 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.33, pruned_loss=0.0851, over 5645960.83 frames. ], batch size: 106, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:18:52,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3424, 1.7263, 1.7046, 1.4703], device='cuda:1'), covar=tensor([0.1780, 0.1637, 0.1925, 0.1812], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0715, 0.0674, 0.0653], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 20:18:53,255 INFO [train.py:968] (1/2) Epoch 17, batch 15850, giga_loss[loss=0.2249, simple_loss=0.3077, pruned_loss=0.07101, over 28943.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3283, pruned_loss=0.08469, over 5665777.30 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3311, pruned_loss=0.08802, over 5785078.10 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3298, pruned_loss=0.08527, over 5649831.68 frames. ], batch size: 112, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:18:54,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-08 20:18:54,662 INFO [optim.py:369] (1/2) 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 20:19:53,763 INFO [train.py:968] (1/2) Epoch 17, batch 15900, giga_loss[loss=0.2678, simple_loss=0.3457, pruned_loss=0.09495, over 29048.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3264, pruned_loss=0.08398, over 5671788.91 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3309, pruned_loss=0.08788, over 5783503.70 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3278, pruned_loss=0.08452, over 5659037.03 frames. ], batch size: 285, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:20:57,647 INFO [train.py:968] (1/2) Epoch 17, batch 15950, giga_loss[loss=0.2752, simple_loss=0.3462, pruned_loss=0.1021, over 26900.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.08505, over 5667703.36 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.331, pruned_loss=0.08792, over 5775715.67 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3302, pruned_loss=0.0854, over 5663644.52 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:21:00,222 INFO [optim.py:369] (1/2) 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 20:21:32,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-08 20:21:50,963 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 16000, giga_loss[loss=0.2644, simple_loss=0.3416, pruned_loss=0.09356, over 28124.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3303, pruned_loss=0.08599, over 5665443.04 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3307, pruned_loss=0.08765, over 5779977.33 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3314, pruned_loss=0.08645, over 5655352.79 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:22:02,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5258, 1.9756, 1.8438, 1.5437], device='cuda:1'), covar=tensor([0.2960, 0.2002, 0.2122, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1783, 0.1694, 0.1848], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 20:22:36,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.92 vs. limit=2.0
+2023-03-08 20:23:05,950 INFO [train.py:968] (1/2) Epoch 17, batch 16050, giga_loss[loss=0.2521, simple_loss=0.3379, pruned_loss=0.08311, over 28742.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3311, pruned_loss=0.08678, over 5665812.74 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3303, pruned_loss=0.0874, over 5779246.92 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3323, pruned_loss=0.08732, over 5655062.91 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:23:07,198 INFO [optim.py:369] (1/2) 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 20:24:01,588 INFO [train.py:968] (1/2) Epoch 17, batch 16100, giga_loss[loss=0.2568, simple_loss=0.3463, pruned_loss=0.08368, over 28933.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3356, pruned_loss=0.08917, over 5662736.19 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3302, pruned_loss=0.08745, over 5782205.91 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3366, pruned_loss=0.08956, over 5649144.39 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:24:14,736 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=746989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:24:54,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2369, 1.4883, 1.5981, 1.2833], device='cuda:1'), covar=tensor([0.1577, 0.1538, 0.1824, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0716, 0.0673, 0.0652], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 20:25:01,891 INFO [train.py:968] (1/2) Epoch 17, batch 16150, giga_loss[loss=0.2932, simple_loss=0.3726, pruned_loss=0.1069, over 28762.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3375, pruned_loss=0.08965, over 5661369.82 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3299, pruned_loss=0.08724, over 5783802.50 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3388, pruned_loss=0.09018, over 5647622.87 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:25:03,988 INFO [optim.py:369] (1/2) 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 20:25:34,915 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 16200, giga_loss[loss=0.2363, simple_loss=0.3232, pruned_loss=0.07467, over 28479.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3373, pruned_loss=0.08948, over 5655408.33 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3298, pruned_loss=0.08711, over 5785533.44 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3385, pruned_loss=0.09007, over 5640680.35 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:26:31,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0745, 3.9111, 3.6872, 1.7975], device='cuda:1'), covar=tensor([0.0623, 0.0744, 0.0812, 0.2076], device='cuda:1'), in_proj_covar=tensor([0.1134, 0.1043, 0.0898, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 20:26:47,200 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 16250, libri_loss[loss=0.2783, simple_loss=0.3653, pruned_loss=0.09563, over 29069.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3345, pruned_loss=0.08804, over 5656764.87 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.33, pruned_loss=0.08718, over 5777671.27 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3354, pruned_loss=0.08845, over 5649761.34 frames. ], batch size: 101, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:27:20,644 INFO [optim.py:369] (1/2) 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 20:28:12,504 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:28:24,033 INFO [train.py:968] (1/2) Epoch 17, batch 16300, giga_loss[loss=0.273, simple_loss=0.3532, pruned_loss=0.09641, over 28728.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3341, pruned_loss=0.08777, over 5667441.09 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3299, pruned_loss=0.08717, over 5780121.86 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3349, pruned_loss=0.08812, over 5657443.30 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:28:45,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:28:47,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:29:25,951 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:29:26,344 INFO [train.py:968] (1/2) Epoch 17, batch 16350, giga_loss[loss=0.2401, simple_loss=0.3233, pruned_loss=0.07846, over 28894.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3338, pruned_loss=0.08842, over 5672020.85 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3296, pruned_loss=0.08699, over 5784418.49 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3349, pruned_loss=0.08889, over 5656955.35 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:29:28,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4684, 1.7815, 1.4467, 1.5585], device='cuda:1'), covar=tensor([0.0734, 0.0305, 0.0327, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0114, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 20:29:29,092 INFO [optim.py:369] (1/2) 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 20:29:46,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:30:25,699 INFO [train.py:968] (1/2) Epoch 17, batch 16400, giga_loss[loss=0.2308, simple_loss=0.3136, pruned_loss=0.074, over 28936.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.331, pruned_loss=0.08772, over 5666186.89 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3294, pruned_loss=0.08696, over 5786579.16 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3322, pruned_loss=0.08815, over 5648551.20 frames. ], batch size: 227, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:31:25,486 INFO [train.py:968] (1/2) Epoch 17, batch 16450, giga_loss[loss=0.2248, simple_loss=0.292, pruned_loss=0.07886, over 24421.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3307, pruned_loss=0.08767, over 5663173.57 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3295, pruned_loss=0.08713, over 5787094.96 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3315, pruned_loss=0.08788, over 5645064.62 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:31:25,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747329.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:31:28,717 INFO [optim.py:369] (1/2) 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 20:32:02,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2952, 1.1382, 3.9521, 3.1982], device='cuda:1'), covar=tensor([0.1647, 0.2837, 0.0423, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0716, 0.0619, 0.0903, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 20:32:09,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:32:13,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-08 20:32:26,999 INFO [train.py:968] (1/2) Epoch 17, batch 16500, giga_loss[loss=0.2358, simple_loss=0.3248, pruned_loss=0.07341, over 28760.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3289, pruned_loss=0.08518, over 5680285.84 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3293, pruned_loss=0.08704, over 5789186.14 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3298, pruned_loss=0.08542, over 5662455.25 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:32:37,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:32:41,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:33:15,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:33:25,807 INFO [train.py:968] (1/2) Epoch 17, batch 16550, giga_loss[loss=0.267, simple_loss=0.3659, pruned_loss=0.08408, over 28746.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3295, pruned_loss=0.08329, over 5680823.89 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3296, pruned_loss=0.08716, over 5787232.18 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3299, pruned_loss=0.0833, over 5666026.93 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:33:27,974 INFO [optim.py:369] (1/2) 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 20:34:20,943 INFO [train.py:968] (1/2) Epoch 17, batch 16600, giga_loss[loss=0.2352, simple_loss=0.3282, pruned_loss=0.07107, over 28980.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3311, pruned_loss=0.08275, over 5692867.28 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3292, pruned_loss=0.08696, over 5791131.36 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3317, pruned_loss=0.08281, over 5674599.85 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:34:23,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:34:54,426 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:34:59,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:35:23,084 INFO [train.py:968] (1/2) Epoch 17, batch 16650, giga_loss[loss=0.2215, simple_loss=0.2935, pruned_loss=0.07475, over 24375.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.332, pruned_loss=0.08342, over 5677245.81 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3292, pruned_loss=0.08694, over 5791220.38 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3325, pruned_loss=0.08345, over 5661980.62 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:35:27,903 INFO [optim.py:369] (1/2) 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 20:35:37,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:35:43,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:36:31,628 INFO [train.py:968] (1/2) Epoch 17, batch 16700, giga_loss[loss=0.2359, simple_loss=0.3262, pruned_loss=0.07276, over 28971.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3335, pruned_loss=0.08506, over 5669502.90 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3295, pruned_loss=0.08708, over 5790065.08 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3337, pruned_loss=0.0849, over 5656401.84 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:36:31,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0458, 3.2541, 2.0476, 0.9886], device='cuda:1'), covar=tensor([0.6040, 0.2695, 0.3685, 0.6177], device='cuda:1'), in_proj_covar=tensor([0.1663, 0.1575, 0.1546, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 20:36:40,647 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747587.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:37:28,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:37:33,169 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 16750, giga_loss[loss=0.2396, simple_loss=0.3262, pruned_loss=0.07652, over 29046.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3324, pruned_loss=0.08475, over 5664798.31 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3293, pruned_loss=0.08699, over 5793628.01 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3329, pruned_loss=0.08466, over 5648002.71 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:37:40,752 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:38:46,660 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=747678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:38:47,037 INFO [train.py:968] (1/2) Epoch 17, batch 16800, giga_loss[loss=0.2498, simple_loss=0.3176, pruned_loss=0.09097, over 26961.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3333, pruned_loss=0.08441, over 5659023.73 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3294, pruned_loss=0.0871, over 5783120.75 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3336, pruned_loss=0.08418, over 5653204.06 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:39:01,839 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-08 20:39:02,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:39:06,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:39:21,007 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-08 20:39:23,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:39:44,283 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:39:57,955 INFO [train.py:968] (1/2) Epoch 17, batch 16850, giga_loss[loss=0.2578, simple_loss=0.3184, pruned_loss=0.0986, over 24316.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3344, pruned_loss=0.08513, over 5656096.46 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3295, pruned_loss=0.08716, over 5786618.03 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3347, pruned_loss=0.08483, over 5644895.11 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:40:04,260 INFO [optim.py:369] (1/2) 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 20:40:58,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 20:41:11,722 INFO [train.py:968] (1/2) Epoch 17, batch 16900, giga_loss[loss=0.2867, simple_loss=0.376, pruned_loss=0.09869, over 28755.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3381, pruned_loss=0.08637, over 5664156.54 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3296, pruned_loss=0.08719, over 5787431.77 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3383, pruned_loss=0.08608, over 5653529.17 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:42:16,295 INFO [train.py:968] (1/2) Epoch 17, batch 16950, giga_loss[loss=0.2289, simple_loss=0.3179, pruned_loss=0.06988, over 28177.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3368, pruned_loss=0.08575, over 5671910.16 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3294, pruned_loss=0.08696, over 5782062.81 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3375, pruned_loss=0.08569, over 5664863.52 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:42:19,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3230, 1.1777, 3.9134, 3.2188], device='cuda:1'), covar=tensor([0.1596, 0.2793, 0.0448, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0715, 0.0618, 0.0900, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 20:42:21,459 INFO [optim.py:369] (1/2) 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 20:42:43,927 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=747847.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:42:48,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=747850.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:43:29,495 INFO [train.py:968] (1/2) Epoch 17, batch 17000, giga_loss[loss=0.2154, simple_loss=0.3041, pruned_loss=0.06329, over 29081.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3356, pruned_loss=0.08603, over 5671491.34 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3293, pruned_loss=0.08682, over 5783372.09 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3363, pruned_loss=0.08608, over 5663433.72 frames. ], batch size: 120, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:43:30,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=747879.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:43:45,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3186, 1.9439, 1.5001, 1.6493], device='cuda:1'), covar=tensor([0.0753, 0.0283, 0.0323, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0114, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 20:44:36,710 INFO [train.py:968] (1/2) Epoch 17, batch 17050, giga_loss[loss=0.2423, simple_loss=0.3335, pruned_loss=0.0755, over 28770.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3334, pruned_loss=0.08427, over 5673961.14 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3292, pruned_loss=0.08677, over 5778558.38 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3342, pruned_loss=0.08428, over 5668319.61 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:44:43,931 INFO [optim.py:369] (1/2) 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 20:45:23,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=747962.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:45:37,921 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-08 20:45:43,554 INFO [train.py:968] (1/2) Epoch 17, batch 17100, giga_loss[loss=0.24, simple_loss=0.3292, pruned_loss=0.07543, over 28623.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3323, pruned_loss=0.0836, over 5664635.21 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3292, pruned_loss=0.0868, over 5770920.33 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.333, pruned_loss=0.08352, over 5664727.35 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:45:53,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4329, 1.6854, 1.3782, 1.3305], device='cuda:1'), covar=tensor([0.2636, 0.2458, 0.2879, 0.2215], device='cuda:1'), in_proj_covar=tensor([0.1412, 0.1023, 0.1261, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 20:46:33,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8809, 1.1470, 1.1201, 0.8345], device='cuda:1'), covar=tensor([0.2232, 0.2091, 0.1276, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1765, 0.1674, 0.1832], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 20:46:43,476 INFO [train.py:968] (1/2) Epoch 17, batch 17150, giga_loss[loss=0.2511, simple_loss=0.332, pruned_loss=0.08508, over 28971.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3335, pruned_loss=0.08469, over 5672950.13 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08677, over 5774224.21 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3342, pruned_loss=0.08458, over 5667216.34 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:46:49,094 INFO [optim.py:369] (1/2) 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 20:47:12,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:47:37,802 INFO [train.py:968] (1/2) Epoch 17, batch 17200, giga_loss[loss=0.276, simple_loss=0.3563, pruned_loss=0.09784, over 28094.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3366, pruned_loss=0.08639, over 5676090.67 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.329, pruned_loss=0.08676, over 5778281.97 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3375, pruned_loss=0.0863, over 5664402.47 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:48:11,241 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748105.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:48:14,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748108.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 20:48:33,591 INFO [train.py:968] (1/2) Epoch 17, batch 17250, giga_loss[loss=0.2524, simple_loss=0.3195, pruned_loss=0.09261, over 26839.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3357, pruned_loss=0.08657, over 5678868.42 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08665, over 5781585.81 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3367, pruned_loss=0.08661, over 5663415.18 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:48:38,887 INFO [optim.py:369] (1/2) 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 20:48:44,000 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748137.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 20:49:31,068 INFO [train.py:968] (1/2) Epoch 17, batch 17300, libri_loss[loss=0.2613, simple_loss=0.3359, pruned_loss=0.0933, over 29758.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3336, pruned_loss=0.08691, over 5676972.99 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3287, pruned_loss=0.08657, over 5785596.59 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3348, pruned_loss=0.087, over 5658063.93 frames. ], batch size: 87, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:49:52,669 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:49:57,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:50:04,613 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:50:27,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:50:28,052 INFO [train.py:968] (1/2) Epoch 17, batch 17350, giga_loss[loss=0.2405, simple_loss=0.3237, pruned_loss=0.07864, over 28876.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3336, pruned_loss=0.08751, over 5668768.96 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08663, over 5785060.97 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3345, pruned_loss=0.08755, over 5651675.58 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:50:32,480 INFO [optim.py:369] (1/2) 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 20:50:34,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6638, 1.7929, 1.9418, 1.4354], device='cuda:1'), covar=tensor([0.1793, 0.2438, 0.1405, 0.1746], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0681, 0.0909, 0.0810], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 20:51:16,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4792, 2.2134, 1.6614, 0.5453], device='cuda:1'), covar=tensor([0.4879, 0.2568, 0.3471, 0.5918], device='cuda:1'), in_proj_covar=tensor([0.1677, 0.1592, 0.1561, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 20:51:19,940 INFO [train.py:968] (1/2) Epoch 17, batch 17400, giga_loss[loss=0.2754, simple_loss=0.3616, pruned_loss=0.09461, over 28527.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3385, pruned_loss=0.09081, over 5659550.00 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08675, over 5777379.62 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3393, pruned_loss=0.09084, over 5647755.32 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:52:04,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1073, 1.2690, 1.1257, 1.0359], device='cuda:1'), covar=tensor([0.1767, 0.1849, 0.1478, 0.1704], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1766, 0.1677, 0.1841], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 20:52:12,272 INFO [train.py:968] (1/2) Epoch 17, batch 17450, giga_loss[loss=0.3092, simple_loss=0.3917, pruned_loss=0.1134, over 29109.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3475, pruned_loss=0.09587, over 5668704.29 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08661, over 5777113.47 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3487, pruned_loss=0.0962, over 5657100.78 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:52:17,522 INFO [optim.py:369] (1/2) 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 20:52:55,803 INFO [train.py:968] (1/2) Epoch 17, batch 17500, giga_loss[loss=0.274, simple_loss=0.3634, pruned_loss=0.09237, over 29092.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3519, pruned_loss=0.09821, over 5678148.81 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3288, pruned_loss=0.08649, over 5778617.14 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3535, pruned_loss=0.09886, over 5664986.43 frames. ], batch size: 155, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:52:59,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:53:42,675 INFO [train.py:968] (1/2) Epoch 17, batch 17550, giga_loss[loss=0.2309, simple_loss=0.3118, pruned_loss=0.07499, over 28812.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3496, pruned_loss=0.09774, over 5680087.56 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3291, pruned_loss=0.08669, over 5781627.89 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3511, pruned_loss=0.09834, over 5664691.60 frames. ], batch size: 243, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 20:53:47,441 INFO [optim.py:369] (1/2) 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 20:54:23,089 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:54:27,284 INFO [train.py:968] (1/2) Epoch 17, batch 17600, giga_loss[loss=0.2652, simple_loss=0.3276, pruned_loss=0.1014, over 26606.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3425, pruned_loss=0.09503, over 5676873.75 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08672, over 5773800.54 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3439, pruned_loss=0.0956, over 5670169.38 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:55:14,351 INFO [train.py:968] (1/2) Epoch 17, batch 17650, giga_loss[loss=0.2584, simple_loss=0.33, pruned_loss=0.09339, over 28866.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.336, pruned_loss=0.09216, over 5684619.30 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3295, pruned_loss=0.08683, over 5773464.30 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.337, pruned_loss=0.09268, over 5677535.64 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:55:17,835 INFO [optim.py:369] (1/2) 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 20:55:57,332 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 17700, giga_loss[loss=0.2453, simple_loss=0.3076, pruned_loss=0.09148, over 26609.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3284, pruned_loss=0.0891, over 5689007.98 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3294, pruned_loss=0.08679, over 5776209.39 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3293, pruned_loss=0.08962, over 5679061.84 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:56:05,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:56:16,820 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=748598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:56:41,133 INFO [train.py:968] (1/2) Epoch 17, batch 17750, giga_loss[loss=0.2229, simple_loss=0.2965, pruned_loss=0.07464, over 28915.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3212, pruned_loss=0.08569, over 5696954.66 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3296, pruned_loss=0.08671, over 5778475.03 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3216, pruned_loss=0.08616, over 5684644.44 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:56:47,557 INFO [optim.py:369] (1/2) 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 20:56:55,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5964, 1.9630, 1.5246, 1.6002], device='cuda:1'), covar=tensor([0.2546, 0.2520, 0.2945, 0.2389], device='cuda:1'), in_proj_covar=tensor([0.1419, 0.1030, 0.1262, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 20:56:56,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 20:57:22,955 INFO [train.py:968] (1/2) Epoch 17, batch 17800, giga_loss[loss=0.2473, simple_loss=0.307, pruned_loss=0.09379, over 26581.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3164, pruned_loss=0.08348, over 5698674.72 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3295, pruned_loss=0.08652, over 5781018.59 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3165, pruned_loss=0.08398, over 5684663.09 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:58:02,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:03,923 INFO [train.py:968] (1/2) Epoch 17, batch 17850, giga_loss[loss=0.2286, simple_loss=0.3052, pruned_loss=0.07603, over 28201.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3132, pruned_loss=0.08177, over 5705743.10 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3294, pruned_loss=0.08635, over 5784513.91 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3129, pruned_loss=0.08223, over 5689180.63 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 20:58:04,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:07,173 INFO [optim.py:369] (1/2) 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 20:58:28,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1515, 4.9276, 4.6852, 2.4024], device='cuda:1'), covar=tensor([0.0388, 0.0631, 0.0662, 0.1798], device='cuda:1'), in_proj_covar=tensor([0.1136, 0.1050, 0.0900, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 20:58:32,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:32,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 20:58:49,135 INFO [train.py:968] (1/2) Epoch 17, batch 17900, giga_loss[loss=0.2322, simple_loss=0.2964, pruned_loss=0.08397, over 28447.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3106, pruned_loss=0.08071, over 5705306.00 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3297, pruned_loss=0.08634, over 5786332.19 frames. ], giga_tot_loss[loss=0.2358, simple_loss=0.3098, pruned_loss=0.08095, over 5688945.34 frames. ], batch size: 60, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:59:29,127 INFO [train.py:968] (1/2) Epoch 17, batch 17950, giga_loss[loss=0.2519, simple_loss=0.3183, pruned_loss=0.09272, over 28746.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3094, pruned_loss=0.08053, over 5683913.74 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3302, pruned_loss=0.08661, over 5768901.35 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3078, pruned_loss=0.08034, over 5683897.05 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 20:59:35,488 INFO [optim.py:369] (1/2) 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 20:59:46,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-08 20:59:47,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:00:13,239 INFO [train.py:968] (1/2) Epoch 17, batch 18000, giga_loss[loss=0.2242, simple_loss=0.2996, pruned_loss=0.07443, over 28875.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3084, pruned_loss=0.08005, over 5690461.42 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3315, pruned_loss=0.08724, over 5764527.06 frames. ], giga_tot_loss[loss=0.2316, simple_loss=0.305, pruned_loss=0.07908, over 5691164.41 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:00:13,240 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 21:00:17,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2728, 1.8232, 1.3546, 0.4142], device='cuda:1'), covar=tensor([0.4704, 0.3590, 0.4683, 0.6118], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1589, 0.1554, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 21:00:21,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2461, 1.7899, 1.2944, 0.3953], device='cuda:1'), covar=tensor([0.4768, 0.3691, 0.4431, 0.5814], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1589, 0.1554, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 21:00:21,828 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 21:00:41,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:00:44,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2709, 0.7769, 0.9434, 1.4208], device='cuda:1'), covar=tensor([0.0798, 0.0389, 0.0360, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0060, 0.0101], device='cuda:1')
+2023-03-08 21:00:44,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:06,765 INFO [train.py:968] (1/2) Epoch 17, batch 18050, giga_loss[loss=0.2061, simple_loss=0.2795, pruned_loss=0.06638, over 28876.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.3052, pruned_loss=0.07896, over 5681939.02 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3317, pruned_loss=0.08725, over 5764188.28 frames. ], giga_tot_loss[loss=0.2292, simple_loss=0.3021, pruned_loss=0.0781, over 5682212.28 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:01:10,685 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=748933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:12,598 INFO [optim.py:369] (1/2) 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 21:01:26,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:45,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=748973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:01:49,366 INFO [train.py:968] (1/2) Epoch 17, batch 18100, giga_loss[loss=0.1927, simple_loss=0.267, pruned_loss=0.05917, over 28769.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3018, pruned_loss=0.07709, over 5691448.88 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3319, pruned_loss=0.08731, over 5768232.17 frames. ], giga_tot_loss[loss=0.2252, simple_loss=0.2983, pruned_loss=0.07602, over 5685641.35 frames. ], batch size: 119, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:02:02,908 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=748992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:02:04,842 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=748995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:02:14,791 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2555, 1.0334, 4.2382, 3.2854], device='cuda:1'), covar=tensor([0.1748, 0.3064, 0.0412, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0716, 0.0617, 0.0907, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 21:02:31,334 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:02:34,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2162, 1.5076, 0.9734, 1.1538], device='cuda:1'), covar=tensor([0.1131, 0.0681, 0.1553, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0436, 0.0507, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 21:02:35,928 INFO [train.py:968] (1/2) Epoch 17, batch 18150, giga_loss[loss=0.1971, simple_loss=0.2746, pruned_loss=0.05974, over 28787.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.2999, pruned_loss=0.07619, over 5703006.14 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3322, pruned_loss=0.08734, over 5771342.67 frames. ], giga_tot_loss[loss=0.2231, simple_loss=0.2961, pruned_loss=0.07508, over 5693880.69 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:02:44,578 INFO [optim.py:369] (1/2) 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 21:03:01,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 21:03:03,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3994, 3.3277, 1.4964, 1.5790], device='cuda:1'), covar=tensor([0.0977, 0.0323, 0.0898, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0528, 0.0365, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 21:03:06,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-08 21:03:22,798 INFO [train.py:968] (1/2) Epoch 17, batch 18200, giga_loss[loss=0.2798, simple_loss=0.3449, pruned_loss=0.1073, over 28587.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.2974, pruned_loss=0.07552, over 5702289.23 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3323, pruned_loss=0.08737, over 5772890.37 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.2937, pruned_loss=0.07444, over 5692718.01 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:03:36,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:03:39,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:04:03,533 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:04:05,474 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:04:12,100 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 18250, giga_loss[loss=0.3082, simple_loss=0.3754, pruned_loss=0.1205, over 27903.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3055, pruned_loss=0.07987, over 5704403.69 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3324, pruned_loss=0.08735, over 5775836.48 frames. ], giga_tot_loss[loss=0.2298, simple_loss=0.3019, pruned_loss=0.07885, over 5692822.30 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:04:21,823 INFO [optim.py:369] (1/2) 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 21:04:32,429 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 18300, giga_loss[loss=0.3046, simple_loss=0.3816, pruned_loss=0.1138, over 28822.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3181, pruned_loss=0.08577, over 5708803.04 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3326, pruned_loss=0.08731, over 5779508.54 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3146, pruned_loss=0.08493, over 5694015.49 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:05:10,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:05:41,831 INFO [train.py:968] (1/2) Epoch 17, batch 18350, giga_loss[loss=0.2931, simple_loss=0.3683, pruned_loss=0.109, over 28771.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3301, pruned_loss=0.09225, over 5709818.79 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3327, pruned_loss=0.0874, over 5782048.98 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.327, pruned_loss=0.09153, over 5694637.13 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:05:47,888 INFO [optim.py:369] (1/2) 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 21:06:23,653 INFO [train.py:968] (1/2) Epoch 17, batch 18400, giga_loss[loss=0.2868, simple_loss=0.3654, pruned_loss=0.1041, over 27874.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3369, pruned_loss=0.09484, over 5701944.61 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3329, pruned_loss=0.08734, over 5785078.69 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3343, pruned_loss=0.09448, over 5685178.08 frames. ], batch size: 412, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:07:07,071 INFO [train.py:968] (1/2) Epoch 17, batch 18450, giga_loss[loss=0.2539, simple_loss=0.341, pruned_loss=0.08336, over 28925.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3398, pruned_loss=0.09486, over 5700578.15 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3327, pruned_loss=0.08726, over 5785784.30 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3379, pruned_loss=0.0947, over 5686371.84 frames. ], batch size: 213, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:07:13,424 INFO [optim.py:369] (1/2) 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 21:07:50,933 INFO [train.py:968] (1/2) Epoch 17, batch 18500, giga_loss[loss=0.2658, simple_loss=0.3387, pruned_loss=0.09651, over 28413.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3421, pruned_loss=0.0951, over 5696325.30 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.0879, over 5785189.88 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3398, pruned_loss=0.09467, over 5682447.04 frames. ], batch size: 65, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:08:30,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-08 21:08:38,494 INFO [train.py:968] (1/2) Epoch 17, batch 18550, giga_loss[loss=0.2682, simple_loss=0.3457, pruned_loss=0.0954, over 28871.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3445, pruned_loss=0.09659, over 5698711.15 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3337, pruned_loss=0.08784, over 5786953.99 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.343, pruned_loss=0.09647, over 5684295.13 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:08:43,606 INFO [optim.py:369] (1/2) 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 21:08:51,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1106, 2.2364, 2.3209, 1.8495], device='cuda:1'), covar=tensor([0.1708, 0.2179, 0.1371, 0.1591], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0689, 0.0919, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 21:08:54,050 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 21:09:09,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7006, 1.7272, 1.3558, 1.3142], device='cuda:1'), covar=tensor([0.0936, 0.0672, 0.1041, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0437, 0.0509, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 21:09:22,512 INFO [train.py:968] (1/2) Epoch 17, batch 18600, giga_loss[loss=0.2785, simple_loss=0.349, pruned_loss=0.104, over 28678.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3465, pruned_loss=0.09806, over 5702632.18 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.334, pruned_loss=0.08782, over 5791111.02 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3455, pruned_loss=0.09831, over 5684230.09 frames. ], batch size: 78, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:10:08,498 INFO [train.py:968] (1/2) Epoch 17, batch 18650, giga_loss[loss=0.2723, simple_loss=0.3511, pruned_loss=0.09678, over 28808.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3498, pruned_loss=0.1001, over 5696771.24 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3343, pruned_loss=0.08784, over 5780718.51 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.349, pruned_loss=0.1005, over 5690073.28 frames. ], batch size: 99, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:10:15,064 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:10:27,253 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:10:39,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=749564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:10:42,798 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 18700, giga_loss[loss=0.2642, simple_loss=0.3498, pruned_loss=0.08929, over 28247.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3539, pruned_loss=0.1022, over 5694645.77 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3351, pruned_loss=0.08834, over 5773097.73 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3529, pruned_loss=0.1023, over 5694398.98 frames. ], batch size: 77, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:11:22,663 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-08 21:11:35,760 INFO [train.py:968] (1/2) Epoch 17, batch 18750, giga_loss[loss=0.2774, simple_loss=0.3638, pruned_loss=0.09547, over 28984.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3554, pruned_loss=0.1018, over 5700269.71 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3353, pruned_loss=0.08837, over 5774569.79 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3545, pruned_loss=0.102, over 5698069.43 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:11:42,948 INFO [optim.py:369] (1/2) 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 21:11:52,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2961, 1.5378, 1.2612, 1.4487], device='cuda:1'), covar=tensor([0.0782, 0.0351, 0.0345, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 21:12:00,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9630, 1.0529, 3.2567, 2.8167], device='cuda:1'), covar=tensor([0.1729, 0.2795, 0.0480, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0712, 0.0612, 0.0899, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 21:12:16,012 INFO [train.py:968] (1/2) Epoch 17, batch 18800, libri_loss[loss=0.2589, simple_loss=0.3505, pruned_loss=0.08362, over 29560.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3566, pruned_loss=0.1019, over 5706534.72 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3355, pruned_loss=0.08833, over 5774981.85 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3564, pruned_loss=0.1026, over 5701727.32 frames. ], batch size: 89, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:12:40,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:42,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:43,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=749711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:43,682 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=749711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:45,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=749714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:12:57,644 INFO [train.py:968] (1/2) Epoch 17, batch 18850, giga_loss[loss=0.2714, simple_loss=0.3313, pruned_loss=0.1058, over 23600.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3563, pruned_loss=0.1011, over 5698615.25 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3359, pruned_loss=0.0885, over 5775922.01 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3563, pruned_loss=0.1018, over 5691910.02 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:13:04,675 INFO [optim.py:369] (1/2) 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 21:13:05,671 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:13:09,092 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=749743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:13:40,807 INFO [train.py:968] (1/2) Epoch 17, batch 18900, giga_loss[loss=0.2507, simple_loss=0.3399, pruned_loss=0.08072, over 28995.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3539, pruned_loss=0.09855, over 5704311.54 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3357, pruned_loss=0.0884, over 5778115.42 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3544, pruned_loss=0.09935, over 5696171.57 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:14:17,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5894, 4.3330, 4.1638, 1.9650], device='cuda:1'), covar=tensor([0.0573, 0.0798, 0.0790, 0.2095], device='cuda:1'), in_proj_covar=tensor([0.1120, 0.1040, 0.0890, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 21:14:20,969 INFO [train.py:968] (1/2) Epoch 17, batch 18950, libri_loss[loss=0.2523, simple_loss=0.3436, pruned_loss=0.08051, over 29539.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3522, pruned_loss=0.09709, over 5711146.46 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3359, pruned_loss=0.08833, over 5781831.50 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3528, pruned_loss=0.09806, over 5699402.55 frames. ], batch size: 84, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:14:27,746 INFO [optim.py:369] (1/2) 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 21:14:41,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3327, 0.8421, 0.9756, 1.5016], device='cuda:1'), covar=tensor([0.0756, 0.0363, 0.0330, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 21:14:41,959 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 21:15:05,034 INFO [train.py:968] (1/2) Epoch 17, batch 19000, libri_loss[loss=0.278, simple_loss=0.3632, pruned_loss=0.0964, over 29526.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3543, pruned_loss=0.1001, over 5700564.30 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3361, pruned_loss=0.08841, over 5783089.96 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3548, pruned_loss=0.1009, over 5689515.77 frames. ], batch size: 89, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:15:33,234 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=749910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:15:34,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-08 21:15:46,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=749925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:15:50,229 INFO [train.py:968] (1/2) Epoch 17, batch 19050, giga_loss[loss=0.3661, simple_loss=0.4072, pruned_loss=0.1624, over 26594.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3567, pruned_loss=0.1045, over 5688290.83 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.336, pruned_loss=0.08831, over 5783372.29 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3578, pruned_loss=0.1056, over 5676532.55 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:15:58,287 INFO [optim.py:369] (1/2) 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 21:16:01,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5411, 4.3225, 4.1394, 2.0666], device='cuda:1'), covar=tensor([0.0635, 0.0840, 0.0772, 0.1968], device='cuda:1'), in_proj_covar=tensor([0.1124, 0.1046, 0.0895, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 21:16:29,243 INFO [train.py:968] (1/2) Epoch 17, batch 19100, libri_loss[loss=0.3079, simple_loss=0.3696, pruned_loss=0.1231, over 18995.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3564, pruned_loss=0.1053, over 5681720.01 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3361, pruned_loss=0.08844, over 5767019.09 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3577, pruned_loss=0.1065, over 5684766.61 frames. ], batch size: 187, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:16:48,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750002.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:17:10,346 INFO [train.py:968] (1/2) Epoch 17, batch 19150, giga_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1117, over 28920.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3543, pruned_loss=0.1046, over 5695469.03 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3364, pruned_loss=0.08858, over 5771415.26 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3558, pruned_loss=0.106, over 5691277.60 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:17:19,581 INFO [optim.py:369] (1/2) 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 21:17:34,065 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:37,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:47,056 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:49,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:17:55,548 INFO [train.py:968] (1/2) Epoch 17, batch 19200, giga_loss[loss=0.3112, simple_loss=0.3694, pruned_loss=0.1265, over 23962.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3542, pruned_loss=0.1048, over 5693885.93 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3366, pruned_loss=0.08864, over 5771016.85 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3555, pruned_loss=0.1063, over 5689446.06 frames. ], batch size: 705, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:18:01,143 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:03,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:07,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:12,639 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:18:16,252 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 19250, giga_loss[loss=0.2781, simple_loss=0.3493, pruned_loss=0.1035, over 28834.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3531, pruned_loss=0.1036, over 5688530.42 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3369, pruned_loss=0.08873, over 5773575.22 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3542, pruned_loss=0.1049, over 5681158.47 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:18:41,313 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6386, 2.7382, 2.4144, 1.9859], device='cuda:1'), covar=tensor([0.1959, 0.1766, 0.1922, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.1873, 0.1792, 0.1709, 0.1874], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 21:18:49,161 INFO [optim.py:369] (1/2) 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 21:19:26,182 INFO [train.py:968] (1/2) Epoch 17, batch 19300, libri_loss[loss=0.2556, simple_loss=0.3341, pruned_loss=0.08854, over 29635.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1016, over 5696972.34 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.337, pruned_loss=0.08882, over 5776290.75 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3519, pruned_loss=0.1028, over 5687013.61 frames. ], batch size: 73, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:19:28,984 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-08 21:20:03,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3755, 3.2632, 1.5209, 1.5238], device='cuda:1'), covar=tensor([0.1027, 0.0288, 0.0930, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0530, 0.0366, 0.0410], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 21:20:09,607 INFO [train.py:968] (1/2) Epoch 17, batch 19350, giga_loss[loss=0.2674, simple_loss=0.3364, pruned_loss=0.0992, over 27692.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3477, pruned_loss=0.09956, over 5691401.50 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3373, pruned_loss=0.08901, over 5778948.84 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3486, pruned_loss=0.1008, over 5678237.89 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:20:09,994 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:20:13,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:20:18,567 INFO [optim.py:369] (1/2) 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 21:20:28,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9055, 1.2564, 2.8707, 2.6845], device='cuda:1'), covar=tensor([0.1572, 0.2345, 0.0560, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0716, 0.0616, 0.0905, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 21:20:40,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:20:58,195 INFO [train.py:968] (1/2) Epoch 17, batch 19400, giga_loss[loss=0.2276, simple_loss=0.3052, pruned_loss=0.07498, over 28313.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3421, pruned_loss=0.09669, over 5690574.53 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3376, pruned_loss=0.08917, over 5781197.85 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3428, pruned_loss=0.09766, over 5676447.99 frames. ], batch size: 368, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:21:44,076 INFO [train.py:968] (1/2) Epoch 17, batch 19450, giga_loss[loss=0.2344, simple_loss=0.311, pruned_loss=0.07896, over 28678.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3367, pruned_loss=0.09379, over 5688864.33 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3379, pruned_loss=0.08928, over 5782814.55 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3369, pruned_loss=0.09456, over 5674580.52 frames. ], batch size: 242, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:21:52,604 INFO [optim.py:369] (1/2) 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 21:22:26,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:22:28,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750377.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:22:30,863 INFO [train.py:968] (1/2) Epoch 17, batch 19500, giga_loss[loss=0.2789, simple_loss=0.3558, pruned_loss=0.101, over 28588.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3346, pruned_loss=0.09226, over 5691897.87 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3384, pruned_loss=0.08951, over 5777224.35 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3343, pruned_loss=0.09277, over 5683213.76 frames. ], batch size: 336, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:22:32,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 21:23:12,354 INFO [train.py:968] (1/2) Epoch 17, batch 19550, giga_loss[loss=0.2428, simple_loss=0.3301, pruned_loss=0.07774, over 28897.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3357, pruned_loss=0.09213, over 5681116.03 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3387, pruned_loss=0.0896, over 5760892.18 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3351, pruned_loss=0.0925, over 5687711.25 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:23:23,357 INFO [optim.py:369] (1/2) 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 21:23:30,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0041, 3.8247, 3.6886, 1.5676], device='cuda:1'), covar=tensor([0.0775, 0.0923, 0.0936, 0.2204], device='cuda:1'), in_proj_covar=tensor([0.1136, 0.1053, 0.0901, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 21:23:31,427 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-08 21:23:47,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:23:51,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:23:54,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9507, 2.0669, 1.7814, 1.7123], device='cuda:1'), covar=tensor([0.1990, 0.2511, 0.2370, 0.2485], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0736, 0.0690, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 21:23:58,305 INFO [train.py:968] (1/2) Epoch 17, batch 19600, giga_loss[loss=0.231, simple_loss=0.311, pruned_loss=0.07545, over 28866.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3347, pruned_loss=0.09121, over 5690921.30 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3389, pruned_loss=0.0895, over 5764073.83 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.334, pruned_loss=0.09163, over 5692018.68 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:24:35,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750520.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:24:37,570 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:24:41,845 INFO [train.py:968] (1/2) Epoch 17, batch 19650, giga_loss[loss=0.2343, simple_loss=0.3145, pruned_loss=0.07699, over 28916.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3331, pruned_loss=0.09088, over 5703596.38 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3389, pruned_loss=0.0894, over 5765250.87 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3324, pruned_loss=0.09132, over 5702502.98 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:24:49,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5283, 1.7790, 1.4178, 1.6515], device='cuda:1'), covar=tensor([0.2516, 0.2604, 0.2892, 0.2391], device='cuda:1'), in_proj_covar=tensor([0.1418, 0.1031, 0.1262, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 21:24:50,071 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750552.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:25:19,652 INFO [train.py:968] (1/2) Epoch 17, batch 19700, giga_loss[loss=0.2605, simple_loss=0.328, pruned_loss=0.09653, over 29044.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3305, pruned_loss=0.08953, over 5712135.23 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3395, pruned_loss=0.08946, over 5765504.29 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3293, pruned_loss=0.08983, over 5709777.57 frames. ], batch size: 128, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:25:43,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1335, 1.3431, 1.1801, 0.9657], device='cuda:1'), covar=tensor([0.3018, 0.2464, 0.1682, 0.2616], device='cuda:1'), in_proj_covar=tensor([0.1865, 0.1776, 0.1703, 0.1864], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 21:25:44,556 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:25:47,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:25:48,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:25:50,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:01,328 INFO [train.py:968] (1/2) Epoch 17, batch 19750, giga_loss[loss=0.2384, simple_loss=0.3147, pruned_loss=0.08108, over 28857.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3281, pruned_loss=0.08844, over 5716329.86 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3397, pruned_loss=0.08935, over 5766035.97 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3266, pruned_loss=0.08878, over 5713016.02 frames. ], batch size: 119, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:26:09,183 INFO [optim.py:369] (1/2) 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 21:26:10,166 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:12,560 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750644.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:26:13,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:44,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=750678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:26:44,688 INFO [train.py:968] (1/2) Epoch 17, batch 19800, giga_loss[loss=0.2621, simple_loss=0.3259, pruned_loss=0.0991, over 28447.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3267, pruned_loss=0.08824, over 5719808.16 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.34, pruned_loss=0.08953, over 5768181.50 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3251, pruned_loss=0.08833, over 5714440.18 frames. ], batch size: 71, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:26:59,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1742, 1.1486, 3.7729, 3.2405], device='cuda:1'), covar=tensor([0.1735, 0.2860, 0.0411, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0618, 0.0908, 0.0836], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 21:27:24,065 INFO [train.py:968] (1/2) Epoch 17, batch 19850, giga_loss[loss=0.2386, simple_loss=0.3101, pruned_loss=0.08358, over 28929.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3245, pruned_loss=0.08715, over 5724301.49 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3404, pruned_loss=0.08958, over 5770807.00 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3226, pruned_loss=0.08713, over 5716739.28 frames. ], batch size: 106, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:27:33,348 INFO [optim.py:369] (1/2) 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 21:27:39,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=750749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:27:49,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4726, 3.2996, 1.6124, 1.5840], device='cuda:1'), covar=tensor([0.0976, 0.0317, 0.0844, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0528, 0.0364, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 21:28:03,946 INFO [train.py:968] (1/2) Epoch 17, batch 19900, giga_loss[loss=0.2504, simple_loss=0.3176, pruned_loss=0.09161, over 28908.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3246, pruned_loss=0.08753, over 5715834.22 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3413, pruned_loss=0.08983, over 5765357.24 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3217, pruned_loss=0.08723, over 5712663.77 frames. ], batch size: 112, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:28:42,367 INFO [train.py:968] (1/2) Epoch 17, batch 19950, giga_loss[loss=0.2391, simple_loss=0.3145, pruned_loss=0.08187, over 28899.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3224, pruned_loss=0.08613, over 5724182.06 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3413, pruned_loss=0.08956, over 5769771.04 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3197, pruned_loss=0.08606, over 5716608.59 frames. ], batch size: 145, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:28:52,856 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([5.8708, 5.6802, 5.3652, 3.1598], device='cuda:1'), covar=tensor([0.0386, 0.0516, 0.0549, 0.1454], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.1058, 0.0909, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 21:29:22,726 INFO [train.py:968] (1/2) Epoch 17, batch 20000, giga_loss[loss=0.2309, simple_loss=0.303, pruned_loss=0.07936, over 29016.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3201, pruned_loss=0.08458, over 5732230.02 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3415, pruned_loss=0.08953, over 5772367.72 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3171, pruned_loss=0.08442, over 5722568.70 frames. ], batch size: 136, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:29:35,032 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=750892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:29:37,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=750895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:29:58,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=750924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:30:01,963 INFO [train.py:968] (1/2) Epoch 17, batch 20050, giga_loss[loss=0.2405, simple_loss=0.3189, pruned_loss=0.081, over 28844.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3184, pruned_loss=0.0838, over 5725503.33 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.342, pruned_loss=0.08983, over 5763478.69 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3154, pruned_loss=0.08337, over 5725452.32 frames. ], batch size: 199, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:30:09,967 INFO [optim.py:369] (1/2) 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 21:30:44,492 INFO [train.py:968] (1/2) Epoch 17, batch 20100, libri_loss[loss=0.2765, simple_loss=0.3614, pruned_loss=0.09579, over 29535.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3202, pruned_loss=0.08515, over 5719212.54 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3424, pruned_loss=0.08995, over 5752772.58 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3173, pruned_loss=0.08464, over 5727198.68 frames. ], batch size: 81, lr: 1.89e-03, grad_scale: 8.0
+2023-03-08 21:31:21,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:31:32,045 INFO [train.py:968] (1/2) Epoch 17, batch 20150, giga_loss[loss=0.2691, simple_loss=0.3393, pruned_loss=0.09951, over 28921.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3268, pruned_loss=0.08954, over 5711268.93 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3423, pruned_loss=0.0898, over 5755950.03 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3242, pruned_loss=0.08921, over 5714192.74 frames. ], batch size: 106, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:31:37,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6182, 1.4699, 1.6878, 1.2732], device='cuda:1'), covar=tensor([0.1912, 0.2788, 0.1471, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0693, 0.0922, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 21:31:44,689 INFO [optim.py:369] (1/2) 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 21:31:45,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-08 21:31:57,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:32:05,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6437, 1.5636, 1.2466, 1.2033], device='cuda:1'), covar=tensor([0.0689, 0.0483, 0.0847, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0441, 0.0511, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 21:32:27,508 INFO [train.py:968] (1/2) Epoch 17, batch 20200, giga_loss[loss=0.3141, simple_loss=0.3782, pruned_loss=0.125, over 28622.00 frames. ], tot_loss[loss=0.265, simple_loss=0.337, pruned_loss=0.09655, over 5695567.85 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08975, over 5757451.73 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3348, pruned_loss=0.09635, over 5696139.86 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:32:34,805 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:33:11,423 INFO [train.py:968] (1/2) Epoch 17, batch 20250, giga_loss[loss=0.3243, simple_loss=0.3969, pruned_loss=0.1259, over 28627.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.343, pruned_loss=0.09981, over 5697300.79 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08972, over 5761257.05 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3412, pruned_loss=0.09989, over 5693035.18 frames. ], batch size: 307, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:33:23,898 INFO [optim.py:369] (1/2) 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 21:33:48,749 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751162.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:33:50,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751165.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:34:04,032 INFO [train.py:968] (1/2) Epoch 17, batch 20300, giga_loss[loss=0.2942, simple_loss=0.3668, pruned_loss=0.1108, over 28909.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3479, pruned_loss=0.1017, over 5682510.36 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3423, pruned_loss=0.08971, over 5760809.32 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3466, pruned_loss=0.1019, over 5678925.29 frames. ], batch size: 106, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:34:12,469 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:17,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:34:18,990 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:22,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:36,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4723, 1.5903, 1.7019, 1.3232], device='cuda:1'), covar=tensor([0.1423, 0.1947, 0.1141, 0.1425], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0692, 0.0918, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 21:34:41,250 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 21:34:48,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:34:48,475 INFO [train.py:968] (1/2) Epoch 17, batch 20350, giga_loss[loss=0.4497, simple_loss=0.4627, pruned_loss=0.2183, over 26545.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3523, pruned_loss=0.1038, over 5684059.03 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3425, pruned_loss=0.08974, over 5764263.08 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3512, pruned_loss=0.1042, over 5676479.79 frames. ], batch size: 555, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:34:53,587 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:35:00,890 INFO [optim.py:369] (1/2) 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 21:35:08,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751247.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:35:09,479 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-08 21:35:35,876 INFO [train.py:968] (1/2) Epoch 17, batch 20400, giga_loss[loss=0.3515, simple_loss=0.4053, pruned_loss=0.1489, over 27497.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3579, pruned_loss=0.1071, over 5679047.93 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08964, over 5766012.60 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3573, pruned_loss=0.1078, over 5669961.33 frames. ], batch size: 472, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:36:17,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9175, 4.7445, 4.4551, 2.3543], device='cuda:1'), covar=tensor([0.0469, 0.0552, 0.0567, 0.1989], device='cuda:1'), in_proj_covar=tensor([0.1139, 0.1054, 0.0902, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 21:36:17,399 INFO [train.py:968] (1/2) Epoch 17, batch 20450, giga_loss[loss=0.2352, simple_loss=0.3139, pruned_loss=0.07822, over 28595.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3538, pruned_loss=0.1041, over 5686410.04 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3426, pruned_loss=0.08974, over 5770176.30 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3537, pruned_loss=0.1051, over 5672762.63 frames. ], batch size: 262, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:36:29,102 INFO [optim.py:369] (1/2) 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 21:36:41,162 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7354, 5.5358, 5.2952, 2.4593], device='cuda:1'), covar=tensor([0.0483, 0.0642, 0.0762, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.1139, 0.1055, 0.0902, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 21:36:57,136 INFO [train.py:968] (1/2) Epoch 17, batch 20500, giga_loss[loss=0.2533, simple_loss=0.3326, pruned_loss=0.08695, over 28689.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3507, pruned_loss=0.1012, over 5692643.01 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3429, pruned_loss=0.08998, over 5768705.86 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3505, pruned_loss=0.1022, over 5680500.31 frames. ], batch size: 92, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:37:10,751 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:37:16,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1252, 5.9398, 5.6588, 2.7065], device='cuda:1'), covar=tensor([0.0442, 0.0626, 0.0752, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.1140, 0.1056, 0.0904, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 21:37:41,240 INFO [train.py:968] (1/2) Epoch 17, batch 20550, giga_loss[loss=0.2855, simple_loss=0.3648, pruned_loss=0.1031, over 29044.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3485, pruned_loss=0.09916, over 5700760.75 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3429, pruned_loss=0.09, over 5772656.85 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3486, pruned_loss=0.1001, over 5685985.84 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:37:45,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5276, 1.5384, 1.3229, 1.5493], device='cuda:1'), covar=tensor([0.0771, 0.0326, 0.0325, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 21:37:53,397 INFO [optim.py:369] (1/2) 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 21:38:11,909 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 20600, giga_loss[loss=0.265, simple_loss=0.3414, pruned_loss=0.09429, over 28518.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3491, pruned_loss=0.09919, over 5695473.21 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3428, pruned_loss=0.08991, over 5773772.86 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3493, pruned_loss=0.1001, over 5682214.78 frames. ], batch size: 71, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:38:55,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3423, 1.4293, 1.3563, 1.3301], device='cuda:1'), covar=tensor([0.2293, 0.2176, 0.2066, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1869, 0.1786, 0.1724, 0.1872], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 21:39:11,483 INFO [train.py:968] (1/2) Epoch 17, batch 20650, libri_loss[loss=0.2502, simple_loss=0.3277, pruned_loss=0.08635, over 28233.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.351, pruned_loss=0.1003, over 5697898.72 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3425, pruned_loss=0.08978, over 5773763.98 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3514, pruned_loss=0.1012, over 5686866.93 frames. ], batch size: 62, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:39:24,703 INFO [optim.py:369] (1/2) 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 21:39:41,293 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-08 21:39:44,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:39:56,389 INFO [train.py:968] (1/2) Epoch 17, batch 20700, giga_loss[loss=0.2667, simple_loss=0.3536, pruned_loss=0.08987, over 28906.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3534, pruned_loss=0.1022, over 5702112.27 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.343, pruned_loss=0.09008, over 5766727.21 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3535, pruned_loss=0.1028, over 5698018.85 frames. ], batch size: 174, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:40:24,483 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:27,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:28,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:39,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751622.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:40:45,113 INFO [train.py:968] (1/2) Epoch 17, batch 20750, giga_loss[loss=0.2815, simple_loss=0.3622, pruned_loss=0.1004, over 28997.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3546, pruned_loss=0.1034, over 5683093.94 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3435, pruned_loss=0.09025, over 5764910.03 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3544, pruned_loss=0.104, over 5679646.03 frames. ], batch size: 164, lr: 1.89e-03, grad_scale: 2.0
+2023-03-08 21:40:52,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:40:55,965 INFO [optim.py:369] (1/2) 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 21:41:27,572 INFO [train.py:968] (1/2) Epoch 17, batch 20800, giga_loss[loss=0.2596, simple_loss=0.3398, pruned_loss=0.0897, over 28848.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3559, pruned_loss=0.1049, over 5680509.28 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09017, over 5757147.02 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3559, pruned_loss=0.1056, over 5682685.17 frames. ], batch size: 186, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:41:47,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3787, 1.6937, 1.5721, 1.5389], device='cuda:1'), covar=tensor([0.0789, 0.0314, 0.0296, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 21:41:51,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:41:55,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:01,702 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 20850, giga_loss[loss=0.2711, simple_loss=0.3553, pruned_loss=0.0934, over 28853.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3558, pruned_loss=0.1041, over 5696334.83 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3444, pruned_loss=0.09051, over 5763068.61 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3555, pruned_loss=0.105, over 5690037.97 frames. ], batch size: 66, lr: 1.89e-03, grad_scale: 4.0
+2023-03-08 21:42:16,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:16,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1889, 1.1047, 3.5724, 3.0206], device='cuda:1'), covar=tensor([0.1592, 0.2735, 0.0446, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0617, 0.0909, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 21:42:19,202 INFO [optim.py:369] (1/2) 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 21:42:25,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:27,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:29,401 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:31,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0152, 1.2514, 1.1561, 0.9784], device='cuda:1'), covar=tensor([0.1948, 0.2205, 0.1311, 0.1717], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1790, 0.1725, 0.1873], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 21:42:35,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751765.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:42:37,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=751768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:42:37,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751768.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:42:46,759 INFO [train.py:968] (1/2) Epoch 17, batch 20900, giga_loss[loss=0.3012, simple_loss=0.3672, pruned_loss=0.1176, over 27634.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3558, pruned_loss=0.1038, over 5700343.07 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3445, pruned_loss=0.09069, over 5766969.66 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3558, pruned_loss=0.1047, over 5689916.96 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:42:48,427 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751781.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:42:49,866 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:43:00,666 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:43:27,603 INFO [train.py:968] (1/2) Epoch 17, batch 20950, giga_loss[loss=0.3065, simple_loss=0.3791, pruned_loss=0.117, over 28691.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3554, pruned_loss=0.1021, over 5705170.13 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3442, pruned_loss=0.09051, over 5769702.80 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3558, pruned_loss=0.1032, over 5693393.92 frames. ], batch size: 242, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:43:38,421 INFO [optim.py:369] (1/2) 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 21:43:53,649 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=751862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:44:05,997 INFO [train.py:968] (1/2) Epoch 17, batch 21000, giga_loss[loss=0.2512, simple_loss=0.3281, pruned_loss=0.08711, over 28698.00 frames. ], tot_loss[loss=0.278, simple_loss=0.354, pruned_loss=0.101, over 5695140.17 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3441, pruned_loss=0.0905, over 5759439.22 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3548, pruned_loss=0.1024, over 5691913.08 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:44:05,997 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 21:44:12,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3499, 3.1578, 1.4102, 1.5117], device='cuda:1'), covar=tensor([0.1099, 0.0289, 0.1038, 0.1527], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0527, 0.0363, 0.0406], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 21:44:14,794 INFO [train.py:1012] (1/2) Epoch 17, validation: loss=0.21, simple_loss=0.3167, pruned_loss=0.05164, over 944034.00 frames. 
+2023-03-08 21:44:14,794 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 21:44:38,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=751911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:44:40,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=751914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:44:53,332 INFO [train.py:968] (1/2) Epoch 17, batch 21050, giga_loss[loss=0.3019, simple_loss=0.3626, pruned_loss=0.1206, over 28922.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3518, pruned_loss=0.09994, over 5711883.38 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3444, pruned_loss=0.09075, over 5766068.52 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3528, pruned_loss=0.1013, over 5700582.25 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:45:02,814 INFO [optim.py:369] (1/2) 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 21:45:03,251 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=751943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:45:13,313 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 21100, giga_loss[loss=0.2434, simple_loss=0.3299, pruned_loss=0.07841, over 28958.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.09985, over 5715777.00 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3446, pruned_loss=0.09101, over 5769614.58 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3515, pruned_loss=0.1009, over 5702241.58 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:46:09,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 21:46:09,382 INFO [train.py:968] (1/2) Epoch 17, batch 21150, giga_loss[loss=0.2465, simple_loss=0.3245, pruned_loss=0.08423, over 28433.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.0992, over 5722271.06 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3447, pruned_loss=0.09116, over 5771183.80 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3498, pruned_loss=0.1001, over 5709088.16 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:46:10,277 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:46:15,890 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-08 21:46:22,173 INFO [optim.py:369] (1/2) 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 21:46:53,622 INFO [train.py:968] (1/2) Epoch 17, batch 21200, giga_loss[loss=0.5069, simple_loss=0.5012, pruned_loss=0.2563, over 26536.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3504, pruned_loss=0.1009, over 5711550.69 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.345, pruned_loss=0.09137, over 5769981.62 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3506, pruned_loss=0.1016, over 5701258.53 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:47:05,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:47:34,032 INFO [train.py:968] (1/2) Epoch 17, batch 21250, giga_loss[loss=0.2562, simple_loss=0.3325, pruned_loss=0.08996, over 28513.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1006, over 5725121.01 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.09191, over 5775074.52 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3507, pruned_loss=0.1009, over 5710542.74 frames. ], batch size: 85, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:47:36,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:47:45,302 INFO [optim.py:369] (1/2) 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 21:47:51,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5588, 1.8880, 1.5379, 1.6392], device='cuda:1'), covar=tensor([0.1827, 0.2148, 0.2261, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0731, 0.0689, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 21:47:55,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752156.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:48:13,946 INFO [train.py:968] (1/2) Epoch 17, batch 21300, giga_loss[loss=0.2609, simple_loss=0.3467, pruned_loss=0.08758, over 28801.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3505, pruned_loss=0.1002, over 5720520.99 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3457, pruned_loss=0.09214, over 5777825.89 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3505, pruned_loss=0.1005, over 5704816.86 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:48:44,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5040, 1.8535, 1.4842, 1.6817], device='cuda:1'), covar=tensor([0.2584, 0.2500, 0.2864, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1420, 0.1036, 0.1257, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 21:48:55,857 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 21350, giga_loss[loss=0.2614, simple_loss=0.3373, pruned_loss=0.0928, over 28565.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3486, pruned_loss=0.09807, over 5725510.76 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3459, pruned_loss=0.09238, over 5780724.78 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3484, pruned_loss=0.09825, over 5709127.19 frames. ], batch size: 85, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:49:02,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:02,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:07,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:09,243 INFO [scaling.py:679] (1/2) 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] (1/2) 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 21:49:16,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5771, 4.4267, 1.7235, 1.7687], device='cuda:1'), covar=tensor([0.0983, 0.0236, 0.0906, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0524, 0.0362, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0027], device='cuda:1')
+2023-03-08 21:49:29,399 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:34,317 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:36,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:49:36,730 INFO [train.py:968] (1/2) Epoch 17, batch 21400, giga_loss[loss=0.2869, simple_loss=0.3514, pruned_loss=0.1112, over 28768.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3476, pruned_loss=0.09759, over 5731747.69 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3457, pruned_loss=0.09238, over 5782591.52 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3476, pruned_loss=0.09781, over 5716396.93 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:49:40,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4938, 3.3296, 1.5706, 1.5924], device='cuda:1'), covar=tensor([0.0974, 0.0238, 0.0916, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0525, 0.0363, 0.0407], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0031, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 21:49:53,802 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752299.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:49:56,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752302.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:49:59,469 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:50:08,415 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:50:15,162 INFO [train.py:968] (1/2) Epoch 17, batch 21450, giga_loss[loss=0.2369, simple_loss=0.3148, pruned_loss=0.07953, over 28941.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3455, pruned_loss=0.0967, over 5734138.18 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3459, pruned_loss=0.09264, over 5785155.07 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3454, pruned_loss=0.09681, over 5717670.35 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:50:17,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752331.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:50:17,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752332.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:50:28,470 INFO [optim.py:369] (1/2) 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 21:50:58,135 INFO [train.py:968] (1/2) Epoch 17, batch 21500, giga_loss[loss=0.2697, simple_loss=0.3422, pruned_loss=0.0986, over 28923.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3432, pruned_loss=0.09566, over 5728767.32 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3464, pruned_loss=0.09301, over 5786915.96 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3426, pruned_loss=0.09547, over 5712602.90 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:50:59,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:00,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:19,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:23,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:51:38,576 INFO [train.py:968] (1/2) Epoch 17, batch 21550, giga_loss[loss=0.2777, simple_loss=0.3447, pruned_loss=0.1054, over 28575.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3431, pruned_loss=0.09605, over 5724199.36 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.347, pruned_loss=0.09354, over 5778955.76 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.342, pruned_loss=0.0955, over 5716622.12 frames. ], batch size: 85, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:51:49,608 INFO [optim.py:369] (1/2) 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 21:52:13,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:52:15,655 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752475.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:52:19,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752478.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:52:19,651 INFO [train.py:968] (1/2) Epoch 17, batch 21600, giga_loss[loss=0.2938, simple_loss=0.3715, pruned_loss=0.1081, over 28628.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3437, pruned_loss=0.09709, over 5722584.60 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3473, pruned_loss=0.09388, over 5781155.94 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3424, pruned_loss=0.09637, over 5713959.04 frames. ], batch size: 307, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:52:39,920 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752507.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:52:56,412 INFO [train.py:968] (1/2) Epoch 17, batch 21650, giga_loss[loss=0.2628, simple_loss=0.329, pruned_loss=0.09829, over 28739.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3431, pruned_loss=0.09771, over 5723328.19 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3479, pruned_loss=0.09473, over 5780174.66 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3413, pruned_loss=0.09646, over 5714087.72 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:53:03,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6493, 1.6159, 1.2153, 1.2893], device='cuda:1'), covar=tensor([0.0757, 0.0576, 0.1038, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0440, 0.0509, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 21:53:09,117 INFO [optim.py:369] (1/2) 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 21:53:14,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:53:17,009 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 21700, giga_loss[loss=0.2663, simple_loss=0.3325, pruned_loss=0.1001, over 29059.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3414, pruned_loss=0.09733, over 5723581.78 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3487, pruned_loss=0.09558, over 5783553.33 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.339, pruned_loss=0.09559, over 5711725.92 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:53:38,525 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:53:55,393 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752603.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:54:03,229 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 21:54:16,869 INFO [train.py:968] (1/2) Epoch 17, batch 21750, giga_loss[loss=0.2727, simple_loss=0.3433, pruned_loss=0.101, over 29082.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3399, pruned_loss=0.09731, over 5719215.24 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3488, pruned_loss=0.09575, over 5783362.64 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3377, pruned_loss=0.09581, over 5708230.56 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:54:26,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4395, 1.5417, 1.6429, 1.2713], device='cuda:1'), covar=tensor([0.1654, 0.2236, 0.1339, 0.1567], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0692, 0.0919, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 21:54:28,798 INFO [optim.py:369] (1/2) 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 21:54:46,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9763, 1.9999, 1.7411, 1.6724], device='cuda:1'), covar=tensor([0.1754, 0.2615, 0.2425, 0.2531], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0733, 0.0690, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 21:54:57,687 INFO [train.py:968] (1/2) Epoch 17, batch 21800, libri_loss[loss=0.3572, simple_loss=0.4106, pruned_loss=0.1519, over 26068.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3368, pruned_loss=0.09576, over 5710612.58 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3487, pruned_loss=0.09601, over 5774105.48 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3348, pruned_loss=0.09432, over 5708389.90 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:55:08,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:55:10,924 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-08 21:55:37,790 INFO [train.py:968] (1/2) Epoch 17, batch 21850, libri_loss[loss=0.2778, simple_loss=0.3507, pruned_loss=0.1025, over 29524.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3363, pruned_loss=0.09525, over 5710231.44 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.3491, pruned_loss=0.09636, over 5776687.17 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.334, pruned_loss=0.09376, over 5704049.15 frames. ], batch size: 83, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:55:50,984 INFO [optim.py:369] (1/2) 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 21:55:53,187 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752746.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 21:55:55,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752749.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:56:21,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752778.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 21:56:21,645 INFO [train.py:968] (1/2) Epoch 17, batch 21900, libri_loss[loss=0.2876, simple_loss=0.3579, pruned_loss=0.1087, over 29569.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.338, pruned_loss=0.09574, over 5713760.46 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3494, pruned_loss=0.0966, over 5779544.09 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3356, pruned_loss=0.0943, over 5704646.34 frames. ], batch size: 79, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:56:32,572 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 21:56:42,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=752803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:01,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6029, 1.8105, 1.2776, 1.4042], device='cuda:1'), covar=tensor([0.0930, 0.0590, 0.1047, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0438, 0.0506, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 21:57:06,957 INFO [train.py:968] (1/2) Epoch 17, batch 21950, giga_loss[loss=0.273, simple_loss=0.3441, pruned_loss=0.101, over 28673.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3406, pruned_loss=0.09688, over 5714147.52 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3495, pruned_loss=0.09682, over 5780760.57 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3385, pruned_loss=0.09556, over 5705233.16 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 21:57:15,188 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:17,035 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:57:22,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=752847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:33,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3722, 1.6716, 1.3608, 1.3060], device='cuda:1'), covar=tensor([0.2325, 0.2337, 0.2627, 0.2184], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1035, 0.1259, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 21:57:40,742 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=752869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:57:47,775 INFO [train.py:968] (1/2) Epoch 17, batch 22000, giga_loss[loss=0.2554, simple_loss=0.3397, pruned_loss=0.08562, over 28849.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3425, pruned_loss=0.09688, over 5698284.15 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3497, pruned_loss=0.0972, over 5762172.25 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3405, pruned_loss=0.09545, over 5705145.57 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:57:55,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5446, 1.8553, 1.6945, 1.4595], device='cuda:1'), covar=tensor([0.2984, 0.2416, 0.2541, 0.2631], device='cuda:1'), in_proj_covar=tensor([0.1873, 0.1800, 0.1733, 0.1874], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 21:58:30,395 INFO [train.py:968] (1/2) Epoch 17, batch 22050, giga_loss[loss=0.2498, simple_loss=0.3253, pruned_loss=0.0872, over 28788.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3435, pruned_loss=0.09669, over 5689504.22 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.35, pruned_loss=0.09755, over 5755881.16 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3415, pruned_loss=0.09522, over 5699614.75 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:58:43,015 INFO [optim.py:369] (1/2) 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 21:58:47,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4865, 3.6124, 1.5437, 1.5637], device='cuda:1'), covar=tensor([0.0911, 0.0287, 0.0947, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0531, 0.0364, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 21:59:10,980 INFO [train.py:968] (1/2) Epoch 17, batch 22100, giga_loss[loss=0.2632, simple_loss=0.3469, pruned_loss=0.08971, over 28642.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3447, pruned_loss=0.09769, over 5678499.31 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.351, pruned_loss=0.0986, over 5746198.48 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3417, pruned_loss=0.09544, over 5692526.59 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 21:59:15,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 21:59:21,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=752990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:59:23,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=752993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:59:29,480 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-08 21:59:46,719 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 21:59:50,641 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 22150, giga_loss[loss=0.2238, simple_loss=0.3072, pruned_loss=0.07023, over 28873.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3442, pruned_loss=0.09751, over 5689388.25 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3511, pruned_loss=0.09879, over 5747503.06 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3417, pruned_loss=0.09554, over 5698283.33 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:00:06,108 INFO [optim.py:369] (1/2) 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 22:00:09,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3915, 3.3988, 1.3864, 1.6145], device='cuda:1'), covar=tensor([0.0919, 0.0297, 0.0943, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0532, 0.0365, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 22:00:19,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4271, 1.9883, 1.3451, 0.6132], device='cuda:1'), covar=tensor([0.5430, 0.2477, 0.3892, 0.5988], device='cuda:1'), in_proj_covar=tensor([0.1656, 0.1562, 0.1538, 0.1349], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 22:00:34,574 INFO [train.py:968] (1/2) Epoch 17, batch 22200, giga_loss[loss=0.2749, simple_loss=0.3533, pruned_loss=0.09827, over 28609.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3447, pruned_loss=0.09803, over 5690774.59 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3509, pruned_loss=0.09885, over 5747844.55 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09637, over 5696372.34 frames. ], batch size: 307, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:01:14,877 INFO [train.py:968] (1/2) Epoch 17, batch 22250, giga_loss[loss=0.3137, simple_loss=0.3763, pruned_loss=0.1256, over 28904.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3471, pruned_loss=0.09973, over 5699347.36 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3512, pruned_loss=0.09935, over 5752107.32 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.345, pruned_loss=0.09794, over 5698380.43 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:01:20,404 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-08 22:01:22,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6470, 1.7275, 1.2410, 1.2813], device='cuda:1'), covar=tensor([0.0836, 0.0588, 0.1033, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0373, 0.0437, 0.0505, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-08 22:01:30,115 INFO [optim.py:369] (1/2) 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 22:01:33,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-08 22:01:55,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:01:55,556 INFO [train.py:968] (1/2) Epoch 17, batch 22300, giga_loss[loss=0.2964, simple_loss=0.3708, pruned_loss=0.111, over 28738.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3508, pruned_loss=0.1022, over 5700110.31 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.352, pruned_loss=0.1, over 5753137.98 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3484, pruned_loss=0.1002, over 5697063.15 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:02:02,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3584, 1.5875, 1.3457, 1.5364], device='cuda:1'), covar=tensor([0.0725, 0.0314, 0.0324, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0114, 0.0114, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 22:02:36,906 INFO [train.py:968] (1/2) Epoch 17, batch 22350, giga_loss[loss=0.2998, simple_loss=0.3705, pruned_loss=0.1145, over 29026.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 5710713.07 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3521, pruned_loss=0.1003, over 5756472.95 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3505, pruned_loss=0.1007, over 5704217.58 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:02:50,135 INFO [optim.py:369] (1/2) 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 22:03:07,679 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:03:16,424 INFO [train.py:968] (1/2) Epoch 17, batch 22400, giga_loss[loss=0.3081, simple_loss=0.3742, pruned_loss=0.121, over 28947.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3534, pruned_loss=0.103, over 5705447.15 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3524, pruned_loss=0.1004, over 5747598.63 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3516, pruned_loss=0.1015, over 5707258.48 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:03:40,903 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-08 22:03:42,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6445, 4.4271, 1.7498, 1.7434], device='cuda:1'), covar=tensor([0.0920, 0.0289, 0.0912, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0531, 0.0364, 0.0408], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 22:03:51,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:03:54,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:03:57,475 INFO [train.py:968] (1/2) Epoch 17, batch 22450, giga_loss[loss=0.2846, simple_loss=0.3609, pruned_loss=0.1042, over 28349.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3548, pruned_loss=0.1038, over 5713149.07 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3535, pruned_loss=0.1015, over 5751757.49 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3524, pruned_loss=0.1017, over 5709361.30 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:04:13,315 INFO [optim.py:369] (1/2) 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 22:04:18,021 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:04:18,790 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-08 22:04:19,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:04:27,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3125, 1.5306, 1.5150, 1.3582], device='cuda:1'), covar=tensor([0.2449, 0.2166, 0.1536, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.1877, 0.1804, 0.1735, 0.1875], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 22:04:33,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7106, 1.6912, 1.2711, 1.2900], device='cuda:1'), covar=tensor([0.0844, 0.0682, 0.1080, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0375, 0.0440, 0.0507, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 22:04:40,053 INFO [train.py:968] (1/2) Epoch 17, batch 22500, giga_loss[loss=0.2629, simple_loss=0.3339, pruned_loss=0.09595, over 28989.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3549, pruned_loss=0.1041, over 5712260.36 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3535, pruned_loss=0.1017, over 5753281.20 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3529, pruned_loss=0.1023, over 5707541.98 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:05:00,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:05:09,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1988, 1.5545, 1.5133, 1.1015], device='cuda:1'), covar=tensor([0.1592, 0.2261, 0.1339, 0.1521], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0692, 0.0917, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 22:05:15,530 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-08 22:05:23,634 INFO [train.py:968] (1/2) Epoch 17, batch 22550, giga_loss[loss=0.3565, simple_loss=0.3993, pruned_loss=0.1569, over 27940.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3518, pruned_loss=0.1023, over 5712294.59 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3541, pruned_loss=0.1021, over 5751072.56 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3498, pruned_loss=0.1006, over 5710312.53 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:05:40,161 INFO [optim.py:369] (1/2) 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 22:06:06,901 INFO [train.py:968] (1/2) Epoch 17, batch 22600, giga_loss[loss=0.2549, simple_loss=0.3356, pruned_loss=0.08712, over 29064.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.349, pruned_loss=0.1009, over 5700153.47 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3548, pruned_loss=0.1027, over 5745954.54 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3467, pruned_loss=0.0989, over 5702942.05 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:06:47,128 INFO [train.py:968] (1/2) Epoch 17, batch 22650, giga_loss[loss=0.2657, simple_loss=0.3349, pruned_loss=0.09825, over 28724.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.347, pruned_loss=0.09993, over 5690924.96 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3553, pruned_loss=0.1033, over 5731919.64 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3445, pruned_loss=0.09779, over 5705280.41 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:06:59,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:07:02,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:07:03,346 INFO [optim.py:369] (1/2) 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 22:07:23,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=753570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:07:28,647 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 22700, giga_loss[loss=0.2147, simple_loss=0.3003, pruned_loss=0.0645, over 28296.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3463, pruned_loss=0.09834, over 5688440.32 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3553, pruned_loss=0.1033, over 5732824.05 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3443, pruned_loss=0.09659, over 5698775.80 frames. ], batch size: 60, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:08:15,070 INFO [train.py:968] (1/2) Epoch 17, batch 22750, giga_loss[loss=0.2817, simple_loss=0.3517, pruned_loss=0.1058, over 28811.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3484, pruned_loss=0.09837, over 5692138.75 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.356, pruned_loss=0.1039, over 5734096.54 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3461, pruned_loss=0.09635, over 5698570.02 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:08:23,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:08:30,419 INFO [optim.py:369] (1/2) 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 22:08:53,373 INFO [train.py:968] (1/2) Epoch 17, batch 22800, libri_loss[loss=0.3718, simple_loss=0.4272, pruned_loss=0.1582, over 29762.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3486, pruned_loss=0.09915, over 5684279.84 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3571, pruned_loss=0.1049, over 5725923.26 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3454, pruned_loss=0.09644, over 5696129.89 frames. ], batch size: 87, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:08:56,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-08 22:09:32,614 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-08 22:09:34,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=753726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:09:36,579 INFO [train.py:968] (1/2) Epoch 17, batch 22850, giga_loss[loss=0.2976, simple_loss=0.3652, pruned_loss=0.115, over 28645.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3469, pruned_loss=0.09984, over 5691231.63 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3571, pruned_loss=0.1051, over 5729517.98 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3442, pruned_loss=0.09744, over 5696666.30 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:09:52,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3046, 1.2196, 1.0888, 1.4665], device='cuda:1'), covar=tensor([0.0748, 0.0361, 0.0360, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 22:09:52,897 INFO [optim.py:369] (1/2) 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 22:10:13,856 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 22900, giga_loss[loss=0.2529, simple_loss=0.3182, pruned_loss=0.09381, over 28471.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3444, pruned_loss=0.09939, over 5702637.47 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3568, pruned_loss=0.1051, over 5734190.40 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3422, pruned_loss=0.09731, over 5702002.33 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:10:20,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4475, 2.2948, 2.1853, 1.9344], device='cuda:1'), covar=tensor([0.1652, 0.2466, 0.2224, 0.2503], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0742, 0.0698, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 22:10:21,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:10:23,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:10:47,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:10:57,687 INFO [train.py:968] (1/2) Epoch 17, batch 22950, giga_loss[loss=0.2871, simple_loss=0.3453, pruned_loss=0.1144, over 28811.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3437, pruned_loss=0.1004, over 5708043.55 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3572, pruned_loss=0.1054, over 5736816.23 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3414, pruned_loss=0.09832, over 5704757.27 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:11:14,490 INFO [optim.py:369] (1/2) 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 22:11:30,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=753869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:11:33,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=753872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:11:38,840 INFO [train.py:968] (1/2) Epoch 17, batch 23000, giga_loss[loss=0.2333, simple_loss=0.3036, pruned_loss=0.08147, over 28814.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3426, pruned_loss=0.09976, over 5712243.39 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3573, pruned_loss=0.1055, over 5738194.57 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3405, pruned_loss=0.09798, over 5707954.35 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:11:48,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6215, 1.7197, 1.4246, 1.9343], device='cuda:1'), covar=tensor([0.2531, 0.2686, 0.2964, 0.2437], device='cuda:1'), in_proj_covar=tensor([0.1420, 0.1031, 0.1259, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 22:11:54,598 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=753901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:11:55,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-08 22:12:13,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2087, 1.4248, 1.2913, 1.0875], device='cuda:1'), covar=tensor([0.2607, 0.2258, 0.1468, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1874, 0.1809, 0.1731, 0.1873], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 22:12:16,504 INFO [train.py:968] (1/2) Epoch 17, batch 23050, giga_loss[loss=0.2614, simple_loss=0.345, pruned_loss=0.08892, over 28927.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3404, pruned_loss=0.09854, over 5714826.07 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.358, pruned_loss=0.1062, over 5736903.78 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3375, pruned_loss=0.09632, over 5712015.34 frames. ], batch size: 174, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:12:30,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 17, batch 23100, giga_loss[loss=0.2454, simple_loss=0.3135, pruned_loss=0.08867, over 28867.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.336, pruned_loss=0.09651, over 5708168.76 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.358, pruned_loss=0.1062, over 5737763.26 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3337, pruned_loss=0.09468, over 5704985.41 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:13:30,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3362, 4.1511, 3.9771, 1.7242], device='cuda:1'), covar=tensor([0.0688, 0.0841, 0.0932, 0.2094], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.1066, 0.0915, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 22:13:36,909 INFO [train.py:968] (1/2) Epoch 17, batch 23150, giga_loss[loss=0.2297, simple_loss=0.3139, pruned_loss=0.07275, over 28932.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3322, pruned_loss=0.09412, over 5717852.83 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3579, pruned_loss=0.1064, over 5744154.04 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3295, pruned_loss=0.09205, over 5708318.29 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:13:54,361 INFO [optim.py:369] (1/2) 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 22:14:19,332 INFO [train.py:968] (1/2) Epoch 17, batch 23200, giga_loss[loss=0.2519, simple_loss=0.3252, pruned_loss=0.08931, over 28838.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3337, pruned_loss=0.09468, over 5717332.06 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3585, pruned_loss=0.1068, over 5745421.33 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3304, pruned_loss=0.09229, over 5707817.59 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:14:27,894 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=754088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:14:29,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=754091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:14:54,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=754120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:14:56,719 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 23250, giga_loss[loss=0.2852, simple_loss=0.357, pruned_loss=0.1067, over 28820.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3361, pruned_loss=0.0954, over 5718435.50 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3583, pruned_loss=0.1069, over 5746907.86 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3334, pruned_loss=0.09333, over 5709155.33 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:15:17,479 INFO [optim.py:369] (1/2) 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 22:15:18,504 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 23300, giga_loss[loss=0.2692, simple_loss=0.3477, pruned_loss=0.09537, over 28843.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3407, pruned_loss=0.09756, over 5711216.61 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3591, pruned_loss=0.1077, over 5740985.60 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3372, pruned_loss=0.09491, over 5707884.86 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:16:26,633 INFO [train.py:968] (1/2) Epoch 17, batch 23350, giga_loss[loss=0.2743, simple_loss=0.3543, pruned_loss=0.09713, over 28879.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3444, pruned_loss=0.099, over 5707586.13 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3594, pruned_loss=0.1079, over 5742372.84 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3413, pruned_loss=0.09662, over 5703490.22 frames. ], batch size: 174, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:16:48,996 INFO [optim.py:369] (1/2) 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 22:16:51,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4574, 1.5069, 1.2565, 1.5628], device='cuda:1'), covar=tensor([0.0707, 0.0310, 0.0328, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0100], device='cuda:1')
+2023-03-08 22:17:16,139 INFO [train.py:968] (1/2) Epoch 17, batch 23400, giga_loss[loss=0.2822, simple_loss=0.3416, pruned_loss=0.1114, over 24097.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3477, pruned_loss=0.1007, over 5692871.08 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3596, pruned_loss=0.108, over 5736136.55 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3447, pruned_loss=0.09845, over 5693641.16 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:17:26,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=754292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:17:30,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=754295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:17:33,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2130, 4.0299, 3.8270, 2.0968], device='cuda:1'), covar=tensor([0.0585, 0.0705, 0.0681, 0.1806], device='cuda:1'), in_proj_covar=tensor([0.1154, 0.1068, 0.0918, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 22:17:34,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3284, 3.1511, 2.9807, 1.3502], device='cuda:1'), covar=tensor([0.0815, 0.0949, 0.0805, 0.2259], device='cuda:1'), in_proj_covar=tensor([0.1154, 0.1068, 0.0918, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 22:17:54,930 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=754324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:17:59,786 INFO [train.py:968] (1/2) Epoch 17, batch 23450, giga_loss[loss=0.427, simple_loss=0.4421, pruned_loss=0.2059, over 23639.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3513, pruned_loss=0.1038, over 5694714.69 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3599, pruned_loss=0.1083, over 5739566.64 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3484, pruned_loss=0.1017, over 5691454.94 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:18:17,169 INFO [optim.py:369] (1/2) 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 22:18:46,907 INFO [train.py:968] (1/2) Epoch 17, batch 23500, libri_loss[loss=0.2866, simple_loss=0.3534, pruned_loss=0.1099, over 29591.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.357, pruned_loss=0.1089, over 5685197.04 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3598, pruned_loss=0.1084, over 5735764.84 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3546, pruned_loss=0.107, over 5684340.05 frames. ], batch size: 74, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:19:26,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-08 22:19:38,261 INFO [train.py:968] (1/2) Epoch 17, batch 23550, libri_loss[loss=0.3551, simple_loss=0.4193, pruned_loss=0.1455, over 29629.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3634, pruned_loss=0.1134, over 5692388.28 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.36, pruned_loss=0.1086, over 5740664.11 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3612, pruned_loss=0.1117, over 5685867.84 frames. ], batch size: 91, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:19:57,718 INFO [optim.py:369] (1/2) 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 22:20:29,115 INFO [train.py:968] (1/2) Epoch 17, batch 23600, giga_loss[loss=0.3111, simple_loss=0.381, pruned_loss=0.1206, over 28427.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3691, pruned_loss=0.1178, over 5684242.26 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3604, pruned_loss=0.1089, over 5739614.93 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3672, pruned_loss=0.1163, over 5678833.18 frames. ], batch size: 65, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:20:49,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=754498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:21:20,162 INFO [train.py:968] (1/2) Epoch 17, batch 23650, giga_loss[loss=0.3498, simple_loss=0.4052, pruned_loss=0.1472, over 28346.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3758, pruned_loss=0.124, over 5675457.20 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3609, pruned_loss=0.1093, over 5742330.99 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.374, pruned_loss=0.1227, over 5667296.51 frames. ], batch size: 369, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:21:39,783 INFO [optim.py:369] (1/2) 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 22:22:06,597 INFO [train.py:968] (1/2) Epoch 17, batch 23700, giga_loss[loss=0.3514, simple_loss=0.422, pruned_loss=0.1404, over 28977.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3805, pruned_loss=0.1279, over 5663706.04 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3606, pruned_loss=0.1092, over 5732810.52 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3802, pruned_loss=0.1277, over 5662482.97 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:22:31,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5549, 3.5907, 1.7032, 1.6204], device='cuda:1'), covar=tensor([0.0974, 0.0411, 0.0890, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0535, 0.0366, 0.0409], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 22:22:43,224 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=754616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:22:43,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6043, 4.4192, 4.2247, 2.0520], device='cuda:1'), covar=tensor([0.0536, 0.0678, 0.0725, 0.1959], device='cuda:1'), in_proj_covar=tensor([0.1163, 0.1073, 0.0920, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 22:22:56,158 INFO [train.py:968] (1/2) Epoch 17, batch 23750, giga_loss[loss=0.2909, simple_loss=0.3683, pruned_loss=0.1067, over 28984.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3833, pruned_loss=0.1296, over 5671476.59 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3609, pruned_loss=0.1094, over 5735445.96 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.383, pruned_loss=0.1295, over 5667236.85 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:23:08,389 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=754641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:23:11,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=754644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:23:17,735 INFO [optim.py:369] (1/2) 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 22:23:41,864 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 23800, giga_loss[loss=0.2775, simple_loss=0.3444, pruned_loss=0.1053, over 28554.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3855, pruned_loss=0.1329, over 5664126.30 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3612, pruned_loss=0.1097, over 5739126.44 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3855, pruned_loss=0.133, over 5656153.93 frames. ], batch size: 71, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:23:49,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=754680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:24:19,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-08 22:24:43,142 INFO [train.py:968] (1/2) Epoch 17, batch 23850, libri_loss[loss=0.3267, simple_loss=0.382, pruned_loss=0.1357, over 19772.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3884, pruned_loss=0.1361, over 5639492.57 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3611, pruned_loss=0.1097, over 5728448.17 frames. ], giga_tot_loss[loss=0.3308, simple_loss=0.3887, pruned_loss=0.1365, over 5642358.54 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:25:03,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9254, 1.1359, 1.0853, 0.8045], device='cuda:1'), covar=tensor([0.2118, 0.2448, 0.1388, 0.2088], device='cuda:1'), in_proj_covar=tensor([0.1887, 0.1826, 0.1746, 0.1883], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 22:25:04,938 INFO [optim.py:369] (1/2) 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 22:25:33,557 INFO [train.py:968] (1/2) Epoch 17, batch 23900, giga_loss[loss=0.471, simple_loss=0.4622, pruned_loss=0.2399, over 23502.00 frames. ], tot_loss[loss=0.3359, simple_loss=0.3919, pruned_loss=0.14, over 5636183.12 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3612, pruned_loss=0.1102, over 5735277.94 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3932, pruned_loss=0.1409, over 5628899.65 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:26:32,907 INFO [train.py:968] (1/2) Epoch 17, batch 23950, giga_loss[loss=0.3298, simple_loss=0.3846, pruned_loss=0.1375, over 28847.00 frames. ], tot_loss[loss=0.3404, simple_loss=0.3945, pruned_loss=0.1431, over 5617722.92 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3616, pruned_loss=0.1105, over 5738494.16 frames. ], giga_tot_loss[loss=0.342, simple_loss=0.3958, pruned_loss=0.1441, over 5607558.72 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:26:52,693 INFO [optim.py:369] (1/2) 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 22:27:12,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-08 22:27:21,711 INFO [train.py:968] (1/2) Epoch 17, batch 24000, giga_loss[loss=0.2769, simple_loss=0.3481, pruned_loss=0.1029, over 28896.00 frames. ], tot_loss[loss=0.3375, simple_loss=0.3917, pruned_loss=0.1417, over 5626948.88 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3612, pruned_loss=0.1105, over 5738988.36 frames. ], giga_tot_loss[loss=0.3405, simple_loss=0.3941, pruned_loss=0.1435, over 5614841.38 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:27:21,712 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 22:27:29,869 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 22:27:31,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-08 22:27:45,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=754896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:28:13,441 INFO [train.py:968] (1/2) Epoch 17, batch 24050, libri_loss[loss=0.3019, simple_loss=0.3742, pruned_loss=0.1149, over 29270.00 frames. ], tot_loss[loss=0.3346, simple_loss=0.3895, pruned_loss=0.1399, over 5633649.02 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3617, pruned_loss=0.1108, over 5734196.95 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3919, pruned_loss=0.1421, over 5625434.48 frames. ], batch size: 94, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:28:33,675 INFO [optim.py:369] (1/2) 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 22:28:59,833 INFO [train.py:968] (1/2) Epoch 17, batch 24100, giga_loss[loss=0.3428, simple_loss=0.4012, pruned_loss=0.1422, over 28022.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3889, pruned_loss=0.1383, over 5625831.76 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3619, pruned_loss=0.1111, over 5728281.26 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3916, pruned_loss=0.1408, over 5620876.85 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:29:17,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=754991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:29:52,893 INFO [train.py:968] (1/2) Epoch 17, batch 24150, giga_loss[loss=0.2755, simple_loss=0.3521, pruned_loss=0.09941, over 28330.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3901, pruned_loss=0.1387, over 5611134.97 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3625, pruned_loss=0.1115, over 5721001.12 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3925, pruned_loss=0.1411, over 5610235.03 frames. ], batch size: 65, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:30:04,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4658, 1.1213, 4.8411, 3.5750], device='cuda:1'), covar=tensor([0.1824, 0.3042, 0.0404, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0726, 0.0624, 0.0920, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 22:30:16,114 INFO [optim.py:369] (1/2) 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 22:30:18,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:30:43,469 INFO [train.py:968] (1/2) Epoch 17, batch 24200, giga_loss[loss=0.3242, simple_loss=0.3787, pruned_loss=0.1349, over 28625.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3898, pruned_loss=0.1378, over 5616810.19 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3626, pruned_loss=0.1118, over 5717566.63 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3928, pruned_loss=0.1405, over 5615233.59 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:31:04,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:31:34,939 INFO [train.py:968] (1/2) Epoch 17, batch 24250, giga_loss[loss=0.3188, simple_loss=0.3916, pruned_loss=0.123, over 28832.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.388, pruned_loss=0.1362, over 5619432.21 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.362, pruned_loss=0.1115, over 5720574.46 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3912, pruned_loss=0.139, over 5614185.88 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:31:42,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:31:45,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:31:51,617 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-08 22:31:58,439 INFO [optim.py:369] (1/2) 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 22:32:12,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 24300, giga_loss[loss=0.2929, simple_loss=0.3671, pruned_loss=0.1093, over 28960.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3856, pruned_loss=0.1333, over 5632013.94 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3623, pruned_loss=0.1117, over 5723988.13 frames. ], giga_tot_loss[loss=0.33, simple_loss=0.3885, pruned_loss=0.1358, over 5622908.77 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:32:41,262 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:32:45,594 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:32:47,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:33:06,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2224, 3.8012, 1.3169, 1.4453], device='cuda:1'), covar=tensor([0.1128, 0.0477, 0.0996, 0.1478], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0540, 0.0368, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 22:33:17,058 INFO [train.py:968] (1/2) Epoch 17, batch 24350, giga_loss[loss=0.2894, simple_loss=0.3596, pruned_loss=0.1096, over 28279.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3824, pruned_loss=0.1304, over 5630353.14 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3623, pruned_loss=0.1119, over 5724882.18 frames. ], giga_tot_loss[loss=0.3246, simple_loss=0.3847, pruned_loss=0.1323, over 5622042.66 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:33:18,425 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:33:19,975 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.99 vs. limit=5.0
+2023-03-08 22:33:29,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2403, 1.5187, 1.2275, 0.9717], device='cuda:1'), covar=tensor([0.2370, 0.2366, 0.2649, 0.2112], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1033, 0.1261, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 22:33:37,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8431, 1.8574, 2.0366, 1.5690], device='cuda:1'), covar=tensor([0.1737, 0.2326, 0.1356, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0690, 0.0909, 0.0809], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 22:33:38,541 INFO [optim.py:369] (1/2) 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 22:33:46,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:33:56,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:34:00,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:34:04,655 INFO [train.py:968] (1/2) Epoch 17, batch 24400, giga_loss[loss=0.3039, simple_loss=0.3652, pruned_loss=0.1213, over 29013.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.379, pruned_loss=0.1279, over 5636973.42 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3622, pruned_loss=0.1119, over 5728896.76 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3814, pruned_loss=0.1298, over 5624669.21 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:34:50,314 INFO [train.py:968] (1/2) Epoch 17, batch 24450, giga_loss[loss=0.3208, simple_loss=0.3862, pruned_loss=0.1277, over 28814.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3765, pruned_loss=0.1266, over 5641934.12 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3618, pruned_loss=0.1119, over 5734310.53 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3793, pruned_loss=0.1286, over 5623912.64 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:35:14,028 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 24500, giga_loss[loss=0.3288, simple_loss=0.3888, pruned_loss=0.1343, over 28930.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3778, pruned_loss=0.128, over 5641906.27 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3618, pruned_loss=0.1119, over 5734041.71 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3803, pruned_loss=0.1299, over 5626171.23 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:35:48,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5610, 1.6104, 1.8031, 1.3563], device='cuda:1'), covar=tensor([0.1728, 0.2569, 0.1376, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0867, 0.0694, 0.0913, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 22:36:03,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:36:22,076 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:36:25,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:36:36,669 INFO [train.py:968] (1/2) Epoch 17, batch 24550, giga_loss[loss=0.332, simple_loss=0.3778, pruned_loss=0.1432, over 26604.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3766, pruned_loss=0.1263, over 5648328.71 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3618, pruned_loss=0.112, over 5732601.99 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3787, pruned_loss=0.1279, over 5636224.76 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:36:56,628 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:37:02,673 INFO [optim.py:369] (1/2) 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 22:37:25,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:37:29,594 INFO [train.py:968] (1/2) Epoch 17, batch 24600, giga_loss[loss=0.3477, simple_loss=0.3963, pruned_loss=0.1495, over 26539.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3752, pruned_loss=0.1235, over 5659279.88 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3619, pruned_loss=0.1121, over 5733936.96 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3771, pruned_loss=0.1249, over 5646624.98 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:38:21,948 INFO [train.py:968] (1/2) Epoch 17, batch 24650, giga_loss[loss=0.303, simple_loss=0.3559, pruned_loss=0.125, over 23673.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3766, pruned_loss=0.1224, over 5668227.04 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3623, pruned_loss=0.1125, over 5737841.12 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3781, pruned_loss=0.1234, over 5653062.79 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:38:47,132 INFO [optim.py:369] (1/2) 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 22:39:03,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:39:05,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6594, 1.5982, 1.8368, 1.4353], device='cuda:1'), covar=tensor([0.1793, 0.2487, 0.1416, 0.1656], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0693, 0.0912, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 22:39:14,081 INFO [train.py:968] (1/2) Epoch 17, batch 24700, giga_loss[loss=0.2937, simple_loss=0.3659, pruned_loss=0.1108, over 29049.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3773, pruned_loss=0.1232, over 5663006.10 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3622, pruned_loss=0.1126, over 5739662.85 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3787, pruned_loss=0.124, over 5648834.11 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:39:49,930 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1767, 1.2607, 1.1345, 0.9094], device='cuda:1'), covar=tensor([0.0969, 0.0542, 0.1060, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0442, 0.0508, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 22:39:51,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:39:53,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:02,726 INFO [train.py:968] (1/2) Epoch 17, batch 24750, giga_loss[loss=0.3477, simple_loss=0.3998, pruned_loss=0.1478, over 27979.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3771, pruned_loss=0.1228, over 5672092.18 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3626, pruned_loss=0.1129, over 5732158.79 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3782, pruned_loss=0.1235, over 5664687.47 frames. ], batch size: 412, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:40:08,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:12,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:24,461 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:40:24,475 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:40:55,599 INFO [train.py:968] (1/2) Epoch 17, batch 24800, giga_loss[loss=0.2917, simple_loss=0.3615, pruned_loss=0.111, over 28749.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.375, pruned_loss=0.1218, over 5679697.12 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3626, pruned_loss=0.1129, over 5732158.79 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3759, pruned_loss=0.1223, over 5673933.95 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:41:29,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:41:31,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:41:43,262 INFO [train.py:968] (1/2) Epoch 17, batch 24850, giga_loss[loss=0.3246, simple_loss=0.3652, pruned_loss=0.142, over 23779.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3744, pruned_loss=0.1229, over 5677969.78 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3626, pruned_loss=0.1129, over 5736255.20 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3755, pruned_loss=0.1237, over 5668171.90 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:41:56,952 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:42:20,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=755770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:27,159 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:27,534 INFO [train.py:968] (1/2) Epoch 17, batch 24900, libri_loss[loss=0.2804, simple_loss=0.3426, pruned_loss=0.1091, over 29614.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3728, pruned_loss=0.1216, over 5685438.55 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3621, pruned_loss=0.1127, over 5739541.75 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3744, pruned_loss=0.1225, over 5673073.48 frames. ], batch size: 74, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:42:29,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:40,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:43,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=755795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:42:54,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:43:10,438 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:43:13,656 INFO [train.py:968] (1/2) Epoch 17, batch 24950, giga_loss[loss=0.2854, simple_loss=0.3604, pruned_loss=0.1053, over 28559.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3735, pruned_loss=0.1209, over 5678252.92 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3621, pruned_loss=0.1127, over 5730561.18 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3748, pruned_loss=0.1217, over 5676333.74 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:43:37,818 INFO [optim.py:369] (1/2) 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 22:44:01,577 INFO [train.py:968] (1/2) Epoch 17, batch 25000, giga_loss[loss=0.3115, simple_loss=0.3782, pruned_loss=0.1224, over 28425.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3747, pruned_loss=0.1219, over 5658893.21 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3628, pruned_loss=0.1134, over 5713373.47 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3755, pruned_loss=0.1221, over 5670824.16 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:44:04,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3257, 1.6922, 1.2979, 1.3152], device='cuda:1'), covar=tensor([0.2521, 0.2524, 0.3019, 0.2267], device='cuda:1'), in_proj_covar=tensor([0.1423, 0.1035, 0.1265, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 22:44:33,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=755913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:44:35,799 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 25050, giga_loss[loss=0.2944, simple_loss=0.3624, pruned_loss=0.1133, over 28855.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.373, pruned_loss=0.1208, over 5670953.15 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3626, pruned_loss=0.1133, over 5719691.66 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3741, pruned_loss=0.1213, over 5673147.71 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:45:05,199 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=755945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:45:12,060 INFO [optim.py:369] (1/2) 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 22:45:25,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=755965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:45:40,405 INFO [train.py:968] (1/2) Epoch 17, batch 25100, giga_loss[loss=0.3094, simple_loss=0.3792, pruned_loss=0.1198, over 28674.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3719, pruned_loss=0.1204, over 5676234.75 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3623, pruned_loss=0.1132, over 5721725.87 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3731, pruned_loss=0.121, over 5675786.43 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:46:03,504 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=756006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:46:12,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:46:25,864 INFO [train.py:968] (1/2) Epoch 17, batch 25150, giga_loss[loss=0.3644, simple_loss=0.3879, pruned_loss=0.1705, over 23653.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3713, pruned_loss=0.121, over 5667105.55 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3626, pruned_loss=0.1134, over 5728622.76 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3727, pruned_loss=0.1217, over 5657919.70 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:46:44,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5581, 1.7639, 1.7074, 1.3914], device='cuda:1'), covar=tensor([0.2835, 0.2366, 0.2014, 0.2537], device='cuda:1'), in_proj_covar=tensor([0.1895, 0.1828, 0.1752, 0.1892], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 22:46:49,084 INFO [optim.py:369] (1/2) 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 22:47:13,273 INFO [train.py:968] (1/2) Epoch 17, batch 25200, giga_loss[loss=0.3091, simple_loss=0.368, pruned_loss=0.1251, over 28872.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 5670589.89 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3627, pruned_loss=0.1136, over 5728109.28 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.372, pruned_loss=0.122, over 5662879.90 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 22:48:04,166 INFO [train.py:968] (1/2) Epoch 17, batch 25250, giga_loss[loss=0.2493, simple_loss=0.3273, pruned_loss=0.08571, over 28597.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3698, pruned_loss=0.1215, over 5663954.18 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3628, pruned_loss=0.1137, over 5726354.60 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1219, over 5659114.12 frames. ], batch size: 307, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:48:26,481 INFO [optim.py:369] (1/2) 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 22:48:29,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:48:31,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:48:48,928 INFO [train.py:968] (1/2) Epoch 17, batch 25300, giga_loss[loss=0.2934, simple_loss=0.3649, pruned_loss=0.111, over 28669.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3688, pruned_loss=0.1208, over 5668732.85 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3636, pruned_loss=0.1143, over 5722635.70 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3689, pruned_loss=0.1208, over 5667051.34 frames. ], batch size: 242, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:48:57,242 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=756187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:49:01,784 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:49:19,361 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 25350, giga_loss[loss=0.347, simple_loss=0.3831, pruned_loss=0.1554, over 23812.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3689, pruned_loss=0.1214, over 5657506.92 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3639, pruned_loss=0.1145, over 5722712.53 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3687, pruned_loss=0.1212, over 5655811.62 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:50:09,674 INFO [optim.py:369] (1/2) 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 22:50:29,491 INFO [train.py:968] (1/2) Epoch 17, batch 25400, giga_loss[loss=0.2707, simple_loss=0.3497, pruned_loss=0.09582, over 28587.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3698, pruned_loss=0.1213, over 5661419.09 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.364, pruned_loss=0.1147, over 5724732.16 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1212, over 5656713.78 frames. ], batch size: 307, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:51:08,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5762, 2.3344, 1.7604, 0.8500], device='cuda:1'), covar=tensor([0.4327, 0.2603, 0.3560, 0.4902], device='cuda:1'), in_proj_covar=tensor([0.1681, 0.1592, 0.1558, 0.1370], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 22:51:17,191 INFO [train.py:968] (1/2) Epoch 17, batch 25450, giga_loss[loss=0.2923, simple_loss=0.367, pruned_loss=0.1087, over 28853.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.369, pruned_loss=0.1197, over 5669506.83 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3638, pruned_loss=0.1145, over 5727267.16 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3693, pruned_loss=0.1199, over 5662682.21 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:51:26,332 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756340.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:51:30,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2129, 4.0341, 3.8257, 2.0515], device='cuda:1'), covar=tensor([0.0618, 0.0753, 0.0782, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.1102, 0.0942, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 22:51:31,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-08 22:51:38,655 INFO [optim.py:369] (1/2) 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 22:52:03,146 INFO [train.py:968] (1/2) Epoch 17, batch 25500, giga_loss[loss=0.2621, simple_loss=0.3365, pruned_loss=0.0938, over 28471.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3702, pruned_loss=0.1205, over 5657005.78 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3639, pruned_loss=0.1147, over 5717628.22 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3704, pruned_loss=0.1206, over 5658254.87 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:52:06,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:52:28,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.88 vs. limit=5.0
+2023-03-08 22:52:48,889 INFO [train.py:968] (1/2) Epoch 17, batch 25550, giga_loss[loss=0.3136, simple_loss=0.3778, pruned_loss=0.1247, over 28891.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3717, pruned_loss=0.1222, over 5652991.77 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.364, pruned_loss=0.1148, over 5714503.36 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.372, pruned_loss=0.1223, over 5654456.10 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:52:54,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-08 22:53:14,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-08 22:53:16,441 INFO [optim.py:369] (1/2) 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 22:53:40,676 INFO [train.py:968] (1/2) Epoch 17, batch 25600, giga_loss[loss=0.3196, simple_loss=0.3843, pruned_loss=0.1275, over 28866.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3752, pruned_loss=0.126, over 5648383.29 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3637, pruned_loss=0.1148, over 5717390.93 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3758, pruned_loss=0.1263, over 5645747.93 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 22:53:43,465 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756483.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 22:53:45,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756486.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 22:54:20,462 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756515.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:54:27,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:54:27,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=756525.0, num_to_drop=1, layers_to_drop={1}
+2023-03-08 22:54:29,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:54:30,767 INFO [train.py:968] (1/2) Epoch 17, batch 25650, giga_loss[loss=0.3335, simple_loss=0.3856, pruned_loss=0.1407, over 28470.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3757, pruned_loss=0.1272, over 5664751.64 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3637, pruned_loss=0.1147, over 5723116.60 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3765, pruned_loss=0.1279, over 5655589.37 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:54:56,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4182, 3.4581, 1.5674, 1.5506], device='cuda:1'), covar=tensor([0.0949, 0.0386, 0.0858, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0543, 0.0370, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 22:54:59,246 INFO [optim.py:369] (1/2) 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 22:55:00,565 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:55:05,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:55:20,700 INFO [train.py:968] (1/2) Epoch 17, batch 25700, giga_loss[loss=0.2965, simple_loss=0.3601, pruned_loss=0.1165, over 28867.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3769, pruned_loss=0.1292, over 5651906.57 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3637, pruned_loss=0.1147, over 5714453.00 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3778, pruned_loss=0.1299, over 5651116.36 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:55:25,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:56:07,210 INFO [train.py:968] (1/2) Epoch 17, batch 25750, giga_loss[loss=0.3028, simple_loss=0.3688, pruned_loss=0.1184, over 29092.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3775, pruned_loss=0.1295, over 5645937.70 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.364, pruned_loss=0.1149, over 5708635.38 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3782, pruned_loss=0.1302, over 5649956.40 frames. ], batch size: 128, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:56:31,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-08 22:56:33,760 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 25800, libri_loss[loss=0.3545, simple_loss=0.3999, pruned_loss=0.1545, over 20118.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3772, pruned_loss=0.1297, over 5625110.03 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3641, pruned_loss=0.1153, over 5684304.80 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3779, pruned_loss=0.1302, over 5649633.63 frames. ], batch size: 187, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:57:19,043 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:21,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:35,963 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=756724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:38,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=756727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:39,106 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-08 22:57:39,278 INFO [train.py:968] (1/2) Epoch 17, batch 25850, giga_loss[loss=0.2957, simple_loss=0.369, pruned_loss=0.1112, over 28728.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3766, pruned_loss=0.1275, over 5644547.38 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3641, pruned_loss=0.1153, over 5687791.47 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3774, pruned_loss=0.1282, over 5659982.59 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:57:45,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:57:58,106 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-08 22:58:03,018 INFO [optim.py:369] (1/2) 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 22:58:03,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=756756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 22:58:26,976 INFO [train.py:968] (1/2) Epoch 17, batch 25900, giga_loss[loss=0.3009, simple_loss=0.3449, pruned_loss=0.1285, over 23902.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3744, pruned_loss=0.1253, over 5642019.85 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3643, pruned_loss=0.1153, over 5690834.89 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3751, pruned_loss=0.1261, over 5650508.70 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:59:17,158 INFO [train.py:968] (1/2) Epoch 17, batch 25950, giga_loss[loss=0.2794, simple_loss=0.348, pruned_loss=0.1054, over 28856.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1231, over 5651352.00 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3645, pruned_loss=0.1155, over 5690243.89 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1237, over 5658299.20 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 22:59:38,519 INFO [optim.py:369] (1/2) 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 23:00:02,709 INFO [train.py:968] (1/2) Epoch 17, batch 26000, giga_loss[loss=0.2961, simple_loss=0.3584, pruned_loss=0.117, over 28875.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3688, pruned_loss=0.1216, over 5666995.74 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3646, pruned_loss=0.1155, over 5694555.89 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1222, over 5668274.13 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:00:21,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 23:00:29,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=756900.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 23:00:54,286 INFO [train.py:968] (1/2) Epoch 17, batch 26050, libri_loss[loss=0.2552, simple_loss=0.3306, pruned_loss=0.08988, over 29566.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3692, pruned_loss=0.1223, over 5657758.97 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3647, pruned_loss=0.1158, over 5682023.00 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3697, pruned_loss=0.1227, over 5669401.04 frames. ], batch size: 77, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:01:19,963 INFO [optim.py:369] (1/2) 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 23:01:39,826 INFO [train.py:968] (1/2) Epoch 17, batch 26100, libri_loss[loss=0.3059, simple_loss=0.3723, pruned_loss=0.1197, over 29540.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.373, pruned_loss=0.1241, over 5668865.21 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3653, pruned_loss=0.1165, over 5687711.76 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.373, pruned_loss=0.124, over 5672621.11 frames. ], batch size: 83, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:02:30,160 INFO [train.py:968] (1/2) Epoch 17, batch 26150, giga_loss[loss=0.2859, simple_loss=0.3653, pruned_loss=0.1033, over 28870.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3751, pruned_loss=0.1222, over 5672359.93 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3651, pruned_loss=0.1164, over 5689614.08 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3754, pruned_loss=0.1223, over 5673445.17 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:02:44,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=757043.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 23:02:48,318 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:03:18,980 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=757075.0, num_to_drop=1, layers_to_drop={0}
+2023-03-08 23:03:21,697 INFO [train.py:968] (1/2) Epoch 17, batch 26200, giga_loss[loss=0.3273, simple_loss=0.3907, pruned_loss=0.1319, over 28585.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3775, pruned_loss=0.1233, over 5676516.83 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3651, pruned_loss=0.1165, over 5690739.01 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3777, pruned_loss=0.1234, over 5676189.05 frames. ], batch size: 307, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:03:31,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5306, 1.5587, 1.2743, 1.1168], device='cuda:1'), covar=tensor([0.0902, 0.0558, 0.1047, 0.1086], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0441, 0.0507, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 23:03:43,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6946, 3.5146, 3.3360, 1.7269], device='cuda:1'), covar=tensor([0.0826, 0.0923, 0.0867, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.1179, 0.1086, 0.0936, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 23:03:46,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4409, 4.4876, 1.7545, 1.5876], device='cuda:1'), covar=tensor([0.1002, 0.0340, 0.0887, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0542, 0.0369, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 23:04:09,280 INFO [train.py:968] (1/2) Epoch 17, batch 26250, giga_loss[loss=0.2701, simple_loss=0.3495, pruned_loss=0.09531, over 28160.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3782, pruned_loss=0.1245, over 5667555.49 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3651, pruned_loss=0.1167, over 5683523.10 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3788, pruned_loss=0.1245, over 5673337.65 frames. ], batch size: 77, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:04:33,787 INFO [optim.py:369] (1/2) 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 23:04:56,676 INFO [train.py:968] (1/2) Epoch 17, batch 26300, giga_loss[loss=0.3215, simple_loss=0.3762, pruned_loss=0.1334, over 28553.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3784, pruned_loss=0.1252, over 5671835.49 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3648, pruned_loss=0.1166, over 5687566.93 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3793, pruned_loss=0.1255, over 5672677.61 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:05:14,683 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-08 23:05:43,664 INFO [train.py:968] (1/2) Epoch 17, batch 26350, giga_loss[loss=0.3136, simple_loss=0.375, pruned_loss=0.1261, over 28903.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3783, pruned_loss=0.126, over 5681822.63 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3647, pruned_loss=0.1165, over 5693233.90 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3795, pruned_loss=0.1266, over 5676887.07 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:06:07,110 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-08 23:06:07,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2164, 1.9372, 1.4289, 1.6752], device='cuda:1'), covar=tensor([0.0839, 0.0768, 0.1125, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0443, 0.0509, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 23:06:11,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3101, 2.8350, 1.4371, 1.4655], device='cuda:1'), covar=tensor([0.0950, 0.0358, 0.0865, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0542, 0.0369, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 23:06:15,203 INFO [optim.py:369] (1/2) 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 23:06:31,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-08 23:06:34,830 INFO [train.py:968] (1/2) Epoch 17, batch 26400, giga_loss[loss=0.2933, simple_loss=0.3569, pruned_loss=0.1149, over 28669.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3761, pruned_loss=0.1249, over 5685624.73 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3646, pruned_loss=0.1164, over 5694295.73 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3772, pruned_loss=0.1254, over 5680706.72 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:07:23,468 INFO [train.py:968] (1/2) Epoch 17, batch 26450, giga_loss[loss=0.276, simple_loss=0.3435, pruned_loss=0.1042, over 28583.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3757, pruned_loss=0.1258, over 5688881.60 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3647, pruned_loss=0.1166, over 5696782.92 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3767, pruned_loss=0.1262, over 5682493.85 frames. ], batch size: 78, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:07:52,375 INFO [optim.py:369] (1/2) 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 23:08:07,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2874, 1.4908, 1.5066, 1.3097], device='cuda:1'), covar=tensor([0.1730, 0.1693, 0.2277, 0.1860], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0740, 0.0699, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 23:08:15,259 INFO [train.py:968] (1/2) Epoch 17, batch 26500, giga_loss[loss=0.4049, simple_loss=0.4174, pruned_loss=0.1962, over 23579.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 5671611.84 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3647, pruned_loss=0.1167, over 5687144.84 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.126, over 5674294.69 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:08:54,317 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=757422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:08:59,447 INFO [train.py:968] (1/2) Epoch 17, batch 26550, giga_loss[loss=0.3167, simple_loss=0.3688, pruned_loss=0.1323, over 28575.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.374, pruned_loss=0.1252, over 5676971.82 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3646, pruned_loss=0.1167, over 5689513.09 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3751, pruned_loss=0.1257, over 5676715.25 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:09:14,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=757446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:09:21,505 INFO [optim.py:369] (1/2) 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 23:09:44,794 INFO [train.py:968] (1/2) Epoch 17, batch 26600, giga_loss[loss=0.2802, simple_loss=0.3515, pruned_loss=0.1044, over 28982.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.374, pruned_loss=0.1263, over 5666276.27 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3646, pruned_loss=0.1166, over 5691850.41 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.375, pruned_loss=0.127, over 5663689.86 frames. ], batch size: 213, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:09:58,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=757495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:10:31,930 INFO [train.py:968] (1/2) Epoch 17, batch 26650, giga_loss[loss=0.3, simple_loss=0.3684, pruned_loss=0.1158, over 29002.00 frames. ], tot_loss[loss=0.311, simple_loss=0.372, pruned_loss=0.125, over 5661688.20 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3648, pruned_loss=0.1167, over 5694692.13 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3729, pruned_loss=0.1257, over 5655923.16 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:10:58,941 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 26700, libri_loss[loss=0.3053, simple_loss=0.3734, pruned_loss=0.1186, over 29517.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3715, pruned_loss=0.1237, over 5664934.36 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3649, pruned_loss=0.1166, over 5697694.34 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3724, pruned_loss=0.1246, over 5656537.01 frames. ], batch size: 82, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:12:02,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6647, 2.1403, 1.5966, 2.1521], device='cuda:1'), covar=tensor([0.2302, 0.2340, 0.2640, 0.2048], device='cuda:1'), in_proj_covar=tensor([0.1422, 0.1037, 0.1265, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 23:12:05,387 INFO [train.py:968] (1/2) Epoch 17, batch 26750, giga_loss[loss=0.3314, simple_loss=0.3912, pruned_loss=0.1358, over 28615.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3744, pruned_loss=0.1249, over 5660862.06 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3657, pruned_loss=0.117, over 5691556.28 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 5658567.12 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:12:38,671 INFO [optim.py:369] (1/2) 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 23:12:39,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-08 23:12:55,922 INFO [train.py:968] (1/2) Epoch 17, batch 26800, libri_loss[loss=0.3067, simple_loss=0.3752, pruned_loss=0.1191, over 29639.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.374, pruned_loss=0.1249, over 5647888.22 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3659, pruned_loss=0.1171, over 5682725.76 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3741, pruned_loss=0.1253, over 5654312.11 frames. ], batch size: 91, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:13:40,182 INFO [train.py:968] (1/2) Epoch 17, batch 26850, giga_loss[loss=0.2648, simple_loss=0.3534, pruned_loss=0.08813, over 28310.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3743, pruned_loss=0.1233, over 5659519.47 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3659, pruned_loss=0.1172, over 5678376.48 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3745, pruned_loss=0.1238, over 5668553.64 frames. ], batch size: 77, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:13:58,349 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:14:26,161 INFO [train.py:968] (1/2) Epoch 17, batch 26900, giga_loss[loss=0.3036, simple_loss=0.3779, pruned_loss=0.1147, over 28945.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3752, pruned_loss=0.1216, over 5670766.38 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3656, pruned_loss=0.117, over 5684291.48 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3759, pruned_loss=0.1223, over 5672051.46 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:14:45,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=757797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:15:06,244 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 26950, giga_loss[loss=0.3409, simple_loss=0.4029, pruned_loss=0.1395, over 28776.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3781, pruned_loss=0.122, over 5674670.85 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3656, pruned_loss=0.117, over 5682930.85 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3788, pruned_loss=0.1226, over 5676757.85 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:15:37,284 INFO [optim.py:369] (1/2) 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 23:15:42,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-08 23:15:44,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3734, 2.2050, 1.6123, 0.5700], device='cuda:1'), covar=tensor([0.3938, 0.2586, 0.3752, 0.4760], device='cuda:1'), in_proj_covar=tensor([0.1683, 0.1599, 0.1558, 0.1369], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 23:15:48,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=757870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:15:58,922 INFO [train.py:968] (1/2) Epoch 17, batch 27000, giga_loss[loss=0.4398, simple_loss=0.4488, pruned_loss=0.2154, over 23693.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.381, pruned_loss=0.1249, over 5675866.70 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3652, pruned_loss=0.1169, over 5687346.38 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3822, pruned_loss=0.1257, over 5673550.93 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:15:58,922 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-08 23:16:06,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2635, 1.8281, 1.3842, 0.4533], device='cuda:1'), covar=tensor([0.3726, 0.2704, 0.4499, 0.4855], device='cuda:1'), in_proj_covar=tensor([0.1684, 0.1600, 0.1560, 0.1369], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 23:16:07,088 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-08 23:16:21,730 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4063, 3.0820, 1.5283, 1.5558], device='cuda:1'), covar=tensor([0.0900, 0.0389, 0.0853, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0543, 0.0369, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 23:16:52,725 INFO [train.py:968] (1/2) Epoch 17, batch 27050, libri_loss[loss=0.3447, simple_loss=0.3844, pruned_loss=0.1525, over 29583.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.383, pruned_loss=0.1281, over 5680106.89 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.366, pruned_loss=0.1176, over 5695252.40 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3842, pruned_loss=0.1284, over 5670737.71 frames. ], batch size: 74, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:16:53,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2558, 1.5574, 1.4991, 1.3115], device='cuda:1'), covar=tensor([0.1770, 0.1619, 0.2345, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0740, 0.0699, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 23:17:03,675 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=757940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:06,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=757943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:20,815 INFO [optim.py:369] (1/2) 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 23:17:30,494 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=757964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:32,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=757967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:36,903 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=757972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:17:43,640 INFO [train.py:968] (1/2) Epoch 17, batch 27100, giga_loss[loss=0.3938, simple_loss=0.4151, pruned_loss=0.1863, over 23303.00 frames. ], tot_loss[loss=0.3202, simple_loss=0.3825, pruned_loss=0.129, over 5650094.18 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3655, pruned_loss=0.1173, over 5691125.70 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3843, pruned_loss=0.1298, over 5645647.77 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:18:00,442 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=757996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:18:19,735 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=758013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:18:22,270 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 27150, giga_loss[loss=0.3481, simple_loss=0.3922, pruned_loss=0.152, over 26685.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3807, pruned_loss=0.1274, over 5656766.68 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3656, pruned_loss=0.1174, over 5690475.08 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.382, pruned_loss=0.1281, over 5653708.36 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:18:53,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=758045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:19:04,544 INFO [optim.py:369] (1/2) 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 23:19:25,177 INFO [train.py:968] (1/2) Epoch 17, batch 27200, giga_loss[loss=0.3088, simple_loss=0.3801, pruned_loss=0.1188, over 28895.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3795, pruned_loss=0.1255, over 5656195.65 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3653, pruned_loss=0.1172, over 5693655.92 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.381, pruned_loss=0.1263, over 5650366.19 frames. ], batch size: 186, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:19:45,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8033, 2.3238, 2.1127, 1.5185], device='cuda:1'), covar=tensor([0.3111, 0.2002, 0.2167, 0.2919], device='cuda:1'), in_proj_covar=tensor([0.1872, 0.1810, 0.1738, 0.1880], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-08 23:20:06,057 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 27250, giga_loss[loss=0.2587, simple_loss=0.3441, pruned_loss=0.08668, over 28720.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3783, pruned_loss=0.1232, over 5662318.67 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3649, pruned_loss=0.1171, over 5694706.60 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3802, pruned_loss=0.1241, over 5655626.47 frames. ], batch size: 66, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:20:38,774 INFO [optim.py:369] (1/2) 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 23:20:59,315 INFO [train.py:968] (1/2) Epoch 17, batch 27300, giga_loss[loss=0.2757, simple_loss=0.355, pruned_loss=0.09826, over 28626.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.378, pruned_loss=0.1229, over 5667459.96 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3647, pruned_loss=0.1171, over 5700549.00 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3802, pruned_loss=0.1238, over 5655970.42 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:21:55,297 INFO [train.py:968] (1/2) Epoch 17, batch 27350, giga_loss[loss=0.2947, simple_loss=0.3739, pruned_loss=0.1077, over 28907.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3784, pruned_loss=0.1236, over 5671619.39 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3645, pruned_loss=0.1171, over 5701535.54 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3803, pruned_loss=0.1244, over 5661521.43 frames. ], batch size: 174, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:22:12,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2637, 4.1055, 3.8661, 2.0186], device='cuda:1'), covar=tensor([0.0651, 0.0779, 0.0811, 0.1924], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.1107, 0.0950, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 23:22:20,229 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=758266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:22:30,629 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 27400, giga_loss[loss=0.2567, simple_loss=0.3358, pruned_loss=0.08876, over 28235.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3769, pruned_loss=0.123, over 5670712.08 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3649, pruned_loss=0.1173, over 5694466.73 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3784, pruned_loss=0.1235, over 5667821.02 frames. ], batch size: 77, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:22:59,689 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=758298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:23:09,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-08 23:23:26,164 INFO [train.py:968] (1/2) Epoch 17, batch 27450, giga_loss[loss=0.2494, simple_loss=0.3215, pruned_loss=0.08866, over 28903.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3759, pruned_loss=0.1242, over 5658589.77 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3656, pruned_loss=0.1176, over 5702016.75 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.377, pruned_loss=0.1246, over 5648473.32 frames. ], batch size: 99, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:23:57,749 INFO [optim.py:369] (1/2) 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 23:24:18,523 INFO [train.py:968] (1/2) Epoch 17, batch 27500, giga_loss[loss=0.2736, simple_loss=0.3469, pruned_loss=0.1002, over 29008.00 frames. ], tot_loss[loss=0.312, simple_loss=0.375, pruned_loss=0.1245, over 5642263.67 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3659, pruned_loss=0.118, over 5695318.52 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3757, pruned_loss=0.1247, over 5639664.23 frames. ], batch size: 136, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:24:24,075 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4404, 1.7401, 1.4258, 1.5926], device='cuda:1'), covar=tensor([0.2556, 0.2537, 0.2880, 0.2225], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1039, 0.1270, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 23:25:11,959 INFO [train.py:968] (1/2) Epoch 17, batch 27550, giga_loss[loss=0.3622, simple_loss=0.3966, pruned_loss=0.1639, over 26539.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3742, pruned_loss=0.125, over 5651156.71 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1179, over 5696420.92 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.375, pruned_loss=0.1252, over 5648019.86 frames. ], batch size: 555, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:25:42,202 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 27600, giga_loss[loss=0.2744, simple_loss=0.3499, pruned_loss=0.09946, over 28952.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3733, pruned_loss=0.125, over 5648358.59 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3658, pruned_loss=0.1179, over 5698491.36 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.374, pruned_loss=0.1252, over 5643509.91 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:26:35,192 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-08 23:26:45,227 INFO [train.py:968] (1/2) Epoch 17, batch 27650, libri_loss[loss=0.3151, simple_loss=0.383, pruned_loss=0.1236, over 29534.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1226, over 5659522.29 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3661, pruned_loss=0.1182, over 5702693.35 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3712, pruned_loss=0.1227, over 5650861.23 frames. ], batch size: 89, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:27:13,683 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 27700, giga_loss[loss=0.279, simple_loss=0.3497, pruned_loss=0.1042, over 28302.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 5661289.17 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3662, pruned_loss=0.1183, over 5703868.92 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3675, pruned_loss=0.1185, over 5651979.64 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:28:21,960 INFO [train.py:968] (1/2) Epoch 17, batch 27750, giga_loss[loss=0.3432, simple_loss=0.4019, pruned_loss=0.1423, over 28793.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3671, pruned_loss=0.1179, over 5652483.80 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3662, pruned_loss=0.1183, over 5693086.91 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3673, pruned_loss=0.1179, over 5654790.53 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:28:55,222 INFO [optim.py:369] (1/2) 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 23:29:15,815 INFO [train.py:968] (1/2) Epoch 17, batch 27800, giga_loss[loss=0.2586, simple_loss=0.3295, pruned_loss=0.09388, over 28919.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.117, over 5647123.45 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5696565.94 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3652, pruned_loss=0.1171, over 5644849.14 frames. ], batch size: 106, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:30:08,077 INFO [train.py:968] (1/2) Epoch 17, batch 27850, giga_loss[loss=0.3139, simple_loss=0.3743, pruned_loss=0.1268, over 28293.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3619, pruned_loss=0.1157, over 5663612.50 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1179, over 5699110.39 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3624, pruned_loss=0.116, over 5658845.27 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:30:11,278 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=758730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:30:39,775 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 27900, giga_loss[loss=0.3199, simple_loss=0.3864, pruned_loss=0.1267, over 29088.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3647, pruned_loss=0.1172, over 5661338.72 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3657, pruned_loss=0.118, over 5692599.67 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.365, pruned_loss=0.1173, over 5663069.20 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:31:46,917 INFO [train.py:968] (1/2) Epoch 17, batch 27950, giga_loss[loss=0.2714, simple_loss=0.3525, pruned_loss=0.09513, over 29005.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3674, pruned_loss=0.1187, over 5651867.41 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3657, pruned_loss=0.1178, over 5694397.47 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3677, pruned_loss=0.1189, over 5651071.58 frames. ], batch size: 164, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:32:14,460 INFO [optim.py:369] (1/2) 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 23:32:24,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-08 23:32:33,549 INFO [train.py:968] (1/2) Epoch 17, batch 28000, giga_loss[loss=0.2645, simple_loss=0.3442, pruned_loss=0.0924, over 28914.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3685, pruned_loss=0.1194, over 5639184.46 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3655, pruned_loss=0.1178, over 5679424.05 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3688, pruned_loss=0.1196, over 5651303.11 frames. ], batch size: 145, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:33:17,044 INFO [train.py:968] (1/2) Epoch 17, batch 28050, giga_loss[loss=0.3091, simple_loss=0.3716, pruned_loss=0.1233, over 28639.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3703, pruned_loss=0.1209, over 5626803.57 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3665, pruned_loss=0.1183, over 5664583.15 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3699, pruned_loss=0.1207, over 5648481.73 frames. ], batch size: 336, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:33:25,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3031, 1.6919, 1.2488, 0.8477], device='cuda:1'), covar=tensor([0.3947, 0.2343, 0.2487, 0.4317], device='cuda:1'), in_proj_covar=tensor([0.1680, 0.1594, 0.1554, 0.1366], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-08 23:33:27,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0044, 1.1307, 3.3214, 2.8930], device='cuda:1'), covar=tensor([0.1630, 0.2662, 0.0485, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0626, 0.0924, 0.0854], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 23:33:28,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 23:33:44,767 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 28100, giga_loss[loss=0.3199, simple_loss=0.3849, pruned_loss=0.1275, over 28921.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3715, pruned_loss=0.122, over 5640865.09 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3669, pruned_loss=0.1187, over 5668051.27 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3709, pruned_loss=0.1216, over 5654257.68 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:34:48,883 INFO [train.py:968] (1/2) Epoch 17, batch 28150, giga_loss[loss=0.3158, simple_loss=0.3842, pruned_loss=0.1237, over 28275.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3727, pruned_loss=0.1225, over 5649261.95 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3668, pruned_loss=0.1187, over 5672667.69 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3723, pruned_loss=0.1222, over 5655107.12 frames. ], batch size: 368, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:35:16,023 INFO [optim.py:369] (1/2) 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 23:35:23,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5665, 1.6463, 1.7426, 1.3311], device='cuda:1'), covar=tensor([0.1825, 0.2585, 0.1513, 0.1737], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0700, 0.0920, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 23:35:32,017 INFO [train.py:968] (1/2) Epoch 17, batch 28200, giga_loss[loss=0.2887, simple_loss=0.3581, pruned_loss=0.1097, over 29073.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3746, pruned_loss=0.1237, over 5649059.24 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3671, pruned_loss=0.1191, over 5664334.20 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3742, pruned_loss=0.1232, over 5660158.13 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:35:53,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3389, 1.4726, 1.2665, 1.4903], device='cuda:1'), covar=tensor([0.0703, 0.0398, 0.0325, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 23:36:01,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=759105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:36:22,329 INFO [train.py:968] (1/2) Epoch 17, batch 28250, giga_loss[loss=0.2939, simple_loss=0.3658, pruned_loss=0.111, over 28802.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3759, pruned_loss=0.1253, over 5646810.35 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1195, over 5672679.13 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3755, pruned_loss=0.1248, over 5647325.90 frames. ], batch size: 284, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:36:49,806 INFO [optim.py:369] (1/2) 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 23:37:03,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4531, 1.6218, 1.5220, 1.3701], device='cuda:1'), covar=tensor([0.1560, 0.1777, 0.2238, 0.1901], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0741, 0.0699, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 23:37:09,472 INFO [train.py:968] (1/2) Epoch 17, batch 28300, giga_loss[loss=0.3183, simple_loss=0.3736, pruned_loss=0.1314, over 28689.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3755, pruned_loss=0.1254, over 5639343.92 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1194, over 5667539.39 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3754, pruned_loss=0.1252, over 5642842.00 frames. ], batch size: 92, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:37:24,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-08 23:37:55,205 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-08 23:37:58,160 INFO [train.py:968] (1/2) Epoch 17, batch 28350, libri_loss[loss=0.2382, simple_loss=0.3108, pruned_loss=0.08282, over 29678.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3757, pruned_loss=0.1236, over 5658783.68 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1192, over 5677730.20 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3765, pruned_loss=0.1239, over 5651348.78 frames. ], batch size: 73, lr: 1.88e-03, grad_scale: 2.0
+2023-03-08 23:38:18,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=759248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:38:22,836 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:38:51,211 INFO [train.py:968] (1/2) Epoch 17, batch 28400, giga_loss[loss=0.296, simple_loss=0.3568, pruned_loss=0.1175, over 28611.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3769, pruned_loss=0.1244, over 5663269.33 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3672, pruned_loss=0.1193, over 5676936.00 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3776, pruned_loss=0.1246, over 5658046.96 frames. ], batch size: 85, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:38:53,059 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=759280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:39:35,754 INFO [train.py:968] (1/2) Epoch 17, batch 28450, giga_loss[loss=0.2947, simple_loss=0.3633, pruned_loss=0.1131, over 28807.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3743, pruned_loss=0.1235, over 5674855.95 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3661, pruned_loss=0.1187, over 5686403.47 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3763, pruned_loss=0.1245, over 5661344.09 frames. ], batch size: 119, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:40:12,099 INFO [optim.py:369] (1/2) 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 23:40:32,137 INFO [train.py:968] (1/2) Epoch 17, batch 28500, giga_loss[loss=0.2865, simple_loss=0.3501, pruned_loss=0.1115, over 28938.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3738, pruned_loss=0.124, over 5678619.96 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3656, pruned_loss=0.1183, over 5690612.58 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.376, pruned_loss=0.1253, over 5663731.44 frames. ], batch size: 112, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:40:51,549 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-08 23:41:29,461 INFO [train.py:968] (1/2) Epoch 17, batch 28550, giga_loss[loss=0.2857, simple_loss=0.3563, pruned_loss=0.1075, over 28667.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3711, pruned_loss=0.1222, over 5684593.32 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.1181, over 5695755.26 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3732, pruned_loss=0.1235, over 5667598.50 frames. ], batch size: 262, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:41:35,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3019, 1.5417, 1.5838, 1.3196], device='cuda:1'), covar=tensor([0.1358, 0.1220, 0.1677, 0.1408], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0744, 0.0701, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-08 23:41:59,354 INFO [optim.py:369] (1/2) 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 23:42:13,691 INFO [train.py:968] (1/2) Epoch 17, batch 28600, giga_loss[loss=0.3186, simple_loss=0.3821, pruned_loss=0.1276, over 29076.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1226, over 5683077.25 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3654, pruned_loss=0.118, over 5693677.78 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 5670120.42 frames. ], batch size: 155, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:43:01,741 INFO [train.py:968] (1/2) Epoch 17, batch 28650, giga_loss[loss=0.2683, simple_loss=0.342, pruned_loss=0.09735, over 28957.00 frames. ], tot_loss[loss=0.31, simple_loss=0.372, pruned_loss=0.1239, over 5662752.19 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.1181, over 5688543.70 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3737, pruned_loss=0.1251, over 5656702.52 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:43:34,463 INFO [optim.py:369] (1/2) 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 23:43:50,648 INFO [train.py:968] (1/2) Epoch 17, batch 28700, giga_loss[loss=0.3203, simple_loss=0.3605, pruned_loss=0.14, over 23490.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3718, pruned_loss=0.1238, over 5656910.36 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3654, pruned_loss=0.1181, over 5684166.80 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3734, pruned_loss=0.1249, over 5655493.76 frames. ], batch size: 705, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:43:55,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5117, 1.5801, 1.2375, 1.2273], device='cuda:1'), covar=tensor([0.0890, 0.0564, 0.1086, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0447, 0.0514, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 23:44:32,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0456, 1.0402, 3.3548, 2.9493], device='cuda:1'), covar=tensor([0.1685, 0.2730, 0.0542, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0625, 0.0925, 0.0855], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-08 23:44:40,693 INFO [train.py:968] (1/2) Epoch 17, batch 28750, giga_loss[loss=0.2899, simple_loss=0.3589, pruned_loss=0.1105, over 28814.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3739, pruned_loss=0.1259, over 5658284.59 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3653, pruned_loss=0.1179, over 5687366.72 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3754, pruned_loss=0.127, over 5653632.13 frames. ], batch size: 199, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:45:12,070 INFO [optim.py:369] (1/2) 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 23:45:19,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-08 23:45:33,831 INFO [train.py:968] (1/2) Epoch 17, batch 28800, giga_loss[loss=0.3163, simple_loss=0.3768, pruned_loss=0.1279, over 28909.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3753, pruned_loss=0.1272, over 5649637.16 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3652, pruned_loss=0.1179, over 5688526.89 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5644846.86 frames. ], batch size: 227, lr: 1.88e-03, grad_scale: 8.0
+2023-03-08 23:46:15,241 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=759723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:46:19,996 INFO [train.py:968] (1/2) Epoch 17, batch 28850, giga_loss[loss=0.359, simple_loss=0.4022, pruned_loss=0.1579, over 27726.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3748, pruned_loss=0.1274, over 5639228.52 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5681397.43 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3756, pruned_loss=0.1281, over 5641317.53 frames. ], batch size: 472, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:46:21,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5964, 1.5746, 1.8569, 1.4457], device='cuda:1'), covar=tensor([0.1261, 0.1854, 0.1026, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0870, 0.0700, 0.0919, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-08 23:46:49,655 INFO [optim.py:369] (1/2) 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 23:47:04,444 INFO [train.py:968] (1/2) Epoch 17, batch 28900, giga_loss[loss=0.3085, simple_loss=0.3786, pruned_loss=0.1192, over 28857.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3752, pruned_loss=0.128, over 5631195.52 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1184, over 5665428.14 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3761, pruned_loss=0.1286, over 5647168.26 frames. ], batch size: 174, lr: 1.88e-03, grad_scale: 4.0
+2023-03-08 23:47:46,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7052, 1.9141, 1.6023, 1.8509], device='cuda:1'), covar=tensor([0.2034, 0.1985, 0.2004, 0.1904], device='cuda:1'), in_proj_covar=tensor([0.1439, 0.1044, 0.1274, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-08 23:47:54,148 INFO [train.py:968] (1/2) Epoch 17, batch 28950, giga_loss[loss=0.2993, simple_loss=0.369, pruned_loss=0.1148, over 28615.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3755, pruned_loss=0.1281, over 5628063.93 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3658, pruned_loss=0.1184, over 5666644.00 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3762, pruned_loss=0.1286, over 5639404.17 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:48:30,750 INFO [optim.py:369] (1/2) 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 23:48:43,327 INFO [train.py:968] (1/2) Epoch 17, batch 29000, giga_loss[loss=0.3076, simple_loss=0.3715, pruned_loss=0.1219, over 28587.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3749, pruned_loss=0.1268, over 5640167.12 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 5671688.95 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5643858.50 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:49:29,505 INFO [train.py:968] (1/2) Epoch 17, batch 29050, giga_loss[loss=0.2971, simple_loss=0.3609, pruned_loss=0.1166, over 29050.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3755, pruned_loss=0.1268, over 5641805.24 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3656, pruned_loss=0.1185, over 5668030.67 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3764, pruned_loss=0.1274, over 5647599.87 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:49:29,709 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=759929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:50:00,407 INFO [optim.py:369] (1/2) 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 23:50:13,865 INFO [train.py:968] (1/2) Epoch 17, batch 29100, giga_loss[loss=0.3361, simple_loss=0.3893, pruned_loss=0.1415, over 28670.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3769, pruned_loss=0.128, over 5652334.61 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3656, pruned_loss=0.1184, over 5667411.39 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3779, pruned_loss=0.1289, over 5657730.68 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:50:51,265 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-08 23:50:55,697 INFO [train.py:968] (1/2) Epoch 17, batch 29150, giga_loss[loss=0.2777, simple_loss=0.3525, pruned_loss=0.1015, over 28790.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3781, pruned_loss=0.129, over 5658140.98 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1184, over 5664299.67 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3795, pruned_loss=0.1302, over 5665048.63 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:51:24,627 INFO [optim.py:369] (1/2) 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 23:51:36,945 INFO [train.py:968] (1/2) Epoch 17, batch 29200, giga_loss[loss=0.2916, simple_loss=0.3625, pruned_loss=0.1104, over 28853.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3774, pruned_loss=0.1283, over 5660729.30 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3647, pruned_loss=0.1176, over 5665372.14 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3801, pruned_loss=0.1305, over 5664103.74 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 8.0
+2023-03-08 23:51:57,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:52:20,490 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-08 23:52:26,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3060, 1.6812, 1.3383, 1.5253], device='cuda:1'), covar=tensor([0.0738, 0.0346, 0.0328, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 23:52:30,859 INFO [train.py:968] (1/2) Epoch 17, batch 29250, giga_loss[loss=0.2921, simple_loss=0.3694, pruned_loss=0.1074, over 28985.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3782, pruned_loss=0.1278, over 5663172.74 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3644, pruned_loss=0.1175, over 5667796.02 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3807, pruned_loss=0.1297, over 5663614.96 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:52:52,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-08 23:53:07,491 INFO [optim.py:369] (1/2) 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 23:53:16,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5976, 4.8212, 1.8720, 1.8178], device='cuda:1'), covar=tensor([0.1004, 0.0276, 0.0854, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0544, 0.0370, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-08 23:53:18,912 INFO [train.py:968] (1/2) Epoch 17, batch 29300, giga_loss[loss=0.3164, simple_loss=0.3813, pruned_loss=0.1258, over 28269.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3773, pruned_loss=0.1264, over 5654147.90 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3643, pruned_loss=0.1176, over 5662810.31 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3796, pruned_loss=0.128, over 5658394.47 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:53:46,021 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:08,187 INFO [train.py:968] (1/2) Epoch 17, batch 29350, giga_loss[loss=0.3021, simple_loss=0.3718, pruned_loss=0.1162, over 28533.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1246, over 5653006.43 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3645, pruned_loss=0.1178, over 5662360.23 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3765, pruned_loss=0.1259, over 5656962.35 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:54:17,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:20,160 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=760244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:37,204 INFO [optim.py:369] (1/2) 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 23:54:46,271 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:54:49,713 INFO [train.py:968] (1/2) Epoch 17, batch 29400, giga_loss[loss=0.2944, simple_loss=0.369, pruned_loss=0.1099, over 28604.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3752, pruned_loss=0.1252, over 5644781.07 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.365, pruned_loss=0.1183, over 5654327.93 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3764, pruned_loss=0.126, over 5654878.20 frames. ], batch size: 85, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:55:13,943 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:55:38,359 INFO [train.py:968] (1/2) Epoch 17, batch 29450, giga_loss[loss=0.2752, simple_loss=0.3535, pruned_loss=0.09843, over 28624.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3769, pruned_loss=0.1262, over 5656226.51 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3646, pruned_loss=0.1179, over 5661399.50 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3786, pruned_loss=0.1272, over 5657945.40 frames. ], batch size: 60, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:56:06,252 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2654, 3.0697, 1.4651, 1.4002], device='cuda:1'), covar=tensor([0.1053, 0.0387, 0.0908, 0.1457], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0547, 0.0372, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-08 23:56:14,542 INFO [optim.py:369] (1/2) 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 23:56:26,635 INFO [train.py:968] (1/2) Epoch 17, batch 29500, giga_loss[loss=0.3282, simple_loss=0.3818, pruned_loss=0.1373, over 28613.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3777, pruned_loss=0.1275, over 5643374.96 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3645, pruned_loss=0.1178, over 5655679.66 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3795, pruned_loss=0.1287, over 5650089.76 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:57:08,300 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 29550, giga_loss[loss=0.3102, simple_loss=0.3751, pruned_loss=0.1227, over 28849.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3764, pruned_loss=0.1271, over 5653445.31 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3643, pruned_loss=0.1177, over 5660917.07 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3782, pruned_loss=0.1283, over 5654110.78 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:57:28,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6262, 4.4702, 4.2063, 2.1586], device='cuda:1'), covar=tensor([0.0512, 0.0669, 0.0676, 0.1865], device='cuda:1'), in_proj_covar=tensor([0.1199, 0.1105, 0.0951, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-08 23:57:29,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2945, 1.7460, 1.5048, 1.5867], device='cuda:1'), covar=tensor([0.0780, 0.0311, 0.0310, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0092, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-08 23:57:31,823 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:57:35,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=760450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:57:48,747 INFO [optim.py:369] (1/2) 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 23:58:01,304 INFO [train.py:968] (1/2) Epoch 17, batch 29600, libri_loss[loss=0.3306, simple_loss=0.3998, pruned_loss=0.1307, over 29654.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3778, pruned_loss=0.1282, over 5642483.57 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3649, pruned_loss=0.118, over 5653345.84 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.379, pruned_loss=0.1292, over 5649814.01 frames. ], batch size: 88, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:58:01,586 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:58:49,885 INFO [train.py:968] (1/2) Epoch 17, batch 29650, giga_loss[loss=0.3257, simple_loss=0.3822, pruned_loss=0.1346, over 28929.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3788, pruned_loss=0.1287, over 5641722.43 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3652, pruned_loss=0.1182, over 5645872.42 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3795, pruned_loss=0.1294, over 5653912.06 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 4.0
+2023-03-08 23:59:28,636 INFO [optim.py:369] (1/2) 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 23:59:33,923 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-08 23:59:39,815 INFO [train.py:968] (1/2) Epoch 17, batch 29700, giga_loss[loss=0.2607, simple_loss=0.3425, pruned_loss=0.08946, over 28947.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3796, pruned_loss=0.1293, over 5642373.45 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3653, pruned_loss=0.1183, over 5648581.89 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3804, pruned_loss=0.13, over 5649431.34 frames. ], batch size: 112, lr: 1.87e-03, grad_scale: 2.0
+2023-03-08 23:59:43,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-08 23:59:56,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3293, 1.1983, 3.7382, 3.1604], device='cuda:1'), covar=tensor([0.1552, 0.2704, 0.0516, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0630, 0.0936, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 00:00:16,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7166, 3.5449, 3.3335, 1.7136], device='cuda:1'), covar=tensor([0.0753, 0.0855, 0.0810, 0.2266], device='cuda:1'), in_proj_covar=tensor([0.1206, 0.1113, 0.0957, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 00:00:26,652 INFO [train.py:968] (1/2) Epoch 17, batch 29750, giga_loss[loss=0.3174, simple_loss=0.3733, pruned_loss=0.1307, over 27565.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3773, pruned_loss=0.1268, over 5664635.85 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3646, pruned_loss=0.1178, over 5654552.40 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3788, pruned_loss=0.128, over 5665116.30 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:00:38,182 INFO [scaling.py:679] (1/2) 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] (1/2) 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 00:01:11,457 INFO [train.py:968] (1/2) Epoch 17, batch 29800, giga_loss[loss=0.2858, simple_loss=0.3553, pruned_loss=0.1081, over 29006.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3768, pruned_loss=0.1264, over 5656509.92 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3646, pruned_loss=0.1177, over 5653835.89 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3786, pruned_loss=0.1277, over 5658054.14 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:01:17,388 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-09 00:01:35,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-09 00:01:53,330 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:01:55,299 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 29850, giga_loss[loss=0.3231, simple_loss=0.3844, pruned_loss=0.1309, over 28000.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3768, pruned_loss=0.1261, over 5669651.67 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3648, pruned_loss=0.1179, over 5664111.83 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3787, pruned_loss=0.1274, over 5661477.19 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:02:21,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:02:28,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1597, 1.2573, 1.1324, 0.7904], device='cuda:1'), covar=tensor([0.0934, 0.0519, 0.1035, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0447, 0.0514, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 00:02:29,011 INFO [optim.py:369] (1/2) 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 00:02:40,826 INFO [train.py:968] (1/2) Epoch 17, batch 29900, giga_loss[loss=0.297, simple_loss=0.3669, pruned_loss=0.1135, over 28974.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1249, over 5653781.71 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3648, pruned_loss=0.1178, over 5650962.42 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3767, pruned_loss=0.1263, over 5660366.60 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:02:58,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=760796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:03:06,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1566, 1.2477, 1.0937, 0.8006], device='cuda:1'), covar=tensor([0.0985, 0.0563, 0.1125, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0447, 0.0514, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 00:03:28,292 INFO [train.py:968] (1/2) Epoch 17, batch 29950, giga_loss[loss=0.2825, simple_loss=0.349, pruned_loss=0.108, over 28886.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3744, pruned_loss=0.1254, over 5654229.61 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3644, pruned_loss=0.1175, over 5654543.73 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3764, pruned_loss=0.1269, over 5656411.60 frames. ], batch size: 145, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:03:53,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:03:59,460 INFO [optim.py:369] (1/2) 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 00:04:09,625 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=760875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:04:12,140 INFO [train.py:968] (1/2) Epoch 17, batch 30000, giga_loss[loss=0.2554, simple_loss=0.3337, pruned_loss=0.08855, over 29029.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3712, pruned_loss=0.1232, over 5648993.81 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1173, over 5649163.47 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3731, pruned_loss=0.1247, over 5655306.45 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:04:12,140 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 00:04:17,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1972, 1.5975, 1.4836, 1.0614], device='cuda:1'), covar=tensor([0.1885, 0.2796, 0.1592, 0.1982], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0700, 0.0919, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 00:04:20,411 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 00:04:36,566 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 30050, libri_loss[loss=0.3581, simple_loss=0.4043, pruned_loss=0.156, over 29540.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3678, pruned_loss=0.1215, over 5660601.03 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.364, pruned_loss=0.1173, over 5644908.71 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3695, pruned_loss=0.1228, over 5669602.39 frames. ], batch size: 80, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:05:18,634 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=760939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:05:20,594 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=760942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:05:43,302 INFO [optim.py:369] (1/2) 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 00:05:49,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=760971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:05:49,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7623, 4.6120, 4.3910, 2.0729], device='cuda:1'), covar=tensor([0.0548, 0.0705, 0.0731, 0.1973], device='cuda:1'), in_proj_covar=tensor([0.1199, 0.1108, 0.0951, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 00:05:54,631 INFO [train.py:968] (1/2) Epoch 17, batch 30100, libri_loss[loss=0.2366, simple_loss=0.3097, pruned_loss=0.08181, over 29515.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3654, pruned_loss=0.1203, over 5680608.09 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3639, pruned_loss=0.1171, over 5653506.99 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.367, pruned_loss=0.1216, over 5680901.81 frames. ], batch size: 70, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:06:46,567 INFO [train.py:968] (1/2) Epoch 17, batch 30150, libri_loss[loss=0.3919, simple_loss=0.4253, pruned_loss=0.1792, over 25874.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3657, pruned_loss=0.1204, over 5683841.50 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3646, pruned_loss=0.1177, over 5654769.28 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3664, pruned_loss=0.1209, over 5683549.65 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:07:01,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1700, 1.3052, 1.1566, 0.9469], device='cuda:1'), covar=tensor([0.0964, 0.0520, 0.1058, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0448, 0.0514, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 00:07:23,817 INFO [optim.py:369] (1/2) 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 00:07:37,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:07:39,988 INFO [train.py:968] (1/2) Epoch 17, batch 30200, libri_loss[loss=0.3535, simple_loss=0.384, pruned_loss=0.1615, over 29675.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3659, pruned_loss=0.1189, over 5679215.36 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3647, pruned_loss=0.118, over 5656170.43 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3663, pruned_loss=0.1191, over 5677701.72 frames. ], batch size: 73, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:07:59,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6103, 1.8358, 1.4438, 1.6054], device='cuda:1'), covar=tensor([0.2692, 0.2668, 0.3171, 0.2338], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1043, 0.1276, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 00:08:35,190 INFO [train.py:968] (1/2) Epoch 17, batch 30250, giga_loss[loss=0.2721, simple_loss=0.3482, pruned_loss=0.09804, over 29071.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3634, pruned_loss=0.1156, over 5667159.14 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3646, pruned_loss=0.1181, over 5657150.86 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3639, pruned_loss=0.1157, over 5665666.18 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:09:00,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5229, 2.3136, 1.6259, 0.6741], device='cuda:1'), covar=tensor([0.5233, 0.2548, 0.3766, 0.5721], device='cuda:1'), in_proj_covar=tensor([0.1684, 0.1601, 0.1562, 0.1375], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 00:09:04,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:09:11,048 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 30300, giga_loss[loss=0.2482, simple_loss=0.3317, pruned_loss=0.08235, over 28267.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3602, pruned_loss=0.1121, over 5666417.74 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3646, pruned_loss=0.1181, over 5661842.04 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3605, pruned_loss=0.112, over 5661184.93 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:10:16,147 INFO [train.py:968] (1/2) Epoch 17, batch 30350, giga_loss[loss=0.2253, simple_loss=0.315, pruned_loss=0.06773, over 28862.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.3558, pruned_loss=0.1081, over 5651147.74 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3643, pruned_loss=0.118, over 5654098.38 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3562, pruned_loss=0.1081, over 5654113.71 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 00:10:23,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:10:40,878 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:10:55,992 INFO [optim.py:369] (1/2) 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 00:10:56,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:11:06,249 INFO [train.py:968] (1/2) Epoch 17, batch 30400, giga_loss[loss=0.2697, simple_loss=0.348, pruned_loss=0.09568, over 27714.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3544, pruned_loss=0.1056, over 5645486.86 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3646, pruned_loss=0.1183, over 5647390.56 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3543, pruned_loss=0.1052, over 5654401.41 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:11:57,454 INFO [train.py:968] (1/2) Epoch 17, batch 30450, libri_loss[loss=0.2769, simple_loss=0.3417, pruned_loss=0.106, over 26046.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3533, pruned_loss=0.1043, over 5617712.12 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3641, pruned_loss=0.1182, over 5635935.41 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3533, pruned_loss=0.1035, over 5634282.88 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:12:25,421 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:12:52,883 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:12:53,877 INFO [train.py:968] (1/2) Epoch 17, batch 30500, giga_loss[loss=0.2865, simple_loss=0.3482, pruned_loss=0.1124, over 26657.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3546, pruned_loss=0.1051, over 5626600.40 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3639, pruned_loss=0.1182, over 5637135.73 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3546, pruned_loss=0.1044, over 5638451.38 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:12:55,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:09,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:12,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:25,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:27,618 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:31,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:43,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:13:47,220 INFO [train.py:968] (1/2) Epoch 17, batch 30550, giga_loss[loss=0.2524, simple_loss=0.334, pruned_loss=0.0854, over 28645.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3515, pruned_loss=0.1031, over 5621934.34 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3636, pruned_loss=0.118, over 5641203.46 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3517, pruned_loss=0.1025, over 5627520.82 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:13:57,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:14:06,918 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:14:30,565 INFO [train.py:968] (1/2) Epoch 17, batch 30600, giga_loss[loss=0.3148, simple_loss=0.3851, pruned_loss=0.1222, over 28860.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3501, pruned_loss=0.1027, over 5621430.25 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3635, pruned_loss=0.1184, over 5629634.79 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3498, pruned_loss=0.1011, over 5636157.22 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:15:16,292 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761526.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:15:19,152 INFO [train.py:968] (1/2) Epoch 17, batch 30650, giga_loss[loss=0.2267, simple_loss=0.309, pruned_loss=0.07216, over 28555.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3482, pruned_loss=0.1015, over 5632817.32 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3628, pruned_loss=0.1181, over 5638479.36 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.348, pruned_loss=0.0998, over 5636045.51 frames. ], batch size: 60, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:15:24,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:15:44,358 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:15:55,302 INFO [optim.py:369] (1/2) 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 00:16:08,869 INFO [train.py:968] (1/2) Epoch 17, batch 30700, giga_loss[loss=0.2698, simple_loss=0.3503, pruned_loss=0.09462, over 28959.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3484, pruned_loss=0.1008, over 5628399.72 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3628, pruned_loss=0.1182, over 5631042.96 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3481, pruned_loss=0.09929, over 5637931.13 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:16:23,740 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:16:26,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:16:55,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:16:56,536 INFO [train.py:968] (1/2) Epoch 17, batch 30750, giga_loss[loss=0.3494, simple_loss=0.4097, pruned_loss=0.1446, over 28672.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3457, pruned_loss=0.09817, over 5644451.34 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3624, pruned_loss=0.1181, over 5636683.12 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3456, pruned_loss=0.09669, over 5647001.74 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:17:25,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-09 00:17:37,826 INFO [optim.py:369] (1/2) 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 00:17:38,007 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:17:47,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 30800, giga_loss[loss=0.2659, simple_loss=0.3388, pruned_loss=0.09654, over 28870.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3432, pruned_loss=0.09635, over 5640557.82 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3618, pruned_loss=0.1177, over 5638849.50 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3432, pruned_loss=0.09498, over 5641068.89 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:17:51,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:18:10,623 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 00:18:21,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:18:39,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:18:39,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-09 00:18:39,736 INFO [train.py:968] (1/2) Epoch 17, batch 30850, giga_loss[loss=0.2432, simple_loss=0.3337, pruned_loss=0.07638, over 28986.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3386, pruned_loss=0.09385, over 5641624.30 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3612, pruned_loss=0.1174, over 5643384.21 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3387, pruned_loss=0.09259, over 5637740.29 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:19:11,711 INFO [optim.py:369] (1/2) 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 00:19:23,390 INFO [train.py:968] (1/2) Epoch 17, batch 30900, giga_loss[loss=0.2398, simple_loss=0.3226, pruned_loss=0.07849, over 28839.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3382, pruned_loss=0.0944, over 5648863.74 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3604, pruned_loss=0.1171, over 5651470.79 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3381, pruned_loss=0.09273, over 5638552.52 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:19:51,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-09 00:20:17,325 INFO [train.py:968] (1/2) Epoch 17, batch 30950, giga_loss[loss=0.2718, simple_loss=0.3376, pruned_loss=0.103, over 26624.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.338, pruned_loss=0.09521, over 5637317.10 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3599, pruned_loss=0.1169, over 5655832.16 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.338, pruned_loss=0.09364, over 5624996.27 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:20:18,837 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=761830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:20:35,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 00:20:59,073 INFO [optim.py:369] (1/2) 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 00:21:03,884 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=761871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:21:06,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=761874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:21:14,345 INFO [train.py:968] (1/2) Epoch 17, batch 31000, giga_loss[loss=0.2537, simple_loss=0.3466, pruned_loss=0.08036, over 28788.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3402, pruned_loss=0.09563, over 5636680.35 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3604, pruned_loss=0.1173, over 5659220.97 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3395, pruned_loss=0.09371, over 5623635.79 frames. ], batch size: 284, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:21:35,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4439, 1.7037, 1.3383, 1.7068], device='cuda:1'), covar=tensor([0.2658, 0.2552, 0.2996, 0.2322], device='cuda:1'), in_proj_covar=tensor([0.1433, 0.1038, 0.1278, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 00:21:42,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761901.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:21:44,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=761903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:22:15,247 INFO [train.py:968] (1/2) Epoch 17, batch 31050, giga_loss[loss=0.2318, simple_loss=0.3192, pruned_loss=0.07225, over 28887.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3419, pruned_loss=0.0949, over 5648046.96 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3602, pruned_loss=0.1172, over 5661814.00 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3413, pruned_loss=0.09327, over 5635174.69 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:22:15,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=761929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:22:25,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-09 00:23:05,987 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 31100, giga_loss[loss=0.2961, simple_loss=0.3629, pruned_loss=0.1146, over 28999.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3426, pruned_loss=0.0948, over 5666714.62 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3599, pruned_loss=0.1172, over 5664677.18 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.342, pruned_loss=0.09316, over 5654120.72 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:24:20,323 INFO [train.py:968] (1/2) Epoch 17, batch 31150, giga_loss[loss=0.2284, simple_loss=0.3153, pruned_loss=0.07076, over 28666.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3401, pruned_loss=0.09388, over 5670690.84 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3596, pruned_loss=0.1172, over 5670566.92 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3395, pruned_loss=0.09188, over 5655560.38 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:24:35,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:24:36,910 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762047.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:25:02,754 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=762072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:25:14,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:25:14,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762076.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:25:17,310 INFO [train.py:968] (1/2) Epoch 17, batch 31200, giga_loss[loss=0.2457, simple_loss=0.3333, pruned_loss=0.079, over 28920.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3394, pruned_loss=0.0928, over 5663227.13 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.359, pruned_loss=0.1168, over 5666189.91 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3387, pruned_loss=0.09065, over 5654318.59 frames. ], batch size: 284, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:25:51,468 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:25:59,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5182, 4.3381, 4.0786, 2.1057], device='cuda:1'), covar=tensor([0.0544, 0.0709, 0.0729, 0.1898], device='cuda:1'), in_proj_covar=tensor([0.1163, 0.1074, 0.0917, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 00:26:17,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4864, 2.1670, 1.4326, 0.6689], device='cuda:1'), covar=tensor([0.5554, 0.2705, 0.4297, 0.6327], device='cuda:1'), in_proj_covar=tensor([0.1684, 0.1596, 0.1561, 0.1378], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 00:26:22,366 INFO [train.py:968] (1/2) Epoch 17, batch 31250, giga_loss[loss=0.2297, simple_loss=0.296, pruned_loss=0.08173, over 24819.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3368, pruned_loss=0.09041, over 5661803.73 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.359, pruned_loss=0.1169, over 5668100.38 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.336, pruned_loss=0.08845, over 5653028.28 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:27:05,564 INFO [optim.py:369] (1/2) 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 00:27:15,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 00:27:17,184 INFO [train.py:968] (1/2) Epoch 17, batch 31300, giga_loss[loss=0.2179, simple_loss=0.2852, pruned_loss=0.07536, over 24044.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3336, pruned_loss=0.0898, over 5664445.68 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3584, pruned_loss=0.1165, over 5667959.84 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3325, pruned_loss=0.08741, over 5656949.29 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:27:27,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=762185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:27:30,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:27:35,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1587, 1.2829, 3.3707, 2.9760], device='cuda:1'), covar=tensor([0.1619, 0.2629, 0.0496, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0622, 0.0918, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 00:27:50,501 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:28:02,690 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:28:16,532 INFO [train.py:968] (1/2) Epoch 17, batch 31350, giga_loss[loss=0.251, simple_loss=0.3222, pruned_loss=0.08991, over 28973.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3335, pruned_loss=0.09038, over 5668748.30 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3583, pruned_loss=0.1167, over 5672927.81 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.08756, over 5658028.51 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:28:59,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5217, 1.6692, 1.7896, 1.3439], device='cuda:1'), covar=tensor([0.1881, 0.2659, 0.1509, 0.1921], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0687, 0.0917, 0.0816], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 00:29:05,165 INFO [optim.py:369] (1/2) 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 00:29:13,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5845, 1.9267, 1.7222, 1.6846], device='cuda:1'), covar=tensor([0.1679, 0.2108, 0.1929, 0.1958], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0731, 0.0687, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 00:29:16,343 INFO [train.py:968] (1/2) Epoch 17, batch 31400, giga_loss[loss=0.2569, simple_loss=0.3413, pruned_loss=0.08624, over 28598.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3338, pruned_loss=0.09021, over 5663101.71 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3583, pruned_loss=0.1168, over 5666607.64 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3323, pruned_loss=0.08762, over 5660051.22 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:30:13,192 INFO [train.py:968] (1/2) Epoch 17, batch 31450, giga_loss[loss=0.2497, simple_loss=0.3346, pruned_loss=0.08243, over 28901.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3361, pruned_loss=0.09106, over 5668774.87 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3582, pruned_loss=0.1168, over 5674664.98 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3342, pruned_loss=0.08796, over 5658581.75 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:30:41,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=762348.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:30:44,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=762351.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:31:03,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8762, 3.7320, 3.5162, 1.8072], device='cuda:1'), covar=tensor([0.0674, 0.0750, 0.0796, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.1165, 0.1075, 0.0918, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 00:31:07,781 INFO [optim.py:369] (1/2) 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 00:31:18,731 INFO [train.py:968] (1/2) Epoch 17, batch 31500, giga_loss[loss=0.205, simple_loss=0.2964, pruned_loss=0.05681, over 28974.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3356, pruned_loss=0.09016, over 5668840.77 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3575, pruned_loss=0.1165, over 5675294.14 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3341, pruned_loss=0.08735, over 5660056.53 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:31:21,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=762380.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:32:29,614 INFO [train.py:968] (1/2) Epoch 17, batch 31550, giga_loss[loss=0.2261, simple_loss=0.3272, pruned_loss=0.06247, over 28951.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3326, pruned_loss=0.08803, over 5670618.06 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3574, pruned_loss=0.1164, over 5676483.66 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3314, pruned_loss=0.08579, over 5662760.81 frames. ], batch size: 145, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:33:23,642 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 31600, libri_loss[loss=0.2945, simple_loss=0.3592, pruned_loss=0.1149, over 29206.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3357, pruned_loss=0.09017, over 5659148.02 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3573, pruned_loss=0.1164, over 5668236.90 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3343, pruned_loss=0.08771, over 5660721.28 frames. ], batch size: 97, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:34:02,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4602, 1.7058, 1.4557, 1.3055], device='cuda:1'), covar=tensor([0.2296, 0.2229, 0.2323, 0.2064], device='cuda:1'), in_proj_covar=tensor([0.1433, 0.1038, 0.1274, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 00:34:20,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5661, 1.5492, 1.2479, 1.2191], device='cuda:1'), covar=tensor([0.0765, 0.0420, 0.0841, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0440, 0.0511, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 00:34:36,340 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 00:34:39,148 INFO [train.py:968] (1/2) Epoch 17, batch 31650, giga_loss[loss=0.2518, simple_loss=0.3549, pruned_loss=0.07439, over 29051.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3398, pruned_loss=0.09018, over 5655153.93 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3574, pruned_loss=0.1166, over 5672633.73 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3382, pruned_loss=0.08757, over 5652299.50 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:35:03,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 00:35:28,774 INFO [optim.py:369] (1/2) 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 00:35:41,672 INFO [train.py:968] (1/2) Epoch 17, batch 31700, giga_loss[loss=0.2619, simple_loss=0.3517, pruned_loss=0.08604, over 28732.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3414, pruned_loss=0.08952, over 5662804.21 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.357, pruned_loss=0.1163, over 5676837.93 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3401, pruned_loss=0.08705, over 5656690.92 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:35:51,017 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:36:17,336 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:36:38,480 INFO [train.py:968] (1/2) Epoch 17, batch 31750, giga_loss[loss=0.2752, simple_loss=0.3612, pruned_loss=0.09454, over 28578.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3418, pruned_loss=0.08974, over 5661034.19 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.357, pruned_loss=0.1166, over 5680067.97 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3403, pruned_loss=0.08671, over 5652629.86 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:37:04,777 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:37:29,851 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 31800, giga_loss[loss=0.2499, simple_loss=0.3425, pruned_loss=0.07871, over 28396.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3398, pruned_loss=0.0887, over 5664144.46 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3564, pruned_loss=0.1164, over 5685388.76 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3388, pruned_loss=0.08577, over 5652142.21 frames. ], batch size: 65, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:38:40,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:38:44,361 INFO [train.py:968] (1/2) Epoch 17, batch 31850, giga_loss[loss=0.2442, simple_loss=0.3236, pruned_loss=0.08237, over 29054.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3397, pruned_loss=0.09009, over 5661888.89 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3559, pruned_loss=0.1161, over 5689644.18 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3391, pruned_loss=0.08754, over 5648220.33 frames. ], batch size: 187, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:39:48,368 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:39:48,749 INFO [optim.py:369] (1/2) 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 00:39:58,607 INFO [train.py:968] (1/2) Epoch 17, batch 31900, giga_loss[loss=0.2451, simple_loss=0.3303, pruned_loss=0.07998, over 29023.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3404, pruned_loss=0.09086, over 5674280.53 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3556, pruned_loss=0.116, over 5692339.65 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3398, pruned_loss=0.08848, over 5660523.12 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:40:38,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4426, 3.5454, 1.4853, 1.6401], device='cuda:1'), covar=tensor([0.0933, 0.0391, 0.0898, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0539, 0.0369, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 00:41:13,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-09 00:41:20,454 INFO [train.py:968] (1/2) Epoch 17, batch 31950, giga_loss[loss=0.2729, simple_loss=0.3433, pruned_loss=0.1013, over 26955.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3382, pruned_loss=0.09018, over 5672414.59 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3552, pruned_loss=0.1157, over 5688830.45 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3377, pruned_loss=0.08787, over 5663251.28 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:42:12,709 INFO [optim.py:369] (1/2) 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 00:42:24,199 INFO [train.py:968] (1/2) Epoch 17, batch 32000, giga_loss[loss=0.2089, simple_loss=0.2941, pruned_loss=0.06191, over 28554.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3355, pruned_loss=0.08881, over 5670306.22 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3546, pruned_loss=0.1155, over 5685795.06 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3352, pruned_loss=0.08637, over 5664915.41 frames. ], batch size: 85, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:43:14,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 00:43:28,950 INFO [train.py:968] (1/2) Epoch 17, batch 32050, giga_loss[loss=0.199, simple_loss=0.2845, pruned_loss=0.0568, over 28976.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3336, pruned_loss=0.08811, over 5648840.69 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3549, pruned_loss=0.1158, over 5669613.33 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3324, pruned_loss=0.08513, over 5658577.76 frames. ], batch size: 106, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:43:48,712 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=762946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:43:53,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 00:44:06,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:44:15,319 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 32100, giga_loss[loss=0.2867, simple_loss=0.3553, pruned_loss=0.109, over 26926.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3372, pruned_loss=0.0913, over 5636323.23 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3553, pruned_loss=0.1162, over 5652568.86 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3351, pruned_loss=0.08732, over 5659053.26 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:44:33,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=762984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:45:13,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3929, 1.7121, 1.6098, 1.4971], device='cuda:1'), covar=tensor([0.1773, 0.1964, 0.2086, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0727, 0.0683, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 00:45:21,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:45:25,824 INFO [train.py:968] (1/2) Epoch 17, batch 32150, libri_loss[loss=0.2374, simple_loss=0.305, pruned_loss=0.08495, over 29652.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3403, pruned_loss=0.09212, over 5652442.16 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3549, pruned_loss=0.116, over 5654255.79 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3386, pruned_loss=0.08872, over 5668563.40 frames. ], batch size: 73, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:45:36,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-09 00:46:19,812 INFO [optim.py:369] (1/2) 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 00:46:24,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9680, 1.1944, 1.1130, 0.8443], device='cuda:1'), covar=tensor([0.2365, 0.2258, 0.1271, 0.1977], device='cuda:1'), in_proj_covar=tensor([0.1832, 0.1769, 0.1686, 0.1829], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 00:46:29,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=763078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:46:30,228 INFO [train.py:968] (1/2) Epoch 17, batch 32200, libri_loss[loss=0.3151, simple_loss=0.3681, pruned_loss=0.1311, over 28565.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3386, pruned_loss=0.092, over 5656953.18 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3545, pruned_loss=0.1157, over 5659162.76 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3373, pruned_loss=0.08898, over 5665248.63 frames. ], batch size: 106, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:46:53,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763099.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:47:01,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:04,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:28,357 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 32250, giga_loss[loss=0.3033, simple_loss=0.3622, pruned_loss=0.1222, over 26893.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3395, pruned_loss=0.09389, over 5662305.64 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3542, pruned_loss=0.1156, over 5663712.46 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3383, pruned_loss=0.09106, over 5664683.90 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:47:32,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:39,775 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:47:49,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:48:07,827 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:48:11,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1373, 1.4639, 1.4089, 1.0264], device='cuda:1'), covar=tensor([0.1432, 0.2412, 0.1236, 0.1638], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0686, 0.0913, 0.0811], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 00:48:18,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:48:20,956 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:48:32,090 INFO [train.py:968] (1/2) Epoch 17, batch 32300, giga_loss[loss=0.2944, simple_loss=0.3586, pruned_loss=0.115, over 26865.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3401, pruned_loss=0.09444, over 5665890.98 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3541, pruned_loss=0.1156, over 5670870.71 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3388, pruned_loss=0.09145, over 5661105.47 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:49:01,149 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:49:42,863 INFO [train.py:968] (1/2) Epoch 17, batch 32350, giga_loss[loss=0.2526, simple_loss=0.3244, pruned_loss=0.09037, over 26854.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.09377, over 5663937.91 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.354, pruned_loss=0.1156, over 5673425.85 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3397, pruned_loss=0.09083, over 5657789.58 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:50:01,923 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763242.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 00:50:07,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763245.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:50:45,063 INFO [optim.py:369] (1/2) 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 00:50:49,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763274.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 00:50:55,417 INFO [train.py:968] (1/2) Epoch 17, batch 32400, libri_loss[loss=0.2732, simple_loss=0.3398, pruned_loss=0.1033, over 29513.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.341, pruned_loss=0.09289, over 5668429.47 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3534, pruned_loss=0.1152, over 5676865.07 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3402, pruned_loss=0.09026, over 5659996.36 frames. ], batch size: 81, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:50:58,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4838, 1.6526, 1.7352, 1.3074], device='cuda:1'), covar=tensor([0.1932, 0.2636, 0.1563, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.0865, 0.0687, 0.0915, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 00:51:09,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:51:13,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:51:57,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:51:58,766 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 32450, giga_loss[loss=0.2142, simple_loss=0.303, pruned_loss=0.06266, over 28402.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3385, pruned_loss=0.09178, over 5673578.22 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3536, pruned_loss=0.1154, over 5678018.43 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3375, pruned_loss=0.08924, over 5665891.36 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:53:01,025 INFO [optim.py:369] (1/2) 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 00:53:10,384 INFO [train.py:968] (1/2) Epoch 17, batch 32500, giga_loss[loss=0.2551, simple_loss=0.3238, pruned_loss=0.09318, over 28036.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.334, pruned_loss=0.09048, over 5669517.26 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3533, pruned_loss=0.1152, over 5672929.99 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.333, pruned_loss=0.08797, over 5668117.73 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:54:19,829 INFO [train.py:968] (1/2) Epoch 17, batch 32550, giga_loss[loss=0.2531, simple_loss=0.3322, pruned_loss=0.08699, over 28126.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3287, pruned_loss=0.08769, over 5660862.75 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3528, pruned_loss=0.1149, over 5673926.61 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3279, pruned_loss=0.08546, over 5658596.99 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 00:54:45,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=763453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:55:00,180 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:55:01,167 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 00:55:04,729 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:55:15,715 INFO [train.py:968] (1/2) Epoch 17, batch 32600, giga_loss[loss=0.2573, simple_loss=0.3282, pruned_loss=0.0932, over 28936.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3301, pruned_loss=0.08905, over 5657363.41 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3524, pruned_loss=0.1147, over 5678189.09 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.329, pruned_loss=0.08656, over 5651156.40 frames. ], batch size: 213, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:55:30,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1268, 3.9404, 3.6846, 1.8138], device='cuda:1'), covar=tensor([0.0619, 0.0825, 0.0891, 0.2360], device='cuda:1'), in_proj_covar=tensor([0.1156, 0.1068, 0.0915, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 00:55:35,593 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:56:14,746 INFO [train.py:968] (1/2) Epoch 17, batch 32650, libri_loss[loss=0.268, simple_loss=0.3437, pruned_loss=0.0961, over 29526.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3309, pruned_loss=0.08922, over 5665598.74 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3525, pruned_loss=0.1147, over 5683763.75 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3294, pruned_loss=0.08658, over 5655100.21 frames. ], batch size: 82, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:57:08,626 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 32700, giga_loss[loss=0.2409, simple_loss=0.3239, pruned_loss=0.07897, over 29042.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3281, pruned_loss=0.08687, over 5660567.41 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3522, pruned_loss=0.1146, over 5688477.47 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3268, pruned_loss=0.0844, over 5647640.10 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:57:41,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=763596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:57:45,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=763599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:58:25,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=763628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 00:58:25,809 INFO [train.py:968] (1/2) Epoch 17, batch 32750, giga_loss[loss=0.2589, simple_loss=0.3307, pruned_loss=0.09353, over 28460.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3264, pruned_loss=0.08555, over 5664729.10 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3519, pruned_loss=0.1145, over 5688154.08 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3255, pruned_loss=0.08354, over 5654824.31 frames. ], batch size: 369, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 00:59:25,542 INFO [optim.py:369] (1/2) 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 00:59:32,022 INFO [train.py:968] (1/2) Epoch 17, batch 32800, giga_loss[loss=0.2491, simple_loss=0.3323, pruned_loss=0.08294, over 28788.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3268, pruned_loss=0.08613, over 5669582.46 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3517, pruned_loss=0.1143, over 5693161.71 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3257, pruned_loss=0.08408, over 5656629.34 frames. ], batch size: 263, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:00:15,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4199, 3.5167, 1.4792, 1.5740], device='cuda:1'), covar=tensor([0.0981, 0.0347, 0.0943, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0537, 0.0369, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 01:00:41,557 INFO [train.py:968] (1/2) Epoch 17, batch 32850, giga_loss[loss=0.2553, simple_loss=0.3381, pruned_loss=0.0862, over 28346.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3268, pruned_loss=0.08597, over 5654304.85 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3514, pruned_loss=0.1142, over 5687446.13 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3256, pruned_loss=0.0837, over 5648191.91 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:01:35,674 INFO [optim.py:369] (1/2) 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 01:01:44,081 INFO [train.py:968] (1/2) Epoch 17, batch 32900, giga_loss[loss=0.2402, simple_loss=0.3024, pruned_loss=0.089, over 24576.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3275, pruned_loss=0.08688, over 5662826.26 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3514, pruned_loss=0.1143, over 5693032.43 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.326, pruned_loss=0.08425, over 5652115.65 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:02:41,428 INFO [train.py:968] (1/2) Epoch 17, batch 32950, giga_loss[loss=0.2124, simple_loss=0.2764, pruned_loss=0.07419, over 24384.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3282, pruned_loss=0.08799, over 5666478.18 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3516, pruned_loss=0.1146, over 5694390.89 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3259, pruned_loss=0.08474, over 5655805.87 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:02:54,722 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-09 01:03:39,022 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 33000, giga_loss[loss=0.2305, simple_loss=0.3273, pruned_loss=0.06689, over 28664.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3282, pruned_loss=0.08643, over 5667764.03 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3514, pruned_loss=0.1145, over 5696526.74 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3264, pruned_loss=0.08375, over 5657368.64 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:03:45,688 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 01:03:52,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2885, 1.7042, 1.5524, 1.1690], device='cuda:1'), covar=tensor([0.1561, 0.2088, 0.1243, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.0866, 0.0684, 0.0913, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 01:03:53,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1585, 1.5592, 1.5817, 1.4164], device='cuda:1'), covar=tensor([0.1756, 0.1383, 0.1871, 0.1503], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0719, 0.0677, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 01:03:54,129 INFO [train.py:1012] (1/2) Epoch 17, validation: loss=0.1987, simple_loss=0.2993, pruned_loss=0.04903, over 944034.00 frames. 
+2023-03-09 01:03:54,130 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 01:04:43,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0617, 3.0025, 1.9783, 1.0642], device='cuda:1'), covar=tensor([0.6544, 0.2871, 0.3647, 0.5958], device='cuda:1'), in_proj_covar=tensor([0.1675, 0.1595, 0.1550, 0.1372], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 01:04:49,062 INFO [train.py:968] (1/2) Epoch 17, batch 33050, giga_loss[loss=0.2799, simple_loss=0.3574, pruned_loss=0.1012, over 28689.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3303, pruned_loss=0.08643, over 5653033.75 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3517, pruned_loss=0.1148, over 5686787.52 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3283, pruned_loss=0.08354, over 5653284.57 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:05:40,790 INFO [optim.py:369] (1/2) 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 01:05:49,246 INFO [train.py:968] (1/2) Epoch 17, batch 33100, giga_loss[loss=0.2529, simple_loss=0.3441, pruned_loss=0.08088, over 28490.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3319, pruned_loss=0.08717, over 5645434.11 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3521, pruned_loss=0.1152, over 5682060.67 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3295, pruned_loss=0.08396, over 5648160.33 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:06:35,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-09 01:06:57,522 INFO [train.py:968] (1/2) Epoch 17, batch 33150, giga_loss[loss=0.2902, simple_loss=0.349, pruned_loss=0.1158, over 26742.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3322, pruned_loss=0.08712, over 5651317.43 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3516, pruned_loss=0.1147, over 5685956.83 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3304, pruned_loss=0.08444, over 5649387.65 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:07:50,429 INFO [optim.py:369] (1/2) 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 01:07:59,315 INFO [train.py:968] (1/2) Epoch 17, batch 33200, libri_loss[loss=0.299, simple_loss=0.3615, pruned_loss=0.1182, over 29521.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3317, pruned_loss=0.08714, over 5650139.44 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3513, pruned_loss=0.1146, over 5681050.32 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3301, pruned_loss=0.0845, over 5652111.61 frames. ], batch size: 85, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:08:34,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3785, 3.1556, 1.4434, 1.6013], device='cuda:1'), covar=tensor([0.1021, 0.0345, 0.1009, 0.1350], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0539, 0.0371, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 01:08:41,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2619, 1.3200, 3.4283, 3.1804], device='cuda:1'), covar=tensor([0.1811, 0.3040, 0.0829, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0720, 0.0623, 0.0913, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 01:08:57,194 INFO [train.py:968] (1/2) Epoch 17, batch 33250, libri_loss[loss=0.2724, simple_loss=0.3255, pruned_loss=0.1097, over 28152.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.33, pruned_loss=0.08631, over 5654148.29 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3512, pruned_loss=0.1148, over 5683665.86 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3282, pruned_loss=0.08312, over 5652228.45 frames. ], batch size: 62, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:09:55,862 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 33300, giga_loss[loss=0.2631, simple_loss=0.3328, pruned_loss=0.0967, over 28650.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3281, pruned_loss=0.08557, over 5658367.39 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3511, pruned_loss=0.1147, over 5684794.53 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3266, pruned_loss=0.083, over 5655724.37 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:10:59,219 INFO [train.py:968] (1/2) Epoch 17, batch 33350, giga_loss[loss=0.2211, simple_loss=0.3087, pruned_loss=0.06671, over 28977.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3297, pruned_loss=0.08654, over 5653920.11 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.351, pruned_loss=0.1146, over 5670624.18 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3279, pruned_loss=0.08366, over 5664003.91 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:11:49,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1697, 1.2978, 1.1051, 0.9255], device='cuda:1'), covar=tensor([0.0959, 0.0470, 0.1030, 0.1031], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0437, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 01:12:03,486 INFO [optim.py:369] (1/2) 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 01:12:07,516 INFO [train.py:968] (1/2) Epoch 17, batch 33400, libri_loss[loss=0.3551, simple_loss=0.3955, pruned_loss=0.1574, over 19581.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.333, pruned_loss=0.08815, over 5644304.73 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3512, pruned_loss=0.1148, over 5662561.37 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3313, pruned_loss=0.08553, over 5660260.57 frames. ], batch size: 187, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:12:40,352 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 33450, giga_loss[loss=0.273, simple_loss=0.3535, pruned_loss=0.09623, over 29073.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3336, pruned_loss=0.0887, over 5643088.84 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3513, pruned_loss=0.1149, over 5663417.73 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3319, pruned_loss=0.08629, over 5654554.57 frames. ], batch size: 285, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:14:14,095 INFO [optim.py:369] (1/2) 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 01:14:16,098 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 33500, giga_loss[loss=0.2723, simple_loss=0.3425, pruned_loss=0.1011, over 26900.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3363, pruned_loss=0.09029, over 5664976.56 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3507, pruned_loss=0.1146, over 5671449.50 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3349, pruned_loss=0.08772, over 5666687.97 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:14:21,229 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 01:15:04,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7338, 4.5769, 4.3294, 1.9283], device='cuda:1'), covar=tensor([0.0469, 0.0660, 0.0818, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.1154, 0.1065, 0.0915, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 01:15:20,841 INFO [train.py:968] (1/2) Epoch 17, batch 33550, giga_loss[loss=0.267, simple_loss=0.3255, pruned_loss=0.1043, over 24455.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3392, pruned_loss=0.09148, over 5656800.92 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3508, pruned_loss=0.1147, over 5672768.36 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3377, pruned_loss=0.089, over 5656469.31 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:16:15,748 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 33600, giga_loss[loss=0.2342, simple_loss=0.2964, pruned_loss=0.08596, over 24221.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3392, pruned_loss=0.09114, over 5645975.31 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3512, pruned_loss=0.1149, over 5658202.30 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3373, pruned_loss=0.0881, over 5657936.44 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:16:43,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8519, 1.2541, 1.2803, 1.0327], device='cuda:1'), covar=tensor([0.1885, 0.1365, 0.2292, 0.1727], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0719, 0.0677, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 01:17:29,202 INFO [train.py:968] (1/2) Epoch 17, batch 33650, giga_loss[loss=0.2252, simple_loss=0.3133, pruned_loss=0.06856, over 28943.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3379, pruned_loss=0.09082, over 5650974.47 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3513, pruned_loss=0.115, over 5662039.94 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3361, pruned_loss=0.0878, over 5656989.88 frames. ], batch size: 284, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:17:44,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1926, 1.7293, 1.3196, 0.4124], device='cuda:1'), covar=tensor([0.3932, 0.2286, 0.3869, 0.5054], device='cuda:1'), in_proj_covar=tensor([0.1678, 0.1600, 0.1552, 0.1373], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 01:18:31,057 INFO [optim.py:369] (1/2) 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 01:18:34,933 INFO [train.py:968] (1/2) Epoch 17, batch 33700, giga_loss[loss=0.2252, simple_loss=0.3087, pruned_loss=0.07085, over 28653.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3362, pruned_loss=0.09072, over 5658541.53 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3513, pruned_loss=0.115, over 5670920.44 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3341, pruned_loss=0.08737, over 5654803.89 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:19:37,609 INFO [train.py:968] (1/2) Epoch 17, batch 33750, giga_loss[loss=0.2906, simple_loss=0.3641, pruned_loss=0.1086, over 27573.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3356, pruned_loss=0.09047, over 5659973.87 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3511, pruned_loss=0.115, over 5676996.51 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3337, pruned_loss=0.08714, over 5651159.50 frames. ], batch size: 472, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:20:04,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5170, 1.7124, 1.2437, 1.1923], device='cuda:1'), covar=tensor([0.0878, 0.0493, 0.0969, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0437, 0.0509, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 01:20:41,118 INFO [optim.py:369] (1/2) 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 01:20:45,581 INFO [train.py:968] (1/2) Epoch 17, batch 33800, giga_loss[loss=0.2635, simple_loss=0.3397, pruned_loss=0.09369, over 28928.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3348, pruned_loss=0.09077, over 5665510.30 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3508, pruned_loss=0.1148, over 5682053.17 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3332, pruned_loss=0.08771, over 5653368.02 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:20:46,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=764679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:20:49,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6237, 1.9446, 1.5325, 1.8508], device='cuda:1'), covar=tensor([0.2594, 0.2492, 0.2891, 0.2504], device='cuda:1'), in_proj_covar=tensor([0.1424, 0.1031, 0.1267, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 01:21:18,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-09 01:21:51,561 INFO [train.py:968] (1/2) Epoch 17, batch 33850, giga_loss[loss=0.245, simple_loss=0.3329, pruned_loss=0.07856, over 28702.00 frames. ], tot_loss[loss=0.257, simple_loss=0.333, pruned_loss=0.09049, over 5656280.30 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3507, pruned_loss=0.1148, over 5685556.48 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3316, pruned_loss=0.08778, over 5643436.04 frames. ], batch size: 262, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:22:17,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=764750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:22:45,511 INFO [optim.py:369] (1/2) 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 01:22:50,856 INFO [train.py:968] (1/2) Epoch 17, batch 33900, giga_loss[loss=0.2282, simple_loss=0.3139, pruned_loss=0.07121, over 28938.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3318, pruned_loss=0.08896, over 5660466.84 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3504, pruned_loss=0.1147, over 5690418.32 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3305, pruned_loss=0.08616, over 5644968.44 frames. ], batch size: 145, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:22:52,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8636, 1.2252, 1.3070, 1.0066], device='cuda:1'), covar=tensor([0.1825, 0.1363, 0.2088, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.0438, 0.0720, 0.0679, 0.0661], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 01:23:44,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=764822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:23:49,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=764825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:23:52,383 INFO [train.py:968] (1/2) Epoch 17, batch 33950, giga_loss[loss=0.2446, simple_loss=0.3391, pruned_loss=0.07505, over 29002.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3315, pruned_loss=0.08745, over 5664219.97 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3503, pruned_loss=0.1147, over 5682965.99 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3302, pruned_loss=0.08477, over 5658386.37 frames. ], batch size: 213, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 01:24:21,017 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=764854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:24:45,456 INFO [optim.py:369] (1/2) 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 01:24:48,658 INFO [train.py:968] (1/2) Epoch 17, batch 34000, giga_loss[loss=0.2431, simple_loss=0.3171, pruned_loss=0.08452, over 24531.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3332, pruned_loss=0.08628, over 5670300.41 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3498, pruned_loss=0.1144, over 5686472.89 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3323, pruned_loss=0.08405, over 5662457.98 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:25:08,188 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=764893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:25:11,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=764896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:25:43,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=764925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:25:48,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5184, 4.3560, 1.7128, 1.7318], device='cuda:1'), covar=tensor([0.0962, 0.0297, 0.0925, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0538, 0.0371, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 01:25:49,019 INFO [train.py:968] (1/2) Epoch 17, batch 34050, giga_loss[loss=0.2219, simple_loss=0.3126, pruned_loss=0.06557, over 28857.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3334, pruned_loss=0.08526, over 5667983.23 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3502, pruned_loss=0.1145, over 5686707.59 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3321, pruned_loss=0.08281, over 5661283.85 frames. ], batch size: 213, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:25:58,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1479, 3.9685, 3.7490, 1.7574], device='cuda:1'), covar=tensor([0.0629, 0.0743, 0.0798, 0.2293], device='cuda:1'), in_proj_covar=tensor([0.1155, 0.1063, 0.0913, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 01:26:50,748 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 34100, giga_loss[loss=0.2657, simple_loss=0.3433, pruned_loss=0.09404, over 28697.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3339, pruned_loss=0.08584, over 5663411.29 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3499, pruned_loss=0.1144, over 5682094.56 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3326, pruned_loss=0.08311, over 5661261.81 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:27:44,499 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-09 01:27:58,781 INFO [train.py:968] (1/2) Epoch 17, batch 34150, giga_loss[loss=0.2448, simple_loss=0.3406, pruned_loss=0.07447, over 28061.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3343, pruned_loss=0.0865, over 5675880.09 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3493, pruned_loss=0.1142, over 5690915.12 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3331, pruned_loss=0.08339, over 5665517.97 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:29:03,395 INFO [optim.py:369] (1/2) 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 01:29:07,370 INFO [train.py:968] (1/2) Epoch 17, batch 34200, giga_loss[loss=0.2514, simple_loss=0.3415, pruned_loss=0.08067, over 28738.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3349, pruned_loss=0.08684, over 5666694.32 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3496, pruned_loss=0.1144, over 5682777.31 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3334, pruned_loss=0.08352, over 5664912.63 frames. ], batch size: 307, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:29:13,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5204, 1.8348, 1.4975, 1.4835], device='cuda:1'), covar=tensor([0.2463, 0.2226, 0.2472, 0.2162], device='cuda:1'), in_proj_covar=tensor([0.1425, 0.1033, 0.1270, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 01:29:49,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-09 01:30:06,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4613, 3.5010, 1.6137, 1.5902], device='cuda:1'), covar=tensor([0.0943, 0.0291, 0.0930, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0538, 0.0370, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 01:30:23,149 INFO [train.py:968] (1/2) Epoch 17, batch 34250, giga_loss[loss=0.2807, simple_loss=0.3468, pruned_loss=0.1073, over 27651.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3347, pruned_loss=0.08617, over 5662240.64 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3497, pruned_loss=0.1146, over 5684559.00 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3333, pruned_loss=0.08314, over 5659013.88 frames. ], batch size: 474, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:30:41,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 01:30:47,368 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-09 01:31:24,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4725, 1.5898, 1.7252, 1.3119], device='cuda:1'), covar=tensor([0.1808, 0.2633, 0.1473, 0.1818], device='cuda:1'), in_proj_covar=tensor([0.0864, 0.0682, 0.0912, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 01:31:26,960 INFO [optim.py:369] (1/2) 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 01:31:28,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-09 01:31:30,927 INFO [train.py:968] (1/2) Epoch 17, batch 34300, giga_loss[loss=0.265, simple_loss=0.3569, pruned_loss=0.08653, over 28854.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3385, pruned_loss=0.08771, over 5663509.12 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3499, pruned_loss=0.1147, over 5686721.91 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3371, pruned_loss=0.08499, over 5658887.51 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:32:04,375 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:32:36,093 INFO [train.py:968] (1/2) Epoch 17, batch 34350, giga_loss[loss=0.2452, simple_loss=0.3319, pruned_loss=0.07925, over 28916.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3396, pruned_loss=0.08824, over 5670995.79 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3498, pruned_loss=0.1148, over 5683212.99 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3383, pruned_loss=0.08529, over 5669661.17 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:33:04,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9046, 3.7431, 3.5492, 1.8007], device='cuda:1'), covar=tensor([0.0662, 0.0768, 0.0799, 0.2021], device='cuda:1'), in_proj_covar=tensor([0.1149, 0.1057, 0.0909, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 01:33:33,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5267, 1.7037, 1.1870, 1.3329], device='cuda:1'), covar=tensor([0.0990, 0.0671, 0.1175, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0439, 0.0511, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 01:33:39,253 INFO [optim.py:369] (1/2) 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 01:33:42,166 INFO [train.py:968] (1/2) Epoch 17, batch 34400, giga_loss[loss=0.2444, simple_loss=0.3275, pruned_loss=0.08067, over 29016.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3387, pruned_loss=0.08864, over 5687568.64 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3496, pruned_loss=0.1146, over 5691158.56 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3374, pruned_loss=0.08542, over 5678859.65 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:34:05,306 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:34:06,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1894, 0.8839, 0.9176, 1.3759], device='cuda:1'), covar=tensor([0.0744, 0.0404, 0.0363, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0115, 0.0117, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 01:34:09,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5470, 1.7009, 1.7686, 1.3297], device='cuda:1'), covar=tensor([0.1820, 0.2510, 0.1545, 0.1715], device='cuda:1'), in_proj_covar=tensor([0.0863, 0.0681, 0.0910, 0.0812], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 01:34:15,736 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:34:53,455 INFO [train.py:968] (1/2) Epoch 17, batch 34450, giga_loss[loss=0.2556, simple_loss=0.3406, pruned_loss=0.08527, over 29006.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3367, pruned_loss=0.08774, over 5684291.58 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3494, pruned_loss=0.1145, over 5691236.70 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3358, pruned_loss=0.08507, over 5677336.57 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:35:05,002 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765334.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:35:39,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:36:01,635 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 34500, giga_loss[loss=0.2309, simple_loss=0.3228, pruned_loss=0.06945, over 28885.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.335, pruned_loss=0.08566, over 5692731.23 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3493, pruned_loss=0.1143, over 5693951.25 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3341, pruned_loss=0.08322, over 5684876.87 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:36:32,090 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 01:36:48,325 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 17, batch 34550, libri_loss[loss=0.2848, simple_loss=0.3359, pruned_loss=0.1168, over 29556.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3346, pruned_loss=0.08576, over 5695155.63 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3491, pruned_loss=0.1141, over 5697714.91 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3339, pruned_loss=0.08345, over 5685350.22 frames. ], batch size: 76, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:37:32,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2401, 1.5680, 1.4131, 1.3959], device='cuda:1'), covar=tensor([0.1775, 0.1831, 0.2285, 0.1903], device='cuda:1'), in_proj_covar=tensor([0.0436, 0.0718, 0.0677, 0.0657], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 01:38:10,125 INFO [optim.py:369] (1/2) 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 01:38:13,448 INFO [train.py:968] (1/2) Epoch 17, batch 34600, giga_loss[loss=0.2642, simple_loss=0.339, pruned_loss=0.0947, over 26701.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3359, pruned_loss=0.08626, over 5689875.85 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3491, pruned_loss=0.1141, over 5700882.32 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3351, pruned_loss=0.08404, over 5679149.46 frames. ], batch size: 555, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:39:11,561 INFO [train.py:968] (1/2) Epoch 17, batch 34650, giga_loss[loss=0.2595, simple_loss=0.3384, pruned_loss=0.09023, over 28384.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3381, pruned_loss=0.08841, over 5668344.47 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3492, pruned_loss=0.1142, over 5692719.41 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.337, pruned_loss=0.08574, over 5666974.47 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:39:47,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5672, 1.9506, 1.7316, 1.6633], device='cuda:1'), covar=tensor([0.1659, 0.1679, 0.1941, 0.1797], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0720, 0.0681, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 01:40:00,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5291, 2.3077, 1.7543, 0.7641], device='cuda:1'), covar=tensor([0.4652, 0.2805, 0.3870, 0.5152], device='cuda:1'), in_proj_covar=tensor([0.1669, 0.1588, 0.1549, 0.1368], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 01:40:06,595 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 34700, giga_loss[loss=0.2419, simple_loss=0.3183, pruned_loss=0.08274, over 28501.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3353, pruned_loss=0.08835, over 5673529.81 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3491, pruned_loss=0.1143, over 5698095.33 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3343, pruned_loss=0.08557, over 5667412.73 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:40:12,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765580.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:41:03,146 INFO [train.py:968] (1/2) Epoch 17, batch 34750, giga_loss[loss=0.2355, simple_loss=0.318, pruned_loss=0.07649, over 28625.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3346, pruned_loss=0.08874, over 5676432.58 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3485, pruned_loss=0.1139, over 5701697.19 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3339, pruned_loss=0.08601, over 5667368.20 frames. ], batch size: 242, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:41:29,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7129, 1.8289, 1.3339, 1.3883], device='cuda:1'), covar=tensor([0.0780, 0.0481, 0.0946, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0436, 0.0508, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 01:41:43,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:41:54,418 INFO [optim.py:369] (1/2) 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 01:41:54,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:41:55,682 INFO [train.py:968] (1/2) Epoch 17, batch 34800, giga_loss[loss=0.289, simple_loss=0.3646, pruned_loss=0.1067, over 28448.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3376, pruned_loss=0.09146, over 5662269.20 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3483, pruned_loss=0.1137, over 5697333.87 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3369, pruned_loss=0.0888, over 5658199.77 frames. ], batch size: 71, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:42:25,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765709.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:42:37,159 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765723.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:42:39,127 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765726.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:42:41,163 INFO [train.py:968] (1/2) Epoch 17, batch 34850, giga_loss[loss=0.2786, simple_loss=0.3644, pruned_loss=0.09635, over 28935.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3463, pruned_loss=0.09625, over 5662092.71 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3488, pruned_loss=0.1141, over 5688231.68 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.345, pruned_loss=0.09339, over 5667258.98 frames. ], batch size: 164, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:42:44,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=765732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:43:06,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765755.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:43:27,422 INFO [optim.py:369] (1/2) 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 01:43:28,740 INFO [train.py:968] (1/2) Epoch 17, batch 34900, giga_loss[loss=0.3254, simple_loss=0.395, pruned_loss=0.1279, over 28275.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3531, pruned_loss=0.1001, over 5670149.91 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3488, pruned_loss=0.114, over 5692205.36 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3522, pruned_loss=0.09744, over 5670021.09 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:43:43,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5610, 1.6370, 1.5931, 1.4298], device='cuda:1'), covar=tensor([0.2589, 0.2355, 0.2005, 0.2348], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1771, 0.1683, 0.1826], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 01:43:58,307 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:00,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:04,461 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:08,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:13,226 INFO [train.py:968] (1/2) Epoch 17, batch 34950, giga_loss[loss=0.251, simple_loss=0.326, pruned_loss=0.08804, over 28582.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3523, pruned_loss=0.1001, over 5677344.67 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3487, pruned_loss=0.1138, over 5696625.74 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3518, pruned_loss=0.09786, over 5672993.54 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:44:24,657 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:31,999 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:32,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765852.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:44:34,458 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765855.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 01:44:41,078 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=765862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:51,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=765875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:53,244 INFO [optim.py:369] (1/2) 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 01:44:53,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=765878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:44:53,859 INFO [train.py:968] (1/2) Epoch 17, batch 35000, giga_loss[loss=0.1996, simple_loss=0.2821, pruned_loss=0.05855, over 28856.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3477, pruned_loss=0.09864, over 5686721.39 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3492, pruned_loss=0.1139, over 5703690.96 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3469, pruned_loss=0.09593, over 5675556.68 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:44:58,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 01:45:17,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=765907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:45:35,676 INFO [train.py:968] (1/2) Epoch 17, batch 35050, giga_loss[loss=0.2421, simple_loss=0.3205, pruned_loss=0.08184, over 28973.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3411, pruned_loss=0.09593, over 5679901.51 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3497, pruned_loss=0.1143, over 5699999.68 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3399, pruned_loss=0.09299, over 5673390.39 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:46:03,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-09 01:46:12,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-09 01:46:14,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-09 01:46:16,262 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 35100, giga_loss[loss=0.2636, simple_loss=0.3135, pruned_loss=0.1069, over 24002.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3331, pruned_loss=0.09226, over 5679936.58 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3495, pruned_loss=0.1141, over 5695019.04 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3321, pruned_loss=0.08961, over 5679048.31 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:46:58,199 INFO [train.py:968] (1/2) Epoch 17, batch 35150, giga_loss[loss=0.2192, simple_loss=0.2988, pruned_loss=0.06974, over 29044.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3262, pruned_loss=0.08948, over 5682688.55 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3499, pruned_loss=0.1145, over 5695841.39 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3246, pruned_loss=0.08649, over 5680786.31 frames. ], batch size: 128, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:47:29,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1941, 1.3022, 1.1679, 0.8770], device='cuda:1'), covar=tensor([0.0937, 0.0535, 0.1118, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0376, 0.0435, 0.0507, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 01:47:39,902 INFO [optim.py:369] (1/2) 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 01:47:40,665 INFO [train.py:968] (1/2) Epoch 17, batch 35200, giga_loss[loss=0.2285, simple_loss=0.2982, pruned_loss=0.07941, over 28872.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3215, pruned_loss=0.08746, over 5681687.53 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3504, pruned_loss=0.1147, over 5697105.04 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3191, pruned_loss=0.08426, over 5678883.76 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:48:07,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.52 vs. limit=5.0
+2023-03-09 01:48:23,229 INFO [train.py:968] (1/2) Epoch 17, batch 35250, giga_loss[loss=0.2528, simple_loss=0.3211, pruned_loss=0.0922, over 28298.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3175, pruned_loss=0.08556, over 5694355.23 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3506, pruned_loss=0.1148, over 5699080.79 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3151, pruned_loss=0.08264, over 5690319.93 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:49:08,979 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 35300, giga_loss[loss=0.2271, simple_loss=0.303, pruned_loss=0.07562, over 29015.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3143, pruned_loss=0.08409, over 5686125.15 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3509, pruned_loss=0.1149, over 5692377.20 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3116, pruned_loss=0.08124, over 5689735.04 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:49:40,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 01:49:52,058 INFO [train.py:968] (1/2) Epoch 17, batch 35350, giga_loss[loss=0.2315, simple_loss=0.3038, pruned_loss=0.07959, over 28799.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3116, pruned_loss=0.0831, over 5675466.86 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.351, pruned_loss=0.1148, over 5696502.55 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3088, pruned_loss=0.08034, over 5674289.66 frames. ], batch size: 99, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:49:53,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3139, 3.1366, 2.9423, 1.3580], device='cuda:1'), covar=tensor([0.0895, 0.1068, 0.0943, 0.2366], device='cuda:1'), in_proj_covar=tensor([0.1148, 0.1058, 0.0908, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 01:49:59,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=766237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:50:36,429 INFO [optim.py:369] (1/2) 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 01:50:36,441 INFO [train.py:968] (1/2) Epoch 17, batch 35400, giga_loss[loss=0.2238, simple_loss=0.2969, pruned_loss=0.07531, over 28281.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3075, pruned_loss=0.08092, over 5672828.71 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.351, pruned_loss=0.1147, over 5690159.76 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3045, pruned_loss=0.07825, over 5677487.95 frames. ], batch size: 368, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:51:03,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4315, 1.5961, 1.5885, 1.4490], device='cuda:1'), covar=tensor([0.1814, 0.2188, 0.2252, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0731, 0.0687, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 01:51:08,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4280, 1.8528, 1.4825, 1.6042], device='cuda:1'), covar=tensor([0.0759, 0.0332, 0.0326, 0.0822], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 01:51:18,866 INFO [train.py:968] (1/2) Epoch 17, batch 35450, giga_loss[loss=0.2367, simple_loss=0.3103, pruned_loss=0.08159, over 27924.00 frames. ], tot_loss[loss=0.2327, simple_loss=0.3055, pruned_loss=0.0799, over 5674235.07 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3514, pruned_loss=0.1149, over 5685620.10 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.3016, pruned_loss=0.07674, over 5682435.51 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:51:40,205 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-09 01:51:42,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2752, 1.3838, 1.4738, 1.0655], device='cuda:1'), covar=tensor([0.1851, 0.2959, 0.1470, 0.1592], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0692, 0.0924, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 01:51:58,355 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 35500, giga_loss[loss=0.1747, simple_loss=0.2496, pruned_loss=0.04994, over 28596.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3052, pruned_loss=0.08018, over 5676485.01 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3521, pruned_loss=0.1148, over 5688269.08 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.2996, pruned_loss=0.07629, over 5680439.18 frames. ], batch size: 60, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:51:59,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=766380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:52:01,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=766383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:52:25,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=766412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 01:52:41,206 INFO [train.py:968] (1/2) Epoch 17, batch 35550, giga_loss[loss=0.2105, simple_loss=0.2802, pruned_loss=0.07041, over 28447.00 frames. ], tot_loss[loss=0.2299, simple_loss=0.3023, pruned_loss=0.07871, over 5680009.56 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3524, pruned_loss=0.115, over 5689641.58 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2964, pruned_loss=0.07465, over 5681903.08 frames. ], batch size: 78, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:53:25,472 INFO [optim.py:369] (1/2) 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 01:53:25,485 INFO [train.py:968] (1/2) Epoch 17, batch 35600, libri_loss[loss=0.3254, simple_loss=0.3957, pruned_loss=0.1275, over 29528.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.2994, pruned_loss=0.07757, over 5675433.29 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3528, pruned_loss=0.1151, over 5693329.36 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2929, pruned_loss=0.07329, over 5672971.17 frames. ], batch size: 82, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:54:06,156 INFO [train.py:968] (1/2) Epoch 17, batch 35650, giga_loss[loss=0.2481, simple_loss=0.3286, pruned_loss=0.08375, over 28970.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3025, pruned_loss=0.07986, over 5667112.22 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3534, pruned_loss=0.1154, over 5685355.88 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2951, pruned_loss=0.07498, over 5672343.98 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:54:36,792 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 35700, giga_loss[loss=0.2675, simple_loss=0.3404, pruned_loss=0.09729, over 28850.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.314, pruned_loss=0.08558, over 5670831.26 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3535, pruned_loss=0.1153, over 5678228.32 frames. ], giga_tot_loss[loss=0.2343, simple_loss=0.3068, pruned_loss=0.08087, over 5680959.84 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:54:51,986 INFO [optim.py:369] (1/2) 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 01:55:39,459 INFO [train.py:968] (1/2) Epoch 17, batch 35750, giga_loss[loss=0.2944, simple_loss=0.3685, pruned_loss=0.1101, over 28894.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3277, pruned_loss=0.09285, over 5676045.71 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3539, pruned_loss=0.1156, over 5681792.46 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3211, pruned_loss=0.08838, over 5680705.21 frames. ], batch size: 145, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:56:11,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8594, 1.0856, 0.9150, 0.7371], device='cuda:1'), covar=tensor([0.2048, 0.2262, 0.1586, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.1858, 0.1781, 0.1704, 0.1845], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 01:56:23,106 INFO [train.py:968] (1/2) Epoch 17, batch 35800, giga_loss[loss=0.2778, simple_loss=0.355, pruned_loss=0.1003, over 28886.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3374, pruned_loss=0.09755, over 5676531.10 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3545, pruned_loss=0.116, over 5680625.50 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.331, pruned_loss=0.09314, over 5681487.45 frames. ], batch size: 119, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:56:23,767 INFO [optim.py:369] (1/2) 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 01:56:51,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-09 01:57:07,384 INFO [train.py:968] (1/2) Epoch 17, batch 35850, giga_loss[loss=0.2871, simple_loss=0.3613, pruned_loss=0.1065, over 29031.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3436, pruned_loss=0.09976, over 5677967.16 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3549, pruned_loss=0.1162, over 5684083.86 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3381, pruned_loss=0.09592, over 5678812.76 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:57:33,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9759, 1.2071, 1.3046, 1.0443], device='cuda:1'), covar=tensor([0.1814, 0.1391, 0.2362, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0735, 0.0693, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 01:57:55,697 INFO [train.py:968] (1/2) Epoch 17, batch 35900, libri_loss[loss=0.292, simple_loss=0.3641, pruned_loss=0.11, over 29754.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3451, pruned_loss=0.09924, over 5670520.61 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.355, pruned_loss=0.1162, over 5687457.94 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3404, pruned_loss=0.09598, over 5668060.66 frames. ], batch size: 87, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:57:56,378 INFO [optim.py:369] (1/2) 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 01:58:41,646 INFO [train.py:968] (1/2) Epoch 17, batch 35950, giga_loss[loss=0.2452, simple_loss=0.3257, pruned_loss=0.08231, over 28620.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3475, pruned_loss=0.1004, over 5673518.48 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3554, pruned_loss=0.1164, over 5689325.53 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3432, pruned_loss=0.09722, over 5669764.70 frames. ], batch size: 60, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 01:58:58,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-09 01:59:26,850 INFO [train.py:968] (1/2) Epoch 17, batch 36000, giga_loss[loss=0.3088, simple_loss=0.3865, pruned_loss=0.1156, over 28494.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3493, pruned_loss=0.1017, over 5660082.94 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3557, pruned_loss=0.1167, over 5663682.72 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3454, pruned_loss=0.09855, over 5678570.77 frames. ], batch size: 336, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 01:59:26,851 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 01:59:35,323 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 01:59:36,143 INFO [optim.py:369] (1/2) 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 01:59:53,698 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-09 02:00:16,108 INFO [train.py:968] (1/2) Epoch 17, batch 36050, giga_loss[loss=0.2904, simple_loss=0.3608, pruned_loss=0.11, over 28899.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3522, pruned_loss=0.1039, over 5666227.92 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3558, pruned_loss=0.1166, over 5666641.90 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.349, pruned_loss=0.1013, over 5678112.15 frames. ], batch size: 99, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:00:23,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8066, 4.6096, 4.3624, 1.9622], device='cuda:1'), covar=tensor([0.0449, 0.0588, 0.0637, 0.2240], device='cuda:1'), in_proj_covar=tensor([0.1140, 0.1053, 0.0902, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 02:00:25,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=766937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:00:53,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8593, 1.9631, 1.7601, 1.7358], device='cuda:1'), covar=tensor([0.2047, 0.2483, 0.2472, 0.2341], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0738, 0.0695, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 02:00:58,200 INFO [train.py:968] (1/2) Epoch 17, batch 36100, giga_loss[loss=0.2897, simple_loss=0.3712, pruned_loss=0.1041, over 28804.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3549, pruned_loss=0.1048, over 5673334.59 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.356, pruned_loss=0.1166, over 5665408.60 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3521, pruned_loss=0.1023, over 5684743.05 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:00:58,733 INFO [optim.py:369] (1/2) 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 02:01:41,121 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767027.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 02:01:42,078 INFO [train.py:968] (1/2) Epoch 17, batch 36150, giga_loss[loss=0.3065, simple_loss=0.36, pruned_loss=0.1265, over 23344.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3574, pruned_loss=0.1048, over 5683256.51 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.356, pruned_loss=0.1166, over 5665408.60 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3552, pruned_loss=0.1029, over 5692135.86 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:02:24,052 INFO [train.py:968] (1/2) Epoch 17, batch 36200, giga_loss[loss=0.2921, simple_loss=0.3745, pruned_loss=0.1049, over 28900.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3597, pruned_loss=0.1061, over 5680467.31 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3564, pruned_loss=0.1168, over 5664193.40 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3578, pruned_loss=0.1041, over 5688209.13 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:02:24,966 INFO [optim.py:369] (1/2) 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 02:02:25,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:02:26,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:02:27,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:02:48,896 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:03:03,184 INFO [train.py:968] (1/2) Epoch 17, batch 36250, giga_loss[loss=0.3147, simple_loss=0.3572, pruned_loss=0.1361, over 23637.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3603, pruned_loss=0.1053, over 5687168.30 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.357, pruned_loss=0.117, over 5669507.45 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3582, pruned_loss=0.1032, over 5689058.36 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:03:33,449 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-09 02:03:41,581 INFO [train.py:968] (1/2) Epoch 17, batch 36300, giga_loss[loss=0.2759, simple_loss=0.3563, pruned_loss=0.0977, over 28834.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3585, pruned_loss=0.103, over 5691636.25 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3571, pruned_loss=0.1169, over 5665455.09 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3568, pruned_loss=0.1012, over 5696456.14 frames. ], batch size: 112, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:03:43,789 INFO [optim.py:369] (1/2) 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 02:03:46,755 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:04:00,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2771, 3.0012, 1.4215, 1.4731], device='cuda:1'), covar=tensor([0.1033, 0.0275, 0.0927, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0533, 0.0368, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 02:04:22,201 INFO [train.py:968] (1/2) Epoch 17, batch 36350, giga_loss[loss=0.2637, simple_loss=0.3421, pruned_loss=0.09265, over 28764.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3583, pruned_loss=0.1028, over 5696369.56 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3577, pruned_loss=0.1172, over 5673129.15 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3564, pruned_loss=0.1005, over 5694197.15 frames. ], batch size: 99, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:04:43,207 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 36400, giga_loss[loss=0.278, simple_loss=0.3544, pruned_loss=0.1008, over 28925.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3577, pruned_loss=0.1026, over 5689333.16 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.358, pruned_loss=0.1174, over 5674274.50 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.356, pruned_loss=0.1006, over 5686786.42 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:05:06,865 INFO [optim.py:369] (1/2) 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 02:05:20,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7097, 4.4992, 4.2848, 2.0539], device='cuda:1'), covar=tensor([0.0487, 0.0682, 0.0707, 0.2052], device='cuda:1'), in_proj_covar=tensor([0.1137, 0.1046, 0.0898, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 02:05:21,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-09 02:05:53,768 INFO [train.py:968] (1/2) Epoch 17, batch 36450, giga_loss[loss=0.4012, simple_loss=0.4282, pruned_loss=0.1871, over 27865.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3611, pruned_loss=0.1073, over 5678989.72 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3586, pruned_loss=0.1177, over 5666146.03 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3592, pruned_loss=0.1053, over 5684735.41 frames. ], batch size: 412, lr: 1.87e-03, grad_scale: 8.0
+2023-03-09 02:06:08,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2751, 0.7583, 0.8218, 1.3976], device='cuda:1'), covar=tensor([0.0773, 0.0382, 0.0358, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 02:06:37,504 INFO [train.py:968] (1/2) Epoch 17, batch 36500, giga_loss[loss=0.2473, simple_loss=0.3261, pruned_loss=0.08428, over 28603.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3622, pruned_loss=0.1098, over 5674241.20 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.359, pruned_loss=0.1178, over 5661758.67 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.3604, pruned_loss=0.1079, over 5682735.59 frames. ], batch size: 78, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:06:40,080 INFO [optim.py:369] (1/2) 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 02:06:55,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9691, 1.1858, 1.2488, 1.0498], device='cuda:1'), covar=tensor([0.1422, 0.1272, 0.1909, 0.1408], device='cuda:1'), in_proj_covar=tensor([0.0452, 0.0739, 0.0693, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 02:06:56,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 02:07:21,874 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:07:22,847 INFO [train.py:968] (1/2) Epoch 17, batch 36550, giga_loss[loss=0.2762, simple_loss=0.3533, pruned_loss=0.09961, over 28956.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3604, pruned_loss=0.1096, over 5678755.25 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3588, pruned_loss=0.1176, over 5663267.92 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3593, pruned_loss=0.1082, over 5684241.87 frames. ], batch size: 106, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:07:47,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:08:03,369 INFO [train.py:968] (1/2) Epoch 17, batch 36600, giga_loss[loss=0.2618, simple_loss=0.3335, pruned_loss=0.09507, over 28772.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3584, pruned_loss=0.109, over 5686981.52 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3588, pruned_loss=0.1176, over 5660282.80 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3575, pruned_loss=0.1076, over 5694553.25 frames. ], batch size: 284, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:08:06,829 INFO [optim.py:369] (1/2) 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 02:08:46,572 INFO [train.py:968] (1/2) Epoch 17, batch 36650, giga_loss[loss=0.2664, simple_loss=0.3485, pruned_loss=0.09218, over 28921.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.358, pruned_loss=0.1092, over 5688607.58 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3591, pruned_loss=0.1178, over 5660164.02 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3569, pruned_loss=0.1077, over 5695432.09 frames. ], batch size: 145, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:08:56,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5516, 1.7070, 1.7848, 1.3534], device='cuda:1'), covar=tensor([0.1809, 0.2422, 0.1462, 0.1656], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0691, 0.0918, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 02:09:00,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767545.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 02:09:03,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767548.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 02:09:13,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:09:14,474 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:09:27,650 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767577.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 02:09:28,698 INFO [train.py:968] (1/2) Epoch 17, batch 36700, giga_loss[loss=0.2331, simple_loss=0.3273, pruned_loss=0.06947, over 29000.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3565, pruned_loss=0.1077, over 5693029.94 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3591, pruned_loss=0.1177, over 5669967.71 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3556, pruned_loss=0.1062, over 5690578.54 frames. ], batch size: 174, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:09:32,493 INFO [optim.py:369] (1/2) 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 02:09:32,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8175, 3.6715, 3.4384, 1.7098], device='cuda:1'), covar=tensor([0.0638, 0.0742, 0.0699, 0.2359], device='cuda:1'), in_proj_covar=tensor([0.1147, 0.1060, 0.0909, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 02:09:49,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:09:52,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:10:12,507 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 36750, giga_loss[loss=0.2626, simple_loss=0.3209, pruned_loss=0.1021, over 23434.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.353, pruned_loss=0.1047, over 5698123.80 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3588, pruned_loss=0.1174, over 5673729.24 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3524, pruned_loss=0.1036, over 5693551.07 frames. ], batch size: 705, lr: 1.87e-03, grad_scale: 2.0
+2023-03-09 02:10:21,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:11:02,102 INFO [train.py:968] (1/2) Epoch 17, batch 36800, giga_loss[loss=0.2406, simple_loss=0.3169, pruned_loss=0.08222, over 28858.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3473, pruned_loss=0.1013, over 5699981.25 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3593, pruned_loss=0.1176, over 5677973.37 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3463, pruned_loss=0.09992, over 5692791.36 frames. ], batch size: 199, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:11:05,713 INFO [optim.py:369] (1/2) 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 02:11:19,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-09 02:11:25,322 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:11:28,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:11:35,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5449, 1.7836, 1.1673, 1.3316], device='cuda:1'), covar=tensor([0.0978, 0.0589, 0.1210, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0442, 0.0513, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 02:11:51,637 INFO [train.py:968] (1/2) Epoch 17, batch 36850, libri_loss[loss=0.3428, simple_loss=0.399, pruned_loss=0.1433, over 25694.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3413, pruned_loss=0.09798, over 5691389.28 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3596, pruned_loss=0.1175, over 5677881.28 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3398, pruned_loss=0.09637, over 5686736.02 frames. ], batch size: 136, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:11:58,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:12:36,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:12:40,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:12:45,123 INFO [train.py:968] (1/2) Epoch 17, batch 36900, giga_loss[loss=0.2476, simple_loss=0.3317, pruned_loss=0.08172, over 29059.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3362, pruned_loss=0.09534, over 5675267.75 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3598, pruned_loss=0.1176, over 5677450.73 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3343, pruned_loss=0.09361, over 5671981.55 frames. ], batch size: 155, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:12:48,601 INFO [optim.py:369] (1/2) 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 02:13:01,977 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-09 02:13:07,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:13:07,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:13:30,253 INFO [train.py:968] (1/2) Epoch 17, batch 36950, giga_loss[loss=0.2483, simple_loss=0.3315, pruned_loss=0.08256, over 28902.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3368, pruned_loss=0.09506, over 5679658.42 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3598, pruned_loss=0.1176, over 5679783.34 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3351, pruned_loss=0.09353, over 5674882.26 frames. ], batch size: 186, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:14:11,173 INFO [train.py:968] (1/2) Epoch 17, batch 37000, giga_loss[loss=0.2588, simple_loss=0.34, pruned_loss=0.08884, over 28964.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3371, pruned_loss=0.09493, over 5678013.20 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.36, pruned_loss=0.1177, over 5664937.38 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3351, pruned_loss=0.09316, over 5687613.09 frames. ], batch size: 227, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:14:14,351 INFO [optim.py:369] (1/2) 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 02:14:16,036 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:14:22,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6576, 3.7125, 1.7339, 1.6801], device='cuda:1'), covar=tensor([0.0941, 0.0332, 0.0847, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0534, 0.0369, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 02:14:43,253 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=767916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:14:53,519 INFO [train.py:968] (1/2) Epoch 17, batch 37050, giga_loss[loss=0.2308, simple_loss=0.3081, pruned_loss=0.0767, over 28594.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3373, pruned_loss=0.09574, over 5681106.39 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3605, pruned_loss=0.1179, over 5669837.37 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3349, pruned_loss=0.09374, over 5684606.82 frames. ], batch size: 60, lr: 1.87e-03, grad_scale: 4.0
+2023-03-09 02:14:59,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=767936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:07,521 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=767945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:09,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=767948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:28,622 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 02:15:33,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=767977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:15:34,371 INFO [train.py:968] (1/2) Epoch 17, batch 37100, libri_loss[loss=0.2965, simple_loss=0.3534, pruned_loss=0.1198, over 29586.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3347, pruned_loss=0.0945, over 5691368.09 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3606, pruned_loss=0.1178, over 5676106.40 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3322, pruned_loss=0.09247, over 5688798.48 frames. ], batch size: 74, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:15:37,038 INFO [optim.py:369] (1/2) 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 02:16:15,045 INFO [train.py:968] (1/2) Epoch 17, batch 37150, giga_loss[loss=0.2691, simple_loss=0.3299, pruned_loss=0.1042, over 28759.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3332, pruned_loss=0.09391, over 5706082.93 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3608, pruned_loss=0.1177, over 5680141.40 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3306, pruned_loss=0.092, over 5700730.27 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:16:53,649 INFO [train.py:968] (1/2) Epoch 17, batch 37200, giga_loss[loss=0.263, simple_loss=0.3364, pruned_loss=0.09479, over 28595.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3314, pruned_loss=0.093, over 5710651.14 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3612, pruned_loss=0.1178, over 5680885.69 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3283, pruned_loss=0.09095, over 5706418.41 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:16:53,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=768079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:16:57,123 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:17:20,251 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=768111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:17:34,686 INFO [train.py:968] (1/2) Epoch 17, batch 37250, giga_loss[loss=0.2509, simple_loss=0.3203, pruned_loss=0.09077, over 28895.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3295, pruned_loss=0.09233, over 5710978.87 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3615, pruned_loss=0.1178, over 5685054.88 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3263, pruned_loss=0.09022, over 5704542.74 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:18:09,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8681, 1.9780, 1.7603, 1.6669], device='cuda:1'), covar=tensor([0.1825, 0.2425, 0.2470, 0.2479], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0739, 0.0696, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 02:18:15,182 INFO [train.py:968] (1/2) Epoch 17, batch 37300, giga_loss[loss=0.206, simple_loss=0.2827, pruned_loss=0.06468, over 28853.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3274, pruned_loss=0.09094, over 5695507.84 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3621, pruned_loss=0.118, over 5670022.04 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3235, pruned_loss=0.08852, over 5705589.96 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:18:18,924 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 37350, giga_loss[loss=0.244, simple_loss=0.318, pruned_loss=0.085, over 28943.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3249, pruned_loss=0.08943, over 5706054.88 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3626, pruned_loss=0.118, over 5673635.69 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3209, pruned_loss=0.08714, over 5711149.10 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:19:03,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3536, 1.6300, 1.5690, 1.4859], device='cuda:1'), covar=tensor([0.2059, 0.1928, 0.2435, 0.2016], device='cuda:1'), in_proj_covar=tensor([0.0454, 0.0737, 0.0694, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 02:19:06,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-09 02:19:19,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=768260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:19:34,782 INFO [train.py:968] (1/2) Epoch 17, batch 37400, libri_loss[loss=0.3178, simple_loss=0.3921, pruned_loss=0.1217, over 29520.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3241, pruned_loss=0.08899, over 5719602.10 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.363, pruned_loss=0.1179, over 5680662.22 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3193, pruned_loss=0.08627, over 5718486.38 frames. ], batch size: 82, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:19:37,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-09 02:19:39,130 INFO [optim.py:369] (1/2) 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 02:19:44,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=768291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:20:15,393 INFO [train.py:968] (1/2) Epoch 17, batch 37450, giga_loss[loss=0.2207, simple_loss=0.3007, pruned_loss=0.07033, over 28701.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.323, pruned_loss=0.08836, over 5723717.52 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3638, pruned_loss=0.1183, over 5681720.74 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3178, pruned_loss=0.08532, over 5722580.50 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:20:20,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 02:20:55,926 INFO [train.py:968] (1/2) Epoch 17, batch 37500, giga_loss[loss=0.3331, simple_loss=0.3966, pruned_loss=0.1348, over 28569.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3242, pruned_loss=0.08892, over 5722073.26 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3638, pruned_loss=0.118, over 5685719.47 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3189, pruned_loss=0.08598, over 5718827.88 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:21:00,926 INFO [optim.py:369] (1/2) 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 02:21:14,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=768403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:19,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=768406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:40,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-09 02:21:40,448 INFO [train.py:968] (1/2) Epoch 17, batch 37550, giga_loss[loss=0.234, simple_loss=0.3177, pruned_loss=0.07514, over 28758.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3283, pruned_loss=0.09145, over 5714519.86 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3635, pruned_loss=0.1178, over 5684447.61 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3236, pruned_loss=0.08877, over 5713604.32 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:21:45,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=768434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:46,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=768435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:21:49,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=768437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:22:13,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1268, 3.9598, 3.7383, 1.8175], device='cuda:1'), covar=tensor([0.0623, 0.0719, 0.0647, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1142, 0.1057, 0.0907, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 02:22:13,426 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 37600, giga_loss[loss=0.3255, simple_loss=0.3888, pruned_loss=0.1311, over 28971.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3353, pruned_loss=0.09598, over 5702524.15 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3638, pruned_loss=0.118, over 5688936.62 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3307, pruned_loss=0.09323, over 5698289.66 frames. ], batch size: 213, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:22:34,747 INFO [optim.py:369] (1/2) 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 02:23:08,856 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=768523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:23:14,480 INFO [train.py:968] (1/2) Epoch 17, batch 37650, giga_loss[loss=0.4065, simple_loss=0.4316, pruned_loss=0.1907, over 26631.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3444, pruned_loss=0.1023, over 5690340.30 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3647, pruned_loss=0.1185, over 5684710.86 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3394, pruned_loss=0.09915, over 5691600.12 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:24:03,731 INFO [train.py:968] (1/2) Epoch 17, batch 37700, giga_loss[loss=0.2845, simple_loss=0.3612, pruned_loss=0.1039, over 28922.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3481, pruned_loss=0.1034, over 5681497.24 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3644, pruned_loss=0.1182, over 5690173.39 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3438, pruned_loss=0.1008, over 5677871.12 frames. ], batch size: 213, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:24:09,519 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 37750, libri_loss[loss=0.3463, simple_loss=0.4052, pruned_loss=0.1437, over 19681.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3536, pruned_loss=0.1062, over 5675579.97 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3645, pruned_loss=0.1182, over 5684417.00 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3498, pruned_loss=0.1038, over 5678498.91 frames. ], batch size: 187, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:25:33,616 INFO [train.py:968] (1/2) Epoch 17, batch 37800, giga_loss[loss=0.2995, simple_loss=0.3786, pruned_loss=0.1102, over 28894.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3603, pruned_loss=0.1104, over 5675909.63 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3654, pruned_loss=0.1188, over 5689831.13 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3563, pruned_loss=0.1076, over 5672758.04 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:25:34,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5002, 3.2243, 1.5794, 1.5871], device='cuda:1'), covar=tensor([0.0930, 0.0341, 0.0856, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0533, 0.0366, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 02:25:38,308 INFO [optim.py:369] (1/2) 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 02:25:40,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7717, 1.1350, 4.9579, 3.6135], device='cuda:1'), covar=tensor([0.1626, 0.3108, 0.0341, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0620, 0.0907, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-09 02:26:07,789 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=768718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:26:15,283 INFO [train.py:968] (1/2) Epoch 17, batch 37850, giga_loss[loss=0.2349, simple_loss=0.3182, pruned_loss=0.07578, over 29007.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3593, pruned_loss=0.1093, over 5674776.22 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3656, pruned_loss=0.1191, over 5692049.35 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3558, pruned_loss=0.1067, over 5670055.86 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:26:36,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7001, 1.8934, 1.3122, 1.4131], device='cuda:1'), covar=tensor([0.0961, 0.0544, 0.1058, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0438, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 02:26:47,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6086, 1.6689, 1.8735, 1.4327], device='cuda:1'), covar=tensor([0.1696, 0.2478, 0.1378, 0.1722], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0691, 0.0917, 0.0814], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 02:26:54,621 INFO [train.py:968] (1/2) Epoch 17, batch 37900, giga_loss[loss=0.2745, simple_loss=0.3538, pruned_loss=0.09761, over 28634.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3557, pruned_loss=0.1063, over 5679987.28 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3659, pruned_loss=0.1194, over 5688153.37 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3525, pruned_loss=0.1036, over 5678397.81 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:27:00,970 INFO [optim.py:369] (1/2) 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 02:27:37,332 INFO [train.py:968] (1/2) Epoch 17, batch 37950, giga_loss[loss=0.2666, simple_loss=0.3463, pruned_loss=0.09345, over 28949.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3534, pruned_loss=0.1041, over 5688956.02 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3656, pruned_loss=0.1193, over 5691970.18 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3508, pruned_loss=0.1017, over 5684345.90 frames. ], batch size: 213, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:28:21,449 INFO [train.py:968] (1/2) Epoch 17, batch 38000, giga_loss[loss=0.2698, simple_loss=0.3476, pruned_loss=0.09597, over 28994.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3523, pruned_loss=0.1031, over 5688591.15 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3659, pruned_loss=0.1196, over 5695810.40 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3497, pruned_loss=0.1006, over 5681217.83 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:28:27,223 INFO [optim.py:369] (1/2) 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 02:28:39,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=768898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:29:04,592 INFO [train.py:968] (1/2) Epoch 17, batch 38050, giga_loss[loss=0.2949, simple_loss=0.3631, pruned_loss=0.1133, over 28925.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3545, pruned_loss=0.1043, over 5686944.94 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3662, pruned_loss=0.1199, over 5694973.84 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3519, pruned_loss=0.1018, over 5681864.63 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:29:30,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4367, 3.3167, 1.4712, 1.5039], device='cuda:1'), covar=tensor([0.0994, 0.0327, 0.0880, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0534, 0.0367, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 02:29:47,721 INFO [train.py:968] (1/2) Epoch 17, batch 38100, giga_loss[loss=0.3545, simple_loss=0.4134, pruned_loss=0.1478, over 29052.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3564, pruned_loss=0.1057, over 5680974.64 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3661, pruned_loss=0.1198, over 5687371.30 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.354, pruned_loss=0.1033, over 5684555.43 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:29:54,317 INFO [optim.py:369] (1/2) 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 02:30:35,170 INFO [train.py:968] (1/2) Epoch 17, batch 38150, giga_loss[loss=0.265, simple_loss=0.3421, pruned_loss=0.09394, over 28547.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3571, pruned_loss=0.1061, over 5691075.47 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3662, pruned_loss=0.1196, over 5692725.62 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3549, pruned_loss=0.104, over 5688948.68 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:30:44,241 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:30:45,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:30:48,816 INFO [zipformer.py:1188] (1/2) 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] (1/2) attn_weights_entropy = tensor([2.2899, 1.4740, 3.8060, 3.1893], device='cuda:1'), covar=tensor([0.1603, 0.2539, 0.0442, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0720, 0.0619, 0.0911, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-09 02:31:01,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1893, 2.5928, 1.2772, 1.2686], device='cuda:1'), covar=tensor([0.1014, 0.0329, 0.0872, 0.1458], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0533, 0.0368, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 02:31:11,420 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-09 02:31:13,816 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 38200, libri_loss[loss=0.2599, simple_loss=0.3311, pruned_loss=0.0944, over 29666.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3581, pruned_loss=0.1074, over 5677018.30 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3666, pruned_loss=0.1199, over 5683289.24 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3559, pruned_loss=0.1053, over 5684048.00 frames. ], batch size: 69, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:31:23,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:31:25,540 INFO [optim.py:369] (1/2) 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 02:31:30,153 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:31:32,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:32:01,122 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 02:32:01,334 INFO [train.py:968] (1/2) Epoch 17, batch 38250, giga_loss[loss=0.295, simple_loss=0.3703, pruned_loss=0.1098, over 28793.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3586, pruned_loss=0.1078, over 5682258.71 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3665, pruned_loss=0.1198, over 5678772.29 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3567, pruned_loss=0.1059, over 5691612.39 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:32:07,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=769137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:32:08,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4600, 1.5430, 1.5833, 1.4677], device='cuda:1'), covar=tensor([0.1882, 0.2235, 0.2380, 0.2014], device='cuda:1'), in_proj_covar=tensor([0.0458, 0.0742, 0.0698, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 02:32:33,264 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 38300, giga_loss[loss=0.2878, simple_loss=0.3686, pruned_loss=0.1035, over 28987.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3583, pruned_loss=0.1068, over 5689692.60 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3664, pruned_loss=0.1195, over 5682210.65 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3567, pruned_loss=0.1052, over 5694166.93 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:32:47,446 INFO [optim.py:369] (1/2) 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 02:33:04,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-09 02:33:24,757 INFO [train.py:968] (1/2) Epoch 17, batch 38350, giga_loss[loss=0.3037, simple_loss=0.3713, pruned_loss=0.118, over 28346.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3573, pruned_loss=0.1047, over 5697624.31 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3664, pruned_loss=0.1195, over 5683320.60 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.356, pruned_loss=0.1035, over 5700166.05 frames. ], batch size: 65, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:33:31,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:33:35,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:33:59,676 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:34:07,801 INFO [train.py:968] (1/2) Epoch 17, batch 38400, giga_loss[loss=0.2547, simple_loss=0.3405, pruned_loss=0.08451, over 28832.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3565, pruned_loss=0.1033, over 5700980.32 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3661, pruned_loss=0.1193, over 5686381.75 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3555, pruned_loss=0.1023, over 5700461.66 frames. ], batch size: 174, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:34:10,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7821, 2.7384, 2.4234, 2.6467], device='cuda:1'), covar=tensor([0.1347, 0.1559, 0.1720, 0.1377], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0738, 0.0696, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 02:34:12,607 INFO [optim.py:369] (1/2) 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 02:34:48,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4245, 1.5364, 1.1686, 1.1039], device='cuda:1'), covar=tensor([0.0854, 0.0489, 0.0990, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0437, 0.0508, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 02:34:49,927 INFO [train.py:968] (1/2) Epoch 17, batch 38450, libri_loss[loss=0.2832, simple_loss=0.3571, pruned_loss=0.1046, over 28634.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3544, pruned_loss=0.102, over 5688225.11 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3664, pruned_loss=0.1194, over 5671332.74 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3532, pruned_loss=0.1007, over 5700356.82 frames. ], batch size: 106, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:35:01,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3671, 1.5261, 1.1910, 1.4278], device='cuda:1'), covar=tensor([0.0794, 0.0339, 0.0354, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0065, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 02:35:30,651 INFO [train.py:968] (1/2) Epoch 17, batch 38500, giga_loss[loss=0.2639, simple_loss=0.338, pruned_loss=0.09494, over 28739.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3534, pruned_loss=0.102, over 5688791.08 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3669, pruned_loss=0.1197, over 5662467.00 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3517, pruned_loss=0.1005, over 5706013.25 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:35:36,299 INFO [optim.py:369] (1/2) 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 02:36:01,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:36:11,949 INFO [train.py:968] (1/2) Epoch 17, batch 38550, giga_loss[loss=0.2958, simple_loss=0.3583, pruned_loss=0.1166, over 28931.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3517, pruned_loss=0.1015, over 5693143.49 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.367, pruned_loss=0.1197, over 5656017.32 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3502, pruned_loss=0.1, over 5713321.00 frames. ], batch size: 106, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:36:34,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:36:37,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.88 vs. limit=5.0
+2023-03-09 02:36:42,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:36:53,230 INFO [train.py:968] (1/2) Epoch 17, batch 38600, giga_loss[loss=0.2665, simple_loss=0.3397, pruned_loss=0.0967, over 28868.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3513, pruned_loss=0.1021, over 5695258.58 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3667, pruned_loss=0.1197, over 5659872.11 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.35, pruned_loss=0.1007, over 5708567.44 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:37:01,377 INFO [optim.py:369] (1/2) 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 02:37:14,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1794, 1.6342, 1.2744, 0.3903], device='cuda:1'), covar=tensor([0.3678, 0.2250, 0.3328, 0.5227], device='cuda:1'), in_proj_covar=tensor([0.1662, 0.1571, 0.1548, 0.1361], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 02:37:20,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3616, 1.9931, 1.5658, 1.7547], device='cuda:1'), covar=tensor([0.0804, 0.0272, 0.0310, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0115, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 02:37:21,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:37:33,541 INFO [train.py:968] (1/2) Epoch 17, batch 38650, libri_loss[loss=0.3159, simple_loss=0.3748, pruned_loss=0.1285, over 29685.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3527, pruned_loss=0.1034, over 5699704.97 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3671, pruned_loss=0.1199, over 5658629.63 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3508, pruned_loss=0.1014, over 5712630.89 frames. ], batch size: 91, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:37:46,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=769544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:37:56,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:37:58,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:12,046 INFO [train.py:968] (1/2) Epoch 17, batch 38700, giga_loss[loss=0.2505, simple_loss=0.3356, pruned_loss=0.0827, over 28906.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3522, pruned_loss=0.1022, over 5708691.52 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3668, pruned_loss=0.1197, over 5664820.57 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5714553.17 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:38:18,668 INFO [optim.py:369] (1/2) 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 02:38:20,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:29,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:31,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:35,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4777, 1.7080, 1.7252, 1.3428], device='cuda:1'), covar=tensor([0.1971, 0.1689, 0.1216, 0.1631], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1794, 0.1717, 0.1865], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 02:38:35,388 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:37,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:50,158 INFO [train.py:968] (1/2) Epoch 17, batch 38750, giga_loss[loss=0.2721, simple_loss=0.3502, pruned_loss=0.09702, over 28896.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3514, pruned_loss=0.1015, over 5712415.05 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3659, pruned_loss=0.1192, over 5675389.17 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3503, pruned_loss=0.09972, over 5709500.69 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:38:53,306 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:38:59,885 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:12,149 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:14,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:14,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-09 02:39:30,556 INFO [train.py:968] (1/2) Epoch 17, batch 38800, giga_loss[loss=0.2589, simple_loss=0.3381, pruned_loss=0.08988, over 28881.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3506, pruned_loss=0.1009, over 5720380.79 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3663, pruned_loss=0.1193, over 5679374.74 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.09907, over 5715115.37 frames. ], batch size: 174, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:39:39,750 INFO [optim.py:369] (1/2) 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 02:39:39,980 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:40,022 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=769687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:42,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=769690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:39:49,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2197, 4.0478, 3.8322, 1.8864], device='cuda:1'), covar=tensor([0.0582, 0.0720, 0.0732, 0.2075], device='cuda:1'), in_proj_covar=tensor([0.1145, 0.1060, 0.0906, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 02:40:07,009 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=769719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:40:14,353 INFO [train.py:968] (1/2) Epoch 17, batch 38850, giga_loss[loss=0.2576, simple_loss=0.336, pruned_loss=0.08957, over 28459.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3502, pruned_loss=0.1014, over 5691015.57 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3669, pruned_loss=0.1198, over 5655939.06 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3482, pruned_loss=0.09917, over 5708613.11 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:40:54,801 INFO [train.py:968] (1/2) Epoch 17, batch 38900, giga_loss[loss=0.2626, simple_loss=0.3361, pruned_loss=0.09458, over 28441.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3459, pruned_loss=0.09891, over 5691001.90 frames. ], libri_tot_loss[loss=0.3033, simple_loss=0.367, pruned_loss=0.1198, over 5658581.39 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.097, over 5702850.77 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:41:01,558 INFO [optim.py:369] (1/2) 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 02:41:35,427 INFO [train.py:968] (1/2) Epoch 17, batch 38950, giga_loss[loss=0.2714, simple_loss=0.3416, pruned_loss=0.1006, over 28900.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3427, pruned_loss=0.09704, over 5700071.90 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3667, pruned_loss=0.1196, over 5661106.75 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3414, pruned_loss=0.09556, over 5707544.21 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:42:16,998 INFO [train.py:968] (1/2) Epoch 17, batch 39000, giga_loss[loss=0.2462, simple_loss=0.3185, pruned_loss=0.08693, over 28691.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3437, pruned_loss=0.09792, over 5691569.76 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3669, pruned_loss=0.1198, over 5654077.28 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3417, pruned_loss=0.09587, over 5705248.96 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:42:16,998 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 02:42:25,446 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 02:42:32,758 INFO [optim.py:369] (1/2) 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 02:42:48,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4580, 1.8545, 1.3981, 1.6379], device='cuda:1'), covar=tensor([0.2630, 0.2646, 0.2928, 0.2339], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1044, 0.1272, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 02:43:07,943 INFO [train.py:968] (1/2) Epoch 17, batch 39050, giga_loss[loss=0.2707, simple_loss=0.3439, pruned_loss=0.09875, over 28763.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3439, pruned_loss=0.09915, over 5689595.30 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3675, pruned_loss=0.1203, over 5658377.32 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3414, pruned_loss=0.09662, over 5697329.74 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:43:18,707 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-09 02:43:47,145 INFO [train.py:968] (1/2) Epoch 17, batch 39100, giga_loss[loss=0.2586, simple_loss=0.3304, pruned_loss=0.09338, over 29033.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3422, pruned_loss=0.09853, over 5697287.64 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3676, pruned_loss=0.1203, over 5659924.73 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3391, pruned_loss=0.09577, over 5703650.75 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:43:54,620 INFO [optim.py:369] (1/2) 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 02:44:21,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5144, 1.8560, 1.4647, 1.5958], device='cuda:1'), covar=tensor([0.2621, 0.2513, 0.2957, 0.2256], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1043, 0.1272, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 02:44:25,586 INFO [train.py:968] (1/2) Epoch 17, batch 39150, giga_loss[loss=0.2382, simple_loss=0.3125, pruned_loss=0.08191, over 28417.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3407, pruned_loss=0.09794, over 5697760.10 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3681, pruned_loss=0.1205, over 5653556.92 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3373, pruned_loss=0.09504, over 5709648.20 frames. ], batch size: 65, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:45:05,006 INFO [train.py:968] (1/2) Epoch 17, batch 39200, giga_loss[loss=0.2197, simple_loss=0.3051, pruned_loss=0.06715, over 29147.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3394, pruned_loss=0.09767, over 5691709.45 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.368, pruned_loss=0.1205, over 5655025.98 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.336, pruned_loss=0.09474, over 5701257.26 frames. ], batch size: 113, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:45:11,607 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:45:15,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8441, 2.0770, 1.4435, 1.6871], device='cuda:1'), covar=tensor([0.0911, 0.0668, 0.1092, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0438, 0.0509, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 02:45:46,892 INFO [train.py:968] (1/2) Epoch 17, batch 39250, giga_loss[loss=0.2843, simple_loss=0.3615, pruned_loss=0.1035, over 27867.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3378, pruned_loss=0.09642, over 5683575.32 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3683, pruned_loss=0.1208, over 5638758.99 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3344, pruned_loss=0.09353, over 5705651.81 frames. ], batch size: 412, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:46:29,595 INFO [train.py:968] (1/2) Epoch 17, batch 39300, giga_loss[loss=0.2407, simple_loss=0.3134, pruned_loss=0.084, over 28096.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3409, pruned_loss=0.09782, over 5673798.97 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3686, pruned_loss=0.1211, over 5634112.95 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3371, pruned_loss=0.09464, over 5696202.12 frames. ], batch size: 77, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:46:36,515 INFO [optim.py:369] (1/2) 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 02:47:13,944 INFO [train.py:968] (1/2) Epoch 17, batch 39350, giga_loss[loss=0.2628, simple_loss=0.3406, pruned_loss=0.09253, over 29099.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3437, pruned_loss=0.09886, over 5677251.57 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3686, pruned_loss=0.121, over 5639371.93 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3403, pruned_loss=0.09602, over 5690999.02 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:47:25,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:47:27,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4327, 3.7161, 1.5497, 1.5253], device='cuda:1'), covar=tensor([0.0926, 0.0253, 0.0962, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0535, 0.0367, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 02:47:55,040 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:47:57,035 INFO [train.py:968] (1/2) Epoch 17, batch 39400, giga_loss[loss=0.3911, simple_loss=0.4234, pruned_loss=0.1794, over 26544.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3458, pruned_loss=0.09953, over 5683461.59 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3684, pruned_loss=0.1208, over 5641382.01 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3428, pruned_loss=0.09701, over 5693115.93 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:48:05,903 INFO [optim.py:369] (1/2) 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 02:48:41,656 INFO [train.py:968] (1/2) Epoch 17, batch 39450, giga_loss[loss=0.2647, simple_loss=0.3481, pruned_loss=0.09072, over 28698.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3441, pruned_loss=0.0976, over 5688878.14 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3682, pruned_loss=0.1209, over 5645504.39 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3416, pruned_loss=0.09529, over 5693563.15 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:48:49,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8051, 5.1883, 2.0384, 2.1099], device='cuda:1'), covar=tensor([0.0908, 0.0305, 0.0844, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0535, 0.0367, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 02:49:23,612 INFO [train.py:968] (1/2) Epoch 17, batch 39500, giga_loss[loss=0.2528, simple_loss=0.3227, pruned_loss=0.09142, over 28505.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3426, pruned_loss=0.09686, over 5698562.62 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3678, pruned_loss=0.1206, over 5650459.15 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3404, pruned_loss=0.09475, over 5698802.68 frames. ], batch size: 78, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:49:32,474 INFO [optim.py:369] (1/2) 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 02:50:06,596 INFO [train.py:968] (1/2) Epoch 17, batch 39550, giga_loss[loss=0.2684, simple_loss=0.3476, pruned_loss=0.09455, over 28784.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3436, pruned_loss=0.09761, over 5700748.99 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3682, pruned_loss=0.1209, over 5652480.43 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3409, pruned_loss=0.0951, over 5700059.12 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 02:50:12,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4696, 2.1634, 1.6476, 0.7014], device='cuda:1'), covar=tensor([0.6179, 0.2645, 0.3793, 0.6533], device='cuda:1'), in_proj_covar=tensor([0.1674, 0.1583, 0.1560, 0.1376], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 02:50:14,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1418, 0.8891, 0.8940, 1.4070], device='cuda:1'), covar=tensor([0.0754, 0.0392, 0.0372, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 02:50:22,818 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:50:28,847 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-09 02:50:31,938 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 39600, giga_loss[loss=0.2615, simple_loss=0.3379, pruned_loss=0.09254, over 28988.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3436, pruned_loss=0.09783, over 5704036.17 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3686, pruned_loss=0.121, over 5644168.49 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3408, pruned_loss=0.09541, over 5710839.09 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:50:57,544 INFO [optim.py:369] (1/2) 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 02:51:09,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-09 02:51:22,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=770518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:51:30,948 INFO [train.py:968] (1/2) Epoch 17, batch 39650, giga_loss[loss=0.3054, simple_loss=0.3773, pruned_loss=0.1167, over 28794.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3471, pruned_loss=0.09955, over 5705529.99 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3682, pruned_loss=0.1207, over 5648133.96 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3449, pruned_loss=0.09759, over 5708389.91 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:51:33,175 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 39700, giga_loss[loss=0.2754, simple_loss=0.3446, pruned_loss=0.1031, over 28915.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.351, pruned_loss=0.1016, over 5703877.09 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1207, over 5649881.23 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3485, pruned_loss=0.09956, over 5705991.52 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:52:12,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7103, 4.5390, 4.3647, 1.7406], device='cuda:1'), covar=tensor([0.0566, 0.0761, 0.0844, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.1137, 0.1054, 0.0904, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 02:52:19,198 INFO [optim.py:369] (1/2) 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 02:52:31,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:34,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:41,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:47,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-09 02:52:50,983 INFO [train.py:968] (1/2) Epoch 17, batch 39750, giga_loss[loss=0.3154, simple_loss=0.3889, pruned_loss=0.121, over 29050.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3523, pruned_loss=0.1019, over 5708174.31 frames. ], libri_tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1208, over 5652974.99 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3499, pruned_loss=0.09997, over 5707884.15 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:52:56,258 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=770635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:52:57,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-09 02:53:11,056 INFO [zipformer.py:1188] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.5160, 2.2871, 1.5539, 0.7306], device='cuda:1'), covar=tensor([0.4953, 0.2950, 0.3943, 0.5242], device='cuda:1'), in_proj_covar=tensor([0.1670, 0.1581, 0.1559, 0.1374], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 02:53:32,185 INFO [train.py:968] (1/2) Epoch 17, batch 39800, giga_loss[loss=0.2952, simple_loss=0.3655, pruned_loss=0.1125, over 28727.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1022, over 5710602.97 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3686, pruned_loss=0.1206, over 5658475.93 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1005, over 5706547.53 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:53:42,496 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 39850, giga_loss[loss=0.2853, simple_loss=0.3593, pruned_loss=0.1056, over 28710.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.354, pruned_loss=0.1031, over 5703893.59 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1205, over 5653230.23 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3521, pruned_loss=0.1016, over 5706106.63 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:54:22,828 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.30 vs. limit=5.0
+2023-03-09 02:54:39,982 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:54:41,914 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:54:44,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6373, 2.5568, 2.3995, 2.1620], device='cuda:1'), covar=tensor([0.1325, 0.1763, 0.1669, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0744, 0.0699, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 02:54:55,556 INFO [train.py:968] (1/2) Epoch 17, batch 39900, giga_loss[loss=0.2597, simple_loss=0.3386, pruned_loss=0.09039, over 28672.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.353, pruned_loss=0.1022, over 5708447.20 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1205, over 5654490.07 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3514, pruned_loss=0.1009, over 5709242.03 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:55:03,552 INFO [optim.py:369] (1/2) 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 02:55:04,515 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=770791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:07,347 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:09,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:31,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:33,801 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=770827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:55:34,798 INFO [train.py:968] (1/2) Epoch 17, batch 39950, giga_loss[loss=0.2458, simple_loss=0.3196, pruned_loss=0.08602, over 28846.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3502, pruned_loss=0.1014, over 5696590.10 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1205, over 5638452.81 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3488, pruned_loss=0.1002, over 5712796.12 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:55:47,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1874, 1.9988, 1.7181, 1.6357], device='cuda:1'), covar=tensor([0.0811, 0.0762, 0.0952, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0439, 0.0508, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 02:56:16,768 INFO [train.py:968] (1/2) Epoch 17, batch 40000, giga_loss[loss=0.3149, simple_loss=0.3639, pruned_loss=0.1329, over 23883.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3461, pruned_loss=0.09909, over 5697817.95 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1202, over 5642525.63 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3448, pruned_loss=0.09806, over 5707774.16 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:56:26,891 INFO [optim.py:369] (1/2) 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 02:56:29,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:56:40,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=770907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:56:58,956 INFO [train.py:968] (1/2) Epoch 17, batch 40050, giga_loss[loss=0.2645, simple_loss=0.3492, pruned_loss=0.0899, over 28884.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.09821, over 5708190.44 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1202, over 5646416.76 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3439, pruned_loss=0.09716, over 5713356.01 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:57:01,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6252, 1.3263, 4.9237, 3.4838], device='cuda:1'), covar=tensor([0.1646, 0.2820, 0.0369, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0620, 0.0913, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 02:57:28,811 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=770967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:57:30,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=770970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:57:37,503 INFO [train.py:968] (1/2) Epoch 17, batch 40100, giga_loss[loss=0.2468, simple_loss=0.3376, pruned_loss=0.07799, over 29030.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3472, pruned_loss=0.09749, over 5713354.62 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1201, over 5652732.11 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3457, pruned_loss=0.09619, over 5713334.17 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 02:57:48,557 INFO [optim.py:369] (1/2) 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 02:57:56,577 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 40150, giga_loss[loss=0.2387, simple_loss=0.3166, pruned_loss=0.08039, over 28742.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3478, pruned_loss=0.09808, over 5711020.93 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3685, pruned_loss=0.1203, over 5656316.71 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3457, pruned_loss=0.09611, over 5709145.25 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:58:25,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=771036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:27,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=771039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:36,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=771050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:39,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=771053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:52,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=771068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 02:58:59,570 INFO [train.py:968] (1/2) Epoch 17, batch 40200, giga_loss[loss=0.2599, simple_loss=0.3371, pruned_loss=0.09136, over 28855.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3463, pruned_loss=0.09836, over 5715455.49 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.12, over 5661319.72 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3447, pruned_loss=0.09663, over 5710191.99 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 02:59:01,977 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:59:37,914 INFO [train.py:968] (1/2) Epoch 17, batch 40250, giga_loss[loss=0.3032, simple_loss=0.3664, pruned_loss=0.12, over 28518.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3467, pruned_loss=0.09995, over 5719386.22 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 5670468.69 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3451, pruned_loss=0.0982, over 5708817.26 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:00:22,473 INFO [train.py:968] (1/2) Epoch 17, batch 40300, giga_loss[loss=0.2727, simple_loss=0.3423, pruned_loss=0.1016, over 28744.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3446, pruned_loss=0.09981, over 5714450.62 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5672804.22 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.343, pruned_loss=0.0982, over 5704334.57 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:00:32,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4845, 1.8639, 1.7787, 1.3045], device='cuda:1'), covar=tensor([0.3301, 0.2357, 0.2520, 0.2964], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1819, 0.1734, 0.1875], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:00:34,097 INFO [optim.py:369] (1/2) 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 03:00:42,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6976, 1.8791, 1.9085, 1.4937], device='cuda:1'), covar=tensor([0.1611, 0.2162, 0.1345, 0.1513], device='cuda:1'), in_proj_covar=tensor([0.0871, 0.0693, 0.0916, 0.0815], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 03:01:04,706 INFO [train.py:968] (1/2) Epoch 17, batch 40350, giga_loss[loss=0.2707, simple_loss=0.344, pruned_loss=0.09871, over 28653.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3436, pruned_loss=0.09983, over 5721331.86 frames. ], libri_tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1197, over 5674908.78 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3416, pruned_loss=0.09805, over 5712336.58 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:01:45,376 INFO [train.py:968] (1/2) Epoch 17, batch 40400, giga_loss[loss=0.2615, simple_loss=0.3405, pruned_loss=0.09122, over 28998.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3425, pruned_loss=0.09936, over 5726301.48 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3675, pruned_loss=0.1193, over 5680942.70 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3406, pruned_loss=0.09772, over 5714845.33 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:01:55,772 INFO [optim.py:369] (1/2) 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 03:01:56,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 03:02:23,576 INFO [train.py:968] (1/2) Epoch 17, batch 40450, giga_loss[loss=0.2899, simple_loss=0.3586, pruned_loss=0.1106, over 28239.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3388, pruned_loss=0.09742, over 5729906.36 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3671, pruned_loss=0.119, over 5687928.41 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3368, pruned_loss=0.09579, over 5715719.67 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:02:38,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2456, 1.5896, 1.2378, 1.0552], device='cuda:1'), covar=tensor([0.2722, 0.2692, 0.3085, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.1434, 0.1039, 0.1270, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 03:02:55,368 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:03:03,520 INFO [train.py:968] (1/2) Epoch 17, batch 40500, giga_loss[loss=0.2311, simple_loss=0.3088, pruned_loss=0.07664, over 28729.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3347, pruned_loss=0.09524, over 5731827.17 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3671, pruned_loss=0.1189, over 5690594.49 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3323, pruned_loss=0.09337, over 5719341.72 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:03:14,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8977, 1.8930, 1.4948, 1.4729], device='cuda:1'), covar=tensor([0.0847, 0.0671, 0.1000, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0377, 0.0441, 0.0509, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:03:14,583 INFO [optim.py:369] (1/2) 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 03:03:45,807 INFO [train.py:968] (1/2) Epoch 17, batch 40550, giga_loss[loss=0.2769, simple_loss=0.3356, pruned_loss=0.1091, over 23874.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3317, pruned_loss=0.09336, over 5723591.15 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.367, pruned_loss=0.1189, over 5692819.12 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3296, pruned_loss=0.09174, over 5712144.73 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:04:28,154 INFO [train.py:968] (1/2) Epoch 17, batch 40600, giga_loss[loss=0.2758, simple_loss=0.3559, pruned_loss=0.09791, over 28960.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3341, pruned_loss=0.09421, over 5719298.30 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3668, pruned_loss=0.1186, over 5694097.96 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3319, pruned_loss=0.09257, over 5709789.60 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:04:39,973 INFO [optim.py:369] (1/2) 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 03:04:46,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-09 03:05:09,267 INFO [train.py:968] (1/2) Epoch 17, batch 40650, giga_loss[loss=0.2774, simple_loss=0.3573, pruned_loss=0.0988, over 28821.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3385, pruned_loss=0.0961, over 5711191.76 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3671, pruned_loss=0.1188, over 5687287.60 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3361, pruned_loss=0.09431, over 5710994.50 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:05:51,802 INFO [train.py:968] (1/2) Epoch 17, batch 40700, giga_loss[loss=0.2858, simple_loss=0.3628, pruned_loss=0.1044, over 28996.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3411, pruned_loss=0.09666, over 5711361.82 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3672, pruned_loss=0.1188, over 5686690.89 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3388, pruned_loss=0.09496, over 5711749.84 frames. ], batch size: 174, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:06:02,537 INFO [optim.py:369] (1/2) 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 03:06:16,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-09 03:06:28,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771620.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 03:06:35,535 INFO [train.py:968] (1/2) Epoch 17, batch 40750, giga_loss[loss=0.285, simple_loss=0.349, pruned_loss=0.1105, over 23892.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3435, pruned_loss=0.09768, over 5707845.71 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3673, pruned_loss=0.119, over 5679287.07 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3412, pruned_loss=0.09592, over 5714295.77 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:06:40,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:07:16,235 INFO [train.py:968] (1/2) Epoch 17, batch 40800, giga_loss[loss=0.2624, simple_loss=0.3412, pruned_loss=0.09179, over 28561.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3459, pruned_loss=0.09938, over 5700739.25 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5674525.05 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3434, pruned_loss=0.09727, over 5709836.69 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:07:29,928 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 40850, giga_loss[loss=0.3735, simple_loss=0.4178, pruned_loss=0.1646, over 27583.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3506, pruned_loss=0.1038, over 5691459.38 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3675, pruned_loss=0.1193, over 5677571.74 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3482, pruned_loss=0.1017, over 5696408.03 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:08:22,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=771743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:08:54,073 INFO [train.py:968] (1/2) Epoch 17, batch 40900, giga_loss[loss=0.379, simple_loss=0.4244, pruned_loss=0.1668, over 28664.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3584, pruned_loss=0.1102, over 5682978.66 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1196, over 5683550.06 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.356, pruned_loss=0.1081, over 5681760.46 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:09:07,925 INFO [optim.py:369] (1/2) 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 03:09:40,505 INFO [train.py:968] (1/2) Epoch 17, batch 40950, giga_loss[loss=0.4445, simple_loss=0.4621, pruned_loss=0.2135, over 27876.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.365, pruned_loss=0.1147, over 5680062.19 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3674, pruned_loss=0.1195, over 5678191.35 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3633, pruned_loss=0.1129, over 5683485.93 frames. ], batch size: 412, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:09:51,755 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 41000, giga_loss[loss=0.3787, simple_loss=0.4254, pruned_loss=0.166, over 28862.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3704, pruned_loss=0.1196, over 5668628.71 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3672, pruned_loss=0.1192, over 5678558.29 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3692, pruned_loss=0.1183, over 5670854.78 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:10:31,880 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=771886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:10:34,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=771889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:10:37,223 INFO [optim.py:369] (1/2) 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 03:10:58,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6196, 1.7630, 1.8901, 1.4055], device='cuda:1'), covar=tensor([0.1569, 0.2328, 0.1299, 0.1585], device='cuda:1'), in_proj_covar=tensor([0.0868, 0.0691, 0.0915, 0.0813], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 03:10:58,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4219, 3.1047, 2.5144, 1.8415], device='cuda:1'), covar=tensor([0.2392, 0.1389, 0.1661, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.1874, 0.1821, 0.1732, 0.1876], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:11:00,227 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=771918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:11:08,688 INFO [train.py:968] (1/2) Epoch 17, batch 41050, giga_loss[loss=0.4206, simple_loss=0.4535, pruned_loss=0.1938, over 28574.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3763, pruned_loss=0.1242, over 5677859.68 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3669, pruned_loss=0.1189, over 5683946.81 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3758, pruned_loss=0.1236, over 5674790.99 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:11:17,773 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=771937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:11:59,988 INFO [train.py:968] (1/2) Epoch 17, batch 41100, giga_loss[loss=0.3367, simple_loss=0.3917, pruned_loss=0.1408, over 28595.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3805, pruned_loss=0.1282, over 5656834.45 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3667, pruned_loss=0.1188, over 5679205.16 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3806, pruned_loss=0.1279, over 5658674.68 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:12:14,323 INFO [optim.py:369] (1/2) 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 03:12:15,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=771995.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 03:12:32,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:12:49,062 INFO [train.py:968] (1/2) Epoch 17, batch 41150, giga_loss[loss=0.3123, simple_loss=0.3802, pruned_loss=0.1222, over 28899.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3817, pruned_loss=0.1295, over 5665803.58 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3672, pruned_loss=0.1191, over 5681880.60 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3819, pruned_loss=0.1294, over 5663813.17 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:13:18,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8671, 1.4552, 5.4691, 3.8577], device='cuda:1'), covar=tensor([0.1692, 0.2707, 0.0349, 0.0746], device='cuda:1'), in_proj_covar=tensor([0.0728, 0.0624, 0.0921, 0.0855], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:13:46,583 INFO [train.py:968] (1/2) Epoch 17, batch 41200, giga_loss[loss=0.3148, simple_loss=0.3713, pruned_loss=0.1292, over 28595.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3853, pruned_loss=0.1339, over 5637511.50 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3672, pruned_loss=0.1191, over 5685113.26 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3857, pruned_loss=0.1341, over 5632217.46 frames. ], batch size: 92, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:14:02,865 INFO [optim.py:369] (1/2) 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 03:14:40,419 INFO [train.py:968] (1/2) Epoch 17, batch 41250, giga_loss[loss=0.4475, simple_loss=0.4687, pruned_loss=0.2132, over 27597.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3895, pruned_loss=0.1383, over 5624812.84 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3674, pruned_loss=0.1192, over 5677562.39 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3898, pruned_loss=0.1385, over 5625727.97 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:14:52,810 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772138.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 03:14:55,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772141.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 03:15:06,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:15:09,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:15:22,895 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 41300, giga_loss[loss=0.3284, simple_loss=0.3844, pruned_loss=0.1362, over 28803.00 frames. ], tot_loss[loss=0.3395, simple_loss=0.3944, pruned_loss=0.1423, over 5632304.84 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3675, pruned_loss=0.1194, over 5684041.65 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3953, pruned_loss=0.1429, over 5625647.48 frames. ], batch size: 284, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:15:38,300 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:16:14,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:16:29,754 INFO [train.py:968] (1/2) Epoch 17, batch 41350, giga_loss[loss=0.3937, simple_loss=0.4296, pruned_loss=0.1789, over 28463.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3955, pruned_loss=0.1439, over 5634267.77 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3677, pruned_loss=0.1194, over 5685341.61 frames. ], giga_tot_loss[loss=0.3427, simple_loss=0.3963, pruned_loss=0.1445, over 5627587.79 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:16:35,811 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=772236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:16:56,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5336, 1.8719, 1.4496, 1.6086], device='cuda:1'), covar=tensor([0.0762, 0.0286, 0.0311, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 03:17:15,196 INFO [train.py:968] (1/2) Epoch 17, batch 41400, giga_loss[loss=0.298, simple_loss=0.3641, pruned_loss=0.116, over 28875.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.392, pruned_loss=0.1416, over 5623359.98 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3679, pruned_loss=0.1197, over 5672877.46 frames. ], giga_tot_loss[loss=0.3391, simple_loss=0.3932, pruned_loss=0.1425, over 5627580.05 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:17:27,120 INFO [optim.py:369] (1/2) 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 03:17:44,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 41450, giga_loss[loss=0.3092, simple_loss=0.3804, pruned_loss=0.119, over 28252.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.391, pruned_loss=0.1405, over 5629956.49 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3676, pruned_loss=0.1194, over 5680805.78 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3935, pruned_loss=0.1427, over 5623877.97 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:18:07,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.55 vs. limit=5.0
+2023-03-09 03:18:11,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4965, 1.6614, 1.5788, 1.3216], device='cuda:1'), covar=tensor([0.2575, 0.2476, 0.1966, 0.2411], device='cuda:1'), in_proj_covar=tensor([0.1900, 0.1840, 0.1748, 0.1895], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:18:30,072 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:18:35,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:18:37,109 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=772364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:18:50,454 INFO [train.py:968] (1/2) Epoch 17, batch 41500, giga_loss[loss=0.3303, simple_loss=0.3871, pruned_loss=0.1367, over 28884.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3903, pruned_loss=0.1393, over 5617838.53 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3676, pruned_loss=0.1194, over 5677175.53 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3929, pruned_loss=0.1416, over 5614597.24 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:19:04,301 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:19:47,510 INFO [train.py:968] (1/2) Epoch 17, batch 41550, giga_loss[loss=0.3455, simple_loss=0.3763, pruned_loss=0.1574, over 23595.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3939, pruned_loss=0.1423, over 5590620.56 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3678, pruned_loss=0.1195, over 5675365.28 frames. ], giga_tot_loss[loss=0.3424, simple_loss=0.3962, pruned_loss=0.1444, over 5588640.56 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:20:13,377 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:20:37,038 INFO [train.py:968] (1/2) Epoch 17, batch 41600, giga_loss[loss=0.2878, simple_loss=0.3661, pruned_loss=0.1048, over 28941.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.392, pruned_loss=0.1404, over 5591207.02 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1201, over 5664986.22 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.394, pruned_loss=0.1422, over 5596429.20 frames. ], batch size: 106, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:20:44,488 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:21:25,341 INFO [train.py:968] (1/2) Epoch 17, batch 41650, giga_loss[loss=0.3269, simple_loss=0.3918, pruned_loss=0.131, over 28280.00 frames. ], tot_loss[loss=0.3305, simple_loss=0.3888, pruned_loss=0.1361, over 5610734.53 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3687, pruned_loss=0.1201, over 5667212.78 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3907, pruned_loss=0.138, over 5611515.96 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:21:39,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4877, 1.6894, 1.5897, 1.4204], device='cuda:1'), covar=tensor([0.2589, 0.2152, 0.1676, 0.1997], device='cuda:1'), in_proj_covar=tensor([0.1890, 0.1826, 0.1738, 0.1883], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:21:53,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5627, 5.3828, 5.1202, 2.4357], device='cuda:1'), covar=tensor([0.0445, 0.0573, 0.0708, 0.1816], device='cuda:1'), in_proj_covar=tensor([0.1165, 0.1082, 0.0924, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 03:22:10,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3370, 2.9740, 2.5010, 1.8811], device='cuda:1'), covar=tensor([0.3011, 0.1739, 0.1948, 0.2575], device='cuda:1'), in_proj_covar=tensor([0.1888, 0.1824, 0.1737, 0.1881], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:22:17,991 INFO [train.py:968] (1/2) Epoch 17, batch 41700, giga_loss[loss=0.366, simple_loss=0.4038, pruned_loss=0.1641, over 27572.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3851, pruned_loss=0.1326, over 5620230.43 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.1199, over 5668733.01 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3871, pruned_loss=0.1344, over 5619159.58 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:22:39,177 INFO [optim.py:369] (1/2) 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 03:22:55,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:22:59,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8868, 2.0566, 1.7582, 1.8235], device='cuda:1'), covar=tensor([0.1797, 0.2427, 0.2227, 0.2178], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0740, 0.0695, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 03:23:14,353 INFO [train.py:968] (1/2) Epoch 17, batch 41750, giga_loss[loss=0.3051, simple_loss=0.354, pruned_loss=0.1281, over 23867.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3811, pruned_loss=0.1296, over 5619241.11 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3681, pruned_loss=0.1198, over 5671973.79 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3832, pruned_loss=0.1312, over 5614761.79 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:23:23,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6958, 2.2797, 1.3675, 0.9683], device='cuda:1'), covar=tensor([0.6952, 0.3522, 0.3594, 0.6183], device='cuda:1'), in_proj_covar=tensor([0.1677, 0.1587, 0.1564, 0.1375], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 03:23:39,474 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 03:23:55,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6442, 1.7396, 1.8057, 1.5973], device='cuda:1'), covar=tensor([0.2415, 0.2190, 0.1627, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.1888, 0.1818, 0.1736, 0.1878], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:24:02,466 INFO [train.py:968] (1/2) Epoch 17, batch 41800, giga_loss[loss=0.3532, simple_loss=0.3945, pruned_loss=0.156, over 26657.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3785, pruned_loss=0.1274, over 5634829.66 frames. ], libri_tot_loss[loss=0.3039, simple_loss=0.3681, pruned_loss=0.1198, over 5678997.63 frames. ], giga_tot_loss[loss=0.3193, simple_loss=0.3805, pruned_loss=0.129, over 5623446.79 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:24:20,295 INFO [optim.py:369] (1/2) 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 03:24:53,376 INFO [train.py:968] (1/2) Epoch 17, batch 41850, libri_loss[loss=0.3005, simple_loss=0.348, pruned_loss=0.1265, over 29643.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3778, pruned_loss=0.1269, over 5638208.88 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3678, pruned_loss=0.1197, over 5673226.00 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3799, pruned_loss=0.1284, over 5632503.62 frames. ], batch size: 73, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:25:01,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=772739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:15,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:18,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:41,619 INFO [train.py:968] (1/2) Epoch 17, batch 41900, giga_loss[loss=0.3018, simple_loss=0.3672, pruned_loss=0.1182, over 28682.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1255, over 5630195.63 frames. ], libri_tot_loss[loss=0.3034, simple_loss=0.3676, pruned_loss=0.1196, over 5666644.28 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.378, pruned_loss=0.1269, over 5629968.65 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:25:48,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:25:56,840 INFO [optim.py:369] (1/2) 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 03:26:33,988 INFO [train.py:968] (1/2) Epoch 17, batch 41950, giga_loss[loss=0.2865, simple_loss=0.3676, pruned_loss=0.1027, over 28811.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3736, pruned_loss=0.1234, over 5627969.97 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3672, pruned_loss=0.1193, over 5661915.40 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3757, pruned_loss=0.1249, over 5631106.47 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:26:44,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 03:27:06,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-09 03:27:24,126 INFO [train.py:968] (1/2) Epoch 17, batch 42000, giga_loss[loss=0.3489, simple_loss=0.3999, pruned_loss=0.149, over 26653.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3752, pruned_loss=0.1217, over 5636819.40 frames. ], libri_tot_loss[loss=0.303, simple_loss=0.3673, pruned_loss=0.1193, over 5663748.30 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.377, pruned_loss=0.123, over 5637132.08 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:27:24,126 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 03:27:33,055 INFO [train.py:1012] (1/2) Epoch 17, validation: loss=0.2056, simple_loss=0.3104, pruned_loss=0.05038, over 944034.00 frames. 
+2023-03-09 03:27:33,055 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 03:27:35,665 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=772882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:27:39,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=772885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:27:50,326 INFO [optim.py:369] (1/2) 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 03:28:09,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-09 03:28:10,259 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=772914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:28:22,484 INFO [train.py:968] (1/2) Epoch 17, batch 42050, giga_loss[loss=0.2933, simple_loss=0.3664, pruned_loss=0.1101, over 28612.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.376, pruned_loss=0.1212, over 5656304.02 frames. ], libri_tot_loss[loss=0.3031, simple_loss=0.3672, pruned_loss=0.1195, over 5669349.75 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3778, pruned_loss=0.1222, over 5650914.76 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:29:08,771 INFO [train.py:968] (1/2) Epoch 17, batch 42100, giga_loss[loss=0.2852, simple_loss=0.3578, pruned_loss=0.1063, over 28731.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3767, pruned_loss=0.1225, over 5658027.79 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3668, pruned_loss=0.1193, over 5667216.72 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3789, pruned_loss=0.1236, over 5654974.98 frames. ], batch size: 78, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:29:23,448 INFO [optim.py:369] (1/2) 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 03:29:27,713 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=772999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:29:54,723 INFO [train.py:968] (1/2) Epoch 17, batch 42150, giga_loss[loss=0.2724, simple_loss=0.346, pruned_loss=0.09935, over 28754.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3773, pruned_loss=0.1234, over 5653878.36 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3669, pruned_loss=0.1192, over 5668436.77 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3792, pruned_loss=0.1244, over 5649797.20 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:29:56,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:30:22,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6467, 1.9900, 1.8281, 1.4600], device='cuda:1'), covar=tensor([0.2755, 0.2159, 0.2304, 0.2596], device='cuda:1'), in_proj_covar=tensor([0.1898, 0.1828, 0.1743, 0.1890], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:30:36,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:30:40,610 INFO [train.py:968] (1/2) Epoch 17, batch 42200, giga_loss[loss=0.3426, simple_loss=0.3936, pruned_loss=0.1458, over 28572.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3758, pruned_loss=0.1235, over 5667960.76 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3667, pruned_loss=0.1191, over 5672010.44 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3776, pruned_loss=0.1243, over 5661482.82 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:30:58,119 INFO [optim.py:369] (1/2) 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 03:31:27,681 INFO [train.py:968] (1/2) Epoch 17, batch 42250, giga_loss[loss=0.34, simple_loss=0.4006, pruned_loss=0.1397, over 29051.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3753, pruned_loss=0.1241, over 5668930.01 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1191, over 5675512.28 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3768, pruned_loss=0.125, over 5660154.47 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:31:38,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2245, 1.4748, 1.3237, 1.1173], device='cuda:1'), covar=tensor([0.2183, 0.2005, 0.1538, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.1897, 0.1826, 0.1745, 0.1888], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:32:16,729 INFO [train.py:968] (1/2) Epoch 17, batch 42300, giga_loss[loss=0.2687, simple_loss=0.3453, pruned_loss=0.09604, over 28657.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3747, pruned_loss=0.123, over 5670316.62 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3665, pruned_loss=0.119, over 5678952.44 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3764, pruned_loss=0.124, over 5660118.07 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:32:31,231 INFO [optim.py:369] (1/2) 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 03:32:39,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5138, 1.7402, 1.6542, 1.4680], device='cuda:1'), covar=tensor([0.2371, 0.2106, 0.2262, 0.2357], device='cuda:1'), in_proj_covar=tensor([0.1894, 0.1823, 0.1740, 0.1887], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:32:58,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0484, 2.5022, 2.0530, 1.6136], device='cuda:1'), covar=tensor([0.2753, 0.1933, 0.2313, 0.2850], device='cuda:1'), in_proj_covar=tensor([0.1891, 0.1819, 0.1739, 0.1885], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:32:59,141 INFO [train.py:968] (1/2) Epoch 17, batch 42350, giga_loss[loss=0.2671, simple_loss=0.351, pruned_loss=0.09162, over 29067.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3746, pruned_loss=0.1217, over 5681479.54 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3662, pruned_loss=0.1187, over 5683313.67 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3766, pruned_loss=0.1228, over 5669047.79 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:33:44,232 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2492, 1.5102, 1.5156, 1.3639], device='cuda:1'), covar=tensor([0.1870, 0.1766, 0.2349, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0747, 0.0701, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 03:33:48,927 INFO [train.py:968] (1/2) Epoch 17, batch 42400, giga_loss[loss=0.3369, simple_loss=0.3942, pruned_loss=0.1398, over 28370.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.375, pruned_loss=0.1219, over 5679762.45 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3661, pruned_loss=0.1186, over 5685735.97 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3767, pruned_loss=0.1229, over 5667647.54 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:34:07,125 INFO [optim.py:369] (1/2) 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 03:34:39,651 INFO [train.py:968] (1/2) Epoch 17, batch 42450, giga_loss[loss=0.3271, simple_loss=0.3878, pruned_loss=0.1332, over 28811.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3755, pruned_loss=0.1226, over 5675893.36 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3664, pruned_loss=0.1187, over 5686095.27 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3767, pruned_loss=0.1233, over 5666078.07 frames. ], batch size: 119, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:35:20,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:35:24,935 INFO [train.py:968] (1/2) Epoch 17, batch 42500, giga_loss[loss=0.3139, simple_loss=0.374, pruned_loss=0.1269, over 28824.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3731, pruned_loss=0.1217, over 5678606.42 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3662, pruned_loss=0.1188, over 5682654.76 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3744, pruned_loss=0.1223, over 5673774.13 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:35:43,517 INFO [optim.py:369] (1/2) 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 03:35:52,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=773406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:36:15,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4619, 1.5874, 1.2024, 1.1702], device='cuda:1'), covar=tensor([0.0848, 0.0490, 0.1009, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0442, 0.0507, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:36:17,616 INFO [train.py:968] (1/2) Epoch 17, batch 42550, giga_loss[loss=0.3204, simple_loss=0.3768, pruned_loss=0.132, over 28559.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3717, pruned_loss=0.1219, over 5670139.91 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 5687115.29 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3728, pruned_loss=0.1223, over 5662255.08 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:36:36,334 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 42600, giga_loss[loss=0.2601, simple_loss=0.3373, pruned_loss=0.09141, over 28393.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3709, pruned_loss=0.1216, over 5684252.39 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3663, pruned_loss=0.1189, over 5692719.88 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.372, pruned_loss=0.1221, over 5672685.60 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:37:20,277 INFO [optim.py:369] (1/2) 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 03:37:38,973 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=773517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:37:42,925 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 42650, giga_loss[loss=0.2664, simple_loss=0.3386, pruned_loss=0.09714, over 28882.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5677543.28 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3663, pruned_loss=0.1188, over 5685984.91 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1217, over 5674710.49 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:38:11,060 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=773549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:11,133 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=773549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:13,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=773552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:39,826 INFO [train.py:968] (1/2) Epoch 17, batch 42700, giga_loss[loss=0.3932, simple_loss=0.4117, pruned_loss=0.1873, over 23803.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3682, pruned_loss=0.1209, over 5663013.21 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3663, pruned_loss=0.1188, over 5691561.94 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.369, pruned_loss=0.1214, over 5655207.13 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:38:42,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5012, 1.6078, 1.6471, 1.3642], device='cuda:1'), covar=tensor([0.2290, 0.2235, 0.1391, 0.2047], device='cuda:1'), in_proj_covar=tensor([0.1895, 0.1824, 0.1744, 0.1885], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 03:38:42,833 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=773581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:52,816 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=773593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:55,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=773596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:38:58,303 INFO [optim.py:369] (1/2) 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 03:39:24,545 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=773625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:39:28,055 INFO [train.py:968] (1/2) Epoch 17, batch 42750, giga_loss[loss=0.3012, simple_loss=0.3709, pruned_loss=0.1157, over 28886.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3677, pruned_loss=0.1205, over 5646160.71 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 5674549.03 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3682, pruned_loss=0.1208, over 5655055.93 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:39:31,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5778, 1.6001, 1.1863, 1.2200], device='cuda:1'), covar=tensor([0.0812, 0.0564, 0.1015, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0445, 0.0509, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:40:10,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:40:17,437 INFO [train.py:968] (1/2) Epoch 17, batch 42800, libri_loss[loss=0.2995, simple_loss=0.3726, pruned_loss=0.1132, over 29750.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3677, pruned_loss=0.1193, over 5661102.00 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.1189, over 5678223.44 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3681, pruned_loss=0.1196, over 5664244.70 frames. ], batch size: 87, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:40:32,976 INFO [optim.py:369] (1/2) 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 03:40:58,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3252, 1.6180, 1.2721, 1.0222], device='cuda:1'), covar=tensor([0.2521, 0.2410, 0.2842, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.1434, 0.1040, 0.1272, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 03:41:01,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2676, 2.9018, 1.4198, 1.4480], device='cuda:1'), covar=tensor([0.0991, 0.0401, 0.0896, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0543, 0.0370, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 03:41:01,645 INFO [train.py:968] (1/2) Epoch 17, batch 42850, giga_loss[loss=0.2856, simple_loss=0.3571, pruned_loss=0.1071, over 28680.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1187, over 5661027.49 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3666, pruned_loss=0.1191, over 5674045.18 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1187, over 5667173.99 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:41:47,938 INFO [train.py:968] (1/2) Epoch 17, batch 42900, giga_loss[loss=0.2855, simple_loss=0.3551, pruned_loss=0.108, over 28854.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3677, pruned_loss=0.1182, over 5668228.74 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.366, pruned_loss=0.1187, over 5678034.99 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3685, pruned_loss=0.1186, over 5669197.79 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:41:49,054 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=773779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:41:52,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1545, 2.1200, 1.6561, 1.7838], device='cuda:1'), covar=tensor([0.0867, 0.0724, 0.0966, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0443, 0.0507, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:42:05,280 INFO [optim.py:369] (1/2) 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 03:42:21,169 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-09 03:42:37,269 INFO [train.py:968] (1/2) Epoch 17, batch 42950, giga_loss[loss=0.3418, simple_loss=0.3972, pruned_loss=0.1432, over 28530.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3702, pruned_loss=0.12, over 5682157.77 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.1189, over 5683631.28 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3705, pruned_loss=0.1202, over 5677742.61 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:42:57,435 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 03:43:27,991 INFO [train.py:968] (1/2) Epoch 17, batch 43000, giga_loss[loss=0.2888, simple_loss=0.3599, pruned_loss=0.1088, over 28532.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3727, pruned_loss=0.1229, over 5681009.43 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3658, pruned_loss=0.1184, over 5686490.23 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3736, pruned_loss=0.1235, over 5675059.24 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:43:47,974 INFO [optim.py:369] (1/2) 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 03:44:17,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-09 03:44:20,246 INFO [train.py:968] (1/2) Epoch 17, batch 43050, giga_loss[loss=0.3304, simple_loss=0.3854, pruned_loss=0.1377, over 28761.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5668057.01 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.366, pruned_loss=0.1187, over 5678971.82 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1252, over 5669608.78 frames. ], batch size: 243, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:44:29,833 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-09 03:45:12,480 INFO [train.py:968] (1/2) Epoch 17, batch 43100, giga_loss[loss=0.4441, simple_loss=0.4665, pruned_loss=0.2108, over 24201.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3764, pruned_loss=0.128, over 5651282.94 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3663, pruned_loss=0.1189, over 5672549.65 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3768, pruned_loss=0.1282, over 5657676.57 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:45:32,909 INFO [optim.py:369] (1/2) 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 03:45:48,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-09 03:46:01,685 INFO [train.py:968] (1/2) Epoch 17, batch 43150, giga_loss[loss=0.411, simple_loss=0.4398, pruned_loss=0.1911, over 26427.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3762, pruned_loss=0.1283, over 5651475.88 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3663, pruned_loss=0.1189, over 5673345.90 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3765, pruned_loss=0.1285, over 5655655.74 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 03:46:18,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=774047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:46:47,985 INFO [train.py:968] (1/2) Epoch 17, batch 43200, giga_loss[loss=0.3478, simple_loss=0.3975, pruned_loss=0.149, over 28616.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.375, pruned_loss=0.1269, over 5653631.75 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3666, pruned_loss=0.119, over 5669849.21 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3753, pruned_loss=0.1271, over 5659210.11 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:47:05,637 INFO [optim.py:369] (1/2) 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 03:47:29,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 03:47:31,325 INFO [train.py:968] (1/2) Epoch 17, batch 43250, giga_loss[loss=0.3104, simple_loss=0.3801, pruned_loss=0.1203, over 29016.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.373, pruned_loss=0.1236, over 5663930.05 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.366, pruned_loss=0.1186, over 5675921.39 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3739, pruned_loss=0.1243, over 5662596.23 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:47:58,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=774154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:48:14,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9202, 3.7335, 3.5495, 1.9249], device='cuda:1'), covar=tensor([0.0682, 0.0820, 0.0751, 0.1891], device='cuda:1'), in_proj_covar=tensor([0.1170, 0.1082, 0.0926, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 03:48:20,895 INFO [train.py:968] (1/2) Epoch 17, batch 43300, giga_loss[loss=0.2624, simple_loss=0.3382, pruned_loss=0.09323, over 28350.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1217, over 5659239.95 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1183, over 5678284.37 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1225, over 5656154.60 frames. ], batch size: 60, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:48:34,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=774190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:48:36,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=774193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:48:40,884 INFO [optim.py:369] (1/2) 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 03:49:01,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=774222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:49:07,992 INFO [train.py:968] (1/2) Epoch 17, batch 43350, giga_loss[loss=0.3024, simple_loss=0.3472, pruned_loss=0.1288, over 23638.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3692, pruned_loss=0.1212, over 5669676.52 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3661, pruned_loss=0.1186, over 5681156.59 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3696, pruned_loss=0.1216, over 5664514.19 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:49:10,469 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:49:19,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3271, 2.1944, 1.7262, 1.7311], device='cuda:1'), covar=tensor([0.0861, 0.0774, 0.1045, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0444, 0.0508, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:49:57,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5199, 1.1352, 4.3398, 3.3495], device='cuda:1'), covar=tensor([0.1565, 0.2957, 0.0428, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0629, 0.0926, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:49:58,042 INFO [train.py:968] (1/2) Epoch 17, batch 43400, giga_loss[loss=0.285, simple_loss=0.3565, pruned_loss=0.1068, over 28783.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5670446.03 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 5682831.11 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1213, over 5664624.42 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:50:07,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4094, 1.8474, 1.3197, 0.8132], device='cuda:1'), covar=tensor([0.4216, 0.2507, 0.3038, 0.5041], device='cuda:1'), in_proj_covar=tensor([0.1685, 0.1605, 0.1574, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 03:50:13,042 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=774297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:50:14,021 INFO [optim.py:369] (1/2) 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 03:50:15,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=774300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:50:40,408 INFO [train.py:968] (1/2) Epoch 17, batch 43450, giga_loss[loss=0.2808, simple_loss=0.3578, pruned_loss=0.1019, over 29004.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5676490.52 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1183, over 5689097.53 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3708, pruned_loss=0.1229, over 5665926.83 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:50:41,674 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=774329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:51:26,135 INFO [train.py:968] (1/2) Epoch 17, batch 43500, giga_loss[loss=0.312, simple_loss=0.3905, pruned_loss=0.1168, over 29007.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3748, pruned_loss=0.1237, over 5665215.24 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1184, over 5686790.58 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3755, pruned_loss=0.1243, over 5658642.96 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:51:47,963 INFO [optim.py:369] (1/2) 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 03:52:07,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4484, 1.7148, 1.4899, 1.5746], device='cuda:1'), covar=tensor([0.0796, 0.0314, 0.0321, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 03:52:22,407 INFO [train.py:968] (1/2) Epoch 17, batch 43550, giga_loss[loss=0.288, simple_loss=0.3623, pruned_loss=0.1069, over 29040.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3766, pruned_loss=0.1229, over 5669003.44 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1183, over 5687945.63 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3773, pruned_loss=0.1235, over 5662802.48 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:52:43,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-09 03:53:10,617 INFO [train.py:968] (1/2) Epoch 17, batch 43600, giga_loss[loss=0.3206, simple_loss=0.3861, pruned_loss=0.1275, over 28833.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3783, pruned_loss=0.1246, over 5660827.25 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3654, pruned_loss=0.1182, over 5682047.91 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3792, pruned_loss=0.1251, over 5660364.06 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:53:27,804 INFO [optim.py:369] (1/2) 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 03:53:56,747 INFO [train.py:968] (1/2) Epoch 17, batch 43650, giga_loss[loss=0.301, simple_loss=0.3708, pruned_loss=0.1156, over 28837.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3792, pruned_loss=0.1255, over 5655127.93 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1185, over 5675186.70 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3802, pruned_loss=0.126, over 5659650.44 frames. ], batch size: 112, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:54:43,685 INFO [train.py:968] (1/2) Epoch 17, batch 43700, giga_loss[loss=0.2615, simple_loss=0.336, pruned_loss=0.09349, over 28614.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3799, pruned_loss=0.1273, over 5642512.92 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1185, over 5660827.01 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.381, pruned_loss=0.1278, over 5659643.63 frames. ], batch size: 71, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:55:01,482 INFO [optim.py:369] (1/2) 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 03:55:09,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=774606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:55:32,551 INFO [train.py:968] (1/2) Epoch 17, batch 43750, giga_loss[loss=0.3365, simple_loss=0.3918, pruned_loss=0.1406, over 28964.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3788, pruned_loss=0.1272, over 5637561.96 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3661, pruned_loss=0.1188, over 5654346.18 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3794, pruned_loss=0.1275, over 5657331.11 frames. ], batch size: 213, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:55:40,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:56:16,927 INFO [train.py:968] (1/2) Epoch 17, batch 43800, giga_loss[loss=0.2784, simple_loss=0.3437, pruned_loss=0.1065, over 28114.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.376, pruned_loss=0.1262, over 5641001.73 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3663, pruned_loss=0.1191, over 5655896.30 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3768, pruned_loss=0.1265, over 5654970.38 frames. ], batch size: 77, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:56:35,053 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 43850, giga_loss[loss=0.4073, simple_loss=0.4315, pruned_loss=0.1916, over 26532.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3732, pruned_loss=0.1246, over 5654314.36 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3658, pruned_loss=0.1188, over 5658692.34 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3746, pruned_loss=0.1253, over 5662723.60 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:57:11,936 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:57:22,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=774749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:57:27,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=774752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 03:57:53,299 INFO [train.py:968] (1/2) Epoch 17, batch 43900, giga_loss[loss=0.275, simple_loss=0.3471, pruned_loss=0.1014, over 28736.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3736, pruned_loss=0.125, over 5665013.15 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3655, pruned_loss=0.1186, over 5664164.98 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3752, pruned_loss=0.126, over 5666872.91 frames. ], batch size: 99, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:57:55,725 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:58:43,543 INFO [train.py:968] (1/2) Epoch 17, batch 43950, giga_loss[loss=0.2874, simple_loss=0.3571, pruned_loss=0.1089, over 28683.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3749, pruned_loss=0.1266, over 5662027.02 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3651, pruned_loss=0.1184, over 5664199.26 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3767, pruned_loss=0.1277, over 5663438.22 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 03:58:48,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5463, 2.2702, 1.7459, 0.7367], device='cuda:1'), covar=tensor([0.6061, 0.2845, 0.3704, 0.6254], device='cuda:1'), in_proj_covar=tensor([0.1691, 0.1606, 0.1572, 0.1386], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 03:59:27,540 INFO [train.py:968] (1/2) Epoch 17, batch 44000, giga_loss[loss=0.2624, simple_loss=0.3359, pruned_loss=0.09443, over 29027.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3739, pruned_loss=0.1265, over 5662249.97 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.365, pruned_loss=0.1182, over 5660099.69 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3757, pruned_loss=0.1278, over 5665965.68 frames. ], batch size: 128, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 03:59:31,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2139, 1.2871, 3.9355, 3.2531], device='cuda:1'), covar=tensor([0.1618, 0.2425, 0.0426, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0626, 0.0927, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 03:59:42,503 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=774894.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 03:59:48,102 INFO [optim.py:369] (1/2) 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 04:00:13,923 INFO [train.py:968] (1/2) Epoch 17, batch 44050, giga_loss[loss=0.3244, simple_loss=0.3873, pruned_loss=0.1308, over 28627.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.373, pruned_loss=0.1256, over 5657068.43 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3655, pruned_loss=0.1185, over 5649715.41 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3741, pruned_loss=0.1265, over 5669391.31 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:00:53,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7503, 1.0789, 4.9212, 3.5615], device='cuda:1'), covar=tensor([0.1644, 0.3079, 0.0407, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0628, 0.0931, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:01:00,056 INFO [train.py:968] (1/2) Epoch 17, batch 44100, giga_loss[loss=0.2752, simple_loss=0.3515, pruned_loss=0.09944, over 28548.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3736, pruned_loss=0.1253, over 5649307.83 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3655, pruned_loss=0.1185, over 5645833.29 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3746, pruned_loss=0.1261, over 5662827.79 frames. ], batch size: 65, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:01:27,807 INFO [optim.py:369] (1/2) 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 04:01:38,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=775012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:01:50,850 INFO [train.py:968] (1/2) Epoch 17, batch 44150, giga_loss[loss=0.2863, simple_loss=0.3563, pruned_loss=0.1081, over 28898.00 frames. ], tot_loss[loss=0.313, simple_loss=0.375, pruned_loss=0.1255, over 5658821.27 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3656, pruned_loss=0.1185, over 5651843.40 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3761, pruned_loss=0.1264, over 5664167.02 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:02:43,194 INFO [train.py:968] (1/2) Epoch 17, batch 44200, giga_loss[loss=0.3717, simple_loss=0.4122, pruned_loss=0.1656, over 28288.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1255, over 5662252.31 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3652, pruned_loss=0.1183, over 5654323.55 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3756, pruned_loss=0.1265, over 5664153.71 frames. ], batch size: 368, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:03:06,432 INFO [optim.py:369] (1/2) 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 04:03:16,890 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 44250, giga_loss[loss=0.2867, simple_loss=0.3688, pruned_loss=0.1023, over 29076.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3764, pruned_loss=0.1252, over 5658219.03 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3653, pruned_loss=0.1184, over 5649295.85 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3774, pruned_loss=0.126, over 5664534.19 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:03:48,223 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 04:03:58,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=775155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:03:59,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=775158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:04:16,435 INFO [train.py:968] (1/2) Epoch 17, batch 44300, giga_loss[loss=0.3575, simple_loss=0.413, pruned_loss=0.151, over 28499.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3781, pruned_loss=0.1238, over 5674046.01 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3653, pruned_loss=0.1184, over 5652133.78 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3791, pruned_loss=0.1244, over 5676479.05 frames. ], batch size: 336, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:04:25,934 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=775187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:04:32,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1502, 1.5580, 1.5498, 1.3749], device='cuda:1'), covar=tensor([0.1907, 0.1533, 0.2162, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0741, 0.0698, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 04:04:38,260 INFO [optim.py:369] (1/2) 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 04:05:07,238 INFO [train.py:968] (1/2) Epoch 17, batch 44350, giga_loss[loss=0.3184, simple_loss=0.3893, pruned_loss=0.1238, over 28555.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3813, pruned_loss=0.1253, over 5669265.22 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3657, pruned_loss=0.1188, over 5644495.15 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.382, pruned_loss=0.1256, over 5678416.89 frames. ], batch size: 85, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:05:37,269 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=775256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:05:41,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=775259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:05:51,149 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 44400, giga_loss[loss=0.3076, simple_loss=0.3732, pruned_loss=0.121, over 28699.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3841, pruned_loss=0.1281, over 5669957.53 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3658, pruned_loss=0.1189, over 5645808.68 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3846, pruned_loss=0.1283, over 5676000.96 frames. ], batch size: 242, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:06:07,702 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=775288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:06:14,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3632, 1.5298, 1.3229, 1.5376], device='cuda:1'), covar=tensor([0.0775, 0.0331, 0.0331, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 04:06:20,102 INFO [optim.py:369] (1/2) 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 04:06:25,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2211, 3.9832, 3.7860, 1.8124], device='cuda:1'), covar=tensor([0.0756, 0.1052, 0.1121, 0.2246], device='cuda:1'), in_proj_covar=tensor([0.1178, 0.1093, 0.0937, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 04:06:26,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6691, 1.6143, 1.2553, 1.2800], device='cuda:1'), covar=tensor([0.0854, 0.0639, 0.0976, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0445, 0.0509, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:06:34,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2518, 0.7677, 0.8209, 1.3607], device='cuda:1'), covar=tensor([0.0744, 0.0385, 0.0355, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 04:06:47,992 INFO [train.py:968] (1/2) Epoch 17, batch 44450, giga_loss[loss=0.2748, simple_loss=0.3498, pruned_loss=0.0999, over 29070.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3845, pruned_loss=0.13, over 5645286.75 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3655, pruned_loss=0.1187, over 5643755.76 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3857, pruned_loss=0.1306, over 5652388.32 frames. ], batch size: 164, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:07:33,367 INFO [train.py:968] (1/2) Epoch 17, batch 44500, giga_loss[loss=0.3389, simple_loss=0.4059, pruned_loss=0.1359, over 28446.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3837, pruned_loss=0.13, over 5642262.29 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3654, pruned_loss=0.1186, over 5637573.21 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3852, pruned_loss=0.1308, over 5653946.38 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:07:48,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 04:07:52,488 INFO [optim.py:369] (1/2) 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 04:08:02,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=775412.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 04:08:05,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=775415.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 04:08:13,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2191, 1.4115, 1.3552, 1.1246], device='cuda:1'), covar=tensor([0.2617, 0.2488, 0.1503, 0.2180], device='cuda:1'), in_proj_covar=tensor([0.1895, 0.1832, 0.1754, 0.1896], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 04:08:17,331 INFO [train.py:968] (1/2) Epoch 17, batch 44550, giga_loss[loss=0.2951, simple_loss=0.3686, pruned_loss=0.1108, over 28987.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.382, pruned_loss=0.1284, over 5652918.34 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3657, pruned_loss=0.1189, over 5644198.25 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3833, pruned_loss=0.129, over 5656776.64 frames. ], batch size: 227, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:08:30,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=775444.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 04:09:04,152 INFO [train.py:968] (1/2) Epoch 17, batch 44600, giga_loss[loss=0.2758, simple_loss=0.3659, pruned_loss=0.09284, over 28949.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.381, pruned_loss=0.1258, over 5659987.18 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3653, pruned_loss=0.1186, over 5648171.54 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3827, pruned_loss=0.1267, over 5660030.79 frames. ], batch size: 145, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:09:21,731 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=775497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:09:25,110 INFO [optim.py:369] (1/2) 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 04:09:41,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=775519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:09:50,683 INFO [train.py:968] (1/2) Epoch 17, batch 44650, giga_loss[loss=0.3169, simple_loss=0.3812, pruned_loss=0.1263, over 28948.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3812, pruned_loss=0.1247, over 5665980.82 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3651, pruned_loss=0.1184, over 5652769.83 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.383, pruned_loss=0.1257, over 5662179.98 frames. ], batch size: 213, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:10:40,675 INFO [train.py:968] (1/2) Epoch 17, batch 44700, giga_loss[loss=0.3001, simple_loss=0.3709, pruned_loss=0.1147, over 28801.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3806, pruned_loss=0.1244, over 5659995.23 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3655, pruned_loss=0.1186, over 5644217.55 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3821, pruned_loss=0.1251, over 5665860.56 frames. ], batch size: 243, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:10:59,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6496, 2.2080, 1.3761, 0.8687], device='cuda:1'), covar=tensor([0.4974, 0.2494, 0.2634, 0.5163], device='cuda:1'), in_proj_covar=tensor([0.1687, 0.1603, 0.1564, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 04:11:06,660 INFO [optim.py:369] (1/2) 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 04:11:13,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1975, 4.0197, 3.7970, 1.7381], device='cuda:1'), covar=tensor([0.0603, 0.0691, 0.0720, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.1100, 0.0941, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 04:11:28,708 INFO [train.py:968] (1/2) Epoch 17, batch 44750, libri_loss[loss=0.3444, simple_loss=0.3977, pruned_loss=0.1456, over 29663.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3802, pruned_loss=0.1246, over 5661080.85 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3655, pruned_loss=0.1186, over 5639315.48 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3816, pruned_loss=0.1253, over 5671217.41 frames. ], batch size: 88, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:12:14,401 INFO [train.py:968] (1/2) Epoch 17, batch 44800, giga_loss[loss=0.3376, simple_loss=0.3942, pruned_loss=0.1405, over 28802.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.38, pruned_loss=0.1261, over 5646587.18 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3655, pruned_loss=0.1186, over 5634224.02 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3815, pruned_loss=0.1268, over 5658764.85 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 04:12:23,307 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 04:12:38,018 INFO [optim.py:369] (1/2) 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 04:12:46,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 04:13:02,458 INFO [train.py:968] (1/2) Epoch 17, batch 44850, giga_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.114, over 28541.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3781, pruned_loss=0.1258, over 5650944.81 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3654, pruned_loss=0.1186, over 5642570.63 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3798, pruned_loss=0.1266, over 5653307.23 frames. ], batch size: 307, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 04:13:14,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8858, 2.8837, 1.8353, 0.9329], device='cuda:1'), covar=tensor([0.5525, 0.2326, 0.2897, 0.4761], device='cuda:1'), in_proj_covar=tensor([0.1695, 0.1611, 0.1572, 0.1386], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 04:13:17,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5258, 1.7209, 1.7722, 1.3024], device='cuda:1'), covar=tensor([0.1774, 0.2763, 0.1559, 0.1859], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0702, 0.0922, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 04:13:49,577 INFO [train.py:968] (1/2) Epoch 17, batch 44900, giga_loss[loss=0.2749, simple_loss=0.3439, pruned_loss=0.103, over 28917.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3763, pruned_loss=0.1251, over 5654818.87 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3658, pruned_loss=0.1189, over 5648870.95 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3775, pruned_loss=0.1257, over 5650963.50 frames. ], batch size: 136, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:14:12,034 INFO [optim.py:369] (1/2) 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 04:14:36,539 INFO [train.py:968] (1/2) Epoch 17, batch 44950, giga_loss[loss=0.3729, simple_loss=0.4092, pruned_loss=0.1683, over 27609.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1246, over 5659666.70 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3654, pruned_loss=0.1188, over 5651475.29 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3755, pruned_loss=0.1253, over 5654249.32 frames. ], batch size: 472, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:15:14,703 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 45000, giga_loss[loss=0.3128, simple_loss=0.3542, pruned_loss=0.1357, over 23477.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3735, pruned_loss=0.1243, over 5654419.18 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3657, pruned_loss=0.1188, over 5640956.72 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3749, pruned_loss=0.1251, over 5658468.25 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:15:20,398 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 04:15:28,821 INFO [train.py:1012] (1/2) Epoch 17, validation: loss=0.2079, simple_loss=0.3159, pruned_loss=0.05001, over 944034.00 frames. 
+2023-03-09 04:15:28,821 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 04:15:42,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=775894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:15:50,831 INFO [optim.py:369] (1/2) 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 04:15:53,445 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=775904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:16:11,404 INFO [train.py:968] (1/2) Epoch 17, batch 45050, giga_loss[loss=0.3361, simple_loss=0.3877, pruned_loss=0.1423, over 26651.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1226, over 5665374.87 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3661, pruned_loss=0.119, over 5649275.25 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3733, pruned_loss=0.1232, over 5661492.48 frames. ], batch size: 555, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:16:26,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1607, 0.8721, 0.9902, 1.3646], device='cuda:1'), covar=tensor([0.0781, 0.0381, 0.0350, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 04:16:47,167 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 04:16:54,450 INFO [train.py:968] (1/2) Epoch 17, batch 45100, giga_loss[loss=0.2663, simple_loss=0.3474, pruned_loss=0.09261, over 29105.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3683, pruned_loss=0.1185, over 5665981.60 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3659, pruned_loss=0.1187, over 5651774.10 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3694, pruned_loss=0.1193, over 5660641.97 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:17:18,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4999, 4.4361, 1.6960, 1.6756], device='cuda:1'), covar=tensor([0.0978, 0.0323, 0.0883, 0.1331], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0544, 0.0372, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 04:17:18,717 INFO [optim.py:369] (1/2) 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 04:17:35,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:17:37,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:17:41,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4427, 2.0438, 1.4529, 0.7122], device='cuda:1'), covar=tensor([0.4122, 0.2188, 0.3171, 0.4283], device='cuda:1'), in_proj_covar=tensor([0.1690, 0.1606, 0.1569, 0.1380], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 04:17:45,833 INFO [train.py:968] (1/2) Epoch 17, batch 45150, giga_loss[loss=0.3293, simple_loss=0.3698, pruned_loss=0.1444, over 23667.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3668, pruned_loss=0.1171, over 5665454.94 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3654, pruned_loss=0.1183, over 5655255.64 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3682, pruned_loss=0.1181, over 5658029.38 frames. ], batch size: 705, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:17:54,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:17:58,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:18:04,594 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:18:25,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:18:35,632 INFO [train.py:968] (1/2) Epoch 17, batch 45200, giga_loss[loss=0.2804, simple_loss=0.3486, pruned_loss=0.1061, over 28992.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3661, pruned_loss=0.1178, over 5648805.75 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3656, pruned_loss=0.1185, over 5647519.65 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1185, over 5649373.65 frames. ], batch size: 155, lr: 1.86e-03, grad_scale: 8.0
+2023-03-09 04:18:46,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-09 04:18:55,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:19:04,675 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 17, batch 45250, giga_loss[loss=0.2812, simple_loss=0.3547, pruned_loss=0.1039, over 28924.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3642, pruned_loss=0.1175, over 5644386.16 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1183, over 5652317.73 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.365, pruned_loss=0.1182, over 5640416.32 frames. ], batch size: 199, lr: 1.86e-03, grad_scale: 4.0
+2023-03-09 04:19:32,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:20:11,146 INFO [train.py:968] (1/2) Epoch 17, batch 45300, giga_loss[loss=0.2961, simple_loss=0.3702, pruned_loss=0.111, over 28980.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3662, pruned_loss=0.1187, over 5644873.17 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5657327.56 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3671, pruned_loss=0.1197, over 5636775.29 frames. ], batch size: 186, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:20:32,893 INFO [optim.py:369] (1/2) 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 04:20:54,976 INFO [train.py:968] (1/2) Epoch 17, batch 45350, giga_loss[loss=0.338, simple_loss=0.3974, pruned_loss=0.1393, over 28678.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3685, pruned_loss=0.1196, over 5650703.86 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3664, pruned_loss=0.1187, over 5658740.85 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3682, pruned_loss=0.1195, over 5642712.85 frames. ], batch size: 262, lr: 1.86e-03, grad_scale: 2.0
+2023-03-09 04:21:34,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3662, 1.2559, 1.2680, 1.4142], device='cuda:1'), covar=tensor([0.0800, 0.0363, 0.0331, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 04:21:45,289 INFO [train.py:968] (1/2) Epoch 17, batch 45400, giga_loss[loss=0.3477, simple_loss=0.3863, pruned_loss=0.1546, over 23413.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3688, pruned_loss=0.1201, over 5628954.38 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3666, pruned_loss=0.1189, over 5653102.86 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3685, pruned_loss=0.1199, over 5627181.26 frames. ], batch size: 705, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:21:45,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=776279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:22:02,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6411, 1.7630, 1.1620, 1.3877], device='cuda:1'), covar=tensor([0.0925, 0.0602, 0.1094, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0444, 0.0509, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:22:07,777 INFO [optim.py:369] (1/2) 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 04:22:27,341 INFO [train.py:968] (1/2) Epoch 17, batch 45450, giga_loss[loss=0.282, simple_loss=0.3568, pruned_loss=0.1036, over 28135.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3689, pruned_loss=0.1202, over 5635820.45 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3667, pruned_loss=0.1188, over 5658788.52 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3686, pruned_loss=0.1203, over 5627839.25 frames. ], batch size: 77, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:22:51,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1892, 0.8289, 0.8675, 1.4300], device='cuda:1'), covar=tensor([0.0760, 0.0367, 0.0344, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 04:23:00,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-09 04:23:10,886 INFO [train.py:968] (1/2) Epoch 17, batch 45500, giga_loss[loss=0.2806, simple_loss=0.348, pruned_loss=0.1067, over 28889.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.37, pruned_loss=0.1214, over 5637817.70 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3661, pruned_loss=0.1184, over 5655970.64 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5633699.55 frames. ], batch size: 112, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:23:36,879 INFO [optim.py:369] (1/2) 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 04:23:56,315 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:23:59,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 17, batch 45550, giga_loss[loss=0.3093, simple_loss=0.3796, pruned_loss=0.1195, over 28875.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3734, pruned_loss=0.124, over 5642665.19 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3662, pruned_loss=0.1187, over 5649762.71 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3736, pruned_loss=0.1241, over 5645041.61 frames. ], batch size: 199, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:24:23,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:24:26,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4241, 1.5201, 1.3609, 1.5679], device='cuda:1'), covar=tensor([0.0611, 0.0286, 0.0278, 0.0619], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 04:24:39,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=776472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:24:46,610 INFO [train.py:968] (1/2) Epoch 17, batch 45600, libri_loss[loss=0.3173, simple_loss=0.3845, pruned_loss=0.125, over 29082.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1248, over 5629604.27 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3668, pruned_loss=0.119, over 5628779.50 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5649369.92 frames. ], batch size: 101, lr: 1.85e-03, grad_scale: 4.0
+2023-03-09 04:25:09,088 INFO [optim.py:369] (1/2) 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 04:25:17,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=776510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:25:38,938 INFO [train.py:968] (1/2) Epoch 17, batch 45650, giga_loss[loss=0.2726, simple_loss=0.3453, pruned_loss=0.09992, over 28888.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3765, pruned_loss=0.1263, over 5633728.33 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.367, pruned_loss=0.119, over 5627392.48 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3764, pruned_loss=0.1265, over 5650636.51 frames. ], batch size: 199, lr: 1.85e-03, grad_scale: 4.0
+2023-03-09 04:26:29,396 INFO [train.py:968] (1/2) Epoch 17, batch 45700, giga_loss[loss=0.3222, simple_loss=0.3862, pruned_loss=0.1291, over 28474.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3773, pruned_loss=0.127, over 5634723.60 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3668, pruned_loss=0.1189, over 5630157.35 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3776, pruned_loss=0.1274, over 5645673.21 frames. ], batch size: 71, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:26:36,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0210, 1.0822, 3.3485, 2.8965], device='cuda:1'), covar=tensor([0.1725, 0.2851, 0.0560, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0631, 0.0931, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:26:36,637 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-09 04:26:52,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4233, 1.7111, 1.3115, 1.5176], device='cuda:1'), covar=tensor([0.2830, 0.2699, 0.3162, 0.2370], device='cuda:1'), in_proj_covar=tensor([0.1442, 0.1046, 0.1277, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 04:26:58,435 INFO [optim.py:369] (1/2) 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 04:27:11,415 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:13,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:24,881 INFO [train.py:968] (1/2) Epoch 17, batch 45750, giga_loss[loss=0.2888, simple_loss=0.3416, pruned_loss=0.118, over 23736.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3765, pruned_loss=0.1245, over 5641754.46 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3667, pruned_loss=0.1189, over 5634490.49 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.377, pruned_loss=0.1248, over 5646658.32 frames. ], batch size: 705, lr: 1.85e-03, grad_scale: 2.0
+2023-03-09 04:27:40,563 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:48,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=776653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:50,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=776656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:27:53,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3511, 2.0424, 1.5637, 0.5741], device='cuda:1'), covar=tensor([0.3703, 0.2227, 0.2941, 0.4286], device='cuda:1'), in_proj_covar=tensor([0.1691, 0.1607, 0.1573, 0.1381], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 04:28:11,833 INFO [train.py:968] (1/2) Epoch 17, batch 45800, giga_loss[loss=0.2687, simple_loss=0.3349, pruned_loss=0.1013, over 28964.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3781, pruned_loss=0.1264, over 5580682.59 frames. ], libri_tot_loss[loss=0.3032, simple_loss=0.3673, pruned_loss=0.1196, over 5574168.09 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3782, pruned_loss=0.1262, over 5639491.45 frames. ], batch size: 106, lr: 1.85e-03, grad_scale: 1.0
+2023-03-09 04:28:17,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=776685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:28:32,423 INFO [optim.py:369] (1/2) 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 04:28:37,314 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-09 04:29:59,688 INFO [train.py:968] (1/2) Epoch 18, batch 50, libri_loss[loss=0.2523, simple_loss=0.3313, pruned_loss=0.08662, over 29548.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3762, pruned_loss=0.1111, over 1268334.09 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3325, pruned_loss=0.08423, over 146400.57 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3816, pruned_loss=0.1144, over 1150694.68 frames. ], batch size: 75, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:30:25,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:30:41,141 INFO [optim.py:369] (1/2) 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 04:30:46,682 INFO [train.py:968] (1/2) Epoch 18, batch 100, giga_loss[loss=0.3145, simple_loss=0.3801, pruned_loss=0.1245, over 28990.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3701, pruned_loss=0.1104, over 2234487.30 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3368, pruned_loss=0.08947, over 278589.95 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3741, pruned_loss=0.1128, over 2057797.43 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:31:18,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=776847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:31:31,120 INFO [train.py:968] (1/2) Epoch 18, batch 150, giga_loss[loss=0.242, simple_loss=0.3222, pruned_loss=0.08086, over 28640.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3534, pruned_loss=0.1015, over 3005124.88 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3364, pruned_loss=0.08872, over 418282.58 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.356, pruned_loss=0.1032, over 2790548.87 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:31:51,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6438, 1.7219, 1.9153, 1.4556], device='cuda:1'), covar=tensor([0.1796, 0.2419, 0.1421, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.0878, 0.0701, 0.0924, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 04:32:03,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-09 04:32:10,514 INFO [optim.py:369] (1/2) 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 04:32:14,771 INFO [train.py:968] (1/2) Epoch 18, batch 200, giga_loss[loss=0.2461, simple_loss=0.3117, pruned_loss=0.09028, over 28720.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3388, pruned_loss=0.09404, over 3603577.37 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3352, pruned_loss=0.08752, over 523408.28 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.34, pruned_loss=0.09519, over 3390385.45 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:32:55,615 INFO [train.py:968] (1/2) Epoch 18, batch 250, giga_loss[loss=0.2233, simple_loss=0.2987, pruned_loss=0.074, over 29016.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3296, pruned_loss=0.08958, over 4066999.98 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3403, pruned_loss=0.09045, over 681694.62 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3291, pruned_loss=0.0899, over 3841803.54 frames. ], batch size: 128, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:33:32,333 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 300, giga_loss[loss=0.2155, simple_loss=0.2861, pruned_loss=0.07244, over 28827.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3212, pruned_loss=0.08589, over 4427632.75 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3407, pruned_loss=0.09019, over 860284.69 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3198, pruned_loss=0.08591, over 4195734.63 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:33:58,555 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=777032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:34:26,522 INFO [train.py:968] (1/2) Epoch 18, batch 350, giga_loss[loss=0.2191, simple_loss=0.2911, pruned_loss=0.07353, over 29012.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3134, pruned_loss=0.08234, over 4696870.58 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3423, pruned_loss=0.09106, over 926118.79 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3114, pruned_loss=0.08194, over 4504530.06 frames. ], batch size: 227, lr: 1.80e-03, grad_scale: 1.0
+2023-03-09 04:34:26,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4281, 4.2547, 4.0319, 1.8176], device='cuda:1'), covar=tensor([0.0505, 0.0686, 0.0687, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1163, 0.1081, 0.0922, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 04:35:03,273 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 400, giga_loss[loss=0.2045, simple_loss=0.2741, pruned_loss=0.06741, over 28622.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3082, pruned_loss=0.07926, over 4919108.46 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3422, pruned_loss=0.09046, over 1042280.25 frames. ], giga_tot_loss[loss=0.2315, simple_loss=0.3056, pruned_loss=0.07868, over 4747656.57 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:35:47,650 INFO [train.py:968] (1/2) Epoch 18, batch 450, giga_loss[loss=0.2099, simple_loss=0.2733, pruned_loss=0.07328, over 28548.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.306, pruned_loss=0.0782, over 5097510.75 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3403, pruned_loss=0.08911, over 1186558.18 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.3032, pruned_loss=0.07764, over 4935862.86 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:35:51,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 18, batch 500, giga_loss[loss=0.1843, simple_loss=0.2689, pruned_loss=0.04986, over 28730.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3035, pruned_loss=0.07709, over 5225141.25 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3403, pruned_loss=0.08911, over 1256805.12 frames. ], giga_tot_loss[loss=0.2269, simple_loss=0.3008, pruned_loss=0.07646, over 5087731.26 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:36:44,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=777222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:36:52,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2711, 2.9934, 1.4349, 1.4211], device='cuda:1'), covar=tensor([0.1028, 0.0342, 0.0923, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0539, 0.0370, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 04:37:19,621 INFO [train.py:968] (1/2) Epoch 18, batch 550, giga_loss[loss=0.194, simple_loss=0.2729, pruned_loss=0.05755, over 28849.00 frames. ], tot_loss[loss=0.229, simple_loss=0.303, pruned_loss=0.07747, over 5332467.78 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3427, pruned_loss=0.09097, over 1371596.51 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.2992, pruned_loss=0.0762, over 5208827.44 frames. ], batch size: 145, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:37:42,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9538, 1.1940, 1.1963, 0.9330], device='cuda:1'), covar=tensor([0.1918, 0.1970, 0.1130, 0.1583], device='cuda:1'), in_proj_covar=tensor([0.1896, 0.1824, 0.1752, 0.1892], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 04:37:47,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5733, 1.6887, 1.1891, 1.3116], device='cuda:1'), covar=tensor([0.0847, 0.0569, 0.1027, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0378, 0.0441, 0.0509, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:37:58,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=777306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:37:59,155 INFO [optim.py:369] (1/2) 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 04:38:00,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=777309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:38:02,850 INFO [train.py:968] (1/2) Epoch 18, batch 600, giga_loss[loss=0.2068, simple_loss=0.283, pruned_loss=0.06528, over 28834.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3013, pruned_loss=0.07668, over 5403865.18 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3431, pruned_loss=0.09197, over 1515877.46 frames. ], giga_tot_loss[loss=0.2234, simple_loss=0.2969, pruned_loss=0.07492, over 5299079.27 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:38:17,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6689, 1.7372, 1.3694, 1.4027], device='cuda:1'), covar=tensor([0.0918, 0.0685, 0.0992, 0.1388], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0441, 0.0510, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:38:31,000 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=777338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:38:52,519 INFO [train.py:968] (1/2) Epoch 18, batch 650, giga_loss[loss=0.2442, simple_loss=0.3014, pruned_loss=0.0935, over 28516.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2995, pruned_loss=0.07592, over 5461658.77 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3425, pruned_loss=0.09162, over 1611803.75 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2951, pruned_loss=0.07424, over 5379433.08 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:38:52,989 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-09 04:38:54,710 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=777365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:38:56,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2093, 1.2520, 1.1348, 0.9679], device='cuda:1'), covar=tensor([0.0940, 0.0556, 0.1130, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0379, 0.0441, 0.0510, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:38:56,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=777368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:39:22,515 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:39:32,854 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 700, giga_loss[loss=0.2077, simple_loss=0.2821, pruned_loss=0.06668, over 28936.00 frames. ], tot_loss[loss=0.223, simple_loss=0.297, pruned_loss=0.07446, over 5517491.42 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3433, pruned_loss=0.09222, over 1718150.63 frames. ], giga_tot_loss[loss=0.2187, simple_loss=0.2923, pruned_loss=0.07258, over 5442481.57 frames. ], batch size: 213, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:39:51,910 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3217, 1.3524, 4.1021, 3.3121], device='cuda:1'), covar=tensor([0.1630, 0.2685, 0.0434, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0726, 0.0627, 0.0926, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:40:19,800 INFO [train.py:968] (1/2) Epoch 18, batch 750, giga_loss[loss=0.1969, simple_loss=0.2685, pruned_loss=0.06268, over 28699.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.295, pruned_loss=0.0732, over 5554089.45 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3439, pruned_loss=0.09212, over 1841532.68 frames. ], giga_tot_loss[loss=0.216, simple_loss=0.2896, pruned_loss=0.07117, over 5484636.05 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 04:40:59,084 INFO [optim.py:369] (1/2) 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 04:41:02,500 INFO [train.py:968] (1/2) Epoch 18, batch 800, giga_loss[loss=0.1852, simple_loss=0.2591, pruned_loss=0.05562, over 29006.00 frames. ], tot_loss[loss=0.221, simple_loss=0.2952, pruned_loss=0.07345, over 5584420.68 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3446, pruned_loss=0.09181, over 2014000.17 frames. ], giga_tot_loss[loss=0.2153, simple_loss=0.2884, pruned_loss=0.0711, over 5520289.01 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:41:07,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3626, 1.6222, 1.5156, 1.4616], device='cuda:1'), covar=tensor([0.1922, 0.1687, 0.1940, 0.1789], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0744, 0.0700, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 04:41:40,838 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=777550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:41:43,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=777553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:41:52,267 INFO [train.py:968] (1/2) Epoch 18, batch 850, libri_loss[loss=0.3112, simple_loss=0.383, pruned_loss=0.1197, over 27963.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3024, pruned_loss=0.07742, over 5606362.90 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3451, pruned_loss=0.09215, over 2089919.38 frames. ], giga_tot_loss[loss=0.2232, simple_loss=0.296, pruned_loss=0.07515, over 5550951.48 frames. ], batch size: 116, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:42:10,815 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=777582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:42:16,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5335, 4.4492, 1.6735, 1.9447], device='cuda:1'), covar=tensor([0.1019, 0.0281, 0.0943, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0537, 0.0369, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 04:42:34,381 INFO [optim.py:369] (1/2) 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 04:42:37,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=777610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:42:39,186 INFO [train.py:968] (1/2) Epoch 18, batch 900, giga_loss[loss=0.2633, simple_loss=0.3426, pruned_loss=0.09205, over 28984.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3152, pruned_loss=0.08376, over 5628395.33 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3459, pruned_loss=0.09221, over 2201074.14 frames. ], giga_tot_loss[loss=0.2361, simple_loss=0.3089, pruned_loss=0.08166, over 5577694.92 frames. ], batch size: 112, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:42:40,694 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 04:42:41,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3962, 1.5129, 1.5449, 1.3978], device='cuda:1'), covar=tensor([0.2140, 0.2066, 0.1544, 0.1952], device='cuda:1'), in_proj_covar=tensor([0.1886, 0.1811, 0.1740, 0.1884], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 04:43:23,017 INFO [train.py:968] (1/2) Epoch 18, batch 950, giga_loss[loss=0.2514, simple_loss=0.3317, pruned_loss=0.08552, over 28565.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3259, pruned_loss=0.08922, over 5639475.21 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3445, pruned_loss=0.09157, over 2274057.32 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.321, pruned_loss=0.08771, over 5595022.10 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:43:36,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-09 04:43:59,816 INFO [optim.py:369] (1/2) 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 04:44:03,686 INFO [train.py:968] (1/2) Epoch 18, batch 1000, giga_loss[loss=0.2611, simple_loss=0.3487, pruned_loss=0.08676, over 28590.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3328, pruned_loss=0.09147, over 5657141.21 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3445, pruned_loss=0.09151, over 2362233.44 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3286, pruned_loss=0.09029, over 5617384.88 frames. ], batch size: 307, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:44:43,749 INFO [train.py:968] (1/2) Epoch 18, batch 1050, giga_loss[loss=0.3106, simple_loss=0.3899, pruned_loss=0.1157, over 28745.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3373, pruned_loss=0.09215, over 5658585.51 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3457, pruned_loss=0.09218, over 2405182.60 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3334, pruned_loss=0.09097, over 5633480.87 frames. ], batch size: 262, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:45:13,749 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2680, 1.2814, 3.6827, 3.0373], device='cuda:1'), covar=tensor([0.1571, 0.2851, 0.0425, 0.2181], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0625, 0.0924, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:45:28,185 INFO [optim.py:369] (1/2) 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 04:45:32,019 INFO [train.py:968] (1/2) Epoch 18, batch 1100, giga_loss[loss=0.3228, simple_loss=0.3976, pruned_loss=0.124, over 27879.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3388, pruned_loss=0.09231, over 5656994.36 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3449, pruned_loss=0.09161, over 2505901.89 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3359, pruned_loss=0.0916, over 5632750.52 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:46:06,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3305, 1.5157, 1.4837, 1.4613], device='cuda:1'), covar=tensor([0.1648, 0.1749, 0.2047, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0742, 0.0698, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 04:46:16,347 INFO [train.py:968] (1/2) Epoch 18, batch 1150, giga_loss[loss=0.2803, simple_loss=0.3577, pruned_loss=0.1014, over 28901.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3411, pruned_loss=0.09386, over 5660343.77 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3447, pruned_loss=0.09141, over 2539829.67 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3389, pruned_loss=0.0934, over 5639072.25 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:46:20,975 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=777867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:46:30,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4953, 1.7414, 1.4387, 1.4796], device='cuda:1'), covar=tensor([0.2451, 0.2400, 0.2642, 0.2204], device='cuda:1'), in_proj_covar=tensor([0.1448, 0.1050, 0.1283, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 04:46:55,923 INFO [optim.py:369] (1/2) 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 04:47:00,911 INFO [train.py:968] (1/2) Epoch 18, batch 1200, giga_loss[loss=0.2971, simple_loss=0.373, pruned_loss=0.1106, over 28961.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3441, pruned_loss=0.09634, over 5671782.25 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3447, pruned_loss=0.09133, over 2606597.44 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3422, pruned_loss=0.09609, over 5650585.52 frames. ], batch size: 145, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:47:30,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5287, 2.1899, 1.5766, 0.6988], device='cuda:1'), covar=tensor([0.5835, 0.2975, 0.4348, 0.6129], device='cuda:1'), in_proj_covar=tensor([0.1685, 0.1595, 0.1569, 0.1375], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 04:47:43,014 INFO [train.py:968] (1/2) Epoch 18, batch 1250, giga_loss[loss=0.259, simple_loss=0.3397, pruned_loss=0.08913, over 28986.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3467, pruned_loss=0.09805, over 5681256.97 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3443, pruned_loss=0.09092, over 2718518.43 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3454, pruned_loss=0.09822, over 5658641.70 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:48:01,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=777985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:48:24,050 INFO [optim.py:369] (1/2) 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 04:48:28,309 INFO [train.py:968] (1/2) Epoch 18, batch 1300, giga_loss[loss=0.3147, simple_loss=0.3843, pruned_loss=0.1226, over 28838.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.35, pruned_loss=0.09908, over 5688064.40 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3443, pruned_loss=0.09093, over 2749719.43 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.349, pruned_loss=0.09926, over 5668962.54 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:48:41,493 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5968, 1.9088, 1.7595, 1.6398], device='cuda:1'), covar=tensor([0.1979, 0.1989, 0.2177, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0739, 0.0698, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 04:49:08,133 INFO [train.py:968] (1/2) Epoch 18, batch 1350, libri_loss[loss=0.2475, simple_loss=0.312, pruned_loss=0.09151, over 29352.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3512, pruned_loss=0.09907, over 5690473.86 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3448, pruned_loss=0.09134, over 2827353.65 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3503, pruned_loss=0.09922, over 5670736.81 frames. ], batch size: 67, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:49:44,795 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 1400, giga_loss[loss=0.2638, simple_loss=0.3507, pruned_loss=0.08841, over 29057.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3527, pruned_loss=0.0992, over 5684585.98 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3454, pruned_loss=0.09146, over 2904469.62 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.352, pruned_loss=0.09947, over 5676159.85 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:49:49,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1503, 1.3086, 3.6241, 3.0945], device='cuda:1'), covar=tensor([0.1746, 0.2742, 0.0454, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0720, 0.0622, 0.0916, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:50:01,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:50:04,347 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:50:32,953 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 1450, giga_loss[loss=0.2919, simple_loss=0.3599, pruned_loss=0.1119, over 27566.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3523, pruned_loss=0.09805, over 5684302.51 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09148, over 2918607.27 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3517, pruned_loss=0.09829, over 5677527.64 frames. ], batch size: 472, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:50:35,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1352, 1.2495, 3.9520, 3.1336], device='cuda:1'), covar=tensor([0.1835, 0.2929, 0.0400, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0720, 0.0623, 0.0917, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 04:51:09,051 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 1500, giga_loss[loss=0.2545, simple_loss=0.3382, pruned_loss=0.08537, over 28091.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3503, pruned_loss=0.0958, over 5697381.63 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3454, pruned_loss=0.09143, over 3018968.80 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.35, pruned_loss=0.09617, over 5687626.78 frames. ], batch size: 77, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:51:34,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:51:52,736 INFO [train.py:968] (1/2) Epoch 18, batch 1550, giga_loss[loss=0.2916, simple_loss=0.3648, pruned_loss=0.1092, over 28158.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09415, over 5698583.49 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3446, pruned_loss=0.09115, over 3080591.77 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3485, pruned_loss=0.09463, over 5695070.31 frames. ], batch size: 77, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:52:33,130 INFO [optim.py:369] (1/2) 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 04:52:38,323 INFO [train.py:968] (1/2) Epoch 18, batch 1600, giga_loss[loss=0.3125, simple_loss=0.3733, pruned_loss=0.1259, over 28550.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3499, pruned_loss=0.09626, over 5688154.19 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.345, pruned_loss=0.09135, over 3132004.45 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3499, pruned_loss=0.09661, over 5685686.47 frames. ], batch size: 336, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:53:23,175 INFO [train.py:968] (1/2) Epoch 18, batch 1650, giga_loss[loss=0.2658, simple_loss=0.3349, pruned_loss=0.09835, over 28926.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3522, pruned_loss=0.09997, over 5698415.22 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3449, pruned_loss=0.09113, over 3199306.50 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3525, pruned_loss=0.1005, over 5693058.10 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:53:27,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3966, 2.6019, 2.3324, 2.1217], device='cuda:1'), covar=tensor([0.2440, 0.2026, 0.2120, 0.2115], device='cuda:1'), in_proj_covar=tensor([0.1879, 0.1812, 0.1743, 0.1881], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 04:53:41,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=778382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:42,915 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=778385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:42,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:45,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:53:58,766 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 1700, giga_loss[loss=0.3275, simple_loss=0.3815, pruned_loss=0.1368, over 28025.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3536, pruned_loss=0.1024, over 5707324.19 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3441, pruned_loss=0.09059, over 3302254.71 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3545, pruned_loss=0.1035, over 5698859.44 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:54:08,183 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:54:24,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1339, 1.5550, 1.1612, 0.3801], device='cuda:1'), covar=tensor([0.3151, 0.1786, 0.2605, 0.4868], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1579, 0.1560, 0.1362], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 04:54:32,188 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 1750, giga_loss[loss=0.2331, simple_loss=0.3112, pruned_loss=0.07748, over 28227.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3534, pruned_loss=0.1035, over 5701385.96 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3444, pruned_loss=0.09072, over 3379767.92 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3542, pruned_loss=0.1046, over 5689009.68 frames. ], batch size: 77, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:55:18,103 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=778500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:55:18,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1961, 1.4883, 1.4965, 1.1247], device='cuda:1'), covar=tensor([0.1260, 0.1901, 0.1039, 0.1278], device='cuda:1'), in_proj_covar=tensor([0.0880, 0.0699, 0.0930, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 04:55:25,754 INFO [optim.py:369] (1/2) 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 04:55:28,584 INFO [train.py:968] (1/2) Epoch 18, batch 1800, giga_loss[loss=0.2483, simple_loss=0.3296, pruned_loss=0.0835, over 28840.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3514, pruned_loss=0.103, over 5697200.77 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3444, pruned_loss=0.09081, over 3464614.00 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3524, pruned_loss=0.1043, over 5682739.74 frames. ], batch size: 174, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:55:38,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2672, 2.5328, 1.2801, 1.3564], device='cuda:1'), covar=tensor([0.0973, 0.0325, 0.0874, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0540, 0.0370, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 04:55:52,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8583, 1.2580, 1.1897, 1.0567], device='cuda:1'), covar=tensor([0.1657, 0.1085, 0.2024, 0.1292], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0737, 0.0697, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 04:56:12,021 INFO [train.py:968] (1/2) Epoch 18, batch 1850, giga_loss[loss=0.3013, simple_loss=0.3729, pruned_loss=0.1149, over 28719.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3504, pruned_loss=0.1021, over 5693708.32 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3442, pruned_loss=0.09065, over 3511662.93 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3514, pruned_loss=0.1034, over 5679698.36 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:56:55,881 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 1900, giga_loss[loss=0.2424, simple_loss=0.3277, pruned_loss=0.07857, over 28969.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3475, pruned_loss=0.09944, over 5697951.78 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3437, pruned_loss=0.09037, over 3547361.26 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3486, pruned_loss=0.1008, over 5683999.97 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:57:05,534 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.42 vs. limit=5.0
+2023-03-09 04:57:45,466 INFO [train.py:968] (1/2) Epoch 18, batch 1950, giga_loss[loss=0.2632, simple_loss=0.3382, pruned_loss=0.09415, over 28983.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3439, pruned_loss=0.09726, over 5693414.03 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3435, pruned_loss=0.0901, over 3603225.02 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3449, pruned_loss=0.09866, over 5679791.44 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 04:57:53,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2277, 1.4083, 1.3747, 1.1704], device='cuda:1'), covar=tensor([0.2643, 0.2504, 0.1568, 0.2103], device='cuda:1'), in_proj_covar=tensor([0.1883, 0.1819, 0.1746, 0.1886], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 04:58:03,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 04:58:26,750 INFO [optim.py:369] (1/2) 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 04:58:30,027 INFO [train.py:968] (1/2) Epoch 18, batch 2000, giga_loss[loss=0.2197, simple_loss=0.3021, pruned_loss=0.06861, over 28748.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3382, pruned_loss=0.09399, over 5689240.03 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08997, over 3681983.65 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3391, pruned_loss=0.09538, over 5672215.13 frames. ], batch size: 262, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 04:59:03,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2815, 1.1762, 1.1280, 1.4982], device='cuda:1'), covar=tensor([0.0804, 0.0363, 0.0354, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 04:59:13,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:59:15,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 04:59:16,650 INFO [train.py:968] (1/2) Epoch 18, batch 2050, giga_loss[loss=0.2561, simple_loss=0.3213, pruned_loss=0.0954, over 27920.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3336, pruned_loss=0.09156, over 5687245.69 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3444, pruned_loss=0.0906, over 3725522.86 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3335, pruned_loss=0.09235, over 5670193.01 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:00:01,329 INFO [optim.py:369] (1/2) 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 05:00:03,684 INFO [train.py:968] (1/2) Epoch 18, batch 2100, libri_loss[loss=0.3266, simple_loss=0.3965, pruned_loss=0.1284, over 29521.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3325, pruned_loss=0.09144, over 5661086.50 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3449, pruned_loss=0.0909, over 3808020.03 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3316, pruned_loss=0.09191, over 5649630.70 frames. ], batch size: 84, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:00:10,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:00:44,825 INFO [train.py:968] (1/2) Epoch 18, batch 2150, giga_loss[loss=0.2704, simple_loss=0.3456, pruned_loss=0.09758, over 28955.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3335, pruned_loss=0.09104, over 5675358.91 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3449, pruned_loss=0.09072, over 3849342.33 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3325, pruned_loss=0.09153, over 5662260.63 frames. ], batch size: 106, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:00:56,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=778875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:16,202 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:19,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:19,192 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:22,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:24,057 INFO [optim.py:369] (1/2) 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 05:01:26,601 INFO [train.py:968] (1/2) Epoch 18, batch 2200, giga_loss[loss=0.2571, simple_loss=0.3256, pruned_loss=0.09432, over 28571.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3331, pruned_loss=0.0905, over 5689675.69 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.345, pruned_loss=0.09059, over 3888994.65 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.332, pruned_loss=0.09095, over 5676438.61 frames. ], batch size: 78, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:01:44,896 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:01:47,267 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=778935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:02:04,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-09 05:02:07,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8963, 1.6007, 5.2273, 3.7901], device='cuda:1'), covar=tensor([0.1569, 0.2707, 0.0347, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0622, 0.0916, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 05:02:09,620 INFO [train.py:968] (1/2) Epoch 18, batch 2250, giga_loss[loss=0.2381, simple_loss=0.3141, pruned_loss=0.08109, over 28693.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3309, pruned_loss=0.08944, over 5695913.14 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3451, pruned_loss=0.09058, over 3938620.47 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3297, pruned_loss=0.08979, over 5680770.69 frames. ], batch size: 262, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:02:09,945 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=778962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:02:12,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=778965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:02:17,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3572, 1.5847, 1.5062, 1.1728], device='cuda:1'), covar=tensor([0.3600, 0.2528, 0.1900, 0.2886], device='cuda:1'), in_proj_covar=tensor([0.1875, 0.1805, 0.1739, 0.1878], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 05:02:36,604 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 18, batch 2300, giga_loss[loss=0.2151, simple_loss=0.2911, pruned_loss=0.06952, over 29121.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3289, pruned_loss=0.08865, over 5706086.77 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3454, pruned_loss=0.09061, over 3967971.02 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3276, pruned_loss=0.08888, over 5691422.34 frames. ], batch size: 113, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:02:58,979 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=779018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:03:01,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=779021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:03:16,438 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 05:03:24,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=779050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:03:33,574 INFO [train.py:968] (1/2) Epoch 18, batch 2350, giga_loss[loss=0.2199, simple_loss=0.2878, pruned_loss=0.07597, over 28684.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3263, pruned_loss=0.0877, over 5706774.94 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3462, pruned_loss=0.09114, over 3997220.36 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3245, pruned_loss=0.08753, over 5692342.60 frames. ], batch size: 60, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:03:56,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2698, 1.5264, 1.6256, 1.3633], device='cuda:1'), covar=tensor([0.2064, 0.1773, 0.2330, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0742, 0.0700, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:04:17,395 INFO [optim.py:369] (1/2) 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 05:04:19,477 INFO [train.py:968] (1/2) Epoch 18, batch 2400, giga_loss[loss=0.2379, simple_loss=0.3166, pruned_loss=0.07959, over 28491.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3237, pruned_loss=0.0866, over 5702426.83 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3464, pruned_loss=0.09112, over 4021982.65 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3218, pruned_loss=0.08643, over 5691577.24 frames. ], batch size: 336, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:04:57,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7168, 2.0144, 1.9575, 1.6209], device='cuda:1'), covar=tensor([0.3178, 0.2295, 0.2023, 0.2709], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1795, 0.1731, 0.1872], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 05:04:58,165 INFO [train.py:968] (1/2) Epoch 18, batch 2450, libri_loss[loss=0.268, simple_loss=0.3575, pruned_loss=0.08929, over 29518.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3217, pruned_loss=0.08541, over 5713393.32 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3468, pruned_loss=0.09113, over 4068770.78 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3194, pruned_loss=0.08516, over 5700074.22 frames. ], batch size: 81, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:05:20,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:05:32,825 INFO [optim.py:369] (1/2) 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 05:05:34,108 INFO [train.py:968] (1/2) Epoch 18, batch 2500, giga_loss[loss=0.2634, simple_loss=0.3277, pruned_loss=0.09959, over 28919.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3222, pruned_loss=0.0859, over 5723360.46 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3481, pruned_loss=0.09165, over 4158282.91 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3184, pruned_loss=0.08509, over 5704309.23 frames. ], batch size: 186, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:05:37,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9510, 1.9884, 1.6753, 2.2768], device='cuda:1'), covar=tensor([0.2302, 0.2532, 0.2802, 0.2220], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1047, 0.1282, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 05:06:17,054 INFO [train.py:968] (1/2) Epoch 18, batch 2550, giga_loss[loss=0.2255, simple_loss=0.3033, pruned_loss=0.07386, over 28527.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3204, pruned_loss=0.08479, over 5728475.17 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3482, pruned_loss=0.09151, over 4190788.23 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3167, pruned_loss=0.08411, over 5712292.44 frames. ], batch size: 60, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:06:55,644 INFO [optim.py:369] (1/2) 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 05:06:57,931 INFO [train.py:968] (1/2) Epoch 18, batch 2600, giga_loss[loss=0.2869, simple_loss=0.3472, pruned_loss=0.1134, over 26595.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.319, pruned_loss=0.08421, over 5728927.46 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3485, pruned_loss=0.09161, over 4233297.69 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3152, pruned_loss=0.08343, over 5711960.64 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:07:33,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1989, 2.4309, 2.1692, 2.1172], device='cuda:1'), covar=tensor([0.2180, 0.2587, 0.2390, 0.2466], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0747, 0.0704, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:07:37,277 INFO [train.py:968] (1/2) Epoch 18, batch 2650, giga_loss[loss=0.2454, simple_loss=0.318, pruned_loss=0.08636, over 28744.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3183, pruned_loss=0.08373, over 5734807.68 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3489, pruned_loss=0.09171, over 4274466.18 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.3143, pruned_loss=0.08283, over 5717497.88 frames. ], batch size: 284, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:07:53,291 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 05:08:17,161 INFO [optim.py:369] (1/2) 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 05:08:19,362 INFO [train.py:968] (1/2) Epoch 18, batch 2700, giga_loss[loss=0.3519, simple_loss=0.4003, pruned_loss=0.1518, over 26464.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3219, pruned_loss=0.08599, over 5726023.76 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3493, pruned_loss=0.09179, over 4312572.98 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3176, pruned_loss=0.08503, over 5716943.98 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:08:41,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:08:45,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-09 05:09:02,890 INFO [train.py:968] (1/2) Epoch 18, batch 2750, giga_loss[loss=0.244, simple_loss=0.3205, pruned_loss=0.08378, over 29035.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3266, pruned_loss=0.08888, over 5718983.39 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3499, pruned_loss=0.09213, over 4352818.62 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3221, pruned_loss=0.08776, over 5712988.94 frames. ], batch size: 155, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:09:49,551 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 2800, giga_loss[loss=0.3196, simple_loss=0.3805, pruned_loss=0.1294, over 28325.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3337, pruned_loss=0.09349, over 5713104.47 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.35, pruned_loss=0.09218, over 4376199.96 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3299, pruned_loss=0.09256, over 5705590.87 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:10:10,395 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:10:37,969 INFO [train.py:968] (1/2) Epoch 18, batch 2850, giga_loss[loss=0.2534, simple_loss=0.3323, pruned_loss=0.0872, over 28759.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3398, pruned_loss=0.09733, over 5696070.46 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3496, pruned_loss=0.09223, over 4403120.45 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3367, pruned_loss=0.09665, over 5695245.06 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:10:43,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=779569.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:11:24,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2907, 3.0766, 2.9002, 1.3515], device='cuda:1'), covar=tensor([0.0965, 0.1134, 0.1043, 0.2515], device='cuda:1'), in_proj_covar=tensor([0.1147, 0.1058, 0.0910, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 05:11:25,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-09 05:11:25,369 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 2900, giga_loss[loss=0.2894, simple_loss=0.372, pruned_loss=0.1034, over 28570.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3439, pruned_loss=0.09826, over 5702398.76 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3503, pruned_loss=0.09271, over 4446042.82 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3407, pruned_loss=0.09755, over 5703130.01 frames. ], batch size: 307, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:12:13,299 INFO [train.py:968] (1/2) Epoch 18, batch 2950, giga_loss[loss=0.2996, simple_loss=0.3769, pruned_loss=0.1112, over 29035.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3495, pruned_loss=0.1012, over 5702450.03 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3504, pruned_loss=0.09281, over 4481804.15 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3468, pruned_loss=0.1007, over 5698770.98 frames. ], batch size: 155, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:12:36,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5331, 1.7918, 1.4773, 1.6879], device='cuda:1'), covar=tensor([0.2113, 0.2013, 0.2150, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1038, 0.1273, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 05:12:44,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2651, 1.4539, 3.0871, 2.8533], device='cuda:1'), covar=tensor([0.1284, 0.2355, 0.0465, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0618, 0.0910, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-09 05:13:02,200 INFO [optim.py:369] (1/2) 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 05:13:05,248 INFO [train.py:968] (1/2) Epoch 18, batch 3000, giga_loss[loss=0.2831, simple_loss=0.3605, pruned_loss=0.1028, over 28839.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3559, pruned_loss=0.1058, over 5681080.29 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3496, pruned_loss=0.09237, over 4495059.24 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3543, pruned_loss=0.1058, over 5677188.13 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:13:05,248 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 05:13:13,705 INFO [train.py:1012] (1/2) Epoch 18, validation: loss=0.2163, simple_loss=0.322, pruned_loss=0.0553, over 944034.00 frames. 
+2023-03-09 05:13:13,705 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 05:13:13,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=779712.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:13:16,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=779715.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:13:25,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5181, 2.2434, 1.6230, 0.6573], device='cuda:1'), covar=tensor([0.4792, 0.2812, 0.4139, 0.5679], device='cuda:1'), in_proj_covar=tensor([0.1676, 0.1585, 0.1566, 0.1373], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 05:13:41,100 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=779744.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:13:55,376 INFO [train.py:968] (1/2) Epoch 18, batch 3050, giga_loss[loss=0.2661, simple_loss=0.3419, pruned_loss=0.0952, over 28796.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3543, pruned_loss=0.1039, over 5688477.10 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.35, pruned_loss=0.09263, over 4531968.28 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.353, pruned_loss=0.1042, over 5683957.40 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:14:18,638 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:14:36,983 INFO [optim.py:369] (1/2) 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 05:14:37,731 INFO [train.py:968] (1/2) Epoch 18, batch 3100, giga_loss[loss=0.2709, simple_loss=0.3474, pruned_loss=0.09726, over 28700.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3514, pruned_loss=0.1015, over 5698897.65 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3499, pruned_loss=0.09284, over 4572814.07 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3505, pruned_loss=0.1018, over 5689545.09 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:14:37,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=779812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:14:49,881 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:15:24,167 INFO [train.py:968] (1/2) Epoch 18, batch 3150, giga_loss[loss=0.2342, simple_loss=0.3223, pruned_loss=0.07305, over 28951.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3484, pruned_loss=0.09888, over 5708220.60 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3494, pruned_loss=0.09266, over 4592781.92 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.348, pruned_loss=0.09933, over 5698245.38 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:15:28,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4676, 1.7396, 1.5785, 1.5526], device='cuda:1'), covar=tensor([0.1852, 0.2166, 0.2279, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0742, 0.0698, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:16:00,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5959, 1.7508, 1.5891, 1.5208], device='cuda:1'), covar=tensor([0.1829, 0.2681, 0.2298, 0.2426], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0744, 0.0699, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:16:03,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=779906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:16:07,197 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 3200, giga_loss[loss=0.2675, simple_loss=0.3531, pruned_loss=0.09098, over 29061.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3484, pruned_loss=0.09836, over 5711132.38 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.349, pruned_loss=0.09249, over 4623782.21 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3484, pruned_loss=0.09899, over 5699974.14 frames. ], batch size: 128, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:16:08,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5616, 2.2893, 1.8241, 0.7487], device='cuda:1'), covar=tensor([0.6193, 0.2787, 0.3738, 0.6274], device='cuda:1'), in_proj_covar=tensor([0.1680, 0.1587, 0.1572, 0.1376], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 05:16:10,925 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=779916.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:16:45,143 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=779955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:16:47,965 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 3250, giga_loss[loss=0.2979, simple_loss=0.3662, pruned_loss=0.1148, over 28917.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3508, pruned_loss=0.09962, over 5714536.33 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3488, pruned_loss=0.09234, over 4642971.90 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3509, pruned_loss=0.1003, over 5703259.64 frames. ], batch size: 213, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:17:12,100 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=779987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:17:17,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3119, 1.4752, 1.4385, 1.2034], device='cuda:1'), covar=tensor([0.2622, 0.2524, 0.1874, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.1877, 0.1803, 0.1738, 0.1881], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 05:17:32,538 INFO [optim.py:369] (1/2) 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 05:17:32,775 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 3300, giga_loss[loss=0.2458, simple_loss=0.3275, pruned_loss=0.08205, over 28682.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3521, pruned_loss=0.1006, over 5710573.78 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3486, pruned_loss=0.0922, over 4676499.17 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3523, pruned_loss=0.1014, over 5700302.98 frames. ], batch size: 60, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:18:07,831 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:18:09,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:18:11,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4009, 3.2424, 3.1188, 1.4619], device='cuda:1'), covar=tensor([0.0945, 0.1014, 0.1083, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1142, 0.1058, 0.0909, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 05:18:17,770 INFO [train.py:968] (1/2) Epoch 18, batch 3350, giga_loss[loss=0.2527, simple_loss=0.3336, pruned_loss=0.08584, over 28882.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3527, pruned_loss=0.1014, over 5711421.15 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3484, pruned_loss=0.09197, over 4700962.23 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3532, pruned_loss=0.1024, over 5699683.08 frames. ], batch size: 186, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:18:32,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:18:59,805 INFO [optim.py:369] (1/2) 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 05:19:02,121 INFO [train.py:968] (1/2) Epoch 18, batch 3400, giga_loss[loss=0.2591, simple_loss=0.3444, pruned_loss=0.08693, over 28798.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3525, pruned_loss=0.1016, over 5719460.31 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3469, pruned_loss=0.09109, over 4735635.24 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3542, pruned_loss=0.1033, over 5705941.54 frames. ], batch size: 174, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:19:46,004 INFO [train.py:968] (1/2) Epoch 18, batch 3450, giga_loss[loss=0.2742, simple_loss=0.3573, pruned_loss=0.09558, over 28929.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3514, pruned_loss=0.1007, over 5726307.78 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.346, pruned_loss=0.09066, over 4763490.48 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3535, pruned_loss=0.1026, over 5712435.00 frames. ], batch size: 174, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:19:47,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:20:10,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6097, 4.4464, 4.1832, 2.1037], device='cuda:1'), covar=tensor([0.0555, 0.0701, 0.0747, 0.2024], device='cuda:1'), in_proj_covar=tensor([0.1141, 0.1057, 0.0908, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 05:20:20,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:20:28,917 INFO [optim.py:369] (1/2) 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 05:20:29,528 INFO [train.py:968] (1/2) Epoch 18, batch 3500, giga_loss[loss=0.3341, simple_loss=0.3939, pruned_loss=0.1371, over 28891.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3521, pruned_loss=0.1012, over 5723495.48 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3455, pruned_loss=0.0904, over 4780198.68 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3543, pruned_loss=0.1031, over 5710249.09 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:20:29,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7723, 2.0949, 1.5979, 1.8870], device='cuda:1'), covar=tensor([0.2898, 0.2779, 0.3271, 0.2641], device='cuda:1'), in_proj_covar=tensor([0.1442, 0.1047, 0.1279, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 05:20:46,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-09 05:20:47,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1352, 0.8067, 0.9408, 1.3347], device='cuda:1'), covar=tensor([0.0842, 0.0372, 0.0350, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 05:21:07,797 INFO [train.py:968] (1/2) Epoch 18, batch 3550, giga_loss[loss=0.2717, simple_loss=0.3508, pruned_loss=0.09625, over 28785.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.352, pruned_loss=0.09964, over 5727874.49 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3459, pruned_loss=0.09042, over 4829835.37 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3537, pruned_loss=0.1016, over 5709977.88 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:21:22,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9978, 1.1299, 3.2867, 2.8505], device='cuda:1'), covar=tensor([0.1680, 0.2736, 0.0472, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0719, 0.0615, 0.0909, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-09 05:21:24,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:21:37,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780291.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:21:50,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:21:52,720 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:21:53,749 INFO [train.py:968] (1/2) Epoch 18, batch 3600, giga_loss[loss=0.2866, simple_loss=0.3552, pruned_loss=0.109, over 28769.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3528, pruned_loss=0.09959, over 5731150.10 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3458, pruned_loss=0.09043, over 4845691.87 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1012, over 5714750.30 frames. ], batch size: 99, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:21:55,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5903, 1.8922, 1.5921, 1.6798], device='cuda:1'), covar=tensor([0.0774, 0.0275, 0.0313, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 05:22:16,158 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:20,349 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:23,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:28,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 05:22:34,526 INFO [train.py:968] (1/2) Epoch 18, batch 3650, giga_loss[loss=0.2525, simple_loss=0.3266, pruned_loss=0.0892, over 28672.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3514, pruned_loss=0.0989, over 5732114.19 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3462, pruned_loss=0.09065, over 4871341.69 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3525, pruned_loss=0.1003, over 5715588.75 frames. ], batch size: 85, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:22:47,318 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:22:55,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:23:12,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5392, 1.9451, 1.4449, 1.8121], device='cuda:1'), covar=tensor([0.2764, 0.2602, 0.3090, 0.2291], device='cuda:1'), in_proj_covar=tensor([0.1443, 0.1047, 0.1278, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 05:23:17,684 INFO [optim.py:369] (1/2) 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 05:23:17,696 INFO [train.py:968] (1/2) Epoch 18, batch 3700, libri_loss[loss=0.2748, simple_loss=0.3648, pruned_loss=0.09241, over 29525.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3508, pruned_loss=0.09934, over 5731724.44 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3465, pruned_loss=0.09068, over 4902122.66 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3515, pruned_loss=0.1007, over 5713469.12 frames. ], batch size: 84, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:23:35,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780434.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:23:37,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780437.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:23:57,311 INFO [train.py:968] (1/2) Epoch 18, batch 3750, giga_loss[loss=0.2444, simple_loss=0.3275, pruned_loss=0.08069, over 28723.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3491, pruned_loss=0.09865, over 5732737.93 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3463, pruned_loss=0.09059, over 4912308.25 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3499, pruned_loss=0.09983, over 5716697.36 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:24:00,139 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780466.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:24:19,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-09 05:24:41,366 INFO [optim.py:369] (1/2) 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 05:24:41,378 INFO [train.py:968] (1/2) Epoch 18, batch 3800, giga_loss[loss=0.2782, simple_loss=0.3565, pruned_loss=0.09993, over 28990.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3488, pruned_loss=0.0984, over 5731784.45 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.346, pruned_loss=0.0904, over 4922049.43 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3497, pruned_loss=0.09961, over 5724797.54 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:24:46,990 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:24:52,671 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780525.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:24:55,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-09 05:24:56,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:01,384 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:08,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3019, 1.7440, 1.6386, 1.4046], device='cuda:1'), covar=tensor([0.2134, 0.1814, 0.2107, 0.2162], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0737, 0.0694, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:25:14,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:22,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 05:25:23,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:25:23,442 INFO [train.py:968] (1/2) Epoch 18, batch 3850, giga_loss[loss=0.234, simple_loss=0.3221, pruned_loss=0.07293, over 28830.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3495, pruned_loss=0.09939, over 5728724.25 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3458, pruned_loss=0.09039, over 4926823.19 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3503, pruned_loss=0.1004, over 5722684.02 frames. ], batch size: 119, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:25:46,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3104, 1.4975, 1.3478, 1.5602], device='cuda:1'), covar=tensor([0.0717, 0.0436, 0.0336, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 05:26:05,007 INFO [optim.py:369] (1/2) 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 05:26:05,020 INFO [train.py:968] (1/2) Epoch 18, batch 3900, giga_loss[loss=0.2561, simple_loss=0.3358, pruned_loss=0.08823, over 28856.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3491, pruned_loss=0.09838, over 5728560.32 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3455, pruned_loss=0.0902, over 4941610.15 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.35, pruned_loss=0.09943, over 5721219.86 frames. ], batch size: 112, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:26:43,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:26:51,233 INFO [train.py:968] (1/2) Epoch 18, batch 3950, giga_loss[loss=0.2344, simple_loss=0.321, pruned_loss=0.07389, over 28434.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3473, pruned_loss=0.09694, over 5723236.41 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3455, pruned_loss=0.0902, over 4941610.15 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3481, pruned_loss=0.09776, over 5717523.24 frames. ], batch size: 60, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:27:30,583 INFO [optim.py:369] (1/2) 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 05:27:30,596 INFO [train.py:968] (1/2) Epoch 18, batch 4000, giga_loss[loss=0.2623, simple_loss=0.3377, pruned_loss=0.0935, over 28939.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3464, pruned_loss=0.09659, over 5726314.12 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3455, pruned_loss=0.09027, over 4978903.08 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.347, pruned_loss=0.09744, over 5715723.99 frames. ], batch size: 106, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:27:44,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 05:28:09,267 INFO [train.py:968] (1/2) Epoch 18, batch 4050, giga_loss[loss=0.2478, simple_loss=0.3245, pruned_loss=0.08561, over 28529.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3443, pruned_loss=0.09571, over 5713392.74 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3457, pruned_loss=0.09062, over 4991863.75 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3446, pruned_loss=0.09622, over 5710417.54 frames. ], batch size: 71, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:28:11,492 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=780765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:28:38,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=780796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:28:40,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4946, 1.6851, 1.6275, 1.4843], device='cuda:1'), covar=tensor([0.1785, 0.2067, 0.2295, 0.2101], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0735, 0.0694, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:28:40,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=780799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:28:50,386 INFO [train.py:968] (1/2) Epoch 18, batch 4100, giga_loss[loss=0.2385, simple_loss=0.3209, pruned_loss=0.07806, over 28886.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.341, pruned_loss=0.09387, over 5711976.49 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3458, pruned_loss=0.09069, over 5009818.56 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3411, pruned_loss=0.09427, over 5706178.69 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:28:50,978 INFO [optim.py:369] (1/2) 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 05:29:01,378 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-09 05:29:02,936 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=780828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:29:19,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 05:29:28,874 INFO [train.py:968] (1/2) Epoch 18, batch 4150, giga_loss[loss=0.2912, simple_loss=0.3651, pruned_loss=0.1086, over 28609.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3402, pruned_loss=0.09399, over 5714123.43 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3456, pruned_loss=0.09066, over 5040873.50 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3403, pruned_loss=0.09443, over 5703217.82 frames. ], batch size: 307, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:29:54,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:29:55,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4362, 1.7099, 1.4455, 1.5109], device='cuda:1'), covar=tensor([0.0735, 0.0313, 0.0317, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 05:29:58,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780900.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:30:05,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=780910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:30:06,931 INFO [train.py:968] (1/2) Epoch 18, batch 4200, giga_loss[loss=0.257, simple_loss=0.3318, pruned_loss=0.09108, over 28954.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3398, pruned_loss=0.09422, over 5714813.04 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3458, pruned_loss=0.09069, over 5060093.53 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3395, pruned_loss=0.09464, over 5703618.61 frames. ], batch size: 120, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:30:09,504 INFO [optim.py:369] (1/2) 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 05:30:20,608 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 4250, giga_loss[loss=0.2322, simple_loss=0.3146, pruned_loss=0.07493, over 28983.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3388, pruned_loss=0.09451, over 5713812.89 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3462, pruned_loss=0.09103, over 5077086.67 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3382, pruned_loss=0.09464, over 5701280.30 frames. ], batch size: 213, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:30:57,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6261, 2.5618, 2.4257, 2.1522], device='cuda:1'), covar=tensor([0.1684, 0.2054, 0.1978, 0.2117], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0738, 0.0696, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:31:28,851 INFO [train.py:968] (1/2) Epoch 18, batch 4300, giga_loss[loss=0.3349, simple_loss=0.3807, pruned_loss=0.1446, over 26528.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3362, pruned_loss=0.09349, over 5717345.45 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3457, pruned_loss=0.09073, over 5098061.52 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3359, pruned_loss=0.09391, over 5705608.83 frames. ], batch size: 555, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:31:29,457 INFO [optim.py:369] (1/2) 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 05:31:50,368 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:31:52,280 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:31:55,007 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781043.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:31:56,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781046.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:32:00,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5476, 1.7121, 1.3699, 1.6104], device='cuda:1'), covar=tensor([0.2997, 0.2986, 0.3502, 0.2573], device='cuda:1'), in_proj_covar=tensor([0.1441, 0.1044, 0.1275, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 05:32:01,597 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:04,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:08,673 INFO [train.py:968] (1/2) Epoch 18, batch 4350, giga_loss[loss=0.2201, simple_loss=0.2973, pruned_loss=0.07148, over 28583.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.335, pruned_loss=0.09328, over 5712319.39 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3461, pruned_loss=0.09096, over 5114040.06 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3341, pruned_loss=0.09347, over 5702227.05 frames. ], batch size: 60, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:32:12,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:13,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:15,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:19,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781075.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:32:27,867 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:39,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:32:48,040 INFO [train.py:968] (1/2) Epoch 18, batch 4400, libri_loss[loss=0.2576, simple_loss=0.3436, pruned_loss=0.08578, over 29496.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3335, pruned_loss=0.09226, over 5713296.13 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3461, pruned_loss=0.09107, over 5135629.07 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3323, pruned_loss=0.09239, over 5709388.46 frames. ], batch size: 81, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:32:48,715 INFO [optim.py:369] (1/2) 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 05:33:04,173 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:33:10,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:33:29,173 INFO [train.py:968] (1/2) Epoch 18, batch 4450, giga_loss[loss=0.2323, simple_loss=0.309, pruned_loss=0.07782, over 28431.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3336, pruned_loss=0.09186, over 5710153.79 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3463, pruned_loss=0.09115, over 5149449.41 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3322, pruned_loss=0.09191, over 5707272.73 frames. ], batch size: 60, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:34:14,814 INFO [train.py:968] (1/2) Epoch 18, batch 4500, giga_loss[loss=0.2664, simple_loss=0.3506, pruned_loss=0.0911, over 28813.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3375, pruned_loss=0.09421, over 5699214.19 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3463, pruned_loss=0.09123, over 5159204.50 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3362, pruned_loss=0.09422, over 5698024.36 frames. ], batch size: 186, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:34:15,388 INFO [optim.py:369] (1/2) 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 05:34:16,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5339, 4.3542, 4.1674, 2.2516], device='cuda:1'), covar=tensor([0.0531, 0.0703, 0.0721, 0.1771], device='cuda:1'), in_proj_covar=tensor([0.1151, 0.1064, 0.0914, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 05:34:25,522 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 4550, giga_loss[loss=0.2854, simple_loss=0.3631, pruned_loss=0.1038, over 28670.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3398, pruned_loss=0.09454, over 5706253.50 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3462, pruned_loss=0.09118, over 5174125.41 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3387, pruned_loss=0.09464, over 5701646.40 frames. ], batch size: 242, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:35:17,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:35:19,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:35:39,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:35:45,364 INFO [train.py:968] (1/2) Epoch 18, batch 4600, giga_loss[loss=0.2313, simple_loss=0.322, pruned_loss=0.07037, over 29038.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3419, pruned_loss=0.09521, over 5698266.85 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.346, pruned_loss=0.0911, over 5181474.83 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3412, pruned_loss=0.0954, over 5692715.63 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:35:46,012 INFO [optim.py:369] (1/2) 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 05:35:48,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:36:27,610 INFO [train.py:968] (1/2) Epoch 18, batch 4650, giga_loss[loss=0.2433, simple_loss=0.3345, pruned_loss=0.07612, over 28979.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3417, pruned_loss=0.09475, over 5701526.00 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3458, pruned_loss=0.09106, over 5206176.04 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3411, pruned_loss=0.09507, over 5690842.05 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:36:31,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1859, 1.4651, 1.2771, 1.4851], device='cuda:1'), covar=tensor([0.0748, 0.0314, 0.0338, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 05:37:09,544 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 4700, giga_loss[loss=0.2669, simple_loss=0.3318, pruned_loss=0.101, over 28541.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3414, pruned_loss=0.09486, over 5701960.16 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3461, pruned_loss=0.09126, over 5209307.82 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3406, pruned_loss=0.09497, over 5695353.68 frames. ], batch size: 78, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:37:10,571 INFO [optim.py:369] (1/2) 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 05:37:52,492 INFO [train.py:968] (1/2) Epoch 18, batch 4750, giga_loss[loss=0.2801, simple_loss=0.3412, pruned_loss=0.1095, over 28808.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3427, pruned_loss=0.09602, over 5700570.45 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3466, pruned_loss=0.09162, over 5223165.99 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3416, pruned_loss=0.09587, over 5691678.84 frames. ], batch size: 112, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:37:58,739 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-09 05:38:29,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:38:33,381 INFO [train.py:968] (1/2) Epoch 18, batch 4800, giga_loss[loss=0.2717, simple_loss=0.3485, pruned_loss=0.09752, over 27951.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.344, pruned_loss=0.0972, over 5692687.41 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3468, pruned_loss=0.09176, over 5225758.76 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3429, pruned_loss=0.09707, over 5691513.88 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:38:35,272 INFO [optim.py:369] (1/2) 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 05:38:58,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 05:39:07,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3093, 4.1302, 3.9227, 1.9167], device='cuda:1'), covar=tensor([0.0669, 0.0835, 0.0845, 0.2007], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.1063, 0.0918, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 05:39:14,337 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:39:17,976 INFO [train.py:968] (1/2) Epoch 18, batch 4850, giga_loss[loss=0.2724, simple_loss=0.3521, pruned_loss=0.09629, over 28875.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3474, pruned_loss=0.09887, over 5694416.82 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3468, pruned_loss=0.09178, over 5234082.19 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3465, pruned_loss=0.09885, over 5692270.81 frames. ], batch size: 199, lr: 1.80e-03, grad_scale: 8.0
+2023-03-09 05:39:46,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:39:50,589 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:39:56,046 INFO [train.py:968] (1/2) Epoch 18, batch 4900, giga_loss[loss=0.3172, simple_loss=0.39, pruned_loss=0.1222, over 28910.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3508, pruned_loss=0.1002, over 5710925.89 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3474, pruned_loss=0.092, over 5260177.94 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3497, pruned_loss=0.1003, over 5702147.34 frames. ], batch size: 164, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:39:58,108 INFO [optim.py:369] (1/2) 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 05:40:17,797 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:27,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:29,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:38,147 INFO [train.py:968] (1/2) Epoch 18, batch 4950, giga_loss[loss=0.2707, simple_loss=0.3512, pruned_loss=0.09507, over 28276.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3511, pruned_loss=0.09998, over 5714375.62 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3475, pruned_loss=0.09197, over 5278178.94 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3502, pruned_loss=0.1003, over 5703160.56 frames. ], batch size: 368, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:40:53,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:54,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:40:58,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3646, 2.0544, 1.5001, 0.5308], device='cuda:1'), covar=tensor([0.5910, 0.2989, 0.4269, 0.6493], device='cuda:1'), in_proj_covar=tensor([0.1682, 0.1582, 0.1564, 0.1373], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 05:41:20,499 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 5000, giga_loss[loss=0.2686, simple_loss=0.3493, pruned_loss=0.09394, over 29062.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3512, pruned_loss=0.09974, over 5721342.33 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3479, pruned_loss=0.09216, over 5284409.30 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3502, pruned_loss=0.09988, over 5710900.97 frames. ], batch size: 155, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:41:22,776 INFO [optim.py:369] (1/2) 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 05:41:45,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:41:47,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:42:01,601 INFO [train.py:968] (1/2) Epoch 18, batch 5050, giga_loss[loss=0.2561, simple_loss=0.3291, pruned_loss=0.09159, over 28895.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3513, pruned_loss=0.1001, over 5725450.44 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3481, pruned_loss=0.09235, over 5295260.32 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3503, pruned_loss=0.1002, over 5715355.61 frames. ], batch size: 112, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:42:11,350 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:42:21,063 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781786.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:42:42,899 INFO [train.py:968] (1/2) Epoch 18, batch 5100, giga_loss[loss=0.2718, simple_loss=0.3366, pruned_loss=0.1035, over 28636.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3497, pruned_loss=0.09933, over 5716411.24 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3485, pruned_loss=0.09255, over 5301613.16 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3487, pruned_loss=0.09938, over 5712500.11 frames. ], batch size: 92, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 05:42:45,653 INFO [optim.py:369] (1/2) 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 05:42:51,751 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:42:53,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:43:18,410 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 5150, giga_loss[loss=0.2313, simple_loss=0.3042, pruned_loss=0.07918, over 28504.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3465, pruned_loss=0.09761, over 5713253.30 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3488, pruned_loss=0.09283, over 5303778.33 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.09754, over 5716699.62 frames. ], batch size: 71, lr: 1.80e-03, grad_scale: 2.0
+2023-03-09 05:44:06,571 INFO [train.py:968] (1/2) Epoch 18, batch 5200, giga_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1142, over 27931.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3444, pruned_loss=0.09669, over 5721056.47 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3495, pruned_loss=0.09318, over 5321818.98 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3429, pruned_loss=0.09645, over 5718693.39 frames. ], batch size: 412, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:44:09,190 INFO [optim.py:369] (1/2) 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 05:44:17,275 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=781926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:44:20,309 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=781929.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:44:22,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=781932.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:44:24,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:44:46,609 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=781961.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:44:46,971 INFO [train.py:968] (1/2) Epoch 18, batch 5250, giga_loss[loss=0.2806, simple_loss=0.3624, pruned_loss=0.09935, over 29026.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3443, pruned_loss=0.09624, over 5719658.01 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3496, pruned_loss=0.09318, over 5330547.82 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3429, pruned_loss=0.09609, over 5715327.82 frames. ], batch size: 136, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:45:03,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=781980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:45:29,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:45:30,922 INFO [train.py:968] (1/2) Epoch 18, batch 5300, giga_loss[loss=0.2841, simple_loss=0.3534, pruned_loss=0.1074, over 28498.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3455, pruned_loss=0.0957, over 5713852.07 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.349, pruned_loss=0.09303, over 5346382.61 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3448, pruned_loss=0.09579, over 5706315.36 frames. ], batch size: 60, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:45:31,767 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:45:41,178 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 5350, giga_loss[loss=0.2682, simple_loss=0.3595, pruned_loss=0.0884, over 28788.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3456, pruned_loss=0.09574, over 5693107.67 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3489, pruned_loss=0.09314, over 5340374.42 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.345, pruned_loss=0.09576, over 5701222.61 frames. ], batch size: 243, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:46:27,789 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:29,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:34,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:53,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:46:55,843 INFO [train.py:968] (1/2) Epoch 18, batch 5400, giga_loss[loss=0.2958, simple_loss=0.3601, pruned_loss=0.1157, over 28885.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3446, pruned_loss=0.09637, over 5699302.37 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3487, pruned_loss=0.09312, over 5349078.62 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3443, pruned_loss=0.09644, over 5702654.65 frames. ], batch size: 186, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:46:56,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4895, 3.4128, 1.4545, 1.6010], device='cuda:1'), covar=tensor([0.0910, 0.0350, 0.0976, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0536, 0.0367, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 05:46:59,000 INFO [optim.py:369] (1/2) 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 05:47:06,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:08,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:14,321 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:27,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2333, 1.1726, 4.1755, 3.3304], device='cuda:1'), covar=tensor([0.1646, 0.2833, 0.0377, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0621, 0.0916, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 05:47:32,109 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:32,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:34,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:47:36,617 INFO [train.py:968] (1/2) Epoch 18, batch 5450, giga_loss[loss=0.2774, simple_loss=0.3457, pruned_loss=0.1045, over 28869.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3446, pruned_loss=0.09771, over 5697845.18 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.349, pruned_loss=0.09351, over 5362787.91 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3439, pruned_loss=0.09755, over 5698654.56 frames. ], batch size: 227, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:47:48,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8819, 3.6825, 3.4840, 1.7956], device='cuda:1'), covar=tensor([0.0758, 0.0922, 0.0853, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1148, 0.1057, 0.0912, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 05:47:59,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:48:04,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-09 05:48:19,267 INFO [train.py:968] (1/2) Epoch 18, batch 5500, giga_loss[loss=0.2319, simple_loss=0.3101, pruned_loss=0.07682, over 28924.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3433, pruned_loss=0.09792, over 5699319.88 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3491, pruned_loss=0.09365, over 5366340.82 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3425, pruned_loss=0.09778, over 5704153.35 frames. ], batch size: 227, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:48:23,138 INFO [optim.py:369] (1/2) 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 05:48:28,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:48:34,106 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:48:36,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:48:55,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8602, 2.8587, 1.9226, 0.8255], device='cuda:1'), covar=tensor([0.8112, 0.3019, 0.3720, 0.7655], device='cuda:1'), in_proj_covar=tensor([0.1668, 0.1567, 0.1549, 0.1363], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 05:49:01,247 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:49:01,734 INFO [train.py:968] (1/2) Epoch 18, batch 5550, giga_loss[loss=0.244, simple_loss=0.3093, pruned_loss=0.08931, over 28335.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3406, pruned_loss=0.09715, over 5702378.58 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3488, pruned_loss=0.09347, over 5377362.42 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3402, pruned_loss=0.09726, over 5702251.63 frames. ], batch size: 65, lr: 1.80e-03, grad_scale: 4.0
+2023-03-09 05:49:18,291 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-09 05:49:35,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:49:43,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2819, 1.5536, 1.3573, 1.4731], device='cuda:1'), covar=tensor([0.0757, 0.0331, 0.0343, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0115, 0.0115, 0.0215], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0101], device='cuda:1')
+2023-03-09 05:49:44,741 INFO [train.py:968] (1/2) Epoch 18, batch 5600, giga_loss[loss=0.2625, simple_loss=0.3369, pruned_loss=0.09408, over 27980.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.34, pruned_loss=0.09662, over 5708629.13 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.349, pruned_loss=0.09351, over 5391039.89 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3392, pruned_loss=0.0968, over 5706152.37 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:49:47,673 INFO [optim.py:369] (1/2) 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 05:50:16,834 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-09 05:50:26,081 INFO [train.py:968] (1/2) Epoch 18, batch 5650, giga_loss[loss=0.2534, simple_loss=0.3254, pruned_loss=0.09072, over 28743.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3366, pruned_loss=0.09458, over 5707195.23 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3492, pruned_loss=0.0936, over 5396711.14 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3355, pruned_loss=0.09475, over 5711138.99 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:50:44,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:50:54,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:06,469 INFO [train.py:968] (1/2) Epoch 18, batch 5700, giga_loss[loss=0.2302, simple_loss=0.3053, pruned_loss=0.07752, over 28918.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3331, pruned_loss=0.09291, over 5701059.27 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3494, pruned_loss=0.09387, over 5396208.10 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3314, pruned_loss=0.0928, over 5717060.10 frames. ], batch size: 112, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:51:09,642 INFO [optim.py:369] (1/2) 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 05:51:32,454 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:34,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:38,108 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:51:44,721 INFO [train.py:968] (1/2) Epoch 18, batch 5750, giga_loss[loss=0.2434, simple_loss=0.3178, pruned_loss=0.08448, over 29089.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3309, pruned_loss=0.09147, over 5703435.71 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3492, pruned_loss=0.09381, over 5408580.21 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3291, pruned_loss=0.09139, over 5716773.84 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:51:51,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-09 05:51:59,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:12,108 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782495.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:52:12,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3198, 1.4391, 1.3782, 1.2416], device='cuda:1'), covar=tensor([0.2450, 0.2147, 0.1714, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1877, 0.1821, 0.1751, 0.1882], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 05:52:20,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:23,373 INFO [train.py:968] (1/2) Epoch 18, batch 5800, libri_loss[loss=0.2765, simple_loss=0.3594, pruned_loss=0.0968, over 27830.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3328, pruned_loss=0.09227, over 5708958.53 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3498, pruned_loss=0.09417, over 5420808.73 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3303, pruned_loss=0.09182, over 5716386.62 frames. ], batch size: 116, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:52:27,031 INFO [optim.py:369] (1/2) 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 05:52:36,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:38,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:46,345 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782538.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 05:52:47,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:52:48,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:53:02,949 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 5850, giga_loss[loss=0.3628, simple_loss=0.4089, pruned_loss=0.1584, over 26810.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3358, pruned_loss=0.09328, over 5713922.01 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3495, pruned_loss=0.09396, over 5432518.28 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3336, pruned_loss=0.09306, over 5715679.66 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:53:13,252 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:53:21,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4688, 1.6451, 1.3531, 1.6278], device='cuda:1'), covar=tensor([0.2732, 0.2815, 0.3157, 0.2518], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1039, 0.1272, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 05:53:33,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:53:46,114 INFO [train.py:968] (1/2) Epoch 18, batch 5900, libri_loss[loss=0.3188, simple_loss=0.3878, pruned_loss=0.1249, over 29652.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3379, pruned_loss=0.09374, over 5719678.51 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3493, pruned_loss=0.09382, over 5450993.58 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3359, pruned_loss=0.09365, over 5714365.23 frames. ], batch size: 91, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:53:48,671 INFO [optim.py:369] (1/2) 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 05:53:53,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 05:54:14,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 05:54:19,191 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:54:21,021 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 5950, giga_loss[loss=0.2999, simple_loss=0.3668, pruned_loss=0.1165, over 28785.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3416, pruned_loss=0.0955, over 5710577.94 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3496, pruned_loss=0.09408, over 5451185.00 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3395, pruned_loss=0.09523, over 5710144.28 frames. ], batch size: 99, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:54:30,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1906, 1.2109, 1.0735, 0.8808], device='cuda:1'), covar=tensor([0.0871, 0.0562, 0.1087, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0445, 0.0509, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 05:54:46,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:54:47,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3294, 1.6467, 1.5764, 1.2820], device='cuda:1'), covar=tensor([0.3047, 0.2405, 0.2691, 0.2860], device='cuda:1'), in_proj_covar=tensor([0.1877, 0.1823, 0.1750, 0.1877], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 05:55:14,480 INFO [train.py:968] (1/2) Epoch 18, batch 6000, giga_loss[loss=0.2713, simple_loss=0.3435, pruned_loss=0.09954, over 28402.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.345, pruned_loss=0.09781, over 5712437.05 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3498, pruned_loss=0.0943, over 5459908.54 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3431, pruned_loss=0.09746, over 5709196.19 frames. ], batch size: 65, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:55:14,480 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 05:55:22,920 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 05:55:27,093 INFO [optim.py:369] (1/2) 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 05:55:44,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8584, 2.1658, 2.0817, 1.9236], device='cuda:1'), covar=tensor([0.1502, 0.1493, 0.1688, 0.1593], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0741, 0.0697, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 05:55:47,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:55:50,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:55:59,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4368, 1.8211, 1.3805, 1.6636], device='cuda:1'), covar=tensor([0.2726, 0.2706, 0.3138, 0.2297], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1042, 0.1272, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 05:56:10,955 INFO [train.py:968] (1/2) Epoch 18, batch 6050, giga_loss[loss=0.294, simple_loss=0.358, pruned_loss=0.115, over 28333.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.349, pruned_loss=0.101, over 5707576.77 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.35, pruned_loss=0.09439, over 5460905.31 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3473, pruned_loss=0.1007, over 5706279.77 frames. ], batch size: 71, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:56:19,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=782771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:56:59,962 INFO [train.py:968] (1/2) Epoch 18, batch 6100, giga_loss[loss=0.3119, simple_loss=0.3761, pruned_loss=0.1238, over 28939.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3556, pruned_loss=0.1062, over 5702858.34 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3502, pruned_loss=0.09449, over 5467300.14 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3541, pruned_loss=0.106, over 5699354.47 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:57:04,137 INFO [optim.py:369] (1/2) 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 05:57:14,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:57:24,420 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 6150, giga_loss[loss=0.2872, simple_loss=0.3569, pruned_loss=0.1087, over 28694.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3623, pruned_loss=0.1116, over 5679445.28 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3498, pruned_loss=0.09424, over 5473624.56 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3616, pruned_loss=0.112, over 5675873.48 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 05:57:55,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782870.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:58:34,765 INFO [train.py:968] (1/2) Epoch 18, batch 6200, giga_loss[loss=0.323, simple_loss=0.3903, pruned_loss=0.1279, over 28705.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3677, pruned_loss=0.1158, over 5676016.10 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09429, over 5478555.02 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3677, pruned_loss=0.1166, over 5675953.32 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 05:58:37,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=782913.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 05:58:43,106 INFO [optim.py:369] (1/2) 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 05:59:23,762 INFO [train.py:968] (1/2) Epoch 18, batch 6250, giga_loss[loss=0.3294, simple_loss=0.3912, pruned_loss=0.1338, over 28814.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.372, pruned_loss=0.1199, over 5678289.03 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3497, pruned_loss=0.09432, over 5485318.77 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3724, pruned_loss=0.121, over 5675620.58 frames. ], batch size: 66, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 05:59:32,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=782970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:59:34,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=782973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:59:38,549 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-09 05:59:39,385 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=782978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 05:59:49,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6147, 4.9537, 1.8317, 1.8978], device='cuda:1'), covar=tensor([0.0967, 0.0225, 0.0889, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0539, 0.0368, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 05:59:52,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2228, 1.5093, 1.5710, 1.2858], device='cuda:1'), covar=tensor([0.1779, 0.1642, 0.2133, 0.1866], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0742, 0.0698, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 06:00:00,079 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:00:10,086 INFO [train.py:968] (1/2) Epoch 18, batch 6300, giga_loss[loss=0.3312, simple_loss=0.3972, pruned_loss=0.1326, over 29039.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3774, pruned_loss=0.1241, over 5670865.89 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3498, pruned_loss=0.09422, over 5495684.23 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3784, pruned_loss=0.1259, over 5665548.77 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:00:14,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783013.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 06:00:15,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 06:00:17,531 INFO [optim.py:369] (1/2) 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 06:00:45,182 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783045.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 06:00:57,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783056.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 06:01:01,458 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783059.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 06:01:03,168 INFO [train.py:968] (1/2) Epoch 18, batch 6350, giga_loss[loss=0.2597, simple_loss=0.3377, pruned_loss=0.0908, over 28876.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3818, pruned_loss=0.1285, over 5658227.66 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3504, pruned_loss=0.09454, over 5502083.89 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1305, over 5651857.08 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:01:31,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 06:01:55,831 INFO [train.py:968] (1/2) Epoch 18, batch 6400, giga_loss[loss=0.4303, simple_loss=0.4395, pruned_loss=0.2105, over 26577.00 frames. ], tot_loss[loss=0.3242, simple_loss=0.384, pruned_loss=0.1322, over 5644454.96 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09445, over 5511254.83 frames. ], giga_tot_loss[loss=0.3276, simple_loss=0.3859, pruned_loss=0.1347, over 5634201.59 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:02:01,925 INFO [optim.py:369] (1/2) 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 06:02:51,235 INFO [train.py:968] (1/2) Epoch 18, batch 6450, giga_loss[loss=0.3735, simple_loss=0.4121, pruned_loss=0.1675, over 28257.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3874, pruned_loss=0.1363, over 5607897.96 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09449, over 5500051.68 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3894, pruned_loss=0.1389, over 5612061.31 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:02:55,519 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=783166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:03:42,287 INFO [train.py:968] (1/2) Epoch 18, batch 6500, giga_loss[loss=0.375, simple_loss=0.4005, pruned_loss=0.1748, over 23444.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3897, pruned_loss=0.1379, over 5608440.03 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3498, pruned_loss=0.09437, over 5512517.52 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.393, pruned_loss=0.1416, over 5604399.58 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:03:43,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:03:47,328 INFO [optim.py:369] (1/2) 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 06:04:07,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4860, 1.6983, 1.3661, 1.6060], device='cuda:1'), covar=tensor([0.2126, 0.2100, 0.2326, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1042, 0.1274, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 06:04:18,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3654, 2.1101, 1.5603, 0.6014], device='cuda:1'), covar=tensor([0.5209, 0.2537, 0.3537, 0.5683], device='cuda:1'), in_proj_covar=tensor([0.1684, 0.1588, 0.1564, 0.1381], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 06:04:34,839 INFO [train.py:968] (1/2) Epoch 18, batch 6550, giga_loss[loss=0.3256, simple_loss=0.3815, pruned_loss=0.1349, over 28734.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3914, pruned_loss=0.1395, over 5621600.29 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3498, pruned_loss=0.09432, over 5513671.93 frames. ], giga_tot_loss[loss=0.3398, simple_loss=0.3942, pruned_loss=0.1427, over 5617913.47 frames. ], batch size: 99, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:05:26,446 INFO [train.py:968] (1/2) Epoch 18, batch 6600, giga_loss[loss=0.4215, simple_loss=0.4343, pruned_loss=0.2043, over 23501.00 frames. ], tot_loss[loss=0.3334, simple_loss=0.3893, pruned_loss=0.1387, over 5626681.00 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3498, pruned_loss=0.09446, over 5515813.69 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3925, pruned_loss=0.1424, over 5624210.87 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:05:32,841 INFO [optim.py:369] (1/2) 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 06:06:07,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:06:10,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:06:13,803 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 6650, giga_loss[loss=0.3106, simple_loss=0.378, pruned_loss=0.1216, over 28550.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3886, pruned_loss=0.1382, over 5617529.03 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.35, pruned_loss=0.09457, over 5515156.85 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3919, pruned_loss=0.142, over 5618384.81 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:06:43,544 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:07:09,205 INFO [train.py:968] (1/2) Epoch 18, batch 6700, giga_loss[loss=0.2911, simple_loss=0.3613, pruned_loss=0.1105, over 28799.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3894, pruned_loss=0.1377, over 5628813.06 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3499, pruned_loss=0.09449, over 5517086.06 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3922, pruned_loss=0.1408, over 5628324.65 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:07:14,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6511, 1.8517, 1.5191, 1.6950], device='cuda:1'), covar=tensor([0.2344, 0.2377, 0.2631, 0.2154], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1042, 0.1274, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 06:07:18,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:07:18,533 INFO [optim.py:369] (1/2) 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 06:07:48,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=783447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:08:02,624 INFO [train.py:968] (1/2) Epoch 18, batch 6750, libri_loss[loss=0.3097, simple_loss=0.3664, pruned_loss=0.1265, over 29599.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3902, pruned_loss=0.1379, over 5610962.61 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3503, pruned_loss=0.09479, over 5517861.14 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3927, pruned_loss=0.1407, over 5611238.23 frames. ], batch size: 75, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:08:37,744 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:08:40,185 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:08:50,765 INFO [train.py:968] (1/2) Epoch 18, batch 6800, giga_loss[loss=0.2847, simple_loss=0.3601, pruned_loss=0.1047, over 29033.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3872, pruned_loss=0.1351, over 5617953.24 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3496, pruned_loss=0.09431, over 5528232.26 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3909, pruned_loss=0.139, over 5611066.73 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:08:58,437 INFO [optim.py:369] (1/2) 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 06:09:10,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:09:21,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6772, 1.9101, 1.6034, 1.7320], device='cuda:1'), covar=tensor([0.2172, 0.1954, 0.2077, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1041, 0.1273, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 06:09:27,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:09:31,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-09 06:09:46,986 INFO [train.py:968] (1/2) Epoch 18, batch 6850, giga_loss[loss=0.3283, simple_loss=0.3986, pruned_loss=0.129, over 28776.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.385, pruned_loss=0.1324, over 5615378.45 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3495, pruned_loss=0.09429, over 5530148.10 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3882, pruned_loss=0.1357, over 5608607.46 frames. ], batch size: 99, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:09:47,521 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:09:51,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:10:05,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.2029, 6.0054, 5.7066, 3.1231], device='cuda:1'), covar=tensor([0.0437, 0.0586, 0.0710, 0.1455], device='cuda:1'), in_proj_covar=tensor([0.1169, 0.1081, 0.0928, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 06:10:18,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:10:36,387 INFO [train.py:968] (1/2) Epoch 18, batch 6900, giga_loss[loss=0.3573, simple_loss=0.3983, pruned_loss=0.1582, over 28309.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3817, pruned_loss=0.1285, over 5636633.12 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3495, pruned_loss=0.09425, over 5538813.99 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.385, pruned_loss=0.132, over 5625720.05 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:10:44,000 INFO [optim.py:369] (1/2) 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 06:11:24,447 INFO [train.py:968] (1/2) Epoch 18, batch 6950, giga_loss[loss=0.2851, simple_loss=0.355, pruned_loss=0.1076, over 28895.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3793, pruned_loss=0.1265, over 5646137.24 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3492, pruned_loss=0.09421, over 5546441.21 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3829, pruned_loss=0.1301, over 5632799.55 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:11:48,622 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:11:50,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:12:14,443 INFO [train.py:968] (1/2) Epoch 18, batch 7000, giga_loss[loss=0.3303, simple_loss=0.3908, pruned_loss=0.1349, over 28766.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3756, pruned_loss=0.1237, over 5651838.25 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3491, pruned_loss=0.09412, over 5553538.68 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3791, pruned_loss=0.1272, over 5636626.29 frames. ], batch size: 284, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:12:18,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:13:04,546 INFO [train.py:968] (1/2) Epoch 18, batch 7050, giga_loss[loss=0.2959, simple_loss=0.361, pruned_loss=0.1154, over 28776.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3756, pruned_loss=0.1237, over 5657443.72 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3491, pruned_loss=0.09413, over 5557841.46 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.379, pruned_loss=0.1271, over 5643524.30 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:13:59,716 INFO [train.py:968] (1/2) Epoch 18, batch 7100, giga_loss[loss=0.2882, simple_loss=0.3628, pruned_loss=0.1068, over 28981.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3753, pruned_loss=0.1231, over 5661743.95 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3492, pruned_loss=0.09414, over 5560886.84 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3784, pruned_loss=0.1264, over 5650041.52 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:14:08,421 INFO [optim.py:369] (1/2) 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 06:14:13,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=783822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:14:53,783 INFO [train.py:968] (1/2) Epoch 18, batch 7150, giga_loss[loss=0.3076, simple_loss=0.3837, pruned_loss=0.1157, over 28805.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.373, pruned_loss=0.1204, over 5673265.28 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3491, pruned_loss=0.09407, over 5571361.87 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3762, pruned_loss=0.1239, over 5657067.60 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:15:14,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1494, 2.9898, 2.8527, 1.5267], device='cuda:1'), covar=tensor([0.1064, 0.1123, 0.1096, 0.2487], device='cuda:1'), in_proj_covar=tensor([0.1174, 0.1090, 0.0935, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 06:15:49,969 INFO [train.py:968] (1/2) Epoch 18, batch 7200, libri_loss[loss=0.2573, simple_loss=0.3334, pruned_loss=0.09059, over 29551.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3747, pruned_loss=0.1201, over 5674637.48 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3488, pruned_loss=0.09399, over 5577975.39 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3781, pruned_loss=0.1236, over 5657884.51 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:15:57,964 INFO [optim.py:369] (1/2) 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 06:16:37,565 INFO [train.py:968] (1/2) Epoch 18, batch 7250, giga_loss[loss=0.3016, simple_loss=0.3769, pruned_loss=0.1131, over 29071.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3754, pruned_loss=0.1197, over 5672293.46 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3486, pruned_loss=0.09393, over 5584522.78 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.379, pruned_loss=0.1231, over 5655252.41 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:16:42,205 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=783965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:16:44,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=783968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:17:18,753 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=783997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:17:32,710 INFO [train.py:968] (1/2) Epoch 18, batch 7300, giga_loss[loss=0.3102, simple_loss=0.3757, pruned_loss=0.1224, over 28309.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3755, pruned_loss=0.1203, over 5677412.07 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3486, pruned_loss=0.09395, over 5587657.87 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3786, pruned_loss=0.1233, over 5662264.35 frames. ], batch size: 77, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:17:41,148 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:18:21,633 INFO [train.py:968] (1/2) Epoch 18, batch 7350, giga_loss[loss=0.3293, simple_loss=0.3816, pruned_loss=0.1386, over 28774.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.374, pruned_loss=0.1199, over 5676900.39 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3484, pruned_loss=0.09381, over 5590641.64 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3769, pruned_loss=0.1226, over 5663107.13 frames. ], batch size: 284, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:19:06,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 06:19:12,185 INFO [train.py:968] (1/2) Epoch 18, batch 7400, libri_loss[loss=0.2676, simple_loss=0.3526, pruned_loss=0.09127, over 29529.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3731, pruned_loss=0.1208, over 5673961.83 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3483, pruned_loss=0.0937, over 5599363.06 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3762, pruned_loss=0.1239, over 5657271.44 frames. ], batch size: 82, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:19:19,513 INFO [optim.py:369] (1/2) 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 06:19:54,324 INFO [train.py:968] (1/2) Epoch 18, batch 7450, libri_loss[loss=0.2859, simple_loss=0.3684, pruned_loss=0.1017, over 28680.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1193, over 5677418.16 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3482, pruned_loss=0.09354, over 5600739.80 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3742, pruned_loss=0.1228, over 5665284.90 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:20:21,371 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 06:20:28,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5707, 3.7566, 1.6280, 1.6621], device='cuda:1'), covar=tensor([0.0958, 0.0370, 0.0891, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0545, 0.0371, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 06:20:46,572 INFO [train.py:968] (1/2) Epoch 18, batch 7500, giga_loss[loss=0.2667, simple_loss=0.3555, pruned_loss=0.08891, over 28923.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3701, pruned_loss=0.1174, over 5691849.95 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3481, pruned_loss=0.09331, over 5609666.51 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3736, pruned_loss=0.1212, over 5676061.68 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:20:49,431 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-09 06:20:52,520 INFO [optim.py:369] (1/2) 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 06:21:01,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8641, 3.6610, 3.4734, 1.6824], device='cuda:1'), covar=tensor([0.0682, 0.0860, 0.0869, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.1174, 0.1085, 0.0932, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 06:21:09,817 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 06:21:33,569 INFO [train.py:968] (1/2) Epoch 18, batch 7550, giga_loss[loss=0.2972, simple_loss=0.3659, pruned_loss=0.1143, over 28652.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3701, pruned_loss=0.1165, over 5700256.74 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3481, pruned_loss=0.09333, over 5616210.04 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3735, pruned_loss=0.1201, over 5683923.97 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:22:13,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6689, 1.8625, 1.3472, 1.4337], device='cuda:1'), covar=tensor([0.0957, 0.0645, 0.1078, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0444, 0.0510, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 06:22:21,543 INFO [train.py:968] (1/2) Epoch 18, batch 7600, giga_loss[loss=0.3307, simple_loss=0.3868, pruned_loss=0.1373, over 27966.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3703, pruned_loss=0.1169, over 5695082.15 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09325, over 5617810.99 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3731, pruned_loss=0.1199, over 5681292.76 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:22:28,576 INFO [optim.py:369] (1/2) 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 06:23:07,617 INFO [train.py:968] (1/2) Epoch 18, batch 7650, giga_loss[loss=0.2758, simple_loss=0.3495, pruned_loss=0.1011, over 28937.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3686, pruned_loss=0.1162, over 5688001.95 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3482, pruned_loss=0.09348, over 5613177.18 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3712, pruned_loss=0.1188, over 5683729.17 frames. ], batch size: 227, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:23:53,676 INFO [train.py:968] (1/2) Epoch 18, batch 7700, giga_loss[loss=0.2735, simple_loss=0.3403, pruned_loss=0.1034, over 28873.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3674, pruned_loss=0.1154, over 5696869.09 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3484, pruned_loss=0.09352, over 5624415.24 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.37, pruned_loss=0.1184, over 5685991.63 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:24:08,134 INFO [optim.py:369] (1/2) 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 06:24:10,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5364, 4.3716, 4.1413, 2.0976], device='cuda:1'), covar=tensor([0.0574, 0.0703, 0.0711, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.1084, 0.0930, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 06:24:19,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:24:41,756 INFO [train.py:968] (1/2) Epoch 18, batch 7750, libri_loss[loss=0.2559, simple_loss=0.3447, pruned_loss=0.08354, over 29380.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3678, pruned_loss=0.1163, over 5695452.07 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.349, pruned_loss=0.09371, over 5634641.39 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3703, pruned_loss=0.1196, over 5679545.30 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:24:43,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4278, 1.7947, 1.4285, 1.4962], device='cuda:1'), covar=tensor([0.2502, 0.2503, 0.2878, 0.2291], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1044, 0.1276, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 06:25:15,072 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:25:36,433 INFO [train.py:968] (1/2) Epoch 18, batch 7800, giga_loss[loss=0.2615, simple_loss=0.3334, pruned_loss=0.09482, over 28254.00 frames. ], tot_loss[loss=0.3, simple_loss=0.367, pruned_loss=0.1165, over 5701811.69 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.349, pruned_loss=0.09377, over 5637402.03 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.369, pruned_loss=0.1192, over 5687334.41 frames. ], batch size: 77, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:25:47,103 INFO [optim.py:369] (1/2) 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 06:26:25,472 INFO [train.py:968] (1/2) Epoch 18, batch 7850, giga_loss[loss=0.3316, simple_loss=0.389, pruned_loss=0.137, over 28694.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3649, pruned_loss=0.1157, over 5704459.36 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3488, pruned_loss=0.09353, over 5643784.56 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3671, pruned_loss=0.1186, over 5688559.06 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:26:38,350 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=784575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:26:41,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=784578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:26:49,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:27:02,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:27:02,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 06:27:08,149 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=784607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:27:13,314 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 7900, giga_loss[loss=0.3058, simple_loss=0.3704, pruned_loss=0.1206, over 28674.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5705473.82 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.349, pruned_loss=0.09369, over 5647229.56 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3662, pruned_loss=0.1185, over 5690614.86 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:27:24,454 INFO [optim.py:369] (1/2) 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 06:28:04,352 INFO [train.py:968] (1/2) Epoch 18, batch 7950, giga_loss[loss=0.3178, simple_loss=0.3826, pruned_loss=0.1265, over 28627.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3652, pruned_loss=0.1166, over 5699956.45 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3488, pruned_loss=0.09355, over 5650996.29 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3669, pruned_loss=0.119, over 5685741.64 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:28:52,466 INFO [train.py:968] (1/2) Epoch 18, batch 8000, giga_loss[loss=0.2799, simple_loss=0.3612, pruned_loss=0.09927, over 28978.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1171, over 5692952.96 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3489, pruned_loss=0.09359, over 5654738.37 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3679, pruned_loss=0.1192, over 5679213.14 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:29:03,498 INFO [optim.py:369] (1/2) 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 06:29:13,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9143, 2.9380, 1.9247, 1.1722], device='cuda:1'), covar=tensor([0.6523, 0.2778, 0.3275, 0.6104], device='cuda:1'), in_proj_covar=tensor([0.1688, 0.1598, 0.1569, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 06:29:39,341 INFO [train.py:968] (1/2) Epoch 18, batch 8050, giga_loss[loss=0.2864, simple_loss=0.3573, pruned_loss=0.1077, over 28657.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3654, pruned_loss=0.1156, over 5680665.13 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3488, pruned_loss=0.09353, over 5650034.84 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3671, pruned_loss=0.1178, over 5675277.04 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:30:23,078 INFO [train.py:968] (1/2) Epoch 18, batch 8100, giga_loss[loss=0.3067, simple_loss=0.3684, pruned_loss=0.1225, over 28904.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3649, pruned_loss=0.1153, over 5681978.13 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.348, pruned_loss=0.0933, over 5660575.46 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.368, pruned_loss=0.1185, over 5668754.38 frames. ], batch size: 112, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:30:33,057 INFO [optim.py:369] (1/2) 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 06:30:46,638 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 06:31:13,790 INFO [train.py:968] (1/2) Epoch 18, batch 8150, giga_loss[loss=0.3043, simple_loss=0.3719, pruned_loss=0.1184, over 28930.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3662, pruned_loss=0.1165, over 5689355.12 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3479, pruned_loss=0.09341, over 5664417.61 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.369, pruned_loss=0.1193, over 5676107.54 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:31:19,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:32:03,845 INFO [train.py:968] (1/2) Epoch 18, batch 8200, giga_loss[loss=0.3478, simple_loss=0.4033, pruned_loss=0.1462, over 28368.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3688, pruned_loss=0.1193, over 5683353.26 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3482, pruned_loss=0.09355, over 5667550.70 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3712, pruned_loss=0.1221, over 5670313.44 frames. ], batch size: 71, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:32:16,677 INFO [optim.py:369] (1/2) 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 06:32:40,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:32:54,078 INFO [train.py:968] (1/2) Epoch 18, batch 8250, giga_loss[loss=0.2761, simple_loss=0.3486, pruned_loss=0.1017, over 28905.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3694, pruned_loss=0.1205, over 5687077.68 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3481, pruned_loss=0.09343, over 5673892.56 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1236, over 5671721.46 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:33:06,962 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=784972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:10,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:20,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=784986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:36,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4970, 1.6693, 1.6376, 1.4403], device='cuda:1'), covar=tensor([0.2599, 0.2481, 0.1794, 0.2235], device='cuda:1'), in_proj_covar=tensor([0.1882, 0.1829, 0.1747, 0.1881], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 06:33:44,330 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:44,609 INFO [train.py:968] (1/2) Epoch 18, batch 8300, giga_loss[loss=0.3264, simple_loss=0.3826, pruned_loss=0.1351, over 28295.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1229, over 5679405.65 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09347, over 5679797.76 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.126, over 5661892.52 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:33:46,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:54,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=785020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:33:57,717 INFO [optim.py:369] (1/2) 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 06:34:15,828 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:34:32,882 INFO [train.py:968] (1/2) Epoch 18, batch 8350, libri_loss[loss=0.2614, simple_loss=0.3468, pruned_loss=0.08798, over 29649.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3704, pruned_loss=0.1223, over 5681783.89 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3482, pruned_loss=0.09336, over 5685649.80 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3731, pruned_loss=0.1256, over 5662453.95 frames. ], batch size: 88, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:34:38,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3655, 1.8513, 1.5288, 1.5116], device='cuda:1'), covar=tensor([0.0621, 0.0248, 0.0276, 0.0657], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 06:34:53,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4170, 1.8586, 1.5743, 1.5422], device='cuda:1'), covar=tensor([0.0675, 0.0391, 0.0301, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 06:35:10,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:10,933 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-09 06:35:12,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:18,957 INFO [train.py:968] (1/2) Epoch 18, batch 8400, giga_loss[loss=0.2629, simple_loss=0.3446, pruned_loss=0.09066, over 28983.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3696, pruned_loss=0.1216, over 5678887.84 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3482, pruned_loss=0.09322, over 5686317.48 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3722, pruned_loss=0.1249, over 5662844.45 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:35:23,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:26,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:27,934 INFO [optim.py:369] (1/2) 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 06:35:34,088 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:35,787 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:35:53,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:36:02,282 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:36:02,702 INFO [train.py:968] (1/2) Epoch 18, batch 8450, giga_loss[loss=0.2726, simple_loss=0.35, pruned_loss=0.09763, over 28860.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3702, pruned_loss=0.1207, over 5672364.46 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09347, over 5680934.41 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1235, over 5664768.10 frames. ], batch size: 112, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:36:48,608 INFO [train.py:968] (1/2) Epoch 18, batch 8500, giga_loss[loss=0.2997, simple_loss=0.363, pruned_loss=0.1181, over 28587.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3679, pruned_loss=0.1188, over 5675599.46 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09353, over 5686266.43 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.37, pruned_loss=0.1216, over 5664523.88 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:36:56,964 INFO [optim.py:369] (1/2) 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 06:36:57,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4595, 1.5465, 1.2358, 1.0686], device='cuda:1'), covar=tensor([0.0739, 0.0408, 0.0888, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0444, 0.0510, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 06:37:33,631 INFO [train.py:968] (1/2) Epoch 18, batch 8550, giga_loss[loss=0.2715, simple_loss=0.3406, pruned_loss=0.1013, over 28573.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3648, pruned_loss=0.1171, over 5684920.94 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.348, pruned_loss=0.09327, over 5691852.43 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3673, pruned_loss=0.1202, over 5671055.60 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:37:55,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6711, 1.9505, 1.5811, 1.8040], device='cuda:1'), covar=tensor([0.2488, 0.2582, 0.2842, 0.2289], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1043, 0.1273, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 06:37:59,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2317, 1.4869, 1.5010, 1.3618], device='cuda:1'), covar=tensor([0.1586, 0.1462, 0.1913, 0.1572], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0740, 0.0693, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 06:38:25,547 INFO [train.py:968] (1/2) Epoch 18, batch 8600, giga_loss[loss=0.3521, simple_loss=0.4058, pruned_loss=0.1492, over 28681.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3654, pruned_loss=0.1185, over 5675589.55 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3481, pruned_loss=0.09325, over 5693938.57 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3674, pruned_loss=0.1211, over 5662670.46 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:38:34,831 INFO [optim.py:369] (1/2) 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 06:38:35,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=785324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:38:59,717 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 8650, giga_loss[loss=0.3391, simple_loss=0.397, pruned_loss=0.1406, over 29044.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3665, pruned_loss=0.1196, over 5664364.31 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3481, pruned_loss=0.09329, over 5696507.07 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3688, pruned_loss=0.1227, over 5650555.60 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:39:45,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=785395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:39:58,749 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=785409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:40:00,711 INFO [train.py:968] (1/2) Epoch 18, batch 8700, giga_loss[loss=0.3139, simple_loss=0.3786, pruned_loss=0.1246, over 28261.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3698, pruned_loss=0.1197, over 5669935.56 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3477, pruned_loss=0.09299, over 5698405.42 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3723, pruned_loss=0.1228, over 5656904.18 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:40:15,380 INFO [optim.py:369] (1/2) 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 06:40:20,119 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 06:40:50,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3456, 1.6339, 1.4650, 1.5203], device='cuda:1'), covar=tensor([0.0768, 0.0337, 0.0322, 0.0832], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 06:40:51,098 INFO [train.py:968] (1/2) Epoch 18, batch 8750, giga_loss[loss=0.2639, simple_loss=0.3546, pruned_loss=0.0866, over 28830.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3702, pruned_loss=0.1174, over 5676934.38 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3471, pruned_loss=0.09272, over 5702664.01 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3732, pruned_loss=0.1207, over 5661985.19 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:40:56,725 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:40:59,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:18,704 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:20,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:27,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:35,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2652, 1.4806, 1.3563, 1.1748], device='cuda:1'), covar=tensor([0.2043, 0.2123, 0.1576, 0.1844], device='cuda:1'), in_proj_covar=tensor([0.1896, 0.1838, 0.1760, 0.1894], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 06:41:38,711 INFO [train.py:968] (1/2) Epoch 18, batch 8800, giga_loss[loss=0.2834, simple_loss=0.3603, pruned_loss=0.1033, over 28403.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3722, pruned_loss=0.1189, over 5666747.61 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3471, pruned_loss=0.09285, over 5697218.74 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3752, pruned_loss=0.122, over 5659921.27 frames. ], batch size: 71, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:41:48,495 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:41:50,468 INFO [optim.py:369] (1/2) 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 06:41:59,830 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:42:03,400 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 8850, giga_loss[loss=0.4062, simple_loss=0.4297, pruned_loss=0.1913, over 26750.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3741, pruned_loss=0.121, over 5657130.81 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3465, pruned_loss=0.09257, over 5693360.81 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3777, pruned_loss=0.1243, over 5654900.27 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:42:32,223 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:42:33,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4694, 4.4982, 1.7230, 1.6351], device='cuda:1'), covar=tensor([0.1009, 0.0320, 0.0874, 0.1374], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0544, 0.0368, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 06:43:07,891 INFO [train.py:968] (1/2) Epoch 18, batch 8900, libri_loss[loss=0.2507, simple_loss=0.3405, pruned_loss=0.08045, over 29650.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3737, pruned_loss=0.1209, over 5661703.85 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.346, pruned_loss=0.09219, over 5700651.35 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.378, pruned_loss=0.125, over 5651751.89 frames. ], batch size: 88, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:43:12,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3309, 1.6924, 1.0031, 1.3206], device='cuda:1'), covar=tensor([0.1179, 0.0732, 0.1599, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0445, 0.0511, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 06:43:19,379 INFO [optim.py:369] (1/2) 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 06:43:36,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-09 06:43:49,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5262, 1.7673, 1.4393, 1.7272], device='cuda:1'), covar=tensor([0.2139, 0.2110, 0.2258, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1042, 0.1272, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 06:43:58,382 INFO [train.py:968] (1/2) Epoch 18, batch 8950, giga_loss[loss=0.3043, simple_loss=0.3729, pruned_loss=0.1179, over 29066.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3734, pruned_loss=0.1221, over 5646385.05 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3462, pruned_loss=0.09243, over 5693674.05 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3769, pruned_loss=0.1254, over 5643526.26 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:44:49,160 INFO [train.py:968] (1/2) Epoch 18, batch 9000, giga_loss[loss=0.2998, simple_loss=0.3647, pruned_loss=0.1175, over 28566.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3712, pruned_loss=0.1212, over 5650945.46 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3457, pruned_loss=0.09216, over 5697925.47 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.375, pruned_loss=0.1246, over 5643974.17 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:44:49,160 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 06:44:54,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0923, 1.5625, 1.6165, 1.3750], device='cuda:1'), covar=tensor([0.2035, 0.1573, 0.2072, 0.1791], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0747, 0.0700, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 06:44:57,696 INFO [train.py:1012] (1/2) Epoch 18, validation: loss=0.2106, simple_loss=0.3183, pruned_loss=0.05144, over 944034.00 frames. 
+2023-03-09 06:44:57,697 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 06:45:10,658 INFO [optim.py:369] (1/2) 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 06:45:46,914 INFO [train.py:968] (1/2) Epoch 18, batch 9050, giga_loss[loss=0.3631, simple_loss=0.3958, pruned_loss=0.1652, over 26708.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3701, pruned_loss=0.1214, over 5652334.81 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3458, pruned_loss=0.09213, over 5698325.92 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5645450.72 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:46:07,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4004, 2.1099, 1.5414, 0.6082], device='cuda:1'), covar=tensor([0.5031, 0.2790, 0.3966, 0.5877], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1612, 0.1569, 0.1388], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 06:46:09,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=785784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:46:39,394 INFO [train.py:968] (1/2) Epoch 18, batch 9100, giga_loss[loss=0.3056, simple_loss=0.3746, pruned_loss=0.1183, over 29116.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.122, over 5644695.98 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3462, pruned_loss=0.09236, over 5689676.03 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.373, pruned_loss=0.1247, over 5647260.04 frames. ], batch size: 113, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:46:51,765 INFO [optim.py:369] (1/2) 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 06:47:22,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=785855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:47:26,252 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8038, 1.8804, 1.4171, 1.4989], device='cuda:1'), covar=tensor([0.0990, 0.0776, 0.1055, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0444, 0.0508, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 06:47:29,312 INFO [train.py:968] (1/2) Epoch 18, batch 9150, giga_loss[loss=0.3815, simple_loss=0.4221, pruned_loss=0.1704, over 27862.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3715, pruned_loss=0.1231, over 5630211.67 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3468, pruned_loss=0.09277, over 5680501.21 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1254, over 5639460.94 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:47:42,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6086, 1.9187, 1.7525, 1.7278], device='cuda:1'), covar=tensor([0.1738, 0.1853, 0.2074, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0753, 0.0705, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 06:48:14,585 INFO [train.py:968] (1/2) Epoch 18, batch 9200, giga_loss[loss=0.3191, simple_loss=0.3727, pruned_loss=0.1328, over 28540.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3696, pruned_loss=0.1223, over 5640429.63 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3474, pruned_loss=0.09299, over 5681856.48 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3715, pruned_loss=0.1251, over 5645210.13 frames. ], batch size: 71, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:48:27,018 INFO [optim.py:369] (1/2) 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 06:48:28,349 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=785927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:48:32,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=785930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:49:03,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=785959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:49:05,696 INFO [train.py:968] (1/2) Epoch 18, batch 9250, giga_loss[loss=0.3051, simple_loss=0.3724, pruned_loss=0.119, over 28887.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3685, pruned_loss=0.1215, over 5642462.82 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3474, pruned_loss=0.09297, over 5682970.20 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.37, pruned_loss=0.1238, over 5645075.44 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:49:58,040 INFO [train.py:968] (1/2) Epoch 18, batch 9300, giga_loss[loss=0.3144, simple_loss=0.3766, pruned_loss=0.1261, over 29014.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3692, pruned_loss=0.1209, over 5647421.58 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3475, pruned_loss=0.09306, over 5681324.59 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.123, over 5650573.95 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:50:10,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-09 06:50:11,359 INFO [optim.py:369] (1/2) 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 06:50:19,595 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 9350, giga_loss[loss=0.3046, simple_loss=0.366, pruned_loss=0.1216, over 28733.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1222, over 5654107.35 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3474, pruned_loss=0.09303, over 5683412.63 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3724, pruned_loss=0.1241, over 5654584.37 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:50:54,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3970, 2.2125, 2.1740, 1.8912], device='cuda:1'), covar=tensor([0.1470, 0.2114, 0.1842, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0750, 0.0703, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 06:51:21,107 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-09 06:51:37,113 INFO [train.py:968] (1/2) Epoch 18, batch 9400, giga_loss[loss=0.3672, simple_loss=0.4153, pruned_loss=0.1595, over 28652.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3715, pruned_loss=0.1231, over 5652761.47 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3473, pruned_loss=0.09304, over 5687066.21 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3731, pruned_loss=0.1252, over 5649165.14 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:51:50,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4698, 1.2254, 4.1697, 3.2325], device='cuda:1'), covar=tensor([0.1690, 0.2968, 0.0479, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0634, 0.0936, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 06:51:51,015 INFO [optim.py:369] (1/2) 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 06:52:25,280 INFO [train.py:968] (1/2) Epoch 18, batch 9450, libri_loss[loss=0.2607, simple_loss=0.3482, pruned_loss=0.0866, over 29535.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3728, pruned_loss=0.1212, over 5659501.78 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3474, pruned_loss=0.09309, over 5691138.64 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3746, pruned_loss=0.1234, over 5652162.56 frames. ], batch size: 82, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:53:07,023 INFO [train.py:968] (1/2) Epoch 18, batch 9500, giga_loss[loss=0.2996, simple_loss=0.37, pruned_loss=0.1146, over 28633.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3729, pruned_loss=0.119, over 5672870.04 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.09292, over 5698864.39 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3755, pruned_loss=0.1222, over 5658775.25 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:53:20,830 INFO [optim.py:369] (1/2) 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 06:53:24,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=786230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:53:36,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3767, 1.6572, 1.3110, 1.5334], device='cuda:1'), covar=tensor([0.2658, 0.2593, 0.3041, 0.2098], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1041, 0.1273, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 06:53:59,162 INFO [train.py:968] (1/2) Epoch 18, batch 9550, giga_loss[loss=0.2806, simple_loss=0.3605, pruned_loss=0.1004, over 29032.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3772, pruned_loss=0.1215, over 5676912.89 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3471, pruned_loss=0.09289, over 5699946.07 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3795, pruned_loss=0.1241, over 5664867.92 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 06:54:47,869 INFO [train.py:968] (1/2) Epoch 18, batch 9600, giga_loss[loss=0.3209, simple_loss=0.3889, pruned_loss=0.1264, over 28596.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3814, pruned_loss=0.1258, over 5677801.66 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.347, pruned_loss=0.09282, over 5703147.16 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3839, pruned_loss=0.1283, over 5664949.33 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 06:55:00,216 INFO [optim.py:369] (1/2) 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 06:55:36,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 06:55:37,550 INFO [train.py:968] (1/2) Epoch 18, batch 9650, giga_loss[loss=0.3383, simple_loss=0.3969, pruned_loss=0.1399, over 28677.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3824, pruned_loss=0.1275, over 5671543.09 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3471, pruned_loss=0.09289, over 5706576.41 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3849, pruned_loss=0.1301, over 5658074.41 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:55:50,342 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=786373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:55:51,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-09 06:55:52,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=786376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:56:13,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3216, 3.1424, 1.4840, 1.4694], device='cuda:1'), covar=tensor([0.0989, 0.0357, 0.0899, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0548, 0.0370, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 06:56:20,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=786405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:56:23,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=786409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:56:26,846 INFO [train.py:968] (1/2) Epoch 18, batch 9700, giga_loss[loss=0.3211, simple_loss=0.3969, pruned_loss=0.1226, over 28732.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3821, pruned_loss=0.1276, over 5660955.45 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.09294, over 5698931.98 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3844, pruned_loss=0.13, over 5656922.02 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:56:43,443 INFO [optim.py:369] (1/2) 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 06:56:53,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4269, 1.5939, 1.5726, 1.3916], device='cuda:1'), covar=tensor([0.1676, 0.2006, 0.2011, 0.1975], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0744, 0.0698, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 06:57:12,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4974, 1.4661, 1.5960, 1.1401], device='cuda:1'), covar=tensor([0.2034, 0.3262, 0.1591, 0.1775], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0700, 0.0922, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 06:57:12,383 INFO [train.py:968] (1/2) Epoch 18, batch 9750, giga_loss[loss=0.3114, simple_loss=0.3849, pruned_loss=0.119, over 29037.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3796, pruned_loss=0.125, over 5669496.28 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.09293, over 5701662.59 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.382, pruned_loss=0.1276, over 5663052.21 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:57:56,955 INFO [train.py:968] (1/2) Epoch 18, batch 9800, giga_loss[loss=0.3906, simple_loss=0.4306, pruned_loss=0.1753, over 27916.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3792, pruned_loss=0.123, over 5676944.29 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3469, pruned_loss=0.09276, over 5707712.10 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3824, pruned_loss=0.1261, over 5665393.98 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:58:11,875 INFO [optim.py:369] (1/2) 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 06:58:35,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=786552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:58:36,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=786555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:58:41,721 INFO [train.py:968] (1/2) Epoch 18, batch 9850, giga_loss[loss=0.299, simple_loss=0.37, pruned_loss=0.114, over 28867.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3806, pruned_loss=0.1236, over 5681203.89 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3468, pruned_loss=0.09284, over 5711760.52 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3838, pruned_loss=0.1266, over 5667796.46 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:59:08,758 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=786584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 06:59:34,416 INFO [train.py:968] (1/2) Epoch 18, batch 9900, giga_loss[loss=0.3308, simple_loss=0.3828, pruned_loss=0.1394, over 27880.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3816, pruned_loss=0.1249, over 5666017.91 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3471, pruned_loss=0.09296, over 5706032.72 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3846, pruned_loss=0.1277, over 5660359.64 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 06:59:35,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-09 06:59:51,065 INFO [optim.py:369] (1/2) 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 07:00:02,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3365, 1.4026, 4.0329, 3.1981], device='cuda:1'), covar=tensor([0.1650, 0.2536, 0.0417, 0.0953], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0629, 0.0930, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:00:03,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2039, 0.8066, 0.9255, 1.3412], device='cuda:1'), covar=tensor([0.0793, 0.0395, 0.0360, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 07:00:24,840 INFO [train.py:968] (1/2) Epoch 18, batch 9950, libri_loss[loss=0.3215, simple_loss=0.3915, pruned_loss=0.1257, over 29520.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3806, pruned_loss=0.1248, over 5660600.69 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3473, pruned_loss=0.0931, over 5710086.85 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3834, pruned_loss=0.1275, over 5651368.15 frames. ], batch size: 89, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:01:09,062 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 10000, giga_loss[loss=0.3584, simple_loss=0.3988, pruned_loss=0.159, over 26676.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.38, pruned_loss=0.1261, over 5652422.71 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3472, pruned_loss=0.0931, over 5712097.85 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3828, pruned_loss=0.1287, over 5642726.95 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:01:32,282 INFO [optim.py:369] (1/2) 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 07:02:06,319 INFO [train.py:968] (1/2) Epoch 18, batch 10050, giga_loss[loss=0.3098, simple_loss=0.3716, pruned_loss=0.1241, over 27980.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3767, pruned_loss=0.1244, over 5666086.40 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3472, pruned_loss=0.09311, over 5716302.63 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3796, pruned_loss=0.1272, over 5653486.16 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:02:13,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-09 07:02:55,142 INFO [train.py:968] (1/2) Epoch 18, batch 10100, giga_loss[loss=0.2895, simple_loss=0.3605, pruned_loss=0.1093, over 29093.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3747, pruned_loss=0.1236, over 5660814.60 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3479, pruned_loss=0.09363, over 5720094.91 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3772, pruned_loss=0.1263, over 5646133.44 frames. ], batch size: 155, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:03:12,392 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 10150, giga_loss[loss=0.2937, simple_loss=0.3617, pruned_loss=0.1129, over 28775.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5664530.55 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3482, pruned_loss=0.09366, over 5724692.51 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3762, pruned_loss=0.1264, over 5647194.39 frames. ], batch size: 66, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:04:32,133 INFO [train.py:968] (1/2) Epoch 18, batch 10200, giga_loss[loss=0.2762, simple_loss=0.3568, pruned_loss=0.09778, over 29004.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3718, pruned_loss=0.122, over 5659038.75 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3486, pruned_loss=0.09393, over 5718003.21 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3737, pruned_loss=0.1244, over 5650018.52 frames. ], batch size: 164, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:04:34,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3043, 1.0619, 4.0455, 3.2200], device='cuda:1'), covar=tensor([0.1746, 0.2997, 0.0487, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0736, 0.0633, 0.0936, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:04:47,525 INFO [optim.py:369] (1/2) 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 07:05:05,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3107, 1.6338, 1.3460, 1.4834], device='cuda:1'), covar=tensor([0.0781, 0.0364, 0.0352, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 07:05:19,543 INFO [train.py:968] (1/2) Epoch 18, batch 10250, giga_loss[loss=0.2708, simple_loss=0.3509, pruned_loss=0.09534, over 28708.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3677, pruned_loss=0.1174, over 5660782.62 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3486, pruned_loss=0.09388, over 5717097.52 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3696, pruned_loss=0.1198, over 5653738.24 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:05:41,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8783, 1.1505, 1.0819, 0.8007], device='cuda:1'), covar=tensor([0.2272, 0.2462, 0.1442, 0.2233], device='cuda:1'), in_proj_covar=tensor([0.1891, 0.1835, 0.1765, 0.1898], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 07:06:05,041 INFO [train.py:968] (1/2) Epoch 18, batch 10300, giga_loss[loss=0.283, simple_loss=0.3607, pruned_loss=0.1027, over 29085.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3648, pruned_loss=0.1145, over 5660559.83 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3486, pruned_loss=0.09381, over 5714834.36 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.367, pruned_loss=0.1173, over 5653969.17 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:06:06,001 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:06:21,937 INFO [optim.py:369] (1/2) 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 07:06:33,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2900, 4.1428, 3.9281, 1.9270], device='cuda:1'), covar=tensor([0.0587, 0.0704, 0.0713, 0.2075], device='cuda:1'), in_proj_covar=tensor([0.1195, 0.1102, 0.0949, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:06:54,943 INFO [train.py:968] (1/2) Epoch 18, batch 10350, libri_loss[loss=0.3145, simple_loss=0.3938, pruned_loss=0.1176, over 29537.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3653, pruned_loss=0.1146, over 5669404.12 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.349, pruned_loss=0.09412, over 5719011.32 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3669, pruned_loss=0.1169, over 5659520.48 frames. ], batch size: 84, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:07:02,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-09 07:07:13,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:07:45,863 INFO [train.py:968] (1/2) Epoch 18, batch 10400, giga_loss[loss=0.3124, simple_loss=0.3722, pruned_loss=0.1262, over 28690.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3643, pruned_loss=0.1154, over 5664437.88 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3493, pruned_loss=0.09431, over 5719964.85 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3658, pruned_loss=0.1177, over 5654040.92 frames. ], batch size: 66, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:07:59,420 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:08:00,638 INFO [optim.py:369] (1/2) 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 07:08:34,940 INFO [train.py:968] (1/2) Epoch 18, batch 10450, giga_loss[loss=0.2784, simple_loss=0.3565, pruned_loss=0.1002, over 28556.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3619, pruned_loss=0.1144, over 5661001.86 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3494, pruned_loss=0.09436, over 5718795.03 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3633, pruned_loss=0.1167, over 5652839.25 frames. ], batch size: 60, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:08:36,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5976, 1.7152, 1.6371, 1.5909], device='cuda:1'), covar=tensor([0.1763, 0.2240, 0.2264, 0.2015], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0747, 0.0701, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 07:09:01,353 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:09:16,397 INFO [train.py:968] (1/2) Epoch 18, batch 10500, giga_loss[loss=0.3275, simple_loss=0.3902, pruned_loss=0.1324, over 27985.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3632, pruned_loss=0.1141, over 5664452.21 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3495, pruned_loss=0.09427, over 5718477.90 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3647, pruned_loss=0.1166, over 5656258.19 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:09:26,130 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787222.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:09:30,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787225.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:09:33,463 INFO [optim.py:369] (1/2) 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 07:09:57,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787254.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:09:57,467 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2588, 1.4917, 1.4147, 1.2810], device='cuda:1'), covar=tensor([0.1371, 0.1265, 0.1790, 0.1374], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0748, 0.0703, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 07:10:05,118 INFO [train.py:968] (1/2) Epoch 18, batch 10550, giga_loss[loss=0.3016, simple_loss=0.3658, pruned_loss=0.1188, over 29031.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3658, pruned_loss=0.1157, over 5662099.54 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3495, pruned_loss=0.09418, over 5722492.01 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3673, pruned_loss=0.1183, over 5650664.40 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:10:22,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 07:10:49,997 INFO [train.py:968] (1/2) Epoch 18, batch 10600, giga_loss[loss=0.2826, simple_loss=0.3533, pruned_loss=0.1059, over 28956.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3655, pruned_loss=0.1158, over 5657481.46 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3495, pruned_loss=0.09409, over 5725578.40 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3673, pruned_loss=0.1187, over 5643325.75 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:11:04,440 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-09 07:11:04,631 INFO [optim.py:369] (1/2) 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 07:11:36,574 INFO [train.py:968] (1/2) Epoch 18, batch 10650, giga_loss[loss=0.2626, simple_loss=0.3384, pruned_loss=0.09334, over 28809.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3653, pruned_loss=0.1157, over 5655161.68 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3502, pruned_loss=0.09443, over 5719879.66 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3666, pruned_loss=0.1184, over 5646813.80 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:11:40,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-09 07:12:00,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:12:26,447 INFO [train.py:968] (1/2) Epoch 18, batch 10700, giga_loss[loss=0.3064, simple_loss=0.3628, pruned_loss=0.125, over 28830.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3677, pruned_loss=0.1178, over 5660800.69 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3504, pruned_loss=0.09452, over 5719057.12 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5653974.60 frames. ], batch size: 99, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:12:41,645 INFO [optim.py:369] (1/2) 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 07:12:47,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4387, 1.3364, 3.7936, 3.2941], device='cuda:1'), covar=tensor([0.1406, 0.2614, 0.0427, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0631, 0.0928, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:13:12,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6633, 1.8247, 1.3687, 1.3778], device='cuda:1'), covar=tensor([0.0906, 0.0590, 0.1026, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0445, 0.0509, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:13:16,858 INFO [train.py:968] (1/2) Epoch 18, batch 10750, giga_loss[loss=0.3152, simple_loss=0.38, pruned_loss=0.1252, over 28953.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3702, pruned_loss=0.1193, over 5661926.52 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3508, pruned_loss=0.09467, over 5722302.52 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3711, pruned_loss=0.1215, over 5652716.05 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:13:24,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:13:30,570 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:13:54,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:04,295 INFO [train.py:968] (1/2) Epoch 18, batch 10800, giga_loss[loss=0.296, simple_loss=0.3632, pruned_loss=0.1144, over 28983.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3725, pruned_loss=0.121, over 5657748.27 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3508, pruned_loss=0.09467, over 5716902.73 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3735, pruned_loss=0.1233, over 5654632.31 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:14:23,497 INFO [optim.py:369] (1/2) 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 07:14:23,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:27,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:30,352 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=787538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:14:48,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 07:14:52,701 INFO [train.py:968] (1/2) Epoch 18, batch 10850, giga_loss[loss=0.3204, simple_loss=0.3791, pruned_loss=0.1309, over 28353.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.373, pruned_loss=0.122, over 5660992.87 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3505, pruned_loss=0.09459, over 5712479.69 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1244, over 5661689.30 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:14:55,011 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:15:01,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:15:19,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 07:15:43,175 INFO [train.py:968] (1/2) Epoch 18, batch 10900, giga_loss[loss=0.3057, simple_loss=0.3817, pruned_loss=0.1149, over 28413.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3733, pruned_loss=0.1213, over 5667457.80 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3499, pruned_loss=0.09434, over 5717081.13 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3756, pruned_loss=0.1241, over 5662834.87 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:16:06,297 INFO [optim.py:369] (1/2) 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 07:16:20,236 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:16:23,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:16:30,027 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-09 07:16:38,345 INFO [train.py:968] (1/2) Epoch 18, batch 10950, giga_loss[loss=0.3679, simple_loss=0.42, pruned_loss=0.1579, over 28699.00 frames. ], tot_loss[loss=0.308, simple_loss=0.374, pruned_loss=0.121, over 5655839.95 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09457, over 5711356.57 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.376, pruned_loss=0.1235, over 5656987.00 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:16:52,343 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:17:14,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7563, 1.9057, 1.8737, 1.6576], device='cuda:1'), covar=tensor([0.1751, 0.1993, 0.2125, 0.2143], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0745, 0.0701, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 07:17:28,146 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:17:28,424 INFO [train.py:968] (1/2) Epoch 18, batch 11000, giga_loss[loss=0.328, simple_loss=0.3886, pruned_loss=0.1337, over 28545.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.374, pruned_loss=0.1219, over 5653277.59 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09453, over 5717532.42 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3766, pruned_loss=0.1249, over 5646841.06 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:17:28,865 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 07:17:30,251 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:17:45,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2993, 1.1289, 1.2301, 1.4033], device='cuda:1'), covar=tensor([0.0606, 0.0315, 0.0266, 0.0615], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 07:17:57,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2517, 4.0950, 3.8879, 1.9977], device='cuda:1'), covar=tensor([0.0582, 0.0738, 0.0770, 0.2075], device='cuda:1'), in_proj_covar=tensor([0.1180, 0.1093, 0.0943, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:17:58,893 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=787743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:18:23,088 INFO [train.py:968] (1/2) Epoch 18, batch 11050, giga_loss[loss=0.2956, simple_loss=0.3564, pruned_loss=0.1174, over 28658.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3735, pruned_loss=0.123, over 5639705.74 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3499, pruned_loss=0.09441, over 5719485.81 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3759, pruned_loss=0.1257, over 5632517.65 frames. ], batch size: 85, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:19:15,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4416, 1.7656, 1.5190, 1.6563], device='cuda:1'), covar=tensor([0.0633, 0.0264, 0.0277, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 07:19:19,463 INFO [train.py:968] (1/2) Epoch 18, batch 11100, giga_loss[loss=0.2657, simple_loss=0.3359, pruned_loss=0.09771, over 28874.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1223, over 5637941.97 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3501, pruned_loss=0.09457, over 5712156.65 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3738, pruned_loss=0.1245, over 5638460.38 frames. ], batch size: 112, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:19:33,181 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-09 07:19:35,432 INFO [optim.py:369] (1/2) 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 07:19:47,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:19:51,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:20:00,743 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 11150, giga_loss[loss=0.2881, simple_loss=0.3562, pruned_loss=0.1099, over 28599.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3709, pruned_loss=0.1223, over 5641887.71 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3498, pruned_loss=0.09445, over 5715013.89 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3733, pruned_loss=0.125, over 5637585.66 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 2.0
+2023-03-09 07:20:46,245 INFO [train.py:968] (1/2) Epoch 18, batch 11200, giga_loss[loss=0.3329, simple_loss=0.3879, pruned_loss=0.1389, over 27993.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3699, pruned_loss=0.1211, over 5657356.52 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3502, pruned_loss=0.09451, over 5715796.05 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3722, pruned_loss=0.1241, over 5651028.57 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:20:47,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=787913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:20:53,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2844, 1.8193, 1.3293, 1.3682], device='cuda:1'), covar=tensor([0.2577, 0.2432, 0.2959, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1053, 0.1283, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 07:21:03,778 INFO [optim.py:369] (1/2) 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 07:21:04,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9415, 3.7278, 3.5338, 1.6784], device='cuda:1'), covar=tensor([0.0898, 0.1119, 0.1220, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.1185, 0.1100, 0.0949, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:21:32,109 INFO [train.py:968] (1/2) Epoch 18, batch 11250, giga_loss[loss=0.2776, simple_loss=0.3538, pruned_loss=0.1006, over 28698.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3698, pruned_loss=0.1212, over 5659878.53 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3504, pruned_loss=0.09451, over 5721482.09 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.372, pruned_loss=0.1245, over 5647854.93 frames. ], batch size: 60, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:21:51,898 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-09 07:21:56,367 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:21:57,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8548, 4.7203, 1.8914, 1.9859], device='cuda:1'), covar=tensor([0.0907, 0.0266, 0.0849, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0546, 0.0371, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 07:21:59,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:02,560 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:04,569 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=787995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:08,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=787999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:10,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=788002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:14,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1084, 3.1453, 2.1206, 1.1429], device='cuda:1'), covar=tensor([0.5827, 0.2556, 0.3055, 0.5759], device='cuda:1'), in_proj_covar=tensor([0.1691, 0.1609, 0.1577, 0.1386], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 07:22:20,445 INFO [train.py:968] (1/2) Epoch 18, batch 11300, giga_loss[loss=0.318, simple_loss=0.3792, pruned_loss=0.1284, over 28839.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1217, over 5649194.50 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3502, pruned_loss=0.09429, over 5720429.20 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.372, pruned_loss=0.1249, over 5639815.34 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:22:25,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:31,530 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:22:36,628 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:22:59,810 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=788056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:23:03,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=788059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:23:05,206 INFO [train.py:968] (1/2) Epoch 18, batch 11350, libri_loss[loss=0.2678, simple_loss=0.3526, pruned_loss=0.09153, over 28663.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.123, over 5658923.24 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3495, pruned_loss=0.09388, over 5720745.99 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3741, pruned_loss=0.1266, over 5649235.20 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:23:33,113 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:23:56,827 INFO [train.py:968] (1/2) Epoch 18, batch 11400, giga_loss[loss=0.3386, simple_loss=0.3861, pruned_loss=0.1455, over 29086.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3724, pruned_loss=0.1247, over 5641968.02 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3492, pruned_loss=0.09375, over 5721466.26 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3751, pruned_loss=0.1278, over 5633550.29 frames. ], batch size: 113, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:24:13,399 INFO [optim.py:369] (1/2) 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 07:24:26,729 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-09 07:24:45,151 INFO [train.py:968] (1/2) Epoch 18, batch 11450, giga_loss[loss=0.2868, simple_loss=0.3561, pruned_loss=0.1088, over 28478.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3735, pruned_loss=0.1255, over 5656403.11 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3497, pruned_loss=0.0941, over 5723563.32 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3755, pruned_loss=0.1281, over 5646953.33 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:24:57,334 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-09 07:25:29,833 INFO [train.py:968] (1/2) Epoch 18, batch 11500, giga_loss[loss=0.4662, simple_loss=0.4661, pruned_loss=0.2331, over 26625.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3741, pruned_loss=0.1261, over 5645682.38 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3497, pruned_loss=0.09402, over 5716602.53 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3763, pruned_loss=0.1289, over 5643115.50 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:25:48,257 INFO [optim.py:369] (1/2) 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 07:25:52,806 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=788235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:26:09,404 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4579, 1.7472, 1.3991, 1.4520], device='cuda:1'), covar=tensor([0.2260, 0.2268, 0.2467, 0.2047], device='cuda:1'), in_proj_covar=tensor([0.1446, 0.1052, 0.1281, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 07:26:11,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7348, 2.0754, 1.5860, 2.1566], device='cuda:1'), covar=tensor([0.2472, 0.2531, 0.2854, 0.2267], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1052, 0.1282, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 07:26:17,217 INFO [train.py:968] (1/2) Epoch 18, batch 11550, giga_loss[loss=0.3001, simple_loss=0.3772, pruned_loss=0.1114, over 28976.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3743, pruned_loss=0.1256, over 5653758.96 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3495, pruned_loss=0.09406, over 5720122.56 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3766, pruned_loss=0.1285, over 5647445.17 frames. ], batch size: 164, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:27:00,792 INFO [train.py:968] (1/2) Epoch 18, batch 11600, giga_loss[loss=0.361, simple_loss=0.4069, pruned_loss=0.1575, over 28670.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3736, pruned_loss=0.1241, over 5667138.77 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3497, pruned_loss=0.09407, over 5722085.57 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3761, pruned_loss=0.1273, over 5658711.48 frames. ], batch size: 242, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:27:18,564 INFO [optim.py:369] (1/2) 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 07:27:24,924 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3787, 1.5293, 1.2917, 1.4961], device='cuda:1'), covar=tensor([0.0674, 0.0428, 0.0342, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 07:27:48,411 INFO [train.py:968] (1/2) Epoch 18, batch 11650, giga_loss[loss=0.334, simple_loss=0.3929, pruned_loss=0.1376, over 28650.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3763, pruned_loss=0.1266, over 5663705.15 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3493, pruned_loss=0.09375, over 5726180.58 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3792, pruned_loss=0.1302, over 5651901.89 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:28:28,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-09 07:28:36,742 INFO [train.py:968] (1/2) Epoch 18, batch 11700, giga_loss[loss=0.2873, simple_loss=0.3537, pruned_loss=0.1104, over 28856.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3772, pruned_loss=0.1276, over 5664263.21 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3492, pruned_loss=0.09367, over 5729436.02 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3801, pruned_loss=0.1312, over 5650636.60 frames. ], batch size: 119, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:28:48,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8776, 1.8778, 1.4691, 1.4807], device='cuda:1'), covar=tensor([0.0925, 0.0754, 0.1019, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0446, 0.0510, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:28:53,797 INFO [optim.py:369] (1/2) 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 07:29:20,042 INFO [train.py:968] (1/2) Epoch 18, batch 11750, libri_loss[loss=0.3077, simple_loss=0.3766, pruned_loss=0.1194, over 18969.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3764, pruned_loss=0.1273, over 5644488.52 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.09361, over 5721713.25 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.38, pruned_loss=0.1314, over 5638712.87 frames. ], batch size: 190, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:29:24,367 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=788467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:29:39,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3496, 1.8493, 1.5596, 1.5761], device='cuda:1'), covar=tensor([0.0796, 0.0293, 0.0321, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 07:30:07,547 INFO [train.py:968] (1/2) Epoch 18, batch 11800, giga_loss[loss=0.3246, simple_loss=0.3835, pruned_loss=0.1328, over 27946.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3767, pruned_loss=0.1259, over 5648111.43 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3488, pruned_loss=0.09355, over 5721853.21 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3798, pruned_loss=0.1295, over 5642906.52 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:30:24,385 INFO [optim.py:369] (1/2) 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 07:30:43,843 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 07:30:51,907 INFO [train.py:968] (1/2) Epoch 18, batch 11850, giga_loss[loss=0.3165, simple_loss=0.3798, pruned_loss=0.1266, over 28965.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3758, pruned_loss=0.1248, over 5656108.49 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3484, pruned_loss=0.09332, over 5727520.12 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3794, pruned_loss=0.1288, over 5644808.23 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:31:21,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4995, 4.3331, 4.0678, 2.0524], device='cuda:1'), covar=tensor([0.0683, 0.0837, 0.0979, 0.2020], device='cuda:1'), in_proj_covar=tensor([0.1187, 0.1101, 0.0948, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:31:34,458 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 11900, giga_loss[loss=0.3502, simple_loss=0.404, pruned_loss=0.1482, over 27488.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3741, pruned_loss=0.1234, over 5652952.99 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3481, pruned_loss=0.09311, over 5732333.34 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3782, pruned_loss=0.1278, over 5637133.89 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:31:42,292 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-09 07:31:50,618 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 11950, giga_loss[loss=0.3591, simple_loss=0.414, pruned_loss=0.1521, over 28230.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.372, pruned_loss=0.1217, over 5665532.44 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3481, pruned_loss=0.09302, over 5732671.39 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3758, pruned_loss=0.1259, over 5651043.70 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:33:08,640 INFO [train.py:968] (1/2) Epoch 18, batch 12000, giga_loss[loss=0.294, simple_loss=0.3625, pruned_loss=0.1128, over 28679.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3739, pruned_loss=0.1234, over 5655416.50 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.348, pruned_loss=0.09294, over 5736731.64 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3777, pruned_loss=0.1276, over 5638628.50 frames. ], batch size: 242, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:33:08,640 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 07:33:17,635 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 07:33:34,541 INFO [optim.py:369] (1/2) 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 07:33:56,009 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=788753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:33:58,541 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 12050, giga_loss[loss=0.3339, simple_loss=0.3873, pruned_loss=0.1402, over 28569.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3757, pruned_loss=0.1251, over 5664754.92 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.348, pruned_loss=0.09293, over 5740950.77 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3793, pruned_loss=0.1292, over 5645775.94 frames. ], batch size: 78, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:34:25,072 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=788785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:34:49,135 INFO [train.py:968] (1/2) Epoch 18, batch 12100, giga_loss[loss=0.2969, simple_loss=0.3664, pruned_loss=0.1137, over 28698.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.124, over 5681227.00 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3482, pruned_loss=0.093, over 5744070.77 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3772, pruned_loss=0.1279, over 5661819.03 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:35:10,105 INFO [optim.py:369] (1/2) 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 07:35:18,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=788842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:35:38,112 INFO [train.py:968] (1/2) Epoch 18, batch 12150, giga_loss[loss=0.3603, simple_loss=0.4055, pruned_loss=0.1576, over 28640.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3745, pruned_loss=0.1245, over 5676291.92 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3482, pruned_loss=0.09287, over 5745730.71 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1282, over 5658247.22 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:36:25,839 INFO [train.py:968] (1/2) Epoch 18, batch 12200, libri_loss[loss=0.2877, simple_loss=0.3747, pruned_loss=0.1003, over 29364.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3755, pruned_loss=0.1252, over 5677481.63 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3481, pruned_loss=0.09289, over 5746766.60 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3785, pruned_loss=0.1286, over 5661274.69 frames. ], batch size: 92, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:36:42,769 INFO [optim.py:369] (1/2) 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 07:37:09,076 INFO [train.py:968] (1/2) Epoch 18, batch 12250, giga_loss[loss=0.3145, simple_loss=0.3762, pruned_loss=0.1264, over 28630.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3764, pruned_loss=0.1259, over 5674232.33 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3485, pruned_loss=0.09303, over 5751689.02 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3795, pruned_loss=0.1298, over 5654074.70 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:37:30,069 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=788985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:37:33,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=788988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:37:49,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5179, 1.7508, 1.6209, 1.5334], device='cuda:1'), covar=tensor([0.1802, 0.1998, 0.2336, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0749, 0.0705, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 07:37:57,030 INFO [train.py:968] (1/2) Epoch 18, batch 12300, giga_loss[loss=0.3037, simple_loss=0.3511, pruned_loss=0.1281, over 23479.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3742, pruned_loss=0.1233, over 5685286.41 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3486, pruned_loss=0.09308, over 5756540.82 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3773, pruned_loss=0.1272, over 5662879.05 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:38:02,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:38:16,214 INFO [optim.py:369] (1/2) 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 07:38:33,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5408, 1.6802, 1.7270, 1.3425], device='cuda:1'), covar=tensor([0.1795, 0.2468, 0.1442, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0699, 0.0921, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 07:38:44,548 INFO [train.py:968] (1/2) Epoch 18, batch 12350, libri_loss[loss=0.285, simple_loss=0.3514, pruned_loss=0.1093, over 29656.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3743, pruned_loss=0.1228, over 5684625.97 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3489, pruned_loss=0.09331, over 5760765.64 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3771, pruned_loss=0.1264, over 5661028.46 frames. ], batch size: 73, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:38:55,594 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-09 07:39:17,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2479, 1.4017, 1.3176, 1.1551], device='cuda:1'), covar=tensor([0.2160, 0.2221, 0.1475, 0.1960], device='cuda:1'), in_proj_covar=tensor([0.1903, 0.1854, 0.1778, 0.1921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 07:39:23,351 INFO [train.py:968] (1/2) Epoch 18, batch 12400, giga_loss[loss=0.3141, simple_loss=0.3769, pruned_loss=0.1256, over 28916.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3738, pruned_loss=0.1217, over 5696579.21 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3492, pruned_loss=0.09359, over 5767740.48 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1257, over 5668326.70 frames. ], batch size: 227, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:39:45,693 INFO [optim.py:369] (1/2) 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 07:39:49,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-09 07:39:52,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9081, 1.9528, 1.5415, 1.6299], device='cuda:1'), covar=tensor([0.0926, 0.0758, 0.0986, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0446, 0.0509, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:40:16,880 INFO [train.py:968] (1/2) Epoch 18, batch 12450, giga_loss[loss=0.3278, simple_loss=0.3889, pruned_loss=0.1333, over 28916.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.373, pruned_loss=0.122, over 5682343.34 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3492, pruned_loss=0.09359, over 5767740.48 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3754, pruned_loss=0.1251, over 5660354.06 frames. ], batch size: 106, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:40:21,795 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789168.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:40:31,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:40:37,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2767, 4.1186, 3.9099, 1.6251], device='cuda:1'), covar=tensor([0.0588, 0.0740, 0.0796, 0.2140], device='cuda:1'), in_proj_covar=tensor([0.1194, 0.1105, 0.0951, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:41:03,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4779, 1.5866, 1.5540, 1.4137], device='cuda:1'), covar=tensor([0.1537, 0.1973, 0.2122, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0750, 0.0708, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 07:41:04,253 INFO [train.py:968] (1/2) Epoch 18, batch 12500, giga_loss[loss=0.2925, simple_loss=0.3541, pruned_loss=0.1155, over 29120.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1231, over 5676136.61 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3493, pruned_loss=0.09363, over 5770152.74 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3754, pruned_loss=0.1259, over 5655579.31 frames. ], batch size: 113, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:41:07,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7014, 4.5365, 4.3143, 2.2603], device='cuda:1'), covar=tensor([0.0593, 0.0777, 0.0806, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1194, 0.1106, 0.0952, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:41:22,971 INFO [optim.py:369] (1/2) 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 07:41:47,984 INFO [train.py:968] (1/2) Epoch 18, batch 12550, giga_loss[loss=0.2891, simple_loss=0.3511, pruned_loss=0.1135, over 28972.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3702, pruned_loss=0.1215, over 5682826.64 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.349, pruned_loss=0.09334, over 5770403.32 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3728, pruned_loss=0.1248, over 5663751.46 frames. ], batch size: 164, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:42:37,743 INFO [train.py:968] (1/2) Epoch 18, batch 12600, giga_loss[loss=0.3132, simple_loss=0.375, pruned_loss=0.1257, over 28331.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.367, pruned_loss=0.1199, over 5685498.68 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3492, pruned_loss=0.09343, over 5761633.63 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3694, pruned_loss=0.123, over 5677214.12 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:42:54,821 INFO [optim.py:369] (1/2) 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 07:43:16,378 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:43:16,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5275, 1.7642, 1.6542, 1.4248], device='cuda:1'), covar=tensor([0.2783, 0.2277, 0.1927, 0.2507], device='cuda:1'), in_proj_covar=tensor([0.1904, 0.1855, 0.1784, 0.1927], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 07:43:20,895 INFO [train.py:968] (1/2) Epoch 18, batch 12650, giga_loss[loss=0.2744, simple_loss=0.3396, pruned_loss=0.1046, over 29050.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3649, pruned_loss=0.1187, over 5691599.28 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3488, pruned_loss=0.09325, over 5762305.94 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3677, pruned_loss=0.1222, over 5682521.84 frames. ], batch size: 128, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:43:36,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4579, 1.6258, 1.6988, 1.2778], device='cuda:1'), covar=tensor([0.1651, 0.2437, 0.1391, 0.1625], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0699, 0.0922, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 07:43:37,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5088, 1.7226, 1.7062, 1.3788], device='cuda:1'), covar=tensor([0.2543, 0.2268, 0.1702, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1904, 0.1858, 0.1786, 0.1929], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 07:44:06,328 INFO [train.py:968] (1/2) Epoch 18, batch 12700, giga_loss[loss=0.2857, simple_loss=0.3591, pruned_loss=0.1062, over 28696.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.364, pruned_loss=0.1175, over 5692658.36 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3489, pruned_loss=0.09313, over 5767960.53 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3666, pruned_loss=0.1213, over 5677872.24 frames. ], batch size: 242, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:44:28,673 INFO [optim.py:369] (1/2) 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 07:44:54,920 INFO [train.py:968] (1/2) Epoch 18, batch 12750, giga_loss[loss=0.266, simple_loss=0.349, pruned_loss=0.09147, over 28516.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1156, over 5689370.20 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3488, pruned_loss=0.09319, over 5770803.96 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5673648.63 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:45:46,870 INFO [train.py:968] (1/2) Epoch 18, batch 12800, giga_loss[loss=0.2827, simple_loss=0.3604, pruned_loss=0.1025, over 28775.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.362, pruned_loss=0.113, over 5680321.18 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3487, pruned_loss=0.0931, over 5772095.51 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3642, pruned_loss=0.1159, over 5666227.39 frames. ], batch size: 174, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:45:54,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:46:06,967 INFO [optim.py:369] (1/2) 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 07:46:17,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789543.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:46:28,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:46:36,695 INFO [train.py:968] (1/2) Epoch 18, batch 12850, giga_loss[loss=0.2465, simple_loss=0.3274, pruned_loss=0.0828, over 27905.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3597, pruned_loss=0.1105, over 5674184.13 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3482, pruned_loss=0.09295, over 5773277.17 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3619, pruned_loss=0.1131, over 5660871.99 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:47:05,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7485, 4.5595, 4.3325, 2.1465], device='cuda:1'), covar=tensor([0.0463, 0.0629, 0.0769, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.1181, 0.1093, 0.0939, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:47:16,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6421, 1.6595, 1.3346, 1.3157], device='cuda:1'), covar=tensor([0.0686, 0.0383, 0.0754, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0444, 0.0508, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:47:29,270 INFO [train.py:968] (1/2) Epoch 18, batch 12900, giga_loss[loss=0.33, simple_loss=0.3875, pruned_loss=0.1363, over 28924.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3562, pruned_loss=0.1072, over 5667434.24 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3482, pruned_loss=0.09298, over 5773007.42 frames. ], giga_tot_loss[loss=0.2886, simple_loss=0.3581, pruned_loss=0.1095, over 5655939.18 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:47:53,020 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 12950, giga_loss[loss=0.201, simple_loss=0.2748, pruned_loss=0.0636, over 24219.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3541, pruned_loss=0.1046, over 5671208.30 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3482, pruned_loss=0.09323, over 5772843.03 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3558, pruned_loss=0.1066, over 5658735.04 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:48:40,340 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=789686.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:48:40,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3627, 2.9371, 2.5728, 2.0538], device='cuda:1'), covar=tensor([0.2131, 0.1277, 0.1471, 0.1990], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1819, 0.1745, 0.1888], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 07:48:43,317 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=789689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:48:43,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=789689.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:48:51,922 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=789697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:48:54,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=789700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:49:06,227 INFO [train.py:968] (1/2) Epoch 18, batch 13000, giga_loss[loss=0.2844, simple_loss=0.3586, pruned_loss=0.1051, over 28746.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3544, pruned_loss=0.1031, over 5665662.63 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3481, pruned_loss=0.09317, over 5774185.44 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3559, pruned_loss=0.1049, over 5653180.31 frames. ], batch size: 243, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:49:12,260 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789718.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 07:49:25,568 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:49:26,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:49:30,619 INFO [optim.py:369] (1/2) 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 07:49:34,761 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-09 07:49:55,604 INFO [train.py:968] (1/2) Epoch 18, batch 13050, giga_loss[loss=0.2786, simple_loss=0.3489, pruned_loss=0.1042, over 28610.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3547, pruned_loss=0.1034, over 5666702.29 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3482, pruned_loss=0.09342, over 5773905.59 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3561, pruned_loss=0.1049, over 5654561.47 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:50:43,596 INFO [train.py:968] (1/2) Epoch 18, batch 13100, giga_loss[loss=0.2843, simple_loss=0.3575, pruned_loss=0.1055, over 28892.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 5665193.69 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3475, pruned_loss=0.09314, over 5776619.56 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3544, pruned_loss=0.104, over 5651288.23 frames. ], batch size: 285, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:51:07,343 INFO [optim.py:369] (1/2) 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 07:51:29,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1984, 1.0798, 3.5943, 3.0969], device='cuda:1'), covar=tensor([0.1677, 0.2902, 0.0482, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0632, 0.0931, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:51:31,705 INFO [train.py:968] (1/2) Epoch 18, batch 13150, giga_loss[loss=0.2472, simple_loss=0.3354, pruned_loss=0.07953, over 28858.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3492, pruned_loss=0.09975, over 5664688.27 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.347, pruned_loss=0.09296, over 5767786.87 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3512, pruned_loss=0.1014, over 5659395.29 frames. ], batch size: 145, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:51:43,953 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=789873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:51:46,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=789876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:52:04,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=789893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:52:13,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=789905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:52:20,276 INFO [train.py:968] (1/2) Epoch 18, batch 13200, giga_loss[loss=0.2511, simple_loss=0.315, pruned_loss=0.09365, over 24193.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09874, over 5663225.84 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3464, pruned_loss=0.09287, over 5769686.96 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3497, pruned_loss=0.1002, over 5655840.86 frames. ], batch size: 705, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:52:23,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4912, 1.2952, 4.2842, 3.4724], device='cuda:1'), covar=tensor([0.1611, 0.2826, 0.0420, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0732, 0.0630, 0.0930, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 07:52:37,954 INFO [optim.py:369] (1/2) 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 07:53:00,852 INFO [train.py:968] (1/2) Epoch 18, batch 13250, giga_loss[loss=0.2494, simple_loss=0.3305, pruned_loss=0.08414, over 28937.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3459, pruned_loss=0.0973, over 5671762.23 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3454, pruned_loss=0.09259, over 5767936.02 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3488, pruned_loss=0.09914, over 5661844.93 frames. ], batch size: 112, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:53:39,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0061, 3.8394, 3.6278, 2.0418], device='cuda:1'), covar=tensor([0.0567, 0.0730, 0.0763, 0.2218], device='cuda:1'), in_proj_covar=tensor([0.1165, 0.1075, 0.0922, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 07:53:46,698 INFO [train.py:968] (1/2) Epoch 18, batch 13300, libri_loss[loss=0.2743, simple_loss=0.3533, pruned_loss=0.09764, over 29282.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3449, pruned_loss=0.09633, over 5664660.27 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3451, pruned_loss=0.09261, over 5761774.84 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3475, pruned_loss=0.0979, over 5660270.29 frames. ], batch size: 94, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:54:12,973 INFO [optim.py:369] (1/2) 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 07:54:13,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:54:16,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:54:38,883 INFO [train.py:968] (1/2) Epoch 18, batch 13350, giga_loss[loss=0.2502, simple_loss=0.3179, pruned_loss=0.09125, over 26623.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3423, pruned_loss=0.09435, over 5663359.78 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3449, pruned_loss=0.09256, over 5759452.52 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3444, pruned_loss=0.09567, over 5660786.77 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:54:42,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=790064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:54:46,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:55:07,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:55:11,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4048, 1.7525, 1.3592, 1.3392], device='cuda:1'), covar=tensor([0.2710, 0.2685, 0.3102, 0.2436], device='cuda:1'), in_proj_covar=tensor([0.1446, 0.1047, 0.1287, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 07:55:32,252 INFO [train.py:968] (1/2) Epoch 18, batch 13400, giga_loss[loss=0.227, simple_loss=0.305, pruned_loss=0.07451, over 28003.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3388, pruned_loss=0.09279, over 5656947.31 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3451, pruned_loss=0.09275, over 5760703.92 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3403, pruned_loss=0.09365, over 5652232.54 frames. ], batch size: 412, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:55:56,933 INFO [optim.py:369] (1/2) 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 07:56:22,250 INFO [train.py:968] (1/2) Epoch 18, batch 13450, libri_loss[loss=0.3159, simple_loss=0.3857, pruned_loss=0.123, over 27710.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3382, pruned_loss=0.09317, over 5650150.47 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3451, pruned_loss=0.09307, over 5764096.40 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3391, pruned_loss=0.09356, over 5639110.01 frames. ], batch size: 116, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:56:41,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3017, 2.9880, 1.4406, 1.4063], device='cuda:1'), covar=tensor([0.0956, 0.0345, 0.0966, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0546, 0.0373, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 07:57:05,449 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:57:08,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:57:10,247 INFO [train.py:968] (1/2) Epoch 18, batch 13500, giga_loss[loss=0.269, simple_loss=0.3466, pruned_loss=0.09564, over 28393.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3384, pruned_loss=0.09403, over 5650455.36 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3445, pruned_loss=0.09283, over 5764952.92 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3395, pruned_loss=0.09456, over 5638092.92 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:57:41,012 INFO [optim.py:369] (1/2) 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 07:57:43,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 07:58:06,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 07:58:08,528 INFO [train.py:968] (1/2) Epoch 18, batch 13550, giga_loss[loss=0.2714, simple_loss=0.3594, pruned_loss=0.09167, over 28728.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3397, pruned_loss=0.09414, over 5645736.19 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3438, pruned_loss=0.09249, over 5766085.68 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3411, pruned_loss=0.0949, over 5632675.23 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 07:58:44,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3505, 3.0074, 1.4291, 1.4360], device='cuda:1'), covar=tensor([0.0928, 0.0308, 0.0930, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0543, 0.0372, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 07:59:02,856 INFO [train.py:968] (1/2) Epoch 18, batch 13600, giga_loss[loss=0.2645, simple_loss=0.3436, pruned_loss=0.0927, over 28938.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.09378, over 5649946.99 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.343, pruned_loss=0.09226, over 5768348.59 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3428, pruned_loss=0.09465, over 5633976.96 frames. ], batch size: 199, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 07:59:29,168 INFO [optim.py:369] (1/2) 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 07:59:42,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4582, 1.8635, 1.6189, 1.6100], device='cuda:1'), covar=tensor([0.1821, 0.2169, 0.2034, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0729, 0.0690, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 07:59:57,901 INFO [train.py:968] (1/2) Epoch 18, batch 13650, giga_loss[loss=0.2697, simple_loss=0.3464, pruned_loss=0.09645, over 28356.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3427, pruned_loss=0.09504, over 5651483.82 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3433, pruned_loss=0.09264, over 5773134.23 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3439, pruned_loss=0.09546, over 5630668.45 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:00:25,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-09 08:00:58,309 INFO [train.py:968] (1/2) Epoch 18, batch 13700, giga_loss[loss=0.2472, simple_loss=0.3282, pruned_loss=0.0831, over 28504.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3405, pruned_loss=0.0934, over 5662604.32 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3432, pruned_loss=0.09266, over 5776516.06 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3414, pruned_loss=0.09376, over 5640441.00 frames. ], batch size: 336, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:01:30,023 INFO [optim.py:369] (1/2) 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 08:01:58,916 INFO [train.py:968] (1/2) Epoch 18, batch 13750, giga_loss[loss=0.2387, simple_loss=0.3242, pruned_loss=0.07662, over 27517.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3387, pruned_loss=0.09153, over 5657231.05 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.343, pruned_loss=0.09256, over 5778276.27 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3397, pruned_loss=0.0919, over 5636349.97 frames. ], batch size: 472, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:01:59,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=790463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:02:36,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-09 08:02:53,685 INFO [train.py:968] (1/2) Epoch 18, batch 13800, giga_loss[loss=0.293, simple_loss=0.3606, pruned_loss=0.1127, over 28992.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3377, pruned_loss=0.09038, over 5648667.44 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3428, pruned_loss=0.09268, over 5771663.32 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3384, pruned_loss=0.0905, over 5634429.82 frames. ], batch size: 186, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:03:16,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4777, 1.7853, 1.3756, 1.6188], device='cuda:1'), covar=tensor([0.0751, 0.0289, 0.0354, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 08:03:22,544 INFO [optim.py:369] (1/2) 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 08:03:52,023 INFO [train.py:968] (1/2) Epoch 18, batch 13850, giga_loss[loss=0.2428, simple_loss=0.3266, pruned_loss=0.07953, over 28797.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3353, pruned_loss=0.09006, over 5645202.38 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3428, pruned_loss=0.09273, over 5758458.87 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3357, pruned_loss=0.09005, over 5642390.59 frames. ], batch size: 243, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:03:58,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-09 08:04:12,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:04:24,086 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 08:04:44,535 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:04:48,783 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:04:53,243 INFO [train.py:968] (1/2) Epoch 18, batch 13900, giga_loss[loss=0.252, simple_loss=0.3367, pruned_loss=0.08366, over 28400.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3353, pruned_loss=0.0904, over 5649979.64 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3427, pruned_loss=0.0927, over 5758066.89 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3356, pruned_loss=0.09037, over 5646200.44 frames. ], batch size: 368, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:05:02,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8836, 1.2762, 1.2882, 1.0209], device='cuda:1'), covar=tensor([0.1575, 0.1052, 0.1887, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0449, 0.0725, 0.0685, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 08:05:03,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 08:05:18,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=790634.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:05:20,780 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:05:47,782 INFO [train.py:968] (1/2) Epoch 18, batch 13950, giga_loss[loss=0.2471, simple_loss=0.3333, pruned_loss=0.08039, over 28937.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3347, pruned_loss=0.08975, over 5664965.56 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3419, pruned_loss=0.0924, over 5762014.31 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3355, pruned_loss=0.08995, over 5656059.89 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:06:45,875 INFO [train.py:968] (1/2) Epoch 18, batch 14000, giga_loss[loss=0.274, simple_loss=0.3547, pruned_loss=0.09663, over 28864.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3361, pruned_loss=0.08982, over 5674755.93 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3413, pruned_loss=0.09206, over 5766292.52 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3372, pruned_loss=0.09022, over 5661346.10 frames. ], batch size: 227, lr: 1.79e-03, grad_scale: 8.0
+2023-03-09 08:07:16,099 INFO [optim.py:369] (1/2) 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 08:07:46,669 INFO [train.py:968] (1/2) Epoch 18, batch 14050, giga_loss[loss=0.2565, simple_loss=0.3345, pruned_loss=0.08931, over 28764.00 frames. ], tot_loss[loss=0.257, simple_loss=0.336, pruned_loss=0.08905, over 5675012.07 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3409, pruned_loss=0.09194, over 5767470.44 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.337, pruned_loss=0.08941, over 5661786.30 frames. ], batch size: 263, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:08:07,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=790777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:08:11,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=790780.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 08:08:20,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4018, 1.6687, 1.6554, 1.2366], device='cuda:1'), covar=tensor([0.1821, 0.2710, 0.1518, 0.1761], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0688, 0.0917, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 08:08:50,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=790809.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 08:08:52,951 INFO [train.py:968] (1/2) Epoch 18, batch 14100, giga_loss[loss=0.2212, simple_loss=0.3057, pruned_loss=0.06837, over 29063.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3339, pruned_loss=0.08855, over 5684702.87 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3404, pruned_loss=0.09177, over 5770286.74 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3351, pruned_loss=0.08892, over 5670069.52 frames. ], batch size: 136, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:09:23,575 INFO [optim.py:369] (1/2) 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 08:09:31,885 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 14150, giga_loss[loss=0.2748, simple_loss=0.3395, pruned_loss=0.105, over 26893.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.335, pruned_loss=0.08931, over 5677386.94 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3398, pruned_loss=0.0916, over 5773579.41 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3363, pruned_loss=0.0897, over 5660238.08 frames. ], batch size: 555, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:10:13,195 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 08:10:28,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4059, 1.8848, 1.4007, 1.4833], device='cuda:1'), covar=tensor([0.0784, 0.0291, 0.0338, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 08:10:35,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4625, 1.7524, 1.3895, 1.5296], device='cuda:1'), covar=tensor([0.2599, 0.2463, 0.2850, 0.2380], device='cuda:1'), in_proj_covar=tensor([0.1445, 0.1049, 0.1286, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 08:10:51,639 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=790906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:10:57,775 INFO [train.py:968] (1/2) Epoch 18, batch 14200, giga_loss[loss=0.2745, simple_loss=0.3663, pruned_loss=0.09135, over 28660.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3387, pruned_loss=0.08913, over 5668950.02 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3392, pruned_loss=0.09128, over 5775434.64 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3403, pruned_loss=0.08971, over 5651761.51 frames. ], batch size: 262, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:11:28,598 INFO [optim.py:369] (1/2) 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 08:11:44,132 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 14250, giga_loss[loss=0.2821, simple_loss=0.3683, pruned_loss=0.09797, over 28926.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.339, pruned_loss=0.08737, over 5661095.44 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.339, pruned_loss=0.0911, over 5776190.14 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3405, pruned_loss=0.08793, over 5644935.58 frames. ], batch size: 213, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:12:38,242 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 14300, giga_loss[loss=0.2934, simple_loss=0.3673, pruned_loss=0.1098, over 28730.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3392, pruned_loss=0.08637, over 5655181.89 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3391, pruned_loss=0.09128, over 5768312.80 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3403, pruned_loss=0.08661, over 5648706.20 frames. ], batch size: 307, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:13:27,195 INFO [optim.py:369] (1/2) 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 08:13:39,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2111, 4.0311, 3.8631, 1.7303], device='cuda:1'), covar=tensor([0.0601, 0.0713, 0.0735, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.1058, 0.0911, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 08:13:54,078 INFO [train.py:968] (1/2) Epoch 18, batch 14350, libri_loss[loss=0.2797, simple_loss=0.354, pruned_loss=0.1027, over 29739.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.08754, over 5662163.60 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3384, pruned_loss=0.09103, over 5770005.04 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08777, over 5650653.51 frames. ], batch size: 87, lr: 1.79e-03, grad_scale: 4.0
+2023-03-09 08:14:30,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:14:37,063 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:14:37,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=791099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:14:51,730 INFO [train.py:968] (1/2) Epoch 18, batch 14400, giga_loss[loss=0.2413, simple_loss=0.3192, pruned_loss=0.08171, over 28989.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3391, pruned_loss=0.08816, over 5662215.75 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3384, pruned_loss=0.09106, over 5763653.43 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3404, pruned_loss=0.08827, over 5656558.27 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:15:11,446 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:15:25,239 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 14450, giga_loss[loss=0.2679, simple_loss=0.3524, pruned_loss=0.09165, over 28863.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3401, pruned_loss=0.08948, over 5664310.32 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3382, pruned_loss=0.09097, over 5764984.63 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3412, pruned_loss=0.08962, over 5657918.75 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:16:21,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9852, 2.0214, 1.4984, 1.5777], device='cuda:1'), covar=tensor([0.0808, 0.0577, 0.0901, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0439, 0.0505, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 08:17:20,137 INFO [train.py:968] (1/2) Epoch 18, batch 14500, giga_loss[loss=0.2092, simple_loss=0.2943, pruned_loss=0.06203, over 28504.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3361, pruned_loss=0.08701, over 5680584.27 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3376, pruned_loss=0.09055, over 5769307.41 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3376, pruned_loss=0.08745, over 5669078.54 frames. ], batch size: 370, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:17:29,837 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:18:20,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3402, 3.5384, 1.5375, 1.4833], device='cuda:1'), covar=tensor([0.0955, 0.0302, 0.0932, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0540, 0.0370, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 08:18:21,158 INFO [train.py:968] (1/2) Epoch 18, batch 14550, giga_loss[loss=0.247, simple_loss=0.3357, pruned_loss=0.07917, over 27531.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3324, pruned_loss=0.08534, over 5676338.76 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3367, pruned_loss=0.09022, over 5774752.22 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3342, pruned_loss=0.08575, over 5656922.49 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:18:44,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=791281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:19:24,577 INFO [train.py:968] (1/2) Epoch 18, batch 14600, giga_loss[loss=0.3106, simple_loss=0.3686, pruned_loss=0.1263, over 28965.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.331, pruned_loss=0.0847, over 5679429.93 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3371, pruned_loss=0.09051, over 5776535.76 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.332, pruned_loss=0.08465, over 5660197.01 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:19:59,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 08:19:59,707 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 14650, giga_loss[loss=0.2375, simple_loss=0.3233, pruned_loss=0.07588, over 28929.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.332, pruned_loss=0.08556, over 5687844.78 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3369, pruned_loss=0.09046, over 5776575.63 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3328, pruned_loss=0.08544, over 5670154.93 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:20:26,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:20:29,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:20:33,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=791369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:21:02,564 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:21:21,994 INFO [train.py:968] (1/2) Epoch 18, batch 14700, giga_loss[loss=0.2258, simple_loss=0.3178, pruned_loss=0.06689, over 28986.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.336, pruned_loss=0.08765, over 5682447.97 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3364, pruned_loss=0.09024, over 5770735.51 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3371, pruned_loss=0.08762, over 5670414.27 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:21:37,350 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:21:40,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:21:53,565 INFO [optim.py:369] (1/2) 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 08:21:53,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6341, 1.6582, 1.2718, 1.2747], device='cuda:1'), covar=tensor([0.0865, 0.0564, 0.0981, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0442, 0.0508, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 08:22:13,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:22:19,626 INFO [train.py:968] (1/2) Epoch 18, batch 14750, giga_loss[loss=0.2563, simple_loss=0.3309, pruned_loss=0.09083, over 28948.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3344, pruned_loss=0.08798, over 5683447.32 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.336, pruned_loss=0.09001, over 5770698.36 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3356, pruned_loss=0.08813, over 5671716.27 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:22:36,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=791474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:23:25,070 INFO [train.py:968] (1/2) Epoch 18, batch 14800, giga_loss[loss=0.2604, simple_loss=0.3343, pruned_loss=0.09324, over 28929.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3353, pruned_loss=0.08979, over 5674596.90 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3358, pruned_loss=0.09005, over 5772495.06 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3365, pruned_loss=0.08985, over 5662598.29 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:23:26,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:23:29,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:23:55,221 INFO [optim.py:369] (1/2) 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 08:24:01,548 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:24:24,195 INFO [train.py:968] (1/2) Epoch 18, batch 14850, giga_loss[loss=0.2782, simple_loss=0.3519, pruned_loss=0.1023, over 28909.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3349, pruned_loss=0.08924, over 5678894.63 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3353, pruned_loss=0.08994, over 5775834.34 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3362, pruned_loss=0.08938, over 5663724.04 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:25:02,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.29 vs. limit=5.0
+2023-03-09 08:25:19,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7235, 2.0378, 1.2654, 1.6097], device='cuda:1'), covar=tensor([0.0878, 0.0500, 0.1001, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0441, 0.0507, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 08:25:32,155 INFO [train.py:968] (1/2) Epoch 18, batch 14900, giga_loss[loss=0.2201, simple_loss=0.2921, pruned_loss=0.07406, over 24193.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3371, pruned_loss=0.08955, over 5675941.38 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3353, pruned_loss=0.08997, over 5777963.68 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3382, pruned_loss=0.08963, over 5661066.02 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:25:41,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3270, 3.0904, 1.4163, 1.4950], device='cuda:1'), covar=tensor([0.0965, 0.0370, 0.0931, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0535, 0.0367, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 08:25:41,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=791617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:25:45,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=791620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:26:14,714 INFO [optim.py:369] (1/2) 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 08:26:26,833 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=791649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:26:43,538 INFO [train.py:968] (1/2) Epoch 18, batch 14950, giga_loss[loss=0.2544, simple_loss=0.3364, pruned_loss=0.08624, over 27708.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3387, pruned_loss=0.09016, over 5673868.26 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3353, pruned_loss=0.09001, over 5775708.34 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3396, pruned_loss=0.09018, over 5661337.81 frames. ], batch size: 474, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:27:57,003 INFO [train.py:968] (1/2) Epoch 18, batch 15000, giga_loss[loss=0.2536, simple_loss=0.3282, pruned_loss=0.08945, over 29030.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3353, pruned_loss=0.08848, over 5682989.34 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3354, pruned_loss=0.09015, over 5773692.39 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.336, pruned_loss=0.08837, over 5672640.95 frames. ], batch size: 200, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:27:57,004 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 08:28:05,681 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 08:28:38,121 INFO [optim.py:369] (1/2) 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 08:29:08,804 INFO [train.py:968] (1/2) Epoch 18, batch 15050, giga_loss[loss=0.2338, simple_loss=0.3132, pruned_loss=0.07717, over 28123.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3298, pruned_loss=0.08656, over 5688864.12 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3346, pruned_loss=0.08984, over 5774441.79 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.331, pruned_loss=0.0867, over 5678138.59 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:30:08,083 INFO [train.py:968] (1/2) Epoch 18, batch 15100, libri_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09378, over 29646.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.326, pruned_loss=0.08472, over 5687213.47 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.0895, over 5775242.29 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3274, pruned_loss=0.08503, over 5675753.98 frames. ], batch size: 91, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:30:41,725 INFO [optim.py:369] (1/2) 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 08:30:53,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=791850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:31:06,843 INFO [train.py:968] (1/2) Epoch 18, batch 15150, giga_loss[loss=0.2701, simple_loss=0.3464, pruned_loss=0.0969, over 28994.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3282, pruned_loss=0.08657, over 5680121.52 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.334, pruned_loss=0.08952, over 5772768.55 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3292, pruned_loss=0.08675, over 5671716.14 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:31:15,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.29 vs. limit=5.0
+2023-03-09 08:31:26,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5468, 1.5168, 1.2618, 1.1622], device='cuda:1'), covar=tensor([0.0646, 0.0330, 0.0767, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0440, 0.0506, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 08:32:04,150 INFO [train.py:968] (1/2) Epoch 18, batch 15200, giga_loss[loss=0.1976, simple_loss=0.2677, pruned_loss=0.06377, over 24382.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3272, pruned_loss=0.08584, over 5669699.50 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3337, pruned_loss=0.08943, over 5774742.96 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3281, pruned_loss=0.08603, over 5660002.48 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:32:35,132 INFO [optim.py:369] (1/2) 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 08:33:01,727 INFO [train.py:968] (1/2) Epoch 18, batch 15250, giga_loss[loss=0.2662, simple_loss=0.3458, pruned_loss=0.09333, over 28681.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3254, pruned_loss=0.08373, over 5675258.72 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3333, pruned_loss=0.08914, over 5778033.27 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3262, pruned_loss=0.08399, over 5661113.57 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:34:03,172 INFO [train.py:968] (1/2) Epoch 18, batch 15300, giga_loss[loss=0.2987, simple_loss=0.3652, pruned_loss=0.1161, over 28496.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3248, pruned_loss=0.08377, over 5665957.33 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3333, pruned_loss=0.08924, over 5778914.35 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3254, pruned_loss=0.0838, over 5652122.11 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:34:43,007 INFO [optim.py:369] (1/2) 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 08:35:06,942 INFO [train.py:968] (1/2) Epoch 18, batch 15350, giga_loss[loss=0.2463, simple_loss=0.3322, pruned_loss=0.08025, over 28667.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3261, pruned_loss=0.08446, over 5679636.46 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3333, pruned_loss=0.08929, over 5781811.98 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3263, pruned_loss=0.08431, over 5663611.86 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:35:40,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4713, 1.4202, 1.5943, 1.1949], device='cuda:1'), covar=tensor([0.2025, 0.3504, 0.1658, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0689, 0.0920, 0.0820], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 08:36:11,549 INFO [train.py:968] (1/2) Epoch 18, batch 15400, libri_loss[loss=0.2761, simple_loss=0.3493, pruned_loss=0.1014, over 29213.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3273, pruned_loss=0.08471, over 5694278.25 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3332, pruned_loss=0.08935, over 5784713.46 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3274, pruned_loss=0.08441, over 5676767.92 frames. ], batch size: 97, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:36:33,850 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 08:36:43,454 INFO [optim.py:369] (1/2) 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 08:37:09,148 INFO [train.py:968] (1/2) Epoch 18, batch 15450, libri_loss[loss=0.252, simple_loss=0.3323, pruned_loss=0.08587, over 25902.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3287, pruned_loss=0.08613, over 5694625.61 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.333, pruned_loss=0.08938, over 5781310.04 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3288, pruned_loss=0.08574, over 5681472.86 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:37:22,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3204, 3.0730, 1.4930, 1.4821], device='cuda:1'), covar=tensor([0.1008, 0.0572, 0.0935, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0538, 0.0371, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 08:38:09,833 INFO [train.py:968] (1/2) Epoch 18, batch 15500, giga_loss[loss=0.244, simple_loss=0.3117, pruned_loss=0.08817, over 26992.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3287, pruned_loss=0.08662, over 5692879.02 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.333, pruned_loss=0.08956, over 5783837.47 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3287, pruned_loss=0.08609, over 5678567.52 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:38:18,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3414, 1.7228, 1.6518, 1.6131], device='cuda:1'), covar=tensor([0.1915, 0.2012, 0.1997, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0720, 0.0680, 0.0655], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 08:38:29,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=792225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:38:41,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3924, 1.9203, 1.4185, 0.7823], device='cuda:1'), covar=tensor([0.6290, 0.3030, 0.3880, 0.5902], device='cuda:1'), in_proj_covar=tensor([0.1684, 0.1591, 0.1567, 0.1381], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 08:38:45,568 INFO [optim.py:369] (1/2) 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 08:39:11,079 INFO [train.py:968] (1/2) Epoch 18, batch 15550, giga_loss[loss=0.246, simple_loss=0.3367, pruned_loss=0.07762, over 28903.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3285, pruned_loss=0.08549, over 5680121.05 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3331, pruned_loss=0.08958, over 5784465.28 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3284, pruned_loss=0.08504, over 5668029.36 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 08:39:27,627 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=792276.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 08:40:09,137 INFO [train.py:968] (1/2) Epoch 18, batch 15600, giga_loss[loss=0.2556, simple_loss=0.3263, pruned_loss=0.0924, over 26967.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3306, pruned_loss=0.08557, over 5671407.69 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3329, pruned_loss=0.08943, over 5784211.05 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3307, pruned_loss=0.08524, over 5658923.46 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:40:29,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-09 08:40:45,378 INFO [optim.py:369] (1/2) 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 08:41:10,444 INFO [train.py:968] (1/2) Epoch 18, batch 15650, giga_loss[loss=0.2394, simple_loss=0.323, pruned_loss=0.07794, over 28693.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3324, pruned_loss=0.08593, over 5667222.64 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3328, pruned_loss=0.08936, over 5783477.76 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3324, pruned_loss=0.08568, over 5656659.74 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:41:18,163 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=792368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:41:21,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=792371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:41:56,127 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 15700, giga_loss[loss=0.221, simple_loss=0.2933, pruned_loss=0.07435, over 24500.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3325, pruned_loss=0.08626, over 5659388.38 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3324, pruned_loss=0.08911, over 5784347.89 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3329, pruned_loss=0.08626, over 5648496.46 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:42:27,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3662, 3.7389, 1.5495, 1.5180], device='cuda:1'), covar=tensor([0.0970, 0.0330, 0.0950, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0538, 0.0371, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 08:42:44,416 INFO [optim.py:369] (1/2) 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 08:43:07,856 INFO [train.py:968] (1/2) Epoch 18, batch 15750, giga_loss[loss=0.2172, simple_loss=0.3051, pruned_loss=0.06464, over 28174.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3315, pruned_loss=0.0862, over 5660249.06 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3323, pruned_loss=0.0891, over 5785694.65 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.332, pruned_loss=0.08619, over 5649326.99 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:44:07,577 INFO [train.py:968] (1/2) Epoch 18, batch 15800, giga_loss[loss=0.2643, simple_loss=0.3418, pruned_loss=0.09343, over 29003.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3301, pruned_loss=0.08573, over 5661812.04 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3319, pruned_loss=0.08887, over 5787367.48 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3308, pruned_loss=0.08584, over 5648907.15 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:44:39,532 INFO [optim.py:369] (1/2) 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 08:45:04,054 INFO [train.py:968] (1/2) Epoch 18, batch 15850, giga_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08905, over 28305.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3285, pruned_loss=0.08539, over 5665904.90 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3324, pruned_loss=0.08926, over 5781271.23 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3286, pruned_loss=0.08505, over 5657432.42 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:45:32,784 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=792589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:45:33,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-09 08:45:50,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4520, 1.5656, 1.4193, 1.6556], device='cuda:1'), covar=tensor([0.0782, 0.0323, 0.0335, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0115, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0066, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 08:46:02,454 INFO [train.py:968] (1/2) Epoch 18, batch 15900, giga_loss[loss=0.2849, simple_loss=0.3485, pruned_loss=0.1106, over 26854.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.328, pruned_loss=0.08456, over 5661686.29 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3318, pruned_loss=0.08902, over 5773186.72 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3285, pruned_loss=0.08442, over 5659946.98 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:46:39,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3438, 1.2854, 1.2315, 1.5180], device='cuda:1'), covar=tensor([0.0753, 0.0409, 0.0355, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0115, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0066, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 08:46:40,518 INFO [optim.py:369] (1/2) 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 08:46:47,957 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 15950, giga_loss[loss=0.2184, simple_loss=0.2893, pruned_loss=0.07381, over 28412.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3305, pruned_loss=0.08585, over 5661237.86 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3315, pruned_loss=0.08899, over 5766432.57 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.331, pruned_loss=0.08565, over 5662249.04 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 08:48:02,093 INFO [train.py:968] (1/2) Epoch 18, batch 16000, libri_loss[loss=0.2345, simple_loss=0.3178, pruned_loss=0.07563, over 29530.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3305, pruned_loss=0.08663, over 5657795.10 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.331, pruned_loss=0.08871, over 5770735.46 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3314, pruned_loss=0.08662, over 5651201.27 frames. ], batch size: 82, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:48:35,130 INFO [optim.py:369] (1/2) 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 08:48:56,668 INFO [train.py:968] (1/2) Epoch 18, batch 16050, giga_loss[loss=0.2594, simple_loss=0.343, pruned_loss=0.08787, over 28425.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3339, pruned_loss=0.0887, over 5662955.52 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3308, pruned_loss=0.08872, over 5772840.44 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3347, pruned_loss=0.08867, over 5653190.58 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:49:31,788 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=792794.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 08:49:34,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=792797.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 08:49:47,708 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=792807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:49:52,577 INFO [train.py:968] (1/2) Epoch 18, batch 16100, giga_loss[loss=0.2731, simple_loss=0.354, pruned_loss=0.09604, over 28911.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3366, pruned_loss=0.08959, over 5656810.00 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3311, pruned_loss=0.08876, over 5775829.36 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3372, pruned_loss=0.08955, over 5643832.29 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:50:09,743 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:50:50,015 INFO [train.py:968] (1/2) Epoch 18, batch 16150, giga_loss[loss=0.2581, simple_loss=0.3412, pruned_loss=0.08751, over 28765.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3375, pruned_loss=0.08956, over 5656691.12 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3311, pruned_loss=0.08864, over 5777765.63 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3381, pruned_loss=0.08966, over 5641773.15 frames. ], batch size: 243, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:51:12,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9805, 3.7837, 3.5902, 1.7035], device='cuda:1'), covar=tensor([0.0796, 0.0997, 0.1060, 0.2329], device='cuda:1'), in_proj_covar=tensor([0.1143, 0.1052, 0.0908, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 08:51:56,931 INFO [train.py:968] (1/2) Epoch 18, batch 16200, giga_loss[loss=0.2378, simple_loss=0.3217, pruned_loss=0.07691, over 29098.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3359, pruned_loss=0.08849, over 5652773.32 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3309, pruned_loss=0.08849, over 5770088.50 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3366, pruned_loss=0.0887, over 5645765.64 frames. ], batch size: 285, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:52:35,341 INFO [optim.py:369] (1/2) 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 08:53:00,604 INFO [train.py:968] (1/2) Epoch 18, batch 16250, libri_loss[loss=0.2092, simple_loss=0.2891, pruned_loss=0.06464, over 27793.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3338, pruned_loss=0.08758, over 5652858.20 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3312, pruned_loss=0.08867, over 5763664.61 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3342, pruned_loss=0.08758, over 5650528.82 frames. ], batch size: 61, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:53:03,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=792964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:53:12,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4998, 1.7110, 1.2647, 1.2678], device='cuda:1'), covar=tensor([0.0952, 0.0482, 0.1046, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0440, 0.0508, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 08:53:56,281 INFO [train.py:968] (1/2) Epoch 18, batch 16300, giga_loss[loss=0.2724, simple_loss=0.3487, pruned_loss=0.09804, over 29078.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3329, pruned_loss=0.08697, over 5672188.15 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3305, pruned_loss=0.08826, over 5768665.94 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.334, pruned_loss=0.0873, over 5661141.05 frames. ], batch size: 285, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:54:30,536 INFO [optim.py:369] (1/2) 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 08:54:32,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3766, 1.6941, 1.4085, 1.6012], device='cuda:1'), covar=tensor([0.0751, 0.0335, 0.0342, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0115, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0066, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 08:54:52,595 INFO [train.py:968] (1/2) Epoch 18, batch 16350, giga_loss[loss=0.2499, simple_loss=0.3282, pruned_loss=0.08576, over 28984.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3331, pruned_loss=0.0882, over 5671425.41 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3305, pruned_loss=0.08813, over 5772831.40 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3341, pruned_loss=0.08858, over 5655564.93 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:55:47,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:55:52,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:55:53,364 INFO [train.py:968] (1/2) Epoch 18, batch 16400, giga_loss[loss=0.2967, simple_loss=0.3556, pruned_loss=0.1189, over 26972.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3301, pruned_loss=0.08722, over 5664416.46 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3305, pruned_loss=0.0882, over 5775403.84 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3309, pruned_loss=0.08744, over 5647498.85 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 08:56:23,805 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=793139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:56:26,358 INFO [optim.py:369] (1/2) 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 08:56:50,800 INFO [train.py:968] (1/2) Epoch 18, batch 16450, giga_loss[loss=0.2603, simple_loss=0.3521, pruned_loss=0.08431, over 28647.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3301, pruned_loss=0.0867, over 5670423.08 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3305, pruned_loss=0.08845, over 5777337.41 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3306, pruned_loss=0.08661, over 5652354.63 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:57:11,219 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:57:13,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:57:24,593 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 08:57:26,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7231, 1.9106, 1.2658, 1.3662], device='cuda:1'), covar=tensor([0.0919, 0.0563, 0.1065, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0438, 0.0507, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 08:57:48,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6836, 1.1126, 2.8959, 2.6493], device='cuda:1'), covar=tensor([0.1709, 0.2487, 0.0503, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0625, 0.0912, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 08:57:49,116 INFO [train.py:968] (1/2) Epoch 18, batch 16500, giga_loss[loss=0.3091, simple_loss=0.3989, pruned_loss=0.1097, over 29051.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3296, pruned_loss=0.08507, over 5678273.40 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3302, pruned_loss=0.08833, over 5779218.17 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3303, pruned_loss=0.08507, over 5661116.94 frames. ], batch size: 128, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:57:54,087 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 08:58:09,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3250, 4.1522, 3.9113, 1.8638], device='cuda:1'), covar=tensor([0.0482, 0.0641, 0.0718, 0.1922], device='cuda:1'), in_proj_covar=tensor([0.1140, 0.1050, 0.0907, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 08:58:17,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 08:58:19,788 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:58:24,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7336, 2.3550, 1.6579, 0.9788], device='cuda:1'), covar=tensor([0.4666, 0.2467, 0.3401, 0.4432], device='cuda:1'), in_proj_covar=tensor([0.1700, 0.1600, 0.1571, 0.1387], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 08:58:40,844 INFO [train.py:968] (1/2) Epoch 18, batch 16550, giga_loss[loss=0.2788, simple_loss=0.3566, pruned_loss=0.1004, over 28902.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.331, pruned_loss=0.08369, over 5692278.18 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3301, pruned_loss=0.08811, over 5781235.05 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3317, pruned_loss=0.08375, over 5673517.70 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:59:36,065 INFO [train.py:968] (1/2) Epoch 18, batch 16600, giga_loss[loss=0.2186, simple_loss=0.3114, pruned_loss=0.06293, over 28518.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3322, pruned_loss=0.08382, over 5688123.43 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3297, pruned_loss=0.0879, over 5782884.71 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3331, pruned_loss=0.08397, over 5670286.61 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 08:59:45,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4873, 1.6847, 1.6697, 1.4403], device='cuda:1'), covar=tensor([0.2616, 0.2109, 0.1765, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1787, 0.1703, 0.1847], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 08:59:52,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 08:59:56,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:00:15,779 INFO [optim.py:369] (1/2) 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 09:00:31,850 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 16650, giga_loss[loss=0.2667, simple_loss=0.3387, pruned_loss=0.09732, over 27586.00 frames. ], tot_loss[loss=0.25, simple_loss=0.332, pruned_loss=0.08401, over 5679298.64 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3297, pruned_loss=0.08802, over 5785605.49 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3328, pruned_loss=0.08392, over 5660168.65 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:01:18,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4850, 1.7268, 1.7478, 1.2783], device='cuda:1'), covar=tensor([0.1961, 0.2860, 0.1682, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.0870, 0.0684, 0.0916, 0.0817], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 09:01:38,226 INFO [train.py:968] (1/2) Epoch 18, batch 16700, giga_loss[loss=0.234, simple_loss=0.3251, pruned_loss=0.07148, over 28365.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3323, pruned_loss=0.0843, over 5665659.28 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3301, pruned_loss=0.08823, over 5777205.10 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3327, pruned_loss=0.08397, over 5655829.11 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:01:51,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5224, 2.2747, 1.6193, 0.7097], device='cuda:1'), covar=tensor([0.5566, 0.2633, 0.4374, 0.5930], device='cuda:1'), in_proj_covar=tensor([0.1688, 0.1587, 0.1564, 0.1378], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 09:02:23,610 INFO [optim.py:369] (1/2) 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 09:02:48,209 INFO [train.py:968] (1/2) Epoch 18, batch 16750, giga_loss[loss=0.2183, simple_loss=0.2875, pruned_loss=0.07455, over 24624.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3315, pruned_loss=0.08352, over 5662658.76 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.33, pruned_loss=0.08813, over 5778525.49 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3319, pruned_loss=0.08325, over 5650870.68 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:02:48,410 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:02:56,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3101, 1.6044, 1.2525, 1.5716], device='cuda:1'), covar=tensor([0.2796, 0.2721, 0.3174, 0.2105], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1038, 0.1276, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:03:33,413 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:03:52,490 INFO [train.py:968] (1/2) Epoch 18, batch 16800, giga_loss[loss=0.2531, simple_loss=0.3435, pruned_loss=0.08136, over 28965.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3326, pruned_loss=0.08355, over 5665880.75 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3303, pruned_loss=0.08817, over 5781063.33 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3327, pruned_loss=0.08319, over 5651585.74 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:04:24,921 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 09:04:36,322 INFO [optim.py:369] (1/2) 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 09:04:49,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:04:56,960 INFO [train.py:968] (1/2) Epoch 18, batch 16850, giga_loss[loss=0.2943, simple_loss=0.3747, pruned_loss=0.107, over 28599.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.335, pruned_loss=0.08499, over 5669164.17 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3296, pruned_loss=0.08768, over 5781557.33 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3358, pruned_loss=0.08499, over 5652122.39 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:05:19,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-09 09:05:55,126 INFO [train.py:968] (1/2) Epoch 18, batch 16900, giga_loss[loss=0.2384, simple_loss=0.3204, pruned_loss=0.07822, over 28664.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3369, pruned_loss=0.08579, over 5676918.08 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3296, pruned_loss=0.08759, over 5786078.90 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3379, pruned_loss=0.08581, over 5655218.55 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:06:01,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:06:35,979 INFO [optim.py:369] (1/2) 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 09:06:54,166 INFO [train.py:968] (1/2) Epoch 18, batch 16950, giga_loss[loss=0.2804, simple_loss=0.3571, pruned_loss=0.1019, over 28768.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3368, pruned_loss=0.08588, over 5692280.10 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3294, pruned_loss=0.08723, over 5787971.45 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3381, pruned_loss=0.08616, over 5669347.47 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:07:47,758 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:07:50,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:08:03,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5422, 1.7518, 1.4237, 1.5697], device='cuda:1'), covar=tensor([0.3019, 0.2796, 0.3474, 0.2493], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1040, 0.1275, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:08:06,202 INFO [train.py:968] (1/2) Epoch 18, batch 17000, giga_loss[loss=0.2244, simple_loss=0.3139, pruned_loss=0.06746, over 28227.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3353, pruned_loss=0.0858, over 5686137.88 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3292, pruned_loss=0.08709, over 5789155.89 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3365, pruned_loss=0.08614, over 5666170.45 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:08:29,719 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:09:10,008 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:12,739 INFO [train.py:968] (1/2) Epoch 18, batch 17050, giga_loss[loss=0.267, simple_loss=0.3528, pruned_loss=0.09055, over 29022.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3344, pruned_loss=0.08472, over 5686125.37 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3293, pruned_loss=0.08712, over 5791526.80 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3354, pruned_loss=0.08491, over 5665874.36 frames. ], batch size: 285, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:09:14,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:44,202 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=793788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:48,920 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=793791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:09:58,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5518, 1.6985, 1.4336, 1.7303], device='cuda:1'), covar=tensor([0.2731, 0.2675, 0.3071, 0.2329], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1037, 0.1273, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:10:11,702 INFO [train.py:968] (1/2) Epoch 18, batch 17100, giga_loss[loss=0.3285, simple_loss=0.3928, pruned_loss=0.1321, over 26972.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3344, pruned_loss=0.08503, over 5674883.93 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3298, pruned_loss=0.08746, over 5782736.96 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3349, pruned_loss=0.08482, over 5664108.94 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:10:15,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1997, 1.5609, 0.8973, 1.1281], device='cuda:1'), covar=tensor([0.1186, 0.0690, 0.1643, 0.1519], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0437, 0.0507, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 09:10:39,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:10:50,570 INFO [optim.py:369] (1/2) 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 09:11:00,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-09 09:11:07,910 INFO [train.py:968] (1/2) Epoch 18, batch 17150, giga_loss[loss=0.2709, simple_loss=0.356, pruned_loss=0.09286, over 28645.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3354, pruned_loss=0.08559, over 5676634.71 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3294, pruned_loss=0.08723, over 5785508.49 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3362, pruned_loss=0.08556, over 5663196.28 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 09:11:19,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=793870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:11:31,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9649, 1.2478, 2.7181, 2.6544], device='cuda:1'), covar=tensor([0.1425, 0.2361, 0.0517, 0.1206], device='cuda:1'), in_proj_covar=tensor([0.0721, 0.0624, 0.0907, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0010], device='cuda:1')
+2023-03-09 09:11:54,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4107, 3.4470, 1.5260, 1.5261], device='cuda:1'), covar=tensor([0.0994, 0.0300, 0.0950, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0537, 0.0371, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 09:12:03,564 INFO [train.py:968] (1/2) Epoch 18, batch 17200, giga_loss[loss=0.2843, simple_loss=0.3557, pruned_loss=0.1065, over 28965.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3371, pruned_loss=0.08676, over 5679211.05 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3294, pruned_loss=0.0874, over 5788701.02 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3379, pruned_loss=0.08658, over 5663086.58 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:12:35,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5703, 4.4014, 4.1650, 1.9316], device='cuda:1'), covar=tensor([0.0626, 0.0747, 0.0837, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.1144, 0.1046, 0.0905, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 09:12:41,576 INFO [optim.py:369] (1/2) 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 09:12:59,294 INFO [train.py:968] (1/2) Epoch 18, batch 17250, giga_loss[loss=0.2387, simple_loss=0.3222, pruned_loss=0.07765, over 28486.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3342, pruned_loss=0.08623, over 5680719.89 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3292, pruned_loss=0.08726, over 5790476.07 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3352, pruned_loss=0.08621, over 5664876.19 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:13:17,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7806, 3.6375, 3.4252, 1.6717], device='cuda:1'), covar=tensor([0.0713, 0.0773, 0.0787, 0.2265], device='cuda:1'), in_proj_covar=tensor([0.1142, 0.1044, 0.0902, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 09:13:21,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=793980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:13:24,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=793983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:13:57,211 INFO [train.py:968] (1/2) Epoch 18, batch 17300, giga_loss[loss=0.2366, simple_loss=0.3264, pruned_loss=0.07347, over 28738.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3327, pruned_loss=0.08644, over 5671672.61 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3291, pruned_loss=0.0872, over 5792250.97 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3337, pruned_loss=0.08647, over 5655293.63 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:13:58,882 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:13:59,939 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:14:02,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:14:04,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3526, 1.6514, 1.3613, 0.9520], device='cuda:1'), covar=tensor([0.2218, 0.2093, 0.2348, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.1433, 0.1037, 0.1276, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:14:29,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3396, 1.6412, 1.3177, 1.0324], device='cuda:1'), covar=tensor([0.2550, 0.2438, 0.2824, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1430, 0.1035, 0.1274, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:14:31,877 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:14:47,953 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:14:53,847 INFO [train.py:968] (1/2) Epoch 18, batch 17350, giga_loss[loss=0.2408, simple_loss=0.3223, pruned_loss=0.07968, over 28290.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3332, pruned_loss=0.08767, over 5667476.69 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3292, pruned_loss=0.08736, over 5795178.38 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.334, pruned_loss=0.08756, over 5648569.72 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:15:43,129 INFO [train.py:968] (1/2) Epoch 18, batch 17400, giga_loss[loss=0.39, simple_loss=0.4345, pruned_loss=0.1727, over 27598.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3402, pruned_loss=0.09145, over 5672077.41 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3289, pruned_loss=0.08725, over 5793660.94 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3413, pruned_loss=0.09153, over 5654547.06 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:16:01,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 09:16:12,989 INFO [optim.py:369] (1/2) 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 09:16:15,210 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:16:28,215 INFO [train.py:968] (1/2) Epoch 18, batch 17450, giga_loss[loss=0.2942, simple_loss=0.3867, pruned_loss=0.1008, over 28948.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3492, pruned_loss=0.09642, over 5677497.48 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3289, pruned_loss=0.08719, over 5794892.15 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3503, pruned_loss=0.09663, over 5661560.15 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:16:29,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:17:12,839 INFO [train.py:968] (1/2) Epoch 18, batch 17500, giga_loss[loss=0.3012, simple_loss=0.3821, pruned_loss=0.1101, over 28979.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3504, pruned_loss=0.09813, over 5675723.59 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.329, pruned_loss=0.08719, over 5794896.68 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3518, pruned_loss=0.09858, over 5659821.36 frames. ], batch size: 213, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:17:16,856 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 09:17:43,728 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 17550, giga_loss[loss=0.2179, simple_loss=0.2955, pruned_loss=0.07014, over 28953.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.0964, over 5682938.16 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3291, pruned_loss=0.08729, over 5797305.74 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.347, pruned_loss=0.09682, over 5666055.07 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:18:37,143 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:18:38,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:18:40,387 INFO [train.py:968] (1/2) Epoch 18, batch 17600, giga_loss[loss=0.2803, simple_loss=0.3444, pruned_loss=0.1081, over 28943.00 frames. ], tot_loss[loss=0.262, simple_loss=0.338, pruned_loss=0.09301, over 5693991.34 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3289, pruned_loss=0.0872, over 5798887.45 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3393, pruned_loss=0.0935, over 5677735.99 frames. ], batch size: 213, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:18:45,429 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:19:02,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:19:21,785 INFO [train.py:968] (1/2) Epoch 18, batch 17650, giga_loss[loss=0.2079, simple_loss=0.285, pruned_loss=0.06537, over 28859.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3308, pruned_loss=0.08963, over 5693308.07 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3296, pruned_loss=0.08751, over 5791364.48 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3313, pruned_loss=0.08983, over 5683946.28 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:19:31,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5930, 2.0579, 1.3538, 0.8995], device='cuda:1'), covar=tensor([0.6270, 0.3046, 0.3025, 0.5851], device='cuda:1'), in_proj_covar=tensor([0.1684, 0.1592, 0.1559, 0.1376], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 09:20:04,010 INFO [train.py:968] (1/2) Epoch 18, batch 17700, giga_loss[loss=0.2159, simple_loss=0.2876, pruned_loss=0.07215, over 27616.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3233, pruned_loss=0.08649, over 5694461.90 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3292, pruned_loss=0.08724, over 5793633.89 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.324, pruned_loss=0.08694, over 5682872.79 frames. ], batch size: 472, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:20:22,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:20:33,558 INFO [optim.py:369] (1/2) 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 09:20:44,955 INFO [train.py:968] (1/2) Epoch 18, batch 17750, giga_loss[loss=0.2544, simple_loss=0.3197, pruned_loss=0.0946, over 28711.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3187, pruned_loss=0.08459, over 5692454.26 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3301, pruned_loss=0.08778, over 5794719.13 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3183, pruned_loss=0.08436, over 5680362.69 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:20:56,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0338, 2.1037, 2.2158, 1.8005], device='cuda:1'), covar=tensor([0.1810, 0.2273, 0.1400, 0.1637], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0689, 0.0925, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 09:21:26,648 INFO [train.py:968] (1/2) Epoch 18, batch 17800, giga_loss[loss=0.2343, simple_loss=0.3063, pruned_loss=0.08114, over 28858.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3143, pruned_loss=0.08274, over 5699297.68 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3301, pruned_loss=0.08772, over 5796359.80 frames. ], giga_tot_loss[loss=0.2394, simple_loss=0.3138, pruned_loss=0.08254, over 5687064.66 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:21:34,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:21:42,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3648, 1.5479, 1.5679, 1.3844], device='cuda:1'), covar=tensor([0.1635, 0.1830, 0.2137, 0.1862], device='cuda:1'), in_proj_covar=tensor([0.0441, 0.0729, 0.0687, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 09:21:54,979 INFO [optim.py:369] (1/2) 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 09:22:07,322 INFO [train.py:968] (1/2) Epoch 18, batch 17850, giga_loss[loss=0.2208, simple_loss=0.2791, pruned_loss=0.08126, over 23931.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3122, pruned_loss=0.08138, over 5690579.90 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3304, pruned_loss=0.08769, over 5786682.73 frames. ], giga_tot_loss[loss=0.2365, simple_loss=0.311, pruned_loss=0.08107, over 5685312.06 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:22:12,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:20,369 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:22,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:33,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6575, 1.9409, 1.5160, 1.7627], device='cuda:1'), covar=tensor([0.2437, 0.2504, 0.2861, 0.2267], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1042, 0.1275, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:22:41,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-09 09:22:46,539 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:22:50,177 INFO [train.py:968] (1/2) Epoch 18, batch 17900, libri_loss[loss=0.2207, simple_loss=0.3085, pruned_loss=0.06642, over 29565.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3088, pruned_loss=0.08013, over 5686548.12 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3299, pruned_loss=0.08735, over 5788367.03 frames. ], giga_tot_loss[loss=0.2341, simple_loss=0.3079, pruned_loss=0.0801, over 5679545.16 frames. ], batch size: 76, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:23:16,426 INFO [optim.py:369] (1/2) 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 09:23:18,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 09:23:27,605 INFO [train.py:968] (1/2) Epoch 18, batch 17950, giga_loss[loss=0.2274, simple_loss=0.301, pruned_loss=0.07691, over 29138.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3065, pruned_loss=0.07858, over 5703915.61 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3306, pruned_loss=0.08756, over 5793251.65 frames. ], giga_tot_loss[loss=0.2302, simple_loss=0.3043, pruned_loss=0.07803, over 5690053.80 frames. ], batch size: 128, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:23:31,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:33,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:44,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:53,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=794692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:23:57,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:24:07,863 INFO [train.py:968] (1/2) Epoch 18, batch 18000, giga_loss[loss=0.2197, simple_loss=0.2942, pruned_loss=0.07263, over 27933.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3049, pruned_loss=0.07782, over 5697978.33 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.331, pruned_loss=0.08767, over 5786835.21 frames. ], giga_tot_loss[loss=0.2281, simple_loss=0.3021, pruned_loss=0.07703, over 5691193.47 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:24:07,863 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 09:24:15,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4795, 1.8282, 1.4350, 1.3799], device='cuda:1'), covar=tensor([0.2982, 0.2862, 0.3174, 0.2450], device='cuda:1'), in_proj_covar=tensor([0.1434, 0.1042, 0.1274, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:24:17,119 INFO [train.py:1012] (1/2) Epoch 18, validation: loss=0.2056, simple_loss=0.3113, pruned_loss=0.04996, over 944034.00 frames. 
+2023-03-09 09:24:17,119 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 09:24:36,862 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:24:44,363 INFO [optim.py:369] (1/2) 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 09:24:56,443 INFO [train.py:968] (1/2) Epoch 18, batch 18050, giga_loss[loss=0.1979, simple_loss=0.2705, pruned_loss=0.06264, over 29005.00 frames. ], tot_loss[loss=0.2282, simple_loss=0.3029, pruned_loss=0.07676, over 5692946.44 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.331, pruned_loss=0.08744, over 5785909.63 frames. ], giga_tot_loss[loss=0.2255, simple_loss=0.2994, pruned_loss=0.07585, over 5683647.17 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:25:12,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6344, 1.8018, 1.4630, 1.7052], device='cuda:1'), covar=tensor([0.0737, 0.0305, 0.0322, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 09:25:16,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-09 09:25:31,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5705, 1.7850, 1.6641, 1.4457], device='cuda:1'), covar=tensor([0.3119, 0.2517, 0.2023, 0.2701], device='cuda:1'), in_proj_covar=tensor([0.1880, 0.1796, 0.1719, 0.1874], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 09:25:36,205 INFO [train.py:968] (1/2) Epoch 18, batch 18100, giga_loss[loss=0.2087, simple_loss=0.2912, pruned_loss=0.06313, over 28882.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3003, pruned_loss=0.07545, over 5697287.15 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.331, pruned_loss=0.08737, over 5789055.21 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2965, pruned_loss=0.07441, over 5684157.21 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:25:58,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=794835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:26:01,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=794838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:26:08,852 INFO [optim.py:369] (1/2) 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 09:26:15,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6100, 1.6460, 1.8084, 1.4563], device='cuda:1'), covar=tensor([0.1485, 0.2080, 0.1235, 0.1466], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0694, 0.0930, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 09:26:20,163 INFO [train.py:968] (1/2) Epoch 18, batch 18150, giga_loss[loss=0.1944, simple_loss=0.2674, pruned_loss=0.06071, over 28677.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2979, pruned_loss=0.07415, over 5708707.52 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3315, pruned_loss=0.08744, over 5790236.42 frames. ], giga_tot_loss[loss=0.2195, simple_loss=0.2933, pruned_loss=0.07286, over 5694350.28 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:26:25,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=794867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:26:45,506 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=794892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:27:02,709 INFO [train.py:968] (1/2) Epoch 18, batch 18200, giga_loss[loss=0.2835, simple_loss=0.347, pruned_loss=0.11, over 28695.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.2978, pruned_loss=0.07493, over 5705997.28 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3317, pruned_loss=0.08758, over 5790596.71 frames. ], giga_tot_loss[loss=0.2202, simple_loss=0.2933, pruned_loss=0.07354, over 5692864.68 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:27:34,320 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:27:52,588 INFO [train.py:968] (1/2) Epoch 18, batch 18250, giga_loss[loss=0.2975, simple_loss=0.3776, pruned_loss=0.1087, over 28614.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.3087, pruned_loss=0.08051, over 5696740.05 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.08766, over 5782320.06 frames. ], giga_tot_loss[loss=0.2314, simple_loss=0.3045, pruned_loss=0.07919, over 5691387.78 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:28:31,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0172, 3.8095, 3.6046, 1.7824], device='cuda:1'), covar=tensor([0.0712, 0.0872, 0.0823, 0.2256], device='cuda:1'), in_proj_covar=tensor([0.1141, 0.1051, 0.0905, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 09:28:35,845 INFO [train.py:968] (1/2) Epoch 18, batch 18300, giga_loss[loss=0.2952, simple_loss=0.3676, pruned_loss=0.1114, over 29033.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3234, pruned_loss=0.08854, over 5692808.54 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3326, pruned_loss=0.08803, over 5779471.17 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3192, pruned_loss=0.0871, over 5689748.28 frames. ], batch size: 128, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:29:05,081 INFO [optim.py:369] (1/2) 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 09:29:14,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:29:18,309 INFO [train.py:968] (1/2) Epoch 18, batch 18350, giga_loss[loss=0.2782, simple_loss=0.3566, pruned_loss=0.09995, over 28832.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3339, pruned_loss=0.09377, over 5696091.52 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3326, pruned_loss=0.08803, over 5779471.17 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3307, pruned_loss=0.09265, over 5693709.68 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:29:33,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5322, 3.0537, 1.6469, 1.6604], device='cuda:1'), covar=tensor([0.0800, 0.0299, 0.0730, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0535, 0.0369, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 09:29:36,087 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:29:38,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:29:57,671 INFO [train.py:968] (1/2) Epoch 18, batch 18400, libri_loss[loss=0.206, simple_loss=0.2885, pruned_loss=0.06172, over 29401.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3392, pruned_loss=0.09526, over 5689216.49 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3334, pruned_loss=0.08847, over 5772931.64 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.336, pruned_loss=0.09414, over 5691116.38 frames. ], batch size: 67, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:29:58,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795113.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:30:01,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:30:19,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 09:30:26,820 INFO [optim.py:369] (1/2) 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 09:30:35,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4458, 1.7470, 1.3961, 1.2920], device='cuda:1'), covar=tensor([0.2566, 0.2681, 0.3028, 0.2342], device='cuda:1'), in_proj_covar=tensor([0.1434, 0.1043, 0.1276, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:30:38,686 INFO [train.py:968] (1/2) Epoch 18, batch 18450, giga_loss[loss=0.2731, simple_loss=0.3499, pruned_loss=0.09813, over 28985.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3414, pruned_loss=0.09519, over 5693896.82 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3331, pruned_loss=0.08834, over 5774924.27 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3392, pruned_loss=0.09455, over 5692268.65 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:30:41,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3275, 2.3983, 1.7628, 1.8602], device='cuda:1'), covar=tensor([0.0837, 0.0668, 0.1017, 0.1174], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0437, 0.0506, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 09:31:05,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3875, 1.5364, 1.4591, 1.2571], device='cuda:1'), covar=tensor([0.2446, 0.2481, 0.1767, 0.2375], device='cuda:1'), in_proj_covar=tensor([0.1890, 0.1806, 0.1732, 0.1887], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 09:31:13,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:31:16,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:31:24,277 INFO [train.py:968] (1/2) Epoch 18, batch 18500, giga_loss[loss=0.243, simple_loss=0.328, pruned_loss=0.07899, over 29045.00 frames. ], tot_loss[loss=0.266, simple_loss=0.342, pruned_loss=0.095, over 5687532.84 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3338, pruned_loss=0.0887, over 5777394.98 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3399, pruned_loss=0.09427, over 5682299.84 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:31:40,855 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:31:55,669 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795256.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:32:04,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:32:07,149 INFO [train.py:968] (1/2) Epoch 18, batch 18550, giga_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08538, over 28354.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3441, pruned_loss=0.09662, over 5693166.73 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.334, pruned_loss=0.08881, over 5778916.79 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3423, pruned_loss=0.09604, over 5686547.81 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:32:11,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:32:28,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:32:50,345 INFO [train.py:968] (1/2) Epoch 18, batch 18600, giga_loss[loss=0.2905, simple_loss=0.3608, pruned_loss=0.1101, over 28667.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3476, pruned_loss=0.09929, over 5697371.50 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3338, pruned_loss=0.08866, over 5782379.81 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3466, pruned_loss=0.09917, over 5687177.73 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:32:51,333 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:32:55,987 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-09 09:33:14,835 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-09 09:33:20,427 INFO [optim.py:369] (1/2) 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 09:33:30,958 INFO [train.py:968] (1/2) Epoch 18, batch 18650, giga_loss[loss=0.3128, simple_loss=0.3833, pruned_loss=0.1211, over 28006.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3505, pruned_loss=0.1006, over 5698347.16 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3342, pruned_loss=0.08875, over 5780366.71 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3498, pruned_loss=0.1007, over 5689478.61 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:33:38,528 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:33:54,669 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795389.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:33:58,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5354, 1.6720, 1.5791, 1.4890], device='cuda:1'), covar=tensor([0.1687, 0.2121, 0.2224, 0.1961], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0735, 0.0691, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 09:34:11,400 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 18700, giga_loss[loss=0.2726, simple_loss=0.353, pruned_loss=0.09613, over 28914.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3526, pruned_loss=0.1006, over 5706850.44 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3344, pruned_loss=0.08886, over 5781161.00 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.352, pruned_loss=0.1007, over 5698425.66 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:34:14,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:34:36,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:34:38,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0326, 2.0714, 1.9198, 1.6555], device='cuda:1'), covar=tensor([0.1964, 0.2138, 0.2034, 0.2161], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0734, 0.0690, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 09:34:40,715 INFO [optim.py:369] (1/2) 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 09:34:51,495 INFO [train.py:968] (1/2) Epoch 18, batch 18750, giga_loss[loss=0.2679, simple_loss=0.3542, pruned_loss=0.09077, over 28903.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3531, pruned_loss=0.0999, over 5711510.05 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3347, pruned_loss=0.08871, over 5785230.26 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3532, pruned_loss=0.1006, over 5697850.63 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:35:01,860 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 09:35:17,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0898, 3.1769, 2.2203, 1.3205], device='cuda:1'), covar=tensor([0.7036, 0.2759, 0.3403, 0.5623], device='cuda:1'), in_proj_covar=tensor([0.1683, 0.1590, 0.1567, 0.1377], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 09:35:30,021 INFO [train.py:968] (1/2) Epoch 18, batch 18800, giga_loss[loss=0.2788, simple_loss=0.3633, pruned_loss=0.09717, over 29083.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3546, pruned_loss=0.1001, over 5707738.19 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.335, pruned_loss=0.08866, over 5785770.29 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3552, pruned_loss=0.1012, over 5693100.09 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:35:30,884 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:35:32,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5857, 1.6378, 1.8498, 1.3818], device='cuda:1'), covar=tensor([0.2019, 0.2600, 0.1596, 0.1959], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0690, 0.0922, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 09:35:44,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:35:59,335 INFO [optim.py:369] (1/2) 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 09:36:10,963 INFO [train.py:968] (1/2) Epoch 18, batch 18850, giga_loss[loss=0.257, simple_loss=0.3403, pruned_loss=0.08686, over 28857.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3537, pruned_loss=0.09847, over 5703855.00 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.335, pruned_loss=0.08866, over 5785770.29 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3542, pruned_loss=0.09935, over 5692461.99 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:36:17,731 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:36:32,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=795588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:36:49,913 INFO [train.py:968] (1/2) Epoch 18, batch 18900, giga_loss[loss=0.2672, simple_loss=0.3504, pruned_loss=0.09201, over 28584.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3527, pruned_loss=0.09718, over 5704927.54 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3357, pruned_loss=0.08895, over 5777788.64 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3528, pruned_loss=0.09775, over 5701456.11 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:37:03,082 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:37:28,483 INFO [train.py:968] (1/2) Epoch 18, batch 18950, giga_loss[loss=0.367, simple_loss=0.4141, pruned_loss=0.1599, over 28213.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3549, pruned_loss=0.09887, over 5701446.65 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3366, pruned_loss=0.08951, over 5779234.85 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3544, pruned_loss=0.09896, over 5695856.66 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:37:43,010 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-09 09:37:48,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5915, 1.6528, 1.2303, 1.1922], device='cuda:1'), covar=tensor([0.0860, 0.0526, 0.0982, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0440, 0.0511, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 09:37:49,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7187, 2.0572, 1.9963, 1.6790], device='cuda:1'), covar=tensor([0.2658, 0.2194, 0.2092, 0.2401], device='cuda:1'), in_proj_covar=tensor([0.1886, 0.1799, 0.1728, 0.1885], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 09:37:51,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 19000, giga_loss[loss=0.3108, simple_loss=0.3657, pruned_loss=0.128, over 28845.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3578, pruned_loss=0.1039, over 5690337.71 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3367, pruned_loss=0.08955, over 5779859.68 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3574, pruned_loss=0.1039, over 5685170.37 frames. ], batch size: 112, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:38:41,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:38:44,765 INFO [optim.py:369] (1/2) 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 09:38:54,412 INFO [train.py:968] (1/2) Epoch 18, batch 19050, libri_loss[loss=0.2754, simple_loss=0.3522, pruned_loss=0.09934, over 29516.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3586, pruned_loss=0.106, over 5693584.63 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3368, pruned_loss=0.08949, over 5782003.43 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.3588, pruned_loss=0.1065, over 5685178.09 frames. ], batch size: 80, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:38:55,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795764.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:39:32,736 INFO [train.py:968] (1/2) Epoch 18, batch 19100, giga_loss[loss=0.2752, simple_loss=0.3453, pruned_loss=0.1025, over 28707.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.358, pruned_loss=0.1063, over 5700948.99 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.337, pruned_loss=0.08948, over 5786491.54 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3585, pruned_loss=0.1072, over 5687692.22 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:39:47,312 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:39:50,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:05,686 INFO [optim.py:369] (1/2) 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 09:40:15,411 INFO [train.py:968] (1/2) Epoch 18, batch 19150, giga_loss[loss=0.2539, simple_loss=0.3377, pruned_loss=0.08502, over 28203.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3548, pruned_loss=0.1047, over 5705662.22 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3374, pruned_loss=0.08955, over 5787189.18 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3552, pruned_loss=0.1056, over 5693498.08 frames. ], batch size: 77, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:40:16,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:39,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:39,563 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:43,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:55,115 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:40:57,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=795907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:40:59,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=795910.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:41:00,955 INFO [train.py:968] (1/2) Epoch 18, batch 19200, giga_loss[loss=0.2807, simple_loss=0.3593, pruned_loss=0.101, over 28685.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3545, pruned_loss=0.1045, over 5696158.38 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3377, pruned_loss=0.08957, over 5789419.71 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3549, pruned_loss=0.1055, over 5682524.36 frames. ], batch size: 307, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:41:07,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:41:22,016 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=795939.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:41:27,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:41:32,281 INFO [optim.py:369] (1/2) 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 09:41:35,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3191, 1.5011, 1.0557, 1.0254], device='cuda:1'), covar=tensor([0.1064, 0.0617, 0.1261, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0439, 0.0510, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 09:41:40,737 INFO [train.py:968] (1/2) Epoch 18, batch 19250, giga_loss[loss=0.2503, simple_loss=0.3288, pruned_loss=0.08588, over 28509.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3528, pruned_loss=0.1028, over 5694135.44 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3378, pruned_loss=0.08963, over 5781343.89 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3533, pruned_loss=0.1038, over 5688073.65 frames. ], batch size: 60, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:41:41,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=795963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:41:48,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 09:42:03,721 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2212, 1.4509, 1.2938, 1.1429], device='cuda:1'), covar=tensor([0.2582, 0.2523, 0.1717, 0.2167], device='cuda:1'), in_proj_covar=tensor([0.1892, 0.1809, 0.1741, 0.1889], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 09:42:16,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=796002.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:42:23,884 INFO [train.py:968] (1/2) Epoch 18, batch 19300, giga_loss[loss=0.236, simple_loss=0.3197, pruned_loss=0.07609, over 28881.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3486, pruned_loss=0.09963, over 5697676.27 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3378, pruned_loss=0.08949, over 5786341.24 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3495, pruned_loss=0.101, over 5685299.52 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:42:32,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 09:42:43,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:45,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:54,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6150, 1.7606, 1.8492, 1.4195], device='cuda:1'), covar=tensor([0.1793, 0.2417, 0.1429, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0690, 0.0921, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 09:42:54,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:57,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:42:57,625 INFO [optim.py:369] (1/2) 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 09:43:09,368 INFO [train.py:968] (1/2) Epoch 18, batch 19350, giga_loss[loss=0.2452, simple_loss=0.3234, pruned_loss=0.0835, over 29029.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.09611, over 5695699.23 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3381, pruned_loss=0.08947, over 5789572.61 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3431, pruned_loss=0.09739, over 5680756.99 frames. ], batch size: 128, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:43:10,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:24,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:33,377 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:35,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:47,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:43:48,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7276, 1.0209, 3.0001, 2.8005], device='cuda:1'), covar=tensor([0.1718, 0.2587, 0.0597, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0722, 0.0628, 0.0920, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 09:43:49,542 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 19400, giga_loss[loss=0.2653, simple_loss=0.3304, pruned_loss=0.1001, over 28004.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3382, pruned_loss=0.09401, over 5698706.51 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3386, pruned_loss=0.08974, over 5792587.60 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3382, pruned_loss=0.0949, over 5681368.01 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:44:00,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:44:18,899 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:44:25,487 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796145.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 09:44:27,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796148.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:44:30,143 INFO [optim.py:369] (1/2) 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 09:44:42,136 INFO [train.py:968] (1/2) Epoch 18, batch 19450, giga_loss[loss=0.2627, simple_loss=0.3354, pruned_loss=0.09498, over 28629.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3338, pruned_loss=0.09184, over 5694639.61 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3386, pruned_loss=0.08973, over 5793118.66 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3337, pruned_loss=0.09256, over 5680146.92 frames. ], batch size: 85, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:44:56,705 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796177.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 09:45:15,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:45:24,543 INFO [train.py:968] (1/2) Epoch 18, batch 19500, giga_loss[loss=0.2644, simple_loss=0.3429, pruned_loss=0.09297, over 28940.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3353, pruned_loss=0.09223, over 5698883.48 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.339, pruned_loss=0.08986, over 5794318.13 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3348, pruned_loss=0.09271, over 5684448.62 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:45:28,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4196, 3.4881, 1.4256, 1.6459], device='cuda:1'), covar=tensor([0.0936, 0.0284, 0.0889, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0531, 0.0368, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 09:45:28,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1400, 3.3996, 2.1573, 1.3126], device='cuda:1'), covar=tensor([0.5655, 0.2318, 0.3426, 0.5144], device='cuda:1'), in_proj_covar=tensor([0.1668, 0.1567, 0.1549, 0.1365], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 09:45:42,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3593, 1.8046, 1.4489, 1.6266], device='cuda:1'), covar=tensor([0.0806, 0.0304, 0.0332, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 09:45:58,507 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 19550, giga_loss[loss=0.255, simple_loss=0.3344, pruned_loss=0.08779, over 27909.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3355, pruned_loss=0.09172, over 5704053.28 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3396, pruned_loss=0.09008, over 5795862.15 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3344, pruned_loss=0.09195, over 5687977.00 frames. ], batch size: 412, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:46:45,424 INFO [train.py:968] (1/2) Epoch 18, batch 19600, giga_loss[loss=0.2407, simple_loss=0.3197, pruned_loss=0.08085, over 29008.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3336, pruned_loss=0.0907, over 5712161.58 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3401, pruned_loss=0.09024, over 5795371.01 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3322, pruned_loss=0.09076, over 5698183.83 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:47:18,526 INFO [optim.py:369] (1/2) 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 09:47:20,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:47:27,214 INFO [train.py:968] (1/2) Epoch 18, batch 19650, giga_loss[loss=0.2528, simple_loss=0.3295, pruned_loss=0.08806, over 29005.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3305, pruned_loss=0.08944, over 5719961.49 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3401, pruned_loss=0.09019, over 5796510.47 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3293, pruned_loss=0.08952, over 5707165.81 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:47:44,070 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:48:06,212 INFO [train.py:968] (1/2) Epoch 18, batch 19700, giga_loss[loss=0.2604, simple_loss=0.33, pruned_loss=0.09537, over 28916.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.329, pruned_loss=0.089, over 5713857.26 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3408, pruned_loss=0.09044, over 5787789.40 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3271, pruned_loss=0.08881, over 5710567.41 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:48:24,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3418, 2.0150, 1.5915, 0.6099], device='cuda:1'), covar=tensor([0.4782, 0.2532, 0.4128, 0.5470], device='cuda:1'), in_proj_covar=tensor([0.1668, 0.1571, 0.1552, 0.1367], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 09:48:36,206 INFO [optim.py:369] (1/2) 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 09:48:37,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-09 09:48:45,770 INFO [train.py:968] (1/2) Epoch 18, batch 19750, giga_loss[loss=0.2693, simple_loss=0.3542, pruned_loss=0.0922, over 29013.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3277, pruned_loss=0.08843, over 5715996.26 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3415, pruned_loss=0.0906, over 5788258.44 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3254, pruned_loss=0.08811, over 5711505.73 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:49:25,621 INFO [train.py:968] (1/2) Epoch 18, batch 19800, giga_loss[loss=0.217, simple_loss=0.2951, pruned_loss=0.06945, over 28604.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3248, pruned_loss=0.08719, over 5719527.94 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3417, pruned_loss=0.09051, over 5789946.83 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3225, pruned_loss=0.08698, over 5713425.09 frames. ], batch size: 60, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:49:54,925 INFO [optim.py:369] (1/2) 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 09:50:02,117 INFO [train.py:968] (1/2) Epoch 18, batch 19850, giga_loss[loss=0.2416, simple_loss=0.3165, pruned_loss=0.08339, over 28678.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3238, pruned_loss=0.08673, over 5727036.27 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3429, pruned_loss=0.09097, over 5793290.01 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3204, pruned_loss=0.08605, over 5717387.06 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:50:04,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3683, 3.1911, 2.9760, 1.8247], device='cuda:1'), covar=tensor([0.0738, 0.0846, 0.0747, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.1137, 0.1050, 0.0906, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 09:50:14,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=796575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:50:43,459 INFO [train.py:968] (1/2) Epoch 18, batch 19900, libri_loss[loss=0.2641, simple_loss=0.3445, pruned_loss=0.09186, over 29584.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3223, pruned_loss=0.08597, over 5722576.77 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3436, pruned_loss=0.09122, over 5792684.20 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3185, pruned_loss=0.08508, over 5713846.34 frames. ], batch size: 74, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:51:12,987 INFO [optim.py:369] (1/2) 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 09:51:19,443 INFO [train.py:968] (1/2) Epoch 18, batch 19950, giga_loss[loss=0.2294, simple_loss=0.2975, pruned_loss=0.08068, over 28654.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3211, pruned_loss=0.08463, over 5734265.23 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3442, pruned_loss=0.09112, over 5795326.19 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3164, pruned_loss=0.08371, over 5721465.20 frames. ], batch size: 85, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:51:22,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5304, 1.7896, 1.4230, 1.4204], device='cuda:1'), covar=tensor([0.2445, 0.2562, 0.2845, 0.2410], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1043, 0.1274, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 09:51:56,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2253, 4.0607, 3.8383, 1.7429], device='cuda:1'), covar=tensor([0.0566, 0.0711, 0.0667, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.1141, 0.1054, 0.0909, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 09:51:56,637 INFO [train.py:968] (1/2) Epoch 18, batch 20000, giga_loss[loss=0.2308, simple_loss=0.3057, pruned_loss=0.07794, over 28808.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3203, pruned_loss=0.08414, over 5725662.58 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3453, pruned_loss=0.09155, over 5786592.98 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3149, pruned_loss=0.08283, over 5721991.40 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:52:01,112 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:04,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:06,299 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:10,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=796729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:16,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9273, 1.1671, 1.3381, 1.0387], device='cuda:1'), covar=tensor([0.2099, 0.1433, 0.2392, 0.1818], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0744, 0.0703, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 09:52:26,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:52:27,497 INFO [optim.py:369] (1/2) 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 09:52:35,254 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 20050, giga_loss[loss=0.2796, simple_loss=0.3531, pruned_loss=0.103, over 28770.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.319, pruned_loss=0.08367, over 5734558.27 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3453, pruned_loss=0.09151, over 5787198.02 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.3146, pruned_loss=0.08265, over 5730938.97 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:53:16,722 INFO [train.py:968] (1/2) Epoch 18, batch 20100, giga_loss[loss=0.2259, simple_loss=0.3057, pruned_loss=0.07305, over 28436.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3243, pruned_loss=0.08739, over 5713514.48 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09148, over 5778817.61 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.32, pruned_loss=0.08641, over 5717374.15 frames. ], batch size: 78, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:53:19,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2603, 2.5791, 1.2652, 1.4114], device='cuda:1'), covar=tensor([0.1009, 0.0401, 0.0888, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0535, 0.0370, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 09:53:29,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:53:55,820 INFO [optim.py:369] (1/2) 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 09:54:04,842 INFO [train.py:968] (1/2) Epoch 18, batch 20150, libri_loss[loss=0.2065, simple_loss=0.2926, pruned_loss=0.06021, over 29336.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3313, pruned_loss=0.09181, over 5716211.29 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3458, pruned_loss=0.09153, over 5781769.74 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3272, pruned_loss=0.09095, over 5715133.24 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:54:14,520 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:17,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:44,472 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=796902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:46,668 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:47,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=796905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:48,707 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=796907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:54:52,749 INFO [train.py:968] (1/2) Epoch 18, batch 20200, giga_loss[loss=0.3564, simple_loss=0.4004, pruned_loss=0.1562, over 26677.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.34, pruned_loss=0.09767, over 5694762.65 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3461, pruned_loss=0.09164, over 5779548.47 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3363, pruned_loss=0.09697, over 5694158.58 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:55:04,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4766, 1.6312, 1.7179, 1.3379], device='cuda:1'), covar=tensor([0.1776, 0.2456, 0.1460, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0694, 0.0926, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 09:55:12,829 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=796934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:55:28,539 INFO [optim.py:369] (1/2) 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 09:55:38,463 INFO [train.py:968] (1/2) Epoch 18, batch 20250, libri_loss[loss=0.2702, simple_loss=0.3564, pruned_loss=0.09196, over 26006.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3434, pruned_loss=0.09828, over 5694054.06 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3461, pruned_loss=0.09157, over 5777484.03 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3403, pruned_loss=0.09792, over 5693973.79 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:56:18,487 INFO [train.py:968] (1/2) Epoch 18, batch 20300, giga_loss[loss=0.2965, simple_loss=0.3685, pruned_loss=0.1122, over 29057.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3484, pruned_loss=0.1008, over 5682946.37 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.346, pruned_loss=0.09148, over 5778867.94 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.346, pruned_loss=0.1009, over 5678104.87 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:56:54,761 INFO [optim.py:369] (1/2) 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 09:57:03,422 INFO [train.py:968] (1/2) Epoch 18, batch 20350, giga_loss[loss=0.306, simple_loss=0.3812, pruned_loss=0.1155, over 28503.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3546, pruned_loss=0.105, over 5673000.05 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3463, pruned_loss=0.09161, over 5776765.43 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3525, pruned_loss=0.1051, over 5669328.90 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 09:57:34,102 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 20400, giga_loss[loss=0.2419, simple_loss=0.3322, pruned_loss=0.07577, over 28740.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3572, pruned_loss=0.1064, over 5677298.87 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3468, pruned_loss=0.092, over 5779803.99 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3553, pruned_loss=0.1064, over 5669827.77 frames. ], batch size: 78, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:58:02,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3580, 1.7036, 1.3110, 1.4093], device='cuda:1'), covar=tensor([0.2591, 0.2543, 0.2921, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1429, 0.1038, 0.1269, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 09:58:18,119 INFO [optim.py:369] (1/2) 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 09:58:26,744 INFO [train.py:968] (1/2) Epoch 18, batch 20450, giga_loss[loss=0.2631, simple_loss=0.3401, pruned_loss=0.09299, over 28884.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3525, pruned_loss=0.1029, over 5680433.74 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3469, pruned_loss=0.09214, over 5781660.71 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.351, pruned_loss=0.103, over 5671118.45 frames. ], batch size: 112, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:58:33,014 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-09 09:58:59,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:59:08,116 INFO [train.py:968] (1/2) Epoch 18, batch 20500, giga_loss[loss=0.2686, simple_loss=0.341, pruned_loss=0.09805, over 28916.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3499, pruned_loss=0.1001, over 5701388.18 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3469, pruned_loss=0.09219, over 5785130.24 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3488, pruned_loss=0.1004, over 5688683.65 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 09:59:15,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2008, 2.2539, 1.6502, 1.9311], device='cuda:1'), covar=tensor([0.0888, 0.0714, 0.1024, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0443, 0.0510, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 09:59:34,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:59:38,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=797245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 09:59:43,983 INFO [optim.py:369] (1/2) 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 09:59:50,870 INFO [train.py:968] (1/2) Epoch 18, batch 20550, giga_loss[loss=0.2296, simple_loss=0.3196, pruned_loss=0.06981, over 28815.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3497, pruned_loss=0.09995, over 5694603.05 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3467, pruned_loss=0.09219, over 5787131.31 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3491, pruned_loss=0.1003, over 5681164.87 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:00:00,323 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:00:07,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:00:10,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=797287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:00:31,322 INFO [train.py:968] (1/2) Epoch 18, batch 20600, giga_loss[loss=0.3261, simple_loss=0.3733, pruned_loss=0.1395, over 23835.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3501, pruned_loss=0.09957, over 5695951.24 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09162, over 5788027.10 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3504, pruned_loss=0.1005, over 5682081.98 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:00:31,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4879, 2.1958, 1.7326, 0.7841], device='cuda:1'), covar=tensor([0.5992, 0.2603, 0.3586, 0.6045], device='cuda:1'), in_proj_covar=tensor([0.1674, 0.1580, 0.1567, 0.1378], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 10:00:34,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-09 10:00:59,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:01:01,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=797347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:01:05,660 INFO [optim.py:369] (1/2) 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 10:01:16,230 INFO [train.py:968] (1/2) Epoch 18, batch 20650, giga_loss[loss=0.2653, simple_loss=0.3402, pruned_loss=0.09526, over 28816.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.353, pruned_loss=0.1018, over 5709660.97 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3466, pruned_loss=0.092, over 5792156.54 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3528, pruned_loss=0.1025, over 5692010.81 frames. ], batch size: 66, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:01:27,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:01:59,668 INFO [train.py:968] (1/2) Epoch 18, batch 20700, giga_loss[loss=0.2765, simple_loss=0.3553, pruned_loss=0.09882, over 28875.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3552, pruned_loss=0.1037, over 5699882.12 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3468, pruned_loss=0.09211, over 5793443.19 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3549, pruned_loss=0.1043, over 5683769.74 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:02:02,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4469, 3.5491, 1.7447, 1.5101], device='cuda:1'), covar=tensor([0.1004, 0.0243, 0.0818, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0536, 0.0370, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 10:02:09,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:02:13,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=797428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:02:37,039 INFO [optim.py:369] (1/2) 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 10:02:39,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:02:43,304 INFO [train.py:968] (1/2) Epoch 18, batch 20750, giga_loss[loss=0.3251, simple_loss=0.3825, pruned_loss=0.1339, over 28795.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3554, pruned_loss=0.1041, over 5695832.83 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3467, pruned_loss=0.09189, over 5794483.90 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3555, pruned_loss=0.1051, over 5679789.49 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:03:25,478 INFO [train.py:968] (1/2) Epoch 18, batch 20800, giga_loss[loss=0.2515, simple_loss=0.3279, pruned_loss=0.08752, over 28368.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3554, pruned_loss=0.1045, over 5699503.50 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3467, pruned_loss=0.09183, over 5795597.25 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3556, pruned_loss=0.1055, over 5685110.28 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:03:56,620 INFO [optim.py:369] (1/2) 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 10:03:59,458 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 20850, giga_loss[loss=0.3116, simple_loss=0.3855, pruned_loss=0.1189, over 28671.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3565, pruned_loss=0.1046, over 5707166.28 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3473, pruned_loss=0.0921, over 5795401.03 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3564, pruned_loss=0.1054, over 5693394.00 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:04:18,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 10:04:25,341 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-09 10:04:27,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=797595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:04:28,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5688, 2.1966, 1.6248, 0.6848], device='cuda:1'), covar=tensor([0.5236, 0.2606, 0.4083, 0.6008], device='cuda:1'), in_proj_covar=tensor([0.1670, 0.1577, 0.1559, 0.1373], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 10:04:40,800 INFO [train.py:968] (1/2) Epoch 18, batch 20900, giga_loss[loss=0.2895, simple_loss=0.3656, pruned_loss=0.1067, over 29046.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3551, pruned_loss=0.1026, over 5692369.94 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3474, pruned_loss=0.09217, over 5782970.55 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3551, pruned_loss=0.1036, over 5689131.51 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:04:47,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1940, 4.0361, 3.7908, 1.6727], device='cuda:1'), covar=tensor([0.0598, 0.0741, 0.0734, 0.2279], device='cuda:1'), in_proj_covar=tensor([0.1150, 0.1063, 0.0915, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 10:05:05,881 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1929, 1.3548, 1.0573, 0.9703], device='cuda:1'), covar=tensor([0.0915, 0.0475, 0.1100, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0440, 0.0508, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:05:06,637 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-09 10:05:13,896 INFO [optim.py:369] (1/2) 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 10:05:16,077 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 20950, giga_loss[loss=0.26, simple_loss=0.336, pruned_loss=0.09196, over 28797.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3552, pruned_loss=0.1016, over 5699497.68 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3472, pruned_loss=0.09193, over 5785259.55 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3556, pruned_loss=0.1028, over 5693436.03 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:05:20,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:05:54,187 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-09 10:05:57,800 INFO [train.py:968] (1/2) Epoch 18, batch 21000, giga_loss[loss=0.3046, simple_loss=0.3551, pruned_loss=0.1271, over 26622.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3536, pruned_loss=0.1009, over 5703388.51 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3471, pruned_loss=0.09185, over 5787225.67 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3541, pruned_loss=0.1022, over 5694986.49 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:05:57,800 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 10:06:06,440 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 10:06:31,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7036, 1.9058, 1.5205, 2.0949], device='cuda:1'), covar=tensor([0.2561, 0.2542, 0.2879, 0.2280], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1045, 0.1274, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 10:06:38,729 INFO [optim.py:369] (1/2) 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 10:06:44,033 INFO [train.py:968] (1/2) Epoch 18, batch 21050, giga_loss[loss=0.2604, simple_loss=0.334, pruned_loss=0.09346, over 28780.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3509, pruned_loss=0.09954, over 5706923.24 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3466, pruned_loss=0.09165, over 5779446.92 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3519, pruned_loss=0.1009, over 5706311.51 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:07:17,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=797805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:07:20,090 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 21100, giga_loss[loss=0.23, simple_loss=0.3085, pruned_loss=0.07577, over 28427.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3496, pruned_loss=0.09921, over 5710349.44 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3467, pruned_loss=0.0916, over 5782806.40 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3505, pruned_loss=0.1005, over 5704989.30 frames. ], batch size: 65, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:07:34,162 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3119, 1.6231, 1.3079, 0.9678], device='cuda:1'), covar=tensor([0.2387, 0.2343, 0.2619, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1045, 0.1273, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 10:07:41,482 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=797837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:07:56,461 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 21150, giga_loss[loss=0.2867, simple_loss=0.3383, pruned_loss=0.1175, over 23365.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3495, pruned_loss=0.09982, over 5706615.41 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3471, pruned_loss=0.09204, over 5784419.73 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3498, pruned_loss=0.1006, over 5700074.99 frames. ], batch size: 705, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:08:13,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5286, 1.6002, 1.5450, 1.3960], device='cuda:1'), covar=tensor([0.2451, 0.2490, 0.2002, 0.2347], device='cuda:1'), in_proj_covar=tensor([0.1878, 0.1803, 0.1737, 0.1890], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 10:08:13,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-09 10:08:46,846 INFO [train.py:968] (1/2) Epoch 18, batch 21200, giga_loss[loss=0.2695, simple_loss=0.3517, pruned_loss=0.09368, over 28801.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.351, pruned_loss=0.1011, over 5713462.88 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3469, pruned_loss=0.09192, over 5784957.63 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3514, pruned_loss=0.1018, over 5707595.34 frames. ], batch size: 242, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:09:01,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:09:19,561 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 21250, giga_loss[loss=0.2452, simple_loss=0.3288, pruned_loss=0.08082, over 28783.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3506, pruned_loss=0.1006, over 5711301.24 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3466, pruned_loss=0.09178, over 5787719.57 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3513, pruned_loss=0.1015, over 5703043.39 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:09:32,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=797970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:09:48,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7217, 4.5740, 4.2956, 2.1908], device='cuda:1'), covar=tensor([0.0498, 0.0690, 0.0743, 0.1930], device='cuda:1'), in_proj_covar=tensor([0.1146, 0.1060, 0.0912, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 10:10:05,808 INFO [train.py:968] (1/2) Epoch 18, batch 21300, giga_loss[loss=0.281, simple_loss=0.3613, pruned_loss=0.1004, over 28923.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.349, pruned_loss=0.09861, over 5715284.48 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3467, pruned_loss=0.09182, over 5788919.57 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3495, pruned_loss=0.09932, over 5707098.32 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:10:22,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=798031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:10:27,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798038.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:10:28,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6526, 1.8394, 1.3869, 1.3475], device='cuda:1'), covar=tensor([0.0979, 0.0570, 0.0979, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0440, 0.0507, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:10:39,865 INFO [optim.py:369] (1/2) 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 10:10:45,669 INFO [train.py:968] (1/2) Epoch 18, batch 21350, giga_loss[loss=0.2576, simple_loss=0.3404, pruned_loss=0.08742, over 28777.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3486, pruned_loss=0.09811, over 5725532.28 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3471, pruned_loss=0.09213, over 5791371.42 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3486, pruned_loss=0.09849, over 5715515.24 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:10:49,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6177, 1.7136, 1.4392, 1.5675], device='cuda:1'), covar=tensor([0.3133, 0.2922, 0.3426, 0.2464], device='cuda:1'), in_proj_covar=tensor([0.1433, 0.1044, 0.1271, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 10:10:55,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:10:58,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:11,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4431, 3.8333, 1.6430, 1.6331], device='cuda:1'), covar=tensor([0.0913, 0.0262, 0.0859, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0529, 0.0366, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 10:11:22,078 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 21400, giga_loss[loss=0.264, simple_loss=0.3395, pruned_loss=0.09419, over 28711.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3472, pruned_loss=0.09744, over 5727761.66 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3476, pruned_loss=0.09252, over 5792374.05 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3467, pruned_loss=0.09755, over 5716339.96 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:11:26,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:28,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:49,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=798145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:11:59,242 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 21450, giga_loss[loss=0.2657, simple_loss=0.341, pruned_loss=0.09525, over 28999.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3445, pruned_loss=0.09622, over 5721224.52 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3479, pruned_loss=0.09278, over 5791768.94 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3437, pruned_loss=0.09616, over 5711090.55 frames. ], batch size: 136, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:12:13,394 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:12:15,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:12:32,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5186, 1.7585, 1.4362, 1.7013], device='cuda:1'), covar=tensor([0.2586, 0.2529, 0.2838, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1045, 0.1272, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 10:12:34,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 10:12:37,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=798206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:12:41,890 INFO [train.py:968] (1/2) Epoch 18, batch 21500, giga_loss[loss=0.3314, simple_loss=0.4006, pruned_loss=0.1311, over 26764.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3433, pruned_loss=0.09605, over 5723581.69 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.348, pruned_loss=0.09291, over 5789828.36 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3427, pruned_loss=0.09595, over 5716070.64 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:13:00,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8163, 2.1501, 2.0639, 1.6173], device='cuda:1'), covar=tensor([0.1891, 0.2527, 0.1594, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.0880, 0.0698, 0.0928, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 10:13:01,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6460, 1.9987, 1.7189, 1.7640], device='cuda:1'), covar=tensor([0.0723, 0.0264, 0.0305, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 10:13:15,242 INFO [optim.py:369] (1/2) 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 10:13:20,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 10:13:20,745 INFO [train.py:968] (1/2) Epoch 18, batch 21550, giga_loss[loss=0.2437, simple_loss=0.324, pruned_loss=0.08167, over 28837.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3422, pruned_loss=0.09556, over 5721172.92 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3478, pruned_loss=0.09288, over 5781915.44 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3418, pruned_loss=0.09559, over 5721402.72 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:13:55,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5103, 1.6533, 1.7132, 1.2972], device='cuda:1'), covar=tensor([0.1803, 0.2569, 0.1499, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0697, 0.0927, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 10:14:00,660 INFO [train.py:968] (1/2) Epoch 18, batch 21600, giga_loss[loss=0.2514, simple_loss=0.3291, pruned_loss=0.08691, over 28890.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3414, pruned_loss=0.09601, over 5716823.66 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3479, pruned_loss=0.09312, over 5784078.69 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3408, pruned_loss=0.09586, over 5714407.28 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:14:06,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-09 10:14:24,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-09 10:14:38,317 INFO [optim.py:369] (1/2) 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 10:14:43,054 INFO [train.py:968] (1/2) Epoch 18, batch 21650, giga_loss[loss=0.2497, simple_loss=0.3243, pruned_loss=0.0876, over 28929.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3402, pruned_loss=0.09584, over 5713248.06 frames. ], libri_tot_loss[loss=0.2679, simple_loss=0.3487, pruned_loss=0.09361, over 5781604.90 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3389, pruned_loss=0.09532, over 5712096.29 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:14:55,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2752, 1.2506, 4.0554, 3.2553], device='cuda:1'), covar=tensor([0.1653, 0.2751, 0.0385, 0.0808], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0622, 0.0914, 0.0850], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:15:06,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2230, 1.5267, 1.4915, 1.3740], device='cuda:1'), covar=tensor([0.1801, 0.1585, 0.2194, 0.1854], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0740, 0.0700, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 10:15:23,255 INFO [train.py:968] (1/2) Epoch 18, batch 21700, giga_loss[loss=0.2248, simple_loss=0.3039, pruned_loss=0.07281, over 28678.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3382, pruned_loss=0.09524, over 5712623.86 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.349, pruned_loss=0.09392, over 5783886.52 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3366, pruned_loss=0.09457, over 5708426.59 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:15:24,023 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=798413.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:15:56,126 INFO [optim.py:369] (1/2) 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 10:15:59,086 INFO [train.py:968] (1/2) Epoch 18, batch 21750, giga_loss[loss=0.2624, simple_loss=0.3398, pruned_loss=0.09251, over 28925.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3375, pruned_loss=0.09555, over 5714297.60 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3488, pruned_loss=0.09409, over 5783236.29 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.336, pruned_loss=0.0949, over 5708772.95 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:16:40,895 INFO [train.py:968] (1/2) Epoch 18, batch 21800, giga_loss[loss=0.2415, simple_loss=0.322, pruned_loss=0.08045, over 29074.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.337, pruned_loss=0.09532, over 5701990.36 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3496, pruned_loss=0.09471, over 5774238.56 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3349, pruned_loss=0.09429, over 5703738.32 frames. ], batch size: 155, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:16:46,666 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:17:18,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=798556.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:17:18,703 INFO [optim.py:369] (1/2) 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 10:17:20,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=798559.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:17:22,140 INFO [train.py:968] (1/2) Epoch 18, batch 21850, libri_loss[loss=0.3078, simple_loss=0.3799, pruned_loss=0.1179, over 29379.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3386, pruned_loss=0.09552, over 5705127.13 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3502, pruned_loss=0.09519, over 5773391.60 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3361, pruned_loss=0.09424, over 5705904.45 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:17:44,728 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=798588.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:18:01,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2223, 1.1853, 4.1819, 3.3457], device='cuda:1'), covar=tensor([0.1769, 0.2931, 0.0357, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0622, 0.0916, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:18:05,527 INFO [train.py:968] (1/2) Epoch 18, batch 21900, giga_loss[loss=0.241, simple_loss=0.3243, pruned_loss=0.07884, over 28921.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3413, pruned_loss=0.09654, over 5709499.07 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3504, pruned_loss=0.09544, over 5776341.41 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.339, pruned_loss=0.09532, over 5706127.32 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 10:18:42,908 INFO [optim.py:369] (1/2) 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 10:18:46,267 INFO [train.py:968] (1/2) Epoch 18, batch 21950, giga_loss[loss=0.2642, simple_loss=0.3478, pruned_loss=0.09031, over 28605.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3439, pruned_loss=0.09695, over 5711042.62 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3512, pruned_loss=0.09608, over 5778828.17 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3411, pruned_loss=0.09543, over 5704611.80 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 2.0
+2023-03-09 10:19:26,423 INFO [train.py:968] (1/2) Epoch 18, batch 22000, giga_loss[loss=0.218, simple_loss=0.2951, pruned_loss=0.0705, over 28514.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3443, pruned_loss=0.09681, over 5708401.09 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3513, pruned_loss=0.0964, over 5780699.81 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3418, pruned_loss=0.09533, over 5699798.75 frames. ], batch size: 78, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:20:05,515 INFO [optim.py:369] (1/2) 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 10:20:09,236 INFO [train.py:968] (1/2) Epoch 18, batch 22050, giga_loss[loss=0.2219, simple_loss=0.2986, pruned_loss=0.07265, over 28481.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3441, pruned_loss=0.09672, over 5691191.36 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3517, pruned_loss=0.09673, over 5772101.23 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3418, pruned_loss=0.09527, over 5690651.96 frames. ], batch size: 71, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:20:43,740 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798803.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:20:49,964 INFO [train.py:968] (1/2) Epoch 18, batch 22100, libri_loss[loss=0.2617, simple_loss=0.3447, pruned_loss=0.08931, over 29520.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3448, pruned_loss=0.09739, over 5695814.17 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3518, pruned_loss=0.09682, over 5763894.87 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3427, pruned_loss=0.09615, over 5700613.65 frames. ], batch size: 84, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:21:01,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5223, 2.3928, 1.8022, 0.7585], device='cuda:1'), covar=tensor([0.4799, 0.2713, 0.3600, 0.5389], device='cuda:1'), in_proj_covar=tensor([0.1672, 0.1576, 0.1563, 0.1371], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 10:21:15,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=798844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:21:26,830 INFO [optim.py:369] (1/2) 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 10:21:30,244 INFO [train.py:968] (1/2) Epoch 18, batch 22150, giga_loss[loss=0.251, simple_loss=0.3341, pruned_loss=0.08397, over 28525.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.344, pruned_loss=0.09711, over 5698361.70 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3518, pruned_loss=0.09702, over 5765493.64 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3422, pruned_loss=0.09596, over 5699734.60 frames. ], batch size: 336, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:21:58,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=798894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:22:12,291 INFO [train.py:968] (1/2) Epoch 18, batch 22200, giga_loss[loss=0.2732, simple_loss=0.347, pruned_loss=0.09972, over 28618.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3466, pruned_loss=0.09894, over 5697966.31 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3518, pruned_loss=0.09707, over 5766938.17 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3451, pruned_loss=0.09799, over 5697168.39 frames. ], batch size: 92, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:22:49,262 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 22250, giga_loss[loss=0.3139, simple_loss=0.3834, pruned_loss=0.1222, over 28689.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3492, pruned_loss=0.1001, over 5704561.37 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3519, pruned_loss=0.09726, over 5768092.35 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3479, pruned_loss=0.09923, over 5702321.02 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:22:53,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2531, 4.0756, 3.8921, 1.8136], device='cuda:1'), covar=tensor([0.0621, 0.0786, 0.0786, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.1140, 0.1055, 0.0907, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 10:23:31,975 INFO [train.py:968] (1/2) Epoch 18, batch 22300, giga_loss[loss=0.3078, simple_loss=0.3921, pruned_loss=0.1117, over 28966.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 5696845.84 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3525, pruned_loss=0.09763, over 5759078.03 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3497, pruned_loss=0.1003, over 5702042.46 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:23:50,966 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:23:52,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:24:06,523 INFO [optim.py:369] (1/2) 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 10:24:09,139 INFO [train.py:968] (1/2) Epoch 18, batch 22350, giga_loss[loss=0.3019, simple_loss=0.3738, pruned_loss=0.1151, over 28312.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3527, pruned_loss=0.1018, over 5708950.03 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3534, pruned_loss=0.09835, over 5763044.29 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3505, pruned_loss=0.1005, over 5707987.87 frames. ], batch size: 368, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:24:14,934 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:24:48,661 INFO [train.py:968] (1/2) Epoch 18, batch 22400, giga_loss[loss=0.3407, simple_loss=0.3966, pruned_loss=0.1424, over 26684.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3532, pruned_loss=0.1022, over 5706145.49 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3531, pruned_loss=0.09841, over 5759048.15 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3517, pruned_loss=0.1012, over 5706473.24 frames. ], batch size: 555, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:25:23,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2090, 1.2664, 3.3775, 2.9916], device='cuda:1'), covar=tensor([0.1498, 0.2697, 0.0457, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0627, 0.0921, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:25:25,444 INFO [optim.py:369] (1/2) 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 10:25:27,417 INFO [train.py:968] (1/2) Epoch 18, batch 22450, giga_loss[loss=0.2617, simple_loss=0.3401, pruned_loss=0.09163, over 28887.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3524, pruned_loss=0.1017, over 5687682.97 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3533, pruned_loss=0.09854, over 5740131.86 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.351, pruned_loss=0.1008, over 5704644.82 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:25:42,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=799178.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:26:06,480 INFO [train.py:968] (1/2) Epoch 18, batch 22500, giga_loss[loss=0.2522, simple_loss=0.3404, pruned_loss=0.08201, over 28890.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3499, pruned_loss=0.1004, over 5700650.52 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3538, pruned_loss=0.09925, over 5743249.46 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3482, pruned_loss=0.09913, over 5709958.37 frames. ], batch size: 145, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:26:12,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=799219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:26:36,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:26:44,560 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 22550, giga_loss[loss=0.2886, simple_loss=0.3586, pruned_loss=0.1093, over 28709.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3458, pruned_loss=0.09842, over 5700777.96 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3537, pruned_loss=0.09938, over 5745665.29 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.09726, over 5704952.47 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:27:25,346 INFO [train.py:968] (1/2) Epoch 18, batch 22600, giga_loss[loss=0.2931, simple_loss=0.3682, pruned_loss=0.109, over 28596.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3444, pruned_loss=0.09751, over 5705956.81 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3544, pruned_loss=0.09983, over 5747983.43 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.09614, over 5706103.82 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:27:32,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799321.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:27:34,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799324.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:27:45,586 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 10:27:57,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2598, 4.0724, 3.8306, 1.8811], device='cuda:1'), covar=tensor([0.0506, 0.0648, 0.0661, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.1151, 0.1065, 0.0917, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 10:27:57,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799353.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:28:02,750 INFO [optim.py:369] (1/2) 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 10:28:05,804 INFO [train.py:968] (1/2) Epoch 18, batch 22650, giga_loss[loss=0.2404, simple_loss=0.3317, pruned_loss=0.07459, over 28958.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3445, pruned_loss=0.09603, over 5697319.88 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3544, pruned_loss=0.09996, over 5742929.15 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3427, pruned_loss=0.09473, over 5699948.03 frames. ], batch size: 164, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:28:06,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:28:07,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:28:31,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:28:46,591 INFO [train.py:968] (1/2) Epoch 18, batch 22700, giga_loss[loss=0.2587, simple_loss=0.3424, pruned_loss=0.08754, over 28877.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3459, pruned_loss=0.09622, over 5701431.19 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.354, pruned_loss=0.1, over 5746144.30 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3446, pruned_loss=0.09503, over 5699645.80 frames. ], batch size: 227, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:28:48,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-09 10:29:21,630 INFO [optim.py:369] (1/2) 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 10:29:24,739 INFO [train.py:968] (1/2) Epoch 18, batch 22750, giga_loss[loss=0.2653, simple_loss=0.3435, pruned_loss=0.0935, over 28838.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3449, pruned_loss=0.09647, over 5691991.66 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3544, pruned_loss=0.1005, over 5738404.93 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3434, pruned_loss=0.09507, over 5697456.51 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:29:35,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-09 10:29:41,007 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:30:03,355 INFO [train.py:968] (1/2) Epoch 18, batch 22800, giga_loss[loss=0.2977, simple_loss=0.3616, pruned_loss=0.1169, over 28767.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3449, pruned_loss=0.09832, over 5696408.99 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3551, pruned_loss=0.1011, over 5740363.24 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.343, pruned_loss=0.09662, over 5698144.13 frames. ], batch size: 284, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:30:18,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4707, 1.6816, 1.3719, 1.3206], device='cuda:1'), covar=tensor([0.2491, 0.2558, 0.2978, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1040, 0.1268, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 10:30:23,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8191, 2.0698, 1.6277, 2.0948], device='cuda:1'), covar=tensor([0.2388, 0.2469, 0.2891, 0.2431], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1041, 0.1268, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0012, 0.0011, 0.0011, 0.0009], device='cuda:1')
+2023-03-09 10:30:42,186 INFO [optim.py:369] (1/2) 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 10:30:44,920 INFO [train.py:968] (1/2) Epoch 18, batch 22850, giga_loss[loss=0.3055, simple_loss=0.3643, pruned_loss=0.1233, over 28762.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3432, pruned_loss=0.09841, over 5703909.72 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3554, pruned_loss=0.1014, over 5741782.92 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3411, pruned_loss=0.09671, over 5703091.13 frames. ], batch size: 243, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:31:21,914 INFO [train.py:968] (1/2) Epoch 18, batch 22900, giga_loss[loss=0.261, simple_loss=0.3338, pruned_loss=0.09408, over 28851.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3418, pruned_loss=0.09873, over 5701728.44 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3555, pruned_loss=0.1016, over 5736035.38 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3397, pruned_loss=0.09709, over 5705354.73 frames. ], batch size: 199, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:31:30,378 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:32:00,355 INFO [train.py:968] (1/2) Epoch 18, batch 22950, giga_loss[loss=0.2133, simple_loss=0.2907, pruned_loss=0.06792, over 28814.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3428, pruned_loss=0.09975, over 5703147.48 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3557, pruned_loss=0.1021, over 5733014.32 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3404, pruned_loss=0.09791, over 5707954.35 frames. ], batch size: 99, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:32:02,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-09 10:32:10,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2265, 1.5145, 1.5100, 1.0944], device='cuda:1'), covar=tensor([0.1749, 0.2447, 0.1456, 0.1653], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0695, 0.0922, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 10:32:37,026 INFO [train.py:968] (1/2) Epoch 18, batch 23000, giga_loss[loss=0.2469, simple_loss=0.3226, pruned_loss=0.08563, over 28927.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3397, pruned_loss=0.09779, over 5708870.65 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3562, pruned_loss=0.1025, over 5733174.17 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3369, pruned_loss=0.09577, over 5712015.34 frames. ], batch size: 174, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:32:55,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799736.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:33:11,505 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 23050, giga_loss[loss=0.2339, simple_loss=0.3055, pruned_loss=0.08112, over 28839.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3356, pruned_loss=0.09601, over 5705225.97 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3564, pruned_loss=0.1027, over 5736215.06 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3329, pruned_loss=0.09409, over 5704470.10 frames. ], batch size: 186, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:33:17,245 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=799765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:33:19,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=799768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:33:36,282 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-09 10:33:42,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=799797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:33:52,112 INFO [train.py:968] (1/2) Epoch 18, batch 23100, giga_loss[loss=0.2619, simple_loss=0.3384, pruned_loss=0.09275, over 28650.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3321, pruned_loss=0.09407, over 5701505.10 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3563, pruned_loss=0.1029, over 5727430.04 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3295, pruned_loss=0.09214, over 5707925.92 frames. ], batch size: 262, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:34:11,223 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=799835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:34:31,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=799859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:34:32,042 INFO [optim.py:369] (1/2) 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 10:34:33,949 INFO [train.py:968] (1/2) Epoch 18, batch 23150, giga_loss[loss=0.2654, simple_loss=0.3442, pruned_loss=0.09327, over 28718.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3323, pruned_loss=0.09357, over 5703928.18 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.356, pruned_loss=0.1029, over 5728155.04 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.33, pruned_loss=0.09197, over 5707996.50 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 4.0
+2023-03-09 10:35:14,272 INFO [train.py:968] (1/2) Epoch 18, batch 23200, giga_loss[loss=0.2488, simple_loss=0.3302, pruned_loss=0.08375, over 28820.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3345, pruned_loss=0.09394, over 5708280.09 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3561, pruned_loss=0.103, over 5730418.60 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3323, pruned_loss=0.09239, over 5709155.33 frames. ], batch size: 119, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:35:28,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 10:35:31,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0735, 3.8884, 3.6643, 1.7012], device='cuda:1'), covar=tensor([0.0607, 0.0725, 0.0705, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.1072, 0.0920, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 10:35:55,716 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 23250, giga_loss[loss=0.2639, simple_loss=0.3404, pruned_loss=0.09367, over 28956.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3389, pruned_loss=0.09581, over 5710736.66 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3563, pruned_loss=0.1032, over 5733305.80 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3366, pruned_loss=0.09433, over 5708563.75 frames. ], batch size: 106, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:36:19,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1534, 1.7663, 1.3903, 0.3647], device='cuda:1'), covar=tensor([0.4276, 0.2643, 0.3820, 0.5587], device='cuda:1'), in_proj_covar=tensor([0.1674, 0.1576, 0.1559, 0.1370], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 10:36:29,189 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:36:30,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:36:36,241 INFO [train.py:968] (1/2) Epoch 18, batch 23300, libri_loss[loss=0.2862, simple_loss=0.3605, pruned_loss=0.106, over 19562.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3426, pruned_loss=0.09748, over 5698893.22 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3563, pruned_loss=0.1033, over 5726997.97 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3404, pruned_loss=0.096, over 5703126.86 frames. ], batch size: 187, lr: 1.78e-03, grad_scale: 8.0
+2023-03-09 10:36:37,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5395, 1.4434, 1.2718, 1.1105], device='cuda:1'), covar=tensor([0.0626, 0.0356, 0.0785, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0441, 0.0507, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:36:44,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2599, 0.7829, 0.8518, 1.3831], device='cuda:1'), covar=tensor([0.0738, 0.0385, 0.0369, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 10:36:46,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0774, 2.1998, 1.9525, 1.7574], device='cuda:1'), covar=tensor([0.1740, 0.2254, 0.2212, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0739, 0.0699, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 10:36:55,301 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:37:20,404 INFO [train.py:968] (1/2) Epoch 18, batch 23350, giga_loss[loss=0.2732, simple_loss=0.3557, pruned_loss=0.09532, over 28835.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.346, pruned_loss=0.09938, over 5693263.13 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.356, pruned_loss=0.1032, over 5728723.98 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3444, pruned_loss=0.09826, over 5694932.26 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:38:01,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:38:02,071 INFO [train.py:968] (1/2) Epoch 18, batch 23400, giga_loss[loss=0.4718, simple_loss=0.4735, pruned_loss=0.2351, over 23639.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3505, pruned_loss=0.1031, over 5691508.06 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3571, pruned_loss=0.1042, over 5728640.31 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3479, pruned_loss=0.1011, over 5691454.94 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:38:09,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2000, 1.4710, 1.4969, 1.1087], device='cuda:1'), covar=tensor([0.1361, 0.2170, 0.1173, 0.1448], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0696, 0.0924, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 10:38:10,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4244, 1.5649, 1.2064, 1.1466], device='cuda:1'), covar=tensor([0.0868, 0.0513, 0.1020, 0.1092], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0441, 0.0507, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:38:35,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3294, 3.1267, 1.3945, 1.4658], device='cuda:1'), covar=tensor([0.1004, 0.0429, 0.0924, 0.1377], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0541, 0.0370, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 10:38:44,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4479, 3.9071, 1.6374, 1.6222], device='cuda:1'), covar=tensor([0.0942, 0.0372, 0.0902, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0541, 0.0370, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 10:38:47,498 INFO [optim.py:369] (1/2) 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 10:38:48,186 INFO [train.py:968] (1/2) Epoch 18, batch 23450, giga_loss[loss=0.2866, simple_loss=0.3566, pruned_loss=0.1083, over 29009.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3572, pruned_loss=0.1091, over 5690452.77 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3571, pruned_loss=0.1045, over 5733727.43 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.355, pruned_loss=0.1074, over 5684340.05 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:38:53,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1189, 1.2796, 3.8333, 3.1528], device='cuda:1'), covar=tensor([0.1772, 0.2753, 0.0452, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0626, 0.0922, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:39:18,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5165, 1.8121, 1.6189, 1.5372], device='cuda:1'), covar=tensor([0.0770, 0.0292, 0.0302, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 10:39:23,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 10:39:36,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:39:37,806 INFO [train.py:968] (1/2) Epoch 18, batch 23500, giga_loss[loss=0.3247, simple_loss=0.384, pruned_loss=0.1327, over 28847.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3637, pruned_loss=0.1139, over 5692636.43 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3571, pruned_loss=0.1047, over 5734261.76 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.362, pruned_loss=0.1125, over 5686285.50 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:39:57,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4487, 2.1484, 1.6155, 0.7866], device='cuda:1'), covar=tensor([0.4866, 0.2701, 0.3758, 0.5237], device='cuda:1'), in_proj_covar=tensor([0.1689, 0.1592, 0.1575, 0.1379], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 10:40:21,600 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:40:23,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800257.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:40:27,239 INFO [optim.py:369] (1/2) 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 10:40:28,012 INFO [train.py:968] (1/2) Epoch 18, batch 23550, giga_loss[loss=0.3391, simple_loss=0.401, pruned_loss=0.1386, over 28980.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3693, pruned_loss=0.1184, over 5682294.65 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3572, pruned_loss=0.1049, over 5733000.01 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.368, pruned_loss=0.1172, over 5677900.60 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:40:36,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9566, 1.2206, 0.9455, 0.2186], device='cuda:1'), covar=tensor([0.2512, 0.1898, 0.2540, 0.5005], device='cuda:1'), in_proj_covar=tensor([0.1689, 0.1595, 0.1575, 0.1380], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 10:40:47,331 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:40:48,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:41:14,368 INFO [train.py:968] (1/2) Epoch 18, batch 23600, giga_loss[loss=0.3353, simple_loss=0.3906, pruned_loss=0.14, over 28674.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3754, pruned_loss=0.1238, over 5675745.82 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3573, pruned_loss=0.105, over 5736359.08 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3745, pruned_loss=0.1231, over 5668570.04 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:41:49,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:41:55,051 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:41:59,511 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:42:06,130 INFO [train.py:968] (1/2) Epoch 18, batch 23650, giga_loss[loss=0.3909, simple_loss=0.4328, pruned_loss=0.1745, over 27885.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3809, pruned_loss=0.1282, over 5662833.21 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3574, pruned_loss=0.105, over 5736919.28 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3803, pruned_loss=0.1277, over 5656459.71 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:42:28,692 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:42:32,386 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:42:51,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=800407.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:42:55,211 INFO [train.py:968] (1/2) Epoch 18, batch 23700, giga_loss[loss=0.4792, simple_loss=0.4828, pruned_loss=0.2378, over 26397.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3834, pruned_loss=0.1306, over 5664637.34 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3576, pruned_loss=0.1051, over 5737748.69 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3828, pruned_loss=0.1303, over 5658474.06 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:43:03,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 10:43:13,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2121, 1.2687, 1.1298, 0.9109], device='cuda:1'), covar=tensor([0.0853, 0.0458, 0.0974, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0442, 0.0509, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:43:40,598 INFO [optim.py:369] (1/2) 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 10:43:40,610 INFO [train.py:968] (1/2) Epoch 18, batch 23750, giga_loss[loss=0.3681, simple_loss=0.4147, pruned_loss=0.1608, over 28909.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3848, pruned_loss=0.1326, over 5673744.34 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3579, pruned_loss=0.1057, over 5743421.39 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3851, pruned_loss=0.1328, over 5660729.05 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:43:50,143 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3252, 1.3884, 1.2124, 1.2994], device='cuda:1'), covar=tensor([0.1495, 0.1634, 0.1416, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.1895, 0.1817, 0.1759, 0.1900], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 10:44:28,478 INFO [train.py:968] (1/2) Epoch 18, batch 23800, giga_loss[loss=0.4422, simple_loss=0.4526, pruned_loss=0.2159, over 23374.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3879, pruned_loss=0.1362, over 5642536.01 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3585, pruned_loss=0.1062, over 5736288.14 frames. ], giga_tot_loss[loss=0.3307, simple_loss=0.3882, pruned_loss=0.1366, over 5635997.13 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:45:22,672 INFO [optim.py:369] (1/2) 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 10:45:22,684 INFO [train.py:968] (1/2) Epoch 18, batch 23850, giga_loss[loss=0.3191, simple_loss=0.3862, pruned_loss=0.1261, over 28862.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3927, pruned_loss=0.1405, over 5630702.38 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3586, pruned_loss=0.1063, over 5732272.02 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3937, pruned_loss=0.1415, over 5626799.49 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:46:12,679 INFO [train.py:968] (1/2) Epoch 18, batch 23900, giga_loss[loss=0.3007, simple_loss=0.3661, pruned_loss=0.1177, over 28400.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3922, pruned_loss=0.1414, over 5618240.18 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3585, pruned_loss=0.1067, over 5739762.14 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3946, pruned_loss=0.1434, over 5603816.01 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:46:36,951 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-09 10:46:59,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:47:01,055 INFO [train.py:968] (1/2) Epoch 18, batch 23950, libri_loss[loss=0.2936, simple_loss=0.3656, pruned_loss=0.1108, over 29759.00 frames. ], tot_loss[loss=0.3347, simple_loss=0.3896, pruned_loss=0.1399, over 5634453.88 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3585, pruned_loss=0.107, over 5743059.43 frames. ], giga_tot_loss[loss=0.3384, simple_loss=0.3924, pruned_loss=0.1421, over 5616671.08 frames. ], batch size: 87, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 10:47:01,690 INFO [optim.py:369] (1/2) 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 10:47:46,029 INFO [train.py:968] (1/2) Epoch 18, batch 24000, giga_loss[loss=0.3198, simple_loss=0.3836, pruned_loss=0.128, over 28752.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3886, pruned_loss=0.1392, over 5647968.77 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3585, pruned_loss=0.1072, over 5746949.32 frames. ], giga_tot_loss[loss=0.3377, simple_loss=0.3918, pruned_loss=0.1418, over 5627036.23 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:47:46,030 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 10:47:55,528 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 10:48:01,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:48:20,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5708, 1.7772, 1.4441, 1.5720], device='cuda:1'), covar=tensor([0.2356, 0.2431, 0.2616, 0.2231], device='cuda:1'), in_proj_covar=tensor([0.1436, 0.1048, 0.1276, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 10:48:36,674 INFO [train.py:968] (1/2) Epoch 18, batch 24050, giga_loss[loss=0.2996, simple_loss=0.369, pruned_loss=0.1151, over 28720.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 5634478.14 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.359, pruned_loss=0.1077, over 5738292.91 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3915, pruned_loss=0.1407, over 5622630.64 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:48:39,417 INFO [optim.py:369] (1/2) 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 10:48:41,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:48:50,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2781, 1.4475, 1.4021, 1.1982], device='cuda:1'), covar=tensor([0.2014, 0.2271, 0.1435, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.1894, 0.1820, 0.1759, 0.1903], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 10:48:54,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-09 10:48:59,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=800782.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:49:11,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3596, 1.6505, 1.6052, 1.3830], device='cuda:1'), covar=tensor([0.2567, 0.2222, 0.2172, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.1891, 0.1817, 0.1756, 0.1901], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 10:49:20,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:49:22,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:49:28,157 INFO [train.py:968] (1/2) Epoch 18, batch 24100, giga_loss[loss=0.333, simple_loss=0.4009, pruned_loss=0.1325, over 28640.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3896, pruned_loss=0.1388, over 5627643.57 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3585, pruned_loss=0.1076, over 5740291.68 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.393, pruned_loss=0.1416, over 5614289.11 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:49:32,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6227, 1.6882, 1.2328, 1.2504], device='cuda:1'), covar=tensor([0.0818, 0.0523, 0.1019, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0444, 0.0510, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:49:52,274 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 24150, giga_loss[loss=0.4351, simple_loss=0.4508, pruned_loss=0.2097, over 26418.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3898, pruned_loss=0.1382, over 5619663.19 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3594, pruned_loss=0.1083, over 5730898.40 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3926, pruned_loss=0.1406, over 5614286.84 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:50:20,078 INFO [optim.py:369] (1/2) 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 10:50:20,357 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:50:22,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:50:51,194 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:04,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:06,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:07,700 INFO [train.py:968] (1/2) Epoch 18, batch 24200, giga_loss[loss=0.3056, simple_loss=0.3671, pruned_loss=0.122, over 27968.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.387, pruned_loss=0.1357, over 5611820.94 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3596, pruned_loss=0.1086, over 5723205.18 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.39, pruned_loss=0.1382, over 5610466.22 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:51:19,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=800925.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:51:22,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=800928.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 10:51:33,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:51:49,947 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=800957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 10:51:55,375 INFO [train.py:968] (1/2) Epoch 18, batch 24250, giga_loss[loss=0.3163, simple_loss=0.3946, pruned_loss=0.119, over 28835.00 frames. ], tot_loss[loss=0.326, simple_loss=0.3856, pruned_loss=0.1332, over 5618693.40 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3595, pruned_loss=0.1087, over 5715856.08 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.3885, pruned_loss=0.1355, over 5623081.30 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:51:56,503 INFO [optim.py:369] (1/2) 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 10:52:43,280 INFO [train.py:968] (1/2) Epoch 18, batch 24300, giga_loss[loss=0.2714, simple_loss=0.3475, pruned_loss=0.09769, over 28664.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3804, pruned_loss=0.1289, over 5620037.58 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3591, pruned_loss=0.1087, over 5713286.45 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3839, pruned_loss=0.1315, over 5622596.45 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:53:21,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:53:28,891 INFO [train.py:968] (1/2) Epoch 18, batch 24350, giga_loss[loss=0.3486, simple_loss=0.3955, pruned_loss=0.1509, over 26768.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3772, pruned_loss=0.1264, over 5616096.34 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3589, pruned_loss=0.1087, over 5703905.00 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3805, pruned_loss=0.1288, over 5624161.26 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:53:30,367 INFO [optim.py:369] (1/2) 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 10:53:35,811 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:54:16,417 INFO [train.py:968] (1/2) Epoch 18, batch 24400, giga_loss[loss=0.3062, simple_loss=0.376, pruned_loss=0.1181, over 28811.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3749, pruned_loss=0.125, over 5625926.84 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3585, pruned_loss=0.1085, over 5708888.08 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3783, pruned_loss=0.1275, over 5626005.31 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:54:51,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 10:55:04,350 INFO [train.py:968] (1/2) Epoch 18, batch 24450, giga_loss[loss=0.3199, simple_loss=0.3878, pruned_loss=0.126, over 28501.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3754, pruned_loss=0.1255, over 5627981.89 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3589, pruned_loss=0.1089, over 5706775.69 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3783, pruned_loss=0.1277, over 5627416.12 frames. ], batch size: 336, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 10:55:05,021 INFO [optim.py:369] (1/2) 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 10:55:57,222 INFO [train.py:968] (1/2) Epoch 18, batch 24500, giga_loss[loss=0.3254, simple_loss=0.388, pruned_loss=0.1314, over 28843.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3752, pruned_loss=0.1249, over 5631193.64 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.359, pruned_loss=0.1091, over 5699835.68 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3776, pruned_loss=0.1266, over 5636886.63 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:56:45,171 INFO [train.py:968] (1/2) Epoch 18, batch 24550, giga_loss[loss=0.3644, simple_loss=0.4084, pruned_loss=0.1601, over 26539.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3745, pruned_loss=0.1228, over 5647226.37 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.359, pruned_loss=0.1092, over 5703663.65 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3769, pruned_loss=0.1245, over 5646624.98 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:56:46,621 INFO [optim.py:369] (1/2) 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 10:56:51,165 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:57:36,571 INFO [train.py:968] (1/2) Epoch 18, batch 24600, giga_loss[loss=0.3137, simple_loss=0.3859, pruned_loss=0.1208, over 28650.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3766, pruned_loss=0.1222, over 5657971.49 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3593, pruned_loss=0.1094, over 5707293.59 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3785, pruned_loss=0.1235, over 5653447.48 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:58:06,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1434, 1.3445, 1.1017, 0.9451], device='cuda:1'), covar=tensor([0.0960, 0.0488, 0.1087, 0.1057], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0442, 0.0507, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 10:58:26,356 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 24650, libri_loss[loss=0.3646, simple_loss=0.4068, pruned_loss=0.1611, over 28508.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3765, pruned_loss=0.1225, over 5654420.53 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3592, pruned_loss=0.1097, over 5709507.82 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3784, pruned_loss=0.1236, over 5648025.24 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:58:30,027 INFO [optim.py:369] (1/2) 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 10:59:14,887 INFO [train.py:968] (1/2) Epoch 18, batch 24700, giga_loss[loss=0.2703, simple_loss=0.3402, pruned_loss=0.1002, over 28527.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3759, pruned_loss=0.122, over 5668671.30 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3589, pruned_loss=0.1097, over 5706585.44 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3781, pruned_loss=0.1233, over 5665033.64 frames. ], batch size: 85, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 10:59:28,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 10:59:45,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:00:00,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:00:02,812 INFO [train.py:968] (1/2) Epoch 18, batch 24750, libri_loss[loss=0.286, simple_loss=0.3458, pruned_loss=0.1131, over 29559.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3757, pruned_loss=0.1229, over 5678205.23 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3591, pruned_loss=0.11, over 5709581.88 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3775, pruned_loss=0.1237, over 5672126.55 frames. ], batch size: 74, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:00:04,127 INFO [optim.py:369] (1/2) 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 11:00:26,813 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 11:00:49,238 INFO [train.py:968] (1/2) Epoch 18, batch 24800, giga_loss[loss=0.2793, simple_loss=0.3445, pruned_loss=0.107, over 28894.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3744, pruned_loss=0.1235, over 5674514.63 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3591, pruned_loss=0.1101, over 5710550.97 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3759, pruned_loss=0.1241, over 5668725.04 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:01:32,667 INFO [train.py:968] (1/2) Epoch 18, batch 24850, giga_loss[loss=0.322, simple_loss=0.3798, pruned_loss=0.1321, over 27890.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3731, pruned_loss=0.1219, over 5670803.29 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3591, pruned_loss=0.11, over 5703492.12 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3745, pruned_loss=0.1227, over 5672476.55 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:01:35,202 INFO [optim.py:369] (1/2) 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 11:01:40,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:01:43,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:01:53,877 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:01:55,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:11,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:13,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1225, 2.9578, 2.8181, 1.7291], device='cuda:1'), covar=tensor([0.0861, 0.1043, 0.0983, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.1175, 0.1089, 0.0933, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 11:02:16,974 INFO [train.py:968] (1/2) Epoch 18, batch 24900, giga_loss[loss=0.2898, simple_loss=0.3703, pruned_loss=0.1046, over 28975.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3718, pruned_loss=0.1198, over 5670107.55 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3594, pruned_loss=0.1104, over 5698087.10 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3732, pruned_loss=0.1205, over 5675766.69 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:02:23,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:49,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:02:50,830 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:03:06,550 INFO [train.py:968] (1/2) Epoch 18, batch 24950, giga_loss[loss=0.3997, simple_loss=0.4264, pruned_loss=0.1864, over 26636.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3725, pruned_loss=0.1201, over 5664136.39 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3595, pruned_loss=0.1104, over 5697019.11 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3737, pruned_loss=0.1207, over 5669106.04 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:03:08,653 INFO [optim.py:369] (1/2) 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 11:03:23,650 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:03:23,879 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 11:03:56,036 INFO [train.py:968] (1/2) Epoch 18, batch 25000, giga_loss[loss=0.2897, simple_loss=0.3606, pruned_loss=0.1094, over 28709.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3716, pruned_loss=0.1196, over 5673028.89 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3597, pruned_loss=0.1107, over 5699567.57 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3725, pruned_loss=0.1199, over 5674287.96 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:04:15,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:04:41,916 INFO [train.py:968] (1/2) Epoch 18, batch 25050, libri_loss[loss=0.3605, simple_loss=0.418, pruned_loss=0.1515, over 29543.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3709, pruned_loss=0.1197, over 5680649.56 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3597, pruned_loss=0.1108, over 5704867.74 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 5676077.51 frames. ], batch size: 89, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:04:43,950 INFO [optim.py:369] (1/2) 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 11:05:05,056 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:05:07,066 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 25100, giga_loss[loss=0.2902, simple_loss=0.3472, pruned_loss=0.1166, over 28199.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.371, pruned_loss=0.1208, over 5650888.01 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3605, pruned_loss=0.1113, over 5690178.67 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3713, pruned_loss=0.121, over 5659994.97 frames. ], batch size: 65, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:05:32,190 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:05:46,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=801833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:05:56,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:10,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3521, 1.1756, 3.6997, 3.2535], device='cuda:1'), covar=tensor([0.1494, 0.2704, 0.0450, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0631, 0.0932, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:06:12,516 INFO [train.py:968] (1/2) Epoch 18, batch 25150, giga_loss[loss=0.3116, simple_loss=0.3764, pruned_loss=0.1234, over 28676.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3712, pruned_loss=0.1219, over 5657999.24 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3608, pruned_loss=0.1116, over 5688892.38 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3713, pruned_loss=0.1218, over 5665997.51 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:06:16,116 INFO [optim.py:369] (1/2) 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 11:06:20,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:29,146 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:32,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:06:44,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5077, 1.3656, 4.4386, 3.5553], device='cuda:1'), covar=tensor([0.1597, 0.2677, 0.0417, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0630, 0.0930, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:06:57,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=801909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:07:01,301 INFO [train.py:968] (1/2) Epoch 18, batch 25200, giga_loss[loss=0.2841, simple_loss=0.3519, pruned_loss=0.1081, over 28217.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.1199, over 5659568.17 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3606, pruned_loss=0.1116, over 5692953.94 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1201, over 5661437.36 frames. ], batch size: 77, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:07:23,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:07:46,128 INFO [train.py:968] (1/2) Epoch 18, batch 25250, giga_loss[loss=0.2761, simple_loss=0.3488, pruned_loss=0.1017, over 28995.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3683, pruned_loss=0.1204, over 5657230.89 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.361, pruned_loss=0.112, over 5678469.45 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3685, pruned_loss=0.1203, over 5670350.64 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:07:48,667 INFO [optim.py:369] (1/2) 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 11:07:52,625 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=801968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:02,873 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=801976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:06,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=801979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:12,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2620, 1.6135, 1.3261, 0.8800], device='cuda:1'), covar=tensor([0.2629, 0.2453, 0.2939, 0.2329], device='cuda:1'), in_proj_covar=tensor([0.1438, 0.1045, 0.1278, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:08:32,340 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 25300, libri_loss[loss=0.32, simple_loss=0.3892, pruned_loss=0.1255, over 29273.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3696, pruned_loss=0.1216, over 5652285.15 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3614, pruned_loss=0.1121, over 5683500.68 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1217, over 5657491.83 frames. ], batch size: 94, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:08:40,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:08:42,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5614, 1.9474, 1.5339, 1.3877], device='cuda:1'), covar=tensor([0.2498, 0.2526, 0.2860, 0.2298], device='cuda:1'), in_proj_covar=tensor([0.1435, 0.1044, 0.1276, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:09:09,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:09:14,709 INFO [train.py:968] (1/2) Epoch 18, batch 25350, giga_loss[loss=0.3011, simple_loss=0.3704, pruned_loss=0.1159, over 28587.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3696, pruned_loss=0.1213, over 5659223.33 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3609, pruned_loss=0.1122, over 5689704.71 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3703, pruned_loss=0.1217, over 5656713.78 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:09:18,292 INFO [optim.py:369] (1/2) 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 11:09:18,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-09 11:09:30,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7838, 1.9610, 1.3970, 1.4878], device='cuda:1'), covar=tensor([0.0963, 0.0674, 0.1105, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0445, 0.0510, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:09:36,499 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 11:09:36,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0349, 3.1995, 2.2046, 1.2624], device='cuda:1'), covar=tensor([0.6262, 0.2956, 0.3264, 0.5490], device='cuda:1'), in_proj_covar=tensor([0.1690, 0.1603, 0.1573, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 11:09:40,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4818, 1.7782, 1.4255, 1.3189], device='cuda:1'), covar=tensor([0.2689, 0.2624, 0.3179, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.1433, 0.1041, 0.1274, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:09:59,577 INFO [train.py:968] (1/2) Epoch 18, batch 25400, giga_loss[loss=0.2721, simple_loss=0.3486, pruned_loss=0.09782, over 28782.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3687, pruned_loss=0.1197, over 5668217.48 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3607, pruned_loss=0.1121, over 5694411.40 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3697, pruned_loss=0.1203, over 5661569.76 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:10:45,527 INFO [train.py:968] (1/2) Epoch 18, batch 25450, giga_loss[loss=0.2588, simple_loss=0.3432, pruned_loss=0.0872, over 28857.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3691, pruned_loss=0.1196, over 5666882.05 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.361, pruned_loss=0.1125, over 5697477.91 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3698, pruned_loss=0.1199, over 5657758.03 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:10:47,420 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:10:49,920 INFO [optim.py:369] (1/2) 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 11:10:50,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:15,867 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:18,401 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:20,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:28,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4194, 1.5791, 1.4914, 1.5190], device='cuda:1'), covar=tensor([0.0760, 0.0338, 0.0316, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0116, 0.0115, 0.0216], device='cuda:1'), out_proj_covar=tensor([0.0093, 0.0066, 0.0059, 0.0102], device='cuda:1')
+2023-03-09 11:11:30,956 INFO [train.py:968] (1/2) Epoch 18, batch 25500, giga_loss[loss=0.318, simple_loss=0.3722, pruned_loss=0.1319, over 28495.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1216, over 5665509.63 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3605, pruned_loss=0.1122, over 5700781.74 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3717, pruned_loss=0.1222, over 5654867.59 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:11:37,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:11:48,674 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:12:02,119 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 25550, giga_loss[loss=0.3743, simple_loss=0.4248, pruned_loss=0.1618, over 28770.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3743, pruned_loss=0.1255, over 5658809.27 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3608, pruned_loss=0.1126, over 5703933.64 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3751, pruned_loss=0.1257, over 5647095.40 frames. ], batch size: 242, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:12:22,020 INFO [optim.py:369] (1/2) 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 11:13:06,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3384, 1.6499, 1.5442, 1.3794], device='cuda:1'), covar=tensor([0.1813, 0.1818, 0.2126, 0.2008], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0739, 0.0701, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 11:13:07,031 INFO [train.py:968] (1/2) Epoch 18, batch 25600, giga_loss[loss=0.3689, simple_loss=0.4144, pruned_loss=0.1617, over 28697.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3738, pruned_loss=0.1257, over 5659650.38 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3611, pruned_loss=0.113, over 5695385.73 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3745, pruned_loss=0.1258, over 5656482.89 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:13:09,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:13:38,028 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 25650, giga_loss[loss=0.3095, simple_loss=0.3737, pruned_loss=0.1226, over 28876.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3741, pruned_loss=0.1268, over 5649309.98 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3611, pruned_loss=0.1131, over 5690200.59 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3749, pruned_loss=0.1271, over 5650336.42 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:13:56,657 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:00,466 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:01,437 INFO [optim.py:369] (1/2) 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 11:14:03,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3668, 1.3782, 3.3258, 3.1639], device='cuda:1'), covar=tensor([0.1368, 0.2644, 0.0489, 0.1102], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0631, 0.0935, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:14:05,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3659, 1.7632, 1.3745, 1.2890], device='cuda:1'), covar=tensor([0.2381, 0.2345, 0.2737, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1047, 0.1280, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:14:19,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:21,339 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:24,537 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:14:39,077 INFO [train.py:968] (1/2) Epoch 18, batch 25700, giga_loss[loss=0.2902, simple_loss=0.3604, pruned_loss=0.11, over 28566.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3749, pruned_loss=0.1274, over 5654945.74 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3612, pruned_loss=0.1131, over 5694882.96 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3758, pruned_loss=0.128, over 5650272.62 frames. ], batch size: 336, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:14:45,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:17,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5517, 1.5514, 1.2529, 1.1564], device='cuda:1'), covar=tensor([0.0771, 0.0457, 0.0887, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0445, 0.0511, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:15:20,751 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:24,683 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 25750, giga_loss[loss=0.314, simple_loss=0.3874, pruned_loss=0.1203, over 28876.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3742, pruned_loss=0.1267, over 5639342.68 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3617, pruned_loss=0.1135, over 5677289.57 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3749, pruned_loss=0.1272, over 5649633.63 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:15:29,724 INFO [optim.py:369] (1/2) 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 11:15:45,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=802486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:48,060 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:15:48,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=802489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:16:06,790 INFO [train.py:968] (1/2) Epoch 18, batch 25800, giga_loss[loss=0.2844, simple_loss=0.3604, pruned_loss=0.1042, over 28833.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3739, pruned_loss=0.1251, over 5629037.51 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.362, pruned_loss=0.1139, over 5652131.11 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3744, pruned_loss=0.1254, over 5659552.35 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:16:12,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=802518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:16:49,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3946, 3.1439, 1.4758, 1.5279], device='cuda:1'), covar=tensor([0.0994, 0.0309, 0.0912, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0546, 0.0372, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 11:16:56,389 INFO [train.py:968] (1/2) Epoch 18, batch 25850, giga_loss[loss=0.3121, simple_loss=0.3722, pruned_loss=0.126, over 27888.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1232, over 5629272.75 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3619, pruned_loss=0.1138, over 5653468.46 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3725, pruned_loss=0.1236, over 5651746.85 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:16:59,508 INFO [optim.py:369] (1/2) 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 11:17:39,743 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 25900, giga_loss[loss=0.2612, simple_loss=0.3393, pruned_loss=0.09155, over 29054.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3707, pruned_loss=0.1227, over 5638037.80 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3626, pruned_loss=0.1144, over 5650340.25 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3707, pruned_loss=0.1227, over 5659061.70 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:17:49,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=802623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:18:24,704 INFO [train.py:968] (1/2) Epoch 18, batch 25950, giga_loss[loss=0.2883, simple_loss=0.354, pruned_loss=0.1113, over 28875.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.369, pruned_loss=0.1223, over 5657052.05 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3629, pruned_loss=0.1149, over 5656227.25 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.369, pruned_loss=0.122, over 5668274.13 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:18:30,791 INFO [optim.py:369] (1/2) 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 11:18:37,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4897, 1.6456, 1.6521, 1.4756], device='cuda:1'), covar=tensor([0.1764, 0.1912, 0.2245, 0.2021], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0736, 0.0697, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 11:19:13,587 INFO [train.py:968] (1/2) Epoch 18, batch 26000, giga_loss[loss=0.3199, simple_loss=0.3849, pruned_loss=0.1274, over 28668.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.1219, over 5665447.20 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3636, pruned_loss=0.1154, over 5660716.41 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3688, pruned_loss=0.1214, over 5670472.73 frames. ], batch size: 242, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:19:59,063 INFO [train.py:968] (1/2) Epoch 18, batch 26050, giga_loss[loss=0.2984, simple_loss=0.3762, pruned_loss=0.1103, over 28885.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3736, pruned_loss=0.1239, over 5672596.58 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3639, pruned_loss=0.1155, over 5664268.95 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.373, pruned_loss=0.1235, over 5673586.18 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:20:04,934 INFO [optim.py:369] (1/2) 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 11:20:21,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2903, 1.8220, 1.4000, 0.5419], device='cuda:1'), covar=tensor([0.4197, 0.3013, 0.4005, 0.5686], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1614, 0.1575, 0.1388], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 11:20:46,732 INFO [train.py:968] (1/2) Epoch 18, batch 26100, giga_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.121, over 28519.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.376, pruned_loss=0.1226, over 5675624.99 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3644, pruned_loss=0.1159, over 5668609.31 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3753, pruned_loss=0.1221, over 5672546.28 frames. ], batch size: 85, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:21:00,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=802826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:21:09,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5748, 1.7282, 1.6706, 1.4661], device='cuda:1'), covar=tensor([0.2906, 0.2312, 0.1814, 0.2387], device='cuda:1'), in_proj_covar=tensor([0.1901, 0.1825, 0.1767, 0.1908], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 11:21:30,522 INFO [train.py:968] (1/2) Epoch 18, batch 26150, libri_loss[loss=0.2795, simple_loss=0.3435, pruned_loss=0.1078, over 29558.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3772, pruned_loss=0.1234, over 5670876.94 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3642, pruned_loss=0.1159, over 5661830.30 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3772, pruned_loss=0.1232, over 5675429.38 frames. ], batch size: 78, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:21:34,560 INFO [optim.py:369] (1/2) 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 11:22:16,269 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 26200, giga_loss[loss=0.308, simple_loss=0.3776, pruned_loss=0.1193, over 28901.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.379, pruned_loss=0.1251, over 5674234.74 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3638, pruned_loss=0.1156, over 5666812.09 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3796, pruned_loss=0.1254, over 5673545.37 frames. ], batch size: 112, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:22:27,211 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-09 11:22:58,450 INFO [train.py:968] (1/2) Epoch 18, batch 26250, giga_loss[loss=0.3158, simple_loss=0.3755, pruned_loss=0.128, over 28864.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3788, pruned_loss=0.1255, over 5679943.67 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.363, pruned_loss=0.1151, over 5674138.28 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3805, pruned_loss=0.1264, over 5672958.30 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:23:06,013 INFO [optim.py:369] (1/2) 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 11:23:22,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:23:33,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=802998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:23:44,299 INFO [train.py:968] (1/2) Epoch 18, batch 26300, giga_loss[loss=0.2772, simple_loss=0.3364, pruned_loss=0.109, over 28656.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3768, pruned_loss=0.1249, over 5679362.39 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3623, pruned_loss=0.1146, over 5671620.06 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3794, pruned_loss=0.1264, over 5675107.20 frames. ], batch size: 92, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:24:33,315 INFO [train.py:968] (1/2) Epoch 18, batch 26350, giga_loss[loss=0.3126, simple_loss=0.3731, pruned_loss=0.1261, over 28936.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3755, pruned_loss=0.1245, over 5686253.11 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3623, pruned_loss=0.1146, over 5674005.85 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3777, pruned_loss=0.1259, over 5680897.99 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:24:38,521 INFO [optim.py:369] (1/2) 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 11:24:57,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5719, 2.3302, 1.6073, 0.8539], device='cuda:1'), covar=tensor([0.6450, 0.2973, 0.4621, 0.6796], device='cuda:1'), in_proj_covar=tensor([0.1692, 0.1612, 0.1571, 0.1384], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 11:25:17,756 INFO [train.py:968] (1/2) Epoch 18, batch 26400, giga_loss[loss=0.3929, simple_loss=0.42, pruned_loss=0.1829, over 26653.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3747, pruned_loss=0.1249, over 5690082.94 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5678559.61 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3768, pruned_loss=0.1263, over 5682030.22 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:25:18,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1800, 1.4425, 1.4384, 1.0809], device='cuda:1'), covar=tensor([0.1261, 0.2101, 0.1085, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0699, 0.0922, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 11:25:28,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8892, 2.0297, 1.4660, 1.6636], device='cuda:1'), covar=tensor([0.0959, 0.0741, 0.1080, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0444, 0.0511, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:25:36,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:25:40,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:25:49,562 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:25:52,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:26:07,343 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:26:07,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5764, 1.8810, 1.5143, 1.7012], device='cuda:1'), covar=tensor([0.2405, 0.2432, 0.2795, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.1444, 0.1051, 0.1282, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:26:08,324 INFO [train.py:968] (1/2) Epoch 18, batch 26450, giga_loss[loss=0.3239, simple_loss=0.3764, pruned_loss=0.1357, over 28568.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5685499.78 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5682239.47 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3758, pruned_loss=0.1264, over 5675886.99 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:26:13,736 INFO [optim.py:369] (1/2) 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 11:26:17,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:26:42,344 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 26500, giga_loss[loss=0.35, simple_loss=0.4025, pruned_loss=0.1487, over 27558.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3744, pruned_loss=0.1254, over 5686425.82 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5686567.90 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3761, pruned_loss=0.1266, over 5674692.98 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:27:28,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:27:31,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5266, 1.7631, 1.4558, 1.4230], device='cuda:1'), covar=tensor([0.2382, 0.2453, 0.2742, 0.2186], device='cuda:1'), in_proj_covar=tensor([0.1442, 0.1050, 0.1281, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:27:33,460 INFO [train.py:968] (1/2) Epoch 18, batch 26550, giga_loss[loss=0.3616, simple_loss=0.4, pruned_loss=0.1616, over 26619.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3734, pruned_loss=0.1251, over 5678522.79 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5682273.82 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3754, pruned_loss=0.1266, over 5671975.99 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:27:40,116 INFO [optim.py:369] (1/2) 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 11:27:54,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:28:16,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:28:17,416 INFO [train.py:968] (1/2) Epoch 18, batch 26600, giga_loss[loss=0.2934, simple_loss=0.3681, pruned_loss=0.1093, over 28970.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.371, pruned_loss=0.124, over 5664705.48 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3627, pruned_loss=0.1149, over 5686224.90 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3726, pruned_loss=0.1253, over 5655253.09 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:28:46,513 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:28:46,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-09 11:28:49,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:28:56,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1841, 1.4169, 1.3364, 1.1165], device='cuda:1'), covar=tensor([0.2141, 0.2101, 0.1488, 0.1897], device='cuda:1'), in_proj_covar=tensor([0.1905, 0.1826, 0.1776, 0.1913], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 11:29:01,235 INFO [train.py:968] (1/2) Epoch 18, batch 26650, giga_loss[loss=0.2879, simple_loss=0.3587, pruned_loss=0.1085, over 28621.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3724, pruned_loss=0.1248, over 5660510.88 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3632, pruned_loss=0.1154, over 5680876.50 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3734, pruned_loss=0.1255, over 5656791.55 frames. ], batch size: 78, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:29:08,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8342, 1.4785, 5.0064, 3.5441], device='cuda:1'), covar=tensor([0.1475, 0.2649, 0.0410, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0636, 0.0940, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:29:09,110 INFO [optim.py:369] (1/2) 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 11:29:14,936 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:29:15,798 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 11:29:49,553 INFO [train.py:968] (1/2) Epoch 18, batch 26700, giga_loss[loss=0.3592, simple_loss=0.4069, pruned_loss=0.1558, over 27510.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.125, over 5663206.99 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1154, over 5683135.36 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.375, pruned_loss=0.1256, over 5658138.64 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:30:03,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:30:06,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:30:21,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-09 11:30:30,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 11:30:36,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:30:37,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4150, 1.7888, 1.4115, 1.5105], device='cuda:1'), covar=tensor([0.2569, 0.2611, 0.2895, 0.2310], device='cuda:1'), in_proj_covar=tensor([0.1442, 0.1052, 0.1282, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:30:37,787 INFO [train.py:968] (1/2) Epoch 18, batch 26750, giga_loss[loss=0.2774, simple_loss=0.3456, pruned_loss=0.1046, over 28942.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 5644152.73 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5668778.77 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3752, pruned_loss=0.126, over 5652052.81 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:30:45,128 INFO [optim.py:369] (1/2) 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 11:30:46,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:31:19,318 INFO [train.py:968] (1/2) Epoch 18, batch 26800, giga_loss[loss=0.278, simple_loss=0.3688, pruned_loss=0.09361, over 29076.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3742, pruned_loss=0.1245, over 5666103.64 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1157, over 5675731.49 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1252, over 5665683.83 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:31:34,640 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:32:01,199 INFO [train.py:968] (1/2) Epoch 18, batch 26850, giga_loss[loss=0.318, simple_loss=0.3932, pruned_loss=0.1214, over 29004.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3742, pruned_loss=0.122, over 5670480.64 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5677718.77 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3761, pruned_loss=0.1231, over 5667995.32 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:32:07,902 INFO [optim.py:369] (1/2) 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 11:32:34,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5056, 1.7042, 1.7628, 1.2829], device='cuda:1'), covar=tensor([0.1670, 0.2643, 0.1449, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0702, 0.0926, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 11:32:45,938 INFO [train.py:968] (1/2) Epoch 18, batch 26900, giga_loss[loss=0.5018, simple_loss=0.4909, pruned_loss=0.2563, over 26621.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3769, pruned_loss=0.1225, over 5682540.92 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1151, over 5684311.08 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.379, pruned_loss=0.1238, over 5674670.41 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:32:57,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4480, 1.5715, 1.6598, 1.2525], device='cuda:1'), covar=tensor([0.1829, 0.2551, 0.1528, 0.1797], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0702, 0.0925, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 11:33:00,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:33:28,045 INFO [train.py:968] (1/2) Epoch 18, batch 26950, giga_loss[loss=0.2559, simple_loss=0.3377, pruned_loss=0.08711, over 28995.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.38, pruned_loss=0.1244, over 5685753.74 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3623, pruned_loss=0.1152, over 5689014.62 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3819, pruned_loss=0.1254, over 5675440.97 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:33:36,671 INFO [optim.py:369] (1/2) 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 11:33:51,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:34:11,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4883, 1.7422, 1.4709, 1.3433], device='cuda:1'), covar=tensor([0.2808, 0.2814, 0.3182, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.1442, 0.1051, 0.1280, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 11:34:18,737 INFO [train.py:968] (1/2) Epoch 18, batch 27000, libri_loss[loss=0.3007, simple_loss=0.3658, pruned_loss=0.1178, over 28501.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3822, pruned_loss=0.127, over 5677839.12 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3623, pruned_loss=0.1153, over 5688468.50 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3838, pruned_loss=0.1279, over 5670136.98 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:34:18,738 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 11:34:28,239 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 11:35:17,847 INFO [train.py:968] (1/2) Epoch 18, batch 27050, giga_loss[loss=0.3743, simple_loss=0.4113, pruned_loss=0.1686, over 27468.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3837, pruned_loss=0.1297, over 5660887.26 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.115, over 5692731.42 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3857, pruned_loss=0.1309, over 5650597.76 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:35:23,313 INFO [optim.py:369] (1/2) 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 11:35:25,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:35:28,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:35:42,879 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-09 11:35:54,955 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=803801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:35:56,836 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 27100, giga_loss[loss=0.2865, simple_loss=0.3572, pruned_loss=0.1079, over 28504.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3824, pruned_loss=0.1287, over 5657127.12 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5685073.87 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.384, pruned_loss=0.1296, over 5655739.24 frames. ], batch size: 78, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:36:23,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:25,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:39,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:49,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=803860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:36:50,951 INFO [train.py:968] (1/2) Epoch 18, batch 27150, libri_loss[loss=0.2309, simple_loss=0.3015, pruned_loss=0.08018, over 29358.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3808, pruned_loss=0.127, over 5651951.80 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3622, pruned_loss=0.1152, over 5688445.54 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3829, pruned_loss=0.1282, over 5646389.84 frames. ], batch size: 71, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:36:58,708 INFO [optim.py:369] (1/2) 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 11:37:07,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-09 11:37:29,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0574, 1.2562, 3.3705, 2.8705], device='cuda:1'), covar=tensor([0.1743, 0.2732, 0.0555, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0736, 0.0634, 0.0939, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:37:29,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=803904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:37:29,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1007, 1.4558, 1.4818, 1.2983], device='cuda:1'), covar=tensor([0.2090, 0.1806, 0.2485, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.0456, 0.0735, 0.0697, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 11:37:37,112 INFO [train.py:968] (1/2) Epoch 18, batch 27200, giga_loss[loss=0.2846, simple_loss=0.3762, pruned_loss=0.09652, over 28743.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3806, pruned_loss=0.125, over 5660897.52 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3618, pruned_loss=0.1152, over 5691947.68 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3827, pruned_loss=0.1261, over 5653102.45 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:37:59,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3878, 1.7429, 1.3895, 1.5576], device='cuda:1'), covar=tensor([0.0722, 0.0299, 0.0323, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 11:38:23,086 INFO [train.py:968] (1/2) Epoch 18, batch 27250, giga_loss[loss=0.4195, simple_loss=0.4384, pruned_loss=0.2003, over 26553.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3809, pruned_loss=0.1254, over 5669495.95 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3616, pruned_loss=0.115, over 5700353.81 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.3836, pruned_loss=0.1268, over 5654714.56 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:38:31,830 INFO [optim.py:369] (1/2) 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 11:38:37,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 11:38:42,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-09 11:38:50,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=803989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:38:53,149 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=803992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:10,519 INFO [train.py:968] (1/2) Epoch 18, batch 27300, giga_loss[loss=0.4055, simple_loss=0.4451, pruned_loss=0.1829, over 28288.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.3811, pruned_loss=0.1257, over 5666317.40 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3617, pruned_loss=0.1153, over 5694217.33 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3834, pruned_loss=0.1267, over 5658756.82 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:39:21,525 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:39,725 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:42,321 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:39:45,233 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 27350, giga_loss[loss=0.2695, simple_loss=0.3425, pruned_loss=0.09826, over 28823.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3799, pruned_loss=0.1252, over 5675813.97 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3623, pruned_loss=0.1156, over 5698354.00 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3816, pruned_loss=0.126, over 5665634.81 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:40:04,419 INFO [optim.py:369] (1/2) 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 11:40:14,607 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:40:45,983 INFO [train.py:968] (1/2) Epoch 18, batch 27400, giga_loss[loss=0.2903, simple_loss=0.3584, pruned_loss=0.1111, over 28918.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3772, pruned_loss=0.125, over 5650549.03 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1154, over 5693840.04 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.379, pruned_loss=0.126, over 5645011.60 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:41:35,924 INFO [train.py:968] (1/2) Epoch 18, batch 27450, giga_loss[loss=0.2704, simple_loss=0.3421, pruned_loss=0.09939, over 29065.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3761, pruned_loss=0.1253, over 5643440.63 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3624, pruned_loss=0.1154, over 5693958.84 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3775, pruned_loss=0.1261, over 5638850.49 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:41:42,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 11:41:44,584 INFO [optim.py:369] (1/2) 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 11:41:52,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:42:15,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5978, 1.5723, 1.8192, 1.4254], device='cuda:1'), covar=tensor([0.1363, 0.2043, 0.1106, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0703, 0.0925, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 11:42:26,023 INFO [train.py:968] (1/2) Epoch 18, batch 27500, giga_loss[loss=0.2921, simple_loss=0.3584, pruned_loss=0.1128, over 28864.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1242, over 5654476.92 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3623, pruned_loss=0.1153, over 5695066.04 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.375, pruned_loss=0.125, over 5649729.50 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:42:49,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3918, 1.5975, 1.6707, 1.2285], device='cuda:1'), covar=tensor([0.1663, 0.2481, 0.1382, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0703, 0.0926, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 11:42:56,614 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-09 11:43:08,316 INFO [train.py:968] (1/2) Epoch 18, batch 27550, giga_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1226, over 27922.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5647440.56 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.362, pruned_loss=0.1152, over 5690679.34 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3749, pruned_loss=0.1256, over 5645951.07 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:43:16,369 INFO [optim.py:369] (1/2) 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 11:43:27,657 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804282.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 11:43:49,615 INFO [train.py:968] (1/2) Epoch 18, batch 27600, giga_loss[loss=0.2751, simple_loss=0.3555, pruned_loss=0.09734, over 28730.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5652848.25 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3634, pruned_loss=0.1163, over 5693691.87 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3718, pruned_loss=0.1231, over 5647473.31 frames. ], batch size: 242, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:43:55,514 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:43:58,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:44:23,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:44:31,071 INFO [train.py:968] (1/2) Epoch 18, batch 27650, giga_loss[loss=0.2769, simple_loss=0.356, pruned_loss=0.09891, over 28635.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1184, over 5665668.01 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3632, pruned_loss=0.1161, over 5700203.87 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1187, over 5654079.76 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:44:40,214 INFO [optim.py:369] (1/2) 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 11:44:50,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2725, 1.3375, 1.2335, 1.5136], device='cuda:1'), covar=tensor([0.0769, 0.0390, 0.0352, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 11:45:17,480 INFO [train.py:968] (1/2) Epoch 18, batch 27700, giga_loss[loss=0.3785, simple_loss=0.4109, pruned_loss=0.1731, over 26580.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1161, over 5664605.66 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5702984.83 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3654, pruned_loss=0.1163, over 5652294.16 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:45:24,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:45:29,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:45:58,246 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:46:06,836 INFO [train.py:968] (1/2) Epoch 18, batch 27750, giga_loss[loss=0.2468, simple_loss=0.3252, pruned_loss=0.08421, over 29009.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3641, pruned_loss=0.1158, over 5654587.67 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3628, pruned_loss=0.1158, over 5708135.63 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3649, pruned_loss=0.1163, over 5639335.65 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:46:16,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8865, 1.9345, 1.8388, 1.6859], device='cuda:1'), covar=tensor([0.1747, 0.2532, 0.2168, 0.2339], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0745, 0.0703, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 11:46:17,678 INFO [optim.py:369] (1/2) 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 11:46:56,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=804509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:46:57,847 INFO [train.py:968] (1/2) Epoch 18, batch 27800, giga_loss[loss=0.2678, simple_loss=0.3446, pruned_loss=0.09553, over 29097.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1141, over 5668159.11 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.363, pruned_loss=0.1159, over 5702231.07 frames. ], giga_tot_loss[loss=0.295, simple_loss=0.3611, pruned_loss=0.1145, over 5660232.23 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:47:45,494 INFO [train.py:968] (1/2) Epoch 18, batch 27850, giga_loss[loss=0.3141, simple_loss=0.3792, pruned_loss=0.1245, over 28796.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3624, pruned_loss=0.1158, over 5667439.88 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3628, pruned_loss=0.1158, over 5705187.78 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3629, pruned_loss=0.1161, over 5657830.35 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:47:46,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:47:49,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:47:50,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-09 11:47:52,720 INFO [optim.py:369] (1/2) 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 11:48:13,746 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:48:28,778 INFO [train.py:968] (1/2) Epoch 18, batch 27900, libri_loss[loss=0.2426, simple_loss=0.3162, pruned_loss=0.08452, over 27256.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3663, pruned_loss=0.1182, over 5643776.92 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3634, pruned_loss=0.1163, over 5689327.62 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3662, pruned_loss=0.1181, over 5648084.58 frames. ], batch size: 60, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:49:12,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804657.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 11:49:16,085 INFO [train.py:968] (1/2) Epoch 18, batch 27950, giga_loss[loss=0.2695, simple_loss=0.3463, pruned_loss=0.09633, over 28694.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3685, pruned_loss=0.1196, over 5643203.22 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1161, over 5691482.97 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3686, pruned_loss=0.1196, over 5644527.29 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:49:25,772 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 28000, giga_loss[loss=0.2986, simple_loss=0.371, pruned_loss=0.1131, over 29003.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3693, pruned_loss=0.1201, over 5649656.42 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3637, pruned_loss=0.1164, over 5693508.79 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3691, pruned_loss=0.12, over 5648227.19 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:50:25,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2282, 3.0539, 2.8981, 1.5003], device='cuda:1'), covar=tensor([0.1062, 0.1262, 0.1233, 0.2214], device='cuda:1'), in_proj_covar=tensor([0.1191, 0.1105, 0.0952, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 11:50:44,262 INFO [train.py:968] (1/2) Epoch 18, batch 28050, giga_loss[loss=0.2883, simple_loss=0.3594, pruned_loss=0.1086, over 28941.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3702, pruned_loss=0.1213, over 5655347.89 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5696896.06 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3695, pruned_loss=0.1208, over 5649323.45 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:50:50,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5874, 4.6852, 1.8649, 1.7319], device='cuda:1'), covar=tensor([0.0962, 0.0283, 0.0805, 0.1294], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0543, 0.0371, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 11:50:52,370 INFO [optim.py:369] (1/2) 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 11:51:17,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:51:18,600 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804800.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 11:51:21,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804803.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 11:51:30,018 INFO [train.py:968] (1/2) Epoch 18, batch 28100, giga_loss[loss=0.288, simple_loss=0.3537, pruned_loss=0.1112, over 28695.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1227, over 5651715.76 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.117, over 5690747.32 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1223, over 5652760.69 frames. ], batch size: 92, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 11:51:42,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:51:47,165 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804832.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 11:52:09,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3665, 3.6940, 1.5740, 1.5748], device='cuda:1'), covar=tensor([0.1014, 0.0325, 0.0879, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0544, 0.0371, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 11:52:14,811 INFO [train.py:968] (1/2) Epoch 18, batch 28150, giga_loss[loss=0.3117, simple_loss=0.3806, pruned_loss=0.1214, over 28974.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3746, pruned_loss=0.1236, over 5656506.28 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.365, pruned_loss=0.1171, over 5689902.46 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3739, pruned_loss=0.1233, over 5657576.41 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:52:26,014 INFO [optim.py:369] (1/2) 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 11:52:38,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=804884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:53:05,437 INFO [train.py:968] (1/2) Epoch 18, batch 28200, giga_loss[loss=0.3091, simple_loss=0.375, pruned_loss=0.1217, over 28802.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1254, over 5650001.14 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3649, pruned_loss=0.1171, over 5693341.34 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3756, pruned_loss=0.1253, over 5647325.90 frames. ], batch size: 284, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:53:14,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3438, 2.6133, 2.3773, 1.8625], device='cuda:1'), covar=tensor([0.2686, 0.2050, 0.2297, 0.2702], device='cuda:1'), in_proj_covar=tensor([0.1904, 0.1834, 0.1774, 0.1912], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 11:53:32,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:53:35,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7713, 4.5525, 4.2781, 2.2428], device='cuda:1'), covar=tensor([0.0614, 0.0801, 0.0913, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.1104, 0.0950, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 11:53:35,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:53:50,950 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-09 11:53:54,266 INFO [train.py:968] (1/2) Epoch 18, batch 28250, giga_loss[loss=0.305, simple_loss=0.3871, pruned_loss=0.1114, over 28830.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3753, pruned_loss=0.1252, over 5649189.96 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5695488.68 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3755, pruned_loss=0.1254, over 5644611.01 frames. ], batch size: 285, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:54:02,893 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=804969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:54:06,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=804972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:54:07,138 INFO [optim.py:369] (1/2) 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 11:54:07,631 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=804973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:54:22,128 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-09 11:54:26,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5625, 1.6747, 1.8233, 1.3735], device='cuda:1'), covar=tensor([0.1996, 0.2882, 0.1652, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0704, 0.0927, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 11:54:34,736 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 28300, giga_loss[loss=0.3064, simple_loss=0.3836, pruned_loss=0.1146, over 28695.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3751, pruned_loss=0.1232, over 5649842.75 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3646, pruned_loss=0.1171, over 5689290.57 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3755, pruned_loss=0.1233, over 5650501.08 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:55:00,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=805027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:55:03,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=805030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 11:55:32,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 28350, libri_loss[loss=0.275, simple_loss=0.3398, pruned_loss=0.1051, over 29590.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3747, pruned_loss=0.123, over 5666045.14 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3644, pruned_loss=0.117, over 5695332.90 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3757, pruned_loss=0.1234, over 5659938.86 frames. ], batch size: 77, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 11:55:45,100 INFO [optim.py:369] (1/2) 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 11:55:51,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2508, 1.8186, 1.4317, 0.4771], device='cuda:1'), covar=tensor([0.4133, 0.2634, 0.3927, 0.5410], device='cuda:1'), in_proj_covar=tensor([0.1698, 0.1605, 0.1573, 0.1383], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 11:55:51,331 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 11:56:24,131 INFO [train.py:968] (1/2) Epoch 18, batch 28400, giga_loss[loss=0.3134, simple_loss=0.3815, pruned_loss=0.1226, over 28968.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3738, pruned_loss=0.1232, over 5667536.71 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3644, pruned_loss=0.117, over 5696353.23 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3746, pruned_loss=0.1235, over 5661795.50 frames. ], batch size: 213, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:56:33,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2486, 1.1536, 3.7423, 3.3163], device='cuda:1'), covar=tensor([0.1653, 0.2926, 0.0479, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0634, 0.0932, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 11:57:17,197 INFO [train.py:968] (1/2) Epoch 18, batch 28450, giga_loss[loss=0.2791, simple_loss=0.3412, pruned_loss=0.1085, over 28787.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3732, pruned_loss=0.1234, over 5676399.98 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3643, pruned_loss=0.1172, over 5702430.93 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3741, pruned_loss=0.1237, over 5665787.84 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:57:32,412 INFO [optim.py:369] (1/2) 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 11:58:09,606 INFO [train.py:968] (1/2) Epoch 18, batch 28500, giga_loss[loss=0.3009, simple_loss=0.3647, pruned_loss=0.1186, over 28479.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3713, pruned_loss=0.1221, over 5682031.91 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3645, pruned_loss=0.1173, over 5705153.90 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.372, pruned_loss=0.1223, over 5670749.95 frames. ], batch size: 65, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:58:52,196 INFO [train.py:968] (1/2) Epoch 18, batch 28550, giga_loss[loss=0.2909, simple_loss=0.3692, pruned_loss=0.1063, over 29002.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1228, over 5683871.80 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3641, pruned_loss=0.117, over 5710677.95 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3727, pruned_loss=0.1234, over 5669046.46 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:59:02,644 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 28600, giga_loss[loss=0.3069, simple_loss=0.3695, pruned_loss=0.1221, over 28768.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3723, pruned_loss=0.1244, over 5666573.64 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3636, pruned_loss=0.1168, over 5714846.06 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3738, pruned_loss=0.1252, over 5650174.73 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 11:59:51,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1990, 2.5181, 2.1167, 2.3301], device='cuda:1'), covar=tensor([0.0493, 0.0213, 0.0229, 0.0565], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 11:59:53,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-09 12:00:24,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 12:00:25,819 INFO [train.py:968] (1/2) Epoch 18, batch 28650, giga_loss[loss=0.3414, simple_loss=0.3938, pruned_loss=0.1444, over 28020.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3736, pruned_loss=0.1254, over 5657911.74 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3641, pruned_loss=0.117, over 5708201.44 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3746, pruned_loss=0.126, over 5649574.37 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:00:37,486 INFO [optim.py:369] (1/2) 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 12:01:11,016 INFO [train.py:968] (1/2) Epoch 18, batch 28700, giga_loss[loss=0.3601, simple_loss=0.4122, pruned_loss=0.154, over 28720.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3757, pruned_loss=0.1275, over 5652620.37 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.364, pruned_loss=0.1169, over 5704584.72 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3768, pruned_loss=0.1283, over 5647629.99 frames. ], batch size: 242, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:01:58,702 INFO [train.py:968] (1/2) Epoch 18, batch 28750, giga_loss[loss=0.3454, simple_loss=0.3803, pruned_loss=0.1553, over 23412.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3765, pruned_loss=0.1281, over 5648006.72 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3636, pruned_loss=0.1167, over 5707270.07 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3779, pruned_loss=0.1291, over 5640783.78 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:02:11,234 INFO [optim.py:369] (1/2) 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 12:02:42,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8696, 3.6872, 3.5097, 1.6394], device='cuda:1'), covar=tensor([0.0737, 0.0856, 0.0850, 0.2215], device='cuda:1'), in_proj_covar=tensor([0.1196, 0.1112, 0.0960, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0011], device='cuda:1')
+2023-03-09 12:02:44,109 INFO [train.py:968] (1/2) Epoch 18, batch 28800, giga_loss[loss=0.3207, simple_loss=0.381, pruned_loss=0.1302, over 27926.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3758, pruned_loss=0.1281, over 5648282.92 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3637, pruned_loss=0.1169, over 5704669.26 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3773, pruned_loss=0.1291, over 5643058.37 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:03:14,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 12:03:28,739 INFO [train.py:968] (1/2) Epoch 18, batch 28850, giga_loss[loss=0.3851, simple_loss=0.4166, pruned_loss=0.1768, over 27858.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3768, pruned_loss=0.1294, over 5657900.68 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3641, pruned_loss=0.1175, over 5710204.66 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.378, pruned_loss=0.1301, over 5647119.88 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:03:40,954 INFO [optim.py:369] (1/2) 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 12:03:54,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 12:04:17,422 INFO [train.py:968] (1/2) Epoch 18, batch 28900, giga_loss[loss=0.274, simple_loss=0.3539, pruned_loss=0.09707, over 28965.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3764, pruned_loss=0.1286, over 5648634.40 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3642, pruned_loss=0.1175, over 5708351.66 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3774, pruned_loss=0.1292, over 5641483.89 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:04:21,334 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 12:05:04,371 INFO [train.py:968] (1/2) Epoch 18, batch 28950, giga_loss[loss=0.2766, simple_loss=0.352, pruned_loss=0.1006, over 28967.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3763, pruned_loss=0.1276, over 5649418.36 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3649, pruned_loss=0.1178, over 5706134.38 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3767, pruned_loss=0.128, over 5644117.19 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:05:14,220 INFO [optim.py:369] (1/2) 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 12:05:48,912 INFO [train.py:968] (1/2) Epoch 18, batch 29000, giga_loss[loss=0.3017, simple_loss=0.3671, pruned_loss=0.1182, over 28907.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3784, pruned_loss=0.1289, over 5652801.54 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 5703216.59 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3784, pruned_loss=0.129, over 5649452.99 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:06:35,311 INFO [train.py:968] (1/2) Epoch 18, batch 29050, giga_loss[loss=0.4009, simple_loss=0.4325, pruned_loss=0.1846, over 27565.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3779, pruned_loss=0.1283, over 5664340.01 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3659, pruned_loss=0.1185, over 5704460.08 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3779, pruned_loss=0.1284, over 5660197.69 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:06:42,473 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-09 12:06:46,055 INFO [optim.py:369] (1/2) 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 12:07:21,757 INFO [train.py:968] (1/2) Epoch 18, batch 29100, giga_loss[loss=0.3033, simple_loss=0.3773, pruned_loss=0.1146, over 28904.00 frames. ], tot_loss[loss=0.3175, simple_loss=0.3781, pruned_loss=0.1284, over 5663565.56 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 5696629.46 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3783, pruned_loss=0.1287, over 5666439.49 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:08:01,067 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2091, 3.9998, 3.7927, 2.1203], device='cuda:1'), covar=tensor([0.0579, 0.0809, 0.0832, 0.2396], device='cuda:1'), in_proj_covar=tensor([0.1193, 0.1109, 0.0952, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 12:08:04,762 INFO [train.py:968] (1/2) Epoch 18, batch 29150, giga_loss[loss=0.2752, simple_loss=0.3576, pruned_loss=0.0964, over 28901.00 frames. ], tot_loss[loss=0.318, simple_loss=0.379, pruned_loss=0.1285, over 5670239.28 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3659, pruned_loss=0.1185, over 5707158.26 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3797, pruned_loss=0.1292, over 5660923.68 frames. ], batch size: 66, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:08:18,174 INFO [optim.py:369] (1/2) 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 12:08:19,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4366, 1.5279, 1.1729, 1.0980], device='cuda:1'), covar=tensor([0.0771, 0.0424, 0.0879, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0447, 0.0511, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:08:28,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2879, 1.9195, 1.6292, 1.5106], device='cuda:1'), covar=tensor([0.0725, 0.0306, 0.0295, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 12:08:42,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2484, 1.5623, 1.5584, 1.1299], device='cuda:1'), covar=tensor([0.1663, 0.2520, 0.1395, 0.1644], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0701, 0.0924, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 12:08:45,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3387, 1.6934, 1.6304, 1.1862], device='cuda:1'), covar=tensor([0.1615, 0.2783, 0.1478, 0.1785], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0702, 0.0924, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 12:08:53,877 INFO [train.py:968] (1/2) Epoch 18, batch 29200, libri_loss[loss=0.3162, simple_loss=0.3723, pruned_loss=0.13, over 29561.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3789, pruned_loss=0.1273, over 5673548.69 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3656, pruned_loss=0.1184, over 5710790.39 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.38, pruned_loss=0.1281, over 5662061.87 frames. ], batch size: 79, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:08:54,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6670, 1.7682, 1.3000, 1.3514], device='cuda:1'), covar=tensor([0.0888, 0.0592, 0.1002, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0447, 0.0512, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:09:00,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3599, 3.5111, 1.5927, 1.5197], device='cuda:1'), covar=tensor([0.0915, 0.0339, 0.0845, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0545, 0.0371, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 12:09:36,991 INFO [train.py:968] (1/2) Epoch 18, batch 29250, giga_loss[loss=0.3105, simple_loss=0.3654, pruned_loss=0.1278, over 28700.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3758, pruned_loss=0.1243, over 5667677.20 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3655, pruned_loss=0.1183, over 5705662.65 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3772, pruned_loss=0.1253, over 5661345.15 frames. ], batch size: 92, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:09:49,194 INFO [optim.py:369] (1/2) 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 12:10:03,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6862, 1.7325, 1.8001, 1.4936], device='cuda:1'), covar=tensor([0.1730, 0.2349, 0.2239, 0.2360], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0747, 0.0705, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 12:10:19,758 INFO [train.py:968] (1/2) Epoch 18, batch 29300, giga_loss[loss=0.3243, simple_loss=0.389, pruned_loss=0.1298, over 28926.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5668553.30 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1183, over 5710792.61 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3752, pruned_loss=0.1244, over 5657609.00 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:11:03,679 INFO [train.py:968] (1/2) Epoch 18, batch 29350, giga_loss[loss=0.3079, simple_loss=0.374, pruned_loss=0.1209, over 28614.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.376, pruned_loss=0.1252, over 5665721.19 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.366, pruned_loss=0.1188, over 5713397.19 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3769, pruned_loss=0.1257, over 5653713.05 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:11:18,058 INFO [optim.py:369] (1/2) 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 12:11:21,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3201, 1.3841, 3.3899, 3.1017], device='cuda:1'), covar=tensor([0.1441, 0.2534, 0.0481, 0.1469], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0635, 0.0940, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:11:51,251 INFO [train.py:968] (1/2) Epoch 18, batch 29400, giga_loss[loss=0.3611, simple_loss=0.4072, pruned_loss=0.1575, over 29000.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3775, pruned_loss=0.1263, over 5663937.39 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3661, pruned_loss=0.1188, over 5715621.87 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3785, pruned_loss=0.127, over 5650446.87 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:12:39,498 INFO [train.py:968] (1/2) Epoch 18, batch 29450, giga_loss[loss=0.2979, simple_loss=0.372, pruned_loss=0.1119, over 28862.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3775, pruned_loss=0.1268, over 5661642.81 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3664, pruned_loss=0.1189, over 5709947.94 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3782, pruned_loss=0.1273, over 5655438.66 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:12:52,429 INFO [optim.py:369] (1/2) 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 12:13:23,409 INFO [train.py:968] (1/2) Epoch 18, batch 29500, giga_loss[loss=0.3293, simple_loss=0.3872, pruned_loss=0.1357, over 28994.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1268, over 5661045.48 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3655, pruned_loss=0.1184, over 5706748.88 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3781, pruned_loss=0.1279, over 5657485.30 frames. ], batch size: 106, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:13:53,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7861, 2.3734, 1.5406, 1.0491], device='cuda:1'), covar=tensor([0.6337, 0.3316, 0.3321, 0.5493], device='cuda:1'), in_proj_covar=tensor([0.1705, 0.1616, 0.1578, 0.1384], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 12:14:03,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0289, 1.3381, 1.0960, 0.2118], device='cuda:1'), covar=tensor([0.3261, 0.2892, 0.4151, 0.5541], device='cuda:1'), in_proj_covar=tensor([0.1705, 0.1617, 0.1578, 0.1384], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 12:14:04,413 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-09 12:14:08,228 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 29550, giga_loss[loss=0.2862, simple_loss=0.3615, pruned_loss=0.1054, over 28369.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.377, pruned_loss=0.1274, over 5657690.12 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3654, pruned_loss=0.1182, over 5707924.07 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3787, pruned_loss=0.1287, over 5652530.46 frames. ], batch size: 65, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:14:21,673 INFO [optim.py:369] (1/2) 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 12:14:54,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1848, 2.2929, 1.3196, 1.3277], device='cuda:1'), covar=tensor([0.0945, 0.0473, 0.0855, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0546, 0.0371, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 12:14:57,912 INFO [train.py:968] (1/2) Epoch 18, batch 29600, giga_loss[loss=0.2735, simple_loss=0.346, pruned_loss=0.1005, over 28854.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.377, pruned_loss=0.1272, over 5649719.32 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3654, pruned_loss=0.1183, over 5700628.76 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3784, pruned_loss=0.1282, over 5652073.06 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:15:44,764 INFO [train.py:968] (1/2) Epoch 18, batch 29650, giga_loss[loss=0.2784, simple_loss=0.3594, pruned_loss=0.09868, over 29034.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.377, pruned_loss=0.127, over 5657917.50 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1185, over 5703038.68 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.378, pruned_loss=0.1278, over 5657283.29 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:15:56,552 INFO [optim.py:369] (1/2) 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 12:16:20,218 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:16:28,218 INFO [train.py:968] (1/2) Epoch 18, batch 29700, giga_loss[loss=0.2654, simple_loss=0.3447, pruned_loss=0.09305, over 28716.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3754, pruned_loss=0.1249, over 5668179.38 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1184, over 5700116.69 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3765, pruned_loss=0.1258, over 5669602.15 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:17:15,248 INFO [train.py:968] (1/2) Epoch 18, batch 29750, giga_loss[loss=0.297, simple_loss=0.3703, pruned_loss=0.1119, over 29037.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3754, pruned_loss=0.1251, over 5654614.30 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1183, over 5703331.15 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3765, pruned_loss=0.126, over 5652277.08 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:17:29,186 INFO [optim.py:369] (1/2) 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 12:18:02,420 INFO [train.py:968] (1/2) Epoch 18, batch 29800, giga_loss[loss=0.2943, simple_loss=0.3608, pruned_loss=0.1139, over 28779.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3738, pruned_loss=0.1239, over 5660739.41 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1178, over 5708180.15 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3753, pruned_loss=0.1251, over 5653539.49 frames. ], batch size: 99, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:18:19,224 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:18:45,518 INFO [train.py:968] (1/2) Epoch 18, batch 29850, giga_loss[loss=0.2798, simple_loss=0.3484, pruned_loss=0.1056, over 28357.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.373, pruned_loss=0.1238, over 5666160.53 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1176, over 5706584.16 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3746, pruned_loss=0.1251, over 5660686.00 frames. ], batch size: 65, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:18:58,736 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 29900, giga_loss[loss=0.2815, simple_loss=0.3478, pruned_loss=0.1077, over 28918.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1238, over 5661671.95 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3653, pruned_loss=0.1178, over 5709639.52 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3732, pruned_loss=0.1248, over 5654345.78 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:19:51,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=806636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:20:18,840 INFO [train.py:968] (1/2) Epoch 18, batch 29950, giga_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1199, over 28892.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3685, pruned_loss=0.1219, over 5663985.98 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.118, over 5703163.58 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.369, pruned_loss=0.1226, over 5663914.93 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:20:33,251 INFO [optim.py:369] (1/2) 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 12:21:01,869 INFO [train.py:968] (1/2) Epoch 18, batch 30000, giga_loss[loss=0.3538, simple_loss=0.4097, pruned_loss=0.149, over 28690.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3661, pruned_loss=0.1206, over 5681017.88 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3655, pruned_loss=0.1178, over 5703578.54 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3666, pruned_loss=0.1214, over 5680186.03 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:21:01,869 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 12:21:10,553 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 12:21:35,839 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806737.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:21:58,819 INFO [train.py:968] (1/2) Epoch 18, batch 30050, libri_loss[loss=0.3515, simple_loss=0.3888, pruned_loss=0.1571, over 29613.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3659, pruned_loss=0.1211, over 5689108.22 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3657, pruned_loss=0.118, over 5705712.53 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3663, pruned_loss=0.1215, over 5686229.93 frames. ], batch size: 69, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:22:13,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=806776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:16,010 INFO [optim.py:369] (1/2) 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 12:22:18,536 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=806779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:22,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=806782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:36,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6199, 1.7363, 1.2929, 1.3365], device='cuda:1'), covar=tensor([0.0913, 0.0542, 0.0945, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0449, 0.0514, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:22:46,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=806811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:22:46,844 INFO [train.py:968] (1/2) Epoch 18, batch 30100, giga_loss[loss=0.2686, simple_loss=0.3468, pruned_loss=0.0952, over 28674.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3653, pruned_loss=0.1192, over 5685417.31 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3654, pruned_loss=0.1179, over 5707554.45 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3658, pruned_loss=0.1197, over 5681211.13 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:23:02,947 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=806828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:23:29,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4998, 1.9044, 1.6478, 1.6452], device='cuda:1'), covar=tensor([0.1611, 0.1529, 0.1925, 0.1602], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0745, 0.0706, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 12:23:37,024 INFO [train.py:968] (1/2) Epoch 18, batch 30150, giga_loss[loss=0.2595, simple_loss=0.3446, pruned_loss=0.08723, over 28766.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3632, pruned_loss=0.1154, over 5687777.21 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3649, pruned_loss=0.1175, over 5713365.72 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.364, pruned_loss=0.1161, over 5678801.96 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:23:53,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2333, 1.0834, 3.6182, 3.2036], device='cuda:1'), covar=tensor([0.1701, 0.3051, 0.0488, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0637, 0.0945, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:23:56,008 INFO [optim.py:369] (1/2) 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 12:24:22,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=806905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:24:27,239 INFO [train.py:968] (1/2) Epoch 18, batch 30200, libri_loss[loss=0.3296, simple_loss=0.385, pruned_loss=0.1371, over 29180.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3612, pruned_loss=0.1134, over 5674055.47 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3647, pruned_loss=0.1175, over 5716369.01 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3621, pruned_loss=0.1139, over 5663224.31 frames. ], batch size: 97, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:24:34,954 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=806919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:24:38,452 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=806922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:25:02,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2857, 2.2446, 1.3280, 1.4887], device='cuda:1'), covar=tensor([0.0862, 0.0419, 0.0867, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0547, 0.0372, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 12:25:08,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1939, 1.1998, 3.9768, 3.2024], device='cuda:1'), covar=tensor([0.1802, 0.2891, 0.0455, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0638, 0.0944, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:25:08,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=806951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:25:18,003 INFO [train.py:968] (1/2) Epoch 18, batch 30250, giga_loss[loss=0.2581, simple_loss=0.3439, pruned_loss=0.08616, over 28070.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3571, pruned_loss=0.1095, over 5665765.65 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1173, over 5720196.62 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3581, pruned_loss=0.11, over 5652914.46 frames. ], batch size: 77, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:25:31,484 INFO [optim.py:369] (1/2) 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 12:26:03,599 INFO [train.py:968] (1/2) Epoch 18, batch 30300, giga_loss[loss=0.2679, simple_loss=0.3473, pruned_loss=0.09431, over 28710.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3539, pruned_loss=0.1062, over 5669184.08 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3638, pruned_loss=0.117, over 5725150.20 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3549, pruned_loss=0.1065, over 5653098.05 frames. ], batch size: 262, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:26:17,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0187, 1.0774, 3.3755, 2.8461], device='cuda:1'), covar=tensor([0.1756, 0.2919, 0.0531, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0636, 0.0942, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:26:39,576 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:26:45,298 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 30350, giga_loss[loss=0.2904, simple_loss=0.3686, pruned_loss=0.1061, over 28895.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.1031, over 5659253.81 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3635, pruned_loss=0.1169, over 5729024.98 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3528, pruned_loss=0.1032, over 5641882.15 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:27:10,167 INFO [optim.py:369] (1/2) 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 12:27:13,717 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:27:39,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1083, 1.4637, 1.3708, 1.1123], device='cuda:1'), covar=tensor([0.2126, 0.1635, 0.1253, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.1887, 0.1809, 0.1737, 0.1884], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 12:27:42,718 INFO [train.py:968] (1/2) Epoch 18, batch 30400, giga_loss[loss=0.3909, simple_loss=0.4379, pruned_loss=0.1719, over 27958.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3537, pruned_loss=0.1043, over 5657026.81 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3627, pruned_loss=0.1166, over 5729523.69 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3548, pruned_loss=0.1043, over 5640317.26 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:27:42,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807112.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:28:32,438 INFO [train.py:968] (1/2) Epoch 18, batch 30450, giga_loss[loss=0.2717, simple_loss=0.3469, pruned_loss=0.09823, over 28760.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3531, pruned_loss=0.1039, over 5654620.18 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3628, pruned_loss=0.1167, over 5732821.58 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3537, pruned_loss=0.1036, over 5637165.70 frames. ], batch size: 284, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:28:49,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5651, 1.8723, 1.6677, 1.5899], device='cuda:1'), covar=tensor([0.1636, 0.2020, 0.1956, 0.1934], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0740, 0.0701, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 12:28:49,722 INFO [optim.py:369] (1/2) 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 12:29:05,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3613, 1.5605, 1.6313, 1.2323], device='cuda:1'), covar=tensor([0.1795, 0.2661, 0.1520, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.0873, 0.0693, 0.0919, 0.0819], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 12:29:12,042 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 30500, giga_loss[loss=0.2349, simple_loss=0.3009, pruned_loss=0.08449, over 24191.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3499, pruned_loss=0.102, over 5649775.76 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3621, pruned_loss=0.1164, over 5734757.90 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3507, pruned_loss=0.1017, over 5631862.60 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:29:59,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2036, 2.4601, 2.1948, 2.0625], device='cuda:1'), covar=tensor([0.1745, 0.1988, 0.1888, 0.1982], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0739, 0.0700, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 12:30:03,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807255.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:30:05,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807258.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:30:08,277 INFO [train.py:968] (1/2) Epoch 18, batch 30550, giga_loss[loss=0.3358, simple_loss=0.3899, pruned_loss=0.1409, over 26594.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3484, pruned_loss=0.1012, over 5642887.67 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3619, pruned_loss=0.1165, over 5727339.68 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.349, pruned_loss=0.1006, over 5634723.63 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:30:25,849 INFO [optim.py:369] (1/2) 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 12:30:32,375 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807287.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:30:55,200 INFO [train.py:968] (1/2) Epoch 18, batch 30600, giga_loss[loss=0.2658, simple_loss=0.3339, pruned_loss=0.09884, over 24043.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3475, pruned_loss=0.1, over 5639336.36 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3614, pruned_loss=0.1163, over 5721788.64 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3482, pruned_loss=0.0994, over 5635564.20 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:31:28,524 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:31:28,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:31:30,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:31:42,681 INFO [train.py:968] (1/2) Epoch 18, batch 30650, giga_loss[loss=0.2644, simple_loss=0.3424, pruned_loss=0.09322, over 28913.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3448, pruned_loss=0.09763, over 5651307.16 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3608, pruned_loss=0.116, over 5726545.38 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3456, pruned_loss=0.097, over 5642402.72 frames. ], batch size: 227, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:31:57,722 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:32:14,551 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 30700, giga_loss[loss=0.2609, simple_loss=0.3475, pruned_loss=0.08715, over 28912.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3425, pruned_loss=0.09544, over 5654638.36 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3602, pruned_loss=0.1157, over 5727534.31 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3433, pruned_loss=0.09481, over 5645347.36 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:33:22,272 INFO [train.py:968] (1/2) Epoch 18, batch 30750, giga_loss[loss=0.2437, simple_loss=0.3253, pruned_loss=0.08105, over 28643.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3395, pruned_loss=0.09409, over 5638164.06 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3603, pruned_loss=0.116, over 5718647.77 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3395, pruned_loss=0.09278, over 5636064.36 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:33:39,531 INFO [optim.py:369] (1/2) 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 12:34:11,656 INFO [train.py:968] (1/2) Epoch 18, batch 30800, giga_loss[loss=0.2301, simple_loss=0.3193, pruned_loss=0.07046, over 28862.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3378, pruned_loss=0.09344, over 5637598.62 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3605, pruned_loss=0.1162, over 5711773.86 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3372, pruned_loss=0.09185, over 5641503.87 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:34:48,981 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:34:51,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 12:34:59,637 INFO [train.py:968] (1/2) Epoch 18, batch 30850, giga_loss[loss=0.2106, simple_loss=0.2962, pruned_loss=0.06251, over 28896.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3353, pruned_loss=0.09258, over 5637398.99 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3602, pruned_loss=0.1162, over 5717881.98 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3345, pruned_loss=0.09065, over 5632884.08 frames. ], batch size: 145, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:35:20,180 INFO [optim.py:369] (1/2) 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 12:35:50,154 INFO [train.py:968] (1/2) Epoch 18, batch 30900, giga_loss[loss=0.2699, simple_loss=0.3532, pruned_loss=0.09332, over 28967.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3371, pruned_loss=0.09407, over 5630113.44 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3592, pruned_loss=0.1158, over 5721480.22 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3365, pruned_loss=0.09215, over 5620470.71 frames. ], batch size: 164, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:36:44,846 INFO [train.py:968] (1/2) Epoch 18, batch 30950, giga_loss[loss=0.2699, simple_loss=0.3545, pruned_loss=0.09262, over 28428.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3407, pruned_loss=0.09464, over 5637405.75 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3591, pruned_loss=0.1158, over 5719161.25 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3399, pruned_loss=0.09265, over 5629619.25 frames. ], batch size: 369, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:37:06,803 INFO [optim.py:369] (1/2) 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 12:37:39,185 INFO [train.py:968] (1/2) Epoch 18, batch 31000, giga_loss[loss=0.2326, simple_loss=0.3191, pruned_loss=0.07304, over 28939.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3416, pruned_loss=0.09497, over 5634959.97 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3588, pruned_loss=0.1159, over 5699790.65 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3407, pruned_loss=0.0926, over 5643070.13 frames. ], batch size: 174, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:37:52,353 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:38:31,943 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:38:42,135 INFO [train.py:968] (1/2) Epoch 18, batch 31050, giga_loss[loss=0.2607, simple_loss=0.3438, pruned_loss=0.08878, over 28608.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3417, pruned_loss=0.09485, over 5654952.00 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3588, pruned_loss=0.116, over 5704681.73 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3406, pruned_loss=0.09242, over 5655884.79 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:38:52,304 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 18, batch 31100, giga_loss[loss=0.2481, simple_loss=0.3362, pruned_loss=0.07999, over 28795.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3398, pruned_loss=0.09371, over 5641701.97 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3585, pruned_loss=0.1161, over 5696417.69 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3386, pruned_loss=0.09098, over 5648766.04 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:39:43,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6263, 1.7796, 1.3682, 1.3027], device='cuda:1'), covar=tensor([0.0889, 0.0527, 0.0955, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0444, 0.0511, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:40:09,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=807836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:40:40,401 INFO [train.py:968] (1/2) Epoch 18, batch 31150, giga_loss[loss=0.2508, simple_loss=0.342, pruned_loss=0.07979, over 28970.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3385, pruned_loss=0.09162, over 5651223.25 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3586, pruned_loss=0.1161, over 5699768.79 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3372, pruned_loss=0.08915, over 5653346.79 frames. ], batch size: 155, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 12:40:43,183 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:40:47,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:40:58,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4943, 1.8793, 1.4670, 1.6591], device='cuda:1'), covar=tensor([0.0736, 0.0298, 0.0346, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0066, 0.0060, 0.0102], device='cuda:1')
+2023-03-09 12:41:02,402 INFO [optim.py:369] (1/2) 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 12:41:19,592 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:41:38,764 INFO [train.py:968] (1/2) Epoch 18, batch 31200, giga_loss[loss=0.2143, simple_loss=0.2785, pruned_loss=0.07502, over 24442.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3345, pruned_loss=0.08994, over 5657685.68 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3577, pruned_loss=0.1158, over 5703067.97 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3337, pruned_loss=0.08752, over 5655217.30 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:41:39,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=807912.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 12:41:41,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=807915.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:41:42,898 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 12:41:53,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=807925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:42:19,912 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=807944.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 12:42:38,987 INFO [train.py:968] (1/2) Epoch 18, batch 31250, giga_loss[loss=0.2747, simple_loss=0.3517, pruned_loss=0.09887, over 28196.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3333, pruned_loss=0.08992, over 5662841.84 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3574, pruned_loss=0.1157, over 5708510.23 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3322, pruned_loss=0.0873, over 5654687.27 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:42:59,294 INFO [optim.py:369] (1/2) 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 12:43:00,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6609, 1.9065, 1.9566, 1.4600], device='cuda:1'), covar=tensor([0.1921, 0.2775, 0.1602, 0.2045], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0690, 0.0919, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 12:43:18,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1637, 1.5284, 1.4617, 1.0566], device='cuda:1'), covar=tensor([0.1666, 0.2609, 0.1429, 0.1750], device='cuda:1'), in_proj_covar=tensor([0.0875, 0.0690, 0.0919, 0.0821], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 12:43:36,340 INFO [train.py:968] (1/2) Epoch 18, batch 31300, giga_loss[loss=0.264, simple_loss=0.3452, pruned_loss=0.09136, over 28930.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3328, pruned_loss=0.08969, over 5660632.81 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3571, pruned_loss=0.1155, over 5701042.68 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3319, pruned_loss=0.08745, over 5659480.25 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:44:27,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-09 12:44:29,321 INFO [train.py:968] (1/2) Epoch 18, batch 31350, giga_loss[loss=0.2541, simple_loss=0.3374, pruned_loss=0.08544, over 28371.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.334, pruned_loss=0.0899, over 5664707.46 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3568, pruned_loss=0.1155, over 5707685.48 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3329, pruned_loss=0.08733, over 5656629.99 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:44:34,963 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:44:38,160 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:45:14,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:45:30,671 INFO [train.py:968] (1/2) Epoch 18, batch 31400, giga_loss[loss=0.257, simple_loss=0.3373, pruned_loss=0.08836, over 28837.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3356, pruned_loss=0.09025, over 5669055.64 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3556, pruned_loss=0.1149, over 5708089.76 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3351, pruned_loss=0.08802, over 5661163.24 frames. ], batch size: 284, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:45:39,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=808117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:45:54,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:46:39,553 INFO [train.py:968] (1/2) Epoch 18, batch 31450, giga_loss[loss=0.2019, simple_loss=0.2926, pruned_loss=0.05562, over 29065.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3325, pruned_loss=0.08858, over 5665455.02 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3557, pruned_loss=0.1151, over 5707373.28 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3316, pruned_loss=0.08625, over 5659389.38 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:46:39,914 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:47:39,900 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 12:47:47,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:47:47,619 INFO [train.py:968] (1/2) Epoch 18, batch 31500, giga_loss[loss=0.2949, simple_loss=0.3691, pruned_loss=0.1103, over 28950.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3347, pruned_loss=0.09007, over 5674679.79 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3557, pruned_loss=0.1151, over 5712395.53 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3335, pruned_loss=0.08766, over 5664362.64 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:48:35,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7172, 1.9928, 1.5709, 2.1863], device='cuda:1'), covar=tensor([0.2625, 0.2636, 0.3028, 0.2253], device='cuda:1'), in_proj_covar=tensor([0.1445, 0.1049, 0.1285, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 12:48:38,448 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=808253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:48:47,084 INFO [train.py:968] (1/2) Epoch 18, batch 31550, giga_loss[loss=0.2444, simple_loss=0.3417, pruned_loss=0.07358, over 28070.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3384, pruned_loss=0.0912, over 5658135.66 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3558, pruned_loss=0.1153, over 5704489.08 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3368, pruned_loss=0.08836, over 5654782.76 frames. ], batch size: 412, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:48:50,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9359, 2.4204, 2.1729, 1.6605], device='cuda:1'), covar=tensor([0.2091, 0.1493, 0.1566, 0.2020], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1779, 0.1712, 0.1860], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 12:48:56,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:49:01,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:49:09,152 INFO [optim.py:369] (1/2) 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 12:49:35,344 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:49:47,712 INFO [train.py:968] (1/2) Epoch 18, batch 31600, giga_loss[loss=0.2499, simple_loss=0.3462, pruned_loss=0.07674, over 29035.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3407, pruned_loss=0.0904, over 5667780.21 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.355, pruned_loss=0.1149, over 5709983.95 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3395, pruned_loss=0.0875, over 5658467.66 frames. ], batch size: 128, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:50:21,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 12:50:30,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7286, 3.5545, 3.3794, 1.7897], device='cuda:1'), covar=tensor([0.0662, 0.0820, 0.0737, 0.2437], device='cuda:1'), in_proj_covar=tensor([0.1156, 0.1076, 0.0920, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 12:50:38,131 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:50:42,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:50:47,508 INFO [train.py:968] (1/2) Epoch 18, batch 31650, giga_loss[loss=0.2733, simple_loss=0.3395, pruned_loss=0.1036, over 26623.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3409, pruned_loss=0.08946, over 5659230.98 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3545, pruned_loss=0.1145, over 5712832.52 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3402, pruned_loss=0.08698, over 5648377.81 frames. ], batch size: 555, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:51:09,975 INFO [optim.py:369] (1/2) 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 12:51:18,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:51:18,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5726, 1.5509, 1.2857, 1.2157], device='cuda:1'), covar=tensor([0.0803, 0.0396, 0.0824, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0444, 0.0512, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:51:19,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4880, 1.5322, 1.2164, 1.1654], device='cuda:1'), covar=tensor([0.0861, 0.0516, 0.0985, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0444, 0.0512, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 12:51:49,905 INFO [train.py:968] (1/2) Epoch 18, batch 31700, giga_loss[loss=0.2692, simple_loss=0.3497, pruned_loss=0.09432, over 28800.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3405, pruned_loss=0.08864, over 5649857.73 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3543, pruned_loss=0.1145, over 5702893.58 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3399, pruned_loss=0.0864, over 5650475.77 frames. ], batch size: 243, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:51:50,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8876, 5.4495, 2.0582, 2.1653], device='cuda:1'), covar=tensor([0.0952, 0.0302, 0.0880, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0542, 0.0372, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 12:51:53,397 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 12:51:57,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3351, 4.1436, 3.9383, 2.0025], device='cuda:1'), covar=tensor([0.0533, 0.0698, 0.0734, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1153, 0.1072, 0.0917, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 12:52:46,924 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4941, 1.6009, 1.7624, 1.3590], device='cuda:1'), covar=tensor([0.1693, 0.2487, 0.1441, 0.1714], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0692, 0.0924, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 12:52:49,133 INFO [train.py:968] (1/2) Epoch 18, batch 31750, giga_loss[loss=0.2617, simple_loss=0.3395, pruned_loss=0.09195, over 28370.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3413, pruned_loss=0.0902, over 5644311.59 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.354, pruned_loss=0.1143, over 5697096.37 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08773, over 5649087.14 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:52:50,398 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5330, 1.8961, 1.7577, 1.5787], device='cuda:1'), covar=tensor([0.1742, 0.1696, 0.1993, 0.1894], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0727, 0.0689, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 12:53:13,598 INFO [optim.py:369] (1/2) 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 12:53:29,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:53:30,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8571, 2.7749, 1.8320, 0.8863], device='cuda:1'), covar=tensor([0.6676, 0.2924, 0.3507, 0.6537], device='cuda:1'), in_proj_covar=tensor([0.1697, 0.1605, 0.1575, 0.1390], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 12:53:58,791 INFO [train.py:968] (1/2) Epoch 18, batch 31800, giga_loss[loss=0.2596, simple_loss=0.3426, pruned_loss=0.08828, over 28975.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3404, pruned_loss=0.09072, over 5656939.44 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3534, pruned_loss=0.114, over 5701251.12 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3401, pruned_loss=0.0886, over 5656262.39 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:54:36,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:55:18,291 INFO [train.py:968] (1/2) Epoch 18, batch 31850, giga_loss[loss=0.2679, simple_loss=0.3413, pruned_loss=0.09727, over 27482.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3412, pruned_loss=0.09187, over 5656945.67 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3533, pruned_loss=0.114, over 5694363.68 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3407, pruned_loss=0.08957, over 5662308.70 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:55:46,195 INFO [optim.py:369] (1/2) 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 12:55:58,465 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 12:56:30,155 INFO [train.py:968] (1/2) Epoch 18, batch 31900, giga_loss[loss=0.2179, simple_loss=0.3046, pruned_loss=0.06559, over 28842.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3365, pruned_loss=0.08903, over 5663194.44 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3532, pruned_loss=0.1139, over 5696829.94 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.336, pruned_loss=0.08692, over 5664660.07 frames. ], batch size: 119, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:56:52,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=808628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:56:53,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1382, 1.5478, 1.4382, 1.0335], device='cuda:1'), covar=tensor([0.1828, 0.2623, 0.1553, 0.1777], device='cuda:1'), in_proj_covar=tensor([0.0877, 0.0692, 0.0922, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 12:57:01,798 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:57:05,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:57:35,116 INFO [train.py:968] (1/2) Epoch 18, batch 31950, giga_loss[loss=0.2291, simple_loss=0.3168, pruned_loss=0.0707, over 28701.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3347, pruned_loss=0.08797, over 5669444.20 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3529, pruned_loss=0.1137, over 5702184.16 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3342, pruned_loss=0.08591, over 5665147.13 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 12:57:42,749 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=808667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:57:53,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-09 12:57:59,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-09 12:58:00,197 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:58:01,594 INFO [optim.py:369] (1/2) 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 12:58:02,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:58:02,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3178, 1.8577, 1.3430, 0.4831], device='cuda:1'), covar=tensor([0.4117, 0.2216, 0.3616, 0.5474], device='cuda:1'), in_proj_covar=tensor([0.1689, 0.1598, 0.1565, 0.1383], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 12:58:39,298 INFO [train.py:968] (1/2) Epoch 18, batch 32000, giga_loss[loss=0.2118, simple_loss=0.2842, pruned_loss=0.06969, over 24352.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3337, pruned_loss=0.08801, over 5665079.85 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.353, pruned_loss=0.1139, over 5700758.26 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3327, pruned_loss=0.08551, over 5661786.78 frames. ], batch size: 705, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:58:40,660 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 32050, giga_loss[loss=0.3079, simple_loss=0.3819, pruned_loss=0.117, over 28493.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.338, pruned_loss=0.09027, over 5661906.54 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3528, pruned_loss=0.1139, over 5695687.75 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3371, pruned_loss=0.08787, over 5663144.18 frames. ], batch size: 336, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 12:59:53,887 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=808771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 12:59:56,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=808774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:00:05,438 INFO [optim.py:369] (1/2) 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 13:00:20,362 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=808792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:00:26,032 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-09 13:00:32,411 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 32100, giga_loss[loss=0.2491, simple_loss=0.3241, pruned_loss=0.08706, over 28970.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3387, pruned_loss=0.09108, over 5669610.58 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3529, pruned_loss=0.1139, over 5699670.66 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3375, pruned_loss=0.08853, over 5666268.79 frames. ], batch size: 284, lr: 1.77e-03, grad_scale: 8.0
+2023-03-09 13:01:26,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4931, 3.8030, 1.6662, 1.6720], device='cuda:1'), covar=tensor([0.0980, 0.0351, 0.0982, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0542, 0.0374, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 13:01:34,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 13:01:47,469 INFO [train.py:968] (1/2) Epoch 18, batch 32150, giga_loss[loss=0.253, simple_loss=0.3259, pruned_loss=0.09005, over 28696.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3376, pruned_loss=0.0917, over 5667501.46 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3525, pruned_loss=0.1138, over 5702540.77 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3366, pruned_loss=0.08912, over 5661143.25 frames. ], batch size: 307, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 13:02:09,093 INFO [optim.py:369] (1/2) 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 13:02:45,862 INFO [train.py:968] (1/2) Epoch 18, batch 32200, giga_loss[loss=0.2443, simple_loss=0.3316, pruned_loss=0.07846, over 28942.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3376, pruned_loss=0.09203, over 5670118.91 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3527, pruned_loss=0.1139, over 5707986.25 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3363, pruned_loss=0.08929, over 5659421.26 frames. ], batch size: 186, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 13:03:54,065 INFO [train.py:968] (1/2) Epoch 18, batch 32250, giga_loss[loss=0.3135, simple_loss=0.3808, pruned_loss=0.1231, over 27586.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3392, pruned_loss=0.09239, over 5664964.73 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3526, pruned_loss=0.1139, over 5701327.78 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3379, pruned_loss=0.0896, over 5661379.19 frames. ], batch size: 472, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 13:04:25,205 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 32300, giga_loss[loss=0.2632, simple_loss=0.3525, pruned_loss=0.08692, over 29039.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3408, pruned_loss=0.09237, over 5677977.85 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.352, pruned_loss=0.1137, over 5706161.80 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3398, pruned_loss=0.08957, over 5669495.46 frames. ], batch size: 199, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 13:06:17,159 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 32350, giga_loss[loss=0.2426, simple_loss=0.3259, pruned_loss=0.07969, over 28360.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3396, pruned_loss=0.09137, over 5677224.87 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3516, pruned_loss=0.1134, over 5709741.44 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3388, pruned_loss=0.08884, over 5666775.28 frames. ], batch size: 368, lr: 1.77e-03, grad_scale: 2.0
+2023-03-09 13:06:46,034 INFO [optim.py:369] (1/2) 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 13:07:15,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2869, 1.6495, 1.6048, 1.4536], device='cuda:1'), covar=tensor([0.1595, 0.1555, 0.1740, 0.1574], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0730, 0.0690, 0.0666], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 13:07:27,220 INFO [train.py:968] (1/2) Epoch 18, batch 32400, giga_loss[loss=0.222, simple_loss=0.303, pruned_loss=0.0705, over 28962.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3358, pruned_loss=0.09061, over 5665928.61 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3519, pruned_loss=0.1136, over 5702625.52 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3347, pruned_loss=0.08808, over 5663077.11 frames. ], batch size: 136, lr: 1.77e-03, grad_scale: 4.0
+2023-03-09 13:08:03,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6845, 2.1878, 1.9940, 1.6204], device='cuda:1'), covar=tensor([0.2848, 0.1775, 0.2067, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1772, 0.1710, 0.1862], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 13:08:22,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0737, 3.8973, 3.6956, 1.8206], device='cuda:1'), covar=tensor([0.0643, 0.0784, 0.0803, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.1159, 0.1076, 0.0921, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 13:08:35,039 INFO [train.py:968] (1/2) Epoch 18, batch 32450, giga_loss[loss=0.2113, simple_loss=0.2908, pruned_loss=0.06586, over 28333.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3295, pruned_loss=0.08774, over 5674222.65 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3518, pruned_loss=0.1137, over 5706692.84 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3284, pruned_loss=0.08519, over 5667653.71 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:08:41,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:09:00,110 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-09 13:09:04,139 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 32500, giga_loss[loss=0.2913, simple_loss=0.355, pruned_loss=0.1138, over 27952.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3301, pruned_loss=0.0886, over 5664404.75 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3518, pruned_loss=0.1138, over 5708913.24 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3286, pruned_loss=0.08571, over 5656067.67 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:09:43,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809218.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:10:04,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-09 13:10:34,773 INFO [train.py:968] (1/2) Epoch 18, batch 32550, giga_loss[loss=0.2258, simple_loss=0.3139, pruned_loss=0.06886, over 28968.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3318, pruned_loss=0.08973, over 5661116.39 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3518, pruned_loss=0.1136, over 5710759.60 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3303, pruned_loss=0.08726, over 5652095.58 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:11:02,935 INFO [optim.py:369] (1/2) 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 13:11:23,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-09 13:11:33,838 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 32600, giga_loss[loss=0.2184, simple_loss=0.3133, pruned_loss=0.0617, over 28866.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3292, pruned_loss=0.08775, over 5659571.52 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3519, pruned_loss=0.1138, over 5714466.47 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3275, pruned_loss=0.08516, over 5648230.67 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:11:38,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=809313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:12:07,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 13:12:11,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4974, 1.9419, 1.7839, 1.6408], device='cuda:1'), covar=tensor([0.1843, 0.1827, 0.2181, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.0440, 0.0727, 0.0687, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 13:12:14,164 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=809342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:12:39,959 INFO [train.py:968] (1/2) Epoch 18, batch 32650, giga_loss[loss=0.2409, simple_loss=0.3191, pruned_loss=0.08136, over 28457.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3279, pruned_loss=0.08626, over 5669302.68 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3517, pruned_loss=0.1138, over 5715751.45 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3262, pruned_loss=0.08359, over 5658172.83 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:13:07,750 INFO [optim.py:369] (1/2) 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 13:13:46,440 INFO [train.py:968] (1/2) Epoch 18, batch 32700, giga_loss[loss=0.2779, simple_loss=0.347, pruned_loss=0.1044, over 27582.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3278, pruned_loss=0.08715, over 5669463.00 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3514, pruned_loss=0.1137, over 5715993.87 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3261, pruned_loss=0.08441, over 5658908.25 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:14:19,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:14:54,909 INFO [train.py:968] (1/2) Epoch 18, batch 32750, giga_loss[loss=0.2464, simple_loss=0.3243, pruned_loss=0.08421, over 27675.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3257, pruned_loss=0.08523, over 5661149.15 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3507, pruned_loss=0.1134, over 5716367.55 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3246, pruned_loss=0.08287, over 5651157.99 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:15:25,785 INFO [optim.py:369] (1/2) 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 13:15:32,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-09 13:15:58,529 INFO [train.py:968] (1/2) Epoch 18, batch 32800, giga_loss[loss=0.3126, simple_loss=0.3774, pruned_loss=0.1239, over 28587.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3258, pruned_loss=0.08533, over 5662069.62 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3508, pruned_loss=0.1135, over 5717922.13 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.08254, over 5650730.34 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:16:05,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:17:00,140 INFO [train.py:968] (1/2) Epoch 18, batch 32850, giga_loss[loss=0.2371, simple_loss=0.3203, pruned_loss=0.07693, over 28646.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3272, pruned_loss=0.08659, over 5657212.72 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3507, pruned_loss=0.1134, over 5709333.15 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3256, pruned_loss=0.0841, over 5654777.04 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:17:22,840 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=809578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:17:27,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=809581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:17:29,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3450, 1.6072, 1.3172, 1.0503], device='cuda:1'), covar=tensor([0.2695, 0.2636, 0.3064, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1445, 0.1047, 0.1287, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 13:17:33,111 INFO [optim.py:369] (1/2) 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 13:17:42,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809593.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:17:59,029 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=809610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:18:00,014 INFO [train.py:968] (1/2) Epoch 18, batch 32900, libri_loss[loss=0.2613, simple_loss=0.3287, pruned_loss=0.09696, over 29525.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3255, pruned_loss=0.08562, over 5648152.48 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3499, pruned_loss=0.1131, over 5704563.16 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3242, pruned_loss=0.08304, over 5649061.11 frames. ], batch size: 82, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:18:13,197 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:18:57,765 INFO [train.py:968] (1/2) Epoch 18, batch 32950, giga_loss[loss=0.2455, simple_loss=0.3347, pruned_loss=0.07809, over 28480.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3274, pruned_loss=0.0852, over 5659365.33 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3494, pruned_loss=0.1128, over 5711048.45 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.326, pruned_loss=0.08249, over 5652589.02 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:19:22,718 INFO [optim.py:369] (1/2) 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 13:19:34,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5348, 1.9228, 1.7088, 1.5407], device='cuda:1'), covar=tensor([0.1778, 0.1842, 0.2062, 0.2014], device='cuda:1'), in_proj_covar=tensor([0.0442, 0.0726, 0.0688, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 13:19:49,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1914, 1.2798, 1.0985, 0.8835], device='cuda:1'), covar=tensor([0.0875, 0.0440, 0.0944, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0440, 0.0509, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 13:19:57,057 INFO [train.py:968] (1/2) Epoch 18, batch 33000, giga_loss[loss=0.2507, simple_loss=0.3422, pruned_loss=0.07962, over 28812.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3303, pruned_loss=0.08571, over 5646184.95 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3493, pruned_loss=0.1127, over 5694271.51 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3289, pruned_loss=0.08316, over 5653937.85 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:19:57,057 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 13:20:05,812 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 13:20:33,944 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=809736.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:20:34,567 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=809737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:20:37,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=809739.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:21:08,336 INFO [train.py:968] (1/2) Epoch 18, batch 33050, giga_loss[loss=0.2376, simple_loss=0.3216, pruned_loss=0.0768, over 28373.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3317, pruned_loss=0.087, over 5637754.78 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3489, pruned_loss=0.1125, over 5696080.42 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3307, pruned_loss=0.08454, over 5641017.50 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:21:16,883 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=809768.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:21:35,585 INFO [optim.py:369] (1/2) 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 13:22:04,904 INFO [train.py:968] (1/2) Epoch 18, batch 33100, giga_loss[loss=0.3172, simple_loss=0.3755, pruned_loss=0.1294, over 28035.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3321, pruned_loss=0.08775, over 5658352.32 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3481, pruned_loss=0.112, over 5705423.88 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3312, pruned_loss=0.08517, over 5650474.03 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:22:57,331 INFO [train.py:968] (1/2) Epoch 18, batch 33150, giga_loss[loss=0.2857, simple_loss=0.3576, pruned_loss=0.1069, over 28094.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3309, pruned_loss=0.0874, over 5639956.48 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3474, pruned_loss=0.1117, over 5681566.22 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.33, pruned_loss=0.08434, over 5655029.54 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:23:15,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8680, 2.4340, 2.1984, 1.7074], device='cuda:1'), covar=tensor([0.3042, 0.1774, 0.2042, 0.2661], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1763, 0.1690, 0.1841], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 13:23:29,393 INFO [optim.py:369] (1/2) 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 13:23:35,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:23:38,016 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 13:23:53,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2534, 1.2751, 3.6716, 3.1910], device='cuda:1'), covar=tensor([0.1672, 0.2843, 0.0452, 0.1231], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0631, 0.0925, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 13:24:00,679 INFO [train.py:968] (1/2) Epoch 18, batch 33200, giga_loss[loss=0.2205, simple_loss=0.3104, pruned_loss=0.06533, over 28940.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3293, pruned_loss=0.08629, over 5642173.27 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3473, pruned_loss=0.1117, over 5683007.22 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3285, pruned_loss=0.08366, over 5652278.35 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:25:00,407 INFO [train.py:968] (1/2) Epoch 18, batch 33250, giga_loss[loss=0.2554, simple_loss=0.3304, pruned_loss=0.09021, over 27662.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3271, pruned_loss=0.08563, over 5652993.19 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3474, pruned_loss=0.1118, over 5687273.98 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3261, pruned_loss=0.08302, over 5656571.32 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:25:28,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4049, 1.4807, 1.3088, 1.6345], device='cuda:1'), covar=tensor([0.0766, 0.0319, 0.0350, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 13:25:31,494 INFO [optim.py:369] (1/2) 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 13:25:45,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=809997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:26:02,848 INFO [train.py:968] (1/2) Epoch 18, batch 33300, giga_loss[loss=0.2486, simple_loss=0.3291, pruned_loss=0.08402, over 28409.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3288, pruned_loss=0.08625, over 5667439.14 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3475, pruned_loss=0.1119, over 5693616.76 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3273, pruned_loss=0.08329, over 5663507.15 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:26:31,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:26:33,904 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 33350, giga_loss[loss=0.2427, simple_loss=0.3237, pruned_loss=0.08082, over 28675.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3309, pruned_loss=0.08717, over 5665766.28 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3472, pruned_loss=0.1118, over 5692234.08 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3296, pruned_loss=0.0844, over 5663093.29 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:27:11,779 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:27:34,385 INFO [optim.py:369] (1/2) 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 13:28:07,533 INFO [train.py:968] (1/2) Epoch 18, batch 33400, giga_loss[loss=0.2804, simple_loss=0.3548, pruned_loss=0.103, over 28709.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3314, pruned_loss=0.08797, over 5666505.16 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3469, pruned_loss=0.1115, over 5698465.86 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3302, pruned_loss=0.0853, over 5658250.10 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:28:07,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:28:50,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:28:55,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 33450, giga_loss[loss=0.2839, simple_loss=0.3635, pruned_loss=0.1021, over 28351.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3345, pruned_loss=0.08892, over 5670651.16 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3471, pruned_loss=0.1118, over 5698548.53 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3333, pruned_loss=0.08651, over 5664010.97 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:29:33,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:29:37,069 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:29:40,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:29:47,342 INFO [optim.py:369] (1/2) 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 13:30:14,951 INFO [train.py:968] (1/2) Epoch 18, batch 33500, giga_loss[loss=0.2611, simple_loss=0.3299, pruned_loss=0.0961, over 24605.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3362, pruned_loss=0.08928, over 5668013.62 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3466, pruned_loss=0.1113, over 5705151.78 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3354, pruned_loss=0.08701, over 5655612.87 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:31:09,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:31:13,533 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 33550, libri_loss[loss=0.2539, simple_loss=0.3124, pruned_loss=0.0977, over 29346.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3351, pruned_loss=0.08882, over 5672149.35 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3456, pruned_loss=0.1108, over 5713164.22 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3349, pruned_loss=0.08661, over 5653689.88 frames. ], batch size: 71, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:31:23,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1669, 1.4566, 1.2738, 1.0537], device='cuda:1'), covar=tensor([0.2432, 0.2043, 0.1560, 0.2134], device='cuda:1'), in_proj_covar=tensor([0.1855, 0.1759, 0.1689, 0.1838], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 13:31:49,607 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:32:22,986 INFO [train.py:968] (1/2) Epoch 18, batch 33600, giga_loss[loss=0.2455, simple_loss=0.325, pruned_loss=0.08303, over 28422.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3335, pruned_loss=0.08838, over 5676278.32 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3448, pruned_loss=0.1103, over 5715689.65 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3337, pruned_loss=0.08635, over 5657772.02 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:32:48,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5743, 1.6884, 1.2951, 1.2741], device='cuda:1'), covar=tensor([0.0952, 0.0536, 0.1022, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0442, 0.0513, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 13:33:08,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4447, 3.3540, 1.4686, 1.6093], device='cuda:1'), covar=tensor([0.0934, 0.0464, 0.0929, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0539, 0.0370, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 13:33:25,963 INFO [train.py:968] (1/2) Epoch 18, batch 33650, libri_loss[loss=0.3137, simple_loss=0.3735, pruned_loss=0.127, over 29364.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3346, pruned_loss=0.09026, over 5669540.58 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3452, pruned_loss=0.1105, over 5711422.14 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3339, pruned_loss=0.08752, over 5657014.11 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:33:33,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 13:33:56,587 INFO [optim.py:369] (1/2) 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 13:34:22,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4875, 3.8734, 1.7982, 1.5819], device='cuda:1'), covar=tensor([0.0987, 0.0316, 0.0893, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0539, 0.0370, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 13:34:27,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0730, 5.8343, 5.5587, 2.8762], device='cuda:1'), covar=tensor([0.0518, 0.0737, 0.0932, 0.1756], device='cuda:1'), in_proj_covar=tensor([0.1156, 0.1068, 0.0916, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 13:34:29,026 INFO [train.py:968] (1/2) Epoch 18, batch 33700, giga_loss[loss=0.2443, simple_loss=0.332, pruned_loss=0.07826, over 28676.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.334, pruned_loss=0.08984, over 5656242.99 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3453, pruned_loss=0.1107, over 5705054.04 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.333, pruned_loss=0.087, over 5651064.03 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:34:29,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:35:37,452 INFO [train.py:968] (1/2) Epoch 18, batch 33750, giga_loss[loss=0.2452, simple_loss=0.3243, pruned_loss=0.08309, over 28945.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.333, pruned_loss=0.08968, over 5664932.55 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3446, pruned_loss=0.1101, over 5707877.59 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3327, pruned_loss=0.08755, over 5657359.12 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:35:47,268 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810470.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:36:11,887 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 33800, giga_loss[loss=0.2155, simple_loss=0.2803, pruned_loss=0.07538, over 24361.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3313, pruned_loss=0.08964, over 5653877.92 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3439, pruned_loss=0.1098, over 5712546.48 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3314, pruned_loss=0.08769, over 5642189.92 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:36:44,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3859, 3.9286, 1.5436, 1.6220], device='cuda:1'), covar=tensor([0.0962, 0.0375, 0.0973, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0539, 0.0371, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 13:36:51,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3833, 1.2654, 4.2483, 3.4086], device='cuda:1'), covar=tensor([0.1720, 0.2932, 0.0419, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0630, 0.0927, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 13:36:56,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4794, 1.5930, 1.5169, 1.4488], device='cuda:1'), covar=tensor([0.2298, 0.1897, 0.1777, 0.1964], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1777, 0.1702, 0.1852], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 13:36:58,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7090, 2.3647, 1.7893, 0.8742], device='cuda:1'), covar=tensor([0.5367, 0.2874, 0.3790, 0.6009], device='cuda:1'), in_proj_covar=tensor([0.1690, 0.1603, 0.1568, 0.1383], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 13:37:28,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:37:33,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:37:40,193 INFO [train.py:968] (1/2) Epoch 18, batch 33850, giga_loss[loss=0.2578, simple_loss=0.3482, pruned_loss=0.08375, over 28094.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3309, pruned_loss=0.08838, over 5654872.13 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3437, pruned_loss=0.1097, over 5707983.89 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3308, pruned_loss=0.08635, over 5647204.04 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:37:50,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6095, 4.4755, 4.2168, 2.1107], device='cuda:1'), covar=tensor([0.0567, 0.0721, 0.0824, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.1159, 0.1071, 0.0919, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 13:38:14,932 INFO [optim.py:369] (1/2) 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 13:38:43,843 INFO [train.py:968] (1/2) Epoch 18, batch 33900, giga_loss[loss=0.2478, simple_loss=0.3394, pruned_loss=0.07812, over 28401.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3303, pruned_loss=0.08604, over 5668102.72 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3434, pruned_loss=0.1095, over 5709743.95 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3303, pruned_loss=0.08436, over 5660089.72 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:39:07,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4288, 3.7128, 1.5623, 1.6057], device='cuda:1'), covar=tensor([0.1010, 0.0318, 0.0945, 0.1399], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0540, 0.0372, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 13:39:32,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1825, 1.2945, 1.1163, 0.9050], device='cuda:1'), covar=tensor([0.1032, 0.0533, 0.1171, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0443, 0.0513, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 13:39:40,097 INFO [train.py:968] (1/2) Epoch 18, batch 33950, giga_loss[loss=0.2992, simple_loss=0.3776, pruned_loss=0.1104, over 28166.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3327, pruned_loss=0.08574, over 5676265.00 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3433, pruned_loss=0.1094, over 5715218.37 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3324, pruned_loss=0.08385, over 5663994.57 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 13:40:10,962 INFO [optim.py:369] (1/2) 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 13:40:18,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:23,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:23,462 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:25,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:27,602 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:38,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:40:40,437 INFO [train.py:968] (1/2) Epoch 18, batch 34000, giga_loss[loss=0.2551, simple_loss=0.344, pruned_loss=0.08316, over 28808.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3334, pruned_loss=0.08584, over 5665440.79 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3432, pruned_loss=0.1094, over 5706250.37 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3332, pruned_loss=0.08395, over 5661980.75 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:40:59,285 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:41:45,631 INFO [train.py:968] (1/2) Epoch 18, batch 34050, giga_loss[loss=0.2141, simple_loss=0.2844, pruned_loss=0.07194, over 24607.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3343, pruned_loss=0.08655, over 5665385.16 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3439, pruned_loss=0.1098, over 5709274.04 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3332, pruned_loss=0.08377, over 5658294.81 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:42:20,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:42:20,922 INFO [optim.py:369] (1/2) 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 13:42:32,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:42:52,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5461, 1.7539, 1.3980, 1.5485], device='cuda:1'), covar=tensor([0.2950, 0.2785, 0.3285, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.1442, 0.1044, 0.1282, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 13:42:54,832 INFO [train.py:968] (1/2) Epoch 18, batch 34100, giga_loss[loss=0.2728, simple_loss=0.3545, pruned_loss=0.09557, over 28023.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3343, pruned_loss=0.08607, over 5673758.02 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3437, pruned_loss=0.1097, over 5711559.74 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3334, pruned_loss=0.08376, over 5665776.44 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:43:47,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=810845.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:44:11,689 INFO [train.py:968] (1/2) Epoch 18, batch 34150, giga_loss[loss=0.2486, simple_loss=0.3374, pruned_loss=0.07989, over 29027.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3345, pruned_loss=0.08578, over 5667855.19 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3438, pruned_loss=0.1098, over 5712534.94 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3338, pruned_loss=0.08381, over 5660590.90 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:44:41,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810879.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:44:56,839 INFO [optim.py:369] (1/2) 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 13:45:28,799 INFO [train.py:968] (1/2) Epoch 18, batch 34200, giga_loss[loss=0.2441, simple_loss=0.3295, pruned_loss=0.07939, over 28647.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3348, pruned_loss=0.08591, over 5660958.84 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3433, pruned_loss=0.1096, over 5715188.80 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3345, pruned_loss=0.08419, over 5652099.45 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:45:33,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5101, 4.3275, 4.1287, 2.1452], device='cuda:1'), covar=tensor([0.0595, 0.0724, 0.0793, 0.1814], device='cuda:1'), in_proj_covar=tensor([0.1146, 0.1059, 0.0909, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 13:45:51,391 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:45:56,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:46:15,719 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:46:29,353 INFO [train.py:968] (1/2) Epoch 18, batch 34250, libri_loss[loss=0.3124, simple_loss=0.3689, pruned_loss=0.1279, over 29534.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3389, pruned_loss=0.08808, over 5671612.29 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3444, pruned_loss=0.1103, over 5717405.45 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3374, pruned_loss=0.08508, over 5660141.21 frames. ], batch size: 89, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:46:29,569 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=810962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:47:00,899 INFO [optim.py:369] (1/2) 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 13:47:04,708 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=810988.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:47:05,412 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=810989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:47:09,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=810991.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:47:32,125 INFO [train.py:968] (1/2) Epoch 18, batch 34300, libri_loss[loss=0.2528, simple_loss=0.3231, pruned_loss=0.09123, over 29568.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3389, pruned_loss=0.08759, over 5682631.29 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3441, pruned_loss=0.1099, over 5719825.20 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3378, pruned_loss=0.08484, over 5669686.40 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:47:43,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811020.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:47:53,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2196, 4.0472, 3.8575, 1.7360], device='cuda:1'), covar=tensor([0.0580, 0.0707, 0.0763, 0.2114], device='cuda:1'), in_proj_covar=tensor([0.1153, 0.1064, 0.0913, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 13:48:40,642 INFO [train.py:968] (1/2) Epoch 18, batch 34350, giga_loss[loss=0.2089, simple_loss=0.3003, pruned_loss=0.05873, over 28632.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3376, pruned_loss=0.08783, over 5690503.18 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3437, pruned_loss=0.1098, over 5724448.32 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3368, pruned_loss=0.08502, over 5674940.13 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:49:01,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:49:10,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2539, 1.6545, 1.5178, 1.3201], device='cuda:1'), covar=tensor([0.0738, 0.0376, 0.0292, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 13:49:10,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:49:12,432 INFO [optim.py:369] (1/2) 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 13:49:43,543 INFO [train.py:968] (1/2) Epoch 18, batch 34400, giga_loss[loss=0.2489, simple_loss=0.3436, pruned_loss=0.07714, over 29006.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3364, pruned_loss=0.08772, over 5686558.89 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3439, pruned_loss=0.1098, over 5718754.55 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3354, pruned_loss=0.08466, over 5677336.57 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:51:00,449 INFO [train.py:968] (1/2) Epoch 18, batch 34450, giga_loss[loss=0.2243, simple_loss=0.3124, pruned_loss=0.06815, over 28936.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3335, pruned_loss=0.08484, over 5695257.40 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3437, pruned_loss=0.1096, over 5720688.19 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3328, pruned_loss=0.08233, over 5685887.59 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:51:09,097 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:52:05,589 INFO [train.py:968] (1/2) Epoch 18, batch 34500, giga_loss[loss=0.2759, simple_loss=0.3537, pruned_loss=0.09904, over 28787.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3337, pruned_loss=0.08521, over 5687882.27 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3436, pruned_loss=0.1096, over 5714051.14 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3331, pruned_loss=0.08281, over 5685301.77 frames. ], batch size: 263, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:52:19,502 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:23,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:28,615 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:30,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:53,620 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:52:56,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:53:02,524 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 34550, giga_loss[loss=0.3004, simple_loss=0.3682, pruned_loss=0.1163, over 28919.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3355, pruned_loss=0.08644, over 5688458.38 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3439, pruned_loss=0.1099, over 5719348.94 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3344, pruned_loss=0.08356, over 5680852.72 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:53:08,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4593, 1.8564, 1.5384, 1.6137], device='cuda:1'), covar=tensor([0.1747, 0.2087, 0.2173, 0.1975], device='cuda:1'), in_proj_covar=tensor([0.0437, 0.0725, 0.0687, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 13:53:19,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4965, 1.7751, 1.4341, 1.3210], device='cuda:1'), covar=tensor([0.2414, 0.2446, 0.2751, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1041, 0.1280, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 13:53:34,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:53:36,144 INFO [optim.py:369] (1/2) 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 13:53:36,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2698, 1.5612, 1.4808, 1.1009], device='cuda:1'), covar=tensor([0.1388, 0.2289, 0.1231, 0.1500], device='cuda:1'), in_proj_covar=tensor([0.0874, 0.0689, 0.0920, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 13:53:59,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:53:59,747 INFO [train.py:968] (1/2) Epoch 18, batch 34600, libri_loss[loss=0.2897, simple_loss=0.3529, pruned_loss=0.1132, over 29057.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3373, pruned_loss=0.08821, over 5683968.86 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3441, pruned_loss=0.11, over 5725348.66 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.336, pruned_loss=0.08478, over 5670692.53 frames. ], batch size: 101, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:54:04,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:15,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:38,178 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:52,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:54:58,752 INFO [train.py:968] (1/2) Epoch 18, batch 34650, giga_loss[loss=0.2423, simple_loss=0.316, pruned_loss=0.08432, over 28185.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.08737, over 5667861.51 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3442, pruned_loss=0.1103, over 5709662.02 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3328, pruned_loss=0.08389, over 5671987.84 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:55:02,047 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:55:34,229 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:55:37,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811397.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:55:40,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811400.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:55:53,896 INFO [train.py:968] (1/2) Epoch 18, batch 34700, libri_loss[loss=0.2741, simple_loss=0.3226, pruned_loss=0.1128, over 29622.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3334, pruned_loss=0.08783, over 5658546.99 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3439, pruned_loss=0.1102, over 5702893.26 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3323, pruned_loss=0.08441, over 5666130.59 frames. ], batch size: 69, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:56:15,134 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 13:56:49,652 INFO [train.py:968] (1/2) Epoch 18, batch 34750, giga_loss[loss=0.306, simple_loss=0.3666, pruned_loss=0.1227, over 26630.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3349, pruned_loss=0.08961, over 5657084.18 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3431, pruned_loss=0.1098, over 5707462.11 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3345, pruned_loss=0.08657, over 5657873.50 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:56:53,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:56:55,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:17,111 INFO [optim.py:369] (1/2) 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 13:57:24,326 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:31,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:35,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:57:35,932 INFO [train.py:968] (1/2) Epoch 18, batch 34800, giga_loss[loss=0.2967, simple_loss=0.3774, pruned_loss=0.108, over 28906.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3443, pruned_loss=0.09492, over 5664443.17 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3434, pruned_loss=0.11, over 5706016.32 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3436, pruned_loss=0.09207, over 5666039.47 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 13:57:49,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-09 13:57:49,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5382, 4.5936, 1.8387, 1.6147], device='cuda:1'), covar=tensor([0.1040, 0.0213, 0.0877, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0534, 0.0370, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 13:58:02,592 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 34850, giga_loss[loss=0.3025, simple_loss=0.3774, pruned_loss=0.1138, over 28275.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3516, pruned_loss=0.09881, over 5663148.75 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3436, pruned_loss=0.1102, over 5699444.40 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3508, pruned_loss=0.09618, over 5670021.09 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:58:35,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6440, 1.7684, 1.2608, 1.3528], device='cuda:1'), covar=tensor([0.0930, 0.0600, 0.0979, 0.1421], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0441, 0.0511, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 13:58:42,152 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811579.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 13:58:51,316 INFO [optim.py:369] (1/2) 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 13:59:08,185 INFO [train.py:968] (1/2) Epoch 18, batch 34900, giga_loss[loss=0.2792, simple_loss=0.3525, pruned_loss=0.103, over 28716.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3518, pruned_loss=0.09978, over 5662783.84 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3439, pruned_loss=0.1103, over 5692681.54 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3513, pruned_loss=0.09728, over 5672775.42 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:59:13,870 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 13:59:20,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3803, 1.7919, 1.3816, 1.4884], device='cuda:1'), covar=tensor([0.0762, 0.0346, 0.0346, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 13:59:50,695 INFO [train.py:968] (1/2) Epoch 18, batch 34950, giga_loss[loss=0.2162, simple_loss=0.2935, pruned_loss=0.06946, over 28956.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3461, pruned_loss=0.09747, over 5672608.38 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3441, pruned_loss=0.1103, over 5695616.84 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3456, pruned_loss=0.09536, over 5677685.61 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 13:59:51,885 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:00:36,329 INFO [train.py:968] (1/2) Epoch 18, batch 35000, giga_loss[loss=0.2394, simple_loss=0.3139, pruned_loss=0.08248, over 28906.00 frames. ], tot_loss[loss=0.264, simple_loss=0.339, pruned_loss=0.09451, over 5673780.86 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3444, pruned_loss=0.1105, over 5698138.90 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3383, pruned_loss=0.09229, over 5675408.57 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:00:48,697 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 35050, giga_loss[loss=0.2138, simple_loss=0.2919, pruned_loss=0.06779, over 28571.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3316, pruned_loss=0.09104, over 5684402.51 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3442, pruned_loss=0.1101, over 5704109.51 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3308, pruned_loss=0.08886, over 5679698.33 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:01:21,009 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:01:50,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:01:53,559 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:01:55,782 INFO [train.py:968] (1/2) Epoch 18, batch 35100, libri_loss[loss=0.321, simple_loss=0.3801, pruned_loss=0.131, over 29544.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3247, pruned_loss=0.08825, over 5684555.84 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3447, pruned_loss=0.1103, over 5708169.29 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3232, pruned_loss=0.08584, over 5676441.77 frames. ], batch size: 81, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:02:16,934 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=811847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:02:31,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5044, 1.7442, 1.5225, 1.7301], device='cuda:1'), covar=tensor([0.0770, 0.0316, 0.0328, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 14:02:37,600 INFO [train.py:968] (1/2) Epoch 18, batch 35150, giga_loss[loss=0.2055, simple_loss=0.2767, pruned_loss=0.06714, over 28554.00 frames. ], tot_loss[loss=0.245, simple_loss=0.319, pruned_loss=0.08548, over 5692215.01 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3445, pruned_loss=0.1098, over 5714344.90 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3171, pruned_loss=0.08305, over 5679256.85 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:02:47,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:02:49,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:02:58,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2372, 1.3403, 3.8847, 3.2813], device='cuda:1'), covar=tensor([0.1727, 0.2731, 0.0444, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0732, 0.0634, 0.0931, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:03:03,334 INFO [optim.py:369] (1/2) 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 14:03:13,867 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 35200, giga_loss[loss=0.243, simple_loss=0.3143, pruned_loss=0.08586, over 28298.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3165, pruned_loss=0.08456, over 5694012.26 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3447, pruned_loss=0.1099, over 5705612.86 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3142, pruned_loss=0.08204, over 5690319.93 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:03:19,225 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=811912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:23,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=811915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:42,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4296, 4.2514, 4.0046, 1.9911], device='cuda:1'), covar=tensor([0.0532, 0.0748, 0.0746, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.1076, 0.0919, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 14:03:52,025 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=811944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:03:58,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811954.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:04:04,195 INFO [train.py:968] (1/2) Epoch 18, batch 35250, giga_loss[loss=0.2071, simple_loss=0.2897, pruned_loss=0.06225, over 29015.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3128, pruned_loss=0.08283, over 5693469.19 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3451, pruned_loss=0.11, over 5706611.60 frames. ], giga_tot_loss[loss=0.2355, simple_loss=0.3102, pruned_loss=0.08036, over 5689735.04 frames. ], batch size: 136, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:04:05,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2528, 2.3285, 1.8290, 1.9529], device='cuda:1'), covar=tensor([0.0934, 0.0687, 0.0995, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0440, 0.0510, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:04:27,267 INFO [optim.py:369] (1/2) 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 14:04:28,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=811992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:04:44,475 INFO [train.py:968] (1/2) Epoch 18, batch 35300, giga_loss[loss=0.1997, simple_loss=0.2816, pruned_loss=0.05893, over 28900.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3115, pruned_loss=0.08272, over 5691510.85 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3452, pruned_loss=0.1098, over 5711926.64 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3083, pruned_loss=0.08002, over 5683019.28 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:04:52,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1400, 1.2712, 1.3308, 1.0900], device='cuda:1'), covar=tensor([0.2520, 0.2311, 0.1410, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1769, 0.1702, 0.1855], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 14:05:26,644 INFO [train.py:968] (1/2) Epoch 18, batch 35350, giga_loss[loss=0.1976, simple_loss=0.27, pruned_loss=0.06261, over 28885.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3094, pruned_loss=0.08186, over 5689971.89 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3452, pruned_loss=0.1096, over 5718653.12 frames. ], giga_tot_loss[loss=0.2318, simple_loss=0.3056, pruned_loss=0.079, over 5676187.45 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:05:49,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5736, 2.2368, 1.7984, 0.7895], device='cuda:1'), covar=tensor([0.5505, 0.3082, 0.4018, 0.6016], device='cuda:1'), in_proj_covar=tensor([0.1701, 0.1617, 0.1575, 0.1390], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 14:05:51,687 INFO [optim.py:369] (1/2) 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 14:05:56,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812097.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:05:59,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812100.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:06:09,098 INFO [train.py:968] (1/2) Epoch 18, batch 35400, giga_loss[loss=0.2233, simple_loss=0.2931, pruned_loss=0.07678, over 28977.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3058, pruned_loss=0.07998, over 5687333.13 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3455, pruned_loss=0.1099, over 5711372.80 frames. ], giga_tot_loss[loss=0.2282, simple_loss=0.3021, pruned_loss=0.07717, over 5682496.72 frames. ], batch size: 106, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:06:23,764 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812129.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:06:28,976 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:06:31,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:06:40,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4978, 4.3373, 4.0903, 1.8986], device='cuda:1'), covar=tensor([0.0533, 0.0717, 0.0704, 0.2145], device='cuda:1'), in_proj_covar=tensor([0.1161, 0.1079, 0.0920, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 14:06:49,577 INFO [train.py:968] (1/2) Epoch 18, batch 35450, giga_loss[loss=0.2561, simple_loss=0.3122, pruned_loss=0.09999, over 23912.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.3046, pruned_loss=0.07974, over 5670864.88 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.346, pruned_loss=0.11, over 5695704.36 frames. ], giga_tot_loss[loss=0.2267, simple_loss=0.3003, pruned_loss=0.07659, over 5680244.40 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:06:54,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:07:01,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4091, 4.2148, 3.9927, 1.8352], device='cuda:1'), covar=tensor([0.0555, 0.0810, 0.0750, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.1078, 0.0920, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 14:07:12,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3189, 1.3272, 1.1133, 1.5380], device='cuda:1'), covar=tensor([0.0735, 0.0337, 0.0350, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 14:07:15,255 INFO [optim.py:369] (1/2) 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 14:07:33,177 INFO [train.py:968] (1/2) Epoch 18, batch 35500, giga_loss[loss=0.1986, simple_loss=0.2745, pruned_loss=0.06132, over 28447.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3015, pruned_loss=0.07805, over 5675246.03 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3462, pruned_loss=0.11, over 5695983.74 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.2971, pruned_loss=0.07504, over 5681903.08 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:07:43,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:08:21,560 INFO [train.py:968] (1/2) Epoch 18, batch 35550, giga_loss[loss=0.1841, simple_loss=0.2613, pruned_loss=0.05342, over 28797.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2983, pruned_loss=0.07688, over 5670913.10 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.346, pruned_loss=0.1098, over 5698873.93 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2943, pruned_loss=0.07424, over 5673032.32 frames. ], batch size: 243, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:08:36,542 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:09:06,406 INFO [train.py:968] (1/2) Epoch 18, batch 35600, giga_loss[loss=0.2356, simple_loss=0.3106, pruned_loss=0.08028, over 29008.00 frames. ], tot_loss[loss=0.2292, simple_loss=0.3013, pruned_loss=0.07857, over 5676669.46 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3463, pruned_loss=0.1098, over 5701826.80 frames. ], giga_tot_loss[loss=0.2244, simple_loss=0.297, pruned_loss=0.07587, over 5675074.89 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:09:10,446 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=812316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:09:20,219 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=812326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:09:55,247 INFO [train.py:968] (1/2) Epoch 18, batch 35650, giga_loss[loss=0.2953, simple_loss=0.3678, pruned_loss=0.1114, over 28766.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3139, pruned_loss=0.08475, over 5668121.93 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3468, pruned_loss=0.1101, over 5685042.74 frames. ], giga_tot_loss[loss=0.2369, simple_loss=0.3096, pruned_loss=0.08208, over 5680527.08 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:09:59,941 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:10:01,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:10:20,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-09 14:10:24,548 INFO [optim.py:369] (1/2) 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 14:10:29,163 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 35700, giga_loss[loss=0.2786, simple_loss=0.3531, pruned_loss=0.102, over 28602.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3271, pruned_loss=0.09166, over 5670999.00 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3471, pruned_loss=0.1102, over 5688325.00 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.323, pruned_loss=0.08902, over 5677709.78 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:11:26,441 INFO [train.py:968] (1/2) Epoch 18, batch 35750, giga_loss[loss=0.2797, simple_loss=0.3438, pruned_loss=0.1078, over 23493.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3357, pruned_loss=0.09581, over 5661982.48 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3471, pruned_loss=0.1099, over 5678787.56 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3318, pruned_loss=0.09339, over 5675269.76 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:11:50,014 INFO [optim.py:369] (1/2) 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 14:11:52,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=812495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:12:06,059 INFO [train.py:968] (1/2) Epoch 18, batch 35800, giga_loss[loss=0.2513, simple_loss=0.3433, pruned_loss=0.07965, over 28865.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3418, pruned_loss=0.09787, over 5674138.00 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3476, pruned_loss=0.1103, over 5684320.64 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.338, pruned_loss=0.09523, over 5679693.58 frames. ], batch size: 285, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:12:52,092 INFO [train.py:968] (1/2) Epoch 18, batch 35850, giga_loss[loss=0.2474, simple_loss=0.3363, pruned_loss=0.07924, over 28824.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3442, pruned_loss=0.09799, over 5671890.87 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3482, pruned_loss=0.1105, over 5688161.36 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3405, pruned_loss=0.0953, over 5672705.19 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:13:23,558 INFO [optim.py:369] (1/2) 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 14:13:38,759 INFO [train.py:968] (1/2) Epoch 18, batch 35900, libri_loss[loss=0.2868, simple_loss=0.3586, pruned_loss=0.1075, over 29530.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3452, pruned_loss=0.09823, over 5673261.00 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3478, pruned_loss=0.1102, over 5692671.02 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3425, pruned_loss=0.09614, over 5669492.16 frames. ], batch size: 80, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:13:51,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7147, 1.9623, 1.5355, 2.0934], device='cuda:1'), covar=tensor([0.2498, 0.2551, 0.2796, 0.2256], device='cuda:1'), in_proj_covar=tensor([0.1431, 0.1043, 0.1275, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 14:14:14,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:14:21,670 INFO [train.py:968] (1/2) Epoch 18, batch 35950, giga_loss[loss=0.3494, simple_loss=0.4133, pruned_loss=0.1427, over 28494.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3474, pruned_loss=0.09972, over 5687124.20 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3481, pruned_loss=0.11, over 5698364.03 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.345, pruned_loss=0.09786, over 5678570.77 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:14:48,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:14:49,254 INFO [optim.py:369] (1/2) 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 14:14:56,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:15:04,613 INFO [train.py:968] (1/2) Epoch 18, batch 36000, libri_loss[loss=0.3164, simple_loss=0.3807, pruned_loss=0.1261, over 29565.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.35, pruned_loss=0.1018, over 5688549.89 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3483, pruned_loss=0.1102, over 5702542.66 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3478, pruned_loss=0.1001, over 5677601.16 frames. ], batch size: 83, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:15:04,613 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 14:15:14,356 INFO [train.py:1012] (1/2) Epoch 18, validation: loss=0.2061, simple_loss=0.3137, pruned_loss=0.04927, over 944034.00 frames. 
+2023-03-09 14:15:14,357 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 14:15:57,772 INFO [train.py:968] (1/2) Epoch 18, batch 36050, giga_loss[loss=0.3249, simple_loss=0.3934, pruned_loss=0.1282, over 28804.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3536, pruned_loss=0.1035, over 5696911.20 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3486, pruned_loss=0.11, over 5706062.72 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3518, pruned_loss=0.102, over 5684743.05 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:16:20,962 INFO [optim.py:369] (1/2) 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 14:16:24,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:26,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:36,208 INFO [train.py:968] (1/2) Epoch 18, batch 36100, giga_loss[loss=0.3071, simple_loss=0.3758, pruned_loss=0.1192, over 28888.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3551, pruned_loss=0.1032, over 5712161.34 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3487, pruned_loss=0.1101, over 5711464.10 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3537, pruned_loss=0.1018, over 5697484.19 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:16:50,519 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:55,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:16:58,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:04,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=812844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:07,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=812847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:20,954 INFO [train.py:968] (1/2) Epoch 18, batch 36150, giga_loss[loss=0.2847, simple_loss=0.3661, pruned_loss=0.1017, over 28619.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3563, pruned_loss=0.1032, over 5701560.64 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.349, pruned_loss=0.1102, over 5714736.43 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.355, pruned_loss=0.1018, over 5686889.24 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:17:24,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=812866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:27,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=812870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:17:32,103 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:18:02,007 INFO [train.py:968] (1/2) Epoch 18, batch 36200, giga_loss[loss=0.2943, simple_loss=0.367, pruned_loss=0.1108, over 28256.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3565, pruned_loss=0.1022, over 5710322.69 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3489, pruned_loss=0.1101, over 5720554.45 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3558, pruned_loss=0.1011, over 5693371.09 frames. ], batch size: 368, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:18:31,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-09 14:18:42,381 INFO [train.py:968] (1/2) Epoch 18, batch 36250, giga_loss[loss=0.2524, simple_loss=0.338, pruned_loss=0.08344, over 28865.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3551, pruned_loss=0.1004, over 5704076.21 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3494, pruned_loss=0.1105, over 5713971.70 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3542, pruned_loss=0.09902, over 5696091.00 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:19:03,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7462, 1.8271, 1.9855, 1.5303], device='cuda:1'), covar=tensor([0.2048, 0.2594, 0.1559, 0.1821], device='cuda:1'), in_proj_covar=tensor([0.0886, 0.0695, 0.0931, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 14:19:09,891 INFO [optim.py:369] (1/2) 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 14:19:25,390 INFO [train.py:968] (1/2) Epoch 18, batch 36300, giga_loss[loss=0.2507, simple_loss=0.3363, pruned_loss=0.08259, over 28268.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3532, pruned_loss=0.09861, over 5703331.07 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3493, pruned_loss=0.1104, over 5716893.47 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3526, pruned_loss=0.09734, over 5693994.17 frames. ], batch size: 77, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:19:26,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=813013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:19:28,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=813016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:19:50,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=813045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:19:56,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-09 14:20:02,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=813061.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:20:02,361 INFO [train.py:968] (1/2) Epoch 18, batch 36350, giga_loss[loss=0.2867, simple_loss=0.3688, pruned_loss=0.1023, over 28876.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3544, pruned_loss=0.1002, over 5677242.69 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3504, pruned_loss=0.111, over 5696246.22 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3531, pruned_loss=0.09814, over 5687441.54 frames. ], batch size: 145, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:20:29,376 INFO [optim.py:369] (1/2) 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 14:20:47,621 INFO [train.py:968] (1/2) Epoch 18, batch 36400, giga_loss[loss=0.3053, simple_loss=0.3704, pruned_loss=0.1201, over 28656.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3563, pruned_loss=0.1035, over 5662267.02 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3504, pruned_loss=0.111, over 5683207.60 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3554, pruned_loss=0.1016, over 5681591.53 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:21:01,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2480, 1.2502, 3.8854, 3.2694], device='cuda:1'), covar=tensor([0.1747, 0.2859, 0.0460, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0630, 0.0924, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:21:30,112 INFO [train.py:968] (1/2) Epoch 18, batch 36450, giga_loss[loss=0.3188, simple_loss=0.3784, pruned_loss=0.1296, over 29025.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3597, pruned_loss=0.108, over 5672395.19 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3505, pruned_loss=0.111, over 5685437.46 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3591, pruned_loss=0.1062, over 5685882.52 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:21:59,332 INFO [optim.py:369] (1/2) 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 14:22:14,930 INFO [train.py:968] (1/2) Epoch 18, batch 36500, giga_loss[loss=0.2454, simple_loss=0.3211, pruned_loss=0.08488, over 28510.00 frames. ], tot_loss[loss=0.2883, simple_loss=0.3589, pruned_loss=0.1089, over 5666584.28 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.351, pruned_loss=0.1113, over 5679906.70 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3582, pruned_loss=0.1071, over 5682594.69 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:22:56,650 INFO [train.py:968] (1/2) Epoch 18, batch 36550, giga_loss[loss=0.2615, simple_loss=0.3333, pruned_loss=0.09482, over 28958.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3567, pruned_loss=0.1078, over 5675584.61 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3511, pruned_loss=0.1112, over 5676909.59 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3562, pruned_loss=0.1065, over 5690109.70 frames. ], batch size: 106, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:23:23,971 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 36600, giga_loss[loss=0.3409, simple_loss=0.3896, pruned_loss=0.1461, over 28921.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3558, pruned_loss=0.1072, over 5685536.19 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3516, pruned_loss=0.1112, over 5683319.52 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.355, pruned_loss=0.106, over 5691114.63 frames. ], batch size: 112, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:24:14,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0839, 1.1288, 3.7890, 3.2060], device='cuda:1'), covar=tensor([0.1760, 0.2921, 0.0392, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0629, 0.0925, 0.0864], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:24:20,879 INFO [train.py:968] (1/2) Epoch 18, batch 36650, giga_loss[loss=0.2749, simple_loss=0.3484, pruned_loss=0.1007, over 29067.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3542, pruned_loss=0.1058, over 5691464.09 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3519, pruned_loss=0.1113, over 5689541.82 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3534, pruned_loss=0.1046, over 5690459.01 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:24:53,265 INFO [optim.py:369] (1/2) 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 14:24:53,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4184, 4.0637, 1.6025, 1.6060], device='cuda:1'), covar=tensor([0.1003, 0.0217, 0.0902, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0536, 0.0368, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 14:25:08,139 INFO [train.py:968] (1/2) Epoch 18, batch 36700, giga_loss[loss=0.2739, simple_loss=0.3491, pruned_loss=0.09934, over 28946.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3513, pruned_loss=0.1031, over 5700738.65 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.352, pruned_loss=0.1114, over 5692804.52 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3506, pruned_loss=0.102, over 5697069.51 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:25:31,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=813436.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:25:55,396 INFO [train.py:968] (1/2) Epoch 18, batch 36750, giga_loss[loss=0.2571, simple_loss=0.3274, pruned_loss=0.09339, over 28876.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3465, pruned_loss=0.1003, over 5695992.86 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3521, pruned_loss=0.1113, over 5695560.31 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3458, pruned_loss=0.09932, over 5690924.33 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:26:31,079 INFO [optim.py:369] (1/2) 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 14:26:49,682 INFO [train.py:968] (1/2) Epoch 18, batch 36800, giga_loss[loss=0.2259, simple_loss=0.3048, pruned_loss=0.07354, over 28839.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.339, pruned_loss=0.09605, over 5692917.26 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.352, pruned_loss=0.1113, over 5696759.51 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3384, pruned_loss=0.09518, over 5687865.64 frames. ], batch size: 243, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:27:45,184 INFO [train.py:968] (1/2) Epoch 18, batch 36850, giga_loss[loss=0.2553, simple_loss=0.3342, pruned_loss=0.08825, over 28973.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3339, pruned_loss=0.09351, over 5682845.94 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3523, pruned_loss=0.1114, over 5699875.96 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3329, pruned_loss=0.09246, over 5675639.83 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:27:50,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 14:27:55,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6359, 1.6603, 1.8810, 1.4163], device='cuda:1'), covar=tensor([0.1757, 0.2486, 0.1401, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0696, 0.0929, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 14:28:01,305 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=813579.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:28:01,426 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-09 14:28:03,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=813582.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:28:14,024 INFO [optim.py:369] (1/2) 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 14:28:14,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1866, 1.2261, 3.8849, 3.1649], device='cuda:1'), covar=tensor([0.1810, 0.2890, 0.0431, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0724, 0.0628, 0.0922, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:28:29,489 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 36900, giga_loss[loss=0.2642, simple_loss=0.3349, pruned_loss=0.09679, over 28758.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3346, pruned_loss=0.0935, over 5677916.29 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3524, pruned_loss=0.1114, over 5697962.58 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3335, pruned_loss=0.09238, over 5673627.16 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:29:02,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5969, 1.8682, 1.5246, 1.7672], device='cuda:1'), covar=tensor([0.2568, 0.2582, 0.2789, 0.2240], device='cuda:1'), in_proj_covar=tensor([0.1437, 0.1045, 0.1279, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 14:29:09,005 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-09 14:29:10,472 INFO [train.py:968] (1/2) Epoch 18, batch 36950, giga_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.09847, over 28684.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3353, pruned_loss=0.0933, over 5692824.41 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3524, pruned_loss=0.1112, over 5702367.03 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3338, pruned_loss=0.09205, over 5684960.99 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:29:36,409 INFO [optim.py:369] (1/2) 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 14:29:53,321 INFO [train.py:968] (1/2) Epoch 18, batch 37000, giga_loss[loss=0.2483, simple_loss=0.3224, pruned_loss=0.08714, over 29069.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3367, pruned_loss=0.09474, over 5697671.35 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3529, pruned_loss=0.1113, over 5706815.09 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3348, pruned_loss=0.09334, over 5687395.83 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:30:32,996 INFO [train.py:968] (1/2) Epoch 18, batch 37050, giga_loss[loss=0.2624, simple_loss=0.3437, pruned_loss=0.09051, over 28626.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3346, pruned_loss=0.09374, over 5696748.28 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3534, pruned_loss=0.1115, over 5706604.68 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3324, pruned_loss=0.09214, over 5688370.38 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:31:00,381 INFO [optim.py:369] (1/2) 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 14:31:15,662 INFO [train.py:968] (1/2) Epoch 18, batch 37100, giga_loss[loss=0.2335, simple_loss=0.3084, pruned_loss=0.07928, over 28739.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.332, pruned_loss=0.0923, over 5699481.22 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3536, pruned_loss=0.1116, over 5698789.42 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3299, pruned_loss=0.09081, over 5700114.64 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:31:49,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4554, 3.6632, 1.6712, 1.7518], device='cuda:1'), covar=tensor([0.1014, 0.0310, 0.0825, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0537, 0.0370, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 14:31:54,675 INFO [train.py:968] (1/2) Epoch 18, batch 37150, giga_loss[loss=0.2499, simple_loss=0.3165, pruned_loss=0.09162, over 28483.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3299, pruned_loss=0.09113, over 5710143.08 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3542, pruned_loss=0.1117, over 5704327.97 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.327, pruned_loss=0.0893, over 5705713.46 frames. ], batch size: 71, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:31:55,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=813863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:32:19,978 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3347, 1.3383, 3.8896, 3.2523], device='cuda:1'), covar=tensor([0.1705, 0.2841, 0.0426, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0633, 0.0927, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:32:21,136 INFO [optim.py:369] (1/2) 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 14:32:35,474 INFO [train.py:968] (1/2) Epoch 18, batch 37200, giga_loss[loss=0.2295, simple_loss=0.3065, pruned_loss=0.07623, over 28741.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3284, pruned_loss=0.09053, over 5702167.00 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3548, pruned_loss=0.1118, over 5699971.00 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3252, pruned_loss=0.08865, over 5702966.73 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:33:17,176 INFO [train.py:968] (1/2) Epoch 18, batch 37250, giga_loss[loss=0.2407, simple_loss=0.3139, pruned_loss=0.08374, over 28754.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3251, pruned_loss=0.08858, over 5707534.21 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3548, pruned_loss=0.1116, over 5704258.64 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.322, pruned_loss=0.08688, over 5704338.76 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:33:45,194 INFO [optim.py:369] (1/2) 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 14:33:56,076 INFO [train.py:968] (1/2) Epoch 18, batch 37300, giga_loss[loss=0.207, simple_loss=0.2854, pruned_loss=0.06426, over 28460.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3242, pruned_loss=0.08802, over 5716629.82 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3556, pruned_loss=0.1118, over 5707082.75 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3199, pruned_loss=0.08567, over 5711955.72 frames. ], batch size: 65, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:34:06,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4665, 1.7328, 1.3791, 1.4119], device='cuda:1'), covar=tensor([0.2578, 0.2643, 0.2980, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1444, 0.1053, 0.1283, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 14:34:35,881 INFO [train.py:968] (1/2) Epoch 18, batch 37350, giga_loss[loss=0.2392, simple_loss=0.3067, pruned_loss=0.08579, over 28832.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.323, pruned_loss=0.08759, over 5721399.38 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3567, pruned_loss=0.1123, over 5707284.17 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3181, pruned_loss=0.08486, over 5717365.72 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:35:04,054 INFO [optim.py:369] (1/2) 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 14:35:14,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:35:15,442 INFO [train.py:968] (1/2) Epoch 18, batch 37400, giga_loss[loss=0.2417, simple_loss=0.3134, pruned_loss=0.08502, over 28921.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3223, pruned_loss=0.08731, over 5730672.98 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3576, pruned_loss=0.1127, over 5713307.59 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3167, pruned_loss=0.08411, over 5722518.40 frames. ], batch size: 145, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:35:30,855 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814129.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:35:57,024 INFO [train.py:968] (1/2) Epoch 18, batch 37450, libri_loss[loss=0.3525, simple_loss=0.4088, pruned_loss=0.1481, over 20089.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3237, pruned_loss=0.08858, over 5712331.75 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3585, pruned_loss=0.1132, over 5700458.75 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3173, pruned_loss=0.08472, over 5718156.69 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:36:02,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-09 14:36:26,643 INFO [optim.py:369] (1/2) 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 14:36:32,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6774, 1.9337, 1.3534, 1.4806], device='cuda:1'), covar=tensor([0.1049, 0.0651, 0.1117, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0444, 0.0513, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:36:37,823 INFO [train.py:968] (1/2) Epoch 18, batch 37500, giga_loss[loss=0.296, simple_loss=0.3655, pruned_loss=0.1132, over 28529.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3286, pruned_loss=0.09166, over 5709528.75 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3587, pruned_loss=0.1133, over 5702745.11 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3227, pruned_loss=0.0881, over 5712362.52 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:37:00,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:37:01,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-09 14:37:11,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:37:23,612 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-09 14:37:24,461 INFO [train.py:968] (1/2) Epoch 18, batch 37550, giga_loss[loss=0.2767, simple_loss=0.3518, pruned_loss=0.1008, over 29068.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.334, pruned_loss=0.09508, over 5707558.44 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3589, pruned_loss=0.1134, over 5705825.15 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3286, pruned_loss=0.09183, over 5707230.65 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:37:57,933 INFO [optim.py:369] (1/2) 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 14:38:01,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4258, 4.2765, 4.0321, 1.9449], device='cuda:1'), covar=tensor([0.0567, 0.0697, 0.0694, 0.2100], device='cuda:1'), in_proj_covar=tensor([0.1161, 0.1079, 0.0921, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 14:38:02,319 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:38:04,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2512, 1.2910, 1.0828, 0.9560], device='cuda:1'), covar=tensor([0.0813, 0.0435, 0.0984, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0443, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:38:10,711 INFO [train.py:968] (1/2) Epoch 18, batch 37600, giga_loss[loss=0.2896, simple_loss=0.3648, pruned_loss=0.1072, over 28640.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3425, pruned_loss=0.101, over 5699152.89 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3594, pruned_loss=0.1139, over 5707091.39 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3371, pruned_loss=0.09758, over 5697749.30 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:38:51,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8142, 1.9718, 1.3127, 1.5907], device='cuda:1'), covar=tensor([0.0934, 0.0656, 0.1113, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0443, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:39:00,216 INFO [train.py:968] (1/2) Epoch 18, batch 37650, giga_loss[loss=0.2632, simple_loss=0.3489, pruned_loss=0.08869, over 28850.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3465, pruned_loss=0.1028, over 5665533.55 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3596, pruned_loss=0.114, over 5691946.77 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3419, pruned_loss=0.09981, over 5677911.70 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:39:14,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8459, 3.6569, 3.4798, 1.6366], device='cuda:1'), covar=tensor([0.0652, 0.0799, 0.0751, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.1157, 0.1074, 0.0919, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 14:39:16,180 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:39:19,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:39:29,939 INFO [optim.py:369] (1/2) 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 14:39:45,327 INFO [train.py:968] (1/2) Epoch 18, batch 37700, giga_loss[loss=0.2848, simple_loss=0.3635, pruned_loss=0.1031, over 28124.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.35, pruned_loss=0.1036, over 5670834.89 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3601, pruned_loss=0.1142, over 5696085.49 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3457, pruned_loss=0.1008, over 5676643.41 frames. ], batch size: 77, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:39:46,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:40:32,336 INFO [train.py:968] (1/2) Epoch 18, batch 37750, giga_loss[loss=0.3002, simple_loss=0.3794, pruned_loss=0.1105, over 28956.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.355, pruned_loss=0.1066, over 5671193.49 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3597, pruned_loss=0.1141, over 5700363.40 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3518, pruned_loss=0.1043, over 5671712.93 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:40:53,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:41:02,864 INFO [optim.py:369] (1/2) 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 14:41:08,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814504.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:41:14,010 INFO [train.py:968] (1/2) Epoch 18, batch 37800, libri_loss[loss=0.3136, simple_loss=0.3742, pruned_loss=0.1265, over 29737.00 frames. ], tot_loss[loss=0.2861, simple_loss=0.357, pruned_loss=0.1076, over 5665384.89 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3602, pruned_loss=0.1144, over 5698374.21 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3538, pruned_loss=0.1052, over 5666473.85 frames. ], batch size: 87, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:41:56,247 INFO [train.py:968] (1/2) Epoch 18, batch 37850, giga_loss[loss=0.3073, simple_loss=0.3872, pruned_loss=0.1137, over 28940.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.353, pruned_loss=0.1041, over 5678731.12 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3605, pruned_loss=0.1145, over 5700448.65 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.35, pruned_loss=0.1018, over 5677254.60 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:42:26,779 INFO [optim.py:369] (1/2) 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 14:42:39,388 INFO [train.py:968] (1/2) Epoch 18, batch 37900, libri_loss[loss=0.2503, simple_loss=0.3208, pruned_loss=0.08985, over 29502.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.352, pruned_loss=0.1028, over 5688289.54 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3607, pruned_loss=0.1149, over 5705414.39 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3492, pruned_loss=0.1004, over 5682228.81 frames. ], batch size: 70, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:42:47,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:42:52,169 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:42:56,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:10,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814647.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:43:11,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814650.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:43:13,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:20,005 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:21,239 INFO [train.py:968] (1/2) Epoch 18, batch 37950, giga_loss[loss=0.2903, simple_loss=0.3599, pruned_loss=0.1104, over 28764.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.351, pruned_loss=0.1017, over 5684105.27 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3606, pruned_loss=0.1148, over 5705634.10 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3487, pruned_loss=0.0997, over 5678833.25 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:43:33,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=814677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:43:35,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814679.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:43:41,202 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814685.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:43:45,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814691.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:43:51,996 INFO [optim.py:369] (1/2) 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 14:43:54,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-09 14:44:05,513 INFO [train.py:968] (1/2) Epoch 18, batch 38000, giga_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.09291, over 28720.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3542, pruned_loss=0.1038, over 5687323.74 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.361, pruned_loss=0.115, over 5708863.53 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3518, pruned_loss=0.1017, over 5680074.56 frames. ], batch size: 262, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:44:16,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0651, 2.3834, 1.9041, 2.4164], device='cuda:1'), covar=tensor([0.2314, 0.2344, 0.2677, 0.2223], device='cuda:1'), in_proj_covar=tensor([0.1441, 0.1046, 0.1280, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 14:44:47,443 INFO [train.py:968] (1/2) Epoch 18, batch 38050, giga_loss[loss=0.2564, simple_loss=0.3411, pruned_loss=0.08582, over 28914.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3567, pruned_loss=0.1059, over 5696001.00 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3611, pruned_loss=0.115, over 5715618.39 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3545, pruned_loss=0.1038, over 5683413.20 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:44:51,419 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:44:51,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=814766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:44:53,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:45:18,812 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:45:19,314 INFO [optim.py:369] (1/2) 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 14:45:29,695 INFO [train.py:968] (1/2) Epoch 18, batch 38100, giga_loss[loss=0.3323, simple_loss=0.3959, pruned_loss=0.1344, over 28731.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3585, pruned_loss=0.1071, over 5691555.69 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3617, pruned_loss=0.1153, over 5708026.76 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3561, pruned_loss=0.1051, over 5686979.15 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:45:39,271 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=814820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:45:43,254 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=814823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:46:07,848 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=814852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:46:16,302 INFO [train.py:968] (1/2) Epoch 18, batch 38150, libri_loss[loss=0.3555, simple_loss=0.4048, pruned_loss=0.1531, over 19787.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3592, pruned_loss=0.1078, over 5685474.65 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3622, pruned_loss=0.1155, over 5698318.73 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3568, pruned_loss=0.1058, over 5690918.72 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:46:47,407 INFO [optim.py:369] (1/2) 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 14:46:57,683 INFO [train.py:968] (1/2) Epoch 18, batch 38200, giga_loss[loss=0.3036, simple_loss=0.3723, pruned_loss=0.1175, over 28875.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3587, pruned_loss=0.1081, over 5688339.43 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3619, pruned_loss=0.1154, over 5701291.73 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.357, pruned_loss=0.1064, over 5689627.66 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:47:22,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7157, 4.5322, 4.2663, 2.0777], device='cuda:1'), covar=tensor([0.0442, 0.0600, 0.0627, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.1159, 0.1082, 0.0921, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 14:47:40,586 INFO [train.py:968] (1/2) Epoch 18, batch 38250, giga_loss[loss=0.3189, simple_loss=0.3788, pruned_loss=0.1295, over 28944.00 frames. ], tot_loss[loss=0.2879, simple_loss=0.3594, pruned_loss=0.1082, over 5696204.13 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3619, pruned_loss=0.1152, over 5706592.55 frames. ], giga_tot_loss[loss=0.2859, simple_loss=0.358, pruned_loss=0.1069, over 5692520.28 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:47:43,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1787, 1.2523, 3.8983, 3.2099], device='cuda:1'), covar=tensor([0.1657, 0.2676, 0.0450, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0627, 0.0925, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:48:08,425 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 38300, libri_loss[loss=0.2445, simple_loss=0.3137, pruned_loss=0.08765, over 28252.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3578, pruned_loss=0.1058, over 5698380.41 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3621, pruned_loss=0.1154, over 5701649.78 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3564, pruned_loss=0.1044, over 5700135.89 frames. ], batch size: 62, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:48:31,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:49:00,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815060.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:49:01,203 INFO [train.py:968] (1/2) Epoch 18, batch 38350, giga_loss[loss=0.3532, simple_loss=0.3907, pruned_loss=0.1578, over 26703.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3584, pruned_loss=0.1055, over 5699897.01 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3626, pruned_loss=0.1157, over 5705309.08 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3568, pruned_loss=0.1039, over 5698099.43 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:49:02,158 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815063.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:49:04,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815066.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:49:05,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7830, 4.9119, 1.9836, 2.0507], device='cuda:1'), covar=tensor([0.0971, 0.0193, 0.0880, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0535, 0.0368, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 14:49:10,308 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:49:29,291 INFO [optim.py:369] (1/2) 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 14:49:42,401 INFO [train.py:968] (1/2) Epoch 18, batch 38400, giga_loss[loss=0.306, simple_loss=0.3707, pruned_loss=0.1207, over 27941.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3577, pruned_loss=0.105, over 5701328.24 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3628, pruned_loss=0.1159, over 5699293.64 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3561, pruned_loss=0.1034, over 5704799.10 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:49:58,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:00,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-09 14:50:04,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:09,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9432, 2.0644, 1.5197, 1.5086], device='cuda:1'), covar=tensor([0.0882, 0.0572, 0.0980, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0442, 0.0511, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:50:10,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5310, 1.6447, 1.5574, 1.3955], device='cuda:1'), covar=tensor([0.2333, 0.2220, 0.1936, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.1892, 0.1803, 0.1746, 0.1892], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 14:50:23,052 INFO [train.py:968] (1/2) Epoch 18, batch 38450, giga_loss[loss=0.2326, simple_loss=0.3199, pruned_loss=0.0727, over 28630.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3565, pruned_loss=0.1047, over 5698070.53 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3636, pruned_loss=0.1164, over 5696276.63 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3544, pruned_loss=0.1027, over 5703302.34 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:50:28,468 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:30,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:51,937 INFO [optim.py:369] (1/2) 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 14:50:55,419 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:50:57,704 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815203.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:51:00,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815206.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:51:02,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 38500, giga_loss[loss=0.214, simple_loss=0.3008, pruned_loss=0.06358, over 28569.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3526, pruned_loss=0.1025, over 5701268.36 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.363, pruned_loss=0.1162, over 5690969.97 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1005, over 5709928.08 frames. ], batch size: 60, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 14:51:04,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815212.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:51:21,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815235.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:51:25,854 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 38550, giga_loss[loss=0.258, simple_loss=0.3342, pruned_loss=0.09089, over 28568.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3516, pruned_loss=0.102, over 5698626.12 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3634, pruned_loss=0.1165, over 5683653.28 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3498, pruned_loss=0.09983, over 5712281.18 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:52:01,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:52:05,453 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:52:25,377 INFO [train.py:968] (1/2) Epoch 18, batch 38600, giga_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1044, over 28855.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3526, pruned_loss=0.1032, over 5693994.29 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3634, pruned_loss=0.1165, over 5678179.60 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.351, pruned_loss=0.1013, over 5709884.44 frames. ], batch size: 106, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:52:28,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:53:01,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5610, 1.7234, 1.3058, 1.3052], device='cuda:1'), covar=tensor([0.1023, 0.0645, 0.1073, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0442, 0.0511, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 14:53:04,217 INFO [train.py:968] (1/2) Epoch 18, batch 38650, giga_loss[loss=0.2693, simple_loss=0.3477, pruned_loss=0.09549, over 28858.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3521, pruned_loss=0.1022, over 5693683.58 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3632, pruned_loss=0.1163, over 5672431.19 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3508, pruned_loss=0.1005, over 5712401.92 frames. ], batch size: 145, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:53:08,845 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:53:25,310 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:53:32,625 INFO [optim.py:369] (1/2) 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 14:53:41,886 INFO [train.py:968] (1/2) Epoch 18, batch 38700, giga_loss[loss=0.2665, simple_loss=0.3429, pruned_loss=0.09504, over 28970.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3511, pruned_loss=0.101, over 5694942.95 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1163, over 5678481.67 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3496, pruned_loss=0.0992, over 5705591.22 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:54:01,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815438.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:54:08,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:54:19,750 INFO [train.py:968] (1/2) Epoch 18, batch 38750, giga_loss[loss=0.257, simple_loss=0.3313, pruned_loss=0.09138, over 28243.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3508, pruned_loss=0.1008, over 5704278.75 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3635, pruned_loss=0.1166, over 5680597.40 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3491, pruned_loss=0.09869, over 5711260.38 frames. ], batch size: 77, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 14:54:53,116 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815501.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:54:58,022 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 38800, giga_loss[loss=0.2595, simple_loss=0.3429, pruned_loss=0.08803, over 28503.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3501, pruned_loss=0.1008, over 5707678.39 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3634, pruned_loss=0.1165, over 5685685.13 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3484, pruned_loss=0.09883, over 5708966.59 frames. ], batch size: 65, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:55:04,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5275, 3.3006, 1.5663, 1.7006], device='cuda:1'), covar=tensor([0.0959, 0.0271, 0.0887, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0533, 0.0368, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 14:55:04,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 14:55:18,072 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:55:43,377 INFO [train.py:968] (1/2) Epoch 18, batch 38850, giga_loss[loss=0.2402, simple_loss=0.3178, pruned_loss=0.08133, over 28300.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3463, pruned_loss=0.09884, over 5706076.35 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3633, pruned_loss=0.1164, over 5688430.07 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.345, pruned_loss=0.09716, over 5704839.86 frames. ], batch size: 65, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:55:59,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815581.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:56:01,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815584.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 14:56:06,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:08,740 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 18, batch 38900, giga_loss[loss=0.2389, simple_loss=0.3229, pruned_loss=0.0774, over 28568.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3443, pruned_loss=0.09799, over 5705400.95 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3628, pruned_loss=0.1159, over 5687108.02 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.343, pruned_loss=0.09628, over 5706176.12 frames. ], batch size: 65, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:56:22,053 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815613.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 14:56:30,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:30,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5194, 1.6439, 1.5771, 1.3943], device='cuda:1'), covar=tensor([0.2561, 0.2478, 0.1946, 0.2358], device='cuda:1'), in_proj_covar=tensor([0.1894, 0.1816, 0.1752, 0.1893], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 14:56:49,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:51,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:56:58,339 INFO [train.py:968] (1/2) Epoch 18, batch 38950, giga_loss[loss=0.2434, simple_loss=0.3237, pruned_loss=0.08154, over 28820.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3441, pruned_loss=0.09816, over 5709506.99 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3629, pruned_loss=0.116, over 5695420.37 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3421, pruned_loss=0.09598, over 5703568.09 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:57:09,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7649, 4.9099, 1.8183, 2.1381], device='cuda:1'), covar=tensor([0.0817, 0.0412, 0.0828, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0535, 0.0368, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 14:57:14,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:57:29,352 INFO [optim.py:369] (1/2) 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 14:57:38,268 INFO [train.py:968] (1/2) Epoch 18, batch 39000, giga_loss[loss=0.3269, simple_loss=0.3815, pruned_loss=0.1361, over 28597.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3448, pruned_loss=0.09921, over 5699062.98 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.363, pruned_loss=0.1161, over 5691795.05 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3425, pruned_loss=0.09681, over 5697591.31 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:57:38,268 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 14:57:46,798 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 14:58:11,649 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 39050, giga_loss[loss=0.2594, simple_loss=0.3307, pruned_loss=0.09408, over 28610.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3419, pruned_loss=0.09764, over 5705844.98 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3629, pruned_loss=0.1161, over 5696891.51 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3397, pruned_loss=0.09533, over 5700240.79 frames. ], batch size: 85, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:58:28,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 14:58:42,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 14:58:55,508 INFO [optim.py:369] (1/2) 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 14:59:03,684 INFO [train.py:968] (1/2) Epoch 18, batch 39100, libri_loss[loss=0.3236, simple_loss=0.3943, pruned_loss=0.1265, over 29670.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3393, pruned_loss=0.09643, over 5716232.04 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3629, pruned_loss=0.1161, over 5703556.13 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.337, pruned_loss=0.09405, over 5706001.28 frames. ], batch size: 91, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:59:11,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5958, 3.6426, 1.6061, 1.7738], device='cuda:1'), covar=tensor([0.0889, 0.0385, 0.0874, 0.1218], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0534, 0.0368, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 14:59:38,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5275, 2.7316, 1.6314, 1.6237], device='cuda:1'), covar=tensor([0.0736, 0.0309, 0.0711, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0534, 0.0367, 0.0411], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0023, 0.0028], device='cuda:1')
+2023-03-09 14:59:42,982 INFO [train.py:968] (1/2) Epoch 18, batch 39150, libri_loss[loss=0.2439, simple_loss=0.3137, pruned_loss=0.0871, over 29496.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3377, pruned_loss=0.09591, over 5719940.32 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3622, pruned_loss=0.1155, over 5710510.15 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3355, pruned_loss=0.09377, over 5705827.65 frames. ], batch size: 70, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 14:59:54,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815876.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:00:04,093 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:05,849 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:00:19,941 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=815906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:20,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=815906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:22,145 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=815909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:24,267 INFO [train.py:968] (1/2) Epoch 18, batch 39200, giga_loss[loss=0.2775, simple_loss=0.3613, pruned_loss=0.09685, over 28937.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3359, pruned_loss=0.09477, over 5719260.42 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3623, pruned_loss=0.1155, over 5713290.70 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3337, pruned_loss=0.09281, over 5705607.88 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:00:24,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=815912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:31,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:50,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=815941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:00:56,857 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-09 15:01:10,128 INFO [train.py:968] (1/2) Epoch 18, batch 39250, giga_loss[loss=0.2663, simple_loss=0.3344, pruned_loss=0.09915, over 28714.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3375, pruned_loss=0.09565, over 5715433.85 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3625, pruned_loss=0.1156, over 5715251.88 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3354, pruned_loss=0.09383, over 5702917.84 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:01:28,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8014, 4.6131, 4.4284, 1.9866], device='cuda:1'), covar=tensor([0.0535, 0.0748, 0.0851, 0.1945], device='cuda:1'), in_proj_covar=tensor([0.1157, 0.1077, 0.0919, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 15:01:43,071 INFO [optim.py:369] (1/2) 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 15:01:52,680 INFO [train.py:968] (1/2) Epoch 18, batch 39300, giga_loss[loss=0.2318, simple_loss=0.3236, pruned_loss=0.07, over 28878.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3414, pruned_loss=0.09762, over 5711538.20 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3631, pruned_loss=0.116, over 5718992.50 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3383, pruned_loss=0.09516, over 5697951.21 frames. ], batch size: 145, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:01:58,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:02:00,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816022.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:02:20,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2190, 1.1151, 4.1398, 3.4360], device='cuda:1'), covar=tensor([0.2108, 0.3299, 0.0696, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0723, 0.0625, 0.0921, 0.0860], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:02:27,388 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:02:29,616 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816051.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:02:30,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:02:36,434 INFO [train.py:968] (1/2) Epoch 18, batch 39350, giga_loss[loss=0.2796, simple_loss=0.3537, pruned_loss=0.1027, over 28847.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3451, pruned_loss=0.09921, over 5708624.20 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3627, pruned_loss=0.1158, over 5722725.48 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3424, pruned_loss=0.09696, over 5694332.99 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:02:54,501 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:03:11,243 INFO [optim.py:369] (1/2) 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 15:03:18,637 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=816105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:03:23,833 INFO [train.py:968] (1/2) Epoch 18, batch 39400, giga_loss[loss=0.2956, simple_loss=0.359, pruned_loss=0.1161, over 26782.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3464, pruned_loss=0.09948, over 5699811.15 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3626, pruned_loss=0.1157, over 5722691.17 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3442, pruned_loss=0.0976, over 5688395.43 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:04:04,869 INFO [train.py:968] (1/2) Epoch 18, batch 39450, giga_loss[loss=0.2437, simple_loss=0.3229, pruned_loss=0.0823, over 29036.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3449, pruned_loss=0.098, over 5709706.41 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3628, pruned_loss=0.1158, over 5728454.55 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3424, pruned_loss=0.09576, over 5694687.37 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:04:08,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4887, 1.6751, 1.7126, 1.2662], device='cuda:1'), covar=tensor([0.1867, 0.2443, 0.1569, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.0878, 0.0693, 0.0923, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 15:04:36,801 INFO [optim.py:369] (1/2) 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 15:04:45,500 INFO [train.py:968] (1/2) Epoch 18, batch 39500, giga_loss[loss=0.2875, simple_loss=0.3497, pruned_loss=0.1127, over 28434.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3442, pruned_loss=0.09783, over 5705140.07 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3629, pruned_loss=0.1159, over 5721989.44 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3418, pruned_loss=0.0957, over 5698334.05 frames. ], batch size: 71, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:05:17,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4927, 1.8211, 1.4095, 1.5761], device='cuda:1'), covar=tensor([0.2549, 0.2542, 0.3082, 0.2374], device='cuda:1'), in_proj_covar=tensor([0.1440, 0.1045, 0.1277, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 15:05:18,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4205, 1.6738, 1.3241, 1.6457], device='cuda:1'), covar=tensor([0.2580, 0.2577, 0.3053, 0.2298], device='cuda:1'), in_proj_covar=tensor([0.1439, 0.1045, 0.1277, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 15:05:20,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8176, 1.9513, 2.0026, 1.5690], device='cuda:1'), covar=tensor([0.1891, 0.2196, 0.1487, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.0880, 0.0694, 0.0925, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 15:05:28,820 INFO [train.py:968] (1/2) Epoch 18, batch 39550, giga_loss[loss=0.2718, simple_loss=0.3483, pruned_loss=0.09761, over 28916.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3446, pruned_loss=0.09805, over 5709494.37 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.363, pruned_loss=0.116, over 5722987.28 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.342, pruned_loss=0.09564, over 5702722.19 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:05:42,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=816281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:06:01,069 INFO [optim.py:369] (1/2) 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 15:06:09,262 INFO [train.py:968] (1/2) Epoch 18, batch 39600, giga_loss[loss=0.2772, simple_loss=0.3513, pruned_loss=0.1015, over 28905.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3445, pruned_loss=0.09783, over 5719588.91 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3632, pruned_loss=0.1161, over 5725482.68 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3418, pruned_loss=0.09533, over 5711625.13 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:06:28,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2506, 1.3637, 1.2776, 1.2348], device='cuda:1'), covar=tensor([0.2140, 0.2020, 0.1559, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.1900, 0.1817, 0.1760, 0.1897], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 15:06:48,823 INFO [train.py:968] (1/2) Epoch 18, batch 39650, giga_loss[loss=0.2747, simple_loss=0.3589, pruned_loss=0.09529, over 28731.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3483, pruned_loss=0.09982, over 5719194.01 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3631, pruned_loss=0.1158, over 5730242.42 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3456, pruned_loss=0.09745, over 5708071.24 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:07:20,669 INFO [optim.py:369] (1/2) 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 15:07:28,590 INFO [train.py:968] (1/2) Epoch 18, batch 39700, giga_loss[loss=0.2589, simple_loss=0.3414, pruned_loss=0.08823, over 29089.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3514, pruned_loss=0.1015, over 5720346.33 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3634, pruned_loss=0.116, over 5734273.47 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3487, pruned_loss=0.09916, over 5707696.29 frames. ], batch size: 128, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:07:39,377 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:07:41,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:07:59,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1803, 1.1659, 3.6213, 3.0375], device='cuda:1'), covar=tensor([0.1677, 0.2843, 0.0485, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0728, 0.0629, 0.0927, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:08:03,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:08:10,405 INFO [train.py:968] (1/2) Epoch 18, batch 39750, giga_loss[loss=0.3325, simple_loss=0.3876, pruned_loss=0.1387, over 26710.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3541, pruned_loss=0.103, over 5709441.16 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3638, pruned_loss=0.1162, over 5727851.51 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3514, pruned_loss=0.1007, over 5705636.77 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:08:25,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=816480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:08:27,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3208, 1.0218, 4.2176, 3.2899], device='cuda:1'), covar=tensor([0.1751, 0.3081, 0.0429, 0.1252], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0629, 0.0927, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:08:44,749 INFO [optim.py:369] (1/2) 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 15:08:54,406 INFO [train.py:968] (1/2) Epoch 18, batch 39800, giga_loss[loss=0.2669, simple_loss=0.352, pruned_loss=0.09088, over 28887.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3551, pruned_loss=0.1034, over 5709774.04 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.116, over 5729484.44 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3532, pruned_loss=0.1017, over 5705350.14 frames. ], batch size: 174, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:09:33,455 INFO [train.py:968] (1/2) Epoch 18, batch 39850, libri_loss[loss=0.4006, simple_loss=0.4369, pruned_loss=0.1821, over 18656.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3555, pruned_loss=0.1038, over 5700343.13 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3638, pruned_loss=0.1162, over 5717672.19 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3534, pruned_loss=0.1019, over 5707539.88 frames. ], batch size: 187, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:10:06,934 INFO [optim.py:369] (1/2) 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 15:10:16,119 INFO [train.py:968] (1/2) Epoch 18, batch 39900, giga_loss[loss=0.3028, simple_loss=0.3627, pruned_loss=0.1214, over 28674.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.355, pruned_loss=0.1037, over 5697728.81 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3642, pruned_loss=0.1165, over 5708507.28 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3529, pruned_loss=0.1017, over 5710312.10 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:10:24,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=816623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:10:27,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=816626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:10:50,730 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=816655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:10:55,388 INFO [train.py:968] (1/2) Epoch 18, batch 39950, giga_loss[loss=0.2293, simple_loss=0.3048, pruned_loss=0.07689, over 28417.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3516, pruned_loss=0.1024, over 5696445.17 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3644, pruned_loss=0.1167, over 5701403.13 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3496, pruned_loss=0.1005, over 5713035.13 frames. ], batch size: 77, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:11:25,904 INFO [optim.py:369] (1/2) 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 15:11:27,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-09 15:11:35,644 INFO [train.py:968] (1/2) Epoch 18, batch 40000, giga_loss[loss=0.256, simple_loss=0.3358, pruned_loss=0.08809, over 28857.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3485, pruned_loss=0.1008, over 5696152.28 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5702683.28 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3464, pruned_loss=0.09885, over 5708092.29 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:11:37,345 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 15:11:45,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 15:12:13,818 INFO [train.py:968] (1/2) Epoch 18, batch 40050, giga_loss[loss=0.328, simple_loss=0.3944, pruned_loss=0.1308, over 28689.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3477, pruned_loss=0.1005, over 5701055.64 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3651, pruned_loss=0.1173, over 5700629.04 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3451, pruned_loss=0.09797, over 5712434.75 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:12:43,491 INFO [optim.py:369] (1/2) 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 15:12:48,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5288, 1.3619, 4.4714, 3.4060], device='cuda:1'), covar=tensor([0.1522, 0.2696, 0.0352, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0725, 0.0626, 0.0923, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:12:51,625 INFO [train.py:968] (1/2) Epoch 18, batch 40100, libri_loss[loss=0.317, simple_loss=0.38, pruned_loss=0.127, over 29547.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09976, over 5713467.22 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3649, pruned_loss=0.1172, over 5706208.27 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3474, pruned_loss=0.09732, over 5717693.44 frames. ], batch size: 81, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:13:33,409 INFO [train.py:968] (1/2) Epoch 18, batch 40150, giga_loss[loss=0.2989, simple_loss=0.3754, pruned_loss=0.1112, over 28689.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.35, pruned_loss=0.09981, over 5706260.38 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3652, pruned_loss=0.1176, over 5707510.09 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3474, pruned_loss=0.09708, over 5708423.23 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:14:07,107 INFO [optim.py:369] (1/2) 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 15:14:13,959 INFO [train.py:968] (1/2) Epoch 18, batch 40200, giga_loss[loss=0.2772, simple_loss=0.3441, pruned_loss=0.1051, over 28912.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3485, pruned_loss=0.09979, over 5710571.79 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3652, pruned_loss=0.1175, over 5710596.78 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3462, pruned_loss=0.09733, over 5709410.77 frames. ], batch size: 145, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:14:54,239 INFO [train.py:968] (1/2) Epoch 18, batch 40250, giga_loss[loss=0.2997, simple_loss=0.3735, pruned_loss=0.1129, over 28737.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3482, pruned_loss=0.1011, over 5712574.14 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3652, pruned_loss=0.1176, over 5713490.22 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3461, pruned_loss=0.09888, over 5709109.59 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:15:28,343 INFO [optim.py:369] (1/2) 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 15:15:36,060 INFO [train.py:968] (1/2) Epoch 18, batch 40300, giga_loss[loss=0.2973, simple_loss=0.3564, pruned_loss=0.1191, over 28813.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3479, pruned_loss=0.1023, over 5711527.19 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3655, pruned_loss=0.1177, over 5719075.24 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3455, pruned_loss=0.1, over 5703823.24 frames. ], batch size: 199, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:15:43,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4696, 1.8771, 1.4995, 1.6946], device='cuda:1'), covar=tensor([0.0751, 0.0277, 0.0329, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 15:15:51,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-09 15:16:19,060 INFO [train.py:968] (1/2) Epoch 18, batch 40350, giga_loss[loss=0.2702, simple_loss=0.3429, pruned_loss=0.09876, over 27578.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3455, pruned_loss=0.1013, over 5710177.65 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3656, pruned_loss=0.1178, over 5711589.89 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3434, pruned_loss=0.09921, over 5711352.89 frames. ], batch size: 472, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:16:26,227 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 15:16:35,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9743, 3.1552, 2.0070, 0.9231], device='cuda:1'), covar=tensor([0.7809, 0.2698, 0.4023, 0.7413], device='cuda:1'), in_proj_covar=tensor([0.1694, 0.1593, 0.1568, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 15:16:47,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2477, 1.4980, 1.3212, 1.1448], device='cuda:1'), covar=tensor([0.2879, 0.2368, 0.1583, 0.2332], device='cuda:1'), in_proj_covar=tensor([0.1910, 0.1828, 0.1767, 0.1899], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 15:16:54,962 INFO [optim.py:369] (1/2) 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 15:16:59,985 INFO [train.py:968] (1/2) Epoch 18, batch 40400, giga_loss[loss=0.2504, simple_loss=0.3294, pruned_loss=0.08568, over 28884.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3438, pruned_loss=0.1003, over 5709798.55 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1179, over 5705466.75 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3415, pruned_loss=0.09807, over 5715155.10 frames. ], batch size: 186, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:17:11,185 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 40450, giga_loss[loss=0.2384, simple_loss=0.3138, pruned_loss=0.08154, over 29034.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3392, pruned_loss=0.09756, over 5715310.93 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3656, pruned_loss=0.1178, over 5708447.94 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3371, pruned_loss=0.09566, over 5716971.10 frames. ], batch size: 136, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:18:06,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-09 15:18:14,638 INFO [optim.py:369] (1/2) 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 15:18:19,758 INFO [train.py:968] (1/2) Epoch 18, batch 40500, giga_loss[loss=0.2827, simple_loss=0.3485, pruned_loss=0.1084, over 27893.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3352, pruned_loss=0.09561, over 5718022.52 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3658, pruned_loss=0.118, over 5708583.85 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3328, pruned_loss=0.09348, over 5719423.73 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:19:01,089 INFO [train.py:968] (1/2) Epoch 18, batch 40550, libri_loss[loss=0.3188, simple_loss=0.3868, pruned_loss=0.1253, over 29225.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3341, pruned_loss=0.09481, over 5715623.85 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3663, pruned_loss=0.1182, over 5712777.03 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.331, pruned_loss=0.09238, over 5712914.76 frames. ], batch size: 97, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:19:09,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 15:19:34,331 INFO [optim.py:369] (1/2) 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 15:19:40,305 INFO [train.py:968] (1/2) Epoch 18, batch 40600, giga_loss[loss=0.2564, simple_loss=0.327, pruned_loss=0.09289, over 28682.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3375, pruned_loss=0.0965, over 5713810.20 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.367, pruned_loss=0.1188, over 5714482.72 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3334, pruned_loss=0.09326, over 5709922.65 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:20:19,588 INFO [train.py:968] (1/2) Epoch 18, batch 40650, giga_loss[loss=0.2461, simple_loss=0.332, pruned_loss=0.08011, over 29042.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.341, pruned_loss=0.09767, over 5715308.15 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3674, pruned_loss=0.119, over 5716458.28 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3365, pruned_loss=0.09436, over 5710219.24 frames. ], batch size: 155, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:20:54,758 INFO [optim.py:369] (1/2) 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 15:21:01,601 INFO [train.py:968] (1/2) Epoch 18, batch 40700, giga_loss[loss=0.2739, simple_loss=0.3547, pruned_loss=0.09651, over 29008.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3452, pruned_loss=0.09989, over 5693564.29 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3681, pruned_loss=0.1196, over 5692539.32 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3403, pruned_loss=0.09606, over 5710324.21 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:21:32,689 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 40750, giga_loss[loss=0.2912, simple_loss=0.3624, pruned_loss=0.11, over 28830.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3475, pruned_loss=0.1008, over 5698784.56 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3686, pruned_loss=0.1201, over 5686988.71 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3424, pruned_loss=0.09672, over 5717393.46 frames. ], batch size: 99, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:22:14,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=817500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:22:16,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2496, 1.5659, 1.2367, 1.1157], device='cuda:1'), covar=tensor([0.2605, 0.2679, 0.3081, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.1445, 0.1051, 0.1280, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 15:22:17,248 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 40800, giga_loss[loss=0.2876, simple_loss=0.3643, pruned_loss=0.1055, over 28917.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3504, pruned_loss=0.1025, over 5702922.82 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3688, pruned_loss=0.1201, over 5692251.06 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3457, pruned_loss=0.09872, over 5713124.21 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:22:31,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7454, 1.8108, 1.3363, 1.4328], device='cuda:1'), covar=tensor([0.0863, 0.0642, 0.1058, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0382, 0.0444, 0.0509, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:23:13,586 INFO [train.py:968] (1/2) Epoch 18, batch 40850, giga_loss[loss=0.2953, simple_loss=0.3651, pruned_loss=0.1128, over 28904.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3535, pruned_loss=0.1052, over 5700219.30 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3691, pruned_loss=0.1203, over 5694528.59 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3494, pruned_loss=0.1019, over 5706119.18 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:23:52,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4855, 1.5891, 1.4939, 1.3758], device='cuda:1'), covar=tensor([0.2328, 0.2059, 0.1936, 0.2143], device='cuda:1'), in_proj_covar=tensor([0.1916, 0.1836, 0.1775, 0.1901], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 15:24:01,195 INFO [optim.py:369] (1/2) 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 15:24:07,574 INFO [train.py:968] (1/2) Epoch 18, batch 40900, giga_loss[loss=0.2972, simple_loss=0.3702, pruned_loss=0.1121, over 28692.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3608, pruned_loss=0.1115, over 5667801.04 frames. ], libri_tot_loss[loss=0.305, simple_loss=0.3693, pruned_loss=0.1204, over 5687099.45 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3571, pruned_loss=0.1085, over 5679788.63 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:24:20,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4932, 3.3691, 1.6037, 1.5927], device='cuda:1'), covar=tensor([0.0943, 0.0377, 0.0880, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0540, 0.0370, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 15:24:37,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=817643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:24:40,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=817646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:24:55,660 INFO [train.py:968] (1/2) Epoch 18, batch 40950, giga_loss[loss=0.3214, simple_loss=0.3919, pruned_loss=0.1255, over 28583.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3652, pruned_loss=0.1142, over 5675934.71 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.12, over 5690109.90 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3629, pruned_loss=0.1122, over 5682442.79 frames. ], batch size: 78, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:25:09,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=817675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:25:12,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-03-09 15:25:37,527 INFO [optim.py:369] (1/2) 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 15:25:42,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1531, 2.5508, 1.1869, 1.3850], device='cuda:1'), covar=tensor([0.1031, 0.0440, 0.0900, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0541, 0.0371, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 15:25:42,913 INFO [train.py:968] (1/2) Epoch 18, batch 41000, giga_loss[loss=0.3361, simple_loss=0.3979, pruned_loss=0.1371, over 29065.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1192, over 5667121.82 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1202, over 5693615.66 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3687, pruned_loss=0.1173, over 5667949.62 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:26:29,857 INFO [train.py:968] (1/2) Epoch 18, batch 41050, giga_loss[loss=0.3289, simple_loss=0.402, pruned_loss=0.1279, over 28969.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3774, pruned_loss=0.1248, over 5674397.04 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5693406.73 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3758, pruned_loss=0.1233, over 5675098.43 frames. ], batch size: 164, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:27:13,564 INFO [optim.py:369] (1/2) 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 15:27:21,079 INFO [train.py:968] (1/2) Epoch 18, batch 41100, giga_loss[loss=0.3201, simple_loss=0.3817, pruned_loss=0.1293, over 28595.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3818, pruned_loss=0.129, over 5661106.93 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3683, pruned_loss=0.1199, over 5696607.17 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3811, pruned_loss=0.1281, over 5658674.68 frames. ], batch size: 307, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:27:21,308 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=817812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:27:32,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=817823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:27:59,951 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 15:28:13,904 INFO [train.py:968] (1/2) Epoch 18, batch 41150, giga_loss[loss=0.29, simple_loss=0.3625, pruned_loss=0.1087, over 28891.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3816, pruned_loss=0.1292, over 5671677.33 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3682, pruned_loss=0.1199, over 5701526.40 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3814, pruned_loss=0.1288, over 5664562.15 frames. ], batch size: 119, lr: 1.76e-03, grad_scale: 2.0
+2023-03-09 15:28:43,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3962, 1.5675, 1.4953, 1.2097], device='cuda:1'), covar=tensor([0.2485, 0.2099, 0.1600, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1913, 0.1834, 0.1770, 0.1896], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 15:29:03,373 INFO [optim.py:369] (1/2) 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 15:29:08,894 INFO [train.py:968] (1/2) Epoch 18, batch 41200, giga_loss[loss=0.3183, simple_loss=0.3753, pruned_loss=0.1306, over 28595.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3857, pruned_loss=0.134, over 5634340.52 frames. ], libri_tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1198, over 5695565.31 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3859, pruned_loss=0.1339, over 5632217.46 frames. ], batch size: 92, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:29:10,404 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 15:29:35,323 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-09 15:29:40,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3769, 2.1017, 1.5818, 0.5603], device='cuda:1'), covar=tensor([0.4558, 0.2496, 0.3446, 0.5335], device='cuda:1'), in_proj_covar=tensor([0.1697, 0.1605, 0.1571, 0.1380], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 15:29:59,126 INFO [train.py:968] (1/2) Epoch 18, batch 41250, libri_loss[loss=0.3028, simple_loss=0.3655, pruned_loss=0.12, over 29539.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3895, pruned_loss=0.1383, over 5605175.57 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.1201, over 5670601.75 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3905, pruned_loss=0.1388, over 5622835.99 frames. ], batch size: 83, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:30:02,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=817966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:30:05,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=817969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:30:08,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-09 15:30:34,623 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=817998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:30:44,278 INFO [optim.py:369] (1/2) 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 15:30:49,813 INFO [train.py:968] (1/2) Epoch 18, batch 41300, giga_loss[loss=0.3771, simple_loss=0.4354, pruned_loss=0.1594, over 28709.00 frames. ], tot_loss[loss=0.3386, simple_loss=0.3937, pruned_loss=0.1418, over 5602595.87 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.3684, pruned_loss=0.1203, over 5660293.27 frames. ], giga_tot_loss[loss=0.34, simple_loss=0.395, pruned_loss=0.1425, over 5624208.64 frames. ], batch size: 242, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:31:44,929 INFO [train.py:968] (1/2) Epoch 18, batch 41350, giga_loss[loss=0.314, simple_loss=0.3782, pruned_loss=0.1249, over 28844.00 frames. ], tot_loss[loss=0.3405, simple_loss=0.3947, pruned_loss=0.1431, over 5614374.16 frames. ], libri_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1202, over 5661092.86 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3964, pruned_loss=0.1443, over 5629620.70 frames. ], batch size: 284, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:32:00,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1998, 1.0728, 3.6734, 3.2111], device='cuda:1'), covar=tensor([0.1678, 0.2847, 0.0512, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0631, 0.0935, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:32:03,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4525, 1.6681, 1.4071, 1.6723], device='cuda:1'), covar=tensor([0.0778, 0.0311, 0.0319, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 15:32:23,987 INFO [optim.py:369] (1/2) 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 15:32:30,198 INFO [train.py:968] (1/2) Epoch 18, batch 41400, giga_loss[loss=0.3055, simple_loss=0.3662, pruned_loss=0.1224, over 28878.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3919, pruned_loss=0.1417, over 5615096.60 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.3686, pruned_loss=0.1203, over 5658682.45 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3936, pruned_loss=0.1431, over 5628624.41 frames. ], batch size: 227, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:33:08,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1982, 0.8672, 0.9782, 1.3774], device='cuda:1'), covar=tensor([0.0793, 0.0387, 0.0342, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 15:33:08,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-09 15:33:21,957 INFO [train.py:968] (1/2) Epoch 18, batch 41450, giga_loss[loss=0.3432, simple_loss=0.4102, pruned_loss=0.1381, over 28554.00 frames. ], tot_loss[loss=0.3387, simple_loss=0.3927, pruned_loss=0.1423, over 5608103.13 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5653965.27 frames. ], giga_tot_loss[loss=0.3404, simple_loss=0.394, pruned_loss=0.1434, over 5622900.19 frames. ], batch size: 336, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:33:48,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=818187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:34:04,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0090, 5.8026, 5.5050, 2.9561], device='cuda:1'), covar=tensor([0.0483, 0.0614, 0.0712, 0.1668], device='cuda:1'), in_proj_covar=tensor([0.1174, 0.1096, 0.0936, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 15:34:06,280 INFO [optim.py:369] (1/2) 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 15:34:14,252 INFO [train.py:968] (1/2) Epoch 18, batch 41500, giga_loss[loss=0.3989, simple_loss=0.423, pruned_loss=0.1874, over 26532.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3911, pruned_loss=0.1402, over 5604675.50 frames. ], libri_tot_loss[loss=0.3051, simple_loss=0.3691, pruned_loss=0.1206, over 5655823.62 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3928, pruned_loss=0.1417, over 5613782.15 frames. ], batch size: 555, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:34:46,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3969, 1.6252, 1.3588, 1.1847], device='cuda:1'), covar=tensor([0.2646, 0.2421, 0.2684, 0.2101], device='cuda:1'), in_proj_covar=tensor([0.1441, 0.1050, 0.1282, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 15:34:50,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2009, 1.2734, 1.1847, 0.9825], device='cuda:1'), covar=tensor([0.0769, 0.0427, 0.0901, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0448, 0.0514, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:35:05,164 INFO [train.py:968] (1/2) Epoch 18, batch 41550, giga_loss[loss=0.3366, simple_loss=0.3708, pruned_loss=0.1512, over 23302.00 frames. ], tot_loss[loss=0.3393, simple_loss=0.394, pruned_loss=0.1423, over 5577080.93 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1214, over 5641317.70 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.395, pruned_loss=0.1432, over 5597342.60 frames. ], batch size: 705, lr: 1.76e-03, grad_scale: 4.0
+2023-03-09 15:35:55,026 INFO [optim.py:369] (1/2) 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 15:36:00,902 INFO [train.py:968] (1/2) Epoch 18, batch 41600, giga_loss[loss=0.2681, simple_loss=0.3439, pruned_loss=0.09618, over 28925.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3917, pruned_loss=0.1405, over 5585729.87 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3697, pruned_loss=0.1213, over 5645321.94 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3932, pruned_loss=0.1416, over 5597372.06 frames. ], batch size: 213, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:36:21,303 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=818330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:36:24,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=818333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:36:52,157 INFO [train.py:968] (1/2) Epoch 18, batch 41650, giga_loss[loss=0.3261, simple_loss=0.3809, pruned_loss=0.1356, over 27978.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.389, pruned_loss=0.1364, over 5604262.58 frames. ], libri_tot_loss[loss=0.3063, simple_loss=0.3699, pruned_loss=0.1213, over 5645618.71 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3904, pruned_loss=0.1376, over 5612323.34 frames. ], batch size: 412, lr: 1.76e-03, grad_scale: 8.0
+2023-03-09 15:36:52,341 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=818362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:36:53,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:37:11,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4110, 1.7595, 1.5977, 1.4832], device='cuda:1'), covar=tensor([0.1863, 0.2138, 0.2216, 0.2139], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0744, 0.0704, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 15:37:20,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5111, 1.7229, 1.7490, 1.3134], device='cuda:1'), covar=tensor([0.1871, 0.2573, 0.1580, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.0872, 0.0691, 0.0916, 0.0818], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 15:37:36,472 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 41700, giga_loss[loss=0.3012, simple_loss=0.3691, pruned_loss=0.1166, over 27957.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3859, pruned_loss=0.1333, over 5619625.54 frames. ], libri_tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1215, over 5652083.23 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3873, pruned_loss=0.1345, over 5619686.01 frames. ], batch size: 412, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:37:56,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6372, 4.5474, 1.7957, 1.7576], device='cuda:1'), covar=tensor([0.0960, 0.0401, 0.0915, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0543, 0.0371, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 15:38:05,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 15:38:09,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 15:38:32,400 INFO [train.py:968] (1/2) Epoch 18, batch 41750, giga_loss[loss=0.3144, simple_loss=0.3669, pruned_loss=0.1309, over 23934.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3818, pruned_loss=0.13, over 5620638.79 frames. ], libri_tot_loss[loss=0.3062, simple_loss=0.3697, pruned_loss=0.1214, over 5653393.43 frames. ], giga_tot_loss[loss=0.3229, simple_loss=0.3834, pruned_loss=0.1312, over 5618881.14 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:39:17,424 INFO [optim.py:369] (1/2) 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 15:39:22,324 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-09 15:39:22,483 INFO [train.py:968] (1/2) Epoch 18, batch 41800, giga_loss[loss=0.3398, simple_loss=0.398, pruned_loss=0.1408, over 28758.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3798, pruned_loss=0.1285, over 5629617.26 frames. ], libri_tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1212, over 5657315.25 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3815, pruned_loss=0.1297, over 5624087.56 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:39:36,035 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818524.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:40:01,290 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 15:40:12,979 INFO [train.py:968] (1/2) Epoch 18, batch 41850, libri_loss[loss=0.3108, simple_loss=0.3761, pruned_loss=0.1227, over 29468.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3783, pruned_loss=0.1274, over 5632988.92 frames. ], libri_tot_loss[loss=0.3055, simple_loss=0.3691, pruned_loss=0.1209, over 5653019.56 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3802, pruned_loss=0.1288, over 5632503.62 frames. ], batch size: 85, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:40:18,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9534, 2.5754, 2.6688, 2.3837], device='cuda:1'), covar=tensor([0.1410, 0.2170, 0.1786, 0.1932], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0742, 0.0701, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 15:40:20,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8373, 2.1403, 2.0255, 1.6221], device='cuda:1'), covar=tensor([0.2973, 0.2336, 0.2407, 0.2713], device='cuda:1'), in_proj_covar=tensor([0.1907, 0.1826, 0.1765, 0.1898], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 15:40:53,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1527, 2.3120, 2.4375, 1.9352], device='cuda:1'), covar=tensor([0.1740, 0.2115, 0.1303, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0695, 0.0919, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 15:40:54,473 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 41900, giga_loss[loss=0.3408, simple_loss=0.386, pruned_loss=0.1479, over 26592.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3774, pruned_loss=0.1269, over 5630846.27 frames. ], libri_tot_loss[loss=0.3054, simple_loss=0.369, pruned_loss=0.1209, over 5655061.94 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3792, pruned_loss=0.1282, over 5628410.81 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:41:33,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-09 15:41:39,048 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:41:54,088 INFO [train.py:968] (1/2) Epoch 18, batch 41950, giga_loss[loss=0.2972, simple_loss=0.3685, pruned_loss=0.113, over 28554.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3754, pruned_loss=0.1249, over 5620059.43 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5640097.14 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3771, pruned_loss=0.1259, over 5632222.80 frames. ], batch size: 78, lr: 1.75e-03, grad_scale: 2.0
+2023-03-09 15:42:44,099 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 42000, giga_loss[loss=0.2973, simple_loss=0.3753, pruned_loss=0.1096, over 28897.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3762, pruned_loss=0.1226, over 5633691.65 frames. ], libri_tot_loss[loss=0.3053, simple_loss=0.3686, pruned_loss=0.1209, over 5643112.58 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3779, pruned_loss=0.1235, over 5640253.38 frames. ], batch size: 145, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:42:48,678 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 15:42:57,325 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 15:43:23,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-09 15:43:25,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=818739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:43:32,827 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=818746.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:43:46,823 INFO [train.py:968] (1/2) Epoch 18, batch 42050, giga_loss[loss=0.301, simple_loss=0.3692, pruned_loss=0.1165, over 28551.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3778, pruned_loss=0.1228, over 5644090.78 frames. ], libri_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1206, over 5646295.30 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3796, pruned_loss=0.1238, over 5646145.67 frames. ], batch size: 85, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:44:28,003 INFO [optim.py:369] (1/2) 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 15:44:30,819 INFO [train.py:968] (1/2) Epoch 18, batch 42100, giga_loss[loss=0.2939, simple_loss=0.3669, pruned_loss=0.1104, over 28731.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3775, pruned_loss=0.1231, over 5663707.11 frames. ], libri_tot_loss[loss=0.3044, simple_loss=0.3679, pruned_loss=0.1205, over 5655077.17 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3797, pruned_loss=0.1241, over 5657546.32 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:44:59,075 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0388, 3.8659, 3.7058, 1.7719], device='cuda:1'), covar=tensor([0.0684, 0.0763, 0.0754, 0.2190], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.1104, 0.0940, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 15:45:10,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7843, 2.0120, 1.5915, 2.1875], device='cuda:1'), covar=tensor([0.2518, 0.2653, 0.3044, 0.2299], device='cuda:1'), in_proj_covar=tensor([0.1451, 0.1056, 0.1293, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 15:45:14,732 INFO [train.py:968] (1/2) Epoch 18, batch 42150, giga_loss[loss=0.3451, simple_loss=0.3959, pruned_loss=0.1472, over 28590.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.377, pruned_loss=0.1233, over 5652206.65 frames. ], libri_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1208, over 5649245.89 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3788, pruned_loss=0.124, over 5652619.13 frames. ], batch size: 85, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:45:34,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=818882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:45:36,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=818885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:45:47,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=818899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:45:57,097 INFO [optim.py:369] (1/2) 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 15:46:00,763 INFO [train.py:968] (1/2) Epoch 18, batch 42200, giga_loss[loss=0.3482, simple_loss=0.3964, pruned_loss=0.15, over 28908.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3749, pruned_loss=0.1225, over 5666354.06 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3677, pruned_loss=0.1204, over 5655155.65 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3769, pruned_loss=0.1234, over 5661808.38 frames. ], batch size: 227, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:46:02,437 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=818914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:46:22,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8940, 1.8938, 1.5583, 1.4919], device='cuda:1'), covar=tensor([0.0892, 0.0668, 0.0983, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0449, 0.0516, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:46:51,236 INFO [train.py:968] (1/2) Epoch 18, batch 42250, giga_loss[loss=0.2708, simple_loss=0.3452, pruned_loss=0.09823, over 28897.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3737, pruned_loss=0.1229, over 5663338.41 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3676, pruned_loss=0.1203, over 5657688.04 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3755, pruned_loss=0.1238, over 5657464.24 frames. ], batch size: 174, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:47:22,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5568, 3.4025, 3.2280, 1.8789], device='cuda:1'), covar=tensor([0.0790, 0.0838, 0.0805, 0.2244], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.1104, 0.0941, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 15:47:23,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7736, 4.6965, 1.9225, 1.8898], device='cuda:1'), covar=tensor([0.0957, 0.0340, 0.0908, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0545, 0.0373, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 15:47:30,423 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 15:47:40,962 INFO [optim.py:369] (1/2) 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 15:47:44,176 INFO [train.py:968] (1/2) Epoch 18, batch 42300, libri_loss[loss=0.3698, simple_loss=0.4129, pruned_loss=0.1634, over 19795.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3732, pruned_loss=0.1215, over 5655487.63 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3677, pruned_loss=0.1203, over 5650098.62 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3745, pruned_loss=0.1221, over 5658614.42 frames. ], batch size: 187, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:47:53,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:48:13,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819042.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:48:14,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:48:16,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819045.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:48:31,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4006, 1.7834, 1.3329, 1.5546], device='cuda:1'), covar=tensor([0.0782, 0.0312, 0.0349, 0.0881], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 15:48:32,061 INFO [train.py:968] (1/2) Epoch 18, batch 42350, giga_loss[loss=0.2719, simple_loss=0.3545, pruned_loss=0.09461, over 28916.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3736, pruned_loss=0.1208, over 5662411.68 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.368, pruned_loss=0.1205, over 5645535.68 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3744, pruned_loss=0.121, over 5669786.73 frames. ], batch size: 227, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:48:43,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819074.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:49:16,713 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 42400, giga_loss[loss=0.3467, simple_loss=0.4051, pruned_loss=0.1442, over 28847.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3744, pruned_loss=0.1216, over 5657257.01 frames. ], libri_tot_loss[loss=0.3045, simple_loss=0.368, pruned_loss=0.1205, over 5643024.20 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3753, pruned_loss=0.1219, over 5665492.45 frames. ], batch size: 174, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 15:49:27,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819121.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:50:02,052 INFO [train.py:968] (1/2) Epoch 18, batch 42450, giga_loss[loss=0.2923, simple_loss=0.3563, pruned_loss=0.1141, over 28669.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3735, pruned_loss=0.1216, over 5652525.45 frames. ], libri_tot_loss[loss=0.3046, simple_loss=0.368, pruned_loss=0.1206, over 5635634.89 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3744, pruned_loss=0.1218, over 5666254.10 frames. ], batch size: 92, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 15:50:03,935 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:50:06,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:50:33,331 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 18, batch 42500, libri_loss[loss=0.2593, simple_loss=0.339, pruned_loss=0.08974, over 29566.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1203, over 5671177.54 frames. ], libri_tot_loss[loss=0.3041, simple_loss=0.3676, pruned_loss=0.1203, over 5645244.42 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3723, pruned_loss=0.1208, over 5674046.26 frames. ], batch size: 74, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:50:48,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3554, 3.0338, 1.3518, 1.5988], device='cuda:1'), covar=tensor([0.1001, 0.0364, 0.0905, 0.1279], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0548, 0.0375, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 15:51:35,892 INFO [train.py:968] (1/2) Epoch 18, batch 42550, giga_loss[loss=0.2901, simple_loss=0.3598, pruned_loss=0.1102, over 28654.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3695, pruned_loss=0.1201, over 5662977.53 frames. ], libri_tot_loss[loss=0.3037, simple_loss=0.3673, pruned_loss=0.12, over 5649450.64 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3709, pruned_loss=0.1207, over 5662006.11 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:51:38,827 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819264.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:51:40,531 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819267.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 15:52:06,504 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819296.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 15:52:13,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2105, 1.7112, 1.5310, 1.3578], device='cuda:1'), covar=tensor([0.0812, 0.0306, 0.0284, 0.1016], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 15:52:17,827 INFO [optim.py:369] (1/2) 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 15:52:20,635 INFO [train.py:968] (1/2) Epoch 18, batch 42600, giga_loss[loss=0.3523, simple_loss=0.3969, pruned_loss=0.1539, over 27633.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.369, pruned_loss=0.12, over 5679492.51 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3666, pruned_loss=0.1193, over 5658674.77 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3708, pruned_loss=0.1211, over 5671074.09 frames. ], batch size: 472, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:52:42,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:53:05,217 INFO [train.py:968] (1/2) Epoch 18, batch 42650, giga_loss[loss=0.2912, simple_loss=0.3539, pruned_loss=0.1142, over 29086.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5683230.15 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3667, pruned_loss=0.1195, over 5663150.34 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5673437.81 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:53:41,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4096, 1.4229, 1.2801, 1.4775], device='cuda:1'), covar=tensor([0.0765, 0.0348, 0.0333, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 15:53:52,972 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 42700, giga_loss[loss=0.2724, simple_loss=0.3353, pruned_loss=0.1047, over 29000.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3682, pruned_loss=0.1211, over 5661718.86 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3664, pruned_loss=0.1192, over 5659747.89 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3697, pruned_loss=0.122, over 5656732.84 frames. ], batch size: 106, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:54:02,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:54:15,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:54:36,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4877, 1.7840, 1.4122, 1.6075], device='cuda:1'), covar=tensor([0.2652, 0.2654, 0.3056, 0.2231], device='cuda:1'), in_proj_covar=tensor([0.1446, 0.1053, 0.1287, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 15:54:44,138 INFO [train.py:968] (1/2) Epoch 18, batch 42750, giga_loss[loss=0.2783, simple_loss=0.3478, pruned_loss=0.1044, over 28754.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3663, pruned_loss=0.1197, over 5659122.70 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3656, pruned_loss=0.1186, over 5663574.55 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3683, pruned_loss=0.121, over 5651833.71 frames. ], batch size: 119, lr: 1.75e-03, grad_scale: 2.0
+2023-03-09 15:54:50,779 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:55:23,853 INFO [optim.py:369] (1/2) 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 15:55:25,724 INFO [train.py:968] (1/2) Epoch 18, batch 42800, giga_loss[loss=0.2839, simple_loss=0.3612, pruned_loss=0.1033, over 28895.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3668, pruned_loss=0.1194, over 5672554.31 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3657, pruned_loss=0.1187, over 5671154.98 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3684, pruned_loss=0.1205, over 5659741.32 frames. ], batch size: 213, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:55:33,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3454, 3.1423, 2.9775, 1.3983], device='cuda:1'), covar=tensor([0.1031, 0.1231, 0.1166, 0.2322], device='cuda:1'), in_proj_covar=tensor([0.1188, 0.1110, 0.0945, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 15:55:49,514 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:56:11,715 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:56:12,008 INFO [train.py:968] (1/2) Epoch 18, batch 42850, giga_loss[loss=0.3537, simple_loss=0.4074, pruned_loss=0.15, over 28828.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3678, pruned_loss=0.1188, over 5677463.52 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3658, pruned_loss=0.1187, over 5674686.75 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.369, pruned_loss=0.1197, over 5664389.48 frames. ], batch size: 174, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:56:14,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:56:38,675 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:56:57,419 INFO [train.py:968] (1/2) Epoch 18, batch 42900, giga_loss[loss=0.3073, simple_loss=0.3684, pruned_loss=0.1232, over 28679.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5688546.18 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3656, pruned_loss=0.1184, over 5681284.67 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3694, pruned_loss=0.1191, over 5672455.16 frames. ], batch size: 92, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:57:02,556 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-09 15:57:22,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5540, 1.6838, 1.2169, 1.2762], device='cuda:1'), covar=tensor([0.0869, 0.0550, 0.0993, 0.1199], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0446, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 15:57:49,278 INFO [train.py:968] (1/2) Epoch 18, batch 42950, giga_loss[loss=0.3516, simple_loss=0.4034, pruned_loss=0.1499, over 28505.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3695, pruned_loss=0.1195, over 5691738.57 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.1179, over 5686560.63 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.371, pruned_loss=0.1207, over 5673981.96 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:58:26,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-09 15:58:33,876 INFO [optim.py:369] (1/2) 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 15:58:36,065 INFO [train.py:968] (1/2) Epoch 18, batch 43000, giga_loss[loss=0.2947, simple_loss=0.3621, pruned_loss=0.1136, over 28766.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3727, pruned_loss=0.1223, over 5694653.77 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.118, over 5686750.97 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3739, pruned_loss=0.1233, over 5680350.95 frames. ], batch size: 119, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:58:37,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:59:20,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2529, 1.8792, 1.3859, 0.5061], device='cuda:1'), covar=tensor([0.3729, 0.2096, 0.2968, 0.4876], device='cuda:1'), in_proj_covar=tensor([0.1703, 0.1618, 0.1578, 0.1395], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 15:59:32,778 INFO [train.py:968] (1/2) Epoch 18, batch 43050, giga_loss[loss=0.378, simple_loss=0.3992, pruned_loss=0.1784, over 23797.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.375, pruned_loss=0.126, over 5685403.46 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.365, pruned_loss=0.1179, over 5689769.51 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3763, pruned_loss=0.127, over 5671503.21 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 15:59:42,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=819769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 15:59:45,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 16:00:20,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:00:23,121 INFO [optim.py:369] (1/2) 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 16:00:25,509 INFO [train.py:968] (1/2) Epoch 18, batch 43100, giga_loss[loss=0.3235, simple_loss=0.3859, pruned_loss=0.1305, over 28598.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3755, pruned_loss=0.127, over 5677647.20 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1176, over 5683756.56 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3771, pruned_loss=0.1283, over 5671339.69 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:00:56,474 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:01:08,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:01:10,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:01:12,426 INFO [train.py:968] (1/2) Epoch 18, batch 43150, giga_loss[loss=0.2599, simple_loss=0.3321, pruned_loss=0.09388, over 28925.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.376, pruned_loss=0.1277, over 5670234.71 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3649, pruned_loss=0.1177, over 5689337.92 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3775, pruned_loss=0.1289, over 5659514.51 frames. ], batch size: 199, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:01:37,315 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:01:59,390 INFO [train.py:968] (1/2) Epoch 18, batch 43200, giga_loss[loss=0.3364, simple_loss=0.385, pruned_loss=0.1439, over 28906.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3755, pruned_loss=0.1277, over 5670097.00 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.1179, over 5691499.13 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3766, pruned_loss=0.1285, over 5659404.42 frames. ], batch size: 112, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:01:59,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=819912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:02:18,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2610, 1.1273, 3.7367, 3.2165], device='cuda:1'), covar=tensor([0.1695, 0.2921, 0.0500, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0633, 0.0937, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:02:34,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:02:36,007 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 43250, giga_loss[loss=0.3108, simple_loss=0.3847, pruned_loss=0.1184, over 28645.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1253, over 5673385.60 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.365, pruned_loss=0.1179, over 5689265.44 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.376, pruned_loss=0.1262, over 5665620.84 frames. ], batch size: 242, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:03:03,969 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=819983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:03:07,813 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=819987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:03:09,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=819990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:03:31,994 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 43300, giga_loss[loss=0.2866, simple_loss=0.3628, pruned_loss=0.1052, over 28959.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3728, pruned_loss=0.1232, over 5662882.03 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3653, pruned_loss=0.118, over 5690212.82 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.124, over 5655760.50 frames. ], batch size: 106, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:03:40,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:04:11,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:04:13,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=820058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:04:16,738 INFO [train.py:968] (1/2) Epoch 18, batch 43350, giga_loss[loss=0.2528, simple_loss=0.3354, pruned_loss=0.08505, over 29064.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1219, over 5667061.85 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3655, pruned_loss=0.1181, over 5687034.77 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1226, over 5663885.88 frames. ], batch size: 136, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:04:23,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3824, 4.2158, 4.0160, 1.8859], device='cuda:1'), covar=tensor([0.0602, 0.0707, 0.0744, 0.2054], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.1105, 0.0944, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 16:04:30,468 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=820077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:04:39,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:05:03,298 INFO [optim.py:369] (1/2) 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 16:05:03,994 INFO [train.py:968] (1/2) Epoch 18, batch 43400, giga_loss[loss=0.3001, simple_loss=0.3551, pruned_loss=0.1225, over 28125.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3692, pruned_loss=0.1219, over 5661441.41 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5687378.12 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3702, pruned_loss=0.1228, over 5658419.77 frames. ], batch size: 77, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:05:10,156 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-09 16:05:35,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:05:52,423 INFO [train.py:968] (1/2) Epoch 18, batch 43450, giga_loss[loss=0.3023, simple_loss=0.3707, pruned_loss=0.1169, over 28800.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1226, over 5669230.48 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3655, pruned_loss=0.1182, over 5688277.73 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.123, over 5665851.27 frames. ], batch size: 243, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:05:58,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5552, 4.4069, 4.1603, 2.0181], device='cuda:1'), covar=tensor([0.0608, 0.0720, 0.0762, 0.2029], device='cuda:1'), in_proj_covar=tensor([0.1183, 0.1105, 0.0943, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 16:06:35,065 INFO [optim.py:369] (1/2) 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 16:06:36,210 INFO [train.py:968] (1/2) Epoch 18, batch 43500, giga_loss[loss=0.2775, simple_loss=0.3531, pruned_loss=0.101, over 28750.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3746, pruned_loss=0.125, over 5663530.52 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.366, pruned_loss=0.1186, over 5683299.36 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5664487.21 frames. ], batch size: 92, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:06:54,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3586, 1.6329, 1.3457, 1.5826], device='cuda:1'), covar=tensor([0.0812, 0.0315, 0.0361, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 16:07:23,262 INFO [train.py:968] (1/2) Epoch 18, batch 43550, giga_loss[loss=0.2992, simple_loss=0.3812, pruned_loss=0.1086, over 28380.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3767, pruned_loss=0.1235, over 5666721.18 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3659, pruned_loss=0.1186, over 5683787.92 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3773, pruned_loss=0.1238, over 5666449.16 frames. ], batch size: 368, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:07:46,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:07:51,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=820290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:08:11,745 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 43600, giga_loss[loss=0.3141, simple_loss=0.377, pruned_loss=0.1256, over 28547.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3773, pruned_loss=0.1234, over 5670946.21 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3657, pruned_loss=0.1185, over 5692032.74 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3782, pruned_loss=0.124, over 5663027.54 frames. ], batch size: 85, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:08:19,328 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:08:22,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6266, 1.8080, 1.5869, 1.6075], device='cuda:1'), covar=tensor([0.1568, 0.2029, 0.2079, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0744, 0.0704, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:08:38,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5981, 1.8164, 1.4902, 1.5989], device='cuda:1'), covar=tensor([0.2553, 0.2633, 0.2884, 0.2544], device='cuda:1'), in_proj_covar=tensor([0.1444, 0.1049, 0.1282, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 16:09:00,655 INFO [train.py:968] (1/2) Epoch 18, batch 43650, giga_loss[loss=0.3419, simple_loss=0.3942, pruned_loss=0.1448, over 28276.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3797, pruned_loss=0.1254, over 5663756.13 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3657, pruned_loss=0.1184, over 5686154.31 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3807, pruned_loss=0.126, over 5663068.15 frames. ], batch size: 368, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:09:52,942 INFO [optim.py:369] (1/2) 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 16:09:52,955 INFO [train.py:968] (1/2) Epoch 18, batch 43700, giga_loss[loss=0.2813, simple_loss=0.3562, pruned_loss=0.1032, over 28915.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3805, pruned_loss=0.1269, over 5658848.27 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3653, pruned_loss=0.1182, over 5688401.07 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3817, pruned_loss=0.1276, over 5656200.13 frames. ], batch size: 174, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:10:27,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:10:37,293 INFO [train.py:968] (1/2) Epoch 18, batch 43750, giga_loss[loss=0.3231, simple_loss=0.3885, pruned_loss=0.1288, over 28663.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3792, pruned_loss=0.1267, over 5670280.98 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3654, pruned_loss=0.1182, over 5691471.85 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3804, pruned_loss=0.1274, over 5665204.65 frames. ], batch size: 242, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:10:41,811 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=820466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:11:02,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6138, 2.2532, 1.6745, 0.8173], device='cuda:1'), covar=tensor([0.5439, 0.2684, 0.3764, 0.5955], device='cuda:1'), in_proj_covar=tensor([0.1700, 0.1615, 0.1571, 0.1386], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 16:11:29,102 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 43800, giga_loss[loss=0.3309, simple_loss=0.3896, pruned_loss=0.1361, over 27941.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3774, pruned_loss=0.1263, over 5664434.59 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.118, over 5691326.35 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3788, pruned_loss=0.1272, over 5660090.09 frames. ], batch size: 412, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:11:43,884 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 43850, giga_loss[loss=0.3039, simple_loss=0.3685, pruned_loss=0.1196, over 28667.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5670302.77 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3653, pruned_loss=0.1181, over 5693600.71 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3758, pruned_loss=0.1256, over 5664781.63 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:12:46,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=820592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:12:50,546 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:12:55,385 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:13:12,190 INFO [train.py:968] (1/2) Epoch 18, batch 43900, giga_loss[loss=0.292, simple_loss=0.3591, pruned_loss=0.1125, over 28544.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3738, pruned_loss=0.1247, over 5671572.68 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3654, pruned_loss=0.1181, over 5695660.58 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3748, pruned_loss=0.1254, over 5664778.66 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:13:24,348 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=820627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:13:56,973 INFO [train.py:968] (1/2) Epoch 18, batch 43950, giga_loss[loss=0.3353, simple_loss=0.3856, pruned_loss=0.1424, over 27644.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3736, pruned_loss=0.1246, over 5680036.42 frames. ], libri_tot_loss[loss=0.3009, simple_loss=0.3656, pruned_loss=0.1182, over 5704725.67 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 5665287.11 frames. ], batch size: 472, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:14:38,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3066, 1.4714, 1.1120, 1.0249], device='cuda:1'), covar=tensor([0.0958, 0.0520, 0.1102, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0447, 0.0513, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:14:45,164 INFO [optim.py:369] (1/2) 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 16:14:45,177 INFO [train.py:968] (1/2) Epoch 18, batch 44000, libri_loss[loss=0.3296, simple_loss=0.3813, pruned_loss=0.1389, over 18726.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3736, pruned_loss=0.1252, over 5663334.51 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5687946.41 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3744, pruned_loss=0.1258, over 5668087.44 frames. ], batch size: 187, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:15:30,815 INFO [train.py:968] (1/2) Epoch 18, batch 44050, giga_loss[loss=0.2586, simple_loss=0.3386, pruned_loss=0.08929, over 29027.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3703, pruned_loss=0.1231, over 5672526.16 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5692714.76 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3711, pruned_loss=0.1238, over 5671470.25 frames. ], batch size: 155, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:16:02,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-09 16:16:16,366 INFO [train.py:968] (1/2) Epoch 18, batch 44100, giga_loss[loss=0.3701, simple_loss=0.4055, pruned_loss=0.1673, over 26594.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.123, over 5671866.00 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.365, pruned_loss=0.1179, over 5695573.25 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3718, pruned_loss=0.1239, over 5668059.95 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:16:18,249 INFO [optim.py:369] (1/2) 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 16:16:48,793 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 44150, libri_loss[loss=0.3167, simple_loss=0.389, pruned_loss=0.1222, over 29210.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 5647910.21 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1178, over 5679633.40 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3753, pruned_loss=0.1263, over 5658806.82 frames. ], batch size: 94, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:17:36,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4684, 1.6819, 1.6249, 1.4291], device='cuda:1'), covar=tensor([0.2383, 0.2300, 0.2292, 0.2363], device='cuda:1'), in_proj_covar=tensor([0.1924, 0.1846, 0.1782, 0.1926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 16:17:50,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:17:57,772 INFO [train.py:968] (1/2) Epoch 18, batch 44200, giga_loss[loss=0.3097, simple_loss=0.3751, pruned_loss=0.1221, over 28597.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3743, pruned_loss=0.1251, over 5665034.29 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.365, pruned_loss=0.1178, over 5683113.49 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3754, pruned_loss=0.1261, over 5670140.99 frames. ], batch size: 242, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:17:58,572 INFO [optim.py:369] (1/2) 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 16:18:34,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9274, 1.1744, 1.1024, 0.8541], device='cuda:1'), covar=tensor([0.2265, 0.2385, 0.1399, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.1918, 0.1838, 0.1775, 0.1919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 16:18:42,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5382, 1.7447, 1.7850, 1.5584], device='cuda:1'), covar=tensor([0.1792, 0.2045, 0.1982, 0.2086], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0745, 0.0706, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:18:48,883 INFO [train.py:968] (1/2) Epoch 18, batch 44250, giga_loss[loss=0.2797, simple_loss=0.3683, pruned_loss=0.09552, over 28918.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1252, over 5658458.61 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5684372.48 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3763, pruned_loss=0.126, over 5661130.31 frames. ], batch size: 112, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:18:54,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=820967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:19:10,991 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=820984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:19:12,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=820987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:19:29,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 16:19:36,565 INFO [train.py:968] (1/2) Epoch 18, batch 44300, giga_loss[loss=0.2844, simple_loss=0.3717, pruned_loss=0.09857, over 28919.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3767, pruned_loss=0.1236, over 5671055.11 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3649, pruned_loss=0.1178, over 5685423.62 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3776, pruned_loss=0.1243, over 5672114.37 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:19:37,321 INFO [optim.py:369] (1/2) 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 16:19:40,875 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:20:05,421 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:20:08,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:20:23,799 INFO [train.py:968] (1/2) Epoch 18, batch 44350, giga_loss[loss=0.2818, simple_loss=0.3562, pruned_loss=0.1037, over 28873.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3793, pruned_loss=0.1242, over 5678032.79 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1177, over 5690348.88 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3805, pruned_loss=0.1249, over 5674305.76 frames. ], batch size: 112, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:20:37,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5628, 4.0673, 1.5979, 1.7736], device='cuda:1'), covar=tensor([0.0925, 0.0383, 0.0928, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0547, 0.0374, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 16:20:41,019 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:21:10,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:21:14,409 INFO [train.py:968] (1/2) Epoch 18, batch 44400, giga_loss[loss=0.4074, simple_loss=0.4334, pruned_loss=0.1906, over 23388.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3813, pruned_loss=0.1258, over 5677514.65 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3647, pruned_loss=0.1176, over 5692871.74 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3827, pruned_loss=0.1266, over 5671907.52 frames. ], batch size: 705, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:21:14,994 INFO [optim.py:369] (1/2) 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 16:21:15,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:21:32,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5166, 1.6260, 1.5871, 1.4783], device='cuda:1'), covar=tensor([0.1644, 0.2078, 0.2041, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.0461, 0.0740, 0.0699, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:21:42,805 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 44450, giga_loss[loss=0.3128, simple_loss=0.3764, pruned_loss=0.1246, over 28935.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3824, pruned_loss=0.1277, over 5677870.64 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3644, pruned_loss=0.1174, over 5695061.59 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.384, pruned_loss=0.1286, over 5671333.90 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:22:25,039 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:22:27,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2600, 1.5180, 1.5020, 1.3731], device='cuda:1'), covar=tensor([0.1764, 0.1608, 0.2214, 0.1694], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0742, 0.0701, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:22:51,211 INFO [train.py:968] (1/2) Epoch 18, batch 44500, giga_loss[loss=0.3881, simple_loss=0.4188, pruned_loss=0.1787, over 26536.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3835, pruned_loss=0.1297, over 5647248.19 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3652, pruned_loss=0.1179, over 5686525.04 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3845, pruned_loss=0.1302, over 5648718.24 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:22:52,436 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 18, batch 44550, giga_loss[loss=0.3178, simple_loss=0.3819, pruned_loss=0.1268, over 28327.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3827, pruned_loss=0.1287, over 5660662.83 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3651, pruned_loss=0.1179, over 5689787.23 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.384, pruned_loss=0.1295, over 5657949.05 frames. ], batch size: 65, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:24:16,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:24:17,131 INFO [train.py:968] (1/2) Epoch 18, batch 44600, giga_loss[loss=0.2697, simple_loss=0.3605, pruned_loss=0.08946, over 28901.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3803, pruned_loss=0.1258, over 5667278.35 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3647, pruned_loss=0.1177, over 5692877.54 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3821, pruned_loss=0.1268, over 5661077.14 frames. ], batch size: 145, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:24:18,472 INFO [optim.py:369] (1/2) 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 16:24:28,643 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-09 16:25:06,088 INFO [train.py:968] (1/2) Epoch 18, batch 44650, giga_loss[loss=0.313, simple_loss=0.3876, pruned_loss=0.1192, over 28638.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3814, pruned_loss=0.1252, over 5665701.82 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3648, pruned_loss=0.1178, over 5696691.76 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3833, pruned_loss=0.1262, over 5656727.90 frames. ], batch size: 92, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:25:37,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4309, 3.3993, 1.5400, 1.5600], device='cuda:1'), covar=tensor([0.0971, 0.0345, 0.0898, 0.1352], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0549, 0.0375, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 16:25:53,475 INFO [train.py:968] (1/2) Epoch 18, batch 44700, giga_loss[loss=0.3288, simple_loss=0.3911, pruned_loss=0.1333, over 28933.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3817, pruned_loss=0.1249, over 5677055.42 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3648, pruned_loss=0.1178, over 5699990.11 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3834, pruned_loss=0.1258, over 5666428.86 frames. ], batch size: 136, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:25:55,424 INFO [optim.py:369] (1/2) 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 16:26:40,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-09 16:26:40,812 INFO [train.py:968] (1/2) Epoch 18, batch 44750, giga_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.09695, over 28702.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3814, pruned_loss=0.1253, over 5680691.41 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5703859.55 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3829, pruned_loss=0.1262, over 5667874.38 frames. ], batch size: 92, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:27:14,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1925, 1.4211, 1.2860, 1.0418], device='cuda:1'), covar=tensor([0.2518, 0.2407, 0.1636, 0.2307], device='cuda:1'), in_proj_covar=tensor([0.1914, 0.1831, 0.1771, 0.1911], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 16:27:24,145 INFO [train.py:968] (1/2) Epoch 18, batch 44800, giga_loss[loss=0.3151, simple_loss=0.3725, pruned_loss=0.1288, over 28988.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3804, pruned_loss=0.1253, over 5676250.45 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5708277.77 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3822, pruned_loss=0.1263, over 5661685.14 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:27:26,306 INFO [optim.py:369] (1/2) 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 16:28:14,361 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 44850, giga_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.1219, over 28585.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3795, pruned_loss=0.1267, over 5654608.84 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3647, pruned_loss=0.1176, over 5702621.96 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3813, pruned_loss=0.1276, over 5648251.42 frames. ], batch size: 85, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:28:37,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9422, 1.1987, 1.3026, 1.0092], device='cuda:1'), covar=tensor([0.1942, 0.1537, 0.2533, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0752, 0.0710, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:28:44,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-09 16:29:06,166 INFO [train.py:968] (1/2) Epoch 18, batch 44900, giga_loss[loss=0.2996, simple_loss=0.3662, pruned_loss=0.1165, over 29003.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3781, pruned_loss=0.1267, over 5659825.11 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3646, pruned_loss=0.1176, over 5705105.66 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3799, pruned_loss=0.1276, over 5651998.70 frames. ], batch size: 128, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:29:08,392 INFO [optim.py:369] (1/2) 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 16:29:54,323 INFO [train.py:968] (1/2) Epoch 18, batch 44950, giga_loss[loss=0.2761, simple_loss=0.3467, pruned_loss=0.1028, over 28959.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3759, pruned_loss=0.1258, over 5666378.25 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3648, pruned_loss=0.1177, over 5708929.91 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3773, pruned_loss=0.1266, over 5655982.04 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:30:16,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=821685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:30:38,689 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:30:41,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:30:45,116 INFO [train.py:968] (1/2) Epoch 18, batch 45000, giga_loss[loss=0.2951, simple_loss=0.3663, pruned_loss=0.112, over 28515.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3751, pruned_loss=0.1255, over 5674728.41 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1179, over 5710542.57 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3761, pruned_loss=0.1261, over 5664850.41 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:30:45,116 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 16:30:55,170 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 16:30:56,454 INFO [optim.py:369] (1/2) 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 16:31:16,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=821735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:31:39,243 INFO [train.py:968] (1/2) Epoch 18, batch 45050, giga_loss[loss=0.3502, simple_loss=0.397, pruned_loss=0.1517, over 26651.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1226, over 5672694.55 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5713624.32 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3733, pruned_loss=0.1233, over 5661492.48 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:31:59,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-09 16:31:59,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5487, 4.4044, 4.1529, 2.1556], device='cuda:1'), covar=tensor([0.0539, 0.0653, 0.0688, 0.1949], device='cuda:1'), in_proj_covar=tensor([0.1196, 0.1115, 0.0950, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 16:32:05,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1843, 1.2708, 3.5542, 3.1118], device='cuda:1'), covar=tensor([0.1682, 0.2855, 0.0437, 0.1548], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0638, 0.0944, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:32:06,665 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:32:18,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4154, 3.9991, 1.6481, 1.5740], device='cuda:1'), covar=tensor([0.1011, 0.0352, 0.0913, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0549, 0.0375, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 16:32:29,340 INFO [train.py:968] (1/2) Epoch 18, batch 45100, giga_loss[loss=0.32, simple_loss=0.3808, pruned_loss=0.1296, over 28942.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3685, pruned_loss=0.1187, over 5671281.09 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.365, pruned_loss=0.1177, over 5714581.20 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3692, pruned_loss=0.1192, over 5661389.77 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:32:32,346 INFO [optim.py:369] (1/2) 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 16:32:34,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1032, 2.2034, 2.1714, 1.9129], device='cuda:1'), covar=tensor([0.1943, 0.2483, 0.2030, 0.2251], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0750, 0.0709, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:32:37,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-09 16:32:42,518 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:32:43,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=821828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:32:45,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=821831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:33:18,636 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 45150, giga_loss[loss=0.2792, simple_loss=0.3544, pruned_loss=0.102, over 28951.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3679, pruned_loss=0.1179, over 5663841.54 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3653, pruned_loss=0.1179, over 5708163.77 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3683, pruned_loss=0.1182, over 5660140.92 frames. ], batch size: 164, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:33:34,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=821878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 16:33:54,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9640, 1.3413, 1.1277, 0.1635], device='cuda:1'), covar=tensor([0.3927, 0.2876, 0.4309, 0.6333], device='cuda:1'), in_proj_covar=tensor([0.1698, 0.1609, 0.1567, 0.1385], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 16:34:05,032 INFO [train.py:968] (1/2) Epoch 18, batch 45200, giga_loss[loss=0.3864, simple_loss=0.4076, pruned_loss=0.1826, over 26589.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3668, pruned_loss=0.1181, over 5648201.07 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3655, pruned_loss=0.1179, over 5701802.89 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.367, pruned_loss=0.1182, over 5649404.69 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:34:07,130 INFO [optim.py:369] (1/2) 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 16:34:11,678 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 16:34:20,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5349, 1.6698, 1.2575, 1.2596], device='cuda:1'), covar=tensor([0.0960, 0.0606, 0.1076, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0450, 0.0515, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:34:56,905 INFO [train.py:968] (1/2) Epoch 18, batch 45250, giga_loss[loss=0.3889, simple_loss=0.4179, pruned_loss=0.18, over 26656.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3645, pruned_loss=0.1175, over 5638917.03 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5701560.26 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3647, pruned_loss=0.1177, over 5639264.09 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 8.0
+2023-03-09 16:35:42,863 INFO [train.py:968] (1/2) Epoch 18, batch 45300, giga_loss[loss=0.3153, simple_loss=0.3986, pruned_loss=0.116, over 28980.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3663, pruned_loss=0.1184, over 5640029.22 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3656, pruned_loss=0.1178, over 5704800.91 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3663, pruned_loss=0.1185, over 5636775.29 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:35:44,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9809, 1.3647, 5.4665, 4.0194], device='cuda:1'), covar=tensor([0.1817, 0.2980, 0.0600, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0639, 0.0943, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:35:45,901 INFO [optim.py:369] (1/2) 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 16:36:17,308 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 45350, giga_loss[loss=0.3399, simple_loss=0.4055, pruned_loss=0.1372, over 28678.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 5652220.27 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3657, pruned_loss=0.1178, over 5710332.61 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3684, pruned_loss=0.1192, over 5642712.85 frames. ], batch size: 262, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:36:45,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3442, 1.3722, 1.3171, 1.3005], device='cuda:1'), covar=tensor([0.1757, 0.2033, 0.1583, 0.1727], device='cuda:1'), in_proj_covar=tensor([0.1918, 0.1840, 0.1776, 0.1922], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 16:36:46,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 16:37:12,975 INFO [train.py:968] (1/2) Epoch 18, batch 45400, giga_loss[loss=0.3346, simple_loss=0.386, pruned_loss=0.1416, over 28695.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3696, pruned_loss=0.1199, over 5640606.45 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3662, pruned_loss=0.1181, over 5710513.77 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5630952.07 frames. ], batch size: 307, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:37:15,823 INFO [optim.py:369] (1/2) 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 16:37:58,629 INFO [train.py:968] (1/2) Epoch 18, batch 45450, libri_loss[loss=0.3722, simple_loss=0.4326, pruned_loss=0.1559, over 29675.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3686, pruned_loss=0.1194, over 5633960.38 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3666, pruned_loss=0.1183, over 5706833.64 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.368, pruned_loss=0.1191, over 5627843.47 frames. ], batch size: 88, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:37:59,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:38:30,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3643, 2.0130, 1.6102, 0.5706], device='cuda:1'), covar=tensor([0.4742, 0.2720, 0.3685, 0.5631], device='cuda:1'), in_proj_covar=tensor([0.1691, 0.1604, 0.1564, 0.1383], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 16:38:34,337 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 45500, giga_loss[loss=0.3243, simple_loss=0.3877, pruned_loss=0.1304, over 28234.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5624883.22 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 5690262.77 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.369, pruned_loss=0.1201, over 5631899.67 frames. ], batch size: 368, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:38:45,341 INFO [optim.py:369] (1/2) 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 16:39:25,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822253.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 16:39:31,601 INFO [train.py:968] (1/2) Epoch 18, batch 45550, giga_loss[loss=0.3715, simple_loss=0.4082, pruned_loss=0.1674, over 26482.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3724, pruned_loss=0.1224, over 5642391.89 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3668, pruned_loss=0.1183, over 5694376.14 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.372, pruned_loss=0.1224, over 5643230.22 frames. ], batch size: 555, lr: 1.75e-03, grad_scale: 2.0
+2023-03-09 16:40:13,256 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:40:14,985 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 18, batch 45600, giga_loss[loss=0.3133, simple_loss=0.3743, pruned_loss=0.1261, over 28556.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3736, pruned_loss=0.123, over 5651803.51 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3667, pruned_loss=0.1181, over 5696870.30 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3735, pruned_loss=0.1232, over 5649369.92 frames. ], batch size: 71, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:40:22,957 INFO [optim.py:369] (1/2) 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 16:40:37,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4895, 1.7243, 1.6851, 1.4977], device='cuda:1'), covar=tensor([0.1713, 0.1782, 0.2124, 0.1858], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0749, 0.0706, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:40:42,898 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:40:51,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:40:54,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:41:09,542 INFO [train.py:968] (1/2) Epoch 18, batch 45650, giga_loss[loss=0.3127, simple_loss=0.3766, pruned_loss=0.1244, over 28571.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3765, pruned_loss=0.1257, over 5654532.70 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3668, pruned_loss=0.1181, over 5698644.58 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3764, pruned_loss=0.1259, over 5650704.52 frames. ], batch size: 336, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:41:12,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9999, 1.2077, 1.2631, 1.0450], device='cuda:1'), covar=tensor([0.1237, 0.1111, 0.1788, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0748, 0.0706, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:41:25,302 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:41:43,401 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822396.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 16:41:46,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822399.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 16:41:58,345 INFO [train.py:968] (1/2) Epoch 18, batch 45700, giga_loss[loss=0.2905, simple_loss=0.3668, pruned_loss=0.1071, over 28474.00 frames. ], tot_loss[loss=0.3162, simple_loss=0.378, pruned_loss=0.1272, over 5636177.94 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3669, pruned_loss=0.1183, over 5685206.32 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3781, pruned_loss=0.1274, over 5645673.21 frames. ], batch size: 71, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:42:05,493 INFO [optim.py:369] (1/2) 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 16:42:10,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:42:15,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822428.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 16:42:18,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5888, 2.7462, 1.6350, 1.6943], device='cuda:1'), covar=tensor([0.0675, 0.0411, 0.0658, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0547, 0.0374, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 16:42:36,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2609, 3.6918, 1.5116, 1.4196], device='cuda:1'), covar=tensor([0.1025, 0.0360, 0.0910, 0.1391], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0547, 0.0374, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 16:42:45,563 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=822459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:42:47,808 INFO [train.py:968] (1/2) Epoch 18, batch 45750, giga_loss[loss=0.3206, simple_loss=0.3885, pruned_loss=0.1264, over 28860.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3775, pruned_loss=0.125, over 5577999.28 frames. ], libri_tot_loss[loss=0.3028, simple_loss=0.3676, pruned_loss=0.119, over 5615411.59 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3773, pruned_loss=0.1248, over 5649346.84 frames. ], batch size: 119, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:43:20,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=822496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:43:37,010 INFO [train.py:968] (1/2) Epoch 18, batch 45800, giga_loss[loss=0.3316, simple_loss=0.3893, pruned_loss=0.1369, over 28943.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3768, pruned_loss=0.1247, over 5550438.24 frames. ], libri_tot_loss[loss=0.3035, simple_loss=0.368, pruned_loss=0.1195, over 5573764.83 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3764, pruned_loss=0.1242, over 5642257.42 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:43:43,134 INFO [optim.py:369] (1/2) 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 16:44:28,293 INFO [train.py:968] (1/2) Epoch 18, batch 45850, libri_loss[loss=0.3618, simple_loss=0.4129, pruned_loss=0.1553, over 18822.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3761, pruned_loss=0.1253, over 5552451.28 frames. ], libri_tot_loss[loss=0.3042, simple_loss=0.3685, pruned_loss=0.12, over 5547949.19 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3754, pruned_loss=0.1245, over 5648881.63 frames. ], batch size: 186, lr: 1.75e-03, grad_scale: 4.0
+2023-03-09 16:44:31,778 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=822566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:44:34,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:44:38,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=822571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:45:00,306 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-09 16:45:47,783 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=822598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:46:07,113 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 50, giga_loss[loss=0.2883, simple_loss=0.3752, pruned_loss=0.1007, over 28619.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3736, pruned_loss=0.1077, over 1264971.80 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3452, pruned_loss=0.08697, over 58064.57 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3748, pruned_loss=0.1086, over 1219103.32 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:46:48,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6022, 1.6813, 1.2940, 1.3317], device='cuda:1'), covar=tensor([0.0891, 0.0645, 0.0941, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0448, 0.0514, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:46:53,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-09 16:47:16,109 INFO [train.py:968] (1/2) Epoch 19, batch 100, giga_loss[loss=0.2392, simple_loss=0.3244, pruned_loss=0.07702, over 28931.00 frames. ], tot_loss[loss=0.2871, simple_loss=0.3646, pruned_loss=0.1047, over 2242061.18 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3488, pruned_loss=0.09101, over 202633.90 frames. ], giga_tot_loss[loss=0.289, simple_loss=0.3661, pruned_loss=0.106, over 2112930.15 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:47:43,259 INFO [optim.py:369] (1/2) 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 16:48:03,777 INFO [train.py:968] (1/2) Epoch 19, batch 150, giga_loss[loss=0.2492, simple_loss=0.325, pruned_loss=0.08675, over 28871.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3477, pruned_loss=0.09694, over 3008328.64 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3504, pruned_loss=0.09113, over 259792.27 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3479, pruned_loss=0.09749, over 2877701.79 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:48:44,569 INFO [train.py:968] (1/2) Epoch 19, batch 200, giga_loss[loss=0.2256, simple_loss=0.3122, pruned_loss=0.06952, over 28946.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09138, over 3608854.56 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3409, pruned_loss=0.08736, over 399671.79 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3352, pruned_loss=0.09205, over 3446617.01 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:49:10,509 INFO [optim.py:369] (1/2) 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 16:49:27,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:49:31,081 INFO [train.py:968] (1/2) Epoch 19, batch 250, giga_loss[loss=0.2068, simple_loss=0.2694, pruned_loss=0.07205, over 23869.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3242, pruned_loss=0.08588, over 4064379.11 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.342, pruned_loss=0.08808, over 535938.49 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3234, pruned_loss=0.08611, over 3889195.90 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:49:57,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:50:14,423 INFO [train.py:968] (1/2) Epoch 19, batch 300, giga_loss[loss=0.2076, simple_loss=0.2821, pruned_loss=0.06653, over 28882.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.315, pruned_loss=0.08157, over 4431944.67 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3383, pruned_loss=0.08649, over 669155.52 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.314, pruned_loss=0.08169, over 4255353.33 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:50:21,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3196, 4.8585, 1.4936, 1.7494], device='cuda:1'), covar=tensor([0.1211, 0.0338, 0.1000, 0.1554], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0543, 0.0372, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 16:50:41,539 INFO [optim.py:369] (1/2) 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 16:50:58,789 INFO [train.py:968] (1/2) Epoch 19, batch 350, giga_loss[loss=0.2207, simple_loss=0.2877, pruned_loss=0.07691, over 28672.00 frames. ], tot_loss[loss=0.235, simple_loss=0.3102, pruned_loss=0.07988, over 4713701.94 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3387, pruned_loss=0.08814, over 849768.29 frames. ], giga_tot_loss[loss=0.2334, simple_loss=0.3081, pruned_loss=0.07937, over 4529499.84 frames. ], batch size: 92, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:51:04,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=822946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:51:14,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2229, 1.5116, 1.5660, 1.3468], device='cuda:1'), covar=tensor([0.2078, 0.1833, 0.2454, 0.1959], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0746, 0.0703, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 16:51:30,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5108, 3.4631, 1.6537, 1.6489], device='cuda:1'), covar=tensor([0.0982, 0.0299, 0.0882, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0543, 0.0372, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 16:51:30,941 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=822980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:51:38,663 INFO [train.py:968] (1/2) Epoch 19, batch 400, giga_loss[loss=0.1871, simple_loss=0.2603, pruned_loss=0.05698, over 28622.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3063, pruned_loss=0.07766, over 4940450.26 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3373, pruned_loss=0.08731, over 1071831.51 frames. ], giga_tot_loss[loss=0.2286, simple_loss=0.3034, pruned_loss=0.07688, over 4747656.57 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:51:53,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5075, 1.6124, 1.2726, 1.5453], device='cuda:1'), covar=tensor([0.0780, 0.0333, 0.0352, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0103], device='cuda:1')
+2023-03-09 16:51:56,458 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=823009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:51:59,674 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:52:02,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:52:17,849 INFO [train.py:968] (1/2) Epoch 19, batch 450, libri_loss[loss=0.252, simple_loss=0.3325, pruned_loss=0.08574, over 29567.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3048, pruned_loss=0.0772, over 5113567.40 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3351, pruned_loss=0.08611, over 1236947.57 frames. ], giga_tot_loss[loss=0.2274, simple_loss=0.3018, pruned_loss=0.07649, over 4931974.73 frames. ], batch size: 76, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:52:21,008 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-09 16:52:28,402 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 500, giga_loss[loss=0.2622, simple_loss=0.3292, pruned_loss=0.09764, over 28203.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3026, pruned_loss=0.07646, over 5237687.00 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3335, pruned_loss=0.08546, over 1306557.19 frames. ], giga_tot_loss[loss=0.2259, simple_loss=0.3001, pruned_loss=0.07589, over 5084423.37 frames. ], batch size: 368, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:53:06,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=823089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:53:08,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:53:28,671 INFO [optim.py:369] (1/2) 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 16:53:31,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=823121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:53:46,316 INFO [train.py:968] (1/2) Epoch 19, batch 550, giga_loss[loss=0.2044, simple_loss=0.2815, pruned_loss=0.0637, over 28849.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.3, pruned_loss=0.07507, over 5341472.86 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3347, pruned_loss=0.08598, over 1397861.22 frames. ], giga_tot_loss[loss=0.2228, simple_loss=0.2971, pruned_loss=0.07428, over 5208827.44 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:54:30,558 INFO [train.py:968] (1/2) Epoch 19, batch 600, giga_loss[loss=0.1865, simple_loss=0.2654, pruned_loss=0.0538, over 29062.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2978, pruned_loss=0.07382, over 5420614.68 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3366, pruned_loss=0.08717, over 1529809.39 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.294, pruned_loss=0.07257, over 5301645.87 frames. ], batch size: 136, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:54:48,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8033, 1.9671, 2.0366, 1.6114], device='cuda:1'), covar=tensor([0.2121, 0.2491, 0.1644, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0707, 0.0940, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-09 16:54:59,883 INFO [optim.py:369] (1/2) 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 16:55:00,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2976, 1.2799, 3.4228, 3.0938], device='cuda:1'), covar=tensor([0.1518, 0.2711, 0.0497, 0.1897], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0632, 0.0935, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:55:11,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:55:16,121 INFO [train.py:968] (1/2) Epoch 19, batch 650, libri_loss[loss=0.2464, simple_loss=0.3334, pruned_loss=0.07971, over 29246.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.296, pruned_loss=0.07295, over 5477935.23 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3345, pruned_loss=0.08608, over 1690006.29 frames. ], giga_tot_loss[loss=0.2177, simple_loss=0.292, pruned_loss=0.07171, over 5375773.98 frames. ], batch size: 97, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:55:55,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4313, 1.8271, 1.5666, 1.5768], device='cuda:1'), covar=tensor([0.0717, 0.0391, 0.0322, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 16:55:57,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5674, 1.2865, 4.6355, 3.6362], device='cuda:1'), covar=tensor([0.1679, 0.2984, 0.0398, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0632, 0.0935, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 16:56:02,112 INFO [train.py:968] (1/2) Epoch 19, batch 700, libri_loss[loss=0.2607, simple_loss=0.3504, pruned_loss=0.08549, over 29144.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.295, pruned_loss=0.0721, over 5531207.04 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3347, pruned_loss=0.08556, over 1855384.71 frames. ], giga_tot_loss[loss=0.2158, simple_loss=0.2902, pruned_loss=0.07072, over 5434393.19 frames. ], batch size: 101, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:56:25,515 INFO [optim.py:369] (1/2) 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 16:56:44,637 INFO [train.py:968] (1/2) Epoch 19, batch 750, giga_loss[loss=0.1924, simple_loss=0.2684, pruned_loss=0.05823, over 28919.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2939, pruned_loss=0.07199, over 5557730.91 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3349, pruned_loss=0.08577, over 1986044.31 frames. ], giga_tot_loss[loss=0.2147, simple_loss=0.2887, pruned_loss=0.07038, over 5476157.35 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:56:54,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3842, 1.5301, 1.3809, 1.3912], device='cuda:1'), covar=tensor([0.2675, 0.1982, 0.2245, 0.2287], device='cuda:1'), in_proj_covar=tensor([0.1905, 0.1823, 0.1751, 0.1894], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 16:57:29,532 INFO [train.py:968] (1/2) Epoch 19, batch 800, giga_loss[loss=0.1904, simple_loss=0.2717, pruned_loss=0.05453, over 29102.00 frames. ], tot_loss[loss=0.2168, simple_loss=0.291, pruned_loss=0.07128, over 5588359.70 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3349, pruned_loss=0.08579, over 2005662.09 frames. ], giga_tot_loss[loss=0.2133, simple_loss=0.2867, pruned_loss=0.06995, over 5523031.88 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:57:34,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4668, 1.8712, 1.3295, 0.9918], device='cuda:1'), covar=tensor([0.5301, 0.2878, 0.2455, 0.4667], device='cuda:1'), in_proj_covar=tensor([0.1692, 0.1603, 0.1567, 0.1381], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 16:57:58,512 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 850, giga_loss[loss=0.2863, simple_loss=0.3542, pruned_loss=0.1092, over 27694.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2987, pruned_loss=0.07528, over 5611602.99 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3356, pruned_loss=0.08656, over 2139122.09 frames. ], giga_tot_loss[loss=0.2204, simple_loss=0.2937, pruned_loss=0.07354, over 5549780.38 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 16:58:45,346 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:59:02,512 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 16:59:04,609 INFO [train.py:968] (1/2) Epoch 19, batch 900, giga_loss[loss=0.2905, simple_loss=0.3618, pruned_loss=0.1096, over 29023.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3118, pruned_loss=0.08173, over 5632679.37 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3358, pruned_loss=0.08669, over 2249225.12 frames. ], giga_tot_loss[loss=0.2337, simple_loss=0.307, pruned_loss=0.08016, over 5576593.89 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 16:59:09,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5015, 1.6170, 1.5676, 1.4190], device='cuda:1'), covar=tensor([0.2135, 0.2065, 0.1970, 0.2239], device='cuda:1'), in_proj_covar=tensor([0.1907, 0.1821, 0.1752, 0.1898], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 16:59:17,707 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 16:59:32,754 INFO [optim.py:369] (1/2) 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 16:59:46,913 INFO [train.py:968] (1/2) Epoch 19, batch 950, giga_loss[loss=0.3334, simple_loss=0.3909, pruned_loss=0.1379, over 27551.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3231, pruned_loss=0.08734, over 5645736.63 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3353, pruned_loss=0.08629, over 2335637.14 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3191, pruned_loss=0.08621, over 5598869.59 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:00:01,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1946, 1.2914, 1.0728, 0.8983], device='cuda:1'), covar=tensor([0.0877, 0.0451, 0.1003, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0447, 0.0513, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 17:00:13,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2621, 1.5202, 1.6328, 1.2974], device='cuda:1'), covar=tensor([0.1941, 0.1750, 0.2154, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0743, 0.0700, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 17:00:27,476 INFO [train.py:968] (1/2) Epoch 19, batch 1000, giga_loss[loss=0.3033, simple_loss=0.3791, pruned_loss=0.1137, over 28766.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3313, pruned_loss=0.09093, over 5657889.50 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3367, pruned_loss=0.08715, over 2405830.54 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3275, pruned_loss=0.08978, over 5616879.28 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:00:45,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=823609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:00:52,674 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 1050, giga_loss[loss=0.2568, simple_loss=0.337, pruned_loss=0.08833, over 29010.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3354, pruned_loss=0.09168, over 5670106.24 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3391, pruned_loss=0.08881, over 2541756.53 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3313, pruned_loss=0.09023, over 5629699.04 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:01:50,708 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 17:01:51,441 INFO [train.py:968] (1/2) Epoch 19, batch 1100, giga_loss[loss=0.2622, simple_loss=0.3388, pruned_loss=0.09281, over 28884.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3368, pruned_loss=0.09153, over 5666564.25 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3394, pruned_loss=0.08927, over 2623191.92 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3333, pruned_loss=0.09025, over 5630839.18 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:02:14,592 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 1150, giga_loss[loss=0.2814, simple_loss=0.3571, pruned_loss=0.1028, over 27625.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3395, pruned_loss=0.0926, over 5667530.04 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.34, pruned_loss=0.08923, over 2769651.88 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3363, pruned_loss=0.09169, over 5640088.82 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:02:47,047 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=823752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:02:48,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:03:17,312 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 1200, giga_loss[loss=0.2885, simple_loss=0.361, pruned_loss=0.108, over 28951.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3419, pruned_loss=0.09484, over 5672633.46 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3396, pruned_loss=0.08895, over 2845667.75 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3397, pruned_loss=0.09437, over 5646847.36 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:03:48,701 INFO [optim.py:369] (1/2) 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 17:04:04,774 INFO [train.py:968] (1/2) Epoch 19, batch 1250, giga_loss[loss=0.3133, simple_loss=0.376, pruned_loss=0.1253, over 28588.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3466, pruned_loss=0.09811, over 5679296.04 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3403, pruned_loss=0.08941, over 2904401.72 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3445, pruned_loss=0.09767, over 5656542.81 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:04:10,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=823842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:04:20,189 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:04:42,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=823879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:04:51,860 INFO [train.py:968] (1/2) Epoch 19, batch 1300, giga_loss[loss=0.2638, simple_loss=0.3488, pruned_loss=0.0894, over 28228.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3495, pruned_loss=0.09889, over 5690140.44 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3402, pruned_loss=0.08918, over 2949385.62 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.348, pruned_loss=0.09878, over 5668979.22 frames. ], batch size: 77, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:04:56,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5980, 1.7509, 1.4285, 1.8678], device='cuda:1'), covar=tensor([0.2553, 0.2656, 0.2904, 0.2419], device='cuda:1'), in_proj_covar=tensor([0.1461, 0.1061, 0.1295, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 17:05:15,254 INFO [optim.py:369] (1/2) 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 17:05:23,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2237, 4.9857, 4.7262, 2.3413], device='cuda:1'), covar=tensor([0.0394, 0.0561, 0.0614, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.1157, 0.1080, 0.0922, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 17:05:29,817 INFO [train.py:968] (1/2) Epoch 19, batch 1350, giga_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.089, over 28926.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3507, pruned_loss=0.09914, over 5693300.99 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3403, pruned_loss=0.08909, over 3118169.43 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.09966, over 5667697.25 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:06:03,591 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=823976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:12,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=823985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:13,892 INFO [train.py:968] (1/2) Epoch 19, batch 1400, giga_loss[loss=0.2494, simple_loss=0.3347, pruned_loss=0.08204, over 28522.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3522, pruned_loss=0.09916, over 5696236.36 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3397, pruned_loss=0.08878, over 3173515.54 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3522, pruned_loss=0.0999, over 5672206.05 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:06:14,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=823988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:22,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3534, 1.4851, 1.3787, 1.5214], device='cuda:1'), covar=tensor([0.0798, 0.0337, 0.0337, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 17:06:23,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-09 17:06:29,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:35,016 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:35,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-09 17:06:38,951 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:40,646 INFO [optim.py:369] (1/2) 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 17:06:43,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:45,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:06:57,835 INFO [train.py:968] (1/2) Epoch 19, batch 1450, giga_loss[loss=0.2562, simple_loss=0.3409, pruned_loss=0.08571, over 28533.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3517, pruned_loss=0.0977, over 5695351.29 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3397, pruned_loss=0.08883, over 3244219.49 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.352, pruned_loss=0.09853, over 5679853.89 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:07:10,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:07:28,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-09 17:07:38,248 INFO [train.py:968] (1/2) Epoch 19, batch 1500, giga_loss[loss=0.2633, simple_loss=0.351, pruned_loss=0.08785, over 28709.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3509, pruned_loss=0.09644, over 5703190.70 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3402, pruned_loss=0.08904, over 3336036.29 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3512, pruned_loss=0.09726, over 5684294.29 frames. ], batch size: 92, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:07:44,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3546, 3.2643, 1.4253, 1.4371], device='cuda:1'), covar=tensor([0.1057, 0.0249, 0.0925, 0.1443], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0537, 0.0371, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 17:07:47,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-09 17:07:50,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2164, 0.9257, 0.9984, 1.3596], device='cuda:1'), covar=tensor([0.0746, 0.0450, 0.0358, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 17:08:03,867 INFO [optim.py:369] (1/2) 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 17:08:17,000 INFO [train.py:968] (1/2) Epoch 19, batch 1550, giga_loss[loss=0.2385, simple_loss=0.3272, pruned_loss=0.07487, over 28923.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3488, pruned_loss=0.09419, over 5710307.13 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3398, pruned_loss=0.08849, over 3429219.91 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3497, pruned_loss=0.09541, over 5694867.52 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:09:00,411 INFO [train.py:968] (1/2) Epoch 19, batch 1600, giga_loss[loss=0.2435, simple_loss=0.3273, pruned_loss=0.07981, over 29019.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3489, pruned_loss=0.09446, over 5697949.71 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3397, pruned_loss=0.08837, over 3468691.02 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3497, pruned_loss=0.09558, over 5690185.25 frames. ], batch size: 136, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:09:11,618 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:09:29,001 INFO [optim.py:369] (1/2) 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 17:09:36,009 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 1650, giga_loss[loss=0.2966, simple_loss=0.3612, pruned_loss=0.116, over 28895.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3505, pruned_loss=0.09725, over 5693144.75 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3406, pruned_loss=0.08898, over 3542178.01 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.351, pruned_loss=0.09804, over 5690161.85 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:10:07,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3604, 4.3873, 1.6108, 1.6545], device='cuda:1'), covar=tensor([0.1319, 0.0329, 0.1002, 0.1625], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0538, 0.0370, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 17:10:35,669 INFO [train.py:968] (1/2) Epoch 19, batch 1700, giga_loss[loss=0.2848, simple_loss=0.3568, pruned_loss=0.1064, over 29079.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.354, pruned_loss=0.1019, over 5698993.30 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3406, pruned_loss=0.08898, over 3542178.01 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3544, pruned_loss=0.1025, over 5696671.67 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:11:04,742 INFO [optim.py:369] (1/2) 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 17:11:20,299 INFO [train.py:968] (1/2) Epoch 19, batch 1750, giga_loss[loss=0.2367, simple_loss=0.3139, pruned_loss=0.07975, over 28954.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3537, pruned_loss=0.1032, over 5701065.49 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3405, pruned_loss=0.08894, over 3575874.16 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3542, pruned_loss=0.1039, over 5697789.11 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:11:32,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:11:49,597 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:11:52,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:11:54,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:12:01,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:12:02,184 INFO [train.py:968] (1/2) Epoch 19, batch 1800, giga_loss[loss=0.3214, simple_loss=0.3819, pruned_loss=0.1304, over 27934.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3509, pruned_loss=0.102, over 5694234.69 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3404, pruned_loss=0.08877, over 3666727.76 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3519, pruned_loss=0.1032, over 5684721.78 frames. ], batch size: 412, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:12:14,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:12:20,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1943, 1.0908, 4.0470, 3.3598], device='cuda:1'), covar=tensor([0.1701, 0.2914, 0.0396, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0632, 0.0933, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 17:12:28,659 INFO [optim.py:369] (1/2) 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 17:12:43,947 INFO [train.py:968] (1/2) Epoch 19, batch 1850, giga_loss[loss=0.2437, simple_loss=0.3198, pruned_loss=0.08376, over 28613.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3496, pruned_loss=0.1012, over 5693835.44 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3402, pruned_loss=0.0886, over 3718360.62 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3507, pruned_loss=0.1025, over 5685180.32 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:13:12,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4204, 1.8806, 1.4000, 0.7863], device='cuda:1'), covar=tensor([0.4939, 0.2437, 0.3226, 0.5678], device='cuda:1'), in_proj_covar=tensor([0.1688, 0.1597, 0.1564, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 17:13:20,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8589, 3.6834, 3.4995, 1.8133], device='cuda:1'), covar=tensor([0.0538, 0.0721, 0.0668, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.1162, 0.1081, 0.0925, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 17:13:27,414 INFO [train.py:968] (1/2) Epoch 19, batch 1900, giga_loss[loss=0.2471, simple_loss=0.3253, pruned_loss=0.0844, over 28789.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3495, pruned_loss=0.1009, over 5688921.27 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3404, pruned_loss=0.08868, over 3750105.89 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3504, pruned_loss=0.1021, over 5680136.56 frames. ], batch size: 99, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:13:33,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:13:36,335 INFO [zipformer.py:1188] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.5683, 1.6960, 1.4360, 1.7903], device='cuda:1'), covar=tensor([0.2571, 0.2667, 0.2850, 0.2533], device='cuda:1'), in_proj_covar=tensor([0.1454, 0.1055, 0.1287, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 17:13:59,836 INFO [optim.py:369] (1/2) 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 17:14:00,805 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:00,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6011, 1.6790, 1.8752, 1.4178], device='cuda:1'), covar=tensor([0.1850, 0.2414, 0.1432, 0.1745], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0700, 0.0935, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0013], device='cuda:1')
+2023-03-09 17:14:03,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:06,264 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:09,385 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:13,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:18,617 INFO [train.py:968] (1/2) Epoch 19, batch 1950, giga_loss[loss=0.2501, simple_loss=0.3326, pruned_loss=0.08376, over 28693.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.346, pruned_loss=0.09819, over 5686737.09 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3409, pruned_loss=0.08882, over 3792206.92 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3466, pruned_loss=0.09925, over 5676722.09 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:14:31,631 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:39,159 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:14:50,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:15:03,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-09 17:15:05,233 INFO [train.py:968] (1/2) Epoch 19, batch 2000, giga_loss[loss=0.2596, simple_loss=0.3227, pruned_loss=0.09825, over 26555.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3407, pruned_loss=0.09531, over 5674934.36 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3418, pruned_loss=0.08919, over 3820515.46 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3406, pruned_loss=0.09605, over 5675376.78 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:15:08,898 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9358, 1.2016, 1.2848, 1.1151], device='cuda:1'), covar=tensor([0.1849, 0.1499, 0.2311, 0.1539], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0744, 0.0701, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 17:15:11,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:15:37,963 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 2050, giga_loss[loss=0.2307, simple_loss=0.3106, pruned_loss=0.07535, over 29017.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3361, pruned_loss=0.09314, over 5667058.63 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3424, pruned_loss=0.08942, over 3850755.53 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3356, pruned_loss=0.09364, over 5664885.68 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:16:15,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3823, 2.6244, 2.5319, 1.9730], device='cuda:1'), covar=tensor([0.2380, 0.1844, 0.1779, 0.2373], device='cuda:1'), in_proj_covar=tensor([0.1906, 0.1827, 0.1764, 0.1903], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 17:16:17,191 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=824665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:16:39,108 INFO [train.py:968] (1/2) Epoch 19, batch 2100, giga_loss[loss=0.2215, simple_loss=0.2995, pruned_loss=0.07174, over 28924.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3316, pruned_loss=0.09089, over 5651737.89 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3421, pruned_loss=0.08934, over 3907205.41 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3311, pruned_loss=0.09142, over 5654999.57 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:16:57,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-09 17:17:05,992 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=824718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:17:07,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=824721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:17:08,142 INFO [optim.py:369] (1/2) 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 17:17:19,623 INFO [train.py:968] (1/2) Epoch 19, batch 2150, giga_loss[loss=0.2525, simple_loss=0.3397, pruned_loss=0.08265, over 28779.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3321, pruned_loss=0.09029, over 5665933.21 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3416, pruned_loss=0.08889, over 3995502.87 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3316, pruned_loss=0.09107, over 5658136.93 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:17:29,804 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=824750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:17:35,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 17:17:59,292 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-09 17:18:00,048 INFO [train.py:968] (1/2) Epoch 19, batch 2200, giga_loss[loss=0.26, simple_loss=0.3402, pruned_loss=0.08993, over 28715.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.333, pruned_loss=0.09036, over 5684144.28 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3418, pruned_loss=0.08905, over 4051866.96 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3322, pruned_loss=0.09092, over 5671738.94 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:18:28,391 INFO [optim.py:369] (1/2) 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 17:18:44,315 INFO [train.py:968] (1/2) Epoch 19, batch 2250, giga_loss[loss=0.231, simple_loss=0.3156, pruned_loss=0.07316, over 28878.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3314, pruned_loss=0.08961, over 5691702.39 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3421, pruned_loss=0.08908, over 4098027.32 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3304, pruned_loss=0.09007, over 5676579.11 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:19:23,285 INFO [train.py:968] (1/2) Epoch 19, batch 2300, giga_loss[loss=0.2317, simple_loss=0.314, pruned_loss=0.07472, over 28956.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3292, pruned_loss=0.08827, over 5691355.72 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08893, over 4132049.49 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.328, pruned_loss=0.08874, over 5684378.77 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:19:51,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4033, 1.4746, 1.3034, 1.4846], device='cuda:1'), covar=tensor([0.0811, 0.0353, 0.0350, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 17:19:54,223 INFO [optim.py:369] (1/2) 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 17:20:07,440 INFO [train.py:968] (1/2) Epoch 19, batch 2350, giga_loss[loss=0.2547, simple_loss=0.3275, pruned_loss=0.09098, over 28921.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3257, pruned_loss=0.08635, over 5698299.04 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08906, over 4153811.84 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3243, pruned_loss=0.0866, over 5694285.61 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:20:34,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=824968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:20:48,676 INFO [train.py:968] (1/2) Epoch 19, batch 2400, libri_loss[loss=0.288, simple_loss=0.3772, pruned_loss=0.09945, over 29650.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3238, pruned_loss=0.08496, over 5703162.48 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3435, pruned_loss=0.08905, over 4222891.01 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3212, pruned_loss=0.085, over 5691971.03 frames. ], batch size: 91, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:20:48,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7302, 4.5595, 4.3644, 2.0260], device='cuda:1'), covar=tensor([0.0560, 0.0726, 0.0861, 0.2021], device='cuda:1'), in_proj_covar=tensor([0.1156, 0.1073, 0.0916, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 17:21:17,582 INFO [optim.py:369] (1/2) 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 17:21:30,609 INFO [train.py:968] (1/2) Epoch 19, batch 2450, libri_loss[loss=0.2602, simple_loss=0.3454, pruned_loss=0.08756, over 29563.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3212, pruned_loss=0.08377, over 5708432.55 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3441, pruned_loss=0.08922, over 4261474.58 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3183, pruned_loss=0.08359, over 5697421.16 frames. ], batch size: 79, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:21:32,453 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 2500, giga_loss[loss=0.2149, simple_loss=0.2832, pruned_loss=0.07324, over 28609.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3197, pruned_loss=0.08348, over 5713101.60 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3439, pruned_loss=0.08917, over 4300832.82 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.317, pruned_loss=0.08323, over 5701122.71 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:22:28,621 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=825114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:22:37,616 INFO [optim.py:369] (1/2) 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 17:22:50,248 INFO [train.py:968] (1/2) Epoch 19, batch 2550, libri_loss[loss=0.2691, simple_loss=0.3616, pruned_loss=0.08825, over 29528.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3174, pruned_loss=0.08254, over 5709485.02 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3444, pruned_loss=0.08936, over 4311335.59 frames. ], giga_tot_loss[loss=0.2394, simple_loss=0.3146, pruned_loss=0.08212, over 5710230.49 frames. ], batch size: 83, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:22:54,848 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=825143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:20,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=825175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:26,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=825186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:30,695 INFO [train.py:968] (1/2) Epoch 19, batch 2600, giga_loss[loss=0.2376, simple_loss=0.3048, pruned_loss=0.08523, over 24244.00 frames. ], tot_loss[loss=0.2408, simple_loss=0.3168, pruned_loss=0.08237, over 5715332.05 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3448, pruned_loss=0.08954, over 4351064.61 frames. ], giga_tot_loss[loss=0.2385, simple_loss=0.3136, pruned_loss=0.08175, over 5711214.41 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:23:51,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=825215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:23:57,878 INFO [optim.py:369] (1/2) 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 17:24:10,349 INFO [train.py:968] (1/2) Epoch 19, batch 2650, giga_loss[loss=0.3082, simple_loss=0.3643, pruned_loss=0.126, over 27675.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3166, pruned_loss=0.08257, over 5721268.48 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3455, pruned_loss=0.08985, over 4388083.09 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.313, pruned_loss=0.08173, over 5715370.30 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:24:51,050 INFO [train.py:968] (1/2) Epoch 19, batch 2700, giga_loss[loss=0.2524, simple_loss=0.3308, pruned_loss=0.08701, over 28884.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3171, pruned_loss=0.08329, over 5724420.52 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3454, pruned_loss=0.08974, over 4403238.20 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3141, pruned_loss=0.08263, over 5718177.78 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:25:23,229 INFO [optim.py:369] (1/2) 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 17:25:34,910 INFO [train.py:968] (1/2) Epoch 19, batch 2750, giga_loss[loss=0.303, simple_loss=0.3726, pruned_loss=0.1167, over 28717.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3222, pruned_loss=0.08642, over 5716796.53 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3451, pruned_loss=0.08959, over 4430564.73 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3195, pruned_loss=0.0859, over 5716224.52 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:26:01,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5000, 1.6357, 1.7461, 1.3122], device='cuda:1'), covar=tensor([0.2076, 0.2901, 0.1748, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0705, 0.0942, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-09 17:26:19,255 INFO [train.py:968] (1/2) Epoch 19, batch 2800, giga_loss[loss=0.2839, simple_loss=0.3518, pruned_loss=0.108, over 28972.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.328, pruned_loss=0.08922, over 5711574.80 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3455, pruned_loss=0.08964, over 4485610.81 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3247, pruned_loss=0.08871, over 5712032.47 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:26:52,041 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 2850, giga_loss[loss=0.2803, simple_loss=0.3567, pruned_loss=0.102, over 28907.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3374, pruned_loss=0.09605, over 5695702.18 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3456, pruned_loss=0.08976, over 4506470.90 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3345, pruned_loss=0.0956, over 5693819.92 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:27:52,859 INFO [train.py:968] (1/2) Epoch 19, batch 2900, giga_loss[loss=0.2685, simple_loss=0.3534, pruned_loss=0.09174, over 28526.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3407, pruned_loss=0.09692, over 5704381.09 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3449, pruned_loss=0.08942, over 4541147.12 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3387, pruned_loss=0.09697, over 5698358.09 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:28:30,936 INFO [optim.py:369] (1/2) 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 17:28:42,650 INFO [train.py:968] (1/2) Epoch 19, batch 2950, giga_loss[loss=0.434, simple_loss=0.456, pruned_loss=0.2059, over 26499.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3446, pruned_loss=0.09784, over 5701351.52 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3451, pruned_loss=0.08947, over 4566481.27 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3428, pruned_loss=0.09802, over 5699793.33 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:28:46,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-09 17:28:53,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=825550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:29:28,822 INFO [train.py:968] (1/2) Epoch 19, batch 3000, giga_loss[loss=0.2827, simple_loss=0.3478, pruned_loss=0.1088, over 23401.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3503, pruned_loss=0.1016, over 5690045.78 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3455, pruned_loss=0.08984, over 4603808.80 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3486, pruned_loss=0.1017, over 5685374.88 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:29:28,823 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 17:29:36,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3448, 1.2662, 1.1104, 1.5467], device='cuda:1'), covar=tensor([0.0804, 0.0345, 0.0361, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 17:29:37,563 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 17:29:38,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=825589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:30:05,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4394, 1.6507, 1.4122, 1.6177], device='cuda:1'), covar=tensor([0.0785, 0.0318, 0.0320, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0115, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 17:30:09,229 INFO [optim.py:369] (1/2) 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 17:30:21,778 INFO [train.py:968] (1/2) Epoch 19, batch 3050, giga_loss[loss=0.2699, simple_loss=0.3464, pruned_loss=0.09667, over 28960.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.354, pruned_loss=0.1038, over 5682752.75 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3462, pruned_loss=0.09035, over 4618891.35 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3524, pruned_loss=0.1037, over 5679753.02 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:31:02,210 INFO [train.py:968] (1/2) Epoch 19, batch 3100, giga_loss[loss=0.2706, simple_loss=0.3469, pruned_loss=0.09717, over 29087.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3501, pruned_loss=0.1008, over 5693799.69 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3462, pruned_loss=0.09046, over 4661447.34 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.349, pruned_loss=0.1011, over 5686480.21 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:31:08,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=825693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:31:11,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=825696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:31:31,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.49 vs. limit=5.0
+2023-03-09 17:31:33,146 INFO [optim.py:369] (1/2) 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 17:31:33,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=825725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:31:34,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5460, 1.7943, 1.7421, 1.6106], device='cuda:1'), covar=tensor([0.1922, 0.1995, 0.2261, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0743, 0.0703, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 17:31:47,861 INFO [train.py:968] (1/2) Epoch 19, batch 3150, giga_loss[loss=0.285, simple_loss=0.3527, pruned_loss=0.1087, over 28920.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3482, pruned_loss=0.09863, over 5703211.81 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3461, pruned_loss=0.09044, over 4685563.41 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3474, pruned_loss=0.09903, over 5694200.68 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:31:54,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2185, 1.5784, 1.5585, 1.1135], device='cuda:1'), covar=tensor([0.1846, 0.2583, 0.1482, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0700, 0.0934, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0013], device='cuda:1')
+2023-03-09 17:32:31,703 INFO [train.py:968] (1/2) Epoch 19, batch 3200, giga_loss[loss=0.338, simple_loss=0.4, pruned_loss=0.138, over 28702.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3483, pruned_loss=0.09841, over 5704877.96 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3459, pruned_loss=0.09021, over 4711706.30 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.348, pruned_loss=0.09908, over 5696595.03 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:32:35,882 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-09 17:33:01,984 INFO [optim.py:369] (1/2) 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 17:33:16,275 INFO [train.py:968] (1/2) Epoch 19, batch 3250, giga_loss[loss=0.2426, simple_loss=0.3315, pruned_loss=0.07681, over 29028.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3499, pruned_loss=0.09864, over 5705557.48 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3455, pruned_loss=0.08994, over 4737871.22 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3499, pruned_loss=0.09957, over 5697823.60 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:33:55,312 INFO [train.py:968] (1/2) Epoch 19, batch 3300, giga_loss[loss=0.2635, simple_loss=0.3404, pruned_loss=0.09335, over 28644.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3509, pruned_loss=0.09915, over 5710657.32 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3458, pruned_loss=0.09014, over 4775105.38 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3509, pruned_loss=0.1, over 5700608.87 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:34:30,108 INFO [optim.py:369] (1/2) 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 17:34:42,061 INFO [train.py:968] (1/2) Epoch 19, batch 3350, giga_loss[loss=0.2579, simple_loss=0.3388, pruned_loss=0.08854, over 29005.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.353, pruned_loss=0.101, over 5708741.45 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3456, pruned_loss=0.09007, over 4794061.74 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3532, pruned_loss=0.102, over 5700107.78 frames. ], batch size: 136, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:35:02,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=825964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:35:10,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6381, 2.3413, 1.7912, 0.7187], device='cuda:1'), covar=tensor([0.4553, 0.2744, 0.3786, 0.5784], device='cuda:1'), in_proj_covar=tensor([0.1688, 0.1596, 0.1567, 0.1385], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 17:35:20,038 INFO [train.py:968] (1/2) Epoch 19, batch 3400, giga_loss[loss=0.3038, simple_loss=0.3707, pruned_loss=0.1185, over 28889.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3543, pruned_loss=0.1027, over 5713676.63 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3457, pruned_loss=0.09022, over 4831951.56 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3547, pruned_loss=0.1037, over 5700872.43 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:35:44,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:35:52,009 INFO [optim.py:369] (1/2) 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 17:35:55,230 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 3450, giga_loss[loss=0.2699, simple_loss=0.3448, pruned_loss=0.0975, over 28680.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3541, pruned_loss=0.1027, over 5726544.63 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3457, pruned_loss=0.09031, over 4874163.51 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3548, pruned_loss=0.104, over 5709584.81 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:36:25,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-09 17:36:47,783 INFO [train.py:968] (1/2) Epoch 19, batch 3500, giga_loss[loss=0.2519, simple_loss=0.3327, pruned_loss=0.08552, over 28886.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3546, pruned_loss=0.1027, over 5726553.03 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3457, pruned_loss=0.09023, over 4890190.43 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3553, pruned_loss=0.104, over 5714209.11 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:36:48,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5739, 2.2198, 1.7493, 0.7552], device='cuda:1'), covar=tensor([0.5804, 0.2740, 0.3751, 0.6089], device='cuda:1'), in_proj_covar=tensor([0.1689, 0.1593, 0.1562, 0.1382], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 17:37:02,887 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:37:07,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:37:17,708 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 3550, giga_loss[loss=0.2768, simple_loss=0.3513, pruned_loss=0.1012, over 28714.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3546, pruned_loss=0.1021, over 5721387.27 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3457, pruned_loss=0.09022, over 4903649.65 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3553, pruned_loss=0.1033, over 5710650.01 frames. ], batch size: 92, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:37:29,849 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:37:41,665 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-09 17:37:52,353 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:38:13,584 INFO [train.py:968] (1/2) Epoch 19, batch 3600, giga_loss[loss=0.2937, simple_loss=0.3674, pruned_loss=0.11, over 28689.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3548, pruned_loss=0.1013, over 5720602.80 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3455, pruned_loss=0.09009, over 4924203.01 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3557, pruned_loss=0.1026, over 5711701.83 frames. ], batch size: 262, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:38:35,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5037, 3.4549, 1.5879, 1.6578], device='cuda:1'), covar=tensor([0.1020, 0.0252, 0.0916, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0536, 0.0370, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 17:38:45,797 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 3650, giga_loss[loss=0.243, simple_loss=0.3211, pruned_loss=0.08239, over 28565.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3535, pruned_loss=0.1001, over 5723980.22 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3456, pruned_loss=0.09021, over 4943944.43 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1012, over 5713321.64 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:39:07,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-09 17:39:26,306 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-09 17:39:32,009 INFO [train.py:968] (1/2) Epoch 19, batch 3700, giga_loss[loss=0.2735, simple_loss=0.3511, pruned_loss=0.09797, over 28966.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.351, pruned_loss=0.09856, over 5732661.91 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3448, pruned_loss=0.08978, over 4987493.81 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3526, pruned_loss=0.1003, over 5719081.02 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:40:04,472 INFO [optim.py:369] (1/2) 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 17:40:05,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-09 17:40:12,648 INFO [train.py:968] (1/2) Epoch 19, batch 3750, giga_loss[loss=0.2891, simple_loss=0.3587, pruned_loss=0.1098, over 28966.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3492, pruned_loss=0.09785, over 5725181.41 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3448, pruned_loss=0.08967, over 4994576.42 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3505, pruned_loss=0.09937, over 5714803.83 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:40:15,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9446, 1.2679, 1.2815, 1.0782], device='cuda:1'), covar=tensor([0.1984, 0.1345, 0.2457, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0741, 0.0701, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 17:40:23,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:40:23,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2790, 3.8786, 1.5433, 1.3986], device='cuda:1'), covar=tensor([0.1052, 0.0257, 0.0897, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0535, 0.0369, 0.0412], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 17:40:36,301 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 3800, giga_loss[loss=0.2265, simple_loss=0.3142, pruned_loss=0.06937, over 28598.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3474, pruned_loss=0.09692, over 5733201.67 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.345, pruned_loss=0.08979, over 5025622.59 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3485, pruned_loss=0.09831, over 5719476.48 frames. ], batch size: 60, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:40:53,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:41:08,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:41:18,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0721, 2.0798, 1.5139, 1.6853], device='cuda:1'), covar=tensor([0.0897, 0.0707, 0.1029, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0442, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 17:41:24,586 INFO [optim.py:369] (1/2) 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 17:41:37,495 INFO [train.py:968] (1/2) Epoch 19, batch 3850, giga_loss[loss=0.2769, simple_loss=0.3542, pruned_loss=0.09978, over 29069.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3487, pruned_loss=0.09829, over 5731256.19 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.345, pruned_loss=0.08979, over 5034393.48 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3496, pruned_loss=0.09945, over 5719079.26 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:41:58,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7175, 1.0512, 2.8382, 2.8129], device='cuda:1'), covar=tensor([0.1774, 0.2624, 0.0567, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0628, 0.0923, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 17:42:16,046 INFO [train.py:968] (1/2) Epoch 19, batch 3900, giga_loss[loss=0.2773, simple_loss=0.3561, pruned_loss=0.09922, over 28659.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3495, pruned_loss=0.09845, over 5736878.69 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3447, pruned_loss=0.08966, over 5051527.04 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3505, pruned_loss=0.09963, over 5724374.30 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:42:22,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2667, 1.5940, 1.5770, 1.4771], device='cuda:1'), covar=tensor([0.1946, 0.1494, 0.2224, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0741, 0.0700, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 17:42:45,600 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 17:42:48,006 INFO [optim.py:369] (1/2) 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 17:42:55,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:42:57,373 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 3950, giga_loss[loss=0.2461, simple_loss=0.3347, pruned_loss=0.07875, over 29121.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3488, pruned_loss=0.09779, over 5731992.85 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3441, pruned_loss=0.08952, over 5083301.10 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3503, pruned_loss=0.09918, over 5717217.67 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:42:58,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:05,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:07,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:20,448 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:34,074 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:43:39,899 INFO [train.py:968] (1/2) Epoch 19, batch 4000, giga_loss[loss=0.2879, simple_loss=0.3604, pruned_loss=0.1077, over 28607.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3479, pruned_loss=0.09686, over 5732656.88 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3444, pruned_loss=0.08975, over 5108940.90 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.349, pruned_loss=0.09804, over 5717865.14 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:44:08,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-09 17:44:10,136 INFO [optim.py:369] (1/2) 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 17:44:13,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2285, 1.2219, 3.8428, 3.1933], device='cuda:1'), covar=tensor([0.1653, 0.2742, 0.0438, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0727, 0.0626, 0.0921, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 17:44:19,523 INFO [train.py:968] (1/2) Epoch 19, batch 4050, libri_loss[loss=0.2291, simple_loss=0.3176, pruned_loss=0.07029, over 29562.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3467, pruned_loss=0.09682, over 5731212.38 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3444, pruned_loss=0.08986, over 5124734.75 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3476, pruned_loss=0.09781, over 5716424.82 frames. ], batch size: 76, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:44:33,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-09 17:44:40,681 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:44:54,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:44:56,354 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 4100, giga_loss[loss=0.2318, simple_loss=0.3094, pruned_loss=0.07708, over 28438.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.344, pruned_loss=0.09537, over 5717675.91 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3449, pruned_loss=0.09019, over 5138229.83 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3442, pruned_loss=0.09604, over 5709864.09 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:45:18,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:45:23,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:45:28,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:45:29,856 INFO [optim.py:369] (1/2) 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 17:45:37,508 INFO [train.py:968] (1/2) Epoch 19, batch 4150, giga_loss[loss=0.2414, simple_loss=0.3218, pruned_loss=0.08053, over 28740.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3406, pruned_loss=0.09393, over 5715898.61 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3445, pruned_loss=0.09004, over 5151905.55 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09465, over 5707965.08 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:45:42,501 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=826743.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:45:45,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=826748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:46:16,450 INFO [train.py:968] (1/2) Epoch 19, batch 4200, giga_loss[loss=0.2245, simple_loss=0.3034, pruned_loss=0.07285, over 28430.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3399, pruned_loss=0.09413, over 5715304.28 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3444, pruned_loss=0.09016, over 5176272.14 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3403, pruned_loss=0.09475, over 5704473.12 frames. ], batch size: 65, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:46:48,330 INFO [optim.py:369] (1/2) 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 17:46:57,872 INFO [train.py:968] (1/2) Epoch 19, batch 4250, giga_loss[loss=0.2687, simple_loss=0.3395, pruned_loss=0.09898, over 28868.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3387, pruned_loss=0.09382, over 5712294.95 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3444, pruned_loss=0.09008, over 5190076.05 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3389, pruned_loss=0.09446, over 5700629.67 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:47:25,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:47:27,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:47:38,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-09 17:47:42,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=826886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:47:43,646 INFO [train.py:968] (1/2) Epoch 19, batch 4300, giga_loss[loss=0.2851, simple_loss=0.3487, pruned_loss=0.1108, over 28874.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3369, pruned_loss=0.09318, over 5711083.17 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3442, pruned_loss=0.08997, over 5203648.43 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.337, pruned_loss=0.09384, over 5698894.25 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:47:44,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=826889.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:47:55,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:48:08,421 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=826918.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:48:11,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4852, 2.0846, 1.6026, 0.8054], device='cuda:1'), covar=tensor([0.6203, 0.2964, 0.3801, 0.6135], device='cuda:1'), in_proj_covar=tensor([0.1683, 0.1584, 0.1559, 0.1378], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 17:48:15,693 INFO [optim.py:369] (1/2) 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 17:48:24,451 INFO [train.py:968] (1/2) Epoch 19, batch 4350, giga_loss[loss=0.2388, simple_loss=0.3171, pruned_loss=0.08025, over 28152.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3345, pruned_loss=0.09258, over 5709182.62 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.08989, over 5204774.54 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3347, pruned_loss=0.09319, over 5705718.93 frames. ], batch size: 77, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:49:02,358 INFO [train.py:968] (1/2) Epoch 19, batch 4400, giga_loss[loss=0.238, simple_loss=0.3046, pruned_loss=0.08574, over 28554.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3325, pruned_loss=0.09165, over 5695393.55 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3444, pruned_loss=0.09027, over 5211910.56 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3319, pruned_loss=0.0919, over 5703748.64 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:49:10,281 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:49:42,026 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 4450, giga_loss[loss=0.2574, simple_loss=0.3417, pruned_loss=0.08651, over 28942.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3307, pruned_loss=0.09043, over 5701388.45 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.08993, over 5227204.92 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3304, pruned_loss=0.0909, over 5705190.57 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:49:42,865 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-09 17:50:26,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6450, 2.4526, 1.5730, 0.8696], device='cuda:1'), covar=tensor([0.8124, 0.3873, 0.4008, 0.6784], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1591, 0.1565, 0.1383], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 17:50:27,025 INFO [train.py:968] (1/2) Epoch 19, batch 4500, giga_loss[loss=0.259, simple_loss=0.3448, pruned_loss=0.08663, over 28539.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3325, pruned_loss=0.09066, over 5703571.89 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08974, over 5231752.33 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3324, pruned_loss=0.09118, over 5706561.70 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:50:31,234 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:50:43,310 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:50:57,127 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:51:04,951 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1465, 1.1985, 3.8495, 2.9772], device='cuda:1'), covar=tensor([0.1721, 0.2774, 0.0395, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0628, 0.0925, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 17:51:11,171 INFO [train.py:968] (1/2) Epoch 19, batch 4550, giga_loss[loss=0.2727, simple_loss=0.3567, pruned_loss=0.09432, over 28646.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3357, pruned_loss=0.09202, over 5692898.83 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08988, over 5239051.84 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3352, pruned_loss=0.09237, over 5698320.60 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:51:23,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827154.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:51:32,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4926, 1.5724, 1.7268, 1.3267], device='cuda:1'), covar=tensor([0.1863, 0.2417, 0.1555, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0700, 0.0932, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 17:51:46,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:51:46,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1878, 1.0992, 3.6084, 2.9945], device='cuda:1'), covar=tensor([0.1632, 0.2849, 0.0432, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0628, 0.0925, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 17:51:49,150 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 4600, giga_loss[loss=0.2941, simple_loss=0.3618, pruned_loss=0.1132, over 29056.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.337, pruned_loss=0.09212, over 5698813.28 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08984, over 5248728.66 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3368, pruned_loss=0.09246, over 5700666.46 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 17:52:16,871 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:52:36,411 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 4650, giga_loss[loss=0.2621, simple_loss=0.3421, pruned_loss=0.09101, over 28693.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3396, pruned_loss=0.09287, over 5696952.06 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09011, over 5267226.90 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3389, pruned_loss=0.09297, over 5693650.59 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:52:38,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:04,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:07,176 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:07,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:25,203 INFO [train.py:968] (1/2) Epoch 19, batch 4700, libri_loss[loss=0.3228, simple_loss=0.393, pruned_loss=0.1264, over 19601.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3397, pruned_loss=0.09299, over 5688264.26 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.09029, over 5266633.23 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3389, pruned_loss=0.09294, over 5693015.05 frames. ], batch size: 187, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:53:33,854 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:53:57,743 INFO [optim.py:369] (1/2) 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 17:54:01,067 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 4750, giga_loss[loss=0.2743, simple_loss=0.3453, pruned_loss=0.1016, over 28871.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3399, pruned_loss=0.09381, over 5685951.49 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3449, pruned_loss=0.09082, over 5267088.69 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3385, pruned_loss=0.09334, over 5697359.44 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 17:54:19,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5820, 1.7769, 1.4835, 1.8040], device='cuda:1'), covar=tensor([0.2618, 0.2713, 0.3065, 0.2460], device='cuda:1'), in_proj_covar=tensor([0.1455, 0.1056, 0.1289, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 17:54:34,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:54:47,806 INFO [train.py:968] (1/2) Epoch 19, batch 4800, giga_loss[loss=0.2489, simple_loss=0.3247, pruned_loss=0.08653, over 28616.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3415, pruned_loss=0.09531, over 5684166.67 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3452, pruned_loss=0.09095, over 5273258.93 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3401, pruned_loss=0.09487, over 5693403.57 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:55:07,446 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 19, batch 4850, giga_loss[loss=0.267, simple_loss=0.3394, pruned_loss=0.09729, over 28833.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3439, pruned_loss=0.09675, over 5673407.61 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3454, pruned_loss=0.09104, over 5266302.63 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3426, pruned_loss=0.09635, over 5688394.11 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:56:06,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:56:16,132 INFO [train.py:968] (1/2) Epoch 19, batch 4900, giga_loss[loss=0.2611, simple_loss=0.3423, pruned_loss=0.08996, over 28692.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3462, pruned_loss=0.09778, over 5686666.63 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3451, pruned_loss=0.09089, over 5283673.51 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3454, pruned_loss=0.09775, over 5694142.31 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:56:17,516 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:56:34,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1593, 2.5958, 1.1930, 1.3790], device='cuda:1'), covar=tensor([0.0993, 0.0359, 0.0944, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0539, 0.0370, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0025, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 17:56:40,308 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:56:43,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:56:52,297 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 4950, giga_loss[loss=0.2791, simple_loss=0.346, pruned_loss=0.1061, over 23881.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3485, pruned_loss=0.0986, over 5696878.35 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.345, pruned_loss=0.09082, over 5286848.14 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.348, pruned_loss=0.09866, over 5701728.33 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:57:02,117 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:57:07,752 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 5000, giga_loss[loss=0.2799, simple_loss=0.3425, pruned_loss=0.1087, over 23852.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3499, pruned_loss=0.0992, over 5704322.94 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3458, pruned_loss=0.09112, over 5308959.40 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.349, pruned_loss=0.09933, over 5702992.43 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:57:51,305 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1297, 1.4667, 1.4146, 1.0341], device='cuda:1'), covar=tensor([0.1587, 0.2276, 0.1335, 0.1534], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0699, 0.0930, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 17:57:51,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3301, 1.5491, 1.2596, 0.9488], device='cuda:1'), covar=tensor([0.2574, 0.2715, 0.3044, 0.2449], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1052, 0.1282, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 17:58:08,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:58:11,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:58:14,672 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 5050, giga_loss[loss=0.2887, simple_loss=0.3612, pruned_loss=0.1081, over 28546.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3497, pruned_loss=0.09922, over 5714202.18 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3456, pruned_loss=0.09107, over 5314782.71 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3492, pruned_loss=0.09943, over 5711487.79 frames. ], batch size: 60, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:58:35,148 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 17:58:37,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2916, 4.1110, 3.9129, 2.0779], device='cuda:1'), covar=tensor([0.0589, 0.0726, 0.0709, 0.1896], device='cuda:1'), in_proj_covar=tensor([0.1155, 0.1075, 0.0918, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 17:58:49,124 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827672.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:58:49,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6717, 1.8331, 1.4389, 1.9477], device='cuda:1'), covar=tensor([0.2533, 0.2730, 0.3140, 0.2397], device='cuda:1'), in_proj_covar=tensor([0.1446, 0.1051, 0.1283, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 17:58:51,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827675.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 17:58:56,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5058, 1.7568, 1.3846, 1.4855], device='cuda:1'), covar=tensor([0.2790, 0.2804, 0.3295, 0.2628], device='cuda:1'), in_proj_covar=tensor([0.1447, 0.1052, 0.1283, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 17:59:01,594 INFO [train.py:968] (1/2) Epoch 19, batch 5100, giga_loss[loss=0.2715, simple_loss=0.3398, pruned_loss=0.1016, over 28870.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3483, pruned_loss=0.09791, over 5721350.58 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.346, pruned_loss=0.09135, over 5326303.24 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3475, pruned_loss=0.09798, over 5716133.29 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 17:59:15,530 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:59:18,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827707.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 17:59:36,453 INFO [optim.py:369] (1/2) 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 17:59:43,043 INFO [train.py:968] (1/2) Epoch 19, batch 5150, giga_loss[loss=0.2335, simple_loss=0.319, pruned_loss=0.07403, over 28924.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3466, pruned_loss=0.0971, over 5714515.84 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3469, pruned_loss=0.09192, over 5330811.52 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3453, pruned_loss=0.09679, over 5713172.37 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:00:10,709 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-09 18:00:24,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 18:00:24,762 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:00:25,771 INFO [train.py:968] (1/2) Epoch 19, batch 5200, giga_loss[loss=0.2531, simple_loss=0.3359, pruned_loss=0.08511, over 28869.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3436, pruned_loss=0.09553, over 5720579.37 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3474, pruned_loss=0.09221, over 5335536.99 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3421, pruned_loss=0.09509, over 5720238.47 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:00:32,150 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 18:01:01,345 INFO [optim.py:369] (1/2) 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 18:01:07,713 INFO [train.py:968] (1/2) Epoch 19, batch 5250, giga_loss[loss=0.2549, simple_loss=0.3226, pruned_loss=0.09354, over 23831.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3421, pruned_loss=0.09488, over 5708105.43 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3473, pruned_loss=0.0921, over 5342906.97 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3409, pruned_loss=0.09471, over 5712541.29 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:01:17,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827850.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:01:19,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827853.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:01:28,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=827865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:01:42,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827882.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:01:47,306 INFO [train.py:968] (1/2) Epoch 19, batch 5300, giga_loss[loss=0.3192, simple_loss=0.3953, pruned_loss=0.1216, over 28916.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3435, pruned_loss=0.0948, over 5706511.52 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3476, pruned_loss=0.09228, over 5357407.02 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3422, pruned_loss=0.09461, over 5708563.59 frames. ], batch size: 227, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:02:09,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3635, 1.6396, 1.5152, 1.4162], device='cuda:1'), covar=tensor([0.1725, 0.2128, 0.2267, 0.2203], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0743, 0.0702, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:02:12,370 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:02:23,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=827929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:02:25,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=827932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:02:29,618 INFO [train.py:968] (1/2) Epoch 19, batch 5350, giga_loss[loss=0.2351, simple_loss=0.3039, pruned_loss=0.08313, over 28556.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3462, pruned_loss=0.09565, over 5707359.20 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3481, pruned_loss=0.09276, over 5372845.60 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3445, pruned_loss=0.09515, over 5705911.88 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:02:42,194 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=827953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:02:50,018 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=827961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:03:05,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-09 18:03:12,508 INFO [train.py:968] (1/2) Epoch 19, batch 5400, giga_loss[loss=0.2425, simple_loss=0.328, pruned_loss=0.07851, over 28930.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3458, pruned_loss=0.09558, over 5696316.58 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3487, pruned_loss=0.09324, over 5376358.91 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3439, pruned_loss=0.09484, over 5700046.62 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:03:23,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-09 18:03:27,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:03:30,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:03:45,409 INFO [optim.py:369] (1/2) 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 18:03:52,991 INFO [train.py:968] (1/2) Epoch 19, batch 5450, giga_loss[loss=0.2509, simple_loss=0.3172, pruned_loss=0.09228, over 28866.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3448, pruned_loss=0.09626, over 5699083.23 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3489, pruned_loss=0.09338, over 5381352.10 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3431, pruned_loss=0.09559, over 5700477.95 frames. ], batch size: 112, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:03:56,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:04:11,213 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:04:13,592 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 5500, libri_loss[loss=0.2276, simple_loss=0.316, pruned_loss=0.06966, over 29581.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3452, pruned_loss=0.09808, over 5690751.69 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3495, pruned_loss=0.09384, over 5388344.23 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3432, pruned_loss=0.09727, over 5697500.07 frames. ], batch size: 75, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:04:37,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:04:38,235 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:05:06,865 INFO [optim.py:369] (1/2) 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 18:05:14,196 INFO [train.py:968] (1/2) Epoch 19, batch 5550, giga_loss[loss=0.2359, simple_loss=0.3054, pruned_loss=0.08321, over 28774.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3441, pruned_loss=0.09818, over 5678467.72 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3504, pruned_loss=0.0945, over 5387923.59 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3414, pruned_loss=0.09711, over 5696985.66 frames. ], batch size: 99, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:05:37,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5599, 1.7634, 1.6701, 1.5042], device='cuda:1'), covar=tensor([0.1735, 0.2050, 0.2267, 0.2265], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0745, 0.0706, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:05:54,643 INFO [train.py:968] (1/2) Epoch 19, batch 5600, giga_loss[loss=0.2717, simple_loss=0.3337, pruned_loss=0.1048, over 28547.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3416, pruned_loss=0.0971, over 5690885.92 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3497, pruned_loss=0.09406, over 5404885.19 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3399, pruned_loss=0.09673, over 5699788.52 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:05:59,708 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:06:32,396 INFO [optim.py:369] (1/2) 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 18:06:39,167 INFO [train.py:968] (1/2) Epoch 19, batch 5650, libri_loss[loss=0.2669, simple_loss=0.3493, pruned_loss=0.09224, over 29486.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.34, pruned_loss=0.09571, over 5699074.01 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3495, pruned_loss=0.09392, over 5416022.08 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3383, pruned_loss=0.09565, over 5707490.92 frames. ], batch size: 85, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:07:19,103 INFO [train.py:968] (1/2) Epoch 19, batch 5700, libri_loss[loss=0.2995, simple_loss=0.3809, pruned_loss=0.1091, over 29129.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3365, pruned_loss=0.0942, over 5712063.63 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3502, pruned_loss=0.09432, over 5432710.26 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3343, pruned_loss=0.09385, over 5712702.55 frames. ], batch size: 101, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:07:52,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:07:54,592 INFO [optim.py:369] (1/2) 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 18:07:59,200 INFO [train.py:968] (1/2) Epoch 19, batch 5750, giga_loss[loss=0.2034, simple_loss=0.2848, pruned_loss=0.06095, over 29054.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3305, pruned_loss=0.09107, over 5715594.67 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.35, pruned_loss=0.09425, over 5439792.29 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3286, pruned_loss=0.09082, over 5713543.31 frames. ], batch size: 136, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:08:40,163 INFO [train.py:968] (1/2) Epoch 19, batch 5800, giga_loss[loss=0.2717, simple_loss=0.3476, pruned_loss=0.09787, over 28603.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3308, pruned_loss=0.09122, over 5718728.88 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3502, pruned_loss=0.09458, over 5457116.56 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3283, pruned_loss=0.09061, over 5711190.41 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:08:46,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0858, 1.1450, 3.3496, 2.8836], device='cuda:1'), covar=tensor([0.1632, 0.2727, 0.0484, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0631, 0.0932, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 18:08:47,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1662, 1.2849, 1.1440, 0.9443], device='cuda:1'), covar=tensor([0.0897, 0.0514, 0.1070, 0.1097], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0444, 0.0513, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 18:09:03,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2729, 1.5961, 1.5373, 1.1241], device='cuda:1'), covar=tensor([0.1783, 0.2321, 0.1467, 0.1615], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0698, 0.0929, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 18:09:14,302 INFO [optim.py:369] (1/2) 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 18:09:19,817 INFO [train.py:968] (1/2) Epoch 19, batch 5850, giga_loss[loss=0.2835, simple_loss=0.3649, pruned_loss=0.1011, over 28595.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3336, pruned_loss=0.0927, over 5723815.38 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3503, pruned_loss=0.09461, over 5460737.14 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3315, pruned_loss=0.09218, over 5717037.99 frames. ], batch size: 307, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:09:39,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:09:42,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:09:45,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:09:48,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:10:01,888 INFO [train.py:968] (1/2) Epoch 19, batch 5900, giga_loss[loss=0.2721, simple_loss=0.3509, pruned_loss=0.09664, over 29017.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3367, pruned_loss=0.09361, over 5729724.59 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3501, pruned_loss=0.09448, over 5471928.00 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3348, pruned_loss=0.09327, over 5720297.51 frames. ], batch size: 155, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:10:15,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:10:36,231 INFO [optim.py:369] (1/2) 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 18:10:41,734 INFO [train.py:968] (1/2) Epoch 19, batch 5950, libri_loss[loss=0.3236, simple_loss=0.3933, pruned_loss=0.1269, over 27762.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3398, pruned_loss=0.09481, over 5724421.96 frames. ], libri_tot_loss[loss=0.2695, simple_loss=0.3502, pruned_loss=0.09438, over 5486901.57 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3377, pruned_loss=0.09461, over 5712149.58 frames. ], batch size: 116, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 18:11:08,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828567.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:11:19,131 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 6000, giga_loss[loss=0.2587, simple_loss=0.3367, pruned_loss=0.09037, over 28897.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09634, over 5724196.82 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.35, pruned_loss=0.09432, over 5490970.82 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3411, pruned_loss=0.09625, over 5712911.79 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:11:27,689 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 18:11:35,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3456, 1.2587, 1.1236, 1.5532], device='cuda:1'), covar=tensor([0.0824, 0.0348, 0.0367, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 18:11:36,637 INFO [train.py:1012] (1/2) Epoch 19, validation: loss=0.2128, simple_loss=0.3196, pruned_loss=0.05298, over 944034.00 frames. 
+2023-03-09 18:11:36,638 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 18:11:55,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:11:57,321 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:12:18,123 INFO [train.py:968] (1/2) Epoch 19, batch 6050, giga_loss[loss=0.3081, simple_loss=0.3637, pruned_loss=0.1263, over 23751.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3446, pruned_loss=0.09747, over 5719253.60 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3497, pruned_loss=0.09408, over 5501056.02 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3435, pruned_loss=0.09768, over 5706239.75 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:12:22,524 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:12:43,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-09 18:13:04,847 INFO [train.py:968] (1/2) Epoch 19, batch 6100, giga_loss[loss=0.2745, simple_loss=0.3472, pruned_loss=0.1009, over 28970.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3507, pruned_loss=0.1023, over 5716962.53 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3499, pruned_loss=0.09415, over 5507573.93 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3497, pruned_loss=0.1026, over 5704369.32 frames. ], batch size: 136, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:13:08,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 18:13:27,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828710.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:13:30,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828713.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:13:46,151 INFO [optim.py:369] (1/2) 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 18:13:51,352 INFO [train.py:968] (1/2) Epoch 19, batch 6150, giga_loss[loss=0.2763, simple_loss=0.3563, pruned_loss=0.09814, over 28534.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3557, pruned_loss=0.1067, over 5696056.78 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3496, pruned_loss=0.09409, over 5508846.59 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3552, pruned_loss=0.1073, over 5690481.13 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:13:56,030 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=828742.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:13:56,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5342, 2.4427, 1.8982, 1.9819], device='cuda:1'), covar=tensor([0.0810, 0.0663, 0.1020, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0444, 0.0512, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 18:14:01,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8591, 3.6678, 3.5008, 2.1643], device='cuda:1'), covar=tensor([0.0662, 0.0878, 0.0792, 0.1725], device='cuda:1'), in_proj_covar=tensor([0.1175, 0.1087, 0.0930, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 18:14:31,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9408, 1.3365, 1.1288, 0.1833], device='cuda:1'), covar=tensor([0.3577, 0.2613, 0.3842, 0.5461], device='cuda:1'), in_proj_covar=tensor([0.1704, 0.1606, 0.1575, 0.1390], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 18:14:40,294 INFO [train.py:968] (1/2) Epoch 19, batch 6200, giga_loss[loss=0.3014, simple_loss=0.3715, pruned_loss=0.1157, over 28675.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3633, pruned_loss=0.1123, over 5687106.06 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.349, pruned_loss=0.09376, over 5517138.52 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3637, pruned_loss=0.1134, over 5679399.07 frames. ], batch size: 78, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:14:56,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:15:25,017 INFO [optim.py:369] (1/2) 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 18:15:31,008 INFO [train.py:968] (1/2) Epoch 19, batch 6250, giga_loss[loss=0.3406, simple_loss=0.3987, pruned_loss=0.1413, over 28965.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3702, pruned_loss=0.1179, over 5678681.43 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3496, pruned_loss=0.09408, over 5518251.92 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3703, pruned_loss=0.1189, over 5675227.39 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:15:31,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:16:02,976 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:16:14,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3848, 1.6876, 1.3929, 1.4486], device='cuda:1'), covar=tensor([0.0766, 0.0290, 0.0322, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 18:16:17,296 INFO [train.py:968] (1/2) Epoch 19, batch 6300, giga_loss[loss=0.3895, simple_loss=0.4176, pruned_loss=0.1807, over 23673.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3756, pruned_loss=0.1226, over 5669225.30 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3498, pruned_loss=0.09425, over 5515542.28 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.376, pruned_loss=0.1238, over 5671706.53 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:16:19,996 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:16:23,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=828894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:16:28,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 18:16:58,673 INFO [optim.py:369] (1/2) 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 18:17:05,667 INFO [train.py:968] (1/2) Epoch 19, batch 6350, giga_loss[loss=0.3192, simple_loss=0.3834, pruned_loss=0.1275, over 28591.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3805, pruned_loss=0.1267, over 5665322.48 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3498, pruned_loss=0.09427, over 5524866.59 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3815, pruned_loss=0.1284, over 5662094.57 frames. ], batch size: 336, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:17:21,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=828953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:17:50,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=828981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:17:53,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=828984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:17:56,841 INFO [train.py:968] (1/2) Epoch 19, batch 6400, giga_loss[loss=0.2849, simple_loss=0.3592, pruned_loss=0.1053, over 28915.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3825, pruned_loss=0.1295, over 5655297.63 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3498, pruned_loss=0.09426, over 5536279.23 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3842, pruned_loss=0.1318, over 5646026.59 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:18:18,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6209, 1.6172, 1.7837, 1.3794], device='cuda:1'), covar=tensor([0.1429, 0.2283, 0.1238, 0.1500], device='cuda:1'), in_proj_covar=tensor([0.0880, 0.0696, 0.0925, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 18:18:25,965 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:18:51,572 INFO [train.py:968] (1/2) Epoch 19, batch 6450, giga_loss[loss=0.4799, simple_loss=0.4888, pruned_loss=0.2355, over 27837.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3864, pruned_loss=0.1345, over 5638783.35 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3497, pruned_loss=0.09422, over 5543587.00 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3884, pruned_loss=0.1371, over 5626800.26 frames. ], batch size: 412, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:19:05,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6268, 2.4390, 1.7201, 0.8153], device='cuda:1'), covar=tensor([0.6134, 0.3176, 0.4442, 0.6357], device='cuda:1'), in_proj_covar=tensor([0.1717, 0.1619, 0.1584, 0.1399], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 18:19:40,750 INFO [train.py:968] (1/2) Epoch 19, batch 6500, giga_loss[loss=0.3167, simple_loss=0.3752, pruned_loss=0.1291, over 28069.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3886, pruned_loss=0.1372, over 5626867.88 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3495, pruned_loss=0.09412, over 5553301.93 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3919, pruned_loss=0.141, over 5611352.45 frames. ], batch size: 77, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:19:52,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:19:57,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:20:26,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:20:31,515 INFO [optim.py:369] (1/2) 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 18:20:35,665 INFO [train.py:968] (1/2) Epoch 19, batch 6550, giga_loss[loss=0.3065, simple_loss=0.3697, pruned_loss=0.1216, over 28821.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.391, pruned_loss=0.1387, over 5625227.86 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3494, pruned_loss=0.09406, over 5558387.86 frames. ], giga_tot_loss[loss=0.3395, simple_loss=0.3942, pruned_loss=0.1424, over 5609334.63 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:20:46,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1504, 2.1814, 1.9108, 1.8720], device='cuda:1'), covar=tensor([0.1842, 0.2602, 0.2397, 0.2417], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0748, 0.0707, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:20:46,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5396, 1.8270, 1.7891, 1.4145], device='cuda:1'), covar=tensor([0.2494, 0.2061, 0.1340, 0.2085], device='cuda:1'), in_proj_covar=tensor([0.1919, 0.1857, 0.1790, 0.1918], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 18:21:16,480 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 6600, libri_loss[loss=0.3085, simple_loss=0.3769, pruned_loss=0.1201, over 19964.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3899, pruned_loss=0.1383, over 5630144.78 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3492, pruned_loss=0.09408, over 5561724.55 frames. ], giga_tot_loss[loss=0.3399, simple_loss=0.3942, pruned_loss=0.1429, over 5617743.88 frames. ], batch size: 187, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:22:11,503 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 6650, giga_loss[loss=0.4524, simple_loss=0.4613, pruned_loss=0.2217, over 26541.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3887, pruned_loss=0.1385, over 5636213.64 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3488, pruned_loss=0.09384, over 5565445.59 frames. ], giga_tot_loss[loss=0.3393, simple_loss=0.393, pruned_loss=0.1428, over 5624164.97 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:22:27,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:22:40,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0811, 1.0423, 3.4291, 3.0601], device='cuda:1'), covar=tensor([0.1753, 0.2811, 0.0494, 0.0949], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0636, 0.0942, 0.0884], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 18:22:45,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:22:48,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:22:54,642 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:23:07,556 INFO [train.py:968] (1/2) Epoch 19, batch 6700, giga_loss[loss=0.2878, simple_loss=0.3662, pruned_loss=0.1047, over 28892.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3886, pruned_loss=0.1377, over 5638203.02 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3487, pruned_loss=0.09373, over 5573898.71 frames. ], giga_tot_loss[loss=0.3389, simple_loss=0.393, pruned_loss=0.1424, over 5622269.10 frames. ], batch size: 145, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:23:41,345 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:23:43,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:23:50,418 INFO [optim.py:369] (1/2) 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 18:23:55,191 INFO [train.py:968] (1/2) Epoch 19, batch 6750, giga_loss[loss=0.3357, simple_loss=0.3988, pruned_loss=0.1363, over 28921.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3886, pruned_loss=0.1363, over 5651297.32 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09354, over 5582335.13 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3934, pruned_loss=0.1414, over 5632965.88 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:24:08,954 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:24:45,985 INFO [train.py:968] (1/2) Epoch 19, batch 6800, giga_loss[loss=0.3147, simple_loss=0.3842, pruned_loss=0.1226, over 28912.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3893, pruned_loss=0.1367, over 5629073.04 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3489, pruned_loss=0.09376, over 5589711.67 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3939, pruned_loss=0.1417, over 5609681.88 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:24:49,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:24:50,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-09 18:24:51,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:07,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:11,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:12,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:12,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:15,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:23,623 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:32,456 INFO [optim.py:369] (1/2) 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 18:25:39,433 INFO [train.py:968] (1/2) Epoch 19, batch 6850, giga_loss[loss=0.3074, simple_loss=0.3701, pruned_loss=0.1223, over 28214.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.388, pruned_loss=0.1355, over 5620719.98 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09381, over 5581207.11 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3917, pruned_loss=0.1396, over 5612795.79 frames. ], batch size: 368, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:25:39,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4411, 1.7111, 1.4978, 1.6178], device='cuda:1'), covar=tensor([0.0637, 0.0274, 0.0285, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 18:25:41,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:25:47,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:26:02,591 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 18:26:34,075 INFO [train.py:968] (1/2) Epoch 19, batch 6900, giga_loss[loss=0.2831, simple_loss=0.3651, pruned_loss=0.1006, over 28833.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3852, pruned_loss=0.1321, over 5618908.20 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.09362, over 5582170.45 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3888, pruned_loss=0.136, over 5612380.97 frames. ], batch size: 119, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:26:35,457 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829489.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:26:38,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5122, 1.8156, 1.7952, 1.5668], device='cuda:1'), covar=tensor([0.1931, 0.2075, 0.2081, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0749, 0.0707, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:27:19,079 INFO [optim.py:369] (1/2) 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 18:27:23,920 INFO [train.py:968] (1/2) Epoch 19, batch 6950, giga_loss[loss=0.2948, simple_loss=0.3699, pruned_loss=0.1099, over 28900.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3807, pruned_loss=0.1276, over 5633574.49 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09346, over 5583508.52 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3842, pruned_loss=0.1312, over 5627602.36 frames. ], batch size: 174, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:27:37,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-09 18:28:12,577 INFO [train.py:968] (1/2) Epoch 19, batch 7000, giga_loss[loss=0.3201, simple_loss=0.3651, pruned_loss=0.1376, over 23759.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3781, pruned_loss=0.1252, over 5639202.50 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09346, over 5592031.39 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3818, pruned_loss=0.1291, over 5628244.58 frames. ], batch size: 705, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:28:24,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-09 18:28:36,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1707, 1.3736, 1.1627, 0.9474], device='cuda:1'), covar=tensor([0.0947, 0.0437, 0.1033, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0449, 0.0517, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 18:29:01,058 INFO [optim.py:369] (1/2) 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 18:29:04,974 INFO [train.py:968] (1/2) Epoch 19, batch 7050, giga_loss[loss=0.2559, simple_loss=0.3333, pruned_loss=0.08931, over 28960.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3765, pruned_loss=0.1245, over 5646835.37 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.09363, over 5593239.24 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3793, pruned_loss=0.1275, over 5637431.87 frames. ], batch size: 164, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:29:13,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:29:17,243 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:29:42,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2609, 1.4283, 1.3197, 1.5572], device='cuda:1'), covar=tensor([0.0745, 0.0356, 0.0322, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 18:29:56,683 INFO [train.py:968] (1/2) Epoch 19, batch 7100, giga_loss[loss=0.286, simple_loss=0.3583, pruned_loss=0.1069, over 28845.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.376, pruned_loss=0.1239, over 5651965.58 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3489, pruned_loss=0.0937, over 5593643.37 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3785, pruned_loss=0.1266, over 5644740.14 frames. ], batch size: 186, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:30:10,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:30:50,025 INFO [optim.py:369] (1/2) 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 18:30:53,118 INFO [train.py:968] (1/2) Epoch 19, batch 7150, libri_loss[loss=0.298, simple_loss=0.378, pruned_loss=0.109, over 28621.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3749, pruned_loss=0.1223, over 5661497.96 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.0935, over 5603652.65 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3781, pruned_loss=0.1256, over 5648493.30 frames. ], batch size: 106, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:31:44,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:31:45,086 INFO [train.py:968] (1/2) Epoch 19, batch 7200, giga_loss[loss=0.3493, simple_loss=0.4276, pruned_loss=0.1355, over 28884.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3733, pruned_loss=0.1196, over 5675285.44 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3488, pruned_loss=0.09362, over 5609619.46 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.376, pruned_loss=0.1226, over 5660651.02 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:31:50,839 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:31:55,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=829795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:32:19,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4296, 2.1309, 1.6273, 0.7068], device='cuda:1'), covar=tensor([0.4888, 0.2474, 0.3146, 0.5497], device='cuda:1'), in_proj_covar=tensor([0.1712, 0.1613, 0.1581, 0.1395], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 18:32:26,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:32:37,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:32:39,332 INFO [optim.py:369] (1/2) 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 18:32:42,626 INFO [train.py:968] (1/2) Epoch 19, batch 7250, giga_loss[loss=0.3148, simple_loss=0.3885, pruned_loss=0.1206, over 28814.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3736, pruned_loss=0.1181, over 5663080.57 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3489, pruned_loss=0.09386, over 5609882.92 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3765, pruned_loss=0.1212, over 5652601.02 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:33:05,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=829864.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:33:30,251 INFO [train.py:968] (1/2) Epoch 19, batch 7300, giga_loss[loss=0.2878, simple_loss=0.3607, pruned_loss=0.1074, over 28874.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3738, pruned_loss=0.1178, over 5664674.24 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3486, pruned_loss=0.09369, over 5618274.22 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3773, pruned_loss=0.1212, over 5650394.00 frames. ], batch size: 213, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:34:16,893 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=829930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:34:18,874 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:34:23,661 INFO [train.py:968] (1/2) Epoch 19, batch 7350, giga_loss[loss=0.3432, simple_loss=0.4005, pruned_loss=0.1429, over 28441.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3743, pruned_loss=0.1189, over 5677224.47 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3488, pruned_loss=0.09376, over 5618941.83 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3772, pruned_loss=0.1218, over 5666270.96 frames. ], batch size: 71, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:34:41,140 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=829953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:34:48,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=829962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:35:12,110 INFO [train.py:968] (1/2) Epoch 19, batch 7400, libri_loss[loss=0.2669, simple_loss=0.352, pruned_loss=0.09085, over 27684.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.371, pruned_loss=0.1172, over 5679196.62 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3478, pruned_loss=0.09333, over 5626813.85 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.375, pruned_loss=0.1208, over 5665109.38 frames. ], batch size: 115, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:35:31,812 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830007.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 18:35:34,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830010.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:35:38,933 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=830015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:35:44,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:35:56,505 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 7450, giga_loss[loss=0.3169, simple_loss=0.3733, pruned_loss=0.1302, over 28353.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3702, pruned_loss=0.1182, over 5670450.81 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3477, pruned_loss=0.09317, over 5634394.04 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3743, pruned_loss=0.1221, over 5653681.21 frames. ], batch size: 368, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:36:00,138 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830039.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 18:36:02,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:36:32,030 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 7500, giga_loss[loss=0.2761, simple_loss=0.3486, pruned_loss=0.1018, over 29093.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3683, pruned_loss=0.1172, over 5682764.03 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3476, pruned_loss=0.09313, over 5635765.07 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3716, pruned_loss=0.1203, over 5668642.73 frames. ], batch size: 128, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:37:27,608 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=830128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:37:34,053 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 7550, giga_loss[loss=0.2937, simple_loss=0.3694, pruned_loss=0.109, over 28856.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3677, pruned_loss=0.1152, over 5694864.10 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3478, pruned_loss=0.09315, over 5640792.28 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3707, pruned_loss=0.1184, over 5680210.54 frames. ], batch size: 199, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:37:45,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=830147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:01,177 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:03,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:21,032 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:23,062 INFO [train.py:968] (1/2) Epoch 19, batch 7600, giga_loss[loss=0.2747, simple_loss=0.3534, pruned_loss=0.09799, over 29008.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3679, pruned_loss=0.1144, over 5703616.74 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3479, pruned_loss=0.093, over 5648651.59 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3708, pruned_loss=0.1176, over 5686448.93 frames. ], batch size: 136, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:38:23,762 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:31,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:31,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3007, 1.2372, 1.1499, 1.5410], device='cuda:1'), covar=tensor([0.0756, 0.0365, 0.0348, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 18:38:35,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7454, 1.8227, 1.7120, 1.6299], device='cuda:1'), covar=tensor([0.1777, 0.2260, 0.2355, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0750, 0.0706, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:38:43,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:50,360 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:50,896 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:38:53,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:39:06,344 INFO [optim.py:369] (1/2) 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 18:39:09,318 INFO [train.py:968] (1/2) Epoch 19, batch 7650, giga_loss[loss=0.3024, simple_loss=0.3695, pruned_loss=0.1177, over 28754.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3674, pruned_loss=0.1147, over 5690773.94 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.09311, over 5642463.88 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3702, pruned_loss=0.1176, over 5683698.56 frames. ], batch size: 242, lr: 1.70e-03, grad_scale: 8.0
+2023-03-09 18:39:17,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:39:28,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 18:39:57,163 INFO [train.py:968] (1/2) Epoch 19, batch 7700, giga_loss[loss=0.3194, simple_loss=0.3781, pruned_loss=0.1304, over 27564.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3665, pruned_loss=0.1146, over 5691693.94 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3479, pruned_loss=0.09311, over 5644454.36 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3691, pruned_loss=0.1175, over 5685253.58 frames. ], batch size: 472, lr: 1.70e-03, grad_scale: 4.0
+2023-03-09 18:40:29,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9806, 2.0777, 1.8690, 1.7129], device='cuda:1'), covar=tensor([0.1859, 0.2376, 0.2181, 0.2380], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0748, 0.0705, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:40:37,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:40:46,453 INFO [optim.py:369] (1/2) 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 18:40:47,227 INFO [train.py:968] (1/2) Epoch 19, batch 7750, giga_loss[loss=0.395, simple_loss=0.424, pruned_loss=0.183, over 26734.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3667, pruned_loss=0.1156, over 5687005.01 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3479, pruned_loss=0.09314, over 5649365.58 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3693, pruned_loss=0.1185, over 5678512.96 frames. ], batch size: 555, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 18:40:58,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:41:00,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:41:26,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9542, 1.1677, 1.2205, 1.0827], device='cuda:1'), covar=tensor([0.1357, 0.1023, 0.1723, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0749, 0.0707, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:41:29,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:41:33,708 INFO [train.py:968] (1/2) Epoch 19, batch 7800, libri_loss[loss=0.2147, simple_loss=0.2964, pruned_loss=0.06647, over 28582.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3645, pruned_loss=0.1146, over 5691072.80 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.347, pruned_loss=0.09266, over 5656368.29 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3679, pruned_loss=0.1179, over 5679215.34 frames. ], batch size: 63, lr: 1.70e-03, grad_scale: 2.0
+2023-03-09 18:41:35,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:42:18,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5369, 2.1408, 1.5566, 0.7711], device='cuda:1'), covar=tensor([0.5389, 0.2572, 0.3724, 0.5895], device='cuda:1'), in_proj_covar=tensor([0.1708, 0.1613, 0.1578, 0.1392], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 18:42:24,353 INFO [optim.py:369] (1/2) 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 18:42:25,112 INFO [train.py:968] (1/2) Epoch 19, batch 7850, giga_loss[loss=0.2642, simple_loss=0.3403, pruned_loss=0.09409, over 29104.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3639, pruned_loss=0.1145, over 5701639.62 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3473, pruned_loss=0.09285, over 5658136.66 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3666, pruned_loss=0.1173, over 5691243.77 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:42:55,061 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:42:59,334 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:10,309 INFO [train.py:968] (1/2) Epoch 19, batch 7900, giga_loss[loss=0.268, simple_loss=0.3395, pruned_loss=0.09821, over 28761.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3627, pruned_loss=0.1146, over 5703245.80 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3473, pruned_loss=0.09285, over 5664843.10 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3655, pruned_loss=0.1175, over 5690361.55 frames. ], batch size: 92, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:43:27,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:27,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:42,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=830522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:55,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:58,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:43:58,551 INFO [optim.py:369] (1/2) 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 18:43:59,242 INFO [train.py:968] (1/2) Epoch 19, batch 7950, giga_loss[loss=0.3, simple_loss=0.3743, pruned_loss=0.1128, over 29030.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3623, pruned_loss=0.1145, over 5706937.37 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3474, pruned_loss=0.09285, over 5668231.32 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3645, pruned_loss=0.1171, over 5694348.40 frames. ], batch size: 128, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:44:24,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2208, 1.4492, 1.4546, 1.1552], device='cuda:1'), covar=tensor([0.2661, 0.2380, 0.1427, 0.2219], device='cuda:1'), in_proj_covar=tensor([0.1924, 0.1848, 0.1789, 0.1923], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 18:44:26,678 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 8000, giga_loss[loss=0.2924, simple_loss=0.3666, pruned_loss=0.1091, over 28813.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3641, pruned_loss=0.116, over 5696384.60 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3474, pruned_loss=0.09285, over 5669339.23 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3663, pruned_loss=0.1187, over 5686032.09 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:44:54,162 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-09 18:45:40,180 INFO [optim.py:369] (1/2) 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 18:45:40,998 INFO [train.py:968] (1/2) Epoch 19, batch 8050, giga_loss[loss=0.2823, simple_loss=0.3589, pruned_loss=0.1028, over 28934.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3647, pruned_loss=0.1161, over 5691408.62 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3471, pruned_loss=0.0927, over 5670396.54 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3668, pruned_loss=0.1184, over 5682559.29 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:45:49,183 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=830646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:45:50,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:46:05,122 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=830668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:46:14,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=830678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:46:24,819 INFO [train.py:968] (1/2) Epoch 19, batch 8100, giga_loss[loss=0.3139, simple_loss=0.3819, pruned_loss=0.1229, over 28349.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3642, pruned_loss=0.1147, over 5685636.27 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.347, pruned_loss=0.09264, over 5678613.19 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3666, pruned_loss=0.1175, over 5671682.00 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:46:35,528 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:47:15,066 INFO [train.py:968] (1/2) Epoch 19, batch 8150, giga_loss[loss=0.3139, simple_loss=0.3698, pruned_loss=0.129, over 28659.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3654, pruned_loss=0.1159, over 5678237.30 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3468, pruned_loss=0.09261, over 5671417.91 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3678, pruned_loss=0.1185, over 5674001.99 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:47:54,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5603, 2.3538, 1.7416, 0.6833], device='cuda:1'), covar=tensor([0.5153, 0.2986, 0.3773, 0.5733], device='cuda:1'), in_proj_covar=tensor([0.1712, 0.1618, 0.1584, 0.1396], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 18:48:06,139 INFO [train.py:968] (1/2) Epoch 19, batch 8200, giga_loss[loss=0.3793, simple_loss=0.4213, pruned_loss=0.1686, over 28982.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3679, pruned_loss=0.1181, over 5682680.34 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3464, pruned_loss=0.09247, over 5674839.26 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3703, pruned_loss=0.1207, over 5676350.33 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:48:31,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-09 18:48:40,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9649, 1.3570, 1.1279, 0.1511], device='cuda:1'), covar=tensor([0.3647, 0.2887, 0.3662, 0.5828], device='cuda:1'), in_proj_covar=tensor([0.1712, 0.1618, 0.1584, 0.1396], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 18:48:40,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 18:48:53,374 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4318, 4.2117, 1.6227, 1.6364], device='cuda:1'), covar=tensor([0.1005, 0.0383, 0.0902, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0549, 0.0374, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 18:48:58,318 INFO [optim.py:369] (1/2) 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 18:48:58,936 INFO [train.py:968] (1/2) Epoch 19, batch 8250, giga_loss[loss=0.3117, simple_loss=0.3771, pruned_loss=0.1231, over 28817.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.37, pruned_loss=0.1203, over 5685125.56 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.09288, over 5681021.21 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1226, over 5674716.03 frames. ], batch size: 284, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:49:36,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8954, 1.1584, 2.8564, 2.7107], device='cuda:1'), covar=tensor([0.1619, 0.2486, 0.0654, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0744, 0.0639, 0.0943, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-09 18:49:44,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2616, 1.1757, 3.7235, 3.2723], device='cuda:1'), covar=tensor([0.1628, 0.2835, 0.0484, 0.1163], device='cuda:1'), in_proj_covar=tensor([0.0744, 0.0639, 0.0942, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-09 18:49:51,327 INFO [train.py:968] (1/2) Epoch 19, batch 8300, giga_loss[loss=0.2738, simple_loss=0.3426, pruned_loss=0.1024, over 28123.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3704, pruned_loss=0.1214, over 5677953.99 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3471, pruned_loss=0.09276, over 5684790.45 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.124, over 5666272.66 frames. ], batch size: 77, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:50:41,198 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 8350, giga_loss[loss=0.2873, simple_loss=0.3567, pruned_loss=0.109, over 28486.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3724, pruned_loss=0.124, over 5675884.26 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3472, pruned_loss=0.09275, over 5689156.75 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3744, pruned_loss=0.1267, over 5662275.60 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:51:25,327 INFO [train.py:968] (1/2) Epoch 19, batch 8400, giga_loss[loss=0.2863, simple_loss=0.3557, pruned_loss=0.1084, over 28842.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3701, pruned_loss=0.122, over 5676811.74 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3471, pruned_loss=0.09268, over 5693122.30 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3727, pruned_loss=0.1253, over 5661702.39 frames. ], batch size: 119, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 18:51:51,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3629, 1.1496, 4.1966, 3.2937], device='cuda:1'), covar=tensor([0.1651, 0.2860, 0.0471, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0638, 0.0939, 0.0884], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 18:52:07,794 INFO [optim.py:369] (1/2) 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 18:52:07,807 INFO [train.py:968] (1/2) Epoch 19, batch 8450, giga_loss[loss=0.3002, simple_loss=0.3802, pruned_loss=0.1101, over 28973.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1219, over 5679907.91 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3472, pruned_loss=0.09264, over 5696885.37 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3732, pruned_loss=0.1252, over 5664118.57 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:52:53,143 INFO [train.py:968] (1/2) Epoch 19, batch 8500, giga_loss[loss=0.2965, simple_loss=0.376, pruned_loss=0.1085, over 28883.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3692, pruned_loss=0.12, over 5663683.09 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3472, pruned_loss=0.09273, over 5691058.74 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.123, over 5656671.97 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:53:36,964 INFO [optim.py:369] (1/2) 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 18:53:36,977 INFO [train.py:968] (1/2) Epoch 19, batch 8550, giga_loss[loss=0.2714, simple_loss=0.3375, pruned_loss=0.1027, over 28755.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3679, pruned_loss=0.1189, over 5673846.58 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3476, pruned_loss=0.09295, over 5693418.84 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3698, pruned_loss=0.1216, over 5665759.13 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:54:07,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5956, 1.7560, 1.6578, 1.5115], device='cuda:1'), covar=tensor([0.1820, 0.2174, 0.2401, 0.2259], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0742, 0.0700, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 18:54:23,999 INFO [train.py:968] (1/2) Epoch 19, batch 8600, giga_loss[loss=0.2813, simple_loss=0.3506, pruned_loss=0.106, over 28722.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3645, pruned_loss=0.1167, over 5682242.70 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3476, pruned_loss=0.093, over 5699958.22 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3667, pruned_loss=0.1197, over 5668953.25 frames. ], batch size: 284, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:55:13,812 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 8650, giga_loss[loss=0.2907, simple_loss=0.3508, pruned_loss=0.1153, over 28421.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3648, pruned_loss=0.1178, over 5671560.37 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.09314, over 5701945.58 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3667, pruned_loss=0.1205, over 5658839.81 frames. ], batch size: 78, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:55:27,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6556, 2.4353, 1.7090, 0.9965], device='cuda:1'), covar=tensor([0.4711, 0.2293, 0.3119, 0.4080], device='cuda:1'), in_proj_covar=tensor([0.1700, 0.1612, 0.1575, 0.1391], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 18:56:05,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1990, 1.4635, 1.3761, 1.4602], device='cuda:1'), covar=tensor([0.0850, 0.0332, 0.0306, 0.1069], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 18:56:06,311 INFO [train.py:968] (1/2) Epoch 19, batch 8700, giga_loss[loss=0.3215, simple_loss=0.396, pruned_loss=0.1235, over 29020.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3671, pruned_loss=0.1196, over 5656453.15 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3477, pruned_loss=0.09321, over 5695252.44 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3689, pruned_loss=0.122, over 5652230.77 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:56:39,883 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-09 18:56:57,406 INFO [train.py:968] (1/2) Epoch 19, batch 8750, giga_loss[loss=0.2927, simple_loss=0.375, pruned_loss=0.1052, over 29018.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3698, pruned_loss=0.1193, over 5654248.80 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3472, pruned_loss=0.09308, over 5688992.86 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3719, pruned_loss=0.1217, over 5655735.06 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 18:56:58,102 INFO [optim.py:369] (1/2) 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 18:57:09,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.55 vs. limit=5.0
+2023-03-09 18:57:41,272 INFO [train.py:968] (1/2) Epoch 19, batch 8800, giga_loss[loss=0.2799, simple_loss=0.3544, pruned_loss=0.1027, over 28599.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3696, pruned_loss=0.1172, over 5671828.58 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3464, pruned_loss=0.09276, over 5696467.35 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.373, pruned_loss=0.1205, over 5665090.44 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:58:27,666 INFO [train.py:968] (1/2) Epoch 19, batch 8850, giga_loss[loss=0.331, simple_loss=0.3997, pruned_loss=0.1311, over 28976.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3727, pruned_loss=0.1197, over 5674042.13 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3462, pruned_loss=0.09273, over 5703914.94 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3765, pruned_loss=0.1232, over 5661287.28 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:58:28,857 INFO [optim.py:369] (1/2) 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 18:58:54,955 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=831472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 18:59:13,939 INFO [train.py:968] (1/2) Epoch 19, batch 8900, giga_loss[loss=0.3362, simple_loss=0.3959, pruned_loss=0.1383, over 28854.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3751, pruned_loss=0.1221, over 5664991.23 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3463, pruned_loss=0.09277, over 5706682.99 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3784, pruned_loss=0.1253, over 5652197.62 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:59:57,136 INFO [train.py:968] (1/2) Epoch 19, batch 8950, giga_loss[loss=0.2789, simple_loss=0.347, pruned_loss=0.1054, over 28273.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3739, pruned_loss=0.1214, over 5669161.15 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3465, pruned_loss=0.0927, over 5710364.91 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3776, pruned_loss=0.1252, over 5653671.86 frames. ], batch size: 65, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 18:59:57,798 INFO [optim.py:369] (1/2) 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 19:00:27,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2246, 1.2196, 3.4673, 3.0489], device='cuda:1'), covar=tensor([0.1584, 0.2861, 0.0486, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0638, 0.0942, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-09 19:00:45,988 INFO [train.py:968] (1/2) Epoch 19, batch 9000, giga_loss[loss=0.2802, simple_loss=0.3515, pruned_loss=0.1045, over 28843.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.122, over 5658518.49 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3465, pruned_loss=0.09275, over 5714341.05 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3767, pruned_loss=0.1256, over 5641623.45 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:00:45,988 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 19:00:53,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1060, 1.5265, 1.5497, 1.3210], device='cuda:1'), covar=tensor([0.1659, 0.1529, 0.2006, 0.1517], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0746, 0.0703, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 19:00:54,701 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 19:01:38,823 INFO [train.py:968] (1/2) Epoch 19, batch 9050, giga_loss[loss=0.2987, simple_loss=0.3627, pruned_loss=0.1174, over 28880.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3714, pruned_loss=0.1211, over 5651690.07 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3466, pruned_loss=0.09285, over 5704537.10 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.375, pruned_loss=0.1249, over 5645248.34 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:01:41,107 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 9100, giga_loss[loss=0.2615, simple_loss=0.345, pruned_loss=0.08897, over 28864.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3706, pruned_loss=0.1216, over 5652128.65 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3469, pruned_loss=0.09306, over 5703500.85 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3735, pruned_loss=0.1248, over 5647038.39 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:02:59,464 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-09 19:03:20,971 INFO [train.py:968] (1/2) Epoch 19, batch 9150, giga_loss[loss=0.2956, simple_loss=0.366, pruned_loss=0.1126, over 28939.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1226, over 5654526.56 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3467, pruned_loss=0.09294, over 5706738.46 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3742, pruned_loss=0.1256, over 5647085.09 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:03:22,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 19:03:23,314 INFO [optim.py:369] (1/2) 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 19:03:34,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3817, 1.6333, 1.5412, 1.5402], device='cuda:1'), covar=tensor([0.1771, 0.1754, 0.2203, 0.1703], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0748, 0.0705, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 19:03:49,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2586, 1.5061, 1.2377, 0.9710], device='cuda:1'), covar=tensor([0.2707, 0.2710, 0.3083, 0.2296], device='cuda:1'), in_proj_covar=tensor([0.1458, 0.1056, 0.1293, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 19:04:10,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6781, 1.8118, 1.7482, 1.6252], device='cuda:1'), covar=tensor([0.1714, 0.2117, 0.2327, 0.2207], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0748, 0.0705, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 19:04:11,236 INFO [train.py:968] (1/2) Epoch 19, batch 9200, giga_loss[loss=0.2735, simple_loss=0.334, pruned_loss=0.1065, over 29031.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3714, pruned_loss=0.1231, over 5651156.95 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3467, pruned_loss=0.09293, over 5709774.10 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.1259, over 5641704.75 frames. ], batch size: 128, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:05:01,083 INFO [train.py:968] (1/2) Epoch 19, batch 9250, giga_loss[loss=0.281, simple_loss=0.3486, pruned_loss=0.1067, over 28617.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3681, pruned_loss=0.1212, over 5661490.50 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3467, pruned_loss=0.09291, over 5713874.32 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3705, pruned_loss=0.124, over 5649612.43 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:05:02,659 INFO [optim.py:369] (1/2) 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 19:05:08,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=831847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:05:48,101 INFO [train.py:968] (1/2) Epoch 19, batch 9300, giga_loss[loss=0.3083, simple_loss=0.3706, pruned_loss=0.123, over 28974.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.368, pruned_loss=0.1211, over 5656720.61 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3469, pruned_loss=0.09297, over 5718938.58 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3704, pruned_loss=0.1242, over 5641154.15 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:06:37,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6554, 1.7160, 1.3910, 1.2375], device='cuda:1'), covar=tensor([0.0937, 0.0618, 0.1003, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0447, 0.0512, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 19:06:37,941 INFO [train.py:968] (1/2) Epoch 19, batch 9350, giga_loss[loss=0.2981, simple_loss=0.3683, pruned_loss=0.114, over 29014.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3686, pruned_loss=0.1201, over 5664581.54 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3468, pruned_loss=0.09291, over 5719562.29 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3711, pruned_loss=0.1232, over 5650573.95 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:06:39,824 INFO [optim.py:369] (1/2) 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 19:07:25,092 INFO [train.py:968] (1/2) Epoch 19, batch 9400, giga_loss[loss=0.3196, simple_loss=0.375, pruned_loss=0.1321, over 28407.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 5659704.87 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3472, pruned_loss=0.09327, over 5712477.20 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5653616.06 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:07:27,808 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=831990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:07:32,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=831993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:07:38,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2911, 1.7671, 1.2370, 0.6346], device='cuda:1'), covar=tensor([0.5454, 0.2867, 0.2903, 0.5987], device='cuda:1'), in_proj_covar=tensor([0.1719, 0.1622, 0.1586, 0.1404], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 19:07:58,760 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=832022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:08:14,161 INFO [train.py:968] (1/2) Epoch 19, batch 9450, giga_loss[loss=0.3269, simple_loss=0.3885, pruned_loss=0.1327, over 28785.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3704, pruned_loss=0.1216, over 5658246.70 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3468, pruned_loss=0.09313, over 5717287.43 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3729, pruned_loss=0.1245, over 5647715.12 frames. ], batch size: 284, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:08:17,102 INFO [optim.py:369] (1/2) 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 19:09:02,952 INFO [train.py:968] (1/2) Epoch 19, batch 9500, giga_loss[loss=0.2676, simple_loss=0.3532, pruned_loss=0.09098, over 28967.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3715, pruned_loss=0.1195, over 5667637.20 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3469, pruned_loss=0.09284, over 5720367.95 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3744, pruned_loss=0.1231, over 5654272.62 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:09:21,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4048, 1.6519, 1.6009, 1.5045], device='cuda:1'), covar=tensor([0.1695, 0.1907, 0.2152, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0748, 0.0704, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 19:09:22,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8859, 1.3239, 1.3993, 1.1051], device='cuda:1'), covar=tensor([0.1770, 0.1216, 0.1890, 0.1504], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0748, 0.0704, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 19:09:49,659 INFO [train.py:968] (1/2) Epoch 19, batch 9550, libri_loss[loss=0.2662, simple_loss=0.3544, pruned_loss=0.08906, over 29235.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3743, pruned_loss=0.1195, over 5665148.58 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3473, pruned_loss=0.09306, over 5714980.74 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3768, pruned_loss=0.1228, over 5657700.92 frames. ], batch size: 94, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:09:51,198 INFO [optim.py:369] (1/2) 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 19:10:30,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7642, 4.8417, 1.9322, 1.8998], device='cuda:1'), covar=tensor([0.0915, 0.0418, 0.0885, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0550, 0.0374, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 19:10:38,593 INFO [train.py:968] (1/2) Epoch 19, batch 9600, giga_loss[loss=0.3045, simple_loss=0.3726, pruned_loss=0.1182, over 28683.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3777, pruned_loss=0.1212, over 5672663.94 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3473, pruned_loss=0.09292, over 5718894.50 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3803, pruned_loss=0.1244, over 5662422.49 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:11:07,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4417, 3.3187, 1.4671, 1.6220], device='cuda:1'), covar=tensor([0.0950, 0.0422, 0.0911, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0550, 0.0375, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 19:11:28,996 INFO [train.py:968] (1/2) Epoch 19, batch 9650, giga_loss[loss=0.2943, simple_loss=0.3585, pruned_loss=0.115, over 28863.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3796, pruned_loss=0.1233, over 5675062.91 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3472, pruned_loss=0.09285, over 5719375.21 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3823, pruned_loss=0.1265, over 5665805.25 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:11:34,015 INFO [optim.py:369] (1/2) 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 19:12:17,168 INFO [train.py:968] (1/2) Epoch 19, batch 9700, giga_loss[loss=0.3411, simple_loss=0.3927, pruned_loss=0.1447, over 28278.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3816, pruned_loss=0.126, over 5673222.40 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3472, pruned_loss=0.09287, over 5724058.11 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.3845, pruned_loss=0.1292, over 5660750.62 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:13:10,360 INFO [train.py:968] (1/2) Epoch 19, batch 9750, giga_loss[loss=0.3083, simple_loss=0.3787, pruned_loss=0.119, over 28722.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3821, pruned_loss=0.127, over 5665889.12 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3472, pruned_loss=0.09282, over 5725106.82 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3845, pruned_loss=0.1297, over 5654972.04 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:13:12,864 INFO [optim.py:369] (1/2) 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 19:13:58,631 INFO [train.py:968] (1/2) Epoch 19, batch 9800, giga_loss[loss=0.2545, simple_loss=0.3467, pruned_loss=0.0811, over 28962.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3802, pruned_loss=0.1256, over 5671741.89 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3469, pruned_loss=0.09266, over 5726958.07 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3831, pruned_loss=0.1285, over 5660431.30 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:14:26,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 19:14:44,523 INFO [train.py:968] (1/2) Epoch 19, batch 9850, giga_loss[loss=0.3709, simple_loss=0.4226, pruned_loss=0.1596, over 28626.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3799, pruned_loss=0.1234, over 5677229.26 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3464, pruned_loss=0.09238, over 5729721.95 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.383, pruned_loss=0.1264, over 5665107.55 frames. ], batch size: 92, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:14:47,104 INFO [optim.py:369] (1/2) 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 19:15:17,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=832470.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:15:31,469 INFO [train.py:968] (1/2) Epoch 19, batch 9900, giga_loss[loss=0.2876, simple_loss=0.3637, pruned_loss=0.1058, over 28594.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3797, pruned_loss=0.1227, over 5680591.38 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3465, pruned_loss=0.09254, over 5733357.79 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.383, pruned_loss=0.1257, over 5666262.13 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:16:21,377 INFO [train.py:968] (1/2) Epoch 19, batch 9950, giga_loss[loss=0.2932, simple_loss=0.3657, pruned_loss=0.1103, over 28427.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3808, pruned_loss=0.1239, over 5678566.77 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3466, pruned_loss=0.0926, over 5737322.89 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3845, pruned_loss=0.1273, over 5661558.59 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:16:23,848 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 10000, giga_loss[loss=0.3615, simple_loss=0.4065, pruned_loss=0.1582, over 27614.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3789, pruned_loss=0.1233, over 5663700.38 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3463, pruned_loss=0.09254, over 5730661.25 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.383, pruned_loss=0.1269, over 5654448.66 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:17:13,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3011, 1.6009, 1.3095, 1.0191], device='cuda:1'), covar=tensor([0.2405, 0.2433, 0.2752, 0.2254], device='cuda:1'), in_proj_covar=tensor([0.1457, 0.1057, 0.1291, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 19:17:59,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5006, 1.7832, 1.4916, 1.4135], device='cuda:1'), covar=tensor([0.2400, 0.2273, 0.2519, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.1459, 0.1058, 0.1292, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 19:18:00,565 INFO [train.py:968] (1/2) Epoch 19, batch 10050, giga_loss[loss=0.2883, simple_loss=0.3589, pruned_loss=0.1089, over 28872.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3763, pruned_loss=0.1225, over 5663662.22 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.346, pruned_loss=0.09234, over 5735635.61 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3808, pruned_loss=0.1264, over 5649919.77 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:18:04,938 INFO [optim.py:369] (1/2) 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 19:18:47,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 19:18:48,720 INFO [train.py:968] (1/2) Epoch 19, batch 10100, giga_loss[loss=0.3047, simple_loss=0.374, pruned_loss=0.1177, over 28891.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3752, pruned_loss=0.1224, over 5675438.72 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3463, pruned_loss=0.09246, over 5740710.92 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3794, pruned_loss=0.1263, over 5657841.49 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:19:36,394 INFO [train.py:968] (1/2) Epoch 19, batch 10150, giga_loss[loss=0.3542, simple_loss=0.3997, pruned_loss=0.1544, over 27977.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3729, pruned_loss=0.1215, over 5667493.97 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3466, pruned_loss=0.09242, over 5743247.75 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3773, pruned_loss=0.1261, over 5647574.41 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:19:40,020 INFO [optim.py:369] (1/2) 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 19:20:29,531 INFO [train.py:968] (1/2) Epoch 19, batch 10200, giga_loss[loss=0.3316, simple_loss=0.3934, pruned_loss=0.1348, over 28897.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3724, pruned_loss=0.1219, over 5668493.80 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3468, pruned_loss=0.09242, over 5745383.65 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3761, pruned_loss=0.1259, over 5649970.83 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:21:06,508 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 19:21:17,843 INFO [train.py:968] (1/2) Epoch 19, batch 10250, giga_loss[loss=0.2647, simple_loss=0.3465, pruned_loss=0.09149, over 28931.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3709, pruned_loss=0.121, over 5669091.76 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3468, pruned_loss=0.09229, over 5748054.22 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3744, pruned_loss=0.125, over 5650151.16 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:21:22,188 INFO [optim.py:369] (1/2) 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 19:21:27,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=832845.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:22:05,817 INFO [train.py:968] (1/2) Epoch 19, batch 10300, giga_loss[loss=0.2844, simple_loss=0.3631, pruned_loss=0.1029, over 28986.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3674, pruned_loss=0.1171, over 5680389.02 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3465, pruned_loss=0.09202, over 5752519.47 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3711, pruned_loss=0.1213, over 5658752.93 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:22:28,199 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=832913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:22:51,994 INFO [train.py:968] (1/2) Epoch 19, batch 10350, giga_loss[loss=0.2932, simple_loss=0.3674, pruned_loss=0.1095, over 28974.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3642, pruned_loss=0.114, over 5666315.36 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3465, pruned_loss=0.092, over 5750616.18 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3676, pruned_loss=0.118, over 5648845.32 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 19:23:00,041 INFO [optim.py:369] (1/2) 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 19:23:05,525 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=832948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:23:42,898 INFO [train.py:968] (1/2) Epoch 19, batch 10400, giga_loss[loss=0.325, simple_loss=0.383, pruned_loss=0.1335, over 28905.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3635, pruned_loss=0.1132, over 5673519.89 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3461, pruned_loss=0.09176, over 5752764.07 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.367, pruned_loss=0.117, over 5655713.75 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:23:43,450 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=832988.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:23:45,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=832991.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:24:13,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833020.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:24:32,370 INFO [train.py:968] (1/2) Epoch 19, batch 10450, giga_loss[loss=0.2587, simple_loss=0.3322, pruned_loss=0.09261, over 28874.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3622, pruned_loss=0.1132, over 5675688.21 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09167, over 5757204.48 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3656, pruned_loss=0.117, over 5655419.87 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:24:41,403 INFO [optim.py:369] (1/2) 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 19:25:19,764 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 10500, giga_loss[loss=0.284, simple_loss=0.3525, pruned_loss=0.1077, over 28941.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1129, over 5672917.93 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3463, pruned_loss=0.09183, over 5758166.80 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3627, pruned_loss=0.1161, over 5654536.21 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:26:12,418 INFO [train.py:968] (1/2) Epoch 19, batch 10550, giga_loss[loss=0.3341, simple_loss=0.3925, pruned_loss=0.1379, over 27564.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.36, pruned_loss=0.1122, over 5677994.21 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3457, pruned_loss=0.09155, over 5761757.10 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.363, pruned_loss=0.1156, over 5658180.72 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:26:16,942 INFO [optim.py:369] (1/2) 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 19:26:48,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1917, 1.1080, 3.7206, 3.1737], device='cuda:1'), covar=tensor([0.1670, 0.2785, 0.0443, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0636, 0.0939, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 19:26:57,490 INFO [train.py:968] (1/2) Epoch 19, batch 10600, giga_loss[loss=0.2493, simple_loss=0.3303, pruned_loss=0.08421, over 28461.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3619, pruned_loss=0.1128, over 5676427.79 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09144, over 5766017.64 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.365, pruned_loss=0.1162, over 5654388.52 frames. ], batch size: 78, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:27:49,018 INFO [train.py:968] (1/2) Epoch 19, batch 10650, libri_loss[loss=0.2535, simple_loss=0.3364, pruned_loss=0.08531, over 29563.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3633, pruned_loss=0.1142, over 5664110.21 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3458, pruned_loss=0.09152, over 5767828.79 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1172, over 5643576.75 frames. ], batch size: 78, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:27:53,785 INFO [optim.py:369] (1/2) 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 19:28:36,412 INFO [train.py:968] (1/2) Epoch 19, batch 10700, giga_loss[loss=0.3367, simple_loss=0.3705, pruned_loss=0.1514, over 23596.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3626, pruned_loss=0.1137, over 5667082.98 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3458, pruned_loss=0.09138, over 5772231.35 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3651, pruned_loss=0.1169, over 5643137.47 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:28:36,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:28:48,911 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-09 19:29:10,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:29:27,346 INFO [train.py:968] (1/2) Epoch 19, batch 10750, giga_loss[loss=0.2943, simple_loss=0.3642, pruned_loss=0.1122, over 28960.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3645, pruned_loss=0.1152, over 5672958.30 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3464, pruned_loss=0.09157, over 5772694.46 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3664, pruned_loss=0.1182, over 5651094.46 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:29:31,890 INFO [optim.py:369] (1/2) 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 19:30:19,039 INFO [train.py:968] (1/2) Epoch 19, batch 10800, giga_loss[loss=0.295, simple_loss=0.3713, pruned_loss=0.1093, over 28847.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3681, pruned_loss=0.1175, over 5669233.38 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3467, pruned_loss=0.09176, over 5775357.68 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3696, pruned_loss=0.1202, over 5647445.02 frames. ], batch size: 243, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:31:01,077 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:03,180 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 10850, libri_loss[loss=0.2444, simple_loss=0.3253, pruned_loss=0.08177, over 29585.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3683, pruned_loss=0.1174, over 5678711.33 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3468, pruned_loss=0.09185, over 5779305.83 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3702, pruned_loss=0.1203, over 5654286.48 frames. ], batch size: 74, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:31:12,753 INFO [optim.py:369] (1/2) 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 19:31:24,501 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:25,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833457.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:31:31,211 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:31,260 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:33,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:37,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:31:54,706 INFO [train.py:968] (1/2) Epoch 19, batch 10900, giga_loss[loss=0.3344, simple_loss=0.3831, pruned_loss=0.1428, over 27508.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3707, pruned_loss=0.1196, over 5677759.67 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3468, pruned_loss=0.09201, over 5772125.81 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3726, pruned_loss=0.1222, over 5663867.88 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:32:05,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:32:34,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4260, 2.0238, 1.4004, 0.7973], device='cuda:1'), covar=tensor([0.5429, 0.2756, 0.3875, 0.5762], device='cuda:1'), in_proj_covar=tensor([0.1713, 0.1618, 0.1579, 0.1400], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 19:32:44,223 INFO [train.py:968] (1/2) Epoch 19, batch 10950, libri_loss[loss=0.3713, simple_loss=0.4247, pruned_loss=0.159, over 20195.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.371, pruned_loss=0.12, over 5668972.91 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3466, pruned_loss=0.092, over 5764201.94 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3733, pruned_loss=0.1229, over 5663269.02 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:32:50,093 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 11000, giga_loss[loss=0.321, simple_loss=0.3849, pruned_loss=0.1286, over 27920.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3716, pruned_loss=0.1193, over 5664209.78 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3464, pruned_loss=0.09196, over 5767827.73 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3744, pruned_loss=0.1224, over 5653763.82 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:33:39,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6340, 1.3191, 5.0831, 3.5319], device='cuda:1'), covar=tensor([0.1646, 0.2778, 0.0409, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0643, 0.0945, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-09 19:33:46,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 19:33:48,914 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833600.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:33:51,693 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=833632.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:34:29,920 INFO [train.py:968] (1/2) Epoch 19, batch 11050, giga_loss[loss=0.2886, simple_loss=0.3618, pruned_loss=0.1077, over 29038.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3726, pruned_loss=0.1211, over 5660909.95 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3459, pruned_loss=0.09176, over 5769838.12 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3755, pruned_loss=0.124, over 5649763.10 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:34:35,836 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 11100, giga_loss[loss=0.3117, simple_loss=0.3757, pruned_loss=0.1238, over 28231.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3721, pruned_loss=0.1216, over 5635668.84 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3459, pruned_loss=0.09185, over 5761032.16 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3751, pruned_loss=0.1247, over 5632192.42 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:35:24,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3945, 4.2284, 4.0236, 2.0472], device='cuda:1'), covar=tensor([0.0552, 0.0702, 0.0738, 0.2218], device='cuda:1'), in_proj_covar=tensor([0.1194, 0.1110, 0.0949, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 19:36:14,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:36:19,285 INFO [train.py:968] (1/2) Epoch 19, batch 11150, giga_loss[loss=0.2879, simple_loss=0.362, pruned_loss=0.1069, over 28598.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3709, pruned_loss=0.1213, over 5637201.42 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3461, pruned_loss=0.09193, over 5754891.18 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3738, pruned_loss=0.1243, over 5637844.38 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:36:25,791 INFO [optim.py:369] (1/2) 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 19:37:07,997 INFO [train.py:968] (1/2) Epoch 19, batch 11200, libri_loss[loss=0.2884, simple_loss=0.3691, pruned_loss=0.1038, over 29464.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3679, pruned_loss=0.1194, over 5640447.50 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.346, pruned_loss=0.09185, over 5757141.08 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3705, pruned_loss=0.1222, over 5637363.31 frames. ], batch size: 85, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:37:41,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=833824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:37:48,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=833831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:37:53,224 INFO [train.py:968] (1/2) Epoch 19, batch 11250, giga_loss[loss=0.2955, simple_loss=0.359, pruned_loss=0.116, over 28917.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1192, over 5650297.10 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3461, pruned_loss=0.09179, over 5757524.56 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3702, pruned_loss=0.1223, over 5644653.46 frames. ], batch size: 106, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:37:53,449 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:38:44,566 INFO [train.py:968] (1/2) Epoch 19, batch 11300, giga_loss[loss=0.3719, simple_loss=0.4025, pruned_loss=0.1707, over 26476.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1198, over 5651040.45 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3462, pruned_loss=0.0918, over 5759850.12 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3699, pruned_loss=0.1227, over 5642945.18 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:39:35,652 INFO [train.py:968] (1/2) Epoch 19, batch 11350, giga_loss[loss=0.3247, simple_loss=0.362, pruned_loss=0.1437, over 23648.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1198, over 5656739.61 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09164, over 5763480.29 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3699, pruned_loss=0.1229, over 5644713.18 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:39:40,653 INFO [optim.py:369] (1/2) 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 19:40:01,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2951, 1.5899, 1.2936, 0.9900], device='cuda:1'), covar=tensor([0.2069, 0.2000, 0.2185, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1460, 0.1058, 0.1294, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 19:40:11,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:14,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:16,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=833981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:19,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=833984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:22,117 INFO [train.py:968] (1/2) Epoch 19, batch 11400, giga_loss[loss=0.2871, simple_loss=0.3562, pruned_loss=0.109, over 28678.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3701, pruned_loss=0.1221, over 5660761.92 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.346, pruned_loss=0.09168, over 5764987.19 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3727, pruned_loss=0.1254, over 5646903.21 frames. ], batch size: 92, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:40:25,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 19:40:42,592 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:47,741 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:40:48,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8007, 2.7260, 1.7845, 1.0474], device='cuda:1'), covar=tensor([0.7937, 0.3291, 0.3780, 0.6340], device='cuda:1'), in_proj_covar=tensor([0.1713, 0.1616, 0.1578, 0.1396], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 19:41:00,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4325, 4.2590, 4.0760, 2.0522], device='cuda:1'), covar=tensor([0.0575, 0.0706, 0.0737, 0.1938], device='cuda:1'), in_proj_covar=tensor([0.1195, 0.1112, 0.0952, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 19:41:11,404 INFO [train.py:968] (1/2) Epoch 19, batch 11450, giga_loss[loss=0.3238, simple_loss=0.3806, pruned_loss=0.1335, over 28581.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3708, pruned_loss=0.1228, over 5657619.95 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09165, over 5768305.50 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3735, pruned_loss=0.1262, over 5640931.76 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:41:17,565 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 11500, giga_loss[loss=0.2656, simple_loss=0.3457, pruned_loss=0.09276, over 28982.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3705, pruned_loss=0.1231, over 5657412.48 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3456, pruned_loss=0.09143, over 5770836.34 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3734, pruned_loss=0.1266, over 5639882.77 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:42:26,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:42:45,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5174, 1.6845, 1.3083, 1.2318], device='cuda:1'), covar=tensor([0.0960, 0.0633, 0.1070, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0446, 0.0509, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 19:42:52,268 INFO [train.py:968] (1/2) Epoch 19, batch 11550, giga_loss[loss=0.3113, simple_loss=0.3676, pruned_loss=0.1275, over 28860.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3705, pruned_loss=0.123, over 5661930.45 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3452, pruned_loss=0.09119, over 5770903.58 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3738, pruned_loss=0.1267, over 5645607.55 frames. ], batch size: 112, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:42:58,711 INFO [optim.py:369] (1/2) 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 19:43:13,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5853, 1.7609, 1.3186, 1.3518], device='cuda:1'), covar=tensor([0.0947, 0.0580, 0.1033, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0446, 0.0509, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 19:43:40,925 INFO [train.py:968] (1/2) Epoch 19, batch 11600, giga_loss[loss=0.3053, simple_loss=0.3743, pruned_loss=0.1182, over 28866.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3718, pruned_loss=0.124, over 5659108.10 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3448, pruned_loss=0.09102, over 5773621.50 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3753, pruned_loss=0.1277, over 5641660.43 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:43:53,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:43:59,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1775, 2.1998, 1.9723, 1.8975], device='cuda:1'), covar=tensor([0.1890, 0.2565, 0.2355, 0.2398], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0751, 0.0707, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 19:44:26,209 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:44:31,122 INFO [train.py:968] (1/2) Epoch 19, batch 11650, giga_loss[loss=0.2904, simple_loss=0.3613, pruned_loss=0.1097, over 28840.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1233, over 5675296.74 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3448, pruned_loss=0.09105, over 5775475.69 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3749, pruned_loss=0.1266, over 5658458.90 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:44:40,856 INFO [optim.py:369] (1/2) 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 19:44:45,093 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:44:49,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:44:59,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 19:45:22,405 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:45:25,853 INFO [train.py:968] (1/2) Epoch 19, batch 11700, giga_loss[loss=0.2924, simple_loss=0.3585, pruned_loss=0.1131, over 28685.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.374, pruned_loss=0.1256, over 5660755.53 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3447, pruned_loss=0.09099, over 5775775.46 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3769, pruned_loss=0.1288, over 5645907.72 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:46:16,515 INFO [train.py:968] (1/2) Epoch 19, batch 11750, giga_loss[loss=0.3423, simple_loss=0.3885, pruned_loss=0.1481, over 27912.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3749, pruned_loss=0.1261, over 5663825.92 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3447, pruned_loss=0.09092, over 5778913.99 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3778, pruned_loss=0.1295, over 5646621.97 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:46:20,381 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:46:23,279 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:46:23,699 INFO [optim.py:369] (1/2) 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 19:46:32,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:46:51,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:47:04,429 INFO [train.py:968] (1/2) Epoch 19, batch 11800, giga_loss[loss=0.3465, simple_loss=0.3774, pruned_loss=0.1577, over 23740.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3747, pruned_loss=0.1261, over 5658821.76 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3447, pruned_loss=0.09086, over 5780815.16 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3775, pruned_loss=0.1293, over 5641977.40 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:47:52,018 INFO [train.py:968] (1/2) Epoch 19, batch 11850, giga_loss[loss=0.2948, simple_loss=0.3691, pruned_loss=0.1103, over 28858.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3746, pruned_loss=0.1244, over 5652973.34 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3449, pruned_loss=0.09103, over 5770014.31 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3776, pruned_loss=0.1279, over 5645192.02 frames. ], batch size: 119, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:47:59,364 INFO [optim.py:369] (1/2) 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 19:48:12,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 19:48:17,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:48:39,800 INFO [train.py:968] (1/2) Epoch 19, batch 11900, libri_loss[loss=0.2348, simple_loss=0.313, pruned_loss=0.0783, over 29414.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3736, pruned_loss=0.1227, over 5659238.36 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3443, pruned_loss=0.09072, over 5771450.62 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3776, pruned_loss=0.127, over 5647609.40 frames. ], batch size: 67, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:49:11,105 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:49:27,382 INFO [train.py:968] (1/2) Epoch 19, batch 11950, giga_loss[loss=0.309, simple_loss=0.3767, pruned_loss=0.1207, over 28574.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3738, pruned_loss=0.1229, over 5652885.80 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3446, pruned_loss=0.09084, over 5774267.12 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3775, pruned_loss=0.127, over 5638387.29 frames. ], batch size: 85, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:49:33,802 INFO [optim.py:369] (1/2) 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 19:49:38,699 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834549.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 19:49:41,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:50:13,598 INFO [train.py:968] (1/2) Epoch 19, batch 12000, giga_loss[loss=0.3353, simple_loss=0.394, pruned_loss=0.1383, over 28912.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1226, over 5660957.22 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3447, pruned_loss=0.09103, over 5774713.10 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3759, pruned_loss=0.1261, over 5647620.17 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:50:13,598 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 19:50:23,113 INFO [train.py:1012] (1/2) Epoch 19, validation: loss=0.2101, simple_loss=0.3175, pruned_loss=0.05135, over 944034.00 frames. 
+2023-03-09 19:50:23,113 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 19:50:39,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9949, 1.3058, 1.2332, 1.1480], device='cuda:1'), covar=tensor([0.1331, 0.0985, 0.1789, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0750, 0.0707, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 19:50:41,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:50:45,067 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:50:45,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 19:51:09,354 INFO [train.py:968] (1/2) Epoch 19, batch 12050, giga_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 28808.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3741, pruned_loss=0.1235, over 5661897.43 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3449, pruned_loss=0.09122, over 5776104.63 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3771, pruned_loss=0.1269, over 5647413.11 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:51:18,625 INFO [optim.py:369] (1/2) 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 19:52:00,251 INFO [train.py:968] (1/2) Epoch 19, batch 12100, libri_loss[loss=0.2935, simple_loss=0.3722, pruned_loss=0.1074, over 29670.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3745, pruned_loss=0.1238, over 5651969.73 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3452, pruned_loss=0.09134, over 5768490.88 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5644919.45 frames. ], batch size: 91, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:52:11,754 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:52:41,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:52:48,229 INFO [train.py:968] (1/2) Epoch 19, batch 12150, giga_loss[loss=0.2692, simple_loss=0.3492, pruned_loss=0.09464, over 28566.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.373, pruned_loss=0.1231, over 5671167.47 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3454, pruned_loss=0.09151, over 5773618.50 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3757, pruned_loss=0.1264, over 5657435.18 frames. ], batch size: 60, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:52:55,780 INFO [optim.py:369] (1/2) 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 19:53:00,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:04,466 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:09,818 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=834757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:31,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:53:39,389 INFO [train.py:968] (1/2) Epoch 19, batch 12200, giga_loss[loss=0.3396, simple_loss=0.3703, pruned_loss=0.1545, over 23583.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.375, pruned_loss=0.1253, over 5667292.56 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3457, pruned_loss=0.09161, over 5774942.45 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1282, over 5654277.67 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:54:27,964 INFO [train.py:968] (1/2) Epoch 19, batch 12250, giga_loss[loss=0.3136, simple_loss=0.3804, pruned_loss=0.1234, over 28658.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3746, pruned_loss=0.1245, over 5677431.81 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09147, over 5778337.14 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3773, pruned_loss=0.1278, over 5661424.45 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:54:29,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:54:36,601 INFO [optim.py:369] (1/2) 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 19:54:59,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:01,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=834875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:03,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1516, 1.2382, 3.4140, 2.9736], device='cuda:1'), covar=tensor([0.1606, 0.2707, 0.0495, 0.1508], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0638, 0.0943, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-09 19:55:10,104 INFO [train.py:968] (1/2) Epoch 19, batch 12300, giga_loss[loss=0.371, simple_loss=0.3978, pruned_loss=0.1721, over 23493.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3749, pruned_loss=0.1247, over 5658582.84 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09146, over 5765745.40 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3787, pruned_loss=0.1293, over 5651218.73 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:55:18,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:24,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=834904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:47,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834924.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:55:49,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:55:52,154 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 12350, libri_loss[loss=0.2949, simple_loss=0.3688, pruned_loss=0.1105, over 29508.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.374, pruned_loss=0.1234, over 5668232.20 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3456, pruned_loss=0.09158, over 5758409.81 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3775, pruned_loss=0.1275, over 5667151.91 frames. ], batch size: 82, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 19:56:09,203 INFO [optim.py:369] (1/2) 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 19:56:46,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=834982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:56:49,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=834985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:56:49,724 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 12400, giga_loss[loss=0.2765, simple_loss=0.3432, pruned_loss=0.1049, over 28766.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3726, pruned_loss=0.1219, over 5661299.80 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3458, pruned_loss=0.09166, over 5760833.32 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3757, pruned_loss=0.1257, over 5656523.03 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:56:56,154 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:34,314 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:34,680 INFO [train.py:968] (1/2) Epoch 19, batch 12450, giga_loss[loss=0.3198, simple_loss=0.379, pruned_loss=0.1303, over 28546.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3719, pruned_loss=0.1203, over 5673127.96 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3455, pruned_loss=0.09138, over 5762088.15 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3754, pruned_loss=0.1244, over 5665701.37 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:57:38,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:41,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:57:44,005 INFO [optim.py:369] (1/2) 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 19:57:58,783 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:01,239 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835067.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:58:04,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:05,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835070.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:58:07,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:07,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:58:14,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 12500, giga_loss[loss=0.3101, simple_loss=0.3547, pruned_loss=0.1328, over 23689.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3721, pruned_loss=0.1205, over 5673708.10 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3462, pruned_loss=0.09172, over 5760571.94 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3749, pruned_loss=0.1241, over 5667179.85 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:58:36,080 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835099.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 19:58:37,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:02,966 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:05,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:05,728 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 12550, giga_loss[loss=0.2586, simple_loss=0.3367, pruned_loss=0.09027, over 29085.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3721, pruned_loss=0.1214, over 5676930.59 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3463, pruned_loss=0.09177, over 5760641.11 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3748, pruned_loss=0.1248, over 5669860.26 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 19:59:21,537 INFO [optim.py:369] (1/2) 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 19:59:34,215 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 19:59:58,026 INFO [train.py:968] (1/2) Epoch 19, batch 12600, giga_loss[loss=0.2589, simple_loss=0.3346, pruned_loss=0.0916, over 28837.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.37, pruned_loss=0.1206, over 5665486.33 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3473, pruned_loss=0.09242, over 5754470.65 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3722, pruned_loss=0.1238, over 5662538.07 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:00:31,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:00:34,366 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 12650, giga_loss[loss=0.2703, simple_loss=0.3315, pruned_loss=0.1045, over 28633.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 5676521.21 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3471, pruned_loss=0.09236, over 5756111.23 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3676, pruned_loss=0.121, over 5672131.24 frames. ], batch size: 85, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:01:02,443 INFO [optim.py:369] (1/2) 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 20:01:07,440 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:27,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:29,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:40,676 INFO [train.py:968] (1/2) Epoch 19, batch 12700, giga_loss[loss=0.2459, simple_loss=0.3265, pruned_loss=0.08269, over 28889.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3633, pruned_loss=0.117, over 5687172.80 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.347, pruned_loss=0.09222, over 5760604.76 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3655, pruned_loss=0.1202, over 5677336.69 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:01:56,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:01:59,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:02:01,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 20:02:26,644 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 20:02:29,984 INFO [train.py:968] (1/2) Epoch 19, batch 12750, libri_loss[loss=0.2058, simple_loss=0.2907, pruned_loss=0.06042, over 29640.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3617, pruned_loss=0.1164, over 5693676.09 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3462, pruned_loss=0.09184, over 5763754.58 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3645, pruned_loss=0.1197, over 5681713.13 frames. ], batch size: 69, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:02:40,790 INFO [optim.py:369] (1/2) 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 20:02:47,659 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:03:01,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:03:18,582 INFO [train.py:968] (1/2) Epoch 19, batch 12800, giga_loss[loss=0.2799, simple_loss=0.3567, pruned_loss=0.1015, over 28871.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3622, pruned_loss=0.1154, over 5692216.90 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3466, pruned_loss=0.09229, over 5767576.42 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3646, pruned_loss=0.1184, over 5676712.42 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:03:41,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 20:03:43,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:10,893 INFO [train.py:968] (1/2) Epoch 19, batch 12850, giga_loss[loss=0.3042, simple_loss=0.3755, pruned_loss=0.1165, over 28580.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.362, pruned_loss=0.1134, over 5689566.47 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3467, pruned_loss=0.09249, over 5769922.49 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5673825.87 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:04:12,160 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:20,257 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:21,366 INFO [optim.py:369] (1/2) 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 20:04:22,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:32,208 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:04:55,569 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:05:03,598 INFO [train.py:968] (1/2) Epoch 19, batch 12900, giga_loss[loss=0.2429, simple_loss=0.3222, pruned_loss=0.08181, over 28384.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3591, pruned_loss=0.1106, over 5669627.14 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3465, pruned_loss=0.0924, over 5760033.18 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3612, pruned_loss=0.113, over 5664294.40 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:05:28,508 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:05:33,733 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:05:46,814 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-09 20:05:52,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1323, 5.9381, 5.6451, 3.3556], device='cuda:1'), covar=tensor([0.0495, 0.0662, 0.0882, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.1186, 0.1107, 0.0943, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 20:05:54,766 INFO [train.py:968] (1/2) Epoch 19, batch 12950, giga_loss[loss=0.257, simple_loss=0.338, pruned_loss=0.08794, over 28664.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3555, pruned_loss=0.1071, over 5666645.64 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3459, pruned_loss=0.09222, over 5761056.55 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.358, pruned_loss=0.1098, over 5658099.09 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:06:00,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:05,234 INFO [optim.py:369] (1/2) 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 20:06:11,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:16,377 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:44,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:06:45,468 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 13000, giga_loss[loss=0.2514, simple_loss=0.3349, pruned_loss=0.08392, over 27965.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3512, pruned_loss=0.1031, over 5673796.69 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3451, pruned_loss=0.09193, over 5763691.38 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3543, pruned_loss=0.106, over 5661583.09 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:07:14,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:07:15,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:07:36,878 INFO [train.py:968] (1/2) Epoch 19, batch 13050, giga_loss[loss=0.2279, simple_loss=0.3252, pruned_loss=0.06529, over 28987.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3503, pruned_loss=0.1004, over 5671788.62 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3448, pruned_loss=0.09176, over 5764908.94 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.353, pruned_loss=0.1029, over 5660302.59 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:07:48,732 INFO [optim.py:369] (1/2) 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 20:07:58,420 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-09 20:08:30,832 INFO [train.py:968] (1/2) Epoch 19, batch 13100, giga_loss[loss=0.2904, simple_loss=0.3697, pruned_loss=0.1056, over 28617.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3518, pruned_loss=0.1014, over 5665389.13 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3449, pruned_loss=0.0921, over 5768083.77 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.354, pruned_loss=0.1033, over 5651027.85 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:09:14,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:09:23,955 INFO [train.py:968] (1/2) Epoch 19, batch 13150, giga_loss[loss=0.2265, simple_loss=0.3092, pruned_loss=0.0719, over 28801.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3505, pruned_loss=0.1005, over 5667798.80 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3447, pruned_loss=0.09201, over 5768369.79 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3526, pruned_loss=0.1022, over 5654837.38 frames. ], batch size: 99, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:09:35,634 INFO [optim.py:369] (1/2) 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 20:10:07,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4081, 3.9694, 1.6048, 1.7758], device='cuda:1'), covar=tensor([0.0999, 0.0241, 0.0972, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0550, 0.0375, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-09 20:10:15,598 INFO [train.py:968] (1/2) Epoch 19, batch 13200, giga_loss[loss=0.2575, simple_loss=0.3407, pruned_loss=0.08713, over 28924.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.09847, over 5660428.47 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3445, pruned_loss=0.09198, over 5757668.47 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3496, pruned_loss=0.1, over 5656854.19 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:10:26,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3947, 1.6586, 1.6552, 1.2150], device='cuda:1'), covar=tensor([0.1691, 0.2569, 0.1415, 0.1737], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0695, 0.0922, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 20:11:02,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:11:07,654 INFO [train.py:968] (1/2) Epoch 19, batch 13250, giga_loss[loss=0.2575, simple_loss=0.3384, pruned_loss=0.08829, over 28945.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3461, pruned_loss=0.09748, over 5664751.10 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3441, pruned_loss=0.09178, over 5760033.81 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3481, pruned_loss=0.09898, over 5658359.48 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:11:18,651 INFO [optim.py:369] (1/2) 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 20:11:41,761 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:11:45,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:11:57,688 INFO [train.py:968] (1/2) Epoch 19, batch 13300, giga_loss[loss=0.2236, simple_loss=0.3125, pruned_loss=0.06735, over 28861.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3456, pruned_loss=0.09677, over 5668025.15 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3434, pruned_loss=0.09149, over 5762933.21 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3478, pruned_loss=0.09836, over 5658404.31 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:12:11,955 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=835900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:12:16,495 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=835904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:12:47,140 INFO [train.py:968] (1/2) Epoch 19, batch 13350, giga_loss[loss=0.2495, simple_loss=0.3333, pruned_loss=0.08288, over 28902.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3449, pruned_loss=0.09584, over 5669220.88 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3432, pruned_loss=0.09141, over 5762049.59 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.09734, over 5659706.70 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:13:00,561 INFO [optim.py:369] (1/2) 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 20:13:26,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=835975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:13:28,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=835978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:13:38,994 INFO [train.py:968] (1/2) Epoch 19, batch 13400, giga_loss[loss=0.2506, simple_loss=0.3319, pruned_loss=0.08463, over 28956.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3421, pruned_loss=0.09363, over 5673777.65 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3427, pruned_loss=0.09123, over 5765174.31 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3442, pruned_loss=0.09506, over 5661358.44 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:13:40,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=835990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:13:58,714 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:14:32,698 INFO [train.py:968] (1/2) Epoch 19, batch 13450, giga_loss[loss=0.2256, simple_loss=0.3088, pruned_loss=0.0712, over 28859.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3388, pruned_loss=0.09182, over 5667840.75 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3428, pruned_loss=0.0913, over 5767430.47 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3403, pruned_loss=0.09294, over 5653719.12 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:14:45,384 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 13500, giga_loss[loss=0.316, simple_loss=0.3716, pruned_loss=0.1302, over 27625.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3366, pruned_loss=0.09158, over 5647637.28 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3419, pruned_loss=0.09101, over 5762730.35 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3385, pruned_loss=0.09275, over 5637297.07 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:16:13,914 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=836133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:16:17,539 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 13550, giga_loss[loss=0.2889, simple_loss=0.3661, pruned_loss=0.1059, over 28618.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3364, pruned_loss=0.09212, over 5653421.91 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3419, pruned_loss=0.09111, over 5764488.55 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3377, pruned_loss=0.09298, over 5641690.46 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:16:32,105 INFO [optim.py:369] (1/2) 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 20:16:49,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:17:18,207 INFO [train.py:968] (1/2) Epoch 19, batch 13600, giga_loss[loss=0.2777, simple_loss=0.3614, pruned_loss=0.09698, over 28333.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3377, pruned_loss=0.09264, over 5628025.10 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3422, pruned_loss=0.09137, over 5748444.99 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3385, pruned_loss=0.09312, over 5630249.33 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:17:27,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5404, 1.7239, 1.6161, 1.4749], device='cuda:1'), covar=tensor([0.2120, 0.1805, 0.1570, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.1894, 0.1810, 0.1741, 0.1877], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 20:18:11,164 INFO [train.py:968] (1/2) Epoch 19, batch 13650, libri_loss[loss=0.23, simple_loss=0.3005, pruned_loss=0.07974, over 29372.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3398, pruned_loss=0.09257, over 5633105.78 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3417, pruned_loss=0.09113, over 5742885.41 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3407, pruned_loss=0.09322, over 5635841.64 frames. ], batch size: 67, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:18:28,049 INFO [optim.py:369] (1/2) 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 20:18:54,448 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 13700, libri_loss[loss=0.213, simple_loss=0.2931, pruned_loss=0.06649, over 29341.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.34, pruned_loss=0.09254, over 5632301.44 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.341, pruned_loss=0.09074, over 5738876.69 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3414, pruned_loss=0.09347, over 5634543.44 frames. ], batch size: 67, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:19:29,659 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=836303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:20:16,962 INFO [train.py:968] (1/2) Epoch 19, batch 13750, giga_loss[loss=0.2364, simple_loss=0.324, pruned_loss=0.07444, over 28906.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3391, pruned_loss=0.09233, over 5638487.31 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3405, pruned_loss=0.09045, over 5741321.26 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3407, pruned_loss=0.09334, over 5636645.00 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:20:32,967 INFO [optim.py:369] (1/2) 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 20:20:52,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3188, 1.2136, 1.0963, 1.4750], device='cuda:1'), covar=tensor([0.0745, 0.0403, 0.0369, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:1')
+2023-03-09 20:21:09,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1189, 2.2253, 1.8400, 1.9027], device='cuda:1'), covar=tensor([0.0674, 0.0397, 0.0693, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0440, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 20:21:19,094 INFO [train.py:968] (1/2) Epoch 19, batch 13800, giga_loss[loss=0.2666, simple_loss=0.3486, pruned_loss=0.0923, over 28408.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3378, pruned_loss=0.09114, over 5636111.02 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3406, pruned_loss=0.09072, over 5734322.27 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3389, pruned_loss=0.09171, over 5639763.97 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:21:51,190 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=836418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:21:55,569 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=836421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:22:15,779 INFO [train.py:968] (1/2) Epoch 19, batch 13850, giga_loss[loss=0.2318, simple_loss=0.3131, pruned_loss=0.07526, over 27626.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3365, pruned_loss=0.08929, over 5640644.52 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3405, pruned_loss=0.09071, over 5735812.30 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3373, pruned_loss=0.08975, over 5639253.89 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:22:29,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:22:29,839 INFO [optim.py:369] (1/2) 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 20:23:17,160 INFO [train.py:968] (1/2) Epoch 19, batch 13900, giga_loss[loss=0.2403, simple_loss=0.3191, pruned_loss=0.0808, over 29017.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3338, pruned_loss=0.08858, over 5647469.44 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3403, pruned_loss=0.09072, over 5739263.06 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3345, pruned_loss=0.08888, over 5640336.32 frames. ], batch size: 155, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:24:13,825 INFO [train.py:968] (1/2) Epoch 19, batch 13950, giga_loss[loss=0.2625, simple_loss=0.3391, pruned_loss=0.09294, over 28913.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3323, pruned_loss=0.0883, over 5659557.10 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3401, pruned_loss=0.09068, over 5743561.16 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3328, pruned_loss=0.0885, over 5647591.93 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:24:30,299 INFO [optim.py:369] (1/2) 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 20:25:14,249 INFO [train.py:968] (1/2) Epoch 19, batch 14000, libri_loss[loss=0.2062, simple_loss=0.2812, pruned_loss=0.06561, over 29366.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3317, pruned_loss=0.08799, over 5667826.65 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3397, pruned_loss=0.0905, over 5747487.25 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3323, pruned_loss=0.08826, over 5652752.49 frames. ], batch size: 71, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:25:32,022 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=836605.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:26:03,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-09 20:26:07,602 INFO [train.py:968] (1/2) Epoch 19, batch 14050, giga_loss[loss=0.2536, simple_loss=0.3155, pruned_loss=0.09587, over 24489.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3328, pruned_loss=0.08816, over 5668424.11 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3387, pruned_loss=0.09022, over 5744124.22 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3339, pruned_loss=0.08854, over 5656036.66 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:26:20,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6666, 2.1745, 1.7850, 0.9122], device='cuda:1'), covar=tensor([0.4455, 0.2859, 0.3129, 0.5300], device='cuda:1'), in_proj_covar=tensor([0.1703, 0.1607, 0.1574, 0.1400], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 20:26:21,848 INFO [optim.py:369] (1/2) 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 20:26:30,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3939, 1.8304, 1.0070, 1.3438], device='cuda:1'), covar=tensor([0.1178, 0.0620, 0.1489, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0381, 0.0440, 0.0507, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 20:26:56,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2487, 1.2156, 3.4608, 3.0399], device='cuda:1'), covar=tensor([0.1558, 0.2841, 0.0453, 0.1600], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0637, 0.0933, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 20:26:57,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=836678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:27:11,228 INFO [train.py:968] (1/2) Epoch 19, batch 14100, giga_loss[loss=0.2535, simple_loss=0.3403, pruned_loss=0.08337, over 28980.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3355, pruned_loss=0.0886, over 5670369.72 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3385, pruned_loss=0.09014, over 5733296.58 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3365, pruned_loss=0.08895, over 5668214.59 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:27:34,479 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 14150, giga_loss[loss=0.2671, simple_loss=0.3432, pruned_loss=0.09554, over 29098.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3314, pruned_loss=0.08616, over 5670444.19 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3383, pruned_loss=0.09009, over 5734200.10 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3323, pruned_loss=0.08646, over 5667768.46 frames. ], batch size: 285, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:28:43,222 INFO [optim.py:369] (1/2) 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 20:29:28,080 INFO [train.py:968] (1/2) Epoch 19, batch 14200, giga_loss[loss=0.251, simple_loss=0.3374, pruned_loss=0.0823, over 28914.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.08754, over 5672213.85 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3384, pruned_loss=0.09021, over 5737060.59 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3337, pruned_loss=0.08763, over 5665495.82 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:30:13,853 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=836821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:30:19,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=836824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:30:23,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7203, 1.9218, 1.8860, 1.4896], device='cuda:1'), covar=tensor([0.2503, 0.1938, 0.1672, 0.2303], device='cuda:1'), in_proj_covar=tensor([0.1886, 0.1805, 0.1736, 0.1880], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 20:30:32,828 INFO [train.py:968] (1/2) Epoch 19, batch 14250, giga_loss[loss=0.2532, simple_loss=0.3537, pruned_loss=0.07634, over 28743.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3362, pruned_loss=0.08822, over 5660341.76 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3382, pruned_loss=0.09012, over 5738764.88 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3367, pruned_loss=0.0883, over 5651415.29 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:30:51,715 INFO [optim.py:369] (1/2) 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 20:30:54,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=836853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:31:03,490 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 20:31:26,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2712, 1.3504, 1.3032, 1.2786], device='cuda:1'), covar=tensor([0.1828, 0.1885, 0.1379, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.1882, 0.1798, 0.1732, 0.1876], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 20:31:32,827 INFO [train.py:968] (1/2) Epoch 19, batch 14300, giga_loss[loss=0.2575, simple_loss=0.3465, pruned_loss=0.08429, over 28338.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3393, pruned_loss=0.08774, over 5662002.15 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3383, pruned_loss=0.09013, over 5740832.99 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3396, pruned_loss=0.08773, over 5650513.88 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:32:33,884 INFO [train.py:968] (1/2) Epoch 19, batch 14350, giga_loss[loss=0.2446, simple_loss=0.339, pruned_loss=0.0751, over 28654.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3392, pruned_loss=0.08695, over 5650844.91 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3385, pruned_loss=0.09029, over 5741798.27 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3393, pruned_loss=0.08672, over 5638239.51 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:32:50,032 INFO [optim.py:369] (1/2) 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 20:33:10,159 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4253, 1.9052, 1.3756, 0.6933], device='cuda:1'), covar=tensor([0.5178, 0.2726, 0.3693, 0.6004], device='cuda:1'), in_proj_covar=tensor([0.1702, 0.1603, 0.1572, 0.1397], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 20:33:23,837 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 14400, giga_loss[loss=0.2618, simple_loss=0.3461, pruned_loss=0.08876, over 28692.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3389, pruned_loss=0.08602, over 5666101.60 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3377, pruned_loss=0.08995, over 5745185.27 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3397, pruned_loss=0.08603, over 5651161.52 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:34:35,446 INFO [train.py:968] (1/2) Epoch 19, batch 14450, giga_loss[loss=0.3134, simple_loss=0.3704, pruned_loss=0.1282, over 29087.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3417, pruned_loss=0.08908, over 5656290.85 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3379, pruned_loss=0.09015, over 5729793.04 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08887, over 5656971.94 frames. ], batch size: 200, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:34:54,889 INFO [optim.py:369] (1/2) 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 20:35:17,920 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:35:26,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:35:35,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 20:35:37,134 INFO [train.py:968] (1/2) Epoch 19, batch 14500, giga_loss[loss=0.2688, simple_loss=0.3453, pruned_loss=0.09617, over 28960.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3393, pruned_loss=0.08913, over 5665239.18 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3369, pruned_loss=0.08976, over 5734201.52 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3407, pruned_loss=0.08928, over 5658998.87 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:36:27,922 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837123.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:36:35,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837126.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 20:36:47,107 INFO [train.py:968] (1/2) Epoch 19, batch 14550, giga_loss[loss=0.2712, simple_loss=0.3382, pruned_loss=0.1021, over 28982.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3397, pruned_loss=0.09003, over 5670163.92 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3362, pruned_loss=0.08942, over 5739697.24 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3416, pruned_loss=0.09049, over 5657874.13 frames. ], batch size: 186, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:36:53,044 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:37:08,380 INFO [optim.py:369] (1/2) 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 20:37:14,918 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837155.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 20:37:49,202 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:38:10,755 INFO [train.py:968] (1/2) Epoch 19, batch 14600, giga_loss[loss=0.2313, simple_loss=0.3183, pruned_loss=0.07215, over 28560.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3346, pruned_loss=0.08695, over 5673709.58 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.08911, over 5740087.82 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3365, pruned_loss=0.08757, over 5661897.69 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:38:59,798 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:39:02,204 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 14650, giga_loss[loss=0.2634, simple_loss=0.3425, pruned_loss=0.09216, over 28905.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3326, pruned_loss=0.08572, over 5660246.34 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3361, pruned_loss=0.08933, over 5732944.28 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3338, pruned_loss=0.08595, over 5656219.73 frames. ], batch size: 112, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:39:34,102 INFO [optim.py:369] (1/2) 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 20:39:39,271 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:40:29,697 INFO [train.py:968] (1/2) Epoch 19, batch 14700, giga_loss[loss=0.2694, simple_loss=0.3441, pruned_loss=0.0973, over 29112.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3304, pruned_loss=0.08489, over 5670526.81 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.336, pruned_loss=0.08927, over 5734690.50 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3314, pruned_loss=0.08508, over 5665211.36 frames. ], batch size: 200, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:41:05,180 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:41:32,913 INFO [train.py:968] (1/2) Epoch 19, batch 14750, giga_loss[loss=0.2706, simple_loss=0.356, pruned_loss=0.09263, over 28887.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3355, pruned_loss=0.08729, over 5678814.40 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3361, pruned_loss=0.08928, over 5733346.80 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3362, pruned_loss=0.08738, over 5675225.59 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:41:51,185 INFO [optim.py:369] (1/2) 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 20:42:14,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-09 20:42:33,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6075, 1.7769, 1.5008, 1.7431], device='cuda:1'), covar=tensor([0.2888, 0.2765, 0.3096, 0.2621], device='cuda:1'), in_proj_covar=tensor([0.1457, 0.1051, 0.1293, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 20:42:35,169 INFO [train.py:968] (1/2) Epoch 19, batch 14800, giga_loss[loss=0.2331, simple_loss=0.3106, pruned_loss=0.0778, over 28651.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3354, pruned_loss=0.08797, over 5677593.65 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.08916, over 5737521.43 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3363, pruned_loss=0.08813, over 5669583.22 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:43:34,368 INFO [train.py:968] (1/2) Epoch 19, batch 14850, giga_loss[loss=0.3524, simple_loss=0.4059, pruned_loss=0.1495, over 28544.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3371, pruned_loss=0.09039, over 5675599.34 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3359, pruned_loss=0.08928, over 5737227.73 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3376, pruned_loss=0.09038, over 5667609.16 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:43:38,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 20:43:47,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:43:54,239 INFO [optim.py:369] (1/2) 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 20:44:16,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5625, 4.0338, 1.6226, 1.6446], device='cuda:1'), covar=tensor([0.0912, 0.0303, 0.0939, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0543, 0.0374, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-09 20:44:35,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4819, 1.7386, 1.3929, 1.6792], device='cuda:1'), covar=tensor([0.2647, 0.2559, 0.2907, 0.2183], device='cuda:1'), in_proj_covar=tensor([0.1457, 0.1050, 0.1295, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 20:44:35,447 INFO [train.py:968] (1/2) Epoch 19, batch 14900, giga_loss[loss=0.2735, simple_loss=0.3537, pruned_loss=0.09664, over 28366.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3371, pruned_loss=0.09089, over 5675223.55 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3355, pruned_loss=0.08916, over 5738553.13 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3379, pruned_loss=0.09103, over 5666135.48 frames. ], batch size: 369, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:45:14,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:45:40,079 INFO [train.py:968] (1/2) Epoch 19, batch 14950, giga_loss[loss=0.271, simple_loss=0.3562, pruned_loss=0.0929, over 28495.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3374, pruned_loss=0.09027, over 5680162.24 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3348, pruned_loss=0.08883, over 5743781.91 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3387, pruned_loss=0.09077, over 5665808.47 frames. ], batch size: 369, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:45:49,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:45:55,907 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 15000, giga_loss[loss=0.3089, simple_loss=0.3841, pruned_loss=0.1169, over 28936.00 frames. ], tot_loss[loss=0.261, simple_loss=0.34, pruned_loss=0.091, over 5685737.29 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3345, pruned_loss=0.08879, over 5750541.84 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3416, pruned_loss=0.09149, over 5665991.80 frames. ], batch size: 145, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:46:45,332 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 20:46:50,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2622, 1.8119, 1.4019, 0.4460], device='cuda:1'), covar=tensor([0.4693, 0.3351, 0.4957, 0.6173], device='cuda:1'), in_proj_covar=tensor([0.1704, 0.1603, 0.1577, 0.1404], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 20:46:54,832 INFO [train.py:1012] (1/2) Epoch 19, validation: loss=0.198, simple_loss=0.2981, pruned_loss=0.04895, over 944034.00 frames. 
+2023-03-09 20:46:54,833 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 20:46:57,670 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:47:00,745 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:47:54,187 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:47:56,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8937, 1.0990, 1.0413, 0.8572], device='cuda:1'), covar=tensor([0.1788, 0.1942, 0.1247, 0.1690], device='cuda:1'), in_proj_covar=tensor([0.1883, 0.1794, 0.1727, 0.1868], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 20:48:11,166 INFO [train.py:968] (1/2) Epoch 19, batch 15050, giga_loss[loss=0.2895, simple_loss=0.3609, pruned_loss=0.109, over 28951.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3391, pruned_loss=0.09055, over 5684924.05 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3342, pruned_loss=0.08878, over 5753342.62 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3407, pruned_loss=0.091, over 5665512.79 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:48:32,613 INFO [optim.py:369] (1/2) 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 20:48:47,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:48:51,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:24,229 INFO [train.py:968] (1/2) Epoch 19, batch 15100, giga_loss[loss=0.2666, simple_loss=0.3373, pruned_loss=0.09793, over 28499.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3344, pruned_loss=0.08875, over 5698644.82 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3338, pruned_loss=0.0886, over 5756397.39 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3361, pruned_loss=0.08929, over 5679422.41 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:49:29,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:31,252 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:31,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=837692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:49:34,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:50:07,476 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=837715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:50:16,788 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:50:35,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0362, 1.3626, 1.0950, 0.1942], device='cuda:1'), covar=tensor([0.3766, 0.2972, 0.5075, 0.6349], device='cuda:1'), in_proj_covar=tensor([0.1715, 0.1611, 0.1586, 0.1412], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 20:50:35,721 INFO [train.py:968] (1/2) Epoch 19, batch 15150, giga_loss[loss=0.2425, simple_loss=0.316, pruned_loss=0.08455, over 28837.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3285, pruned_loss=0.086, over 5691083.29 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3337, pruned_loss=0.08858, over 5757375.40 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3298, pruned_loss=0.08643, over 5674995.43 frames. ], batch size: 112, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:50:50,969 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:51:33,497 INFO [train.py:968] (1/2) Epoch 19, batch 15200, giga_loss[loss=0.249, simple_loss=0.3294, pruned_loss=0.08431, over 29119.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3294, pruned_loss=0.08705, over 5680855.81 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3332, pruned_loss=0.08833, over 5751424.27 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3309, pruned_loss=0.08759, over 5671149.00 frames. ], batch size: 113, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 20:51:48,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-09 20:52:28,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=837835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:52:30,527 INFO [train.py:968] (1/2) Epoch 19, batch 15250, giga_loss[loss=0.2522, simple_loss=0.3313, pruned_loss=0.08654, over 28899.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3306, pruned_loss=0.08784, over 5679961.27 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3336, pruned_loss=0.0885, over 5754015.82 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3313, pruned_loss=0.08809, over 5668186.08 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:52:31,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=837838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:52:51,365 INFO [optim.py:369] (1/2) 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 20:53:07,992 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=837867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:53:29,815 INFO [train.py:968] (1/2) Epoch 19, batch 15300, giga_loss[loss=0.2446, simple_loss=0.326, pruned_loss=0.0816, over 28946.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3284, pruned_loss=0.08644, over 5657114.80 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3336, pruned_loss=0.08864, over 5742548.13 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3289, pruned_loss=0.08645, over 5656262.06 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:54:31,477 INFO [train.py:968] (1/2) Epoch 19, batch 15350, giga_loss[loss=0.2572, simple_loss=0.329, pruned_loss=0.09266, over 27668.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.327, pruned_loss=0.08466, over 5664130.09 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3335, pruned_loss=0.08863, over 5746912.58 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3273, pruned_loss=0.0846, over 5657625.12 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:54:48,128 INFO [optim.py:369] (1/2) 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 20:55:01,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8451, 2.2065, 2.1743, 1.6206], device='cuda:1'), covar=tensor([0.1878, 0.2346, 0.1491, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.0878, 0.0688, 0.0922, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 20:55:39,208 INFO [train.py:968] (1/2) Epoch 19, batch 15400, giga_loss[loss=0.2657, simple_loss=0.3375, pruned_loss=0.09701, over 28524.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.326, pruned_loss=0.08447, over 5662077.38 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3337, pruned_loss=0.08887, over 5742452.50 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3258, pruned_loss=0.08412, over 5658418.84 frames. ], batch size: 336, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:56:01,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:56:22,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-09 20:56:44,257 INFO [train.py:968] (1/2) Epoch 19, batch 15450, giga_loss[loss=0.2415, simple_loss=0.3322, pruned_loss=0.0754, over 29116.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3261, pruned_loss=0.08378, over 5679109.50 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3336, pruned_loss=0.08877, over 5745932.66 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3259, pruned_loss=0.08346, over 5671628.84 frames. ], batch size: 176, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:56:52,751 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:57:08,576 INFO [optim.py:369] (1/2) 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 20:57:22,060 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 15500, giga_loss[loss=0.2358, simple_loss=0.3179, pruned_loss=0.0769, over 28876.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3266, pruned_loss=0.08395, over 5685075.71 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3336, pruned_loss=0.08882, over 5744881.49 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3264, pruned_loss=0.08358, over 5679582.27 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:57:56,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:58:37,459 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 15550, giga_loss[loss=0.2393, simple_loss=0.3151, pruned_loss=0.08177, over 28436.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3283, pruned_loss=0.08587, over 5674344.29 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.333, pruned_loss=0.08882, over 5730584.52 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3284, pruned_loss=0.08545, over 5680790.79 frames. ], batch size: 370, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 20:59:08,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:59:12,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 20:59:15,840 INFO [optim.py:369] (1/2) 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 20:59:41,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4851, 1.9207, 1.4390, 0.8405], device='cuda:1'), covar=tensor([0.4921, 0.2688, 0.3463, 0.5447], device='cuda:1'), in_proj_covar=tensor([0.1713, 0.1616, 0.1590, 0.1414], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 20:59:50,832 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 15600, giga_loss[loss=0.312, simple_loss=0.3823, pruned_loss=0.1209, over 28946.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.327, pruned_loss=0.08542, over 5663509.91 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3329, pruned_loss=0.08882, over 5727753.76 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3271, pruned_loss=0.08506, over 5670324.86 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:00:51,421 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:00:54,702 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 15650, libri_loss[loss=0.214, simple_loss=0.2891, pruned_loss=0.06943, over 29665.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3271, pruned_loss=0.08382, over 5658997.41 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3325, pruned_loss=0.08852, over 5731019.79 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3275, pruned_loss=0.08366, over 5659389.51 frames. ], batch size: 73, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:01:15,058 INFO [optim.py:369] (1/2) 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 21:01:20,281 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4577, 1.6901, 1.3514, 1.3215], device='cuda:1'), covar=tensor([0.0979, 0.0506, 0.1031, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0441, 0.0511, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:01:25,733 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:01:27,838 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:01:30,843 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 15700, giga_loss[loss=0.2551, simple_loss=0.3413, pruned_loss=0.08443, over 28947.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3298, pruned_loss=0.08474, over 5661405.78 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3321, pruned_loss=0.08831, over 5735764.56 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3303, pruned_loss=0.08467, over 5655327.27 frames. ], batch size: 285, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:02:10,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:02:57,063 INFO [train.py:968] (1/2) Epoch 19, batch 15750, giga_loss[loss=0.2546, simple_loss=0.3345, pruned_loss=0.08736, over 28950.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3315, pruned_loss=0.08522, over 5665255.57 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3324, pruned_loss=0.08851, over 5737981.13 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3315, pruned_loss=0.08492, over 5657282.05 frames. ], batch size: 213, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:03:15,102 INFO [optim.py:369] (1/2) 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 21:03:46,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7458, 1.9790, 1.3161, 1.5665], device='cuda:1'), covar=tensor([0.0950, 0.0586, 0.0997, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0442, 0.0512, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:03:55,500 INFO [train.py:968] (1/2) Epoch 19, batch 15800, giga_loss[loss=0.2867, simple_loss=0.3512, pruned_loss=0.1111, over 26804.00 frames. ], tot_loss[loss=0.252, simple_loss=0.332, pruned_loss=0.086, over 5656862.74 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3328, pruned_loss=0.08879, over 5740667.08 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3317, pruned_loss=0.08542, over 5645861.36 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:04:29,159 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8947, 2.1981, 1.6035, 1.8452], device='cuda:1'), covar=tensor([0.0984, 0.0632, 0.0958, 0.1122], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0442, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:04:32,137 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:04:56,006 INFO [train.py:968] (1/2) Epoch 19, batch 15850, giga_loss[loss=0.2116, simple_loss=0.2979, pruned_loss=0.06269, over 28981.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3292, pruned_loss=0.08447, over 5659599.48 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3324, pruned_loss=0.08861, over 5743835.35 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3293, pruned_loss=0.08412, over 5646478.63 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:05:01,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:05:12,650 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-09 21:05:19,506 INFO [optim.py:369] (1/2) 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 21:05:40,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2424, 1.4900, 1.3303, 1.5277], device='cuda:1'), covar=tensor([0.0816, 0.0324, 0.0345, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0104], device='cuda:1')
+2023-03-09 21:06:01,695 INFO [train.py:968] (1/2) Epoch 19, batch 15900, giga_loss[loss=0.2627, simple_loss=0.3404, pruned_loss=0.09247, over 28063.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3277, pruned_loss=0.08365, over 5661525.21 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3316, pruned_loss=0.08818, over 5746349.13 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3285, pruned_loss=0.0837, over 5647709.85 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:06:06,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-09 21:06:59,771 INFO [train.py:968] (1/2) Epoch 19, batch 15950, giga_loss[loss=0.2293, simple_loss=0.3071, pruned_loss=0.07574, over 28923.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3262, pruned_loss=0.0834, over 5671424.38 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3312, pruned_loss=0.08794, over 5747155.60 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3271, pruned_loss=0.08352, over 5656994.55 frames. ], batch size: 175, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:07:05,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:21,019 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:29,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:54,691 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=838584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:56,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 21:07:59,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=838587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:07:59,453 INFO [train.py:968] (1/2) Epoch 19, batch 16000, giga_loss[loss=0.2386, simple_loss=0.3277, pruned_loss=0.07479, over 28631.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3278, pruned_loss=0.08398, over 5677201.07 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3308, pruned_loss=0.08774, over 5748421.90 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3287, pruned_loss=0.08416, over 5662340.73 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:08:03,244 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 21:08:07,237 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:08:21,791 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:08:34,632 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=838616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:08:43,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-09 21:08:57,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4852, 1.9960, 1.6768, 1.6790], device='cuda:1'), covar=tensor([0.2236, 0.2342, 0.2303, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.0446, 0.0721, 0.0686, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 21:09:03,080 INFO [train.py:968] (1/2) Epoch 19, batch 16050, giga_loss[loss=0.2627, simple_loss=0.3367, pruned_loss=0.09432, over 28646.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.33, pruned_loss=0.08534, over 5672495.37 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3309, pruned_loss=0.08779, over 5746485.25 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3307, pruned_loss=0.08537, over 5660892.02 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:09:26,486 INFO [optim.py:369] (1/2) 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 21:09:56,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8349, 2.0584, 1.4845, 1.6996], device='cuda:1'), covar=tensor([0.0833, 0.0501, 0.0944, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0443, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:10:10,364 INFO [train.py:968] (1/2) Epoch 19, batch 16100, giga_loss[loss=0.242, simple_loss=0.3278, pruned_loss=0.0781, over 29041.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.331, pruned_loss=0.08646, over 5666511.55 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3307, pruned_loss=0.0877, over 5748038.12 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3317, pruned_loss=0.08653, over 5654582.12 frames. ], batch size: 128, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:10:20,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5007, 1.7495, 1.3831, 1.5834], device='cuda:1'), covar=tensor([0.2575, 0.2553, 0.2945, 0.2232], device='cuda:1'), in_proj_covar=tensor([0.1450, 0.1049, 0.1292, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 21:10:26,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2793, 1.5215, 1.5055, 1.2321], device='cuda:1'), covar=tensor([0.2281, 0.2078, 0.1378, 0.1875], device='cuda:1'), in_proj_covar=tensor([0.1890, 0.1798, 0.1731, 0.1876], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:10:40,683 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 21:11:04,594 INFO [train.py:968] (1/2) Epoch 19, batch 16150, giga_loss[loss=0.339, simple_loss=0.4029, pruned_loss=0.1376, over 28589.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3352, pruned_loss=0.08921, over 5669391.66 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3303, pruned_loss=0.08749, over 5754042.11 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3362, pruned_loss=0.08949, over 5651213.01 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:11:27,921 INFO [optim.py:369] (1/2) 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 21:11:30,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4944, 1.8015, 1.4983, 1.7669], device='cuda:1'), covar=tensor([0.0764, 0.0298, 0.0339, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0104], device='cuda:1')
+2023-03-09 21:11:54,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6145, 1.9414, 1.8670, 1.5439], device='cuda:1'), covar=tensor([0.2603, 0.1997, 0.2210, 0.2394], device='cuda:1'), in_proj_covar=tensor([0.1883, 0.1790, 0.1725, 0.1869], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:12:02,408 INFO [train.py:968] (1/2) Epoch 19, batch 16200, giga_loss[loss=0.2567, simple_loss=0.3433, pruned_loss=0.0851, over 28546.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3362, pruned_loss=0.08888, over 5668692.29 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3298, pruned_loss=0.08724, over 5757365.89 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3376, pruned_loss=0.08939, over 5649096.44 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:12:05,415 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:12:08,060 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-09 21:12:22,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:13:11,346 INFO [train.py:968] (1/2) Epoch 19, batch 16250, giga_loss[loss=0.2199, simple_loss=0.3052, pruned_loss=0.06731, over 28324.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3372, pruned_loss=0.08945, over 5657386.34 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3299, pruned_loss=0.08725, over 5758107.06 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3383, pruned_loss=0.08985, over 5640928.84 frames. ], batch size: 60, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:13:20,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4304, 1.7684, 1.5525, 1.6456], device='cuda:1'), covar=tensor([0.1718, 0.2075, 0.1997, 0.1807], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0722, 0.0685, 0.0660], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 21:13:37,150 INFO [optim.py:369] (1/2) 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 21:13:53,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5902, 1.7693, 1.2491, 1.3029], device='cuda:1'), covar=tensor([0.0935, 0.0588, 0.1014, 0.1209], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0442, 0.0511, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:14:18,989 INFO [train.py:968] (1/2) Epoch 19, batch 16300, giga_loss[loss=0.2694, simple_loss=0.3385, pruned_loss=0.1002, over 28792.00 frames. ], tot_loss[loss=0.257, simple_loss=0.336, pruned_loss=0.08898, over 5668994.49 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3298, pruned_loss=0.08711, over 5761688.87 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3371, pruned_loss=0.08948, over 5650462.14 frames. ], batch size: 262, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:14:32,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7198, 1.7689, 1.4564, 2.0096], device='cuda:1'), covar=tensor([0.2634, 0.2806, 0.3171, 0.2459], device='cuda:1'), in_proj_covar=tensor([0.1455, 0.1051, 0.1295, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 21:14:54,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=838916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:15:27,773 INFO [train.py:968] (1/2) Epoch 19, batch 16350, giga_loss[loss=0.3064, simple_loss=0.3707, pruned_loss=0.121, over 28613.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3337, pruned_loss=0.08743, over 5671017.73 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3299, pruned_loss=0.08714, over 5761539.36 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3345, pruned_loss=0.08782, over 5655811.42 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 2.0
+2023-03-09 21:15:50,969 INFO [optim.py:369] (1/2) 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 21:16:16,030 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 16400, giga_loss[loss=0.2424, simple_loss=0.3071, pruned_loss=0.08883, over 24558.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3332, pruned_loss=0.08758, over 5668007.32 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3298, pruned_loss=0.087, over 5757713.93 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3341, pruned_loss=0.08803, over 5655282.14 frames. ], batch size: 705, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:16:29,933 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=838991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:17:29,508 INFO [train.py:968] (1/2) Epoch 19, batch 16450, giga_loss[loss=0.2329, simple_loss=0.3146, pruned_loss=0.07556, over 29021.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.332, pruned_loss=0.08825, over 5661931.73 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3296, pruned_loss=0.08686, over 5759591.64 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.333, pruned_loss=0.08875, over 5649034.43 frames. ], batch size: 285, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:17:52,060 INFO [optim.py:369] (1/2) 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 21:17:54,851 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:17:58,947 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 16500, giga_loss[loss=0.2618, simple_loss=0.3389, pruned_loss=0.0923, over 29040.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3295, pruned_loss=0.0864, over 5665847.61 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3295, pruned_loss=0.08682, over 5761637.87 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3303, pruned_loss=0.08683, over 5651619.00 frames. ], batch size: 128, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:18:34,467 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:18:38,207 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6933, 0.9585, 2.8903, 2.7638], device='cuda:1'), covar=tensor([0.1840, 0.2596, 0.0600, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0736, 0.0634, 0.0928, 0.0864], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:18:41,166 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-09 21:18:52,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6154, 1.9988, 1.6629, 1.6735], device='cuda:1'), covar=tensor([0.1809, 0.2207, 0.2157, 0.2171], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0721, 0.0686, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 21:19:15,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:19:19,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:19:36,547 INFO [train.py:968] (1/2) Epoch 19, batch 16550, giga_loss[loss=0.218, simple_loss=0.3083, pruned_loss=0.0638, over 28462.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.329, pruned_loss=0.08506, over 5673940.88 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3294, pruned_loss=0.08677, over 5762424.68 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3298, pruned_loss=0.08544, over 5661742.04 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:19:54,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:20:04,668 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839161.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:20:08,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:20:24,393 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:20:34,256 INFO [train.py:968] (1/2) Epoch 19, batch 16600, giga_loss[loss=0.2181, simple_loss=0.32, pruned_loss=0.05806, over 28943.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3289, pruned_loss=0.08335, over 5678832.00 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.329, pruned_loss=0.08657, over 5764983.51 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.33, pruned_loss=0.08378, over 5664662.69 frames. ], batch size: 136, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:21:35,518 INFO [train.py:968] (1/2) Epoch 19, batch 16650, giga_loss[loss=0.2627, simple_loss=0.3425, pruned_loss=0.09141, over 27839.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3318, pruned_loss=0.08339, over 5677539.24 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3291, pruned_loss=0.08666, over 5753873.84 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3325, pruned_loss=0.08364, over 5674590.21 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:21:57,793 INFO [optim.py:369] (1/2) 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 21:22:28,730 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5497, 1.6903, 1.4248, 1.5718], device='cuda:1'), covar=tensor([0.2640, 0.2565, 0.2851, 0.2345], device='cuda:1'), in_proj_covar=tensor([0.1462, 0.1054, 0.1301, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 21:22:37,009 INFO [train.py:968] (1/2) Epoch 19, batch 16700, giga_loss[loss=0.2574, simple_loss=0.3335, pruned_loss=0.09062, over 27991.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3317, pruned_loss=0.08356, over 5665469.06 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3288, pruned_loss=0.08663, over 5755669.74 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3326, pruned_loss=0.08376, over 5660899.64 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:23:01,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3893, 1.2648, 1.2920, 1.6017], device='cuda:1'), covar=tensor([0.0717, 0.0419, 0.0334, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0105], device='cuda:1')
+2023-03-09 21:23:03,288 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:06,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:20,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:25,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:23:45,726 INFO [train.py:968] (1/2) Epoch 19, batch 16750, giga_loss[loss=0.25, simple_loss=0.3323, pruned_loss=0.08381, over 28149.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3313, pruned_loss=0.08358, over 5648153.91 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3287, pruned_loss=0.08655, over 5739815.92 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3322, pruned_loss=0.08375, over 5656268.83 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:23:47,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:24:06,770 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:24:10,330 INFO [optim.py:369] (1/2) 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 21:24:25,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:24:41,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3102, 1.4060, 1.3049, 1.5323], device='cuda:1'), covar=tensor([0.0812, 0.0341, 0.0345, 0.0895], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0105], device='cuda:1')
+2023-03-09 21:24:55,561 INFO [train.py:968] (1/2) Epoch 19, batch 16800, giga_loss[loss=0.2421, simple_loss=0.326, pruned_loss=0.0791, over 28580.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3314, pruned_loss=0.08404, over 5648043.23 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.329, pruned_loss=0.08686, over 5744236.91 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3319, pruned_loss=0.08383, over 5648236.22 frames. ], batch size: 307, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:25:04,978 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3558, 1.5653, 1.4584, 1.3703], device='cuda:1'), covar=tensor([0.2322, 0.1882, 0.1826, 0.2064], device='cuda:1'), in_proj_covar=tensor([0.1887, 0.1787, 0.1717, 0.1863], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:26:05,624 INFO [train.py:968] (1/2) Epoch 19, batch 16850, giga_loss[loss=0.2494, simple_loss=0.3385, pruned_loss=0.08014, over 28926.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3314, pruned_loss=0.08334, over 5652824.85 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3286, pruned_loss=0.08686, over 5740026.92 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3322, pruned_loss=0.08306, over 5654515.63 frames. ], batch size: 227, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:26:32,145 INFO [optim.py:369] (1/2) 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 21:27:13,620 INFO [train.py:968] (1/2) Epoch 19, batch 16900, giga_loss[loss=0.2784, simple_loss=0.354, pruned_loss=0.1014, over 28157.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3331, pruned_loss=0.08458, over 5649430.26 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3289, pruned_loss=0.08697, over 5739399.90 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3336, pruned_loss=0.08418, over 5648823.23 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:27:41,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5591, 2.0143, 1.8999, 1.5592], device='cuda:1'), covar=tensor([0.3088, 0.2039, 0.2135, 0.2378], device='cuda:1'), in_proj_covar=tensor([0.1883, 0.1782, 0.1713, 0.1858], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:27:44,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:27:47,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=839512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:28:17,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8785, 4.7004, 4.4846, 2.0457], device='cuda:1'), covar=tensor([0.0455, 0.0572, 0.0654, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1153, 0.1063, 0.0906, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 21:28:21,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839536.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 21:28:23,621 INFO [train.py:968] (1/2) Epoch 19, batch 16950, giga_loss[loss=0.2578, simple_loss=0.3342, pruned_loss=0.09071, over 26928.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3367, pruned_loss=0.08606, over 5657528.43 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3291, pruned_loss=0.087, over 5744360.01 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3371, pruned_loss=0.08566, over 5649833.61 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:28:27,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:28:50,765 INFO [optim.py:369] (1/2) 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 21:29:34,279 INFO [train.py:968] (1/2) Epoch 19, batch 17000, giga_loss[loss=0.2273, simple_loss=0.314, pruned_loss=0.07028, over 28924.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3358, pruned_loss=0.0852, over 5671916.45 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.329, pruned_loss=0.08684, over 5744519.81 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3363, pruned_loss=0.08503, over 5665011.57 frames. ], batch size: 174, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:29:36,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839589.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 21:30:24,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0360, 1.2869, 5.7412, 3.6650], device='cuda:1'), covar=tensor([0.1571, 0.2943, 0.0365, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0732, 0.0634, 0.0922, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:30:27,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 21:30:40,960 INFO [train.py:968] (1/2) Epoch 19, batch 17050, libri_loss[loss=0.2626, simple_loss=0.3475, pruned_loss=0.08889, over 29768.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3343, pruned_loss=0.0851, over 5674630.33 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3286, pruned_loss=0.08657, over 5748979.35 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3354, pruned_loss=0.08517, over 5662162.23 frames. ], batch size: 87, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:31:05,869 INFO [optim.py:369] (1/2) 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 21:31:37,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=839679.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:31:45,548 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 17100, giga_loss[loss=0.2363, simple_loss=0.3174, pruned_loss=0.07755, over 28913.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3329, pruned_loss=0.08397, over 5677907.58 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3283, pruned_loss=0.08637, over 5743478.67 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.334, pruned_loss=0.08417, over 5671171.76 frames. ], batch size: 106, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:32:27,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=839711.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:32:59,956 INFO [train.py:968] (1/2) Epoch 19, batch 17150, giga_loss[loss=0.2318, simple_loss=0.3209, pruned_loss=0.07135, over 28928.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3317, pruned_loss=0.08307, over 5670619.33 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3284, pruned_loss=0.08635, over 5743967.80 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3327, pruned_loss=0.08318, over 5663265.39 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:33:21,571 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:33:24,820 INFO [optim.py:369] (1/2) 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 21:33:55,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:34:03,434 INFO [train.py:968] (1/2) Epoch 19, batch 17200, giga_loss[loss=0.2851, simple_loss=0.3626, pruned_loss=0.1037, over 28451.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3337, pruned_loss=0.08464, over 5674527.61 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3284, pruned_loss=0.08633, over 5745342.90 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3345, pruned_loss=0.08471, over 5666578.23 frames. ], batch size: 369, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:35:04,655 INFO [train.py:968] (1/2) Epoch 19, batch 17250, giga_loss[loss=0.2341, simple_loss=0.3233, pruned_loss=0.07251, over 29007.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3365, pruned_loss=0.08624, over 5675826.03 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08648, over 5748769.75 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3372, pruned_loss=0.08614, over 5664822.77 frames. ], batch size: 164, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:35:29,049 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 17300, giga_loss[loss=0.2736, simple_loss=0.3326, pruned_loss=0.1073, over 26799.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3355, pruned_loss=0.08668, over 5673913.78 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3282, pruned_loss=0.08635, over 5750143.98 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3364, pruned_loss=0.08672, over 5663266.97 frames. ], batch size: 555, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:36:40,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=839920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:36:41,282 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 17350, giga_loss[loss=0.2554, simple_loss=0.3318, pruned_loss=0.08945, over 27606.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3329, pruned_loss=0.08635, over 5668897.35 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3281, pruned_loss=0.08624, over 5754003.84 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3338, pruned_loss=0.08648, over 5655606.37 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:37:25,409 INFO [optim.py:369] (1/2) 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 21:37:33,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=839964.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 21:37:57,030 INFO [train.py:968] (1/2) Epoch 19, batch 17400, giga_loss[loss=0.2632, simple_loss=0.3396, pruned_loss=0.0934, over 28618.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3324, pruned_loss=0.08684, over 5656617.58 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3278, pruned_loss=0.08617, over 5746670.33 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3335, pruned_loss=0.08704, over 5650666.11 frames. ], batch size: 242, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:38:47,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9385, 3.7722, 3.5288, 1.6686], device='cuda:1'), covar=tensor([0.0656, 0.0774, 0.0794, 0.2409], device='cuda:1'), in_proj_covar=tensor([0.1159, 0.1066, 0.0910, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 21:38:56,956 INFO [train.py:968] (1/2) Epoch 19, batch 17450, giga_loss[loss=0.2707, simple_loss=0.3537, pruned_loss=0.09386, over 28904.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3396, pruned_loss=0.09093, over 5652679.72 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.08615, over 5736647.82 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3406, pruned_loss=0.09114, over 5654360.65 frames. ], batch size: 199, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:39:14,908 INFO [optim.py:369] (1/2) 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 21:39:18,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5842, 1.8318, 1.2516, 1.3008], device='cuda:1'), covar=tensor([0.0978, 0.0567, 0.1142, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0441, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:39:25,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5604, 1.7504, 1.5158, 1.3544], device='cuda:1'), covar=tensor([0.2448, 0.2295, 0.2170, 0.2469], device='cuda:1'), in_proj_covar=tensor([0.1883, 0.1782, 0.1713, 0.1865], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:39:36,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3589, 1.4515, 1.2087, 1.5235], device='cuda:1'), covar=tensor([0.0815, 0.0345, 0.0348, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0116, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0067, 0.0061, 0.0104], device='cuda:1')
+2023-03-09 21:39:41,972 INFO [train.py:968] (1/2) Epoch 19, batch 17500, libri_loss[loss=0.2989, simple_loss=0.367, pruned_loss=0.1154, over 19337.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.348, pruned_loss=0.09565, over 5645532.81 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3278, pruned_loss=0.08633, over 5720404.90 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3492, pruned_loss=0.0958, over 5660076.18 frames. ], batch size: 187, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:39:58,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840107.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:40:01,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-09 21:40:01,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 21:40:17,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:40:23,442 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:40:26,578 INFO [train.py:968] (1/2) Epoch 19, batch 17550, giga_loss[loss=0.3019, simple_loss=0.3637, pruned_loss=0.1201, over 27676.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3516, pruned_loss=0.09829, over 5653875.72 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3282, pruned_loss=0.08668, over 5724109.98 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3525, pruned_loss=0.09828, over 5660836.39 frames. ], batch size: 472, lr: 1.69e-03, grad_scale: 4.0
+2023-03-09 21:40:27,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 21:40:34,167 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:40:40,098 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:41:09,418 INFO [train.py:968] (1/2) Epoch 19, batch 17600, giga_loss[loss=0.3012, simple_loss=0.3616, pruned_loss=0.1204, over 28008.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3474, pruned_loss=0.09668, over 5658797.45 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3289, pruned_loss=0.08692, over 5722977.11 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3483, pruned_loss=0.09685, over 5663277.77 frames. ], batch size: 412, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:41:14,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3644, 1.6748, 1.3581, 1.5792], device='cuda:1'), covar=tensor([0.0780, 0.0322, 0.0334, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:1')
+2023-03-09 21:41:54,126 INFO [train.py:968] (1/2) Epoch 19, batch 17650, giga_loss[loss=0.2287, simple_loss=0.302, pruned_loss=0.07772, over 28825.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3396, pruned_loss=0.09315, over 5674808.74 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3287, pruned_loss=0.08675, over 5726901.67 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3408, pruned_loss=0.09358, over 5673691.85 frames. ], batch size: 119, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:42:09,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-09 21:42:15,655 INFO [optim.py:369] (1/2) 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 21:42:27,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:29,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:36,327 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-09 21:42:37,693 INFO [train.py:968] (1/2) Epoch 19, batch 17700, giga_loss[loss=0.2401, simple_loss=0.3111, pruned_loss=0.08456, over 28317.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3333, pruned_loss=0.09016, over 5685680.63 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3293, pruned_loss=0.08684, over 5730602.64 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3338, pruned_loss=0.09055, over 5679958.94 frames. ], batch size: 368, lr: 1.69e-03, grad_scale: 8.0
+2023-03-09 21:42:43,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:44,369 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:46,385 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:49,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:42:52,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:43:04,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5096, 1.8319, 1.5194, 1.3541], device='cuda:1'), covar=tensor([0.2111, 0.2009, 0.2285, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.1450, 0.1050, 0.1291, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 21:43:16,942 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 17750, giga_loss[loss=0.2446, simple_loss=0.3155, pruned_loss=0.08689, over 28590.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3265, pruned_loss=0.08739, over 5690429.61 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3292, pruned_loss=0.08664, over 5735899.23 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3272, pruned_loss=0.08797, over 5679604.49 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:43:24,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:43:42,132 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 17800, giga_loss[loss=0.2183, simple_loss=0.2998, pruned_loss=0.06837, over 28785.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3195, pruned_loss=0.08449, over 5688690.51 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3294, pruned_loss=0.08668, over 5735710.23 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3197, pruned_loss=0.08489, over 5680011.91 frames. ], batch size: 66, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:44:29,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8149, 2.6725, 1.7424, 1.0477], device='cuda:1'), covar=tensor([0.7962, 0.3732, 0.3896, 0.6742], device='cuda:1'), in_proj_covar=tensor([0.1698, 0.1612, 0.1578, 0.1399], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 21:44:49,282 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-09 21:44:49,392 INFO [train.py:968] (1/2) Epoch 19, batch 17850, giga_loss[loss=0.2102, simple_loss=0.2823, pruned_loss=0.06905, over 28389.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3148, pruned_loss=0.08223, over 5691175.78 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3293, pruned_loss=0.08642, over 5735811.45 frames. ], giga_tot_loss[loss=0.2401, simple_loss=0.3148, pruned_loss=0.08271, over 5683415.40 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:44:49,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:44:50,335 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:44:51,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:44:51,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6150, 3.2271, 2.6556, 2.2243], device='cuda:1'), covar=tensor([0.2264, 0.1296, 0.1791, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.1904, 0.1799, 0.1733, 0.1888], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:44:52,247 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:45:12,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:45:12,867 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:45:25,506 INFO [train.py:968] (1/2) Epoch 19, batch 17900, libri_loss[loss=0.2314, simple_loss=0.3154, pruned_loss=0.07374, over 29537.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3122, pruned_loss=0.08078, over 5706009.03 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3286, pruned_loss=0.08583, over 5743415.41 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3122, pruned_loss=0.08148, over 5691071.32 frames. ], batch size: 80, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:45:45,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:45:58,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:46:10,482 INFO [train.py:968] (1/2) Epoch 19, batch 17950, giga_loss[loss=0.2195, simple_loss=0.2794, pruned_loss=0.07977, over 23698.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3089, pruned_loss=0.07926, over 5693392.49 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3284, pruned_loss=0.08565, over 5738723.04 frames. ], giga_tot_loss[loss=0.2341, simple_loss=0.3085, pruned_loss=0.07981, over 5685135.35 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:46:29,107 INFO [optim.py:369] (1/2) 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 21:46:37,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4365, 3.2654, 1.6104, 1.5289], device='cuda:1'), covar=tensor([0.0980, 0.0338, 0.0871, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0538, 0.0373, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 21:46:50,355 INFO [train.py:968] (1/2) Epoch 19, batch 18000, giga_loss[loss=0.2378, simple_loss=0.2987, pruned_loss=0.08847, over 28556.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3068, pruned_loss=0.07847, over 5686327.61 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3287, pruned_loss=0.08584, over 5732499.09 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.3055, pruned_loss=0.0785, over 5684558.44 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:46:50,355 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 21:46:58,872 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 21:47:44,708 INFO [train.py:968] (1/2) Epoch 19, batch 18050, giga_loss[loss=0.2199, simple_loss=0.277, pruned_loss=0.08143, over 24155.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3043, pruned_loss=0.07736, over 5694209.71 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3286, pruned_loss=0.08569, over 5736064.44 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.303, pruned_loss=0.07737, over 5688946.24 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:47:56,509 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:47:58,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:01,231 INFO [optim.py:369] (1/2) 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 21:48:05,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:08,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=840668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:24,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:26,398 INFO [train.py:968] (1/2) Epoch 19, batch 18100, giga_loss[loss=0.2076, simple_loss=0.284, pruned_loss=0.06565, over 28746.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3017, pruned_loss=0.07647, over 5682659.65 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.329, pruned_loss=0.0859, over 5728568.29 frames. ], giga_tot_loss[loss=0.2262, simple_loss=0.3, pruned_loss=0.07619, over 5684166.00 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:48:36,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:48:45,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2206, 1.2178, 3.4959, 3.0447], device='cuda:1'), covar=tensor([0.1644, 0.2875, 0.0490, 0.1433], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0634, 0.0930, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:48:52,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=840715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:49:10,022 INFO [train.py:968] (1/2) Epoch 19, batch 18150, giga_loss[loss=0.269, simple_loss=0.3338, pruned_loss=0.1021, over 28834.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3, pruned_loss=0.07575, over 5684849.42 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3293, pruned_loss=0.08593, over 5725757.26 frames. ], giga_tot_loss[loss=0.2241, simple_loss=0.2976, pruned_loss=0.07525, over 5687510.58 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:49:10,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2715, 1.3360, 3.4568, 3.0762], device='cuda:1'), covar=tensor([0.1578, 0.2756, 0.0497, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0635, 0.0930, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:49:19,059 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=840748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:49:30,347 INFO [optim.py:369] (1/2) 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 21:49:57,282 INFO [train.py:968] (1/2) Epoch 19, batch 18200, giga_loss[loss=0.2245, simple_loss=0.2943, pruned_loss=0.0774, over 28773.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.298, pruned_loss=0.07458, over 5696840.26 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3298, pruned_loss=0.08604, over 5731046.46 frames. ], giga_tot_loss[loss=0.2212, simple_loss=0.2948, pruned_loss=0.07377, over 5693201.62 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:50:41,601 INFO [train.py:968] (1/2) Epoch 19, batch 18250, giga_loss[loss=0.3065, simple_loss=0.3569, pruned_loss=0.1281, over 26666.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.2966, pruned_loss=0.0743, over 5698344.37 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3301, pruned_loss=0.08623, over 5735232.70 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.293, pruned_loss=0.07323, over 5691116.34 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:51:02,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=840858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:51:03,894 INFO [optim.py:369] (1/2) 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 21:51:06,154 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 18300, giga_loss[loss=0.3087, simple_loss=0.3725, pruned_loss=0.1225, over 27645.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3066, pruned_loss=0.07992, over 5700925.76 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3302, pruned_loss=0.08629, over 5737018.46 frames. ], giga_tot_loss[loss=0.2306, simple_loss=0.3033, pruned_loss=0.0789, over 5693027.11 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:51:36,729 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=840890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:52:14,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 21:52:20,247 INFO [train.py:968] (1/2) Epoch 19, batch 18350, giga_loss[loss=0.3495, simple_loss=0.3858, pruned_loss=0.1566, over 23878.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3206, pruned_loss=0.08749, over 5696691.35 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3303, pruned_loss=0.08638, over 5739994.67 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3175, pruned_loss=0.08654, over 5687208.55 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:52:38,702 INFO [optim.py:369] (1/2) 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 21:52:59,665 INFO [train.py:968] (1/2) Epoch 19, batch 18400, giga_loss[loss=0.2975, simple_loss=0.3671, pruned_loss=0.1139, over 28793.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3307, pruned_loss=0.09235, over 5700389.57 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3307, pruned_loss=0.08661, over 5735696.50 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3277, pruned_loss=0.09149, over 5694899.20 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:53:40,920 INFO [train.py:968] (1/2) Epoch 19, batch 18450, giga_loss[loss=0.3087, simple_loss=0.3641, pruned_loss=0.1267, over 23412.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3365, pruned_loss=0.09424, over 5696322.75 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.331, pruned_loss=0.08655, over 5740702.32 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.334, pruned_loss=0.09395, over 5685354.52 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:53:49,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-09 21:54:02,216 INFO [optim.py:369] (1/2) 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 21:54:22,047 INFO [train.py:968] (1/2) Epoch 19, batch 18500, giga_loss[loss=0.259, simple_loss=0.3454, pruned_loss=0.08627, over 28914.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3398, pruned_loss=0.09462, over 5701646.46 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3311, pruned_loss=0.08666, over 5739832.81 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3378, pruned_loss=0.09454, over 5692446.60 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:54:22,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:54:55,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:55:07,073 INFO [train.py:968] (1/2) Epoch 19, batch 18550, giga_loss[loss=0.2413, simple_loss=0.3242, pruned_loss=0.0792, over 28566.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3414, pruned_loss=0.09466, over 5690334.54 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.331, pruned_loss=0.08667, over 5735152.45 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3402, pruned_loss=0.09477, over 5686234.12 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:55:28,174 INFO [optim.py:369] (1/2) 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 21:55:29,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9877, 1.1555, 1.1161, 0.9213], device='cuda:1'), covar=tensor([0.1832, 0.2359, 0.1430, 0.1916], device='cuda:1'), in_proj_covar=tensor([0.1913, 0.1814, 0.1747, 0.1905], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:55:49,785 INFO [train.py:968] (1/2) Epoch 19, batch 18600, giga_loss[loss=0.2584, simple_loss=0.3386, pruned_loss=0.08914, over 28505.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3436, pruned_loss=0.09608, over 5688514.66 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3313, pruned_loss=0.08663, over 5736521.27 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3428, pruned_loss=0.09651, over 5682016.11 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:56:20,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8345, 1.0829, 2.8633, 2.7139], device='cuda:1'), covar=tensor([0.1723, 0.2689, 0.0569, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0729, 0.0628, 0.0921, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:56:34,241 INFO [train.py:968] (1/2) Epoch 19, batch 18650, giga_loss[loss=0.2907, simple_loss=0.3597, pruned_loss=0.1108, over 28954.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3467, pruned_loss=0.09866, over 5691181.34 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3317, pruned_loss=0.08696, over 5740343.83 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3459, pruned_loss=0.09894, over 5681415.41 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:56:55,570 INFO [optim.py:369] (1/2) 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 21:56:58,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:57:00,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-09 21:57:00,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:57:10,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3115, 2.2686, 2.2593, 1.9989], device='cuda:1'), covar=tensor([0.1687, 0.2317, 0.1951, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0735, 0.0700, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 21:57:17,061 INFO [train.py:968] (1/2) Epoch 19, batch 18700, giga_loss[loss=0.2919, simple_loss=0.371, pruned_loss=0.1064, over 28636.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3488, pruned_loss=0.09932, over 5695089.16 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3327, pruned_loss=0.08728, over 5735979.40 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.348, pruned_loss=0.09979, over 5689295.96 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:57:26,314 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 18750, giga_loss[loss=0.295, simple_loss=0.3741, pruned_loss=0.108, over 29052.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3516, pruned_loss=0.1002, over 5696203.07 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3333, pruned_loss=0.08746, over 5733582.33 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.351, pruned_loss=0.1009, over 5691764.23 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 21:58:01,505 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-09 21:58:19,770 INFO [optim.py:369] (1/2) 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 21:58:38,672 INFO [train.py:968] (1/2) Epoch 19, batch 18800, giga_loss[loss=0.2722, simple_loss=0.3531, pruned_loss=0.09565, over 28720.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3532, pruned_loss=0.1002, over 5705114.91 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3337, pruned_loss=0.08754, over 5736337.31 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.353, pruned_loss=0.1011, over 5698001.66 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:58:54,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:59:01,224 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:59:12,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4543, 2.9207, 2.6120, 2.0426], device='cuda:1'), covar=tensor([0.2260, 0.1577, 0.1782, 0.1978], device='cuda:1'), in_proj_covar=tensor([0.1905, 0.1807, 0.1740, 0.1896], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 21:59:20,005 INFO [train.py:968] (1/2) Epoch 19, batch 18850, giga_loss[loss=0.2898, simple_loss=0.3709, pruned_loss=0.1043, over 28872.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3536, pruned_loss=0.09974, over 5707896.05 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3334, pruned_loss=0.08735, over 5739438.50 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3544, pruned_loss=0.101, over 5698604.26 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 21:59:37,520 INFO [optim.py:369] (1/2) 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 21:59:38,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 21:59:51,276 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1661, 1.1369, 3.8008, 3.1657], device='cuda:1'), covar=tensor([0.1778, 0.2961, 0.0444, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0730, 0.0629, 0.0922, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 21:59:57,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4517, 1.6838, 1.3549, 1.6976], device='cuda:1'), covar=tensor([0.2726, 0.2922, 0.3205, 0.2378], device='cuda:1'), in_proj_covar=tensor([0.1452, 0.1057, 0.1292, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 21:59:59,345 INFO [train.py:968] (1/2) Epoch 19, batch 18900, giga_loss[loss=0.2664, simple_loss=0.3453, pruned_loss=0.09372, over 28826.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3537, pruned_loss=0.09897, over 5698840.81 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3335, pruned_loss=0.08731, over 5736359.75 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3549, pruned_loss=0.1005, over 5694026.37 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:00:38,361 INFO [train.py:968] (1/2) Epoch 19, batch 18950, giga_loss[loss=0.3044, simple_loss=0.3815, pruned_loss=0.1136, over 28511.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3536, pruned_loss=0.09791, over 5701333.13 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3339, pruned_loss=0.08744, over 5739780.20 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3547, pruned_loss=0.09929, over 5693275.24 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:00:51,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8876, 3.7035, 3.5084, 1.8122], device='cuda:1'), covar=tensor([0.0649, 0.0814, 0.0747, 0.2302], device='cuda:1'), in_proj_covar=tensor([0.1153, 0.1072, 0.0913, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 22:00:57,277 INFO [optim.py:369] (1/2) 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 22:01:06,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-09 22:01:07,262 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 19000, giga_loss[loss=0.2742, simple_loss=0.3571, pruned_loss=0.09568, over 28952.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3519, pruned_loss=0.09616, over 5712628.13 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3339, pruned_loss=0.08728, over 5745385.85 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3536, pruned_loss=0.09782, over 5699538.00 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:01:25,982 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:01:31,474 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:01:34,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:01:56,846 INFO [train.py:968] (1/2) Epoch 19, batch 19050, giga_loss[loss=0.2909, simple_loss=0.3655, pruned_loss=0.1082, over 28930.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3518, pruned_loss=0.09609, over 5707914.68 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3339, pruned_loss=0.08726, over 5739528.70 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3534, pruned_loss=0.09755, over 5701791.44 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:01:57,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=841638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:02:20,672 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 19100, giga_loss[loss=0.3622, simple_loss=0.3879, pruned_loss=0.1682, over 23675.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3546, pruned_loss=0.1005, over 5691067.03 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.08737, over 5742027.13 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3558, pruned_loss=0.1018, over 5683522.35 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:02:57,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2208, 1.4478, 1.4843, 1.1347], device='cuda:1'), covar=tensor([0.1219, 0.1958, 0.1017, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0690, 0.0927, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 22:03:31,544 INFO [train.py:968] (1/2) Epoch 19, batch 19150, giga_loss[loss=0.2889, simple_loss=0.3725, pruned_loss=0.1027, over 28107.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3585, pruned_loss=0.1057, over 5680942.72 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3344, pruned_loss=0.08748, over 5732642.26 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3597, pruned_loss=0.1068, over 5683345.71 frames. ], batch size: 77, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:03:51,167 INFO [optim.py:369] (1/2) 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 22:04:05,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:04:11,226 INFO [train.py:968] (1/2) Epoch 19, batch 19200, giga_loss[loss=0.2494, simple_loss=0.3255, pruned_loss=0.08663, over 28335.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3575, pruned_loss=0.106, over 5691187.47 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3347, pruned_loss=0.08781, over 5737073.84 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3587, pruned_loss=0.1071, over 5687726.06 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:04:13,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:04:55,026 INFO [train.py:968] (1/2) Epoch 19, batch 19250, giga_loss[loss=0.3068, simple_loss=0.3692, pruned_loss=0.1223, over 28895.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3549, pruned_loss=0.1049, over 5699236.55 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3349, pruned_loss=0.08786, over 5740205.02 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3561, pruned_loss=0.106, over 5692847.62 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:05:06,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841851.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 22:05:15,666 INFO [optim.py:369] (1/2) 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 22:05:38,193 INFO [train.py:968] (1/2) Epoch 19, batch 19300, giga_loss[loss=0.2414, simple_loss=0.327, pruned_loss=0.07786, over 28901.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3538, pruned_loss=0.1038, over 5696684.90 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3352, pruned_loss=0.08771, over 5744028.98 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3551, pruned_loss=0.1054, over 5686683.51 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:05:51,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6284, 1.6763, 1.8690, 1.4382], device='cuda:1'), covar=tensor([0.1707, 0.2259, 0.1368, 0.1624], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0689, 0.0925, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 22:06:01,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4625, 1.7414, 1.4030, 1.4107], device='cuda:1'), covar=tensor([0.2660, 0.2697, 0.3092, 0.2355], device='cuda:1'), in_proj_covar=tensor([0.1446, 0.1052, 0.1284, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 22:06:07,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:10,057 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:14,315 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=841932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:16,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1348, 1.1957, 1.1254, 0.8557], device='cuda:1'), covar=tensor([0.1040, 0.0562, 0.1065, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0443, 0.0515, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:06:16,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=841935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:19,253 INFO [train.py:968] (1/2) Epoch 19, batch 19350, giga_loss[loss=0.2336, simple_loss=0.3179, pruned_loss=0.07468, over 28916.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3525, pruned_loss=0.1022, over 5689457.48 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3357, pruned_loss=0.08782, over 5738793.93 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3538, pruned_loss=0.104, over 5684313.58 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:06:28,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:33,234 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=841956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:40,559 INFO [optim.py:369] (1/2) 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 22:06:41,974 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=841964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:48,904 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=841970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:06:54,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=841976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:07:05,207 INFO [train.py:968] (1/2) Epoch 19, batch 19400, giga_loss[loss=0.2454, simple_loss=0.3294, pruned_loss=0.08067, over 28998.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3501, pruned_loss=0.1002, over 5690725.95 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3358, pruned_loss=0.08778, over 5739083.07 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3513, pruned_loss=0.1019, over 5685362.07 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:07:50,284 INFO [train.py:968] (1/2) Epoch 19, batch 19450, giga_loss[loss=0.2246, simple_loss=0.3037, pruned_loss=0.07274, over 28779.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3444, pruned_loss=0.09714, over 5678470.80 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.336, pruned_loss=0.08787, over 5730772.38 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3453, pruned_loss=0.09849, over 5680791.23 frames. ], batch size: 243, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:08:18,862 INFO [optim.py:369] (1/2) 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 22:08:41,157 INFO [train.py:968] (1/2) Epoch 19, batch 19500, giga_loss[loss=0.2433, simple_loss=0.315, pruned_loss=0.08577, over 28908.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3387, pruned_loss=0.09433, over 5670421.71 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3361, pruned_loss=0.08797, over 5721297.52 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3393, pruned_loss=0.09533, over 5680604.86 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:08:47,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:08:50,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:08:55,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3241, 1.3697, 1.2576, 1.5425], device='cuda:1'), covar=tensor([0.0809, 0.0350, 0.0344, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0104], device='cuda:1')
+2023-03-09 22:09:15,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:17,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:19,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:32,325 INFO [train.py:968] (1/2) Epoch 19, batch 19550, giga_loss[loss=0.2394, simple_loss=0.3266, pruned_loss=0.07613, over 28846.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3337, pruned_loss=0.09185, over 5677997.19 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3359, pruned_loss=0.08787, over 5723990.13 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3344, pruned_loss=0.0928, over 5683029.90 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:09:46,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:09:57,063 INFO [optim.py:369] (1/2) 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 22:10:15,624 INFO [train.py:968] (1/2) Epoch 19, batch 19600, giga_loss[loss=0.2249, simple_loss=0.3063, pruned_loss=0.07172, over 28866.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3339, pruned_loss=0.09117, over 5685993.03 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.336, pruned_loss=0.08788, over 5731054.22 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3342, pruned_loss=0.09208, over 5681692.05 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:10:23,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5976, 1.7681, 1.4711, 1.8487], device='cuda:1'), covar=tensor([0.2417, 0.2531, 0.2708, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.1449, 0.1053, 0.1286, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 22:10:35,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:10:47,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 22:11:00,705 INFO [train.py:968] (1/2) Epoch 19, batch 19650, giga_loss[loss=0.257, simple_loss=0.3364, pruned_loss=0.08878, over 29068.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3344, pruned_loss=0.09079, over 5696721.19 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3363, pruned_loss=0.0879, over 5732690.35 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3344, pruned_loss=0.09152, over 5691435.77 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:11:23,515 INFO [optim.py:369] (1/2) 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 22:11:40,577 INFO [train.py:968] (1/2) Epoch 19, batch 19700, libri_loss[loss=0.2961, simple_loss=0.3781, pruned_loss=0.107, over 29166.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3348, pruned_loss=0.09123, over 5706381.38 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3372, pruned_loss=0.08819, over 5737123.02 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3339, pruned_loss=0.09166, over 5696774.88 frames. ], batch size: 101, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:12:20,464 INFO [train.py:968] (1/2) Epoch 19, batch 19750, giga_loss[loss=0.2159, simple_loss=0.3013, pruned_loss=0.06524, over 28928.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3325, pruned_loss=0.09006, over 5710744.53 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3377, pruned_loss=0.08838, over 5733490.66 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3313, pruned_loss=0.09033, over 5705503.73 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:12:26,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:12:39,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-09 22:12:40,997 INFO [optim.py:369] (1/2) 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 22:12:44,250 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842369.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 22:12:46,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 22:12:51,540 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:12:59,797 INFO [train.py:968] (1/2) Epoch 19, batch 19800, giga_loss[loss=0.2609, simple_loss=0.3279, pruned_loss=0.09693, over 28796.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3313, pruned_loss=0.08961, over 5721814.19 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.338, pruned_loss=0.08827, over 5739655.59 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3298, pruned_loss=0.08999, over 5711377.86 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:13:11,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842401.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 22:13:16,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4885, 1.7427, 1.4088, 1.4972], device='cuda:1'), covar=tensor([0.2641, 0.2683, 0.3044, 0.2377], device='cuda:1'), in_proj_covar=tensor([0.1451, 0.1054, 0.1288, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 22:13:41,481 INFO [train.py:968] (1/2) Epoch 19, batch 19850, giga_loss[loss=0.2393, simple_loss=0.3076, pruned_loss=0.08551, over 28468.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3287, pruned_loss=0.08876, over 5719300.89 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3384, pruned_loss=0.08836, over 5738041.26 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3271, pruned_loss=0.08899, over 5712277.41 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:14:06,505 INFO [optim.py:369] (1/2) 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 22:14:22,214 INFO [train.py:968] (1/2) Epoch 19, batch 19900, giga_loss[loss=0.2119, simple_loss=0.2896, pruned_loss=0.06708, over 28708.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.326, pruned_loss=0.08758, over 5721253.90 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3386, pruned_loss=0.08835, over 5736125.21 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3242, pruned_loss=0.08776, over 5717062.77 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:14:22,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:14:25,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:14:48,633 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:15:02,674 INFO [train.py:968] (1/2) Epoch 19, batch 19950, giga_loss[loss=0.2309, simple_loss=0.3105, pruned_loss=0.07565, over 28910.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3252, pruned_loss=0.08716, over 5713494.59 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3397, pruned_loss=0.08879, over 5731221.00 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3224, pruned_loss=0.08688, over 5713837.56 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:15:14,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3766, 5.2134, 4.9326, 2.2090], device='cuda:1'), covar=tensor([0.0403, 0.0537, 0.0515, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.1077, 0.0918, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0014, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 22:15:26,292 INFO [optim.py:369] (1/2) 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 22:15:43,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:15:43,857 INFO [train.py:968] (1/2) Epoch 19, batch 20000, giga_loss[loss=0.2267, simple_loss=0.3042, pruned_loss=0.07456, over 28698.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3233, pruned_loss=0.08666, over 5715600.88 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3399, pruned_loss=0.08883, over 5734124.17 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3207, pruned_loss=0.08638, over 5713231.70 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:15:58,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7536, 1.2797, 1.1776, 1.1294], device='cuda:1'), covar=tensor([0.1823, 0.1145, 0.2164, 0.1447], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0737, 0.0702, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 22:16:24,684 INFO [train.py:968] (1/2) Epoch 19, batch 20050, giga_loss[loss=0.229, simple_loss=0.2968, pruned_loss=0.08057, over 28524.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3215, pruned_loss=0.08524, over 5715653.76 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3406, pruned_loss=0.08908, over 5729080.59 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3183, pruned_loss=0.08472, over 5718420.68 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:16:46,261 INFO [optim.py:369] (1/2) 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 22:16:47,412 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:04,949 INFO [train.py:968] (1/2) Epoch 19, batch 20100, giga_loss[loss=0.2295, simple_loss=0.3084, pruned_loss=0.07526, over 28564.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3197, pruned_loss=0.08449, over 5720621.51 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3411, pruned_loss=0.08934, over 5730900.25 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3166, pruned_loss=0.08384, over 5721218.69 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:17:19,414 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=842708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:36,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:40,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:17:43,361 INFO [train.py:968] (1/2) Epoch 19, batch 20150, giga_loss[loss=0.2535, simple_loss=0.3293, pruned_loss=0.08886, over 28896.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3187, pruned_loss=0.08385, over 5728632.25 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3416, pruned_loss=0.08946, over 5730893.35 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3153, pruned_loss=0.08308, over 5728807.91 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:17:54,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=842753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:18:02,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:18:05,032 INFO [optim.py:369] (1/2) 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 22:18:11,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9033, 3.7272, 3.5149, 1.5957], device='cuda:1'), covar=tensor([0.0692, 0.0827, 0.0778, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1167, 0.1080, 0.0923, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 22:18:24,494 INFO [train.py:968] (1/2) Epoch 19, batch 20200, libri_loss[loss=0.2214, simple_loss=0.307, pruned_loss=0.06793, over 29572.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3235, pruned_loss=0.08627, over 5733173.23 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.08957, over 5740329.52 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3189, pruned_loss=0.08531, over 5724733.15 frames. ], batch size: 74, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:18:42,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6627, 4.4036, 1.7316, 1.8631], device='cuda:1'), covar=tensor([0.0954, 0.0253, 0.0880, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0534, 0.0371, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 22:19:08,824 INFO [train.py:968] (1/2) Epoch 19, batch 20250, giga_loss[loss=0.325, simple_loss=0.3863, pruned_loss=0.1318, over 28739.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3298, pruned_loss=0.09039, over 5722607.80 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3435, pruned_loss=0.09024, over 5740588.76 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3247, pruned_loss=0.08899, over 5715150.98 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:19:26,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5930, 1.7236, 1.5854, 1.4680], device='cuda:1'), covar=tensor([0.2111, 0.2013, 0.1848, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.1907, 0.1799, 0.1753, 0.1904], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 22:19:36,385 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 20300, giga_loss[loss=0.2843, simple_loss=0.3678, pruned_loss=0.1004, over 29021.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3389, pruned_loss=0.09669, over 5706806.87 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3438, pruned_loss=0.09034, over 5743652.08 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3343, pruned_loss=0.09555, over 5697450.85 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:20:07,950 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=842896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:20:11,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=842899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:20:35,456 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=842928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:20:40,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1537, 1.2244, 3.2940, 2.9640], device='cuda:1'), covar=tensor([0.1583, 0.2704, 0.0492, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0731, 0.0630, 0.0923, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:20:45,593 INFO [train.py:968] (1/2) Epoch 19, batch 20350, giga_loss[loss=0.269, simple_loss=0.3524, pruned_loss=0.09284, over 28672.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3452, pruned_loss=0.1001, over 5689144.83 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3445, pruned_loss=0.09071, over 5735179.69 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.341, pruned_loss=0.09905, over 5688071.05 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:21:05,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-09 22:21:14,576 INFO [optim.py:369] (1/2) 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 22:21:33,438 INFO [train.py:968] (1/2) Epoch 19, batch 20400, giga_loss[loss=0.3118, simple_loss=0.387, pruned_loss=0.1183, over 28864.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3491, pruned_loss=0.1017, over 5685658.71 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3445, pruned_loss=0.09082, over 5739701.59 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3458, pruned_loss=0.101, over 5679394.66 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:22:08,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3501, 1.5971, 1.2957, 0.9723], device='cuda:1'), covar=tensor([0.2703, 0.2800, 0.3110, 0.2361], device='cuda:1'), in_proj_covar=tensor([0.1454, 0.1057, 0.1291, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 22:22:17,289 INFO [train.py:968] (1/2) Epoch 19, batch 20450, giga_loss[loss=0.2829, simple_loss=0.3573, pruned_loss=0.1042, over 28781.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3524, pruned_loss=0.1029, over 5689781.25 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3449, pruned_loss=0.09116, over 5744859.72 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3495, pruned_loss=0.1025, over 5677795.40 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:22:22,915 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=843041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:22:47,703 INFO [optim.py:369] (1/2) 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 22:23:01,493 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 20500, giga_loss[loss=0.258, simple_loss=0.3382, pruned_loss=0.08891, over 29056.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3582, pruned_loss=0.107, over 5681230.32 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3447, pruned_loss=0.09109, over 5745483.42 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3562, pruned_loss=0.1068, over 5670978.16 frames. ], batch size: 128, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:23:52,329 INFO [train.py:968] (1/2) Epoch 19, batch 20550, giga_loss[loss=0.2638, simple_loss=0.3398, pruned_loss=0.0939, over 28015.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.353, pruned_loss=0.103, over 5683903.14 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3449, pruned_loss=0.09127, over 5747094.90 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3514, pruned_loss=0.1029, over 5673119.50 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:24:17,137 INFO [optim.py:369] (1/2) 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 22:24:30,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=843184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:24:33,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=843187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:24:33,980 INFO [train.py:968] (1/2) Epoch 19, batch 20600, giga_loss[loss=0.315, simple_loss=0.3836, pruned_loss=0.1233, over 28915.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3502, pruned_loss=0.1007, over 5686646.22 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3449, pruned_loss=0.0915, over 5739637.65 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3491, pruned_loss=0.1006, over 5683473.16 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:24:51,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-09 22:24:57,791 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=843216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:08,124 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=843226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:11,570 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=843229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:18,618 INFO [train.py:968] (1/2) Epoch 19, batch 20650, giga_loss[loss=0.2616, simple_loss=0.3417, pruned_loss=0.09074, over 28535.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3504, pruned_loss=0.1008, over 5691258.00 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.346, pruned_loss=0.09234, over 5742397.14 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3487, pruned_loss=0.1002, over 5684822.73 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:25:36,623 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=843258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:25:43,141 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 20700, giga_loss[loss=0.2919, simple_loss=0.3525, pruned_loss=0.1157, over 28743.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3509, pruned_loss=0.1006, over 5691411.54 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.346, pruned_loss=0.09256, over 5743877.23 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3496, pruned_loss=0.1001, over 5683348.17 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:26:09,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2128, 2.5958, 1.2856, 1.3523], device='cuda:1'), covar=tensor([0.1051, 0.0340, 0.0862, 0.1418], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0536, 0.0371, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 22:26:38,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3565, 1.3327, 4.0414, 3.2658], device='cuda:1'), covar=tensor([0.1601, 0.2638, 0.0413, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0735, 0.0632, 0.0929, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:26:45,711 INFO [train.py:968] (1/2) Epoch 19, batch 20750, giga_loss[loss=0.2623, simple_loss=0.3418, pruned_loss=0.09138, over 28562.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3527, pruned_loss=0.1018, over 5696992.51 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3458, pruned_loss=0.09249, over 5746741.07 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.352, pruned_loss=0.1017, over 5686994.59 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:27:10,189 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:27:10,554 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 20800, giga_loss[loss=0.2837, simple_loss=0.3618, pruned_loss=0.1028, over 28887.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3536, pruned_loss=0.1026, over 5702796.01 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.346, pruned_loss=0.09275, over 5742031.68 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3531, pruned_loss=0.1025, over 5697003.87 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:28:03,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4472, 1.5039, 1.5954, 1.3720], device='cuda:1'), covar=tensor([0.1742, 0.1975, 0.2167, 0.1924], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0742, 0.0708, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 22:28:14,658 INFO [train.py:968] (1/2) Epoch 19, batch 20850, libri_loss[loss=0.2633, simple_loss=0.3434, pruned_loss=0.09165, over 29565.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3546, pruned_loss=0.1039, over 5685677.41 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3463, pruned_loss=0.09304, over 5743604.98 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3541, pruned_loss=0.1038, over 5678196.49 frames. ], batch size: 79, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:28:18,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-09 22:28:40,337 INFO [optim.py:369] (1/2) 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 22:28:59,156 INFO [train.py:968] (1/2) Epoch 19, batch 20900, giga_loss[loss=0.28, simple_loss=0.3521, pruned_loss=0.1039, over 29005.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3562, pruned_loss=0.1055, over 5690721.69 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3463, pruned_loss=0.09307, over 5745922.55 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3559, pruned_loss=0.1056, over 5681667.18 frames. ], batch size: 106, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:29:16,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-09 22:29:39,247 INFO [train.py:968] (1/2) Epoch 19, batch 20950, giga_loss[loss=0.2844, simple_loss=0.355, pruned_loss=0.1069, over 28852.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3558, pruned_loss=0.1051, over 5698604.16 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3466, pruned_loss=0.0934, over 5746261.00 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3555, pruned_loss=0.1051, over 5690439.78 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:30:02,280 INFO [optim.py:369] (1/2) 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 22:30:21,110 INFO [train.py:968] (1/2) Epoch 19, batch 21000, giga_loss[loss=0.2398, simple_loss=0.321, pruned_loss=0.07932, over 28661.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3548, pruned_loss=0.1037, over 5699805.92 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3469, pruned_loss=0.09369, over 5748533.78 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3544, pruned_loss=0.1036, over 5690725.60 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:30:21,110 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 22:30:31,411 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 22:30:31,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6129, 1.7273, 1.8507, 1.4099], device='cuda:1'), covar=tensor([0.1752, 0.2486, 0.1397, 0.1727], device='cuda:1'), in_proj_covar=tensor([0.0886, 0.0697, 0.0929, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-09 22:30:34,791 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:31:11,572 INFO [train.py:968] (1/2) Epoch 19, batch 21050, giga_loss[loss=0.2464, simple_loss=0.3295, pruned_loss=0.08166, over 28576.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3555, pruned_loss=0.1026, over 5694369.40 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3473, pruned_loss=0.09397, over 5740015.07 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3549, pruned_loss=0.1024, over 5694449.59 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:31:37,803 INFO [optim.py:369] (1/2) 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 22:31:38,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6776, 1.8166, 1.4462, 2.1026], device='cuda:1'), covar=tensor([0.2582, 0.2654, 0.3024, 0.2231], device='cuda:1'), in_proj_covar=tensor([0.1457, 0.1059, 0.1292, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 22:31:41,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3271, 1.1547, 4.3893, 3.4445], device='cuda:1'), covar=tensor([0.1706, 0.2981, 0.0359, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0634, 0.0931, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:31:44,964 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:31:52,276 INFO [train.py:968] (1/2) Epoch 19, batch 21100, giga_loss[loss=0.273, simple_loss=0.3426, pruned_loss=0.1017, over 28649.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3549, pruned_loss=0.1024, over 5696764.05 frames. ], libri_tot_loss[loss=0.2683, simple_loss=0.3478, pruned_loss=0.09442, over 5743842.49 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3542, pruned_loss=0.102, over 5692338.27 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:32:08,861 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:32:30,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.92 vs. limit=5.0
+2023-03-09 22:32:32,726 INFO [train.py:968] (1/2) Epoch 19, batch 21150, giga_loss[loss=0.2696, simple_loss=0.3486, pruned_loss=0.09527, over 28324.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3528, pruned_loss=0.1012, over 5708810.40 frames. ], libri_tot_loss[loss=0.2687, simple_loss=0.3481, pruned_loss=0.09465, over 5744669.21 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.352, pruned_loss=0.1007, over 5704206.84 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:32:34,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=843741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:32:50,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4682, 1.5996, 1.7093, 1.3007], device='cuda:1'), covar=tensor([0.1648, 0.2577, 0.1350, 0.1670], device='cuda:1'), in_proj_covar=tensor([0.0886, 0.0698, 0.0931, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-09 22:32:56,505 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 21200, giga_loss[loss=0.2683, simple_loss=0.3379, pruned_loss=0.09936, over 28615.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 5715076.99 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3487, pruned_loss=0.09521, over 5749528.01 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09967, over 5705974.79 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:33:36,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2432, 2.7339, 1.8070, 1.9399], device='cuda:1'), covar=tensor([0.0915, 0.0543, 0.0929, 0.1097], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0440, 0.0511, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:33:38,834 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=843821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:33:52,573 INFO [train.py:968] (1/2) Epoch 19, batch 21250, giga_loss[loss=0.2911, simple_loss=0.3567, pruned_loss=0.1128, over 28965.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3491, pruned_loss=0.09958, over 5718272.60 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3491, pruned_loss=0.0955, over 5754361.56 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3479, pruned_loss=0.09885, over 5705600.81 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:34:16,894 INFO [optim.py:369] (1/2) 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 22:34:29,508 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=843884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:34:31,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 21300, giga_loss[loss=0.2427, simple_loss=0.3231, pruned_loss=0.08116, over 28492.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3514, pruned_loss=0.1015, over 5719291.31 frames. ], libri_tot_loss[loss=0.2708, simple_loss=0.3496, pruned_loss=0.096, over 5758920.25 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.35, pruned_loss=0.1006, over 5704205.30 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:34:34,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9476, 2.0223, 1.8390, 1.7820], device='cuda:1'), covar=tensor([0.1978, 0.2552, 0.2369, 0.2275], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0738, 0.0703, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 22:34:59,285 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=843916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:35:11,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4062, 1.8407, 1.3047, 0.9305], device='cuda:1'), covar=tensor([0.4545, 0.2394, 0.2216, 0.3876], device='cuda:1'), in_proj_covar=tensor([0.1691, 0.1586, 0.1568, 0.1384], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 22:35:16,412 INFO [train.py:968] (1/2) Epoch 19, batch 21350, giga_loss[loss=0.2738, simple_loss=0.3569, pruned_loss=0.09536, over 28883.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3513, pruned_loss=0.1011, over 5709564.72 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3501, pruned_loss=0.09639, over 5748261.93 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3497, pruned_loss=0.1001, over 5706859.39 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:35:42,161 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=843968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:35:42,555 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 21400, giga_loss[loss=0.2508, simple_loss=0.3378, pruned_loss=0.08184, over 28560.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09965, over 5705381.98 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3503, pruned_loss=0.09666, over 5747580.28 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3484, pruned_loss=0.09871, over 5703038.73 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:36:39,568 INFO [train.py:968] (1/2) Epoch 19, batch 21450, giga_loss[loss=0.2529, simple_loss=0.3363, pruned_loss=0.08475, over 28550.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3494, pruned_loss=0.09887, over 5716749.09 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3502, pruned_loss=0.09674, over 5750110.28 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3483, pruned_loss=0.0981, over 5712109.67 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:36:44,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4035, 1.3724, 1.2736, 1.5195], device='cuda:1'), covar=tensor([0.0797, 0.0349, 0.0341, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0115, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 22:36:49,299 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:37:14,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:17,439 INFO [train.py:968] (1/2) Epoch 19, batch 21500, giga_loss[loss=0.2663, simple_loss=0.3381, pruned_loss=0.09725, over 28798.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3482, pruned_loss=0.09831, over 5721888.45 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3505, pruned_loss=0.09713, over 5747603.66 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.09739, over 5719473.42 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:37:36,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:39,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:55,400 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:37:55,857 INFO [train.py:968] (1/2) Epoch 19, batch 21550, giga_loss[loss=0.2377, simple_loss=0.3148, pruned_loss=0.08024, over 28890.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3449, pruned_loss=0.09693, over 5714198.45 frames. ], libri_tot_loss[loss=0.2731, simple_loss=0.3508, pruned_loss=0.09771, over 5742913.46 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3435, pruned_loss=0.09565, over 5715551.08 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:37:59,510 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:13,776 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1971, 1.7592, 1.3848, 0.4663], device='cuda:1'), covar=tensor([0.4792, 0.2715, 0.4482, 0.5889], device='cuda:1'), in_proj_covar=tensor([0.1688, 0.1583, 0.1563, 0.1381], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 22:38:22,143 INFO [optim.py:369] (1/2) 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 22:38:38,313 INFO [train.py:968] (1/2) Epoch 19, batch 21600, giga_loss[loss=0.2971, simple_loss=0.3764, pruned_loss=0.1089, over 28377.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3428, pruned_loss=0.0961, over 5706605.19 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3514, pruned_loss=0.09837, over 5736318.94 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3411, pruned_loss=0.09446, over 5714040.32 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:38:41,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2807, 1.5397, 1.2652, 1.4328], device='cuda:1'), covar=tensor([0.0781, 0.0382, 0.0341, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0115, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 22:38:43,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4035, 3.2308, 1.5778, 1.4562], device='cuda:1'), covar=tensor([0.0958, 0.0265, 0.0858, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0534, 0.0371, 0.0413], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 22:38:43,149 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:43,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:45,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:38:59,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2439, 1.6408, 1.0381, 1.1766], device='cuda:1'), covar=tensor([0.1250, 0.0630, 0.1597, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0441, 0.0512, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:39:08,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:39:08,851 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:39:10,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:39:16,277 INFO [train.py:968] (1/2) Epoch 19, batch 21650, giga_loss[loss=0.2759, simple_loss=0.3499, pruned_loss=0.101, over 29068.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3427, pruned_loss=0.09642, over 5714069.38 frames. ], libri_tot_loss[loss=0.2745, simple_loss=0.3516, pruned_loss=0.09874, over 5737047.72 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3409, pruned_loss=0.09473, over 5718554.64 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:39:33,604 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 19, batch 21700, giga_loss[loss=0.2981, simple_loss=0.3672, pruned_loss=0.1145, over 28236.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3418, pruned_loss=0.09654, over 5708043.17 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3516, pruned_loss=0.09891, over 5731946.92 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3402, pruned_loss=0.09498, over 5714812.15 frames. ], batch size: 77, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:40:29,455 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:40:39,281 INFO [train.py:968] (1/2) Epoch 19, batch 21750, giga_loss[loss=0.2377, simple_loss=0.3185, pruned_loss=0.07848, over 29050.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3412, pruned_loss=0.09707, over 5707611.02 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3517, pruned_loss=0.09915, over 5730137.93 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3395, pruned_loss=0.09551, over 5713908.39 frames. ], batch size: 128, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:40:40,300 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:40:42,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:41:04,248 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:41:05,331 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 21800, giga_loss[loss=0.2719, simple_loss=0.341, pruned_loss=0.1014, over 27648.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3392, pruned_loss=0.09613, over 5712951.96 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3518, pruned_loss=0.09943, over 5734217.01 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3373, pruned_loss=0.09454, over 5713758.58 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:41:56,635 INFO [train.py:968] (1/2) Epoch 19, batch 21850, libri_loss[loss=0.2859, simple_loss=0.3562, pruned_loss=0.1078, over 29545.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3372, pruned_loss=0.09536, over 5708987.92 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3526, pruned_loss=0.09999, over 5731045.24 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3343, pruned_loss=0.09337, over 5710936.58 frames. ], batch size: 89, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:42:22,743 INFO [optim.py:369] (1/2) 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 22:42:38,526 INFO [train.py:968] (1/2) Epoch 19, batch 21900, giga_loss[loss=0.2436, simple_loss=0.3246, pruned_loss=0.08127, over 28953.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3351, pruned_loss=0.09444, over 5705267.82 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3527, pruned_loss=0.1001, over 5731495.02 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3326, pruned_loss=0.09279, over 5706310.93 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:43:00,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:43:21,619 INFO [train.py:968] (1/2) Epoch 19, batch 21950, giga_loss[loss=0.2514, simple_loss=0.3359, pruned_loss=0.08343, over 28704.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3369, pruned_loss=0.09558, over 5700222.16 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3532, pruned_loss=0.1006, over 5730082.54 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3341, pruned_loss=0.09362, over 5701644.73 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:43:21,781 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:43:52,013 INFO [optim.py:369] (1/2) 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 22:44:04,868 INFO [train.py:968] (1/2) Epoch 19, batch 22000, giga_loss[loss=0.2599, simple_loss=0.3414, pruned_loss=0.08927, over 28847.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3399, pruned_loss=0.09663, over 5703355.21 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.354, pruned_loss=0.1014, over 5730953.82 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3365, pruned_loss=0.09426, over 5703157.29 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:44:17,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-09 22:44:31,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3691, 1.4224, 1.2140, 1.4872], device='cuda:1'), covar=tensor([0.0746, 0.0317, 0.0358, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0115, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0094, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 22:44:43,102 INFO [train.py:968] (1/2) Epoch 19, batch 22050, giga_loss[loss=0.2465, simple_loss=0.3281, pruned_loss=0.08244, over 28820.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3415, pruned_loss=0.09732, over 5714981.41 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3538, pruned_loss=0.1017, over 5737844.71 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3382, pruned_loss=0.09476, over 5707125.99 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:44:56,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:44:59,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:45:09,087 INFO [optim.py:369] (1/2) 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 22:45:14,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-09 22:45:23,875 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 22100, giga_loss[loss=0.3018, simple_loss=0.36, pruned_loss=0.1218, over 23651.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3436, pruned_loss=0.09784, over 5709899.85 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3538, pruned_loss=0.102, over 5743085.61 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3407, pruned_loss=0.09547, over 5698320.67 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:45:35,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:45:41,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-09 22:45:41,978 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 22:46:07,936 INFO [train.py:968] (1/2) Epoch 19, batch 22150, giga_loss[loss=0.2605, simple_loss=0.3278, pruned_loss=0.0966, over 28986.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3434, pruned_loss=0.09693, over 5702798.93 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.354, pruned_loss=0.1022, over 5744668.66 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3407, pruned_loss=0.0948, over 5691949.09 frames. ], batch size: 106, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:46:14,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:46:38,348 INFO [optim.py:369] (1/2) 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 22:46:54,611 INFO [train.py:968] (1/2) Epoch 19, batch 22200, giga_loss[loss=0.2401, simple_loss=0.3189, pruned_loss=0.08065, over 28781.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09633, over 5704304.84 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3543, pruned_loss=0.1025, over 5746001.37 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3403, pruned_loss=0.09432, over 5694273.50 frames. ], batch size: 112, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:47:13,409 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-09 22:47:34,936 INFO [train.py:968] (1/2) Epoch 19, batch 22250, libri_loss[loss=0.3324, simple_loss=0.3934, pruned_loss=0.1357, over 29528.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3446, pruned_loss=0.09783, over 5698264.75 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3554, pruned_loss=0.1033, over 5738550.15 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3414, pruned_loss=0.09534, over 5695763.15 frames. ], batch size: 81, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:47:40,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:47:42,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:03,266 INFO [optim.py:369] (1/2) 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 22:48:06,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=844875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:16,682 INFO [train.py:968] (1/2) Epoch 19, batch 22300, giga_loss[loss=0.2616, simple_loss=0.3415, pruned_loss=0.09078, over 28650.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3463, pruned_loss=0.09947, over 5692143.71 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3555, pruned_loss=0.1035, over 5727289.29 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3435, pruned_loss=0.09724, over 5699767.03 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:48:16,972 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=844888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:19,416 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=844891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:38,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=844913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:42,738 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=844918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:48:43,974 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 22350, giga_loss[loss=0.2926, simple_loss=0.3658, pruned_loss=0.1097, over 29086.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3493, pruned_loss=0.1011, over 5688144.37 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3567, pruned_loss=0.1044, over 5722890.14 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3458, pruned_loss=0.09845, over 5697460.25 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:49:02,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9206, 1.1396, 1.1305, 0.8629], device='cuda:1'), covar=tensor([0.2127, 0.2237, 0.1315, 0.1955], device='cuda:1'), in_proj_covar=tensor([0.1922, 0.1829, 0.1780, 0.1915], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 22:49:07,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 22:49:09,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3879, 2.4509, 1.9641, 1.9456], device='cuda:1'), covar=tensor([0.0808, 0.0659, 0.0912, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0442, 0.0511, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:49:09,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-09 22:49:25,077 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 22400, giga_loss[loss=0.2757, simple_loss=0.3604, pruned_loss=0.09546, over 28607.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3513, pruned_loss=0.1018, over 5694801.23 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3563, pruned_loss=0.1043, over 5721307.84 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3487, pruned_loss=0.09968, over 5702848.08 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:50:18,436 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:50:20,148 INFO [train.py:968] (1/2) Epoch 19, batch 22450, giga_loss[loss=0.3098, simple_loss=0.3787, pruned_loss=0.1205, over 28805.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3531, pruned_loss=0.1027, over 5695189.21 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3565, pruned_loss=0.1045, over 5718865.81 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3507, pruned_loss=0.1007, over 5703551.45 frames. ], batch size: 199, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:50:33,975 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=845056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:50:35,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=845059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:50:43,650 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 22500, giga_loss[loss=0.2706, simple_loss=0.3484, pruned_loss=0.09643, over 28812.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5703073.29 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3567, pruned_loss=0.1047, over 5724249.54 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 5704518.90 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:50:58,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=845088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:51:28,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1029, 1.1537, 3.4351, 2.9768], device='cuda:1'), covar=tensor([0.1594, 0.2588, 0.0544, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0633, 0.0933, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 22:51:38,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4396, 1.7582, 1.3644, 1.2588], device='cuda:1'), covar=tensor([0.2779, 0.2785, 0.3170, 0.2340], device='cuda:1'), in_proj_covar=tensor([0.1454, 0.1053, 0.1287, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 22:51:42,588 INFO [train.py:968] (1/2) Epoch 19, batch 22550, giga_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1044, over 28970.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3549, pruned_loss=0.1036, over 5709867.88 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3574, pruned_loss=0.1053, over 5727901.51 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3525, pruned_loss=0.1016, over 5707259.33 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:52:10,119 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 22600, giga_loss[loss=0.2951, simple_loss=0.3785, pruned_loss=0.1059, over 28738.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3533, pruned_loss=0.1029, over 5698926.24 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3575, pruned_loss=0.1056, over 5719284.83 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3512, pruned_loss=0.101, over 5704185.89 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:53:06,477 INFO [train.py:968] (1/2) Epoch 19, batch 22650, giga_loss[loss=0.2975, simple_loss=0.3709, pruned_loss=0.112, over 28911.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3505, pruned_loss=0.1013, over 5704652.21 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3578, pruned_loss=0.1058, over 5720647.61 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3484, pruned_loss=0.09948, over 5707265.53 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:53:33,878 INFO [optim.py:369] (1/2) 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 22:53:48,535 INFO [train.py:968] (1/2) Epoch 19, batch 22700, giga_loss[loss=0.2252, simple_loss=0.2999, pruned_loss=0.07524, over 28495.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.346, pruned_loss=0.09882, over 5705476.60 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3579, pruned_loss=0.1059, over 5721566.38 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3442, pruned_loss=0.0973, over 5706554.28 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:53:54,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=845293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:54:29,117 INFO [train.py:968] (1/2) Epoch 19, batch 22750, giga_loss[loss=0.2741, simple_loss=0.3559, pruned_loss=0.09614, over 28264.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3445, pruned_loss=0.09757, over 5705519.24 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3581, pruned_loss=0.1062, over 5724623.37 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3427, pruned_loss=0.09591, over 5703366.10 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 22:54:59,961 INFO [optim.py:369] (1/2) 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 22:55:11,934 INFO [train.py:968] (1/2) Epoch 19, batch 22800, giga_loss[loss=0.2582, simple_loss=0.3528, pruned_loss=0.08174, over 28810.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3462, pruned_loss=0.0971, over 5700467.26 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3585, pruned_loss=0.1068, over 5723508.13 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3438, pruned_loss=0.09487, over 5698761.16 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:55:20,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-09 22:55:33,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=845410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:55:55,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=845436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:55:56,008 INFO [train.py:968] (1/2) Epoch 19, batch 22850, giga_loss[loss=0.256, simple_loss=0.3367, pruned_loss=0.08766, over 28718.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3475, pruned_loss=0.09718, over 5698868.16 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3584, pruned_loss=0.1068, over 5724281.59 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3456, pruned_loss=0.09539, over 5696723.89 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:55:57,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=845439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:56:19,929 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:56:37,386 INFO [train.py:968] (1/2) Epoch 19, batch 22900, giga_loss[loss=0.2633, simple_loss=0.3367, pruned_loss=0.09498, over 28905.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3448, pruned_loss=0.09706, over 5699442.34 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3584, pruned_loss=0.107, over 5727069.45 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3431, pruned_loss=0.09531, over 5694848.84 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 22:57:00,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:57:06,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8434, 5.6400, 5.3071, 3.3953], device='cuda:1'), covar=tensor([0.0442, 0.0631, 0.0680, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.1160, 0.1083, 0.0923, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 22:57:19,925 INFO [train.py:968] (1/2) Epoch 19, batch 22950, giga_loss[loss=0.2529, simple_loss=0.3245, pruned_loss=0.09069, over 28996.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3445, pruned_loss=0.09849, over 5696799.32 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3594, pruned_loss=0.1078, over 5720937.65 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.342, pruned_loss=0.09615, over 5698596.25 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:57:31,992 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=845553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:57:35,107 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:57:52,949 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-09 22:57:58,455 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 23000, giga_loss[loss=0.2735, simple_loss=0.3425, pruned_loss=0.1023, over 28933.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3422, pruned_loss=0.09788, over 5700581.57 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3596, pruned_loss=0.1078, over 5714258.33 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3397, pruned_loss=0.09578, over 5707963.11 frames. ], batch size: 145, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:58:13,548 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 22:58:15,722 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-09 22:58:42,798 INFO [train.py:968] (1/2) Epoch 19, batch 23050, giga_loss[loss=0.3037, simple_loss=0.3714, pruned_loss=0.118, over 28664.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3408, pruned_loss=0.09809, over 5699053.78 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3598, pruned_loss=0.1081, over 5716123.92 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3385, pruned_loss=0.09615, over 5703141.73 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:59:10,956 INFO [optim.py:369] (1/2) 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 22:59:21,941 INFO [train.py:968] (1/2) Epoch 19, batch 23100, giga_loss[loss=0.2427, simple_loss=0.3235, pruned_loss=0.08098, over 28896.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3404, pruned_loss=0.09769, over 5708126.92 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3604, pruned_loss=0.1087, over 5714061.97 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3375, pruned_loss=0.09532, over 5713311.13 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 22:59:41,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4338, 1.8338, 1.6000, 1.5005], device='cuda:1'), covar=tensor([0.1643, 0.1331, 0.1871, 0.1613], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0741, 0.0706, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 22:59:56,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0337, 2.1696, 1.5441, 1.8474], device='cuda:1'), covar=tensor([0.0948, 0.0840, 0.1138, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0443, 0.0510, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:00:02,510 INFO [train.py:968] (1/2) Epoch 19, batch 23150, libri_loss[loss=0.2735, simple_loss=0.3495, pruned_loss=0.0988, over 29547.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.338, pruned_loss=0.0968, over 5699218.53 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3605, pruned_loss=0.1088, over 5710413.59 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3347, pruned_loss=0.0943, over 5706519.18 frames. ], batch size: 79, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:00:17,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845758.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:00:23,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9939, 1.2039, 1.2074, 0.9707], device='cuda:1'), covar=tensor([0.2014, 0.1958, 0.1206, 0.1681], device='cuda:1'), in_proj_covar=tensor([0.1929, 0.1841, 0.1790, 0.1919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:00:28,880 INFO [optim.py:369] (1/2) 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 23:00:37,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-09 23:00:41,161 INFO [train.py:968] (1/2) Epoch 19, batch 23200, giga_loss[loss=0.254, simple_loss=0.3253, pruned_loss=0.09134, over 28981.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.335, pruned_loss=0.09578, over 5699085.24 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3607, pruned_loss=0.1092, over 5708390.91 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3312, pruned_loss=0.09298, over 5706250.67 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:00:55,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2939, 0.9350, 1.0530, 1.4207], device='cuda:1'), covar=tensor([0.0727, 0.0372, 0.0349, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0217], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0103], device='cuda:1')
+2023-03-09 23:01:07,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6454, 2.1845, 1.4286, 0.9030], device='cuda:1'), covar=tensor([0.7333, 0.3866, 0.3415, 0.6703], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1584, 0.1569, 0.1384], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 23:01:21,758 INFO [train.py:968] (1/2) Epoch 19, batch 23250, giga_loss[loss=0.2983, simple_loss=0.3586, pruned_loss=0.119, over 26633.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3324, pruned_loss=0.09404, over 5702326.69 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3608, pruned_loss=0.1093, over 5710349.06 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3291, pruned_loss=0.0916, over 5706140.45 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:01:32,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=845851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:01:41,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1503, 2.4697, 2.2147, 1.7618], device='cuda:1'), covar=tensor([0.2821, 0.2037, 0.2198, 0.2735], device='cuda:1'), in_proj_covar=tensor([0.1928, 0.1845, 0.1793, 0.1922], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:01:53,968 INFO [optim.py:369] (1/2) 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 23:02:04,709 INFO [train.py:968] (1/2) Epoch 19, batch 23300, giga_loss[loss=0.3057, simple_loss=0.3732, pruned_loss=0.1191, over 28720.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3333, pruned_loss=0.09365, over 5707873.46 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3607, pruned_loss=0.1093, over 5713251.22 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3304, pruned_loss=0.09147, over 5708167.49 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:02:05,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=845889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:02:34,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1678, 1.3024, 3.7243, 3.1397], device='cuda:1'), covar=tensor([0.1719, 0.2757, 0.0422, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0634, 0.0932, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:02:45,851 INFO [train.py:968] (1/2) Epoch 19, batch 23350, giga_loss[loss=0.2842, simple_loss=0.3668, pruned_loss=0.1008, over 28818.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3367, pruned_loss=0.09519, over 5706805.72 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3608, pruned_loss=0.1096, over 5710878.48 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3332, pruned_loss=0.09253, over 5708801.16 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:03:16,423 INFO [optim.py:369] (1/2) 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 23:03:19,519 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 23400, giga_loss[loss=0.2569, simple_loss=0.3322, pruned_loss=0.09085, over 28839.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3411, pruned_loss=0.09718, over 5700002.09 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3607, pruned_loss=0.1096, over 5704278.37 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.338, pruned_loss=0.09483, over 5706527.45 frames. ], batch size: 66, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:03:44,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-09 23:03:46,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3323, 1.4583, 1.4135, 1.2544], device='cuda:1'), covar=tensor([0.2605, 0.2411, 0.1714, 0.2245], device='cuda:1'), in_proj_covar=tensor([0.1924, 0.1845, 0.1792, 0.1918], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:04:05,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:04:08,428 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 23450, giga_loss[loss=0.2747, simple_loss=0.3509, pruned_loss=0.09928, over 28550.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3442, pruned_loss=0.09859, over 5701621.08 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3604, pruned_loss=0.1095, over 5707971.36 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3417, pruned_loss=0.09658, over 5703145.71 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:04:34,884 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=846064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:04:43,981 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 23500, giga_loss[loss=0.2724, simple_loss=0.3501, pruned_loss=0.09738, over 29047.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.347, pruned_loss=0.09989, over 5694786.18 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3603, pruned_loss=0.1094, over 5709713.78 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.345, pruned_loss=0.09828, over 5694265.70 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:04:58,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-09 23:05:29,201 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:05:31,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=846124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:05:38,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=846133.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:05:44,892 INFO [train.py:968] (1/2) Epoch 19, batch 23550, giga_loss[loss=0.3818, simple_loss=0.4242, pruned_loss=0.1697, over 27920.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3531, pruned_loss=0.1054, over 5691579.83 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3605, pruned_loss=0.1096, over 5710168.41 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3512, pruned_loss=0.1039, over 5690602.29 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:05:59,262 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:06:22,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-09 23:06:34,592 INFO [train.py:968] (1/2) Epoch 19, batch 23600, libri_loss[loss=0.3, simple_loss=0.3612, pruned_loss=0.1194, over 29558.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3582, pruned_loss=0.1097, over 5687671.92 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3606, pruned_loss=0.1101, over 5712271.83 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3564, pruned_loss=0.108, over 5684286.64 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:06:47,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1619, 1.4120, 1.4180, 1.0549], device='cuda:1'), covar=tensor([0.1207, 0.1905, 0.0997, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0695, 0.0927, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:07:01,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3810, 3.2593, 1.5974, 1.4885], device='cuda:1'), covar=tensor([0.0965, 0.0366, 0.0898, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0542, 0.0374, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 23:07:10,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=846226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:07:23,858 INFO [train.py:968] (1/2) Epoch 19, batch 23650, giga_loss[loss=0.3169, simple_loss=0.3761, pruned_loss=0.1288, over 28825.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.366, pruned_loss=0.1151, over 5682098.69 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3613, pruned_loss=0.1105, over 5711080.12 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.364, pruned_loss=0.1134, over 5679718.00 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:07:31,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7214, 3.7787, 1.7020, 1.7293], device='cuda:1'), covar=tensor([0.0896, 0.0362, 0.0862, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0541, 0.0373, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 23:07:36,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6953, 1.7734, 1.9008, 1.4819], device='cuda:1'), covar=tensor([0.1750, 0.2294, 0.1345, 0.1608], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0694, 0.0926, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:07:59,611 INFO [optim.py:369] (1/2) 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 23:08:01,967 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846276.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:08:05,842 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=846279.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:08:10,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2630, 0.8243, 0.9546, 1.3673], device='cuda:1'), covar=tensor([0.0767, 0.0358, 0.0344, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0104], device='cuda:1')
+2023-03-09 23:08:13,324 INFO [train.py:968] (1/2) Epoch 19, batch 23700, giga_loss[loss=0.3063, simple_loss=0.369, pruned_loss=0.1218, over 28881.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.372, pruned_loss=0.1207, over 5684813.89 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3618, pruned_loss=0.1111, over 5716820.21 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3702, pruned_loss=0.119, over 5677190.03 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:08:35,995 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=846308.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:08:49,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2840, 1.2094, 3.5626, 3.0795], device='cuda:1'), covar=tensor([0.1526, 0.2739, 0.0477, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0736, 0.0631, 0.0929, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:09:02,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-09 23:09:03,456 INFO [train.py:968] (1/2) Epoch 19, batch 23750, giga_loss[loss=0.3444, simple_loss=0.4154, pruned_loss=0.1367, over 29023.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3792, pruned_loss=0.1271, over 5669085.30 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3619, pruned_loss=0.1114, over 5716676.02 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3779, pruned_loss=0.1258, over 5662112.80 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:09:32,233 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=846369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:09:36,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=846372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:09:40,856 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 23800, giga_loss[loss=0.3529, simple_loss=0.4124, pruned_loss=0.1467, over 28956.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3828, pruned_loss=0.1298, over 5671617.01 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3618, pruned_loss=0.1114, over 5720761.94 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3825, pruned_loss=0.1293, over 5661005.77 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:10:06,545 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=846401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:10:35,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-09 23:10:44,674 INFO [train.py:968] (1/2) Epoch 19, batch 23850, giga_loss[loss=0.3782, simple_loss=0.4191, pruned_loss=0.1686, over 27965.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3852, pruned_loss=0.1325, over 5657008.20 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3621, pruned_loss=0.1116, over 5713892.02 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3849, pruned_loss=0.1321, over 5654619.92 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:10:47,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8515, 3.6758, 3.4920, 1.6526], device='cuda:1'), covar=tensor([0.0745, 0.0904, 0.0828, 0.2245], device='cuda:1'), in_proj_covar=tensor([0.1180, 0.1101, 0.0935, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 23:10:51,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3152, 1.4598, 1.4734, 1.3159], device='cuda:1'), covar=tensor([0.1450, 0.1418, 0.1847, 0.1497], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0746, 0.0709, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 23:11:21,084 INFO [optim.py:369] (1/2) 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 23:11:34,724 INFO [train.py:968] (1/2) Epoch 19, batch 23900, libri_loss[loss=0.3601, simple_loss=0.3994, pruned_loss=0.1604, over 29532.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3865, pruned_loss=0.1349, over 5651011.84 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3623, pruned_loss=0.112, over 5718291.36 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3868, pruned_loss=0.1348, over 5643489.70 frames. ], batch size: 81, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:11:53,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5541, 1.7786, 1.6890, 1.4658], device='cuda:1'), covar=tensor([0.2323, 0.1959, 0.1693, 0.2108], device='cuda:1'), in_proj_covar=tensor([0.1930, 0.1854, 0.1796, 0.1928], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:12:10,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4354, 1.3043, 3.8803, 3.3167], device='cuda:1'), covar=tensor([0.1492, 0.2557, 0.0460, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0632, 0.0933, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:12:10,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1704, 0.7748, 0.8586, 1.3153], device='cuda:1'), covar=tensor([0.0673, 0.0350, 0.0324, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:1')
+2023-03-09 23:12:23,929 INFO [train.py:968] (1/2) Epoch 19, batch 23950, giga_loss[loss=0.4342, simple_loss=0.4557, pruned_loss=0.2063, over 27669.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3886, pruned_loss=0.1375, over 5647961.09 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3623, pruned_loss=0.1122, over 5719432.08 frames. ], giga_tot_loss[loss=0.3325, simple_loss=0.3894, pruned_loss=0.1377, over 5639517.62 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:12:29,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7410, 2.0222, 1.5227, 2.1091], device='cuda:1'), covar=tensor([0.2415, 0.2547, 0.2899, 0.2295], device='cuda:1'), in_proj_covar=tensor([0.1452, 0.1055, 0.1288, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 23:12:47,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4920, 1.7096, 1.5435, 1.4438], device='cuda:1'), covar=tensor([0.1916, 0.2036, 0.2237, 0.2001], device='cuda:1'), in_proj_covar=tensor([0.1925, 0.1851, 0.1796, 0.1926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:13:10,221 INFO [optim.py:369] (1/2) 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 23:13:12,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-09 23:13:23,134 INFO [train.py:968] (1/2) Epoch 19, batch 24000, giga_loss[loss=0.3469, simple_loss=0.4011, pruned_loss=0.1463, over 28784.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3919, pruned_loss=0.1405, over 5629980.37 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3622, pruned_loss=0.1123, over 5714599.44 frames. ], giga_tot_loss[loss=0.338, simple_loss=0.3934, pruned_loss=0.1413, over 5625916.41 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:13:23,135 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-09 23:13:32,001 INFO [train.py:1012] (1/2) Epoch 19, validation: loss=0.2059, simple_loss=0.313, pruned_loss=0.04942, over 944034.00 frames. 
+2023-03-09 23:13:32,002 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-09 23:13:40,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-09 23:14:10,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2204, 2.8696, 1.4136, 1.3385], device='cuda:1'), covar=tensor([0.0979, 0.0323, 0.0827, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0545, 0.0375, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0028], device='cuda:1')
+2023-03-09 23:14:25,682 INFO [train.py:968] (1/2) Epoch 19, batch 24050, giga_loss[loss=0.3015, simple_loss=0.3644, pruned_loss=0.1192, over 28645.00 frames. ], tot_loss[loss=0.3362, simple_loss=0.3909, pruned_loss=0.1408, over 5615511.52 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3621, pruned_loss=0.1123, over 5718892.95 frames. ], giga_tot_loss[loss=0.3387, simple_loss=0.3929, pruned_loss=0.1423, over 5605810.96 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:15:00,277 INFO [optim.py:369] (1/2) 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 23:15:12,446 INFO [train.py:968] (1/2) Epoch 19, batch 24100, giga_loss[loss=0.4649, simple_loss=0.4628, pruned_loss=0.2335, over 26600.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.389, pruned_loss=0.14, over 5622749.00 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3623, pruned_loss=0.1126, over 5714917.86 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3914, pruned_loss=0.1418, over 5615733.01 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:15:26,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4395, 1.7615, 1.0422, 1.3123], device='cuda:1'), covar=tensor([0.1219, 0.0764, 0.1585, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0448, 0.0514, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:15:39,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6597, 1.6620, 1.9173, 1.5056], device='cuda:1'), covar=tensor([0.1101, 0.1568, 0.0956, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0876, 0.0692, 0.0920, 0.0822], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:16:03,936 INFO [train.py:968] (1/2) Epoch 19, batch 24150, giga_loss[loss=0.3613, simple_loss=0.3934, pruned_loss=0.1646, over 23517.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3882, pruned_loss=0.1391, over 5625707.64 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3628, pruned_loss=0.113, over 5716809.12 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3902, pruned_loss=0.1408, over 5616898.06 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:16:06,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-09 23:16:37,083 INFO [optim.py:369] (1/2) 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 23:16:50,058 INFO [train.py:968] (1/2) Epoch 19, batch 24200, libri_loss[loss=0.3011, simple_loss=0.3685, pruned_loss=0.1169, over 29663.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3874, pruned_loss=0.1372, over 5616915.66 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3631, pruned_loss=0.1133, over 5709171.41 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3896, pruned_loss=0.1391, over 5613761.50 frames. ], batch size: 91, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:17:30,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2073, 4.0677, 3.8414, 1.7244], device='cuda:1'), covar=tensor([0.0607, 0.0747, 0.0705, 0.2334], device='cuda:1'), in_proj_covar=tensor([0.1185, 0.1105, 0.0938, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 23:17:46,510 INFO [train.py:968] (1/2) Epoch 19, batch 24250, giga_loss[loss=0.2737, simple_loss=0.3483, pruned_loss=0.09961, over 28865.00 frames. ], tot_loss[loss=0.3338, simple_loss=0.3902, pruned_loss=0.1387, over 5619388.84 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.363, pruned_loss=0.1131, over 5711076.80 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3922, pruned_loss=0.1406, over 5614525.88 frames. ], batch size: 66, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:18:17,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5404, 4.3732, 4.1571, 1.7342], device='cuda:1'), covar=tensor([0.0705, 0.0849, 0.0917, 0.2222], device='cuda:1'), in_proj_covar=tensor([0.1183, 0.1105, 0.0938, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 23:18:27,416 INFO [optim.py:369] (1/2) 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 23:18:38,275 INFO [train.py:968] (1/2) Epoch 19, batch 24300, giga_loss[loss=0.2737, simple_loss=0.3526, pruned_loss=0.09744, over 28795.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3893, pruned_loss=0.1377, over 5630306.65 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.363, pruned_loss=0.1132, over 5714841.06 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3914, pruned_loss=0.1397, over 5621352.87 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:19:31,416 INFO [train.py:968] (1/2) Epoch 19, batch 24350, giga_loss[loss=0.2687, simple_loss=0.3509, pruned_loss=0.09326, over 28803.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.387, pruned_loss=0.1348, over 5630460.99 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.363, pruned_loss=0.1132, over 5716986.84 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3889, pruned_loss=0.1365, over 5620582.89 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:20:09,661 INFO [optim.py:369] (1/2) 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 23:20:15,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7631, 1.9178, 2.0047, 1.5245], device='cuda:1'), covar=tensor([0.1876, 0.2553, 0.1504, 0.1840], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0694, 0.0921, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:20:21,929 INFO [train.py:968] (1/2) Epoch 19, batch 24400, giga_loss[loss=0.3138, simple_loss=0.3585, pruned_loss=0.1346, over 23727.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3836, pruned_loss=0.1317, over 5632574.59 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.363, pruned_loss=0.1134, over 5721717.01 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3857, pruned_loss=0.1333, over 5618329.00 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:20:51,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4061, 1.7352, 1.5466, 1.5331], device='cuda:1'), covar=tensor([0.1739, 0.1851, 0.2185, 0.1925], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0742, 0.0704, 0.0675], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 23:21:06,679 INFO [train.py:968] (1/2) Epoch 19, batch 24450, giga_loss[loss=0.3197, simple_loss=0.3767, pruned_loss=0.1313, over 28568.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3803, pruned_loss=0.1289, over 5623133.75 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3639, pruned_loss=0.1143, over 5703584.82 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3818, pruned_loss=0.13, over 5625323.28 frames. ], batch size: 78, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:21:22,499 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-09 23:21:45,324 INFO [optim.py:369] (1/2) 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 23:21:45,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1797, 1.1749, 3.5002, 3.0821], device='cuda:1'), covar=tensor([0.1625, 0.2743, 0.0475, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0637, 0.0939, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:21:52,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 23:21:54,632 INFO [train.py:968] (1/2) Epoch 19, batch 24500, giga_loss[loss=0.261, simple_loss=0.3404, pruned_loss=0.09073, over 29036.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3768, pruned_loss=0.1263, over 5634338.73 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3633, pruned_loss=0.1139, over 5708039.10 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3789, pruned_loss=0.1279, over 5630418.09 frames. ], batch size: 128, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:22:34,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4254, 1.5830, 1.4989, 1.3516], device='cuda:1'), covar=tensor([0.2704, 0.2303, 0.2109, 0.2316], device='cuda:1'), in_proj_covar=tensor([0.1929, 0.1850, 0.1792, 0.1924], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:22:44,858 INFO [train.py:968] (1/2) Epoch 19, batch 24550, giga_loss[loss=0.3256, simple_loss=0.3887, pruned_loss=0.1312, over 29082.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3771, pruned_loss=0.1268, over 5630256.75 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3632, pruned_loss=0.1138, over 5707392.72 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.379, pruned_loss=0.1283, over 5626267.69 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:22:53,782 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-09 23:23:28,097 INFO [optim.py:369] (1/2) 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 23:23:31,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-09 23:23:38,239 INFO [train.py:968] (1/2) Epoch 19, batch 24600, giga_loss[loss=0.3032, simple_loss=0.3693, pruned_loss=0.1185, over 28604.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3768, pruned_loss=0.1264, over 5635988.19 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3636, pruned_loss=0.1142, over 5709786.67 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3785, pruned_loss=0.1277, over 5628448.83 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:23:53,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0483, 2.2531, 2.3221, 1.7962], device='cuda:1'), covar=tensor([0.1753, 0.2059, 0.1282, 0.1627], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0693, 0.0922, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:24:26,645 INFO [train.py:968] (1/2) Epoch 19, batch 24650, giga_loss[loss=0.3194, simple_loss=0.3922, pruned_loss=0.1233, over 28788.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3737, pruned_loss=0.1233, over 5658158.37 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3627, pruned_loss=0.1138, over 5716549.84 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3762, pruned_loss=0.1251, over 5643803.18 frames. ], batch size: 243, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:25:04,856 INFO [optim.py:369] (1/2) 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 23:25:16,510 INFO [train.py:968] (1/2) Epoch 19, batch 24700, giga_loss[loss=0.3121, simple_loss=0.3836, pruned_loss=0.1203, over 28537.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3738, pruned_loss=0.1215, over 5664776.72 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3625, pruned_loss=0.1137, over 5723534.12 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3765, pruned_loss=0.1233, over 5644983.74 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:26:12,395 INFO [train.py:968] (1/2) Epoch 19, batch 24750, libri_loss[loss=0.3025, simple_loss=0.3538, pruned_loss=0.1256, over 29556.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3765, pruned_loss=0.1221, over 5668745.02 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3625, pruned_loss=0.114, over 5727362.08 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3787, pruned_loss=0.1235, over 5648641.76 frames. ], batch size: 77, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:26:31,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-09 23:26:47,962 INFO [optim.py:369] (1/2) 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 23:26:58,599 INFO [train.py:968] (1/2) Epoch 19, batch 24800, libri_loss[loss=0.3086, simple_loss=0.3739, pruned_loss=0.1217, over 29768.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3762, pruned_loss=0.1226, over 5673335.37 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3621, pruned_loss=0.1138, over 5732086.44 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3792, pruned_loss=0.1243, over 5650076.40 frames. ], batch size: 87, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:27:26,420 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=847420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:27:44,899 INFO [train.py:968] (1/2) Epoch 19, batch 24850, giga_loss[loss=0.3008, simple_loss=0.3684, pruned_loss=0.1166, over 28436.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3773, pruned_loss=0.1235, over 5689667.88 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3624, pruned_loss=0.114, over 5736896.76 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3798, pruned_loss=0.1248, over 5665306.18 frames. ], batch size: 60, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:28:01,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-09 23:28:06,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0749, 2.4673, 2.2419, 1.7687], device='cuda:1'), covar=tensor([0.2914, 0.2222, 0.2259, 0.2786], device='cuda:1'), in_proj_covar=tensor([0.1926, 0.1846, 0.1785, 0.1924], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:28:18,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 23:28:25,803 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 19, batch 24900, giga_loss[loss=0.2731, simple_loss=0.34, pruned_loss=0.1031, over 28646.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3756, pruned_loss=0.123, over 5686161.55 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3627, pruned_loss=0.1142, over 5728850.97 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3775, pruned_loss=0.1239, over 5673326.68 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:28:38,671 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=847488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:29:18,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-09 23:29:23,107 INFO [train.py:968] (1/2) Epoch 19, batch 24950, giga_loss[loss=0.3116, simple_loss=0.3921, pruned_loss=0.1156, over 28866.00 frames. ], tot_loss[loss=0.31, simple_loss=0.374, pruned_loss=0.123, over 5683163.42 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3627, pruned_loss=0.1143, over 5729028.77 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3757, pruned_loss=0.124, over 5671893.56 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:30:01,432 INFO [optim.py:369] (1/2) 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 23:30:09,933 INFO [train.py:968] (1/2) Epoch 19, batch 25000, giga_loss[loss=0.2805, simple_loss=0.3618, pruned_loss=0.09955, over 28603.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3739, pruned_loss=0.1223, over 5681776.10 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.363, pruned_loss=0.1145, over 5731652.23 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3751, pruned_loss=0.123, over 5669968.27 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:30:20,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0421, 1.2958, 1.0743, 0.3125], device='cuda:1'), covar=tensor([0.3234, 0.2675, 0.3544, 0.5624], device='cuda:1'), in_proj_covar=tensor([0.1705, 0.1604, 0.1576, 0.1392], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 23:30:55,501 INFO [train.py:968] (1/2) Epoch 19, batch 25050, giga_loss[loss=0.2859, simple_loss=0.3692, pruned_loss=0.1013, over 28960.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.372, pruned_loss=0.1193, over 5695249.40 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3631, pruned_loss=0.1146, over 5737918.62 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3732, pruned_loss=0.12, over 5678538.10 frames. ], batch size: 106, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:31:38,002 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 25100, giga_loss[loss=0.2985, simple_loss=0.3705, pruned_loss=0.1133, over 29137.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3721, pruned_loss=0.1193, over 5679412.22 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3631, pruned_loss=0.1146, over 5729426.04 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3731, pruned_loss=0.1198, over 5672786.20 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:31:47,118 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-09 23:32:35,497 INFO [train.py:968] (1/2) Epoch 19, batch 25150, giga_loss[loss=0.317, simple_loss=0.3748, pruned_loss=0.1296, over 28874.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3713, pruned_loss=0.1193, over 5683840.56 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3633, pruned_loss=0.1148, over 5731794.37 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3722, pruned_loss=0.1197, over 5674848.98 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:33:15,185 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 25200, giga_loss[loss=0.3196, simple_loss=0.3669, pruned_loss=0.1362, over 23781.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3705, pruned_loss=0.1193, over 5678715.09 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3632, pruned_loss=0.1148, over 5735233.47 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3714, pruned_loss=0.1197, over 5667231.12 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:33:28,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=847792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:33:31,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=847795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:34:10,108 INFO [train.py:968] (1/2) Epoch 19, batch 25250, giga_loss[loss=0.3763, simple_loss=0.4074, pruned_loss=0.1726, over 26738.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3688, pruned_loss=0.1189, over 5675130.08 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.363, pruned_loss=0.1147, over 5739304.10 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.37, pruned_loss=0.1194, over 5660773.90 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:34:21,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=847850.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 23:34:31,948 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:34:53,677 INFO [train.py:968] (1/2) Epoch 19, batch 25300, giga_loss[loss=0.293, simple_loss=0.37, pruned_loss=0.1081, over 29044.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3696, pruned_loss=0.1201, over 5685893.26 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3636, pruned_loss=0.1153, over 5744924.16 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3702, pruned_loss=0.1203, over 5666711.52 frames. ], batch size: 155, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:35:23,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6501, 1.9717, 1.9987, 1.5184], device='cuda:1'), covar=tensor([0.3547, 0.2467, 0.2539, 0.3028], device='cuda:1'), in_proj_covar=tensor([0.1935, 0.1856, 0.1797, 0.1937], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:35:40,884 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=847935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:35:43,030 INFO [train.py:968] (1/2) Epoch 19, batch 25350, giga_loss[loss=0.3076, simple_loss=0.3604, pruned_loss=0.1274, over 28217.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3673, pruned_loss=0.1192, over 5679128.96 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3631, pruned_loss=0.115, over 5745356.45 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3684, pruned_loss=0.1197, over 5661437.36 frames. ], batch size: 77, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:35:43,889 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:35:43,903 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=847938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:35:47,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:09,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=847967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:11,945 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=847970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:20,218 INFO [optim.py:369] (1/2) 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 23:36:26,222 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-09 23:36:26,814 INFO [train.py:968] (1/2) Epoch 19, batch 25400, giga_loss[loss=0.3001, simple_loss=0.3656, pruned_loss=0.1173, over 28582.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3671, pruned_loss=0.1193, over 5679647.85 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3634, pruned_loss=0.1153, over 5739362.49 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3677, pruned_loss=0.1194, over 5669704.16 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:36:34,978 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=847993.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:36:38,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847996.0, num_to_drop=1, layers_to_drop={0}
+2023-03-09 23:36:48,735 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=848006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:36:50,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=848009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:37:07,898 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=848025.0, num_to_drop=1, layers_to_drop={1}
+2023-03-09 23:37:10,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3537, 1.5106, 1.5479, 1.1808], device='cuda:1'), covar=tensor([0.1606, 0.2459, 0.1326, 0.1559], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0698, 0.0926, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:37:19,829 INFO [train.py:968] (1/2) Epoch 19, batch 25450, giga_loss[loss=0.311, simple_loss=0.3733, pruned_loss=0.1244, over 28898.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1206, over 5668202.62 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3635, pruned_loss=0.1155, over 5740202.68 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3686, pruned_loss=0.1207, over 5659428.51 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:37:20,011 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=848038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:38:00,407 INFO [optim.py:369] (1/2) 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 23:38:09,331 INFO [train.py:968] (1/2) Epoch 19, batch 25500, giga_loss[loss=0.3676, simple_loss=0.4039, pruned_loss=0.1656, over 26570.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3687, pruned_loss=0.1199, over 5658873.58 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3636, pruned_loss=0.1155, over 5730707.11 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3691, pruned_loss=0.12, over 5659605.29 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:38:34,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-09 23:38:51,227 INFO [train.py:968] (1/2) Epoch 19, batch 25550, giga_loss[loss=0.2771, simple_loss=0.3557, pruned_loss=0.09927, over 28813.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3685, pruned_loss=0.1193, over 5662867.31 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.363, pruned_loss=0.1153, over 5727247.69 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3696, pruned_loss=0.1197, over 5663230.75 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:38:57,590 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-09 23:39:07,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3412, 1.6075, 1.6118, 1.3270], device='cuda:1'), covar=tensor([0.1793, 0.1849, 0.2126, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0742, 0.0702, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 23:39:30,927 INFO [optim.py:369] (1/2) 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 23:39:40,240 INFO [train.py:968] (1/2) Epoch 19, batch 25600, giga_loss[loss=0.3287, simple_loss=0.3819, pruned_loss=0.1377, over 28065.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3686, pruned_loss=0.1192, over 5646928.21 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3631, pruned_loss=0.1153, over 5719108.37 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3694, pruned_loss=0.1196, over 5654026.39 frames. ], batch size: 412, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:40:22,981 INFO [train.py:968] (1/2) Epoch 19, batch 25650, giga_loss[loss=0.3082, simple_loss=0.3746, pruned_loss=0.1209, over 28683.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3682, pruned_loss=0.1191, over 5656306.40 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3627, pruned_loss=0.1148, over 5723518.18 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.1201, over 5655276.25 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:40:36,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-09 23:40:38,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7989, 1.9631, 1.5884, 2.2074], device='cuda:1'), covar=tensor([0.2642, 0.2761, 0.3042, 0.2293], device='cuda:1'), in_proj_covar=tensor([0.1460, 0.1061, 0.1293, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 23:41:02,116 INFO [optim.py:369] (1/2) 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 23:41:04,903 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-09 23:41:09,590 INFO [train.py:968] (1/2) Epoch 19, batch 25700, giga_loss[loss=0.406, simple_loss=0.4216, pruned_loss=0.1952, over 23580.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.371, pruned_loss=0.1222, over 5645810.71 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1145, over 5721243.20 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1234, over 5643901.14 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:41:57,641 INFO [train.py:968] (1/2) Epoch 19, batch 25750, giga_loss[loss=0.3121, simple_loss=0.3782, pruned_loss=0.123, over 28898.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3708, pruned_loss=0.1227, over 5663625.04 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.362, pruned_loss=0.1145, over 5726839.78 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5654971.43 frames. ], batch size: 213, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:42:32,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3309, 3.1531, 3.0090, 1.3332], device='cuda:1'), covar=tensor([0.0963, 0.1075, 0.1016, 0.2367], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.1118, 0.0951, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-09 23:42:40,382 INFO [optim.py:369] (1/2) 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 23:42:49,055 INFO [train.py:968] (1/2) Epoch 19, batch 25800, giga_loss[loss=0.4256, simple_loss=0.4539, pruned_loss=0.1987, over 23944.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1245, over 5656708.28 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3623, pruned_loss=0.1146, over 5725924.50 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3735, pruned_loss=0.1256, over 5649181.78 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:43:26,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6169, 2.3515, 1.6477, 0.8869], device='cuda:1'), covar=tensor([0.4493, 0.2588, 0.3349, 0.4491], device='cuda:1'), in_proj_covar=tensor([0.1716, 0.1621, 0.1582, 0.1398], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 23:43:33,769 INFO [train.py:968] (1/2) Epoch 19, batch 25850, giga_loss[loss=0.2885, simple_loss=0.3532, pruned_loss=0.1119, over 28488.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3727, pruned_loss=0.1251, over 5652459.63 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3623, pruned_loss=0.1146, over 5719050.50 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.374, pruned_loss=0.1262, over 5650302.93 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:44:12,616 INFO [optim.py:369] (1/2) 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 23:44:19,728 INFO [train.py:968] (1/2) Epoch 19, batch 25900, giga_loss[loss=0.3084, simple_loss=0.369, pruned_loss=0.1239, over 28811.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3721, pruned_loss=0.1249, over 5658871.72 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5722403.88 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.373, pruned_loss=0.1255, over 5652393.52 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:44:48,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5022, 1.8630, 1.5184, 1.6254], device='cuda:1'), covar=tensor([0.0735, 0.0272, 0.0306, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-09 23:44:54,898 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:44:56,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1237, 1.0794, 3.6945, 3.2035], device='cuda:1'), covar=tensor([0.1773, 0.2826, 0.0542, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0642, 0.0945, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-09 23:45:05,498 INFO [train.py:968] (1/2) Epoch 19, batch 25950, giga_loss[loss=0.2834, simple_loss=0.3586, pruned_loss=0.1041, over 28922.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3726, pruned_loss=0.1237, over 5666714.25 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3631, pruned_loss=0.1152, over 5725023.47 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3731, pruned_loss=0.1243, over 5658405.45 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:45:43,907 INFO [optim.py:369] (1/2) 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 23:45:51,861 INFO [train.py:968] (1/2) Epoch 19, batch 26000, giga_loss[loss=0.2729, simple_loss=0.3432, pruned_loss=0.1013, over 28529.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3698, pruned_loss=0.1213, over 5665775.66 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5728770.85 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3707, pruned_loss=0.1221, over 5654222.99 frames. ], batch size: 336, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:46:06,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:46:09,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0738, 3.1880, 2.0789, 1.0196], device='cuda:1'), covar=tensor([0.7360, 0.3073, 0.3702, 0.7143], device='cuda:1'), in_proj_covar=tensor([0.1709, 0.1623, 0.1579, 0.1398], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-09 23:46:40,973 INFO [train.py:968] (1/2) Epoch 19, batch 26050, giga_loss[loss=0.2766, simple_loss=0.3505, pruned_loss=0.1013, over 28776.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3677, pruned_loss=0.1201, over 5667914.60 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5729855.99 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3685, pruned_loss=0.1207, over 5657731.86 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:47:17,410 INFO [optim.py:369] (1/2) 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 23:47:24,648 INFO [train.py:968] (1/2) Epoch 19, batch 26100, giga_loss[loss=0.3111, simple_loss=0.371, pruned_loss=0.1256, over 28704.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3669, pruned_loss=0.1204, over 5668876.57 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3633, pruned_loss=0.1155, over 5723043.57 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3672, pruned_loss=0.1207, over 5665469.77 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:47:53,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:48:13,906 INFO [train.py:968] (1/2) Epoch 19, batch 26150, giga_loss[loss=0.2758, simple_loss=0.3397, pruned_loss=0.106, over 28336.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3665, pruned_loss=0.1202, over 5663507.86 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5713064.56 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3666, pruned_loss=0.1204, over 5666939.48 frames. ], batch size: 65, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:48:51,377 INFO [optim.py:369] (1/2) 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 23:48:57,389 INFO [train.py:968] (1/2) Epoch 19, batch 26200, giga_loss[loss=0.3363, simple_loss=0.3964, pruned_loss=0.1381, over 28921.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3685, pruned_loss=0.1207, over 5666079.70 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3632, pruned_loss=0.1155, over 5707411.61 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.369, pruned_loss=0.1212, over 5672318.50 frames. ], batch size: 186, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:49:00,375 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-09 23:49:10,736 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:49:48,105 INFO [train.py:968] (1/2) Epoch 19, batch 26250, giga_loss[loss=0.2758, simple_loss=0.3608, pruned_loss=0.09538, over 28704.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3722, pruned_loss=0.12, over 5669384.14 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1156, over 5703103.34 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3725, pruned_loss=0.1203, over 5677104.47 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:50:01,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-09 23:50:16,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5480, 1.5097, 1.7576, 1.3643], device='cuda:1'), covar=tensor([0.1401, 0.2316, 0.1227, 0.1571], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0696, 0.0925, 0.0824], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:50:27,045 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:50:29,841 INFO [optim.py:369] (1/2) 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 23:50:36,185 INFO [train.py:968] (1/2) Epoch 19, batch 26300, giga_loss[loss=0.3819, simple_loss=0.4098, pruned_loss=0.177, over 23764.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.374, pruned_loss=0.1205, over 5665121.30 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3637, pruned_loss=0.1159, over 5697982.70 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3742, pruned_loss=0.1207, over 5674377.31 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:50:46,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=848901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:51:02,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4458, 1.5670, 1.2487, 1.1018], device='cuda:1'), covar=tensor([0.1048, 0.0646, 0.1146, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0446, 0.0511, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:51:21,840 INFO [train.py:968] (1/2) Epoch 19, batch 26350, giga_loss[loss=0.4023, simple_loss=0.4324, pruned_loss=0.1861, over 26649.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3751, pruned_loss=0.1221, over 5664962.78 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1158, over 5692664.31 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3762, pruned_loss=0.1225, over 5676742.21 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 2.0
+2023-03-09 23:52:00,732 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=848978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:52:01,369 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=848979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:52:03,778 INFO [optim.py:369] (1/2) 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 23:52:08,511 INFO [train.py:968] (1/2) Epoch 19, batch 26400, giga_loss[loss=0.3049, simple_loss=0.371, pruned_loss=0.1194, over 28622.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.377, pruned_loss=0.1238, over 5674756.04 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5693831.98 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3781, pruned_loss=0.1243, over 5682827.05 frames. ], batch size: 92, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:52:45,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4235, 1.5294, 1.6666, 1.2335], device='cuda:1'), covar=tensor([0.1572, 0.2442, 0.1333, 0.1579], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0699, 0.0927, 0.0825], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-09 23:53:00,744 INFO [train.py:968] (1/2) Epoch 19, batch 26450, giga_loss[loss=0.3534, simple_loss=0.4087, pruned_loss=0.1491, over 28662.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3778, pruned_loss=0.1257, over 5662448.95 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3628, pruned_loss=0.1154, over 5686962.95 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3793, pruned_loss=0.1266, over 5674022.93 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:53:06,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:53:11,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:53:39,463 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:53:43,823 INFO [optim.py:369] (1/2) 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 23:53:47,886 INFO [train.py:968] (1/2) Epoch 19, batch 26500, giga_loss[loss=0.3145, simple_loss=0.3728, pruned_loss=0.1281, over 28915.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3763, pruned_loss=0.125, over 5675418.11 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3629, pruned_loss=0.1155, over 5688216.34 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3779, pruned_loss=0.126, over 5682881.48 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:53:50,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:54:17,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:54:19,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:54:32,196 INFO [train.py:968] (1/2) Epoch 19, batch 26550, giga_loss[loss=0.3112, simple_loss=0.366, pruned_loss=0.1282, over 28388.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1247, over 5667176.49 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3634, pruned_loss=0.1159, over 5680711.78 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3761, pruned_loss=0.1255, over 5679213.06 frames. ], batch size: 71, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:54:47,523 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:55:12,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:55:16,723 INFO [optim.py:369] (1/2) 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 23:55:17,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-09 23:55:20,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1636, 1.2456, 1.1511, 0.8808], device='cuda:1'), covar=tensor([0.0998, 0.0547, 0.1093, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0449, 0.0514, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-09 23:55:21,691 INFO [train.py:968] (1/2) Epoch 19, batch 26600, giga_loss[loss=0.3505, simple_loss=0.4015, pruned_loss=0.1497, over 28847.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5672487.24 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.1159, over 5676361.02 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3752, pruned_loss=0.1253, over 5685564.62 frames. ], batch size: 227, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:56:01,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6559, 1.7455, 1.7721, 1.5581], device='cuda:1'), covar=tensor([0.2581, 0.2429, 0.2007, 0.2377], device='cuda:1'), in_proj_covar=tensor([0.1923, 0.1848, 0.1795, 0.1932], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-09 23:56:04,853 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:07,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:07,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3517, 1.7020, 1.3414, 1.3911], device='cuda:1'), covar=tensor([0.2439, 0.2366, 0.2701, 0.2021], device='cuda:1'), in_proj_covar=tensor([0.1461, 0.1062, 0.1295, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 23:56:07,597 INFO [train.py:968] (1/2) Epoch 19, batch 26650, libri_loss[loss=0.3375, simple_loss=0.3905, pruned_loss=0.1422, over 25784.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3755, pruned_loss=0.126, over 5668013.47 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3639, pruned_loss=0.1162, over 5680420.41 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3764, pruned_loss=0.1266, over 5674723.58 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:56:22,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:22,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-09 23:56:32,802 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:56:44,984 INFO [optim.py:369] (1/2) 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 23:56:50,330 INFO [train.py:968] (1/2) Epoch 19, batch 26700, giga_loss[loss=0.3456, simple_loss=0.3996, pruned_loss=0.1458, over 28765.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3756, pruned_loss=0.1268, over 5666274.68 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3632, pruned_loss=0.1159, over 5674305.64 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3772, pruned_loss=0.1278, over 5676301.10 frames. ], batch size: 262, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:57:17,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:19,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:31,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=849336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:34,544 INFO [train.py:968] (1/2) Epoch 19, batch 26750, giga_loss[loss=0.332, simple_loss=0.3699, pruned_loss=0.147, over 23697.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3727, pruned_loss=0.1256, over 5653301.83 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3631, pruned_loss=0.1158, over 5681139.72 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3746, pruned_loss=0.127, over 5655213.72 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:57:47,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:57:50,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:58:17,589 INFO [optim.py:369] (1/2) 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 23:58:24,693 INFO [train.py:968] (1/2) Epoch 19, batch 26800, giga_loss[loss=0.2877, simple_loss=0.3618, pruned_loss=0.1069, over 28651.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3724, pruned_loss=0.1251, over 5647860.81 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3637, pruned_loss=0.1161, over 5674802.00 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3734, pruned_loss=0.126, over 5655337.80 frames. ], batch size: 242, lr: 1.68e-03, grad_scale: 8.0
+2023-03-09 23:58:33,031 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:58:37,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:59:00,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4199, 1.7319, 1.3090, 1.5244], device='cuda:1'), covar=tensor([0.2632, 0.2680, 0.3068, 0.2329], device='cuda:1'), in_proj_covar=tensor([0.1462, 0.1064, 0.1298, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-09 23:59:02,773 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-09 23:59:10,921 INFO [train.py:968] (1/2) Epoch 19, batch 26850, giga_loss[loss=0.3497, simple_loss=0.3806, pruned_loss=0.1594, over 23578.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3735, pruned_loss=0.1247, over 5655477.11 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3641, pruned_loss=0.1164, over 5679412.85 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3741, pruned_loss=0.1254, over 5656728.52 frames. ], batch size: 705, lr: 1.68e-03, grad_scale: 4.0
+2023-03-09 23:59:32,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7813, 2.8325, 2.7010, 2.6378], device='cuda:1'), covar=tensor([0.1556, 0.1745, 0.1579, 0.1568], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0741, 0.0705, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-09 23:59:57,467 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 26900, giga_loss[loss=0.2575, simple_loss=0.3374, pruned_loss=0.08879, over 28854.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3765, pruned_loss=0.1271, over 5638413.20 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3641, pruned_loss=0.1163, over 5665022.92 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3772, pruned_loss=0.1278, over 5651746.94 frames. ], batch size: 174, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:00:09,946 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:00:12,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:00:41,980 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:00:48,510 INFO [train.py:968] (1/2) Epoch 19, batch 26950, giga_loss[loss=0.3085, simple_loss=0.3901, pruned_loss=0.1135, over 28993.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3761, pruned_loss=0.1268, over 5654617.12 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1163, over 5669513.04 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3769, pruned_loss=0.1277, over 5660532.49 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:01:28,720 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 27000, giga_loss[loss=0.3983, simple_loss=0.4281, pruned_loss=0.1842, over 26757.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3762, pruned_loss=0.1241, over 5662936.05 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3636, pruned_loss=0.116, over 5674421.75 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3776, pruned_loss=0.1253, over 5663162.27 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:01:34,802 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 00:01:43,567 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 00:01:45,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3711, 1.7126, 1.6098, 1.3371], device='cuda:1'), covar=tensor([0.3588, 0.2731, 0.2318, 0.2796], device='cuda:1'), in_proj_covar=tensor([0.1919, 0.1845, 0.1791, 0.1926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 00:02:12,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-10 00:02:28,961 INFO [train.py:968] (1/2) Epoch 19, batch 27050, giga_loss[loss=0.3192, simple_loss=0.3887, pruned_loss=0.1249, over 28711.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3776, pruned_loss=0.1227, over 5679207.09 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3636, pruned_loss=0.1158, over 5681614.79 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3792, pruned_loss=0.1241, over 5672926.26 frames. ], batch size: 284, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:02:54,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6494, 4.3670, 1.6558, 1.7688], device='cuda:1'), covar=tensor([0.0918, 0.0428, 0.0940, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0549, 0.0376, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 00:03:08,690 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 27100, giga_loss[loss=0.2925, simple_loss=0.3639, pruned_loss=0.1106, over 29032.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3804, pruned_loss=0.1241, over 5690281.49 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3634, pruned_loss=0.1158, over 5682741.88 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3822, pruned_loss=0.1253, over 5684163.50 frames. ], batch size: 164, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:03:19,176 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-10 00:03:35,366 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=849711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:04:02,789 INFO [train.py:968] (1/2) Epoch 19, batch 27150, giga_loss[loss=0.4378, simple_loss=0.467, pruned_loss=0.2043, over 28624.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3835, pruned_loss=0.1272, over 5684518.36 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3639, pruned_loss=0.1161, over 5684295.40 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3848, pruned_loss=0.1282, over 5678270.90 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:04:50,567 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 27200, libri_loss[loss=0.3505, simple_loss=0.4009, pruned_loss=0.1501, over 29543.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3854, pruned_loss=0.1305, over 5658706.07 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1162, over 5686242.30 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3864, pruned_loss=0.1312, over 5651835.56 frames. ], batch size: 82, lr: 1.68e-03, grad_scale: 8.0
+2023-03-10 00:05:50,021 INFO [train.py:968] (1/2) Epoch 19, batch 27250, giga_loss[loss=0.2856, simple_loss=0.3509, pruned_loss=0.1102, over 28803.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3828, pruned_loss=0.1291, over 5666436.77 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3638, pruned_loss=0.1159, over 5689636.49 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3842, pruned_loss=0.1301, over 5657525.50 frames. ], batch size: 99, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:06:05,206 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=849854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:10,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=849857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:33,788 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=849884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:37,037 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=849886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:06:37,965 INFO [train.py:968] (1/2) Epoch 19, batch 27300, giga_loss[loss=0.2867, simple_loss=0.3673, pruned_loss=0.103, over 28622.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.382, pruned_loss=0.1284, over 5651518.02 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3639, pruned_loss=0.116, over 5690123.31 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3833, pruned_loss=0.1293, over 5643550.05 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:06:42,522 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 00:07:22,568 INFO [train.py:968] (1/2) Epoch 19, batch 27350, giga_loss[loss=0.2779, simple_loss=0.363, pruned_loss=0.09641, over 29007.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3802, pruned_loss=0.1252, over 5664281.34 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3637, pruned_loss=0.116, over 5692188.29 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3818, pruned_loss=0.1263, over 5655384.57 frames. ], batch size: 136, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:07:39,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 00:08:05,616 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 27400, giga_loss[loss=0.275, simple_loss=0.3566, pruned_loss=0.09674, over 28574.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3794, pruned_loss=0.1234, over 5678118.73 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.116, over 5697717.67 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3815, pruned_loss=0.1246, over 5665006.14 frames. ], batch size: 85, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:09:01,480 INFO [train.py:968] (1/2) Epoch 19, batch 27450, giga_loss[loss=0.3029, simple_loss=0.3687, pruned_loss=0.1185, over 28276.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3803, pruned_loss=0.1246, over 5674310.20 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5703082.58 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3828, pruned_loss=0.126, over 5658666.88 frames. ], batch size: 368, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:09:16,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5439, 2.1280, 1.5232, 0.7031], device='cuda:1'), covar=tensor([0.5964, 0.2873, 0.3896, 0.6390], device='cuda:1'), in_proj_covar=tensor([0.1723, 0.1629, 0.1593, 0.1402], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 00:09:44,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6575, 1.8255, 1.7404, 1.5470], device='cuda:1'), covar=tensor([0.1875, 0.2081, 0.2281, 0.2183], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0747, 0.0708, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 00:09:44,952 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 27500, giga_loss[loss=0.43, simple_loss=0.4477, pruned_loss=0.2062, over 26540.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3801, pruned_loss=0.125, over 5678792.78 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.363, pruned_loss=0.1156, over 5703756.13 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3823, pruned_loss=0.1263, over 5665454.33 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:10:40,187 INFO [train.py:968] (1/2) Epoch 19, batch 27550, giga_loss[loss=0.268, simple_loss=0.3389, pruned_loss=0.09857, over 28843.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3768, pruned_loss=0.1241, over 5666829.77 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.363, pruned_loss=0.1155, over 5707642.03 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3789, pruned_loss=0.1253, over 5652253.32 frames. ], batch size: 119, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:10:48,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:11:21,157 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 00:11:23,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8563, 1.8541, 1.8296, 1.6871], device='cuda:1'), covar=tensor([0.1718, 0.2284, 0.2251, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0750, 0.0710, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 00:11:24,586 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 27600, libri_loss[loss=0.2547, simple_loss=0.3207, pruned_loss=0.09432, over 29365.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3763, pruned_loss=0.1253, over 5635403.66 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3638, pruned_loss=0.1164, over 5684486.74 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3777, pruned_loss=0.1258, over 5642736.92 frames. ], batch size: 67, lr: 1.68e-03, grad_scale: 8.0
+2023-03-10 00:11:42,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4150, 1.5978, 1.5325, 1.6579], device='cuda:1'), covar=tensor([0.0625, 0.0284, 0.0268, 0.0642], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:1')
+2023-03-10 00:12:19,659 INFO [train.py:968] (1/2) Epoch 19, batch 27650, giga_loss[loss=0.3626, simple_loss=0.4089, pruned_loss=0.1582, over 27603.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3732, pruned_loss=0.1232, over 5647292.20 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3631, pruned_loss=0.1159, over 5689951.62 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3752, pruned_loss=0.1243, over 5647334.83 frames. ], batch size: 472, lr: 1.68e-03, grad_scale: 8.0
+2023-03-10 00:12:39,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:12:46,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7702, 1.8824, 1.9122, 1.6058], device='cuda:1'), covar=tensor([0.1767, 0.2147, 0.2136, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0749, 0.0710, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 00:13:03,621 INFO [optim.py:369] (1/2) 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 00:13:07,194 INFO [train.py:968] (1/2) Epoch 19, batch 27700, giga_loss[loss=0.326, simple_loss=0.3933, pruned_loss=0.1293, over 28576.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1244, over 5646497.21 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3633, pruned_loss=0.1159, over 5694162.13 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3747, pruned_loss=0.1254, over 5641818.23 frames. ], batch size: 307, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:13:17,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4536, 1.5840, 1.2837, 1.1065], device='cuda:1'), covar=tensor([0.0912, 0.0550, 0.0987, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0443, 0.0511, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 00:13:46,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4973, 1.6283, 1.3029, 1.2076], device='cuda:1'), covar=tensor([0.0948, 0.0597, 0.1079, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0444, 0.0512, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 00:13:52,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4416, 1.7465, 1.7362, 1.2878], device='cuda:1'), covar=tensor([0.1786, 0.2387, 0.1446, 0.1642], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0705, 0.0933, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0013], device='cuda:1')
+2023-03-10 00:13:53,585 INFO [train.py:968] (1/2) Epoch 19, batch 27750, giga_loss[loss=0.3908, simple_loss=0.4222, pruned_loss=0.1797, over 26637.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3717, pruned_loss=0.1236, over 5650416.16 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.363, pruned_loss=0.1157, over 5697165.85 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3734, pruned_loss=0.1247, over 5643397.19 frames. ], batch size: 555, lr: 1.68e-03, grad_scale: 4.0
+2023-03-10 00:14:36,966 INFO [optim.py:369] (1/2) 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 00:14:39,892 INFO [train.py:968] (1/2) Epoch 19, batch 27800, giga_loss[loss=0.2911, simple_loss=0.378, pruned_loss=0.1021, over 28693.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3694, pruned_loss=0.1204, over 5649427.23 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3635, pruned_loss=0.1162, over 5688034.27 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3704, pruned_loss=0.121, over 5651299.94 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:14:50,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:14:53,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=850405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:15:03,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850415.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 00:15:19,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=850434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:15:22,623 INFO [train.py:968] (1/2) Epoch 19, batch 27850, libri_loss[loss=0.3525, simple_loss=0.4015, pruned_loss=0.1517, over 19760.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3669, pruned_loss=0.1173, over 5644255.68 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3641, pruned_loss=0.1165, over 5674173.23 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3673, pruned_loss=0.1175, over 5657725.80 frames. ], batch size: 187, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:15:28,528 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-10 00:15:57,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5857, 1.6294, 1.8739, 1.4085], device='cuda:1'), covar=tensor([0.1519, 0.2081, 0.1219, 0.1615], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0703, 0.0931, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 00:16:06,483 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:16:11,859 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 27900, giga_loss[loss=0.3074, simple_loss=0.3776, pruned_loss=0.1186, over 28833.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1176, over 5642489.78 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3648, pruned_loss=0.1171, over 5678534.16 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3666, pruned_loss=0.1172, over 5648543.54 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 00:16:46,494 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 27950, giga_loss[loss=0.2816, simple_loss=0.3555, pruned_loss=0.1038, over 28664.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3636, pruned_loss=0.1159, over 5651788.53 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3646, pruned_loss=0.117, over 5681516.64 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5653385.27 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 00:17:16,175 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-10 00:17:35,322 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:17:58,921 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 28000, giga_loss[loss=0.3549, simple_loss=0.3988, pruned_loss=0.1555, over 27511.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3631, pruned_loss=0.1163, over 5653853.81 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3648, pruned_loss=0.1171, over 5685752.44 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3628, pruned_loss=0.116, over 5650976.06 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:18:46,511 INFO [train.py:968] (1/2) Epoch 19, batch 28050, giga_loss[loss=0.3199, simple_loss=0.3635, pruned_loss=0.1382, over 23573.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3657, pruned_loss=0.1173, over 5669738.08 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3655, pruned_loss=0.1177, over 5691176.03 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3648, pruned_loss=0.1165, over 5661936.14 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:19:12,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:19:15,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=850666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:19:21,371 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 00:19:32,167 INFO [optim.py:369] (1/2) 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 00:19:34,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-10 00:19:34,526 INFO [train.py:968] (1/2) Epoch 19, batch 28100, giga_loss[loss=0.3848, simple_loss=0.4146, pruned_loss=0.1775, over 23467.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3682, pruned_loss=0.1189, over 5640880.57 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3662, pruned_loss=0.1182, over 5672806.57 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3669, pruned_loss=0.1178, over 5650790.75 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:19:41,313 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=850695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:20:04,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:20:04,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3765, 4.2197, 4.0359, 1.9208], device='cuda:1'), covar=tensor([0.0524, 0.0647, 0.0607, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.1199, 0.1116, 0.0952, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 00:20:20,890 INFO [train.py:968] (1/2) Epoch 19, batch 28150, giga_loss[loss=0.2939, simple_loss=0.3616, pruned_loss=0.1131, over 28902.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1196, over 5635056.17 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3665, pruned_loss=0.1183, over 5666394.41 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1187, over 5647756.31 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:20:32,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:20:43,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 00:21:04,144 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:21:07,082 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 28200, giga_loss[loss=0.3419, simple_loss=0.3871, pruned_loss=0.1483, over 26599.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3695, pruned_loss=0.1202, over 5629612.76 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3667, pruned_loss=0.1183, over 5663213.41 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3684, pruned_loss=0.1194, over 5642316.89 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:21:10,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850790.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 00:21:53,570 INFO [train.py:968] (1/2) Epoch 19, batch 28250, giga_loss[loss=0.3092, simple_loss=0.3744, pruned_loss=0.122, over 28874.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3705, pruned_loss=0.1206, over 5655845.01 frames. ], libri_tot_loss[loss=0.3018, simple_loss=0.3669, pruned_loss=0.1184, over 5668204.22 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3695, pruned_loss=0.1201, over 5661042.39 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:22:07,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:22:37,176 INFO [optim.py:369] (1/2) 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 00:22:38,520 INFO [train.py:968] (1/2) Epoch 19, batch 28300, giga_loss[loss=0.3168, simple_loss=0.3872, pruned_loss=0.1232, over 28456.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3712, pruned_loss=0.1209, over 5655188.58 frames. ], libri_tot_loss[loss=0.3011, simple_loss=0.3663, pruned_loss=0.1179, over 5673933.33 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.371, pruned_loss=0.1208, over 5654149.24 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:23:22,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850933.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 00:23:25,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=850936.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 00:23:28,785 INFO [train.py:968] (1/2) Epoch 19, batch 28350, giga_loss[loss=0.2808, simple_loss=0.3557, pruned_loss=0.1029, over 28764.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.372, pruned_loss=0.1213, over 5657833.08 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3667, pruned_loss=0.1183, over 5676526.48 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5654278.55 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 00:23:30,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=850939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:23:58,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=850965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 00:23:59,390 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=850966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:24:18,243 INFO [optim.py:369] (1/2) 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 00:24:18,890 INFO [train.py:968] (1/2) Epoch 19, batch 28400, giga_loss[loss=0.3064, simple_loss=0.3653, pruned_loss=0.1238, over 28946.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3725, pruned_loss=0.1224, over 5644366.90 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3667, pruned_loss=0.1182, over 5670642.66 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3723, pruned_loss=0.1223, over 5646337.10 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:24:29,643 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=850998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:24:34,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:25:04,853 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 28450, giga_loss[loss=0.3079, simple_loss=0.3755, pruned_loss=0.1201, over 28703.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3735, pruned_loss=0.1221, over 5650659.37 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3666, pruned_loss=0.1182, over 5673646.46 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3735, pruned_loss=0.1222, over 5649027.83 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:26:00,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:03,300 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:26:05,029 INFO [train.py:968] (1/2) Epoch 19, batch 28500, giga_loss[loss=0.2967, simple_loss=0.3714, pruned_loss=0.111, over 28682.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3731, pruned_loss=0.1209, over 5660399.42 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3666, pruned_loss=0.1183, over 5681005.83 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3734, pruned_loss=0.121, over 5651542.77 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:26:11,556 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:33,117 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:43,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:26:54,691 INFO [train.py:968] (1/2) Epoch 19, batch 28550, giga_loss[loss=0.2843, simple_loss=0.3535, pruned_loss=0.1075, over 28891.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3736, pruned_loss=0.1223, over 5668019.61 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.3663, pruned_loss=0.1181, over 5681823.26 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3741, pruned_loss=0.1225, over 5660216.48 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:27:10,086 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3657, 3.2899, 1.5575, 1.5678], device='cuda:1'), covar=tensor([0.1020, 0.0349, 0.0876, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0552, 0.0378, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 00:27:16,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:27:48,127 INFO [optim.py:369] (1/2) 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 00:27:48,929 INFO [train.py:968] (1/2) Epoch 19, batch 28600, giga_loss[loss=0.3243, simple_loss=0.3849, pruned_loss=0.1318, over 28273.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3729, pruned_loss=0.1227, over 5660947.45 frames. ], libri_tot_loss[loss=0.3012, simple_loss=0.3663, pruned_loss=0.1181, over 5672081.55 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3735, pruned_loss=0.1229, over 5662595.43 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:28:41,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=851228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:28:55,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3323, 1.5012, 1.3676, 1.5243], device='cuda:1'), covar=tensor([0.0771, 0.0344, 0.0322, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 00:28:55,342 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 28650, giga_loss[loss=0.272, simple_loss=0.3409, pruned_loss=0.1015, over 28934.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.121, over 5667694.40 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3658, pruned_loss=0.1178, over 5672482.22 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3709, pruned_loss=0.1215, over 5668678.43 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:28:58,970 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:22,953 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:25,269 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:25,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:40,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 00:29:40,747 INFO [optim.py:369] (1/2) 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 00:29:41,434 INFO [train.py:968] (1/2) Epoch 19, batch 28700, giga_loss[loss=0.3347, simple_loss=0.388, pruned_loss=0.1407, over 27600.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1201, over 5674158.51 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3654, pruned_loss=0.1174, over 5679203.43 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.1209, over 5668662.49 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:29:49,884 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:50,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:29:53,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:30:07,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-10 00:30:23,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 28750, giga_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1177, over 28709.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3681, pruned_loss=0.1205, over 5663811.30 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3649, pruned_loss=0.1169, over 5684716.78 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3695, pruned_loss=0.1217, over 5654506.20 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:30:33,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:31:15,740 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 28800, giga_loss[loss=0.2914, simple_loss=0.3649, pruned_loss=0.1089, over 28867.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.368, pruned_loss=0.1204, over 5669150.46 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3644, pruned_loss=0.1166, over 5687092.41 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1217, over 5659197.32 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:31:34,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-10 00:31:38,910 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=851408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:32:09,448 INFO [train.py:968] (1/2) Epoch 19, batch 28850, giga_loss[loss=0.3118, simple_loss=0.3812, pruned_loss=0.1212, over 29043.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3706, pruned_loss=0.1228, over 5661108.07 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3646, pruned_loss=0.1166, over 5686765.38 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1239, over 5653197.51 frames. ], batch size: 155, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:32:49,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2300, 1.3243, 3.3768, 3.0083], device='cuda:1'), covar=tensor([0.1581, 0.2625, 0.0502, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0641, 0.0946, 0.0884], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 00:32:53,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:32:55,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:32:55,795 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 28900, giga_loss[loss=0.332, simple_loss=0.3686, pruned_loss=0.1477, over 23485.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3728, pruned_loss=0.1247, over 5649511.20 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3643, pruned_loss=0.1164, over 5679781.92 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3742, pruned_loss=0.1259, over 5648598.47 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:32:58,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2358, 1.6441, 1.2098, 0.7995], device='cuda:1'), covar=tensor([0.3947, 0.2435, 0.2459, 0.4680], device='cuda:1'), in_proj_covar=tensor([0.1724, 0.1631, 0.1592, 0.1403], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 00:33:26,929 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:33:33,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3697, 2.0132, 1.4688, 0.5263], device='cuda:1'), covar=tensor([0.4677, 0.2715, 0.4019, 0.5742], device='cuda:1'), in_proj_covar=tensor([0.1722, 0.1629, 0.1590, 0.1400], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 00:33:49,225 INFO [train.py:968] (1/2) Epoch 19, batch 28950, giga_loss[loss=0.3757, simple_loss=0.4122, pruned_loss=0.1696, over 27916.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3727, pruned_loss=0.1254, over 5643466.98 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3644, pruned_loss=0.1165, over 5681332.08 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3737, pruned_loss=0.1263, over 5641147.11 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:34:03,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6453, 1.8796, 1.9174, 1.4263], device='cuda:1'), covar=tensor([0.1782, 0.2464, 0.1482, 0.1746], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0704, 0.0930, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 00:34:35,232 INFO [optim.py:369] (1/2) 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 00:34:35,246 INFO [train.py:968] (1/2) Epoch 19, batch 29000, giga_loss[loss=0.2998, simple_loss=0.3658, pruned_loss=0.1169, over 28551.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3725, pruned_loss=0.1254, over 5653748.58 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3644, pruned_loss=0.1164, over 5687091.31 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3736, pruned_loss=0.1264, over 5645915.56 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:34:48,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=851603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:35:02,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3781, 4.2010, 3.9884, 1.9718], device='cuda:1'), covar=tensor([0.0602, 0.0768, 0.0788, 0.2058], device='cuda:1'), in_proj_covar=tensor([0.1216, 0.1128, 0.0964, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 00:35:18,328 INFO [train.py:968] (1/2) Epoch 19, batch 29050, libri_loss[loss=0.2756, simple_loss=0.3383, pruned_loss=0.1064, over 29670.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3729, pruned_loss=0.1256, over 5654501.79 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.364, pruned_loss=0.1162, over 5694190.83 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3745, pruned_loss=0.1271, over 5639763.14 frames. ], batch size: 73, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:36:07,294 INFO [optim.py:369] (1/2) 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 00:36:07,309 INFO [train.py:968] (1/2) Epoch 19, batch 29100, giga_loss[loss=0.3187, simple_loss=0.3599, pruned_loss=0.1388, over 23618.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3732, pruned_loss=0.1251, over 5649624.29 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3643, pruned_loss=0.1163, over 5695373.18 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3745, pruned_loss=0.1265, over 5635752.65 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:36:54,302 INFO [train.py:968] (1/2) Epoch 19, batch 29150, giga_loss[loss=0.3227, simple_loss=0.3889, pruned_loss=0.1283, over 28824.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.125, over 5660124.06 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.364, pruned_loss=0.1161, over 5698873.73 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3753, pruned_loss=0.1264, over 5645511.41 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:37:01,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:37:03,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:37:04,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 00:37:31,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:37:36,771 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 19, batch 29200, giga_loss[loss=0.2823, simple_loss=0.3526, pruned_loss=0.106, over 28886.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3746, pruned_loss=0.1255, over 5661715.84 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1163, over 5693324.51 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3758, pruned_loss=0.1266, over 5653803.36 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:37:46,185 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-10 00:38:25,762 INFO [train.py:968] (1/2) Epoch 19, batch 29250, giga_loss[loss=0.3551, simple_loss=0.4061, pruned_loss=0.152, over 28194.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1262, over 5674465.83 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3642, pruned_loss=0.1162, over 5697553.55 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3768, pruned_loss=0.1273, over 5664069.88 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:38:41,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-10 00:39:12,213 INFO [train.py:968] (1/2) Epoch 19, batch 29300, giga_loss[loss=0.3368, simple_loss=0.3943, pruned_loss=0.1396, over 27953.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3763, pruned_loss=0.1267, over 5675818.82 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5701592.39 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3776, pruned_loss=0.128, over 5663610.22 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:39:13,116 INFO [optim.py:369] (1/2) 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 00:39:18,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-10 00:39:53,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5770, 4.3897, 4.1468, 1.9230], device='cuda:1'), covar=tensor([0.0587, 0.0759, 0.0843, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.1222, 0.1135, 0.0967, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 00:39:53,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=851926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:39:56,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=851929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:40:05,803 INFO [train.py:968] (1/2) Epoch 19, batch 29350, giga_loss[loss=0.2576, simple_loss=0.3388, pruned_loss=0.08819, over 29014.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3759, pruned_loss=0.1249, over 5678340.26 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3637, pruned_loss=0.1158, over 5705559.31 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3774, pruned_loss=0.1263, over 5664610.30 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:40:15,739 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 00:40:25,273 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=851958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:40:34,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4545, 1.6200, 1.6318, 1.2556], device='cuda:1'), covar=tensor([0.1553, 0.2574, 0.1355, 0.1646], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0703, 0.0931, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 00:40:51,869 INFO [train.py:968] (1/2) Epoch 19, batch 29400, giga_loss[loss=0.2728, simple_loss=0.3479, pruned_loss=0.09888, over 28891.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3752, pruned_loss=0.1238, over 5667170.19 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3642, pruned_loss=0.1162, over 5700944.15 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3763, pruned_loss=0.1247, over 5658993.49 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:40:53,440 INFO [optim.py:369] (1/2) 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 00:41:18,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3654, 1.4656, 1.3909, 1.2920], device='cuda:1'), covar=tensor([0.2180, 0.2028, 0.2019, 0.2052], device='cuda:1'), in_proj_covar=tensor([0.1928, 0.1866, 0.1799, 0.1937], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 00:41:36,096 INFO [train.py:968] (1/2) Epoch 19, batch 29450, libri_loss[loss=0.2505, simple_loss=0.3242, pruned_loss=0.08837, over 29554.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3722, pruned_loss=0.1221, over 5670638.40 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5706895.20 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3738, pruned_loss=0.1232, over 5657162.99 frames. ], batch size: 77, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:42:21,663 INFO [train.py:968] (1/2) Epoch 19, batch 29500, giga_loss[loss=0.4167, simple_loss=0.4417, pruned_loss=0.1958, over 26595.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1236, over 5666185.05 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1163, over 5709196.68 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.375, pruned_loss=0.1244, over 5653184.98 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:42:22,270 INFO [optim.py:369] (1/2) 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 00:43:04,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4021, 1.9398, 1.3460, 0.6808], device='cuda:1'), covar=tensor([0.4305, 0.2396, 0.3443, 0.5513], device='cuda:1'), in_proj_covar=tensor([0.1726, 0.1633, 0.1594, 0.1406], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 00:43:10,337 INFO [train.py:968] (1/2) Epoch 19, batch 29550, giga_loss[loss=0.3842, simple_loss=0.4164, pruned_loss=0.176, over 26557.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.374, pruned_loss=0.1231, over 5660603.72 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3645, pruned_loss=0.1166, over 5699888.05 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3748, pruned_loss=0.1237, over 5657609.45 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:44:01,635 INFO [train.py:968] (1/2) Epoch 19, batch 29600, giga_loss[loss=0.267, simple_loss=0.3422, pruned_loss=0.09592, over 28423.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3751, pruned_loss=0.1248, over 5648421.06 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3644, pruned_loss=0.1167, over 5694456.69 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.376, pruned_loss=0.1253, over 5650262.31 frames. ], batch size: 65, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:44:02,172 INFO [optim.py:369] (1/2) 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 00:44:44,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6359, 2.3139, 1.6336, 0.8516], device='cuda:1'), covar=tensor([0.4753, 0.2751, 0.3905, 0.5394], device='cuda:1'), in_proj_covar=tensor([0.1719, 0.1627, 0.1587, 0.1401], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 00:44:48,942 INFO [train.py:968] (1/2) Epoch 19, batch 29650, giga_loss[loss=0.3208, simple_loss=0.3846, pruned_loss=0.1285, over 28849.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.374, pruned_loss=0.1247, over 5649067.97 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3644, pruned_loss=0.1167, over 5689456.74 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.375, pruned_loss=0.1253, over 5654110.78 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:45:03,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2387, 1.5706, 1.2408, 1.0010], device='cuda:1'), covar=tensor([0.2448, 0.2439, 0.2765, 0.2221], device='cuda:1'), in_proj_covar=tensor([0.1465, 0.1067, 0.1300, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 00:45:38,609 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:45:38,970 INFO [train.py:968] (1/2) Epoch 19, batch 29700, giga_loss[loss=0.3003, simple_loss=0.3667, pruned_loss=0.1169, over 28720.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.375, pruned_loss=0.1258, over 5649652.41 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3642, pruned_loss=0.1166, over 5691452.58 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.376, pruned_loss=0.1264, over 5651557.09 frames. ], batch size: 284, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:45:40,195 INFO [optim.py:369] (1/2) 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 00:45:50,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3237, 2.0219, 1.6930, 1.5264], device='cuda:1'), covar=tensor([0.0801, 0.0279, 0.0300, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0117, 0.0115, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0060, 0.0104], device='cuda:1')
+2023-03-10 00:46:22,624 INFO [train.py:968] (1/2) Epoch 19, batch 29750, giga_loss[loss=0.3424, simple_loss=0.3933, pruned_loss=0.1458, over 27649.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3767, pruned_loss=0.1269, over 5655432.15 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3648, pruned_loss=0.1169, over 5694360.51 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3774, pruned_loss=0.1275, over 5652853.20 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:47:05,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3023, 1.6573, 1.5522, 1.1268], device='cuda:1'), covar=tensor([0.1619, 0.2726, 0.1479, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0704, 0.0932, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 00:47:09,757 INFO [train.py:968] (1/2) Epoch 19, batch 29800, giga_loss[loss=0.2855, simple_loss=0.3588, pruned_loss=0.1061, over 28868.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3773, pruned_loss=0.1273, over 5646185.88 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1174, over 5691163.15 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3776, pruned_loss=0.1278, over 5644569.63 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:47:10,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 00:47:11,209 INFO [optim.py:369] (1/2) 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 00:47:57,281 INFO [train.py:968] (1/2) Epoch 19, batch 29850, giga_loss[loss=0.2751, simple_loss=0.3528, pruned_loss=0.0987, over 29044.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3751, pruned_loss=0.1248, over 5667643.50 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3655, pruned_loss=0.1173, over 5693641.96 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3757, pruned_loss=0.1255, over 5663703.02 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:48:50,115 INFO [train.py:968] (1/2) Epoch 19, batch 29900, giga_loss[loss=0.2818, simple_loss=0.3612, pruned_loss=0.1012, over 28957.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.124, over 5664984.19 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.365, pruned_loss=0.1169, over 5698247.68 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3756, pruned_loss=0.1249, over 5656925.44 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:48:51,371 INFO [optim.py:369] (1/2) 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 00:49:35,367 INFO [train.py:968] (1/2) Epoch 19, batch 29950, libri_loss[loss=0.3267, simple_loss=0.3885, pruned_loss=0.1324, over 29518.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3739, pruned_loss=0.1233, over 5672139.54 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1167, over 5700593.23 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3753, pruned_loss=0.1246, over 5662115.70 frames. ], batch size: 89, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:50:17,333 INFO [train.py:968] (1/2) Epoch 19, batch 30000, giga_loss[loss=0.268, simple_loss=0.339, pruned_loss=0.09851, over 28331.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5663185.52 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3653, pruned_loss=0.1171, over 5696006.48 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3743, pruned_loss=0.1243, over 5658173.23 frames. ], batch size: 65, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:50:17,334 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 00:50:28,073 INFO [train.py:1012] (1/2) Epoch 19, validation: loss=0.2051, simple_loss=0.3127, pruned_loss=0.04877, over 944034.00 frames. 
+2023-03-10 00:50:28,074 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 00:50:29,402 INFO [optim.py:369] (1/2) 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 00:51:14,149 INFO [train.py:968] (1/2) Epoch 19, batch 30050, giga_loss[loss=0.3331, simple_loss=0.3714, pruned_loss=0.1474, over 23445.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3722, pruned_loss=0.1233, over 5655671.26 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3651, pruned_loss=0.1171, over 5691834.05 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3735, pruned_loss=0.1243, over 5653871.44 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:51:40,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=852662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:51:46,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:52:01,352 INFO [train.py:968] (1/2) Epoch 19, batch 30100, giga_loss[loss=0.2916, simple_loss=0.3341, pruned_loss=0.1245, over 23519.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.369, pruned_loss=0.1212, over 5659949.27 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.365, pruned_loss=0.1169, over 5695657.17 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5654424.00 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:52:04,933 INFO [optim.py:369] (1/2) 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 00:52:18,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3486, 1.5010, 1.5964, 1.1984], device='cuda:1'), covar=tensor([0.1629, 0.2425, 0.1337, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0705, 0.0934, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 00:52:50,598 INFO [train.py:968] (1/2) Epoch 19, batch 30150, giga_loss[loss=0.2583, simple_loss=0.3272, pruned_loss=0.09475, over 28589.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3649, pruned_loss=0.1189, over 5675818.48 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3645, pruned_loss=0.1166, over 5699691.88 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3663, pruned_loss=0.12, over 5667297.37 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:53:26,215 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:53:35,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-10 00:53:38,066 INFO [train.py:968] (1/2) Epoch 19, batch 30200, giga_loss[loss=0.289, simple_loss=0.3563, pruned_loss=0.1108, over 28866.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3628, pruned_loss=0.118, over 5690454.90 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3646, pruned_loss=0.1167, over 5702694.43 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3638, pruned_loss=0.1188, over 5680706.35 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:53:39,712 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-10 00:53:40,488 INFO [optim.py:369] (1/2) 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 00:53:56,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=852805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:53:58,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=852808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:54:12,209 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:54:30,069 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=852837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:54:30,402 INFO [train.py:968] (1/2) Epoch 19, batch 30250, giga_loss[loss=0.3231, simple_loss=0.3867, pruned_loss=0.1298, over 27598.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3627, pruned_loss=0.1179, over 5691718.86 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3643, pruned_loss=0.1165, over 5704225.20 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3638, pruned_loss=0.1187, over 5682729.91 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:54:55,529 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:55:18,433 INFO [train.py:968] (1/2) Epoch 19, batch 30300, giga_loss[loss=0.261, simple_loss=0.3408, pruned_loss=0.0906, over 28870.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3635, pruned_loss=0.117, over 5688757.46 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3642, pruned_loss=0.1165, over 5709363.40 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3643, pruned_loss=0.1177, over 5676579.29 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:55:23,928 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=852906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:56:03,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3416, 1.5724, 1.2154, 1.5014], device='cuda:1'), covar=tensor([0.0773, 0.0337, 0.0356, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0117, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 00:56:11,363 INFO [train.py:968] (1/2) Epoch 19, batch 30350, giga_loss[loss=0.2702, simple_loss=0.34, pruned_loss=0.1002, over 26779.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3616, pruned_loss=0.1139, over 5683744.65 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3642, pruned_loss=0.1166, over 5713292.45 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3622, pruned_loss=0.1143, over 5669419.46 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:56:47,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9004, 1.1475, 1.1299, 0.9021], device='cuda:1'), covar=tensor([0.2251, 0.2012, 0.1238, 0.1798], device='cuda:1'), in_proj_covar=tensor([0.1922, 0.1853, 0.1785, 0.1925], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 00:56:58,936 INFO [train.py:968] (1/2) Epoch 19, batch 30400, giga_loss[loss=0.2781, simple_loss=0.3402, pruned_loss=0.108, over 26751.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3583, pruned_loss=0.1107, over 5663593.65 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3635, pruned_loss=0.1164, over 5700265.21 frames. ], giga_tot_loss[loss=0.2907, simple_loss=0.3592, pruned_loss=0.1111, over 5664203.18 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:57:01,196 INFO [optim.py:369] (1/2) 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 00:57:12,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:57:43,485 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:57:48,111 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:57:52,456 INFO [train.py:968] (1/2) Epoch 19, batch 30450, giga_loss[loss=0.2813, simple_loss=0.3586, pruned_loss=0.102, over 28780.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3551, pruned_loss=0.1075, over 5640480.08 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3638, pruned_loss=0.1167, over 5682376.35 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.3556, pruned_loss=0.1075, over 5655475.86 frames. ], batch size: 284, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 00:57:54,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3788, 1.9680, 1.5436, 1.5589], device='cuda:1'), covar=tensor([0.0807, 0.0309, 0.0339, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 00:58:00,390 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 30500, giga_loss[loss=0.2595, simple_loss=0.3548, pruned_loss=0.08211, over 28928.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.352, pruned_loss=0.1043, over 5627333.18 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3638, pruned_loss=0.1169, over 5663722.23 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3521, pruned_loss=0.1038, over 5655589.54 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:58:44,191 INFO [optim.py:369] (1/2) 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 00:58:54,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 00:59:26,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 00:59:31,117 INFO [train.py:968] (1/2) Epoch 19, batch 30550, giga_loss[loss=0.2773, simple_loss=0.3602, pruned_loss=0.09718, over 28947.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3503, pruned_loss=0.1016, over 5625150.55 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3631, pruned_loss=0.1166, over 5669830.48 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3506, pruned_loss=0.1011, over 5641516.22 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 00:59:45,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:02,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3154, 3.0444, 1.3934, 1.4547], device='cuda:1'), covar=tensor([0.0993, 0.0353, 0.0995, 0.1411], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0551, 0.0376, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 01:00:24,018 INFO [train.py:968] (1/2) Epoch 19, batch 30600, giga_loss[loss=0.274, simple_loss=0.3523, pruned_loss=0.0978, over 28639.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3517, pruned_loss=0.1027, over 5629430.76 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3627, pruned_loss=0.1165, over 5672339.12 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.352, pruned_loss=0.1021, over 5639220.27 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:00:27,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:29,075 INFO [optim.py:369] (1/2) 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 01:00:31,057 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:33,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:50,615 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:00:59,300 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:01:16,462 INFO [train.py:968] (1/2) Epoch 19, batch 30650, giga_loss[loss=0.2745, simple_loss=0.353, pruned_loss=0.09805, over 28242.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3493, pruned_loss=0.1012, over 5628204.21 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3623, pruned_loss=0.1163, over 5675736.09 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3497, pruned_loss=0.1006, over 5632267.49 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:01:17,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:01:46,632 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4290, 1.8641, 1.9235, 1.5156], device='cuda:1'), covar=tensor([0.2441, 0.1663, 0.1508, 0.1893], device='cuda:1'), in_proj_covar=tensor([0.1906, 0.1833, 0.1765, 0.1905], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 01:01:58,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:05,251 INFO [train.py:968] (1/2) Epoch 19, batch 30700, giga_loss[loss=0.2625, simple_loss=0.3395, pruned_loss=0.09278, over 28965.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3464, pruned_loss=0.09947, over 5628386.25 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3616, pruned_loss=0.1161, over 5671473.29 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3471, pruned_loss=0.09884, over 5635102.39 frames. ], batch size: 200, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:02:07,893 INFO [optim.py:369] (1/2) 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 01:02:09,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:13,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:35,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0263, 1.3140, 1.0563, 0.1907], device='cuda:1'), covar=tensor([0.3181, 0.2698, 0.3839, 0.5645], device='cuda:1'), in_proj_covar=tensor([0.1723, 0.1634, 0.1590, 0.1409], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 01:02:42,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:02:43,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-10 01:02:52,873 INFO [train.py:968] (1/2) Epoch 19, batch 30750, giga_loss[loss=0.2351, simple_loss=0.3015, pruned_loss=0.08432, over 24192.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3451, pruned_loss=0.09895, over 5632078.09 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3608, pruned_loss=0.1158, over 5673870.13 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3459, pruned_loss=0.09818, over 5633532.94 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:02:53,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1968, 3.4483, 2.1616, 1.1061], device='cuda:1'), covar=tensor([0.6299, 0.2533, 0.3903, 0.6146], device='cuda:1'), in_proj_covar=tensor([0.1718, 0.1629, 0.1588, 0.1405], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 01:02:53,892 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:16,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-10 01:03:21,013 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:25,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:33,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:37,384 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 30800, giga_loss[loss=0.2724, simple_loss=0.3371, pruned_loss=0.1038, over 26776.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3455, pruned_loss=0.09892, over 5626330.52 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3602, pruned_loss=0.1154, over 5663121.94 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3462, pruned_loss=0.09803, over 5635567.18 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:03:44,372 INFO [optim.py:369] (1/2) 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 01:03:50,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9097, 3.7405, 3.5423, 1.7671], device='cuda:1'), covar=tensor([0.0679, 0.0825, 0.0838, 0.2374], device='cuda:1'), in_proj_covar=tensor([0.1188, 0.1106, 0.0939, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 01:03:55,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:03:58,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:04:04,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:04:13,146 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:04:16,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 30850, libri_loss[loss=0.2275, simple_loss=0.2915, pruned_loss=0.08174, over 29321.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3437, pruned_loss=0.09726, over 5634898.16 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3596, pruned_loss=0.1154, over 5662258.85 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3442, pruned_loss=0.09595, over 5641698.22 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:04:45,928 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:07,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:17,318 INFO [train.py:968] (1/2) Epoch 19, batch 30900, giga_loss[loss=0.2641, simple_loss=0.3341, pruned_loss=0.097, over 27575.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09471, over 5640454.01 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3596, pruned_loss=0.1153, over 5662951.52 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3412, pruned_loss=0.09335, over 5644658.65 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:05:22,553 INFO [optim.py:369] (1/2) 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 01:05:25,925 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:41,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2037, 1.5949, 1.5361, 1.0734], device='cuda:1'), covar=tensor([0.1819, 0.2860, 0.1564, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0697, 0.0927, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 01:05:44,131 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:05:47,407 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 30950, giga_loss[loss=0.2537, simple_loss=0.3342, pruned_loss=0.08663, over 28892.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3384, pruned_loss=0.09349, over 5638772.04 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3592, pruned_loss=0.1152, over 5665026.20 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.338, pruned_loss=0.09161, over 5639178.95 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:06:12,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:15,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:16,292 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:16,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:20,987 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:43,918 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:47,887 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:52,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:06:52,581 INFO [train.py:968] (1/2) Epoch 19, batch 31000, giga_loss[loss=0.2417, simple_loss=0.3204, pruned_loss=0.08145, over 28855.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3363, pruned_loss=0.09271, over 5634671.70 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3587, pruned_loss=0.1151, over 5660043.66 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3361, pruned_loss=0.09096, over 5639234.60 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:06:56,661 INFO [optim.py:369] (1/2) 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 01:07:22,982 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:07:26,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:07:41,842 INFO [train.py:968] (1/2) Epoch 19, batch 31050, giga_loss[loss=0.287, simple_loss=0.3642, pruned_loss=0.1049, over 28704.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3353, pruned_loss=0.09283, over 5633353.26 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3583, pruned_loss=0.115, over 5666974.58 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3348, pruned_loss=0.09088, over 5629971.86 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:07:56,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:08:36,461 INFO [train.py:968] (1/2) Epoch 19, batch 31100, libri_loss[loss=0.3008, simple_loss=0.3601, pruned_loss=0.1208, over 25401.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3359, pruned_loss=0.09319, over 5609658.54 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3586, pruned_loss=0.1154, over 5649228.34 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3349, pruned_loss=0.09086, over 5621195.70 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:08:41,060 INFO [optim.py:369] (1/2) 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 01:09:23,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=853730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:09:27,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=853733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:09:32,189 INFO [train.py:968] (1/2) Epoch 19, batch 31150, giga_loss[loss=0.263, simple_loss=0.3442, pruned_loss=0.09089, over 27615.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3391, pruned_loss=0.09382, over 5626293.68 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3578, pruned_loss=0.1151, over 5655998.13 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3384, pruned_loss=0.09165, over 5628644.89 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:10:00,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=853762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:10:32,501 INFO [train.py:968] (1/2) Epoch 19, batch 31200, giga_loss[loss=0.2945, simple_loss=0.3825, pruned_loss=0.1032, over 28662.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3409, pruned_loss=0.09394, over 5634883.18 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3575, pruned_loss=0.115, over 5647471.13 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3404, pruned_loss=0.09209, over 5643516.77 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:10:38,653 INFO [optim.py:369] (1/2) 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 01:11:12,537 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 31250, giga_loss[loss=0.2407, simple_loss=0.3234, pruned_loss=0.07898, over 29004.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3405, pruned_loss=0.09366, over 5649071.52 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3574, pruned_loss=0.1149, over 5649692.68 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3401, pruned_loss=0.0921, over 5653949.54 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:12:26,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:12:28,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=853871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:12:48,127 INFO [train.py:968] (1/2) Epoch 19, batch 31300, giga_loss[loss=0.2884, simple_loss=0.3631, pruned_loss=0.1069, over 28957.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3379, pruned_loss=0.09149, over 5649081.33 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3572, pruned_loss=0.1147, over 5652317.13 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09015, over 5650462.61 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:12:56,795 INFO [optim.py:369] (1/2) 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 01:13:29,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=853919.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:13:50,899 INFO [train.py:968] (1/2) Epoch 19, batch 31350, giga_loss[loss=0.2137, simple_loss=0.3016, pruned_loss=0.06294, over 28750.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3357, pruned_loss=0.08918, over 5650109.43 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3568, pruned_loss=0.1147, over 5649290.59 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3352, pruned_loss=0.08757, over 5654192.26 frames. ], batch size: 119, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:14:53,283 INFO [train.py:968] (1/2) Epoch 19, batch 31400, giga_loss[loss=0.228, simple_loss=0.3051, pruned_loss=0.07544, over 28863.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3326, pruned_loss=0.08842, over 5653667.12 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3565, pruned_loss=0.1145, over 5647188.53 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08689, over 5658743.54 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:14:58,894 INFO [optim.py:369] (1/2) 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 01:15:26,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:15:29,303 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 31450, giga_loss[loss=0.318, simple_loss=0.3843, pruned_loss=0.1259, over 28332.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3326, pruned_loss=0.08889, over 5649432.32 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3563, pruned_loss=0.1145, over 5641541.38 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3318, pruned_loss=0.0869, over 5658187.50 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:16:01,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:16:23,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4181, 3.0310, 1.3916, 1.6234], device='cuda:1'), covar=tensor([0.0933, 0.0373, 0.0961, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0547, 0.0377, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 01:16:48,687 INFO [train.py:968] (1/2) Epoch 19, batch 31500, giga_loss[loss=0.2894, simple_loss=0.3677, pruned_loss=0.1056, over 28952.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3331, pruned_loss=0.08965, over 5659677.66 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3562, pruned_loss=0.1147, over 5646170.06 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3319, pruned_loss=0.0873, over 5662833.71 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:16:56,346 INFO [optim.py:369] (1/2) 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 01:17:45,800 INFO [train.py:968] (1/2) Epoch 19, batch 31550, giga_loss[loss=0.2414, simple_loss=0.3281, pruned_loss=0.07728, over 28932.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.334, pruned_loss=0.08943, over 5658899.18 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3561, pruned_loss=0.1148, over 5650055.93 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3326, pruned_loss=0.08685, over 5658370.30 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:17:52,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-10 01:18:51,266 INFO [train.py:968] (1/2) Epoch 19, batch 31600, giga_loss[loss=0.2292, simple_loss=0.2978, pruned_loss=0.08028, over 24650.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3348, pruned_loss=0.08966, over 5659600.14 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3553, pruned_loss=0.1145, over 5654093.38 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3338, pruned_loss=0.08714, over 5655741.93 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:18:55,543 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:19:53,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 01:19:53,781 INFO [train.py:968] (1/2) Epoch 19, batch 31650, libri_loss[loss=0.2407, simple_loss=0.3032, pruned_loss=0.08914, over 29388.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3313, pruned_loss=0.08763, over 5670368.41 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3544, pruned_loss=0.114, over 5662212.74 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3306, pruned_loss=0.0852, over 5660234.73 frames. ], batch size: 67, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:19:54,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5528, 4.3964, 4.1538, 2.1347], device='cuda:1'), covar=tensor([0.0502, 0.0634, 0.0731, 0.1875], device='cuda:1'), in_proj_covar=tensor([0.1178, 0.1093, 0.0930, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 01:20:03,675 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:21:03,795 INFO [train.py:968] (1/2) Epoch 19, batch 31700, libri_loss[loss=0.287, simple_loss=0.3571, pruned_loss=0.1085, over 29620.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3339, pruned_loss=0.08954, over 5667492.67 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3546, pruned_loss=0.1142, over 5656520.18 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3328, pruned_loss=0.08702, over 5664136.69 frames. ], batch size: 91, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:21:10,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854294.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:21:10,581 INFO [optim.py:369] (1/2) 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 01:21:24,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2894, 1.4982, 3.1999, 3.0179], device='cuda:1'), covar=tensor([0.1412, 0.2511, 0.0413, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0638, 0.0935, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 01:21:55,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:21:59,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:22:00,897 INFO [train.py:968] (1/2) Epoch 19, batch 31750, giga_loss[loss=0.3324, simple_loss=0.3913, pruned_loss=0.1368, over 26953.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3366, pruned_loss=0.08999, over 5664710.70 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3534, pruned_loss=0.1135, over 5665007.41 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3361, pruned_loss=0.08766, over 5654655.89 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:22:37,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:04,333 INFO [train.py:968] (1/2) Epoch 19, batch 31800, giga_loss[loss=0.2629, simple_loss=0.337, pruned_loss=0.09442, over 26943.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3402, pruned_loss=0.08971, over 5654324.70 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3531, pruned_loss=0.1135, over 5651903.74 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.08729, over 5657690.50 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:23:05,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:09,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:13,799 INFO [optim.py:369] (1/2) 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 01:23:19,022 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=854400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:40,518 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:23:51,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2005, 1.5305, 1.4670, 1.3723], device='cuda:1'), covar=tensor([0.1808, 0.1860, 0.2239, 0.1848], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0735, 0.0698, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 01:23:57,094 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=854434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:24:00,573 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854437.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 01:24:00,821 INFO [train.py:968] (1/2) Epoch 19, batch 31850, giga_loss[loss=0.2864, simple_loss=0.354, pruned_loss=0.1094, over 26623.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3407, pruned_loss=0.08981, over 5653856.73 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3529, pruned_loss=0.1135, over 5661120.39 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.34, pruned_loss=0.08672, over 5648377.81 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:24:04,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854440.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:24:26,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5961, 3.4394, 3.2627, 1.9956], device='cuda:1'), covar=tensor([0.0757, 0.0971, 0.0921, 0.1846], device='cuda:1'), in_proj_covar=tensor([0.1177, 0.1093, 0.0930, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 01:24:35,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854469.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 01:24:42,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7205, 2.0868, 1.4541, 1.6856], device='cuda:1'), covar=tensor([0.0888, 0.0526, 0.0916, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0441, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 01:24:58,390 INFO [train.py:968] (1/2) Epoch 19, batch 31900, giga_loss[loss=0.2273, simple_loss=0.3216, pruned_loss=0.06651, over 29050.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3408, pruned_loss=0.08929, over 5650683.21 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3528, pruned_loss=0.1135, over 5658228.60 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3399, pruned_loss=0.08609, over 5648550.50 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:25:06,488 INFO [optim.py:369] (1/2) 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 01:25:59,195 INFO [train.py:968] (1/2) Epoch 19, batch 31950, giga_loss[loss=0.2339, simple_loss=0.3218, pruned_loss=0.07303, over 28401.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3414, pruned_loss=0.09017, over 5635599.19 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3528, pruned_loss=0.1137, over 5641489.31 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3403, pruned_loss=0.08684, over 5648854.37 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:26:35,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8355, 2.1693, 1.7302, 2.1906], device='cuda:1'), covar=tensor([0.2571, 0.2558, 0.2971, 0.2342], device='cuda:1'), in_proj_covar=tensor([0.1468, 0.1061, 0.1303, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 01:27:05,093 INFO [train.py:968] (1/2) Epoch 19, batch 32000, giga_loss[loss=0.241, simple_loss=0.3266, pruned_loss=0.07771, over 28893.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3398, pruned_loss=0.09054, over 5651293.33 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3518, pruned_loss=0.113, over 5648784.65 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3394, pruned_loss=0.08762, over 5655016.41 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:27:13,459 INFO [optim.py:369] (1/2) 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 01:28:10,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 01:28:23,973 INFO [train.py:968] (1/2) Epoch 19, batch 32050, giga_loss[loss=0.259, simple_loss=0.3395, pruned_loss=0.0893, over 28686.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3411, pruned_loss=0.09172, over 5665880.74 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3518, pruned_loss=0.113, over 5654551.59 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3405, pruned_loss=0.08887, over 5664011.91 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:29:15,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3448, 3.1807, 3.0150, 1.5736], device='cuda:1'), covar=tensor([0.0922, 0.1071, 0.0998, 0.2292], device='cuda:1'), in_proj_covar=tensor([0.1177, 0.1095, 0.0932, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 01:29:38,059 INFO [train.py:968] (1/2) Epoch 19, batch 32100, giga_loss[loss=0.2142, simple_loss=0.2819, pruned_loss=0.07327, over 23807.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3373, pruned_loss=0.08977, over 5665829.84 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3518, pruned_loss=0.113, over 5658459.99 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3367, pruned_loss=0.08708, over 5661080.54 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 01:29:43,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2742, 1.6389, 1.4449, 1.5953], device='cuda:1'), covar=tensor([0.0792, 0.0343, 0.0338, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 01:29:46,364 INFO [optim.py:369] (1/2) 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 01:30:26,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 01:30:44,683 INFO [train.py:968] (1/2) Epoch 19, batch 32150, giga_loss[loss=0.2489, simple_loss=0.3316, pruned_loss=0.08306, over 28775.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3353, pruned_loss=0.08853, over 5652198.28 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3516, pruned_loss=0.1131, over 5641436.48 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3348, pruned_loss=0.08615, over 5663116.64 frames. ], batch size: 243, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:31:24,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4862, 1.6589, 1.7364, 1.2969], device='cuda:1'), covar=tensor([0.1886, 0.2754, 0.1576, 0.1941], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0694, 0.0928, 0.0830], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 01:31:34,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:31:49,698 INFO [train.py:968] (1/2) Epoch 19, batch 32200, giga_loss[loss=0.3342, simple_loss=0.4021, pruned_loss=0.1331, over 28668.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3339, pruned_loss=0.08852, over 5660271.85 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3514, pruned_loss=0.113, over 5647361.74 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3331, pruned_loss=0.08593, over 5663883.00 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:31:59,172 INFO [optim.py:369] (1/2) 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 01:32:13,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=854809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:32:45,693 INFO [train.py:968] (1/2) Epoch 19, batch 32250, libri_loss[loss=0.2736, simple_loss=0.3311, pruned_loss=0.108, over 29565.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3376, pruned_loss=0.09139, over 5664927.80 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3515, pruned_loss=0.1135, over 5651738.21 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3361, pruned_loss=0.08759, over 5663977.90 frames. ], batch size: 77, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:32:59,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2982, 1.4419, 1.2858, 1.5743], device='cuda:1'), covar=tensor([0.0781, 0.0351, 0.0344, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 01:33:39,290 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-10 01:33:44,136 INFO [train.py:968] (1/2) Epoch 19, batch 32300, giga_loss[loss=0.2181, simple_loss=0.3031, pruned_loss=0.06651, over 28920.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3395, pruned_loss=0.09211, over 5662967.10 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3509, pruned_loss=0.1131, over 5651790.65 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3385, pruned_loss=0.08885, over 5662540.85 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:33:51,101 INFO [optim.py:369] (1/2) 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 01:34:28,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=854918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:34:33,125 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 32350, giga_loss[loss=0.2718, simple_loss=0.3441, pruned_loss=0.09973, over 28860.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3379, pruned_loss=0.09228, over 5661302.03 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3508, pruned_loss=0.113, over 5652774.16 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.337, pruned_loss=0.08946, over 5660035.24 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:35:06,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:35:08,403 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=854955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:35:46,429 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=854984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:35:50,645 INFO [train.py:968] (1/2) Epoch 19, batch 32400, giga_loss[loss=0.2475, simple_loss=0.332, pruned_loss=0.0815, over 28575.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3387, pruned_loss=0.09348, over 5665668.03 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3503, pruned_loss=0.1128, over 5657193.71 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3381, pruned_loss=0.09075, over 5660972.19 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:36:00,909 INFO [optim.py:369] (1/2) 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 01:36:53,562 INFO [train.py:968] (1/2) Epoch 19, batch 32450, giga_loss[loss=0.2808, simple_loss=0.3634, pruned_loss=0.09916, over 28888.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3392, pruned_loss=0.09333, over 5669568.26 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3497, pruned_loss=0.1125, over 5661500.12 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.339, pruned_loss=0.0909, over 5662181.13 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:38:08,807 INFO [train.py:968] (1/2) Epoch 19, batch 32500, giga_loss[loss=0.2508, simple_loss=0.3404, pruned_loss=0.08058, over 28927.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.341, pruned_loss=0.09263, over 5664095.46 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3498, pruned_loss=0.1127, over 5652325.62 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3407, pruned_loss=0.09054, over 5666849.22 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:38:26,965 INFO [optim.py:369] (1/2) 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 01:39:29,287 INFO [train.py:968] (1/2) Epoch 19, batch 32550, giga_loss[loss=0.2635, simple_loss=0.3438, pruned_loss=0.09164, over 28611.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3398, pruned_loss=0.09138, over 5658441.10 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3499, pruned_loss=0.1128, over 5647060.60 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3393, pruned_loss=0.08923, over 5665710.72 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:40:42,848 INFO [train.py:968] (1/2) Epoch 19, batch 32600, giga_loss[loss=0.2772, simple_loss=0.3467, pruned_loss=0.1039, over 28953.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3361, pruned_loss=0.09052, over 5658897.49 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3501, pruned_loss=0.1129, over 5649514.51 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3355, pruned_loss=0.08853, over 5662427.24 frames. ], batch size: 284, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:40:55,202 INFO [optim.py:369] (1/2) 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 01:41:50,400 INFO [train.py:968] (1/2) Epoch 19, batch 32650, giga_loss[loss=0.2645, simple_loss=0.3307, pruned_loss=0.09917, over 27863.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3302, pruned_loss=0.08773, over 5666606.96 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3504, pruned_loss=0.1132, over 5652033.26 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3292, pruned_loss=0.08566, over 5667178.44 frames. ], batch size: 476, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:42:27,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:42:45,951 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4246, 1.5553, 1.6013, 1.4574], device='cuda:1'), covar=tensor([0.2271, 0.2013, 0.1690, 0.1961], device='cuda:1'), in_proj_covar=tensor([0.1890, 0.1812, 0.1736, 0.1880], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 01:42:52,034 INFO [train.py:968] (1/2) Epoch 19, batch 32700, giga_loss[loss=0.2316, simple_loss=0.3173, pruned_loss=0.07293, over 27952.00 frames. ], tot_loss[loss=0.253, simple_loss=0.33, pruned_loss=0.08796, over 5649689.85 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3502, pruned_loss=0.1132, over 5645273.30 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3288, pruned_loss=0.0855, over 5656067.67 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:43:04,284 INFO [optim.py:369] (1/2) 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 01:43:18,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-10 01:43:31,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3484, 1.6440, 1.4772, 1.6298], device='cuda:1'), covar=tensor([0.0761, 0.0346, 0.0325, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 01:43:53,249 INFO [train.py:968] (1/2) Epoch 19, batch 32750, giga_loss[loss=0.2502, simple_loss=0.3316, pruned_loss=0.0844, over 28968.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3324, pruned_loss=0.09007, over 5650304.08 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3501, pruned_loss=0.1132, over 5648767.91 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3313, pruned_loss=0.08777, over 5652095.58 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:44:22,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2572, 1.7131, 1.2615, 0.6286], device='cuda:1'), covar=tensor([0.5217, 0.2526, 0.3618, 0.5694], device='cuda:1'), in_proj_covar=tensor([0.1713, 0.1615, 0.1578, 0.1400], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 01:44:53,926 INFO [train.py:968] (1/2) Epoch 19, batch 32800, giga_loss[loss=0.2394, simple_loss=0.3277, pruned_loss=0.07555, over 28680.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3299, pruned_loss=0.08822, over 5657033.95 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3502, pruned_loss=0.1133, over 5656704.04 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3284, pruned_loss=0.08568, over 5651586.62 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:45:06,852 INFO [optim.py:369] (1/2) 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 01:45:41,044 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-10 01:45:56,293 INFO [train.py:968] (1/2) Epoch 19, batch 32850, giga_loss[loss=0.2804, simple_loss=0.3463, pruned_loss=0.1072, over 28914.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3286, pruned_loss=0.08652, over 5663519.86 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3502, pruned_loss=0.1133, over 5659085.18 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3269, pruned_loss=0.08385, over 5657276.24 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:46:32,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2341, 1.0948, 3.5890, 3.1805], device='cuda:1'), covar=tensor([0.1622, 0.2797, 0.0534, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0635, 0.0931, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 01:46:33,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1723, 1.4661, 1.5894, 1.2401], device='cuda:1'), covar=tensor([0.1836, 0.1660, 0.1950, 0.1717], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0729, 0.0692, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 01:47:04,815 INFO [train.py:968] (1/2) Epoch 19, batch 32900, giga_loss[loss=0.2531, simple_loss=0.3288, pruned_loss=0.08874, over 27582.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3278, pruned_loss=0.08649, over 5666250.85 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3502, pruned_loss=0.1133, over 5661974.93 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.326, pruned_loss=0.08392, over 5658908.25 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:47:22,227 INFO [optim.py:369] (1/2) 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 01:48:15,143 INFO [train.py:968] (1/2) Epoch 19, batch 32950, giga_loss[loss=0.2736, simple_loss=0.3496, pruned_loss=0.09883, over 27675.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3262, pruned_loss=0.085, over 5661765.33 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3499, pruned_loss=0.1131, over 5666969.95 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3247, pruned_loss=0.08259, over 5651157.99 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:49:24,796 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:49:25,232 INFO [train.py:968] (1/2) Epoch 19, batch 33000, giga_loss[loss=0.2192, simple_loss=0.2965, pruned_loss=0.07101, over 28947.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.326, pruned_loss=0.08479, over 5655673.55 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3498, pruned_loss=0.113, over 5660577.26 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3245, pruned_loss=0.08256, over 5651965.32 frames. ], batch size: 106, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:49:25,232 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 01:49:31,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1082, 1.5479, 1.5886, 1.3308], device='cuda:1'), covar=tensor([0.2131, 0.1668, 0.2318, 0.1910], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0728, 0.0692, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 01:49:33,753 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 01:49:45,775 INFO [optim.py:369] (1/2) 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 01:50:28,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5526, 1.9659, 1.7423, 1.6138], device='cuda:1'), covar=tensor([0.1925, 0.2180, 0.2123, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0727, 0.0692, 0.0665], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 01:50:39,559 INFO [train.py:968] (1/2) Epoch 19, batch 33050, giga_loss[loss=0.2739, simple_loss=0.349, pruned_loss=0.09945, over 28347.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3274, pruned_loss=0.08616, over 5656546.42 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3498, pruned_loss=0.113, over 5657578.38 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3262, pruned_loss=0.08433, over 5655987.07 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:50:40,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=855639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:50:50,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9771, 1.2587, 1.0413, 0.1725], device='cuda:1'), covar=tensor([0.3651, 0.2807, 0.4137, 0.6170], device='cuda:1'), in_proj_covar=tensor([0.1711, 0.1614, 0.1578, 0.1401], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 01:51:42,099 INFO [train.py:968] (1/2) Epoch 19, batch 33100, libri_loss[loss=0.3009, simple_loss=0.3671, pruned_loss=0.1173, over 29172.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3262, pruned_loss=0.08496, over 5656070.80 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3495, pruned_loss=0.1129, over 5660647.19 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3251, pruned_loss=0.08308, over 5652713.07 frames. ], batch size: 97, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:51:57,529 INFO [optim.py:369] (1/2) 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 01:52:41,089 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 33150, libri_loss[loss=0.2381, simple_loss=0.3039, pruned_loss=0.08608, over 29534.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3287, pruned_loss=0.08489, over 5655353.04 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3495, pruned_loss=0.113, over 5660503.47 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3275, pruned_loss=0.08297, over 5652750.42 frames. ], batch size: 70, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:53:24,446 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:53:33,238 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=855782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:53:36,676 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 33200, giga_loss[loss=0.2347, simple_loss=0.3212, pruned_loss=0.07414, over 28144.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3313, pruned_loss=0.08599, over 5660604.03 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3492, pruned_loss=0.1126, over 5666721.00 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3301, pruned_loss=0.08401, over 5653068.19 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:53:56,843 INFO [optim.py:369] (1/2) 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 01:54:15,628 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 33250, giga_loss[loss=0.2727, simple_loss=0.3528, pruned_loss=0.09624, over 28918.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3325, pruned_loss=0.08684, over 5656440.95 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3491, pruned_loss=0.1125, over 5669286.55 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3313, pruned_loss=0.08488, over 5647615.85 frames. ], batch size: 284, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:55:00,939 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=855850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 01:55:48,209 INFO [train.py:968] (1/2) Epoch 19, batch 33300, giga_loss[loss=0.2692, simple_loss=0.3439, pruned_loss=0.09726, over 27681.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3326, pruned_loss=0.08714, over 5662932.24 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3488, pruned_loss=0.1124, over 5674052.52 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3316, pruned_loss=0.08521, over 5651589.03 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 01:55:59,321 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 33350, giga_loss[loss=0.2402, simple_loss=0.3067, pruned_loss=0.08687, over 24380.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3299, pruned_loss=0.08547, over 5663704.22 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3482, pruned_loss=0.1121, over 5675772.75 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3291, pruned_loss=0.08353, over 5653058.14 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:57:17,389 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 33400, giga_loss[loss=0.3429, simple_loss=0.4034, pruned_loss=0.1412, over 28484.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3292, pruned_loss=0.08572, over 5663664.87 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3479, pruned_loss=0.1118, over 5672267.13 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3283, pruned_loss=0.08354, over 5658143.91 frames. ], batch size: 370, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:57:59,121 INFO [optim.py:369] (1/2) 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 01:58:37,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2358, 1.4554, 1.4823, 1.3017], device='cuda:1'), covar=tensor([0.2643, 0.2010, 0.1631, 0.2024], device='cuda:1'), in_proj_covar=tensor([0.1894, 0.1810, 0.1731, 0.1881], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 01:58:43,412 INFO [train.py:968] (1/2) Epoch 19, batch 33450, giga_loss[loss=0.2421, simple_loss=0.3298, pruned_loss=0.07718, over 28701.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3272, pruned_loss=0.08514, over 5661395.75 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3475, pruned_loss=0.1117, over 5668661.05 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3263, pruned_loss=0.08279, over 5660442.54 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:58:57,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2455, 1.1973, 3.5245, 3.0414], device='cuda:1'), covar=tensor([0.1596, 0.2769, 0.0511, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0636, 0.0930, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 01:59:41,033 INFO [train.py:968] (1/2) Epoch 19, batch 33500, giga_loss[loss=0.2495, simple_loss=0.3359, pruned_loss=0.08158, over 28411.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3311, pruned_loss=0.08735, over 5668160.87 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3478, pruned_loss=0.1119, over 5674725.56 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3296, pruned_loss=0.08445, over 5661749.48 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 01:59:57,833 INFO [optim.py:369] (1/2) 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 02:00:07,354 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:10,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:14,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:30,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0636, 2.9370, 1.9744, 1.2630], device='cuda:1'), covar=tensor([0.7012, 0.3173, 0.3910, 0.6342], device='cuda:1'), in_proj_covar=tensor([0.1702, 0.1612, 0.1570, 0.1398], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 02:00:48,152 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:00:48,496 INFO [train.py:968] (1/2) Epoch 19, batch 33550, giga_loss[loss=0.2273, simple_loss=0.2989, pruned_loss=0.07786, over 24342.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3317, pruned_loss=0.08713, over 5667508.32 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3478, pruned_loss=0.1119, over 5675889.93 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3304, pruned_loss=0.08476, over 5661419.86 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:00:51,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5542, 1.6452, 1.8500, 1.4138], device='cuda:1'), covar=tensor([0.1756, 0.2483, 0.1463, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0690, 0.0928, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 02:01:07,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:01:53,529 INFO [train.py:968] (1/2) Epoch 19, batch 33600, giga_loss[loss=0.294, simple_loss=0.3681, pruned_loss=0.1099, over 28996.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3327, pruned_loss=0.08798, over 5672798.60 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3472, pruned_loss=0.1116, over 5680032.54 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3317, pruned_loss=0.08568, over 5663779.73 frames. ], batch size: 285, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:02:10,299 INFO [optim.py:369] (1/2) 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 02:02:12,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5208, 1.7743, 1.8001, 1.3116], device='cuda:1'), covar=tensor([0.1913, 0.2539, 0.1567, 0.1871], device='cuda:1'), in_proj_covar=tensor([0.0881, 0.0689, 0.0926, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 02:02:41,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:02:44,212 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 33650, giga_loss[loss=0.2318, simple_loss=0.3174, pruned_loss=0.07311, over 28786.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3355, pruned_loss=0.08906, over 5661751.70 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3473, pruned_loss=0.1118, over 5674283.37 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3345, pruned_loss=0.08667, over 5660378.15 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:03:13,877 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:03:17,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:03:55,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:03:58,742 INFO [train.py:968] (1/2) Epoch 19, batch 33700, giga_loss[loss=0.2434, simple_loss=0.3322, pruned_loss=0.07727, over 28992.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3367, pruned_loss=0.0888, over 5662690.00 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3472, pruned_loss=0.1116, over 5677406.97 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3358, pruned_loss=0.08675, over 5658751.25 frames. ], batch size: 213, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:04:05,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:04:12,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:04:18,069 INFO [optim.py:369] (1/2) 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 02:04:23,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=5.81 vs. limit=5.0
+2023-03-10 02:04:31,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 02:04:49,323 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:05:10,657 INFO [train.py:968] (1/2) Epoch 19, batch 33750, giga_loss[loss=0.2086, simple_loss=0.2793, pruned_loss=0.06897, over 24465.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3361, pruned_loss=0.08908, over 5651865.03 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3471, pruned_loss=0.1116, over 5673888.67 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3351, pruned_loss=0.0868, over 5651535.45 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:05:27,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9076, 3.7345, 3.5092, 1.6183], device='cuda:1'), covar=tensor([0.0736, 0.0875, 0.0878, 0.2270], device='cuda:1'), in_proj_covar=tensor([0.1168, 0.1079, 0.0922, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 02:05:48,189 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:05:53,852 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 33800, giga_loss[loss=0.2829, simple_loss=0.3449, pruned_loss=0.1105, over 26903.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3347, pruned_loss=0.0889, over 5662820.36 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3473, pruned_loss=0.1118, over 5676945.56 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3334, pruned_loss=0.08624, over 5659623.84 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:06:31,973 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:07:16,637 INFO [train.py:968] (1/2) Epoch 19, batch 33850, giga_loss[loss=0.2473, simple_loss=0.3256, pruned_loss=0.08455, over 28156.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3345, pruned_loss=0.08906, over 5649244.53 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.347, pruned_loss=0.1116, over 5667748.10 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3334, pruned_loss=0.08655, over 5654898.04 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:07:39,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:07:56,889 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:08:26,176 INFO [train.py:968] (1/2) Epoch 19, batch 33900, libri_loss[loss=0.2343, simple_loss=0.2972, pruned_loss=0.08576, over 29599.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3334, pruned_loss=0.08919, over 5646801.49 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3465, pruned_loss=0.1113, over 5672046.74 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08709, over 5646871.43 frames. ], batch size: 76, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:08:42,859 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 33950, giga_loss[loss=0.2198, simple_loss=0.3066, pruned_loss=0.06652, over 28832.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3304, pruned_loss=0.08802, over 5639677.23 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3462, pruned_loss=0.1112, over 5664224.06 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3301, pruned_loss=0.08626, over 5646562.74 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:09:38,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4127, 1.6403, 1.1308, 1.1920], device='cuda:1'), covar=tensor([0.1044, 0.0613, 0.1224, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0439, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 02:09:48,152 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 34000, giga_loss[loss=0.2741, simple_loss=0.3433, pruned_loss=0.1024, over 26811.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3307, pruned_loss=0.08812, over 5636395.53 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3461, pruned_loss=0.1112, over 5665706.26 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3302, pruned_loss=0.08621, over 5639748.26 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:10:35,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8307, 2.1987, 2.1939, 1.7493], device='cuda:1'), covar=tensor([0.3237, 0.2192, 0.2383, 0.2511], device='cuda:1'), in_proj_covar=tensor([0.1897, 0.1808, 0.1738, 0.1879], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 02:10:46,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:10:48,883 INFO [optim.py:369] (1/2) 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 02:10:50,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3184, 3.5353, 1.5659, 1.5192], device='cuda:1'), covar=tensor([0.1028, 0.0290, 0.0947, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0543, 0.0376, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 02:11:34,206 INFO [train.py:968] (1/2) Epoch 19, batch 34050, giga_loss[loss=0.2162, simple_loss=0.3062, pruned_loss=0.06308, over 28161.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3288, pruned_loss=0.0857, over 5660779.69 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3455, pruned_loss=0.1107, over 5671350.88 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3285, pruned_loss=0.08392, over 5657877.74 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:11:41,413 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:12:29,620 INFO [train.py:968] (1/2) Epoch 19, batch 34100, giga_loss[loss=0.2563, simple_loss=0.3414, pruned_loss=0.08564, over 28599.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3299, pruned_loss=0.08444, over 5662493.29 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3449, pruned_loss=0.1104, over 5664276.78 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3298, pruned_loss=0.08258, over 5666444.27 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:12:44,909 INFO [optim.py:369] (1/2) 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 02:12:47,877 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-10 02:13:29,294 INFO [train.py:968] (1/2) Epoch 19, batch 34150, giga_loss[loss=0.2269, simple_loss=0.3203, pruned_loss=0.06673, over 28866.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3319, pruned_loss=0.08434, over 5662314.14 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3451, pruned_loss=0.1105, over 5668364.00 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3314, pruned_loss=0.08242, over 5661841.54 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:13:34,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:13:38,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:14:14,578 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=856773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:14:37,682 INFO [train.py:968] (1/2) Epoch 19, batch 34200, giga_loss[loss=0.2581, simple_loss=0.3399, pruned_loss=0.0881, over 28482.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3315, pruned_loss=0.08384, over 5666508.52 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3449, pruned_loss=0.1106, over 5671787.36 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3311, pruned_loss=0.08184, over 5662966.22 frames. ], batch size: 369, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:14:56,086 INFO [optim.py:369] (1/2) 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 02:15:16,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 02:15:35,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:15:50,815 INFO [train.py:968] (1/2) Epoch 19, batch 34250, giga_loss[loss=0.2664, simple_loss=0.3329, pruned_loss=0.09994, over 26897.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3331, pruned_loss=0.0851, over 5660302.82 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3453, pruned_loss=0.111, over 5666154.45 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3321, pruned_loss=0.0827, over 5662554.12 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:15:55,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=856842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:16:57,901 INFO [train.py:968] (1/2) Epoch 19, batch 34300, giga_loss[loss=0.2702, simple_loss=0.3449, pruned_loss=0.09777, over 28516.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3334, pruned_loss=0.08524, over 5663913.83 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3454, pruned_loss=0.111, over 5668507.31 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3324, pruned_loss=0.08294, over 5663740.02 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:17:16,801 INFO [optim.py:369] (1/2) 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 02:17:47,784 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 34350, giga_loss[loss=0.2493, simple_loss=0.3383, pruned_loss=0.08012, over 28543.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3337, pruned_loss=0.08494, over 5665790.22 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3454, pruned_loss=0.1111, over 5674168.77 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3327, pruned_loss=0.08246, over 5660436.89 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:18:32,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5434, 1.6033, 1.7978, 1.3537], device='cuda:1'), covar=tensor([0.1768, 0.2541, 0.1437, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0882, 0.0690, 0.0928, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 02:18:54,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:18:58,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:11,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=856985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:15,003 INFO [train.py:968] (1/2) Epoch 19, batch 34400, libri_loss[loss=0.2622, simple_loss=0.3331, pruned_loss=0.09564, over 29494.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3349, pruned_loss=0.08651, over 5661694.14 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.345, pruned_loss=0.111, over 5678742.26 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3339, pruned_loss=0.08339, over 5652513.57 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:19:15,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=856988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:17,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5227, 1.6240, 1.7430, 1.3190], device='cuda:1'), covar=tensor([0.1826, 0.2489, 0.1472, 0.1816], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0690, 0.0927, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 02:19:27,581 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=856999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:29,364 INFO [optim.py:369] (1/2) 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 02:19:33,127 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:37,257 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 02:19:42,318 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857011.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:19:49,716 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:19:57,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:20:11,790 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2368, 1.5723, 1.5521, 1.3880], device='cuda:1'), covar=tensor([0.1870, 0.1850, 0.2134, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.0444, 0.0723, 0.0689, 0.0659], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 02:20:15,156 INFO [train.py:968] (1/2) Epoch 19, batch 34450, giga_loss[loss=0.2476, simple_loss=0.3321, pruned_loss=0.08155, over 29007.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3379, pruned_loss=0.08752, over 5670577.03 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3452, pruned_loss=0.1112, over 5677030.16 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3367, pruned_loss=0.0841, over 5663772.47 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:20:56,538 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:21:01,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:21:23,519 INFO [train.py:968] (1/2) Epoch 19, batch 34500, giga_loss[loss=0.2383, simple_loss=0.321, pruned_loss=0.07782, over 29196.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3382, pruned_loss=0.08798, over 5678904.99 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3455, pruned_loss=0.1114, over 5678794.41 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3367, pruned_loss=0.08451, over 5672066.79 frames. ], batch size: 200, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:21:42,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 19, batch 34550, giga_loss[loss=0.2609, simple_loss=0.3243, pruned_loss=0.09879, over 24733.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3354, pruned_loss=0.08683, over 5685926.98 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3451, pruned_loss=0.1111, over 5684558.96 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3343, pruned_loss=0.08378, over 5675527.36 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:23:14,694 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:23:17,935 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:23:50,313 INFO [train.py:968] (1/2) Epoch 19, batch 34600, giga_loss[loss=0.2414, simple_loss=0.3275, pruned_loss=0.07761, over 28890.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3338, pruned_loss=0.08559, over 5692539.16 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3454, pruned_loss=0.1112, over 5687756.79 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3324, pruned_loss=0.08228, over 5681269.10 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:24:01,733 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:24:15,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4148, 1.6585, 1.2344, 1.2227], device='cuda:1'), covar=tensor([0.0945, 0.0459, 0.1000, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0380, 0.0436, 0.0504, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0008, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 02:24:52,360 INFO [train.py:968] (1/2) Epoch 19, batch 34650, giga_loss[loss=0.235, simple_loss=0.3165, pruned_loss=0.07679, over 28976.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3327, pruned_loss=0.08468, over 5697454.84 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3455, pruned_loss=0.1112, over 5683973.16 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3311, pruned_loss=0.08128, over 5691537.61 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:25:55,626 INFO [train.py:968] (1/2) Epoch 19, batch 34700, giga_loss[loss=0.2678, simple_loss=0.3491, pruned_loss=0.09323, over 28912.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3326, pruned_loss=0.08474, over 5688499.45 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3455, pruned_loss=0.1112, over 5685238.39 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3311, pruned_loss=0.08156, over 5682432.10 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:26:02,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:26:12,527 INFO [optim.py:369] (1/2) 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 02:26:53,701 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 34750, libri_loss[loss=0.2444, simple_loss=0.321, pruned_loss=0.08391, over 29520.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3354, pruned_loss=0.08648, over 5680012.85 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3452, pruned_loss=0.1109, over 5688741.79 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3343, pruned_loss=0.08361, over 5671833.61 frames. ], batch size: 83, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:27:38,143 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:27:52,416 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857386.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:27:53,714 INFO [train.py:968] (1/2) Epoch 19, batch 34800, giga_loss[loss=0.225, simple_loss=0.3051, pruned_loss=0.0725, over 28078.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.335, pruned_loss=0.08713, over 5657684.93 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3454, pruned_loss=0.1111, over 5672536.74 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3338, pruned_loss=0.08417, over 5665901.90 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:28:11,713 INFO [optim.py:369] (1/2) 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 02:28:51,598 INFO [train.py:968] (1/2) Epoch 19, batch 34850, giga_loss[loss=0.296, simple_loss=0.3548, pruned_loss=0.1186, over 26819.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3329, pruned_loss=0.08685, over 5663842.31 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3455, pruned_loss=0.1111, over 5676016.19 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3317, pruned_loss=0.0842, over 5666939.68 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:29:31,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4645, 3.3519, 1.4666, 1.6549], device='cuda:1'), covar=tensor([0.0947, 0.0372, 0.0949, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0542, 0.0377, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 02:29:51,464 INFO [train.py:968] (1/2) Epoch 19, batch 34900, giga_loss[loss=0.2753, simple_loss=0.349, pruned_loss=0.1008, over 28463.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3347, pruned_loss=0.0886, over 5664325.45 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3454, pruned_loss=0.111, over 5678517.32 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3336, pruned_loss=0.08637, over 5664409.66 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:30:03,072 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-10 02:30:08,789 INFO [optim.py:369] (1/2) 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 02:30:11,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1286, 2.5711, 1.2154, 1.3219], device='cuda:1'), covar=tensor([0.1100, 0.0382, 0.1007, 0.1485], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0543, 0.0378, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 02:30:21,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:30:24,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:30:31,776 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857529.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:30:34,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857532.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:30:40,156 INFO [train.py:968] (1/2) Epoch 19, batch 34950, giga_loss[loss=0.3099, simple_loss=0.3883, pruned_loss=0.1157, over 28394.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3422, pruned_loss=0.09342, over 5665407.29 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.345, pruned_loss=0.1108, over 5683142.47 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3415, pruned_loss=0.09135, over 5660978.52 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:30:48,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:30:59,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857561.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:31:28,905 INFO [train.py:968] (1/2) Epoch 19, batch 35000, giga_loss[loss=0.2759, simple_loss=0.3555, pruned_loss=0.09815, over 28256.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3515, pruned_loss=0.09816, over 5678269.78 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3451, pruned_loss=0.1108, over 5685370.21 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3508, pruned_loss=0.09638, over 5672852.59 frames. ], batch size: 77, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:31:44,634 INFO [optim.py:369] (1/2) 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 02:32:00,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:32:13,052 INFO [train.py:968] (1/2) Epoch 19, batch 35050, giga_loss[loss=0.2538, simple_loss=0.3385, pruned_loss=0.08454, over 28905.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3529, pruned_loss=0.09963, over 5674264.66 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3454, pruned_loss=0.1109, over 5681739.63 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3524, pruned_loss=0.09782, over 5673505.91 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:32:38,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:32:53,269 INFO [train.py:968] (1/2) Epoch 19, batch 35100, giga_loss[loss=0.2432, simple_loss=0.32, pruned_loss=0.08315, over 28117.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3471, pruned_loss=0.09718, over 5669145.82 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3451, pruned_loss=0.1106, over 5676614.31 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3471, pruned_loss=0.09562, over 5673320.42 frames. ], batch size: 77, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:33:05,916 INFO [optim.py:369] (1/2) 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 02:33:11,436 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 35150, giga_loss[loss=0.2685, simple_loss=0.3194, pruned_loss=0.1088, over 23853.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3414, pruned_loss=0.09528, over 5656854.14 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3457, pruned_loss=0.1108, over 5662744.06 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3409, pruned_loss=0.09343, over 5672066.11 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 02:34:18,682 INFO [train.py:968] (1/2) Epoch 19, batch 35200, giga_loss[loss=0.2071, simple_loss=0.2842, pruned_loss=0.06503, over 28977.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3337, pruned_loss=0.09177, over 5669964.64 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3459, pruned_loss=0.1109, over 5663480.42 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3329, pruned_loss=0.0899, over 5681555.79 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:34:30,299 INFO [optim.py:369] (1/2) 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 02:34:32,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4527, 2.4239, 2.4441, 2.1220], device='cuda:1'), covar=tensor([0.1822, 0.2384, 0.1966, 0.2200], device='cuda:1'), in_proj_covar=tensor([0.0447, 0.0727, 0.0693, 0.0664], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 02:34:37,375 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:34:39,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:34:56,609 INFO [train.py:968] (1/2) Epoch 19, batch 35250, giga_loss[loss=0.2223, simple_loss=0.2989, pruned_loss=0.07283, over 29019.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3282, pruned_loss=0.08998, over 5676663.35 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3465, pruned_loss=0.1112, over 5673412.01 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3263, pruned_loss=0.08724, over 5677506.52 frames. ], batch size: 155, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:35:01,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:10,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=857853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:12,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=857856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:34,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857880.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:35:39,482 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=857885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:41,287 INFO [train.py:968] (1/2) Epoch 19, batch 35300, libri_loss[loss=0.2658, simple_loss=0.3346, pruned_loss=0.09857, over 29582.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3207, pruned_loss=0.08617, over 5681019.46 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3464, pruned_loss=0.111, over 5677104.89 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3189, pruned_loss=0.0838, over 5678055.63 frames. ], batch size: 74, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:35:52,039 INFO [optim.py:369] (1/2) 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 02:35:54,155 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=857907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:35:54,925 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2281, 1.7421, 1.4062, 0.4176], device='cuda:1'), covar=tensor([0.4249, 0.2895, 0.4075, 0.5961], device='cuda:1'), in_proj_covar=tensor([0.1701, 0.1612, 0.1573, 0.1390], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 02:36:20,986 INFO [train.py:968] (1/2) Epoch 19, batch 35350, giga_loss[loss=0.2362, simple_loss=0.3072, pruned_loss=0.08254, over 28640.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3164, pruned_loss=0.08407, over 5695407.98 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3461, pruned_loss=0.1106, over 5682059.45 frames. ], giga_tot_loss[loss=0.2392, simple_loss=0.3146, pruned_loss=0.08196, over 5688957.31 frames. ], batch size: 242, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:36:53,865 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 02:37:03,284 INFO [train.py:968] (1/2) Epoch 19, batch 35400, giga_loss[loss=0.198, simple_loss=0.2762, pruned_loss=0.05986, over 28839.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3132, pruned_loss=0.08263, over 5691174.90 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3463, pruned_loss=0.1106, over 5681349.58 frames. ], giga_tot_loss[loss=0.2362, simple_loss=0.3112, pruned_loss=0.08064, over 5686770.25 frames. ], batch size: 119, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:37:04,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 02:37:09,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=857997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:37:14,961 INFO [optim.py:369] (1/2) 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 02:37:43,099 INFO [train.py:968] (1/2) Epoch 19, batch 35450, giga_loss[loss=0.2291, simple_loss=0.3031, pruned_loss=0.07758, over 28744.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3105, pruned_loss=0.08159, over 5688746.16 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3464, pruned_loss=0.1106, over 5683476.31 frames. ], giga_tot_loss[loss=0.2341, simple_loss=0.3086, pruned_loss=0.07978, over 5683463.54 frames. ], batch size: 284, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:38:28,092 INFO [train.py:968] (1/2) Epoch 19, batch 35500, giga_loss[loss=0.1935, simple_loss=0.2795, pruned_loss=0.05375, over 28871.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3087, pruned_loss=0.08121, over 5671366.84 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.347, pruned_loss=0.1108, over 5675712.05 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.3053, pruned_loss=0.07857, over 5674492.77 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:38:40,028 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 35550, giga_loss[loss=0.2024, simple_loss=0.2749, pruned_loss=0.06493, over 28533.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3075, pruned_loss=0.08096, over 5668581.50 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3475, pruned_loss=0.1111, over 5665390.55 frames. ], giga_tot_loss[loss=0.2291, simple_loss=0.3031, pruned_loss=0.07761, over 5679625.24 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:39:10,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:39:12,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:39:39,210 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 35600, giga_loss[loss=0.218, simple_loss=0.288, pruned_loss=0.07399, over 28596.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3043, pruned_loss=0.07909, over 5671051.61 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.348, pruned_loss=0.1113, over 5658860.10 frames. ], giga_tot_loss[loss=0.2257, simple_loss=0.2998, pruned_loss=0.07582, over 5684805.72 frames. ], batch size: 307, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:39:56,950 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:39:59,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6628, 1.6408, 1.9248, 1.5466], device='cuda:1'), covar=tensor([0.1524, 0.2013, 0.1243, 0.1523], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0698, 0.0940, 0.0836], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 02:40:07,058 INFO [optim.py:369] (1/2) 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 02:40:38,936 INFO [train.py:968] (1/2) Epoch 19, batch 35650, giga_loss[loss=0.1933, simple_loss=0.2777, pruned_loss=0.05443, over 28996.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3007, pruned_loss=0.07719, over 5674963.23 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3482, pruned_loss=0.1114, over 5661981.94 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.2963, pruned_loss=0.07408, over 5683066.45 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:40:50,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 02:40:52,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=858255.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:41:17,403 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 35700, giga_loss[loss=0.1959, simple_loss=0.2733, pruned_loss=0.05921, over 28841.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.2998, pruned_loss=0.07754, over 5673473.53 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3487, pruned_loss=0.1117, over 5668805.01 frames. ], giga_tot_loss[loss=0.2208, simple_loss=0.2943, pruned_loss=0.07365, over 5674466.24 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:41:39,194 INFO [optim.py:369] (1/2) 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 02:42:05,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:42:09,318 INFO [train.py:968] (1/2) Epoch 19, batch 35750, giga_loss[loss=0.2488, simple_loss=0.3231, pruned_loss=0.08726, over 28452.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.2996, pruned_loss=0.07766, over 5671419.98 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3488, pruned_loss=0.1117, over 5670119.98 frames. ], giga_tot_loss[loss=0.2219, simple_loss=0.295, pruned_loss=0.07442, over 5671032.81 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:42:38,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 02:42:53,566 INFO [train.py:968] (1/2) Epoch 19, batch 35800, giga_loss[loss=0.3239, simple_loss=0.3881, pruned_loss=0.1298, over 27902.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3119, pruned_loss=0.08376, over 5682727.38 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3495, pruned_loss=0.112, over 5673311.42 frames. ], giga_tot_loss[loss=0.233, simple_loss=0.3062, pruned_loss=0.07992, over 5679659.02 frames. ], batch size: 412, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:43:05,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858398.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:43:07,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858401.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 02:43:11,077 INFO [optim.py:369] (1/2) 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 02:43:26,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6003, 1.7186, 1.2033, 1.2873], device='cuda:1'), covar=tensor([0.0870, 0.0585, 0.1027, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0444, 0.0513, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 02:43:30,952 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:43:34,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:43:34,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8379, 1.9081, 2.0178, 1.5779], device='cuda:1'), covar=tensor([0.1808, 0.2283, 0.1434, 0.1748], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0696, 0.0935, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 02:43:36,045 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 35850, giga_loss[loss=0.3155, simple_loss=0.3887, pruned_loss=0.1211, over 28670.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3257, pruned_loss=0.09107, over 5684702.17 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3496, pruned_loss=0.112, over 5675771.51 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3208, pruned_loss=0.08784, over 5680212.27 frames. ], batch size: 78, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:44:00,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:44:28,191 INFO [train.py:968] (1/2) Epoch 19, batch 35900, libri_loss[loss=0.3114, simple_loss=0.3807, pruned_loss=0.121, over 25775.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3355, pruned_loss=0.09593, over 5676041.58 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3501, pruned_loss=0.1121, over 5668267.16 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3306, pruned_loss=0.09276, over 5680491.94 frames. ], batch size: 136, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:44:32,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 02:44:42,001 INFO [optim.py:369] (1/2) 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 02:44:43,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9367, 2.9159, 1.9507, 0.9396], device='cuda:1'), covar=tensor([0.8684, 0.3711, 0.4088, 0.7253], device='cuda:1'), in_proj_covar=tensor([0.1713, 0.1618, 0.1578, 0.1396], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 02:45:10,786 INFO [train.py:968] (1/2) Epoch 19, batch 35950, giga_loss[loss=0.2557, simple_loss=0.337, pruned_loss=0.08719, over 28918.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3407, pruned_loss=0.09736, over 5678025.74 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3499, pruned_loss=0.1119, over 5671978.57 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3369, pruned_loss=0.0948, over 5678401.63 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:45:23,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5349, 1.8299, 1.4337, 1.7244], device='cuda:1'), covar=tensor([0.2601, 0.2670, 0.3037, 0.2417], device='cuda:1'), in_proj_covar=tensor([0.1462, 0.1060, 0.1299, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 02:45:37,131 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 36000, giga_loss[loss=0.246, simple_loss=0.3343, pruned_loss=0.07885, over 28691.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.343, pruned_loss=0.09769, over 5667632.94 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.35, pruned_loss=0.1119, over 5669181.11 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3397, pruned_loss=0.09525, over 5671147.98 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:45:55,605 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 02:46:04,211 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 02:46:19,367 INFO [optim.py:369] (1/2) 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 02:46:25,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5368, 1.7430, 1.4180, 1.6097], device='cuda:1'), covar=tensor([0.2540, 0.2565, 0.2750, 0.2315], device='cuda:1'), in_proj_covar=tensor([0.1460, 0.1060, 0.1298, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 02:46:50,117 INFO [train.py:968] (1/2) Epoch 19, batch 36050, libri_loss[loss=0.2636, simple_loss=0.325, pruned_loss=0.1011, over 29654.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3443, pruned_loss=0.0979, over 5670814.98 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3502, pruned_loss=0.112, over 5676100.98 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3412, pruned_loss=0.09544, over 5667143.58 frames. ], batch size: 69, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:47:31,292 INFO [train.py:968] (1/2) Epoch 19, batch 36100, giga_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1114, over 28946.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.346, pruned_loss=0.09926, over 5676894.95 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3502, pruned_loss=0.1119, over 5673605.50 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3433, pruned_loss=0.09693, over 5676505.55 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:47:47,319 INFO [optim.py:369] (1/2) 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 02:47:48,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=858707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:47:50,565 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:47:52,571 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:48:14,910 INFO [train.py:968] (1/2) Epoch 19, batch 36150, giga_loss[loss=0.2688, simple_loss=0.3487, pruned_loss=0.09447, over 29077.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3489, pruned_loss=0.1013, over 5679918.91 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3504, pruned_loss=0.1118, over 5679005.40 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3466, pruned_loss=0.0993, over 5674775.94 frames. ], batch size: 155, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:48:18,195 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:48:20,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:48:42,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2631, 1.3669, 1.1909, 1.1512], device='cuda:1'), covar=tensor([0.1993, 0.1948, 0.1698, 0.1894], device='cuda:1'), in_proj_covar=tensor([0.1909, 0.1820, 0.1749, 0.1898], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 02:48:48,923 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 36200, giga_loss[loss=0.295, simple_loss=0.3783, pruned_loss=0.1059, over 28945.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.103, over 5681580.23 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3506, pruned_loss=0.112, over 5673259.38 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3501, pruned_loss=0.1011, over 5682355.23 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:49:07,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6597, 1.7172, 1.3581, 1.2800], device='cuda:1'), covar=tensor([0.0921, 0.0584, 0.0991, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0444, 0.0513, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 02:49:11,887 INFO [optim.py:369] (1/2) 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 02:49:20,085 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=858816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:49:20,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4562, 3.5665, 1.5477, 1.5925], device='cuda:1'), covar=tensor([0.0988, 0.0278, 0.0936, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0542, 0.0376, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 02:49:36,541 INFO [train.py:968] (1/2) Epoch 19, batch 36250, libri_loss[loss=0.2748, simple_loss=0.352, pruned_loss=0.09885, over 29148.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3553, pruned_loss=0.1034, over 5700737.65 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3509, pruned_loss=0.1121, over 5678985.09 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3535, pruned_loss=0.1015, over 5696521.49 frames. ], batch size: 101, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:49:46,361 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=858850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:49:48,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=858853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:50:13,094 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=858882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:50:16,994 INFO [train.py:968] (1/2) Epoch 19, batch 36300, giga_loss[loss=0.2772, simple_loss=0.3609, pruned_loss=0.09671, over 28787.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.356, pruned_loss=0.1033, over 5691014.92 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3505, pruned_loss=0.1117, over 5677788.20 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3551, pruned_loss=0.1019, over 5689812.62 frames. ], batch size: 243, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:50:32,620 INFO [optim.py:369] (1/2) 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 02:50:46,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7087, 1.7997, 1.7388, 1.5176], device='cuda:1'), covar=tensor([0.1868, 0.2170, 0.2367, 0.2187], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0736, 0.0700, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 02:50:58,110 INFO [train.py:968] (1/2) Epoch 19, batch 36350, giga_loss[loss=0.2598, simple_loss=0.3346, pruned_loss=0.09257, over 27590.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.357, pruned_loss=0.1028, over 5693156.82 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3507, pruned_loss=0.1117, over 5677953.71 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3562, pruned_loss=0.1015, over 5692383.22 frames. ], batch size: 472, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:51:38,667 INFO [train.py:968] (1/2) Epoch 19, batch 36400, giga_loss[loss=0.2813, simple_loss=0.3703, pruned_loss=0.09621, over 28992.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3571, pruned_loss=0.1023, over 5690908.64 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3511, pruned_loss=0.112, over 5674648.32 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3563, pruned_loss=0.1008, over 5693223.09 frames. ], batch size: 164, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:51:52,063 INFO [optim.py:369] (1/2) 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 02:52:15,911 INFO [train.py:968] (1/2) Epoch 19, batch 36450, giga_loss[loss=0.2523, simple_loss=0.3403, pruned_loss=0.08219, over 28492.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.356, pruned_loss=0.1008, over 5698148.53 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3521, pruned_loss=0.1124, over 5678744.74 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3547, pruned_loss=0.09876, over 5696897.87 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:52:33,522 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=859060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:52:59,553 INFO [train.py:968] (1/2) Epoch 19, batch 36500, giga_loss[loss=0.2803, simple_loss=0.3603, pruned_loss=0.1002, over 28880.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3557, pruned_loss=0.1008, over 5687834.91 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3523, pruned_loss=0.1124, over 5680750.67 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3545, pruned_loss=0.09902, over 5685329.96 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:53:08,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3553, 1.5559, 1.5699, 1.3808], device='cuda:1'), covar=tensor([0.1772, 0.1766, 0.2105, 0.1807], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0740, 0.0703, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 02:53:14,234 INFO [optim.py:369] (1/2) 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 02:53:14,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1610, 1.1249, 3.8476, 3.2649], device='cuda:1'), covar=tensor([0.1762, 0.2651, 0.0773, 0.1008], device='cuda:1'), in_proj_covar=tensor([0.0733, 0.0631, 0.0927, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 02:53:23,008 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=859116.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:53:27,451 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 36550, giga_loss[loss=0.376, simple_loss=0.4062, pruned_loss=0.173, over 26514.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.358, pruned_loss=0.1041, over 5683028.53 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3526, pruned_loss=0.1126, over 5681390.21 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3568, pruned_loss=0.1023, over 5680904.42 frames. ], batch size: 555, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:53:48,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3417, 1.1538, 4.1189, 3.4038], device='cuda:1'), covar=tensor([0.1680, 0.3002, 0.0442, 0.1218], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0632, 0.0929, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 02:53:48,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3291, 4.1464, 3.9804, 1.5873], device='cuda:1'), covar=tensor([0.0702, 0.0807, 0.0899, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1164, 0.1081, 0.0920, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 02:53:57,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:54:27,956 INFO [train.py:968] (1/2) Epoch 19, batch 36600, giga_loss[loss=0.2837, simple_loss=0.3474, pruned_loss=0.11, over 28830.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.36, pruned_loss=0.1075, over 5690882.41 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3525, pruned_loss=0.1126, over 5686153.80 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3592, pruned_loss=0.1059, over 5685098.43 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:54:30,458 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:54:44,162 INFO [optim.py:369] (1/2) 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 02:54:47,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 02:55:13,325 INFO [train.py:968] (1/2) Epoch 19, batch 36650, giga_loss[loss=0.2958, simple_loss=0.3545, pruned_loss=0.1186, over 28510.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.36, pruned_loss=0.1088, over 5689015.80 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.353, pruned_loss=0.1129, over 5688338.77 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.359, pruned_loss=0.1073, over 5682594.69 frames. ], batch size: 78, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:55:28,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-10 02:55:32,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1435, 1.7296, 1.3846, 0.4017], device='cuda:1'), covar=tensor([0.4244, 0.2523, 0.3896, 0.5430], device='cuda:1'), in_proj_covar=tensor([0.1704, 0.1606, 0.1579, 0.1390], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 02:55:38,426 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:55:40,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:55:41,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-10 02:56:01,208 INFO [train.py:968] (1/2) Epoch 19, batch 36700, giga_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.09573, over 28881.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3574, pruned_loss=0.1077, over 5698249.20 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3532, pruned_loss=0.1129, over 5690512.91 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3565, pruned_loss=0.1064, over 5691238.18 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:56:07,273 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:07,910 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:09,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:16,293 INFO [optim.py:369] (1/2) 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 02:56:33,338 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:39,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:56:42,341 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 36750, giga_loss[loss=0.3057, simple_loss=0.3684, pruned_loss=0.1214, over 28662.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.356, pruned_loss=0.1073, over 5703322.47 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3533, pruned_loss=0.1127, over 5696818.26 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3553, pruned_loss=0.1063, over 5692344.96 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:57:06,530 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:57:17,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-10 02:57:26,162 INFO [train.py:968] (1/2) Epoch 19, batch 36800, giga_loss[loss=0.252, simple_loss=0.3353, pruned_loss=0.08433, over 28834.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3544, pruned_loss=0.1056, over 5697510.60 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3536, pruned_loss=0.1127, over 5693425.90 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3537, pruned_loss=0.1046, over 5691434.96 frames. ], batch size: 145, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 02:57:33,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3582, 1.4181, 1.3835, 1.2518], device='cuda:1'), covar=tensor([0.2344, 0.2357, 0.1781, 0.2137], device='cuda:1'), in_proj_covar=tensor([0.1903, 0.1817, 0.1742, 0.1898], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 02:57:44,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8030, 2.0103, 1.3659, 1.5535], device='cuda:1'), covar=tensor([0.0924, 0.0628, 0.1057, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0444, 0.0514, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 02:57:45,510 INFO [optim.py:369] (1/2) 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 02:58:01,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=859425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:58:09,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 02:58:11,340 INFO [train.py:968] (1/2) Epoch 19, batch 36850, libri_loss[loss=0.2586, simple_loss=0.3207, pruned_loss=0.09824, over 29357.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3517, pruned_loss=0.1031, over 5692352.65 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3539, pruned_loss=0.1129, over 5683671.97 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3508, pruned_loss=0.1019, over 5697259.16 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:58:44,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3848, 2.0120, 1.4917, 0.5852], device='cuda:1'), covar=tensor([0.5354, 0.2557, 0.3862, 0.6312], device='cuda:1'), in_proj_covar=tensor([0.1704, 0.1609, 0.1573, 0.1391], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 02:58:55,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3618, 1.4336, 1.3066, 1.5578], device='cuda:1'), covar=tensor([0.0785, 0.0374, 0.0351, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 02:58:57,503 INFO [train.py:968] (1/2) Epoch 19, batch 36900, giga_loss[loss=0.2303, simple_loss=0.3082, pruned_loss=0.07618, over 28738.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3471, pruned_loss=0.1004, over 5691137.99 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3541, pruned_loss=0.113, over 5689177.03 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3461, pruned_loss=0.09915, over 5690500.84 frames. ], batch size: 262, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 02:59:00,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859491.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 02:59:17,098 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 02:59:20,155 INFO [optim.py:369] (1/2) 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 02:59:47,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6590, 4.2625, 1.8077, 1.7819], device='cuda:1'), covar=tensor([0.0954, 0.0217, 0.0868, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0540, 0.0374, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 02:59:47,929 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 02:59:50,323 INFO [train.py:968] (1/2) Epoch 19, batch 36950, giga_loss[loss=0.235, simple_loss=0.293, pruned_loss=0.08851, over 23352.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3404, pruned_loss=0.09677, over 5689345.71 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3546, pruned_loss=0.1133, over 5691168.57 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3391, pruned_loss=0.09538, over 5687200.96 frames. ], batch size: 705, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:00:38,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:00:40,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:00:45,035 INFO [train.py:968] (1/2) Epoch 19, batch 37000, libri_loss[loss=0.2578, simple_loss=0.3348, pruned_loss=0.09041, over 29550.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3346, pruned_loss=0.09376, over 5672790.71 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3543, pruned_loss=0.1129, over 5686764.13 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3335, pruned_loss=0.09262, over 5674788.46 frames. ], batch size: 79, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:00:54,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7997, 1.8474, 1.7966, 1.5048], device='cuda:1'), covar=tensor([0.2505, 0.2324, 0.1989, 0.2675], device='cuda:1'), in_proj_covar=tensor([0.1898, 0.1812, 0.1740, 0.1899], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 03:01:05,450 INFO [optim.py:369] (1/2) 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 03:01:06,979 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:01:27,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859634.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:01:30,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859637.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:01:30,817 INFO [train.py:968] (1/2) Epoch 19, batch 37050, giga_loss[loss=0.26, simple_loss=0.3378, pruned_loss=0.09109, over 28496.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3364, pruned_loss=0.09432, over 5670931.69 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3554, pruned_loss=0.1134, over 5687527.30 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3338, pruned_loss=0.09233, over 5671622.44 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:01:54,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:02:12,052 INFO [train.py:968] (1/2) Epoch 19, batch 37100, giga_loss[loss=0.2469, simple_loss=0.3304, pruned_loss=0.08174, over 28695.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3369, pruned_loss=0.09426, over 5684276.78 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3556, pruned_loss=0.1134, over 5689658.79 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3343, pruned_loss=0.09224, over 5682444.85 frames. ], batch size: 242, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:02:28,498 INFO [optim.py:369] (1/2) 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 03:02:55,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6287, 1.8691, 1.2939, 1.4319], device='cuda:1'), covar=tensor([0.0971, 0.0623, 0.0994, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0445, 0.0516, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 03:02:55,994 INFO [train.py:968] (1/2) Epoch 19, batch 37150, giga_loss[loss=0.2949, simple_loss=0.3599, pruned_loss=0.1149, over 28283.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3364, pruned_loss=0.09413, over 5691418.07 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3556, pruned_loss=0.1133, over 5692627.30 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3341, pruned_loss=0.09234, over 5687455.34 frames. ], batch size: 368, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:03:08,643 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 37200, giga_loss[loss=0.2423, simple_loss=0.3158, pruned_loss=0.08437, over 28579.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3352, pruned_loss=0.09369, over 5691810.74 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3566, pruned_loss=0.1136, over 5693215.16 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3318, pruned_loss=0.09141, over 5688160.43 frames. ], batch size: 71, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:03:47,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=859800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:03:53,284 INFO [optim.py:369] (1/2) 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 03:04:16,943 INFO [train.py:968] (1/2) Epoch 19, batch 37250, libri_loss[loss=0.3353, simple_loss=0.4112, pruned_loss=0.1297, over 29542.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3333, pruned_loss=0.0929, over 5692187.22 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3574, pruned_loss=0.1141, over 5682683.21 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3296, pruned_loss=0.09047, over 5698140.22 frames. ], batch size: 89, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:04:23,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1629, 1.3344, 1.3337, 1.0837], device='cuda:1'), covar=tensor([0.2655, 0.2378, 0.1512, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.1912, 0.1818, 0.1749, 0.1916], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 03:04:59,009 INFO [train.py:968] (1/2) Epoch 19, batch 37300, giga_loss[loss=0.2559, simple_loss=0.325, pruned_loss=0.09339, over 28973.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3311, pruned_loss=0.0921, over 5701317.42 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3579, pruned_loss=0.1144, over 5683760.43 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3275, pruned_loss=0.08959, over 5705135.48 frames. ], batch size: 128, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:05:15,312 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 37350, giga_loss[loss=0.2356, simple_loss=0.3054, pruned_loss=0.08294, over 28801.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3304, pruned_loss=0.09204, over 5683604.75 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3591, pruned_loss=0.1151, over 5668714.57 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3259, pruned_loss=0.08905, over 5701912.47 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:05:43,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=859943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:05:45,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=859946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:06:07,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=859975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:06:20,516 INFO [train.py:968] (1/2) Epoch 19, batch 37400, giga_loss[loss=0.2324, simple_loss=0.3219, pruned_loss=0.0714, over 28853.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.328, pruned_loss=0.09068, over 5689837.47 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3599, pruned_loss=0.1154, over 5671116.10 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3231, pruned_loss=0.08756, over 5702587.67 frames. ], batch size: 174, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:06:22,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4579, 1.8072, 1.4669, 1.6420], device='cuda:1'), covar=tensor([0.0763, 0.0308, 0.0325, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 03:06:36,349 INFO [optim.py:369] (1/2) 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 03:06:51,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3181, 1.2030, 3.9845, 3.2729], device='cuda:1'), covar=tensor([0.1722, 0.2871, 0.0430, 0.0895], device='cuda:1'), in_proj_covar=tensor([0.0735, 0.0634, 0.0927, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 03:07:00,827 INFO [train.py:968] (1/2) Epoch 19, batch 37450, giga_loss[loss=0.2839, simple_loss=0.3459, pruned_loss=0.1109, over 28396.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3291, pruned_loss=0.0919, over 5686930.70 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3622, pruned_loss=0.1167, over 5657302.83 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3219, pruned_loss=0.08729, over 5711405.78 frames. ], batch size: 65, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:07:33,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5000, 3.5600, 1.5839, 1.5669], device='cuda:1'), covar=tensor([0.1020, 0.0331, 0.0917, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0538, 0.0373, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 03:07:40,683 INFO [train.py:968] (1/2) Epoch 19, batch 37500, giga_loss[loss=0.2268, simple_loss=0.3008, pruned_loss=0.07639, over 28859.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3269, pruned_loss=0.09081, over 5700974.87 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3629, pruned_loss=0.1172, over 5662190.53 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.32, pruned_loss=0.08632, over 5716674.99 frames. ], batch size: 199, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:07:57,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6835, 1.8502, 1.5283, 1.7446], device='cuda:1'), covar=tensor([0.2557, 0.2734, 0.3043, 0.2494], device='cuda:1'), in_proj_covar=tensor([0.1468, 0.1064, 0.1303, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 03:07:58,132 INFO [optim.py:369] (1/2) 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 03:07:59,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:08:13,918 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 37550, giga_loss[loss=0.2362, simple_loss=0.3129, pruned_loss=0.07974, over 28465.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3236, pruned_loss=0.08856, over 5709693.21 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3632, pruned_loss=0.1172, over 5662592.60 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3174, pruned_loss=0.08465, over 5722042.58 frames. ], batch size: 65, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:08:49,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-10 03:08:52,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5175, 1.6070, 1.7554, 1.3316], device='cuda:1'), covar=tensor([0.1913, 0.2600, 0.1508, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0699, 0.0937, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 03:09:04,930 INFO [train.py:968] (1/2) Epoch 19, batch 37600, libri_loss[loss=0.2981, simple_loss=0.378, pruned_loss=0.1091, over 29473.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3249, pruned_loss=0.08933, over 5705635.98 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3642, pruned_loss=0.1176, over 5659617.78 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3172, pruned_loss=0.08457, over 5720468.50 frames. ], batch size: 85, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:09:12,034 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:09:20,083 INFO [optim.py:369] (1/2) 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 03:09:47,283 INFO [train.py:968] (1/2) Epoch 19, batch 37650, giga_loss[loss=0.2613, simple_loss=0.3319, pruned_loss=0.09536, over 28820.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3285, pruned_loss=0.09138, over 5703734.95 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3641, pruned_loss=0.1173, over 5663282.11 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3218, pruned_loss=0.08733, over 5713053.97 frames. ], batch size: 112, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:10:16,171 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=860270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:10:19,440 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 37700, giga_loss[loss=0.3093, simple_loss=0.3756, pruned_loss=0.1215, over 28871.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3335, pruned_loss=0.09445, over 5701389.88 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3642, pruned_loss=0.1172, over 5667213.25 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3277, pruned_loss=0.09097, over 5705854.93 frames. ], batch size: 227, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:10:49,508 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=860302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:10:55,420 INFO [optim.py:369] (1/2) 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 03:11:25,252 INFO [train.py:968] (1/2) Epoch 19, batch 37750, giga_loss[loss=0.2831, simple_loss=0.3489, pruned_loss=0.1086, over 28749.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3431, pruned_loss=0.1012, over 5696458.26 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3643, pruned_loss=0.1172, over 5670431.50 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3379, pruned_loss=0.09822, over 5697597.29 frames. ], batch size: 99, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:12:16,622 INFO [train.py:968] (1/2) Epoch 19, batch 37800, giga_loss[loss=0.26, simple_loss=0.3467, pruned_loss=0.0867, over 28850.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.347, pruned_loss=0.103, over 5678373.85 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3639, pruned_loss=0.117, over 5672004.99 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.343, pruned_loss=0.1007, over 5677911.70 frames. ], batch size: 186, lr: 1.67e-03, grad_scale: 8.0
+2023-03-10 03:12:34,612 INFO [optim.py:369] (1/2) 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 03:12:44,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-10 03:12:59,661 INFO [train.py:968] (1/2) Epoch 19, batch 37850, giga_loss[loss=0.281, simple_loss=0.3524, pruned_loss=0.1048, over 28778.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3514, pruned_loss=0.1046, over 5688830.25 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3641, pruned_loss=0.1169, over 5679972.20 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3476, pruned_loss=0.1023, over 5681544.57 frames. ], batch size: 92, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:13:41,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:13:44,759 INFO [train.py:968] (1/2) Epoch 19, batch 37900, giga_loss[loss=0.2958, simple_loss=0.3746, pruned_loss=0.1085, over 28576.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3553, pruned_loss=0.1065, over 5672359.52 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3643, pruned_loss=0.117, over 5676775.27 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3515, pruned_loss=0.1041, over 5669595.22 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:13:52,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860498.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:13:55,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=860502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:14:03,094 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 37950, giga_loss[loss=0.253, simple_loss=0.3349, pruned_loss=0.08551, over 28510.00 frames. ], tot_loss[loss=0.2887, simple_loss=0.3591, pruned_loss=0.1092, over 5673515.21 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3637, pruned_loss=0.1166, over 5679394.64 frames. ], giga_tot_loss[loss=0.2855, simple_loss=0.3565, pruned_loss=0.1073, over 5668909.49 frames. ], batch size: 336, lr: 1.67e-03, grad_scale: 2.0
+2023-03-10 03:14:53,690 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 38000, giga_loss[loss=0.309, simple_loss=0.3756, pruned_loss=0.1212, over 28705.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3537, pruned_loss=0.105, over 5681117.39 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3633, pruned_loss=0.1165, over 5682406.82 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3519, pruned_loss=0.1035, over 5674743.09 frames. ], batch size: 60, lr: 1.67e-03, grad_scale: 4.0
+2023-03-10 03:15:24,287 INFO [optim.py:369] (1/2) 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 03:15:30,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8762, 2.0952, 1.8418, 1.8621], device='cuda:1'), covar=tensor([0.2053, 0.2376, 0.2298, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0744, 0.0708, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 03:15:38,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=860628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:15:41,439 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 38050, giga_loss[loss=0.236, simple_loss=0.3206, pruned_loss=0.07571, over 29079.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3532, pruned_loss=0.1042, over 5684242.16 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3636, pruned_loss=0.1169, over 5680494.48 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3511, pruned_loss=0.1022, over 5680874.62 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:16:06,362 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=860660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:16:28,190 INFO [train.py:968] (1/2) Epoch 19, batch 38100, giga_loss[loss=0.2589, simple_loss=0.3399, pruned_loss=0.08897, over 28695.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.352, pruned_loss=0.1033, over 5689988.72 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3634, pruned_loss=0.1169, over 5688078.92 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3501, pruned_loss=0.1012, over 5680264.02 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:16:38,832 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:16:54,283 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=860716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:16:56,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=860719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:17:13,210 INFO [train.py:968] (1/2) Epoch 19, batch 38150, giga_loss[loss=0.2736, simple_loss=0.3534, pruned_loss=0.09692, over 28949.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3529, pruned_loss=0.1034, over 5695403.94 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3631, pruned_loss=0.1168, over 5688218.52 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3516, pruned_loss=0.1018, over 5687696.11 frames. ], batch size: 145, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:17:23,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=860748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:17:55,947 INFO [train.py:968] (1/2) Epoch 19, batch 38200, giga_loss[loss=0.3819, simple_loss=0.4215, pruned_loss=0.1711, over 26452.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3542, pruned_loss=0.1041, over 5688926.32 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3631, pruned_loss=0.1167, over 5687907.16 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3528, pruned_loss=0.1023, over 5683267.54 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:18:03,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3560, 3.5059, 1.5659, 1.4752], device='cuda:1'), covar=tensor([0.0963, 0.0322, 0.0827, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0538, 0.0372, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 03:18:09,066 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 03:18:18,344 INFO [optim.py:369] (1/2) 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 03:18:36,494 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-10 03:18:43,036 INFO [train.py:968] (1/2) Epoch 19, batch 38250, giga_loss[loss=0.3045, simple_loss=0.3653, pruned_loss=0.1218, over 28662.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3556, pruned_loss=0.1051, over 5693771.41 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.363, pruned_loss=0.1166, over 5691969.45 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3544, pruned_loss=0.1036, over 5685721.01 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:18:48,252 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3591, 3.1957, 1.5476, 1.4822], device='cuda:1'), covar=tensor([0.1013, 0.0314, 0.0862, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0538, 0.0372, 0.0415], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 03:19:16,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860873.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:19:19,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=860877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:19:29,832 INFO [train.py:968] (1/2) Epoch 19, batch 38300, giga_loss[loss=0.3196, simple_loss=0.3739, pruned_loss=0.1326, over 28723.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3567, pruned_loss=0.1062, over 5698628.47 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3632, pruned_loss=0.1167, over 5693082.60 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3556, pruned_loss=0.1049, over 5691392.51 frames. ], batch size: 99, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:19:51,467 INFO [optim.py:369] (1/2) 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 03:20:11,649 INFO [train.py:968] (1/2) Epoch 19, batch 38350, giga_loss[loss=0.3457, simple_loss=0.3909, pruned_loss=0.1502, over 26546.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3582, pruned_loss=0.1076, over 5696574.06 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3636, pruned_loss=0.1167, over 5698675.82 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3567, pruned_loss=0.1062, over 5685788.57 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:20:53,216 INFO [train.py:968] (1/2) Epoch 19, batch 38400, giga_loss[loss=0.2798, simple_loss=0.3539, pruned_loss=0.1029, over 28941.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.359, pruned_loss=0.108, over 5704271.17 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3636, pruned_loss=0.1165, over 5700675.76 frames. ], giga_tot_loss[loss=0.2856, simple_loss=0.3577, pruned_loss=0.1068, over 5694174.34 frames. ], batch size: 112, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:21:13,594 INFO [optim.py:369] (1/2) 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 03:21:17,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:21:20,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861019.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:21:21,617 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:21:23,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:21:33,422 INFO [train.py:968] (1/2) Epoch 19, batch 38450, giga_loss[loss=0.2767, simple_loss=0.3513, pruned_loss=0.1011, over 28539.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3574, pruned_loss=0.1058, over 5713892.41 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.363, pruned_loss=0.1162, over 5704689.80 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3567, pruned_loss=0.105, over 5702062.94 frames. ], batch size: 85, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:21:42,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 03:21:44,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:22:02,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:22:16,659 INFO [train.py:968] (1/2) Epoch 19, batch 38500, giga_loss[loss=0.2763, simple_loss=0.3585, pruned_loss=0.097, over 28943.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3584, pruned_loss=0.1055, over 5716597.24 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3635, pruned_loss=0.1165, over 5711074.58 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.3573, pruned_loss=0.1042, over 5701538.67 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:22:20,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:22:35,646 INFO [optim.py:369] (1/2) 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 03:22:44,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9838, 1.2980, 1.1262, 0.2257], device='cuda:1'), covar=tensor([0.3771, 0.2901, 0.4363, 0.6166], device='cuda:1'), in_proj_covar=tensor([0.1691, 0.1590, 0.1561, 0.1381], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 03:22:56,206 INFO [train.py:968] (1/2) Epoch 19, batch 38550, giga_loss[loss=0.2746, simple_loss=0.3553, pruned_loss=0.09693, over 28919.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3574, pruned_loss=0.1045, over 5709004.51 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3637, pruned_loss=0.1168, over 5702159.66 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3563, pruned_loss=0.1031, over 5704755.73 frames. ], batch size: 145, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:23:00,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5541, 2.5401, 2.4427, 2.2858], device='cuda:1'), covar=tensor([0.1800, 0.2304, 0.2042, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0743, 0.0708, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 03:23:22,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3352, 3.1362, 2.9757, 1.4100], device='cuda:1'), covar=tensor([0.0899, 0.1087, 0.0957, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.1163, 0.1084, 0.0920, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 03:23:42,663 INFO [train.py:968] (1/2) Epoch 19, batch 38600, giga_loss[loss=0.2571, simple_loss=0.337, pruned_loss=0.08866, over 28348.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3551, pruned_loss=0.1035, over 5706651.14 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3637, pruned_loss=0.1167, over 5702371.87 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3541, pruned_loss=0.1022, over 5703145.91 frames. ], batch size: 77, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:24:00,821 INFO [optim.py:369] (1/2) 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 03:24:02,918 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:24:05,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:24:22,410 INFO [train.py:968] (1/2) Epoch 19, batch 38650, giga_loss[loss=0.2955, simple_loss=0.3629, pruned_loss=0.114, over 28938.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3532, pruned_loss=0.1024, over 5718453.11 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3639, pruned_loss=0.1167, over 5710024.54 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3519, pruned_loss=0.1009, over 5709009.44 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:24:29,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6481, 1.9563, 1.5690, 1.9389], device='cuda:1'), covar=tensor([0.2620, 0.2645, 0.2992, 0.2269], device='cuda:1'), in_proj_covar=tensor([0.1462, 0.1065, 0.1300, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 03:24:31,140 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:25:02,293 INFO [train.py:968] (1/2) Epoch 19, batch 38700, giga_loss[loss=0.2778, simple_loss=0.3563, pruned_loss=0.09972, over 29037.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3507, pruned_loss=0.1011, over 5717692.12 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3636, pruned_loss=0.1165, over 5708968.13 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3496, pruned_loss=0.09973, over 5710949.58 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:25:24,558 INFO [optim.py:369] (1/2) 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 03:25:43,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4115, 2.6433, 1.7655, 2.1537], device='cuda:1'), covar=tensor([0.0906, 0.0650, 0.1025, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0385, 0.0442, 0.0513, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 03:25:44,855 INFO [train.py:968] (1/2) Epoch 19, batch 38750, giga_loss[loss=0.2816, simple_loss=0.3598, pruned_loss=0.1017, over 28944.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3522, pruned_loss=0.1025, over 5705004.01 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3639, pruned_loss=0.1167, over 5697236.65 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3509, pruned_loss=0.101, over 5710459.99 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:26:26,505 INFO [train.py:968] (1/2) Epoch 19, batch 38800, giga_loss[loss=0.29, simple_loss=0.3639, pruned_loss=0.1081, over 28887.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1023, over 5709482.11 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3635, pruned_loss=0.1165, over 5700989.81 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3513, pruned_loss=0.1011, over 5710768.82 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:26:42,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3000, 3.8968, 1.4528, 1.4627], device='cuda:1'), covar=tensor([0.1058, 0.0217, 0.1023, 0.1461], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0537, 0.0372, 0.0414], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 03:26:44,003 INFO [optim.py:369] (1/2) 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 03:26:50,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:27:06,233 INFO [train.py:968] (1/2) Epoch 19, batch 38850, giga_loss[loss=0.3499, simple_loss=0.4088, pruned_loss=0.1455, over 28609.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3512, pruned_loss=0.1008, over 5704145.49 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3635, pruned_loss=0.1165, over 5700989.81 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3505, pruned_loss=0.09986, over 5705146.96 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:27:22,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:27:28,024 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 38900, giga_loss[loss=0.2436, simple_loss=0.3248, pruned_loss=0.08117, over 28243.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3507, pruned_loss=0.1002, over 5714970.29 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3638, pruned_loss=0.1164, over 5706207.58 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3496, pruned_loss=0.09914, over 5711260.38 frames. ], batch size: 77, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:27:45,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 03:28:06,401 INFO [optim.py:369] (1/2) 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 03:28:23,576 INFO [train.py:968] (1/2) Epoch 19, batch 38950, giga_loss[loss=0.2722, simple_loss=0.3517, pruned_loss=0.09636, over 28503.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3495, pruned_loss=0.1001, over 5706563.06 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5700964.70 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3486, pruned_loss=0.09899, over 5708966.59 frames. ], batch size: 65, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:28:39,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5852, 2.1227, 1.5440, 0.9369], device='cuda:1'), covar=tensor([0.5659, 0.2572, 0.3668, 0.5886], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1584, 0.1562, 0.1381], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 03:28:43,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9662, 2.9809, 1.9632, 0.9264], device='cuda:1'), covar=tensor([0.7267, 0.2403, 0.3611, 0.6930], device='cuda:1'), in_proj_covar=tensor([0.1693, 0.1583, 0.1562, 0.1380], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 03:29:05,162 INFO [train.py:968] (1/2) Epoch 19, batch 39000, giga_loss[loss=0.2772, simple_loss=0.3507, pruned_loss=0.1019, over 28324.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.347, pruned_loss=0.0992, over 5702819.85 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3635, pruned_loss=0.1164, over 5700754.63 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3459, pruned_loss=0.09784, over 5705065.19 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:29:05,162 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 03:29:11,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2415, 1.4197, 1.3670, 1.2241], device='cuda:1'), covar=tensor([0.2479, 0.2285, 0.1546, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.1915, 0.1837, 0.1765, 0.1913], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 03:29:14,767 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 03:29:34,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:35,718 INFO [optim.py:369] (1/2) 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 03:29:36,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:37,374 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:37,624 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 03:29:52,708 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:29:54,597 INFO [train.py:968] (1/2) Epoch 19, batch 39050, giga_loss[loss=0.2645, simple_loss=0.339, pruned_loss=0.09497, over 28731.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3436, pruned_loss=0.09751, over 5690240.02 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3638, pruned_loss=0.1167, over 5684594.47 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3424, pruned_loss=0.09603, over 5705972.77 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:30:00,190 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:30:29,691 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:30:36,405 INFO [train.py:968] (1/2) Epoch 19, batch 39100, giga_loss[loss=0.2739, simple_loss=0.353, pruned_loss=0.09741, over 27987.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3423, pruned_loss=0.0966, over 5692837.68 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3639, pruned_loss=0.1167, over 5684706.31 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.341, pruned_loss=0.0952, over 5705066.87 frames. ], batch size: 412, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:31:00,263 INFO [optim.py:369] (1/2) 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 03:31:19,670 INFO [train.py:968] (1/2) Epoch 19, batch 39150, libri_loss[loss=0.2623, simple_loss=0.3335, pruned_loss=0.09552, over 29585.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3416, pruned_loss=0.0968, over 5693377.84 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3639, pruned_loss=0.1167, over 5689101.08 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3401, pruned_loss=0.09537, over 5699128.68 frames. ], batch size: 74, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:31:41,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-10 03:31:45,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-10 03:31:58,832 INFO [train.py:968] (1/2) Epoch 19, batch 39200, giga_loss[loss=0.2688, simple_loss=0.3331, pruned_loss=0.1022, over 28444.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3397, pruned_loss=0.09615, over 5704036.03 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3642, pruned_loss=0.1168, over 5695330.02 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3376, pruned_loss=0.09424, over 5702956.95 frames. ], batch size: 65, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:32:04,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:32:14,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-10 03:32:19,470 INFO [optim.py:369] (1/2) 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 03:32:37,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:32:38,173 INFO [train.py:968] (1/2) Epoch 19, batch 39250, giga_loss[loss=0.2149, simple_loss=0.2879, pruned_loss=0.07095, over 28685.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3382, pruned_loss=0.0957, over 5695006.27 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.365, pruned_loss=0.1174, over 5681581.49 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3349, pruned_loss=0.09296, over 5707034.61 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:32:57,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:33:17,365 INFO [train.py:968] (1/2) Epoch 19, batch 39300, giga_loss[loss=0.24, simple_loss=0.3195, pruned_loss=0.08025, over 28646.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3351, pruned_loss=0.09405, over 5685766.79 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3649, pruned_loss=0.1172, over 5673191.90 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3319, pruned_loss=0.09149, over 5703845.09 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:33:37,995 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:33:59,852 INFO [train.py:968] (1/2) Epoch 19, batch 39350, libri_loss[loss=0.3156, simple_loss=0.3846, pruned_loss=0.1233, over 29520.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3343, pruned_loss=0.09363, over 5701426.23 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3654, pruned_loss=0.1177, over 5682374.73 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.33, pruned_loss=0.09018, over 5708122.86 frames. ], batch size: 89, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:34:00,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:02,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=861941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:15,365 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:18,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=861961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:27,713 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=861970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:35,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=861980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:34:38,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 39400, giga_loss[loss=0.3251, simple_loss=0.3925, pruned_loss=0.1289, over 27522.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3361, pruned_loss=0.09395, over 5691990.86 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3656, pruned_loss=0.1178, over 5676014.06 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.332, pruned_loss=0.09077, over 5702879.53 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:34:46,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=861990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:05,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:06,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:08,588 INFO [optim.py:369] (1/2) 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 03:35:30,910 INFO [train.py:968] (1/2) Epoch 19, batch 39450, giga_loss[loss=0.3462, simple_loss=0.3958, pruned_loss=0.1483, over 26777.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3398, pruned_loss=0.09588, over 5681828.01 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3658, pruned_loss=0.1179, over 5674890.14 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3362, pruned_loss=0.09305, over 5691677.34 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:35:45,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:49,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:53,826 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:35:55,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5863, 2.2891, 1.6828, 0.7389], device='cuda:1'), covar=tensor([0.6199, 0.2772, 0.4284, 0.6752], device='cuda:1'), in_proj_covar=tensor([0.1707, 0.1602, 0.1580, 0.1393], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 03:36:13,122 INFO [train.py:968] (1/2) Epoch 19, batch 39500, giga_loss[loss=0.3022, simple_loss=0.3769, pruned_loss=0.1137, over 28955.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3432, pruned_loss=0.09712, over 5674990.32 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3664, pruned_loss=0.1183, over 5666926.22 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3391, pruned_loss=0.09388, over 5690443.76 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:36:18,084 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:36:20,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1795, 0.9267, 0.9789, 1.3952], device='cuda:1'), covar=tensor([0.0764, 0.0349, 0.0367, 0.0899], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0115, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0104], device='cuda:1')
+2023-03-10 03:36:22,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-10 03:36:35,746 INFO [optim.py:369] (1/2) 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 03:36:54,108 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:36:56,940 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 39550, giga_loss[loss=0.2481, simple_loss=0.3346, pruned_loss=0.08075, over 29066.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3438, pruned_loss=0.09705, over 5676076.03 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3664, pruned_loss=0.1184, over 5664063.94 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3399, pruned_loss=0.09386, over 5691296.59 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:37:09,020 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:12,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:19,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:22,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:35,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:37,770 INFO [train.py:968] (1/2) Epoch 19, batch 39600, giga_loss[loss=0.2619, simple_loss=0.3449, pruned_loss=0.08944, over 28718.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3424, pruned_loss=0.09615, over 5688736.61 frames. ], libri_tot_loss[loss=0.301, simple_loss=0.3657, pruned_loss=0.1181, over 5669638.10 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.339, pruned_loss=0.09297, over 5696494.76 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:37:48,134 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:50,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:50,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:52,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:37:58,457 INFO [optim.py:369] (1/2) 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 03:38:11,981 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:14,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:16,062 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 39650, giga_loss[loss=0.2496, simple_loss=0.3222, pruned_loss=0.08854, over 28432.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3424, pruned_loss=0.09643, over 5694554.39 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1178, over 5676448.03 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3391, pruned_loss=0.09349, over 5695212.43 frames. ], batch size: 60, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:38:41,186 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-10 03:38:43,396 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:38:59,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:39:02,901 INFO [train.py:968] (1/2) Epoch 19, batch 39700, giga_loss[loss=0.2406, simple_loss=0.3203, pruned_loss=0.08049, over 28965.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3425, pruned_loss=0.09647, over 5708370.85 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3654, pruned_loss=0.1177, over 5679969.71 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3397, pruned_loss=0.09401, over 5706203.47 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:39:25,892 INFO [optim.py:369] (1/2) 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 03:39:40,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:39:42,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:39:43,716 INFO [train.py:968] (1/2) Epoch 19, batch 39750, giga_loss[loss=0.3436, simple_loss=0.3993, pruned_loss=0.1439, over 28857.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3446, pruned_loss=0.09809, over 5711555.14 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1178, over 5677382.45 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3411, pruned_loss=0.095, over 5712757.10 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:39:56,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3509, 1.6359, 1.3432, 0.9792], device='cuda:1'), covar=tensor([0.2510, 0.2479, 0.2925, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.1466, 0.1063, 0.1300, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 03:40:04,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 03:40:21,356 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:40:21,442 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:40:24,403 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 39800, giga_loss[loss=0.2761, simple_loss=0.3436, pruned_loss=0.1043, over 28721.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3478, pruned_loss=0.09923, over 5707909.95 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1179, over 5676740.53 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.0967, over 5709573.64 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:40:48,907 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:40:59,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6406, 1.6658, 1.2825, 1.2678], device='cuda:1'), covar=tensor([0.0874, 0.0658, 0.1041, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0383, 0.0441, 0.0510, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 03:41:04,087 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:05,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:11,892 INFO [train.py:968] (1/2) Epoch 19, batch 39850, giga_loss[loss=0.2632, simple_loss=0.343, pruned_loss=0.09173, over 28876.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3504, pruned_loss=0.1006, over 5712117.88 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3663, pruned_loss=0.1182, over 5680210.21 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3474, pruned_loss=0.09809, over 5710839.77 frames. ], batch size: 174, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:41:12,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:30,658 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:35,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:44,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:46,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:46,133 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:49,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:41:52,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3833, 1.7258, 1.3338, 1.6357], device='cuda:1'), covar=tensor([0.2591, 0.2555, 0.3090, 0.2271], device='cuda:1'), in_proj_covar=tensor([0.1468, 0.1064, 0.1301, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 03:41:54,470 INFO [train.py:968] (1/2) Epoch 19, batch 39900, giga_loss[loss=0.253, simple_loss=0.331, pruned_loss=0.08755, over 28968.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1011, over 5703556.66 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3668, pruned_loss=0.1185, over 5674646.36 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3483, pruned_loss=0.0985, over 5708120.34 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:42:04,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-10 03:42:11,195 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:42:13,100 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 19, batch 39950, giga_loss[loss=0.2853, simple_loss=0.371, pruned_loss=0.09982, over 28895.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3532, pruned_loss=0.1017, over 5701750.49 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1185, over 5673018.73 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3501, pruned_loss=0.09924, over 5707464.34 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:42:40,896 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:42:59,497 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5777, 1.7844, 1.2639, 1.3976], device='cuda:1'), covar=tensor([0.0863, 0.0585, 0.1032, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0384, 0.0442, 0.0511, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 03:43:11,658 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:13,021 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-10 03:43:13,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:16,613 INFO [train.py:968] (1/2) Epoch 19, batch 40000, giga_loss[loss=0.3162, simple_loss=0.3888, pruned_loss=0.1218, over 28887.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3531, pruned_loss=0.1019, over 5703528.15 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3666, pruned_loss=0.1181, over 5678447.68 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.09989, over 5703705.11 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:43:33,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:36,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:37,168 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:43:54,919 INFO [train.py:968] (1/2) Epoch 19, batch 40050, libri_loss[loss=0.3232, simple_loss=0.3866, pruned_loss=0.1299, over 29226.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3528, pruned_loss=0.1021, over 5704740.88 frames. ], libri_tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1184, over 5675610.43 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.35, pruned_loss=0.09967, over 5708838.55 frames. ], batch size: 94, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:43:57,143 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:43:57,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:44:36,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:44:36,907 INFO [train.py:968] (1/2) Epoch 19, batch 40100, giga_loss[loss=0.2483, simple_loss=0.3314, pruned_loss=0.08264, over 28763.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.0999, over 5713809.65 frames. ], libri_tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.1181, over 5679134.42 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3465, pruned_loss=0.09795, over 5714318.74 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:44:39,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:44:40,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2624, 4.0781, 3.9311, 2.5805], device='cuda:1'), covar=tensor([0.0658, 0.0879, 0.0864, 0.1595], device='cuda:1'), in_proj_covar=tensor([0.1166, 0.1084, 0.0922, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 03:44:48,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 03:45:00,337 INFO [optim.py:369] (1/2) 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 03:45:03,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:45:04,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2921, 1.5968, 1.2955, 1.1032], device='cuda:1'), covar=tensor([0.2602, 0.2621, 0.3064, 0.2366], device='cuda:1'), in_proj_covar=tensor([0.1462, 0.1062, 0.1298, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 03:45:19,918 INFO [train.py:968] (1/2) Epoch 19, batch 40150, giga_loss[loss=0.2328, simple_loss=0.313, pruned_loss=0.07628, over 28712.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3462, pruned_loss=0.09917, over 5698958.44 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3666, pruned_loss=0.1183, over 5668475.50 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.344, pruned_loss=0.09694, over 5708564.74 frames. ], batch size: 119, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:45:33,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=862754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:45:57,428 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:46:00,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=862787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:46:00,452 INFO [train.py:968] (1/2) Epoch 19, batch 40200, giga_loss[loss=0.2564, simple_loss=0.3434, pruned_loss=0.08473, over 28880.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3459, pruned_loss=0.09849, over 5702888.28 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.367, pruned_loss=0.1185, over 5665272.49 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3432, pruned_loss=0.09598, over 5714112.37 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:46:18,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 03:46:22,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=862816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:46:22,562 INFO [optim.py:369] (1/2) 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 03:46:40,424 INFO [train.py:968] (1/2) Epoch 19, batch 40250, giga_loss[loss=0.2729, simple_loss=0.356, pruned_loss=0.09491, over 28733.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3484, pruned_loss=0.0985, over 5709922.79 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.367, pruned_loss=0.1186, over 5673168.25 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3457, pruned_loss=0.09573, over 5713235.16 frames. ], batch size: 242, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:47:26,063 INFO [train.py:968] (1/2) Epoch 19, batch 40300, giga_loss[loss=0.2784, simple_loss=0.3516, pruned_loss=0.1026, over 28918.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3471, pruned_loss=0.09758, over 5704575.89 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1186, over 5675396.03 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3448, pruned_loss=0.0953, over 5705649.56 frames. ], batch size: 186, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:47:28,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2907, 1.3201, 3.7765, 3.1487], device='cuda:1'), covar=tensor([0.1600, 0.2748, 0.0453, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0735, 0.0631, 0.0927, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 03:47:32,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=862897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:47:34,541 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:47:57,717 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 40350, giga_loss[loss=0.3, simple_loss=0.3575, pruned_loss=0.1213, over 28791.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3467, pruned_loss=0.09872, over 5712397.86 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3668, pruned_loss=0.1186, over 5679850.34 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3446, pruned_loss=0.09653, over 5709994.33 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 03:48:06,763 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=862938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:48:11,389 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-10 03:48:45,410 INFO [train.py:968] (1/2) Epoch 19, batch 40400, giga_loss[loss=0.3017, simple_loss=0.3683, pruned_loss=0.1176, over 28343.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.346, pruned_loss=0.09964, over 5711911.82 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3667, pruned_loss=0.1183, over 5685458.84 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.344, pruned_loss=0.0977, over 5705595.39 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:49:07,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5930, 4.4423, 4.1774, 2.0587], device='cuda:1'), covar=tensor([0.0555, 0.0704, 0.0712, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.1172, 0.1088, 0.0924, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 03:49:08,061 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 03:49:12,323 INFO [optim.py:369] (1/2) 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 03:49:31,421 INFO [train.py:968] (1/2) Epoch 19, batch 40450, giga_loss[loss=0.2626, simple_loss=0.3346, pruned_loss=0.09532, over 28264.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3446, pruned_loss=0.09978, over 5713071.80 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3668, pruned_loss=0.1182, over 5687421.82 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3426, pruned_loss=0.09803, over 5706586.95 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:50:12,268 INFO [train.py:968] (1/2) Epoch 19, batch 40500, giga_loss[loss=0.3135, simple_loss=0.378, pruned_loss=0.1245, over 27600.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3425, pruned_loss=0.09881, over 5718473.03 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3668, pruned_loss=0.1182, over 5688807.42 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3407, pruned_loss=0.09728, over 5712331.34 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:50:36,196 INFO [optim.py:369] (1/2) 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 03:50:52,517 INFO [train.py:968] (1/2) Epoch 19, batch 40550, giga_loss[loss=0.2371, simple_loss=0.316, pruned_loss=0.07912, over 28907.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3416, pruned_loss=0.09861, over 5726552.58 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3676, pruned_loss=0.1189, over 5695360.52 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3386, pruned_loss=0.0962, over 5716514.48 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:51:26,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4617, 1.5891, 1.4087, 1.5793], device='cuda:1'), covar=tensor([0.0754, 0.0327, 0.0347, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 03:51:33,104 INFO [train.py:968] (1/2) Epoch 19, batch 40600, giga_loss[loss=0.3185, simple_loss=0.3742, pruned_loss=0.1314, over 27671.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3377, pruned_loss=0.09688, over 5726608.29 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3676, pruned_loss=0.1188, over 5696046.15 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3349, pruned_loss=0.09463, over 5718579.72 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:51:44,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2373, 3.4621, 2.4105, 1.2466], device='cuda:1'), covar=tensor([0.6592, 0.2711, 0.3014, 0.6016], device='cuda:1'), in_proj_covar=tensor([0.1701, 0.1596, 0.1571, 0.1385], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 03:51:56,533 INFO [optim.py:369] (1/2) 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 03:52:16,037 INFO [train.py:968] (1/2) Epoch 19, batch 40650, giga_loss[loss=0.2574, simple_loss=0.3315, pruned_loss=0.09161, over 28745.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3335, pruned_loss=0.09467, over 5726296.85 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3672, pruned_loss=0.1186, over 5702361.44 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3306, pruned_loss=0.09228, over 5715163.72 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:52:46,410 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863277.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:52:55,626 INFO [train.py:968] (1/2) Epoch 19, batch 40700, giga_loss[loss=0.2201, simple_loss=0.2971, pruned_loss=0.07153, over 28592.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3328, pruned_loss=0.09409, over 5723717.83 frames. ], libri_tot_loss[loss=0.3016, simple_loss=0.3666, pruned_loss=0.1183, over 5704451.41 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3299, pruned_loss=0.09171, over 5713352.05 frames. ], batch size: 60, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:53:17,024 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 19, batch 40750, giga_loss[loss=0.2463, simple_loss=0.3203, pruned_loss=0.08612, over 28430.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3361, pruned_loss=0.09543, over 5713394.91 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1186, over 5697805.17 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3331, pruned_loss=0.09299, over 5711654.16 frames. ], batch size: 78, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:53:55,732 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9436, 5.1063, 2.1145, 2.1975], device='cuda:1'), covar=tensor([0.0887, 0.0292, 0.0869, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0543, 0.0375, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 03:54:18,310 INFO [train.py:968] (1/2) Epoch 19, batch 40800, libri_loss[loss=0.3257, simple_loss=0.38, pruned_loss=0.1357, over 19573.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3401, pruned_loss=0.09726, over 5710969.01 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1185, over 5696691.59 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3366, pruned_loss=0.09444, over 5711839.27 frames. ], batch size: 187, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 03:54:26,861 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:54:40,590 INFO [optim.py:369] (1/2) 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 03:54:57,211 INFO [train.py:968] (1/2) Epoch 19, batch 40850, giga_loss[loss=0.2947, simple_loss=0.3798, pruned_loss=0.1048, over 28996.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3445, pruned_loss=0.09912, over 5706018.75 frames. ], libri_tot_loss[loss=0.3027, simple_loss=0.3674, pruned_loss=0.119, over 5690624.54 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3404, pruned_loss=0.09588, over 5711749.84 frames. ], batch size: 174, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:55:07,574 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:55:14,189 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:55:16,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=863459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:55:40,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6718, 1.7982, 1.4955, 1.8179], device='cuda:1'), covar=tensor([0.2725, 0.2907, 0.3227, 0.2790], device='cuda:1'), in_proj_covar=tensor([0.1469, 0.1065, 0.1302, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 03:55:42,542 INFO [train.py:968] (1/2) Epoch 19, batch 40900, giga_loss[loss=0.2825, simple_loss=0.3501, pruned_loss=0.1074, over 28814.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3467, pruned_loss=0.09992, over 5709955.66 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.1189, over 5684932.50 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3432, pruned_loss=0.09709, over 5718998.76 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:55:43,512 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=863488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 03:56:07,138 INFO [optim.py:369] (1/2) 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 03:56:24,579 INFO [train.py:968] (1/2) Epoch 19, batch 40950, giga_loss[loss=0.2522, simple_loss=0.3241, pruned_loss=0.0901, over 24006.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3491, pruned_loss=0.1014, over 5693623.88 frames. ], libri_tot_loss[loss=0.3024, simple_loss=0.3673, pruned_loss=0.1188, over 5676417.08 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3457, pruned_loss=0.09864, over 5708802.29 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:57:16,420 INFO [train.py:968] (1/2) Epoch 19, batch 41000, giga_loss[loss=0.3919, simple_loss=0.4155, pruned_loss=0.1842, over 23616.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3519, pruned_loss=0.1041, over 5687103.52 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3671, pruned_loss=0.1186, over 5678684.99 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3492, pruned_loss=0.1019, over 5697350.33 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:57:45,046 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 41050, giga_loss[loss=0.2913, simple_loss=0.3567, pruned_loss=0.1129, over 28098.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3595, pruned_loss=0.1105, over 5673892.12 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3674, pruned_loss=0.1188, over 5681096.46 frames. ], giga_tot_loss[loss=0.2865, simple_loss=0.3567, pruned_loss=0.1081, over 5680126.54 frames. ], batch size: 77, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:58:18,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=863652.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 03:58:19,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1171, 1.3319, 1.2866, 1.0456], device='cuda:1'), covar=tensor([0.2593, 0.2285, 0.1671, 0.2181], device='cuda:1'), in_proj_covar=tensor([0.1947, 0.1874, 0.1796, 0.1926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 03:58:52,892 INFO [train.py:968] (1/2) Epoch 19, batch 41100, giga_loss[loss=0.3027, simple_loss=0.3691, pruned_loss=0.1181, over 28827.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3662, pruned_loss=0.1152, over 5681324.67 frames. ], libri_tot_loss[loss=0.3021, simple_loss=0.3671, pruned_loss=0.1186, over 5684373.96 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3641, pruned_loss=0.1134, over 5683455.36 frames. ], batch size: 112, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 03:59:20,302 INFO [optim.py:369] (1/2) 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 03:59:38,324 INFO [train.py:968] (1/2) Epoch 19, batch 41150, giga_loss[loss=0.3921, simple_loss=0.4339, pruned_loss=0.1752, over 28283.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3719, pruned_loss=0.12, over 5663562.04 frames. ], libri_tot_loss[loss=0.3023, simple_loss=0.3672, pruned_loss=0.1187, over 5680609.92 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3701, pruned_loss=0.1185, over 5668989.55 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:00:11,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=863775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:00:23,346 INFO [train.py:968] (1/2) Epoch 19, batch 41200, giga_loss[loss=0.3229, simple_loss=0.3968, pruned_loss=0.1245, over 28969.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3785, pruned_loss=0.1258, over 5673905.80 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3671, pruned_loss=0.1187, over 5683752.25 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3774, pruned_loss=0.1247, over 5675098.43 frames. ], batch size: 164, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:00:32,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863795.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:00:34,276 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=863825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:01:03,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=863827.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 04:01:03,915 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=863828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:01:15,523 INFO [train.py:968] (1/2) Epoch 19, batch 41250, giga_loss[loss=0.4028, simple_loss=0.435, pruned_loss=0.1853, over 28595.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3838, pruned_loss=0.1305, over 5659625.24 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3674, pruned_loss=0.1188, over 5685732.45 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3829, pruned_loss=0.1296, over 5658674.68 frames. ], batch size: 307, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:01:28,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9023, 1.8933, 2.0545, 1.6514], device='cuda:1'), covar=tensor([0.1654, 0.2466, 0.1362, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.0881, 0.0693, 0.0926, 0.0826], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 04:01:52,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1915, 1.3696, 1.3251, 1.0906], device='cuda:1'), covar=tensor([0.2361, 0.2170, 0.1466, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.1936, 0.1867, 0.1790, 0.1919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 04:01:56,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3505, 2.4978, 1.2791, 1.5043], device='cuda:1'), covar=tensor([0.0858, 0.0465, 0.0851, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0544, 0.0375, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 04:02:10,283 INFO [train.py:968] (1/2) Epoch 19, batch 41300, giga_loss[loss=0.308, simple_loss=0.3714, pruned_loss=0.1223, over 28920.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3845, pruned_loss=0.1314, over 5666917.99 frames. ], libri_tot_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.1189, over 5686424.59 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.384, pruned_loss=0.1308, over 5665159.34 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:02:46,400 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:02:46,739 INFO [optim.py:369] (1/2) 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 04:02:49,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=863921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:02:54,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4150, 1.1714, 4.4424, 3.4684], device='cuda:1'), covar=tensor([0.1736, 0.2928, 0.0384, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0638, 0.0940, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 04:03:12,745 INFO [train.py:968] (1/2) Epoch 19, batch 41350, libri_loss[loss=0.3697, simple_loss=0.4221, pruned_loss=0.1587, over 29255.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.388, pruned_loss=0.1359, over 5633635.53 frames. ], libri_tot_loss[loss=0.3029, simple_loss=0.3676, pruned_loss=0.1191, over 5689585.32 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3876, pruned_loss=0.1354, over 5629052.12 frames. ], batch size: 94, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:03:24,062 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=863950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:03:43,412 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=863968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:03:46,345 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 41400, giga_loss[loss=0.4168, simple_loss=0.4443, pruned_loss=0.1947, over 28246.00 frames. ], tot_loss[loss=0.3352, simple_loss=0.3911, pruned_loss=0.1396, over 5627855.30 frames. ], libri_tot_loss[loss=0.3025, simple_loss=0.3673, pruned_loss=0.1189, over 5685488.05 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3916, pruned_loss=0.1399, over 5625845.33 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:04:16,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:04:26,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:04:35,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9758, 2.3794, 2.0403, 1.6802], device='cuda:1'), covar=tensor([0.2940, 0.2012, 0.2503, 0.2770], device='cuda:1'), in_proj_covar=tensor([0.1934, 0.1868, 0.1789, 0.1921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 04:04:36,784 INFO [optim.py:369] (1/2) 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 04:04:55,963 INFO [train.py:968] (1/2) Epoch 19, batch 41450, libri_loss[loss=0.2831, simple_loss=0.3502, pruned_loss=0.108, over 29552.00 frames. ], tot_loss[loss=0.3384, simple_loss=0.3936, pruned_loss=0.1416, over 5632107.48 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3664, pruned_loss=0.1182, over 5688409.57 frames. ], giga_tot_loss[loss=0.3408, simple_loss=0.3955, pruned_loss=0.143, over 5626417.24 frames. ], batch size: 78, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:04:59,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4975, 1.5252, 1.2403, 1.1228], device='cuda:1'), covar=tensor([0.0802, 0.0484, 0.0925, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0449, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 04:05:10,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5608, 1.8687, 1.7360, 1.6463], device='cuda:1'), covar=tensor([0.0764, 0.0289, 0.0293, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0115, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0067, 0.0060, 0.0104], device='cuda:1')
+2023-03-10 04:05:36,311 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:05:47,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5235, 1.6945, 1.5073, 1.5538], device='cuda:1'), covar=tensor([0.1307, 0.1534, 0.1907, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0745, 0.0710, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 04:05:49,007 INFO [train.py:968] (1/2) Epoch 19, batch 41500, giga_loss[loss=0.3004, simple_loss=0.363, pruned_loss=0.1189, over 28746.00 frames. ], tot_loss[loss=0.338, simple_loss=0.393, pruned_loss=0.1415, over 5625861.63 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3666, pruned_loss=0.1184, over 5676822.08 frames. ], giga_tot_loss[loss=0.3407, simple_loss=0.3951, pruned_loss=0.1431, over 5630051.13 frames. ], batch size: 99, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:06:12,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:06:19,548 INFO [optim.py:369] (1/2) 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 04:06:20,440 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:06:34,416 INFO [train.py:968] (1/2) Epoch 19, batch 41550, giga_loss[loss=0.3377, simple_loss=0.3932, pruned_loss=0.1411, over 28749.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.391, pruned_loss=0.1406, over 5633429.20 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3669, pruned_loss=0.1187, over 5681717.91 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3931, pruned_loss=0.1423, over 5630606.05 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:06:40,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6560, 1.6683, 1.2643, 1.2911], device='cuda:1'), covar=tensor([0.0809, 0.0542, 0.0995, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0448, 0.0515, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 04:07:22,477 INFO [train.py:968] (1/2) Epoch 19, batch 41600, libri_loss[loss=0.2449, simple_loss=0.3206, pruned_loss=0.08462, over 29349.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3906, pruned_loss=0.1399, over 5624329.19 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1186, over 5678435.61 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3932, pruned_loss=0.1422, over 5623877.97 frames. ], batch size: 67, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:07:31,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:07:39,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:07:58,174 INFO [optim.py:369] (1/2) 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 04:08:18,819 INFO [train.py:968] (1/2) Epoch 19, batch 41650, giga_loss[loss=0.3771, simple_loss=0.4142, pruned_loss=0.17, over 26535.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3899, pruned_loss=0.1387, over 5601231.84 frames. ], libri_tot_loss[loss=0.3022, simple_loss=0.3672, pruned_loss=0.1187, over 5668174.30 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.392, pruned_loss=0.1407, over 5609051.74 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:08:22,926 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:09:10,796 INFO [train.py:968] (1/2) Epoch 19, batch 41700, giga_loss[loss=0.3502, simple_loss=0.411, pruned_loss=0.1447, over 29057.00 frames. ], tot_loss[loss=0.337, simple_loss=0.3924, pruned_loss=0.1408, over 5591032.57 frames. ], libri_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1185, over 5674125.76 frames. ], giga_tot_loss[loss=0.3406, simple_loss=0.3949, pruned_loss=0.1431, over 5589754.36 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:09:16,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7980, 4.9009, 1.9195, 2.1687], device='cuda:1'), covar=tensor([0.1052, 0.0321, 0.0832, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0544, 0.0374, 0.0416], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0028], device='cuda:1')
+2023-03-10 04:09:27,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5415, 1.6530, 1.8230, 1.4354], device='cuda:1'), covar=tensor([0.1384, 0.1948, 0.1156, 0.1597], device='cuda:1'), in_proj_covar=tensor([0.0879, 0.0693, 0.0923, 0.0823], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 04:09:35,625 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:09:42,420 INFO [optim.py:369] (1/2) 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 04:10:01,089 INFO [train.py:968] (1/2) Epoch 19, batch 41750, giga_loss[loss=0.2785, simple_loss=0.3558, pruned_loss=0.1006, over 28769.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3897, pruned_loss=0.1385, over 5606409.90 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3665, pruned_loss=0.1182, over 5679431.57 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3927, pruned_loss=0.1411, over 5598996.92 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:10:09,064 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:12,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:16,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3221, 1.7825, 1.4278, 1.4858], device='cuda:1'), covar=tensor([0.0667, 0.0382, 0.0318, 0.0735], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 04:10:40,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:42,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-10 04:10:47,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:10:49,516 INFO [train.py:968] (1/2) Epoch 19, batch 41800, giga_loss[loss=0.3149, simple_loss=0.3739, pruned_loss=0.128, over 27978.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3866, pruned_loss=0.1346, over 5619133.54 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.3659, pruned_loss=0.1179, over 5677999.26 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3903, pruned_loss=0.1376, over 5612323.34 frames. ], batch size: 412, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:11:23,609 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 41850, giga_loss[loss=0.2943, simple_loss=0.3668, pruned_loss=0.1109, over 28216.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3833, pruned_loss=0.1313, over 5628845.40 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3656, pruned_loss=0.1177, over 5681675.56 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3867, pruned_loss=0.134, over 5619279.79 frames. ], batch size: 77, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:11:57,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:12:33,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:12:35,018 INFO [train.py:968] (1/2) Epoch 19, batch 41900, giga_loss[loss=0.2903, simple_loss=0.3441, pruned_loss=0.1183, over 23715.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3798, pruned_loss=0.1286, over 5621774.67 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3654, pruned_loss=0.1176, over 5685025.56 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.383, pruned_loss=0.131, over 5610402.98 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:12:42,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:13:03,436 INFO [optim.py:369] (1/2) 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 04:13:05,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6312, 1.7795, 1.7120, 1.6082], device='cuda:1'), covar=tensor([0.1711, 0.2296, 0.2018, 0.2146], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0748, 0.0713, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 04:13:14,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:13:16,699 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 41950, giga_loss[loss=0.3042, simple_loss=0.3669, pruned_loss=0.1208, over 28919.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3769, pruned_loss=0.1262, over 5639829.21 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.365, pruned_loss=0.1174, over 5689482.80 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3801, pruned_loss=0.1287, over 5624886.12 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:13:47,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:13:56,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:14:13,797 INFO [train.py:968] (1/2) Epoch 19, batch 42000, libri_loss[loss=0.2352, simple_loss=0.3014, pruned_loss=0.0845, over 29374.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3758, pruned_loss=0.1256, over 5646537.69 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3646, pruned_loss=0.1172, over 5692502.02 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.379, pruned_loss=0.1278, over 5631262.32 frames. ], batch size: 67, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:14:13,797 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 04:14:23,598 INFO [train.py:1012] (1/2) Epoch 19, validation: loss=0.2044, simple_loss=0.3117, pruned_loss=0.04848, over 944034.00 frames. 
+2023-03-10 04:14:23,599 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 04:14:29,802 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:14:31,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:14:47,912 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864616.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 04:14:53,291 INFO [optim.py:369] (1/2) 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 04:15:00,196 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:02,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:04,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2668, 1.6430, 1.2515, 0.7345], device='cuda:1'), covar=tensor([0.4166, 0.2109, 0.2472, 0.5200], device='cuda:1'), in_proj_covar=tensor([0.1707, 0.1604, 0.1581, 0.1396], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 04:15:07,561 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 42050, giga_loss[loss=0.3397, simple_loss=0.398, pruned_loss=0.1407, over 28837.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1251, over 5641839.32 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.117, over 5691687.39 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3781, pruned_loss=0.1272, over 5629816.13 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:15:15,511 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:17,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:40,205 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:15:49,971 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 42100, giga_loss[loss=0.3535, simple_loss=0.4235, pruned_loss=0.1418, over 28836.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3726, pruned_loss=0.1224, over 5647461.81 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.364, pruned_loss=0.1168, over 5695077.84 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3754, pruned_loss=0.1243, over 5633618.95 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:16:08,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=864689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:16:35,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:16:36,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:16:41,147 INFO [optim.py:369] (1/2) 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 04:16:57,311 INFO [train.py:968] (1/2) Epoch 19, batch 42150, giga_loss[loss=0.2614, simple_loss=0.3404, pruned_loss=0.0912, over 28931.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3723, pruned_loss=0.1195, over 5645838.56 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3635, pruned_loss=0.1166, over 5687745.49 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3753, pruned_loss=0.1214, over 5639554.15 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:17:04,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:17:18,998 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864759.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:17:23,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864762.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:17:45,385 INFO [train.py:968] (1/2) Epoch 19, batch 42200, libri_loss[loss=0.2577, simple_loss=0.3354, pruned_loss=0.09002, over 29562.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.374, pruned_loss=0.1194, over 5656307.26 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3629, pruned_loss=0.1161, over 5686524.98 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3774, pruned_loss=0.1215, over 5651484.19 frames. ], batch size: 75, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:17:50,808 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864791.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 04:18:18,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5593, 1.7572, 1.4331, 1.7075], device='cuda:1'), covar=tensor([0.2420, 0.2465, 0.2628, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.1462, 0.1063, 0.1301, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 04:18:19,265 INFO [optim.py:369] (1/2) 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 04:18:21,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5404, 1.6344, 1.3294, 1.2011], device='cuda:1'), covar=tensor([0.0940, 0.0639, 0.1017, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0448, 0.0514, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 04:18:29,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=864832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:18:31,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=864835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:18:33,510 INFO [train.py:968] (1/2) Epoch 19, batch 42250, giga_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.1221, over 28739.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3752, pruned_loss=0.1211, over 5665468.81 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3628, pruned_loss=0.1162, over 5691150.32 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3783, pruned_loss=0.1228, over 5656668.17 frames. ], batch size: 92, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:18:59,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=864864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:19:19,699 INFO [train.py:968] (1/2) Epoch 19, batch 42300, libri_loss[loss=0.3344, simple_loss=0.3978, pruned_loss=0.1355, over 29100.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3745, pruned_loss=0.121, over 5663503.78 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3629, pruned_loss=0.1162, over 5695885.02 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3773, pruned_loss=0.1226, over 5650995.30 frames. ], batch size: 101, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:19:21,458 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:20:08,340 INFO [train.py:968] (1/2) Epoch 19, batch 42350, giga_loss[loss=0.2611, simple_loss=0.3338, pruned_loss=0.09423, over 28908.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.374, pruned_loss=0.1218, over 5673856.12 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3628, pruned_loss=0.1161, over 5697877.46 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3764, pruned_loss=0.1232, over 5661808.38 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:20:34,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-10 04:20:45,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 04:20:56,344 INFO [train.py:968] (1/2) Epoch 19, batch 42400, giga_loss[loss=0.3157, simple_loss=0.3742, pruned_loss=0.1286, over 28734.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3734, pruned_loss=0.1227, over 5668355.50 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3631, pruned_loss=0.1163, over 5699816.78 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3752, pruned_loss=0.1237, over 5656851.50 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:20:56,762 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=864988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:21:09,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3579, 2.3874, 2.1866, 2.1592], device='cuda:1'), covar=tensor([0.1872, 0.2480, 0.2310, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0752, 0.0715, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-10 04:21:28,565 INFO [optim.py:369] (1/2) 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 04:21:46,695 INFO [train.py:968] (1/2) Epoch 19, batch 42450, giga_loss[loss=0.3236, simple_loss=0.3686, pruned_loss=0.1393, over 23985.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3716, pruned_loss=0.1204, over 5671995.41 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.363, pruned_loss=0.1161, over 5704950.63 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3734, pruned_loss=0.1215, over 5657425.32 frames. ], batch size: 705, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:22:31,213 INFO [train.py:968] (1/2) Epoch 19, batch 42500, libri_loss[loss=0.2554, simple_loss=0.3268, pruned_loss=0.092, over 29617.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3715, pruned_loss=0.1191, over 5683790.30 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3627, pruned_loss=0.1158, over 5707328.57 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3735, pruned_loss=0.1204, over 5669322.36 frames. ], batch size: 73, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:22:36,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-10 04:23:06,587 INFO [optim.py:369] (1/2) 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 04:23:20,965 INFO [train.py:968] (1/2) Epoch 19, batch 42550, libri_loss[loss=0.3873, simple_loss=0.423, pruned_loss=0.1758, over 18822.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3714, pruned_loss=0.1191, over 5661387.14 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.363, pruned_loss=0.1162, over 5691330.48 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3729, pruned_loss=0.1198, over 5664212.25 frames. ], batch size: 187, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:24:02,433 INFO [train.py:968] (1/2) Epoch 19, batch 42600, giga_loss[loss=0.3413, simple_loss=0.3943, pruned_loss=0.1442, over 27684.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3715, pruned_loss=0.1199, over 5651236.05 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.363, pruned_loss=0.1165, over 5679989.42 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3731, pruned_loss=0.1203, over 5663085.11 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:24:29,061 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=865218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:24:33,830 INFO [optim.py:369] (1/2) 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 04:24:48,854 INFO [train.py:968] (1/2) Epoch 19, batch 42650, giga_loss[loss=0.4221, simple_loss=0.4441, pruned_loss=0.2001, over 27578.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3695, pruned_loss=0.1189, over 5658980.43 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3633, pruned_loss=0.1167, over 5673430.30 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3707, pruned_loss=0.1191, over 5673695.58 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:25:16,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=865265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:25:21,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4182, 1.6474, 1.6763, 1.2523], device='cuda:1'), covar=tensor([0.1812, 0.2591, 0.1490, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0701, 0.0933, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 04:25:41,253 INFO [train.py:968] (1/2) Epoch 19, batch 42700, giga_loss[loss=0.3064, simple_loss=0.37, pruned_loss=0.1214, over 28654.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3685, pruned_loss=0.1191, over 5652662.77 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3635, pruned_loss=0.1167, over 5675693.57 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3693, pruned_loss=0.1193, over 5662006.11 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:26:14,870 INFO [optim.py:369] (1/2) 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 04:26:30,342 INFO [train.py:968] (1/2) Epoch 19, batch 42750, giga_loss[loss=0.2638, simple_loss=0.3391, pruned_loss=0.09426, over 28996.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3678, pruned_loss=0.1189, over 5669182.76 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3634, pruned_loss=0.1166, over 5678064.77 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3686, pruned_loss=0.1192, over 5674179.83 frames. ], batch size: 164, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:26:58,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=865363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:27:20,628 INFO [train.py:968] (1/2) Epoch 19, batch 42800, libri_loss[loss=0.2939, simple_loss=0.3712, pruned_loss=0.1083, over 27442.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.367, pruned_loss=0.1192, over 5668035.02 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3633, pruned_loss=0.1166, over 5680045.68 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3677, pruned_loss=0.1195, over 5669863.31 frames. ], batch size: 115, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:27:40,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=865408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:27:45,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=865411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:27:59,379 INFO [optim.py:369] (1/2) 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 04:28:12,073 INFO [train.py:968] (1/2) Epoch 19, batch 42850, libri_loss[loss=0.306, simple_loss=0.3805, pruned_loss=0.1158, over 29123.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3675, pruned_loss=0.1203, over 5659332.73 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3636, pruned_loss=0.1166, over 5682728.38 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.368, pruned_loss=0.1206, over 5657227.65 frames. ], batch size: 101, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:28:14,592 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:28:59,424 INFO [train.py:968] (1/2) Epoch 19, batch 42900, libri_loss[loss=0.2963, simple_loss=0.3608, pruned_loss=0.116, over 19754.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3676, pruned_loss=0.1204, over 5651949.03 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3637, pruned_loss=0.1166, over 5678573.51 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.368, pruned_loss=0.1208, over 5654152.89 frames. ], batch size: 187, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:29:15,894 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=865506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:29:18,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=865509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:29:30,732 INFO [optim.py:369] (1/2) 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 04:29:38,064 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-10 04:29:46,412 INFO [train.py:968] (1/2) Epoch 19, batch 42950, giga_loss[loss=0.3547, simple_loss=0.4022, pruned_loss=0.1536, over 28250.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3676, pruned_loss=0.1191, over 5666779.41 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3632, pruned_loss=0.1162, over 5683141.32 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3684, pruned_loss=0.1198, over 5664173.48 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:29:46,719 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:30:31,228 INFO [train.py:968] (1/2) Epoch 19, batch 43000, giga_loss[loss=0.4589, simple_loss=0.4591, pruned_loss=0.2294, over 26575.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3676, pruned_loss=0.1186, over 5671062.05 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3634, pruned_loss=0.1163, over 5686620.89 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3682, pruned_loss=0.1191, over 5665413.12 frames. ], batch size: 555, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:30:37,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=865593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:31:03,319 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 43050, libri_loss[loss=0.3147, simple_loss=0.3796, pruned_loss=0.1249, over 28713.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3679, pruned_loss=0.1183, over 5660254.16 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3641, pruned_loss=0.1168, over 5671864.84 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1182, over 5668687.23 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:32:11,555 INFO [train.py:968] (1/2) Epoch 19, batch 43100, giga_loss[loss=0.3773, simple_loss=0.4202, pruned_loss=0.1672, over 28269.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3707, pruned_loss=0.1203, over 5672577.58 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3642, pruned_loss=0.1169, over 5673899.98 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3706, pruned_loss=0.1203, over 5677419.24 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:32:43,443 INFO [optim.py:369] (1/2) 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 04:32:47,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4004, 1.4841, 3.4902, 3.2940], device='cuda:1'), covar=tensor([0.1314, 0.2410, 0.0450, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0642, 0.0948, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 04:32:58,912 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=865736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:33:00,326 INFO [train.py:968] (1/2) Epoch 19, batch 43150, giga_loss[loss=0.3382, simple_loss=0.3858, pruned_loss=0.1453, over 28782.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3715, pruned_loss=0.1219, over 5676453.34 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3639, pruned_loss=0.1165, over 5678362.36 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3718, pruned_loss=0.1222, over 5676424.38 frames. ], batch size: 284, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:33:02,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=865739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:33:36,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=865768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:33:58,213 INFO [train.py:968] (1/2) Epoch 19, batch 43200, giga_loss[loss=0.3382, simple_loss=0.3882, pruned_loss=0.1441, over 28585.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5672144.97 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.364, pruned_loss=0.1166, over 5679731.54 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3736, pruned_loss=0.1248, over 5670955.30 frames. ], batch size: 85, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:34:34,628 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 43250, giga_loss[loss=0.2922, simple_loss=0.3649, pruned_loss=0.1098, over 28930.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3768, pruned_loss=0.1279, over 5659892.13 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.364, pruned_loss=0.1166, over 5683909.10 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.377, pruned_loss=0.1282, over 5655232.10 frames. ], batch size: 155, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:35:05,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6761, 1.7488, 1.8832, 1.4130], device='cuda:1'), covar=tensor([0.1720, 0.2424, 0.1392, 0.1649], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0698, 0.0928, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 04:35:12,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6742, 1.7655, 1.3425, 1.3354], device='cuda:1'), covar=tensor([0.0987, 0.0648, 0.1065, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0450, 0.0517, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 04:35:33,103 INFO [train.py:968] (1/2) Epoch 19, batch 43300, giga_loss[loss=0.2988, simple_loss=0.3637, pruned_loss=0.117, over 28879.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3757, pruned_loss=0.1274, over 5654196.70 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3645, pruned_loss=0.1169, over 5678105.54 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3759, pruned_loss=0.1277, over 5655581.58 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:36:07,978 INFO [optim.py:369] (1/2) 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 04:36:20,861 INFO [train.py:968] (1/2) Epoch 19, batch 43350, giga_loss[loss=0.3138, simple_loss=0.3909, pruned_loss=0.1184, over 28991.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3738, pruned_loss=0.1254, over 5660027.20 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3643, pruned_loss=0.1168, over 5677748.60 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3742, pruned_loss=0.1259, over 5661154.80 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:37:03,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5501, 1.9942, 1.7210, 1.6441], device='cuda:1'), covar=tensor([0.0797, 0.0284, 0.0308, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 04:37:06,449 INFO [train.py:968] (1/2) Epoch 19, batch 43400, giga_loss[loss=0.31, simple_loss=0.3667, pruned_loss=0.1267, over 27670.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3737, pruned_loss=0.1237, over 5661712.62 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3645, pruned_loss=0.1168, over 5681383.86 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.374, pruned_loss=0.1241, over 5659068.33 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:37:38,569 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 43450, giga_loss[loss=0.2753, simple_loss=0.3473, pruned_loss=0.1017, over 28350.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1222, over 5655865.26 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3645, pruned_loss=0.1169, over 5678832.89 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1226, over 5656154.60 frames. ], batch size: 60, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:38:16,925 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 43500, giga_loss[loss=0.3116, simple_loss=0.3741, pruned_loss=0.1245, over 28890.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3685, pruned_loss=0.1202, over 5672024.36 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3643, pruned_loss=0.1167, over 5681194.75 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5669890.75 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:39:12,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-10 04:39:12,203 INFO [optim.py:369] (1/2) 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 04:39:25,374 INFO [train.py:968] (1/2) Epoch 19, batch 43550, libri_loss[loss=0.2821, simple_loss=0.3552, pruned_loss=0.1045, over 29509.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3677, pruned_loss=0.1203, over 5668433.25 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3645, pruned_loss=0.1168, over 5683568.59 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3681, pruned_loss=0.1208, over 5664162.23 frames. ], batch size: 84, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:40:10,848 INFO [train.py:968] (1/2) Epoch 19, batch 43600, giga_loss[loss=0.3109, simple_loss=0.3796, pruned_loss=0.1211, over 28844.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3701, pruned_loss=0.1221, over 5672787.47 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3644, pruned_loss=0.117, over 5687104.32 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5666074.41 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:40:46,873 INFO [optim.py:369] (1/2) 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 04:40:59,192 INFO [train.py:968] (1/2) Epoch 19, batch 43650, libri_loss[loss=0.2785, simple_loss=0.3489, pruned_loss=0.1041, over 29218.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.374, pruned_loss=0.1228, over 5660978.52 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.117, over 5682466.41 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3746, pruned_loss=0.1232, over 5659524.44 frames. ], batch size: 97, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:41:47,929 INFO [train.py:968] (1/2) Epoch 19, batch 43700, giga_loss[loss=0.3648, simple_loss=0.4163, pruned_loss=0.1566, over 27517.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3754, pruned_loss=0.1219, over 5666046.19 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1175, over 5684604.85 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3757, pruned_loss=0.122, over 5662225.21 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:42:25,222 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 19, batch 43750, giga_loss[loss=0.2869, simple_loss=0.362, pruned_loss=0.1059, over 28526.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.378, pruned_loss=0.1242, over 5664674.58 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.365, pruned_loss=0.1175, over 5687946.63 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3785, pruned_loss=0.1244, over 5658389.46 frames. ], batch size: 71, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:43:25,062 INFO [train.py:968] (1/2) Epoch 19, batch 43800, giga_loss[loss=0.3002, simple_loss=0.3739, pruned_loss=0.1133, over 28926.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3785, pruned_loss=0.1249, over 5659182.52 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3643, pruned_loss=0.117, over 5682933.48 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3799, pruned_loss=0.1257, over 5658494.46 frames. ], batch size: 174, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:44:01,361 INFO [optim.py:369] (1/2) 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 04:44:12,032 INFO [train.py:968] (1/2) Epoch 19, batch 43850, libri_loss[loss=0.2964, simple_loss=0.3696, pruned_loss=0.1116, over 29195.00 frames. ], tot_loss[loss=0.3159, simple_loss=0.3792, pruned_loss=0.1263, over 5662748.13 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3646, pruned_loss=0.1171, over 5685576.40 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3804, pruned_loss=0.127, over 5658929.91 frames. ], batch size: 101, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:44:16,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=866443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:44:59,438 INFO [train.py:968] (1/2) Epoch 19, batch 43900, giga_loss[loss=0.3331, simple_loss=0.3874, pruned_loss=0.1394, over 28721.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3771, pruned_loss=0.1257, over 5663649.30 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3645, pruned_loss=0.1171, over 5690473.79 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3784, pruned_loss=0.1265, over 5655856.10 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:45:17,240 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=866506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:45:35,365 INFO [optim.py:369] (1/2) 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 04:45:46,012 INFO [train.py:968] (1/2) Epoch 19, batch 43950, giga_loss[loss=0.2723, simple_loss=0.3431, pruned_loss=0.1008, over 28587.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1249, over 5657224.46 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3646, pruned_loss=0.117, over 5684435.19 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3761, pruned_loss=0.1258, over 5655282.53 frames. ], batch size: 85, lr: 1.66e-03, grad_scale: 2.0
+2023-03-10 04:45:51,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8858, 1.9350, 1.8300, 1.6935], device='cuda:1'), covar=tensor([0.1625, 0.1957, 0.2083, 0.1934], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0744, 0.0710, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 04:46:35,909 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 19, batch 44000, giga_loss[loss=0.2803, simple_loss=0.3477, pruned_loss=0.1064, over 28741.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1242, over 5659969.35 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.365, pruned_loss=0.1173, over 5678481.78 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3739, pruned_loss=0.1247, over 5662867.63 frames. ], batch size: 262, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:46:38,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=866589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:47:09,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2153, 1.0648, 3.7107, 3.1902], device='cuda:1'), covar=tensor([0.1663, 0.2886, 0.0462, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0643, 0.0951, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 04:47:09,493 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=866618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:47:16,073 INFO [optim.py:369] (1/2) 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 04:47:21,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 04:47:26,881 INFO [train.py:968] (1/2) Epoch 19, batch 44050, giga_loss[loss=0.3133, simple_loss=0.3776, pruned_loss=0.1245, over 28937.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5661580.87 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1172, over 5673118.30 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3733, pruned_loss=0.1245, over 5668115.59 frames. ], batch size: 227, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:48:13,887 INFO [train.py:968] (1/2) Epoch 19, batch 44100, giga_loss[loss=0.249, simple_loss=0.328, pruned_loss=0.08503, over 29018.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3731, pruned_loss=0.1244, over 5663719.60 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3646, pruned_loss=0.1169, over 5677242.11 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3743, pruned_loss=0.1255, over 5664637.77 frames. ], batch size: 136, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:48:29,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2502, 1.5688, 1.1873, 1.3085], device='cuda:1'), covar=tensor([0.2639, 0.2507, 0.3091, 0.1930], device='cuda:1'), in_proj_covar=tensor([0.1471, 0.1066, 0.1308, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 04:48:48,701 INFO [optim.py:369] (1/2) 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 04:48:49,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3213, 4.1334, 3.9650, 1.7400], device='cuda:1'), covar=tensor([0.0703, 0.0859, 0.0830, 0.2184], device='cuda:1'), in_proj_covar=tensor([0.1204, 0.1118, 0.0950, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 04:49:00,177 INFO [train.py:968] (1/2) Epoch 19, batch 44150, giga_loss[loss=0.3019, simple_loss=0.3683, pruned_loss=0.1178, over 28279.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3709, pruned_loss=0.1238, over 5665897.20 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3645, pruned_loss=0.117, over 5679679.74 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1246, over 5664203.44 frames. ], batch size: 368, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:49:44,020 INFO [train.py:968] (1/2) Epoch 19, batch 44200, giga_loss[loss=0.2815, simple_loss=0.3502, pruned_loss=0.1064, over 28834.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3699, pruned_loss=0.1226, over 5656155.66 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3652, pruned_loss=0.1174, over 5663229.16 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3705, pruned_loss=0.1232, over 5668525.90 frames. ], batch size: 199, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:50:19,224 INFO [optim.py:369] (1/2) 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 04:50:30,880 INFO [train.py:968] (1/2) Epoch 19, batch 44250, giga_loss[loss=0.3049, simple_loss=0.3733, pruned_loss=0.1183, over 28954.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5651494.79 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3648, pruned_loss=0.1171, over 5661342.67 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.123, over 5663800.84 frames. ], batch size: 213, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:51:16,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=866881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:51:21,272 INFO [train.py:968] (1/2) Epoch 19, batch 44300, giga_loss[loss=0.2865, simple_loss=0.3538, pruned_loss=0.1096, over 29007.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.372, pruned_loss=0.1229, over 5643855.85 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3649, pruned_loss=0.1173, over 5645914.44 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1235, over 5666484.35 frames. ], batch size: 106, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:52:03,429 INFO [optim.py:369] (1/2) 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 04:52:09,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5671, 1.7208, 1.3359, 1.3339], device='cuda:1'), covar=tensor([0.0910, 0.0573, 0.0946, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0448, 0.0514, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 04:52:12,815 INFO [train.py:968] (1/2) Epoch 19, batch 44350, giga_loss[loss=0.2858, simple_loss=0.3555, pruned_loss=0.108, over 28370.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1232, over 5638572.50 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.365, pruned_loss=0.1172, over 5637891.35 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5664486.65 frames. ], batch size: 65, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:52:37,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1143, 3.9335, 3.7181, 1.8222], device='cuda:1'), covar=tensor([0.0697, 0.0807, 0.0787, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1121, 0.0951, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 04:53:02,534 INFO [train.py:968] (1/2) Epoch 19, batch 44400, giga_loss[loss=0.2859, simple_loss=0.3657, pruned_loss=0.103, over 29024.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3747, pruned_loss=0.1236, over 5642815.22 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3646, pruned_loss=0.117, over 5642174.74 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3756, pruned_loss=0.1243, over 5659472.66 frames. ], batch size: 128, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:53:08,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3183, 1.5295, 1.2943, 1.4410], device='cuda:1'), covar=tensor([0.0752, 0.0369, 0.0345, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 04:53:30,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5201, 1.7042, 1.7028, 1.2945], device='cuda:1'), covar=tensor([0.2062, 0.2578, 0.1750, 0.1968], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0703, 0.0934, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 04:53:33,710 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=867024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:53:35,981 INFO [optim.py:369] (1/2) 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 04:53:36,915 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=867027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:53:45,434 INFO [train.py:968] (1/2) Epoch 19, batch 44450, giga_loss[loss=0.288, simple_loss=0.3716, pruned_loss=0.1022, over 28880.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3765, pruned_loss=0.1222, over 5631422.37 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3652, pruned_loss=0.1176, over 5606487.16 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3768, pruned_loss=0.1223, over 5677567.29 frames. ], batch size: 66, lr: 1.66e-03, grad_scale: 8.0
+2023-03-10 04:54:03,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=867056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:54:07,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-10 04:54:36,061 INFO [train.py:968] (1/2) Epoch 19, batch 44500, libri_loss[loss=0.3291, simple_loss=0.3782, pruned_loss=0.14, over 19873.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3815, pruned_loss=0.125, over 5600811.89 frames. ], libri_tot_loss[loss=0.3013, simple_loss=0.366, pruned_loss=0.1183, over 5563706.37 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3812, pruned_loss=0.1246, over 5677647.80 frames. ], batch size: 187, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:55:13,695 INFO [optim.py:369] (1/2) 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 04:55:24,295 INFO [train.py:968] (1/2) Epoch 19, batch 44550, giga_loss[loss=0.4362, simple_loss=0.4615, pruned_loss=0.2055, over 27487.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3838, pruned_loss=0.1276, over 5585556.02 frames. ], libri_tot_loss[loss=0.3017, simple_loss=0.3663, pruned_loss=0.1186, over 5531747.72 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3834, pruned_loss=0.1271, over 5675649.61 frames. ], batch size: 472, lr: 1.66e-03, grad_scale: 4.0
+2023-03-10 04:55:45,866 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-10 04:56:39,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6001, 1.7471, 1.8196, 1.3684], device='cuda:1'), covar=tensor([0.2034, 0.2562, 0.1610, 0.1940], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0698, 0.0928, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0013], device='cuda:1')
+2023-03-10 04:57:00,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3614, 3.6201, 1.5706, 1.5477], device='cuda:1'), covar=tensor([0.1071, 0.0262, 0.0964, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0550, 0.0378, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 04:57:04,575 INFO [train.py:968] (1/2) Epoch 20, batch 50, giga_loss[loss=0.3274, simple_loss=0.4076, pruned_loss=0.1236, over 28692.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3705, pruned_loss=0.1067, over 1265483.44 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3541, pruned_loss=0.09555, over 199048.24 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3732, pruned_loss=0.1085, over 1104715.67 frames. ], batch size: 262, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 04:57:22,680 INFO [optim.py:369] (1/2) 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 04:57:39,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8722, 3.6801, 3.4985, 1.8397], device='cuda:1'), covar=tensor([0.0652, 0.0806, 0.0720, 0.2341], device='cuda:1'), in_proj_covar=tensor([0.1201, 0.1114, 0.0946, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 04:57:53,947 INFO [train.py:968] (1/2) Epoch 20, batch 100, giga_loss[loss=0.2557, simple_loss=0.336, pruned_loss=0.08774, over 28593.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3656, pruned_loss=0.1063, over 2241213.57 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3456, pruned_loss=0.09154, over 340902.09 frames. ], giga_tot_loss[loss=0.2928, simple_loss=0.3686, pruned_loss=0.1084, over 2020233.61 frames. ], batch size: 307, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 04:58:03,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7830, 2.0001, 1.6147, 2.0254], device='cuda:1'), covar=tensor([0.2673, 0.2763, 0.3161, 0.2586], device='cuda:1'), in_proj_covar=tensor([0.1468, 0.1066, 0.1305, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 04:58:24,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8860, 2.0712, 2.1697, 1.6547], device='cuda:1'), covar=tensor([0.1989, 0.2353, 0.1508, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0700, 0.0933, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 04:58:40,033 INFO [train.py:968] (1/2) Epoch 20, batch 150, giga_loss[loss=0.2859, simple_loss=0.3554, pruned_loss=0.1082, over 28640.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3485, pruned_loss=0.0977, over 2999723.44 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3418, pruned_loss=0.09062, over 414398.97 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3498, pruned_loss=0.09868, over 2790548.87 frames. ], batch size: 242, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 04:58:45,889 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 04:58:54,856 INFO [optim.py:369] (1/2) 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 04:59:05,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3392, 1.6621, 1.2831, 1.3645], device='cuda:1'), covar=tensor([0.2664, 0.2636, 0.3134, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.1469, 0.1067, 0.1306, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 04:59:19,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 04:59:21,780 INFO [train.py:968] (1/2) Epoch 20, batch 200, libri_loss[loss=0.2286, simple_loss=0.3095, pruned_loss=0.07384, over 29651.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3348, pruned_loss=0.0915, over 3601788.46 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3379, pruned_loss=0.08853, over 548430.57 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3354, pruned_loss=0.09225, over 3378558.25 frames. ], batch size: 69, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:00:02,422 INFO [train.py:968] (1/2) Epoch 20, batch 250, giga_loss[loss=0.1929, simple_loss=0.2735, pruned_loss=0.0561, over 28947.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3259, pruned_loss=0.08714, over 4056415.57 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3392, pruned_loss=0.08846, over 720590.81 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3252, pruned_loss=0.08749, over 3822837.47 frames. ], batch size: 145, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:00:15,154 INFO [optim.py:369] (1/2) 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 05:00:43,233 INFO [train.py:968] (1/2) Epoch 20, batch 300, giga_loss[loss=0.207, simple_loss=0.2802, pruned_loss=0.06694, over 28672.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3175, pruned_loss=0.08321, over 4420380.09 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3364, pruned_loss=0.08715, over 924134.27 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3164, pruned_loss=0.08342, over 4171952.04 frames. ], batch size: 92, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:01:27,613 INFO [train.py:968] (1/2) Epoch 20, batch 350, giga_loss[loss=0.2157, simple_loss=0.2841, pruned_loss=0.07362, over 27660.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.311, pruned_loss=0.08034, over 4699341.38 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3364, pruned_loss=0.0871, over 1046382.85 frames. ], giga_tot_loss[loss=0.2348, simple_loss=0.3092, pruned_loss=0.08022, over 4472114.70 frames. ], batch size: 472, lr: 1.62e-03, grad_scale: 2.0
+2023-03-10 05:01:33,966 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867520.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:01:40,543 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 400, libri_loss[loss=0.2604, simple_loss=0.3447, pruned_loss=0.08812, over 29541.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3059, pruned_loss=0.07783, over 4917360.66 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3373, pruned_loss=0.08779, over 1084422.02 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.304, pruned_loss=0.07745, over 4737374.18 frames. ], batch size: 80, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:02:44,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:02:46,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5013, 1.8132, 1.7809, 1.4652], device='cuda:1'), covar=tensor([0.1788, 0.1849, 0.2048, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0740, 0.0707, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 05:02:50,070 INFO [train.py:968] (1/2) Epoch 20, batch 450, giga_loss[loss=0.1981, simple_loss=0.271, pruned_loss=0.06259, over 28573.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.305, pruned_loss=0.07759, over 5095740.11 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3392, pruned_loss=0.08862, over 1226659.56 frames. ], giga_tot_loss[loss=0.2277, simple_loss=0.3019, pruned_loss=0.07676, over 4927738.42 frames. ], batch size: 85, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:03:06,543 INFO [optim.py:369] (1/2) 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 05:03:32,223 INFO [train.py:968] (1/2) Epoch 20, batch 500, giga_loss[loss=0.2104, simple_loss=0.2891, pruned_loss=0.06585, over 29071.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3034, pruned_loss=0.07683, over 5219490.64 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3408, pruned_loss=0.08968, over 1377963.23 frames. ], giga_tot_loss[loss=0.2253, simple_loss=0.2995, pruned_loss=0.07553, over 5069094.96 frames. ], batch size: 128, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:03:59,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=867693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:04:16,266 INFO [train.py:968] (1/2) Epoch 20, batch 550, giga_loss[loss=0.2111, simple_loss=0.2886, pruned_loss=0.06678, over 28868.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3026, pruned_loss=0.0767, over 5325444.32 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3414, pruned_loss=0.09004, over 1510152.28 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.2983, pruned_loss=0.07522, over 5189787.39 frames. ], batch size: 285, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:04:30,556 INFO [optim.py:369] (1/2) 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 05:04:58,434 INFO [train.py:968] (1/2) Epoch 20, batch 600, giga_loss[loss=0.222, simple_loss=0.2942, pruned_loss=0.07489, over 28848.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3004, pruned_loss=0.07552, over 5406009.26 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3415, pruned_loss=0.08997, over 1658963.32 frames. ], giga_tot_loss[loss=0.2216, simple_loss=0.2955, pruned_loss=0.07389, over 5282279.05 frames. ], batch size: 112, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:05:04,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3464, 1.5499, 1.3249, 1.5657], device='cuda:1'), covar=tensor([0.0773, 0.0341, 0.0349, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 05:05:38,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0768, 2.2095, 1.6229, 1.8405], device='cuda:1'), covar=tensor([0.0973, 0.0701, 0.1123, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0444, 0.0511, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:05:45,929 INFO [train.py:968] (1/2) Epoch 20, batch 650, giga_loss[loss=0.1867, simple_loss=0.2649, pruned_loss=0.05425, over 28878.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2985, pruned_loss=0.0743, over 5474668.00 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3418, pruned_loss=0.08988, over 1783218.01 frames. ], giga_tot_loss[loss=0.2192, simple_loss=0.2932, pruned_loss=0.07257, over 5364365.79 frames. ], batch size: 112, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:06:01,863 INFO [optim.py:369] (1/2) 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 05:06:10,380 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=867836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:06:11,974 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=867839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:06:28,854 INFO [train.py:968] (1/2) Epoch 20, batch 700, giga_loss[loss=0.2382, simple_loss=0.2992, pruned_loss=0.08858, over 29001.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2977, pruned_loss=0.07413, over 5524870.96 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3434, pruned_loss=0.09116, over 1883370.83 frames. ], giga_tot_loss[loss=0.218, simple_loss=0.292, pruned_loss=0.07199, over 5429523.55 frames. ], batch size: 106, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:06:38,374 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=867868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:07:02,369 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=867895.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:07:05,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-10 05:07:15,495 INFO [train.py:968] (1/2) Epoch 20, batch 750, giga_loss[loss=0.2144, simple_loss=0.2833, pruned_loss=0.0728, over 28919.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2947, pruned_loss=0.0728, over 5558486.77 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3413, pruned_loss=0.08997, over 1963217.93 frames. ], giga_tot_loss[loss=0.2161, simple_loss=0.2899, pruned_loss=0.07113, over 5476157.35 frames. ], batch size: 227, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:07:17,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-10 05:07:32,254 INFO [optim.py:369] (1/2) 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 05:07:43,643 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:07:44,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=867943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:08:00,291 INFO [train.py:968] (1/2) Epoch 20, batch 800, giga_loss[loss=0.2405, simple_loss=0.315, pruned_loss=0.08296, over 28850.00 frames. ], tot_loss[loss=0.2176, simple_loss=0.2918, pruned_loss=0.07176, over 5589405.40 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3417, pruned_loss=0.08995, over 2002803.63 frames. ], giga_tot_loss[loss=0.214, simple_loss=0.2874, pruned_loss=0.07028, over 5521537.57 frames. ], batch size: 174, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:08:17,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=867979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:08:45,836 INFO [train.py:968] (1/2) Epoch 20, batch 850, giga_loss[loss=0.3032, simple_loss=0.3668, pruned_loss=0.1198, over 27694.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2993, pruned_loss=0.07572, over 5603341.59 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3425, pruned_loss=0.0903, over 2108494.68 frames. ], giga_tot_loss[loss=0.2212, simple_loss=0.2943, pruned_loss=0.07399, over 5549780.38 frames. ], batch size: 472, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:09:00,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3988, 2.6418, 1.5503, 1.5377], device='cuda:1'), covar=tensor([0.0829, 0.0298, 0.0750, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0547, 0.0379, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:09:03,656 INFO [optim.py:369] (1/2) 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 05:09:11,433 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868038.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:09:14,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868041.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:09:18,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3968, 1.5402, 4.0553, 3.1964], device='cuda:1'), covar=tensor([0.1680, 0.2671, 0.0447, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0634, 0.0936, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:09:35,604 INFO [train.py:968] (1/2) Epoch 20, batch 900, giga_loss[loss=0.3022, simple_loss=0.3767, pruned_loss=0.1138, over 28813.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3124, pruned_loss=0.08197, over 5625308.11 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3422, pruned_loss=0.09019, over 2182647.69 frames. ], giga_tot_loss[loss=0.2345, simple_loss=0.308, pruned_loss=0.0805, over 5578619.45 frames. ], batch size: 112, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:09:35,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6885, 1.7543, 1.9300, 1.4549], device='cuda:1'), covar=tensor([0.1929, 0.2444, 0.1512, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0706, 0.0946, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 05:09:41,680 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868070.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:10:04,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-10 05:10:15,713 INFO [train.py:968] (1/2) Epoch 20, batch 950, giga_loss[loss=0.3141, simple_loss=0.3845, pruned_loss=0.1219, over 27955.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3224, pruned_loss=0.0868, over 5640927.12 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3401, pruned_loss=0.08914, over 2322678.04 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3187, pruned_loss=0.08581, over 5599315.17 frames. ], batch size: 412, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:10:26,955 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:10:28,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:10:30,685 INFO [optim.py:369] (1/2) 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 05:10:54,706 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:10:59,054 INFO [train.py:968] (1/2) Epoch 20, batch 1000, giga_loss[loss=0.274, simple_loss=0.3536, pruned_loss=0.09718, over 28590.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3301, pruned_loss=0.08972, over 5652274.20 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3402, pruned_loss=0.08899, over 2392014.44 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.327, pruned_loss=0.08898, over 5616194.25 frames. ], batch size: 336, lr: 1.62e-03, grad_scale: 8.0
+2023-03-10 05:11:17,152 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 1050, giga_loss[loss=0.2569, simple_loss=0.3412, pruned_loss=0.08629, over 28958.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3333, pruned_loss=0.0898, over 5668318.67 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3392, pruned_loss=0.0883, over 2546854.04 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3309, pruned_loss=0.08954, over 5628709.75 frames. ], batch size: 145, lr: 1.62e-03, grad_scale: 4.0
+2023-03-10 05:11:54,845 INFO [optim.py:369] (1/2) 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 05:12:23,244 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-10 05:12:25,775 INFO [train.py:968] (1/2) Epoch 20, batch 1100, giga_loss[loss=0.2535, simple_loss=0.325, pruned_loss=0.09099, over 28314.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3353, pruned_loss=0.09006, over 5666694.65 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3393, pruned_loss=0.08828, over 2630416.18 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3332, pruned_loss=0.0899, over 5629301.21 frames. ], batch size: 65, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:13:10,732 INFO [train.py:968] (1/2) Epoch 20, batch 1150, giga_loss[loss=0.2807, simple_loss=0.3576, pruned_loss=0.1019, over 28485.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.338, pruned_loss=0.09153, over 5673904.99 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3393, pruned_loss=0.08818, over 2695530.76 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3363, pruned_loss=0.09148, over 5640157.74 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:13:12,834 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:13:15,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:13:16,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:13:23,971 INFO [optim.py:369] (1/2) 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 05:13:53,104 INFO [train.py:968] (1/2) Epoch 20, batch 1200, giga_loss[loss=0.3074, simple_loss=0.375, pruned_loss=0.12, over 28606.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3404, pruned_loss=0.09351, over 5678160.26 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.34, pruned_loss=0.08856, over 2786779.40 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3387, pruned_loss=0.09345, over 5648334.19 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:14:40,082 INFO [train.py:968] (1/2) Epoch 20, batch 1250, giga_loss[loss=0.2741, simple_loss=0.3485, pruned_loss=0.09979, over 28532.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3441, pruned_loss=0.09608, over 5680827.22 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3391, pruned_loss=0.08805, over 2818015.12 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3432, pruned_loss=0.09629, over 5655427.30 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:14:50,657 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:15:08,900 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:24,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:24,409 INFO [train.py:968] (1/2) Epoch 20, batch 1300, libri_loss[loss=0.2261, simple_loss=0.3147, pruned_loss=0.06876, over 29560.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.347, pruned_loss=0.09664, over 5693879.62 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3394, pruned_loss=0.08807, over 2920853.32 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3463, pruned_loss=0.0971, over 5669509.32 frames. ], batch size: 76, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:15:24,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:26,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:27,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:39,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:45,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0105, 2.3487, 2.0981, 1.7010], device='cuda:1'), covar=tensor([0.2600, 0.2289, 0.2542, 0.2616], device='cuda:1'), in_proj_covar=tensor([0.1918, 0.1849, 0.1770, 0.1910], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 05:15:50,619 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:15:51,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:16:04,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1919, 2.3180, 1.5647, 1.8646], device='cuda:1'), covar=tensor([0.0947, 0.0690, 0.1078, 0.1124], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0446, 0.0515, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:16:05,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 05:16:05,359 INFO [train.py:968] (1/2) Epoch 20, batch 1350, giga_loss[loss=0.2694, simple_loss=0.3504, pruned_loss=0.09422, over 28769.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3494, pruned_loss=0.0976, over 5691043.71 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3398, pruned_loss=0.08808, over 3008758.82 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.349, pruned_loss=0.09824, over 5666017.93 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:16:08,694 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:16:20,717 INFO [optim.py:369] (1/2) 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 05:16:42,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:16:49,903 INFO [train.py:968] (1/2) Epoch 20, batch 1400, giga_loss[loss=0.2645, simple_loss=0.35, pruned_loss=0.08949, over 28423.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3515, pruned_loss=0.09831, over 5697924.81 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.34, pruned_loss=0.0882, over 3080672.78 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3513, pruned_loss=0.09901, over 5673003.98 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:17:32,755 INFO [train.py:968] (1/2) Epoch 20, batch 1450, giga_loss[loss=0.2915, simple_loss=0.3728, pruned_loss=0.1051, over 28221.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.352, pruned_loss=0.0976, over 5704255.74 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.34, pruned_loss=0.08824, over 3123007.85 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.352, pruned_loss=0.09826, over 5681898.17 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:17:34,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4220, 1.1875, 4.2597, 3.4480], device='cuda:1'), covar=tensor([0.1515, 0.2618, 0.0397, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0734, 0.0632, 0.0931, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:17:35,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6405, 1.6856, 1.8797, 1.4203], device='cuda:1'), covar=tensor([0.1981, 0.2566, 0.1559, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0702, 0.0942, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 05:17:48,547 INFO [optim.py:369] (1/2) 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 05:18:12,385 INFO [train.py:968] (1/2) Epoch 20, batch 1500, giga_loss[loss=0.2349, simple_loss=0.329, pruned_loss=0.07036, over 28506.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3499, pruned_loss=0.0953, over 5709355.37 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3397, pruned_loss=0.08786, over 3218701.82 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3504, pruned_loss=0.09629, over 5685223.46 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:18:24,130 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:18:27,354 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=868679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:36,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:43,532 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:45,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:18:52,232 INFO [train.py:968] (1/2) Epoch 20, batch 1550, giga_loss[loss=0.2664, simple_loss=0.3449, pruned_loss=0.09393, over 28887.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09348, over 5718747.27 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3396, pruned_loss=0.08772, over 3307971.72 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3491, pruned_loss=0.09455, over 5696265.67 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:19:09,086 INFO [optim.py:369] (1/2) 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 05:19:09,287 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:19:35,539 INFO [train.py:968] (1/2) Epoch 20, batch 1600, giga_loss[loss=0.2314, simple_loss=0.3206, pruned_loss=0.07108, over 28609.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3491, pruned_loss=0.09495, over 5708767.37 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3399, pruned_loss=0.0879, over 3383234.83 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3499, pruned_loss=0.09589, over 5687172.55 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:19:52,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6131, 1.7155, 1.3113, 1.2570], device='cuda:1'), covar=tensor([0.0963, 0.0606, 0.0983, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0443, 0.0512, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:20:06,868 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:09,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3654, 2.0804, 1.6285, 0.6488], device='cuda:1'), covar=tensor([0.5394, 0.3020, 0.3605, 0.5926], device='cuda:1'), in_proj_covar=tensor([0.1704, 0.1611, 0.1578, 0.1390], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 05:20:20,589 INFO [train.py:968] (1/2) Epoch 20, batch 1650, giga_loss[loss=0.3169, simple_loss=0.3813, pruned_loss=0.1263, over 29032.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3505, pruned_loss=0.09792, over 5709151.24 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3397, pruned_loss=0.08805, over 3455140.90 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3514, pruned_loss=0.09881, over 5689212.83 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:20:29,247 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:39,407 INFO [optim.py:369] (1/2) 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 05:20:41,186 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:43,474 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:20:59,299 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 1700, giga_loss[loss=0.31, simple_loss=0.3654, pruned_loss=0.1273, over 28860.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3521, pruned_loss=0.1008, over 5714195.02 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3395, pruned_loss=0.08806, over 3503159.83 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3531, pruned_loss=0.1018, over 5695515.17 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:21:09,471 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:21:16,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1202, 2.9158, 1.3293, 1.3252], device='cuda:1'), covar=tensor([0.1274, 0.0426, 0.1026, 0.1617], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0544, 0.0376, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:21:31,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=868891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:21:49,679 INFO [train.py:968] (1/2) Epoch 20, batch 1750, giga_loss[loss=0.2538, simple_loss=0.3253, pruned_loss=0.09114, over 28819.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3518, pruned_loss=0.1022, over 5703457.80 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3394, pruned_loss=0.08811, over 3552384.89 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.353, pruned_loss=0.1032, over 5693764.97 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:22:09,454 INFO [optim.py:369] (1/2) 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 05:22:17,627 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:19,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:24,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-10 05:22:33,487 INFO [train.py:968] (1/2) Epoch 20, batch 1800, giga_loss[loss=0.3393, simple_loss=0.3819, pruned_loss=0.1484, over 26645.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3494, pruned_loss=0.1013, over 5686860.25 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3398, pruned_loss=0.0885, over 3588575.92 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3503, pruned_loss=0.1021, over 5684480.01 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:22:36,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:38,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:22:43,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:03,862 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=868995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:04,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=868996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:06,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=868999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:16,893 INFO [train.py:968] (1/2) Epoch 20, batch 1850, giga_loss[loss=0.2665, simple_loss=0.346, pruned_loss=0.09348, over 28788.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3478, pruned_loss=0.1002, over 5688659.60 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3395, pruned_loss=0.08819, over 3635067.86 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3489, pruned_loss=0.1012, over 5682425.60 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:23:30,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:33,474 INFO [optim.py:369] (1/2) 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 05:23:35,188 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:35,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4151, 1.6448, 1.7078, 1.2737], device='cuda:1'), covar=tensor([0.1714, 0.2502, 0.1416, 0.1648], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0704, 0.0942, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 05:23:38,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:23:47,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:23:51,473 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 1900, giga_loss[loss=0.2838, simple_loss=0.3485, pruned_loss=0.1095, over 27631.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3471, pruned_loss=0.09873, over 5680043.37 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3398, pruned_loss=0.08802, over 3698475.53 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.348, pruned_loss=0.1, over 5681153.80 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:24:01,692 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:24:47,512 INFO [train.py:968] (1/2) Epoch 20, batch 1950, libri_loss[loss=0.2401, simple_loss=0.335, pruned_loss=0.07261, over 26095.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3449, pruned_loss=0.0972, over 5677822.04 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3403, pruned_loss=0.08815, over 3749520.42 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3455, pruned_loss=0.09837, over 5677419.90 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:25:04,909 INFO [optim.py:369] (1/2) 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 05:25:20,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4294, 1.5845, 1.2306, 1.2188], device='cuda:1'), covar=tensor([0.1009, 0.0596, 0.1120, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0446, 0.0515, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:25:33,112 INFO [train.py:968] (1/2) Epoch 20, batch 2000, libri_loss[loss=0.2918, simple_loss=0.3518, pruned_loss=0.1159, over 29378.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3404, pruned_loss=0.09507, over 5681369.39 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3411, pruned_loss=0.0888, over 3802431.56 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3404, pruned_loss=0.09579, over 5676415.15 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:26:03,360 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:26:06,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869195.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:26:07,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:26:11,121 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 2050, giga_loss[loss=0.2244, simple_loss=0.3013, pruned_loss=0.07373, over 28830.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3346, pruned_loss=0.09224, over 5673063.60 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.341, pruned_loss=0.0887, over 3843999.07 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3347, pruned_loss=0.09295, over 5665391.25 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:26:35,387 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:26:39,682 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:26:42,304 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 2100, giga_loss[loss=0.2411, simple_loss=0.3073, pruned_loss=0.08738, over 27559.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3312, pruned_loss=0.09081, over 5663138.60 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3411, pruned_loss=0.08883, over 3885355.93 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3309, pruned_loss=0.09135, over 5652466.37 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:27:16,026 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 05:27:49,708 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:27:55,424 INFO [train.py:968] (1/2) Epoch 20, batch 2150, giga_loss[loss=0.2523, simple_loss=0.3284, pruned_loss=0.08807, over 29030.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3321, pruned_loss=0.09018, over 5674779.36 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3415, pruned_loss=0.08892, over 3933683.32 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3315, pruned_loss=0.09058, over 5662538.15 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:27:56,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1865, 3.6757, 1.4086, 1.5355], device='cuda:1'), covar=tensor([0.1170, 0.0408, 0.0988, 0.1539], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0543, 0.0377, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:28:11,233 INFO [optim.py:369] (1/2) 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 05:28:35,543 INFO [train.py:968] (1/2) Epoch 20, batch 2200, giga_loss[loss=0.2569, simple_loss=0.3273, pruned_loss=0.09329, over 28588.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3329, pruned_loss=0.09045, over 5688910.38 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3417, pruned_loss=0.0889, over 3999208.67 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.332, pruned_loss=0.09084, over 5674366.31 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:29:17,557 INFO [train.py:968] (1/2) Epoch 20, batch 2250, giga_loss[loss=0.2271, simple_loss=0.305, pruned_loss=0.07453, over 28714.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.331, pruned_loss=0.08902, over 5697543.10 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.08895, over 4064664.75 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3297, pruned_loss=0.08933, over 5679381.31 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:29:17,833 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:29:22,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2378, 1.5664, 1.3722, 1.4670], device='cuda:1'), covar=tensor([0.0753, 0.0387, 0.0333, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0218], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 05:29:35,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-10 05:29:35,247 INFO [optim.py:369] (1/2) 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 05:29:59,263 INFO [train.py:968] (1/2) Epoch 20, batch 2300, giga_loss[loss=0.222, simple_loss=0.3019, pruned_loss=0.07106, over 28992.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3288, pruned_loss=0.08818, over 5703517.66 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.342, pruned_loss=0.08877, over 4083282.89 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3277, pruned_loss=0.08854, over 5687176.25 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:30:27,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9088, 1.0630, 1.0544, 0.8242], device='cuda:1'), covar=tensor([0.2180, 0.2595, 0.1504, 0.2307], device='cuda:1'), in_proj_covar=tensor([0.1917, 0.1842, 0.1775, 0.1915], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 05:30:42,423 INFO [train.py:968] (1/2) Epoch 20, batch 2350, giga_loss[loss=0.2268, simple_loss=0.3057, pruned_loss=0.074, over 28763.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3268, pruned_loss=0.08725, over 5707017.41 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08886, over 4126510.44 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3253, pruned_loss=0.08745, over 5691186.25 frames. ], batch size: 243, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:30:52,470 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:30:59,695 INFO [optim.py:369] (1/2) 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 05:31:16,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4032, 1.4997, 1.4576, 1.3685], device='cuda:1'), covar=tensor([0.2538, 0.2130, 0.1909, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.1917, 0.1842, 0.1775, 0.1916], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 05:31:25,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3915, 3.1259, 1.4809, 1.4833], device='cuda:1'), covar=tensor([0.1031, 0.0353, 0.0935, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0542, 0.0377, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:31:25,427 INFO [train.py:968] (1/2) Epoch 20, batch 2400, giga_loss[loss=0.2194, simple_loss=0.303, pruned_loss=0.06785, over 28726.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3236, pruned_loss=0.08571, over 5706819.48 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3423, pruned_loss=0.08866, over 4153029.15 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3222, pruned_loss=0.08596, over 5691790.42 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:32:05,920 INFO [train.py:968] (1/2) Epoch 20, batch 2450, giga_loss[loss=0.2546, simple_loss=0.3307, pruned_loss=0.08925, over 28408.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3221, pruned_loss=0.08537, over 5709635.72 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3421, pruned_loss=0.08834, over 4179227.83 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3209, pruned_loss=0.08574, over 5694910.11 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:32:22,232 INFO [optim.py:369] (1/2) 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 05:32:43,046 INFO [train.py:968] (1/2) Epoch 20, batch 2500, giga_loss[loss=0.2219, simple_loss=0.2968, pruned_loss=0.07346, over 28927.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3214, pruned_loss=0.08568, over 5711906.41 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3429, pruned_loss=0.08893, over 4221263.70 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3193, pruned_loss=0.08551, over 5702666.49 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:32:58,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869679.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:33:25,775 INFO [train.py:968] (1/2) Epoch 20, batch 2550, giga_loss[loss=0.2323, simple_loss=0.3135, pruned_loss=0.07559, over 28758.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3184, pruned_loss=0.08407, over 5721745.08 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3427, pruned_loss=0.08895, over 4246433.66 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3166, pruned_loss=0.08386, over 5711776.95 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:33:36,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6292, 1.8568, 1.4687, 1.9991], device='cuda:1'), covar=tensor([0.2601, 0.2701, 0.3031, 0.2424], device='cuda:1'), in_proj_covar=tensor([0.1478, 0.1069, 0.1310, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 05:33:43,382 INFO [optim.py:369] (1/2) 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 05:33:50,971 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=869743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:34:05,986 INFO [train.py:968] (1/2) Epoch 20, batch 2600, giga_loss[loss=0.2465, simple_loss=0.3242, pruned_loss=0.08436, over 27983.00 frames. ], tot_loss[loss=0.2414, simple_loss=0.3167, pruned_loss=0.08306, over 5723685.82 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3428, pruned_loss=0.08886, over 4271289.49 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3149, pruned_loss=0.08287, over 5713444.33 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:34:28,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:34:47,379 INFO [train.py:968] (1/2) Epoch 20, batch 2650, giga_loss[loss=0.2263, simple_loss=0.3035, pruned_loss=0.07451, over 28942.00 frames. ], tot_loss[loss=0.24, simple_loss=0.3154, pruned_loss=0.08228, over 5725001.06 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3427, pruned_loss=0.08849, over 4307178.86 frames. ], giga_tot_loss[loss=0.2389, simple_loss=0.3134, pruned_loss=0.08225, over 5716876.06 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:34:57,173 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869822.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:34:59,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869825.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:35:06,734 INFO [optim.py:369] (1/2) 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 05:35:25,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869854.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:35:31,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3920, 3.4972, 1.4758, 1.5633], device='cuda:1'), covar=tensor([0.0967, 0.0265, 0.0934, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0542, 0.0377, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:35:31,694 INFO [train.py:968] (1/2) Epoch 20, batch 2700, libri_loss[loss=0.232, simple_loss=0.3249, pruned_loss=0.06958, over 29570.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3193, pruned_loss=0.08444, over 5729674.78 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08874, over 4354221.62 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3165, pruned_loss=0.0841, over 5718595.11 frames. ], batch size: 75, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:35:53,815 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4942, 1.5602, 1.3130, 1.0908], device='cuda:1'), covar=tensor([0.1001, 0.0586, 0.1040, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0443, 0.0514, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:36:06,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=869899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:36:18,295 INFO [train.py:968] (1/2) Epoch 20, batch 2750, giga_loss[loss=0.2487, simple_loss=0.3217, pruned_loss=0.08787, over 28827.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3237, pruned_loss=0.08719, over 5722778.58 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3435, pruned_loss=0.08871, over 4369717.50 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3212, pruned_loss=0.08691, over 5712516.53 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:36:34,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=869929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:36:36,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=869932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:36:38,068 INFO [optim.py:369] (1/2) 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 05:37:01,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-10 05:37:04,802 INFO [train.py:968] (1/2) Epoch 20, batch 2800, giga_loss[loss=0.3015, simple_loss=0.3701, pruned_loss=0.1164, over 28824.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3311, pruned_loss=0.09143, over 5714628.98 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3441, pruned_loss=0.08882, over 4404946.68 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3283, pruned_loss=0.09113, over 5710251.36 frames. ], batch size: 243, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:37:05,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=869961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:37:11,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7973, 2.0307, 1.9717, 1.8414], device='cuda:1'), covar=tensor([0.1950, 0.1966, 0.2196, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0742, 0.0708, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 05:37:48,383 INFO [train.py:968] (1/2) Epoch 20, batch 2850, giga_loss[loss=0.2937, simple_loss=0.3714, pruned_loss=0.108, over 28930.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3377, pruned_loss=0.09572, over 5704020.70 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3436, pruned_loss=0.08853, over 4463229.93 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3354, pruned_loss=0.09591, over 5693936.34 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:38:02,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3878, 3.4808, 1.5597, 1.4513], device='cuda:1'), covar=tensor([0.1028, 0.0271, 0.0908, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0544, 0.0376, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:38:10,873 INFO [optim.py:369] (1/2) 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 05:38:15,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 05:38:20,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:38:22,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:38:37,572 INFO [train.py:968] (1/2) Epoch 20, batch 2900, giga_loss[loss=0.2611, simple_loss=0.3485, pruned_loss=0.08687, over 29127.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.341, pruned_loss=0.09628, over 5702936.12 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.344, pruned_loss=0.08878, over 4474237.78 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3389, pruned_loss=0.09634, over 5702582.33 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:38:53,377 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:39:24,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5409, 3.4381, 1.6546, 1.5671], device='cuda:1'), covar=tensor([0.1012, 0.0275, 0.0894, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0543, 0.0376, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:39:24,978 INFO [train.py:968] (1/2) Epoch 20, batch 2950, giga_loss[loss=0.2954, simple_loss=0.3683, pruned_loss=0.1112, over 28265.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3451, pruned_loss=0.09764, over 5700888.08 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3433, pruned_loss=0.08829, over 4512110.29 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3439, pruned_loss=0.09828, over 5699090.43 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:39:30,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=870118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:39:45,603 INFO [optim.py:369] (1/2) 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 05:40:12,112 INFO [train.py:968] (1/2) Epoch 20, batch 3000, giga_loss[loss=0.3303, simple_loss=0.3995, pruned_loss=0.1306, over 28744.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3518, pruned_loss=0.1022, over 5686104.15 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3434, pruned_loss=0.08828, over 4551252.73 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3509, pruned_loss=0.1031, over 5680952.27 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:40:12,112 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 05:40:20,895 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 05:40:24,442 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:40:52,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0677, 4.8667, 4.5940, 2.3747], device='cuda:1'), covar=tensor([0.0467, 0.0561, 0.0616, 0.1859], device='cuda:1'), in_proj_covar=tensor([0.1175, 0.1089, 0.0922, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 05:41:05,821 INFO [train.py:968] (1/2) Epoch 20, batch 3050, libri_loss[loss=0.2651, simple_loss=0.3377, pruned_loss=0.09619, over 29531.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3521, pruned_loss=0.1021, over 5692312.35 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3437, pruned_loss=0.08866, over 4590926.73 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3514, pruned_loss=0.1029, over 5682549.74 frames. ], batch size: 80, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:41:24,856 INFO [optim.py:369] (1/2) 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 05:41:32,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5334, 1.8996, 1.7652, 1.5630], device='cuda:1'), covar=tensor([0.1971, 0.1955, 0.2210, 0.2115], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0739, 0.0705, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 05:41:47,080 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870260.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:41:47,490 INFO [train.py:968] (1/2) Epoch 20, batch 3100, giga_loss[loss=0.2699, simple_loss=0.3483, pruned_loss=0.09572, over 29014.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3484, pruned_loss=0.09932, over 5696144.36 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.343, pruned_loss=0.08831, over 4613398.00 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3485, pruned_loss=0.1004, over 5687756.53 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:41:47,743 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:41:50,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:42:15,737 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:42:34,342 INFO [train.py:968] (1/2) Epoch 20, batch 3150, giga_loss[loss=0.2946, simple_loss=0.3675, pruned_loss=0.1108, over 27522.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3465, pruned_loss=0.0974, over 5702363.77 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3428, pruned_loss=0.08828, over 4635785.83 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3468, pruned_loss=0.09846, over 5695075.77 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:42:52,895 INFO [optim.py:369] (1/2) 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 05:42:54,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 05:43:19,734 INFO [train.py:968] (1/2) Epoch 20, batch 3200, giga_loss[loss=0.2731, simple_loss=0.3501, pruned_loss=0.09803, over 28924.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09676, over 5705780.61 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3428, pruned_loss=0.08833, over 4647962.80 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3464, pruned_loss=0.09764, over 5698766.18 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:43:40,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 05:44:04,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-10 05:44:04,884 INFO [train.py:968] (1/2) Epoch 20, batch 3250, giga_loss[loss=0.2609, simple_loss=0.3409, pruned_loss=0.09045, over 28851.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3482, pruned_loss=0.09764, over 5711831.15 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.343, pruned_loss=0.08843, over 4673143.81 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3484, pruned_loss=0.09843, over 5702334.25 frames. ], batch size: 199, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:44:26,440 INFO [optim.py:369] (1/2) 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 05:44:48,837 INFO [train.py:968] (1/2) Epoch 20, batch 3300, giga_loss[loss=0.2734, simple_loss=0.3518, pruned_loss=0.09753, over 28881.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3491, pruned_loss=0.09811, over 5700989.77 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3432, pruned_loss=0.08842, over 4687440.32 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09892, over 5699937.20 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:44:56,128 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 05:45:01,440 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 05:45:21,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4236, 1.4885, 1.4052, 1.3783], device='cuda:1'), covar=tensor([0.1992, 0.1936, 0.1953, 0.1945], device='cuda:1'), in_proj_covar=tensor([0.1902, 0.1834, 0.1771, 0.1904], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 05:45:32,677 INFO [train.py:968] (1/2) Epoch 20, batch 3350, giga_loss[loss=0.2482, simple_loss=0.3307, pruned_loss=0.08287, over 28788.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3504, pruned_loss=0.09938, over 5705717.81 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3438, pruned_loss=0.08875, over 4723398.20 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3501, pruned_loss=0.1001, over 5699547.87 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:45:50,254 INFO [optim.py:369] (1/2) 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 05:45:55,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=870541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:46:14,263 INFO [train.py:968] (1/2) Epoch 20, batch 3400, giga_loss[loss=0.2666, simple_loss=0.3412, pruned_loss=0.09599, over 28515.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3512, pruned_loss=0.1003, over 5712415.39 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3436, pruned_loss=0.08873, over 4770799.32 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3514, pruned_loss=0.1013, over 5703667.14 frames. ], batch size: 78, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:46:52,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.65 vs. limit=5.0
+2023-03-10 05:47:00,198 INFO [train.py:968] (1/2) Epoch 20, batch 3450, giga_loss[loss=0.2549, simple_loss=0.3397, pruned_loss=0.08507, over 28921.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3518, pruned_loss=0.1008, over 5720094.82 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3438, pruned_loss=0.08878, over 4786014.73 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.352, pruned_loss=0.1018, over 5712149.95 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:47:06,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2094, 2.5431, 1.2423, 1.3748], device='cuda:1'), covar=tensor([0.0992, 0.0332, 0.0873, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0544, 0.0377, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:47:08,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3660, 1.5626, 1.6261, 1.2116], device='cuda:1'), covar=tensor([0.1760, 0.2524, 0.1427, 0.1679], device='cuda:1'), in_proj_covar=tensor([0.0891, 0.0698, 0.0937, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 05:47:20,280 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:47:39,431 INFO [train.py:968] (1/2) Epoch 20, batch 3500, giga_loss[loss=0.2964, simple_loss=0.3722, pruned_loss=0.1103, over 28938.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3515, pruned_loss=0.1, over 5725215.55 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3438, pruned_loss=0.08879, over 4824589.61 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3519, pruned_loss=0.1011, over 5712913.21 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:47:58,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:48:01,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:48:21,521 INFO [train.py:968] (1/2) Epoch 20, batch 3550, giga_loss[loss=0.2975, simple_loss=0.376, pruned_loss=0.1095, over 29025.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3515, pruned_loss=0.09894, over 5717973.44 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.344, pruned_loss=0.0889, over 4836408.30 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3517, pruned_loss=0.09989, over 5709503.56 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 05:48:26,117 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:48:42,125 INFO [optim.py:369] (1/2) 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 05:49:02,327 INFO [train.py:968] (1/2) Epoch 20, batch 3600, giga_loss[loss=0.2558, simple_loss=0.3353, pruned_loss=0.08811, over 28642.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3523, pruned_loss=0.09879, over 5725585.17 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3441, pruned_loss=0.08902, over 4867843.74 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3526, pruned_loss=0.09973, over 5714117.21 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:49:18,607 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=870778.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 05:49:21,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=870781.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:49:25,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-10 05:49:33,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-10 05:49:42,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=870810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 05:49:43,221 INFO [train.py:968] (1/2) Epoch 20, batch 3650, giga_loss[loss=0.3142, simple_loss=0.378, pruned_loss=0.1252, over 29008.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3506, pruned_loss=0.09784, over 5725939.48 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08872, over 4887347.76 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3514, pruned_loss=0.09898, over 5714667.88 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:49:59,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:50:03,528 INFO [optim.py:369] (1/2) 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 05:50:25,551 INFO [train.py:968] (1/2) Epoch 20, batch 3700, giga_loss[loss=0.2625, simple_loss=0.3152, pruned_loss=0.1049, over 23684.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3483, pruned_loss=0.09727, over 5721735.80 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3437, pruned_loss=0.08905, over 4912204.89 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3491, pruned_loss=0.09819, over 5712904.68 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:50:27,658 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=870864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:51:02,114 INFO [train.py:968] (1/2) Epoch 20, batch 3750, libri_loss[loss=0.2635, simple_loss=0.3387, pruned_loss=0.09411, over 29562.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3465, pruned_loss=0.09615, over 5727216.57 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3442, pruned_loss=0.08926, over 4945920.58 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3468, pruned_loss=0.09697, over 5714893.32 frames. ], batch size: 78, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:51:22,342 INFO [optim.py:369] (1/2) 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 05:51:47,411 INFO [train.py:968] (1/2) Epoch 20, batch 3800, giga_loss[loss=0.2338, simple_loss=0.3173, pruned_loss=0.07511, over 28248.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.346, pruned_loss=0.09605, over 5726137.34 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3442, pruned_loss=0.08934, over 4946246.74 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3463, pruned_loss=0.09668, over 5723524.55 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:52:29,783 INFO [train.py:968] (1/2) Epoch 20, batch 3850, giga_loss[loss=0.297, simple_loss=0.3658, pruned_loss=0.1141, over 28889.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3464, pruned_loss=0.09674, over 5724947.94 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3432, pruned_loss=0.08882, over 4973388.60 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3474, pruned_loss=0.09786, over 5722675.52 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:52:32,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3650, 1.0852, 4.1865, 3.3904], device='cuda:1'), covar=tensor([0.1570, 0.2740, 0.0369, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0739, 0.0631, 0.0930, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:52:34,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6019, 1.6463, 1.8107, 1.3998], device='cuda:1'), covar=tensor([0.1906, 0.2593, 0.1508, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0702, 0.0939, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 05:52:41,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2173, 1.4534, 1.3606, 1.0915], device='cuda:1'), covar=tensor([0.2580, 0.2361, 0.1653, 0.2311], device='cuda:1'), in_proj_covar=tensor([0.1915, 0.1850, 0.1785, 0.1915], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 05:52:48,047 INFO [optim.py:369] (1/2) 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 05:52:50,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 05:53:11,084 INFO [train.py:968] (1/2) Epoch 20, batch 3900, giga_loss[loss=0.2403, simple_loss=0.3239, pruned_loss=0.07836, over 29087.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3472, pruned_loss=0.09714, over 5723956.16 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3431, pruned_loss=0.08882, over 4992104.77 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3481, pruned_loss=0.09817, over 5718640.56 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:53:55,432 INFO [train.py:968] (1/2) Epoch 20, batch 3950, giga_loss[loss=0.3015, simple_loss=0.3684, pruned_loss=0.1173, over 28658.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3464, pruned_loss=0.096, over 5722726.63 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.08872, over 5006139.89 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3474, pruned_loss=0.097, over 5715869.23 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:54:16,724 INFO [optim.py:369] (1/2) 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 05:54:16,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:54:34,420 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 05:54:39,433 INFO [train.py:968] (1/2) Epoch 20, batch 4000, giga_loss[loss=0.2665, simple_loss=0.3433, pruned_loss=0.09488, over 28868.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3455, pruned_loss=0.09568, over 5724731.56 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.08867, over 5010623.19 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3463, pruned_loss=0.09653, over 5718582.68 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:55:14,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:55:19,262 INFO [train.py:968] (1/2) Epoch 20, batch 4050, giga_loss[loss=0.243, simple_loss=0.3187, pruned_loss=0.08366, over 29086.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3436, pruned_loss=0.09532, over 5716574.34 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3427, pruned_loss=0.08855, over 5023519.19 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3444, pruned_loss=0.09616, over 5709817.39 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:55:37,592 INFO [optim.py:369] (1/2) 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 05:55:41,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:55:59,150 INFO [train.py:968] (1/2) Epoch 20, batch 4100, giga_loss[loss=0.2535, simple_loss=0.3284, pruned_loss=0.08931, over 28700.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3407, pruned_loss=0.09387, over 5711201.77 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3428, pruned_loss=0.08856, over 5034829.65 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3412, pruned_loss=0.09461, over 5705073.55 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:56:02,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 05:56:04,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-10 05:56:21,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6053, 1.7845, 1.8587, 1.4020], device='cuda:1'), covar=tensor([0.1852, 0.2530, 0.1454, 0.1736], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0704, 0.0942, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 05:56:22,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2596, 1.1301, 4.0591, 3.2285], device='cuda:1'), covar=tensor([0.1694, 0.2913, 0.0411, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0738, 0.0631, 0.0927, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 05:56:39,652 INFO [train.py:968] (1/2) Epoch 20, batch 4150, giga_loss[loss=0.2666, simple_loss=0.3392, pruned_loss=0.097, over 29075.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3396, pruned_loss=0.09396, over 5708862.14 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3422, pruned_loss=0.08821, over 5046765.25 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3404, pruned_loss=0.09489, over 5702371.89 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 05:56:59,239 INFO [optim.py:369] (1/2) 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 05:57:09,656 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=871348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:11,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=871351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:17,788 INFO [train.py:968] (1/2) Epoch 20, batch 4200, giga_loss[loss=0.2476, simple_loss=0.32, pruned_loss=0.08758, over 28681.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3393, pruned_loss=0.09403, over 5695984.84 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.342, pruned_loss=0.08813, over 5067884.49 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.34, pruned_loss=0.09511, over 5701068.22 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:57:33,464 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=871380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:37,410 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=871382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:39,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=871385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:57:57,335 INFO [train.py:968] (1/2) Epoch 20, batch 4250, giga_loss[loss=0.2547, simple_loss=0.3272, pruned_loss=0.09112, over 28634.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3378, pruned_loss=0.09344, over 5704849.10 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3428, pruned_loss=0.08878, over 5094736.69 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3376, pruned_loss=0.09398, over 5704270.33 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:58:01,983 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:58:39,875 INFO [train.py:968] (1/2) Epoch 20, batch 4300, giga_loss[loss=0.2843, simple_loss=0.3453, pruned_loss=0.1116, over 28763.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3352, pruned_loss=0.09224, over 5700517.99 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3429, pruned_loss=0.08879, over 5117928.54 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3348, pruned_loss=0.09283, over 5703788.43 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:58:56,761 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:59:15,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5557, 3.7625, 1.6335, 1.6940], device='cuda:1'), covar=tensor([0.0921, 0.0374, 0.0916, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0546, 0.0378, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 05:59:19,098 INFO [train.py:968] (1/2) Epoch 20, batch 4350, giga_loss[loss=0.2355, simple_loss=0.3126, pruned_loss=0.0792, over 28430.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3345, pruned_loss=0.09247, over 5699824.44 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.08893, over 5132317.56 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3336, pruned_loss=0.09291, over 5702781.50 frames. ], batch size: 65, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 05:59:19,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 05:59:28,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5474, 1.6925, 1.6972, 1.4823], device='cuda:1'), covar=tensor([0.3215, 0.2556, 0.2082, 0.2511], device='cuda:1'), in_proj_covar=tensor([0.1902, 0.1843, 0.1769, 0.1905], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 05:59:38,362 INFO [optim.py:369] (1/2) 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 05:59:38,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-10 05:59:57,908 INFO [train.py:968] (1/2) Epoch 20, batch 4400, giga_loss[loss=0.2479, simple_loss=0.3224, pruned_loss=0.0867, over 29056.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3319, pruned_loss=0.09097, over 5710522.77 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3437, pruned_loss=0.08919, over 5154691.02 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3306, pruned_loss=0.09119, over 5707894.93 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:00:38,261 INFO [train.py:968] (1/2) Epoch 20, batch 4450, giga_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1012, over 29103.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3314, pruned_loss=0.09067, over 5706892.10 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3441, pruned_loss=0.08937, over 5155101.29 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3299, pruned_loss=0.09072, over 5711238.90 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:01:01,073 INFO [optim.py:369] (1/2) 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 06:01:18,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=871654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:01:21,081 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 4500, giga_loss[loss=0.2502, simple_loss=0.3228, pruned_loss=0.08877, over 28909.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3365, pruned_loss=0.09343, over 5702456.35 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3445, pruned_loss=0.08961, over 5178893.63 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3347, pruned_loss=0.09339, over 5701509.18 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:01:44,151 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=871686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:02:03,365 INFO [train.py:968] (1/2) Epoch 20, batch 4550, giga_loss[loss=0.233, simple_loss=0.3079, pruned_loss=0.07902, over 28460.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.338, pruned_loss=0.09375, over 5689957.98 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3441, pruned_loss=0.08966, over 5196598.16 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3365, pruned_loss=0.09386, over 5698748.96 frames. ], batch size: 78, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:02:13,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2965, 1.5185, 1.2568, 1.4361], device='cuda:1'), covar=tensor([0.0740, 0.0366, 0.0367, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 06:02:25,512 INFO [optim.py:369] (1/2) 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 06:02:44,361 INFO [train.py:968] (1/2) Epoch 20, batch 4600, giga_loss[loss=0.2668, simple_loss=0.3515, pruned_loss=0.09104, over 28961.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3406, pruned_loss=0.09453, over 5698551.90 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08979, over 5214298.39 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3394, pruned_loss=0.09461, over 5703077.93 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:03:28,968 INFO [train.py:968] (1/2) Epoch 20, batch 4650, giga_loss[loss=0.247, simple_loss=0.3288, pruned_loss=0.08262, over 28800.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3418, pruned_loss=0.09459, over 5689052.74 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3444, pruned_loss=0.09002, over 5231123.56 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3404, pruned_loss=0.09459, over 5687974.53 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:03:39,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3587, 2.8668, 1.5030, 1.4426], device='cuda:1'), covar=tensor([0.0883, 0.0309, 0.0880, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0545, 0.0377, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 06:03:53,177 INFO [optim.py:369] (1/2) 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 06:04:08,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=871859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:04:09,963 INFO [train.py:968] (1/2) Epoch 20, batch 4700, giga_loss[loss=0.2478, simple_loss=0.3317, pruned_loss=0.08193, over 28627.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3408, pruned_loss=0.09344, over 5702940.90 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08999, over 5262678.51 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3395, pruned_loss=0.09363, over 5693651.51 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:04:22,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:04:34,848 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=871890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:04:51,550 INFO [train.py:968] (1/2) Epoch 20, batch 4750, giga_loss[loss=0.2677, simple_loss=0.3445, pruned_loss=0.09552, over 28611.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3399, pruned_loss=0.09338, over 5699791.82 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.09, over 5268218.78 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3388, pruned_loss=0.0936, over 5697483.64 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:05:12,837 INFO [optim.py:369] (1/2) 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 06:05:29,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3168, 1.6047, 1.2845, 1.0479], device='cuda:1'), covar=tensor([0.2979, 0.2986, 0.3430, 0.2566], device='cuda:1'), in_proj_covar=tensor([0.1474, 0.1067, 0.1307, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 06:05:32,531 INFO [train.py:968] (1/2) Epoch 20, batch 4800, giga_loss[loss=0.2807, simple_loss=0.3424, pruned_loss=0.1095, over 28651.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3415, pruned_loss=0.09442, over 5700460.89 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3445, pruned_loss=0.09006, over 5285782.18 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3403, pruned_loss=0.09465, over 5694119.43 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:06:01,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4994, 1.5304, 1.7075, 1.3594], device='cuda:1'), covar=tensor([0.1493, 0.2101, 0.1295, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0696, 0.0935, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 06:06:06,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:06:10,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:06:15,020 INFO [train.py:968] (1/2) Epoch 20, batch 4850, giga_loss[loss=0.3007, simple_loss=0.3712, pruned_loss=0.1151, over 28723.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3434, pruned_loss=0.09564, over 5701631.09 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3446, pruned_loss=0.09019, over 5298380.68 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3423, pruned_loss=0.09579, over 5692750.49 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:06:36,253 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:06:40,036 INFO [optim.py:369] (1/2) 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 06:06:53,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-10 06:07:01,850 INFO [train.py:968] (1/2) Epoch 20, batch 4900, giga_loss[loss=0.2705, simple_loss=0.3498, pruned_loss=0.09555, over 28692.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3464, pruned_loss=0.09724, over 5702012.27 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3445, pruned_loss=0.09013, over 5301459.77 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3457, pruned_loss=0.09745, over 5694142.31 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:07:15,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1550, 1.3901, 1.3945, 1.0366], device='cuda:1'), covar=tensor([0.1708, 0.2547, 0.1444, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0696, 0.0935, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 06:07:44,603 INFO [train.py:968] (1/2) Epoch 20, batch 4950, giga_loss[loss=0.2689, simple_loss=0.3479, pruned_loss=0.09494, over 28832.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3495, pruned_loss=0.09843, over 5715877.97 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3449, pruned_loss=0.09021, over 5313790.02 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3487, pruned_loss=0.09868, over 5705702.29 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:08:02,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3611, 1.0243, 4.5615, 3.3783], device='cuda:1'), covar=tensor([0.1800, 0.3119, 0.0393, 0.1043], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0633, 0.0935, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 06:08:08,279 INFO [optim.py:369] (1/2) 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 06:08:10,734 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 5000, giga_loss[loss=0.2639, simple_loss=0.3336, pruned_loss=0.09705, over 23852.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3507, pruned_loss=0.09907, over 5714966.23 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3452, pruned_loss=0.09029, over 5327898.74 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3499, pruned_loss=0.0994, over 5702992.43 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:08:40,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8204, 1.8429, 1.3912, 1.4189], device='cuda:1'), covar=tensor([0.0826, 0.0630, 0.0987, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0442, 0.0511, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 06:09:06,438 INFO [train.py:968] (1/2) Epoch 20, batch 5050, giga_loss[loss=0.2713, simple_loss=0.3557, pruned_loss=0.09345, over 28878.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3508, pruned_loss=0.09864, over 5724792.01 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3459, pruned_loss=0.09057, over 5344186.08 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3497, pruned_loss=0.0989, over 5710842.94 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:09:27,661 INFO [optim.py:369] (1/2) 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 06:09:36,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:09:37,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4743, 1.5520, 1.6869, 1.2809], device='cuda:1'), covar=tensor([0.1911, 0.2545, 0.1548, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0699, 0.0937, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 06:09:46,249 INFO [train.py:968] (1/2) Epoch 20, batch 5100, giga_loss[loss=0.2618, simple_loss=0.3416, pruned_loss=0.09101, over 29013.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.349, pruned_loss=0.09765, over 5727278.83 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3458, pruned_loss=0.09058, over 5352955.01 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3483, pruned_loss=0.09801, over 5716945.28 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:09:50,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:10:03,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-10 06:10:11,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5051, 4.3291, 4.1375, 2.0523], device='cuda:1'), covar=tensor([0.0491, 0.0648, 0.0666, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.1174, 0.1093, 0.0927, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 06:10:27,555 INFO [train.py:968] (1/2) Epoch 20, batch 5150, giga_loss[loss=0.2734, simple_loss=0.3477, pruned_loss=0.09959, over 28583.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3479, pruned_loss=0.09746, over 5721945.83 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3461, pruned_loss=0.09081, over 5366256.85 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.09774, over 5711314.33 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:10:51,879 INFO [optim.py:369] (1/2) 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 06:11:11,305 INFO [train.py:968] (1/2) Epoch 20, batch 5200, giga_loss[loss=0.251, simple_loss=0.3214, pruned_loss=0.09028, over 28916.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3442, pruned_loss=0.09587, over 5727444.70 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3462, pruned_loss=0.09083, over 5368494.56 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3435, pruned_loss=0.0961, over 5718743.27 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:11:37,808 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:41,419 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:50,955 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:52,557 INFO [train.py:968] (1/2) Epoch 20, batch 5250, giga_loss[loss=0.2635, simple_loss=0.3493, pruned_loss=0.08889, over 28898.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.0945, over 5724165.97 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3465, pruned_loss=0.09092, over 5375565.44 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.341, pruned_loss=0.09472, over 5719808.88 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:11:53,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:11:59,334 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:12:03,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4922, 3.5880, 1.5821, 1.6544], device='cuda:1'), covar=tensor([0.0952, 0.0304, 0.0942, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0545, 0.0377, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 06:12:03,940 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:12:04,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6597, 1.7897, 1.5118, 1.7006], device='cuda:1'), covar=tensor([0.2578, 0.2769, 0.3092, 0.2494], device='cuda:1'), in_proj_covar=tensor([0.1467, 0.1063, 0.1301, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 06:12:05,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-10 06:12:12,036 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:12:14,467 INFO [optim.py:369] (1/2) 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 06:12:15,208 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 5300, giga_loss[loss=0.2555, simple_loss=0.3452, pruned_loss=0.08292, over 28877.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.342, pruned_loss=0.09371, over 5720704.17 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3464, pruned_loss=0.09088, over 5388875.15 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.094, over 5712921.14 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:13:13,669 INFO [train.py:968] (1/2) Epoch 20, batch 5350, giga_loss[loss=0.3137, simple_loss=0.3761, pruned_loss=0.1256, over 26780.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3441, pruned_loss=0.09387, over 5713129.91 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3468, pruned_loss=0.09121, over 5406613.48 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.343, pruned_loss=0.09395, over 5704773.47 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:13:17,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:13:35,133 INFO [optim.py:369] (1/2) 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 06:13:54,706 INFO [train.py:968] (1/2) Epoch 20, batch 5400, giga_loss[loss=0.2615, simple_loss=0.3387, pruned_loss=0.09218, over 29054.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3449, pruned_loss=0.09481, over 5701903.53 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.347, pruned_loss=0.09142, over 5410149.32 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3437, pruned_loss=0.09477, over 5700175.27 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:13:58,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-10 06:14:22,208 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:14:29,280 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:14:34,687 INFO [train.py:968] (1/2) Epoch 20, batch 5450, giga_loss[loss=0.3049, simple_loss=0.3711, pruned_loss=0.1194, over 28520.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3451, pruned_loss=0.09634, over 5707821.06 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3467, pruned_loss=0.0912, over 5421782.15 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3445, pruned_loss=0.09662, over 5702364.95 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:14:57,113 INFO [optim.py:369] (1/2) 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 06:15:15,341 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:15:16,540 INFO [train.py:968] (1/2) Epoch 20, batch 5500, giga_loss[loss=0.2776, simple_loss=0.3579, pruned_loss=0.09859, over 28627.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3439, pruned_loss=0.09689, over 5705920.02 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.347, pruned_loss=0.09129, over 5433671.13 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3431, pruned_loss=0.09716, over 5697225.63 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:15:18,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:15:41,700 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:15:58,139 INFO [train.py:968] (1/2) Epoch 20, batch 5550, giga_loss[loss=0.2294, simple_loss=0.3044, pruned_loss=0.07719, over 28824.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3424, pruned_loss=0.09723, over 5698414.89 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3472, pruned_loss=0.09149, over 5434275.79 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3415, pruned_loss=0.0974, over 5696808.18 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:16:19,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 06:16:22,051 INFO [optim.py:369] (1/2) 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 06:16:42,809 INFO [train.py:968] (1/2) Epoch 20, batch 5600, libri_loss[loss=0.2185, simple_loss=0.3017, pruned_loss=0.06767, over 29385.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3396, pruned_loss=0.09632, over 5703261.64 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3469, pruned_loss=0.09134, over 5438825.30 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.339, pruned_loss=0.09664, over 5700268.28 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:17:10,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:17:26,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:17:27,034 INFO [train.py:968] (1/2) Epoch 20, batch 5650, giga_loss[loss=0.2805, simple_loss=0.3466, pruned_loss=0.1072, over 28311.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3374, pruned_loss=0.09498, over 5711027.23 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.347, pruned_loss=0.09141, over 5443653.11 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3368, pruned_loss=0.09521, over 5706822.70 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:17:51,965 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 5700, giga_loss[loss=0.1996, simple_loss=0.2706, pruned_loss=0.06427, over 28844.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3319, pruned_loss=0.09218, over 5720614.70 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3471, pruned_loss=0.09157, over 5452698.70 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3312, pruned_loss=0.09226, over 5713932.72 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:18:22,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:18:49,080 INFO [train.py:968] (1/2) Epoch 20, batch 5750, libri_loss[loss=0.2834, simple_loss=0.3673, pruned_loss=0.09976, over 27767.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3297, pruned_loss=0.09074, over 5725529.76 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3476, pruned_loss=0.09184, over 5466946.94 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.328, pruned_loss=0.09053, over 5715439.60 frames. ], batch size: 116, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:19:07,482 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:09,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=872938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:11,375 INFO [optim.py:369] (1/2) 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 06:19:23,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=872953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:26,037 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 5800, giga_loss[loss=0.2596, simple_loss=0.3355, pruned_loss=0.0919, over 28830.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3301, pruned_loss=0.09093, over 5726137.07 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3481, pruned_loss=0.09214, over 5475596.74 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3281, pruned_loss=0.09051, over 5715098.79 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:19:33,740 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:34,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:35,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4858, 1.8295, 1.7863, 1.3700], device='cuda:1'), covar=tensor([0.3471, 0.2443, 0.2694, 0.3054], device='cuda:1'), in_proj_covar=tensor([0.1920, 0.1856, 0.1780, 0.1912], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 06:19:41,312 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=872977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:42,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=872979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:19:46,963 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=872985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:20:09,018 INFO [train.py:968] (1/2) Epoch 20, batch 5850, giga_loss[loss=0.2596, simple_loss=0.3383, pruned_loss=0.09047, over 28416.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.334, pruned_loss=0.09265, over 5718889.90 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3483, pruned_loss=0.09243, over 5475750.91 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3316, pruned_loss=0.09203, over 5718020.08 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:20:17,354 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:20:33,920 INFO [optim.py:369] (1/2) 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 06:20:51,684 INFO [train.py:968] (1/2) Epoch 20, batch 5900, giga_loss[loss=0.2755, simple_loss=0.3554, pruned_loss=0.09776, over 28658.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3372, pruned_loss=0.09392, over 5717050.90 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3484, pruned_loss=0.09239, over 5479707.16 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.335, pruned_loss=0.09347, over 5717086.10 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:21:02,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 06:21:31,334 INFO [train.py:968] (1/2) Epoch 20, batch 5950, giga_loss[loss=0.2423, simple_loss=0.3175, pruned_loss=0.08354, over 28513.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3408, pruned_loss=0.09535, over 5718932.28 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3484, pruned_loss=0.09226, over 5495479.27 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3387, pruned_loss=0.0952, over 5713093.72 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:21:31,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:33,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:40,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:42,588 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:21:56,087 INFO [optim.py:369] (1/2) 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 06:21:58,362 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:22:08,497 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:22:14,861 INFO [train.py:968] (1/2) Epoch 20, batch 6000, giga_loss[loss=0.2685, simple_loss=0.3504, pruned_loss=0.09332, over 28764.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3432, pruned_loss=0.09614, over 5709906.61 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3486, pruned_loss=0.09242, over 5497957.78 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3411, pruned_loss=0.09596, over 5708868.42 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:22:14,861 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 06:22:23,593 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 06:22:50,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-10 06:23:07,640 INFO [train.py:968] (1/2) Epoch 20, batch 6050, giga_loss[loss=0.4042, simple_loss=0.4379, pruned_loss=0.1852, over 27627.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3449, pruned_loss=0.0974, over 5702466.25 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3485, pruned_loss=0.09254, over 5499170.60 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3433, pruned_loss=0.09727, over 5705684.40 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:23:32,676 INFO [optim.py:369] (1/2) 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 06:23:42,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:23:52,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0805, 3.3100, 2.2203, 1.1970], device='cuda:1'), covar=tensor([0.7280, 0.2284, 0.3443, 0.6191], device='cuda:1'), in_proj_covar=tensor([0.1707, 0.1604, 0.1573, 0.1391], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 06:23:52,788 INFO [train.py:968] (1/2) Epoch 20, batch 6100, giga_loss[loss=0.2677, simple_loss=0.342, pruned_loss=0.0967, over 28460.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3508, pruned_loss=0.1021, over 5704862.77 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3483, pruned_loss=0.09246, over 5509032.61 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3496, pruned_loss=0.1023, over 5704217.08 frames. ], batch size: 78, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:24:44,397 INFO [train.py:968] (1/2) Epoch 20, batch 6150, giga_loss[loss=0.2823, simple_loss=0.352, pruned_loss=0.1063, over 28534.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.356, pruned_loss=0.1066, over 5685762.50 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3482, pruned_loss=0.09239, over 5506228.96 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3552, pruned_loss=0.1069, over 5690799.15 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:25:10,426 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:25:15,436 INFO [optim.py:369] (1/2) 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 06:25:23,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 06:25:24,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:25:26,738 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873355.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:25:32,174 INFO [train.py:968] (1/2) Epoch 20, batch 6200, giga_loss[loss=0.3702, simple_loss=0.4185, pruned_loss=0.1609, over 28287.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.364, pruned_loss=0.1126, over 5682521.05 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3483, pruned_loss=0.09258, over 5518217.35 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3636, pruned_loss=0.1133, over 5680036.96 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:25:37,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873366.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:25:47,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5612, 2.0411, 1.2961, 0.9328], device='cuda:1'), covar=tensor([0.5224, 0.2849, 0.2549, 0.5100], device='cuda:1'), in_proj_covar=tensor([0.1707, 0.1606, 0.1576, 0.1392], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 06:25:50,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3533, 3.4385, 1.4923, 1.4177], device='cuda:1'), covar=tensor([0.1014, 0.0337, 0.0885, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0545, 0.0377, 0.0417], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 06:26:06,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:09,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:09,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:22,880 INFO [train.py:968] (1/2) Epoch 20, batch 6250, giga_loss[loss=0.3934, simple_loss=0.4085, pruned_loss=0.1891, over 23567.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3699, pruned_loss=0.1179, over 5676300.45 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3483, pruned_loss=0.09252, over 5525023.73 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.37, pruned_loss=0.1189, over 5671276.02 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:26:25,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6452, 1.6302, 1.9006, 1.4843], device='cuda:1'), covar=tensor([0.1162, 0.1740, 0.0995, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0694, 0.0931, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 06:26:35,250 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:26:47,785 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 6300, giga_loss[loss=0.2842, simple_loss=0.3679, pruned_loss=0.1003, over 29039.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3748, pruned_loss=0.1219, over 5676105.63 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3482, pruned_loss=0.09239, over 5527721.08 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3755, pruned_loss=0.1234, over 5673183.64 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:27:44,047 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:27:45,979 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:27:49,169 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-10 06:27:59,791 INFO [train.py:968] (1/2) Epoch 20, batch 6350, giga_loss[loss=0.3164, simple_loss=0.3894, pruned_loss=0.1218, over 29058.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3792, pruned_loss=0.1258, over 5657580.90 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3486, pruned_loss=0.09279, over 5528846.27 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3801, pruned_loss=0.1275, over 5658691.72 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:28:17,749 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:28:32,725 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:28:34,871 INFO [optim.py:369] (1/2) 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 06:28:35,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:28:49,372 INFO [train.py:968] (1/2) Epoch 20, batch 6400, giga_loss[loss=0.2944, simple_loss=0.3613, pruned_loss=0.1137, over 28751.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3809, pruned_loss=0.1282, over 5653947.81 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3486, pruned_loss=0.09289, over 5540949.44 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3828, pruned_loss=0.1307, over 5647239.84 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:29:01,622 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 6450, giga_loss[loss=0.3723, simple_loss=0.4203, pruned_loss=0.1621, over 27837.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3846, pruned_loss=0.1331, over 5629681.92 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3484, pruned_loss=0.09298, over 5541534.07 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3873, pruned_loss=0.1361, over 5626800.26 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:29:46,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5146, 1.6773, 1.7168, 1.3106], device='cuda:1'), covar=tensor([0.1700, 0.2609, 0.1424, 0.1752], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0695, 0.0931, 0.0828], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 06:30:07,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873635.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:30:16,409 INFO [optim.py:369] (1/2) 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 06:30:37,384 INFO [train.py:968] (1/2) Epoch 20, batch 6500, giga_loss[loss=0.4234, simple_loss=0.4451, pruned_loss=0.2009, over 27565.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3876, pruned_loss=0.1368, over 5613953.70 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3481, pruned_loss=0.09279, over 5548266.01 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3907, pruned_loss=0.1401, over 5607195.77 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:31:24,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=873704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:31:31,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:31:31,961 INFO [train.py:968] (1/2) Epoch 20, batch 6550, giga_loss[loss=0.3979, simple_loss=0.429, pruned_loss=0.1835, over 27574.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3893, pruned_loss=0.1373, over 5619794.98 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3481, pruned_loss=0.09276, over 5551963.81 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3921, pruned_loss=0.1403, over 5611797.87 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:31:53,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873730.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:32:02,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=873741.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:32:02,557 INFO [optim.py:369] (1/2) 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 06:32:21,764 INFO [train.py:968] (1/2) Epoch 20, batch 6600, giga_loss[loss=0.3087, simple_loss=0.368, pruned_loss=0.1247, over 28931.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3883, pruned_loss=0.1373, over 5624787.48 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.348, pruned_loss=0.09278, over 5549948.75 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3915, pruned_loss=0.1407, over 5622371.51 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:33:14,666 INFO [train.py:968] (1/2) Epoch 20, batch 6650, giga_loss[loss=0.3621, simple_loss=0.3946, pruned_loss=0.1648, over 23475.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3869, pruned_loss=0.1368, over 5626923.89 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3479, pruned_loss=0.09273, over 5555808.20 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3902, pruned_loss=0.1404, over 5621202.04 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:33:44,445 INFO [optim.py:369] (1/2) 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 06:33:54,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:33:56,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-10 06:33:57,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:34:01,213 INFO [train.py:968] (1/2) Epoch 20, batch 6700, libri_loss[loss=0.2379, simple_loss=0.3193, pruned_loss=0.07821, over 29566.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3858, pruned_loss=0.1353, over 5637850.23 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3474, pruned_loss=0.09251, over 5568553.72 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3907, pruned_loss=0.1402, over 5624473.08 frames. ], batch size: 76, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:34:14,289 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873873.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:34:18,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873876.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:34:26,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=873884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:34:27,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:34:29,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=873887.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:34:45,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873905.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:34:50,681 INFO [train.py:968] (1/2) Epoch 20, batch 6750, giga_loss[loss=0.3267, simple_loss=0.3803, pruned_loss=0.1366, over 28592.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3865, pruned_loss=0.1345, over 5649956.75 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3475, pruned_loss=0.09245, over 5574849.55 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.391, pruned_loss=0.1393, over 5634798.39 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 06:34:56,560 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=873916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:35:09,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6625, 4.4993, 4.2583, 1.9676], device='cuda:1'), covar=tensor([0.0581, 0.0752, 0.0772, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.1208, 0.1121, 0.0949, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-10 06:35:15,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4143, 1.4221, 1.2862, 1.3251], device='cuda:1'), covar=tensor([0.1486, 0.1570, 0.1496, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.1941, 0.1879, 0.1803, 0.1926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 06:35:23,356 INFO [optim.py:369] (1/2) 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 06:35:42,252 INFO [train.py:968] (1/2) Epoch 20, batch 6800, giga_loss[loss=0.3224, simple_loss=0.39, pruned_loss=0.1274, over 29033.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3874, pruned_loss=0.1348, over 5630691.50 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3476, pruned_loss=0.09253, over 5582907.26 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3919, pruned_loss=0.1396, over 5612417.25 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:36:09,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3399, 1.4778, 1.4212, 1.2910], device='cuda:1'), covar=tensor([0.2412, 0.2218, 0.1964, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.1944, 0.1878, 0.1801, 0.1925], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 06:36:32,320 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 6850, giga_loss[loss=0.2892, simple_loss=0.3627, pruned_loss=0.1078, over 28985.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3853, pruned_loss=0.1328, over 5628893.65 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3476, pruned_loss=0.09246, over 5588003.00 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3893, pruned_loss=0.1372, over 5610603.71 frames. ], batch size: 128, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:36:36,401 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-10 06:37:08,581 INFO [optim.py:369] (1/2) 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 06:37:18,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 06:37:24,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5569, 1.9366, 1.7789, 1.5862], device='cuda:1'), covar=tensor([0.1767, 0.1681, 0.2008, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0742, 0.0707, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 06:37:25,013 INFO [train.py:968] (1/2) Epoch 20, batch 6900, libri_loss[loss=0.2711, simple_loss=0.3548, pruned_loss=0.09372, over 29574.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3824, pruned_loss=0.1289, over 5634969.86 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3476, pruned_loss=0.09244, over 5597633.27 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3869, pruned_loss=0.1337, over 5612380.97 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:37:31,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3921, 1.8079, 1.6484, 1.5644], device='cuda:1'), covar=tensor([0.0790, 0.0307, 0.0309, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 06:37:42,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=874079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:38:07,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3961, 1.4607, 1.3327, 1.5161], device='cuda:1'), covar=tensor([0.0766, 0.0340, 0.0325, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 06:38:15,016 INFO [train.py:968] (1/2) Epoch 20, batch 6950, giga_loss[loss=0.2895, simple_loss=0.3587, pruned_loss=0.1102, over 27542.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3785, pruned_loss=0.1255, over 5640392.19 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3471, pruned_loss=0.09234, over 5594377.40 frames. ], giga_tot_loss[loss=0.3215, simple_loss=0.383, pruned_loss=0.13, over 5626836.55 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:38:47,062 INFO [optim.py:369] (1/2) 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 06:38:58,892 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874153.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:39:03,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=874156.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 06:39:07,145 INFO [train.py:968] (1/2) Epoch 20, batch 7000, giga_loss[loss=0.2666, simple_loss=0.3527, pruned_loss=0.0903, over 28942.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3776, pruned_loss=0.1246, over 5642559.39 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3473, pruned_loss=0.09255, over 5597561.46 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3813, pruned_loss=0.1284, over 5629653.08 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:39:30,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874185.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 06:39:48,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6706, 4.5163, 4.3033, 2.0094], device='cuda:1'), covar=tensor([0.0550, 0.0685, 0.0685, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1211, 0.1124, 0.0954, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-10 06:39:58,925 INFO [train.py:968] (1/2) Epoch 20, batch 7050, libri_loss[loss=0.2601, simple_loss=0.3517, pruned_loss=0.08424, over 29532.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.376, pruned_loss=0.1239, over 5648993.50 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3476, pruned_loss=0.09261, over 5599793.35 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3792, pruned_loss=0.1273, over 5637431.87 frames. ], batch size: 84, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:40:11,540 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:40:14,301 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:40:44,288 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:40:51,150 INFO [train.py:968] (1/2) Epoch 20, batch 7100, giga_loss[loss=0.3269, simple_loss=0.3946, pruned_loss=0.1296, over 28969.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3751, pruned_loss=0.1231, over 5647326.91 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3476, pruned_loss=0.09272, over 5592797.56 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3778, pruned_loss=0.126, over 5645485.44 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:41:03,216 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 06:41:03,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:41:48,951 INFO [train.py:968] (1/2) Epoch 20, batch 7150, giga_loss[loss=0.2742, simple_loss=0.3512, pruned_loss=0.09859, over 28878.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3739, pruned_loss=0.1219, over 5649194.57 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3474, pruned_loss=0.09259, over 5593557.74 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3767, pruned_loss=0.1248, over 5648254.09 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:41:52,882 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:42:23,784 INFO [optim.py:369] (1/2) 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 06:42:35,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1940, 1.5654, 1.3616, 1.3274], device='cuda:1'), covar=tensor([0.2239, 0.2111, 0.2354, 0.2251], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0743, 0.0707, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 06:42:40,644 INFO [train.py:968] (1/2) Epoch 20, batch 7200, giga_loss[loss=0.3763, simple_loss=0.416, pruned_loss=0.1683, over 27634.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3729, pruned_loss=0.1196, over 5664549.30 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09299, over 5598045.29 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3751, pruned_loss=0.1222, over 5661171.42 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:43:37,400 INFO [train.py:968] (1/2) Epoch 20, batch 7250, giga_loss[loss=0.3004, simple_loss=0.3785, pruned_loss=0.1112, over 29038.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3743, pruned_loss=0.1187, over 5661554.42 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3484, pruned_loss=0.09325, over 5603523.12 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3763, pruned_loss=0.1211, over 5655151.79 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:44:10,563 INFO [optim.py:369] (1/2) 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 06:44:24,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0938, 1.1566, 3.3582, 2.9401], device='cuda:1'), covar=tensor([0.1655, 0.2613, 0.0532, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0745, 0.0637, 0.0942, 0.0891], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 06:44:28,795 INFO [train.py:968] (1/2) Epoch 20, batch 7300, giga_loss[loss=0.2878, simple_loss=0.3566, pruned_loss=0.1095, over 28773.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3753, pruned_loss=0.1197, over 5661721.01 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3483, pruned_loss=0.09327, over 5608366.35 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3774, pruned_loss=0.1219, over 5653388.86 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:45:25,565 INFO [train.py:968] (1/2) Epoch 20, batch 7350, giga_loss[loss=0.2772, simple_loss=0.3461, pruned_loss=0.1041, over 28923.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3741, pruned_loss=0.1191, over 5673355.97 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3482, pruned_loss=0.0932, over 5609845.52 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3759, pruned_loss=0.121, over 5665814.42 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:45:54,411 INFO [optim.py:369] (1/2) 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 06:46:10,784 INFO [train.py:968] (1/2) Epoch 20, batch 7400, giga_loss[loss=0.3455, simple_loss=0.3804, pruned_loss=0.1553, over 23861.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3714, pruned_loss=0.118, over 5659264.08 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3481, pruned_loss=0.09307, over 5605329.04 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3737, pruned_loss=0.1204, over 5658332.35 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:46:21,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5430, 4.3700, 4.1994, 2.1763], device='cuda:1'), covar=tensor([0.0584, 0.0728, 0.0736, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.1220, 0.1133, 0.0961, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 06:46:58,118 INFO [train.py:968] (1/2) Epoch 20, batch 7450, giga_loss[loss=0.3003, simple_loss=0.3632, pruned_loss=0.1186, over 28606.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3699, pruned_loss=0.1183, over 5665348.08 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3479, pruned_loss=0.09294, over 5609374.93 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3722, pruned_loss=0.1206, over 5661880.83 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:47:09,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3581, 1.5335, 1.4038, 1.2541], device='cuda:1'), covar=tensor([0.2544, 0.2370, 0.1934, 0.2320], device='cuda:1'), in_proj_covar=tensor([0.1946, 0.1880, 0.1804, 0.1931], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 06:47:27,962 INFO [optim.py:369] (1/2) 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 06:47:30,313 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 7500, giga_loss[loss=0.2954, simple_loss=0.3715, pruned_loss=0.1096, over 28800.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3695, pruned_loss=0.1178, over 5673541.12 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09295, over 5615728.27 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3716, pruned_loss=0.1202, over 5666451.12 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:47:49,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3209, 1.1976, 3.7548, 3.2071], device='cuda:1'), covar=tensor([0.1615, 0.2744, 0.0497, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0744, 0.0636, 0.0942, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 06:47:55,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5027, 1.9182, 1.4489, 1.6438], device='cuda:1'), covar=tensor([0.2733, 0.2603, 0.3198, 0.2372], device='cuda:1'), in_proj_covar=tensor([0.1473, 0.1070, 0.1310, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 06:48:13,754 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:48:15,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=874690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:48:34,653 INFO [train.py:968] (1/2) Epoch 20, batch 7550, giga_loss[loss=0.2674, simple_loss=0.3521, pruned_loss=0.09131, over 28782.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3692, pruned_loss=0.1162, over 5690796.56 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3481, pruned_loss=0.09295, over 5623035.03 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3714, pruned_loss=0.1187, over 5679889.07 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:48:38,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=874715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:49:03,075 INFO [optim.py:369] (1/2) 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 06:49:19,798 INFO [train.py:968] (1/2) Epoch 20, batch 7600, giga_loss[loss=0.285, simple_loss=0.3579, pruned_loss=0.106, over 28610.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3699, pruned_loss=0.116, over 5694938.04 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3481, pruned_loss=0.09305, over 5624491.79 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3724, pruned_loss=0.1187, over 5687939.58 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:49:45,739 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:49:48,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=874793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:03,029 INFO [train.py:968] (1/2) Epoch 20, batch 7650, giga_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 28074.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3697, pruned_loss=0.1163, over 5689842.14 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3485, pruned_loss=0.09329, over 5627210.42 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.372, pruned_loss=0.1188, over 5683690.46 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:50:10,939 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:25,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=874833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:26,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0997, 3.9152, 3.7432, 1.9622], device='cuda:1'), covar=tensor([0.0661, 0.0789, 0.0774, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.1129, 0.0958, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 06:50:27,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=874836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:50:36,036 INFO [optim.py:369] (1/2) 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 06:50:53,267 INFO [train.py:968] (1/2) Epoch 20, batch 7700, giga_loss[loss=0.2767, simple_loss=0.3528, pruned_loss=0.1003, over 28943.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.368, pruned_loss=0.1159, over 5685550.28 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3485, pruned_loss=0.09347, over 5624245.34 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3704, pruned_loss=0.1184, over 5685789.44 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:50:57,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=874865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:51:08,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2283, 2.9666, 1.3881, 1.4605], device='cuda:1'), covar=tensor([0.1026, 0.0351, 0.0875, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0550, 0.0379, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 06:51:43,913 INFO [train.py:968] (1/2) Epoch 20, batch 7750, giga_loss[loss=0.2742, simple_loss=0.3497, pruned_loss=0.09936, over 28913.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3684, pruned_loss=0.1169, over 5673371.68 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3488, pruned_loss=0.09364, over 5622092.78 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3704, pruned_loss=0.1192, over 5676307.13 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:52:16,119 INFO [optim.py:369] (1/2) 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 06:52:32,524 INFO [train.py:968] (1/2) Epoch 20, batch 7800, giga_loss[loss=0.3219, simple_loss=0.3869, pruned_loss=0.1284, over 28311.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.368, pruned_loss=0.1174, over 5675863.02 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3484, pruned_loss=0.09337, over 5618557.91 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3704, pruned_loss=0.12, over 5683208.03 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:53:20,063 INFO [train.py:968] (1/2) Epoch 20, batch 7850, giga_loss[loss=0.3466, simple_loss=0.3896, pruned_loss=0.1518, over 27537.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3669, pruned_loss=0.1169, over 5685953.64 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3483, pruned_loss=0.09325, over 5626398.54 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3696, pruned_loss=0.12, over 5687232.15 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:53:32,013 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:53:33,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:53:45,952 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:53:54,346 INFO [optim.py:369] (1/2) 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 06:54:10,358 INFO [train.py:968] (1/2) Epoch 20, batch 7900, giga_loss[loss=0.3015, simple_loss=0.371, pruned_loss=0.116, over 28830.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3643, pruned_loss=0.1158, over 5690560.00 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.348, pruned_loss=0.09313, over 5629267.92 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3669, pruned_loss=0.1186, over 5689924.82 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:54:13,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:54:19,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0881, 2.2600, 1.5782, 1.7893], device='cuda:1'), covar=tensor([0.0982, 0.0728, 0.1075, 0.1258], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0448, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 06:54:36,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:54:46,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1448, 1.2583, 1.0989, 0.8871], device='cuda:1'), covar=tensor([0.0993, 0.0550, 0.1099, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0447, 0.0516, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 06:54:56,196 INFO [train.py:968] (1/2) Epoch 20, batch 7950, giga_loss[loss=0.298, simple_loss=0.3681, pruned_loss=0.1139, over 28230.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3625, pruned_loss=0.1147, over 5697196.90 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3477, pruned_loss=0.09297, over 5637356.46 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3652, pruned_loss=0.1177, over 5691309.98 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:55:01,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5832, 1.6492, 1.7819, 1.3500], device='cuda:1'), covar=tensor([0.1884, 0.2514, 0.1533, 0.1721], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0700, 0.0933, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 06:55:10,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5469, 1.1983, 4.1334, 3.3807], device='cuda:1'), covar=tensor([0.1589, 0.2869, 0.0437, 0.0954], device='cuda:1'), in_proj_covar=tensor([0.0746, 0.0639, 0.0943, 0.0887], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 06:55:30,913 INFO [optim.py:369] (1/2) 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 06:55:47,002 INFO [train.py:968] (1/2) Epoch 20, batch 8000, giga_loss[loss=0.3298, simple_loss=0.386, pruned_loss=0.1368, over 28264.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3636, pruned_loss=0.1154, over 5688515.44 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3475, pruned_loss=0.09286, over 5642608.98 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3662, pruned_loss=0.1183, over 5680231.97 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 06:55:52,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:55:55,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:24,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:28,159 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:30,816 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 8050, giga_loss[loss=0.2712, simple_loss=0.3478, pruned_loss=0.09729, over 28927.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3643, pruned_loss=0.115, over 5690448.25 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3475, pruned_loss=0.09273, over 5649597.24 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3668, pruned_loss=0.1179, over 5678782.06 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:56:44,655 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:52,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:54,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:56:55,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:57:00,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1920, 2.4009, 1.8455, 1.9558], device='cuda:1'), covar=tensor([0.0953, 0.0686, 0.0977, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0446, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 06:57:00,893 INFO [optim.py:369] (1/2) 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 06:57:16,082 INFO [train.py:968] (1/2) Epoch 20, batch 8100, giga_loss[loss=0.2877, simple_loss=0.356, pruned_loss=0.1097, over 28651.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3644, pruned_loss=0.1141, over 5678874.70 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3476, pruned_loss=0.0929, over 5649785.86 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3672, pruned_loss=0.1174, over 5670806.37 frames. ], batch size: 85, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:57:22,049 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:57:37,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3752, 2.8963, 1.3697, 1.4995], device='cuda:1'), covar=tensor([0.0956, 0.0381, 0.0924, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0551, 0.0379, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 06:58:03,543 INFO [train.py:968] (1/2) Epoch 20, batch 8150, giga_loss[loss=0.2978, simple_loss=0.3632, pruned_loss=0.1162, over 28659.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3662, pruned_loss=0.1152, over 5676684.99 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3482, pruned_loss=0.0932, over 5645841.59 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3683, pruned_loss=0.1182, over 5674001.99 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:58:23,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8172, 2.2741, 2.0684, 1.6431], device='cuda:1'), covar=tensor([0.3154, 0.2238, 0.2511, 0.2829], device='cuda:1'), in_proj_covar=tensor([0.1939, 0.1874, 0.1798, 0.1926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 06:58:37,242 INFO [optim.py:369] (1/2) 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 06:58:48,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3775, 1.7492, 1.3900, 1.5357], device='cuda:1'), covar=tensor([0.0734, 0.0326, 0.0325, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0095, 0.0068, 0.0061, 0.0104], device='cuda:1')
+2023-03-10 06:58:55,044 INFO [train.py:968] (1/2) Epoch 20, batch 8200, giga_loss[loss=0.3911, simple_loss=0.4352, pruned_loss=0.1735, over 27545.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3688, pruned_loss=0.1176, over 5674282.88 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3482, pruned_loss=0.09333, over 5642431.67 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3711, pruned_loss=0.1205, over 5676044.18 frames. ], batch size: 472, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:59:36,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-10 06:59:38,012 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 06:59:50,685 INFO [train.py:968] (1/2) Epoch 20, batch 8250, giga_loss[loss=0.3604, simple_loss=0.4111, pruned_loss=0.1549, over 28817.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3715, pruned_loss=0.1207, over 5676313.71 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.09362, over 5646069.96 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1231, over 5674716.03 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 06:59:50,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:00:22,500 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 8300, giga_loss[loss=0.3081, simple_loss=0.3713, pruned_loss=0.1224, over 28653.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3716, pruned_loss=0.1217, over 5652692.03 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3485, pruned_loss=0.09359, over 5627382.19 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.374, pruned_loss=0.1247, over 5669980.32 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:00:49,562 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=875475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:01:27,861 INFO [train.py:968] (1/2) Epoch 20, batch 8350, giga_loss[loss=0.4188, simple_loss=0.4447, pruned_loss=0.1965, over 26459.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3726, pruned_loss=0.1235, over 5653864.22 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3483, pruned_loss=0.09338, over 5633080.52 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3751, pruned_loss=0.1266, over 5662922.48 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:01:57,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:01:59,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:00,714 INFO [optim.py:369] (1/2) 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 07:02:07,385 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:09,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:13,136 INFO [train.py:968] (1/2) Epoch 20, batch 8400, giga_loss[loss=0.4117, simple_loss=0.4391, pruned_loss=0.1921, over 26426.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3717, pruned_loss=0.1231, over 5649411.60 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09345, over 5628420.16 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.374, pruned_loss=0.126, over 5661166.22 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:02:23,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2136, 2.2245, 1.9201, 2.3422], device='cuda:1'), covar=tensor([0.2174, 0.2491, 0.2691, 0.2179], device='cuda:1'), in_proj_covar=tensor([0.1472, 0.1071, 0.1305, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 07:02:25,178 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:36,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:48,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:02:59,726 INFO [train.py:968] (1/2) Epoch 20, batch 8450, giga_loss[loss=0.3036, simple_loss=0.3781, pruned_loss=0.1145, over 28732.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3713, pruned_loss=0.1225, over 5657881.82 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3481, pruned_loss=0.09339, over 5631148.31 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3735, pruned_loss=0.1251, over 5665082.60 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:03:04,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9789, 4.8148, 4.5835, 2.1387], device='cuda:1'), covar=tensor([0.0489, 0.0628, 0.0663, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.1215, 0.1126, 0.0956, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 07:03:31,274 INFO [optim.py:369] (1/2) 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 07:03:44,662 INFO [train.py:968] (1/2) Epoch 20, batch 8500, giga_loss[loss=0.2812, simple_loss=0.3633, pruned_loss=0.09951, over 28971.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3688, pruned_loss=0.1195, over 5658482.07 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3479, pruned_loss=0.09318, over 5640691.46 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3716, pruned_loss=0.1228, over 5656069.31 frames. ], batch size: 136, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:04:27,300 INFO [train.py:968] (1/2) Epoch 20, batch 8550, giga_loss[loss=0.3643, simple_loss=0.398, pruned_loss=0.1653, over 27640.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3672, pruned_loss=0.1183, over 5669789.01 frames. ], libri_tot_loss[loss=0.2672, simple_loss=0.3479, pruned_loss=0.09325, over 5642512.13 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3698, pruned_loss=0.1214, over 5666997.77 frames. ], batch size: 474, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:04:58,428 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:05:00,214 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:05:15,581 INFO [train.py:968] (1/2) Epoch 20, batch 8600, giga_loss[loss=0.2778, simple_loss=0.3531, pruned_loss=0.1013, over 29011.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.118, over 5674273.87 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3482, pruned_loss=0.09342, over 5645852.83 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3677, pruned_loss=0.1208, over 5669619.48 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:05:22,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5096, 1.6611, 1.6313, 1.3727], device='cuda:1'), covar=tensor([0.3012, 0.2620, 0.1862, 0.2662], device='cuda:1'), in_proj_covar=tensor([0.1942, 0.1875, 0.1808, 0.1930], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 07:05:28,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=875774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:05:32,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2088, 1.2789, 1.1285, 1.1789], device='cuda:1'), covar=tensor([0.1658, 0.1861, 0.1328, 0.1428], device='cuda:1'), in_proj_covar=tensor([0.1941, 0.1875, 0.1807, 0.1928], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 07:05:38,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 07:06:02,690 INFO [train.py:968] (1/2) Epoch 20, batch 8650, giga_loss[loss=0.3961, simple_loss=0.42, pruned_loss=0.1861, over 26557.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3658, pruned_loss=0.1187, over 5661894.11 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3484, pruned_loss=0.09346, over 5647194.31 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.368, pruned_loss=0.1218, over 5657900.35 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:06:06,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2037, 1.5175, 1.4693, 1.0751], device='cuda:1'), covar=tensor([0.1579, 0.2365, 0.1314, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0703, 0.0935, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 07:06:39,875 INFO [optim.py:369] (1/2) 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 07:06:45,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=875850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:06:55,792 INFO [train.py:968] (1/2) Epoch 20, batch 8700, giga_loss[loss=0.3426, simple_loss=0.4042, pruned_loss=0.1405, over 28632.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3675, pruned_loss=0.1201, over 5657432.87 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3482, pruned_loss=0.09335, over 5651913.28 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3697, pruned_loss=0.1231, over 5650374.10 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:07:44,796 INFO [train.py:968] (1/2) Epoch 20, batch 8750, giga_loss[loss=0.2973, simple_loss=0.3573, pruned_loss=0.1186, over 23825.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1203, over 5654964.63 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3483, pruned_loss=0.09347, over 5648144.96 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3731, pruned_loss=0.1232, over 5652733.19 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:08:15,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4364, 1.6445, 1.2944, 1.3162], device='cuda:1'), covar=tensor([0.1011, 0.0525, 0.1057, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0446, 0.0513, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 07:08:15,837 INFO [optim.py:369] (1/2) 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 07:08:21,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6988, 4.5495, 4.2891, 2.1392], device='cuda:1'), covar=tensor([0.0618, 0.0783, 0.0866, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.1213, 0.1126, 0.0954, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-10 07:08:27,912 INFO [train.py:968] (1/2) Epoch 20, batch 8800, giga_loss[loss=0.3152, simple_loss=0.3855, pruned_loss=0.1225, over 28840.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3715, pruned_loss=0.1184, over 5672453.80 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3477, pruned_loss=0.09308, over 5656698.13 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3746, pruned_loss=0.122, over 5663447.52 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:09:00,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=875993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:09:02,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=875996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:09:16,526 INFO [train.py:968] (1/2) Epoch 20, batch 8850, giga_loss[loss=0.3139, simple_loss=0.3853, pruned_loss=0.1212, over 28745.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3732, pruned_loss=0.1195, over 5661174.32 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3477, pruned_loss=0.09322, over 5650188.21 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3761, pruned_loss=0.1228, over 5660427.49 frames. ], batch size: 284, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:09:27,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=876025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:09:35,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-10 07:09:40,594 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-10 07:09:47,385 INFO [optim.py:369] (1/2) 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 07:10:04,974 INFO [train.py:968] (1/2) Epoch 20, batch 8900, giga_loss[loss=0.345, simple_loss=0.3973, pruned_loss=0.1464, over 27838.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3753, pruned_loss=0.1219, over 5652589.62 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3475, pruned_loss=0.09306, over 5650838.80 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.378, pruned_loss=0.1248, over 5651601.63 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:10:21,799 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-10 07:10:49,837 INFO [train.py:968] (1/2) Epoch 20, batch 8950, libri_loss[loss=0.2553, simple_loss=0.3381, pruned_loss=0.08622, over 29538.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3749, pruned_loss=0.1219, over 5660293.49 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3477, pruned_loss=0.09299, over 5656245.94 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3776, pruned_loss=0.1251, over 5654752.26 frames. ], batch size: 81, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:10:53,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6145, 2.0036, 1.9702, 1.7401], device='cuda:1'), covar=tensor([0.1905, 0.1823, 0.1865, 0.1869], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0740, 0.0706, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 07:11:22,412 INFO [optim.py:369] (1/2) 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 07:11:35,766 INFO [train.py:968] (1/2) Epoch 20, batch 9000, libri_loss[loss=0.2525, simple_loss=0.3435, pruned_loss=0.08069, over 29222.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3714, pruned_loss=0.1201, over 5653040.47 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3473, pruned_loss=0.09279, over 5666192.80 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3751, pruned_loss=0.1242, over 5639518.13 frames. ], batch size: 97, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:11:35,767 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 07:11:44,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5248, 1.6558, 1.2923, 1.2849], device='cuda:1'), covar=tensor([0.0873, 0.0442, 0.0968, 0.0984], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0447, 0.0514, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 07:11:45,121 INFO [train.py:1012] (1/2) Epoch 20, validation: loss=0.2064, simple_loss=0.3134, pruned_loss=0.04969, over 944034.00 frames. 
+2023-03-10 07:11:45,121 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 07:12:12,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-10 07:12:34,117 INFO [train.py:968] (1/2) Epoch 20, batch 9050, giga_loss[loss=0.2955, simple_loss=0.3514, pruned_loss=0.1198, over 28669.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3698, pruned_loss=0.1195, over 5659971.02 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3474, pruned_loss=0.09276, over 5669215.92 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3729, pruned_loss=0.1231, over 5646372.44 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:12:52,790 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9029, 3.0964, 1.8851, 1.1440], device='cuda:1'), covar=tensor([0.6576, 0.2420, 0.3863, 0.5852], device='cuda:1'), in_proj_covar=tensor([0.1724, 0.1635, 0.1582, 0.1405], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 07:13:09,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 07:13:09,535 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 9100, giga_loss[loss=0.2828, simple_loss=0.3681, pruned_loss=0.09875, over 28864.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5658132.72 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09302, over 5669013.28 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3727, pruned_loss=0.1236, over 5647038.39 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:14:12,688 INFO [train.py:968] (1/2) Epoch 20, batch 9150, giga_loss[loss=0.421, simple_loss=0.4407, pruned_loss=0.2007, over 26666.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3714, pruned_loss=0.1223, over 5658099.44 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09294, over 5672220.61 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3739, pruned_loss=0.1254, over 5646377.98 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:14:53,554 INFO [optim.py:369] (1/2) 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 07:15:05,287 INFO [train.py:968] (1/2) Epoch 20, batch 9200, giga_loss[loss=0.2729, simple_loss=0.3448, pruned_loss=0.1005, over 28990.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3706, pruned_loss=0.1221, over 5653205.21 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3477, pruned_loss=0.09289, over 5675328.05 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3731, pruned_loss=0.1251, over 5640878.14 frames. ], batch size: 164, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:15:51,539 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:15:53,214 INFO [train.py:968] (1/2) Epoch 20, batch 9250, giga_loss[loss=0.2688, simple_loss=0.3466, pruned_loss=0.09549, over 28897.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3688, pruned_loss=0.1216, over 5661367.69 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09299, over 5678444.41 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3712, pruned_loss=0.1247, over 5648413.64 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:16:32,165 INFO [optim.py:369] (1/2) 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 07:16:36,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:16:39,721 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 9300, giga_loss[loss=0.323, simple_loss=0.3897, pruned_loss=0.1282, over 28954.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3679, pruned_loss=0.1211, over 5655431.39 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09302, over 5681543.75 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3702, pruned_loss=0.1239, over 5641902.38 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:16:46,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7051, 1.7506, 1.9073, 1.4615], device='cuda:1'), covar=tensor([0.1943, 0.2628, 0.1516, 0.1776], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0702, 0.0934, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 07:17:22,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2156, 1.4130, 3.3604, 2.9993], device='cuda:1'), covar=tensor([0.1552, 0.2526, 0.0517, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0641, 0.0947, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 07:17:35,097 INFO [train.py:968] (1/2) Epoch 20, batch 9350, giga_loss[loss=0.3669, simple_loss=0.4148, pruned_loss=0.1595, over 27915.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.12, over 5664963.15 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3477, pruned_loss=0.09292, over 5685206.55 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.1229, over 5650697.89 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:18:00,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7371, 1.8074, 1.8371, 1.6585], device='cuda:1'), covar=tensor([0.2674, 0.2328, 0.1807, 0.2124], device='cuda:1'), in_proj_covar=tensor([0.1940, 0.1873, 0.1809, 0.1926], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 07:18:04,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4490, 3.0634, 1.6316, 1.5876], device='cuda:1'), covar=tensor([0.0818, 0.0365, 0.0696, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0555, 0.0381, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 07:18:11,867 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 9400, giga_loss[loss=0.2904, simple_loss=0.3597, pruned_loss=0.1105, over 28733.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1209, over 5658372.34 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3477, pruned_loss=0.09292, over 5675186.56 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3719, pruned_loss=0.1236, over 5654584.37 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:19:08,294 INFO [train.py:968] (1/2) Epoch 20, batch 9450, giga_loss[loss=0.3281, simple_loss=0.3865, pruned_loss=0.1348, over 28649.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3699, pruned_loss=0.1217, over 5656730.52 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3472, pruned_loss=0.09259, over 5680886.74 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3726, pruned_loss=0.1249, over 5648125.55 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:19:19,314 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:19:32,427 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:19:47,159 INFO [optim.py:369] (1/2) 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 07:19:51,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.62 vs. limit=5.0
+2023-03-10 07:19:58,609 INFO [train.py:968] (1/2) Epoch 20, batch 9500, giga_loss[loss=0.3058, simple_loss=0.3824, pruned_loss=0.1146, over 28930.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3724, pruned_loss=0.1208, over 5652642.80 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3477, pruned_loss=0.09299, over 5674024.50 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3744, pruned_loss=0.1233, over 5651407.60 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:20:43,049 INFO [train.py:968] (1/2) Epoch 20, batch 9550, giga_loss[loss=0.2842, simple_loss=0.3696, pruned_loss=0.09935, over 28630.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3739, pruned_loss=0.1197, over 5663926.69 frames. ], libri_tot_loss[loss=0.2668, simple_loss=0.3478, pruned_loss=0.0929, over 5679220.35 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.376, pruned_loss=0.1224, over 5657826.67 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 2.0
+2023-03-10 07:21:20,505 INFO [optim.py:369] (1/2) 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 07:21:28,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4449, 1.5676, 1.5757, 1.4323], device='cuda:1'), covar=tensor([0.2534, 0.2072, 0.1872, 0.2187], device='cuda:1'), in_proj_covar=tensor([0.1943, 0.1877, 0.1809, 0.1931], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 07:21:32,530 INFO [train.py:968] (1/2) Epoch 20, batch 9600, giga_loss[loss=0.2931, simple_loss=0.3708, pruned_loss=0.1076, over 28940.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3763, pruned_loss=0.1201, over 5673262.02 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3476, pruned_loss=0.09273, over 5681460.78 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3785, pruned_loss=0.1228, over 5666212.58 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:21:54,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:22:21,083 INFO [train.py:968] (1/2) Epoch 20, batch 9650, giga_loss[loss=0.3254, simple_loss=0.386, pruned_loss=0.1323, over 28973.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3788, pruned_loss=0.1229, over 5676063.34 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3479, pruned_loss=0.09298, over 5685625.97 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3815, pruned_loss=0.1259, over 5666223.18 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:22:34,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:22:35,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:22:55,144 INFO [optim.py:369] (1/2) 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 07:23:04,662 INFO [train.py:968] (1/2) Epoch 20, batch 9700, giga_loss[loss=0.3432, simple_loss=0.3975, pruned_loss=0.1444, over 28278.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3799, pruned_loss=0.125, over 5673099.48 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3472, pruned_loss=0.09273, over 5690110.19 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3839, pruned_loss=0.1288, over 5660750.62 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:23:33,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=876890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:23:51,964 INFO [train.py:968] (1/2) Epoch 20, batch 9750, giga_loss[loss=0.2792, simple_loss=0.3555, pruned_loss=0.1014, over 28705.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3787, pruned_loss=0.1247, over 5671007.14 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3466, pruned_loss=0.09239, over 5698514.37 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3837, pruned_loss=0.1293, over 5652474.25 frames. ], batch size: 60, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:24:05,051 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=876926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:06,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=876929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:26,428 INFO [optim.py:369] (1/2) 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 07:24:34,710 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=876958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:36,805 INFO [train.py:968] (1/2) Epoch 20, batch 9800, giga_loss[loss=0.2842, simple_loss=0.3594, pruned_loss=0.1045, over 28551.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3786, pruned_loss=0.1245, over 5676312.41 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3471, pruned_loss=0.09275, over 5701113.35 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3828, pruned_loss=0.1286, over 5658769.66 frames. ], batch size: 71, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:24:45,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=876970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:48,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=876972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:48,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=876973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:24:50,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=876975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:05,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=876996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:11,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:13,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:17,740 INFO [train.py:968] (1/2) Epoch 20, batch 9850, giga_loss[loss=0.2766, simple_loss=0.3655, pruned_loss=0.09386, over 29145.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3775, pruned_loss=0.1222, over 5683776.36 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3467, pruned_loss=0.09247, over 5708837.69 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3825, pruned_loss=0.1269, over 5661772.39 frames. ], batch size: 113, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:25:18,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:24,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:25:31,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3165, 1.5664, 1.2650, 1.5062], device='cuda:1'), covar=tensor([0.0727, 0.0379, 0.0344, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 07:25:53,137 INFO [optim.py:369] (1/2) 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 07:26:02,263 INFO [train.py:968] (1/2) Epoch 20, batch 9900, giga_loss[loss=0.2595, simple_loss=0.3411, pruned_loss=0.08896, over 28715.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3781, pruned_loss=0.1215, over 5681445.71 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3466, pruned_loss=0.09245, over 5709102.76 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3827, pruned_loss=0.1258, over 5663339.26 frames. ], batch size: 66, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:26:15,863 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:26:47,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-10 07:26:51,630 INFO [train.py:968] (1/2) Epoch 20, batch 9950, giga_loss[loss=0.299, simple_loss=0.3625, pruned_loss=0.1178, over 28733.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3795, pruned_loss=0.1228, over 5672656.81 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3468, pruned_loss=0.09261, over 5704947.37 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3838, pruned_loss=0.1269, over 5660286.63 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:26:56,648 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:05,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8386, 2.0291, 1.4918, 2.0815], device='cuda:1'), covar=tensor([0.2763, 0.2846, 0.3375, 0.2532], device='cuda:1'), in_proj_covar=tensor([0.1473, 0.1072, 0.1309, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 07:27:15,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7950, 1.8262, 1.3706, 1.3944], device='cuda:1'), covar=tensor([0.0953, 0.0693, 0.1059, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0449, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 07:27:22,462 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:26,672 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:42,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:27:46,230 INFO [train.py:968] (1/2) Epoch 20, batch 10000, giga_loss[loss=0.2764, simple_loss=0.3449, pruned_loss=0.1039, over 28682.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3794, pruned_loss=0.1236, over 5669585.06 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.3469, pruned_loss=0.09256, over 5707194.83 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3831, pruned_loss=0.1271, over 5657517.91 frames. ], batch size: 92, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:27:58,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:28:13,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:28:26,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 07:28:33,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5024, 0.9990, 4.5236, 3.4475], device='cuda:1'), covar=tensor([0.1673, 0.3025, 0.0457, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0640, 0.0948, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 07:28:34,934 INFO [train.py:968] (1/2) Epoch 20, batch 10050, libri_loss[loss=0.2307, simple_loss=0.3195, pruned_loss=0.07092, over 29567.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3774, pruned_loss=0.1233, over 5665310.58 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3469, pruned_loss=0.09248, over 5712681.61 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3813, pruned_loss=0.1271, over 5649212.22 frames. ], batch size: 77, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:28:42,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-10 07:29:16,660 INFO [optim.py:369] (1/2) 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 07:29:25,583 INFO [train.py:968] (1/2) Epoch 20, batch 10100, libri_loss[loss=0.2752, simple_loss=0.3603, pruned_loss=0.09505, over 29533.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3756, pruned_loss=0.1227, over 5671848.81 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3468, pruned_loss=0.09233, over 5715965.19 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3795, pruned_loss=0.1266, over 5654702.81 frames. ], batch size: 84, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:29:28,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:29:39,225 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4442, 2.1075, 1.5333, 0.6655], device='cuda:1'), covar=tensor([0.4322, 0.2181, 0.3390, 0.5307], device='cuda:1'), in_proj_covar=tensor([0.1718, 0.1625, 0.1577, 0.1401], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 07:29:46,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5203, 4.3529, 4.1492, 2.1695], device='cuda:1'), covar=tensor([0.0637, 0.0743, 0.0789, 0.1895], device='cuda:1'), in_proj_covar=tensor([0.1220, 0.1137, 0.0966, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 07:29:50,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 07:30:16,425 INFO [train.py:968] (1/2) Epoch 20, batch 10150, libri_loss[loss=0.2194, simple_loss=0.2998, pruned_loss=0.06953, over 29685.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3725, pruned_loss=0.1215, over 5658129.17 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3466, pruned_loss=0.09227, over 5711304.01 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3765, pruned_loss=0.1255, over 5647260.90 frames. ], batch size: 73, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:30:57,775 INFO [optim.py:369] (1/2) 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 07:31:07,376 INFO [train.py:968] (1/2) Epoch 20, batch 10200, giga_loss[loss=0.3165, simple_loss=0.3782, pruned_loss=0.1274, over 28616.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3718, pruned_loss=0.1219, over 5662994.66 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3464, pruned_loss=0.09204, over 5716120.05 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3759, pruned_loss=0.126, over 5648803.68 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:31:30,265 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:31:37,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:31:51,834 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 10250, giga_loss[loss=0.2954, simple_loss=0.3677, pruned_loss=0.1115, over 28866.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 5669035.25 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3463, pruned_loss=0.09199, over 5720380.61 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3749, pruned_loss=0.1256, over 5652853.89 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:31:53,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:32:21,066 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:32:28,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 07:32:30,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:32:30,707 INFO [optim.py:369] (1/2) 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 07:32:40,522 INFO [train.py:968] (1/2) Epoch 20, batch 10300, giga_loss[loss=0.3612, simple_loss=0.4093, pruned_loss=0.1566, over 28909.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3678, pruned_loss=0.1182, over 5676770.91 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.346, pruned_loss=0.09183, over 5726855.08 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3719, pruned_loss=0.1225, over 5656152.24 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:33:03,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:33:22,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1050, 1.0786, 3.9754, 3.1360], device='cuda:1'), covar=tensor([0.1849, 0.2938, 0.0430, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0751, 0.0641, 0.0951, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 07:33:25,561 INFO [train.py:968] (1/2) Epoch 20, batch 10350, giga_loss[loss=0.2583, simple_loss=0.3405, pruned_loss=0.08801, over 28851.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3644, pruned_loss=0.1147, over 5658183.16 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.346, pruned_loss=0.09175, over 5719675.72 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 5645923.41 frames. ], batch size: 174, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:33:52,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:33:54,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:34:07,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4087, 1.6924, 1.3697, 1.3680], device='cuda:1'), covar=tensor([0.2505, 0.2580, 0.2947, 0.2206], device='cuda:1'), in_proj_covar=tensor([0.1473, 0.1071, 0.1308, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 07:34:07,916 INFO [optim.py:369] (1/2) 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 07:34:17,320 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 10400, giga_loss[loss=0.2966, simple_loss=0.3644, pruned_loss=0.1144, over 28697.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3641, pruned_loss=0.1138, over 5672803.37 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3463, pruned_loss=0.09201, over 5724116.34 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3673, pruned_loss=0.1175, over 5657783.52 frames. ], batch size: 242, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:34:21,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5003, 1.6052, 1.7503, 1.3063], device='cuda:1'), covar=tensor([0.1581, 0.2553, 0.1320, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0705, 0.0938, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 07:34:22,987 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:34:50,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:34:52,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:06,698 INFO [train.py:968] (1/2) Epoch 20, batch 10450, giga_loss[loss=0.2993, simple_loss=0.3599, pruned_loss=0.1194, over 28775.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3625, pruned_loss=0.1134, over 5672324.16 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3461, pruned_loss=0.09175, over 5724704.83 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3658, pruned_loss=0.1171, over 5658585.40 frames. ], batch size: 262, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:35:23,774 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:25,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4658, 1.7813, 1.3568, 1.6079], device='cuda:1'), covar=tensor([0.2977, 0.2949, 0.3397, 0.2556], device='cuda:1'), in_proj_covar=tensor([0.1477, 0.1073, 0.1310, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 07:35:28,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:31,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=877635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:35:46,951 INFO [optim.py:369] (1/2) 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 07:35:56,108 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 10500, giga_loss[loss=0.2597, simple_loss=0.3251, pruned_loss=0.09716, over 28703.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5669555.17 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3461, pruned_loss=0.09163, over 5725103.88 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3631, pruned_loss=0.1165, over 5657072.76 frames. ], batch size: 99, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:36:00,770 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:36:40,930 INFO [train.py:968] (1/2) Epoch 20, batch 10550, libri_loss[loss=0.2548, simple_loss=0.3387, pruned_loss=0.08541, over 29553.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3621, pruned_loss=0.1146, over 5666283.92 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3456, pruned_loss=0.09151, over 5721881.36 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1184, over 5656741.04 frames. ], batch size: 76, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:37:01,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3029, 1.5443, 1.4588, 1.6048], device='cuda:1'), covar=tensor([0.0799, 0.0347, 0.0321, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 07:37:14,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6752, 1.8109, 1.6750, 1.6041], device='cuda:1'), covar=tensor([0.1775, 0.2350, 0.2124, 0.2168], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0747, 0.0711, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 07:37:19,521 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:37:19,920 INFO [optim.py:369] (1/2) 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 07:37:27,570 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 07:37:27,765 INFO [train.py:968] (1/2) Epoch 20, batch 10600, libri_loss[loss=0.2962, simple_loss=0.3794, pruned_loss=0.1065, over 29322.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3643, pruned_loss=0.1152, over 5669503.17 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.09141, over 5727082.58 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3675, pruned_loss=0.119, over 5655409.48 frames. ], batch size: 97, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:37:28,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:38:13,970 INFO [train.py:968] (1/2) Epoch 20, batch 10650, giga_loss[loss=0.2786, simple_loss=0.3509, pruned_loss=0.1031, over 28757.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3658, pruned_loss=0.1163, over 5663958.46 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.346, pruned_loss=0.09168, over 5730038.12 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3687, pruned_loss=0.12, over 5647741.65 frames. ], batch size: 119, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:38:28,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4807, 2.3162, 2.2925, 2.1028], device='cuda:1'), covar=tensor([0.1897, 0.2590, 0.2174, 0.2332], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0747, 0.0711, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 07:38:36,760 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:38:48,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0012, 3.8358, 3.6320, 1.9412], device='cuda:1'), covar=tensor([0.0681, 0.0781, 0.0790, 0.2106], device='cuda:1'), in_proj_covar=tensor([0.1215, 0.1130, 0.0961, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 07:38:52,612 INFO [optim.py:369] (1/2) 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 07:39:02,936 INFO [train.py:968] (1/2) Epoch 20, batch 10700, giga_loss[loss=0.2655, simple_loss=0.3394, pruned_loss=0.09585, over 28963.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3652, pruned_loss=0.1158, over 5660067.65 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3465, pruned_loss=0.09176, over 5727796.56 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1191, over 5647736.55 frames. ], batch size: 145, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:39:13,130 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:39:43,664 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=877904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:39:47,311 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 10750, libri_loss[loss=0.2627, simple_loss=0.3546, pruned_loss=0.0854, over 29738.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3667, pruned_loss=0.1173, over 5662452.89 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3464, pruned_loss=0.09156, over 5732140.74 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3691, pruned_loss=0.1208, over 5646961.70 frames. ], batch size: 87, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:40:13,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=877935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:40:13,739 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=877936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:40:27,383 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 10800, giga_loss[loss=0.3499, simple_loss=0.3821, pruned_loss=0.1588, over 23498.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3685, pruned_loss=0.1184, over 5667059.65 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3466, pruned_loss=0.09184, over 5736794.72 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1225, over 5646247.18 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:40:40,881 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=877967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:41:14,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1935, 1.3034, 1.2460, 1.0078], device='cuda:1'), covar=tensor([0.0971, 0.0478, 0.0982, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0448, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 07:41:20,444 INFO [train.py:968] (1/2) Epoch 20, batch 10850, giga_loss[loss=0.3196, simple_loss=0.3746, pruned_loss=0.1323, over 28946.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3685, pruned_loss=0.1178, over 5678077.30 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3463, pruned_loss=0.0918, over 5743354.08 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3722, pruned_loss=0.1223, over 5652508.49 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:41:43,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:41:56,496 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:42:05,092 INFO [train.py:968] (1/2) Epoch 20, batch 10900, libri_loss[loss=0.277, simple_loss=0.3562, pruned_loss=0.09886, over 29556.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3714, pruned_loss=0.1203, over 5685014.85 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3464, pruned_loss=0.09189, over 5748392.42 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.375, pruned_loss=0.1247, over 5658076.12 frames. ], batch size: 80, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:42:23,681 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:42:26,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:42:56,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:42:57,065 INFO [train.py:968] (1/2) Epoch 20, batch 10950, giga_loss[loss=0.3216, simple_loss=0.3851, pruned_loss=0.129, over 28327.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3727, pruned_loss=0.1221, over 5686732.25 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.346, pruned_loss=0.09171, over 5750881.16 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3762, pruned_loss=0.1261, over 5662290.93 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:43:13,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:43:40,626 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 11000, giga_loss[loss=0.364, simple_loss=0.4041, pruned_loss=0.1619, over 26658.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3737, pruned_loss=0.1217, over 5679069.20 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3459, pruned_loss=0.09174, over 5752295.56 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3771, pruned_loss=0.1255, over 5657241.48 frames. ], batch size: 555, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:44:08,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:44:10,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:44:42,146 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 11050, giga_loss[loss=0.2813, simple_loss=0.3576, pruned_loss=0.1025, over 28997.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3731, pruned_loss=0.1208, over 5677426.97 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3461, pruned_loss=0.0918, over 5751945.10 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.376, pruned_loss=0.124, over 5659486.44 frames. ], batch size: 227, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:44:43,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:45:18,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:45:24,735 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 11100, giga_loss[loss=0.3447, simple_loss=0.3983, pruned_loss=0.1456, over 28640.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3722, pruned_loss=0.1211, over 5669188.56 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3454, pruned_loss=0.09151, over 5755405.70 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3762, pruned_loss=0.1251, over 5648495.98 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:45:41,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:45:45,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:46:19,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:46:29,672 INFO [train.py:968] (1/2) Epoch 20, batch 11150, giga_loss[loss=0.3379, simple_loss=0.3925, pruned_loss=0.1417, over 28927.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3709, pruned_loss=0.1205, over 5659811.09 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3456, pruned_loss=0.09156, over 5758985.34 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3744, pruned_loss=0.1244, over 5637911.56 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:46:58,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:09,060 INFO [optim.py:369] (1/2) 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 07:47:12,229 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:13,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2066, 1.2576, 3.6392, 3.0859], device='cuda:1'), covar=tensor([0.1727, 0.2722, 0.0518, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0753, 0.0644, 0.0951, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 07:47:15,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:17,336 INFO [train.py:968] (1/2) Epoch 20, batch 11200, giga_loss[loss=0.3434, simple_loss=0.3928, pruned_loss=0.1469, over 28570.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3691, pruned_loss=0.1196, over 5664161.52 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3452, pruned_loss=0.09133, over 5765102.51 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3733, pruned_loss=0.124, over 5637664.13 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 8.0
+2023-03-10 07:47:37,606 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=878383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:41,509 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:43,675 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:47:45,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:48:02,549 INFO [train.py:968] (1/2) Epoch 20, batch 11250, giga_loss[loss=0.3163, simple_loss=0.3667, pruned_loss=0.133, over 28918.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1198, over 5663200.80 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3451, pruned_loss=0.09125, over 5764129.98 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1244, over 5639538.24 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:48:12,555 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:48:15,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:48:40,264 INFO [optim.py:369] (1/2) 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 07:48:43,032 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 11300, giga_loss[loss=0.2987, simple_loss=0.3649, pruned_loss=0.1162, over 28509.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3677, pruned_loss=0.1192, over 5670250.37 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3454, pruned_loss=0.09156, over 5759474.57 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3716, pruned_loss=0.1235, over 5651987.29 frames. ], batch size: 307, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:49:00,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1590, 2.5891, 1.2228, 1.3740], device='cuda:1'), covar=tensor([0.1043, 0.0382, 0.0913, 0.1443], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0550, 0.0379, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 07:49:11,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:49:14,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:49:38,578 INFO [train.py:968] (1/2) Epoch 20, batch 11350, giga_loss[loss=0.3455, simple_loss=0.3795, pruned_loss=0.1558, over 23377.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3685, pruned_loss=0.1205, over 5659587.54 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3457, pruned_loss=0.0918, over 5762504.64 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3717, pruned_loss=0.1242, over 5640712.26 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:49:45,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:50:18,523 INFO [optim.py:369] (1/2) 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 07:50:25,502 INFO [train.py:968] (1/2) Epoch 20, batch 11400, giga_loss[loss=0.3471, simple_loss=0.3973, pruned_loss=0.1485, over 28613.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3684, pruned_loss=0.1205, over 5663619.68 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3457, pruned_loss=0.09169, over 5767153.43 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3717, pruned_loss=0.1244, over 5641319.37 frames. ], batch size: 336, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:50:34,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:50:37,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:50:45,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=878579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:51:03,104 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=878600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:51:14,408 INFO [train.py:968] (1/2) Epoch 20, batch 11450, giga_loss[loss=0.2835, simple_loss=0.3546, pruned_loss=0.1062, over 28988.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3703, pruned_loss=0.1221, over 5664745.08 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3453, pruned_loss=0.09162, over 5760876.10 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3738, pruned_loss=0.126, over 5651024.63 frames. ], batch size: 106, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:51:53,656 INFO [optim.py:369] (1/2) 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 07:52:02,931 INFO [train.py:968] (1/2) Epoch 20, batch 11500, giga_loss[loss=0.3041, simple_loss=0.3665, pruned_loss=0.1209, over 28960.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3716, pruned_loss=0.1236, over 5638108.14 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3456, pruned_loss=0.0918, over 5751418.12 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3748, pruned_loss=0.1274, over 5632627.42 frames. ], batch size: 213, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:52:12,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0219, 1.3547, 1.1011, 0.1987], device='cuda:1'), covar=tensor([0.3519, 0.2957, 0.4310, 0.5897], device='cuda:1'), in_proj_covar=tensor([0.1722, 0.1627, 0.1585, 0.1403], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 07:52:28,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8448, 2.8017, 1.7495, 0.9708], device='cuda:1'), covar=tensor([0.7663, 0.3357, 0.4017, 0.6671], device='cuda:1'), in_proj_covar=tensor([0.1721, 0.1626, 0.1585, 0.1403], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 07:52:51,702 INFO [train.py:968] (1/2) Epoch 20, batch 11550, giga_loss[loss=0.2927, simple_loss=0.3579, pruned_loss=0.1137, over 28873.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3714, pruned_loss=0.1235, over 5654137.29 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3456, pruned_loss=0.09171, over 5752970.23 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3747, pruned_loss=0.1275, over 5645557.98 frames. ], batch size: 186, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:53:32,854 INFO [optim.py:369] (1/2) 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 07:53:36,490 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 11600, giga_loss[loss=0.3614, simple_loss=0.3926, pruned_loss=0.1651, over 23585.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3714, pruned_loss=0.1233, over 5653813.90 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3459, pruned_loss=0.09188, over 5754081.82 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3742, pruned_loss=0.1268, over 5644353.40 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:54:24,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9384, 1.1602, 2.8263, 2.6934], device='cuda:1'), covar=tensor([0.1539, 0.2498, 0.0595, 0.1553], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0640, 0.0945, 0.0891], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 07:54:28,454 INFO [train.py:968] (1/2) Epoch 20, batch 11650, giga_loss[loss=0.3382, simple_loss=0.3913, pruned_loss=0.1426, over 27943.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3722, pruned_loss=0.1233, over 5656069.24 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.346, pruned_loss=0.09178, over 5755153.91 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.375, pruned_loss=0.127, over 5644941.40 frames. ], batch size: 412, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:54:47,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:55:08,161 INFO [optim.py:369] (1/2) 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 07:55:15,049 INFO [train.py:968] (1/2) Epoch 20, batch 11700, giga_loss[loss=0.325, simple_loss=0.3665, pruned_loss=0.1418, over 23595.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3712, pruned_loss=0.1216, over 5663474.94 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3461, pruned_loss=0.09188, over 5749708.14 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3741, pruned_loss=0.1254, over 5657019.89 frames. ], batch size: 705, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:55:56,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:55:58,442 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 11750, giga_loss[loss=0.3367, simple_loss=0.3977, pruned_loss=0.1379, over 28957.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3743, pruned_loss=0.1242, over 5657881.24 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3464, pruned_loss=0.092, over 5751125.67 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3769, pruned_loss=0.1278, over 5649648.12 frames. ], batch size: 155, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:56:29,070 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=878954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:56:48,848 INFO [optim.py:369] (1/2) 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 07:56:55,058 INFO [train.py:968] (1/2) Epoch 20, batch 11800, giga_loss[loss=0.3402, simple_loss=0.3936, pruned_loss=0.1435, over 28269.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3753, pruned_loss=0.1253, over 5660823.17 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3462, pruned_loss=0.09189, over 5755309.86 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3782, pruned_loss=0.129, over 5648222.22 frames. ], batch size: 368, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:57:08,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=878974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:57:10,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=878977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:57:38,043 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:57:39,280 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 11850, libri_loss[loss=0.2318, simple_loss=0.3162, pruned_loss=0.07374, over 29554.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1263, over 5648926.11 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.346, pruned_loss=0.09174, over 5747818.92 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3792, pruned_loss=0.1303, over 5643664.71 frames. ], batch size: 76, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:58:00,087 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-10 07:58:18,596 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 11900, giga_loss[loss=0.3123, simple_loss=0.3815, pruned_loss=0.1216, over 28839.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3753, pruned_loss=0.1246, over 5651434.73 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3457, pruned_loss=0.09174, over 5749465.29 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3794, pruned_loss=0.1291, over 5642166.48 frames. ], batch size: 112, lr: 1.61e-03, grad_scale: 4.0
+2023-03-10 07:59:04,483 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 07:59:07,222 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 11950, giga_loss[loss=0.3019, simple_loss=0.3665, pruned_loss=0.1186, over 28575.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3758, pruned_loss=0.1246, over 5650967.50 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3457, pruned_loss=0.09173, over 5749900.96 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3791, pruned_loss=0.1282, over 5642957.06 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 07:59:34,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:00:00,843 INFO [optim.py:369] (1/2) 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 08:00:05,053 INFO [train.py:968] (1/2) Epoch 20, batch 12000, libri_loss[loss=0.2541, simple_loss=0.3469, pruned_loss=0.0807, over 28787.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3736, pruned_loss=0.1231, over 5645861.41 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.345, pruned_loss=0.09133, over 5744149.70 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3776, pruned_loss=0.1272, over 5642340.77 frames. ], batch size: 106, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:00:05,054 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 08:00:14,019 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 08:00:59,291 INFO [train.py:968] (1/2) Epoch 20, batch 12050, giga_loss[loss=0.3018, simple_loss=0.3645, pruned_loss=0.1195, over 28699.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3717, pruned_loss=0.1214, over 5648892.44 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3451, pruned_loss=0.09135, over 5736830.82 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3753, pruned_loss=0.1251, over 5651069.05 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:01:01,355 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 12100, giga_loss[loss=0.2839, simple_loss=0.3592, pruned_loss=0.1044, over 28589.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3724, pruned_loss=0.1214, over 5653221.80 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3454, pruned_loss=0.09143, over 5742008.77 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1254, over 5647452.35 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:02:34,017 INFO [train.py:968] (1/2) Epoch 20, batch 12150, giga_loss[loss=0.3174, simple_loss=0.381, pruned_loss=0.1269, over 28634.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3726, pruned_loss=0.1219, over 5647508.46 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3455, pruned_loss=0.09155, over 5735247.50 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3763, pruned_loss=0.1261, over 5645328.85 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:02:36,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1771, 1.0729, 3.9223, 3.1902], device='cuda:1'), covar=tensor([0.1799, 0.2976, 0.0459, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0645, 0.0955, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 08:03:18,935 INFO [optim.py:369] (1/2) 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 08:03:24,782 INFO [train.py:968] (1/2) Epoch 20, batch 12200, giga_loss[loss=0.299, simple_loss=0.3652, pruned_loss=0.1165, over 28655.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1226, over 5656112.13 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3457, pruned_loss=0.09158, over 5727902.69 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3757, pruned_loss=0.1264, over 5659432.47 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:03:34,796 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:03:46,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 12250, giga_loss[loss=0.3545, simple_loss=0.3826, pruned_loss=0.1632, over 23583.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1236, over 5649342.51 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3457, pruned_loss=0.09151, over 5723294.14 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3765, pruned_loss=0.1275, over 5654277.67 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:04:23,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2369, 0.8255, 0.8955, 1.3767], device='cuda:1'), covar=tensor([0.0699, 0.0435, 0.0352, 0.0770], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 08:04:54,397 INFO [optim.py:369] (1/2) 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 08:04:59,128 INFO [train.py:968] (1/2) Epoch 20, batch 12300, giga_loss[loss=0.3805, simple_loss=0.4121, pruned_loss=0.1744, over 26736.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3733, pruned_loss=0.1235, over 5663142.73 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3452, pruned_loss=0.09127, over 5727794.56 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3772, pruned_loss=0.1279, over 5661071.81 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:05:49,860 INFO [train.py:968] (1/2) Epoch 20, batch 12350, giga_loss[loss=0.2745, simple_loss=0.3419, pruned_loss=0.1036, over 28588.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.374, pruned_loss=0.1243, over 5658820.57 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3453, pruned_loss=0.09128, over 5729196.22 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3773, pruned_loss=0.1281, over 5655433.46 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:06:03,509 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:06:06,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=879529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:06:22,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5387, 1.6297, 1.7086, 1.3044], device='cuda:1'), covar=tensor([0.1642, 0.2518, 0.1396, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0703, 0.0937, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 08:06:34,281 INFO [optim.py:369] (1/2) 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 08:06:37,675 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:06:39,315 INFO [train.py:968] (1/2) Epoch 20, batch 12400, giga_loss[loss=0.3176, simple_loss=0.3784, pruned_loss=0.1284, over 28643.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3733, pruned_loss=0.1226, over 5679653.67 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3458, pruned_loss=0.09142, over 5733574.14 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3761, pruned_loss=0.1262, over 5671588.37 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:07:07,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=879588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:07:31,685 INFO [train.py:968] (1/2) Epoch 20, batch 12450, giga_loss[loss=0.334, simple_loss=0.4028, pruned_loss=0.1326, over 28762.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3731, pruned_loss=0.1222, over 5668248.66 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3458, pruned_loss=0.09143, over 5734376.27 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3754, pruned_loss=0.1251, over 5661028.46 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:07:40,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0260, 3.3193, 2.1654, 1.0871], device='cuda:1'), covar=tensor([0.6936, 0.2886, 0.3690, 0.6854], device='cuda:1'), in_proj_covar=tensor([0.1719, 0.1630, 0.1584, 0.1404], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 08:07:47,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2251, 4.0498, 3.8416, 2.0053], device='cuda:1'), covar=tensor([0.0550, 0.0682, 0.0727, 0.2283], device='cuda:1'), in_proj_covar=tensor([0.1227, 0.1141, 0.0970, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 08:08:10,920 INFO [optim.py:369] (1/2) 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 08:08:14,361 INFO [train.py:968] (1/2) Epoch 20, batch 12500, giga_loss[loss=0.276, simple_loss=0.3535, pruned_loss=0.09923, over 29028.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3724, pruned_loss=0.1209, over 5678684.36 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3458, pruned_loss=0.0913, over 5737936.53 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.375, pruned_loss=0.1243, over 5667929.41 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:09:04,227 INFO [train.py:968] (1/2) Epoch 20, batch 12550, libri_loss[loss=0.2788, simple_loss=0.3659, pruned_loss=0.09588, over 27951.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3703, pruned_loss=0.1194, over 5671694.18 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3462, pruned_loss=0.09157, over 5739729.14 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3731, pruned_loss=0.1231, over 5659104.11 frames. ], batch size: 116, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:09:21,424 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:09:22,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-10 08:09:24,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=879734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:09:34,847 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=879745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:09:47,260 INFO [optim.py:369] (1/2) 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 08:09:51,496 INFO [train.py:968] (1/2) Epoch 20, batch 12600, libri_loss[loss=0.2507, simple_loss=0.3344, pruned_loss=0.08352, over 29766.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3699, pruned_loss=0.1197, over 5676128.11 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3461, pruned_loss=0.09148, over 5741721.30 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3727, pruned_loss=0.1232, over 5663141.22 frames. ], batch size: 87, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:09:54,231 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:10:04,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2540, 1.5049, 1.5514, 1.3619], device='cuda:1'), covar=tensor([0.1936, 0.1682, 0.2350, 0.1910], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0751, 0.0715, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 08:10:18,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-10 08:10:37,120 INFO [train.py:968] (1/2) Epoch 20, batch 12650, libri_loss[loss=0.2714, simple_loss=0.3571, pruned_loss=0.09284, over 29173.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3681, pruned_loss=0.1192, over 5677677.03 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3463, pruned_loss=0.09158, over 5744361.78 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5663423.06 frames. ], batch size: 97, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:11:18,267 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:11:25,656 INFO [train.py:968] (1/2) Epoch 20, batch 12700, giga_loss[loss=0.2902, simple_loss=0.3537, pruned_loss=0.1133, over 27819.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3653, pruned_loss=0.1181, over 5690013.94 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3463, pruned_loss=0.09158, over 5747296.64 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3678, pruned_loss=0.1214, over 5674789.27 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:11:43,940 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:11:49,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6341, 1.7150, 1.7886, 1.3615], device='cuda:1'), covar=tensor([0.1685, 0.2446, 0.1412, 0.1688], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0703, 0.0939, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 08:11:52,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=879888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:11:54,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=879891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:12:03,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1856, 0.8388, 0.9552, 1.3884], device='cuda:1'), covar=tensor([0.0759, 0.0366, 0.0343, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 08:12:15,313 INFO [train.py:968] (1/2) Epoch 20, batch 12750, giga_loss[loss=0.3065, simple_loss=0.3683, pruned_loss=0.1223, over 28285.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3644, pruned_loss=0.1179, over 5693602.33 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3463, pruned_loss=0.09155, over 5746368.09 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3667, pruned_loss=0.1209, over 5681397.91 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:12:25,623 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=879920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:12:46,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-10 08:12:46,595 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=879940.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 08:13:04,274 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 12800, giga_loss[loss=0.2907, simple_loss=0.3433, pruned_loss=0.1191, over 24051.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3642, pruned_loss=0.1181, over 5683359.26 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3459, pruned_loss=0.09139, over 5746549.70 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3666, pruned_loss=0.121, over 5672797.22 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:13:32,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2261, 1.2424, 3.6526, 3.1715], device='cuda:1'), covar=tensor([0.1725, 0.2838, 0.0467, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0645, 0.0953, 0.0898], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 08:13:58,199 INFO [train.py:968] (1/2) Epoch 20, batch 12850, libri_loss[loss=0.2872, simple_loss=0.3688, pruned_loss=0.1028, over 29506.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3637, pruned_loss=0.1158, over 5688111.78 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.346, pruned_loss=0.09143, over 5751195.88 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.366, pruned_loss=0.1188, over 5673648.63 frames. ], batch size: 84, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:14:07,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 08:14:17,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3561, 4.1810, 3.9972, 1.8950], device='cuda:1'), covar=tensor([0.0643, 0.0811, 0.0809, 0.2006], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.1137, 0.0965, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 08:14:44,312 INFO [optim.py:369] (1/2) 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 08:14:46,474 INFO [train.py:968] (1/2) Epoch 20, batch 12900, giga_loss[loss=0.2345, simple_loss=0.3198, pruned_loss=0.07462, over 28953.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3617, pruned_loss=0.1126, over 5677598.55 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3464, pruned_loss=0.09162, over 5751244.00 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1156, over 5663890.45 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:15:41,940 INFO [train.py:968] (1/2) Epoch 20, batch 12950, giga_loss[loss=0.3279, simple_loss=0.3735, pruned_loss=0.1412, over 26889.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3588, pruned_loss=0.1098, over 5672608.07 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3461, pruned_loss=0.09149, over 5752680.95 frames. ], giga_tot_loss[loss=0.2929, simple_loss=0.3609, pruned_loss=0.1124, over 5659714.38 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:16:29,344 INFO [optim.py:369] (1/2) 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 08:16:31,767 INFO [train.py:968] (1/2) Epoch 20, batch 13000, giga_loss[loss=0.2533, simple_loss=0.3358, pruned_loss=0.0854, over 28563.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3549, pruned_loss=0.1061, over 5675608.34 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3448, pruned_loss=0.09103, over 5757522.60 frames. ], giga_tot_loss[loss=0.2881, simple_loss=0.358, pruned_loss=0.109, over 5658845.48 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:17:26,948 INFO [train.py:968] (1/2) Epoch 20, batch 13050, giga_loss[loss=0.2428, simple_loss=0.3277, pruned_loss=0.07901, over 28916.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3528, pruned_loss=0.1034, over 5679871.99 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3446, pruned_loss=0.09089, over 5759665.12 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3557, pruned_loss=0.1061, over 5663101.83 frames. ], batch size: 106, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:17:46,253 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:17:53,956 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-10 08:18:11,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:18:17,270 INFO [optim.py:369] (1/2) 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 08:18:20,482 INFO [train.py:968] (1/2) Epoch 20, batch 13100, giga_loss[loss=0.2511, simple_loss=0.335, pruned_loss=0.0836, over 28063.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3515, pruned_loss=0.1007, over 5672287.87 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3442, pruned_loss=0.09081, over 5759615.14 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3541, pruned_loss=0.103, over 5658394.79 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:19:14,301 INFO [train.py:968] (1/2) Epoch 20, batch 13150, libri_loss[loss=0.2826, simple_loss=0.3628, pruned_loss=0.1012, over 25688.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3526, pruned_loss=0.1017, over 5667114.50 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.344, pruned_loss=0.09073, over 5759461.72 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3551, pruned_loss=0.1039, over 5654146.77 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:19:18,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880315.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 08:19:22,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.45 vs. limit=5.0
+2023-03-10 08:19:41,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5776, 3.4194, 3.1957, 2.1573], device='cuda:1'), covar=tensor([0.0670, 0.0889, 0.0907, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1126, 0.0954, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-10 08:19:47,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5166, 1.8537, 1.8639, 1.5230], device='cuda:1'), covar=tensor([0.2974, 0.1987, 0.1580, 0.2235], device='cuda:1'), in_proj_covar=tensor([0.1921, 0.1855, 0.1771, 0.1911], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 08:19:53,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6513, 4.4884, 4.2582, 2.2751], device='cuda:1'), covar=tensor([0.0537, 0.0721, 0.0813, 0.1801], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1127, 0.0955, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 08:20:02,432 INFO [optim.py:369] (1/2) 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 08:20:05,597 INFO [train.py:968] (1/2) Epoch 20, batch 13200, giga_loss[loss=0.2854, simple_loss=0.3465, pruned_loss=0.1122, over 26621.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.351, pruned_loss=0.1008, over 5663565.06 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3441, pruned_loss=0.09105, over 5761726.29 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.353, pruned_loss=0.1024, over 5649965.31 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:20:17,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=880372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:19,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=880375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:42,183 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=880396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:44,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=880399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:48,504 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=880404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:20:55,292 INFO [train.py:968] (1/2) Epoch 20, batch 13250, giga_loss[loss=0.2731, simple_loss=0.3496, pruned_loss=0.09825, over 28927.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.09769, over 5670965.87 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.344, pruned_loss=0.0911, over 5759607.26 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3489, pruned_loss=0.09912, over 5659133.12 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:21:14,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=880428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:21:44,116 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=880458.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 08:21:44,389 INFO [optim.py:369] (1/2) 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 08:21:45,821 INFO [train.py:968] (1/2) Epoch 20, batch 13300, giga_loss[loss=0.2679, simple_loss=0.3456, pruned_loss=0.09511, over 28544.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3468, pruned_loss=0.09778, over 5661283.14 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3441, pruned_loss=0.0914, over 5750883.23 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3482, pruned_loss=0.09875, over 5658850.40 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:21:46,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=880461.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 08:21:53,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4913, 1.6108, 1.7459, 1.3412], device='cuda:1'), covar=tensor([0.1727, 0.2459, 0.1443, 0.1792], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0694, 0.0929, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 08:22:12,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7749, 1.9110, 1.5762, 2.3186], device='cuda:1'), covar=tensor([0.2735, 0.2765, 0.3125, 0.2162], device='cuda:1'), in_proj_covar=tensor([0.1484, 0.1075, 0.1317, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 08:22:14,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=880490.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 08:22:34,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7639, 1.9402, 1.4108, 1.4713], device='cuda:1'), covar=tensor([0.0915, 0.0533, 0.0938, 0.1116], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0443, 0.0510, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 08:22:34,567 INFO [train.py:968] (1/2) Epoch 20, batch 13350, giga_loss[loss=0.2486, simple_loss=0.3338, pruned_loss=0.0817, over 28707.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3465, pruned_loss=0.09732, over 5670340.19 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3437, pruned_loss=0.09131, over 5755582.81 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3481, pruned_loss=0.09839, over 5661301.67 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:22:47,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5347, 2.0237, 1.8328, 1.4750], device='cuda:1'), covar=tensor([0.2635, 0.1782, 0.2060, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.1913, 0.1845, 0.1761, 0.1900], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 08:23:24,908 INFO [optim.py:369] (1/2) 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 08:23:27,297 INFO [train.py:968] (1/2) Epoch 20, batch 13400, giga_loss[loss=0.235, simple_loss=0.3183, pruned_loss=0.07586, over 28868.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3445, pruned_loss=0.09572, over 5662707.46 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3436, pruned_loss=0.09143, over 5749325.41 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3459, pruned_loss=0.09657, over 5659311.67 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:23:32,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=880564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:24:17,130 INFO [train.py:968] (1/2) Epoch 20, batch 13450, giga_loss[loss=0.2138, simple_loss=0.3059, pruned_loss=0.0609, over 29007.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.341, pruned_loss=0.09305, over 5669017.45 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3427, pruned_loss=0.09101, over 5751708.60 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.343, pruned_loss=0.09421, over 5661489.84 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:25:08,956 INFO [optim.py:369] (1/2) 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 08:25:10,636 INFO [train.py:968] (1/2) Epoch 20, batch 13500, giga_loss[loss=0.226, simple_loss=0.306, pruned_loss=0.07303, over 28003.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3372, pruned_loss=0.09108, over 5659309.98 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3426, pruned_loss=0.09107, over 5751263.92 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3389, pruned_loss=0.09197, over 5652232.54 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:25:59,440 INFO [train.py:968] (1/2) Epoch 20, batch 13550, giga_loss[loss=0.2469, simple_loss=0.3199, pruned_loss=0.08693, over 28646.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3364, pruned_loss=0.09139, over 5636664.94 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3422, pruned_loss=0.09111, over 5738401.54 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3378, pruned_loss=0.09209, over 5638186.16 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:26:49,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4370, 3.0306, 1.5757, 1.4626], device='cuda:1'), covar=tensor([0.0881, 0.0352, 0.0899, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0548, 0.0379, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 08:26:53,306 INFO [optim.py:369] (1/2) 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 08:26:55,420 INFO [train.py:968] (1/2) Epoch 20, batch 13600, giga_loss[loss=0.2818, simple_loss=0.352, pruned_loss=0.1057, over 28393.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3363, pruned_loss=0.09202, over 5635164.05 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3422, pruned_loss=0.0911, over 5734617.50 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3374, pruned_loss=0.0926, over 5638092.92 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:27:54,555 INFO [train.py:968] (1/2) Epoch 20, batch 13650, giga_loss[loss=0.2669, simple_loss=0.3515, pruned_loss=0.09118, over 27974.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3385, pruned_loss=0.09263, over 5633136.01 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3419, pruned_loss=0.09088, over 5736734.76 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3395, pruned_loss=0.09328, over 5632485.85 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:28:40,033 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-10 08:28:55,504 INFO [optim.py:369] (1/2) 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 08:28:56,513 INFO [train.py:968] (1/2) Epoch 20, batch 13700, giga_loss[loss=0.2545, simple_loss=0.3392, pruned_loss=0.08493, over 28903.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3412, pruned_loss=0.09342, over 5635837.73 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3421, pruned_loss=0.09127, over 5739621.97 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3417, pruned_loss=0.0936, over 5631419.79 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:29:25,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2102, 1.2500, 3.1799, 2.9007], device='cuda:1'), covar=tensor([0.1474, 0.2646, 0.0481, 0.1970], device='cuda:1'), in_proj_covar=tensor([0.0746, 0.0638, 0.0940, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 08:29:45,281 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1427, 2.4636, 1.2499, 1.2880], device='cuda:1'), covar=tensor([0.0992, 0.0361, 0.0934, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0549, 0.0381, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 08:29:58,766 INFO [train.py:968] (1/2) Epoch 20, batch 13750, giga_loss[loss=0.3152, simple_loss=0.3819, pruned_loss=0.1242, over 29033.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3415, pruned_loss=0.09407, over 5638700.02 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3415, pruned_loss=0.091, over 5741961.98 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3425, pruned_loss=0.09449, over 5631897.74 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:30:37,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=880939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:30:58,743 INFO [optim.py:369] (1/2) 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 08:30:59,568 INFO [train.py:968] (1/2) Epoch 20, batch 13800, giga_loss[loss=0.2459, simple_loss=0.3303, pruned_loss=0.08081, over 28390.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3389, pruned_loss=0.09229, over 5648319.89 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3411, pruned_loss=0.09081, over 5739342.63 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3401, pruned_loss=0.09289, over 5641973.40 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:31:58,101 INFO [train.py:968] (1/2) Epoch 20, batch 13850, giga_loss[loss=0.2217, simple_loss=0.2997, pruned_loss=0.0719, over 24243.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3383, pruned_loss=0.09118, over 5626130.79 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3405, pruned_loss=0.09061, over 5723048.69 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3397, pruned_loss=0.09188, over 5632411.22 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:32:17,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-10 08:32:55,403 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 13900, giga_loss[loss=0.2779, simple_loss=0.3476, pruned_loss=0.1041, over 28992.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3367, pruned_loss=0.08978, over 5630006.55 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3403, pruned_loss=0.09084, over 5719825.36 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3378, pruned_loss=0.09011, over 5634429.82 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:33:23,119 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:33:28,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:33:58,400 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:33:59,509 INFO [train.py:968] (1/2) Epoch 20, batch 13950, giga_loss[loss=0.2422, simple_loss=0.3234, pruned_loss=0.08051, over 29145.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3332, pruned_loss=0.08888, over 5641633.81 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.34, pruned_loss=0.09077, over 5721193.56 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3344, pruned_loss=0.08919, over 5642803.51 frames. ], batch size: 165, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:34:04,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=881114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:34:26,328 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:34:59,871 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 14000, giga_loss[loss=0.2167, simple_loss=0.3017, pruned_loss=0.06581, over 28887.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3328, pruned_loss=0.08881, over 5652323.47 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3396, pruned_loss=0.09065, over 5725792.30 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3339, pruned_loss=0.08911, over 5647513.14 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:36:01,067 INFO [train.py:968] (1/2) Epoch 20, batch 14050, giga_loss[loss=0.2837, simple_loss=0.3546, pruned_loss=0.1064, over 27545.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3329, pruned_loss=0.08879, over 5661957.22 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3394, pruned_loss=0.09084, over 5726747.27 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3337, pruned_loss=0.08882, over 5655659.61 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:36:41,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-10 08:37:04,810 INFO [optim.py:369] (1/2) 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 08:37:05,713 INFO [train.py:968] (1/2) Epoch 20, batch 14100, giga_loss[loss=0.2499, simple_loss=0.3424, pruned_loss=0.07868, over 28716.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3366, pruned_loss=0.09011, over 5662370.25 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3395, pruned_loss=0.09097, over 5720512.76 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3372, pruned_loss=0.09, over 5662903.53 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:37:10,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-10 08:37:16,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8022, 2.0016, 1.6673, 1.8333], device='cuda:1'), covar=tensor([0.2618, 0.2657, 0.3064, 0.2331], device='cuda:1'), in_proj_covar=tensor([0.1483, 0.1073, 0.1315, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 08:38:03,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1949, 1.4493, 1.5706, 1.3382], device='cuda:1'), covar=tensor([0.1463, 0.1069, 0.1496, 0.1207], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0731, 0.0696, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 08:38:13,294 INFO [train.py:968] (1/2) Epoch 20, batch 14150, giga_loss[loss=0.2132, simple_loss=0.2913, pruned_loss=0.06756, over 27581.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.335, pruned_loss=0.08865, over 5666691.91 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3391, pruned_loss=0.09088, over 5724105.32 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3357, pruned_loss=0.08862, over 5662985.61 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:38:55,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-10 08:39:14,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5822, 1.7905, 1.4865, 1.5741], device='cuda:1'), covar=tensor([0.2663, 0.2588, 0.3013, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1483, 0.1074, 0.1317, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 08:39:18,366 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 14200, giga_loss[loss=0.262, simple_loss=0.3395, pruned_loss=0.09226, over 28886.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3341, pruned_loss=0.08849, over 5679702.47 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3391, pruned_loss=0.09096, over 5728491.84 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3346, pruned_loss=0.08832, over 5671291.27 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:40:20,956 INFO [train.py:968] (1/2) Epoch 20, batch 14250, giga_loss[loss=0.2977, simple_loss=0.382, pruned_loss=0.1067, over 28814.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3369, pruned_loss=0.09019, over 5671806.03 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3391, pruned_loss=0.09095, over 5735289.31 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.09002, over 5657084.12 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:41:22,615 INFO [optim.py:369] (1/2) 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 08:41:22,629 INFO [train.py:968] (1/2) Epoch 20, batch 14300, giga_loss[loss=0.2702, simple_loss=0.3532, pruned_loss=0.09364, over 27649.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3392, pruned_loss=0.08961, over 5670097.71 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3382, pruned_loss=0.09066, over 5739913.90 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3402, pruned_loss=0.0897, over 5650921.67 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:41:32,764 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:41:52,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:42:21,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:42:27,036 INFO [train.py:968] (1/2) Epoch 20, batch 14350, giga_loss[loss=0.2634, simple_loss=0.3562, pruned_loss=0.08536, over 28884.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3398, pruned_loss=0.08793, over 5660068.38 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3379, pruned_loss=0.09052, over 5741113.31 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3409, pruned_loss=0.0881, over 5643422.85 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:43:02,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6138, 1.7915, 1.2826, 1.3776], device='cuda:1'), covar=tensor([0.1032, 0.0622, 0.1035, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0386, 0.0441, 0.0509, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 08:43:16,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4263, 4.5756, 1.6535, 1.7137], device='cuda:1'), covar=tensor([0.1045, 0.0208, 0.0932, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0547, 0.0380, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 08:43:30,093 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 14400, giga_loss[loss=0.2581, simple_loss=0.3448, pruned_loss=0.08572, over 28922.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3399, pruned_loss=0.08674, over 5664179.29 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3378, pruned_loss=0.09058, over 5743757.75 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3408, pruned_loss=0.08678, over 5647786.21 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:43:56,341 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:44:34,229 INFO [train.py:968] (1/2) Epoch 20, batch 14450, giga_loss[loss=0.2536, simple_loss=0.3325, pruned_loss=0.08733, over 28957.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08802, over 5669703.29 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3379, pruned_loss=0.09059, over 5747204.14 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.342, pruned_loss=0.08796, over 5651648.51 frames. ], batch size: 106, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:44:55,363 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:44:59,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:05,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2328, 3.0558, 1.3940, 1.4281], device='cuda:1'), covar=tensor([0.1040, 0.0432, 0.0956, 0.1402], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0548, 0.0380, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 08:45:16,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3645, 3.4208, 1.5654, 1.5996], device='cuda:1'), covar=tensor([0.0995, 0.0293, 0.0907, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0548, 0.0380, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 08:45:22,069 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:24,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:32,975 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=881659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:45:34,188 INFO [optim.py:369] (1/2) 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 08:45:34,201 INFO [train.py:968] (1/2) Epoch 20, batch 14500, giga_loss[loss=0.2481, simple_loss=0.3244, pruned_loss=0.08592, over 28442.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3398, pruned_loss=0.08835, over 5660805.59 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3372, pruned_loss=0.09031, over 5733471.05 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3411, pruned_loss=0.0885, over 5655848.51 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:46:03,747 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=881681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:46:11,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:46:47,579 INFO [train.py:968] (1/2) Epoch 20, batch 14550, giga_loss[loss=0.2958, simple_loss=0.3657, pruned_loss=0.1129, over 28934.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3409, pruned_loss=0.09001, over 5661768.09 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3369, pruned_loss=0.09014, over 5733461.79 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3422, pruned_loss=0.09027, over 5657342.46 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 08:48:12,392 INFO [train.py:968] (1/2) Epoch 20, batch 14600, giga_loss[loss=0.2155, simple_loss=0.3019, pruned_loss=0.0646, over 28773.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3383, pruned_loss=0.08838, over 5669621.69 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3371, pruned_loss=0.09029, over 5728216.23 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3394, pruned_loss=0.08844, over 5668454.41 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:48:14,311 INFO [optim.py:369] (1/2) 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 08:48:54,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5882, 1.6745, 1.8668, 1.4634], device='cuda:1'), covar=tensor([0.1823, 0.2185, 0.1501, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0692, 0.0933, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 08:49:26,967 INFO [train.py:968] (1/2) Epoch 20, batch 14650, giga_loss[loss=0.2542, simple_loss=0.3364, pruned_loss=0.08602, over 29074.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3336, pruned_loss=0.08581, over 5659754.60 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3368, pruned_loss=0.09021, over 5728100.81 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3346, pruned_loss=0.08584, over 5658025.21 frames. ], batch size: 187, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:49:50,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4337, 3.4780, 1.5942, 1.5648], device='cuda:1'), covar=tensor([0.0965, 0.0365, 0.0925, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0549, 0.0381, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 08:50:05,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:50:28,323 INFO [train.py:968] (1/2) Epoch 20, batch 14700, giga_loss[loss=0.2302, simple_loss=0.3164, pruned_loss=0.072, over 28752.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3316, pruned_loss=0.08514, over 5671662.47 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3361, pruned_loss=0.08988, over 5735227.59 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3329, pruned_loss=0.08528, over 5661245.04 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:50:31,252 INFO [optim.py:369] (1/2) 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 08:50:54,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-10 08:51:31,805 INFO [train.py:968] (1/2) Epoch 20, batch 14750, libri_loss[loss=0.2595, simple_loss=0.3383, pruned_loss=0.09032, over 29539.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3327, pruned_loss=0.08588, over 5683182.02 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.336, pruned_loss=0.0899, over 5738124.10 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3337, pruned_loss=0.08585, over 5670503.15 frames. ], batch size: 82, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 08:51:51,971 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=881927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:51:52,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-10 08:52:30,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=881956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:52:37,935 INFO [train.py:968] (1/2) Epoch 20, batch 14800, giga_loss[loss=0.2306, simple_loss=0.2971, pruned_loss=0.08208, over 24381.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.336, pruned_loss=0.08764, over 5679283.85 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3356, pruned_loss=0.08973, over 5740835.20 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3372, pruned_loss=0.08773, over 5666120.86 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:52:39,521 INFO [optim.py:369] (1/2) 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 08:53:12,472 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=881986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:53:14,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=881989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:53:38,617 INFO [train.py:968] (1/2) Epoch 20, batch 14850, giga_loss[loss=0.239, simple_loss=0.321, pruned_loss=0.0785, over 28913.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3342, pruned_loss=0.08774, over 5687864.42 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3352, pruned_loss=0.08961, over 5741999.34 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3354, pruned_loss=0.08786, over 5674346.47 frames. ], batch size: 199, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:53:48,073 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:54:42,416 INFO [train.py:968] (1/2) Epoch 20, batch 14900, giga_loss[loss=0.2488, simple_loss=0.3283, pruned_loss=0.08464, over 28982.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3365, pruned_loss=0.09013, over 5668270.14 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3355, pruned_loss=0.08995, over 5735041.81 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3371, pruned_loss=0.08989, over 5662770.12 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:54:45,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:54:45,568 INFO [optim.py:369] (1/2) 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 08:55:26,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:55:29,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:55:32,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4001, 1.9402, 1.4287, 1.4988], device='cuda:1'), covar=tensor([0.0766, 0.0325, 0.0343, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0118, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 08:55:40,029 INFO [train.py:968] (1/2) Epoch 20, batch 14950, giga_loss[loss=0.3279, simple_loss=0.3947, pruned_loss=0.1306, over 27509.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3365, pruned_loss=0.09004, over 5663694.59 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3354, pruned_loss=0.08982, over 5731914.79 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3372, pruned_loss=0.08998, over 5660143.04 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:56:08,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:56:47,386 INFO [train.py:968] (1/2) Epoch 20, batch 15000, giga_loss[loss=0.2621, simple_loss=0.3547, pruned_loss=0.08477, over 28902.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3393, pruned_loss=0.09072, over 5658970.48 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3356, pruned_loss=0.09002, over 5723417.57 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3398, pruned_loss=0.0905, over 5661662.69 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:56:47,387 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 08:56:56,184 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 08:56:57,902 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:57:20,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6437, 1.7359, 1.8785, 1.4382], device='cuda:1'), covar=tensor([0.1901, 0.2595, 0.1569, 0.1850], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0689, 0.0929, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 08:57:40,888 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:57:57,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-10 08:58:06,783 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:58:11,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:58:13,931 INFO [train.py:968] (1/2) Epoch 20, batch 15050, giga_loss[loss=0.219, simple_loss=0.3077, pruned_loss=0.06512, over 28784.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3396, pruned_loss=0.09036, over 5661403.45 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3355, pruned_loss=0.08993, over 5722863.20 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3401, pruned_loss=0.09028, over 5663058.39 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:58:52,717 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3775, 3.6054, 1.5386, 1.5575], device='cuda:1'), covar=tensor([0.0991, 0.0397, 0.0963, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0546, 0.0380, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 08:59:02,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 08:59:38,646 INFO [train.py:968] (1/2) Epoch 20, batch 15100, giga_loss[loss=0.2664, simple_loss=0.3364, pruned_loss=0.0982, over 28866.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3357, pruned_loss=0.08851, over 5674650.62 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3353, pruned_loss=0.0898, over 5722837.16 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3363, pruned_loss=0.08856, over 5675488.90 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 08:59:40,261 INFO [optim.py:369] (1/2) 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 08:59:46,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 09:00:29,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:00:44,100 INFO [train.py:968] (1/2) Epoch 20, batch 15150, giga_loss[loss=0.2524, simple_loss=0.328, pruned_loss=0.08837, over 28908.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3308, pruned_loss=0.08688, over 5682563.59 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.335, pruned_loss=0.08967, over 5729833.78 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3315, pruned_loss=0.08698, over 5675801.49 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:01:42,910 INFO [train.py:968] (1/2) Epoch 20, batch 15200, giga_loss[loss=0.2612, simple_loss=0.3387, pruned_loss=0.09181, over 28874.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3283, pruned_loss=0.08586, over 5676276.94 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3348, pruned_loss=0.08962, over 5722844.00 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.329, pruned_loss=0.08592, over 5675753.49 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:01:48,155 INFO [optim.py:369] (1/2) 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 09:02:21,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4942, 1.8345, 1.7647, 1.5102], device='cuda:1'), covar=tensor([0.2782, 0.2028, 0.1730, 0.2053], device='cuda:1'), in_proj_covar=tensor([0.1904, 0.1821, 0.1741, 0.1880], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 09:02:41,590 INFO [train.py:968] (1/2) Epoch 20, batch 15250, giga_loss[loss=0.2449, simple_loss=0.3101, pruned_loss=0.08987, over 24454.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.33, pruned_loss=0.08771, over 5675127.10 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3346, pruned_loss=0.08969, over 5728886.86 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3305, pruned_loss=0.08761, over 5667811.56 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:03:19,445 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:03:22,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:03:27,075 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3761, 1.6898, 1.6822, 1.2291], device='cuda:1'), covar=tensor([0.1900, 0.2764, 0.1580, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0690, 0.0932, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 09:03:37,371 INFO [train.py:968] (1/2) Epoch 20, batch 15300, giga_loss[loss=0.2474, simple_loss=0.327, pruned_loss=0.08387, over 28770.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3292, pruned_loss=0.0871, over 5672555.14 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3341, pruned_loss=0.08945, over 5730259.58 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3299, pruned_loss=0.08721, over 5663940.18 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:03:43,388 INFO [optim.py:369] (1/2) 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 09:04:01,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:04:45,884 INFO [train.py:968] (1/2) Epoch 20, batch 15350, giga_loss[loss=0.2172, simple_loss=0.3094, pruned_loss=0.06252, over 28624.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3269, pruned_loss=0.08488, over 5670540.58 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.334, pruned_loss=0.08939, over 5732384.33 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3276, pruned_loss=0.08498, over 5661317.37 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:05:02,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2652, 3.0588, 1.3526, 1.4983], device='cuda:1'), covar=tensor([0.1010, 0.0330, 0.0969, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0545, 0.0380, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 09:05:18,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:05:25,252 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:05:48,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3407, 3.1133, 1.5528, 1.4728], device='cuda:1'), covar=tensor([0.0958, 0.0387, 0.0909, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0544, 0.0379, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 09:05:50,735 INFO [train.py:968] (1/2) Epoch 20, batch 15400, giga_loss[loss=0.2371, simple_loss=0.3198, pruned_loss=0.07721, over 28713.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3263, pruned_loss=0.08492, over 5648653.20 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3344, pruned_loss=0.08972, over 5717777.70 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3263, pruned_loss=0.0846, over 5652284.07 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:05:54,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:05:55,132 INFO [optim.py:369] (1/2) 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 09:06:59,776 INFO [train.py:968] (1/2) Epoch 20, batch 15450, giga_loss[loss=0.2114, simple_loss=0.2975, pruned_loss=0.06267, over 28440.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3257, pruned_loss=0.08419, over 5665101.15 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3342, pruned_loss=0.08974, over 5719978.28 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3256, pruned_loss=0.08376, over 5664333.72 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:07:37,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882638.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 09:08:04,557 INFO [train.py:968] (1/2) Epoch 20, batch 15500, giga_loss[loss=0.2603, simple_loss=0.3372, pruned_loss=0.09167, over 28857.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3256, pruned_loss=0.08355, over 5681721.52 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3333, pruned_loss=0.0894, over 5723157.26 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3261, pruned_loss=0.08339, over 5677281.04 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:08:09,836 INFO [optim.py:369] (1/2) 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 09:08:32,398 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:08:32,458 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:08:34,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=882682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:08:49,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=882693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:09:05,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4238, 3.1170, 2.5466, 1.9436], device='cuda:1'), covar=tensor([0.2701, 0.1480, 0.1880, 0.2299], device='cuda:1'), in_proj_covar=tensor([0.1895, 0.1808, 0.1730, 0.1868], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 09:09:09,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=882706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:09:12,500 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 15550, giga_loss[loss=0.2354, simple_loss=0.3152, pruned_loss=0.07777, over 28910.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3263, pruned_loss=0.08435, over 5684225.82 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3328, pruned_loss=0.08907, over 5726589.44 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3272, pruned_loss=0.08443, over 5676801.52 frames. ], batch size: 199, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:09:14,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:09:21,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-10 09:09:48,569 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=882738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:10:15,846 INFO [train.py:968] (1/2) Epoch 20, batch 15600, giga_loss[loss=0.2213, simple_loss=0.3113, pruned_loss=0.0656, over 28852.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3252, pruned_loss=0.08395, over 5690333.64 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.332, pruned_loss=0.08863, over 5733034.67 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3264, pruned_loss=0.08427, over 5677536.34 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:10:21,010 INFO [optim.py:369] (1/2) 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 09:10:53,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8747, 1.1385, 1.0791, 0.8374], device='cuda:1'), covar=tensor([0.2307, 0.2341, 0.1431, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1894, 0.1804, 0.1730, 0.1867], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 09:11:16,589 INFO [train.py:968] (1/2) Epoch 20, batch 15650, giga_loss[loss=0.2174, simple_loss=0.305, pruned_loss=0.06491, over 27633.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.325, pruned_loss=0.08276, over 5672092.96 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3319, pruned_loss=0.08857, over 5726636.86 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3259, pruned_loss=0.08297, over 5667463.67 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:12:11,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6938, 2.4155, 1.7352, 0.7313], device='cuda:1'), covar=tensor([0.5908, 0.3017, 0.4119, 0.6511], device='cuda:1'), in_proj_covar=tensor([0.1704, 0.1613, 0.1573, 0.1391], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 09:12:12,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1137, 1.5041, 1.5111, 1.3057], device='cuda:1'), covar=tensor([0.1854, 0.1523, 0.2013, 0.1694], device='cuda:1'), in_proj_covar=tensor([0.0450, 0.0724, 0.0690, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 09:12:15,411 INFO [train.py:968] (1/2) Epoch 20, batch 15700, giga_loss[loss=0.267, simple_loss=0.3478, pruned_loss=0.09309, over 28959.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3287, pruned_loss=0.08355, over 5658135.29 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3323, pruned_loss=0.08867, over 5718854.98 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3289, pruned_loss=0.08346, over 5660257.75 frames. ], batch size: 199, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:12:20,114 INFO [optim.py:369] (1/2) 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 09:13:21,764 INFO [train.py:968] (1/2) Epoch 20, batch 15750, giga_loss[loss=0.2254, simple_loss=0.3139, pruned_loss=0.06844, over 28693.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3308, pruned_loss=0.0846, over 5657469.01 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.332, pruned_loss=0.08856, over 5720852.23 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3312, pruned_loss=0.08459, over 5656659.74 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:13:28,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=882917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:14:20,521 INFO [train.py:968] (1/2) Epoch 20, batch 15800, giga_loss[loss=0.2301, simple_loss=0.3204, pruned_loss=0.06995, over 28632.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3308, pruned_loss=0.08464, over 5656974.99 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3318, pruned_loss=0.08837, over 5723745.88 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3313, pruned_loss=0.08475, over 5652257.75 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:14:23,397 INFO [optim.py:369] (1/2) 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 09:14:29,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-10 09:14:33,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1602, 1.4703, 1.4714, 1.0712], device='cuda:1'), covar=tensor([0.1598, 0.2456, 0.1376, 0.1689], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0689, 0.0933, 0.0832], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 09:14:48,313 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-10 09:15:21,913 INFO [train.py:968] (1/2) Epoch 20, batch 15850, giga_loss[loss=0.2262, simple_loss=0.3167, pruned_loss=0.0679, over 28619.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3302, pruned_loss=0.08463, over 5655830.99 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3315, pruned_loss=0.08814, over 5727496.87 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3309, pruned_loss=0.08487, over 5647743.90 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:15:24,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883013.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:16:18,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:27,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:28,001 INFO [train.py:968] (1/2) Epoch 20, batch 15900, giga_loss[loss=0.2179, simple_loss=0.3061, pruned_loss=0.0648, over 28958.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3274, pruned_loss=0.08306, over 5658638.79 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3313, pruned_loss=0.08808, over 5731128.04 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3281, pruned_loss=0.08321, over 5647478.26 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:16:31,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:32,462 INFO [optim.py:369] (1/2) 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 09:16:36,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:16:52,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 09:17:06,075 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:17:25,249 INFO [train.py:968] (1/2) Epoch 20, batch 15950, libri_loss[loss=0.2026, simple_loss=0.292, pruned_loss=0.05656, over 29362.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3259, pruned_loss=0.08314, over 5671774.99 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3307, pruned_loss=0.0877, over 5734693.34 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3269, pruned_loss=0.08345, over 5657294.91 frames. ], batch size: 71, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:17:29,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3924, 1.9664, 1.4041, 0.6138], device='cuda:1'), covar=tensor([0.4985, 0.2558, 0.3915, 0.5750], device='cuda:1'), in_proj_covar=tensor([0.1710, 0.1620, 0.1580, 0.1397], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 09:17:44,129 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:18:24,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883156.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 09:18:28,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883159.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:18:29,887 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:18:30,330 INFO [train.py:968] (1/2) Epoch 20, batch 16000, giga_loss[loss=0.2485, simple_loss=0.3277, pruned_loss=0.08462, over 28825.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3265, pruned_loss=0.08305, over 5674991.85 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3305, pruned_loss=0.0876, over 5736576.98 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3274, pruned_loss=0.08335, over 5661399.98 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:18:35,151 INFO [optim.py:369] (1/2) 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 09:19:05,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883188.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:19:08,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 09:19:14,830 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:19,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:32,876 INFO [train.py:968] (1/2) Epoch 20, batch 16050, giga_loss[loss=0.2241, simple_loss=0.3131, pruned_loss=0.06754, over 28704.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3283, pruned_loss=0.08386, over 5675617.83 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3305, pruned_loss=0.08758, over 5737372.28 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.329, pruned_loss=0.08402, over 5662641.80 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:19:33,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:37,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:38,644 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9470, 2.8347, 1.7710, 1.1573], device='cuda:1'), covar=tensor([0.6983, 0.3104, 0.4203, 0.5717], device='cuda:1'), in_proj_covar=tensor([0.1710, 0.1618, 0.1579, 0.1397], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 09:19:58,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:19:58,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4875, 1.8596, 1.4864, 1.4719], device='cuda:1'), covar=tensor([0.2690, 0.2546, 0.2994, 0.2431], device='cuda:1'), in_proj_covar=tensor([0.1474, 0.1066, 0.1308, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 09:20:14,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:20:17,914 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 16100, giga_loss[loss=0.2555, simple_loss=0.3373, pruned_loss=0.08685, over 28476.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3285, pruned_loss=0.08493, over 5669826.38 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3293, pruned_loss=0.08706, over 5743797.59 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.33, pruned_loss=0.08539, over 5651201.27 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:20:41,657 INFO [optim.py:369] (1/2) 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 09:21:39,365 INFO [train.py:968] (1/2) Epoch 20, batch 16150, giga_loss[loss=0.2606, simple_loss=0.3452, pruned_loss=0.08801, over 28951.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3326, pruned_loss=0.08744, over 5668858.40 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3292, pruned_loss=0.08702, over 5745479.40 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3339, pruned_loss=0.08782, over 5652175.24 frames. ], batch size: 285, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:22:35,747 INFO [train.py:968] (1/2) Epoch 20, batch 16200, giga_loss[loss=0.245, simple_loss=0.3341, pruned_loss=0.07796, over 28642.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3347, pruned_loss=0.08801, over 5660775.02 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3291, pruned_loss=0.08691, over 5745650.14 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.336, pruned_loss=0.08844, over 5645387.66 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:22:42,835 INFO [optim.py:369] (1/2) 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 09:23:29,751 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 09:23:34,333 INFO [train.py:968] (1/2) Epoch 20, batch 16250, libri_loss[loss=0.2749, simple_loss=0.3478, pruned_loss=0.101, over 29264.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3349, pruned_loss=0.0878, over 5662631.49 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3285, pruned_loss=0.08666, over 5751658.59 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3367, pruned_loss=0.0884, over 5640835.40 frames. ], batch size: 94, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:24:00,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-10 09:24:17,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-10 09:24:42,583 INFO [train.py:968] (1/2) Epoch 20, batch 16300, giga_loss[loss=0.2471, simple_loss=0.3283, pruned_loss=0.08293, over 28402.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.334, pruned_loss=0.08749, over 5647903.52 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3284, pruned_loss=0.08664, over 5734660.65 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3357, pruned_loss=0.08803, over 5643092.89 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:24:48,064 INFO [optim.py:369] (1/2) 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 09:25:36,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:25:50,006 INFO [train.py:968] (1/2) Epoch 20, batch 16350, giga_loss[loss=0.2402, simple_loss=0.3319, pruned_loss=0.07429, over 28856.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3322, pruned_loss=0.087, over 5644562.59 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.08692, over 5725196.15 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3332, pruned_loss=0.08716, over 5649075.64 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:26:14,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:26:27,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:26:37,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5764, 1.6238, 1.8645, 1.4057], device='cuda:1'), covar=tensor([0.1794, 0.2512, 0.1429, 0.1815], device='cuda:1'), in_proj_covar=tensor([0.0883, 0.0686, 0.0928, 0.0827], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 09:27:02,304 INFO [train.py:968] (1/2) Epoch 20, batch 16400, libri_loss[loss=0.2459, simple_loss=0.3264, pruned_loss=0.08267, over 29526.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08687, over 5659921.39 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.329, pruned_loss=0.08702, over 5726101.16 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3329, pruned_loss=0.0869, over 5661912.27 frames. ], batch size: 84, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:27:06,608 INFO [optim.py:369] (1/2) 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 09:28:05,003 INFO [train.py:968] (1/2) Epoch 20, batch 16450, giga_loss[loss=0.2782, simple_loss=0.3291, pruned_loss=0.1137, over 24757.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3313, pruned_loss=0.08741, over 5652881.35 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3289, pruned_loss=0.08695, over 5728083.24 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.332, pruned_loss=0.0875, over 5651503.25 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:28:20,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:28:38,967 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:28:50,668 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:28:53,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:29:10,559 INFO [train.py:968] (1/2) Epoch 20, batch 16500, giga_loss[loss=0.2297, simple_loss=0.2972, pruned_loss=0.08115, over 24439.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3291, pruned_loss=0.0866, over 5652401.32 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3287, pruned_loss=0.08685, over 5731720.40 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3299, pruned_loss=0.08675, over 5646721.54 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:29:17,300 INFO [optim.py:369] (1/2) 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 09:29:29,784 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:29:32,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:29:35,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:30:13,953 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:30:14,334 INFO [train.py:968] (1/2) Epoch 20, batch 16550, giga_loss[loss=0.2582, simple_loss=0.3384, pruned_loss=0.089, over 28926.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.329, pruned_loss=0.08543, over 5663652.67 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3284, pruned_loss=0.08667, over 5733796.97 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3299, pruned_loss=0.08572, over 5656293.84 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:31:20,132 INFO [train.py:968] (1/2) Epoch 20, batch 16600, giga_loss[loss=0.2883, simple_loss=0.3806, pruned_loss=0.09804, over 28956.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3282, pruned_loss=0.08294, over 5674127.39 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3283, pruned_loss=0.08655, over 5735809.46 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3289, pruned_loss=0.08324, over 5666086.55 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:31:23,498 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=883764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:31:26,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=883767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:31:26,293 INFO [optim.py:369] (1/2) 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 09:32:00,754 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=883796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:32:19,424 INFO [train.py:968] (1/2) Epoch 20, batch 16650, giga_loss[loss=0.2491, simple_loss=0.3372, pruned_loss=0.08047, over 28134.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3309, pruned_loss=0.08263, over 5684358.79 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3279, pruned_loss=0.08635, over 5739407.11 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3319, pruned_loss=0.08298, over 5673751.12 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:32:57,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6083, 1.8183, 1.2897, 1.3606], device='cuda:1'), covar=tensor([0.0965, 0.0546, 0.0969, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0439, 0.0511, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 09:33:15,977 INFO [train.py:968] (1/2) Epoch 20, batch 16700, giga_loss[loss=0.2766, simple_loss=0.3519, pruned_loss=0.1007, over 28730.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3317, pruned_loss=0.08304, over 5671418.76 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3279, pruned_loss=0.08629, over 5734172.74 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3327, pruned_loss=0.08333, over 5666466.76 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:33:25,545 INFO [optim.py:369] (1/2) 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 09:34:13,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=883900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:34:27,057 INFO [train.py:968] (1/2) Epoch 20, batch 16750, giga_loss[loss=0.2826, simple_loss=0.3739, pruned_loss=0.09566, over 28801.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3322, pruned_loss=0.08374, over 5662560.07 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3278, pruned_loss=0.08625, over 5735874.13 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3331, pruned_loss=0.08396, over 5656518.66 frames. ], batch size: 263, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:34:43,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3533, 1.5370, 1.3370, 1.5129], device='cuda:1'), covar=tensor([0.0759, 0.0327, 0.0335, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 09:35:26,481 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=883955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:35:32,704 INFO [train.py:968] (1/2) Epoch 20, batch 16800, libri_loss[loss=0.287, simple_loss=0.3611, pruned_loss=0.1064, over 29262.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3324, pruned_loss=0.0841, over 5659046.47 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3281, pruned_loss=0.08632, over 5739416.36 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3329, pruned_loss=0.08412, over 5648680.75 frames. ], batch size: 94, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:35:41,856 INFO [optim.py:369] (1/2) 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 09:36:09,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3661, 1.6468, 1.3446, 1.2980], device='cuda:1'), covar=tensor([0.2585, 0.2472, 0.2817, 0.2198], device='cuda:1'), in_proj_covar=tensor([0.1478, 0.1070, 0.1313, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 09:36:36,364 INFO [train.py:968] (1/2) Epoch 20, batch 16850, giga_loss[loss=0.2577, simple_loss=0.3443, pruned_loss=0.0856, over 27648.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3325, pruned_loss=0.08362, over 5664423.06 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3281, pruned_loss=0.08625, over 5736831.62 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3331, pruned_loss=0.08356, over 5654587.45 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:36:40,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:37:26,547 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:37:30,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=884046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:37:49,824 INFO [train.py:968] (1/2) Epoch 20, batch 16900, giga_loss[loss=0.2762, simple_loss=0.3544, pruned_loss=0.09897, over 28058.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3323, pruned_loss=0.08295, over 5662393.11 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3277, pruned_loss=0.08606, over 5739689.91 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3332, pruned_loss=0.08302, over 5650565.05 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:38:00,980 INFO [optim.py:369] (1/2) 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 09:38:13,727 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=884075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:39:05,822 INFO [train.py:968] (1/2) Epoch 20, batch 16950, giga_loss[loss=0.2451, simple_loss=0.3386, pruned_loss=0.07584, over 28852.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3368, pruned_loss=0.08501, over 5668564.20 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3278, pruned_loss=0.08614, over 5740558.00 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3375, pruned_loss=0.08497, over 5657750.75 frames. ], batch size: 164, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:39:44,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:40:07,037 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:40:10,520 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 17000, giga_loss[loss=0.2664, simple_loss=0.347, pruned_loss=0.0929, over 28437.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3373, pruned_loss=0.08556, over 5672611.60 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.328, pruned_loss=0.0862, over 5739397.99 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3379, pruned_loss=0.08547, over 5663484.83 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:40:25,926 INFO [optim.py:369] (1/2) 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 09:40:51,252 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 17050, giga_loss[loss=0.288, simple_loss=0.3567, pruned_loss=0.1097, over 27953.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3371, pruned_loss=0.08663, over 5662700.11 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.328, pruned_loss=0.08616, over 5730889.59 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3377, pruned_loss=0.0866, over 5662159.46 frames. ], batch size: 476, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:42:20,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 09:42:39,158 INFO [train.py:968] (1/2) Epoch 20, batch 17100, giga_loss[loss=0.2687, simple_loss=0.341, pruned_loss=0.09821, over 26893.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3348, pruned_loss=0.08517, over 5672335.51 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3278, pruned_loss=0.08604, over 5733986.14 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3356, pruned_loss=0.08523, over 5668070.29 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:42:50,335 INFO [optim.py:369] (1/2) 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 09:43:51,514 INFO [train.py:968] (1/2) Epoch 20, batch 17150, giga_loss[loss=0.2452, simple_loss=0.3316, pruned_loss=0.07937, over 28764.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3325, pruned_loss=0.08329, over 5674416.00 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.08613, over 5738419.33 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3331, pruned_loss=0.08317, over 5665287.16 frames. ], batch size: 263, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:44:18,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:44:54,566 INFO [train.py:968] (1/2) Epoch 20, batch 17200, giga_loss[loss=0.2536, simple_loss=0.3367, pruned_loss=0.08523, over 28043.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3325, pruned_loss=0.08361, over 5669563.32 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3278, pruned_loss=0.08603, over 5733287.30 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3333, pruned_loss=0.08352, over 5665849.86 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:45:03,970 INFO [optim.py:369] (1/2) 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 09:45:55,445 INFO [train.py:968] (1/2) Epoch 20, batch 17250, giga_loss[loss=0.2737, simple_loss=0.3572, pruned_loss=0.09514, over 29013.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3343, pruned_loss=0.08437, over 5669128.67 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3274, pruned_loss=0.08588, over 5735501.39 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3353, pruned_loss=0.08443, over 5663353.71 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:46:53,097 INFO [train.py:968] (1/2) Epoch 20, batch 17300, giga_loss[loss=0.2666, simple_loss=0.3346, pruned_loss=0.09935, over 28650.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3345, pruned_loss=0.08512, over 5675572.52 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.327, pruned_loss=0.0857, over 5738730.84 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3358, pruned_loss=0.08529, over 5666891.54 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:47:04,619 INFO [optim.py:369] (1/2) 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 09:47:07,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:47:13,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=884476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:47:44,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=884505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:47:50,117 INFO [train.py:968] (1/2) Epoch 20, batch 17350, giga_loss[loss=0.2697, simple_loss=0.3414, pruned_loss=0.09898, over 28881.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3312, pruned_loss=0.08466, over 5663588.50 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3271, pruned_loss=0.08579, over 5732340.68 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3322, pruned_loss=0.08468, over 5661485.55 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:47:54,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:48:26,349 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:48:44,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2952, 1.4126, 1.2602, 1.2973], device='cuda:1'), covar=tensor([0.1902, 0.1660, 0.1620, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.1911, 0.1810, 0.1730, 0.1881], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 09:48:44,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6878, 1.9697, 1.9590, 1.5041], device='cuda:1'), covar=tensor([0.1945, 0.2523, 0.1501, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0884, 0.0683, 0.0929, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 09:48:50,655 INFO [train.py:968] (1/2) Epoch 20, batch 17400, giga_loss[loss=0.2526, simple_loss=0.3351, pruned_loss=0.08501, over 29010.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3305, pruned_loss=0.08507, over 5658650.73 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3271, pruned_loss=0.08578, over 5736598.17 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3314, pruned_loss=0.08507, over 5651888.15 frames. ], batch size: 285, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:48:58,553 INFO [optim.py:369] (1/2) 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 09:49:07,776 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:49:44,587 INFO [train.py:968] (1/2) Epoch 20, batch 17450, giga_loss[loss=0.3043, simple_loss=0.3854, pruned_loss=0.1116, over 28842.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.08732, over 5657372.12 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3264, pruned_loss=0.08537, over 5737724.27 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3359, pruned_loss=0.08776, over 5648358.55 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:50:33,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=884656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:50:36,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=884659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:50:38,487 INFO [train.py:968] (1/2) Epoch 20, batch 17500, giga_loss[loss=0.3452, simple_loss=0.3997, pruned_loss=0.1454, over 26729.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3433, pruned_loss=0.09194, over 5666923.63 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3264, pruned_loss=0.08533, over 5740265.85 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3447, pruned_loss=0.09237, over 5656701.35 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:50:43,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0731, 2.1333, 1.8068, 2.0671], device='cuda:1'), covar=tensor([0.2561, 0.2745, 0.3130, 0.2408], device='cuda:1'), in_proj_covar=tensor([0.1469, 0.1066, 0.1306, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 09:50:45,023 INFO [optim.py:369] (1/2) 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 09:51:01,640 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=884688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:51:20,183 INFO [train.py:968] (1/2) Epoch 20, batch 17550, giga_loss[loss=0.3029, simple_loss=0.3684, pruned_loss=0.1187, over 28900.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3489, pruned_loss=0.09507, over 5678678.31 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3269, pruned_loss=0.08536, over 5742851.41 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3501, pruned_loss=0.0956, over 5666723.40 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 09:51:37,780 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-10 09:52:07,673 INFO [train.py:968] (1/2) Epoch 20, batch 17600, giga_loss[loss=0.2556, simple_loss=0.3325, pruned_loss=0.08933, over 28866.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3472, pruned_loss=0.09579, over 5678519.98 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3272, pruned_loss=0.0857, over 5745149.18 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3482, pruned_loss=0.09604, over 5666046.52 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:52:18,140 INFO [optim.py:369] (1/2) 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 09:52:55,324 INFO [train.py:968] (1/2) Epoch 20, batch 17650, giga_loss[loss=0.2364, simple_loss=0.3146, pruned_loss=0.07914, over 28960.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3393, pruned_loss=0.09248, over 5682941.27 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3272, pruned_loss=0.0857, over 5745149.18 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3401, pruned_loss=0.09267, over 5673233.01 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:52:58,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-10 09:53:41,248 INFO [train.py:968] (1/2) Epoch 20, batch 17700, giga_loss[loss=0.2182, simple_loss=0.2976, pruned_loss=0.06944, over 28317.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3323, pruned_loss=0.08954, over 5692429.15 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3273, pruned_loss=0.08572, over 5748929.94 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.333, pruned_loss=0.08979, over 5679958.94 frames. ], batch size: 368, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:53:48,776 INFO [optim.py:369] (1/2) 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 09:54:26,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.08 vs. limit=2.0
+2023-03-10 09:54:26,706 INFO [train.py:968] (1/2) Epoch 20, batch 17750, libri_loss[loss=0.2458, simple_loss=0.3344, pruned_loss=0.07855, over 29648.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3259, pruned_loss=0.08718, over 5687448.02 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3275, pruned_loss=0.08576, over 5743941.01 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3263, pruned_loss=0.08739, over 5680226.33 frames. ], batch size: 88, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:54:30,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=884916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:54:41,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2996, 1.4051, 1.2761, 1.2808], device='cuda:1'), covar=tensor([0.2515, 0.2361, 0.1886, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1928, 0.1824, 0.1754, 0.1904], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 09:54:41,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5219, 1.5996, 1.7335, 1.3607], device='cuda:1'), covar=tensor([0.1539, 0.2174, 0.1281, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0689, 0.0936, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 09:54:58,034 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 17800, giga_loss[loss=0.2284, simple_loss=0.2953, pruned_loss=0.08072, over 27583.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3199, pruned_loss=0.08445, over 5692168.81 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3279, pruned_loss=0.08592, over 5748224.48 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3195, pruned_loss=0.08447, over 5680291.09 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:55:14,512 INFO [optim.py:369] (1/2) 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 09:55:18,801 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=884976.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 09:55:49,422 INFO [train.py:968] (1/2) Epoch 20, batch 17850, libri_loss[loss=0.322, simple_loss=0.3935, pruned_loss=0.1253, over 29687.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.3148, pruned_loss=0.0821, over 5695726.02 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3282, pruned_loss=0.08602, over 5751504.59 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.314, pruned_loss=0.08195, over 5682116.39 frames. ], batch size: 88, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:56:27,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:56:28,097 INFO [train.py:968] (1/2) Epoch 20, batch 17900, libri_loss[loss=0.3363, simple_loss=0.3999, pruned_loss=0.1363, over 29543.00 frames. ], tot_loss[loss=0.237, simple_loss=0.3121, pruned_loss=0.08091, over 5705403.64 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3286, pruned_loss=0.08625, over 5753629.17 frames. ], giga_tot_loss[loss=0.2358, simple_loss=0.3107, pruned_loss=0.08041, over 5691408.97 frames. ], batch size: 89, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:56:30,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:56:39,035 INFO [optim.py:369] (1/2) 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 09:56:57,494 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:01,440 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:03,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:14,390 INFO [train.py:968] (1/2) Epoch 20, batch 17950, giga_loss[loss=0.2282, simple_loss=0.2984, pruned_loss=0.07898, over 28338.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3093, pruned_loss=0.07978, over 5699187.72 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3291, pruned_loss=0.08641, over 5757045.35 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.3074, pruned_loss=0.07912, over 5683804.39 frames. ], batch size: 65, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:57:22,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6088, 1.8599, 1.5479, 1.7639], device='cuda:1'), covar=tensor([0.2557, 0.2718, 0.2982, 0.2427], device='cuda:1'), in_proj_covar=tensor([0.1475, 0.1069, 0.1308, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 09:57:25,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 09:57:58,432 INFO [train.py:968] (1/2) Epoch 20, batch 18000, giga_loss[loss=0.2279, simple_loss=0.3063, pruned_loss=0.07475, over 29004.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3064, pruned_loss=0.07815, over 5702365.12 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3292, pruned_loss=0.08634, over 5759047.03 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.3044, pruned_loss=0.07754, over 5687373.43 frames. ], batch size: 128, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:57:58,433 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 09:58:03,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4279, 1.7467, 1.4084, 1.3951], device='cuda:1'), covar=tensor([0.3089, 0.2958, 0.3451, 0.2614], device='cuda:1'), in_proj_covar=tensor([0.1473, 0.1068, 0.1307, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 09:58:07,477 INFO [train.py:1012] (1/2) Epoch 20, validation: loss=0.2054, simple_loss=0.3118, pruned_loss=0.04946, over 944034.00 frames. 
+2023-03-10 09:58:07,477 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 09:58:12,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5093, 1.7033, 1.3094, 1.2568], device='cuda:1'), covar=tensor([0.1004, 0.0556, 0.1076, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0387, 0.0438, 0.0509, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 09:58:13,665 INFO [optim.py:369] (1/2) 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 09:58:50,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7625, 2.8106, 2.9617, 2.5922], device='cuda:1'), covar=tensor([0.1675, 0.2095, 0.1685, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0733, 0.0700, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 09:58:52,559 INFO [train.py:968] (1/2) Epoch 20, batch 18050, giga_loss[loss=0.1949, simple_loss=0.2731, pruned_loss=0.0584, over 29054.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3029, pruned_loss=0.07649, over 5705351.85 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3292, pruned_loss=0.08622, over 5762072.98 frames. ], giga_tot_loss[loss=0.2263, simple_loss=0.3007, pruned_loss=0.07591, over 5689692.67 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 09:59:12,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6789, 1.9810, 1.6386, 1.6995], device='cuda:1'), covar=tensor([0.2672, 0.2756, 0.3162, 0.2478], device='cuda:1'), in_proj_covar=tensor([0.1473, 0.1068, 0.1308, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 09:59:33,364 INFO [train.py:968] (1/2) Epoch 20, batch 18100, giga_loss[loss=0.2298, simple_loss=0.2965, pruned_loss=0.08151, over 28516.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3024, pruned_loss=0.07638, over 5687035.36 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3299, pruned_loss=0.08657, over 5748664.99 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.299, pruned_loss=0.07523, over 5683839.19 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 09:59:43,257 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 18150, giga_loss[loss=0.2006, simple_loss=0.2609, pruned_loss=0.07014, over 23987.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.299, pruned_loss=0.07502, over 5687660.01 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3299, pruned_loss=0.0866, over 5750355.28 frames. ], giga_tot_loss[loss=0.222, simple_loss=0.296, pruned_loss=0.07399, over 5683035.09 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:00:22,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4497, 2.0293, 1.8072, 1.6701], device='cuda:1'), covar=tensor([0.0771, 0.0275, 0.0299, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0105], device='cuda:1')
+2023-03-10 10:00:38,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4012, 1.5912, 1.3665, 0.9812], device='cuda:1'), covar=tensor([0.2574, 0.2651, 0.3008, 0.2448], device='cuda:1'), in_proj_covar=tensor([0.1480, 0.1073, 0.1313, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 10:00:46,863 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:01:01,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=885351.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:01:08,791 INFO [train.py:968] (1/2) Epoch 20, batch 18200, giga_loss[loss=0.2406, simple_loss=0.2978, pruned_loss=0.09166, over 26533.00 frames. ], tot_loss[loss=0.2225, simple_loss=0.2968, pruned_loss=0.0741, over 5699230.19 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3302, pruned_loss=0.08668, over 5752934.59 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2936, pruned_loss=0.07297, over 5692110.00 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:01:16,928 INFO [optim.py:369] (1/2) 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 10:01:48,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885407.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:01:51,700 INFO [train.py:968] (1/2) Epoch 20, batch 18250, giga_loss[loss=0.2782, simple_loss=0.3476, pruned_loss=0.1044, over 28534.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2959, pruned_loss=0.07415, over 5700939.18 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3302, pruned_loss=0.08659, over 5754502.74 frames. ], giga_tot_loss[loss=0.2195, simple_loss=0.2928, pruned_loss=0.0731, over 5692949.54 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:02:45,937 INFO [train.py:968] (1/2) Epoch 20, batch 18300, giga_loss[loss=0.2762, simple_loss=0.3548, pruned_loss=0.09874, over 29050.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3059, pruned_loss=0.07939, over 5698241.27 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3299, pruned_loss=0.08644, over 5755969.30 frames. ], giga_tot_loss[loss=0.2303, simple_loss=0.3034, pruned_loss=0.0786, over 5690318.43 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:02:54,539 INFO [optim.py:369] (1/2) 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 10:03:18,972 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885494.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:03:21,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:03:33,785 INFO [train.py:968] (1/2) Epoch 20, batch 18350, giga_loss[loss=0.2841, simple_loss=0.3651, pruned_loss=0.1016, over 28769.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3209, pruned_loss=0.08719, over 5694283.89 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3305, pruned_loss=0.08663, over 5755759.71 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.318, pruned_loss=0.08632, over 5687264.98 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:03:47,069 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885526.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:04:06,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-10 10:04:15,674 INFO [train.py:968] (1/2) Epoch 20, batch 18400, giga_loss[loss=0.3027, simple_loss=0.3726, pruned_loss=0.1164, over 28807.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3311, pruned_loss=0.09195, over 5706408.68 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3305, pruned_loss=0.08655, over 5758699.61 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3288, pruned_loss=0.09142, over 5697138.51 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:04:24,626 INFO [optim.py:369] (1/2) 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 10:04:29,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4364, 3.0613, 1.4648, 1.4252], device='cuda:1'), covar=tensor([0.0976, 0.0308, 0.0906, 0.1424], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0544, 0.0378, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 10:04:34,924 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8578, 1.9523, 1.4593, 1.5611], device='cuda:1'), covar=tensor([0.0938, 0.0662, 0.1080, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0440, 0.0512, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 10:04:55,048 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 10:04:58,394 INFO [train.py:968] (1/2) Epoch 20, batch 18450, giga_loss[loss=0.2954, simple_loss=0.3648, pruned_loss=0.113, over 28938.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3371, pruned_loss=0.09398, over 5700614.18 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3309, pruned_loss=0.08677, over 5761166.12 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.335, pruned_loss=0.09354, over 5688966.33 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:05:19,976 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 10:05:23,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:05:27,001 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-10 10:05:39,504 INFO [train.py:968] (1/2) Epoch 20, batch 18500, giga_loss[loss=0.2516, simple_loss=0.3444, pruned_loss=0.07944, over 29066.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.34, pruned_loss=0.09414, over 5702242.84 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3309, pruned_loss=0.08671, over 5764007.42 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3385, pruned_loss=0.09403, over 5688799.74 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:05:48,357 INFO [optim.py:369] (1/2) 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 10:05:53,492 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:05:54,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:06:07,575 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:06:27,440 INFO [train.py:968] (1/2) Epoch 20, batch 18550, libri_loss[loss=0.2261, simple_loss=0.3119, pruned_loss=0.07016, over 29558.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.341, pruned_loss=0.09392, over 5693266.60 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3312, pruned_loss=0.08672, over 5762019.90 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3397, pruned_loss=0.09407, over 5682104.03 frames. ], batch size: 77, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:06:29,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=885712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:06:53,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6412, 1.7574, 1.2680, 1.2732], device='cuda:1'), covar=tensor([0.0865, 0.0502, 0.0944, 0.1099], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0439, 0.0512, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 10:07:13,542 INFO [train.py:968] (1/2) Epoch 20, batch 18600, giga_loss[loss=0.2779, simple_loss=0.3536, pruned_loss=0.1011, over 28793.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.343, pruned_loss=0.09535, over 5695346.36 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3316, pruned_loss=0.08687, over 5764889.53 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3419, pruned_loss=0.09549, over 5682649.64 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:07:21,345 INFO [optim.py:369] (1/2) 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 10:07:30,429 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 18650, giga_loss[loss=0.2892, simple_loss=0.3593, pruned_loss=0.1095, over 28820.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3471, pruned_loss=0.09852, over 5703947.75 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3317, pruned_loss=0.08694, over 5768806.21 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3465, pruned_loss=0.09877, over 5689170.92 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:08:37,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:08:40,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:08:42,568 INFO [train.py:968] (1/2) Epoch 20, batch 18700, giga_loss[loss=0.2582, simple_loss=0.3462, pruned_loss=0.0851, over 28444.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3506, pruned_loss=0.1005, over 5707789.07 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3319, pruned_loss=0.08693, over 5771030.37 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3502, pruned_loss=0.1009, over 5693439.85 frames. ], batch size: 71, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:08:52,110 INFO [optim.py:369] (1/2) 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 10:09:05,854 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=885887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:09:25,776 INFO [train.py:968] (1/2) Epoch 20, batch 18750, giga_loss[loss=0.2799, simple_loss=0.3695, pruned_loss=0.09517, over 28423.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3532, pruned_loss=0.1005, over 5711673.25 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3328, pruned_loss=0.08732, over 5773171.47 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3524, pruned_loss=0.1007, over 5697251.49 frames. ], batch size: 71, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:09:37,230 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=885925.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:09:42,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=885928.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:10:07,094 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 18800, giga_loss[loss=0.2674, simple_loss=0.3474, pruned_loss=0.09377, over 28548.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.355, pruned_loss=0.1011, over 5703765.56 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3329, pruned_loss=0.08734, over 5767856.73 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3544, pruned_loss=0.1013, over 5696294.75 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:10:20,141 INFO [optim.py:369] (1/2) 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 10:10:29,451 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=885983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:10:33,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 10:10:53,576 INFO [train.py:968] (1/2) Epoch 20, batch 18850, giga_loss[loss=0.2551, simple_loss=0.3224, pruned_loss=0.09385, over 23554.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3558, pruned_loss=0.1009, over 5701541.91 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.333, pruned_loss=0.08742, over 5770416.48 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3557, pruned_loss=0.1014, over 5692258.02 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:10:56,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:24,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:25,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:37,156 INFO [train.py:968] (1/2) Epoch 20, batch 18900, giga_loss[loss=0.2668, simple_loss=0.3281, pruned_loss=0.1027, over 24029.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3552, pruned_loss=0.09962, over 5701504.22 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3333, pruned_loss=0.0875, over 5771474.34 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3551, pruned_loss=0.09999, over 5692664.71 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:11:38,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:11:45,362 INFO [optim.py:369] (1/2) 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 10:12:17,093 INFO [train.py:968] (1/2) Epoch 20, batch 18950, giga_loss[loss=0.2624, simple_loss=0.3436, pruned_loss=0.0906, over 28952.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3537, pruned_loss=0.09793, over 5702239.26 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3338, pruned_loss=0.08756, over 5765026.36 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3538, pruned_loss=0.09851, over 5699538.00 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:12:24,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4717, 1.6948, 1.4191, 1.6247], device='cuda:1'), covar=tensor([0.2571, 0.2662, 0.2932, 0.2191], device='cuda:1'), in_proj_covar=tensor([0.1470, 0.1069, 0.1304, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 10:12:41,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2263, 2.1331, 1.7328, 1.4355], device='cuda:1'), covar=tensor([0.0878, 0.0266, 0.0295, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0105], device='cuda:1')
+2023-03-10 10:12:54,788 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:12:56,491 INFO [train.py:968] (1/2) Epoch 20, batch 19000, giga_loss[loss=0.3198, simple_loss=0.3952, pruned_loss=0.1221, over 28627.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3531, pruned_loss=0.09707, over 5691537.86 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3344, pruned_loss=0.08769, over 5750003.33 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09772, over 5701420.17 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:12:56,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3543, 1.5717, 1.1575, 1.1203], device='cuda:1'), covar=tensor([0.1217, 0.0640, 0.1300, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0441, 0.0514, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 10:12:56,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:07,669 INFO [optim.py:369] (1/2) 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 10:13:22,695 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:23,947 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:24,518 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:25,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:26,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3109, 1.1704, 1.2500, 1.5905], device='cuda:1'), covar=tensor([0.0804, 0.0379, 0.0346, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0105], device='cuda:1')
+2023-03-10 10:13:26,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:35,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4673, 2.2173, 1.6399, 0.6961], device='cuda:1'), covar=tensor([0.5019, 0.2440, 0.3267, 0.4835], device='cuda:1'), in_proj_covar=tensor([0.1706, 0.1615, 0.1580, 0.1388], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 10:13:36,537 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:40,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:41,748 INFO [train.py:968] (1/2) Epoch 20, batch 19050, giga_loss[loss=0.3095, simple_loss=0.3721, pruned_loss=0.1235, over 28884.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.355, pruned_loss=0.1009, over 5675787.49 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3344, pruned_loss=0.08774, over 5745001.51 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3556, pruned_loss=0.1017, over 5686587.46 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:13:50,012 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:53,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:13:54,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:14:05,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:14:15,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3461, 1.2663, 1.3275, 1.5153], device='cuda:1'), covar=tensor([0.0683, 0.0450, 0.0323, 0.0721], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:14:25,931 INFO [train.py:968] (1/2) Epoch 20, batch 19100, giga_loss[loss=0.2716, simple_loss=0.3424, pruned_loss=0.1003, over 28892.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3569, pruned_loss=0.1044, over 5677126.49 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3344, pruned_loss=0.08768, over 5748565.05 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.358, pruned_loss=0.1055, over 5680264.41 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:14:36,821 INFO [optim.py:369] (1/2) 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 10:14:39,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5170, 2.0779, 1.8877, 1.9416], device='cuda:1'), covar=tensor([0.0791, 0.0280, 0.0290, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:15:00,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9629, 2.5206, 1.0259, 1.2695], device='cuda:1'), covar=tensor([0.1316, 0.0465, 0.1049, 0.1629], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0543, 0.0378, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 10:15:04,957 INFO [train.py:968] (1/2) Epoch 20, batch 19150, libri_loss[loss=0.2378, simple_loss=0.3163, pruned_loss=0.0797, over 28438.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3564, pruned_loss=0.1048, over 5688453.31 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3346, pruned_loss=0.0877, over 5749193.08 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3578, pruned_loss=0.1062, over 5688666.21 frames. ], batch size: 63, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:15:24,937 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:15:47,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:15:50,299 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:15:50,724 INFO [train.py:968] (1/2) Epoch 20, batch 19200, giga_loss[loss=0.2912, simple_loss=0.3526, pruned_loss=0.1149, over 28735.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3548, pruned_loss=0.1045, over 5694042.42 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3352, pruned_loss=0.08793, over 5751263.73 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3556, pruned_loss=0.1056, over 5691552.40 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:16:01,571 INFO [optim.py:369] (1/2) 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 10:16:01,930 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7295, 2.0403, 1.5973, 2.0650], device='cuda:1'), covar=tensor([0.2520, 0.2671, 0.2962, 0.2473], device='cuda:1'), in_proj_covar=tensor([0.1472, 0.1071, 0.1305, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 10:16:04,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5644, 1.8935, 1.8042, 1.6173], device='cuda:1'), covar=tensor([0.1986, 0.2024, 0.2275, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0737, 0.0704, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 10:16:31,333 INFO [train.py:968] (1/2) Epoch 20, batch 19250, libri_loss[loss=0.2358, simple_loss=0.3099, pruned_loss=0.08086, over 29660.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3527, pruned_loss=0.1033, over 5700913.90 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3354, pruned_loss=0.08806, over 5758213.92 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.354, pruned_loss=0.1048, over 5690195.91 frames. ], batch size: 69, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:16:40,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:16:48,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886429.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:16:49,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2121, 0.8246, 0.9816, 1.4583], device='cuda:1'), covar=tensor([0.0806, 0.0380, 0.0350, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:17:00,763 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:17:13,741 INFO [train.py:968] (1/2) Epoch 20, batch 19300, giga_loss[loss=0.2988, simple_loss=0.3639, pruned_loss=0.1169, over 27873.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3515, pruned_loss=0.102, over 5695731.11 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.335, pruned_loss=0.08773, over 5761202.23 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3532, pruned_loss=0.1037, over 5683344.24 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:17:24,899 INFO [optim.py:369] (1/2) 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 10:17:51,367 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:17:54,330 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 19350, giga_loss[loss=0.2456, simple_loss=0.3285, pruned_loss=0.08138, over 28557.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3482, pruned_loss=0.09929, over 5695188.09 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3352, pruned_loss=0.08782, over 5762689.48 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3495, pruned_loss=0.1007, over 5683618.76 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:18:17,354 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=886529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:18:20,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:18:46,550 INFO [train.py:968] (1/2) Epoch 20, batch 19400, giga_loss[loss=0.2383, simple_loss=0.3146, pruned_loss=0.081, over 28608.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3431, pruned_loss=0.09622, over 5693092.07 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3352, pruned_loss=0.08771, over 5765990.76 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3443, pruned_loss=0.09762, over 5679811.67 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:18:58,118 INFO [optim.py:369] (1/2) 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 10:19:21,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:19:34,972 INFO [train.py:968] (1/2) Epoch 20, batch 19450, libri_loss[loss=0.3102, simple_loss=0.3795, pruned_loss=0.1204, over 18955.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.339, pruned_loss=0.09416, over 5680296.07 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3359, pruned_loss=0.088, over 5755496.22 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3395, pruned_loss=0.0952, over 5678291.67 frames. ], batch size: 187, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:20:21,192 INFO [train.py:968] (1/2) Epoch 20, batch 19500, giga_loss[loss=0.2205, simple_loss=0.3032, pruned_loss=0.06896, over 28887.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3336, pruned_loss=0.09139, over 5678927.23 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3362, pruned_loss=0.088, over 5752137.51 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3337, pruned_loss=0.09237, over 5678170.07 frames. ], batch size: 199, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:20:34,065 INFO [optim.py:369] (1/2) 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 10:21:09,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:12,148 INFO [train.py:968] (1/2) Epoch 20, batch 19550, giga_loss[loss=0.2699, simple_loss=0.3475, pruned_loss=0.09612, over 28552.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3324, pruned_loss=0.09024, over 5682525.86 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3362, pruned_loss=0.08798, over 5753736.12 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3325, pruned_loss=0.09104, over 5680045.91 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:21:23,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7901, 1.2902, 2.8627, 2.6926], device='cuda:1'), covar=tensor([0.1764, 0.2475, 0.0576, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0735, 0.0631, 0.0927, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 10:21:32,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:34,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:36,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:21:54,236 INFO [train.py:968] (1/2) Epoch 20, batch 19600, giga_loss[loss=0.2528, simple_loss=0.325, pruned_loss=0.09034, over 28557.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.334, pruned_loss=0.09071, over 5690450.17 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3364, pruned_loss=0.08795, over 5746324.17 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3338, pruned_loss=0.09145, over 5693202.79 frames. ], batch size: 71, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:22:04,008 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:22:07,093 INFO [optim.py:369] (1/2) 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 10:22:24,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:22:31,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886804.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:22:36,611 INFO [train.py:968] (1/2) Epoch 20, batch 19650, giga_loss[loss=0.2406, simple_loss=0.3177, pruned_loss=0.08178, over 28693.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3334, pruned_loss=0.09026, over 5688143.17 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3367, pruned_loss=0.08793, over 5741238.04 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3329, pruned_loss=0.09094, over 5693419.55 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:22:43,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:13,041 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:15,476 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:19,519 INFO [train.py:968] (1/2) Epoch 20, batch 19700, giga_loss[loss=0.2449, simple_loss=0.3201, pruned_loss=0.08483, over 29014.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3317, pruned_loss=0.08956, over 5704730.58 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3373, pruned_loss=0.08803, over 5745096.77 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3307, pruned_loss=0.09006, over 5704441.89 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:23:29,319 INFO [optim.py:369] (1/2) 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 10:23:34,334 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:37,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:38,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:23:52,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6994, 1.6588, 1.9414, 1.4776], device='cuda:1'), covar=tensor([0.1876, 0.2520, 0.1446, 0.1696], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0692, 0.0936, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 10:23:54,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=886904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:02,438 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:02,920 INFO [train.py:968] (1/2) Epoch 20, batch 19750, libri_loss[loss=0.2545, simple_loss=0.3478, pruned_loss=0.08063, over 29570.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3293, pruned_loss=0.08851, over 5714355.95 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3375, pruned_loss=0.08806, over 5747719.64 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3281, pruned_loss=0.0889, over 5711137.55 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:24:24,713 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:25,546 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 10:24:26,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:29,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886947.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:24:33,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886950.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:24:40,560 INFO [train.py:968] (1/2) Epoch 20, batch 19800, giga_loss[loss=0.2468, simple_loss=0.3175, pruned_loss=0.0881, over 28921.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3284, pruned_loss=0.08859, over 5713020.46 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3387, pruned_loss=0.08861, over 5743740.68 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3262, pruned_loss=0.0884, over 5713174.36 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:24:40,865 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=886961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:43,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=886964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:51,296 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:24:52,377 INFO [optim.py:369] (1/2) 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 10:24:58,011 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886979.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:25:09,810 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=886993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:25:22,877 INFO [train.py:968] (1/2) Epoch 20, batch 19850, giga_loss[loss=0.2203, simple_loss=0.2985, pruned_loss=0.07108, over 28969.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.327, pruned_loss=0.0878, over 5714014.29 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3394, pruned_loss=0.0886, over 5741115.86 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3243, pruned_loss=0.08765, over 5715635.46 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:25:37,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 10:25:52,466 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=887047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:25:56,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=887050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:26:05,448 INFO [train.py:968] (1/2) Epoch 20, batch 19900, libri_loss[loss=0.2549, simple_loss=0.3365, pruned_loss=0.08665, over 29337.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3246, pruned_loss=0.08706, over 5713343.60 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3393, pruned_loss=0.08859, over 5741747.28 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3224, pruned_loss=0.08694, over 5713837.56 frames. ], batch size: 71, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:26:15,907 INFO [optim.py:369] (1/2) 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 10:26:18,652 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:26:22,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-10 10:26:44,926 INFO [train.py:968] (1/2) Epoch 20, batch 19950, giga_loss[loss=0.2824, simple_loss=0.3517, pruned_loss=0.1066, over 28008.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3236, pruned_loss=0.0866, over 5714107.24 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3397, pruned_loss=0.08862, over 5743337.86 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3211, pruned_loss=0.08641, over 5712409.61 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:26:48,448 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:26:50,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3801, 1.8042, 1.4577, 1.5713], device='cuda:1'), covar=tensor([0.0799, 0.0324, 0.0339, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:26:55,632 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2645, 2.6708, 1.8723, 2.1256], device='cuda:1'), covar=tensor([0.0966, 0.0604, 0.0943, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0440, 0.0515, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 10:27:16,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6471, 1.8813, 1.5369, 1.7755], device='cuda:1'), covar=tensor([0.2617, 0.2741, 0.3068, 0.2460], device='cuda:1'), in_proj_covar=tensor([0.1479, 0.1074, 0.1310, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 10:27:23,584 INFO [train.py:968] (1/2) Epoch 20, batch 20000, giga_loss[loss=0.2095, simple_loss=0.2883, pruned_loss=0.06531, over 28582.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3222, pruned_loss=0.08522, over 5723367.55 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3408, pruned_loss=0.08896, over 5748021.14 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3184, pruned_loss=0.08467, over 5716528.00 frames. ], batch size: 60, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:27:35,427 INFO [optim.py:369] (1/2) 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 10:27:44,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8211, 2.5522, 1.6220, 0.9580], device='cuda:1'), covar=tensor([0.7968, 0.4797, 0.4350, 0.6856], device='cuda:1'), in_proj_covar=tensor([0.1709, 0.1614, 0.1580, 0.1390], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 10:27:52,249 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:27:59,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5191, 1.3604, 4.5341, 3.4328], device='cuda:1'), covar=tensor([0.1720, 0.2796, 0.0386, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0737, 0.0632, 0.0928, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 10:28:03,472 INFO [train.py:968] (1/2) Epoch 20, batch 20050, giga_loss[loss=0.2744, simple_loss=0.3422, pruned_loss=0.1033, over 27935.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3197, pruned_loss=0.08393, over 5730747.01 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3408, pruned_loss=0.08885, over 5750473.84 frames. ], giga_tot_loss[loss=0.2418, simple_loss=0.3165, pruned_loss=0.08353, over 5722900.38 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:28:16,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2584, 0.7951, 0.8942, 1.3566], device='cuda:1'), covar=tensor([0.0811, 0.0403, 0.0369, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:28:40,616 INFO [train.py:968] (1/2) Epoch 20, batch 20100, giga_loss[loss=0.2572, simple_loss=0.3305, pruned_loss=0.0919, over 28962.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3196, pruned_loss=0.08374, over 5737710.53 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3415, pruned_loss=0.08901, over 5754662.28 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3157, pruned_loss=0.08313, over 5727103.31 frames. ], batch size: 106, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:28:51,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3902, 1.6297, 1.6730, 1.2392], device='cuda:1'), covar=tensor([0.1754, 0.2357, 0.1386, 0.1612], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0694, 0.0938, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 10:28:52,317 INFO [optim.py:369] (1/2) 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 10:29:19,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 10:29:26,213 INFO [train.py:968] (1/2) Epoch 20, batch 20150, giga_loss[loss=0.2949, simple_loss=0.36, pruned_loss=0.1149, over 28971.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.322, pruned_loss=0.08562, over 5734204.31 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3417, pruned_loss=0.08906, over 5755900.92 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3185, pruned_loss=0.08506, over 5724551.91 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:30:18,182 INFO [train.py:968] (1/2) Epoch 20, batch 20200, giga_loss[loss=0.2992, simple_loss=0.3656, pruned_loss=0.1164, over 27592.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3271, pruned_loss=0.08909, over 5715481.01 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3419, pruned_loss=0.08918, over 5746068.42 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3242, pruned_loss=0.08854, over 5715537.03 frames. ], batch size: 472, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:30:22,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-10 10:30:31,584 INFO [optim.py:369] (1/2) 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 10:31:09,341 INFO [train.py:968] (1/2) Epoch 20, batch 20250, libri_loss[loss=0.2939, simple_loss=0.3826, pruned_loss=0.1025, over 29383.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3379, pruned_loss=0.09657, over 5698254.94 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.342, pruned_loss=0.08907, over 5751039.01 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3351, pruned_loss=0.09632, over 5692732.94 frames. ], batch size: 92, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:31:55,146 INFO [train.py:968] (1/2) Epoch 20, batch 20300, libri_loss[loss=0.2903, simple_loss=0.3724, pruned_loss=0.1042, over 29233.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3438, pruned_loss=0.09942, over 5695878.94 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08941, over 5752623.39 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.341, pruned_loss=0.09908, over 5689219.49 frames. ], batch size: 101, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:32:08,323 INFO [optim.py:369] (1/2) 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 10:32:16,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1685, 1.2554, 1.1314, 1.1586], device='cuda:1'), covar=tensor([0.1518, 0.1682, 0.1223, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.1921, 0.1830, 0.1768, 0.1911], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 10:32:25,585 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=887491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:32:43,876 INFO [train.py:968] (1/2) Epoch 20, batch 20350, giga_loss[loss=0.3057, simple_loss=0.3812, pruned_loss=0.1151, over 28875.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3485, pruned_loss=0.1014, over 5687551.88 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.08945, over 5755701.97 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3463, pruned_loss=0.1014, over 5678362.43 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:33:28,790 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-10 10:33:30,910 INFO [train.py:968] (1/2) Epoch 20, batch 20400, giga_loss[loss=0.3254, simple_loss=0.3945, pruned_loss=0.1282, over 28940.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3536, pruned_loss=0.1043, over 5682485.56 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3427, pruned_loss=0.08944, over 5757090.06 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.352, pruned_loss=0.1044, over 5672700.49 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:33:44,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=887573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:33:45,463 INFO [optim.py:369] (1/2) 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 10:33:51,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3640, 1.6077, 1.6253, 1.2138], device='cuda:1'), covar=tensor([0.1512, 0.2292, 0.1286, 0.1604], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0695, 0.0935, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 10:34:16,380 INFO [train.py:968] (1/2) Epoch 20, batch 20450, giga_loss[loss=0.3275, simple_loss=0.3934, pruned_loss=0.1308, over 28556.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3589, pruned_loss=0.1077, over 5675386.99 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3427, pruned_loss=0.08947, over 5758144.11 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3578, pruned_loss=0.1081, over 5665567.21 frames. ], batch size: 336, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:34:26,009 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887623.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:34:39,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=887634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:34:42,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=887637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:35:03,568 INFO [train.py:968] (1/2) Epoch 20, batch 20500, giga_loss[loss=0.2287, simple_loss=0.3106, pruned_loss=0.0734, over 28744.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3528, pruned_loss=0.103, over 5683897.10 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3428, pruned_loss=0.08957, over 5759300.59 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.352, pruned_loss=0.1034, over 5674573.99 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:35:08,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:35:16,600 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 20550, giga_loss[loss=0.2655, simple_loss=0.3464, pruned_loss=0.09234, over 28830.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3511, pruned_loss=0.1013, over 5697029.52 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3429, pruned_loss=0.08977, over 5762110.08 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3506, pruned_loss=0.1017, over 5684973.15 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:35:50,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=887716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:35:52,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=887719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:36:10,028 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:36:13,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5570, 1.9034, 1.5090, 1.5020], device='cuda:1'), covar=tensor([0.2722, 0.2637, 0.3074, 0.2448], device='cuda:1'), in_proj_covar=tensor([0.1474, 0.1070, 0.1308, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 10:36:18,264 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=887748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:36:27,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-10 10:36:27,937 INFO [train.py:968] (1/2) Epoch 20, batch 20600, giga_loss[loss=0.2478, simple_loss=0.3287, pruned_loss=0.08342, over 28758.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3494, pruned_loss=0.09974, over 5689789.51 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.343, pruned_loss=0.08993, over 5752819.77 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.349, pruned_loss=0.1002, over 5685680.77 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:36:43,281 INFO [optim.py:369] (1/2) 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 10:36:44,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:37:10,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9925, 2.1994, 1.7653, 2.1313], device='cuda:1'), covar=tensor([0.2569, 0.2565, 0.3033, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1476, 0.1072, 0.1309, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 10:37:12,389 INFO [train.py:968] (1/2) Epoch 20, batch 20650, giga_loss[loss=0.2691, simple_loss=0.3505, pruned_loss=0.09384, over 28913.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3503, pruned_loss=0.09982, over 5692156.07 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3428, pruned_loss=0.08984, over 5754364.72 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3502, pruned_loss=0.1004, over 5686264.99 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:37:25,062 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-10 10:37:44,164 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:37:54,766 INFO [train.py:968] (1/2) Epoch 20, batch 20700, giga_loss[loss=0.3005, simple_loss=0.3699, pruned_loss=0.1155, over 28681.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3513, pruned_loss=0.1005, over 5697379.64 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3436, pruned_loss=0.09042, over 5757277.50 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.351, pruned_loss=0.1009, over 5687240.62 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:38:05,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:38:07,925 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4555, 1.6289, 1.7248, 1.2879], device='cuda:1'), covar=tensor([0.1690, 0.2415, 0.1324, 0.1593], device='cuda:1'), in_proj_covar=tensor([0.0887, 0.0692, 0.0932, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 10:38:09,949 INFO [optim.py:369] (1/2) 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 10:38:37,851 INFO [train.py:968] (1/2) Epoch 20, batch 20750, giga_loss[loss=0.275, simple_loss=0.3599, pruned_loss=0.09509, over 28405.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3531, pruned_loss=0.1015, over 5705998.02 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3438, pruned_loss=0.09041, over 5756280.70 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3528, pruned_loss=0.1021, over 5697933.76 frames. ], batch size: 65, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:38:53,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=887925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:39:24,366 INFO [train.py:968] (1/2) Epoch 20, batch 20800, giga_loss[loss=0.2563, simple_loss=0.3381, pruned_loss=0.08725, over 28947.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3539, pruned_loss=0.103, over 5691137.22 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.344, pruned_loss=0.09086, over 5760869.26 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3537, pruned_loss=0.1034, over 5679044.25 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:39:38,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5431, 4.3687, 4.1591, 1.8575], device='cuda:1'), covar=tensor([0.0553, 0.0675, 0.0693, 0.2129], device='cuda:1'), in_proj_covar=tensor([0.1175, 0.1092, 0.0926, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 10:39:39,492 INFO [optim.py:369] (1/2) 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 10:40:00,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=887998.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:40:10,071 INFO [train.py:968] (1/2) Epoch 20, batch 20850, giga_loss[loss=0.2593, simple_loss=0.3379, pruned_loss=0.09032, over 28470.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3547, pruned_loss=0.1039, over 5685630.29 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3441, pruned_loss=0.09091, over 5752554.40 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3547, pruned_loss=0.1044, over 5682248.41 frames. ], batch size: 71, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:40:49,350 INFO [train.py:968] (1/2) Epoch 20, batch 20900, giga_loss[loss=0.3097, simple_loss=0.373, pruned_loss=0.1232, over 28765.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3542, pruned_loss=0.1031, over 5698307.46 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3444, pruned_loss=0.0911, over 5756533.71 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3542, pruned_loss=0.1036, over 5690752.58 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:41:03,824 INFO [optim.py:369] (1/2) 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 10:41:14,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3196, 1.4701, 1.3635, 1.3099], device='cuda:1'), covar=tensor([0.2522, 0.2150, 0.2322, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.1928, 0.1837, 0.1779, 0.1915], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 10:41:30,210 INFO [train.py:968] (1/2) Epoch 20, batch 20950, giga_loss[loss=0.2765, simple_loss=0.3584, pruned_loss=0.09731, over 28881.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3539, pruned_loss=0.1023, over 5700453.38 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3447, pruned_loss=0.0913, over 5758970.30 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3539, pruned_loss=0.1028, over 5690334.86 frames. ], batch size: 227, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:41:33,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:41:55,979 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888141.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:41:58,908 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888144.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:42:05,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888152.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:42:12,251 INFO [train.py:968] (1/2) Epoch 20, batch 21000, giga_loss[loss=0.2785, simple_loss=0.3627, pruned_loss=0.09712, over 28932.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3535, pruned_loss=0.1006, over 5705555.52 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3442, pruned_loss=0.0911, over 5761687.65 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3541, pruned_loss=0.1014, over 5693858.95 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:42:12,251 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 10:42:20,610 INFO [train.py:1012] (1/2) Epoch 20, validation: loss=0.2084, simple_loss=0.3162, pruned_loss=0.05028, over 944034.00 frames. 
+2023-03-10 10:42:20,611 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 10:42:30,971 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888173.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:42:33,587 INFO [optim.py:369] (1/2) 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 10:43:01,230 INFO [train.py:968] (1/2) Epoch 20, batch 21050, giga_loss[loss=0.2591, simple_loss=0.345, pruned_loss=0.0866, over 28768.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.353, pruned_loss=0.1003, over 5705918.02 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3446, pruned_loss=0.09143, over 5764636.98 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3533, pruned_loss=0.1008, over 5692966.49 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:43:10,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:31,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:39,540 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:41,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:43:42,106 INFO [train.py:968] (1/2) Epoch 20, batch 21100, giga_loss[loss=0.2718, simple_loss=0.3512, pruned_loss=0.09617, over 29046.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3515, pruned_loss=0.09975, over 5713637.07 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3453, pruned_loss=0.0919, over 5763504.59 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3513, pruned_loss=0.09989, over 5703438.27 frames. ], batch size: 155, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:43:53,833 INFO [optim.py:369] (1/2) 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 10:44:06,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:44:10,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888295.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:44:12,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888298.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 10:44:13,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:44:20,938 INFO [train.py:968] (1/2) Epoch 20, batch 21150, giga_loss[loss=0.2697, simple_loss=0.3466, pruned_loss=0.09641, over 28615.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3494, pruned_loss=0.0987, over 5717426.29 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3455, pruned_loss=0.09219, over 5766452.31 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3491, pruned_loss=0.09879, over 5704985.56 frames. ], batch size: 242, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:44:32,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 10:44:35,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888327.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 10:45:00,405 INFO [train.py:968] (1/2) Epoch 20, batch 21200, giga_loss[loss=0.2412, simple_loss=0.3243, pruned_loss=0.07906, over 28316.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3477, pruned_loss=0.09821, over 5718974.19 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3456, pruned_loss=0.09232, over 5770165.95 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3474, pruned_loss=0.09834, over 5704288.74 frames. ], batch size: 77, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 10:45:04,494 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:08,678 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:45:26,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:28,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:45:31,215 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 21250, giga_loss[loss=0.2335, simple_loss=0.3209, pruned_loss=0.07306, over 28710.00 frames. ], tot_loss[loss=0.273, simple_loss=0.348, pruned_loss=0.09896, over 5717510.47 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3455, pruned_loss=0.09235, over 5772481.16 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.348, pruned_loss=0.09922, over 5702236.27 frames. ], batch size: 60, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:45:51,712 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:10,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=888443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:12,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=888446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:22,523 INFO [train.py:968] (1/2) Epoch 20, batch 21300, giga_loss[loss=0.2724, simple_loss=0.3499, pruned_loss=0.09749, over 28899.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3485, pruned_loss=0.09884, over 5728593.00 frames. ], libri_tot_loss[loss=0.2656, simple_loss=0.3458, pruned_loss=0.09269, over 5775946.30 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3483, pruned_loss=0.09894, over 5711951.80 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:46:34,226 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=888475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:46:37,092 INFO [optim.py:369] (1/2) 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 10:47:06,833 INFO [train.py:968] (1/2) Epoch 20, batch 21350, giga_loss[loss=0.268, simple_loss=0.3462, pruned_loss=0.09489, over 28835.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3474, pruned_loss=0.09767, over 5716063.08 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3456, pruned_loss=0.09271, over 5776702.57 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3474, pruned_loss=0.09775, over 5702241.79 frames. ], batch size: 119, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:47:08,046 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=888512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:47:30,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-10 10:47:47,526 INFO [train.py:968] (1/2) Epoch 20, batch 21400, giga_loss[loss=0.2822, simple_loss=0.3547, pruned_loss=0.1048, over 28601.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3464, pruned_loss=0.09676, over 5712238.83 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3456, pruned_loss=0.09299, over 5761742.57 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3465, pruned_loss=0.09675, over 5711263.12 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:48:01,447 INFO [optim.py:369] (1/2) 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 10:48:24,506 INFO [train.py:968] (1/2) Epoch 20, batch 21450, libri_loss[loss=0.3559, simple_loss=0.4104, pruned_loss=0.1507, over 29547.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3463, pruned_loss=0.09681, over 5714211.10 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3459, pruned_loss=0.09337, over 5756413.74 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3461, pruned_loss=0.09652, over 5716663.42 frames. ], batch size: 89, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:48:34,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6240, 5.3985, 5.1186, 2.6487], device='cuda:1'), covar=tensor([0.0449, 0.0645, 0.0705, 0.1798], device='cuda:1'), in_proj_covar=tensor([0.1177, 0.1095, 0.0930, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 10:49:04,569 INFO [train.py:968] (1/2) Epoch 20, batch 21500, giga_loss[loss=0.2793, simple_loss=0.3545, pruned_loss=0.1021, over 28783.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3445, pruned_loss=0.09631, over 5715391.96 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3465, pruned_loss=0.09403, over 5752164.71 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3437, pruned_loss=0.09557, over 5719271.46 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:49:20,259 INFO [optim.py:369] (1/2) 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 10:49:45,906 INFO [train.py:968] (1/2) Epoch 20, batch 21550, giga_loss[loss=0.2495, simple_loss=0.3189, pruned_loss=0.09001, over 28903.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3413, pruned_loss=0.09474, over 5715243.77 frames. ], libri_tot_loss[loss=0.2675, simple_loss=0.3466, pruned_loss=0.09423, over 5755396.17 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09398, over 5714716.71 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 1.0
+2023-03-10 10:50:25,983 INFO [train.py:968] (1/2) Epoch 20, batch 21600, giga_loss[loss=0.2559, simple_loss=0.3364, pruned_loss=0.08767, over 28554.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3413, pruned_loss=0.09512, over 5721780.71 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3469, pruned_loss=0.09459, over 5758260.25 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3403, pruned_loss=0.09423, over 5717800.21 frames. ], batch size: 65, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:50:31,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4968, 1.7078, 1.4361, 1.7163], device='cuda:1'), covar=tensor([0.0766, 0.0310, 0.0340, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:50:40,196 INFO [optim.py:369] (1/2) 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 10:50:48,933 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=888790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:51:04,619 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-10 10:51:05,438 INFO [train.py:968] (1/2) Epoch 20, batch 21650, giga_loss[loss=0.2555, simple_loss=0.3279, pruned_loss=0.09151, over 28796.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3416, pruned_loss=0.09588, over 5720299.93 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3473, pruned_loss=0.09499, over 5760113.48 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3402, pruned_loss=0.09482, over 5714378.61 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:51:18,704 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=888829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:51:46,457 INFO [train.py:968] (1/2) Epoch 20, batch 21700, giga_loss[loss=0.2474, simple_loss=0.3199, pruned_loss=0.0875, over 28865.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3401, pruned_loss=0.09559, over 5720718.27 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3474, pruned_loss=0.09509, over 5761184.61 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3388, pruned_loss=0.09467, over 5714839.12 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:52:02,863 INFO [optim.py:369] (1/2) 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 10:52:09,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=888887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:52:27,002 INFO [train.py:968] (1/2) Epoch 20, batch 21750, giga_loss[loss=0.2404, simple_loss=0.3175, pruned_loss=0.08168, over 28736.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3382, pruned_loss=0.09482, over 5722103.76 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3477, pruned_loss=0.09541, over 5763879.66 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3367, pruned_loss=0.09379, over 5713940.48 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:53:08,514 INFO [train.py:968] (1/2) Epoch 20, batch 21800, giga_loss[loss=0.2662, simple_loss=0.3262, pruned_loss=0.1031, over 28653.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3344, pruned_loss=0.09311, over 5708364.27 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3477, pruned_loss=0.09555, over 5754186.06 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3329, pruned_loss=0.09214, over 5710429.72 frames. ], batch size: 92, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:53:10,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-10 10:53:24,109 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 21850, giga_loss[loss=0.297, simple_loss=0.358, pruned_loss=0.118, over 23641.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3327, pruned_loss=0.09242, over 5708879.48 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3476, pruned_loss=0.09557, over 5758198.74 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3314, pruned_loss=0.09157, over 5706046.67 frames. ], batch size: 705, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:54:04,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:54:07,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:54:30,259 INFO [train.py:968] (1/2) Epoch 20, batch 21900, giga_loss[loss=0.2535, simple_loss=0.3434, pruned_loss=0.0818, over 28534.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3326, pruned_loss=0.0921, over 5709600.80 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3479, pruned_loss=0.09607, over 5757702.54 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3309, pruned_loss=0.09089, over 5706584.15 frames. ], batch size: 307, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:54:32,741 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:54:35,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-10 10:54:47,702 INFO [optim.py:369] (1/2) 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 10:55:05,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7927, 1.8726, 1.5848, 1.9351], device='cuda:1'), covar=tensor([0.2629, 0.2803, 0.3200, 0.2575], device='cuda:1'), in_proj_covar=tensor([0.1475, 0.1070, 0.1305, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 10:55:14,028 INFO [train.py:968] (1/2) Epoch 20, batch 21950, giga_loss[loss=0.3125, simple_loss=0.3736, pruned_loss=0.1257, over 26714.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3355, pruned_loss=0.09337, over 5708317.29 frames. ], libri_tot_loss[loss=0.2702, simple_loss=0.348, pruned_loss=0.09615, over 5760777.38 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3338, pruned_loss=0.09228, over 5702162.66 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 2.0
+2023-03-10 10:55:33,314 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=889131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:55:44,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=889143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:55:51,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3303, 1.2193, 1.1383, 1.4257], device='cuda:1'), covar=tensor([0.0721, 0.0341, 0.0328, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:55:59,420 INFO [train.py:968] (1/2) Epoch 20, batch 22000, giga_loss[loss=0.2661, simple_loss=0.3437, pruned_loss=0.09424, over 28831.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3382, pruned_loss=0.09397, over 5717863.65 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3488, pruned_loss=0.09679, over 5765451.98 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3358, pruned_loss=0.09243, over 5707242.63 frames. ], batch size: 199, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:56:02,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:56:06,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3273, 1.4118, 1.2993, 1.5062], device='cuda:1'), covar=tensor([0.0735, 0.0321, 0.0342, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0185, 0.0116, 0.0116, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0068, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 10:56:13,755 INFO [optim.py:369] (1/2) 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 10:56:34,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5141, 5.3341, 5.0927, 2.7708], device='cuda:1'), covar=tensor([0.0373, 0.0562, 0.0594, 0.1652], device='cuda:1'), in_proj_covar=tensor([0.1183, 0.1099, 0.0936, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 10:56:36,830 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 22050, giga_loss[loss=0.2949, simple_loss=0.3687, pruned_loss=0.1105, over 28731.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3411, pruned_loss=0.09511, over 5710266.86 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3493, pruned_loss=0.09733, over 5767963.76 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3386, pruned_loss=0.0934, over 5698912.69 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:57:27,682 INFO [train.py:968] (1/2) Epoch 20, batch 22100, giga_loss[loss=0.2742, simple_loss=0.356, pruned_loss=0.09619, over 28902.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09464, over 5702359.96 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3494, pruned_loss=0.09751, over 5767928.79 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3389, pruned_loss=0.09307, over 5692627.39 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:57:46,603 INFO [optim.py:369] (1/2) 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 10:58:11,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:13,723 INFO [train.py:968] (1/2) Epoch 20, batch 22150, giga_loss[loss=0.238, simple_loss=0.3184, pruned_loss=0.07881, over 28781.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3405, pruned_loss=0.09415, over 5704878.60 frames. ], libri_tot_loss[loss=0.2729, simple_loss=0.3498, pruned_loss=0.098, over 5769332.61 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3381, pruned_loss=0.09234, over 5694273.50 frames. ], batch size: 112, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:58:13,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:39,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:45,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 10:58:47,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 22200, giga_loss[loss=0.2674, simple_loss=0.3363, pruned_loss=0.09923, over 28488.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3426, pruned_loss=0.09614, over 5704136.85 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3503, pruned_loss=0.09839, over 5768129.66 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.34, pruned_loss=0.09427, over 5695772.33 frames. ], batch size: 78, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:59:09,777 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:59:16,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7634, 4.5989, 4.4036, 1.9070], device='cuda:1'), covar=tensor([0.0557, 0.0653, 0.0757, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.1177, 0.1095, 0.0932, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 10:59:26,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 10:59:35,547 INFO [train.py:968] (1/2) Epoch 20, batch 22250, giga_loss[loss=0.277, simple_loss=0.3557, pruned_loss=0.09911, over 28930.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3432, pruned_loss=0.09689, over 5709917.96 frames. ], libri_tot_loss[loss=0.2735, simple_loss=0.3502, pruned_loss=0.09838, over 5770640.82 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3411, pruned_loss=0.09534, over 5699615.11 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 10:59:58,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=889440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:00:15,911 INFO [train.py:968] (1/2) Epoch 20, batch 22300, giga_loss[loss=0.3108, simple_loss=0.3849, pruned_loss=0.1184, over 28802.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3465, pruned_loss=0.09897, over 5713770.39 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.351, pruned_loss=0.09923, over 5775686.31 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3438, pruned_loss=0.09691, over 5698552.03 frames. ], batch size: 284, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:00:30,639 INFO [optim.py:369] (1/2) 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 11:00:51,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:00:54,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-10 11:00:55,215 INFO [train.py:968] (1/2) Epoch 20, batch 22350, giga_loss[loss=0.2945, simple_loss=0.3723, pruned_loss=0.1083, over 28689.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.09988, over 5717920.26 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3508, pruned_loss=0.09928, over 5777101.84 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3466, pruned_loss=0.09816, over 5702321.02 frames. ], batch size: 262, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:01:00,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:01:35,498 INFO [train.py:968] (1/2) Epoch 20, batch 22400, libri_loss[loss=0.3008, simple_loss=0.3711, pruned_loss=0.1152, over 29222.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3502, pruned_loss=0.1005, over 5695090.41 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3519, pruned_loss=0.1002, over 5753495.65 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3474, pruned_loss=0.09825, over 5702111.94 frames. ], batch size: 97, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 11:01:49,715 INFO [optim.py:369] (1/2) 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 11:02:19,515 INFO [train.py:968] (1/2) Epoch 20, batch 22450, giga_loss[loss=0.2766, simple_loss=0.3514, pruned_loss=0.1009, over 28858.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3524, pruned_loss=0.1017, over 5690316.22 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3527, pruned_loss=0.1009, over 5738105.65 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3494, pruned_loss=0.0993, over 5707187.30 frames. ], batch size: 186, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:02:52,570 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-10 11:02:54,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:02:56,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:02:58,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9767, 3.7850, 3.6484, 1.8186], device='cuda:1'), covar=tensor([0.0815, 0.0962, 0.0833, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.1181, 0.1098, 0.0936, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 11:03:05,103 INFO [train.py:968] (1/2) Epoch 20, batch 22500, giga_loss[loss=0.3495, simple_loss=0.4012, pruned_loss=0.1489, over 26684.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3527, pruned_loss=0.1018, over 5693849.74 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.353, pruned_loss=0.1011, over 5738209.60 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3501, pruned_loss=0.09973, over 5706473.24 frames. ], batch size: 555, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:03:05,395 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:08,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:22,874 INFO [optim.py:369] (1/2) 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 11:03:23,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:32,883 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:03:48,812 INFO [train.py:968] (1/2) Epoch 20, batch 22550, giga_loss[loss=0.2761, simple_loss=0.3501, pruned_loss=0.101, over 28643.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3516, pruned_loss=0.1013, over 5696059.79 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3531, pruned_loss=0.1013, over 5739053.55 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3495, pruned_loss=0.09962, over 5705069.87 frames. ], batch size: 85, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:04:15,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3443, 3.1519, 3.0086, 1.4471], device='cuda:1'), covar=tensor([0.0885, 0.1046, 0.0877, 0.2173], device='cuda:1'), in_proj_covar=tensor([0.1177, 0.1094, 0.0933, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 11:04:24,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6615, 4.6043, 1.9092, 1.8314], device='cuda:1'), covar=tensor([0.0931, 0.0287, 0.0886, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0546, 0.0379, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 11:04:32,965 INFO [train.py:968] (1/2) Epoch 20, batch 22600, libri_loss[loss=0.2614, simple_loss=0.3351, pruned_loss=0.09382, over 29562.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3498, pruned_loss=0.1006, over 5701735.74 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3535, pruned_loss=0.1016, over 5741489.95 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3478, pruned_loss=0.09893, over 5705815.33 frames. ], batch size: 76, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:04:50,763 INFO [optim.py:369] (1/2) 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 11:05:15,652 INFO [train.py:968] (1/2) Epoch 20, batch 22650, giga_loss[loss=0.2729, simple_loss=0.3452, pruned_loss=0.1003, over 28010.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3456, pruned_loss=0.09842, over 5706372.34 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.354, pruned_loss=0.1021, over 5742753.23 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3435, pruned_loss=0.0966, over 5708105.87 frames. ], batch size: 412, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:05:18,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=889815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:05:21,975 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-10 11:05:54,064 INFO [train.py:968] (1/2) Epoch 20, batch 22700, giga_loss[loss=0.2889, simple_loss=0.353, pruned_loss=0.1125, over 28510.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3446, pruned_loss=0.09756, over 5708730.31 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3543, pruned_loss=0.1026, over 5744962.09 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3421, pruned_loss=0.09545, over 5706846.34 frames. ], batch size: 60, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:06:10,109 INFO [optim.py:369] (1/2) 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 11:06:27,632 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8992, 1.2683, 1.2600, 1.1341], device='cuda:1'), covar=tensor([0.1582, 0.1029, 0.1902, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0744, 0.0710, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 11:06:35,427 INFO [train.py:968] (1/2) Epoch 20, batch 22750, giga_loss[loss=0.2983, simple_loss=0.379, pruned_loss=0.1088, over 28958.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3462, pruned_loss=0.09758, over 5705676.12 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3541, pruned_loss=0.1028, over 5746275.44 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3441, pruned_loss=0.09553, over 5701754.52 frames. ], batch size: 213, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:06:59,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7414, 1.9784, 1.5529, 1.8521], device='cuda:1'), covar=tensor([0.2748, 0.2840, 0.3300, 0.2651], device='cuda:1'), in_proj_covar=tensor([0.1475, 0.1071, 0.1306, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 11:07:15,933 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=889958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:07:18,062 INFO [train.py:968] (1/2) Epoch 20, batch 22800, giga_loss[loss=0.2273, simple_loss=0.3139, pruned_loss=0.07034, over 28804.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09788, over 5707186.56 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3541, pruned_loss=0.1028, over 5748117.58 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3468, pruned_loss=0.09622, over 5702084.83 frames. ], batch size: 174, lr: 1.60e-03, grad_scale: 8.0
+2023-03-10 11:07:19,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=889961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:07:36,176 INFO [optim.py:369] (1/2) 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 11:07:42,690 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=889990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:07:59,536 INFO [train.py:968] (1/2) Epoch 20, batch 22850, giga_loss[loss=0.2289, simple_loss=0.311, pruned_loss=0.07337, over 29042.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3466, pruned_loss=0.09787, over 5687961.99 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3543, pruned_loss=0.1031, over 5730789.87 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.345, pruned_loss=0.09623, over 5698533.40 frames. ], batch size: 136, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:08:06,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:08:40,719 INFO [train.py:968] (1/2) Epoch 20, batch 22900, giga_loss[loss=0.234, simple_loss=0.3131, pruned_loss=0.07748, over 28991.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3446, pruned_loss=0.09806, over 5695105.64 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.355, pruned_loss=0.1037, over 5733650.21 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3424, pruned_loss=0.09604, over 5699806.00 frames. ], batch size: 145, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:08:48,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3832, 1.6425, 1.5473, 1.3482], device='cuda:1'), covar=tensor([0.3201, 0.2413, 0.2114, 0.2653], device='cuda:1'), in_proj_covar=tensor([0.1946, 0.1863, 0.1794, 0.1928], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 11:08:57,518 INFO [optim.py:369] (1/2) 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 11:09:22,194 INFO [train.py:968] (1/2) Epoch 20, batch 22950, giga_loss[loss=0.2494, simple_loss=0.3122, pruned_loss=0.09333, over 28781.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3425, pruned_loss=0.09775, over 5701317.47 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3552, pruned_loss=0.1039, over 5733774.98 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3403, pruned_loss=0.09579, over 5704302.26 frames. ], batch size: 99, lr: 1.60e-03, grad_scale: 4.0
+2023-03-10 11:10:05,404 INFO [train.py:968] (1/2) Epoch 20, batch 23000, giga_loss[loss=0.2539, simple_loss=0.3317, pruned_loss=0.08803, over 28864.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3419, pruned_loss=0.09876, over 5700692.71 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3552, pruned_loss=0.104, over 5736331.52 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.34, pruned_loss=0.0971, over 5700421.71 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:10:22,363 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 23050, giga_loss[loss=0.2515, simple_loss=0.33, pruned_loss=0.08654, over 28837.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3411, pruned_loss=0.09821, over 5714364.77 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3556, pruned_loss=0.1043, over 5738960.05 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3389, pruned_loss=0.09644, over 5711161.08 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:11:02,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2508, 1.6180, 1.2726, 0.9530], device='cuda:1'), covar=tensor([0.2613, 0.2625, 0.3063, 0.2418], device='cuda:1'), in_proj_covar=tensor([0.1474, 0.1069, 0.1305, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 11:11:26,587 INFO [train.py:968] (1/2) Epoch 20, batch 23100, giga_loss[loss=0.2356, simple_loss=0.3114, pruned_loss=0.07991, over 29000.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.338, pruned_loss=0.09671, over 5708084.30 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3559, pruned_loss=0.1047, over 5736828.42 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3356, pruned_loss=0.09481, over 5707111.70 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:11:44,087 INFO [optim.py:369] (1/2) 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 11:12:08,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2368, 4.0933, 3.8487, 1.9371], device='cuda:1'), covar=tensor([0.0651, 0.0742, 0.0728, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.1194, 0.1107, 0.0942, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 11:12:08,465 INFO [train.py:968] (1/2) Epoch 20, batch 23150, giga_loss[loss=0.2113, simple_loss=0.2947, pruned_loss=0.06399, over 28957.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3332, pruned_loss=0.09424, over 5711025.30 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.356, pruned_loss=0.1049, over 5739536.00 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3309, pruned_loss=0.09241, over 5707515.75 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:12:48,257 INFO [train.py:968] (1/2) Epoch 20, batch 23200, libri_loss[loss=0.3061, simple_loss=0.355, pruned_loss=0.1286, over 29331.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3313, pruned_loss=0.09321, over 5708114.27 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3564, pruned_loss=0.1054, over 5733521.96 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3284, pruned_loss=0.09096, over 5708998.64 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:13:04,438 INFO [optim.py:369] (1/2) 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 11:13:17,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=890395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:13:31,331 INFO [train.py:968] (1/2) Epoch 20, batch 23250, giga_loss[loss=0.2859, simple_loss=0.3556, pruned_loss=0.1081, over 28938.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3325, pruned_loss=0.09329, over 5713550.44 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3563, pruned_loss=0.1053, over 5736725.67 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3298, pruned_loss=0.09129, over 5710843.73 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:13:52,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9158, 1.1303, 1.2412, 1.0452], device='cuda:1'), covar=tensor([0.1478, 0.1064, 0.1770, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0747, 0.0713, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 11:14:15,146 INFO [train.py:968] (1/2) Epoch 20, batch 23300, giga_loss[loss=0.2805, simple_loss=0.352, pruned_loss=0.1045, over 28418.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3356, pruned_loss=0.09459, over 5715650.83 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3564, pruned_loss=0.1055, over 5740049.79 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.333, pruned_loss=0.09264, over 5710099.16 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:14:18,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-10 11:14:32,227 INFO [optim.py:369] (1/2) 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 11:14:42,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:14:56,930 INFO [train.py:968] (1/2) Epoch 20, batch 23350, giga_loss[loss=0.2703, simple_loss=0.3518, pruned_loss=0.09437, over 28823.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.341, pruned_loss=0.09732, over 5708340.18 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3574, pruned_loss=0.1064, over 5735486.09 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3374, pruned_loss=0.09473, over 5707628.25 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:15:20,343 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=890538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:15:23,112 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 23400, libri_loss[loss=0.2991, simple_loss=0.3656, pruned_loss=0.1162, over 29532.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3437, pruned_loss=0.09853, over 5707238.25 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3573, pruned_loss=0.1065, over 5742325.85 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3404, pruned_loss=0.09593, over 5699120.69 frames. ], batch size: 81, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:15:46,176 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=890570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:15:57,382 INFO [optim.py:369] (1/2) 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 11:16:21,994 INFO [train.py:968] (1/2) Epoch 20, batch 23450, giga_loss[loss=0.2611, simple_loss=0.3291, pruned_loss=0.09653, over 24097.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3467, pruned_loss=0.1001, over 5688346.55 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3579, pruned_loss=0.1072, over 5728629.05 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.343, pruned_loss=0.0971, over 5693641.16 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:17:04,310 INFO [train.py:968] (1/2) Epoch 20, batch 23500, giga_loss[loss=0.3236, simple_loss=0.3834, pruned_loss=0.1319, over 28606.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3487, pruned_loss=0.1017, over 5697515.59 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3578, pruned_loss=0.1074, over 5733743.01 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3454, pruned_loss=0.09884, over 5695941.97 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:17:09,918 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:17:25,336 INFO [optim.py:369] (1/2) 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 11:17:52,901 INFO [train.py:968] (1/2) Epoch 20, batch 23550, giga_loss[loss=0.3162, simple_loss=0.3735, pruned_loss=0.1295, over 29009.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3536, pruned_loss=0.106, over 5689257.07 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3576, pruned_loss=0.1074, over 5736837.32 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3509, pruned_loss=0.1036, over 5684340.05 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:18:14,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-10 11:18:18,996 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:18:24,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-10 11:18:25,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6935, 1.8403, 1.3933, 1.3599], device='cuda:1'), covar=tensor([0.0899, 0.0606, 0.0973, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0447, 0.0516, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 11:18:42,509 INFO [train.py:968] (1/2) Epoch 20, batch 23600, giga_loss[loss=0.3302, simple_loss=0.4036, pruned_loss=0.1284, over 28731.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3609, pruned_loss=0.1115, over 5694363.93 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3581, pruned_loss=0.1081, over 5740594.68 frames. ], giga_tot_loss[loss=0.2882, simple_loss=0.3584, pruned_loss=0.1091, over 5685380.74 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:18:52,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:19:04,231 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.03 vs. limit=2.0
+2023-03-10 11:19:06,081 INFO [optim.py:369] (1/2) 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 11:19:21,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:19:29,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=890803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:19:35,699 INFO [train.py:968] (1/2) Epoch 20, batch 23650, giga_loss[loss=0.462, simple_loss=0.4721, pruned_loss=0.226, over 26537.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3678, pruned_loss=0.117, over 5687267.32 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3582, pruned_loss=0.1081, over 5743342.77 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3658, pruned_loss=0.1151, over 5676976.02 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:20:13,799 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 11:20:24,627 INFO [train.py:968] (1/2) Epoch 20, batch 23700, giga_loss[loss=0.3607, simple_loss=0.4073, pruned_loss=0.157, over 28346.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3757, pruned_loss=0.124, over 5680126.91 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3588, pruned_loss=0.1086, over 5747326.81 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3739, pruned_loss=0.1223, over 5667296.51 frames. ], batch size: 369, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:20:35,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=890869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:20:40,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 11:20:46,361 INFO [optim.py:369] (1/2) 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 11:21:09,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3241, 2.4949, 1.8627, 2.1449], device='cuda:1'), covar=tensor([0.0832, 0.0555, 0.0944, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0448, 0.0518, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 11:21:13,763 INFO [train.py:968] (1/2) Epoch 20, batch 23750, giga_loss[loss=0.3655, simple_loss=0.4158, pruned_loss=0.1576, over 28190.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.381, pruned_loss=0.1281, over 5678053.63 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.359, pruned_loss=0.1088, over 5749260.04 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3799, pruned_loss=0.1271, over 5664388.70 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:21:36,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3979, 1.3017, 3.5454, 3.1724], device='cuda:1'), covar=tensor([0.1356, 0.2519, 0.0455, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0636, 0.0941, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 11:22:03,684 INFO [train.py:968] (1/2) Epoch 20, batch 23800, giga_loss[loss=0.3449, simple_loss=0.3967, pruned_loss=0.1466, over 28533.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.383, pruned_loss=0.1298, over 5669438.54 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3593, pruned_loss=0.1091, over 5742663.35 frames. ], giga_tot_loss[loss=0.3203, simple_loss=0.3822, pruned_loss=0.1292, over 5662176.17 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:22:27,416 INFO [optim.py:369] (1/2) 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 11:22:57,067 INFO [train.py:968] (1/2) Epoch 20, batch 23850, giga_loss[loss=0.2886, simple_loss=0.3616, pruned_loss=0.1078, over 28962.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3846, pruned_loss=0.1325, over 5666032.59 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3593, pruned_loss=0.1091, over 5743379.59 frames. ], giga_tot_loss[loss=0.3243, simple_loss=0.3843, pruned_loss=0.1321, over 5658994.09 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:23:00,048 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:23:02,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:23:30,439 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 23900, giga_loss[loss=0.3046, simple_loss=0.3709, pruned_loss=0.1191, over 28780.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.386, pruned_loss=0.1344, over 5645375.55 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3594, pruned_loss=0.1093, over 5735692.89 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3865, pruned_loss=0.1347, over 5642926.75 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:24:12,784 INFO [optim.py:369] (1/2) 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 11:24:48,168 INFO [train.py:968] (1/2) Epoch 20, batch 23950, giga_loss[loss=0.3103, simple_loss=0.3781, pruned_loss=0.1213, over 28637.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3901, pruned_loss=0.138, over 5626715.16 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3593, pruned_loss=0.1093, over 5727924.31 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3908, pruned_loss=0.1386, over 5629980.35 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:24:49,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:25:27,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:25:42,840 INFO [train.py:968] (1/2) Epoch 20, batch 24000, giga_loss[loss=0.4274, simple_loss=0.4333, pruned_loss=0.2107, over 23532.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.391, pruned_loss=0.14, over 5599547.31 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3596, pruned_loss=0.1096, over 5722982.73 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3923, pruned_loss=0.1411, over 5603433.17 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:25:42,840 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 11:25:51,388 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 11:26:01,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:08,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:14,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:14,522 INFO [optim.py:369] (1/2) 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 11:26:17,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:26:27,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6850, 1.8377, 1.6986, 1.6600], device='cuda:1'), covar=tensor([0.1489, 0.1840, 0.1965, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0748, 0.0713, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 11:26:40,359 INFO [train.py:968] (1/2) Epoch 20, batch 24050, giga_loss[loss=0.3229, simple_loss=0.3793, pruned_loss=0.1333, over 28563.00 frames. ], tot_loss[loss=0.3335, simple_loss=0.3888, pruned_loss=0.1391, over 5617571.50 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3595, pruned_loss=0.1095, over 5724374.49 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3905, pruned_loss=0.1405, over 5617150.73 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:26:45,151 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:27:01,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2975, 4.0950, 3.9441, 1.8197], device='cuda:1'), covar=tensor([0.0688, 0.0839, 0.0832, 0.2066], device='cuda:1'), in_proj_covar=tensor([0.1211, 0.1125, 0.0956, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-10 11:27:08,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0855, 3.1763, 1.2793, 1.2971], device='cuda:1'), covar=tensor([0.1239, 0.0444, 0.0987, 0.1627], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0551, 0.0380, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 11:27:28,105 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:27:30,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:27:32,847 INFO [train.py:968] (1/2) Epoch 20, batch 24100, giga_loss[loss=0.4928, simple_loss=0.4736, pruned_loss=0.2559, over 23637.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3881, pruned_loss=0.1386, over 5627305.73 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3595, pruned_loss=0.1095, over 5724374.49 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3894, pruned_loss=0.1398, over 5626978.23 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:27:55,009 INFO [optim.py:369] (1/2) 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 11:27:56,634 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:00,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:23,337 INFO [train.py:968] (1/2) Epoch 20, batch 24150, giga_loss[loss=0.3763, simple_loss=0.4189, pruned_loss=0.1668, over 27589.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3874, pruned_loss=0.1377, over 5616595.94 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3596, pruned_loss=0.1098, over 5727201.40 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3893, pruned_loss=0.1393, over 5610883.05 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:28:28,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:31,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:34,159 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:34,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=891321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:28:37,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=891324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:29:01,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:29:08,509 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=891353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:29:16,131 INFO [train.py:968] (1/2) Epoch 20, batch 24200, giga_loss[loss=0.2806, simple_loss=0.3565, pruned_loss=0.1023, over 28442.00 frames. ], tot_loss[loss=0.3318, simple_loss=0.3886, pruned_loss=0.1375, over 5622318.34 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3595, pruned_loss=0.1098, over 5725733.70 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3905, pruned_loss=0.1391, over 5617681.73 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:29:37,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 11:29:44,439 INFO [optim.py:369] (1/2) 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 11:30:06,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-10 11:30:07,239 INFO [train.py:968] (1/2) Epoch 20, batch 24250, giga_loss[loss=0.2728, simple_loss=0.3503, pruned_loss=0.09765, over 28950.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.3886, pruned_loss=0.1373, over 5623193.45 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3595, pruned_loss=0.1099, over 5725051.38 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1392, over 5617807.71 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:30:58,501 INFO [train.py:968] (1/2) Epoch 20, batch 24300, giga_loss[loss=0.2917, simple_loss=0.3697, pruned_loss=0.1069, over 28905.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3859, pruned_loss=0.1345, over 5623592.74 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3598, pruned_loss=0.1102, over 5726261.69 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.388, pruned_loss=0.1363, over 5615780.22 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:31:25,655 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 24350, giga_loss[loss=0.3569, simple_loss=0.4044, pruned_loss=0.1547, over 27986.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3837, pruned_loss=0.1316, over 5631442.22 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3601, pruned_loss=0.1105, over 5727131.35 frames. ], giga_tot_loss[loss=0.3257, simple_loss=0.3854, pruned_loss=0.133, over 5623567.88 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:31:53,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3394, 2.0659, 1.5581, 0.5359], device='cuda:1'), covar=tensor([0.4868, 0.2655, 0.3812, 0.6167], device='cuda:1'), in_proj_covar=tensor([0.1716, 0.1615, 0.1584, 0.1398], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 11:32:38,196 INFO [train.py:968] (1/2) Epoch 20, batch 24400, giga_loss[loss=0.3139, simple_loss=0.3778, pruned_loss=0.125, over 29004.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3807, pruned_loss=0.1291, over 5614604.90 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3607, pruned_loss=0.1111, over 5708198.32 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.382, pruned_loss=0.1301, over 5623044.21 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:33:03,933 INFO [optim.py:369] (1/2) 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 11:33:29,399 INFO [train.py:968] (1/2) Epoch 20, batch 24450, giga_loss[loss=0.2659, simple_loss=0.3425, pruned_loss=0.09459, over 29003.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3773, pruned_loss=0.1267, over 5618749.93 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3608, pruned_loss=0.1114, over 5704082.63 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3788, pruned_loss=0.1277, over 5626764.15 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:33:58,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 24500, giga_loss[loss=0.2916, simple_loss=0.3577, pruned_loss=0.1128, over 28462.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3755, pruned_loss=0.1256, over 5621486.89 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3604, pruned_loss=0.1112, over 5704520.47 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3772, pruned_loss=0.1267, over 5626114.01 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:34:33,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4182, 1.6819, 1.3813, 1.2472], device='cuda:1'), covar=tensor([0.2368, 0.2419, 0.2677, 0.2291], device='cuda:1'), in_proj_covar=tensor([0.1476, 0.1071, 0.1309, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 11:34:44,915 INFO [optim.py:369] (1/2) 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 11:34:51,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 11:35:03,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-10 11:35:15,423 INFO [train.py:968] (1/2) Epoch 20, batch 24550, giga_loss[loss=0.4234, simple_loss=0.4414, pruned_loss=0.2027, over 26605.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1251, over 5621142.16 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3603, pruned_loss=0.1112, over 5702289.71 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3769, pruned_loss=0.1263, over 5625433.47 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:35:32,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8203, 4.6417, 4.4199, 1.9326], device='cuda:1'), covar=tensor([0.0557, 0.0712, 0.0748, 0.2179], device='cuda:1'), in_proj_covar=tensor([0.1208, 0.1126, 0.0955, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-10 11:36:04,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=891759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:36:05,881 INFO [train.py:968] (1/2) Epoch 20, batch 24600, giga_loss[loss=0.3312, simple_loss=0.388, pruned_loss=0.1372, over 27976.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3735, pruned_loss=0.1231, over 5635693.59 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3606, pruned_loss=0.1114, over 5696358.70 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3749, pruned_loss=0.1241, over 5643231.34 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:36:21,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5398, 4.3543, 4.1521, 2.0553], device='cuda:1'), covar=tensor([0.0566, 0.0743, 0.0762, 0.1993], device='cuda:1'), in_proj_covar=tensor([0.1210, 0.1128, 0.0956, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 11:36:36,047 INFO [optim.py:369] (1/2) 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 11:36:48,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4847, 1.7293, 1.4266, 1.6028], device='cuda:1'), covar=tensor([0.2504, 0.2497, 0.2637, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.1476, 0.1071, 0.1309, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 11:36:59,638 INFO [train.py:968] (1/2) Epoch 20, batch 24650, giga_loss[loss=0.309, simple_loss=0.3852, pruned_loss=0.1164, over 28828.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3738, pruned_loss=0.1211, over 5646099.08 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3602, pruned_loss=0.1113, over 5700719.61 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3756, pruned_loss=0.1222, over 5647021.50 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:37:54,245 INFO [train.py:968] (1/2) Epoch 20, batch 24700, libri_loss[loss=0.3043, simple_loss=0.3745, pruned_loss=0.1171, over 29527.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3753, pruned_loss=0.1208, over 5654637.21 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3602, pruned_loss=0.1113, over 5704950.25 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3769, pruned_loss=0.1218, over 5650569.52 frames. ], batch size: 82, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:38:14,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=891880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:38:20,082 INFO [optim.py:369] (1/2) 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 11:38:35,305 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6042, 1.7056, 1.6893, 1.5169], device='cuda:1'), covar=tensor([0.1847, 0.2253, 0.2260, 0.2290], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0747, 0.0711, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 11:38:47,854 INFO [train.py:968] (1/2) Epoch 20, batch 24750, giga_loss[loss=0.3189, simple_loss=0.3823, pruned_loss=0.1278, over 27960.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3757, pruned_loss=0.1216, over 5652343.47 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3605, pruned_loss=0.1115, over 5699616.54 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3771, pruned_loss=0.1224, over 5653147.83 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:38:48,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5858, 3.3251, 1.6988, 1.7343], device='cuda:1'), covar=tensor([0.0854, 0.0421, 0.0805, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0554, 0.0382, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 11:39:32,864 INFO [train.py:968] (1/2) Epoch 20, batch 24800, giga_loss[loss=0.3705, simple_loss=0.4129, pruned_loss=0.1641, over 28327.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3752, pruned_loss=0.1211, over 5659771.23 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3612, pruned_loss=0.1121, over 5689707.27 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3759, pruned_loss=0.1214, over 5668061.95 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:39:59,290 INFO [optim.py:369] (1/2) 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 11:40:23,429 INFO [train.py:968] (1/2) Epoch 20, batch 24850, giga_loss[loss=0.3106, simple_loss=0.3668, pruned_loss=0.1272, over 28564.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3735, pruned_loss=0.1211, over 5667577.55 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3609, pruned_loss=0.112, over 5688207.17 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3749, pruned_loss=0.1218, over 5674043.80 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:40:30,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:40:37,767 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:40:40,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:40:58,873 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.62 vs. limit=5.0
+2023-03-10 11:41:04,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:41:10,482 INFO [train.py:968] (1/2) Epoch 20, batch 24900, giga_loss[loss=0.3081, simple_loss=0.3811, pruned_loss=0.1176, over 28866.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3718, pruned_loss=0.1211, over 5672541.70 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3604, pruned_loss=0.1118, over 5693590.30 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3737, pruned_loss=0.1221, over 5671893.56 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:41:34,355 INFO [optim.py:369] (1/2) 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 11:41:49,255 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:41:56,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3693, 1.5725, 1.4331, 1.2180], device='cuda:1'), covar=tensor([0.2632, 0.2280, 0.1936, 0.2373], device='cuda:1'), in_proj_covar=tensor([0.1946, 0.1873, 0.1804, 0.1931], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 11:41:57,224 INFO [train.py:968] (1/2) Epoch 20, batch 24950, giga_loss[loss=0.2675, simple_loss=0.3449, pruned_loss=0.0951, over 28336.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3707, pruned_loss=0.1199, over 5671415.99 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3603, pruned_loss=0.1118, over 5694360.01 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3726, pruned_loss=0.1209, over 5669713.84 frames. ], batch size: 77, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:42:07,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2847, 1.7801, 1.4725, 1.4713], device='cuda:1'), covar=tensor([0.0799, 0.0332, 0.0331, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0118, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 11:42:17,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:42:42,852 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 25000, giga_loss[loss=0.29, simple_loss=0.3689, pruned_loss=0.1055, over 28860.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3703, pruned_loss=0.1182, over 5675120.55 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3605, pruned_loss=0.112, over 5688982.65 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3719, pruned_loss=0.119, over 5677380.84 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:42:44,735 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:43:08,916 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 11:43:12,323 INFO [optim.py:369] (1/2) 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 11:43:15,942 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:43:34,339 INFO [train.py:968] (1/2) Epoch 20, batch 25050, giga_loss[loss=0.2958, simple_loss=0.3696, pruned_loss=0.111, over 28688.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3709, pruned_loss=0.1188, over 5656447.49 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3613, pruned_loss=0.1128, over 5677290.87 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.372, pruned_loss=0.119, over 5669448.51 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:44:20,576 INFO [train.py:968] (1/2) Epoch 20, batch 25100, giga_loss[loss=0.3368, simple_loss=0.3971, pruned_loss=0.1383, over 28855.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3697, pruned_loss=0.118, over 5670111.48 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3611, pruned_loss=0.1129, over 5685246.06 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3711, pruned_loss=0.1184, over 5673147.71 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:44:36,533 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:44:39,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:44:44,027 INFO [optim.py:369] (1/2) 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 11:45:06,273 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:45:07,367 INFO [train.py:968] (1/2) Epoch 20, batch 25150, giga_loss[loss=0.3157, simple_loss=0.3723, pruned_loss=0.1296, over 29115.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3695, pruned_loss=0.1187, over 5678717.94 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3613, pruned_loss=0.113, over 5692328.59 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3708, pruned_loss=0.1191, over 5674562.13 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:45:14,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-10 11:45:58,806 INFO [train.py:968] (1/2) Epoch 20, batch 25200, giga_loss[loss=0.245, simple_loss=0.3164, pruned_loss=0.08677, over 28547.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3686, pruned_loss=0.1191, over 5662578.86 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.361, pruned_loss=0.1129, over 5692976.52 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.37, pruned_loss=0.1196, over 5658455.11 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:46:23,146 INFO [optim.py:369] (1/2) 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 11:46:46,088 INFO [train.py:968] (1/2) Epoch 20, batch 25250, giga_loss[loss=0.3451, simple_loss=0.398, pruned_loss=0.1461, over 28872.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3698, pruned_loss=0.1205, over 5667832.86 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3612, pruned_loss=0.1129, over 5694254.97 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3709, pruned_loss=0.1211, over 5662879.90 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:47:23,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:47:36,561 INFO [train.py:968] (1/2) Epoch 20, batch 25300, giga_loss[loss=0.2593, simple_loss=0.336, pruned_loss=0.09135, over 28700.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3686, pruned_loss=0.1202, over 5662787.58 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3613, pruned_loss=0.113, over 5694564.82 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5658143.89 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:47:43,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5276, 1.8559, 1.4536, 1.7131], device='cuda:1'), covar=tensor([0.2579, 0.2666, 0.3053, 0.2335], device='cuda:1'), in_proj_covar=tensor([0.1479, 0.1073, 0.1312, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 11:47:53,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:48:02,428 INFO [optim.py:369] (1/2) 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 11:48:23,239 INFO [train.py:968] (1/2) Epoch 20, batch 25350, giga_loss[loss=0.3153, simple_loss=0.3743, pruned_loss=0.1282, over 28585.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3668, pruned_loss=0.1192, over 5677195.34 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.361, pruned_loss=0.1129, over 5701057.42 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5666715.92 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:48:56,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-10 11:49:12,064 INFO [train.py:968] (1/2) Epoch 20, batch 25400, libri_loss[loss=0.3022, simple_loss=0.3691, pruned_loss=0.1176, over 29749.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3672, pruned_loss=0.1198, over 5671534.00 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3614, pruned_loss=0.1131, over 5704150.71 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.368, pruned_loss=0.1203, over 5658963.52 frames. ], batch size: 87, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:49:35,622 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 25450, giga_loss[loss=0.2773, simple_loss=0.3595, pruned_loss=0.09749, over 28827.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.369, pruned_loss=0.1209, over 5647694.12 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3617, pruned_loss=0.1136, over 5682800.05 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3695, pruned_loss=0.1212, over 5655108.24 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:50:06,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:50:09,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:50:35,917 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=892652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:50:45,873 INFO [train.py:968] (1/2) Epoch 20, batch 25500, giga_loss[loss=0.3853, simple_loss=0.4153, pruned_loss=0.1777, over 26633.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3691, pruned_loss=0.1201, over 5656231.88 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3617, pruned_loss=0.1138, over 5684922.26 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3696, pruned_loss=0.1202, over 5659998.59 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:50:52,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3021, 4.1265, 3.9228, 2.0469], device='cuda:1'), covar=tensor([0.0581, 0.0703, 0.0723, 0.1980], device='cuda:1'), in_proj_covar=tensor([0.1215, 0.1128, 0.0956, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 11:50:55,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7823, 1.8888, 1.3931, 1.4939], device='cuda:1'), covar=tensor([0.1024, 0.0757, 0.1058, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0446, 0.0514, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 11:51:12,406 INFO [optim.py:369] (1/2) 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 11:51:32,752 INFO [train.py:968] (1/2) Epoch 20, batch 25550, libri_loss[loss=0.2886, simple_loss=0.3597, pruned_loss=0.1088, over 27593.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3689, pruned_loss=0.1193, over 5659674.67 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3622, pruned_loss=0.1141, over 5688348.16 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.369, pruned_loss=0.1192, over 5658848.17 frames. ], batch size: 116, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:51:43,072 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:51:54,290 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=892733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:52:08,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4512, 1.6310, 1.6666, 1.2364], device='cuda:1'), covar=tensor([0.1615, 0.2687, 0.1417, 0.1762], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0696, 0.0928, 0.0829], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 11:52:22,295 INFO [train.py:968] (1/2) Epoch 20, batch 25600, giga_loss[loss=0.3331, simple_loss=0.3893, pruned_loss=0.1385, over 28433.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3698, pruned_loss=0.1203, over 5653989.72 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3624, pruned_loss=0.1143, over 5682982.80 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3698, pruned_loss=0.1202, over 5657223.26 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:52:47,835 INFO [optim.py:369] (1/2) 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 11:53:12,115 INFO [train.py:968] (1/2) Epoch 20, batch 25650, giga_loss[loss=0.2654, simple_loss=0.34, pruned_loss=0.09538, over 28832.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1238, over 5650410.17 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3625, pruned_loss=0.1143, over 5687016.55 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3736, pruned_loss=0.1239, over 5648675.84 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 11:53:14,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4188, 3.2498, 3.1237, 2.0329], device='cuda:1'), covar=tensor([0.0795, 0.0940, 0.0873, 0.1666], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.1130, 0.0959, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 11:53:22,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=892824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:53:46,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-10 11:53:55,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-10 11:54:00,550 INFO [train.py:968] (1/2) Epoch 20, batch 25700, giga_loss[loss=0.4016, simple_loss=0.436, pruned_loss=0.1836, over 28742.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3739, pruned_loss=0.1254, over 5630643.35 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3623, pruned_loss=0.1143, over 5660226.61 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3745, pruned_loss=0.1257, over 5653307.53 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:54:33,534 INFO [optim.py:369] (1/2) 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 11:54:57,695 INFO [train.py:968] (1/2) Epoch 20, batch 25750, giga_loss[loss=0.3941, simple_loss=0.4315, pruned_loss=0.1783, over 23944.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3759, pruned_loss=0.128, over 5634006.66 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3622, pruned_loss=0.1143, over 5662855.88 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3765, pruned_loss=0.1284, over 5649181.78 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:55:42,871 INFO [train.py:968] (1/2) Epoch 20, batch 25800, giga_loss[loss=0.2836, simple_loss=0.3511, pruned_loss=0.108, over 28731.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3759, pruned_loss=0.1281, over 5641674.44 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3621, pruned_loss=0.1141, over 5666027.43 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3769, pruned_loss=0.1289, over 5650065.25 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:55:49,775 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=892967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:55:52,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=892970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:56:13,011 INFO [optim.py:369] (1/2) 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 11:56:20,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3026, 4.1073, 3.9065, 1.7528], device='cuda:1'), covar=tensor([0.0801, 0.0985, 0.1199, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1221, 0.1132, 0.0962, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 11:56:21,807 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 25850, giga_loss[loss=0.3956, simple_loss=0.4192, pruned_loss=0.186, over 24157.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3752, pruned_loss=0.1278, over 5635939.48 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3623, pruned_loss=0.1144, over 5659776.17 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3758, pruned_loss=0.1284, over 5648311.20 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:56:43,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3461, 1.4645, 1.4110, 1.4327], device='cuda:1'), covar=tensor([0.0732, 0.0381, 0.0324, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0069, 0.0061, 0.0105], device='cuda:1')
+2023-03-10 11:57:14,939 INFO [train.py:968] (1/2) Epoch 20, batch 25900, giga_loss[loss=0.2706, simple_loss=0.3503, pruned_loss=0.09546, over 28907.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3737, pruned_loss=0.1255, over 5635421.13 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3617, pruned_loss=0.1142, over 5645334.85 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3753, pruned_loss=0.1267, over 5657840.56 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:57:24,713 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=893071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:57:42,243 INFO [optim.py:369] (1/2) 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 11:57:47,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=893093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 11:58:03,526 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 25950, giga_loss[loss=0.3136, simple_loss=0.3697, pruned_loss=0.1288, over 27957.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1225, over 5637986.53 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5647316.42 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3722, pruned_loss=0.1234, over 5653963.81 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 11:58:11,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-10 11:58:32,898 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7392, 2.5106, 1.8275, 0.9356], device='cuda:1'), covar=tensor([0.5091, 0.2783, 0.3812, 0.5474], device='cuda:1'), in_proj_covar=tensor([0.1725, 0.1637, 0.1594, 0.1408], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 11:58:36,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6626, 1.6032, 1.8986, 1.4903], device='cuda:1'), covar=tensor([0.1440, 0.2000, 0.1206, 0.1548], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0699, 0.0932, 0.0831], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 11:58:50,926 INFO [train.py:968] (1/2) Epoch 20, batch 26000, giga_loss[loss=0.3231, simple_loss=0.3761, pruned_loss=0.135, over 27934.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3695, pruned_loss=0.1215, over 5637171.64 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1145, over 5642803.63 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3706, pruned_loss=0.1223, over 5654220.26 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 11:59:17,529 INFO [optim.py:369] (1/2) 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 11:59:39,590 INFO [train.py:968] (1/2) Epoch 20, batch 26050, giga_loss[loss=0.3303, simple_loss=0.3815, pruned_loss=0.1396, over 27450.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3676, pruned_loss=0.1209, over 5642715.70 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.115, over 5633014.55 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.368, pruned_loss=0.1211, over 5664467.57 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:00:09,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:12,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:25,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:28,639 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:34,871 INFO [train.py:968] (1/2) Epoch 20, batch 26100, giga_loss[loss=0.3282, simple_loss=0.3893, pruned_loss=0.1335, over 28896.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3674, pruned_loss=0.1205, over 5653017.11 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3632, pruned_loss=0.1152, over 5634486.88 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3674, pruned_loss=0.1207, over 5669025.08 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:00:39,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=893268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:40,567 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=893269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:00:50,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3260, 3.1552, 2.9828, 1.5055], device='cuda:1'), covar=tensor([0.0955, 0.1078, 0.1063, 0.2173], device='cuda:1'), in_proj_covar=tensor([0.1221, 0.1133, 0.0966, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 12:00:53,875 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:01:18,261 INFO [train.py:968] (1/2) Epoch 20, batch 26150, libri_loss[loss=0.3339, simple_loss=0.3852, pruned_loss=0.1413, over 19093.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1218, over 5655804.67 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3633, pruned_loss=0.1153, over 5631773.15 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3704, pruned_loss=0.122, over 5673477.80 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:02:06,889 INFO [train.py:968] (1/2) Epoch 20, batch 26200, giga_loss[loss=0.3523, simple_loss=0.4223, pruned_loss=0.1411, over 28830.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3733, pruned_loss=0.1209, over 5657989.69 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3638, pruned_loss=0.116, over 5628024.00 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3731, pruned_loss=0.1206, over 5676784.39 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:02:33,811 INFO [optim.py:369] (1/2) 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 12:02:57,993 INFO [train.py:968] (1/2) Epoch 20, batch 26250, giga_loss[loss=0.3308, simple_loss=0.3934, pruned_loss=0.1341, over 28920.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.375, pruned_loss=0.1213, over 5665427.58 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3637, pruned_loss=0.1159, over 5630277.18 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3752, pruned_loss=0.1212, over 5679008.35 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:03:33,749 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 26300, giga_loss[loss=0.2937, simple_loss=0.3677, pruned_loss=0.1099, over 29030.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.376, pruned_loss=0.1228, over 5668427.57 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1156, over 5633515.13 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3772, pruned_loss=0.1231, over 5677003.87 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:03:49,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2773, 2.6350, 1.3176, 1.4284], device='cuda:1'), covar=tensor([0.0953, 0.0434, 0.0915, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0553, 0.0380, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 12:04:14,776 INFO [optim.py:369] (1/2) 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 12:04:26,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9212, 1.2166, 1.2908, 1.0837], device='cuda:1'), covar=tensor([0.1641, 0.1254, 0.2118, 0.1442], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0748, 0.0710, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 12:04:31,765 INFO [train.py:968] (1/2) Epoch 20, batch 26350, giga_loss[loss=0.3036, simple_loss=0.3718, pruned_loss=0.1177, over 28937.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3769, pruned_loss=0.1236, over 5680340.75 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.363, pruned_loss=0.1157, over 5639816.02 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.378, pruned_loss=0.124, over 5682618.14 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:05:22,585 INFO [train.py:968] (1/2) Epoch 20, batch 26400, giga_loss[loss=0.329, simple_loss=0.3916, pruned_loss=0.1332, over 28630.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3779, pruned_loss=0.1255, over 5675468.55 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.363, pruned_loss=0.1156, over 5644624.06 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3791, pruned_loss=0.1261, over 5673729.58 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:05:48,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:05:49,722 INFO [optim.py:369] (1/2) 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 12:05:50,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1272, 1.2246, 3.3179, 2.9669], device='cuda:1'), covar=tensor([0.1594, 0.2617, 0.0520, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0751, 0.0643, 0.0953, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 12:05:50,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:06:09,730 INFO [train.py:968] (1/2) Epoch 20, batch 26450, libri_loss[loss=0.303, simple_loss=0.3763, pruned_loss=0.1149, over 29350.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3752, pruned_loss=0.1238, over 5690247.01 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3628, pruned_loss=0.1154, over 5655729.36 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3771, pruned_loss=0.125, over 5680643.10 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:06:16,370 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=893621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:06:36,946 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 26500, libri_loss[loss=0.2688, simple_loss=0.3298, pruned_loss=0.1039, over 29656.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3729, pruned_loss=0.1229, over 5685735.71 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3622, pruned_loss=0.1151, over 5654849.07 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3755, pruned_loss=0.1245, over 5679625.58 frames. ], batch size: 69, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:07:18,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6557, 4.4786, 4.2711, 2.0685], device='cuda:1'), covar=tensor([0.0488, 0.0623, 0.0668, 0.2137], device='cuda:1'), in_proj_covar=tensor([0.1222, 0.1138, 0.0968, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 12:07:25,974 INFO [optim.py:369] (1/2) 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 12:07:50,664 INFO [train.py:968] (1/2) Epoch 20, batch 26550, giga_loss[loss=0.2736, simple_loss=0.3434, pruned_loss=0.1019, over 28265.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1226, over 5690562.21 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3623, pruned_loss=0.1152, over 5656083.82 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1238, over 5684842.55 frames. ], batch size: 65, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:08:37,454 INFO [train.py:968] (1/2) Epoch 20, batch 26600, giga_loss[loss=0.2676, simple_loss=0.3499, pruned_loss=0.09269, over 28873.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3719, pruned_loss=0.1233, over 5684183.99 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1151, over 5662696.59 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1245, over 5674606.38 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:09:02,522 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=893787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:09:03,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4601, 4.2372, 4.0078, 1.9093], device='cuda:1'), covar=tensor([0.0699, 0.0893, 0.0990, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1221, 0.1134, 0.0964, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 12:09:04,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=893790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:09:04,906 INFO [optim.py:369] (1/2) 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 12:09:22,656 INFO [train.py:968] (1/2) Epoch 20, batch 26650, giga_loss[loss=0.348, simple_loss=0.402, pruned_loss=0.147, over 27944.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3718, pruned_loss=0.1236, over 5687877.08 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3623, pruned_loss=0.1152, over 5668528.36 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3735, pruned_loss=0.1248, over 5675602.85 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:09:29,374 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=893819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:09:38,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 12:10:11,459 INFO [train.py:968] (1/2) Epoch 20, batch 26700, giga_loss[loss=0.3178, simple_loss=0.383, pruned_loss=0.1263, over 28602.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.37, pruned_loss=0.1234, over 5668815.60 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5672756.78 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3719, pruned_loss=0.1248, over 5655539.65 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:10:43,515 INFO [optim.py:369] (1/2) 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 12:11:03,997 INFO [train.py:968] (1/2) Epoch 20, batch 26750, giga_loss[loss=0.2784, simple_loss=0.3509, pruned_loss=0.1029, over 28661.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3699, pruned_loss=0.123, over 5666425.09 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3622, pruned_loss=0.1151, over 5672340.24 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3711, pruned_loss=0.1239, over 5656061.71 frames. ], batch size: 60, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:11:09,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 12:11:41,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6842, 1.9707, 2.0012, 1.7215], device='cuda:1'), covar=tensor([0.1709, 0.1709, 0.1848, 0.1696], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0753, 0.0716, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-10 12:11:50,059 INFO [train.py:968] (1/2) Epoch 20, batch 26800, giga_loss[loss=0.331, simple_loss=0.3867, pruned_loss=0.1377, over 28881.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3705, pruned_loss=0.1222, over 5669248.15 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.1151, over 5676290.54 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1232, over 5657325.87 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:12:18,284 INFO [optim.py:369] (1/2) 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 12:12:41,248 INFO [train.py:968] (1/2) Epoch 20, batch 26850, giga_loss[loss=0.3403, simple_loss=0.3952, pruned_loss=0.1427, over 28193.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5665930.13 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3619, pruned_loss=0.1147, over 5681372.11 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3743, pruned_loss=0.1251, over 5651681.20 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:13:24,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6997, 2.0728, 1.5395, 1.9952], device='cuda:1'), covar=tensor([0.2620, 0.2646, 0.3074, 0.2374], device='cuda:1'), in_proj_covar=tensor([0.1483, 0.1077, 0.1317, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 12:13:26,031 INFO [train.py:968] (1/2) Epoch 20, batch 26900, giga_loss[loss=0.2551, simple_loss=0.332, pruned_loss=0.08903, over 28607.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3729, pruned_loss=0.1243, over 5666171.21 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3621, pruned_loss=0.1149, over 5676222.99 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1255, over 5659451.24 frames. ], batch size: 60, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:13:54,179 INFO [optim.py:369] (1/2) 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 12:14:10,364 INFO [train.py:968] (1/2) Epoch 20, batch 26950, giga_loss[loss=0.3221, simple_loss=0.3841, pruned_loss=0.13, over 28311.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3734, pruned_loss=0.122, over 5665426.98 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.362, pruned_loss=0.1147, over 5673222.07 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.375, pruned_loss=0.1234, over 5663152.10 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:14:58,715 INFO [train.py:968] (1/2) Epoch 20, batch 27000, giga_loss[loss=0.3039, simple_loss=0.377, pruned_loss=0.1154, over 28893.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3742, pruned_loss=0.1202, over 5682591.36 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1141, over 5678187.55 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3768, pruned_loss=0.122, over 5676322.90 frames. ], batch size: 285, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:14:58,715 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 12:15:07,804 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 12:15:35,842 INFO [optim.py:369] (1/2) 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 12:15:52,948 INFO [train.py:968] (1/2) Epoch 20, batch 27050, libri_loss[loss=0.3041, simple_loss=0.3692, pruned_loss=0.1195, over 29295.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.377, pruned_loss=0.1213, over 5690179.94 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3604, pruned_loss=0.1138, over 5680654.46 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3797, pruned_loss=0.1231, over 5683158.11 frames. ], batch size: 94, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:16:12,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-10 12:16:41,066 INFO [train.py:968] (1/2) Epoch 20, batch 27100, libri_loss[loss=0.2554, simple_loss=0.3255, pruned_loss=0.09261, over 29554.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3801, pruned_loss=0.1249, over 5677494.92 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3602, pruned_loss=0.1136, over 5675562.46 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3829, pruned_loss=0.1267, over 5676391.37 frames. ], batch size: 76, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:16:45,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-10 12:17:14,599 INFO [optim.py:369] (1/2) 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 12:17:16,361 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:17:30,485 INFO [train.py:968] (1/2) Epoch 20, batch 27150, giga_loss[loss=0.4112, simple_loss=0.4222, pruned_loss=0.2001, over 23368.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3824, pruned_loss=0.1281, over 5664976.26 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3601, pruned_loss=0.1137, over 5678311.93 frames. ], giga_tot_loss[loss=0.3223, simple_loss=0.3851, pruned_loss=0.1297, over 5661456.84 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:17:57,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4241, 3.8868, 1.5537, 1.5600], device='cuda:1'), covar=tensor([0.0930, 0.0389, 0.0919, 0.1301], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0554, 0.0380, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 12:18:21,566 INFO [train.py:968] (1/2) Epoch 20, batch 27200, giga_loss[loss=0.2857, simple_loss=0.363, pruned_loss=0.1042, over 28901.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3808, pruned_loss=0.1276, over 5663224.73 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3599, pruned_loss=0.1136, over 5684426.09 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3839, pruned_loss=0.1295, over 5654232.49 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:18:57,313 INFO [optim.py:369] (1/2) 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 12:19:10,878 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 12:19:12,290 INFO [train.py:968] (1/2) Epoch 20, batch 27250, giga_loss[loss=0.3096, simple_loss=0.3754, pruned_loss=0.1219, over 27988.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3796, pruned_loss=0.1269, over 5658166.61 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.36, pruned_loss=0.1137, over 5690356.07 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3823, pruned_loss=0.1286, over 5644661.36 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:19:58,724 INFO [train.py:968] (1/2) Epoch 20, batch 27300, giga_loss[loss=0.3229, simple_loss=0.3904, pruned_loss=0.1277, over 27923.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3791, pruned_loss=0.1248, over 5664891.05 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3604, pruned_loss=0.1139, over 5694072.88 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3814, pruned_loss=0.1262, over 5650415.58 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:20:27,555 INFO [optim.py:369] (1/2) 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 12:20:41,452 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 27350, giga_loss[loss=0.2886, simple_loss=0.3692, pruned_loss=0.104, over 28826.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3775, pruned_loss=0.1219, over 5672471.35 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.114, over 5689147.87 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3798, pruned_loss=0.1231, over 5664736.23 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:20:56,434 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:21:37,675 INFO [train.py:968] (1/2) Epoch 20, batch 27400, giga_loss[loss=0.326, simple_loss=0.3884, pruned_loss=0.1318, over 28701.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3788, pruned_loss=0.1236, over 5669825.61 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3605, pruned_loss=0.1142, over 5691168.47 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3809, pruned_loss=0.1246, over 5661502.10 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:22:12,586 INFO [optim.py:369] (1/2) 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 12:22:29,084 INFO [train.py:968] (1/2) Epoch 20, batch 27450, libri_loss[loss=0.2907, simple_loss=0.3555, pruned_loss=0.1129, over 29568.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3796, pruned_loss=0.1248, over 5665456.06 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1141, over 5685679.08 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3819, pruned_loss=0.1259, over 5662982.56 frames. ], batch size: 77, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:23:13,336 INFO [train.py:968] (1/2) Epoch 20, batch 27500, giga_loss[loss=0.3225, simple_loss=0.3768, pruned_loss=0.1341, over 28497.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3782, pruned_loss=0.1248, over 5673298.73 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3613, pruned_loss=0.1147, over 5694375.60 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.38, pruned_loss=0.1256, over 5662523.79 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:23:21,417 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=894669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:23:23,746 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=894672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:23:28,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7285, 1.9482, 1.6748, 1.8128], device='cuda:1'), covar=tensor([0.1990, 0.1940, 0.1964, 0.1856], device='cuda:1'), in_proj_covar=tensor([0.1482, 0.1074, 0.1316, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 12:23:45,850 INFO [optim.py:369] (1/2) 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 12:24:00,624 INFO [train.py:968] (1/2) Epoch 20, batch 27550, giga_loss[loss=0.3125, simple_loss=0.3508, pruned_loss=0.1371, over 23296.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3769, pruned_loss=0.1254, over 5659977.51 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5700589.16 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3789, pruned_loss=0.1264, over 5644444.17 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 1.0
+2023-03-10 12:24:57,536 INFO [train.py:968] (1/2) Epoch 20, batch 27600, giga_loss[loss=0.2666, simple_loss=0.3369, pruned_loss=0.09813, over 28333.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1247, over 5649169.72 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3612, pruned_loss=0.1148, over 5690375.59 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1257, over 5644321.01 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:25:31,102 INFO [optim.py:369] (1/2) 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 12:25:37,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2221, 0.7457, 0.8641, 1.3908], device='cuda:1'), covar=tensor([0.0783, 0.0404, 0.0369, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0105], device='cuda:1')
+2023-03-10 12:25:44,898 INFO [train.py:968] (1/2) Epoch 20, batch 27650, giga_loss[loss=0.2942, simple_loss=0.3672, pruned_loss=0.1106, over 28826.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1245, over 5662327.19 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3611, pruned_loss=0.1148, over 5698768.45 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3756, pruned_loss=0.1256, over 5649829.58 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:25:46,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=894812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:25:48,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=894815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:26:07,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4002, 2.4088, 2.0559, 2.1907], device='cuda:1'), covar=tensor([0.0598, 0.0434, 0.0661, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0447, 0.0515, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 12:26:14,437 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=894844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:26:31,017 INFO [train.py:968] (1/2) Epoch 20, batch 27700, giga_loss[loss=0.4381, simple_loss=0.452, pruned_loss=0.212, over 26411.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3734, pruned_loss=0.1251, over 5647616.48 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3612, pruned_loss=0.1148, over 5693156.97 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3754, pruned_loss=0.1263, over 5641703.36 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:26:50,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=894880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:27:02,163 INFO [optim.py:369] (1/2) 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 12:27:03,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=894896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:27:14,909 INFO [train.py:968] (1/2) Epoch 20, batch 27750, libri_loss[loss=0.2555, simple_loss=0.3234, pruned_loss=0.09384, over 29549.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3703, pruned_loss=0.1222, over 5650234.00 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3611, pruned_loss=0.1147, over 5687354.71 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3725, pruned_loss=0.1237, over 5648269.77 frames. ], batch size: 79, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:27:50,810 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-10 12:28:04,649 INFO [train.py:968] (1/2) Epoch 20, batch 27800, libri_loss[loss=0.2981, simple_loss=0.3694, pruned_loss=0.1134, over 29541.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1177, over 5655849.98 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3611, pruned_loss=0.1147, over 5689478.05 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3683, pruned_loss=0.1188, over 5651937.33 frames. ], batch size: 89, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:28:06,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2119, 1.2094, 3.3693, 3.0141], device='cuda:1'), covar=tensor([0.1554, 0.2817, 0.0454, 0.1766], device='cuda:1'), in_proj_covar=tensor([0.0751, 0.0644, 0.0954, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 12:28:23,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-10 12:28:36,065 INFO [optim.py:369] (1/2) 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 12:28:38,935 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 27850, giga_loss[loss=0.3389, simple_loss=0.393, pruned_loss=0.1424, over 27540.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3647, pruned_loss=0.1159, over 5650581.55 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.361, pruned_loss=0.1147, over 5682525.97 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3663, pruned_loss=0.1169, over 5652846.55 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:29:08,650 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:11,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:24,360 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:26,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:32,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:39,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:29:44,587 INFO [train.py:968] (1/2) Epoch 20, batch 27900, giga_loss[loss=0.2799, simple_loss=0.3469, pruned_loss=0.1065, over 29055.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3639, pruned_loss=0.116, over 5635038.65 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3608, pruned_loss=0.1146, over 5676469.05 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3654, pruned_loss=0.117, over 5641674.08 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:29:57,709 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:30:21,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:30:23,059 INFO [optim.py:369] (1/2) 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 12:30:38,828 INFO [train.py:968] (1/2) Epoch 20, batch 27950, giga_loss[loss=0.2811, simple_loss=0.3522, pruned_loss=0.105, over 28852.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3601, pruned_loss=0.1138, over 5663673.66 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3604, pruned_loss=0.1142, over 5684007.38 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3617, pruned_loss=0.1149, over 5661270.67 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:31:32,419 INFO [train.py:968] (1/2) Epoch 20, batch 28000, libri_loss[loss=0.2956, simple_loss=0.3568, pruned_loss=0.1172, over 29555.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3629, pruned_loss=0.116, over 5665333.47 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3605, pruned_loss=0.1142, over 5686936.73 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3641, pruned_loss=0.1168, over 5660458.27 frames. ], batch size: 75, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:31:37,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3972, 2.3664, 1.8071, 2.1649], device='cuda:1'), covar=tensor([0.0770, 0.0577, 0.0897, 0.0936], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0448, 0.0516, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 12:32:01,833 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:32:06,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:32:07,095 INFO [optim.py:369] (1/2) 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 12:32:17,193 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 28050, giga_loss[loss=0.3079, simple_loss=0.3779, pruned_loss=0.119, over 28930.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3676, pruned_loss=0.1192, over 5658402.46 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.1141, over 5691208.59 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3689, pruned_loss=0.1201, over 5650169.70 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:32:34,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:33:12,538 INFO [train.py:968] (1/2) Epoch 20, batch 28100, giga_loss[loss=0.3098, simple_loss=0.3767, pruned_loss=0.1215, over 28915.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.1199, over 5657289.15 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.1141, over 5693495.19 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.37, pruned_loss=0.1207, over 5648173.32 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:33:45,696 INFO [optim.py:369] (1/2) 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 12:34:01,529 INFO [train.py:968] (1/2) Epoch 20, batch 28150, giga_loss[loss=0.3467, simple_loss=0.3961, pruned_loss=0.1487, over 28939.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3701, pruned_loss=0.1209, over 5658845.87 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3609, pruned_loss=0.1143, over 5696930.99 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1214, over 5648083.15 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:34:01,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2046, 2.5036, 1.3309, 1.3683], device='cuda:1'), covar=tensor([0.0948, 0.0364, 0.0835, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0554, 0.0382, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 12:34:44,505 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-10 12:34:47,608 INFO [train.py:968] (1/2) Epoch 20, batch 28200, giga_loss[loss=0.3043, simple_loss=0.3752, pruned_loss=0.1167, over 28982.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.371, pruned_loss=0.1217, over 5665765.42 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5698906.64 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1222, over 5654968.39 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:34:58,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:35:21,656 INFO [optim.py:369] (1/2) 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 12:35:36,807 INFO [train.py:968] (1/2) Epoch 20, batch 28250, giga_loss[loss=0.3352, simple_loss=0.3953, pruned_loss=0.1376, over 28275.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1226, over 5669082.57 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3612, pruned_loss=0.1142, over 5703967.04 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5655107.12 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:36:20,998 INFO [train.py:968] (1/2) Epoch 20, batch 28300, giga_loss[loss=0.3165, simple_loss=0.3833, pruned_loss=0.1248, over 28348.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3729, pruned_loss=0.1223, over 5673093.69 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1142, over 5704857.28 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3737, pruned_loss=0.1231, over 5660205.34 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:36:25,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3238, 1.4744, 1.3104, 1.7073], device='cuda:1'), covar=tensor([0.0727, 0.0323, 0.0331, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0069, 0.0062, 0.0105], device='cuda:1')
+2023-03-10 12:36:28,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:36:52,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6784, 1.8427, 1.4762, 1.3546], device='cuda:1'), covar=tensor([0.0919, 0.0565, 0.0904, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0447, 0.0516, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 12:36:56,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:36:57,077 INFO [optim.py:369] (1/2) 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 12:37:13,132 INFO [train.py:968] (1/2) Epoch 20, batch 28350, giga_loss[loss=0.2938, simple_loss=0.3621, pruned_loss=0.1128, over 28986.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.374, pruned_loss=0.1238, over 5652717.58 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3615, pruned_loss=0.1146, over 5697774.33 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1242, over 5647188.22 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:37:17,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:19,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:50,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:50,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:37:59,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:38:05,546 INFO [train.py:968] (1/2) Epoch 20, batch 28400, giga_loss[loss=0.3473, simple_loss=0.4072, pruned_loss=0.1437, over 27932.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3751, pruned_loss=0.1247, over 5653097.48 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1145, over 5700934.99 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3757, pruned_loss=0.1253, over 5645222.84 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:38:09,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 12:38:24,464 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 28450, giga_loss[loss=0.3007, simple_loss=0.3714, pruned_loss=0.1149, over 28714.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3752, pruned_loss=0.1232, over 5662220.29 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1145, over 5704855.57 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3764, pruned_loss=0.1241, over 5650962.57 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:39:00,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:39:02,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:39:31,992 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:39:51,799 INFO [train.py:968] (1/2) Epoch 20, batch 28500, giga_loss[loss=0.2767, simple_loss=0.3483, pruned_loss=0.1026, over 29043.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3747, pruned_loss=0.1231, over 5663663.61 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1145, over 5705292.92 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3761, pruned_loss=0.1239, over 5653400.41 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:40:28,788 INFO [optim.py:369] (1/2) 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 12:40:41,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-10 12:40:44,165 INFO [train.py:968] (1/2) Epoch 20, batch 28550, giga_loss[loss=0.2657, simple_loss=0.338, pruned_loss=0.09667, over 28909.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3741, pruned_loss=0.1235, over 5668392.81 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3613, pruned_loss=0.1146, over 5707320.81 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3751, pruned_loss=0.1241, over 5658133.45 frames. ], batch size: 112, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:40:53,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-10 12:40:59,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=895722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:41:02,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=895725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:41:37,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=895754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:41:44,506 INFO [train.py:968] (1/2) Epoch 20, batch 28600, giga_loss[loss=0.2853, simple_loss=0.3537, pruned_loss=0.1084, over 28641.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3731, pruned_loss=0.1232, over 5678741.20 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1145, over 5710811.52 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3743, pruned_loss=0.1241, over 5666453.52 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:42:24,203 INFO [optim.py:369] (1/2) 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 12:42:40,480 INFO [train.py:968] (1/2) Epoch 20, batch 28650, giga_loss[loss=0.2915, simple_loss=0.3522, pruned_loss=0.1154, over 28880.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3721, pruned_loss=0.123, over 5673882.26 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5703411.61 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3726, pruned_loss=0.1233, over 5669889.32 frames. ], batch size: 99, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:43:20,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895853.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 12:43:26,823 INFO [train.py:968] (1/2) Epoch 20, batch 28700, giga_loss[loss=0.3347, simple_loss=0.3927, pruned_loss=0.1383, over 28800.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3728, pruned_loss=0.1241, over 5677373.83 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5706085.01 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1244, over 5671312.41 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:43:36,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:43:47,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-10 12:44:01,571 INFO [optim.py:369] (1/2) 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 12:44:17,838 INFO [train.py:968] (1/2) Epoch 20, batch 28750, giga_loss[loss=0.3024, simple_loss=0.3747, pruned_loss=0.1151, over 28954.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1239, over 5652556.58 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5699304.32 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3722, pruned_loss=0.1242, over 5652392.17 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 12:44:29,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:44:33,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2381, 5.0749, 4.8750, 2.3868], device='cuda:1'), covar=tensor([0.0492, 0.0621, 0.0681, 0.1854], device='cuda:1'), in_proj_covar=tensor([0.1227, 0.1140, 0.0970, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 12:44:36,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=895929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:44:59,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=895953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:45:08,473 INFO [train.py:968] (1/2) Epoch 20, batch 28800, giga_loss[loss=0.2725, simple_loss=0.3519, pruned_loss=0.09654, over 29043.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3732, pruned_loss=0.1252, over 5650482.74 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3627, pruned_loss=0.1155, over 5699407.41 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3734, pruned_loss=0.1255, over 5649345.79 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:45:44,859 INFO [optim.py:369] (1/2) 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 12:45:57,314 INFO [train.py:968] (1/2) Epoch 20, batch 28850, libri_loss[loss=0.313, simple_loss=0.3868, pruned_loss=0.1196, over 29214.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.374, pruned_loss=0.1262, over 5656769.43 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3623, pruned_loss=0.1153, over 5703935.80 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.375, pruned_loss=0.127, over 5650305.53 frames. ], batch size: 101, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:45:59,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:01,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:30,923 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:34,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3498, 3.1152, 1.4758, 1.4488], device='cuda:1'), covar=tensor([0.0964, 0.0325, 0.0871, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0554, 0.0380, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 12:46:45,446 INFO [train.py:968] (1/2) Epoch 20, batch 28900, giga_loss[loss=0.3389, simple_loss=0.3841, pruned_loss=0.1469, over 23412.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3756, pruned_loss=0.1276, over 5642476.70 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3628, pruned_loss=0.1158, over 5698272.41 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3764, pruned_loss=0.1282, over 5640783.78 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:46:48,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:53,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:46:57,397 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:02,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:18,813 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:19,240 INFO [optim.py:369] (1/2) 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 12:47:26,285 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:32,031 INFO [train.py:968] (1/2) Epoch 20, batch 28950, giga_loss[loss=0.3111, simple_loss=0.3751, pruned_loss=0.1235, over 27926.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3754, pruned_loss=0.1275, over 5648477.50 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3632, pruned_loss=0.116, over 5700585.95 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.376, pruned_loss=0.1282, over 5643058.37 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:47:35,691 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:47:39,638 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 12:48:22,217 INFO [train.py:968] (1/2) Epoch 20, batch 29000, giga_loss[loss=0.2924, simple_loss=0.3552, pruned_loss=0.1147, over 28635.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3749, pruned_loss=0.1275, over 5642936.31 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1161, over 5692662.06 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3754, pruned_loss=0.128, over 5644495.86 frames. ], batch size: 71, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:48:28,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4740, 2.1785, 1.5805, 0.6267], device='cuda:1'), covar=tensor([0.5462, 0.2563, 0.4083, 0.6060], device='cuda:1'), in_proj_covar=tensor([0.1727, 0.1634, 0.1584, 0.1404], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 12:48:56,706 INFO [optim.py:369] (1/2) 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 12:49:10,031 INFO [train.py:968] (1/2) Epoch 20, batch 29050, giga_loss[loss=0.3263, simple_loss=0.3715, pruned_loss=0.1405, over 23430.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.3753, pruned_loss=0.1272, over 5638348.24 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3633, pruned_loss=0.116, over 5693966.32 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3761, pruned_loss=0.128, over 5636706.47 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:49:27,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=896228.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 12:49:59,177 INFO [train.py:968] (1/2) Epoch 20, batch 29100, giga_loss[loss=0.3022, simple_loss=0.3669, pruned_loss=0.1188, over 28967.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.375, pruned_loss=0.1261, over 5652845.44 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3632, pruned_loss=0.1159, over 5700462.39 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.3762, pruned_loss=0.1272, over 5644117.19 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:50:31,409 INFO [optim.py:369] (1/2) 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 12:50:32,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1792, 1.6365, 1.4362, 1.3012], device='cuda:1'), covar=tensor([0.0827, 0.0320, 0.0304, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0096, 0.0069, 0.0062, 0.0105], device='cuda:1')
+2023-03-10 12:50:43,612 INFO [train.py:968] (1/2) Epoch 20, batch 29150, giga_loss[loss=0.2936, simple_loss=0.364, pruned_loss=0.1116, over 28949.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3769, pruned_loss=0.1272, over 5653941.07 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3633, pruned_loss=0.1158, over 5695481.43 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1284, over 5650354.46 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:51:01,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=896328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:51:32,357 INFO [train.py:968] (1/2) Epoch 20, batch 29200, giga_loss[loss=0.3019, simple_loss=0.3654, pruned_loss=0.1192, over 28566.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3774, pruned_loss=0.1273, over 5666650.48 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5698751.36 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3785, pruned_loss=0.1285, over 5660462.70 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:51:40,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896371.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 12:51:42,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896374.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 12:52:00,709 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:52:09,911 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896403.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 12:52:17,307 INFO [train.py:968] (1/2) Epoch 20, batch 29250, giga_loss[loss=0.2868, simple_loss=0.3598, pruned_loss=0.1069, over 28914.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3774, pruned_loss=0.1272, over 5674335.69 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3634, pruned_loss=0.1157, over 5702102.43 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3784, pruned_loss=0.1284, over 5665864.81 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 12:52:45,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6230, 1.8543, 1.5400, 1.6789], device='cuda:1'), covar=tensor([0.2203, 0.2277, 0.2352, 0.2178], device='cuda:1'), in_proj_covar=tensor([0.1476, 0.1074, 0.1308, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 12:53:08,230 INFO [train.py:968] (1/2) Epoch 20, batch 29300, giga_loss[loss=0.3128, simple_loss=0.3818, pruned_loss=0.1219, over 28942.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.379, pruned_loss=0.1281, over 5673879.35 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3634, pruned_loss=0.1157, over 5705659.83 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3801, pruned_loss=0.1293, over 5663318.94 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:53:17,886 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:53:21,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:53:36,517 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:53:49,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5743, 1.6573, 1.7741, 1.3626], device='cuda:1'), covar=tensor([0.1785, 0.2484, 0.1448, 0.1697], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0700, 0.0933, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 12:53:50,287 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:53:58,746 INFO [train.py:968] (1/2) Epoch 20, batch 29350, libri_loss[loss=0.2787, simple_loss=0.3504, pruned_loss=0.1035, over 29746.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3786, pruned_loss=0.1267, over 5659774.79 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3633, pruned_loss=0.1156, over 5696314.94 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3799, pruned_loss=0.128, over 5658036.53 frames. ], batch size: 87, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:54:47,927 INFO [train.py:968] (1/2) Epoch 20, batch 29400, giga_loss[loss=0.3308, simple_loss=0.3848, pruned_loss=0.1385, over 28299.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3764, pruned_loss=0.1246, over 5666514.07 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3637, pruned_loss=0.1159, over 5696584.71 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3773, pruned_loss=0.1255, over 5664522.78 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:55:22,253 INFO [optim.py:369] (1/2) 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 12:55:36,469 INFO [train.py:968] (1/2) Epoch 20, batch 29450, giga_loss[loss=0.3347, simple_loss=0.373, pruned_loss=0.1481, over 23418.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3748, pruned_loss=0.1241, over 5655840.93 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3643, pruned_loss=0.1165, over 5701862.30 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3752, pruned_loss=0.1245, over 5648790.06 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:55:56,029 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:55:57,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:56:24,325 INFO [train.py:968] (1/2) Epoch 20, batch 29500, giga_loss[loss=0.3244, simple_loss=0.3759, pruned_loss=0.1365, over 27641.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3765, pruned_loss=0.1251, over 5667596.21 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3648, pruned_loss=0.1166, over 5706337.29 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3767, pruned_loss=0.1255, over 5656950.32 frames. ], batch size: 474, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:56:24,486 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:56:28,040 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=896664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:56:28,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3886, 2.0541, 1.5506, 0.6073], device='cuda:1'), covar=tensor([0.5089, 0.3002, 0.4296, 0.5949], device='cuda:1'), in_proj_covar=tensor([0.1735, 0.1644, 0.1592, 0.1411], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 12:57:01,469 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-10 12:57:01,551 INFO [optim.py:369] (1/2) 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 12:57:14,042 INFO [train.py:968] (1/2) Epoch 20, batch 29550, giga_loss[loss=0.2985, simple_loss=0.3722, pruned_loss=0.1124, over 29032.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3768, pruned_loss=0.1256, over 5667323.87 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1163, over 5710257.08 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3776, pruned_loss=0.1264, over 5654395.02 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:58:03,061 INFO [train.py:968] (1/2) Epoch 20, batch 29600, giga_loss[loss=0.2732, simple_loss=0.3423, pruned_loss=0.102, over 29006.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3755, pruned_loss=0.1257, over 5672664.09 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3636, pruned_loss=0.116, over 5714265.80 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5658237.15 frames. ], batch size: 106, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:58:11,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=896768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:58:38,901 INFO [optim.py:369] (1/2) 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 12:58:50,841 INFO [train.py:968] (1/2) Epoch 20, batch 29650, giga_loss[loss=0.3732, simple_loss=0.4182, pruned_loss=0.1641, over 28900.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3767, pruned_loss=0.1273, over 5656735.89 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3636, pruned_loss=0.116, over 5707901.66 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3781, pruned_loss=0.1283, over 5650487.60 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 12:58:52,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 12:59:42,682 INFO [train.py:968] (1/2) Epoch 20, batch 29700, giga_loss[loss=0.3026, simple_loss=0.3691, pruned_loss=0.1181, over 28933.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3777, pruned_loss=0.1281, over 5658606.08 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3639, pruned_loss=0.1163, over 5700544.55 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3787, pruned_loss=0.1287, over 5658499.23 frames. ], batch size: 227, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:00:24,159 INFO [optim.py:369] (1/2) 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 13:00:39,068 INFO [train.py:968] (1/2) Epoch 20, batch 29750, giga_loss[loss=0.3165, simple_loss=0.3804, pruned_loss=0.1263, over 28929.00 frames. ], tot_loss[loss=0.3171, simple_loss=0.3778, pruned_loss=0.1282, over 5642596.01 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3639, pruned_loss=0.1163, over 5701572.83 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3787, pruned_loss=0.1289, over 5641395.02 frames. ], batch size: 199, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:00:39,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=896911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:00:42,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=896914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:00:44,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2768, 1.3262, 3.8162, 3.3595], device='cuda:1'), covar=tensor([0.1656, 0.2737, 0.0441, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0643, 0.0954, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 13:01:09,105 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 29800, giga_loss[loss=0.3174, simple_loss=0.379, pruned_loss=0.1279, over 28854.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3764, pruned_loss=0.1261, over 5665340.60 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3641, pruned_loss=0.1164, over 5703333.96 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.377, pruned_loss=0.1266, over 5662439.05 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:01:48,137 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=896979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:02:03,258 INFO [optim.py:369] (1/2) 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 13:02:14,909 INFO [train.py:968] (1/2) Epoch 20, batch 29850, giga_loss[loss=0.3019, simple_loss=0.3679, pruned_loss=0.118, over 28279.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.376, pruned_loss=0.1253, over 5654683.54 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3646, pruned_loss=0.1165, over 5696993.77 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3765, pruned_loss=0.1259, over 5657016.77 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:02:42,207 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:03:05,961 INFO [train.py:968] (1/2) Epoch 20, batch 29900, giga_loss[loss=0.3044, simple_loss=0.3843, pruned_loss=0.1123, over 28915.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5652073.74 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3645, pruned_loss=0.1167, over 5689553.86 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3747, pruned_loss=0.1239, over 5660631.49 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:03:20,782 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 13:03:41,536 INFO [optim.py:369] (1/2) 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 13:03:53,687 INFO [train.py:968] (1/2) Epoch 20, batch 29950, giga_loss[loss=0.2598, simple_loss=0.3311, pruned_loss=0.09422, over 28331.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1232, over 5656826.95 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3643, pruned_loss=0.1166, over 5694651.26 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3739, pruned_loss=0.1239, over 5658173.23 frames. ], batch size: 65, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:04:14,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-03-10 13:04:42,938 INFO [train.py:968] (1/2) Epoch 20, batch 30000, giga_loss[loss=0.2793, simple_loss=0.3484, pruned_loss=0.1051, over 29012.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3732, pruned_loss=0.1242, over 5652954.34 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3644, pruned_loss=0.1167, over 5693154.24 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3739, pruned_loss=0.1247, over 5655133.29 frames. ], batch size: 128, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:04:42,939 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 13:04:50,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2575, 1.5096, 1.5006, 1.2028], device='cuda:1'), covar=tensor([0.3199, 0.2236, 0.1576, 0.2306], device='cuda:1'), in_proj_covar=tensor([0.1963, 0.1889, 0.1811, 0.1955], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 13:04:51,922 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 13:04:52,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:07,991 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:11,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:15,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:27,289 INFO [optim.py:369] (1/2) 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 13:05:39,103 INFO [train.py:968] (1/2) Epoch 20, batch 30050, giga_loss[loss=0.2735, simple_loss=0.3475, pruned_loss=0.09971, over 28916.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3711, pruned_loss=0.123, over 5655809.38 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.365, pruned_loss=0.1169, over 5695901.82 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1233, over 5654550.20 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:05:39,337 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:05:51,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-10 13:06:28,687 INFO [train.py:968] (1/2) Epoch 20, batch 30100, giga_loss[loss=0.2927, simple_loss=0.3525, pruned_loss=0.1164, over 28744.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3675, pruned_loss=0.1209, over 5665064.58 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3653, pruned_loss=0.1171, over 5689787.36 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3675, pruned_loss=0.1211, over 5669290.84 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:07:03,500 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 30150, giga_loss[loss=0.3104, simple_loss=0.3683, pruned_loss=0.1263, over 27915.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3668, pruned_loss=0.121, over 5670969.98 frames. ], libri_tot_loss[loss=0.3005, simple_loss=0.3659, pruned_loss=0.1175, over 5682162.70 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3663, pruned_loss=0.1209, over 5680241.55 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:07:31,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:07:54,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:07:58,756 INFO [train.py:968] (1/2) Epoch 20, batch 30200, giga_loss[loss=0.2951, simple_loss=0.3676, pruned_loss=0.1113, over 28946.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3652, pruned_loss=0.1196, over 5684634.50 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3652, pruned_loss=0.1171, over 5687319.61 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3654, pruned_loss=0.1201, over 5687223.73 frames. ], batch size: 174, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:08:00,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:08:31,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:08:38,457 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 30250, giga_loss[loss=0.2675, simple_loss=0.3456, pruned_loss=0.09471, over 28006.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3643, pruned_loss=0.1174, over 5679724.53 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.365, pruned_loss=0.1169, over 5690228.49 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3647, pruned_loss=0.118, over 5678913.36 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:09:22,380 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 30300, giga_loss[loss=0.3006, simple_loss=0.3752, pruned_loss=0.113, over 28623.00 frames. ], tot_loss[loss=0.295, simple_loss=0.362, pruned_loss=0.114, over 5671997.67 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.117, over 5685538.61 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3625, pruned_loss=0.1144, over 5674438.29 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:10:17,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-10 13:10:22,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:10:24,746 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 30350, giga_loss[loss=0.2901, simple_loss=0.3595, pruned_loss=0.1103, over 27934.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3592, pruned_loss=0.1114, over 5668972.43 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3644, pruned_loss=0.117, over 5691838.78 frames. ], giga_tot_loss[loss=0.2914, simple_loss=0.3598, pruned_loss=0.1115, over 5664113.51 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:10:48,880 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-10 13:10:52,142 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:10:59,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:11:02,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2186, 1.6360, 5.5357, 3.9341], device='cuda:1'), covar=tensor([0.1537, 0.2734, 0.0374, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0643, 0.0950, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 13:11:22,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2891, 1.4645, 1.2210, 1.5376], device='cuda:1'), covar=tensor([0.0807, 0.0343, 0.0374, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0118, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 13:11:26,025 INFO [train.py:968] (1/2) Epoch 20, batch 30400, giga_loss[loss=0.2458, simple_loss=0.3248, pruned_loss=0.08337, over 28565.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3541, pruned_loss=0.1072, over 5663457.75 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3633, pruned_loss=0.1165, over 5698121.19 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3553, pruned_loss=0.1075, over 5653214.89 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:12:04,003 INFO [optim.py:369] (1/2) 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 13:12:13,142 INFO [train.py:968] (1/2) Epoch 20, batch 30450, libri_loss[loss=0.328, simple_loss=0.3845, pruned_loss=0.1358, over 29371.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3511, pruned_loss=0.1042, over 5651882.00 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3625, pruned_loss=0.1162, over 5687589.27 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3524, pruned_loss=0.1043, over 5652424.43 frames. ], batch size: 92, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:12:22,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-10 13:12:37,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:12:45,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4411, 2.0934, 1.4580, 0.6936], device='cuda:1'), covar=tensor([0.5768, 0.2871, 0.4299, 0.6123], device='cuda:1'), in_proj_covar=tensor([0.1724, 0.1631, 0.1584, 0.1406], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 13:13:02,790 INFO [train.py:968] (1/2) Epoch 20, batch 30500, giga_loss[loss=0.2428, simple_loss=0.3265, pruned_loss=0.07952, over 28562.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.351, pruned_loss=0.1026, over 5642317.91 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3626, pruned_loss=0.1164, over 5683444.42 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3517, pruned_loss=0.1022, over 5644778.61 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:13:23,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2609, 1.6454, 1.0130, 1.2193], device='cuda:1'), covar=tensor([0.1123, 0.0602, 0.1430, 0.1302], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0441, 0.0510, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 13:13:27,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:13:30,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:13:49,642 INFO [optim.py:369] (1/2) 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 13:14:00,726 INFO [train.py:968] (1/2) Epoch 20, batch 30550, giga_loss[loss=0.2421, simple_loss=0.332, pruned_loss=0.07611, over 28744.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3525, pruned_loss=0.1033, over 5640050.70 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3628, pruned_loss=0.1167, over 5684545.20 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3528, pruned_loss=0.1027, over 5640743.67 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:14:01,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:14:51,428 INFO [train.py:968] (1/2) Epoch 20, batch 30600, giga_loss[loss=0.2535, simple_loss=0.3365, pruned_loss=0.08522, over 28760.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3507, pruned_loss=0.1023, over 5640808.97 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3622, pruned_loss=0.1165, over 5687702.60 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3512, pruned_loss=0.1015, over 5637165.70 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:14:58,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:15:17,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=897786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:15:31,098 INFO [optim.py:369] (1/2) 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 13:15:39,922 INFO [train.py:968] (1/2) Epoch 20, batch 30650, libri_loss[loss=0.3287, simple_loss=0.3727, pruned_loss=0.1423, over 29528.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3478, pruned_loss=0.1004, over 5640113.31 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3616, pruned_loss=0.1163, over 5691567.61 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3484, pruned_loss=0.09964, over 5632410.28 frames. ], batch size: 81, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:15:46,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=897816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:16:14,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6066, 1.7705, 1.7778, 1.5548], device='cuda:1'), covar=tensor([0.2477, 0.2043, 0.1811, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.1934, 0.1857, 0.1780, 0.1925], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 13:16:27,901 INFO [train.py:968] (1/2) Epoch 20, batch 30700, giga_loss[loss=0.2401, simple_loss=0.3269, pruned_loss=0.07661, over 28866.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3462, pruned_loss=0.09937, over 5649089.80 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3615, pruned_loss=0.1164, over 5697280.23 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3464, pruned_loss=0.09815, over 5636311.19 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:16:38,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-10 13:17:08,116 INFO [optim.py:369] (1/2) 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 13:17:16,336 INFO [train.py:968] (1/2) Epoch 20, batch 30750, giga_loss[loss=0.2472, simple_loss=0.3358, pruned_loss=0.07928, over 28565.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3463, pruned_loss=0.099, over 5653454.13 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3612, pruned_loss=0.1164, over 5700022.57 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3463, pruned_loss=0.09767, over 5639719.80 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:17:17,449 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:17:20,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:17:39,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-10 13:17:51,812 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:08,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=897959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:09,224 INFO [train.py:968] (1/2) Epoch 20, batch 30800, giga_loss[loss=0.234, simple_loss=0.32, pruned_loss=0.07403, over 28603.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3441, pruned_loss=0.09688, over 5656698.07 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3609, pruned_loss=0.1162, over 5702686.10 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3442, pruned_loss=0.09571, over 5642793.71 frames. ], batch size: 242, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:18:11,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=897962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:43,065 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=897991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:50,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6588, 2.7966, 2.4557, 2.4982], device='cuda:1'), covar=tensor([0.1547, 0.1779, 0.1749, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0740, 0.0703, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 13:18:50,461 INFO [optim.py:369] (1/2) 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 13:18:58,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3777, 3.2118, 1.4338, 1.6061], device='cuda:1'), covar=tensor([0.0984, 0.0294, 0.0992, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0552, 0.0380, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 13:18:58,786 INFO [train.py:968] (1/2) Epoch 20, batch 30850, giga_loss[loss=0.2411, simple_loss=0.322, pruned_loss=0.08013, over 28697.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3414, pruned_loss=0.0947, over 5651420.15 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3605, pruned_loss=0.116, over 5696391.26 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3413, pruned_loss=0.09317, over 5644686.64 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:18:59,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=898011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:18:59,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3359, 2.0309, 1.4595, 0.5755], device='cuda:1'), covar=tensor([0.4885, 0.3068, 0.4059, 0.5589], device='cuda:1'), in_proj_covar=tensor([0.1719, 0.1628, 0.1579, 0.1404], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 13:19:13,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4265, 2.1408, 1.5457, 0.6078], device='cuda:1'), covar=tensor([0.4357, 0.3046, 0.3969, 0.5417], device='cuda:1'), in_proj_covar=tensor([0.1717, 0.1626, 0.1578, 0.1403], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 13:19:37,128 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=898048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:19:48,610 INFO [train.py:968] (1/2) Epoch 20, batch 30900, giga_loss[loss=0.2872, simple_loss=0.3411, pruned_loss=0.1167, over 26664.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3386, pruned_loss=0.09331, over 5640335.07 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.36, pruned_loss=0.1159, over 5691303.91 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3383, pruned_loss=0.09159, over 5637647.05 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:19:53,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1555, 1.1759, 3.2054, 2.8978], device='cuda:1'), covar=tensor([0.1513, 0.2706, 0.0502, 0.1638], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0640, 0.0944, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 13:19:57,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7322, 1.9994, 1.4318, 1.4910], device='cuda:1'), covar=tensor([0.1015, 0.0563, 0.0984, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0443, 0.0514, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 13:20:20,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5412, 1.6315, 1.7462, 1.3817], device='cuda:1'), covar=tensor([0.1711, 0.2432, 0.1428, 0.1794], device='cuda:1'), in_proj_covar=tensor([0.0888, 0.0693, 0.0930, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 13:20:32,631 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 30950, giga_loss[loss=0.2629, simple_loss=0.3377, pruned_loss=0.09404, over 28624.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3363, pruned_loss=0.09219, over 5646159.77 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3599, pruned_loss=0.1158, over 5695112.41 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3357, pruned_loss=0.09047, over 5639662.43 frames. ], batch size: 242, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:21:25,658 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=898154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:21:28,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=898157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:21:31,269 INFO [train.py:968] (1/2) Epoch 20, batch 31000, giga_loss[loss=0.2456, simple_loss=0.3387, pruned_loss=0.07625, over 28981.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3345, pruned_loss=0.09168, over 5638999.08 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3593, pruned_loss=0.1154, over 5696110.31 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3341, pruned_loss=0.09009, over 5631231.99 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:21:31,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=898161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:22:01,026 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=898186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:22:04,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1283, 1.5334, 1.4965, 1.3133], device='cuda:1'), covar=tensor([0.1971, 0.1543, 0.2149, 0.1785], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0737, 0.0700, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 13:22:18,572 INFO [optim.py:369] (1/2) 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 13:22:26,155 INFO [train.py:968] (1/2) Epoch 20, batch 31050, giga_loss[loss=0.261, simple_loss=0.3466, pruned_loss=0.08769, over 28967.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3358, pruned_loss=0.09279, over 5631069.21 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3589, pruned_loss=0.1154, over 5699581.22 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3352, pruned_loss=0.09103, over 5620470.71 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:22:52,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-10 13:23:24,862 INFO [train.py:968] (1/2) Epoch 20, batch 31100, giga_loss[loss=0.2704, simple_loss=0.3552, pruned_loss=0.09279, over 28428.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3394, pruned_loss=0.09369, over 5643717.20 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3591, pruned_loss=0.1157, over 5704193.62 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3384, pruned_loss=0.09147, over 5629619.25 frames. ], batch size: 369, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:23:38,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1262, 3.3733, 1.3635, 1.3805], device='cuda:1'), covar=tensor([0.1090, 0.0312, 0.0995, 0.1428], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0551, 0.0381, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 13:24:10,935 INFO [optim.py:369] (1/2) 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 13:24:13,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=898304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:24:19,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=898307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:24:22,532 INFO [train.py:968] (1/2) Epoch 20, batch 31150, libri_loss[loss=0.2532, simple_loss=0.3179, pruned_loss=0.09422, over 29469.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3405, pruned_loss=0.09391, over 5656842.74 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3586, pruned_loss=0.1155, over 5704883.11 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3394, pruned_loss=0.09165, over 5643070.13 frames. ], batch size: 70, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:24:58,859 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=898336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:25:25,445 INFO [train.py:968] (1/2) Epoch 20, batch 31200, libri_loss[loss=0.2534, simple_loss=0.3154, pruned_loss=0.09568, over 29354.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3403, pruned_loss=0.09372, over 5669632.29 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3585, pruned_loss=0.1155, over 5707494.27 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.339, pruned_loss=0.09115, over 5654824.68 frames. ], batch size: 67, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:26:15,684 INFO [optim.py:369] (1/2) 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 13:26:26,776 INFO [train.py:968] (1/2) Epoch 20, batch 31250, giga_loss[loss=0.237, simple_loss=0.3276, pruned_loss=0.07325, over 28795.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3377, pruned_loss=0.09201, over 5660523.67 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3579, pruned_loss=0.1152, over 5706425.86 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3368, pruned_loss=0.08975, over 5648766.04 frames. ], batch size: 243, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:26:44,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=898423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:27:31,514 INFO [train.py:968] (1/2) Epoch 20, batch 31300, giga_loss[loss=0.2304, simple_loss=0.3211, pruned_loss=0.06982, over 29080.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3361, pruned_loss=0.08983, over 5655061.49 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3574, pruned_loss=0.1149, over 5698020.08 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3353, pruned_loss=0.08764, over 5651659.31 frames. ], batch size: 113, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:27:39,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9212, 3.7577, 3.5345, 1.9153], device='cuda:1'), covar=tensor([0.0657, 0.0818, 0.0816, 0.2440], device='cuda:1'), in_proj_covar=tensor([0.1195, 0.1111, 0.0940, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 13:28:06,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4580, 1.6888, 1.7101, 1.2749], device='cuda:1'), covar=tensor([0.1941, 0.2755, 0.1652, 0.1930], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0694, 0.0935, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 13:28:18,504 INFO [optim.py:369] (1/2) 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 13:28:27,840 INFO [train.py:968] (1/2) Epoch 20, batch 31350, giga_loss[loss=0.2324, simple_loss=0.307, pruned_loss=0.07893, over 29002.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3343, pruned_loss=0.08964, over 5664305.01 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3564, pruned_loss=0.1144, over 5701546.52 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3337, pruned_loss=0.08726, over 5656716.45 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:29:29,948 INFO [train.py:968] (1/2) Epoch 20, batch 31400, giga_loss[loss=0.2448, simple_loss=0.3288, pruned_loss=0.08042, over 28937.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.333, pruned_loss=0.0899, over 5657158.62 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3559, pruned_loss=0.1143, over 5697802.82 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3323, pruned_loss=0.08745, over 5653687.90 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:29:37,490 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=898566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:29:41,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=898569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:29:46,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5471, 4.5698, 1.8939, 1.7391], device='cuda:1'), covar=tensor([0.1009, 0.0363, 0.0902, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0553, 0.0381, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 13:30:04,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-10 13:30:14,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=898598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:30:21,401 INFO [optim.py:369] (1/2) 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 13:30:22,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 13:30:32,670 INFO [train.py:968] (1/2) Epoch 20, batch 31450, giga_loss[loss=0.2893, simple_loss=0.3682, pruned_loss=0.1052, over 28530.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3318, pruned_loss=0.08929, over 5654688.26 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.356, pruned_loss=0.1144, over 5689910.76 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3311, pruned_loss=0.0871, over 5658559.02 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:31:25,338 INFO [train.py:968] (1/2) Epoch 20, batch 31500, giga_loss[loss=0.2743, simple_loss=0.3662, pruned_loss=0.09119, over 28667.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3343, pruned_loss=0.09048, over 5651955.00 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3554, pruned_loss=0.1142, over 5688386.19 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3332, pruned_loss=0.08787, over 5655220.10 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:31:46,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4272, 3.7262, 1.4641, 1.6219], device='cuda:1'), covar=tensor([0.1019, 0.0315, 0.0961, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0551, 0.0380, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 13:32:20,767 INFO [optim.py:369] (1/2) 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 13:32:23,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-10 13:32:26,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3421, 1.2568, 3.9895, 3.2281], device='cuda:1'), covar=tensor([0.1576, 0.2897, 0.0412, 0.1493], device='cuda:1'), in_proj_covar=tensor([0.0744, 0.0639, 0.0939, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 13:32:32,645 INFO [train.py:968] (1/2) Epoch 20, batch 31550, giga_loss[loss=0.2186, simple_loss=0.31, pruned_loss=0.06361, over 28920.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3362, pruned_loss=0.09021, over 5659928.28 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3558, pruned_loss=0.1145, over 5691314.67 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3347, pruned_loss=0.08753, over 5659527.00 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:33:14,552 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7874, 1.9478, 1.6551, 1.9930], device='cuda:1'), covar=tensor([0.2774, 0.2865, 0.3102, 0.2644], device='cuda:1'), in_proj_covar=tensor([0.1487, 0.1075, 0.1321, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 13:33:31,927 INFO [train.py:968] (1/2) Epoch 20, batch 31600, giga_loss[loss=0.2153, simple_loss=0.3023, pruned_loss=0.06415, over 28648.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3327, pruned_loss=0.08821, over 5669824.76 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.355, pruned_loss=0.1142, over 5695975.99 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3314, pruned_loss=0.08542, over 5664241.91 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:33:43,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-10 13:33:59,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0016, 5.7714, 5.5186, 2.7830], device='cuda:1'), covar=tensor([0.0523, 0.0718, 0.0849, 0.1813], device='cuda:1'), in_proj_covar=tensor([0.1193, 0.1110, 0.0940, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 13:34:35,149 INFO [optim.py:369] (1/2) 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 13:34:45,590 INFO [train.py:968] (1/2) Epoch 20, batch 31650, giga_loss[loss=0.3006, simple_loss=0.3589, pruned_loss=0.1211, over 27705.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3333, pruned_loss=0.08887, over 5674956.66 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3543, pruned_loss=0.1137, over 5699765.21 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3325, pruned_loss=0.08645, over 5666345.48 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 13:35:52,282 INFO [train.py:968] (1/2) Epoch 20, batch 31700, giga_loss[loss=0.3314, simple_loss=0.3904, pruned_loss=0.1362, over 26869.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3366, pruned_loss=0.08971, over 5668284.55 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3543, pruned_loss=0.1138, over 5698662.19 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3357, pruned_loss=0.08741, over 5661940.30 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:36:30,189 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=898891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:36:35,288 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=898894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:36:46,459 INFO [optim.py:369] (1/2) 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 13:36:55,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4682, 1.5847, 1.1473, 1.2300], device='cuda:1'), covar=tensor([0.0978, 0.0585, 0.1057, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0514, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 13:36:56,569 INFO [train.py:968] (1/2) Epoch 20, batch 31750, giga_loss[loss=0.2442, simple_loss=0.3382, pruned_loss=0.07512, over 28216.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3393, pruned_loss=0.08866, over 5666108.97 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3537, pruned_loss=0.1135, over 5702385.95 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3387, pruned_loss=0.08647, over 5657039.46 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:37:00,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3706, 1.5538, 1.5008, 1.3246], device='cuda:1'), covar=tensor([0.2299, 0.2113, 0.1673, 0.2183], device='cuda:1'), in_proj_covar=tensor([0.1916, 0.1834, 0.1753, 0.1899], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 13:38:01,060 INFO [train.py:968] (1/2) Epoch 20, batch 31800, giga_loss[loss=0.2417, simple_loss=0.3347, pruned_loss=0.07434, over 28426.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3393, pruned_loss=0.08757, over 5661052.78 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3531, pruned_loss=0.1132, over 5705777.55 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3391, pruned_loss=0.08568, over 5650004.83 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:38:48,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-10 13:38:55,289 INFO [optim.py:369] (1/2) 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 13:39:01,662 INFO [train.py:968] (1/2) Epoch 20, batch 31850, giga_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08891, over 28989.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3387, pruned_loss=0.08691, over 5652738.75 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3531, pruned_loss=0.1133, over 5698649.68 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3383, pruned_loss=0.08479, over 5649296.75 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:39:13,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7821, 3.5905, 3.4275, 1.7229], device='cuda:1'), covar=tensor([0.0729, 0.0872, 0.0816, 0.2249], device='cuda:1'), in_proj_covar=tensor([0.1192, 0.1107, 0.0936, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 13:39:24,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:39:58,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:40:02,829 INFO [train.py:968] (1/2) Epoch 20, batch 31900, giga_loss[loss=0.2529, simple_loss=0.3417, pruned_loss=0.082, over 28401.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3398, pruned_loss=0.08854, over 5654715.05 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3526, pruned_loss=0.1131, over 5700363.61 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3395, pruned_loss=0.08616, over 5648854.37 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:41:04,967 INFO [optim.py:369] (1/2) 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 13:41:13,579 INFO [train.py:968] (1/2) Epoch 20, batch 31950, giga_loss[loss=0.259, simple_loss=0.3342, pruned_loss=0.09196, over 27697.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.34, pruned_loss=0.09009, over 5655491.58 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.353, pruned_loss=0.1135, over 5694583.69 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3391, pruned_loss=0.0873, over 5655593.54 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:41:30,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3764, 1.8267, 1.5316, 1.4782], device='cuda:1'), covar=tensor([0.0760, 0.0306, 0.0321, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 13:42:37,676 INFO [train.py:968] (1/2) Epoch 20, batch 32000, giga_loss[loss=0.2522, simple_loss=0.3312, pruned_loss=0.08655, over 27482.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3409, pruned_loss=0.09103, over 5664362.42 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.353, pruned_loss=0.1135, over 5696700.58 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3401, pruned_loss=0.08866, over 5662308.70 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:42:47,927 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:43:44,017 INFO [optim.py:369] (1/2) 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 13:43:51,369 INFO [train.py:968] (1/2) Epoch 20, batch 32050, giga_loss[loss=0.2188, simple_loss=0.305, pruned_loss=0.06626, over 28842.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.336, pruned_loss=0.0882, over 5668478.88 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3526, pruned_loss=0.1133, over 5698737.65 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3355, pruned_loss=0.08615, over 5664660.07 frames. ], batch size: 119, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:44:59,402 INFO [train.py:968] (1/2) Epoch 20, batch 32100, giga_loss[loss=0.2425, simple_loss=0.3228, pruned_loss=0.0811, over 27727.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3335, pruned_loss=0.08677, over 5662786.72 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3524, pruned_loss=0.1133, over 5692718.81 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3329, pruned_loss=0.08466, over 5664548.39 frames. ], batch size: 474, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:45:07,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:45:11,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:45:57,278 INFO [optim.py:369] (1/2) 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 13:46:00,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 13:46:05,599 INFO [train.py:968] (1/2) Epoch 20, batch 32150, giga_loss[loss=0.3042, simple_loss=0.3752, pruned_loss=0.1165, over 28859.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3324, pruned_loss=0.08681, over 5665165.81 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3519, pruned_loss=0.1131, over 5696150.98 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3319, pruned_loss=0.08466, over 5662731.93 frames. ], batch size: 213, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:46:53,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4308, 1.5634, 1.6291, 1.2508], device='cuda:1'), covar=tensor([0.1788, 0.2569, 0.1501, 0.1793], device='cuda:1'), in_proj_covar=tensor([0.0891, 0.0691, 0.0935, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 13:47:13,861 INFO [train.py:968] (1/2) Epoch 20, batch 32200, giga_loss[loss=0.2903, simple_loss=0.3679, pruned_loss=0.1063, over 29042.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3375, pruned_loss=0.08942, over 5669609.90 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3521, pruned_loss=0.1132, over 5697978.55 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3368, pruned_loss=0.08741, over 5665890.24 frames. ], batch size: 155, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:48:02,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 13:48:07,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4034, 1.5565, 1.3839, 1.5743], device='cuda:1'), covar=tensor([0.0718, 0.0406, 0.0344, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 13:48:07,584 INFO [optim.py:369] (1/2) 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 13:48:11,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:13,514 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:14,821 INFO [train.py:968] (1/2) Epoch 20, batch 32250, giga_loss[loss=0.245, simple_loss=0.3165, pruned_loss=0.08672, over 28970.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3372, pruned_loss=0.08998, over 5671037.50 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3519, pruned_loss=0.1131, over 5702116.14 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3365, pruned_loss=0.08811, over 5663957.30 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:48:17,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:17,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:22,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:46,720 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 13:48:47,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:50,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=899435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:48:52,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3711, 1.4206, 1.3245, 1.5335], device='cuda:1'), covar=tensor([0.0739, 0.0342, 0.0337, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 13:48:58,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:49:04,300 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:49:19,460 INFO [train.py:968] (1/2) Epoch 20, batch 32300, giga_loss[loss=0.2476, simple_loss=0.3284, pruned_loss=0.08342, over 28554.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3361, pruned_loss=0.09061, over 5673681.25 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3516, pruned_loss=0.113, over 5707024.12 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.08863, over 5662672.40 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:49:37,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0238, 3.1190, 1.9553, 0.9427], device='cuda:1'), covar=tensor([0.7166, 0.3182, 0.3994, 0.6839], device='cuda:1'), in_proj_covar=tensor([0.1715, 0.1620, 0.1579, 0.1402], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 13:50:19,064 INFO [optim.py:369] (1/2) 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 13:50:26,158 INFO [train.py:968] (1/2) Epoch 20, batch 32350, giga_loss[loss=0.251, simple_loss=0.3386, pruned_loss=0.0817, over 29054.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3355, pruned_loss=0.09024, over 5671906.78 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3512, pruned_loss=0.1128, over 5707328.22 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3351, pruned_loss=0.08845, over 5662113.03 frames. ], batch size: 285, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 13:51:03,679 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:51:18,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:51:30,534 INFO [train.py:968] (1/2) Epoch 20, batch 32400, giga_loss[loss=0.2439, simple_loss=0.3296, pruned_loss=0.07913, over 28666.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.337, pruned_loss=0.09092, over 5673672.08 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3504, pruned_loss=0.1126, over 5712816.91 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3367, pruned_loss=0.08871, over 5659241.35 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:51:51,410 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:51:54,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:52:00,384 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:52:37,902 INFO [optim.py:369] (1/2) 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 13:52:39,646 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:52:43,653 INFO [train.py:968] (1/2) Epoch 20, batch 32450, giga_loss[loss=0.2479, simple_loss=0.3377, pruned_loss=0.0791, over 29017.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3383, pruned_loss=0.09054, over 5673560.79 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3502, pruned_loss=0.1125, over 5704579.64 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.338, pruned_loss=0.08839, over 5668524.46 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:53:45,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2664, 1.4981, 1.4656, 1.2193], device='cuda:1'), covar=tensor([0.2534, 0.2217, 0.1323, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.1909, 0.1821, 0.1747, 0.1890], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 13:54:00,464 INFO [train.py:968] (1/2) Epoch 20, batch 32500, giga_loss[loss=0.2676, simple_loss=0.3428, pruned_loss=0.09615, over 28100.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.337, pruned_loss=0.09015, over 5664092.36 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3498, pruned_loss=0.1124, over 5698864.90 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3368, pruned_loss=0.08783, over 5664651.62 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:54:33,460 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:54:36,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:55:00,718 INFO [optim.py:369] (1/2) 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 13:55:01,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5677, 1.7941, 1.7116, 1.4621], device='cuda:1'), covar=tensor([0.2696, 0.2104, 0.1791, 0.2230], device='cuda:1'), in_proj_covar=tensor([0.1912, 0.1821, 0.1746, 0.1890], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 13:55:05,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-10 13:55:07,006 INFO [train.py:968] (1/2) Epoch 20, batch 32550, giga_loss[loss=0.2305, simple_loss=0.3126, pruned_loss=0.07421, over 28421.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3327, pruned_loss=0.08888, over 5668712.64 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3493, pruned_loss=0.1121, over 5702735.25 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3327, pruned_loss=0.08686, over 5665193.04 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:55:14,978 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:55:32,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=2.08 vs. limit=2.0
+2023-03-10 13:56:10,870 INFO [train.py:968] (1/2) Epoch 20, batch 32600, giga_loss[loss=0.2445, simple_loss=0.3208, pruned_loss=0.08414, over 27863.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3277, pruned_loss=0.08691, over 5646831.07 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3493, pruned_loss=0.1123, over 5673376.49 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.327, pruned_loss=0.08429, over 5667178.44 frames. ], batch size: 476, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:56:44,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3221, 1.2697, 1.2144, 1.4713], device='cuda:1'), covar=tensor([0.0760, 0.0410, 0.0357, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 13:57:06,083 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=899810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 13:57:12,356 INFO [train.py:968] (1/2) Epoch 20, batch 32650, giga_loss[loss=0.2137, simple_loss=0.305, pruned_loss=0.06118, over 28876.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3281, pruned_loss=0.08749, over 5645020.81 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3487, pruned_loss=0.112, over 5677559.85 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3275, pruned_loss=0.0849, over 5656663.34 frames. ], batch size: 145, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:58:05,678 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-10 13:58:11,585 INFO [train.py:968] (1/2) Epoch 20, batch 32700, giga_loss[loss=0.2627, simple_loss=0.3295, pruned_loss=0.09792, over 27006.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3298, pruned_loss=0.08871, over 5648348.49 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3482, pruned_loss=0.1117, over 5684401.14 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3292, pruned_loss=0.08623, over 5650725.13 frames. ], batch size: 555, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:58:28,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5900, 2.0588, 1.9068, 1.8450], device='cuda:1'), covar=tensor([0.1922, 0.1894, 0.1978, 0.1805], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0729, 0.0692, 0.0663], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 13:58:52,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5447, 1.6868, 1.6463, 1.5263], device='cuda:1'), covar=tensor([0.2442, 0.2213, 0.1642, 0.1939], device='cuda:1'), in_proj_covar=tensor([0.1905, 0.1813, 0.1741, 0.1885], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 13:59:05,657 INFO [optim.py:369] (1/2) 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 13:59:11,142 INFO [train.py:968] (1/2) Epoch 20, batch 32750, giga_loss[loss=0.2188, simple_loss=0.2997, pruned_loss=0.06896, over 27570.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3272, pruned_loss=0.08652, over 5654597.57 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3478, pruned_loss=0.1115, over 5688565.21 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3265, pruned_loss=0.08408, over 5651755.20 frames. ], batch size: 472, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 13:59:13,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5557, 1.8269, 1.4700, 1.5656], device='cuda:1'), covar=tensor([0.2713, 0.2553, 0.3000, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.1482, 0.1070, 0.1315, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 14:00:08,348 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=899953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:00:12,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=899956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:00:17,854 INFO [train.py:968] (1/2) Epoch 20, batch 32800, giga_loss[loss=0.2318, simple_loss=0.3078, pruned_loss=0.07795, over 29016.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3261, pruned_loss=0.08479, over 5662489.60 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3479, pruned_loss=0.1115, over 5691020.24 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3253, pruned_loss=0.08269, over 5658005.86 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:00:55,178 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=899985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:01:13,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3046, 3.3475, 1.5768, 1.4781], device='cuda:1'), covar=tensor([0.1009, 0.0354, 0.0911, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0546, 0.0379, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 14:01:23,777 INFO [optim.py:369] (1/2) 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 14:01:32,111 INFO [train.py:968] (1/2) Epoch 20, batch 32850, giga_loss[loss=0.2214, simple_loss=0.2862, pruned_loss=0.07831, over 24292.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3255, pruned_loss=0.08491, over 5656118.46 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3478, pruned_loss=0.1114, over 5689622.45 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3248, pruned_loss=0.08309, over 5653734.26 frames. ], batch size: 705, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:02:08,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6399, 2.0028, 1.8868, 1.6074], device='cuda:1'), covar=tensor([0.2982, 0.2103, 0.2240, 0.2427], device='cuda:1'), in_proj_covar=tensor([0.1903, 0.1812, 0.1742, 0.1886], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 14:02:12,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:02:41,905 INFO [train.py:968] (1/2) Epoch 20, batch 32900, giga_loss[loss=0.2093, simple_loss=0.298, pruned_loss=0.06028, over 28432.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3249, pruned_loss=0.08378, over 5655135.23 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3478, pruned_loss=0.1113, over 5689843.30 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.324, pruned_loss=0.08191, over 5652158.25 frames. ], batch size: 60, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:03:47,257 INFO [optim.py:369] (1/2) 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 14:03:51,603 INFO [train.py:968] (1/2) Epoch 20, batch 32950, giga_loss[loss=0.2607, simple_loss=0.3416, pruned_loss=0.08992, over 28801.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3244, pruned_loss=0.08352, over 5658813.09 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3474, pruned_loss=0.1111, over 5692005.79 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3238, pruned_loss=0.082, over 5654358.90 frames. ], batch size: 243, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:04:09,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6549, 1.6126, 1.3068, 1.2555], device='cuda:1'), covar=tensor([0.0655, 0.0378, 0.0788, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0442, 0.0514, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 14:04:23,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 14:04:49,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:04:50,780 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:04:56,172 INFO [train.py:968] (1/2) Epoch 20, batch 33000, giga_loss[loss=0.2409, simple_loss=0.3185, pruned_loss=0.08166, over 28347.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3265, pruned_loss=0.08558, over 5664545.85 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3477, pruned_loss=0.1112, over 5696518.74 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3253, pruned_loss=0.0836, over 5655987.07 frames. ], batch size: 368, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:04:56,173 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 14:05:02,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2691, 1.8188, 1.3797, 0.4192], device='cuda:1'), covar=tensor([0.4810, 0.3509, 0.5048, 0.6436], device='cuda:1'), in_proj_covar=tensor([0.1717, 0.1622, 0.1581, 0.1405], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 14:05:05,151 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 14:05:58,880 INFO [optim.py:369] (1/2) 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 14:06:05,574 INFO [train.py:968] (1/2) Epoch 20, batch 33050, giga_loss[loss=0.2219, simple_loss=0.3177, pruned_loss=0.06305, over 28547.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3264, pruned_loss=0.08504, over 5654591.41 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3479, pruned_loss=0.1114, over 5691228.20 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3247, pruned_loss=0.08278, over 5652028.22 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:06:12,020 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 14:06:18,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4497, 1.6640, 1.6001, 1.4892], device='cuda:1'), covar=tensor([0.2708, 0.2323, 0.1865, 0.2086], device='cuda:1'), in_proj_covar=tensor([0.1910, 0.1817, 0.1746, 0.1892], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 14:07:03,752 INFO [train.py:968] (1/2) Epoch 20, batch 33100, giga_loss[loss=0.2484, simple_loss=0.3359, pruned_loss=0.08046, over 28901.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3289, pruned_loss=0.08509, over 5662944.42 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3474, pruned_loss=0.111, over 5693815.68 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3275, pruned_loss=0.08289, over 5657410.05 frames. ], batch size: 284, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:07:35,662 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:07:59,371 INFO [optim.py:369] (1/2) 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 14:08:05,082 INFO [train.py:968] (1/2) Epoch 20, batch 33150, giga_loss[loss=0.2436, simple_loss=0.3338, pruned_loss=0.07672, over 28689.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3311, pruned_loss=0.08593, over 5662409.31 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3468, pruned_loss=0.1108, over 5696858.99 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3301, pruned_loss=0.08392, over 5654740.84 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:09:11,468 INFO [train.py:968] (1/2) Epoch 20, batch 33200, giga_loss[loss=0.2493, simple_loss=0.3278, pruned_loss=0.08543, over 28958.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3309, pruned_loss=0.08594, over 5646885.10 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3462, pruned_loss=0.1104, over 5688997.98 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3305, pruned_loss=0.08425, over 5646458.80 frames. ], batch size: 186, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:09:27,459 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.82 vs. limit=5.0
+2023-03-10 14:10:08,038 INFO [optim.py:369] (1/2) 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 14:10:12,234 INFO [train.py:968] (1/2) Epoch 20, batch 33250, giga_loss[loss=0.2307, simple_loss=0.3201, pruned_loss=0.07068, over 28541.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3323, pruned_loss=0.08767, over 5647564.02 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3461, pruned_loss=0.1104, over 5685012.14 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3315, pruned_loss=0.08533, over 5648704.73 frames. ], batch size: 60, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:10:16,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:10:37,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6646, 2.1174, 2.0058, 1.6327], device='cuda:1'), covar=tensor([0.3412, 0.2170, 0.2318, 0.2618], device='cuda:1'), in_proj_covar=tensor([0.1906, 0.1815, 0.1741, 0.1892], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 14:11:11,704 INFO [train.py:968] (1/2) Epoch 20, batch 33300, giga_loss[loss=0.253, simple_loss=0.338, pruned_loss=0.08402, over 28690.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3303, pruned_loss=0.08619, over 5648284.52 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3461, pruned_loss=0.1104, over 5678202.89 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3294, pruned_loss=0.08391, over 5654724.28 frames. ], batch size: 262, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:11:43,762 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:12:12,007 INFO [optim.py:369] (1/2) 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 14:12:14,124 INFO [train.py:968] (1/2) Epoch 20, batch 33350, libri_loss[loss=0.2352, simple_loss=0.306, pruned_loss=0.08226, over 29570.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3292, pruned_loss=0.08575, over 5655299.44 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3461, pruned_loss=0.1104, over 5680375.51 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3281, pruned_loss=0.08341, over 5657839.42 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:12:31,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:12:32,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:13:08,627 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:13:11,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:13:11,977 INFO [train.py:968] (1/2) Epoch 20, batch 33400, giga_loss[loss=0.2447, simple_loss=0.332, pruned_loss=0.07868, over 29024.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08527, over 5655667.42 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3459, pruned_loss=0.1103, over 5676739.06 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3263, pruned_loss=0.08294, over 5660440.24 frames. ], batch size: 285, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:13:45,746 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=900589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:14:12,126 INFO [optim.py:369] (1/2) 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 14:14:16,669 INFO [train.py:968] (1/2) Epoch 20, batch 33450, giga_loss[loss=0.2393, simple_loss=0.333, pruned_loss=0.07284, over 28644.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3312, pruned_loss=0.08721, over 5660650.54 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.346, pruned_loss=0.1104, over 5679220.27 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3299, pruned_loss=0.08492, over 5662084.73 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:14:52,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:05,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-10 14:15:12,390 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-10 14:15:20,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 33500, giga_loss[loss=0.2304, simple_loss=0.3132, pruned_loss=0.07382, over 28984.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3315, pruned_loss=0.08721, over 5659911.58 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3458, pruned_loss=0.1103, over 5673509.77 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3306, pruned_loss=0.08516, over 5665222.00 frames. ], batch size: 136, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:15:33,049 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:37,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:38,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:40,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:42,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:15:55,641 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:16:18,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5024, 1.7796, 1.7388, 1.3285], device='cuda:1'), covar=tensor([0.1748, 0.2569, 0.1522, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.0886, 0.0688, 0.0929, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 14:16:20,423 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=900704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:16:21,271 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 33550, giga_loss[loss=0.2836, simple_loss=0.3584, pruned_loss=0.1044, over 28996.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3326, pruned_loss=0.0881, over 5654707.16 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3458, pruned_loss=0.1103, over 5667096.77 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3317, pruned_loss=0.08604, over 5663779.73 frames. ], batch size: 285, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:17:10,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900738.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 14:17:22,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:17:25,472 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:17:34,671 INFO [train.py:968] (1/2) Epoch 20, batch 33600, giga_loss[loss=0.2437, simple_loss=0.3372, pruned_loss=0.07512, over 28591.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3358, pruned_loss=0.08974, over 5656883.15 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3455, pruned_loss=0.1102, over 5671949.50 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.335, pruned_loss=0.08764, over 5659320.21 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:18:19,141 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=900803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:18:23,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=900806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:18:25,691 INFO [optim.py:369] (1/2) 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 14:18:28,023 INFO [train.py:968] (1/2) Epoch 20, batch 33650, giga_loss[loss=0.2295, simple_loss=0.3163, pruned_loss=0.07129, over 27744.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3371, pruned_loss=0.08957, over 5661555.66 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3453, pruned_loss=0.1099, over 5679770.96 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3362, pruned_loss=0.0873, over 5655850.47 frames. ], batch size: 474, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:19:05,877 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=900835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:19:21,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.52 vs. limit=5.0
+2023-03-10 14:19:44,137 INFO [train.py:968] (1/2) Epoch 20, batch 33700, giga_loss[loss=0.2224, simple_loss=0.3065, pruned_loss=0.06918, over 28659.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3358, pruned_loss=0.08875, over 5662301.68 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.345, pruned_loss=0.1098, over 5680749.11 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3354, pruned_loss=0.08702, over 5656824.31 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:19:44,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=900861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:19:58,019 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:20:48,772 INFO [train.py:968] (1/2) Epoch 20, batch 33750, giga_loss[loss=0.2361, simple_loss=0.3167, pruned_loss=0.07777, over 28618.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3347, pruned_loss=0.08875, over 5670294.23 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3453, pruned_loss=0.1099, over 5685364.24 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3338, pruned_loss=0.08668, over 5661025.97 frames. ], batch size: 85, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:21:52,451 INFO [train.py:968] (1/2) Epoch 20, batch 33800, giga_loss[loss=0.2371, simple_loss=0.3237, pruned_loss=0.07529, over 28834.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3347, pruned_loss=0.0891, over 5667917.18 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.345, pruned_loss=0.1097, over 5691104.50 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.334, pruned_loss=0.08701, over 5654718.72 frames. ], batch size: 243, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:22:13,172 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=900976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:22:55,707 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:22:58,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:22:59,114 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 33850, giga_loss[loss=0.244, simple_loss=0.3262, pruned_loss=0.08095, over 28694.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.335, pruned_loss=0.09015, over 5659070.99 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3449, pruned_loss=0.1097, over 5693056.68 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3344, pruned_loss=0.08837, over 5646871.43 frames. ], batch size: 307, lr: 1.59e-03, grad_scale: 8.0
+2023-03-10 14:23:07,941 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:23:52,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:24:10,271 INFO [train.py:968] (1/2) Epoch 20, batch 33900, giga_loss[loss=0.2563, simple_loss=0.3377, pruned_loss=0.0874, over 28892.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.332, pruned_loss=0.08908, over 5649515.59 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3448, pruned_loss=0.1096, over 5686137.05 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3314, pruned_loss=0.08741, over 5645960.55 frames. ], batch size: 164, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 14:24:12,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:24:43,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-10 14:24:58,189 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.21 vs. limit=5.0
+2023-03-10 14:25:08,595 INFO [optim.py:369] (1/2) 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 14:25:10,511 INFO [train.py:968] (1/2) Epoch 20, batch 33950, giga_loss[loss=0.2459, simple_loss=0.334, pruned_loss=0.0789, over 27985.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3322, pruned_loss=0.08916, over 5650448.45 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3447, pruned_loss=0.1099, over 5692893.16 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3315, pruned_loss=0.08706, over 5640394.95 frames. ], batch size: 412, lr: 1.59e-03, grad_scale: 2.0
+2023-03-10 14:25:14,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 14:25:26,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:25:29,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:04,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:08,057 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 34000, libri_loss[loss=0.2642, simple_loss=0.3294, pruned_loss=0.09947, over 29566.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.331, pruned_loss=0.08758, over 5674550.44 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3445, pruned_loss=0.1096, over 5702491.20 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3301, pruned_loss=0.08517, over 5656008.90 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:26:39,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:41,056 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:42,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:26:55,462 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=901201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:01,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:05,506 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:27:07,398 INFO [train.py:968] (1/2) Epoch 20, batch 34050, giga_loss[loss=0.2494, simple_loss=0.3441, pruned_loss=0.07732, over 28595.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3325, pruned_loss=0.08629, over 5684860.58 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3442, pruned_loss=0.1094, over 5706095.64 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3318, pruned_loss=0.08415, over 5666458.44 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:27:15,095 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:36,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:27:48,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:00,531 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:28:03,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:28:04,780 INFO [train.py:968] (1/2) Epoch 20, batch 34100, giga_loss[loss=0.2066, simple_loss=0.3056, pruned_loss=0.05381, over 28545.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3346, pruned_loss=0.08668, over 5682365.75 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3443, pruned_loss=0.1094, over 5712721.51 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3336, pruned_loss=0.08412, over 5660675.43 frames. ], batch size: 78, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:28:09,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:12,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:13,055 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:16,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:32,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 14:28:34,127 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901288.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:28:47,954 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:28:52,766 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:29:04,714 INFO [train.py:968] (1/2) Epoch 20, batch 34150, giga_loss[loss=0.2437, simple_loss=0.3296, pruned_loss=0.07886, over 28490.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3336, pruned_loss=0.0855, over 5678127.57 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3442, pruned_loss=0.1092, over 5714623.83 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3328, pruned_loss=0.08324, over 5658714.46 frames. ], batch size: 336, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:30:02,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:30:15,602 INFO [train.py:968] (1/2) Epoch 20, batch 34200, giga_loss[loss=0.2706, simple_loss=0.3491, pruned_loss=0.09605, over 28801.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.334, pruned_loss=0.08581, over 5680249.76 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3442, pruned_loss=0.1093, over 5716091.26 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.333, pruned_loss=0.0834, over 5662534.99 frames. ], batch size: 243, lr: 1.59e-03, grad_scale: 4.0
+2023-03-10 14:30:57,630 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:31:00,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:31:03,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-10 14:31:16,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9536, 1.9371, 1.3910, 1.6416], device='cuda:1'), covar=tensor([0.0945, 0.0676, 0.1005, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0443, 0.0517, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 14:31:25,302 INFO [optim.py:369] (1/2) 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 14:31:25,315 INFO [train.py:968] (1/2) Epoch 20, batch 34250, giga_loss[loss=0.2651, simple_loss=0.3491, pruned_loss=0.09059, over 28484.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3324, pruned_loss=0.08497, over 5671554.20 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3436, pruned_loss=0.1091, over 5711608.06 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3319, pruned_loss=0.08257, over 5660801.86 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:31:37,973 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:32:32,992 INFO [train.py:968] (1/2) Epoch 20, batch 34300, giga_loss[loss=0.2425, simple_loss=0.3325, pruned_loss=0.0763, over 28096.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.333, pruned_loss=0.08505, over 5661788.44 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3438, pruned_loss=0.1092, over 5706656.70 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3321, pruned_loss=0.08225, over 5657072.67 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:32:46,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5444, 1.6774, 1.7884, 1.3411], device='cuda:1'), covar=tensor([0.1912, 0.2645, 0.1574, 0.1925], device='cuda:1'), in_proj_covar=tensor([0.0885, 0.0686, 0.0930, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0015, 0.0014], device='cuda:1')
+2023-03-10 14:33:26,051 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 34350, giga_loss[loss=0.2602, simple_loss=0.3261, pruned_loss=0.09719, over 24384.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3336, pruned_loss=0.08501, over 5654757.97 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3434, pruned_loss=0.109, over 5709471.95 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3331, pruned_loss=0.08267, over 5647935.38 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:34:08,077 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:34:26,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 14:34:34,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-10 14:34:51,365 INFO [train.py:968] (1/2) Epoch 20, batch 34400, giga_loss[loss=0.2425, simple_loss=0.3352, pruned_loss=0.07494, over 29148.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3371, pruned_loss=0.0865, over 5658853.40 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3436, pruned_loss=0.1091, over 5700478.98 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3363, pruned_loss=0.08394, over 5660115.62 frames. ], batch size: 113, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:35:18,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=901576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:36:02,492 INFO [optim.py:369] (1/2) 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 14:36:02,504 INFO [train.py:968] (1/2) Epoch 20, batch 34450, giga_loss[loss=0.2652, simple_loss=0.3397, pruned_loss=0.09529, over 28986.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.338, pruned_loss=0.08722, over 5667760.40 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3436, pruned_loss=0.1092, over 5699755.05 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3372, pruned_loss=0.08471, over 5668881.65 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:36:32,287 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-10 14:36:35,626 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=901636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:36:43,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-10 14:37:08,858 INFO [train.py:968] (1/2) Epoch 20, batch 34500, libri_loss[loss=0.2922, simple_loss=0.3554, pruned_loss=0.1145, over 29529.00 frames. ], tot_loss[loss=0.255, simple_loss=0.336, pruned_loss=0.08696, over 5664410.82 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3438, pruned_loss=0.1094, over 5685177.96 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.335, pruned_loss=0.08402, over 5677556.17 frames. ], batch size: 89, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:37:57,508 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 34550, giga_loss[loss=0.2308, simple_loss=0.3244, pruned_loss=0.0686, over 29005.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3343, pruned_loss=0.08513, over 5663209.47 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3438, pruned_loss=0.1094, over 5677312.53 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3334, pruned_loss=0.08265, over 5679940.18 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:38:41,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=901719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:38:45,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=901722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:38:47,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3230, 1.5516, 1.6274, 1.3571], device='cuda:1'), covar=tensor([0.3685, 0.2353, 0.1835, 0.2624], device='cuda:1'), in_proj_covar=tensor([0.1916, 0.1819, 0.1751, 0.1899], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 14:39:16,756 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=901751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:39:28,145 INFO [train.py:968] (1/2) Epoch 20, batch 34600, giga_loss[loss=0.262, simple_loss=0.3463, pruned_loss=0.08889, over 28885.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.334, pruned_loss=0.08527, over 5673085.20 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3448, pruned_loss=0.1102, over 5674750.42 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3319, pruned_loss=0.0813, over 5689590.86 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:40:05,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4775, 1.7131, 1.6506, 1.4647], device='cuda:1'), covar=tensor([0.2890, 0.2244, 0.1754, 0.2190], device='cuda:1'), in_proj_covar=tensor([0.1921, 0.1825, 0.1756, 0.1904], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 14:40:10,819 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-10 14:40:35,565 INFO [optim.py:369] (1/2) 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 14:40:35,583 INFO [train.py:968] (1/2) Epoch 20, batch 34650, giga_loss[loss=0.2398, simple_loss=0.3307, pruned_loss=0.0745, over 28939.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3342, pruned_loss=0.08573, over 5674628.34 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3446, pruned_loss=0.11, over 5679217.26 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3326, pruned_loss=0.0823, over 5683715.10 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:41:37,382 INFO [train.py:968] (1/2) Epoch 20, batch 34700, giga_loss[loss=0.2559, simple_loss=0.345, pruned_loss=0.08337, over 28586.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3364, pruned_loss=0.08709, over 5666818.64 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3442, pruned_loss=0.1099, over 5675688.84 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3352, pruned_loss=0.08392, over 5677030.19 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:42:38,043 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 34750, giga_loss[loss=0.2143, simple_loss=0.3033, pruned_loss=0.06264, over 29025.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3362, pruned_loss=0.08739, over 5663468.01 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3444, pruned_loss=0.11, over 5677083.26 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.335, pruned_loss=0.08454, over 5670096.06 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:42:44,480 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-10 14:43:22,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3444, 1.2642, 1.2304, 1.5094], device='cuda:1'), covar=tensor([0.0760, 0.0369, 0.0369, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0117, 0.0119, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 14:43:33,539 INFO [train.py:968] (1/2) Epoch 20, batch 34800, giga_loss[loss=0.2241, simple_loss=0.3113, pruned_loss=0.06849, over 28496.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.08716, over 5656339.94 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3444, pruned_loss=0.1102, over 5671228.75 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3318, pruned_loss=0.08399, over 5666036.01 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 14:44:32,692 INFO [train.py:968] (1/2) Epoch 20, batch 34850, giga_loss[loss=0.2653, simple_loss=0.3422, pruned_loss=0.09422, over 27572.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3332, pruned_loss=0.08774, over 5653017.64 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3443, pruned_loss=0.1102, over 5664466.10 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.332, pruned_loss=0.08485, over 5666663.72 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:44:33,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:44:33,297 INFO [optim.py:369] (1/2) 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 14:45:26,964 INFO [train.py:968] (1/2) Epoch 20, batch 34900, giga_loss[loss=0.336, simple_loss=0.3884, pruned_loss=0.1418, over 26749.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3385, pruned_loss=0.09096, over 5647842.98 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.344, pruned_loss=0.11, over 5665850.82 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3376, pruned_loss=0.08842, over 5657355.69 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:45:33,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:46:08,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5805, 1.6788, 1.2229, 1.2056], device='cuda:1'), covar=tensor([0.0896, 0.0602, 0.0936, 0.1438], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0441, 0.0512, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 14:46:18,094 INFO [train.py:968] (1/2) Epoch 20, batch 34950, giga_loss[loss=0.3471, simple_loss=0.4223, pruned_loss=0.136, over 28796.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.348, pruned_loss=0.09548, over 5662451.80 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.344, pruned_loss=0.11, over 5665850.82 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3473, pruned_loss=0.0935, over 5669855.66 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:46:18,610 INFO [optim.py:369] (1/2) 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 14:46:51,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3431, 1.6312, 1.3361, 1.0981], device='cuda:1'), covar=tensor([0.2838, 0.2759, 0.3243, 0.2225], device='cuda:1'), in_proj_covar=tensor([0.1479, 0.1070, 0.1312, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 14:47:00,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:03,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:05,570 INFO [train.py:968] (1/2) Epoch 20, batch 35000, libri_loss[loss=0.299, simple_loss=0.3665, pruned_loss=0.1158, over 29530.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3508, pruned_loss=0.09751, over 5668506.92 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3439, pruned_loss=0.11, over 5669371.35 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3505, pruned_loss=0.09583, over 5671044.23 frames. ], batch size: 84, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:47:28,626 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:43,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4172, 1.7099, 1.4199, 1.2733], device='cuda:1'), covar=tensor([0.2628, 0.2714, 0.3061, 0.2455], device='cuda:1'), in_proj_covar=tensor([0.1478, 0.1071, 0.1311, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 14:47:47,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:47:47,882 INFO [train.py:968] (1/2) Epoch 20, batch 35050, giga_loss[loss=0.2358, simple_loss=0.3107, pruned_loss=0.08047, over 28664.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3477, pruned_loss=0.09687, over 5679744.97 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3446, pruned_loss=0.1103, over 5674878.33 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3469, pruned_loss=0.0949, over 5676704.04 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:47:48,449 INFO [optim.py:369] (1/2) 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 14:47:50,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:48:13,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.50 vs. limit=5.0
+2023-03-10 14:48:15,509 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:48:28,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9906, 1.2419, 1.3395, 1.0790], device='cuda:1'), covar=tensor([0.1942, 0.1394, 0.2344, 0.1726], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0735, 0.0700, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 14:48:33,501 INFO [train.py:968] (1/2) Epoch 20, batch 35100, giga_loss[loss=0.2599, simple_loss=0.3299, pruned_loss=0.09495, over 27860.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3397, pruned_loss=0.09325, over 5675638.94 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3445, pruned_loss=0.1102, over 5676070.94 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3392, pruned_loss=0.09173, over 5672228.00 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:48:39,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2486, 0.7955, 0.9049, 1.4388], device='cuda:1'), covar=tensor([0.0795, 0.0400, 0.0383, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 14:48:40,336 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-10 14:49:00,540 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:49:16,585 INFO [train.py:968] (1/2) Epoch 20, batch 35150, giga_loss[loss=0.2229, simple_loss=0.2932, pruned_loss=0.07627, over 28995.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3318, pruned_loss=0.08965, over 5688551.51 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3446, pruned_loss=0.1102, over 5679559.84 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3312, pruned_loss=0.08821, over 5682850.88 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:49:18,787 INFO [optim.py:369] (1/2) 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 14:50:00,366 INFO [train.py:968] (1/2) Epoch 20, batch 35200, giga_loss[loss=0.1989, simple_loss=0.2767, pruned_loss=0.06049, over 28646.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3238, pruned_loss=0.08632, over 5687789.31 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3449, pruned_loss=0.1104, over 5681582.77 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3229, pruned_loss=0.08483, over 5681544.85 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:50:08,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 14:50:23,825 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902386.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:50:37,868 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:50:38,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.2079, 6.0009, 5.7523, 2.7744], device='cuda:1'), covar=tensor([0.0508, 0.0654, 0.0808, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.1184, 0.1090, 0.0927, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 14:50:43,606 INFO [train.py:968] (1/2) Epoch 20, batch 35250, giga_loss[loss=0.2251, simple_loss=0.2947, pruned_loss=0.07775, over 28609.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3186, pruned_loss=0.08446, over 5679584.40 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3448, pruned_loss=0.1103, over 5677030.36 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.3173, pruned_loss=0.08265, over 5679098.34 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:50:45,085 INFO [optim.py:369] (1/2) 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 14:50:53,872 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 14:50:56,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3827, 1.4339, 1.4082, 1.6294], device='cuda:1'), covar=tensor([0.0760, 0.0389, 0.0342, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 14:51:26,265 INFO [train.py:968] (1/2) Epoch 20, batch 35300, giga_loss[loss=0.256, simple_loss=0.3207, pruned_loss=0.09568, over 27541.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3146, pruned_loss=0.08254, over 5692001.20 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3447, pruned_loss=0.1103, over 5679170.17 frames. ], giga_tot_loss[loss=0.2376, simple_loss=0.3134, pruned_loss=0.08091, over 5689937.08 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:52:09,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4707, 1.5972, 1.3390, 1.6388], device='cuda:1'), covar=tensor([0.0811, 0.0337, 0.0349, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 14:52:11,589 INFO [train.py:968] (1/2) Epoch 20, batch 35350, libri_loss[loss=0.3453, simple_loss=0.4009, pruned_loss=0.1449, over 18758.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3113, pruned_loss=0.08104, over 5676080.13 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3453, pruned_loss=0.1107, over 5662398.27 frames. ], giga_tot_loss[loss=0.2337, simple_loss=0.3093, pruned_loss=0.07906, over 5690430.45 frames. ], batch size: 187, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:52:13,278 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:52:13,735 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 35400, libri_loss[loss=0.2641, simple_loss=0.3338, pruned_loss=0.09715, over 29583.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.3094, pruned_loss=0.08086, over 5668561.78 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3457, pruned_loss=0.1106, over 5668434.07 frames. ], giga_tot_loss[loss=0.232, simple_loss=0.3067, pruned_loss=0.07861, over 5674641.21 frames. ], batch size: 76, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:53:19,055 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:53:40,151 INFO [train.py:968] (1/2) Epoch 20, batch 35450, giga_loss[loss=0.2263, simple_loss=0.3019, pruned_loss=0.07541, over 29084.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3063, pruned_loss=0.0794, over 5675529.98 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3463, pruned_loss=0.111, over 5670859.92 frames. ], giga_tot_loss[loss=0.2282, simple_loss=0.3029, pruned_loss=0.07674, over 5678184.51 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:53:43,209 INFO [optim.py:369] (1/2) 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 14:54:24,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4379, 1.3939, 3.7310, 3.1140], device='cuda:1'), covar=tensor([0.1479, 0.2655, 0.0451, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0642, 0.0945, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 14:54:26,603 INFO [train.py:968] (1/2) Epoch 20, batch 35500, giga_loss[loss=0.2344, simple_loss=0.2991, pruned_loss=0.08484, over 26636.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3024, pruned_loss=0.07729, over 5678064.39 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3464, pruned_loss=0.111, over 5670422.62 frames. ], giga_tot_loss[loss=0.2248, simple_loss=0.2995, pruned_loss=0.07506, over 5680530.36 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:54:30,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:54:34,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-10 14:54:54,200 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:55:06,893 INFO [train.py:968] (1/2) Epoch 20, batch 35550, giga_loss[loss=0.2177, simple_loss=0.2846, pruned_loss=0.07538, over 28971.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3002, pruned_loss=0.07645, over 5683036.79 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3468, pruned_loss=0.111, over 5672289.54 frames. ], giga_tot_loss[loss=0.2221, simple_loss=0.2965, pruned_loss=0.07387, over 5683687.41 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 14:55:09,282 INFO [optim.py:369] (1/2) 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 14:55:09,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5770, 1.6830, 1.6880, 1.5278], device='cuda:1'), covar=tensor([0.3230, 0.2655, 0.2354, 0.2871], device='cuda:1'), in_proj_covar=tensor([0.1935, 0.1833, 0.1766, 0.1921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 14:55:51,657 INFO [train.py:968] (1/2) Epoch 20, batch 35600, giga_loss[loss=0.2012, simple_loss=0.2761, pruned_loss=0.06317, over 28391.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2978, pruned_loss=0.0754, over 5676968.10 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.347, pruned_loss=0.111, over 5666398.39 frames. ], giga_tot_loss[loss=0.2195, simple_loss=0.2937, pruned_loss=0.07262, over 5683523.67 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:55:51,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902761.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:55:57,992 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=902769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:56:05,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:56:08,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2974, 2.3573, 2.1666, 2.0820], device='cuda:1'), covar=tensor([0.1913, 0.2376, 0.2311, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0739, 0.0702, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 14:56:30,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-10 14:56:31,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:56:33,880 INFO [train.py:968] (1/2) Epoch 20, batch 35650, giga_loss[loss=0.2087, simple_loss=0.2861, pruned_loss=0.06567, over 29076.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2961, pruned_loss=0.07476, over 5673719.98 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3476, pruned_loss=0.111, over 5671233.26 frames. ], giga_tot_loss[loss=0.2167, simple_loss=0.2907, pruned_loss=0.07134, over 5674869.45 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:56:35,786 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:56:59,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:57:21,893 INFO [train.py:968] (1/2) Epoch 20, batch 35700, giga_loss[loss=0.2714, simple_loss=0.3549, pruned_loss=0.09393, over 28915.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3009, pruned_loss=0.07784, over 5667004.41 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3485, pruned_loss=0.1116, over 5666070.05 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2949, pruned_loss=0.07398, over 5672133.42 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:57:47,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:02,477 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902904.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:58:05,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 14:58:09,701 INFO [train.py:968] (1/2) Epoch 20, batch 35750, giga_loss[loss=0.3778, simple_loss=0.421, pruned_loss=0.1673, over 26498.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3142, pruned_loss=0.0845, over 5678443.34 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3489, pruned_loss=0.1117, over 5669434.80 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.3085, pruned_loss=0.08089, over 5679704.06 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:58:12,562 INFO [optim.py:369] (1/2) 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 14:58:20,482 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=902920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:22,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=902923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:34,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2735, 3.4454, 1.4421, 1.5338], device='cuda:1'), covar=tensor([0.1071, 0.0313, 0.0944, 0.1395], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0545, 0.0377, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 14:58:34,727 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902936.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 14:58:50,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=902952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:58:56,402 INFO [train.py:968] (1/2) Epoch 20, batch 35800, giga_loss[loss=0.3052, simple_loss=0.3793, pruned_loss=0.1156, over 28911.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3275, pruned_loss=0.09171, over 5670703.61 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3488, pruned_loss=0.1116, over 5668217.23 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3227, pruned_loss=0.08863, over 5673109.99 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:58:56,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=902961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 14:59:36,549 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-10 14:59:41,560 INFO [train.py:968] (1/2) Epoch 20, batch 35850, giga_loss[loss=0.2756, simple_loss=0.3568, pruned_loss=0.0972, over 28936.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3356, pruned_loss=0.095, over 5670444.27 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3491, pruned_loss=0.1118, over 5661423.17 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3314, pruned_loss=0.09224, over 5678028.86 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 14:59:44,517 INFO [optim.py:369] (1/2) 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 14:59:59,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:01,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=903034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:25,932 INFO [train.py:968] (1/2) Epoch 20, batch 35900, libri_loss[loss=0.3216, simple_loss=0.3859, pruned_loss=0.1287, over 27717.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3393, pruned_loss=0.09555, over 5679179.38 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3492, pruned_loss=0.1117, over 5664636.20 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3356, pruned_loss=0.09301, over 5682708.06 frames. ], batch size: 115, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:00:27,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:31,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=903069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:00:50,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7406, 4.5398, 4.2765, 2.0374], device='cuda:1'), covar=tensor([0.0483, 0.0685, 0.0706, 0.2025], device='cuda:1'), in_proj_covar=tensor([0.1176, 0.1086, 0.0925, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 15:01:01,713 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:01:04,686 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 35950, giga_loss[loss=0.2408, simple_loss=0.3347, pruned_loss=0.07342, over 28652.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3411, pruned_loss=0.09554, over 5667854.11 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.349, pruned_loss=0.1113, over 5665429.71 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3378, pruned_loss=0.09311, over 5670985.03 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:01:11,462 INFO [optim.py:369] (1/2) 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 15:01:35,643 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:01:42,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=903144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:01:55,309 INFO [train.py:968] (1/2) Epoch 20, batch 36000, giga_loss[loss=0.2737, simple_loss=0.3515, pruned_loss=0.09798, over 28651.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3429, pruned_loss=0.0963, over 5667031.62 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3491, pruned_loss=0.1112, over 5667727.18 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3401, pruned_loss=0.09425, over 5667438.09 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:01:55,310 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 15:02:04,853 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 15:02:25,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 15:02:49,294 INFO [train.py:968] (1/2) Epoch 20, batch 36050, giga_loss[loss=0.2805, simple_loss=0.3439, pruned_loss=0.1086, over 28607.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3448, pruned_loss=0.09758, over 5673454.16 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3494, pruned_loss=0.1113, over 5660568.26 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3422, pruned_loss=0.09556, over 5680362.40 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:02:50,208 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:02:51,289 INFO [optim.py:369] (1/2) 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 15:02:52,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=903215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:03:17,142 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:03:23,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3279, 3.2147, 1.4912, 1.3974], device='cuda:1'), covar=tensor([0.1051, 0.0279, 0.0936, 0.1480], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0545, 0.0379, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 15:03:31,548 INFO [train.py:968] (1/2) Epoch 20, batch 36100, giga_loss[loss=0.3253, simple_loss=0.3948, pruned_loss=0.1279, over 28855.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3491, pruned_loss=0.1009, over 5674190.31 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3501, pruned_loss=0.1117, over 5663059.00 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3464, pruned_loss=0.0986, over 5677728.60 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:03:54,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=903287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:03:56,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=903290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:04:03,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6319, 1.7662, 1.6605, 1.5554], device='cuda:1'), covar=tensor([0.1842, 0.2146, 0.2407, 0.2115], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0742, 0.0706, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:04:14,577 INFO [train.py:968] (1/2) Epoch 20, batch 36150, giga_loss[loss=0.2577, simple_loss=0.3457, pruned_loss=0.08483, over 28960.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1023, over 5688882.85 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3506, pruned_loss=0.1119, over 5665241.61 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3506, pruned_loss=0.1002, over 5689894.42 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:04:17,248 INFO [optim.py:369] (1/2) 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 15:04:22,045 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=903319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:04:42,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6203, 1.3614, 4.6199, 3.4171], device='cuda:1'), covar=tensor([0.1722, 0.2959, 0.0388, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0743, 0.0638, 0.0938, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 15:04:56,280 INFO [train.py:968] (1/2) Epoch 20, batch 36200, giga_loss[loss=0.2672, simple_loss=0.3458, pruned_loss=0.09433, over 28577.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3547, pruned_loss=0.1021, over 5691891.39 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3506, pruned_loss=0.1117, over 5670324.90 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3527, pruned_loss=0.1005, over 5688763.05 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:05:40,627 INFO [train.py:968] (1/2) Epoch 20, batch 36250, giga_loss[loss=0.2743, simple_loss=0.3545, pruned_loss=0.09703, over 28832.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3558, pruned_loss=0.1017, over 5698592.60 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3509, pruned_loss=0.1118, over 5674555.70 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.354, pruned_loss=0.1002, over 5692746.77 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:05:43,787 INFO [optim.py:369] (1/2) 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 15:06:21,671 INFO [train.py:968] (1/2) Epoch 20, batch 36300, giga_loss[loss=0.2469, simple_loss=0.3342, pruned_loss=0.07983, over 28986.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3552, pruned_loss=0.1004, over 5689549.88 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3508, pruned_loss=0.1118, over 5668559.71 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3539, pruned_loss=0.09906, over 5690549.57 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:06:31,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6813, 1.7895, 1.8099, 1.5355], device='cuda:1'), covar=tensor([0.3036, 0.2903, 0.2689, 0.2862], device='cuda:1'), in_proj_covar=tensor([0.1933, 0.1836, 0.1770, 0.1919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 15:06:36,341 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-10 15:06:39,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1406, 2.5142, 1.2343, 1.2341], device='cuda:1'), covar=tensor([0.1123, 0.0326, 0.1002, 0.1558], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0545, 0.0378, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 15:06:40,962 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 15:07:03,529 INFO [train.py:968] (1/2) Epoch 20, batch 36350, giga_loss[loss=0.2561, simple_loss=0.3384, pruned_loss=0.08689, over 28733.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3546, pruned_loss=0.09919, over 5690507.91 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3515, pruned_loss=0.1122, over 5660781.15 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.353, pruned_loss=0.09749, over 5699318.84 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:07:06,154 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 36400, giga_loss[loss=0.2369, simple_loss=0.3257, pruned_loss=0.07403, over 28533.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3529, pruned_loss=0.09791, over 5694067.58 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3516, pruned_loss=0.1121, over 5669376.46 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3516, pruned_loss=0.09617, over 5694283.23 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:08:26,563 INFO [train.py:968] (1/2) Epoch 20, batch 36450, giga_loss[loss=0.289, simple_loss=0.3584, pruned_loss=0.1098, over 28908.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3552, pruned_loss=0.1007, over 5691099.98 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.352, pruned_loss=0.1125, over 5672007.60 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3538, pruned_loss=0.09876, over 5689167.60 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:08:29,977 INFO [optim.py:369] (1/2) 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 15:08:31,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1568, 1.5301, 1.4341, 1.5923], device='cuda:1'), covar=tensor([0.0856, 0.0319, 0.0297, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 15:08:36,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3824, 1.1747, 1.1501, 1.5448], device='cuda:1'), covar=tensor([0.0809, 0.0377, 0.0363, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 15:09:13,065 INFO [train.py:968] (1/2) Epoch 20, batch 36500, giga_loss[loss=0.3131, simple_loss=0.3704, pruned_loss=0.1279, over 28672.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3574, pruned_loss=0.1045, over 5687624.55 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.352, pruned_loss=0.1125, over 5671284.77 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3564, pruned_loss=0.1028, over 5686821.88 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:09:54,101 INFO [train.py:968] (1/2) Epoch 20, batch 36550, giga_loss[loss=0.3725, simple_loss=0.4081, pruned_loss=0.1685, over 28792.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3593, pruned_loss=0.1077, over 5683720.70 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3526, pruned_loss=0.1127, over 5666041.32 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3582, pruned_loss=0.106, over 5687790.40 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:09:59,639 INFO [optim.py:369] (1/2) 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 15:10:05,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5534, 1.8262, 1.5117, 1.5921], device='cuda:1'), covar=tensor([0.2152, 0.2102, 0.2276, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.1481, 0.1075, 0.1310, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 15:10:32,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 15:10:42,032 INFO [train.py:968] (1/2) Epoch 20, batch 36600, giga_loss[loss=0.3655, simple_loss=0.4031, pruned_loss=0.1639, over 26724.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3569, pruned_loss=0.1071, over 5678849.38 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3526, pruned_loss=0.1127, over 5663751.34 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.356, pruned_loss=0.1057, over 5684375.52 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:10:52,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-10 15:11:17,479 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-10 15:11:22,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7304, 1.8135, 1.9499, 1.4983], device='cuda:1'), covar=tensor([0.1814, 0.2354, 0.1431, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0891, 0.0693, 0.0935, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 15:11:24,037 INFO [train.py:968] (1/2) Epoch 20, batch 36650, giga_loss[loss=0.2699, simple_loss=0.3238, pruned_loss=0.108, over 23925.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3551, pruned_loss=0.1066, over 5691836.99 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3528, pruned_loss=0.1128, over 5669395.10 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3543, pruned_loss=0.1053, over 5691651.47 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:11:27,605 INFO [optim.py:369] (1/2) 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 15:11:37,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-10 15:11:51,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3659, 2.2341, 2.0608, 1.8760], device='cuda:1'), covar=tensor([0.1803, 0.2504, 0.2304, 0.2356], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0745, 0.0706, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:11:52,327 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=903843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:11:56,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6650, 4.5855, 1.9748, 1.9256], device='cuda:1'), covar=tensor([0.1012, 0.0228, 0.0820, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0547, 0.0378, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 15:12:07,267 INFO [train.py:968] (1/2) Epoch 20, batch 36700, giga_loss[loss=0.2884, simple_loss=0.3575, pruned_loss=0.1096, over 27574.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3538, pruned_loss=0.1056, over 5687708.64 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3526, pruned_loss=0.1126, over 5662369.51 frames. ], giga_tot_loss[loss=0.2812, simple_loss=0.3534, pruned_loss=0.1045, over 5694563.86 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:12:51,289 INFO [train.py:968] (1/2) Epoch 20, batch 36750, giga_loss[loss=0.2432, simple_loss=0.3219, pruned_loss=0.08226, over 28898.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3515, pruned_loss=0.1033, over 5684306.66 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3529, pruned_loss=0.1126, over 5659092.74 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.351, pruned_loss=0.1022, over 5694159.83 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:12:58,133 INFO [optim.py:369] (1/2) 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 15:13:12,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9355, 2.9339, 2.1058, 1.1958], device='cuda:1'), covar=tensor([0.7645, 0.3013, 0.3492, 0.6315], device='cuda:1'), in_proj_covar=tensor([0.1726, 0.1632, 0.1587, 0.1406], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 15:13:37,231 INFO [train.py:968] (1/2) Epoch 20, batch 36800, giga_loss[loss=0.2232, simple_loss=0.304, pruned_loss=0.07121, over 28734.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3481, pruned_loss=0.1011, over 5683687.51 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3535, pruned_loss=0.1129, over 5664661.33 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3471, pruned_loss=0.09963, over 5687326.04 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:13:56,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-10 15:14:19,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-10 15:14:24,571 INFO [train.py:968] (1/2) Epoch 20, batch 36850, giga_loss[loss=0.2662, simple_loss=0.3404, pruned_loss=0.096, over 28974.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3422, pruned_loss=0.09736, over 5697021.42 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3534, pruned_loss=0.1128, over 5670414.55 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3412, pruned_loss=0.09599, over 5695294.35 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:14:28,563 INFO [optim.py:369] (1/2) 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 15:14:31,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5301, 1.6812, 1.7352, 1.3517], device='cuda:1'), covar=tensor([0.1171, 0.1777, 0.1035, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0891, 0.0694, 0.0936, 0.0836], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 15:15:04,299 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-10 15:15:19,014 INFO [train.py:968] (1/2) Epoch 20, batch 36900, giga_loss[loss=0.2177, simple_loss=0.3037, pruned_loss=0.06589, over 29017.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3357, pruned_loss=0.0941, over 5662839.84 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3535, pruned_loss=0.1129, over 5653170.85 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3346, pruned_loss=0.09273, over 5677052.97 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:15:55,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-10 15:16:09,744 INFO [train.py:968] (1/2) Epoch 20, batch 36950, giga_loss[loss=0.2452, simple_loss=0.3266, pruned_loss=0.08186, over 28958.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3324, pruned_loss=0.09219, over 5660616.02 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3538, pruned_loss=0.1131, over 5649610.92 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3308, pruned_loss=0.09042, over 5675329.32 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:16:13,996 INFO [optim.py:369] (1/2) 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 15:16:55,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2600, 2.9534, 1.3999, 1.4390], device='cuda:1'), covar=tensor([0.1039, 0.0323, 0.0900, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0547, 0.0378, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 15:16:57,029 INFO [train.py:968] (1/2) Epoch 20, batch 37000, giga_loss[loss=0.3293, simple_loss=0.3759, pruned_loss=0.1414, over 26582.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3337, pruned_loss=0.09227, over 5665987.21 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3541, pruned_loss=0.1131, over 5652732.03 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3319, pruned_loss=0.09058, over 5674903.86 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:17:38,387 INFO [train.py:968] (1/2) Epoch 20, batch 37050, giga_loss[loss=0.2634, simple_loss=0.3403, pruned_loss=0.09326, over 27647.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3335, pruned_loss=0.09143, over 5680990.56 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3544, pruned_loss=0.113, over 5654569.77 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3311, pruned_loss=0.08946, over 5687384.58 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:17:43,915 INFO [optim.py:369] (1/2) 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 15:17:44,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=904218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:17:44,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3156, 1.5141, 1.5156, 1.3302], device='cuda:1'), covar=tensor([0.1960, 0.1963, 0.2360, 0.2078], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0746, 0.0709, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:18:19,639 INFO [train.py:968] (1/2) Epoch 20, batch 37100, giga_loss[loss=0.2634, simple_loss=0.3232, pruned_loss=0.1018, over 28329.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.334, pruned_loss=0.09258, over 5685590.47 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3549, pruned_loss=0.1131, over 5658912.41 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3313, pruned_loss=0.09057, over 5687226.92 frames. ], batch size: 65, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:18:29,939 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=904271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:19:01,321 INFO [train.py:968] (1/2) Epoch 20, batch 37150, giga_loss[loss=0.2351, simple_loss=0.3136, pruned_loss=0.0783, over 28793.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3325, pruned_loss=0.09208, over 5692307.18 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3554, pruned_loss=0.1132, over 5660612.17 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3295, pruned_loss=0.08999, over 5692468.10 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:19:06,796 INFO [optim.py:369] (1/2) 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 15:19:14,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=904326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:19:32,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6685, 1.7969, 1.7972, 1.5455], device='cuda:1'), covar=tensor([0.2037, 0.2484, 0.2453, 0.2550], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0747, 0.0710, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:19:35,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4790, 1.6310, 1.7091, 1.3087], device='cuda:1'), covar=tensor([0.1884, 0.2700, 0.1518, 0.1793], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0693, 0.0936, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 15:19:44,121 INFO [train.py:968] (1/2) Epoch 20, batch 37200, giga_loss[loss=0.2725, simple_loss=0.344, pruned_loss=0.1005, over 28718.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3308, pruned_loss=0.09117, over 5692703.40 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3563, pruned_loss=0.1137, over 5653907.54 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.327, pruned_loss=0.08848, over 5700459.24 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:19:44,411 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=904361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:19:46,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=904364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:20:07,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=904393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:20:08,180 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-10 15:20:21,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 15:20:22,624 INFO [train.py:968] (1/2) Epoch 20, batch 37250, giga_loss[loss=0.2366, simple_loss=0.311, pruned_loss=0.08111, over 28863.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3282, pruned_loss=0.08981, over 5705957.97 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3568, pruned_loss=0.1139, over 5661156.52 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.324, pruned_loss=0.08684, over 5706923.48 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:20:26,519 INFO [optim.py:369] (1/2) 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 15:21:02,472 INFO [train.py:968] (1/2) Epoch 20, batch 37300, giga_loss[loss=0.2498, simple_loss=0.3256, pruned_loss=0.08698, over 27980.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3279, pruned_loss=0.09027, over 5698501.13 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3573, pruned_loss=0.114, over 5657955.48 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3231, pruned_loss=0.08703, over 5704355.88 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:21:42,512 INFO [train.py:968] (1/2) Epoch 20, batch 37350, giga_loss[loss=0.2278, simple_loss=0.3072, pruned_loss=0.07423, over 28687.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3258, pruned_loss=0.08923, over 5705445.85 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3578, pruned_loss=0.1143, over 5662882.28 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.321, pruned_loss=0.08597, over 5706511.21 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:21:46,554 INFO [optim.py:369] (1/2) 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 15:22:21,669 INFO [train.py:968] (1/2) Epoch 20, batch 37400, libri_loss[loss=0.3101, simple_loss=0.3878, pruned_loss=0.1162, over 27721.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3229, pruned_loss=0.08753, over 5711246.80 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3582, pruned_loss=0.1142, over 5664967.90 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3179, pruned_loss=0.08438, over 5711338.78 frames. ], batch size: 115, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:22:41,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3500, 4.1871, 4.0269, 1.5566], device='cuda:1'), covar=tensor([0.0667, 0.0803, 0.0898, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.1189, 0.1103, 0.0936, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 15:22:57,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0343, 1.9514, 2.2217, 1.7762], device='cuda:1'), covar=tensor([0.1885, 0.2415, 0.1418, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0696, 0.0940, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 15:23:02,262 INFO [train.py:968] (1/2) Epoch 20, batch 37450, giga_loss[loss=0.2203, simple_loss=0.2952, pruned_loss=0.07275, over 28547.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3206, pruned_loss=0.08627, over 5721248.18 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3585, pruned_loss=0.1144, over 5668341.30 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3158, pruned_loss=0.08329, over 5719016.49 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:23:06,997 INFO [optim.py:369] (1/2) 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 15:23:30,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=904646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:23:43,603 INFO [train.py:968] (1/2) Epoch 20, batch 37500, giga_loss[loss=0.2458, simple_loss=0.3216, pruned_loss=0.08502, over 28714.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3212, pruned_loss=0.08688, over 5720266.71 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3596, pruned_loss=0.115, over 5664792.70 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3155, pruned_loss=0.08314, over 5723348.95 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:24:17,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4354, 1.7958, 1.4152, 1.6234], device='cuda:1'), covar=tensor([0.2843, 0.2882, 0.3312, 0.2413], device='cuda:1'), in_proj_covar=tensor([0.1485, 0.1076, 0.1314, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 15:24:18,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=904701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:24:25,545 INFO [train.py:968] (1/2) Epoch 20, batch 37550, giga_loss[loss=0.2766, simple_loss=0.3496, pruned_loss=0.1018, over 28885.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.323, pruned_loss=0.08803, over 5720148.53 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3596, pruned_loss=0.1149, over 5668249.41 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3179, pruned_loss=0.08483, over 5720051.98 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:24:31,680 INFO [optim.py:369] (1/2) 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 15:25:08,425 INFO [train.py:968] (1/2) Epoch 20, batch 37600, giga_loss[loss=0.2864, simple_loss=0.3479, pruned_loss=0.1125, over 28181.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3279, pruned_loss=0.09102, over 5719092.51 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3598, pruned_loss=0.1148, over 5675399.37 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3227, pruned_loss=0.0878, over 5713882.61 frames. ], batch size: 77, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:25:20,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4596, 1.7030, 1.6613, 1.5569], device='cuda:1'), covar=tensor([0.1893, 0.1768, 0.2241, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0749, 0.0713, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:25:32,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3711, 1.4079, 1.2601, 1.6228], device='cuda:1'), covar=tensor([0.0800, 0.0359, 0.0350, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 15:25:32,389 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=904789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:25:34,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=904792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:25:50,969 INFO [train.py:968] (1/2) Epoch 20, batch 37650, libri_loss[loss=0.2949, simple_loss=0.3655, pruned_loss=0.1121, over 29506.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3362, pruned_loss=0.09653, over 5693367.53 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3606, pruned_loss=0.1152, over 5667384.62 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3301, pruned_loss=0.09272, over 5698289.66 frames. ], batch size: 84, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:25:59,283 INFO [optim.py:369] (1/2) 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 15:26:01,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=904821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:26:26,818 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=904844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:26:28,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=904847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:26:41,364 INFO [train.py:968] (1/2) Epoch 20, batch 37700, libri_loss[loss=0.2322, simple_loss=0.3038, pruned_loss=0.08032, over 29653.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3448, pruned_loss=0.1025, over 5691754.18 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3603, pruned_loss=0.115, over 5670769.31 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.34, pruned_loss=0.09953, over 5692770.14 frames. ], batch size: 69, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:26:57,153 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=904876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:27:29,250 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=904908.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 15:27:32,156 INFO [train.py:968] (1/2) Epoch 20, batch 37750, giga_loss[loss=0.2548, simple_loss=0.3373, pruned_loss=0.08611, over 29057.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3483, pruned_loss=0.1034, over 5680704.19 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3602, pruned_loss=0.1148, over 5672157.77 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3444, pruned_loss=0.101, over 5680349.22 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:27:33,252 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6405, 1.6879, 1.8648, 1.4169], device='cuda:1'), covar=tensor([0.1717, 0.2562, 0.1442, 0.1765], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0696, 0.0940, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 15:27:38,953 INFO [optim.py:369] (1/2) 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 15:28:18,285 INFO [train.py:968] (1/2) Epoch 20, batch 37800, giga_loss[loss=0.2756, simple_loss=0.356, pruned_loss=0.09753, over 28855.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3527, pruned_loss=0.105, over 5681667.73 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3602, pruned_loss=0.1147, over 5674513.29 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3496, pruned_loss=0.1031, over 5679318.44 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:28:31,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3828, 1.1206, 4.5802, 3.2782], device='cuda:1'), covar=tensor([0.1779, 0.3188, 0.0378, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0741, 0.0634, 0.0935, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 15:28:50,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5695, 2.3454, 1.7081, 0.7899], device='cuda:1'), covar=tensor([0.3397, 0.1999, 0.2536, 0.4113], device='cuda:1'), in_proj_covar=tensor([0.1714, 0.1625, 0.1582, 0.1402], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 15:29:01,746 INFO [train.py:968] (1/2) Epoch 20, batch 37850, giga_loss[loss=0.2976, simple_loss=0.3742, pruned_loss=0.1105, over 28823.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3582, pruned_loss=0.1082, over 5673255.14 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.361, pruned_loss=0.1151, over 5670099.81 frames. ], giga_tot_loss[loss=0.2835, simple_loss=0.3548, pruned_loss=0.1061, over 5674618.67 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:29:09,047 INFO [optim.py:369] (1/2) 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 15:29:28,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:29:45,886 INFO [train.py:968] (1/2) Epoch 20, batch 37900, giga_loss[loss=0.2343, simple_loss=0.3192, pruned_loss=0.07474, over 28831.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3568, pruned_loss=0.1069, over 5670635.75 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3614, pruned_loss=0.1154, over 5668466.20 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3537, pruned_loss=0.1049, over 5672932.02 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:29:47,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1643, 3.4880, 2.5897, 1.2492], device='cuda:1'), covar=tensor([0.7636, 0.2662, 0.3147, 0.6717], device='cuda:1'), in_proj_covar=tensor([0.1718, 0.1624, 0.1588, 0.1404], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 15:30:07,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9150, 1.1411, 1.0431, 0.8614], device='cuda:1'), covar=tensor([0.2413, 0.2609, 0.1584, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1951, 0.1848, 0.1794, 0.1943], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 15:30:25,599 INFO [train.py:968] (1/2) Epoch 20, batch 37950, giga_loss[loss=0.2618, simple_loss=0.3408, pruned_loss=0.09141, over 28994.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3535, pruned_loss=0.1041, over 5674460.66 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3615, pruned_loss=0.1156, over 5666584.18 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3508, pruned_loss=0.102, over 5678150.38 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:30:35,493 INFO [optim.py:369] (1/2) 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 15:31:02,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1431, 1.2545, 3.8196, 3.2224], device='cuda:1'), covar=tensor([0.1862, 0.2956, 0.0468, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0638, 0.0939, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 15:31:09,009 INFO [train.py:968] (1/2) Epoch 20, batch 38000, giga_loss[loss=0.26, simple_loss=0.3354, pruned_loss=0.09232, over 28883.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3514, pruned_loss=0.1019, over 5681579.44 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3613, pruned_loss=0.1154, over 5669729.44 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3492, pruned_loss=0.1002, over 5681912.43 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:31:51,008 INFO [train.py:968] (1/2) Epoch 20, batch 38050, giga_loss[loss=0.2715, simple_loss=0.3538, pruned_loss=0.09457, over 28737.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.351, pruned_loss=0.1013, over 5687054.24 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3613, pruned_loss=0.1153, over 5673053.95 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3492, pruned_loss=0.09982, over 5684684.29 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:31:59,138 INFO [optim.py:369] (1/2) 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 15:32:14,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 15:32:19,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 15:32:33,487 INFO [train.py:968] (1/2) Epoch 20, batch 38100, giga_loss[loss=0.3144, simple_loss=0.3756, pruned_loss=0.1266, over 28833.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3538, pruned_loss=0.1032, over 5686013.80 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3623, pruned_loss=0.116, over 5674051.57 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.351, pruned_loss=0.1007, over 5683951.07 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:32:52,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905283.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 15:33:16,074 INFO [train.py:968] (1/2) Epoch 20, batch 38150, giga_loss[loss=0.2889, simple_loss=0.3641, pruned_loss=0.1068, over 29024.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3561, pruned_loss=0.1048, over 5679626.63 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3627, pruned_loss=0.1163, over 5668491.09 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3533, pruned_loss=0.1024, over 5683221.07 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:33:24,355 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:33:51,946 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:33:55,503 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 15:34:01,668 INFO [train.py:968] (1/2) Epoch 20, batch 38200, giga_loss[loss=0.2827, simple_loss=0.3592, pruned_loss=0.1031, over 29017.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3584, pruned_loss=0.1067, over 5692649.54 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3632, pruned_loss=0.1164, over 5674983.05 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3556, pruned_loss=0.1044, over 5690009.01 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:34:08,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8442, 1.1738, 1.2675, 0.9966], device='cuda:1'), covar=tensor([0.2215, 0.1405, 0.2390, 0.1752], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0746, 0.0711, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:34:42,600 INFO [train.py:968] (1/2) Epoch 20, batch 38250, giga_loss[loss=0.2871, simple_loss=0.3563, pruned_loss=0.109, over 28845.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3592, pruned_loss=0.108, over 5688627.67 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1163, over 5677591.94 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3568, pruned_loss=0.1059, over 5684997.00 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:34:47,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:34:51,599 INFO [optim.py:369] (1/2) 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 15:34:54,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 15:34:57,399 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905426.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 15:34:59,488 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905429.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 15:35:13,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-10 15:35:24,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2829, 1.0288, 4.1452, 3.3347], device='cuda:1'), covar=tensor([0.1754, 0.3083, 0.0412, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0637, 0.0939, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 15:35:24,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905458.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 15:35:26,276 INFO [train.py:968] (1/2) Epoch 20, batch 38300, giga_loss[loss=0.3235, simple_loss=0.3863, pruned_loss=0.1303, over 28631.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3598, pruned_loss=0.1086, over 5690898.47 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3631, pruned_loss=0.1163, over 5672398.78 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3578, pruned_loss=0.1068, over 5692889.48 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:36:06,536 INFO [train.py:968] (1/2) Epoch 20, batch 38350, giga_loss[loss=0.2657, simple_loss=0.3533, pruned_loss=0.08906, over 28921.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.3591, pruned_loss=0.1073, over 5698442.56 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3633, pruned_loss=0.1162, over 5678106.62 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3574, pruned_loss=0.1058, over 5695436.12 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:36:11,413 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:36:14,163 INFO [optim.py:369] (1/2) 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 15:36:46,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:36:47,970 INFO [train.py:968] (1/2) Epoch 20, batch 38400, giga_loss[loss=0.2858, simple_loss=0.3727, pruned_loss=0.09949, over 29000.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3577, pruned_loss=0.1051, over 5703360.99 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3633, pruned_loss=0.1162, over 5679030.63 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3561, pruned_loss=0.1037, over 5700665.22 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:36:48,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:37:10,502 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:37:13,146 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:37:13,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5175, 4.3115, 4.1848, 1.7147], device='cuda:1'), covar=tensor([0.0671, 0.0863, 0.0954, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.1186, 0.1100, 0.0935, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 15:37:19,884 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=905598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:37:29,507 INFO [train.py:968] (1/2) Epoch 20, batch 38450, giga_loss[loss=0.277, simple_loss=0.358, pruned_loss=0.09795, over 29011.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3575, pruned_loss=0.1041, over 5705084.53 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5681090.43 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3562, pruned_loss=0.1028, over 5701364.06 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:37:36,371 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-10 15:37:37,520 INFO [optim.py:369] (1/2) 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 15:37:39,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4801, 1.7497, 1.3835, 1.4954], device='cuda:1'), covar=tensor([0.2884, 0.2759, 0.3182, 0.2381], device='cuda:1'), in_proj_covar=tensor([0.1484, 0.1076, 0.1310, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 15:37:43,340 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 38500, libri_loss[loss=0.242, simple_loss=0.313, pruned_loss=0.08552, over 29367.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3556, pruned_loss=0.1032, over 5709024.43 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3629, pruned_loss=0.1158, over 5688307.52 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3548, pruned_loss=0.1022, over 5700356.82 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:38:35,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:38:48,612 INFO [train.py:968] (1/2) Epoch 20, batch 38550, giga_loss[loss=0.2447, simple_loss=0.3252, pruned_loss=0.08204, over 28407.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.353, pruned_loss=0.1021, over 5719037.10 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3619, pruned_loss=0.1152, over 5696359.92 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3529, pruned_loss=0.1013, over 5705528.58 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:38:55,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-10 15:38:55,949 INFO [optim.py:369] (1/2) 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 15:38:59,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:39:27,063 INFO [train.py:968] (1/2) Epoch 20, batch 38600, giga_loss[loss=0.2764, simple_loss=0.3467, pruned_loss=0.103, over 28692.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3503, pruned_loss=0.1004, over 5718845.09 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3619, pruned_loss=0.1151, over 5693663.62 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3499, pruned_loss=0.09941, over 5711305.53 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:40:09,637 INFO [train.py:968] (1/2) Epoch 20, batch 38650, giga_loss[loss=0.3129, simple_loss=0.3787, pruned_loss=0.1236, over 28759.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3505, pruned_loss=0.1011, over 5712151.56 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5692175.09 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.35, pruned_loss=0.1001, over 5707859.87 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:40:16,917 INFO [optim.py:369] (1/2) 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 15:40:24,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4906, 1.7488, 1.8341, 1.5491], device='cuda:1'), covar=tensor([0.1582, 0.1465, 0.1777, 0.1612], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0745, 0.0709, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:40:28,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:31,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:47,365 INFO [train.py:968] (1/2) Epoch 20, batch 38700, giga_loss[loss=0.2528, simple_loss=0.3386, pruned_loss=0.08353, over 29003.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3518, pruned_loss=0.1025, over 5712249.62 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3622, pruned_loss=0.1153, over 5689336.79 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3507, pruned_loss=0.101, over 5711635.15 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:40:53,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=905868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:54,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9478, 3.7301, 3.5674, 1.8833], device='cuda:1'), covar=tensor([0.0750, 0.0976, 0.0973, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1191, 0.1103, 0.0936, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 15:40:54,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:40:57,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=905871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:41:13,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:41:18,041 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=905900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:41:27,672 INFO [train.py:968] (1/2) Epoch 20, batch 38750, giga_loss[loss=0.2457, simple_loss=0.3268, pruned_loss=0.0823, over 28987.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3519, pruned_loss=0.102, over 5714027.20 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3621, pruned_loss=0.1153, over 5689928.94 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3511, pruned_loss=0.1007, over 5713249.24 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:41:35,074 INFO [optim.py:369] (1/2) 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 15:42:05,693 INFO [train.py:968] (1/2) Epoch 20, batch 38800, giga_loss[loss=0.3331, simple_loss=0.3943, pruned_loss=0.1359, over 28893.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1008, over 5694721.10 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1155, over 5675544.74 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.35, pruned_loss=0.09915, over 5708333.67 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:42:05,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:14,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=905973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:40,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:43,952 INFO [train.py:968] (1/2) Epoch 20, batch 38850, libri_loss[loss=0.2438, simple_loss=0.3201, pruned_loss=0.08374, over 29550.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3495, pruned_loss=0.09951, over 5706886.07 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1152, over 5680368.22 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3486, pruned_loss=0.09806, over 5713888.13 frames. ], batch size: 75, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:42:47,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:42:52,929 INFO [optim.py:369] (1/2) 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 15:42:55,775 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:43:07,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:43:09,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:43:28,640 INFO [train.py:968] (1/2) Epoch 20, batch 38900, giga_loss[loss=0.2567, simple_loss=0.3344, pruned_loss=0.08953, over 29087.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3483, pruned_loss=0.09915, over 5685910.12 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3622, pruned_loss=0.1154, over 5662236.91 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3472, pruned_loss=0.0976, over 5708959.88 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:43:32,971 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:03,630 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:05,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:09,191 INFO [train.py:968] (1/2) Epoch 20, batch 38950, libri_loss[loss=0.2978, simple_loss=0.3728, pruned_loss=0.1114, over 29544.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3448, pruned_loss=0.09735, over 5691798.83 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3618, pruned_loss=0.1152, over 5667092.15 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.344, pruned_loss=0.09597, over 5706208.78 frames. ], batch size: 82, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:44:13,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:15,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:17,152 INFO [optim.py:369] (1/2) 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 15:44:17,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5513, 1.7512, 1.4752, 1.7964], device='cuda:1'), covar=tensor([0.2299, 0.2347, 0.2450, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.1478, 0.1074, 0.1307, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 15:44:29,323 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:37,450 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:37,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:40,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:44:42,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3128, 1.3765, 1.2009, 1.5625], device='cuda:1'), covar=tensor([0.0803, 0.0366, 0.0355, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 15:44:47,676 INFO [train.py:968] (1/2) Epoch 20, batch 39000, giga_loss[loss=0.2388, simple_loss=0.3143, pruned_loss=0.08164, over 28824.00 frames. ], tot_loss[loss=0.267, simple_loss=0.342, pruned_loss=0.096, over 5688837.10 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3619, pruned_loss=0.1152, over 5662097.23 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3407, pruned_loss=0.09448, over 5706193.14 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:44:47,676 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 15:44:56,765 INFO [train.py:1012] (1/2) Epoch 20, validation: loss=0.2055, simple_loss=0.3137, pruned_loss=0.04862, over 944034.00 frames. 
+2023-03-10 15:44:56,766 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 15:45:12,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:45:12,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:45:13,707 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 15:45:16,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7622, 2.0463, 1.3253, 1.5935], device='cuda:1'), covar=tensor([0.0835, 0.0476, 0.0971, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0442, 0.0518, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 15:45:35,428 INFO [train.py:968] (1/2) Epoch 20, batch 39050, giga_loss[loss=0.2838, simple_loss=0.3601, pruned_loss=0.1037, over 29104.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3431, pruned_loss=0.09717, over 5690204.59 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3623, pruned_loss=0.1153, over 5661658.37 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3411, pruned_loss=0.09512, over 5705645.54 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:45:44,275 INFO [optim.py:369] (1/2) 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 15:46:16,049 INFO [train.py:968] (1/2) Epoch 20, batch 39100, giga_loss[loss=0.2325, simple_loss=0.3088, pruned_loss=0.0781, over 28660.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3431, pruned_loss=0.09778, over 5688348.97 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3626, pruned_loss=0.1155, over 5664105.40 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3408, pruned_loss=0.09564, over 5699128.68 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:46:54,315 INFO [train.py:968] (1/2) Epoch 20, batch 39150, giga_loss[loss=0.251, simple_loss=0.3195, pruned_loss=0.09125, over 28672.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3399, pruned_loss=0.09625, over 5698311.76 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5669167.90 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3376, pruned_loss=0.0941, over 5703028.09 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:47:02,800 INFO [optim.py:369] (1/2) 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 15:47:23,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:47:33,647 INFO [train.py:968] (1/2) Epoch 20, batch 39200, giga_loss[loss=0.2621, simple_loss=0.3321, pruned_loss=0.09605, over 28553.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.338, pruned_loss=0.09574, over 5696317.36 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3626, pruned_loss=0.1155, over 5660869.89 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3357, pruned_loss=0.09367, over 5707410.66 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:47:58,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:48:03,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:48:12,205 INFO [train.py:968] (1/2) Epoch 20, batch 39250, giga_loss[loss=0.2646, simple_loss=0.3315, pruned_loss=0.09882, over 28913.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3369, pruned_loss=0.09548, over 5686987.78 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3627, pruned_loss=0.1154, over 5652773.00 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3342, pruned_loss=0.0933, over 5704591.86 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:48:23,188 INFO [optim.py:369] (1/2) 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 15:48:58,018 INFO [train.py:968] (1/2) Epoch 20, batch 39300, giga_loss[loss=0.2523, simple_loss=0.3318, pruned_loss=0.08645, over 28930.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3354, pruned_loss=0.09451, over 5692965.36 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3629, pruned_loss=0.1156, over 5655323.37 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3327, pruned_loss=0.09232, over 5704985.55 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:49:11,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8356, 2.7399, 1.7641, 1.0951], device='cuda:1'), covar=tensor([0.8893, 0.3586, 0.4212, 0.6896], device='cuda:1'), in_proj_covar=tensor([0.1710, 0.1610, 0.1574, 0.1392], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 15:49:37,633 INFO [train.py:968] (1/2) Epoch 20, batch 39350, giga_loss[loss=0.2415, simple_loss=0.3282, pruned_loss=0.07738, over 29015.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3367, pruned_loss=0.09447, over 5693691.26 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3627, pruned_loss=0.1155, over 5660266.49 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3339, pruned_loss=0.09217, over 5699851.35 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:49:46,752 INFO [optim.py:369] (1/2) 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 15:49:58,849 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:49:58,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:00,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:05,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:07,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:18,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:22,329 INFO [train.py:968] (1/2) Epoch 20, batch 39400, giga_loss[loss=0.247, simple_loss=0.3399, pruned_loss=0.07704, over 29021.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3407, pruned_loss=0.0964, over 5690744.57 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5664345.88 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3381, pruned_loss=0.09434, over 5692646.52 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 15:50:29,035 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:50:37,006 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:51:01,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9898, 1.3151, 1.1360, 0.2406], device='cuda:1'), covar=tensor([0.4121, 0.3179, 0.4957, 0.6556], device='cuda:1'), in_proj_covar=tensor([0.1710, 0.1609, 0.1574, 0.1392], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 15:51:06,218 INFO [train.py:968] (1/2) Epoch 20, batch 39450, libri_loss[loss=0.3304, simple_loss=0.3938, pruned_loss=0.1335, over 29791.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3442, pruned_loss=0.0979, over 5696503.01 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1152, over 5673008.48 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3415, pruned_loss=0.09576, over 5690661.54 frames. ], batch size: 87, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:51:14,793 INFO [optim.py:369] (1/2) 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 15:51:30,789 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=906638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:51:39,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4076, 1.7049, 1.3514, 1.2520], device='cuda:1'), covar=tensor([0.2962, 0.2919, 0.3374, 0.2617], device='cuda:1'), in_proj_covar=tensor([0.1481, 0.1073, 0.1307, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 15:51:49,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3282, 1.5313, 1.3205, 1.4926], device='cuda:1'), covar=tensor([0.0696, 0.0428, 0.0364, 0.0787], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 15:51:50,282 INFO [train.py:968] (1/2) Epoch 20, batch 39500, giga_loss[loss=0.2667, simple_loss=0.3509, pruned_loss=0.09129, over 28717.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.09731, over 5696631.45 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1151, over 5675379.50 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3421, pruned_loss=0.0955, over 5690128.91 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:52:22,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:52:25,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 39550, giga_loss[loss=0.2576, simple_loss=0.3392, pruned_loss=0.08797, over 28972.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.344, pruned_loss=0.09681, over 5686010.23 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5657708.40 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3411, pruned_loss=0.09434, over 5696967.30 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:52:41,257 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:52:47,719 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:53:13,265 INFO [train.py:968] (1/2) Epoch 20, batch 39600, giga_loss[loss=0.2546, simple_loss=0.3321, pruned_loss=0.08853, over 28878.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3428, pruned_loss=0.09625, over 5689865.82 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5659687.86 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3403, pruned_loss=0.0941, over 5697036.98 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:53:53,872 INFO [train.py:968] (1/2) Epoch 20, batch 39650, giga_loss[loss=0.2748, simple_loss=0.3523, pruned_loss=0.09871, over 28694.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3427, pruned_loss=0.09622, over 5696003.03 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5654208.52 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3403, pruned_loss=0.09397, over 5708027.64 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:54:00,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1541, 2.5621, 1.2124, 1.3142], device='cuda:1'), covar=tensor([0.1010, 0.0424, 0.0964, 0.1411], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0547, 0.0378, 0.0419], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 15:54:04,596 INFO [optim.py:369] (1/2) 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 15:54:07,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3236, 1.4638, 1.3058, 1.5412], device='cuda:1'), covar=tensor([0.0705, 0.0394, 0.0354, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 15:54:35,328 INFO [train.py:968] (1/2) Epoch 20, batch 39700, libri_loss[loss=0.3151, simple_loss=0.3813, pruned_loss=0.1244, over 29223.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3457, pruned_loss=0.09796, over 5704659.26 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3641, pruned_loss=0.1164, over 5659496.25 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3424, pruned_loss=0.09511, over 5710625.43 frames. ], batch size: 94, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:54:38,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=906865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:54:40,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=906868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:55:01,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=906897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:55:11,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=906909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:55:12,832 INFO [train.py:968] (1/2) Epoch 20, batch 39750, libri_loss[loss=0.2714, simple_loss=0.3416, pruned_loss=0.1006, over 29573.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3506, pruned_loss=0.1013, over 5696497.92 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3638, pruned_loss=0.1164, over 5658008.39 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3474, pruned_loss=0.09827, over 5704984.17 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:55:25,711 INFO [optim.py:369] (1/2) 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 15:55:31,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 15:55:54,228 INFO [train.py:968] (1/2) Epoch 20, batch 39800, giga_loss[loss=0.2718, simple_loss=0.3438, pruned_loss=0.09993, over 28705.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3516, pruned_loss=0.1015, over 5701765.72 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3638, pruned_loss=0.1162, over 5659667.01 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.349, pruned_loss=0.09904, over 5707263.35 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:56:14,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3666, 3.7606, 1.5931, 1.5362], device='cuda:1'), covar=tensor([0.0999, 0.0293, 0.0931, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0547, 0.0378, 0.0418], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 15:56:33,403 INFO [train.py:968] (1/2) Epoch 20, batch 39850, giga_loss[loss=0.2883, simple_loss=0.3598, pruned_loss=0.1084, over 28731.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1023, over 5707203.37 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3643, pruned_loss=0.1164, over 5663441.40 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3504, pruned_loss=0.09984, over 5709036.56 frames. ], batch size: 243, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:56:35,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:56:46,689 INFO [optim.py:369] (1/2) 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 15:57:03,481 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:09,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:11,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:16,176 INFO [train.py:968] (1/2) Epoch 20, batch 39900, giga_loss[loss=0.2684, simple_loss=0.3511, pruned_loss=0.09286, over 28849.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3536, pruned_loss=0.102, over 5707682.06 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3642, pruned_loss=0.1162, over 5664948.31 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3512, pruned_loss=0.1, over 5708382.34 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:57:16,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4139, 1.6446, 1.4591, 1.6532], device='cuda:1'), covar=tensor([0.0758, 0.0309, 0.0328, 0.0851], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 15:57:22,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7657, 1.8414, 1.8267, 1.5984], device='cuda:1'), covar=tensor([0.1776, 0.2148, 0.2153, 0.2203], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0745, 0.0710, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 15:57:34,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:57:54,272 INFO [train.py:968] (1/2) Epoch 20, batch 39950, giga_loss[loss=0.2816, simple_loss=0.3652, pruned_loss=0.09894, over 28966.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3533, pruned_loss=0.1018, over 5710155.72 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3643, pruned_loss=0.1161, over 5671934.96 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.351, pruned_loss=0.09993, over 5705432.17 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 15:58:01,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:58:07,258 INFO [optim.py:369] (1/2) 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 15:58:31,546 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:58:33,677 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 40000, giga_loss[loss=0.3478, simple_loss=0.3981, pruned_loss=0.1488, over 26805.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3526, pruned_loss=0.1021, over 5706070.82 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1163, over 5665492.31 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3503, pruned_loss=0.1002, over 5708455.09 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:59:00,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 15:59:20,101 INFO [train.py:968] (1/2) Epoch 20, batch 40050, giga_loss[loss=0.2885, simple_loss=0.3546, pruned_loss=0.1112, over 27631.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.348, pruned_loss=0.09973, over 5708504.65 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3644, pruned_loss=0.1162, over 5667772.27 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3462, pruned_loss=0.09817, over 5708759.42 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 15:59:31,619 INFO [optim.py:369] (1/2) 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 15:59:57,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-10 16:00:01,885 INFO [train.py:968] (1/2) Epoch 20, batch 40100, giga_loss[loss=0.2754, simple_loss=0.3523, pruned_loss=0.09927, over 28858.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3442, pruned_loss=0.09764, over 5712220.09 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3642, pruned_loss=0.116, over 5669742.28 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09631, over 5711074.26 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:00:42,039 INFO [train.py:968] (1/2) Epoch 20, batch 40150, giga_loss[loss=0.2793, simple_loss=0.3652, pruned_loss=0.09675, over 28969.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3449, pruned_loss=0.09671, over 5716507.83 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3644, pruned_loss=0.1161, over 5672741.50 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.343, pruned_loss=0.09509, over 5714220.94 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:00:49,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 16:00:49,761 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:00:53,658 INFO [optim.py:369] (1/2) 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 16:01:07,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8172, 1.9779, 1.4199, 1.5725], device='cuda:1'), covar=tensor([0.0902, 0.0603, 0.1053, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0442, 0.0515, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 16:01:25,533 INFO [train.py:968] (1/2) Epoch 20, batch 40200, giga_loss[loss=0.2525, simple_loss=0.3362, pruned_loss=0.08443, over 28992.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3459, pruned_loss=0.09622, over 5710447.55 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3638, pruned_loss=0.1157, over 5677488.19 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3445, pruned_loss=0.09483, over 5705121.46 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:01:43,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2952, 1.8902, 1.5001, 1.4801], device='cuda:1'), covar=tensor([0.0733, 0.0351, 0.0333, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 16:01:53,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4927, 1.5781, 3.2684, 3.2260], device='cuda:1'), covar=tensor([0.1219, 0.2361, 0.0435, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0742, 0.0634, 0.0937, 0.0884], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 16:02:07,959 INFO [train.py:968] (1/2) Epoch 20, batch 40250, giga_loss[loss=0.3107, simple_loss=0.3706, pruned_loss=0.1254, over 26752.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3447, pruned_loss=0.09568, over 5712845.97 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3638, pruned_loss=0.1157, over 5679801.84 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3432, pruned_loss=0.09423, over 5707106.57 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:02:15,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-10 16:02:15,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:02:18,715 INFO [optim.py:369] (1/2) 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 16:02:24,944 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=907432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:02:31,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 16:02:48,983 INFO [train.py:968] (1/2) Epoch 20, batch 40300, giga_loss[loss=0.2494, simple_loss=0.3178, pruned_loss=0.09044, over 28918.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3447, pruned_loss=0.09684, over 5712223.04 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3641, pruned_loss=0.1158, over 5683051.96 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3431, pruned_loss=0.09538, over 5705382.41 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:03:13,450 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:03:28,038 INFO [train.py:968] (1/2) Epoch 20, batch 40350, giga_loss[loss=0.2707, simple_loss=0.3347, pruned_loss=0.1033, over 28760.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3444, pruned_loss=0.09854, over 5710126.35 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.364, pruned_loss=0.1158, over 5686109.56 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3427, pruned_loss=0.09687, over 5702450.10 frames. ], batch size: 99, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:03:40,092 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 16:03:40,252 INFO [optim.py:369] (1/2) 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 16:03:50,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-10 16:04:03,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3178, 1.6853, 1.5765, 1.4920], device='cuda:1'), covar=tensor([0.1936, 0.1680, 0.2282, 0.2028], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0746, 0.0711, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 16:04:10,969 INFO [train.py:968] (1/2) Epoch 20, batch 40400, giga_loss[loss=0.2519, simple_loss=0.3215, pruned_loss=0.09114, over 29041.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3436, pruned_loss=0.09933, over 5715875.27 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3641, pruned_loss=0.1159, over 5686879.97 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3419, pruned_loss=0.09765, over 5709587.83 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:04:14,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:04:16,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:04:42,461 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 40450, giga_loss[loss=0.2841, simple_loss=0.3552, pruned_loss=0.1065, over 28960.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3423, pruned_loss=0.09904, over 5722383.72 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3636, pruned_loss=0.1156, over 5690028.96 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.341, pruned_loss=0.09778, over 5715097.98 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:05:06,487 INFO [optim.py:369] (1/2) 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 16:05:15,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:05:17,691 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 40500, giga_loss[loss=0.2121, simple_loss=0.2919, pruned_loss=0.0662, over 28984.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3398, pruned_loss=0.09761, over 5714485.74 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5684071.51 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3384, pruned_loss=0.09626, over 5715180.47 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:05:40,424 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:06:04,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:06:16,800 INFO [train.py:968] (1/2) Epoch 20, batch 40550, giga_loss[loss=0.2488, simple_loss=0.3281, pruned_loss=0.08473, over 28306.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3364, pruned_loss=0.09607, over 5716736.50 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5686512.25 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.335, pruned_loss=0.09474, over 5715504.71 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:06:28,725 INFO [optim.py:369] (1/2) 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 16:06:43,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9020, 1.1266, 1.0077, 0.8191], device='cuda:1'), covar=tensor([0.2352, 0.2508, 0.1640, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1973, 0.1887, 0.1834, 0.1965], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 16:06:57,790 INFO [train.py:968] (1/2) Epoch 20, batch 40600, giga_loss[loss=0.229, simple_loss=0.3019, pruned_loss=0.07811, over 28703.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.333, pruned_loss=0.09419, over 5711564.99 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3644, pruned_loss=0.1162, over 5680396.77 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3302, pruned_loss=0.09198, over 5717984.22 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:07:35,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=907807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:07:38,420 INFO [train.py:968] (1/2) Epoch 20, batch 40650, giga_loss[loss=0.2541, simple_loss=0.3236, pruned_loss=0.09234, over 24031.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3337, pruned_loss=0.09399, over 5704013.71 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3642, pruned_loss=0.116, over 5681772.03 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3311, pruned_loss=0.092, over 5708479.87 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:07:51,838 INFO [optim.py:369] (1/2) 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 16:08:02,193 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:08:05,606 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 40700, giga_loss[loss=0.2373, simple_loss=0.3167, pruned_loss=0.07894, over 28515.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3359, pruned_loss=0.09439, over 5707442.28 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3639, pruned_loss=0.1158, over 5684026.43 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3336, pruned_loss=0.09261, over 5709387.88 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:08:29,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:08:34,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8922, 3.7076, 3.5148, 1.7220], device='cuda:1'), covar=tensor([0.0676, 0.0826, 0.0723, 0.2159], device='cuda:1'), in_proj_covar=tensor([0.1204, 0.1111, 0.0944, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 16:09:00,493 INFO [train.py:968] (1/2) Epoch 20, batch 40750, giga_loss[loss=0.2863, simple_loss=0.3595, pruned_loss=0.1066, over 28989.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3405, pruned_loss=0.09638, over 5707299.57 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3644, pruned_loss=0.1159, over 5686899.13 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3373, pruned_loss=0.09418, over 5707143.43 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:09:14,038 INFO [optim.py:369] (1/2) 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 16:09:33,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=907950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:09:36,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=907953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:09:42,772 INFO [train.py:968] (1/2) Epoch 20, batch 40800, giga_loss[loss=0.225, simple_loss=0.3071, pruned_loss=0.07152, over 28609.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3431, pruned_loss=0.09728, over 5709026.43 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3643, pruned_loss=0.116, over 5681803.21 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3402, pruned_loss=0.09509, over 5713580.80 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:10:03,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=907982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:10:25,666 INFO [train.py:968] (1/2) Epoch 20, batch 40850, giga_loss[loss=0.2433, simple_loss=0.3296, pruned_loss=0.07854, over 28979.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3464, pruned_loss=0.09927, over 5704215.47 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3647, pruned_loss=0.1163, over 5675764.06 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3433, pruned_loss=0.09691, over 5713545.51 frames. ], batch size: 155, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:10:39,847 INFO [optim.py:369] (1/2) 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 16:11:13,311 INFO [train.py:968] (1/2) Epoch 20, batch 40900, giga_loss[loss=0.3105, simple_loss=0.3781, pruned_loss=0.1215, over 28930.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3484, pruned_loss=0.1009, over 5702848.91 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5677954.50 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3461, pruned_loss=0.09912, over 5708447.19 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:11:24,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4006, 1.9555, 1.4637, 0.6854], device='cuda:1'), covar=tensor([0.4383, 0.2509, 0.3552, 0.5643], device='cuda:1'), in_proj_covar=tensor([0.1716, 0.1616, 0.1577, 0.1392], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 16:12:03,774 INFO [train.py:968] (1/2) Epoch 20, batch 40950, giga_loss[loss=0.2835, simple_loss=0.3574, pruned_loss=0.1048, over 28942.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3568, pruned_loss=0.1084, over 5688031.50 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 5683017.35 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3543, pruned_loss=0.1064, over 5688569.45 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:12:19,448 INFO [optim.py:369] (1/2) 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 16:12:51,549 INFO [train.py:968] (1/2) Epoch 20, batch 41000, giga_loss[loss=0.3052, simple_loss=0.3755, pruned_loss=0.1174, over 28761.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3627, pruned_loss=0.1125, over 5688111.95 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3647, pruned_loss=0.1162, over 5688613.18 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3606, pruned_loss=0.1109, over 5682998.83 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:13:39,611 INFO [train.py:968] (1/2) Epoch 20, batch 41050, giga_loss[loss=0.322, simple_loss=0.3928, pruned_loss=0.1256, over 28893.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3699, pruned_loss=0.1185, over 5677237.17 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3647, pruned_loss=0.1163, over 5692748.89 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3682, pruned_loss=0.1171, over 5669698.28 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:13:53,520 INFO [optim.py:369] (1/2) 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 16:14:25,508 INFO [train.py:968] (1/2) Epoch 20, batch 41100, giga_loss[loss=0.3101, simple_loss=0.3739, pruned_loss=0.1232, over 28691.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3755, pruned_loss=0.123, over 5678830.27 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3648, pruned_loss=0.1162, over 5695178.78 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3741, pruned_loss=0.122, over 5670609.73 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:15:14,836 INFO [train.py:968] (1/2) Epoch 20, batch 41150, giga_loss[loss=0.3724, simple_loss=0.4208, pruned_loss=0.162, over 28783.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3801, pruned_loss=0.1266, over 5678025.66 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1162, over 5698382.53 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3792, pruned_loss=0.126, over 5668317.23 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:15:33,807 INFO [optim.py:369] (1/2) 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 16:16:07,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-10 16:16:07,931 INFO [train.py:968] (1/2) Epoch 20, batch 41200, giga_loss[loss=0.3433, simple_loss=0.4015, pruned_loss=0.1426, over 28705.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3815, pruned_loss=0.1285, over 5663780.34 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3646, pruned_loss=0.1159, over 5695238.42 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3816, pruned_loss=0.1287, over 5656843.43 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:17:00,483 INFO [train.py:968] (1/2) Epoch 20, batch 41250, giga_loss[loss=0.2846, simple_loss=0.3531, pruned_loss=0.1081, over 28481.00 frames. ], tot_loss[loss=0.3232, simple_loss=0.3835, pruned_loss=0.1315, over 5648268.19 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3644, pruned_loss=0.1158, over 5701048.70 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3844, pruned_loss=0.1322, over 5635971.62 frames. ], batch size: 65, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:17:15,904 INFO [optim.py:369] (1/2) 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 16:17:52,549 INFO [train.py:968] (1/2) Epoch 20, batch 41300, giga_loss[loss=0.3306, simple_loss=0.3899, pruned_loss=0.1357, over 28737.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3847, pruned_loss=0.1333, over 5643551.07 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3637, pruned_loss=0.1153, over 5707415.29 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3869, pruned_loss=0.1349, over 5626187.13 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:18:44,112 INFO [train.py:968] (1/2) Epoch 20, batch 41350, giga_loss[loss=0.3153, simple_loss=0.3905, pruned_loss=0.1201, over 28928.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3904, pruned_loss=0.1387, over 5637365.44 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3639, pruned_loss=0.1154, over 5711393.50 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3924, pruned_loss=0.1403, over 5618707.59 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:18:48,331 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=908515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:18:56,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4950, 3.4621, 1.5759, 1.5321], device='cuda:1'), covar=tensor([0.0938, 0.0333, 0.0875, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0552, 0.0380, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 16:19:03,056 INFO [optim.py:369] (1/2) 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 16:19:20,577 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-10 16:19:41,489 INFO [train.py:968] (1/2) Epoch 20, batch 41400, giga_loss[loss=0.3434, simple_loss=0.3945, pruned_loss=0.1461, over 28594.00 frames. ], tot_loss[loss=0.3381, simple_loss=0.3935, pruned_loss=0.1414, over 5643849.12 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3635, pruned_loss=0.1152, over 5713425.30 frames. ], giga_tot_loss[loss=0.341, simple_loss=0.3958, pruned_loss=0.1431, over 5626515.36 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:19:50,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-10 16:20:35,074 INFO [train.py:968] (1/2) Epoch 20, batch 41450, giga_loss[loss=0.4201, simple_loss=0.434, pruned_loss=0.2031, over 23427.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3933, pruned_loss=0.1425, over 5631981.78 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3633, pruned_loss=0.1151, over 5707545.41 frames. ], giga_tot_loss[loss=0.3425, simple_loss=0.3959, pruned_loss=0.1445, over 5622361.44 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:20:50,924 INFO [optim.py:369] (1/2) 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 16:21:22,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4831, 1.6474, 1.3615, 1.5745], device='cuda:1'), covar=tensor([0.0763, 0.0321, 0.0327, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0068, 0.0062, 0.0105], device='cuda:1')
+2023-03-10 16:21:23,766 INFO [train.py:968] (1/2) Epoch 20, batch 41500, giga_loss[loss=0.262, simple_loss=0.3399, pruned_loss=0.09202, over 28171.00 frames. ], tot_loss[loss=0.336, simple_loss=0.3909, pruned_loss=0.1406, over 5633365.89 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3634, pruned_loss=0.1151, over 5708678.10 frames. ], giga_tot_loss[loss=0.3397, simple_loss=0.3936, pruned_loss=0.1428, over 5622857.98 frames. ], batch size: 65, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:21:51,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1801, 1.2325, 1.0823, 0.9160], device='cuda:1'), covar=tensor([0.1050, 0.0590, 0.1180, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0449, 0.0522, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 16:22:14,830 INFO [train.py:968] (1/2) Epoch 20, batch 41550, giga_loss[loss=0.3277, simple_loss=0.3886, pruned_loss=0.1335, over 28001.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3898, pruned_loss=0.1388, over 5628299.39 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3639, pruned_loss=0.1154, over 5712963.55 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3921, pruned_loss=0.141, over 5614479.04 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:22:31,207 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 41600, giga_loss[loss=0.3219, simple_loss=0.3864, pruned_loss=0.1287, over 28758.00 frames. ], tot_loss[loss=0.3371, simple_loss=0.3924, pruned_loss=0.1409, over 5606623.32 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.364, pruned_loss=0.1157, over 5693352.92 frames. ], giga_tot_loss[loss=0.3403, simple_loss=0.3949, pruned_loss=0.1429, over 5610627.16 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:23:56,053 INFO [train.py:968] (1/2) Epoch 20, batch 41650, giga_loss[loss=0.3214, simple_loss=0.385, pruned_loss=0.1289, over 28894.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3903, pruned_loss=0.139, over 5595037.59 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3638, pruned_loss=0.1156, over 5694877.14 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3932, pruned_loss=0.1415, over 5593909.71 frames. ], batch size: 112, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:24:12,628 INFO [optim.py:369] (1/2) 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 16:24:41,870 INFO [train.py:968] (1/2) Epoch 20, batch 41700, giga_loss[loss=0.319, simple_loss=0.3892, pruned_loss=0.1244, over 29066.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3854, pruned_loss=0.1346, over 5615054.99 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3632, pruned_loss=0.1155, over 5701211.52 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3898, pruned_loss=0.138, over 5603397.34 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:25:09,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=908890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:25:27,810 INFO [train.py:968] (1/2) Epoch 20, batch 41750, giga_loss[loss=0.3476, simple_loss=0.4115, pruned_loss=0.1419, over 28892.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3839, pruned_loss=0.1319, over 5635236.95 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5706564.70 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3882, pruned_loss=0.1352, over 5618817.05 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:25:35,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6044, 1.7883, 1.2933, 1.4521], device='cuda:1'), covar=tensor([0.0976, 0.0597, 0.1126, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0449, 0.0521, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 16:25:35,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5815, 1.8123, 1.5131, 1.5754], device='cuda:1'), covar=tensor([0.2686, 0.2763, 0.3094, 0.2330], device='cuda:1'), in_proj_covar=tensor([0.1484, 0.1080, 0.1312, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 16:25:48,145 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=908930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:25:59,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2343, 0.8354, 0.9006, 1.4054], device='cuda:1'), covar=tensor([0.0734, 0.0425, 0.0370, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 16:26:20,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3054, 4.1564, 3.9516, 2.1706], device='cuda:1'), covar=tensor([0.0613, 0.0694, 0.0718, 0.1801], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.1126, 0.0956, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 16:26:22,037 INFO [train.py:968] (1/2) Epoch 20, batch 41800, giga_loss[loss=0.2792, simple_loss=0.3529, pruned_loss=0.1027, over 28817.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3811, pruned_loss=0.1295, over 5632142.36 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3631, pruned_loss=0.1155, over 5708660.16 frames. ], giga_tot_loss[loss=0.3245, simple_loss=0.3846, pruned_loss=0.1322, over 5616671.09 frames. ], batch size: 284, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:27:10,746 INFO [train.py:968] (1/2) Epoch 20, batch 41850, giga_loss[loss=0.2749, simple_loss=0.3451, pruned_loss=0.1024, over 28383.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3784, pruned_loss=0.1273, over 5621902.03 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1155, over 5702532.14 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3819, pruned_loss=0.1298, over 5612536.30 frames. ], batch size: 71, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:27:28,286 INFO [optim.py:369] (1/2) 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 16:27:33,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=909033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:27:38,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=909036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:28:01,670 INFO [train.py:968] (1/2) Epoch 20, batch 41900, giga_loss[loss=0.3173, simple_loss=0.3787, pruned_loss=0.128, over 28263.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3753, pruned_loss=0.1247, over 5646410.51 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3627, pruned_loss=0.1155, over 5706078.51 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3785, pruned_loss=0.1269, over 5634406.58 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:28:04,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=909065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:28:31,593 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-10 16:28:49,248 INFO [train.py:968] (1/2) Epoch 20, batch 41950, giga_loss[loss=0.3476, simple_loss=0.3959, pruned_loss=0.1496, over 27586.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3757, pruned_loss=0.1251, over 5653025.15 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5709315.45 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3787, pruned_loss=0.1272, over 5639862.24 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:29:05,648 INFO [optim.py:369] (1/2) 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 16:29:12,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-10 16:29:15,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6462, 1.9504, 1.9954, 1.6693], device='cuda:1'), covar=tensor([0.1618, 0.1634, 0.1818, 0.1779], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0751, 0.0714, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 16:29:29,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0219, 1.1991, 3.3196, 2.9750], device='cuda:1'), covar=tensor([0.1802, 0.2764, 0.0571, 0.0953], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0643, 0.0955, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 16:29:39,989 INFO [train.py:968] (1/2) Epoch 20, batch 42000, giga_loss[loss=0.2876, simple_loss=0.3553, pruned_loss=0.1099, over 28599.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3744, pruned_loss=0.1238, over 5651925.18 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5713104.54 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3772, pruned_loss=0.1259, over 5636452.47 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:29:39,989 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 16:29:48,874 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 16:30:33,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 16:30:46,597 INFO [train.py:968] (1/2) Epoch 20, batch 42050, giga_loss[loss=0.3592, simple_loss=0.3947, pruned_loss=0.1618, over 23795.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3732, pruned_loss=0.1217, over 5641353.14 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.1151, over 5715102.64 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3756, pruned_loss=0.1235, over 5626947.77 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:31:05,762 INFO [optim.py:369] (1/2) 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 16:31:11,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8853, 3.7011, 3.5146, 1.9095], device='cuda:1'), covar=tensor([0.0769, 0.0938, 0.0992, 0.2066], device='cuda:1'), in_proj_covar=tensor([0.1224, 0.1136, 0.0963, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 16:31:41,470 INFO [train.py:968] (1/2) Epoch 20, batch 42100, giga_loss[loss=0.307, simple_loss=0.3897, pruned_loss=0.1121, over 28943.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3742, pruned_loss=0.1198, over 5656395.02 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5714085.64 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3768, pruned_loss=0.1215, over 5644757.89 frames. ], batch size: 213, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:32:22,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=909305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:32:26,930 INFO [train.py:968] (1/2) Epoch 20, batch 42150, giga_loss[loss=0.3143, simple_loss=0.3853, pruned_loss=0.1216, over 28967.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3744, pruned_loss=0.1198, over 5670730.50 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1143, over 5721325.06 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3775, pruned_loss=0.1218, over 5653336.23 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 16:32:47,351 INFO [optim.py:369] (1/2) 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 16:33:16,640 INFO [train.py:968] (1/2) Epoch 20, batch 42200, giga_loss[loss=0.2822, simple_loss=0.3613, pruned_loss=0.1015, over 28857.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3754, pruned_loss=0.1211, over 5661721.33 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1145, over 5713397.19 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3778, pruned_loss=0.1226, over 5654023.69 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:34:05,901 INFO [train.py:968] (1/2) Epoch 20, batch 42250, giga_loss[loss=0.32, simple_loss=0.383, pruned_loss=0.1285, over 28545.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3742, pruned_loss=0.1213, over 5669302.16 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 5715496.77 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3767, pruned_loss=0.1228, over 5660882.81 frames. ], batch size: 336, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:34:07,440 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=909412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:34:25,472 INFO [optim.py:369] (1/2) 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 16:34:35,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3992, 1.5394, 1.4980, 1.4394], device='cuda:1'), covar=tensor([0.1590, 0.1950, 0.2089, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.0473, 0.0749, 0.0713, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 16:34:43,961 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=909448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:34:48,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=909451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:34:57,753 INFO [train.py:968] (1/2) Epoch 20, batch 42300, giga_loss[loss=0.3573, simple_loss=0.4045, pruned_loss=0.155, over 27986.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3738, pruned_loss=0.1228, over 5664161.22 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3609, pruned_loss=0.1142, over 5717427.95 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3758, pruned_loss=0.124, over 5655300.88 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:35:14,772 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=909480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:35:29,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0384, 1.3599, 5.2512, 3.6609], device='cuda:1'), covar=tensor([0.1505, 0.2660, 0.0467, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0643, 0.0954, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 16:35:45,395 INFO [train.py:968] (1/2) Epoch 20, batch 42350, giga_loss[loss=0.2907, simple_loss=0.3707, pruned_loss=0.1054, over 28746.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3719, pruned_loss=0.1213, over 5662122.89 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1142, over 5714360.15 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3741, pruned_loss=0.1227, over 5655153.54 frames. ], batch size: 92, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:35:57,161 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6112, 4.4492, 4.2249, 1.9326], device='cuda:1'), covar=tensor([0.0553, 0.0704, 0.0760, 0.2125], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1135, 0.0961, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 16:36:06,890 INFO [optim.py:369] (1/2) 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 16:36:35,406 INFO [train.py:968] (1/2) Epoch 20, batch 42400, giga_loss[loss=0.2972, simple_loss=0.3671, pruned_loss=0.1137, over 28907.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3728, pruned_loss=0.1208, over 5675253.17 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3612, pruned_loss=0.1146, over 5719163.63 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3745, pruned_loss=0.1217, over 5664412.48 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:36:35,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-10 16:37:21,905 INFO [train.py:968] (1/2) Epoch 20, batch 42450, giga_loss[loss=0.2925, simple_loss=0.3785, pruned_loss=0.1032, over 28970.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3732, pruned_loss=0.1203, over 5685275.85 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5722072.78 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3746, pruned_loss=0.121, over 5673526.09 frames. ], batch size: 164, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:37:32,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5514, 1.5710, 1.7511, 1.3976], device='cuda:1'), covar=tensor([0.1359, 0.2051, 0.1130, 0.1406], device='cuda:1'), in_proj_covar=tensor([0.0891, 0.0697, 0.0935, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 16:37:43,529 INFO [optim.py:369] (1/2) 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 16:38:09,614 INFO [train.py:968] (1/2) Epoch 20, batch 42500, giga_loss[loss=0.31, simple_loss=0.3735, pruned_loss=0.1233, over 28737.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3728, pruned_loss=0.1205, over 5674208.08 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3616, pruned_loss=0.1148, over 5722971.60 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.374, pruned_loss=0.1211, over 5662517.07 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:38:52,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5283, 1.4068, 1.5567, 1.1668], device='cuda:1'), covar=tensor([0.1725, 0.3386, 0.1434, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0697, 0.0935, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 16:38:53,887 INFO [train.py:968] (1/2) Epoch 20, batch 42550, giga_loss[loss=0.2729, simple_loss=0.3388, pruned_loss=0.1035, over 28994.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3719, pruned_loss=0.1204, over 5665953.70 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3617, pruned_loss=0.115, over 5707174.27 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3731, pruned_loss=0.1209, over 5670480.64 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:39:15,864 INFO [optim.py:369] (1/2) 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 16:39:46,436 INFO [train.py:968] (1/2) Epoch 20, batch 42600, giga_loss[loss=0.3046, simple_loss=0.379, pruned_loss=0.1151, over 28909.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3714, pruned_loss=0.1205, over 5664268.19 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3619, pruned_loss=0.115, over 5709296.73 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3724, pruned_loss=0.121, over 5665085.12 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:40:12,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=909787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:40:33,361 INFO [train.py:968] (1/2) Epoch 20, batch 42650, giga_loss[loss=0.3035, simple_loss=0.3713, pruned_loss=0.1178, over 28978.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 5669693.99 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3615, pruned_loss=0.1146, over 5706969.77 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3698, pruned_loss=0.1199, over 5670236.60 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:40:54,573 INFO [optim.py:369] (1/2) 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 16:40:54,827 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=909832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:40:58,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5134, 2.2246, 1.7024, 0.7810], device='cuda:1'), covar=tensor([0.4836, 0.2541, 0.3552, 0.5680], device='cuda:1'), in_proj_covar=tensor([0.1736, 0.1641, 0.1594, 0.1410], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 16:41:12,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7912, 2.2153, 1.7321, 2.0271], device='cuda:1'), covar=tensor([0.0670, 0.0252, 0.0306, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 16:41:19,973 INFO [train.py:968] (1/2) Epoch 20, batch 42700, giga_loss[loss=0.2659, simple_loss=0.3362, pruned_loss=0.09783, over 28650.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3661, pruned_loss=0.1177, over 5679201.80 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1143, over 5711681.33 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3678, pruned_loss=0.1188, over 5674372.32 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:42:13,320 INFO [train.py:968] (1/2) Epoch 20, batch 42750, giga_loss[loss=0.292, simple_loss=0.3665, pruned_loss=0.1087, over 28909.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3654, pruned_loss=0.1177, over 5678030.31 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3611, pruned_loss=0.1142, over 5714294.48 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3668, pruned_loss=0.1188, over 5671456.38 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:42:33,862 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=909930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:42:34,722 INFO [optim.py:369] (1/2) 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 16:42:36,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=909933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:42:48,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7253, 1.9166, 1.6128, 1.6214], device='cuda:1'), covar=tensor([0.1926, 0.2494, 0.2450, 0.2375], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0749, 0.0713, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 16:43:04,701 INFO [train.py:968] (1/2) Epoch 20, batch 42800, giga_loss[loss=0.3204, simple_loss=0.3818, pruned_loss=0.1295, over 28212.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3652, pruned_loss=0.1183, over 5661253.23 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.361, pruned_loss=0.1141, over 5714810.48 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3665, pruned_loss=0.1193, over 5654820.13 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:43:07,804 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=909962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:43:36,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-10 16:43:53,497 INFO [train.py:968] (1/2) Epoch 20, batch 42850, giga_loss[loss=0.3578, simple_loss=0.395, pruned_loss=0.1602, over 26667.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3662, pruned_loss=0.1184, over 5664601.64 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3613, pruned_loss=0.1142, over 5716764.47 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.367, pruned_loss=0.1191, over 5657466.93 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:44:16,756 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 42900, libri_loss[loss=0.3412, simple_loss=0.397, pruned_loss=0.1427, over 19615.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3678, pruned_loss=0.1187, over 5666247.97 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1144, over 5710566.61 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1192, over 5665849.98 frames. ], batch size: 187, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:45:25,285 INFO [train.py:968] (1/2) Epoch 20, batch 42950, libri_loss[loss=0.2365, simple_loss=0.3101, pruned_loss=0.08146, over 29679.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.368, pruned_loss=0.1185, over 5665809.23 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3613, pruned_loss=0.1142, over 5707686.50 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3689, pruned_loss=0.1193, over 5666444.47 frames. ], batch size: 73, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:45:48,189 INFO [optim.py:369] (1/2) 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 16:46:21,361 INFO [train.py:968] (1/2) Epoch 20, batch 43000, giga_loss[loss=0.2801, simple_loss=0.3547, pruned_loss=0.1027, over 28784.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3671, pruned_loss=0.1176, over 5679696.09 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3608, pruned_loss=0.1139, over 5711843.47 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3685, pruned_loss=0.1186, over 5675874.76 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:46:38,629 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=910179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:47:04,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:47:09,559 INFO [train.py:968] (1/2) Epoch 20, batch 43050, giga_loss[loss=0.3067, simple_loss=0.3649, pruned_loss=0.1242, over 28927.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3706, pruned_loss=0.1204, over 5684487.65 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3607, pruned_loss=0.1137, over 5715508.55 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3719, pruned_loss=0.1214, over 5677287.72 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:47:12,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 20, batch 43100, giga_loss[loss=0.2847, simple_loss=0.3505, pruned_loss=0.1094, over 28943.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3716, pruned_loss=0.1225, over 5687674.76 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3609, pruned_loss=0.1138, over 5716241.21 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3726, pruned_loss=0.1233, over 5681071.41 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:48:56,383 INFO [train.py:968] (1/2) Epoch 20, batch 43150, giga_loss[loss=0.3123, simple_loss=0.3788, pruned_loss=0.1229, over 28594.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3723, pruned_loss=0.1241, over 5681961.93 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3612, pruned_loss=0.114, over 5715446.04 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3731, pruned_loss=0.1249, over 5676294.94 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:49:10,701 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-10 16:49:21,900 INFO [optim.py:369] (1/2) 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 16:49:40,258 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=910350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:49:42,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=910353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:49:45,928 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-10 16:49:50,670 INFO [train.py:968] (1/2) Epoch 20, batch 43200, giga_loss[loss=0.3008, simple_loss=0.3657, pruned_loss=0.118, over 27871.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3738, pruned_loss=0.126, over 5664652.24 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3612, pruned_loss=0.114, over 5716426.81 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3746, pruned_loss=0.1266, over 5659041.40 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:50:09,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=910382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:50:19,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6612, 1.7449, 1.2667, 1.3797], device='cuda:1'), covar=tensor([0.0904, 0.0600, 0.1104, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0448, 0.0518, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 16:50:34,696 INFO [train.py:968] (1/2) Epoch 20, batch 43250, giga_loss[loss=0.4061, simple_loss=0.4432, pruned_loss=0.1845, over 26427.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3731, pruned_loss=0.1257, over 5662455.32 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5711676.11 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3738, pruned_loss=0.1264, over 5660850.02 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:50:54,889 INFO [optim.py:369] (1/2) 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 16:51:17,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7906, 1.8331, 1.9375, 1.5273], device='cuda:1'), covar=tensor([0.2037, 0.2548, 0.1600, 0.1885], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0699, 0.0935, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 16:51:22,965 INFO [train.py:968] (1/2) Epoch 20, batch 43300, giga_loss[loss=0.2802, simple_loss=0.3621, pruned_loss=0.09913, over 28535.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3725, pruned_loss=0.1236, over 5665713.86 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1143, over 5712629.73 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5663458.32 frames. ], batch size: 85, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:52:13,976 INFO [train.py:968] (1/2) Epoch 20, batch 43350, giga_loss[loss=0.3244, simple_loss=0.3634, pruned_loss=0.1427, over 26621.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3704, pruned_loss=0.1212, over 5657331.78 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3618, pruned_loss=0.1144, over 5706148.46 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3707, pruned_loss=0.1216, over 5661027.49 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:52:33,173 INFO [optim.py:369] (1/2) 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 16:52:53,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:53:00,303 INFO [train.py:968] (1/2) Epoch 20, batch 43400, giga_loss[loss=0.2709, simple_loss=0.345, pruned_loss=0.09838, over 28642.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5663673.59 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3615, pruned_loss=0.1142, over 5709251.16 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3694, pruned_loss=0.1212, over 5663137.57 frames. ], batch size: 60, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:53:24,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:53:25,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5567, 2.0912, 1.3223, 0.7999], device='cuda:1'), covar=tensor([0.6448, 0.3853, 0.3020, 0.6374], device='cuda:1'), in_proj_covar=tensor([0.1719, 0.1632, 0.1581, 0.1399], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 16:53:29,626 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=910592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:53:47,496 INFO [train.py:968] (1/2) Epoch 20, batch 43450, giga_loss[loss=0.3414, simple_loss=0.3737, pruned_loss=0.1545, over 23585.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3674, pruned_loss=0.1204, over 5663337.20 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3617, pruned_loss=0.1142, over 5713311.68 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3678, pruned_loss=0.121, over 5658587.71 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:54:09,504 INFO [optim.py:369] (1/2) 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 16:54:32,435 INFO [train.py:968] (1/2) Epoch 20, batch 43500, giga_loss[loss=0.3203, simple_loss=0.3808, pruned_loss=0.1298, over 28865.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3669, pruned_loss=0.12, over 5662970.24 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.1141, over 5706620.96 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5664303.45 frames. ], batch size: 227, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:54:37,479 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 16:54:38,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 16:54:41,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6152, 1.6819, 1.8243, 1.4131], device='cuda:1'), covar=tensor([0.1776, 0.2474, 0.1411, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0700, 0.0937, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 16:54:52,849 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=910682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:09,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=910697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:12,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=910700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:19,666 INFO [train.py:968] (1/2) Epoch 20, batch 43550, libri_loss[loss=0.2359, simple_loss=0.317, pruned_loss=0.07741, over 29524.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3697, pruned_loss=0.1216, over 5661040.36 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5704827.48 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3709, pruned_loss=0.1228, over 5661780.04 frames. ], batch size: 80, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 16:55:37,646 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=910729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:39,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:55:41,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=910734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:55:48,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2396, 0.8439, 0.8557, 1.3524], device='cuda:1'), covar=tensor([0.0816, 0.0397, 0.0396, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 16:56:00,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1198, 3.0740, 1.2260, 1.4435], device='cuda:1'), covar=tensor([0.1110, 0.0504, 0.1048, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0556, 0.0382, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 16:56:08,673 INFO [train.py:968] (1/2) Epoch 20, batch 43600, giga_loss[loss=0.3056, simple_loss=0.3834, pruned_loss=0.1139, over 28749.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3724, pruned_loss=0.1203, over 5666645.43 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3614, pruned_loss=0.1139, over 5705924.88 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3734, pruned_loss=0.1213, over 5665431.90 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:56:10,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=910763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 16:57:01,377 INFO [train.py:968] (1/2) Epoch 20, batch 43650, giga_loss[loss=0.2809, simple_loss=0.3629, pruned_loss=0.09945, over 28940.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.373, pruned_loss=0.1201, over 5666788.43 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5708473.30 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3744, pruned_loss=0.1211, over 5662052.78 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:57:23,977 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 43700, giga_loss[loss=0.3464, simple_loss=0.3985, pruned_loss=0.1471, over 27904.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3753, pruned_loss=0.1221, over 5649976.04 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3608, pruned_loss=0.114, over 5690445.41 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3773, pruned_loss=0.1232, over 5660808.29 frames. ], batch size: 412, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:58:37,886 INFO [train.py:968] (1/2) Epoch 20, batch 43750, giga_loss[loss=0.3752, simple_loss=0.4053, pruned_loss=0.1725, over 23584.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3774, pruned_loss=0.1242, over 5649339.80 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3611, pruned_loss=0.1141, over 5690119.50 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.379, pruned_loss=0.1251, over 5657815.61 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:58:47,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5521, 2.2417, 1.6460, 0.8122], device='cuda:1'), covar=tensor([0.6166, 0.3202, 0.4412, 0.6578], device='cuda:1'), in_proj_covar=tensor([0.1735, 0.1638, 0.1585, 0.1405], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 16:58:58,042 INFO [optim.py:369] (1/2) 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 16:59:23,298 INFO [train.py:968] (1/2) Epoch 20, batch 43800, giga_loss[loss=0.3306, simple_loss=0.3882, pruned_loss=0.1365, over 28566.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3775, pruned_loss=0.1251, over 5646482.73 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3614, pruned_loss=0.1144, over 5674350.75 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3787, pruned_loss=0.1257, over 5666024.61 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 16:59:27,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=910967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:00:13,391 INFO [train.py:968] (1/2) Epoch 20, batch 43850, giga_loss[loss=0.4217, simple_loss=0.4393, pruned_loss=0.202, over 26587.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3759, pruned_loss=0.125, over 5647545.20 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5678269.17 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3771, pruned_loss=0.1257, over 5658909.82 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:00:35,434 INFO [optim.py:369] (1/2) 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 17:00:57,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:01:01,050 INFO [train.py:968] (1/2) Epoch 20, batch 43900, giga_loss[loss=0.3349, simple_loss=0.3871, pruned_loss=0.1414, over 28713.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3729, pruned_loss=0.1235, over 5662132.43 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 5684724.66 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3748, pruned_loss=0.1247, over 5664664.26 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:01:03,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5243, 1.6208, 1.2884, 1.2211], device='cuda:1'), covar=tensor([0.0944, 0.0567, 0.1080, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0451, 0.0520, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 17:01:50,182 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:01:50,586 INFO [train.py:968] (1/2) Epoch 20, batch 43950, giga_loss[loss=0.3172, simple_loss=0.3718, pruned_loss=0.1312, over 27569.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3721, pruned_loss=0.1236, over 5659461.81 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.114, over 5683634.95 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1248, over 5661822.46 frames. ], batch size: 472, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:01:56,299 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:02:15,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3356, 2.2153, 2.1812, 2.0282], device='cuda:1'), covar=tensor([0.1745, 0.2510, 0.2105, 0.2268], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0749, 0.0712, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 17:02:16,272 INFO [optim.py:369] (1/2) 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 17:02:24,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:02:29,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 17:02:41,585 INFO [train.py:968] (1/2) Epoch 20, batch 44000, giga_loss[loss=0.2955, simple_loss=0.369, pruned_loss=0.111, over 28878.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1236, over 5674792.82 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3609, pruned_loss=0.114, over 5688531.27 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3741, pruned_loss=0.1248, over 5671871.45 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 17:03:19,698 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:03:22,859 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 44050, giga_loss[loss=0.3108, simple_loss=0.3674, pruned_loss=0.1271, over 28928.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.371, pruned_loss=0.1236, over 5671495.42 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5694612.88 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3738, pruned_loss=0.1254, over 5662817.87 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 8.0
+2023-03-10 17:03:47,984 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:03:50,166 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 44100, giga_loss[loss=0.2901, simple_loss=0.3564, pruned_loss=0.1119, over 28596.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3686, pruned_loss=0.1221, over 5677261.01 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.36, pruned_loss=0.1135, over 5697898.69 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3708, pruned_loss=0.1236, over 5667095.80 frames. ], batch size: 242, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:05:06,081 INFO [train.py:968] (1/2) Epoch 20, batch 44150, giga_loss[loss=0.3795, simple_loss=0.4186, pruned_loss=0.1702, over 26694.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3688, pruned_loss=0.1219, over 5676220.05 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3601, pruned_loss=0.1136, over 5698950.20 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3704, pruned_loss=0.123, over 5667249.64 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:05:30,747 INFO [optim.py:369] (1/2) 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 17:06:01,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6200, 1.5848, 1.8602, 1.4623], device='cuda:1'), covar=tensor([0.1322, 0.1859, 0.1080, 0.1428], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0702, 0.0940, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 17:06:01,524 INFO [train.py:968] (1/2) Epoch 20, batch 44200, giga_loss[loss=0.3799, simple_loss=0.4031, pruned_loss=0.1783, over 23496.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3707, pruned_loss=0.1226, over 5671102.88 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3601, pruned_loss=0.1135, over 5701614.03 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3722, pruned_loss=0.1237, over 5661543.08 frames. ], batch size: 705, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:06:04,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5019, 1.6282, 1.2023, 1.2053], device='cuda:1'), covar=tensor([0.0831, 0.0478, 0.0912, 0.1241], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0450, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 17:06:49,710 INFO [train.py:968] (1/2) Epoch 20, batch 44250, giga_loss[loss=0.4302, simple_loss=0.4463, pruned_loss=0.2071, over 26741.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 5680929.63 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3597, pruned_loss=0.1134, over 5704922.60 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3739, pruned_loss=0.1252, over 5669999.85 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:06:52,450 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:07:16,149 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 44300, giga_loss[loss=0.3051, simple_loss=0.3844, pruned_loss=0.1129, over 28880.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3727, pruned_loss=0.1235, over 5676824.58 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3602, pruned_loss=0.1137, over 5706924.32 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1242, over 5666144.54 frames. ], batch size: 199, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:07:52,985 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:08:22,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:08:27,249 INFO [train.py:968] (1/2) Epoch 20, batch 44350, libri_loss[loss=0.2799, simple_loss=0.3509, pruned_loss=0.1045, over 29552.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3738, pruned_loss=0.1219, over 5678661.10 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3605, pruned_loss=0.1139, over 5708891.82 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3747, pruned_loss=0.1225, over 5667482.51 frames. ], batch size: 77, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:08:28,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1613, 1.5994, 1.3619, 1.3439], device='cuda:1'), covar=tensor([0.2329, 0.2157, 0.2603, 0.2505], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0747, 0.0709, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 17:08:47,178 INFO [optim.py:369] (1/2) 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 17:08:59,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6601, 1.7442, 1.8586, 1.4265], device='cuda:1'), covar=tensor([0.2098, 0.2702, 0.1748, 0.1986], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0702, 0.0939, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 17:09:02,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4409, 1.6726, 1.6068, 1.3450], device='cuda:1'), covar=tensor([0.2998, 0.2490, 0.1940, 0.2584], device='cuda:1'), in_proj_covar=tensor([0.1963, 0.1877, 0.1826, 0.1965], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 17:09:06,527 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:09:08,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3189, 1.8442, 1.5556, 1.6912], device='cuda:1'), covar=tensor([0.0832, 0.0294, 0.0342, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 17:09:13,009 INFO [train.py:968] (1/2) Epoch 20, batch 44400, giga_loss[loss=0.3198, simple_loss=0.3958, pruned_loss=0.1219, over 28590.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.376, pruned_loss=0.1214, over 5677135.92 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1142, over 5702883.88 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3767, pruned_loss=0.1218, over 5672993.73 frames. ], batch size: 78, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:10:02,775 INFO [train.py:968] (1/2) Epoch 20, batch 44450, giga_loss[loss=0.5275, simple_loss=0.5063, pruned_loss=0.2743, over 26510.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3786, pruned_loss=0.1234, over 5681154.36 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3606, pruned_loss=0.1141, over 5699292.05 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3797, pruned_loss=0.124, over 5681179.16 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:10:10,021 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=911617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:10:27,924 INFO [optim.py:369] (1/2) 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 17:10:38,261 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:10:42,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=911651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:10:51,707 INFO [train.py:968] (1/2) Epoch 20, batch 44500, libri_loss[loss=0.3085, simple_loss=0.3722, pruned_loss=0.1224, over 29208.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3804, pruned_loss=0.1261, over 5674378.70 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3604, pruned_loss=0.114, over 5705651.82 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3822, pruned_loss=0.127, over 5667797.99 frames. ], batch size: 97, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:11:10,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:11:43,601 INFO [train.py:968] (1/2) Epoch 20, batch 44550, giga_loss[loss=0.2927, simple_loss=0.362, pruned_loss=0.1117, over 28929.00 frames. ], tot_loss[loss=0.3189, simple_loss=0.3817, pruned_loss=0.128, over 5660604.36 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3606, pruned_loss=0.1141, over 5707803.22 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.3834, pruned_loss=0.129, over 5652580.16 frames. ], batch size: 145, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:12:08,630 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 44600, giga_loss[loss=0.3309, simple_loss=0.3921, pruned_loss=0.1349, over 28548.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.381, pruned_loss=0.1278, over 5665115.84 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.1141, over 5709632.38 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3828, pruned_loss=0.1289, over 5656111.10 frames. ], batch size: 307, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:12:47,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 17:12:51,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:13:09,946 INFO [train.py:968] (1/2) Epoch 20, batch 44650, libri_loss[loss=0.2376, simple_loss=0.3058, pruned_loss=0.08468, over 29655.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3788, pruned_loss=0.1257, over 5668488.92 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3601, pruned_loss=0.1138, over 5709136.14 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3817, pruned_loss=0.1274, over 5659692.78 frames. ], batch size: 73, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:13:34,161 INFO [optim.py:369] (1/2) 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 17:13:36,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5773, 1.8071, 1.5231, 1.5460], device='cuda:1'), covar=tensor([0.2283, 0.2244, 0.2373, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.1485, 0.1076, 0.1313, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 17:13:42,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:13:56,234 INFO [train.py:968] (1/2) Epoch 20, batch 44700, giga_loss[loss=0.2908, simple_loss=0.3662, pruned_loss=0.1076, over 28277.00 frames. ], tot_loss[loss=0.313, simple_loss=0.379, pruned_loss=0.1235, over 5674284.65 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3605, pruned_loss=0.1142, over 5712211.37 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3814, pruned_loss=0.1249, over 5663342.65 frames. ], batch size: 368, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:14:44,071 INFO [train.py:968] (1/2) Epoch 20, batch 44750, giga_loss[loss=0.3177, simple_loss=0.3817, pruned_loss=0.1269, over 28706.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3807, pruned_loss=0.1243, over 5664292.28 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3609, pruned_loss=0.1146, over 5705605.78 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3826, pruned_loss=0.1252, over 5661450.35 frames. ], batch size: 262, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:15:06,093 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:15:06,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:15:11,596 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:15:15,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6705, 4.4597, 4.2820, 1.9607], device='cuda:1'), covar=tensor([0.0623, 0.0817, 0.0849, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1229, 0.1142, 0.0967, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 17:15:37,898 INFO [train.py:968] (1/2) Epoch 20, batch 44800, giga_loss[loss=0.2899, simple_loss=0.365, pruned_loss=0.1074, over 29010.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3799, pruned_loss=0.1243, over 5671692.61 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3603, pruned_loss=0.1143, over 5709985.32 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3824, pruned_loss=0.1254, over 5664647.71 frames. ], batch size: 128, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:15:42,606 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=911963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:15:49,405 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-10 17:15:50,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-10 17:16:08,395 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=911991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:09,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=911992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:10,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=911994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:25,917 INFO [train.py:968] (1/2) Epoch 20, batch 44850, giga_loss[loss=0.2861, simple_loss=0.3562, pruned_loss=0.1081, over 28960.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3786, pruned_loss=0.1241, over 5681493.44 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3599, pruned_loss=0.114, over 5712277.11 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3812, pruned_loss=0.1254, over 5673386.94 frames. ], batch size: 106, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:16:37,984 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=912023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:16:49,128 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 44900, giga_loss[loss=0.2723, simple_loss=0.3392, pruned_loss=0.1027, over 28933.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3774, pruned_loss=0.1246, over 5651508.54 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1142, over 5703099.83 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3796, pruned_loss=0.1258, over 5652560.41 frames. ], batch size: 136, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:17:27,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=912073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:17:29,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=912076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:17:57,494 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 20, batch 44950, giga_loss[loss=0.3047, simple_loss=0.3635, pruned_loss=0.123, over 28906.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3759, pruned_loss=0.1244, over 5659213.60 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3604, pruned_loss=0.1145, over 5707117.68 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3778, pruned_loss=0.1253, over 5655514.54 frames. ], batch size: 112, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:18:13,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6569, 3.5079, 3.3381, 1.9902], device='cuda:1'), covar=tensor([0.0713, 0.0845, 0.0833, 0.1796], device='cuda:1'), in_proj_covar=tensor([0.1231, 0.1146, 0.0970, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 17:18:26,483 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=912135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:18:27,452 INFO [optim.py:369] (1/2) 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 17:18:29,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=912138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:18:40,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 17:18:53,899 INFO [train.py:968] (1/2) Epoch 20, batch 45000, giga_loss[loss=0.3048, simple_loss=0.365, pruned_loss=0.1223, over 28765.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1234, over 5644202.36 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3604, pruned_loss=0.1145, over 5691438.57 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3754, pruned_loss=0.1243, over 5653264.02 frames. ], batch size: 119, lr: 1.58e-03, grad_scale: 4.0
+2023-03-10 17:18:53,899 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 17:19:01,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4947, 1.8694, 1.4777, 1.3557], device='cuda:1'), covar=tensor([0.3014, 0.2888, 0.3087, 0.2462], device='cuda:1'), in_proj_covar=tensor([0.1486, 0.1078, 0.1313, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 17:19:02,810 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 17:19:07,987 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=912167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:19:48,472 INFO [train.py:968] (1/2) Epoch 20, batch 45050, libri_loss[loss=0.3621, simple_loss=0.4179, pruned_loss=0.1532, over 19440.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3726, pruned_loss=0.1235, over 5628840.61 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3608, pruned_loss=0.1149, over 5669565.23 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.374, pruned_loss=0.1242, over 5655387.22 frames. ], batch size: 186, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:20:09,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1951, 2.6083, 2.1654, 2.4038], device='cuda:1'), covar=tensor([0.0513, 0.0216, 0.0231, 0.0566], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 17:20:11,475 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 20, batch 45100, giga_loss[loss=0.4541, simple_loss=0.4596, pruned_loss=0.2243, over 26609.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1235, over 5598510.98 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3616, pruned_loss=0.1155, over 5624137.66 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3731, pruned_loss=0.1236, over 5657789.33 frames. ], batch size: 555, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:21:19,585 INFO [train.py:968] (1/2) Epoch 20, batch 45150, giga_loss[loss=0.2776, simple_loss=0.3599, pruned_loss=0.09763, over 28966.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.37, pruned_loss=0.1208, over 5545086.62 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3627, pruned_loss=0.1164, over 5557322.42 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3698, pruned_loss=0.1203, over 5652297.00 frames. ], batch size: 174, lr: 1.58e-03, grad_scale: 2.0
+2023-03-10 17:21:22,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 17:21:43,534 INFO [optim.py:369] (1/2) 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 17:21:59,492 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-10 17:23:23,371 INFO [train.py:968] (1/2) Epoch 21, batch 50, giga_loss[loss=0.2969, simple_loss=0.3764, pruned_loss=0.1088, over 28692.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3633, pruned_loss=0.1033, over 1267840.73 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.325, pruned_loss=0.08326, over 202134.76 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3694, pruned_loss=0.1065, over 1104715.67 frames. ], batch size: 262, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:24:00,068 INFO [optim.py:369] (1/2) 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 17:24:14,474 INFO [train.py:968] (1/2) Epoch 21, batch 100, giga_loss[loss=0.2488, simple_loss=0.3323, pruned_loss=0.08264, over 28593.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3623, pruned_loss=0.1047, over 2232992.26 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.335, pruned_loss=0.08906, over 332322.10 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3657, pruned_loss=0.1066, over 2020233.61 frames. ], batch size: 307, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:24:34,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 17:24:45,911 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-10 17:24:50,099 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 150, giga_loss[loss=0.2294, simple_loss=0.3127, pruned_loss=0.07303, over 28640.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3468, pruned_loss=0.09672, over 3002312.28 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3367, pruned_loss=0.08813, over 415637.62 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.348, pruned_loss=0.09772, over 2790548.87 frames. ], batch size: 242, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:25:21,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2851, 1.5053, 1.4128, 1.4912], device='cuda:1'), covar=tensor([0.0755, 0.0409, 0.0341, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 17:25:33,584 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 200, giga_loss[loss=0.2168, simple_loss=0.2945, pruned_loss=0.0696, over 29033.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3342, pruned_loss=0.09067, over 3605334.01 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3406, pruned_loss=0.09007, over 629523.14 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3338, pruned_loss=0.09096, over 3342486.08 frames. ], batch size: 136, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:26:27,258 INFO [train.py:968] (1/2) Epoch 21, batch 250, giga_loss[loss=0.2005, simple_loss=0.2784, pruned_loss=0.06134, over 28947.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3251, pruned_loss=0.08619, over 4064339.17 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3432, pruned_loss=0.09091, over 729880.59 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3235, pruned_loss=0.08592, over 3822837.47 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:27:00,311 INFO [optim.py:369] (1/2) 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 17:27:12,206 INFO [train.py:968] (1/2) Epoch 21, batch 300, giga_loss[loss=0.1992, simple_loss=0.2771, pruned_loss=0.06066, over 28874.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.3163, pruned_loss=0.08217, over 4424947.26 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3429, pruned_loss=0.0904, over 832572.49 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3141, pruned_loss=0.0817, over 4203629.96 frames. ], batch size: 213, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:27:45,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=912684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:27:59,554 INFO [train.py:968] (1/2) Epoch 21, batch 350, giga_loss[loss=0.2059, simple_loss=0.2848, pruned_loss=0.06354, over 28738.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3097, pruned_loss=0.07912, over 4694969.47 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3425, pruned_loss=0.08985, over 971102.69 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3069, pruned_loss=0.07842, over 4491531.76 frames. ], batch size: 262, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:28:29,186 INFO [optim.py:369] (1/2) 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 17:28:38,616 INFO [train.py:968] (1/2) Epoch 21, batch 400, giga_loss[loss=0.2057, simple_loss=0.2839, pruned_loss=0.06376, over 29007.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3063, pruned_loss=0.07754, over 4916623.61 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3429, pruned_loss=0.09028, over 1107082.25 frames. ], giga_tot_loss[loss=0.2279, simple_loss=0.3028, pruned_loss=0.07651, over 4732709.73 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:28:42,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4817, 4.3189, 4.0800, 1.9354], device='cuda:1'), covar=tensor([0.0550, 0.0742, 0.0786, 0.2032], device='cuda:1'), in_proj_covar=tensor([0.1211, 0.1125, 0.0953, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-10 17:29:20,866 INFO [train.py:968] (1/2) Epoch 21, batch 450, giga_loss[loss=0.2424, simple_loss=0.3134, pruned_loss=0.08565, over 29044.00 frames. ], tot_loss[loss=0.2297, simple_loss=0.3052, pruned_loss=0.0771, over 5096239.55 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3421, pruned_loss=0.09021, over 1272575.24 frames. ], giga_tot_loss[loss=0.2264, simple_loss=0.3012, pruned_loss=0.07579, over 4919549.09 frames. ], batch size: 106, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:29:55,110 INFO [optim.py:369] (1/2) 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 17:30:05,752 INFO [train.py:968] (1/2) Epoch 21, batch 500, giga_loss[loss=0.209, simple_loss=0.2877, pruned_loss=0.06511, over 29071.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3021, pruned_loss=0.07559, over 5226811.72 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3412, pruned_loss=0.08928, over 1387094.67 frames. ], giga_tot_loss[loss=0.2235, simple_loss=0.2981, pruned_loss=0.07442, over 5069094.96 frames. ], batch size: 128, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:30:19,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=912864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:30:53,167 INFO [train.py:968] (1/2) Epoch 21, batch 550, giga_loss[loss=0.2323, simple_loss=0.3136, pruned_loss=0.07545, over 28500.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.3006, pruned_loss=0.07514, over 5330086.48 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3411, pruned_loss=0.08908, over 1474822.56 frames. ], giga_tot_loss[loss=0.2223, simple_loss=0.2967, pruned_loss=0.074, over 5195412.07 frames. ], batch size: 307, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:30:58,525 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=912908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:31:25,288 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 600, giga_loss[loss=0.207, simple_loss=0.2823, pruned_loss=0.06585, over 28922.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2987, pruned_loss=0.07435, over 5405239.64 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3396, pruned_loss=0.08821, over 1618291.84 frames. ], giga_tot_loss[loss=0.2205, simple_loss=0.2946, pruned_loss=0.07321, over 5287222.04 frames. ], batch size: 213, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:31:56,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5130, 2.1637, 1.6338, 0.7377], device='cuda:1'), covar=tensor([0.7573, 0.3540, 0.4475, 0.7443], device='cuda:1'), in_proj_covar=tensor([0.1720, 0.1622, 0.1574, 0.1395], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 17:32:23,778 INFO [train.py:968] (1/2) Epoch 21, batch 650, giga_loss[loss=0.2315, simple_loss=0.3039, pruned_loss=0.07962, over 28535.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2976, pruned_loss=0.07399, over 5472236.95 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3394, pruned_loss=0.08808, over 1742491.92 frames. ], giga_tot_loss[loss=0.2193, simple_loss=0.2931, pruned_loss=0.0727, over 5367631.33 frames. ], batch size: 307, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:32:32,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:32:34,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:32:56,812 INFO [optim.py:369] (1/2) 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 17:32:57,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:33:06,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2858, 1.2550, 3.3016, 2.9907], device='cuda:1'), covar=tensor([0.1468, 0.2678, 0.0502, 0.1919], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0639, 0.0949, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 17:33:09,485 INFO [train.py:968] (1/2) Epoch 21, batch 700, giga_loss[loss=0.1894, simple_loss=0.2691, pruned_loss=0.05487, over 28992.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2955, pruned_loss=0.07284, over 5522970.76 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3386, pruned_loss=0.08772, over 1862292.08 frames. ], giga_tot_loss[loss=0.217, simple_loss=0.291, pruned_loss=0.07151, over 5431367.93 frames. ], batch size: 136, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:33:19,137 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:33:57,764 INFO [train.py:968] (1/2) Epoch 21, batch 750, giga_loss[loss=0.185, simple_loss=0.2665, pruned_loss=0.05171, over 29053.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2924, pruned_loss=0.07115, over 5556620.32 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3383, pruned_loss=0.08752, over 1922536.36 frames. ], giga_tot_loss[loss=0.2141, simple_loss=0.2883, pruned_loss=0.06993, over 5479420.78 frames. ], batch size: 128, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:34:30,474 INFO [optim.py:369] (1/2) 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 17:34:42,696 INFO [train.py:968] (1/2) Epoch 21, batch 800, giga_loss[loss=0.1867, simple_loss=0.2662, pruned_loss=0.05359, over 29102.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2901, pruned_loss=0.07046, over 5588077.57 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3383, pruned_loss=0.08734, over 1981744.98 frames. ], giga_tot_loss[loss=0.2125, simple_loss=0.2863, pruned_loss=0.06934, over 5523031.88 frames. ], batch size: 128, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:35:31,611 INFO [train.py:968] (1/2) Epoch 21, batch 850, giga_loss[loss=0.2648, simple_loss=0.3455, pruned_loss=0.09202, over 28920.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2989, pruned_loss=0.07532, over 5611152.76 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3403, pruned_loss=0.08877, over 2097609.75 frames. ], giga_tot_loss[loss=0.2205, simple_loss=0.2939, pruned_loss=0.07354, over 5550951.48 frames. ], batch size: 213, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:35:32,560 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:35:34,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:36:00,565 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=913230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:36:03,826 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:36:07,978 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 900, giga_loss[loss=0.271, simple_loss=0.3441, pruned_loss=0.09892, over 28717.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3119, pruned_loss=0.08159, over 5631342.45 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3406, pruned_loss=0.08888, over 2154094.54 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3076, pruned_loss=0.08006, over 5580318.85 frames. ], batch size: 99, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:36:53,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:37:08,419 INFO [train.py:968] (1/2) Epoch 21, batch 950, giga_loss[loss=0.2762, simple_loss=0.3584, pruned_loss=0.097, over 28601.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3231, pruned_loss=0.08724, over 5634315.06 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3407, pruned_loss=0.08892, over 2237786.76 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3191, pruned_loss=0.08593, over 5596394.13 frames. ], batch size: 60, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:37:23,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=913317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:37:40,718 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 1000, giga_loss[loss=0.2899, simple_loss=0.3502, pruned_loss=0.1148, over 23371.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3308, pruned_loss=0.09016, over 5645521.88 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3407, pruned_loss=0.08876, over 2309684.11 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3274, pruned_loss=0.08919, over 5611642.69 frames. ], batch size: 705, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:38:33,110 INFO [train.py:968] (1/2) Epoch 21, batch 1050, giga_loss[loss=0.283, simple_loss=0.3735, pruned_loss=0.09619, over 28609.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3346, pruned_loss=0.09048, over 5653270.19 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3411, pruned_loss=0.08891, over 2383821.41 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3317, pruned_loss=0.08968, over 5633922.46 frames. ], batch size: 307, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:39:02,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:39:05,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:39:14,903 INFO [optim.py:369] (1/2) 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 17:39:24,281 INFO [train.py:968] (1/2) Epoch 21, batch 1100, giga_loss[loss=0.2554, simple_loss=0.3317, pruned_loss=0.08951, over 28592.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3365, pruned_loss=0.09091, over 5653516.32 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3408, pruned_loss=0.08862, over 2435561.69 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3342, pruned_loss=0.09044, over 5635508.18 frames. ], batch size: 78, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:39:29,509 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:39:42,837 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=913472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:40:05,591 INFO [train.py:968] (1/2) Epoch 21, batch 1150, giga_loss[loss=0.2428, simple_loss=0.3255, pruned_loss=0.08006, over 28886.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3388, pruned_loss=0.09245, over 5660874.77 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3398, pruned_loss=0.08799, over 2571509.95 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3373, pruned_loss=0.09245, over 5637831.23 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:40:39,536 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 1200, giga_loss[loss=0.2677, simple_loss=0.3453, pruned_loss=0.09509, over 28537.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3415, pruned_loss=0.09425, over 5671673.64 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3402, pruned_loss=0.08804, over 2696874.88 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.34, pruned_loss=0.09441, over 5649303.24 frames. ], batch size: 71, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:41:35,247 INFO [train.py:968] (1/2) Epoch 21, batch 1250, giga_loss[loss=0.2852, simple_loss=0.3638, pruned_loss=0.1033, over 28517.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3451, pruned_loss=0.09683, over 5675429.24 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3397, pruned_loss=0.08769, over 2741351.56 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3443, pruned_loss=0.09724, over 5657431.59 frames. ], batch size: 78, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:41:35,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3432, 3.1480, 1.6153, 1.4036], device='cuda:1'), covar=tensor([0.1077, 0.0295, 0.0896, 0.1491], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0549, 0.0381, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 17:41:38,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:41:42,294 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-10 17:42:12,316 INFO [optim.py:369] (1/2) 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 17:42:19,417 INFO [train.py:968] (1/2) Epoch 21, batch 1300, giga_loss[loss=0.2495, simple_loss=0.3358, pruned_loss=0.08157, over 28936.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3477, pruned_loss=0.09729, over 5683981.85 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3403, pruned_loss=0.08848, over 2839961.62 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3469, pruned_loss=0.09755, over 5671598.08 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:42:40,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3482, 1.5817, 1.5824, 1.5626], device='cuda:1'), covar=tensor([0.0845, 0.0345, 0.0323, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 17:42:55,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=913692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:43:01,318 INFO [train.py:968] (1/2) Epoch 21, batch 1350, giga_loss[loss=0.306, simple_loss=0.3924, pruned_loss=0.1098, over 29015.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3493, pruned_loss=0.09766, over 5683952.30 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3404, pruned_loss=0.08853, over 2929903.27 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3488, pruned_loss=0.0981, over 5668701.17 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:43:36,250 INFO [optim.py:369] (1/2) 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 17:43:37,490 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 17:43:44,950 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:43:46,612 INFO [train.py:968] (1/2) Epoch 21, batch 1400, giga_loss[loss=0.264, simple_loss=0.3544, pruned_loss=0.08678, over 28884.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3505, pruned_loss=0.09754, over 5690714.05 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3398, pruned_loss=0.08813, over 3002250.53 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3507, pruned_loss=0.09833, over 5674681.05 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:43:46,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:43:58,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2379, 1.3328, 3.7566, 3.3109], device='cuda:1'), covar=tensor([0.1738, 0.2836, 0.0429, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0639, 0.0943, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 17:44:11,992 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=913780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:44:28,776 INFO [train.py:968] (1/2) Epoch 21, batch 1450, giga_loss[loss=0.2619, simple_loss=0.3458, pruned_loss=0.08896, over 28763.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3509, pruned_loss=0.09663, over 5698483.19 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3395, pruned_loss=0.08785, over 3045514.24 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3514, pruned_loss=0.09752, over 5682862.96 frames. ], batch size: 92, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:44:57,749 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:44:59,681 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:45:05,534 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 17:45:07,161 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 1500, giga_loss[loss=0.2708, simple_loss=0.3528, pruned_loss=0.09435, over 28889.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3492, pruned_loss=0.09488, over 5696791.40 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3395, pruned_loss=0.08789, over 3120243.40 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3499, pruned_loss=0.09578, over 5687084.32 frames. ], batch size: 186, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:45:23,416 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 1550, giga_loss[loss=0.2581, simple_loss=0.3401, pruned_loss=0.08805, over 28925.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09264, over 5712051.44 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3382, pruned_loss=0.08695, over 3228836.07 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3484, pruned_loss=0.09406, over 5697678.43 frames. ], batch size: 136, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:45:52,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4802, 1.7129, 1.3727, 1.6097], device='cuda:1'), covar=tensor([0.3023, 0.3018, 0.3423, 0.2505], device='cuda:1'), in_proj_covar=tensor([0.1494, 0.1079, 0.1318, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 17:46:03,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2592, 1.1339, 4.2255, 3.3965], device='cuda:1'), covar=tensor([0.1668, 0.2772, 0.0424, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0641, 0.0944, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 17:46:13,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7490, 1.9177, 1.6537, 1.8207], device='cuda:1'), covar=tensor([0.2027, 0.1925, 0.1962, 0.1879], device='cuda:1'), in_proj_covar=tensor([0.1492, 0.1079, 0.1317, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 17:46:24,693 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 1600, giga_loss[loss=0.3433, simple_loss=0.3738, pruned_loss=0.1564, over 23562.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3481, pruned_loss=0.09441, over 5697477.60 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3378, pruned_loss=0.08654, over 3294778.47 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3497, pruned_loss=0.0959, over 5681879.09 frames. ], batch size: 705, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:47:09,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=913990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:47:14,793 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=913993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:47:22,930 INFO [train.py:968] (1/2) Epoch 21, batch 1650, giga_loss[loss=0.3902, simple_loss=0.4411, pruned_loss=0.1696, over 28886.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3504, pruned_loss=0.09771, over 5708724.28 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.338, pruned_loss=0.08689, over 3384145.43 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3519, pruned_loss=0.09901, over 5690695.59 frames. ], batch size: 227, lr: 1.54e-03, grad_scale: 8.0
+2023-03-10 17:47:43,776 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=914022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:48:00,211 INFO [optim.py:369] (1/2) 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 17:48:09,384 INFO [train.py:968] (1/2) Epoch 21, batch 1700, giga_loss[loss=0.305, simple_loss=0.3695, pruned_loss=0.1203, over 28842.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3525, pruned_loss=0.1007, over 5712496.50 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3386, pruned_loss=0.08696, over 3442916.91 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3538, pruned_loss=0.102, over 5696845.13 frames. ], batch size: 119, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:48:54,362 INFO [train.py:968] (1/2) Epoch 21, batch 1750, giga_loss[loss=0.265, simple_loss=0.3432, pruned_loss=0.09335, over 28877.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3517, pruned_loss=0.1013, over 5712072.10 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.338, pruned_loss=0.08657, over 3515762.43 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3534, pruned_loss=0.1029, over 5694398.19 frames. ], batch size: 145, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:49:31,419 INFO [optim.py:369] (1/2) 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 17:49:39,799 INFO [train.py:968] (1/2) Epoch 21, batch 1800, giga_loss[loss=0.2457, simple_loss=0.3234, pruned_loss=0.08398, over 28962.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.349, pruned_loss=0.1004, over 5702311.42 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3382, pruned_loss=0.08664, over 3573162.97 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3505, pruned_loss=0.102, over 5685019.61 frames. ], batch size: 227, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:50:03,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4067, 1.4922, 3.5058, 3.2333], device='cuda:1'), covar=tensor([0.1331, 0.2511, 0.0398, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0640, 0.0946, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 17:50:06,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-10 17:50:21,281 INFO [train.py:968] (1/2) Epoch 21, batch 1850, giga_loss[loss=0.25, simple_loss=0.3345, pruned_loss=0.0827, over 28788.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.349, pruned_loss=0.1004, over 5695008.41 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3391, pruned_loss=0.0869, over 3624705.69 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3499, pruned_loss=0.1019, over 5682425.60 frames. ], batch size: 186, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:50:56,589 INFO [optim.py:369] (1/2) 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 17:51:05,334 INFO [train.py:968] (1/2) Epoch 21, batch 1900, giga_loss[loss=0.2401, simple_loss=0.3191, pruned_loss=0.08052, over 27631.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09871, over 5694109.25 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3391, pruned_loss=0.08672, over 3698739.33 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3485, pruned_loss=0.1004, over 5681153.80 frames. ], batch size: 472, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:51:52,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2601, 1.4564, 1.4901, 1.2746], device='cuda:1'), covar=tensor([0.3279, 0.2586, 0.2156, 0.2717], device='cuda:1'), in_proj_covar=tensor([0.1951, 0.1865, 0.1817, 0.1953], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 17:51:57,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4678, 4.3081, 4.0882, 2.1467], device='cuda:1'), covar=tensor([0.0520, 0.0645, 0.0685, 0.1802], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.1112, 0.0943, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 17:51:59,100 INFO [train.py:968] (1/2) Epoch 21, batch 1950, giga_loss[loss=0.2454, simple_loss=0.32, pruned_loss=0.08544, over 27987.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3441, pruned_loss=0.09671, over 5690035.65 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3393, pruned_loss=0.08669, over 3731553.48 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3449, pruned_loss=0.09821, over 5677198.80 frames. ], batch size: 412, lr: 1.54e-03, grad_scale: 2.0
+2023-03-10 17:52:31,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3018, 1.4253, 1.3531, 1.4742], device='cuda:1'), covar=tensor([0.0823, 0.0354, 0.0348, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 17:52:39,947 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 2000, giga_loss[loss=0.2554, simple_loss=0.3111, pruned_loss=0.09983, over 23438.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.339, pruned_loss=0.09413, over 5686070.32 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3394, pruned_loss=0.08677, over 3774797.90 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3396, pruned_loss=0.09542, over 5671687.80 frames. ], batch size: 705, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:52:53,115 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-10 17:53:35,006 INFO [train.py:968] (1/2) Epoch 21, batch 2050, giga_loss[loss=0.2296, simple_loss=0.3056, pruned_loss=0.07683, over 28896.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.334, pruned_loss=0.09167, over 5681458.57 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3396, pruned_loss=0.08694, over 3816167.60 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3343, pruned_loss=0.09269, over 5666834.88 frames. ], batch size: 199, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:54:15,315 INFO [optim.py:369] (1/2) 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 17:54:25,843 INFO [train.py:968] (1/2) Epoch 21, batch 2100, giga_loss[loss=0.2452, simple_loss=0.3304, pruned_loss=0.07997, over 28899.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3313, pruned_loss=0.0907, over 5671312.91 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3395, pruned_loss=0.08689, over 3878483.94 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3313, pruned_loss=0.09163, over 5653103.04 frames. ], batch size: 227, lr: 1.54e-03, grad_scale: 4.0
+2023-03-10 17:55:09,456 INFO [train.py:968] (1/2) Epoch 21, batch 2150, giga_loss[loss=0.2084, simple_loss=0.2959, pruned_loss=0.0604, over 28863.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.332, pruned_loss=0.09029, over 5682354.11 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3395, pruned_loss=0.08695, over 3928271.74 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3319, pruned_loss=0.09108, over 5663088.46 frames. ], batch size: 66, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:55:44,754 INFO [optim.py:369] (1/2) 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 17:55:52,245 INFO [train.py:968] (1/2) Epoch 21, batch 2200, giga_loss[loss=0.252, simple_loss=0.3333, pruned_loss=0.0854, over 28669.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3318, pruned_loss=0.08982, over 5692978.51 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3398, pruned_loss=0.08702, over 3956400.11 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3314, pruned_loss=0.09044, over 5675992.83 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:55:55,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2528, 3.1438, 1.3768, 1.4549], device='cuda:1'), covar=tensor([0.1007, 0.0299, 0.0908, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0548, 0.0380, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 17:56:00,067 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 2250, giga_loss[loss=0.2478, simple_loss=0.3146, pruned_loss=0.09046, over 28576.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3298, pruned_loss=0.08861, over 5697294.44 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3401, pruned_loss=0.08694, over 4011679.53 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.329, pruned_loss=0.08922, over 5680159.18 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:56:45,440 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=914614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 17:57:07,204 INFO [optim.py:369] (1/2) 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 17:57:08,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3495, 1.1289, 1.0914, 1.5265], device='cuda:1'), covar=tensor([0.0810, 0.0396, 0.0377, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 17:57:16,305 INFO [train.py:968] (1/2) Epoch 21, batch 2300, libri_loss[loss=0.2801, simple_loss=0.3712, pruned_loss=0.09448, over 29521.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3284, pruned_loss=0.08794, over 5703502.09 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3409, pruned_loss=0.08706, over 4046831.39 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3271, pruned_loss=0.08836, over 5688334.33 frames. ], batch size: 89, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:57:18,915 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-10 17:57:55,957 INFO [train.py:968] (1/2) Epoch 21, batch 2350, libri_loss[loss=0.2492, simple_loss=0.3493, pruned_loss=0.07453, over 29403.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3254, pruned_loss=0.08621, over 5709254.69 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3406, pruned_loss=0.0867, over 4092937.84 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3241, pruned_loss=0.08677, over 5692750.73 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:58:31,356 INFO [optim.py:369] (1/2) 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 17:58:38,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-10 17:58:39,107 INFO [train.py:968] (1/2) Epoch 21, batch 2400, libri_loss[loss=0.2629, simple_loss=0.343, pruned_loss=0.0914, over 29544.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3243, pruned_loss=0.08608, over 5709522.11 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3419, pruned_loss=0.0876, over 4136757.67 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3219, pruned_loss=0.08586, over 5692487.24 frames. ], batch size: 74, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 17:58:39,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-10 17:58:46,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6535, 2.0452, 1.7379, 2.0395], device='cuda:1'), covar=tensor([0.0767, 0.0289, 0.0306, 0.0777], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 17:58:56,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3743, 4.1869, 3.9610, 2.0748], device='cuda:1'), covar=tensor([0.0474, 0.0609, 0.0580, 0.2298], device='cuda:1'), in_proj_covar=tensor([0.1196, 0.1107, 0.0939, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 17:59:15,333 INFO [train.py:968] (1/2) Epoch 21, batch 2450, giga_loss[loss=0.24, simple_loss=0.3148, pruned_loss=0.08265, over 28988.00 frames. ], tot_loss[loss=0.247, simple_loss=0.323, pruned_loss=0.08545, over 5715899.82 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3424, pruned_loss=0.08767, over 4205945.56 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3201, pruned_loss=0.08516, over 5695507.89 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 17:59:45,612 INFO [optim.py:369] (1/2) 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 17:59:51,711 INFO [train.py:968] (1/2) Epoch 21, batch 2500, giga_loss[loss=0.2125, simple_loss=0.2919, pruned_loss=0.0666, over 28927.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3218, pruned_loss=0.08535, over 5716191.99 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3431, pruned_loss=0.08788, over 4225766.53 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3187, pruned_loss=0.08494, over 5702666.49 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:00:32,149 INFO [train.py:968] (1/2) Epoch 21, batch 2550, giga_loss[loss=0.2234, simple_loss=0.3028, pruned_loss=0.07199, over 28975.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3192, pruned_loss=0.08393, over 5711535.62 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3438, pruned_loss=0.08814, over 4247079.60 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.3158, pruned_loss=0.08338, over 5713475.21 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:00:45,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4748, 1.7478, 1.4710, 1.5952], device='cuda:1'), covar=tensor([0.0791, 0.0318, 0.0340, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 18:01:00,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=914936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:01:06,284 INFO [optim.py:369] (1/2) 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 18:01:11,570 INFO [train.py:968] (1/2) Epoch 21, batch 2600, giga_loss[loss=0.2206, simple_loss=0.2987, pruned_loss=0.07122, over 28868.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3175, pruned_loss=0.08306, over 5708394.98 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.344, pruned_loss=0.0883, over 4287033.23 frames. ], giga_tot_loss[loss=0.2394, simple_loss=0.314, pruned_loss=0.08238, over 5713423.33 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:01:42,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=914989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:01:51,377 INFO [train.py:968] (1/2) Epoch 21, batch 2650, giga_loss[loss=0.224, simple_loss=0.2962, pruned_loss=0.07594, over 28687.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3172, pruned_loss=0.08293, over 5713445.73 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3444, pruned_loss=0.08837, over 4323410.66 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3136, pruned_loss=0.08224, over 5716680.74 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:02:25,149 INFO [optim.py:369] (1/2) 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 18:02:31,127 INFO [train.py:968] (1/2) Epoch 21, batch 2700, giga_loss[loss=0.2273, simple_loss=0.3066, pruned_loss=0.07398, over 28958.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3194, pruned_loss=0.08428, over 5717731.89 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3445, pruned_loss=0.08836, over 4360381.26 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3158, pruned_loss=0.08362, over 5718245.93 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:02:59,773 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:01,806 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 2750, giga_loss[loss=0.259, simple_loss=0.3391, pruned_loss=0.08944, over 28916.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3241, pruned_loss=0.08735, over 5712653.20 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3439, pruned_loss=0.08801, over 4382098.55 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3214, pruned_loss=0.08706, over 5711612.78 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:03:28,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=915111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:47,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:03:49,736 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:04:04,931 INFO [train.py:968] (1/2) Epoch 21, batch 2800, giga_loss[loss=0.3454, simple_loss=0.4014, pruned_loss=0.1447, over 27571.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3309, pruned_loss=0.09166, over 5712342.00 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3434, pruned_loss=0.08774, over 4397037.39 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.329, pruned_loss=0.09164, over 5709977.24 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:04:10,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8055, 4.8690, 2.1346, 1.8376], device='cuda:1'), covar=tensor([0.0935, 0.0233, 0.0774, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0548, 0.0380, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 18:04:18,402 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=915164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:04:52,602 INFO [train.py:968] (1/2) Epoch 21, batch 2850, giga_loss[loss=0.2699, simple_loss=0.3413, pruned_loss=0.09928, over 28663.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3388, pruned_loss=0.09671, over 5699340.79 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3439, pruned_loss=0.08803, over 4419012.54 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3369, pruned_loss=0.0966, over 5695020.03 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:05:30,294 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 2900, giga_loss[loss=0.2438, simple_loss=0.334, pruned_loss=0.07685, over 28538.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3426, pruned_loss=0.09745, over 5707693.89 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3441, pruned_loss=0.08812, over 4458834.66 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3408, pruned_loss=0.09754, over 5702607.42 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:06:22,632 INFO [train.py:968] (1/2) Epoch 21, batch 2950, giga_loss[loss=0.274, simple_loss=0.35, pruned_loss=0.09896, over 28942.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3487, pruned_loss=0.1006, over 5705369.70 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3442, pruned_loss=0.08821, over 4493582.04 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3472, pruned_loss=0.1009, over 5697814.19 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:07:02,167 INFO [optim.py:369] (1/2) 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 18:07:08,320 INFO [train.py:968] (1/2) Epoch 21, batch 3000, giga_loss[loss=0.2936, simple_loss=0.3759, pruned_loss=0.1057, over 28947.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3536, pruned_loss=0.1037, over 5689111.23 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3437, pruned_loss=0.08788, over 4537149.32 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.353, pruned_loss=0.1046, over 5681494.51 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:07:08,320 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 18:07:17,023 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 18:07:58,740 INFO [train.py:968] (1/2) Epoch 21, batch 3050, giga_loss[loss=0.2349, simple_loss=0.3144, pruned_loss=0.07769, over 27924.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3522, pruned_loss=0.102, over 5692514.39 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3432, pruned_loss=0.0875, over 4564059.06 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3523, pruned_loss=0.1033, over 5682777.70 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:08:13,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4919, 1.5637, 1.7961, 1.3836], device='cuda:1'), covar=tensor([0.1392, 0.2228, 0.1201, 0.1729], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0701, 0.0946, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 18:08:34,716 INFO [optim.py:369] (1/2) 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 18:08:36,408 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 3100, giga_loss[loss=0.278, simple_loss=0.3511, pruned_loss=0.1025, over 28386.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3495, pruned_loss=0.09983, over 5699553.09 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3438, pruned_loss=0.08795, over 4590489.51 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3493, pruned_loss=0.1008, over 5688399.46 frames. ], batch size: 65, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:08:43,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0090, 1.1966, 3.2935, 2.9319], device='cuda:1'), covar=tensor([0.1806, 0.2889, 0.0525, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0635, 0.0937, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 18:08:47,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:09:04,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5954, 1.8851, 1.2972, 1.4370], device='cuda:1'), covar=tensor([0.1171, 0.0695, 0.1085, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0447, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 18:09:24,588 INFO [train.py:968] (1/2) Epoch 21, batch 3150, libri_loss[loss=0.2783, simple_loss=0.3588, pruned_loss=0.09888, over 29533.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3483, pruned_loss=0.09863, over 5700229.80 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3444, pruned_loss=0.08853, over 4618659.89 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3478, pruned_loss=0.09926, over 5695511.39 frames. ], batch size: 82, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:10:02,017 INFO [optim.py:369] (1/2) 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 18:10:08,165 INFO [train.py:968] (1/2) Epoch 21, batch 3200, giga_loss[loss=0.3296, simple_loss=0.3931, pruned_loss=0.133, over 28924.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09895, over 5696077.52 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3444, pruned_loss=0.0886, over 4628032.66 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3488, pruned_loss=0.09953, over 5698766.18 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:10:10,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:10:48,100 INFO [train.py:968] (1/2) Epoch 21, batch 3250, giga_loss[loss=0.2653, simple_loss=0.346, pruned_loss=0.09234, over 28689.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3513, pruned_loss=0.09974, over 5705129.99 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3446, pruned_loss=0.08869, over 4666052.63 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1004, over 5701431.37 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:11:10,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0777, 3.9020, 3.6810, 1.9449], device='cuda:1'), covar=tensor([0.0661, 0.0785, 0.0709, 0.2029], device='cuda:1'), in_proj_covar=tensor([0.1196, 0.1108, 0.0939, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 18:11:27,226 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 3300, libri_loss[loss=0.2737, simple_loss=0.3608, pruned_loss=0.09327, over 27938.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3532, pruned_loss=0.1013, over 5694240.35 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3449, pruned_loss=0.08893, over 4667624.36 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3528, pruned_loss=0.1017, over 5699937.20 frames. ], batch size: 116, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:11:48,233 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-10 18:12:01,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9821, 3.1510, 1.9713, 1.0697], device='cuda:1'), covar=tensor([0.7710, 0.2624, 0.3989, 0.6921], device='cuda:1'), in_proj_covar=tensor([0.1729, 0.1622, 0.1583, 0.1404], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 18:12:07,679 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:12:13,558 INFO [train.py:968] (1/2) Epoch 21, batch 3350, giga_loss[loss=0.3099, simple_loss=0.3812, pruned_loss=0.1193, over 28788.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.354, pruned_loss=0.1021, over 5699469.66 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.345, pruned_loss=0.08883, over 4702934.84 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3539, pruned_loss=0.103, over 5699547.87 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:12:41,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2826, 2.3894, 1.9714, 2.1504], device='cuda:1'), covar=tensor([0.0654, 0.0436, 0.0707, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0445, 0.0516, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 18:12:48,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=915743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:12:50,906 INFO [optim.py:369] (1/2) 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 18:12:54,988 INFO [train.py:968] (1/2) Epoch 21, batch 3400, giga_loss[loss=0.2948, simple_loss=0.3625, pruned_loss=0.1135, over 28648.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3539, pruned_loss=0.1023, over 5693560.44 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3449, pruned_loss=0.08889, over 4720347.88 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3541, pruned_loss=0.1033, over 5704672.88 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:13:37,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7775, 1.9407, 1.4516, 1.5436], device='cuda:1'), covar=tensor([0.0955, 0.0670, 0.0986, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0444, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 18:13:41,087 INFO [train.py:968] (1/2) Epoch 21, batch 3450, giga_loss[loss=0.2241, simple_loss=0.3148, pruned_loss=0.06669, over 28551.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3539, pruned_loss=0.1024, over 5704868.30 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3448, pruned_loss=0.08884, over 4732235.59 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3542, pruned_loss=0.1033, over 5711705.86 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:13:59,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=915821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:14:09,937 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=915835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:14:20,499 INFO [optim.py:369] (1/2) 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 18:14:23,822 INFO [train.py:968] (1/2) Epoch 21, batch 3500, giga_loss[loss=0.2384, simple_loss=0.3204, pruned_loss=0.07821, over 28619.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3545, pruned_loss=0.1026, over 5695251.35 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3452, pruned_loss=0.08923, over 4732422.71 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3545, pruned_loss=0.1031, over 5709907.63 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:14:48,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-10 18:15:03,271 INFO [train.py:968] (1/2) Epoch 21, batch 3550, giga_loss[loss=0.2475, simple_loss=0.3325, pruned_loss=0.08127, over 28769.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3538, pruned_loss=0.1007, over 5703057.66 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3454, pruned_loss=0.08924, over 4755614.47 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3538, pruned_loss=0.1014, over 5711021.55 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:15:26,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=915927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:15:28,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2339, 2.2164, 1.6531, 1.8106], device='cuda:1'), covar=tensor([0.0871, 0.0672, 0.1004, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0445, 0.0516, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 18:15:40,787 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 3600, giga_loss[loss=0.2654, simple_loss=0.3411, pruned_loss=0.09486, over 28993.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3537, pruned_loss=0.09993, over 5696730.59 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3453, pruned_loss=0.08918, over 4764773.90 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.354, pruned_loss=0.1007, over 5715145.91 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:15:56,076 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:15:58,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=915967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:07,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=915978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:08,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8656, 1.9685, 1.8375, 1.6559], device='cuda:1'), covar=tensor([0.1978, 0.2450, 0.2260, 0.2288], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0745, 0.0708, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:16:09,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=915981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:11,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3421, 1.4450, 1.2817, 1.6072], device='cuda:1'), covar=tensor([0.0815, 0.0349, 0.0342, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0061, 0.0106], device='cuda:1')
+2023-03-10 18:16:20,479 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=915996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:16:24,969 INFO [train.py:968] (1/2) Epoch 21, batch 3650, libri_loss[loss=0.2761, simple_loss=0.362, pruned_loss=0.09509, over 29523.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.352, pruned_loss=0.0991, over 5703958.85 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3453, pruned_loss=0.08916, over 4776190.43 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3524, pruned_loss=0.09984, over 5716367.34 frames. ], batch size: 83, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:16:32,118 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:04,443 INFO [optim.py:369] (1/2) 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 18:17:07,484 INFO [train.py:968] (1/2) Epoch 21, batch 3700, giga_loss[loss=0.2376, simple_loss=0.3261, pruned_loss=0.07457, over 28994.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3513, pruned_loss=0.09977, over 5706701.34 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3452, pruned_loss=0.08931, over 4798781.63 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3518, pruned_loss=0.1005, over 5712917.07 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:17:18,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:19,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:22,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:42,776 INFO [train.py:968] (1/2) Epoch 21, batch 3750, giga_loss[loss=0.2493, simple_loss=0.3273, pruned_loss=0.08564, over 28821.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3485, pruned_loss=0.09805, over 5715159.13 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3443, pruned_loss=0.08893, over 4852172.86 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3498, pruned_loss=0.09941, over 5716161.95 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:17:43,655 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:17:54,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:18:22,819 INFO [optim.py:369] (1/2) 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 18:18:26,210 INFO [train.py:968] (1/2) Epoch 21, batch 3800, giga_loss[loss=0.289, simple_loss=0.3658, pruned_loss=0.1061, over 28974.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09792, over 5727258.22 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3439, pruned_loss=0.08881, over 4878045.62 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3498, pruned_loss=0.09927, over 5724375.47 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:18:30,719 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 3850, giga_loss[loss=0.2806, simple_loss=0.3574, pruned_loss=0.1019, over 28982.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3502, pruned_loss=0.09952, over 5726802.82 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.344, pruned_loss=0.08876, over 4893002.28 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3514, pruned_loss=0.1008, over 5722460.67 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:19:13,820 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:15,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:30,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:38,454 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:39,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4473, 2.5263, 2.4354, 2.2138], device='cuda:1'), covar=tensor([0.1755, 0.2087, 0.1914, 0.1934], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0743, 0.0706, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:19:40,742 INFO [optim.py:369] (1/2) 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 18:19:45,127 INFO [train.py:968] (1/2) Epoch 21, batch 3900, giga_loss[loss=0.2482, simple_loss=0.3313, pruned_loss=0.08257, over 28524.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3494, pruned_loss=0.09829, over 5727101.20 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3438, pruned_loss=0.08874, over 4918167.23 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3506, pruned_loss=0.09953, over 5719509.58 frames. ], batch size: 60, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:19:49,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5856, 1.8466, 1.4702, 1.5463], device='cuda:1'), covar=tensor([0.2719, 0.2662, 0.3136, 0.2407], device='cuda:1'), in_proj_covar=tensor([0.1486, 0.1077, 0.1312, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 18:19:53,222 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:19:55,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:20:14,368 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:20:21,153 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 3950, giga_loss[loss=0.2644, simple_loss=0.3452, pruned_loss=0.09181, over 29034.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3471, pruned_loss=0.09633, over 5724123.05 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3439, pruned_loss=0.08887, over 4927174.06 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.348, pruned_loss=0.0973, over 5717219.41 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:21:03,633 INFO [optim.py:369] (1/2) 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 18:21:08,587 INFO [train.py:968] (1/2) Epoch 21, batch 4000, giga_loss[loss=0.2609, simple_loss=0.3395, pruned_loss=0.09108, over 28799.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3465, pruned_loss=0.0962, over 5726868.01 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.344, pruned_loss=0.08886, over 4961545.46 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3473, pruned_loss=0.09724, over 5718668.39 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:21:21,429 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:21:47,510 INFO [train.py:968] (1/2) Epoch 21, batch 4050, libri_loss[loss=0.2832, simple_loss=0.3627, pruned_loss=0.1019, over 25770.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3445, pruned_loss=0.09526, over 5714610.56 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.344, pruned_loss=0.08873, over 4990397.19 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3452, pruned_loss=0.09643, over 5708795.76 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:22:02,895 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-10 18:22:16,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0033, 1.2508, 0.9194, 0.8563], device='cuda:1'), covar=tensor([0.1180, 0.0549, 0.1533, 0.1166], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0446, 0.0517, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 18:22:20,624 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 4100, giga_loss[loss=0.2394, simple_loss=0.3191, pruned_loss=0.07982, over 28984.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3423, pruned_loss=0.09437, over 5720623.89 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3442, pruned_loss=0.08889, over 5017761.67 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3426, pruned_loss=0.09533, over 5710558.85 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:23:05,640 INFO [train.py:968] (1/2) Epoch 21, batch 4150, giga_loss[loss=0.2554, simple_loss=0.3282, pruned_loss=0.09127, over 28703.00 frames. ], tot_loss[loss=0.265, simple_loss=0.341, pruned_loss=0.09447, over 5710005.26 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3443, pruned_loss=0.089, over 5022317.86 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3411, pruned_loss=0.09516, over 5701270.61 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:23:31,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:23:43,170 INFO [optim.py:369] (1/2) 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 18:23:46,551 INFO [train.py:968] (1/2) Epoch 21, batch 4200, giga_loss[loss=0.2442, simple_loss=0.3248, pruned_loss=0.08178, over 28762.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3392, pruned_loss=0.09396, over 5709679.52 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3443, pruned_loss=0.089, over 5026760.27 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3393, pruned_loss=0.09452, over 5702053.70 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:24:06,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 18:24:28,767 INFO [train.py:968] (1/2) Epoch 21, batch 4250, giga_loss[loss=0.2291, simple_loss=0.3077, pruned_loss=0.07526, over 28831.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3372, pruned_loss=0.09331, over 5707988.57 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3443, pruned_loss=0.0889, over 5034775.66 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3372, pruned_loss=0.09386, over 5700993.28 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:24:34,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:25:04,468 INFO [optim.py:369] (1/2) 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 18:25:07,872 INFO [train.py:968] (1/2) Epoch 21, batch 4300, libri_loss[loss=0.2607, simple_loss=0.3388, pruned_loss=0.09127, over 29490.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.335, pruned_loss=0.09228, over 5716874.53 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3439, pruned_loss=0.0887, over 5062618.81 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3351, pruned_loss=0.093, over 5707065.04 frames. ], batch size: 70, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:25:13,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:25:27,254 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:25:29,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:25:36,925 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7676, 2.4846, 1.5161, 0.8045], device='cuda:1'), covar=tensor([0.8082, 0.3851, 0.4291, 0.7382], device='cuda:1'), in_proj_covar=tensor([0.1725, 0.1614, 0.1577, 0.1397], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 18:25:46,060 INFO [train.py:968] (1/2) Epoch 21, batch 4350, libri_loss[loss=0.2632, simple_loss=0.3496, pruned_loss=0.08838, over 25538.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3333, pruned_loss=0.09199, over 5697653.64 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3441, pruned_loss=0.08891, over 5065915.81 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3329, pruned_loss=0.09248, over 5701911.39 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:25:50,489 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:07,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1288, 1.8035, 1.4076, 0.3949], device='cuda:1'), covar=tensor([0.5459, 0.3166, 0.4777, 0.6279], device='cuda:1'), in_proj_covar=tensor([0.1729, 0.1618, 0.1579, 0.1401], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 18:26:18,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=916742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:22,069 INFO [optim.py:369] (1/2) 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 18:26:24,475 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:24,808 INFO [train.py:968] (1/2) Epoch 21, batch 4400, giga_loss[loss=0.261, simple_loss=0.3324, pruned_loss=0.09474, over 28938.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.331, pruned_loss=0.09071, over 5700266.91 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3444, pruned_loss=0.089, over 5067951.01 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3304, pruned_loss=0.09105, over 5710349.85 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:26:27,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:31,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-10 18:26:39,341 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=916769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:26:48,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:04,779 INFO [train.py:968] (1/2) Epoch 21, batch 4450, giga_loss[loss=0.2457, simple_loss=0.3307, pruned_loss=0.08034, over 28919.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3313, pruned_loss=0.09005, over 5698085.87 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3449, pruned_loss=0.0893, over 5087425.86 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.33, pruned_loss=0.09012, over 5707655.36 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:27:05,256 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:07,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=916804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:21,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9898, 1.0075, 3.2930, 3.0508], device='cuda:1'), covar=tensor([0.2102, 0.3273, 0.0868, 0.1218], device='cuda:1'), in_proj_covar=tensor([0.0746, 0.0636, 0.0939, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 18:27:32,212 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:27:46,323 INFO [optim.py:369] (1/2) 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 18:27:46,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2022, 1.2011, 4.0196, 3.2184], device='cuda:1'), covar=tensor([0.1711, 0.2815, 0.0410, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0746, 0.0635, 0.0939, 0.0891], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 18:27:48,446 INFO [train.py:968] (1/2) Epoch 21, batch 4500, giga_loss[loss=0.3108, simple_loss=0.3787, pruned_loss=0.1214, over 27667.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3357, pruned_loss=0.09252, over 5683486.86 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3454, pruned_loss=0.08967, over 5098264.75 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3339, pruned_loss=0.09234, over 5697699.86 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:28:17,573 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=916885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:28:20,757 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 4550, libri_loss[loss=0.2246, simple_loss=0.3088, pruned_loss=0.07015, over 28067.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3384, pruned_loss=0.09313, over 5691963.20 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3452, pruned_loss=0.08955, over 5108871.23 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.337, pruned_loss=0.09314, over 5701646.40 frames. ], batch size: 62, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:28:41,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1864, 0.9539, 1.0020, 1.3727], device='cuda:1'), covar=tensor([0.0700, 0.0426, 0.0355, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0061, 0.0106], device='cuda:1')
+2023-03-10 18:28:43,850 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=916917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:29:01,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5577, 1.7961, 1.4207, 1.3014], device='cuda:1'), covar=tensor([0.1013, 0.0611, 0.0997, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0445, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 18:29:10,693 INFO [optim.py:369] (1/2) 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 18:29:12,615 INFO [train.py:968] (1/2) Epoch 21, batch 4600, giga_loss[loss=0.2756, simple_loss=0.362, pruned_loss=0.09461, over 28795.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3413, pruned_loss=0.09444, over 5686945.81 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3456, pruned_loss=0.08986, over 5128543.24 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3398, pruned_loss=0.0943, over 5692257.02 frames. ], batch size: 243, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:29:31,065 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 18:29:49,247 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6417, 1.7245, 1.3222, 1.2939], device='cuda:1'), covar=tensor([0.0933, 0.0626, 0.1030, 0.1215], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0445, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 18:29:57,398 INFO [train.py:968] (1/2) Epoch 21, batch 4650, giga_loss[loss=0.2695, simple_loss=0.3514, pruned_loss=0.09382, over 28637.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3407, pruned_loss=0.09352, over 5691101.29 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3458, pruned_loss=0.09006, over 5147517.49 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3391, pruned_loss=0.09333, over 5690492.08 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:30:35,019 INFO [optim.py:369] (1/2) 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 18:30:36,754 INFO [train.py:968] (1/2) Epoch 21, batch 4700, giga_loss[loss=0.2682, simple_loss=0.3527, pruned_loss=0.0918, over 28766.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3397, pruned_loss=0.09279, over 5701034.05 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3458, pruned_loss=0.08999, over 5163002.85 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3384, pruned_loss=0.09277, over 5698560.11 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:31:12,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5162, 1.7411, 1.4297, 1.6445], device='cuda:1'), covar=tensor([0.2617, 0.2708, 0.3135, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.1488, 0.1075, 0.1314, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 18:31:17,033 INFO [train.py:968] (1/2) Epoch 21, batch 4750, giga_loss[loss=0.2765, simple_loss=0.3427, pruned_loss=0.1052, over 23901.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3403, pruned_loss=0.09348, over 5695452.38 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3459, pruned_loss=0.09006, over 5175933.55 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.339, pruned_loss=0.09349, over 5691544.30 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:31:53,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:31:56,094 INFO [optim.py:369] (1/2) 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 18:31:58,140 INFO [train.py:968] (1/2) Epoch 21, batch 4800, giga_loss[loss=0.2691, simple_loss=0.3422, pruned_loss=0.09799, over 28737.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3416, pruned_loss=0.09497, over 5701244.79 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3459, pruned_loss=0.09008, over 5186402.92 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3406, pruned_loss=0.09502, over 5695819.39 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:32:18,653 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:32:34,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2231, 3.4250, 1.3133, 1.4328], device='cuda:1'), covar=tensor([0.1046, 0.0365, 0.1000, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0547, 0.0380, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 18:32:37,360 INFO [train.py:968] (1/2) Epoch 21, batch 4850, giga_loss[loss=0.3042, simple_loss=0.3781, pruned_loss=0.1151, over 28350.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.345, pruned_loss=0.09668, over 5705607.78 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3457, pruned_loss=0.09007, over 5217977.06 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3441, pruned_loss=0.09695, over 5692991.62 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:32:59,379 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:33:18,014 INFO [optim.py:369] (1/2) 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 18:33:20,153 INFO [train.py:968] (1/2) Epoch 21, batch 4900, giga_loss[loss=0.2471, simple_loss=0.3399, pruned_loss=0.07715, over 29080.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3472, pruned_loss=0.09727, over 5711201.40 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3459, pruned_loss=0.09018, over 5223677.38 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3464, pruned_loss=0.09749, over 5702686.73 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:33:22,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9262, 1.2952, 1.0793, 0.1895], device='cuda:1'), covar=tensor([0.4381, 0.3032, 0.4264, 0.6638], device='cuda:1'), in_proj_covar=tensor([0.1717, 0.1608, 0.1573, 0.1396], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 18:33:33,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5383, 1.7476, 1.6159, 1.3701], device='cuda:1'), covar=tensor([0.2682, 0.2294, 0.2171, 0.2582], device='cuda:1'), in_proj_covar=tensor([0.1946, 0.1859, 0.1805, 0.1943], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 18:33:34,708 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-10 18:33:35,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 18:33:48,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4341, 1.7117, 1.5658, 1.5422], device='cuda:1'), covar=tensor([0.1873, 0.2059, 0.2279, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0744, 0.0708, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:33:49,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:33:51,974 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 4950, giga_loss[loss=0.238, simple_loss=0.3153, pruned_loss=0.08036, over 28783.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3498, pruned_loss=0.09871, over 5716093.39 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3463, pruned_loss=0.09042, over 5246789.32 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3489, pruned_loss=0.09896, over 5703223.74 frames. ], batch size: 66, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:34:04,035 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:34:14,607 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:34:37,426 INFO [optim.py:369] (1/2) 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 18:34:38,774 INFO [train.py:968] (1/2) Epoch 21, batch 5000, giga_loss[loss=0.2631, simple_loss=0.3446, pruned_loss=0.09077, over 28808.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3502, pruned_loss=0.0985, over 5713975.26 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3467, pruned_loss=0.09064, over 5255450.85 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3492, pruned_loss=0.09874, over 5708053.00 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:34:45,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6500, 1.8245, 1.5389, 1.6908], device='cuda:1'), covar=tensor([0.3001, 0.2981, 0.3341, 0.2557], device='cuda:1'), in_proj_covar=tensor([0.1488, 0.1076, 0.1314, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 18:35:19,058 INFO [train.py:968] (1/2) Epoch 21, batch 5050, giga_loss[loss=0.2769, simple_loss=0.3639, pruned_loss=0.09494, over 28922.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.351, pruned_loss=0.09926, over 5721250.14 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3469, pruned_loss=0.09075, over 5263530.22 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3501, pruned_loss=0.09947, over 5715540.68 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:35:27,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2401, 1.1906, 3.6452, 3.1433], device='cuda:1'), covar=tensor([0.1641, 0.2887, 0.0435, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0751, 0.0638, 0.0944, 0.0898], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 18:35:58,608 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 5100, giga_loss[loss=0.261, simple_loss=0.3337, pruned_loss=0.09413, over 28627.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3488, pruned_loss=0.09803, over 5718389.32 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.347, pruned_loss=0.09079, over 5266729.57 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3481, pruned_loss=0.0982, over 5713162.98 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:36:23,085 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:36:25,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:36:44,544 INFO [train.py:968] (1/2) Epoch 21, batch 5150, giga_loss[loss=0.2763, simple_loss=0.351, pruned_loss=0.1008, over 28890.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3462, pruned_loss=0.09663, over 5722772.44 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3472, pruned_loss=0.09089, over 5275998.77 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3455, pruned_loss=0.09677, over 5716240.40 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:37:09,561 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:22,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:24,228 INFO [optim.py:369] (1/2) 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 18:37:24,814 INFO [train.py:968] (1/2) Epoch 21, batch 5200, giga_loss[loss=0.2384, simple_loss=0.3178, pruned_loss=0.07953, over 28878.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3426, pruned_loss=0.09498, over 5725708.09 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3472, pruned_loss=0.09093, over 5282243.27 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.342, pruned_loss=0.09511, over 5718977.93 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:37:32,230 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:33,434 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:37:35,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:38:00,313 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3320, 1.1797, 3.8055, 3.2659], device='cuda:1'), covar=tensor([0.1575, 0.2872, 0.0416, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0640, 0.0946, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 18:38:04,956 INFO [train.py:968] (1/2) Epoch 21, batch 5250, giga_loss[loss=0.3464, simple_loss=0.3952, pruned_loss=0.1488, over 26531.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3416, pruned_loss=0.0942, over 5724714.44 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3471, pruned_loss=0.09094, over 5302742.15 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.341, pruned_loss=0.09441, over 5714515.92 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:38:05,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:38:34,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4739, 1.6679, 1.3604, 1.6279], device='cuda:1'), covar=tensor([0.2638, 0.2748, 0.3109, 0.2379], device='cuda:1'), in_proj_covar=tensor([0.1489, 0.1076, 0.1315, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 18:38:44,215 INFO [optim.py:369] (1/2) 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 18:38:44,794 INFO [train.py:968] (1/2) Epoch 21, batch 5300, giga_loss[loss=0.2532, simple_loss=0.3426, pruned_loss=0.08191, over 29023.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3434, pruned_loss=0.09371, over 5706129.62 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3473, pruned_loss=0.09104, over 5303690.71 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3427, pruned_loss=0.09384, over 5706516.69 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:39:10,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:18,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:22,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:23,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:26,655 INFO [train.py:968] (1/2) Epoch 21, batch 5350, libri_loss[loss=0.2854, simple_loss=0.3666, pruned_loss=0.1021, over 26353.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3438, pruned_loss=0.09388, over 5702851.29 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3473, pruned_loss=0.09111, over 5318074.69 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.343, pruned_loss=0.094, over 5701132.09 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:39:32,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:35,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:44,447 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:56,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:39:58,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5400, 2.8767, 3.0431, 2.4342], device='cuda:1'), covar=tensor([0.1885, 0.1743, 0.1428, 0.1839], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0746, 0.0708, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:39:59,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:40:01,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:40:04,970 INFO [optim.py:369] (1/2) 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 18:40:05,699 INFO [train.py:968] (1/2) Epoch 21, batch 5400, libri_loss[loss=0.2834, simple_loss=0.3613, pruned_loss=0.1027, over 19839.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3427, pruned_loss=0.09402, over 5697258.36 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3472, pruned_loss=0.09114, over 5328473.33 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3421, pruned_loss=0.0942, over 5701007.01 frames. ], batch size: 187, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:40:09,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5710, 1.6685, 1.7198, 1.5681], device='cuda:1'), covar=tensor([0.3236, 0.2735, 0.2356, 0.2760], device='cuda:1'), in_proj_covar=tensor([0.1951, 0.1865, 0.1811, 0.1947], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 18:40:24,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:40:45,431 INFO [train.py:968] (1/2) Epoch 21, batch 5450, giga_loss[loss=0.2552, simple_loss=0.3217, pruned_loss=0.09435, over 28445.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3414, pruned_loss=0.09445, over 5696534.41 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3467, pruned_loss=0.09091, over 5337721.33 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3412, pruned_loss=0.09485, over 5697786.82 frames. ], batch size: 78, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:41:00,168 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-10 18:41:05,119 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:07,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:24,832 INFO [optim.py:369] (1/2) 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 18:41:25,606 INFO [train.py:968] (1/2) Epoch 21, batch 5500, giga_loss[loss=0.2834, simple_loss=0.3562, pruned_loss=0.1053, over 27938.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3408, pruned_loss=0.09508, over 5707467.81 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3467, pruned_loss=0.09083, over 5356702.78 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3406, pruned_loss=0.0956, over 5702544.36 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:41:27,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:30,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=917856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:30,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:41:45,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6677, 1.8312, 1.4512, 1.7429], device='cuda:1'), covar=tensor([0.2524, 0.2653, 0.3096, 0.2384], device='cuda:1'), in_proj_covar=tensor([0.1490, 0.1078, 0.1315, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 18:42:06,233 INFO [train.py:968] (1/2) Epoch 21, batch 5550, giga_loss[loss=0.263, simple_loss=0.3358, pruned_loss=0.09515, over 28837.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3398, pruned_loss=0.09535, over 5705630.59 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3466, pruned_loss=0.09081, over 5373132.72 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3394, pruned_loss=0.09597, over 5698932.79 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:42:09,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:36,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:40,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=917938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:40,773 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=917939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:42:44,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7457, 1.8988, 1.7733, 1.6036], device='cuda:1'), covar=tensor([0.1918, 0.2563, 0.2267, 0.2571], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0746, 0.0709, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:42:48,361 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 5600, giga_loss[loss=0.2454, simple_loss=0.3136, pruned_loss=0.08855, over 28860.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3379, pruned_loss=0.09441, over 5702969.89 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3469, pruned_loss=0.09092, over 5368736.41 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3373, pruned_loss=0.09483, over 5706701.45 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:43:11,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3448, 1.4759, 1.4456, 1.3045], device='cuda:1'), covar=tensor([0.2632, 0.2328, 0.1967, 0.2437], device='cuda:1'), in_proj_covar=tensor([0.1956, 0.1872, 0.1815, 0.1952], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 18:43:12,895 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-10 18:43:15,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0876, 1.3889, 1.1779, 0.2774], device='cuda:1'), covar=tensor([0.3248, 0.3002, 0.4037, 0.5641], device='cuda:1'), in_proj_covar=tensor([0.1718, 0.1607, 0.1573, 0.1395], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 18:43:26,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=917996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:30,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=917999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:31,086 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 5650, giga_loss[loss=0.2272, simple_loss=0.3078, pruned_loss=0.07327, over 28895.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3343, pruned_loss=0.09261, over 5711252.71 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3466, pruned_loss=0.09086, over 5379374.70 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3338, pruned_loss=0.09305, over 5710548.05 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:43:34,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:53,019 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:43:56,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:10,031 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:44:10,907 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-10 18:44:10,962 INFO [train.py:968] (1/2) Epoch 21, batch 5700, giga_loss[loss=0.2161, simple_loss=0.2946, pruned_loss=0.06877, over 28918.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3299, pruned_loss=0.09054, over 5722056.51 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3463, pruned_loss=0.09079, over 5392610.39 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3294, pruned_loss=0.09098, over 5717060.10 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:44:12,994 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:29,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:34,997 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:36,223 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:36,307 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:37,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:38,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:44:51,727 INFO [train.py:968] (1/2) Epoch 21, batch 5750, giga_loss[loss=0.2586, simple_loss=0.3307, pruned_loss=0.09319, over 28799.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3272, pruned_loss=0.08928, over 5717484.17 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3459, pruned_loss=0.09056, over 5397880.22 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.327, pruned_loss=0.08981, over 5711720.65 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:45:01,942 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:45:03,396 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:45:09,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918122.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 18:45:16,944 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:45:32,452 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 5800, giga_loss[loss=0.2755, simple_loss=0.3593, pruned_loss=0.09588, over 28790.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3298, pruned_loss=0.09041, over 5721739.79 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.346, pruned_loss=0.09059, over 5400415.82 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3295, pruned_loss=0.0908, over 5716389.29 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:45:39,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8407, 1.1108, 2.8271, 2.7278], device='cuda:1'), covar=tensor([0.1690, 0.2670, 0.0569, 0.1267], device='cuda:1'), in_proj_covar=tensor([0.0751, 0.0639, 0.0945, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 18:46:13,899 INFO [train.py:968] (1/2) Epoch 21, batch 5850, giga_loss[loss=0.2208, simple_loss=0.3053, pruned_loss=0.0682, over 28903.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3331, pruned_loss=0.09149, over 5724449.33 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3457, pruned_loss=0.09045, over 5407934.91 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3328, pruned_loss=0.09193, over 5719118.94 frames. ], batch size: 66, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:46:27,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:46:30,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:46:41,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1286, 1.2456, 1.1297, 1.1453], device='cuda:1'), covar=tensor([0.1810, 0.1890, 0.1409, 0.1553], device='cuda:1'), in_proj_covar=tensor([0.1952, 0.1874, 0.1813, 0.1951], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 18:46:45,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3975, 1.5158, 4.2701, 3.3743], device='cuda:1'), covar=tensor([0.1620, 0.2475, 0.0412, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0638, 0.0944, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 18:46:54,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:46:58,014 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 5900, giga_loss[loss=0.2573, simple_loss=0.3413, pruned_loss=0.08667, over 29022.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.336, pruned_loss=0.09254, over 5718490.77 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3456, pruned_loss=0.09036, over 5414971.71 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3358, pruned_loss=0.093, over 5712149.58 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:47:10,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 18:47:40,955 INFO [train.py:968] (1/2) Epoch 21, batch 5950, giga_loss[loss=0.3022, simple_loss=0.3724, pruned_loss=0.116, over 27651.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3384, pruned_loss=0.09347, over 5718954.59 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3453, pruned_loss=0.09022, over 5422854.91 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3385, pruned_loss=0.09403, over 5712193.65 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:47:51,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:47:51,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9465, 1.8757, 1.8334, 1.7104], device='cuda:1'), covar=tensor([0.1898, 0.2747, 0.2327, 0.2441], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0751, 0.0713, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:48:00,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3962, 3.2759, 1.5061, 1.4744], device='cuda:1'), covar=tensor([0.0956, 0.0342, 0.0938, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0551, 0.0382, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 18:48:21,988 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 6000, giga_loss[loss=0.31, simple_loss=0.3722, pruned_loss=0.124, over 28903.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3401, pruned_loss=0.09428, over 5719853.51 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3453, pruned_loss=0.09022, over 5432673.01 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3399, pruned_loss=0.09479, over 5711043.97 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:48:22,001 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 18:48:30,631 INFO [train.py:1012] (1/2) Epoch 21, validation: loss=0.2114, simple_loss=0.3187, pruned_loss=0.05201, over 944034.00 frames. 
+2023-03-10 18:48:30,632 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 18:49:14,689 INFO [train.py:968] (1/2) Epoch 21, batch 6050, giga_loss[loss=0.2642, simple_loss=0.3386, pruned_loss=0.09493, over 28641.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3468, pruned_loss=0.09959, over 5714384.54 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3458, pruned_loss=0.09037, over 5445726.05 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3462, pruned_loss=0.1001, over 5703046.12 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:49:34,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9101, 2.0198, 1.7847, 1.7321], device='cuda:1'), covar=tensor([0.1780, 0.2376, 0.2252, 0.2233], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0751, 0.0714, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:49:35,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-10 18:49:51,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2344, 1.5327, 1.4572, 1.3467], device='cuda:1'), covar=tensor([0.1475, 0.1386, 0.2036, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0750, 0.0713, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 18:50:03,728 INFO [optim.py:369] (1/2) 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 18:50:03,741 INFO [train.py:968] (1/2) Epoch 21, batch 6100, giga_loss[loss=0.317, simple_loss=0.3833, pruned_loss=0.1253, over 28958.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3542, pruned_loss=0.1056, over 5701562.18 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3456, pruned_loss=0.09032, over 5450357.15 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3539, pruned_loss=0.1061, over 5690986.72 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:50:10,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:50:14,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:50:38,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:50:46,102 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 6150, giga_loss[loss=0.31, simple_loss=0.3787, pruned_loss=0.1207, over 28675.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3604, pruned_loss=0.1101, over 5693854.52 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3452, pruned_loss=0.09025, over 5464038.32 frames. ], giga_tot_loss[loss=0.2916, simple_loss=0.3609, pruned_loss=0.1111, over 5679399.07 frames. ], batch size: 78, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:50:54,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=918507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:51:05,837 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:51:37,273 INFO [train.py:968] (1/2) Epoch 21, batch 6200, giga_loss[loss=0.3246, simple_loss=0.3827, pruned_loss=0.1332, over 28580.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3674, pruned_loss=0.116, over 5688093.30 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3458, pruned_loss=0.0906, over 5470070.23 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3676, pruned_loss=0.1169, over 5675431.26 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:51:37,976 INFO [optim.py:369] (1/2) 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 18:52:22,606 INFO [train.py:968] (1/2) Epoch 21, batch 6250, giga_loss[loss=0.3474, simple_loss=0.4025, pruned_loss=0.1462, over 28669.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3744, pruned_loss=0.122, over 5681615.29 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.346, pruned_loss=0.09064, over 5475040.15 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3749, pruned_loss=0.1233, over 5672017.00 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:52:46,193 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=918627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:52:59,817 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=918640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 18:53:03,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918643.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 18:53:09,432 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 6300, libri_loss[loss=0.244, simple_loss=0.3215, pruned_loss=0.08325, over 29569.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3781, pruned_loss=0.1253, over 5675724.18 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3459, pruned_loss=0.09057, over 5487635.29 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3795, pruned_loss=0.1273, over 5662094.57 frames. ], batch size: 75, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:53:11,076 INFO [optim.py:369] (1/2) 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 18:53:12,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=918653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:53:33,650 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918672.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 18:53:43,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:53:50,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3311, 1.6424, 1.3016, 1.6449], device='cuda:1'), covar=tensor([0.0782, 0.0325, 0.0351, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0117, 0.0219], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 18:54:03,538 INFO [train.py:968] (1/2) Epoch 21, batch 6350, giga_loss[loss=0.4184, simple_loss=0.4417, pruned_loss=0.1975, over 27526.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3797, pruned_loss=0.1274, over 5652031.37 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3461, pruned_loss=0.09066, over 5484304.90 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.381, pruned_loss=0.1293, over 5646529.64 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 18:54:52,780 INFO [train.py:968] (1/2) Epoch 21, batch 6400, giga_loss[loss=0.3987, simple_loss=0.4377, pruned_loss=0.1799, over 27837.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3824, pruned_loss=0.1311, over 5635798.87 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3453, pruned_loss=0.09018, over 5494792.26 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3853, pruned_loss=0.1343, over 5626800.26 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 18:54:54,065 INFO [optim.py:369] (1/2) 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 18:55:03,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4254, 4.3816, 1.6693, 1.6100], device='cuda:1'), covar=tensor([0.0994, 0.0298, 0.0864, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0554, 0.0383, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 18:55:09,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3162, 1.4974, 1.5126, 1.3980], device='cuda:1'), covar=tensor([0.1568, 0.1277, 0.1708, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0753, 0.0716, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-10 18:55:44,206 INFO [train.py:968] (1/2) Epoch 21, batch 6450, giga_loss[loss=0.2896, simple_loss=0.3507, pruned_loss=0.1143, over 28836.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3861, pruned_loss=0.1354, over 5620849.56 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3451, pruned_loss=0.08994, over 5505257.42 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3898, pruned_loss=0.1395, over 5607459.41 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:56:37,187 INFO [train.py:968] (1/2) Epoch 21, batch 6500, libri_loss[loss=0.3096, simple_loss=0.3755, pruned_loss=0.1218, over 29536.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3898, pruned_loss=0.1378, over 5625540.49 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3452, pruned_loss=0.0901, over 5511129.97 frames. ], giga_tot_loss[loss=0.3381, simple_loss=0.3932, pruned_loss=0.1415, over 5611114.67 frames. ], batch size: 79, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:56:39,522 INFO [optim.py:369] (1/2) 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 18:57:17,233 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 6550, giga_loss[loss=0.2947, simple_loss=0.3605, pruned_loss=0.1145, over 28865.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3897, pruned_loss=0.1383, over 5638434.52 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3454, pruned_loss=0.09029, over 5519121.68 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.393, pruned_loss=0.142, over 5622045.48 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:58:13,416 INFO [train.py:968] (1/2) Epoch 21, batch 6600, giga_loss[loss=0.3154, simple_loss=0.3749, pruned_loss=0.1279, over 28927.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3875, pruned_loss=0.1371, over 5636169.55 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3453, pruned_loss=0.09025, over 5518575.45 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3912, pruned_loss=0.1411, over 5626731.36 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:58:17,553 INFO [optim.py:369] (1/2) 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 18:59:03,128 INFO [train.py:968] (1/2) Epoch 21, batch 6650, giga_loss[loss=0.3346, simple_loss=0.4033, pruned_loss=0.133, over 28783.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3869, pruned_loss=0.1357, over 5626546.17 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3455, pruned_loss=0.0903, over 5516158.70 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3908, pruned_loss=0.1401, over 5624473.08 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:59:05,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=919002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:59:38,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=919037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:59:41,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=919040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 18:59:45,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1945, 4.0236, 3.8257, 1.9506], device='cuda:1'), covar=tensor([0.0630, 0.0742, 0.0775, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.1218, 0.1127, 0.0958, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 18:59:50,708 INFO [train.py:968] (1/2) Epoch 21, batch 6700, giga_loss[loss=0.2941, simple_loss=0.3636, pruned_loss=0.1123, over 28592.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.387, pruned_loss=0.1348, over 5641776.37 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3451, pruned_loss=0.09005, over 5524252.78 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3913, pruned_loss=0.1394, over 5634798.39 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 18:59:53,387 INFO [optim.py:369] (1/2) 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 19:00:07,193 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=919069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:00:36,607 INFO [train.py:968] (1/2) Epoch 21, batch 6750, giga_loss[loss=0.3559, simple_loss=0.4078, pruned_loss=0.152, over 28256.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3873, pruned_loss=0.1346, over 5624574.08 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3453, pruned_loss=0.09014, over 5539171.91 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3923, pruned_loss=0.1401, over 5610263.18 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:01:17,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=919145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:01:19,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=919148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:01:22,596 INFO [train.py:968] (1/2) Epoch 21, batch 6800, giga_loss[loss=0.2906, simple_loss=0.3671, pruned_loss=0.107, over 28799.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3862, pruned_loss=0.1334, over 5627689.56 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3455, pruned_loss=0.09028, over 5547736.47 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3916, pruned_loss=0.1394, over 5610924.85 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:01:24,867 INFO [optim.py:369] (1/2) 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 19:01:51,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=919177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:02:07,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-10 19:02:12,780 INFO [train.py:968] (1/2) Epoch 21, batch 6850, giga_loss[loss=0.3231, simple_loss=0.3958, pruned_loss=0.1252, over 28776.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3832, pruned_loss=0.1299, over 5629559.95 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3457, pruned_loss=0.0904, over 5558861.79 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3888, pruned_loss=0.1359, over 5608607.46 frames. ], batch size: 99, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:03:02,085 INFO [train.py:968] (1/2) Epoch 21, batch 6900, libri_loss[loss=0.2031, simple_loss=0.2863, pruned_loss=0.05998, over 29501.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3788, pruned_loss=0.1256, over 5646711.67 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3456, pruned_loss=0.09044, over 5565688.26 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.384, pruned_loss=0.1311, over 5625235.45 frames. ], batch size: 70, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:03:04,120 INFO [optim.py:369] (1/2) 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 19:03:51,761 INFO [train.py:968] (1/2) Epoch 21, batch 6950, giga_loss[loss=0.2809, simple_loss=0.3573, pruned_loss=0.1023, over 28975.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3776, pruned_loss=0.1246, over 5648489.61 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3459, pruned_loss=0.09069, over 5569180.00 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3817, pruned_loss=0.1291, over 5629274.64 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:04:12,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6047, 1.8643, 1.4992, 1.8838], device='cuda:1'), covar=tensor([0.2564, 0.2751, 0.3007, 0.2590], device='cuda:1'), in_proj_covar=tensor([0.1492, 0.1078, 0.1318, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 19:04:24,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5684, 1.7807, 1.4152, 1.5019], device='cuda:1'), covar=tensor([0.2633, 0.2788, 0.3179, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.1492, 0.1078, 0.1318, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 19:04:39,525 INFO [train.py:968] (1/2) Epoch 21, batch 7000, giga_loss[loss=0.2984, simple_loss=0.3621, pruned_loss=0.1174, over 28905.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3758, pruned_loss=0.1234, over 5641512.93 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3462, pruned_loss=0.09095, over 5557901.93 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3794, pruned_loss=0.1274, over 5637906.60 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:04:42,239 INFO [optim.py:369] (1/2) 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 19:04:50,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3935, 1.1457, 4.6778, 3.6262], device='cuda:1'), covar=tensor([0.1766, 0.2961, 0.0372, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0643, 0.0952, 0.0901], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 19:04:59,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2398, 1.5033, 1.5418, 1.1107], device='cuda:1'), covar=tensor([0.1697, 0.2536, 0.1413, 0.1651], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0698, 0.0936, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 19:05:24,044 INFO [train.py:968] (1/2) Epoch 21, batch 7050, giga_loss[loss=0.398, simple_loss=0.4467, pruned_loss=0.1746, over 28657.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3745, pruned_loss=0.1226, over 5655371.40 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.346, pruned_loss=0.09106, over 5570100.30 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3786, pruned_loss=0.1267, over 5643963.68 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:05:45,573 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-10 19:06:02,053 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=919434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:06:19,000 INFO [train.py:968] (1/2) Epoch 21, batch 7100, giga_loss[loss=0.2808, simple_loss=0.3647, pruned_loss=0.09843, over 28853.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.374, pruned_loss=0.1218, over 5653737.74 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3461, pruned_loss=0.09115, over 5565930.93 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3776, pruned_loss=0.1255, over 5650644.32 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:06:20,956 INFO [optim.py:369] (1/2) 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 19:06:47,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5562, 1.7363, 1.7739, 1.3103], device='cuda:1'), covar=tensor([0.1883, 0.2497, 0.1543, 0.1777], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0698, 0.0936, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 19:07:13,016 INFO [train.py:968] (1/2) Epoch 21, batch 7150, giga_loss[loss=0.262, simple_loss=0.3495, pruned_loss=0.08725, over 28943.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3707, pruned_loss=0.1184, over 5666091.33 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3459, pruned_loss=0.09106, over 5573939.42 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3743, pruned_loss=0.1221, over 5658375.66 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:08:09,374 INFO [train.py:968] (1/2) Epoch 21, batch 7200, giga_loss[loss=0.3242, simple_loss=0.3998, pruned_loss=0.1243, over 28678.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3723, pruned_loss=0.1174, over 5669933.69 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3462, pruned_loss=0.09135, over 5580367.69 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3755, pruned_loss=0.1207, over 5659293.38 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:08:13,837 INFO [optim.py:369] (1/2) 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 19:08:46,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-10 19:08:58,828 INFO [train.py:968] (1/2) Epoch 21, batch 7250, giga_loss[loss=0.2987, simple_loss=0.3752, pruned_loss=0.1111, over 28887.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3746, pruned_loss=0.1187, over 5663344.14 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3464, pruned_loss=0.09143, over 5583168.18 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3772, pruned_loss=0.1214, over 5653224.42 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:09:39,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4247, 1.5890, 1.6323, 1.2399], device='cuda:1'), covar=tensor([0.1649, 0.2411, 0.1355, 0.1634], device='cuda:1'), in_proj_covar=tensor([0.0893, 0.0698, 0.0936, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 19:09:48,139 INFO [train.py:968] (1/2) Epoch 21, batch 7300, giga_loss[loss=0.3215, simple_loss=0.3904, pruned_loss=0.1263, over 28693.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3746, pruned_loss=0.1194, over 5676816.32 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3458, pruned_loss=0.09123, over 5592188.64 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3782, pruned_loss=0.1227, over 5663192.95 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:09:51,667 INFO [optim.py:369] (1/2) 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 19:09:52,815 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 19:10:08,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6267, 4.7693, 1.8230, 2.0119], device='cuda:1'), covar=tensor([0.0996, 0.0290, 0.0861, 0.1198], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0556, 0.0384, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-10 19:10:34,862 INFO [train.py:968] (1/2) Epoch 21, batch 7350, giga_loss[loss=0.2627, simple_loss=0.3394, pruned_loss=0.09301, over 28948.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3718, pruned_loss=0.1177, over 5678366.18 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3453, pruned_loss=0.09105, over 5599774.85 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.376, pruned_loss=0.1214, over 5663131.63 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:10:38,908 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=919706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:11:13,001 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-10 19:11:22,077 INFO [train.py:968] (1/2) Epoch 21, batch 7400, giga_loss[loss=0.3292, simple_loss=0.3827, pruned_loss=0.1378, over 28913.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3702, pruned_loss=0.1178, over 5671917.72 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3454, pruned_loss=0.09106, over 5603532.14 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3739, pruned_loss=0.1213, over 5657638.64 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:11:27,635 INFO [optim.py:369] (1/2) 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 19:11:30,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9916, 1.1970, 0.9557, 0.7574], device='cuda:1'), covar=tensor([0.1066, 0.0537, 0.1325, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0448, 0.0515, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 19:11:58,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9568, 2.0067, 1.5036, 1.5349], device='cuda:1'), covar=tensor([0.0935, 0.0717, 0.1021, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0449, 0.0517, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 19:12:06,802 INFO [train.py:968] (1/2) Epoch 21, batch 7450, giga_loss[loss=0.2758, simple_loss=0.349, pruned_loss=0.1013, over 29009.00 frames. ], tot_loss[loss=0.302, simple_loss=0.369, pruned_loss=0.1175, over 5677600.01 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3458, pruned_loss=0.09117, over 5603857.20 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3721, pruned_loss=0.1206, over 5667410.72 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:12:13,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=919809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:12:35,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2914, 1.3033, 3.9400, 3.2417], device='cuda:1'), covar=tensor([0.1737, 0.2811, 0.0490, 0.0899], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0644, 0.0957, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 19:12:45,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3178, 2.6575, 2.4190, 2.2931], device='cuda:1'), covar=tensor([0.1919, 0.2071, 0.2042, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0755, 0.0718, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-10 19:12:57,643 INFO [train.py:968] (1/2) Epoch 21, batch 7500, giga_loss[loss=0.2825, simple_loss=0.363, pruned_loss=0.101, over 28709.00 frames. ], tot_loss[loss=0.301, simple_loss=0.369, pruned_loss=0.1165, over 5680426.49 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3455, pruned_loss=0.09114, over 5598307.30 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3718, pruned_loss=0.1193, over 5678155.74 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:13:02,135 INFO [optim.py:369] (1/2) 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 19:13:42,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4075, 1.5817, 1.1834, 1.0982], device='cuda:1'), covar=tensor([0.0937, 0.0522, 0.1060, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0448, 0.0516, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 19:13:43,947 INFO [train.py:968] (1/2) Epoch 21, batch 7550, giga_loss[loss=0.3088, simple_loss=0.3886, pruned_loss=0.1145, over 29018.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3687, pruned_loss=0.1153, over 5682434.85 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3456, pruned_loss=0.09118, over 5597735.16 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3717, pruned_loss=0.1182, over 5684390.18 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:14:12,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-10 19:14:16,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 19:14:28,852 INFO [train.py:968] (1/2) Epoch 21, batch 7600, giga_loss[loss=0.2781, simple_loss=0.3509, pruned_loss=0.1027, over 28588.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3686, pruned_loss=0.1152, over 5683724.17 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3457, pruned_loss=0.0912, over 5603347.81 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3713, pruned_loss=0.118, over 5681726.56 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:14:29,814 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=919952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:14:32,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=919955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:14:32,845 INFO [optim.py:369] (1/2) 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 19:14:59,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=919984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:15:12,159 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=919997.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:15:16,473 INFO [train.py:968] (1/2) Epoch 21, batch 7650, giga_loss[loss=0.2655, simple_loss=0.3392, pruned_loss=0.0959, over 28953.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3677, pruned_loss=0.1153, over 5689582.01 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3459, pruned_loss=0.0913, over 5607709.48 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.37, pruned_loss=0.1177, over 5685255.52 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:16:07,382 INFO [train.py:968] (1/2) Epoch 21, batch 7700, giga_loss[loss=0.3807, simple_loss=0.414, pruned_loss=0.1737, over 26734.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3676, pruned_loss=0.1165, over 5685046.69 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3459, pruned_loss=0.0913, over 5612512.36 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3697, pruned_loss=0.1187, over 5678512.96 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:16:11,821 INFO [optim.py:369] (1/2) 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 19:16:32,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2821, 1.3606, 3.4294, 2.9845], device='cuda:1'), covar=tensor([0.1500, 0.2590, 0.0453, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0644, 0.0955, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 19:16:36,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=920081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:16:53,593 INFO [train.py:968] (1/2) Epoch 21, batch 7750, giga_loss[loss=0.2968, simple_loss=0.3671, pruned_loss=0.1132, over 29000.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3675, pruned_loss=0.1169, over 5684991.91 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3459, pruned_loss=0.09122, over 5614426.09 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3696, pruned_loss=0.1193, over 5679819.26 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:17:03,782 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 19:17:07,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-10 19:17:25,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1510, 2.3778, 2.2505, 1.8415], device='cuda:1'), covar=tensor([0.2793, 0.2355, 0.2271, 0.2806], device='cuda:1'), in_proj_covar=tensor([0.1950, 0.1879, 0.1808, 0.1948], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 19:17:38,486 INFO [train.py:968] (1/2) Epoch 21, batch 7800, giga_loss[loss=0.2923, simple_loss=0.3603, pruned_loss=0.1121, over 28618.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3654, pruned_loss=0.1154, over 5701956.14 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.0907, over 5625977.71 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3687, pruned_loss=0.1188, over 5690159.68 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:17:44,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4928, 1.8208, 1.4683, 1.7029], device='cuda:1'), covar=tensor([0.2544, 0.2678, 0.2967, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1080, 0.1317, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 19:17:44,387 INFO [optim.py:369] (1/2) 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 19:18:29,940 INFO [train.py:968] (1/2) Epoch 21, batch 7850, giga_loss[loss=0.2553, simple_loss=0.3272, pruned_loss=0.0917, over 28746.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3639, pruned_loss=0.1151, over 5699567.77 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.09071, over 5626651.46 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3667, pruned_loss=0.1181, over 5690655.74 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:18:44,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=920216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:18:51,575 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=920224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:18:53,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=920227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:19:00,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4655, 1.6198, 1.5185, 1.4327], device='cuda:1'), covar=tensor([0.1677, 0.2162, 0.2208, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0754, 0.0717, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-10 19:19:18,206 INFO [train.py:968] (1/2) Epoch 21, batch 7900, giga_loss[loss=0.2931, simple_loss=0.3605, pruned_loss=0.1128, over 28656.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3633, pruned_loss=0.1152, over 5699762.78 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3452, pruned_loss=0.09075, over 5625555.21 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3655, pruned_loss=0.1178, over 5694532.66 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:19:22,674 INFO [optim.py:369] (1/2) 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 19:19:23,313 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=920256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:19:23,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8530, 1.9318, 1.4873, 1.5559], device='cuda:1'), covar=tensor([0.0941, 0.0668, 0.1055, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0449, 0.0516, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 19:20:04,743 INFO [train.py:968] (1/2) Epoch 21, batch 7950, giga_loss[loss=0.3187, simple_loss=0.3744, pruned_loss=0.1315, over 27685.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3643, pruned_loss=0.1159, over 5684891.40 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3457, pruned_loss=0.09104, over 5622330.19 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3661, pruned_loss=0.1182, over 5685286.93 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:20:50,263 INFO [train.py:968] (1/2) Epoch 21, batch 8000, giga_loss[loss=0.3059, simple_loss=0.378, pruned_loss=0.1169, over 28649.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3649, pruned_loss=0.1157, over 5684114.38 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3455, pruned_loss=0.09097, over 5628872.32 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 5680304.61 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:20:55,670 INFO [optim.py:369] (1/2) 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 19:21:06,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3267, 3.1226, 2.9797, 1.4538], device='cuda:1'), covar=tensor([0.1062, 0.1288, 0.1245, 0.2390], device='cuda:1'), in_proj_covar=tensor([0.1221, 0.1132, 0.0961, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 19:21:06,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=920372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:21:33,960 INFO [train.py:968] (1/2) Epoch 21, batch 8050, giga_loss[loss=0.2749, simple_loss=0.3553, pruned_loss=0.09728, over 29036.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3647, pruned_loss=0.1148, over 5679332.43 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3458, pruned_loss=0.09107, over 5635244.08 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3667, pruned_loss=0.1175, over 5671691.45 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:22:06,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6988, 2.5642, 2.7542, 2.2494], device='cuda:1'), covar=tensor([0.1697, 0.2344, 0.1818, 0.2338], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0751, 0.0715, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-10 19:22:17,327 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-10 19:22:18,897 INFO [train.py:968] (1/2) Epoch 21, batch 8100, giga_loss[loss=0.318, simple_loss=0.3782, pruned_loss=0.129, over 28680.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3641, pruned_loss=0.1141, over 5685247.52 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3454, pruned_loss=0.09088, over 5643547.73 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3666, pruned_loss=0.1172, over 5672462.53 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:22:24,346 INFO [optim.py:369] (1/2) 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 19:23:07,378 INFO [train.py:968] (1/2) Epoch 21, batch 8150, giga_loss[loss=0.3204, simple_loss=0.3846, pruned_loss=0.1281, over 28842.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3651, pruned_loss=0.1149, over 5684631.60 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3455, pruned_loss=0.09089, over 5641336.55 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3677, pruned_loss=0.118, over 5677598.57 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:23:22,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=920515.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:23:24,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=920518.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:23:58,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=920547.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:24:01,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 19:24:01,295 INFO [train.py:968] (1/2) Epoch 21, batch 8200, giga_loss[loss=0.2743, simple_loss=0.3466, pruned_loss=0.101, over 28925.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3685, pruned_loss=0.1191, over 5679241.98 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3452, pruned_loss=0.09075, over 5642999.78 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.1219, over 5672534.82 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:24:08,094 INFO [optim.py:369] (1/2) 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 19:24:24,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6730, 1.2866, 5.0449, 3.5775], device='cuda:1'), covar=tensor([0.1678, 0.2841, 0.0410, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0647, 0.0959, 0.0906], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 19:24:39,949 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=920591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:24:48,999 INFO [train.py:968] (1/2) Epoch 21, batch 8250, giga_loss[loss=0.4106, simple_loss=0.4375, pruned_loss=0.1918, over 24048.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3691, pruned_loss=0.1204, over 5677107.20 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09044, over 5646317.01 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1238, over 5669675.70 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:25:33,634 INFO [train.py:968] (1/2) Epoch 21, batch 8300, giga_loss[loss=0.2787, simple_loss=0.3566, pruned_loss=0.1004, over 29035.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.37, pruned_loss=0.1217, over 5678909.21 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3446, pruned_loss=0.09045, over 5656532.50 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3735, pruned_loss=0.1255, over 5664708.04 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:25:42,093 INFO [optim.py:369] (1/2) 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 19:26:21,609 INFO [train.py:968] (1/2) Epoch 21, batch 8350, giga_loss[loss=0.3734, simple_loss=0.4165, pruned_loss=0.1652, over 23836.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3687, pruned_loss=0.121, over 5677008.16 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3441, pruned_loss=0.0902, over 5662852.05 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3725, pruned_loss=0.1251, over 5660443.02 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:26:33,116 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=920715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:26:49,853 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=920734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:26:52,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=920737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:27:02,929 INFO [train.py:968] (1/2) Epoch 21, batch 8400, giga_loss[loss=0.2732, simple_loss=0.3498, pruned_loss=0.09834, over 28983.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3674, pruned_loss=0.1198, over 5679944.88 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.344, pruned_loss=0.09004, over 5667028.23 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3718, pruned_loss=0.1246, over 5662844.45 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:27:07,775 INFO [optim.py:369] (1/2) 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 19:27:15,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=920766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:27:46,890 INFO [train.py:968] (1/2) Epoch 21, batch 8450, libri_loss[loss=0.2068, simple_loss=0.2883, pruned_loss=0.06264, over 29631.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3677, pruned_loss=0.1187, over 5686751.44 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3436, pruned_loss=0.08986, over 5672957.40 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3725, pruned_loss=0.1238, over 5668001.42 frames. ], batch size: 69, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:27:48,046 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-10 19:27:54,560 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=920809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:28:14,797 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-10 19:28:31,991 INFO [train.py:968] (1/2) Epoch 21, batch 8500, giga_loss[loss=0.3096, simple_loss=0.3848, pruned_loss=0.1172, over 28584.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3662, pruned_loss=0.1171, over 5681916.08 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3441, pruned_loss=0.09009, over 5676959.35 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3701, pruned_loss=0.1216, over 5663453.55 frames. ], batch size: 85, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:28:36,414 INFO [optim.py:369] (1/2) 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 19:29:14,233 INFO [train.py:968] (1/2) Epoch 21, batch 8550, giga_loss[loss=0.286, simple_loss=0.3584, pruned_loss=0.1068, over 28953.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3639, pruned_loss=0.1159, over 5690908.47 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3438, pruned_loss=0.09001, over 5685833.95 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3681, pruned_loss=0.1206, over 5668003.92 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:29:30,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-10 19:29:46,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1974, 1.2738, 1.1246, 0.8700], device='cuda:1'), covar=tensor([0.0884, 0.0441, 0.0999, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0446, 0.0513, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 19:30:03,572 INFO [train.py:968] (1/2) Epoch 21, batch 8600, giga_loss[loss=0.3929, simple_loss=0.4296, pruned_loss=0.1781, over 27964.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3635, pruned_loss=0.1168, over 5675037.95 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08987, over 5678182.92 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3674, pruned_loss=0.1211, over 5663433.01 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:30:09,280 INFO [optim.py:369] (1/2) 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 19:30:30,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-10 19:30:49,964 INFO [train.py:968] (1/2) Epoch 21, batch 8650, giga_loss[loss=0.2982, simple_loss=0.3656, pruned_loss=0.1154, over 28760.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.365, pruned_loss=0.1181, over 5663180.74 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3444, pruned_loss=0.09041, over 5683209.56 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.368, pruned_loss=0.1221, over 5648850.83 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:31:11,464 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 19:31:27,398 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6347, 1.7083, 1.4138, 1.6656], device='cuda:1'), covar=tensor([0.2879, 0.3056, 0.3285, 0.2776], device='cuda:1'), in_proj_covar=tensor([0.1494, 0.1081, 0.1318, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 19:31:31,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9262, 2.1080, 1.9172, 1.8123], device='cuda:1'), covar=tensor([0.1977, 0.2512, 0.2298, 0.2580], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0746, 0.0712, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 19:31:34,479 INFO [train.py:968] (1/2) Epoch 21, batch 8700, giga_loss[loss=0.3432, simple_loss=0.4179, pruned_loss=0.1343, over 28898.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3672, pruned_loss=0.118, over 5674440.31 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08986, over 5691847.34 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1228, over 5654462.87 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:31:37,639 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:31:43,265 INFO [optim.py:369] (1/2) 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 19:32:10,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:32:18,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3788, 1.7289, 1.4783, 1.5834], device='cuda:1'), covar=tensor([0.0673, 0.0292, 0.0293, 0.0688], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 19:32:19,541 INFO [train.py:968] (1/2) Epoch 21, batch 8750, giga_loss[loss=0.354, simple_loss=0.4087, pruned_loss=0.1497, over 28736.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3686, pruned_loss=0.1168, over 5665293.79 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3428, pruned_loss=0.08967, over 5685700.37 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3734, pruned_loss=0.1217, over 5654949.52 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:32:35,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:32:58,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3956, 1.2485, 4.1888, 3.4376], device='cuda:1'), covar=tensor([0.1647, 0.2789, 0.0452, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0645, 0.0957, 0.0906], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 19:33:04,192 INFO [train.py:968] (1/2) Epoch 21, batch 8800, giga_loss[loss=0.2929, simple_loss=0.3551, pruned_loss=0.1154, over 28456.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3693, pruned_loss=0.1161, over 5679548.26 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08954, over 5691997.10 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3744, pruned_loss=0.1211, over 5665170.44 frames. ], batch size: 78, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:33:13,138 INFO [optim.py:369] (1/2) 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 19:33:15,005 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:33:35,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:33:49,868 INFO [train.py:968] (1/2) Epoch 21, batch 8850, giga_loss[loss=0.348, simple_loss=0.3842, pruned_loss=0.1559, over 23465.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3712, pruned_loss=0.1178, over 5673676.03 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.08945, over 5695094.14 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3758, pruned_loss=0.1222, over 5659268.44 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:33:55,963 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:15,891 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:20,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:22,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:34:36,241 INFO [train.py:968] (1/2) Epoch 21, batch 8900, giga_loss[loss=0.3532, simple_loss=0.3877, pruned_loss=0.1594, over 23513.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3721, pruned_loss=0.1192, over 5662153.63 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.342, pruned_loss=0.08936, over 5698692.60 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3766, pruned_loss=0.1235, over 5646954.81 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:34:44,946 INFO [optim.py:369] (1/2) 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 19:34:50,235 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:35:20,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 19:35:22,298 INFO [train.py:968] (1/2) Epoch 21, batch 8950, giga_loss[loss=0.2779, simple_loss=0.3381, pruned_loss=0.1088, over 28822.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3713, pruned_loss=0.1198, over 5666310.35 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3421, pruned_loss=0.08938, over 5702154.03 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3753, pruned_loss=0.1236, over 5650653.56 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:35:49,847 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:35:53,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6003, 2.1927, 1.5793, 0.9022], device='cuda:1'), covar=tensor([0.5746, 0.2900, 0.3937, 0.5991], device='cuda:1'), in_proj_covar=tensor([0.1736, 0.1637, 0.1583, 0.1403], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 19:35:58,920 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:36:11,432 INFO [train.py:968] (1/2) Epoch 21, batch 9000, giga_loss[loss=0.2749, simple_loss=0.337, pruned_loss=0.1064, over 28764.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3707, pruned_loss=0.1202, over 5653632.78 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3424, pruned_loss=0.08946, over 5705460.14 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.374, pruned_loss=0.1235, over 5637541.92 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:36:11,432 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 19:36:19,998 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 19:36:26,241 INFO [optim.py:369] (1/2) 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 19:36:26,503 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:37:08,670 INFO [train.py:968] (1/2) Epoch 21, batch 9050, giga_loss[loss=0.2832, simple_loss=0.3494, pruned_loss=0.1085, over 28827.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3694, pruned_loss=0.1201, over 5663887.12 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.08958, over 5707588.43 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.372, pruned_loss=0.1229, over 5649011.96 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:37:22,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7663, 4.6072, 4.3837, 2.2060], device='cuda:1'), covar=tensor([0.0455, 0.0601, 0.0656, 0.1966], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.1131, 0.0960, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 19:37:28,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:37:38,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:37:58,045 INFO [train.py:968] (1/2) Epoch 21, batch 9100, giga_loss[loss=0.3374, simple_loss=0.3874, pruned_loss=0.1437, over 27676.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3704, pruned_loss=0.1219, over 5668690.99 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08954, over 5709692.98 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1244, over 5654781.87 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:38:07,934 INFO [optim.py:369] (1/2) 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 19:38:36,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:38:48,807 INFO [train.py:968] (1/2) Epoch 21, batch 9150, giga_loss[loss=0.2805, simple_loss=0.3547, pruned_loss=0.1031, over 28792.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3721, pruned_loss=0.1233, over 5649329.44 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08989, over 5712555.39 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3742, pruned_loss=0.126, over 5633984.56 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:38:52,886 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:38:54,508 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-10 19:39:02,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8391, 5.0006, 1.9447, 2.1195], device='cuda:1'), covar=tensor([0.0943, 0.0317, 0.0848, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0557, 0.0385, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-10 19:39:20,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 9200, giga_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1154, over 28936.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3692, pruned_loss=0.1216, over 5668496.50 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3435, pruned_loss=0.09011, over 5715911.09 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3711, pruned_loss=0.1242, over 5652467.41 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:39:36,881 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5692, 1.7460, 1.7808, 1.3652], device='cuda:1'), covar=tensor([0.1810, 0.2442, 0.1473, 0.1679], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0702, 0.0938, 0.0837], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 19:39:45,637 INFO [optim.py:369] (1/2) 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 19:39:50,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7406, 2.4455, 1.6739, 0.8967], device='cuda:1'), covar=tensor([0.7859, 0.3777, 0.3560, 0.6908], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1641, 0.1588, 0.1408], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 19:39:55,922 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:39:57,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:08,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:25,526 INFO [train.py:968] (1/2) Epoch 21, batch 9250, giga_loss[loss=0.3193, simple_loss=0.3777, pruned_loss=0.1305, over 27563.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3673, pruned_loss=0.1208, over 5659727.77 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3433, pruned_loss=0.09006, over 5717677.06 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3692, pruned_loss=0.1231, over 5645077.30 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:40:27,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:27,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:52,504 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:54,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:40:57,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:41:06,869 INFO [train.py:968] (1/2) Epoch 21, batch 9300, giga_loss[loss=0.3023, simple_loss=0.3808, pruned_loss=0.1119, over 28999.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3668, pruned_loss=0.1191, over 5665450.53 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3432, pruned_loss=0.09002, over 5723450.11 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3692, pruned_loss=0.122, over 5646711.51 frames. ], batch size: 155, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:41:19,293 INFO [optim.py:369] (1/2) 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 19:41:26,070 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:41:38,649 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:41:41,483 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 9350, giga_loss[loss=0.3101, simple_loss=0.38, pruned_loss=0.1201, over 28729.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3676, pruned_loss=0.1189, over 5668785.48 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3432, pruned_loss=0.09002, over 5727008.72 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1217, over 5649593.30 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:42:05,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4561, 3.0212, 1.4103, 1.5903], device='cuda:1'), covar=tensor([0.0953, 0.0367, 0.0908, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0556, 0.0384, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-10 19:42:08,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:08,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:18,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4831, 1.9657, 1.6500, 1.6023], device='cuda:1'), covar=tensor([0.0613, 0.0252, 0.0265, 0.0657], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 19:42:25,758 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:27,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:32,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:40,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:43,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:44,530 INFO [train.py:968] (1/2) Epoch 21, batch 9400, giga_loss[loss=0.2618, simple_loss=0.3335, pruned_loss=0.09508, over 28847.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3673, pruned_loss=0.119, over 5667501.67 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3428, pruned_loss=0.08976, over 5728629.62 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3702, pruned_loss=0.1222, over 5649424.56 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:42:51,969 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:42:53,228 INFO [optim.py:369] (1/2) 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 19:43:12,358 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:43:26,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=921795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:43:32,528 INFO [train.py:968] (1/2) Epoch 21, batch 9450, giga_loss[loss=0.3057, simple_loss=0.3773, pruned_loss=0.117, over 28657.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3696, pruned_loss=0.1189, over 5669196.33 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08999, over 5731405.88 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.372, pruned_loss=0.1216, over 5651460.13 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:43:37,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6898, 1.8194, 1.9223, 1.4345], device='cuda:1'), covar=tensor([0.2163, 0.2593, 0.1729, 0.2016], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0706, 0.0942, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 19:43:56,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2580, 3.1585, 1.3968, 1.3917], device='cuda:1'), covar=tensor([0.1104, 0.0459, 0.1013, 0.1511], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0558, 0.0385, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-10 19:44:16,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0000, 1.5279, 5.0069, 3.8108], device='cuda:1'), covar=tensor([0.1479, 0.2763, 0.0437, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0646, 0.0961, 0.0910], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 19:44:17,189 INFO [train.py:968] (1/2) Epoch 21, batch 9500, giga_loss[loss=0.3059, simple_loss=0.3937, pruned_loss=0.1091, over 28758.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3707, pruned_loss=0.1176, over 5672978.68 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3427, pruned_loss=0.08974, over 5735289.24 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3734, pruned_loss=0.1205, over 5654465.93 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:44:18,123 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=921855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:44:20,544 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:44:25,254 INFO [optim.py:369] (1/2) 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 19:44:43,793 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 9550, giga_loss[loss=0.3045, simple_loss=0.3734, pruned_loss=0.1178, over 28724.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3739, pruned_loss=0.1185, over 5678155.83 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3428, pruned_loss=0.08973, over 5737582.25 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3764, pruned_loss=0.1211, over 5660800.28 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:45:42,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=921938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:45:44,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=921941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:45:53,530 INFO [train.py:968] (1/2) Epoch 21, batch 9600, libri_loss[loss=0.2382, simple_loss=0.3223, pruned_loss=0.07708, over 29579.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3763, pruned_loss=0.1206, over 5682605.62 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08959, over 5742781.95 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3794, pruned_loss=0.1238, over 5662160.17 frames. ], batch size: 75, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:46:01,207 INFO [optim.py:369] (1/2) 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 19:46:05,892 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 19:46:10,521 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=921970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:46:36,826 INFO [train.py:968] (1/2) Epoch 21, batch 9650, giga_loss[loss=0.2961, simple_loss=0.3648, pruned_loss=0.1137, over 28844.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3791, pruned_loss=0.1238, over 5690330.67 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3429, pruned_loss=0.08992, over 5745867.07 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3824, pruned_loss=0.127, over 5669704.44 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:46:51,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:47:00,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:47:03,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:47:26,265 INFO [train.py:968] (1/2) Epoch 21, batch 9700, giga_loss[loss=0.3286, simple_loss=0.3915, pruned_loss=0.1328, over 28930.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3795, pruned_loss=0.1252, over 5670145.06 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3429, pruned_loss=0.08984, over 5747859.91 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.383, pruned_loss=0.1286, over 5650446.62 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:47:30,757 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:47:36,844 INFO [optim.py:369] (1/2) 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 19:47:37,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 19:48:14,177 INFO [train.py:968] (1/2) Epoch 21, batch 9750, giga_loss[loss=0.284, simple_loss=0.3589, pruned_loss=0.1046, over 28905.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3779, pruned_loss=0.1239, over 5674962.28 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.343, pruned_loss=0.08986, over 5749315.96 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.381, pruned_loss=0.127, over 5657149.81 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:48:21,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:48:29,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2089, 1.7492, 1.5816, 1.5455], device='cuda:1'), covar=tensor([0.2181, 0.1767, 0.2371, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0745, 0.0709, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 19:48:47,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6523, 1.5417, 1.8482, 1.4337], device='cuda:1'), covar=tensor([0.1705, 0.2409, 0.1368, 0.1651], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0704, 0.0940, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 19:48:55,873 INFO [train.py:968] (1/2) Epoch 21, batch 9800, giga_loss[loss=0.3142, simple_loss=0.3897, pruned_loss=0.1193, over 29004.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.376, pruned_loss=0.1207, over 5679662.32 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3429, pruned_loss=0.08988, over 5752622.54 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3792, pruned_loss=0.1239, over 5660932.05 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:48:56,773 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-10 19:49:03,742 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922159.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:49:05,218 INFO [optim.py:369] (1/2) 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 19:49:05,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922162.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:49:30,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922191.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 19:49:38,220 INFO [train.py:968] (1/2) Epoch 21, batch 9850, giga_loss[loss=0.3021, simple_loss=0.3757, pruned_loss=0.1143, over 28975.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3769, pruned_loss=0.1203, over 5671276.39 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09015, over 5742897.22 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.38, pruned_loss=0.1234, over 5662634.10 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:50:04,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:50:28,114 INFO [train.py:968] (1/2) Epoch 21, batch 9900, giga_loss[loss=0.3406, simple_loss=0.395, pruned_loss=0.1431, over 28733.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.378, pruned_loss=0.1215, over 5670027.67 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3436, pruned_loss=0.09033, over 5745201.61 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3807, pruned_loss=0.1242, over 5660286.63 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:50:30,964 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:50:34,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:50:36,954 INFO [optim.py:369] (1/2) 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 19:50:55,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5980, 1.8228, 1.5237, 1.7274], device='cuda:1'), covar=tensor([0.2263, 0.2322, 0.2403, 0.2384], device='cuda:1'), in_proj_covar=tensor([0.1495, 0.1080, 0.1320, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 19:50:59,417 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 9950, giga_loss[loss=0.3222, simple_loss=0.3871, pruned_loss=0.1286, over 28880.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3768, pruned_loss=0.121, over 5636861.05 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3442, pruned_loss=0.09071, over 5707450.34 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3794, pruned_loss=0.1237, over 5660619.41 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 19:51:15,679 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 10000, giga_loss[loss=0.2843, simple_loss=0.3515, pruned_loss=0.1086, over 29036.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3745, pruned_loss=0.1203, over 5632350.15 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3443, pruned_loss=0.09064, over 5707199.29 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3773, pruned_loss=0.1233, over 5649159.50 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:52:09,899 INFO [optim.py:369] (1/2) 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 19:52:22,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:52:24,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=922376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:52:47,816 INFO [train.py:968] (1/2) Epoch 21, batch 10050, giga_loss[loss=0.2954, simple_loss=0.3612, pruned_loss=0.1148, over 28898.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3739, pruned_loss=0.1208, over 5641501.50 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3446, pruned_loss=0.09078, over 5706692.60 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3767, pruned_loss=0.1239, over 5653519.58 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:52:51,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:52:53,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8949, 1.9081, 1.7863, 1.6813], device='cuda:1'), covar=tensor([0.1685, 0.2044, 0.2209, 0.2097], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0744, 0.0709, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 19:52:56,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1960, 1.3416, 1.3768, 1.1747], device='cuda:1'), covar=tensor([0.2510, 0.2276, 0.1761, 0.2292], device='cuda:1'), in_proj_covar=tensor([0.1961, 0.1897, 0.1826, 0.1968], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 19:53:34,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-10 19:53:34,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6655, 1.9824, 1.9112, 1.4864], device='cuda:1'), covar=tensor([0.2109, 0.2800, 0.1769, 0.2053], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0706, 0.0944, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 19:53:35,330 INFO [train.py:968] (1/2) Epoch 21, batch 10100, libri_loss[loss=0.2304, simple_loss=0.3238, pruned_loss=0.06845, over 29526.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.37, pruned_loss=0.119, over 5649930.44 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3438, pruned_loss=0.09029, over 5713379.55 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3738, pruned_loss=0.1229, over 5651278.90 frames. ], batch size: 82, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:53:39,910 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.28 vs. limit=5.0
+2023-03-10 19:53:44,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5633, 4.3743, 4.1762, 1.9351], device='cuda:1'), covar=tensor([0.0527, 0.0674, 0.0735, 0.2270], device='cuda:1'), in_proj_covar=tensor([0.1231, 0.1143, 0.0967, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 19:53:45,537 INFO [optim.py:369] (1/2) 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 19:54:20,518 INFO [train.py:968] (1/2) Epoch 21, batch 10150, giga_loss[loss=0.3134, simple_loss=0.3724, pruned_loss=0.1272, over 28295.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3688, pruned_loss=0.1187, over 5642219.22 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3441, pruned_loss=0.09044, over 5711775.32 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3729, pruned_loss=0.1232, over 5641802.34 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:54:29,928 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-10 19:55:08,668 INFO [train.py:968] (1/2) Epoch 21, batch 10200, giga_loss[loss=0.2767, simple_loss=0.345, pruned_loss=0.1043, over 29013.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3694, pruned_loss=0.1197, over 5655811.46 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3443, pruned_loss=0.09055, over 5711112.55 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1235, over 5655238.50 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:55:17,630 INFO [optim.py:369] (1/2) 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 19:55:28,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-10 19:55:58,832 INFO [train.py:968] (1/2) Epoch 21, batch 10250, giga_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09707, over 28946.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3672, pruned_loss=0.1175, over 5655626.12 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3443, pruned_loss=0.09055, over 5711112.55 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3698, pruned_loss=0.1204, over 5655180.18 frames. ], batch size: 227, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:56:08,506 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=922612.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 19:56:43,976 INFO [train.py:968] (1/2) Epoch 21, batch 10300, giga_loss[loss=0.2757, simple_loss=0.3515, pruned_loss=0.09996, over 28676.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3634, pruned_loss=0.1137, over 5639244.58 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3443, pruned_loss=0.09054, over 5703113.69 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3661, pruned_loss=0.1167, over 5644827.04 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:56:54,849 INFO [optim.py:369] (1/2) 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 19:57:10,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:57:33,017 INFO [train.py:968] (1/2) Epoch 21, batch 10350, giga_loss[loss=0.2466, simple_loss=0.3342, pruned_loss=0.07952, over 29002.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3624, pruned_loss=0.1125, over 5649565.65 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3439, pruned_loss=0.09042, over 5698310.02 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3651, pruned_loss=0.1153, over 5657373.38 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:57:33,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4239, 1.7602, 1.3777, 1.4427], device='cuda:1'), covar=tensor([0.2952, 0.2956, 0.3374, 0.2542], device='cuda:1'), in_proj_covar=tensor([0.1495, 0.1082, 0.1319, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 19:58:20,856 INFO [train.py:968] (1/2) Epoch 21, batch 10400, giga_loss[loss=0.2803, simple_loss=0.3443, pruned_loss=0.1081, over 28954.00 frames. ], tot_loss[loss=0.294, simple_loss=0.362, pruned_loss=0.1131, over 5657329.30 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.344, pruned_loss=0.09046, over 5703436.84 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3645, pruned_loss=0.1158, over 5658100.91 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 19:58:33,095 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 10450, giga_loss[loss=0.2442, simple_loss=0.3253, pruned_loss=0.08149, over 28890.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.358, pruned_loss=0.1112, over 5667009.88 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3428, pruned_loss=0.08962, over 5709880.47 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3619, pruned_loss=0.1151, over 5660079.02 frames. ], batch size: 174, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:59:24,643 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=922820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 19:59:26,707 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 10500, giga_loss[loss=0.3629, simple_loss=0.4083, pruned_loss=0.1588, over 26658.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3587, pruned_loss=0.1117, over 5649099.83 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3431, pruned_loss=0.08981, over 5694673.45 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.115, over 5655453.33 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 19:59:54,133 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=922852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:00:04,167 INFO [optim.py:369] (1/2) 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 20:00:38,334 INFO [train.py:968] (1/2) Epoch 21, batch 10550, giga_loss[loss=0.288, simple_loss=0.3637, pruned_loss=0.1062, over 28987.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3623, pruned_loss=0.1135, over 5653975.39 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09011, over 5697540.69 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 5655409.48 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:00:52,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6771, 1.7090, 1.8734, 1.4509], device='cuda:1'), covar=tensor([0.1817, 0.2620, 0.1463, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0705, 0.0943, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 20:01:14,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4400, 1.6846, 1.5508, 1.5531], device='cuda:1'), covar=tensor([0.1976, 0.2148, 0.2485, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0746, 0.0711, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 20:01:24,532 INFO [train.py:968] (1/2) Epoch 21, batch 10600, giga_loss[loss=0.2903, simple_loss=0.3686, pruned_loss=0.106, over 28756.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3651, pruned_loss=0.1156, over 5652472.19 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09016, over 5701076.24 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1183, over 5649383.74 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:01:29,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-10 20:01:36,971 INFO [optim.py:369] (1/2) 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 20:01:48,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5687, 1.6612, 1.2649, 1.1713], device='cuda:1'), covar=tensor([0.0936, 0.0562, 0.1008, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0446, 0.0512, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 20:02:02,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=922987.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:02:15,156 INFO [train.py:968] (1/2) Epoch 21, batch 10650, giga_loss[loss=0.3573, simple_loss=0.3885, pruned_loss=0.1631, over 23458.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3656, pruned_loss=0.1167, over 5649426.76 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3442, pruned_loss=0.09035, over 5703084.48 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3671, pruned_loss=0.1189, over 5644830.84 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:02:33,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2753, 1.3830, 1.2870, 1.0260], device='cuda:1'), covar=tensor([0.0919, 0.0420, 0.0971, 0.0929], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0445, 0.0511, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 20:02:57,692 INFO [train.py:968] (1/2) Epoch 21, batch 10700, giga_loss[loss=0.3895, simple_loss=0.4208, pruned_loss=0.1791, over 26469.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3664, pruned_loss=0.1176, over 5646412.66 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3442, pruned_loss=0.09027, over 5696206.35 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.12, over 5647604.23 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:03:12,069 INFO [optim.py:369] (1/2) 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 20:03:17,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-10 20:03:26,795 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=923075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:03:36,909 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-10 20:03:45,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4124, 1.1782, 4.0872, 3.3467], device='cuda:1'), covar=tensor([0.1627, 0.2838, 0.0461, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0645, 0.0955, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 20:03:51,390 INFO [train.py:968] (1/2) Epoch 21, batch 10750, giga_loss[loss=0.3321, simple_loss=0.3898, pruned_loss=0.1372, over 28354.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3694, pruned_loss=0.1196, over 5648981.63 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09019, over 5699430.18 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3712, pruned_loss=0.122, over 5646115.74 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:04:17,802 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=923130.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:04:19,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=923133.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:04:35,072 INFO [train.py:968] (1/2) Epoch 21, batch 10800, giga_loss[loss=0.3237, simple_loss=0.391, pruned_loss=0.1282, over 28689.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3702, pruned_loss=0.1194, over 5649081.55 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3446, pruned_loss=0.09032, over 5691671.11 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3718, pruned_loss=0.1219, over 5651788.76 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:04:44,378 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:05:15,924 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6290, 1.8996, 1.6533, 1.9813], device='cuda:1'), covar=tensor([0.0729, 0.0289, 0.0310, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 20:05:21,325 INFO [train.py:968] (1/2) Epoch 21, batch 10850, libri_loss[loss=0.2297, simple_loss=0.3098, pruned_loss=0.07477, over 28609.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3707, pruned_loss=0.1196, over 5667478.47 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3446, pruned_loss=0.09033, over 5695978.81 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3726, pruned_loss=0.1224, over 5665029.04 frames. ], batch size: 63, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:05:26,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-10 20:06:05,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3694, 1.0425, 3.9769, 3.4249], device='cuda:1'), covar=tensor([0.1677, 0.2925, 0.0481, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0648, 0.0958, 0.0907], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 20:06:07,910 INFO [train.py:968] (1/2) Epoch 21, batch 10900, giga_loss[loss=0.3262, simple_loss=0.3849, pruned_loss=0.1337, over 28471.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3723, pruned_loss=0.1211, over 5676186.67 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3443, pruned_loss=0.09018, over 5702363.11 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3748, pruned_loss=0.1243, over 5667682.79 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:06:21,413 INFO [optim.py:369] (1/2) 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 20:06:59,851 INFO [train.py:968] (1/2) Epoch 21, batch 10950, giga_loss[loss=0.3205, simple_loss=0.3633, pruned_loss=0.1389, over 23864.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3725, pruned_loss=0.1201, over 5662350.70 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.08994, over 5704462.27 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3753, pruned_loss=0.1233, over 5653358.56 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:07:16,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5273, 4.3573, 4.0925, 1.9501], device='cuda:1'), covar=tensor([0.0718, 0.0834, 0.1071, 0.2134], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1138, 0.0967, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 20:07:48,347 INFO [train.py:968] (1/2) Epoch 21, batch 11000, giga_loss[loss=0.2891, simple_loss=0.3566, pruned_loss=0.1108, over 28829.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3712, pruned_loss=0.1194, over 5663937.63 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.08989, over 5711077.57 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3746, pruned_loss=0.123, over 5649645.16 frames. ], batch size: 119, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:07:59,560 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 11050, giga_loss[loss=0.2837, simple_loss=0.3574, pruned_loss=0.105, over 28948.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3709, pruned_loss=0.1195, over 5651909.50 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3443, pruned_loss=0.09023, over 5699656.31 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3743, pruned_loss=0.1235, over 5648704.93 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:09:03,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-10 20:09:27,537 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 11100, giga_loss[loss=0.2989, simple_loss=0.3616, pruned_loss=0.1181, over 28338.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3718, pruned_loss=0.1213, over 5639470.56 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3447, pruned_loss=0.0904, over 5694453.70 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3747, pruned_loss=0.1248, over 5639880.00 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:09:43,009 INFO [optim.py:369] (1/2) 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 20:10:15,326 INFO [train.py:968] (1/2) Epoch 21, batch 11150, libri_loss[loss=0.2681, simple_loss=0.3591, pruned_loss=0.08852, over 29352.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3696, pruned_loss=0.1201, over 5644941.60 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3446, pruned_loss=0.09019, over 5696186.21 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5642392.08 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:10:44,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8900, 5.7205, 5.4613, 3.0742], device='cuda:1'), covar=tensor([0.0468, 0.0621, 0.0649, 0.1471], device='cuda:1'), in_proj_covar=tensor([0.1224, 0.1135, 0.0963, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 20:11:01,514 INFO [train.py:968] (1/2) Epoch 21, batch 11200, giga_loss[loss=0.3053, simple_loss=0.3691, pruned_loss=0.1207, over 28596.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3693, pruned_loss=0.1208, over 5651122.10 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3441, pruned_loss=0.08994, over 5698453.57 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3723, pruned_loss=0.1242, over 5646716.56 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:11:16,424 INFO [optim.py:369] (1/2) 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 20:11:19,334 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=923569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:11:41,557 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=923593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:11:44,821 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 11250, giga_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1033, over 28932.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3682, pruned_loss=0.1202, over 5645379.75 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3446, pruned_loss=0.09025, over 5692959.76 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3706, pruned_loss=0.1233, over 5645960.79 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:12:15,327 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 11300, giga_loss[loss=0.3162, simple_loss=0.3761, pruned_loss=0.1282, over 28972.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5652272.51 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3446, pruned_loss=0.09021, over 5696899.14 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3709, pruned_loss=0.1237, over 5648173.16 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:12:42,098 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-10 20:12:50,567 INFO [optim.py:369] (1/2) 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 20:13:25,848 INFO [train.py:968] (1/2) Epoch 21, batch 11350, giga_loss[loss=0.3599, simple_loss=0.4119, pruned_loss=0.1539, over 28832.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3692, pruned_loss=0.1216, over 5648148.13 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3443, pruned_loss=0.09012, over 5696174.24 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3719, pruned_loss=0.1248, over 5644488.89 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:13:36,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-10 20:14:13,616 INFO [train.py:968] (1/2) Epoch 21, batch 11400, giga_loss[loss=0.3023, simple_loss=0.3703, pruned_loss=0.1171, over 28935.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3708, pruned_loss=0.1227, over 5643451.75 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3444, pruned_loss=0.09009, over 5699213.33 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3732, pruned_loss=0.1257, over 5637468.63 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:14:26,321 INFO [optim.py:369] (1/2) 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 20:14:30,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-10 20:15:00,207 INFO [train.py:968] (1/2) Epoch 21, batch 11450, giga_loss[loss=0.3153, simple_loss=0.3733, pruned_loss=0.1287, over 28646.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3705, pruned_loss=0.1228, over 5649730.58 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08999, over 5705472.82 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3733, pruned_loss=0.1262, over 5637941.00 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:15:35,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7405, 1.8840, 1.2983, 1.4970], device='cuda:1'), covar=tensor([0.0933, 0.0646, 0.1023, 0.1267], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0448, 0.0514, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 20:15:50,786 INFO [train.py:968] (1/2) Epoch 21, batch 11500, libri_loss[loss=0.2255, simple_loss=0.3026, pruned_loss=0.07422, over 29522.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3704, pruned_loss=0.1231, over 5659555.76 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.08991, over 5706467.46 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.373, pruned_loss=0.126, over 5649103.29 frames. ], batch size: 70, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:16:03,295 INFO [optim.py:369] (1/2) 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 20:16:16,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3513, 1.1540, 4.0866, 3.4498], device='cuda:1'), covar=tensor([0.1636, 0.2923, 0.0453, 0.1287], device='cuda:1'), in_proj_covar=tensor([0.0759, 0.0646, 0.0956, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 20:16:28,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4241, 1.2662, 3.9268, 3.1984], device='cuda:1'), covar=tensor([0.1646, 0.2809, 0.0497, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0759, 0.0646, 0.0955, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 20:16:35,778 INFO [train.py:968] (1/2) Epoch 21, batch 11550, giga_loss[loss=0.2717, simple_loss=0.3446, pruned_loss=0.09938, over 28868.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3715, pruned_loss=0.1236, over 5656858.17 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08945, over 5712316.34 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3751, pruned_loss=0.1274, over 5641506.22 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:16:44,341 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-10 20:17:15,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=923944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:17:18,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1007, 1.1242, 3.5902, 3.1930], device='cuda:1'), covar=tensor([0.1706, 0.2831, 0.0489, 0.0998], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0646, 0.0957, 0.0906], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 20:17:21,841 INFO [train.py:968] (1/2) Epoch 21, batch 11600, giga_loss[loss=0.3161, simple_loss=0.3776, pruned_loss=0.1273, over 28270.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3713, pruned_loss=0.1225, over 5673365.49 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3434, pruned_loss=0.08933, over 5716803.10 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3747, pruned_loss=0.1265, over 5656037.97 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:17:28,721 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4506, 3.5777, 1.5104, 1.5830], device='cuda:1'), covar=tensor([0.1000, 0.0373, 0.0912, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0558, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-10 20:17:34,643 INFO [optim.py:369] (1/2) 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 20:18:08,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2103, 1.7988, 1.6040, 1.3862], device='cuda:1'), covar=tensor([0.0836, 0.0293, 0.0283, 0.1017], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 20:18:13,378 INFO [train.py:968] (1/2) Epoch 21, batch 11650, giga_loss[loss=0.4033, simple_loss=0.4291, pruned_loss=0.1888, over 26641.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1235, over 5658717.69 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3435, pruned_loss=0.08941, over 5718516.69 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3756, pruned_loss=0.1272, over 5642466.99 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:18:13,620 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:19:03,978 INFO [train.py:968] (1/2) Epoch 21, batch 11700, giga_loss[loss=0.4118, simple_loss=0.4266, pruned_loss=0.1985, over 26537.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1253, over 5649202.79 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3437, pruned_loss=0.08957, over 5710563.73 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3772, pruned_loss=0.1282, over 5642902.73 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:19:08,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6986, 4.7141, 1.8080, 1.9288], device='cuda:1'), covar=tensor([0.0962, 0.0413, 0.0868, 0.1245], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0559, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-10 20:19:17,600 INFO [optim.py:369] (1/2) 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 20:19:36,197 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:19:39,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:19:45,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:19:48,876 INFO [train.py:968] (1/2) Epoch 21, batch 11750, giga_loss[loss=0.346, simple_loss=0.3998, pruned_loss=0.1461, over 28971.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.374, pruned_loss=0.125, over 5648409.05 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3438, pruned_loss=0.08955, over 5710929.35 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3766, pruned_loss=0.1282, over 5641431.52 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:20:07,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:20:08,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:20:30,794 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-10 20:20:35,456 INFO [train.py:968] (1/2) Epoch 21, batch 11800, giga_loss[loss=0.3724, simple_loss=0.4001, pruned_loss=0.1724, over 23814.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3753, pruned_loss=0.1245, over 5650531.31 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08975, over 5713221.05 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3777, pruned_loss=0.1276, over 5641825.08 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:20:49,038 INFO [optim.py:369] (1/2) 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 20:21:06,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6553, 1.4990, 1.6985, 1.2571], device='cuda:1'), covar=tensor([0.2150, 0.3343, 0.1668, 0.1803], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0705, 0.0943, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 20:21:19,194 INFO [train.py:968] (1/2) Epoch 21, batch 11850, giga_loss[loss=0.3147, simple_loss=0.3828, pruned_loss=0.1233, over 28683.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3751, pruned_loss=0.1233, over 5658495.36 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3446, pruned_loss=0.09009, over 5718540.99 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3777, pruned_loss=0.1266, over 5644709.11 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:22:06,140 INFO [train.py:968] (1/2) Epoch 21, batch 11900, giga_loss[loss=0.2746, simple_loss=0.35, pruned_loss=0.09955, over 28730.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3739, pruned_loss=0.1225, over 5657940.47 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3442, pruned_loss=0.08986, over 5718819.57 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3769, pruned_loss=0.126, over 5645846.44 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:22:07,244 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:22:50,141 INFO [train.py:968] (1/2) Epoch 21, batch 11950, libri_loss[loss=0.2753, simple_loss=0.3628, pruned_loss=0.09385, over 28672.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3712, pruned_loss=0.1208, over 5658247.52 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3441, pruned_loss=0.08986, over 5720741.29 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3742, pruned_loss=0.1243, over 5645456.36 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:23:26,673 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-10 20:23:34,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-10 20:23:38,082 INFO [train.py:968] (1/2) Epoch 21, batch 12000, giga_loss[loss=0.2852, simple_loss=0.3615, pruned_loss=0.1045, over 29005.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.371, pruned_loss=0.1202, over 5658118.71 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08948, over 5717471.73 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3751, pruned_loss=0.1246, over 5648925.93 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:23:38,082 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 20:23:46,876 INFO [train.py:1012] (1/2) Epoch 21, validation: loss=0.2074, simple_loss=0.3145, pruned_loss=0.05016, over 944034.00 frames. 
+2023-03-10 20:23:46,876 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 20:23:59,101 INFO [optim.py:369] (1/2) 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 20:24:09,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:24:22,767 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924391.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:24:32,027 INFO [train.py:968] (1/2) Epoch 21, batch 12050, giga_loss[loss=0.3143, simple_loss=0.377, pruned_loss=0.1258, over 28750.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3724, pruned_loss=0.1215, over 5646130.74 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08984, over 5709707.00 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1252, over 5645117.67 frames. ], batch size: 243, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:25:21,522 INFO [train.py:968] (1/2) Epoch 21, batch 12100, giga_loss[loss=0.2963, simple_loss=0.3552, pruned_loss=0.1187, over 28579.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3725, pruned_loss=0.1224, over 5664065.56 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3439, pruned_loss=0.08976, over 5712213.02 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3756, pruned_loss=0.1258, over 5660265.73 frames. ], batch size: 78, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:25:35,396 INFO [optim.py:369] (1/2) 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 20:25:39,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:26:02,971 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 12150, giga_loss[loss=0.2836, simple_loss=0.3572, pruned_loss=0.1049, over 28954.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3718, pruned_loss=0.1222, over 5668214.62 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3436, pruned_loss=0.08954, over 5719046.27 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3756, pruned_loss=0.1264, over 5657127.28 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:26:25,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:26:30,204 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 12200, giga_loss[loss=0.3238, simple_loss=0.3849, pruned_loss=0.1313, over 28709.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3726, pruned_loss=0.1229, over 5673956.72 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3432, pruned_loss=0.08923, over 5723180.82 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3767, pruned_loss=0.1273, over 5660167.31 frames. ], batch size: 242, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:26:55,068 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:27:06,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3147, 1.8085, 1.4781, 1.4918], device='cuda:1'), covar=tensor([0.0699, 0.0397, 0.0325, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 20:27:11,084 INFO [optim.py:369] (1/2) 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 20:27:17,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:27:19,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7860, 3.6186, 3.4320, 1.9292], device='cuda:1'), covar=tensor([0.0689, 0.0831, 0.0809, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.1134, 0.0965, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-10 20:27:39,079 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-10 20:27:41,867 INFO [train.py:968] (1/2) Epoch 21, batch 12250, giga_loss[loss=0.3208, simple_loss=0.3785, pruned_loss=0.1316, over 27823.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3735, pruned_loss=0.1236, over 5670345.98 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08911, over 5726961.81 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3774, pruned_loss=0.1278, over 5655108.73 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:27:57,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:27:59,844 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=924621.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:28:07,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:18,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:22,062 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:28:28,894 INFO [train.py:968] (1/2) Epoch 21, batch 12300, giga_loss[loss=0.3046, simple_loss=0.3698, pruned_loss=0.1197, over 28226.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3727, pruned_loss=0.1224, over 5673486.54 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3436, pruned_loss=0.08942, over 5720843.07 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3761, pruned_loss=0.1264, over 5664996.96 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:28:42,903 INFO [optim.py:369] (1/2) 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 20:28:47,983 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 12350, giga_loss[loss=0.2817, simple_loss=0.3653, pruned_loss=0.09903, over 28910.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3727, pruned_loss=0.1222, over 5648350.75 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08984, over 5707999.56 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3761, pruned_loss=0.1262, over 5650835.45 frames. ], batch size: 199, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:29:55,506 INFO [train.py:968] (1/2) Epoch 21, batch 12400, giga_loss[loss=0.2681, simple_loss=0.3479, pruned_loss=0.09418, over 29032.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3732, pruned_loss=0.1214, over 5656560.98 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3447, pruned_loss=0.09016, over 5701364.14 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3761, pruned_loss=0.1253, over 5662723.67 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:30:10,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924766.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:30:10,385 INFO [optim.py:369] (1/2) 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 20:30:13,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:30:16,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:30:45,189 INFO [train.py:968] (1/2) Epoch 21, batch 12450, giga_loss[loss=0.3155, simple_loss=0.3816, pruned_loss=0.1247, over 28699.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3725, pruned_loss=0.1207, over 5667530.82 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3447, pruned_loss=0.09015, over 5698938.42 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3752, pruned_loss=0.1242, over 5674219.69 frames. ], batch size: 284, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:30:46,166 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:31:29,986 INFO [train.py:968] (1/2) Epoch 21, batch 12500, giga_loss[loss=0.3045, simple_loss=0.3712, pruned_loss=0.1189, over 28578.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3715, pruned_loss=0.1209, over 5654753.18 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3448, pruned_loss=0.09027, over 5692532.29 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3742, pruned_loss=0.1243, over 5664314.46 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:31:38,107 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 12550, giga_loss[loss=0.2882, simple_loss=0.3518, pruned_loss=0.1123, over 28683.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3685, pruned_loss=0.1191, over 5662625.13 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3446, pruned_loss=0.09023, over 5699020.34 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3716, pruned_loss=0.1227, over 5663014.33 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:32:24,831 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=924909.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:32:26,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=924912.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:32:50,156 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924941.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:32:57,811 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 12600, giga_loss[loss=0.2559, simple_loss=0.3237, pruned_loss=0.09409, over 28679.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3666, pruned_loss=0.1185, over 5657977.02 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3451, pruned_loss=0.09052, over 5686443.67 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3696, pruned_loss=0.1223, over 5668131.93 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:33:16,887 INFO [optim.py:369] (1/2) 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 20:33:45,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=924996.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:33:48,179 INFO [train.py:968] (1/2) Epoch 21, batch 12650, giga_loss[loss=0.2801, simple_loss=0.3502, pruned_loss=0.105, over 28975.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3648, pruned_loss=0.1184, over 5660425.11 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3451, pruned_loss=0.09057, over 5678961.36 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3673, pruned_loss=0.1215, over 5674511.46 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:33:58,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:34:06,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-10 20:34:10,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7471, 2.5220, 1.5790, 0.7848], device='cuda:1'), covar=tensor([0.7570, 0.3593, 0.3854, 0.7080], device='cuda:1'), in_proj_covar=tensor([0.1736, 0.1634, 0.1582, 0.1411], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 20:34:31,548 INFO [train.py:968] (1/2) Epoch 21, batch 12700, giga_loss[loss=0.3414, simple_loss=0.3919, pruned_loss=0.1454, over 28700.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3638, pruned_loss=0.1171, over 5668578.83 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.346, pruned_loss=0.09074, over 5676827.86 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3661, pruned_loss=0.1209, over 5681584.29 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:34:51,257 INFO [optim.py:369] (1/2) 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 20:35:14,291 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:35:15,071 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:35:18,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:35:22,646 INFO [train.py:968] (1/2) Epoch 21, batch 12750, giga_loss[loss=0.3014, simple_loss=0.3702, pruned_loss=0.1163, over 28921.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3643, pruned_loss=0.1172, over 5668020.96 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3459, pruned_loss=0.09066, over 5679164.88 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3663, pruned_loss=0.1205, over 5675866.99 frames. ], batch size: 213, lr: 1.53e-03, grad_scale: 2.0
+2023-03-10 20:35:36,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1365, 1.4064, 1.4535, 1.1270], device='cuda:1'), covar=tensor([0.2544, 0.2143, 0.1336, 0.2045], device='cuda:1'), in_proj_covar=tensor([0.1963, 0.1899, 0.1821, 0.1969], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 20:35:46,648 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:36:00,550 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925139.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:36:02,994 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925142.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:36:11,635 INFO [train.py:968] (1/2) Epoch 21, batch 12800, giga_loss[loss=0.3056, simple_loss=0.3838, pruned_loss=0.1137, over 28515.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3623, pruned_loss=0.1137, over 5665896.69 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3456, pruned_loss=0.09058, over 5674234.62 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3644, pruned_loss=0.1167, over 5675873.46 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:36:27,809 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:36:28,790 INFO [optim.py:369] (1/2) 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 20:36:31,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6016, 1.6188, 1.2914, 1.2695], device='cuda:1'), covar=tensor([0.0783, 0.0390, 0.0809, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0448, 0.0514, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 20:36:32,852 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:36:59,674 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 12850, giga_loss[loss=0.2945, simple_loss=0.3535, pruned_loss=0.1177, over 26778.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3607, pruned_loss=0.1115, over 5658340.07 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3458, pruned_loss=0.09082, over 5678753.02 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3625, pruned_loss=0.1142, over 5662030.09 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:37:32,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925233.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:37:40,449 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 12900, libri_loss[loss=0.21, simple_loss=0.2915, pruned_loss=0.06423, over 29365.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3566, pruned_loss=0.1076, over 5665175.57 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3449, pruned_loss=0.09044, over 5687398.98 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3595, pruned_loss=0.1109, over 5659344.35 frames. ], batch size: 71, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:38:06,510 INFO [optim.py:369] (1/2) 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 20:38:35,895 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-10 20:38:37,480 INFO [train.py:968] (1/2) Epoch 21, batch 12950, giga_loss[loss=0.2777, simple_loss=0.337, pruned_loss=0.1092, over 26578.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.354, pruned_loss=0.1053, over 5654243.81 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3448, pruned_loss=0.09069, over 5679298.49 frames. ], giga_tot_loss[loss=0.2871, simple_loss=0.3569, pruned_loss=0.1086, over 5655814.62 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:39:27,200 INFO [train.py:968] (1/2) Epoch 21, batch 13000, giga_loss[loss=0.2621, simple_loss=0.3455, pruned_loss=0.08937, over 28872.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3522, pruned_loss=0.1026, over 5656438.97 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3447, pruned_loss=0.09067, over 5679473.99 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3547, pruned_loss=0.1052, over 5657398.57 frames. ], batch size: 112, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:39:41,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5276, 1.9990, 1.4574, 0.8126], device='cuda:1'), covar=tensor([0.5118, 0.2842, 0.3740, 0.5790], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1633, 0.1586, 0.1414], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 20:39:45,000 INFO [optim.py:369] (1/2) 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 20:39:53,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925376.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:39:55,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925379.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 20:39:57,708 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-10 20:40:04,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:40:18,127 INFO [train.py:968] (1/2) Epoch 21, batch 13050, giga_loss[loss=0.3297, simple_loss=0.3897, pruned_loss=0.1349, over 28329.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3514, pruned_loss=0.1009, over 5655946.61 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.344, pruned_loss=0.09036, over 5685176.10 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3542, pruned_loss=0.1035, over 5651188.86 frames. ], batch size: 369, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:40:27,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925408.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 20:40:28,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3719, 1.6066, 1.5363, 1.3651], device='cuda:1'), covar=tensor([0.2264, 0.1852, 0.1443, 0.1793], device='cuda:1'), in_proj_covar=tensor([0.1932, 0.1866, 0.1789, 0.1935], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 20:40:29,689 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:41:07,598 INFO [train.py:968] (1/2) Epoch 21, batch 13100, giga_loss[loss=0.25, simple_loss=0.3316, pruned_loss=0.08415, over 28991.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3511, pruned_loss=0.1003, over 5650880.04 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3439, pruned_loss=0.09046, over 5676122.91 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3536, pruned_loss=0.1025, over 5655177.88 frames. ], batch size: 136, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:41:24,667 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:41:41,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3523, 1.3746, 3.3906, 3.0856], device='cuda:1'), covar=tensor([0.1448, 0.2668, 0.0471, 0.1442], device='cuda:1'), in_proj_covar=tensor([0.0759, 0.0646, 0.0954, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 20:41:57,370 INFO [train.py:968] (1/2) Epoch 21, batch 13150, giga_loss[loss=0.2453, simple_loss=0.3204, pruned_loss=0.08511, over 27963.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3488, pruned_loss=0.09896, over 5653163.53 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3438, pruned_loss=0.09065, over 5679263.05 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3509, pruned_loss=0.1007, over 5653205.34 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:42:29,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:42:33,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:42:38,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:42:46,034 INFO [train.py:968] (1/2) Epoch 21, batch 13200, giga_loss[loss=0.2669, simple_loss=0.3461, pruned_loss=0.09388, over 28957.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3454, pruned_loss=0.09652, over 5667440.72 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3435, pruned_loss=0.09052, over 5684976.28 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3475, pruned_loss=0.09818, over 5661829.79 frames. ], batch size: 164, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:42:48,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3290, 1.6945, 1.3137, 0.8803], device='cuda:1'), covar=tensor([0.5107, 0.2782, 0.2871, 0.5099], device='cuda:1'), in_proj_covar=tensor([0.1745, 0.1638, 0.1591, 0.1417], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 20:42:58,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9643, 2.8510, 1.7007, 1.1676], device='cuda:1'), covar=tensor([0.7541, 0.3408, 0.4552, 0.6455], device='cuda:1'), in_proj_covar=tensor([0.1743, 0.1636, 0.1589, 0.1416], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 20:42:59,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:03,285 INFO [optim.py:369] (1/2) 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 20:43:05,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:37,041 INFO [train.py:968] (1/2) Epoch 21, batch 13250, giga_loss[loss=0.2659, simple_loss=0.3494, pruned_loss=0.09119, over 28899.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3452, pruned_loss=0.0961, over 5665941.70 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3431, pruned_loss=0.09037, over 5688438.87 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3472, pruned_loss=0.09762, over 5658085.59 frames. ], batch size: 145, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:43:46,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:48,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:43:51,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:44:06,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4770, 2.1310, 1.5747, 0.6124], device='cuda:1'), covar=tensor([0.4580, 0.2883, 0.4162, 0.5838], device='cuda:1'), in_proj_covar=tensor([0.1739, 0.1636, 0.1589, 0.1415], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 20:44:16,321 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:44:25,577 INFO [train.py:968] (1/2) Epoch 21, batch 13300, giga_loss[loss=0.253, simple_loss=0.3312, pruned_loss=0.08738, over 28780.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3435, pruned_loss=0.09516, over 5654174.87 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3425, pruned_loss=0.09021, over 5680523.63 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3457, pruned_loss=0.09661, over 5654374.26 frames. ], batch size: 92, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:44:44,393 INFO [optim.py:369] (1/2) 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 20:45:02,807 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:04,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:07,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:17,475 INFO [train.py:968] (1/2) Epoch 21, batch 13350, giga_loss[loss=0.2306, simple_loss=0.3182, pruned_loss=0.07152, over 28596.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3415, pruned_loss=0.09311, over 5661772.82 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3425, pruned_loss=0.09033, over 5684059.05 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3432, pruned_loss=0.09423, over 5658366.12 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:45:29,909 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:32,739 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:45:32,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:04,081 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 13400, giga_loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08817, over 28633.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3381, pruned_loss=0.09086, over 5662947.37 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3423, pruned_loss=0.09022, over 5687472.24 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3396, pruned_loss=0.0919, over 5656885.62 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:46:15,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:17,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:24,612 INFO [optim.py:369] (1/2) 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 20:46:40,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3784, 3.6907, 1.5727, 1.5591], device='cuda:1'), covar=tensor([0.1000, 0.0333, 0.0975, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0552, 0.0383, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 20:46:41,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=925786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:44,332 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:46:57,563 INFO [train.py:968] (1/2) Epoch 21, batch 13450, giga_loss[loss=0.2561, simple_loss=0.3337, pruned_loss=0.08928, over 28321.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3349, pruned_loss=0.08968, over 5657708.43 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3412, pruned_loss=0.08976, over 5695452.38 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3368, pruned_loss=0.09094, over 5644669.67 frames. ], batch size: 368, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:47:20,021 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=925820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:47:34,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2276, 1.5309, 1.3972, 1.1608], device='cuda:1'), covar=tensor([0.2230, 0.1899, 0.1398, 0.1983], device='cuda:1'), in_proj_covar=tensor([0.1926, 0.1850, 0.1770, 0.1921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 20:47:51,806 INFO [train.py:968] (1/2) Epoch 21, batch 13500, giga_loss[loss=0.3035, simple_loss=0.3564, pruned_loss=0.1253, over 26746.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3348, pruned_loss=0.09068, over 5650003.75 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3412, pruned_loss=0.08976, over 5696509.12 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3364, pruned_loss=0.09168, over 5638594.20 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:48:09,996 INFO [optim.py:369] (1/2) 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 20:48:49,700 INFO [train.py:968] (1/2) Epoch 21, batch 13550, giga_loss[loss=0.2673, simple_loss=0.3482, pruned_loss=0.09323, over 28947.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3354, pruned_loss=0.0911, over 5635937.38 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3413, pruned_loss=0.08989, over 5699365.90 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3365, pruned_loss=0.0918, over 5624084.65 frames. ], batch size: 186, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:49:20,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=925929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:49:25,588 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=925932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:49:46,614 INFO [train.py:968] (1/2) Epoch 21, batch 13600, libri_loss[loss=0.3231, simple_loss=0.3826, pruned_loss=0.1318, over 28680.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3382, pruned_loss=0.09143, over 5648052.56 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3416, pruned_loss=0.09019, over 5699769.14 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3387, pruned_loss=0.09169, over 5637868.92 frames. ], batch size: 106, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:49:57,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=925961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:50:06,413 INFO [optim.py:369] (1/2) 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 20:50:23,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4444, 1.7536, 1.4112, 1.5625], device='cuda:1'), covar=tensor([0.0774, 0.0325, 0.0345, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 20:50:41,421 INFO [train.py:968] (1/2) Epoch 21, batch 13650, giga_loss[loss=0.2501, simple_loss=0.3361, pruned_loss=0.08207, over 27984.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.339, pruned_loss=0.09175, over 5633244.60 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3412, pruned_loss=0.09026, over 5688157.87 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3396, pruned_loss=0.09195, over 5632731.96 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:51:41,341 INFO [train.py:968] (1/2) Epoch 21, batch 13700, giga_loss[loss=0.2575, simple_loss=0.3259, pruned_loss=0.09458, over 26740.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3389, pruned_loss=0.09198, over 5638227.31 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3406, pruned_loss=0.08998, over 5692418.87 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.34, pruned_loss=0.09246, over 5632810.69 frames. ], batch size: 555, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:51:53,283 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:52:23,512 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:52:27,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3712, 1.0066, 4.4592, 3.4269], device='cuda:1'), covar=tensor([0.1764, 0.3240, 0.0430, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0643, 0.0948, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 20:52:34,823 INFO [train.py:968] (1/2) Epoch 21, batch 13750, giga_loss[loss=0.2279, simple_loss=0.2931, pruned_loss=0.08132, over 24473.00 frames. ], tot_loss[loss=0.258, simple_loss=0.336, pruned_loss=0.08995, over 5654365.22 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3396, pruned_loss=0.08942, over 5701471.00 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3378, pruned_loss=0.0909, over 5639827.59 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:52:46,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:53:35,185 INFO [train.py:968] (1/2) Epoch 21, batch 13800, giga_loss[loss=0.2334, simple_loss=0.3233, pruned_loss=0.0717, over 28649.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3353, pruned_loss=0.08823, over 5650219.50 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3397, pruned_loss=0.08956, over 5702554.65 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3365, pruned_loss=0.08883, over 5637530.16 frames. ], batch size: 262, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:53:35,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4983, 1.6089, 1.7271, 1.3213], device='cuda:1'), covar=tensor([0.2051, 0.2841, 0.1695, 0.2020], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0695, 0.0941, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 20:53:57,868 INFO [optim.py:369] (1/2) 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 20:54:29,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:54:36,988 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.78 vs. limit=5.0
+2023-03-10 20:54:37,096 INFO [train.py:968] (1/2) Epoch 21, batch 13850, giga_loss[loss=0.2644, simple_loss=0.3149, pruned_loss=0.1069, over 24355.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3327, pruned_loss=0.08723, over 5655879.14 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3392, pruned_loss=0.08946, over 5708789.37 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3339, pruned_loss=0.08775, over 5639023.08 frames. ], batch size: 705, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:54:43,164 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:54:46,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:55:19,872 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 13900, giga_loss[loss=0.2531, simple_loss=0.3308, pruned_loss=0.08773, over 27668.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3321, pruned_loss=0.08791, over 5658055.92 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3392, pruned_loss=0.08947, over 5707159.88 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.333, pruned_loss=0.08828, over 5645355.77 frames. ], batch size: 472, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:56:01,183 INFO [optim.py:369] (1/2) 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 20:56:38,048 INFO [train.py:968] (1/2) Epoch 21, batch 13950, giga_loss[loss=0.255, simple_loss=0.3372, pruned_loss=0.08639, over 27771.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3311, pruned_loss=0.08733, over 5662144.41 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3392, pruned_loss=0.08947, over 5707159.88 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3319, pruned_loss=0.08762, over 5652259.72 frames. ], batch size: 474, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:57:23,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:57:25,205 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 14000, giga_loss[loss=0.2469, simple_loss=0.3352, pruned_loss=0.07929, over 28558.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3333, pruned_loss=0.08805, over 5672281.24 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3389, pruned_loss=0.08942, over 5711385.46 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.334, pruned_loss=0.08829, over 5659848.45 frames. ], batch size: 307, lr: 1.53e-03, grad_scale: 8.0
+2023-03-10 20:58:00,937 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=926370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 20:58:02,138 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 14050, giga_loss[loss=0.2564, simple_loss=0.3406, pruned_loss=0.08612, over 28509.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3354, pruned_loss=0.08854, over 5680741.39 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3385, pruned_loss=0.08934, over 5715006.62 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3361, pruned_loss=0.0888, over 5666936.58 frames. ], batch size: 336, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:59:44,460 INFO [train.py:968] (1/2) Epoch 21, batch 14100, giga_loss[loss=0.2778, simple_loss=0.3559, pruned_loss=0.0998, over 28969.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3321, pruned_loss=0.08665, over 5682361.60 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3387, pruned_loss=0.08946, over 5716884.94 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3326, pruned_loss=0.08674, over 5669755.89 frames. ], batch size: 120, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 20:59:45,660 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:00:05,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:00:09,147 INFO [optim.py:369] (1/2) 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 21:00:23,977 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-10 21:00:24,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:00:24,925 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 21:00:40,342 INFO [train.py:968] (1/2) Epoch 21, batch 14150, giga_loss[loss=0.2701, simple_loss=0.3448, pruned_loss=0.09771, over 28133.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3334, pruned_loss=0.08783, over 5682784.99 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3379, pruned_loss=0.08915, over 5723486.50 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3343, pruned_loss=0.08812, over 5665102.86 frames. ], batch size: 412, lr: 1.53e-03, grad_scale: 4.0
+2023-03-10 21:01:45,168 INFO [train.py:968] (1/2) Epoch 21, batch 14200, giga_loss[loss=0.2715, simple_loss=0.3672, pruned_loss=0.0879, over 28426.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3365, pruned_loss=0.08827, over 5654093.97 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3376, pruned_loss=0.08907, over 5707558.06 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3373, pruned_loss=0.08853, over 5652774.98 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:02:11,188 INFO [optim.py:369] (1/2) 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 21:02:49,804 INFO [train.py:968] (1/2) Epoch 21, batch 14250, giga_loss[loss=0.225, simple_loss=0.3308, pruned_loss=0.05962, over 28931.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3389, pruned_loss=0.08689, over 5658464.26 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3372, pruned_loss=0.08889, over 5710400.54 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3399, pruned_loss=0.08723, over 5654408.63 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:02:57,072 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:00,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:22,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:25,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:36,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=926641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:03:47,764 INFO [train.py:968] (1/2) Epoch 21, batch 14300, giga_loss[loss=0.2682, simple_loss=0.3565, pruned_loss=0.09, over 28644.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3392, pruned_loss=0.08653, over 5647463.28 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3371, pruned_loss=0.08883, over 5715069.02 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3401, pruned_loss=0.08681, over 5639074.66 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:03:58,441 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:04:37,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5266, 1.8164, 1.7296, 1.4513], device='cuda:1'), covar=tensor([0.2389, 0.1772, 0.1466, 0.1982], device='cuda:1'), in_proj_covar=tensor([0.1922, 0.1844, 0.1765, 0.1921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 21:04:49,631 INFO [train.py:968] (1/2) Epoch 21, batch 14350, giga_loss[loss=0.2853, simple_loss=0.3613, pruned_loss=0.1046, over 28176.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3394, pruned_loss=0.0861, over 5659740.82 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.337, pruned_loss=0.08884, over 5716153.67 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3402, pruned_loss=0.08631, over 5651943.55 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:05:27,467 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1743, 1.4581, 1.3566, 1.1372], device='cuda:1'), covar=tensor([0.2437, 0.1949, 0.1360, 0.1972], device='cuda:1'), in_proj_covar=tensor([0.1923, 0.1843, 0.1764, 0.1922], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 21:05:51,678 INFO [train.py:968] (1/2) Epoch 21, batch 14400, giga_loss[loss=0.2735, simple_loss=0.3499, pruned_loss=0.09848, over 28792.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3402, pruned_loss=0.08779, over 5659809.50 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3368, pruned_loss=0.08879, over 5710941.32 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3411, pruned_loss=0.08796, over 5657510.38 frames. ], batch size: 263, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:06:16,630 INFO [optim.py:369] (1/2) 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 21:06:31,124 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:06:33,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4515, 1.5343, 1.6481, 1.3208], device='cuda:1'), covar=tensor([0.1562, 0.2323, 0.1335, 0.1634], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0695, 0.0942, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 21:06:41,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-10 21:06:45,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4250, 1.6928, 1.7537, 1.4587], device='cuda:1'), covar=tensor([0.3313, 0.2110, 0.1658, 0.2474], device='cuda:1'), in_proj_covar=tensor([0.1918, 0.1840, 0.1760, 0.1918], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 21:06:54,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2574, 5.0981, 4.8188, 2.2398], device='cuda:1'), covar=tensor([0.0458, 0.0607, 0.0722, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.1100, 0.0933, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 21:06:55,765 INFO [train.py:968] (1/2) Epoch 21, batch 14450, giga_loss[loss=0.267, simple_loss=0.3433, pruned_loss=0.09532, over 28734.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.34, pruned_loss=0.08894, over 5639088.64 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3373, pruned_loss=0.08916, over 5691668.68 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3403, pruned_loss=0.08871, over 5653889.78 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:07:28,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=926821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:07:33,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=926826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:08:07,581 INFO [train.py:968] (1/2) Epoch 21, batch 14500, giga_loss[loss=0.2309, simple_loss=0.3209, pruned_loss=0.07045, over 28651.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3387, pruned_loss=0.08821, over 5655407.65 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3372, pruned_loss=0.08922, over 5694246.07 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.339, pruned_loss=0.08796, over 5663944.99 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:08:35,258 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-10 21:08:47,384 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 14550, giga_loss[loss=0.2427, simple_loss=0.3251, pruned_loss=0.08016, over 28443.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3333, pruned_loss=0.08504, over 5652413.76 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3372, pruned_loss=0.08922, over 5695357.84 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3336, pruned_loss=0.08484, over 5657992.86 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:10:29,615 INFO [train.py:968] (1/2) Epoch 21, batch 14600, giga_loss[loss=0.2389, simple_loss=0.321, pruned_loss=0.07841, over 28878.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3315, pruned_loss=0.08453, over 5658908.43 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3367, pruned_loss=0.08925, over 5699504.36 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.332, pruned_loss=0.08414, over 5658191.42 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:10:53,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3960, 1.6995, 1.6136, 1.2060], device='cuda:1'), covar=tensor([0.1793, 0.2662, 0.1487, 0.1842], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0696, 0.0944, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 21:10:54,133 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=926969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:10:57,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3146, 1.4654, 1.2888, 1.5811], device='cuda:1'), covar=tensor([0.0787, 0.0346, 0.0349, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 21:10:57,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=926972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:10:57,486 INFO [optim.py:369] (1/2) 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 21:11:04,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1745, 1.2659, 1.1628, 0.9077], device='cuda:1'), covar=tensor([0.0985, 0.0505, 0.1048, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0443, 0.0512, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 21:11:30,622 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-10 21:11:32,992 INFO [train.py:968] (1/2) Epoch 21, batch 14650, giga_loss[loss=0.2813, simple_loss=0.3642, pruned_loss=0.09919, over 28957.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3317, pruned_loss=0.08521, over 5663467.61 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3365, pruned_loss=0.08927, over 5694499.27 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3321, pruned_loss=0.08475, over 5667378.80 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:11:33,503 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:11:51,015 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 14700, giga_loss[loss=0.2495, simple_loss=0.3164, pruned_loss=0.09128, over 24702.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3359, pruned_loss=0.08694, over 5668082.28 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3364, pruned_loss=0.08913, over 5696596.81 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3362, pruned_loss=0.08661, over 5668786.12 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:12:57,968 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-10 21:12:58,698 INFO [optim.py:369] (1/2) 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 21:13:28,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-10 21:13:35,605 INFO [train.py:968] (1/2) Epoch 21, batch 14750, giga_loss[loss=0.2285, simple_loss=0.3108, pruned_loss=0.07306, over 28401.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3347, pruned_loss=0.08749, over 5672314.64 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3364, pruned_loss=0.08915, over 5697767.83 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3349, pruned_loss=0.08721, over 5671758.69 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:14:40,324 INFO [train.py:968] (1/2) Epoch 21, batch 14800, giga_loss[loss=0.2348, simple_loss=0.3178, pruned_loss=0.07594, over 28929.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3359, pruned_loss=0.08953, over 5665232.21 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3362, pruned_loss=0.08911, over 5699701.02 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3362, pruned_loss=0.08933, over 5662598.29 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:14:48,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:15:05,841 INFO [optim.py:369] (1/2) 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 21:15:37,242 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 14850, giga_loss[loss=0.2481, simple_loss=0.3304, pruned_loss=0.0829, over 28892.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3358, pruned_loss=0.08912, over 5663004.82 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3365, pruned_loss=0.08928, over 5699647.03 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3358, pruned_loss=0.08882, over 5660734.33 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:16:45,878 INFO [train.py:968] (1/2) Epoch 21, batch 14900, giga_loss[loss=0.2439, simple_loss=0.3387, pruned_loss=0.07458, over 28095.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3375, pruned_loss=0.08873, over 5668303.87 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3369, pruned_loss=0.08954, over 5704523.54 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3372, pruned_loss=0.08827, over 5661449.38 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:16:55,966 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-10 21:17:01,345 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:17:22,024 INFO [optim.py:369] (1/2) 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 21:18:00,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:01,101 INFO [train.py:968] (1/2) Epoch 21, batch 14950, libri_loss[loss=0.2097, simple_loss=0.2898, pruned_loss=0.06477, over 29529.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3377, pruned_loss=0.08883, over 5659977.25 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3368, pruned_loss=0.08968, over 5698571.70 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3375, pruned_loss=0.08832, over 5658954.09 frames. ], batch size: 70, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:18:04,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:14,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3555, 3.1813, 3.0328, 1.3540], device='cuda:1'), covar=tensor([0.0997, 0.1151, 0.1110, 0.2372], device='cuda:1'), in_proj_covar=tensor([0.1191, 0.1100, 0.0934, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 21:18:26,114 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:43,834 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:49,676 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:18:57,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:19:06,810 INFO [train.py:968] (1/2) Epoch 21, batch 15000, libri_loss[loss=0.2687, simple_loss=0.346, pruned_loss=0.09567, over 29344.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3342, pruned_loss=0.08725, over 5665081.56 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.336, pruned_loss=0.08934, over 5688687.92 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3348, pruned_loss=0.08706, over 5671970.80 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:19:06,811 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 21:19:15,256 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 21:19:42,005 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:19:44,071 INFO [optim.py:369] (1/2) 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 21:20:11,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:20:26,660 INFO [train.py:968] (1/2) Epoch 21, batch 15050, giga_loss[loss=0.2081, simple_loss=0.2878, pruned_loss=0.06425, over 28952.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3288, pruned_loss=0.08551, over 5670825.52 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3359, pruned_loss=0.08931, over 5689777.48 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3294, pruned_loss=0.08538, over 5675269.84 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:20:34,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:21:25,420 INFO [train.py:968] (1/2) Epoch 21, batch 15100, giga_loss[loss=0.3105, simple_loss=0.3693, pruned_loss=0.1259, over 26756.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08527, over 5669073.33 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.336, pruned_loss=0.08944, over 5686971.90 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3275, pruned_loss=0.08493, over 5674381.19 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 21:21:28,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4955, 1.4048, 4.3922, 3.3557], device='cuda:1'), covar=tensor([0.1603, 0.2697, 0.0395, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0640, 0.0940, 0.0887], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 21:21:53,373 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 15150, giga_loss[loss=0.2329, simple_loss=0.3231, pruned_loss=0.07139, over 28994.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3289, pruned_loss=0.08676, over 5670105.92 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3359, pruned_loss=0.08953, over 5688862.68 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08632, over 5671716.14 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 21:22:58,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:23:01,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:23:14,664 INFO [train.py:968] (1/2) Epoch 21, batch 15200, giga_loss[loss=0.2449, simple_loss=0.3231, pruned_loss=0.08338, over 28510.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3282, pruned_loss=0.08627, over 5667064.59 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3353, pruned_loss=0.08933, over 5692030.95 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3286, pruned_loss=0.08603, over 5664993.58 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:23:38,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:23:47,164 INFO [optim.py:369] (1/2) 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 21:24:17,284 INFO [train.py:968] (1/2) Epoch 21, batch 15250, giga_loss[loss=0.2342, simple_loss=0.3233, pruned_loss=0.07253, over 28707.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3258, pruned_loss=0.08394, over 5660266.26 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3353, pruned_loss=0.08932, over 5687934.17 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3259, pruned_loss=0.08365, over 5661542.97 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:24:40,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2240, 1.7000, 1.2808, 0.4855], device='cuda:1'), covar=tensor([0.3997, 0.2367, 0.3933, 0.5496], device='cuda:1'), in_proj_covar=tensor([0.1727, 0.1624, 0.1585, 0.1413], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 21:24:59,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5041, 1.6751, 1.4075, 1.6716], device='cuda:1'), covar=tensor([0.0757, 0.0299, 0.0338, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 21:24:59,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:25:21,777 INFO [train.py:968] (1/2) Epoch 21, batch 15300, giga_loss[loss=0.274, simple_loss=0.3498, pruned_loss=0.09908, over 28713.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3239, pruned_loss=0.08295, over 5655480.76 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3347, pruned_loss=0.08925, over 5693593.43 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3243, pruned_loss=0.08267, over 5650926.55 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:25:53,879 INFO [optim.py:369] (1/2) 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 21:26:24,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:26:31,177 INFO [train.py:968] (1/2) Epoch 21, batch 15350, giga_loss[loss=0.2272, simple_loss=0.3108, pruned_loss=0.07184, over 28708.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3238, pruned_loss=0.08274, over 5669807.57 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3343, pruned_loss=0.08912, over 5696683.11 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3243, pruned_loss=0.08255, over 5663261.16 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:27:34,395 INFO [train.py:968] (1/2) Epoch 21, batch 15400, giga_loss[loss=0.2464, simple_loss=0.3228, pruned_loss=0.08498, over 29048.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3243, pruned_loss=0.08222, over 5677668.88 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3344, pruned_loss=0.08913, over 5691136.73 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3245, pruned_loss=0.08194, over 5676767.92 frames. ], batch size: 120, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:27:39,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1260, 1.2983, 1.1394, 0.9600], device='cuda:1'), covar=tensor([0.0964, 0.0489, 0.1049, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0441, 0.0509, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 21:27:43,981 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:28:13,508 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:28:15,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:28:17,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=927784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:28:40,643 INFO [train.py:968] (1/2) Epoch 21, batch 15450, giga_loss[loss=0.2418, simple_loss=0.3198, pruned_loss=0.08192, over 27639.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3257, pruned_loss=0.08341, over 5681233.05 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3343, pruned_loss=0.08902, over 5694809.02 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3256, pruned_loss=0.08315, over 5676829.60 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:28:51,735 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:29:25,052 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:29:28,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:29:41,691 INFO [train.py:968] (1/2) Epoch 21, batch 15500, giga_loss[loss=0.2809, simple_loss=0.345, pruned_loss=0.1084, over 26895.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3255, pruned_loss=0.08366, over 5671884.85 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3342, pruned_loss=0.08894, over 5684997.21 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3254, pruned_loss=0.08339, over 5677069.69 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:30:05,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:30:08,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6957, 1.9932, 1.2816, 1.5121], device='cuda:1'), covar=tensor([0.1002, 0.0572, 0.1058, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0442, 0.0510, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 21:30:09,725 INFO [optim.py:369] (1/2) 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 21:30:32,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5389, 1.7246, 1.3595, 1.6916], device='cuda:1'), covar=tensor([0.2903, 0.2758, 0.3211, 0.2511], device='cuda:1'), in_proj_covar=tensor([0.1495, 0.1076, 0.1322, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 21:30:40,050 INFO [train.py:968] (1/2) Epoch 21, batch 15550, giga_loss[loss=0.2408, simple_loss=0.3317, pruned_loss=0.07497, over 28335.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3251, pruned_loss=0.08227, over 5667729.17 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3342, pruned_loss=0.08894, over 5688695.47 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3249, pruned_loss=0.08195, over 5668171.53 frames. ], batch size: 369, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:31:07,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2657, 1.8764, 1.4673, 0.4833], device='cuda:1'), covar=tensor([0.4441, 0.2765, 0.4175, 0.5932], device='cuda:1'), in_proj_covar=tensor([0.1722, 0.1623, 0.1583, 0.1411], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 21:31:12,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=927927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:31:16,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=927930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:31:41,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=927949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:31:43,089 INFO [train.py:968] (1/2) Epoch 21, batch 15600, giga_loss[loss=0.2265, simple_loss=0.3252, pruned_loss=0.06393, over 28772.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3286, pruned_loss=0.08308, over 5660953.50 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3342, pruned_loss=0.08893, over 5689722.99 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3283, pruned_loss=0.0828, over 5660380.31 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:31:58,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=927959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:32:12,817 INFO [optim.py:369] (1/2) 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 21:32:43,759 INFO [train.py:968] (1/2) Epoch 21, batch 15650, giga_loss[loss=0.2332, simple_loss=0.3201, pruned_loss=0.07312, over 28693.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3306, pruned_loss=0.08415, over 5663024.72 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3339, pruned_loss=0.08886, over 5695420.06 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3305, pruned_loss=0.08385, over 5656659.74 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:33:34,042 INFO [train.py:968] (1/2) Epoch 21, batch 15700, giga_loss[loss=0.2605, simple_loss=0.3489, pruned_loss=0.08604, over 28334.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3313, pruned_loss=0.08518, over 5648581.09 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3336, pruned_loss=0.08875, over 5685269.64 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3314, pruned_loss=0.08483, over 5651331.15 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:33:37,311 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-10 21:34:04,387 INFO [optim.py:369] (1/2) 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 21:34:31,689 INFO [train.py:968] (1/2) Epoch 21, batch 15750, giga_loss[loss=0.2428, simple_loss=0.3222, pruned_loss=0.08168, over 28916.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3311, pruned_loss=0.08563, over 5649825.85 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3331, pruned_loss=0.08852, over 5689897.63 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3316, pruned_loss=0.0855, over 5647143.01 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:35:13,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:35:33,212 INFO [train.py:968] (1/2) Epoch 21, batch 15800, libri_loss[loss=0.2316, simple_loss=0.2977, pruned_loss=0.0827, over 29375.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3282, pruned_loss=0.08395, over 5652567.32 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3328, pruned_loss=0.08842, over 5694094.05 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3288, pruned_loss=0.08383, over 5645142.79 frames. ], batch size: 67, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:36:03,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-10 21:36:03,359 INFO [optim.py:369] (1/2) 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 21:36:31,253 INFO [train.py:968] (1/2) Epoch 21, batch 15850, giga_loss[loss=0.2657, simple_loss=0.3304, pruned_loss=0.1004, over 26930.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3269, pruned_loss=0.08355, over 5666399.52 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3325, pruned_loss=0.08816, over 5699132.71 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3275, pruned_loss=0.08359, over 5655090.15 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:37:14,765 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-10 21:37:34,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4266, 1.3027, 4.1939, 3.4653], device='cuda:1'), covar=tensor([0.1667, 0.2879, 0.0451, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0640, 0.0939, 0.0887], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 21:37:35,499 INFO [train.py:968] (1/2) Epoch 21, batch 15900, giga_loss[loss=0.236, simple_loss=0.3267, pruned_loss=0.07264, over 28831.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3261, pruned_loss=0.08299, over 5670794.96 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3323, pruned_loss=0.08805, over 5701200.20 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3267, pruned_loss=0.08306, over 5659977.47 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:37:41,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3430, 3.3932, 1.4667, 1.6093], device='cuda:1'), covar=tensor([0.0983, 0.0294, 0.0929, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0549, 0.0383, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 21:38:04,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4246, 2.0373, 1.4839, 0.6620], device='cuda:1'), covar=tensor([0.5408, 0.2781, 0.3998, 0.6158], device='cuda:1'), in_proj_covar=tensor([0.1732, 0.1634, 0.1588, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 21:38:07,533 INFO [optim.py:369] (1/2) 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 21:38:08,948 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:38:12,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:38:36,288 INFO [train.py:968] (1/2) Epoch 21, batch 15950, giga_loss[loss=0.2671, simple_loss=0.3426, pruned_loss=0.09577, over 28325.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3286, pruned_loss=0.08402, over 5676759.40 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3321, pruned_loss=0.08793, over 5704509.28 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3292, pruned_loss=0.0841, over 5664582.82 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:38:41,874 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:38:56,713 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=928319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:39:06,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:39:14,562 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=928331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:39:41,869 INFO [train.py:968] (1/2) Epoch 21, batch 16000, giga_loss[loss=0.23, simple_loss=0.3157, pruned_loss=0.07212, over 28873.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3293, pruned_loss=0.08502, over 5667720.79 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3319, pruned_loss=0.08777, over 5708673.06 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3299, pruned_loss=0.08516, over 5653954.67 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:40:10,981 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-10 21:40:11,093 INFO [optim.py:369] (1/2) 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 21:40:37,888 INFO [train.py:968] (1/2) Epoch 21, batch 16050, giga_loss[loss=0.3065, simple_loss=0.3703, pruned_loss=0.1214, over 28944.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3308, pruned_loss=0.08593, over 5670973.64 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3309, pruned_loss=0.08733, over 5714065.98 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3321, pruned_loss=0.08637, over 5654307.60 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:40:43,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4699, 1.8520, 1.6316, 1.5940], device='cuda:1'), covar=tensor([0.1918, 0.2256, 0.2153, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.0453, 0.0728, 0.0693, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 21:41:39,933 INFO [train.py:968] (1/2) Epoch 21, batch 16100, giga_loss[loss=0.2383, simple_loss=0.3309, pruned_loss=0.07283, over 28911.00 frames. ], tot_loss[loss=0.254, simple_loss=0.334, pruned_loss=0.08701, over 5659217.28 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3312, pruned_loss=0.08746, over 5716093.30 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3348, pruned_loss=0.08724, over 5643832.29 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:42:01,191 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:42:03,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:42:09,743 INFO [optim.py:369] (1/2) 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 21:42:39,470 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:42:42,331 INFO [train.py:968] (1/2) Epoch 21, batch 16150, giga_loss[loss=0.2645, simple_loss=0.3432, pruned_loss=0.09286, over 28687.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3349, pruned_loss=0.08694, over 5657760.78 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3311, pruned_loss=0.08733, over 5718535.54 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3356, pruned_loss=0.08723, over 5642251.29 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:43:46,120 INFO [train.py:968] (1/2) Epoch 21, batch 16200, giga_loss[loss=0.2363, simple_loss=0.319, pruned_loss=0.07683, over 28360.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3339, pruned_loss=0.0866, over 5663559.45 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3309, pruned_loss=0.0872, over 5724574.43 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3349, pruned_loss=0.08694, over 5643237.42 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:44:17,071 INFO [optim.py:369] (1/2) 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 21:44:41,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5573, 1.6180, 1.7754, 1.4131], device='cuda:1'), covar=tensor([0.1558, 0.2281, 0.1315, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0691, 0.0939, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 21:44:49,245 INFO [train.py:968] (1/2) Epoch 21, batch 16250, giga_loss[loss=0.2667, simple_loss=0.35, pruned_loss=0.09172, over 28630.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3331, pruned_loss=0.08676, over 5668162.76 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3312, pruned_loss=0.08739, over 5727234.80 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3336, pruned_loss=0.08686, over 5648461.95 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:45:51,945 INFO [train.py:968] (1/2) Epoch 21, batch 16300, libri_loss[loss=0.2211, simple_loss=0.2931, pruned_loss=0.07459, over 29395.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3329, pruned_loss=0.08681, over 5683076.31 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.331, pruned_loss=0.08729, over 5732238.27 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3337, pruned_loss=0.08697, over 5661141.05 frames. ], batch size: 67, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:46:25,000 INFO [optim.py:369] (1/2) 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 21:46:48,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:46:54,750 INFO [train.py:968] (1/2) Epoch 21, batch 16350, giga_loss[loss=0.2131, simple_loss=0.2931, pruned_loss=0.06657, over 29084.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3331, pruned_loss=0.08822, over 5661650.58 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3314, pruned_loss=0.0876, over 5719820.40 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3333, pruned_loss=0.08805, over 5654723.37 frames. ], batch size: 200, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:47:03,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=928706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:47:59,458 INFO [train.py:968] (1/2) Epoch 21, batch 16400, giga_loss[loss=0.2893, simple_loss=0.357, pruned_loss=0.1108, over 28122.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3303, pruned_loss=0.08698, over 5650305.71 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3313, pruned_loss=0.0876, over 5713188.88 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3306, pruned_loss=0.08685, over 5649565.21 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:48:02,441 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=928755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:48:27,889 INFO [optim.py:369] (1/2) 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 21:48:42,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3390, 1.3916, 1.2317, 1.5377], device='cuda:1'), covar=tensor([0.0812, 0.0350, 0.0363, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 21:48:58,005 INFO [train.py:968] (1/2) Epoch 21, batch 16450, giga_loss[loss=0.235, simple_loss=0.3207, pruned_loss=0.07459, over 28959.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3301, pruned_loss=0.08607, over 5658206.22 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3312, pruned_loss=0.0876, over 5715808.98 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3304, pruned_loss=0.08597, over 5654593.10 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:49:17,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8245, 1.6365, 5.1438, 3.5653], device='cuda:1'), covar=tensor([0.1554, 0.2614, 0.0391, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0641, 0.0940, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 21:49:28,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4201, 1.5230, 1.3248, 1.5653], device='cuda:1'), covar=tensor([0.0780, 0.0372, 0.0354, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 21:49:42,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:49:45,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:49:55,578 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=928849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:49:56,534 INFO [train.py:968] (1/2) Epoch 21, batch 16500, giga_loss[loss=0.2296, simple_loss=0.328, pruned_loss=0.0656, over 28974.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3291, pruned_loss=0.08394, over 5670317.99 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3311, pruned_loss=0.08758, over 5719338.54 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3293, pruned_loss=0.08381, over 5663092.89 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:49:57,914 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=928852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:50:17,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:50:24,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3458, 1.5214, 1.2699, 1.6748], device='cuda:1'), covar=tensor([0.0773, 0.0343, 0.0360, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 21:50:25,535 INFO [optim.py:369] (1/2) 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 21:50:31,104 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=928881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:50:53,270 INFO [train.py:968] (1/2) Epoch 21, batch 16550, giga_loss[loss=0.2448, simple_loss=0.3383, pruned_loss=0.07567, over 28896.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3297, pruned_loss=0.08208, over 5682558.48 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3304, pruned_loss=0.08732, over 5723752.24 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3305, pruned_loss=0.08211, over 5671883.40 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:51:34,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3435, 1.6224, 1.4442, 1.5424], device='cuda:1'), covar=tensor([0.0739, 0.0388, 0.0337, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 21:51:46,072 INFO [train.py:968] (1/2) Epoch 21, batch 16600, giga_loss[loss=0.2936, simple_loss=0.3538, pruned_loss=0.1167, over 26755.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3306, pruned_loss=0.08217, over 5685409.17 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3299, pruned_loss=0.08701, over 5729555.27 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3317, pruned_loss=0.08232, over 5670879.43 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:52:20,482 INFO [optim.py:369] (1/2) 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 21:52:44,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8366, 2.3426, 2.1442, 1.6837], device='cuda:1'), covar=tensor([0.3151, 0.1991, 0.2132, 0.2497], device='cuda:1'), in_proj_covar=tensor([0.1918, 0.1830, 0.1755, 0.1902], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 21:52:47,853 INFO [train.py:968] (1/2) Epoch 21, batch 16650, giga_loss[loss=0.2459, simple_loss=0.3255, pruned_loss=0.08314, over 27797.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3306, pruned_loss=0.08247, over 5680544.00 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3299, pruned_loss=0.08701, over 5735840.03 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3316, pruned_loss=0.08245, over 5661655.82 frames. ], batch size: 474, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:53:09,419 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:53:52,187 INFO [train.py:968] (1/2) Epoch 21, batch 16700, giga_loss[loss=0.2828, simple_loss=0.366, pruned_loss=0.09985, over 28631.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3309, pruned_loss=0.08286, over 5672878.10 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3299, pruned_loss=0.08713, over 5736471.99 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3317, pruned_loss=0.08264, over 5655407.01 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:54:31,347 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 16750, giga_loss[loss=0.2862, simple_loss=0.3572, pruned_loss=0.1076, over 28445.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3306, pruned_loss=0.08253, over 5666128.09 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3296, pruned_loss=0.08697, over 5735007.67 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3315, pruned_loss=0.08237, over 5652348.10 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:55:43,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=929130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:55:57,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3005, 1.5480, 1.2521, 0.9480], device='cuda:1'), covar=tensor([0.3081, 0.2687, 0.3270, 0.2414], device='cuda:1'), in_proj_covar=tensor([0.1489, 0.1074, 0.1317, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 21:56:10,865 INFO [train.py:968] (1/2) Epoch 21, batch 16800, giga_loss[loss=0.2701, simple_loss=0.3476, pruned_loss=0.09629, over 27701.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3312, pruned_loss=0.08245, over 5657468.26 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3291, pruned_loss=0.08662, over 5727185.86 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3324, pruned_loss=0.08252, over 5651028.81 frames. ], batch size: 474, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 21:56:12,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6301, 1.8968, 1.5975, 1.5429], device='cuda:1'), covar=tensor([0.0769, 0.0291, 0.0324, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 21:56:55,101 INFO [optim.py:369] (1/2) 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 21:57:16,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3297, 4.0382, 1.5778, 1.4938], device='cuda:1'), covar=tensor([0.1029, 0.0294, 0.0971, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0547, 0.0383, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 21:57:23,864 INFO [train.py:968] (1/2) Epoch 21, batch 16850, libri_loss[loss=0.2417, simple_loss=0.327, pruned_loss=0.07825, over 29547.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3356, pruned_loss=0.08489, over 5662315.32 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3292, pruned_loss=0.0866, over 5729188.00 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3367, pruned_loss=0.08492, over 5653868.83 frames. ], batch size: 89, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:57:54,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:58:23,535 INFO [train.py:968] (1/2) Epoch 21, batch 16900, giga_loss[loss=0.2707, simple_loss=0.3512, pruned_loss=0.09512, over 29086.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3367, pruned_loss=0.0854, over 5660813.79 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3291, pruned_loss=0.0866, over 5721980.47 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3378, pruned_loss=0.0854, over 5658548.63 frames. ], batch size: 285, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 21:58:46,393 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 21:58:56,627 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=929273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:59:01,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=929276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 21:59:05,058 INFO [optim.py:369] (1/2) 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 21:59:36,920 INFO [train.py:968] (1/2) Epoch 21, batch 16950, giga_loss[loss=0.2282, simple_loss=0.3105, pruned_loss=0.07292, over 28953.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3358, pruned_loss=0.08567, over 5666738.66 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3288, pruned_loss=0.08644, over 5723452.05 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3371, pruned_loss=0.0858, over 5663194.63 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 21:59:42,369 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=929305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:00:25,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2068, 1.6917, 1.1964, 0.6088], device='cuda:1'), covar=tensor([0.4141, 0.2128, 0.3330, 0.5572], device='cuda:1'), in_proj_covar=tensor([0.1726, 0.1619, 0.1582, 0.1411], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 22:00:47,388 INFO [train.py:968] (1/2) Epoch 21, batch 17000, libri_loss[loss=0.2421, simple_loss=0.3244, pruned_loss=0.07986, over 29547.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3349, pruned_loss=0.08557, over 5675610.60 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3289, pruned_loss=0.08648, over 5726421.38 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3359, pruned_loss=0.08563, over 5669223.71 frames. ], batch size: 80, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:01:29,518 INFO [optim.py:369] (1/2) 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 22:01:47,667 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 17050, giga_loss[loss=0.2625, simple_loss=0.3384, pruned_loss=0.0933, over 29245.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3321, pruned_loss=0.0831, over 5668884.92 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3287, pruned_loss=0.08641, over 5728309.15 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.333, pruned_loss=0.08319, over 5661883.57 frames. ], batch size: 113, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:02:08,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4314, 3.2151, 1.4972, 1.5340], device='cuda:1'), covar=tensor([0.0967, 0.0317, 0.0963, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0548, 0.0384, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 22:02:59,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 22:03:00,173 INFO [train.py:968] (1/2) Epoch 21, batch 17100, giga_loss[loss=0.2245, simple_loss=0.3154, pruned_loss=0.06679, over 28965.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3311, pruned_loss=0.08239, over 5671684.56 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3285, pruned_loss=0.08626, over 5728746.78 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3321, pruned_loss=0.08254, over 5665023.72 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:03:08,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2816, 4.0964, 3.9011, 1.8600], device='cuda:1'), covar=tensor([0.0639, 0.0751, 0.0875, 0.2206], device='cuda:1'), in_proj_covar=tensor([0.1181, 0.1085, 0.0925, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 22:03:30,824 INFO [optim.py:369] (1/2) 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 22:03:54,148 INFO [train.py:968] (1/2) Epoch 21, batch 17150, giga_loss[loss=0.2948, simple_loss=0.3659, pruned_loss=0.1118, over 28915.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3338, pruned_loss=0.0841, over 5676147.35 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3282, pruned_loss=0.08607, over 5734856.33 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3349, pruned_loss=0.08429, over 5663637.71 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:04:33,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=929534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:04:36,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=929537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:04:49,173 INFO [train.py:968] (1/2) Epoch 21, batch 17200, giga_loss[loss=0.2397, simple_loss=0.3269, pruned_loss=0.07625, over 29037.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3347, pruned_loss=0.08502, over 5680780.49 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.328, pruned_loss=0.08594, over 5740024.45 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.336, pruned_loss=0.08524, over 5664037.55 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:05:05,837 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=929566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:05:21,252 INFO [optim.py:369] (1/2) 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 22:05:41,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=929597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:05:44,579 INFO [train.py:968] (1/2) Epoch 21, batch 17250, giga_loss[loss=0.2385, simple_loss=0.317, pruned_loss=0.07999, over 28547.00 frames. ], tot_loss[loss=0.251, simple_loss=0.332, pruned_loss=0.08502, over 5666540.35 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3281, pruned_loss=0.08596, over 5733742.44 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.333, pruned_loss=0.08515, over 5658115.83 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:06:15,136 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:06:40,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4311, 1.6196, 1.5449, 1.3382], device='cuda:1'), covar=tensor([0.2375, 0.2206, 0.1572, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.1912, 0.1821, 0.1745, 0.1895], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 22:06:40,558 INFO [train.py:968] (1/2) Epoch 21, batch 17300, giga_loss[loss=0.2558, simple_loss=0.3374, pruned_loss=0.08716, over 29031.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3327, pruned_loss=0.08663, over 5645303.96 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3282, pruned_loss=0.08607, over 5717317.59 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3335, pruned_loss=0.08664, over 5652934.00 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:07:15,386 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 17350, giga_loss[loss=0.2862, simple_loss=0.3666, pruned_loss=0.1029, over 28141.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3354, pruned_loss=0.08854, over 5647391.17 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.328, pruned_loss=0.08592, over 5719778.80 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3362, pruned_loss=0.08869, over 5650058.78 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:07:54,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5496, 2.2172, 1.6335, 0.8412], device='cuda:1'), covar=tensor([0.6792, 0.3495, 0.4223, 0.6604], device='cuda:1'), in_proj_covar=tensor([0.1736, 0.1630, 0.1588, 0.1415], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 22:08:18,513 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=929740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:08:20,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=929743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:08:27,393 INFO [train.py:968] (1/2) Epoch 21, batch 17400, giga_loss[loss=0.3032, simple_loss=0.379, pruned_loss=0.1137, over 28900.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3431, pruned_loss=0.09257, over 5658251.61 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3273, pruned_loss=0.08559, over 5719881.81 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3448, pruned_loss=0.09311, over 5658498.15 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:08:30,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-10 22:08:45,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=929772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:08:51,704 INFO [optim.py:369] (1/2) 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 22:09:11,626 INFO [train.py:968] (1/2) Epoch 21, batch 17450, giga_loss[loss=0.2665, simple_loss=0.3417, pruned_loss=0.09563, over 28834.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3492, pruned_loss=0.09623, over 5667379.99 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3271, pruned_loss=0.0855, over 5725978.31 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3513, pruned_loss=0.09702, over 5660401.46 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:09:51,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:09:54,894 INFO [train.py:968] (1/2) Epoch 21, batch 17500, giga_loss[loss=0.2669, simple_loss=0.3457, pruned_loss=0.09404, over 28861.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3483, pruned_loss=0.09654, over 5669383.87 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3274, pruned_loss=0.08562, over 5728104.48 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3501, pruned_loss=0.09723, over 5661296.08 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:10:22,988 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:10:33,008 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=929892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:10:39,886 INFO [train.py:968] (1/2) Epoch 21, batch 17550, giga_loss[loss=0.2451, simple_loss=0.3173, pruned_loss=0.08648, over 28745.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3416, pruned_loss=0.09365, over 5678185.45 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3273, pruned_loss=0.08558, over 5727414.62 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3435, pruned_loss=0.09445, over 5670961.52 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:11:22,407 INFO [train.py:968] (1/2) Epoch 21, batch 17600, giga_loss[loss=0.264, simple_loss=0.3275, pruned_loss=0.1003, over 27687.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3344, pruned_loss=0.09014, over 5690108.36 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3273, pruned_loss=0.0854, over 5732635.10 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3362, pruned_loss=0.09116, over 5678145.30 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:11:34,995 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5956, 1.9148, 1.8019, 1.4636], device='cuda:1'), covar=tensor([0.3390, 0.2694, 0.2501, 0.2973], device='cuda:1'), in_proj_covar=tensor([0.1931, 0.1838, 0.1764, 0.1919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 22:11:45,882 INFO [optim.py:369] (1/2) 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 22:12:02,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3907, 1.4661, 1.4986, 1.1288], device='cuda:1'), covar=tensor([0.1850, 0.3182, 0.1478, 0.1697], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0690, 0.0943, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 22:12:03,474 INFO [train.py:968] (1/2) Epoch 21, batch 17650, libri_loss[loss=0.2305, simple_loss=0.3167, pruned_loss=0.07221, over 29521.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3274, pruned_loss=0.08717, over 5688741.77 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3273, pruned_loss=0.08528, over 5724096.78 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3289, pruned_loss=0.08817, over 5685322.24 frames. ], batch size: 80, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:12:03,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:12:43,794 INFO [train.py:968] (1/2) Epoch 21, batch 17700, giga_loss[loss=0.2194, simple_loss=0.2916, pruned_loss=0.07354, over 28796.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3207, pruned_loss=0.08417, over 5686364.84 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3272, pruned_loss=0.08505, over 5721078.02 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3219, pruned_loss=0.08523, over 5684456.05 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:13:08,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3584, 3.3324, 1.4975, 1.6133], device='cuda:1'), covar=tensor([0.1041, 0.0314, 0.0938, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0547, 0.0381, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 22:13:09,924 INFO [optim.py:369] (1/2) 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 22:13:25,615 INFO [train.py:968] (1/2) Epoch 21, batch 17750, libri_loss[loss=0.2225, simple_loss=0.3089, pruned_loss=0.06805, over 29564.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.316, pruned_loss=0.08209, over 5687840.56 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3272, pruned_loss=0.08496, over 5724667.21 frames. ], giga_tot_loss[loss=0.2413, simple_loss=0.3167, pruned_loss=0.08296, over 5681865.22 frames. ], batch size: 76, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:13:27,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6487, 2.0124, 1.6030, 1.6971], device='cuda:1'), covar=tensor([0.2668, 0.2700, 0.3157, 0.2579], device='cuda:1'), in_proj_covar=tensor([0.1492, 0.1078, 0.1319, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 22:14:01,313 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:14:01,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2332, 1.5319, 1.5647, 1.1559], device='cuda:1'), covar=tensor([0.1603, 0.2528, 0.1327, 0.1637], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0694, 0.0948, 0.0845], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 22:14:03,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:14:07,036 INFO [train.py:968] (1/2) Epoch 21, batch 17800, giga_loss[loss=0.2779, simple_loss=0.3407, pruned_loss=0.1076, over 27930.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.3133, pruned_loss=0.08087, over 5696404.18 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3276, pruned_loss=0.08493, over 5728508.25 frames. ], giga_tot_loss[loss=0.2381, simple_loss=0.3132, pruned_loss=0.08151, over 5687520.48 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:14:28,975 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:14:33,326 INFO [optim.py:369] (1/2) 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 22:14:46,657 INFO [train.py:968] (1/2) Epoch 21, batch 17850, giga_loss[loss=0.277, simple_loss=0.3327, pruned_loss=0.1107, over 26631.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3118, pruned_loss=0.08036, over 5698212.85 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.328, pruned_loss=0.08499, over 5734671.22 frames. ], giga_tot_loss[loss=0.2361, simple_loss=0.3108, pruned_loss=0.08066, over 5684159.96 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:15:03,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:15:25,316 INFO [train.py:968] (1/2) Epoch 21, batch 17900, giga_loss[loss=0.2142, simple_loss=0.2927, pruned_loss=0.06786, over 29103.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3093, pruned_loss=0.07924, over 5702266.06 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3278, pruned_loss=0.08465, over 5741139.84 frames. ], giga_tot_loss[loss=0.2337, simple_loss=0.3081, pruned_loss=0.07962, over 5683904.68 frames. ], batch size: 113, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:15:29,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5456, 1.5993, 1.2127, 1.1367], device='cuda:1'), covar=tensor([0.0833, 0.0501, 0.0965, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0442, 0.0513, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 22:15:34,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:15:38,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=930267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:15:50,538 INFO [optim.py:369] (1/2) 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 22:16:04,997 INFO [train.py:968] (1/2) Epoch 21, batch 17950, giga_loss[loss=0.2174, simple_loss=0.3014, pruned_loss=0.06674, over 28960.00 frames. ], tot_loss[loss=0.231, simple_loss=0.3064, pruned_loss=0.07778, over 5710672.81 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3283, pruned_loss=0.08469, over 5745654.45 frames. ], giga_tot_loss[loss=0.23, simple_loss=0.3043, pruned_loss=0.07787, over 5690754.30 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:16:37,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4030, 3.3258, 1.6305, 1.5924], device='cuda:1'), covar=tensor([0.0995, 0.0307, 0.0902, 0.1314], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0546, 0.0381, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 22:16:49,180 INFO [train.py:968] (1/2) Epoch 21, batch 18000, giga_loss[loss=0.2258, simple_loss=0.2993, pruned_loss=0.07608, over 28635.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3033, pruned_loss=0.07669, over 5704694.53 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3286, pruned_loss=0.08491, over 5747165.38 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.3012, pruned_loss=0.07648, over 5687422.49 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:16:49,181 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 22:16:57,370 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 22:17:11,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:13,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:15,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1453, 2.2883, 1.9276, 2.4357], device='cuda:1'), covar=tensor([0.2404, 0.2628, 0.2970, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.1492, 0.1078, 0.1318, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 22:17:22,559 INFO [optim.py:369] (1/2) 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 22:17:36,762 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:38,363 INFO [train.py:968] (1/2) Epoch 21, batch 18050, giga_loss[loss=0.2283, simple_loss=0.3051, pruned_loss=0.07577, over 28972.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3012, pruned_loss=0.07604, over 5692415.95 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3283, pruned_loss=0.08477, over 5742415.70 frames. ], giga_tot_loss[loss=0.2253, simple_loss=0.299, pruned_loss=0.07577, over 5681905.94 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:17:42,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:44,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:46,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=930410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:46,445 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:48,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:17:56,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5928, 1.7017, 1.8062, 1.4014], device='cuda:1'), covar=tensor([0.1838, 0.2415, 0.1499, 0.1702], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0697, 0.0952, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 22:18:08,633 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:18:15,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:18:21,904 INFO [train.py:968] (1/2) Epoch 21, batch 18100, giga_loss[loss=0.1992, simple_loss=0.2788, pruned_loss=0.05978, over 28960.00 frames. ], tot_loss[loss=0.2227, simple_loss=0.2974, pruned_loss=0.07399, over 5701600.91 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3283, pruned_loss=0.08469, over 5744848.53 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.2953, pruned_loss=0.0737, over 5690272.59 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:18:29,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1193, 1.1133, 3.6443, 3.2171], device='cuda:1'), covar=tensor([0.1779, 0.2892, 0.0501, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0646, 0.0949, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:18:53,044 INFO [optim.py:369] (1/2) 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 22:19:07,563 INFO [train.py:968] (1/2) Epoch 21, batch 18150, giga_loss[loss=0.2177, simple_loss=0.2948, pruned_loss=0.07032, over 28008.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.294, pruned_loss=0.07254, over 5696690.91 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3286, pruned_loss=0.08473, over 5741280.83 frames. ], giga_tot_loss[loss=0.2177, simple_loss=0.2914, pruned_loss=0.07204, over 5690226.35 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:19:52,566 INFO [train.py:968] (1/2) Epoch 21, batch 18200, giga_loss[loss=0.2423, simple_loss=0.3233, pruned_loss=0.08063, over 28596.00 frames. ], tot_loss[loss=0.225, simple_loss=0.2991, pruned_loss=0.07546, over 5700729.96 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3294, pruned_loss=0.08515, over 5745469.80 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.2956, pruned_loss=0.0744, over 5690863.75 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:20:22,423 INFO [optim.py:369] (1/2) 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 22:20:40,533 INFO [train.py:968] (1/2) Epoch 21, batch 18250, giga_loss[loss=0.3313, simple_loss=0.3968, pruned_loss=0.1329, over 28759.00 frames. ], tot_loss[loss=0.2371, simple_loss=0.3114, pruned_loss=0.08142, over 5701512.35 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3294, pruned_loss=0.08513, over 5747086.64 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3083, pruned_loss=0.08056, over 5691450.97 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:20:49,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0621, 3.3861, 2.0418, 1.2337], device='cuda:1'), covar=tensor([0.8199, 0.2668, 0.4310, 0.6743], device='cuda:1'), in_proj_covar=tensor([0.1729, 0.1628, 0.1586, 0.1405], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 22:21:20,093 INFO [train.py:968] (1/2) Epoch 21, batch 18300, giga_loss[loss=0.2646, simple_loss=0.3422, pruned_loss=0.09354, over 28491.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3243, pruned_loss=0.08836, over 5696766.88 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3295, pruned_loss=0.08507, over 5742516.79 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3216, pruned_loss=0.08775, over 5691768.44 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:21:22,705 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=930654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:21:24,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2265, 1.2707, 3.8303, 3.2302], device='cuda:1'), covar=tensor([0.1779, 0.2893, 0.0488, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0753, 0.0647, 0.0950, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:21:43,508 INFO [optim.py:369] (1/2) 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 22:21:58,026 INFO [train.py:968] (1/2) Epoch 21, batch 18350, giga_loss[loss=0.2731, simple_loss=0.3484, pruned_loss=0.09893, over 28621.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3332, pruned_loss=0.09222, over 5702720.03 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3299, pruned_loss=0.08531, over 5746858.11 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3307, pruned_loss=0.09168, over 5693646.57 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:22:01,011 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 22:22:38,738 INFO [train.py:968] (1/2) Epoch 21, batch 18400, giga_loss[loss=0.2992, simple_loss=0.3775, pruned_loss=0.1104, over 27561.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3385, pruned_loss=0.09417, over 5701960.90 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3298, pruned_loss=0.08538, over 5750697.75 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3366, pruned_loss=0.09386, over 5690397.14 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:23:02,565 INFO [optim.py:369] (1/2) 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 22:23:05,170 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 18450, giga_loss[loss=0.267, simple_loss=0.3468, pruned_loss=0.09364, over 27624.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3415, pruned_loss=0.09473, over 5688995.84 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3306, pruned_loss=0.0858, over 5744803.80 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3395, pruned_loss=0.0943, over 5683359.08 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:23:46,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2458, 2.2739, 2.1719, 2.0793], device='cuda:1'), covar=tensor([0.1913, 0.2452, 0.2225, 0.2151], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0742, 0.0706, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 22:24:03,002 INFO [train.py:968] (1/2) Epoch 21, batch 18500, giga_loss[loss=0.3318, simple_loss=0.3816, pruned_loss=0.141, over 26440.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3425, pruned_loss=0.09524, over 5687800.09 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3307, pruned_loss=0.08586, over 5746215.46 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.341, pruned_loss=0.09496, over 5681388.63 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:24:29,204 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 18550, giga_loss[loss=0.2734, simple_loss=0.3535, pruned_loss=0.09662, over 28711.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.345, pruned_loss=0.09715, over 5691108.96 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3308, pruned_loss=0.08584, over 5749445.07 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.344, pruned_loss=0.09714, over 5682253.92 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:25:07,414 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=930928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:25:09,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=930931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:25:24,914 INFO [train.py:968] (1/2) Epoch 21, batch 18600, libri_loss[loss=0.2333, simple_loss=0.3134, pruned_loss=0.07658, over 29531.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3482, pruned_loss=0.09907, over 5701628.48 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3315, pruned_loss=0.08611, over 5753477.40 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3474, pruned_loss=0.09934, over 5688781.40 frames. ], batch size: 77, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:25:32,161 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=930960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:25:51,591 INFO [optim.py:369] (1/2) 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 22:26:06,520 INFO [train.py:968] (1/2) Epoch 21, batch 18650, giga_loss[loss=0.2685, simple_loss=0.3522, pruned_loss=0.09238, over 28283.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3517, pruned_loss=0.1008, over 5701932.97 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3317, pruned_loss=0.08625, over 5754464.01 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.351, pruned_loss=0.101, over 5690691.20 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:26:14,455 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-10 22:26:27,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:26:29,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:26:33,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:26:47,467 INFO [train.py:968] (1/2) Epoch 21, batch 18700, giga_loss[loss=0.2744, simple_loss=0.3583, pruned_loss=0.09526, over 28268.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3532, pruned_loss=0.1003, over 5713384.99 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.332, pruned_loss=0.08628, over 5757550.08 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3527, pruned_loss=0.1006, over 5700879.92 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:27:13,278 INFO [optim.py:369] (1/2) 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 22:27:18,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3216, 2.0805, 1.5616, 0.5442], device='cuda:1'), covar=tensor([0.5382, 0.3161, 0.4525, 0.6231], device='cuda:1'), in_proj_covar=tensor([0.1734, 0.1635, 0.1592, 0.1410], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 22:27:27,411 INFO [train.py:968] (1/2) Epoch 21, batch 18750, giga_loss[loss=0.2891, simple_loss=0.3667, pruned_loss=0.1057, over 28819.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.355, pruned_loss=0.1006, over 5713232.42 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3321, pruned_loss=0.08629, over 5760726.84 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3549, pruned_loss=0.1011, over 5699618.44 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:27:32,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3331, 1.6806, 1.3845, 1.6332], device='cuda:1'), covar=tensor([0.0738, 0.0352, 0.0331, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 22:28:09,232 INFO [train.py:968] (1/2) Epoch 21, batch 18800, giga_loss[loss=0.294, simple_loss=0.3685, pruned_loss=0.1097, over 28960.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3561, pruned_loss=0.1008, over 5710472.22 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3322, pruned_loss=0.08627, over 5762980.78 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3564, pruned_loss=0.1014, over 5696838.64 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:28:25,989 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:28:26,883 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-10 22:28:29,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:28:32,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4925, 2.0620, 1.5692, 0.8021], device='cuda:1'), covar=tensor([0.5669, 0.2893, 0.4315, 0.6148], device='cuda:1'), in_proj_covar=tensor([0.1735, 0.1635, 0.1595, 0.1411], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 22:28:35,060 INFO [optim.py:369] (1/2) 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 22:28:38,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2266, 2.5391, 1.3085, 1.4677], device='cuda:1'), covar=tensor([0.1063, 0.0321, 0.0936, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0543, 0.0380, 0.0421], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 22:28:49,033 INFO [train.py:968] (1/2) Epoch 21, batch 18850, giga_loss[loss=0.2603, simple_loss=0.3427, pruned_loss=0.08898, over 28890.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.0989, over 5708115.72 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3322, pruned_loss=0.08618, over 5765061.19 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.355, pruned_loss=0.0997, over 5694610.16 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:28:52,450 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 18900, giga_loss[loss=0.2591, simple_loss=0.3453, pruned_loss=0.08643, over 28570.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3532, pruned_loss=0.09734, over 5711704.27 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3327, pruned_loss=0.08623, over 5765193.82 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3537, pruned_loss=0.09825, over 5699340.62 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:29:33,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:29:42,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2845, 1.7962, 1.4854, 1.3889], device='cuda:1'), covar=tensor([0.0831, 0.0328, 0.0305, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0116, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 22:29:44,861 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931275.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:29:51,421 INFO [optim.py:369] (1/2) 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 22:30:05,559 INFO [train.py:968] (1/2) Epoch 21, batch 18950, giga_loss[loss=0.2522, simple_loss=0.3383, pruned_loss=0.08298, over 28774.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3553, pruned_loss=0.09951, over 5703789.16 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3338, pruned_loss=0.08701, over 5760505.93 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3554, pruned_loss=0.09996, over 5695761.09 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:30:47,557 INFO [train.py:968] (1/2) Epoch 21, batch 19000, giga_loss[loss=0.3285, simple_loss=0.388, pruned_loss=0.1345, over 28227.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3588, pruned_loss=0.1049, over 5696966.54 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3345, pruned_loss=0.08738, over 5763326.16 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.359, pruned_loss=0.1054, over 5686025.91 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:31:15,036 INFO [optim.py:369] (1/2) 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 22:31:25,455 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-10 22:31:29,550 INFO [train.py:968] (1/2) Epoch 21, batch 19050, libri_loss[loss=0.2927, simple_loss=0.3737, pruned_loss=0.1058, over 28891.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3586, pruned_loss=0.1063, over 5693102.04 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3342, pruned_loss=0.08716, over 5764477.36 frames. ], giga_tot_loss[loss=0.287, simple_loss=0.3595, pruned_loss=0.1072, over 5681855.97 frames. ], batch size: 107, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:31:29,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:31:36,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:32:07,421 INFO [train.py:968] (1/2) Epoch 21, batch 19100, giga_loss[loss=0.2842, simple_loss=0.3551, pruned_loss=0.1067, over 28757.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3577, pruned_loss=0.1064, over 5698480.29 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3342, pruned_loss=0.08709, over 5764492.20 frames. ], giga_tot_loss[loss=0.2874, simple_loss=0.3592, pruned_loss=0.1078, over 5687557.35 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:32:33,463 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 19150, giga_loss[loss=0.3195, simple_loss=0.3888, pruned_loss=0.1251, over 28203.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3552, pruned_loss=0.1049, over 5709616.48 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.335, pruned_loss=0.08741, over 5770107.53 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3565, pruned_loss=0.1064, over 5693498.08 frames. ], batch size: 77, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:32:46,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:32:57,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2842, 1.3801, 1.4686, 1.1711], device='cuda:1'), covar=tensor([0.3281, 0.2874, 0.2002, 0.2714], device='cuda:1'), in_proj_covar=tensor([0.1960, 0.1869, 0.1802, 0.1954], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 22:33:13,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2525, 1.1510, 4.0449, 3.3548], device='cuda:1'), covar=tensor([0.1702, 0.2941, 0.0401, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0644, 0.0947, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:33:20,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:23,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:27,212 INFO [train.py:968] (1/2) Epoch 21, batch 19200, libri_loss[loss=0.2937, simple_loss=0.3754, pruned_loss=0.106, over 27525.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3557, pruned_loss=0.1052, over 5697929.88 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3355, pruned_loss=0.08747, over 5771684.90 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3573, pruned_loss=0.1073, over 5680313.49 frames. ], batch size: 115, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:33:28,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:30,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:33:47,750 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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 22:34:03,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0073, 1.3598, 1.3332, 1.1957], device='cuda:1'), covar=tensor([0.2144, 0.1661, 0.2524, 0.1853], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0741, 0.0708, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 22:34:04,994 INFO [train.py:968] (1/2) Epoch 21, batch 19250, giga_loss[loss=0.2573, simple_loss=0.339, pruned_loss=0.08775, over 28716.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3539, pruned_loss=0.1031, over 5704145.83 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3354, pruned_loss=0.08731, over 5773842.48 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3558, pruned_loss=0.1054, over 5686110.11 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:34:34,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:34:49,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:34:49,938 INFO [train.py:968] (1/2) Epoch 21, batch 19300, giga_loss[loss=0.2702, simple_loss=0.3503, pruned_loss=0.09505, over 28566.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3504, pruned_loss=0.1005, over 5691382.78 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3356, pruned_loss=0.08735, over 5766119.52 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3519, pruned_loss=0.1024, over 5683255.60 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:35:19,543 INFO [optim.py:369] (1/2) 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 22:35:33,209 INFO [train.py:968] (1/2) Epoch 21, batch 19350, giga_loss[loss=0.2389, simple_loss=0.3164, pruned_loss=0.08074, over 28966.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3452, pruned_loss=0.09775, over 5690334.36 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3356, pruned_loss=0.08724, over 5769513.07 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3467, pruned_loss=0.09967, over 5679076.45 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:35:56,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1687, 1.2990, 3.3341, 3.0060], device='cuda:1'), covar=tensor([0.1598, 0.2698, 0.0499, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0643, 0.0948, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:36:14,445 INFO [train.py:968] (1/2) Epoch 21, batch 19400, giga_loss[loss=0.2227, simple_loss=0.3015, pruned_loss=0.07198, over 28787.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3403, pruned_loss=0.09487, over 5692818.68 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3358, pruned_loss=0.08709, over 5770216.84 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3417, pruned_loss=0.09694, over 5679719.08 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:36:40,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:36:43,809 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:36:54,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=931791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:36:56,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=931793.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:36:59,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=931796.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:37:02,779 INFO [train.py:968] (1/2) Epoch 21, batch 19450, giga_loss[loss=0.23, simple_loss=0.3128, pruned_loss=0.07359, over 28589.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3349, pruned_loss=0.09221, over 5693524.69 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3358, pruned_loss=0.08705, over 5770445.42 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.336, pruned_loss=0.09397, over 5681762.17 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:37:10,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=931809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:37:13,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-10 22:37:25,152 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=931825.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:37:33,200 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 19500, giga_loss[loss=0.2448, simple_loss=0.3259, pruned_loss=0.08189, over 28857.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3355, pruned_loss=0.09194, over 5693618.33 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3362, pruned_loss=0.08723, over 5770717.33 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.336, pruned_loss=0.09326, over 5682631.66 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:38:09,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=931876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:38:11,874 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 19550, giga_loss[loss=0.245, simple_loss=0.3274, pruned_loss=0.08129, over 28637.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.335, pruned_loss=0.09083, over 5702258.34 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3364, pruned_loss=0.08734, over 5771236.25 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3352, pruned_loss=0.09182, over 5692070.13 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:38:36,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 22:38:59,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7236, 4.8867, 2.1110, 2.0645], device='cuda:1'), covar=tensor([0.1001, 0.0263, 0.0829, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0542, 0.0381, 0.0420], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 22:39:11,010 INFO [train.py:968] (1/2) Epoch 21, batch 19600, giga_loss[loss=0.2681, simple_loss=0.3383, pruned_loss=0.09894, over 28916.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3335, pruned_loss=0.08988, over 5701785.04 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3366, pruned_loss=0.08727, over 5764868.63 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3334, pruned_loss=0.09084, over 5697295.69 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:39:37,618 INFO [optim.py:369] (1/2) 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 22:39:50,228 INFO [train.py:968] (1/2) Epoch 21, batch 19650, giga_loss[loss=0.2171, simple_loss=0.2916, pruned_loss=0.07132, over 28656.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3309, pruned_loss=0.08891, over 5712748.61 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08728, over 5766668.95 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3306, pruned_loss=0.08972, over 5706244.19 frames. ], batch size: 60, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:40:00,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3803, 1.5962, 1.3190, 0.9987], device='cuda:1'), covar=tensor([0.2599, 0.2780, 0.3059, 0.2480], device='cuda:1'), in_proj_covar=tensor([0.1492, 0.1082, 0.1314, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 22:40:02,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3782, 1.6350, 1.6594, 1.4870], device='cuda:1'), covar=tensor([0.2163, 0.2137, 0.2550, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0744, 0.0712, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 22:40:05,874 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:07,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:28,403 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:30,400 INFO [train.py:968] (1/2) Epoch 21, batch 19700, giga_loss[loss=0.2837, simple_loss=0.3429, pruned_loss=0.1122, over 24137.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3293, pruned_loss=0.08842, over 5716073.32 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3375, pruned_loss=0.08758, over 5768722.34 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3283, pruned_loss=0.0888, over 5708478.01 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:40:30,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:40:40,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8207, 1.0329, 2.9061, 2.8701], device='cuda:1'), covar=tensor([0.1591, 0.2479, 0.0595, 0.0954], device='cuda:1'), in_proj_covar=tensor([0.0747, 0.0640, 0.0943, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:40:54,865 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 19750, giga_loss[loss=0.2224, simple_loss=0.3022, pruned_loss=0.07129, over 28821.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3291, pruned_loss=0.08886, over 5722024.73 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3383, pruned_loss=0.08786, over 5771854.67 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3272, pruned_loss=0.08893, over 5711214.62 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:41:44,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-10 22:41:47,081 INFO [train.py:968] (1/2) Epoch 21, batch 19800, giga_loss[loss=0.2156, simple_loss=0.2982, pruned_loss=0.06651, over 28900.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3257, pruned_loss=0.08723, over 5730476.08 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3386, pruned_loss=0.08797, over 5775231.44 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3237, pruned_loss=0.08721, over 5717882.98 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:42:00,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:42:15,873 INFO [optim.py:369] (1/2) 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 22:42:25,951 INFO [train.py:968] (1/2) Epoch 21, batch 19850, giga_loss[loss=0.2524, simple_loss=0.3211, pruned_loss=0.09184, over 29097.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.324, pruned_loss=0.08645, over 5726320.17 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3388, pruned_loss=0.0879, over 5774786.60 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3218, pruned_loss=0.08647, over 5715586.92 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:42:33,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:42:43,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-10 22:42:59,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3084, 1.2858, 3.7356, 3.1082], device='cuda:1'), covar=tensor([0.1596, 0.2693, 0.0439, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0748, 0.0639, 0.0944, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:43:06,159 INFO [train.py:968] (1/2) Epoch 21, batch 19900, giga_loss[loss=0.2203, simple_loss=0.3007, pruned_loss=0.06997, over 28816.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3219, pruned_loss=0.08583, over 5723038.02 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3391, pruned_loss=0.08791, over 5776087.48 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3199, pruned_loss=0.08583, over 5713062.13 frames. ], batch size: 66, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:43:08,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:43:34,393 INFO [optim.py:369] (1/2) 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 22:43:38,822 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:43:44,647 INFO [train.py:968] (1/2) Epoch 21, batch 19950, giga_loss[loss=0.2111, simple_loss=0.2923, pruned_loss=0.06494, over 28963.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3188, pruned_loss=0.08359, over 5731360.13 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3395, pruned_loss=0.08796, over 5777738.88 frames. ], giga_tot_loss[loss=0.2418, simple_loss=0.3166, pruned_loss=0.08351, over 5721387.22 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:43:50,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:43:53,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:16,422 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 20000, giga_loss[loss=0.2065, simple_loss=0.2861, pruned_loss=0.06343, over 28947.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3168, pruned_loss=0.08257, over 5731152.30 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3397, pruned_loss=0.08798, over 5778134.47 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3147, pruned_loss=0.08244, over 5722462.81 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:44:24,613 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:26,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:47,055 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:44:49,870 INFO [optim.py:369] (1/2) 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 22:44:58,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:01,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:02,057 INFO [train.py:968] (1/2) Epoch 21, batch 20050, giga_loss[loss=0.2529, simple_loss=0.3334, pruned_loss=0.08625, over 28770.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3159, pruned_loss=0.08186, over 5739603.42 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3399, pruned_loss=0.08795, over 5779659.13 frames. ], giga_tot_loss[loss=0.2386, simple_loss=0.3137, pruned_loss=0.08171, over 5730938.97 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:45:15,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:19,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:26,527 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:45:28,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0860, 2.1844, 1.6146, 1.6551], device='cuda:1'), covar=tensor([0.0949, 0.0734, 0.1078, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0445, 0.0516, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 22:45:44,916 INFO [train.py:968] (1/2) Epoch 21, batch 20100, giga_loss[loss=0.3032, simple_loss=0.3663, pruned_loss=0.12, over 28596.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3211, pruned_loss=0.08558, over 5720564.31 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08799, over 5780268.84 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3192, pruned_loss=0.0854, over 5713099.34 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:46:06,217 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:46:30,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2094, 1.3047, 3.8443, 3.2349], device='cuda:1'), covar=tensor([0.1681, 0.2731, 0.0447, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0640, 0.0946, 0.0891], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:46:34,678 INFO [train.py:968] (1/2) Epoch 21, batch 20150, giga_loss[loss=0.2998, simple_loss=0.3599, pruned_loss=0.1199, over 28627.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3301, pruned_loss=0.09116, over 5716716.58 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3402, pruned_loss=0.08796, over 5781321.25 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3283, pruned_loss=0.09106, over 5709418.17 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:46:56,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-10 22:47:18,365 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:47:18,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6233, 1.8226, 1.5633, 1.4275], device='cuda:1'), covar=tensor([0.2209, 0.2168, 0.2254, 0.2095], device='cuda:1'), in_proj_covar=tensor([0.1493, 0.1081, 0.1317, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 22:47:23,993 INFO [train.py:968] (1/2) Epoch 21, batch 20200, giga_loss[loss=0.2364, simple_loss=0.3163, pruned_loss=0.07828, over 28579.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3389, pruned_loss=0.09717, over 5697017.21 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3406, pruned_loss=0.08826, over 5782577.88 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3372, pruned_loss=0.09689, over 5689690.70 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:47:38,280 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:47:40,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:47:57,320 INFO [optim.py:369] (1/2) 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 22:48:05,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:48:07,687 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 20250, giga_loss[loss=0.3402, simple_loss=0.41, pruned_loss=0.1352, over 28861.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3433, pruned_loss=0.09824, over 5694302.76 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3406, pruned_loss=0.08823, over 5784459.97 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3418, pruned_loss=0.09818, over 5685579.30 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:48:19,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3339, 1.9569, 1.4982, 0.5446], device='cuda:1'), covar=tensor([0.4538, 0.2545, 0.3602, 0.5667], device='cuda:1'), in_proj_covar=tensor([0.1742, 0.1640, 0.1599, 0.1420], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 22:48:30,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.24 vs. limit=5.0
+2023-03-10 22:48:30,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1983, 3.9867, 3.7858, 1.8714], device='cuda:1'), covar=tensor([0.0603, 0.0799, 0.0756, 0.2333], device='cuda:1'), in_proj_covar=tensor([0.1193, 0.1104, 0.0937, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 22:48:51,401 INFO [train.py:968] (1/2) Epoch 21, batch 20300, giga_loss[loss=0.2849, simple_loss=0.3635, pruned_loss=0.1031, over 28951.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1008, over 5684603.82 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.341, pruned_loss=0.0883, over 5780034.87 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3478, pruned_loss=0.1011, over 5678071.23 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:48:57,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2882, 1.5129, 0.9378, 1.0780], device='cuda:1'), covar=tensor([0.1094, 0.0652, 0.1445, 0.1558], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0444, 0.0516, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 22:49:07,586 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:49:15,588 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932677.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:49:22,770 INFO [optim.py:369] (1/2) 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 22:49:33,812 INFO [train.py:968] (1/2) Epoch 21, batch 20350, libri_loss[loss=0.2714, simple_loss=0.3575, pruned_loss=0.09265, over 29688.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3555, pruned_loss=0.1051, over 5673130.12 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3414, pruned_loss=0.08846, over 5771850.31 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3546, pruned_loss=0.1058, over 5671803.76 frames. ], batch size: 88, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:49:42,337 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932713.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:50:11,008 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:50:15,410 INFO [train.py:968] (1/2) Epoch 21, batch 20400, giga_loss[loss=0.254, simple_loss=0.336, pruned_loss=0.08601, over 28665.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3553, pruned_loss=0.1048, over 5676034.67 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3416, pruned_loss=0.08868, over 5773336.69 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3546, pruned_loss=0.1054, over 5672572.30 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 22:50:21,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6424, 1.8912, 1.5269, 1.6120], device='cuda:1'), covar=tensor([0.2738, 0.2768, 0.3077, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.1487, 0.1078, 0.1313, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 22:50:46,930 INFO [optim.py:369] (1/2) 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 22:50:49,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:50:56,703 INFO [train.py:968] (1/2) Epoch 21, batch 20450, giga_loss[loss=0.2716, simple_loss=0.3425, pruned_loss=0.1003, over 28874.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3513, pruned_loss=0.1017, over 5681438.11 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.08906, over 5774868.89 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3508, pruned_loss=0.1022, over 5675212.24 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:51:04,910 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:07,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:15,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5534, 1.8427, 1.5009, 1.4898], device='cuda:1'), covar=tensor([0.2625, 0.2584, 0.2810, 0.2350], device='cuda:1'), in_proj_covar=tensor([0.1487, 0.1078, 0.1312, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 22:51:30,810 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:32,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:51:34,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2087, 1.3533, 3.6751, 3.2972], device='cuda:1'), covar=tensor([0.1694, 0.2696, 0.0472, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0642, 0.0947, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 22:51:37,147 INFO [train.py:968] (1/2) Epoch 21, batch 20500, giga_loss[loss=0.2758, simple_loss=0.3439, pruned_loss=0.1038, over 23447.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.349, pruned_loss=0.09972, over 5692967.72 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3418, pruned_loss=0.08914, over 5776274.21 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3487, pruned_loss=0.1001, over 5686099.01 frames. ], batch size: 710, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:51:46,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3325, 3.8957, 1.5260, 1.5123], device='cuda:1'), covar=tensor([0.1038, 0.0296, 0.0903, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0545, 0.0381, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 22:52:08,237 INFO [optim.py:369] (1/2) 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 22:52:11,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-10 22:52:17,588 INFO [train.py:968] (1/2) Epoch 21, batch 20550, giga_loss[loss=0.2686, simple_loss=0.3527, pruned_loss=0.0923, over 29064.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09969, over 5685545.63 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3422, pruned_loss=0.08941, over 5769752.02 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3492, pruned_loss=0.1001, over 5683871.73 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:52:32,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:44,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=932935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:45,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=932936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:46,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:52:58,614 INFO [train.py:968] (1/2) Epoch 21, batch 20600, giga_loss[loss=0.2742, simple_loss=0.3503, pruned_loss=0.09901, over 28599.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3505, pruned_loss=0.09996, over 5686649.76 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3422, pruned_loss=0.08942, over 5770331.89 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3503, pruned_loss=0.1005, over 5682856.02 frames. ], batch size: 60, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:53:13,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=932967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:53:15,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=932970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:53:32,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:53:34,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=932991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:53:43,195 INFO [train.py:968] (1/2) Epoch 21, batch 20650, giga_loss[loss=0.2733, simple_loss=0.349, pruned_loss=0.09886, over 28881.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3529, pruned_loss=0.1018, over 5693799.78 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08962, over 5763858.32 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3527, pruned_loss=0.1022, over 5695195.73 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:53:58,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:23,895 INFO [train.py:968] (1/2) Epoch 21, batch 20700, giga_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1225, over 28704.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3548, pruned_loss=0.1032, over 5684162.44 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3435, pruned_loss=0.09022, over 5759503.74 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3541, pruned_loss=0.1034, over 5686410.87 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:54:24,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933052.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:54:33,473 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:35,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:36,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:54:54,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:54:54,994 INFO [optim.py:369] (1/2) 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 22:55:00,025 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 20750, giga_loss[loss=0.3137, simple_loss=0.3882, pruned_loss=0.1196, over 28907.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3572, pruned_loss=0.1057, over 5673263.53 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3433, pruned_loss=0.09021, over 5752473.62 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.357, pruned_loss=0.1061, over 5679932.96 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 22:55:17,836 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:19,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:21,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:29,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-10 22:55:40,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:44,127 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:55:45,182 INFO [train.py:968] (1/2) Epoch 21, batch 20800, giga_loss[loss=0.283, simple_loss=0.357, pruned_loss=0.1046, over 28734.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3558, pruned_loss=0.1047, over 5687837.20 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3432, pruned_loss=0.09009, over 5757741.21 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3562, pruned_loss=0.1057, over 5685910.37 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:56:13,276 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1894, 1.3383, 1.1929, 1.0025], device='cuda:1'), covar=tensor([0.0910, 0.0427, 0.0942, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0514, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 22:56:14,323 INFO [optim.py:369] (1/2) 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 22:56:18,977 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933195.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:56:20,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933198.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:56:22,734 INFO [train.py:968] (1/2) Epoch 21, batch 20850, giga_loss[loss=0.2673, simple_loss=0.3499, pruned_loss=0.09235, over 28555.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3556, pruned_loss=0.104, over 5698175.93 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3431, pruned_loss=0.09011, over 5759977.80 frames. ], giga_tot_loss[loss=0.2831, simple_loss=0.3562, pruned_loss=0.105, over 5693750.36 frames. ], batch size: 71, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:56:44,354 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933227.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:56:46,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:56:49,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 22:57:00,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0942, 1.3227, 1.0892, 0.9133], device='cuda:1'), covar=tensor([0.1101, 0.0536, 0.1106, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0444, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 22:57:02,892 INFO [train.py:968] (1/2) Epoch 21, batch 20900, libri_loss[loss=0.2847, simple_loss=0.3513, pruned_loss=0.109, over 29561.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3545, pruned_loss=0.1022, over 5700135.44 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.343, pruned_loss=0.09, over 5763747.23 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3553, pruned_loss=0.1033, over 5692009.91 frames. ], batch size: 79, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:57:12,306 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:57:12,945 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933263.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 22:57:14,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:57:30,531 INFO [optim.py:369] (1/2) 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 22:57:34,508 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:57:39,792 INFO [train.py:968] (1/2) Epoch 21, batch 20950, giga_loss[loss=0.2762, simple_loss=0.3478, pruned_loss=0.1023, over 28792.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3544, pruned_loss=0.101, over 5707402.02 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.08963, over 5768568.10 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3557, pruned_loss=0.1027, over 5694397.56 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:57:49,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:58:19,540 INFO [train.py:968] (1/2) Epoch 21, batch 21000, giga_loss[loss=0.2608, simple_loss=0.3394, pruned_loss=0.09115, over 29002.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3532, pruned_loss=0.1008, over 5710705.84 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3429, pruned_loss=0.08979, over 5770471.74 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3542, pruned_loss=0.1021, over 5697483.37 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:58:19,540 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 22:58:24,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4601, 1.7976, 1.4169, 1.4266], device='cuda:1'), covar=tensor([0.3090, 0.2971, 0.3433, 0.2588], device='cuda:1'), in_proj_covar=tensor([0.1494, 0.1084, 0.1317, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 22:58:28,140 INFO [train.py:1012] (1/2) Epoch 21, validation: loss=0.2087, simple_loss=0.3155, pruned_loss=0.05099, over 944034.00 frames. 
+2023-03-10 22:58:28,141 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 22:58:41,436 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:58:43,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8587, 1.9491, 1.4287, 1.4277], device='cuda:1'), covar=tensor([0.0987, 0.0672, 0.1063, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0445, 0.0516, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 22:58:44,865 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:58:55,188 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 21050, giga_loss[loss=0.339, simple_loss=0.3885, pruned_loss=0.1448, over 26649.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3507, pruned_loss=0.09967, over 5705344.33 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3429, pruned_loss=0.08988, over 5763467.70 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3516, pruned_loss=0.1008, over 5700424.05 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:59:27,411 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-10 22:59:35,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:59:41,680 INFO [train.py:968] (1/2) Epoch 21, batch 21100, giga_loss[loss=0.2742, simple_loss=0.3484, pruned_loss=0.09997, over 28603.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3489, pruned_loss=0.09831, over 5710129.39 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.09, over 5761362.05 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3494, pruned_loss=0.09933, over 5706795.71 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 22:59:44,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 22:59:46,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:00:09,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:00:11,785 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 21150, giga_loss[loss=0.2742, simple_loss=0.3508, pruned_loss=0.09878, over 28818.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3483, pruned_loss=0.09859, over 5705258.03 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09005, over 5761937.59 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3488, pruned_loss=0.09949, over 5701132.58 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:00:39,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:00:43,343 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:01:02,802 INFO [train.py:968] (1/2) Epoch 21, batch 21200, giga_loss[loss=0.307, simple_loss=0.3668, pruned_loss=0.1236, over 28890.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3498, pruned_loss=0.09948, over 5696423.66 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3442, pruned_loss=0.09051, over 5743225.51 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3495, pruned_loss=0.09999, over 5708691.85 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:01:28,767 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:01:30,915 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:01:31,332 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 21250, giga_loss[loss=0.2306, simple_loss=0.3195, pruned_loss=0.0708, over 28568.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3486, pruned_loss=0.09853, over 5695629.75 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3442, pruned_loss=0.09056, over 5744881.98 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3485, pruned_loss=0.09897, over 5703431.45 frames. ], batch size: 60, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:01:52,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-10 23:01:55,270 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:02:20,310 INFO [train.py:968] (1/2) Epoch 21, batch 21300, giga_loss[loss=0.2425, simple_loss=0.3276, pruned_loss=0.07871, over 28906.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3468, pruned_loss=0.09659, over 5704380.27 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3444, pruned_loss=0.09072, over 5747018.71 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3466, pruned_loss=0.09691, over 5707800.34 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:02:37,007 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:02:38,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-10 23:02:38,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:02:48,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-10 23:02:52,243 INFO [optim.py:369] (1/2) 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 23:02:59,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:03:00,898 INFO [train.py:968] (1/2) Epoch 21, batch 21350, giga_loss[loss=0.2461, simple_loss=0.3247, pruned_loss=0.08371, over 28678.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3464, pruned_loss=0.09629, over 5714707.78 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3446, pruned_loss=0.09091, over 5747611.34 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.346, pruned_loss=0.09645, over 5716296.90 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:03:30,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1892, 1.8392, 1.7071, 1.3390], device='cuda:1'), covar=tensor([0.0768, 0.0260, 0.0256, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0116, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 23:03:33,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:03:36,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:03:37,616 INFO [train.py:968] (1/2) Epoch 21, batch 21400, giga_loss[loss=0.2719, simple_loss=0.3396, pruned_loss=0.1021, over 28682.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3449, pruned_loss=0.09596, over 5720794.48 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3447, pruned_loss=0.09103, over 5752388.31 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3445, pruned_loss=0.09613, over 5717039.11 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:03:45,448 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-10 23:04:06,861 INFO [optim.py:369] (1/2) 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 23:04:15,077 INFO [train.py:968] (1/2) Epoch 21, batch 21450, giga_loss[loss=0.2603, simple_loss=0.3385, pruned_loss=0.09108, over 28620.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3422, pruned_loss=0.0944, over 5718891.03 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3454, pruned_loss=0.09136, over 5756162.24 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3413, pruned_loss=0.09433, over 5711678.25 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:04:53,982 INFO [train.py:968] (1/2) Epoch 21, batch 21500, giga_loss[loss=0.2787, simple_loss=0.354, pruned_loss=0.1017, over 28013.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3402, pruned_loss=0.09311, over 5725397.94 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3455, pruned_loss=0.09139, over 5758706.93 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3393, pruned_loss=0.09306, over 5716630.28 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:05:01,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=933862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:21,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:22,669 INFO [optim.py:369] (1/2) 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 23:05:22,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:24,369 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=933892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:26,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=933895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:29,897 INFO [train.py:968] (1/2) Epoch 21, batch 21550, giga_loss[loss=0.2536, simple_loss=0.3305, pruned_loss=0.08831, over 28933.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.34, pruned_loss=0.0932, over 5715559.64 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09166, over 5742022.77 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3388, pruned_loss=0.09294, over 5721921.72 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:05:32,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=933904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:43,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=933919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:05:46,685 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 21600, giga_loss[loss=0.2663, simple_loss=0.3418, pruned_loss=0.09542, over 28680.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3408, pruned_loss=0.09481, over 5718994.65 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.346, pruned_loss=0.09203, over 5751015.08 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3394, pruned_loss=0.09441, over 5714678.27 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:06:23,596 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-10 23:06:38,507 INFO [optim.py:369] (1/2) 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 23:06:47,534 INFO [train.py:968] (1/2) Epoch 21, batch 21650, giga_loss[loss=0.2531, simple_loss=0.3331, pruned_loss=0.08658, over 28720.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3382, pruned_loss=0.0938, over 5717161.24 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3457, pruned_loss=0.09183, over 5753758.77 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3372, pruned_loss=0.09371, over 5710685.11 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:07:23,144 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 21700, giga_loss[loss=0.2739, simple_loss=0.3454, pruned_loss=0.1012, over 28848.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3355, pruned_loss=0.09298, over 5712845.06 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.0919, over 5751703.21 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3346, pruned_loss=0.09286, over 5709396.98 frames. ], batch size: 285, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:07:50,776 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=934079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:07:58,662 INFO [optim.py:369] (1/2) 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 23:08:07,547 INFO [train.py:968] (1/2) Epoch 21, batch 21750, giga_loss[loss=0.2432, simple_loss=0.3173, pruned_loss=0.08452, over 29041.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3339, pruned_loss=0.0926, over 5710709.92 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3463, pruned_loss=0.09215, over 5750161.27 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3326, pruned_loss=0.09231, over 5708389.90 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:08:16,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1086, 2.0938, 1.6353, 1.5066], device='cuda:1'), covar=tensor([0.0642, 0.0209, 0.0232, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-10 23:08:46,399 INFO [train.py:968] (1/2) Epoch 21, batch 21800, giga_loss[loss=0.2533, simple_loss=0.3295, pruned_loss=0.08855, over 28963.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3341, pruned_loss=0.0929, over 5710720.64 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3465, pruned_loss=0.09254, over 5754029.80 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3324, pruned_loss=0.09235, over 5704491.90 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:08:57,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-10 23:09:17,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3909, 1.6577, 1.3659, 1.0043], device='cuda:1'), covar=tensor([0.2643, 0.2731, 0.3097, 0.2502], device='cuda:1'), in_proj_covar=tensor([0.1490, 0.1079, 0.1314, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 23:09:19,508 INFO [optim.py:369] (1/2) 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 23:09:27,863 INFO [train.py:968] (1/2) Epoch 21, batch 21850, giga_loss[loss=0.2346, simple_loss=0.3116, pruned_loss=0.07878, over 28670.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3357, pruned_loss=0.09358, over 5712419.62 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3464, pruned_loss=0.09263, over 5756095.57 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3342, pruned_loss=0.09307, over 5704793.11 frames. ], batch size: 60, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:10:01,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=934237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:10:11,928 INFO [train.py:968] (1/2) Epoch 21, batch 21900, giga_loss[loss=0.242, simple_loss=0.3283, pruned_loss=0.07786, over 28964.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3379, pruned_loss=0.09428, over 5704936.85 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3468, pruned_loss=0.09291, over 5748886.93 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3362, pruned_loss=0.09363, over 5705085.59 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:10:46,573 INFO [optim.py:369] (1/2) 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 23:10:55,192 INFO [train.py:968] (1/2) Epoch 21, batch 21950, giga_loss[loss=0.2565, simple_loss=0.3428, pruned_loss=0.08514, over 28605.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3402, pruned_loss=0.09427, over 5704251.97 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3471, pruned_loss=0.09314, over 5748228.67 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3385, pruned_loss=0.09357, over 5704611.80 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:11:36,957 INFO [train.py:968] (1/2) Epoch 21, batch 22000, giga_loss[loss=0.2353, simple_loss=0.3145, pruned_loss=0.07805, over 28868.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3401, pruned_loss=0.09365, over 5703105.62 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3472, pruned_loss=0.09334, over 5750828.97 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3386, pruned_loss=0.09295, over 5700167.76 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:12:03,653 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=934380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:12:05,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=934383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:12:12,870 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 22050, giga_loss[loss=0.2621, simple_loss=0.3366, pruned_loss=0.09383, over 28826.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3389, pruned_loss=0.09259, over 5703530.65 frames. ], libri_tot_loss[loss=0.2674, simple_loss=0.3476, pruned_loss=0.09365, over 5755493.18 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3371, pruned_loss=0.0917, over 5695064.90 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:12:28,313 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=934412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:12:43,016 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-10 23:13:00,206 INFO [train.py:968] (1/2) Epoch 21, batch 22100, giga_loss[loss=0.2884, simple_loss=0.3617, pruned_loss=0.1075, over 28853.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3408, pruned_loss=0.09406, over 5711395.65 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3481, pruned_loss=0.09413, over 5759643.06 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3386, pruned_loss=0.09292, over 5699600.70 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:13:30,661 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 22150, giga_loss[loss=0.2221, simple_loss=0.3049, pruned_loss=0.06967, over 28905.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3426, pruned_loss=0.09577, over 5710961.98 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3488, pruned_loss=0.09487, over 5762324.08 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.34, pruned_loss=0.0942, over 5697512.86 frames. ], batch size: 66, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:14:18,634 INFO [train.py:968] (1/2) Epoch 21, batch 22200, giga_loss[loss=0.2715, simple_loss=0.3463, pruned_loss=0.09833, over 28800.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3451, pruned_loss=0.09746, over 5712199.74 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3492, pruned_loss=0.09512, over 5763317.18 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3426, pruned_loss=0.096, over 5700225.89 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:14:52,354 INFO [optim.py:369] (1/2) 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 23:14:59,060 INFO [train.py:968] (1/2) Epoch 21, batch 22250, giga_loss[loss=0.269, simple_loss=0.3359, pruned_loss=0.101, over 28712.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3496, pruned_loss=0.1004, over 5714395.66 frames. ], libri_tot_loss[loss=0.2709, simple_loss=0.35, pruned_loss=0.09585, over 5767229.72 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3469, pruned_loss=0.09866, over 5700087.40 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:15:08,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6455, 1.6642, 1.8301, 1.3869], device='cuda:1'), covar=tensor([0.1857, 0.2356, 0.1489, 0.1723], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0694, 0.0940, 0.0838], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 23:15:19,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2552, 2.2793, 2.0228, 1.9677], device='cuda:1'), covar=tensor([0.0737, 0.0556, 0.0858, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0443, 0.0511, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 23:15:38,465 INFO [train.py:968] (1/2) Epoch 21, batch 22300, giga_loss[loss=0.2396, simple_loss=0.3245, pruned_loss=0.07732, over 28995.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3523, pruned_loss=0.1015, over 5719610.83 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3507, pruned_loss=0.09642, over 5769646.45 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3495, pruned_loss=0.09973, over 5704318.46 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:16:11,085 INFO [optim.py:369] (1/2) 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 23:16:16,015 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:16:18,570 INFO [train.py:968] (1/2) Epoch 21, batch 22350, giga_loss[loss=0.2748, simple_loss=0.3535, pruned_loss=0.09809, over 28999.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3527, pruned_loss=0.1013, over 5721734.19 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3508, pruned_loss=0.09649, over 5769989.96 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3504, pruned_loss=0.09993, over 5708735.91 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:16:35,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6068, 1.7331, 1.4083, 1.7404], device='cuda:1'), covar=tensor([0.2829, 0.3000, 0.3308, 0.2809], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1080, 0.1314, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 23:17:00,606 INFO [train.py:968] (1/2) Epoch 21, batch 22400, libri_loss[loss=0.2966, simple_loss=0.3688, pruned_loss=0.1121, over 29543.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3534, pruned_loss=0.1017, over 5714610.05 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3514, pruned_loss=0.09693, over 5764501.29 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3512, pruned_loss=0.1004, over 5707787.54 frames. ], batch size: 83, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:17:09,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9471, 2.2164, 1.3785, 1.7230], device='cuda:1'), covar=tensor([0.0902, 0.0669, 0.1129, 0.1206], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0446, 0.0514, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 23:17:24,923 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:17:36,563 INFO [optim.py:369] (1/2) 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 23:17:39,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:17:44,490 INFO [train.py:968] (1/2) Epoch 21, batch 22450, giga_loss[loss=0.3184, simple_loss=0.3711, pruned_loss=0.1329, over 23907.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3535, pruned_loss=0.1023, over 5711429.67 frames. ], libri_tot_loss[loss=0.2732, simple_loss=0.3518, pruned_loss=0.09734, over 5766400.39 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3513, pruned_loss=0.1009, over 5703662.50 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:18:04,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4195, 2.1336, 1.6734, 0.6480], device='cuda:1'), covar=tensor([0.6051, 0.2864, 0.4428, 0.6769], device='cuda:1'), in_proj_covar=tensor([0.1736, 0.1630, 0.1596, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 23:18:23,162 INFO [train.py:968] (1/2) Epoch 21, batch 22500, giga_loss[loss=0.2564, simple_loss=0.3276, pruned_loss=0.09259, over 28462.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.351, pruned_loss=0.1007, over 5722359.42 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3526, pruned_loss=0.09794, over 5770406.19 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3486, pruned_loss=0.09924, over 5710927.17 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:18:58,195 INFO [optim.py:369] (1/2) 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 23:19:03,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:19:04,210 INFO [train.py:968] (1/2) Epoch 21, batch 22550, giga_loss[loss=0.2416, simple_loss=0.3134, pruned_loss=0.08495, over 28776.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3486, pruned_loss=0.09983, over 5717094.87 frames. ], libri_tot_loss[loss=0.275, simple_loss=0.3531, pruned_loss=0.09841, over 5775298.61 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3461, pruned_loss=0.09828, over 5702195.24 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:19:20,587 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:19:41,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2752, 1.7047, 1.4530, 1.4613], device='cuda:1'), covar=tensor([0.0768, 0.0296, 0.0338, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 23:19:46,175 INFO [train.py:968] (1/2) Epoch 21, batch 22600, giga_loss[loss=0.2901, simple_loss=0.357, pruned_loss=0.1116, over 28872.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3448, pruned_loss=0.09765, over 5718276.39 frames. ], libri_tot_loss[loss=0.2752, simple_loss=0.3533, pruned_loss=0.09854, over 5776067.12 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09633, over 5705766.90 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:19:52,697 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=934960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:20:19,280 INFO [optim.py:369] (1/2) 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 23:20:25,267 INFO [train.py:968] (1/2) Epoch 21, batch 22650, giga_loss[loss=0.2883, simple_loss=0.3746, pruned_loss=0.101, over 28623.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3462, pruned_loss=0.09737, over 5715658.77 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3541, pruned_loss=0.09936, over 5779378.95 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3435, pruned_loss=0.09551, over 5700971.46 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:20:31,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-10 23:20:43,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3232, 3.2858, 1.4177, 1.5242], device='cuda:1'), covar=tensor([0.0968, 0.0286, 0.0990, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0550, 0.0383, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-10 23:21:08,585 INFO [train.py:968] (1/2) Epoch 21, batch 22700, giga_loss[loss=0.2925, simple_loss=0.3594, pruned_loss=0.1128, over 28811.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3477, pruned_loss=0.09696, over 5710942.54 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3543, pruned_loss=0.09948, over 5780025.05 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3454, pruned_loss=0.09538, over 5698570.02 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:21:25,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:21:40,590 INFO [optim.py:369] (1/2) 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 23:21:46,594 INFO [train.py:968] (1/2) Epoch 21, batch 22750, giga_loss[loss=0.3262, simple_loss=0.3846, pruned_loss=0.1339, over 27721.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3461, pruned_loss=0.09691, over 5708430.74 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3539, pruned_loss=0.09943, over 5781265.93 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3445, pruned_loss=0.09565, over 5696321.38 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:22:27,579 INFO [train.py:968] (1/2) Epoch 21, batch 22800, giga_loss[loss=0.255, simple_loss=0.3116, pruned_loss=0.09925, over 28152.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3452, pruned_loss=0.09792, over 5710371.14 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3544, pruned_loss=0.09999, over 5783679.94 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3432, pruned_loss=0.09636, over 5696944.82 frames. ], batch size: 77, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:22:27,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:22:44,647 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:23:07,780 INFO [train.py:968] (1/2) Epoch 21, batch 22850, giga_loss[loss=0.2341, simple_loss=0.3006, pruned_loss=0.08382, over 28498.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3433, pruned_loss=0.0981, over 5716760.84 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3548, pruned_loss=0.1003, over 5785817.93 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3412, pruned_loss=0.09655, over 5702763.16 frames. ], batch size: 85, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:23:20,072 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:23:22,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:23:43,021 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.29 vs. limit=5.0
+2023-03-10 23:23:46,413 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 22900, giga_loss[loss=0.2416, simple_loss=0.3139, pruned_loss=0.08461, over 29003.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3407, pruned_loss=0.09753, over 5711003.55 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3549, pruned_loss=0.1006, over 5778716.82 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3386, pruned_loss=0.09591, over 5704324.28 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:24:07,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:22,235 INFO [optim.py:369] (1/2) 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 23:24:22,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:23,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:24,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:27,903 INFO [train.py:968] (1/2) Epoch 21, batch 22950, giga_loss[loss=0.2393, simple_loss=0.3165, pruned_loss=0.08106, over 28996.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.34, pruned_loss=0.09712, over 5716358.33 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3546, pruned_loss=0.1005, over 5780619.33 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3384, pruned_loss=0.09589, over 5708495.38 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:24:36,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:40,050 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:47,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:24:53,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6176, 1.0105, 4.7728, 3.6393], device='cuda:1'), covar=tensor([0.1568, 0.3070, 0.0374, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0643, 0.0953, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 23:24:54,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=935335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:25:02,527 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 23000, giga_loss[loss=0.2491, simple_loss=0.3197, pruned_loss=0.08922, over 28910.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3375, pruned_loss=0.096, over 5702520.25 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3549, pruned_loss=0.1009, over 5762522.58 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3356, pruned_loss=0.09457, over 5710969.01 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:25:38,424 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 23050, giga_loss[loss=0.2163, simple_loss=0.3019, pruned_loss=0.0653, over 28741.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3334, pruned_loss=0.0942, over 5701096.51 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.355, pruned_loss=0.1011, over 5763932.03 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3315, pruned_loss=0.09276, over 5705645.33 frames. ], batch size: 284, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:25:57,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:00,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:04,340 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-03-10 23:26:14,188 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:16,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:21,722 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:22,117 INFO [train.py:968] (1/2) Epoch 21, batch 23100, giga_loss[loss=0.241, simple_loss=0.3215, pruned_loss=0.08026, over 29028.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3304, pruned_loss=0.09224, over 5709518.08 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.355, pruned_loss=0.1012, over 5767365.54 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3283, pruned_loss=0.09081, over 5708529.74 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:26:38,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:44,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=935478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:46,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=935481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:26:57,502 INFO [optim.py:369] (1/2) 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 23:27:02,445 INFO [train.py:968] (1/2) Epoch 21, batch 23150, giga_loss[loss=0.2098, simple_loss=0.2926, pruned_loss=0.06356, over 28944.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3314, pruned_loss=0.0924, over 5712456.48 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3553, pruned_loss=0.1015, over 5768691.54 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3289, pruned_loss=0.09074, over 5709196.05 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:27:08,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5663, 4.4132, 4.2108, 2.0558], device='cuda:1'), covar=tensor([0.0614, 0.0730, 0.0810, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.1194, 0.1101, 0.0939, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-10 23:27:09,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5588, 1.6165, 1.5867, 1.4606], device='cuda:1'), covar=tensor([0.2870, 0.2435, 0.1965, 0.2363], device='cuda:1'), in_proj_covar=tensor([0.1981, 0.1900, 0.1841, 0.1975], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 23:27:10,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=935510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:27:43,468 INFO [train.py:968] (1/2) Epoch 21, batch 23200, libri_loss[loss=0.2537, simple_loss=0.3226, pruned_loss=0.09238, over 28071.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3344, pruned_loss=0.09369, over 5715147.03 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3553, pruned_loss=0.1017, over 5770458.63 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3319, pruned_loss=0.092, over 5709843.30 frames. ], batch size: 62, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:27:45,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-10 23:28:04,494 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:28:12,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2839, 1.2764, 3.8868, 3.2496], device='cuda:1'), covar=tensor([0.1648, 0.2734, 0.0455, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0642, 0.0951, 0.0902], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 23:28:19,971 INFO [optim.py:369] (1/2) 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 23:28:22,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6459, 1.6562, 1.8980, 1.4551], device='cuda:1'), covar=tensor([0.1749, 0.2276, 0.1439, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0692, 0.0937, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 23:28:25,980 INFO [train.py:968] (1/2) Epoch 21, batch 23250, giga_loss[loss=0.2807, simple_loss=0.366, pruned_loss=0.09771, over 28823.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3391, pruned_loss=0.09568, over 5711756.21 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3553, pruned_loss=0.1017, over 5770458.63 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3371, pruned_loss=0.09437, over 5707628.25 frames. ], batch size: 174, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:29:05,165 INFO [train.py:968] (1/2) Epoch 21, batch 23300, giga_loss[loss=0.2625, simple_loss=0.3469, pruned_loss=0.08909, over 28573.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3426, pruned_loss=0.09733, over 5707736.77 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3557, pruned_loss=0.1021, over 5773920.28 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3402, pruned_loss=0.09573, over 5699198.08 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:29:09,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9446, 1.9497, 1.9272, 1.8504], device='cuda:1'), covar=tensor([0.2012, 0.2796, 0.2298, 0.2474], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0744, 0.0712, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-10 23:29:43,029 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 23350, giga_loss[loss=0.3063, simple_loss=0.3819, pruned_loss=0.1153, over 28965.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3465, pruned_loss=0.0993, over 5696951.64 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3565, pruned_loss=0.1029, over 5766444.19 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3437, pruned_loss=0.09731, over 5695598.26 frames. ], batch size: 227, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:30:32,040 INFO [train.py:968] (1/2) Epoch 21, batch 23400, giga_loss[loss=0.4769, simple_loss=0.4724, pruned_loss=0.2407, over 23429.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3509, pruned_loss=0.1033, over 5674709.28 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3568, pruned_loss=0.1032, over 5750006.29 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3482, pruned_loss=0.1013, over 5686630.63 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:31:13,316 INFO [optim.py:369] (1/2) 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 23:31:19,379 INFO [train.py:968] (1/2) Epoch 21, batch 23450, giga_loss[loss=0.3485, simple_loss=0.3838, pruned_loss=0.1566, over 23572.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3559, pruned_loss=0.1076, over 5677205.41 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3568, pruned_loss=0.1036, over 5753773.79 frames. ], giga_tot_loss[loss=0.2825, simple_loss=0.3536, pruned_loss=0.1057, over 5681198.95 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:31:41,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5725, 1.8158, 1.4810, 1.5891], device='cuda:1'), covar=tensor([0.2664, 0.2738, 0.3054, 0.2495], device='cuda:1'), in_proj_covar=tensor([0.1494, 0.1085, 0.1317, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 23:32:10,048 INFO [train.py:968] (1/2) Epoch 21, batch 23500, giga_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1217, over 28721.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3623, pruned_loss=0.1115, over 5685481.95 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3573, pruned_loss=0.104, over 5752978.55 frames. ], giga_tot_loss[loss=0.2898, simple_loss=0.36, pruned_loss=0.1098, over 5688284.90 frames. ], batch size: 66, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:32:22,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:32:50,666 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 23550, giga_loss[loss=0.3337, simple_loss=0.3944, pruned_loss=0.1365, over 28875.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3672, pruned_loss=0.1153, over 5684189.73 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3575, pruned_loss=0.1041, over 5758308.76 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3655, pruned_loss=0.1141, over 5679259.06 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:33:07,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2194, 1.2038, 3.7497, 3.2592], device='cuda:1'), covar=tensor([0.1676, 0.2775, 0.0476, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0750, 0.0642, 0.0954, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 23:33:32,667 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:33:46,266 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 23600, giga_loss[loss=0.4029, simple_loss=0.4229, pruned_loss=0.1915, over 23587.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3744, pruned_loss=0.1222, over 5649080.41 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.358, pruned_loss=0.1046, over 5739488.02 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3727, pruned_loss=0.121, over 5659776.96 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:33:54,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=935960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:34:28,031 INFO [optim.py:369] (1/2) 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 23:34:34,186 INFO [train.py:968] (1/2) Epoch 21, batch 23650, giga_loss[loss=0.3736, simple_loss=0.4206, pruned_loss=0.1633, over 27885.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3788, pruned_loss=0.1256, over 5643226.21 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.358, pruned_loss=0.1048, over 5732495.39 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.378, pruned_loss=0.1249, over 5656459.71 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:35:10,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6542, 1.8424, 1.3716, 1.3389], device='cuda:1'), covar=tensor([0.0885, 0.0529, 0.0944, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0446, 0.0515, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 23:35:19,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8915, 2.0510, 1.5694, 1.5197], device='cuda:1'), covar=tensor([0.0924, 0.0652, 0.0964, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0446, 0.0515, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 23:35:19,540 INFO [train.py:968] (1/2) Epoch 21, batch 23700, giga_loss[loss=0.2705, simple_loss=0.3457, pruned_loss=0.09763, over 28889.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.38, pruned_loss=0.1269, over 5653293.03 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3583, pruned_loss=0.1051, over 5735831.36 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3795, pruned_loss=0.1265, over 5659382.05 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:35:35,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7383, 2.0062, 1.3864, 1.6290], device='cuda:1'), covar=tensor([0.0816, 0.0460, 0.0955, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0446, 0.0515, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 23:35:58,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:01,651 INFO [optim.py:369] (1/2) 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 23:36:02,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:07,189 INFO [train.py:968] (1/2) Epoch 21, batch 23750, giga_loss[loss=0.3604, simple_loss=0.4066, pruned_loss=0.1571, over 28878.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3821, pruned_loss=0.1298, over 5658833.40 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3582, pruned_loss=0.1053, over 5737150.64 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3824, pruned_loss=0.13, over 5660120.65 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:36:12,902 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:31,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:36:31,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4396, 1.5518, 1.6760, 1.3341], device='cuda:1'), covar=tensor([0.0957, 0.1600, 0.0820, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0692, 0.0935, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 23:36:44,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3642, 2.6618, 1.4555, 1.4080], device='cuda:1'), covar=tensor([0.0865, 0.0365, 0.0836, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0552, 0.0384, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-10 23:36:54,406 INFO [train.py:968] (1/2) Epoch 21, batch 23800, giga_loss[loss=0.3416, simple_loss=0.3948, pruned_loss=0.1442, over 28872.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3846, pruned_loss=0.133, over 5635871.95 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3586, pruned_loss=0.1058, over 5731762.49 frames. ], giga_tot_loss[loss=0.3262, simple_loss=0.3853, pruned_loss=0.1336, over 5639422.10 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:37:40,018 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 23850, giga_loss[loss=0.3241, simple_loss=0.3875, pruned_loss=0.1303, over 28624.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3873, pruned_loss=0.1359, over 5634088.38 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3579, pruned_loss=0.1058, over 5736751.47 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3893, pruned_loss=0.1372, over 5629490.80 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:38:31,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:38:37,438 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-10 23:38:43,253 INFO [train.py:968] (1/2) Epoch 21, batch 23900, giga_loss[loss=0.3591, simple_loss=0.4036, pruned_loss=0.1573, over 28628.00 frames. ], tot_loss[loss=0.3333, simple_loss=0.3891, pruned_loss=0.1387, over 5611690.66 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3579, pruned_loss=0.1058, over 5738318.57 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3911, pruned_loss=0.1401, over 5605416.22 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:39:25,586 INFO [optim.py:369] (1/2) 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 23:39:28,901 INFO [train.py:968] (1/2) Epoch 21, batch 23950, giga_loss[loss=0.3767, simple_loss=0.4147, pruned_loss=0.1693, over 28563.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3887, pruned_loss=0.1394, over 5620004.85 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3584, pruned_loss=0.1066, over 5731260.96 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3909, pruned_loss=0.1409, over 5617150.73 frames. ], batch size: 336, lr: 1.52e-03, grad_scale: 2.0
+2023-03-10 23:39:40,594 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:40:01,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:40:09,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:40:14,503 INFO [train.py:968] (1/2) Epoch 21, batch 24000, giga_loss[loss=0.33, simple_loss=0.3881, pruned_loss=0.136, over 28659.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3869, pruned_loss=0.138, over 5635193.88 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3584, pruned_loss=0.1066, over 5732768.63 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3895, pruned_loss=0.1401, over 5628468.31 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:40:14,503 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-10 23:40:23,183 INFO [train.py:1012] (1/2) Epoch 21, validation: loss=0.2051, simple_loss=0.3129, pruned_loss=0.0486, over 944034.00 frames. 
+2023-03-10 23:40:23,184 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-10 23:40:37,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.62 vs. limit=5.0
+2023-03-10 23:40:49,556 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:40:52,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:41:04,345 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 24050, libri_loss[loss=0.2864, simple_loss=0.3584, pruned_loss=0.1072, over 27688.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3858, pruned_loss=0.1358, over 5630560.05 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3587, pruned_loss=0.107, over 5735857.45 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.3885, pruned_loss=0.1381, over 5618935.70 frames. ], batch size: 116, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:41:16,135 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5818, 1.8089, 1.6443, 1.6325], device='cuda:1'), covar=tensor([0.1868, 0.2148, 0.2069, 0.1908], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0748, 0.0715, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-10 23:41:22,149 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 24100, giga_loss[loss=0.3178, simple_loss=0.3868, pruned_loss=0.1244, over 28640.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3871, pruned_loss=0.1363, over 5631212.25 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3587, pruned_loss=0.107, over 5740574.12 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3901, pruned_loss=0.139, over 5614289.11 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:42:00,972 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:05,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:24,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:25,667 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:28,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:29,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:35,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:42:41,889 INFO [optim.py:369] (1/2) 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 23:42:48,307 INFO [train.py:968] (1/2) Epoch 21, batch 24150, giga_loss[loss=0.2887, simple_loss=0.3597, pruned_loss=0.1089, over 28421.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3861, pruned_loss=0.1349, over 5634527.20 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3586, pruned_loss=0.1072, over 5742013.94 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3896, pruned_loss=0.1379, over 5615948.59 frames. ], batch size: 71, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:42:56,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:43:39,202 INFO [train.py:968] (1/2) Epoch 21, batch 24200, giga_loss[loss=0.2966, simple_loss=0.3685, pruned_loss=0.1123, over 28780.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3833, pruned_loss=0.1321, over 5630902.34 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3589, pruned_loss=0.1074, over 5745161.73 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3863, pruned_loss=0.1347, over 5611767.68 frames. ], batch size: 99, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:44:04,667 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936582.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 23:44:18,710 INFO [optim.py:369] (1/2) 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 23:44:22,761 INFO [train.py:968] (1/2) Epoch 21, batch 24250, giga_loss[loss=0.3607, simple_loss=0.3869, pruned_loss=0.1673, over 24154.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.381, pruned_loss=0.1292, over 5643334.63 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3586, pruned_loss=0.1075, over 5747753.40 frames. ], giga_tot_loss[loss=0.3242, simple_loss=0.3844, pruned_loss=0.132, over 5622358.09 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:44:28,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9301, 2.4767, 2.2449, 1.6874], device='cuda:1'), covar=tensor([0.3361, 0.2213, 0.2373, 0.2899], device='cuda:1'), in_proj_covar=tensor([0.1976, 0.1892, 0.1831, 0.1962], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 23:44:46,606 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:44:48,875 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 24300, giga_loss[loss=0.3508, simple_loss=0.3881, pruned_loss=0.1568, over 23394.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.379, pruned_loss=0.1277, over 5638458.26 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3584, pruned_loss=0.1076, over 5749488.91 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3822, pruned_loss=0.1302, over 5618274.49 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:45:16,471 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=936657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:45:54,151 INFO [optim.py:369] (1/2) 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 23:45:58,963 INFO [train.py:968] (1/2) Epoch 21, batch 24350, giga_loss[loss=0.2512, simple_loss=0.3363, pruned_loss=0.08301, over 28776.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3761, pruned_loss=0.1253, over 5636253.05 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3584, pruned_loss=0.1077, over 5740755.92 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3788, pruned_loss=0.1273, over 5627296.89 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:46:10,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5398, 2.5251, 1.8536, 2.1203], device='cuda:1'), covar=tensor([0.0841, 0.0673, 0.0982, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0447, 0.0516, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-10 23:46:15,847 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:46:31,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3255, 1.2657, 3.8581, 3.3409], device='cuda:1'), covar=tensor([0.1894, 0.2935, 0.0763, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0648, 0.0963, 0.0912], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-10 23:46:44,448 INFO [train.py:968] (1/2) Epoch 21, batch 24400, giga_loss[loss=0.275, simple_loss=0.3562, pruned_loss=0.09687, over 28860.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.375, pruned_loss=0.1249, over 5634375.66 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3586, pruned_loss=0.1078, over 5742657.44 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3774, pruned_loss=0.1268, over 5623214.42 frames. ], batch size: 145, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:46:48,819 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:47:34,641 INFO [train.py:968] (1/2) Epoch 21, batch 24450, giga_loss[loss=0.2989, simple_loss=0.3712, pruned_loss=0.1133, over 28741.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3751, pruned_loss=0.1245, over 5641400.32 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3592, pruned_loss=0.1081, over 5740507.60 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3769, pruned_loss=0.1261, over 5631983.04 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:47:39,535 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:47:53,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=936817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:48:12,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5706, 1.7913, 1.4402, 1.5936], device='cuda:1'), covar=tensor([0.2625, 0.2698, 0.3056, 0.2359], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1082, 0.1316, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 23:48:24,275 INFO [train.py:968] (1/2) Epoch 21, batch 24500, giga_loss[loss=0.2828, simple_loss=0.355, pruned_loss=0.1053, over 28645.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.374, pruned_loss=0.1232, over 5652044.44 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3588, pruned_loss=0.1081, over 5743891.54 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3762, pruned_loss=0.125, over 5638646.29 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:48:25,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:48:36,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:48:39,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=936865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:49:09,846 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 24550, libri_loss[loss=0.3324, simple_loss=0.391, pruned_loss=0.1369, over 29674.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3744, pruned_loss=0.1219, over 5654404.72 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3593, pruned_loss=0.1086, over 5742288.96 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3758, pruned_loss=0.1231, over 5644049.81 frames. ], batch size: 88, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:49:39,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-10 23:49:45,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-10 23:50:09,500 INFO [train.py:968] (1/2) Epoch 21, batch 24600, giga_loss[loss=0.3614, simple_loss=0.417, pruned_loss=0.1529, over 28869.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3762, pruned_loss=0.1213, over 5663338.85 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3588, pruned_loss=0.1083, over 5744495.27 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3781, pruned_loss=0.1226, over 5651989.86 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:50:18,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=936957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 23:50:54,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=936995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:50:56,138 INFO [optim.py:369] (1/2) 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 23:50:57,170 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 24650, giga_loss[loss=0.3088, simple_loss=0.3706, pruned_loss=0.1236, over 28851.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3765, pruned_loss=0.1222, over 5651573.51 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3589, pruned_loss=0.1086, over 5734400.39 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3782, pruned_loss=0.1232, over 5649433.11 frames. ], batch size: 186, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:51:22,825 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:51:44,629 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 24700, giga_loss[loss=0.2873, simple_loss=0.3558, pruned_loss=0.1094, over 28984.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3741, pruned_loss=0.1204, over 5673282.82 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3581, pruned_loss=0.1082, over 5737972.54 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.377, pruned_loss=0.1222, over 5665106.86 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:52:10,398 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=937075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:52:31,947 INFO [optim.py:369] (1/2) 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 23:52:34,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937100.0, num_to_drop=1, layers_to_drop={1}
+2023-03-10 23:52:34,383 INFO [train.py:968] (1/2) Epoch 21, batch 24750, giga_loss[loss=0.2901, simple_loss=0.3618, pruned_loss=0.1092, over 28971.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3717, pruned_loss=0.1192, over 5675151.82 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3578, pruned_loss=0.1081, over 5730366.71 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3745, pruned_loss=0.121, over 5673552.71 frames. ], batch size: 136, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:52:37,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937103.0, num_to_drop=1, layers_to_drop={0}
+2023-03-10 23:52:50,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0225, 1.3426, 1.1228, 0.2027], device='cuda:1'), covar=tensor([0.3918, 0.3031, 0.4544, 0.6476], device='cuda:1'), in_proj_covar=tensor([0.1745, 0.1642, 0.1600, 0.1414], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 23:53:03,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:53:04,296 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 24800, giga_loss[loss=0.3934, simple_loss=0.4127, pruned_loss=0.1871, over 23779.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3714, pruned_loss=0.1206, over 5665063.33 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3575, pruned_loss=0.108, over 5724270.08 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3743, pruned_loss=0.1224, over 5668171.90 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:53:29,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4577, 1.9866, 1.4184, 0.7910], device='cuda:1'), covar=tensor([0.5868, 0.2731, 0.3406, 0.6142], device='cuda:1'), in_proj_covar=tensor([0.1745, 0.1642, 0.1600, 0.1413], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-10 23:53:42,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:53:54,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:54:00,414 INFO [optim.py:369] (1/2) 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 23:54:02,479 INFO [train.py:968] (1/2) Epoch 21, batch 24850, giga_loss[loss=0.3669, simple_loss=0.4142, pruned_loss=0.1598, over 27654.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3709, pruned_loss=0.1203, over 5665485.08 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3576, pruned_loss=0.1081, over 5717262.29 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3734, pruned_loss=0.1218, over 5672362.11 frames. ], batch size: 472, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:54:42,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1526, 1.4452, 1.3003, 1.0816], device='cuda:1'), covar=tensor([0.2886, 0.2546, 0.1759, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.1976, 0.1891, 0.1832, 0.1964], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 23:54:46,020 INFO [train.py:968] (1/2) Epoch 21, batch 24900, giga_loss[loss=0.3006, simple_loss=0.3811, pruned_loss=0.1101, over 28975.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3705, pruned_loss=0.1183, over 5676155.69 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.358, pruned_loss=0.1084, over 5721946.95 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3724, pruned_loss=0.1195, over 5675766.69 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:55:08,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:10,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:24,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5455, 1.7060, 1.6348, 1.5116], device='cuda:1'), covar=tensor([0.2769, 0.2462, 0.2244, 0.2474], device='cuda:1'), in_proj_covar=tensor([0.1972, 0.1888, 0.1829, 0.1960], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-10 23:55:29,112 INFO [optim.py:369] (1/2) 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 23:55:31,164 INFO [train.py:968] (1/2) Epoch 21, batch 24950, giga_loss[loss=0.2821, simple_loss=0.3609, pruned_loss=0.1016, over 28679.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3708, pruned_loss=0.1183, over 5679360.86 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3581, pruned_loss=0.1087, over 5727951.04 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3727, pruned_loss=0.1194, over 5672270.48 frames. ], batch size: 242, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:55:35,470 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:46,967 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:55:50,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:01,705 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:04,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:17,023 INFO [train.py:968] (1/2) Epoch 21, batch 25000, giga_loss[loss=0.223, simple_loss=0.3077, pruned_loss=0.06913, over 28570.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3702, pruned_loss=0.1181, over 5679680.42 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3584, pruned_loss=0.109, over 5728704.09 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3717, pruned_loss=0.1188, over 5672655.99 frames. ], batch size: 60, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:56:20,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:23,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3326, 1.5793, 1.4443, 1.4600], device='cuda:1'), covar=tensor([0.0768, 0.0347, 0.0327, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 23:56:33,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:56:44,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3838, 1.5672, 1.2652, 1.5672], device='cuda:1'), covar=tensor([0.0762, 0.0348, 0.0335, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0106], device='cuda:1')
+2023-03-10 23:56:46,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5328, 1.4770, 1.7723, 1.3371], device='cuda:1'), covar=tensor([0.1739, 0.2557, 0.1408, 0.1732], device='cuda:1'), in_proj_covar=tensor([0.0889, 0.0696, 0.0936, 0.0833], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 23:56:55,661 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=937391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:57:02,052 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 25050, giga_loss[loss=0.2979, simple_loss=0.3765, pruned_loss=0.1096, over 28777.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3696, pruned_loss=0.1183, over 5686056.03 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3587, pruned_loss=0.1092, over 5733095.77 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3708, pruned_loss=0.119, over 5675482.92 frames. ], batch size: 243, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:57:25,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:57:48,881 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 25100, giga_loss[loss=0.3349, simple_loss=0.3678, pruned_loss=0.151, over 23653.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3697, pruned_loss=0.1193, over 5671049.27 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3592, pruned_loss=0.1096, over 5736290.34 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3706, pruned_loss=0.1198, over 5657919.70 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:58:16,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-10 23:58:31,599 INFO [optim.py:369] (1/2) 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 23:58:33,138 INFO [train.py:968] (1/2) Epoch 21, batch 25150, libri_loss[loss=0.3651, simple_loss=0.4174, pruned_loss=0.1564, over 29203.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3699, pruned_loss=0.12, over 5660736.64 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3598, pruned_loss=0.1102, over 5721080.36 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3704, pruned_loss=0.1201, over 5662079.87 frames. ], batch size: 97, lr: 1.52e-03, grad_scale: 4.0
+2023-03-10 23:58:34,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5659, 1.6209, 1.7622, 1.3540], device='cuda:1'), covar=tensor([0.1705, 0.2442, 0.1387, 0.1584], device='cuda:1'), in_proj_covar=tensor([0.0890, 0.0697, 0.0937, 0.0834], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-10 23:59:16,787 INFO [train.py:968] (1/2) Epoch 21, batch 25200, giga_loss[loss=0.285, simple_loss=0.3566, pruned_loss=0.1067, over 28970.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3681, pruned_loss=0.1195, over 5666347.36 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3597, pruned_loss=0.1104, over 5724868.73 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3689, pruned_loss=0.1197, over 5662310.01 frames. ], batch size: 199, lr: 1.52e-03, grad_scale: 8.0
+2023-03-10 23:59:34,180 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-10 23:59:34,669 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:59:36,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:59:37,260 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3103, 1.6034, 1.2835, 0.9164], device='cuda:1'), covar=tensor([0.2707, 0.2776, 0.3151, 0.2424], device='cuda:1'), in_proj_covar=tensor([0.1497, 0.1086, 0.1322, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-10 23:59:57,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-10 23:59:59,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:00:01,701 INFO [optim.py:369] (1/2) 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 00:00:03,294 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:00:04,539 INFO [train.py:968] (1/2) Epoch 21, batch 25250, giga_loss[loss=0.2799, simple_loss=0.35, pruned_loss=0.1049, over 28972.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.366, pruned_loss=0.1183, over 5671513.07 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3598, pruned_loss=0.1105, over 5726360.41 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3665, pruned_loss=0.1184, over 5666463.23 frames. ], batch size: 106, lr: 1.52e-03, grad_scale: 8.0
+2023-03-11 00:00:28,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4206, 2.0733, 1.5414, 0.6109], device='cuda:1'), covar=tensor([0.5564, 0.2767, 0.3836, 0.6376], device='cuda:1'), in_proj_covar=tensor([0.1740, 0.1638, 0.1592, 0.1408], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 00:00:31,472 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:00:53,650 INFO [train.py:968] (1/2) Epoch 21, batch 25300, giga_loss[loss=0.3089, simple_loss=0.3752, pruned_loss=0.1213, over 28617.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3661, pruned_loss=0.1188, over 5667410.22 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3596, pruned_loss=0.1104, over 5729134.21 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3667, pruned_loss=0.1191, over 5659996.31 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:01:38,298 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 25350, giga_loss[loss=0.2851, simple_loss=0.3637, pruned_loss=0.1033, over 28899.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.367, pruned_loss=0.1191, over 5668318.93 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3588, pruned_loss=0.11, over 5733771.75 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3684, pruned_loss=0.12, over 5656393.80 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:01:55,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=937721.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:02:09,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1146, 2.3217, 1.6704, 1.8916], device='cuda:1'), covar=tensor([0.1042, 0.0780, 0.1123, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0447, 0.0515, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 00:02:23,782 INFO [train.py:968] (1/2) Epoch 21, batch 25400, giga_loss[loss=0.3716, simple_loss=0.4061, pruned_loss=0.1686, over 26633.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1184, over 5661750.81 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3587, pruned_loss=0.11, over 5722820.76 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3687, pruned_loss=0.1192, over 5659998.59 frames. ], batch size: 555, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:02:35,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=937766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:03:05,936 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 25450, giga_loss[loss=0.2799, simple_loss=0.3542, pruned_loss=0.1028, over 28625.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3666, pruned_loss=0.1172, over 5657744.86 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3588, pruned_loss=0.11, over 5718963.30 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3679, pruned_loss=0.1181, over 5658456.45 frames. ], batch size: 307, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:03:53,054 INFO [train.py:968] (1/2) Epoch 21, batch 25500, giga_loss[loss=0.2783, simple_loss=0.3544, pruned_loss=0.1011, over 28775.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3681, pruned_loss=0.1188, over 5659684.34 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3587, pruned_loss=0.11, over 5721144.33 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1196, over 5657075.64 frames. ], batch size: 119, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:04:16,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 00:04:42,225 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 25550, giga_loss[loss=0.31, simple_loss=0.3683, pruned_loss=0.1258, over 28467.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3716, pruned_loss=0.1225, over 5654222.53 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3584, pruned_loss=0.1098, over 5724200.26 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3731, pruned_loss=0.1234, over 5648899.46 frames. ], batch size: 65, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:04:50,539 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=937909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:04:52,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=937912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:05:23,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=937941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:05:30,697 INFO [train.py:968] (1/2) Epoch 21, batch 25600, giga_loss[loss=0.3241, simple_loss=0.3871, pruned_loss=0.1305, over 28898.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3725, pruned_loss=0.1242, over 5662089.63 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3585, pruned_loss=0.11, over 5726295.57 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3737, pruned_loss=0.1251, over 5654971.43 frames. ], batch size: 213, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:05:56,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5285, 1.6353, 1.7261, 1.3159], device='cuda:1'), covar=tensor([0.1665, 0.2403, 0.1375, 0.1645], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0700, 0.0938, 0.0835], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 00:06:20,793 INFO [optim.py:369] (1/2) 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 00:06:20,805 INFO [train.py:968] (1/2) Epoch 21, batch 25650, libri_loss[loss=0.2859, simple_loss=0.3637, pruned_loss=0.104, over 29647.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.373, pruned_loss=0.1254, over 5666191.56 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3585, pruned_loss=0.11, over 5732274.51 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3745, pruned_loss=0.1267, over 5652742.02 frames. ], batch size: 91, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:06:36,512 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-11 00:07:07,003 INFO [train.py:968] (1/2) Epoch 21, batch 25700, giga_loss[loss=0.3348, simple_loss=0.3899, pruned_loss=0.1399, over 29037.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3738, pruned_loss=0.1262, over 5664353.66 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3586, pruned_loss=0.11, over 5735393.58 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3752, pruned_loss=0.1274, over 5649582.17 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:07:23,841 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=938071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:07:46,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=938096.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:07:51,875 INFO [train.py:968] (1/2) Epoch 21, batch 25750, giga_loss[loss=0.3471, simple_loss=0.3894, pruned_loss=0.1524, over 23922.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3731, pruned_loss=0.1256, over 5667862.62 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3589, pruned_loss=0.1102, over 5738786.87 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3745, pruned_loss=0.127, over 5650549.40 frames. ], batch size: 705, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:07:52,487 INFO [optim.py:369] (1/2) 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 00:08:33,515 INFO [train.py:968] (1/2) Epoch 21, batch 25800, giga_loss[loss=0.3053, simple_loss=0.3771, pruned_loss=0.1167, over 29033.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.373, pruned_loss=0.1248, over 5659603.18 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3588, pruned_loss=0.1105, over 5724326.11 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3745, pruned_loss=0.1261, over 5656586.12 frames. ], batch size: 155, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:09:19,655 INFO [train.py:968] (1/2) Epoch 21, batch 25850, giga_loss[loss=0.3043, simple_loss=0.3746, pruned_loss=0.117, over 28840.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3711, pruned_loss=0.1222, over 5663238.93 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.359, pruned_loss=0.1104, over 5726512.19 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1235, over 5657991.18 frames. ], batch size: 112, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:09:21,908 INFO [optim.py:369] (1/2) 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 00:09:58,280 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=938239.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:10:03,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=938242.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:10:10,281 INFO [train.py:968] (1/2) Epoch 21, batch 25900, giga_loss[loss=0.3269, simple_loss=0.3819, pruned_loss=0.136, over 27934.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1205, over 5663292.06 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3592, pruned_loss=0.1107, over 5730338.71 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1215, over 5654220.26 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:10:27,968 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=938271.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:10:56,606 INFO [train.py:968] (1/2) Epoch 21, batch 25950, giga_loss[loss=0.2646, simple_loss=0.3366, pruned_loss=0.09627, over 28946.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3666, pruned_loss=0.1197, over 5673575.72 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3593, pruned_loss=0.1107, over 5731245.64 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3674, pruned_loss=0.1206, over 5665091.23 frames. ], batch size: 164, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:10:58,375 INFO [optim.py:369] (1/2) 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 00:11:13,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1452, 1.3072, 1.1004, 0.9186], device='cuda:1'), covar=tensor([0.1016, 0.0537, 0.1108, 0.1099], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0446, 0.0513, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 00:11:40,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-11 00:11:48,101 INFO [train.py:968] (1/2) Epoch 21, batch 26000, giga_loss[loss=0.2818, simple_loss=0.3476, pruned_loss=0.108, over 28720.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3666, pruned_loss=0.1199, over 5669442.50 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3596, pruned_loss=0.1111, over 5724000.41 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3671, pruned_loss=0.1205, over 5667870.79 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:12:15,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6834, 1.8303, 1.2991, 1.4129], device='cuda:1'), covar=tensor([0.0870, 0.0551, 0.0984, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0446, 0.0513, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 00:12:32,707 INFO [train.py:968] (1/2) Epoch 21, batch 26050, giga_loss[loss=0.2943, simple_loss=0.365, pruned_loss=0.1117, over 29021.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1208, over 5679048.74 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3595, pruned_loss=0.1112, over 5726661.89 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1214, over 5673477.80 frames. ], batch size: 128, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:12:34,228 INFO [optim.py:369] (1/2) 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 00:13:10,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6200, 1.8792, 1.7834, 1.4876], device='cuda:1'), covar=tensor([0.3301, 0.2479, 0.2216, 0.2816], device='cuda:1'), in_proj_covar=tensor([0.1982, 0.1900, 0.1842, 0.1973], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 00:13:12,105 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 26100, giga_loss[loss=0.3047, simple_loss=0.3813, pruned_loss=0.114, over 28313.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3717, pruned_loss=0.1198, over 5689111.53 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.36, pruned_loss=0.1118, over 5734130.50 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3724, pruned_loss=0.1201, over 5675427.82 frames. ], batch size: 368, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:14:05,682 INFO [train.py:968] (1/2) Epoch 21, batch 26150, libri_loss[loss=0.2652, simple_loss=0.3364, pruned_loss=0.09698, over 29570.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3729, pruned_loss=0.1193, over 5691521.65 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3596, pruned_loss=0.1116, over 5736545.01 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.374, pruned_loss=0.1199, over 5677881.16 frames. ], batch size: 78, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:14:07,389 INFO [optim.py:369] (1/2) 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 00:14:19,887 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=938519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:14:29,350 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=938527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:14:53,172 INFO [train.py:968] (1/2) Epoch 21, batch 26200, giga_loss[loss=0.3338, simple_loss=0.3945, pruned_loss=0.1366, over 27847.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3748, pruned_loss=0.1211, over 5691746.61 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3599, pruned_loss=0.1118, over 5740767.59 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3757, pruned_loss=0.1216, over 5676205.50 frames. ], batch size: 412, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:15:28,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=938589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:15:30,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=938592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:15:36,995 INFO [train.py:968] (1/2) Epoch 21, batch 26250, giga_loss[loss=0.3506, simple_loss=0.4027, pruned_loss=0.1493, over 28622.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3764, pruned_loss=0.1225, over 5687040.48 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.36, pruned_loss=0.1119, over 5732201.95 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3772, pruned_loss=0.1229, over 5682827.05 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:15:38,330 INFO [optim.py:369] (1/2) 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 00:15:58,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=938621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:15:58,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6087, 1.8310, 1.6871, 1.6235], device='cuda:1'), covar=tensor([0.2046, 0.2275, 0.2523, 0.2222], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0751, 0.0716, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 00:16:25,590 INFO [train.py:968] (1/2) Epoch 21, batch 26300, giga_loss[loss=0.3576, simple_loss=0.4166, pruned_loss=0.1493, over 28630.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.378, pruned_loss=0.1252, over 5671514.71 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3609, pruned_loss=0.1126, over 5723030.92 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3783, pruned_loss=0.1251, over 5673729.58 frames. ], batch size: 262, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:17:09,542 INFO [train.py:968] (1/2) Epoch 21, batch 26350, giga_loss[loss=0.284, simple_loss=0.3437, pruned_loss=0.1121, over 28640.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3766, pruned_loss=0.1248, over 5671211.07 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.361, pruned_loss=0.1129, over 5711643.90 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3774, pruned_loss=0.1251, over 5681338.36 frames. ], batch size: 92, lr: 1.52e-03, grad_scale: 2.0
+2023-03-11 00:17:10,823 INFO [optim.py:369] (1/2) 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 00:17:47,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5722, 1.6385, 1.7886, 1.3654], device='cuda:1'), covar=tensor([0.1777, 0.2598, 0.1439, 0.1681], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0705, 0.0945, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 00:17:56,336 INFO [train.py:968] (1/2) Epoch 21, batch 26400, giga_loss[loss=0.3121, simple_loss=0.3675, pruned_loss=0.1284, over 28388.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3747, pruned_loss=0.1243, over 5674807.80 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3611, pruned_loss=0.1131, over 5714990.60 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3755, pruned_loss=0.1245, over 5679213.06 frames. ], batch size: 71, lr: 1.52e-03, grad_scale: 4.0
+2023-03-11 00:18:44,023 INFO [train.py:968] (1/2) Epoch 21, batch 26450, giga_loss[loss=0.3253, simple_loss=0.3893, pruned_loss=0.1307, over 29034.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3728, pruned_loss=0.1233, over 5676309.70 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3612, pruned_loss=0.1132, over 5708668.64 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3739, pruned_loss=0.1237, over 5684722.17 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:18:46,338 INFO [optim.py:369] (1/2) 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 00:18:49,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.04 vs. limit=2.0
+2023-03-11 00:19:28,962 INFO [train.py:968] (1/2) Epoch 21, batch 26500, libri_loss[loss=0.2985, simple_loss=0.3727, pruned_loss=0.1121, over 29204.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3735, pruned_loss=0.1243, over 5670165.69 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3614, pruned_loss=0.1135, over 5710831.40 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3745, pruned_loss=0.1247, over 5674103.90 frames. ], batch size: 97, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:20:06,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=938894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:20:11,299 INFO [train.py:968] (1/2) Epoch 21, batch 26550, libri_loss[loss=0.2769, simple_loss=0.357, pruned_loss=0.09835, over 29647.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3725, pruned_loss=0.1236, over 5678687.09 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3613, pruned_loss=0.1133, over 5716430.88 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3737, pruned_loss=0.1244, over 5676039.05 frames. ], batch size: 91, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:20:12,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=938902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:20:13,101 INFO [optim.py:369] (1/2) 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 00:20:57,256 INFO [train.py:968] (1/2) Epoch 21, batch 26600, giga_loss[loss=0.3174, simple_loss=0.3591, pruned_loss=0.1379, over 23697.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3699, pruned_loss=0.123, over 5656615.93 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3612, pruned_loss=0.1133, over 5714773.85 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3712, pruned_loss=0.1238, over 5655213.72 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:21:41,280 INFO [train.py:968] (1/2) Epoch 21, batch 26650, giga_loss[loss=0.2984, simple_loss=0.3673, pruned_loss=0.1147, over 28705.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.37, pruned_loss=0.1232, over 5653109.66 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3614, pruned_loss=0.1136, over 5708599.44 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.371, pruned_loss=0.1239, over 5655172.96 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:21:43,058 INFO [optim.py:369] (1/2) 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 00:22:10,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3658, 2.0456, 1.5333, 0.5872], device='cuda:1'), covar=tensor([0.5761, 0.2776, 0.3827, 0.6517], device='cuda:1'), in_proj_covar=tensor([0.1755, 0.1652, 0.1600, 0.1420], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 00:22:13,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:19,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:23,310 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:26,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:28,012 INFO [train.py:968] (1/2) Epoch 21, batch 26700, giga_loss[loss=0.2839, simple_loss=0.3657, pruned_loss=0.101, over 28947.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3695, pruned_loss=0.1214, over 5664722.73 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3616, pruned_loss=0.1138, over 5711667.53 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3703, pruned_loss=0.122, over 5662592.76 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:22:45,701 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=939069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:22:47,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5884, 1.7860, 1.5173, 1.5011], device='cuda:1'), covar=tensor([0.2442, 0.2272, 0.2391, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1491, 0.1083, 0.1318, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 00:22:52,309 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=939077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:23:17,176 INFO [train.py:968] (1/2) Epoch 21, batch 26750, giga_loss[loss=0.2874, simple_loss=0.3555, pruned_loss=0.1097, over 28977.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1235, over 5657021.93 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3619, pruned_loss=0.1139, over 5715022.46 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.373, pruned_loss=0.124, over 5651394.65 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:23:18,435 INFO [optim.py:369] (1/2) 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 00:23:58,340 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-11 00:24:03,449 INFO [train.py:968] (1/2) Epoch 21, batch 26800, giga_loss[loss=0.3463, simple_loss=0.4201, pruned_loss=0.1362, over 28905.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3711, pruned_loss=0.1224, over 5665229.96 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3616, pruned_loss=0.1138, over 5714476.97 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5660058.98 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:24:18,482 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=939166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:24:18,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6885, 1.8868, 1.7735, 1.6487], device='cuda:1'), covar=tensor([0.2082, 0.2617, 0.2428, 0.2685], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0751, 0.0716, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 00:24:46,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2865, 2.5554, 2.4240, 2.2736], device='cuda:1'), covar=tensor([0.2037, 0.2250, 0.2092, 0.2330], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0750, 0.0714, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 00:24:47,777 INFO [train.py:968] (1/2) Epoch 21, batch 26850, libri_loss[loss=0.3085, simple_loss=0.3647, pruned_loss=0.1262, over 29577.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3719, pruned_loss=0.1203, over 5669277.49 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5714729.16 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3727, pruned_loss=0.121, over 5663771.34 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:24:49,116 INFO [optim.py:369] (1/2) 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 00:24:57,078 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 00:25:09,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-11 00:25:17,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4922, 1.5930, 1.7408, 1.2792], device='cuda:1'), covar=tensor([0.1938, 0.2828, 0.1700, 0.2002], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0701, 0.0941, 0.0836], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 00:25:34,177 INFO [train.py:968] (1/2) Epoch 21, batch 26900, giga_loss[loss=0.2891, simple_loss=0.3745, pruned_loss=0.1019, over 28917.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3738, pruned_loss=0.1194, over 5680828.80 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3616, pruned_loss=0.1138, over 5716812.73 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3748, pruned_loss=0.12, over 5674363.78 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:26:16,272 INFO [train.py:968] (1/2) Epoch 21, batch 26950, giga_loss[loss=0.3259, simple_loss=0.3939, pruned_loss=0.1289, over 29056.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3762, pruned_loss=0.1204, over 5687714.81 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3616, pruned_loss=0.1138, over 5714627.65 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3772, pruned_loss=0.121, over 5683591.25 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:26:20,376 INFO [optim.py:369] (1/2) 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 00:26:56,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8873, 1.1271, 2.8724, 2.7775], device='cuda:1'), covar=tensor([0.1745, 0.2623, 0.0649, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0652, 0.0971, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 00:27:07,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6025, 1.8328, 1.5516, 1.5720], device='cuda:1'), covar=tensor([0.2002, 0.2088, 0.2164, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.1490, 0.1082, 0.1317, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 00:27:07,944 INFO [train.py:968] (1/2) Epoch 21, batch 27000, giga_loss[loss=0.4298, simple_loss=0.447, pruned_loss=0.2063, over 26564.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3807, pruned_loss=0.1252, over 5665347.41 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3619, pruned_loss=0.114, over 5703567.39 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3813, pruned_loss=0.1255, over 5671186.96 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:27:07,945 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 00:27:16,795 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 00:28:02,629 INFO [train.py:968] (1/2) Epoch 21, batch 27050, giga_loss[loss=0.3313, simple_loss=0.3715, pruned_loss=0.1456, over 23480.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3814, pruned_loss=0.1274, over 5646248.37 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3617, pruned_loss=0.1141, over 5699646.91 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3828, pruned_loss=0.128, over 5652544.28 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:28:06,655 INFO [optim.py:369] (1/2) 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 00:28:23,309 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 00:28:51,699 INFO [train.py:968] (1/2) Epoch 21, batch 27100, libri_loss[loss=0.2862, simple_loss=0.3578, pruned_loss=0.1073, over 28606.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.38, pruned_loss=0.1269, over 5653871.51 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3619, pruned_loss=0.1143, over 5701390.78 frames. ], giga_tot_loss[loss=0.3182, simple_loss=0.3813, pruned_loss=0.1275, over 5656272.90 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 1.0
+2023-03-11 00:29:15,283 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 27150, giga_loss[loss=0.2978, simple_loss=0.3706, pruned_loss=0.1125, over 28714.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3801, pruned_loss=0.127, over 5639228.44 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3627, pruned_loss=0.1149, over 5692916.35 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3809, pruned_loss=0.1273, over 5646738.48 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 1.0
+2023-03-11 00:29:40,673 INFO [optim.py:369] (1/2) 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 00:30:13,502 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 27200, giga_loss[loss=0.2906, simple_loss=0.3664, pruned_loss=0.1074, over 28694.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3797, pruned_loss=0.1253, over 5650818.21 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3628, pruned_loss=0.115, over 5695263.70 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3805, pruned_loss=0.1256, over 5654181.73 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:31:08,043 INFO [train.py:968] (1/2) Epoch 21, batch 27250, giga_loss[loss=0.3145, simple_loss=0.3815, pruned_loss=0.1237, over 28925.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3779, pruned_loss=0.1223, over 5666079.71 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5697316.78 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3792, pruned_loss=0.123, over 5666002.09 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:31:12,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7964, 1.8751, 1.4915, 1.5272], device='cuda:1'), covar=tensor([0.0978, 0.0713, 0.0981, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0446, 0.0514, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 00:31:14,259 INFO [optim.py:369] (1/2) 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 00:31:43,534 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=939633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:31:59,393 INFO [train.py:968] (1/2) Epoch 21, batch 27300, giga_loss[loss=0.2947, simple_loss=0.3695, pruned_loss=0.11, over 28168.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.378, pruned_loss=0.1231, over 5663497.59 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3618, pruned_loss=0.1144, over 5700283.92 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3799, pruned_loss=0.1241, over 5660048.39 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:32:32,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:32:35,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:32:46,412 INFO [train.py:968] (1/2) Epoch 21, batch 27350, libri_loss[loss=0.2505, simple_loss=0.3219, pruned_loss=0.08953, over 29666.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3766, pruned_loss=0.1227, over 5665191.34 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1145, over 5700893.24 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3785, pruned_loss=0.1235, over 5661397.10 frames. ], batch size: 73, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:32:49,729 INFO [optim.py:369] (1/2) 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 00:32:53,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3508, 3.1532, 2.9830, 1.5527], device='cuda:1'), covar=tensor([0.0925, 0.1082, 0.0987, 0.2040], device='cuda:1'), in_proj_covar=tensor([0.1238, 0.1149, 0.0971, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 00:32:55,200 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-11 00:32:58,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=939716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:33:34,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6702, 1.7183, 1.8490, 1.4237], device='cuda:1'), covar=tensor([0.1766, 0.2541, 0.1469, 0.1730], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0703, 0.0943, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 00:33:35,389 INFO [train.py:968] (1/2) Epoch 21, batch 27400, giga_loss[loss=0.282, simple_loss=0.3564, pruned_loss=0.1038, over 28564.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3737, pruned_loss=0.1218, over 5660142.67 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3619, pruned_loss=0.1146, over 5704110.92 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3753, pruned_loss=0.1226, over 5654007.86 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:34:25,639 INFO [train.py:968] (1/2) Epoch 21, batch 27450, giga_loss[loss=0.4011, simple_loss=0.4359, pruned_loss=0.1832, over 27965.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.373, pruned_loss=0.1226, over 5645433.24 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.362, pruned_loss=0.1146, over 5703676.76 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3743, pruned_loss=0.1233, over 5640415.39 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:34:33,412 INFO [optim.py:369] (1/2) 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 00:35:16,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=939850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:35:17,029 INFO [train.py:968] (1/2) Epoch 21, batch 27500, giga_loss[loss=0.3006, simple_loss=0.3673, pruned_loss=0.1169, over 28807.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.122, over 5656301.73 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3625, pruned_loss=0.1148, over 5707501.97 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3725, pruned_loss=0.1225, over 5647537.30 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:35:25,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-11 00:36:01,995 INFO [train.py:968] (1/2) Epoch 21, batch 27550, giga_loss[loss=0.2833, simple_loss=0.355, pruned_loss=0.1058, over 28723.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5652671.85 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5710695.64 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3717, pruned_loss=0.123, over 5642312.90 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:36:06,820 INFO [optim.py:369] (1/2) 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 00:36:30,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7442, 1.9083, 1.2705, 1.4904], device='cuda:1'), covar=tensor([0.0942, 0.0630, 0.1088, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0447, 0.0516, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 00:36:42,073 INFO [train.py:968] (1/2) Epoch 21, batch 27600, giga_loss[loss=0.2683, simple_loss=0.3466, pruned_loss=0.09499, over 28894.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3703, pruned_loss=0.1223, over 5653801.86 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3625, pruned_loss=0.1149, over 5709105.34 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1229, over 5645210.35 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:36:42,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-11 00:36:51,975 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=939962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:37:15,540 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-11 00:37:21,140 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=939993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:37:24,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=939996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:37:28,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5756, 4.4372, 4.2255, 1.9689], device='cuda:1'), covar=tensor([0.0570, 0.0656, 0.0713, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.1237, 0.1149, 0.0971, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 00:37:29,345 INFO [train.py:968] (1/2) Epoch 21, batch 27650, giga_loss[loss=0.247, simple_loss=0.3288, pruned_loss=0.08257, over 28785.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3667, pruned_loss=0.1177, over 5663025.19 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5710086.81 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3674, pruned_loss=0.1182, over 5655261.83 frames. ], batch size: 60, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:37:33,130 INFO [optim.py:369] (1/2) 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 00:37:34,811 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:38:15,514 INFO [train.py:968] (1/2) Epoch 21, batch 27700, giga_loss[loss=0.3308, simple_loss=0.3938, pruned_loss=0.1339, over 28245.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3646, pruned_loss=0.1154, over 5669551.34 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3627, pruned_loss=0.1151, over 5712778.74 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3652, pruned_loss=0.1157, over 5660069.42 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:39:06,992 INFO [train.py:968] (1/2) Epoch 21, batch 27750, giga_loss[loss=0.2577, simple_loss=0.3319, pruned_loss=0.09174, over 28595.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3647, pruned_loss=0.1161, over 5648473.85 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5709763.05 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3653, pruned_loss=0.1165, over 5642320.47 frames. ], batch size: 78, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:39:12,448 INFO [optim.py:369] (1/2) 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 00:39:50,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:39:58,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5457, 1.4683, 3.9875, 3.4191], device='cuda:1'), covar=tensor([0.1532, 0.2542, 0.0452, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0649, 0.0966, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 00:39:58,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4913, 2.1685, 1.7314, 0.7547], device='cuda:1'), covar=tensor([0.6291, 0.2921, 0.3666, 0.6700], device='cuda:1'), in_proj_covar=tensor([0.1735, 0.1634, 0.1588, 0.1408], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 00:39:59,214 INFO [train.py:968] (1/2) Epoch 21, batch 27800, giga_loss[loss=0.2788, simple_loss=0.3556, pruned_loss=0.101, over 28855.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3607, pruned_loss=0.114, over 5667678.96 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3627, pruned_loss=0.1149, over 5714427.51 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1143, over 5657385.88 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:39:59,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:40:04,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:40:07,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3907, 1.2298, 4.2570, 3.4402], device='cuda:1'), covar=tensor([0.1666, 0.2803, 0.0412, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0649, 0.0966, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 00:40:34,567 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:40:52,419 INFO [train.py:968] (1/2) Epoch 21, batch 27850, giga_loss[loss=0.2594, simple_loss=0.3294, pruned_loss=0.09472, over 28939.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3614, pruned_loss=0.1152, over 5667307.20 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3627, pruned_loss=0.115, over 5717420.57 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3617, pruned_loss=0.1154, over 5655898.72 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:40:56,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 00:40:57,793 INFO [optim.py:369] (1/2) 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 00:41:10,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4682, 1.8066, 1.4058, 1.4915], device='cuda:1'), covar=tensor([0.2728, 0.2725, 0.3175, 0.2482], device='cuda:1'), in_proj_covar=tensor([0.1497, 0.1086, 0.1321, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 00:41:38,451 INFO [train.py:968] (1/2) Epoch 21, batch 27900, giga_loss[loss=0.3966, simple_loss=0.4281, pruned_loss=0.1825, over 27442.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3641, pruned_loss=0.1165, over 5662411.70 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3625, pruned_loss=0.1148, over 5721466.06 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3645, pruned_loss=0.1168, over 5648564.38 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:42:23,141 INFO [train.py:968] (1/2) Epoch 21, batch 27950, giga_loss[loss=0.3987, simple_loss=0.4288, pruned_loss=0.1843, over 26669.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3666, pruned_loss=0.1181, over 5660691.74 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3626, pruned_loss=0.115, over 5722419.61 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.367, pruned_loss=0.1183, over 5646399.75 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:42:25,255 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:42:28,745 INFO [optim.py:369] (1/2) 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 00:42:46,741 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:42:57,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:43:09,714 INFO [train.py:968] (1/2) Epoch 21, batch 28000, giga_loss[loss=0.3, simple_loss=0.3695, pruned_loss=0.1152, over 29015.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3671, pruned_loss=0.1182, over 5664488.19 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3628, pruned_loss=0.115, over 5725724.02 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3673, pruned_loss=0.1184, over 5649104.81 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:43:53,484 INFO [train.py:968] (1/2) Epoch 21, batch 28050, giga_loss[loss=0.276, simple_loss=0.3521, pruned_loss=0.09995, over 29097.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1184, over 5660231.08 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3636, pruned_loss=0.1153, over 5725315.81 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3668, pruned_loss=0.1184, over 5646084.67 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:43:58,101 INFO [optim.py:369] (1/2) 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 00:44:01,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3836, 3.2118, 3.0538, 1.3666], device='cuda:1'), covar=tensor([0.1033, 0.1165, 0.1215, 0.2231], device='cuda:1'), in_proj_covar=tensor([0.1238, 0.1149, 0.0973, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 00:44:20,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:44:35,017 INFO [train.py:968] (1/2) Epoch 21, batch 28100, giga_loss[loss=0.3171, simple_loss=0.3761, pruned_loss=0.1291, over 28483.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3698, pruned_loss=0.1201, over 5663611.53 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3637, pruned_loss=0.1154, over 5719898.82 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3694, pruned_loss=0.1201, over 5654893.36 frames. ], batch size: 60, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:45:01,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:05,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:15,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940494.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:45:21,545 INFO [train.py:968] (1/2) Epoch 21, batch 28150, giga_loss[loss=0.266, simple_loss=0.3463, pruned_loss=0.09284, over 28939.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3721, pruned_loss=0.1215, over 5665172.96 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.364, pruned_loss=0.1158, over 5720438.86 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3717, pruned_loss=0.1214, over 5656379.16 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:45:27,419 INFO [optim.py:369] (1/2) 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 00:45:31,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:39,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:45:48,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5042, 2.7917, 1.6907, 1.5800], device='cuda:1'), covar=tensor([0.0769, 0.0333, 0.0664, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0557, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 00:46:08,616 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-11 00:46:11,662 INFO [train.py:968] (1/2) Epoch 21, batch 28200, giga_loss[loss=0.3382, simple_loss=0.3969, pruned_loss=0.1397, over 28928.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3729, pruned_loss=0.1222, over 5662051.77 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.364, pruned_loss=0.1156, over 5722516.77 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3727, pruned_loss=0.1223, over 5652870.41 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:46:44,325 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 28250, giga_loss[loss=0.3054, simple_loss=0.372, pruned_loss=0.1194, over 28993.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.1239, over 5644968.48 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3644, pruned_loss=0.1162, over 5708372.42 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3737, pruned_loss=0.1236, over 5649310.16 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:47:03,456 INFO [optim.py:369] (1/2) 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 00:47:47,925 INFO [train.py:968] (1/2) Epoch 21, batch 28300, giga_loss[loss=0.2947, simple_loss=0.3729, pruned_loss=0.1082, over 28883.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3767, pruned_loss=0.1243, over 5648868.11 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3646, pruned_loss=0.1162, over 5710892.92 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3764, pruned_loss=0.1242, over 5649215.55 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:47:55,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4954, 1.6614, 1.7185, 1.2994], device='cuda:1'), covar=tensor([0.1838, 0.2617, 0.1545, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0705, 0.0947, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 00:48:03,036 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:05,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:15,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:33,269 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:33,284 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:35,027 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1649, 1.2568, 1.1234, 0.8574], device='cuda:1'), covar=tensor([0.1032, 0.0560, 0.1109, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0449, 0.0517, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 00:48:39,215 INFO [train.py:968] (1/2) Epoch 21, batch 28350, giga_loss[loss=0.311, simple_loss=0.3803, pruned_loss=0.1208, over 28723.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3758, pruned_loss=0.123, over 5660666.51 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3639, pruned_loss=0.1158, over 5713753.50 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3764, pruned_loss=0.1233, over 5657640.97 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 00:48:40,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:48:45,240 INFO [optim.py:369] (1/2) 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 00:49:30,963 INFO [train.py:968] (1/2) Epoch 21, batch 28400, giga_loss[loss=0.2937, simple_loss=0.3574, pruned_loss=0.115, over 28784.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3741, pruned_loss=0.1229, over 5666666.22 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1161, over 5715465.81 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3744, pruned_loss=0.123, over 5662108.20 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:49:38,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7719, 2.0010, 1.6390, 2.1258], device='cuda:1'), covar=tensor([0.2527, 0.2689, 0.3008, 0.2402], device='cuda:1'), in_proj_covar=tensor([0.1501, 0.1088, 0.1325, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 00:50:22,778 INFO [train.py:968] (1/2) Epoch 21, batch 28450, giga_loss[loss=0.3083, simple_loss=0.372, pruned_loss=0.1223, over 28899.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.123, over 5671921.04 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3647, pruned_loss=0.1165, over 5719256.69 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3734, pruned_loss=0.1229, over 5663108.99 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:50:30,554 INFO [optim.py:369] (1/2) 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 00:50:33,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:50:46,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:50:49,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:15,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:16,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:18,511 INFO [train.py:968] (1/2) Epoch 21, batch 28500, giga_loss[loss=0.3195, simple_loss=0.371, pruned_loss=0.134, over 28895.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3713, pruned_loss=0.1219, over 5677986.39 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3652, pruned_loss=0.117, over 5721841.68 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.371, pruned_loss=0.1215, over 5668155.50 frames. ], batch size: 112, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:51:19,988 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:34,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940869.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:51:43,438 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:51:51,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-11 00:52:02,947 INFO [train.py:968] (1/2) Epoch 21, batch 28550, giga_loss[loss=0.3038, simple_loss=0.3675, pruned_loss=0.1201, over 28734.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3705, pruned_loss=0.1215, over 5678204.85 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3651, pruned_loss=0.1169, over 5715864.88 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3705, pruned_loss=0.1214, over 5674229.61 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:52:12,204 INFO [optim.py:369] (1/2) 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 00:52:50,412 INFO [train.py:968] (1/2) Epoch 21, batch 28600, giga_loss[loss=0.2847, simple_loss=0.3494, pruned_loss=0.11, over 28834.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3702, pruned_loss=0.1221, over 5661772.58 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.365, pruned_loss=0.1168, over 5719962.97 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3704, pruned_loss=0.1222, over 5653541.40 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:52:52,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=940953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:52:52,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4885, 1.7706, 1.4238, 1.3596], device='cuda:1'), covar=tensor([0.2733, 0.2688, 0.3055, 0.2418], device='cuda:1'), in_proj_covar=tensor([0.1498, 0.1087, 0.1323, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 00:52:54,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=940956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:02,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=940964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:09,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4685, 2.9510, 1.4785, 1.5942], device='cuda:1'), covar=tensor([0.0858, 0.0428, 0.0846, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0557, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 00:53:19,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=940985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:26,766 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=940993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:53:32,140 INFO [train.py:968] (1/2) Epoch 21, batch 28650, giga_loss[loss=0.3467, simple_loss=0.4036, pruned_loss=0.1449, over 28529.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3699, pruned_loss=0.1218, over 5668353.62 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3647, pruned_loss=0.1164, over 5723447.11 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5655952.97 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:53:38,707 INFO [optim.py:369] (1/2) 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 00:53:40,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3037, 1.6163, 1.5232, 1.4219], device='cuda:1'), covar=tensor([0.0781, 0.0321, 0.0307, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 00:53:42,816 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941012.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:53:45,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 00:54:13,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941044.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 00:54:17,914 INFO [train.py:968] (1/2) Epoch 21, batch 28700, giga_loss[loss=0.3702, simple_loss=0.4072, pruned_loss=0.1666, over 26725.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1239, over 5667537.21 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 5726154.16 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3728, pruned_loss=0.1247, over 5653083.55 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:54:20,198 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:54:53,440 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:55:02,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-11 00:55:04,868 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-11 00:55:08,524 INFO [train.py:968] (1/2) Epoch 21, batch 28750, giga_loss[loss=0.3328, simple_loss=0.3903, pruned_loss=0.1376, over 28909.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3743, pruned_loss=0.1258, over 5656096.32 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 5726154.16 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.375, pruned_loss=0.1265, over 5644846.86 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:55:15,707 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:55:17,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:55:47,127 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:55:56,911 INFO [train.py:968] (1/2) Epoch 21, batch 28800, giga_loss[loss=0.2894, simple_loss=0.362, pruned_loss=0.1084, over 29022.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3737, pruned_loss=0.1262, over 5649913.69 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3642, pruned_loss=0.1161, over 5729155.46 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3747, pruned_loss=0.1271, over 5637414.07 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 00:56:22,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3604, 3.0440, 1.4281, 1.4835], device='cuda:1'), covar=tensor([0.0996, 0.0347, 0.0916, 0.1354], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0559, 0.0386, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 00:56:42,198 INFO [train.py:968] (1/2) Epoch 21, batch 28850, giga_loss[loss=0.2918, simple_loss=0.3587, pruned_loss=0.1125, over 27913.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.374, pruned_loss=0.1266, over 5649825.37 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3644, pruned_loss=0.1164, over 5720787.35 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3746, pruned_loss=0.1271, over 5647443.42 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:56:49,115 INFO [optim.py:369] (1/2) 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 00:56:53,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1840, 1.4795, 1.1644, 0.5676], device='cuda:1'), covar=tensor([0.2097, 0.1528, 0.2040, 0.4268], device='cuda:1'), in_proj_covar=tensor([0.1751, 0.1648, 0.1598, 0.1420], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 00:56:54,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:56:56,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:56:56,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:57:04,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3163, 1.7152, 1.2871, 0.7842], device='cuda:1'), covar=tensor([0.3896, 0.2189, 0.2282, 0.4891], device='cuda:1'), in_proj_covar=tensor([0.1751, 0.1648, 0.1599, 0.1420], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 00:57:11,318 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:57:26,056 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:57:29,974 INFO [train.py:968] (1/2) Epoch 21, batch 28900, giga_loss[loss=0.2888, simple_loss=0.3616, pruned_loss=0.108, over 28716.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3744, pruned_loss=0.1266, over 5632309.45 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3646, pruned_loss=0.1165, over 5712569.28 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3749, pruned_loss=0.127, over 5637553.21 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:57:35,099 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:58:09,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 00:58:14,801 INFO [train.py:968] (1/2) Epoch 21, batch 28950, giga_loss[loss=0.3173, simple_loss=0.3586, pruned_loss=0.138, over 23440.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3739, pruned_loss=0.1254, over 5631217.64 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3647, pruned_loss=0.1165, over 5705110.89 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3746, pruned_loss=0.1261, over 5639079.21 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:58:23,157 INFO [optim.py:369] (1/2) 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 00:59:02,002 INFO [train.py:968] (1/2) Epoch 21, batch 29000, giga_loss[loss=0.2604, simple_loss=0.3347, pruned_loss=0.09307, over 29022.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3751, pruned_loss=0.1257, over 5647971.82 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1167, over 5709114.89 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1262, over 5649236.58 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:59:05,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-11 00:59:15,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 00:59:44,841 INFO [train.py:968] (1/2) Epoch 21, batch 29050, giga_loss[loss=0.4539, simple_loss=0.4595, pruned_loss=0.2242, over 26421.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3761, pruned_loss=0.1267, over 5661113.04 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3648, pruned_loss=0.1165, over 5713468.90 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3769, pruned_loss=0.1276, over 5656775.14 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 00:59:50,932 INFO [optim.py:369] (1/2) 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 01:00:10,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:00:27,920 INFO [train.py:968] (1/2) Epoch 21, batch 29100, giga_loss[loss=0.2685, simple_loss=0.3445, pruned_loss=0.09619, over 28671.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3764, pruned_loss=0.1272, over 5666360.81 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1164, over 5709280.55 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3776, pruned_loss=0.1283, over 5666213.77 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:00:28,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2896, 1.5665, 1.2772, 1.5923], device='cuda:1'), covar=tensor([0.0789, 0.0333, 0.0358, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 01:00:39,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:00:41,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6126, 1.6958, 1.7975, 1.3954], device='cuda:1'), covar=tensor([0.1814, 0.2592, 0.1467, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0704, 0.0944, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 01:00:46,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5748, 1.7836, 1.3902, 1.7555], device='cuda:1'), covar=tensor([0.2779, 0.2795, 0.3206, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.1503, 0.1090, 0.1326, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 01:01:14,345 INFO [train.py:968] (1/2) Epoch 21, batch 29150, giga_loss[loss=0.283, simple_loss=0.3598, pruned_loss=0.1031, over 28542.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3775, pruned_loss=0.1276, over 5662719.91 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3648, pruned_loss=0.1166, over 5711757.66 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3783, pruned_loss=0.1284, over 5659733.27 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:01:24,023 INFO [optim.py:369] (1/2) 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 01:01:26,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:01:28,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:01:35,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-11 01:01:58,414 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:01,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 01:02:05,900 INFO [train.py:968] (1/2) Epoch 21, batch 29200, giga_loss[loss=0.2809, simple_loss=0.3619, pruned_loss=0.09997, over 28873.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3771, pruned_loss=0.1258, over 5663440.32 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.365, pruned_loss=0.1167, over 5713209.19 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3778, pruned_loss=0.1265, over 5659034.08 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:02:14,915 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:25,393 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:28,428 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:30,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:42,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:51,775 INFO [train.py:968] (1/2) Epoch 21, batch 29250, giga_loss[loss=0.2547, simple_loss=0.3254, pruned_loss=0.09195, over 28711.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3752, pruned_loss=0.1238, over 5666974.94 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3651, pruned_loss=0.1168, over 5715946.97 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3757, pruned_loss=0.1243, over 5660064.06 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:02:54,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:54,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:56,539 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:58,481 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:02:58,836 INFO [optim.py:369] (1/2) 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 01:03:19,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:03:24,557 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:03:34,209 INFO [train.py:968] (1/2) Epoch 21, batch 29300, giga_loss[loss=0.283, simple_loss=0.3545, pruned_loss=0.1057, over 28990.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3726, pruned_loss=0.1224, over 5661567.94 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3649, pruned_loss=0.1166, over 5713215.86 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3736, pruned_loss=0.1232, over 5656120.65 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:04:02,891 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:04:14,694 INFO [train.py:968] (1/2) Epoch 21, batch 29350, giga_loss[loss=0.3933, simple_loss=0.4388, pruned_loss=0.1739, over 28215.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.373, pruned_loss=0.1227, over 5655801.64 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.365, pruned_loss=0.1166, over 5710281.61 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.374, pruned_loss=0.1236, over 5652521.11 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:04:21,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6734, 1.7130, 1.8443, 1.4567], device='cuda:1'), covar=tensor([0.1740, 0.2560, 0.1397, 0.1699], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0707, 0.0947, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 01:04:25,786 INFO [optim.py:369] (1/2) 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 01:04:45,410 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:04:47,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:00,715 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:04,093 INFO [train.py:968] (1/2) Epoch 21, batch 29400, giga_loss[loss=0.374, simple_loss=0.3985, pruned_loss=0.1747, over 23409.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5651054.86 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3646, pruned_loss=0.1163, over 5709007.90 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3762, pruned_loss=0.1254, over 5648352.71 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:05:04,407 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:18,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:26,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=941774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:29,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=941777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:31,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:05:33,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3220, 2.9291, 1.4318, 1.4530], device='cuda:1'), covar=tensor([0.0983, 0.0376, 0.0885, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0559, 0.0385, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 01:05:51,433 INFO [train.py:968] (1/2) Epoch 21, batch 29450, giga_loss[loss=0.2935, simple_loss=0.3574, pruned_loss=0.1148, over 28887.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.374, pruned_loss=0.1241, over 5649559.29 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3645, pruned_loss=0.1163, over 5699501.57 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3753, pruned_loss=0.125, over 5654801.56 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:05:59,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=941806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:06:03,158 INFO [optim.py:369] (1/2) 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 01:06:36,921 INFO [train.py:968] (1/2) Epoch 21, batch 29500, giga_loss[loss=0.297, simple_loss=0.3621, pruned_loss=0.116, over 28314.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3737, pruned_loss=0.1246, over 5660679.01 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3648, pruned_loss=0.1167, over 5705350.05 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1252, over 5657944.12 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:06:52,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 01:07:12,597 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:07:26,297 INFO [train.py:968] (1/2) Epoch 21, batch 29550, giga_loss[loss=0.3262, simple_loss=0.3842, pruned_loss=0.1341, over 28712.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.376, pruned_loss=0.1266, over 5657647.06 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.365, pruned_loss=0.1167, over 5708060.99 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3768, pruned_loss=0.1273, over 5652423.57 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:07:34,896 INFO [optim.py:369] (1/2) 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 01:07:58,213 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:08:06,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=941944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:08:14,477 INFO [train.py:968] (1/2) Epoch 21, batch 29600, giga_loss[loss=0.3216, simple_loss=0.3869, pruned_loss=0.1281, over 28881.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.376, pruned_loss=0.1263, over 5658364.79 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3651, pruned_loss=0.1169, over 5710671.90 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3767, pruned_loss=0.1268, over 5651476.44 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:08:18,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=941955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:09:02,098 INFO [train.py:968] (1/2) Epoch 21, batch 29650, libri_loss[loss=0.3359, simple_loss=0.3958, pruned_loss=0.138, over 29534.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3761, pruned_loss=0.1261, over 5657975.48 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3655, pruned_loss=0.1172, over 5705746.85 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3764, pruned_loss=0.1264, over 5654962.42 frames. ], batch size: 83, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:09:11,022 INFO [optim.py:369] (1/2) 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 01:09:50,613 INFO [train.py:968] (1/2) Epoch 21, batch 29700, giga_loss[loss=0.2891, simple_loss=0.3569, pruned_loss=0.1107, over 28764.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3747, pruned_loss=0.1241, over 5674354.74 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3654, pruned_loss=0.117, over 5707571.80 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3753, pruned_loss=0.1246, over 5670018.13 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:09:53,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-11 01:09:59,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:05,592 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:13,401 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:18,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:24,382 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:27,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:10:38,282 INFO [train.py:968] (1/2) Epoch 21, batch 29750, giga_loss[loss=0.2943, simple_loss=0.3666, pruned_loss=0.111, over 28596.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3742, pruned_loss=0.124, over 5657449.00 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.117, over 5709144.50 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3749, pruned_loss=0.1245, over 5651497.57 frames. ], batch size: 78, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:10:43,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1299, 1.0789, 3.6345, 3.1179], device='cuda:1'), covar=tensor([0.2140, 0.3175, 0.0919, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0653, 0.0971, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 01:10:45,463 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:10:53,655 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:11:24,957 INFO [train.py:968] (1/2) Epoch 21, batch 29800, giga_loss[loss=0.3097, simple_loss=0.3701, pruned_loss=0.1246, over 28779.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3727, pruned_loss=0.1226, over 5656385.80 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3655, pruned_loss=0.1173, over 5706007.98 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3732, pruned_loss=0.1228, over 5653539.49 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:12:06,180 INFO [train.py:968] (1/2) Epoch 21, batch 29850, giga_loss[loss=0.2676, simple_loss=0.339, pruned_loss=0.0981, over 28749.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5662314.92 frames. ], libri_tot_loss[loss=0.3006, simple_loss=0.3659, pruned_loss=0.1176, over 5704935.41 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3717, pruned_loss=0.1222, over 5659794.12 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:12:09,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:12:12,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:12:14,930 INFO [optim.py:369] (1/2) 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 01:12:17,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4013, 1.5653, 1.6771, 1.4744], device='cuda:1'), covar=tensor([0.1614, 0.1490, 0.1671, 0.1582], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0754, 0.0718, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 01:12:31,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4211, 1.7260, 1.4001, 1.3511], device='cuda:1'), covar=tensor([0.2732, 0.2690, 0.3070, 0.2292], device='cuda:1'), in_proj_covar=tensor([0.1501, 0.1089, 0.1325, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 01:12:38,900 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:12:51,682 INFO [train.py:968] (1/2) Epoch 21, batch 29900, giga_loss[loss=0.3316, simple_loss=0.3808, pruned_loss=0.1412, over 28533.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.121, over 5654924.14 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5703069.59 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1213, over 5653590.53 frames. ], batch size: 71, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:13:02,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:13:40,450 INFO [train.py:968] (1/2) Epoch 21, batch 29950, giga_loss[loss=0.2605, simple_loss=0.328, pruned_loss=0.09647, over 28767.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3666, pruned_loss=0.1199, over 5667465.04 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3658, pruned_loss=0.1175, over 5707038.41 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3669, pruned_loss=0.1201, over 5662183.68 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:13:50,273 INFO [optim.py:369] (1/2) 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 01:14:07,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:14:17,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1764, 4.0095, 3.8227, 1.8027], device='cuda:1'), covar=tensor([0.0670, 0.0792, 0.0777, 0.2054], device='cuda:1'), in_proj_covar=tensor([0.1241, 0.1148, 0.0973, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 01:14:24,957 INFO [train.py:968] (1/2) Epoch 21, batch 30000, giga_loss[loss=0.3002, simple_loss=0.3671, pruned_loss=0.1167, over 28695.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3642, pruned_loss=0.119, over 5686622.12 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 5711295.34 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3647, pruned_loss=0.1193, over 5678068.46 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:14:24,957 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 01:14:33,836 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 01:15:13,798 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:15:22,424 INFO [train.py:968] (1/2) Epoch 21, batch 30050, giga_loss[loss=0.3268, simple_loss=0.3802, pruned_loss=0.1367, over 28961.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3638, pruned_loss=0.1191, over 5690617.58 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.366, pruned_loss=0.1177, over 5707809.74 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3637, pruned_loss=0.1191, over 5686891.36 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:15:29,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:15:32,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:15:34,525 INFO [optim.py:369] (1/2) 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 01:15:52,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0503, 4.8700, 4.6044, 2.0349], device='cuda:1'), covar=tensor([0.0511, 0.0630, 0.0758, 0.2165], device='cuda:1'), in_proj_covar=tensor([0.1248, 0.1154, 0.0978, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 01:15:58,804 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:16:03,783 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 30100, giga_loss[loss=0.2673, simple_loss=0.3479, pruned_loss=0.09337, over 28932.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3627, pruned_loss=0.1178, over 5681374.19 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3652, pruned_loss=0.1173, over 5704744.86 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3634, pruned_loss=0.1183, over 5680449.25 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:16:31,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:16:35,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:17:00,581 INFO [train.py:968] (1/2) Epoch 21, batch 30150, giga_loss[loss=0.2763, simple_loss=0.3497, pruned_loss=0.1014, over 28854.00 frames. ], tot_loss[loss=0.295, simple_loss=0.361, pruned_loss=0.1145, over 5681937.18 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3645, pruned_loss=0.1168, over 5710292.25 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3621, pruned_loss=0.1153, over 5675737.11 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:17:06,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:17:15,431 INFO [optim.py:369] (1/2) 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 01:17:48,068 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942545.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:17:53,262 INFO [train.py:968] (1/2) Epoch 21, batch 30200, giga_loss[loss=0.2579, simple_loss=0.3279, pruned_loss=0.09398, over 28303.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3591, pruned_loss=0.1121, over 5667135.52 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3642, pruned_loss=0.1168, over 5710944.93 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3601, pruned_loss=0.1127, over 5661159.18 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:17:58,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2902, 5.0857, 4.8355, 2.5676], device='cuda:1'), covar=tensor([0.0489, 0.0649, 0.0811, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.1244, 0.1150, 0.0975, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 01:18:27,546 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:18:29,927 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 30250, giga_loss[loss=0.2439, simple_loss=0.333, pruned_loss=0.07742, over 28946.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3555, pruned_loss=0.1081, over 5659754.86 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3642, pruned_loss=0.1168, over 5710555.12 frames. ], giga_tot_loss[loss=0.2866, simple_loss=0.3563, pruned_loss=0.1085, over 5654926.56 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:18:55,316 INFO [optim.py:369] (1/2) 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 01:19:02,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:19:19,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=942637.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:19:32,742 INFO [train.py:968] (1/2) Epoch 21, batch 30300, giga_loss[loss=0.2466, simple_loss=0.3329, pruned_loss=0.08017, over 28931.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3517, pruned_loss=0.1043, over 5656331.67 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3642, pruned_loss=0.1168, over 5711650.34 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3522, pruned_loss=0.1045, over 5651252.63 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:20:21,830 INFO [train.py:968] (1/2) Epoch 21, batch 30350, giga_loss[loss=0.2865, simple_loss=0.3756, pruned_loss=0.0987, over 28818.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3504, pruned_loss=0.1016, over 5653176.55 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3638, pruned_loss=0.1166, over 5716453.04 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1016, over 5642918.63 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:20:31,441 INFO [optim.py:369] (1/2) 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 01:21:11,139 INFO [train.py:968] (1/2) Epoch 21, batch 30400, giga_loss[loss=0.3313, simple_loss=0.3973, pruned_loss=0.1327, over 28272.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3516, pruned_loss=0.1022, over 5643497.93 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1164, over 5710992.12 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3522, pruned_loss=0.1021, over 5638369.21 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:21:30,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:22:05,805 INFO [train.py:968] (1/2) Epoch 21, batch 30450, giga_loss[loss=0.3218, simple_loss=0.3845, pruned_loss=0.1296, over 27975.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3507, pruned_loss=0.1015, over 5638306.61 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3633, pruned_loss=0.1165, over 5708364.16 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.351, pruned_loss=0.1012, over 5636097.13 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:22:19,032 INFO [optim.py:369] (1/2) 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 01:22:54,631 INFO [train.py:968] (1/2) Epoch 21, batch 30500, giga_loss[loss=0.23, simple_loss=0.315, pruned_loss=0.07251, over 28566.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.348, pruned_loss=0.1001, over 5641737.64 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3632, pruned_loss=0.1168, over 5710932.64 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3481, pruned_loss=0.09917, over 5635850.85 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:23:37,417 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 01:23:42,297 INFO [train.py:968] (1/2) Epoch 21, batch 30550, giga_loss[loss=0.2691, simple_loss=0.3497, pruned_loss=0.09426, over 29040.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3472, pruned_loss=0.09988, over 5627929.45 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3632, pruned_loss=0.1172, over 5696590.51 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3467, pruned_loss=0.09827, over 5634743.40 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:23:49,866 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=942909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:23:53,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=942912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:23:53,661 INFO [optim.py:369] (1/2) 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 01:23:59,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-11 01:24:01,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=942920.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:24:17,700 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=942941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:24:27,743 INFO [train.py:968] (1/2) Epoch 21, batch 30600, giga_loss[loss=0.2802, simple_loss=0.355, pruned_loss=0.1027, over 28777.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3476, pruned_loss=0.09973, over 5635902.82 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3628, pruned_loss=0.1172, over 5697907.41 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3471, pruned_loss=0.09807, over 5638527.17 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:25:16,960 INFO [train.py:968] (1/2) Epoch 21, batch 30650, giga_loss[loss=0.2125, simple_loss=0.3042, pruned_loss=0.06037, over 28840.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3454, pruned_loss=0.09777, over 5642892.23 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3628, pruned_loss=0.1173, over 5701313.18 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3449, pruned_loss=0.09611, over 5641137.91 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:25:27,805 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:26:06,920 INFO [train.py:968] (1/2) Epoch 21, batch 30700, giga_loss[loss=0.2555, simple_loss=0.3312, pruned_loss=0.08989, over 27575.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3418, pruned_loss=0.09451, over 5648520.10 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3624, pruned_loss=0.1171, over 5703400.76 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3416, pruned_loss=0.09305, over 5644658.65 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:26:20,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=943063.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:26:22,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=943066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:26:49,738 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=943095.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:26:53,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-11 01:26:56,207 INFO [train.py:968] (1/2) Epoch 21, batch 30750, giga_loss[loss=0.2586, simple_loss=0.3346, pruned_loss=0.09124, over 28643.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3392, pruned_loss=0.09338, over 5644546.68 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3621, pruned_loss=0.117, over 5707952.14 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3387, pruned_loss=0.09181, over 5636064.36 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:27:08,760 INFO [optim.py:369] (1/2) 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 01:27:43,213 INFO [train.py:968] (1/2) Epoch 21, batch 30800, giga_loss[loss=0.2686, simple_loss=0.3495, pruned_loss=0.0938, over 28844.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3378, pruned_loss=0.09334, over 5644698.74 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3616, pruned_loss=0.1169, over 5702599.70 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3374, pruned_loss=0.09171, over 5640846.10 frames. ], batch size: 112, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:27:47,456 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=943155.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 01:27:49,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3120, 2.3335, 1.8574, 2.0554], device='cuda:1'), covar=tensor([0.0775, 0.0532, 0.0836, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0447, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 01:27:49,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=943158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:28:18,845 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=943187.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:28:20,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8534, 2.3978, 2.1902, 1.8165], device='cuda:1'), covar=tensor([0.2907, 0.1760, 0.1845, 0.2197], device='cuda:1'), in_proj_covar=tensor([0.1936, 0.1866, 0.1791, 0.1928], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 01:28:30,383 INFO [train.py:968] (1/2) Epoch 21, batch 30850, giga_loss[loss=0.2718, simple_loss=0.3473, pruned_loss=0.09814, over 28581.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.336, pruned_loss=0.09315, over 5630965.08 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3609, pruned_loss=0.1166, over 5696348.14 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3355, pruned_loss=0.09131, over 5630403.00 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:28:42,886 INFO [optim.py:369] (1/2) 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 01:29:22,905 INFO [train.py:968] (1/2) Epoch 21, batch 30900, giga_loss[loss=0.2892, simple_loss=0.3615, pruned_loss=0.1084, over 28967.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3361, pruned_loss=0.09325, over 5625548.56 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.36, pruned_loss=0.1161, over 5700453.07 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.336, pruned_loss=0.09172, over 5620470.71 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:29:38,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-11 01:29:45,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3927, 1.5211, 1.1928, 1.1374], device='cuda:1'), covar=tensor([0.0993, 0.0536, 0.1010, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0447, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 01:30:18,669 INFO [train.py:968] (1/2) Epoch 21, batch 30950, giga_loss[loss=0.2756, simple_loss=0.358, pruned_loss=0.09663, over 28534.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3384, pruned_loss=0.09277, over 5627813.09 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3598, pruned_loss=0.1161, over 5692122.27 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3381, pruned_loss=0.0912, over 5630005.15 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:30:31,065 INFO [optim.py:369] (1/2) 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 01:31:13,997 INFO [train.py:968] (1/2) Epoch 21, batch 31000, giga_loss[loss=0.2222, simple_loss=0.3123, pruned_loss=0.06609, over 28939.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3396, pruned_loss=0.09309, over 5639737.82 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3595, pruned_loss=0.116, over 5688526.27 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.339, pruned_loss=0.09115, over 5643070.13 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:32:21,452 INFO [train.py:968] (1/2) Epoch 21, batch 31050, giga_loss[loss=0.2236, simple_loss=0.2894, pruned_loss=0.07895, over 24663.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3394, pruned_loss=0.0929, over 5653874.88 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3588, pruned_loss=0.1157, over 5692610.26 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3391, pruned_loss=0.09117, over 5652268.36 frames. ], batch size: 705, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:32:35,727 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 31100, giga_loss[loss=0.3071, simple_loss=0.3757, pruned_loss=0.1193, over 28704.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3367, pruned_loss=0.09154, over 5654149.87 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3581, pruned_loss=0.1156, over 5695688.79 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3363, pruned_loss=0.08919, over 5648212.10 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:33:37,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7614, 4.5955, 4.3661, 2.2451], device='cuda:1'), covar=tensor([0.0537, 0.0669, 0.0774, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.1218, 0.1127, 0.0951, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 01:34:08,409 INFO [train.py:968] (1/2) Epoch 21, batch 31150, giga_loss[loss=0.2557, simple_loss=0.3377, pruned_loss=0.0868, over 28559.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3361, pruned_loss=0.09071, over 5656257.00 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3572, pruned_loss=0.1153, over 5693156.42 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3353, pruned_loss=0.08777, over 5651199.45 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:34:24,605 INFO [optim.py:369] (1/2) 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 01:34:58,467 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9660, 1.1784, 1.1536, 0.9312], device='cuda:1'), covar=tensor([0.2051, 0.1998, 0.1231, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.1925, 0.1858, 0.1780, 0.1919], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 01:35:08,675 INFO [train.py:968] (1/2) Epoch 21, batch 31200, giga_loss[loss=0.2575, simple_loss=0.3279, pruned_loss=0.09352, over 28951.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3337, pruned_loss=0.08936, over 5663094.69 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3568, pruned_loss=0.1151, over 5696215.80 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3333, pruned_loss=0.08683, over 5655994.42 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:36:00,890 INFO [train.py:968] (1/2) Epoch 21, batch 31250, giga_loss[loss=0.2438, simple_loss=0.3274, pruned_loss=0.08011, over 28916.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3313, pruned_loss=0.08926, over 5662336.75 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3556, pruned_loss=0.1147, over 5699628.54 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3308, pruned_loss=0.08634, over 5651890.51 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:36:08,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.29 vs. limit=5.0
+2023-03-11 01:36:17,779 INFO [optim.py:369] (1/2) 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 01:36:59,463 INFO [train.py:968] (1/2) Epoch 21, batch 31300, giga_loss[loss=0.2315, simple_loss=0.3152, pruned_loss=0.07385, over 28736.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3306, pruned_loss=0.08882, over 5669598.20 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3552, pruned_loss=0.1145, over 5702406.83 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3303, pruned_loss=0.08638, over 5658586.89 frames. ], batch size: 263, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:37:57,843 INFO [train.py:968] (1/2) Epoch 21, batch 31350, giga_loss[loss=0.2355, simple_loss=0.3283, pruned_loss=0.07135, over 28667.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3321, pruned_loss=0.08878, over 5666311.59 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3551, pruned_loss=0.1145, over 5704544.68 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3316, pruned_loss=0.08646, over 5655220.10 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:38:12,486 INFO [optim.py:369] (1/2) 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 01:38:54,508 INFO [train.py:968] (1/2) Epoch 21, batch 31400, libri_loss[loss=0.3422, simple_loss=0.392, pruned_loss=0.1462, over 29517.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3344, pruned_loss=0.08936, over 5675345.07 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.355, pruned_loss=0.1145, over 5711932.66 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3333, pruned_loss=0.08654, over 5658371.13 frames. ], batch size: 89, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:39:53,669 INFO [train.py:968] (1/2) Epoch 21, batch 31450, giga_loss[loss=0.2136, simple_loss=0.2986, pruned_loss=0.06431, over 28894.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3323, pruned_loss=0.08809, over 5681137.25 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3548, pruned_loss=0.1145, over 5715791.31 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3311, pruned_loss=0.08518, over 5663090.44 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:40:15,042 INFO [optim.py:369] (1/2) 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 01:41:00,830 INFO [train.py:968] (1/2) Epoch 21, batch 31500, giga_loss[loss=0.2665, simple_loss=0.3449, pruned_loss=0.09404, over 28088.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3329, pruned_loss=0.08847, over 5681101.54 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3548, pruned_loss=0.1145, over 5715222.80 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3314, pruned_loss=0.08566, over 5666425.49 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:42:00,303 INFO [train.py:968] (1/2) Epoch 21, batch 31550, giga_loss[loss=0.2642, simple_loss=0.3567, pruned_loss=0.08585, over 28893.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3355, pruned_loss=0.08921, over 5678587.09 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3543, pruned_loss=0.1142, over 5719377.26 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3341, pruned_loss=0.08642, over 5661927.66 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 01:42:18,576 INFO [optim.py:369] (1/2) 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 01:42:59,912 INFO [train.py:968] (1/2) Epoch 21, batch 31600, giga_loss[loss=0.2415, simple_loss=0.3399, pruned_loss=0.07152, over 28449.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3393, pruned_loss=0.08931, over 5673291.28 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3541, pruned_loss=0.1141, over 5722199.03 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3378, pruned_loss=0.08622, over 5655068.23 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:43:11,106 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=943960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:43:55,752 INFO [train.py:968] (1/2) Epoch 21, batch 31650, giga_loss[loss=0.2585, simple_loss=0.3252, pruned_loss=0.09586, over 26786.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3413, pruned_loss=0.08947, over 5674528.54 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3539, pruned_loss=0.114, over 5724122.52 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3397, pruned_loss=0.08616, over 5656391.42 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:44:13,285 INFO [optim.py:369] (1/2) 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 01:44:16,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2121, 4.0529, 3.8589, 2.1266], device='cuda:1'), covar=tensor([0.0502, 0.0671, 0.0705, 0.2331], device='cuda:1'), in_proj_covar=tensor([0.1212, 0.1126, 0.0949, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 01:44:54,369 INFO [train.py:968] (1/2) Epoch 21, batch 31700, giga_loss[loss=0.2617, simple_loss=0.3472, pruned_loss=0.08811, over 28718.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3409, pruned_loss=0.08848, over 5665085.36 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3538, pruned_loss=0.1139, over 5723753.29 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3396, pruned_loss=0.08551, over 5649815.77 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:44:58,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4592, 2.6002, 1.5355, 1.6138], device='cuda:1'), covar=tensor([0.0828, 0.0310, 0.0807, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0553, 0.0384, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 01:45:10,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-11 01:45:20,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0903, 1.2279, 3.4259, 2.9930], device='cuda:1'), covar=tensor([0.1726, 0.2676, 0.0526, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0646, 0.0955, 0.0898], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 01:45:28,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2737, 1.5545, 1.3799, 1.5292], device='cuda:1'), covar=tensor([0.0787, 0.0330, 0.0339, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 01:45:53,194 INFO [train.py:968] (1/2) Epoch 21, batch 31750, giga_loss[loss=0.293, simple_loss=0.3678, pruned_loss=0.1091, over 28037.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3413, pruned_loss=0.08892, over 5668785.22 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3536, pruned_loss=0.1138, over 5723774.68 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3401, pruned_loss=0.08607, over 5654989.27 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:46:08,957 INFO [optim.py:369] (1/2) 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 01:46:54,153 INFO [train.py:968] (1/2) Epoch 21, batch 31800, giga_loss[loss=0.2665, simple_loss=0.3435, pruned_loss=0.0947, over 28921.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3405, pruned_loss=0.08994, over 5665633.74 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3531, pruned_loss=0.1134, over 5724664.51 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3397, pruned_loss=0.08729, over 5651566.84 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:48:11,071 INFO [train.py:968] (1/2) Epoch 21, batch 31850, giga_loss[loss=0.2557, simple_loss=0.3359, pruned_loss=0.08779, over 29164.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3418, pruned_loss=0.09139, over 5670098.66 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3532, pruned_loss=0.1134, over 5721174.99 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.341, pruned_loss=0.08907, over 5661456.09 frames. ], batch size: 200, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:48:37,665 INFO [optim.py:369] (1/2) 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 01:49:28,229 INFO [train.py:968] (1/2) Epoch 21, batch 31900, giga_loss[loss=0.2715, simple_loss=0.3307, pruned_loss=0.1061, over 26970.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3376, pruned_loss=0.0892, over 5674403.52 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3531, pruned_loss=0.1134, over 5723290.46 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3368, pruned_loss=0.08722, over 5665351.10 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:50:17,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:50:29,999 INFO [train.py:968] (1/2) Epoch 21, batch 31950, giga_loss[loss=0.2237, simple_loss=0.3071, pruned_loss=0.07018, over 28461.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3353, pruned_loss=0.08809, over 5669413.79 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3531, pruned_loss=0.1134, over 5723721.88 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3344, pruned_loss=0.08591, over 5660974.42 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:50:50,104 INFO [optim.py:369] (1/2) 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 01:51:13,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:51:26,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3713, 3.5592, 1.5600, 1.5139], device='cuda:1'), covar=tensor([0.0954, 0.0428, 0.0882, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0555, 0.0386, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 01:51:31,557 INFO [train.py:968] (1/2) Epoch 21, batch 32000, giga_loss[loss=0.2465, simple_loss=0.3292, pruned_loss=0.08189, over 28497.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.08719, over 5667465.31 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3531, pruned_loss=0.1135, over 5718437.78 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3319, pruned_loss=0.08474, over 5664097.26 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:52:32,243 INFO [train.py:968] (1/2) Epoch 21, batch 32050, giga_loss[loss=0.2516, simple_loss=0.3411, pruned_loss=0.0811, over 28078.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3363, pruned_loss=0.08936, over 5673000.09 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3533, pruned_loss=0.1137, over 5723419.81 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3345, pruned_loss=0.0863, over 5663977.90 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 01:52:51,861 INFO [optim.py:369] (1/2) 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 01:53:29,938 INFO [train.py:968] (1/2) Epoch 21, batch 32100, giga_loss[loss=0.3084, simple_loss=0.3686, pruned_loss=0.1241, over 27763.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3389, pruned_loss=0.09098, over 5669210.48 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3529, pruned_loss=0.1136, over 5722224.86 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3375, pruned_loss=0.0882, over 5661991.14 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:54:05,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=944478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:54:10,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=944481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:54:21,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 01:54:34,111 INFO [train.py:968] (1/2) Epoch 21, batch 32150, libri_loss[loss=0.2402, simple_loss=0.3078, pruned_loss=0.08629, over 29591.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3358, pruned_loss=0.09007, over 5669233.89 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3525, pruned_loss=0.1133, over 5725033.55 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3348, pruned_loss=0.08773, over 5660035.24 frames. ], batch size: 75, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:54:45,594 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=944510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:54:51,307 INFO [optim.py:369] (1/2) 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 01:55:31,776 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:55:34,476 INFO [train.py:968] (1/2) Epoch 21, batch 32200, giga_loss[loss=0.2749, simple_loss=0.3424, pruned_loss=0.1037, over 26975.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3366, pruned_loss=0.09098, over 5662100.98 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3526, pruned_loss=0.1133, over 5717448.48 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3355, pruned_loss=0.08876, over 5659751.94 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:55:37,654 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 01:56:39,096 INFO [train.py:968] (1/2) Epoch 21, batch 32250, giga_loss[loss=0.2785, simple_loss=0.3556, pruned_loss=0.1007, over 28652.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3384, pruned_loss=0.0917, over 5665115.03 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3522, pruned_loss=0.1131, over 5718137.30 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3376, pruned_loss=0.08969, over 5661896.15 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:56:58,750 INFO [optim.py:369] (1/2) 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 01:57:47,117 INFO [train.py:968] (1/2) Epoch 21, batch 32300, libri_loss[loss=0.3076, simple_loss=0.3568, pruned_loss=0.1292, over 29547.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.339, pruned_loss=0.09086, over 5675308.46 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3517, pruned_loss=0.1129, over 5722639.87 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3385, pruned_loss=0.08886, over 5667459.97 frames. ], batch size: 81, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:58:06,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944666.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 01:58:50,853 INFO [train.py:968] (1/2) Epoch 21, batch 32350, libri_loss[loss=0.3792, simple_loss=0.4038, pruned_loss=0.1773, over 19936.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.338, pruned_loss=0.09065, over 5658196.74 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3509, pruned_loss=0.1126, over 5707609.41 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3375, pruned_loss=0.08796, over 5663268.87 frames. ], batch size: 188, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 01:59:14,457 INFO [optim.py:369] (1/2) 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 01:59:19,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2476, 4.0970, 3.8596, 2.0471], device='cuda:1'), covar=tensor([0.0578, 0.0724, 0.0768, 0.2162], device='cuda:1'), in_proj_covar=tensor([0.1209, 0.1119, 0.0944, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 01:59:57,577 INFO [train.py:968] (1/2) Epoch 21, batch 32400, giga_loss[loss=0.2386, simple_loss=0.3191, pruned_loss=0.079, over 28725.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3348, pruned_loss=0.08978, over 5669329.59 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3504, pruned_loss=0.1125, over 5710569.53 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3347, pruned_loss=0.08752, over 5670078.46 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:00:37,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5549, 1.3155, 4.1563, 3.3909], device='cuda:1'), covar=tensor([0.1538, 0.2826, 0.0446, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0647, 0.0956, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 02:01:03,344 INFO [train.py:968] (1/2) Epoch 21, batch 32450, giga_loss[loss=0.2018, simple_loss=0.285, pruned_loss=0.05933, over 28679.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3292, pruned_loss=0.08744, over 5662218.07 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3504, pruned_loss=0.1126, over 5704339.02 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3289, pruned_loss=0.08527, over 5668035.74 frames. ], batch size: 262, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:01:14,821 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=944809.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:01:17,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=944812.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 02:01:26,302 INFO [optim.py:369] (1/2) 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 02:01:40,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5961, 1.7860, 1.4869, 1.6007], device='cuda:1'), covar=tensor([0.2651, 0.2633, 0.2950, 0.2449], device='cuda:1'), in_proj_covar=tensor([0.1496, 0.1076, 0.1322, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 02:01:57,243 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=944841.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:02:07,952 INFO [train.py:968] (1/2) Epoch 21, batch 32500, giga_loss[loss=0.2834, simple_loss=0.3584, pruned_loss=0.1042, over 28351.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3295, pruned_loss=0.088, over 5652650.18 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3506, pruned_loss=0.1128, over 5705188.32 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.329, pruned_loss=0.08596, over 5656232.69 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:02:24,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6248, 1.8933, 1.3152, 1.4777], device='cuda:1'), covar=tensor([0.1003, 0.0607, 0.0987, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0444, 0.0514, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 02:02:35,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3761, 1.8714, 1.7055, 1.6363], device='cuda:1'), covar=tensor([0.1594, 0.1364, 0.1764, 0.1539], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0730, 0.0699, 0.0668], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 02:03:05,924 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:03:06,952 INFO [train.py:968] (1/2) Epoch 21, batch 32550, libri_loss[loss=0.257, simple_loss=0.3293, pruned_loss=0.0924, over 29514.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3317, pruned_loss=0.0895, over 5651290.78 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3504, pruned_loss=0.1126, over 5706442.60 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3312, pruned_loss=0.08773, over 5652219.12 frames. ], batch size: 81, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:03:27,626 INFO [optim.py:369] (1/2) 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 02:03:34,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:03:39,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=944928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:03:57,706 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-11 02:04:05,654 INFO [train.py:968] (1/2) Epoch 21, batch 32600, giga_loss[loss=0.2071, simple_loss=0.2915, pruned_loss=0.06131, over 27570.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3304, pruned_loss=0.08816, over 5653322.86 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3505, pruned_loss=0.1126, over 5707674.61 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3296, pruned_loss=0.08632, over 5651755.20 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:04:26,940 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=944967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:04:31,320 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2664, 1.5380, 1.4130, 1.0912], device='cuda:1'), covar=tensor([0.2182, 0.1986, 0.1547, 0.1994], device='cuda:1'), in_proj_covar=tensor([0.1919, 0.1845, 0.1770, 0.1908], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 02:05:04,263 INFO [train.py:968] (1/2) Epoch 21, batch 32650, giga_loss[loss=0.2514, simple_loss=0.3285, pruned_loss=0.08714, over 28850.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3297, pruned_loss=0.08712, over 5664678.06 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3507, pruned_loss=0.1127, over 5712929.30 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3283, pruned_loss=0.08483, over 5657084.46 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:05:22,610 INFO [optim.py:369] (1/2) 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 02:06:03,038 INFO [train.py:968] (1/2) Epoch 21, batch 32700, giga_loss[loss=0.2501, simple_loss=0.3287, pruned_loss=0.08575, over 28455.00 frames. ], tot_loss[loss=0.252, simple_loss=0.329, pruned_loss=0.08749, over 5659760.44 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3505, pruned_loss=0.1127, over 5705290.92 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3273, pruned_loss=0.08475, over 5657991.11 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:06:23,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:06:23,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5525, 4.3528, 4.1606, 1.8213], device='cuda:1'), covar=tensor([0.0662, 0.0875, 0.0951, 0.2226], device='cuda:1'), in_proj_covar=tensor([0.1202, 0.1113, 0.0941, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 02:06:25,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:06:28,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:06:32,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:07:05,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:07:07,105 INFO [train.py:968] (1/2) Epoch 21, batch 32750, giga_loss[loss=0.2562, simple_loss=0.336, pruned_loss=0.08822, over 28676.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3276, pruned_loss=0.08613, over 5655675.92 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3501, pruned_loss=0.1125, over 5711647.89 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3259, pruned_loss=0.08339, over 5647183.30 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:07:09,197 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:07:43,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3717, 4.2107, 3.9836, 1.9313], device='cuda:1'), covar=tensor([0.0600, 0.0698, 0.0766, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.1199, 0.1111, 0.0939, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 02:08:08,997 INFO [train.py:968] (1/2) Epoch 21, batch 32800, giga_loss[loss=0.2235, simple_loss=0.3028, pruned_loss=0.07212, over 28857.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3264, pruned_loss=0.08512, over 5661703.48 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3499, pruned_loss=0.1124, over 5714838.54 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3248, pruned_loss=0.08253, over 5651254.22 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:08:18,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6217, 1.7406, 1.3084, 1.2883], device='cuda:1'), covar=tensor([0.0915, 0.0504, 0.0965, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0444, 0.0514, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 02:08:50,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4746, 3.5287, 1.5996, 1.5530], device='cuda:1'), covar=tensor([0.1005, 0.0496, 0.0934, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0549, 0.0384, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 02:09:05,590 INFO [train.py:968] (1/2) Epoch 21, batch 32850, libri_loss[loss=0.2706, simple_loss=0.3394, pruned_loss=0.1009, over 29532.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3278, pruned_loss=0.08697, over 5664417.30 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3493, pruned_loss=0.1123, over 5717212.58 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3265, pruned_loss=0.08428, over 5652345.68 frames. ], batch size: 82, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:09:27,096 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 32900, giga_loss[loss=0.2489, simple_loss=0.3277, pruned_loss=0.08499, over 28136.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3263, pruned_loss=0.08569, over 5667484.34 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3493, pruned_loss=0.1123, over 5720460.74 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3248, pruned_loss=0.08307, over 5654156.62 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:10:32,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=945274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:11:00,244 INFO [train.py:968] (1/2) Epoch 21, batch 32950, giga_loss[loss=0.2543, simple_loss=0.3383, pruned_loss=0.08514, over 28404.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08631, over 5662401.75 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3494, pruned_loss=0.1122, over 5714963.70 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3268, pruned_loss=0.08316, over 5655208.93 frames. ], batch size: 369, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:11:18,833 INFO [optim.py:369] (1/2) 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 02:11:46,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=945341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:11:47,515 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 33000, giga_loss[loss=0.269, simple_loss=0.3492, pruned_loss=0.09447, over 28900.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3322, pruned_loss=0.08701, over 5665805.86 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3496, pruned_loss=0.1123, over 5718192.12 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.33, pruned_loss=0.08398, over 5656527.12 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:11:57,923 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 02:12:06,527 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 02:13:03,143 INFO [train.py:968] (1/2) Epoch 21, batch 33050, giga_loss[loss=0.2803, simple_loss=0.3446, pruned_loss=0.108, over 26810.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3331, pruned_loss=0.08747, over 5657655.46 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3493, pruned_loss=0.1122, over 5721112.23 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3313, pruned_loss=0.08472, over 5646473.94 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:13:25,647 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:13:27,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:13:49,380 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=945436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:13:56,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-11 02:14:05,004 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:14:06,734 INFO [train.py:968] (1/2) Epoch 21, batch 33100, giga_loss[loss=0.2849, simple_loss=0.3482, pruned_loss=0.1108, over 28608.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3327, pruned_loss=0.08746, over 5653405.93 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3488, pruned_loss=0.112, over 5712698.05 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3313, pruned_loss=0.08491, over 5651165.78 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:14:44,447 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:14:48,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:15:01,259 INFO [train.py:968] (1/2) Epoch 21, batch 33150, giga_loss[loss=0.253, simple_loss=0.3339, pruned_loss=0.08607, over 28094.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3319, pruned_loss=0.08734, over 5663490.25 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3486, pruned_loss=0.1119, over 5718101.75 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3305, pruned_loss=0.08454, over 5655029.54 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:15:21,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:15:24,381 INFO [optim.py:369] (1/2) 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 02:16:00,653 INFO [train.py:968] (1/2) Epoch 21, batch 33200, giga_loss[loss=0.2259, simple_loss=0.312, pruned_loss=0.06992, over 28960.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3302, pruned_loss=0.08654, over 5653692.93 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3484, pruned_loss=0.1119, over 5712620.69 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3288, pruned_loss=0.08372, over 5650360.99 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:17:00,580 INFO [train.py:968] (1/2) Epoch 21, batch 33250, libri_loss[loss=0.2759, simple_loss=0.3404, pruned_loss=0.1057, over 29540.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3288, pruned_loss=0.08658, over 5649832.82 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3487, pruned_loss=0.1121, over 5701082.80 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3271, pruned_loss=0.08376, over 5656113.06 frames. ], batch size: 84, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:17:21,304 INFO [optim.py:369] (1/2) 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 02:17:49,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-11 02:17:59,396 INFO [train.py:968] (1/2) Epoch 21, batch 33300, giga_loss[loss=0.2893, simple_loss=0.3549, pruned_loss=0.1119, over 26735.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3302, pruned_loss=0.0873, over 5663201.20 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.348, pruned_loss=0.1118, over 5705707.25 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.329, pruned_loss=0.08467, over 5663069.57 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:19:01,343 INFO [train.py:968] (1/2) Epoch 21, batch 33350, giga_loss[loss=0.2378, simple_loss=0.3193, pruned_loss=0.07821, over 28965.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3329, pruned_loss=0.08848, over 5667474.08 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3482, pruned_loss=0.1121, over 5709826.42 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3314, pruned_loss=0.08565, over 5662969.80 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:19:20,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=945716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:19:24,973 INFO [optim.py:369] (1/2) 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 02:20:03,627 INFO [train.py:968] (1/2) Epoch 21, batch 33400, giga_loss[loss=0.2674, simple_loss=0.3433, pruned_loss=0.0957, over 28936.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3337, pruned_loss=0.08921, over 5665079.32 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3483, pruned_loss=0.1121, over 5711828.27 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3322, pruned_loss=0.0866, over 5658894.85 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:20:18,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2407, 1.2187, 3.7152, 3.1003], device='cuda:1'), covar=tensor([0.1706, 0.2946, 0.0463, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.0753, 0.0644, 0.0951, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 02:21:09,755 INFO [train.py:968] (1/2) Epoch 21, batch 33450, giga_loss[loss=0.2525, simple_loss=0.3468, pruned_loss=0.07914, over 28969.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3372, pruned_loss=0.09046, over 5666933.65 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3487, pruned_loss=0.1123, over 5709322.27 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3355, pruned_loss=0.0879, over 5663979.87 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:21:20,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=945811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:21:29,403 INFO [optim.py:369] (1/2) 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 02:22:01,076 INFO [train.py:968] (1/2) Epoch 21, batch 33500, giga_loss[loss=0.2403, simple_loss=0.3364, pruned_loss=0.07207, over 28838.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3384, pruned_loss=0.09082, over 5658544.34 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3481, pruned_loss=0.1122, over 5703564.16 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3373, pruned_loss=0.08827, over 5659304.39 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:22:10,446 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:22:14,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:22:39,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-11 02:22:54,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:23:05,861 INFO [train.py:968] (1/2) Epoch 21, batch 33550, giga_loss[loss=0.2514, simple_loss=0.3351, pruned_loss=0.08379, over 28430.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3374, pruned_loss=0.09006, over 5658307.15 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3479, pruned_loss=0.112, over 5707886.25 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3365, pruned_loss=0.08767, over 5653897.74 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:23:20,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-11 02:23:32,746 INFO [optim.py:369] (1/2) 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 02:24:10,755 INFO [train.py:968] (1/2) Epoch 21, batch 33600, giga_loss[loss=0.2278, simple_loss=0.3136, pruned_loss=0.07104, over 28878.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3346, pruned_loss=0.08837, over 5667680.35 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3474, pruned_loss=0.1116, over 5710997.37 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3341, pruned_loss=0.08623, over 5660022.37 frames. ], batch size: 120, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:24:18,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=945954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:24:22,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=945957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:24:23,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9286, 3.7555, 3.5672, 1.6808], device='cuda:1'), covar=tensor([0.0715, 0.0880, 0.0858, 0.2177], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.1107, 0.0937, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 02:24:54,282 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=945986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:25:13,318 INFO [train.py:968] (1/2) Epoch 21, batch 33650, giga_loss[loss=0.2482, simple_loss=0.336, pruned_loss=0.08025, over 28733.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3339, pruned_loss=0.08858, over 5656997.66 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3475, pruned_loss=0.1118, over 5702226.01 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3331, pruned_loss=0.08619, over 5657526.56 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:25:39,068 INFO [optim.py:369] (1/2) 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 02:26:19,340 INFO [train.py:968] (1/2) Epoch 21, batch 33700, giga_loss[loss=0.2639, simple_loss=0.3414, pruned_loss=0.09323, over 28375.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3336, pruned_loss=0.08851, over 5656169.56 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3475, pruned_loss=0.1117, over 5704732.99 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3328, pruned_loss=0.08633, over 5653753.93 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:27:00,097 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 33750, giga_loss[loss=0.3336, simple_loss=0.3992, pruned_loss=0.1339, over 28971.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3323, pruned_loss=0.08849, over 5654474.23 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3472, pruned_loss=0.1115, over 5705051.18 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3317, pruned_loss=0.08674, over 5651622.99 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:27:52,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2311, 3.1468, 1.3126, 1.5499], device='cuda:1'), covar=tensor([0.1049, 0.0482, 0.0960, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0550, 0.0384, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 02:27:52,511 INFO [optim.py:369] (1/2) 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 02:28:28,891 INFO [train.py:968] (1/2) Epoch 21, batch 33800, giga_loss[loss=0.2556, simple_loss=0.3394, pruned_loss=0.08588, over 28747.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.331, pruned_loss=0.0881, over 5646985.66 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3472, pruned_loss=0.1115, over 5707547.63 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3304, pruned_loss=0.08638, over 5641404.54 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:29:30,119 INFO [train.py:968] (1/2) Epoch 21, batch 33850, giga_loss[loss=0.2432, simple_loss=0.3262, pruned_loss=0.08012, over 28923.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3304, pruned_loss=0.0864, over 5660261.89 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3471, pruned_loss=0.1114, over 5708926.47 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3298, pruned_loss=0.0848, over 5653937.04 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:29:52,223 INFO [optim.py:369] (1/2) 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 02:30:20,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 02:30:23,210 INFO [train.py:968] (1/2) Epoch 21, batch 33900, giga_loss[loss=0.208, simple_loss=0.3025, pruned_loss=0.05675, over 28519.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3315, pruned_loss=0.08562, over 5673752.31 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3468, pruned_loss=0.1112, over 5712416.50 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3306, pruned_loss=0.08371, over 5664448.51 frames. ], batch size: 65, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:31:08,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5757, 1.6899, 1.2726, 1.3337], device='cuda:1'), covar=tensor([0.0925, 0.0528, 0.0954, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0440, 0.0511, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 02:31:16,410 INFO [train.py:968] (1/2) Epoch 21, batch 33950, giga_loss[loss=0.2574, simple_loss=0.3393, pruned_loss=0.08768, over 28090.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3332, pruned_loss=0.08527, over 5675150.28 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3465, pruned_loss=0.1109, over 5716420.56 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3324, pruned_loss=0.08318, over 5662975.85 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 02:31:41,741 INFO [optim.py:369] (1/2) 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 02:31:52,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5973, 1.3540, 4.9018, 3.5043], device='cuda:1'), covar=tensor([0.1685, 0.2865, 0.0358, 0.0975], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0643, 0.0950, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 02:32:10,140 INFO [train.py:968] (1/2) Epoch 21, batch 34000, giga_loss[loss=0.2271, simple_loss=0.3151, pruned_loss=0.06959, over 28919.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3323, pruned_loss=0.08434, over 5650035.79 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3466, pruned_loss=0.1111, over 5691646.79 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3311, pruned_loss=0.08177, over 5661038.01 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:33:16,602 INFO [train.py:968] (1/2) Epoch 21, batch 34050, giga_loss[loss=0.2422, simple_loss=0.3313, pruned_loss=0.07653, over 28936.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3318, pruned_loss=0.08419, over 5658604.42 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3462, pruned_loss=0.1109, over 5696575.96 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3309, pruned_loss=0.08168, over 5662088.03 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:33:47,570 INFO [optim.py:369] (1/2) 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 02:33:51,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6249, 1.9123, 1.7403, 1.6116], device='cuda:1'), covar=tensor([0.1941, 0.2419, 0.2190, 0.2346], device='cuda:1'), in_proj_covar=tensor([0.0455, 0.0728, 0.0697, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 02:34:21,514 INFO [train.py:968] (1/2) Epoch 21, batch 34100, giga_loss[loss=0.2445, simple_loss=0.332, pruned_loss=0.07855, over 28863.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3312, pruned_loss=0.08383, over 5653590.63 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3458, pruned_loss=0.1108, over 5689913.09 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3307, pruned_loss=0.08153, over 5662096.40 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:34:28,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=946455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:35:27,910 INFO [train.py:968] (1/2) Epoch 21, batch 34150, giga_loss[loss=0.265, simple_loss=0.3496, pruned_loss=0.09018, over 28985.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3322, pruned_loss=0.08415, over 5656972.64 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3459, pruned_loss=0.1108, over 5692162.54 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3313, pruned_loss=0.08164, over 5661087.67 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:36:02,551 INFO [optim.py:369] (1/2) 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 02:36:21,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4223, 1.5177, 1.1933, 1.1123], device='cuda:1'), covar=tensor([0.0961, 0.0521, 0.1023, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0441, 0.0513, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 02:36:40,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3756, 4.1828, 3.9906, 1.9906], device='cuda:1'), covar=tensor([0.0634, 0.0765, 0.0821, 0.2034], device='cuda:1'), in_proj_covar=tensor([0.1191, 0.1102, 0.0931, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 02:36:42,744 INFO [train.py:968] (1/2) Epoch 21, batch 34200, giga_loss[loss=0.256, simple_loss=0.3472, pruned_loss=0.0824, over 28436.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3323, pruned_loss=0.08365, over 5652002.35 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3459, pruned_loss=0.1108, over 5693223.72 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3315, pruned_loss=0.08151, over 5653971.27 frames. ], batch size: 369, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:37:37,941 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=946598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:37:44,333 INFO [train.py:968] (1/2) Epoch 21, batch 34250, giga_loss[loss=0.2802, simple_loss=0.3541, pruned_loss=0.1031, over 27015.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3356, pruned_loss=0.08572, over 5663315.75 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3452, pruned_loss=0.1104, over 5701888.24 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3351, pruned_loss=0.0833, over 5655722.68 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:37:44,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=946601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:38:09,121 INFO [optim.py:369] (1/2) 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 02:38:22,604 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:38:31,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 02:38:45,070 INFO [train.py:968] (1/2) Epoch 21, batch 34300, giga_loss[loss=0.2408, simple_loss=0.3264, pruned_loss=0.07756, over 28849.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3361, pruned_loss=0.08538, over 5680204.77 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3454, pruned_loss=0.1103, over 5703661.36 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3354, pruned_loss=0.08289, over 5671894.87 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:38:46,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2802, 2.9673, 1.3879, 1.3744], device='cuda:1'), covar=tensor([0.0993, 0.0462, 0.0958, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0551, 0.0385, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 02:39:38,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6096, 1.9555, 1.6430, 1.6334], device='cuda:1'), covar=tensor([0.2234, 0.2036, 0.2292, 0.2108], device='cuda:1'), in_proj_covar=tensor([0.1494, 0.1073, 0.1320, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 02:39:51,044 INFO [train.py:968] (1/2) Epoch 21, batch 34350, libri_loss[loss=0.2294, simple_loss=0.297, pruned_loss=0.08083, over 29459.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3346, pruned_loss=0.08558, over 5674596.80 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.345, pruned_loss=0.11, over 5694931.06 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.334, pruned_loss=0.08303, over 5675784.33 frames. ], batch size: 70, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:40:17,377 INFO [optim.py:369] (1/2) 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 02:40:59,291 INFO [train.py:968] (1/2) Epoch 21, batch 34400, giga_loss[loss=0.2398, simple_loss=0.3319, pruned_loss=0.07379, over 28749.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3328, pruned_loss=0.08448, over 5684836.16 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3448, pruned_loss=0.1098, over 5699990.96 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3322, pruned_loss=0.08204, over 5680927.49 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:41:03,369 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=946753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:42:03,649 INFO [train.py:968] (1/2) Epoch 21, batch 34450, giga_loss[loss=0.2171, simple_loss=0.3106, pruned_loss=0.0618, over 28960.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3316, pruned_loss=0.08331, over 5686939.83 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3451, pruned_loss=0.1101, over 5694642.57 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3307, pruned_loss=0.08057, over 5687641.94 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:42:24,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3453, 1.7046, 1.6313, 1.5392], device='cuda:1'), covar=tensor([0.1792, 0.1649, 0.1692, 0.1586], device='cuda:1'), in_proj_covar=tensor([0.0451, 0.0726, 0.0694, 0.0662], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 02:42:33,546 INFO [optim.py:369] (1/2) 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 02:43:07,763 INFO [train.py:968] (1/2) Epoch 21, batch 34500, giga_loss[loss=0.2566, simple_loss=0.3369, pruned_loss=0.08811, over 28998.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3314, pruned_loss=0.08328, over 5690171.99 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3453, pruned_loss=0.1102, over 5698833.67 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3302, pruned_loss=0.08055, over 5686782.96 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:43:57,001 INFO [train.py:968] (1/2) Epoch 21, batch 34550, libri_loss[loss=0.238, simple_loss=0.2998, pruned_loss=0.08814, over 29351.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3328, pruned_loss=0.0851, over 5687884.96 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3445, pruned_loss=0.1097, over 5702278.43 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3317, pruned_loss=0.08158, over 5680335.40 frames. ], batch size: 67, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:44:19,725 INFO [optim.py:369] (1/2) 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 02:44:50,582 INFO [train.py:968] (1/2) Epoch 21, batch 34600, libri_loss[loss=0.2491, simple_loss=0.3131, pruned_loss=0.09255, over 29573.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3362, pruned_loss=0.08789, over 5680312.27 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3447, pruned_loss=0.1099, over 5705334.46 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3348, pruned_loss=0.0841, over 5670782.00 frames. ], batch size: 75, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:44:57,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6150, 1.7271, 1.8623, 1.4187], device='cuda:1'), covar=tensor([0.1920, 0.2712, 0.1562, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0690, 0.0940, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 02:45:08,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=946965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:45:38,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7743, 1.8982, 1.5777, 1.9759], device='cuda:1'), covar=tensor([0.2803, 0.2745, 0.3145, 0.2466], device='cuda:1'), in_proj_covar=tensor([0.1493, 0.1072, 0.1322, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 02:45:47,570 INFO [train.py:968] (1/2) Epoch 21, batch 34650, libri_loss[loss=0.3762, simple_loss=0.408, pruned_loss=0.1722, over 29487.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3333, pruned_loss=0.08723, over 5680265.31 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3445, pruned_loss=0.1099, over 5708822.47 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3321, pruned_loss=0.08366, over 5668765.80 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:46:05,000 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-11 02:46:10,058 INFO [optim.py:369] (1/2) 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 02:46:40,992 INFO [train.py:968] (1/2) Epoch 21, batch 34700, giga_loss[loss=0.2378, simple_loss=0.3211, pruned_loss=0.07727, over 28881.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3317, pruned_loss=0.08675, over 5680091.93 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3438, pruned_loss=0.1094, over 5711911.78 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3312, pruned_loss=0.08371, over 5667078.60 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:47:31,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4556, 1.7684, 1.4469, 1.3491], device='cuda:1'), covar=tensor([0.2402, 0.2239, 0.2480, 0.2178], device='cuda:1'), in_proj_covar=tensor([0.1496, 0.1073, 0.1324, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 02:47:31,540 INFO [train.py:968] (1/2) Epoch 21, batch 34750, giga_loss[loss=0.3269, simple_loss=0.3828, pruned_loss=0.1355, over 26630.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3327, pruned_loss=0.08791, over 5676266.29 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3428, pruned_loss=0.1088, over 5718988.58 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3326, pruned_loss=0.08509, over 5657873.50 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:47:52,071 INFO [optim.py:369] (1/2) 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 02:47:55,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:48:13,477 INFO [train.py:968] (1/2) Epoch 21, batch 34800, giga_loss[loss=0.3184, simple_loss=0.3822, pruned_loss=0.1273, over 27944.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3408, pruned_loss=0.09292, over 5681688.33 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3425, pruned_loss=0.1086, over 5720657.97 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3408, pruned_loss=0.09019, over 5664288.79 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:48:57,414 INFO [train.py:968] (1/2) Epoch 21, batch 34850, giga_loss[loss=0.3071, simple_loss=0.3819, pruned_loss=0.1162, over 27898.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3483, pruned_loss=0.09678, over 5681921.99 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3426, pruned_loss=0.1086, over 5713638.29 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09416, over 5673329.27 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:49:05,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4762, 1.7172, 1.4621, 1.5749], device='cuda:1'), covar=tensor([0.0820, 0.0316, 0.0337, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 02:49:17,429 INFO [optim.py:369] (1/2) 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 02:49:38,064 INFO [train.py:968] (1/2) Epoch 21, batch 34900, libri_loss[loss=0.2215, simple_loss=0.2845, pruned_loss=0.07926, over 28620.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3488, pruned_loss=0.09734, over 5684764.60 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3423, pruned_loss=0.1083, over 5716859.47 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3494, pruned_loss=0.09505, over 5673505.91 frames. ], batch size: 63, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:49:52,060 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-11 02:49:52,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6384, 2.3796, 1.8079, 0.8679], device='cuda:1'), covar=tensor([0.4804, 0.2837, 0.3500, 0.5169], device='cuda:1'), in_proj_covar=tensor([0.1743, 0.1647, 0.1596, 0.1422], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 02:49:54,708 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=947271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:49:56,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:50:12,021 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947292.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:50:18,103 INFO [train.py:968] (1/2) Epoch 21, batch 34950, giga_loss[loss=0.2787, simple_loss=0.3554, pruned_loss=0.101, over 28957.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3448, pruned_loss=0.09596, over 5678846.84 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3425, pruned_loss=0.1084, over 5710563.78 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3452, pruned_loss=0.09385, over 5674349.02 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:50:20,492 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:50:49,734 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 35000, giga_loss[loss=0.2182, simple_loss=0.301, pruned_loss=0.06766, over 28935.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3385, pruned_loss=0.09348, over 5673939.87 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3429, pruned_loss=0.1087, over 5706433.75 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3385, pruned_loss=0.09124, over 5672801.86 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:51:30,366 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:51:39,559 INFO [train.py:968] (1/2) Epoch 21, batch 35050, giga_loss[loss=0.2384, simple_loss=0.3107, pruned_loss=0.08303, over 27967.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3322, pruned_loss=0.0907, over 5675827.86 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3435, pruned_loss=0.1089, over 5699029.34 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3313, pruned_loss=0.08807, over 5680983.71 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:51:48,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0583, 1.3936, 5.2638, 3.8155], device='cuda:1'), covar=tensor([0.1551, 0.2995, 0.0381, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0644, 0.0957, 0.0898], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 02:51:57,359 INFO [optim.py:369] (1/2) 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 02:52:18,643 INFO [train.py:968] (1/2) Epoch 21, batch 35100, giga_loss[loss=0.2122, simple_loss=0.2807, pruned_loss=0.07189, over 28572.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3261, pruned_loss=0.08833, over 5678505.11 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3442, pruned_loss=0.1091, over 5703168.29 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3244, pruned_loss=0.08549, over 5678245.53 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:52:20,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2667, 0.7823, 0.8277, 1.4150], device='cuda:1'), covar=tensor([0.0681, 0.0365, 0.0341, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 02:52:35,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6996, 1.7488, 1.6598, 1.5957], device='cuda:1'), covar=tensor([0.2880, 0.2746, 0.2171, 0.2472], device='cuda:1'), in_proj_covar=tensor([0.1947, 0.1861, 0.1780, 0.1928], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 02:52:43,366 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=947483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:52:47,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:53:00,248 INFO [train.py:968] (1/2) Epoch 21, batch 35150, libri_loss[loss=0.2849, simple_loss=0.3564, pruned_loss=0.1067, over 29736.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3206, pruned_loss=0.08596, over 5673863.15 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.345, pruned_loss=0.1095, over 5696446.72 frames. ], giga_tot_loss[loss=0.2418, simple_loss=0.318, pruned_loss=0.08276, over 5678055.63 frames. ], batch size: 87, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:53:12,720 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:53:34,814 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:53:38,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3046, 2.5776, 1.3218, 1.4182], device='cuda:1'), covar=tensor([0.0982, 0.0395, 0.0919, 0.1351], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0548, 0.0382, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 02:53:41,146 INFO [train.py:968] (1/2) Epoch 21, batch 35200, libri_loss[loss=0.3212, simple_loss=0.3786, pruned_loss=0.132, over 29504.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3173, pruned_loss=0.08443, over 5681763.10 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3452, pruned_loss=0.1095, over 5692459.80 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3142, pruned_loss=0.08118, over 5688761.11 frames. ], batch size: 81, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:54:21,828 INFO [train.py:968] (1/2) Epoch 21, batch 35250, giga_loss[loss=0.2534, simple_loss=0.3126, pruned_loss=0.09706, over 26530.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3154, pruned_loss=0.08399, over 5682056.35 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3461, pruned_loss=0.1099, over 5695232.64 frames. ], giga_tot_loss[loss=0.2355, simple_loss=0.311, pruned_loss=0.08, over 5684662.94 frames. ], batch size: 555, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:54:38,755 INFO [optim.py:369] (1/2) 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 02:54:54,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-11 02:55:00,720 INFO [train.py:968] (1/2) Epoch 21, batch 35300, giga_loss[loss=0.2109, simple_loss=0.2959, pruned_loss=0.06289, over 28831.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3125, pruned_loss=0.08252, over 5683272.05 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3465, pruned_loss=0.11, over 5699999.03 frames. ], giga_tot_loss[loss=0.2324, simple_loss=0.3077, pruned_loss=0.07854, over 5680636.70 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 02:55:10,564 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-11 02:55:13,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947667.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 02:55:45,083 INFO [train.py:968] (1/2) Epoch 21, batch 35350, libri_loss[loss=0.4247, simple_loss=0.4528, pruned_loss=0.1983, over 29222.00 frames. ], tot_loss[loss=0.2379, simple_loss=0.3109, pruned_loss=0.08241, over 5669408.34 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3471, pruned_loss=0.1104, over 5696465.00 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3053, pruned_loss=0.07784, over 5670717.97 frames. ], batch size: 97, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:56:03,333 INFO [optim.py:369] (1/2) 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 02:56:23,726 INFO [train.py:968] (1/2) Epoch 21, batch 35400, giga_loss[loss=0.2129, simple_loss=0.2898, pruned_loss=0.06797, over 28764.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3074, pruned_loss=0.08065, over 5674613.42 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3466, pruned_loss=0.11, over 5692115.66 frames. ], giga_tot_loss[loss=0.2272, simple_loss=0.3019, pruned_loss=0.07622, over 5679047.22 frames. ], batch size: 243, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:56:35,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=947764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:56:57,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5423, 1.6767, 1.7741, 1.3761], device='cuda:1'), covar=tensor([0.1606, 0.2258, 0.1371, 0.1567], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0700, 0.0953, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 02:57:06,934 INFO [train.py:968] (1/2) Epoch 21, batch 35450, giga_loss[loss=0.1903, simple_loss=0.2732, pruned_loss=0.05376, over 29087.00 frames. ], tot_loss[loss=0.2312, simple_loss=0.3044, pruned_loss=0.07906, over 5681634.69 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3466, pruned_loss=0.11, over 5694271.21 frames. ], giga_tot_loss[loss=0.2252, simple_loss=0.2996, pruned_loss=0.07536, over 5682891.70 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:57:14,136 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=947810.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 02:57:16,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947813.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:57:24,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6111, 1.8562, 1.5560, 1.5849], device='cuda:1'), covar=tensor([0.2722, 0.2809, 0.3116, 0.2551], device='cuda:1'), in_proj_covar=tensor([0.1499, 0.1079, 0.1324, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 02:57:24,545 INFO [optim.py:369] (1/2) 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 02:57:39,922 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=947842.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 02:57:46,156 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-11 02:57:46,933 INFO [train.py:968] (1/2) Epoch 21, batch 35500, giga_loss[loss=0.2135, simple_loss=0.2889, pruned_loss=0.06903, over 28979.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.302, pruned_loss=0.07829, over 5688094.99 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3467, pruned_loss=0.11, over 5699740.60 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2969, pruned_loss=0.07443, over 5683887.97 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:57:48,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6942, 1.7966, 1.3590, 1.3433], device='cuda:1'), covar=tensor([0.0959, 0.0623, 0.1055, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0514, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 02:57:51,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7831, 2.0241, 1.4969, 1.5599], device='cuda:1'), covar=tensor([0.0978, 0.0614, 0.1100, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0515, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 02:58:28,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2578, 0.8485, 0.9814, 1.4504], device='cuda:1'), covar=tensor([0.0732, 0.0436, 0.0370, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 02:58:30,363 INFO [train.py:968] (1/2) Epoch 21, batch 35550, giga_loss[loss=0.1906, simple_loss=0.2741, pruned_loss=0.05348, over 29015.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.2993, pruned_loss=0.07697, over 5685744.22 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.347, pruned_loss=0.1101, over 5703311.95 frames. ], giga_tot_loss[loss=0.2202, simple_loss=0.2941, pruned_loss=0.07317, over 5678891.12 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:58:37,592 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=947910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:58:40,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=947912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 02:58:45,615 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:59:04,358 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 35600, giga_loss[loss=0.1848, simple_loss=0.2591, pruned_loss=0.0553, over 28472.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.2989, pruned_loss=0.0774, over 5673889.78 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3479, pruned_loss=0.1106, over 5694260.37 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.2928, pruned_loss=0.07317, over 5675605.37 frames. ], batch size: 78, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 02:59:48,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.79 vs. limit=5.0
+2023-03-11 02:59:56,586 INFO [train.py:968] (1/2) Epoch 21, batch 35650, giga_loss[loss=0.279, simple_loss=0.3636, pruned_loss=0.09717, over 28907.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3074, pruned_loss=0.08155, over 5657498.04 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3483, pruned_loss=0.1108, over 5672617.89 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.3009, pruned_loss=0.07709, over 5676886.69 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:00:17,795 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 35700, giga_loss[loss=0.3438, simple_loss=0.4026, pruned_loss=0.1425, over 28355.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3213, pruned_loss=0.0889, over 5665958.20 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3479, pruned_loss=0.1105, over 5678522.91 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3156, pruned_loss=0.08503, over 5676100.16 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:00:49,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=948060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:00:52,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=948063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:01:16,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=948092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:01:22,426 INFO [train.py:968] (1/2) Epoch 21, batch 35750, giga_loss[loss=0.2762, simple_loss=0.3481, pruned_loss=0.1022, over 28569.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.332, pruned_loss=0.09462, over 5658644.51 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3495, pruned_loss=0.1115, over 5667554.29 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3251, pruned_loss=0.08981, over 5676573.93 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:01:42,618 INFO [optim.py:369] (1/2) 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 03:02:04,583 INFO [train.py:968] (1/2) Epoch 21, batch 35800, giga_loss[loss=0.2503, simple_loss=0.3366, pruned_loss=0.08195, over 28517.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3385, pruned_loss=0.09665, over 5670243.92 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3493, pruned_loss=0.1113, over 5672453.24 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3329, pruned_loss=0.09272, over 5680119.97 frames. ], batch size: 71, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:02:44,064 INFO [train.py:968] (1/2) Epoch 21, batch 35850, giga_loss[loss=0.2613, simple_loss=0.3424, pruned_loss=0.09008, over 29050.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3413, pruned_loss=0.09675, over 5675784.22 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3493, pruned_loss=0.1109, over 5672798.53 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3363, pruned_loss=0.09324, over 5683269.16 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:02:45,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2232, 1.0857, 3.9964, 3.0746], device='cuda:1'), covar=tensor([0.1792, 0.3102, 0.0459, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0749, 0.0640, 0.0954, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 03:02:52,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0153, 2.0198, 2.2842, 1.7640], device='cuda:1'), covar=tensor([0.1889, 0.2462, 0.1416, 0.1697], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0698, 0.0949, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 03:03:08,211 INFO [optim.py:369] (1/2) 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 03:03:32,258 INFO [train.py:968] (1/2) Epoch 21, batch 35900, giga_loss[loss=0.266, simple_loss=0.341, pruned_loss=0.09547, over 28492.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3418, pruned_loss=0.09563, over 5664806.19 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.349, pruned_loss=0.1106, over 5675249.08 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.338, pruned_loss=0.09299, over 5668579.60 frames. ], batch size: 78, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:04:05,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=948287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:04:15,454 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-11 03:04:18,056 INFO [train.py:968] (1/2) Epoch 21, batch 35950, giga_loss[loss=0.2554, simple_loss=0.3398, pruned_loss=0.08548, over 28748.00 frames. ], tot_loss[loss=0.269, simple_loss=0.344, pruned_loss=0.09702, over 5667114.54 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3494, pruned_loss=0.1109, over 5672758.10 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3405, pruned_loss=0.09433, over 5672138.38 frames. ], batch size: 284, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:04:18,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=948302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:04:42,000 INFO [optim.py:369] (1/2) 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 03:05:00,312 INFO [train.py:968] (1/2) Epoch 21, batch 36000, giga_loss[loss=0.3154, simple_loss=0.3805, pruned_loss=0.1252, over 28930.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3465, pruned_loss=0.09883, over 5670498.57 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3497, pruned_loss=0.111, over 5667769.09 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3433, pruned_loss=0.09625, over 5678841.74 frames. ], batch size: 213, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:05:00,312 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 03:05:09,655 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 03:05:51,983 INFO [train.py:968] (1/2) Epoch 21, batch 36050, giga_loss[loss=0.2754, simple_loss=0.352, pruned_loss=0.09943, over 28763.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3502, pruned_loss=0.1014, over 5672998.57 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3499, pruned_loss=0.111, over 5670377.37 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3475, pruned_loss=0.09929, over 5677273.64 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:06:02,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6045, 1.7800, 1.7651, 1.5878], device='cuda:1'), covar=tensor([0.2036, 0.2310, 0.2438, 0.2335], device='cuda:1'), in_proj_covar=tensor([0.0466, 0.0741, 0.0709, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 03:06:12,503 INFO [optim.py:369] (1/2) 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 03:06:15,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=948430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:06:17,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=948433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:06:28,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1938, 2.9840, 2.8394, 1.5211], device='cuda:1'), covar=tensor([0.0970, 0.1098, 0.0908, 0.2362], device='cuda:1'), in_proj_covar=tensor([0.1190, 0.1109, 0.0931, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 03:06:29,537 INFO [train.py:968] (1/2) Epoch 21, batch 36100, giga_loss[loss=0.2684, simple_loss=0.3549, pruned_loss=0.09091, over 28942.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3548, pruned_loss=0.1032, over 5689996.85 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3507, pruned_loss=0.1113, over 5673342.02 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.352, pruned_loss=0.101, over 5691406.48 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:06:38,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=948462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:06:58,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2035, 3.2599, 1.9458, 1.1101], device='cuda:1'), covar=tensor([0.8042, 0.2911, 0.4690, 0.7173], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1644, 0.1600, 0.1419], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 03:07:09,113 INFO [train.py:968] (1/2) Epoch 21, batch 36150, giga_loss[loss=0.321, simple_loss=0.3885, pruned_loss=0.1267, over 27916.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.357, pruned_loss=0.104, over 5693273.75 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3516, pruned_loss=0.1117, over 5677778.88 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3541, pruned_loss=0.1015, over 5690839.45 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:07:32,151 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 36200, giga_loss[loss=0.2811, simple_loss=0.364, pruned_loss=0.09912, over 29004.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3576, pruned_loss=0.103, over 5698080.90 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.352, pruned_loss=0.112, over 5678420.32 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.355, pruned_loss=0.1008, over 5695661.52 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:08:27,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1605, 0.8747, 0.9784, 1.3291], device='cuda:1'), covar=tensor([0.0845, 0.0391, 0.0368, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 03:08:28,975 INFO [train.py:968] (1/2) Epoch 21, batch 36250, giga_loss[loss=0.2926, simple_loss=0.3688, pruned_loss=0.1082, over 28842.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3569, pruned_loss=0.1018, over 5697481.86 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.352, pruned_loss=0.1119, over 5683003.39 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3549, pruned_loss=0.0998, over 5691918.81 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:08:46,966 INFO [optim.py:369] (1/2) 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 03:09:05,754 INFO [train.py:968] (1/2) Epoch 21, batch 36300, giga_loss[loss=0.2592, simple_loss=0.3465, pruned_loss=0.08596, over 28951.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3548, pruned_loss=0.09976, over 5709812.00 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3521, pruned_loss=0.1118, over 5690723.77 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3533, pruned_loss=0.09773, over 5699081.25 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:09:23,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1573, 1.2621, 1.1092, 1.1529], device='cuda:1'), covar=tensor([0.1898, 0.1917, 0.1353, 0.1701], device='cuda:1'), in_proj_covar=tensor([0.1960, 0.1877, 0.1796, 0.1943], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 03:09:28,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=948677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:09:46,150 INFO [train.py:968] (1/2) Epoch 21, batch 36350, giga_loss[loss=0.2622, simple_loss=0.3379, pruned_loss=0.09326, over 27975.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3525, pruned_loss=0.09793, over 5704193.08 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.352, pruned_loss=0.1115, over 5693320.85 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3514, pruned_loss=0.09622, over 5693786.81 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:10:06,583 INFO [optim.py:369] (1/2) 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 03:10:24,062 INFO [train.py:968] (1/2) Epoch 21, batch 36400, libri_loss[loss=0.4186, simple_loss=0.4537, pruned_loss=0.1917, over 29322.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3535, pruned_loss=0.09959, over 5705196.69 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3523, pruned_loss=0.1118, over 5702635.54 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3523, pruned_loss=0.09738, over 5688422.43 frames. ], batch size: 94, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:10:30,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.95 vs. limit=5.0
+2023-03-11 03:10:40,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 03:11:02,016 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=948790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:11:08,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5558, 2.2258, 1.6315, 0.8683], device='cuda:1'), covar=tensor([0.6624, 0.3081, 0.4742, 0.6563], device='cuda:1'), in_proj_covar=tensor([0.1744, 0.1643, 0.1602, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 03:11:12,209 INFO [train.py:968] (1/2) Epoch 21, batch 36450, giga_loss[loss=0.3414, simple_loss=0.3938, pruned_loss=0.1445, over 28345.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3564, pruned_loss=0.1038, over 5700241.82 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3521, pruned_loss=0.1117, over 5703764.37 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3556, pruned_loss=0.1021, over 5686189.74 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:11:27,738 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=948820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:11:29,640 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:11:50,710 INFO [train.py:968] (1/2) Epoch 21, batch 36500, libri_loss[loss=0.3066, simple_loss=0.3645, pruned_loss=0.1244, over 29553.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3579, pruned_loss=0.1068, over 5690992.81 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3528, pruned_loss=0.1121, over 5697836.14 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3568, pruned_loss=0.1048, over 5685282.02 frames. ], batch size: 78, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:11:51,640 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 36550, giga_loss[loss=0.2958, simple_loss=0.3639, pruned_loss=0.1139, over 28870.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3562, pruned_loss=0.1066, over 5692280.99 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3526, pruned_loss=0.112, over 5699954.55 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3556, pruned_loss=0.1051, over 5685907.49 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:12:43,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-11 03:13:00,969 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 36600, libri_loss[loss=0.2883, simple_loss=0.3509, pruned_loss=0.1129, over 29572.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3548, pruned_loss=0.1061, over 5704146.59 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3529, pruned_loss=0.1121, over 5703181.69 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3541, pruned_loss=0.1047, over 5696207.51 frames. ], batch size: 79, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:13:21,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5494, 4.9056, 1.7578, 2.0822], device='cuda:1'), covar=tensor([0.0946, 0.0384, 0.0849, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0548, 0.0382, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 03:13:57,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-11 03:13:59,763 INFO [train.py:968] (1/2) Epoch 21, batch 36650, giga_loss[loss=0.2623, simple_loss=0.3405, pruned_loss=0.09203, over 27574.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3535, pruned_loss=0.105, over 5701664.38 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3532, pruned_loss=0.1121, over 5703869.62 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3527, pruned_loss=0.1037, over 5694563.86 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:14:13,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:14:23,637 INFO [optim.py:369] (1/2) 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 03:14:33,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6980, 1.9506, 1.3248, 1.5139], device='cuda:1'), covar=tensor([0.1041, 0.0617, 0.1110, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0516, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 03:14:43,371 INFO [train.py:968] (1/2) Epoch 21, batch 36700, giga_loss[loss=0.2594, simple_loss=0.3374, pruned_loss=0.09076, over 28962.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3525, pruned_loss=0.1037, over 5694911.80 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3537, pruned_loss=0.1123, over 5699559.24 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3514, pruned_loss=0.1022, over 5693730.49 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:14:43,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949051.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:15:28,888 INFO [train.py:968] (1/2) Epoch 21, batch 36750, libri_loss[loss=0.2917, simple_loss=0.3521, pruned_loss=0.1156, over 19584.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.349, pruned_loss=0.1013, over 5682488.81 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3535, pruned_loss=0.1122, over 5694375.37 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3482, pruned_loss=0.09999, over 5687027.17 frames. ], batch size: 186, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:15:54,396 INFO [optim.py:369] (1/2) 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 03:16:17,161 INFO [train.py:968] (1/2) Epoch 21, batch 36800, giga_loss[loss=0.222, simple_loss=0.3041, pruned_loss=0.06994, over 28863.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3436, pruned_loss=0.09808, over 5695179.28 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3539, pruned_loss=0.1124, over 5697714.89 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3426, pruned_loss=0.09666, over 5695680.87 frames. ], batch size: 199, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:16:29,345 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:17:08,455 INFO [train.py:968] (1/2) Epoch 21, batch 36850, giga_loss[loss=0.241, simple_loss=0.3094, pruned_loss=0.08626, over 28731.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3364, pruned_loss=0.09442, over 5669020.80 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3539, pruned_loss=0.1124, over 5689739.81 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3354, pruned_loss=0.09313, over 5677398.71 frames. ], batch size: 92, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:17:13,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4187, 1.7508, 1.5592, 1.5447], device='cuda:1'), covar=tensor([0.0793, 0.0307, 0.0316, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 03:17:40,392 INFO [optim.py:369] (1/2) 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 03:17:57,637 INFO [train.py:968] (1/2) Epoch 21, batch 36900, libri_loss[loss=0.2956, simple_loss=0.3726, pruned_loss=0.1093, over 29258.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3335, pruned_loss=0.09251, over 5674594.11 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3537, pruned_loss=0.1121, over 5694891.63 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3324, pruned_loss=0.09125, over 5675907.09 frames. ], batch size: 94, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:18:39,120 INFO [train.py:968] (1/2) Epoch 21, batch 36950, giga_loss[loss=0.2952, simple_loss=0.3627, pruned_loss=0.1138, over 27913.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3362, pruned_loss=0.09384, over 5671983.19 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3542, pruned_loss=0.1122, over 5689886.37 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3343, pruned_loss=0.09222, over 5676705.39 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:18:44,133 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:18:47,364 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:19:11,130 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=949340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:19:18,641 INFO [train.py:968] (1/2) Epoch 21, batch 37000, giga_loss[loss=0.3134, simple_loss=0.3819, pruned_loss=0.1224, over 28523.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3359, pruned_loss=0.09329, over 5678893.71 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3548, pruned_loss=0.1125, over 5683005.00 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3332, pruned_loss=0.09115, over 5688178.47 frames. ], batch size: 336, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:19:39,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2516, 1.6101, 1.5564, 1.1430], device='cuda:1'), covar=tensor([0.1695, 0.2519, 0.1417, 0.1636], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0698, 0.0946, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 03:19:51,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949390.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:19:59,305 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:19:59,778 INFO [train.py:968] (1/2) Epoch 21, batch 37050, libri_loss[loss=0.3617, simple_loss=0.424, pruned_loss=0.1497, over 29642.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3364, pruned_loss=0.0942, over 5684161.72 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3559, pruned_loss=0.113, over 5687444.35 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3329, pruned_loss=0.09165, over 5687334.59 frames. ], batch size: 91, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:20:20,326 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:20:38,922 INFO [train.py:968] (1/2) Epoch 21, batch 37100, giga_loss[loss=0.3125, simple_loss=0.3783, pruned_loss=0.1233, over 28613.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3354, pruned_loss=0.09392, over 5692399.51 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3565, pruned_loss=0.1132, over 5690783.37 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3316, pruned_loss=0.09129, over 5691853.04 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:20:55,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-11 03:20:57,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3300, 2.5562, 1.4045, 1.4304], device='cuda:1'), covar=tensor([0.1021, 0.0391, 0.0916, 0.1371], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0547, 0.0383, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 03:21:18,401 INFO [train.py:968] (1/2) Epoch 21, batch 37150, giga_loss[loss=0.2418, simple_loss=0.3218, pruned_loss=0.08094, over 28568.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3328, pruned_loss=0.09228, over 5704423.54 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3565, pruned_loss=0.1128, over 5693757.48 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3291, pruned_loss=0.08984, over 5701287.10 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:21:39,310 INFO [optim.py:369] (1/2) 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 03:21:42,840 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949533.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:21:44,734 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 37200, giga_loss[loss=0.2194, simple_loss=0.2954, pruned_loss=0.07168, over 28250.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.331, pruned_loss=0.09153, over 5702670.97 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3567, pruned_loss=0.1128, over 5686629.54 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3273, pruned_loss=0.08921, over 5706183.36 frames. ], batch size: 65, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:22:07,046 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=949565.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:22:09,924 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949569.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:22:12,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=949572.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:22:35,024 INFO [train.py:968] (1/2) Epoch 21, batch 37250, giga_loss[loss=0.2657, simple_loss=0.337, pruned_loss=0.09718, over 27980.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3295, pruned_loss=0.09125, over 5694691.51 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3573, pruned_loss=0.1131, over 5678479.61 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3256, pruned_loss=0.08876, over 5704355.88 frames. ], batch size: 412, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:22:35,283 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=949601.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:22:50,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:22:57,489 INFO [optim.py:369] (1/2) 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 03:23:03,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1870, 2.5729, 1.2259, 1.3889], device='cuda:1'), covar=tensor([0.1097, 0.0360, 0.0967, 0.1429], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0545, 0.0382, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 03:23:15,836 INFO [train.py:968] (1/2) Epoch 21, batch 37300, giga_loss[loss=0.2211, simple_loss=0.3034, pruned_loss=0.06934, over 28687.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3268, pruned_loss=0.08972, over 5700643.83 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.358, pruned_loss=0.1133, over 5680619.51 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3225, pruned_loss=0.08714, over 5706511.21 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:23:55,048 INFO [train.py:968] (1/2) Epoch 21, batch 37350, giga_loss[loss=0.2167, simple_loss=0.2913, pruned_loss=0.07102, over 28521.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3254, pruned_loss=0.08879, over 5711651.74 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3587, pruned_loss=0.1134, over 5686962.07 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3202, pruned_loss=0.08585, over 5711114.35 frames. ], batch size: 71, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:24:11,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6770, 1.8473, 1.5095, 1.7094], device='cuda:1'), covar=tensor([0.2560, 0.2751, 0.2974, 0.2486], device='cuda:1'), in_proj_covar=tensor([0.1506, 0.1089, 0.1325, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 03:24:18,685 INFO [optim.py:369] (1/2) 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 03:24:36,431 INFO [train.py:968] (1/2) Epoch 21, batch 37400, giga_loss[loss=0.2329, simple_loss=0.2968, pruned_loss=0.08454, over 28679.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3226, pruned_loss=0.08707, over 5721717.75 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3591, pruned_loss=0.1134, over 5690360.90 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3176, pruned_loss=0.08428, over 5718877.20 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:24:39,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5273, 2.0898, 1.4903, 0.7840], device='cuda:1'), covar=tensor([0.6098, 0.2710, 0.4492, 0.6751], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1643, 0.1597, 0.1420], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 03:24:55,176 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:25:14,858 INFO [train.py:968] (1/2) Epoch 21, batch 37450, giga_loss[loss=0.2251, simple_loss=0.303, pruned_loss=0.0736, over 29039.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3224, pruned_loss=0.08703, over 5723014.26 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5688749.50 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.317, pruned_loss=0.08402, over 5723251.21 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:25:17,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 03:25:24,252 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:25:32,998 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:25:48,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=949839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:25:48,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3966, 1.5352, 1.6369, 1.2458], device='cuda:1'), covar=tensor([0.1815, 0.2468, 0.1447, 0.1678], device='cuda:1'), in_proj_covar=tensor([0.0904, 0.0701, 0.0952, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 03:25:56,309 INFO [train.py:968] (1/2) Epoch 21, batch 37500, giga_loss[loss=0.3391, simple_loss=0.3975, pruned_loss=0.1403, over 28270.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3247, pruned_loss=0.08854, over 5718912.14 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3602, pruned_loss=0.1138, over 5689587.94 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3193, pruned_loss=0.08554, over 5719385.73 frames. ], batch size: 368, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:26:10,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-11 03:26:43,048 INFO [train.py:968] (1/2) Epoch 21, batch 37550, giga_loss[loss=0.2914, simple_loss=0.3594, pruned_loss=0.1117, over 29072.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3288, pruned_loss=0.09128, over 5716462.09 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5692749.35 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3236, pruned_loss=0.08833, over 5714385.20 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:26:43,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2221, 1.4719, 1.4694, 1.3202], device='cuda:1'), covar=tensor([0.2053, 0.1837, 0.2559, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0749, 0.0717, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 03:26:58,798 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=949918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:27:01,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:27:25,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-11 03:27:30,543 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 37600, giga_loss[loss=0.3404, simple_loss=0.3944, pruned_loss=0.1432, over 27560.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3355, pruned_loss=0.09561, over 5701861.30 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1142, over 5695760.17 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3309, pruned_loss=0.09292, over 5697671.68 frames. ], batch size: 472, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:27:59,207 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 03:28:04,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5967, 1.8260, 1.4820, 1.9562], device='cuda:1'), covar=tensor([0.2373, 0.2449, 0.2567, 0.2152], device='cuda:1'), in_proj_covar=tensor([0.1499, 0.1083, 0.1320, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 03:28:09,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=949995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:28:15,386 INFO [train.py:968] (1/2) Epoch 21, batch 37650, giga_loss[loss=0.278, simple_loss=0.3542, pruned_loss=0.1009, over 29085.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3436, pruned_loss=0.1009, over 5692945.34 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3612, pruned_loss=0.1143, over 5696326.30 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.339, pruned_loss=0.09817, over 5689144.11 frames. ], batch size: 155, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:28:49,353 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 37700, giga_loss[loss=0.2724, simple_loss=0.3511, pruned_loss=0.09687, over 28622.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3476, pruned_loss=0.1024, over 5672388.02 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5687712.03 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3436, pruned_loss=0.09993, over 5677154.63 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:29:12,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 03:29:35,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3210, 3.0990, 2.9582, 1.3013], device='cuda:1'), covar=tensor([0.0886, 0.1089, 0.0900, 0.2518], device='cuda:1'), in_proj_covar=tensor([0.1198, 0.1113, 0.0937, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 03:29:54,550 INFO [train.py:968] (1/2) Epoch 21, batch 37750, libri_loss[loss=0.2515, simple_loss=0.3185, pruned_loss=0.09221, over 29389.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3521, pruned_loss=0.1044, over 5667903.40 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1142, over 5688408.19 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3493, pruned_loss=0.1026, over 5670762.46 frames. ], batch size: 67, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:30:21,367 INFO [optim.py:369] (1/2) 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 03:30:25,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2937, 1.5503, 1.2244, 1.0450], device='cuda:1'), covar=tensor([0.2605, 0.2664, 0.2950, 0.2333], device='cuda:1'), in_proj_covar=tensor([0.1496, 0.1081, 0.1317, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 03:30:26,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:30:27,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:30:34,887 INFO [train.py:968] (1/2) Epoch 21, batch 37800, giga_loss[loss=0.3046, simple_loss=0.3741, pruned_loss=0.1176, over 28999.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3587, pruned_loss=0.1085, over 5681423.46 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1144, over 5695484.99 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3561, pruned_loss=0.1067, over 5676525.91 frames. ], batch size: 136, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:30:51,546 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:31:04,393 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:31:16,910 INFO [train.py:968] (1/2) Epoch 21, batch 37850, giga_loss[loss=0.232, simple_loss=0.3164, pruned_loss=0.07374, over 28966.00 frames. ], tot_loss[loss=0.2825, simple_loss=0.3547, pruned_loss=0.1052, over 5680421.24 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1145, over 5696573.56 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3526, pruned_loss=0.1037, over 5675545.54 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:31:27,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:31:33,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5189, 1.8077, 1.4244, 1.5252], device='cuda:1'), covar=tensor([0.2680, 0.2583, 0.2879, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.1496, 0.1080, 0.1318, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 03:31:40,776 INFO [optim.py:369] (1/2) 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 03:31:55,802 INFO [train.py:968] (1/2) Epoch 21, batch 37900, giga_loss[loss=0.2786, simple_loss=0.3553, pruned_loss=0.101, over 28211.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3522, pruned_loss=0.1028, over 5686571.93 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5701034.42 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3501, pruned_loss=0.1011, over 5678305.39 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:32:15,654 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:32:37,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6792, 1.7978, 1.5054, 1.7944], device='cuda:1'), covar=tensor([0.2641, 0.2804, 0.3054, 0.2365], device='cuda:1'), in_proj_covar=tensor([0.1499, 0.1082, 0.1320, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 03:32:39,560 INFO [train.py:968] (1/2) Epoch 21, batch 37950, giga_loss[loss=0.275, simple_loss=0.3523, pruned_loss=0.09883, over 28940.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3516, pruned_loss=0.1019, over 5689457.68 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3619, pruned_loss=0.1149, over 5702362.85 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3494, pruned_loss=0.1001, over 5681350.31 frames. ], batch size: 145, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:33:04,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:06,454 INFO [optim.py:369] (1/2) 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 03:33:06,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:12,071 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:14,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:22,716 INFO [train.py:968] (1/2) Epoch 21, batch 38000, giga_loss[loss=0.2803, simple_loss=0.3616, pruned_loss=0.09948, over 28862.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3517, pruned_loss=0.1017, over 5691846.87 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1148, over 5702984.46 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3501, pruned_loss=0.1002, over 5684834.05 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:33:27,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:31,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:32,206 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:43,469 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:33:44,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9332, 4.7633, 4.4851, 2.1975], device='cuda:1'), covar=tensor([0.0464, 0.0595, 0.0624, 0.2058], device='cuda:1'), in_proj_covar=tensor([0.1202, 0.1121, 0.0942, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 03:33:56,228 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:34:05,233 INFO [train.py:968] (1/2) Epoch 21, batch 38050, libri_loss[loss=0.2928, simple_loss=0.3649, pruned_loss=0.1103, over 29521.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3537, pruned_loss=0.103, over 5681678.93 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1147, over 5695471.04 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3521, pruned_loss=0.1015, over 5682363.46 frames. ], batch size: 84, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:34:31,895 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 38100, giga_loss[loss=0.2804, simple_loss=0.3522, pruned_loss=0.1043, over 29027.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3548, pruned_loss=0.1039, over 5685091.00 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1145, over 5700621.56 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3537, pruned_loss=0.1027, over 5680823.96 frames. ], batch size: 128, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:34:58,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5952, 4.4539, 4.1832, 1.9888], device='cuda:1'), covar=tensor([0.0562, 0.0668, 0.0659, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.1199, 0.1119, 0.0941, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 03:35:31,728 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:35:32,861 INFO [train.py:968] (1/2) Epoch 21, batch 38150, giga_loss[loss=0.2663, simple_loss=0.3513, pruned_loss=0.09069, over 28981.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3559, pruned_loss=0.1047, over 5696132.10 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1144, over 5701767.75 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3549, pruned_loss=0.1036, over 5691651.35 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:35:59,773 INFO [optim.py:369] (1/2) 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 03:36:13,305 INFO [train.py:968] (1/2) Epoch 21, batch 38200, libri_loss[loss=0.3154, simple_loss=0.3721, pruned_loss=0.1294, over 29521.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3563, pruned_loss=0.1058, over 5692371.64 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1146, over 5702284.55 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3551, pruned_loss=0.1044, over 5687672.58 frames. ], batch size: 79, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:36:52,659 INFO [train.py:968] (1/2) Epoch 21, batch 38250, giga_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1034, over 28771.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3562, pruned_loss=0.1057, over 5703964.48 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3612, pruned_loss=0.1143, over 5707309.03 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3554, pruned_loss=0.1047, over 5695487.10 frames. ], batch size: 119, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:37:19,726 INFO [optim.py:369] (1/2) 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 03:37:24,869 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 03:37:25,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=950639.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:37:30,355 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 38300, giga_loss[loss=0.2461, simple_loss=0.3312, pruned_loss=0.08048, over 28567.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3562, pruned_loss=0.1051, over 5692276.78 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3616, pruned_loss=0.1146, over 5692836.08 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3551, pruned_loss=0.1038, over 5698764.49 frames. ], batch size: 60, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:37:44,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-11 03:38:01,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4056, 4.5265, 1.7151, 1.8283], device='cuda:1'), covar=tensor([0.1110, 0.0213, 0.0921, 0.1302], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0548, 0.0383, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 03:38:03,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8869, 3.7023, 3.4793, 1.6006], device='cuda:1'), covar=tensor([0.0669, 0.0823, 0.0782, 0.2283], device='cuda:1'), in_proj_covar=tensor([0.1200, 0.1121, 0.0944, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 03:38:12,718 INFO [train.py:968] (1/2) Epoch 21, batch 38350, giga_loss[loss=0.2599, simple_loss=0.3485, pruned_loss=0.08563, over 28651.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3554, pruned_loss=0.1032, over 5702752.16 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3616, pruned_loss=0.1144, over 5697919.77 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3544, pruned_loss=0.1021, over 5703369.17 frames. ], batch size: 242, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:38:35,569 INFO [optim.py:369] (1/2) 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 03:38:40,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 03:38:48,052 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 38400, giga_loss[loss=0.26, simple_loss=0.3402, pruned_loss=0.08995, over 28863.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3558, pruned_loss=0.1031, over 5692124.15 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1145, over 5688414.71 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3545, pruned_loss=0.1016, over 5701663.71 frames. ], batch size: 99, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:39:08,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-11 03:39:22,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:39:24,879 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 38450, giga_loss[loss=0.2509, simple_loss=0.3303, pruned_loss=0.08576, over 28928.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3532, pruned_loss=0.1017, over 5688054.92 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.362, pruned_loss=0.1148, over 5682451.89 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3519, pruned_loss=0.1002, over 5701762.66 frames. ], batch size: 112, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:39:46,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=950822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:39:54,551 INFO [optim.py:369] (1/2) 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 03:40:07,745 INFO [train.py:968] (1/2) Epoch 21, batch 38500, giga_loss[loss=0.2953, simple_loss=0.3653, pruned_loss=0.1127, over 28649.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3519, pruned_loss=0.1012, over 5693105.72 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3616, pruned_loss=0.1144, over 5680139.58 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3509, pruned_loss=0.09981, over 5706132.18 frames. ], batch size: 307, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:40:18,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3197, 1.6112, 0.9562, 1.2813], device='cuda:1'), covar=tensor([0.1162, 0.0767, 0.1458, 0.1546], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0444, 0.0515, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 03:40:27,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=950874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:40:41,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=950892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:40:43,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=950895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:40:48,426 INFO [train.py:968] (1/2) Epoch 21, batch 38550, giga_loss[loss=0.2541, simple_loss=0.3362, pruned_loss=0.08598, over 28985.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3504, pruned_loss=0.1005, over 5697385.67 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3621, pruned_loss=0.1148, over 5672838.56 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3491, pruned_loss=0.09888, over 5713739.40 frames. ], batch size: 227, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:41:06,818 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 38600, giga_loss[loss=0.2592, simple_loss=0.3358, pruned_loss=0.09132, over 29020.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3498, pruned_loss=0.1006, over 5698924.34 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3619, pruned_loss=0.1146, over 5676392.30 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3488, pruned_loss=0.0993, over 5709204.95 frames. ], batch size: 106, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:42:09,682 INFO [train.py:968] (1/2) Epoch 21, batch 38650, giga_loss[loss=0.2535, simple_loss=0.3332, pruned_loss=0.08693, over 28704.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3503, pruned_loss=0.1011, over 5704210.12 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1145, over 5677584.99 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3495, pruned_loss=0.09992, over 5711865.90 frames. ], batch size: 60, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:42:11,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7490, 1.0750, 2.8366, 2.6387], device='cuda:1'), covar=tensor([0.1814, 0.2743, 0.0588, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0751, 0.0640, 0.0951, 0.0901], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 03:42:21,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=951014.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:42:23,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=951017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:42:25,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=951020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:42:31,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9588, 5.1069, 2.1329, 2.1466], device='cuda:1'), covar=tensor([0.0965, 0.0238, 0.0844, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0543, 0.0381, 0.0422], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0032, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 03:42:35,256 INFO [optim.py:369] (1/2) 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 03:42:47,822 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 38700, giga_loss[loss=0.277, simple_loss=0.3541, pruned_loss=0.09998, over 28172.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1004, over 5691462.10 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1146, over 5666181.42 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3493, pruned_loss=0.09921, over 5707988.59 frames. ], batch size: 77, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:43:26,091 INFO [train.py:968] (1/2) Epoch 21, batch 38750, giga_loss[loss=0.2616, simple_loss=0.3366, pruned_loss=0.09325, over 28684.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09979, over 5695705.29 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3619, pruned_loss=0.1147, over 5668411.47 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3486, pruned_loss=0.09821, over 5707780.04 frames. ], batch size: 85, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:43:26,279 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=951101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:43:51,439 INFO [optim.py:369] (1/2) 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 03:44:04,491 INFO [train.py:968] (1/2) Epoch 21, batch 38800, giga_loss[loss=0.2678, simple_loss=0.3497, pruned_loss=0.09295, over 28976.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3486, pruned_loss=0.09927, over 5702684.84 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5666915.36 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09742, over 5715344.27 frames. ], batch size: 164, lr: 1.51e-03, grad_scale: 8.0
+2023-03-11 03:44:10,893 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=951157.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:44:13,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=951160.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 03:44:36,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=951189.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 03:44:38,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 03:44:45,160 INFO [train.py:968] (1/2) Epoch 21, batch 38850, giga_loss[loss=0.267, simple_loss=0.3413, pruned_loss=0.09634, over 28459.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3483, pruned_loss=0.09982, over 5676515.90 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3617, pruned_loss=0.1149, over 5645139.10 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3467, pruned_loss=0.0976, over 5708613.11 frames. ], batch size: 60, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:45:10,559 INFO [optim.py:369] (1/2) 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 03:45:20,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-11 03:45:25,617 INFO [train.py:968] (1/2) Epoch 21, batch 38900, giga_loss[loss=0.2569, simple_loss=0.3298, pruned_loss=0.09197, over 28441.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.344, pruned_loss=0.0973, over 5679760.69 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.1151, over 5647746.15 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3424, pruned_loss=0.09525, over 5702850.77 frames. ], batch size: 71, lr: 1.51e-03, grad_scale: 4.0
+2023-03-11 03:46:03,786 INFO [train.py:968] (1/2) Epoch 21, batch 38950, giga_loss[loss=0.2244, simple_loss=0.3087, pruned_loss=0.07004, over 28856.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3411, pruned_loss=0.09569, over 5692423.53 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1148, over 5653208.29 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3396, pruned_loss=0.09387, over 5706688.94 frames. ], batch size: 174, lr: 1.51e-03, grad_scale: 2.0
+2023-03-11 03:46:29,251 INFO [optim.py:369] (1/2) 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 03:46:32,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4798, 1.6102, 1.3488, 1.6084], device='cuda:1'), covar=tensor([0.0748, 0.0328, 0.0345, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0116, 0.0116, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 03:46:42,487 INFO [train.py:968] (1/2) Epoch 21, batch 39000, giga_loss[loss=0.3269, simple_loss=0.3872, pruned_loss=0.1333, over 27640.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3418, pruned_loss=0.09656, over 5696952.36 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3611, pruned_loss=0.1145, over 5657930.01 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3402, pruned_loss=0.09456, over 5706037.26 frames. ], batch size: 472, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:46:42,487 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 03:46:52,169 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 03:47:16,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0845, 3.1309, 2.2324, 1.3773], device='cuda:1'), covar=tensor([0.7718, 0.2862, 0.3698, 0.6398], device='cuda:1'), in_proj_covar=tensor([0.1735, 0.1631, 0.1594, 0.1413], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 03:47:30,438 INFO [train.py:968] (1/2) Epoch 21, batch 39050, giga_loss[loss=0.2631, simple_loss=0.3357, pruned_loss=0.09528, over 28906.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3409, pruned_loss=0.09676, over 5689831.50 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5653668.56 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3389, pruned_loss=0.09458, over 5701179.97 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:47:57,152 INFO [optim.py:369] (1/2) 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 03:48:10,916 INFO [train.py:968] (1/2) Epoch 21, batch 39100, giga_loss[loss=0.2242, simple_loss=0.3039, pruned_loss=0.07224, over 29033.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3376, pruned_loss=0.09498, over 5696433.35 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1147, over 5655967.78 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3359, pruned_loss=0.09314, over 5703650.75 frames. ], batch size: 155, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:48:30,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=951476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:48:49,269 INFO [train.py:968] (1/2) Epoch 21, batch 39150, libri_loss[loss=0.3378, simple_loss=0.3852, pruned_loss=0.1452, over 27716.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3358, pruned_loss=0.09452, over 5705597.28 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1148, over 5657947.13 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.334, pruned_loss=0.09265, over 5710349.69 frames. ], batch size: 115, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 03:48:50,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5504, 2.2111, 1.6370, 0.7353], device='cuda:1'), covar=tensor([0.6421, 0.3074, 0.4369, 0.7286], device='cuda:1'), in_proj_covar=tensor([0.1732, 0.1633, 0.1592, 0.1410], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 03:49:17,151 INFO [optim.py:369] (1/2) 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 03:49:33,889 INFO [train.py:968] (1/2) Epoch 21, batch 39200, giga_loss[loss=0.2753, simple_loss=0.3414, pruned_loss=0.1046, over 28846.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3337, pruned_loss=0.09343, over 5700660.17 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3616, pruned_loss=0.1149, over 5659172.40 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3321, pruned_loss=0.09183, over 5703486.65 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:49:48,832 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=951570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:50:13,957 INFO [train.py:968] (1/2) Epoch 21, batch 39250, giga_loss[loss=0.2851, simple_loss=0.3599, pruned_loss=0.1052, over 28923.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3339, pruned_loss=0.09317, over 5688654.29 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5648288.47 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.332, pruned_loss=0.09147, over 5701705.87 frames. ], batch size: 145, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:50:25,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-11 03:50:28,783 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=951619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:50:31,762 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:50:50,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7434, 3.5895, 3.3999, 2.0369], device='cuda:1'), covar=tensor([0.0574, 0.0726, 0.0681, 0.1754], device='cuda:1'), in_proj_covar=tensor([0.1196, 0.1112, 0.0937, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 03:50:55,104 INFO [train.py:968] (1/2) Epoch 21, batch 39300, giga_loss[loss=0.2558, simple_loss=0.3306, pruned_loss=0.09051, over 28797.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3383, pruned_loss=0.09548, over 5689842.29 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3615, pruned_loss=0.1147, over 5652606.64 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3359, pruned_loss=0.09353, over 5697380.28 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:50:55,402 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=951651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:51:28,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6210, 1.9089, 1.8352, 1.5910], device='cuda:1'), covar=tensor([0.2089, 0.2098, 0.2296, 0.2381], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0749, 0.0713, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 03:51:40,526 INFO [train.py:968] (1/2) Epoch 21, batch 39350, giga_loss[loss=0.2634, simple_loss=0.3467, pruned_loss=0.09, over 27887.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3413, pruned_loss=0.09673, over 5688401.69 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5656311.52 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3393, pruned_loss=0.09501, over 5691788.67 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:52:08,823 INFO [optim.py:369] (1/2) 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 03:52:21,816 INFO [train.py:968] (1/2) Epoch 21, batch 39400, giga_loss[loss=0.2593, simple_loss=0.3488, pruned_loss=0.08491, over 28911.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.344, pruned_loss=0.09755, over 5688247.84 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3611, pruned_loss=0.1144, over 5653646.79 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3421, pruned_loss=0.0959, over 5693992.54 frames. ], batch size: 227, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:52:32,426 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-11 03:53:05,265 INFO [train.py:968] (1/2) Epoch 21, batch 39450, giga_loss[loss=0.2444, simple_loss=0.3286, pruned_loss=0.08016, over 28358.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3436, pruned_loss=0.09641, over 5692641.17 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5658287.77 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3414, pruned_loss=0.09453, over 5693636.59 frames. ], batch size: 369, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:53:07,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 03:53:32,262 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=951829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:53:36,780 INFO [optim.py:369] (1/2) 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 03:53:46,850 INFO [train.py:968] (1/2) Epoch 21, batch 39500, libri_loss[loss=0.3315, simple_loss=0.3839, pruned_loss=0.1395, over 19568.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3417, pruned_loss=0.09536, over 5693304.95 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3619, pruned_loss=0.115, over 5654093.53 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3392, pruned_loss=0.09311, over 5699458.38 frames. ], batch size: 187, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:54:26,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8855, 2.0193, 2.0663, 1.6394], device='cuda:1'), covar=tensor([0.1620, 0.2540, 0.1420, 0.1780], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0699, 0.0947, 0.0845], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 03:54:26,466 INFO [train.py:968] (1/2) Epoch 21, batch 39550, libri_loss[loss=0.2987, simple_loss=0.3704, pruned_loss=0.1135, over 27822.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3433, pruned_loss=0.09696, over 5693070.01 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1148, over 5653909.84 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3408, pruned_loss=0.09462, over 5699430.14 frames. ], batch size: 116, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:54:53,890 INFO [optim.py:369] (1/2) 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 03:55:01,490 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 39600, giga_loss[loss=0.2657, simple_loss=0.3453, pruned_loss=0.09307, over 28751.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3432, pruned_loss=0.09688, over 5710090.68 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1149, over 5659483.84 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3406, pruned_loss=0.09457, over 5711069.20 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 03:55:41,535 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-11 03:55:48,643 INFO [train.py:968] (1/2) Epoch 21, batch 39650, libri_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.1109, over 29575.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.346, pruned_loss=0.09828, over 5712315.06 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.115, over 5667993.61 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3432, pruned_loss=0.09579, over 5706838.57 frames. ], batch size: 78, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 03:55:51,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2587, 1.3332, 3.7561, 3.3696], device='cuda:1'), covar=tensor([0.1693, 0.2815, 0.0414, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0641, 0.0954, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 03:56:17,562 INFO [optim.py:369] (1/2) 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 03:56:29,995 INFO [train.py:968] (1/2) Epoch 21, batch 39700, giga_loss[loss=0.2717, simple_loss=0.3473, pruned_loss=0.09811, over 28796.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3504, pruned_loss=0.1005, over 5713996.26 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3626, pruned_loss=0.1151, over 5675433.34 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3473, pruned_loss=0.09785, over 5704031.21 frames. ], batch size: 186, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:56:43,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4683, 1.5915, 1.6465, 1.4228], device='cuda:1'), covar=tensor([0.2033, 0.2388, 0.2351, 0.2361], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0752, 0.0716, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 03:56:47,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3858, 1.6618, 1.3243, 1.3045], device='cuda:1'), covar=tensor([0.2594, 0.2718, 0.3098, 0.2422], device='cuda:1'), in_proj_covar=tensor([0.1500, 0.1085, 0.1324, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 03:56:59,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=952088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:57:02,090 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 39750, giga_loss[loss=0.2761, simple_loss=0.3431, pruned_loss=0.1045, over 28564.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3522, pruned_loss=0.1016, over 5712232.03 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3628, pruned_loss=0.1152, over 5669916.56 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.09917, over 5709739.10 frames. ], batch size: 78, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:57:10,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5737, 3.1151, 1.5738, 1.6765], device='cuda:1'), covar=tensor([0.0752, 0.0298, 0.0740, 0.1047], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0547, 0.0382, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 03:57:24,891 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=952120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:57:36,055 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-11 03:57:37,432 INFO [optim.py:369] (1/2) 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 03:57:50,161 INFO [train.py:968] (1/2) Epoch 21, batch 39800, giga_loss[loss=0.3117, simple_loss=0.3854, pruned_loss=0.119, over 28942.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3526, pruned_loss=0.1014, over 5710237.94 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3626, pruned_loss=0.115, over 5673727.33 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3503, pruned_loss=0.09948, over 5705438.16 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:58:31,942 INFO [train.py:968] (1/2) Epoch 21, batch 39850, libri_loss[loss=0.2488, simple_loss=0.3223, pruned_loss=0.08761, over 29628.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3533, pruned_loss=0.1015, over 5705677.00 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3627, pruned_loss=0.1149, over 5671202.02 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3512, pruned_loss=0.09972, over 5704910.36 frames. ], batch size: 69, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:58:34,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 03:58:46,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2606, 2.3119, 2.1809, 2.1168], device='cuda:1'), covar=tensor([0.1980, 0.2505, 0.2214, 0.2253], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0751, 0.0716, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 03:58:47,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5677, 1.2615, 4.6992, 3.6101], device='cuda:1'), covar=tensor([0.1672, 0.2907, 0.0375, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0642, 0.0957, 0.0908], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 03:58:55,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1527, 1.7779, 1.3675, 0.4834], device='cuda:1'), covar=tensor([0.3361, 0.2045, 0.3105, 0.4057], device='cuda:1'), in_proj_covar=tensor([0.1738, 0.1638, 0.1601, 0.1416], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 03:58:59,314 INFO [optim.py:369] (1/2) 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 03:59:02,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3052, 1.5510, 1.5212, 1.1951], device='cuda:1'), covar=tensor([0.3924, 0.2631, 0.2227, 0.3062], device='cuda:1'), in_proj_covar=tensor([0.1963, 0.1890, 0.1814, 0.1950], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 03:59:12,695 INFO [train.py:968] (1/2) Epoch 21, batch 39900, giga_loss[loss=0.2718, simple_loss=0.3452, pruned_loss=0.09914, over 28716.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3537, pruned_loss=0.102, over 5700839.68 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3633, pruned_loss=0.1153, over 5662852.58 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3513, pruned_loss=0.1001, over 5707722.40 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 03:59:49,259 INFO [train.py:968] (1/2) Epoch 21, batch 39950, giga_loss[loss=0.3039, simple_loss=0.3673, pruned_loss=0.1202, over 27569.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3527, pruned_loss=0.1022, over 5709825.10 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1161, over 5670655.04 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3496, pruned_loss=0.0995, over 5709722.82 frames. ], batch size: 472, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:00:16,029 INFO [optim.py:369] (1/2) 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 04:00:26,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=952347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:00:29,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=952350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:00:29,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3469, 1.6569, 1.3236, 1.0330], device='cuda:1'), covar=tensor([0.2720, 0.2739, 0.3246, 0.2444], device='cuda:1'), in_proj_covar=tensor([0.1499, 0.1085, 0.1323, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 04:00:29,623 INFO [train.py:968] (1/2) Epoch 21, batch 40000, giga_loss[loss=0.2633, simple_loss=0.3345, pruned_loss=0.09604, over 28785.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3494, pruned_loss=0.1007, over 5715739.00 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3648, pruned_loss=0.1164, over 5678118.05 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3461, pruned_loss=0.09778, over 5710468.47 frames. ], batch size: 99, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:00:53,423 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=952379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:01:08,203 INFO [train.py:968] (1/2) Epoch 21, batch 40050, giga_loss[loss=0.2586, simple_loss=0.3401, pruned_loss=0.08858, over 28858.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3464, pruned_loss=0.09919, over 5720534.65 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3651, pruned_loss=0.1167, over 5685186.27 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3428, pruned_loss=0.09591, over 5711074.26 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:01:34,710 INFO [optim.py:369] (1/2) 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 04:01:48,182 INFO [train.py:968] (1/2) Epoch 21, batch 40100, giga_loss[loss=0.2126, simple_loss=0.2968, pruned_loss=0.06421, over 28801.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3469, pruned_loss=0.09822, over 5723163.08 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3648, pruned_loss=0.1166, over 5687285.90 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3438, pruned_loss=0.09531, over 5714450.83 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:01:54,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=952460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:02:27,307 INFO [train.py:968] (1/2) Epoch 21, batch 40150, giga_loss[loss=0.2438, simple_loss=0.3317, pruned_loss=0.07793, over 28906.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3475, pruned_loss=0.09723, over 5715681.50 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3644, pruned_loss=0.1163, over 5691602.14 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.345, pruned_loss=0.09479, over 5705755.97 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:02:30,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=952503.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:02:30,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8774, 1.3181, 1.3124, 1.0592], device='cuda:1'), covar=tensor([0.1952, 0.1244, 0.2437, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0751, 0.0716, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 04:02:56,464 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 40200, giga_loss[loss=0.3258, simple_loss=0.3828, pruned_loss=0.1344, over 26752.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09637, over 5719233.36 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1162, over 5696839.35 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3433, pruned_loss=0.09407, over 5707106.57 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:03:45,272 INFO [train.py:968] (1/2) Epoch 21, batch 40250, giga_loss[loss=0.3568, simple_loss=0.4051, pruned_loss=0.1542, over 28908.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3465, pruned_loss=0.09851, over 5714709.81 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1162, over 5697331.06 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3443, pruned_loss=0.09621, over 5704959.16 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:03:50,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=952608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:04:12,669 INFO [optim.py:369] (1/2) 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 04:04:24,140 INFO [train.py:968] (1/2) Epoch 21, batch 40300, giga_loss[loss=0.265, simple_loss=0.3313, pruned_loss=0.09938, over 28504.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3457, pruned_loss=0.09957, over 5711990.69 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1162, over 5699642.03 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3436, pruned_loss=0.09755, over 5702441.85 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:04:41,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-11 04:04:50,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6269, 1.8209, 1.4794, 1.9138], device='cuda:1'), covar=tensor([0.2602, 0.2835, 0.3152, 0.2602], device='cuda:1'), in_proj_covar=tensor([0.1502, 0.1087, 0.1324, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 04:05:05,669 INFO [train.py:968] (1/2) Epoch 21, batch 40350, giga_loss[loss=0.2395, simple_loss=0.3151, pruned_loss=0.08194, over 28913.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3435, pruned_loss=0.09927, over 5693775.16 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1167, over 5673489.41 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.341, pruned_loss=0.09692, over 5709952.89 frames. ], batch size: 145, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:05:33,457 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 40400, giga_loss[loss=0.2644, simple_loss=0.3295, pruned_loss=0.09965, over 28666.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3419, pruned_loss=0.09869, over 5692968.20 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3648, pruned_loss=0.1167, over 5664093.55 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3397, pruned_loss=0.09645, over 5714711.53 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:06:26,522 INFO [train.py:968] (1/2) Epoch 21, batch 40450, giga_loss[loss=0.2371, simple_loss=0.3122, pruned_loss=0.08107, over 28907.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3383, pruned_loss=0.09623, over 5700377.72 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3644, pruned_loss=0.1165, over 5666627.53 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3367, pruned_loss=0.09449, over 5715511.57 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:06:45,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3196, 1.6080, 1.2983, 1.1158], device='cuda:1'), covar=tensor([0.2672, 0.2685, 0.3130, 0.2333], device='cuda:1'), in_proj_covar=tensor([0.1498, 0.1083, 0.1321, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 04:06:54,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:06:56,929 INFO [optim.py:369] (1/2) 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 04:07:07,289 INFO [train.py:968] (1/2) Epoch 21, batch 40500, giga_loss[loss=0.2238, simple_loss=0.2974, pruned_loss=0.07509, over 28325.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3351, pruned_loss=0.09504, over 5704119.59 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3651, pruned_loss=0.1171, over 5667930.51 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3328, pruned_loss=0.09287, over 5715252.19 frames. ], batch size: 65, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:07:11,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2395, 1.1168, 3.6617, 3.0604], device='cuda:1'), covar=tensor([0.1629, 0.2825, 0.0456, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0642, 0.0954, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 04:07:26,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952878.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:07:43,822 INFO [train.py:968] (1/2) Epoch 21, batch 40550, giga_loss[loss=0.258, simple_loss=0.3256, pruned_loss=0.09513, over 28703.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3323, pruned_loss=0.09369, over 5707296.93 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.3651, pruned_loss=0.1171, over 5668035.12 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3293, pruned_loss=0.09109, over 5717984.22 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:07:55,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4942, 1.5872, 1.2200, 1.2383], device='cuda:1'), covar=tensor([0.0922, 0.0570, 0.1117, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0448, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 04:08:12,813 INFO [optim.py:369] (1/2) 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 04:08:23,327 INFO [train.py:968] (1/2) Epoch 21, batch 40600, giga_loss[loss=0.2382, simple_loss=0.3271, pruned_loss=0.07464, over 29012.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3339, pruned_loss=0.0944, over 5708355.88 frames. ], libri_tot_loss[loss=0.3, simple_loss=0.3654, pruned_loss=0.1173, over 5673404.47 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3305, pruned_loss=0.09158, over 5713013.94 frames. ], batch size: 164, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:08:45,196 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=952978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:08:47,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=952981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:08:49,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:08:53,533 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-11 04:08:59,059 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 40650, giga_loss[loss=0.2504, simple_loss=0.3288, pruned_loss=0.086, over 29009.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3368, pruned_loss=0.09563, over 5698886.63 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3655, pruned_loss=0.1176, over 5661886.72 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3329, pruned_loss=0.09231, over 5713313.78 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:09:11,675 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:09:19,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953021.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:09:21,122 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:09:42,734 INFO [train.py:968] (1/2) Epoch 21, batch 40700, giga_loss[loss=0.2904, simple_loss=0.362, pruned_loss=0.1094, over 28189.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3392, pruned_loss=0.09595, over 5697632.59 frames. ], libri_tot_loss[loss=0.3003, simple_loss=0.3655, pruned_loss=0.1175, over 5665446.10 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3355, pruned_loss=0.09301, over 5706669.21 frames. ], batch size: 77, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:09:44,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953053.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:10:19,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 04:10:21,019 INFO [train.py:968] (1/2) Epoch 21, batch 40750, giga_loss[loss=0.2774, simple_loss=0.3576, pruned_loss=0.09866, over 28808.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.343, pruned_loss=0.09735, over 5709403.27 frames. ], libri_tot_loss[loss=0.3007, simple_loss=0.3658, pruned_loss=0.1177, over 5669529.74 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3394, pruned_loss=0.09444, over 5713539.50 frames. ], batch size: 243, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:10:43,461 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:10:45,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=953129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:10:51,520 INFO [optim.py:369] (1/2) 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 04:11:01,479 INFO [train.py:968] (1/2) Epoch 21, batch 40800, giga_loss[loss=0.2588, simple_loss=0.3336, pruned_loss=0.09198, over 28660.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3456, pruned_loss=0.09863, over 5713616.32 frames. ], libri_tot_loss[loss=0.3004, simple_loss=0.3656, pruned_loss=0.1176, over 5674069.86 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3424, pruned_loss=0.09594, over 5713813.24 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:11:07,285 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:11:40,081 INFO [train.py:968] (1/2) Epoch 21, batch 40850, libri_loss[loss=0.2368, simple_loss=0.3018, pruned_loss=0.08589, over 29625.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3465, pruned_loss=0.09946, over 5712225.01 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3648, pruned_loss=0.1171, over 5681016.78 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3439, pruned_loss=0.09698, over 5707270.13 frames. ], batch size: 69, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:12:08,633 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.97 vs. limit=5.0
+2023-03-11 04:12:19,140 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 40900, giga_loss[loss=0.3659, simple_loss=0.4281, pruned_loss=0.1519, over 27942.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3544, pruned_loss=0.1062, over 5692920.69 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3652, pruned_loss=0.1173, over 5684008.86 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3517, pruned_loss=0.1038, over 5686695.70 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:13:13,615 INFO [train.py:968] (1/2) Epoch 21, batch 40950, giga_loss[loss=0.3666, simple_loss=0.4048, pruned_loss=0.1642, over 23644.00 frames. ], tot_loss[loss=0.29, simple_loss=0.3594, pruned_loss=0.1103, over 5684664.46 frames. ], libri_tot_loss[loss=0.2995, simple_loss=0.3649, pruned_loss=0.1171, over 5685528.49 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3572, pruned_loss=0.1082, over 5678826.47 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:13:47,846 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 41000, giga_loss[loss=0.3124, simple_loss=0.3878, pruned_loss=0.1185, over 28949.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3665, pruned_loss=0.1155, over 5682817.65 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3653, pruned_loss=0.1175, over 5686760.44 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3643, pruned_loss=0.1134, over 5677245.17 frames. ], batch size: 213, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:14:17,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=953369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:14:18,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5458, 1.5355, 1.7363, 1.3303], device='cuda:1'), covar=tensor([0.1314, 0.2230, 0.1160, 0.1499], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0698, 0.0943, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 04:14:45,202 INFO [train.py:968] (1/2) Epoch 21, batch 41050, giga_loss[loss=0.3055, simple_loss=0.3738, pruned_loss=0.1186, over 28990.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3736, pruned_loss=0.1216, over 5677626.23 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3655, pruned_loss=0.1175, over 5688223.08 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3718, pruned_loss=0.12, over 5671873.62 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:14:49,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5987, 1.6958, 1.5773, 1.5257], device='cuda:1'), covar=tensor([0.2205, 0.2191, 0.2033, 0.2038], device='cuda:1'), in_proj_covar=tensor([0.1977, 0.1911, 0.1833, 0.1967], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 04:14:51,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2425, 1.4942, 1.5364, 1.3720], device='cuda:1'), covar=tensor([0.1816, 0.1667, 0.2229, 0.1709], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0755, 0.0719, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 04:15:14,735 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 41100, giga_loss[loss=0.4061, simple_loss=0.4395, pruned_loss=0.1864, over 27559.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3793, pruned_loss=0.1264, over 5663980.35 frames. ], libri_tot_loss[loss=0.3008, simple_loss=0.366, pruned_loss=0.1178, over 5677240.92 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3777, pruned_loss=0.125, over 5669806.43 frames. ], batch size: 472, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:15:32,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-11 04:16:14,943 INFO [train.py:968] (1/2) Epoch 21, batch 41150, giga_loss[loss=0.3159, simple_loss=0.3817, pruned_loss=0.1251, over 28776.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3812, pruned_loss=0.1288, over 5649683.12 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.1171, over 5683215.21 frames. ], giga_tot_loss[loss=0.3195, simple_loss=0.3813, pruned_loss=0.1288, over 5648103.38 frames. ], batch size: 284, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:16:16,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=953503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:22,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6109, 4.4131, 4.1968, 2.0054], device='cuda:1'), covar=tensor([0.0651, 0.0833, 0.0949, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1215, 0.1129, 0.0955, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 04:16:25,472 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:29,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=953515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:52,927 INFO [optim.py:369] (1/2) 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 04:16:56,580 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:16:58,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2532, 1.2271, 3.4320, 3.0879], device='cuda:1'), covar=tensor([0.1470, 0.2591, 0.0463, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.0754, 0.0643, 0.0957, 0.0906], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 04:17:04,750 INFO [train.py:968] (1/2) Epoch 21, batch 41200, giga_loss[loss=0.4495, simple_loss=0.4548, pruned_loss=0.2221, over 26598.00 frames. ], tot_loss[loss=0.3234, simple_loss=0.3837, pruned_loss=0.1316, over 5658123.30 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3655, pruned_loss=0.1174, over 5689179.13 frames. ], giga_tot_loss[loss=0.3236, simple_loss=0.384, pruned_loss=0.1316, over 5650808.10 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:17:30,175 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 04:17:58,453 INFO [train.py:968] (1/2) Epoch 21, batch 41250, giga_loss[loss=0.41, simple_loss=0.4255, pruned_loss=0.1972, over 23446.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3859, pruned_loss=0.1347, over 5632982.47 frames. ], libri_tot_loss[loss=0.3002, simple_loss=0.3655, pruned_loss=0.1175, over 5693407.07 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3864, pruned_loss=0.1349, over 5622391.65 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:18:34,262 INFO [optim.py:369] (1/2) 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 04:18:43,403 INFO [train.py:968] (1/2) Epoch 21, batch 41300, libri_loss[loss=0.2644, simple_loss=0.3187, pruned_loss=0.105, over 29350.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3885, pruned_loss=0.1375, over 5644884.98 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3647, pruned_loss=0.1169, over 5701723.74 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3905, pruned_loss=0.1389, over 5626590.45 frames. ], batch size: 67, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:19:15,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0815, 1.2581, 3.7232, 3.1003], device='cuda:1'), covar=tensor([0.1864, 0.2727, 0.0478, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0644, 0.0960, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 04:19:33,267 INFO [train.py:968] (1/2) Epoch 21, batch 41350, giga_loss[loss=0.3601, simple_loss=0.4025, pruned_loss=0.1588, over 28429.00 frames. ], tot_loss[loss=0.3361, simple_loss=0.3921, pruned_loss=0.1401, over 5642967.12 frames. ], libri_tot_loss[loss=0.2998, simple_loss=0.3652, pruned_loss=0.1172, over 5701686.72 frames. ], giga_tot_loss[loss=0.3382, simple_loss=0.3938, pruned_loss=0.1413, over 5627464.95 frames. ], batch size: 369, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:20:09,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3853, 3.4056, 1.5403, 1.4820], device='cuda:1'), covar=tensor([0.0947, 0.0302, 0.0852, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0552, 0.0384, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 04:20:17,016 INFO [optim.py:369] (1/2) 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 04:20:25,904 INFO [train.py:968] (1/2) Epoch 21, batch 41400, giga_loss[loss=0.3538, simple_loss=0.4021, pruned_loss=0.1528, over 28646.00 frames. ], tot_loss[loss=0.3356, simple_loss=0.3912, pruned_loss=0.14, over 5637910.45 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3648, pruned_loss=0.117, over 5696269.25 frames. ], giga_tot_loss[loss=0.3386, simple_loss=0.3936, pruned_loss=0.1418, over 5628502.86 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:20:50,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 04:21:09,131 INFO [train.py:968] (1/2) Epoch 21, batch 41450, libri_loss[loss=0.2347, simple_loss=0.3103, pruned_loss=0.07955, over 29562.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3884, pruned_loss=0.1383, over 5645462.53 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3649, pruned_loss=0.117, over 5704979.85 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3914, pruned_loss=0.1407, over 5627580.05 frames. ], batch size: 76, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:21:17,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=953809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:21:47,497 INFO [optim.py:369] (1/2) 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 04:21:57,357 INFO [train.py:968] (1/2) Epoch 21, batch 41500, giga_loss[loss=0.2783, simple_loss=0.3558, pruned_loss=0.1003, over 29082.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3891, pruned_loss=0.1383, over 5637087.18 frames. ], libri_tot_loss[loss=0.2996, simple_loss=0.365, pruned_loss=0.1171, over 5700419.79 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3917, pruned_loss=0.1404, over 5626040.81 frames. ], batch size: 128, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:22:28,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3365, 3.0071, 1.4241, 1.4358], device='cuda:1'), covar=tensor([0.1027, 0.0346, 0.0923, 0.1442], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0553, 0.0384, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 04:22:28,743 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 41550, giga_loss[loss=0.3587, simple_loss=0.4134, pruned_loss=0.1519, over 28919.00 frames. ], tot_loss[loss=0.3343, simple_loss=0.3906, pruned_loss=0.139, over 5617046.66 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1171, over 5703208.56 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3929, pruned_loss=0.141, over 5604840.22 frames. ], batch size: 227, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:23:33,791 INFO [optim.py:369] (1/2) 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 04:23:43,402 INFO [train.py:968] (1/2) Epoch 21, batch 41600, giga_loss[loss=0.299, simple_loss=0.3736, pruned_loss=0.1122, over 28910.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3922, pruned_loss=0.1403, over 5592998.95 frames. ], libri_tot_loss[loss=0.2999, simple_loss=0.3653, pruned_loss=0.1172, over 5696206.33 frames. ], giga_tot_loss[loss=0.3392, simple_loss=0.3943, pruned_loss=0.1421, over 5588692.90 frames. ], batch size: 174, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:23:45,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=953952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:23:49,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=953955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:24:18,812 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=953984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:24:32,839 INFO [train.py:968] (1/2) Epoch 21, batch 41650, giga_loss[loss=0.2889, simple_loss=0.3649, pruned_loss=0.1064, over 28957.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3874, pruned_loss=0.1363, over 5604139.23 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3645, pruned_loss=0.1169, over 5694195.22 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3908, pruned_loss=0.1388, over 5599096.21 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:24:53,130 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=954021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:24:56,336 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=954024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:25:09,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-11 04:25:09,539 INFO [optim.py:369] (1/2) 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 04:25:20,341 INFO [train.py:968] (1/2) Epoch 21, batch 41700, giga_loss[loss=0.2861, simple_loss=0.3557, pruned_loss=0.1082, over 28021.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3848, pruned_loss=0.1327, over 5615632.13 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3642, pruned_loss=0.1167, over 5688196.03 frames. ], giga_tot_loss[loss=0.3296, simple_loss=0.3884, pruned_loss=0.1354, over 5613965.21 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:25:22,247 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=954053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:25:32,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8238, 1.0498, 2.8565, 2.7549], device='cuda:1'), covar=tensor([0.1695, 0.2632, 0.0612, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0759, 0.0647, 0.0964, 0.0913], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 04:26:02,907 INFO [train.py:968] (1/2) Epoch 21, batch 41750, giga_loss[loss=0.3478, simple_loss=0.3931, pruned_loss=0.1513, over 27572.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3818, pruned_loss=0.13, over 5626837.01 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3637, pruned_loss=0.1165, over 5691634.82 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3862, pruned_loss=0.1332, over 5619159.58 frames. ], batch size: 472, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:26:39,878 INFO [optim.py:369] (1/2) 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 04:26:43,944 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:26:49,605 INFO [train.py:968] (1/2) Epoch 21, batch 41800, giga_loss[loss=0.3284, simple_loss=0.3752, pruned_loss=0.1408, over 23934.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3795, pruned_loss=0.1281, over 5630258.70 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.364, pruned_loss=0.1167, over 5695959.87 frames. ], giga_tot_loss[loss=0.3224, simple_loss=0.3832, pruned_loss=0.1308, over 5618881.14 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:26:52,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-11 04:27:13,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3650, 3.4201, 1.4118, 1.6306], device='cuda:1'), covar=tensor([0.0966, 0.0414, 0.0937, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0553, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 04:27:40,378 INFO [train.py:968] (1/2) Epoch 21, batch 41850, giga_loss[loss=0.3226, simple_loss=0.3733, pruned_loss=0.136, over 28591.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3771, pruned_loss=0.1261, over 5636931.28 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3641, pruned_loss=0.1167, over 5698132.40 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.38, pruned_loss=0.1283, over 5625352.69 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:28:20,127 INFO [optim.py:369] (1/2) 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 04:28:28,804 INFO [train.py:968] (1/2) Epoch 21, batch 41900, giga_loss[loss=0.3128, simple_loss=0.3796, pruned_loss=0.123, over 28871.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3767, pruned_loss=0.1257, over 5645418.68 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3643, pruned_loss=0.1168, over 5702456.35 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.379, pruned_loss=0.1275, over 5631262.32 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:29:02,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3573, 3.3101, 1.4628, 1.5638], device='cuda:1'), covar=tensor([0.0997, 0.0325, 0.0910, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0555, 0.0385, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 04:29:17,975 INFO [train.py:968] (1/2) Epoch 21, batch 41950, giga_loss[loss=0.2878, simple_loss=0.3662, pruned_loss=0.1047, over 28876.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.375, pruned_loss=0.1243, over 5645152.06 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3638, pruned_loss=0.1165, over 5704643.10 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3773, pruned_loss=0.1261, over 5631450.41 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:29:58,642 INFO [optim.py:369] (1/2) 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 04:30:06,114 INFO [train.py:968] (1/2) Epoch 21, batch 42000, giga_loss[loss=0.2657, simple_loss=0.3429, pruned_loss=0.09426, over 28937.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3735, pruned_loss=0.1227, over 5650156.20 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3638, pruned_loss=0.1165, over 5708245.63 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3757, pruned_loss=0.1243, over 5634387.85 frames. ], batch size: 213, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:30:06,114 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 04:30:15,117 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 04:30:49,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5832, 2.1654, 1.8549, 1.5963], device='cuda:1'), covar=tensor([0.0712, 0.0259, 0.0282, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 04:31:01,574 INFO [train.py:968] (1/2) Epoch 21, batch 42050, giga_loss[loss=0.2972, simple_loss=0.365, pruned_loss=0.1147, over 28931.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3749, pruned_loss=0.1211, over 5654581.89 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3637, pruned_loss=0.1166, over 5708565.98 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.377, pruned_loss=0.1224, over 5639554.15 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:31:42,652 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 42100, giga_loss[loss=0.308, simple_loss=0.3733, pruned_loss=0.1213, over 28551.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3756, pruned_loss=0.1205, over 5663943.10 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3633, pruned_loss=0.1162, over 5713773.80 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.378, pruned_loss=0.122, over 5646145.67 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:32:40,800 INFO [train.py:968] (1/2) Epoch 21, batch 42150, libri_loss[loss=0.2861, simple_loss=0.3622, pruned_loss=0.105, over 29271.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3756, pruned_loss=0.1209, over 5670983.30 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3632, pruned_loss=0.1162, over 5717214.53 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3778, pruned_loss=0.1222, over 5652637.49 frames. ], batch size: 94, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:32:53,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2806, 1.2817, 1.2963, 1.4102], device='cuda:1'), covar=tensor([0.0791, 0.0357, 0.0325, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 04:32:56,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=954521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:33:16,095 INFO [optim.py:369] (1/2) 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 04:33:24,246 INFO [train.py:968] (1/2) Epoch 21, batch 42200, giga_loss[loss=0.2934, simple_loss=0.368, pruned_loss=0.1094, over 28468.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3755, pruned_loss=0.1213, over 5663690.23 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3635, pruned_loss=0.1163, over 5711753.65 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3773, pruned_loss=0.1223, over 5652711.79 frames. ], batch size: 78, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:34:06,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:34:07,180 INFO [train.py:968] (1/2) Epoch 21, batch 42250, libri_loss[loss=0.3139, simple_loss=0.3844, pruned_loss=0.1218, over 29691.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3729, pruned_loss=0.1205, over 5679447.78 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.363, pruned_loss=0.1158, over 5718239.17 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3751, pruned_loss=0.1219, over 5663371.02 frames. ], batch size: 91, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:34:11,860 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:34:37,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:34:45,236 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 42300, libri_loss[loss=0.2693, simple_loss=0.3374, pruned_loss=0.1006, over 29564.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3717, pruned_loss=0.1208, over 5674358.77 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3622, pruned_loss=0.1153, over 5721528.09 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3745, pruned_loss=0.1227, over 5656851.50 frames. ], batch size: 76, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:34:59,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6117, 1.9500, 1.4938, 1.8535], device='cuda:1'), covar=tensor([0.2703, 0.2736, 0.3162, 0.2425], device='cuda:1'), in_proj_covar=tensor([0.1496, 0.1083, 0.1321, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 04:35:04,484 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=954664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:35:06,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=954667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:35:10,440 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:35:25,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 04:35:39,594 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 42350, giga_loss[loss=0.3481, simple_loss=0.4085, pruned_loss=0.1439, over 28562.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1188, over 5674713.23 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5723490.56 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3727, pruned_loss=0.1204, over 5658964.24 frames. ], batch size: 336, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:36:20,236 INFO [optim.py:369] (1/2) 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 04:36:30,255 INFO [train.py:968] (1/2) Epoch 21, batch 42400, giga_loss[loss=0.3172, simple_loss=0.3775, pruned_loss=0.1284, over 28956.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3708, pruned_loss=0.1181, over 5680686.34 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.115, over 5721676.43 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3729, pruned_loss=0.1196, over 5668765.64 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:37:16,258 INFO [train.py:968] (1/2) Epoch 21, batch 42450, giga_loss[loss=0.3249, simple_loss=0.3892, pruned_loss=0.1303, over 28662.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3708, pruned_loss=0.118, over 5677935.47 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3621, pruned_loss=0.1149, over 5724411.94 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3727, pruned_loss=0.1193, over 5665005.33 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:37:17,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:37:47,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 04:37:55,307 INFO [optim.py:369] (1/2) 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 04:38:01,952 INFO [train.py:968] (1/2) Epoch 21, batch 42500, libri_loss[loss=0.3045, simple_loss=0.3681, pruned_loss=0.1205, over 29534.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.371, pruned_loss=0.1186, over 5673726.95 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5718017.08 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3723, pruned_loss=0.1195, over 5667238.07 frames. ], batch size: 81, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:38:45,143 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 04:38:49,253 INFO [train.py:968] (1/2) Epoch 21, batch 42550, giga_loss[loss=0.2901, simple_loss=0.3602, pruned_loss=0.11, over 28761.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3685, pruned_loss=0.1175, over 5675249.98 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.1151, over 5716379.24 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3699, pruned_loss=0.1184, over 5670818.77 frames. ], batch size: 284, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:38:54,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5309, 1.6924, 1.6773, 1.5999], device='cuda:1'), covar=tensor([0.2041, 0.2289, 0.2370, 0.2137], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0755, 0.0716, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 04:39:08,932 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=954923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:39:14,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7547, 1.8212, 1.3651, 1.3440], device='cuda:1'), covar=tensor([0.0898, 0.0564, 0.1022, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0448, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 04:39:30,916 INFO [optim.py:369] (1/2) 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 04:39:37,651 INFO [train.py:968] (1/2) Epoch 21, batch 42600, giga_loss[loss=0.3018, simple_loss=0.3632, pruned_loss=0.1202, over 28593.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5671931.49 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.1151, over 5719809.87 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3686, pruned_loss=0.1187, over 5664618.30 frames. ], batch size: 307, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:40:03,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=954975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:40:07,913 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=954980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:40:16,975 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 42650, giga_loss[loss=0.3056, simple_loss=0.3675, pruned_loss=0.1218, over 28721.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3655, pruned_loss=0.1168, over 5686500.94 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1149, over 5723622.67 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3669, pruned_loss=0.1177, over 5675764.39 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:40:30,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:40:59,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4580, 1.1968, 3.8483, 3.2844], device='cuda:1'), covar=tensor([0.1551, 0.2813, 0.0516, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0651, 0.0969, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 04:41:03,514 INFO [optim.py:369] (1/2) 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 04:41:04,742 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 42700, giga_loss[loss=0.2777, simple_loss=0.3464, pruned_loss=0.1045, over 28925.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3643, pruned_loss=0.1166, over 5675831.42 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5718834.40 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3654, pruned_loss=0.1174, over 5670806.83 frames. ], batch size: 213, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:41:14,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5203, 4.3878, 4.1358, 1.9924], device='cuda:1'), covar=tensor([0.0555, 0.0639, 0.0716, 0.1982], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1147, 0.0965, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 04:41:59,816 INFO [train.py:968] (1/2) Epoch 21, batch 42750, giga_loss[loss=0.3373, simple_loss=0.3803, pruned_loss=0.1471, over 26536.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3639, pruned_loss=0.1173, over 5666882.39 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5724831.54 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3654, pruned_loss=0.1184, over 5655477.51 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 04:42:18,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:20,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:23,965 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:41,769 INFO [optim.py:369] (1/2) 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 04:42:47,789 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:47,862 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:48,194 INFO [train.py:968] (1/2) Epoch 21, batch 42800, giga_loss[loss=0.3233, simple_loss=0.3821, pruned_loss=0.1322, over 28754.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.364, pruned_loss=0.1174, over 5669152.61 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3614, pruned_loss=0.1143, over 5727572.18 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3653, pruned_loss=0.1184, over 5656830.22 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:42:49,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:42:51,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:12,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:17,142 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:21,350 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:43:27,102 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 42850, giga_loss[loss=0.2972, simple_loss=0.3599, pruned_loss=0.1173, over 28852.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1163, over 5675893.91 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3614, pruned_loss=0.1143, over 5726079.68 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3655, pruned_loss=0.1172, over 5666672.14 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:43:51,079 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:44:10,435 INFO [optim.py:369] (1/2) 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 04:44:17,292 INFO [train.py:968] (1/2) Epoch 21, batch 42900, libri_loss[loss=0.2592, simple_loss=0.3301, pruned_loss=0.0941, over 29648.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3637, pruned_loss=0.1152, over 5673131.96 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.361, pruned_loss=0.1141, over 5721754.79 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.365, pruned_loss=0.1161, over 5667268.61 frames. ], batch size: 73, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:44:21,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5118, 5.3330, 5.0360, 2.5494], device='cuda:1'), covar=tensor([0.0455, 0.0626, 0.0737, 0.1761], device='cuda:1'), in_proj_covar=tensor([0.1239, 0.1150, 0.0967, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 04:45:00,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:45:05,229 INFO [train.py:968] (1/2) Epoch 21, batch 42950, giga_loss[loss=0.3077, simple_loss=0.3784, pruned_loss=0.1184, over 28747.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3646, pruned_loss=0.1156, over 5671342.76 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1144, over 5714012.50 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3655, pruned_loss=0.1161, over 5672371.87 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:45:21,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:45:26,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:45:44,515 INFO [optim.py:369] (1/2) 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 04:45:52,198 INFO [train.py:968] (1/2) Epoch 21, batch 43000, giga_loss[loss=0.3302, simple_loss=0.3891, pruned_loss=0.1356, over 28822.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1183, over 5673553.54 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1147, over 5708481.12 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3682, pruned_loss=0.1184, over 5678251.06 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:45:52,454 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:46:05,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=955366.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:46:37,299 INFO [train.py:968] (1/2) Epoch 21, batch 43050, giga_loss[loss=0.299, simple_loss=0.3598, pruned_loss=0.1191, over 28782.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3695, pruned_loss=0.1205, over 5677482.52 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3621, pruned_loss=0.115, over 5712828.05 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3697, pruned_loss=0.1205, over 5676424.38 frames. ], batch size: 284, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:46:52,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8117, 1.9573, 1.3962, 1.5731], device='cuda:1'), covar=tensor([0.0919, 0.0616, 0.1058, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0448, 0.0517, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 04:47:22,804 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:47:24,573 INFO [optim.py:369] (1/2) 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 04:47:24,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:47:30,830 INFO [train.py:968] (1/2) Epoch 21, batch 43100, giga_loss[loss=0.2897, simple_loss=0.3624, pruned_loss=0.1086, over 28899.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1218, over 5673989.22 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3623, pruned_loss=0.1151, over 5715360.87 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.37, pruned_loss=0.1218, over 5670391.88 frames. ], batch size: 227, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:47:50,970 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:48:01,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4387, 1.7014, 1.6686, 1.2297], device='cuda:1'), covar=tensor([0.1721, 0.2681, 0.1491, 0.1791], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0701, 0.0942, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 04:48:18,408 INFO [train.py:968] (1/2) Epoch 21, batch 43150, giga_loss[loss=0.3791, simple_loss=0.4195, pruned_loss=0.1693, over 27406.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3735, pruned_loss=0.1258, over 5657285.45 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5710562.67 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3736, pruned_loss=0.1258, over 5658466.05 frames. ], batch size: 472, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:48:26,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=955509.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:48:29,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=955512.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 04:48:51,962 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=955541.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 04:48:54,541 INFO [optim.py:369] (1/2) 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 04:49:00,420 INFO [train.py:968] (1/2) Epoch 21, batch 43200, giga_loss[loss=0.3217, simple_loss=0.3826, pruned_loss=0.1304, over 28543.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3712, pruned_loss=0.124, over 5644594.74 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.1151, over 5695323.95 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3718, pruned_loss=0.1244, over 5657516.32 frames. ], batch size: 336, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 04:49:42,778 INFO [train.py:968] (1/2) Epoch 21, batch 43250, giga_loss[loss=0.2886, simple_loss=0.3592, pruned_loss=0.109, over 28721.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.371, pruned_loss=0.1232, over 5646744.75 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3622, pruned_loss=0.115, over 5690934.84 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3718, pruned_loss=0.1239, over 5658964.92 frames. ], batch size: 242, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:50:21,560 INFO [optim.py:369] (1/2) 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 04:50:25,414 INFO [train.py:968] (1/2) Epoch 21, batch 43300, giga_loss[loss=0.2715, simple_loss=0.3469, pruned_loss=0.09803, over 28849.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3715, pruned_loss=0.1221, over 5657979.52 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3619, pruned_loss=0.1149, over 5698672.45 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3727, pruned_loss=0.123, over 5659537.02 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:50:32,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3458, 3.4532, 1.4205, 1.5857], device='cuda:1'), covar=tensor([0.1076, 0.0380, 0.0987, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0557, 0.0386, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 04:51:12,338 INFO [train.py:968] (1/2) Epoch 21, batch 43350, giga_loss[loss=0.265, simple_loss=0.341, pruned_loss=0.09451, over 28724.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3693, pruned_loss=0.1203, over 5656987.62 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5703004.45 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3705, pruned_loss=0.1213, over 5653799.86 frames. ], batch size: 262, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:51:49,265 INFO [optim.py:369] (1/2) 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 04:51:55,617 INFO [train.py:968] (1/2) Epoch 21, batch 43400, giga_loss[loss=0.3273, simple_loss=0.377, pruned_loss=0.1388, over 28507.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3674, pruned_loss=0.1194, over 5674396.26 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3615, pruned_loss=0.1147, over 5707890.17 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5666501.40 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:52:40,960 INFO [train.py:968] (1/2) Epoch 21, batch 43450, giga_loss[loss=0.3494, simple_loss=0.389, pruned_loss=0.155, over 26561.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3668, pruned_loss=0.12, over 5670693.92 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5711469.98 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3681, pruned_loss=0.1209, over 5660268.09 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:53:18,042 INFO [optim.py:369] (1/2) 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 04:53:23,407 INFO [train.py:968] (1/2) Epoch 21, batch 43500, libri_loss[loss=0.3018, simple_loss=0.3726, pruned_loss=0.1155, over 29356.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3689, pruned_loss=0.1214, over 5680169.99 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1148, over 5714434.89 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3699, pruned_loss=0.1223, over 5667783.06 frames. ], batch size: 92, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:53:36,165 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-11 04:53:55,813 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 04:54:00,270 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-11 04:54:10,464 INFO [train.py:968] (1/2) Epoch 21, batch 43550, giga_loss[loss=0.4204, simple_loss=0.4358, pruned_loss=0.2025, over 23585.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3728, pruned_loss=0.1232, over 5663510.40 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1148, over 5708428.12 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.124, over 5657385.05 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:54:40,366 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-11 04:54:55,672 INFO [optim.py:369] (1/2) 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 04:55:02,804 INFO [train.py:968] (1/2) Epoch 21, batch 43600, libri_loss[loss=0.3613, simple_loss=0.4139, pruned_loss=0.1544, over 29541.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3745, pruned_loss=0.1211, over 5672817.18 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3618, pruned_loss=0.1149, over 5710340.02 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3753, pruned_loss=0.1218, over 5665636.63 frames. ], batch size: 89, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:55:07,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6261, 2.3337, 1.7131, 0.8363], device='cuda:1'), covar=tensor([0.6638, 0.3264, 0.4115, 0.6857], device='cuda:1'), in_proj_covar=tensor([0.1742, 0.1656, 0.1598, 0.1422], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 04:55:08,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 04:55:34,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5781, 1.7574, 1.4638, 1.6504], device='cuda:1'), covar=tensor([0.2706, 0.2671, 0.2937, 0.2210], device='cuda:1'), in_proj_covar=tensor([0.1500, 0.1085, 0.1322, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 04:55:54,930 INFO [train.py:968] (1/2) Epoch 21, batch 43650, giga_loss[loss=0.3302, simple_loss=0.3975, pruned_loss=0.1314, over 28881.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3766, pruned_loss=0.1225, over 5668949.75 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5711371.32 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3775, pruned_loss=0.1232, over 5662092.01 frames. ], batch size: 199, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:56:10,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5563, 1.8384, 1.4535, 1.6358], device='cuda:1'), covar=tensor([0.2506, 0.2598, 0.2909, 0.2367], device='cuda:1'), in_proj_covar=tensor([0.1499, 0.1085, 0.1321, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 04:56:29,767 INFO [optim.py:369] (1/2) 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 04:56:35,005 INFO [train.py:968] (1/2) Epoch 21, batch 43700, giga_loss[loss=0.3453, simple_loss=0.4002, pruned_loss=0.1452, over 27864.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3776, pruned_loss=0.1237, over 5669329.11 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.115, over 5715946.57 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3786, pruned_loss=0.1243, over 5658107.06 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:56:43,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:57:23,504 INFO [train.py:968] (1/2) Epoch 21, batch 43750, libri_loss[loss=0.2876, simple_loss=0.3615, pruned_loss=0.1068, over 28581.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3775, pruned_loss=0.1244, over 5659836.11 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1151, over 5708203.43 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3784, pruned_loss=0.125, over 5657646.68 frames. ], batch size: 106, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:57:50,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 04:58:04,858 INFO [optim.py:369] (1/2) 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 04:58:09,998 INFO [train.py:968] (1/2) Epoch 21, batch 43800, giga_loss[loss=0.3034, simple_loss=0.3665, pruned_loss=0.1202, over 28135.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3768, pruned_loss=0.125, over 5656516.30 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3622, pruned_loss=0.1152, over 5709748.65 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3775, pruned_loss=0.1254, over 5652977.96 frames. ], batch size: 77, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:59:00,061 INFO [train.py:968] (1/2) Epoch 21, batch 43850, giga_loss[loss=0.2953, simple_loss=0.357, pruned_loss=0.1168, over 28114.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1237, over 5658794.51 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1151, over 5710746.91 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3745, pruned_loss=0.1241, over 5654970.38 frames. ], batch size: 77, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 04:59:21,092 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:59:47,979 INFO [train.py:968] (1/2) Epoch 21, batch 43900, giga_loss[loss=0.3386, simple_loss=0.3936, pruned_loss=0.1418, over 28394.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3732, pruned_loss=0.124, over 5660471.98 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1156, over 5704174.98 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3734, pruned_loss=0.1241, over 5662512.80 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:00:13,157 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:00:23,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:00:26,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3531, 1.5601, 1.4036, 1.2611], device='cuda:1'), covar=tensor([0.2335, 0.2281, 0.1892, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.1975, 0.1904, 0.1839, 0.1975], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 05:00:34,744 INFO [train.py:968] (1/2) Epoch 21, batch 43950, giga_loss[loss=0.282, simple_loss=0.3536, pruned_loss=0.1052, over 28908.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3721, pruned_loss=0.1233, over 5673027.15 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3623, pruned_loss=0.1152, over 5710101.43 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.373, pruned_loss=0.124, over 5668307.94 frames. ], batch size: 112, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:01:22,594 INFO [optim.py:369] (1/2) 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 05:01:23,096 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 05:01:23,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:01:26,328 INFO [train.py:968] (1/2) Epoch 21, batch 44000, giga_loss[loss=0.2901, simple_loss=0.3509, pruned_loss=0.1147, over 28626.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3742, pruned_loss=0.1255, over 5669862.74 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5710694.92 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3749, pruned_loss=0.126, over 5665559.28 frames. ], batch size: 78, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:02:12,517 INFO [train.py:968] (1/2) Epoch 21, batch 44050, giga_loss[loss=0.3126, simple_loss=0.376, pruned_loss=0.1247, over 28279.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3719, pruned_loss=0.1244, over 5670566.75 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3617, pruned_loss=0.1149, over 5713284.69 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3733, pruned_loss=0.1254, over 5664203.44 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:02:22,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 05:02:33,365 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:02:40,120 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:02:57,139 INFO [train.py:968] (1/2) Epoch 21, batch 44100, giga_loss[loss=0.3244, simple_loss=0.3871, pruned_loss=0.1308, over 28918.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3712, pruned_loss=0.1237, over 5670985.41 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3622, pruned_loss=0.1151, over 5707031.71 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1245, over 5669642.92 frames. ], batch size: 227, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:03:44,190 INFO [train.py:968] (1/2) Epoch 21, batch 44150, giga_loss[loss=0.2855, simple_loss=0.3675, pruned_loss=0.1018, over 29002.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3711, pruned_loss=0.1227, over 5672470.05 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1146, over 5713241.31 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3726, pruned_loss=0.1241, over 5664483.83 frames. ], batch size: 164, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:03:50,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5736, 1.6663, 1.1737, 1.2510], device='cuda:1'), covar=tensor([0.0860, 0.0492, 0.1013, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0453, 0.0521, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 05:03:51,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:04:28,928 INFO [optim.py:369] (1/2) 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 05:04:31,893 INFO [train.py:968] (1/2) Epoch 21, batch 44200, giga_loss[loss=0.2738, simple_loss=0.3464, pruned_loss=0.1006, over 28710.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3733, pruned_loss=0.1242, over 5674848.99 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3616, pruned_loss=0.1146, over 5714250.03 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1254, over 5666861.93 frames. ], batch size: 242, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:04:55,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:04:57,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:05:16,782 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 44250, libri_loss[loss=0.279, simple_loss=0.357, pruned_loss=0.1005, over 29650.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3721, pruned_loss=0.1239, over 5677287.71 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1141, over 5719830.32 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3742, pruned_loss=0.1256, over 5664619.94 frames. ], batch size: 88, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:05:24,637 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:05:41,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9318, 3.6912, 3.4838, 1.8140], device='cuda:1'), covar=tensor([0.0846, 0.1003, 0.0972, 0.2212], device='cuda:1'), in_proj_covar=tensor([0.1248, 0.1158, 0.0979, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 05:06:01,692 INFO [optim.py:369] (1/2) 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 05:06:05,569 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:05,597 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:05,849 INFO [train.py:968] (1/2) Epoch 21, batch 44300, giga_loss[loss=0.3756, simple_loss=0.4112, pruned_loss=0.1699, over 26746.00 frames. ], tot_loss[loss=0.31, simple_loss=0.374, pruned_loss=0.123, over 5670212.37 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1143, over 5721090.05 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3755, pruned_loss=0.1244, over 5658742.37 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:06:07,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:16,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:33,732 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:34,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:06:48,882 INFO [train.py:968] (1/2) Epoch 21, batch 44350, giga_loss[loss=0.2698, simple_loss=0.3555, pruned_loss=0.09206, over 28880.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3747, pruned_loss=0.1205, over 5689791.27 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3612, pruned_loss=0.1142, over 5724061.47 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3761, pruned_loss=0.1217, over 5677567.29 frames. ], batch size: 66, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:07:08,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:07:27,894 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:07:30,714 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 44400, giga_loss[loss=0.4638, simple_loss=0.4727, pruned_loss=0.2274, over 27976.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.378, pruned_loss=0.1222, over 5670008.25 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3611, pruned_loss=0.1142, over 5700881.01 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3795, pruned_loss=0.1234, over 5678253.15 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 05:07:55,902 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:07:58,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-11 05:08:15,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2347, 1.8105, 1.3232, 0.4526], device='cuda:1'), covar=tensor([0.4672, 0.2922, 0.4063, 0.6238], device='cuda:1'), in_proj_covar=tensor([0.1753, 0.1663, 0.1602, 0.1429], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 05:08:15,746 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:18,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:20,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=956797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:23,154 INFO [train.py:968] (1/2) Epoch 21, batch 44450, giga_loss[loss=0.3467, simple_loss=0.402, pruned_loss=0.1457, over 28830.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3802, pruned_loss=0.1248, over 5665796.88 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5696687.51 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3825, pruned_loss=0.1262, over 5674667.57 frames. ], batch size: 119, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:08:26,177 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:28,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:43,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=956825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:49,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=956830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:08:56,490 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 21, batch 44500, giga_loss[loss=0.384, simple_loss=0.407, pruned_loss=0.1806, over 23199.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3817, pruned_loss=0.1269, over 5663959.80 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.361, pruned_loss=0.114, over 5701345.39 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3839, pruned_loss=0.1284, over 5666101.16 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:09:24,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:09:27,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:09:42,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5384, 1.6704, 1.3629, 1.2782], device='cuda:1'), covar=tensor([0.0891, 0.0471, 0.0938, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0452, 0.0519, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 05:09:53,916 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 44550, giga_loss[loss=0.2966, simple_loss=0.3622, pruned_loss=0.1154, over 28971.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.382, pruned_loss=0.1284, over 5643718.54 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1142, over 5693982.14 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3838, pruned_loss=0.1296, over 5651227.98 frames. ], batch size: 213, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:10:21,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3523, 1.5308, 1.5294, 1.3845], device='cuda:1'), covar=tensor([0.1503, 0.1669, 0.1850, 0.1637], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0752, 0.0712, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 05:10:28,651 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=956940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:10:31,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=956943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:10:35,223 INFO [optim.py:369] (1/2) 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 05:10:37,115 INFO [train.py:968] (1/2) Epoch 21, batch 44600, giga_loss[loss=0.2559, simple_loss=0.3411, pruned_loss=0.0854, over 28867.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3802, pruned_loss=0.1265, over 5661615.91 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3606, pruned_loss=0.1138, over 5700679.30 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.383, pruned_loss=0.1284, over 5660101.01 frames. ], batch size: 164, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:10:55,298 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 21, batch 44650, libri_loss[loss=0.2833, simple_loss=0.3534, pruned_loss=0.1065, over 29484.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.379, pruned_loss=0.1242, over 5667273.18 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3607, pruned_loss=0.1139, over 5702663.17 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3817, pruned_loss=0.1259, over 5663007.35 frames. ], batch size: 85, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:12:04,119 INFO [optim.py:369] (1/2) 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 05:12:05,877 INFO [train.py:968] (1/2) Epoch 21, batch 44700, giga_loss[loss=0.3039, simple_loss=0.375, pruned_loss=0.1164, over 28856.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3803, pruned_loss=0.124, over 5651134.62 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.361, pruned_loss=0.1143, over 5689722.39 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3825, pruned_loss=0.1252, over 5659398.40 frames. ], batch size: 186, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:12:11,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:12:41,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5102, 1.9252, 1.7312, 1.6861], device='cuda:1'), covar=tensor([0.2144, 0.2130, 0.2377, 0.2301], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0755, 0.0714, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 05:12:52,570 INFO [train.py:968] (1/2) Epoch 21, batch 44750, giga_loss[loss=0.3112, simple_loss=0.3722, pruned_loss=0.1251, over 28302.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3801, pruned_loss=0.1239, over 5667431.36 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5695985.29 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3824, pruned_loss=0.125, over 5667238.55 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:13:35,293 INFO [optim.py:369] (1/2) 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 05:13:38,206 INFO [train.py:968] (1/2) Epoch 21, batch 44800, giga_loss[loss=0.3031, simple_loss=0.372, pruned_loss=0.1171, over 28958.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3783, pruned_loss=0.1233, over 5673129.93 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1143, over 5699137.92 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3811, pruned_loss=0.1246, over 5669242.96 frames. ], batch size: 227, lr: 1.50e-03, grad_scale: 8.0
+2023-03-11 05:14:00,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5638, 1.2492, 4.2453, 3.4495], device='cuda:1'), covar=tensor([0.1536, 0.2845, 0.0433, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0651, 0.0973, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 05:14:22,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8588, 3.6737, 3.4875, 1.7327], device='cuda:1'), covar=tensor([0.0768, 0.0918, 0.0849, 0.2197], device='cuda:1'), in_proj_covar=tensor([0.1246, 0.1156, 0.0979, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 05:14:23,570 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:23,963 INFO [train.py:968] (1/2) Epoch 21, batch 44850, giga_loss[loss=0.3122, simple_loss=0.3719, pruned_loss=0.1263, over 28706.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3776, pruned_loss=0.1238, over 5666720.28 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3605, pruned_loss=0.1142, over 5701253.18 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.38, pruned_loss=0.1249, over 5661670.14 frames. ], batch size: 243, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:14:24,193 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:26,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:28,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:14:56,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:15:11,862 INFO [optim.py:369] (1/2) 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 05:15:13,526 INFO [train.py:968] (1/2) Epoch 21, batch 44900, giga_loss[loss=0.3273, simple_loss=0.3836, pruned_loss=0.1355, over 28287.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3768, pruned_loss=0.1247, over 5652911.12 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3609, pruned_loss=0.1143, over 5699945.90 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3786, pruned_loss=0.1257, over 5649637.55 frames. ], batch size: 368, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:15:24,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:15:34,378 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:16:00,006 INFO [train.py:968] (1/2) Epoch 21, batch 44950, libri_loss[loss=0.3104, simple_loss=0.3742, pruned_loss=0.1232, over 29183.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3746, pruned_loss=0.1238, over 5651830.02 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5694975.47 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3763, pruned_loss=0.1248, over 5652936.92 frames. ], batch size: 97, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:16:26,625 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:16:45,084 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:16:47,571 INFO [train.py:968] (1/2) Epoch 21, batch 45000, giga_loss[loss=0.2828, simple_loss=0.3506, pruned_loss=0.1075, over 28823.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3729, pruned_loss=0.1233, over 5655337.71 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 5698173.62 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3746, pruned_loss=0.1244, over 5652914.10 frames. ], batch size: 186, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:16:47,572 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 05:16:56,694 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 05:16:56,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:17:20,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7765, 1.9682, 1.2772, 1.5583], device='cuda:1'), covar=tensor([0.0995, 0.0639, 0.1138, 0.1143], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0451, 0.0519, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 05:17:24,231 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:17:39,787 INFO [train.py:968] (1/2) Epoch 21, batch 45050, giga_loss[loss=0.3905, simple_loss=0.4287, pruned_loss=0.1761, over 27955.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3733, pruned_loss=0.1241, over 5670204.87 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3608, pruned_loss=0.1143, over 5702015.87 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.375, pruned_loss=0.1252, over 5663754.56 frames. ], batch size: 412, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:17:55,353 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:18:23,231 INFO [train.py:968] (1/2) Epoch 21, batch 45100, giga_loss[loss=0.3217, simple_loss=0.3793, pruned_loss=0.132, over 26651.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1217, over 5671333.41 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3601, pruned_loss=0.1139, over 5706809.44 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1231, over 5661492.48 frames. ], batch size: 555, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:18:32,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1952, 1.0098, 1.0120, 1.3684], device='cuda:1'), covar=tensor([0.0685, 0.0366, 0.0330, 0.0702], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 05:19:06,499 INFO [train.py:968] (1/2) Epoch 21, batch 45150, libri_loss[loss=0.2583, simple_loss=0.3277, pruned_loss=0.09444, over 29492.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3668, pruned_loss=0.1178, over 5668978.26 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3597, pruned_loss=0.1136, over 5705188.01 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3694, pruned_loss=0.1194, over 5660883.06 frames. ], batch size: 70, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:19:28,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3340, 1.4166, 1.3699, 1.2863], device='cuda:1'), covar=tensor([0.2306, 0.2118, 0.2078, 0.2168], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1897, 0.1839, 0.1965], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 05:19:54,313 INFO [optim.py:369] (1/2) 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 05:19:55,470 INFO [train.py:968] (1/2) Epoch 21, batch 45200, giga_loss[loss=0.282, simple_loss=0.3521, pruned_loss=0.106, over 28761.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3661, pruned_loss=0.1165, over 5669714.48 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5706511.30 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3681, pruned_loss=0.1178, over 5661957.55 frames. ], batch size: 66, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:19:55,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3751, 1.3727, 4.0989, 3.1954], device='cuda:1'), covar=tensor([0.1638, 0.2639, 0.0421, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0650, 0.0969, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 05:20:20,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:20:41,720 INFO [train.py:968] (1/2) Epoch 21, batch 45250, giga_loss[loss=0.2761, simple_loss=0.3454, pruned_loss=0.1034, over 28862.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3652, pruned_loss=0.1166, over 5650020.71 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1138, over 5698432.29 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3664, pruned_loss=0.1175, over 5650297.25 frames. ], batch size: 186, lr: 1.50e-03, grad_scale: 4.0
+2023-03-11 05:20:51,911 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-11 05:21:18,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:21:26,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:21:28,949 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 21, batch 45300, giga_loss[loss=0.2694, simple_loss=0.3472, pruned_loss=0.09575, over 28983.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.364, pruned_loss=0.1173, over 5584397.28 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3615, pruned_loss=0.1149, over 5634758.72 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3642, pruned_loss=0.1172, over 5638482.42 frames. ], batch size: 136, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:21:46,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1542, 1.2530, 1.1296, 0.8294], device='cuda:1'), covar=tensor([0.0984, 0.0540, 0.1067, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0451, 0.0520, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 05:22:11,750 INFO [train.py:968] (1/2) Epoch 21, batch 45350, giga_loss[loss=0.3379, simple_loss=0.3744, pruned_loss=0.1507, over 23584.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.367, pruned_loss=0.1197, over 5548753.88 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3625, pruned_loss=0.1158, over 5583521.48 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3663, pruned_loss=0.1189, over 5637214.57 frames. ], batch size: 705, lr: 1.50e-03, grad_scale: 2.0
+2023-03-11 05:22:16,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:22:27,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:22:30,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:22:38,601 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-11 05:23:32,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:23:32,514 INFO [optim.py:369] (1/2) 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 05:24:01,358 INFO [train.py:968] (1/2) Epoch 22, batch 50, libri_loss[loss=0.2202, simple_loss=0.3017, pruned_loss=0.06937, over 29337.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3754, pruned_loss=0.111, over 1267788.20 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3331, pruned_loss=0.0859, over 145888.66 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3807, pruned_loss=0.114, over 1150694.68 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:24:03,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:05,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:09,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:12,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=957794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:12,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:24:32,899 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 100, giga_loss[loss=0.2647, simple_loss=0.3339, pruned_loss=0.09778, over 28479.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3642, pruned_loss=0.1054, over 2242263.18 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3327, pruned_loss=0.08843, over 259663.69 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3678, pruned_loss=0.1073, over 2075987.45 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:24:53,148 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=957837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:25:02,347 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:25:04,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:25:04,757 INFO [optim.py:369] (1/2) 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 05:25:30,668 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 150, giga_loss[loss=0.2315, simple_loss=0.3058, pruned_loss=0.07855, over 28494.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3481, pruned_loss=0.09696, over 3000750.30 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3332, pruned_loss=0.08646, over 387431.20 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3502, pruned_loss=0.09838, over 2805090.12 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:26:14,249 INFO [train.py:968] (1/2) Epoch 22, batch 200, libri_loss[loss=0.2542, simple_loss=0.3441, pruned_loss=0.08214, over 29543.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3343, pruned_loss=0.09027, over 3600692.92 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3351, pruned_loss=0.08533, over 493183.28 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3349, pruned_loss=0.0912, over 3402425.53 frames. ], batch size: 83, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:26:19,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=957937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:26:23,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=957940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:26:30,437 INFO [optim.py:369] (1/2) 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 05:26:41,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2920, 2.7931, 1.4630, 1.4002], device='cuda:1'), covar=tensor([0.0990, 0.0336, 0.0900, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0552, 0.0384, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 05:26:45,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=957969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:26:57,332 INFO [train.py:968] (1/2) Epoch 22, batch 250, giga_loss[loss=0.1963, simple_loss=0.278, pruned_loss=0.05729, over 28916.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3244, pruned_loss=0.086, over 4060735.38 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3408, pruned_loss=0.08959, over 574060.75 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3235, pruned_loss=0.08594, over 3876498.14 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:27:37,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=958029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:27:38,017 INFO [train.py:968] (1/2) Epoch 22, batch 300, giga_loss[loss=0.1904, simple_loss=0.271, pruned_loss=0.0549, over 28874.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3179, pruned_loss=0.08272, over 4418015.04 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3438, pruned_loss=0.09126, over 822457.78 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3151, pruned_loss=0.0819, over 4203629.96 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:27:49,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3663, 1.3070, 3.9858, 3.3884], device='cuda:1'), covar=tensor([0.1634, 0.2858, 0.0423, 0.1412], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0650, 0.0971, 0.0915], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 05:27:55,182 INFO [optim.py:369] (1/2) 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 05:28:15,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=958072.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 05:28:23,789 INFO [train.py:968] (1/2) Epoch 22, batch 350, giga_loss[loss=0.2092, simple_loss=0.2838, pruned_loss=0.0673, over 28934.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.311, pruned_loss=0.07969, over 4697141.92 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3432, pruned_loss=0.09052, over 946467.70 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.308, pruned_loss=0.07883, over 4498008.10 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:28:59,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2933, 3.1159, 2.9755, 1.4878], device='cuda:1'), covar=tensor([0.0949, 0.1158, 0.0994, 0.2332], device='cuda:1'), in_proj_covar=tensor([0.1222, 0.1138, 0.0961, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 05:29:04,359 INFO [train.py:968] (1/2) Epoch 22, batch 400, libri_loss[loss=0.276, simple_loss=0.3601, pruned_loss=0.09596, over 28560.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3072, pruned_loss=0.07785, over 4925296.72 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3424, pruned_loss=0.09015, over 1068387.52 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.3039, pruned_loss=0.07686, over 4742748.31 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:29:21,765 INFO [optim.py:369] (1/2) 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 05:29:45,627 INFO [train.py:968] (1/2) Epoch 22, batch 450, libri_loss[loss=0.2611, simple_loss=0.3459, pruned_loss=0.08815, over 29289.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3047, pruned_loss=0.07695, over 5099753.36 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3414, pruned_loss=0.08915, over 1140441.13 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.3018, pruned_loss=0.07617, over 4943676.11 frames. ], batch size: 94, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:30:13,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=958212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:30:31,237 INFO [train.py:968] (1/2) Epoch 22, batch 500, libri_loss[loss=0.2662, simple_loss=0.3429, pruned_loss=0.09474, over 29552.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3019, pruned_loss=0.07592, over 5220647.07 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08935, over 1187566.75 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.2991, pruned_loss=0.07512, over 5091621.98 frames. ], batch size: 79, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:30:47,549 INFO [optim.py:369] (1/2) 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 05:30:52,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2611, 1.6098, 1.2247, 0.8273], device='cuda:1'), covar=tensor([0.4525, 0.2188, 0.2834, 0.5669], device='cuda:1'), in_proj_covar=tensor([0.1743, 0.1648, 0.1591, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 05:31:15,151 INFO [train.py:968] (1/2) Epoch 22, batch 550, libri_loss[loss=0.2635, simple_loss=0.3448, pruned_loss=0.09107, over 29517.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3002, pruned_loss=0.07522, over 5328125.05 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08985, over 1281188.00 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2968, pruned_loss=0.07419, over 5214447.12 frames. ], batch size: 81, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:31:58,466 INFO [train.py:968] (1/2) Epoch 22, batch 600, libri_loss[loss=0.2325, simple_loss=0.3192, pruned_loss=0.07285, over 29565.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2987, pruned_loss=0.07427, over 5405758.81 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08874, over 1428724.67 frames. ], giga_tot_loss[loss=0.2206, simple_loss=0.2948, pruned_loss=0.0732, over 5308727.61 frames. ], batch size: 76, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:32:15,783 INFO [optim.py:369] (1/2) 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 05:32:19,866 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=958355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:32:23,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9115, 1.2467, 1.3309, 1.0996], device='cuda:1'), covar=tensor([0.1963, 0.1328, 0.2269, 0.1620], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0748, 0.0709, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 05:32:24,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=958358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:32:42,754 INFO [train.py:968] (1/2) Epoch 22, batch 650, giga_loss[loss=0.2211, simple_loss=0.2971, pruned_loss=0.0725, over 28680.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2968, pruned_loss=0.07328, over 5475126.90 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3419, pruned_loss=0.08842, over 1560847.94 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2924, pruned_loss=0.07206, over 5385299.44 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:32:49,839 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=958387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:33:02,238 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 700, giga_loss[loss=0.2043, simple_loss=0.2791, pruned_loss=0.06477, over 28720.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.293, pruned_loss=0.07134, over 5525739.73 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.0885, over 1626002.14 frames. ], giga_tot_loss[loss=0.2146, simple_loss=0.2889, pruned_loss=0.07011, over 5449108.53 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:33:32,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3409, 1.2462, 3.8611, 3.1534], device='cuda:1'), covar=tensor([0.1651, 0.2868, 0.0459, 0.1045], device='cuda:1'), in_proj_covar=tensor([0.0759, 0.0648, 0.0969, 0.0912], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 05:33:39,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7708, 2.1263, 2.0548, 1.8395], device='cuda:1'), covar=tensor([0.1728, 0.1680, 0.1774, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0747, 0.0709, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 05:33:41,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=958447.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 05:33:44,124 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 750, libri_loss[loss=0.2499, simple_loss=0.3393, pruned_loss=0.08024, over 26279.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.2925, pruned_loss=0.07098, over 5552217.15 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3439, pruned_loss=0.08949, over 1767300.88 frames. ], giga_tot_loss[loss=0.2126, simple_loss=0.287, pruned_loss=0.06913, over 5484799.90 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:34:55,173 INFO [train.py:968] (1/2) Epoch 22, batch 800, giga_loss[loss=0.2582, simple_loss=0.3291, pruned_loss=0.09362, over 29043.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2925, pruned_loss=0.07119, over 5594347.51 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3428, pruned_loss=0.0888, over 1949955.24 frames. ], giga_tot_loss[loss=0.2122, simple_loss=0.2862, pruned_loss=0.0691, over 5525723.21 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:35:11,446 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=958547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:35:14,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=958550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:35:15,252 INFO [optim.py:369] (1/2) 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 05:35:42,692 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=958579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:35:43,974 INFO [train.py:968] (1/2) Epoch 22, batch 850, giga_loss[loss=0.2862, simple_loss=0.3633, pruned_loss=0.1045, over 28297.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3003, pruned_loss=0.07553, over 5613067.07 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3426, pruned_loss=0.08865, over 1969727.46 frames. ], giga_tot_loss[loss=0.2216, simple_loss=0.2953, pruned_loss=0.07392, over 5558041.13 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:35:46,126 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9728, 1.2571, 1.3298, 1.0431], device='cuda:1'), covar=tensor([0.1752, 0.1270, 0.2197, 0.1655], device='cuda:1'), in_proj_covar=tensor([0.0474, 0.0746, 0.0709, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 05:35:55,274 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=958590.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 05:35:57,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=958593.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 05:36:25,003 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 900, giga_loss[loss=0.2728, simple_loss=0.3515, pruned_loss=0.09706, over 28902.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3133, pruned_loss=0.08204, over 5630034.71 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3422, pruned_loss=0.08865, over 2117143.35 frames. ], giga_tot_loss[loss=0.2345, simple_loss=0.3081, pruned_loss=0.08044, over 5581966.41 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:36:49,637 INFO [optim.py:369] (1/2) 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 05:36:57,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5880, 1.7967, 1.4822, 1.5629], device='cuda:1'), covar=tensor([0.2631, 0.2687, 0.3025, 0.2346], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1091, 0.1334, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 05:37:03,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 05:37:12,030 INFO [train.py:968] (1/2) Epoch 22, batch 950, giga_loss[loss=0.2892, simple_loss=0.3698, pruned_loss=0.1043, over 28557.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3253, pruned_loss=0.08863, over 5641402.27 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3423, pruned_loss=0.0891, over 2227848.13 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3207, pruned_loss=0.08716, over 5596969.16 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:37:13,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3972, 1.5348, 1.1451, 1.1410], device='cuda:1'), covar=tensor([0.0916, 0.0486, 0.1053, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0449, 0.0520, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 05:37:53,934 INFO [train.py:968] (1/2) Epoch 22, batch 1000, giga_loss[loss=0.3075, simple_loss=0.3796, pruned_loss=0.1177, over 28286.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3328, pruned_loss=0.09162, over 5650546.28 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.342, pruned_loss=0.08903, over 2264576.65 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3292, pruned_loss=0.09053, over 5613215.65 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:38:02,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-11 05:38:09,973 INFO [optim.py:369] (1/2) 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 05:38:29,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 05:38:33,841 INFO [train.py:968] (1/2) Epoch 22, batch 1050, giga_loss[loss=0.2653, simple_loss=0.3536, pruned_loss=0.08852, over 27650.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3356, pruned_loss=0.09134, over 5662444.22 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.341, pruned_loss=0.08894, over 2362925.02 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3329, pruned_loss=0.09057, over 5633903.72 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:38:45,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=958793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:39:05,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-11 05:39:22,536 INFO [train.py:968] (1/2) Epoch 22, batch 1100, giga_loss[loss=0.2735, simple_loss=0.3398, pruned_loss=0.1036, over 28509.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3367, pruned_loss=0.09122, over 5660738.32 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3404, pruned_loss=0.08878, over 2397344.17 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3347, pruned_loss=0.09069, over 5637064.30 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:39:39,412 INFO [optim.py:369] (1/2) 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 05:40:03,637 INFO [train.py:968] (1/2) Epoch 22, batch 1150, giga_loss[loss=0.2858, simple_loss=0.3539, pruned_loss=0.1088, over 28498.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3395, pruned_loss=0.09306, over 5664803.44 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3402, pruned_loss=0.08868, over 2466998.58 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3379, pruned_loss=0.09274, over 5641776.05 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:40:14,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6126, 1.8037, 1.6435, 1.5690], device='cuda:1'), covar=tensor([0.2072, 0.2534, 0.2506, 0.2413], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0752, 0.0715, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 05:40:26,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 05:40:30,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2176, 1.1962, 3.5540, 3.0896], device='cuda:1'), covar=tensor([0.1701, 0.2852, 0.0493, 0.1142], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0642, 0.0959, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 05:40:32,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6056, 1.4729, 4.5311, 3.4097], device='cuda:1'), covar=tensor([0.1674, 0.2796, 0.0389, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0642, 0.0959, 0.0905], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 05:40:34,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3297, 1.6642, 1.3787, 1.0932], device='cuda:1'), covar=tensor([0.2587, 0.2570, 0.2931, 0.2248], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1096, 0.1336, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 05:40:47,409 INFO [train.py:968] (1/2) Epoch 22, batch 1200, giga_loss[loss=0.2743, simple_loss=0.3575, pruned_loss=0.09554, over 28836.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3426, pruned_loss=0.0952, over 5668615.26 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.341, pruned_loss=0.08938, over 2575958.17 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.341, pruned_loss=0.09483, over 5650427.39 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:41:07,509 INFO [optim.py:369] (1/2) 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 05:41:29,242 INFO [train.py:968] (1/2) Epoch 22, batch 1250, giga_loss[loss=0.2934, simple_loss=0.3775, pruned_loss=0.1047, over 28981.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3454, pruned_loss=0.09687, over 5679570.93 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3411, pruned_loss=0.08947, over 2657910.10 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3442, pruned_loss=0.0967, over 5660259.98 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:41:40,820 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-11 05:42:12,641 INFO [train.py:968] (1/2) Epoch 22, batch 1300, libri_loss[loss=0.2683, simple_loss=0.3564, pruned_loss=0.09013, over 29761.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.348, pruned_loss=0.09726, over 5688749.98 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3409, pruned_loss=0.08897, over 2767281.35 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09762, over 5668466.88 frames. ], batch size: 87, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:42:27,977 INFO [optim.py:369] (1/2) 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 05:42:45,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0655, 5.8126, 5.5214, 3.2603], device='cuda:1'), covar=tensor([0.0404, 0.0557, 0.0666, 0.1549], device='cuda:1'), in_proj_covar=tensor([0.1204, 0.1121, 0.0949, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 05:42:50,690 INFO [train.py:968] (1/2) Epoch 22, batch 1350, giga_loss[loss=0.2841, simple_loss=0.3633, pruned_loss=0.1025, over 28966.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3501, pruned_loss=0.09787, over 5690238.70 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3408, pruned_loss=0.08897, over 2814095.58 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3498, pruned_loss=0.09828, over 5671349.13 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:43:28,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3457, 1.4692, 1.2471, 1.5512], device='cuda:1'), covar=tensor([0.0759, 0.0408, 0.0365, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 05:43:36,296 INFO [train.py:968] (1/2) Epoch 22, batch 1400, giga_loss[loss=0.2944, simple_loss=0.3715, pruned_loss=0.1086, over 28940.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3521, pruned_loss=0.09844, over 5693266.30 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3407, pruned_loss=0.08883, over 2843487.90 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.352, pruned_loss=0.09892, over 5677760.51 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:43:49,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.33 vs. limit=5.0
+2023-03-11 05:43:51,618 INFO [optim.py:369] (1/2) 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 05:44:05,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=959168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:44:16,464 INFO [train.py:968] (1/2) Epoch 22, batch 1450, giga_loss[loss=0.2414, simple_loss=0.3274, pruned_loss=0.07768, over 28505.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.351, pruned_loss=0.09685, over 5692700.49 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3399, pruned_loss=0.08831, over 2958401.02 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3517, pruned_loss=0.0978, over 5678048.65 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:44:21,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-11 05:44:41,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5652, 2.2459, 1.6677, 0.7792], device='cuda:1'), covar=tensor([0.7143, 0.3026, 0.4785, 0.7271], device='cuda:1'), in_proj_covar=tensor([0.1742, 0.1637, 0.1592, 0.1415], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 05:44:55,493 INFO [train.py:968] (1/2) Epoch 22, batch 1500, giga_loss[loss=0.2403, simple_loss=0.3324, pruned_loss=0.07406, over 28792.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3498, pruned_loss=0.09514, over 5701863.56 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3398, pruned_loss=0.08818, over 2987583.55 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3506, pruned_loss=0.09602, over 5688758.05 frames. ], batch size: 66, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:45:11,979 INFO [optim.py:369] (1/2) 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 05:45:23,992 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 1550, giga_loss[loss=0.296, simple_loss=0.367, pruned_loss=0.1125, over 28833.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3489, pruned_loss=0.09422, over 5710581.63 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3398, pruned_loss=0.08812, over 3059688.87 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3496, pruned_loss=0.09508, over 5695913.16 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:45:46,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 05:45:53,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4463, 1.5841, 1.5120, 1.3303], device='cuda:1'), covar=tensor([0.2992, 0.2722, 0.2470, 0.2846], device='cuda:1'), in_proj_covar=tensor([0.1956, 0.1876, 0.1817, 0.1952], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 05:46:02,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=959311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:46:05,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=959314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:46:17,361 INFO [train.py:968] (1/2) Epoch 22, batch 1600, libri_loss[loss=0.2542, simple_loss=0.3348, pruned_loss=0.08677, over 29525.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.349, pruned_loss=0.09537, over 5704949.27 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3388, pruned_loss=0.08747, over 3171518.38 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3504, pruned_loss=0.09667, over 5685564.30 frames. ], batch size: 80, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:46:25,307 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=959343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:46:25,556 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-11 05:46:33,752 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 1650, giga_loss[loss=0.3236, simple_loss=0.3823, pruned_loss=0.1325, over 28855.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3517, pruned_loss=0.09935, over 5709916.15 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3391, pruned_loss=0.0876, over 3236053.77 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.353, pruned_loss=0.1006, over 5692595.07 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:47:41,563 INFO [train.py:968] (1/2) Epoch 22, batch 1700, giga_loss[loss=0.3307, simple_loss=0.3823, pruned_loss=0.1395, over 28025.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3533, pruned_loss=0.1018, over 5715724.38 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3396, pruned_loss=0.08774, over 3324748.87 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3546, pruned_loss=0.1032, over 5698859.44 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:48:00,559 INFO [optim.py:369] (1/2) 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 05:48:26,234 INFO [train.py:968] (1/2) Epoch 22, batch 1750, giga_loss[loss=0.2671, simple_loss=0.3419, pruned_loss=0.09616, over 28749.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3524, pruned_loss=0.1024, over 5702173.53 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3399, pruned_loss=0.08775, over 3358885.28 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3534, pruned_loss=0.1037, over 5690215.06 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 05:49:04,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8112, 2.0450, 1.4149, 1.5446], device='cuda:1'), covar=tensor([0.0977, 0.0587, 0.1063, 0.1143], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0450, 0.0522, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 05:49:07,373 INFO [train.py:968] (1/2) Epoch 22, batch 1800, giga_loss[loss=0.2642, simple_loss=0.3434, pruned_loss=0.09248, over 29010.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3493, pruned_loss=0.1008, over 5691834.39 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3393, pruned_loss=0.08737, over 3448928.89 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3508, pruned_loss=0.1025, over 5683632.85 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:49:27,625 INFO [optim.py:369] (1/2) 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 05:49:29,733 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=959556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:49:37,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6669, 2.0834, 1.8309, 1.8475], device='cuda:1'), covar=tensor([0.0782, 0.0274, 0.0298, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 05:49:49,582 INFO [train.py:968] (1/2) Epoch 22, batch 1850, giga_loss[loss=0.242, simple_loss=0.327, pruned_loss=0.0785, over 28878.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3485, pruned_loss=0.1001, over 5677168.76 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08748, over 3461391.39 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3496, pruned_loss=0.1015, over 5680157.64 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:50:35,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1382, 1.5692, 1.2043, 0.4533], device='cuda:1'), covar=tensor([0.3743, 0.1979, 0.3014, 0.5741], device='cuda:1'), in_proj_covar=tensor([0.1747, 0.1643, 0.1596, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 05:50:36,166 INFO [train.py:968] (1/2) Epoch 22, batch 1900, giga_loss[loss=0.2681, simple_loss=0.3405, pruned_loss=0.09789, over 28505.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3468, pruned_loss=0.09846, over 5681631.33 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.34, pruned_loss=0.08761, over 3506572.19 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3475, pruned_loss=0.09967, over 5682748.24 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:50:40,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-11 05:50:46,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=959642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:50:58,764 INFO [optim.py:369] (1/2) 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 05:51:21,143 INFO [train.py:968] (1/2) Epoch 22, batch 1950, giga_loss[loss=0.2475, simple_loss=0.3187, pruned_loss=0.08819, over 28870.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3427, pruned_loss=0.09591, over 5685893.92 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3406, pruned_loss=0.08799, over 3577538.11 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3431, pruned_loss=0.09695, over 5680830.67 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:52:07,049 INFO [train.py:968] (1/2) Epoch 22, batch 2000, giga_loss[loss=0.2877, simple_loss=0.3432, pruned_loss=0.1161, over 26541.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.337, pruned_loss=0.09304, over 5677590.28 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3409, pruned_loss=0.0882, over 3642953.29 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3373, pruned_loss=0.09396, over 5670913.73 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:52:27,657 INFO [optim.py:369] (1/2) 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 05:52:51,933 INFO [train.py:968] (1/2) Epoch 22, batch 2050, giga_loss[loss=0.2283, simple_loss=0.2859, pruned_loss=0.08538, over 23363.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3315, pruned_loss=0.09022, over 5675435.14 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3401, pruned_loss=0.08774, over 3687796.29 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.332, pruned_loss=0.09129, over 5665956.09 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:52:55,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1732, 2.5420, 1.2624, 1.3332], device='cuda:1'), covar=tensor([0.1060, 0.0324, 0.0972, 0.1515], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0547, 0.0382, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 05:52:56,175 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=959785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:52:58,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=959788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:53:27,065 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 2100, giga_loss[loss=0.2575, simple_loss=0.3364, pruned_loss=0.0893, over 28868.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3301, pruned_loss=0.0897, over 5660680.68 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3406, pruned_loss=0.08787, over 3721103.94 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3301, pruned_loss=0.09049, over 5649875.12 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:53:56,652 INFO [optim.py:369] (1/2) 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 05:54:18,691 INFO [train.py:968] (1/2) Epoch 22, batch 2150, giga_loss[loss=0.2275, simple_loss=0.3061, pruned_loss=0.0744, over 28532.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3314, pruned_loss=0.08955, over 5678121.34 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3412, pruned_loss=0.08827, over 3764171.95 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3309, pruned_loss=0.08995, over 5665639.74 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:55:00,849 INFO [train.py:968] (1/2) Epoch 22, batch 2200, giga_loss[loss=0.2834, simple_loss=0.3509, pruned_loss=0.108, over 28039.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3316, pruned_loss=0.08958, over 5687669.27 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08903, over 3815603.73 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.33, pruned_loss=0.08948, over 5674349.40 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:55:01,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=959931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:55:20,977 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 2250, giga_loss[loss=0.2399, simple_loss=0.3153, pruned_loss=0.08229, over 28942.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3302, pruned_loss=0.08887, over 5695758.07 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3428, pruned_loss=0.08888, over 3866997.31 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3286, pruned_loss=0.08889, over 5680645.39 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:56:06,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9174, 2.1090, 2.1525, 1.6645], device='cuda:1'), covar=tensor([0.1850, 0.2388, 0.1450, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.0904, 0.0703, 0.0952, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 05:56:20,515 INFO [train.py:968] (1/2) Epoch 22, batch 2300, giga_loss[loss=0.2247, simple_loss=0.3108, pruned_loss=0.06936, over 29095.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3286, pruned_loss=0.08799, over 5703832.83 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.343, pruned_loss=0.08873, over 3947242.14 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3266, pruned_loss=0.08808, over 5692298.91 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:56:39,723 INFO [optim.py:369] (1/2) 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 05:56:54,794 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:56:54,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3658, 1.7298, 1.4584, 1.5290], device='cuda:1'), covar=tensor([0.0799, 0.0354, 0.0354, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 05:56:56,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:56:58,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:56:59,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2820, 3.0472, 2.9060, 1.3159], device='cuda:1'), covar=tensor([0.0946, 0.1155, 0.0942, 0.2427], device='cuda:1'), in_proj_covar=tensor([0.1205, 0.1120, 0.0949, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 05:57:01,163 INFO [train.py:968] (1/2) Epoch 22, batch 2350, giga_loss[loss=0.2478, simple_loss=0.3186, pruned_loss=0.08847, over 28809.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3251, pruned_loss=0.0861, over 5705173.84 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.343, pruned_loss=0.08873, over 3986006.90 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3232, pruned_loss=0.08615, over 5692689.89 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:57:10,701 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-11 05:57:22,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=960106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:57:36,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9672, 1.4960, 5.2465, 3.6709], device='cuda:1'), covar=tensor([0.1574, 0.2805, 0.0358, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0752, 0.0642, 0.0954, 0.0901], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 05:57:38,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4488, 1.4277, 1.1602, 1.0577], device='cuda:1'), covar=tensor([0.0797, 0.0425, 0.0918, 0.1122], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0447, 0.0518, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 05:57:43,871 INFO [train.py:968] (1/2) Epoch 22, batch 2400, giga_loss[loss=0.228, simple_loss=0.3018, pruned_loss=0.07707, over 28306.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3237, pruned_loss=0.0856, over 5706402.80 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3436, pruned_loss=0.08866, over 4042709.05 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3213, pruned_loss=0.08559, over 5690972.83 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:58:01,947 INFO [optim.py:369] (1/2) 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 05:58:02,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 05:58:02,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2150, 1.1961, 1.1488, 1.4576], device='cuda:1'), covar=tensor([0.0806, 0.0368, 0.0360, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 05:58:21,953 INFO [train.py:968] (1/2) Epoch 22, batch 2450, giga_loss[loss=0.2289, simple_loss=0.3093, pruned_loss=0.0742, over 29010.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3232, pruned_loss=0.08545, over 5711697.47 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3452, pruned_loss=0.08962, over 4092926.62 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3196, pruned_loss=0.08474, over 5698585.31 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:58:47,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2580, 1.3381, 1.4505, 1.2712], device='cuda:1'), covar=tensor([0.3451, 0.3033, 0.2229, 0.3115], device='cuda:1'), in_proj_covar=tensor([0.1949, 0.1873, 0.1812, 0.1948], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 05:59:00,119 INFO [train.py:968] (1/2) Epoch 22, batch 2500, giga_loss[loss=0.2384, simple_loss=0.3168, pruned_loss=0.07998, over 28608.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3208, pruned_loss=0.08459, over 5718066.97 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3453, pruned_loss=0.08945, over 4119988.12 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3176, pruned_loss=0.08407, over 5705212.46 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 05:59:19,283 INFO [optim.py:369] (1/2) 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 05:59:40,465 INFO [train.py:968] (1/2) Epoch 22, batch 2550, giga_loss[loss=0.2161, simple_loss=0.2946, pruned_loss=0.06881, over 28809.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3184, pruned_loss=0.08353, over 5716398.30 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3457, pruned_loss=0.08982, over 4128033.65 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3154, pruned_loss=0.08286, over 5713338.24 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 05:59:44,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1805, 1.5392, 1.5099, 1.0768], device='cuda:1'), covar=tensor([0.1600, 0.2820, 0.1454, 0.1791], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0706, 0.0954, 0.0850], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 05:59:58,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8049, 0.9347, 0.8148, 0.7544], device='cuda:1'), covar=tensor([0.2061, 0.2356, 0.1658, 0.2259], device='cuda:1'), in_proj_covar=tensor([0.1953, 0.1876, 0.1815, 0.1951], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 06:00:18,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:00:19,492 INFO [train.py:968] (1/2) Epoch 22, batch 2600, giga_loss[loss=0.2224, simple_loss=0.3051, pruned_loss=0.06991, over 28997.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3183, pruned_loss=0.0835, over 5711744.91 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3469, pruned_loss=0.09032, over 4170908.83 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.3145, pruned_loss=0.08249, over 5712795.57 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:00:38,571 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.5571, 1.8297, 1.7821, 1.5174], device='cuda:1'), covar=tensor([0.2576, 0.2132, 0.1582, 0.2115], device='cuda:1'), in_proj_covar=tensor([0.1953, 0.1875, 0.1813, 0.1948], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 06:00:57,691 INFO [train.py:968] (1/2) Epoch 22, batch 2650, giga_loss[loss=0.2293, simple_loss=0.2981, pruned_loss=0.08028, over 28792.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3179, pruned_loss=0.08346, over 5720742.17 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3474, pruned_loss=0.0903, over 4222818.25 frames. ], giga_tot_loss[loss=0.2393, simple_loss=0.3138, pruned_loss=0.08246, over 5715951.68 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:01:02,428 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960388.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 06:01:03,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.54 vs. limit=5.0
+2023-03-11 06:01:18,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7167, 1.9728, 1.5264, 1.9956], device='cuda:1'), covar=tensor([0.2684, 0.2785, 0.3137, 0.2521], device='cuda:1'), in_proj_covar=tensor([0.1510, 0.1094, 0.1334, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 06:01:40,460 INFO [train.py:968] (1/2) Epoch 22, batch 2700, giga_loss[loss=0.2814, simple_loss=0.3473, pruned_loss=0.1078, over 29059.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3219, pruned_loss=0.08618, over 5711256.09 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3472, pruned_loss=0.09014, over 4252638.65 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3181, pruned_loss=0.08538, over 5713040.31 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:01:54,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:02:00,386 INFO [optim.py:369] (1/2) 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 06:02:22,430 INFO [train.py:968] (1/2) Epoch 22, batch 2750, giga_loss[loss=0.2899, simple_loss=0.3608, pruned_loss=0.1095, over 28594.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3264, pruned_loss=0.08884, over 5713878.44 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3466, pruned_loss=0.0897, over 4285183.92 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3234, pruned_loss=0.08847, over 5711986.98 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:02:24,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 06:02:29,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3639, 3.0506, 1.4712, 1.4540], device='cuda:1'), covar=tensor([0.1018, 0.0330, 0.0893, 0.1439], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0548, 0.0384, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 06:03:00,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4491, 5.2839, 5.0290, 2.3003], device='cuda:1'), covar=tensor([0.0372, 0.0502, 0.0563, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.1204, 0.1119, 0.0948, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 06:03:07,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:03:08,263 INFO [train.py:968] (1/2) Epoch 22, batch 2800, giga_loss[loss=0.3068, simple_loss=0.3748, pruned_loss=0.1194, over 27960.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3349, pruned_loss=0.094, over 5708642.93 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3474, pruned_loss=0.09009, over 4324930.59 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3315, pruned_loss=0.09346, over 5703173.11 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:03:31,434 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 2850, giga_loss[loss=0.2599, simple_loss=0.3433, pruned_loss=0.08829, over 29153.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3416, pruned_loss=0.09825, over 5694020.82 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3477, pruned_loss=0.0902, over 4344275.12 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3387, pruned_loss=0.09785, over 5696963.18 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:04:02,353 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:04:05,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:04:32,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=960622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:04:43,781 INFO [train.py:968] (1/2) Epoch 22, batch 2900, giga_loss[loss=0.274, simple_loss=0.3529, pruned_loss=0.09752, over 28945.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3448, pruned_loss=0.09848, over 5706130.35 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3475, pruned_loss=0.09022, over 4382486.95 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3424, pruned_loss=0.09832, over 5704217.71 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:04:59,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.97 vs. limit=5.0
+2023-03-11 06:05:05,617 INFO [optim.py:369] (1/2) 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 06:05:17,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:05:19,584 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 2950, giga_loss[loss=0.3594, simple_loss=0.4152, pruned_loss=0.1518, over 28350.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3495, pruned_loss=0.1006, over 5707651.01 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3476, pruned_loss=0.09032, over 4441335.48 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3475, pruned_loss=0.1008, over 5699685.51 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:05:44,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:05:47,149 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 3000, giga_loss[loss=0.2895, simple_loss=0.3661, pruned_loss=0.1065, over 28839.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3551, pruned_loss=0.1046, over 5688380.62 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3475, pruned_loss=0.09044, over 4489549.75 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3538, pruned_loss=0.1051, over 5677188.13 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:06:07,629 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 06:06:17,126 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 06:06:37,842 INFO [optim.py:369] (1/2) 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 06:06:43,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=960763.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:06:57,498 INFO [train.py:968] (1/2) Epoch 22, batch 3050, giga_loss[loss=0.2288, simple_loss=0.3213, pruned_loss=0.06814, over 28910.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3526, pruned_loss=0.1022, over 5688991.16 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3472, pruned_loss=0.09046, over 4507819.25 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3518, pruned_loss=0.1028, over 5685045.82 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:07:32,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 06:07:35,637 INFO [train.py:968] (1/2) Epoch 22, batch 3100, giga_loss[loss=0.2422, simple_loss=0.3326, pruned_loss=0.07587, over 28895.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3493, pruned_loss=0.0995, over 5690627.68 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3465, pruned_loss=0.09029, over 4575997.48 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1007, over 5688455.74 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:07:48,972 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:07:51,639 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:07:54,129 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=960854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:07:56,759 INFO [optim.py:369] (1/2) 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 06:07:59,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-11 06:08:16,515 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 3150, giga_loss[loss=0.3377, simple_loss=0.3954, pruned_loss=0.14, over 27522.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3475, pruned_loss=0.09779, over 5694435.85 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3465, pruned_loss=0.09048, over 4605951.25 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3476, pruned_loss=0.09881, over 5695075.77 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:08:36,198 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=960906.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 06:08:38,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=960909.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:08:59,343 INFO [train.py:968] (1/2) Epoch 22, batch 3200, libri_loss[loss=0.2445, simple_loss=0.3372, pruned_loss=0.07591, over 29194.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3468, pruned_loss=0.09672, over 5700858.87 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.346, pruned_loss=0.09012, over 4647385.40 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3473, pruned_loss=0.09802, over 5697967.52 frames. ], batch size: 94, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:09:04,752 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:09:36,952 INFO [train.py:968] (1/2) Epoch 22, batch 3250, giga_loss[loss=0.2782, simple_loss=0.3521, pruned_loss=0.1022, over 28650.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3481, pruned_loss=0.0972, over 5699305.76 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3448, pruned_loss=0.08957, over 4702297.75 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3495, pruned_loss=0.09904, over 5698735.19 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:10:17,801 INFO [train.py:968] (1/2) Epoch 22, batch 3300, giga_loss[loss=0.2809, simple_loss=0.359, pruned_loss=0.1014, over 28795.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3489, pruned_loss=0.09756, over 5706844.33 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3449, pruned_loss=0.08993, over 4731862.16 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3501, pruned_loss=0.09898, over 5702086.30 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:10:34,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0058, 1.0815, 1.0961, 0.9338], device='cuda:1'), covar=tensor([0.2050, 0.2809, 0.1571, 0.2009], device='cuda:1'), in_proj_covar=tensor([0.1962, 0.1891, 0.1827, 0.1963], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 06:10:37,457 INFO [optim.py:369] (1/2) 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 06:11:01,467 INFO [train.py:968] (1/2) Epoch 22, batch 3350, giga_loss[loss=0.2601, simple_loss=0.3268, pruned_loss=0.09671, over 23926.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3496, pruned_loss=0.09855, over 5701768.61 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3446, pruned_loss=0.08986, over 4757605.30 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3507, pruned_loss=0.0999, over 5696738.06 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:11:14,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9922, 1.3218, 1.1051, 0.2408], device='cuda:1'), covar=tensor([0.3794, 0.2789, 0.4088, 0.6031], device='cuda:1'), in_proj_covar=tensor([0.1730, 0.1627, 0.1584, 0.1411], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 06:11:42,594 INFO [train.py:968] (1/2) Epoch 22, batch 3400, libri_loss[loss=0.2721, simple_loss=0.3517, pruned_loss=0.09624, over 29554.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3509, pruned_loss=0.09997, over 5709150.03 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3446, pruned_loss=0.0899, over 4788375.89 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 5702435.87 frames. ], batch size: 83, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:11:55,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3436, 3.1813, 2.9966, 1.4573], device='cuda:1'), covar=tensor([0.0869, 0.1013, 0.0853, 0.2244], device='cuda:1'), in_proj_covar=tensor([0.1206, 0.1127, 0.0951, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 06:12:05,129 INFO [optim.py:369] (1/2) 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 06:12:25,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-11 06:12:25,766 INFO [train.py:968] (1/2) Epoch 22, batch 3450, giga_loss[loss=0.2524, simple_loss=0.3312, pruned_loss=0.08678, over 28493.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3517, pruned_loss=0.1007, over 5713741.09 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3445, pruned_loss=0.08976, over 4806802.69 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3529, pruned_loss=0.1021, over 5711866.32 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:13:02,988 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 3500, giga_loss[loss=0.2668, simple_loss=0.343, pruned_loss=0.09527, over 28796.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3511, pruned_loss=0.09997, over 5710970.04 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.08961, over 4824030.81 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3523, pruned_loss=0.1014, over 5714817.33 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:13:13,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3307, 1.5014, 1.5736, 1.2064], device='cuda:1'), covar=tensor([0.1712, 0.2501, 0.1391, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0704, 0.0952, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 06:13:26,864 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 3550, giga_loss[loss=0.2856, simple_loss=0.3639, pruned_loss=0.1037, over 28770.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3508, pruned_loss=0.0988, over 5713999.57 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3441, pruned_loss=0.0895, over 4850919.47 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3521, pruned_loss=0.1003, over 5712357.32 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:13:45,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-11 06:14:27,493 INFO [train.py:968] (1/2) Epoch 22, batch 3600, giga_loss[loss=0.2427, simple_loss=0.3293, pruned_loss=0.07803, over 28890.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3503, pruned_loss=0.09759, over 5715941.64 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3436, pruned_loss=0.08914, over 4878759.98 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.352, pruned_loss=0.09932, over 5712369.03 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:14:45,362 INFO [optim.py:369] (1/2) 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 06:14:57,768 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=961372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:14:58,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 06:14:59,593 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 3650, giga_loss[loss=0.2455, simple_loss=0.3267, pruned_loss=0.0822, over 28523.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3498, pruned_loss=0.09693, over 5720881.27 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08891, over 4912378.52 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3513, pruned_loss=0.09876, over 5713636.62 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:15:03,815 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-11 06:15:21,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=961404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:15:25,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5737, 2.6193, 2.7180, 2.3880], device='cuda:1'), covar=tensor([0.1895, 0.2286, 0.1795, 0.2021], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0751, 0.0716, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 06:15:43,825 INFO [train.py:968] (1/2) Epoch 22, batch 3700, giga_loss[loss=0.2899, simple_loss=0.3523, pruned_loss=0.1137, over 27637.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3484, pruned_loss=0.09655, over 5727069.32 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08871, over 4941437.54 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3499, pruned_loss=0.09838, over 5716791.93 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:15:52,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7122, 1.9471, 1.6175, 1.7083], device='cuda:1'), covar=tensor([0.2614, 0.2626, 0.2928, 0.2396], device='cuda:1'), in_proj_covar=tensor([0.1510, 0.1093, 0.1331, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 06:16:04,384 INFO [optim.py:369] (1/2) 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 06:16:10,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4417, 1.6346, 1.6530, 1.2426], device='cuda:1'), covar=tensor([0.1969, 0.3079, 0.1693, 0.2101], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0704, 0.0951, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 06:16:22,748 INFO [train.py:968] (1/2) Epoch 22, batch 3750, giga_loss[loss=0.2643, simple_loss=0.3353, pruned_loss=0.09664, over 28800.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3465, pruned_loss=0.09557, over 5728514.19 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3439, pruned_loss=0.08893, over 4974073.58 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3474, pruned_loss=0.0971, over 5715029.81 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:16:59,314 INFO [train.py:968] (1/2) Epoch 22, batch 3800, libri_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.08897, over 29575.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3452, pruned_loss=0.09525, over 5736154.00 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3434, pruned_loss=0.08878, over 5006287.06 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3464, pruned_loss=0.09681, over 5719771.09 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:17:18,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0663, 1.2390, 3.3041, 3.0304], device='cuda:1'), covar=tensor([0.1676, 0.2742, 0.0519, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0641, 0.0957, 0.0906], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 06:17:24,191 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 3850, libri_loss[loss=0.2291, simple_loss=0.3168, pruned_loss=0.07072, over 29550.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3456, pruned_loss=0.09568, over 5727826.23 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08885, over 5018385.12 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3465, pruned_loss=0.09703, over 5719628.41 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:18:22,625 INFO [train.py:968] (1/2) Epoch 22, batch 3900, libri_loss[loss=0.295, simple_loss=0.3672, pruned_loss=0.1114, over 29536.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3467, pruned_loss=0.09614, over 5733907.14 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08886, over 5036008.43 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3475, pruned_loss=0.09734, over 5724055.26 frames. ], batch size: 89, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:18:41,327 INFO [optim.py:369] (1/2) 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 06:19:00,522 INFO [train.py:968] (1/2) Epoch 22, batch 3950, giga_loss[loss=0.2797, simple_loss=0.3499, pruned_loss=0.1048, over 28582.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3473, pruned_loss=0.09621, over 5718745.01 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3437, pruned_loss=0.08904, over 5060740.92 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3479, pruned_loss=0.09734, over 5716136.31 frames. ], batch size: 85, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:19:40,581 INFO [train.py:968] (1/2) Epoch 22, batch 4000, giga_loss[loss=0.2481, simple_loss=0.3269, pruned_loss=0.08467, over 28956.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3459, pruned_loss=0.09505, over 5724070.03 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3433, pruned_loss=0.08882, over 5084549.24 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3468, pruned_loss=0.09632, over 5718107.92 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:19:56,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-11 06:20:01,699 INFO [optim.py:369] (1/2) 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 06:20:20,650 INFO [train.py:968] (1/2) Epoch 22, batch 4050, giga_loss[loss=0.2453, simple_loss=0.3286, pruned_loss=0.08098, over 28578.00 frames. ], tot_loss[loss=0.268, simple_loss=0.345, pruned_loss=0.09548, over 5722802.99 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.343, pruned_loss=0.08864, over 5095914.02 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.346, pruned_loss=0.0967, over 5716536.51 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:20:35,534 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961800.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:21:00,364 INFO [train.py:968] (1/2) Epoch 22, batch 4100, libri_loss[loss=0.2115, simple_loss=0.2966, pruned_loss=0.06325, over 28544.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.342, pruned_loss=0.09391, over 5708892.61 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3427, pruned_loss=0.0884, over 5104343.90 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3431, pruned_loss=0.09521, over 5709864.09 frames. ], batch size: 63, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:21:08,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5011, 1.7371, 1.6720, 1.5990], device='cuda:1'), covar=tensor([0.2044, 0.2093, 0.2430, 0.2087], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0747, 0.0711, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 06:21:22,489 INFO [optim.py:369] (1/2) 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 06:21:37,364 INFO [train.py:968] (1/2) Epoch 22, batch 4150, libri_loss[loss=0.2483, simple_loss=0.3356, pruned_loss=0.08047, over 29532.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3393, pruned_loss=0.09256, over 5711136.74 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3426, pruned_loss=0.08822, over 5120552.26 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3402, pruned_loss=0.09383, over 5707965.08 frames. ], batch size: 81, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:21:38,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4388, 3.4520, 1.5795, 1.5623], device='cuda:1'), covar=tensor([0.1013, 0.0269, 0.0876, 0.1356], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0545, 0.0382, 0.0423], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 06:21:50,714 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:22:16,589 INFO [train.py:968] (1/2) Epoch 22, batch 4200, giga_loss[loss=0.2749, simple_loss=0.3418, pruned_loss=0.104, over 28883.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09372, over 5711317.67 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3427, pruned_loss=0.08838, over 5136356.00 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3401, pruned_loss=0.09469, over 5705188.48 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:22:37,748 INFO [optim.py:369] (1/2) 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 06:22:47,478 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:22:54,556 INFO [train.py:968] (1/2) Epoch 22, batch 4250, giga_loss[loss=0.2771, simple_loss=0.3522, pruned_loss=0.101, over 28735.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.339, pruned_loss=0.09387, over 5707590.22 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3429, pruned_loss=0.0886, over 5155559.62 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3392, pruned_loss=0.09461, over 5700861.53 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:23:03,786 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=961991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:23:06,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9451, 1.1272, 1.0502, 0.8726], device='cuda:1'), covar=tensor([0.2380, 0.2759, 0.1775, 0.2347], device='cuda:1'), in_proj_covar=tensor([0.1972, 0.1900, 0.1834, 0.1975], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 06:23:37,650 INFO [train.py:968] (1/2) Epoch 22, batch 4300, giga_loss[loss=0.2636, simple_loss=0.332, pruned_loss=0.09762, over 28270.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3375, pruned_loss=0.09375, over 5707448.36 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3424, pruned_loss=0.08831, over 5170168.89 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3379, pruned_loss=0.0947, over 5698669.77 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:23:46,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2495, 4.0599, 3.8335, 1.6641], device='cuda:1'), covar=tensor([0.0681, 0.0895, 0.0849, 0.2374], device='cuda:1'), in_proj_covar=tensor([0.1208, 0.1122, 0.0950, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 06:23:58,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5108, 1.7881, 1.4252, 1.6192], device='cuda:1'), covar=tensor([0.2671, 0.2736, 0.3145, 0.2496], device='cuda:1'), in_proj_covar=tensor([0.1510, 0.1092, 0.1330, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 06:24:00,908 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 4350, giga_loss[loss=0.2632, simple_loss=0.335, pruned_loss=0.09564, over 28376.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3345, pruned_loss=0.09262, over 5715187.20 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3422, pruned_loss=0.08816, over 5181302.03 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3349, pruned_loss=0.09355, over 5705566.34 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:24:56,844 INFO [train.py:968] (1/2) Epoch 22, batch 4400, giga_loss[loss=0.2409, simple_loss=0.3201, pruned_loss=0.08085, over 28952.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3327, pruned_loss=0.09206, over 5713777.80 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3423, pruned_loss=0.08823, over 5188561.80 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3329, pruned_loss=0.09277, over 5704305.89 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:25:07,447 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=962143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:25:10,227 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-11 06:25:21,114 INFO [optim.py:369] (1/2) 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 06:25:31,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6668, 4.5025, 4.3132, 2.0738], device='cuda:1'), covar=tensor([0.0583, 0.0724, 0.0852, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.1213, 0.1126, 0.0956, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 06:25:31,815 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 4450, giga_loss[loss=0.2632, simple_loss=0.3274, pruned_loss=0.09945, over 28733.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3309, pruned_loss=0.09084, over 5713290.20 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3427, pruned_loss=0.08844, over 5199885.47 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3305, pruned_loss=0.09129, over 5707473.71 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:26:06,879 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 06:26:08,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3743, 1.9595, 1.5473, 1.5053], device='cuda:1'), covar=tensor([0.0765, 0.0283, 0.0340, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0186, 0.0117, 0.0116, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0097, 0.0069, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 06:26:16,582 INFO [train.py:968] (1/2) Epoch 22, batch 4500, giga_loss[loss=0.2916, simple_loss=0.3728, pruned_loss=0.1052, over 28645.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3343, pruned_loss=0.09214, over 5716286.61 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3427, pruned_loss=0.08857, over 5216823.82 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3337, pruned_loss=0.09248, over 5707846.60 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:26:39,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-11 06:26:42,467 INFO [optim.py:369] (1/2) 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 06:26:52,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:27:00,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-11 06:27:00,981 INFO [train.py:968] (1/2) Epoch 22, batch 4550, giga_loss[loss=0.2692, simple_loss=0.3514, pruned_loss=0.09348, over 28693.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3376, pruned_loss=0.09367, over 5704612.32 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3427, pruned_loss=0.08858, over 5219368.70 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3371, pruned_loss=0.09394, over 5697918.16 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:27:29,209 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962318.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 06:27:30,944 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 4600, giga_loss[loss=0.2628, simple_loss=0.3431, pruned_loss=0.09127, over 28956.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3398, pruned_loss=0.09412, over 5710491.60 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3428, pruned_loss=0.08884, over 5242140.43 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3392, pruned_loss=0.0943, over 5701688.76 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:27:51,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:27:55,308 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:28:12,096 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 4650, giga_loss[loss=0.2646, simple_loss=0.3458, pruned_loss=0.09171, over 28934.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3408, pruned_loss=0.09387, over 5698236.05 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3424, pruned_loss=0.08866, over 5248401.89 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3406, pruned_loss=0.09424, over 5691665.70 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:28:51,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:28:53,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:29:06,828 INFO [train.py:968] (1/2) Epoch 22, batch 4700, giga_loss[loss=0.2521, simple_loss=0.3385, pruned_loss=0.08282, over 28662.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3403, pruned_loss=0.09323, over 5703640.07 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3428, pruned_loss=0.08882, over 5264318.21 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3398, pruned_loss=0.09349, over 5694238.03 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:29:14,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1527, 3.9934, 3.7812, 2.0339], device='cuda:1'), covar=tensor([0.0654, 0.0772, 0.0752, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.1127, 0.0956, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 06:29:19,516 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 4750, giga_loss[loss=0.3434, simple_loss=0.4079, pruned_loss=0.1394, over 28547.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3395, pruned_loss=0.09299, over 5706389.72 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.343, pruned_loss=0.08893, over 5276581.78 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3389, pruned_loss=0.0932, over 5697410.47 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:29:54,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:29:56,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:06,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3983, 2.0369, 1.5138, 0.7636], device='cuda:1'), covar=tensor([0.5845, 0.2678, 0.4056, 0.6193], device='cuda:1'), in_proj_covar=tensor([0.1738, 0.1624, 0.1589, 0.1415], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 06:30:10,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:12,317 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-11 06:30:13,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:17,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=962518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:19,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=962521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:27,323 INFO [train.py:968] (1/2) Epoch 22, batch 4800, giga_loss[loss=0.2466, simple_loss=0.3272, pruned_loss=0.08299, over 28983.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3401, pruned_loss=0.09342, over 5704987.67 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3431, pruned_loss=0.08892, over 5289030.86 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3394, pruned_loss=0.09366, over 5694309.57 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:30:29,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-11 06:30:36,993 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=962541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:30:49,431 INFO [optim.py:369] (1/2) 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 06:31:05,547 INFO [train.py:968] (1/2) Epoch 22, batch 4850, giga_loss[loss=0.3058, simple_loss=0.378, pruned_loss=0.1168, over 28723.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3418, pruned_loss=0.09447, over 5707562.43 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3426, pruned_loss=0.08874, over 5312136.52 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3416, pruned_loss=0.09503, over 5692750.49 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:31:48,331 INFO [train.py:968] (1/2) Epoch 22, batch 4900, giga_loss[loss=0.2969, simple_loss=0.3713, pruned_loss=0.1112, over 28802.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3464, pruned_loss=0.09727, over 5706089.81 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3432, pruned_loss=0.08908, over 5316593.25 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3458, pruned_loss=0.0975, over 5693919.91 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:32:12,055 INFO [optim.py:369] (1/2) 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 06:32:12,713 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=962661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:32:15,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=962664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:32:28,719 INFO [train.py:968] (1/2) Epoch 22, batch 4950, giga_loss[loss=0.2504, simple_loss=0.3339, pruned_loss=0.08348, over 28832.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.349, pruned_loss=0.09825, over 5716965.61 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3433, pruned_loss=0.08915, over 5324431.65 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3485, pruned_loss=0.0985, over 5705702.29 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:32:39,698 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 5000, giga_loss[loss=0.3075, simple_loss=0.3798, pruned_loss=0.1176, over 28903.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3503, pruned_loss=0.0989, over 5705751.47 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3435, pruned_loss=0.08929, over 5317471.49 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3498, pruned_loss=0.09902, over 5704154.76 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:33:37,310 INFO [optim.py:369] (1/2) 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 06:33:39,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=962763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:33:52,349 INFO [train.py:968] (1/2) Epoch 22, batch 5050, giga_loss[loss=0.2511, simple_loss=0.3269, pruned_loss=0.08767, over 28546.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3504, pruned_loss=0.09875, over 5717530.75 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3439, pruned_loss=0.08933, over 5334389.91 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3498, pruned_loss=0.09906, over 5711487.79 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:34:32,943 INFO [train.py:968] (1/2) Epoch 22, batch 5100, giga_loss[loss=0.2682, simple_loss=0.3428, pruned_loss=0.09679, over 28870.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3485, pruned_loss=0.09757, over 5724342.21 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3436, pruned_loss=0.08921, over 5345816.50 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3483, pruned_loss=0.09809, over 5716133.29 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:34:54,995 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:34:59,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3675, 1.5730, 1.1989, 1.1396], device='cuda:1'), covar=tensor([0.0943, 0.0525, 0.1062, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0445, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 06:35:11,164 INFO [train.py:968] (1/2) Epoch 22, batch 5150, libri_loss[loss=0.2461, simple_loss=0.3352, pruned_loss=0.07846, over 29521.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3465, pruned_loss=0.09643, over 5720306.77 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3437, pruned_loss=0.08914, over 5362593.56 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3465, pruned_loss=0.09721, over 5711774.53 frames. ], batch size: 79, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:35:13,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-11 06:35:14,982 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-11 06:35:38,438 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-11 06:35:50,675 INFO [train.py:968] (1/2) Epoch 22, batch 5200, giga_loss[loss=0.2663, simple_loss=0.335, pruned_loss=0.09877, over 28885.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3437, pruned_loss=0.09519, over 5727405.73 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.344, pruned_loss=0.08939, over 5374037.04 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3435, pruned_loss=0.09576, over 5718419.36 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:36:08,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3302, 3.0514, 1.5022, 1.3725], device='cuda:1'), covar=tensor([0.0973, 0.0349, 0.0920, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0550, 0.0384, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 06:36:15,058 INFO [optim.py:369] (1/2) 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 06:36:18,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5411, 2.4915, 2.4108, 2.1255], device='cuda:1'), covar=tensor([0.1838, 0.2428, 0.2046, 0.2457], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0748, 0.0713, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 06:36:29,664 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 5250, giga_loss[loss=0.2409, simple_loss=0.3346, pruned_loss=0.07361, over 28898.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3408, pruned_loss=0.09342, over 5728448.68 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3437, pruned_loss=0.08926, over 5384217.11 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3408, pruned_loss=0.09409, over 5719808.88 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:37:10,541 INFO [train.py:968] (1/2) Epoch 22, batch 5300, giga_loss[loss=0.2387, simple_loss=0.3147, pruned_loss=0.08131, over 23832.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3406, pruned_loss=0.09273, over 5718192.35 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3435, pruned_loss=0.08917, over 5394411.32 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3407, pruned_loss=0.09342, over 5708188.54 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:37:36,221 INFO [optim.py:369] (1/2) 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 06:37:53,734 INFO [train.py:968] (1/2) Epoch 22, batch 5350, giga_loss[loss=0.2428, simple_loss=0.3327, pruned_loss=0.0765, over 29043.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3418, pruned_loss=0.09245, over 5714361.93 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3434, pruned_loss=0.08916, over 5397564.22 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.342, pruned_loss=0.09303, over 5706435.15 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:38:23,924 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6183, 1.8314, 1.4993, 1.7415], device='cuda:1'), covar=tensor([0.2590, 0.2686, 0.3059, 0.2444], device='cuda:1'), in_proj_covar=tensor([0.1503, 0.1085, 0.1322, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 06:38:35,768 INFO [train.py:968] (1/2) Epoch 22, batch 5400, giga_loss[loss=0.2593, simple_loss=0.3151, pruned_loss=0.1017, over 28484.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3421, pruned_loss=0.09345, over 5709402.14 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08908, over 5402792.76 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3425, pruned_loss=0.09403, over 5701192.46 frames. ], batch size: 65, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:38:41,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:38:48,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-11 06:38:57,446 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:38:59,078 INFO [optim.py:369] (1/2) 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 06:39:02,728 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:39:02,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5613, 1.8148, 1.5137, 1.5574], device='cuda:1'), covar=tensor([0.2327, 0.2283, 0.2473, 0.2221], device='cuda:1'), in_proj_covar=tensor([0.1504, 0.1087, 0.1324, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 06:39:15,162 INFO [train.py:968] (1/2) Epoch 22, batch 5450, giga_loss[loss=0.2754, simple_loss=0.3513, pruned_loss=0.09972, over 28189.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3415, pruned_loss=0.09427, over 5708402.07 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3433, pruned_loss=0.08906, over 5409774.74 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3417, pruned_loss=0.09482, over 5701143.02 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:39:54,937 INFO [train.py:968] (1/2) Epoch 22, batch 5500, giga_loss[loss=0.2597, simple_loss=0.3372, pruned_loss=0.09109, over 28572.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.341, pruned_loss=0.095, over 5698325.40 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3436, pruned_loss=0.08931, over 5413704.12 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3408, pruned_loss=0.0954, over 5698923.68 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:39:57,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:40:34,601 INFO [train.py:968] (1/2) Epoch 22, batch 5550, giga_loss[loss=0.2851, simple_loss=0.3524, pruned_loss=0.1088, over 28973.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3394, pruned_loss=0.09498, over 5704747.94 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3435, pruned_loss=0.08934, over 5430035.14 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3391, pruned_loss=0.09545, over 5699572.81 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:40:34,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:40:37,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:00,957 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 5600, giga_loss[loss=0.2541, simple_loss=0.3301, pruned_loss=0.08908, over 29060.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3383, pruned_loss=0.09455, over 5707709.72 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08972, over 5443998.81 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3374, pruned_loss=0.0948, over 5700904.87 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:41:19,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4443, 1.2266, 4.1486, 3.4881], device='cuda:1'), covar=tensor([0.1603, 0.2822, 0.0442, 0.0768], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0642, 0.0956, 0.0904], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 06:41:38,177 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:41:53,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:55,186 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:41:57,862 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 5650, giga_loss[loss=0.2328, simple_loss=0.3143, pruned_loss=0.07565, over 28311.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3366, pruned_loss=0.09359, over 5712347.78 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3442, pruned_loss=0.08975, over 5452122.50 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3357, pruned_loss=0.09388, over 5706822.70 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:42:21,066 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 5700, giga_loss[loss=0.2524, simple_loss=0.3282, pruned_loss=0.08825, over 29110.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3318, pruned_loss=0.09146, over 5720008.46 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.08966, over 5455628.18 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3312, pruned_loss=0.09179, over 5714919.15 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:43:04,919 INFO [optim.py:369] (1/2) 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 06:43:18,697 INFO [train.py:968] (1/2) Epoch 22, batch 5750, giga_loss[loss=0.233, simple_loss=0.3123, pruned_loss=0.07682, over 28727.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3274, pruned_loss=0.08897, over 5713684.55 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3437, pruned_loss=0.08949, over 5457262.52 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3268, pruned_loss=0.08939, over 5716009.43 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:43:31,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:43:33,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:43:55,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:43:55,540 INFO [train.py:968] (1/2) Epoch 22, batch 5800, giga_loss[loss=0.2617, simple_loss=0.3346, pruned_loss=0.09442, over 28830.00 frames. ], tot_loss[loss=0.253, simple_loss=0.328, pruned_loss=0.08902, over 5714672.87 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3436, pruned_loss=0.08948, over 5467456.13 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3271, pruned_loss=0.08935, over 5715098.79 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:43:58,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:44:02,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:44:20,105 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([3.7813, 3.6164, 3.4014, 1.7761], device='cuda:1'), covar=tensor([0.0731, 0.0843, 0.0765, 0.2142], device='cuda:1'), in_proj_covar=tensor([0.1220, 0.1135, 0.0962, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 06:44:26,642 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-11 06:44:34,382 INFO [train.py:968] (1/2) Epoch 22, batch 5850, giga_loss[loss=0.2473, simple_loss=0.3324, pruned_loss=0.08109, over 29061.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3314, pruned_loss=0.09039, over 5721572.00 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08982, over 5478690.44 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.33, pruned_loss=0.0904, over 5718490.98 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:44:56,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-11 06:45:13,514 INFO [train.py:968] (1/2) Epoch 22, batch 5900, giga_loss[loss=0.2709, simple_loss=0.341, pruned_loss=0.1004, over 28837.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3339, pruned_loss=0.09138, over 5721228.24 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3434, pruned_loss=0.08954, over 5483821.11 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3331, pruned_loss=0.09165, over 5717409.87 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:45:40,727 INFO [optim.py:369] (1/2) 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 06:45:51,806 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:45:53,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:45:54,172 INFO [train.py:968] (1/2) Epoch 22, batch 5950, giga_loss[loss=0.2876, simple_loss=0.3713, pruned_loss=0.1019, over 28717.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3379, pruned_loss=0.09309, over 5720533.61 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08968, over 5494652.33 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3368, pruned_loss=0.09326, over 5713798.08 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 06:45:56,718 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:45:59,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=963686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:46:09,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4991, 3.8935, 1.5789, 1.6247], device='cuda:1'), covar=tensor([0.0945, 0.0268, 0.0883, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0552, 0.0384, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 06:46:17,203 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:46:23,114 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 6000, giga_loss[loss=0.2834, simple_loss=0.3549, pruned_loss=0.1059, over 28808.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.34, pruned_loss=0.09408, over 5719443.47 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3436, pruned_loss=0.08967, over 5504863.18 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3391, pruned_loss=0.09431, over 5710020.30 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:46:36,745 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 06:46:45,308 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 06:47:01,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=963750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:47:12,702 INFO [optim.py:369] (1/2) 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 06:47:24,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4909, 2.2100, 1.6049, 0.6269], device='cuda:1'), covar=tensor([0.5465, 0.3198, 0.4147, 0.6596], device='cuda:1'), in_proj_covar=tensor([0.1746, 0.1634, 0.1593, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 06:47:24,758 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 6050, giga_loss[loss=0.2868, simple_loss=0.3594, pruned_loss=0.1072, over 28602.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.09558, over 5720412.07 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08965, over 5521268.96 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3417, pruned_loss=0.09601, over 5706173.47 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:47:28,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9814, 1.3052, 1.0864, 0.2143], device='cuda:1'), covar=tensor([0.3739, 0.2863, 0.4453, 0.6496], device='cuda:1'), in_proj_covar=tensor([0.1746, 0.1634, 0.1593, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 06:47:31,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2814, 1.5386, 1.2705, 1.4703], device='cuda:1'), covar=tensor([0.0746, 0.0357, 0.0343, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 06:47:35,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-11 06:48:09,941 INFO [train.py:968] (1/2) Epoch 22, batch 6100, giga_loss[loss=0.3118, simple_loss=0.3791, pruned_loss=0.1222, over 28801.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3479, pruned_loss=0.1001, over 5720908.78 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08958, over 5531057.70 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3475, pruned_loss=0.1007, over 5705268.10 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:48:43,082 INFO [optim.py:369] (1/2) 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 06:48:56,324 INFO [train.py:968] (1/2) Epoch 22, batch 6150, giga_loss[loss=0.413, simple_loss=0.462, pruned_loss=0.182, over 28309.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3544, pruned_loss=0.1056, over 5704542.44 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08975, over 5536723.33 frames. ], giga_tot_loss[loss=0.2834, simple_loss=0.3543, pruned_loss=0.1063, over 5690741.72 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:49:08,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=963893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:49:12,688 INFO [zipformer.py:1188] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.2152, 1.6053, 1.5321, 1.0877], device='cuda:1'), covar=tensor([0.1665, 0.2627, 0.1454, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0702, 0.0950, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 06:49:19,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:49:38,249 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=963925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:49:45,497 INFO [train.py:968] (1/2) Epoch 22, batch 6200, giga_loss[loss=0.2986, simple_loss=0.3723, pruned_loss=0.1125, over 28958.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3623, pruned_loss=0.1113, over 5695504.78 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08966, over 5543955.43 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3625, pruned_loss=0.1122, over 5681058.25 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:49:56,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7150, 1.6493, 1.9387, 1.4891], device='cuda:1'), covar=tensor([0.1384, 0.2068, 0.1162, 0.1536], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0701, 0.0948, 0.0845], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 06:50:15,888 INFO [optim.py:369] (1/2) 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 06:50:29,350 INFO [train.py:968] (1/2) Epoch 22, batch 6250, giga_loss[loss=0.339, simple_loss=0.3731, pruned_loss=0.1525, over 23567.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3672, pruned_loss=0.1158, over 5680472.32 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08983, over 5547680.99 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.368, pruned_loss=0.1172, over 5671276.02 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:50:32,799 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=963983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:51:15,730 INFO [train.py:968] (1/2) Epoch 22, batch 6300, giga_loss[loss=0.3242, simple_loss=0.3882, pruned_loss=0.1301, over 28948.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3733, pruned_loss=0.1207, over 5684572.21 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08966, over 5552961.85 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3744, pruned_loss=0.1224, over 5674338.92 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:51:45,028 INFO [optim.py:369] (1/2) 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 06:52:01,510 INFO [train.py:968] (1/2) Epoch 22, batch 6350, libri_loss[loss=0.2922, simple_loss=0.371, pruned_loss=0.1067, over 29541.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3757, pruned_loss=0.1228, over 5670945.80 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08936, over 5563301.11 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3783, pruned_loss=0.1257, over 5657923.34 frames. ], batch size: 89, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:52:51,621 INFO [train.py:968] (1/2) Epoch 22, batch 6400, giga_loss[loss=0.2996, simple_loss=0.3717, pruned_loss=0.1138, over 29055.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3783, pruned_loss=0.126, over 5660407.21 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08937, over 5567347.46 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3808, pruned_loss=0.1288, over 5647803.08 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:53:05,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3143, 1.1001, 4.0351, 3.3585], device='cuda:1'), covar=tensor([0.1708, 0.2851, 0.0447, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0643, 0.0960, 0.0910], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 06:53:18,279 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=964155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:53:28,201 INFO [optim.py:369] (1/2) 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 06:53:44,291 INFO [train.py:968] (1/2) Epoch 22, batch 6450, libri_loss[loss=0.249, simple_loss=0.3356, pruned_loss=0.08127, over 29587.00 frames. ], tot_loss[loss=0.321, simple_loss=0.3817, pruned_loss=0.1301, over 5642074.78 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08925, over 5571780.91 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3847, pruned_loss=0.1334, over 5629844.80 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:54:37,720 INFO [train.py:968] (1/2) Epoch 22, batch 6500, giga_loss[loss=0.3489, simple_loss=0.402, pruned_loss=0.1479, over 28774.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3852, pruned_loss=0.1339, over 5626517.45 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08939, over 5580474.17 frames. ], giga_tot_loss[loss=0.3318, simple_loss=0.3885, pruned_loss=0.1376, over 5609768.06 frames. ], batch size: 243, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:55:13,691 INFO [optim.py:369] (1/2) 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 06:55:27,271 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 6550, giga_loss[loss=0.3261, simple_loss=0.3822, pruned_loss=0.135, over 28805.00 frames. ], tot_loss[loss=0.3297, simple_loss=0.3879, pruned_loss=0.1358, over 5630945.81 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3429, pruned_loss=0.08942, over 5586798.01 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3915, pruned_loss=0.1395, over 5612543.88 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:55:48,365 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:55:51,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:56:17,056 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=964330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:56:17,379 INFO [train.py:968] (1/2) Epoch 22, batch 6600, giga_loss[loss=0.4035, simple_loss=0.434, pruned_loss=0.1865, over 27613.00 frames. ], tot_loss[loss=0.3301, simple_loss=0.3877, pruned_loss=0.1363, over 5638517.73 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3431, pruned_loss=0.0896, over 5590933.68 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3919, pruned_loss=0.1406, over 5621548.25 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:56:22,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3482, 3.0609, 1.4938, 1.4394], device='cuda:1'), covar=tensor([0.0966, 0.0325, 0.0819, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0554, 0.0384, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 06:56:25,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3895, 1.3856, 3.9756, 3.2897], device='cuda:1'), covar=tensor([0.1544, 0.2539, 0.0519, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0650, 0.0971, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 06:56:45,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=964358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:56:52,214 INFO [optim.py:369] (1/2) 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 06:57:05,955 INFO [train.py:968] (1/2) Epoch 22, batch 6650, giga_loss[loss=0.3212, simple_loss=0.3809, pruned_loss=0.1307, over 28927.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3854, pruned_loss=0.1349, over 5647400.21 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08964, over 5600435.87 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3898, pruned_loss=0.1396, over 5626731.36 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:57:44,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:57:48,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:57:54,618 INFO [train.py:968] (1/2) Epoch 22, batch 6700, giga_loss[loss=0.3019, simple_loss=0.3784, pruned_loss=0.1127, over 28969.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3855, pruned_loss=0.1345, over 5649531.27 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.08994, over 5612685.21 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3904, pruned_loss=0.1398, over 5623126.76 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:58:14,807 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=964453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:58:17,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7005, 1.7763, 1.8810, 1.4550], device='cuda:1'), covar=tensor([0.1753, 0.2430, 0.1416, 0.1735], device='cuda:1'), in_proj_covar=tensor([0.0894, 0.0700, 0.0941, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 06:58:23,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=964463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:58:26,159 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 6750, giga_loss[loss=0.3361, simple_loss=0.399, pruned_loss=0.1366, over 28921.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3858, pruned_loss=0.1337, over 5656390.16 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.08999, over 5616006.30 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.391, pruned_loss=0.1393, over 5632965.88 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 06:58:59,636 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:59:02,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 06:59:14,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2935, 1.6050, 1.2725, 0.7941], device='cuda:1'), covar=tensor([0.3351, 0.1784, 0.2017, 0.4368], device='cuda:1'), in_proj_covar=tensor([0.1756, 0.1642, 0.1601, 0.1423], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 06:59:29,772 INFO [train.py:968] (1/2) Epoch 22, batch 6800, giga_loss[loss=0.3498, simple_loss=0.4069, pruned_loss=0.1464, over 28256.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3868, pruned_loss=0.1344, over 5633519.05 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09006, over 5621839.41 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3917, pruned_loss=0.1396, over 5610263.18 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 06:59:31,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=964533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:00:01,769 INFO [optim.py:369] (1/2) 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 07:00:12,959 INFO [train.py:968] (1/2) Epoch 22, batch 6850, giga_loss[loss=0.2951, simple_loss=0.3737, pruned_loss=0.1082, over 28174.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3843, pruned_loss=0.1321, over 5635634.36 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09011, over 5628409.05 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3897, pruned_loss=0.1377, over 5611044.22 frames. ], batch size: 77, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:00:59,958 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.93 vs. limit=2.0
+2023-03-11 07:01:02,742 INFO [train.py:968] (1/2) Epoch 22, batch 6900, giga_loss[loss=0.3434, simple_loss=0.3862, pruned_loss=0.1503, over 26606.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3812, pruned_loss=0.1289, over 5631204.46 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3437, pruned_loss=0.09022, over 5632712.39 frames. ], giga_tot_loss[loss=0.3277, simple_loss=0.3865, pruned_loss=0.1344, over 5607923.60 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:01:36,941 INFO [optim.py:369] (1/2) 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 07:01:38,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-11 07:01:45,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 07:01:48,290 INFO [train.py:968] (1/2) Epoch 22, batch 6950, giga_loss[loss=0.2907, simple_loss=0.3643, pruned_loss=0.1086, over 28309.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3765, pruned_loss=0.124, over 5651039.11 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3432, pruned_loss=0.09005, over 5640324.59 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3824, pruned_loss=0.1297, over 5625720.05 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:02:36,815 INFO [train.py:968] (1/2) Epoch 22, batch 7000, giga_loss[loss=0.2891, simple_loss=0.3448, pruned_loss=0.1167, over 23524.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3744, pruned_loss=0.1225, over 5652241.09 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08992, over 5644372.86 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3796, pruned_loss=0.1276, over 5628649.10 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:03:06,900 INFO [optim.py:369] (1/2) 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 07:03:11,353 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=964770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:03:14,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8779, 1.0623, 2.8834, 2.7707], device='cuda:1'), covar=tensor([0.1652, 0.2598, 0.0590, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0764, 0.0650, 0.0968, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 07:03:22,727 INFO [train.py:968] (1/2) Epoch 22, batch 7050, giga_loss[loss=0.3002, simple_loss=0.3468, pruned_loss=0.1268, over 23637.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3714, pruned_loss=0.12, over 5661711.83 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.343, pruned_loss=0.08982, over 5652323.38 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3766, pruned_loss=0.1252, over 5636040.50 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:03:51,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3053, 1.6132, 1.2881, 0.9592], device='cuda:1'), covar=tensor([0.2847, 0.2711, 0.3224, 0.2416], device='cuda:1'), in_proj_covar=tensor([0.1499, 0.1084, 0.1323, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 07:04:01,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0819, 1.4670, 1.3474, 1.2716], device='cuda:1'), covar=tensor([0.1979, 0.1572, 0.2338, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0759, 0.0721, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 07:04:08,719 INFO [train.py:968] (1/2) Epoch 22, batch 7100, giga_loss[loss=0.3563, simple_loss=0.4108, pruned_loss=0.1509, over 28955.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3725, pruned_loss=0.1206, over 5665245.80 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3432, pruned_loss=0.08982, over 5654673.40 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3771, pruned_loss=0.1254, over 5642963.11 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:04:15,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=964838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:04:45,841 INFO [optim.py:369] (1/2) 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 07:04:54,587 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 07:05:03,440 INFO [train.py:968] (1/2) Epoch 22, batch 7150, giga_loss[loss=0.3002, simple_loss=0.3731, pruned_loss=0.1136, over 28673.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3731, pruned_loss=0.121, over 5662605.42 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08992, over 5651129.93 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3773, pruned_loss=0.1254, over 5648541.77 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:05:27,934 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=964907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:05:51,018 INFO [train.py:968] (1/2) Epoch 22, batch 7200, giga_loss[loss=0.2947, simple_loss=0.3732, pruned_loss=0.108, over 28928.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3717, pruned_loss=0.1195, over 5672520.05 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.344, pruned_loss=0.0904, over 5659053.85 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3758, pruned_loss=0.1239, over 5654729.75 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:06:22,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4175, 1.6131, 1.5951, 1.4824], device='cuda:1'), covar=tensor([0.1639, 0.1613, 0.1702, 0.1735], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0756, 0.0719, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 07:06:27,791 INFO [optim.py:369] (1/2) 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 07:06:44,767 INFO [train.py:968] (1/2) Epoch 22, batch 7250, giga_loss[loss=0.3152, simple_loss=0.3941, pruned_loss=0.1182, over 28906.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3721, pruned_loss=0.1177, over 5675285.62 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09021, over 5666433.93 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3764, pruned_loss=0.1222, over 5655022.57 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:06:45,080 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=964981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:06:48,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=964984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:07:18,745 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=965013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:07:35,503 INFO [train.py:968] (1/2) Epoch 22, batch 7300, giga_loss[loss=0.3234, simple_loss=0.3867, pruned_loss=0.13, over 29011.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3741, pruned_loss=0.1181, over 5671387.18 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3437, pruned_loss=0.09024, over 5665931.76 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3775, pruned_loss=0.1217, over 5656055.96 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:08:12,976 INFO [optim.py:369] (1/2) 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 07:08:27,950 INFO [train.py:968] (1/2) Epoch 22, batch 7350, giga_loss[loss=0.2841, simple_loss=0.3555, pruned_loss=0.1064, over 28729.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3741, pruned_loss=0.1187, over 5677932.10 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3437, pruned_loss=0.0902, over 5670372.58 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3774, pruned_loss=0.1222, over 5661912.62 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:09:04,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3896, 1.7604, 1.4610, 1.7075], device='cuda:1'), covar=tensor([0.0778, 0.0316, 0.0318, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 07:09:17,136 INFO [train.py:968] (1/2) Epoch 22, batch 7400, giga_loss[loss=0.2558, simple_loss=0.3322, pruned_loss=0.08965, over 28948.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3735, pruned_loss=0.1191, over 5676684.29 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3436, pruned_loss=0.09012, over 5671579.72 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3764, pruned_loss=0.122, over 5663131.63 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:09:30,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:09:54,136 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 7450, giga_loss[loss=0.2829, simple_loss=0.3498, pruned_loss=0.1079, over 28775.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3705, pruned_loss=0.1181, over 5667307.08 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3435, pruned_loss=0.09002, over 5670561.00 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3739, pruned_loss=0.1216, over 5657069.98 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:10:24,822 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-11 07:10:43,881 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4333, 1.5120, 1.2063, 1.0772], device='cuda:1'), covar=tensor([0.0934, 0.0566, 0.1068, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0449, 0.0520, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 07:10:54,465 INFO [train.py:968] (1/2) Epoch 22, batch 7500, giga_loss[loss=0.2785, simple_loss=0.351, pruned_loss=0.103, over 28651.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3687, pruned_loss=0.1174, over 5674894.12 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09015, over 5672411.88 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3716, pruned_loss=0.1204, over 5665284.90 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:11:31,498 INFO [optim.py:369] (1/2) 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 07:11:43,536 INFO [train.py:968] (1/2) Epoch 22, batch 7550, giga_loss[loss=0.3149, simple_loss=0.3738, pruned_loss=0.128, over 28759.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3676, pruned_loss=0.1154, over 5685126.27 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08995, over 5673244.81 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3705, pruned_loss=0.1184, over 5676780.35 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:11:45,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:11:52,009 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=965288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:11:55,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=965291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:12:10,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=965307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:12:22,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=965320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:12:31,720 INFO [train.py:968] (1/2) Epoch 22, batch 7600, giga_loss[loss=0.302, simple_loss=0.3743, pruned_loss=0.1149, over 28648.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3672, pruned_loss=0.1142, over 5694835.64 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08997, over 5678211.72 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3701, pruned_loss=0.1171, over 5683791.14 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:13:03,713 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 7650, giga_loss[loss=0.355, simple_loss=0.3952, pruned_loss=0.1574, over 26668.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3674, pruned_loss=0.1144, over 5692796.26 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09022, over 5680769.18 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.37, pruned_loss=0.1171, over 5681735.44 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:13:57,403 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=965425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:13:59,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=965428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:14:02,445 INFO [train.py:968] (1/2) Epoch 22, batch 7700, giga_loss[loss=0.2898, simple_loss=0.3544, pruned_loss=0.1126, over 28873.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3668, pruned_loss=0.1146, over 5693597.01 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3436, pruned_loss=0.09007, over 5681619.90 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3693, pruned_loss=0.1171, over 5684364.97 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:14:29,278 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:14:51,762 INFO [train.py:968] (1/2) Epoch 22, batch 7750, libri_loss[loss=0.2726, simple_loss=0.3562, pruned_loss=0.09444, over 29623.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3678, pruned_loss=0.1159, over 5697233.48 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09024, over 5685132.10 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.37, pruned_loss=0.1185, over 5686746.59 frames. ], batch size: 91, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:14:58,046 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 7800, libri_loss[loss=0.294, simple_loss=0.3742, pruned_loss=0.1069, over 25956.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3672, pruned_loss=0.1163, over 5686486.86 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.0901, over 5686222.99 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3698, pruned_loss=0.1191, over 5677524.91 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:15:45,902 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=965539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:16:11,908 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 7850, giga_loss[loss=0.3029, simple_loss=0.3709, pruned_loss=0.1175, over 28880.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3649, pruned_loss=0.1149, over 5701096.79 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3436, pruned_loss=0.0901, over 5693039.58 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3684, pruned_loss=0.1185, over 5687518.00 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:17:09,382 INFO [train.py:968] (1/2) Epoch 22, batch 7900, giga_loss[loss=0.3149, simple_loss=0.3751, pruned_loss=0.1274, over 28694.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3622, pruned_loss=0.1134, over 5703218.90 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3435, pruned_loss=0.09002, over 5697589.02 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3653, pruned_loss=0.1167, over 5688559.06 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:17:42,289 INFO [optim.py:369] (1/2) 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 07:17:44,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 07:17:52,260 INFO [train.py:968] (1/2) Epoch 22, batch 7950, giga_loss[loss=0.2768, simple_loss=0.3539, pruned_loss=0.09988, over 28818.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3615, pruned_loss=0.1135, over 5709971.24 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3439, pruned_loss=0.09025, over 5701708.62 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3642, pruned_loss=0.1167, over 5694705.88 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:17:54,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:18:14,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4498, 1.7669, 1.2734, 1.4336], device='cuda:1'), covar=tensor([0.0976, 0.0461, 0.1041, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0450, 0.0520, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 07:18:35,629 INFO [train.py:968] (1/2) Epoch 22, batch 8000, giga_loss[loss=0.2678, simple_loss=0.343, pruned_loss=0.0963, over 28993.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.362, pruned_loss=0.1139, over 5703165.05 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3442, pruned_loss=0.09036, over 5708934.96 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3647, pruned_loss=0.1173, over 5684501.50 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:19:11,062 INFO [optim.py:369] (1/2) 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 07:19:20,926 INFO [train.py:968] (1/2) Epoch 22, batch 8050, giga_loss[loss=0.3266, simple_loss=0.3844, pruned_loss=0.1344, over 28860.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3629, pruned_loss=0.1141, over 5701119.80 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09026, over 5713939.65 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3659, pruned_loss=0.1176, over 5681309.15 frames. ], batch size: 285, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:20:01,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=965825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:04,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=965828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:05,752 INFO [train.py:968] (1/2) Epoch 22, batch 8100, giga_loss[loss=0.3237, simple_loss=0.3881, pruned_loss=0.1297, over 28920.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3638, pruned_loss=0.1142, over 5692114.25 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3439, pruned_loss=0.09031, over 5715837.57 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3662, pruned_loss=0.1171, over 5674765.85 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:20:32,412 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=965857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:36,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:20:41,076 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 8150, giga_loss[loss=0.2773, simple_loss=0.3512, pruned_loss=0.1017, over 28871.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3658, pruned_loss=0.1155, over 5667703.09 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3444, pruned_loss=0.09056, over 5698585.09 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3678, pruned_loss=0.1181, over 5668117.11 frames. ], batch size: 66, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:21:25,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=965914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:21:41,015 INFO [train.py:968] (1/2) Epoch 22, batch 8200, giga_loss[loss=0.3752, simple_loss=0.4319, pruned_loss=0.1593, over 28577.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3666, pruned_loss=0.1161, over 5681931.53 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3441, pruned_loss=0.09042, over 5701532.40 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3688, pruned_loss=0.1188, over 5679194.17 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:22:14,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4003, 1.7591, 5.5460, 4.3856], device='cuda:1'), covar=tensor([0.1539, 0.2826, 0.0701, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0652, 0.0971, 0.0920], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 07:22:18,483 INFO [optim.py:369] (1/2) 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 07:22:31,724 INFO [train.py:968] (1/2) Epoch 22, batch 8250, giga_loss[loss=0.3106, simple_loss=0.3666, pruned_loss=0.1273, over 29072.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3676, pruned_loss=0.1179, over 5678839.07 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09015, over 5706575.86 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3704, pruned_loss=0.1209, over 5671547.59 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:22:55,817 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:22:57,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:15,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4476, 3.1046, 1.5118, 1.6537], device='cuda:1'), covar=tensor([0.0926, 0.0395, 0.0848, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0561, 0.0388, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 07:23:20,639 INFO [train.py:968] (1/2) Epoch 22, batch 8300, giga_loss[loss=0.4633, simple_loss=0.4588, pruned_loss=0.2339, over 26541.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3696, pruned_loss=0.1207, over 5677091.70 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.08999, over 5707529.03 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3728, pruned_loss=0.1242, over 5669757.63 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:23:27,151 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=966038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:46,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:48,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:23:51,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-11 07:23:57,019 INFO [optim.py:369] (1/2) 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 07:23:59,847 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-11 07:24:09,224 INFO [train.py:968] (1/2) Epoch 22, batch 8350, giga_loss[loss=0.2996, simple_loss=0.3604, pruned_loss=0.1194, over 29056.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3721, pruned_loss=0.1237, over 5671966.35 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08995, over 5713881.26 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3755, pruned_loss=0.1275, over 5659384.04 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:24:16,157 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=966089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:24:52,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=966125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:24:56,342 INFO [train.py:968] (1/2) Epoch 22, batch 8400, giga_loss[loss=0.262, simple_loss=0.3336, pruned_loss=0.09524, over 29103.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3709, pruned_loss=0.123, over 5675163.04 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.09004, over 5717699.78 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3741, pruned_loss=0.1265, over 5661005.46 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:25:29,609 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=966167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:25:31,271 INFO [optim.py:369] (1/2) 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 07:25:38,974 INFO [train.py:968] (1/2) Epoch 22, batch 8450, giga_loss[loss=0.2964, simple_loss=0.3869, pruned_loss=0.103, over 28990.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3694, pruned_loss=0.1214, over 5671738.62 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3432, pruned_loss=0.08987, over 5716023.02 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3725, pruned_loss=0.1248, over 5661429.26 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:26:21,491 INFO [train.py:968] (1/2) Epoch 22, batch 8500, giga_loss[loss=0.2772, simple_loss=0.3554, pruned_loss=0.09953, over 28992.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3701, pruned_loss=0.1207, over 5676193.63 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3432, pruned_loss=0.08985, over 5715035.22 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1242, over 5668116.00 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:26:51,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 07:26:53,183 INFO [optim.py:369] (1/2) 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 07:27:02,276 INFO [train.py:968] (1/2) Epoch 22, batch 8550, giga_loss[loss=0.2738, simple_loss=0.3402, pruned_loss=0.1037, over 28278.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3684, pruned_loss=0.1187, over 5669329.92 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.09, over 5713783.19 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3715, pruned_loss=0.1225, over 5662118.57 frames. ], batch size: 369, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:27:48,109 INFO [train.py:968] (1/2) Epoch 22, batch 8600, giga_loss[loss=0.3745, simple_loss=0.4141, pruned_loss=0.1674, over 27910.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3676, pruned_loss=0.1188, over 5678598.35 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.09, over 5718718.91 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3705, pruned_loss=0.1226, over 5667362.67 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:28:18,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6364, 1.7530, 1.7279, 1.4974], device='cuda:1'), covar=tensor([0.3032, 0.2514, 0.1990, 0.2665], device='cuda:1'), in_proj_covar=tensor([0.1976, 0.1919, 0.1846, 0.1976], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 07:28:22,509 INFO [optim.py:369] (1/2) 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 07:28:31,635 INFO [train.py:968] (1/2) Epoch 22, batch 8650, giga_loss[loss=0.4565, simple_loss=0.4624, pruned_loss=0.2253, over 26646.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.366, pruned_loss=0.1181, over 5681502.36 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.08988, over 5722963.11 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3692, pruned_loss=0.1221, over 5667363.33 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:28:45,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2727, 2.5271, 1.2408, 1.4320], device='cuda:1'), covar=tensor([0.1003, 0.0386, 0.0906, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0559, 0.0388, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 07:29:18,659 INFO [train.py:968] (1/2) Epoch 22, batch 8700, giga_loss[loss=0.3427, simple_loss=0.3823, pruned_loss=0.1515, over 23375.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.366, pruned_loss=0.1184, over 5664540.63 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08986, over 5729523.54 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3695, pruned_loss=0.123, over 5644649.48 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:29:53,137 INFO [optim.py:369] (1/2) 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 07:30:01,690 INFO [train.py:968] (1/2) Epoch 22, batch 8750, giga_loss[loss=0.304, simple_loss=0.3758, pruned_loss=0.1161, over 28683.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5672509.75 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3436, pruned_loss=0.08973, over 5725261.45 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3711, pruned_loss=0.123, over 5657602.83 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:30:19,051 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=966500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:30:20,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=966502.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 07:30:47,013 INFO [train.py:968] (1/2) Epoch 22, batch 8800, giga_loss[loss=0.2927, simple_loss=0.3694, pruned_loss=0.108, over 28984.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3716, pruned_loss=0.1185, over 5676837.64 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.08989, over 5729369.44 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.375, pruned_loss=0.1231, over 5660102.64 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:30:59,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=966542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:31:26,097 INFO [optim.py:369] (1/2) 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 07:31:36,626 INFO [train.py:968] (1/2) Epoch 22, batch 8850, giga_loss[loss=0.2816, simple_loss=0.3501, pruned_loss=0.1065, over 28456.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3739, pruned_loss=0.1192, over 5680588.07 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3444, pruned_loss=0.0901, over 5731184.45 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3767, pruned_loss=0.1229, over 5665170.44 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:32:17,742 INFO [train.py:968] (1/2) Epoch 22, batch 8900, giga_loss[loss=0.376, simple_loss=0.4234, pruned_loss=0.1643, over 28791.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.375, pruned_loss=0.12, over 5677692.24 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.0899, over 5723925.75 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3784, pruned_loss=0.1241, over 5669198.68 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:32:29,738 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:32:33,501 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:32:38,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9823, 1.1999, 1.3108, 1.0554], device='cuda:1'), covar=tensor([0.1477, 0.1141, 0.1821, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0753, 0.0716, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 07:32:57,975 INFO [optim.py:369] (1/2) 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 07:33:00,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=966675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:33:05,617 INFO [train.py:968] (1/2) Epoch 22, batch 8950, giga_loss[loss=0.3041, simple_loss=0.3749, pruned_loss=0.1167, over 28964.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3764, pruned_loss=0.1222, over 5657192.33 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3442, pruned_loss=0.09006, over 5716984.83 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3795, pruned_loss=0.1257, over 5656364.14 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:33:09,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=966685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:33:12,353 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=966688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:33:39,453 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 9000, giga_loss[loss=0.3875, simple_loss=0.4108, pruned_loss=0.1821, over 23487.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3743, pruned_loss=0.1217, over 5645762.81 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08991, over 5711638.87 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3776, pruned_loss=0.1253, over 5648706.25 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:33:54,508 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 07:33:59,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2838, 1.4516, 1.4401, 1.2396], device='cuda:1'), covar=tensor([0.2407, 0.2436, 0.1600, 0.2083], device='cuda:1'), in_proj_covar=tensor([0.1962, 0.1911, 0.1835, 0.1965], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 07:34:03,091 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 07:34:48,580 INFO [optim.py:369] (1/2) 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 07:34:56,971 INFO [train.py:968] (1/2) Epoch 22, batch 9050, giga_loss[loss=0.2577, simple_loss=0.3424, pruned_loss=0.08645, over 28882.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3734, pruned_loss=0.1226, over 5634671.82 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08991, over 5711638.87 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.376, pruned_loss=0.1254, over 5636962.73 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:35:23,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7102, 1.6786, 1.2700, 1.3035], device='cuda:1'), covar=tensor([0.0892, 0.0610, 0.1018, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0450, 0.0521, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 07:35:46,110 INFO [train.py:968] (1/2) Epoch 22, batch 9100, giga_loss[loss=0.2834, simple_loss=0.3538, pruned_loss=0.1065, over 28998.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3716, pruned_loss=0.1218, over 5651105.76 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3442, pruned_loss=0.09008, over 5715261.68 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3739, pruned_loss=0.1246, over 5648171.82 frames. ], batch size: 106, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:36:13,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2701, 1.2996, 3.4664, 3.0588], device='cuda:1'), covar=tensor([0.1535, 0.2637, 0.0468, 0.1459], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0653, 0.0972, 0.0921], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 07:36:24,355 INFO [optim.py:369] (1/2) 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 07:36:29,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=966877.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 07:36:33,325 INFO [train.py:968] (1/2) Epoch 22, batch 9150, giga_loss[loss=0.3294, simple_loss=0.3832, pruned_loss=0.1378, over 28850.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3719, pruned_loss=0.1227, over 5661629.42 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.344, pruned_loss=0.08997, over 5718312.14 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3746, pruned_loss=0.1257, over 5655134.53 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:36:51,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1080, 1.7237, 1.5676, 1.2906], device='cuda:1'), covar=tensor([0.0857, 0.0306, 0.0292, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0107], device='cuda:1')
+2023-03-11 07:37:26,191 INFO [train.py:968] (1/2) Epoch 22, batch 9200, giga_loss[loss=0.3072, simple_loss=0.3523, pruned_loss=0.131, over 23496.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5632775.51 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3443, pruned_loss=0.0902, over 5710379.46 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3747, pruned_loss=0.1264, over 5633028.50 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:37:42,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3502, 1.6824, 1.4311, 1.5113], device='cuda:1'), covar=tensor([0.0766, 0.0341, 0.0325, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 07:38:03,798 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 9250, giga_loss[loss=0.2504, simple_loss=0.3196, pruned_loss=0.09054, over 28554.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3691, pruned_loss=0.1216, over 5652578.84 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09037, over 5710435.55 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3709, pruned_loss=0.124, over 5651790.36 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:38:13,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3992, 4.2357, 4.0400, 2.0542], device='cuda:1'), covar=tensor([0.0595, 0.0710, 0.0725, 0.1865], device='cuda:1'), in_proj_covar=tensor([0.1247, 0.1153, 0.0976, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 07:38:52,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=967020.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 07:38:56,178 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 9300, giga_loss[loss=0.275, simple_loss=0.3396, pruned_loss=0.1052, over 28437.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3675, pruned_loss=0.121, over 5649400.86 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09035, over 5713456.03 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3693, pruned_loss=0.1233, over 5645081.43 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:39:22,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=967052.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 07:39:36,039 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 9350, giga_loss[loss=0.31, simple_loss=0.3769, pruned_loss=0.1215, over 28603.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1193, over 5640815.55 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09037, over 5703119.57 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.369, pruned_loss=0.122, over 5645306.75 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:40:11,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7017, 3.5411, 3.3584, 1.7288], device='cuda:1'), covar=tensor([0.0775, 0.0863, 0.0829, 0.2171], device='cuda:1'), in_proj_covar=tensor([0.1246, 0.1153, 0.0977, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 07:40:35,305 INFO [train.py:968] (1/2) Epoch 22, batch 9400, giga_loss[loss=0.3099, simple_loss=0.3719, pruned_loss=0.1239, over 28486.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3692, pruned_loss=0.1204, over 5649151.79 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3446, pruned_loss=0.09016, over 5707023.76 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3714, pruned_loss=0.1233, over 5648348.60 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:40:43,867 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:41:15,623 INFO [optim.py:369] (1/2) 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 07:41:22,425 INFO [train.py:968] (1/2) Epoch 22, batch 9450, giga_loss[loss=0.2509, simple_loss=0.3212, pruned_loss=0.09026, over 28763.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3686, pruned_loss=0.1199, over 5644395.08 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3447, pruned_loss=0.0902, over 5700952.36 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3707, pruned_loss=0.1227, over 5648821.67 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:41:48,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4441, 3.3409, 1.5946, 1.5144], device='cuda:1'), covar=tensor([0.1050, 0.0384, 0.0980, 0.1497], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0559, 0.0388, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 07:42:06,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7282, 2.0355, 1.9588, 1.5197], device='cuda:1'), covar=tensor([0.2037, 0.2703, 0.1691, 0.2006], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0704, 0.0947, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 07:42:06,975 INFO [train.py:968] (1/2) Epoch 22, batch 9500, libri_loss[loss=0.2394, simple_loss=0.3217, pruned_loss=0.07856, over 29587.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3698, pruned_loss=0.1192, over 5644457.93 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3445, pruned_loss=0.09011, over 5696168.83 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3722, pruned_loss=0.1222, over 5650418.29 frames. ], batch size: 74, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:42:39,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:42:47,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5897, 1.9280, 1.5121, 1.5804], device='cuda:1'), covar=tensor([0.2890, 0.2730, 0.3309, 0.2588], device='cuda:1'), in_proj_covar=tensor([0.1502, 0.1085, 0.1327, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 07:42:47,449 INFO [optim.py:369] (1/2) 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 07:42:53,488 INFO [train.py:968] (1/2) Epoch 22, batch 9550, giga_loss[loss=0.3613, simple_loss=0.408, pruned_loss=0.1573, over 27587.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3707, pruned_loss=0.1175, over 5656551.18 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3448, pruned_loss=0.09016, over 5697957.54 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3729, pruned_loss=0.1205, over 5658620.26 frames. ], batch size: 472, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:43:36,122 INFO [train.py:968] (1/2) Epoch 22, batch 9600, giga_loss[loss=0.2939, simple_loss=0.3788, pruned_loss=0.1045, over 28786.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3729, pruned_loss=0.1174, over 5660067.69 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3446, pruned_loss=0.09006, over 5688687.24 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3756, pruned_loss=0.1206, over 5669066.37 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:44:15,900 INFO [optim.py:369] (1/2) 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 07:44:23,796 INFO [train.py:968] (1/2) Epoch 22, batch 9650, giga_loss[loss=0.396, simple_loss=0.4327, pruned_loss=0.1797, over 27985.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3756, pruned_loss=0.12, over 5657497.75 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08991, over 5693968.91 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3786, pruned_loss=0.1234, over 5659434.69 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 07:44:53,396 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-11 07:45:08,838 INFO [train.py:968] (1/2) Epoch 22, batch 9700, giga_loss[loss=0.3213, simple_loss=0.388, pruned_loss=0.1273, over 28961.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3785, pruned_loss=0.1233, over 5668648.84 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3442, pruned_loss=0.08987, over 5696574.64 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3815, pruned_loss=0.1265, over 5667463.91 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:45:10,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4342, 3.3455, 1.4977, 1.5813], device='cuda:1'), covar=tensor([0.0978, 0.0409, 0.0885, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0559, 0.0388, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 07:45:48,954 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 9750, giga_loss[loss=0.2937, simple_loss=0.364, pruned_loss=0.1117, over 28628.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3773, pruned_loss=0.1235, over 5659164.26 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08946, over 5700554.57 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3815, pruned_loss=0.1275, over 5653593.57 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:46:22,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:46:35,897 INFO [train.py:968] (1/2) Epoch 22, batch 9800, giga_loss[loss=0.3843, simple_loss=0.4173, pruned_loss=0.1756, over 26669.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3755, pruned_loss=0.1217, over 5667604.71 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.0895, over 5708037.86 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3805, pruned_loss=0.1266, over 5654597.68 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:46:41,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-11 07:46:47,129 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:46:54,505 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 07:47:06,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3723, 1.7483, 1.4628, 1.5224], device='cuda:1'), covar=tensor([0.0827, 0.0310, 0.0340, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 07:47:14,726 INFO [optim.py:369] (1/2) 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 07:47:20,744 INFO [train.py:968] (1/2) Epoch 22, batch 9850, libri_loss[loss=0.2583, simple_loss=0.3417, pruned_loss=0.08744, over 19612.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3746, pruned_loss=0.12, over 5662592.15 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3434, pruned_loss=0.08955, over 5698147.18 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3793, pruned_loss=0.1245, over 5661495.33 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:47:49,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7348, 2.7478, 1.6648, 0.8255], device='cuda:1'), covar=tensor([0.9294, 0.4123, 0.4720, 0.8343], device='cuda:1'), in_proj_covar=tensor([0.1761, 0.1657, 0.1598, 0.1422], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 07:47:52,480 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:47:58,520 INFO [train.py:968] (1/2) Epoch 22, batch 9900, giga_loss[loss=0.2852, simple_loss=0.3673, pruned_loss=0.1015, over 28682.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3732, pruned_loss=0.1178, over 5675260.81 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3423, pruned_loss=0.08911, over 5706936.71 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3794, pruned_loss=0.123, over 5665255.00 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:48:02,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-11 07:48:06,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:48:18,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 07:48:24,095 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=967658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:48:26,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=967661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:48:37,738 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 9950, giga_loss[loss=0.3463, simple_loss=0.3831, pruned_loss=0.1548, over 23470.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3754, pruned_loss=0.1191, over 5666703.32 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3425, pruned_loss=0.08924, over 5701118.08 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3807, pruned_loss=0.1238, over 5663522.55 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:48:56,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=967690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:49:36,141 INFO [train.py:968] (1/2) Epoch 22, batch 10000, giga_loss[loss=0.3131, simple_loss=0.3782, pruned_loss=0.124, over 28793.00 frames. ], tot_loss[loss=0.31, simple_loss=0.377, pruned_loss=0.1214, over 5662066.73 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3424, pruned_loss=0.08917, over 5704120.84 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3818, pruned_loss=0.1256, over 5656115.76 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:50:09,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3547, 1.4155, 1.2743, 1.2740], device='cuda:1'), covar=tensor([0.1831, 0.1936, 0.1870, 0.1773], device='cuda:1'), in_proj_covar=tensor([0.1969, 0.1915, 0.1839, 0.1971], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 07:50:18,262 INFO [optim.py:369] (1/2) 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 07:50:20,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-11 07:50:25,975 INFO [train.py:968] (1/2) Epoch 22, batch 10050, giga_loss[loss=0.3297, simple_loss=0.389, pruned_loss=0.1352, over 28580.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3756, pruned_loss=0.1215, over 5658280.65 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3422, pruned_loss=0.08906, over 5704974.23 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3797, pruned_loss=0.125, over 5652752.42 frames. ], batch size: 307, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:50:26,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=967781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:50:28,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=967784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:51:01,620 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 10100, giga_loss[loss=0.2834, simple_loss=0.3518, pruned_loss=0.1075, over 28837.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3746, pruned_loss=0.1222, over 5657460.18 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.08896, over 5707025.16 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3782, pruned_loss=0.1253, over 5651022.38 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:52:01,599 INFO [optim.py:369] (1/2) 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 07:52:10,481 INFO [train.py:968] (1/2) Epoch 22, batch 10150, giga_loss[loss=0.2879, simple_loss=0.3578, pruned_loss=0.109, over 28808.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.372, pruned_loss=0.1212, over 5659135.50 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3423, pruned_loss=0.08909, over 5708964.20 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3749, pruned_loss=0.1239, over 5651830.51 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:52:49,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3300, 1.3819, 1.2280, 1.5275], device='cuda:1'), covar=tensor([0.0714, 0.0398, 0.0337, 0.0766], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0107], device='cuda:1')
+2023-03-11 07:52:49,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:52:58,658 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:53:00,427 INFO [train.py:968] (1/2) Epoch 22, batch 10200, giga_loss[loss=0.3064, simple_loss=0.3739, pruned_loss=0.1195, over 28656.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3708, pruned_loss=0.1212, over 5656095.77 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3424, pruned_loss=0.08916, over 5712036.63 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3734, pruned_loss=0.1237, over 5646660.52 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:53:14,749 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:53:48,634 INFO [train.py:968] (1/2) Epoch 22, batch 10250, giga_loss[loss=0.3274, simple_loss=0.3813, pruned_loss=0.1367, over 26558.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3701, pruned_loss=0.1209, over 5670245.32 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08917, over 5718411.64 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.1239, over 5655173.06 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:53:50,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6069, 1.6291, 1.7223, 1.4281], device='cuda:1'), covar=tensor([0.2966, 0.2734, 0.2266, 0.2731], device='cuda:1'), in_proj_covar=tensor([0.1974, 0.1918, 0.1846, 0.1980], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 07:53:59,952 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=967994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:54:05,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=967999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:54:35,666 INFO [train.py:968] (1/2) Epoch 22, batch 10300, giga_loss[loss=0.3076, simple_loss=0.374, pruned_loss=0.1206, over 27984.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3674, pruned_loss=0.1179, over 5669097.54 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08914, over 5720161.91 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3701, pruned_loss=0.1207, over 5654888.28 frames. ], batch size: 412, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:54:40,216 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:55:03,222 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:55:06,584 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:55:17,256 INFO [optim.py:369] (1/2) 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 07:55:22,932 INFO [train.py:968] (1/2) Epoch 22, batch 10350, giga_loss[loss=0.281, simple_loss=0.356, pruned_loss=0.103, over 28749.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3637, pruned_loss=0.1143, over 5659793.17 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08904, over 5721092.10 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.366, pruned_loss=0.1168, over 5647623.93 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 07:55:37,073 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:10,649 INFO [train.py:968] (1/2) Epoch 22, batch 10400, giga_loss[loss=0.2812, simple_loss=0.3547, pruned_loss=0.1038, over 28917.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.363, pruned_loss=0.1128, over 5659574.82 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08921, over 5716249.98 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.365, pruned_loss=0.1153, over 5653493.92 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:56:20,395 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:24,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:50,116 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:56:50,590 INFO [optim.py:369] (1/2) 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 07:56:57,354 INFO [train.py:968] (1/2) Epoch 22, batch 10450, giga_loss[loss=0.3177, simple_loss=0.3659, pruned_loss=0.1348, over 23508.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3613, pruned_loss=0.1125, over 5663435.80 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3429, pruned_loss=0.08939, over 5720020.15 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3633, pruned_loss=0.1149, over 5653973.68 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:57:48,815 INFO [train.py:968] (1/2) Epoch 22, batch 10500, giga_loss[loss=0.2917, simple_loss=0.3604, pruned_loss=0.1115, over 29035.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1126, over 5661514.01 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.0894, over 5720259.34 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3611, pruned_loss=0.1147, over 5653504.05 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:58:15,435 INFO [scaling.py:679] (1/2) 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] (1/2) 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 07:58:33,385 INFO [train.py:968] (1/2) Epoch 22, batch 10550, giga_loss[loss=0.2611, simple_loss=0.3481, pruned_loss=0.08707, over 28905.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.362, pruned_loss=0.1139, over 5662938.53 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.08987, over 5716546.05 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3631, pruned_loss=0.1157, over 5658393.40 frames. ], batch size: 145, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:58:43,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2577, 1.1590, 4.0179, 3.2262], device='cuda:1'), covar=tensor([0.1743, 0.2892, 0.0429, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0649, 0.0970, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 07:58:51,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:58:52,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6298, 1.4651, 1.7543, 1.2978], device='cuda:1'), covar=tensor([0.1841, 0.2912, 0.1408, 0.1596], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0710, 0.0954, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 07:59:06,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968320.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 07:59:16,523 INFO [train.py:968] (1/2) Epoch 22, batch 10600, giga_loss[loss=0.3597, simple_loss=0.4111, pruned_loss=0.1542, over 28578.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.365, pruned_loss=0.115, over 5665478.87 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3442, pruned_loss=0.09006, over 5718531.94 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3659, pruned_loss=0.1169, over 5659062.82 frames. ], batch size: 336, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 07:59:30,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:59:30,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 07:59:56,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:00:03,256 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 10650, giga_loss[loss=0.2807, simple_loss=0.3489, pruned_loss=0.1063, over 28683.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3657, pruned_loss=0.1162, over 5655501.41 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3441, pruned_loss=0.09004, over 5720271.71 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3667, pruned_loss=0.1179, over 5648394.73 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:00:40,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:00:40,667 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 10700, giga_loss[loss=0.25, simple_loss=0.3233, pruned_loss=0.08841, over 28850.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3654, pruned_loss=0.1164, over 5649586.83 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3444, pruned_loss=0.0902, over 5719354.93 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3662, pruned_loss=0.118, over 5643998.67 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:01:11,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:01:13,303 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:01:27,124 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968463.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 08:01:30,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968466.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 08:01:39,935 INFO [optim.py:369] (1/2) 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 08:01:42,751 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 10750, giga_loss[loss=0.2899, simple_loss=0.3563, pruned_loss=0.1117, over 28429.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.1181, over 5659743.82 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3446, pruned_loss=0.09033, over 5722868.75 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3682, pruned_loss=0.1197, over 5650971.08 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:01:58,503 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968495.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 08:02:08,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:15,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:17,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:30,069 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:31,129 INFO [train.py:968] (1/2) Epoch 22, batch 10800, libri_loss[loss=0.2763, simple_loss=0.3565, pruned_loss=0.09809, over 29521.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3699, pruned_loss=0.1196, over 5661891.93 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3448, pruned_loss=0.09049, over 5728240.42 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3713, pruned_loss=0.1218, over 5647445.02 frames. ], batch size: 82, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:02:34,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3252, 2.3467, 2.1399, 2.1522], device='cuda:1'), covar=tensor([0.1874, 0.2335, 0.2050, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0755, 0.0719, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 08:02:43,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:54,124 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:02:56,161 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:03:09,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7751, 1.9992, 2.0161, 1.6576], device='cuda:1'), covar=tensor([0.2633, 0.2399, 0.2276, 0.2450], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1912, 0.1838, 0.1976], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 08:03:13,367 INFO [optim.py:369] (1/2) 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 08:03:18,220 INFO [train.py:968] (1/2) Epoch 22, batch 10850, giga_loss[loss=0.3499, simple_loss=0.4057, pruned_loss=0.1471, over 28914.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3704, pruned_loss=0.1196, over 5653279.54 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3451, pruned_loss=0.09067, over 5711648.72 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3716, pruned_loss=0.1217, over 5656162.27 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:03:23,431 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:04:06,576 INFO [train.py:968] (1/2) Epoch 22, batch 10900, libri_loss[loss=0.2919, simple_loss=0.3665, pruned_loss=0.1087, over 20891.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3717, pruned_loss=0.1211, over 5652208.93 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3452, pruned_loss=0.09087, over 5707819.50 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3731, pruned_loss=0.1232, over 5657403.26 frames. ], batch size: 187, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:04:23,589 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:04:26,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:04:50,445 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 10950, giga_loss[loss=0.2877, simple_loss=0.3707, pruned_loss=0.1023, over 28674.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3731, pruned_loss=0.1218, over 5664990.98 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3456, pruned_loss=0.09116, over 5711358.83 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3745, pruned_loss=0.1238, over 5664990.82 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:05:00,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2887, 1.2887, 3.7359, 3.2543], device='cuda:1'), covar=tensor([0.1664, 0.2808, 0.0492, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0651, 0.0968, 0.0915], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 08:05:33,068 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 11000, giga_loss[loss=0.2936, simple_loss=0.3626, pruned_loss=0.1123, over 28810.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3734, pruned_loss=0.1209, over 5660790.68 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3456, pruned_loss=0.09118, over 5713806.66 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3749, pruned_loss=0.1229, over 5657927.11 frames. ], batch size: 119, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:06:30,871 INFO [optim.py:369] (1/2) 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 08:06:35,720 INFO [train.py:968] (1/2) Epoch 22, batch 11050, giga_loss[loss=0.2923, simple_loss=0.3612, pruned_loss=0.1117, over 28740.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3741, pruned_loss=0.1221, over 5653736.83 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3454, pruned_loss=0.0911, over 5714774.08 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.376, pruned_loss=0.1245, over 5649519.55 frames. ], batch size: 284, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:06:42,625 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:06:43,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9482, 1.2861, 5.2670, 3.7236], device='cuda:1'), covar=tensor([0.1564, 0.2885, 0.0370, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0652, 0.0969, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 08:06:45,243 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:24,254 INFO [train.py:968] (1/2) Epoch 22, batch 11100, libri_loss[loss=0.2894, simple_loss=0.3694, pruned_loss=0.1047, over 29132.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3728, pruned_loss=0.1219, over 5642668.02 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3454, pruned_loss=0.0912, over 5719773.98 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3753, pruned_loss=0.1247, over 5632343.37 frames. ], batch size: 101, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:07:55,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:56,458 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=968861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:58,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:07:59,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:15,381 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 11150, giga_loss[loss=0.3472, simple_loss=0.403, pruned_loss=0.1457, over 28953.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3717, pruned_loss=0.1212, over 5643440.22 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3457, pruned_loss=0.09142, over 5714282.55 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3744, pruned_loss=0.1244, over 5638608.15 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:08:27,859 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:28,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:29,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5101, 1.6609, 1.5959, 1.4102], device='cuda:1'), covar=tensor([0.2878, 0.2397, 0.1973, 0.2410], device='cuda:1'), in_proj_covar=tensor([0.1970, 0.1913, 0.1835, 0.1976], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 08:08:39,112 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=968904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:43,981 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=968909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:08:46,456 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-11 08:09:03,246 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 11200, giga_loss[loss=0.3189, simple_loss=0.3604, pruned_loss=0.1386, over 23399.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3694, pruned_loss=0.1205, over 5639911.55 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3454, pruned_loss=0.09128, over 5717379.26 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3722, pruned_loss=0.1235, over 5632575.49 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:09:05,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=968932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:09:31,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=968961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:09:42,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2205, 1.3279, 3.2543, 2.9246], device='cuda:1'), covar=tensor([0.1563, 0.2609, 0.0553, 0.1511], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0652, 0.0969, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 08:09:44,409 INFO [optim.py:369] (1/2) 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 08:09:49,241 INFO [train.py:968] (1/2) Epoch 22, batch 11250, giga_loss[loss=0.2699, simple_loss=0.344, pruned_loss=0.0979, over 28513.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3685, pruned_loss=0.12, over 5656540.61 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3455, pruned_loss=0.09134, over 5721232.54 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5645545.90 frames. ], batch size: 71, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:09:51,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4950, 2.1009, 1.5342, 0.6953], device='cuda:1'), covar=tensor([0.5457, 0.2850, 0.4042, 0.6056], device='cuda:1'), in_proj_covar=tensor([0.1760, 0.1653, 0.1594, 0.1427], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:09:55,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4387, 1.5488, 1.4926, 1.3410], device='cuda:1'), covar=tensor([0.2651, 0.2295, 0.2030, 0.2435], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1910, 0.1833, 0.1975], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 08:10:29,742 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-11 08:10:35,490 INFO [train.py:968] (1/2) Epoch 22, batch 11300, libri_loss[loss=0.287, simple_loss=0.3626, pruned_loss=0.1056, over 29510.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5660759.84 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.346, pruned_loss=0.09152, over 5727218.62 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1235, over 5644692.64 frames. ], batch size: 84, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:10:51,664 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:10:54,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:11:16,060 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 08:11:18,538 INFO [optim.py:369] (1/2) 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 08:11:20,224 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:11:22,699 INFO [train.py:968] (1/2) Epoch 22, batch 11350, giga_loss[loss=0.3541, simple_loss=0.3814, pruned_loss=0.1634, over 23648.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5661451.28 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3458, pruned_loss=0.09134, over 5730528.38 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3712, pruned_loss=0.1237, over 5644713.18 frames. ], batch size: 705, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:11:30,900 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 08:11:40,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 08:11:46,144 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-11 08:11:56,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3699, 2.0026, 1.6590, 0.6089], device='cuda:1'), covar=tensor([0.4918, 0.2923, 0.3640, 0.5998], device='cuda:1'), in_proj_covar=tensor([0.1760, 0.1655, 0.1596, 0.1427], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:12:02,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5487, 2.2116, 1.6841, 0.7413], device='cuda:1'), covar=tensor([0.5743, 0.3017, 0.3805, 0.6488], device='cuda:1'), in_proj_covar=tensor([0.1760, 0.1655, 0.1597, 0.1427], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:12:08,294 INFO [train.py:968] (1/2) Epoch 22, batch 11400, giga_loss[loss=0.3123, simple_loss=0.3783, pruned_loss=0.1231, over 28678.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3709, pruned_loss=0.1222, over 5666918.13 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.346, pruned_loss=0.09145, over 5735499.13 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3733, pruned_loss=0.1255, over 5646903.21 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:12:40,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:12:51,423 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 11450, giga_loss[loss=0.2343, simple_loss=0.311, pruned_loss=0.07879, over 28624.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3708, pruned_loss=0.1219, over 5662374.93 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3459, pruned_loss=0.09129, over 5740146.38 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3739, pruned_loss=0.1259, over 5639124.76 frames. ], batch size: 60, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:13:44,419 INFO [train.py:968] (1/2) Epoch 22, batch 11500, libri_loss[loss=0.2538, simple_loss=0.3358, pruned_loss=0.08588, over 29672.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3719, pruned_loss=0.1236, over 5659704.81 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.346, pruned_loss=0.09134, over 5741951.63 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3747, pruned_loss=0.1272, over 5638427.38 frames. ], batch size: 91, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:14:18,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5148, 1.8729, 1.4665, 1.6876], device='cuda:1'), covar=tensor([0.2419, 0.2404, 0.2745, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.1507, 0.1091, 0.1329, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 08:14:28,416 INFO [optim.py:369] (1/2) 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 08:14:32,097 INFO [train.py:968] (1/2) Epoch 22, batch 11550, giga_loss[loss=0.2904, simple_loss=0.3588, pruned_loss=0.111, over 28684.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3712, pruned_loss=0.1228, over 5665031.43 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.346, pruned_loss=0.09135, over 5743549.25 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3736, pruned_loss=0.1259, over 5646063.51 frames. ], batch size: 242, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:14:35,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:15:02,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:15:04,226 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 11600, giga_loss[loss=0.2991, simple_loss=0.3661, pruned_loss=0.1161, over 28746.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1228, over 5661818.10 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3459, pruned_loss=0.09124, over 5745731.08 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3744, pruned_loss=0.1264, over 5641699.93 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:15:20,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 08:15:30,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:15:39,083 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:16:08,578 INFO [train.py:968] (1/2) Epoch 22, batch 11650, giga_loss[loss=0.2405, simple_loss=0.3286, pruned_loss=0.07619, over 28886.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1214, over 5675615.28 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3458, pruned_loss=0.09129, over 5747023.55 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3732, pruned_loss=0.1245, over 5657908.44 frames. ], batch size: 174, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:16:54,255 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:16:54,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4033, 1.8273, 1.4317, 1.7282], device='cuda:1'), covar=tensor([0.0782, 0.0293, 0.0323, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0062, 0.0108], device='cuda:1')
+2023-03-11 08:16:54,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0196, 2.1859, 1.4539, 1.7912], device='cuda:1'), covar=tensor([0.0935, 0.0648, 0.1062, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0446, 0.0516, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 08:16:58,046 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:17:00,321 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 11700, giga_loss[loss=0.2845, simple_loss=0.3547, pruned_loss=0.1072, over 28685.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3729, pruned_loss=0.1237, over 5661708.66 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3458, pruned_loss=0.09122, over 5748334.45 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3752, pruned_loss=0.1265, over 5645907.72 frames. ], batch size: 99, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:17:16,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2495, 1.6198, 1.2177, 0.7513], device='cuda:1'), covar=tensor([0.3410, 0.2188, 0.2494, 0.4816], device='cuda:1'), in_proj_covar=tensor([0.1762, 0.1659, 0.1598, 0.1428], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:17:29,700 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:17:49,762 INFO [train.py:968] (1/2) Epoch 22, batch 11750, giga_loss[loss=0.3028, simple_loss=0.354, pruned_loss=0.1258, over 28442.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3749, pruned_loss=0.1256, over 5662107.63 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3454, pruned_loss=0.09091, over 5749800.42 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3776, pruned_loss=0.1288, over 5646830.86 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:18:02,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:18:23,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4497, 1.7413, 1.6929, 1.2447], device='cuda:1'), covar=tensor([0.1957, 0.2891, 0.1685, 0.2010], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0707, 0.0949, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:18:34,863 INFO [train.py:968] (1/2) Epoch 22, batch 11800, libri_loss[loss=0.2664, simple_loss=0.3561, pruned_loss=0.08828, over 29546.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1253, over 5664588.06 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3454, pruned_loss=0.09083, over 5754259.28 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3773, pruned_loss=0.1288, over 5645904.27 frames. ], batch size: 83, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:19:16,605 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 11850, libri_loss[loss=0.2779, simple_loss=0.3621, pruned_loss=0.09688, over 29308.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.1239, over 5650340.96 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3458, pruned_loss=0.09122, over 5740531.44 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3774, pruned_loss=0.1275, over 5644521.41 frames. ], batch size: 94, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:19:43,069 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 11900, giga_loss[loss=0.3845, simple_loss=0.4128, pruned_loss=0.1782, over 26628.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3749, pruned_loss=0.1238, over 5645290.24 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3458, pruned_loss=0.09112, over 5740486.46 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3775, pruned_loss=0.127, over 5639961.02 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:20:17,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4416, 4.0312, 1.5411, 1.6009], device='cuda:1'), covar=tensor([0.0993, 0.0303, 0.0925, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0559, 0.0388, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 08:20:43,455 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-11 08:20:45,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5287, 1.8257, 1.4873, 1.5029], device='cuda:1'), covar=tensor([0.2518, 0.2542, 0.2858, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1092, 0.1332, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 08:20:49,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=2.00 vs. limit=2.0
+2023-03-11 08:20:53,264 INFO [optim.py:369] (1/2) 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 08:20:55,147 INFO [train.py:968] (1/2) Epoch 22, batch 11950, giga_loss[loss=0.2802, simple_loss=0.3487, pruned_loss=0.1059, over 28551.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3724, pruned_loss=0.1216, over 5649420.33 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3455, pruned_loss=0.09089, over 5744874.27 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3753, pruned_loss=0.125, over 5639360.83 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 2.0
+2023-03-11 08:21:34,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:21:39,703 INFO [train.py:968] (1/2) Epoch 22, batch 12000, giga_loss[loss=0.2863, simple_loss=0.3581, pruned_loss=0.1073, over 28848.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3705, pruned_loss=0.1204, over 5659670.60 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3455, pruned_loss=0.09098, over 5746394.43 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3733, pruned_loss=0.1237, over 5648836.05 frames. ], batch size: 199, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:21:39,703 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 08:21:48,238 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 08:22:08,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3387, 3.1680, 3.0133, 1.3719], device='cuda:1'), covar=tensor([0.0950, 0.1169, 0.1041, 0.2348], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.1163, 0.0982, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 08:22:10,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:22:10,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5489, 3.1027, 1.6613, 1.6983], device='cuda:1'), covar=tensor([0.0784, 0.0352, 0.0726, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0560, 0.0389, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 08:22:32,876 INFO [optim.py:369] (1/2) 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 08:22:34,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-11 08:22:34,488 INFO [train.py:968] (1/2) Epoch 22, batch 12050, giga_loss[loss=0.27, simple_loss=0.3494, pruned_loss=0.09524, over 28589.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3714, pruned_loss=0.1211, over 5657831.16 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3453, pruned_loss=0.09086, over 5747803.37 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.374, pruned_loss=0.124, over 5647452.35 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:22:36,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5633, 1.5267, 1.2190, 1.1789], device='cuda:1'), covar=tensor([0.0735, 0.0417, 0.0837, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0445, 0.0515, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 08:22:52,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:22:56,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:23:18,304 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:23:21,419 INFO [train.py:968] (1/2) Epoch 22, batch 12100, giga_loss[loss=0.2956, simple_loss=0.3559, pruned_loss=0.1177, over 28686.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3718, pruned_loss=0.1216, over 5657442.63 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3456, pruned_loss=0.09095, over 5748892.22 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3743, pruned_loss=0.1246, over 5645484.09 frames. ], batch size: 92, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:24:02,112 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:02,199 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:02,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:04,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:10,359 INFO [optim.py:369] (1/2) 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 08:24:11,738 INFO [train.py:968] (1/2) Epoch 22, batch 12150, giga_loss[loss=0.279, simple_loss=0.3484, pruned_loss=0.1048, over 29052.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3695, pruned_loss=0.1204, over 5673389.02 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3455, pruned_loss=0.09092, over 5750720.15 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.372, pruned_loss=0.1233, over 5661238.36 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:24:26,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2228, 1.2601, 1.0759, 0.9584], device='cuda:1'), covar=tensor([0.0849, 0.0464, 0.0983, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0447, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 08:24:32,881 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:24:58,885 INFO [train.py:968] (1/2) Epoch 22, batch 12200, giga_loss[loss=0.2627, simple_loss=0.3398, pruned_loss=0.09285, over 28940.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3692, pruned_loss=0.1204, over 5671215.01 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09047, over 5754400.04 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1238, over 5656392.01 frames. ], batch size: 136, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:25:08,699 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=969942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:11,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=969945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:41,824 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=969974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:45,890 INFO [optim.py:369] (1/2) 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 08:25:46,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=969980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:25:47,141 INFO [train.py:968] (1/2) Epoch 22, batch 12250, giga_loss[loss=0.3437, simple_loss=0.3998, pruned_loss=0.1438, over 28889.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3725, pruned_loss=0.1226, over 5674413.55 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.345, pruned_loss=0.09047, over 5756174.03 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3751, pruned_loss=0.1257, over 5660165.46 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:25:47,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=969981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:26:19,339 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:26:21,225 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 12300, giga_loss[loss=0.3125, simple_loss=0.379, pruned_loss=0.1231, over 28923.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1238, over 5668937.83 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3447, pruned_loss=0.09033, over 5757564.18 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3759, pruned_loss=0.1266, over 5655975.00 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:26:47,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3118, 1.6082, 1.3105, 1.0441], device='cuda:1'), covar=tensor([0.2487, 0.2515, 0.2882, 0.2256], device='cuda:1'), in_proj_covar=tensor([0.1515, 0.1095, 0.1335, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 08:26:47,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:27:21,194 INFO [optim.py:369] (1/2) 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 08:27:22,526 INFO [train.py:968] (1/2) Epoch 22, batch 12350, giga_loss[loss=0.2738, simple_loss=0.3552, pruned_loss=0.09619, over 28970.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3718, pruned_loss=0.1214, over 5680361.85 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.345, pruned_loss=0.09036, over 5761331.15 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3743, pruned_loss=0.1246, over 5664383.39 frames. ], batch size: 164, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:28:01,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 08:28:02,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:06,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:09,242 INFO [train.py:968] (1/2) Epoch 22, batch 12400, giga_loss[loss=0.2802, simple_loss=0.3573, pruned_loss=0.1015, over 28924.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3718, pruned_loss=0.1208, over 5679377.16 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3452, pruned_loss=0.09051, over 5763208.56 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1237, over 5663543.86 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:28:10,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:30,612 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:28:52,808 INFO [optim.py:369] (1/2) 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 08:28:53,046 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 12450, giga_loss[loss=0.3426, simple_loss=0.3975, pruned_loss=0.1439, over 28863.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3723, pruned_loss=0.121, over 5674389.66 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3455, pruned_loss=0.09074, over 5754967.68 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3741, pruned_loss=0.1234, over 5669112.45 frames. ], batch size: 186, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:29:15,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:29:32,711 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 12500, giga_loss[loss=0.2898, simple_loss=0.3626, pruned_loss=0.1085, over 28885.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3708, pruned_loss=0.1202, over 5669770.04 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.346, pruned_loss=0.09095, over 5757148.48 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3722, pruned_loss=0.1223, over 5662557.64 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:29:57,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:29:59,953 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:24,245 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:27,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:30,140 INFO [optim.py:369] (1/2) 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 08:30:30,845 INFO [train.py:968] (1/2) Epoch 22, batch 12550, giga_loss[loss=0.2725, simple_loss=0.3361, pruned_loss=0.1045, over 28533.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3708, pruned_loss=0.121, over 5665792.88 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3465, pruned_loss=0.09127, over 5760111.53 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1235, over 5654508.53 frames. ], batch size: 78, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:30:40,846 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:30:52,561 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:07,877 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:10,359 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 12600, giga_loss[loss=0.3381, simple_loss=0.3967, pruned_loss=0.1397, over 28339.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3668, pruned_loss=0.1187, over 5675479.00 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3464, pruned_loss=0.09122, over 5762428.99 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3685, pruned_loss=0.1212, over 5662786.89 frames. ], batch size: 368, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:31:16,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2443, 1.7962, 1.3504, 0.4015], device='cuda:1'), covar=tensor([0.4113, 0.2728, 0.3951, 0.5861], device='cuda:1'), in_proj_covar=tensor([0.1759, 0.1660, 0.1596, 0.1426], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:31:20,226 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:27,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:30,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:34,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5049, 2.1007, 1.6375, 0.8159], device='cuda:1'), covar=tensor([0.5220, 0.2974, 0.3892, 0.6006], device='cuda:1'), in_proj_covar=tensor([0.1761, 0.1661, 0.1599, 0.1428], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:31:36,891 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:31:37,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:00,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:03,115 INFO [optim.py:369] (1/2) 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 08:32:03,883 INFO [train.py:968] (1/2) Epoch 22, batch 12650, giga_loss[loss=0.2633, simple_loss=0.3335, pruned_loss=0.09655, over 28855.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3638, pruned_loss=0.117, over 5687549.19 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3467, pruned_loss=0.09128, over 5764777.25 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3653, pruned_loss=0.1195, over 5673969.83 frames. ], batch size: 112, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:32:10,284 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:12,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:32:21,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5354, 1.7297, 1.3557, 1.6800], device='cuda:1'), covar=tensor([0.2707, 0.2809, 0.3175, 0.2388], device='cuda:1'), in_proj_covar=tensor([0.1517, 0.1095, 0.1336, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 08:32:22,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3191, 1.5103, 1.5210, 1.1885], device='cuda:1'), covar=tensor([0.1466, 0.2224, 0.1210, 0.1477], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0707, 0.0949, 0.0845], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:32:39,278 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 12700, giga_loss[loss=0.2882, simple_loss=0.3571, pruned_loss=0.1096, over 28696.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3618, pruned_loss=0.1159, over 5696512.75 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3464, pruned_loss=0.09116, over 5769404.61 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3639, pruned_loss=0.119, over 5678466.11 frames. ], batch size: 262, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:32:57,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4718, 1.8257, 1.3789, 1.6991], device='cuda:1'), covar=tensor([0.0748, 0.0289, 0.0312, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 08:33:34,645 INFO [optim.py:369] (1/2) 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 08:33:35,236 INFO [train.py:968] (1/2) Epoch 22, batch 12750, giga_loss[loss=0.3143, simple_loss=0.3734, pruned_loss=0.1276, over 28875.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3613, pruned_loss=0.1156, over 5681779.10 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3464, pruned_loss=0.0912, over 5758701.99 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3632, pruned_loss=0.1184, over 5675244.27 frames. ], batch size: 227, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:33:53,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:33:55,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:34:08,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-11 08:34:22,848 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:34:23,183 INFO [train.py:968] (1/2) Epoch 22, batch 12800, giga_loss[loss=0.296, simple_loss=0.3659, pruned_loss=0.113, over 28991.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3605, pruned_loss=0.1128, over 5684818.10 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3466, pruned_loss=0.0914, over 5762405.25 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3622, pruned_loss=0.1156, over 5674207.98 frames. ], batch size: 155, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:34:50,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5664, 1.6483, 1.8693, 1.3922], device='cuda:1'), covar=tensor([0.1976, 0.2761, 0.1618, 0.1977], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0704, 0.0949, 0.0845], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:35:10,890 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 12850, giga_loss[loss=0.2671, simple_loss=0.3442, pruned_loss=0.09496, over 28857.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3586, pruned_loss=0.11, over 5680778.15 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3458, pruned_loss=0.09109, over 5766734.71 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3611, pruned_loss=0.113, over 5666044.68 frames. ], batch size: 213, lr: 1.46e-03, grad_scale: 8.0
+2023-03-11 08:35:25,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:35:36,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6316, 1.8424, 1.4444, 1.8319], device='cuda:1'), covar=tensor([0.2775, 0.2732, 0.3113, 0.2559], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1094, 0.1338, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 08:35:51,149 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 12900, giga_loss[loss=0.2853, simple_loss=0.3461, pruned_loss=0.1123, over 26889.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3558, pruned_loss=0.1071, over 5668902.69 frames. ], libri_tot_loss[loss=0.2641, simple_loss=0.3457, pruned_loss=0.09119, over 5760240.18 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3583, pruned_loss=0.1102, over 5659714.38 frames. ], batch size: 555, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:36:12,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5713, 1.8282, 1.2397, 1.3570], device='cuda:1'), covar=tensor([0.1018, 0.0589, 0.1057, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0443, 0.0514, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 08:36:38,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:36:52,158 INFO [optim.py:369] (1/2) 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 08:36:52,171 INFO [train.py:968] (1/2) Epoch 22, batch 12950, giga_loss[loss=0.234, simple_loss=0.3206, pruned_loss=0.07375, over 28966.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3531, pruned_loss=0.1044, over 5663974.66 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3457, pruned_loss=0.09114, over 5761192.98 frames. ], giga_tot_loss[loss=0.2845, simple_loss=0.3552, pruned_loss=0.1069, over 5655580.54 frames. ], batch size: 128, lr: 1.46e-03, grad_scale: 4.0
+2023-03-11 08:36:54,067 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8626, 3.6852, 3.4938, 1.8002], device='cuda:1'), covar=tensor([0.0760, 0.0926, 0.0948, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1239, 0.1152, 0.0974, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 08:37:22,115 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=970711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:37:30,350 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=970722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:37:37,868 INFO [train.py:968] (1/2) Epoch 22, batch 13000, giga_loss[loss=0.2454, simple_loss=0.3336, pruned_loss=0.0786, over 28716.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3502, pruned_loss=0.101, over 5675502.03 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3451, pruned_loss=0.09096, over 5763376.35 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3527, pruned_loss=0.1036, over 5663502.32 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:37:45,065 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:37:49,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:38:14,648 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:38:15,733 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:38:17,548 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:38:28,249 INFO [train.py:968] (1/2) Epoch 22, batch 13050, giga_loss[loss=0.2541, simple_loss=0.3368, pruned_loss=0.08568, over 28063.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3501, pruned_loss=0.09916, over 5671204.87 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3448, pruned_loss=0.09084, over 5765637.91 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3524, pruned_loss=0.1015, over 5658394.79 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:38:37,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-11 08:38:43,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6379, 1.6963, 1.8792, 1.4683], device='cuda:1'), covar=tensor([0.1476, 0.2232, 0.1272, 0.1641], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0699, 0.0943, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:38:47,498 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:00,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:03,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:03,426 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-11 08:39:20,260 INFO [train.py:968] (1/2) Epoch 22, batch 13100, giga_loss[loss=0.2511, simple_loss=0.3307, pruned_loss=0.0857, over 28614.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3504, pruned_loss=0.09933, over 5668449.11 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3445, pruned_loss=0.09072, over 5767794.77 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3526, pruned_loss=0.1014, over 5654490.03 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:39:34,037 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:43,455 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=970854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:39:46,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=970857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:40:09,474 INFO [optim.py:369] (1/2) 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 08:40:09,486 INFO [train.py:968] (1/2) Epoch 22, batch 13150, giga_loss[loss=0.2618, simple_loss=0.3366, pruned_loss=0.09352, over 28934.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3484, pruned_loss=0.09854, over 5664929.02 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.344, pruned_loss=0.09055, over 5769979.56 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3507, pruned_loss=0.1005, over 5650649.48 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:40:15,989 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=970886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:40:32,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4427, 1.6738, 1.7072, 1.2637], device='cuda:1'), covar=tensor([0.1876, 0.2761, 0.1581, 0.1922], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0698, 0.0942, 0.0841], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:40:58,884 INFO [train.py:968] (1/2) Epoch 22, batch 13200, giga_loss[loss=0.2446, simple_loss=0.3229, pruned_loss=0.08313, over 28346.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3447, pruned_loss=0.09566, over 5673594.67 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3437, pruned_loss=0.09045, over 5771194.77 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.09741, over 5659444.32 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:41:19,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5998, 2.0116, 1.6302, 1.5883], device='cuda:1'), covar=tensor([0.2183, 0.2035, 0.2106, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0749, 0.0713, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 08:41:46,906 INFO [optim.py:369] (1/2) 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 08:41:46,918 INFO [train.py:968] (1/2) Epoch 22, batch 13250, giga_loss[loss=0.2852, simple_loss=0.3577, pruned_loss=0.1063, over 28735.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3445, pruned_loss=0.0959, over 5676386.69 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3428, pruned_loss=0.0901, over 5774489.50 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.09775, over 5659947.60 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:42:18,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-11 08:42:19,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4582, 4.5184, 1.5762, 1.7934], device='cuda:1'), covar=tensor([0.1042, 0.0261, 0.1004, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0557, 0.0387, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 08:42:34,839 INFO [train.py:968] (1/2) Epoch 22, batch 13300, giga_loss[loss=0.2688, simple_loss=0.3469, pruned_loss=0.09537, over 28885.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3448, pruned_loss=0.09577, over 5673613.02 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3428, pruned_loss=0.09019, over 5773677.81 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3468, pruned_loss=0.09725, over 5659510.22 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:43:27,850 INFO [train.py:968] (1/2) Epoch 22, batch 13350, giga_loss[loss=0.2682, simple_loss=0.3494, pruned_loss=0.09352, over 28935.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3425, pruned_loss=0.0937, over 5670264.55 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3425, pruned_loss=0.09004, over 5772046.86 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3444, pruned_loss=0.09507, over 5659259.12 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:43:28,439 INFO [optim.py:369] (1/2) 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 08:43:33,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=971087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:43:42,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:44:02,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3792, 1.6457, 1.6148, 1.2379], device='cuda:1'), covar=tensor([0.1830, 0.2670, 0.1548, 0.1916], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0699, 0.0945, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:44:15,629 INFO [train.py:968] (1/2) Epoch 22, batch 13400, giga_loss[loss=0.2385, simple_loss=0.3193, pruned_loss=0.07881, over 28736.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3399, pruned_loss=0.09211, over 5675841.46 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3427, pruned_loss=0.09031, over 5775442.87 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3412, pruned_loss=0.09302, over 5661821.27 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:45:06,998 INFO [train.py:968] (1/2) Epoch 22, batch 13450, giga_loss[loss=0.2511, simple_loss=0.3097, pruned_loss=0.09625, over 24013.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3346, pruned_loss=0.08936, over 5663285.52 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3415, pruned_loss=0.08985, over 5777475.50 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3366, pruned_loss=0.09051, over 5647827.33 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:45:07,667 INFO [optim.py:369] (1/2) 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 08:45:24,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5846, 1.6508, 1.8366, 1.4040], device='cuda:1'), covar=tensor([0.1961, 0.2665, 0.1604, 0.1935], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0699, 0.0945, 0.0843], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:46:00,135 INFO [train.py:968] (1/2) Epoch 22, batch 13500, giga_loss[loss=0.2353, simple_loss=0.3182, pruned_loss=0.0762, over 27953.00 frames. ], tot_loss[loss=0.257, simple_loss=0.334, pruned_loss=0.08994, over 5654383.09 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3412, pruned_loss=0.08978, over 5776311.68 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3357, pruned_loss=0.09092, over 5641315.22 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:46:10,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=971240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:46:13,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=971243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:46:33,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5537, 2.0542, 1.3783, 0.9615], device='cuda:1'), covar=tensor([0.7455, 0.4146, 0.3552, 0.6355], device='cuda:1'), in_proj_covar=tensor([0.1744, 0.1638, 0.1584, 0.1414], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:46:46,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=971272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:46:58,384 INFO [train.py:968] (1/2) Epoch 22, batch 13550, giga_loss[loss=0.3141, simple_loss=0.3707, pruned_loss=0.1287, over 26645.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3345, pruned_loss=0.09073, over 5638383.27 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.341, pruned_loss=0.08966, over 5776806.12 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.336, pruned_loss=0.09161, over 5627232.21 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:46:59,344 INFO [optim.py:369] (1/2) 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 08:46:59,854 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-11 08:47:00,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4635, 1.6684, 1.6566, 1.4344], device='cuda:1'), covar=tensor([0.2404, 0.1971, 0.1896, 0.2093], device='cuda:1'), in_proj_covar=tensor([0.1929, 0.1861, 0.1783, 0.1920], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 08:47:49,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4551, 1.5434, 1.6332, 1.2739], device='cuda:1'), covar=tensor([0.1819, 0.2599, 0.1526, 0.1883], device='cuda:1'), in_proj_covar=tensor([0.0898, 0.0699, 0.0945, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 08:47:51,159 INFO [train.py:968] (1/2) Epoch 22, batch 13600, giga_loss[loss=0.267, simple_loss=0.351, pruned_loss=0.09156, over 28871.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3363, pruned_loss=0.09035, over 5648396.83 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3407, pruned_loss=0.08952, over 5778583.35 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3376, pruned_loss=0.09121, over 5634171.43 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:48:38,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5561, 1.8507, 1.4926, 1.5745], device='cuda:1'), covar=tensor([0.2651, 0.2573, 0.3061, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.1511, 0.1087, 0.1334, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 08:48:47,716 INFO [train.py:968] (1/2) Epoch 22, batch 13650, giga_loss[loss=0.2318, simple_loss=0.3187, pruned_loss=0.07246, over 28708.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3371, pruned_loss=0.09012, over 5652299.72 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3402, pruned_loss=0.08933, over 5781805.79 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3385, pruned_loss=0.09098, over 5635199.44 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:48:49,872 INFO [optim.py:369] (1/2) 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 08:48:58,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1557, 1.3047, 1.0994, 0.9952], device='cuda:1'), covar=tensor([0.0959, 0.0458, 0.1022, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0388, 0.0439, 0.0510, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 08:49:27,570 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 13700, giga_loss[loss=0.2493, simple_loss=0.3277, pruned_loss=0.08544, over 28125.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3375, pruned_loss=0.09047, over 5649255.82 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3399, pruned_loss=0.08924, over 5784967.76 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3389, pruned_loss=0.0913, over 5628975.03 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:50:21,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:50:41,203 INFO [train.py:968] (1/2) Epoch 22, batch 13750, libri_loss[loss=0.222, simple_loss=0.2955, pruned_loss=0.07423, over 28074.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3346, pruned_loss=0.08855, over 5654103.02 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3391, pruned_loss=0.08894, over 5776261.77 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3363, pruned_loss=0.0895, over 5641842.36 frames. ], batch size: 62, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:50:42,336 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 08:50:42,583 INFO [optim.py:369] (1/2) 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 08:51:15,220 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-11 08:51:39,025 INFO [train.py:968] (1/2) Epoch 22, batch 13800, giga_loss[loss=0.2507, simple_loss=0.3497, pruned_loss=0.07587, over 28932.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3344, pruned_loss=0.08763, over 5649245.31 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3384, pruned_loss=0.08865, over 5778026.95 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3363, pruned_loss=0.08864, over 5634502.92 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:51:43,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3238, 1.2162, 3.6741, 3.2458], device='cuda:1'), covar=tensor([0.1961, 0.3494, 0.0857, 0.2469], device='cuda:1'), in_proj_covar=tensor([0.0759, 0.0647, 0.0958, 0.0902], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 08:51:49,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8185, 2.0163, 1.4209, 1.6314], device='cuda:1'), covar=tensor([0.1043, 0.0640, 0.1017, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0440, 0.0512, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 08:52:32,657 INFO [train.py:968] (1/2) Epoch 22, batch 13850, giga_loss[loss=0.2594, simple_loss=0.3333, pruned_loss=0.09276, over 29051.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3326, pruned_loss=0.08659, over 5652387.11 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3379, pruned_loss=0.08873, over 5771966.49 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3344, pruned_loss=0.08726, over 5639760.26 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:52:36,700 INFO [optim.py:369] (1/2) 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 08:53:04,886 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=971605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:53:07,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=971608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:53:10,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=971610.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 08:53:23,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-11 08:53:33,671 INFO [train.py:968] (1/2) Epoch 22, batch 13900, giga_loss[loss=0.2549, simple_loss=0.3326, pruned_loss=0.08864, over 28455.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3316, pruned_loss=0.08757, over 5650838.75 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.338, pruned_loss=0.08895, over 5770098.67 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3328, pruned_loss=0.0879, over 5640547.87 frames. ], batch size: 369, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:53:42,312 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=971637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:54:23,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4658, 1.7296, 1.4864, 1.7044], device='cuda:1'), covar=tensor([0.0664, 0.0287, 0.0303, 0.0700], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0117, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 08:54:33,429 INFO [train.py:968] (1/2) Epoch 22, batch 13950, giga_loss[loss=0.2511, simple_loss=0.3352, pruned_loss=0.08353, over 28612.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3313, pruned_loss=0.0878, over 5655771.46 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3382, pruned_loss=0.08925, over 5771158.31 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3321, pruned_loss=0.08778, over 5645572.99 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:54:37,004 INFO [optim.py:369] (1/2) 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 08:55:30,030 INFO [train.py:968] (1/2) Epoch 22, batch 14000, libri_loss[loss=0.2325, simple_loss=0.3125, pruned_loss=0.07624, over 29528.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3327, pruned_loss=0.08791, over 5666635.35 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3382, pruned_loss=0.08928, over 5770767.20 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3332, pruned_loss=0.08784, over 5656923.29 frames. ], batch size: 80, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 08:56:28,035 INFO [train.py:968] (1/2) Epoch 22, batch 14050, giga_loss[loss=0.2721, simple_loss=0.3558, pruned_loss=0.09419, over 28695.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.335, pruned_loss=0.08825, over 5678669.77 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3378, pruned_loss=0.0891, over 5774096.13 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3356, pruned_loss=0.0883, over 5664829.80 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:56:31,200 INFO [optim.py:369] (1/2) 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 08:56:32,152 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=971784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:56:40,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:56:51,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5451, 2.2637, 1.6952, 0.6888], device='cuda:1'), covar=tensor([0.5466, 0.3084, 0.4332, 0.6493], device='cuda:1'), in_proj_covar=tensor([0.1752, 0.1644, 0.1591, 0.1420], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 08:57:09,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3487, 1.3776, 3.8878, 3.0987], device='cuda:1'), covar=tensor([0.1585, 0.2602, 0.0416, 0.0929], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0651, 0.0960, 0.0906], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 08:57:30,446 INFO [train.py:968] (1/2) Epoch 22, batch 14100, giga_loss[loss=0.2328, simple_loss=0.3161, pruned_loss=0.07471, over 28931.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3326, pruned_loss=0.08659, over 5682171.89 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3372, pruned_loss=0.08889, over 5778099.64 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3335, pruned_loss=0.08677, over 5663465.44 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:58:34,762 INFO [train.py:968] (1/2) Epoch 22, batch 14150, libri_loss[loss=0.2973, simple_loss=0.3581, pruned_loss=0.1183, over 29563.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3318, pruned_loss=0.08628, over 5688545.78 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3372, pruned_loss=0.08901, over 5778119.13 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3324, pruned_loss=0.08624, over 5671803.57 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:58:38,168 INFO [optim.py:369] (1/2) 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 08:59:38,710 INFO [train.py:968] (1/2) Epoch 22, batch 14200, giga_loss[loss=0.2475, simple_loss=0.3327, pruned_loss=0.08117, over 28911.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3343, pruned_loss=0.08781, over 5670088.96 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.337, pruned_loss=0.08886, over 5779361.75 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3349, pruned_loss=0.0879, over 5653742.74 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 08:59:43,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=971933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 08:59:47,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=971936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:00:24,205 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=971965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:00:44,079 INFO [train.py:968] (1/2) Epoch 22, batch 14250, giga_loss[loss=0.2275, simple_loss=0.3219, pruned_loss=0.06657, over 27623.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3383, pruned_loss=0.08752, over 5663745.04 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3368, pruned_loss=0.08878, over 5780217.05 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.339, pruned_loss=0.08765, over 5649026.76 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:00:47,356 INFO [optim.py:369] (1/2) 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 09:00:49,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=971985.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:01:45,137 INFO [train.py:968] (1/2) Epoch 22, batch 14300, giga_loss[loss=0.2614, simple_loss=0.348, pruned_loss=0.08735, over 28901.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3379, pruned_loss=0.0861, over 5650141.53 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3369, pruned_loss=0.08885, over 5781835.97 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3384, pruned_loss=0.08612, over 5634998.52 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:02:42,230 INFO [train.py:968] (1/2) Epoch 22, batch 14350, giga_loss[loss=0.241, simple_loss=0.3312, pruned_loss=0.07544, over 28716.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3387, pruned_loss=0.08556, over 5645689.12 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.337, pruned_loss=0.08902, over 5761488.21 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.339, pruned_loss=0.08535, over 5648171.71 frames. ], batch size: 243, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:02:45,394 INFO [optim.py:369] (1/2) 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 09:03:38,903 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 09:03:43,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=972128.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:03:45,898 INFO [train.py:968] (1/2) Epoch 22, batch 14400, giga_loss[loss=0.2502, simple_loss=0.3369, pruned_loss=0.08177, over 28860.00 frames. ], tot_loss[loss=0.258, simple_loss=0.341, pruned_loss=0.08755, over 5655898.65 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3367, pruned_loss=0.08894, over 5763535.01 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3415, pruned_loss=0.08741, over 5654793.14 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:03:46,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=972131.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:04:08,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8291, 1.1677, 2.9295, 2.8008], device='cuda:1'), covar=tensor([0.1641, 0.2428, 0.0608, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0646, 0.0955, 0.0898], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 09:04:18,779 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=972160.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:04:21,591 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.82 vs. limit=5.0
+2023-03-11 09:04:47,112 INFO [train.py:968] (1/2) Epoch 22, batch 14450, giga_loss[loss=0.2587, simple_loss=0.3385, pruned_loss=0.08942, over 28357.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3392, pruned_loss=0.0879, over 5659268.33 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3365, pruned_loss=0.08891, over 5764667.58 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3398, pruned_loss=0.08779, over 5656088.21 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:04:50,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9253, 2.0622, 1.7831, 2.4636], device='cuda:1'), covar=tensor([0.2513, 0.2749, 0.3000, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.1510, 0.1088, 0.1332, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 09:04:50,360 INFO [optim.py:369] (1/2) 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 09:06:01,683 INFO [train.py:968] (1/2) Epoch 22, batch 14500, giga_loss[loss=0.2793, simple_loss=0.3539, pruned_loss=0.1023, over 28890.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3405, pruned_loss=0.08955, over 5664012.71 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3367, pruned_loss=0.08911, over 5767756.59 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3409, pruned_loss=0.08928, over 5657178.02 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:06:21,878 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=972245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:06:34,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4794, 1.9073, 1.7578, 1.5947], device='cuda:1'), covar=tensor([0.2356, 0.2590, 0.2253, 0.2520], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0742, 0.0707, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 09:07:22,030 INFO [train.py:968] (1/2) Epoch 22, batch 14550, giga_loss[loss=0.2491, simple_loss=0.3327, pruned_loss=0.08274, over 28560.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3359, pruned_loss=0.08676, over 5671659.40 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3365, pruned_loss=0.08904, over 5770640.01 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3364, pruned_loss=0.08658, over 5661897.69 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:07:25,903 INFO [optim.py:369] (1/2) 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 09:07:49,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=972302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:07:54,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=972305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:08:25,707 INFO [train.py:968] (1/2) Epoch 22, batch 14600, giga_loss[loss=0.2337, simple_loss=0.3223, pruned_loss=0.07256, over 28468.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3334, pruned_loss=0.08561, over 5666715.96 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3357, pruned_loss=0.08864, over 5771360.77 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3345, pruned_loss=0.08576, over 5655592.69 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:08:29,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=972334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:08:47,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5850, 1.7590, 1.2646, 1.3666], device='cuda:1'), covar=tensor([0.0957, 0.0581, 0.0961, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0515, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 09:09:31,758 INFO [train.py:968] (1/2) Epoch 22, batch 14650, giga_loss[loss=0.2942, simple_loss=0.3496, pruned_loss=0.1194, over 26777.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3303, pruned_loss=0.08414, over 5675234.05 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3359, pruned_loss=0.08878, over 5772841.44 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.331, pruned_loss=0.08408, over 5663843.69 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:09:32,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1682, 4.0017, 3.8398, 1.8593], device='cuda:1'), covar=tensor([0.0683, 0.0771, 0.0849, 0.2112], device='cuda:1'), in_proj_covar=tensor([0.1215, 0.1122, 0.0952, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 09:09:36,903 INFO [optim.py:369] (1/2) 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 09:10:19,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5583, 4.3990, 4.1939, 1.8597], device='cuda:1'), covar=tensor([0.0559, 0.0692, 0.0828, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.1209, 0.1119, 0.0950, 0.0710], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 09:10:31,383 INFO [train.py:968] (1/2) Epoch 22, batch 14700, giga_loss[loss=0.2473, simple_loss=0.3359, pruned_loss=0.07933, over 28888.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3361, pruned_loss=0.0875, over 5689640.73 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.08867, over 5776269.79 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3369, pruned_loss=0.08748, over 5674991.59 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:11:27,129 INFO [train.py:968] (1/2) Epoch 22, batch 14750, giga_loss[loss=0.2307, simple_loss=0.3068, pruned_loss=0.0773, over 28943.00 frames. ], tot_loss[loss=0.256, simple_loss=0.336, pruned_loss=0.08796, over 5679781.48 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3349, pruned_loss=0.08831, over 5772471.62 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3372, pruned_loss=0.08825, over 5668572.68 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:11:31,749 INFO [optim.py:369] (1/2) 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 09:11:35,211 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=972487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:12:22,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8984, 3.7484, 3.5489, 1.8033], device='cuda:1'), covar=tensor([0.0683, 0.0785, 0.0830, 0.2299], device='cuda:1'), in_proj_covar=tensor([0.1206, 0.1116, 0.0947, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 09:12:28,317 INFO [train.py:968] (1/2) Epoch 22, batch 14800, giga_loss[loss=0.2571, simple_loss=0.3328, pruned_loss=0.09069, over 28910.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3353, pruned_loss=0.08846, over 5686690.04 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3348, pruned_loss=0.08819, over 5775656.47 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3364, pruned_loss=0.08879, over 5673118.02 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:13:00,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2366, 1.3280, 3.7789, 3.2460], device='cuda:1'), covar=tensor([0.1710, 0.2769, 0.0492, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0649, 0.0957, 0.0900], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 09:13:20,400 INFO [train.py:968] (1/2) Epoch 22, batch 14850, giga_loss[loss=0.262, simple_loss=0.3395, pruned_loss=0.09228, over 28832.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3369, pruned_loss=0.09027, over 5676217.80 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08818, over 5768129.43 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3381, pruned_loss=0.09061, over 5667357.02 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:13:26,113 INFO [optim.py:369] (1/2) 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 09:14:10,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=972620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:14:25,433 INFO [train.py:968] (1/2) Epoch 22, batch 14900, giga_loss[loss=0.2562, simple_loss=0.3411, pruned_loss=0.08569, over 29019.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3368, pruned_loss=0.0898, over 5671079.38 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3345, pruned_loss=0.08812, over 5768537.24 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3378, pruned_loss=0.09014, over 5662948.82 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:14:32,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3589, 1.6424, 1.5765, 1.2499], device='cuda:1'), covar=tensor([0.2605, 0.2152, 0.1658, 0.2225], device='cuda:1'), in_proj_covar=tensor([0.1918, 0.1839, 0.1750, 0.1903], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 09:15:39,393 INFO [train.py:968] (1/2) Epoch 22, batch 14950, giga_loss[loss=0.2832, simple_loss=0.3451, pruned_loss=0.1107, over 26785.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3395, pruned_loss=0.09018, over 5671835.13 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3345, pruned_loss=0.08808, over 5769244.56 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3403, pruned_loss=0.09048, over 5664446.84 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:15:47,767 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 15000, giga_loss[loss=0.212, simple_loss=0.298, pruned_loss=0.06294, over 29056.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3382, pruned_loss=0.08915, over 5665214.92 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3348, pruned_loss=0.08834, over 5760056.44 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3386, pruned_loss=0.08919, over 5666171.14 frames. ], batch size: 165, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:17:00,993 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 09:17:05,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4707, 1.7982, 1.4268, 1.3627], device='cuda:1'), covar=tensor([0.3046, 0.2890, 0.3263, 0.2625], device='cuda:1'), in_proj_covar=tensor([0.1515, 0.1090, 0.1336, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 09:17:09,798 INFO [train.py:1012] (1/2) Epoch 22, validation: loss=0.1979, simple_loss=0.2984, pruned_loss=0.04871, over 944034.00 frames. 
+2023-03-11 09:17:09,799 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 09:17:25,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3113, 1.3838, 3.9132, 3.1590], device='cuda:1'), covar=tensor([0.1590, 0.2668, 0.0463, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0647, 0.0953, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 09:17:53,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=972763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:17:56,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=972766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:18:10,836 INFO [train.py:968] (1/2) Epoch 22, batch 15050, giga_loss[loss=0.2539, simple_loss=0.3296, pruned_loss=0.08907, over 28499.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.334, pruned_loss=0.08777, over 5682696.53 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3346, pruned_loss=0.08826, over 5761576.81 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08787, over 5679422.41 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:18:16,616 INFO [optim.py:369] (1/2) 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 09:18:28,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=972795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:19:10,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5376, 1.6373, 1.6567, 1.5032], device='cuda:1'), covar=tensor([0.2725, 0.2315, 0.1936, 0.2140], device='cuda:1'), in_proj_covar=tensor([0.1921, 0.1841, 0.1752, 0.1904], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 09:19:12,361 INFO [train.py:968] (1/2) Epoch 22, batch 15100, giga_loss[loss=0.2138, simple_loss=0.2913, pruned_loss=0.06817, over 28570.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.328, pruned_loss=0.08516, over 5685614.87 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3342, pruned_loss=0.08818, over 5767664.25 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3286, pruned_loss=0.08525, over 5674515.29 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:19:50,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=972862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:20:11,110 INFO [train.py:968] (1/2) Epoch 22, batch 15150, giga_loss[loss=0.2588, simple_loss=0.3383, pruned_loss=0.08969, over 28551.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3285, pruned_loss=0.08585, over 5682256.91 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3343, pruned_loss=0.08834, over 5767292.31 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3288, pruned_loss=0.08572, over 5671435.82 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:20:16,318 INFO [optim.py:369] (1/2) 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 09:20:30,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-11 09:21:06,118 INFO [train.py:968] (1/2) Epoch 22, batch 15200, libri_loss[loss=0.2344, simple_loss=0.3045, pruned_loss=0.08215, over 29375.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3302, pruned_loss=0.08754, over 5684011.20 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3342, pruned_loss=0.08836, over 5770095.50 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3305, pruned_loss=0.08738, over 5670844.74 frames. ], batch size: 67, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:21:58,240 INFO [train.py:968] (1/2) Epoch 22, batch 15250, giga_loss[loss=0.2026, simple_loss=0.2961, pruned_loss=0.05459, over 28910.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3271, pruned_loss=0.08563, over 5670428.74 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3335, pruned_loss=0.08813, over 5770855.08 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3278, pruned_loss=0.08567, over 5655666.42 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:22:05,485 INFO [optim.py:369] (1/2) 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 09:22:13,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3156, 2.8853, 1.4395, 1.4657], device='cuda:1'), covar=tensor([0.0980, 0.0351, 0.0945, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0552, 0.0387, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 09:22:23,611 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:22:28,177 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:22:31,502 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 15300, giga_loss[loss=0.2093, simple_loss=0.2949, pruned_loss=0.06183, over 28407.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3262, pruned_loss=0.08415, over 5669157.84 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3331, pruned_loss=0.08796, over 5766338.07 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3269, pruned_loss=0.08425, over 5658593.45 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:23:02,606 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973037.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 09:23:02,615 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:23:20,409 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 15350, giga_loss[loss=0.2432, simple_loss=0.3217, pruned_loss=0.08234, over 28930.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3257, pruned_loss=0.0842, over 5667688.35 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3332, pruned_loss=0.08797, over 5766892.43 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.326, pruned_loss=0.08415, over 5655779.01 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:24:06,574 INFO [optim.py:369] (1/2) 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 09:24:43,591 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-11 09:25:04,017 INFO [train.py:968] (1/2) Epoch 22, batch 15400, giga_loss[loss=0.2108, simple_loss=0.2941, pruned_loss=0.06379, over 28807.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3248, pruned_loss=0.083, over 5681175.14 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3331, pruned_loss=0.08785, over 5769505.00 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.325, pruned_loss=0.083, over 5667794.99 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:26:03,345 INFO [train.py:968] (1/2) Epoch 22, batch 15450, giga_loss[loss=0.2531, simple_loss=0.3354, pruned_loss=0.08537, over 28868.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3254, pruned_loss=0.08307, over 5689480.53 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3326, pruned_loss=0.08764, over 5769156.47 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3257, pruned_loss=0.08312, over 5676390.11 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:26:08,862 INFO [optim.py:369] (1/2) 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 09:27:06,064 INFO [train.py:968] (1/2) Epoch 22, batch 15500, giga_loss[loss=0.2548, simple_loss=0.3333, pruned_loss=0.08813, over 28870.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3268, pruned_loss=0.08458, over 5691667.53 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3324, pruned_loss=0.08759, over 5771298.69 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3272, pruned_loss=0.08458, over 5677718.61 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:27:15,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6447, 1.8409, 1.8871, 1.5062], device='cuda:1'), covar=tensor([0.2926, 0.2371, 0.1881, 0.2528], device='cuda:1'), in_proj_covar=tensor([0.1932, 0.1859, 0.1768, 0.1921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 09:28:10,377 INFO [train.py:968] (1/2) Epoch 22, batch 15550, giga_loss[loss=0.2152, simple_loss=0.2782, pruned_loss=0.07608, over 24381.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3248, pruned_loss=0.08332, over 5687459.10 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3322, pruned_loss=0.08748, over 5772992.84 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3252, pruned_loss=0.08337, over 5673670.69 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:28:18,349 INFO [optim.py:369] (1/2) 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 09:28:55,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5743, 1.8309, 1.5070, 1.6615], device='cuda:1'), covar=tensor([0.2709, 0.2555, 0.2796, 0.2424], device='cuda:1'), in_proj_covar=tensor([0.1510, 0.1088, 0.1331, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 09:29:10,665 INFO [train.py:968] (1/2) Epoch 22, batch 15600, giga_loss[loss=0.2628, simple_loss=0.3584, pruned_loss=0.08361, over 28912.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3265, pruned_loss=0.08306, over 5675291.22 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3323, pruned_loss=0.08761, over 5774563.59 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3266, pruned_loss=0.08294, over 5662208.88 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:30:08,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=973377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:30:14,568 INFO [train.py:968] (1/2) Epoch 22, batch 15650, giga_loss[loss=0.2345, simple_loss=0.3266, pruned_loss=0.07119, over 28947.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3296, pruned_loss=0.08422, over 5667858.83 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.332, pruned_loss=0.08752, over 5775859.50 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3299, pruned_loss=0.08418, over 5655327.27 frames. ], batch size: 285, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:30:21,161 INFO [optim.py:369] (1/2) 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 09:30:46,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=973412.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:31:08,523 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 15700, giga_loss[loss=0.3076, simple_loss=0.3632, pruned_loss=0.126, over 26914.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3322, pruned_loss=0.08562, over 5646029.03 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3323, pruned_loss=0.0878, over 5748707.79 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3322, pruned_loss=0.08526, over 5655925.05 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:32:06,220 INFO [train.py:968] (1/2) Epoch 22, batch 15750, giga_loss[loss=0.2557, simple_loss=0.3385, pruned_loss=0.08642, over 28989.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3325, pruned_loss=0.08641, over 5644059.18 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3322, pruned_loss=0.08786, over 5751896.08 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3327, pruned_loss=0.08605, over 5647209.88 frames. ], batch size: 285, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:32:17,320 INFO [optim.py:369] (1/2) 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 09:32:53,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:32:56,779 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 15800, giga_loss[loss=0.2091, simple_loss=0.2927, pruned_loss=0.06274, over 28768.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3294, pruned_loss=0.08438, over 5646426.00 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3321, pruned_loss=0.0878, over 5753919.50 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3296, pruned_loss=0.08412, over 5645726.26 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:33:32,103 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:33:35,968 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973555.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:33:41,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=973558.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:33:57,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=973573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:00,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=973576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:04,717 INFO [train.py:968] (1/2) Epoch 22, batch 15850, giga_loss[loss=0.2224, simple_loss=0.2977, pruned_loss=0.07352, over 24389.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3282, pruned_loss=0.08394, over 5646449.33 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3319, pruned_loss=0.08797, over 5748702.69 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3285, pruned_loss=0.08339, over 5646529.60 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:34:12,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973587.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 09:34:12,881 INFO [optim.py:369] (1/2) 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 09:34:19,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:35,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=973605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:34:43,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0946, 3.9278, 3.7546, 1.7386], device='cuda:1'), covar=tensor([0.0626, 0.0807, 0.0834, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.1209, 0.1117, 0.0945, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 09:34:52,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2985, 3.5347, 1.6125, 1.4761], device='cuda:1'), covar=tensor([0.1019, 0.0357, 0.0917, 0.1391], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0551, 0.0386, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 09:35:01,531 INFO [train.py:968] (1/2) Epoch 22, batch 15900, giga_loss[loss=0.2344, simple_loss=0.3173, pruned_loss=0.07579, over 28892.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3269, pruned_loss=0.08367, over 5663514.50 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3321, pruned_loss=0.08813, over 5752137.42 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3267, pruned_loss=0.083, over 5658633.90 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:36:02,325 INFO [train.py:968] (1/2) Epoch 22, batch 15950, giga_loss[loss=0.3118, simple_loss=0.3883, pruned_loss=0.1177, over 28368.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.328, pruned_loss=0.08395, over 5660546.03 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3319, pruned_loss=0.08804, over 5745587.54 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3279, pruned_loss=0.08345, over 5662019.83 frames. ], batch size: 369, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:36:11,608 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 16000, giga_loss[loss=0.2353, simple_loss=0.3241, pruned_loss=0.07324, over 28922.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3295, pruned_loss=0.08453, over 5662374.13 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3318, pruned_loss=0.08784, over 5746050.53 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3295, pruned_loss=0.0842, over 5661252.02 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:37:37,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-11 09:37:58,787 INFO [train.py:968] (1/2) Epoch 22, batch 16050, giga_loss[loss=0.2714, simple_loss=0.3446, pruned_loss=0.09906, over 28021.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3288, pruned_loss=0.08476, over 5655937.47 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3307, pruned_loss=0.08723, over 5744260.23 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3296, pruned_loss=0.08493, over 5651769.27 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:38:06,467 INFO [optim.py:369] (1/2) 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 09:38:55,478 INFO [train.py:968] (1/2) Epoch 22, batch 16100, giga_loss[loss=0.2649, simple_loss=0.3584, pruned_loss=0.08572, over 28927.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3331, pruned_loss=0.08746, over 5661336.32 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3302, pruned_loss=0.08698, over 5748171.45 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3343, pruned_loss=0.08777, over 5652841.36 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:39:48,556 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=973879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:39:50,552 INFO [train.py:968] (1/2) Epoch 22, batch 16150, giga_loss[loss=0.267, simple_loss=0.3445, pruned_loss=0.09473, over 28403.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3351, pruned_loss=0.08808, over 5653163.85 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3302, pruned_loss=0.08706, over 5746972.31 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3362, pruned_loss=0.08828, over 5645563.72 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:39:59,329 INFO [optim.py:369] (1/2) 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 09:40:55,422 INFO [train.py:968] (1/2) Epoch 22, batch 16200, giga_loss[loss=0.248, simple_loss=0.3356, pruned_loss=0.08017, over 28547.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3361, pruned_loss=0.08804, over 5652037.48 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3302, pruned_loss=0.08705, over 5747764.45 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3369, pruned_loss=0.08821, over 5644880.52 frames. ], batch size: 370, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:41:50,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=973969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:42:07,630 INFO [train.py:968] (1/2) Epoch 22, batch 16250, giga_loss[loss=0.2254, simple_loss=0.3063, pruned_loss=0.07227, over 29082.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3344, pruned_loss=0.08709, over 5656457.17 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3304, pruned_loss=0.08713, over 5748570.10 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.335, pruned_loss=0.08717, over 5649240.02 frames. ], batch size: 100, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:42:18,065 INFO [optim.py:369] (1/2) 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 09:43:08,155 INFO [train.py:968] (1/2) Epoch 22, batch 16300, giga_loss[loss=0.2234, simple_loss=0.3047, pruned_loss=0.07108, over 28611.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3318, pruned_loss=0.08579, over 5655590.51 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3301, pruned_loss=0.08693, over 5741057.31 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3326, pruned_loss=0.086, over 5654595.37 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:43:21,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6087, 2.0340, 1.2485, 1.4893], device='cuda:1'), covar=tensor([0.1059, 0.0529, 0.1075, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0390, 0.0438, 0.0512, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 09:44:03,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5909, 1.7983, 1.2394, 1.3517], device='cuda:1'), covar=tensor([0.0922, 0.0533, 0.0969, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0389, 0.0436, 0.0510, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 09:44:13,375 INFO [train.py:968] (1/2) Epoch 22, batch 16350, giga_loss[loss=0.2337, simple_loss=0.3005, pruned_loss=0.08341, over 24558.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3312, pruned_loss=0.0858, over 5659476.30 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.33, pruned_loss=0.08687, over 5743164.10 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3319, pruned_loss=0.086, over 5655282.14 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:44:21,293 INFO [optim.py:369] (1/2) 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 09:44:34,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-11 09:44:46,371 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:44:50,106 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=974112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:44:53,452 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=974115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:45:14,202 INFO [train.py:968] (1/2) Epoch 22, batch 16400, giga_loss[loss=0.2342, simple_loss=0.3149, pruned_loss=0.07679, over 29021.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3303, pruned_loss=0.0866, over 5654802.29 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.33, pruned_loss=0.08682, over 5744931.14 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3309, pruned_loss=0.08679, over 5649034.43 frames. ], batch size: 285, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:45:28,194 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=974144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:46:10,061 INFO [train.py:968] (1/2) Epoch 22, batch 16450, giga_loss[loss=0.2596, simple_loss=0.3459, pruned_loss=0.0866, over 28663.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3285, pruned_loss=0.08546, over 5659473.89 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3299, pruned_loss=0.08679, over 5747198.26 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.329, pruned_loss=0.08561, over 5650073.53 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:46:20,669 INFO [optim.py:369] (1/2) 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 09:47:15,196 INFO [train.py:968] (1/2) Epoch 22, batch 16500, giga_loss[loss=0.2152, simple_loss=0.3015, pruned_loss=0.06445, over 28857.00 frames. ], tot_loss[loss=0.248, simple_loss=0.328, pruned_loss=0.08402, over 5669353.77 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.33, pruned_loss=0.08685, over 5747987.27 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3283, pruned_loss=0.08407, over 5660928.62 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:47:42,641 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 16550, giga_loss[loss=0.262, simple_loss=0.3532, pruned_loss=0.08538, over 28046.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3282, pruned_loss=0.08218, over 5665493.50 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3303, pruned_loss=0.08708, over 5741315.72 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3281, pruned_loss=0.08193, over 5664039.89 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:48:21,619 INFO [optim.py:369] (1/2) 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 09:48:28,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-11 09:49:08,832 INFO [train.py:968] (1/2) Epoch 22, batch 16600, giga_loss[loss=0.2348, simple_loss=0.3288, pruned_loss=0.07035, over 28980.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3304, pruned_loss=0.08182, over 5679171.85 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3301, pruned_loss=0.08692, over 5743978.34 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3305, pruned_loss=0.08172, over 5674599.85 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:50:04,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-11 09:50:08,419 INFO [train.py:968] (1/2) Epoch 22, batch 16650, giga_loss[loss=0.2318, simple_loss=0.3201, pruned_loss=0.07179, over 28857.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3311, pruned_loss=0.08218, over 5671787.05 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3298, pruned_loss=0.0868, over 5745289.08 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3314, pruned_loss=0.08213, over 5665935.29 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:50:19,171 INFO [optim.py:369] (1/2) 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 09:50:28,751 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=974397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:50:32,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=974400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:50:38,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2861, 1.2012, 4.0348, 3.2580], device='cuda:1'), covar=tensor([0.1707, 0.2768, 0.0453, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0647, 0.0948, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 09:51:01,190 INFO [scaling.py:679] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=974429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:51:12,397 INFO [train.py:968] (1/2) Epoch 22, batch 16700, giga_loss[loss=0.2475, simple_loss=0.3306, pruned_loss=0.08221, over 29091.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3307, pruned_loss=0.08243, over 5666432.99 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3294, pruned_loss=0.08672, over 5748275.87 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3314, pruned_loss=0.08235, over 5657220.48 frames. ], batch size: 200, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:51:44,965 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-11 09:52:00,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5151, 3.6651, 1.5706, 1.6938], device='cuda:1'), covar=tensor([0.0945, 0.0368, 0.0903, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0548, 0.0386, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 09:52:07,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 09:52:16,098 INFO [train.py:968] (1/2) Epoch 22, batch 16750, giga_loss[loss=0.26, simple_loss=0.3432, pruned_loss=0.08843, over 28900.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.33, pruned_loss=0.08237, over 5664934.52 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3294, pruned_loss=0.0868, over 5752844.24 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3306, pruned_loss=0.0821, over 5650899.17 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:52:20,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=974484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:52:27,837 INFO [optim.py:369] (1/2) 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 09:53:19,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1953, 1.4143, 3.4966, 3.0179], device='cuda:1'), covar=tensor([0.1642, 0.2697, 0.0479, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0649, 0.0951, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 09:53:26,109 INFO [train.py:968] (1/2) Epoch 22, batch 16800, giga_loss[loss=0.2639, simple_loss=0.3492, pruned_loss=0.0893, over 28926.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3301, pruned_loss=0.08149, over 5668569.29 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3292, pruned_loss=0.0867, over 5754791.90 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3308, pruned_loss=0.0813, over 5654515.63 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 09:53:33,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6574, 2.0333, 1.7645, 1.6805], device='cuda:1'), covar=tensor([0.2144, 0.2536, 0.2312, 0.2396], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0730, 0.0696, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 09:53:35,224 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 16850, giga_loss[loss=0.2951, simple_loss=0.3855, pruned_loss=0.1023, over 28877.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3307, pruned_loss=0.08172, over 5660791.94 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08676, over 5757449.43 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3313, pruned_loss=0.0814, over 5645547.63 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:54:52,374 INFO [optim.py:369] (1/2) 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 09:55:40,840 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=974627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:55:45,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=974630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:55:46,141 INFO [train.py:968] (1/2) Epoch 22, batch 16900, giga_loss[loss=0.2451, simple_loss=0.3385, pruned_loss=0.07583, over 29023.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3341, pruned_loss=0.08347, over 5662749.91 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3287, pruned_loss=0.08658, over 5753175.88 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3351, pruned_loss=0.08327, over 5651258.00 frames. ], batch size: 165, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:56:26,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=974659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 09:56:53,913 INFO [train.py:968] (1/2) Epoch 22, batch 16950, giga_loss[loss=0.2205, simple_loss=0.3122, pruned_loss=0.06444, over 28177.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3341, pruned_loss=0.08338, over 5677369.27 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3288, pruned_loss=0.08664, over 5755558.19 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3349, pruned_loss=0.08311, over 5664863.52 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:57:06,997 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 17000, libri_loss[loss=0.2419, simple_loss=0.3165, pruned_loss=0.08365, over 29579.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3332, pruned_loss=0.08401, over 5673293.90 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3288, pruned_loss=0.0867, over 5754779.54 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.334, pruned_loss=0.0837, over 5662014.12 frames. ], batch size: 76, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:59:14,256 INFO [train.py:968] (1/2) Epoch 22, batch 17050, giga_loss[loss=0.2204, simple_loss=0.2989, pruned_loss=0.07092, over 24579.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3316, pruned_loss=0.08254, over 5673652.79 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3287, pruned_loss=0.08644, over 5748721.20 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3324, pruned_loss=0.08242, over 5667889.05 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 09:59:30,120 INFO [optim.py:369] (1/2) 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 09:59:33,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:00:22,144 INFO [train.py:968] (1/2) Epoch 22, batch 17100, giga_loss[loss=0.3044, simple_loss=0.387, pruned_loss=0.111, over 28730.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3317, pruned_loss=0.08274, over 5674085.63 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3287, pruned_loss=0.0866, over 5752577.46 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3324, pruned_loss=0.08241, over 5664426.49 frames. ], batch size: 263, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:00:28,409 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-11 10:00:52,280 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=974855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:01:10,526 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 17150, giga_loss[loss=0.2617, simple_loss=0.3447, pruned_loss=0.08938, over 28886.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3325, pruned_loss=0.08348, over 5678945.34 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3287, pruned_loss=0.08651, over 5754900.72 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3331, pruned_loss=0.08325, over 5668344.28 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:01:24,324 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-11 10:01:35,262 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=974911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:02:00,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-11 10:02:15,151 INFO [train.py:968] (1/2) Epoch 22, batch 17200, giga_loss[loss=0.2128, simple_loss=0.308, pruned_loss=0.05886, over 29017.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3337, pruned_loss=0.08419, over 5672264.58 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.08618, over 5749701.02 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3351, pruned_loss=0.08419, over 5664631.63 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:03:11,434 INFO [train.py:968] (1/2) Epoch 22, batch 17250, giga_loss[loss=0.2451, simple_loss=0.3298, pruned_loss=0.0802, over 28907.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.334, pruned_loss=0.0853, over 5669278.25 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3284, pruned_loss=0.08653, over 5747901.35 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3347, pruned_loss=0.08494, over 5663825.93 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:03:23,594 INFO [optim.py:369] (1/2) 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 10:04:03,346 INFO [train.py:968] (1/2) Epoch 22, batch 17300, giga_loss[loss=0.2416, simple_loss=0.3309, pruned_loss=0.07621, over 28956.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3305, pruned_loss=0.08458, over 5659694.12 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3283, pruned_loss=0.08652, over 5741531.14 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3314, pruned_loss=0.08427, over 5658063.93 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:04:26,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0066, 3.8577, 3.6213, 1.8593], device='cuda:1'), covar=tensor([0.0624, 0.0739, 0.0789, 0.2238], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1110, 0.0938, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 10:04:33,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:04:36,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:04:38,369 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=975059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:05:02,635 INFO [train.py:968] (1/2) Epoch 22, batch 17350, giga_loss[loss=0.2687, simple_loss=0.3406, pruned_loss=0.09837, over 28616.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3297, pruned_loss=0.08473, over 5657877.94 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3279, pruned_loss=0.08628, over 5744804.71 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3307, pruned_loss=0.08466, over 5652033.20 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:05:03,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-11 10:05:10,160 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:05:15,520 INFO [optim.py:369] (1/2) 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 10:05:31,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3582, 1.1902, 3.6503, 3.1743], device='cuda:1'), covar=tensor([0.1604, 0.2952, 0.0482, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0651, 0.0955, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 10:05:55,764 INFO [train.py:968] (1/2) Epoch 22, batch 17400, giga_loss[loss=0.3034, simple_loss=0.3805, pruned_loss=0.1132, over 28950.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3344, pruned_loss=0.08764, over 5663150.67 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3277, pruned_loss=0.08617, over 5751156.64 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3356, pruned_loss=0.08766, over 5649375.41 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:06:35,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:06:45,683 INFO [train.py:968] (1/2) Epoch 22, batch 17450, giga_loss[loss=0.294, simple_loss=0.3812, pruned_loss=0.1034, over 29031.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3438, pruned_loss=0.09296, over 5661595.29 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3275, pruned_loss=0.08623, over 5742745.79 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3451, pruned_loss=0.09299, over 5657448.84 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:06:54,304 INFO [optim.py:369] (1/2) 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 10:07:27,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:07:27,491 INFO [train.py:968] (1/2) Epoch 22, batch 17500, giga_loss[loss=0.2811, simple_loss=0.3515, pruned_loss=0.1054, over 27620.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3476, pruned_loss=0.09486, over 5666582.99 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3277, pruned_loss=0.08624, over 5737417.71 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3489, pruned_loss=0.09506, over 5666009.78 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:07:43,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:08:13,306 INFO [train.py:968] (1/2) Epoch 22, batch 17550, giga_loss[loss=0.2427, simple_loss=0.3201, pruned_loss=0.08271, over 28866.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3444, pruned_loss=0.09381, over 5670658.87 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3275, pruned_loss=0.08598, over 5740601.80 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3459, pruned_loss=0.09436, over 5666046.52 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:08:17,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5646, 1.7327, 1.6969, 1.4912], device='cuda:1'), covar=tensor([0.3015, 0.2678, 0.2152, 0.2510], device='cuda:1'), in_proj_covar=tensor([0.1931, 0.1851, 0.1762, 0.1921], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 10:08:24,568 INFO [optim.py:369] (1/2) 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 10:08:40,305 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:08:42,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:08:42,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5047, 1.7951, 1.4949, 1.3602], device='cuda:1'), covar=tensor([0.2685, 0.2594, 0.2865, 0.2511], device='cuda:1'), in_proj_covar=tensor([0.1502, 0.1085, 0.1330, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:08:54,077 INFO [train.py:968] (1/2) Epoch 22, batch 17600, giga_loss[loss=0.2679, simple_loss=0.328, pruned_loss=0.1039, over 26606.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3393, pruned_loss=0.09195, over 5669273.48 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3282, pruned_loss=0.08629, over 5733037.62 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3405, pruned_loss=0.09235, over 5670169.38 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:09:07,872 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:36,280 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:38,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:41,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4369, 1.4004, 1.2792, 1.6507], device='cuda:1'), covar=tensor([0.0787, 0.0360, 0.0352, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0117, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 10:09:41,685 INFO [train.py:968] (1/2) Epoch 22, batch 17650, giga_loss[loss=0.2043, simple_loss=0.2854, pruned_loss=0.06158, over 28910.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3316, pruned_loss=0.08855, over 5670524.45 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3283, pruned_loss=0.0863, over 5724647.92 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3325, pruned_loss=0.08888, over 5678980.80 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:09:48,199 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:50,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:09:50,949 INFO [optim.py:369] (1/2) 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 10:10:03,872 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:10:17,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:10:25,549 INFO [train.py:968] (1/2) Epoch 22, batch 17700, libri_loss[loss=0.3318, simple_loss=0.3926, pruned_loss=0.1355, over 19235.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3248, pruned_loss=0.08607, over 5667529.42 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3286, pruned_loss=0.08641, over 5718730.79 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3251, pruned_loss=0.08622, over 5679252.81 frames. ], batch size: 187, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:10:30,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=975434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:11:05,451 INFO [train.py:968] (1/2) Epoch 22, batch 17750, giga_loss[loss=0.2152, simple_loss=0.2921, pruned_loss=0.0692, over 28577.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3184, pruned_loss=0.08292, over 5684514.28 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3287, pruned_loss=0.08634, over 5727405.80 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3182, pruned_loss=0.08301, over 5684150.19 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:11:15,770 INFO [optim.py:369] (1/2) 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 10:11:17,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7265, 1.5582, 1.7890, 1.3454], device='cuda:1'), covar=tensor([0.2397, 0.3327, 0.1823, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0697, 0.0953, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 10:11:44,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6152, 1.8498, 1.5838, 1.6284], device='cuda:1'), covar=tensor([0.2375, 0.2346, 0.2485, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.1508, 0.1089, 0.1335, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:11:45,289 INFO [train.py:968] (1/2) Epoch 22, batch 17800, giga_loss[loss=0.2028, simple_loss=0.276, pruned_loss=0.06479, over 28964.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3142, pruned_loss=0.08127, over 5687960.79 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3291, pruned_loss=0.08643, over 5729398.04 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3135, pruned_loss=0.08119, over 5685203.77 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:11:56,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 10:12:24,833 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=975577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:12:27,929 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=975580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:12:28,227 INFO [train.py:968] (1/2) Epoch 22, batch 17850, giga_loss[loss=0.2062, simple_loss=0.2888, pruned_loss=0.06174, over 28717.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3114, pruned_loss=0.08008, over 5695772.13 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3292, pruned_loss=0.0864, over 5732075.97 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.3105, pruned_loss=0.07998, over 5690634.49 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:12:41,569 INFO [optim.py:369] (1/2) 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 10:12:55,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=975609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:13:13,793 INFO [train.py:968] (1/2) Epoch 22, batch 17900, giga_loss[loss=0.2291, simple_loss=0.3064, pruned_loss=0.07589, over 28930.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3081, pruned_loss=0.07881, over 5690389.43 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.329, pruned_loss=0.0862, over 5734410.17 frames. ], giga_tot_loss[loss=0.2325, simple_loss=0.3073, pruned_loss=0.07882, over 5683885.81 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:13:52,623 INFO [train.py:968] (1/2) Epoch 22, batch 17950, giga_loss[loss=0.2452, simple_loss=0.3132, pruned_loss=0.08864, over 28860.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3071, pruned_loss=0.07818, over 5692886.12 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3297, pruned_loss=0.0863, over 5735698.15 frames. ], giga_tot_loss[loss=0.2303, simple_loss=0.305, pruned_loss=0.07777, over 5684995.65 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:14:04,880 INFO [optim.py:369] (1/2) 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 10:14:36,962 INFO [train.py:968] (1/2) Epoch 22, batch 18000, giga_loss[loss=0.1972, simple_loss=0.2802, pruned_loss=0.05709, over 28987.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3039, pruned_loss=0.07642, over 5703591.14 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3294, pruned_loss=0.08603, over 5739166.15 frames. ], giga_tot_loss[loss=0.227, simple_loss=0.3017, pruned_loss=0.0761, over 5693257.53 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:14:36,963 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 10:14:45,417 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 10:15:07,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-11 10:15:28,540 INFO [train.py:968] (1/2) Epoch 22, batch 18050, giga_loss[loss=0.2238, simple_loss=0.2963, pruned_loss=0.07567, over 28516.00 frames. ], tot_loss[loss=0.2256, simple_loss=0.3008, pruned_loss=0.07524, over 5694852.83 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3295, pruned_loss=0.08586, over 5741801.95 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.2986, pruned_loss=0.075, over 5683839.19 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:15:38,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3656, 1.2927, 1.2424, 1.5786], device='cuda:1'), covar=tensor([0.0683, 0.0439, 0.0337, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 10:15:40,140 INFO [optim.py:369] (1/2) 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 10:16:11,021 INFO [train.py:968] (1/2) Epoch 22, batch 18100, libri_loss[loss=0.3108, simple_loss=0.3857, pruned_loss=0.1179, over 19146.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2986, pruned_loss=0.07412, over 5687423.79 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3303, pruned_loss=0.08626, over 5732097.94 frames. ], giga_tot_loss[loss=0.221, simple_loss=0.2954, pruned_loss=0.0733, over 5687510.58 frames. ], batch size: 188, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:16:58,219 INFO [train.py:968] (1/2) Epoch 22, batch 18150, giga_loss[loss=0.1923, simple_loss=0.268, pruned_loss=0.05834, over 28563.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2953, pruned_loss=0.07251, over 5697448.03 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3303, pruned_loss=0.08613, over 5735747.12 frames. ], giga_tot_loss[loss=0.2178, simple_loss=0.2922, pruned_loss=0.07173, over 5693399.10 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:17:10,707 INFO [optim.py:369] (1/2) 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 10:17:43,092 INFO [train.py:968] (1/2) Epoch 22, batch 18200, giga_loss[loss=0.216, simple_loss=0.2941, pruned_loss=0.06892, over 28881.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.294, pruned_loss=0.07242, over 5688128.07 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3306, pruned_loss=0.08625, over 5727173.74 frames. ], giga_tot_loss[loss=0.217, simple_loss=0.2909, pruned_loss=0.07152, over 5692532.60 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:17:55,791 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5943, 1.8981, 1.5371, 1.7140], device='cuda:1'), covar=tensor([0.2666, 0.2719, 0.3087, 0.2363], device='cuda:1'), in_proj_covar=tensor([0.1508, 0.1090, 0.1335, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:18:33,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1842, 0.8971, 1.0140, 1.4455], device='cuda:1'), covar=tensor([0.0786, 0.0409, 0.0357, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 10:18:34,710 INFO [train.py:968] (1/2) Epoch 22, batch 18250, giga_loss[loss=0.2574, simple_loss=0.3374, pruned_loss=0.08872, over 28676.00 frames. ], tot_loss[loss=0.2304, simple_loss=0.3045, pruned_loss=0.07818, over 5685936.37 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3307, pruned_loss=0.08629, over 5727264.07 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.3015, pruned_loss=0.07728, over 5688838.98 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:18:45,822 INFO [optim.py:369] (1/2) 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 10:18:46,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2845, 1.1512, 1.1501, 1.4973], device='cuda:1'), covar=tensor([0.0805, 0.0384, 0.0358, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 10:19:19,557 INFO [train.py:968] (1/2) Epoch 22, batch 18300, giga_loss[loss=0.3351, simple_loss=0.4019, pruned_loss=0.1342, over 28911.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3183, pruned_loss=0.08531, over 5688321.22 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3308, pruned_loss=0.08625, over 5731502.05 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.3153, pruned_loss=0.08451, over 5686162.72 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:19:58,237 INFO [train.py:968] (1/2) Epoch 22, batch 18350, giga_loss[loss=0.2656, simple_loss=0.3487, pruned_loss=0.09129, over 28594.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3285, pruned_loss=0.0901, over 5691984.97 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3308, pruned_loss=0.08617, over 5724276.78 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3259, pruned_loss=0.08962, over 5695315.80 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:20:09,655 INFO [optim.py:369] (1/2) 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 10:20:26,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:20:40,271 INFO [train.py:968] (1/2) Epoch 22, batch 18400, giga_loss[loss=0.2625, simple_loss=0.345, pruned_loss=0.08996, over 28938.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3352, pruned_loss=0.09263, over 5689798.98 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3311, pruned_loss=0.08621, over 5729551.27 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.333, pruned_loss=0.09234, over 5687003.76 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:21:22,505 INFO [train.py:968] (1/2) Epoch 22, batch 18450, giga_loss[loss=0.268, simple_loss=0.356, pruned_loss=0.09005, over 28627.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3384, pruned_loss=0.09283, over 5690771.76 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3312, pruned_loss=0.08621, over 5729803.03 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3365, pruned_loss=0.09266, over 5687991.57 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:21:33,229 INFO [optim.py:369] (1/2) 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 10:22:06,927 INFO [train.py:968] (1/2) Epoch 22, batch 18500, giga_loss[loss=0.2759, simple_loss=0.3549, pruned_loss=0.09847, over 27956.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.339, pruned_loss=0.09232, over 5678757.45 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.331, pruned_loss=0.08606, over 5722576.82 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3379, pruned_loss=0.0925, over 5681200.24 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:22:14,768 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5745, 1.8092, 1.4735, 1.5915], device='cuda:1'), covar=tensor([0.2827, 0.2847, 0.3257, 0.2487], device='cuda:1'), in_proj_covar=tensor([0.1505, 0.1088, 0.1332, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:22:48,020 INFO [train.py:968] (1/2) Epoch 22, batch 18550, libri_loss[loss=0.1947, simple_loss=0.2795, pruned_loss=0.05497, over 29401.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3413, pruned_loss=0.09376, over 5683995.67 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3313, pruned_loss=0.08609, over 5720955.84 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3406, pruned_loss=0.0942, over 5685486.63 frames. ], batch size: 67, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:23:01,114 INFO [optim.py:369] (1/2) 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 10:23:30,814 INFO [train.py:968] (1/2) Epoch 22, batch 18600, giga_loss[loss=0.2799, simple_loss=0.3609, pruned_loss=0.09943, over 28663.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3444, pruned_loss=0.09616, over 5689927.06 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3316, pruned_loss=0.08618, over 5726305.39 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3438, pruned_loss=0.09666, over 5685516.80 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:23:42,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:24:07,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7366, 1.1901, 2.8193, 2.6770], device='cuda:1'), covar=tensor([0.2168, 0.2870, 0.1074, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0648, 0.0955, 0.0900], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 10:24:13,141 INFO [train.py:968] (1/2) Epoch 22, batch 18650, giga_loss[loss=0.2919, simple_loss=0.3671, pruned_loss=0.1083, over 28941.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3474, pruned_loss=0.09817, over 5699611.59 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.332, pruned_loss=0.08636, over 5730079.22 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3469, pruned_loss=0.09868, over 5691876.52 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:24:14,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5878, 5.3685, 5.0652, 2.6676], device='cuda:1'), covar=tensor([0.0384, 0.0560, 0.0626, 0.1796], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1119, 0.0944, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 10:24:25,159 INFO [optim.py:369] (1/2) 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 10:24:43,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1391, 1.3151, 3.4374, 3.0221], device='cuda:1'), covar=tensor([0.1672, 0.2778, 0.0467, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0647, 0.0955, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 10:24:54,454 INFO [train.py:968] (1/2) Epoch 22, batch 18700, giga_loss[loss=0.3492, simple_loss=0.4149, pruned_loss=0.1417, over 28712.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3506, pruned_loss=0.09946, over 5706218.79 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3324, pruned_loss=0.08654, over 5735069.54 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3504, pruned_loss=0.1001, over 5694757.40 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:24:58,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976435.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 10:25:37,327 INFO [train.py:968] (1/2) Epoch 22, batch 18750, giga_loss[loss=0.269, simple_loss=0.3498, pruned_loss=0.09414, over 28916.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3525, pruned_loss=0.09929, over 5709608.28 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3324, pruned_loss=0.08646, over 5736763.68 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3525, pruned_loss=0.09997, over 5698729.55 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:25:45,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:25:48,925 INFO [optim.py:369] (1/2) 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 10:25:49,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5173, 1.8353, 1.4578, 1.5163], device='cuda:1'), covar=tensor([0.2749, 0.2777, 0.3180, 0.2445], device='cuda:1'), in_proj_covar=tensor([0.1510, 0.1093, 0.1335, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:26:01,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1715, 1.1858, 3.6453, 3.2020], device='cuda:1'), covar=tensor([0.1791, 0.3013, 0.0454, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0647, 0.0955, 0.0900], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 10:26:15,731 INFO [train.py:968] (1/2) Epoch 22, batch 18800, giga_loss[loss=0.3115, simple_loss=0.3946, pruned_loss=0.1142, over 28913.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3536, pruned_loss=0.09901, over 5715930.58 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3322, pruned_loss=0.08626, over 5740705.61 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3542, pruned_loss=0.1, over 5703198.51 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:26:19,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2312, 1.4761, 1.5116, 1.2986], device='cuda:1'), covar=tensor([0.2143, 0.1830, 0.2502, 0.2043], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0745, 0.0712, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 10:26:29,957 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:26:51,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:26:54,029 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 18850, giga_loss[loss=0.251, simple_loss=0.3186, pruned_loss=0.09171, over 23600.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.354, pruned_loss=0.0985, over 5703573.12 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.333, pruned_loss=0.08644, over 5739850.08 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3548, pruned_loss=0.09975, over 5691910.02 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:27:04,472 INFO [optim.py:369] (1/2) 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 10:27:34,520 INFO [train.py:968] (1/2) Epoch 22, batch 18900, giga_loss[loss=0.2546, simple_loss=0.3427, pruned_loss=0.08325, over 28694.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3535, pruned_loss=0.09751, over 5708013.40 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3337, pruned_loss=0.0869, over 5742844.08 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3538, pruned_loss=0.09829, over 5695654.84 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:27:37,374 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=976634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:27:39,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=976637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:27:44,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=976643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:28:00,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=976666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:28:08,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 10:28:14,983 INFO [train.py:968] (1/2) Epoch 22, batch 18950, giga_loss[loss=0.3211, simple_loss=0.3811, pruned_loss=0.1305, over 27710.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3517, pruned_loss=0.09618, over 5712819.00 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3339, pruned_loss=0.08705, over 5746132.98 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3522, pruned_loss=0.09688, over 5699402.55 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:28:25,420 INFO [optim.py:369] (1/2) 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 10:28:46,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:28:58,189 INFO [train.py:968] (1/2) Epoch 22, batch 19000, giga_loss[loss=0.3018, simple_loss=0.3677, pruned_loss=0.1179, over 28915.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3536, pruned_loss=0.09889, over 5701502.81 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.334, pruned_loss=0.08699, over 5747451.63 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3542, pruned_loss=0.09958, over 5689515.77 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:28:59,157 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 10:29:38,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-11 10:29:43,318 INFO [train.py:968] (1/2) Epoch 22, batch 19050, giga_loss[loss=0.2821, simple_loss=0.3506, pruned_loss=0.1068, over 28939.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3567, pruned_loss=0.1036, over 5681509.35 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3341, pruned_loss=0.0869, over 5740699.18 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3573, pruned_loss=0.1045, over 5677384.74 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:29:56,605 INFO [optim.py:369] (1/2) 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 10:30:07,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976810.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 10:30:08,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-11 10:30:21,880 INFO [train.py:968] (1/2) Epoch 22, batch 19100, giga_loss[loss=0.2672, simple_loss=0.3431, pruned_loss=0.09569, over 29079.00 frames. ], tot_loss[loss=0.2831, simple_loss=0.3567, pruned_loss=0.1047, over 5692728.14 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3342, pruned_loss=0.08695, over 5742195.73 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3575, pruned_loss=0.1056, over 5687349.51 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:30:46,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5611, 4.3749, 4.1604, 2.1155], device='cuda:1'), covar=tensor([0.0589, 0.0746, 0.0688, 0.1988], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1120, 0.0946, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 10:30:47,319 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=976861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:30:48,747 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-11 10:30:49,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=976864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:30:57,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3912, 1.7212, 1.3652, 1.4127], device='cuda:1'), covar=tensor([0.2643, 0.2691, 0.3068, 0.2347], device='cuda:1'), in_proj_covar=tensor([0.1506, 0.1091, 0.1331, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:31:02,510 INFO [train.py:968] (1/2) Epoch 22, batch 19150, giga_loss[loss=0.2555, simple_loss=0.3353, pruned_loss=0.08788, over 28949.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3551, pruned_loss=0.1044, over 5687812.00 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3349, pruned_loss=0.0872, over 5734753.01 frames. ], giga_tot_loss[loss=0.283, simple_loss=0.3555, pruned_loss=0.1053, over 5689437.17 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:31:13,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=976893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:31:17,394 INFO [optim.py:369] (1/2) 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 10:31:39,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:31:46,657 INFO [train.py:968] (1/2) Epoch 22, batch 19200, libri_loss[loss=0.2629, simple_loss=0.3427, pruned_loss=0.09151, over 19724.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3536, pruned_loss=0.1036, over 5673801.22 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3352, pruned_loss=0.08736, over 5718705.54 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3541, pruned_loss=0.1045, over 5689746.58 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:32:06,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=976952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:32:07,169 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=976953.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 10:32:08,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 19250, giga_loss[loss=0.2833, simple_loss=0.3577, pruned_loss=0.1044, over 28880.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3538, pruned_loss=0.1033, over 5664663.79 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.336, pruned_loss=0.08779, over 5712300.49 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3539, pruned_loss=0.104, over 5682165.37 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:32:31,623 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=976985.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 10:32:38,838 INFO [optim.py:369] (1/2) 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 10:32:57,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=977018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:33:09,689 INFO [train.py:968] (1/2) Epoch 22, batch 19300, giga_loss[loss=0.275, simple_loss=0.3509, pruned_loss=0.09948, over 28629.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5669656.14 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3362, pruned_loss=0.08782, over 5705220.06 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3511, pruned_loss=0.1014, over 5688073.71 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:33:32,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4428, 2.8830, 2.7410, 2.1489], device='cuda:1'), covar=tensor([0.2557, 0.1667, 0.1632, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1957, 0.1879, 0.1799, 0.1956], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 10:33:43,143 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:33:46,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:33:52,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 10:33:57,059 INFO [train.py:968] (1/2) Epoch 22, batch 19350, giga_loss[loss=0.2726, simple_loss=0.3415, pruned_loss=0.1018, over 28552.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3455, pruned_loss=0.09718, over 5667271.92 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3362, pruned_loss=0.08771, over 5709419.88 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3461, pruned_loss=0.09834, over 5677568.52 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:34:08,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:10,621 INFO [optim.py:369] (1/2) 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 10:34:11,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:11,473 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:12,234 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:13,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3497, 1.6150, 1.3895, 1.5089], device='cuda:1'), covar=tensor([0.0786, 0.0372, 0.0343, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 10:34:13,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:39,989 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:42,439 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:34:42,806 INFO [train.py:968] (1/2) Epoch 22, batch 19400, giga_loss[loss=0.2209, simple_loss=0.304, pruned_loss=0.06893, over 28801.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3406, pruned_loss=0.09469, over 5672191.82 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3369, pruned_loss=0.08798, over 5711209.37 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3405, pruned_loss=0.0955, over 5677963.73 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:35:11,169 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:35:14,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:35:29,946 INFO [train.py:968] (1/2) Epoch 22, batch 19450, giga_loss[loss=0.2082, simple_loss=0.2908, pruned_loss=0.06279, over 29049.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3355, pruned_loss=0.09255, over 5675973.88 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3369, pruned_loss=0.08801, over 5714927.95 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3355, pruned_loss=0.09329, over 5676540.17 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:35:41,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:35:44,758 INFO [optim.py:369] (1/2) 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 10:36:15,977 INFO [train.py:968] (1/2) Epoch 22, batch 19500, giga_loss[loss=0.245, simple_loss=0.3265, pruned_loss=0.08178, over 28996.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.333, pruned_loss=0.09077, over 5686263.84 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3364, pruned_loss=0.08772, over 5720725.41 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3334, pruned_loss=0.09171, over 5680702.16 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:36:30,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4238, 1.2868, 4.1345, 3.3628], device='cuda:1'), covar=tensor([0.1694, 0.2915, 0.0411, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0648, 0.0957, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 10:36:55,664 INFO [train.py:968] (1/2) Epoch 22, batch 19550, libri_loss[loss=0.2732, simple_loss=0.3593, pruned_loss=0.09358, over 29040.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3348, pruned_loss=0.09081, over 5701022.23 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3368, pruned_loss=0.08757, over 5729113.10 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3345, pruned_loss=0.09186, over 5687711.25 frames. ], batch size: 101, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:37:08,732 INFO [optim.py:369] (1/2) 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 10:37:40,309 INFO [train.py:968] (1/2) Epoch 22, batch 19600, giga_loss[loss=0.2411, simple_loss=0.3186, pruned_loss=0.08182, over 28838.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.335, pruned_loss=0.09087, over 5702046.41 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3373, pruned_loss=0.08765, over 5728325.12 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3344, pruned_loss=0.0917, over 5691610.73 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:37:48,910 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=977342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:38:21,029 INFO [train.py:968] (1/2) Epoch 22, batch 19650, giga_loss[loss=0.2366, simple_loss=0.3149, pruned_loss=0.07908, over 28967.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3334, pruned_loss=0.09029, over 5712183.52 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3378, pruned_loss=0.08776, over 5730181.72 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3325, pruned_loss=0.09089, over 5702097.46 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:38:32,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 10:38:33,664 INFO [optim.py:369] (1/2) 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 10:38:38,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5738, 2.8148, 2.5050, 2.6199], device='cuda:1'), covar=tensor([0.2038, 0.2242, 0.2381, 0.2171], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0749, 0.0712, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 10:38:59,630 INFO [train.py:968] (1/2) Epoch 22, batch 19700, libri_loss[loss=0.2933, simple_loss=0.3803, pruned_loss=0.1032, over 29178.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3308, pruned_loss=0.08904, over 5710364.40 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3382, pruned_loss=0.08782, over 5723091.75 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3294, pruned_loss=0.08948, over 5708806.00 frames. ], batch size: 101, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:39:37,528 INFO [train.py:968] (1/2) Epoch 22, batch 19750, giga_loss[loss=0.2161, simple_loss=0.2955, pruned_loss=0.0684, over 28756.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3278, pruned_loss=0.08783, over 5716279.41 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.338, pruned_loss=0.08755, over 5725525.22 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3268, pruned_loss=0.08843, over 5712722.63 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:39:51,605 INFO [optim.py:369] (1/2) 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 10:40:20,637 INFO [train.py:968] (1/2) Epoch 22, batch 19800, giga_loss[loss=0.2278, simple_loss=0.31, pruned_loss=0.07284, over 28959.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3264, pruned_loss=0.08734, over 5721123.77 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3381, pruned_loss=0.08751, over 5729458.48 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3252, pruned_loss=0.08786, over 5714559.81 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:40:48,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2760, 4.1239, 3.8788, 1.9190], device='cuda:1'), covar=tensor([0.0564, 0.0686, 0.0630, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.1203, 0.1115, 0.0941, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 10:40:57,451 INFO [train.py:968] (1/2) Epoch 22, batch 19850, giga_loss[loss=0.223, simple_loss=0.3034, pruned_loss=0.07132, over 28856.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3239, pruned_loss=0.086, over 5723870.72 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.339, pruned_loss=0.08783, over 5732070.61 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3219, pruned_loss=0.08612, over 5716177.80 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:41:12,119 INFO [optim.py:369] (1/2) 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 10:41:38,917 INFO [train.py:968] (1/2) Epoch 22, batch 19900, giga_loss[loss=0.2889, simple_loss=0.3543, pruned_loss=0.1118, over 28605.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.322, pruned_loss=0.08547, over 5719003.41 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3389, pruned_loss=0.08773, over 5732972.26 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3204, pruned_loss=0.08563, over 5712236.17 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:41:49,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3053, 3.1607, 1.4108, 1.4778], device='cuda:1'), covar=tensor([0.1091, 0.0324, 0.0900, 0.1404], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0547, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 10:42:08,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=977667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:42:19,763 INFO [train.py:968] (1/2) Epoch 22, batch 19950, giga_loss[loss=0.2454, simple_loss=0.3213, pruned_loss=0.08477, over 28656.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3211, pruned_loss=0.0853, over 5725106.74 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3393, pruned_loss=0.08778, over 5735688.87 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3193, pruned_loss=0.08536, over 5717150.40 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 10:42:33,341 INFO [optim.py:369] (1/2) 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 10:42:48,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=977717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:42:52,923 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-11 10:42:57,539 INFO [train.py:968] (1/2) Epoch 22, batch 20000, giga_loss[loss=0.2302, simple_loss=0.2988, pruned_loss=0.0808, over 28572.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3186, pruned_loss=0.08386, over 5729699.28 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3395, pruned_loss=0.08785, over 5735043.19 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3168, pruned_loss=0.08382, over 5723711.85 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:43:36,433 INFO [train.py:968] (1/2) Epoch 22, batch 20050, giga_loss[loss=0.21, simple_loss=0.2885, pruned_loss=0.06578, over 29099.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3183, pruned_loss=0.08346, over 5735796.60 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3402, pruned_loss=0.08805, over 5739467.03 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3157, pruned_loss=0.08315, over 5726775.18 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:43:51,889 INFO [optim.py:369] (1/2) 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 10:43:57,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2974, 3.1368, 2.9648, 1.4023], device='cuda:1'), covar=tensor([0.0885, 0.0983, 0.0852, 0.2285], device='cuda:1'), in_proj_covar=tensor([0.1208, 0.1115, 0.0941, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 10:44:22,150 INFO [train.py:968] (1/2) Epoch 22, batch 20100, giga_loss[loss=0.2541, simple_loss=0.3329, pruned_loss=0.08763, over 28804.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3219, pruned_loss=0.08607, over 5731754.63 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3402, pruned_loss=0.08805, over 5739467.03 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3199, pruned_loss=0.08583, over 5724733.15 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:44:47,454 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=977860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:44:50,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=977863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:45:09,447 INFO [train.py:968] (1/2) Epoch 22, batch 20150, giga_loss[loss=0.2581, simple_loss=0.3369, pruned_loss=0.08969, over 29016.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3278, pruned_loss=0.08982, over 5725469.67 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08797, over 5743494.83 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3259, pruned_loss=0.08972, over 5715850.17 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:45:17,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=977892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:45:22,850 INFO [optim.py:369] (1/2) 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 10:45:54,653 INFO [train.py:968] (1/2) Epoch 22, batch 20200, giga_loss[loss=0.282, simple_loss=0.3588, pruned_loss=0.1026, over 29021.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3369, pruned_loss=0.09576, over 5708872.53 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.34, pruned_loss=0.08781, over 5746898.65 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3352, pruned_loss=0.09595, over 5697450.85 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:45:58,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3021, 1.8620, 1.4089, 0.5515], device='cuda:1'), covar=tensor([0.4626, 0.2266, 0.3332, 0.5692], device='cuda:1'), in_proj_covar=tensor([0.1757, 0.1646, 0.1600, 0.1426], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 10:46:39,445 INFO [train.py:968] (1/2) Epoch 22, batch 20250, giga_loss[loss=0.3062, simple_loss=0.386, pruned_loss=0.1132, over 28672.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3428, pruned_loss=0.09884, over 5698606.30 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3405, pruned_loss=0.08803, over 5747524.11 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.341, pruned_loss=0.09898, over 5688071.05 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:46:55,515 INFO [optim.py:369] (1/2) 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 10:47:08,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3521, 1.5941, 1.5178, 1.4468], device='cuda:1'), covar=tensor([0.1586, 0.1627, 0.1919, 0.1667], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0750, 0.0712, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 10:47:25,010 INFO [train.py:968] (1/2) Epoch 22, batch 20300, giga_loss[loss=0.2939, simple_loss=0.3661, pruned_loss=0.1109, over 28864.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3469, pruned_loss=0.1003, over 5692649.53 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3404, pruned_loss=0.0879, over 5751672.92 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3457, pruned_loss=0.1008, over 5679394.66 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:47:28,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2407, 1.9362, 1.4770, 0.4591], device='cuda:1'), covar=tensor([0.5652, 0.3073, 0.4215, 0.6389], device='cuda:1'), in_proj_covar=tensor([0.1759, 0.1651, 0.1602, 0.1429], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 10:47:34,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=978042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:48:01,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4851, 1.7949, 1.4369, 1.5740], device='cuda:1'), covar=tensor([0.2658, 0.2660, 0.3016, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.1503, 0.1090, 0.1327, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:48:11,531 INFO [train.py:968] (1/2) Epoch 22, batch 20350, giga_loss[loss=0.287, simple_loss=0.3608, pruned_loss=0.1066, over 28452.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3518, pruned_loss=0.1029, over 5679929.86 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3405, pruned_loss=0.08796, over 5742938.30 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3508, pruned_loss=0.1034, over 5677145.82 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:48:26,978 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 20400, giga_loss[loss=0.2644, simple_loss=0.3439, pruned_loss=0.0924, over 28612.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3567, pruned_loss=0.1059, over 5676466.24 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3403, pruned_loss=0.08791, over 5744428.94 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3565, pruned_loss=0.1067, over 5671235.27 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:49:00,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2042, 1.8831, 1.6221, 1.4090], device='cuda:1'), covar=tensor([0.0928, 0.0301, 0.0305, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 10:49:40,100 INFO [train.py:968] (1/2) Epoch 22, batch 20450, giga_loss[loss=0.2627, simple_loss=0.3422, pruned_loss=0.09157, over 28709.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3521, pruned_loss=0.1021, over 5679605.91 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3405, pruned_loss=0.08806, over 5746550.11 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3519, pruned_loss=0.1028, over 5672730.98 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:49:43,229 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=978185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:49:46,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=978188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:49:52,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1516, 1.4343, 1.4889, 1.3445], device='cuda:1'), covar=tensor([0.1720, 0.1358, 0.2015, 0.1483], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0750, 0.0713, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 10:49:55,181 INFO [optim.py:369] (1/2) 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 10:50:02,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5790, 1.7713, 1.7351, 1.4377], device='cuda:1'), covar=tensor([0.3159, 0.2703, 0.2179, 0.2901], device='cuda:1'), in_proj_covar=tensor([0.1949, 0.1872, 0.1795, 0.1958], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 10:50:05,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6595, 1.6846, 1.9057, 1.4449], device='cuda:1'), covar=tensor([0.2018, 0.2521, 0.1685, 0.1996], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0699, 0.0951, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 10:50:08,708 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=978217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:50:18,387 INFO [train.py:968] (1/2) Epoch 22, batch 20500, giga_loss[loss=0.3515, simple_loss=0.4121, pruned_loss=0.1454, over 28915.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3492, pruned_loss=0.09961, over 5691949.04 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3407, pruned_loss=0.08836, over 5747997.89 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3491, pruned_loss=0.1003, over 5683473.16 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:50:31,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2860, 1.3472, 3.5468, 3.1852], device='cuda:1'), covar=tensor([0.1646, 0.2845, 0.0480, 0.1585], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0644, 0.0954, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 10:50:52,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 10:50:58,370 INFO [train.py:968] (1/2) Epoch 22, batch 20550, giga_loss[loss=0.2876, simple_loss=0.3587, pruned_loss=0.1083, over 28535.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3478, pruned_loss=0.09826, over 5693936.84 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3408, pruned_loss=0.08836, over 5749076.34 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3479, pruned_loss=0.09908, over 5684822.73 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:51:10,032 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-11 10:51:14,423 INFO [optim.py:369] (1/2) 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 10:51:26,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4311, 1.6893, 1.3993, 1.3524], device='cuda:1'), covar=tensor([0.2716, 0.2784, 0.3065, 0.2440], device='cuda:1'), in_proj_covar=tensor([0.1505, 0.1090, 0.1330, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:51:31,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4949, 1.7786, 1.4348, 1.4561], device='cuda:1'), covar=tensor([0.2467, 0.2510, 0.2743, 0.2330], device='cuda:1'), in_proj_covar=tensor([0.1506, 0.1091, 0.1331, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:51:32,846 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:51:41,289 INFO [train.py:968] (1/2) Epoch 22, batch 20600, giga_loss[loss=0.2662, simple_loss=0.3487, pruned_loss=0.09184, over 28941.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3482, pruned_loss=0.09781, over 5694715.98 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3409, pruned_loss=0.08833, over 5751486.55 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3483, pruned_loss=0.09864, over 5684561.81 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:51:45,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-11 10:52:25,160 INFO [train.py:968] (1/2) Epoch 22, batch 20650, giga_loss[loss=0.274, simple_loss=0.3454, pruned_loss=0.1013, over 28827.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3515, pruned_loss=0.1003, over 5699111.00 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3411, pruned_loss=0.08842, over 5753425.29 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3515, pruned_loss=0.101, over 5688693.46 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:52:38,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8940, 3.7484, 3.5143, 1.6330], device='cuda:1'), covar=tensor([0.0712, 0.0777, 0.0745, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1212, 0.1125, 0.0948, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 10:52:40,446 INFO [optim.py:369] (1/2) 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 10:52:46,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5210, 1.8000, 1.4795, 1.6180], device='cuda:1'), covar=tensor([0.2497, 0.2547, 0.2763, 0.2323], device='cuda:1'), in_proj_covar=tensor([0.1506, 0.1092, 0.1332, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:52:57,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9789, 1.2923, 3.3660, 3.0078], device='cuda:1'), covar=tensor([0.1789, 0.2652, 0.0533, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0642, 0.0953, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 10:53:06,756 INFO [train.py:968] (1/2) Epoch 22, batch 20700, giga_loss[loss=0.2521, simple_loss=0.3268, pruned_loss=0.08872, over 28591.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3532, pruned_loss=0.1015, over 5710211.12 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3414, pruned_loss=0.08849, over 5756486.50 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1023, over 5698035.09 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:53:35,070 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978460.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 10:53:51,223 INFO [train.py:968] (1/2) Epoch 22, batch 20750, giga_loss[loss=0.318, simple_loss=0.3833, pruned_loss=0.1263, over 28761.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.355, pruned_loss=0.1033, over 5689613.91 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3418, pruned_loss=0.08858, over 5755103.94 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3549, pruned_loss=0.1041, over 5680190.66 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:54:06,478 INFO [optim.py:369] (1/2) 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 10:54:35,178 INFO [train.py:968] (1/2) Epoch 22, batch 20800, giga_loss[loss=0.2706, simple_loss=0.356, pruned_loss=0.09257, over 28702.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3564, pruned_loss=0.1046, over 5695501.48 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08918, over 5758973.02 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.3558, pruned_loss=0.1052, over 5682189.36 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:55:12,967 INFO [train.py:968] (1/2) Epoch 22, batch 20850, giga_loss[loss=0.3089, simple_loss=0.3782, pruned_loss=0.1198, over 28962.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3562, pruned_loss=0.1042, over 5704499.85 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3432, pruned_loss=0.08927, over 5761054.75 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3557, pruned_loss=0.1049, over 5691260.82 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 10:55:29,130 INFO [optim.py:369] (1/2) 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 10:55:45,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2362, 1.3595, 1.3897, 1.1887], device='cuda:1'), covar=tensor([0.2390, 0.2546, 0.1659, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.1965, 0.1891, 0.1818, 0.1972], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 10:55:51,825 INFO [train.py:968] (1/2) Epoch 22, batch 20900, giga_loss[loss=0.2441, simple_loss=0.3272, pruned_loss=0.08049, over 28436.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3548, pruned_loss=0.1024, over 5708436.75 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3432, pruned_loss=0.08919, over 5766951.11 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3549, pruned_loss=0.1036, over 5689903.38 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:56:12,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9095, 2.2076, 1.6796, 2.2928], device='cuda:1'), covar=tensor([0.2833, 0.2882, 0.3263, 0.2412], device='cuda:1'), in_proj_covar=tensor([0.1509, 0.1095, 0.1333, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 10:56:27,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=978675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:56:31,891 INFO [train.py:968] (1/2) Epoch 22, batch 20950, giga_loss[loss=0.2555, simple_loss=0.345, pruned_loss=0.08295, over 28932.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3543, pruned_loss=0.1008, over 5711864.24 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08893, over 5769545.85 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.355, pruned_loss=0.1022, over 5693858.95 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:56:33,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2215, 1.5525, 1.5250, 1.1093], device='cuda:1'), covar=tensor([0.1770, 0.2648, 0.1485, 0.1707], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0699, 0.0950, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 10:56:43,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=978694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:56:46,175 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:56:48,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4179, 3.3714, 1.4834, 1.5989], device='cuda:1'), covar=tensor([0.1006, 0.0270, 0.0944, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0549, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 10:57:12,016 INFO [train.py:968] (1/2) Epoch 22, batch 21000, giga_loss[loss=0.2701, simple_loss=0.3517, pruned_loss=0.0943, over 28768.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3524, pruned_loss=0.09974, over 5710011.86 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08894, over 5771310.25 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3534, pruned_loss=0.101, over 5692966.49 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:57:12,016 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 10:57:20,795 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 10:57:57,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3812, 2.0874, 1.5030, 0.6557], device='cuda:1'), covar=tensor([0.4294, 0.2635, 0.3517, 0.4773], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1630, 0.1586, 0.1418], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 10:58:00,150 INFO [train.py:968] (1/2) Epoch 22, batch 21050, giga_loss[loss=0.272, simple_loss=0.3455, pruned_loss=0.09925, over 28906.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.35, pruned_loss=0.09863, over 5718832.42 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08884, over 5772607.55 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3512, pruned_loss=0.09997, over 5702906.81 frames. ], batch size: 285, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:58:06,740 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:58:36,227 INFO [train.py:968] (1/2) Epoch 22, batch 21100, giga_loss[loss=0.2591, simple_loss=0.3291, pruned_loss=0.09458, over 28602.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.09751, over 5720606.72 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08896, over 5773933.82 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3491, pruned_loss=0.09877, over 5704895.04 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:58:40,156 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=978835.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 10:58:41,607 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=978837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:58:43,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=978840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:59:06,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=978869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 10:59:14,985 INFO [train.py:968] (1/2) Epoch 22, batch 21150, giga_loss[loss=0.3149, simple_loss=0.3745, pruned_loss=0.1276, over 28692.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3472, pruned_loss=0.09749, over 5720221.79 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08912, over 5776509.69 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.348, pruned_loss=0.09849, over 5704459.29 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 10:59:18,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5686, 1.6550, 1.5764, 1.3932], device='cuda:1'), covar=tensor([0.2662, 0.2626, 0.2107, 0.2535], device='cuda:1'), in_proj_covar=tensor([0.1947, 0.1875, 0.1800, 0.1956], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 10:59:31,034 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 21200, giga_loss[loss=0.2893, simple_loss=0.3549, pruned_loss=0.1118, over 28916.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.349, pruned_loss=0.09915, over 5716373.14 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08978, over 5776249.69 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3489, pruned_loss=0.09972, over 5701577.34 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:00:22,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3073, 1.5478, 1.5038, 1.3901], device='cuda:1'), covar=tensor([0.2152, 0.2055, 0.2592, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0749, 0.0711, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 11:00:27,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2777, 2.6029, 1.2930, 1.4378], device='cuda:1'), covar=tensor([0.0991, 0.0322, 0.0903, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0545, 0.0383, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0024, 0.0029], device='cuda:1')
+2023-03-11 11:00:35,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=978978.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:00:36,845 INFO [train.py:968] (1/2) Epoch 22, batch 21250, giga_loss[loss=0.2798, simple_loss=0.3522, pruned_loss=0.1037, over 28764.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3504, pruned_loss=0.09983, over 5726467.52 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.09002, over 5778245.78 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3503, pruned_loss=0.1003, over 5711610.85 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:00:37,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=978981.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:00:52,569 INFO [optim.py:369] (1/2) 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 11:01:00,462 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979010.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:01:01,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6323, 1.7068, 1.8706, 1.4315], device='cuda:1'), covar=tensor([0.1799, 0.2481, 0.1456, 0.1761], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0702, 0.0952, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 11:01:16,190 INFO [train.py:968] (1/2) Epoch 22, batch 21300, giga_loss[loss=0.2696, simple_loss=0.3533, pruned_loss=0.09299, over 28705.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3501, pruned_loss=0.09927, over 5706060.03 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3443, pruned_loss=0.09063, over 5769509.13 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3495, pruned_loss=0.09933, over 5700359.36 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:01:30,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:01:34,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3356, 1.3741, 3.6380, 3.2317], device='cuda:1'), covar=tensor([0.2072, 0.3175, 0.0803, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0645, 0.0955, 0.0904], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 11:01:51,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:01:53,349 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:01:57,338 INFO [train.py:968] (1/2) Epoch 22, batch 21350, giga_loss[loss=0.2611, simple_loss=0.3456, pruned_loss=0.08828, over 28909.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3483, pruned_loss=0.09748, over 5720101.46 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3444, pruned_loss=0.09084, over 5773061.15 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.09755, over 5710922.91 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:02:06,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5215, 1.6474, 1.5288, 1.4458], device='cuda:1'), covar=tensor([0.2721, 0.2479, 0.2037, 0.2288], device='cuda:1'), in_proj_covar=tensor([0.1949, 0.1876, 0.1803, 0.1956], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:02:14,826 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 21400, libri_loss[loss=0.2769, simple_loss=0.3503, pruned_loss=0.1017, over 29575.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3471, pruned_loss=0.0967, over 5726537.01 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3447, pruned_loss=0.09106, over 5775160.21 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3465, pruned_loss=0.09662, over 5716618.10 frames. ], batch size: 76, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:02:57,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:06,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:08,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-11 11:03:12,218 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:16,826 INFO [train.py:968] (1/2) Epoch 22, batch 21450, giga_loss[loss=0.285, simple_loss=0.3487, pruned_loss=0.1106, over 28880.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3441, pruned_loss=0.09538, over 5729081.13 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3448, pruned_loss=0.09115, over 5776030.60 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3435, pruned_loss=0.09529, over 5720005.80 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:03:27,403 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:29,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:34,320 INFO [optim.py:369] (1/2) 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 11:03:46,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:48,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:53,655 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:03:53,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5593, 3.7004, 1.6164, 1.6193], device='cuda:1'), covar=tensor([0.0964, 0.0249, 0.0903, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0543, 0.0383, 0.0424], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 11:03:58,677 INFO [train.py:968] (1/2) Epoch 22, batch 21500, giga_loss[loss=0.2469, simple_loss=0.3229, pruned_loss=0.08544, over 28455.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3405, pruned_loss=0.09377, over 5722545.53 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3446, pruned_loss=0.09112, over 5777343.14 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3403, pruned_loss=0.09375, over 5713810.74 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:04:01,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 11:04:11,498 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:04:37,294 INFO [train.py:968] (1/2) Epoch 22, batch 21550, giga_loss[loss=0.2258, simple_loss=0.3063, pruned_loss=0.07262, over 28631.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09395, over 5725124.25 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3446, pruned_loss=0.09123, over 5776310.84 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3402, pruned_loss=0.09388, over 5718246.26 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:04:52,934 INFO [optim.py:369] (1/2) 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 11:05:01,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:05:04,158 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 21600, giga_loss[loss=0.3056, simple_loss=0.3739, pruned_loss=0.1186, over 28812.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3401, pruned_loss=0.09447, over 5722350.20 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3445, pruned_loss=0.09119, over 5777707.97 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3399, pruned_loss=0.09449, over 5715050.09 frames. ], batch size: 243, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:05:24,584 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 21650, giga_loss[loss=0.2234, simple_loss=0.2974, pruned_loss=0.07474, over 28729.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.339, pruned_loss=0.09475, over 5718466.80 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3452, pruned_loss=0.09166, over 5780005.53 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3383, pruned_loss=0.09443, over 5709924.93 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:06:16,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4230, 2.7527, 2.3477, 1.9730], device='cuda:1'), covar=tensor([0.2755, 0.1959, 0.2370, 0.2720], device='cuda:1'), in_proj_covar=tensor([0.1957, 0.1882, 0.1808, 0.1963], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:06:17,948 INFO [optim.py:369] (1/2) 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 11:06:26,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 11:06:39,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6273, 1.7602, 1.8039, 1.3705], device='cuda:1'), covar=tensor([0.1646, 0.2599, 0.1447, 0.1653], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0701, 0.0952, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 11:06:39,923 INFO [train.py:968] (1/2) Epoch 22, batch 21700, giga_loss[loss=0.2601, simple_loss=0.3256, pruned_loss=0.09729, over 28660.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3379, pruned_loss=0.0946, over 5721733.97 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3459, pruned_loss=0.09221, over 5778530.97 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3364, pruned_loss=0.09392, over 5714263.40 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:06:57,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:07:21,053 INFO [train.py:968] (1/2) Epoch 22, batch 21750, giga_loss[loss=0.2967, simple_loss=0.3542, pruned_loss=0.1196, over 23684.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.335, pruned_loss=0.09333, over 5706175.80 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3463, pruned_loss=0.09264, over 5771282.81 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3332, pruned_loss=0.09238, over 5706340.35 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:07:21,258 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979481.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:07:36,096 INFO [optim.py:369] (1/2) 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 11:07:57,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:07:58,084 INFO [train.py:968] (1/2) Epoch 22, batch 21800, giga_loss[loss=0.2685, simple_loss=0.3405, pruned_loss=0.09827, over 28931.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3336, pruned_loss=0.09253, over 5713051.34 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3463, pruned_loss=0.09293, over 5775811.92 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3318, pruned_loss=0.09153, over 5707414.24 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:07:59,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8951, 1.9498, 2.1055, 1.6400], device='cuda:1'), covar=tensor([0.1892, 0.2475, 0.1506, 0.1796], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0702, 0.0953, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 11:08:14,679 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 21850, libri_loss[loss=0.2697, simple_loss=0.3403, pruned_loss=0.09956, over 29555.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3343, pruned_loss=0.09323, over 5710029.28 frames. ], libri_tot_loss[loss=0.266, simple_loss=0.346, pruned_loss=0.09302, over 5777560.62 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3328, pruned_loss=0.09236, over 5702798.46 frames. ], batch size: 76, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:08:52,344 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:08:55,036 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:09:19,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:09:25,098 INFO [train.py:968] (1/2) Epoch 22, batch 21900, giga_loss[loss=0.361, simple_loss=0.4088, pruned_loss=0.1566, over 26663.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3375, pruned_loss=0.09436, over 5709441.29 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3464, pruned_loss=0.09334, over 5778407.96 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3359, pruned_loss=0.09338, over 5702282.31 frames. ], batch size: 555, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:09:59,604 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:02,316 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 21950, giga_loss[loss=0.2936, simple_loss=0.3546, pruned_loss=0.1163, over 24008.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3392, pruned_loss=0.09453, over 5714757.24 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3469, pruned_loss=0.09383, over 5781809.11 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3373, pruned_loss=0.09333, over 5704651.21 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:10:13,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5341, 1.6805, 1.7118, 1.3044], device='cuda:1'), covar=tensor([0.2005, 0.2558, 0.1652, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0699, 0.0949, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 11:10:16,273 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:18,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=979695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:23,240 INFO [optim.py:369] (1/2) 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 11:10:25,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=979704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:10:45,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1774, 1.3498, 3.4211, 3.2728], device='cuda:1'), covar=tensor([0.1801, 0.2940, 0.0814, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0646, 0.0957, 0.0903], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 11:10:45,175 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 22000, giga_loss[loss=0.2478, simple_loss=0.3275, pruned_loss=0.08404, over 29037.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3414, pruned_loss=0.09499, over 5708029.75 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3471, pruned_loss=0.09411, over 5783872.27 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3395, pruned_loss=0.09381, over 5696861.73 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:11:24,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3535, 1.5102, 1.4720, 1.2970], device='cuda:1'), covar=tensor([0.3123, 0.2589, 0.2139, 0.2675], device='cuda:1'), in_proj_covar=tensor([0.1955, 0.1881, 0.1811, 0.1963], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:11:32,032 INFO [train.py:968] (1/2) Epoch 22, batch 22050, giga_loss[loss=0.2568, simple_loss=0.3438, pruned_loss=0.08494, over 28599.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3413, pruned_loss=0.09467, over 5700089.98 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3474, pruned_loss=0.09452, over 5777224.35 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3395, pruned_loss=0.09337, over 5694960.69 frames. ], batch size: 336, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:11:50,585 INFO [optim.py:369] (1/2) 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 11:12:12,818 INFO [train.py:968] (1/2) Epoch 22, batch 22100, giga_loss[loss=0.2665, simple_loss=0.3337, pruned_loss=0.0996, over 28567.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3416, pruned_loss=0.09488, over 5706211.34 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3478, pruned_loss=0.09505, over 5779639.56 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3396, pruned_loss=0.09335, over 5698117.79 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:12:32,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=979856.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:12:51,860 INFO [train.py:968] (1/2) Epoch 22, batch 22150, giga_loss[loss=0.262, simple_loss=0.3488, pruned_loss=0.08756, over 29001.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3435, pruned_loss=0.09632, over 5699834.59 frames. ], libri_tot_loss[loss=0.2694, simple_loss=0.3481, pruned_loss=0.0954, over 5773070.54 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3416, pruned_loss=0.09476, over 5698196.04 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:13:12,219 INFO [optim.py:369] (1/2) 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 11:13:34,261 INFO [train.py:968] (1/2) Epoch 22, batch 22200, giga_loss[loss=0.3061, simple_loss=0.3699, pruned_loss=0.1212, over 28735.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3445, pruned_loss=0.09721, over 5707193.83 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3478, pruned_loss=0.09527, over 5775899.90 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3432, pruned_loss=0.09614, over 5701801.79 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:13:57,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2873, 1.3299, 1.1535, 1.4876], device='cuda:1'), covar=tensor([0.0751, 0.0335, 0.0363, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0117, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 11:14:03,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 22250, giga_loss[loss=0.2853, simple_loss=0.354, pruned_loss=0.1083, over 28408.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09789, over 5703706.75 frames. ], libri_tot_loss[loss=0.2701, simple_loss=0.3484, pruned_loss=0.09589, over 5779419.70 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3444, pruned_loss=0.09654, over 5694411.03 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:14:17,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=979985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:14:20,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4366, 1.6039, 1.5581, 1.2799], device='cuda:1'), covar=tensor([0.3185, 0.2617, 0.2088, 0.2915], device='cuda:1'), in_proj_covar=tensor([0.1966, 0.1896, 0.1825, 0.1977], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:14:28,908 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=979999.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:14:32,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980002.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:14:32,439 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 22300, giga_loss[loss=0.3111, simple_loss=0.3801, pruned_loss=0.121, over 28460.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3492, pruned_loss=0.09963, over 5705510.27 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.349, pruned_loss=0.09663, over 5775508.71 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3472, pruned_loss=0.09796, over 5698246.31 frames. ], batch size: 65, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:14:52,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=980031.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:15:30,103 INFO [train.py:968] (1/2) Epoch 22, batch 22350, giga_loss[loss=0.2615, simple_loss=0.3457, pruned_loss=0.08863, over 29050.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3501, pruned_loss=0.09972, over 5713711.94 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3497, pruned_loss=0.09714, over 5775802.93 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.348, pruned_loss=0.09801, over 5705396.10 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:15:41,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9497, 3.7748, 3.5773, 1.9300], device='cuda:1'), covar=tensor([0.0666, 0.0796, 0.0760, 0.1867], device='cuda:1'), in_proj_covar=tensor([0.1207, 0.1118, 0.0949, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 11:15:46,202 INFO [optim.py:369] (1/2) 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 11:16:09,208 INFO [train.py:968] (1/2) Epoch 22, batch 22400, giga_loss[loss=0.2715, simple_loss=0.3522, pruned_loss=0.09543, over 28862.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3512, pruned_loss=0.1001, over 5714410.40 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3496, pruned_loss=0.09712, over 5776503.92 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3495, pruned_loss=0.09883, over 5706967.50 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:16:31,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-11 11:16:44,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6270, 1.6868, 1.6826, 1.5502], device='cuda:1'), covar=tensor([0.2965, 0.2578, 0.2116, 0.2570], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1900, 0.1829, 0.1976], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:16:54,078 INFO [train.py:968] (1/2) Epoch 22, batch 22450, giga_loss[loss=0.2621, simple_loss=0.3423, pruned_loss=0.09095, over 28314.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3523, pruned_loss=0.1011, over 5708469.90 frames. ], libri_tot_loss[loss=0.2721, simple_loss=0.3498, pruned_loss=0.09722, over 5767887.40 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3509, pruned_loss=0.09996, over 5710306.64 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:17:13,512 INFO [optim.py:369] (1/2) 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 11:17:35,793 INFO [train.py:968] (1/2) Epoch 22, batch 22500, giga_loss[loss=0.2414, simple_loss=0.3276, pruned_loss=0.07758, over 29015.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3511, pruned_loss=0.1005, over 5701208.23 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3505, pruned_loss=0.09774, over 5761540.81 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3493, pruned_loss=0.09923, over 5706877.51 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:17:48,843 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:18:14,621 INFO [train.py:968] (1/2) Epoch 22, batch 22550, libri_loss[loss=0.2805, simple_loss=0.3542, pruned_loss=0.1034, over 29548.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3486, pruned_loss=0.09936, over 5696248.48 frames. ], libri_tot_loss[loss=0.2737, simple_loss=0.351, pruned_loss=0.09825, over 5750000.07 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3468, pruned_loss=0.09794, over 5708889.13 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:18:32,693 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 22600, giga_loss[loss=0.2561, simple_loss=0.3342, pruned_loss=0.08897, over 28556.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3442, pruned_loss=0.09721, over 5700004.60 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3507, pruned_loss=0.0984, over 5751700.04 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3429, pruned_loss=0.09597, over 5707191.18 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:19:03,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:19:16,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:19:23,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 11:19:33,163 INFO [train.py:968] (1/2) Epoch 22, batch 22650, giga_loss[loss=0.249, simple_loss=0.3243, pruned_loss=0.08691, over 28722.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3426, pruned_loss=0.09546, over 5684928.47 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3516, pruned_loss=0.09888, over 5736339.33 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3406, pruned_loss=0.09401, over 5703571.27 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:19:55,757 INFO [optim.py:369] (1/2) 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 11:20:21,256 INFO [train.py:968] (1/2) Epoch 22, batch 22700, giga_loss[loss=0.2425, simple_loss=0.3279, pruned_loss=0.07852, over 28796.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3443, pruned_loss=0.09464, over 5687419.00 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3516, pruned_loss=0.09895, over 5737218.63 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3426, pruned_loss=0.09341, over 5700974.98 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:20:31,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-11 11:20:51,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1244, 3.2431, 1.3563, 1.3127], device='cuda:1'), covar=tensor([0.1253, 0.0417, 0.1070, 0.1633], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0550, 0.0386, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 11:21:00,759 INFO [train.py:968] (1/2) Epoch 22, batch 22750, libri_loss[loss=0.2726, simple_loss=0.3474, pruned_loss=0.09893, over 29568.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3458, pruned_loss=0.09584, over 5685940.82 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.352, pruned_loss=0.09943, over 5736404.81 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3441, pruned_loss=0.09437, over 5696132.56 frames. ], batch size: 77, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:21:04,601 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:07,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7100, 1.7649, 1.3668, 1.3533], device='cuda:1'), covar=tensor([0.0992, 0.0702, 0.1058, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0516, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 11:21:17,309 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=980503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:17,739 INFO [optim.py:369] (1/2) 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 11:21:19,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:21,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:21:30,696 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 22800, giga_loss[loss=0.2519, simple_loss=0.3284, pruned_loss=0.08771, over 29077.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3441, pruned_loss=0.0962, over 5691288.59 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3522, pruned_loss=0.09986, over 5740085.57 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3423, pruned_loss=0.09458, over 5695081.58 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:21:40,347 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-11 11:21:43,994 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=980535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:22:21,621 INFO [train.py:968] (1/2) Epoch 22, batch 22850, giga_loss[loss=0.2839, simple_loss=0.3475, pruned_loss=0.1101, over 29062.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3425, pruned_loss=0.0967, over 5701437.15 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3521, pruned_loss=0.09984, over 5743243.51 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3409, pruned_loss=0.09532, over 5700500.94 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:22:40,771 INFO [optim.py:369] (1/2) 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 11:22:55,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5347, 1.7365, 1.7928, 1.3086], device='cuda:1'), covar=tensor([0.1734, 0.2877, 0.1542, 0.1931], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0699, 0.0947, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 11:22:56,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:23:01,365 INFO [train.py:968] (1/2) Epoch 22, batch 22900, giga_loss[loss=0.238, simple_loss=0.3067, pruned_loss=0.08459, over 28537.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3409, pruned_loss=0.09686, over 5712063.34 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3523, pruned_loss=0.1001, over 5745591.65 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3391, pruned_loss=0.09538, over 5707960.30 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:23:02,837 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-11 11:23:40,687 INFO [train.py:968] (1/2) Epoch 22, batch 22950, giga_loss[loss=0.2321, simple_loss=0.3098, pruned_loss=0.07725, over 28969.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3398, pruned_loss=0.09719, over 5711210.97 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.352, pruned_loss=0.1002, over 5749696.95 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3383, pruned_loss=0.09584, over 5703523.31 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:23:54,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9119, 1.1184, 2.8035, 2.6503], device='cuda:1'), covar=tensor([0.1672, 0.2744, 0.0588, 0.1255], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0643, 0.0956, 0.0902], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 11:23:58,524 INFO [optim.py:369] (1/2) 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 11:24:03,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3304, 2.9845, 1.3738, 1.4281], device='cuda:1'), covar=tensor([0.0960, 0.0363, 0.0980, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0549, 0.0386, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 11:24:20,884 INFO [train.py:968] (1/2) Epoch 22, batch 23000, libri_loss[loss=0.3152, simple_loss=0.3808, pruned_loss=0.1248, over 29515.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3387, pruned_loss=0.09624, over 5712866.00 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3522, pruned_loss=0.1004, over 5742735.98 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3371, pruned_loss=0.09494, over 5713311.13 frames. ], batch size: 84, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:24:33,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3280, 3.1618, 2.9817, 1.4426], device='cuda:1'), covar=tensor([0.0892, 0.1044, 0.0929, 0.2253], device='cuda:1'), in_proj_covar=tensor([0.1220, 0.1126, 0.0954, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 11:24:48,625 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=980768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:24:50,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=980771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:24:57,850 INFO [train.py:968] (1/2) Epoch 22, batch 23050, giga_loss[loss=0.2101, simple_loss=0.2913, pruned_loss=0.06443, over 28819.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3353, pruned_loss=0.09476, over 5706767.66 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3524, pruned_loss=0.1007, over 5741341.24 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3334, pruned_loss=0.09327, over 5707267.67 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:25:03,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9590, 3.8200, 3.5884, 1.7527], device='cuda:1'), covar=tensor([0.0704, 0.0830, 0.0788, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.1223, 0.1130, 0.0957, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 11:25:12,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:25:15,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2037, 1.3810, 1.2978, 1.1497], device='cuda:1'), covar=tensor([0.2921, 0.2575, 0.1700, 0.2468], device='cuda:1'), in_proj_covar=tensor([0.1966, 0.1896, 0.1819, 0.1972], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:25:29,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.54 vs. limit=5.0
+2023-03-11 11:25:39,963 INFO [train.py:968] (1/2) Epoch 22, batch 23100, giga_loss[loss=0.2292, simple_loss=0.303, pruned_loss=0.07769, over 28489.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3316, pruned_loss=0.09297, over 5710225.65 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.353, pruned_loss=0.1013, over 5746063.94 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3291, pruned_loss=0.09112, over 5705705.60 frames. ], batch size: 60, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:26:08,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 11:26:18,979 INFO [train.py:968] (1/2) Epoch 22, batch 23150, giga_loss[loss=0.256, simple_loss=0.3317, pruned_loss=0.09018, over 28828.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3303, pruned_loss=0.09211, over 5713230.88 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3534, pruned_loss=0.1017, over 5748292.23 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3277, pruned_loss=0.09022, over 5707221.25 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:26:22,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=980884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:26:39,301 INFO [optim.py:369] (1/2) 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 11:26:40,117 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:26:53,751 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=980919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:27:01,462 INFO [train.py:968] (1/2) Epoch 22, batch 23200, giga_loss[loss=0.2963, simple_loss=0.3687, pruned_loss=0.112, over 28973.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.332, pruned_loss=0.09223, over 5708199.83 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3538, pruned_loss=0.1019, over 5741489.55 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3291, pruned_loss=0.09025, over 5708599.66 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:27:40,596 INFO [train.py:968] (1/2) Epoch 22, batch 23250, giga_loss[loss=0.279, simple_loss=0.3564, pruned_loss=0.1008, over 28999.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3352, pruned_loss=0.09352, over 5714207.01 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3538, pruned_loss=0.1022, over 5745969.35 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3323, pruned_loss=0.09144, over 5709500.57 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:27:53,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7987, 1.8898, 1.6262, 1.8734], device='cuda:1'), covar=tensor([0.2544, 0.2785, 0.3109, 0.2464], device='cuda:1'), in_proj_covar=tensor([0.1503, 0.1085, 0.1325, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 11:28:01,613 INFO [optim.py:369] (1/2) 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 11:28:20,143 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:28:22,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:28:22,613 INFO [train.py:968] (1/2) Epoch 22, batch 23300, giga_loss[loss=0.2666, simple_loss=0.3458, pruned_loss=0.09368, over 28676.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3399, pruned_loss=0.09567, over 5712837.36 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.354, pruned_loss=0.1023, over 5746646.50 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3373, pruned_loss=0.09388, over 5708401.64 frames. ], batch size: 242, lr: 1.45e-03, grad_scale: 8.0
+2023-03-11 11:28:46,166 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:29:04,390 INFO [train.py:968] (1/2) Epoch 22, batch 23350, giga_loss[loss=0.2857, simple_loss=0.3556, pruned_loss=0.1079, over 28912.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3429, pruned_loss=0.09704, over 5707428.65 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3542, pruned_loss=0.1027, over 5750821.99 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3402, pruned_loss=0.09502, over 5699014.09 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:29:07,806 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:29:23,157 INFO [optim.py:369] (1/2) 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 11:29:45,467 INFO [train.py:968] (1/2) Epoch 22, batch 23400, giga_loss[loss=0.3154, simple_loss=0.3784, pruned_loss=0.1262, over 28873.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3469, pruned_loss=0.09946, over 5696287.65 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3556, pruned_loss=0.1039, over 5744627.60 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3432, pruned_loss=0.09655, over 5693687.14 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:29:58,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2749, 1.5988, 1.2745, 1.0692], device='cuda:1'), covar=tensor([0.2419, 0.2482, 0.2858, 0.2197], device='cuda:1'), in_proj_covar=tensor([0.1507, 0.1088, 0.1329, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 11:30:18,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981168.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:30:28,499 INFO [train.py:968] (1/2) Epoch 22, batch 23450, giga_loss[loss=0.3031, simple_loss=0.3765, pruned_loss=0.1148, over 28891.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.352, pruned_loss=0.1042, over 5694365.22 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3553, pruned_loss=0.1038, over 5745798.55 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3491, pruned_loss=0.1019, over 5690183.86 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:30:53,147 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 23500, giga_loss[loss=0.3084, simple_loss=0.3819, pruned_loss=0.1175, over 29048.00 frames. ], tot_loss[loss=0.2869, simple_loss=0.357, pruned_loss=0.1084, over 5683415.60 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3554, pruned_loss=0.1041, over 5739341.09 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3546, pruned_loss=0.1063, over 5683844.86 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:31:30,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-11 11:31:31,843 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:31:34,259 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:32:07,674 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 23550, giga_loss[loss=0.2938, simple_loss=0.3695, pruned_loss=0.1091, over 28865.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3649, pruned_loss=0.1143, over 5680298.44 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3556, pruned_loss=0.1044, over 5738570.94 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.363, pruned_loss=0.1125, over 5680184.41 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:32:15,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:32:21,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:32:36,009 INFO [optim.py:369] (1/2) 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 11:32:43,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4070, 3.2145, 1.5425, 1.5239], device='cuda:1'), covar=tensor([0.0994, 0.0352, 0.0875, 0.1320], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0549, 0.0385, 0.0425], device='cuda:1'), out_proj_covar=tensor([0.0026, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 11:32:58,374 INFO [train.py:968] (1/2) Epoch 22, batch 23600, giga_loss[loss=0.3985, simple_loss=0.4257, pruned_loss=0.1857, over 27628.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3712, pruned_loss=0.1197, over 5679549.98 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.356, pruned_loss=0.1048, over 5741562.80 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3695, pruned_loss=0.1182, over 5675795.55 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:33:11,198 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=981341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:33:50,418 INFO [train.py:968] (1/2) Epoch 22, batch 23650, giga_loss[loss=0.4296, simple_loss=0.4561, pruned_loss=0.2016, over 27887.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3774, pruned_loss=0.1251, over 5661881.32 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3564, pruned_loss=0.1052, over 5734930.10 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3762, pruned_loss=0.124, over 5662677.04 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:33:56,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4684, 1.4911, 1.4560, 1.4495], device='cuda:1'), covar=tensor([0.1688, 0.1882, 0.1523, 0.1597], device='cuda:1'), in_proj_covar=tensor([0.1971, 0.1906, 0.1823, 0.1970], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:34:13,769 INFO [optim.py:369] (1/2) 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 11:34:34,281 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:34:38,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:34:41,033 INFO [train.py:968] (1/2) Epoch 22, batch 23700, giga_loss[loss=0.2834, simple_loss=0.3582, pruned_loss=0.1043, over 29089.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3806, pruned_loss=0.1275, over 5660096.67 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3565, pruned_loss=0.1054, over 5731810.15 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3801, pruned_loss=0.1269, over 5661365.05 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:34:46,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:34:48,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:02,658 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:09,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:18,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:35:26,979 INFO [train.py:968] (1/2) Epoch 22, batch 23750, giga_loss[loss=0.3243, simple_loss=0.3899, pruned_loss=0.1294, over 28886.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3824, pruned_loss=0.1296, over 5651941.36 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3565, pruned_loss=0.1055, over 5726183.34 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3825, pruned_loss=0.1295, over 5656461.29 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:35:51,634 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 23800, giga_loss[loss=0.3648, simple_loss=0.4054, pruned_loss=0.1621, over 28761.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3849, pruned_loss=0.1331, over 5630123.77 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3564, pruned_loss=0.1055, over 5710918.65 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3854, pruned_loss=0.1333, over 5645018.07 frames. ], batch size: 243, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:36:20,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4845, 4.2988, 4.0842, 1.8514], device='cuda:1'), covar=tensor([0.0710, 0.0884, 0.1040, 0.2160], device='cuda:1'), in_proj_covar=tensor([0.1228, 0.1138, 0.0963, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 11:36:31,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:37:07,312 INFO [train.py:968] (1/2) Epoch 22, batch 23850, giga_loss[loss=0.2857, simple_loss=0.362, pruned_loss=0.1047, over 28796.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3865, pruned_loss=0.1351, over 5633141.63 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3567, pruned_loss=0.1061, over 5714714.91 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3874, pruned_loss=0.1356, over 5639562.07 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:37:36,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5899, 2.0676, 1.3193, 1.0120], device='cuda:1'), covar=tensor([0.6358, 0.3948, 0.3241, 0.5654], device='cuda:1'), in_proj_covar=tensor([0.1755, 0.1647, 0.1602, 0.1424], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 11:37:36,458 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:37:37,358 INFO [optim.py:369] (1/2) 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 11:37:39,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:37:51,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:37:52,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:38:04,170 INFO [train.py:968] (1/2) Epoch 22, batch 23900, giga_loss[loss=0.371, simple_loss=0.4146, pruned_loss=0.1637, over 28276.00 frames. ], tot_loss[loss=0.3351, simple_loss=0.3909, pruned_loss=0.1396, over 5613579.55 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3568, pruned_loss=0.1063, over 5715106.36 frames. ], giga_tot_loss[loss=0.3366, simple_loss=0.3923, pruned_loss=0.1405, over 5616652.59 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:38:11,189 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:38:37,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:38:48,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7678, 2.1058, 1.3727, 1.6092], device='cuda:1'), covar=tensor([0.1012, 0.0608, 0.1102, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0449, 0.0521, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 11:38:56,311 INFO [train.py:968] (1/2) Epoch 22, batch 23950, libri_loss[loss=0.3012, simple_loss=0.3726, pruned_loss=0.1149, over 26095.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3896, pruned_loss=0.1395, over 5608688.18 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3563, pruned_loss=0.1063, over 5716232.15 frames. ], giga_tot_loss[loss=0.3369, simple_loss=0.3919, pruned_loss=0.141, over 5607993.92 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:39:03,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981686.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 11:39:07,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981689.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:39:23,166 INFO [optim.py:369] (1/2) 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 11:39:32,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=981716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:39:35,032 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981718.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:39:46,116 INFO [train.py:968] (1/2) Epoch 22, batch 24000, giga_loss[loss=0.3417, simple_loss=0.3954, pruned_loss=0.144, over 28861.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3866, pruned_loss=0.1376, over 5627165.44 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3566, pruned_loss=0.1065, over 5719215.27 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3887, pruned_loss=0.139, over 5622641.10 frames. ], batch size: 227, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:39:46,116 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 11:39:55,526 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 11:39:58,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6905, 1.7752, 1.7233, 1.5715], device='cuda:1'), covar=tensor([0.2906, 0.2852, 0.2355, 0.2650], device='cuda:1'), in_proj_covar=tensor([0.1980, 0.1921, 0.1833, 0.1982], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 11:40:18,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9651, 1.3424, 1.0988, 0.2167], device='cuda:1'), covar=tensor([0.3551, 0.2712, 0.3581, 0.4996], device='cuda:1'), in_proj_covar=tensor([0.1757, 0.1647, 0.1602, 0.1424], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 11:40:23,099 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:24,404 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:25,640 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:27,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:34,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-11 11:40:40,642 INFO [train.py:968] (1/2) Epoch 22, batch 24050, giga_loss[loss=0.3306, simple_loss=0.3822, pruned_loss=0.1395, over 28934.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.385, pruned_loss=0.1357, over 5627776.97 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.356, pruned_loss=0.1063, over 5722698.52 frames. ], giga_tot_loss[loss=0.3314, simple_loss=0.3877, pruned_loss=0.1376, over 5619578.68 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:40:44,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4320, 4.2482, 1.5631, 1.6614], device='cuda:1'), covar=tensor([0.1058, 0.0373, 0.0921, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0553, 0.0388, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 11:40:51,143 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:54,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:40:54,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3213, 1.2780, 1.2798, 1.5049], device='cuda:1'), covar=tensor([0.0683, 0.0456, 0.0330, 0.0737], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0098, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 11:41:00,933 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:41:08,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:12,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2223, 2.3119, 1.7752, 1.9116], device='cuda:1'), covar=tensor([0.1017, 0.0730, 0.1024, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0448, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 11:41:20,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-03-11 11:41:30,098 INFO [train.py:968] (1/2) Epoch 22, batch 24100, giga_loss[loss=0.2813, simple_loss=0.3506, pruned_loss=0.106, over 28805.00 frames. ], tot_loss[loss=0.327, simple_loss=0.3848, pruned_loss=0.1345, over 5621267.03 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3563, pruned_loss=0.1066, over 5728578.86 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3877, pruned_loss=0.1368, over 5606441.08 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:41:32,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4851, 1.7545, 1.4284, 1.3700], device='cuda:1'), covar=tensor([0.2344, 0.2319, 0.2512, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1502, 0.1086, 0.1327, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 11:41:32,398 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4888, 1.5618, 1.7158, 1.3032], device='cuda:1'), covar=tensor([0.1409, 0.2216, 0.1154, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0696, 0.0941, 0.0840], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 11:41:35,124 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:35,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9359, 1.2758, 1.3110, 1.0587], device='cuda:1'), covar=tensor([0.1792, 0.1276, 0.2055, 0.1520], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0758, 0.0722, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 11:41:57,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=981859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:41:59,731 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=981862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:42:00,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-11 11:42:19,762 INFO [train.py:968] (1/2) Epoch 22, batch 24150, giga_loss[loss=0.3531, simple_loss=0.4056, pruned_loss=0.1503, over 28775.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3874, pruned_loss=0.1359, over 5629461.01 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3562, pruned_loss=0.1065, over 5732269.54 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3903, pruned_loss=0.1382, over 5612583.58 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:42:32,008 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=981891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:42:48,666 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 24200, giga_loss[loss=0.3491, simple_loss=0.3985, pruned_loss=0.1499, over 27909.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.386, pruned_loss=0.1344, over 5626288.72 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3562, pruned_loss=0.1065, over 5733254.19 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3884, pruned_loss=0.1363, over 5611829.38 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:44:05,306 INFO [train.py:968] (1/2) Epoch 22, batch 24250, giga_loss[loss=0.2936, simple_loss=0.365, pruned_loss=0.1111, over 28794.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3835, pruned_loss=0.1311, over 5636723.78 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3558, pruned_loss=0.1063, over 5735235.64 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.386, pruned_loss=0.133, over 5622750.46 frames. ], batch size: 284, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:44:29,100 INFO [optim.py:369] (1/2) 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 11:44:41,588 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:44:51,759 INFO [train.py:968] (1/2) Epoch 22, batch 24300, giga_loss[loss=0.3291, simple_loss=0.393, pruned_loss=0.1326, over 28560.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3807, pruned_loss=0.1288, over 5636394.88 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3563, pruned_loss=0.1069, over 5740322.59 frames. ], giga_tot_loss[loss=0.3225, simple_loss=0.3834, pruned_loss=0.1308, over 5616453.04 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:45:37,183 INFO [train.py:968] (1/2) Epoch 22, batch 24350, giga_loss[loss=0.3277, simple_loss=0.3848, pruned_loss=0.1352, over 28628.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3774, pruned_loss=0.1263, over 5642278.87 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3567, pruned_loss=0.1074, over 5740471.47 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3798, pruned_loss=0.1279, over 5623970.30 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:46:01,094 INFO [optim.py:369] (1/2) 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 11:46:06,086 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:46:09,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6754, 1.0273, 2.8689, 2.7583], device='cuda:1'), covar=tensor([0.2131, 0.2917, 0.0976, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0650, 0.0966, 0.0912], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 11:46:20,854 INFO [train.py:968] (1/2) Epoch 22, batch 24400, giga_loss[loss=0.2959, simple_loss=0.3532, pruned_loss=0.1193, over 28694.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3756, pruned_loss=0.1254, over 5646214.83 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3573, pruned_loss=0.1082, over 5735850.59 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3777, pruned_loss=0.1267, over 5630746.87 frames. ], batch size: 92, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:46:24,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3644, 1.5697, 1.5002, 1.4389], device='cuda:1'), covar=tensor([0.1696, 0.1868, 0.1965, 0.1756], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0752, 0.0716, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 11:46:59,255 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 24450, giga_loss[loss=0.2975, simple_loss=0.3662, pruned_loss=0.1144, over 28723.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.377, pruned_loss=0.1271, over 5640289.53 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3571, pruned_loss=0.1082, over 5736826.33 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3788, pruned_loss=0.1283, over 5626955.24 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:47:13,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3106, 1.5013, 1.3342, 1.4935], device='cuda:1'), covar=tensor([0.0791, 0.0350, 0.0342, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 11:47:41,203 INFO [optim.py:369] (1/2) 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 11:48:04,220 INFO [train.py:968] (1/2) Epoch 22, batch 24500, libri_loss[loss=0.3115, simple_loss=0.386, pruned_loss=0.1185, over 29675.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3754, pruned_loss=0.1256, over 5649578.59 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.357, pruned_loss=0.1083, over 5740927.88 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3774, pruned_loss=0.1268, over 5633333.54 frames. ], batch size: 88, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:48:18,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5024, 1.6303, 1.7552, 1.2926], device='cuda:1'), covar=tensor([0.1797, 0.2657, 0.1499, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0892, 0.0697, 0.0941, 0.0839], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 11:48:26,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2605, 1.1839, 3.9387, 3.2665], device='cuda:1'), covar=tensor([0.1725, 0.2979, 0.0463, 0.1551], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0651, 0.0968, 0.0913], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 11:48:56,572 INFO [train.py:968] (1/2) Epoch 22, batch 24550, giga_loss[loss=0.3528, simple_loss=0.4195, pruned_loss=0.1431, over 28126.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3729, pruned_loss=0.1221, over 5654060.64 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3572, pruned_loss=0.1084, over 5733499.45 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3745, pruned_loss=0.1232, over 5646532.27 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:49:26,222 INFO [optim.py:369] (1/2) 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 11:49:47,046 INFO [train.py:968] (1/2) Epoch 22, batch 24600, giga_loss[loss=0.3746, simple_loss=0.4198, pruned_loss=0.1646, over 28746.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3737, pruned_loss=0.1202, over 5667758.16 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3572, pruned_loss=0.1085, over 5738146.25 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3754, pruned_loss=0.1213, over 5655237.30 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:50:35,798 INFO [train.py:968] (1/2) Epoch 22, batch 24650, giga_loss[loss=0.2882, simple_loss=0.361, pruned_loss=0.1077, over 29015.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3745, pruned_loss=0.1206, over 5645022.20 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3574, pruned_loss=0.1087, over 5720072.17 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3763, pruned_loss=0.1217, over 5648564.92 frames. ], batch size: 155, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:50:49,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=982395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:51:00,724 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 24700, giga_loss[loss=0.3124, simple_loss=0.3805, pruned_loss=0.1221, over 28630.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3753, pruned_loss=0.1214, over 5639250.16 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.358, pruned_loss=0.1092, over 5701519.16 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3769, pruned_loss=0.1223, over 5656287.69 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:51:48,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 11:52:10,764 INFO [train.py:968] (1/2) Epoch 22, batch 24750, giga_loss[loss=0.3841, simple_loss=0.4248, pruned_loss=0.1717, over 27505.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3741, pruned_loss=0.1207, over 5663148.50 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.358, pruned_loss=0.1093, over 5707731.05 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3759, pruned_loss=0.1217, over 5669841.50 frames. ], batch size: 472, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:52:13,969 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-11 11:52:17,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=982488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:52:24,163 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.35 vs. limit=5.0
+2023-03-11 11:52:34,594 INFO [optim.py:369] (1/2) 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 11:52:54,531 INFO [train.py:968] (1/2) Epoch 22, batch 24800, giga_loss[loss=0.2788, simple_loss=0.3511, pruned_loss=0.1032, over 29094.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3722, pruned_loss=0.1206, over 5663168.45 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3581, pruned_loss=0.1095, over 5704891.99 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3741, pruned_loss=0.1216, over 5668991.35 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:53:02,273 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=982538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:05,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=982541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:08,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=982544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:30,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=982570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:53:35,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.26 vs. limit=5.0
+2023-03-11 11:53:39,125 INFO [train.py:968] (1/2) Epoch 22, batch 24850, giga_loss[loss=0.2879, simple_loss=0.3532, pruned_loss=0.1113, over 28793.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3722, pruned_loss=0.1216, over 5653690.56 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3582, pruned_loss=0.1096, over 5694410.19 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3737, pruned_loss=0.1224, over 5667118.15 frames. ], batch size: 119, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:54:05,908 INFO [optim.py:369] (1/2) 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 11:54:23,632 INFO [train.py:968] (1/2) Epoch 22, batch 24900, giga_loss[loss=0.3052, simple_loss=0.3778, pruned_loss=0.1163, over 28692.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3721, pruned_loss=0.1202, over 5667265.27 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3584, pruned_loss=0.1098, over 5697524.47 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3733, pruned_loss=0.1208, over 5674480.35 frames. ], batch size: 262, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:54:24,022 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=982631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:54:26,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=982634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:54:56,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=982663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:55:12,062 INFO [train.py:968] (1/2) Epoch 22, batch 24950, giga_loss[loss=0.3726, simple_loss=0.4008, pruned_loss=0.1722, over 23609.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3723, pruned_loss=0.1194, over 5674006.60 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3586, pruned_loss=0.11, over 5700251.82 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3733, pruned_loss=0.1198, over 5676739.20 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:55:12,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2320, 3.0702, 1.4011, 1.3500], device='cuda:1'), covar=tensor([0.1079, 0.0458, 0.0967, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0557, 0.0389, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 11:55:19,257 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=982687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:55:20,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2443, 1.5469, 1.1555, 0.7006], device='cuda:1'), covar=tensor([0.2980, 0.2020, 0.2396, 0.5132], device='cuda:1'), in_proj_covar=tensor([0.1763, 0.1649, 0.1604, 0.1427], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 11:55:21,465 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:55:47,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=982719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:55:55,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1953, 1.2309, 3.6874, 3.2690], device='cuda:1'), covar=tensor([0.1682, 0.2744, 0.0517, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0654, 0.0974, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 11:55:58,322 INFO [train.py:968] (1/2) Epoch 22, batch 25000, giga_loss[loss=0.3228, simple_loss=0.3891, pruned_loss=0.1282, over 28624.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3727, pruned_loss=0.1199, over 5654562.91 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3591, pruned_loss=0.1105, over 5684248.12 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3733, pruned_loss=0.12, over 5671370.41 frames. ], batch size: 307, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:56:03,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3976, 2.6744, 1.5770, 1.5602], device='cuda:1'), covar=tensor([0.0862, 0.0378, 0.0758, 0.1163], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0557, 0.0390, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 11:56:10,673 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:56:46,003 INFO [train.py:968] (1/2) Epoch 22, batch 25050, giga_loss[loss=0.248, simple_loss=0.3293, pruned_loss=0.08333, over 28875.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3707, pruned_loss=0.1187, over 5668809.66 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3594, pruned_loss=0.1109, over 5688589.66 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3711, pruned_loss=0.1186, over 5677787.09 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:57:12,401 INFO [optim.py:369] (1/2) 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 11:57:31,420 INFO [train.py:968] (1/2) Epoch 22, batch 25100, giga_loss[loss=0.2934, simple_loss=0.3496, pruned_loss=0.1186, over 28786.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3706, pruned_loss=0.1198, over 5640166.56 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.36, pruned_loss=0.1115, over 5670688.17 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3708, pruned_loss=0.1194, over 5663084.36 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:57:52,700 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 11:58:22,879 INFO [train.py:968] (1/2) Epoch 22, batch 25150, giga_loss[loss=0.3549, simple_loss=0.3781, pruned_loss=0.1659, over 23709.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3712, pruned_loss=0.1212, over 5642610.87 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3602, pruned_loss=0.1116, over 5672085.72 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3712, pruned_loss=0.1209, over 5659259.87 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 11:58:28,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 11:58:38,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4126, 1.5379, 1.3323, 1.5782], device='cuda:1'), covar=tensor([0.0765, 0.0337, 0.0321, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 11:58:47,642 INFO [optim.py:369] (1/2) 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 11:59:05,645 INFO [train.py:968] (1/2) Epoch 22, batch 25200, giga_loss[loss=0.2908, simple_loss=0.3569, pruned_loss=0.1123, over 28377.00 frames. ], tot_loss[loss=0.306, simple_loss=0.37, pruned_loss=0.1209, over 5657186.36 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3603, pruned_loss=0.1117, over 5679725.49 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.121, over 5662678.21 frames. ], batch size: 368, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:59:07,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2534, 1.1182, 3.5570, 3.1749], device='cuda:1'), covar=tensor([0.1552, 0.2838, 0.0461, 0.1776], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0655, 0.0976, 0.0920], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 11:59:29,648 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=982954.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 11:59:53,223 INFO [train.py:968] (1/2) Epoch 22, batch 25250, giga_loss[loss=0.2622, simple_loss=0.3349, pruned_loss=0.09478, over 28519.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3677, pruned_loss=0.1198, over 5665643.57 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3599, pruned_loss=0.1115, over 5683005.94 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1201, over 5666506.04 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 11:59:56,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2239, 1.4446, 1.4359, 1.2795], device='cuda:1'), covar=tensor([0.1749, 0.1702, 0.2321, 0.1818], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0753, 0.0716, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 12:00:18,892 INFO [optim.py:369] (1/2) 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 12:00:40,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7490, 1.9833, 1.9107, 1.7018], device='cuda:1'), covar=tensor([0.2054, 0.1734, 0.1518, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.1985, 0.1923, 0.1836, 0.1987], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 12:00:40,917 INFO [train.py:968] (1/2) Epoch 22, batch 25300, libri_loss[loss=0.3325, simple_loss=0.3936, pruned_loss=0.1357, over 25827.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3677, pruned_loss=0.1202, over 5657300.32 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3604, pruned_loss=0.1119, over 5684933.16 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3681, pruned_loss=0.1204, over 5655812.35 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:01:30,980 INFO [train.py:968] (1/2) Epoch 22, batch 25350, giga_loss[loss=0.3795, simple_loss=0.401, pruned_loss=0.179, over 23597.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3672, pruned_loss=0.1196, over 5658750.85 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3605, pruned_loss=0.1122, over 5685163.41 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3675, pruned_loss=0.1196, over 5657161.84 frames. ], batch size: 705, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:01:57,446 INFO [optim.py:369] (1/2) 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 12:02:06,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:02:15,678 INFO [train.py:968] (1/2) Epoch 22, batch 25400, giga_loss[loss=0.3283, simple_loss=0.3936, pruned_loss=0.1315, over 28782.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3672, pruned_loss=0.1182, over 5664163.14 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3602, pruned_loss=0.1119, over 5686719.34 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3678, pruned_loss=0.1185, over 5661139.46 frames. ], batch size: 99, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:02:20,499 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-11 12:02:49,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3302, 1.7222, 1.3889, 1.0874], device='cuda:1'), covar=tensor([0.2658, 0.2656, 0.3051, 0.2301], device='cuda:1'), in_proj_covar=tensor([0.1507, 0.1090, 0.1329, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 12:03:04,215 INFO [train.py:968] (1/2) Epoch 22, batch 25450, libri_loss[loss=0.3141, simple_loss=0.3837, pruned_loss=0.1222, over 29266.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3678, pruned_loss=0.1182, over 5658082.26 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3607, pruned_loss=0.1122, over 5685920.80 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3679, pruned_loss=0.1182, over 5655969.99 frames. ], batch size: 94, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:03:14,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 12:03:21,488 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 12:03:34,222 INFO [optim.py:369] (1/2) 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 12:03:49,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:03:52,391 INFO [train.py:968] (1/2) Epoch 22, batch 25500, giga_loss[loss=0.2717, simple_loss=0.337, pruned_loss=0.1032, over 28304.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3683, pruned_loss=0.1187, over 5663543.98 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3609, pruned_loss=0.1124, over 5687026.19 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3682, pruned_loss=0.1186, over 5660739.77 frames. ], batch size: 77, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:04:22,189 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:04:25,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:04:40,435 INFO [train.py:968] (1/2) Epoch 22, batch 25550, libri_loss[loss=0.2973, simple_loss=0.3705, pruned_loss=0.1121, over 28642.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3725, pruned_loss=0.1229, over 5652700.54 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3608, pruned_loss=0.1124, over 5690195.87 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3727, pruned_loss=0.123, over 5647025.81 frames. ], batch size: 106, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:04:53,138 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:05:24,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:05:27,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983329.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 12:05:29,843 INFO [train.py:968] (1/2) Epoch 22, batch 25600, giga_loss[loss=0.2671, simple_loss=0.3408, pruned_loss=0.09673, over 28932.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3722, pruned_loss=0.1238, over 5659556.13 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3603, pruned_loss=0.1121, over 5694832.35 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.373, pruned_loss=0.1243, over 5650281.96 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:06:07,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:06:09,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:06:16,388 INFO [train.py:968] (1/2) Epoch 22, batch 25650, giga_loss[loss=0.2926, simple_loss=0.3551, pruned_loss=0.115, over 28904.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3732, pruned_loss=0.125, over 5673745.27 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3603, pruned_loss=0.1121, over 5701013.90 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3744, pruned_loss=0.1259, over 5659387.70 frames. ], batch size: 186, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:06:38,188 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:06:56,812 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 25700, giga_loss[loss=0.2692, simple_loss=0.3468, pruned_loss=0.0958, over 28871.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3747, pruned_loss=0.1265, over 5651905.61 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3605, pruned_loss=0.1123, over 5694305.23 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3757, pruned_loss=0.1273, over 5645609.68 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:07:40,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983472.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 12:07:43,505 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 25750, giga_loss[loss=0.3222, simple_loss=0.3928, pruned_loss=0.1258, over 28980.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3724, pruned_loss=0.1245, over 5665446.14 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3599, pruned_loss=0.1119, over 5699266.47 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.374, pruned_loss=0.1258, over 5655244.57 frames. ], batch size: 128, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:07:52,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 12:08:12,606 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983504.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 12:08:17,384 INFO [optim.py:369] (1/2) 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 12:08:33,865 INFO [train.py:968] (1/2) Epoch 22, batch 25800, giga_loss[loss=0.3112, simple_loss=0.3933, pruned_loss=0.1146, over 28896.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1245, over 5649220.64 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3603, pruned_loss=0.1123, over 5682119.18 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3745, pruned_loss=0.1254, over 5655505.53 frames. ], batch size: 174, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:09:00,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5975, 2.2035, 1.5448, 0.7776], device='cuda:1'), covar=tensor([0.5776, 0.3345, 0.4370, 0.6936], device='cuda:1'), in_proj_covar=tensor([0.1764, 0.1658, 0.1607, 0.1434], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 12:09:21,649 INFO [train.py:968] (1/2) Epoch 22, batch 25850, giga_loss[loss=0.3142, simple_loss=0.3796, pruned_loss=0.1244, over 28840.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1221, over 5654140.45 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3604, pruned_loss=0.1124, over 5683272.58 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3723, pruned_loss=0.1227, over 5657991.18 frames. ], batch size: 112, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:09:46,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8369, 3.6853, 3.4919, 1.9701], device='cuda:1'), covar=tensor([0.0712, 0.0787, 0.0804, 0.2083], device='cuda:1'), in_proj_covar=tensor([0.1244, 0.1150, 0.0973, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 12:09:49,196 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 25900, giga_loss[loss=0.3688, simple_loss=0.4089, pruned_loss=0.1643, over 27934.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3699, pruned_loss=0.1214, over 5654375.31 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3605, pruned_loss=0.1126, over 5686339.51 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3706, pruned_loss=0.1219, over 5654220.26 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:10:25,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8009, 4.8042, 1.8494, 2.2081], device='cuda:1'), covar=tensor([0.0891, 0.0359, 0.0848, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0560, 0.0390, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:10:53,594 INFO [train.py:968] (1/2) Epoch 22, batch 25950, giga_loss[loss=0.2661, simple_loss=0.3403, pruned_loss=0.09595, over 28868.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3677, pruned_loss=0.1203, over 5660169.33 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.361, pruned_loss=0.1128, over 5681005.46 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3681, pruned_loss=0.1207, over 5663420.83 frames. ], batch size: 145, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:11:11,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:11:23,914 INFO [optim.py:369] (1/2) 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 12:11:38,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3750, 1.6635, 1.4099, 1.6115], device='cuda:1'), covar=tensor([0.0807, 0.0332, 0.0335, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 12:11:40,956 INFO [train.py:968] (1/2) Epoch 22, batch 26000, giga_loss[loss=0.2903, simple_loss=0.3579, pruned_loss=0.1114, over 28473.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3663, pruned_loss=0.1195, over 5672794.62 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3608, pruned_loss=0.1127, over 5687370.10 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3671, pruned_loss=0.1203, over 5669206.10 frames. ], batch size: 78, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:12:10,119 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 12:12:13,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3428, 1.7026, 1.3359, 1.2513], device='cuda:1'), covar=tensor([0.2447, 0.2408, 0.2744, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.1505, 0.1088, 0.1327, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 12:12:15,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4341, 1.7638, 1.6997, 1.2381], device='cuda:1'), covar=tensor([0.1732, 0.2747, 0.1499, 0.1845], device='cuda:1'), in_proj_covar=tensor([0.0895, 0.0701, 0.0944, 0.0842], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 12:12:24,188 INFO [train.py:968] (1/2) Epoch 22, batch 26050, giga_loss[loss=0.3752, simple_loss=0.4154, pruned_loss=0.1674, over 27863.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3691, pruned_loss=0.1212, over 5656837.11 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.361, pruned_loss=0.113, over 5665698.78 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1218, over 5673951.14 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:12:38,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=983797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:12:51,849 INFO [optim.py:369] (1/2) 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 12:13:06,598 INFO [train.py:968] (1/2) Epoch 22, batch 26100, libri_loss[loss=0.2382, simple_loss=0.3066, pruned_loss=0.0849, over 29334.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3714, pruned_loss=0.1206, over 5666710.89 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3603, pruned_loss=0.1127, over 5674645.75 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3731, pruned_loss=0.1218, over 5671906.83 frames. ], batch size: 71, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:13:15,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-11 12:13:16,892 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:13:19,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:13:44,313 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:13:46,256 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 26150, giga_loss[loss=0.3006, simple_loss=0.3738, pruned_loss=0.1137, over 29003.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3719, pruned_loss=0.1191, over 5673150.34 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3602, pruned_loss=0.1128, over 5677296.40 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3741, pruned_loss=0.1204, over 5673925.79 frames. ], batch size: 213, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:14:18,012 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 26200, giga_loss[loss=0.2973, simple_loss=0.368, pruned_loss=0.1133, over 28952.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3736, pruned_loss=0.1205, over 5679609.38 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.36, pruned_loss=0.1128, over 5676131.57 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3756, pruned_loss=0.1216, over 5681214.77 frames. ], batch size: 164, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:14:46,878 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=983940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:14:51,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=983943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:14:54,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=983946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:15:16,033 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=983972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:15:22,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4007, 1.2713, 3.9886, 3.3684], device='cuda:1'), covar=tensor([0.1540, 0.2779, 0.0460, 0.1427], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0652, 0.0970, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 12:15:24,143 INFO [train.py:968] (1/2) Epoch 22, batch 26250, giga_loss[loss=0.3521, simple_loss=0.4064, pruned_loss=0.1489, over 28896.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3755, pruned_loss=0.1222, over 5674764.53 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5674166.93 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3772, pruned_loss=0.1232, over 5677737.75 frames. ], batch size: 199, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:15:54,935 INFO [optim.py:369] (1/2) 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 12:16:14,011 INFO [train.py:968] (1/2) Epoch 22, batch 26300, giga_loss[loss=0.2765, simple_loss=0.3523, pruned_loss=0.1003, over 29037.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3752, pruned_loss=0.123, over 5659730.31 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.1129, over 5664486.80 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3767, pruned_loss=0.1238, over 5669853.56 frames. ], batch size: 136, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:16:18,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8333, 1.9412, 1.7546, 1.7080], device='cuda:1'), covar=tensor([0.1876, 0.2411, 0.2285, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0748, 0.0715, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 12:17:03,423 INFO [train.py:968] (1/2) Epoch 22, batch 26350, giga_loss[loss=0.3483, simple_loss=0.4026, pruned_loss=0.147, over 27954.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.374, pruned_loss=0.1225, over 5673249.11 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1126, over 5667504.10 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3758, pruned_loss=0.1236, over 5678775.15 frames. ], batch size: 412, lr: 1.45e-03, grad_scale: 2.0
+2023-03-11 12:17:32,302 INFO [optim.py:369] (1/2) 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 12:17:48,981 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-11 12:17:50,021 INFO [train.py:968] (1/2) Epoch 22, batch 26400, giga_loss[loss=0.2663, simple_loss=0.3314, pruned_loss=0.1006, over 28501.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1226, over 5676361.10 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5668746.58 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5679728.45 frames. ], batch size: 85, lr: 1.45e-03, grad_scale: 4.0
+2023-03-11 12:18:23,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4562, 3.3383, 1.4996, 1.5022], device='cuda:1'), covar=tensor([0.0995, 0.0416, 0.0898, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0558, 0.0388, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:18:43,464 INFO [train.py:968] (1/2) Epoch 22, batch 26450, giga_loss[loss=0.3062, simple_loss=0.381, pruned_loss=0.1157, over 28847.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3718, pruned_loss=0.1227, over 5682943.76 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5668746.58 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.373, pruned_loss=0.1234, over 5685564.62 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:18:43,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0211, 1.2396, 3.3567, 2.8974], device='cuda:1'), covar=tensor([0.1717, 0.2700, 0.0518, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0766, 0.0653, 0.0970, 0.0915], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 12:18:58,890 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 12:19:00,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:19:15,515 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 26500, giga_loss[loss=0.2702, simple_loss=0.3359, pruned_loss=0.1023, over 28587.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3735, pruned_loss=0.1243, over 5677392.31 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5672260.10 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3741, pruned_loss=0.1246, over 5676157.04 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:19:46,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5263, 1.3412, 4.5943, 3.3365], device='cuda:1'), covar=tensor([0.1715, 0.2861, 0.0448, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0655, 0.0973, 0.0918], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 12:19:52,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:20:18,104 INFO [train.py:968] (1/2) Epoch 22, batch 26550, giga_loss[loss=0.2684, simple_loss=0.3398, pruned_loss=0.09851, over 28979.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3738, pruned_loss=0.1249, over 5680887.91 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.1131, over 5674550.77 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3742, pruned_loss=0.1252, over 5678001.72 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:20:46,383 INFO [optim.py:369] (1/2) 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 12:20:54,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:21:01,671 INFO [train.py:968] (1/2) Epoch 22, batch 26600, giga_loss[loss=0.397, simple_loss=0.4245, pruned_loss=0.1847, over 26681.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3723, pruned_loss=0.125, over 5653559.98 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3612, pruned_loss=0.1137, over 5668529.35 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3722, pruned_loss=0.125, over 5656999.01 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:21:11,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1402, 2.0723, 1.6067, 1.9507], device='cuda:1'), covar=tensor([0.0817, 0.0643, 0.0967, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0448, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 12:21:16,336 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 22, batch 26650, libri_loss[loss=0.2474, simple_loss=0.3216, pruned_loss=0.08662, over 29595.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3718, pruned_loss=0.1245, over 5657854.31 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3615, pruned_loss=0.1138, over 5676227.52 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3719, pruned_loss=0.1248, over 5652963.33 frames. ], batch size: 74, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:22:01,253 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:04,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=984399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:17,867 INFO [optim.py:369] (1/2) 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 12:22:22,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:32,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:22:34,835 INFO [train.py:968] (1/2) Epoch 22, batch 26700, giga_loss[loss=0.3077, simple_loss=0.3786, pruned_loss=0.1184, over 28976.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3713, pruned_loss=0.1227, over 5666406.33 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3613, pruned_loss=0.1137, over 5679056.92 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5659904.83 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:23:07,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:23:13,035 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 26750, libri_loss[loss=0.2726, simple_loss=0.3426, pruned_loss=0.1013, over 29545.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3731, pruned_loss=0.124, over 5664836.93 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3615, pruned_loss=0.1138, over 5685602.42 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3736, pruned_loss=0.1246, over 5652898.59 frames. ], batch size: 76, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:23:38,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:23:53,429 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:23:56,018 INFO [optim.py:369] (1/2) 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 12:24:10,050 INFO [train.py:968] (1/2) Epoch 22, batch 26800, libri_loss[loss=0.2982, simple_loss=0.3525, pruned_loss=0.122, over 29539.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3733, pruned_loss=0.1241, over 5673882.67 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.362, pruned_loss=0.1142, over 5686911.45 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1245, over 5662889.71 frames. ], batch size: 79, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 12:24:27,966 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:24:34,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5178, 1.7524, 1.4246, 1.4661], device='cuda:1'), covar=tensor([0.2838, 0.2802, 0.3316, 0.2379], device='cuda:1'), in_proj_covar=tensor([0.1505, 0.1090, 0.1330, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 12:24:47,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:24:55,020 INFO [train.py:968] (1/2) Epoch 22, batch 26850, giga_loss[loss=0.2776, simple_loss=0.352, pruned_loss=0.1016, over 28611.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3739, pruned_loss=0.1217, over 5676359.69 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3618, pruned_loss=0.1141, over 5688143.89 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3743, pruned_loss=0.1221, over 5666595.56 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:25:18,288 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:25:27,736 INFO [optim.py:369] (1/2) 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 12:25:28,016 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:25:42,939 INFO [train.py:968] (1/2) Epoch 22, batch 26900, giga_loss[loss=0.4233, simple_loss=0.4419, pruned_loss=0.2023, over 26621.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3764, pruned_loss=0.1217, over 5684647.55 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3619, pruned_loss=0.1142, over 5690634.58 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3768, pruned_loss=0.122, over 5674670.41 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:26:24,512 INFO [train.py:968] (1/2) Epoch 22, batch 26950, giga_loss[loss=0.302, simple_loss=0.3719, pruned_loss=0.1161, over 28893.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3803, pruned_loss=0.1243, over 5677763.69 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3623, pruned_loss=0.1145, over 5685858.68 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3806, pruned_loss=0.1245, over 5674211.12 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:26:58,126 INFO [optim.py:369] (1/2) 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 12:26:59,135 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:27:01,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=984718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:27:12,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5463, 1.5945, 1.4009, 1.1564], device='cuda:1'), covar=tensor([0.0969, 0.0599, 0.1020, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0450, 0.0520, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 12:27:15,998 INFO [train.py:968] (1/2) Epoch 22, batch 27000, giga_loss[loss=0.2888, simple_loss=0.3627, pruned_loss=0.1074, over 28935.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3824, pruned_loss=0.1272, over 5673504.15 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1144, over 5686984.38 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3828, pruned_loss=0.1274, over 5669734.65 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:27:15,999 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 12:27:24,930 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 12:27:34,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0307, 1.3484, 1.1055, 0.2502], device='cuda:1'), covar=tensor([0.3492, 0.3031, 0.4033, 0.6208], device='cuda:1'), in_proj_covar=tensor([0.1750, 0.1645, 0.1594, 0.1423], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 12:27:39,635 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:28:11,976 INFO [train.py:968] (1/2) Epoch 22, batch 27050, libri_loss[loss=0.2482, simple_loss=0.3163, pruned_loss=0.09002, over 29492.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3827, pruned_loss=0.1289, over 5654185.93 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5687313.28 frames. ], giga_tot_loss[loss=0.3221, simple_loss=0.3845, pruned_loss=0.1299, over 5650077.03 frames. ], batch size: 70, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:28:13,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-11 12:28:23,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:28:41,994 INFO [optim.py:369] (1/2) 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 12:28:56,252 INFO [train.py:968] (1/2) Epoch 22, batch 27100, giga_loss[loss=0.2975, simple_loss=0.3672, pruned_loss=0.1139, over 27997.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3798, pruned_loss=0.1269, over 5669068.13 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1142, over 5694383.78 frames. ], giga_tot_loss[loss=0.3194, simple_loss=0.3823, pruned_loss=0.1282, over 5658139.14 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:29:11,248 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:29:27,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9296, 1.2679, 0.9882, 0.2304], device='cuda:1'), covar=tensor([0.2779, 0.2173, 0.2883, 0.4600], device='cuda:1'), in_proj_covar=tensor([0.1757, 0.1649, 0.1600, 0.1428], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 12:29:44,816 INFO [train.py:968] (1/2) Epoch 22, batch 27150, libri_loss[loss=0.2833, simple_loss=0.3543, pruned_loss=0.1062, over 29254.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3791, pruned_loss=0.1261, over 5655007.03 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1142, over 5696089.64 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3816, pruned_loss=0.1275, over 5643602.50 frames. ], batch size: 94, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:29:48,004 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:30:25,690 INFO [train.py:968] (1/2) Epoch 22, batch 27200, giga_loss[loss=0.315, simple_loss=0.3883, pruned_loss=0.1208, over 29004.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3786, pruned_loss=0.1239, over 5672862.96 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.361, pruned_loss=0.1141, over 5702085.30 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3813, pruned_loss=0.1254, over 5657518.62 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 12:30:33,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=984935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:30:37,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=984938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:30:40,468 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=984942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:31:04,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=984967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:31:16,542 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 27250, libri_loss[loss=0.304, simple_loss=0.3679, pruned_loss=0.1201, over 29566.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3793, pruned_loss=0.1238, over 5650504.33 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5686040.48 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3811, pruned_loss=0.1246, over 5652979.39 frames. ], batch size: 89, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:31:22,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=984989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:31:50,602 INFO [optim.py:369] (1/2) 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 12:32:05,385 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:08,483 INFO [train.py:968] (1/2) Epoch 22, batch 27300, giga_loss[loss=0.3586, simple_loss=0.3949, pruned_loss=0.1612, over 23578.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3807, pruned_loss=0.1253, over 5657208.84 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1147, over 5690383.23 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3823, pruned_loss=0.126, over 5655004.99 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:32:09,334 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:22,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3387, 1.5304, 1.5424, 1.3128], device='cuda:1'), covar=tensor([0.3060, 0.2806, 0.1944, 0.2581], device='cuda:1'), in_proj_covar=tensor([0.1980, 0.1931, 0.1841, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 12:32:34,543 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:40,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:43,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:32:49,516 INFO [train.py:968] (1/2) Epoch 22, batch 27350, giga_loss[loss=0.3215, simple_loss=0.3821, pruned_loss=0.1305, over 28556.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3797, pruned_loss=0.1249, over 5661245.13 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3626, pruned_loss=0.1155, over 5685693.60 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3808, pruned_loss=0.1251, over 5662190.39 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:33:09,923 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:24,034 INFO [optim.py:369] (1/2) 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 12:33:30,989 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:34,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:38,543 INFO [train.py:968] (1/2) Epoch 22, batch 27400, giga_loss[loss=0.3656, simple_loss=0.4043, pruned_loss=0.1634, over 27933.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3765, pruned_loss=0.124, over 5656458.36 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3623, pruned_loss=0.1154, over 5689145.16 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3779, pruned_loss=0.1244, over 5653670.82 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:33:41,438 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:45,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:33:58,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3581, 1.4988, 1.4212, 1.2353], device='cuda:1'), covar=tensor([0.2694, 0.2555, 0.2053, 0.2568], device='cuda:1'), in_proj_covar=tensor([0.1983, 0.1933, 0.1848, 0.1991], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 12:34:02,179 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:34:12,660 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 27450, giga_loss[loss=0.3152, simple_loss=0.3779, pruned_loss=0.1263, over 27965.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3754, pruned_loss=0.1244, over 5646965.08 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1155, over 5693128.69 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3768, pruned_loss=0.1249, over 5640415.39 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:35:03,797 INFO [optim.py:369] (1/2) 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 12:35:08,826 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 27500, giga_loss[loss=0.3933, simple_loss=0.4193, pruned_loss=0.1836, over 26733.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3735, pruned_loss=0.1238, over 5644082.57 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3622, pruned_loss=0.1155, over 5684343.00 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5645445.87 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:36:03,811 INFO [train.py:968] (1/2) Epoch 22, batch 27550, giga_loss[loss=0.3061, simple_loss=0.3713, pruned_loss=0.1205, over 28723.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3727, pruned_loss=0.124, over 5635963.82 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3627, pruned_loss=0.1157, over 5677123.27 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3736, pruned_loss=0.1243, over 5642312.90 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:36:33,093 INFO [optim.py:369] (1/2) 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 12:36:34,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:36:45,589 INFO [train.py:968] (1/2) Epoch 22, batch 27600, libri_loss[loss=0.2912, simple_loss=0.3634, pruned_loss=0.1095, over 29526.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3713, pruned_loss=0.1227, over 5649343.58 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3629, pruned_loss=0.1155, over 5686171.41 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1234, over 5644539.05 frames. ], batch size: 84, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:37:10,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:37:12,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:37:14,322 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 27650, giga_loss[loss=0.3034, simple_loss=0.3721, pruned_loss=0.1174, over 28693.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 5645287.64 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3628, pruned_loss=0.1156, over 5673903.29 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.369, pruned_loss=0.1197, over 5651299.94 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:37:39,983 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:37:59,043 INFO [optim.py:369] (1/2) 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 12:38:12,444 INFO [train.py:968] (1/2) Epoch 22, batch 27700, giga_loss[loss=0.3791, simple_loss=0.4121, pruned_loss=0.1731, over 26573.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3651, pruned_loss=0.1164, over 5660291.95 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3624, pruned_loss=0.1155, over 5682792.36 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3664, pruned_loss=0.1172, over 5656012.53 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:38:37,317 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:38:38,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-11 12:38:40,628 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 27750, giga_loss[loss=0.32, simple_loss=0.3818, pruned_loss=0.1291, over 27838.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.366, pruned_loss=0.1174, over 5653522.39 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3629, pruned_loss=0.116, over 5684621.95 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3668, pruned_loss=0.1176, over 5648138.82 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:39:11,652 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=985492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:39:13,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6800, 4.9028, 1.9315, 1.8851], device='cuda:1'), covar=tensor([0.0977, 0.0253, 0.0845, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0559, 0.0389, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:39:34,290 INFO [optim.py:369] (1/2) 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 12:39:40,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:39:41,318 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:39:46,096 INFO [train.py:968] (1/2) Epoch 22, batch 27800, giga_loss[loss=0.2752, simple_loss=0.3432, pruned_loss=0.1036, over 28945.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3627, pruned_loss=0.1154, over 5661468.01 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3632, pruned_loss=0.1162, over 5688403.27 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5652005.28 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:40:37,329 INFO [train.py:968] (1/2) Epoch 22, batch 27850, giga_loss[loss=0.2765, simple_loss=0.3512, pruned_loss=0.1009, over 28728.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3622, pruned_loss=0.116, over 5654572.87 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3637, pruned_loss=0.1167, over 5686026.25 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.362, pruned_loss=0.1155, over 5648762.02 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 1.0
+2023-03-11 12:41:11,128 INFO [optim.py:369] (1/2) 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 12:41:20,108 INFO [train.py:968] (1/2) Epoch 22, batch 27900, giga_loss[loss=0.3371, simple_loss=0.3957, pruned_loss=0.1393, over 28972.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3644, pruned_loss=0.1172, over 5662868.38 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3634, pruned_loss=0.1168, over 5678225.42 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3645, pruned_loss=0.1167, over 5664611.89 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 1.0
+2023-03-11 12:42:04,239 INFO [train.py:968] (1/2) Epoch 22, batch 27950, giga_loss[loss=0.2975, simple_loss=0.3795, pruned_loss=0.1078, over 29005.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3659, pruned_loss=0.1178, over 5647348.09 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3632, pruned_loss=0.1165, over 5674549.45 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3663, pruned_loss=0.1177, over 5651071.58 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 1.0
+2023-03-11 12:42:09,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-11 12:42:39,689 INFO [optim.py:369] (1/2) 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 12:42:51,270 INFO [train.py:968] (1/2) Epoch 22, batch 28000, giga_loss[loss=0.2936, simple_loss=0.365, pruned_loss=0.1111, over 28854.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.1181, over 5651297.48 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3628, pruned_loss=0.1163, over 5676945.40 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3674, pruned_loss=0.1182, over 5651860.28 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:42:56,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:43:32,313 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 28050, giga_loss[loss=0.2367, simple_loss=0.3217, pruned_loss=0.07588, over 28346.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3681, pruned_loss=0.1193, over 5640236.98 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3634, pruned_loss=0.1166, over 5671429.38 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3682, pruned_loss=0.1191, over 5644531.13 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:43:42,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3086, 4.1704, 3.9662, 1.9716], device='cuda:1'), covar=tensor([0.0590, 0.0686, 0.0715, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1249, 0.1160, 0.0980, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 12:43:50,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-11 12:44:08,515 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 28100, giga_loss[loss=0.41, simple_loss=0.4384, pruned_loss=0.1908, over 27530.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3699, pruned_loss=0.1207, over 5657381.80 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3638, pruned_loss=0.1169, over 5676165.97 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3698, pruned_loss=0.1204, over 5656170.55 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:44:26,284 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=985837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:44:38,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1557, 2.5567, 1.2005, 1.2703], device='cuda:1'), covar=tensor([0.1086, 0.0425, 0.0955, 0.1524], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0558, 0.0388, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:45:04,335 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=985878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:04,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5739, 1.7091, 1.7836, 1.3659], device='cuda:1'), covar=tensor([0.1863, 0.2596, 0.1580, 0.1769], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0705, 0.0951, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 12:45:05,964 INFO [train.py:968] (1/2) Epoch 22, batch 28150, giga_loss[loss=0.278, simple_loss=0.3499, pruned_loss=0.103, over 28665.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3708, pruned_loss=0.1208, over 5656080.00 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3638, pruned_loss=0.1167, over 5673138.65 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3709, pruned_loss=0.1209, over 5657346.92 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:45:06,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=985881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:20,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:21,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=985899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:45:30,644 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 28200, giga_loss[loss=0.2797, simple_loss=0.3551, pruned_loss=0.1022, over 28943.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3728, pruned_loss=0.122, over 5646063.11 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3641, pruned_loss=0.117, over 5655458.60 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3728, pruned_loss=0.1219, over 5662039.01 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:45:58,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6709, 1.8839, 1.7979, 1.4854], device='cuda:1'), covar=tensor([0.3306, 0.2807, 0.2336, 0.2940], device='cuda:1'), in_proj_covar=tensor([0.1979, 0.1929, 0.1840, 0.1982], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 12:46:41,398 INFO [train.py:968] (1/2) Epoch 22, batch 28250, giga_loss[loss=0.2942, simple_loss=0.3602, pruned_loss=0.114, over 28492.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3745, pruned_loss=0.1239, over 5633167.83 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.364, pruned_loss=0.1169, over 5656070.59 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3747, pruned_loss=0.1239, over 5645037.42 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:47:15,023 INFO [optim.py:369] (1/2) 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 12:47:27,387 INFO [train.py:968] (1/2) Epoch 22, batch 28300, giga_loss[loss=0.2903, simple_loss=0.3618, pruned_loss=0.1094, over 28283.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.375, pruned_loss=0.1243, over 5628227.87 frames. ], libri_tot_loss[loss=0.2994, simple_loss=0.3644, pruned_loss=0.1172, over 5647895.84 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3752, pruned_loss=0.1244, over 5645518.43 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:47:39,367 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:47:41,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:47:42,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:47:44,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6880, 1.9306, 1.5382, 1.8642], device='cuda:1'), covar=tensor([0.2620, 0.2765, 0.3073, 0.2613], device='cuda:1'), in_proj_covar=tensor([0.1518, 0.1098, 0.1339, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 12:47:55,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2683, 2.6864, 1.4554, 1.4159], device='cuda:1'), covar=tensor([0.0951, 0.0410, 0.0887, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0559, 0.0388, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:48:10,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:48:12,579 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 28350, giga_loss[loss=0.3154, simple_loss=0.3848, pruned_loss=0.1231, over 28560.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3747, pruned_loss=0.1223, over 5641902.88 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3641, pruned_loss=0.117, over 5652554.73 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3754, pruned_loss=0.1227, over 5651267.76 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:48:37,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2637, 2.8819, 1.4378, 1.3971], device='cuda:1'), covar=tensor([0.0967, 0.0390, 0.0865, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0558, 0.0388, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:48:51,972 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:49:08,581 INFO [train.py:968] (1/2) Epoch 22, batch 28400, giga_loss[loss=0.2722, simple_loss=0.3464, pruned_loss=0.09898, over 28950.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3743, pruned_loss=0.1222, over 5648390.90 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3642, pruned_loss=0.1171, over 5652746.05 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3749, pruned_loss=0.1225, over 5655342.84 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:49:28,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:50:00,028 INFO [train.py:968] (1/2) Epoch 22, batch 28450, giga_loss[loss=0.2963, simple_loss=0.3541, pruned_loss=0.1193, over 28532.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3728, pruned_loss=0.122, over 5642952.08 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3642, pruned_loss=0.1172, over 5640680.98 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3734, pruned_loss=0.1223, over 5658798.99 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:50:36,839 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:50:58,457 INFO [train.py:968] (1/2) Epoch 22, batch 28500, giga_loss[loss=0.2885, simple_loss=0.3559, pruned_loss=0.1105, over 29010.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.372, pruned_loss=0.1219, over 5660580.66 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3642, pruned_loss=0.1171, over 5647116.55 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3726, pruned_loss=0.1222, over 5667659.26 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:51:11,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7150, 4.4573, 1.8431, 1.7745], device='cuda:1'), covar=tensor([0.0934, 0.0291, 0.0842, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0558, 0.0388, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:51:23,904 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:51:48,848 INFO [train.py:968] (1/2) Epoch 22, batch 28550, giga_loss[loss=0.3254, simple_loss=0.3641, pruned_loss=0.1433, over 23458.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3698, pruned_loss=0.1207, over 5654806.98 frames. ], libri_tot_loss[loss=0.2997, simple_loss=0.3646, pruned_loss=0.1173, over 5640823.46 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3701, pruned_loss=0.1209, over 5665565.57 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:52:00,531 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:52:03,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:52:23,937 INFO [optim.py:369] (1/2) 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 12:52:29,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:52:34,265 INFO [train.py:968] (1/2) Epoch 22, batch 28600, libri_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.1201, over 27549.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3695, pruned_loss=0.1209, over 5666543.04 frames. ], libri_tot_loss[loss=0.2992, simple_loss=0.3643, pruned_loss=0.1171, over 5644243.61 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3702, pruned_loss=0.1213, over 5672359.67 frames. ], batch size: 116, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:52:58,130 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:53:01,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:53:04,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-11 12:53:22,542 INFO [train.py:968] (1/2) Epoch 22, batch 28650, giga_loss[loss=0.3109, simple_loss=0.3714, pruned_loss=0.1252, over 28250.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3692, pruned_loss=0.1214, over 5652176.29 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.3639, pruned_loss=0.1168, over 5648879.75 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3702, pruned_loss=0.1221, over 5652794.54 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:53:27,335 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:53:31,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4509, 1.6441, 1.3777, 1.6427], device='cuda:1'), covar=tensor([0.0778, 0.0320, 0.0337, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 12:53:44,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4339, 1.7828, 1.3580, 1.6982], device='cuda:1'), covar=tensor([0.2749, 0.2751, 0.3189, 0.2308], device='cuda:1'), in_proj_covar=tensor([0.1517, 0.1096, 0.1338, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 12:53:54,030 INFO [optim.py:369] (1/2) 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 12:54:06,953 INFO [train.py:968] (1/2) Epoch 22, batch 28700, giga_loss[loss=0.3015, simple_loss=0.359, pruned_loss=0.122, over 28878.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3689, pruned_loss=0.1214, over 5655729.17 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3632, pruned_loss=0.1162, over 5657479.70 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3704, pruned_loss=0.1227, over 5648545.51 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:54:21,696 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 28750, giga_loss[loss=0.3169, simple_loss=0.3823, pruned_loss=0.1257, over 28750.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.1229, over 5651594.54 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1163, over 5650597.97 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.372, pruned_loss=0.1239, over 5651547.74 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:55:02,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:55:04,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6096, 3.3749, 1.5971, 1.8074], device='cuda:1'), covar=tensor([0.0919, 0.0346, 0.0872, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0559, 0.0388, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 12:55:06,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 12:55:08,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3819, 1.5669, 1.6332, 1.2171], device='cuda:1'), covar=tensor([0.1654, 0.2460, 0.1374, 0.1669], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0706, 0.0949, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 12:55:30,559 INFO [optim.py:369] (1/2) 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 12:55:42,218 INFO [train.py:968] (1/2) Epoch 22, batch 28800, giga_loss[loss=0.2998, simple_loss=0.3674, pruned_loss=0.1161, over 28722.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3715, pruned_loss=0.1237, over 5647061.37 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3633, pruned_loss=0.1161, over 5656291.17 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3728, pruned_loss=0.1249, over 5641663.03 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 12:56:30,968 INFO [train.py:968] (1/2) Epoch 22, batch 28850, giga_loss[loss=0.2701, simple_loss=0.343, pruned_loss=0.09862, over 29072.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3716, pruned_loss=0.1244, over 5649475.45 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3635, pruned_loss=0.1162, over 5659022.47 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3726, pruned_loss=0.1254, over 5642589.46 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:57:06,673 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:57:17,149 INFO [train.py:968] (1/2) Epoch 22, batch 28900, giga_loss[loss=0.3194, simple_loss=0.3843, pruned_loss=0.1272, over 28928.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.373, pruned_loss=0.1258, over 5644299.59 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3639, pruned_loss=0.1165, over 5650573.75 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3735, pruned_loss=0.1264, over 5645635.69 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:57:19,950 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:57:22,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:57:50,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:58:06,809 INFO [train.py:968] (1/2) Epoch 22, batch 28950, giga_loss[loss=0.3229, simple_loss=0.3818, pruned_loss=0.132, over 28452.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3726, pruned_loss=0.1248, over 5633276.21 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.364, pruned_loss=0.1165, over 5647243.70 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3731, pruned_loss=0.1255, over 5636336.03 frames. ], batch size: 78, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 12:58:43,805 INFO [optim.py:369] (1/2) 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 12:58:54,372 INFO [train.py:968] (1/2) Epoch 22, batch 29000, giga_loss[loss=0.2944, simple_loss=0.3599, pruned_loss=0.1144, over 28849.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 5639924.60 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3637, pruned_loss=0.1165, over 5642074.71 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3731, pruned_loss=0.1248, over 5646652.48 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:59:30,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 12:59:33,085 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 29050, giga_loss[loss=0.3188, simple_loss=0.3814, pruned_loss=0.1281, over 28937.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.125, over 5649453.55 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3637, pruned_loss=0.1164, over 5647019.70 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3747, pruned_loss=0.1257, over 5650452.31 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 12:59:55,976 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:00:13,062 INFO [optim.py:369] (1/2) 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 13:00:13,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=986821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:00:15,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9243, 1.2044, 1.3183, 1.0620], device='cuda:1'), covar=tensor([0.1799, 0.1324, 0.2176, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0759, 0.0724, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 13:00:22,162 INFO [train.py:968] (1/2) Epoch 22, batch 29100, giga_loss[loss=0.275, simple_loss=0.355, pruned_loss=0.0975, over 29002.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3743, pruned_loss=0.1253, over 5666600.61 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5650873.58 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3755, pruned_loss=0.1263, over 5664097.85 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:01:00,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 13:01:04,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:01:09,337 INFO [train.py:968] (1/2) Epoch 22, batch 29150, giga_loss[loss=0.3392, simple_loss=0.3953, pruned_loss=0.1415, over 28860.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3749, pruned_loss=0.126, over 5670055.99 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1161, over 5653391.68 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.376, pruned_loss=0.127, over 5665976.98 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:01:09,657 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986881.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:01:17,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2763, 0.8906, 0.9458, 1.3920], device='cuda:1'), covar=tensor([0.0772, 0.0385, 0.0366, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 13:01:37,233 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 13:01:41,563 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:01:43,262 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-11 13:01:47,785 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 29200, giga_loss[loss=0.3479, simple_loss=0.3942, pruned_loss=0.1509, over 26472.00 frames. ], tot_loss[loss=0.315, simple_loss=0.377, pruned_loss=0.1266, over 5667987.61 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3639, pruned_loss=0.1164, over 5658621.92 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3776, pruned_loss=0.1273, over 5660180.67 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:02:33,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=986964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:02:36,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=986967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:02:50,327 INFO [train.py:968] (1/2) Epoch 22, batch 29250, giga_loss[loss=0.2806, simple_loss=0.3555, pruned_loss=0.1028, over 29007.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3762, pruned_loss=0.1249, over 5661872.91 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3637, pruned_loss=0.1163, over 5661000.22 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.377, pruned_loss=0.1258, over 5653819.73 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:02:51,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2758, 0.8104, 0.8777, 1.3804], device='cuda:1'), covar=tensor([0.0799, 0.0397, 0.0374, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 13:02:52,833 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=986983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:05,160 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=986996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:09,997 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=987002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:23,045 INFO [optim.py:369] (1/2) 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 13:03:28,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=987026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:03:34,244 INFO [train.py:968] (1/2) Epoch 22, batch 29300, giga_loss[loss=0.3683, simple_loss=0.3857, pruned_loss=0.1755, over 23885.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3728, pruned_loss=0.1224, over 5674228.00 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3633, pruned_loss=0.1161, over 5668367.45 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3742, pruned_loss=0.1234, over 5661305.44 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:03:50,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5818, 1.3031, 4.3680, 3.4235], device='cuda:1'), covar=tensor([0.1626, 0.2879, 0.0442, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0654, 0.0977, 0.0920], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 13:04:20,008 INFO [train.py:968] (1/2) Epoch 22, batch 29350, giga_loss[loss=0.2897, simple_loss=0.3634, pruned_loss=0.108, over 28819.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1227, over 5665760.01 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1161, over 5671925.75 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3736, pruned_loss=0.1236, over 5652366.22 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:04:25,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1749, 1.2658, 3.2665, 2.9154], device='cuda:1'), covar=tensor([0.1610, 0.2696, 0.0539, 0.1467], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0655, 0.0978, 0.0921], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 13:04:31,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4174, 1.6216, 1.2576, 1.1607], device='cuda:1'), covar=tensor([0.1024, 0.0561, 0.1056, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0448, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 13:04:53,411 INFO [optim.py:369] (1/2) 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 13:05:02,172 INFO [train.py:968] (1/2) Epoch 22, batch 29400, giga_loss[loss=0.2807, simple_loss=0.357, pruned_loss=0.1021, over 28751.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3728, pruned_loss=0.1223, over 5677636.53 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3628, pruned_loss=0.1157, over 5678589.15 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3745, pruned_loss=0.1236, over 5660734.02 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:05:54,933 INFO [train.py:968] (1/2) Epoch 22, batch 29450, giga_loss[loss=0.3074, simple_loss=0.3645, pruned_loss=0.1252, over 28733.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3744, pruned_loss=0.1239, over 5668746.35 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3631, pruned_loss=0.1158, over 5681453.15 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3759, pruned_loss=0.1251, over 5652377.80 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:06:22,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9787, 1.3262, 1.1101, 0.2148], device='cuda:1'), covar=tensor([0.3954, 0.3080, 0.4544, 0.6395], device='cuda:1'), in_proj_covar=tensor([0.1771, 0.1666, 0.1608, 0.1441], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 13:06:31,931 INFO [optim.py:369] (1/2) 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 13:06:42,888 INFO [train.py:968] (1/2) Epoch 22, batch 29500, libri_loss[loss=0.3331, simple_loss=0.3904, pruned_loss=0.1379, over 29135.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.374, pruned_loss=0.1244, over 5668140.19 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3634, pruned_loss=0.116, over 5682781.39 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3752, pruned_loss=0.1253, over 5653492.46 frames. ], batch size: 101, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:06:58,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:07:03,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987256.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:07:24,970 INFO [train.py:968] (1/2) Epoch 22, batch 29550, giga_loss[loss=0.3362, simple_loss=0.4026, pruned_loss=0.1349, over 29043.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3748, pruned_loss=0.1252, over 5660568.64 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3637, pruned_loss=0.116, over 5686777.16 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3758, pruned_loss=0.1262, over 5644870.23 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:07:31,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:07:55,062 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-11 13:08:00,590 INFO [optim.py:369] (1/2) 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 13:08:11,614 INFO [train.py:968] (1/2) Epoch 22, batch 29600, giga_loss[loss=0.2982, simple_loss=0.3665, pruned_loss=0.115, over 28551.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.375, pruned_loss=0.125, over 5672968.30 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.364, pruned_loss=0.1163, over 5688350.51 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3757, pruned_loss=0.1258, over 5658756.04 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 13:08:37,261 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:08:38,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4244, 1.2961, 4.1292, 3.3650], device='cuda:1'), covar=tensor([0.1684, 0.2830, 0.0459, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0657, 0.0981, 0.0925], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 13:08:53,712 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 29650, giga_loss[loss=0.3089, simple_loss=0.3792, pruned_loss=0.1193, over 28899.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3766, pruned_loss=0.1263, over 5654013.14 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3645, pruned_loss=0.1166, over 5684494.52 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1269, over 5645371.23 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:09:10,683 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:13,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:16,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987399.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:09:18,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=987401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:19,337 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 13:09:36,641 INFO [optim.py:369] (1/2) 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 13:09:40,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:45,337 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:09:45,625 INFO [train.py:968] (1/2) Epoch 22, batch 29700, giga_loss[loss=0.2771, simple_loss=0.3559, pruned_loss=0.09914, over 28817.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3749, pruned_loss=0.1246, over 5671382.93 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3641, pruned_loss=0.1165, over 5688601.57 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3759, pruned_loss=0.1253, over 5660289.28 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:09:45,971 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987431.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:09:47,324 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 29750, giga_loss[loss=0.2823, simple_loss=0.3585, pruned_loss=0.103, over 28443.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3749, pruned_loss=0.1243, over 5672136.88 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3641, pruned_loss=0.1164, over 5693019.82 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1251, over 5658815.97 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:10:37,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3527, 1.5427, 1.5855, 1.2390], device='cuda:1'), covar=tensor([0.1234, 0.1848, 0.1072, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0704, 0.0948, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 13:10:50,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:10:50,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8004, 1.0232, 2.8304, 2.6767], device='cuda:1'), covar=tensor([0.1800, 0.2774, 0.0636, 0.1081], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0655, 0.0977, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 13:10:52,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:07,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5399, 1.7591, 1.3978, 1.6638], device='cuda:1'), covar=tensor([0.2555, 0.2701, 0.3003, 0.2230], device='cuda:1'), in_proj_covar=tensor([0.1520, 0.1100, 0.1340, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 13:11:09,547 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:09,894 INFO [optim.py:369] (1/2) 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 13:11:11,478 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:17,079 INFO [train.py:968] (1/2) Epoch 22, batch 29800, giga_loss[loss=0.28, simple_loss=0.3564, pruned_loss=0.1018, over 28845.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3751, pruned_loss=0.124, over 5672240.24 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3643, pruned_loss=0.1165, over 5696424.32 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.376, pruned_loss=0.1248, over 5658212.69 frames. ], batch size: 285, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:11:19,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:32,026 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=987544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:35,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=987547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:11:40,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:12:02,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=987576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:12:07,173 INFO [train.py:968] (1/2) Epoch 22, batch 29850, giga_loss[loss=0.3007, simple_loss=0.3666, pruned_loss=0.1174, over 28631.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3732, pruned_loss=0.123, over 5670690.71 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3641, pruned_loss=0.1164, over 5697266.20 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1237, over 5658900.55 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:12:42,943 INFO [optim.py:369] (1/2) 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 13:12:51,040 INFO [train.py:968] (1/2) Epoch 22, batch 29900, libri_loss[loss=0.309, simple_loss=0.3738, pruned_loss=0.1221, over 28568.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3716, pruned_loss=0.1224, over 5661918.76 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3643, pruned_loss=0.1165, over 5694688.86 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3725, pruned_loss=0.1231, over 5653871.44 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:13:35,088 INFO [train.py:968] (1/2) Epoch 22, batch 29950, giga_loss[loss=0.3286, simple_loss=0.3882, pruned_loss=0.1345, over 28316.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3691, pruned_loss=0.1209, over 5664325.96 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3646, pruned_loss=0.1167, over 5692466.42 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1215, over 5659201.00 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:14:13,214 INFO [optim.py:369] (1/2) 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 13:14:20,328 INFO [train.py:968] (1/2) Epoch 22, batch 30000, giga_loss[loss=0.3052, simple_loss=0.3671, pruned_loss=0.1216, over 28597.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3651, pruned_loss=0.1186, over 5674638.70 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3646, pruned_loss=0.1166, over 5695910.82 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3656, pruned_loss=0.1192, over 5666226.08 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:14:20,328 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 13:14:25,961 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2264, 1.3183, 3.2802, 3.0696], device='cuda:1'), covar=tensor([0.1812, 0.3080, 0.0618, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0659, 0.0981, 0.0925], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 13:14:29,028 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 13:14:29,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5480, 1.6194, 1.7676, 1.3715], device='cuda:1'), covar=tensor([0.1656, 0.2554, 0.1368, 0.1718], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0706, 0.0949, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 13:14:55,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-11 13:14:58,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2543, 2.9856, 1.3645, 1.3873], device='cuda:1'), covar=tensor([0.1007, 0.0391, 0.0893, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0561, 0.0389, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 13:15:13,776 INFO [train.py:968] (1/2) Epoch 22, batch 30050, giga_loss[loss=0.2656, simple_loss=0.3333, pruned_loss=0.09892, over 28494.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3628, pruned_loss=0.1177, over 5688838.16 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3643, pruned_loss=0.1164, over 5699140.87 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3635, pruned_loss=0.1184, over 5679271.36 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:15:43,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8995, 1.8955, 2.0490, 1.6414], device='cuda:1'), covar=tensor([0.1880, 0.2519, 0.1462, 0.1735], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0705, 0.0948, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 13:15:52,440 INFO [optim.py:369] (1/2) 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 13:15:59,684 INFO [train.py:968] (1/2) Epoch 22, batch 30100, giga_loss[loss=0.2572, simple_loss=0.3346, pruned_loss=0.08993, over 28800.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3616, pruned_loss=0.1169, over 5689365.70 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3641, pruned_loss=0.1162, over 5694086.14 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3623, pruned_loss=0.1177, over 5686868.63 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:16:26,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6716, 1.8125, 1.3719, 1.4222], device='cuda:1'), covar=tensor([0.0968, 0.0591, 0.0961, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0448, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 13:16:45,423 INFO [train.py:968] (1/2) Epoch 22, batch 30150, giga_loss[loss=0.2938, simple_loss=0.3709, pruned_loss=0.1083, over 28637.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3626, pruned_loss=0.1163, over 5680565.69 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3641, pruned_loss=0.1162, over 5689865.20 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3631, pruned_loss=0.1169, over 5681918.63 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:17:25,149 INFO [optim.py:369] (1/2) 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 13:17:34,121 INFO [train.py:968] (1/2) Epoch 22, batch 30200, giga_loss[loss=0.2689, simple_loss=0.3511, pruned_loss=0.0933, over 28766.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3603, pruned_loss=0.1127, over 5685068.98 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3639, pruned_loss=0.1162, over 5697495.72 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5678801.96 frames. ], batch size: 243, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:18:22,822 INFO [train.py:968] (1/2) Epoch 22, batch 30250, giga_loss[loss=0.2528, simple_loss=0.3361, pruned_loss=0.08473, over 28880.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3582, pruned_loss=0.1108, over 5659586.63 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3638, pruned_loss=0.1165, over 5688776.05 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.3586, pruned_loss=0.1108, over 5662184.29 frames. ], batch size: 285, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:19:00,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-11 13:19:02,711 INFO [optim.py:369] (1/2) 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 13:19:13,376 INFO [train.py:968] (1/2) Epoch 22, batch 30300, giga_loss[loss=0.259, simple_loss=0.3398, pruned_loss=0.08911, over 28520.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3547, pruned_loss=0.1074, over 5635784.37 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3637, pruned_loss=0.1166, over 5672607.26 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.355, pruned_loss=0.1072, over 5651050.02 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:20:01,097 INFO [train.py:968] (1/2) Epoch 22, batch 30350, giga_loss[loss=0.2522, simple_loss=0.3395, pruned_loss=0.08244, over 28649.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3515, pruned_loss=0.1042, over 5641311.13 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3631, pruned_loss=0.1163, over 5673128.88 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.352, pruned_loss=0.104, over 5652424.43 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:20:39,238 INFO [optim.py:369] (1/2) 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 13:20:49,432 INFO [train.py:968] (1/2) Epoch 22, batch 30400, giga_loss[loss=0.2584, simple_loss=0.3359, pruned_loss=0.09047, over 28562.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3496, pruned_loss=0.1016, over 5642015.12 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3619, pruned_loss=0.1159, over 5679094.93 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3508, pruned_loss=0.1016, over 5644778.61 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 13:21:00,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5429, 2.2356, 1.6389, 0.7057], device='cuda:1'), covar=tensor([0.6282, 0.3021, 0.3887, 0.6752], device='cuda:1'), in_proj_covar=tensor([0.1770, 0.1661, 0.1605, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 13:21:31,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6347, 2.5996, 2.5419, 2.2218], device='cuda:1'), covar=tensor([0.1508, 0.1829, 0.1576, 0.1771], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0748, 0.0713, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 13:21:41,542 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=988179.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 13:21:43,237 INFO [train.py:968] (1/2) Epoch 22, batch 30450, giga_loss[loss=0.3474, simple_loss=0.4032, pruned_loss=0.1458, over 27958.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3506, pruned_loss=0.1019, over 5641466.53 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3619, pruned_loss=0.116, over 5682431.57 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3514, pruned_loss=0.1016, over 5640317.26 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:22:29,003 INFO [optim.py:369] (1/2) 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 13:22:35,512 INFO [train.py:968] (1/2) Epoch 22, batch 30500, giga_loss[loss=0.2387, simple_loss=0.3214, pruned_loss=0.07795, over 28980.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1008, over 5641885.83 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3617, pruned_loss=0.116, over 5684697.46 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3497, pruned_loss=0.1004, over 5638480.43 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:23:13,516 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 13:23:29,406 INFO [train.py:968] (1/2) Epoch 22, batch 30550, giga_loss[loss=0.2614, simple_loss=0.3419, pruned_loss=0.09042, over 28724.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3465, pruned_loss=0.09888, over 5638390.13 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3617, pruned_loss=0.116, over 5685497.97 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.347, pruned_loss=0.09849, over 5634802.75 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:23:48,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8398, 1.2713, 1.4021, 1.0680], device='cuda:1'), covar=tensor([0.2122, 0.1316, 0.2045, 0.1685], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0746, 0.0712, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 13:24:04,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 13:24:05,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2229, 2.3784, 1.8352, 2.1497], device='cuda:1'), covar=tensor([0.0802, 0.0552, 0.0868, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0445, 0.0516, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 13:24:09,136 INFO [optim.py:369] (1/2) 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 13:24:15,825 INFO [train.py:968] (1/2) Epoch 22, batch 30600, libri_loss[loss=0.2684, simple_loss=0.3361, pruned_loss=0.1004, over 29526.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3453, pruned_loss=0.09878, over 5636211.20 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.361, pruned_loss=0.1159, over 5681343.21 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3456, pruned_loss=0.0977, over 5634723.63 frames. ], batch size: 81, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:25:00,040 INFO [train.py:968] (1/2) Epoch 22, batch 30650, giga_loss[loss=0.2476, simple_loss=0.3331, pruned_loss=0.08108, over 28777.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3452, pruned_loss=0.0985, over 5648701.13 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3601, pruned_loss=0.1156, over 5689273.49 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3456, pruned_loss=0.09716, over 5638527.17 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:25:21,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9046, 2.1098, 2.1119, 1.7072], device='cuda:1'), covar=tensor([0.1272, 0.1990, 0.1140, 0.1454], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0699, 0.0944, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 13:25:33,858 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=988419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:25:37,651 INFO [optim.py:369] (1/2) 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 13:25:46,130 INFO [train.py:968] (1/2) Epoch 22, batch 30700, giga_loss[loss=0.2546, simple_loss=0.3325, pruned_loss=0.08833, over 28959.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3442, pruned_loss=0.09773, over 5634409.91 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.36, pruned_loss=0.1158, over 5674398.32 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3441, pruned_loss=0.09591, over 5638736.41 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:26:37,215 INFO [train.py:968] (1/2) Epoch 22, batch 30750, giga_loss[loss=0.235, simple_loss=0.3227, pruned_loss=0.07368, over 29067.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3413, pruned_loss=0.09493, over 5642176.86 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3596, pruned_loss=0.1155, over 5676160.40 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3413, pruned_loss=0.09342, over 5643723.79 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:27:20,212 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-11 13:27:20,415 INFO [optim.py:369] (1/2) 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 13:27:25,656 INFO [train.py:968] (1/2) Epoch 22, batch 30800, libri_loss[loss=0.3058, simple_loss=0.3599, pruned_loss=0.1259, over 20018.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3386, pruned_loss=0.09336, over 5633470.11 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3591, pruned_loss=0.1154, over 5672605.23 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3384, pruned_loss=0.09159, over 5637356.21 frames. ], batch size: 187, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:27:46,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=988554.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 13:28:08,119 INFO [train.py:968] (1/2) Epoch 22, batch 30850, giga_loss[loss=0.2376, simple_loss=0.3183, pruned_loss=0.07847, over 28077.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.336, pruned_loss=0.09263, over 5646436.35 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3577, pruned_loss=0.1147, over 5681768.49 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3359, pruned_loss=0.09045, over 5639148.43 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:28:47,995 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 30900, giga_loss[loss=0.2773, simple_loss=0.3444, pruned_loss=0.1051, over 29047.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3352, pruned_loss=0.09271, over 5653065.54 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3571, pruned_loss=0.1145, over 5685836.92 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3352, pruned_loss=0.09063, over 5643067.72 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:29:47,616 INFO [train.py:968] (1/2) Epoch 22, batch 30950, giga_loss[loss=0.3123, simple_loss=0.3728, pruned_loss=0.126, over 28296.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3362, pruned_loss=0.09373, over 5617041.74 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3568, pruned_loss=0.1145, over 5667553.35 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.336, pruned_loss=0.09161, over 5625170.37 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:30:04,108 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=988697.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 13:30:07,092 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=988700.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 13:30:35,431 INFO [optim.py:369] (1/2) 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 13:30:44,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=988729.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 13:30:45,413 INFO [train.py:968] (1/2) Epoch 22, batch 31000, giga_loss[loss=0.2618, simple_loss=0.3508, pruned_loss=0.08646, over 28533.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3395, pruned_loss=0.09525, over 5612494.29 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3565, pruned_loss=0.1146, over 5660934.78 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.339, pruned_loss=0.09287, over 5623696.61 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:31:23,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-11 13:31:29,175 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 13:31:40,531 INFO [train.py:968] (1/2) Epoch 22, batch 31050, giga_loss[loss=0.254, simple_loss=0.3388, pruned_loss=0.08462, over 28629.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3412, pruned_loss=0.09457, over 5631093.63 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3566, pruned_loss=0.1147, over 5662061.57 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3405, pruned_loss=0.09235, over 5638556.46 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:31:59,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=988794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:32:40,877 INFO [optim.py:369] (1/2) 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 13:32:49,878 INFO [train.py:968] (1/2) Epoch 22, batch 31100, giga_loss[loss=0.2742, simple_loss=0.3487, pruned_loss=0.09986, over 28432.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09381, over 5654343.51 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3561, pruned_loss=0.1144, over 5664347.65 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3406, pruned_loss=0.09211, over 5657991.03 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:33:02,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-11 13:33:38,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1572, 2.3314, 1.6583, 1.9213], device='cuda:1'), covar=tensor([0.1011, 0.0642, 0.0999, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0443, 0.0514, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 13:33:47,516 INFO [train.py:968] (1/2) Epoch 22, batch 31150, giga_loss[loss=0.2576, simple_loss=0.3383, pruned_loss=0.08846, over 28703.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3385, pruned_loss=0.09248, over 5633739.24 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3559, pruned_loss=0.1143, over 5651680.09 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.338, pruned_loss=0.09064, over 5647917.41 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:34:46,820 INFO [optim.py:369] (1/2) 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 13:34:54,456 INFO [train.py:968] (1/2) Epoch 22, batch 31200, giga_loss[loss=0.2289, simple_loss=0.3193, pruned_loss=0.06925, over 28962.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3377, pruned_loss=0.09075, over 5644810.99 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3559, pruned_loss=0.1143, over 5655756.19 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.337, pruned_loss=0.08893, over 5652160.18 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:34:59,794 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=988937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:35:02,280 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=988940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:35:38,489 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 31250, giga_loss[loss=0.2591, simple_loss=0.3284, pruned_loss=0.09491, over 29002.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09027, over 5655265.08 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3558, pruned_loss=0.1143, over 5660761.28 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3354, pruned_loss=0.08846, over 5656716.45 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:36:47,238 INFO [optim.py:369] (1/2) 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 13:36:56,424 INFO [train.py:968] (1/2) Epoch 22, batch 31300, giga_loss[loss=0.2611, simple_loss=0.3307, pruned_loss=0.09577, over 28590.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3344, pruned_loss=0.09027, over 5656294.15 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.356, pruned_loss=0.1145, over 5664642.40 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3332, pruned_loss=0.08805, over 5654009.46 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:37:55,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7699, 4.5640, 4.3835, 2.0951], device='cuda:1'), covar=tensor([0.0507, 0.0631, 0.0787, 0.2018], device='cuda:1'), in_proj_covar=tensor([0.1233, 0.1142, 0.0960, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 13:37:56,316 INFO [train.py:968] (1/2) Epoch 22, batch 31350, giga_loss[loss=0.2699, simple_loss=0.3546, pruned_loss=0.09259, over 28530.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3334, pruned_loss=0.08978, over 5661070.18 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3558, pruned_loss=0.1144, over 5665644.04 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3323, pruned_loss=0.08779, over 5658559.02 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:38:45,055 INFO [optim.py:369] (1/2) 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 13:38:51,274 INFO [train.py:968] (1/2) Epoch 22, batch 31400, giga_loss[loss=0.235, simple_loss=0.3205, pruned_loss=0.07479, over 27605.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3346, pruned_loss=0.08982, over 5651985.21 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3554, pruned_loss=0.1143, over 5660580.08 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3335, pruned_loss=0.08769, over 5655211.25 frames. ], batch size: 474, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:39:56,842 INFO [train.py:968] (1/2) Epoch 22, batch 31450, giga_loss[loss=0.2719, simple_loss=0.35, pruned_loss=0.09694, over 28698.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3365, pruned_loss=0.08994, over 5649656.62 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3555, pruned_loss=0.1145, over 5652012.76 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3352, pruned_loss=0.08764, over 5660496.87 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:40:50,044 INFO [optim.py:369] (1/2) 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 13:41:00,966 INFO [train.py:968] (1/2) Epoch 22, batch 31500, libri_loss[loss=0.3838, simple_loss=0.4184, pruned_loss=0.1746, over 29666.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3323, pruned_loss=0.08766, over 5653375.52 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3549, pruned_loss=0.1144, over 5657578.29 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3313, pruned_loss=0.08531, over 5656975.69 frames. ], batch size: 88, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:42:08,042 INFO [train.py:968] (1/2) Epoch 22, batch 31550, giga_loss[loss=0.2806, simple_loss=0.3544, pruned_loss=0.1034, over 27681.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3352, pruned_loss=0.08981, over 5660320.04 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3549, pruned_loss=0.1145, over 5657536.17 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.334, pruned_loss=0.08746, over 5663045.53 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:42:09,699 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:43:06,009 INFO [optim.py:369] (1/2) 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 13:43:11,363 INFO [train.py:968] (1/2) Epoch 22, batch 31600, giga_loss[loss=0.2388, simple_loss=0.3371, pruned_loss=0.07028, over 29049.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3375, pruned_loss=0.08932, over 5657508.94 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3548, pruned_loss=0.1145, over 5661686.45 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3365, pruned_loss=0.08719, over 5655982.80 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 13:44:11,982 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-11 13:44:13,711 INFO [train.py:968] (1/2) Epoch 22, batch 31650, giga_loss[loss=0.2614, simple_loss=0.3514, pruned_loss=0.08566, over 28626.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3411, pruned_loss=0.08929, over 5654386.40 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3545, pruned_loss=0.1143, over 5655398.13 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3401, pruned_loss=0.08711, over 5659042.71 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:44:59,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-11 13:45:10,249 INFO [optim.py:369] (1/2) 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 13:45:12,947 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-11 13:45:14,875 INFO [train.py:968] (1/2) Epoch 22, batch 31700, giga_loss[loss=0.2582, simple_loss=0.348, pruned_loss=0.08417, over 29028.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3423, pruned_loss=0.08926, over 5641171.94 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3546, pruned_loss=0.1144, over 5650245.27 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3411, pruned_loss=0.08693, over 5649664.30 frames. ], batch size: 285, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:45:43,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-11 13:45:48,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1000, 1.4239, 1.3721, 0.9938], device='cuda:1'), covar=tensor([0.1746, 0.2784, 0.1508, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.0903, 0.0701, 0.0951, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 13:46:15,942 INFO [train.py:968] (1/2) Epoch 22, batch 31750, giga_loss[loss=0.259, simple_loss=0.3448, pruned_loss=0.08658, over 28094.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3409, pruned_loss=0.08783, over 5645167.46 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3545, pruned_loss=0.1145, over 5651942.49 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3399, pruned_loss=0.08546, over 5650243.48 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:47:06,849 INFO [optim.py:369] (1/2) 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 13:47:10,643 INFO [train.py:968] (1/2) Epoch 22, batch 31800, giga_loss[loss=0.2282, simple_loss=0.3174, pruned_loss=0.06949, over 29019.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.341, pruned_loss=0.08927, over 5645479.96 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3538, pruned_loss=0.1143, over 5650503.81 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3403, pruned_loss=0.0865, over 5651343.28 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:47:39,429 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:48:19,877 INFO [train.py:968] (1/2) Epoch 22, batch 31850, giga_loss[loss=0.2768, simple_loss=0.3535, pruned_loss=0.1001, over 28924.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3401, pruned_loss=0.0898, over 5647242.62 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3539, pruned_loss=0.1145, over 5645179.97 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3393, pruned_loss=0.0871, over 5655950.41 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:48:47,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2716, 1.3307, 3.5485, 3.1470], device='cuda:1'), covar=tensor([0.1540, 0.2709, 0.0471, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0655, 0.0967, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 13:49:02,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5680, 1.8574, 1.8647, 1.6226], device='cuda:1'), covar=tensor([0.3286, 0.2217, 0.1701, 0.2297], device='cuda:1'), in_proj_covar=tensor([0.1935, 0.1877, 0.1790, 0.1940], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 13:49:05,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-11 13:49:19,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7354, 1.9785, 1.8270, 1.5960], device='cuda:1'), covar=tensor([0.2418, 0.2029, 0.1860, 0.2125], device='cuda:1'), in_proj_covar=tensor([0.1934, 0.1877, 0.1790, 0.1940], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 13:49:35,572 INFO [optim.py:369] (1/2) 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 13:49:39,907 INFO [train.py:968] (1/2) Epoch 22, batch 31900, giga_loss[loss=0.2117, simple_loss=0.2985, pruned_loss=0.06244, over 28390.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3397, pruned_loss=0.08976, over 5659546.77 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3539, pruned_loss=0.1145, over 5647335.67 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.339, pruned_loss=0.08741, over 5664500.18 frames. ], batch size: 369, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:50:10,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4889, 1.7668, 1.4229, 1.6599], device='cuda:1'), covar=tensor([0.2818, 0.2813, 0.3286, 0.2390], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1095, 0.1339, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 13:50:13,107 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-11 13:50:21,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=989657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:50:29,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-11 13:50:54,935 INFO [train.py:968] (1/2) Epoch 22, batch 31950, libri_loss[loss=0.2845, simple_loss=0.3391, pruned_loss=0.1149, over 29576.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3346, pruned_loss=0.08684, over 5664556.89 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3537, pruned_loss=0.1143, over 5650907.87 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3339, pruned_loss=0.08471, over 5665288.08 frames. ], batch size: 74, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 13:51:51,888 INFO [optim.py:369] (1/2) 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 13:51:57,556 INFO [train.py:968] (1/2) Epoch 22, batch 32000, giga_loss[loss=0.2575, simple_loss=0.3254, pruned_loss=0.09475, over 26915.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3322, pruned_loss=0.0855, over 5663951.36 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3537, pruned_loss=0.1145, over 5650892.70 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3314, pruned_loss=0.08335, over 5664607.21 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:52:23,950 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:01,200 INFO [train.py:968] (1/2) Epoch 22, batch 32050, giga_loss[loss=0.2847, simple_loss=0.3645, pruned_loss=0.1024, over 28845.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3315, pruned_loss=0.08576, over 5665019.24 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3532, pruned_loss=0.1141, over 5654198.08 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3306, pruned_loss=0.0835, over 5662989.60 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:53:03,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:07,579 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=989786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:22,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=989800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:25,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=989803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:53:26,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 13:53:52,613 INFO [optim.py:369] (1/2) 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 13:53:57,677 INFO [train.py:968] (1/2) Epoch 22, batch 32100, giga_loss[loss=0.2788, simple_loss=0.3582, pruned_loss=0.09963, over 28525.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3362, pruned_loss=0.08834, over 5664420.32 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3534, pruned_loss=0.1143, over 5650545.94 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3348, pruned_loss=0.08569, over 5667084.07 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:54:00,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=989832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:54:40,391 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-11 13:54:58,750 INFO [train.py:968] (1/2) Epoch 22, batch 32150, giga_loss[loss=0.2265, simple_loss=0.3129, pruned_loss=0.06999, over 28903.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3356, pruned_loss=0.08882, over 5666102.15 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3532, pruned_loss=0.1142, over 5653813.52 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3343, pruned_loss=0.0863, over 5665547.23 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:55:26,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4405, 5.2564, 4.9986, 2.7566], device='cuda:1'), covar=tensor([0.0458, 0.0615, 0.0825, 0.1687], device='cuda:1'), in_proj_covar=tensor([0.1228, 0.1134, 0.0958, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 13:55:56,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1845, 4.0185, 3.8208, 1.8349], device='cuda:1'), covar=tensor([0.0660, 0.0770, 0.0896, 0.2160], device='cuda:1'), in_proj_covar=tensor([0.1227, 0.1133, 0.0958, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 13:55:57,440 INFO [optim.py:369] (1/2) 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 13:55:57,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=989927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:56:00,846 INFO [train.py:968] (1/2) Epoch 22, batch 32200, giga_loss[loss=0.2883, simple_loss=0.361, pruned_loss=0.1078, over 28926.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3351, pruned_loss=0.08976, over 5669370.40 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3528, pruned_loss=0.114, over 5658489.36 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3341, pruned_loss=0.08747, over 5664935.15 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:57:00,327 INFO [train.py:968] (1/2) Epoch 22, batch 32250, giga_loss[loss=0.2671, simple_loss=0.3519, pruned_loss=0.09117, over 29054.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3358, pruned_loss=0.09062, over 5673352.66 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3526, pruned_loss=0.1139, over 5667648.99 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3346, pruned_loss=0.08798, over 5662113.03 frames. ], batch size: 285, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:58:02,319 INFO [optim.py:369] (1/2) 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 13:58:06,503 INFO [train.py:968] (1/2) Epoch 22, batch 32300, libri_loss[loss=0.2495, simple_loss=0.3193, pruned_loss=0.08986, over 29549.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3369, pruned_loss=0.09077, over 5677136.54 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3516, pruned_loss=0.1134, over 5676011.47 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3363, pruned_loss=0.08835, over 5660632.40 frames. ], batch size: 76, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:58:58,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:59:05,672 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 32350, libri_loss[loss=0.2752, simple_loss=0.3442, pruned_loss=0.1031, over 25999.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3378, pruned_loss=0.09048, over 5684725.06 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3506, pruned_loss=0.1129, over 5680274.53 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3376, pruned_loss=0.08805, over 5667846.69 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 13:59:47,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 13:59:51,697 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:00:18,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:00:19,895 INFO [optim.py:369] (1/2) 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 14:00:27,306 INFO [train.py:968] (1/2) Epoch 22, batch 32400, giga_loss[loss=0.2959, simple_loss=0.3489, pruned_loss=0.1214, over 26944.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3363, pruned_loss=0.08947, over 5681544.38 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3503, pruned_loss=0.1126, over 5683983.49 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.336, pruned_loss=0.08703, over 5664876.00 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:00:59,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:01:05,026 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 32450, libri_loss[loss=0.2555, simple_loss=0.3137, pruned_loss=0.09866, over 28474.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3318, pruned_loss=0.08816, over 5680862.96 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3495, pruned_loss=0.1122, over 5687458.49 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3319, pruned_loss=0.08602, over 5664364.35 frames. ], batch size: 63, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:01:56,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8335, 1.9646, 1.5477, 1.5117], device='cuda:1'), covar=tensor([0.0966, 0.0592, 0.0977, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0441, 0.0515, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 14:02:20,941 INFO [optim.py:369] (1/2) 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 14:02:21,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6297, 1.6878, 1.3699, 1.3542], device='cuda:1'), covar=tensor([0.0699, 0.0402, 0.0759, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0443, 0.0517, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 14:02:23,042 INFO [train.py:968] (1/2) Epoch 22, batch 32500, giga_loss[loss=0.2302, simple_loss=0.3097, pruned_loss=0.07539, over 29052.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3287, pruned_loss=0.08757, over 5677142.00 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3493, pruned_loss=0.1121, over 5680853.35 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3278, pruned_loss=0.08445, over 5670241.22 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:02:45,948 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:16,520 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:18,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:28,759 INFO [train.py:968] (1/2) Epoch 22, batch 32550, giga_loss[loss=0.2204, simple_loss=0.308, pruned_loss=0.0664, over 28490.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3289, pruned_loss=0.08798, over 5658963.33 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3493, pruned_loss=0.1121, over 5677604.11 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3276, pruned_loss=0.08477, over 5656162.50 frames. ], batch size: 369, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:03:38,345 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:50,296 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:51,285 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:52,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:03:55,069 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:04:00,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:04:23,171 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 32600, giga_loss[loss=0.2367, simple_loss=0.324, pruned_loss=0.07475, over 28989.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3311, pruned_loss=0.08954, over 5656881.84 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3493, pruned_loss=0.1121, over 5681077.44 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3298, pruned_loss=0.08676, over 5651553.89 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:04:29,646 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:04:36,884 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:04:45,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1530, 1.5044, 1.4424, 1.0729], device='cuda:1'), covar=tensor([0.1546, 0.2387, 0.1264, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0696, 0.0946, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 14:04:53,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 14:05:05,332 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:05:33,584 INFO [train.py:968] (1/2) Epoch 22, batch 32650, giga_loss[loss=0.2385, simple_loss=0.3278, pruned_loss=0.07457, over 28680.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3292, pruned_loss=0.08764, over 5656880.68 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3493, pruned_loss=0.1121, over 5682149.05 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.328, pruned_loss=0.08532, over 5651586.62 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:06:36,736 INFO [optim.py:369] (1/2) 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 14:06:40,047 INFO [train.py:968] (1/2) Epoch 22, batch 32700, giga_loss[loss=0.2354, simple_loss=0.3177, pruned_loss=0.07651, over 28457.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3276, pruned_loss=0.08567, over 5662336.90 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3491, pruned_loss=0.1121, over 5681829.93 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3265, pruned_loss=0.0835, over 5658172.83 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:07:25,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5338, 3.9181, 1.6348, 1.6517], device='cuda:1'), covar=tensor([0.0954, 0.0409, 0.0910, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0552, 0.0386, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 14:07:39,439 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-11 14:07:39,620 INFO [train.py:968] (1/2) Epoch 22, batch 32750, giga_loss[loss=0.213, simple_loss=0.2945, pruned_loss=0.06577, over 29007.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3263, pruned_loss=0.08577, over 5666343.40 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3482, pruned_loss=0.1116, over 5684738.66 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3256, pruned_loss=0.08365, over 5659624.37 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:07:39,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:07:42,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 14:08:45,975 INFO [optim.py:369] (1/2) 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 14:08:50,901 INFO [train.py:968] (1/2) Epoch 22, batch 32800, giga_loss[loss=0.2533, simple_loss=0.3284, pruned_loss=0.08917, over 27675.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3254, pruned_loss=0.08456, over 5657906.59 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3478, pruned_loss=0.1113, over 5686152.62 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3249, pruned_loss=0.08274, over 5651157.99 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:09:17,125 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:09:17,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3795, 1.6298, 1.6692, 1.2024], device='cuda:1'), covar=tensor([0.1698, 0.2690, 0.1466, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0695, 0.0945, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 14:09:52,883 INFO [train.py:968] (1/2) Epoch 22, batch 32850, giga_loss[loss=0.2979, simple_loss=0.3697, pruned_loss=0.1131, over 28587.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3255, pruned_loss=0.08473, over 5643255.17 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3479, pruned_loss=0.1115, over 5670878.69 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3245, pruned_loss=0.08246, over 5650730.34 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:10:19,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-11 14:10:41,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:10:43,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6482, 1.8816, 1.5481, 1.7015], device='cuda:1'), covar=tensor([0.2752, 0.2712, 0.3068, 0.2312], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1094, 0.1341, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 14:10:47,555 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:10:50,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:10:51,303 INFO [optim.py:369] (1/2) 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 14:10:52,808 INFO [train.py:968] (1/2) Epoch 22, batch 32900, giga_loss[loss=0.2389, simple_loss=0.3266, pruned_loss=0.07564, over 28646.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3266, pruned_loss=0.0859, over 5652031.56 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3478, pruned_loss=0.1115, over 5674134.77 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3255, pruned_loss=0.08376, over 5654777.04 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:11:26,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:11:34,954 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 32950, giga_loss[loss=0.2128, simple_loss=0.306, pruned_loss=0.05978, over 28455.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3249, pruned_loss=0.08451, over 5647091.58 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3479, pruned_loss=0.1116, over 5672555.58 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3236, pruned_loss=0.08224, over 5650521.82 frames. ], batch size: 369, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:12:52,892 INFO [optim.py:369] (1/2) 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 14:12:56,124 INFO [train.py:968] (1/2) Epoch 22, batch 33000, giga_loss[loss=0.2457, simple_loss=0.3339, pruned_loss=0.07874, over 28901.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3272, pruned_loss=0.08366, over 5656553.00 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3479, pruned_loss=0.1115, over 5674440.14 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.326, pruned_loss=0.08169, over 5657410.05 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:12:56,124 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 14:13:04,890 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 14:13:08,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=990734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:13:10,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2987, 1.4271, 1.2757, 1.5067], device='cuda:1'), covar=tensor([0.0794, 0.0378, 0.0364, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 14:13:13,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2707, 4.1081, 3.8905, 1.9313], device='cuda:1'), covar=tensor([0.0677, 0.0794, 0.1016, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1128, 0.0951, 0.0714], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 14:13:43,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:13:45,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:14:01,202 INFO [train.py:968] (1/2) Epoch 22, batch 33050, giga_loss[loss=0.2342, simple_loss=0.3145, pruned_loss=0.07692, over 28701.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3297, pruned_loss=0.08464, over 5658619.36 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3476, pruned_loss=0.1113, over 5679581.16 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3285, pruned_loss=0.08267, over 5654323.46 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:14:23,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:14:38,519 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:14:42,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:14:55,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4902, 1.6619, 1.7214, 1.3384], device='cuda:1'), covar=tensor([0.1841, 0.2524, 0.1546, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.0899, 0.0696, 0.0947, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 14:15:05,797 INFO [optim.py:369] (1/2) 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 14:15:07,191 INFO [train.py:968] (1/2) Epoch 22, batch 33100, giga_loss[loss=0.2912, simple_loss=0.3595, pruned_loss=0.1115, over 28706.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3303, pruned_loss=0.08493, over 5649638.59 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3474, pruned_loss=0.1112, over 5678436.38 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3294, pruned_loss=0.08323, over 5646932.51 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:15:19,495 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:03,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=990877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:07,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=990880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:07,489 INFO [train.py:968] (1/2) Epoch 22, batch 33150, giga_loss[loss=0.2274, simple_loss=0.319, pruned_loss=0.06789, over 28964.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3303, pruned_loss=0.08533, over 5658500.66 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3466, pruned_loss=0.1107, over 5683118.28 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.33, pruned_loss=0.08378, over 5651541.58 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:16:18,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2305, 1.5864, 1.6043, 1.4125], device='cuda:1'), covar=tensor([0.1908, 0.1785, 0.2007, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.0462, 0.0734, 0.0701, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 14:16:22,291 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5848, 1.7191, 1.4475, 1.5790], device='cuda:1'), covar=tensor([0.2901, 0.2856, 0.3289, 0.2593], device='cuda:1'), in_proj_covar=tensor([0.1520, 0.1097, 0.1344, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 14:16:35,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=990908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:37,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=990909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:16:56,308 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:17:00,596 INFO [train.py:968] (1/2) Epoch 22, batch 33200, giga_loss[loss=0.2092, simple_loss=0.3, pruned_loss=0.05921, over 29010.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3289, pruned_loss=0.08492, over 5670540.99 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3461, pruned_loss=0.1104, over 5693783.57 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3282, pruned_loss=0.08261, over 5653677.20 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:17:51,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-11 14:17:58,599 INFO [train.py:968] (1/2) Epoch 22, batch 33250, libri_loss[loss=0.2916, simple_loss=0.3579, pruned_loss=0.1126, over 29663.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3285, pruned_loss=0.08521, over 5673453.78 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.346, pruned_loss=0.1104, over 5697187.00 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3275, pruned_loss=0.0825, over 5655884.62 frames. ], batch size: 88, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:18:56,540 INFO [optim.py:369] (1/2) 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 14:18:57,462 INFO [train.py:968] (1/2) Epoch 22, batch 33300, giga_loss[loss=0.2199, simple_loss=0.2984, pruned_loss=0.07076, over 28396.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3261, pruned_loss=0.08453, over 5678392.10 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.346, pruned_loss=0.1104, over 5702264.57 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3247, pruned_loss=0.08168, over 5658859.37 frames. ], batch size: 65, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:19:43,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=991069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:19:46,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=991072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:19:54,443 INFO [train.py:968] (1/2) Epoch 22, batch 33350, giga_loss[loss=0.2701, simple_loss=0.3389, pruned_loss=0.1007, over 26981.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3281, pruned_loss=0.08551, over 5677842.72 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3452, pruned_loss=0.11, over 5700271.02 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.327, pruned_loss=0.08262, over 5663232.38 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:20:23,302 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=991101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:20:53,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 14:20:56,068 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 33400, giga_loss[loss=0.2666, simple_loss=0.3446, pruned_loss=0.09423, over 28083.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3301, pruned_loss=0.08652, over 5678288.40 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3445, pruned_loss=0.1096, over 5705674.92 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3294, pruned_loss=0.08391, over 5660788.20 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:21:13,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0660, 1.3422, 1.3778, 1.1017], device='cuda:1'), covar=tensor([0.2221, 0.1931, 0.1272, 0.1822], device='cuda:1'), in_proj_covar=tensor([0.1937, 0.1869, 0.1785, 0.1934], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 14:22:01,588 INFO [train.py:968] (1/2) Epoch 22, batch 33450, giga_loss[loss=0.2647, simple_loss=0.3536, pruned_loss=0.08786, over 28721.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.331, pruned_loss=0.08732, over 5670306.40 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3446, pruned_loss=0.1096, over 5699097.50 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3301, pruned_loss=0.08475, over 5660763.52 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:22:14,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3733, 1.6704, 1.4825, 1.5385], device='cuda:1'), covar=tensor([0.0801, 0.0334, 0.0339, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 14:23:06,032 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 33500, giga_loss[loss=0.2622, simple_loss=0.3536, pruned_loss=0.08539, over 29055.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3341, pruned_loss=0.08838, over 5674327.75 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3444, pruned_loss=0.1094, over 5699688.67 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3334, pruned_loss=0.08619, over 5665933.72 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:24:06,318 INFO [train.py:968] (1/2) Epoch 22, batch 33550, giga_loss[loss=0.2483, simple_loss=0.3374, pruned_loss=0.07963, over 28868.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3359, pruned_loss=0.08837, over 5666930.81 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3444, pruned_loss=0.1095, over 5700202.60 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3353, pruned_loss=0.08642, over 5659611.20 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:24:10,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=991283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:24:13,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5700, 4.3780, 4.1973, 2.0427], device='cuda:1'), covar=tensor([0.0636, 0.0823, 0.0900, 0.1885], device='cuda:1'), in_proj_covar=tensor([0.1218, 0.1121, 0.0946, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 14:24:23,668 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 14:25:13,202 INFO [optim.py:369] (1/2) 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 14:25:15,076 INFO [train.py:968] (1/2) Epoch 22, batch 33600, giga_loss[loss=0.2475, simple_loss=0.3288, pruned_loss=0.08312, over 28094.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3353, pruned_loss=0.08809, over 5669194.67 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3445, pruned_loss=0.1095, over 5704458.51 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3346, pruned_loss=0.0861, over 5659106.20 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:26:27,133 INFO [train.py:968] (1/2) Epoch 22, batch 33650, giga_loss[loss=0.2757, simple_loss=0.3474, pruned_loss=0.102, over 28651.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3335, pruned_loss=0.08791, over 5662846.29 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3443, pruned_loss=0.1095, over 5698427.10 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3329, pruned_loss=0.08595, over 5659077.98 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:27:24,074 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=991426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:27:27,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=991429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:27:29,033 INFO [optim.py:369] (1/2) 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 14:27:29,046 INFO [train.py:968] (1/2) Epoch 22, batch 33700, giga_loss[loss=0.2671, simple_loss=0.3479, pruned_loss=0.09313, over 29015.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3334, pruned_loss=0.08788, over 5666366.46 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3442, pruned_loss=0.1095, over 5699127.43 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3327, pruned_loss=0.08583, over 5661771.23 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:27:53,256 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8220, 1.9946, 1.4411, 1.6639], device='cuda:1'), covar=tensor([0.1065, 0.0713, 0.1037, 0.1243], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0443, 0.0518, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 14:28:10,889 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 33750, giga_loss[loss=0.2531, simple_loss=0.3264, pruned_loss=0.08992, over 28090.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3322, pruned_loss=0.08766, over 5640985.70 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3444, pruned_loss=0.1097, over 5689938.79 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3315, pruned_loss=0.08581, over 5645978.37 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:28:56,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1238, 2.3498, 1.6819, 1.9742], device='cuda:1'), covar=tensor([0.0956, 0.0616, 0.1016, 0.1103], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0443, 0.0518, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 14:29:44,973 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 33800, giga_loss[loss=0.2592, simple_loss=0.3398, pruned_loss=0.08927, over 28668.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3297, pruned_loss=0.08698, over 5648467.72 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.344, pruned_loss=0.1095, over 5693127.81 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3292, pruned_loss=0.08536, over 5649065.35 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:30:22,567 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=991560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:30:43,185 INFO [train.py:968] (1/2) Epoch 22, batch 33850, giga_loss[loss=0.2224, simple_loss=0.3123, pruned_loss=0.06621, over 28492.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3298, pruned_loss=0.08701, over 5637008.99 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3441, pruned_loss=0.1095, over 5685119.89 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.329, pruned_loss=0.08504, over 5643567.29 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:30:57,109 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 14:31:49,113 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 33900, giga_loss[loss=0.2642, simple_loss=0.346, pruned_loss=0.09121, over 28663.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3284, pruned_loss=0.08485, over 5655259.14 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3443, pruned_loss=0.1097, over 5687427.44 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3274, pruned_loss=0.08292, over 5658006.77 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:31:51,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2142, 1.2586, 3.8202, 3.1752], device='cuda:1'), covar=tensor([0.1700, 0.2785, 0.0409, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0646, 0.0953, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 14:32:15,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8299, 1.2016, 1.3108, 0.9878], device='cuda:1'), covar=tensor([0.2093, 0.1479, 0.2499, 0.1945], device='cuda:1'), in_proj_covar=tensor([0.0459, 0.0731, 0.0698, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 14:32:41,293 INFO [train.py:968] (1/2) Epoch 22, batch 33950, giga_loss[loss=0.2654, simple_loss=0.3582, pruned_loss=0.08625, over 28050.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3303, pruned_loss=0.08418, over 5660964.12 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3444, pruned_loss=0.1099, over 5683693.24 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3289, pruned_loss=0.08161, over 5666176.15 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:33:38,966 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 34000, giga_loss[loss=0.1979, simple_loss=0.2928, pruned_loss=0.05146, over 28457.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3325, pruned_loss=0.08462, over 5661982.98 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3445, pruned_loss=0.1101, over 5688956.98 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.331, pruned_loss=0.08183, over 5661158.90 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:34:39,018 INFO [train.py:968] (1/2) Epoch 22, batch 34050, giga_loss[loss=0.2415, simple_loss=0.3287, pruned_loss=0.07722, over 28950.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3326, pruned_loss=0.08451, over 5664542.44 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3442, pruned_loss=0.1098, over 5692106.53 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3315, pruned_loss=0.08208, over 5660802.28 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:35:26,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2015, 1.7715, 1.2953, 0.4695], device='cuda:1'), covar=tensor([0.4624, 0.2923, 0.4683, 0.6238], device='cuda:1'), in_proj_covar=tensor([0.1759, 0.1659, 0.1597, 0.1442], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 14:35:48,310 INFO [train.py:968] (1/2) Epoch 22, batch 34100, giga_loss[loss=0.2483, simple_loss=0.331, pruned_loss=0.08283, over 28940.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3328, pruned_loss=0.08481, over 5663025.08 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.344, pruned_loss=0.1097, over 5686779.99 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3318, pruned_loss=0.08247, over 5664336.71 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:35:52,026 INFO [optim.py:369] (1/2) 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 14:35:58,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2780, 1.5912, 1.5603, 1.4764], device='cuda:1'), covar=tensor([0.1888, 0.1795, 0.1979, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0732, 0.0698, 0.0669], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 14:36:55,808 INFO [train.py:968] (1/2) Epoch 22, batch 34150, giga_loss[loss=0.3016, simple_loss=0.3576, pruned_loss=0.1228, over 26856.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.334, pruned_loss=0.0859, over 5659778.05 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3443, pruned_loss=0.11, over 5687956.44 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3327, pruned_loss=0.08305, over 5659045.75 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:37:04,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3858, 1.6028, 1.5968, 1.3986], device='cuda:1'), covar=tensor([0.2815, 0.2242, 0.1896, 0.2270], device='cuda:1'), in_proj_covar=tensor([0.1936, 0.1864, 0.1776, 0.1924], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 14:37:44,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1635, 1.3136, 3.3029, 2.9581], device='cuda:1'), covar=tensor([0.1628, 0.2694, 0.0494, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0647, 0.0952, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 14:38:09,863 INFO [train.py:968] (1/2) Epoch 22, batch 34200, giga_loss[loss=0.2455, simple_loss=0.329, pruned_loss=0.08097, over 29012.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.334, pruned_loss=0.08532, over 5663334.03 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3439, pruned_loss=0.1098, over 5691342.55 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3331, pruned_loss=0.08289, over 5659468.23 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:38:11,614 INFO [optim.py:369] (1/2) 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 14:38:17,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=991935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:39:15,364 INFO [train.py:968] (1/2) Epoch 22, batch 34250, giga_loss[loss=0.2243, simple_loss=0.3194, pruned_loss=0.06458, over 28364.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3343, pruned_loss=0.08546, over 5658394.39 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3436, pruned_loss=0.1096, over 5695043.88 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3336, pruned_loss=0.08301, over 5650931.05 frames. ], batch size: 65, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:40:20,036 INFO [train.py:968] (1/2) Epoch 22, batch 34300, giga_loss[loss=0.2502, simple_loss=0.341, pruned_loss=0.07964, over 28848.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.338, pruned_loss=0.08726, over 5674838.58 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.344, pruned_loss=0.1101, over 5700625.64 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3369, pruned_loss=0.08417, over 5663080.87 frames. ], batch size: 243, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:40:21,497 INFO [optim.py:369] (1/2) 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 14:40:29,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5943, 3.0597, 2.7640, 2.3002], device='cuda:1'), covar=tensor([0.2358, 0.1528, 0.1646, 0.2075], device='cuda:1'), in_proj_covar=tensor([0.1924, 0.1854, 0.1765, 0.1916], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 14:40:32,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6346, 1.4835, 4.1816, 3.3963], device='cuda:1'), covar=tensor([0.1485, 0.2637, 0.0463, 0.1055], device='cuda:1'), in_proj_covar=tensor([0.0755, 0.0647, 0.0951, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 14:40:42,336 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=992048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:41:18,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3088, 1.8177, 1.3380, 0.7147], device='cuda:1'), covar=tensor([0.4903, 0.2656, 0.3494, 0.6189], device='cuda:1'), in_proj_covar=tensor([0.1758, 0.1662, 0.1601, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 14:41:20,844 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=992078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:41:23,237 INFO [train.py:968] (1/2) Epoch 22, batch 34350, giga_loss[loss=0.2563, simple_loss=0.3383, pruned_loss=0.0872, over 29109.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3373, pruned_loss=0.08752, over 5681876.09 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3428, pruned_loss=0.1093, over 5702433.44 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3373, pruned_loss=0.08489, over 5669646.24 frames. ], batch size: 285, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:41:23,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=992081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:42:00,900 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=992110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:42:26,598 INFO [train.py:968] (1/2) Epoch 22, batch 34400, libri_loss[loss=0.3114, simple_loss=0.3725, pruned_loss=0.1251, over 29113.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3361, pruned_loss=0.08745, over 5690709.52 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3431, pruned_loss=0.1096, over 5704692.59 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3355, pruned_loss=0.08435, over 5678170.39 frames. ], batch size: 101, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:42:28,415 INFO [optim.py:369] (1/2) 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 14:42:46,166 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-11 14:43:02,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7117, 2.1436, 2.0372, 1.6217], device='cuda:1'), covar=tensor([0.3268, 0.1995, 0.2118, 0.2635], device='cuda:1'), in_proj_covar=tensor([0.1923, 0.1851, 0.1762, 0.1917], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 14:43:34,413 INFO [train.py:968] (1/2) Epoch 22, batch 34450, giga_loss[loss=0.2237, simple_loss=0.3185, pruned_loss=0.06446, over 28980.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3343, pruned_loss=0.08632, over 5687417.28 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3429, pruned_loss=0.1094, over 5701434.14 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3339, pruned_loss=0.08329, over 5678891.27 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:43:47,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8511, 1.2565, 1.3307, 1.0401], device='cuda:1'), covar=tensor([0.2024, 0.1365, 0.2317, 0.1766], device='cuda:1'), in_proj_covar=tensor([0.0457, 0.0730, 0.0696, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 14:44:39,333 INFO [train.py:968] (1/2) Epoch 22, batch 34500, giga_loss[loss=0.2419, simple_loss=0.328, pruned_loss=0.07794, over 28291.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3321, pruned_loss=0.08417, over 5698456.57 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3426, pruned_loss=0.1093, over 5702838.12 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3319, pruned_loss=0.08154, over 5690403.87 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:44:41,512 INFO [optim.py:369] (1/2) 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 14:45:42,513 INFO [train.py:968] (1/2) Epoch 22, batch 34550, giga_loss[loss=0.2272, simple_loss=0.3179, pruned_loss=0.06823, over 28710.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3328, pruned_loss=0.08446, over 5690107.98 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3428, pruned_loss=0.1093, over 5705715.27 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3323, pruned_loss=0.08202, over 5681114.44 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:46:31,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5173, 4.2492, 1.7068, 1.6613], device='cuda:1'), covar=tensor([0.1003, 0.0270, 0.0932, 0.1320], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0553, 0.0388, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 14:46:43,749 INFO [train.py:968] (1/2) Epoch 22, batch 34600, giga_loss[loss=0.2655, simple_loss=0.342, pruned_loss=0.09445, over 28207.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3362, pruned_loss=0.08675, over 5684879.48 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3429, pruned_loss=0.1094, over 5709501.04 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3357, pruned_loss=0.08427, over 5674037.12 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:46:45,141 INFO [optim.py:369] (1/2) 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 14:47:38,991 INFO [train.py:968] (1/2) Epoch 22, batch 34650, giga_loss[loss=0.22, simple_loss=0.306, pruned_loss=0.06699, over 28744.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3361, pruned_loss=0.0877, over 5676187.12 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3429, pruned_loss=0.1096, over 5705794.93 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3355, pruned_loss=0.08499, over 5670112.99 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 14:48:07,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1921, 2.4019, 1.2940, 1.3786], device='cuda:1'), covar=tensor([0.0975, 0.0597, 0.0964, 0.1359], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0551, 0.0388, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 14:48:24,436 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 34700, giga_loss[loss=0.2655, simple_loss=0.3445, pruned_loss=0.09326, over 28493.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.334, pruned_loss=0.08757, over 5653373.98 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3433, pruned_loss=0.1098, over 5684916.00 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3328, pruned_loss=0.08443, over 5666198.83 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 14:48:34,208 INFO [optim.py:369] (1/2) 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 14:49:11,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2855, 1.2677, 3.6484, 3.0868], device='cuda:1'), covar=tensor([0.1964, 0.3067, 0.0867, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0651, 0.0959, 0.0901], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 14:49:29,986 INFO [train.py:968] (1/2) Epoch 22, batch 34750, libri_loss[loss=0.3118, simple_loss=0.3535, pruned_loss=0.1351, over 29567.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3344, pruned_loss=0.08893, over 5662140.19 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3436, pruned_loss=0.1102, over 5690593.01 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.333, pruned_loss=0.08556, over 5666663.72 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 14:49:47,007 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-11 14:50:18,350 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.93 vs. limit=5.0
+2023-03-11 14:50:23,368 INFO [train.py:968] (1/2) Epoch 22, batch 34800, giga_loss[loss=0.3537, simple_loss=0.403, pruned_loss=0.1522, over 26749.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3399, pruned_loss=0.09221, over 5653756.84 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3434, pruned_loss=0.11, over 5689918.45 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3388, pruned_loss=0.08931, over 5657355.69 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:50:26,536 INFO [optim.py:369] (1/2) 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 14:50:53,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=992566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:50:56,415 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=992569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:51:07,484 INFO [train.py:968] (1/2) Epoch 22, batch 34850, giga_loss[loss=0.2872, simple_loss=0.3709, pruned_loss=0.1018, over 28653.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3487, pruned_loss=0.09695, over 5663217.83 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3432, pruned_loss=0.11, over 5684980.86 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3479, pruned_loss=0.09427, over 5669699.75 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:51:21,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4144, 1.5755, 1.6392, 1.2402], device='cuda:1'), covar=tensor([0.1776, 0.2530, 0.1487, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0694, 0.0946, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 14:51:23,961 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=992598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 14:51:37,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 14:51:51,964 INFO [train.py:968] (1/2) Epoch 22, batch 34900, giga_loss[loss=0.2619, simple_loss=0.3402, pruned_loss=0.09179, over 28997.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3523, pruned_loss=0.09915, over 5669657.60 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.343, pruned_loss=0.1098, over 5688710.08 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3521, pruned_loss=0.09697, over 5671153.58 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:51:54,308 INFO [optim.py:369] (1/2) 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 14:52:09,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9543, 5.7435, 5.4374, 3.0407], device='cuda:1'), covar=tensor([0.0355, 0.0514, 0.0635, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.1221, 0.1120, 0.0945, 0.0711], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 14:52:16,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4914, 3.0877, 1.6574, 1.6031], device='cuda:1'), covar=tensor([0.0838, 0.0326, 0.0740, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0550, 0.0386, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 14:52:30,620 INFO [train.py:968] (1/2) Epoch 22, batch 34950, giga_loss[loss=0.2697, simple_loss=0.3422, pruned_loss=0.09866, over 28760.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3488, pruned_loss=0.09779, over 5676269.81 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.343, pruned_loss=0.1096, over 5690889.08 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3489, pruned_loss=0.09578, over 5674620.20 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:52:44,932 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 35000, giga_loss[loss=0.261, simple_loss=0.3371, pruned_loss=0.09238, over 28845.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3419, pruned_loss=0.09493, over 5681859.08 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3431, pruned_loss=0.1095, over 5693160.66 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.342, pruned_loss=0.0932, over 5678226.45 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:53:14,818 INFO [optim.py:369] (1/2) 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 14:53:20,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 14:53:24,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-11 14:53:54,794 INFO [train.py:968] (1/2) Epoch 22, batch 35050, giga_loss[loss=0.2839, simple_loss=0.3348, pruned_loss=0.1165, over 26652.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3346, pruned_loss=0.09174, over 5678349.10 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3432, pruned_loss=0.1094, over 5688916.84 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3345, pruned_loss=0.09008, over 5678546.63 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:54:36,122 INFO [train.py:968] (1/2) Epoch 22, batch 35100, giga_loss[loss=0.2327, simple_loss=0.2846, pruned_loss=0.09046, over 23853.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3274, pruned_loss=0.08878, over 5672074.42 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3429, pruned_loss=0.1092, over 5685729.89 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3272, pruned_loss=0.08718, over 5674751.25 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:54:37,863 INFO [optim.py:369] (1/2) 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 14:55:18,687 INFO [train.py:968] (1/2) Epoch 22, batch 35150, giga_loss[loss=0.1919, simple_loss=0.2745, pruned_loss=0.05462, over 28809.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3209, pruned_loss=0.08599, over 5679152.88 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3436, pruned_loss=0.1096, over 5687428.19 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3201, pruned_loss=0.08411, over 5679672.32 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:56:01,216 INFO [train.py:968] (1/2) Epoch 22, batch 35200, giga_loss[loss=0.2287, simple_loss=0.3044, pruned_loss=0.07656, over 28911.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3175, pruned_loss=0.08465, over 5685060.63 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3438, pruned_loss=0.1096, over 5686967.13 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.3157, pruned_loss=0.08217, over 5685752.34 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:56:03,203 INFO [optim.py:369] (1/2) 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 14:56:41,164 INFO [train.py:968] (1/2) Epoch 22, batch 35250, giga_loss[loss=0.211, simple_loss=0.2886, pruned_loss=0.06675, over 28935.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3155, pruned_loss=0.08381, over 5696428.39 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3438, pruned_loss=0.1094, over 5695611.54 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3129, pruned_loss=0.08103, over 5689165.22 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 14:57:22,773 INFO [train.py:968] (1/2) Epoch 22, batch 35300, giga_loss[loss=0.2509, simple_loss=0.2988, pruned_loss=0.1015, over 23936.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.3129, pruned_loss=0.08296, over 5689165.68 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3442, pruned_loss=0.1096, over 5699248.04 frames. ], giga_tot_loss[loss=0.2348, simple_loss=0.3098, pruned_loss=0.07991, over 5679940.67 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:57:25,415 INFO [optim.py:369] (1/2) 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 14:57:26,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 14:57:45,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-11 14:57:59,402 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 35350, giga_loss[loss=0.2147, simple_loss=0.2881, pruned_loss=0.07064, over 28938.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3098, pruned_loss=0.08158, over 5686069.74 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3448, pruned_loss=0.11, over 5704398.93 frames. ], giga_tot_loss[loss=0.2312, simple_loss=0.306, pruned_loss=0.07819, over 5673646.52 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:58:28,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0516, 1.2920, 3.4094, 2.9669], device='cuda:1'), covar=tensor([0.1739, 0.2693, 0.0507, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0647, 0.0957, 0.0898], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 14:58:47,766 INFO [train.py:968] (1/2) Epoch 22, batch 35400, giga_loss[loss=0.2317, simple_loss=0.2994, pruned_loss=0.08205, over 28960.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3071, pruned_loss=0.08044, over 5687024.17 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3448, pruned_loss=0.1098, over 5703095.68 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.3032, pruned_loss=0.07724, over 5677769.54 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 14:58:50,552 INFO [optim.py:369] (1/2) 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 14:58:55,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3568, 4.1801, 3.9557, 2.1187], device='cuda:1'), covar=tensor([0.0568, 0.0760, 0.0753, 0.1931], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1129, 0.0952, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 14:59:25,674 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 14:59:31,477 INFO [train.py:968] (1/2) Epoch 22, batch 35450, libri_loss[loss=0.2417, simple_loss=0.3089, pruned_loss=0.0872, over 29325.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3047, pruned_loss=0.07911, over 5694369.49 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.345, pruned_loss=0.1097, over 5708079.75 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.3005, pruned_loss=0.0759, over 5681915.09 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:00:00,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=993215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:02,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=993218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:16,303 INFO [train.py:968] (1/2) Epoch 22, batch 35500, giga_loss[loss=0.2056, simple_loss=0.2784, pruned_loss=0.06635, over 28918.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3006, pruned_loss=0.07714, over 5694870.86 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3449, pruned_loss=0.1097, over 5709093.36 frames. ], giga_tot_loss[loss=0.2231, simple_loss=0.2972, pruned_loss=0.07454, over 5684157.39 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:00:18,965 INFO [optim.py:369] (1/2) 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 15:00:28,406 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=993247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:35,250 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=993254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:00:58,652 INFO [train.py:968] (1/2) Epoch 22, batch 35550, giga_loss[loss=0.2204, simple_loss=0.2785, pruned_loss=0.08115, over 23787.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.2986, pruned_loss=0.07649, over 5691199.25 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3449, pruned_loss=0.1096, over 5713813.28 frames. ], giga_tot_loss[loss=0.2208, simple_loss=0.2946, pruned_loss=0.07353, over 5677901.99 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:01:42,190 INFO [train.py:968] (1/2) Epoch 22, batch 35600, giga_loss[loss=0.2891, simple_loss=0.3547, pruned_loss=0.1117, over 28893.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.2984, pruned_loss=0.07694, over 5690524.23 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3456, pruned_loss=0.1099, over 5718040.12 frames. ], giga_tot_loss[loss=0.2203, simple_loss=0.2935, pruned_loss=0.07356, over 5675451.70 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:01:45,377 INFO [optim.py:369] (1/2) 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 15:01:47,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4226, 1.6498, 1.3884, 1.5311], device='cuda:1'), covar=tensor([0.0779, 0.0345, 0.0344, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 15:02:14,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-11 15:02:21,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2887, 1.5319, 1.6179, 1.3882], device='cuda:1'), covar=tensor([0.1936, 0.1731, 0.2222, 0.1905], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0743, 0.0708, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 15:02:29,139 INFO [train.py:968] (1/2) Epoch 22, batch 35650, giga_loss[loss=0.3009, simple_loss=0.3738, pruned_loss=0.114, over 28765.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3065, pruned_loss=0.08106, over 5690750.99 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3454, pruned_loss=0.1097, over 5719948.73 frames. ], giga_tot_loss[loss=0.229, simple_loss=0.3019, pruned_loss=0.07801, over 5676361.49 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:02:41,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-11 15:03:05,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=993419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:03:17,229 INFO [train.py:968] (1/2) Epoch 22, batch 35700, giga_loss[loss=0.2946, simple_loss=0.3685, pruned_loss=0.1104, over 28962.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3212, pruned_loss=0.08907, over 5689574.25 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3454, pruned_loss=0.1096, over 5721031.94 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3175, pruned_loss=0.08651, over 5677069.21 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:03:20,685 INFO [optim.py:369] (1/2) 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 15:04:00,704 INFO [train.py:968] (1/2) Epoch 22, batch 35750, giga_loss[loss=0.3042, simple_loss=0.3687, pruned_loss=0.1198, over 29035.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3315, pruned_loss=0.0937, over 5691077.59 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3458, pruned_loss=0.1098, over 5722768.85 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3278, pruned_loss=0.09119, over 5679099.37 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:04:11,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4316, 3.5901, 1.6596, 1.5429], device='cuda:1'), covar=tensor([0.1045, 0.0283, 0.0892, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0549, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 15:04:33,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8115, 5.5852, 5.2561, 2.7300], device='cuda:1'), covar=tensor([0.0398, 0.0557, 0.0664, 0.1718], device='cuda:1'), in_proj_covar=tensor([0.1222, 0.1129, 0.0949, 0.0716], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 15:04:43,879 INFO [train.py:968] (1/2) Epoch 22, batch 35800, giga_loss[loss=0.301, simple_loss=0.3532, pruned_loss=0.1244, over 23527.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3387, pruned_loss=0.09657, over 5684423.94 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3464, pruned_loss=0.11, over 5721147.13 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3349, pruned_loss=0.09393, over 5675743.72 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:04:44,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4710, 1.6695, 1.3240, 1.1934], device='cuda:1'), covar=tensor([0.1024, 0.0614, 0.1025, 0.1235], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0442, 0.0520, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 15:04:47,787 INFO [optim.py:369] (1/2) 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 15:05:12,435 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=993564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:05:27,923 INFO [train.py:968] (1/2) Epoch 22, batch 35850, giga_loss[loss=0.2784, simple_loss=0.363, pruned_loss=0.09695, over 28706.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3422, pruned_loss=0.09721, over 5677287.38 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3471, pruned_loss=0.1105, over 5713882.97 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3385, pruned_loss=0.09452, over 5675980.11 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:06:14,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=993629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:06:16,941 INFO [train.py:968] (1/2) Epoch 22, batch 35900, giga_loss[loss=0.2987, simple_loss=0.3744, pruned_loss=0.1114, over 27982.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3426, pruned_loss=0.09632, over 5667392.00 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3474, pruned_loss=0.1105, over 5708049.73 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3392, pruned_loss=0.09378, over 5670892.25 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:06:21,946 INFO [optim.py:369] (1/2) 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 15:06:41,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2528, 1.3808, 1.2611, 1.0983], device='cuda:1'), covar=tensor([0.2344, 0.2560, 0.1720, 0.2394], device='cuda:1'), in_proj_covar=tensor([0.1952, 0.1873, 0.1795, 0.1948], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:07:02,030 INFO [train.py:968] (1/2) Epoch 22, batch 35950, giga_loss[loss=0.2519, simple_loss=0.3241, pruned_loss=0.08985, over 28064.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3451, pruned_loss=0.09797, over 5672140.68 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3474, pruned_loss=0.1105, over 5708589.18 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3424, pruned_loss=0.09591, over 5673932.77 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:07:47,757 INFO [train.py:968] (1/2) Epoch 22, batch 36000, giga_loss[loss=0.3211, simple_loss=0.398, pruned_loss=0.1221, over 28607.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.348, pruned_loss=0.1006, over 5675692.21 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3473, pruned_loss=0.1104, over 5711856.06 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.346, pruned_loss=0.09882, over 5673980.16 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:07:47,758 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 15:07:56,283 INFO [train.py:1012] (1/2) Epoch 22, validation: loss=0.2052, simple_loss=0.313, pruned_loss=0.04868, over 944034.00 frames. 
+2023-03-11 15:07:56,283 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 15:07:59,540 INFO [optim.py:369] (1/2) 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 15:08:20,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3585, 1.6115, 1.4634, 1.5269], device='cuda:1'), covar=tensor([0.0840, 0.0343, 0.0339, 0.0921], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 15:08:29,743 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=993772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:08:32,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=993775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:08:36,919 INFO [train.py:968] (1/2) Epoch 22, batch 36050, giga_loss[loss=0.2697, simple_loss=0.3515, pruned_loss=0.09393, over 28794.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3509, pruned_loss=0.102, over 5668890.96 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3477, pruned_loss=0.1105, over 5696038.14 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3489, pruned_loss=0.1004, over 5680290.86 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:08:43,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8494, 2.6322, 1.6403, 0.9517], device='cuda:1'), covar=tensor([0.7565, 0.3743, 0.4524, 0.6872], device='cuda:1'), in_proj_covar=tensor([0.1750, 0.1660, 0.1595, 0.1433], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 15:08:48,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=993794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:08:57,529 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=993804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:09:18,314 INFO [train.py:968] (1/2) Epoch 22, batch 36100, giga_loss[loss=0.2644, simple_loss=0.3519, pruned_loss=0.08847, over 29085.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3543, pruned_loss=0.1026, over 5685459.88 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3482, pruned_loss=0.1108, over 5694624.81 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3524, pruned_loss=0.101, over 5695674.51 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:09:22,737 INFO [optim.py:369] (1/2) 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 15:10:02,214 INFO [train.py:968] (1/2) Epoch 22, batch 36150, giga_loss[loss=0.2352, simple_loss=0.3251, pruned_loss=0.0726, over 28554.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3556, pruned_loss=0.1025, over 5684414.21 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3485, pruned_loss=0.1108, over 5697261.38 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3539, pruned_loss=0.1011, over 5689917.56 frames. ], batch size: 65, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:10:41,969 INFO [train.py:968] (1/2) Epoch 22, batch 36200, giga_loss[loss=0.2918, simple_loss=0.3639, pruned_loss=0.1098, over 27590.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3563, pruned_loss=0.1017, over 5684157.48 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3489, pruned_loss=0.1109, over 5692848.34 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3548, pruned_loss=0.1003, over 5692383.22 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:10:45,545 INFO [optim.py:369] (1/2) 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 15:10:48,580 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=993937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:10:49,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=993939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:10:50,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=993940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:10:56,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3112, 1.1210, 1.0800, 1.4186], device='cuda:1'), covar=tensor([0.0826, 0.0394, 0.0377, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 15:11:12,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=993969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:11:21,900 INFO [train.py:968] (1/2) Epoch 22, batch 36250, libri_loss[loss=0.3303, simple_loss=0.3918, pruned_loss=0.1344, over 29399.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3553, pruned_loss=0.1002, over 5693531.57 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3497, pruned_loss=0.1113, over 5699945.08 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3535, pruned_loss=0.09832, over 5693223.09 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:11:58,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 15:12:04,630 INFO [train.py:968] (1/2) Epoch 22, batch 36300, giga_loss[loss=0.2503, simple_loss=0.3391, pruned_loss=0.08079, over 28799.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3531, pruned_loss=0.09784, over 5694069.85 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3501, pruned_loss=0.1113, over 5700371.05 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3514, pruned_loss=0.09608, over 5693548.66 frames. ], batch size: 66, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:12:09,055 INFO [optim.py:369] (1/2) 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 15:12:10,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5752, 4.3781, 4.1936, 2.2823], device='cuda:1'), covar=tensor([0.0575, 0.0741, 0.0756, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.1209, 0.1119, 0.0942, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0017, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 15:12:46,244 INFO [train.py:968] (1/2) Epoch 22, batch 36350, giga_loss[loss=0.3243, simple_loss=0.3923, pruned_loss=0.1282, over 28872.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3546, pruned_loss=0.09946, over 5680841.82 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3504, pruned_loss=0.1115, over 5693365.74 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.353, pruned_loss=0.09775, over 5686387.99 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:12:47,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=994082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:12:48,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=994085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:12:53,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8616, 1.1072, 0.9796, 0.7801], device='cuda:1'), covar=tensor([0.2598, 0.3036, 0.1925, 0.2604], device='cuda:1'), in_proj_covar=tensor([0.1946, 0.1875, 0.1799, 0.1948], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:13:13,877 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=994114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:13:23,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-11 15:13:29,674 INFO [train.py:968] (1/2) Epoch 22, batch 36400, giga_loss[loss=0.2659, simple_loss=0.3503, pruned_loss=0.09076, over 28144.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3562, pruned_loss=0.1024, over 5676781.37 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3507, pruned_loss=0.1115, over 5693167.46 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3548, pruned_loss=0.1008, over 5680990.64 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:13:35,033 INFO [optim.py:369] (1/2) 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 15:13:37,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5033, 2.2207, 1.7214, 0.7128], device='cuda:1'), covar=tensor([0.6845, 0.3841, 0.4336, 0.6853], device='cuda:1'), in_proj_covar=tensor([0.1744, 0.1651, 0.1593, 0.1428], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 15:14:13,552 INFO [train.py:968] (1/2) Epoch 22, batch 36450, giga_loss[loss=0.2958, simple_loss=0.3616, pruned_loss=0.115, over 28617.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3598, pruned_loss=0.107, over 5683907.97 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3511, pruned_loss=0.1117, over 5694074.59 frames. ], giga_tot_loss[loss=0.2847, simple_loss=0.3585, pruned_loss=0.1054, over 5686070.11 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:14:57,334 INFO [train.py:968] (1/2) Epoch 22, batch 36500, giga_loss[loss=0.2475, simple_loss=0.3093, pruned_loss=0.09284, over 23995.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3591, pruned_loss=0.1081, over 5677530.57 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3514, pruned_loss=0.1121, over 5690715.97 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3579, pruned_loss=0.1064, over 5682497.17 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:15:05,880 INFO [optim.py:369] (1/2) 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 15:15:40,138 INFO [train.py:968] (1/2) Epoch 22, batch 36550, libri_loss[loss=0.2232, simple_loss=0.3014, pruned_loss=0.07247, over 29577.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3565, pruned_loss=0.1072, over 5682676.23 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3509, pruned_loss=0.1116, over 5687401.01 frames. ], giga_tot_loss[loss=0.2843, simple_loss=0.3563, pruned_loss=0.1062, over 5689599.70 frames. ], batch size: 74, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:15:55,862 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5232, 1.6878, 1.7603, 1.3608], device='cuda:1'), covar=tensor([0.1708, 0.2441, 0.1390, 0.1605], device='cuda:1'), in_proj_covar=tensor([0.0903, 0.0699, 0.0951, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 15:16:20,194 INFO [train.py:968] (1/2) Epoch 22, batch 36600, giga_loss[loss=0.2841, simple_loss=0.3554, pruned_loss=0.1064, over 29025.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3551, pruned_loss=0.1067, over 5684017.67 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3513, pruned_loss=0.1118, over 5686583.49 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3547, pruned_loss=0.1055, over 5690355.64 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:16:27,655 INFO [optim.py:369] (1/2) 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 15:16:54,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2643, 1.3032, 3.8461, 3.1552], device='cuda:1'), covar=tensor([0.2027, 0.2995, 0.0741, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0650, 0.0966, 0.0907], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 15:17:02,530 INFO [train.py:968] (1/2) Epoch 22, batch 36650, giga_loss[loss=0.2922, simple_loss=0.3551, pruned_loss=0.1146, over 28677.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3537, pruned_loss=0.1053, over 5686089.80 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3515, pruned_loss=0.1117, over 5688353.90 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3533, pruned_loss=0.1043, over 5689841.22 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:17:03,693 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=994381.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:17:20,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-11 15:17:45,309 INFO [train.py:968] (1/2) Epoch 22, batch 36700, giga_loss[loss=0.2755, simple_loss=0.3454, pruned_loss=0.1028, over 28741.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3516, pruned_loss=0.103, over 5696014.69 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3518, pruned_loss=0.1116, over 5691745.35 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 5695953.93 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:17:51,604 INFO [optim.py:369] (1/2) 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 15:18:18,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5916, 1.6729, 1.2685, 1.2153], device='cuda:1'), covar=tensor([0.1023, 0.0635, 0.1094, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0445, 0.0523, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 15:18:33,805 INFO [train.py:968] (1/2) Epoch 22, batch 36750, giga_loss[loss=0.2356, simple_loss=0.3147, pruned_loss=0.07821, over 27935.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.348, pruned_loss=0.1012, over 5679600.30 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3524, pruned_loss=0.1119, over 5684210.88 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.347, pruned_loss=0.09991, over 5686868.58 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:19:18,248 INFO [train.py:968] (1/2) Epoch 22, batch 36800, giga_loss[loss=0.221, simple_loss=0.2975, pruned_loss=0.07223, over 28809.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3419, pruned_loss=0.09756, over 5699468.02 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3526, pruned_loss=0.1119, over 5691000.08 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3407, pruned_loss=0.09614, over 5699287.41 frames. ], batch size: 112, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:19:21,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2584, 1.3310, 1.3543, 1.2256], device='cuda:1'), covar=tensor([0.2333, 0.2675, 0.1667, 0.2276], device='cuda:1'), in_proj_covar=tensor([0.1954, 0.1884, 0.1813, 0.1958], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:19:27,369 INFO [optim.py:369] (1/2) 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 15:20:10,635 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-11 15:20:10,750 INFO [train.py:968] (1/2) Epoch 22, batch 36850, giga_loss[loss=0.2442, simple_loss=0.3216, pruned_loss=0.08342, over 28912.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3363, pruned_loss=0.09484, over 5671080.15 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3529, pruned_loss=0.112, over 5685887.42 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3348, pruned_loss=0.09331, over 5675313.11 frames. ], batch size: 227, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:20:44,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8269, 2.6669, 1.6543, 0.9208], device='cuda:1'), covar=tensor([0.8191, 0.3402, 0.4468, 0.7378], device='cuda:1'), in_proj_covar=tensor([0.1738, 0.1646, 0.1582, 0.1422], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 15:20:57,767 INFO [train.py:968] (1/2) Epoch 22, batch 36900, giga_loss[loss=0.2426, simple_loss=0.3246, pruned_loss=0.08027, over 28884.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3352, pruned_loss=0.09374, over 5670927.26 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3528, pruned_loss=0.1118, over 5681572.66 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3336, pruned_loss=0.09228, over 5677368.61 frames. ], batch size: 199, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:21:06,546 INFO [optim.py:369] (1/2) 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 15:21:23,957 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=994659.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:21:29,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-11 15:21:43,886 INFO [train.py:968] (1/2) Epoch 22, batch 36950, giga_loss[loss=0.2426, simple_loss=0.322, pruned_loss=0.08158, over 28948.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3353, pruned_loss=0.09349, over 5676685.00 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3527, pruned_loss=0.1117, over 5683800.14 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.334, pruned_loss=0.09225, over 5679615.86 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:22:01,532 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-11 15:22:26,842 INFO [train.py:968] (1/2) Epoch 22, batch 37000, libri_loss[loss=0.3135, simple_loss=0.3834, pruned_loss=0.1218, over 29683.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.336, pruned_loss=0.09377, over 5685969.24 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3529, pruned_loss=0.1117, over 5687304.26 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3345, pruned_loss=0.09247, over 5684935.86 frames. ], batch size: 88, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:22:33,415 INFO [optim.py:369] (1/2) 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 15:22:47,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=994756.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:22:48,483 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=994757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:23:08,006 INFO [train.py:968] (1/2) Epoch 22, batch 37050, libri_loss[loss=0.2336, simple_loss=0.3074, pruned_loss=0.07987, over 29329.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3348, pruned_loss=0.09337, over 5689492.84 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3533, pruned_loss=0.1117, over 5690170.54 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3329, pruned_loss=0.09198, over 5686057.32 frames. ], batch size: 67, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:23:46,487 INFO [train.py:968] (1/2) Epoch 22, batch 37100, giga_loss[loss=0.2399, simple_loss=0.3135, pruned_loss=0.08312, over 28790.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3329, pruned_loss=0.09239, over 5703289.95 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3533, pruned_loss=0.1115, over 5696143.29 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3307, pruned_loss=0.09088, over 5695308.86 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:23:51,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4987, 1.5920, 1.5151, 1.3865], device='cuda:1'), covar=tensor([0.2764, 0.2731, 0.1888, 0.2560], device='cuda:1'), in_proj_covar=tensor([0.1950, 0.1881, 0.1810, 0.1963], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:23:54,805 INFO [optim.py:369] (1/2) 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 15:24:19,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3915, 1.1758, 3.7734, 3.3143], device='cuda:1'), covar=tensor([0.1413, 0.2567, 0.0442, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0757, 0.0645, 0.0956, 0.0900], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 15:24:29,055 INFO [train.py:968] (1/2) Epoch 22, batch 37150, giga_loss[loss=0.2544, simple_loss=0.3305, pruned_loss=0.08918, over 28897.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3304, pruned_loss=0.09112, over 5711028.37 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3537, pruned_loss=0.1117, over 5697876.37 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.328, pruned_loss=0.08951, over 5703332.88 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:24:43,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=994899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:24:46,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=994902.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:24:59,997 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 15:25:12,427 INFO [train.py:968] (1/2) Epoch 22, batch 37200, giga_loss[loss=0.2483, simple_loss=0.3228, pruned_loss=0.08694, over 29009.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3279, pruned_loss=0.08988, over 5706826.84 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3541, pruned_loss=0.1119, over 5688625.47 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3255, pruned_loss=0.08823, over 5708370.99 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:25:14,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=994931.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:25:22,192 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 37250, libri_loss[loss=0.4434, simple_loss=0.4805, pruned_loss=0.2031, over 27607.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.328, pruned_loss=0.09073, over 5702369.66 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3554, pruned_loss=0.1126, over 5690749.58 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3243, pruned_loss=0.08829, over 5701992.83 frames. ], batch size: 115, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:26:38,168 INFO [train.py:968] (1/2) Epoch 22, batch 37300, giga_loss[loss=0.2386, simple_loss=0.3143, pruned_loss=0.08139, over 28808.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3265, pruned_loss=0.08972, over 5716163.93 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3558, pruned_loss=0.1125, over 5698347.71 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3222, pruned_loss=0.08709, over 5709402.64 frames. ], batch size: 285, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:26:40,346 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:26:54,652 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3885, 1.5325, 1.3091, 1.4982], device='cuda:1'), covar=tensor([0.0731, 0.0432, 0.0353, 0.0778], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 15:27:16,317 INFO [train.py:968] (1/2) Epoch 22, batch 37350, libri_loss[loss=0.2632, simple_loss=0.3432, pruned_loss=0.09159, over 29583.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3244, pruned_loss=0.08841, over 5722935.72 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3558, pruned_loss=0.1122, over 5702816.84 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3198, pruned_loss=0.08569, over 5714194.28 frames. ], batch size: 75, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:27:17,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8597, 1.2814, 2.8451, 2.7026], device='cuda:1'), covar=tensor([0.1716, 0.2460, 0.0587, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0645, 0.0957, 0.0900], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 15:27:29,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3118, 4.1457, 3.9822, 1.8429], device='cuda:1'), covar=tensor([0.0726, 0.0841, 0.0916, 0.1921], device='cuda:1'), in_proj_covar=tensor([0.1213, 0.1122, 0.0943, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 15:27:58,086 INFO [train.py:968] (1/2) Epoch 22, batch 37400, libri_loss[loss=0.27, simple_loss=0.3533, pruned_loss=0.0933, over 28856.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3231, pruned_loss=0.08747, over 5729174.89 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3563, pruned_loss=0.1124, over 5704891.34 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3178, pruned_loss=0.08427, over 5720920.15 frames. ], batch size: 107, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:27:58,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=995132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:28:04,995 INFO [optim.py:369] (1/2) 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 15:28:31,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3153, 3.5293, 1.5135, 1.4456], device='cuda:1'), covar=tensor([0.1001, 0.0365, 0.0898, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0549, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 15:28:36,088 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=995177.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:28:39,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1791, 1.3861, 1.5148, 1.4084], device='cuda:1'), covar=tensor([0.1598, 0.1178, 0.1781, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0753, 0.0718, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 15:28:40,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=995180.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:28:40,915 INFO [train.py:968] (1/2) Epoch 22, batch 37450, giga_loss[loss=0.2223, simple_loss=0.3014, pruned_loss=0.07162, over 29059.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3224, pruned_loss=0.08709, over 5718346.55 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3567, pruned_loss=0.1124, over 5698998.07 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3169, pruned_loss=0.08383, over 5717981.98 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:28:51,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-11 15:29:02,015 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=995209.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:29:21,826 INFO [train.py:968] (1/2) Epoch 22, batch 37500, giga_loss[loss=0.2462, simple_loss=0.317, pruned_loss=0.0877, over 28702.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3267, pruned_loss=0.08979, over 5701403.63 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3575, pruned_loss=0.1126, over 5685059.49 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3204, pruned_loss=0.08604, over 5714899.98 frames. ], batch size: 92, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:29:28,368 INFO [optim.py:369] (1/2) 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 15:29:35,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-11 15:29:41,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3434, 1.6197, 1.3413, 1.0666], device='cuda:1'), covar=tensor([0.2047, 0.2047, 0.2193, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1519, 0.1099, 0.1340, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 15:29:41,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7964, 2.1493, 1.5335, 1.7236], device='cuda:1'), covar=tensor([0.1064, 0.0662, 0.1006, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0445, 0.0522, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 15:30:04,474 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=995275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:30:06,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=995278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:30:08,134 INFO [train.py:968] (1/2) Epoch 22, batch 37550, giga_loss[loss=0.315, simple_loss=0.3807, pruned_loss=0.1246, over 28280.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3304, pruned_loss=0.09184, over 5700910.31 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3577, pruned_loss=0.1126, over 5687520.77 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3248, pruned_loss=0.08869, over 5709447.97 frames. ], batch size: 368, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:30:37,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=995307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:31:04,457 INFO [train.py:968] (1/2) Epoch 22, batch 37600, giga_loss[loss=0.3058, simple_loss=0.3705, pruned_loss=0.1205, over 29026.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3385, pruned_loss=0.09738, over 5693528.03 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3577, pruned_loss=0.1126, over 5688644.16 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.334, pruned_loss=0.09485, over 5699294.18 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:31:11,326 INFO [optim.py:369] (1/2) 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 15:31:50,817 INFO [train.py:968] (1/2) Epoch 22, batch 37650, giga_loss[loss=0.2988, simple_loss=0.3786, pruned_loss=0.1094, over 28696.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3444, pruned_loss=0.101, over 5681116.85 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3574, pruned_loss=0.1122, over 5694281.28 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3404, pruned_loss=0.09878, over 5680986.48 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:32:43,051 INFO [train.py:968] (1/2) Epoch 22, batch 37700, giga_loss[loss=0.3032, simple_loss=0.3802, pruned_loss=0.1131, over 28629.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3479, pruned_loss=0.1021, over 5674883.04 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3577, pruned_loss=0.1124, over 5687527.83 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3443, pruned_loss=0.1, over 5679818.83 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:32:51,042 INFO [optim.py:369] (1/2) 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 15:33:00,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7931, 4.6845, 2.0218, 1.9542], device='cuda:1'), covar=tensor([0.0979, 0.0194, 0.0845, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0550, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 15:33:21,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4921, 4.3532, 1.7876, 1.6575], device='cuda:1'), covar=tensor([0.1065, 0.0260, 0.0915, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0551, 0.0385, 0.0426], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0029], device='cuda:1')
+2023-03-11 15:33:27,201 INFO [train.py:968] (1/2) Epoch 22, batch 37750, giga_loss[loss=0.2603, simple_loss=0.345, pruned_loss=0.08778, over 28587.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3536, pruned_loss=0.1054, over 5674932.82 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3586, pruned_loss=0.1131, over 5695982.25 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3495, pruned_loss=0.1026, over 5670995.65 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:33:52,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.66 vs. limit=5.0
+2023-03-11 15:34:12,491 INFO [train.py:968] (1/2) Epoch 22, batch 37800, giga_loss[loss=0.312, simple_loss=0.383, pruned_loss=0.1205, over 28538.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.359, pruned_loss=0.1086, over 5680012.19 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3589, pruned_loss=0.1133, over 5700216.87 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.3554, pruned_loss=0.1061, over 5672667.11 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:34:20,425 INFO [optim.py:369] (1/2) 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 15:34:57,286 INFO [train.py:968] (1/2) Epoch 22, batch 37850, giga_loss[loss=0.2397, simple_loss=0.337, pruned_loss=0.07121, over 28931.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3538, pruned_loss=0.1045, over 5684993.77 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3588, pruned_loss=0.1134, over 5703065.44 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3511, pruned_loss=0.1024, over 5676648.32 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:35:37,878 INFO [train.py:968] (1/2) Epoch 22, batch 37900, giga_loss[loss=0.3027, simple_loss=0.3675, pruned_loss=0.119, over 28211.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3522, pruned_loss=0.1029, over 5671204.13 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3593, pruned_loss=0.1138, over 5685951.69 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3493, pruned_loss=0.1006, over 5678305.39 frames. ], batch size: 77, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:35:41,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-11 15:35:50,604 INFO [optim.py:369] (1/2) 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 15:36:01,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8827, 1.1390, 1.3155, 0.9871], device='cuda:1'), covar=tensor([0.1951, 0.1464, 0.2219, 0.1729], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0751, 0.0716, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 15:36:26,259 INFO [train.py:968] (1/2) Epoch 22, batch 37950, giga_loss[loss=0.2907, simple_loss=0.3647, pruned_loss=0.1084, over 27565.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3509, pruned_loss=0.1017, over 5679394.63 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3587, pruned_loss=0.1134, over 5690317.78 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.349, pruned_loss=0.0999, over 5680814.38 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:36:38,190 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=995696.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:36:39,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 15:36:51,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5526, 4.4071, 4.1273, 2.1847], device='cuda:1'), covar=tensor([0.0574, 0.0713, 0.0726, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.1133, 0.0950, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 15:37:08,925 INFO [train.py:968] (1/2) Epoch 22, batch 38000, giga_loss[loss=0.3148, simple_loss=0.3716, pruned_loss=0.129, over 26719.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3504, pruned_loss=0.1012, over 5687322.82 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.358, pruned_loss=0.1131, over 5694620.08 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3493, pruned_loss=0.09973, over 5684394.03 frames. ], batch size: 555, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:37:11,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=995734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:37:18,214 INFO [optim.py:369] (1/2) 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 15:37:43,097 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=995768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:37:54,145 INFO [train.py:968] (1/2) Epoch 22, batch 38050, giga_loss[loss=0.2716, simple_loss=0.3496, pruned_loss=0.09683, over 29067.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3525, pruned_loss=0.1021, over 5686656.59 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3582, pruned_loss=0.113, over 5697189.48 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3513, pruned_loss=0.1008, over 5681918.05 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:38:22,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4850, 1.8652, 1.4707, 1.6042], device='cuda:1'), covar=tensor([0.2772, 0.2734, 0.3167, 0.2392], device='cuda:1'), in_proj_covar=tensor([0.1517, 0.1100, 0.1338, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 15:38:41,590 INFO [train.py:968] (1/2) Epoch 22, batch 38100, giga_loss[loss=0.3063, simple_loss=0.3741, pruned_loss=0.1192, over 28902.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3545, pruned_loss=0.1037, over 5678143.19 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3585, pruned_loss=0.1131, over 5689244.43 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3532, pruned_loss=0.1024, over 5681321.84 frames. ], batch size: 145, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:38:51,994 INFO [optim.py:369] (1/2) 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 15:39:01,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 15:39:25,369 INFO [train.py:968] (1/2) Epoch 22, batch 38150, giga_loss[loss=0.2918, simple_loss=0.3708, pruned_loss=0.1064, over 28993.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3563, pruned_loss=0.105, over 5684191.47 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3589, pruned_loss=0.1132, over 5685407.45 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3548, pruned_loss=0.1036, over 5690623.09 frames. ], batch size: 164, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:40:14,805 INFO [train.py:968] (1/2) Epoch 22, batch 38200, giga_loss[loss=0.2747, simple_loss=0.3452, pruned_loss=0.1021, over 28624.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3568, pruned_loss=0.1058, over 5681049.75 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.359, pruned_loss=0.1131, over 5687957.44 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3554, pruned_loss=0.1047, over 5683793.52 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:40:24,247 INFO [optim.py:369] (1/2) 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 15:40:44,688 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-11 15:40:49,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-11 15:41:01,021 INFO [train.py:968] (1/2) Epoch 22, batch 38250, giga_loss[loss=0.2673, simple_loss=0.3404, pruned_loss=0.09711, over 28452.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3576, pruned_loss=0.1063, over 5691877.35 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3593, pruned_loss=0.1132, over 5689720.29 frames. ], giga_tot_loss[loss=0.2833, simple_loss=0.3562, pruned_loss=0.1052, over 5692546.03 frames. ], batch size: 71, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:41:42,774 INFO [train.py:968] (1/2) Epoch 22, batch 38300, giga_loss[loss=0.2911, simple_loss=0.3619, pruned_loss=0.1101, over 28573.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3577, pruned_loss=0.1054, over 5698950.14 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3598, pruned_loss=0.1134, over 5690023.90 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3562, pruned_loss=0.1043, over 5699716.75 frames. ], batch size: 85, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:41:53,964 INFO [optim.py:369] (1/2) 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 15:42:20,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996071.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:42:28,726 INFO [train.py:968] (1/2) Epoch 22, batch 38350, giga_loss[loss=0.273, simple_loss=0.3556, pruned_loss=0.09523, over 28822.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3577, pruned_loss=0.1042, over 5699877.32 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3602, pruned_loss=0.1137, over 5689954.59 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3562, pruned_loss=0.1029, over 5700613.24 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:42:35,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4411, 2.0942, 1.7236, 1.6966], device='cuda:1'), covar=tensor([0.0830, 0.0270, 0.0317, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 15:42:52,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:43:11,093 INFO [train.py:968] (1/2) Epoch 22, batch 38400, giga_loss[loss=0.2551, simple_loss=0.3413, pruned_loss=0.08448, over 28818.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3561, pruned_loss=0.1029, over 5702182.44 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3601, pruned_loss=0.1138, over 5688401.84 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.355, pruned_loss=0.1017, over 5704358.52 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 15:43:21,296 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-11 15:43:22,063 INFO [optim.py:369] (1/2) 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 15:43:23,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:43:53,533 INFO [train.py:968] (1/2) Epoch 22, batch 38450, giga_loss[loss=0.2971, simple_loss=0.3664, pruned_loss=0.1139, over 28835.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3545, pruned_loss=0.1026, over 5700965.59 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 5689686.42 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3533, pruned_loss=0.1012, over 5702378.18 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:44:05,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3330, 2.8049, 2.5115, 2.0294], device='cuda:1'), covar=tensor([0.2462, 0.1558, 0.1756, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.1953, 0.1888, 0.1817, 0.1974], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:44:16,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-11 15:44:21,309 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996214.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:44:24,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996217.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:44:36,602 INFO [train.py:968] (1/2) Epoch 22, batch 38500, giga_loss[loss=0.2658, simple_loss=0.3484, pruned_loss=0.09157, over 28939.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3521, pruned_loss=0.1011, over 5709931.66 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3602, pruned_loss=0.1138, over 5692641.26 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3511, pruned_loss=0.09998, over 5708614.51 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:44:46,435 INFO [optim.py:369] (1/2) 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 15:44:48,758 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996246.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 15:44:49,331 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=996247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:44:53,021 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:44:55,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:15,961 INFO [train.py:968] (1/2) Epoch 22, batch 38550, giga_loss[loss=0.264, simple_loss=0.335, pruned_loss=0.09649, over 28496.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3497, pruned_loss=0.1, over 5707268.05 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5688572.06 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09899, over 5710639.01 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:45:19,682 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:20,976 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:23,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:51,088 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:45:55,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1385, 1.1241, 3.8375, 3.2582], device='cuda:1'), covar=tensor([0.1859, 0.3045, 0.0448, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0644, 0.0958, 0.0904], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 15:46:00,875 INFO [train.py:968] (1/2) Epoch 22, batch 38600, giga_loss[loss=0.3089, simple_loss=0.375, pruned_loss=0.1214, over 28920.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3511, pruned_loss=0.1019, over 5708916.48 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1135, over 5690240.91 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3504, pruned_loss=0.1007, over 5710591.96 frames. ], batch size: 186, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:46:12,254 INFO [optim.py:369] (1/2) 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 15:46:43,285 INFO [train.py:968] (1/2) Epoch 22, batch 38650, giga_loss[loss=0.3044, simple_loss=0.3728, pruned_loss=0.118, over 28929.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3517, pruned_loss=0.1021, over 5710537.04 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3602, pruned_loss=0.1138, over 5693935.18 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3503, pruned_loss=0.1006, over 5708897.73 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:46:46,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-11 15:47:09,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8752, 3.7228, 3.4832, 1.9739], device='cuda:1'), covar=tensor([0.0593, 0.0729, 0.0705, 0.2405], device='cuda:1'), in_proj_covar=tensor([0.1215, 0.1130, 0.0951, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 15:47:20,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6877, 1.7231, 1.7397, 1.5937], device='cuda:1'), covar=tensor([0.2945, 0.2609, 0.2019, 0.2493], device='cuda:1'), in_proj_covar=tensor([0.1965, 0.1899, 0.1830, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:47:22,883 INFO [train.py:968] (1/2) Epoch 22, batch 38700, giga_loss[loss=0.2833, simple_loss=0.3619, pruned_loss=0.1023, over 27863.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3505, pruned_loss=0.1004, over 5712741.69 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3599, pruned_loss=0.1136, over 5697999.45 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3495, pruned_loss=0.09912, over 5708292.04 frames. ], batch size: 412, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:47:33,610 INFO [optim.py:369] (1/2) 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 15:47:36,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5467, 1.7240, 1.5307, 1.3411], device='cuda:1'), covar=tensor([0.2932, 0.2644, 0.2382, 0.2715], device='cuda:1'), in_proj_covar=tensor([0.1966, 0.1900, 0.1832, 0.1987], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:48:01,925 INFO [train.py:968] (1/2) Epoch 22, batch 38750, giga_loss[loss=0.2981, simple_loss=0.3748, pruned_loss=0.1108, over 28668.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09971, over 5718758.55 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3604, pruned_loss=0.1139, over 5704243.74 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3483, pruned_loss=0.09796, over 5710192.41 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:48:13,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=996495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:48:42,276 INFO [train.py:968] (1/2) Epoch 22, batch 38800, giga_loss[loss=0.2761, simple_loss=0.3546, pruned_loss=0.09883, over 29080.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3481, pruned_loss=0.09907, over 5715385.62 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3595, pruned_loss=0.1134, over 5701120.93 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.09733, over 5712127.06 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:48:54,591 INFO [optim.py:369] (1/2) 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 15:49:23,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6499, 1.7463, 1.5282, 1.6514], device='cuda:1'), covar=tensor([0.0748, 0.0310, 0.0315, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 15:49:23,935 INFO [train.py:968] (1/2) Epoch 22, batch 38850, giga_loss[loss=0.2224, simple_loss=0.3019, pruned_loss=0.07146, over 28493.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3463, pruned_loss=0.0986, over 5716012.07 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5705020.03 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.09663, over 5710156.38 frames. ], batch size: 60, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:49:57,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:50:03,647 INFO [train.py:968] (1/2) Epoch 22, batch 38900, giga_loss[loss=0.2397, simple_loss=0.3222, pruned_loss=0.07862, over 29085.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.344, pruned_loss=0.0977, over 5702379.38 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3604, pruned_loss=0.1144, over 5700198.36 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.342, pruned_loss=0.09523, over 5703011.00 frames. ], batch size: 155, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:50:14,871 INFO [optim.py:369] (1/2) 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 15:50:43,284 INFO [train.py:968] (1/2) Epoch 22, batch 38950, giga_loss[loss=0.2599, simple_loss=0.3225, pruned_loss=0.09869, over 23396.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3422, pruned_loss=0.09696, over 5705469.80 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3606, pruned_loss=0.1144, over 5703276.04 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3398, pruned_loss=0.09432, over 5703570.63 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:51:29,014 INFO [train.py:968] (1/2) Epoch 22, batch 39000, giga_loss[loss=0.269, simple_loss=0.34, pruned_loss=0.099, over 28987.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3419, pruned_loss=0.09656, over 5706720.97 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3609, pruned_loss=0.1145, over 5706102.21 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3395, pruned_loss=0.09415, over 5702762.49 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:51:29,014 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 15:51:37,978 INFO [train.py:1012] (1/2) Epoch 22, validation: loss=0.2064, simple_loss=0.3137, pruned_loss=0.04954, over 944034.00 frames. 
+2023-03-11 15:51:37,978 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 15:51:49,797 INFO [optim.py:369] (1/2) 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 15:52:00,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5667, 1.8016, 1.8168, 1.5660], device='cuda:1'), covar=tensor([0.2381, 0.2021, 0.1468, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.1963, 0.1898, 0.1825, 0.1980], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 15:52:01,982 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 15:52:05,956 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=996765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:52:07,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=996768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:52:18,176 INFO [train.py:968] (1/2) Epoch 22, batch 39050, libri_loss[loss=0.3431, simple_loss=0.3994, pruned_loss=0.1434, over 29674.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3411, pruned_loss=0.09664, over 5699202.16 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1148, over 5703418.39 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.338, pruned_loss=0.09386, over 5698832.71 frames. ], batch size: 91, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:52:32,286 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=996797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:53:02,072 INFO [train.py:968] (1/2) Epoch 22, batch 39100, giga_loss[loss=0.2502, simple_loss=0.3299, pruned_loss=0.08519, over 28620.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3385, pruned_loss=0.09534, over 5708228.83 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3617, pruned_loss=0.1149, over 5707550.20 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3352, pruned_loss=0.09245, over 5704197.02 frames. ], batch size: 336, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:53:12,295 INFO [optim.py:369] (1/2) 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 15:53:35,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=996870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:53:44,875 INFO [train.py:968] (1/2) Epoch 22, batch 39150, giga_loss[loss=0.3331, simple_loss=0.3835, pruned_loss=0.1414, over 23723.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3354, pruned_loss=0.0938, over 5710064.93 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3617, pruned_loss=0.1149, over 5707550.20 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3329, pruned_loss=0.09155, over 5706926.92 frames. ], batch size: 705, lr: 1.44e-03, grad_scale: 2.0
+2023-03-11 15:54:17,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-11 15:54:22,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2538, 1.9514, 1.5282, 0.5511], device='cuda:1'), covar=tensor([0.6180, 0.3413, 0.4431, 0.6778], device='cuda:1'), in_proj_covar=tensor([0.1740, 0.1640, 0.1587, 0.1421], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 15:54:24,638 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=996927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:54:26,996 INFO [train.py:968] (1/2) Epoch 22, batch 39200, libri_loss[loss=0.3104, simple_loss=0.3802, pruned_loss=0.1203, over 28490.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3355, pruned_loss=0.09399, over 5706687.26 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.1151, over 5707355.14 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3318, pruned_loss=0.09113, over 5703942.45 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:54:36,770 INFO [optim.py:369] (1/2) 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 15:55:09,551 INFO [train.py:968] (1/2) Epoch 22, batch 39250, giga_loss[loss=0.2816, simple_loss=0.3675, pruned_loss=0.09786, over 28577.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3357, pruned_loss=0.09407, over 5698746.29 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.115, over 5700785.76 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3322, pruned_loss=0.09129, over 5702530.68 frames. ], batch size: 307, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:55:40,899 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:55:41,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-11 15:55:42,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:55:55,841 INFO [train.py:968] (1/2) Epoch 22, batch 39300, giga_loss[loss=0.2881, simple_loss=0.3659, pruned_loss=0.1051, over 28770.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3389, pruned_loss=0.09547, over 5701886.67 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.115, over 5707926.32 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3352, pruned_loss=0.09256, over 5698234.61 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:56:06,825 INFO [optim.py:369] (1/2) 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 15:56:07,080 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:56:37,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.28 vs. limit=5.0
+2023-03-11 15:56:37,617 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=997076.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 15:56:41,967 INFO [train.py:968] (1/2) Epoch 22, batch 39350, giga_loss[loss=0.2545, simple_loss=0.3404, pruned_loss=0.08432, over 28803.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3417, pruned_loss=0.0965, over 5695154.15 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.115, over 5707667.95 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3384, pruned_loss=0.09385, over 5692132.73 frames. ], batch size: 243, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:56:56,189 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-11 15:57:16,376 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-11 15:57:25,743 INFO [train.py:968] (1/2) Epoch 22, batch 39400, giga_loss[loss=0.2425, simple_loss=0.3315, pruned_loss=0.07674, over 28719.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3443, pruned_loss=0.0972, over 5693081.50 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.115, over 5703014.26 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3412, pruned_loss=0.09464, over 5694241.57 frames. ], batch size: 284, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:57:39,169 INFO [optim.py:369] (1/2) 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 15:57:58,852 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=997168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 15:58:08,186 INFO [train.py:968] (1/2) Epoch 22, batch 39450, giga_loss[loss=0.2475, simple_loss=0.3249, pruned_loss=0.08501, over 28831.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3446, pruned_loss=0.0972, over 5688721.24 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3619, pruned_loss=0.1149, over 5699844.12 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3414, pruned_loss=0.09427, over 5692527.57 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:58:27,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2016, 4.0169, 3.7963, 1.9368], device='cuda:1'), covar=tensor([0.0664, 0.0791, 0.0772, 0.1989], device='cuda:1'), in_proj_covar=tensor([0.1216, 0.1130, 0.0951, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 15:58:53,119 INFO [train.py:968] (1/2) Epoch 22, batch 39500, giga_loss[loss=0.235, simple_loss=0.3122, pruned_loss=0.07887, over 28949.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3419, pruned_loss=0.09543, over 5696295.16 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3619, pruned_loss=0.1149, over 5699844.12 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3394, pruned_loss=0.09314, over 5699257.67 frames. ], batch size: 106, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:59:04,435 INFO [optim.py:369] (1/2) 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 15:59:15,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1509, 1.1789, 1.1096, 0.9249], device='cuda:1'), covar=tensor([0.0776, 0.0412, 0.0884, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0392, 0.0442, 0.0519, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 15:59:20,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5088, 1.5699, 1.7032, 1.3125], device='cuda:1'), covar=tensor([0.1749, 0.2390, 0.1495, 0.1755], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0700, 0.0950, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 15:59:37,043 INFO [train.py:968] (1/2) Epoch 22, batch 39550, libri_loss[loss=0.2816, simple_loss=0.357, pruned_loss=0.1031, over 29531.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3414, pruned_loss=0.09502, over 5699801.93 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3616, pruned_loss=0.1146, over 5702093.31 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3391, pruned_loss=0.09291, over 5700059.12 frames. ], batch size: 82, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 15:59:40,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3074, 1.5067, 1.3919, 1.5507], device='cuda:1'), covar=tensor([0.0767, 0.0340, 0.0339, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 15:59:55,159 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 39600, giga_loss[loss=0.306, simple_loss=0.3709, pruned_loss=0.1206, over 27611.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3425, pruned_loss=0.09605, over 5713564.08 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3623, pruned_loss=0.1149, over 5705795.04 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09372, over 5710403.11 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 16:00:31,206 INFO [optim.py:369] (1/2) 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 16:00:51,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0376, 2.2619, 1.6493, 1.8656], device='cuda:1'), covar=tensor([0.0966, 0.0684, 0.1022, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.0391, 0.0442, 0.0517, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 16:00:58,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5000, 2.1052, 1.5383, 0.6063], device='cuda:1'), covar=tensor([0.6213, 0.2842, 0.4002, 0.7148], device='cuda:1'), in_proj_covar=tensor([0.1739, 0.1635, 0.1588, 0.1419], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 16:01:02,878 INFO [train.py:968] (1/2) Epoch 22, batch 39650, giga_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 29053.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3461, pruned_loss=0.09799, over 5713615.39 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3625, pruned_loss=0.1151, over 5709871.08 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3433, pruned_loss=0.09571, over 5707655.87 frames. ], batch size: 136, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:01:45,048 INFO [train.py:968] (1/2) Epoch 22, batch 39700, giga_loss[loss=0.2703, simple_loss=0.3505, pruned_loss=0.09506, over 29047.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3491, pruned_loss=0.09936, over 5711022.01 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3628, pruned_loss=0.1152, over 5711780.83 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.346, pruned_loss=0.09678, over 5704561.61 frames. ], batch size: 128, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:01:55,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:01:55,541 INFO [optim.py:369] (1/2) 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 16:01:57,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:02:00,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=997451.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:02:21,388 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:02:24,370 INFO [train.py:968] (1/2) Epoch 22, batch 39750, giga_loss[loss=0.2804, simple_loss=0.3631, pruned_loss=0.09886, over 28626.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3507, pruned_loss=0.1003, over 5711276.01 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3626, pruned_loss=0.115, over 5712266.54 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3481, pruned_loss=0.09803, over 5705824.68 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:03:06,534 INFO [train.py:968] (1/2) Epoch 22, batch 39800, giga_loss[loss=0.2693, simple_loss=0.3533, pruned_loss=0.09262, over 28732.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3533, pruned_loss=0.102, over 5706213.59 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3628, pruned_loss=0.1149, over 5707451.70 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.09991, over 5705642.97 frames. ], batch size: 242, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:03:16,740 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:03:49,169 INFO [train.py:968] (1/2) Epoch 22, batch 39850, giga_loss[loss=0.2752, simple_loss=0.3421, pruned_loss=0.1041, over 28729.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.354, pruned_loss=0.1022, over 5709394.43 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3625, pruned_loss=0.1147, over 5711409.49 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.352, pruned_loss=0.1005, over 5705567.98 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:03:59,768 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997594.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:04:01,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997597.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:04:14,199 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4069, 1.2883, 1.1720, 1.5017], device='cuda:1'), covar=tensor([0.0693, 0.0322, 0.0335, 0.0834], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 16:04:26,251 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997626.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:04:29,315 INFO [train.py:968] (1/2) Epoch 22, batch 39900, giga_loss[loss=0.2707, simple_loss=0.3413, pruned_loss=0.1001, over 28772.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3534, pruned_loss=0.1021, over 5713868.78 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3625, pruned_loss=0.1147, over 5714709.50 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3514, pruned_loss=0.1002, over 5707519.72 frames. ], batch size: 99, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:04:37,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 16:04:42,706 INFO [optim.py:369] (1/2) 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 16:05:09,885 INFO [train.py:968] (1/2) Epoch 22, batch 39950, giga_loss[loss=0.2864, simple_loss=0.3544, pruned_loss=0.1092, over 27569.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3517, pruned_loss=0.1016, over 5710990.39 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5709487.72 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09972, over 5709722.82 frames. ], batch size: 472, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:05:10,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-11 16:05:13,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=997686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:05:15,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=997689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:05:41,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=997718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:05:51,422 INFO [train.py:968] (1/2) Epoch 22, batch 40000, giga_loss[loss=0.2145, simple_loss=0.301, pruned_loss=0.06401, over 28895.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3484, pruned_loss=0.1002, over 5704729.80 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.363, pruned_loss=0.1153, over 5702088.72 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3462, pruned_loss=0.09796, over 5710831.68 frames. ], batch size: 174, lr: 1.44e-03, grad_scale: 8.0
+2023-03-11 16:06:04,338 INFO [optim.py:369] (1/2) 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 16:06:04,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-11 16:06:07,742 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 40050, giga_loss[loss=0.2858, simple_loss=0.3636, pruned_loss=0.104, over 28645.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3457, pruned_loss=0.09872, over 5705042.34 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3629, pruned_loss=0.1152, over 5700553.18 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3435, pruned_loss=0.09643, over 5711241.13 frames. ], batch size: 262, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:06:57,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-11 16:07:06,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0084, 2.1248, 2.0333, 1.7873], device='cuda:1'), covar=tensor([0.2132, 0.2942, 0.2421, 0.2913], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0750, 0.0716, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 16:07:13,989 INFO [train.py:968] (1/2) Epoch 22, batch 40100, giga_loss[loss=0.2799, simple_loss=0.3673, pruned_loss=0.09624, over 28926.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3457, pruned_loss=0.09712, over 5712192.98 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3629, pruned_loss=0.1151, over 5702651.74 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3436, pruned_loss=0.09495, over 5715302.82 frames. ], batch size: 213, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:07:27,730 INFO [optim.py:369] (1/2) 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 16:07:56,994 INFO [train.py:968] (1/2) Epoch 22, batch 40150, giga_loss[loss=0.2363, simple_loss=0.3225, pruned_loss=0.07499, over 28801.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3471, pruned_loss=0.09686, over 5703130.54 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.363, pruned_loss=0.1152, over 5702307.06 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3451, pruned_loss=0.0948, over 5706107.15 frames. ], batch size: 119, lr: 1.44e-03, grad_scale: 4.0
+2023-03-11 16:08:38,288 INFO [train.py:968] (1/2) Epoch 22, batch 40200, giga_loss[loss=0.2779, simple_loss=0.3507, pruned_loss=0.1026, over 28747.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3464, pruned_loss=0.09707, over 5707178.65 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3632, pruned_loss=0.1153, over 5703910.10 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3443, pruned_loss=0.09506, over 5708092.41 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:08:52,455 INFO [optim.py:369] (1/2) 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 16:08:54,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0021, 2.4386, 2.2732, 1.8419], device='cuda:1'), covar=tensor([0.3821, 0.2422, 0.2625, 0.3191], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1904, 0.1836, 0.1978], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 16:09:20,721 INFO [train.py:968] (1/2) Epoch 22, batch 40250, giga_loss[loss=0.2383, simple_loss=0.3102, pruned_loss=0.08319, over 28840.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3465, pruned_loss=0.09857, over 5706198.83 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3638, pruned_loss=0.1158, over 5705175.04 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3439, pruned_loss=0.0961, over 5705775.19 frames. ], batch size: 119, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:10:02,960 INFO [train.py:968] (1/2) Epoch 22, batch 40300, giga_loss[loss=0.2568, simple_loss=0.3248, pruned_loss=0.09436, over 29035.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3445, pruned_loss=0.09883, over 5705166.52 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3635, pruned_loss=0.1157, over 5706683.14 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3424, pruned_loss=0.0967, over 5703395.56 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:10:19,630 INFO [optim.py:369] (1/2) 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 16:10:38,966 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-11 16:10:45,560 INFO [train.py:968] (1/2) Epoch 22, batch 40350, giga_loss[loss=0.2581, simple_loss=0.329, pruned_loss=0.09358, over 28931.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3425, pruned_loss=0.09824, over 5706409.58 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1156, over 5701370.35 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3404, pruned_loss=0.09627, over 5710500.12 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:11:21,824 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 40400, libri_loss[loss=0.3139, simple_loss=0.3841, pruned_loss=0.1219, over 28711.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3407, pruned_loss=0.09736, over 5707232.65 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5695450.32 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3383, pruned_loss=0.09529, over 5715661.66 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:11:40,607 INFO [optim.py:369] (1/2) 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 16:12:07,678 INFO [train.py:968] (1/2) Epoch 22, batch 40450, giga_loss[loss=0.2271, simple_loss=0.3055, pruned_loss=0.07432, over 28907.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3372, pruned_loss=0.09522, over 5711591.66 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1156, over 5698846.59 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3348, pruned_loss=0.09313, over 5715511.57 frames. ], batch size: 112, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:12:50,400 INFO [train.py:968] (1/2) Epoch 22, batch 40500, giga_loss[loss=0.2406, simple_loss=0.3119, pruned_loss=0.08462, over 29074.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3339, pruned_loss=0.09375, over 5712571.21 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.364, pruned_loss=0.1159, over 5698894.52 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3314, pruned_loss=0.09165, over 5715749.92 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:13:02,629 INFO [optim.py:369] (1/2) 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 16:13:20,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=998269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:13:23,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=998272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:13:28,995 INFO [train.py:968] (1/2) Epoch 22, batch 40550, giga_loss[loss=0.2429, simple_loss=0.3212, pruned_loss=0.08232, over 28692.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3307, pruned_loss=0.09219, over 5703332.86 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3639, pruned_loss=0.1159, over 5684558.67 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3277, pruned_loss=0.08981, over 5718251.54 frames. ], batch size: 242, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:13:35,169 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 16:13:45,140 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=998301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:14:09,941 INFO [train.py:968] (1/2) Epoch 22, batch 40600, giga_loss[loss=0.2631, simple_loss=0.3298, pruned_loss=0.09819, over 24031.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3315, pruned_loss=0.09239, over 5699856.19 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3634, pruned_loss=0.1155, over 5689039.87 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3287, pruned_loss=0.09023, over 5708479.87 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:14:23,333 INFO [optim.py:369] (1/2) 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 16:14:51,899 INFO [train.py:968] (1/2) Epoch 22, batch 40650, giga_loss[loss=0.308, simple_loss=0.3651, pruned_loss=0.1254, over 24051.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3347, pruned_loss=0.0936, over 5706967.46 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1156, over 5694474.26 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3317, pruned_loss=0.09125, over 5709419.98 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:15:33,911 INFO [train.py:968] (1/2) Epoch 22, batch 40700, giga_loss[loss=0.2585, simple_loss=0.3274, pruned_loss=0.09482, over 28721.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3384, pruned_loss=0.09509, over 5709454.81 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3632, pruned_loss=0.1155, over 5700236.59 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3351, pruned_loss=0.09248, over 5707015.29 frames. ], batch size: 99, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:15:43,845 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=998444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:15:47,324 INFO [optim.py:369] (1/2) 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 16:15:58,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4520, 1.8204, 1.6956, 1.6608], device='cuda:1'), covar=tensor([0.1987, 0.2046, 0.2243, 0.2085], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0750, 0.0716, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 16:16:02,528 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-11 16:16:14,462 INFO [train.py:968] (1/2) Epoch 22, batch 40750, giga_loss[loss=0.2697, simple_loss=0.3501, pruned_loss=0.09468, over 28924.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3426, pruned_loss=0.0969, over 5717143.03 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3638, pruned_loss=0.1158, over 5702642.78 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3391, pruned_loss=0.09415, over 5713297.99 frames. ], batch size: 66, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:16:40,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5479, 1.8334, 1.5079, 1.5407], device='cuda:1'), covar=tensor([0.0724, 0.0287, 0.0326, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 16:16:57,107 INFO [train.py:968] (1/2) Epoch 22, batch 40800, giga_loss[loss=0.2745, simple_loss=0.3533, pruned_loss=0.09789, over 28799.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3455, pruned_loss=0.09842, over 5709424.91 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5695411.76 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3418, pruned_loss=0.09553, over 5713434.71 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:17:13,451 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 40850, libri_loss[loss=0.3096, simple_loss=0.3706, pruned_loss=0.1243, over 29145.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3469, pruned_loss=0.09926, over 5707575.75 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3638, pruned_loss=0.1158, over 5698922.60 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3439, pruned_loss=0.09681, over 5707662.12 frames. ], batch size: 101, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:18:31,400 INFO [train.py:968] (1/2) Epoch 22, batch 40900, giga_loss[loss=0.324, simple_loss=0.384, pruned_loss=0.132, over 28886.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3524, pruned_loss=0.1047, over 5692583.21 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3631, pruned_loss=0.1153, over 5705625.70 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3499, pruned_loss=0.1025, over 5686526.76 frames. ], batch size: 199, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:18:47,689 INFO [optim.py:369] (1/2) 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 16:19:22,275 INFO [train.py:968] (1/2) Epoch 22, batch 40950, giga_loss[loss=0.3403, simple_loss=0.4038, pruned_loss=0.1384, over 29040.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3587, pruned_loss=0.1095, over 5686359.67 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.363, pruned_loss=0.1152, over 5707503.22 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3568, pruned_loss=0.1078, over 5679976.67 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:19:45,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5618, 1.8113, 1.4316, 1.7956], device='cuda:1'), covar=tensor([0.2530, 0.2518, 0.2850, 0.2147], device='cuda:1'), in_proj_covar=tensor([0.1515, 0.1094, 0.1336, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 16:19:48,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-11 16:20:07,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-11 16:20:13,073 INFO [train.py:968] (1/2) Epoch 22, batch 41000, giga_loss[loss=0.304, simple_loss=0.3701, pruned_loss=0.119, over 28887.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3663, pruned_loss=0.1152, over 5678582.61 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3632, pruned_loss=0.1154, over 5699364.31 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3647, pruned_loss=0.1136, over 5679817.12 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:20:28,306 INFO [optim.py:369] (1/2) 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 16:20:55,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-11 16:20:57,295 INFO [train.py:968] (1/2) Epoch 22, batch 41050, giga_loss[loss=0.3278, simple_loss=0.3903, pruned_loss=0.1326, over 28478.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3728, pruned_loss=0.1209, over 5676140.21 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3631, pruned_loss=0.1153, over 5703968.62 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3716, pruned_loss=0.1198, over 5672608.04 frames. ], batch size: 71, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:21:32,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=998819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:21:32,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.53 vs. limit=5.0
+2023-03-11 16:21:46,312 INFO [train.py:968] (1/2) Epoch 22, batch 41100, giga_loss[loss=0.3246, simple_loss=0.392, pruned_loss=0.1286, over 28825.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3795, pruned_loss=0.1265, over 5666663.49 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3634, pruned_loss=0.1155, over 5694993.93 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3786, pruned_loss=0.1256, over 5670954.91 frames. ], batch size: 199, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:22:04,566 INFO [optim.py:369] (1/2) 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 16:22:09,746 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=998851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:22:20,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4471, 1.6881, 1.6222, 1.2391], device='cuda:1'), covar=tensor([0.1561, 0.2639, 0.1427, 0.1775], device='cuda:1'), in_proj_covar=tensor([0.0897, 0.0698, 0.0944, 0.0844], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 16:22:39,222 INFO [train.py:968] (1/2) Epoch 22, batch 41150, giga_loss[loss=0.3076, simple_loss=0.3709, pruned_loss=0.1222, over 28977.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3815, pruned_loss=0.1288, over 5657439.59 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3633, pruned_loss=0.1154, over 5699666.35 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3812, pruned_loss=0.1284, over 5655780.92 frames. ], batch size: 213, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:23:34,270 INFO [train.py:968] (1/2) Epoch 22, batch 41200, giga_loss[loss=0.371, simple_loss=0.4167, pruned_loss=0.1627, over 28671.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3844, pruned_loss=0.1323, over 5645945.86 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3633, pruned_loss=0.1153, over 5702915.83 frames. ], giga_tot_loss[loss=0.3248, simple_loss=0.3847, pruned_loss=0.1324, over 5640706.50 frames. ], batch size: 284, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:23:53,883 INFO [optim.py:369] (1/2) 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 16:24:07,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=998962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:24:11,352 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 41250, giga_loss[loss=0.3304, simple_loss=0.3873, pruned_loss=0.1367, over 28677.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3867, pruned_loss=0.1349, over 5630493.60 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3634, pruned_loss=0.1153, over 5706850.39 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3872, pruned_loss=0.1353, over 5621934.37 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:24:41,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4404, 1.8486, 1.4387, 1.4677], device='cuda:1'), covar=tensor([0.2300, 0.2247, 0.2528, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1095, 0.1335, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 16:24:44,801 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=998994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:25:21,916 INFO [train.py:968] (1/2) Epoch 22, batch 41300, giga_loss[loss=0.3797, simple_loss=0.4174, pruned_loss=0.171, over 27559.00 frames. ], tot_loss[loss=0.3337, simple_loss=0.3901, pruned_loss=0.1386, over 5629873.56 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3629, pruned_loss=0.1151, over 5710297.47 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3916, pruned_loss=0.1396, over 5618173.06 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:25:42,133 INFO [optim.py:369] (1/2) 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 16:25:47,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5701, 1.7430, 1.6337, 1.4807], device='cuda:1'), covar=tensor([0.1760, 0.1847, 0.2111, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0748, 0.0714, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 16:25:55,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1800, 1.2772, 1.0984, 0.8987], device='cuda:1'), covar=tensor([0.0886, 0.0428, 0.1016, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0450, 0.0523, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 16:26:10,445 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-11 16:26:12,437 INFO [train.py:968] (1/2) Epoch 22, batch 41350, giga_loss[loss=0.3692, simple_loss=0.422, pruned_loss=0.1582, over 28821.00 frames. ], tot_loss[loss=0.3345, simple_loss=0.3911, pruned_loss=0.1389, over 5639712.47 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3631, pruned_loss=0.1151, over 5711182.46 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.3929, pruned_loss=0.1404, over 5627544.34 frames. ], batch size: 285, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:27:01,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-11 16:27:07,362 INFO [train.py:968] (1/2) Epoch 22, batch 41400, giga_loss[loss=0.3252, simple_loss=0.3749, pruned_loss=0.1378, over 28626.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3892, pruned_loss=0.1382, over 5637078.76 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3631, pruned_loss=0.1151, over 5704730.27 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3911, pruned_loss=0.1398, over 5631606.82 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:27:26,757 INFO [optim.py:369] (1/2) 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 16:27:58,662 INFO [train.py:968] (1/2) Epoch 22, batch 41450, giga_loss[loss=0.4454, simple_loss=0.46, pruned_loss=0.2154, over 26516.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3878, pruned_loss=0.1378, over 5624651.64 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3624, pruned_loss=0.1146, over 5700254.81 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3908, pruned_loss=0.1403, over 5621166.71 frames. ], batch size: 555, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:28:43,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=999226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:28:50,983 INFO [train.py:968] (1/2) Epoch 22, batch 41500, giga_loss[loss=0.264, simple_loss=0.3493, pruned_loss=0.08932, over 28986.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3875, pruned_loss=0.1369, over 5618183.96 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3628, pruned_loss=0.1149, over 5695865.01 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3901, pruned_loss=0.1392, over 5617845.39 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:29:11,010 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 41550, giga_loss[loss=0.341, simple_loss=0.4001, pruned_loss=0.1409, over 28563.00 frames. ], tot_loss[loss=0.3353, simple_loss=0.3911, pruned_loss=0.1397, over 5595035.19 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3629, pruned_loss=0.115, over 5678390.75 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3934, pruned_loss=0.1417, over 5609918.75 frames. ], batch size: 336, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:29:55,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1992, 1.6297, 1.2398, 0.4976], device='cuda:1'), covar=tensor([0.3235, 0.1967, 0.2650, 0.4813], device='cuda:1'), in_proj_covar=tensor([0.1763, 0.1663, 0.1605, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 16:30:07,331 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:30:38,845 INFO [train.py:968] (1/2) Epoch 22, batch 41600, giga_loss[loss=0.3147, simple_loss=0.3759, pruned_loss=0.1267, over 28894.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3901, pruned_loss=0.1389, over 5589661.59 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3628, pruned_loss=0.1149, over 5684128.88 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3927, pruned_loss=0.1411, over 5593909.71 frames. ], batch size: 112, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:31:03,799 INFO [optim.py:369] (1/2) 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 16:31:24,268 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=999369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:31:26,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=999372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:31:35,022 INFO [train.py:968] (1/2) Epoch 22, batch 41650, giga_loss[loss=0.3187, simple_loss=0.3892, pruned_loss=0.1242, over 28658.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.386, pruned_loss=0.1344, over 5597359.13 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3629, pruned_loss=0.115, over 5675511.24 frames. ], giga_tot_loss[loss=0.3302, simple_loss=0.3881, pruned_loss=0.1361, over 5608481.47 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:31:54,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=999401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:32:17,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5114, 1.8356, 1.4376, 1.5807], device='cuda:1'), covar=tensor([0.2759, 0.2794, 0.3234, 0.2421], device='cuda:1'), in_proj_covar=tensor([0.1512, 0.1094, 0.1335, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 16:32:21,714 INFO [train.py:968] (1/2) Epoch 22, batch 41700, giga_loss[loss=0.3458, simple_loss=0.3957, pruned_loss=0.1479, over 27698.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3832, pruned_loss=0.131, over 5610956.71 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3627, pruned_loss=0.1151, over 5670408.03 frames. ], giga_tot_loss[loss=0.3256, simple_loss=0.3856, pruned_loss=0.1328, over 5622835.23 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:32:43,997 INFO [optim.py:369] (1/2) 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 16:32:48,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3160, 2.2496, 1.3090, 1.4350], device='cuda:1'), covar=tensor([0.0828, 0.0414, 0.0770, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0557, 0.0387, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 16:33:16,036 INFO [train.py:968] (1/2) Epoch 22, batch 41750, giga_loss[loss=0.3537, simple_loss=0.3959, pruned_loss=0.1558, over 26562.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3802, pruned_loss=0.1285, over 5611697.54 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1148, over 5670521.87 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3826, pruned_loss=0.1304, over 5620503.16 frames. ], batch size: 555, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:34:05,558 INFO [train.py:968] (1/2) Epoch 22, batch 41800, libri_loss[loss=0.2459, simple_loss=0.3225, pruned_loss=0.08466, over 29559.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1252, over 5620291.03 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5677328.28 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3793, pruned_loss=0.1274, over 5619476.02 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:34:23,036 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:34:27,513 INFO [optim.py:369] (1/2) 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 16:34:56,297 INFO [train.py:968] (1/2) Epoch 22, batch 41850, giga_loss[loss=0.3333, simple_loss=0.3893, pruned_loss=0.1386, over 27644.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3742, pruned_loss=0.1237, over 5625279.48 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1145, over 5667865.60 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3771, pruned_loss=0.1258, over 5632103.06 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:35:13,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4180, 1.5284, 1.4801, 1.4218], device='cuda:1'), covar=tensor([0.2351, 0.2040, 0.2067, 0.2115], device='cuda:1'), in_proj_covar=tensor([0.1977, 0.1914, 0.1843, 0.1980], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 16:35:43,073 INFO [train.py:968] (1/2) Epoch 22, batch 41900, giga_loss[loss=0.2593, simple_loss=0.3424, pruned_loss=0.08805, over 28933.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3747, pruned_loss=0.124, over 5628352.22 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3617, pruned_loss=0.1145, over 5662977.49 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3772, pruned_loss=0.1259, over 5636647.66 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:36:03,201 INFO [optim.py:369] (1/2) 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 16:36:22,588 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999666.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:36:30,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=999674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:36:38,189 INFO [train.py:968] (1/2) Epoch 22, batch 41950, giga_loss[loss=0.2744, simple_loss=0.3601, pruned_loss=0.09436, over 28857.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3734, pruned_loss=0.1227, over 5631956.78 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3617, pruned_loss=0.1145, over 5663600.31 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3756, pruned_loss=0.1244, over 5637635.51 frames. ], batch size: 119, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:37:36,260 INFO [train.py:968] (1/2) Epoch 22, batch 42000, giga_loss[loss=0.3272, simple_loss=0.3966, pruned_loss=0.1289, over 28928.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3732, pruned_loss=0.1208, over 5630960.64 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3617, pruned_loss=0.1144, over 5667625.57 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3751, pruned_loss=0.1223, over 5631284.22 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:37:36,260 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 16:37:42,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5539, 1.9367, 1.5619, 1.4631], device='cuda:1'), covar=tensor([0.2546, 0.2435, 0.2742, 0.2320], device='cuda:1'), in_proj_covar=tensor([0.1508, 0.1091, 0.1332, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 16:37:44,528 INFO [train.py:1012] (1/2) Epoch 22, validation: loss=0.2037, simple_loss=0.3101, pruned_loss=0.04862, over 944034.00 frames. 
+2023-03-11 16:37:44,529 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 16:38:04,821 INFO [optim.py:369] (1/2) 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 16:38:29,649 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:38:35,500 INFO [train.py:968] (1/2) Epoch 22, batch 42050, giga_loss[loss=0.3038, simple_loss=0.3769, pruned_loss=0.1153, over 28729.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3758, pruned_loss=0.1204, over 5648823.88 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3622, pruned_loss=0.1148, over 5668752.74 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3771, pruned_loss=0.1214, over 5647676.50 frames. ], batch size: 99, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:39:09,326 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=999817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:39:09,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-11 16:39:12,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=999820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:39:25,576 INFO [train.py:968] (1/2) Epoch 22, batch 42100, giga_loss[loss=0.2816, simple_loss=0.358, pruned_loss=0.1026, over 28791.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3769, pruned_loss=0.1218, over 5655865.51 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3621, pruned_loss=0.1148, over 5672282.12 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3784, pruned_loss=0.1226, over 5651319.47 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:39:41,426 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:39:42,124 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:40:12,447 INFO [train.py:968] (1/2) Epoch 22, batch 42150, giga_loss[loss=0.2853, simple_loss=0.3594, pruned_loss=0.1056, over 28996.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3766, pruned_loss=0.1224, over 5663505.59 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.362, pruned_loss=0.1148, over 5677750.92 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3782, pruned_loss=0.1232, over 5654696.29 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:40:50,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=999922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:40:57,776 INFO [train.py:968] (1/2) Epoch 22, batch 42200, giga_loss[loss=0.2925, simple_loss=0.3595, pruned_loss=0.1127, over 28906.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.374, pruned_loss=0.1215, over 5671684.75 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1147, over 5681006.30 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3757, pruned_loss=0.1223, over 5661582.30 frames. ], batch size: 199, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:40:59,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8114, 2.0891, 1.4036, 1.8022], device='cuda:1'), covar=tensor([0.0886, 0.0521, 0.0994, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0449, 0.0520, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 16:41:16,976 INFO [optim.py:369] (1/2) 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 16:41:27,161 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=999962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:41:46,885 INFO [train.py:968] (1/2) Epoch 22, batch 42250, libri_loss[loss=0.3422, simple_loss=0.4055, pruned_loss=0.1394, over 26023.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3733, pruned_loss=0.1225, over 5648333.03 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.362, pruned_loss=0.1149, over 5664536.10 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3746, pruned_loss=0.1231, over 5655582.37 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:42:35,306 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-11 16:42:40,602 INFO [train.py:968] (1/2) Epoch 22, batch 42300, giga_loss[loss=0.256, simple_loss=0.3483, pruned_loss=0.08192, over 28582.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3722, pruned_loss=0.1214, over 5650975.46 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.115, over 5665436.42 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3731, pruned_loss=0.1218, over 5655874.68 frames. ], batch size: 60, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:42:48,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6465, 1.9642, 1.5064, 1.7398], device='cuda:1'), covar=tensor([0.2795, 0.2760, 0.3375, 0.2537], device='cuda:1'), in_proj_covar=tensor([0.1514, 0.1095, 0.1336, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 16:42:50,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000041.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:43:00,923 INFO [optim.py:369] (1/2) 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 16:43:13,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:43:18,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:43:29,699 INFO [train.py:968] (1/2) Epoch 22, batch 42350, libri_loss[loss=0.373, simple_loss=0.4129, pruned_loss=0.1666, over 29554.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3715, pruned_loss=0.1192, over 5659439.87 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3626, pruned_loss=0.1154, over 5661997.80 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.372, pruned_loss=0.1194, over 5665660.11 frames. ], batch size: 83, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:43:47,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:44:21,971 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5648, 1.7182, 1.2995, 1.2924], device='cuda:1'), covar=tensor([0.0906, 0.0538, 0.1017, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0448, 0.0518, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 16:44:22,278 INFO [train.py:968] (1/2) Epoch 22, batch 42400, giga_loss[loss=0.3165, simple_loss=0.3829, pruned_loss=0.125, over 28684.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3732, pruned_loss=0.1205, over 5662412.94 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5664278.70 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3738, pruned_loss=0.1208, over 5665063.63 frames. ], batch size: 307, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:44:37,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:44:40,532 INFO [optim.py:369] (1/2) 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 16:45:10,325 INFO [train.py:968] (1/2) Epoch 22, batch 42450, giga_loss[loss=0.3026, simple_loss=0.3777, pruned_loss=0.1137, over 28593.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3731, pruned_loss=0.1207, over 5657261.57 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1152, over 5658992.23 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3738, pruned_loss=0.1211, over 5664659.45 frames. ], batch size: 307, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:45:14,043 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000184.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:45:15,516 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:45:16,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000187.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:45:18,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-11 16:45:40,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000216.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:45:46,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:45:55,165 INFO [train.py:968] (1/2) Epoch 22, batch 42500, giga_loss[loss=0.2471, simple_loss=0.3186, pruned_loss=0.08785, over 28144.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3701, pruned_loss=0.119, over 5672905.47 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3626, pruned_loss=0.115, over 5665767.94 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.371, pruned_loss=0.1196, over 5672697.88 frames. ], batch size: 77, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:45:56,745 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:46:17,515 INFO [optim.py:369] (1/2) 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 16:46:34,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-11 16:46:46,703 INFO [train.py:968] (1/2) Epoch 22, batch 42550, giga_loss[loss=0.3142, simple_loss=0.3804, pruned_loss=0.124, over 29041.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3687, pruned_loss=0.1185, over 5667946.96 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3626, pruned_loss=0.115, over 5666982.10 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3694, pruned_loss=0.119, over 5666589.13 frames. ], batch size: 155, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:46:57,710 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:46:59,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:47:27,536 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:47:35,015 INFO [train.py:968] (1/2) Epoch 22, batch 42600, giga_loss[loss=0.258, simple_loss=0.3286, pruned_loss=0.09367, over 28575.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3679, pruned_loss=0.1188, over 5680646.51 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3625, pruned_loss=0.1151, over 5670515.47 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3687, pruned_loss=0.1192, over 5676628.23 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:47:40,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:47:54,693 INFO [optim.py:369] (1/2) 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 16:48:06,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:48:09,431 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 42650, giga_loss[loss=0.3649, simple_loss=0.4108, pruned_loss=0.1596, over 28374.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3683, pruned_loss=0.1199, over 5682375.61 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1154, over 5677735.92 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3687, pruned_loss=0.1201, over 5673239.65 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:48:41,811 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:49:15,206 INFO [train.py:968] (1/2) Epoch 22, batch 42700, giga_loss[loss=0.3482, simple_loss=0.3773, pruned_loss=0.1595, over 23560.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3681, pruned_loss=0.1205, over 5659201.45 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3632, pruned_loss=0.1155, over 5672935.96 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3683, pruned_loss=0.1207, over 5655264.24 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:49:33,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4118, 2.1969, 1.5849, 0.6859], device='cuda:1'), covar=tensor([0.4899, 0.3106, 0.4438, 0.5726], device='cuda:1'), in_proj_covar=tensor([0.1765, 0.1665, 0.1604, 0.1436], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 16:49:34,272 INFO [optim.py:369] (1/2) 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 16:49:59,963 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:50:00,334 INFO [train.py:968] (1/2) Epoch 22, batch 42750, giga_loss[loss=0.2794, simple_loss=0.3469, pruned_loss=0.1059, over 28291.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3673, pruned_loss=0.12, over 5658860.18 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3632, pruned_loss=0.1154, over 5675628.70 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3677, pruned_loss=0.1205, over 5653115.00 frames. ], batch size: 77, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 16:50:02,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:50:27,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:50:46,997 INFO [train.py:968] (1/2) Epoch 22, batch 42800, giga_loss[loss=0.3019, simple_loss=0.3668, pruned_loss=0.1185, over 28509.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3678, pruned_loss=0.1195, over 5664025.68 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3631, pruned_loss=0.1152, over 5677459.38 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.1201, over 5657154.73 frames. ], batch size: 336, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:51:09,568 INFO [optim.py:369] (1/2) 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 16:51:16,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:51:17,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000562.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:51:29,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3633, 1.6753, 1.5604, 1.4247], device='cuda:1'), covar=tensor([0.1790, 0.1751, 0.2186, 0.1978], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0750, 0.0716, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 16:51:33,325 INFO [train.py:968] (1/2) Epoch 22, batch 42850, giga_loss[loss=0.2992, simple_loss=0.3632, pruned_loss=0.1176, over 27629.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3669, pruned_loss=0.1177, over 5664917.04 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3628, pruned_loss=0.115, over 5670408.64 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3677, pruned_loss=0.1185, over 5665519.62 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:51:59,548 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 42900, giga_loss[loss=0.3106, simple_loss=0.384, pruned_loss=0.1186, over 28978.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3671, pruned_loss=0.1173, over 5673221.56 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3627, pruned_loss=0.115, over 5675390.53 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3679, pruned_loss=0.118, over 5669228.55 frames. ], batch size: 164, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:52:31,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1533, 3.3978, 2.3151, 1.0944], device='cuda:1'), covar=tensor([0.7302, 0.3160, 0.3642, 0.6910], device='cuda:1'), in_proj_covar=tensor([0.1766, 0.1666, 0.1603, 0.1436], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 16:52:43,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 16:52:44,107 INFO [optim.py:369] (1/2) 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 16:52:50,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:53:13,169 INFO [train.py:968] (1/2) Epoch 22, batch 42950, giga_loss[loss=0.3229, simple_loss=0.386, pruned_loss=0.1299, over 28499.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3681, pruned_loss=0.1182, over 5681279.54 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3625, pruned_loss=0.1148, over 5679732.08 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.369, pruned_loss=0.119, over 5674261.16 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:53:34,653 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:53:36,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:53:36,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3692, 1.5249, 1.4005, 1.3234], device='cuda:1'), covar=tensor([0.1702, 0.1796, 0.1465, 0.1491], device='cuda:1'), in_proj_covar=tensor([0.1976, 0.1913, 0.1845, 0.1987], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 16:54:01,153 INFO [train.py:968] (1/2) Epoch 22, batch 43000, giga_loss[loss=0.3005, simple_loss=0.3713, pruned_loss=0.1148, over 28866.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3716, pruned_loss=0.1211, over 5679672.24 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3628, pruned_loss=0.1149, over 5674590.29 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3723, pruned_loss=0.1217, over 5678968.96 frames. ], batch size: 112, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:54:03,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2686, 1.4622, 1.3533, 1.1902], device='cuda:1'), covar=tensor([0.2273, 0.2341, 0.1688, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1977, 0.1913, 0.1844, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 16:54:05,973 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:54:22,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1000751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:54:25,808 INFO [optim.py:369] (1/2) 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 16:54:26,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1000754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:54:49,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3282, 1.8961, 1.3732, 0.6714], device='cuda:1'), covar=tensor([0.4178, 0.2504, 0.3322, 0.5580], device='cuda:1'), in_proj_covar=tensor([0.1768, 0.1671, 0.1607, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 16:54:53,975 INFO [train.py:968] (1/2) Epoch 22, batch 43050, giga_loss[loss=0.2799, simple_loss=0.3505, pruned_loss=0.1046, over 28943.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 5687725.00 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3621, pruned_loss=0.1144, over 5681443.26 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3723, pruned_loss=0.123, over 5681071.41 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:54:57,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1000783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:55:44,662 INFO [train.py:968] (1/2) Epoch 22, batch 43100, giga_loss[loss=0.2881, simple_loss=0.3524, pruned_loss=0.112, over 28788.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.371, pruned_loss=0.1227, over 5686505.18 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3623, pruned_loss=0.1146, over 5686121.51 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3722, pruned_loss=0.1237, over 5676701.46 frames. ], batch size: 99, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:56:07,013 INFO [optim.py:369] (1/2) 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 16:56:34,938 INFO [train.py:968] (1/2) Epoch 22, batch 43150, giga_loss[loss=0.2759, simple_loss=0.348, pruned_loss=0.102, over 28881.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3739, pruned_loss=0.1254, over 5664631.21 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5681323.52 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.375, pruned_loss=0.1263, over 5660514.16 frames. ], batch size: 119, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:57:20,047 INFO [train.py:968] (1/2) Epoch 22, batch 43200, giga_loss[loss=0.2767, simple_loss=0.3461, pruned_loss=0.1036, over 28530.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3735, pruned_loss=0.1255, over 5653532.78 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3629, pruned_loss=0.115, over 5668176.51 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3743, pruned_loss=0.1263, over 5661499.76 frames. ], batch size: 336, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:57:26,272 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1000937.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 16:57:41,253 INFO [optim.py:369] (1/2) 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 16:57:56,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 16:58:07,141 INFO [train.py:968] (1/2) Epoch 22, batch 43250, giga_loss[loss=0.323, simple_loss=0.3799, pruned_loss=0.133, over 28570.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3732, pruned_loss=0.1243, over 5648399.89 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1155, over 5659817.43 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3735, pruned_loss=0.1247, over 5662311.95 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 16:58:11,922 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-11 16:58:36,572 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-03-11 16:58:54,624 INFO [train.py:968] (1/2) Epoch 22, batch 43300, giga_loss[loss=0.3028, simple_loss=0.3661, pruned_loss=0.1197, over 28530.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3715, pruned_loss=0.1216, over 5652348.88 frames. ], libri_tot_loss[loss=0.2971, simple_loss=0.3633, pruned_loss=0.1154, over 5663417.14 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3721, pruned_loss=0.1221, over 5659952.79 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:58:57,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 16:59:19,323 INFO [optim.py:369] (1/2) 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 16:59:43,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001080.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 16:59:43,634 INFO [train.py:968] (1/2) Epoch 22, batch 43350, giga_loss[loss=0.2883, simple_loss=0.359, pruned_loss=0.1088, over 28484.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3695, pruned_loss=0.1208, over 5657214.27 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1157, over 5666906.65 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3698, pruned_loss=0.121, over 5660082.24 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 16:59:46,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001083.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:00:13,177 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001112.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 17:00:16,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5766, 1.8388, 1.6553, 1.6398], device='cuda:1'), covar=tensor([0.1677, 0.1891, 0.1908, 0.1884], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0752, 0.0718, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 17:00:32,573 INFO [train.py:968] (1/2) Epoch 22, batch 43400, giga_loss[loss=0.3186, simple_loss=0.3615, pruned_loss=0.1378, over 23744.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3684, pruned_loss=0.1207, over 5652074.59 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3641, pruned_loss=0.116, over 5659382.09 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3682, pruned_loss=0.1207, over 5661670.46 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:00:45,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 17:00:57,281 INFO [optim.py:369] (1/2) 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 17:01:18,120 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:01:21,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:01:22,557 INFO [train.py:968] (1/2) Epoch 22, batch 43450, giga_loss[loss=0.3277, simple_loss=0.3985, pruned_loss=0.1284, over 28884.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3685, pruned_loss=0.1209, over 5659168.05 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3643, pruned_loss=0.1161, over 5662696.45 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3682, pruned_loss=0.1209, over 5663757.24 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:01:48,449 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:02:10,277 INFO [train.py:968] (1/2) Epoch 22, batch 43500, giga_loss[loss=0.2978, simple_loss=0.3745, pruned_loss=0.1106, over 28801.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3725, pruned_loss=0.1233, over 5660509.42 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3641, pruned_loss=0.1159, over 5663726.84 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3727, pruned_loss=0.1237, over 5662839.38 frames. ], batch size: 284, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:02:30,626 INFO [optim.py:369] (1/2) 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 17:02:46,987 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:02:57,339 INFO [train.py:968] (1/2) Epoch 22, batch 43550, giga_loss[loss=0.2914, simple_loss=0.3662, pruned_loss=0.1083, over 28858.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3739, pruned_loss=0.1211, over 5660718.38 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3638, pruned_loss=0.1156, over 5659090.36 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3745, pruned_loss=0.1218, over 5666122.78 frames. ], batch size: 66, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:03:14,172 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:03:47,002 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:03:47,494 INFO [train.py:968] (1/2) Epoch 22, batch 43600, giga_loss[loss=0.3148, simple_loss=0.3804, pruned_loss=0.1246, over 28547.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3742, pruned_loss=0.1209, over 5656288.03 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1156, over 5655929.46 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3752, pruned_loss=0.1216, over 5663027.54 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:04:15,570 INFO [optim.py:369] (1/2) 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 17:04:40,564 INFO [train.py:968] (1/2) Epoch 22, batch 43650, giga_loss[loss=0.4289, simple_loss=0.4356, pruned_loss=0.2111, over 23475.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3767, pruned_loss=0.123, over 5646666.58 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3637, pruned_loss=0.1157, over 5649793.29 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3775, pruned_loss=0.1235, over 5658227.81 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:04:42,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6856, 2.2237, 1.8097, 1.9366], device='cuda:1'), covar=tensor([0.0730, 0.0255, 0.0287, 0.0799], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 17:05:22,890 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5990, 1.6728, 1.7830, 1.4187], device='cuda:1'), covar=tensor([0.1846, 0.2466, 0.1492, 0.1704], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0705, 0.0947, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 17:05:27,597 INFO [train.py:968] (1/2) Epoch 22, batch 43700, giga_loss[loss=0.2755, simple_loss=0.364, pruned_loss=0.09349, over 28880.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3775, pruned_loss=0.1243, over 5649469.48 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1156, over 5654094.54 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3784, pruned_loss=0.125, over 5654925.57 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:05:50,374 INFO [optim.py:369] (1/2) 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 17:06:13,459 INFO [train.py:968] (1/2) Epoch 22, batch 43750, giga_loss[loss=0.2962, simple_loss=0.3682, pruned_loss=0.1121, over 29009.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3762, pruned_loss=0.1241, over 5663669.49 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3638, pruned_loss=0.1157, over 5658069.26 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3769, pruned_loss=0.1247, over 5664411.03 frames. ], batch size: 164, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:06:41,304 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001510.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 17:06:59,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.55 vs. limit=5.0
+2023-03-11 17:07:02,058 INFO [train.py:968] (1/2) Epoch 22, batch 43800, giga_loss[loss=0.2979, simple_loss=0.359, pruned_loss=0.1184, over 27998.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3746, pruned_loss=0.1238, over 5651688.78 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3641, pruned_loss=0.1158, over 5649853.97 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3753, pruned_loss=0.1244, over 5659244.86 frames. ], batch size: 412, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:07:27,591 INFO [optim.py:369] (1/2) 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 17:07:34,279 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 17:07:42,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4168, 1.7521, 1.3573, 1.3776], device='cuda:1'), covar=tensor([0.2696, 0.2649, 0.3028, 0.2435], device='cuda:1'), in_proj_covar=tensor([0.1517, 0.1098, 0.1340, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 17:07:51,815 INFO [train.py:968] (1/2) Epoch 22, batch 43850, giga_loss[loss=0.2493, simple_loss=0.3262, pruned_loss=0.08619, over 28622.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3719, pruned_loss=0.1222, over 5664971.47 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3639, pruned_loss=0.1156, over 5654423.52 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3728, pruned_loss=0.123, over 5666816.30 frames. ], batch size: 60, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:08:41,031 INFO [train.py:968] (1/2) Epoch 22, batch 43900, giga_loss[loss=0.284, simple_loss=0.3659, pruned_loss=0.1011, over 28862.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3718, pruned_loss=0.1229, over 5657186.48 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3639, pruned_loss=0.1156, over 5647043.29 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3728, pruned_loss=0.1238, over 5665154.61 frames. ], batch size: 164, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:08:54,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:09:04,746 INFO [optim.py:369] (1/2) 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 17:09:18,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:09:28,654 INFO [train.py:968] (1/2) Epoch 22, batch 43950, giga_loss[loss=0.2886, simple_loss=0.3542, pruned_loss=0.1115, over 28811.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3724, pruned_loss=0.1233, over 5662082.86 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3632, pruned_loss=0.1151, over 5652629.96 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.374, pruned_loss=0.1246, over 5663914.01 frames. ], batch size: 199, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:09:54,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:09:57,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6629, 1.8579, 1.8460, 1.6140], device='cuda:1'), covar=tensor([0.3306, 0.2574, 0.2105, 0.2504], device='cuda:1'), in_proj_covar=tensor([0.1979, 0.1913, 0.1850, 0.1989], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 17:10:17,563 INFO [train.py:968] (1/2) Epoch 22, batch 44000, giga_loss[loss=0.3714, simple_loss=0.4186, pruned_loss=0.1621, over 27587.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3727, pruned_loss=0.1244, over 5665158.65 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.363, pruned_loss=0.1152, over 5653986.13 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3744, pruned_loss=0.1256, over 5665623.11 frames. ], batch size: 472, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:10:34,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 17:10:39,230 INFO [optim.py:369] (1/2) 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 17:11:04,190 INFO [train.py:968] (1/2) Epoch 22, batch 44050, giga_loss[loss=0.2545, simple_loss=0.3355, pruned_loss=0.0868, over 28915.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1222, over 5671293.75 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5656941.38 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3712, pruned_loss=0.1235, over 5669315.82 frames. ], batch size: 186, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:11:11,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:14,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:38,653 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:41,559 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:45,261 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:11:56,766 INFO [train.py:968] (1/2) Epoch 22, batch 44100, giga_loss[loss=0.3008, simple_loss=0.3724, pruned_loss=0.1146, over 28684.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.369, pruned_loss=0.1212, over 5669943.10 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5656941.38 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3704, pruned_loss=0.1222, over 5668403.65 frames. ], batch size: 284, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:12:10,232 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:17,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1001848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:19,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1001851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:26,403 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1001880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:12:50,467 INFO [train.py:968] (1/2) Epoch 22, batch 44150, giga_loss[loss=0.2782, simple_loss=0.3545, pruned_loss=0.1009, over 28935.00 frames. ], tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5665886.54 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3628, pruned_loss=0.115, over 5662994.98 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3732, pruned_loss=0.1239, over 5659447.79 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:12:54,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1001885.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:13:04,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0048, 2.1672, 1.7610, 1.8397], device='cuda:1'), covar=tensor([0.0982, 0.0758, 0.0998, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0451, 0.0521, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 17:13:17,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3598, 1.5448, 1.5263, 1.4135], device='cuda:1'), covar=tensor([0.1711, 0.1750, 0.2027, 0.1778], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0755, 0.0718, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 17:13:36,913 INFO [train.py:968] (1/2) Epoch 22, batch 44200, giga_loss[loss=0.3777, simple_loss=0.4213, pruned_loss=0.167, over 28597.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.1241, over 5670904.41 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.363, pruned_loss=0.1152, over 5657735.57 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1248, over 5670140.99 frames. ], batch size: 242, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:14:04,597 INFO [optim.py:369] (1/2) 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 17:14:23,793 INFO [train.py:968] (1/2) Epoch 22, batch 44250, giga_loss[loss=0.2769, simple_loss=0.3739, pruned_loss=0.08994, over 29039.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3733, pruned_loss=0.1239, over 5637536.64 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3633, pruned_loss=0.1157, over 5627713.97 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3741, pruned_loss=0.1244, over 5664368.99 frames. ], batch size: 106, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:14:27,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1001985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:15:10,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1002028.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 17:15:11,922 INFO [train.py:968] (1/2) Epoch 22, batch 44300, giga_loss[loss=0.2652, simple_loss=0.3533, pruned_loss=0.0885, over 28441.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3738, pruned_loss=0.1218, over 5648622.43 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5631850.65 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3752, pruned_loss=0.1226, over 5666420.20 frames. ], batch size: 65, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:15:14,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1002031.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 17:15:37,029 INFO [optim.py:369] (1/2) 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 17:15:38,673 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1002060.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 17:16:00,685 INFO [train.py:968] (1/2) Epoch 22, batch 44350, libri_loss[loss=0.2852, simple_loss=0.3545, pruned_loss=0.108, over 29515.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1216, over 5658932.75 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.363, pruned_loss=0.1155, over 5632644.74 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3773, pruned_loss=0.1221, over 5672767.56 frames. ], batch size: 84, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:16:13,583 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-11 17:16:36,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8973, 3.7085, 3.4876, 1.6665], device='cuda:1'), covar=tensor([0.0847, 0.1004, 0.1031, 0.2244], device='cuda:1'), in_proj_covar=tensor([0.1256, 0.1164, 0.0984, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 17:16:49,145 INFO [train.py:968] (1/2) Epoch 22, batch 44400, giga_loss[loss=0.3081, simple_loss=0.3764, pruned_loss=0.1199, over 28909.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3784, pruned_loss=0.1227, over 5670788.58 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3627, pruned_loss=0.1154, over 5632443.26 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3799, pruned_loss=0.1235, over 5683007.62 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:16:54,422 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:17:42,134 INFO [train.py:968] (1/2) Epoch 22, batch 44450, giga_loss[loss=0.3187, simple_loss=0.3749, pruned_loss=0.1313, over 28922.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3811, pruned_loss=0.1264, over 5656337.88 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3631, pruned_loss=0.1157, over 5630343.99 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3823, pruned_loss=0.1269, over 5668793.93 frames. ], batch size: 112, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:18:31,624 INFO [train.py:968] (1/2) Epoch 22, batch 44500, giga_loss[loss=0.3169, simple_loss=0.3809, pruned_loss=0.1264, over 28636.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3817, pruned_loss=0.1278, over 5648071.16 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3632, pruned_loss=0.1157, over 5637199.43 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3829, pruned_loss=0.1285, over 5652953.26 frames. ], batch size: 242, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:18:59,924 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 22, batch 44550, giga_loss[loss=0.2877, simple_loss=0.3642, pruned_loss=0.1056, over 28548.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3817, pruned_loss=0.1278, over 5646868.21 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3634, pruned_loss=0.1159, over 5632428.02 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3829, pruned_loss=0.1285, over 5656111.10 frames. ], batch size: 307, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:20:07,823 INFO [train.py:968] (1/2) Epoch 22, batch 44600, giga_loss[loss=0.3218, simple_loss=0.396, pruned_loss=0.1238, over 28687.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3787, pruned_loss=0.1248, over 5651765.46 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.1159, over 5634523.62 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3799, pruned_loss=0.1254, over 5657149.22 frames. ], batch size: 262, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:20:21,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4609, 1.5213, 1.6665, 1.2857], device='cuda:1'), covar=tensor([0.1841, 0.2642, 0.1523, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0705, 0.0948, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 17:20:33,437 INFO [optim.py:369] (1/2) 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 17:20:36,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1002360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:20:40,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5913, 1.7292, 1.4322, 1.7355], device='cuda:1'), covar=tensor([0.2697, 0.2872, 0.3214, 0.2440], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1095, 0.1339, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0013, 0.0011, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 17:20:53,017 INFO [train.py:968] (1/2) Epoch 22, batch 44650, giga_loss[loss=0.3006, simple_loss=0.3873, pruned_loss=0.107, over 28570.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3783, pruned_loss=0.1228, over 5642075.13 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.116, over 5621006.45 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3797, pruned_loss=0.1234, over 5659701.55 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:21:41,704 INFO [train.py:968] (1/2) Epoch 22, batch 44700, giga_loss[loss=0.2967, simple_loss=0.3735, pruned_loss=0.11, over 28898.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3796, pruned_loss=0.1232, over 5653199.18 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.116, over 5624390.56 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.381, pruned_loss=0.1239, over 5664593.68 frames. ], batch size: 227, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:22:07,184 INFO [optim.py:369] (1/2) 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 17:22:33,340 INFO [train.py:968] (1/2) Epoch 22, batch 44750, giga_loss[loss=0.3093, simple_loss=0.3787, pruned_loss=0.12, over 28324.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3804, pruned_loss=0.1245, over 5659187.25 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3637, pruned_loss=0.1162, over 5628701.09 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3815, pruned_loss=0.1249, over 5664846.36 frames. ], batch size: 368, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:22:52,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3405, 1.9241, 1.3484, 0.6571], device='cuda:1'), covar=tensor([0.5259, 0.2498, 0.3338, 0.6113], device='cuda:1'), in_proj_covar=tensor([0.1754, 0.1656, 0.1596, 0.1430], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 17:22:53,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1002503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:22:57,115 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1002506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:23:01,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1002512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:23:14,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.7668, 5.6100, 5.3447, 2.8215], device='cuda:1'), covar=tensor([0.0479, 0.0598, 0.0684, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.1170, 0.0989, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 17:23:22,299 INFO [train.py:968] (1/2) Epoch 22, batch 44800, giga_loss[loss=0.3127, simple_loss=0.3571, pruned_loss=0.1341, over 23501.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3796, pruned_loss=0.1247, over 5664606.75 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3636, pruned_loss=0.1161, over 5633766.78 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3809, pruned_loss=0.1252, over 5665490.52 frames. ], batch size: 705, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:23:27,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1002535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:23:36,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3391, 4.3529, 1.5261, 1.8064], device='cuda:1'), covar=tensor([0.1234, 0.0441, 0.0980, 0.1453], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0563, 0.0390, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 17:23:49,867 INFO [optim.py:369] (1/2) 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 17:23:51,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4538, 1.3641, 3.6058, 3.2282], device='cuda:1'), covar=tensor([0.1506, 0.2705, 0.0522, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0651, 0.0973, 0.0920], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 17:24:16,558 INFO [train.py:968] (1/2) Epoch 22, batch 44850, giga_loss[loss=0.3174, simple_loss=0.3837, pruned_loss=0.1256, over 28776.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3777, pruned_loss=0.125, over 5645894.45 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.116, over 5634537.83 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3791, pruned_loss=0.1256, over 5646271.92 frames. ], batch size: 243, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:25:03,756 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-11 17:25:04,559 INFO [train.py:968] (1/2) Epoch 22, batch 44900, libri_loss[loss=0.2777, simple_loss=0.3409, pruned_loss=0.1073, over 29383.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3755, pruned_loss=0.124, over 5652988.99 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3637, pruned_loss=0.1161, over 5637804.33 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3767, pruned_loss=0.1247, over 5650582.01 frames. ], batch size: 71, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:25:28,892 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1002655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:25:31,567 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:25:51,695 INFO [train.py:968] (1/2) Epoch 22, batch 44950, giga_loss[loss=0.2853, simple_loss=0.3454, pruned_loss=0.1125, over 28536.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.374, pruned_loss=0.1233, over 5648355.48 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.364, pruned_loss=0.1163, over 5627329.80 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3749, pruned_loss=0.1238, over 5657192.77 frames. ], batch size: 85, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:25:53,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-11 17:25:57,960 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1002687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:26:12,246 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1002702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:26:32,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6078, 1.8096, 1.6607, 1.6134], device='cuda:1'), covar=tensor([0.1819, 0.2229, 0.2415, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0753, 0.0717, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 17:26:40,672 INFO [train.py:968] (1/2) Epoch 22, batch 45000, libri_loss[loss=0.2749, simple_loss=0.3472, pruned_loss=0.1013, over 29552.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3734, pruned_loss=0.1235, over 5657060.52 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.364, pruned_loss=0.1162, over 5631518.83 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3742, pruned_loss=0.124, over 5660466.13 frames. ], batch size: 78, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:26:40,672 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 17:26:50,445 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 17:27:17,062 INFO [optim.py:369] (1/2) 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 17:27:36,228 INFO [train.py:968] (1/2) Epoch 22, batch 45050, giga_loss[loss=0.2603, simple_loss=0.3469, pruned_loss=0.08686, over 28927.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.372, pruned_loss=0.1221, over 5665124.57 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3644, pruned_loss=0.1164, over 5636543.61 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3725, pruned_loss=0.1226, over 5663745.49 frames. ], batch size: 136, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:28:23,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2096, 3.0429, 1.3569, 1.3982], device='cuda:1'), covar=tensor([0.1124, 0.0469, 0.0973, 0.1539], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0562, 0.0390, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 17:28:25,799 INFO [train.py:968] (1/2) Epoch 22, batch 45100, giga_loss[loss=0.2896, simple_loss=0.3602, pruned_loss=0.1095, over 28626.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3687, pruned_loss=0.1187, over 5663448.28 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3643, pruned_loss=0.1163, over 5641681.76 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3693, pruned_loss=0.1192, over 5658121.66 frames. ], batch size: 307, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:28:48,271 INFO [optim.py:369] (1/2) 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 17:29:09,361 INFO [train.py:968] (1/2) Epoch 22, batch 45150, giga_loss[loss=0.3397, simple_loss=0.3941, pruned_loss=0.1426, over 28551.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3676, pruned_loss=0.1176, over 5661927.23 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.3648, pruned_loss=0.1167, over 5638560.78 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3678, pruned_loss=0.1177, over 5661827.64 frames. ], batch size: 336, lr: 1.43e-03, grad_scale: 2.0
+2023-03-11 17:29:18,987 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1002890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:30:01,950 INFO [train.py:968] (1/2) Epoch 22, batch 45200, giga_loss[loss=0.2593, simple_loss=0.3389, pruned_loss=0.08982, over 28877.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3657, pruned_loss=0.1166, over 5647489.53 frames. ], libri_tot_loss[loss=0.299, simple_loss=0.3647, pruned_loss=0.1166, over 5639115.04 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3659, pruned_loss=0.1168, over 5647343.33 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:30:32,386 INFO [optim.py:369] (1/2) 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 17:30:54,671 INFO [train.py:968] (1/2) Epoch 22, batch 45250, giga_loss[loss=0.3013, simple_loss=0.367, pruned_loss=0.1178, over 28007.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3637, pruned_loss=0.1164, over 5626894.91 frames. ], libri_tot_loss[loss=0.2993, simple_loss=0.3651, pruned_loss=0.1168, over 5632625.12 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3635, pruned_loss=0.1164, over 5633476.44 frames. ], batch size: 412, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:31:38,942 INFO [train.py:968] (1/2) Epoch 22, batch 45300, giga_loss[loss=0.2749, simple_loss=0.3529, pruned_loss=0.0985, over 28848.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3632, pruned_loss=0.1165, over 5643655.75 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1164, over 5640699.59 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3635, pruned_loss=0.1168, over 5641823.43 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:32:08,069 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:32:22,827 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 22, batch 45350, giga_loss[loss=0.2697, simple_loss=0.3527, pruned_loss=0.09332, over 28933.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3658, pruned_loss=0.1173, over 5634095.53 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3642, pruned_loss=0.1161, over 5628327.51 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3664, pruned_loss=0.1178, over 5644409.37 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:33:04,515 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:33:16,466 INFO [train.py:968] (1/2) Epoch 22, batch 45400, libri_loss[loss=0.2857, simple_loss=0.3564, pruned_loss=0.1075, over 29527.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.368, pruned_loss=0.1188, over 5630342.38 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3641, pruned_loss=0.1161, over 5635508.47 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3686, pruned_loss=0.1193, over 5631959.44 frames. ], batch size: 83, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:33:44,583 INFO [optim.py:369] (1/2) 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 17:34:06,394 INFO [train.py:968] (1/2) Epoch 22, batch 45450, giga_loss[loss=0.3198, simple_loss=0.3835, pruned_loss=0.128, over 29031.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3681, pruned_loss=0.1194, over 5624752.61 frames. ], libri_tot_loss[loss=0.2987, simple_loss=0.3645, pruned_loss=0.1164, over 5638218.63 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3683, pruned_loss=0.1196, over 5623481.66 frames. ], batch size: 128, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:34:12,642 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-11 17:34:44,583 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:34:46,947 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:34:55,368 INFO [train.py:968] (1/2) Epoch 22, batch 45500, giga_loss[loss=0.2652, simple_loss=0.3441, pruned_loss=0.0931, over 28974.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3675, pruned_loss=0.1187, over 5637756.62 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3648, pruned_loss=0.1165, over 5641482.31 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3674, pruned_loss=0.1188, over 5633746.72 frames. ], batch size: 174, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:35:13,460 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:35:20,394 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 22, batch 45550, giga_loss[loss=0.3303, simple_loss=0.3833, pruned_loss=0.1387, over 28829.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1213, over 5647812.52 frames. ], libri_tot_loss[loss=0.2991, simple_loss=0.365, pruned_loss=0.1166, over 5646407.67 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3704, pruned_loss=0.1213, over 5640116.06 frames. ], batch size: 66, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:36:25,797 INFO [train.py:968] (1/2) Epoch 22, batch 45600, giga_loss[loss=0.2996, simple_loss=0.37, pruned_loss=0.1146, over 28898.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3728, pruned_loss=0.1227, over 5597040.03 frames. ], libri_tot_loss[loss=0.3001, simple_loss=0.3657, pruned_loss=0.1173, over 5586703.64 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3722, pruned_loss=0.1224, over 5644783.93 frames. ], batch size: 112, lr: 1.43e-03, grad_scale: 8.0
+2023-03-11 17:36:52,394 INFO [optim.py:369] (1/2) 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 17:37:10,689 INFO [train.py:968] (1/2) Epoch 22, batch 45650, giga_loss[loss=0.2973, simple_loss=0.3694, pruned_loss=0.1126, over 28944.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3745, pruned_loss=0.1244, over 5550345.80 frames. ], libri_tot_loss[loss=0.3014, simple_loss=0.3666, pruned_loss=0.1181, over 5520391.16 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3734, pruned_loss=0.1234, over 5649253.11 frames. ], batch size: 213, lr: 1.43e-03, grad_scale: 4.0
+2023-03-11 17:37:22,022 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-11 17:38:10,683 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:13,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:40,297 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:40,652 INFO [train.py:968] (1/2) Epoch 23, batch 50, giga_loss[loss=0.3191, simple_loss=0.3929, pruned_loss=0.1227, over 28689.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3723, pruned_loss=0.1082, over 1260751.55 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3403, pruned_loss=0.08886, over 165329.29 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3765, pruned_loss=0.1107, over 1127881.09 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:38:49,023 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1003448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:38:59,165 INFO [optim.py:369] (1/2) 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 17:39:28,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6497, 1.9354, 1.3219, 1.6139], device='cuda:1'), covar=tensor([0.1090, 0.0739, 0.1161, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0451, 0.0522, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 17:39:29,500 INFO [train.py:968] (1/2) Epoch 23, batch 100, giga_loss[loss=0.2673, simple_loss=0.3478, pruned_loss=0.0934, over 28643.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3645, pruned_loss=0.1049, over 2238752.90 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3392, pruned_loss=0.08783, over 389122.22 frames. ], giga_tot_loss[loss=0.2923, simple_loss=0.3689, pruned_loss=0.1078, over 1983987.80 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:39:33,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1003494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:39:34,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5317, 5.3522, 5.0430, 2.5716], device='cuda:1'), covar=tensor([0.0409, 0.0567, 0.0640, 0.1785], device='cuda:1'), in_proj_covar=tensor([0.1251, 0.1160, 0.0981, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 17:40:14,348 INFO [train.py:968] (1/2) Epoch 23, batch 150, giga_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09928, over 28301.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3493, pruned_loss=0.09744, over 3003508.52 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3404, pruned_loss=0.08933, over 515732.85 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3511, pruned_loss=0.09881, over 2735699.96 frames. ], batch size: 369, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:40:34,434 INFO [optim.py:369] (1/2) 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 17:40:49,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.2426, 6.0225, 5.6888, 3.4429], device='cuda:1'), covar=tensor([0.0351, 0.0601, 0.0701, 0.1410], device='cuda:1'), in_proj_covar=tensor([0.1250, 0.1158, 0.0980, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 17:40:59,910 INFO [train.py:968] (1/2) Epoch 23, batch 200, libri_loss[loss=0.2609, simple_loss=0.3456, pruned_loss=0.08812, over 29179.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3348, pruned_loss=0.09064, over 3599283.22 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3407, pruned_loss=0.08926, over 542333.19 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3349, pruned_loss=0.09111, over 3378558.25 frames. ], batch size: 97, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:41:00,163 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:02,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:25,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:25,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7647, 2.1098, 1.9181, 1.8387], device='cuda:1'), covar=tensor([0.2329, 0.2571, 0.2376, 0.2546], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0754, 0.0718, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 17:41:28,908 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:39,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1003637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:42,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1003640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:41:43,150 INFO [train.py:968] (1/2) Epoch 23, batch 250, giga_loss[loss=0.2139, simple_loss=0.2926, pruned_loss=0.0676, over 29016.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3244, pruned_loss=0.08551, over 4060057.41 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3402, pruned_loss=0.08793, over 670765.63 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3237, pruned_loss=0.08574, over 3841803.54 frames. ], batch size: 128, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:41:58,591 INFO [optim.py:369] (1/2) 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 17:42:06,969 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1003669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:42:27,426 INFO [train.py:968] (1/2) Epoch 23, batch 300, giga_loss[loss=0.2217, simple_loss=0.2853, pruned_loss=0.07903, over 26604.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3163, pruned_loss=0.08224, over 4419901.24 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3404, pruned_loss=0.08856, over 774634.92 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.315, pruned_loss=0.08208, over 4217050.29 frames. ], batch size: 555, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:42:31,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:42:34,152 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 350, giga_loss[loss=0.2167, simple_loss=0.3029, pruned_loss=0.06523, over 28773.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.3087, pruned_loss=0.07855, over 4701695.07 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3394, pruned_loss=0.08787, over 876135.91 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3069, pruned_loss=0.07823, over 4517143.78 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:43:31,396 INFO [optim.py:369] (1/2) 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 17:43:41,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3886, 3.2249, 3.0233, 1.8796], device='cuda:1'), covar=tensor([0.0813, 0.0980, 0.0949, 0.1874], device='cuda:1'), in_proj_covar=tensor([0.1244, 0.1152, 0.0973, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 17:43:57,396 INFO [train.py:968] (1/2) Epoch 23, batch 400, giga_loss[loss=0.2247, simple_loss=0.3062, pruned_loss=0.07162, over 28439.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3053, pruned_loss=0.07701, over 4925738.90 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.34, pruned_loss=0.08803, over 996878.93 frames. ], giga_tot_loss[loss=0.2278, simple_loss=0.3028, pruned_loss=0.07639, over 4757442.45 frames. ], batch size: 78, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:44:40,958 INFO [train.py:968] (1/2) Epoch 23, batch 450, giga_loss[loss=0.2149, simple_loss=0.2914, pruned_loss=0.06917, over 28820.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3043, pruned_loss=0.07738, over 5093997.26 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3403, pruned_loss=0.08815, over 1070191.05 frames. ], giga_tot_loss[loss=0.2276, simple_loss=0.3018, pruned_loss=0.07672, over 4948977.29 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:44:59,904 INFO [optim.py:369] (1/2) 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 17:45:10,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:45:16,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:45:28,686 INFO [train.py:968] (1/2) Epoch 23, batch 500, giga_loss[loss=0.1909, simple_loss=0.2716, pruned_loss=0.05507, over 28844.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.3013, pruned_loss=0.07598, over 5221499.00 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08806, over 1094497.10 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.299, pruned_loss=0.0754, over 5104595.20 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:45:32,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6723, 1.8305, 1.9130, 1.4635], device='cuda:1'), covar=tensor([0.2044, 0.2606, 0.1581, 0.1804], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0711, 0.0962, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 17:46:11,886 INFO [train.py:968] (1/2) Epoch 23, batch 550, giga_loss[loss=0.2165, simple_loss=0.292, pruned_loss=0.07043, over 29009.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.2992, pruned_loss=0.07462, over 5330130.76 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3401, pruned_loss=0.08766, over 1237167.71 frames. ], giga_tot_loss[loss=0.222, simple_loss=0.2963, pruned_loss=0.07386, over 5219585.59 frames. ], batch size: 136, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:46:24,012 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1003951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:46:34,352 INFO [optim.py:369] (1/2) 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 17:46:38,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6187, 1.7635, 1.6614, 1.4977], device='cuda:1'), covar=tensor([0.2695, 0.2579, 0.1908, 0.2507], device='cuda:1'), in_proj_covar=tensor([0.1987, 0.1914, 0.1855, 0.1995], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 17:47:03,104 INFO [train.py:968] (1/2) Epoch 23, batch 600, giga_loss[loss=0.1854, simple_loss=0.2674, pruned_loss=0.05175, over 28601.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2969, pruned_loss=0.07345, over 5416216.15 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3398, pruned_loss=0.08749, over 1283852.74 frames. ], giga_tot_loss[loss=0.2199, simple_loss=0.2943, pruned_loss=0.07275, over 5324486.20 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:47:13,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:47:49,899 INFO [train.py:968] (1/2) Epoch 23, batch 650, giga_loss[loss=0.2202, simple_loss=0.2904, pruned_loss=0.07505, over 28830.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2959, pruned_loss=0.07294, over 5482635.53 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.342, pruned_loss=0.08844, over 1396199.94 frames. ], giga_tot_loss[loss=0.2179, simple_loss=0.2922, pruned_loss=0.07177, over 5400729.56 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:48:08,193 INFO [optim.py:369] (1/2) 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 17:48:18,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:48:20,004 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:48:34,785 INFO [train.py:968] (1/2) Epoch 23, batch 700, giga_loss[loss=0.1737, simple_loss=0.2516, pruned_loss=0.04789, over 28873.00 frames. ], tot_loss[loss=0.2186, simple_loss=0.2935, pruned_loss=0.07188, over 5517738.03 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3427, pruned_loss=0.0887, over 1453500.24 frames. ], giga_tot_loss[loss=0.2156, simple_loss=0.2898, pruned_loss=0.07069, over 5456879.95 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 2.0
+2023-03-11 17:49:21,271 INFO [train.py:968] (1/2) Epoch 23, batch 750, giga_loss[loss=0.2226, simple_loss=0.2906, pruned_loss=0.07726, over 28629.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2927, pruned_loss=0.07169, over 5540874.86 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3441, pruned_loss=0.08924, over 1593250.70 frames. ], giga_tot_loss[loss=0.214, simple_loss=0.2878, pruned_loss=0.07006, over 5492167.44 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 2.0
+2023-03-11 17:49:25,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:49:27,328 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:49:53,850 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:05,802 INFO [train.py:968] (1/2) Epoch 23, batch 800, giga_loss[loss=0.2449, simple_loss=0.3225, pruned_loss=0.0837, over 28862.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2933, pruned_loss=0.07228, over 5577492.08 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3424, pruned_loss=0.08825, over 1678691.26 frames. ], giga_tot_loss[loss=0.2154, simple_loss=0.2888, pruned_loss=0.07094, over 5532667.60 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:50:27,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:29,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:30,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:30,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5562, 1.6903, 1.6610, 1.5214], device='cuda:1'), covar=tensor([0.2597, 0.2419, 0.2141, 0.2377], device='cuda:1'), in_proj_covar=tensor([0.1995, 0.1919, 0.1861, 0.2001], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 17:50:32,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:54,536 INFO [train.py:968] (1/2) Epoch 23, batch 850, giga_loss[loss=0.2541, simple_loss=0.3318, pruned_loss=0.08823, over 28576.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.3029, pruned_loss=0.07721, over 5594441.30 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3413, pruned_loss=0.08788, over 1795730.90 frames. ], giga_tot_loss[loss=0.2252, simple_loss=0.2986, pruned_loss=0.07588, over 5557531.60 frames. ], batch size: 60, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:50:59,923 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:50:59,937 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:51:02,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:51:07,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:51:16,581 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 900, giga_loss[loss=0.3127, simple_loss=0.3889, pruned_loss=0.1183, over 28849.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3152, pruned_loss=0.0833, over 5621861.65 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3401, pruned_loss=0.08752, over 1877492.43 frames. ], giga_tot_loss[loss=0.2381, simple_loss=0.3116, pruned_loss=0.08226, over 5587116.28 frames. ], batch size: 186, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:51:50,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3530, 1.9275, 1.6542, 1.6254], device='cuda:1'), covar=tensor([0.0783, 0.0316, 0.0304, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 17:51:55,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-11 17:52:17,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1004326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:52:28,489 INFO [train.py:968] (1/2) Epoch 23, batch 950, giga_loss[loss=0.2832, simple_loss=0.3613, pruned_loss=0.1026, over 28838.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3261, pruned_loss=0.0888, over 5633402.32 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3389, pruned_loss=0.08695, over 1917654.76 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3236, pruned_loss=0.08821, over 5603411.23 frames. ], batch size: 186, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:52:46,662 INFO [optim.py:369] (1/2) 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 17:52:56,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2192, 1.1133, 3.9040, 3.3576], device='cuda:1'), covar=tensor([0.1771, 0.3078, 0.0448, 0.0921], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0650, 0.0969, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 17:53:10,123 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:12,026 INFO [train.py:968] (1/2) Epoch 23, batch 1000, giga_loss[loss=0.2536, simple_loss=0.3394, pruned_loss=0.08388, over 28741.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3316, pruned_loss=0.09014, over 5644846.66 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3393, pruned_loss=0.08722, over 1996518.88 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3293, pruned_loss=0.08964, over 5616740.32 frames. ], batch size: 242, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:53:12,880 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:16,743 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1004397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:18,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:35,345 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:42,737 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:53:53,531 INFO [train.py:968] (1/2) Epoch 23, batch 1050, giga_loss[loss=0.2589, simple_loss=0.334, pruned_loss=0.09191, over 28567.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.335, pruned_loss=0.0904, over 5658012.66 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3409, pruned_loss=0.0884, over 2109057.27 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3325, pruned_loss=0.0897, over 5631727.65 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:54:14,679 INFO [optim.py:369] (1/2) 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 17:54:20,122 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1004472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:54:38,427 INFO [train.py:968] (1/2) Epoch 23, batch 1100, giga_loss[loss=0.2922, simple_loss=0.364, pruned_loss=0.1102, over 28985.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3366, pruned_loss=0.09056, over 5663244.92 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08816, over 2238509.87 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3345, pruned_loss=0.09017, over 5635619.47 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:54:49,138 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1004501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:55:23,673 INFO [train.py:968] (1/2) Epoch 23, batch 1150, giga_loss[loss=0.2694, simple_loss=0.3395, pruned_loss=0.09961, over 28385.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3392, pruned_loss=0.09245, over 5669469.58 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.08833, over 2328440.34 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3372, pruned_loss=0.09216, over 5642770.08 frames. ], batch size: 65, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 17:55:44,560 INFO [optim.py:369] (1/2) 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 17:56:02,890 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-11 17:56:09,361 INFO [train.py:968] (1/2) Epoch 23, batch 1200, giga_loss[loss=0.2702, simple_loss=0.3551, pruned_loss=0.09265, over 29004.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3426, pruned_loss=0.09488, over 5674458.20 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08869, over 2381790.57 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3408, pruned_loss=0.0946, over 5650214.04 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:56:50,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2255, 1.1111, 3.8662, 3.2084], device='cuda:1'), covar=tensor([0.1806, 0.3037, 0.0463, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0756, 0.0645, 0.0959, 0.0909], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 17:56:52,672 INFO [train.py:968] (1/2) Epoch 23, batch 1250, giga_loss[loss=0.2956, simple_loss=0.37, pruned_loss=0.1106, over 29090.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3459, pruned_loss=0.09675, over 5683816.94 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08847, over 2468313.21 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3447, pruned_loss=0.09682, over 5660101.25 frames. ], batch size: 155, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:56:52,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3874, 1.7159, 1.4252, 1.5824], device='cuda:1'), covar=tensor([0.0847, 0.0333, 0.0344, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 17:57:13,067 INFO [optim.py:369] (1/2) 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 17:57:30,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3676, 2.4581, 2.2634, 2.0122], device='cuda:1'), covar=tensor([0.2082, 0.2266, 0.2266, 0.2352], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0751, 0.0716, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 17:57:36,216 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1004687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:57:38,602 INFO [train.py:968] (1/2) Epoch 23, batch 1300, libri_loss[loss=0.2818, simple_loss=0.3655, pruned_loss=0.09905, over 29658.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3494, pruned_loss=0.09799, over 5687160.50 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08924, over 2568468.65 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3481, pruned_loss=0.09798, over 5663602.74 frames. ], batch size: 91, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:57:39,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1004692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 17:58:02,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4198, 1.5225, 1.5376, 1.3431], device='cuda:1'), covar=tensor([0.2914, 0.3050, 0.2338, 0.2998], device='cuda:1'), in_proj_covar=tensor([0.1973, 0.1906, 0.1845, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 17:58:21,455 INFO [train.py:968] (1/2) Epoch 23, batch 1350, giga_loss[loss=0.2649, simple_loss=0.3543, pruned_loss=0.08772, over 28966.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3506, pruned_loss=0.09812, over 5689692.15 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08937, over 2634557.32 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3497, pruned_loss=0.09822, over 5667327.22 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:58:40,479 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 1400, giga_loss[loss=0.2309, simple_loss=0.3242, pruned_loss=0.06885, over 28792.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3503, pruned_loss=0.09655, over 5698306.05 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08927, over 2663719.52 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3497, pruned_loss=0.09673, over 5681760.81 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:59:46,899 INFO [train.py:968] (1/2) Epoch 23, batch 1450, giga_loss[loss=0.2649, simple_loss=0.3546, pruned_loss=0.08762, over 29008.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3491, pruned_loss=0.09493, over 5696706.01 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08891, over 2726327.12 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3489, pruned_loss=0.09534, over 5681411.52 frames. ], batch size: 164, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 17:59:49,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 23, batch 1500, giga_loss[loss=0.2428, simple_loss=0.3325, pruned_loss=0.07653, over 28632.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3468, pruned_loss=0.09254, over 5708628.29 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3417, pruned_loss=0.08868, over 2773724.55 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.347, pruned_loss=0.09303, over 5693986.54 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:01:11,615 INFO [train.py:968] (1/2) Epoch 23, batch 1550, libri_loss[loss=0.2145, simple_loss=0.2934, pruned_loss=0.06783, over 28139.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3466, pruned_loss=0.09274, over 5702793.44 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3408, pruned_loss=0.08794, over 2863818.74 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3474, pruned_loss=0.09356, over 5688289.72 frames. ], batch size: 62, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:01:32,737 INFO [optim.py:369] (1/2) 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 18:01:56,083 INFO [train.py:968] (1/2) Epoch 23, batch 1600, giga_loss[loss=0.3246, simple_loss=0.3917, pruned_loss=0.1288, over 28895.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.349, pruned_loss=0.09604, over 5695379.25 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.341, pruned_loss=0.0881, over 2898676.34 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3496, pruned_loss=0.09672, over 5690161.85 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:02:35,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9606, 2.1670, 2.0802, 1.7240], device='cuda:1'), covar=tensor([0.2983, 0.2468, 0.2518, 0.2922], device='cuda:1'), in_proj_covar=tensor([0.1968, 0.1902, 0.1842, 0.1981], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 18:02:44,120 INFO [train.py:968] (1/2) Epoch 23, batch 1650, giga_loss[loss=0.2539, simple_loss=0.3348, pruned_loss=0.08647, over 29022.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3517, pruned_loss=0.1001, over 5703785.10 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3413, pruned_loss=0.08853, over 2958349.61 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3522, pruned_loss=0.1007, over 5695950.55 frames. ], batch size: 155, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:03:04,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:03:05,019 INFO [optim.py:369] (1/2) 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 18:03:07,878 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:03:15,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4644, 1.5850, 1.2558, 1.6028], device='cuda:1'), covar=tensor([0.0729, 0.0314, 0.0326, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 18:03:16,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1923, 1.3456, 3.7084, 3.1548], device='cuda:1'), covar=tensor([0.1781, 0.2757, 0.0473, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0649, 0.0966, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 18:03:27,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4329, 1.5899, 1.3071, 1.1706], device='cuda:1'), covar=tensor([0.1064, 0.0598, 0.1069, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:03:27,830 INFO [train.py:968] (1/2) Epoch 23, batch 1700, giga_loss[loss=0.276, simple_loss=0.3469, pruned_loss=0.1025, over 28844.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3518, pruned_loss=0.1013, over 5707234.10 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3417, pruned_loss=0.08859, over 3016939.10 frames. ], giga_tot_loss[loss=0.2781, simple_loss=0.3523, pruned_loss=0.1019, over 5697236.07 frames. ], batch size: 112, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:04:01,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5332, 1.6725, 1.7577, 1.3350], device='cuda:1'), covar=tensor([0.1763, 0.2463, 0.1455, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.0909, 0.0704, 0.0955, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 18:04:13,615 INFO [train.py:968] (1/2) Epoch 23, batch 1750, giga_loss[loss=0.2805, simple_loss=0.3557, pruned_loss=0.1027, over 28691.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3496, pruned_loss=0.1005, over 5698064.82 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.342, pruned_loss=0.08879, over 3101953.29 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3501, pruned_loss=0.1013, over 5685074.68 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:04:28,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:04:33,086 INFO [optim.py:369] (1/2) 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 18:04:56,478 INFO [train.py:968] (1/2) Epoch 23, batch 1800, libri_loss[loss=0.2442, simple_loss=0.3291, pruned_loss=0.07972, over 29541.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3482, pruned_loss=0.0998, over 5695672.87 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08868, over 3204824.60 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3488, pruned_loss=0.1009, over 5684535.00 frames. ], batch size: 80, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:05:08,172 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:11,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:12,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:14,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:19,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:23,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2672, 2.6790, 1.4136, 1.3947], device='cuda:1'), covar=tensor([0.0966, 0.0356, 0.0872, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0556, 0.0388, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 18:05:38,911 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:41,217 INFO [train.py:968] (1/2) Epoch 23, batch 1850, giga_loss[loss=0.2545, simple_loss=0.3319, pruned_loss=0.08857, over 28547.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3469, pruned_loss=0.09844, over 5688572.35 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3429, pruned_loss=0.08919, over 3231775.95 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.347, pruned_loss=0.09917, over 5678023.54 frames. ], batch size: 71, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:05:42,105 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:05:47,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9462, 2.2395, 2.0765, 1.6754], device='cuda:1'), covar=tensor([0.3418, 0.2493, 0.2694, 0.3281], device='cuda:1'), in_proj_covar=tensor([0.1973, 0.1907, 0.1844, 0.1982], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 18:06:06,313 INFO [optim.py:369] (1/2) 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 18:06:19,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9436, 1.3025, 1.1250, 0.1888], device='cuda:1'), covar=tensor([0.4749, 0.3388, 0.4932, 0.7146], device='cuda:1'), in_proj_covar=tensor([0.1754, 0.1656, 0.1597, 0.1434], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 18:06:33,984 INFO [train.py:968] (1/2) Epoch 23, batch 1900, giga_loss[loss=0.2287, simple_loss=0.3099, pruned_loss=0.07374, over 28849.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3435, pruned_loss=0.09614, over 5687517.79 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08923, over 3245178.07 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3436, pruned_loss=0.09672, over 5678212.21 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:06:53,698 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:06:56,932 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:07:09,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-11 18:07:23,051 INFO [train.py:968] (1/2) Epoch 23, batch 1950, giga_loss[loss=0.2266, simple_loss=0.3089, pruned_loss=0.07215, over 28861.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3378, pruned_loss=0.09285, over 5679630.27 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08938, over 3287830.06 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3377, pruned_loss=0.09333, over 5677126.45 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:07:37,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-11 18:07:44,037 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:07:55,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5762, 1.7380, 1.4074, 1.7368], device='cuda:1'), covar=tensor([0.0747, 0.0303, 0.0338, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 18:08:09,874 INFO [train.py:968] (1/2) Epoch 23, batch 2000, giga_loss[loss=0.2314, simple_loss=0.3068, pruned_loss=0.07805, over 28832.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3336, pruned_loss=0.09109, over 5666913.41 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08971, over 3360310.80 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3331, pruned_loss=0.09136, over 5663122.02 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:08:13,511 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:08:56,895 INFO [train.py:968] (1/2) Epoch 23, batch 2050, giga_loss[loss=0.267, simple_loss=0.3361, pruned_loss=0.09898, over 27884.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3297, pruned_loss=0.08911, over 5665272.21 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08945, over 3469492.92 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.329, pruned_loss=0.08948, over 5655835.84 frames. ], batch size: 412, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:09:18,257 INFO [optim.py:369] (1/2) 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 18:09:38,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2915, 3.3949, 1.4652, 1.4395], device='cuda:1'), covar=tensor([0.1117, 0.0321, 0.0960, 0.1463], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0553, 0.0387, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 18:09:40,862 INFO [train.py:968] (1/2) Epoch 23, batch 2100, giga_loss[loss=0.2512, simple_loss=0.3358, pruned_loss=0.08329, over 28685.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3313, pruned_loss=0.08966, over 5657856.83 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3439, pruned_loss=0.08973, over 3516063.01 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3301, pruned_loss=0.0898, over 5657541.28 frames. ], batch size: 60, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:10:08,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3734, 2.0472, 1.5736, 0.6746], device='cuda:1'), covar=tensor([0.5961, 0.2866, 0.4253, 0.6331], device='cuda:1'), in_proj_covar=tensor([0.1740, 0.1642, 0.1584, 0.1422], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 18:10:16,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:10:17,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:10:22,195 INFO [train.py:968] (1/2) Epoch 23, batch 2150, giga_loss[loss=0.2266, simple_loss=0.3024, pruned_loss=0.07538, over 28716.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3326, pruned_loss=0.0898, over 5676429.61 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3441, pruned_loss=0.08967, over 3574998.67 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3312, pruned_loss=0.08995, over 5670842.06 frames. ], batch size: 92, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:10:28,153 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:10:41,093 INFO [optim.py:369] (1/2) 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 18:11:02,251 INFO [train.py:968] (1/2) Epoch 23, batch 2200, giga_loss[loss=0.2352, simple_loss=0.3094, pruned_loss=0.08051, over 28405.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3316, pruned_loss=0.08897, over 5688267.11 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3451, pruned_loss=0.08988, over 3677656.68 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3294, pruned_loss=0.08895, over 5674610.00 frames. ], batch size: 65, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:11:16,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2912, 1.8209, 1.4264, 0.4682], device='cuda:1'), covar=tensor([0.4492, 0.2728, 0.4722, 0.6345], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1644, 0.1587, 0.1425], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 18:11:33,358 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-11 18:11:40,627 INFO [train.py:968] (1/2) Epoch 23, batch 2250, giga_loss[loss=0.2409, simple_loss=0.3135, pruned_loss=0.0841, over 28904.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3299, pruned_loss=0.08802, over 5698597.91 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3461, pruned_loss=0.09039, over 3761705.83 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3269, pruned_loss=0.08766, over 5683177.63 frames. ], batch size: 112, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:11:42,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 18:11:42,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6069, 2.3688, 1.6306, 0.6899], device='cuda:1'), covar=tensor([0.5682, 0.3385, 0.4511, 0.6463], device='cuda:1'), in_proj_covar=tensor([0.1741, 0.1645, 0.1587, 0.1425], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 18:11:58,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4526, 1.6218, 1.5100, 1.2339], device='cuda:1'), covar=tensor([0.3211, 0.2670, 0.2130, 0.3089], device='cuda:1'), in_proj_covar=tensor([0.1969, 0.1897, 0.1833, 0.1975], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 18:12:03,263 INFO [optim.py:369] (1/2) 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 18:12:12,067 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 18:12:12,606 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:12:15,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:12:24,010 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 2300, giga_loss[loss=0.2276, simple_loss=0.31, pruned_loss=0.07257, over 28532.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3263, pruned_loss=0.086, over 5707282.17 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3459, pruned_loss=0.09002, over 3804172.87 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3238, pruned_loss=0.08587, over 5691536.93 frames. ], batch size: 336, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:12:24,571 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:12:30,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1277, 5.9285, 5.6712, 3.0560], device='cuda:1'), covar=tensor([0.0475, 0.0573, 0.0781, 0.1487], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.1131, 0.0958, 0.0722], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 18:12:38,380 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 2350, giga_loss[loss=0.2492, simple_loss=0.329, pruned_loss=0.08469, over 28734.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.324, pruned_loss=0.08487, over 5711393.35 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3458, pruned_loss=0.08971, over 3866238.16 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3215, pruned_loss=0.08483, over 5693919.68 frames. ], batch size: 284, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:13:25,719 INFO [optim.py:369] (1/2) 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 18:13:48,843 INFO [train.py:968] (1/2) Epoch 23, batch 2400, giga_loss[loss=0.2851, simple_loss=0.3547, pruned_loss=0.1077, over 27558.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3223, pruned_loss=0.08428, over 5711673.94 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3463, pruned_loss=0.08973, over 3924362.29 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3195, pruned_loss=0.0841, over 5694739.60 frames. ], batch size: 472, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:14:14,332 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:23,897 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:24,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1005834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:26,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1005836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:27,168 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 2450, giga_loss[loss=0.2155, simple_loss=0.2984, pruned_loss=0.0663, over 28629.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3205, pruned_loss=0.08367, over 5717443.34 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.346, pruned_loss=0.08934, over 3963750.19 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3179, pruned_loss=0.08367, over 5700400.29 frames. ], batch size: 242, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:14:35,494 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-11 18:14:35,984 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1005849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:47,216 INFO [optim.py:369] (1/2) 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 18:14:47,505 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:48,150 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1005866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:14:54,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5895, 1.6241, 1.8476, 1.3969], device='cuda:1'), covar=tensor([0.1904, 0.2547, 0.1525, 0.1824], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0704, 0.0954, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 18:14:58,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6076, 1.6640, 1.5976, 1.4088], device='cuda:1'), covar=tensor([0.2813, 0.2632, 0.2009, 0.2870], device='cuda:1'), in_proj_covar=tensor([0.1968, 0.1894, 0.1829, 0.1972], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 18:15:08,416 INFO [train.py:968] (1/2) Epoch 23, batch 2500, giga_loss[loss=0.2177, simple_loss=0.2846, pruned_loss=0.0754, over 28627.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3189, pruned_loss=0.08287, over 5727610.44 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3468, pruned_loss=0.08966, over 4030911.72 frames. ], giga_tot_loss[loss=0.2402, simple_loss=0.3154, pruned_loss=0.08248, over 5707881.51 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:15:17,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6223, 1.7125, 1.3063, 1.3943], device='cuda:1'), covar=tensor([0.1030, 0.0677, 0.1084, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0450, 0.0525, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:15:22,289 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1005924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:15:49,348 INFO [train.py:968] (1/2) Epoch 23, batch 2550, giga_loss[loss=0.223, simple_loss=0.299, pruned_loss=0.07345, over 28430.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3178, pruned_loss=0.08212, over 5735003.49 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3471, pruned_loss=0.0895, over 4086218.43 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.314, pruned_loss=0.0817, over 5714801.55 frames. ], batch size: 78, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:16:05,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4325, 1.7196, 1.6555, 1.3316], device='cuda:1'), covar=tensor([0.2582, 0.2056, 0.1512, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1894, 0.1830, 0.1972], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 18:16:07,773 INFO [optim.py:369] (1/2) 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 18:16:27,984 INFO [train.py:968] (1/2) Epoch 23, batch 2600, giga_loss[loss=0.274, simple_loss=0.3562, pruned_loss=0.0959, over 28749.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3167, pruned_loss=0.0816, over 5733922.24 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3472, pruned_loss=0.08931, over 4130914.49 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.3129, pruned_loss=0.08122, over 5714131.44 frames. ], batch size: 243, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:17:02,173 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:11,277 INFO [train.py:968] (1/2) Epoch 23, batch 2650, libri_loss[loss=0.2263, simple_loss=0.3185, pruned_loss=0.06706, over 29595.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3162, pruned_loss=0.08171, over 5734776.55 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3469, pruned_loss=0.08913, over 4157417.53 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.313, pruned_loss=0.08142, over 5716622.05 frames. ], batch size: 74, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:17:11,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2595, 2.9973, 1.4273, 1.3981], device='cuda:1'), covar=tensor([0.1111, 0.0433, 0.0943, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0553, 0.0387, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 18:17:21,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:23,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:33,012 INFO [optim.py:369] (1/2) 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 18:17:33,856 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:36,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:50,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:17:57,977 INFO [train.py:968] (1/2) Epoch 23, batch 2700, giga_loss[loss=0.2348, simple_loss=0.3168, pruned_loss=0.07642, over 29019.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.32, pruned_loss=0.08392, over 5730696.64 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3468, pruned_loss=0.08891, over 4191323.59 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.317, pruned_loss=0.08373, over 5713966.97 frames. ], batch size: 136, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:18:04,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1402, 1.2062, 1.1055, 0.8355], device='cuda:1'), covar=tensor([0.0929, 0.0465, 0.0966, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0450, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:18:04,580 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:18:09,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8741, 1.9246, 2.0599, 1.6527], device='cuda:1'), covar=tensor([0.2062, 0.2586, 0.1608, 0.1851], device='cuda:1'), in_proj_covar=tensor([0.0913, 0.0708, 0.0960, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 18:18:18,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:18:45,946 INFO [train.py:968] (1/2) Epoch 23, batch 2750, libri_loss[loss=0.3127, simple_loss=0.3913, pruned_loss=0.117, over 29539.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3266, pruned_loss=0.08822, over 5724565.94 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3472, pruned_loss=0.08911, over 4199905.97 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3239, pruned_loss=0.08793, over 5710558.83 frames. ], batch size: 83, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:19:07,697 INFO [optim.py:369] (1/2) 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 18:19:30,451 INFO [train.py:968] (1/2) Epoch 23, batch 2800, libri_loss[loss=0.2743, simple_loss=0.3578, pruned_loss=0.0954, over 29204.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3349, pruned_loss=0.09405, over 5714775.37 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3471, pruned_loss=0.08915, over 4252985.80 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3323, pruned_loss=0.09387, over 5702774.96 frames. ], batch size: 97, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:19:36,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:19:40,088 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:20:02,393 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:20:15,310 INFO [train.py:968] (1/2) Epoch 23, batch 2850, giga_loss[loss=0.2991, simple_loss=0.3762, pruned_loss=0.111, over 28583.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3391, pruned_loss=0.0958, over 5711687.92 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3463, pruned_loss=0.08869, over 4293286.07 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3374, pruned_loss=0.09616, over 5698036.71 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:20:38,975 INFO [optim.py:369] (1/2) 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 18:20:59,326 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-11 18:21:03,679 INFO [train.py:968] (1/2) Epoch 23, batch 2900, giga_loss[loss=0.2889, simple_loss=0.3616, pruned_loss=0.1081, over 28306.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3431, pruned_loss=0.09672, over 5705084.15 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3463, pruned_loss=0.08857, over 4322496.55 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3416, pruned_loss=0.09726, over 5699370.36 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:21:50,096 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:21:51,712 INFO [train.py:968] (1/2) Epoch 23, batch 2950, giga_loss[loss=0.2695, simple_loss=0.3462, pruned_loss=0.0964, over 28522.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3497, pruned_loss=0.101, over 5701421.74 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3461, pruned_loss=0.08858, over 4338312.81 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3487, pruned_loss=0.1015, over 5695056.07 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:21:51,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:15,444 INFO [optim.py:369] (1/2) 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 18:22:18,844 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:21,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:21,919 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:25,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3870, 3.6198, 1.5386, 1.6155], device='cuda:1'), covar=tensor([0.1042, 0.0282, 0.0893, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0556, 0.0388, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 18:22:37,925 INFO [train.py:968] (1/2) Epoch 23, batch 3000, libri_loss[loss=0.2545, simple_loss=0.3335, pruned_loss=0.08779, over 29558.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3527, pruned_loss=0.1028, over 5689675.48 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3453, pruned_loss=0.08838, over 4398591.22 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3528, pruned_loss=0.104, over 5677670.75 frames. ], batch size: 75, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:22:37,925 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 18:22:46,528 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 18:22:54,930 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:22:59,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:23:30,870 INFO [train.py:968] (1/2) Epoch 23, batch 3050, giga_loss[loss=0.2427, simple_loss=0.3249, pruned_loss=0.08026, over 28876.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3493, pruned_loss=0.1003, over 5696577.27 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3452, pruned_loss=0.08838, over 4413652.66 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3495, pruned_loss=0.1013, over 5685289.47 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:23:44,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:23:57,307 INFO [optim.py:369] (1/2) 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 18:24:15,325 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 3100, giga_loss[loss=0.2549, simple_loss=0.3387, pruned_loss=0.08556, over 28571.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3477, pruned_loss=0.09849, over 5703598.99 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3453, pruned_loss=0.08858, over 4428356.82 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3478, pruned_loss=0.09925, over 5692979.01 frames. ], batch size: 336, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:25:00,181 INFO [train.py:968] (1/2) Epoch 23, batch 3150, giga_loss[loss=0.2483, simple_loss=0.3334, pruned_loss=0.08153, over 28745.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3465, pruned_loss=0.09756, over 5703361.23 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3446, pruned_loss=0.08845, over 4469819.08 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3472, pruned_loss=0.09856, over 5697097.40 frames. ], batch size: 119, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:25:06,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-11 18:25:08,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:08,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-11 18:25:11,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:23,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4991, 1.7645, 1.4179, 1.5614], device='cuda:1'), covar=tensor([0.2650, 0.2650, 0.2996, 0.2498], device='cuda:1'), in_proj_covar=tensor([0.1521, 0.1098, 0.1346, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 18:25:24,933 INFO [optim.py:369] (1/2) 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 18:25:30,196 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:35,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1006580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:25:46,237 INFO [train.py:968] (1/2) Epoch 23, batch 3200, giga_loss[loss=0.3033, simple_loss=0.3755, pruned_loss=0.1156, over 28902.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3487, pruned_loss=0.09839, over 5701290.40 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3448, pruned_loss=0.08867, over 4487693.10 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.349, pruned_loss=0.09914, over 5696907.32 frames. ], batch size: 186, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:26:18,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:26:20,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:26:26,559 INFO [train.py:968] (1/2) Epoch 23, batch 3250, giga_loss[loss=0.3187, simple_loss=0.3831, pruned_loss=0.1272, over 28697.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3505, pruned_loss=0.09913, over 5702234.72 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3448, pruned_loss=0.08859, over 4527344.25 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3509, pruned_loss=0.1001, over 5701196.68 frames. ], batch size: 242, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:26:47,125 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:26:53,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5938, 1.7393, 1.8348, 1.3953], device='cuda:1'), covar=tensor([0.1738, 0.2473, 0.1423, 0.1691], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0704, 0.0953, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 18:27:13,575 INFO [train.py:968] (1/2) Epoch 23, batch 3300, giga_loss[loss=0.3123, simple_loss=0.3845, pruned_loss=0.12, over 28832.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.352, pruned_loss=0.1005, over 5701915.85 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3448, pruned_loss=0.08856, over 4546240.65 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3524, pruned_loss=0.1015, over 5699756.57 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:27:37,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:27:40,889 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 3350, giga_loss[loss=0.2886, simple_loss=0.3574, pruned_loss=0.1098, over 28542.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3527, pruned_loss=0.1014, over 5703467.19 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.345, pruned_loss=0.08861, over 4558016.20 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.353, pruned_loss=0.1022, over 5701591.78 frames. ], batch size: 336, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:28:04,972 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:28:39,697 INFO [train.py:968] (1/2) Epoch 23, batch 3400, giga_loss[loss=0.2919, simple_loss=0.3617, pruned_loss=0.111, over 28957.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3521, pruned_loss=0.1009, over 5717741.24 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3445, pruned_loss=0.0882, over 4603986.71 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1023, over 5710607.69 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:29:09,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1006827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:29:14,444 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-11 18:29:21,744 INFO [train.py:968] (1/2) Epoch 23, batch 3450, giga_loss[loss=0.2631, simple_loss=0.3479, pruned_loss=0.08917, over 28956.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3526, pruned_loss=0.1009, over 5712843.19 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3451, pruned_loss=0.08856, over 4629341.27 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.353, pruned_loss=0.102, over 5714727.35 frames. ], batch size: 136, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:29:46,608 INFO [optim.py:369] (1/2) 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 18:30:04,525 INFO [train.py:968] (1/2) Epoch 23, batch 3500, giga_loss[loss=0.2872, simple_loss=0.3642, pruned_loss=0.1051, over 28993.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3524, pruned_loss=0.1001, over 5713127.28 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3451, pruned_loss=0.08862, over 4654673.18 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3529, pruned_loss=0.1012, over 5711089.76 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:30:07,756 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1006895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:30:42,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3875, 1.6157, 1.3465, 1.0093], device='cuda:1'), covar=tensor([0.2960, 0.2739, 0.3284, 0.2426], device='cuda:1'), in_proj_covar=tensor([0.1524, 0.1102, 0.1348, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 18:30:48,361 INFO [train.py:968] (1/2) Epoch 23, batch 3550, giga_loss[loss=0.2832, simple_loss=0.3591, pruned_loss=0.1036, over 28613.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3538, pruned_loss=0.1003, over 5716958.74 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3458, pruned_loss=0.08912, over 4679770.26 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3539, pruned_loss=0.1011, over 5711676.21 frames. ], batch size: 336, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:31:13,720 INFO [optim.py:369] (1/2) 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 18:31:16,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1006970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:31:18,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1006973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:31:25,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3247, 1.7966, 1.0440, 1.3917], device='cuda:1'), covar=tensor([0.1302, 0.0679, 0.1594, 0.1226], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0447, 0.0523, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:31:27,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9750, 2.2154, 1.8667, 1.9425], device='cuda:1'), covar=tensor([0.2067, 0.2536, 0.2440, 0.2357], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0750, 0.0719, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 18:31:34,219 INFO [train.py:968] (1/2) Epoch 23, batch 3600, giga_loss[loss=0.2843, simple_loss=0.3517, pruned_loss=0.1085, over 28910.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3524, pruned_loss=0.09903, over 5716934.73 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3458, pruned_loss=0.08912, over 4679770.26 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3524, pruned_loss=0.09959, over 5712823.27 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:31:42,993 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 3650, giga_loss[loss=0.2544, simple_loss=0.3307, pruned_loss=0.089, over 28959.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3509, pruned_loss=0.09856, over 5724570.64 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3458, pruned_loss=0.08905, over 4710195.73 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3511, pruned_loss=0.09929, over 5717118.94 frames. ], batch size: 227, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:32:24,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9911, 2.4247, 1.5505, 2.0036], device='cuda:1'), covar=tensor([0.0992, 0.0605, 0.1036, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0446, 0.0523, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:32:38,205 INFO [optim.py:369] (1/2) 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 18:32:47,810 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 18:32:55,236 INFO [train.py:968] (1/2) Epoch 23, batch 3700, giga_loss[loss=0.252, simple_loss=0.3317, pruned_loss=0.08616, over 28635.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3484, pruned_loss=0.09753, over 5715531.40 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3456, pruned_loss=0.08909, over 4739166.24 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3488, pruned_loss=0.09838, over 5715343.87 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:33:05,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6507, 1.7460, 1.4041, 1.2659], device='cuda:1'), covar=tensor([0.1010, 0.0596, 0.1039, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0447, 0.0524, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:33:14,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3567, 1.6231, 1.3441, 1.0021], device='cuda:1'), covar=tensor([0.2822, 0.2595, 0.3068, 0.2391], device='cuda:1'), in_proj_covar=tensor([0.1522, 0.1103, 0.1348, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 18:33:36,003 INFO [train.py:968] (1/2) Epoch 23, batch 3750, giga_loss[loss=0.2581, simple_loss=0.3401, pruned_loss=0.08803, over 29015.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3477, pruned_loss=0.09713, over 5717031.55 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3459, pruned_loss=0.08924, over 4756928.90 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3479, pruned_loss=0.09786, over 5722161.79 frames. ], batch size: 136, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:33:48,403 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1007151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:33:54,680 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-11 18:34:00,630 INFO [optim.py:369] (1/2) 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 18:34:05,459 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 3800, giga_loss[loss=0.3117, simple_loss=0.3772, pruned_loss=0.1231, over 28715.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3486, pruned_loss=0.09838, over 5718987.70 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.346, pruned_loss=0.08926, over 4768644.67 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3487, pruned_loss=0.09902, over 5721006.95 frames. ], batch size: 262, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:35:01,612 INFO [train.py:968] (1/2) Epoch 23, batch 3850, libri_loss[loss=0.2224, simple_loss=0.2994, pruned_loss=0.07268, over 29373.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3484, pruned_loss=0.09781, over 5722032.89 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3453, pruned_loss=0.08904, over 4791060.78 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3491, pruned_loss=0.09868, over 5720231.64 frames. ], batch size: 67, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:35:09,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-11 18:35:23,382 INFO [optim.py:369] (1/2) 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 18:35:25,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:35:44,380 INFO [train.py:968] (1/2) Epoch 23, batch 3900, giga_loss[loss=0.2379, simple_loss=0.3221, pruned_loss=0.07683, over 28976.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3471, pruned_loss=0.0963, over 5718970.31 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3455, pruned_loss=0.0892, over 4827132.15 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3477, pruned_loss=0.09715, over 5713998.72 frames. ], batch size: 213, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:36:25,908 INFO [train.py:968] (1/2) Epoch 23, batch 3950, giga_loss[loss=0.2496, simple_loss=0.3278, pruned_loss=0.08564, over 28912.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3469, pruned_loss=0.09605, over 5722685.85 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3457, pruned_loss=0.08927, over 4847149.14 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3471, pruned_loss=0.09679, over 5716691.57 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:36:49,480 INFO [optim.py:369] (1/2) 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 18:37:06,404 INFO [train.py:968] (1/2) Epoch 23, batch 4000, giga_loss[loss=0.2534, simple_loss=0.3403, pruned_loss=0.08324, over 28514.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3452, pruned_loss=0.09526, over 5720308.97 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3456, pruned_loss=0.08926, over 4884457.92 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3455, pruned_loss=0.09614, over 5712750.16 frames. ], batch size: 78, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:37:23,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1007413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:37:26,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1007416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:37:45,173 INFO [train.py:968] (1/2) Epoch 23, batch 4050, giga_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09166, over 28843.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3422, pruned_loss=0.0938, over 5713517.32 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3448, pruned_loss=0.08901, over 4920847.38 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3431, pruned_loss=0.09497, over 5710581.48 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:37:47,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:38:23,012 INFO [train.py:968] (1/2) Epoch 23, batch 4100, giga_loss[loss=0.2736, simple_loss=0.3564, pruned_loss=0.09543, over 28931.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3398, pruned_loss=0.09257, over 5712018.54 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3448, pruned_loss=0.089, over 4941232.88 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3404, pruned_loss=0.0936, over 5709061.00 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:38:53,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:39:05,499 INFO [train.py:968] (1/2) Epoch 23, batch 4150, giga_loss[loss=0.2754, simple_loss=0.3481, pruned_loss=0.1014, over 28854.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.339, pruned_loss=0.09267, over 5707878.67 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3448, pruned_loss=0.08899, over 4952322.57 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3394, pruned_loss=0.09354, over 5705778.72 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:39:06,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7371, 2.0316, 1.4449, 1.5077], device='cuda:1'), covar=tensor([0.1074, 0.0653, 0.1092, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0446, 0.0524, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:39:12,891 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1007549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:39:12,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:39:20,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3186, 2.2238, 2.1198, 1.9487], device='cuda:1'), covar=tensor([0.1795, 0.2610, 0.2294, 0.2503], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0744, 0.0713, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 18:39:29,444 INFO [optim.py:369] (1/2) 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 18:39:45,752 INFO [train.py:968] (1/2) Epoch 23, batch 4200, giga_loss[loss=0.2111, simple_loss=0.2894, pruned_loss=0.06637, over 28365.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3373, pruned_loss=0.09226, over 5707612.98 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3448, pruned_loss=0.08916, over 4984177.35 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3374, pruned_loss=0.09295, over 5700575.55 frames. ], batch size: 65, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:40:21,785 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1007630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:40:25,790 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8967, 1.1270, 1.0725, 0.8322], device='cuda:1'), covar=tensor([0.2524, 0.2566, 0.1609, 0.2412], device='cuda:1'), in_proj_covar=tensor([0.1985, 0.1921, 0.1857, 0.1992], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 18:40:29,515 INFO [train.py:968] (1/2) Epoch 23, batch 4250, giga_loss[loss=0.2856, simple_loss=0.3529, pruned_loss=0.1092, over 28665.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09134, over 5707850.03 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3452, pruned_loss=0.08948, over 4993474.35 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3346, pruned_loss=0.09165, over 5700374.11 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:40:42,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6009, 1.8227, 1.7975, 1.5494], device='cuda:1'), covar=tensor([0.1915, 0.2090, 0.2191, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0743, 0.0712, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 18:40:53,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2114, 0.9624, 4.0261, 3.2964], device='cuda:1'), covar=tensor([0.1647, 0.2936, 0.0468, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0758, 0.0645, 0.0957, 0.0911], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 18:40:53,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1007669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:40:54,144 INFO [optim.py:369] (1/2) 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 18:40:55,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1007672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:12,767 INFO [train.py:968] (1/2) Epoch 23, batch 4300, giga_loss[loss=0.2274, simple_loss=0.3025, pruned_loss=0.07618, over 28898.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3329, pruned_loss=0.09118, over 5705981.76 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3451, pruned_loss=0.08948, over 4998326.64 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3326, pruned_loss=0.09145, over 5704620.74 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:41:13,616 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1007692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:15,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1007695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:20,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1007701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:39,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1007724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:41:45,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5248, 1.7714, 1.5763, 1.5663], device='cuda:1'), covar=tensor([0.1993, 0.2533, 0.2405, 0.2483], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0743, 0.0712, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 18:41:51,323 INFO [train.py:968] (1/2) Epoch 23, batch 4350, giga_loss[loss=0.2377, simple_loss=0.3068, pruned_loss=0.08431, over 28713.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3305, pruned_loss=0.08996, over 5709099.11 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.345, pruned_loss=0.08941, over 5023309.93 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3301, pruned_loss=0.09024, over 5704497.36 frames. ], batch size: 92, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:42:05,899 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-11 18:42:15,155 INFO [optim.py:369] (1/2) 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 18:42:31,418 INFO [train.py:968] (1/2) Epoch 23, batch 4400, giga_loss[loss=0.2564, simple_loss=0.3361, pruned_loss=0.08829, over 28711.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3291, pruned_loss=0.08903, over 5714705.25 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3452, pruned_loss=0.08962, over 5040483.43 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3283, pruned_loss=0.0891, over 5707962.39 frames. ], batch size: 284, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:43:14,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 18:43:15,607 INFO [train.py:968] (1/2) Epoch 23, batch 4450, giga_loss[loss=0.2259, simple_loss=0.3066, pruned_loss=0.07262, over 28897.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3319, pruned_loss=0.08985, over 5717178.30 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3451, pruned_loss=0.08956, over 5053737.43 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3312, pruned_loss=0.08995, over 5708843.53 frames. ], batch size: 106, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:43:45,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.16 vs. limit=5.0
+2023-03-11 18:43:45,163 INFO [optim.py:369] (1/2) 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 18:44:03,566 INFO [train.py:968] (1/2) Epoch 23, batch 4500, giga_loss[loss=0.2741, simple_loss=0.364, pruned_loss=0.09215, over 28693.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3342, pruned_loss=0.09104, over 5706499.73 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3448, pruned_loss=0.08942, over 5058117.75 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3338, pruned_loss=0.09124, over 5698948.21 frames. ], batch size: 284, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:44:34,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1007924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:44:50,108 INFO [train.py:968] (1/2) Epoch 23, batch 4550, giga_loss[loss=0.2452, simple_loss=0.3344, pruned_loss=0.07803, over 28898.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3376, pruned_loss=0.09199, over 5706137.75 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3449, pruned_loss=0.08951, over 5062314.16 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3371, pruned_loss=0.09208, over 5699405.97 frames. ], batch size: 174, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:44:54,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3149, 2.0152, 1.6112, 0.6588], device='cuda:1'), covar=tensor([0.5413, 0.3240, 0.4309, 0.6079], device='cuda:1'), in_proj_covar=tensor([0.1745, 0.1636, 0.1594, 0.1426], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 18:45:16,938 INFO [optim.py:369] (1/2) 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 18:45:35,150 INFO [train.py:968] (1/2) Epoch 23, batch 4600, giga_loss[loss=0.245, simple_loss=0.3303, pruned_loss=0.07991, over 27547.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3386, pruned_loss=0.09212, over 5699622.11 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3446, pruned_loss=0.08932, over 5090439.28 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3383, pruned_loss=0.09245, over 5688512.44 frames. ], batch size: 472, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:45:46,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1008005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:46:03,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6632, 1.8783, 1.3425, 1.4612], device='cuda:1'), covar=tensor([0.1030, 0.0690, 0.1130, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0444, 0.0521, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 18:46:17,746 INFO [train.py:968] (1/2) Epoch 23, batch 4650, giga_loss[loss=0.2665, simple_loss=0.3455, pruned_loss=0.09376, over 28912.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3374, pruned_loss=0.09106, over 5707822.76 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3449, pruned_loss=0.08936, over 5101753.69 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3368, pruned_loss=0.09132, over 5697031.33 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:46:39,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1008067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:46:43,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1008070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:46:43,357 INFO [optim.py:369] (1/2) 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 18:46:59,983 INFO [train.py:968] (1/2) Epoch 23, batch 4700, giga_loss[loss=0.2639, simple_loss=0.339, pruned_loss=0.09443, over 28864.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3377, pruned_loss=0.09186, over 5710269.21 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3444, pruned_loss=0.08922, over 5117527.45 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3375, pruned_loss=0.09219, over 5698455.81 frames. ], batch size: 186, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:47:07,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1008099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:47:10,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-11 18:47:38,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6730, 1.8143, 1.5132, 1.6154], device='cuda:1'), covar=tensor([0.2629, 0.2751, 0.3110, 0.2611], device='cuda:1'), in_proj_covar=tensor([0.1527, 0.1103, 0.1347, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 18:47:45,482 INFO [train.py:968] (1/2) Epoch 23, batch 4750, giga_loss[loss=0.2591, simple_loss=0.335, pruned_loss=0.09165, over 28432.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3388, pruned_loss=0.09297, over 5705631.11 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3442, pruned_loss=0.08905, over 5125411.31 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3387, pruned_loss=0.0934, over 5694448.44 frames. ], batch size: 71, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:47:50,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1008148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:47:53,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1008151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:48:09,011 INFO [optim.py:369] (1/2) 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 18:48:13,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0324, 1.3485, 5.3990, 3.6019], device='cuda:1'), covar=tensor([0.1475, 0.2751, 0.0382, 0.1077], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0648, 0.0961, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 18:48:15,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-11 18:48:20,246 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 4800, giga_loss[loss=0.2739, simple_loss=0.3449, pruned_loss=0.1015, over 28725.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.341, pruned_loss=0.09378, over 5706100.80 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3436, pruned_loss=0.0886, over 5162551.26 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3413, pruned_loss=0.09472, over 5688763.34 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:49:07,855 INFO [train.py:968] (1/2) Epoch 23, batch 4850, giga_loss[loss=0.2646, simple_loss=0.3379, pruned_loss=0.09567, over 28852.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3434, pruned_loss=0.09492, over 5713847.54 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3432, pruned_loss=0.08842, over 5177528.37 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.344, pruned_loss=0.09593, over 5696443.31 frames. ], batch size: 112, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:49:32,540 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 4900, giga_loss[loss=0.2839, simple_loss=0.3466, pruned_loss=0.1106, over 28773.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3452, pruned_loss=0.09551, over 5721015.52 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3426, pruned_loss=0.0881, over 5199010.22 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09678, over 5702231.84 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:50:30,609 INFO [train.py:968] (1/2) Epoch 23, batch 4950, giga_loss[loss=0.3201, simple_loss=0.3734, pruned_loss=0.1334, over 28829.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3471, pruned_loss=0.09652, over 5715099.10 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3429, pruned_loss=0.08828, over 5206794.82 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3477, pruned_loss=0.09757, over 5704462.98 frames. ], batch size: 99, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:50:48,907 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1008365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:50:54,741 INFO [optim.py:369] (1/2) 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 18:51:11,017 INFO [train.py:968] (1/2) Epoch 23, batch 5000, giga_loss[loss=0.279, simple_loss=0.3467, pruned_loss=0.1056, over 28421.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3476, pruned_loss=0.09681, over 5720357.65 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3432, pruned_loss=0.08844, over 5213587.12 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3479, pruned_loss=0.09762, over 5712641.48 frames. ], batch size: 71, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:51:28,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2217, 2.5453, 1.2612, 1.2883], device='cuda:1'), covar=tensor([0.0956, 0.0382, 0.0933, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0553, 0.0387, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 18:51:50,481 INFO [train.py:968] (1/2) Epoch 23, batch 5050, giga_loss[loss=0.2672, simple_loss=0.3444, pruned_loss=0.09502, over 28937.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3473, pruned_loss=0.09682, over 5716805.37 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3435, pruned_loss=0.08857, over 5221307.42 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3475, pruned_loss=0.09759, over 5716879.99 frames. ], batch size: 199, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:52:14,639 INFO [optim.py:369] (1/2) 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 18:52:30,438 INFO [train.py:968] (1/2) Epoch 23, batch 5100, libri_loss[loss=0.2644, simple_loss=0.3482, pruned_loss=0.09027, over 29634.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3452, pruned_loss=0.09598, over 5717374.21 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3431, pruned_loss=0.08843, over 5246408.42 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3457, pruned_loss=0.097, over 5712344.15 frames. ], batch size: 88, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:53:10,915 INFO [train.py:968] (1/2) Epoch 23, batch 5150, giga_loss[loss=0.2352, simple_loss=0.3097, pruned_loss=0.08031, over 28916.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3415, pruned_loss=0.0939, over 5726005.04 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3433, pruned_loss=0.08847, over 5264112.79 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3418, pruned_loss=0.09488, over 5718743.27 frames. ], batch size: 112, lr: 1.40e-03, grad_scale: 4.0
+2023-03-11 18:53:37,417 INFO [optim.py:369] (1/2) 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 18:53:52,285 INFO [train.py:968] (1/2) Epoch 23, batch 5200, giga_loss[loss=0.2396, simple_loss=0.3168, pruned_loss=0.08115, over 28940.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3394, pruned_loss=0.09271, over 5728299.16 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3435, pruned_loss=0.08879, over 5276837.28 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3394, pruned_loss=0.09328, over 5719178.67 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:54:28,445 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 5250, giga_loss[loss=0.3011, simple_loss=0.388, pruned_loss=0.107, over 29003.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3402, pruned_loss=0.09228, over 5718754.30 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3439, pruned_loss=0.08914, over 5286156.14 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3398, pruned_loss=0.09249, over 5709325.44 frames. ], batch size: 213, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:55:04,645 INFO [optim.py:369] (1/2) 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 18:55:12,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5390, 1.7690, 1.4497, 1.5520], device='cuda:1'), covar=tensor([0.2885, 0.3041, 0.3446, 0.2700], device='cuda:1'), in_proj_covar=tensor([0.1520, 0.1099, 0.1344, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 18:55:14,039 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1008682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:55:21,032 INFO [train.py:968] (1/2) Epoch 23, batch 5300, giga_loss[loss=0.2333, simple_loss=0.3223, pruned_loss=0.07216, over 28958.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3416, pruned_loss=0.09236, over 5714747.34 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3441, pruned_loss=0.08927, over 5295080.04 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3411, pruned_loss=0.09248, over 5706860.97 frames. ], batch size: 145, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:56:03,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1008740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:56:03,862 INFO [train.py:968] (1/2) Epoch 23, batch 5350, giga_loss[loss=0.2629, simple_loss=0.342, pruned_loss=0.0919, over 28361.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3425, pruned_loss=0.09338, over 5711865.44 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3443, pruned_loss=0.08927, over 5313284.60 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3419, pruned_loss=0.09357, over 5700440.83 frames. ], batch size: 368, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:56:24,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4222, 1.7687, 1.4158, 1.0404], device='cuda:1'), covar=tensor([0.2678, 0.2791, 0.3114, 0.2464], device='cuda:1'), in_proj_covar=tensor([0.1516, 0.1096, 0.1340, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 18:56:28,727 INFO [optim.py:369] (1/2) 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 18:56:35,678 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1008779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:56:45,942 INFO [train.py:968] (1/2) Epoch 23, batch 5400, giga_loss[loss=0.2461, simple_loss=0.3173, pruned_loss=0.08748, over 28499.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.341, pruned_loss=0.09395, over 5713479.85 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3438, pruned_loss=0.08914, over 5325185.25 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3408, pruned_loss=0.09429, over 5700997.55 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:57:29,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3610, 1.6858, 1.3670, 1.5438], device='cuda:1'), covar=tensor([0.0724, 0.0318, 0.0332, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0187, 0.0118, 0.0117, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0108], device='cuda:1')
+2023-03-11 18:57:29,451 INFO [train.py:968] (1/2) Epoch 23, batch 5450, giga_loss[loss=0.2578, simple_loss=0.3324, pruned_loss=0.09165, over 28620.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3402, pruned_loss=0.09496, over 5711446.58 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3435, pruned_loss=0.08899, over 5333705.97 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3403, pruned_loss=0.09544, over 5699049.06 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:57:52,754 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-11 18:57:57,246 INFO [optim.py:369] (1/2) 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 18:58:07,221 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1008883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:58:09,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1008886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:58:12,059 INFO [train.py:968] (1/2) Epoch 23, batch 5500, giga_loss[loss=0.2672, simple_loss=0.3405, pruned_loss=0.09697, over 28548.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3386, pruned_loss=0.09492, over 5713018.58 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3433, pruned_loss=0.08885, over 5346310.39 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3388, pruned_loss=0.09556, over 5700085.41 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:58:33,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1008915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 18:58:56,821 INFO [train.py:968] (1/2) Epoch 23, batch 5550, giga_loss[loss=0.2579, simple_loss=0.3331, pruned_loss=0.09134, over 28572.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3368, pruned_loss=0.0941, over 5715628.71 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3436, pruned_loss=0.0891, over 5356373.21 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3366, pruned_loss=0.09449, over 5702873.62 frames. ], batch size: 71, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:58:58,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-11 18:59:22,969 INFO [optim.py:369] (1/2) 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 18:59:40,261 INFO [train.py:968] (1/2) Epoch 23, batch 5600, giga_loss[loss=0.2216, simple_loss=0.2918, pruned_loss=0.07573, over 28432.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3337, pruned_loss=0.09225, over 5720043.68 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3431, pruned_loss=0.08884, over 5364739.92 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3338, pruned_loss=0.09282, over 5707413.42 frames. ], batch size: 85, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 18:59:54,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:00:17,967 INFO [train.py:968] (1/2) Epoch 23, batch 5650, giga_loss[loss=0.2451, simple_loss=0.3258, pruned_loss=0.08217, over 28628.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3308, pruned_loss=0.09069, over 5730733.97 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3437, pruned_loss=0.08929, over 5389730.18 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3297, pruned_loss=0.09089, over 5714972.55 frames. ], batch size: 307, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:00:30,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:00:32,959 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-11 19:00:33,544 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 19:00:41,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:00:42,049 INFO [optim.py:369] (1/2) 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 19:00:49,676 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-11 19:00:56,533 INFO [train.py:968] (1/2) Epoch 23, batch 5700, giga_loss[loss=0.2351, simple_loss=0.3135, pruned_loss=0.07835, over 28727.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.327, pruned_loss=0.08874, over 5726267.97 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3437, pruned_loss=0.08946, over 5398604.46 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3257, pruned_loss=0.08877, over 5716009.43 frames. ], batch size: 284, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:01:37,436 INFO [train.py:968] (1/2) Epoch 23, batch 5750, giga_loss[loss=0.246, simple_loss=0.3286, pruned_loss=0.08176, over 28839.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3269, pruned_loss=0.08814, over 5727301.76 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3434, pruned_loss=0.08932, over 5412183.64 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3257, pruned_loss=0.08827, over 5715362.30 frames. ], batch size: 186, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:01:44,818 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:01:46,263 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009152.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:01:46,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:01:47,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:02,086 INFO [optim.py:369] (1/2) 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 19:02:10,232 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:12,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1991, 1.3003, 1.1341, 0.9917], device='cuda:1'), covar=tensor([0.0975, 0.0540, 0.1116, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0445, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 19:02:16,778 INFO [train.py:968] (1/2) Epoch 23, batch 5800, giga_loss[loss=0.2907, simple_loss=0.3619, pruned_loss=0.1098, over 28416.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3303, pruned_loss=0.08937, over 5732759.70 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3437, pruned_loss=0.08957, over 5430079.54 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3286, pruned_loss=0.08925, over 5718020.08 frames. ], batch size: 71, lr: 1.40e-03, grad_scale: 8.0
+2023-03-11 19:02:25,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:28,484 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:36,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4583, 3.7079, 1.5480, 1.5770], device='cuda:1'), covar=tensor([0.0966, 0.0349, 0.0920, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0554, 0.0388, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 19:02:51,171 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:02:58,564 INFO [train.py:968] (1/2) Epoch 23, batch 5850, libri_loss[loss=0.2285, simple_loss=0.3155, pruned_loss=0.07074, over 29560.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3336, pruned_loss=0.09075, over 5732541.43 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08915, over 5445213.99 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3324, pruned_loss=0.09107, over 5716248.88 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:03:03,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5010, 1.7271, 1.7357, 1.2960], device='cuda:1'), covar=tensor([0.1850, 0.2467, 0.1546, 0.1727], device='cuda:1'), in_proj_covar=tensor([0.0904, 0.0701, 0.0950, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 19:03:25,502 INFO [optim.py:369] (1/2) 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 19:03:41,422 INFO [train.py:968] (1/2) Epoch 23, batch 5900, giga_loss[loss=0.2567, simple_loss=0.3463, pruned_loss=0.08361, over 28837.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3368, pruned_loss=0.0922, over 5728150.40 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3428, pruned_loss=0.08911, over 5449855.43 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.336, pruned_loss=0.09253, over 5713649.33 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:03:49,700 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:03:51,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:04:13,367 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.10 vs. limit=5.0
+2023-03-11 19:04:15,018 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 5950, giga_loss[loss=0.2678, simple_loss=0.3489, pruned_loss=0.09337, over 28764.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3396, pruned_loss=0.09341, over 5718605.58 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3429, pruned_loss=0.08911, over 5460082.57 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3387, pruned_loss=0.09383, over 5708868.42 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:04:29,809 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:05:10,528 INFO [train.py:968] (1/2) Epoch 23, batch 6000, giga_loss[loss=0.244, simple_loss=0.3253, pruned_loss=0.08138, over 28860.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3432, pruned_loss=0.09623, over 5713811.14 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3428, pruned_loss=0.08911, over 5464758.07 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3425, pruned_loss=0.09659, over 5704476.91 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:05:10,528 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 19:05:18,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5062, 1.7012, 1.3883, 1.4134], device='cuda:1'), covar=tensor([0.0825, 0.0390, 0.0857, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0446, 0.0520, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 19:05:19,129 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 19:06:04,417 INFO [train.py:968] (1/2) Epoch 23, batch 6050, giga_loss[loss=0.3657, simple_loss=0.41, pruned_loss=0.1607, over 27520.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3493, pruned_loss=0.1014, over 5716437.81 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3425, pruned_loss=0.08884, over 5475874.30 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3491, pruned_loss=0.1021, over 5704628.57 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:06:04,619 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:06:10,056 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:06:54,319 INFO [train.py:968] (1/2) Epoch 23, batch 6100, giga_loss[loss=0.3132, simple_loss=0.3607, pruned_loss=0.1328, over 23822.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3555, pruned_loss=0.1064, over 5694368.16 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3428, pruned_loss=0.089, over 5479756.71 frames. ], giga_tot_loss[loss=0.2848, simple_loss=0.3553, pruned_loss=0.1071, over 5686652.58 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:07:04,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5048, 1.6410, 1.5232, 1.3348], device='cuda:1'), covar=tensor([0.2604, 0.2469, 0.2136, 0.2487], device='cuda:1'), in_proj_covar=tensor([0.1974, 0.1923, 0.1852, 0.1977], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 19:07:30,560 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009527.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:07:42,582 INFO [train.py:968] (1/2) Epoch 23, batch 6150, giga_loss[loss=0.3414, simple_loss=0.3993, pruned_loss=0.1417, over 28958.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3631, pruned_loss=0.1117, over 5689817.32 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08904, over 5488491.90 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3635, pruned_loss=0.1129, over 5681058.25 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:08:20,249 INFO [optim.py:369] (1/2) 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 19:08:36,149 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:08:37,400 INFO [train.py:968] (1/2) Epoch 23, batch 6200, giga_loss[loss=0.3501, simple_loss=0.4048, pruned_loss=0.1477, over 28888.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3695, pruned_loss=0.1177, over 5671226.46 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08922, over 5487623.63 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.37, pruned_loss=0.1189, over 5668106.67 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:08:38,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:09:06,500 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:09:24,544 INFO [train.py:968] (1/2) Epoch 23, batch 6250, libri_loss[loss=0.2232, simple_loss=0.3052, pruned_loss=0.0706, over 29354.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3742, pruned_loss=0.1215, over 5676679.60 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08911, over 5489624.37 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3752, pruned_loss=0.123, over 5675129.46 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:09:53,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009670.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:09:55,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009673.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:09:56,961 INFO [optim.py:369] (1/2) 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 19:10:14,581 INFO [train.py:968] (1/2) Epoch 23, batch 6300, giga_loss[loss=0.2977, simple_loss=0.3653, pruned_loss=0.115, over 28826.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3763, pruned_loss=0.1239, over 5657660.99 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3423, pruned_loss=0.089, over 5503621.54 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3787, pruned_loss=0.1265, over 5649809.15 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:10:26,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009702.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:10:48,603 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:11:05,798 INFO [train.py:968] (1/2) Epoch 23, batch 6350, giga_loss[loss=0.2889, simple_loss=0.3593, pruned_loss=0.1092, over 28956.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3789, pruned_loss=0.1268, over 5658428.38 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08909, over 5509488.08 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3813, pruned_loss=0.1293, over 5648968.69 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:11:07,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1036, 1.0271, 3.4665, 3.0838], device='cuda:1'), covar=tensor([0.1677, 0.2807, 0.0498, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0764, 0.0650, 0.0964, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 19:11:16,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:11:20,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4070, 1.5241, 1.6904, 1.4317], device='cuda:1'), covar=tensor([0.1477, 0.1496, 0.1461, 0.1475], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0753, 0.0717, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 19:11:44,075 INFO [optim.py:369] (1/2) 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 19:12:01,386 INFO [train.py:968] (1/2) Epoch 23, batch 6400, giga_loss[loss=0.448, simple_loss=0.4655, pruned_loss=0.2153, over 26620.00 frames. ], tot_loss[loss=0.324, simple_loss=0.3836, pruned_loss=0.1322, over 5638081.07 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3426, pruned_loss=0.08918, over 5518930.52 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3862, pruned_loss=0.135, over 5625204.48 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:12:27,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1009816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:12:43,477 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:12:58,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-11 19:12:58,710 INFO [train.py:968] (1/2) Epoch 23, batch 6450, giga_loss[loss=0.3803, simple_loss=0.4008, pruned_loss=0.1799, over 23289.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3877, pruned_loss=0.1364, over 5602801.88 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3427, pruned_loss=0.08927, over 5508047.18 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3904, pruned_loss=0.1393, over 5603567.23 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:13:16,396 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009856.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:13:27,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:13:29,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:13:38,964 INFO [optim.py:369] (1/2) 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 19:13:57,162 INFO [train.py:968] (1/2) Epoch 23, batch 6500, giga_loss[loss=0.4791, simple_loss=0.4814, pruned_loss=0.2384, over 26444.00 frames. ], tot_loss[loss=0.3339, simple_loss=0.3907, pruned_loss=0.1386, over 5607206.16 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3426, pruned_loss=0.08922, over 5508199.54 frames. ], giga_tot_loss[loss=0.3379, simple_loss=0.3932, pruned_loss=0.1413, over 5608850.92 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:14:03,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:14:32,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 19:14:39,765 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1009932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:14:46,542 INFO [train.py:968] (1/2) Epoch 23, batch 6550, giga_loss[loss=0.2954, simple_loss=0.3592, pruned_loss=0.1158, over 28083.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3887, pruned_loss=0.1375, over 5630566.81 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3426, pruned_loss=0.08921, over 5516019.66 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3916, pruned_loss=0.1406, over 5626777.65 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:15:06,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1009959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:15:09,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1009962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:15:22,733 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 6600, giga_loss[loss=0.3066, simple_loss=0.3757, pruned_loss=0.1188, over 28896.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3875, pruned_loss=0.137, over 5635769.04 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3427, pruned_loss=0.08927, over 5524984.93 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3908, pruned_loss=0.1405, over 5627237.70 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:15:38,930 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1009991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:15:53,038 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3329, 1.9785, 1.4537, 0.5283], device='cuda:1'), covar=tensor([0.4881, 0.2810, 0.3827, 0.6178], device='cuda:1'), in_proj_covar=tensor([0.1749, 0.1650, 0.1600, 0.1432], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 19:16:28,190 INFO [train.py:968] (1/2) Epoch 23, batch 6650, libri_loss[loss=0.2678, simple_loss=0.3551, pruned_loss=0.09028, over 29132.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3875, pruned_loss=0.1357, over 5640188.59 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08938, over 5534152.43 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3912, pruned_loss=0.1398, over 5627809.65 frames. ], batch size: 101, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:17:04,103 INFO [optim.py:369] (1/2) 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 19:17:12,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:20,558 INFO [train.py:968] (1/2) Epoch 23, batch 6700, giga_loss[loss=0.391, simple_loss=0.4146, pruned_loss=0.1837, over 23299.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3884, pruned_loss=0.1358, over 5638893.07 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08949, over 5538934.89 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.392, pruned_loss=0.1398, over 5626726.91 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:17:21,333 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:34,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:56,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:17:57,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:18:13,742 INFO [train.py:968] (1/2) Epoch 23, batch 6750, giga_loss[loss=0.4169, simple_loss=0.4393, pruned_loss=0.1972, over 24296.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3895, pruned_loss=0.1367, over 5597178.17 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3438, pruned_loss=0.09004, over 5518513.33 frames. ], giga_tot_loss[loss=0.3363, simple_loss=0.3924, pruned_loss=0.1401, over 5605813.89 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:18:47,245 INFO [optim.py:369] (1/2) 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 19:19:05,531 INFO [train.py:968] (1/2) Epoch 23, batch 6800, giga_loss[loss=0.3344, simple_loss=0.3714, pruned_loss=0.1487, over 23893.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3847, pruned_loss=0.1324, over 5596431.41 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3435, pruned_loss=0.08984, over 5519612.68 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3887, pruned_loss=0.1366, over 5604371.69 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:19:13,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9343, 3.7491, 3.5454, 1.7864], device='cuda:1'), covar=tensor([0.0700, 0.0838, 0.0870, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.1244, 0.1148, 0.0979, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 19:19:16,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:19:47,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:19:54,858 INFO [train.py:968] (1/2) Epoch 23, batch 6850, giga_loss[loss=0.2799, simple_loss=0.3582, pruned_loss=0.1008, over 28776.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.382, pruned_loss=0.1285, over 5603185.86 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09018, over 5509770.98 frames. ], giga_tot_loss[loss=0.325, simple_loss=0.3854, pruned_loss=0.1323, over 5618917.79 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:20:05,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:08,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:21,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:21,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6089, 1.8821, 1.2676, 1.4729], device='cuda:1'), covar=tensor([0.0947, 0.0553, 0.1055, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0449, 0.0522, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 19:20:23,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:27,932 INFO [optim.py:369] (1/2) 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 19:20:38,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:20:38,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-11 19:20:46,927 INFO [train.py:968] (1/2) Epoch 23, batch 6900, giga_loss[loss=0.2632, simple_loss=0.3411, pruned_loss=0.09266, over 28886.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3771, pruned_loss=0.1243, over 5623215.45 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.09007, over 5515631.75 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3804, pruned_loss=0.1278, over 5631932.55 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:20:49,114 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010293.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:20:57,406 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:21:02,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-11 19:21:03,950 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3606, 1.7104, 1.7172, 1.4993], device='cuda:1'), covar=tensor([0.2145, 0.2184, 0.2454, 0.2296], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0756, 0.0720, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 19:21:04,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:21:25,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-11 19:21:37,794 INFO [train.py:968] (1/2) Epoch 23, batch 6950, libri_loss[loss=0.2351, simple_loss=0.3198, pruned_loss=0.0752, over 29525.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3744, pruned_loss=0.122, over 5630342.48 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3442, pruned_loss=0.09025, over 5523408.91 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3778, pruned_loss=0.1258, over 5633355.32 frames. ], batch size: 80, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:21:42,498 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:21:44,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:22:07,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010374.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:22:08,590 INFO [optim.py:369] (1/2) 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 19:22:10,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:22:11,588 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:22:23,136 INFO [train.py:968] (1/2) Epoch 23, batch 7000, giga_loss[loss=0.3294, simple_loss=0.3944, pruned_loss=0.1322, over 28501.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1202, over 5646160.24 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3437, pruned_loss=0.08998, over 5534960.01 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3758, pruned_loss=0.1244, over 5641058.46 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:22:38,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:23:00,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:03,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-11 19:23:16,236 INFO [train.py:968] (1/2) Epoch 23, batch 7050, giga_loss[loss=0.3073, simple_loss=0.3759, pruned_loss=0.1193, over 28480.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.372, pruned_loss=0.1204, over 5649907.73 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3439, pruned_loss=0.09007, over 5535237.85 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3752, pruned_loss=0.1238, over 5646530.16 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:23:28,300 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:31,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:35,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:45,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:23:54,821 INFO [optim.py:369] (1/2) 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 19:24:03,855 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:24:11,625 INFO [train.py:968] (1/2) Epoch 23, batch 7100, libri_loss[loss=0.2908, simple_loss=0.3757, pruned_loss=0.1029, over 29527.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3713, pruned_loss=0.1192, over 5646164.86 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3442, pruned_loss=0.09024, over 5531567.66 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3745, pruned_loss=0.1227, over 5650041.93 frames. ], batch size: 89, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:24:23,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:25:06,860 INFO [train.py:968] (1/2) Epoch 23, batch 7150, giga_loss[loss=0.3119, simple_loss=0.3843, pruned_loss=0.1198, over 27651.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3707, pruned_loss=0.1174, over 5655607.73 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3443, pruned_loss=0.09039, over 5536947.07 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3735, pruned_loss=0.1205, over 5655462.78 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:25:29,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1010557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:25:52,112 INFO [optim.py:369] (1/2) 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 19:25:53,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6178, 1.7863, 1.4975, 1.8400], device='cuda:1'), covar=tensor([0.2392, 0.2563, 0.2673, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.1520, 0.1097, 0.1346, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 19:26:05,054 INFO [train.py:968] (1/2) Epoch 23, batch 7200, giga_loss[loss=0.2924, simple_loss=0.3756, pruned_loss=0.1046, over 28680.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3721, pruned_loss=0.1167, over 5659575.99 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09022, over 5540444.36 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3748, pruned_loss=0.1194, over 5657262.83 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:26:16,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:18,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:23,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:30,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:31,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.52 vs. limit=5.0
+2023-03-11 19:26:49,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:26:58,896 INFO [train.py:968] (1/2) Epoch 23, batch 7250, giga_loss[loss=0.2846, simple_loss=0.357, pruned_loss=0.1061, over 28533.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3727, pruned_loss=0.1174, over 5653541.40 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3441, pruned_loss=0.09041, over 5538569.82 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3754, pruned_loss=0.12, over 5656406.49 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:27:00,293 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:03,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:06,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:25,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010668.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:27:34,948 INFO [optim.py:369] (1/2) 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 19:27:35,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:27:49,112 INFO [train.py:968] (1/2) Epoch 23, batch 7300, giga_loss[loss=0.2961, simple_loss=0.3627, pruned_loss=0.1148, over 28821.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3725, pruned_loss=0.1178, over 5663292.28 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3442, pruned_loss=0.09051, over 5544204.04 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3749, pruned_loss=0.1202, over 5662148.33 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:28:37,546 INFO [train.py:968] (1/2) Epoch 23, batch 7350, giga_loss[loss=0.334, simple_loss=0.3703, pruned_loss=0.1489, over 23617.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3707, pruned_loss=0.1175, over 5661746.52 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3444, pruned_loss=0.09072, over 5555910.62 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3735, pruned_loss=0.1203, over 5654304.77 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:29:11,881 INFO [optim.py:369] (1/2) 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 19:29:21,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4908, 2.2476, 1.6777, 0.7628], device='cuda:1'), covar=tensor([0.6996, 0.3296, 0.4364, 0.6882], device='cuda:1'), in_proj_covar=tensor([0.1754, 0.1650, 0.1601, 0.1431], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 19:29:22,565 INFO [train.py:968] (1/2) Epoch 23, batch 7400, giga_loss[loss=0.284, simple_loss=0.358, pruned_loss=0.105, over 29022.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3684, pruned_loss=0.117, over 5661950.89 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.344, pruned_loss=0.09056, over 5554985.72 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1202, over 5660889.19 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:29:23,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3170, 3.6429, 1.5758, 1.5949], device='cuda:1'), covar=tensor([0.1070, 0.0369, 0.0876, 0.1331], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0563, 0.0392, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 19:29:34,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:29:39,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010811.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:29:41,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010814.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 19:30:10,011 INFO [train.py:968] (1/2) Epoch 23, batch 7450, giga_loss[loss=0.3165, simple_loss=0.3731, pruned_loss=0.1299, over 27496.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3661, pruned_loss=0.1152, over 5664666.24 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3438, pruned_loss=0.09035, over 5557611.88 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3697, pruned_loss=0.1187, over 5664575.63 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:30:11,889 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1010843.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 19:30:48,274 INFO [optim.py:369] (1/2) 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 19:31:03,153 INFO [train.py:968] (1/2) Epoch 23, batch 7500, giga_loss[loss=0.2918, simple_loss=0.3681, pruned_loss=0.1078, over 28810.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3659, pruned_loss=0.1134, over 5681459.53 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3437, pruned_loss=0.09035, over 5560875.94 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.369, pruned_loss=0.1164, over 5679504.59 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:31:42,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1010932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:31:51,308 INFO [train.py:968] (1/2) Epoch 23, batch 7550, giga_loss[loss=0.3011, simple_loss=0.3726, pruned_loss=0.1148, over 28896.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3651, pruned_loss=0.1122, over 5694611.20 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3434, pruned_loss=0.09018, over 5567906.68 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3683, pruned_loss=0.1152, over 5688943.57 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:31:59,084 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1010947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:32:01,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1010950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:32:27,250 INFO [optim.py:369] (1/2) 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 19:32:29,466 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 7600, giga_loss[loss=0.2794, simple_loss=0.35, pruned_loss=0.1044, over 28858.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.365, pruned_loss=0.1127, over 5693000.83 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3436, pruned_loss=0.09029, over 5570990.51 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3675, pruned_loss=0.1152, over 5686847.05 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:33:20,495 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 7650, giga_loss[loss=0.2741, simple_loss=0.3383, pruned_loss=0.1049, over 28782.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.364, pruned_loss=0.1131, over 5678206.02 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.344, pruned_loss=0.09064, over 5559278.89 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.366, pruned_loss=0.1152, over 5686825.31 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:34:04,380 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:34:05,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-11 19:34:06,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4965, 1.9122, 1.4344, 1.6662], device='cuda:1'), covar=tensor([0.2926, 0.2838, 0.3401, 0.2396], device='cuda:1'), in_proj_covar=tensor([0.1520, 0.1096, 0.1347, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 19:34:06,780 INFO [optim.py:369] (1/2) 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 19:34:07,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:34:19,213 INFO [train.py:968] (1/2) Epoch 23, batch 7700, giga_loss[loss=0.2621, simple_loss=0.3333, pruned_loss=0.09542, over 28983.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3641, pruned_loss=0.1135, over 5668490.28 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3439, pruned_loss=0.09052, over 5561647.83 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3666, pruned_loss=0.1162, over 5678082.15 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:34:33,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-11 19:34:34,906 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:35:06,808 INFO [train.py:968] (1/2) Epoch 23, batch 7750, giga_loss[loss=0.2745, simple_loss=0.3452, pruned_loss=0.1019, over 29014.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3638, pruned_loss=0.1139, over 5683260.20 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3436, pruned_loss=0.0904, over 5572063.54 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3666, pruned_loss=0.1168, over 5684634.01 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:35:37,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:35:37,873 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 19:35:45,035 INFO [optim.py:369] (1/2) 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 19:35:55,903 INFO [train.py:968] (1/2) Epoch 23, batch 7800, libri_loss[loss=0.2481, simple_loss=0.3227, pruned_loss=0.08677, over 29701.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3618, pruned_loss=0.1128, over 5683919.00 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3434, pruned_loss=0.09034, over 5573173.57 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3651, pruned_loss=0.1161, over 5688529.07 frames. ], batch size: 69, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:36:44,743 INFO [train.py:968] (1/2) Epoch 23, batch 7850, giga_loss[loss=0.2611, simple_loss=0.3331, pruned_loss=0.09452, over 28204.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3608, pruned_loss=0.113, over 5678651.45 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3433, pruned_loss=0.0904, over 5566364.68 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3636, pruned_loss=0.1158, over 5690324.22 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:36:45,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:37:19,549 INFO [optim.py:369] (1/2) 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 19:37:30,834 INFO [train.py:968] (1/2) Epoch 23, batch 7900, giga_loss[loss=0.2913, simple_loss=0.3633, pruned_loss=0.1096, over 28908.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1124, over 5688320.52 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3429, pruned_loss=0.09008, over 5574916.53 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3628, pruned_loss=0.1155, over 5692207.67 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:38:21,367 INFO [train.py:968] (1/2) Epoch 23, batch 7950, giga_loss[loss=0.3317, simple_loss=0.3847, pruned_loss=0.1393, over 27579.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3613, pruned_loss=0.1134, over 5677989.36 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3434, pruned_loss=0.09038, over 5578657.54 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3636, pruned_loss=0.1161, over 5680456.52 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:38:56,894 INFO [optim.py:369] (1/2) 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 19:39:09,543 INFO [train.py:968] (1/2) Epoch 23, batch 8000, libri_loss[loss=0.2411, simple_loss=0.3268, pruned_loss=0.07769, over 29588.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3615, pruned_loss=0.1128, over 5677024.93 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3434, pruned_loss=0.09033, over 5585978.78 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3639, pruned_loss=0.1156, over 5674583.34 frames. ], batch size: 75, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:39:21,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1011404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:39:58,792 INFO [train.py:968] (1/2) Epoch 23, batch 8050, giga_loss[loss=0.2926, simple_loss=0.3733, pruned_loss=0.106, over 28838.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3632, pruned_loss=0.1137, over 5672370.90 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3438, pruned_loss=0.09058, over 5590435.77 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3651, pruned_loss=0.1161, over 5668301.45 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 19:40:37,034 INFO [optim.py:369] (1/2) 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 19:40:47,763 INFO [train.py:968] (1/2) Epoch 23, batch 8100, giga_loss[loss=0.2844, simple_loss=0.3536, pruned_loss=0.1076, over 28964.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3633, pruned_loss=0.1139, over 5682857.29 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3434, pruned_loss=0.09046, over 5599879.68 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3656, pruned_loss=0.1166, over 5673232.75 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:41:39,292 INFO [train.py:968] (1/2) Epoch 23, batch 8150, giga_loss[loss=0.2681, simple_loss=0.3423, pruned_loss=0.09697, over 28865.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.367, pruned_loss=0.1171, over 5683548.12 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3433, pruned_loss=0.09027, over 5604673.21 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1197, over 5672858.17 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:41:40,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1011542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:41:45,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:41:51,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:42:21,588 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:42:35,485 INFO [train.py:968] (1/2) Epoch 23, batch 8200, giga_loss[loss=0.281, simple_loss=0.3538, pruned_loss=0.1041, over 28894.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3685, pruned_loss=0.1194, over 5686612.54 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3436, pruned_loss=0.09045, over 5606176.85 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5677283.06 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:42:41,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3990, 3.8211, 1.5906, 1.4978], device='cuda:1'), covar=tensor([0.0956, 0.0410, 0.0880, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0561, 0.0391, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 19:42:52,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8742, 5.7116, 5.4022, 2.7420], device='cuda:1'), covar=tensor([0.0446, 0.0607, 0.0769, 0.1714], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1166, 0.0992, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 19:43:02,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1011616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:43:02,630 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 19:43:08,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6337, 1.7370, 1.8589, 1.4236], device='cuda:1'), covar=tensor([0.1761, 0.2442, 0.1428, 0.1649], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0704, 0.0948, 0.0847], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 19:43:27,032 INFO [train.py:968] (1/2) Epoch 23, batch 8250, giga_loss[loss=0.2907, simple_loss=0.355, pruned_loss=0.1132, over 28736.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3701, pruned_loss=0.1218, over 5674498.08 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3437, pruned_loss=0.09043, over 5608076.47 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3718, pruned_loss=0.1241, over 5666253.91 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:43:43,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3495, 2.0019, 1.4935, 0.5654], device='cuda:1'), covar=tensor([0.5205, 0.2942, 0.4009, 0.6623], device='cuda:1'), in_proj_covar=tensor([0.1764, 0.1664, 0.1605, 0.1440], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 19:44:07,367 INFO [optim.py:369] (1/2) 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 19:44:13,769 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:44:17,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:44:20,148 INFO [train.py:968] (1/2) Epoch 23, batch 8300, giga_loss[loss=0.372, simple_loss=0.4091, pruned_loss=0.1674, over 24099.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3705, pruned_loss=0.1228, over 5665131.77 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3437, pruned_loss=0.09042, over 5611045.33 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3721, pruned_loss=0.1249, over 5656637.34 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:44:44,832 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:07,786 INFO [train.py:968] (1/2) Epoch 23, batch 8350, giga_loss[loss=0.2587, simple_loss=0.3263, pruned_loss=0.09554, over 28724.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3689, pruned_loss=0.1218, over 5669992.37 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09011, over 5616762.04 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3709, pruned_loss=0.1243, over 5659076.08 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:45:10,130 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:23,178 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1011759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:25,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1011762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:45:41,627 INFO [optim.py:369] (1/2) 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 19:45:50,793 INFO [train.py:968] (1/2) Epoch 23, batch 8400, giga_loss[loss=0.2812, simple_loss=0.359, pruned_loss=0.1017, over 28923.00 frames. ], tot_loss[loss=0.305, simple_loss=0.369, pruned_loss=0.1205, over 5681241.20 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3441, pruned_loss=0.09063, over 5624437.53 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3706, pruned_loss=0.1228, over 5667114.53 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:45:53,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1011791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:46:03,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1011804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:46:20,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4029, 4.1147, 1.6156, 1.5884], device='cuda:1'), covar=tensor([0.1055, 0.0309, 0.0920, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0562, 0.0391, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 19:46:34,503 INFO [train.py:968] (1/2) Epoch 23, batch 8450, giga_loss[loss=0.3162, simple_loss=0.3779, pruned_loss=0.1272, over 28870.00 frames. ], tot_loss[loss=0.3, simple_loss=0.366, pruned_loss=0.1171, over 5677711.52 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3444, pruned_loss=0.09065, over 5635381.25 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.368, pruned_loss=0.1202, over 5657849.45 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:47:09,600 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 8500, giga_loss[loss=0.3371, simple_loss=0.3841, pruned_loss=0.1451, over 27506.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.366, pruned_loss=0.1177, over 5672213.76 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3448, pruned_loss=0.09088, over 5627381.39 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3676, pruned_loss=0.1203, over 5664476.43 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:47:25,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6198, 4.6097, 1.7786, 1.7439], device='cuda:1'), covar=tensor([0.0988, 0.0298, 0.0846, 0.1297], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0561, 0.0390, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 19:47:49,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2149, 1.5305, 1.5067, 1.3021], device='cuda:1'), covar=tensor([0.2018, 0.1704, 0.2416, 0.2007], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0748, 0.0714, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 19:47:55,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6870, 1.8740, 1.6164, 1.7634], device='cuda:1'), covar=tensor([0.2247, 0.2290, 0.2338, 0.2335], device='cuda:1'), in_proj_covar=tensor([0.1523, 0.1097, 0.1345, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 19:48:08,893 INFO [train.py:968] (1/2) Epoch 23, batch 8550, giga_loss[loss=0.3456, simple_loss=0.394, pruned_loss=0.1486, over 27870.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3631, pruned_loss=0.1162, over 5667170.26 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3447, pruned_loss=0.0908, over 5621591.44 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3647, pruned_loss=0.1187, over 5666628.48 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:48:47,667 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 8600, giga_loss[loss=0.36, simple_loss=0.3902, pruned_loss=0.1649, over 23529.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3647, pruned_loss=0.1181, over 5645682.93 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3448, pruned_loss=0.09085, over 5621534.40 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3668, pruned_loss=0.1211, over 5647137.78 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:49:53,167 INFO [train.py:968] (1/2) Epoch 23, batch 8650, giga_loss[loss=0.3463, simple_loss=0.3814, pruned_loss=0.1557, over 23704.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3682, pruned_loss=0.1197, over 5654193.39 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3448, pruned_loss=0.09079, over 5624406.84 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.37, pruned_loss=0.1224, over 5653018.82 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:50:28,660 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-11 19:50:32,398 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 8700, giga_loss[loss=0.2887, simple_loss=0.3633, pruned_loss=0.1071, over 28736.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3709, pruned_loss=0.1184, over 5654493.30 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3451, pruned_loss=0.09091, over 5623034.90 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3723, pruned_loss=0.1207, over 5654949.52 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:51:07,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:51:27,072 INFO [train.py:968] (1/2) Epoch 23, batch 8750, giga_loss[loss=0.3066, simple_loss=0.381, pruned_loss=0.1161, over 29068.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3733, pruned_loss=0.1188, over 5671064.86 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3451, pruned_loss=0.09091, over 5629865.13 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.375, pruned_loss=0.1212, over 5666528.64 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:51:44,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8709, 3.7194, 3.5538, 1.8666], device='cuda:1'), covar=tensor([0.0671, 0.0796, 0.0772, 0.2324], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.1157, 0.0986, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 19:52:01,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:02,193 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 8800, giga_loss[loss=0.4536, simple_loss=0.4557, pruned_loss=0.2257, over 23465.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3758, pruned_loss=0.1215, over 5658846.20 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3456, pruned_loss=0.0914, over 5626156.07 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3773, pruned_loss=0.1236, over 5659268.44 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:52:19,929 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:32,745 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:34,410 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:52:37,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.26 vs. limit=5.0
+2023-03-11 19:52:59,056 INFO [train.py:968] (1/2) Epoch 23, batch 8850, giga_loss[loss=0.2921, simple_loss=0.3662, pruned_loss=0.1089, over 28964.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3762, pruned_loss=0.1222, over 5663753.46 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.345, pruned_loss=0.09097, over 5635789.36 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3788, pruned_loss=0.1253, over 5656364.14 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:53:21,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:53:23,817 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:53:27,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-11 19:53:32,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2032, 1.5316, 3.0401, 3.0075], device='cuda:1'), covar=tensor([0.1578, 0.2340, 0.0936, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0655, 0.0972, 0.0925], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 19:53:39,635 INFO [optim.py:369] (1/2) 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 19:53:48,463 INFO [train.py:968] (1/2) Epoch 23, batch 8900, libri_loss[loss=0.2547, simple_loss=0.3296, pruned_loss=0.08992, over 29540.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3753, pruned_loss=0.1228, over 5660208.96 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3444, pruned_loss=0.09078, over 5642344.46 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3788, pruned_loss=0.1262, over 5648706.25 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:53:51,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:54:18,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8516, 4.6983, 4.4871, 2.0994], device='cuda:1'), covar=tensor([0.0605, 0.0723, 0.0818, 0.1993], device='cuda:1'), in_proj_covar=tensor([0.1256, 0.1161, 0.0989, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 19:54:18,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6311, 1.8932, 1.5313, 1.7405], device='cuda:1'), covar=tensor([0.2858, 0.2883, 0.3311, 0.2534], device='cuda:1'), in_proj_covar=tensor([0.1524, 0.1100, 0.1347, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 19:54:21,863 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:54:26,276 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 8950, giga_loss[loss=0.2925, simple_loss=0.3531, pruned_loss=0.1159, over 28902.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3743, pruned_loss=0.1232, over 5649233.25 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3446, pruned_loss=0.09082, over 5647859.76 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3775, pruned_loss=0.1264, over 5635407.64 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:54:48,683 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-11 19:54:52,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:55:19,828 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 9000, giga_loss[loss=0.3365, simple_loss=0.3933, pruned_loss=0.1399, over 28877.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.372, pruned_loss=0.1216, over 5664048.98 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3448, pruned_loss=0.09076, over 5655504.51 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3751, pruned_loss=0.1251, over 5646484.45 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:55:30,040 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 19:55:38,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3405, 1.2330, 1.1633, 1.5143], device='cuda:1'), covar=tensor([0.0814, 0.0356, 0.0359, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 19:55:39,578 INFO [train.py:1012] (1/2) Epoch 23, validation: loss=0.2075, simple_loss=0.3151, pruned_loss=0.04992, over 944034.00 frames. 
+2023-03-11 19:55:39,578 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 19:56:26,471 INFO [train.py:968] (1/2) Epoch 23, batch 9050, giga_loss[loss=0.3022, simple_loss=0.369, pruned_loss=0.1177, over 28612.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3723, pruned_loss=0.1225, over 5669319.11 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3449, pruned_loss=0.09075, over 5659849.75 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3753, pruned_loss=0.126, over 5651729.42 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:57:05,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1012480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:57:06,335 INFO [optim.py:369] (1/2) 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 19:57:15,235 INFO [train.py:968] (1/2) Epoch 23, batch 9100, giga_loss[loss=0.2813, simple_loss=0.3496, pruned_loss=0.1065, over 29039.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3711, pruned_loss=0.122, over 5647699.55 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3447, pruned_loss=0.09059, over 5655797.34 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3747, pruned_loss=0.1261, over 5637325.07 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:58:03,536 INFO [train.py:968] (1/2) Epoch 23, batch 9150, giga_loss[loss=0.2897, simple_loss=0.3612, pruned_loss=0.109, over 28699.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3704, pruned_loss=0.1221, over 5658698.92 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3446, pruned_loss=0.09054, over 5659901.06 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3738, pruned_loss=0.126, over 5646816.84 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 19:58:35,573 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:58:47,773 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:58:50,369 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 19:58:52,785 INFO [train.py:968] (1/2) Epoch 23, batch 9200, giga_loss[loss=0.2928, simple_loss=0.3547, pruned_loss=0.1155, over 28933.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3679, pruned_loss=0.1208, over 5653928.96 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3447, pruned_loss=0.09053, over 5654061.70 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3708, pruned_loss=0.1243, over 5650015.95 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 19:59:08,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7449, 3.5753, 3.3633, 1.7446], device='cuda:1'), covar=tensor([0.0784, 0.0880, 0.0859, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1256, 0.1160, 0.0987, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 19:59:10,708 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-11 19:59:40,090 INFO [train.py:968] (1/2) Epoch 23, batch 9250, giga_loss[loss=0.2913, simple_loss=0.3626, pruned_loss=0.11, over 28922.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3672, pruned_loss=0.1194, over 5654095.37 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3447, pruned_loss=0.09053, over 5656615.87 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1224, over 5648973.79 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:00:20,098 INFO [optim.py:369] (1/2) 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 20:00:27,961 INFO [train.py:968] (1/2) Epoch 23, batch 9300, giga_loss[loss=0.3303, simple_loss=0.3937, pruned_loss=0.1334, over 28259.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3687, pruned_loss=0.1191, over 5667259.82 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.345, pruned_loss=0.09065, over 5669524.50 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3717, pruned_loss=0.1229, over 5651231.39 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:00:52,702 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:00:54,878 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:00:55,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0183, 5.8725, 5.5712, 3.1779], device='cuda:1'), covar=tensor([0.0468, 0.0649, 0.0739, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.1259, 0.1165, 0.0990, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 20:01:04,918 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:05,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:07,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1012732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:08,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 9350, giga_loss[loss=0.371, simple_loss=0.4157, pruned_loss=0.1632, over 27888.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3701, pruned_loss=0.1205, over 5650545.34 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.09069, over 5666009.24 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3731, pruned_loss=0.1243, over 5641453.70 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:01:24,559 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:37,026 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1012761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:01:38,172 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:02:07,014 INFO [train.py:968] (1/2) Epoch 23, batch 9400, giga_loss[loss=0.331, simple_loss=0.3715, pruned_loss=0.1452, over 23553.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3715, pruned_loss=0.1211, over 5654474.46 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3452, pruned_loss=0.09078, over 5667154.19 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.1241, over 5646384.31 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:02:57,134 INFO [train.py:968] (1/2) Epoch 23, batch 9450, giga_loss[loss=0.2755, simple_loss=0.3662, pruned_loss=0.09235, over 28932.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3724, pruned_loss=0.1194, over 5656158.39 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3449, pruned_loss=0.09065, over 5660499.20 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3748, pruned_loss=0.1223, over 5654971.02 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:03:10,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1012855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:03:11,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9752, 1.3194, 1.1268, 0.1258], device='cuda:1'), covar=tensor([0.4664, 0.3622, 0.4809, 0.7808], device='cuda:1'), in_proj_covar=tensor([0.1748, 0.1653, 0.1593, 0.1430], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 20:03:35,463 INFO [optim.py:369] (1/2) 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 20:03:40,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7206, 1.7538, 1.9406, 1.4605], device='cuda:1'), covar=tensor([0.1888, 0.2492, 0.1507, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0709, 0.0954, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 20:03:41,753 INFO [train.py:968] (1/2) Epoch 23, batch 9500, giga_loss[loss=0.292, simple_loss=0.3681, pruned_loss=0.108, over 28783.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3744, pruned_loss=0.1188, over 5672707.85 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3448, pruned_loss=0.09057, over 5665584.97 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.377, pruned_loss=0.1217, over 5667215.43 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:04:27,518 INFO [train.py:968] (1/2) Epoch 23, batch 9550, giga_loss[loss=0.372, simple_loss=0.4102, pruned_loss=0.1669, over 28275.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3764, pruned_loss=0.1203, over 5667287.62 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3452, pruned_loss=0.09075, over 5665428.09 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3795, pruned_loss=0.1237, over 5662588.30 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:04:34,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2169, 1.4786, 1.6044, 1.3082], device='cuda:1'), covar=tensor([0.1992, 0.1751, 0.2204, 0.1986], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0750, 0.0716, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 20:04:50,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 20:05:09,053 INFO [optim.py:369] (1/2) 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 20:05:18,102 INFO [train.py:968] (1/2) Epoch 23, batch 9600, giga_loss[loss=0.3342, simple_loss=0.3871, pruned_loss=0.1406, over 28849.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3781, pruned_loss=0.1226, over 5665633.77 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3448, pruned_loss=0.09067, over 5659966.46 frames. ], giga_tot_loss[loss=0.3169, simple_loss=0.3816, pruned_loss=0.1261, over 5666698.55 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:05:26,350 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1012998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:05:28,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1013001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:05:58,752 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1013030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:06:08,308 INFO [train.py:968] (1/2) Epoch 23, batch 9650, giga_loss[loss=0.3642, simple_loss=0.4123, pruned_loss=0.1581, over 27969.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3798, pruned_loss=0.1252, over 5657153.63 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3448, pruned_loss=0.09062, over 5665322.80 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3834, pruned_loss=0.1287, over 5653176.88 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:06:46,786 INFO [optim.py:369] (1/2) 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 20:06:53,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 20:06:55,045 INFO [train.py:968] (1/2) Epoch 23, batch 9700, giga_loss[loss=0.2734, simple_loss=0.3526, pruned_loss=0.09704, over 28271.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3786, pruned_loss=0.1246, over 5662670.48 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3445, pruned_loss=0.09046, over 5668812.16 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3821, pruned_loss=0.128, over 5656361.77 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:07:14,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6977, 1.1879, 4.9790, 3.7175], device='cuda:1'), covar=tensor([0.1673, 0.3008, 0.0402, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0654, 0.0971, 0.0923], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 20:07:22,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3725, 1.6263, 1.5726, 1.3085], device='cuda:1'), covar=tensor([0.3496, 0.2958, 0.2436, 0.2911], device='cuda:1'), in_proj_covar=tensor([0.1983, 0.1931, 0.1867, 0.1990], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 20:07:31,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9061, 2.9083, 1.7920, 1.1811], device='cuda:1'), covar=tensor([0.7511, 0.3259, 0.3542, 0.5658], device='cuda:1'), in_proj_covar=tensor([0.1755, 0.1659, 0.1597, 0.1436], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 20:07:34,273 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013132.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:07:41,067 INFO [train.py:968] (1/2) Epoch 23, batch 9750, libri_loss[loss=0.2615, simple_loss=0.3509, pruned_loss=0.08602, over 29627.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.377, pruned_loss=0.1221, over 5665589.42 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3443, pruned_loss=0.09029, over 5663087.81 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3805, pruned_loss=0.1256, over 5665121.57 frames. ], batch size: 91, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:07:48,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3965, 2.0161, 1.4094, 0.7142], device='cuda:1'), covar=tensor([0.7106, 0.3588, 0.4083, 0.7112], device='cuda:1'), in_proj_covar=tensor([0.1759, 0.1662, 0.1600, 0.1440], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 20:08:10,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5899, 4.3987, 4.1513, 1.9743], device='cuda:1'), covar=tensor([0.0608, 0.0792, 0.0822, 0.2146], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.1167, 0.0991, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 20:08:14,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4716, 2.0151, 1.4915, 0.7461], device='cuda:1'), covar=tensor([0.5510, 0.3005, 0.3531, 0.6494], device='cuda:1'), in_proj_covar=tensor([0.1758, 0.1661, 0.1599, 0.1440], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 20:08:22,490 INFO [optim.py:369] (1/2) 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 20:08:28,923 INFO [train.py:968] (1/2) Epoch 23, batch 9800, giga_loss[loss=0.3746, simple_loss=0.4108, pruned_loss=0.1692, over 26703.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3769, pruned_loss=0.1206, over 5660849.02 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3444, pruned_loss=0.09043, over 5659121.18 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3799, pruned_loss=0.1236, over 5664438.36 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:09:15,993 INFO [train.py:968] (1/2) Epoch 23, batch 9850, giga_loss[loss=0.3182, simple_loss=0.384, pruned_loss=0.1262, over 28833.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3766, pruned_loss=0.1201, over 5665017.73 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3445, pruned_loss=0.09049, over 5662646.14 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3797, pruned_loss=0.1231, over 5664616.42 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:09:39,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3309, 1.3977, 1.0722, 1.4764], device='cuda:1'), covar=tensor([0.0795, 0.0354, 0.0373, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 20:09:58,607 INFO [optim.py:369] (1/2) 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 20:10:08,080 INFO [train.py:968] (1/2) Epoch 23, batch 9900, giga_loss[loss=0.2956, simple_loss=0.3647, pruned_loss=0.1132, over 28560.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3778, pruned_loss=0.1217, over 5660785.70 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3446, pruned_loss=0.09037, over 5667309.84 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3808, pruned_loss=0.1248, over 5656395.18 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:10:35,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-11 20:10:57,909 INFO [train.py:968] (1/2) Epoch 23, batch 9950, giga_loss[loss=0.2823, simple_loss=0.3542, pruned_loss=0.1052, over 28850.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3764, pruned_loss=0.1215, over 5657557.89 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3447, pruned_loss=0.09045, over 5668814.88 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3791, pruned_loss=0.1243, over 5652434.59 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:11:03,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-11 20:11:30,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-11 20:11:41,574 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 10000, giga_loss[loss=0.3121, simple_loss=0.3734, pruned_loss=0.1254, over 28959.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3753, pruned_loss=0.122, over 5658088.68 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09043, over 5673706.08 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.378, pruned_loss=0.1249, over 5649264.50 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:11:56,740 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-11 20:12:35,419 INFO [train.py:968] (1/2) Epoch 23, batch 10050, libri_loss[loss=0.2282, simple_loss=0.3096, pruned_loss=0.0734, over 28582.00 frames. ], tot_loss[loss=0.305, simple_loss=0.371, pruned_loss=0.1195, over 5666273.30 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3441, pruned_loss=0.09005, over 5678908.19 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3747, pruned_loss=0.1231, over 5654232.36 frames. ], batch size: 63, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:13:04,200 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-11 20:13:21,693 INFO [optim.py:369] (1/2) 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 20:13:29,758 INFO [train.py:968] (1/2) Epoch 23, batch 10100, giga_loss[loss=0.3469, simple_loss=0.3774, pruned_loss=0.1582, over 23544.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3704, pruned_loss=0.1202, over 5654627.39 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.09, over 5682769.35 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3738, pruned_loss=0.1238, over 5640883.34 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:13:44,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013507.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:14:09,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-11 20:14:13,634 INFO [train.py:968] (1/2) Epoch 23, batch 10150, giga_loss[loss=0.2721, simple_loss=0.3426, pruned_loss=0.1007, over 28425.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3689, pruned_loss=0.1191, over 5662407.55 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08995, over 5680797.31 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3725, pruned_loss=0.1232, over 5652439.69 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:14:36,132 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:14:47,184 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:14:48,459 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:15:02,509 INFO [train.py:968] (1/2) Epoch 23, batch 10200, libri_loss[loss=0.286, simple_loss=0.3673, pruned_loss=0.1024, over 29534.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3677, pruned_loss=0.1182, over 5651872.37 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3445, pruned_loss=0.09027, over 5675348.71 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3708, pruned_loss=0.1216, over 5648604.28 frames. ], batch size: 83, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:15:50,435 INFO [train.py:968] (1/2) Epoch 23, batch 10250, giga_loss[loss=0.2345, simple_loss=0.3162, pruned_loss=0.07641, over 28888.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3639, pruned_loss=0.1141, over 5650643.06 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3446, pruned_loss=0.0903, over 5676170.97 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3663, pruned_loss=0.1168, over 5647339.18 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:15:57,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:16:02,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1013650.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:16:04,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1013653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:16:06,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4127, 2.1088, 1.5595, 0.6080], device='cuda:1'), covar=tensor([0.6226, 0.3047, 0.4282, 0.7311], device='cuda:1'), in_proj_covar=tensor([0.1759, 0.1662, 0.1596, 0.1441], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 20:16:35,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1013682.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:16:37,430 INFO [optim.py:369] (1/2) 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 20:16:42,990 INFO [train.py:968] (1/2) Epoch 23, batch 10300, libri_loss[loss=0.2669, simple_loss=0.3514, pruned_loss=0.09126, over 25989.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3626, pruned_loss=0.1125, over 5654539.26 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3448, pruned_loss=0.09032, over 5673976.93 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3648, pruned_loss=0.1152, over 5653635.24 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:17:28,838 INFO [train.py:968] (1/2) Epoch 23, batch 10350, giga_loss[loss=0.2616, simple_loss=0.3314, pruned_loss=0.09593, over 29030.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3613, pruned_loss=0.1117, over 5667059.53 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3447, pruned_loss=0.09029, over 5681859.56 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3637, pruned_loss=0.1146, over 5658665.86 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:17:49,164 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:18:12,246 INFO [optim.py:369] (1/2) 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 20:18:20,698 INFO [train.py:968] (1/2) Epoch 23, batch 10400, giga_loss[loss=0.3491, simple_loss=0.3847, pruned_loss=0.1568, over 26665.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3583, pruned_loss=0.1108, over 5660311.65 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.08996, over 5677712.56 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.361, pruned_loss=0.1141, over 5656910.83 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:18:54,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8149, 2.0258, 1.7450, 1.8467], device='cuda:1'), covar=tensor([0.1904, 0.2400, 0.2391, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0750, 0.0716, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 20:19:09,318 INFO [train.py:968] (1/2) Epoch 23, batch 10450, giga_loss[loss=0.2879, simple_loss=0.3597, pruned_loss=0.108, over 28976.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1115, over 5666213.38 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08977, over 5681433.01 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3612, pruned_loss=0.1146, over 5659975.22 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:19:45,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2894, 1.2068, 3.9395, 3.2432], device='cuda:1'), covar=tensor([0.1713, 0.2904, 0.0434, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0657, 0.0975, 0.0927], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 20:19:49,768 INFO [optim.py:369] (1/2) 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 20:19:56,211 INFO [train.py:968] (1/2) Epoch 23, batch 10500, giga_loss[loss=0.2793, simple_loss=0.3509, pruned_loss=0.1039, over 29094.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3624, pruned_loss=0.1136, over 5666706.61 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3442, pruned_loss=0.08995, over 5686485.16 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3647, pruned_loss=0.1164, over 5656859.56 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:20:03,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2878, 1.6374, 1.2860, 1.0623], device='cuda:1'), covar=tensor([0.2571, 0.2592, 0.2906, 0.2370], device='cuda:1'), in_proj_covar=tensor([0.1528, 0.1102, 0.1353, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 20:20:10,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4997, 1.7383, 1.4095, 1.5474], device='cuda:1'), covar=tensor([0.2627, 0.2530, 0.2819, 0.2324], device='cuda:1'), in_proj_covar=tensor([0.1529, 0.1103, 0.1353, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 20:20:42,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3867, 1.4967, 1.4468, 1.3416], device='cuda:1'), covar=tensor([0.2093, 0.2078, 0.2213, 0.2071], device='cuda:1'), in_proj_covar=tensor([0.1997, 0.1945, 0.1876, 0.2005], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 20:20:44,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:20:45,898 INFO [train.py:968] (1/2) Epoch 23, batch 10550, libri_loss[loss=0.2887, simple_loss=0.3736, pruned_loss=0.1019, over 29660.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3645, pruned_loss=0.1148, over 5658533.09 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3448, pruned_loss=0.09015, over 5690719.16 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1178, over 5645396.36 frames. ], batch size: 88, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:20:53,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:20:56,435 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1013952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:20:56,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1013952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:21:27,627 INFO [optim.py:369] (1/2) 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 20:21:32,495 INFO [train.py:968] (1/2) Epoch 23, batch 10600, giga_loss[loss=0.2688, simple_loss=0.3443, pruned_loss=0.09671, over 28889.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3635, pruned_loss=0.1138, over 5664936.89 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3451, pruned_loss=0.09019, over 5696765.65 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3654, pruned_loss=0.117, over 5648060.78 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:22:03,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:22:05,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-11 20:22:22,089 INFO [train.py:968] (1/2) Epoch 23, batch 10650, giga_loss[loss=0.2631, simple_loss=0.3413, pruned_loss=0.09247, over 29113.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3639, pruned_loss=0.1151, over 5660062.27 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.345, pruned_loss=0.09011, over 5698326.61 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1179, over 5644944.63 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:22:50,257 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-11 20:22:52,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014071.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:23:04,291 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:07,560 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:07,870 INFO [optim.py:369] (1/2) 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 20:23:14,299 INFO [train.py:968] (1/2) Epoch 23, batch 10700, giga_loss[loss=0.398, simple_loss=0.4231, pruned_loss=0.1865, over 26729.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3669, pruned_loss=0.1174, over 5661823.83 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3453, pruned_loss=0.09018, over 5702581.05 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.1201, over 5645306.89 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:23:16,026 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:17,365 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:18,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:19,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:34,223 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:44,783 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:46,228 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:23:52,598 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-11 20:23:54,885 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 10750, giga_loss[loss=0.2681, simple_loss=0.3458, pruned_loss=0.09517, over 29109.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3663, pruned_loss=0.116, over 5673298.47 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3448, pruned_loss=0.09, over 5706888.63 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3687, pruned_loss=0.1191, over 5655152.67 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:24:21,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:24:25,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:24:41,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6091, 1.7248, 1.6816, 1.5007], device='cuda:1'), covar=tensor([0.2770, 0.2618, 0.2311, 0.2584], device='cuda:1'), in_proj_covar=tensor([0.1998, 0.1951, 0.1882, 0.2008], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 20:24:42,005 INFO [optim.py:369] (1/2) 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 20:24:47,953 INFO [train.py:968] (1/2) Epoch 23, batch 10800, giga_loss[loss=0.2596, simple_loss=0.3377, pruned_loss=0.09074, over 28857.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3694, pruned_loss=0.1185, over 5671057.74 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3452, pruned_loss=0.0903, over 5702966.40 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3712, pruned_loss=0.1211, over 5659884.85 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:24:51,944 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:25:38,916 INFO [train.py:968] (1/2) Epoch 23, batch 10850, giga_loss[loss=0.2685, simple_loss=0.3493, pruned_loss=0.09389, over 28948.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3709, pruned_loss=0.1203, over 5675358.16 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3453, pruned_loss=0.09036, over 5704151.05 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3727, pruned_loss=0.1229, over 5664654.09 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:26:17,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:26:20,051 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:26:30,025 INFO [train.py:968] (1/2) Epoch 23, batch 10900, giga_loss[loss=0.3032, simple_loss=0.3796, pruned_loss=0.1134, over 28549.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3732, pruned_loss=0.1204, over 5664483.48 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3457, pruned_loss=0.09053, over 5703975.65 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3748, pruned_loss=0.1229, over 5655258.29 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:26:48,294 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:27:06,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:27:19,389 INFO [train.py:968] (1/2) Epoch 23, batch 10950, giga_loss[loss=0.3637, simple_loss=0.3988, pruned_loss=0.1643, over 23550.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3728, pruned_loss=0.1202, over 5664908.82 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3459, pruned_loss=0.09063, over 5705455.76 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3746, pruned_loss=0.1227, over 5655193.43 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:27:47,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6305, 1.1580, 4.9953, 3.6025], device='cuda:1'), covar=tensor([0.1704, 0.2990, 0.0424, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0655, 0.0969, 0.0924], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 20:27:55,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:28:05,033 INFO [optim.py:369] (1/2) 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 20:28:08,921 INFO [train.py:968] (1/2) Epoch 23, batch 11000, giga_loss[loss=0.2639, simple_loss=0.3452, pruned_loss=0.0913, over 28956.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3712, pruned_loss=0.1197, over 5663968.91 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3459, pruned_loss=0.09078, over 5709956.24 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3733, pruned_loss=0.1224, over 5650777.24 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:29:05,721 INFO [train.py:968] (1/2) Epoch 23, batch 11050, giga_loss[loss=0.3293, simple_loss=0.3804, pruned_loss=0.1391, over 28727.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3712, pruned_loss=0.1206, over 5651912.63 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3459, pruned_loss=0.09077, over 5709782.26 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3733, pruned_loss=0.1232, over 5640686.77 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:29:15,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014446.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:29:38,002 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:29:40,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:29:50,570 INFO [optim.py:369] (1/2) 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 20:29:57,244 INFO [train.py:968] (1/2) Epoch 23, batch 11100, giga_loss[loss=0.3102, simple_loss=0.3746, pruned_loss=0.1228, over 28672.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3683, pruned_loss=0.119, over 5656788.58 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3457, pruned_loss=0.09067, over 5716493.83 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.371, pruned_loss=0.1222, over 5639721.55 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:30:05,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3627, 1.5107, 1.5997, 1.2220], device='cuda:1'), covar=tensor([0.1420, 0.2231, 0.1210, 0.1480], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0711, 0.0955, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 20:30:09,322 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:30:28,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:30:44,173 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:30:44,665 INFO [train.py:968] (1/2) Epoch 23, batch 11150, libri_loss[loss=0.2824, simple_loss=0.3605, pruned_loss=0.1021, over 29571.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3676, pruned_loss=0.119, over 5662814.60 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3458, pruned_loss=0.09076, over 5720094.67 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3701, pruned_loss=0.122, over 5644607.87 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:31:30,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5217, 1.6574, 1.6331, 1.5244], device='cuda:1'), covar=tensor([0.2002, 0.2468, 0.2445, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0759, 0.0724, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 20:31:31,083 INFO [optim.py:369] (1/2) 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 20:31:34,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:31:35,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4129, 1.7795, 1.3603, 1.5783], device='cuda:1'), covar=tensor([0.2545, 0.2465, 0.2863, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.1528, 0.1103, 0.1351, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 20:31:35,596 INFO [train.py:968] (1/2) Epoch 23, batch 11200, giga_loss[loss=0.3382, simple_loss=0.3912, pruned_loss=0.1426, over 28799.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3673, pruned_loss=0.1194, over 5665271.03 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3456, pruned_loss=0.09074, over 5721756.52 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3696, pruned_loss=0.122, over 5649015.33 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:31:38,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014592.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:32:11,673 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014621.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:32:27,053 INFO [train.py:968] (1/2) Epoch 23, batch 11250, giga_loss[loss=0.3101, simple_loss=0.3746, pruned_loss=0.1228, over 28254.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.368, pruned_loss=0.1202, over 5661270.91 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3457, pruned_loss=0.09071, over 5723421.71 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.37, pruned_loss=0.1226, over 5646333.65 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:32:54,940 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 20:33:16,608 INFO [optim.py:369] (1/2) 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 20:33:19,331 INFO [train.py:968] (1/2) Epoch 23, batch 11300, giga_loss[loss=0.3616, simple_loss=0.4139, pruned_loss=0.1547, over 28260.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3694, pruned_loss=0.1218, over 5658967.41 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3451, pruned_loss=0.0905, over 5726725.75 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.372, pruned_loss=0.1245, over 5642701.73 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:33:35,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2651, 3.1093, 2.9635, 1.4100], device='cuda:1'), covar=tensor([0.1008, 0.1094, 0.0982, 0.2339], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.1172, 0.0993, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 20:33:57,472 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:34:08,172 INFO [train.py:968] (1/2) Epoch 23, batch 11350, giga_loss[loss=0.3141, simple_loss=0.3776, pruned_loss=0.1253, over 28942.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3704, pruned_loss=0.1223, over 5650777.99 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.345, pruned_loss=0.0903, over 5722913.03 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3733, pruned_loss=0.1256, over 5638944.25 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:34:16,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:34:32,666 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1014766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:34:54,999 INFO [optim.py:369] (1/2) 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 20:34:57,233 INFO [train.py:968] (1/2) Epoch 23, batch 11400, giga_loss[loss=0.2884, simple_loss=0.3557, pruned_loss=0.1105, over 28893.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3708, pruned_loss=0.1231, over 5643069.41 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3447, pruned_loss=0.09023, over 5716436.05 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3739, pruned_loss=0.1264, over 5637724.25 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:35:47,511 INFO [train.py:968] (1/2) Epoch 23, batch 11450, giga_loss[loss=0.3515, simple_loss=0.39, pruned_loss=0.1565, over 23596.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3705, pruned_loss=0.123, over 5649049.43 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3448, pruned_loss=0.0903, over 5715866.29 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3737, pruned_loss=0.1266, over 5643092.41 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:36:22,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2439, 4.0607, 3.8778, 1.8987], device='cuda:1'), covar=tensor([0.0654, 0.0790, 0.0791, 0.1934], device='cuda:1'), in_proj_covar=tensor([0.1264, 0.1169, 0.0989, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 20:36:34,089 INFO [optim.py:369] (1/2) 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 20:36:35,366 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-11 20:36:36,239 INFO [train.py:968] (1/2) Epoch 23, batch 11500, giga_loss[loss=0.2826, simple_loss=0.3571, pruned_loss=0.104, over 28399.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3714, pruned_loss=0.1235, over 5638668.19 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09034, over 5709974.46 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3746, pruned_loss=0.1272, over 5637527.29 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:36:38,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:36:38,744 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1014894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:36:42,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1014897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:36:44,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4245, 2.4592, 2.3011, 2.1171], device='cuda:1'), covar=tensor([0.1929, 0.2346, 0.2227, 0.2305], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0758, 0.0723, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 20:37:01,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1014915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:37:10,320 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1014926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:37:24,670 INFO [train.py:968] (1/2) Epoch 23, batch 11550, libri_loss[loss=0.2656, simple_loss=0.3519, pruned_loss=0.08967, over 29530.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3717, pruned_loss=0.1229, over 5657479.94 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3449, pruned_loss=0.09029, over 5713852.20 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1265, over 5651800.01 frames. ], batch size: 84, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 20:37:25,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2883, 3.8499, 1.5544, 1.5019], device='cuda:1'), covar=tensor([0.1017, 0.0366, 0.0897, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0561, 0.0391, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 20:38:05,119 INFO [optim.py:369] (1/2) 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 20:38:07,020 INFO [train.py:968] (1/2) Epoch 23, batch 11600, giga_loss[loss=0.3842, simple_loss=0.4232, pruned_loss=0.1726, over 28636.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1227, over 5658895.99 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3454, pruned_loss=0.09057, over 5709759.85 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3754, pruned_loss=0.1265, over 5657072.53 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:38:59,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:01,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:02,078 INFO [train.py:968] (1/2) Epoch 23, batch 11650, giga_loss[loss=0.3097, simple_loss=0.3762, pruned_loss=0.1216, over 28772.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3748, pruned_loss=0.1251, over 5649153.93 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3456, pruned_loss=0.09059, over 5713705.38 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3775, pruned_loss=0.1287, over 5642871.18 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:39:22,040 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:23,920 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:23,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:32,355 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:39:33,632 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:39:53,014 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 11700, giga_loss[loss=0.3127, simple_loss=0.3709, pruned_loss=0.1273, over 28680.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3757, pruned_loss=0.1266, over 5651414.61 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3456, pruned_loss=0.09055, over 5714691.85 frames. ], giga_tot_loss[loss=0.3186, simple_loss=0.378, pruned_loss=0.1297, over 5645227.47 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:40:09,280 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:40:10,500 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:40:11,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-11 20:40:40,731 INFO [train.py:968] (1/2) Epoch 23, batch 11750, giga_loss[loss=0.3725, simple_loss=0.4123, pruned_loss=0.1663, over 26586.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3757, pruned_loss=0.1256, over 5646402.65 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3457, pruned_loss=0.09077, over 5708855.13 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.378, pruned_loss=0.1285, over 5645393.21 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:40:40,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:41:26,372 INFO [optim.py:369] (1/2) 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 20:41:29,673 INFO [train.py:968] (1/2) Epoch 23, batch 11800, giga_loss[loss=0.2941, simple_loss=0.3642, pruned_loss=0.112, over 28660.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3757, pruned_loss=0.1245, over 5643764.75 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3459, pruned_loss=0.09084, over 5708171.22 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3778, pruned_loss=0.1273, over 5642370.59 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:41:51,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3160, 1.6778, 1.4794, 1.3953], device='cuda:1'), covar=tensor([0.2422, 0.2308, 0.2497, 0.2412], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0756, 0.0720, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 20:42:16,530 INFO [train.py:968] (1/2) Epoch 23, batch 11850, giga_loss[loss=0.3357, simple_loss=0.4042, pruned_loss=0.1336, over 29152.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3756, pruned_loss=0.1241, over 5641871.17 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3463, pruned_loss=0.09119, over 5700844.66 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3777, pruned_loss=0.127, over 5646198.81 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:42:28,198 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:42:31,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:42:47,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6512, 1.9015, 1.4803, 1.5102], device='cuda:1'), covar=tensor([0.0989, 0.0493, 0.0973, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0451, 0.0522, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 20:42:57,902 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:00,495 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:03,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:04,100 INFO [optim.py:369] (1/2) 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 20:43:06,670 INFO [train.py:968] (1/2) Epoch 23, batch 11900, giga_loss[loss=0.2687, simple_loss=0.337, pruned_loss=0.1002, over 28823.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3728, pruned_loss=0.1223, over 5642518.67 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.346, pruned_loss=0.09099, over 5704630.99 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3753, pruned_loss=0.1253, over 5641563.16 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:43:29,658 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:43:53,600 INFO [train.py:968] (1/2) Epoch 23, batch 11950, giga_loss[loss=0.3253, simple_loss=0.3713, pruned_loss=0.1396, over 23632.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3718, pruned_loss=0.1215, over 5652415.33 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.345, pruned_loss=0.09041, over 5708425.80 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3757, pruned_loss=0.1256, over 5646231.78 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:44:37,394 INFO [optim.py:369] (1/2) 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 20:44:40,324 INFO [train.py:968] (1/2) Epoch 23, batch 12000, giga_loss[loss=0.3285, simple_loss=0.3901, pruned_loss=0.1334, over 28218.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3734, pruned_loss=0.1225, over 5647578.89 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3451, pruned_loss=0.09037, over 5709470.23 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.377, pruned_loss=0.1266, over 5640464.48 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:44:40,324 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 20:44:49,124 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 20:45:29,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:45:34,725 INFO [train.py:968] (1/2) Epoch 23, batch 12050, libri_loss[loss=0.2679, simple_loss=0.3515, pruned_loss=0.09213, over 28696.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3729, pruned_loss=0.1225, over 5665377.17 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3449, pruned_loss=0.09022, over 5713830.35 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.377, pruned_loss=0.127, over 5653816.90 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:45:38,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:45:41,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:46:00,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015467.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:46:11,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5588, 1.6878, 1.8114, 1.3590], device='cuda:1'), covar=tensor([0.1737, 0.2672, 0.1453, 0.1793], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0708, 0.0952, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 20:46:17,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:46:21,693 INFO [optim.py:369] (1/2) 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 20:46:23,764 INFO [train.py:968] (1/2) Epoch 23, batch 12100, giga_loss[loss=0.3426, simple_loss=0.3918, pruned_loss=0.1467, over 28577.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3723, pruned_loss=0.1228, over 5668281.23 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3447, pruned_loss=0.09018, over 5717661.81 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3764, pruned_loss=0.1272, over 5654565.64 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:47:13,383 INFO [train.py:968] (1/2) Epoch 23, batch 12150, giga_loss[loss=0.3554, simple_loss=0.4117, pruned_loss=0.1496, over 27871.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5664629.64 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3452, pruned_loss=0.09062, over 5711536.49 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3782, pruned_loss=0.1286, over 5658188.15 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:47:29,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:47:54,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:47:56,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:02,464 INFO [optim.py:369] (1/2) 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 20:48:02,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:03,799 INFO [train.py:968] (1/2) Epoch 23, batch 12200, giga_loss[loss=0.2882, simple_loss=0.3677, pruned_loss=0.1044, over 29010.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3767, pruned_loss=0.1261, over 5664779.77 frames. ], libri_tot_loss[loss=0.2634, simple_loss=0.3454, pruned_loss=0.09071, over 5713819.76 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3794, pruned_loss=0.1293, over 5657249.92 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:48:05,421 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:24,523 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:33,661 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:36,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:39,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:42,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:48:50,833 INFO [train.py:968] (1/2) Epoch 23, batch 12250, giga_loss[loss=0.3894, simple_loss=0.4299, pruned_loss=0.1745, over 29066.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3765, pruned_loss=0.1256, over 5665193.61 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.346, pruned_loss=0.09103, over 5707502.59 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3793, pruned_loss=0.1291, over 5662261.99 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:49:08,973 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:49:36,497 INFO [optim.py:369] (1/2) 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 20:49:39,112 INFO [train.py:968] (1/2) Epoch 23, batch 12300, giga_loss[loss=0.2959, simple_loss=0.3599, pruned_loss=0.1159, over 28554.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3755, pruned_loss=0.1243, over 5661303.36 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3466, pruned_loss=0.09141, over 5711657.29 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.378, pruned_loss=0.1278, over 5653899.64 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:49:43,959 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1015695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:50:18,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1588, 2.2814, 2.4024, 1.8835], device='cuda:1'), covar=tensor([0.1794, 0.2299, 0.1369, 0.1686], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0708, 0.0952, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 20:50:26,625 INFO [train.py:968] (1/2) Epoch 23, batch 12350, giga_loss[loss=0.3893, simple_loss=0.4165, pruned_loss=0.181, over 23687.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3753, pruned_loss=0.1233, over 5669338.60 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3466, pruned_loss=0.09139, over 5712843.85 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3776, pruned_loss=0.1263, over 5662001.67 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:50:54,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9485, 1.1671, 1.0965, 0.8680], device='cuda:1'), covar=tensor([0.2078, 0.2329, 0.1486, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1990, 0.1949, 0.1871, 0.2007], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 20:51:14,652 INFO [optim.py:369] (1/2) 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 20:51:16,544 INFO [train.py:968] (1/2) Epoch 23, batch 12400, giga_loss[loss=0.3083, simple_loss=0.3704, pruned_loss=0.1231, over 28230.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3751, pruned_loss=0.1229, over 5679117.33 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3467, pruned_loss=0.09144, over 5711813.91 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3772, pruned_loss=0.1258, over 5673890.14 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:51:42,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:51:45,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-11 20:52:03,435 INFO [train.py:968] (1/2) Epoch 23, batch 12450, giga_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1218, over 29008.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.122, over 5674650.12 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3469, pruned_loss=0.09154, over 5718193.56 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3756, pruned_loss=0.1251, over 5663826.25 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:52:06,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015842.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:52:09,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5752, 1.3499, 4.9277, 3.5917], device='cuda:1'), covar=tensor([0.1742, 0.2891, 0.0411, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0660, 0.0979, 0.0933], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 20:52:55,003 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 12500, giga_loss[loss=0.3279, simple_loss=0.3792, pruned_loss=0.1383, over 28683.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3718, pruned_loss=0.1215, over 5674194.06 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.347, pruned_loss=0.09154, over 5720675.08 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3739, pruned_loss=0.1244, over 5663014.33 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:53:05,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7492, 1.0171, 2.8918, 2.7079], device='cuda:1'), covar=tensor([0.1772, 0.2665, 0.0611, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0659, 0.0977, 0.0931], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 20:53:39,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:53:49,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-11 20:53:49,868 INFO [train.py:968] (1/2) Epoch 23, batch 12550, giga_loss[loss=0.2753, simple_loss=0.3457, pruned_loss=0.1025, over 28857.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3668, pruned_loss=0.1189, over 5681431.76 frames. ], libri_tot_loss[loss=0.2647, simple_loss=0.3466, pruned_loss=0.09135, over 5722141.56 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.369, pruned_loss=0.1216, over 5670905.98 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:54:10,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:54:13,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1015964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:54:32,923 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1015985.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:54:35,063 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:54:37,757 INFO [train.py:968] (1/2) Epoch 23, batch 12600, giga_loss[loss=0.2769, simple_loss=0.3464, pruned_loss=0.1037, over 28914.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3646, pruned_loss=0.118, over 5689066.30 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3465, pruned_loss=0.09133, over 5726354.24 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3668, pruned_loss=0.1207, over 5676529.58 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:54:40,008 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1015993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:54:46,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1015999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:55:04,748 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016017.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 20:55:28,080 INFO [train.py:968] (1/2) Epoch 23, batch 12650, giga_loss[loss=0.2957, simple_loss=0.363, pruned_loss=0.1142, over 28812.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3639, pruned_loss=0.1177, over 5688155.39 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3468, pruned_loss=0.09154, over 5719897.81 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3658, pruned_loss=0.1203, over 5682272.33 frames. ], batch size: 285, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:55:34,900 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:55:57,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4219, 1.2508, 1.1093, 1.5355], device='cuda:1'), covar=tensor([0.0736, 0.0374, 0.0364, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0119, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 20:55:59,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:56:03,164 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:56:05,569 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:56:20,962 INFO [optim.py:369] (1/2) 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 20:56:22,731 INFO [train.py:968] (1/2) Epoch 23, batch 12700, giga_loss[loss=0.2501, simple_loss=0.3328, pruned_loss=0.08371, over 29039.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5680801.40 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3468, pruned_loss=0.09154, over 5719897.81 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3649, pruned_loss=0.1182, over 5676222.54 frames. ], batch size: 136, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:56:39,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:01,713 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016128.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 20:57:08,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 20:57:13,648 INFO [train.py:968] (1/2) Epoch 23, batch 12750, giga_loss[loss=0.2879, simple_loss=0.3645, pruned_loss=0.1056, over 28580.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3627, pruned_loss=0.1134, over 5680033.31 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3468, pruned_loss=0.09164, over 5722103.24 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3641, pruned_loss=0.1152, over 5673825.87 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:57:14,670 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:17,292 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:27,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1643, 4.0111, 3.7816, 1.7732], device='cuda:1'), covar=tensor([0.0665, 0.0779, 0.0893, 0.2215], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1165, 0.0989, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 20:57:31,644 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:57:49,653 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:58:04,387 INFO [optim.py:369] (1/2) 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 20:58:04,965 INFO [train.py:968] (1/2) Epoch 23, batch 12800, giga_loss[loss=0.2729, simple_loss=0.3468, pruned_loss=0.09947, over 28350.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3601, pruned_loss=0.1106, over 5674322.87 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3464, pruned_loss=0.09156, over 5727373.10 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3621, pruned_loss=0.1128, over 5662980.30 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 20:58:29,324 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:58:31,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:58:48,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3584, 3.4791, 1.4809, 1.4863], device='cuda:1'), covar=tensor([0.1059, 0.0374, 0.1043, 0.1471], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0560, 0.0391, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 20:58:59,901 INFO [train.py:968] (1/2) Epoch 23, batch 12850, giga_loss[loss=0.2736, simple_loss=0.3449, pruned_loss=0.1012, over 27976.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3572, pruned_loss=0.1076, over 5669669.01 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3465, pruned_loss=0.09166, over 5730000.24 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3588, pruned_loss=0.1095, over 5657784.60 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 20:59:03,774 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 20:59:47,016 INFO [optim.py:369] (1/2) 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 20:59:47,028 INFO [train.py:968] (1/2) Epoch 23, batch 12900, giga_loss[loss=0.265, simple_loss=0.3566, pruned_loss=0.08667, over 28763.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3538, pruned_loss=0.1042, over 5676535.35 frames. ], libri_tot_loss[loss=0.2649, simple_loss=0.3462, pruned_loss=0.09176, over 5732707.97 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3557, pruned_loss=0.1062, over 5661927.73 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:00:34,341 INFO [train.py:968] (1/2) Epoch 23, batch 12950, giga_loss[loss=0.2733, simple_loss=0.3621, pruned_loss=0.09221, over 28924.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3515, pruned_loss=0.1007, over 5679631.60 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3456, pruned_loss=0.0916, over 5737787.47 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3538, pruned_loss=0.1028, over 5661525.28 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:01:30,962 INFO [optim.py:369] (1/2) 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 21:01:30,975 INFO [train.py:968] (1/2) Epoch 23, batch 13000, giga_loss[loss=0.3032, simple_loss=0.375, pruned_loss=0.1156, over 28735.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3514, pruned_loss=0.1007, over 5663875.67 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3453, pruned_loss=0.09154, over 5737127.50 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1025, over 5649296.81 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:01:51,854 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:02:02,059 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:02:20,527 INFO [train.py:968] (1/2) Epoch 23, batch 13050, giga_loss[loss=0.2622, simple_loss=0.3373, pruned_loss=0.09358, over 28510.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3501, pruned_loss=0.09971, over 5669839.66 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3447, pruned_loss=0.09135, over 5740530.39 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3524, pruned_loss=0.1015, over 5654069.78 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:02:52,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:02:54,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3586, 1.7154, 1.4657, 1.6294], device='cuda:1'), covar=tensor([0.0787, 0.0325, 0.0339, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 21:03:10,547 INFO [optim.py:369] (1/2) 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 21:03:10,560 INFO [train.py:968] (1/2) Epoch 23, batch 13100, giga_loss[loss=0.2526, simple_loss=0.3312, pruned_loss=0.08703, over 28907.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.09738, over 5674153.54 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3445, pruned_loss=0.0913, over 5744182.86 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3491, pruned_loss=0.09902, over 5656504.54 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:03:23,182 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:04:00,050 INFO [train.py:968] (1/2) Epoch 23, batch 13150, giga_loss[loss=0.2635, simple_loss=0.3419, pruned_loss=0.09254, over 28659.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3444, pruned_loss=0.09573, over 5673445.92 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3441, pruned_loss=0.09125, over 5745466.43 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3464, pruned_loss=0.09715, over 5656952.64 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:04:26,733 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:04:29,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:04:50,833 INFO [train.py:968] (1/2) Epoch 23, batch 13200, giga_loss[loss=0.2354, simple_loss=0.3183, pruned_loss=0.07619, over 28473.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3448, pruned_loss=0.09573, over 5677013.88 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3439, pruned_loss=0.09109, over 5746934.72 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3466, pruned_loss=0.09708, over 5661554.89 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:04:51,540 INFO [optim.py:369] (1/2) 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 21:04:56,715 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:05:35,746 INFO [train.py:968] (1/2) Epoch 23, batch 13250, giga_loss[loss=0.2373, simple_loss=0.3264, pruned_loss=0.07411, over 28822.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3434, pruned_loss=0.09454, over 5669417.65 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.343, pruned_loss=0.09079, over 5741145.12 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3457, pruned_loss=0.09607, over 5659141.76 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:05:40,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016646.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:05:43,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016649.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 21:05:47,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4596, 1.6117, 1.3292, 1.5979], device='cuda:1'), covar=tensor([0.0752, 0.0394, 0.0367, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 21:05:49,992 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 21:06:12,995 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:06:15,716 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016678.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:06:15,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:06:28,660 INFO [train.py:968] (1/2) Epoch 23, batch 13300, giga_loss[loss=0.2508, simple_loss=0.3287, pruned_loss=0.08648, over 28293.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3411, pruned_loss=0.09273, over 5667856.78 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3431, pruned_loss=0.09092, over 5740640.86 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3429, pruned_loss=0.09389, over 5658755.62 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:06:30,141 INFO [optim.py:369] (1/2) 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 21:06:42,879 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:07:20,180 INFO [train.py:968] (1/2) Epoch 23, batch 13350, giga_loss[loss=0.2702, simple_loss=0.3423, pruned_loss=0.09903, over 28746.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3381, pruned_loss=0.09081, over 5662043.04 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.343, pruned_loss=0.09099, over 5735556.27 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3395, pruned_loss=0.0917, over 5657703.95 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:08:15,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:08:19,953 INFO [train.py:968] (1/2) Epoch 23, batch 13400, giga_loss[loss=0.2374, simple_loss=0.3178, pruned_loss=0.0785, over 28877.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3366, pruned_loss=0.09111, over 5645498.52 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3423, pruned_loss=0.09069, over 5738831.59 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3382, pruned_loss=0.09207, over 5637914.33 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:08:23,536 INFO [optim.py:369] (1/2) 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 21:08:23,852 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:09:13,758 INFO [train.py:968] (1/2) Epoch 23, batch 13450, giga_loss[loss=0.3114, simple_loss=0.3659, pruned_loss=0.1285, over 26676.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3359, pruned_loss=0.09146, over 5650152.61 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.342, pruned_loss=0.09045, over 5741377.06 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3374, pruned_loss=0.09245, over 5640572.72 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:09:18,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1016846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:10:12,276 INFO [train.py:968] (1/2) Epoch 23, batch 13500, giga_loss[loss=0.2838, simple_loss=0.3668, pruned_loss=0.1004, over 28621.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3366, pruned_loss=0.09142, over 5643364.51 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3416, pruned_loss=0.09032, over 5745586.79 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.338, pruned_loss=0.09238, over 5629398.66 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:10:16,211 INFO [optim.py:369] (1/2) 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 21:10:51,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1016924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:10:58,562 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:11:02,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:11:10,130 INFO [train.py:968] (1/2) Epoch 23, batch 13550, giga_loss[loss=0.2875, simple_loss=0.3688, pruned_loss=0.1032, over 28302.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3388, pruned_loss=0.09139, over 5650984.91 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3409, pruned_loss=0.08995, over 5747572.32 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3405, pruned_loss=0.09254, over 5635841.64 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 1.0
+2023-03-11 21:11:39,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1016962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:12:08,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1016989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:12:09,741 INFO [train.py:968] (1/2) Epoch 23, batch 13600, giga_loss[loss=0.2949, simple_loss=0.3694, pruned_loss=0.1102, over 28508.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09161, over 5653216.70 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3399, pruned_loss=0.08952, over 5751440.82 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3412, pruned_loss=0.09303, over 5633749.18 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:12:12,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1016992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:12:13,819 INFO [optim.py:369] (1/2) 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 21:12:49,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:13:06,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:13:09,693 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-11 21:13:13,893 INFO [train.py:968] (1/2) Epoch 23, batch 13650, giga_loss[loss=0.2407, simple_loss=0.3194, pruned_loss=0.081, over 29006.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3378, pruned_loss=0.09106, over 5655332.94 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3398, pruned_loss=0.0895, over 5751858.86 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3397, pruned_loss=0.09224, over 5636939.58 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:14:18,111 INFO [train.py:968] (1/2) Epoch 23, batch 13700, giga_loss[loss=0.2419, simple_loss=0.3353, pruned_loss=0.07426, over 29176.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3361, pruned_loss=0.08923, over 5657525.40 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3397, pruned_loss=0.08955, over 5754644.48 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3376, pruned_loss=0.09012, over 5638722.35 frames. ], batch size: 113, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:14:21,161 INFO [optim.py:369] (1/2) 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 21:15:22,122 INFO [train.py:968] (1/2) Epoch 23, batch 13750, giga_loss[loss=0.2643, simple_loss=0.3458, pruned_loss=0.09139, over 28595.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3344, pruned_loss=0.08722, over 5656426.29 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3395, pruned_loss=0.08949, over 5755516.25 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3357, pruned_loss=0.08796, over 5640421.27 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:15:36,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0787, 1.3998, 1.1108, 0.3038], device='cuda:1'), covar=tensor([0.3636, 0.3374, 0.5223, 0.6872], device='cuda:1'), in_proj_covar=tensor([0.1745, 0.1648, 0.1589, 0.1427], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 21:15:57,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1017169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:16:25,085 INFO [train.py:968] (1/2) Epoch 23, batch 13800, giga_loss[loss=0.2651, simple_loss=0.3317, pruned_loss=0.09928, over 28864.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3319, pruned_loss=0.08676, over 5662804.50 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3396, pruned_loss=0.08976, over 5760341.04 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3327, pruned_loss=0.08701, over 5642480.38 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:16:29,905 INFO [optim.py:369] (1/2) 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 21:17:01,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5169, 1.7860, 1.4577, 1.3282], device='cuda:1'), covar=tensor([0.2900, 0.2777, 0.3325, 0.2467], device='cuda:1'), in_proj_covar=tensor([0.1528, 0.1099, 0.1355, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 21:17:27,604 INFO [train.py:968] (1/2) Epoch 23, batch 13850, giga_loss[loss=0.2345, simple_loss=0.3172, pruned_loss=0.07593, over 29040.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3318, pruned_loss=0.08722, over 5666708.97 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3399, pruned_loss=0.09004, over 5762727.68 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3319, pruned_loss=0.08711, over 5646216.30 frames. ], batch size: 285, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:18:10,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-11 21:18:31,029 INFO [train.py:968] (1/2) Epoch 23, batch 13900, giga_loss[loss=0.2197, simple_loss=0.3031, pruned_loss=0.06816, over 28861.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.332, pruned_loss=0.08748, over 5672502.74 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3399, pruned_loss=0.09012, over 5761916.39 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.332, pruned_loss=0.08726, over 5655309.71 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:18:35,069 INFO [optim.py:369] (1/2) 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 21:18:39,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3580, 1.9690, 1.5204, 1.4905], device='cuda:1'), covar=tensor([0.0764, 0.0298, 0.0331, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 21:18:40,739 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1017299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:18:54,942 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1017312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:19:00,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1017315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:19:31,041 INFO [train.py:968] (1/2) Epoch 23, batch 13950, giga_loss[loss=0.2918, simple_loss=0.3643, pruned_loss=0.1096, over 28997.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3341, pruned_loss=0.088, over 5668341.78 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3398, pruned_loss=0.0902, over 5753451.84 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.334, pruned_loss=0.0877, over 5659621.07 frames. ], batch size: 285, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:19:35,920 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:19:37,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6361, 2.2518, 1.5626, 0.7779], device='cuda:1'), covar=tensor([0.4659, 0.2708, 0.3540, 0.5466], device='cuda:1'), in_proj_covar=tensor([0.1745, 0.1648, 0.1588, 0.1427], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 21:19:40,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5394, 1.8734, 1.7812, 1.3498], device='cuda:1'), covar=tensor([0.1998, 0.2638, 0.1623, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0700, 0.0952, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 21:19:42,497 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6302, 1.8668, 1.7583, 1.5723], device='cuda:1'), covar=tensor([0.2371, 0.1931, 0.1910, 0.2124], device='cuda:1'), in_proj_covar=tensor([0.1940, 0.1889, 0.1810, 0.1945], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 21:20:31,382 INFO [train.py:968] (1/2) Epoch 23, batch 14000, giga_loss[loss=0.2193, simple_loss=0.2877, pruned_loss=0.0755, over 24764.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3352, pruned_loss=0.08815, over 5668353.84 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3391, pruned_loss=0.0899, over 5746721.09 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3357, pruned_loss=0.08814, over 5665414.58 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:20:36,387 INFO [optim.py:369] (1/2) 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 21:21:00,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1017410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:21:45,041 INFO [train.py:968] (1/2) Epoch 23, batch 14050, libri_loss[loss=0.1961, simple_loss=0.2753, pruned_loss=0.05849, over 29485.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3311, pruned_loss=0.08584, over 5676981.10 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3383, pruned_loss=0.08966, over 5749656.08 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3321, pruned_loss=0.08602, over 5670519.17 frames. ], batch size: 70, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:21:47,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1017442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:21:53,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1017445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:22:31,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:22:54,317 INFO [train.py:968] (1/2) Epoch 23, batch 14100, giga_loss[loss=0.2661, simple_loss=0.3538, pruned_loss=0.08922, over 28922.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3337, pruned_loss=0.08834, over 5668027.05 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.338, pruned_loss=0.08963, over 5751583.79 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3347, pruned_loss=0.08848, over 5660114.36 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:23:00,141 INFO [optim.py:369] (1/2) 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 21:24:02,539 INFO [train.py:968] (1/2) Epoch 23, batch 14150, giga_loss[loss=0.2622, simple_loss=0.3628, pruned_loss=0.08083, over 28846.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.337, pruned_loss=0.08817, over 5660431.56 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3375, pruned_loss=0.08942, over 5751677.85 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3382, pruned_loss=0.08844, over 5651444.03 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:24:16,997 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1017553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:24:20,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1017556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:24:57,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1017585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:24:57,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3800, 1.9486, 1.3527, 0.6281], device='cuda:1'), covar=tensor([0.6205, 0.3049, 0.4662, 0.6570], device='cuda:1'), in_proj_covar=tensor([0.1755, 0.1658, 0.1596, 0.1434], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 21:25:02,207 INFO [train.py:968] (1/2) Epoch 23, batch 14200, giga_loss[loss=0.214, simple_loss=0.2867, pruned_loss=0.0707, over 24453.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3392, pruned_loss=0.08774, over 5666567.24 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3372, pruned_loss=0.08933, over 5757236.00 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3406, pruned_loss=0.08798, over 5650589.59 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:25:05,824 INFO [optim.py:369] (1/2) 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 21:25:19,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7047, 4.5535, 4.3079, 1.8091], device='cuda:1'), covar=tensor([0.0633, 0.0791, 0.0997, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1133, 0.0959, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 21:25:23,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3669, 4.2140, 4.0269, 1.7240], device='cuda:1'), covar=tensor([0.0528, 0.0663, 0.0650, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.1224, 0.1133, 0.0959, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 21:25:52,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3108, 1.4189, 1.3702, 1.2841], device='cuda:1'), covar=tensor([0.2098, 0.1934, 0.1582, 0.1964], device='cuda:1'), in_proj_covar=tensor([0.1936, 0.1883, 0.1799, 0.1940], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 21:26:01,372 INFO [train.py:968] (1/2) Epoch 23, batch 14250, giga_loss[loss=0.2471, simple_loss=0.3349, pruned_loss=0.07967, over 28694.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3377, pruned_loss=0.08592, over 5657413.78 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3364, pruned_loss=0.08895, over 5760505.32 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3395, pruned_loss=0.08638, over 5638623.78 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:26:11,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:26:31,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-11 21:27:03,905 INFO [train.py:968] (1/2) Epoch 23, batch 14300, giga_loss[loss=0.25, simple_loss=0.3318, pruned_loss=0.08412, over 28809.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3381, pruned_loss=0.08529, over 5669951.00 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3362, pruned_loss=0.0888, over 5761784.95 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3397, pruned_loss=0.08572, over 5652146.77 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:27:09,061 INFO [optim.py:369] (1/2) 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 21:27:15,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6300, 1.8055, 1.2817, 1.4544], device='cuda:1'), covar=tensor([0.0923, 0.0527, 0.1024, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0393, 0.0443, 0.0517, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 21:27:27,874 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017708.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:27:43,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-11 21:28:06,387 INFO [train.py:968] (1/2) Epoch 23, batch 14350, giga_loss[loss=0.2786, simple_loss=0.3539, pruned_loss=0.1017, over 28681.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3392, pruned_loss=0.08703, over 5676621.66 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.336, pruned_loss=0.08876, over 5762118.33 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3408, pruned_loss=0.08736, over 5659858.85 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 21:29:07,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-11 21:29:12,194 INFO [train.py:968] (1/2) Epoch 23, batch 14400, giga_loss[loss=0.2448, simple_loss=0.3288, pruned_loss=0.0804, over 28857.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3381, pruned_loss=0.08783, over 5674368.41 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3355, pruned_loss=0.08862, over 5764905.80 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3398, pruned_loss=0.08817, over 5656172.99 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:29:18,544 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 14450, giga_loss[loss=0.2247, simple_loss=0.3124, pruned_loss=0.06847, over 28850.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3384, pruned_loss=0.08828, over 5677015.70 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3352, pruned_loss=0.08849, over 5765494.27 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.34, pruned_loss=0.08865, over 5661932.15 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:31:34,099 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:31:54,808 INFO [train.py:968] (1/2) Epoch 23, batch 14500, giga_loss[loss=0.2688, simple_loss=0.3444, pruned_loss=0.09657, over 28948.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.333, pruned_loss=0.08502, over 5679917.83 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3348, pruned_loss=0.08829, over 5768796.63 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3347, pruned_loss=0.08543, over 5662476.54 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:32:00,821 INFO [optim.py:369] (1/2) 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 21:33:01,552 INFO [train.py:968] (1/2) Epoch 23, batch 14550, giga_loss[loss=0.1974, simple_loss=0.2837, pruned_loss=0.05559, over 28959.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3304, pruned_loss=0.08349, over 5677234.64 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3343, pruned_loss=0.08804, over 5771524.85 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3321, pruned_loss=0.08393, over 5657652.90 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:33:16,971 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1017949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:34:03,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5391, 3.8110, 1.7090, 1.6469], device='cuda:1'), covar=tensor([0.0964, 0.0431, 0.0894, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0556, 0.0391, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 21:34:05,586 INFO [train.py:968] (1/2) Epoch 23, batch 14600, giga_loss[loss=0.2772, simple_loss=0.3594, pruned_loss=0.09746, over 28616.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3275, pruned_loss=0.08251, over 5685758.69 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3329, pruned_loss=0.08745, over 5773234.41 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.33, pruned_loss=0.08323, over 5665537.15 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:34:10,288 INFO [optim.py:369] (1/2) 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 21:34:18,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6479, 1.7461, 1.8468, 1.4063], device='cuda:1'), covar=tensor([0.1909, 0.2623, 0.1524, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0699, 0.0953, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 21:34:22,632 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:34:35,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3562, 1.7548, 1.4438, 1.5821], device='cuda:1'), covar=tensor([0.0719, 0.0397, 0.0336, 0.0764], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-11 21:34:42,993 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:35:06,168 INFO [train.py:968] (1/2) Epoch 23, batch 14650, giga_loss[loss=0.2844, simple_loss=0.3573, pruned_loss=0.1057, over 27695.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3336, pruned_loss=0.08562, over 5688046.81 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3332, pruned_loss=0.08759, over 5772444.01 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3353, pruned_loss=0.086, over 5671062.90 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:35:55,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 21:36:05,962 INFO [train.py:968] (1/2) Epoch 23, batch 14700, giga_loss[loss=0.2615, simple_loss=0.3323, pruned_loss=0.09532, over 27560.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.333, pruned_loss=0.08602, over 5688983.66 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3325, pruned_loss=0.08725, over 5776428.12 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3349, pruned_loss=0.08657, over 5668632.21 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:36:10,372 INFO [optim.py:369] (1/2) 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 21:36:56,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2675, 1.7103, 1.2568, 0.6354], device='cuda:1'), covar=tensor([0.3503, 0.1989, 0.2814, 0.5055], device='cuda:1'), in_proj_covar=tensor([0.1753, 0.1658, 0.1595, 0.1438], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 21:37:11,165 INFO [train.py:968] (1/2) Epoch 23, batch 14750, giga_loss[loss=0.2491, simple_loss=0.3267, pruned_loss=0.08572, over 28020.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08707, over 5684197.40 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3328, pruned_loss=0.08744, over 5775510.70 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3344, pruned_loss=0.08736, over 5667150.78 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:37:16,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6496, 4.8873, 1.8232, 1.9874], device='cuda:1'), covar=tensor([0.0999, 0.0337, 0.0936, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0555, 0.0392, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 21:37:44,575 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:37:47,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:38:13,594 INFO [train.py:968] (1/2) Epoch 23, batch 14800, giga_loss[loss=0.2278, simple_loss=0.2924, pruned_loss=0.08155, over 24361.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3327, pruned_loss=0.08726, over 5680649.70 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3322, pruned_loss=0.08718, over 5776867.50 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3344, pruned_loss=0.08774, over 5663284.52 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:38:17,997 INFO [optim.py:369] (1/2) 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 21:38:22,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:39:05,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:39:08,578 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018229.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:39:22,614 INFO [train.py:968] (1/2) Epoch 23, batch 14850, giga_loss[loss=0.2694, simple_loss=0.3542, pruned_loss=0.09228, over 28907.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3348, pruned_loss=0.08749, over 5681364.16 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3321, pruned_loss=0.08707, over 5777528.16 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3362, pruned_loss=0.08796, over 5666862.76 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:39:38,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:39:48,661 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018258.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 21:40:06,120 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:40:10,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5311, 1.9134, 1.4625, 1.5954], device='cuda:1'), covar=tensor([0.2702, 0.2619, 0.3165, 0.2468], device='cuda:1'), in_proj_covar=tensor([0.1526, 0.1097, 0.1351, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 21:40:38,721 INFO [train.py:968] (1/2) Epoch 23, batch 14900, giga_loss[loss=0.2669, simple_loss=0.3473, pruned_loss=0.09325, over 28729.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3362, pruned_loss=0.08779, over 5676651.56 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.332, pruned_loss=0.08708, over 5776354.81 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3374, pruned_loss=0.08816, over 5664343.01 frames. ], batch size: 243, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:40:47,025 INFO [optim.py:369] (1/2) 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 21:41:29,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:41:53,399 INFO [train.py:968] (1/2) Epoch 23, batch 14950, giga_loss[loss=0.2321, simple_loss=0.3102, pruned_loss=0.07702, over 29050.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3342, pruned_loss=0.08698, over 5681598.18 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3318, pruned_loss=0.08718, over 5779466.22 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3355, pruned_loss=0.08721, over 5665857.24 frames. ], batch size: 285, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:42:41,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:42:56,305 INFO [train.py:968] (1/2) Epoch 23, batch 15000, giga_loss[loss=0.21, simple_loss=0.2925, pruned_loss=0.0638, over 29082.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3301, pruned_loss=0.08558, over 5697177.74 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3315, pruned_loss=0.08713, over 5784077.89 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3315, pruned_loss=0.08577, over 5677323.35 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:42:56,305 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 21:43:05,334 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 21:43:08,022 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:43:09,662 INFO [optim.py:369] (1/2) 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 21:43:09,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:43:52,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:44:11,537 INFO [train.py:968] (1/2) Epoch 23, batch 15050, giga_loss[loss=0.2621, simple_loss=0.3409, pruned_loss=0.09168, over 28401.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3258, pruned_loss=0.08389, over 5693217.88 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3316, pruned_loss=0.08716, over 5785625.41 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3267, pruned_loss=0.08397, over 5674529.13 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:44:31,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2927, 4.1297, 3.9348, 1.9821], device='cuda:1'), covar=tensor([0.0561, 0.0735, 0.0805, 0.2117], device='cuda:1'), in_proj_covar=tensor([0.1226, 0.1131, 0.0959, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 21:44:44,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:44:48,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:45:10,969 INFO [train.py:968] (1/2) Epoch 23, batch 15100, giga_loss[loss=0.252, simple_loss=0.3271, pruned_loss=0.08845, over 28936.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3268, pruned_loss=0.08498, over 5682889.32 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3314, pruned_loss=0.08705, over 5779380.12 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08508, over 5671227.70 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:45:12,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3560, 2.1246, 1.6110, 0.5324], device='cuda:1'), covar=tensor([0.5376, 0.2974, 0.4180, 0.6345], device='cuda:1'), in_proj_covar=tensor([0.1762, 0.1662, 0.1601, 0.1446], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 21:45:16,261 INFO [optim.py:369] (1/2) 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 21:45:20,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:45:45,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:45:48,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:46:06,531 INFO [train.py:968] (1/2) Epoch 23, batch 15150, giga_loss[loss=0.2372, simple_loss=0.3197, pruned_loss=0.0773, over 28572.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3271, pruned_loss=0.08554, over 5681349.61 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3306, pruned_loss=0.08681, over 5782424.72 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3284, pruned_loss=0.08582, over 5667259.63 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:46:22,863 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:47:01,977 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4395, 3.3249, 1.5661, 1.6788], device='cuda:1'), covar=tensor([0.0940, 0.0284, 0.0908, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0553, 0.0390, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 21:47:10,733 INFO [train.py:968] (1/2) Epoch 23, batch 15200, giga_loss[loss=0.2745, simple_loss=0.3531, pruned_loss=0.09795, over 28505.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3244, pruned_loss=0.08343, over 5675024.73 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3299, pruned_loss=0.08643, over 5785599.72 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3259, pruned_loss=0.08393, over 5658758.51 frames. ], batch size: 370, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:47:17,100 INFO [optim.py:369] (1/2) 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 21:47:35,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:04,513 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:12,008 INFO [train.py:968] (1/2) Epoch 23, batch 15250, giga_loss[loss=0.2152, simple_loss=0.3003, pruned_loss=0.06503, over 28052.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3236, pruned_loss=0.08226, over 5665381.69 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.33, pruned_loss=0.08654, over 5774917.49 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3247, pruned_loss=0.08249, over 5659981.26 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:48:17,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:17,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:48:21,879 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:49:10,953 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1018680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:49:25,507 INFO [train.py:968] (1/2) Epoch 23, batch 15300, giga_loss[loss=0.2774, simple_loss=0.3574, pruned_loss=0.09867, over 28169.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.323, pruned_loss=0.08224, over 5668006.05 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3301, pruned_loss=0.08667, over 5776727.91 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3235, pruned_loss=0.08223, over 5659997.42 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:49:34,555 INFO [optim.py:369] (1/2) 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 21:49:44,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4957, 1.6957, 1.7183, 1.3192], device='cuda:1'), covar=tensor([0.1843, 0.2666, 0.1590, 0.1861], device='cuda:1'), in_proj_covar=tensor([0.0903, 0.0695, 0.0949, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 21:50:31,234 INFO [train.py:968] (1/2) Epoch 23, batch 15350, giga_loss[loss=0.2463, simple_loss=0.3329, pruned_loss=0.07985, over 28936.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3242, pruned_loss=0.08227, over 5683188.39 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3302, pruned_loss=0.08668, over 5777618.11 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3245, pruned_loss=0.08217, over 5674207.41 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:50:44,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2426, 1.8639, 1.4267, 0.4758], device='cuda:1'), covar=tensor([0.4507, 0.3168, 0.4398, 0.6215], device='cuda:1'), in_proj_covar=tensor([0.1754, 0.1658, 0.1593, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 21:50:50,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-11 21:51:33,312 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1018787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:51:35,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1018790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:51:35,852 INFO [train.py:968] (1/2) Epoch 23, batch 15400, giga_loss[loss=0.299, simple_loss=0.3693, pruned_loss=0.1143, over 28672.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3257, pruned_loss=0.08341, over 5684548.24 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3302, pruned_loss=0.08676, over 5770178.52 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3258, pruned_loss=0.08319, over 5681780.58 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:51:43,540 INFO [optim.py:369] (1/2) 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 21:52:14,911 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1018819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:52:44,305 INFO [train.py:968] (1/2) Epoch 23, batch 15450, giga_loss[loss=0.2279, simple_loss=0.312, pruned_loss=0.07192, over 28923.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3267, pruned_loss=0.08458, over 5685023.74 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3304, pruned_loss=0.08687, over 5771060.99 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3266, pruned_loss=0.08425, over 5680521.29 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:52:56,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3338, 1.6447, 1.6614, 1.2158], device='cuda:1'), covar=tensor([0.1761, 0.2630, 0.1473, 0.1862], device='cuda:1'), in_proj_covar=tensor([0.0902, 0.0695, 0.0947, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 21:53:16,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9051, 1.4423, 1.2704, 1.1689], device='cuda:1'), covar=tensor([0.2339, 0.1748, 0.2356, 0.2060], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0732, 0.0701, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 21:53:23,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4474, 1.5311, 1.2278, 1.0961], device='cuda:1'), covar=tensor([0.0934, 0.0455, 0.0944, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0445, 0.0520, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 21:53:44,937 INFO [train.py:968] (1/2) Epoch 23, batch 15500, giga_loss[loss=0.2243, simple_loss=0.3146, pruned_loss=0.06693, over 28644.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3261, pruned_loss=0.08391, over 5678099.47 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3304, pruned_loss=0.08686, over 5770641.00 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3259, pruned_loss=0.08361, over 5673000.21 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:53:52,961 INFO [optim.py:369] (1/2) 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 21:54:22,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1054, 1.4782, 1.3594, 1.0362], device='cuda:1'), covar=tensor([0.1508, 0.2225, 0.1299, 0.1633], device='cuda:1'), in_proj_covar=tensor([0.0904, 0.0696, 0.0950, 0.0850], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 21:54:50,811 INFO [train.py:968] (1/2) Epoch 23, batch 15550, giga_loss[loss=0.2645, simple_loss=0.3453, pruned_loss=0.09189, over 27656.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3279, pruned_loss=0.08362, over 5652664.79 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3307, pruned_loss=0.08715, over 5762354.18 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3275, pruned_loss=0.08311, over 5655056.88 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:55:21,007 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-11 21:55:47,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1018986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:55:52,171 INFO [train.py:968] (1/2) Epoch 23, batch 15600, giga_loss[loss=0.266, simple_loss=0.3567, pruned_loss=0.08769, over 29048.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3304, pruned_loss=0.08469, over 5662971.45 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.33, pruned_loss=0.08683, over 5766621.14 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3306, pruned_loss=0.0845, over 5658047.69 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 21:55:58,822 INFO [optim.py:369] (1/2) 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 21:56:13,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:56:28,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:56:34,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:56:53,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4908, 1.8149, 1.4381, 1.6628], device='cuda:1'), covar=tensor([0.0791, 0.0299, 0.0335, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-11 21:56:53,538 INFO [train.py:968] (1/2) Epoch 23, batch 15650, giga_loss[loss=0.2194, simple_loss=0.3088, pruned_loss=0.06506, over 28915.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3308, pruned_loss=0.08458, over 5659536.42 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.33, pruned_loss=0.08684, over 5767262.87 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3311, pruned_loss=0.08439, over 5653696.85 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:57:12,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:57:58,310 INFO [train.py:968] (1/2) Epoch 23, batch 15700, giga_loss[loss=0.2492, simple_loss=0.3355, pruned_loss=0.08145, over 28483.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3305, pruned_loss=0.08501, over 5650883.61 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3297, pruned_loss=0.08683, over 5767361.28 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3309, pruned_loss=0.08481, over 5643184.49 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:58:04,092 INFO [optim.py:369] (1/2) 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 21:58:24,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2728, 2.3889, 2.4714, 2.1171], device='cuda:1'), covar=tensor([0.1945, 0.2440, 0.2008, 0.2279], device='cuda:1'), in_proj_covar=tensor([0.0464, 0.0731, 0.0698, 0.0670], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 21:58:45,888 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:58:49,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:58:57,918 INFO [train.py:968] (1/2) Epoch 23, batch 15750, giga_loss[loss=0.2832, simple_loss=0.361, pruned_loss=0.1026, over 28899.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3275, pruned_loss=0.08326, over 5657771.75 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3298, pruned_loss=0.08693, over 5769284.42 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3279, pruned_loss=0.08297, over 5647349.61 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 21:59:16,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:21,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:28,850 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:31,619 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:31,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:35,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:35,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:41,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 21:59:57,273 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 15800, giga_loss[loss=0.271, simple_loss=0.3315, pruned_loss=0.1052, over 26719.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3264, pruned_loss=0.08299, over 5663421.14 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.33, pruned_loss=0.08711, over 5769795.00 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3265, pruned_loss=0.0826, over 5654401.63 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:00:13,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:15,958 INFO [optim.py:369] (1/2) 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 22:00:16,474 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:16,521 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:18,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-11 22:00:19,817 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:28,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:00:54,267 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:01:06,649 INFO [train.py:968] (1/2) Epoch 23, batch 15850, giga_loss[loss=0.2661, simple_loss=0.3458, pruned_loss=0.09324, over 28976.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3258, pruned_loss=0.08278, over 5671633.41 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3299, pruned_loss=0.08701, over 5773080.03 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3258, pruned_loss=0.08242, over 5658248.29 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:02:13,021 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:02:13,482 INFO [train.py:968] (1/2) Epoch 23, batch 15900, libri_loss[loss=0.2432, simple_loss=0.3182, pruned_loss=0.08408, over 29616.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3277, pruned_loss=0.08328, over 5677546.01 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3293, pruned_loss=0.08672, over 5774754.55 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3282, pruned_loss=0.08319, over 5662818.26 frames. ], batch size: 74, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:02:21,398 INFO [optim.py:369] (1/2) 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 22:03:20,416 INFO [train.py:968] (1/2) Epoch 23, batch 15950, giga_loss[loss=0.2959, simple_loss=0.3563, pruned_loss=0.1178, over 26904.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3287, pruned_loss=0.08437, over 5665462.74 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3287, pruned_loss=0.08643, over 5776384.64 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3296, pruned_loss=0.08449, over 5649915.50 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:04:17,281 INFO [train.py:968] (1/2) Epoch 23, batch 16000, giga_loss[loss=0.2526, simple_loss=0.3412, pruned_loss=0.08198, over 28831.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3304, pruned_loss=0.08571, over 5677811.11 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3284, pruned_loss=0.08631, over 5780548.70 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3314, pruned_loss=0.08588, over 5657567.23 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:04:25,254 INFO [optim.py:369] (1/2) 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 22:04:41,750 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019410.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:04:51,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4421, 1.6425, 1.5947, 1.3428], device='cuda:1'), covar=tensor([0.2709, 0.2316, 0.2048, 0.2471], device='cuda:1'), in_proj_covar=tensor([0.1943, 0.1878, 0.1800, 0.1943], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 22:04:53,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4178, 1.5576, 1.2042, 1.1513], device='cuda:1'), covar=tensor([0.0985, 0.0536, 0.1026, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0444, 0.0519, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 22:05:20,432 INFO [train.py:968] (1/2) Epoch 23, batch 16050, giga_loss[loss=0.2636, simple_loss=0.3522, pruned_loss=0.08748, over 28902.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3348, pruned_loss=0.0879, over 5662503.37 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3285, pruned_loss=0.08625, over 5779436.24 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3356, pruned_loss=0.08811, over 5645392.31 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:06:21,448 INFO [train.py:968] (1/2) Epoch 23, batch 16100, giga_loss[loss=0.2432, simple_loss=0.3254, pruned_loss=0.08051, over 28755.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3359, pruned_loss=0.0876, over 5663956.13 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3287, pruned_loss=0.08635, over 5779764.26 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3365, pruned_loss=0.0877, over 5647834.51 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:06:31,342 INFO [optim.py:369] (1/2) 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 22:07:14,788 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-11 22:07:29,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:07:33,964 INFO [train.py:968] (1/2) Epoch 23, batch 16150, giga_loss[loss=0.2355, simple_loss=0.2948, pruned_loss=0.08809, over 24546.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3359, pruned_loss=0.08832, over 5653885.22 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3288, pruned_loss=0.08644, over 5781998.69 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3364, pruned_loss=0.08836, over 5636447.37 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:08:26,223 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1019579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:08:30,503 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 16200, giga_loss[loss=0.2757, simple_loss=0.3412, pruned_loss=0.1052, over 28036.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3332, pruned_loss=0.08692, over 5671809.29 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3288, pruned_loss=0.08655, over 5785615.87 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3338, pruned_loss=0.08685, over 5650978.29 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:08:47,406 INFO [optim.py:369] (1/2) 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 22:08:54,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.90 vs. limit=2.0
+2023-03-11 22:09:45,809 INFO [train.py:968] (1/2) Epoch 23, batch 16250, giga_loss[loss=0.256, simple_loss=0.332, pruned_loss=0.09003, over 28043.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3318, pruned_loss=0.0859, over 5666809.41 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3286, pruned_loss=0.08653, over 5776909.80 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3325, pruned_loss=0.08586, over 5656646.93 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:10:18,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:10:42,154 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:10:45,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:10:49,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 22:10:55,178 INFO [train.py:968] (1/2) Epoch 23, batch 16300, giga_loss[loss=0.2618, simple_loss=0.327, pruned_loss=0.09834, over 26834.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3317, pruned_loss=0.08659, over 5667683.62 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.08643, over 5778373.70 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3323, pruned_loss=0.08666, over 5657307.61 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:11:04,538 INFO [optim.py:369] (1/2) 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 22:11:25,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:11:43,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:11:45,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3003, 1.8722, 1.4876, 0.4828], device='cuda:1'), covar=tensor([0.5022, 0.3230, 0.4839, 0.7603], device='cuda:1'), in_proj_covar=tensor([0.1762, 0.1664, 0.1598, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 22:11:48,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:12:00,615 INFO [train.py:968] (1/2) Epoch 23, batch 16350, libri_loss[loss=0.2308, simple_loss=0.3093, pruned_loss=0.0762, over 29571.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3294, pruned_loss=0.08619, over 5660069.24 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3285, pruned_loss=0.0864, over 5778793.27 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.33, pruned_loss=0.08628, over 5650359.29 frames. ], batch size: 75, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:12:22,714 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:12:55,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019785.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:13:02,643 INFO [train.py:968] (1/2) Epoch 23, batch 16400, giga_loss[loss=0.2708, simple_loss=0.3395, pruned_loss=0.1011, over 26802.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3293, pruned_loss=0.08594, over 5651423.04 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3288, pruned_loss=0.08678, over 5768249.02 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3295, pruned_loss=0.08566, over 5649663.74 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:13:12,815 INFO [optim.py:369] (1/2) 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 22:13:23,250 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:13:25,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:14:02,234 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:14:02,583 INFO [train.py:968] (1/2) Epoch 23, batch 16450, giga_loss[loss=0.2391, simple_loss=0.3331, pruned_loss=0.07254, over 28751.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3277, pruned_loss=0.08365, over 5665263.00 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3286, pruned_loss=0.08675, over 5767204.46 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3281, pruned_loss=0.08341, over 5662503.42 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:14:36,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0459, 1.3053, 1.3158, 1.1042], device='cuda:1'), covar=tensor([0.2704, 0.2043, 0.1480, 0.2161], device='cuda:1'), in_proj_covar=tensor([0.1943, 0.1870, 0.1796, 0.1939], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 22:14:58,402 INFO [train.py:968] (1/2) Epoch 23, batch 16500, giga_loss[loss=0.2604, simple_loss=0.3546, pruned_loss=0.08312, over 28637.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3295, pruned_loss=0.08242, over 5676558.26 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3287, pruned_loss=0.08686, over 5767213.06 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3296, pruned_loss=0.08199, over 5671572.57 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:15:08,047 INFO [optim.py:369] (1/2) 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 22:15:42,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5694, 5.4024, 5.1539, 2.4601], device='cuda:1'), covar=tensor([0.0463, 0.0604, 0.0709, 0.1844], device='cuda:1'), in_proj_covar=tensor([0.1219, 0.1126, 0.0952, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 22:15:42,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1019928.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:15:48,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1019931.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:15:55,269 INFO [train.py:968] (1/2) Epoch 23, batch 16550, libri_loss[loss=0.2526, simple_loss=0.3359, pruned_loss=0.0847, over 29513.00 frames. ], tot_loss[loss=0.248, simple_loss=0.331, pruned_loss=0.08243, over 5677493.78 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3288, pruned_loss=0.08686, over 5770647.23 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3312, pruned_loss=0.08196, over 5667705.46 frames. ], batch size: 89, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:16:09,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1019954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:16:16,080 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1019960.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:16:56,352 INFO [train.py:968] (1/2) Epoch 23, batch 16600, giga_loss[loss=0.2405, simple_loss=0.3276, pruned_loss=0.07675, over 29000.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3313, pruned_loss=0.08271, over 5671736.18 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3287, pruned_loss=0.08688, over 5768129.94 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3315, pruned_loss=0.08223, over 5664217.06 frames. ], batch size: 199, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:17:09,661 INFO [optim.py:369] (1/2) 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 22:17:16,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4438, 1.3450, 1.3041, 1.5454], device='cuda:1'), covar=tensor([0.0709, 0.0342, 0.0327, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-11 22:17:28,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3445, 1.2817, 3.6537, 3.1380], device='cuda:1'), covar=tensor([0.1522, 0.2780, 0.0474, 0.1219], device='cuda:1'), in_proj_covar=tensor([0.0760, 0.0652, 0.0954, 0.0908], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 22:17:50,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 22:18:07,184 INFO [train.py:968] (1/2) Epoch 23, batch 16650, giga_loss[loss=0.2484, simple_loss=0.3234, pruned_loss=0.0867, over 26910.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3313, pruned_loss=0.0831, over 5664853.60 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3288, pruned_loss=0.08689, over 5767790.68 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3315, pruned_loss=0.08264, over 5657347.81 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 22:19:02,551 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-11 22:19:12,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7388, 1.9573, 1.4476, 1.5362], device='cuda:1'), covar=tensor([0.1007, 0.0556, 0.0993, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0443, 0.0520, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 22:19:13,114 INFO [train.py:968] (1/2) Epoch 23, batch 16700, giga_loss[loss=0.2251, simple_loss=0.309, pruned_loss=0.07054, over 28720.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3296, pruned_loss=0.08195, over 5663398.44 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3281, pruned_loss=0.0865, over 5771508.90 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3303, pruned_loss=0.08177, over 5650936.98 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 22:19:23,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1020097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:19:25,495 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:19:40,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3871, 1.5272, 1.2299, 1.4853], device='cuda:1'), covar=tensor([0.0793, 0.0353, 0.0369, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-11 22:20:09,350 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 16750, giga_loss[loss=0.2319, simple_loss=0.292, pruned_loss=0.08595, over 24643.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3298, pruned_loss=0.08142, over 5662909.76 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.328, pruned_loss=0.08634, over 5773705.37 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3305, pruned_loss=0.08127, over 5647692.11 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 22:21:37,966 INFO [train.py:968] (1/2) Epoch 23, batch 16800, giga_loss[loss=0.2766, simple_loss=0.3634, pruned_loss=0.09488, over 28878.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3312, pruned_loss=0.08224, over 5667183.11 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3272, pruned_loss=0.08596, over 5777484.42 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3327, pruned_loss=0.08235, over 5647849.37 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:21:53,287 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 16850, giga_loss[loss=0.2354, simple_loss=0.3274, pruned_loss=0.07174, over 28182.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.334, pruned_loss=0.0831, over 5669928.73 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3272, pruned_loss=0.08595, over 5776638.32 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3353, pruned_loss=0.08314, over 5653671.07 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:23:10,445 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:23:55,683 INFO [train.py:968] (1/2) Epoch 23, batch 16900, giga_loss[loss=0.2153, simple_loss=0.3028, pruned_loss=0.06388, over 28765.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3341, pruned_loss=0.08354, over 5686851.31 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3268, pruned_loss=0.08556, over 5780101.73 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3358, pruned_loss=0.08386, over 5667333.11 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:24:02,151 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-11 22:24:11,606 INFO [optim.py:369] (1/2) 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 22:24:27,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5392, 1.7994, 1.4712, 1.7312], device='cuda:1'), covar=tensor([0.2731, 0.2865, 0.3346, 0.2579], device='cuda:1'), in_proj_covar=tensor([0.1521, 0.1096, 0.1346, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 22:24:44,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3385, 3.1929, 3.0340, 1.4086], device='cuda:1'), covar=tensor([0.0929, 0.1038, 0.0956, 0.2239], device='cuda:1'), in_proj_covar=tensor([0.1216, 0.1120, 0.0950, 0.0713], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0011], device='cuda:1')
+2023-03-11 22:25:08,913 INFO [train.py:968] (1/2) Epoch 23, batch 16950, giga_loss[loss=0.2103, simple_loss=0.3011, pruned_loss=0.05977, over 29081.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3336, pruned_loss=0.08427, over 5676930.53 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3271, pruned_loss=0.08582, over 5774204.30 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3348, pruned_loss=0.08423, over 5663433.72 frames. ], batch size: 120, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:25:12,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1047, 1.2959, 3.2620, 2.9923], device='cuda:1'), covar=tensor([0.1601, 0.2670, 0.0520, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0653, 0.0956, 0.0908], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 22:25:50,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5841, 1.6734, 1.8147, 1.3898], device='cuda:1'), covar=tensor([0.2108, 0.2843, 0.1695, 0.1951], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0692, 0.0947, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 22:26:15,315 INFO [train.py:968] (1/2) Epoch 23, batch 17000, giga_loss[loss=0.268, simple_loss=0.3477, pruned_loss=0.09419, over 28770.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3319, pruned_loss=0.08285, over 5685275.50 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3274, pruned_loss=0.08584, over 5777765.85 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3328, pruned_loss=0.08273, over 5668319.61 frames. ], batch size: 243, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:26:31,470 INFO [optim.py:369] (1/2) 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 22:27:11,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5150, 1.5409, 1.7346, 1.3124], device='cuda:1'), covar=tensor([0.1976, 0.3016, 0.1682, 0.2043], device='cuda:1'), in_proj_covar=tensor([0.0901, 0.0693, 0.0948, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0015, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 22:27:13,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4635, 1.8651, 1.8521, 1.5613], device='cuda:1'), covar=tensor([0.2034, 0.1971, 0.2106, 0.2013], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0728, 0.0698, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 22:27:25,538 INFO [train.py:968] (1/2) Epoch 23, batch 17050, giga_loss[loss=0.2513, simple_loss=0.3404, pruned_loss=0.0811, over 28623.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3309, pruned_loss=0.08212, over 5680151.06 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3276, pruned_loss=0.08586, over 5777289.57 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3316, pruned_loss=0.08194, over 5664727.35 frames. ], batch size: 307, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:27:35,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5509, 4.3852, 4.1548, 1.9705], device='cuda:1'), covar=tensor([0.0598, 0.0734, 0.0804, 0.2123], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.1120, 0.0952, 0.0715], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0012, 0.0012], device='cuda:1')
+2023-03-11 22:27:44,576 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4815, 1.8765, 1.6866, 1.6457], device='cuda:1'), covar=tensor([0.1949, 0.2201, 0.2048, 0.2088], device='cuda:1'), in_proj_covar=tensor([0.0460, 0.0729, 0.0698, 0.0667], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 22:28:07,731 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:28:16,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5516, 4.6958, 1.7366, 1.8291], device='cuda:1'), covar=tensor([0.0982, 0.0277, 0.0915, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0552, 0.0391, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 22:28:27,520 INFO [train.py:968] (1/2) Epoch 23, batch 17100, giga_loss[loss=0.2578, simple_loss=0.3416, pruned_loss=0.08706, over 28886.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3318, pruned_loss=0.08275, over 5683218.05 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3276, pruned_loss=0.08596, over 5778408.81 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3324, pruned_loss=0.08245, over 5668344.28 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:28:40,249 INFO [optim.py:369] (1/2) 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 22:29:28,932 INFO [train.py:968] (1/2) Epoch 23, batch 17150, giga_loss[loss=0.2574, simple_loss=0.3407, pruned_loss=0.08705, over 28388.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3345, pruned_loss=0.08456, over 5676201.44 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.327, pruned_loss=0.08561, over 5780795.98 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3356, pruned_loss=0.08461, over 5660503.39 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:29:44,641 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:30:24,715 INFO [train.py:968] (1/2) Epoch 23, batch 17200, giga_loss[loss=0.2813, simple_loss=0.3333, pruned_loss=0.1147, over 26799.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3328, pruned_loss=0.08442, over 5678176.05 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3269, pruned_loss=0.08547, over 5780969.57 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3339, pruned_loss=0.08455, over 5663266.97 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:30:37,311 INFO [optim.py:369] (1/2) 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 22:31:13,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1020629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:31:27,893 INFO [train.py:968] (1/2) Epoch 23, batch 17250, giga_loss[loss=0.2575, simple_loss=0.339, pruned_loss=0.08803, over 28719.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3303, pruned_loss=0.08422, over 5672249.50 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3271, pruned_loss=0.08545, over 5783326.57 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3312, pruned_loss=0.08432, over 5656047.33 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:32:31,998 INFO [train.py:968] (1/2) Epoch 23, batch 17300, giga_loss[loss=0.2526, simple_loss=0.3135, pruned_loss=0.09585, over 24324.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3306, pruned_loss=0.08514, over 5664328.48 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3271, pruned_loss=0.08547, over 5785635.24 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3313, pruned_loss=0.08518, over 5647443.90 frames. ], batch size: 705, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:32:45,768 INFO [optim.py:369] (1/2) 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 22:33:11,959 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1862, 1.5827, 1.3642, 1.3418], device='cuda:1'), covar=tensor([0.0834, 0.0356, 0.0317, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0111], device='cuda:1')
+2023-03-11 22:33:28,588 INFO [train.py:968] (1/2) Epoch 23, batch 17350, giga_loss[loss=0.2828, simple_loss=0.3738, pruned_loss=0.09592, over 28845.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3363, pruned_loss=0.08802, over 5674626.49 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3264, pruned_loss=0.08499, over 5788694.73 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3378, pruned_loss=0.08858, over 5654933.85 frames. ], batch size: 174, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:33:58,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1020772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:34:01,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1020775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:34:11,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9483, 1.3273, 1.0974, 0.2885], device='cuda:1'), covar=tensor([0.3316, 0.2747, 0.3799, 0.5107], device='cuda:1'), in_proj_covar=tensor([0.1765, 0.1667, 0.1604, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 22:34:13,413 INFO [train.py:968] (1/2) Epoch 23, batch 17400, giga_loss[loss=0.311, simple_loss=0.389, pruned_loss=0.1165, over 28292.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3447, pruned_loss=0.09253, over 5680474.12 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.326, pruned_loss=0.08479, over 5791141.78 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3465, pruned_loss=0.09328, over 5660585.82 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:34:20,692 INFO [optim.py:369] (1/2) 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 22:34:23,041 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1020804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:34:48,238 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1793, 1.4804, 1.5054, 1.3040], device='cuda:1'), covar=tensor([0.2093, 0.1739, 0.2363, 0.1945], device='cuda:1'), in_proj_covar=tensor([0.0463, 0.0733, 0.0703, 0.0671], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-11 22:34:59,671 INFO [train.py:968] (1/2) Epoch 23, batch 17450, libri_loss[loss=0.2792, simple_loss=0.3562, pruned_loss=0.1011, over 28727.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3483, pruned_loss=0.09526, over 5675726.02 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3263, pruned_loss=0.08498, over 5783473.24 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3502, pruned_loss=0.09602, over 5661876.11 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:35:05,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1020849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:35:14,694 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-11 22:35:44,827 INFO [train.py:968] (1/2) Epoch 23, batch 17500, giga_loss[loss=0.2581, simple_loss=0.3296, pruned_loss=0.09334, over 27649.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3454, pruned_loss=0.0947, over 5673989.05 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3268, pruned_loss=0.08515, over 5780600.24 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3469, pruned_loss=0.09539, over 5662610.38 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:35:54,171 INFO [optim.py:369] (1/2) 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 22:36:03,150 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:36:10,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1020922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:36:17,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1020930.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:36:25,861 INFO [train.py:968] (1/2) Epoch 23, batch 17550, giga_loss[loss=0.2298, simple_loss=0.3015, pruned_loss=0.07909, over 28496.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.09141, over 5679054.08 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3267, pruned_loss=0.08498, over 5775181.84 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.34, pruned_loss=0.09234, over 5672636.19 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:37:11,501 INFO [train.py:968] (1/2) Epoch 23, batch 17600, libri_loss[loss=0.2459, simple_loss=0.3441, pruned_loss=0.07391, over 29545.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3313, pruned_loss=0.08844, over 5688888.05 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3266, pruned_loss=0.08478, over 5777572.86 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3329, pruned_loss=0.08944, over 5679958.94 frames. ], batch size: 82, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:37:12,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1020992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:37:14,927 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:37:33,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4393, 4.4505, 1.7213, 1.6888], device='cuda:1'), covar=tensor([0.1041, 0.0271, 0.0904, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0550, 0.0389, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 22:37:42,072 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:37:57,225 INFO [train.py:968] (1/2) Epoch 23, batch 17650, libri_loss[loss=0.2506, simple_loss=0.3367, pruned_loss=0.08221, over 29503.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3254, pruned_loss=0.08608, over 5691120.14 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3273, pruned_loss=0.08496, over 5779026.52 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.326, pruned_loss=0.08679, over 5679826.47 frames. ], batch size: 84, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:38:24,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021073.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:38:27,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021076.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 22:38:40,212 INFO [train.py:968] (1/2) Epoch 23, batch 17700, giga_loss[loss=0.1988, simple_loss=0.2817, pruned_loss=0.05795, over 28785.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3188, pruned_loss=0.08348, over 5683708.17 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3277, pruned_loss=0.08514, over 5771163.42 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3189, pruned_loss=0.08388, over 5680011.91 frames. ], batch size: 66, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:38:49,468 INFO [optim.py:369] (1/2) 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 22:38:52,878 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021105.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 22:39:02,900 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-11 22:39:23,147 INFO [train.py:968] (1/2) Epoch 23, batch 17750, giga_loss[loss=0.2224, simple_loss=0.293, pruned_loss=0.07593, over 28949.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.3142, pruned_loss=0.08138, over 5691726.49 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08508, over 5774593.02 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.314, pruned_loss=0.08166, over 5683443.62 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:39:27,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5002, 2.0585, 1.6927, 0.7309], device='cuda:1'), covar=tensor([0.6304, 0.3393, 0.4010, 0.7160], device='cuda:1'), in_proj_covar=tensor([0.1754, 0.1660, 0.1597, 0.1433], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 22:39:53,406 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4317, 3.1683, 1.4100, 1.6170], device='cuda:1'), covar=tensor([0.0967, 0.0293, 0.0938, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0550, 0.0390, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 22:40:03,417 INFO [train.py:968] (1/2) Epoch 23, batch 17800, giga_loss[loss=0.268, simple_loss=0.3241, pruned_loss=0.106, over 26663.00 frames. ], tot_loss[loss=0.2358, simple_loss=0.3115, pruned_loss=0.08006, over 5692423.52 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.328, pruned_loss=0.08523, over 5767613.60 frames. ], giga_tot_loss[loss=0.2353, simple_loss=0.3106, pruned_loss=0.08004, over 5689752.19 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:40:10,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2770, 4.1121, 3.8600, 2.0749], device='cuda:1'), covar=tensor([0.0590, 0.0799, 0.0765, 0.1999], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1130, 0.0960, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 22:40:14,151 INFO [optim.py:369] (1/2) 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 22:40:46,331 INFO [train.py:968] (1/2) Epoch 23, batch 17850, giga_loss[loss=0.1992, simple_loss=0.2749, pruned_loss=0.0617, over 28338.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3087, pruned_loss=0.07902, over 5691429.38 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.328, pruned_loss=0.08516, over 5771472.54 frames. ], giga_tot_loss[loss=0.2327, simple_loss=0.3075, pruned_loss=0.07891, over 5683804.39 frames. ], batch size: 65, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:40:56,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-11 22:41:14,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6288, 1.9546, 1.6035, 1.6276], device='cuda:1'), covar=tensor([0.2818, 0.2752, 0.3195, 0.2408], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1101, 0.1351, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 22:41:17,494 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:41:26,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 17900, giga_loss[loss=0.2219, simple_loss=0.3002, pruned_loss=0.07184, over 29004.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.3059, pruned_loss=0.07756, over 5694000.31 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3283, pruned_loss=0.08527, over 5770816.41 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3043, pruned_loss=0.07724, over 5687373.43 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:41:35,143 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:41:37,024 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:41:42,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3529, 1.2272, 3.7487, 3.1405], device='cuda:1'), covar=tensor([0.1648, 0.2919, 0.0460, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0652, 0.0961, 0.0914], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 22:42:12,917 INFO [train.py:968] (1/2) Epoch 23, batch 17950, giga_loss[loss=0.242, simple_loss=0.3083, pruned_loss=0.08792, over 28952.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3023, pruned_loss=0.07585, over 5690665.15 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3287, pruned_loss=0.08549, over 5763373.09 frames. ], giga_tot_loss[loss=0.2255, simple_loss=0.3003, pruned_loss=0.07532, over 5690446.49 frames. ], batch size: 213, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:42:59,487 INFO [train.py:968] (1/2) Epoch 23, batch 18000, giga_loss[loss=0.2241, simple_loss=0.3045, pruned_loss=0.07186, over 28884.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3006, pruned_loss=0.07566, over 5677288.46 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.329, pruned_loss=0.08575, over 5754214.13 frames. ], giga_tot_loss[loss=0.2241, simple_loss=0.2985, pruned_loss=0.0749, over 5683646.76 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:42:59,488 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 22:43:08,693 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 22:43:17,230 INFO [optim.py:369] (1/2) 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 22:43:29,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3356, 1.6586, 1.0026, 1.3234], device='cuda:1'), covar=tensor([0.1312, 0.0742, 0.1621, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0441, 0.0516, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 22:43:31,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8374, 1.9697, 2.0546, 1.6172], device='cuda:1'), covar=tensor([0.1873, 0.2437, 0.1482, 0.1672], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0700, 0.0959, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 22:43:41,307 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:43:45,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:43:50,284 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 18050, giga_loss[loss=0.2438, simple_loss=0.3119, pruned_loss=0.08779, over 27639.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2989, pruned_loss=0.07486, over 5676633.85 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3292, pruned_loss=0.08575, over 5751085.63 frames. ], giga_tot_loss[loss=0.2217, simple_loss=0.2958, pruned_loss=0.07381, over 5682258.91 frames. ], batch size: 472, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:43:52,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:44:12,396 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:44:19,462 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:44:38,252 INFO [train.py:968] (1/2) Epoch 23, batch 18100, giga_loss[loss=0.2365, simple_loss=0.3161, pruned_loss=0.07846, over 28988.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2951, pruned_loss=0.07301, over 5691520.54 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3291, pruned_loss=0.08568, over 5753291.34 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2924, pruned_loss=0.07209, over 5693178.19 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:44:47,369 INFO [optim.py:369] (1/2) 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 22:45:24,496 INFO [train.py:968] (1/2) Epoch 23, batch 18150, giga_loss[loss=0.2714, simple_loss=0.331, pruned_loss=0.1059, over 28189.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2957, pruned_loss=0.07402, over 5694222.58 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3298, pruned_loss=0.08598, over 5755646.90 frames. ], giga_tot_loss[loss=0.2191, simple_loss=0.2925, pruned_loss=0.07288, over 5692850.09 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:45:27,589 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 18200, giga_loss[loss=0.2742, simple_loss=0.3513, pruned_loss=0.09859, over 29008.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.3063, pruned_loss=0.07929, over 5692222.18 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3298, pruned_loss=0.08583, over 5755395.42 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.3029, pruned_loss=0.07823, over 5689717.01 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:46:26,374 INFO [optim.py:369] (1/2) 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 22:47:01,961 INFO [train.py:968] (1/2) Epoch 23, batch 18250, giga_loss[loss=0.2948, simple_loss=0.3696, pruned_loss=0.11, over 28709.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3197, pruned_loss=0.08586, over 5690561.36 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3303, pruned_loss=0.08592, over 5755718.23 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3162, pruned_loss=0.08485, over 5686930.63 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:47:05,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3132, 1.5637, 1.1999, 1.5727], device='cuda:1'), covar=tensor([0.0804, 0.0336, 0.0357, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-11 22:47:12,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:47:28,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:47:34,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4968, 4.3074, 4.0948, 1.9381], device='cuda:1'), covar=tensor([0.0682, 0.0870, 0.0944, 0.2134], device='cuda:1'), in_proj_covar=tensor([0.1218, 0.1126, 0.0955, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 22:47:42,770 INFO [train.py:968] (1/2) Epoch 23, batch 18300, giga_loss[loss=0.2689, simple_loss=0.3451, pruned_loss=0.09635, over 28800.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3312, pruned_loss=0.09138, over 5686426.61 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3314, pruned_loss=0.08656, over 5740658.79 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3273, pruned_loss=0.09005, over 5696421.70 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:47:53,290 INFO [optim.py:369] (1/2) 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 22:47:54,427 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:48:09,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6433, 1.8346, 1.1946, 1.4626], device='cuda:1'), covar=tensor([0.1119, 0.0757, 0.1180, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0442, 0.0519, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-11 22:48:26,318 INFO [train.py:968] (1/2) Epoch 23, batch 18350, giga_loss[loss=0.256, simple_loss=0.3395, pruned_loss=0.0862, over 28938.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3368, pruned_loss=0.093, over 5684268.09 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3317, pruned_loss=0.08656, over 5743011.61 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3335, pruned_loss=0.09206, over 5688966.33 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:49:10,210 INFO [train.py:968] (1/2) Epoch 23, batch 18400, giga_loss[loss=0.2631, simple_loss=0.3449, pruned_loss=0.09065, over 28813.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3396, pruned_loss=0.09311, over 5688120.96 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3315, pruned_loss=0.08641, over 5744743.79 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3373, pruned_loss=0.09257, over 5689833.75 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:49:15,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:20,442 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:49:39,285 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1021818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:41,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1021821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:49,341 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:49:58,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1021839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:50:00,025 INFO [train.py:968] (1/2) Epoch 23, batch 18450, libri_loss[loss=0.2251, simple_loss=0.3052, pruned_loss=0.07253, over 29575.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3405, pruned_loss=0.09322, over 5685105.40 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3317, pruned_loss=0.08645, over 5745555.90 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3387, pruned_loss=0.09292, over 5684011.17 frames. ], batch size: 78, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:50:10,511 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1021850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:50:29,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5608, 2.1100, 1.6024, 0.7195], device='cuda:1'), covar=tensor([0.5397, 0.2825, 0.3674, 0.6201], device='cuda:1'), in_proj_covar=tensor([0.1751, 0.1656, 0.1595, 0.1428], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 22:50:43,717 INFO [train.py:968] (1/2) Epoch 23, batch 18500, giga_loss[loss=0.2699, simple_loss=0.3502, pruned_loss=0.09475, over 28869.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.343, pruned_loss=0.09512, over 5689984.52 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3321, pruned_loss=0.08671, over 5749883.48 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3414, pruned_loss=0.09483, over 5683806.48 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:50:54,932 INFO [optim.py:369] (1/2) 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 22:51:08,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1021918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:51:29,382 INFO [train.py:968] (1/2) Epoch 23, batch 18550, libri_loss[loss=0.2265, simple_loss=0.3102, pruned_loss=0.07134, over 29358.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3464, pruned_loss=0.0974, over 5699016.32 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.332, pruned_loss=0.08655, over 5753786.61 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3455, pruned_loss=0.09759, over 5689044.44 frames. ], batch size: 71, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:52:11,148 INFO [train.py:968] (1/2) Epoch 23, batch 18600, libri_loss[loss=0.2575, simple_loss=0.3472, pruned_loss=0.08386, over 29211.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3485, pruned_loss=0.0984, over 5705875.28 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3322, pruned_loss=0.08655, over 5756938.86 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3481, pruned_loss=0.09888, over 5693439.85 frames. ], batch size: 97, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:52:19,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1022001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:52:21,061 INFO [optim.py:369] (1/2) 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 22:52:52,945 INFO [train.py:968] (1/2) Epoch 23, batch 18650, giga_loss[loss=0.3231, simple_loss=0.3954, pruned_loss=0.1254, over 28620.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3509, pruned_loss=0.09892, over 5705533.52 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3321, pruned_loss=0.08634, over 5752837.73 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3515, pruned_loss=0.1001, over 5695808.11 frames. ], batch size: 60, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:53:07,802 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:53:12,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1022064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:53:25,111 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 18700, giga_loss[loss=0.2537, simple_loss=0.3406, pruned_loss=0.08339, over 28809.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3528, pruned_loss=0.09954, over 5683292.38 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3324, pruned_loss=0.08663, over 5724511.41 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3533, pruned_loss=0.1004, over 5699314.49 frames. ], batch size: 119, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:53:35,740 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1022093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:53:44,947 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 18750, giga_loss[loss=0.2816, simple_loss=0.359, pruned_loss=0.1021, over 28309.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3545, pruned_loss=0.09965, over 5683748.33 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.333, pruned_loss=0.08684, over 5722860.24 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3547, pruned_loss=0.1004, over 5697315.05 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:54:43,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-11 22:54:56,551 INFO [train.py:968] (1/2) Epoch 23, batch 18800, giga_loss[loss=0.2367, simple_loss=0.3268, pruned_loss=0.07326, over 28844.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3532, pruned_loss=0.09779, over 5682710.16 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3333, pruned_loss=0.0868, over 5719759.18 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3537, pruned_loss=0.09881, over 5694648.59 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:55:05,306 INFO [optim.py:369] (1/2) 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 22:55:16,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1022214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:55:21,711 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:55:24,624 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 18850, giga_loss[loss=0.2632, simple_loss=0.3494, pruned_loss=0.08847, over 29072.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.352, pruned_loss=0.09615, over 5683271.88 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3335, pruned_loss=0.08669, over 5712743.61 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3528, pruned_loss=0.09738, over 5697358.94 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:55:48,227 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1022253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:56:19,196 INFO [train.py:968] (1/2) Epoch 23, batch 18900, giga_loss[loss=0.2747, simple_loss=0.3546, pruned_loss=0.09739, over 28028.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3503, pruned_loss=0.09491, over 5691605.24 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3337, pruned_loss=0.08673, over 5715865.64 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3511, pruned_loss=0.09597, over 5699557.74 frames. ], batch size: 77, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:56:28,424 INFO [optim.py:369] (1/2) 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 22:57:03,365 INFO [train.py:968] (1/2) Epoch 23, batch 18950, giga_loss[loss=0.3455, simple_loss=0.3899, pruned_loss=0.1506, over 27821.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3521, pruned_loss=0.09797, over 5677703.72 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3337, pruned_loss=0.08674, over 5709395.28 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3531, pruned_loss=0.099, over 5689522.16 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:57:19,183 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:21,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1022360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:35,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1022376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:48,559 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1022389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:57:49,929 INFO [train.py:968] (1/2) Epoch 23, batch 19000, giga_loss[loss=0.2515, simple_loss=0.3312, pruned_loss=0.08594, over 29079.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.355, pruned_loss=0.1021, over 5664033.39 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3342, pruned_loss=0.087, over 5703795.35 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3557, pruned_loss=0.103, over 5677650.03 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 22:57:51,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-11 22:57:59,786 INFO [optim.py:369] (1/2) 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 22:58:18,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6555, 1.6191, 1.8611, 1.4476], device='cuda:1'), covar=tensor([0.1595, 0.2222, 0.1291, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0697, 0.0954, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 22:58:27,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5499, 1.8341, 1.4881, 1.6899], device='cuda:1'), covar=tensor([0.2569, 0.2614, 0.2929, 0.2283], device='cuda:1'), in_proj_covar=tensor([0.1523, 0.1099, 0.1346, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 22:58:29,206 INFO [train.py:968] (1/2) Epoch 23, batch 19050, giga_loss[loss=0.2664, simple_loss=0.3448, pruned_loss=0.094, over 29079.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3548, pruned_loss=0.103, over 5672454.23 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3345, pruned_loss=0.0872, over 5697921.07 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3556, pruned_loss=0.1039, over 5687349.51 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:59:13,671 INFO [train.py:968] (1/2) Epoch 23, batch 19100, giga_loss[loss=0.2705, simple_loss=0.3379, pruned_loss=0.1015, over 28497.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3525, pruned_loss=0.1025, over 5681932.68 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3344, pruned_loss=0.08707, over 5702008.83 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3536, pruned_loss=0.1036, over 5689486.98 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 22:59:17,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2565, 1.4244, 1.4833, 1.2995], device='cuda:1'), covar=tensor([0.1904, 0.1815, 0.2365, 0.1908], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0746, 0.0714, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 22:59:25,867 INFO [optim.py:369] (1/2) 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 22:59:30,599 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-11 22:59:39,382 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1022519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:59:41,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1022522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 22:59:44,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5717, 1.8008, 1.4249, 1.8203], device='cuda:1'), covar=tensor([0.2653, 0.2782, 0.3031, 0.2590], device='cuda:1'), in_proj_covar=tensor([0.1522, 0.1100, 0.1346, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 22:59:59,687 INFO [train.py:968] (1/2) Epoch 23, batch 19150, giga_loss[loss=0.2702, simple_loss=0.3483, pruned_loss=0.09605, over 28835.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3509, pruned_loss=0.1015, over 5677441.81 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3342, pruned_loss=0.08703, over 5692945.76 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.352, pruned_loss=0.1025, over 5690579.93 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:00:10,364 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 19200, giga_loss[loss=0.2532, simple_loss=0.3407, pruned_loss=0.08286, over 29084.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3506, pruned_loss=0.1008, over 5674567.80 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3345, pruned_loss=0.087, over 5693501.05 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3515, pruned_loss=0.1019, over 5684011.52 frames. ], batch size: 155, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:00:56,441 INFO [optim.py:369] (1/2) 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 23:01:26,590 INFO [train.py:968] (1/2) Epoch 23, batch 19250, libri_loss[loss=0.2857, simple_loss=0.3652, pruned_loss=0.1032, over 29224.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3488, pruned_loss=0.09883, over 5683612.35 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3352, pruned_loss=0.08714, over 5700469.65 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.1, over 5684346.06 frames. ], batch size: 94, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:01:26,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8426, 5.6366, 5.3541, 2.9197], device='cuda:1'), covar=tensor([0.0503, 0.0678, 0.0823, 0.1592], device='cuda:1'), in_proj_covar=tensor([0.1221, 0.1129, 0.0957, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 23:01:41,842 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1022657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:01:55,959 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4513, 3.5548, 1.5111, 1.6383], device='cuda:1'), covar=tensor([0.1070, 0.0332, 0.0995, 0.1431], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0553, 0.0390, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 23:02:02,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7808, 1.7997, 1.9418, 1.5400], device='cuda:1'), covar=tensor([0.1755, 0.2532, 0.1433, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0698, 0.0954, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 23:02:12,296 INFO [train.py:968] (1/2) Epoch 23, batch 19300, giga_loss[loss=0.2446, simple_loss=0.3195, pruned_loss=0.08481, over 28740.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3431, pruned_loss=0.09551, over 5682591.97 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3349, pruned_loss=0.08691, over 5704546.42 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3441, pruned_loss=0.09682, over 5679452.50 frames. ], batch size: 99, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:02:24,090 INFO [optim.py:369] (1/2) 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 23:02:54,095 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 23, batch 19350, giga_loss[loss=0.2399, simple_loss=0.3105, pruned_loss=0.08464, over 28729.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3383, pruned_loss=0.09309, over 5686526.12 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3349, pruned_loss=0.08684, over 5709119.64 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3393, pruned_loss=0.09442, over 5679180.29 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:03:19,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1022764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:03:41,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2969, 1.4551, 1.3172, 1.1769], device='cuda:1'), covar=tensor([0.2431, 0.2493, 0.1863, 0.2332], device='cuda:1'), in_proj_covar=tensor([0.1970, 0.1903, 0.1823, 0.1974], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 23:03:46,296 INFO [train.py:968] (1/2) Epoch 23, batch 19400, giga_loss[loss=0.2273, simple_loss=0.3061, pruned_loss=0.07426, over 28721.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3336, pruned_loss=0.09092, over 5685788.80 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3353, pruned_loss=0.08702, over 5711282.75 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.334, pruned_loss=0.09191, over 5677589.37 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:04:00,789 INFO [optim.py:369] (1/2) 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 23:04:16,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3439, 1.3624, 4.1035, 3.2568], device='cuda:1'), covar=tensor([0.2122, 0.3071, 0.0670, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0652, 0.0961, 0.0917], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 23:04:33,101 INFO [train.py:968] (1/2) Epoch 23, batch 19450, giga_loss[loss=0.2596, simple_loss=0.3358, pruned_loss=0.09169, over 28758.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3322, pruned_loss=0.08937, over 5679092.90 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3358, pruned_loss=0.08705, over 5701077.78 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3319, pruned_loss=0.09022, over 5681224.68 frames. ], batch size: 262, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:04:40,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1268, 1.7095, 1.3446, 0.4075], device='cuda:1'), covar=tensor([0.5174, 0.2951, 0.4058, 0.6141], device='cuda:1'), in_proj_covar=tensor([0.1739, 0.1639, 0.1577, 0.1415], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 23:04:47,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3164, 3.1376, 2.9844, 1.3871], device='cuda:1'), covar=tensor([0.0934, 0.1064, 0.0833, 0.2379], device='cuda:1'), in_proj_covar=tensor([0.1217, 0.1127, 0.0955, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 23:05:04,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-11 23:05:16,217 INFO [train.py:968] (1/2) Epoch 23, batch 19500, giga_loss[loss=0.2403, simple_loss=0.3223, pruned_loss=0.0792, over 28886.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3333, pruned_loss=0.08963, over 5693247.41 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3358, pruned_loss=0.08698, over 5704467.25 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.333, pruned_loss=0.09038, over 5691751.44 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:05:27,860 INFO [optim.py:369] (1/2) 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 23:05:56,878 INFO [train.py:968] (1/2) Epoch 23, batch 19550, giga_loss[loss=0.3381, simple_loss=0.3871, pruned_loss=0.1445, over 26595.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3329, pruned_loss=0.08934, over 5697109.27 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3359, pruned_loss=0.08683, over 5708408.91 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3326, pruned_loss=0.09019, over 5691865.37 frames. ], batch size: 555, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:06:40,012 INFO [train.py:968] (1/2) Epoch 23, batch 19600, libri_loss[loss=0.2769, simple_loss=0.3664, pruned_loss=0.09373, over 29361.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3309, pruned_loss=0.08824, over 5709696.82 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3365, pruned_loss=0.08699, over 5710974.94 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3299, pruned_loss=0.08879, over 5703001.74 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 8.0
+2023-03-11 23:06:50,600 INFO [optim.py:369] (1/2) 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 23:07:11,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:07:16,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6002, 1.6148, 1.8288, 1.4198], device='cuda:1'), covar=tensor([0.1951, 0.2665, 0.1570, 0.1822], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0699, 0.0952, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 23:07:18,376 INFO [train.py:968] (1/2) Epoch 23, batch 19650, giga_loss[loss=0.2351, simple_loss=0.3115, pruned_loss=0.07929, over 29024.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3272, pruned_loss=0.08644, over 5718352.38 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3363, pruned_loss=0.08681, over 5712601.78 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3265, pruned_loss=0.08705, over 5711670.06 frames. ], batch size: 128, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:07:30,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7224, 1.9131, 1.9826, 1.4927], device='cuda:1'), covar=tensor([0.1950, 0.2755, 0.1651, 0.2021], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0698, 0.0952, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 23:07:34,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9003, 1.3054, 1.1090, 0.1417], device='cuda:1'), covar=tensor([0.4436, 0.3028, 0.4733, 0.6954], device='cuda:1'), in_proj_covar=tensor([0.1742, 0.1645, 0.1582, 0.1422], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 23:07:44,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2026, 1.4748, 1.5819, 1.3016], device='cuda:1'), covar=tensor([0.2215, 0.1833, 0.2459, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0753, 0.0720, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 23:07:45,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-11 23:07:59,600 INFO [train.py:968] (1/2) Epoch 23, batch 19700, giga_loss[loss=0.2922, simple_loss=0.3621, pruned_loss=0.1112, over 27975.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3264, pruned_loss=0.08617, over 5720243.53 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3373, pruned_loss=0.08707, over 5716353.94 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3245, pruned_loss=0.08641, over 5711408.97 frames. ], batch size: 412, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:08:08,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-11 23:08:13,375 INFO [optim.py:369] (1/2) 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 23:08:41,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:08:42,524 INFO [train.py:968] (1/2) Epoch 23, batch 19750, giga_loss[loss=0.2366, simple_loss=0.3069, pruned_loss=0.08312, over 28650.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3243, pruned_loss=0.08548, over 5706408.69 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3377, pruned_loss=0.08725, over 5700492.48 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3223, pruned_loss=0.08549, over 5714575.49 frames. ], batch size: 92, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:09:10,468 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1023175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:09:12,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1023178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:09:21,301 INFO [train.py:968] (1/2) Epoch 23, batch 19800, giga_loss[loss=0.2143, simple_loss=0.2925, pruned_loss=0.06806, over 28834.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3222, pruned_loss=0.08433, over 5711538.67 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3383, pruned_loss=0.08732, over 5701895.25 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3197, pruned_loss=0.08421, over 5716989.58 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:09:35,146 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:09:42,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 23:10:03,365 INFO [train.py:968] (1/2) Epoch 23, batch 19850, giga_loss[loss=0.2575, simple_loss=0.3284, pruned_loss=0.09335, over 28605.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3211, pruned_loss=0.08418, over 5700730.73 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3392, pruned_loss=0.08771, over 5694691.63 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3178, pruned_loss=0.08365, over 5712236.17 frames. ], batch size: 336, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:10:36,581 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1023282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:10:39,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1023285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:10:45,164 INFO [train.py:968] (1/2) Epoch 23, batch 19900, giga_loss[loss=0.2139, simple_loss=0.2873, pruned_loss=0.07022, over 28619.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3186, pruned_loss=0.08286, over 5708579.86 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3394, pruned_loss=0.08756, over 5695449.40 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3156, pruned_loss=0.0825, over 5717079.79 frames. ], batch size: 85, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:10:59,000 INFO [optim.py:369] (1/2) 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 23:11:04,206 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 19950, giga_loss[loss=0.2641, simple_loss=0.3378, pruned_loss=0.09524, over 28273.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3171, pruned_loss=0.08182, over 5720904.78 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3398, pruned_loss=0.08757, over 5700469.81 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3138, pruned_loss=0.08139, over 5723461.30 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:11:37,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-11 23:11:52,039 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:12:03,514 INFO [train.py:968] (1/2) Epoch 23, batch 20000, giga_loss[loss=0.2291, simple_loss=0.3118, pruned_loss=0.07323, over 28894.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3157, pruned_loss=0.08127, over 5726758.25 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3394, pruned_loss=0.08717, over 5704237.13 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.313, pruned_loss=0.08118, over 5725863.04 frames. ], batch size: 145, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:12:16,496 INFO [optim.py:369] (1/2) 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 23:12:36,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4615, 2.0541, 1.5580, 0.6364], device='cuda:1'), covar=tensor([0.5122, 0.2651, 0.4800, 0.6579], device='cuda:1'), in_proj_covar=tensor([0.1748, 0.1646, 0.1589, 0.1427], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 23:12:44,597 INFO [train.py:968] (1/2) Epoch 23, batch 20050, giga_loss[loss=0.2725, simple_loss=0.3439, pruned_loss=0.1005, over 28899.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3184, pruned_loss=0.08281, over 5733420.33 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3403, pruned_loss=0.08748, over 5711562.30 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3147, pruned_loss=0.0823, over 5727198.68 frames. ], batch size: 186, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:13:34,526 INFO [train.py:968] (1/2) Epoch 23, batch 20100, giga_loss[loss=0.2667, simple_loss=0.3345, pruned_loss=0.09942, over 28894.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3245, pruned_loss=0.08713, over 5720646.82 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.34, pruned_loss=0.08728, over 5713286.85 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3216, pruned_loss=0.08688, over 5714515.78 frames. ], batch size: 112, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:13:49,432 INFO [optim.py:369] (1/2) 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 23:13:59,247 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:14:25,680 INFO [train.py:968] (1/2) Epoch 23, batch 20150, giga_loss[loss=0.3052, simple_loss=0.3762, pruned_loss=0.1171, over 28764.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3335, pruned_loss=0.09324, over 5706157.94 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3397, pruned_loss=0.08716, over 5719091.62 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3311, pruned_loss=0.0932, over 5696443.22 frames. ], batch size: 284, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:15:11,373 INFO [train.py:968] (1/2) Epoch 23, batch 20200, giga_loss[loss=0.2988, simple_loss=0.37, pruned_loss=0.1138, over 28905.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3402, pruned_loss=0.09706, over 5705359.85 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3401, pruned_loss=0.08714, over 5724454.99 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3379, pruned_loss=0.09725, over 5692513.20 frames. ], batch size: 106, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:15:20,413 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:15:36,775 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-11 23:15:55,695 INFO [train.py:968] (1/2) Epoch 23, batch 20250, giga_loss[loss=0.3027, simple_loss=0.376, pruned_loss=0.1147, over 28772.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3442, pruned_loss=0.09829, over 5697059.23 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3403, pruned_loss=0.08718, over 5730232.85 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3422, pruned_loss=0.09878, over 5680604.64 frames. ], batch size: 242, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:16:42,865 INFO [train.py:968] (1/2) Epoch 23, batch 20300, giga_loss[loss=0.2822, simple_loss=0.3582, pruned_loss=0.1031, over 28962.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3487, pruned_loss=0.1005, over 5690352.75 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3399, pruned_loss=0.08696, over 5731863.31 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3475, pruned_loss=0.1012, over 5675419.96 frames. ], batch size: 164, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:17:00,410 INFO [optim.py:369] (1/2) 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 23:17:01,235 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:17:15,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-11 23:17:29,187 INFO [train.py:968] (1/2) Epoch 23, batch 20350, giga_loss[loss=0.3253, simple_loss=0.3858, pruned_loss=0.1323, over 28158.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3563, pruned_loss=0.1054, over 5684647.04 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3401, pruned_loss=0.08722, over 5735078.72 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3554, pruned_loss=0.106, over 5669389.69 frames. ], batch size: 368, lr: 1.39e-03, grad_scale: 2.0
+2023-03-11 23:17:39,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:17:56,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2999, 0.7996, 0.8130, 1.4146], device='cuda:1'), covar=tensor([0.0805, 0.0388, 0.0377, 0.0899], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 23:18:14,715 INFO [train.py:968] (1/2) Epoch 23, batch 20400, giga_loss[loss=0.2581, simple_loss=0.3454, pruned_loss=0.08539, over 28861.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3531, pruned_loss=0.1025, over 5683234.37 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3408, pruned_loss=0.0877, over 5728516.03 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.103, over 5675499.13 frames. ], batch size: 227, lr: 1.39e-03, grad_scale: 4.0
+2023-03-11 23:18:32,593 INFO [optim.py:369] (1/2) 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 23:18:57,005 INFO [train.py:968] (1/2) Epoch 23, batch 20450, giga_loss[loss=0.2598, simple_loss=0.3474, pruned_loss=0.08611, over 28728.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3495, pruned_loss=0.09949, over 5688717.37 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3411, pruned_loss=0.08783, over 5730791.62 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3487, pruned_loss=0.1, over 5679489.46 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:19:21,010 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1023867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:39,010 INFO [train.py:968] (1/2) Epoch 23, batch 20500, giga_loss[loss=0.2311, simple_loss=0.3156, pruned_loss=0.07331, over 28539.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3482, pruned_loss=0.09805, over 5691104.32 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3419, pruned_loss=0.08842, over 5726239.94 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3472, pruned_loss=0.09845, over 5686207.52 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:19:42,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:42,816 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1023894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:46,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1023897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:19:54,215 INFO [optim.py:369] (1/2) 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 23:20:11,052 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1023926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:20:25,254 INFO [train.py:968] (1/2) Epoch 23, batch 20550, giga_loss[loss=0.2626, simple_loss=0.3476, pruned_loss=0.08882, over 28613.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3488, pruned_loss=0.09809, over 5687815.46 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08852, over 5729269.68 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.348, pruned_loss=0.09849, over 5680638.49 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:20:45,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5490, 1.7802, 1.4787, 1.4898], device='cuda:1'), covar=tensor([0.2455, 0.2486, 0.2695, 0.2340], device='cuda:1'), in_proj_covar=tensor([0.1527, 0.1102, 0.1352, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 23:20:50,299 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-11 23:20:54,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2725, 2.3800, 2.2043, 2.1354], device='cuda:1'), covar=tensor([0.1934, 0.2263, 0.2056, 0.1979], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0749, 0.0715, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 23:20:55,430 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1023976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:06,530 INFO [train.py:968] (1/2) Epoch 23, batch 20600, libri_loss[loss=0.2422, simple_loss=0.326, pruned_loss=0.07918, over 29572.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3507, pruned_loss=0.09927, over 5694479.34 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3421, pruned_loss=0.08846, over 5734558.37 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3502, pruned_loss=0.09999, over 5682682.35 frames. ], batch size: 76, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:21:19,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7647, 1.7852, 1.9790, 1.5389], device='cuda:1'), covar=tensor([0.1838, 0.2566, 0.1511, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0909, 0.0701, 0.0956, 0.0854], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 23:21:20,865 INFO [optim.py:369] (1/2) 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 23:21:29,864 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:36,559 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:41,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:45,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:47,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:21:49,430 INFO [train.py:968] (1/2) Epoch 23, batch 20650, giga_loss[loss=0.3124, simple_loss=0.3683, pruned_loss=0.1283, over 26560.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3521, pruned_loss=0.1005, over 5696435.41 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3422, pruned_loss=0.0885, over 5727267.39 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3518, pruned_loss=0.1012, over 5693271.00 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:21:53,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4582, 1.1526, 4.2888, 3.3812], device='cuda:1'), covar=tensor([0.1628, 0.2951, 0.0421, 0.1174], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0650, 0.0959, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 23:22:12,382 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:22:30,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:22:34,608 INFO [train.py:968] (1/2) Epoch 23, batch 20700, giga_loss[loss=0.2888, simple_loss=0.3602, pruned_loss=0.1086, over 27723.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3533, pruned_loss=0.1015, over 5686013.26 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3421, pruned_loss=0.08842, over 5722686.41 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3534, pruned_loss=0.1026, over 5686058.28 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:22:53,513 INFO [optim.py:369] (1/2) 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 23:23:04,069 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:23:06,120 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 20750, giga_loss[loss=0.2746, simple_loss=0.3525, pruned_loss=0.09835, over 28924.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3552, pruned_loss=0.1037, over 5684467.59 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3421, pruned_loss=0.08831, over 5725161.90 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3555, pruned_loss=0.1048, over 5681937.18 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:23:33,027 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:24:03,298 INFO [train.py:968] (1/2) Epoch 23, batch 20800, giga_loss[loss=0.2912, simple_loss=0.37, pruned_loss=0.1062, over 28770.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3538, pruned_loss=0.1024, over 5694921.67 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3423, pruned_loss=0.08807, over 5731769.64 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.1041, over 5685399.53 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:24:10,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-11 23:24:17,109 INFO [optim.py:369] (1/2) 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 23:24:33,138 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:24:35,333 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 20850, giga_loss[loss=0.2902, simple_loss=0.3607, pruned_loss=0.1098, over 28958.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3538, pruned_loss=0.1016, over 5704140.78 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3423, pruned_loss=0.08806, over 5732778.16 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3543, pruned_loss=0.1032, over 5695234.80 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:24:43,786 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:25:07,259 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:25:24,504 INFO [train.py:968] (1/2) Epoch 23, batch 20900, giga_loss[loss=0.2653, simple_loss=0.3508, pruned_loss=0.08991, over 28877.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3529, pruned_loss=0.1003, over 5705936.16 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3427, pruned_loss=0.08846, over 5737909.89 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3534, pruned_loss=0.1015, over 5692875.55 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:25:40,184 INFO [optim.py:369] (1/2) 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 23:25:46,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5151, 1.6736, 1.7368, 1.4083], device='cuda:1'), covar=tensor([0.2907, 0.2547, 0.1982, 0.2649], device='cuda:1'), in_proj_covar=tensor([0.1978, 0.1913, 0.1838, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-11 23:26:06,598 INFO [train.py:968] (1/2) Epoch 23, batch 20950, giga_loss[loss=0.3424, simple_loss=0.386, pruned_loss=0.1494, over 26807.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3535, pruned_loss=0.09989, over 5705176.92 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3432, pruned_loss=0.08868, over 5738987.68 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3538, pruned_loss=0.1011, over 5692605.14 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:26:41,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:26:42,357 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:26:44,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:26:46,763 INFO [train.py:968] (1/2) Epoch 23, batch 21000, giga_loss[loss=0.2348, simple_loss=0.3271, pruned_loss=0.07122, over 28636.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3518, pruned_loss=0.09904, over 5711348.70 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3435, pruned_loss=0.08895, over 5741889.24 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3519, pruned_loss=0.1, over 5698123.07 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:26:46,764 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-11 23:26:56,891 INFO [train.py:1012] (1/2) Epoch 23, validation: loss=0.2072, simple_loss=0.314, pruned_loss=0.05021, over 944034.00 frames. 
+2023-03-11 23:26:56,891 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-11 23:26:59,566 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:05,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:05,783 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:09,255 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:27:16,529 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:27:35,816 INFO [train.py:968] (1/2) Epoch 23, batch 21050, giga_loss[loss=0.3814, simple_loss=0.4108, pruned_loss=0.1761, over 26649.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3497, pruned_loss=0.09837, over 5706159.70 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3436, pruned_loss=0.08915, over 5735026.05 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3499, pruned_loss=0.09919, over 5700424.05 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:28:09,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1024483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:28:15,365 INFO [train.py:968] (1/2) Epoch 23, batch 21100, giga_loss[loss=0.2514, simple_loss=0.3362, pruned_loss=0.08335, over 29043.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3489, pruned_loss=0.09808, over 5711564.30 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3439, pruned_loss=0.08921, over 5733945.86 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.349, pruned_loss=0.09879, over 5707786.28 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:28:30,482 INFO [optim.py:369] (1/2) 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 23:28:55,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:28:57,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:28:58,217 INFO [train.py:968] (1/2) Epoch 23, batch 21150, giga_loss[loss=0.2843, simple_loss=0.355, pruned_loss=0.1069, over 28752.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3487, pruned_loss=0.09872, over 5699572.06 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3443, pruned_loss=0.08948, over 5728478.58 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3485, pruned_loss=0.09925, over 5701378.92 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:28:59,750 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:01,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:04,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:05,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3999, 2.0233, 1.5160, 0.6381], device='cuda:1'), covar=tensor([0.5081, 0.2749, 0.4319, 0.5997], device='cuda:1'), in_proj_covar=tensor([0.1750, 0.1641, 0.1586, 0.1425], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 23:29:08,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:20,238 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:24,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:28,723 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:29:39,425 INFO [train.py:968] (1/2) Epoch 23, batch 21200, giga_loss[loss=0.2607, simple_loss=0.3377, pruned_loss=0.09185, over 28902.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3488, pruned_loss=0.09875, over 5710389.87 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3444, pruned_loss=0.08975, over 5731468.39 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3487, pruned_loss=0.09931, over 5707959.36 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:29:52,225 INFO [optim.py:369] (1/2) 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 23:30:18,877 INFO [train.py:968] (1/2) Epoch 23, batch 21250, giga_loss[loss=0.2494, simple_loss=0.3306, pruned_loss=0.08405, over 28783.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3485, pruned_loss=0.09834, over 5709549.66 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3441, pruned_loss=0.08962, over 5735905.26 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3488, pruned_loss=0.09914, over 5703043.39 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:30:20,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:30:57,811 INFO [train.py:968] (1/2) Epoch 23, batch 21300, giga_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.08808, over 28611.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3463, pruned_loss=0.09594, over 5713641.95 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3445, pruned_loss=0.08992, over 5737559.49 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3463, pruned_loss=0.09646, over 5706313.57 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:31:13,523 INFO [optim.py:369] (1/2) 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 23:31:19,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5929, 1.8836, 1.6896, 1.5995], device='cuda:1'), covar=tensor([0.1940, 0.2174, 0.2214, 0.2174], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0751, 0.0716, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-11 23:31:25,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-11 23:31:38,808 INFO [train.py:968] (1/2) Epoch 23, batch 21350, giga_loss[loss=0.2718, simple_loss=0.3459, pruned_loss=0.09887, over 28946.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3458, pruned_loss=0.09546, over 5721713.07 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3449, pruned_loss=0.09027, over 5738896.46 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3454, pruned_loss=0.09571, over 5714169.13 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:31:52,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:09,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:18,442 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:20,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:21,462 INFO [train.py:968] (1/2) Epoch 23, batch 21400, giga_loss[loss=0.2561, simple_loss=0.3351, pruned_loss=0.08855, over 28898.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3438, pruned_loss=0.09466, over 5722976.85 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3449, pruned_loss=0.09027, over 5738896.46 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3435, pruned_loss=0.09486, over 5717105.31 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:32:34,990 INFO [optim.py:369] (1/2) 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 23:32:38,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1777, 1.1809, 3.7631, 3.1860], device='cuda:1'), covar=tensor([0.1737, 0.2897, 0.0444, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0648, 0.0958, 0.0915], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 23:32:42,559 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:32:43,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4519, 3.4567, 1.5963, 1.5805], device='cuda:1'), covar=tensor([0.1029, 0.0291, 0.0900, 0.1390], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0548, 0.0387, 0.0427], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 23:33:00,762 INFO [train.py:968] (1/2) Epoch 23, batch 21450, giga_loss[loss=0.2166, simple_loss=0.2982, pruned_loss=0.06752, over 28627.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3409, pruned_loss=0.09338, over 5710342.50 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3448, pruned_loss=0.09041, over 5731284.47 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3407, pruned_loss=0.09348, over 5711312.89 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:33:13,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1024858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:33:40,748 INFO [train.py:968] (1/2) Epoch 23, batch 21500, giga_loss[loss=0.2717, simple_loss=0.3462, pruned_loss=0.09855, over 29006.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3401, pruned_loss=0.093, over 5716402.31 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3451, pruned_loss=0.09056, over 5731537.37 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3396, pruned_loss=0.093, over 5716496.29 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:33:50,804 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:33:52,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:33:56,137 INFO [optim.py:369] (1/2) 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 23:34:04,064 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1024919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:34:05,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1024922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:34:15,501 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:34:20,781 INFO [train.py:968] (1/2) Epoch 23, batch 21550, giga_loss[loss=0.2812, simple_loss=0.3526, pruned_loss=0.1049, over 28933.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3398, pruned_loss=0.09325, over 5722476.55 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3449, pruned_loss=0.09058, over 5732032.54 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3394, pruned_loss=0.09328, over 5721921.72 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:34:30,268 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1024951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:05,591 INFO [train.py:968] (1/2) Epoch 23, batch 21600, giga_loss[loss=0.2956, simple_loss=0.3646, pruned_loss=0.1132, over 28573.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3389, pruned_loss=0.09357, over 5717817.04 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.345, pruned_loss=0.09066, over 5735218.45 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3385, pruned_loss=0.09358, over 5714408.25 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:35:14,644 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:16,626 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:35:23,690 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:42,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:35:49,138 INFO [train.py:968] (1/2) Epoch 23, batch 21650, giga_loss[loss=0.2338, simple_loss=0.3162, pruned_loss=0.07566, over 28958.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3379, pruned_loss=0.09383, over 5713256.65 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3453, pruned_loss=0.09087, over 5732148.95 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3372, pruned_loss=0.0937, over 5713123.05 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:36:11,122 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:36:29,574 INFO [train.py:968] (1/2) Epoch 23, batch 21700, giga_loss[loss=0.2375, simple_loss=0.3134, pruned_loss=0.08077, over 28817.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.336, pruned_loss=0.09337, over 5713639.62 frames. ], libri_tot_loss[loss=0.2635, simple_loss=0.3452, pruned_loss=0.0909, over 5737122.98 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3353, pruned_loss=0.0933, over 5708701.45 frames. ], batch size: 112, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:36:44,880 INFO [optim.py:369] (1/2) 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 23:36:50,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7745, 2.7489, 1.7605, 1.0832], device='cuda:1'), covar=tensor([0.9381, 0.3937, 0.4523, 0.7245], device='cuda:1'), in_proj_covar=tensor([0.1764, 0.1659, 0.1601, 0.1436], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 23:37:09,615 INFO [train.py:968] (1/2) Epoch 23, batch 21750, giga_loss[loss=0.2232, simple_loss=0.3025, pruned_loss=0.07193, over 29013.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3344, pruned_loss=0.09268, over 5710882.31 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3454, pruned_loss=0.09096, over 5732996.99 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3334, pruned_loss=0.09262, over 5709803.16 frames. ], batch size: 66, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:37:12,784 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.33 vs. limit=5.0
+2023-03-11 23:37:26,338 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-11 23:37:34,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3645, 3.6387, 1.4381, 1.6595], device='cuda:1'), covar=tensor([0.0926, 0.0353, 0.0873, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0553, 0.0390, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 23:37:52,063 INFO [train.py:968] (1/2) Epoch 23, batch 21800, giga_loss[loss=0.2559, simple_loss=0.3276, pruned_loss=0.09211, over 28844.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3345, pruned_loss=0.09318, over 5706703.86 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3458, pruned_loss=0.09146, over 5733679.06 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3331, pruned_loss=0.09274, over 5704498.75 frames. ], batch size: 112, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:38:08,638 INFO [optim.py:369] (1/2) 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 23:38:22,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5922, 1.6686, 1.8014, 1.3860], device='cuda:1'), covar=tensor([0.1846, 0.2555, 0.1557, 0.1902], device='cuda:1'), in_proj_covar=tensor([0.0910, 0.0704, 0.0957, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 23:38:34,730 INFO [train.py:968] (1/2) Epoch 23, batch 21850, giga_loss[loss=0.2729, simple_loss=0.3503, pruned_loss=0.09773, over 28821.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3362, pruned_loss=0.09364, over 5711078.25 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.346, pruned_loss=0.09175, over 5735660.56 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3347, pruned_loss=0.09306, over 5706853.96 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:38:57,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:39:07,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6823, 1.8113, 1.9009, 1.4753], device='cuda:1'), covar=tensor([0.2061, 0.2462, 0.1699, 0.1970], device='cuda:1'), in_proj_covar=tensor([0.0910, 0.0704, 0.0957, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 23:39:11,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5432, 1.5726, 1.7607, 1.3961], device='cuda:1'), covar=tensor([0.1539, 0.2108, 0.1289, 0.1595], device='cuda:1'), in_proj_covar=tensor([0.0910, 0.0704, 0.0957, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-11 23:39:21,421 INFO [train.py:968] (1/2) Epoch 23, batch 21900, giga_loss[loss=0.3192, simple_loss=0.3961, pruned_loss=0.1212, over 28485.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3383, pruned_loss=0.09419, over 5712521.19 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3458, pruned_loss=0.09171, over 5737529.40 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3372, pruned_loss=0.09379, over 5707150.72 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:39:39,429 INFO [optim.py:369] (1/2) 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 23:39:49,709 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.52 vs. limit=5.0
+2023-03-11 23:40:04,872 INFO [train.py:968] (1/2) Epoch 23, batch 21950, giga_loss[loss=0.258, simple_loss=0.3383, pruned_loss=0.08883, over 28852.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3397, pruned_loss=0.09378, over 5712037.10 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3458, pruned_loss=0.09175, over 5739977.46 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3387, pruned_loss=0.09345, over 5705158.39 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:40:11,444 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025349.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 23:40:43,464 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 22000, giga_loss[loss=0.2416, simple_loss=0.3272, pruned_loss=0.078, over 28945.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3404, pruned_loss=0.09357, over 5700189.98 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3467, pruned_loss=0.09239, over 5732407.28 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3388, pruned_loss=0.09282, over 5700062.65 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:41:08,465 INFO [optim.py:369] (1/2) 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 23:41:33,668 INFO [train.py:968] (1/2) Epoch 23, batch 22050, giga_loss[loss=0.2574, simple_loss=0.341, pruned_loss=0.08686, over 28690.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3398, pruned_loss=0.09316, over 5694819.65 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3469, pruned_loss=0.09262, over 5733764.16 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09235, over 5692919.88 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:41:36,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1025444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:42:14,207 INFO [train.py:968] (1/2) Epoch 23, batch 22100, giga_loss[loss=0.2545, simple_loss=0.335, pruned_loss=0.08703, over 28832.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3404, pruned_loss=0.09366, over 5706098.86 frames. ], libri_tot_loss[loss=0.2666, simple_loss=0.3472, pruned_loss=0.09298, over 5735608.33 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3386, pruned_loss=0.09271, over 5702031.41 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:42:32,838 INFO [optim.py:369] (1/2) 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 23:42:46,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:42:48,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:42:54,591 INFO [train.py:968] (1/2) Epoch 23, batch 22150, giga_loss[loss=0.2814, simple_loss=0.3533, pruned_loss=0.1048, over 28879.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09381, over 5702697.56 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3476, pruned_loss=0.09348, over 5734001.93 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3386, pruned_loss=0.09258, over 5699140.84 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:43:12,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:43:32,899 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:43:35,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:43:35,352 INFO [train.py:968] (1/2) Epoch 23, batch 22200, giga_loss[loss=0.2708, simple_loss=0.3554, pruned_loss=0.09308, over 28690.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3441, pruned_loss=0.09627, over 5709188.77 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3478, pruned_loss=0.09395, over 5739451.34 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.342, pruned_loss=0.09488, over 5700613.63 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:43:51,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3568, 3.0254, 1.3921, 1.3813], device='cuda:1'), covar=tensor([0.0976, 0.0335, 0.0964, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0553, 0.0390, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 23:43:54,809 INFO [optim.py:369] (1/2) 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 23:44:00,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:44:15,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1025639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:44:17,474 INFO [train.py:968] (1/2) Epoch 23, batch 22250, giga_loss[loss=0.2651, simple_loss=0.339, pruned_loss=0.09563, over 28650.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3481, pruned_loss=0.09875, over 5710167.17 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3483, pruned_loss=0.09431, over 5741791.70 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.346, pruned_loss=0.09739, over 5700933.42 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:45:01,839 INFO [train.py:968] (1/2) Epoch 23, batch 22300, giga_loss[loss=0.2571, simple_loss=0.3288, pruned_loss=0.0927, over 28843.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.35, pruned_loss=0.09956, over 5710566.91 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3488, pruned_loss=0.09463, over 5743737.32 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.09828, over 5701222.19 frames. ], batch size: 112, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:45:18,714 INFO [optim.py:369] (1/2) 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 23:45:27,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1025724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 23:45:41,210 INFO [train.py:968] (1/2) Epoch 23, batch 22350, libri_loss[loss=0.2836, simple_loss=0.3557, pruned_loss=0.1057, over 29545.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.352, pruned_loss=0.1003, over 5701959.42 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3495, pruned_loss=0.09522, over 5730423.43 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3497, pruned_loss=0.09897, over 5706863.61 frames. ], batch size: 80, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:45:50,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5184, 1.7767, 1.5262, 1.6124], device='cuda:1'), covar=tensor([0.0615, 0.0270, 0.0281, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 23:46:15,795 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:46:19,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:46:23,455 INFO [train.py:968] (1/2) Epoch 23, batch 22400, giga_loss[loss=0.2924, simple_loss=0.3683, pruned_loss=0.1082, over 28826.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3531, pruned_loss=0.1013, over 5700575.38 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.35, pruned_loss=0.0956, over 5724803.85 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.351, pruned_loss=0.1, over 5708824.89 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:46:25,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4097, 1.6421, 1.3592, 1.2786], device='cuda:1'), covar=tensor([0.2374, 0.2561, 0.2746, 0.2338], device='cuda:1'), in_proj_covar=tensor([0.1521, 0.1100, 0.1343, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-11 23:46:43,349 INFO [optim.py:369] (1/2) 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 23:46:45,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:08,131 INFO [train.py:968] (1/2) Epoch 23, batch 22450, giga_loss[loss=0.2511, simple_loss=0.3309, pruned_loss=0.08567, over 28983.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3528, pruned_loss=0.1014, over 5692046.49 frames. ], libri_tot_loss[loss=0.2713, simple_loss=0.3504, pruned_loss=0.09604, over 5719831.19 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3507, pruned_loss=0.1001, over 5702658.31 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:47:13,654 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:26,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:29,552 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1025867.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 23:47:31,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1025868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:47:32,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1025870.0, num_to_drop=1, layers_to_drop={0}
+2023-03-11 23:47:48,595 INFO [train.py:968] (1/2) Epoch 23, batch 22500, giga_loss[loss=0.259, simple_loss=0.3363, pruned_loss=0.09082, over 27940.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3496, pruned_loss=0.09944, over 5692660.10 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3512, pruned_loss=0.09685, over 5707634.51 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3473, pruned_loss=0.09785, over 5710312.53 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:47:56,763 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1025899.0, num_to_drop=1, layers_to_drop={1}
+2023-03-11 23:48:09,699 INFO [optim.py:369] (1/2) 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 23:48:22,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3812, 1.4757, 1.2930, 1.6009], device='cuda:1'), covar=tensor([0.0763, 0.0335, 0.0350, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 23:48:22,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3529, 2.0285, 1.5495, 0.6027], device='cuda:1'), covar=tensor([0.5910, 0.2875, 0.5026, 0.7098], device='cuda:1'), in_proj_covar=tensor([0.1765, 0.1661, 0.1605, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-11 23:48:30,918 INFO [train.py:968] (1/2) Epoch 23, batch 22550, giga_loss[loss=0.2592, simple_loss=0.3296, pruned_loss=0.09437, over 28847.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3473, pruned_loss=0.09874, over 5690687.62 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3514, pruned_loss=0.09707, over 5709477.52 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3452, pruned_loss=0.09733, over 5702390.00 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:48:37,323 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.12 vs. limit=5.0
+2023-03-11 23:49:11,816 INFO [train.py:968] (1/2) Epoch 23, batch 22600, giga_loss[loss=0.2525, simple_loss=0.3261, pruned_loss=0.08948, over 28724.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3444, pruned_loss=0.09714, over 5689358.05 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3515, pruned_loss=0.09723, over 5701117.77 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3425, pruned_loss=0.0959, over 5705280.41 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:49:23,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5042, 3.4076, 1.4729, 1.7119], device='cuda:1'), covar=tensor([0.0905, 0.0272, 0.0962, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0554, 0.0390, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-11 23:49:31,007 INFO [optim.py:369] (1/2) 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 23:49:56,982 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:49:57,538 INFO [train.py:968] (1/2) Epoch 23, batch 22650, giga_loss[loss=0.2076, simple_loss=0.2921, pruned_loss=0.06161, over 28296.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3446, pruned_loss=0.0962, over 5687460.14 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3515, pruned_loss=0.09725, over 5702281.18 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3431, pruned_loss=0.0952, over 5698775.80 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:50:23,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6936, 3.5453, 3.3291, 1.8903], device='cuda:1'), covar=tensor([0.0713, 0.0825, 0.0776, 0.2483], device='cuda:1'), in_proj_covar=tensor([0.1225, 0.1131, 0.0959, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-11 23:50:40,440 INFO [train.py:968] (1/2) Epoch 23, batch 22700, giga_loss[loss=0.3011, simple_loss=0.3869, pruned_loss=0.1077, over 28932.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3471, pruned_loss=0.09642, over 5692566.01 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.352, pruned_loss=0.09758, over 5705081.62 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3454, pruned_loss=0.09535, over 5699009.17 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:50:58,907 INFO [optim.py:369] (1/2) 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 23:51:19,649 INFO [train.py:968] (1/2) Epoch 23, batch 22750, giga_loss[loss=0.234, simple_loss=0.3101, pruned_loss=0.07895, over 29122.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3469, pruned_loss=0.09731, over 5687496.65 frames. ], libri_tot_loss[loss=0.2744, simple_loss=0.3523, pruned_loss=0.09824, over 5699412.93 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3452, pruned_loss=0.09584, over 5696771.24 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:51:28,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6615, 1.1942, 5.1443, 3.7116], device='cuda:1'), covar=tensor([0.1649, 0.2876, 0.0352, 0.0787], device='cuda:1'), in_proj_covar=tensor([0.0761, 0.0649, 0.0959, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 23:51:40,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4513, 1.7132, 1.3596, 1.6080], device='cuda:1'), covar=tensor([0.0725, 0.0315, 0.0337, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0117, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0099, 0.0070, 0.0063, 0.0109], device='cuda:1')
+2023-03-11 23:52:02,264 INFO [train.py:968] (1/2) Epoch 23, batch 22800, libri_loss[loss=0.2479, simple_loss=0.3245, pruned_loss=0.08566, over 29529.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3454, pruned_loss=0.09811, over 5692618.02 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3519, pruned_loss=0.09826, over 5703042.27 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3442, pruned_loss=0.09689, over 5696503.82 frames. ], batch size: 76, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:52:16,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2871, 1.2610, 3.9263, 3.2467], device='cuda:1'), covar=tensor([0.1609, 0.2728, 0.0430, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0762, 0.0650, 0.0961, 0.0916], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-11 23:52:20,841 INFO [optim.py:369] (1/2) 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 23:52:28,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:52:39,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:52:42,665 INFO [train.py:968] (1/2) Epoch 23, batch 22850, giga_loss[loss=0.2458, simple_loss=0.3108, pruned_loss=0.09037, over 28471.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3432, pruned_loss=0.09775, over 5702078.95 frames. ], libri_tot_loss[loss=0.2741, simple_loss=0.3517, pruned_loss=0.09819, over 5706477.38 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3422, pruned_loss=0.09685, over 5702002.33 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:52:45,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:53:23,868 INFO [train.py:968] (1/2) Epoch 23, batch 22900, giga_loss[loss=0.2561, simple_loss=0.3199, pruned_loss=0.09617, over 28777.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3409, pruned_loss=0.09732, over 5708036.03 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3519, pruned_loss=0.09861, over 5710465.58 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3397, pruned_loss=0.09622, over 5704391.23 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:53:45,816 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 22950, giga_loss[loss=0.2774, simple_loss=0.3522, pruned_loss=0.1013, over 28727.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3423, pruned_loss=0.09861, over 5715240.37 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3529, pruned_loss=0.09956, over 5716266.58 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3401, pruned_loss=0.09683, over 5706943.67 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:54:28,396 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:30,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:39,106 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:41,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:43,131 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:46,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:54:47,511 INFO [train.py:968] (1/2) Epoch 23, batch 23000, giga_loss[loss=0.2066, simple_loss=0.2795, pruned_loss=0.06688, over 28497.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3394, pruned_loss=0.09717, over 5719928.64 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3533, pruned_loss=0.09998, over 5718282.84 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3372, pruned_loss=0.09535, over 5711510.13 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:54:53,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:03,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:05,547 INFO [optim.py:369] (1/2) 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 23:55:06,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:08,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:23,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:55:24,582 INFO [train.py:968] (1/2) Epoch 23, batch 23050, giga_loss[loss=0.2083, simple_loss=0.2918, pruned_loss=0.06244, over 28934.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3359, pruned_loss=0.09573, over 5721282.50 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.353, pruned_loss=0.1001, over 5724398.57 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3339, pruned_loss=0.09406, over 5709039.22 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:56:07,259 INFO [train.py:968] (1/2) Epoch 23, batch 23100, giga_loss[loss=0.2312, simple_loss=0.3029, pruned_loss=0.07974, over 29041.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3331, pruned_loss=0.09433, over 5720315.72 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3536, pruned_loss=0.1007, over 5728870.33 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3304, pruned_loss=0.09231, over 5706311.99 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:56:15,666 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-11 23:56:18,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-11 23:56:27,633 INFO [optim.py:369] (1/2) 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 23:56:48,161 INFO [train.py:968] (1/2) Epoch 23, batch 23150, giga_loss[loss=0.2223, simple_loss=0.2954, pruned_loss=0.07459, over 28495.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3334, pruned_loss=0.09406, over 5725088.71 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3535, pruned_loss=0.1008, over 5731758.64 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3309, pruned_loss=0.09226, over 5711183.72 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:57:04,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:57:07,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:57:12,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-11 23:57:32,431 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-11 23:57:32,771 INFO [train.py:968] (1/2) Epoch 23, batch 23200, giga_loss[loss=0.2933, simple_loss=0.3656, pruned_loss=0.1105, over 27968.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3363, pruned_loss=0.09526, over 5713239.56 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3536, pruned_loss=0.1008, over 5725850.24 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3339, pruned_loss=0.09365, over 5707293.55 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:57:54,257 INFO [optim.py:369] (1/2) 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 23:58:02,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-11 23:58:14,592 INFO [train.py:968] (1/2) Epoch 23, batch 23250, giga_loss[loss=0.2635, simple_loss=0.3488, pruned_loss=0.08912, over 28675.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3399, pruned_loss=0.09676, over 5711622.18 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3539, pruned_loss=0.1013, over 5726700.23 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3373, pruned_loss=0.09496, over 5705572.17 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 8.0
+2023-03-11 23:58:56,150 INFO [train.py:968] (1/2) Epoch 23, batch 23300, giga_loss[loss=0.3176, simple_loss=0.3848, pruned_loss=0.1252, over 28750.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3439, pruned_loss=0.09843, over 5711419.13 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3542, pruned_loss=0.1015, over 5731245.15 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3412, pruned_loss=0.09664, over 5702114.47 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-11 23:59:17,427 INFO [optim.py:369] (1/2) 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 23:59:40,444 INFO [train.py:968] (1/2) Epoch 23, batch 23350, giga_loss[loss=0.2638, simple_loss=0.3442, pruned_loss=0.09169, over 28910.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3458, pruned_loss=0.09885, over 5706046.22 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3544, pruned_loss=0.1018, over 5731403.23 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3433, pruned_loss=0.09713, over 5698160.41 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:00:25,454 INFO [train.py:968] (1/2) Epoch 23, batch 23400, giga_loss[loss=0.4673, simple_loss=0.4761, pruned_loss=0.2292, over 26692.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3481, pruned_loss=0.1005, over 5705186.91 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3548, pruned_loss=0.1021, over 5734219.47 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3455, pruned_loss=0.09877, over 5695889.27 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:00:49,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1026814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:00:51,080 INFO [optim.py:369] (1/2) 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 00:01:13,941 INFO [train.py:968] (1/2) Epoch 23, batch 23450, libri_loss[loss=0.2837, simple_loss=0.3609, pruned_loss=0.1032, over 29759.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3532, pruned_loss=0.1052, over 5700161.90 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3546, pruned_loss=0.102, over 5739422.97 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3513, pruned_loss=0.104, over 5686637.40 frames. ], batch size: 87, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:01:30,360 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:02:00,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8778, 1.7948, 2.0506, 1.6119], device='cuda:1'), covar=tensor([0.1807, 0.2461, 0.1409, 0.1698], device='cuda:1'), in_proj_covar=tensor([0.0904, 0.0700, 0.0951, 0.0850], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 00:02:01,025 INFO [train.py:968] (1/2) Epoch 23, batch 23500, giga_loss[loss=0.264, simple_loss=0.3446, pruned_loss=0.09167, over 29086.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3576, pruned_loss=0.1086, over 5693752.25 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3542, pruned_loss=0.1021, over 5734797.16 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3565, pruned_loss=0.1077, over 5685354.22 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:02:10,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5442, 1.8982, 1.4969, 1.4923], device='cuda:1'), covar=tensor([0.2702, 0.2718, 0.3076, 0.2443], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1101, 0.1345, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 00:02:22,999 INFO [optim.py:369] (1/2) 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 00:02:47,707 INFO [train.py:968] (1/2) Epoch 23, batch 23550, giga_loss[loss=0.3086, simple_loss=0.3759, pruned_loss=0.1207, over 28531.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.363, pruned_loss=0.1124, over 5691009.39 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3539, pruned_loss=0.1021, over 5736094.75 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3626, pruned_loss=0.1119, over 5681533.77 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:03:04,557 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1026957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:06,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1026960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:16,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1026967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:40,130 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1026989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:03:42,040 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 00:03:42,164 INFO [train.py:968] (1/2) Epoch 23, batch 23600, giga_loss[loss=0.319, simple_loss=0.3863, pruned_loss=0.1258, over 28693.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3705, pruned_loss=0.1191, over 5673663.88 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3543, pruned_loss=0.1024, over 5729213.53 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.37, pruned_loss=0.1187, over 5671124.53 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:03:52,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5685, 1.7548, 1.4470, 1.7314], device='cuda:1'), covar=tensor([0.2562, 0.2513, 0.2805, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1524, 0.1101, 0.1346, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 00:04:07,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2768, 2.6542, 1.4165, 1.4218], device='cuda:1'), covar=tensor([0.0931, 0.0410, 0.0847, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0557, 0.0391, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 00:04:08,871 INFO [optim.py:369] (1/2) 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 00:04:19,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2516, 1.4832, 1.5434, 1.1293], device='cuda:1'), covar=tensor([0.1610, 0.2513, 0.1328, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0703, 0.0953, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 00:04:32,734 INFO [train.py:968] (1/2) Epoch 23, batch 23650, giga_loss[loss=0.2992, simple_loss=0.3798, pruned_loss=0.1093, over 29059.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3774, pruned_loss=0.1249, over 5666174.18 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3544, pruned_loss=0.1025, over 5730735.54 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3772, pruned_loss=0.1247, over 5661965.95 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:04:58,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0795, 1.2707, 3.3042, 2.9451], device='cuda:1'), covar=tensor([0.1659, 0.2659, 0.0511, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0651, 0.0963, 0.0919], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 00:05:18,466 INFO [train.py:968] (1/2) Epoch 23, batch 23700, giga_loss[loss=0.2997, simple_loss=0.374, pruned_loss=0.1127, over 28676.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3799, pruned_loss=0.1266, over 5666877.94 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3542, pruned_loss=0.1026, over 5725412.16 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3809, pruned_loss=0.1273, over 5665412.48 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:05:46,096 INFO [optim.py:369] (1/2) 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 00:06:12,028 INFO [train.py:968] (1/2) Epoch 23, batch 23750, giga_loss[loss=0.4048, simple_loss=0.4289, pruned_loss=0.1903, over 27612.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3834, pruned_loss=0.1304, over 5659231.57 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3545, pruned_loss=0.1029, over 5723860.94 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3842, pruned_loss=0.131, over 5658457.68 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:06:53,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6260, 1.8850, 1.5437, 1.6156], device='cuda:1'), covar=tensor([0.2316, 0.2284, 0.2521, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.1526, 0.1104, 0.1347, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 00:07:04,636 INFO [train.py:968] (1/2) Epoch 23, batch 23800, giga_loss[loss=0.3192, simple_loss=0.3818, pruned_loss=0.1283, over 28834.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3848, pruned_loss=0.1327, over 5634675.70 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3549, pruned_loss=0.1032, over 5716142.78 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3856, pruned_loss=0.1335, over 5639669.61 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:07:27,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2992, 1.4644, 1.6246, 1.2957], device='cuda:1'), covar=tensor([0.1779, 0.1641, 0.2008, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0756, 0.0724, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 00:07:31,895 INFO [optim.py:369] (1/2) 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 00:07:50,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:07:56,644 INFO [train.py:968] (1/2) Epoch 23, batch 23850, giga_loss[loss=0.4686, simple_loss=0.4703, pruned_loss=0.2335, over 26753.00 frames. ], tot_loss[loss=0.3321, simple_loss=0.3894, pruned_loss=0.1374, over 5635422.82 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3553, pruned_loss=0.1037, over 5720251.14 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3905, pruned_loss=0.1383, over 5633846.19 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:08:59,216 INFO [train.py:968] (1/2) Epoch 23, batch 23900, giga_loss[loss=0.4051, simple_loss=0.4236, pruned_loss=0.1933, over 23646.00 frames. ], tot_loss[loss=0.3374, simple_loss=0.3924, pruned_loss=0.1412, over 5603681.90 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3551, pruned_loss=0.1036, over 5714772.74 frames. ], giga_tot_loss[loss=0.3396, simple_loss=0.394, pruned_loss=0.1425, over 5606042.58 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:09:27,394 INFO [optim.py:369] (1/2) 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 00:09:48,334 INFO [train.py:968] (1/2) Epoch 23, batch 23950, giga_loss[loss=0.3662, simple_loss=0.408, pruned_loss=0.1622, over 27594.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3899, pruned_loss=0.14, over 5617620.83 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3557, pruned_loss=0.1043, over 5719618.88 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3918, pruned_loss=0.1416, over 5611535.42 frames. ], batch size: 474, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:09:49,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:09:55,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3953, 1.5581, 1.5920, 1.2850], device='cuda:1'), covar=tensor([0.2575, 0.2392, 0.1666, 0.2372], device='cuda:1'), in_proj_covar=tensor([0.1989, 0.1943, 0.1862, 0.1995], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 00:10:00,946 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:10:22,785 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1027376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:10:25,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:10:35,883 INFO [train.py:968] (1/2) Epoch 23, batch 24000, giga_loss[loss=0.3319, simple_loss=0.3873, pruned_loss=0.1382, over 28957.00 frames. ], tot_loss[loss=0.333, simple_loss=0.3882, pruned_loss=0.1388, over 5630040.45 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3557, pruned_loss=0.1043, over 5713306.19 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3902, pruned_loss=0.1406, over 5629165.77 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:10:35,884 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 00:10:44,786 INFO [train.py:1012] (1/2) Epoch 23, validation: loss=0.2055, simple_loss=0.314, pruned_loss=0.0485, over 944034.00 frames. 
+2023-03-12 00:10:44,786 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 00:10:58,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8976, 3.7411, 3.5535, 2.0041], device='cuda:1'), covar=tensor([0.0736, 0.0804, 0.0839, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.1247, 0.1151, 0.0978, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 00:11:00,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:11:04,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:11:12,753 INFO [optim.py:369] (1/2) 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 00:11:13,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8106, 1.1695, 1.2878, 0.9550], device='cuda:1'), covar=tensor([0.2201, 0.1452, 0.2259, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0758, 0.0725, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 00:11:31,062 INFO [train.py:968] (1/2) Epoch 23, batch 24050, giga_loss[loss=0.3114, simple_loss=0.3805, pruned_loss=0.1212, over 28999.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3874, pruned_loss=0.1372, over 5626606.95 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3561, pruned_loss=0.1047, over 5715778.66 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3898, pruned_loss=0.1393, over 5620959.65 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:12:21,590 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1027485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:12:23,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:12:25,899 INFO [train.py:968] (1/2) Epoch 23, batch 24100, giga_loss[loss=0.3487, simple_loss=0.4069, pruned_loss=0.1453, over 28794.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3876, pruned_loss=0.1367, over 5610953.41 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3562, pruned_loss=0.105, over 5708734.95 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3898, pruned_loss=0.1387, over 5610977.85 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:12:52,854 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:12:53,275 INFO [optim.py:369] (1/2) 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 00:13:10,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:13:18,096 INFO [train.py:968] (1/2) Epoch 23, batch 24150, giga_loss[loss=0.2789, simple_loss=0.3492, pruned_loss=0.1043, over 28896.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3881, pruned_loss=0.1366, over 5615181.60 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3564, pruned_loss=0.1052, over 5708244.37 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3907, pruned_loss=0.1389, over 5613133.29 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:14:15,360 INFO [train.py:968] (1/2) Epoch 23, batch 24200, giga_loss[loss=0.2947, simple_loss=0.3608, pruned_loss=0.1143, over 28648.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3858, pruned_loss=0.1345, over 5614302.59 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3563, pruned_loss=0.1054, over 5710372.97 frames. ], giga_tot_loss[loss=0.3305, simple_loss=0.3881, pruned_loss=0.1365, over 5609995.96 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:14:43,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2914, 1.8893, 1.3279, 0.5419], device='cuda:1'), covar=tensor([0.4943, 0.2607, 0.3927, 0.6122], device='cuda:1'), in_proj_covar=tensor([0.1768, 0.1669, 0.1607, 0.1440], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 00:14:43,398 INFO [optim.py:369] (1/2) 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 00:15:09,636 INFO [train.py:968] (1/2) Epoch 23, batch 24250, giga_loss[loss=0.2772, simple_loss=0.3601, pruned_loss=0.09721, over 28795.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.383, pruned_loss=0.1309, over 5628546.19 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3561, pruned_loss=0.1052, over 5711377.10 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3851, pruned_loss=0.1326, over 5623760.89 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:15:44,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1441, 2.2898, 1.7786, 1.7465], device='cuda:1'), covar=tensor([0.1003, 0.0712, 0.1025, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0450, 0.0524, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 00:16:00,231 INFO [train.py:968] (1/2) Epoch 23, batch 24300, giga_loss[loss=0.2504, simple_loss=0.3229, pruned_loss=0.08898, over 28490.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3788, pruned_loss=0.1274, over 5628596.41 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3565, pruned_loss=0.1058, over 5715992.65 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3808, pruned_loss=0.1289, over 5618673.11 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:16:03,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4630, 1.5990, 1.7336, 1.2967], device='cuda:1'), covar=tensor([0.1607, 0.2547, 0.1383, 0.1687], device='cuda:1'), in_proj_covar=tensor([0.0900, 0.0699, 0.0947, 0.0846], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 00:16:24,512 INFO [optim.py:369] (1/2) 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 00:16:35,575 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 24350, giga_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 28978.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3762, pruned_loss=0.1254, over 5628206.70 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3567, pruned_loss=0.1062, over 5706560.51 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3779, pruned_loss=0.1266, over 5626654.17 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:16:52,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4651, 1.7586, 1.7618, 1.3975], device='cuda:1'), covar=tensor([0.2739, 0.2231, 0.2367, 0.2620], device='cuda:1'), in_proj_covar=tensor([0.1989, 0.1942, 0.1860, 0.2000], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 00:17:12,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2125, 4.0323, 3.8563, 2.0101], device='cuda:1'), covar=tensor([0.0639, 0.0792, 0.0788, 0.1913], device='cuda:1'), in_proj_covar=tensor([0.1254, 0.1160, 0.0983, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 00:17:20,127 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:17:34,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:17:37,455 INFO [train.py:968] (1/2) Epoch 23, batch 24400, giga_loss[loss=0.323, simple_loss=0.388, pruned_loss=0.129, over 28257.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1247, over 5625566.47 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3564, pruned_loss=0.1061, over 5706294.65 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3768, pruned_loss=0.1262, over 5622466.75 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:17:47,066 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-12 00:18:03,547 INFO [optim.py:369] (1/2) 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 00:18:27,785 INFO [train.py:968] (1/2) Epoch 23, batch 24450, libri_loss[loss=0.2312, simple_loss=0.3042, pruned_loss=0.07907, over 29323.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 5641706.07 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3557, pruned_loss=0.1059, over 5714772.38 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3766, pruned_loss=0.1262, over 5628135.04 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:18:48,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1936, 1.5753, 1.1756, 0.4703], device='cuda:1'), covar=tensor([0.3268, 0.1885, 0.2801, 0.5788], device='cuda:1'), in_proj_covar=tensor([0.1764, 0.1665, 0.1603, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 00:18:48,615 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-12 00:18:52,981 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:02,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1027872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:06,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:23,326 INFO [train.py:968] (1/2) Epoch 23, batch 24500, giga_loss[loss=0.2809, simple_loss=0.3546, pruned_loss=0.1036, over 28645.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3724, pruned_loss=0.1223, over 5650423.76 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3557, pruned_loss=0.1059, over 5715743.52 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3751, pruned_loss=0.1241, over 5638646.29 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:19:36,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:41,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1027909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:19:52,415 INFO [optim.py:369] (1/2) 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 00:19:58,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1027925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:20:06,048 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1027933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:20:18,529 INFO [train.py:968] (1/2) Epoch 23, batch 24550, giga_loss[loss=0.2797, simple_loss=0.359, pruned_loss=0.1002, over 28412.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3717, pruned_loss=0.1199, over 5654769.47 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3559, pruned_loss=0.1062, over 5716494.09 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3737, pruned_loss=0.1213, over 5644049.81 frames. ], batch size: 65, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:20:36,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9522, 4.8306, 2.0050, 2.1371], device='cuda:1'), covar=tensor([0.0896, 0.0402, 0.0844, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0559, 0.0392, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 00:20:41,119 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1027962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:21:09,138 INFO [train.py:968] (1/2) Epoch 23, batch 24600, giga_loss[loss=0.2571, simple_loss=0.3352, pruned_loss=0.08946, over 28167.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3731, pruned_loss=0.119, over 5664093.38 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3559, pruned_loss=0.1064, over 5718718.16 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3751, pruned_loss=0.1202, over 5652163.57 frames. ], batch size: 77, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:21:40,489 INFO [optim.py:369] (1/2) 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 00:22:03,289 INFO [train.py:968] (1/2) Epoch 23, batch 24650, giga_loss[loss=0.3433, simple_loss=0.3988, pruned_loss=0.1439, over 27860.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.373, pruned_loss=0.1193, over 5661288.10 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3555, pruned_loss=0.1063, over 5721468.89 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3752, pruned_loss=0.1206, over 5648826.24 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:22:15,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:22:17,838 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:22:45,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:22:50,243 INFO [train.py:968] (1/2) Epoch 23, batch 24700, giga_loss[loss=0.2696, simple_loss=0.3462, pruned_loss=0.09651, over 28920.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.373, pruned_loss=0.1192, over 5680025.26 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3561, pruned_loss=0.1067, over 5723994.18 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3747, pruned_loss=0.1202, over 5666230.13 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:23:16,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2381, 2.2840, 2.1066, 2.2029], device='cuda:1'), covar=tensor([0.2079, 0.2649, 0.2466, 0.2322], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0756, 0.0722, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 00:23:19,880 INFO [optim.py:369] (1/2) 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 00:23:37,917 INFO [train.py:968] (1/2) Epoch 23, batch 24750, giga_loss[loss=0.3171, simple_loss=0.3781, pruned_loss=0.1281, over 28879.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3723, pruned_loss=0.1194, over 5687975.22 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3565, pruned_loss=0.1069, over 5726139.34 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3736, pruned_loss=0.1202, over 5674443.28 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:23:44,444 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:24:15,279 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:24:26,038 INFO [train.py:968] (1/2) Epoch 23, batch 24800, giga_loss[loss=0.2477, simple_loss=0.3254, pruned_loss=0.08503, over 28894.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3708, pruned_loss=0.1201, over 5683832.80 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3562, pruned_loss=0.107, over 5729808.62 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3725, pruned_loss=0.121, over 5668725.04 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:24:29,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2132, 1.5624, 1.5016, 1.0698], device='cuda:1'), covar=tensor([0.1591, 0.2662, 0.1404, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0704, 0.0952, 0.0850], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 00:24:53,071 INFO [optim.py:369] (1/2) 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 00:25:06,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3201, 3.4322, 1.4878, 1.3680], device='cuda:1'), covar=tensor([0.1077, 0.0379, 0.0969, 0.1500], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0560, 0.0393, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 00:25:09,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:25:11,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3450, 1.7949, 1.4845, 1.4935], device='cuda:1'), covar=tensor([0.0741, 0.0362, 0.0323, 0.0785], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 00:25:12,969 INFO [train.py:968] (1/2) Epoch 23, batch 24850, giga_loss[loss=0.2882, simple_loss=0.3715, pruned_loss=0.1025, over 28951.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3696, pruned_loss=0.119, over 5688649.76 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3565, pruned_loss=0.1073, over 5732916.29 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.371, pruned_loss=0.1197, over 5672951.30 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:25:58,241 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 24900, libri_loss[loss=0.3021, simple_loss=0.3743, pruned_loss=0.1149, over 28655.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3693, pruned_loss=0.1175, over 5688412.28 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3563, pruned_loss=0.1074, over 5731437.67 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3708, pruned_loss=0.1183, over 5676155.52 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:26:01,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:26:08,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:26:25,985 INFO [optim.py:369] (1/2) 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 00:26:29,315 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:26:46,477 INFO [train.py:968] (1/2) Epoch 23, batch 24950, giga_loss[loss=0.3023, simple_loss=0.372, pruned_loss=0.1163, over 28425.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3691, pruned_loss=0.1172, over 5677717.99 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3564, pruned_loss=0.1076, over 5726986.04 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3706, pruned_loss=0.1178, over 5670824.16 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:26:58,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:26:59,455 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-12 00:27:25,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:27:29,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:27:38,891 INFO [train.py:968] (1/2) Epoch 23, batch 25000, giga_loss[loss=0.2858, simple_loss=0.3575, pruned_loss=0.1071, over 28709.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3682, pruned_loss=0.1166, over 5681588.54 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3564, pruned_loss=0.1076, over 5728574.01 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3695, pruned_loss=0.1171, over 5674287.96 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:27:56,118 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:28:01,564 INFO [optim.py:369] (1/2) 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 00:28:26,867 INFO [train.py:968] (1/2) Epoch 23, batch 25050, libri_loss[loss=0.2818, simple_loss=0.3477, pruned_loss=0.1079, over 29668.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3675, pruned_loss=0.117, over 5687214.78 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3559, pruned_loss=0.1073, over 5734014.02 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3693, pruned_loss=0.1179, over 5675398.50 frames. ], batch size: 73, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:28:28,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:28:31,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:28:48,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6024, 1.7710, 1.8178, 1.5869], device='cuda:1'), covar=tensor([0.1958, 0.2060, 0.2266, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0759, 0.0723, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 00:29:00,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:29:18,637 INFO [train.py:968] (1/2) Epoch 23, batch 25100, giga_loss[loss=0.2683, simple_loss=0.3339, pruned_loss=0.1013, over 28199.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3674, pruned_loss=0.1181, over 5671580.25 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3561, pruned_loss=0.1075, over 5735147.14 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.369, pruned_loss=0.1189, over 5659994.97 frames. ], batch size: 65, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:29:46,805 INFO [optim.py:369] (1/2) 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 00:29:58,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3528, 1.7239, 1.3296, 1.4377], device='cuda:1'), covar=tensor([0.0755, 0.0329, 0.0327, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-12 00:30:05,106 INFO [train.py:968] (1/2) Epoch 23, batch 25150, giga_loss[loss=0.3054, simple_loss=0.3679, pruned_loss=0.1215, over 28209.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3667, pruned_loss=0.1181, over 5679977.00 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3554, pruned_loss=0.1072, over 5740766.08 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3689, pruned_loss=0.1195, over 5663553.49 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:30:20,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:30:58,744 INFO [train.py:968] (1/2) Epoch 23, batch 25200, giga_loss[loss=0.2586, simple_loss=0.3341, pruned_loss=0.09153, over 29031.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3653, pruned_loss=0.1178, over 5675536.53 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3552, pruned_loss=0.1071, over 5741722.24 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3672, pruned_loss=0.1191, over 5661528.00 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:31:25,852 INFO [optim.py:369] (1/2) 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 00:31:46,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5218, 1.6299, 1.6337, 1.4599], device='cuda:1'), covar=tensor([0.2710, 0.2673, 0.1964, 0.2317], device='cuda:1'), in_proj_covar=tensor([0.1985, 0.1938, 0.1856, 0.2001], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 00:31:47,761 INFO [train.py:968] (1/2) Epoch 23, batch 25250, giga_loss[loss=0.3181, simple_loss=0.3736, pruned_loss=0.1313, over 28582.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1177, over 5676511.22 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3557, pruned_loss=0.1074, over 5735807.57 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3665, pruned_loss=0.1187, over 5669704.16 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:32:43,278 INFO [train.py:968] (1/2) Epoch 23, batch 25300, giga_loss[loss=0.3088, simple_loss=0.3703, pruned_loss=0.1236, over 28943.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3656, pruned_loss=0.1185, over 5665444.10 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3556, pruned_loss=0.1073, over 5737040.64 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.367, pruned_loss=0.1194, over 5658147.37 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:32:43,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4757, 1.6910, 1.2488, 1.2181], device='cuda:1'), covar=tensor([0.0993, 0.0526, 0.1079, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 00:32:50,665 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:32:54,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:33:14,844 INFO [optim.py:369] (1/2) 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 00:33:19,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 00:33:23,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1028729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:33:25,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:33:29,945 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-12 00:33:34,463 INFO [train.py:968] (1/2) Epoch 23, batch 25350, giga_loss[loss=0.2899, simple_loss=0.3637, pruned_loss=0.108, over 28835.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.366, pruned_loss=0.1177, over 5657487.79 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3557, pruned_loss=0.1074, over 5726478.94 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1185, over 5660672.36 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:33:38,357 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:34:17,517 INFO [train.py:968] (1/2) Epoch 23, batch 25400, giga_loss[loss=0.3166, simple_loss=0.3827, pruned_loss=0.1253, over 28813.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3664, pruned_loss=0.1172, over 5667774.38 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3559, pruned_loss=0.1076, over 5732272.66 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3674, pruned_loss=0.118, over 5663230.75 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:34:23,070 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 00:34:29,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9067, 1.1672, 1.0819, 0.8681], device='cuda:1'), covar=tensor([0.2291, 0.2738, 0.1772, 0.2222], device='cuda:1'), in_proj_covar=tensor([0.1986, 0.1939, 0.1858, 0.2002], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 00:34:29,586 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-12 00:34:37,209 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:34:45,989 INFO [optim.py:369] (1/2) 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 00:35:07,456 INFO [train.py:968] (1/2) Epoch 23, batch 25450, giga_loss[loss=0.3155, simple_loss=0.3732, pruned_loss=0.1289, over 27504.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3672, pruned_loss=0.1179, over 5662561.40 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3559, pruned_loss=0.1078, over 5736576.11 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3684, pruned_loss=0.1187, over 5652841.57 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:35:35,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1028872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:35:37,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8012, 1.1399, 2.8920, 2.7401], device='cuda:1'), covar=tensor([0.1721, 0.2551, 0.0610, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0659, 0.0974, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 00:35:39,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1028875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:35:52,829 INFO [train.py:968] (1/2) Epoch 23, batch 25500, giga_loss[loss=0.283, simple_loss=0.3571, pruned_loss=0.1044, over 28683.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3675, pruned_loss=0.1187, over 5660336.58 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3558, pruned_loss=0.1079, over 5731595.12 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3689, pruned_loss=0.1195, over 5655276.25 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:36:07,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1028904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:36:16,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4809, 2.0467, 1.4361, 1.5447], device='cuda:1'), covar=tensor([0.2693, 0.2545, 0.3131, 0.2480], device='cuda:1'), in_proj_covar=tensor([0.1526, 0.1105, 0.1348, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 00:36:22,467 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-12 00:36:24,946 INFO [optim.py:369] (1/2) 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 00:36:34,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0594, 4.9141, 4.6843, 2.5989], device='cuda:1'), covar=tensor([0.0499, 0.0650, 0.0726, 0.1627], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1165, 0.0987, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 00:36:45,502 INFO [train.py:968] (1/2) Epoch 23, batch 25550, giga_loss[loss=0.2499, simple_loss=0.3259, pruned_loss=0.08694, over 28669.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3707, pruned_loss=0.1218, over 5654968.02 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3558, pruned_loss=0.108, over 5734245.16 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.372, pruned_loss=0.1226, over 5647528.92 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:37:13,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1028967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:37:33,476 INFO [train.py:968] (1/2) Epoch 23, batch 25600, giga_loss[loss=0.3248, simple_loss=0.3875, pruned_loss=0.1311, over 28704.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3715, pruned_loss=0.1231, over 5667887.25 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3559, pruned_loss=0.1082, over 5738211.13 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3729, pruned_loss=0.1241, over 5656066.22 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:38:07,666 INFO [optim.py:369] (1/2) 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 00:38:27,843 INFO [train.py:968] (1/2) Epoch 23, batch 25650, giga_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1202, over 28876.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3733, pruned_loss=0.1256, over 5653447.37 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3568, pruned_loss=0.1089, over 5729944.51 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.126, over 5650336.42 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:38:28,769 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 25700, giga_loss[loss=0.3411, simple_loss=0.3921, pruned_loss=0.1451, over 27912.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3732, pruned_loss=0.1255, over 5652657.44 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3564, pruned_loss=0.1088, over 5733349.30 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3743, pruned_loss=0.1262, over 5645891.20 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:39:32,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-12 00:39:41,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:39:43,152 INFO [optim.py:369] (1/2) 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 00:40:03,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:40:05,146 INFO [train.py:968] (1/2) Epoch 23, batch 25750, giga_loss[loss=0.2691, simple_loss=0.3322, pruned_loss=0.103, over 28805.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3727, pruned_loss=0.1254, over 5654100.36 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.357, pruned_loss=0.1092, over 5733596.41 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3733, pruned_loss=0.1258, over 5647417.50 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:40:44,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:40:48,521 INFO [train.py:968] (1/2) Epoch 23, batch 25800, giga_loss[loss=0.2878, simple_loss=0.3642, pruned_loss=0.1056, over 29013.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3718, pruned_loss=0.1228, over 5676138.75 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3574, pruned_loss=0.1094, over 5741080.24 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3726, pruned_loss=0.1235, over 5660695.67 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:41:01,128 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:41:18,969 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 00:41:19,156 INFO [optim.py:369] (1/2) 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 00:41:39,569 INFO [train.py:968] (1/2) Epoch 23, batch 25850, giga_loss[loss=0.3804, simple_loss=0.4155, pruned_loss=0.1727, over 27888.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3702, pruned_loss=0.1212, over 5666850.82 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3579, pruned_loss=0.1098, over 5743244.54 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1216, over 5651746.85 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:41:57,973 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:02,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:12,094 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:25,702 INFO [train.py:968] (1/2) Epoch 23, batch 25900, giga_loss[loss=0.2919, simple_loss=0.3648, pruned_loss=0.1095, over 29059.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5674450.12 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3582, pruned_loss=0.1101, over 5744298.27 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.1201, over 5659825.39 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:42:28,871 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:42:53,740 INFO [optim.py:369] (1/2) 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 00:43:01,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:43:03,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:43:13,836 INFO [train.py:968] (1/2) Epoch 23, batch 25950, giga_loss[loss=0.3353, simple_loss=0.3892, pruned_loss=0.1407, over 28479.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3666, pruned_loss=0.1198, over 5684017.52 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3586, pruned_loss=0.1105, over 5746570.94 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3665, pruned_loss=0.1198, over 5669446.64 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:43:15,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:43:32,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:44:02,583 INFO [train.py:968] (1/2) Epoch 23, batch 26000, giga_loss[loss=0.3218, simple_loss=0.3932, pruned_loss=0.1252, over 28668.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.366, pruned_loss=0.1192, over 5688440.90 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3583, pruned_loss=0.1105, over 5751347.73 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3665, pruned_loss=0.1195, over 5670472.73 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:44:24,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:44:49,813 INFO [train.py:968] (1/2) Epoch 23, batch 26050, giga_loss[loss=0.3308, simple_loss=0.3981, pruned_loss=0.1318, over 27572.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3695, pruned_loss=0.1206, over 5688723.50 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3584, pruned_loss=0.1106, over 5751903.39 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 5672544.24 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:45:32,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:45:35,003 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 26100, giga_loss[loss=0.2881, simple_loss=0.3678, pruned_loss=0.1042, over 28864.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3713, pruned_loss=0.1185, over 5691685.74 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3588, pruned_loss=0.1109, over 5755313.15 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3717, pruned_loss=0.1186, over 5673941.94 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:45:57,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:01,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:08,553 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:25,195 INFO [train.py:968] (1/2) Epoch 23, batch 26150, giga_loss[loss=0.2592, simple_loss=0.3453, pruned_loss=0.08657, over 28903.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3726, pruned_loss=0.1193, over 5690994.66 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3583, pruned_loss=0.1109, over 5754735.84 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3737, pruned_loss=0.1196, over 5675364.20 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:46:41,289 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:46:46,919 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:47:01,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:47:08,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7474, 2.3894, 1.5041, 1.0428], device='cuda:1'), covar=tensor([0.6869, 0.3413, 0.3329, 0.6052], device='cuda:1'), in_proj_covar=tensor([0.1753, 0.1663, 0.1596, 0.1432], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 00:47:14,515 INFO [train.py:968] (1/2) Epoch 23, batch 26200, giga_loss[loss=0.3483, simple_loss=0.3854, pruned_loss=0.1556, over 23413.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3751, pruned_loss=0.1218, over 5689647.27 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3583, pruned_loss=0.111, over 5757672.81 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3764, pruned_loss=0.1222, over 5672846.71 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:47:15,969 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:47:42,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3609, 1.3776, 1.2026, 1.5437], device='cuda:1'), covar=tensor([0.0733, 0.0392, 0.0347, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0223], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0109], device='cuda:1')
+2023-03-12 00:47:44,197 INFO [optim.py:369] (1/2) 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 00:47:50,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0668, 2.2103, 2.3260, 1.8544], device='cuda:1'), covar=tensor([0.1761, 0.2208, 0.1347, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0705, 0.0951, 0.0849], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 00:47:59,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9495, 1.2178, 1.3475, 0.9787], device='cuda:1'), covar=tensor([0.1988, 0.1519, 0.2453, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0754, 0.0721, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 00:48:00,165 INFO [train.py:968] (1/2) Epoch 23, batch 26250, giga_loss[loss=0.3213, simple_loss=0.3771, pruned_loss=0.1327, over 28864.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3749, pruned_loss=0.122, over 5689053.40 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3581, pruned_loss=0.1109, over 5759949.47 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3763, pruned_loss=0.1226, over 5672958.30 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:48:12,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:19,016 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:21,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:30,402 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:48:50,117 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 26300, giga_loss[loss=0.2914, simple_loss=0.3602, pruned_loss=0.1114, over 28903.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3746, pruned_loss=0.1226, over 5691502.17 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3581, pruned_loss=0.111, over 5761583.17 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3758, pruned_loss=0.1231, over 5676887.07 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:49:21,651 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:49:21,970 INFO [optim.py:369] (1/2) 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 00:49:23,062 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1029725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:49:23,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:49:31,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5228, 3.3713, 3.2162, 2.0371], device='cuda:1'), covar=tensor([0.0720, 0.0841, 0.0787, 0.1699], device='cuda:1'), in_proj_covar=tensor([0.1267, 0.1172, 0.0995, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 00:49:37,688 INFO [train.py:968] (1/2) Epoch 23, batch 26350, giga_loss[loss=0.3183, simple_loss=0.3773, pruned_loss=0.1297, over 28814.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3725, pruned_loss=0.1219, over 5690692.74 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3583, pruned_loss=0.1112, over 5756029.43 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3738, pruned_loss=0.1224, over 5681307.50 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 00:49:52,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:50:27,134 INFO [train.py:968] (1/2) Epoch 23, batch 26400, giga_loss[loss=0.2988, simple_loss=0.3622, pruned_loss=0.1177, over 28803.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3723, pruned_loss=0.1232, over 5691064.07 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3582, pruned_loss=0.1112, over 5757669.36 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3735, pruned_loss=0.1238, over 5681683.73 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:50:29,568 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1029794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:50:33,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1029797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:50:47,989 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-12 00:50:50,632 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:51:00,294 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1029826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:51:14,416 INFO [train.py:968] (1/2) Epoch 23, batch 26450, giga_loss[loss=0.2888, simple_loss=0.3584, pruned_loss=0.1097, over 28928.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3713, pruned_loss=0.123, over 5677163.20 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3586, pruned_loss=0.1115, over 5747386.97 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1234, over 5676435.56 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:51:27,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5273, 3.4363, 1.6103, 1.6378], device='cuda:1'), covar=tensor([0.0971, 0.0337, 0.0877, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0562, 0.0393, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 00:52:01,939 INFO [train.py:968] (1/2) Epoch 23, batch 26500, giga_loss[loss=0.332, simple_loss=0.3981, pruned_loss=0.133, over 28821.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1237, over 5679679.23 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1118, over 5750855.58 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3732, pruned_loss=0.124, over 5674595.80 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:52:14,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1029907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:52:29,320 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 26550, giga_loss[loss=0.2735, simple_loss=0.3457, pruned_loss=0.1006, over 29013.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3719, pruned_loss=0.1242, over 5682082.81 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1118, over 5753636.77 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.373, pruned_loss=0.1249, over 5673639.92 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:52:47,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-12 00:53:30,461 INFO [train.py:968] (1/2) Epoch 23, batch 26600, giga_loss[loss=0.3223, simple_loss=0.3889, pruned_loss=0.1278, over 28910.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3706, pruned_loss=0.1239, over 5668993.01 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3588, pruned_loss=0.1118, over 5756727.30 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3719, pruned_loss=0.1249, over 5655837.81 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:53:50,963 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:00,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-12 00:54:01,520 INFO [optim.py:369] (1/2) 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 00:54:16,469 INFO [train.py:968] (1/2) Epoch 23, batch 26650, giga_loss[loss=0.281, simple_loss=0.363, pruned_loss=0.09955, over 28872.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3698, pruned_loss=0.1228, over 5670359.56 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1117, over 5757097.35 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3714, pruned_loss=0.124, over 5657379.60 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:54:20,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:25,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:28,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:54:43,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7972, 2.0729, 1.4648, 1.5970], device='cuda:1'), covar=tensor([0.0969, 0.0581, 0.1030, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 00:54:45,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-12 00:54:56,004 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 26700, libri_loss[loss=0.275, simple_loss=0.3507, pruned_loss=0.09966, over 29571.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3716, pruned_loss=0.1231, over 5671731.10 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3587, pruned_loss=0.1118, over 5759386.60 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.373, pruned_loss=0.1242, over 5658138.64 frames. ], batch size: 74, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:55:13,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:55:21,787 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:55:55,990 INFO [train.py:968] (1/2) Epoch 23, batch 26750, giga_loss[loss=0.3407, simple_loss=0.3908, pruned_loss=0.1452, over 28267.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.373, pruned_loss=0.1243, over 5656612.44 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.359, pruned_loss=0.112, over 5750344.55 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.374, pruned_loss=0.1251, over 5652059.55 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:56:30,232 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-12 00:56:41,518 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:56:43,370 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 26800, giga_loss[loss=0.3057, simple_loss=0.3839, pruned_loss=0.1137, over 28925.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1223, over 5663863.11 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1125, over 5739957.23 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3726, pruned_loss=0.1227, over 5667202.05 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:56:43,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:57:08,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4450, 1.4268, 4.0809, 3.4105], device='cuda:1'), covar=tensor([0.1718, 0.2869, 0.0482, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0660, 0.0979, 0.0933], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 00:57:10,449 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:57:31,991 INFO [train.py:968] (1/2) Epoch 23, batch 26850, giga_loss[loss=0.3202, simple_loss=0.3988, pruned_loss=0.1208, over 28977.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3727, pruned_loss=0.1195, over 5671266.31 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1125, over 5741979.79 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.373, pruned_loss=0.1199, over 5671114.35 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 00:57:33,811 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:57:38,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:57:38,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9901, 1.3154, 1.1180, 0.1873], device='cuda:1'), covar=tensor([0.4495, 0.3515, 0.4620, 0.7404], device='cuda:1'), in_proj_covar=tensor([0.1762, 0.1664, 0.1596, 0.1434], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 00:58:06,179 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:58:21,239 INFO [train.py:968] (1/2) Epoch 23, batch 26900, giga_loss[loss=0.3197, simple_loss=0.3877, pruned_loss=0.1259, over 28700.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3767, pruned_loss=0.1211, over 5677965.91 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3596, pruned_loss=0.1124, over 5742762.72 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3774, pruned_loss=0.1217, over 5676173.53 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:58:35,042 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:58:52,001 INFO [optim.py:369] (1/2) 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 00:58:59,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:02,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:08,338 INFO [train.py:968] (1/2) Epoch 23, batch 26950, giga_loss[loss=0.4357, simple_loss=0.4488, pruned_loss=0.2113, over 23693.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3795, pruned_loss=0.1237, over 5669472.60 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3595, pruned_loss=0.1124, over 5734875.13 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3805, pruned_loss=0.1244, over 5673550.93 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:59:29,535 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:50,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 00:59:54,431 INFO [train.py:968] (1/2) Epoch 23, batch 27000, giga_loss[loss=0.3809, simple_loss=0.4203, pruned_loss=0.1707, over 27520.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3812, pruned_loss=0.1262, over 5670907.49 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5737165.45 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3824, pruned_loss=0.1269, over 5670167.01 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 00:59:54,431 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 01:00:04,550 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 01:00:23,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8309, 2.6540, 1.6670, 0.8228], device='cuda:1'), covar=tensor([0.7840, 0.3472, 0.4122, 0.7517], device='cuda:1'), in_proj_covar=tensor([0.1771, 0.1673, 0.1603, 0.1441], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 01:00:38,699 INFO [optim.py:369] (1/2) 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 01:00:42,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:00:52,042 INFO [train.py:968] (1/2) Epoch 23, batch 27050, giga_loss[loss=0.3328, simple_loss=0.3902, pruned_loss=0.1377, over 28299.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3814, pruned_loss=0.1276, over 5661607.27 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3596, pruned_loss=0.1125, over 5744460.35 frames. ], giga_tot_loss[loss=0.3207, simple_loss=0.3835, pruned_loss=0.1289, over 5651386.69 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:01:00,367 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:01:33,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:01:40,063 INFO [train.py:968] (1/2) Epoch 23, batch 27100, giga_loss[loss=0.2795, simple_loss=0.3539, pruned_loss=0.1026, over 28642.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3795, pruned_loss=0.1267, over 5649346.07 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1124, over 5726459.07 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3819, pruned_loss=0.1281, over 5655638.42 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:02:09,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3358, 4.1514, 3.9525, 1.9402], device='cuda:1'), covar=tensor([0.0635, 0.0778, 0.0795, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.1171, 0.0992, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 01:02:14,829 INFO [optim.py:369] (1/2) 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 01:02:20,999 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:02:23,762 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 27150, giga_loss[loss=0.2688, simple_loss=0.3519, pruned_loss=0.09281, over 28635.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3797, pruned_loss=0.1263, over 5645200.23 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3595, pruned_loss=0.1126, over 5729790.36 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3819, pruned_loss=0.1276, over 5645797.82 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:02:51,039 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:03:18,923 INFO [train.py:968] (1/2) Epoch 23, batch 27200, libri_loss[loss=0.2997, simple_loss=0.3705, pruned_loss=0.1144, over 29384.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3783, pruned_loss=0.1234, over 5655309.02 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5732172.99 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3803, pruned_loss=0.1244, over 5652188.51 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:03:54,473 INFO [optim.py:369] (1/2) 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 01:03:55,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:04:00,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:04:11,869 INFO [train.py:968] (1/2) Epoch 23, batch 27250, giga_loss[loss=0.3019, simple_loss=0.3747, pruned_loss=0.1146, over 28704.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3785, pruned_loss=0.1229, over 5658823.30 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5732172.99 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.38, pruned_loss=0.1237, over 5656394.57 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:04:28,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:04:37,851 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1030667.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 01:04:52,372 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 27300, libri_loss[loss=0.2374, simple_loss=0.3155, pruned_loss=0.07969, over 29492.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3776, pruned_loss=0.1225, over 5667129.41 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1125, over 5736168.26 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.38, pruned_loss=0.1238, over 5659178.63 frames. ], batch size: 70, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:05:31,173 INFO [optim.py:369] (1/2) 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 01:05:45,933 INFO [train.py:968] (1/2) Epoch 23, batch 27350, giga_loss[loss=0.2819, simple_loss=0.3528, pruned_loss=0.1055, over 28618.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3752, pruned_loss=0.1211, over 5671033.52 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1121, over 5732071.39 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3778, pruned_loss=0.1227, over 5666734.29 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:06:36,559 INFO [train.py:968] (1/2) Epoch 23, batch 27400, giga_loss[loss=0.299, simple_loss=0.3672, pruned_loss=0.1154, over 28957.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3731, pruned_loss=0.1213, over 5646388.98 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5725400.70 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3752, pruned_loss=0.1226, over 5646478.79 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:06:49,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:14,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1030825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:14,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:14,713 INFO [optim.py:369] (1/2) 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 01:07:17,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:07:32,285 INFO [train.py:968] (1/2) Epoch 23, batch 27450, giga_loss[loss=0.3017, simple_loss=0.3659, pruned_loss=0.1188, over 29008.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3734, pruned_loss=0.1225, over 5641743.62 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3598, pruned_loss=0.1129, over 5726851.70 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3747, pruned_loss=0.1232, over 5639664.23 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:07:49,350 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:08:22,396 INFO [train.py:968] (1/2) Epoch 23, batch 27500, giga_loss[loss=0.2935, simple_loss=0.3593, pruned_loss=0.1139, over 28864.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3719, pruned_loss=0.1226, over 5653428.60 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5728596.98 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5649729.50 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:08:41,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3393, 3.3401, 1.4658, 1.5467], device='cuda:1'), covar=tensor([0.1012, 0.0341, 0.0903, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0564, 0.0394, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 01:08:58,814 INFO [optim.py:369] (1/2) 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 01:09:08,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8159, 2.0210, 1.6421, 1.9937], device='cuda:1'), covar=tensor([0.2552, 0.2698, 0.3026, 0.2528], device='cuda:1'), in_proj_covar=tensor([0.1525, 0.1105, 0.1350, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 01:09:10,814 INFO [train.py:968] (1/2) Epoch 23, batch 27550, giga_loss[loss=0.3011, simple_loss=0.3657, pruned_loss=0.1183, over 27922.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3713, pruned_loss=0.1228, over 5653743.82 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5731308.23 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3725, pruned_loss=0.1235, over 5645951.07 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:09:14,619 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:17,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:35,344 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1030968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:38,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1030971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:42,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8681, 2.0843, 1.4680, 1.6346], device='cuda:1'), covar=tensor([0.0981, 0.0627, 0.1066, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0448, 0.0521, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 01:09:44,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1030978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:09:49,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 01:09:54,730 INFO [train.py:968] (1/2) Epoch 23, batch 27600, giga_loss[loss=0.2885, simple_loss=0.3586, pruned_loss=0.1092, over 28599.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3692, pruned_loss=0.1207, over 5656409.23 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 5731708.90 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3702, pruned_loss=0.1214, over 5647834.94 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:10:04,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:10:18,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2604, 1.4611, 1.3864, 1.4350], device='cuda:1'), covar=tensor([0.0773, 0.0408, 0.0344, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 01:10:28,664 INFO [optim.py:369] (1/2) 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 01:10:41,203 INFO [train.py:968] (1/2) Epoch 23, batch 27650, giga_loss[loss=0.2377, simple_loss=0.322, pruned_loss=0.07666, over 28504.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3648, pruned_loss=0.1163, over 5667869.65 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3594, pruned_loss=0.1128, over 5735812.41 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3663, pruned_loss=0.1172, over 5655596.21 frames. ], batch size: 71, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:10:42,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1031042.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 01:11:32,714 INFO [train.py:968] (1/2) Epoch 23, batch 27700, giga_loss[loss=0.2924, simple_loss=0.3644, pruned_loss=0.1102, over 28492.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3637, pruned_loss=0.115, over 5657017.89 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5728952.81 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3648, pruned_loss=0.1157, over 5652389.16 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:12:11,310 INFO [optim.py:369] (1/2) 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 01:12:27,175 INFO [train.py:968] (1/2) Epoch 23, batch 27750, giga_loss[loss=0.2816, simple_loss=0.359, pruned_loss=0.1021, over 29055.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3624, pruned_loss=0.1146, over 5638597.64 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1127, over 5719884.98 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3635, pruned_loss=0.1153, over 5641674.08 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:12:33,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3523, 3.1796, 3.0524, 1.3879], device='cuda:1'), covar=tensor([0.0921, 0.1012, 0.0914, 0.2203], device='cuda:1'), in_proj_covar=tensor([0.1265, 0.1171, 0.0992, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 01:13:13,299 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1031185.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 01:13:14,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-12 01:13:15,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1031188.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 01:13:16,892 INFO [train.py:968] (1/2) Epoch 23, batch 27800, giga_loss[loss=0.2728, simple_loss=0.3333, pruned_loss=0.1061, over 28427.00 frames. ], tot_loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1127, over 5664066.43 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 5724050.88 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5660358.07 frames. ], batch size: 60, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:13:18,953 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:13:44,137 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031217.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 01:13:54,569 INFO [optim.py:369] (1/2) 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 01:14:05,069 INFO [train.py:968] (1/2) Epoch 23, batch 27850, giga_loss[loss=0.2702, simple_loss=0.344, pruned_loss=0.09826, over 28689.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3616, pruned_loss=0.1148, over 5651703.39 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3605, pruned_loss=0.1132, over 5711753.89 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3615, pruned_loss=0.1149, over 5657626.51 frames. ], batch size: 92, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:14:07,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3523, 1.4522, 1.3120, 1.5098], device='cuda:1'), covar=tensor([0.0694, 0.0444, 0.0338, 0.0745], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 01:14:09,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6278, 1.8778, 1.3497, 1.4506], device='cuda:1'), covar=tensor([0.0998, 0.0580, 0.1063, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0450, 0.0524, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 01:14:56,042 INFO [train.py:968] (1/2) Epoch 23, batch 27900, giga_loss[loss=0.28, simple_loss=0.3553, pruned_loss=0.1024, over 29053.00 frames. ], tot_loss[loss=0.299, simple_loss=0.365, pruned_loss=0.1166, over 5647284.92 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1135, over 5714429.25 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3645, pruned_loss=0.1164, over 5648897.16 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:15:32,281 INFO [optim.py:369] (1/2) 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 01:15:45,128 INFO [train.py:968] (1/2) Epoch 23, batch 27950, giga_loss[loss=0.3727, simple_loss=0.4101, pruned_loss=0.1677, over 27499.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3664, pruned_loss=0.1172, over 5637684.81 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3612, pruned_loss=0.1136, over 5707195.36 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3658, pruned_loss=0.1171, over 5644053.56 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:16:20,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3972, 1.5088, 1.6260, 1.2345], device='cuda:1'), covar=tensor([0.1495, 0.2249, 0.1215, 0.1523], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0707, 0.0955, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 01:16:31,299 INFO [train.py:968] (1/2) Epoch 23, batch 28000, giga_loss[loss=0.3231, simple_loss=0.3862, pruned_loss=0.13, over 28947.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1186, over 5644901.10 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3617, pruned_loss=0.114, over 5709478.45 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3673, pruned_loss=0.1182, over 5646096.13 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:16:46,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2548, 1.3474, 1.4415, 1.0512], device='cuda:1'), covar=tensor([0.1921, 0.3215, 0.1480, 0.1660], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0706, 0.0954, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 01:16:52,515 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:17:04,519 INFO [optim.py:369] (1/2) 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 01:17:14,402 INFO [train.py:968] (1/2) Epoch 23, batch 28050, giga_loss[loss=0.2947, simple_loss=0.3585, pruned_loss=0.1155, over 28512.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1194, over 5643827.24 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3611, pruned_loss=0.1137, over 5702603.56 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3684, pruned_loss=0.1196, over 5648633.97 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:18:01,535 INFO [train.py:968] (1/2) Epoch 23, batch 28100, libri_loss[loss=0.2755, simple_loss=0.3438, pruned_loss=0.1036, over 29543.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3707, pruned_loss=0.121, over 5656898.55 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3613, pruned_loss=0.1137, over 5707202.56 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3709, pruned_loss=0.1213, over 5655088.78 frames. ], batch size: 79, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:18:37,298 INFO [optim.py:369] (1/2) 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 01:18:48,406 INFO [train.py:968] (1/2) Epoch 23, batch 28150, giga_loss[loss=0.3105, simple_loss=0.3746, pruned_loss=0.1232, over 28777.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3729, pruned_loss=0.1221, over 5655305.14 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3617, pruned_loss=0.114, over 5700032.06 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3729, pruned_loss=0.1222, over 5660014.66 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:19:11,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1031566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:19:37,737 INFO [train.py:968] (1/2) Epoch 23, batch 28200, libri_loss[loss=0.2913, simple_loss=0.3688, pruned_loss=0.1069, over 27685.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3743, pruned_loss=0.1232, over 5652513.24 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3624, pruned_loss=0.1143, over 5701634.80 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.374, pruned_loss=0.1233, over 5653241.06 frames. ], batch size: 116, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:20:10,543 INFO [optim.py:369] (1/2) 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 01:20:22,555 INFO [train.py:968] (1/2) Epoch 23, batch 28250, giga_loss[loss=0.3892, simple_loss=0.4152, pruned_loss=0.1816, over 26458.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3732, pruned_loss=0.1228, over 5647939.38 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3624, pruned_loss=0.1144, over 5699198.99 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3734, pruned_loss=0.1232, over 5648559.96 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:20:42,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3062, 1.5718, 1.5186, 1.3961], device='cuda:1'), covar=tensor([0.1803, 0.1938, 0.2199, 0.1984], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0760, 0.0722, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 01:20:54,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3152, 1.1897, 4.0608, 3.3344], device='cuda:1'), covar=tensor([0.1719, 0.3045, 0.0452, 0.1406], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0662, 0.0981, 0.0933], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 01:21:14,665 INFO [train.py:968] (1/2) Epoch 23, batch 28300, giga_loss[loss=0.3142, simple_loss=0.3845, pruned_loss=0.1219, over 28060.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3748, pruned_loss=0.1232, over 5637666.82 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3622, pruned_loss=0.1144, over 5684964.99 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3755, pruned_loss=0.1237, over 5649548.48 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:21:35,943 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1031709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:21:38,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1031712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:21:56,422 INFO [optim.py:369] (1/2) 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 01:22:06,301 INFO [train.py:968] (1/2) Epoch 23, batch 28350, giga_loss[loss=0.3482, simple_loss=0.4052, pruned_loss=0.1456, over 28973.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3749, pruned_loss=0.1222, over 5651952.14 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3618, pruned_loss=0.1142, over 5689326.44 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.376, pruned_loss=0.123, over 5656637.17 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:22:06,557 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:22:55,544 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1031789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:22:56,694 INFO [train.py:968] (1/2) Epoch 23, batch 28400, giga_loss[loss=0.4376, simple_loss=0.4463, pruned_loss=0.2144, over 26572.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3761, pruned_loss=0.1246, over 5657402.15 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1145, over 5689481.55 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3768, pruned_loss=0.125, over 5660299.99 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:23:34,854 INFO [optim.py:369] (1/2) 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 01:23:49,248 INFO [train.py:968] (1/2) Epoch 23, batch 28450, giga_loss[loss=0.2968, simple_loss=0.3628, pruned_loss=0.1154, over 28958.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3736, pruned_loss=0.1233, over 5667596.70 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.362, pruned_loss=0.1145, over 5693903.37 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3746, pruned_loss=0.1239, over 5665460.96 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:24:03,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3066, 1.3099, 1.2530, 1.4319], device='cuda:1'), covar=tensor([0.0783, 0.0351, 0.0334, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 01:24:50,694 INFO [train.py:968] (1/2) Epoch 23, batch 28500, giga_loss[loss=0.3349, simple_loss=0.3977, pruned_loss=0.136, over 28979.00 frames. ], tot_loss[loss=0.307, simple_loss=0.371, pruned_loss=0.1215, over 5671254.51 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3619, pruned_loss=0.1143, over 5697214.32 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3721, pruned_loss=0.1223, over 5666371.97 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:25:26,746 INFO [optim.py:369] (1/2) 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 01:25:29,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1031932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:25:33,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1031935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:25:38,031 INFO [train.py:968] (1/2) Epoch 23, batch 28550, giga_loss[loss=0.2774, simple_loss=0.3588, pruned_loss=0.09802, over 29019.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3716, pruned_loss=0.1223, over 5679640.36 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3624, pruned_loss=0.1146, over 5701947.61 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3723, pruned_loss=0.1228, over 5671206.12 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:25:39,106 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:25:42,617 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-12 01:25:56,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 01:26:01,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1031964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:26:21,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1031981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:26:28,817 INFO [train.py:968] (1/2) Epoch 23, batch 28600, giga_loss[loss=0.2971, simple_loss=0.363, pruned_loss=0.1156, over 28722.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3721, pruned_loss=0.1235, over 5667571.09 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3626, pruned_loss=0.1146, over 5707048.39 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3726, pruned_loss=0.124, over 5655495.76 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:26:48,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-12 01:27:07,943 INFO [optim.py:369] (1/2) 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 01:27:18,865 INFO [train.py:968] (1/2) Epoch 23, batch 28650, giga_loss[loss=0.3201, simple_loss=0.3848, pruned_loss=0.1277, over 28591.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.371, pruned_loss=0.1224, over 5668699.24 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3622, pruned_loss=0.1145, over 5709078.02 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3718, pruned_loss=0.1231, over 5656922.13 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:27:53,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:28:10,274 INFO [train.py:968] (1/2) Epoch 23, batch 28700, giga_loss[loss=0.3382, simple_loss=0.3935, pruned_loss=0.1414, over 28846.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3739, pruned_loss=0.1253, over 5664915.67 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3626, pruned_loss=0.1147, over 5709304.95 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3744, pruned_loss=0.1258, over 5654621.16 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:28:25,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3926, 3.4068, 1.6590, 1.4498], device='cuda:1'), covar=tensor([0.1023, 0.0348, 0.0867, 0.1399], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0562, 0.0393, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 01:28:30,100 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:28:48,975 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 28750, giga_loss[loss=0.3429, simple_loss=0.4024, pruned_loss=0.1417, over 29047.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3758, pruned_loss=0.1269, over 5643990.59 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3632, pruned_loss=0.1151, over 5697756.87 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3758, pruned_loss=0.1271, over 5645669.62 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:29:52,471 INFO [train.py:968] (1/2) Epoch 23, batch 28800, giga_loss[loss=0.2676, simple_loss=0.3414, pruned_loss=0.09687, over 29022.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3739, pruned_loss=0.1263, over 5640724.61 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3629, pruned_loss=0.1151, over 5701883.38 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3744, pruned_loss=0.1268, over 5637414.07 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:30:24,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2904, 0.8030, 0.8562, 1.3658], device='cuda:1'), covar=tensor([0.0746, 0.0385, 0.0369, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 01:30:27,265 INFO [optim.py:369] (1/2) 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 01:30:36,644 INFO [train.py:968] (1/2) Epoch 23, batch 28850, giga_loss[loss=0.2818, simple_loss=0.356, pruned_loss=0.1039, over 28844.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3729, pruned_loss=0.1254, over 5658146.39 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3625, pruned_loss=0.1147, over 5708045.22 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.374, pruned_loss=0.1264, over 5648471.90 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:31:25,758 INFO [train.py:968] (1/2) Epoch 23, batch 28900, giga_loss[loss=0.3012, simple_loss=0.3717, pruned_loss=0.1154, over 28615.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3745, pruned_loss=0.1268, over 5647749.41 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.363, pruned_loss=0.1151, over 5707689.09 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.375, pruned_loss=0.1275, over 5639404.17 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:31:53,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:32:06,485 INFO [optim.py:369] (1/2) 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 01:32:15,647 INFO [train.py:968] (1/2) Epoch 23, batch 28950, libri_loss[loss=0.3235, simple_loss=0.398, pruned_loss=0.1245, over 29213.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3746, pruned_loss=0.1258, over 5656032.93 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3634, pruned_loss=0.1152, over 5712426.89 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.375, pruned_loss=0.1265, over 5643488.94 frames. ], batch size: 97, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:32:30,334 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:33:02,837 INFO [train.py:968] (1/2) Epoch 23, batch 29000, giga_loss[loss=0.3198, simple_loss=0.3785, pruned_loss=0.1306, over 29050.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3748, pruned_loss=0.1255, over 5660865.36 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3634, pruned_loss=0.1152, over 5714932.36 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3753, pruned_loss=0.1262, over 5647599.87 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:33:07,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8035, 2.1626, 1.8380, 1.8429], device='cuda:1'), covar=tensor([0.1977, 0.1936, 0.2165, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.1534, 0.1112, 0.1357, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 01:33:21,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4918, 1.6949, 1.6364, 1.5877], device='cuda:1'), covar=tensor([0.1995, 0.2265, 0.2371, 0.2124], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0758, 0.0723, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 01:33:24,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4513, 1.5740, 1.2448, 1.1455], device='cuda:1'), covar=tensor([0.0999, 0.0582, 0.1078, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0451, 0.0524, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 01:33:40,300 INFO [optim.py:369] (1/2) 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 01:33:48,780 INFO [train.py:968] (1/2) Epoch 23, batch 29050, giga_loss[loss=0.3005, simple_loss=0.361, pruned_loss=0.12, over 28741.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3754, pruned_loss=0.1262, over 5658580.54 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1157, over 5700512.15 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3758, pruned_loss=0.1265, over 5658191.33 frames. ], batch size: 78, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:33:58,712 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:06,919 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:10,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:30,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:36,245 INFO [train.py:968] (1/2) Epoch 23, batch 29100, giga_loss[loss=0.3482, simple_loss=0.4017, pruned_loss=0.1474, over 28664.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3765, pruned_loss=0.1271, over 5670588.87 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1157, over 5704715.32 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3773, pruned_loss=0.1277, over 5665404.15 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:34:36,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1032491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:41,955 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:34:46,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:35:09,574 INFO [optim.py:369] (1/2) 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 01:35:11,693 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 29150, libri_loss[loss=0.3213, simple_loss=0.3868, pruned_loss=0.1279, over 29651.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3766, pruned_loss=0.1272, over 5662889.50 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3639, pruned_loss=0.1161, over 5702642.64 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3775, pruned_loss=0.1278, over 5658578.62 frames. ], batch size: 88, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:36:12,573 INFO [train.py:968] (1/2) Epoch 23, batch 29200, giga_loss[loss=0.3966, simple_loss=0.4338, pruned_loss=0.1797, over 27607.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3761, pruned_loss=0.1252, over 5667856.54 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3633, pruned_loss=0.1156, over 5706764.63 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3775, pruned_loss=0.1264, over 5659858.68 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 01:36:14,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:15,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:17,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:19,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-12 01:36:29,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3413, 1.6580, 1.3503, 1.0145], device='cuda:1'), covar=tensor([0.2715, 0.2697, 0.3142, 0.2418], device='cuda:1'), in_proj_covar=tensor([0.1533, 0.1111, 0.1356, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 01:36:44,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1032625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:46,438 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1032627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:48,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1032630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:36:48,946 INFO [optim.py:369] (1/2) 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 01:36:57,305 INFO [train.py:968] (1/2) Epoch 23, batch 29250, giga_loss[loss=0.2823, simple_loss=0.3592, pruned_loss=0.1027, over 29036.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3748, pruned_loss=0.1234, over 5673483.87 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3631, pruned_loss=0.1154, over 5713468.54 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3766, pruned_loss=0.1248, over 5659734.53 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:37:12,658 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1032659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:37:42,778 INFO [train.py:968] (1/2) Epoch 23, batch 29300, giga_loss[loss=0.2735, simple_loss=0.3455, pruned_loss=0.1008, over 28297.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3714, pruned_loss=0.1214, over 5671447.57 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3624, pruned_loss=0.115, over 5716694.75 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3738, pruned_loss=0.1231, over 5656145.64 frames. ], batch size: 77, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:38:15,767 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:38:18,524 INFO [optim.py:369] (1/2) 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 01:38:24,949 INFO [train.py:968] (1/2) Epoch 23, batch 29350, giga_loss[loss=0.3389, simple_loss=0.3967, pruned_loss=0.1406, over 28295.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3721, pruned_loss=0.1222, over 5664447.69 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3628, pruned_loss=0.1153, over 5711679.87 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.374, pruned_loss=0.1236, over 5654241.51 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:38:42,822 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1032760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:39:13,376 INFO [train.py:968] (1/2) Epoch 23, batch 29400, giga_loss[loss=0.3206, simple_loss=0.3866, pruned_loss=0.1273, over 28774.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1217, over 5664329.14 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1151, over 5706889.05 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3741, pruned_loss=0.1231, over 5658999.19 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:39:22,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4472, 2.9971, 1.5213, 1.5508], device='cuda:1'), covar=tensor([0.0950, 0.0378, 0.0875, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0563, 0.0393, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 01:39:58,041 INFO [optim.py:369] (1/2) 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 01:40:08,952 INFO [train.py:968] (1/2) Epoch 23, batch 29450, giga_loss[loss=0.3448, simple_loss=0.3964, pruned_loss=0.1467, over 28278.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3742, pruned_loss=0.1239, over 5655488.35 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1151, over 5706889.05 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.125, over 5651339.99 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:40:57,375 INFO [train.py:968] (1/2) Epoch 23, batch 29500, giga_loss[loss=0.3239, simple_loss=0.3855, pruned_loss=0.1312, over 28849.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3734, pruned_loss=0.1238, over 5664874.19 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3625, pruned_loss=0.115, over 5713376.37 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.375, pruned_loss=0.1251, over 5654110.78 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:41:37,700 INFO [optim.py:369] (1/2) 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 01:41:46,039 INFO [train.py:968] (1/2) Epoch 23, batch 29550, giga_loss[loss=0.2842, simple_loss=0.3479, pruned_loss=0.1103, over 28866.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3755, pruned_loss=0.1258, over 5662592.44 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3629, pruned_loss=0.1152, over 5715800.99 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1268, over 5651092.55 frames. ], batch size: 112, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:42:14,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1032969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:42:32,972 INFO [train.py:968] (1/2) Epoch 23, batch 29600, giga_loss[loss=0.3151, simple_loss=0.3731, pruned_loss=0.1286, over 28794.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3755, pruned_loss=0.1255, over 5666982.96 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.363, pruned_loss=0.1151, over 5718777.90 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3765, pruned_loss=0.1265, over 5654436.50 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:43:14,652 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 29650, libri_loss[loss=0.3218, simple_loss=0.3845, pruned_loss=0.1296, over 27549.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3767, pruned_loss=0.1267, over 5656020.45 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3632, pruned_loss=0.1153, over 5712731.21 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3777, pruned_loss=0.1276, over 5648727.98 frames. ], batch size: 115, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:44:10,152 INFO [train.py:968] (1/2) Epoch 23, batch 29700, giga_loss[loss=0.3225, simple_loss=0.3844, pruned_loss=0.1303, over 28671.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3749, pruned_loss=0.1246, over 5671244.91 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3631, pruned_loss=0.1154, over 5712120.89 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3759, pruned_loss=0.1253, over 5665461.72 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:44:23,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1033104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:29,578 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1033112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:32,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1033115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:40,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1484, 3.9867, 3.8083, 2.0291], device='cuda:1'), covar=tensor([0.0616, 0.0742, 0.0728, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.1274, 0.1180, 0.1001, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 01:44:46,489 INFO [optim.py:369] (1/2) 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 01:44:49,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7841, 1.8004, 2.0050, 1.5665], device='cuda:1'), covar=tensor([0.1939, 0.2564, 0.1563, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0707, 0.0953, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 01:44:50,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1033135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:44:54,772 INFO [train.py:968] (1/2) Epoch 23, batch 29750, giga_loss[loss=0.2828, simple_loss=0.3585, pruned_loss=0.1036, over 28956.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3749, pruned_loss=0.1241, over 5659675.71 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3635, pruned_loss=0.1156, over 5706309.48 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3757, pruned_loss=0.1248, over 5659039.08 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:44:59,711 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1033144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:45:46,094 INFO [train.py:968] (1/2) Epoch 23, batch 29800, giga_loss[loss=0.2603, simple_loss=0.3365, pruned_loss=0.0921, over 28742.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3733, pruned_loss=0.1226, over 5665996.50 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3631, pruned_loss=0.1155, over 5710342.57 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3744, pruned_loss=0.1235, over 5661002.35 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:46:27,842 INFO [optim.py:369] (1/2) 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 01:46:34,978 INFO [train.py:968] (1/2) Epoch 23, batch 29850, giga_loss[loss=0.2878, simple_loss=0.3544, pruned_loss=0.1106, over 28895.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3709, pruned_loss=0.1216, over 5663465.08 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.363, pruned_loss=0.1153, over 5710883.39 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.372, pruned_loss=0.1224, over 5658583.67 frames. ], batch size: 106, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:46:43,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1033247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:46:45,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1033250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:12,929 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1033278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:13,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1033279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:15,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1033281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:47:27,590 INFO [train.py:968] (1/2) Epoch 23, batch 29900, giga_loss[loss=0.3717, simple_loss=0.4, pruned_loss=0.1717, over 23497.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3689, pruned_loss=0.1207, over 5658226.49 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3626, pruned_loss=0.115, over 5712886.56 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3701, pruned_loss=0.1216, over 5652218.26 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:47:44,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1033310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:48:04,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-12 01:48:07,465 INFO [optim.py:369] (1/2) 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 01:48:15,560 INFO [train.py:968] (1/2) Epoch 23, batch 29950, giga_loss[loss=0.3498, simple_loss=0.385, pruned_loss=0.1573, over 26437.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3656, pruned_loss=0.1188, over 5668804.25 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3627, pruned_loss=0.1149, over 5716083.02 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3666, pruned_loss=0.1198, over 5660105.81 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:48:29,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 01:48:30,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4578, 3.6796, 1.6081, 1.6117], device='cuda:1'), covar=tensor([0.0937, 0.0351, 0.0848, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0563, 0.0394, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 01:48:49,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8227, 1.4223, 5.1176, 3.8657], device='cuda:1'), covar=tensor([0.1657, 0.2807, 0.0403, 0.0724], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0659, 0.0980, 0.0935], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 01:48:52,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2318, 1.6584, 1.2289, 0.4912], device='cuda:1'), covar=tensor([0.3940, 0.2462, 0.3562, 0.5686], device='cuda:1'), in_proj_covar=tensor([0.1773, 0.1677, 0.1612, 0.1444], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 01:48:58,456 INFO [train.py:968] (1/2) Epoch 23, batch 30000, giga_loss[loss=0.3251, simple_loss=0.3823, pruned_loss=0.1339, over 28789.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3631, pruned_loss=0.1174, over 5677434.71 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3629, pruned_loss=0.1151, over 5708930.22 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3638, pruned_loss=0.1182, over 5674434.33 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:48:58,456 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 01:49:06,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4193, 2.9768, 1.4672, 1.5908], device='cuda:1'), covar=tensor([0.1030, 0.0392, 0.0974, 0.1438], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0564, 0.0394, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 01:49:07,292 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 01:49:21,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 01:49:52,392 INFO [optim.py:369] (1/2) 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 01:49:59,009 INFO [train.py:968] (1/2) Epoch 23, batch 30050, giga_loss[loss=0.3393, simple_loss=0.3928, pruned_loss=0.1428, over 28661.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3622, pruned_loss=0.1173, over 5690087.98 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3631, pruned_loss=0.1153, over 5711141.81 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3627, pruned_loss=0.1178, over 5685050.35 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:50:48,004 INFO [train.py:968] (1/2) Epoch 23, batch 30100, giga_loss[loss=0.2938, simple_loss=0.3593, pruned_loss=0.1141, over 28434.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3624, pruned_loss=0.1172, over 5680755.85 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3632, pruned_loss=0.1154, over 5704644.31 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3626, pruned_loss=0.1175, over 5682148.54 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:51:29,936 INFO [optim.py:369] (1/2) 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 01:51:37,085 INFO [train.py:968] (1/2) Epoch 23, batch 30150, giga_loss[loss=0.3072, simple_loss=0.3781, pruned_loss=0.1182, over 27919.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.361, pruned_loss=0.1143, over 5677031.02 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1152, over 5708267.54 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5674323.61 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:52:00,856 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1033565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:52:32,125 INFO [train.py:968] (1/2) Epoch 23, batch 30200, giga_loss[loss=0.2689, simple_loss=0.3494, pruned_loss=0.09418, over 28844.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1111, over 5668035.88 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1152, over 5709571.71 frames. ], giga_tot_loss[loss=0.291, simple_loss=0.3591, pruned_loss=0.1114, over 5663704.13 frames. ], batch size: 227, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:53:12,762 INFO [optim.py:369] (1/2) 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 01:53:21,416 INFO [train.py:968] (1/2) Epoch 23, batch 30250, giga_loss[loss=0.2698, simple_loss=0.3465, pruned_loss=0.09652, over 28916.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3548, pruned_loss=0.1075, over 5653618.10 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3624, pruned_loss=0.1152, over 5701988.40 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3553, pruned_loss=0.1076, over 5656371.24 frames. ], batch size: 145, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:54:03,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8068, 2.9537, 1.9075, 0.9439], device='cuda:1'), covar=tensor([0.7565, 0.3052, 0.3520, 0.5872], device='cuda:1'), in_proj_covar=tensor([0.1764, 0.1662, 0.1602, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 01:54:11,804 INFO [train.py:968] (1/2) Epoch 23, batch 30300, giga_loss[loss=0.2653, simple_loss=0.3538, pruned_loss=0.0884, over 28914.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3518, pruned_loss=0.1047, over 5653662.63 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.362, pruned_loss=0.1152, over 5706129.35 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3524, pruned_loss=0.1046, over 5651237.47 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:54:31,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1033709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 01:54:35,650 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 01:54:52,784 INFO [optim.py:369] (1/2) 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 01:55:00,507 INFO [train.py:968] (1/2) Epoch 23, batch 30350, giga_loss[loss=0.2492, simple_loss=0.3203, pruned_loss=0.08905, over 24046.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3492, pruned_loss=0.1014, over 5655655.87 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3611, pruned_loss=0.1149, over 5709437.29 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3502, pruned_loss=0.1013, over 5649361.43 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 01:55:49,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1833, 1.4934, 1.2012, 0.5374], device='cuda:1'), covar=tensor([0.2928, 0.1758, 0.2507, 0.4963], device='cuda:1'), in_proj_covar=tensor([0.1770, 0.1665, 0.1606, 0.1442], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 01:55:59,430 INFO [train.py:968] (1/2) Epoch 23, batch 30400, giga_loss[loss=0.2308, simple_loss=0.3232, pruned_loss=0.06924, over 27939.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3492, pruned_loss=0.1003, over 5633230.29 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.361, pruned_loss=0.1148, over 5701580.21 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3499, pruned_loss=0.1002, over 5635069.06 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:56:25,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.55 vs. limit=5.0
+2023-03-12 01:56:43,548 INFO [optim.py:369] (1/2) 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 01:56:51,714 INFO [train.py:968] (1/2) Epoch 23, batch 30450, giga_loss[loss=0.2847, simple_loss=0.3447, pruned_loss=0.1124, over 24248.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3502, pruned_loss=0.1012, over 5637503.28 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3606, pruned_loss=0.1147, over 5702871.93 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3509, pruned_loss=0.1008, over 5635951.57 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:57:42,872 INFO [train.py:968] (1/2) Epoch 23, batch 30500, giga_loss[loss=0.257, simple_loss=0.3391, pruned_loss=0.08747, over 28645.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3478, pruned_loss=0.09995, over 5635054.06 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3602, pruned_loss=0.1146, over 5707423.58 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3484, pruned_loss=0.09925, over 5627520.82 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:58:25,498 INFO [optim.py:369] (1/2) 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 01:58:33,042 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 30550, giga_loss[loss=0.2741, simple_loss=0.3482, pruned_loss=0.1, over 28970.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3443, pruned_loss=0.09732, over 5639434.65 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3596, pruned_loss=0.1144, over 5701615.36 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.345, pruned_loss=0.09664, over 5637347.67 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 01:59:25,320 INFO [train.py:968] (1/2) Epoch 23, batch 30600, giga_loss[loss=0.2937, simple_loss=0.3677, pruned_loss=0.1098, over 28703.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3439, pruned_loss=0.09712, over 5638023.42 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3594, pruned_loss=0.1144, over 5704266.32 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3444, pruned_loss=0.0963, over 5632790.27 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:00:02,920 INFO [optim.py:369] (1/2) 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 02:00:04,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5366, 1.6097, 1.7627, 1.3698], device='cuda:1'), covar=tensor([0.1690, 0.2598, 0.1470, 0.1819], device='cuda:1'), in_proj_covar=tensor([0.0903, 0.0700, 0.0948, 0.0848], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 02:00:08,396 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1034039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:00:10,329 INFO [train.py:968] (1/2) Epoch 23, batch 30650, giga_loss[loss=0.2797, simple_loss=0.3587, pruned_loss=0.1003, over 28790.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3448, pruned_loss=0.09761, over 5642208.04 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3591, pruned_loss=0.1147, over 5698428.92 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3449, pruned_loss=0.09611, over 5641202.38 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:00:17,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8752, 4.7098, 4.5072, 2.2765], device='cuda:1'), covar=tensor([0.0467, 0.0616, 0.0731, 0.2000], device='cuda:1'), in_proj_covar=tensor([0.1254, 0.1162, 0.0982, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 02:00:40,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3433, 1.2845, 3.5156, 3.1262], device='cuda:1'), covar=tensor([0.1488, 0.2554, 0.0523, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0659, 0.0977, 0.0927], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 02:00:50,782 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1034083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:00:51,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1034084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:00:54,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 30700, giga_loss[loss=0.2591, simple_loss=0.3329, pruned_loss=0.09264, over 28527.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3423, pruned_loss=0.09539, over 5655464.36 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3586, pruned_loss=0.1145, over 5704745.26 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3425, pruned_loss=0.09385, over 5647487.75 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:01:22,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1034115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:01:43,299 INFO [optim.py:369] (1/2) 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 02:01:48,902 INFO [train.py:968] (1/2) Epoch 23, batch 30750, libri_loss[loss=0.2923, simple_loss=0.3639, pruned_loss=0.1104, over 29466.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3394, pruned_loss=0.09359, over 5655772.13 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3582, pruned_loss=0.1145, over 5711150.60 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3393, pruned_loss=0.09166, over 5642117.72 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:02:41,585 INFO [train.py:968] (1/2) Epoch 23, batch 30800, giga_loss[loss=0.2579, simple_loss=0.3287, pruned_loss=0.09354, over 27917.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3355, pruned_loss=0.09142, over 5645167.73 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3577, pruned_loss=0.1143, over 5705420.09 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3355, pruned_loss=0.08975, over 5637940.51 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:02:53,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8123, 2.0421, 1.3604, 1.7017], device='cuda:1'), covar=tensor([0.0869, 0.0505, 0.0994, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0447, 0.0519, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:03:16,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1034227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:03:21,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1034230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:03:25,984 INFO [optim.py:369] (1/2) 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 02:03:32,340 INFO [train.py:968] (1/2) Epoch 23, batch 30850, giga_loss[loss=0.2503, simple_loss=0.3236, pruned_loss=0.08848, over 28802.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3349, pruned_loss=0.09182, over 5649272.55 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3576, pruned_loss=0.1143, over 5708896.98 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3347, pruned_loss=0.09004, over 5639606.99 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:03:52,618 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1034259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:03:55,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2373, 1.1463, 3.4321, 2.9701], device='cuda:1'), covar=tensor([0.1678, 0.3012, 0.0523, 0.1511], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0655, 0.0969, 0.0921], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 02:03:58,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7213, 2.0261, 1.5763, 1.9626], device='cuda:1'), covar=tensor([0.2772, 0.2733, 0.3173, 0.2521], device='cuda:1'), in_proj_covar=tensor([0.1538, 0.1109, 0.1362, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 02:04:00,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4210, 1.7186, 1.5721, 1.5787], device='cuda:1'), covar=tensor([0.0785, 0.0308, 0.0330, 0.0913], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 02:04:28,107 INFO [train.py:968] (1/2) Epoch 23, batch 30900, giga_loss[loss=0.2138, simple_loss=0.2837, pruned_loss=0.07197, over 24157.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3345, pruned_loss=0.09185, over 5619544.96 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3579, pruned_loss=0.1147, over 5698733.56 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3336, pruned_loss=0.08983, over 5620857.81 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:05:13,747 INFO [optim.py:369] (1/2) 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 02:05:20,972 INFO [train.py:968] (1/2) Epoch 23, batch 30950, giga_loss[loss=0.2462, simple_loss=0.3313, pruned_loss=0.0805, over 28897.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3367, pruned_loss=0.09249, over 5632267.92 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3571, pruned_loss=0.1141, over 5703880.92 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.336, pruned_loss=0.09056, over 5626092.69 frames. ], batch size: 112, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:06:19,435 INFO [train.py:968] (1/2) Epoch 23, batch 31000, libri_loss[loss=0.3036, simple_loss=0.3689, pruned_loss=0.1192, over 29520.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3394, pruned_loss=0.09269, over 5637594.21 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3571, pruned_loss=0.1143, over 5698950.14 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3385, pruned_loss=0.09065, over 5635535.82 frames. ], batch size: 80, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:06:49,242 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1034414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:07:14,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 02:07:17,673 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 31050, libri_loss[loss=0.282, simple_loss=0.3405, pruned_loss=0.1118, over 29535.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3408, pruned_loss=0.09344, over 5657805.21 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3567, pruned_loss=0.1142, over 5699311.98 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3401, pruned_loss=0.09146, over 5654853.11 frames. ], batch size: 79, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:08:29,441 INFO [train.py:968] (1/2) Epoch 23, batch 31100, giga_loss[loss=0.2607, simple_loss=0.3454, pruned_loss=0.08801, over 28035.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.338, pruned_loss=0.09187, over 5648559.59 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3566, pruned_loss=0.1143, over 5694804.06 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3373, pruned_loss=0.08997, over 5649715.01 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:09:22,968 INFO [optim.py:369] (1/2) 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 02:09:29,077 INFO [train.py:968] (1/2) Epoch 23, batch 31150, giga_loss[loss=0.2656, simple_loss=0.3548, pruned_loss=0.08826, over 28393.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3381, pruned_loss=0.09105, over 5642001.25 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3563, pruned_loss=0.1141, over 5678761.88 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3373, pruned_loss=0.08903, over 5655701.54 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:09:52,706 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1034557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:09:54,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1034560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:10:32,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1034589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:10:34,025 INFO [train.py:968] (1/2) Epoch 23, batch 31200, giga_loss[loss=0.2518, simple_loss=0.3234, pruned_loss=0.09012, over 29042.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3355, pruned_loss=0.08931, over 5648607.22 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3564, pruned_loss=0.1142, over 5681909.95 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3345, pruned_loss=0.08726, over 5656188.58 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:11:11,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5218, 1.7668, 1.3049, 1.3429], device='cuda:1'), covar=tensor([0.0984, 0.0515, 0.0978, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0447, 0.0520, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:11:33,299 INFO [optim.py:369] (1/2) 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 02:11:38,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5395, 1.7801, 1.2377, 1.3742], device='cuda:1'), covar=tensor([0.1011, 0.0559, 0.1003, 0.1124], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0446, 0.0520, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:11:38,822 INFO [train.py:968] (1/2) Epoch 23, batch 31250, giga_loss[loss=0.2621, simple_loss=0.3405, pruned_loss=0.09186, over 28524.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3327, pruned_loss=0.08895, over 5654779.24 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3562, pruned_loss=0.1143, over 5687607.62 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3315, pruned_loss=0.08656, over 5654696.75 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:12:25,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6464, 1.9470, 1.5188, 1.7839], device='cuda:1'), covar=tensor([0.2719, 0.2678, 0.3169, 0.2434], device='cuda:1'), in_proj_covar=tensor([0.1534, 0.1107, 0.1357, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 02:12:39,051 INFO [train.py:968] (1/2) Epoch 23, batch 31300, giga_loss[loss=0.22, simple_loss=0.3053, pruned_loss=0.06739, over 28055.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3329, pruned_loss=0.0896, over 5662731.23 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3561, pruned_loss=0.1145, over 5691194.37 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3314, pruned_loss=0.08694, over 5658826.25 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:12:44,807 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1034694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:12:49,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4662, 1.6354, 1.6744, 1.2856], device='cuda:1'), covar=tensor([0.1885, 0.2774, 0.1606, 0.1952], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0700, 0.0950, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 02:13:35,755 INFO [optim.py:369] (1/2) 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 02:13:41,351 INFO [train.py:968] (1/2) Epoch 23, batch 31350, giga_loss[loss=0.2527, simple_loss=0.3382, pruned_loss=0.08359, over 28397.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3327, pruned_loss=0.08921, over 5660809.36 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3555, pruned_loss=0.1144, over 5691821.91 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3317, pruned_loss=0.08674, over 5656709.68 frames. ], batch size: 369, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:13:43,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8876, 2.0367, 1.4404, 1.5696], device='cuda:1'), covar=tensor([0.1034, 0.0679, 0.1025, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0447, 0.0521, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:14:40,625 INFO [train.py:968] (1/2) Epoch 23, batch 31400, libri_loss[loss=0.2667, simple_loss=0.3244, pruned_loss=0.1045, over 29572.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3354, pruned_loss=0.09023, over 5645522.90 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3557, pruned_loss=0.1147, over 5678364.15 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3338, pruned_loss=0.08717, over 5653417.00 frames. ], batch size: 76, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:14:52,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8475, 1.2678, 1.2966, 1.0558], device='cuda:1'), covar=tensor([0.2021, 0.1311, 0.2079, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.0477, 0.0744, 0.0710, 0.0680], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 02:14:57,576 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:15:39,724 INFO [train.py:968] (1/2) Epoch 23, batch 31450, giga_loss[loss=0.1895, simple_loss=0.2787, pruned_loss=0.05016, over 28903.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3339, pruned_loss=0.08921, over 5659493.98 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.355, pruned_loss=0.1145, over 5684795.43 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3324, pruned_loss=0.08595, over 5659379.45 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:16:49,969 INFO [train.py:968] (1/2) Epoch 23, batch 31500, giga_loss[loss=0.2611, simple_loss=0.342, pruned_loss=0.09013, over 28585.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3331, pruned_loss=0.08898, over 5658233.57 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.355, pruned_loss=0.1145, over 5679514.13 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3314, pruned_loss=0.08588, over 5662860.07 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:17:04,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9647, 1.1688, 1.1206, 0.9241], device='cuda:1'), covar=tensor([0.2089, 0.2360, 0.1428, 0.1977], device='cuda:1'), in_proj_covar=tensor([0.1946, 0.1891, 0.1805, 0.1953], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 02:17:40,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4339, 1.8054, 1.6844, 1.2792], device='cuda:1'), covar=tensor([0.1645, 0.2604, 0.1459, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0701, 0.0952, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 02:17:47,900 INFO [optim.py:369] (1/2) 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 02:17:57,272 INFO [train.py:968] (1/2) Epoch 23, batch 31550, giga_loss[loss=0.2308, simple_loss=0.3102, pruned_loss=0.07572, over 24445.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3353, pruned_loss=0.09003, over 5655089.90 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3541, pruned_loss=0.114, over 5680879.84 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3342, pruned_loss=0.08734, over 5656956.60 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:19:03,665 INFO [train.py:968] (1/2) Epoch 23, batch 31600, giga_loss[loss=0.2532, simple_loss=0.3496, pruned_loss=0.07839, over 28587.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.339, pruned_loss=0.08971, over 5642504.11 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3543, pruned_loss=0.1141, over 5670809.00 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3376, pruned_loss=0.08699, over 5653305.09 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:19:55,928 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 31650, giga_loss[loss=0.2645, simple_loss=0.355, pruned_loss=0.08693, over 28764.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3406, pruned_loss=0.08929, over 5645947.58 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3535, pruned_loss=0.1137, over 5668343.91 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3397, pruned_loss=0.08651, over 5655390.07 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:20:17,033 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-12 02:20:37,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:20:51,797 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:21:04,808 INFO [train.py:968] (1/2) Epoch 23, batch 31700, giga_loss[loss=0.2356, simple_loss=0.299, pruned_loss=0.08611, over 24511.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3396, pruned_loss=0.08764, over 5645444.96 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3534, pruned_loss=0.1135, over 5671874.15 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3389, pruned_loss=0.08524, over 5649545.21 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:21:20,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2492, 1.5381, 1.6180, 1.4132], device='cuda:1'), covar=tensor([0.1892, 0.1597, 0.1628, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0744, 0.0708, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 02:22:07,066 INFO [optim.py:369] (1/2) 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 02:22:11,394 INFO [train.py:968] (1/2) Epoch 23, batch 31750, giga_loss[loss=0.2364, simple_loss=0.3275, pruned_loss=0.07263, over 28887.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3394, pruned_loss=0.08741, over 5654454.81 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3534, pruned_loss=0.1136, over 5674212.58 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3386, pruned_loss=0.08518, over 5655173.57 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:22:35,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 02:22:54,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5882, 1.6041, 1.8469, 1.4350], device='cuda:1'), covar=tensor([0.1763, 0.2456, 0.1409, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0906, 0.0700, 0.0952, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 02:23:09,607 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 31800, giga_loss[loss=0.2458, simple_loss=0.3266, pruned_loss=0.08249, over 28364.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3385, pruned_loss=0.08848, over 5655667.18 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3528, pruned_loss=0.1135, over 5680063.75 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3381, pruned_loss=0.0862, over 5650164.17 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:23:42,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3364, 2.8058, 1.4471, 1.4473], device='cuda:1'), covar=tensor([0.0962, 0.0365, 0.0925, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0558, 0.0393, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 02:23:47,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3792, 1.7642, 1.5309, 1.5786], device='cuda:1'), covar=tensor([0.0754, 0.0350, 0.0332, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 02:23:51,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1035212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:23:56,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:24:20,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-12 02:24:27,738 INFO [optim.py:369] (1/2) 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 02:24:36,335 INFO [train.py:968] (1/2) Epoch 23, batch 31850, giga_loss[loss=0.2281, simple_loss=0.298, pruned_loss=0.07908, over 24639.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3391, pruned_loss=0.08946, over 5665653.11 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3526, pruned_loss=0.1133, over 5683307.08 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3385, pruned_loss=0.08716, over 5657564.35 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:24:41,267 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:25:51,847 INFO [train.py:968] (1/2) Epoch 23, batch 31900, giga_loss[loss=0.2436, simple_loss=0.3232, pruned_loss=0.08203, over 28983.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3361, pruned_loss=0.08841, over 5675586.47 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3521, pruned_loss=0.113, over 5688093.44 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3357, pruned_loss=0.08614, over 5664609.29 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:26:32,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-12 02:26:35,101 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1035323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:26:37,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:26:49,571 INFO [optim.py:369] (1/2) 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 02:26:54,836 INFO [train.py:968] (1/2) Epoch 23, batch 31950, giga_loss[loss=0.2417, simple_loss=0.3245, pruned_loss=0.07947, over 28554.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3342, pruned_loss=0.08754, over 5669596.93 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3519, pruned_loss=0.1129, over 5682752.73 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3334, pruned_loss=0.08476, over 5664915.41 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:27:12,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:27:17,222 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:28:07,787 INFO [train.py:968] (1/2) Epoch 23, batch 32000, giga_loss[loss=0.2357, simple_loss=0.3191, pruned_loss=0.07621, over 28379.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3315, pruned_loss=0.08645, over 5665764.90 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3518, pruned_loss=0.113, over 5684323.50 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3307, pruned_loss=0.08385, over 5660455.74 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:29:13,486 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 32050, giga_loss[loss=0.3135, simple_loss=0.3952, pruned_loss=0.1159, over 29116.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3342, pruned_loss=0.08781, over 5671475.71 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3519, pruned_loss=0.1131, over 5686667.66 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3333, pruned_loss=0.08545, over 5665028.63 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:29:40,329 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:29:46,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4278, 1.6129, 1.1831, 1.1383], device='cuda:1'), covar=tensor([0.0990, 0.0507, 0.1037, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0446, 0.0520, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:30:23,878 INFO [train.py:968] (1/2) Epoch 23, batch 32100, giga_loss[loss=0.2759, simple_loss=0.3409, pruned_loss=0.1054, over 28418.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3371, pruned_loss=0.08915, over 5673376.88 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3517, pruned_loss=0.1129, over 5689596.81 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3365, pruned_loss=0.08709, over 5665229.84 frames. ], batch size: 369, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:31:29,744 INFO [optim.py:369] (1/2) 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 02:31:31,840 INFO [train.py:968] (1/2) Epoch 23, batch 32150, libri_loss[loss=0.2553, simple_loss=0.3206, pruned_loss=0.09505, over 29547.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3347, pruned_loss=0.08902, over 5671941.80 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3509, pruned_loss=0.1125, over 5694042.56 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3345, pruned_loss=0.08729, over 5660935.71 frames. ], batch size: 77, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:31:40,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:32:37,655 INFO [train.py:968] (1/2) Epoch 23, batch 32200, giga_loss[loss=0.247, simple_loss=0.3311, pruned_loss=0.08147, over 28728.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3354, pruned_loss=0.0899, over 5663650.44 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3512, pruned_loss=0.1128, over 5686362.06 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3349, pruned_loss=0.08809, over 5661395.33 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:32:49,259 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1035599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:32:53,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:33:29,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:33:40,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3205, 1.6385, 1.6452, 1.5067], device='cuda:1'), covar=tensor([0.1869, 0.1565, 0.1624, 0.1524], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0740, 0.0706, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 02:33:41,691 INFO [optim.py:369] (1/2) 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 02:33:44,673 INFO [train.py:968] (1/2) Epoch 23, batch 32250, giga_loss[loss=0.2469, simple_loss=0.3338, pruned_loss=0.07997, over 28083.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3374, pruned_loss=0.09087, over 5668930.13 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3514, pruned_loss=0.1129, over 5691545.96 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3362, pruned_loss=0.08853, over 5661601.13 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:34:42,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:34:58,838 INFO [train.py:968] (1/2) Epoch 23, batch 32300, giga_loss[loss=0.2938, simple_loss=0.3645, pruned_loss=0.1116, over 28358.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3389, pruned_loss=0.09068, over 5675514.64 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3511, pruned_loss=0.1126, over 5696182.00 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.338, pruned_loss=0.08861, over 5665085.16 frames. ], batch size: 369, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:35:57,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:36:09,353 INFO [optim.py:369] (1/2) 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 02:36:14,947 INFO [train.py:968] (1/2) Epoch 23, batch 32350, giga_loss[loss=0.283, simple_loss=0.3641, pruned_loss=0.101, over 28363.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3381, pruned_loss=0.08978, over 5667747.28 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3507, pruned_loss=0.1125, over 5691455.82 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3374, pruned_loss=0.08764, over 5663139.07 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 02:37:20,776 INFO [train.py:968] (1/2) Epoch 23, batch 32400, giga_loss[loss=0.2735, simple_loss=0.3442, pruned_loss=0.1014, over 28974.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3361, pruned_loss=0.09018, over 5679420.78 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3509, pruned_loss=0.1128, over 5698252.45 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3349, pruned_loss=0.08751, over 5669027.10 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:37:49,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0192, 3.8502, 3.6579, 1.9507], device='cuda:1'), covar=tensor([0.0651, 0.0789, 0.0787, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1231, 0.1141, 0.0963, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 02:38:02,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4050, 1.7726, 1.7234, 1.5228], device='cuda:1'), covar=tensor([0.1979, 0.2025, 0.2005, 0.1969], device='cuda:1'), in_proj_covar=tensor([0.0471, 0.0741, 0.0706, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 02:38:30,878 INFO [optim.py:369] (1/2) 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 02:38:33,786 INFO [train.py:968] (1/2) Epoch 23, batch 32450, giga_loss[loss=0.2144, simple_loss=0.2927, pruned_loss=0.0681, over 29096.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3309, pruned_loss=0.08832, over 5674858.42 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3509, pruned_loss=0.1128, over 5696605.54 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3298, pruned_loss=0.08596, over 5667695.21 frames. ], batch size: 113, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:39:05,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1035862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:39:20,367 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1035874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:39:37,571 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 02:39:42,676 INFO [train.py:968] (1/2) Epoch 23, batch 32500, giga_loss[loss=0.2552, simple_loss=0.3344, pruned_loss=0.08796, over 28307.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3291, pruned_loss=0.08745, over 5660947.00 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3507, pruned_loss=0.1129, over 5690991.61 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3281, pruned_loss=0.08515, over 5660012.33 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:39:56,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1035903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:40:02,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5360, 1.8022, 1.4167, 1.8377], device='cuda:1'), covar=tensor([0.2572, 0.2588, 0.2930, 0.2449], device='cuda:1'), in_proj_covar=tensor([0.1533, 0.1107, 0.1357, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 02:40:22,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1035922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:40:41,277 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 32550, libri_loss[loss=0.3285, simple_loss=0.3767, pruned_loss=0.1402, over 19199.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.33, pruned_loss=0.08805, over 5649500.18 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3505, pruned_loss=0.1128, over 5685174.52 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.329, pruned_loss=0.08578, over 5654176.51 frames. ], batch size: 186, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:40:48,491 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 32600, giga_loss[loss=0.238, simple_loss=0.3034, pruned_loss=0.0863, over 24345.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3285, pruned_loss=0.08714, over 5651922.52 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3506, pruned_loss=0.1129, over 5690396.99 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3271, pruned_loss=0.0847, over 5650307.49 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:42:01,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4129, 1.5671, 1.5770, 1.4457], device='cuda:1'), covar=tensor([0.2321, 0.2053, 0.1820, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.1947, 0.1879, 0.1803, 0.1949], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 02:42:42,836 INFO [optim.py:369] (1/2) 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 02:42:45,449 INFO [train.py:968] (1/2) Epoch 23, batch 32650, giga_loss[loss=0.2312, simple_loss=0.3219, pruned_loss=0.0702, over 28877.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3271, pruned_loss=0.08534, over 5653538.48 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3505, pruned_loss=0.113, over 5685136.22 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3256, pruned_loss=0.08276, over 5656099.43 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:42:58,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1036053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:43:11,141 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:43:13,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:43:41,049 INFO [train.py:968] (1/2) Epoch 23, batch 32700, giga_loss[loss=0.2423, simple_loss=0.3201, pruned_loss=0.08219, over 29084.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3269, pruned_loss=0.08619, over 5663766.06 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3496, pruned_loss=0.1124, over 5695639.85 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3251, pruned_loss=0.08296, over 5654581.71 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:43:50,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:44:49,198 INFO [optim.py:369] (1/2) 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 02:44:53,606 INFO [train.py:968] (1/2) Epoch 23, batch 32750, giga_loss[loss=0.239, simple_loss=0.3023, pruned_loss=0.08786, over 24305.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3259, pruned_loss=0.0851, over 5663057.76 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3495, pruned_loss=0.1123, over 5697745.22 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3244, pruned_loss=0.0824, over 5653605.94 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:45:40,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3192, 3.1551, 2.9781, 1.5170], device='cuda:1'), covar=tensor([0.0948, 0.1045, 0.0938, 0.2244], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1140, 0.0963, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 02:46:10,647 INFO [train.py:968] (1/2) Epoch 23, batch 32800, giga_loss[loss=0.2239, simple_loss=0.3116, pruned_loss=0.06807, over 28887.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3255, pruned_loss=0.08424, over 5658147.89 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3495, pruned_loss=0.1123, over 5697745.22 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3243, pruned_loss=0.08213, over 5650791.43 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:46:18,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8912, 1.2849, 1.3809, 1.0616], device='cuda:1'), covar=tensor([0.1946, 0.1370, 0.2159, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0736, 0.0702, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 02:46:18,853 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:46:21,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:46:37,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1036213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:46:56,363 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:47:09,245 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:47:13,128 INFO [train.py:968] (1/2) Epoch 23, batch 32850, giga_loss[loss=0.2784, simple_loss=0.3482, pruned_loss=0.1043, over 28901.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3275, pruned_loss=0.08621, over 5653354.43 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3497, pruned_loss=0.1126, over 5690737.11 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.326, pruned_loss=0.08385, over 5652345.68 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:48:11,850 INFO [train.py:968] (1/2) Epoch 23, batch 32900, libri_loss[loss=0.3591, simple_loss=0.3998, pruned_loss=0.1592, over 19250.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3264, pruned_loss=0.08563, over 5639309.62 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3501, pruned_loss=0.1129, over 5674289.65 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3241, pruned_loss=0.08254, over 5653861.23 frames. ], batch size: 187, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:48:45,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1036320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:49:08,381 INFO [optim.py:369] (1/2) 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 02:49:09,090 INFO [train.py:968] (1/2) Epoch 23, batch 32950, libri_loss[loss=0.3069, simple_loss=0.3702, pruned_loss=0.1217, over 29047.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3285, pruned_loss=0.0855, over 5649511.05 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3503, pruned_loss=0.1129, over 5679787.63 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3257, pruned_loss=0.08209, over 5655299.83 frames. ], batch size: 101, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:49:45,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4575, 4.3022, 4.0713, 2.1624], device='cuda:1'), covar=tensor([0.0595, 0.0772, 0.0913, 0.1926], device='cuda:1'), in_proj_covar=tensor([0.1231, 0.1138, 0.0961, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 02:49:55,321 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:49:59,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:50:00,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-12 02:50:07,965 INFO [train.py:968] (1/2) Epoch 23, batch 33000, giga_loss[loss=0.2508, simple_loss=0.3256, pruned_loss=0.08805, over 26827.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3313, pruned_loss=0.08586, over 5649595.38 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3506, pruned_loss=0.1131, over 5677472.73 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3286, pruned_loss=0.08265, over 5655906.65 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:50:07,965 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 02:50:16,958 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 02:50:30,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4409, 1.6368, 1.2060, 1.2211], device='cuda:1'), covar=tensor([0.0907, 0.0423, 0.0892, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0444, 0.0519, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:50:31,860 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 02:50:34,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.10 vs. limit=5.0
+2023-03-12 02:50:43,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:51:19,074 INFO [optim.py:369] (1/2) 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 02:51:19,873 INFO [train.py:968] (1/2) Epoch 23, batch 33050, giga_loss[loss=0.2556, simple_loss=0.3414, pruned_loss=0.08488, over 28637.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3323, pruned_loss=0.08652, over 5647719.77 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3503, pruned_loss=0.113, over 5682026.57 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3301, pruned_loss=0.08362, over 5647903.46 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:51:43,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8941, 1.0549, 2.8630, 2.6675], device='cuda:1'), covar=tensor([0.1678, 0.2714, 0.0578, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0764, 0.0652, 0.0960, 0.0908], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 02:51:49,451 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:51:54,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1036466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:52:21,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6850, 2.0459, 1.3043, 1.6579], device='cuda:1'), covar=tensor([0.1078, 0.0642, 0.1073, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0444, 0.0520, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:52:22,786 INFO [train.py:968] (1/2) Epoch 23, batch 33100, giga_loss[loss=0.2529, simple_loss=0.3336, pruned_loss=0.08615, over 28653.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3317, pruned_loss=0.08642, over 5655720.79 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3497, pruned_loss=0.1126, over 5687589.68 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3299, pruned_loss=0.08352, over 5649672.95 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:52:28,387 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:53:15,794 INFO [optim.py:369] (1/2) 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 02:53:17,815 INFO [train.py:968] (1/2) Epoch 23, batch 33150, giga_loss[loss=0.2013, simple_loss=0.2911, pruned_loss=0.05572, over 28976.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3313, pruned_loss=0.08649, over 5664605.08 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3493, pruned_loss=0.1124, over 5691697.01 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3296, pruned_loss=0.08346, over 5655274.00 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 02:54:00,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3466, 1.2668, 1.2659, 1.5091], device='cuda:1'), covar=tensor([0.0788, 0.0370, 0.0356, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 02:54:18,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1036588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:54:22,412 INFO [train.py:968] (1/2) Epoch 23, batch 33200, giga_loss[loss=0.2389, simple_loss=0.3173, pruned_loss=0.08024, over 27730.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3295, pruned_loss=0.08554, over 5656676.01 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3492, pruned_loss=0.1123, over 5691162.56 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3279, pruned_loss=0.08278, over 5649247.48 frames. ], batch size: 474, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:55:17,748 INFO [optim.py:369] (1/2) 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 02:55:19,339 INFO [train.py:968] (1/2) Epoch 23, batch 33250, giga_loss[loss=0.2225, simple_loss=0.3051, pruned_loss=0.06998, over 28135.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.327, pruned_loss=0.08501, over 5658590.12 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3486, pruned_loss=0.1119, over 5688321.28 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3256, pruned_loss=0.08236, over 5653549.99 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:56:14,718 INFO [train.py:968] (1/2) Epoch 23, batch 33300, giga_loss[loss=0.21, simple_loss=0.2923, pruned_loss=0.06385, over 28650.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3283, pruned_loss=0.08552, over 5672883.11 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3483, pruned_loss=0.1116, over 5695100.98 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3268, pruned_loss=0.08262, over 5661772.89 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:57:07,754 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1036731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:57:12,233 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:57:21,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0395, 2.3138, 1.6234, 1.8984], device='cuda:1'), covar=tensor([0.0943, 0.0581, 0.0959, 0.1094], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0445, 0.0521, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 02:57:22,073 INFO [train.py:968] (1/2) Epoch 23, batch 33350, giga_loss[loss=0.209, simple_loss=0.3019, pruned_loss=0.05805, over 28809.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3303, pruned_loss=0.08615, over 5673577.72 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3479, pruned_loss=0.1114, over 5699282.93 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3292, pruned_loss=0.0836, over 5660530.16 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:57:49,644 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1036763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 02:58:25,548 INFO [train.py:968] (1/2) Epoch 23, batch 33400, giga_loss[loss=0.2414, simple_loss=0.326, pruned_loss=0.07836, over 29017.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.331, pruned_loss=0.08708, over 5669855.92 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3476, pruned_loss=0.1112, over 5704252.91 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3297, pruned_loss=0.08441, over 5654006.80 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 02:59:29,396 INFO [optim.py:369] (1/2) 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 02:59:30,926 INFO [train.py:968] (1/2) Epoch 23, batch 33450, giga_loss[loss=0.2912, simple_loss=0.3715, pruned_loss=0.1054, over 28646.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08701, over 5681684.74 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3469, pruned_loss=0.1108, over 5704954.91 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3311, pruned_loss=0.08469, over 5668128.61 frames. ], batch size: 242, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 03:00:22,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1036889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:00:24,127 INFO [train.py:968] (1/2) Epoch 23, batch 33500, giga_loss[loss=0.2615, simple_loss=0.3509, pruned_loss=0.08604, over 28599.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3355, pruned_loss=0.0888, over 5670148.96 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.347, pruned_loss=0.1109, over 5697370.18 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3344, pruned_loss=0.08612, over 5664760.62 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:00:51,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3365, 4.1792, 3.9939, 1.8540], device='cuda:1'), covar=tensor([0.0568, 0.0694, 0.0733, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1136, 0.0961, 0.0717], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 03:01:27,698 INFO [optim.py:369] (1/2) 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 03:01:27,711 INFO [train.py:968] (1/2) Epoch 23, batch 33550, giga_loss[loss=0.2811, simple_loss=0.3547, pruned_loss=0.1038, over 28964.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3361, pruned_loss=0.08908, over 5662020.87 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3464, pruned_loss=0.1107, over 5693236.11 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3355, pruned_loss=0.08645, over 5660850.12 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:02:11,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1036969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:02:35,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 03:02:43,722 INFO [train.py:968] (1/2) Epoch 23, batch 33600, giga_loss[loss=0.2496, simple_loss=0.332, pruned_loss=0.08359, over 28943.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.335, pruned_loss=0.08865, over 5660022.63 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3463, pruned_loss=0.1107, over 5695213.84 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3344, pruned_loss=0.0864, over 5656989.88 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:02:46,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2833, 1.2295, 1.1878, 1.5226], device='cuda:1'), covar=tensor([0.0760, 0.0379, 0.0358, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 03:04:02,262 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2242, 0.7954, 0.8927, 1.4034], device='cuda:1'), covar=tensor([0.0750, 0.0443, 0.0382, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0118, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 03:04:03,504 INFO [train.py:968] (1/2) Epoch 23, batch 33650, giga_loss[loss=0.2342, simple_loss=0.3251, pruned_loss=0.07169, over 28894.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3331, pruned_loss=0.08779, over 5659983.47 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3463, pruned_loss=0.1107, over 5695213.84 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3327, pruned_loss=0.08604, over 5657623.04 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:04:06,891 INFO [optim.py:369] (1/2) 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 03:04:23,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:04:37,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3789, 1.1884, 3.9726, 3.3136], device='cuda:1'), covar=tensor([0.1636, 0.2852, 0.0511, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0767, 0.0653, 0.0962, 0.0908], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 03:05:10,585 INFO [train.py:968] (1/2) Epoch 23, batch 33700, giga_loss[loss=0.223, simple_loss=0.3108, pruned_loss=0.06754, over 29006.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3321, pruned_loss=0.08716, over 5659062.39 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3459, pruned_loss=0.1104, over 5700273.97 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3318, pruned_loss=0.08542, over 5651814.71 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:06:17,421 INFO [train.py:968] (1/2) Epoch 23, batch 33750, giga_loss[loss=0.2154, simple_loss=0.2976, pruned_loss=0.06664, over 28987.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3315, pruned_loss=0.08791, over 5667442.62 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3451, pruned_loss=0.11, over 5703438.32 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3318, pruned_loss=0.08648, over 5658059.32 frames. ], batch size: 128, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:06:18,145 INFO [optim.py:369] (1/2) 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 03:07:23,266 INFO [train.py:968] (1/2) Epoch 23, batch 33800, giga_loss[loss=0.2963, simple_loss=0.361, pruned_loss=0.1158, over 26768.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3312, pruned_loss=0.08898, over 5652024.51 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.345, pruned_loss=0.1099, over 5706263.85 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3311, pruned_loss=0.08734, over 5640746.97 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:07:35,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2794, 1.6055, 1.6486, 1.4292], device='cuda:1'), covar=tensor([0.1895, 0.1674, 0.1594, 0.1721], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0735, 0.0701, 0.0673], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 03:08:02,307 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4464, 4.3031, 4.0664, 1.8799], device='cuda:1'), covar=tensor([0.0626, 0.0758, 0.0884, 0.2115], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1136, 0.0961, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 03:08:17,327 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:08:25,825 INFO [train.py:968] (1/2) Epoch 23, batch 33850, giga_loss[loss=0.2312, simple_loss=0.3242, pruned_loss=0.06907, over 29008.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3292, pruned_loss=0.0864, over 5662174.17 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3445, pruned_loss=0.1097, over 5710283.96 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3293, pruned_loss=0.08489, over 5648576.03 frames. ], batch size: 284, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:08:28,730 INFO [optim.py:369] (1/2) 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 03:08:54,651 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:09:05,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-12 03:09:19,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7486, 2.1801, 1.8693, 1.9178], device='cuda:1'), covar=tensor([0.2210, 0.2624, 0.2489, 0.2507], device='cuda:1'), in_proj_covar=tensor([0.0465, 0.0736, 0.0702, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 03:09:24,169 INFO [train.py:968] (1/2) Epoch 23, batch 33900, giga_loss[loss=0.2384, simple_loss=0.3309, pruned_loss=0.07295, over 28912.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3291, pruned_loss=0.08492, over 5671224.50 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3445, pruned_loss=0.1097, over 5709689.17 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3287, pruned_loss=0.08296, over 5659317.25 frames. ], batch size: 164, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:10:23,390 INFO [train.py:968] (1/2) Epoch 23, batch 33950, giga_loss[loss=0.2613, simple_loss=0.3523, pruned_loss=0.08514, over 28414.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.331, pruned_loss=0.08385, over 5676678.31 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3444, pruned_loss=0.1096, over 5712743.83 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3306, pruned_loss=0.08202, over 5664088.60 frames. ], batch size: 368, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:10:26,778 INFO [optim.py:369] (1/2) 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 03:10:29,979 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:11:14,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5861, 1.7278, 1.8028, 1.3682], device='cuda:1'), covar=tensor([0.2068, 0.2803, 0.1692, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0696, 0.0951, 0.0852], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 03:11:22,838 INFO [train.py:968] (1/2) Epoch 23, batch 34000, giga_loss[loss=0.2668, simple_loss=0.3441, pruned_loss=0.09472, over 27612.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3312, pruned_loss=0.0836, over 5672362.23 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.344, pruned_loss=0.1093, over 5712551.61 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3308, pruned_loss=0.08167, over 5661480.15 frames. ], batch size: 472, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:11:43,125 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:11:47,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:11:50,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:16,410 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:19,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:34,030 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:12:36,556 INFO [train.py:968] (1/2) Epoch 23, batch 34050, giga_loss[loss=0.2385, simple_loss=0.3257, pruned_loss=0.07565, over 28050.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3309, pruned_loss=0.08329, over 5671633.01 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3438, pruned_loss=0.1091, over 5715232.47 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3307, pruned_loss=0.08157, over 5659954.64 frames. ], batch size: 412, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:12:39,194 INFO [optim.py:369] (1/2) 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 03:13:35,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:13:39,580 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 34100, giga_loss[loss=0.2839, simple_loss=0.3651, pruned_loss=0.1013, over 28437.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3322, pruned_loss=0.08411, over 5670593.14 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3437, pruned_loss=0.1091, over 5709597.41 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3319, pruned_loss=0.08228, over 5665884.56 frames. ], batch size: 369, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:14:04,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7451, 1.9855, 1.8446, 1.6737], device='cuda:1'), covar=tensor([0.1739, 0.2125, 0.2155, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.0467, 0.0737, 0.0703, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 03:14:20,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:14:53,239 INFO [train.py:968] (1/2) Epoch 23, batch 34150, giga_loss[loss=0.1976, simple_loss=0.2691, pruned_loss=0.06311, over 24954.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3335, pruned_loss=0.08531, over 5658765.92 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3443, pruned_loss=0.1095, over 5704844.38 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3324, pruned_loss=0.08281, over 5658744.06 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:14:56,430 INFO [optim.py:369] (1/2) 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 03:15:37,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3557, 2.1597, 1.6053, 0.4955], device='cuda:1'), covar=tensor([0.4657, 0.2725, 0.3693, 0.6013], device='cuda:1'), in_proj_covar=tensor([0.1770, 0.1667, 0.1610, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 03:15:40,751 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:15:44,937 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:16:05,199 INFO [train.py:968] (1/2) Epoch 23, batch 34200, giga_loss[loss=0.2529, simple_loss=0.3503, pruned_loss=0.07773, over 28879.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3339, pruned_loss=0.08523, over 5653588.75 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3443, pruned_loss=0.1098, over 5698150.44 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3329, pruned_loss=0.08267, over 5658401.71 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:16:28,628 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:16:35,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:17:15,889 INFO [train.py:968] (1/2) Epoch 23, batch 34250, giga_loss[loss=0.2731, simple_loss=0.3591, pruned_loss=0.09356, over 29025.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3374, pruned_loss=0.08663, over 5659925.73 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3441, pruned_loss=0.1096, over 5701387.36 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3367, pruned_loss=0.08442, over 5660141.21 frames. ], batch size: 285, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:17:22,669 INFO [optim.py:369] (1/2) 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 03:17:38,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4367, 1.6768, 1.3826, 1.3256], device='cuda:1'), covar=tensor([0.2840, 0.2947, 0.3515, 0.2570], device='cuda:1'), in_proj_covar=tensor([0.1530, 0.1105, 0.1355, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 03:17:39,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0657, 1.0781, 3.3310, 2.9930], device='cuda:1'), covar=tensor([0.1715, 0.2890, 0.0486, 0.1155], device='cuda:1'), in_proj_covar=tensor([0.0765, 0.0652, 0.0959, 0.0907], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 03:18:27,998 INFO [train.py:968] (1/2) Epoch 23, batch 34300, giga_loss[loss=0.2343, simple_loss=0.3226, pruned_loss=0.07305, over 28995.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3385, pruned_loss=0.08636, over 5665170.06 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3444, pruned_loss=0.1098, over 5693753.03 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3376, pruned_loss=0.08429, over 5670980.79 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:18:47,898 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6704, 1.8184, 1.5061, 1.9655], device='cuda:1'), covar=tensor([0.2653, 0.2770, 0.3184, 0.2511], device='cuda:1'), in_proj_covar=tensor([0.1528, 0.1104, 0.1353, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 03:19:39,690 INFO [train.py:968] (1/2) Epoch 23, batch 34350, giga_loss[loss=0.2132, simple_loss=0.2993, pruned_loss=0.0635, over 28533.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3366, pruned_loss=0.08634, over 5674329.35 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3444, pruned_loss=0.1098, over 5696765.50 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3358, pruned_loss=0.08432, over 5675938.41 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:19:42,836 INFO [optim.py:369] (1/2) 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 03:19:56,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:04,106 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:35,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:44,232 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:20:53,454 INFO [train.py:968] (1/2) Epoch 23, batch 34400, giga_loss[loss=0.2312, simple_loss=0.3242, pruned_loss=0.06911, over 28987.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3342, pruned_loss=0.08494, over 5678803.56 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3448, pruned_loss=0.1101, over 5699053.23 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3331, pruned_loss=0.0827, over 5677726.93 frames. ], batch size: 136, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:21:07,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1037802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:21:17,948 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1037809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:21:21,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3380, 1.6226, 1.1806, 1.1859], device='cuda:1'), covar=tensor([0.1127, 0.0551, 0.1130, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0442, 0.0518, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 03:21:56,992 INFO [train.py:968] (1/2) Epoch 23, batch 34450, giga_loss[loss=0.2418, simple_loss=0.3219, pruned_loss=0.08086, over 28798.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3318, pruned_loss=0.08256, over 5692399.97 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3446, pruned_loss=0.11, over 5701631.70 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3309, pruned_loss=0.08058, over 5689151.62 frames. ], batch size: 243, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:22:00,407 INFO [optim.py:369] (1/2) 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 03:23:01,820 INFO [train.py:968] (1/2) Epoch 23, batch 34500, giga_loss[loss=0.2676, simple_loss=0.3498, pruned_loss=0.09268, over 28930.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3325, pruned_loss=0.08346, over 5689573.34 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3447, pruned_loss=0.11, over 5705761.23 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3314, pruned_loss=0.08135, over 5683192.06 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:23:40,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:23:42,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:03,240 INFO [train.py:968] (1/2) Epoch 23, batch 34550, giga_loss[loss=0.241, simple_loss=0.3313, pruned_loss=0.07534, over 28606.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3346, pruned_loss=0.08487, over 5683809.91 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3446, pruned_loss=0.1098, over 5708012.79 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3335, pruned_loss=0.08264, over 5676229.48 frames. ], batch size: 307, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:24:06,836 INFO [optim.py:369] (1/2) 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 03:24:09,179 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1037945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:14,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1037948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:19,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-12 03:24:20,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:24:47,580 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1037977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:25:04,267 INFO [train.py:968] (1/2) Epoch 23, batch 34600, giga_loss[loss=0.2572, simple_loss=0.3251, pruned_loss=0.09465, over 24203.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3353, pruned_loss=0.08616, over 5665164.59 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3447, pruned_loss=0.1101, over 5699346.82 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3341, pruned_loss=0.08339, over 5666740.32 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:25:59,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9553, 2.8297, 1.8472, 1.0204], device='cuda:1'), covar=tensor([0.7466, 0.3756, 0.4324, 0.6773], device='cuda:1'), in_proj_covar=tensor([0.1772, 0.1671, 0.1610, 0.1447], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 03:26:01,924 INFO [train.py:968] (1/2) Epoch 23, batch 34650, giga_loss[loss=0.2393, simple_loss=0.3174, pruned_loss=0.0806, over 28907.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3321, pruned_loss=0.08568, over 5650183.23 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3448, pruned_loss=0.1103, over 5685468.50 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3308, pruned_loss=0.08278, over 5664267.28 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:26:04,898 INFO [optim.py:369] (1/2) 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 03:26:28,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 03:26:55,552 INFO [train.py:968] (1/2) Epoch 23, batch 34700, giga_loss[loss=0.2676, simple_loss=0.3472, pruned_loss=0.09395, over 28863.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3317, pruned_loss=0.08626, over 5664003.13 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3441, pruned_loss=0.1098, over 5692920.38 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3308, pruned_loss=0.08337, over 5667627.57 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:27:47,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:27:51,972 INFO [train.py:968] (1/2) Epoch 23, batch 34750, giga_loss[loss=0.2612, simple_loss=0.3596, pruned_loss=0.08144, over 28848.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3359, pruned_loss=0.08919, over 5642462.79 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3445, pruned_loss=0.1101, over 5675704.18 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3345, pruned_loss=0.08599, over 5660113.71 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:27:54,002 INFO [optim.py:369] (1/2) 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 03:28:31,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:28:39,070 INFO [train.py:968] (1/2) Epoch 23, batch 34800, giga_loss[loss=0.2976, simple_loss=0.3819, pruned_loss=0.1067, over 28820.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3445, pruned_loss=0.09373, over 5657001.11 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3443, pruned_loss=0.11, over 5674437.06 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3434, pruned_loss=0.09096, over 5671454.56 frames. ], batch size: 199, lr: 1.38e-03, grad_scale: 8.0
+2023-03-12 03:28:43,061 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-12 03:28:46,570 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 34850, giga_loss[loss=0.2576, simple_loss=0.3411, pruned_loss=0.08705, over 28617.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3502, pruned_loss=0.09678, over 5664362.16 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3443, pruned_loss=0.1099, over 5678264.38 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3494, pruned_loss=0.09445, over 5672259.31 frames. ], batch size: 336, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:29:30,808 INFO [optim.py:369] (1/2) 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 03:30:12,449 INFO [train.py:968] (1/2) Epoch 23, batch 34900, giga_loss[loss=0.2411, simple_loss=0.3267, pruned_loss=0.07778, over 29078.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3477, pruned_loss=0.09631, over 5667688.60 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3443, pruned_loss=0.1099, over 5679146.88 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3471, pruned_loss=0.09424, over 5672848.60 frames. ], batch size: 155, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:30:42,721 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:30:45,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:30:48,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:30:54,545 INFO [train.py:968] (1/2) Epoch 23, batch 34950, giga_loss[loss=0.2763, simple_loss=0.3393, pruned_loss=0.1066, over 26583.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3411, pruned_loss=0.09364, over 5677312.63 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3445, pruned_loss=0.1098, over 5685688.99 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3405, pruned_loss=0.09163, over 5675190.92 frames. ], batch size: 555, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:30:57,967 INFO [optim.py:369] (1/2) 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 03:31:12,569 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:31:37,358 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:31:38,975 INFO [train.py:968] (1/2) Epoch 23, batch 35000, giga_loss[loss=0.2257, simple_loss=0.3119, pruned_loss=0.06973, over 28954.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3336, pruned_loss=0.09035, over 5683091.93 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3441, pruned_loss=0.1095, over 5688077.07 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3333, pruned_loss=0.08863, over 5679067.57 frames. ], batch size: 213, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:32:18,659 INFO [train.py:968] (1/2) Epoch 23, batch 35050, giga_loss[loss=0.2152, simple_loss=0.2945, pruned_loss=0.06789, over 28683.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3277, pruned_loss=0.08834, over 5670946.73 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3445, pruned_loss=0.1097, over 5676581.59 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3265, pruned_loss=0.08597, over 5676747.08 frames. ], batch size: 119, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:32:19,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3532, 1.5728, 1.5615, 1.4092], device='cuda:1'), covar=tensor([0.1882, 0.1878, 0.2296, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0744, 0.0709, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 03:32:23,043 INFO [optim.py:369] (1/2) 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 03:32:33,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:32:36,624 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 03:32:42,009 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 03:33:00,110 INFO [train.py:968] (1/2) Epoch 23, batch 35100, giga_loss[loss=0.1964, simple_loss=0.2817, pruned_loss=0.05558, over 28904.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3204, pruned_loss=0.08504, over 5679566.52 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3444, pruned_loss=0.1097, over 5680490.33 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3191, pruned_loss=0.0827, over 5680564.65 frames. ], batch size: 174, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:33:20,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:33:45,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:33:46,189 INFO [train.py:968] (1/2) Epoch 23, batch 35150, giga_loss[loss=0.2053, simple_loss=0.2819, pruned_loss=0.06435, over 28734.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.3153, pruned_loss=0.08285, over 5687614.51 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3439, pruned_loss=0.1092, over 5686300.09 frames. ], giga_tot_loss[loss=0.2378, simple_loss=0.3141, pruned_loss=0.08076, over 5683350.70 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 2.0
+2023-03-12 03:33:49,649 INFO [optim.py:369] (1/2) 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 03:34:11,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:34:27,601 INFO [train.py:968] (1/2) Epoch 23, batch 35200, giga_loss[loss=0.2062, simple_loss=0.2809, pruned_loss=0.06578, over 28771.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3127, pruned_loss=0.08183, over 5692378.30 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3444, pruned_loss=0.1094, over 5686941.18 frames. ], giga_tot_loss[loss=0.235, simple_loss=0.3109, pruned_loss=0.0796, over 5688348.32 frames. ], batch size: 99, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:34:28,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3779, 1.6830, 1.3712, 1.0004], device='cuda:1'), covar=tensor([0.2753, 0.2788, 0.3195, 0.2553], device='cuda:1'), in_proj_covar=tensor([0.1530, 0.1104, 0.1356, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 03:34:38,978 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3931, 1.5040, 1.3362, 1.5601], device='cuda:1'), covar=tensor([0.0774, 0.0345, 0.0349, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 03:34:48,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1236, 1.3521, 1.1399, 0.9117], device='cuda:1'), covar=tensor([0.1070, 0.0542, 0.1168, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0443, 0.0519, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 03:35:13,963 INFO [train.py:968] (1/2) Epoch 23, batch 35250, giga_loss[loss=0.234, simple_loss=0.2892, pruned_loss=0.08941, over 23936.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.3088, pruned_loss=0.07988, over 5683790.70 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3444, pruned_loss=0.1093, over 5687742.47 frames. ], giga_tot_loss[loss=0.2317, simple_loss=0.3072, pruned_loss=0.07807, over 5679940.67 frames. ], batch size: 705, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:35:18,719 INFO [optim.py:369] (1/2) 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 03:35:19,232 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-12 03:35:25,848 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:35:28,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:35:55,604 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 35300, giga_loss[loss=0.2103, simple_loss=0.294, pruned_loss=0.0633, over 28676.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.3072, pruned_loss=0.07944, over 5682098.18 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3447, pruned_loss=0.1093, over 5695350.23 frames. ], giga_tot_loss[loss=0.2293, simple_loss=0.3046, pruned_loss=0.07704, over 5672375.77 frames. ], batch size: 262, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:36:15,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:36:20,736 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:36:22,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:36:42,715 INFO [train.py:968] (1/2) Epoch 23, batch 35350, giga_loss[loss=0.1941, simple_loss=0.2681, pruned_loss=0.06001, over 28542.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3039, pruned_loss=0.07782, over 5688121.77 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3447, pruned_loss=0.1092, over 5699577.41 frames. ], giga_tot_loss[loss=0.2256, simple_loss=0.3007, pruned_loss=0.07521, over 5676348.88 frames. ], batch size: 85, lr: 1.38e-03, grad_scale: 4.0
+2023-03-12 03:36:46,950 INFO [optim.py:369] (1/2) 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 03:36:47,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:37:00,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1038761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:37:02,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:37:26,914 INFO [train.py:968] (1/2) Epoch 23, batch 35400, giga_loss[loss=0.2186, simple_loss=0.2992, pruned_loss=0.06897, over 28896.00 frames. ], tot_loss[loss=0.2273, simple_loss=0.3015, pruned_loss=0.0765, over 5693621.00 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3451, pruned_loss=0.1094, over 5702757.14 frames. ], giga_tot_loss[loss=0.2226, simple_loss=0.2979, pruned_loss=0.07369, over 5681353.89 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:37:59,765 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:07,921 INFO [train.py:968] (1/2) Epoch 23, batch 35450, giga_loss[loss=0.185, simple_loss=0.272, pruned_loss=0.04901, over 28893.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3003, pruned_loss=0.07618, over 5690048.20 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3459, pruned_loss=0.1097, over 5694952.47 frames. ], giga_tot_loss[loss=0.2199, simple_loss=0.295, pruned_loss=0.07239, over 5686785.50 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:38:12,259 INFO [optim.py:369] (1/2) 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 03:38:18,892 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:20,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:46,238 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:38:51,527 INFO [train.py:968] (1/2) Epoch 23, batch 35500, giga_loss[loss=0.226, simple_loss=0.2991, pruned_loss=0.0764, over 28213.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2986, pruned_loss=0.0755, over 5676077.85 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3462, pruned_loss=0.1096, over 5686157.17 frames. ], giga_tot_loss[loss=0.2184, simple_loss=0.2931, pruned_loss=0.07181, over 5681808.48 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:39:09,577 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:11,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1038910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:15,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1038915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:15,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 03:39:40,725 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1038939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:39:41,667 INFO [train.py:968] (1/2) Epoch 23, batch 35550, giga_loss[loss=0.2212, simple_loss=0.3053, pruned_loss=0.06856, over 29107.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2961, pruned_loss=0.07475, over 5674363.50 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3469, pruned_loss=0.11, over 5689649.07 frames. ], giga_tot_loss[loss=0.2163, simple_loss=0.2906, pruned_loss=0.07105, over 5675573.58 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:39:44,774 INFO [optim.py:369] (1/2) 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 03:40:09,096 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1038974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:40:11,126 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 35600, giga_loss[loss=0.2757, simple_loss=0.3574, pruned_loss=0.09702, over 28700.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3038, pruned_loss=0.0789, over 5676495.39 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3473, pruned_loss=0.11, over 5692374.84 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.2974, pruned_loss=0.07489, over 5674431.68 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:40:34,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 03:40:41,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:12,880 INFO [train.py:968] (1/2) Epoch 23, batch 35650, giga_loss[loss=0.2917, simple_loss=0.3466, pruned_loss=0.1183, over 23854.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3162, pruned_loss=0.08545, over 5670603.93 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3475, pruned_loss=0.1102, over 5686431.79 frames. ], giga_tot_loss[loss=0.236, simple_loss=0.3098, pruned_loss=0.08116, over 5674048.99 frames. ], batch size: 710, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:41:15,988 INFO [optim.py:369] (1/2) 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 03:41:17,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2820, 1.4457, 1.3129, 1.4381], device='cuda:1'), covar=tensor([0.0834, 0.0366, 0.0362, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 03:41:28,957 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:31,353 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:58,352 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:41:58,720 INFO [train.py:968] (1/2) Epoch 23, batch 35700, giga_loss[loss=0.2851, simple_loss=0.3678, pruned_loss=0.1013, over 28876.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.328, pruned_loss=0.09127, over 5666540.32 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3476, pruned_loss=0.1101, over 5680390.08 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3224, pruned_loss=0.08752, over 5674128.85 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:42:40,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:42:40,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-12 03:42:43,999 INFO [train.py:968] (1/2) Epoch 23, batch 35750, giga_loss[loss=0.2899, simple_loss=0.3726, pruned_loss=0.1036, over 28979.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3365, pruned_loss=0.09482, over 5676924.86 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3477, pruned_loss=0.1101, over 5683815.76 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3317, pruned_loss=0.09164, over 5679916.82 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:42:48,184 INFO [optim.py:369] (1/2) 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 03:43:09,438 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-12 03:43:27,714 INFO [train.py:968] (1/2) Epoch 23, batch 35800, giga_loss[loss=0.286, simple_loss=0.3675, pruned_loss=0.1023, over 28694.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.339, pruned_loss=0.09466, over 5684040.04 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3477, pruned_loss=0.11, over 5687867.81 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.335, pruned_loss=0.09185, over 5682632.32 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:43:59,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7716, 1.9386, 1.8779, 1.6936], device='cuda:1'), covar=tensor([0.2459, 0.2042, 0.1735, 0.1874], device='cuda:1'), in_proj_covar=tensor([0.1986, 0.1908, 0.1818, 0.1975], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 03:44:13,622 INFO [train.py:968] (1/2) Epoch 23, batch 35850, giga_loss[loss=0.2515, simple_loss=0.3399, pruned_loss=0.08154, over 28863.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3409, pruned_loss=0.09477, over 5677919.66 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.348, pruned_loss=0.1099, over 5692855.06 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.337, pruned_loss=0.092, over 5671589.93 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:44:18,527 INFO [optim.py:369] (1/2) 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 03:44:30,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4846, 1.7553, 1.7261, 1.2632], device='cuda:1'), covar=tensor([0.1494, 0.2500, 0.1320, 0.1637], device='cuda:1'), in_proj_covar=tensor([0.0910, 0.0700, 0.0958, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 03:44:43,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1039274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:48,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:50,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:54,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1039289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:44:56,119 INFO [train.py:968] (1/2) Epoch 23, batch 35900, giga_loss[loss=0.2747, simple_loss=0.353, pruned_loss=0.09824, over 28651.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3424, pruned_loss=0.09587, over 5680956.98 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3482, pruned_loss=0.1099, over 5701152.25 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3388, pruned_loss=0.093, over 5667438.09 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:45:14,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:45:40,659 INFO [train.py:968] (1/2) Epoch 23, batch 35950, giga_loss[loss=0.3227, simple_loss=0.3877, pruned_loss=0.1288, over 29018.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3445, pruned_loss=0.09747, over 5697067.07 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3478, pruned_loss=0.1097, over 5707452.73 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3418, pruned_loss=0.09501, over 5680156.18 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:45:44,774 INFO [optim.py:369] (1/2) 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 03:46:21,618 INFO [train.py:968] (1/2) Epoch 23, batch 36000, giga_loss[loss=0.2882, simple_loss=0.3639, pruned_loss=0.1062, over 28579.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.348, pruned_loss=0.1001, over 5686889.90 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3485, pruned_loss=0.1101, over 5702176.54 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3453, pruned_loss=0.09755, over 5677259.90 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:46:21,618 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 03:46:30,745 INFO [train.py:1012] (1/2) Epoch 23, validation: loss=0.2031, simple_loss=0.3104, pruned_loss=0.04787, over 944034.00 frames. 
+2023-03-12 03:46:30,746 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 03:46:53,720 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-12 03:47:10,516 INFO [train.py:968] (1/2) Epoch 23, batch 36050, giga_loss[loss=0.2763, simple_loss=0.3585, pruned_loss=0.09707, over 28674.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3515, pruned_loss=0.1016, over 5681474.83 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.349, pruned_loss=0.1104, over 5688056.08 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3488, pruned_loss=0.09889, over 5686194.55 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:47:15,165 INFO [optim.py:369] (1/2) 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 03:47:48,537 INFO [train.py:968] (1/2) Epoch 23, batch 36100, giga_loss[loss=0.2745, simple_loss=0.3546, pruned_loss=0.09718, over 28522.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.354, pruned_loss=0.1018, over 5699277.36 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3496, pruned_loss=0.1104, over 5693726.25 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3515, pruned_loss=0.09928, over 5697810.04 frames. ], batch size: 78, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:48:34,103 INFO [train.py:968] (1/2) Epoch 23, batch 36150, giga_loss[loss=0.2643, simple_loss=0.3507, pruned_loss=0.08898, over 28940.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3544, pruned_loss=0.1012, over 5692456.61 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3492, pruned_loss=0.1101, over 5696728.30 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3528, pruned_loss=0.09941, over 5688708.08 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:48:38,682 INFO [optim.py:369] (1/2) 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 03:49:14,880 INFO [train.py:968] (1/2) Epoch 23, batch 36200, giga_loss[loss=0.2278, simple_loss=0.3253, pruned_loss=0.06516, over 28860.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3549, pruned_loss=0.1003, over 5692249.97 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3495, pruned_loss=0.1102, over 5697961.20 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3534, pruned_loss=0.09873, over 5688270.23 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:49:54,151 INFO [train.py:968] (1/2) Epoch 23, batch 36250, giga_loss[loss=0.2639, simple_loss=0.3509, pruned_loss=0.08845, over 28787.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3539, pruned_loss=0.09881, over 5700548.93 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3499, pruned_loss=0.1103, over 5698786.54 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3525, pruned_loss=0.09711, over 5696724.76 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:49:57,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 03:50:01,135 INFO [optim.py:369] (1/2) 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 03:50:03,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:50:13,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-12 03:50:14,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1039664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:50:37,586 INFO [train.py:968] (1/2) Epoch 23, batch 36300, giga_loss[loss=0.2766, simple_loss=0.357, pruned_loss=0.0981, over 28764.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3513, pruned_loss=0.097, over 5695961.77 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3494, pruned_loss=0.1099, over 5696817.49 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3507, pruned_loss=0.09546, over 5694197.15 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 03:51:12,797 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-12 03:51:17,578 INFO [train.py:968] (1/2) Epoch 23, batch 36350, giga_loss[loss=0.2578, simple_loss=0.3483, pruned_loss=0.08364, over 28602.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3519, pruned_loss=0.09772, over 5686094.06 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3499, pruned_loss=0.11, over 5693464.43 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.351, pruned_loss=0.09598, over 5687676.77 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 03:51:25,068 INFO [optim.py:369] (1/2) 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 03:52:05,099 INFO [train.py:968] (1/2) Epoch 23, batch 36400, giga_loss[loss=0.2946, simple_loss=0.3623, pruned_loss=0.1134, over 28938.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3549, pruned_loss=0.102, over 5681651.10 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3499, pruned_loss=0.1099, over 5693298.27 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3543, pruned_loss=0.1004, over 5682709.33 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:52:06,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:07,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:11,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8635, 4.2461, 1.8021, 2.0147], device='cuda:1'), covar=tensor([0.0942, 0.0268, 0.0870, 0.1199], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0553, 0.0392, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 03:52:20,824 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1039807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:23,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1039810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:34,316 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1039824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 03:52:47,124 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 36450, libri_loss[loss=0.2694, simple_loss=0.3315, pruned_loss=0.1037, over 28646.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.357, pruned_loss=0.1052, over 5687291.09 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3501, pruned_loss=0.11, over 5692836.70 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3564, pruned_loss=0.1038, over 5688216.64 frames. ], batch size: 63, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:52:54,732 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5352, 1.7612, 1.4727, 1.6011], device='cuda:1'), covar=tensor([0.2535, 0.2504, 0.2793, 0.2269], device='cuda:1'), in_proj_covar=tensor([0.1528, 0.1102, 0.1352, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 03:52:55,658 INFO [optim.py:369] (1/2) 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 03:53:36,943 INFO [train.py:968] (1/2) Epoch 23, batch 36500, giga_loss[loss=0.265, simple_loss=0.3433, pruned_loss=0.09335, over 29059.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3573, pruned_loss=0.1071, over 5674813.04 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3504, pruned_loss=0.1101, over 5685986.23 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3566, pruned_loss=0.1058, over 5682234.32 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:54:21,357 INFO [train.py:968] (1/2) Epoch 23, batch 36550, giga_loss[loss=0.2698, simple_loss=0.3419, pruned_loss=0.09886, over 28853.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3545, pruned_loss=0.1055, over 5690942.32 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3507, pruned_loss=0.1101, over 5690046.01 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3539, pruned_loss=0.1043, over 5693079.97 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:54:22,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1039943.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 03:54:27,780 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 36600, libri_loss[loss=0.3589, simple_loss=0.4116, pruned_loss=0.1531, over 28627.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3529, pruned_loss=0.1046, over 5695413.65 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3507, pruned_loss=0.1101, over 5693515.90 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3524, pruned_loss=0.1036, over 5693978.64 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:55:50,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5838, 1.4180, 1.7549, 1.2946], device='cuda:1'), covar=tensor([0.2186, 0.3253, 0.1732, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0701, 0.0955, 0.0855], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 03:55:51,296 INFO [train.py:968] (1/2) Epoch 23, batch 36650, giga_loss[loss=0.2807, simple_loss=0.3539, pruned_loss=0.1037, over 28770.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3513, pruned_loss=0.1026, over 5694416.31 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3507, pruned_loss=0.11, over 5696027.85 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3509, pruned_loss=0.1018, over 5691158.44 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:55:59,607 INFO [optim.py:369] (1/2) 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 03:56:34,285 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.25 vs. limit=5.0
+2023-03-12 03:56:41,301 INFO [train.py:968] (1/2) Epoch 23, batch 36700, giga_loss[loss=0.2593, simple_loss=0.3282, pruned_loss=0.09522, over 26547.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3488, pruned_loss=0.1006, over 5692240.94 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3509, pruned_loss=0.11, over 5699229.52 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3483, pruned_loss=0.09983, over 5686891.94 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:56:52,021 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-12 03:57:22,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3408, 1.2846, 1.2942, 1.5160], device='cuda:1'), covar=tensor([0.0762, 0.0372, 0.0345, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 03:57:29,409 INFO [train.py:968] (1/2) Epoch 23, batch 36750, giga_loss[loss=0.221, simple_loss=0.3057, pruned_loss=0.06814, over 28770.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3428, pruned_loss=0.09714, over 5700157.80 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3508, pruned_loss=0.1099, over 5700288.37 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3424, pruned_loss=0.09655, over 5695066.66 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:57:35,836 INFO [optim.py:369] (1/2) 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 03:58:19,840 INFO [train.py:968] (1/2) Epoch 23, batch 36800, giga_loss[loss=0.2319, simple_loss=0.3123, pruned_loss=0.07578, over 29109.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3364, pruned_loss=0.09404, over 5686073.73 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3511, pruned_loss=0.11, over 5703365.30 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3357, pruned_loss=0.09314, over 5679269.75 frames. ], batch size: 113, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 03:59:11,773 INFO [train.py:968] (1/2) Epoch 23, batch 36850, libri_loss[loss=0.2862, simple_loss=0.3588, pruned_loss=0.1068, over 29552.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3325, pruned_loss=0.09193, over 5681798.69 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3514, pruned_loss=0.1102, over 5706134.95 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3314, pruned_loss=0.09084, over 5673683.86 frames. ], batch size: 83, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 03:59:18,746 INFO [optim.py:369] (1/2) 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 03:59:53,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4577, 4.3046, 4.1529, 2.6076], device='cuda:1'), covar=tensor([0.0658, 0.0808, 0.0835, 0.1591], device='cuda:1'), in_proj_covar=tensor([0.1230, 0.1136, 0.0961, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 03:59:54,838 INFO [train.py:968] (1/2) Epoch 23, batch 36900, giga_loss[loss=0.2435, simple_loss=0.324, pruned_loss=0.08153, over 28725.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3329, pruned_loss=0.09144, over 5675931.01 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.351, pruned_loss=0.1098, over 5698835.67 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3319, pruned_loss=0.0904, over 5676314.36 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:00:18,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1040318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:00:32,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6467, 4.4475, 4.2673, 1.8918], device='cuda:1'), covar=tensor([0.0580, 0.0773, 0.0694, 0.2151], device='cuda:1'), in_proj_covar=tensor([0.1230, 0.1136, 0.0961, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:00:34,496 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1040336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:00:39,170 INFO [train.py:968] (1/2) Epoch 23, batch 36950, giga_loss[loss=0.2197, simple_loss=0.303, pruned_loss=0.06822, over 28577.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.333, pruned_loss=0.09094, over 5689396.91 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3511, pruned_loss=0.1098, over 5700196.45 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3318, pruned_loss=0.08987, over 5688095.95 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:00:40,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4956, 1.5363, 1.6905, 1.3241], device='cuda:1'), covar=tensor([0.1673, 0.2431, 0.1367, 0.1618], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0702, 0.0956, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 04:00:46,931 INFO [optim.py:369] (1/2) 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 04:01:22,880 INFO [train.py:968] (1/2) Epoch 23, batch 37000, libri_loss[loss=0.3579, simple_loss=0.4134, pruned_loss=0.1512, over 20060.00 frames. ], tot_loss[loss=0.258, simple_loss=0.333, pruned_loss=0.09152, over 5683195.30 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3514, pruned_loss=0.1098, over 5696373.47 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3314, pruned_loss=0.09016, over 5686208.84 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:01:53,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4093, 1.2554, 4.3126, 3.3809], device='cuda:1'), covar=tensor([0.1500, 0.2665, 0.0404, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0763, 0.0651, 0.0964, 0.0912], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 04:02:04,077 INFO [train.py:968] (1/2) Epoch 23, batch 37050, giga_loss[loss=0.2525, simple_loss=0.3308, pruned_loss=0.08711, over 29009.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3317, pruned_loss=0.09086, over 5688186.66 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3521, pruned_loss=0.1099, over 5696743.03 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3289, pruned_loss=0.08901, over 5689594.34 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:02:11,096 INFO [optim.py:369] (1/2) 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 04:02:21,181 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1040461.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:02:23,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1040464.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:02:44,734 INFO [train.py:968] (1/2) Epoch 23, batch 37100, giga_loss[loss=0.2391, simple_loss=0.3199, pruned_loss=0.07914, over 28752.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3301, pruned_loss=0.09006, over 5704747.60 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3522, pruned_loss=0.1098, over 5702077.56 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3273, pruned_loss=0.08821, over 5700978.62 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:02:46,628 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1040493.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:02:56,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6261, 2.3178, 1.6885, 0.7383], device='cuda:1'), covar=tensor([0.6123, 0.2737, 0.4696, 0.6893], device='cuda:1'), in_proj_covar=tensor([0.1768, 0.1668, 0.1605, 0.1440], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 04:03:01,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1040511.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:03:04,969 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 04:03:10,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6779, 1.4772, 1.6707, 1.2949], device='cuda:1'), covar=tensor([0.2573, 0.3393, 0.1907, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0700, 0.0956, 0.0855], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 04:03:23,833 INFO [train.py:968] (1/2) Epoch 23, batch 37150, giga_loss[loss=0.2893, simple_loss=0.3553, pruned_loss=0.1116, over 27561.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3285, pruned_loss=0.08945, over 5712575.28 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3527, pruned_loss=0.1098, over 5705887.94 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3251, pruned_loss=0.08736, over 5706355.19 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:03:31,192 INFO [optim.py:369] (1/2) 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 04:04:07,273 INFO [train.py:968] (1/2) Epoch 23, batch 37200, giga_loss[loss=0.2352, simple_loss=0.3052, pruned_loss=0.08261, over 28657.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3273, pruned_loss=0.08925, over 5707849.62 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3535, pruned_loss=0.1101, over 5704727.00 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3235, pruned_loss=0.08698, over 5704168.58 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:04:48,773 INFO [train.py:968] (1/2) Epoch 23, batch 37250, giga_loss[loss=0.2239, simple_loss=0.3007, pruned_loss=0.07357, over 28708.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3251, pruned_loss=0.08827, over 5712420.91 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3539, pruned_loss=0.1103, over 5709643.18 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3212, pruned_loss=0.08582, over 5705263.28 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:04:56,718 INFO [optim.py:369] (1/2) 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 04:04:59,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9504, 1.2306, 1.2578, 1.0646], device='cuda:1'), covar=tensor([0.1679, 0.1319, 0.2191, 0.1557], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0753, 0.0721, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 04:05:29,258 INFO [train.py:968] (1/2) Epoch 23, batch 37300, giga_loss[loss=0.2141, simple_loss=0.2829, pruned_loss=0.07268, over 28611.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3236, pruned_loss=0.08765, over 5719160.72 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3548, pruned_loss=0.1108, over 5713227.37 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.319, pruned_loss=0.08488, over 5710260.20 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:05:45,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1040711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:05:47,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 04:05:54,583 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-12 04:06:09,463 INFO [train.py:968] (1/2) Epoch 23, batch 37350, giga_loss[loss=0.2752, simple_loss=0.3428, pruned_loss=0.1038, over 28614.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3224, pruned_loss=0.08706, over 5724898.39 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3553, pruned_loss=0.1109, over 5712877.20 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3177, pruned_loss=0.08435, over 5718507.95 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:06:17,686 INFO [optim.py:369] (1/2) 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 04:06:30,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6870, 1.8826, 1.7189, 1.6133], device='cuda:1'), covar=tensor([0.2200, 0.2582, 0.2637, 0.2619], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0754, 0.0721, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 04:06:40,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8982, 1.1120, 1.0461, 0.8283], device='cuda:1'), covar=tensor([0.2654, 0.2940, 0.1742, 0.2678], device='cuda:1'), in_proj_covar=tensor([0.1987, 0.1908, 0.1826, 0.1990], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 04:06:49,354 INFO [train.py:968] (1/2) Epoch 23, batch 37400, giga_loss[loss=0.2264, simple_loss=0.2973, pruned_loss=0.0777, over 28682.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3218, pruned_loss=0.08698, over 5727797.01 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3556, pruned_loss=0.1108, over 5713452.40 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3167, pruned_loss=0.08406, over 5722490.41 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:07:30,029 INFO [train.py:968] (1/2) Epoch 23, batch 37450, giga_loss[loss=0.2766, simple_loss=0.3423, pruned_loss=0.1054, over 28986.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3232, pruned_loss=0.08803, over 5727898.85 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3561, pruned_loss=0.111, over 5719627.15 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3176, pruned_loss=0.08479, over 5718518.34 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:07:38,480 INFO [optim.py:369] (1/2) 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 04:07:44,172 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1040854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:07:46,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1040857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:08:10,500 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1040886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:08:10,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1040886.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:08:13,720 INFO [train.py:968] (1/2) Epoch 23, batch 37500, libri_loss[loss=0.2891, simple_loss=0.3695, pruned_loss=0.1044, over 29191.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3269, pruned_loss=0.0901, over 5705655.97 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3563, pruned_loss=0.1111, over 5703927.66 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3214, pruned_loss=0.08686, over 5712893.62 frames. ], batch size: 101, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:08:15,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1031, 2.2466, 1.9029, 2.2366], device='cuda:1'), covar=tensor([0.2435, 0.2549, 0.2902, 0.2447], device='cuda:1'), in_proj_covar=tensor([0.1529, 0.1104, 0.1352, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 04:08:41,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3638, 1.6631, 1.5275, 1.2330], device='cuda:1'), covar=tensor([0.2238, 0.2006, 0.1412, 0.1848], device='cuda:1'), in_proj_covar=tensor([0.1982, 0.1903, 0.1824, 0.1985], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 04:08:43,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7974, 1.7489, 1.9415, 1.5475], device='cuda:1'), covar=tensor([0.1814, 0.2566, 0.1467, 0.1745], device='cuda:1'), in_proj_covar=tensor([0.0912, 0.0703, 0.0959, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 04:08:57,153 INFO [train.py:968] (1/2) Epoch 23, batch 37550, giga_loss[loss=0.2932, simple_loss=0.3684, pruned_loss=0.109, over 28751.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3321, pruned_loss=0.09299, over 5705776.83 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3565, pruned_loss=0.111, over 5708562.04 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3268, pruned_loss=0.08987, over 5707445.50 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:08:59,286 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5820, 1.7901, 1.7850, 1.3941], device='cuda:1'), covar=tensor([0.1560, 0.2442, 0.1388, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0703, 0.0959, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 04:09:06,431 INFO [optim.py:369] (1/2) 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 04:09:41,044 INFO [train.py:968] (1/2) Epoch 23, batch 37600, giga_loss[loss=0.2884, simple_loss=0.3617, pruned_loss=0.1075, over 28858.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3402, pruned_loss=0.09832, over 5703315.72 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3568, pruned_loss=0.111, over 5713794.71 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3346, pruned_loss=0.09505, over 5699309.51 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:09:50,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:10:18,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041029.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:10:20,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041032.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:10:30,923 INFO [train.py:968] (1/2) Epoch 23, batch 37650, giga_loss[loss=0.2565, simple_loss=0.3332, pruned_loss=0.08988, over 28439.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3448, pruned_loss=0.1011, over 5678532.13 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3574, pruned_loss=0.1116, over 5709542.04 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3394, pruned_loss=0.09764, over 5678242.39 frames. ], batch size: 78, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:10:41,990 INFO [optim.py:369] (1/2) 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 04:10:50,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9545, 1.1418, 1.1284, 0.9437], device='cuda:1'), covar=tensor([0.2289, 0.2641, 0.1599, 0.2161], device='cuda:1'), in_proj_covar=tensor([0.1977, 0.1899, 0.1822, 0.1979], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 04:10:51,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041061.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:11:20,802 INFO [train.py:968] (1/2) Epoch 23, batch 37700, giga_loss[loss=0.3281, simple_loss=0.3825, pruned_loss=0.1368, over 23647.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3498, pruned_loss=0.1032, over 5678159.06 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3573, pruned_loss=0.1115, over 5710591.33 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3456, pruned_loss=0.1006, over 5676783.85 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:11:57,425 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041133.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:12:06,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-12 04:12:08,217 INFO [train.py:968] (1/2) Epoch 23, batch 37750, giga_loss[loss=0.3273, simple_loss=0.3959, pruned_loss=0.1293, over 28644.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3546, pruned_loss=0.1061, over 5676221.02 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3575, pruned_loss=0.1119, over 5715278.42 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.351, pruned_loss=0.1035, over 5670303.60 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:12:10,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-12 04:12:17,177 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 37800, giga_loss[loss=0.3325, simple_loss=0.3729, pruned_loss=0.146, over 23907.00 frames. ], tot_loss[loss=0.2855, simple_loss=0.3567, pruned_loss=0.1072, over 5676487.33 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3573, pruned_loss=0.1117, over 5720079.63 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3541, pruned_loss=0.1051, over 5666473.85 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:13:32,370 INFO [train.py:968] (1/2) Epoch 23, batch 37850, giga_loss[loss=0.2409, simple_loss=0.3326, pruned_loss=0.0746, over 28940.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.352, pruned_loss=0.1032, over 5679595.15 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3578, pruned_loss=0.1122, over 5712213.16 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3493, pruned_loss=0.101, over 5677254.60 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:13:41,476 INFO [optim.py:369] (1/2) 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 04:13:51,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041264.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:13:56,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:14:00,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4292, 1.4304, 1.2462, 1.5590], device='cuda:1'), covar=tensor([0.0829, 0.0351, 0.0354, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 04:14:06,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1067, 5.9415, 5.6649, 3.1039], device='cuda:1'), covar=tensor([0.0407, 0.0568, 0.0600, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.1237, 0.1144, 0.0966, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:14:15,995 INFO [train.py:968] (1/2) Epoch 23, batch 37900, giga_loss[loss=0.2597, simple_loss=0.3392, pruned_loss=0.09008, over 28995.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3513, pruned_loss=0.1019, over 5686076.58 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3584, pruned_loss=0.1127, over 5713738.74 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3485, pruned_loss=0.09952, over 5682228.81 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:14:54,020 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:15:01,584 INFO [train.py:968] (1/2) Epoch 23, batch 37950, giga_loss[loss=0.2728, simple_loss=0.3531, pruned_loss=0.09621, over 28827.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3504, pruned_loss=0.1009, over 5675240.95 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3591, pruned_loss=0.1133, over 5705912.85 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3475, pruned_loss=0.09831, over 5679266.08 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:15:11,736 INFO [optim.py:369] (1/2) 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 04:15:29,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:15:42,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4789, 1.6812, 1.1629, 1.2737], device='cuda:1'), covar=tensor([0.1020, 0.0580, 0.1067, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0445, 0.0522, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:15:42,953 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-12 04:15:44,581 INFO [train.py:968] (1/2) Epoch 23, batch 38000, libri_loss[loss=0.3371, simple_loss=0.4005, pruned_loss=0.1368, over 29780.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3513, pruned_loss=0.1009, over 5671696.67 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5699155.58 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.348, pruned_loss=0.09815, over 5680074.56 frames. ], batch size: 87, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:16:26,405 INFO [train.py:968] (1/2) Epoch 23, batch 38050, giga_loss[loss=0.2968, simple_loss=0.3691, pruned_loss=0.1123, over 28563.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3529, pruned_loss=0.1023, over 5680019.16 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1136, over 5702908.62 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3504, pruned_loss=0.09986, over 5683129.31 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:16:29,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3791, 1.7489, 1.6434, 1.5362], device='cuda:1'), covar=tensor([0.2354, 0.1982, 0.2608, 0.2139], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0755, 0.0723, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 04:16:36,417 INFO [optim.py:369] (1/2) 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 04:17:08,268 INFO [train.py:968] (1/2) Epoch 23, batch 38100, libri_loss[loss=0.282, simple_loss=0.3518, pruned_loss=0.1061, over 29509.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3544, pruned_loss=0.1036, over 5679209.34 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3595, pruned_loss=0.1135, over 5696395.06 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3522, pruned_loss=0.1013, over 5686979.15 frames. ], batch size: 81, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:17:12,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-12 04:17:20,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:17:24,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041508.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:17:33,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:17:35,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:17:45,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6020, 5.4556, 5.1268, 2.6912], device='cuda:1'), covar=tensor([0.0412, 0.0547, 0.0570, 0.1715], device='cuda:1'), in_proj_covar=tensor([0.1235, 0.1144, 0.0967, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:17:52,030 INFO [train.py:968] (1/2) Epoch 23, batch 38150, giga_loss[loss=0.2952, simple_loss=0.3674, pruned_loss=0.1116, over 28610.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3553, pruned_loss=0.1045, over 5690468.28 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3593, pruned_loss=0.1135, over 5703658.88 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3536, pruned_loss=0.1024, over 5689697.87 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:18:01,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:18:03,321 INFO [optim.py:369] (1/2) 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 04:18:35,240 INFO [train.py:968] (1/2) Epoch 23, batch 38200, giga_loss[loss=0.2692, simple_loss=0.342, pruned_loss=0.09824, over 28830.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3565, pruned_loss=0.1057, over 5692517.71 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3592, pruned_loss=0.1133, over 5707557.26 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3551, pruned_loss=0.104, over 5688337.96 frames. ], batch size: 112, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:19:05,674 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1041624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:18,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041639.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:19:19,529 INFO [train.py:968] (1/2) Epoch 23, batch 38250, libri_loss[loss=0.3259, simple_loss=0.3905, pruned_loss=0.1307, over 29558.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.3573, pruned_loss=0.106, over 5694714.99 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3595, pruned_loss=0.1134, over 5702606.12 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3559, pruned_loss=0.1044, over 5696212.50 frames. ], batch size: 82, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:19:21,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:25,016 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:28,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:28,647 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041651.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:19:29,706 INFO [optim.py:369] (1/2) 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 04:19:30,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041654.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:19:48,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4666, 1.7266, 1.2325, 1.2768], device='cuda:1'), covar=tensor([0.1031, 0.0568, 0.1063, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0445, 0.0522, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:19:50,985 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:19:53,541 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 38300, giga_loss[loss=0.2684, simple_loss=0.3573, pruned_loss=0.0898, over 28980.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.3568, pruned_loss=0.1048, over 5690911.71 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3602, pruned_loss=0.114, over 5694256.29 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.355, pruned_loss=0.1028, over 5698940.20 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:20:16,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:20:45,573 INFO [train.py:968] (1/2) Epoch 23, batch 38350, giga_loss[loss=0.2727, simple_loss=0.3542, pruned_loss=0.09558, over 28617.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3569, pruned_loss=0.1036, over 5695781.79 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3607, pruned_loss=0.1142, over 5697266.41 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.355, pruned_loss=0.1017, over 5699455.19 frames. ], batch size: 78, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:20:55,616 INFO [optim.py:369] (1/2) 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 04:21:12,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5534, 1.5590, 1.7649, 1.3636], device='cuda:1'), covar=tensor([0.1839, 0.2729, 0.1522, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0705, 0.0958, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 04:21:18,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041782.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:21:21,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041785.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 04:21:22,268 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:21:26,314 INFO [train.py:968] (1/2) Epoch 23, batch 38400, giga_loss[loss=0.225, simple_loss=0.3113, pruned_loss=0.06938, over 28963.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3556, pruned_loss=0.1024, over 5705737.07 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3609, pruned_loss=0.1142, over 5701687.42 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3538, pruned_loss=0.1005, over 5704803.71 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:21:34,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3911, 1.7837, 1.1287, 1.2860], device='cuda:1'), covar=tensor([0.1379, 0.0772, 0.1637, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0446, 0.0523, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:21:43,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041814.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 04:21:44,516 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 04:21:46,228 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:02,232 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2662, 4.0736, 3.8559, 1.7607], device='cuda:1'), covar=tensor([0.0650, 0.0832, 0.0810, 0.2252], device='cuda:1'), in_proj_covar=tensor([0.1236, 0.1145, 0.0969, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:22:06,121 INFO [train.py:968] (1/2) Epoch 23, batch 38450, giga_loss[loss=0.2576, simple_loss=0.3347, pruned_loss=0.09023, over 28408.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3535, pruned_loss=0.102, over 5710997.84 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 5708781.58 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3519, pruned_loss=0.1001, over 5703854.25 frames. ], batch size: 65, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:22:12,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5173, 1.6057, 1.6038, 1.4082], device='cuda:1'), covar=tensor([0.2382, 0.2460, 0.1756, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1983, 0.1909, 0.1835, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 04:22:12,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1041850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:15,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1041853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:15,383 INFO [optim.py:369] (1/2) 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 04:22:38,966 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1041882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:22:46,740 INFO [train.py:968] (1/2) Epoch 23, batch 38500, giga_loss[loss=0.2515, simple_loss=0.3322, pruned_loss=0.08536, over 28708.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3512, pruned_loss=0.1006, over 5707747.42 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3609, pruned_loss=0.1142, over 5699290.35 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3496, pruned_loss=0.09885, over 5709760.13 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:23:26,595 INFO [train.py:968] (1/2) Epoch 23, batch 38550, giga_loss[loss=0.2768, simple_loss=0.3538, pruned_loss=0.0999, over 28709.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3496, pruned_loss=0.09974, over 5713587.49 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.361, pruned_loss=0.1142, over 5702970.28 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.348, pruned_loss=0.0981, over 5712274.55 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:23:34,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2996, 2.3795, 1.9710, 2.0818], device='cuda:1'), covar=tensor([0.0601, 0.0415, 0.0676, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0445, 0.0522, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:23:39,473 INFO [optim.py:369] (1/2) 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 04:23:54,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9030, 3.7167, 3.5474, 1.6884], device='cuda:1'), covar=tensor([0.0675, 0.0868, 0.0777, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1140, 0.0966, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:24:08,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-12 04:24:08,809 INFO [train.py:968] (1/2) Epoch 23, batch 38600, giga_loss[loss=0.2846, simple_loss=0.356, pruned_loss=0.1066, over 28212.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3513, pruned_loss=0.1018, over 5705614.28 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1144, over 5696776.98 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3495, pruned_loss=0.09994, over 5709687.53 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:24:15,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1041999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:24:36,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5081, 4.3511, 4.1025, 2.2081], device='cuda:1'), covar=tensor([0.0494, 0.0626, 0.0668, 0.2107], device='cuda:1'), in_proj_covar=tensor([0.1236, 0.1140, 0.0966, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:24:48,063 INFO [train.py:968] (1/2) Epoch 23, batch 38650, giga_loss[loss=0.2617, simple_loss=0.3403, pruned_loss=0.09156, over 28876.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3511, pruned_loss=0.1009, over 5708364.90 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3616, pruned_loss=0.1144, over 5696821.77 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.09916, over 5711891.76 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:24:58,441 INFO [optim.py:369] (1/2) 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 04:25:06,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8994, 1.9125, 2.0777, 1.6437], device='cuda:1'), covar=tensor([0.1959, 0.2604, 0.1561, 0.1846], device='cuda:1'), in_proj_covar=tensor([0.0912, 0.0706, 0.0958, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 04:25:27,848 INFO [train.py:968] (1/2) Epoch 23, batch 38700, giga_loss[loss=0.2388, simple_loss=0.3196, pruned_loss=0.07901, over 28970.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.349, pruned_loss=0.09885, over 5704739.27 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3611, pruned_loss=0.1141, over 5699255.30 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09743, over 5705591.22 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:25:33,305 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 04:25:51,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1042121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:08,944 INFO [train.py:968] (1/2) Epoch 23, batch 38750, giga_loss[loss=0.2701, simple_loss=0.3479, pruned_loss=0.09613, over 28664.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3479, pruned_loss=0.09799, over 5715030.63 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3608, pruned_loss=0.1138, over 5704363.28 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.347, pruned_loss=0.09676, over 5711368.07 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:26:09,938 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1042142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:11,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1042145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:19,746 INFO [optim.py:369] (1/2) 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 04:26:38,305 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1042174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:26:41,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3840, 3.2696, 1.4968, 1.5293], device='cuda:1'), covar=tensor([0.1073, 0.0289, 0.0944, 0.1439], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0550, 0.0390, 0.0428], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 04:26:50,820 INFO [train.py:968] (1/2) Epoch 23, batch 38800, giga_loss[loss=0.2432, simple_loss=0.3196, pruned_loss=0.08342, over 28545.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.348, pruned_loss=0.09861, over 5704930.63 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.361, pruned_loss=0.1141, over 5696743.78 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3469, pruned_loss=0.09729, over 5709760.61 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:27:31,667 INFO [train.py:968] (1/2) Epoch 23, batch 38850, giga_loss[loss=0.2699, simple_loss=0.339, pruned_loss=0.1004, over 28300.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3472, pruned_loss=0.09914, over 5698297.07 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5693924.70 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3452, pruned_loss=0.09697, over 5704839.86 frames. ], batch size: 65, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:27:42,235 INFO [optim.py:369] (1/2) 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 04:27:42,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1042255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:28:10,120 INFO [train.py:968] (1/2) Epoch 23, batch 38900, giga_loss[loss=0.2547, simple_loss=0.3314, pruned_loss=0.08899, over 28931.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.344, pruned_loss=0.09777, over 5701201.50 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1146, over 5694422.90 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3423, pruned_loss=0.09573, over 5705987.56 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:28:49,364 INFO [train.py:968] (1/2) Epoch 23, batch 38950, giga_loss[loss=0.2187, simple_loss=0.301, pruned_loss=0.06822, over 28471.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3436, pruned_loss=0.09748, over 5706728.06 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3621, pruned_loss=0.115, over 5701143.83 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3411, pruned_loss=0.09497, over 5704652.44 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:28:52,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-12 04:28:59,678 INFO [optim.py:369] (1/2) 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 04:29:31,995 INFO [train.py:968] (1/2) Epoch 23, batch 39000, giga_loss[loss=0.242, simple_loss=0.3176, pruned_loss=0.08323, over 28992.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3448, pruned_loss=0.09898, over 5696551.81 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5697794.42 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3419, pruned_loss=0.09623, over 5697424.15 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:29:31,996 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 04:29:40,812 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 04:29:44,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4443, 1.5708, 1.2268, 1.5805], device='cuda:1'), covar=tensor([0.0748, 0.0324, 0.0366, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0117, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 04:30:07,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7308, 1.8483, 1.6894, 1.5318], device='cuda:1'), covar=tensor([0.2939, 0.2591, 0.2285, 0.2629], device='cuda:1'), in_proj_covar=tensor([0.1987, 0.1920, 0.1846, 0.1994], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 04:30:21,703 INFO [train.py:968] (1/2) Epoch 23, batch 39050, giga_loss[loss=0.2696, simple_loss=0.3398, pruned_loss=0.09975, over 28955.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3416, pruned_loss=0.09716, over 5696065.01 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5693836.43 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3387, pruned_loss=0.0944, over 5700747.13 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:30:27,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9084, 1.2241, 1.3088, 0.9795], device='cuda:1'), covar=tensor([0.1913, 0.1327, 0.2230, 0.1769], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0754, 0.0722, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 04:30:33,157 INFO [optim.py:369] (1/2) 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 04:31:01,237 INFO [train.py:968] (1/2) Epoch 23, batch 39100, giga_loss[loss=0.2433, simple_loss=0.316, pruned_loss=0.0853, over 28996.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3395, pruned_loss=0.09644, over 5699255.24 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3623, pruned_loss=0.1153, over 5689996.20 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3364, pruned_loss=0.09346, over 5706467.28 frames. ], batch size: 106, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:31:06,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1042496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:31:40,266 INFO [train.py:968] (1/2) Epoch 23, batch 39150, giga_loss[loss=0.2363, simple_loss=0.3049, pruned_loss=0.08385, over 28673.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3377, pruned_loss=0.09583, over 5701929.24 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3617, pruned_loss=0.1148, over 5692311.88 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3348, pruned_loss=0.09312, over 5705971.51 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:31:53,406 INFO [optim.py:369] (1/2) 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 04:31:58,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5575, 1.7220, 1.4338, 1.5659], device='cuda:1'), covar=tensor([0.2930, 0.2991, 0.3358, 0.2790], device='cuda:1'), in_proj_covar=tensor([0.1533, 0.1106, 0.1353, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 04:32:04,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7220, 4.5448, 4.2992, 2.0423], device='cuda:1'), covar=tensor([0.0459, 0.0601, 0.0703, 0.2118], device='cuda:1'), in_proj_covar=tensor([0.1235, 0.1142, 0.0964, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:32:15,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-12 04:32:24,337 INFO [train.py:968] (1/2) Epoch 23, batch 39200, giga_loss[loss=0.2355, simple_loss=0.3232, pruned_loss=0.07385, over 28937.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3347, pruned_loss=0.09379, over 5706418.01 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1144, over 5696816.52 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3325, pruned_loss=0.09168, over 5705607.88 frames. ], batch size: 164, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:32:26,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3519, 1.7046, 0.9596, 1.2548], device='cuda:1'), covar=tensor([0.1312, 0.0838, 0.1740, 0.1472], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0445, 0.0521, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:32:58,330 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1042630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:07,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1042639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:08,179 INFO [train.py:968] (1/2) Epoch 23, batch 39250, libri_loss[loss=0.2591, simple_loss=0.3242, pruned_loss=0.09701, over 29368.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3353, pruned_loss=0.09381, over 5698858.17 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3609, pruned_loss=0.1143, over 5691778.71 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3332, pruned_loss=0.09172, over 5702045.36 frames. ], batch size: 67, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:33:09,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1042642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:20,280 INFO [optim.py:369] (1/2) 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 04:33:36,508 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1042671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:33:50,403 INFO [train.py:968] (1/2) Epoch 23, batch 39300, giga_loss[loss=0.3057, simple_loss=0.3747, pruned_loss=0.1184, over 29050.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3381, pruned_loss=0.09536, over 5700660.25 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3604, pruned_loss=0.1142, over 5699303.04 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3359, pruned_loss=0.09298, over 5696651.75 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:34:10,175 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 04:34:36,369 INFO [train.py:968] (1/2) Epoch 23, batch 39350, giga_loss[loss=0.2941, simple_loss=0.3763, pruned_loss=0.106, over 28429.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3427, pruned_loss=0.09751, over 5692175.79 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1141, over 5695762.77 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3402, pruned_loss=0.09504, over 5692643.34 frames. ], batch size: 368, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:34:48,306 INFO [optim.py:369] (1/2) 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 04:35:00,828 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-12 04:35:04,332 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1042773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:35:06,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1042776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:35:10,227 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1042778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:35:10,393 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 04:35:22,013 INFO [train.py:968] (1/2) Epoch 23, batch 39400, giga_loss[loss=0.2448, simple_loss=0.3289, pruned_loss=0.08032, over 28898.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.345, pruned_loss=0.09835, over 5689996.75 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3603, pruned_loss=0.1141, over 5698026.92 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3428, pruned_loss=0.09611, over 5688497.88 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:35:34,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1042805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:36:03,085 INFO [train.py:968] (1/2) Epoch 23, batch 39450, giga_loss[loss=0.2513, simple_loss=0.3255, pruned_loss=0.08857, over 28685.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.344, pruned_loss=0.0969, over 5696986.77 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1143, over 5698678.55 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3416, pruned_loss=0.09462, over 5695359.69 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:36:17,077 INFO [optim.py:369] (1/2) 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 04:36:46,119 INFO [train.py:968] (1/2) Epoch 23, batch 39500, giga_loss[loss=0.2498, simple_loss=0.3279, pruned_loss=0.08581, over 28905.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3426, pruned_loss=0.09615, over 5700521.42 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.114, over 5700604.60 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3408, pruned_loss=0.0943, over 5697564.05 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:37:29,851 INFO [train.py:968] (1/2) Epoch 23, batch 39550, giga_loss[loss=0.2649, simple_loss=0.3328, pruned_loss=0.0985, over 28647.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.343, pruned_loss=0.09665, over 5701231.50 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.114, over 5696149.78 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09478, over 5701955.56 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:37:39,992 INFO [optim.py:369] (1/2) 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 04:38:07,899 INFO [train.py:968] (1/2) Epoch 23, batch 39600, giga_loss[loss=0.2865, simple_loss=0.3653, pruned_loss=0.1038, over 27969.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3441, pruned_loss=0.09764, over 5718491.92 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.114, over 5705901.47 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3416, pruned_loss=0.09525, over 5710901.20 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:38:27,887 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 04:38:53,887 INFO [train.py:968] (1/2) Epoch 23, batch 39650, giga_loss[loss=0.327, simple_loss=0.3822, pruned_loss=0.1359, over 26698.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3467, pruned_loss=0.09854, over 5714713.53 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3604, pruned_loss=0.114, over 5706998.96 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3447, pruned_loss=0.09666, over 5707879.64 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:38:57,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 04:39:04,654 INFO [optim.py:369] (1/2) 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 04:39:19,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 04:39:34,921 INFO [train.py:968] (1/2) Epoch 23, batch 39700, giga_loss[loss=0.28, simple_loss=0.3591, pruned_loss=0.1005, over 28554.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3492, pruned_loss=0.09948, over 5710246.39 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3603, pruned_loss=0.1136, over 5704601.32 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3473, pruned_loss=0.09781, over 5706511.00 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:39:57,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-12 04:40:17,588 INFO [train.py:968] (1/2) Epoch 23, batch 39750, giga_loss[loss=0.2944, simple_loss=0.3673, pruned_loss=0.1108, over 28048.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.0998, over 5702501.24 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3603, pruned_loss=0.1137, over 5695439.32 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3489, pruned_loss=0.09841, over 5707464.09 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:40:29,110 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:40:59,687 INFO [train.py:968] (1/2) Epoch 23, batch 39800, giga_loss[loss=0.2901, simple_loss=0.3663, pruned_loss=0.107, over 29028.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3519, pruned_loss=0.1009, over 5707629.24 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3606, pruned_loss=0.1138, over 5701888.27 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3502, pruned_loss=0.09923, over 5705995.31 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:41:42,389 INFO [train.py:968] (1/2) Epoch 23, batch 39850, libri_loss[loss=0.2911, simple_loss=0.3702, pruned_loss=0.106, over 29240.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3524, pruned_loss=0.1013, over 5710999.01 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3605, pruned_loss=0.1136, over 5705488.27 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.351, pruned_loss=0.09991, over 5706562.44 frames. ], batch size: 97, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 04:41:52,789 INFO [optim.py:369] (1/2) 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 04:42:21,562 INFO [train.py:968] (1/2) Epoch 23, batch 39900, giga_loss[loss=0.2357, simple_loss=0.3165, pruned_loss=0.07743, over 28990.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1012, over 5713584.67 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3608, pruned_loss=0.1137, over 5707400.28 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3505, pruned_loss=0.09965, over 5708401.38 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:42:24,925 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1043296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:42:26,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1043299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:42:52,077 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1043328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:43:02,521 INFO [train.py:968] (1/2) Epoch 23, batch 39950, giga_loss[loss=0.2461, simple_loss=0.3328, pruned_loss=0.07975, over 28755.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.349, pruned_loss=0.09985, over 5690906.40 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3611, pruned_loss=0.114, over 5678857.82 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3473, pruned_loss=0.09823, over 5713098.38 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:43:16,458 INFO [optim.py:369] (1/2) 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 04:43:25,698 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1043368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:43:40,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 04:43:44,904 INFO [train.py:968] (1/2) Epoch 23, batch 40000, giga_loss[loss=0.2466, simple_loss=0.3247, pruned_loss=0.08429, over 28920.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3451, pruned_loss=0.09784, over 5697425.25 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3615, pruned_loss=0.1143, over 5680992.86 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3433, pruned_loss=0.09618, over 5713050.68 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:43:52,355 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 04:43:54,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5631, 1.7625, 1.4447, 1.8123], device='cuda:1'), covar=tensor([0.2685, 0.2783, 0.3152, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.1523, 0.1099, 0.1346, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 04:44:07,880 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.06 vs. limit=5.0
+2023-03-12 04:44:19,258 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1043433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:44:26,953 INFO [train.py:968] (1/2) Epoch 23, batch 40050, giga_loss[loss=0.2189, simple_loss=0.3029, pruned_loss=0.06741, over 28857.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3439, pruned_loss=0.09679, over 5706058.19 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5688970.83 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3421, pruned_loss=0.09507, over 5712199.03 frames. ], batch size: 112, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:44:38,619 INFO [optim.py:369] (1/2) 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 04:44:56,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-12 04:45:06,899 INFO [train.py:968] (1/2) Epoch 23, batch 40100, giga_loss[loss=0.2861, simple_loss=0.37, pruned_loss=0.1011, over 27662.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3457, pruned_loss=0.09579, over 5714130.20 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5691958.18 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3439, pruned_loss=0.09423, over 5716766.97 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:45:44,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 04:45:50,511 INFO [train.py:968] (1/2) Epoch 23, batch 40150, giga_loss[loss=0.2503, simple_loss=0.3348, pruned_loss=0.0829, over 28942.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3454, pruned_loss=0.09525, over 5711073.81 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3613, pruned_loss=0.1139, over 5697415.71 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3437, pruned_loss=0.09373, over 5708823.11 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:46:04,225 INFO [optim.py:369] (1/2) 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 04:46:32,386 INFO [train.py:968] (1/2) Epoch 23, batch 40200, giga_loss[loss=0.2968, simple_loss=0.3736, pruned_loss=0.11, over 27938.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3436, pruned_loss=0.0955, over 5715966.61 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1135, over 5704401.45 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3425, pruned_loss=0.09406, over 5708479.39 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:46:38,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5379, 1.6080, 1.2359, 1.2139], device='cuda:1'), covar=tensor([0.0941, 0.0597, 0.1010, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0445, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:47:05,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3559, 1.2116, 1.1713, 1.4733], device='cuda:1'), covar=tensor([0.0738, 0.0351, 0.0353, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 04:47:07,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8302, 1.3307, 5.3139, 3.7372], device='cuda:1'), covar=tensor([0.1551, 0.2712, 0.0422, 0.0962], device='cuda:1'), in_proj_covar=tensor([0.0764, 0.0650, 0.0966, 0.0915], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 04:47:13,247 INFO [train.py:968] (1/2) Epoch 23, batch 40250, giga_loss[loss=0.2943, simple_loss=0.3584, pruned_loss=0.1152, over 28798.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3443, pruned_loss=0.09723, over 5718201.96 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3606, pruned_loss=0.1135, over 5708408.77 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3428, pruned_loss=0.0957, over 5708687.92 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:47:28,317 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 40300, giga_loss[loss=0.2456, simple_loss=0.316, pruned_loss=0.08763, over 29003.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3432, pruned_loss=0.09784, over 5713818.91 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3603, pruned_loss=0.1132, over 5712084.14 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.342, pruned_loss=0.09654, over 5703126.51 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:48:40,568 INFO [train.py:968] (1/2) Epoch 23, batch 40350, giga_loss[loss=0.2872, simple_loss=0.3555, pruned_loss=0.1095, over 28576.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3413, pruned_loss=0.09759, over 5722743.39 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1133, over 5714919.58 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3401, pruned_loss=0.0963, over 5711927.44 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:48:43,070 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:49:22,821 INFO [train.py:968] (1/2) Epoch 23, batch 40400, libri_loss[loss=0.2965, simple_loss=0.3693, pruned_loss=0.1118, over 29483.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3394, pruned_loss=0.09664, over 5726767.57 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.113, over 5719881.88 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3382, pruned_loss=0.09539, over 5713711.64 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:49:28,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-12 04:49:36,559 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:49:42,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 04:50:03,270 INFO [train.py:968] (1/2) Epoch 23, batch 40450, giga_loss[loss=0.2358, simple_loss=0.3121, pruned_loss=0.07975, over 28700.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3359, pruned_loss=0.09462, over 5727502.85 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 5721561.62 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.334, pruned_loss=0.093, over 5715767.81 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:50:08,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1043848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:50:18,404 INFO [optim.py:369] (1/2) 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 04:50:40,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1043886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:50:42,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1043889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:50:43,429 INFO [train.py:968] (1/2) Epoch 23, batch 40500, giga_loss[loss=0.2384, simple_loss=0.3176, pruned_loss=0.07963, over 28605.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3332, pruned_loss=0.09356, over 5723799.71 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1136, over 5716209.44 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3302, pruned_loss=0.09123, over 5718602.42 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:51:05,280 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1043918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:51:24,338 INFO [train.py:968] (1/2) Epoch 23, batch 40550, giga_loss[loss=0.2139, simple_loss=0.2948, pruned_loss=0.06655, over 29038.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3305, pruned_loss=0.09215, over 5722565.52 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3611, pruned_loss=0.1138, over 5719603.79 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3271, pruned_loss=0.08955, over 5715314.85 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:51:32,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1043951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:51:36,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1043954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:51:40,474 INFO [optim.py:369] (1/2) 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 04:51:58,599 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1043983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:52:05,278 INFO [train.py:968] (1/2) Epoch 23, batch 40600, giga_loss[loss=0.2997, simple_loss=0.3704, pruned_loss=0.1145, over 28926.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3334, pruned_loss=0.09362, over 5711434.95 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5715981.94 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3291, pruned_loss=0.0904, over 5708817.14 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:52:44,721 INFO [train.py:968] (1/2) Epoch 23, batch 40650, giga_loss[loss=0.2964, simple_loss=0.3721, pruned_loss=0.1103, over 28589.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3372, pruned_loss=0.09537, over 5698204.63 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5700960.66 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3326, pruned_loss=0.09194, over 5710494.91 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:53:03,339 INFO [optim.py:369] (1/2) 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 04:53:29,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5644, 1.6592, 1.7888, 1.3604], device='cuda:1'), covar=tensor([0.1967, 0.2594, 0.1581, 0.1840], device='cuda:1'), in_proj_covar=tensor([0.0909, 0.0703, 0.0954, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 04:53:30,227 INFO [train.py:968] (1/2) Epoch 23, batch 40700, giga_loss[loss=0.2429, simple_loss=0.328, pruned_loss=0.07889, over 29058.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3395, pruned_loss=0.09573, over 5697064.48 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1143, over 5699188.14 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3356, pruned_loss=0.09279, over 5708455.98 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:53:41,959 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:53:44,448 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:54:10,370 INFO [train.py:968] (1/2) Epoch 23, batch 40750, giga_loss[loss=0.2714, simple_loss=0.3495, pruned_loss=0.09663, over 28877.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3411, pruned_loss=0.09582, over 5699721.97 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3617, pruned_loss=0.1142, over 5692213.53 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3377, pruned_loss=0.09314, over 5714951.39 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 04:54:13,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1233, 1.3443, 1.1673, 0.9270], device='cuda:1'), covar=tensor([0.1021, 0.0518, 0.1057, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0446, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:54:28,950 INFO [optim.py:369] (1/2) 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 04:54:53,020 INFO [train.py:968] (1/2) Epoch 23, batch 40800, giga_loss[loss=0.2947, simple_loss=0.367, pruned_loss=0.1112, over 27721.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3435, pruned_loss=0.09713, over 5696430.67 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5687109.61 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3401, pruned_loss=0.09438, over 5713649.33 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:55:22,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:55:36,843 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:55:37,230 INFO [train.py:968] (1/2) Epoch 23, batch 40850, giga_loss[loss=0.2615, simple_loss=0.3416, pruned_loss=0.09075, over 28933.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3445, pruned_loss=0.09798, over 5691094.11 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5686892.21 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3416, pruned_loss=0.09568, over 5704931.89 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:56:02,850 INFO [optim.py:369] (1/2) 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 04:56:36,175 INFO [train.py:968] (1/2) Epoch 23, batch 40900, giga_loss[loss=0.3043, simple_loss=0.3722, pruned_loss=0.1182, over 28737.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3534, pruned_loss=0.1056, over 5671861.43 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5686892.21 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3511, pruned_loss=0.1038, over 5682631.55 frames. ], batch size: 243, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:56:47,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3270, 3.5101, 1.5142, 1.5762], device='cuda:1'), covar=tensor([0.1066, 0.0393, 0.0924, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0555, 0.0391, 0.0429], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 04:57:24,967 INFO [train.py:968] (1/2) Epoch 23, batch 40950, giga_loss[loss=0.3254, simple_loss=0.3899, pruned_loss=0.1304, over 28951.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.36, pruned_loss=0.1103, over 5674622.93 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3616, pruned_loss=0.1141, over 5690065.44 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3581, pruned_loss=0.1088, over 5680138.45 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:57:38,989 INFO [optim.py:369] (1/2) 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 04:57:47,382 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:57:49,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:57:51,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5916, 1.7604, 1.3177, 1.2930], device='cuda:1'), covar=tensor([0.0975, 0.0572, 0.1022, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0446, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 04:58:09,107 INFO [train.py:968] (1/2) Epoch 23, batch 41000, libri_loss[loss=0.3186, simple_loss=0.3851, pruned_loss=0.1261, over 29538.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3654, pruned_loss=0.1147, over 5657225.37 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5683055.35 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3639, pruned_loss=0.1133, over 5666841.88 frames. ], batch size: 79, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:58:14,333 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:58:38,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5688, 2.2726, 1.7387, 0.8562], device='cuda:1'), covar=tensor([0.5778, 0.3020, 0.3909, 0.5976], device='cuda:1'), in_proj_covar=tensor([0.1764, 0.1667, 0.1608, 0.1435], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 04:58:50,668 INFO [train.py:968] (1/2) Epoch 23, batch 41050, giga_loss[loss=0.34, simple_loss=0.3953, pruned_loss=0.1423, over 28799.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3723, pruned_loss=0.1206, over 5660297.13 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1145, over 5680369.37 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3708, pruned_loss=0.1192, over 5670073.03 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 04:59:09,580 INFO [optim.py:369] (1/2) 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 04:59:19,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9880, 1.2573, 1.3205, 1.0959], device='cuda:1'), covar=tensor([0.1436, 0.1116, 0.1826, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0752, 0.0720, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 04:59:28,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4103, 4.2642, 4.0517, 1.8675], device='cuda:1'), covar=tensor([0.0644, 0.0820, 0.0836, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1244, 0.1151, 0.0972, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 04:59:30,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:59:36,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 04:59:41,838 INFO [train.py:968] (1/2) Epoch 23, batch 41100, giga_loss[loss=0.3841, simple_loss=0.4064, pruned_loss=0.1809, over 23761.00 frames. ], tot_loss[loss=0.3139, simple_loss=0.3778, pruned_loss=0.125, over 5655541.48 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1143, over 5682538.66 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3771, pruned_loss=0.1242, over 5661168.05 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:00:33,404 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4700, 4.3132, 4.1200, 1.8596], device='cuda:1'), covar=tensor([0.0618, 0.0769, 0.0799, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.1245, 0.1150, 0.0972, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 05:00:34,409 INFO [train.py:968] (1/2) Epoch 23, batch 41150, giga_loss[loss=0.2669, simple_loss=0.3507, pruned_loss=0.09155, over 28816.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3792, pruned_loss=0.1267, over 5650026.05 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3618, pruned_loss=0.1144, over 5674034.74 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3787, pruned_loss=0.126, over 5661562.43 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:00:55,490 INFO [optim.py:369] (1/2) 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 05:01:17,978 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8634, 1.5496, 5.3045, 3.9769], device='cuda:1'), covar=tensor([0.1715, 0.2820, 0.0371, 0.0722], device='cuda:1'), in_proj_covar=tensor([0.0768, 0.0653, 0.0972, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 05:01:18,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:01:33,088 INFO [train.py:968] (1/2) Epoch 23, batch 41200, giga_loss[loss=0.397, simple_loss=0.4132, pruned_loss=0.1904, over 23510.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3828, pruned_loss=0.1312, over 5626816.70 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1148, over 5681001.42 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3825, pruned_loss=0.1307, over 5628977.56 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:01:59,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:08,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:10,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:11,835 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:14,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:26,206 INFO [train.py:968] (1/2) Epoch 23, batch 41250, giga_loss[loss=0.3548, simple_loss=0.406, pruned_loss=0.1518, over 28709.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3864, pruned_loss=0.1352, over 5618946.67 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1149, over 5679305.13 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3864, pruned_loss=0.135, over 5621617.17 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:02:40,534 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:49,920 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:02:50,298 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 23, batch 41300, giga_loss[loss=0.3278, simple_loss=0.3913, pruned_loss=0.1322, over 28709.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3898, pruned_loss=0.1376, over 5610271.90 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3621, pruned_loss=0.1148, over 5664047.63 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3903, pruned_loss=0.1378, over 5624208.64 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:04:13,240 INFO [train.py:968] (1/2) Epoch 23, batch 41350, giga_loss[loss=0.3206, simple_loss=0.3836, pruned_loss=0.1288, over 28844.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3916, pruned_loss=0.1397, over 5615992.30 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3624, pruned_loss=0.1151, over 5660427.62 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3924, pruned_loss=0.1401, over 5629620.70 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:04:33,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1044758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:04:35,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1044761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:04:35,944 INFO [optim.py:369] (1/2) 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 05:05:03,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1044790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:05:03,450 INFO [train.py:968] (1/2) Epoch 23, batch 41400, giga_loss[loss=0.3571, simple_loss=0.4078, pruned_loss=0.1532, over 27849.00 frames. ], tot_loss[loss=0.332, simple_loss=0.3882, pruned_loss=0.1379, over 5623620.11 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3622, pruned_loss=0.115, over 5666163.69 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3897, pruned_loss=0.1389, over 5628330.29 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:05:18,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5166, 1.2240, 4.2823, 3.5415], device='cuda:1'), covar=tensor([0.1520, 0.2676, 0.0442, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0659, 0.0980, 0.0927], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 05:05:55,191 INFO [train.py:968] (1/2) Epoch 23, batch 41450, giga_loss[loss=0.3094, simple_loss=0.3764, pruned_loss=0.1212, over 28713.00 frames. ], tot_loss[loss=0.3327, simple_loss=0.3885, pruned_loss=0.1384, over 5624841.99 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.115, over 5672578.38 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3905, pruned_loss=0.14, over 5621575.36 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:06:06,927 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1044853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:06:17,677 INFO [optim.py:369] (1/2) 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 05:06:30,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.95 vs. limit=2.0
+2023-03-12 05:06:53,461 INFO [train.py:968] (1/2) Epoch 23, batch 41500, giga_loss[loss=0.3855, simple_loss=0.4142, pruned_loss=0.1784, over 26532.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3876, pruned_loss=0.137, over 5619851.92 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3617, pruned_loss=0.1146, over 5675881.18 frames. ], giga_tot_loss[loss=0.334, simple_loss=0.3901, pruned_loss=0.1389, over 5613782.15 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:06:59,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6785, 1.9449, 1.8723, 1.5887], device='cuda:1'), covar=tensor([0.2541, 0.2081, 0.1798, 0.2129], device='cuda:1'), in_proj_covar=tensor([0.2001, 0.1943, 0.1864, 0.2007], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 05:07:47,276 INFO [train.py:968] (1/2) Epoch 23, batch 41550, giga_loss[loss=0.346, simple_loss=0.3795, pruned_loss=0.1562, over 23236.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3903, pruned_loss=0.1391, over 5600655.02 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5678125.95 frames. ], giga_tot_loss[loss=0.3376, simple_loss=0.3928, pruned_loss=0.1412, over 5592591.89 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:08:02,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1044954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:08:09,810 INFO [optim.py:369] (1/2) 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 05:08:10,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 05:08:43,826 INFO [train.py:968] (1/2) Epoch 23, batch 41600, giga_loss[loss=0.3169, simple_loss=0.3714, pruned_loss=0.1312, over 28052.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.389, pruned_loss=0.1379, over 5607633.86 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1146, over 5681302.68 frames. ], giga_tot_loss[loss=0.3354, simple_loss=0.3913, pruned_loss=0.1398, over 5597437.20 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:09:28,985 INFO [train.py:968] (1/2) Epoch 23, batch 41650, libri_loss[loss=0.2628, simple_loss=0.33, pruned_loss=0.09774, over 29567.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3846, pruned_loss=0.1329, over 5628548.42 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1147, over 5689094.83 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3877, pruned_loss=0.1353, over 5609829.96 frames. ], batch size: 78, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:09:29,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1427, 1.1784, 3.3328, 2.9739], device='cuda:1'), covar=tensor([0.1667, 0.2816, 0.0554, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0660, 0.0978, 0.0926], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 05:09:48,338 INFO [optim.py:369] (1/2) 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 05:09:50,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045063.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:09:51,121 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:10:16,363 INFO [train.py:968] (1/2) Epoch 23, batch 41700, libri_loss[loss=0.2942, simple_loss=0.3688, pruned_loss=0.1098, over 29673.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3812, pruned_loss=0.1292, over 5644264.89 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1145, over 5694747.43 frames. ], giga_tot_loss[loss=0.3241, simple_loss=0.3846, pruned_loss=0.1318, over 5622213.06 frames. ], batch size: 91, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:10:23,044 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:10:25,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:10:59,433 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:11:10,075 INFO [train.py:968] (1/2) Epoch 23, batch 41750, giga_loss[loss=0.3343, simple_loss=0.3806, pruned_loss=0.144, over 26580.00 frames. ], tot_loss[loss=0.3169, simple_loss=0.3787, pruned_loss=0.1275, over 5640687.55 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5696901.63 frames. ], giga_tot_loss[loss=0.3209, simple_loss=0.3819, pruned_loss=0.1299, over 5620951.95 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:11:33,882 INFO [optim.py:369] (1/2) 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 05:11:49,834 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:11:50,399 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 41800, giga_loss[loss=0.3058, simple_loss=0.3624, pruned_loss=0.1246, over 27540.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3757, pruned_loss=0.1249, over 5645250.49 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1144, over 5701957.70 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3789, pruned_loss=0.1272, over 5622863.28 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:12:35,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:12:44,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:12:50,564 INFO [train.py:968] (1/2) Epoch 23, batch 41850, giga_loss[loss=0.3312, simple_loss=0.3844, pruned_loss=0.139, over 27589.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3749, pruned_loss=0.1243, over 5646534.61 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 5701852.65 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3777, pruned_loss=0.1263, over 5628316.18 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:13:09,948 INFO [optim.py:369] (1/2) 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 05:13:41,215 INFO [train.py:968] (1/2) Epoch 23, batch 41900, giga_loss[loss=0.3258, simple_loss=0.3816, pruned_loss=0.135, over 27865.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3747, pruned_loss=0.1237, over 5656031.63 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1142, over 5703924.91 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.377, pruned_loss=0.1254, over 5639332.14 frames. ], batch size: 412, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:14:25,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4805, 1.4347, 3.9258, 3.2879], device='cuda:1'), covar=tensor([0.1612, 0.2771, 0.0507, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0661, 0.0980, 0.0926], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 05:14:35,109 INFO [train.py:968] (1/2) Epoch 23, batch 41950, giga_loss[loss=0.272, simple_loss=0.3462, pruned_loss=0.09886, over 28636.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3723, pruned_loss=0.1218, over 5655109.37 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3606, pruned_loss=0.1138, over 5711325.23 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.375, pruned_loss=0.1239, over 5632785.43 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:14:39,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:14:49,636 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.96 vs. limit=2.0
+2023-03-12 05:14:53,635 INFO [optim.py:369] (1/2) 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 05:15:08,043 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:15:10,237 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 42000, giga_loss[loss=0.3034, simple_loss=0.387, pruned_loss=0.1099, over 28480.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3733, pruned_loss=0.1198, over 5647306.56 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5701124.84 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3755, pruned_loss=0.1213, over 5636785.91 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:15:27,896 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 05:15:36,895 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 05:15:48,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:16:20,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045438.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:16:22,021 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:16:23,020 INFO [train.py:968] (1/2) Epoch 23, batch 42050, giga_loss[loss=0.3454, simple_loss=0.4086, pruned_loss=0.1411, over 28958.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.374, pruned_loss=0.1193, over 5666025.61 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1137, over 5710643.13 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3772, pruned_loss=0.1212, over 5646792.75 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:16:24,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4785, 1.7658, 1.4165, 1.4962], device='cuda:1'), covar=tensor([0.2512, 0.2503, 0.2790, 0.2211], device='cuda:1'), in_proj_covar=tensor([0.1535, 0.1108, 0.1356, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 05:16:43,810 INFO [optim.py:369] (1/2) 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 05:16:58,608 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-12 05:17:12,485 INFO [train.py:968] (1/2) Epoch 23, batch 42100, giga_loss[loss=0.2612, simple_loss=0.3372, pruned_loss=0.09259, over 28798.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3753, pruned_loss=0.1204, over 5665655.69 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.1141, over 5704347.61 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3778, pruned_loss=0.1217, over 5655116.76 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:17:57,087 INFO [train.py:968] (1/2) Epoch 23, batch 42150, giga_loss[loss=0.3039, simple_loss=0.3763, pruned_loss=0.1157, over 28822.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3745, pruned_loss=0.1205, over 5670100.89 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3599, pruned_loss=0.1139, over 5708830.08 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3774, pruned_loss=0.122, over 5656090.95 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:18:10,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:11,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:19,126 INFO [optim.py:369] (1/2) 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 05:18:34,836 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045581.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:18:35,400 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:37,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045584.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:18:38,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:18:43,188 INFO [train.py:968] (1/2) Epoch 23, batch 42200, giga_loss[loss=0.2656, simple_loss=0.3403, pruned_loss=0.09543, over 28906.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3727, pruned_loss=0.1201, over 5677657.97 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3601, pruned_loss=0.114, over 5712861.44 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3753, pruned_loss=0.1214, over 5661582.30 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:19:03,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:19:05,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045613.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:19:06,144 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 42250, giga_loss[loss=0.2981, simple_loss=0.3624, pruned_loss=0.1169, over 28855.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3714, pruned_loss=0.1208, over 5664677.12 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1137, over 5707230.43 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.374, pruned_loss=0.1222, over 5656148.60 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:19:55,522 INFO [optim.py:369] (1/2) 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 05:19:59,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:25,095 INFO [train.py:968] (1/2) Epoch 23, batch 42300, libri_loss[loss=0.2911, simple_loss=0.3696, pruned_loss=0.1063, over 29527.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3705, pruned_loss=0.12, over 5664671.31 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5709652.47 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3727, pruned_loss=0.1214, over 5654368.72 frames. ], batch size: 84, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:20:33,281 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:33,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:35,990 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:36,618 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:20:54,808 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1045720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:06,230 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:06,950 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:15,652 INFO [train.py:968] (1/2) Epoch 23, batch 42350, giga_loss[loss=0.3008, simple_loss=0.3739, pruned_loss=0.1138, over 28686.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3705, pruned_loss=0.1183, over 5676168.18 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5711585.95 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3723, pruned_loss=0.1194, over 5666017.81 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:21:18,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1864, 1.4782, 1.4557, 1.0679], device='cuda:1'), covar=tensor([0.1903, 0.2770, 0.1634, 0.1876], device='cuda:1'), in_proj_covar=tensor([0.0909, 0.0706, 0.0954, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 05:21:27,971 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:33,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:21:37,519 INFO [optim.py:369] (1/2) 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 05:21:59,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5236, 3.0876, 1.6769, 1.6440], device='cuda:1'), covar=tensor([0.0811, 0.0355, 0.0697, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0560, 0.0394, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 05:22:03,471 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 42400, libri_loss[loss=0.2496, simple_loss=0.315, pruned_loss=0.0921, over 29339.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3699, pruned_loss=0.1177, over 5676655.88 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3596, pruned_loss=0.1135, over 5715208.73 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3718, pruned_loss=0.1189, over 5664703.15 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:22:25,743 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1045808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:22:45,530 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 23, batch 42450, giga_loss[loss=0.3021, simple_loss=0.3727, pruned_loss=0.1158, over 28593.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3689, pruned_loss=0.1173, over 5675993.87 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3595, pruned_loss=0.1135, over 5716269.72 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3708, pruned_loss=0.1183, over 5664659.45 frames. ], batch size: 307, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:23:17,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1045863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:23:17,408 INFO [optim.py:369] (1/2) 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 05:23:19,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1045866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:23:45,414 INFO [train.py:968] (1/2) Epoch 23, batch 42500, giga_loss[loss=0.2966, simple_loss=0.3651, pruned_loss=0.1141, over 28900.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3668, pruned_loss=0.1163, over 5681223.23 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3595, pruned_loss=0.1135, over 5715258.40 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3684, pruned_loss=0.1171, over 5672457.81 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:23:49,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1045895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:24:30,406 INFO [train.py:968] (1/2) Epoch 23, batch 42550, giga_loss[loss=0.2731, simple_loss=0.3389, pruned_loss=0.1036, over 28812.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3662, pruned_loss=0.1164, over 5675308.33 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3596, pruned_loss=0.1133, over 5717933.02 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3676, pruned_loss=0.1173, over 5664502.25 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:24:52,393 INFO [optim.py:369] (1/2) 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 05:24:56,654 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-12 05:25:17,991 INFO [train.py:968] (1/2) Epoch 23, batch 42600, libri_loss[loss=0.2652, simple_loss=0.3309, pruned_loss=0.09979, over 29657.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3651, pruned_loss=0.1163, over 5681881.57 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1135, over 5714980.91 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3663, pruned_loss=0.117, over 5673908.81 frames. ], batch size: 73, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:25:28,125 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7809, 1.9970, 1.5906, 1.9182], device='cuda:1'), covar=tensor([0.2442, 0.2577, 0.2925, 0.2386], device='cuda:1'), in_proj_covar=tensor([0.1537, 0.1109, 0.1358, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 05:26:01,057 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 23, batch 42650, giga_loss[loss=0.2624, simple_loss=0.3393, pruned_loss=0.09272, over 28925.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.364, pruned_loss=0.1162, over 5681472.80 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3595, pruned_loss=0.1133, over 5715483.79 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3652, pruned_loss=0.117, over 5674006.84 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:26:11,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:26:32,915 INFO [optim.py:369] (1/2) 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 05:26:54,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7465, 3.5982, 3.4409, 2.1724], device='cuda:1'), covar=tensor([0.0632, 0.0764, 0.0742, 0.1580], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1173, 0.0990, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 05:27:03,296 INFO [train.py:968] (1/2) Epoch 23, batch 42700, libri_loss[loss=0.3093, simple_loss=0.3753, pruned_loss=0.1217, over 29293.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3642, pruned_loss=0.1174, over 5664177.47 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3596, pruned_loss=0.1133, over 5718134.05 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.118, over 5655264.24 frames. ], batch size: 94, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:27:31,191 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-12 05:27:53,379 INFO [train.py:968] (1/2) Epoch 23, batch 42750, giga_loss[loss=0.3223, simple_loss=0.3612, pruned_loss=0.1417, over 23511.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3636, pruned_loss=0.1171, over 5659089.53 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3592, pruned_loss=0.1131, over 5717920.09 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3649, pruned_loss=0.118, over 5650709.94 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:28:14,484 INFO [optim.py:369] (1/2) 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 05:28:34,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:28:36,506 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:28:40,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:28:42,265 INFO [train.py:968] (1/2) Epoch 23, batch 42800, giga_loss[loss=0.3089, simple_loss=0.3685, pruned_loss=0.1246, over 27458.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3639, pruned_loss=0.1165, over 5667110.11 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3592, pruned_loss=0.1131, over 5717477.09 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3649, pruned_loss=0.1172, over 5660004.55 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:28:54,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:29:07,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:29:21,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3033, 0.9087, 1.0205, 1.5092], device='cuda:1'), covar=tensor([0.0733, 0.0362, 0.0340, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 05:29:30,169 INFO [train.py:968] (1/2) Epoch 23, batch 42850, libri_loss[loss=0.3548, simple_loss=0.4103, pruned_loss=0.1496, over 29234.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3641, pruned_loss=0.1157, over 5672841.53 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3595, pruned_loss=0.1136, over 5719562.79 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3647, pruned_loss=0.1159, over 5664382.73 frames. ], batch size: 97, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:29:51,771 INFO [optim.py:369] (1/2) 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 05:30:05,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1046279.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:30:08,699 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-12 05:30:16,821 INFO [train.py:968] (1/2) Epoch 23, batch 42900, giga_loss[loss=0.3378, simple_loss=0.3805, pruned_loss=0.1476, over 26687.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3651, pruned_loss=0.116, over 5677113.10 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3594, pruned_loss=0.1135, over 5723348.46 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3658, pruned_loss=0.1163, over 5666102.29 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:30:54,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:30:56,225 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:07,418 INFO [train.py:968] (1/2) Epoch 23, batch 42950, giga_loss[loss=0.2965, simple_loss=0.3667, pruned_loss=0.1132, over 28765.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3663, pruned_loss=0.1168, over 5684693.71 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1136, over 5722959.22 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3668, pruned_loss=0.1169, over 5675104.21 frames. ], batch size: 262, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:31:11,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:14,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:22,764 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:27,566 INFO [optim.py:369] (1/2) 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 05:31:43,978 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:31:55,489 INFO [train.py:968] (1/2) Epoch 23, batch 43000, giga_loss[loss=0.3562, simple_loss=0.3874, pruned_loss=0.1625, over 23618.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3686, pruned_loss=0.1192, over 5668998.50 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3593, pruned_loss=0.1134, over 5706060.96 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3695, pruned_loss=0.1196, over 5676290.71 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:32:48,656 INFO [train.py:968] (1/2) Epoch 23, batch 43050, giga_loss[loss=0.276, simple_loss=0.3456, pruned_loss=0.1032, over 28826.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1216, over 5662091.43 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3593, pruned_loss=0.1133, over 5700835.56 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3708, pruned_loss=0.122, over 5671971.69 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:33:14,177 INFO [optim.py:369] (1/2) 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 05:33:36,581 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1046486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:33:41,072 INFO [train.py:968] (1/2) Epoch 23, batch 43100, giga_loss[loss=0.3179, simple_loss=0.3778, pruned_loss=0.129, over 28954.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3712, pruned_loss=0.1236, over 5667904.78 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3587, pruned_loss=0.113, over 5704012.92 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3728, pruned_loss=0.1246, over 5671937.00 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:33:50,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9880, 1.2641, 1.3556, 1.0430], device='cuda:1'), covar=tensor([0.1868, 0.1376, 0.2327, 0.1833], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0755, 0.0723, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 05:34:10,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3127, 4.1582, 3.9546, 1.8127], device='cuda:1'), covar=tensor([0.0675, 0.0781, 0.0740, 0.2142], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1178, 0.0995, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 05:34:31,152 INFO [train.py:968] (1/2) Epoch 23, batch 43150, giga_loss[loss=0.2517, simple_loss=0.3267, pruned_loss=0.08835, over 28756.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3721, pruned_loss=0.1245, over 5661055.68 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3585, pruned_loss=0.1128, over 5706743.02 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3738, pruned_loss=0.1256, over 5661397.27 frames. ], batch size: 99, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:34:52,266 INFO [optim.py:369] (1/2) 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 05:35:20,235 INFO [train.py:968] (1/2) Epoch 23, batch 43200, libri_loss[loss=0.3283, simple_loss=0.3882, pruned_loss=0.1342, over 29378.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3723, pruned_loss=0.1247, over 5658014.26 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3591, pruned_loss=0.1133, over 5704500.00 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3735, pruned_loss=0.1256, over 5658282.80 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:36:01,437 INFO [train.py:968] (1/2) Epoch 23, batch 43250, giga_loss[loss=0.2801, simple_loss=0.3578, pruned_loss=0.1011, over 28535.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3724, pruned_loss=0.1236, over 5672176.49 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3594, pruned_loss=0.1137, over 5712453.50 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1244, over 5663458.32 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:36:12,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5897, 1.8468, 1.3082, 1.4321], device='cuda:1'), covar=tensor([0.1012, 0.0580, 0.1062, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0446, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 05:36:16,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1046654.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:36:22,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4897, 2.0970, 1.8159, 1.5751], device='cuda:1'), covar=tensor([0.0809, 0.0285, 0.0302, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 05:36:27,832 INFO [optim.py:369] (1/2) 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 05:36:44,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-12 05:36:50,682 INFO [train.py:968] (1/2) Epoch 23, batch 43300, giga_loss[loss=0.3031, simple_loss=0.3488, pruned_loss=0.1287, over 26621.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3695, pruned_loss=0.1202, over 5671003.59 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3589, pruned_loss=0.1133, over 5714805.52 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3712, pruned_loss=0.1213, over 5661027.49 frames. ], batch size: 555, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:37:01,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7322, 1.9060, 1.6885, 1.6639], device='cuda:1'), covar=tensor([0.2240, 0.2053, 0.2086, 0.1971], device='cuda:1'), in_proj_covar=tensor([0.1542, 0.1113, 0.1359, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 05:37:05,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6096, 1.8859, 1.4955, 1.6616], device='cuda:1'), covar=tensor([0.2620, 0.2633, 0.3043, 0.2426], device='cuda:1'), in_proj_covar=tensor([0.1542, 0.1113, 0.1359, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 05:37:10,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3666, 1.5966, 1.3213, 0.9752], device='cuda:1'), covar=tensor([0.2552, 0.2698, 0.3048, 0.2365], device='cuda:1'), in_proj_covar=tensor([0.1542, 0.1113, 0.1360, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 05:37:35,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5760, 1.7639, 1.6687, 1.5998], device='cuda:1'), covar=tensor([0.1917, 0.2162, 0.2366, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0752, 0.0720, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 05:37:36,657 INFO [train.py:968] (1/2) Epoch 23, batch 43350, giga_loss[loss=0.2932, simple_loss=0.3627, pruned_loss=0.1119, over 29042.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3681, pruned_loss=0.1198, over 5671818.36 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3585, pruned_loss=0.113, over 5714900.75 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.37, pruned_loss=0.121, over 5662809.62 frames. ], batch size: 128, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:37:59,480 INFO [optim.py:369] (1/2) 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 05:38:14,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5006, 1.7934, 1.4513, 1.5697], device='cuda:1'), covar=tensor([0.2394, 0.2405, 0.2621, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1113, 0.1361, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 05:38:24,930 INFO [train.py:968] (1/2) Epoch 23, batch 43400, giga_loss[loss=0.3237, simple_loss=0.3609, pruned_loss=0.1433, over 23585.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3669, pruned_loss=0.1197, over 5665927.79 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.359, pruned_loss=0.1132, over 5714317.15 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3681, pruned_loss=0.1206, over 5658587.71 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:38:30,934 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1046797.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:38:35,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1046800.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 05:39:05,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1046829.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:39:15,756 INFO [train.py:968] (1/2) Epoch 23, batch 43450, giga_loss[loss=0.302, simple_loss=0.3712, pruned_loss=0.1164, over 28946.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.367, pruned_loss=0.1196, over 5672554.75 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.359, pruned_loss=0.1132, over 5714317.15 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3679, pruned_loss=0.1203, over 5666841.87 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:39:35,794 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:39:55,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2866, 0.8757, 0.8876, 1.3942], device='cuda:1'), covar=tensor([0.0769, 0.0385, 0.0386, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 05:40:06,594 INFO [train.py:968] (1/2) Epoch 23, batch 43500, giga_loss[loss=0.2918, simple_loss=0.378, pruned_loss=0.1028, over 29076.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3708, pruned_loss=0.1212, over 5665593.93 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3592, pruned_loss=0.1133, over 5712168.30 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3716, pruned_loss=0.1219, over 5661613.34 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:40:20,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8440, 1.2005, 1.2975, 1.0189], device='cuda:1'), covar=tensor([0.2122, 0.1493, 0.2447, 0.1968], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0753, 0.0720, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 05:40:49,608 INFO [train.py:968] (1/2) Epoch 23, batch 43550, giga_loss[loss=0.2928, simple_loss=0.3711, pruned_loss=0.1073, over 28918.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3725, pruned_loss=0.1193, over 5674373.11 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3592, pruned_loss=0.1133, over 5713113.65 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3734, pruned_loss=0.12, over 5669019.57 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:41:17,678 INFO [optim.py:369] (1/2) 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 05:41:41,715 INFO [train.py:968] (1/2) Epoch 23, batch 43600, giga_loss[loss=0.3787, simple_loss=0.4016, pruned_loss=0.178, over 23810.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.374, pruned_loss=0.1203, over 5667879.08 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3594, pruned_loss=0.1133, over 5715488.49 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3749, pruned_loss=0.121, over 5660529.91 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:41:55,331 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:42:00,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:42:24,189 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:42:27,176 INFO [train.py:968] (1/2) Epoch 23, batch 43650, giga_loss[loss=0.3394, simple_loss=0.4078, pruned_loss=0.1355, over 28911.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3763, pruned_loss=0.1224, over 5673465.35 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3588, pruned_loss=0.1129, over 5722684.88 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3782, pruned_loss=0.1237, over 5659167.73 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:42:52,280 INFO [optim.py:369] (1/2) 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 05:42:58,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-12 05:43:00,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 05:43:05,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:43:15,334 INFO [train.py:968] (1/2) Epoch 23, batch 43700, libri_loss[loss=0.2873, simple_loss=0.3539, pruned_loss=0.1103, over 29544.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3764, pruned_loss=0.1229, over 5666286.23 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3594, pruned_loss=0.1132, over 5720601.66 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3781, pruned_loss=0.1241, over 5654925.57 frames. ], batch size: 84, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:43:32,999 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4083, 1.6013, 1.6091, 1.4319], device='cuda:1'), covar=tensor([0.1983, 0.2002, 0.2386, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0755, 0.0722, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 05:43:58,234 INFO [train.py:968] (1/2) Epoch 23, batch 43750, giga_loss[loss=0.2598, simple_loss=0.3275, pruned_loss=0.09607, over 28620.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3765, pruned_loss=0.124, over 5664290.37 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3593, pruned_loss=0.113, over 5711477.75 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3783, pruned_loss=0.1253, over 5663001.65 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:44:15,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5198, 1.8089, 1.4918, 1.4820], device='cuda:1'), covar=tensor([0.2538, 0.2526, 0.2829, 0.2282], device='cuda:1'), in_proj_covar=tensor([0.1535, 0.1109, 0.1355, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 05:44:27,352 INFO [optim.py:369] (1/2) 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 05:44:31,288 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-12 05:44:39,975 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:44:46,975 INFO [train.py:968] (1/2) Epoch 23, batch 43800, giga_loss[loss=0.3423, simple_loss=0.3994, pruned_loss=0.1426, over 28796.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.375, pruned_loss=0.124, over 5646958.70 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3594, pruned_loss=0.1131, over 5694455.93 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3767, pruned_loss=0.1252, over 5660316.38 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:44:55,434 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:45:33,918 INFO [train.py:968] (1/2) Epoch 23, batch 43850, giga_loss[loss=0.2558, simple_loss=0.3239, pruned_loss=0.09384, over 28950.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3712, pruned_loss=0.1218, over 5665593.19 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3585, pruned_loss=0.1124, over 5705964.29 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.1239, over 5663584.55 frames. ], batch size: 86, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:45:57,621 INFO [optim.py:369] (1/2) 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 05:46:20,444 INFO [train.py:968] (1/2) Epoch 23, batch 43900, giga_loss[loss=0.3337, simple_loss=0.3638, pruned_loss=0.1518, over 23445.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1213, over 5669461.75 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.112, over 5711040.52 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3729, pruned_loss=0.1238, over 5661958.03 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:46:29,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5120, 1.8392, 1.4513, 1.7872], device='cuda:1'), covar=tensor([0.2334, 0.2421, 0.2606, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.1536, 0.1108, 0.1356, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 05:47:10,755 INFO [train.py:968] (1/2) Epoch 23, batch 43950, giga_loss[loss=0.3456, simple_loss=0.4043, pruned_loss=0.1434, over 28601.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1218, over 5679378.94 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3574, pruned_loss=0.1115, over 5712504.44 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1244, over 5670968.05 frames. ], batch size: 336, lr: 1.37e-03, grad_scale: 2.0
+2023-03-12 05:47:18,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:47:26,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 05:47:37,914 INFO [optim.py:369] (1/2) 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 05:47:38,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3191, 3.2112, 1.4766, 1.4602], device='cuda:1'), covar=tensor([0.0994, 0.0371, 0.0899, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0561, 0.0393, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 05:47:54,740 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:48:01,447 INFO [train.py:968] (1/2) Epoch 23, batch 44000, giga_loss[loss=0.2919, simple_loss=0.3609, pruned_loss=0.1114, over 28985.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.371, pruned_loss=0.1232, over 5672024.26 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3576, pruned_loss=0.1116, over 5714308.24 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3737, pruned_loss=0.1253, over 5663488.78 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:48:17,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:48:19,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5770, 1.7474, 1.6168, 1.7126], device='cuda:1'), covar=tensor([0.0732, 0.0309, 0.0294, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 05:48:39,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3693, 1.3115, 3.7214, 3.2980], device='cuda:1'), covar=tensor([0.1526, 0.2766, 0.0443, 0.2000], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0662, 0.0979, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 05:48:51,088 INFO [train.py:968] (1/2) Epoch 23, batch 44050, giga_loss[loss=0.2719, simple_loss=0.3457, pruned_loss=0.0991, over 28917.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3672, pruned_loss=0.1206, over 5674602.23 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3576, pruned_loss=0.1116, over 5714308.24 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1222, over 5667958.97 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:49:04,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:49:13,289 INFO [optim.py:369] (1/2) 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 05:49:37,311 INFO [train.py:968] (1/2) Epoch 23, batch 44100, libri_loss[loss=0.3087, simple_loss=0.3813, pruned_loss=0.118, over 29277.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3669, pruned_loss=0.1197, over 5672745.14 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3577, pruned_loss=0.1117, over 5711801.31 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3688, pruned_loss=0.1213, over 5667750.27 frames. ], batch size: 97, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:49:40,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:49:42,425 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 05:50:26,132 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:50:29,678 INFO [train.py:968] (1/2) Epoch 23, batch 44150, giga_loss[loss=0.2819, simple_loss=0.355, pruned_loss=0.1044, over 28926.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 5662854.91 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3579, pruned_loss=0.1119, over 5704843.59 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3715, pruned_loss=0.1225, over 5664698.61 frames. ], batch size: 136, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:50:37,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5647, 1.7479, 1.2546, 1.3360], device='cuda:1'), covar=tensor([0.0924, 0.0581, 0.0948, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0449, 0.0520, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 05:50:44,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:50:45,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2378, 1.5103, 1.5005, 1.1126], device='cuda:1'), covar=tensor([0.1681, 0.2632, 0.1401, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.0909, 0.0707, 0.0954, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 05:50:53,388 INFO [optim.py:369] (1/2) 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 05:50:59,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:51:14,816 INFO [train.py:968] (1/2) Epoch 23, batch 44200, giga_loss[loss=0.3016, simple_loss=0.3721, pruned_loss=0.1155, over 29012.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.1209, over 5674564.16 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3574, pruned_loss=0.1116, over 5710078.15 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3717, pruned_loss=0.1225, over 5670184.94 frames. ], batch size: 155, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:51:23,146 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:51:25,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:51:33,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7428, 2.5836, 2.6887, 2.3536], device='cuda:1'), covar=tensor([0.1864, 0.2580, 0.2016, 0.2316], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0755, 0.0722, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 05:51:56,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:52:03,001 INFO [train.py:968] (1/2) Epoch 23, batch 44250, giga_loss[loss=0.291, simple_loss=0.3638, pruned_loss=0.1091, over 28716.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3704, pruned_loss=0.1218, over 5667968.51 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3576, pruned_loss=0.1117, over 5712713.85 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3722, pruned_loss=0.1232, over 5660913.42 frames. ], batch size: 284, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:52:27,166 INFO [optim.py:369] (1/2) 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 05:52:29,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7799, 1.9283, 1.5535, 2.2604], device='cuda:1'), covar=tensor([0.2820, 0.2926, 0.3405, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.1534, 0.1109, 0.1355, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 05:52:49,404 INFO [train.py:968] (1/2) Epoch 23, batch 44300, giga_loss[loss=0.2843, simple_loss=0.3698, pruned_loss=0.09938, over 28864.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3728, pruned_loss=0.121, over 5672889.43 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3581, pruned_loss=0.1121, over 5714612.87 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3742, pruned_loss=0.122, over 5664628.44 frames. ], batch size: 186, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:52:58,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:00,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:12,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:14,866 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:16,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:25,043 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:33,062 INFO [train.py:968] (1/2) Epoch 23, batch 44350, giga_loss[loss=0.4163, simple_loss=0.4325, pruned_loss=0.2001, over 23436.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3747, pruned_loss=0.1203, over 5681525.40 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3578, pruned_loss=0.1119, over 5715008.40 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3765, pruned_loss=0.1216, over 5673339.74 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:53:40,070 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:48,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:53:52,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2283, 1.4315, 1.4833, 1.1526], device='cuda:1'), covar=tensor([0.1052, 0.1711, 0.0880, 0.1183], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0706, 0.0954, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 05:54:01,202 INFO [optim.py:369] (1/2) 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 05:54:14,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:54:23,212 INFO [train.py:968] (1/2) Epoch 23, batch 44400, giga_loss[loss=0.2717, simple_loss=0.3636, pruned_loss=0.08994, over 28851.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3774, pruned_loss=0.1216, over 5683968.85 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3579, pruned_loss=0.1121, over 5710791.46 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3792, pruned_loss=0.1226, over 5679762.76 frames. ], batch size: 174, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 05:54:54,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2081, 1.8201, 1.4277, 0.4228], device='cuda:1'), covar=tensor([0.5032, 0.3101, 0.3994, 0.6495], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1685, 0.1611, 0.1446], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 05:55:11,263 INFO [train.py:968] (1/2) Epoch 23, batch 44450, giga_loss[loss=0.3008, simple_loss=0.3635, pruned_loss=0.1191, over 28608.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3796, pruned_loss=0.1244, over 5681973.25 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3573, pruned_loss=0.1117, over 5715078.18 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.382, pruned_loss=0.1258, over 5674290.43 frames. ], batch size: 78, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:55:18,650 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4192, 1.4707, 1.3919, 1.5342], device='cuda:1'), covar=tensor([0.0763, 0.0351, 0.0321, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 05:55:37,109 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:55:38,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:55:41,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:04,197 INFO [train.py:968] (1/2) Epoch 23, batch 44500, giga_loss[loss=0.2944, simple_loss=0.371, pruned_loss=0.1089, over 28680.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3816, pruned_loss=0.1278, over 5660606.88 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3579, pruned_loss=0.1122, over 5716197.45 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3835, pruned_loss=0.1287, over 5652764.31 frames. ], batch size: 60, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:56:05,054 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047892.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:56:11,057 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:15,369 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:16,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2113, 1.3114, 1.1523, 0.9485], device='cuda:1'), covar=tensor([0.1003, 0.0519, 0.1061, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0449, 0.0519, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 05:56:17,450 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:20,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1047907.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 05:56:23,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1047912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:38,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1047925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:40,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1047928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:45,284 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:56:50,549 INFO [train.py:968] (1/2) Epoch 23, batch 44550, giga_loss[loss=0.2842, simple_loss=0.3652, pruned_loss=0.1016, over 28878.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3816, pruned_loss=0.1283, over 5665939.84 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3579, pruned_loss=0.1123, over 5718127.66 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3838, pruned_loss=0.1294, over 5656218.91 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:57:08,227 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1047957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:57:17,679 INFO [optim.py:369] (1/2) 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 05:57:38,387 INFO [train.py:968] (1/2) Epoch 23, batch 44600, giga_loss[loss=0.275, simple_loss=0.3608, pruned_loss=0.09457, over 28930.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3779, pruned_loss=0.1246, over 5668202.48 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3577, pruned_loss=0.1121, over 5718873.24 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3803, pruned_loss=0.126, over 5658526.94 frames. ], batch size: 213, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:57:42,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0556, 2.6539, 1.9867, 1.6109], device='cuda:1'), covar=tensor([0.4612, 0.3012, 0.2770, 0.4222], device='cuda:1'), in_proj_covar=tensor([0.1783, 0.1686, 0.1613, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 05:58:00,061 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:01,911 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:18,635 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:24,519 INFO [train.py:968] (1/2) Epoch 23, batch 44650, libri_loss[loss=0.3026, simple_loss=0.3753, pruned_loss=0.115, over 29365.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3766, pruned_loss=0.1214, over 5668452.07 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3578, pruned_loss=0.1122, over 5714611.26 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.379, pruned_loss=0.1227, over 5662249.73 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:58:30,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:36,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-12 05:58:37,286 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:39,332 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:39,546 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 05:58:41,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:58:54,648 INFO [optim.py:369] (1/2) 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 05:59:02,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3467, 1.5123, 1.4710, 1.3917], device='cuda:1'), covar=tensor([0.1655, 0.1876, 0.2136, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0752, 0.0718, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 05:59:12,213 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 05:59:14,706 INFO [train.py:968] (1/2) Epoch 23, batch 44700, giga_loss[loss=0.2979, simple_loss=0.3772, pruned_loss=0.1092, over 28624.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3779, pruned_loss=0.1214, over 5666462.77 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3578, pruned_loss=0.1122, over 5716200.02 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3799, pruned_loss=0.1226, over 5659939.19 frames. ], batch size: 92, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 05:59:25,911 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:00:05,787 INFO [train.py:968] (1/2) Epoch 23, batch 44750, giga_loss[loss=0.2791, simple_loss=0.3517, pruned_loss=0.1032, over 28675.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3788, pruned_loss=0.1229, over 5670396.82 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3576, pruned_loss=0.112, over 5717896.85 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.381, pruned_loss=0.1242, over 5662570.70 frames. ], batch size: 242, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:00:31,282 INFO [optim.py:369] (1/2) 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 06:00:50,636 INFO [train.py:968] (1/2) Epoch 23, batch 44800, giga_loss[loss=0.2703, simple_loss=0.3482, pruned_loss=0.09622, over 28999.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3767, pruned_loss=0.1217, over 5684850.77 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3576, pruned_loss=0.112, over 5721916.88 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3793, pruned_loss=0.1232, over 5673273.96 frames. ], batch size: 164, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 06:01:16,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6447, 4.4773, 4.2327, 1.8764], device='cuda:1'), covar=tensor([0.0598, 0.0710, 0.0716, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1184, 0.1000, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 06:01:46,902 INFO [train.py:968] (1/2) Epoch 23, batch 44850, giga_loss[loss=0.3457, simple_loss=0.396, pruned_loss=0.1477, over 27546.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3768, pruned_loss=0.1235, over 5650418.22 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3577, pruned_loss=0.1121, over 5712843.29 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3788, pruned_loss=0.1246, over 5649867.89 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 06:01:56,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3968, 2.4042, 2.2597, 2.0516], device='cuda:1'), covar=tensor([0.1678, 0.2097, 0.1913, 0.2064], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0756, 0.0721, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 06:02:15,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048267.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:02:18,087 INFO [optim.py:369] (1/2) 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 06:02:27,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048282.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:02:36,322 INFO [train.py:968] (1/2) Epoch 23, batch 44900, giga_loss[loss=0.285, simple_loss=0.3493, pruned_loss=0.1103, over 28831.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3743, pruned_loss=0.1226, over 5656648.74 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3573, pruned_loss=0.1117, over 5716002.92 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3766, pruned_loss=0.1241, over 5651983.64 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:02:40,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1957, 1.2877, 1.1084, 0.8887], device='cuda:1'), covar=tensor([0.0953, 0.0486, 0.1016, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0451, 0.0522, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 06:03:25,203 INFO [train.py:968] (1/2) Epoch 23, batch 44950, libri_loss[loss=0.3056, simple_loss=0.3746, pruned_loss=0.1183, over 29742.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3723, pruned_loss=0.1219, over 5657151.15 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3578, pruned_loss=0.112, over 5718011.64 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3742, pruned_loss=0.1231, over 5649853.01 frames. ], batch size: 87, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:03:50,671 INFO [optim.py:369] (1/2) 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 06:03:58,663 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:04:08,889 INFO [train.py:968] (1/2) Epoch 23, batch 45000, giga_loss[loss=0.332, simple_loss=0.3851, pruned_loss=0.1395, over 27545.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3712, pruned_loss=0.1217, over 5667914.40 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3582, pruned_loss=0.1121, over 5721668.78 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3727, pruned_loss=0.1228, over 5657473.43 frames. ], batch size: 472, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:04:08,889 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 06:04:17,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4892, 1.8749, 1.4696, 1.3925], device='cuda:1'), covar=tensor([0.3107, 0.2969, 0.3276, 0.2605], device='cuda:1'), in_proj_covar=tensor([0.1536, 0.1107, 0.1357, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 06:04:19,228 INFO [train.py:1012] (1/2) Epoch 23, validation: loss=0.2058, simple_loss=0.3149, pruned_loss=0.04835, over 944034.00 frames. 
+2023-03-12 06:04:19,228 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 06:04:34,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:04:35,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048410.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:04:37,441 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048413.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:04:46,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048425.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:04:48,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:04:49,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048428.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:05:00,399 INFO [train.py:968] (1/2) Epoch 23, batch 45050, giga_loss[loss=0.2636, simple_loss=0.339, pruned_loss=0.09413, over 28762.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3712, pruned_loss=0.1216, over 5657069.23 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1127, over 5703151.04 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3723, pruned_loss=0.1224, over 5662728.31 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:05:02,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048442.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:05:04,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:05:14,603 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:05:14,618 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048457.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:05:24,822 INFO [optim.py:369] (1/2) 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 06:05:30,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:05:42,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4932, 3.6244, 1.5322, 1.7048], device='cuda:1'), covar=tensor([0.1033, 0.0348, 0.0941, 0.1360], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0563, 0.0394, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 06:05:44,380 INFO [train.py:968] (1/2) Epoch 23, batch 45100, giga_loss[loss=0.2492, simple_loss=0.3155, pruned_loss=0.09141, over 24034.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.368, pruned_loss=0.1186, over 5646782.39 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3591, pruned_loss=0.1129, over 5697984.96 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3692, pruned_loss=0.1192, over 5654828.14 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:06:29,097 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:06:30,993 INFO [train.py:968] (1/2) Epoch 23, batch 45150, giga_loss[loss=0.268, simple_loss=0.3475, pruned_loss=0.09427, over 28951.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3642, pruned_loss=0.1146, over 5659432.13 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3592, pruned_loss=0.1129, over 5700465.02 frames. ], giga_tot_loss[loss=0.2977, simple_loss=0.365, pruned_loss=0.1152, over 5662847.67 frames. ], batch size: 145, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:06:38,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6315, 1.8746, 1.6998, 1.6072], device='cuda:1'), covar=tensor([0.1793, 0.2034, 0.2218, 0.2129], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0758, 0.0722, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 06:06:40,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:06:43,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:06:59,650 INFO [optim.py:369] (1/2) 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 06:06:59,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:02,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:13,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:21,614 INFO [train.py:968] (1/2) Epoch 23, batch 45200, libri_loss[loss=0.3278, simple_loss=0.3869, pruned_loss=0.1344, over 29241.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3637, pruned_loss=0.1143, over 5663165.18 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3592, pruned_loss=0.1129, over 5706955.58 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3645, pruned_loss=0.1148, over 5658693.58 frames. ], batch size: 97, lr: 1.37e-03, grad_scale: 8.0
+2023-03-12 06:07:32,983 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:36,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6035, 1.7196, 1.8077, 1.4139], device='cuda:1'), covar=tensor([0.1836, 0.2607, 0.1483, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0710, 0.0957, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 06:07:45,438 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:47,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1048619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:07:49,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:08:08,529 INFO [train.py:968] (1/2) Epoch 23, batch 45250, giga_loss[loss=0.2328, simple_loss=0.3059, pruned_loss=0.07987, over 28595.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1142, over 5652239.72 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3588, pruned_loss=0.1127, over 5703431.70 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3633, pruned_loss=0.1149, over 5650302.00 frames. ], batch size: 85, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:08:19,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:08:43,686 INFO [optim.py:369] (1/2) 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 06:08:53,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7632, 2.1438, 1.8863, 1.9570], device='cuda:1'), covar=tensor([0.0673, 0.0265, 0.0289, 0.0760], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 06:08:55,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8168, 2.6585, 1.6800, 0.9293], device='cuda:1'), covar=tensor([0.8575, 0.4021, 0.4330, 0.7648], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1686, 0.1613, 0.1447], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:09:01,082 INFO [train.py:968] (1/2) Epoch 23, batch 45300, giga_loss[loss=0.2855, simple_loss=0.3566, pruned_loss=0.1072, over 28924.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1147, over 5626176.69 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1128, over 5683754.38 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3621, pruned_loss=0.1152, over 5640416.32 frames. ], batch size: 199, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:09:46,485 INFO [train.py:968] (1/2) Epoch 23, batch 45350, giga_loss[loss=0.288, simple_loss=0.3626, pruned_loss=0.1067, over 28937.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.1161, over 5576610.99 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5631035.58 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3634, pruned_loss=0.1158, over 5635204.19 frames. ], batch size: 227, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:09:57,423 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:10:13,788 INFO [optim.py:369] (1/2) 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 06:10:34,562 INFO [train.py:968] (1/2) Epoch 23, batch 45400, giga_loss[loss=0.308, simple_loss=0.3702, pruned_loss=0.1229, over 28759.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3659, pruned_loss=0.1171, over 5584041.60 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3601, pruned_loss=0.1138, over 5614317.67 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3655, pruned_loss=0.1166, over 5643907.46 frames. ], batch size: 119, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:11:05,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:11:14,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7043, 2.1269, 1.6999, 1.1999], device='cuda:1'), covar=tensor([0.4088, 0.2931, 0.2890, 0.4677], device='cuda:1'), in_proj_covar=tensor([0.1780, 0.1679, 0.1609, 0.1442], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:11:16,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:11:26,370 INFO [train.py:968] (1/2) Epoch 23, batch 45450, giga_loss[loss=0.33, simple_loss=0.3739, pruned_loss=0.143, over 23413.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3681, pruned_loss=0.1191, over 5541811.88 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3612, pruned_loss=0.1146, over 5572326.96 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.367, pruned_loss=0.1181, over 5627181.26 frames. ], batch size: 705, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:11:42,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.01 vs. limit=2.0
+2023-03-12 06:11:46,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-12 06:11:50,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7384, 1.9622, 1.5388, 2.0366], device='cuda:1'), covar=tensor([0.2618, 0.2720, 0.3037, 0.2504], device='cuda:1'), in_proj_covar=tensor([0.1539, 0.1112, 0.1361, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 06:11:54,823 INFO [optim.py:369] (1/2) 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 06:12:11,998 INFO [train.py:968] (1/2) Epoch 23, batch 45500, giga_loss[loss=0.2777, simple_loss=0.349, pruned_loss=0.1032, over 28438.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3683, pruned_loss=0.1195, over 5532375.40 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.362, pruned_loss=0.1152, over 5538609.41 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3668, pruned_loss=0.1182, over 5630888.56 frames. ], batch size: 71, lr: 1.37e-03, grad_scale: 4.0
+2023-03-12 06:12:17,699 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:12:19,984 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-12 06:12:54,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:08,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:12,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0698, 1.4230, 0.8941, 0.9568], device='cuda:1'), covar=tensor([0.1292, 0.0650, 0.1604, 0.1441], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0451, 0.0523, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 06:13:24,251 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:39,801 INFO [train.py:968] (1/2) Epoch 24, batch 50, giga_loss[loss=0.3172, simple_loss=0.3908, pruned_loss=0.1218, over 28689.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3716, pruned_loss=0.1061, over 1267235.64 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08897, over 173134.40 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3756, pruned_loss=0.1084, over 1127881.09 frames. ], batch size: 262, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:13:55,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:13:57,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:02,156 INFO [optim.py:369] (1/2) 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 06:14:08,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1048975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:12,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1048978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:30,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1048994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:31,109 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1048995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:33,434 INFO [train.py:968] (1/2) Epoch 24, batch 100, giga_loss[loss=0.2554, simple_loss=0.3405, pruned_loss=0.08513, over 28990.00 frames. ], tot_loss[loss=0.2858, simple_loss=0.3643, pruned_loss=0.1036, over 2241287.43 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3436, pruned_loss=0.08819, over 286209.47 frames. ], giga_tot_loss[loss=0.2887, simple_loss=0.3667, pruned_loss=0.1054, over 2057797.43 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:14:40,654 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:14:42,165 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:15:17,369 INFO [train.py:968] (1/2) Epoch 24, batch 150, giga_loss[loss=0.2197, simple_loss=0.2851, pruned_loss=0.07719, over 23911.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3489, pruned_loss=0.09653, over 2993627.56 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3422, pruned_loss=0.08798, over 485845.85 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3505, pruned_loss=0.09798, over 2745932.46 frames. ], batch size: 705, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:15:24,821 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:15:26,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:15:38,495 INFO [optim.py:369] (1/2) 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 06:15:52,243 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 200, giga_loss[loss=0.244, simple_loss=0.3133, pruned_loss=0.08741, over 28697.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3351, pruned_loss=0.08971, over 3596270.87 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3403, pruned_loss=0.08636, over 668966.99 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3354, pruned_loss=0.09074, over 3318167.43 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:16:07,165 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:16:32,271 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:16:34,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:16:39,286 INFO [train.py:968] (1/2) Epoch 24, batch 250, libri_loss[loss=0.3095, simple_loss=0.3928, pruned_loss=0.1131, over 29153.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3275, pruned_loss=0.08587, over 4073233.49 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3395, pruned_loss=0.08525, over 923305.09 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3267, pruned_loss=0.0867, over 3754088.15 frames. ], batch size: 97, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:17:02,054 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 24, batch 300, giga_loss[loss=0.2169, simple_loss=0.2835, pruned_loss=0.07511, over 28592.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3192, pruned_loss=0.0826, over 4428439.86 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3402, pruned_loss=0.08595, over 1045292.53 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3174, pruned_loss=0.08278, over 4137963.60 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:18:11,764 INFO [train.py:968] (1/2) Epoch 24, batch 350, giga_loss[loss=0.2055, simple_loss=0.2834, pruned_loss=0.06377, over 28848.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3113, pruned_loss=0.07896, over 4703893.26 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3407, pruned_loss=0.08596, over 1184427.14 frames. ], giga_tot_loss[loss=0.2329, simple_loss=0.3084, pruned_loss=0.07867, over 4439280.31 frames. ], batch size: 213, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:18:28,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1751, 3.3689, 2.4208, 1.2762], device='cuda:1'), covar=tensor([0.8892, 0.3595, 0.3868, 0.7187], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1682, 0.1614, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:18:31,704 INFO [optim.py:369] (1/2) 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 06:18:40,060 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 06:18:51,440 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2836, 1.0805, 3.9091, 3.1757], device='cuda:1'), covar=tensor([0.1759, 0.3112, 0.0449, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0660, 0.0978, 0.0926], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 06:18:53,770 INFO [train.py:968] (1/2) Epoch 24, batch 400, libri_loss[loss=0.289, simple_loss=0.3644, pruned_loss=0.1068, over 29126.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.3078, pruned_loss=0.07763, over 4925695.10 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3419, pruned_loss=0.08691, over 1300674.49 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3041, pruned_loss=0.07684, over 4690614.00 frames. ], batch size: 101, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:19:09,349 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 450, giga_loss[loss=0.2302, simple_loss=0.3038, pruned_loss=0.07825, over 28274.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3056, pruned_loss=0.07673, over 5101487.07 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3421, pruned_loss=0.08732, over 1413935.77 frames. ], giga_tot_loss[loss=0.2265, simple_loss=0.3016, pruned_loss=0.07571, over 4894507.28 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:20:00,464 INFO [optim.py:369] (1/2) 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 06:20:05,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6698, 2.4451, 1.5937, 0.8563], device='cuda:1'), covar=tensor([0.9654, 0.4241, 0.4614, 0.8287], device='cuda:1'), in_proj_covar=tensor([0.1777, 0.1679, 0.1611, 0.1441], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:20:06,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:20:18,513 INFO [train.py:968] (1/2) Epoch 24, batch 500, giga_loss[loss=0.2166, simple_loss=0.2874, pruned_loss=0.07293, over 28468.00 frames. ], tot_loss[loss=0.229, simple_loss=0.3045, pruned_loss=0.07673, over 5216885.26 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3438, pruned_loss=0.08858, over 1554900.91 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.2995, pruned_loss=0.07515, over 5039946.95 frames. ], batch size: 65, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:20:40,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:21:02,486 INFO [train.py:968] (1/2) Epoch 24, batch 550, libri_loss[loss=0.2933, simple_loss=0.3633, pruned_loss=0.1117, over 29526.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3026, pruned_loss=0.07637, over 5324155.90 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3431, pruned_loss=0.08881, over 1662380.10 frames. ], giga_tot_loss[loss=0.2236, simple_loss=0.2977, pruned_loss=0.07473, over 5168993.07 frames. ], batch size: 80, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:21:04,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1811, 1.4887, 1.5729, 1.3132], device='cuda:1'), covar=tensor([0.2220, 0.1828, 0.2483, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0753, 0.0718, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 06:21:22,989 INFO [optim.py:369] (1/2) 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 06:21:35,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:21:49,548 INFO [train.py:968] (1/2) Epoch 24, batch 600, giga_loss[loss=0.1805, simple_loss=0.2541, pruned_loss=0.05347, over 28489.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.2998, pruned_loss=0.07523, over 5404908.70 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3428, pruned_loss=0.08885, over 1725821.89 frames. ], giga_tot_loss[loss=0.2215, simple_loss=0.2955, pruned_loss=0.07372, over 5275106.78 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:22:05,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3765, 1.5782, 1.3446, 1.5837], device='cuda:1'), covar=tensor([0.0799, 0.0363, 0.0362, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 06:22:15,330 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:22:17,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:22:34,536 INFO [train.py:968] (1/2) Epoch 24, batch 650, giga_loss[loss=0.2089, simple_loss=0.2819, pruned_loss=0.0679, over 28890.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2979, pruned_loss=0.07391, over 5472431.19 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3424, pruned_loss=0.08828, over 1866361.96 frames. ], giga_tot_loss[loss=0.2189, simple_loss=0.293, pruned_loss=0.07236, over 5356252.03 frames. ], batch size: 213, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:22:42,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:22:56,870 INFO [optim.py:369] (1/2) 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 06:22:57,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7357, 2.0285, 1.6879, 1.6926], device='cuda:1'), covar=tensor([0.2664, 0.2763, 0.3131, 0.2700], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1118, 0.1370, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 06:23:20,884 INFO [train.py:968] (1/2) Epoch 24, batch 700, giga_loss[loss=0.2004, simple_loss=0.2743, pruned_loss=0.0633, over 28777.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2952, pruned_loss=0.07248, over 5524693.59 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3419, pruned_loss=0.08785, over 1945439.13 frames. ], giga_tot_loss[loss=0.2164, simple_loss=0.2906, pruned_loss=0.07109, over 5426279.15 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:23:43,597 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:23:48,919 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:24:05,871 INFO [train.py:968] (1/2) Epoch 24, batch 750, giga_loss[loss=0.1937, simple_loss=0.2657, pruned_loss=0.06086, over 28816.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2931, pruned_loss=0.07128, over 5555931.12 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3398, pruned_loss=0.08682, over 2070359.11 frames. ], giga_tot_loss[loss=0.2143, simple_loss=0.2886, pruned_loss=0.07001, over 5475498.07 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:24:16,072 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:24:16,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6757, 1.9271, 1.4373, 1.5197], device='cuda:1'), covar=tensor([0.1040, 0.0677, 0.1108, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0448, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 06:24:29,253 INFO [optim.py:369] (1/2) 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 06:24:44,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:24:52,033 INFO [train.py:968] (1/2) Epoch 24, batch 800, giga_loss[loss=0.2012, simple_loss=0.2712, pruned_loss=0.06555, over 27715.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2912, pruned_loss=0.07094, over 5583281.05 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3392, pruned_loss=0.08691, over 2199418.45 frames. ], giga_tot_loss[loss=0.2125, simple_loss=0.2862, pruned_loss=0.06937, over 5510658.55 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:25:12,896 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:25:36,651 INFO [train.py:968] (1/2) Epoch 24, batch 850, libri_loss[loss=0.2624, simple_loss=0.3419, pruned_loss=0.09141, over 29769.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.295, pruned_loss=0.07288, over 5609884.53 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3382, pruned_loss=0.08645, over 2326742.88 frames. ], giga_tot_loss[loss=0.2163, simple_loss=0.2899, pruned_loss=0.07131, over 5542114.43 frames. ], batch size: 87, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:25:41,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3731, 1.4588, 1.3367, 1.3603], device='cuda:1'), covar=tensor([0.1949, 0.1777, 0.1955, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.1997, 0.1933, 0.1858, 0.2001], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 06:25:58,555 INFO [optim.py:369] (1/2) 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 06:26:11,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 06:26:19,407 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 06:26:24,028 INFO [train.py:968] (1/2) Epoch 24, batch 900, giga_loss[loss=0.2678, simple_loss=0.3481, pruned_loss=0.09374, over 28931.00 frames. ], tot_loss[loss=0.2326, simple_loss=0.3074, pruned_loss=0.07892, over 5631872.47 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3379, pruned_loss=0.08644, over 2413300.88 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3028, pruned_loss=0.0775, over 5573207.84 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:26:24,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1049798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:26:58,245 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:27:03,716 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 950, giga_loss[loss=0.2771, simple_loss=0.3619, pruned_loss=0.09608, over 28956.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.32, pruned_loss=0.08557, over 5645065.31 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3373, pruned_loss=0.08628, over 2482873.87 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3162, pruned_loss=0.08448, over 5593587.19 frames. ], batch size: 227, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:27:26,088 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1049864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:27:27,261 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:27:53,607 INFO [train.py:968] (1/2) Epoch 24, batch 1000, giga_loss[loss=0.2836, simple_loss=0.3636, pruned_loss=0.1018, over 29016.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3296, pruned_loss=0.09, over 5639725.62 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3381, pruned_loss=0.08702, over 2564288.66 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.326, pruned_loss=0.08891, over 5609605.95 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:28:16,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4575, 1.7444, 1.6784, 1.5170], device='cuda:1'), covar=tensor([0.2176, 0.2154, 0.2485, 0.2400], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0751, 0.0719, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 06:28:30,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7165, 1.8491, 1.8102, 1.6460], device='cuda:1'), covar=tensor([0.2072, 0.2262, 0.2423, 0.2225], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0751, 0.0719, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 06:28:31,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1049941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:28:33,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1049944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:28:36,052 INFO [train.py:968] (1/2) Epoch 24, batch 1050, giga_loss[loss=0.2482, simple_loss=0.3306, pruned_loss=0.08287, over 28587.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3333, pruned_loss=0.09027, over 5652667.92 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3378, pruned_loss=0.08683, over 2612992.71 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3306, pruned_loss=0.08954, over 5626359.00 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:28:42,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3561, 2.0740, 1.5307, 0.5659], device='cuda:1'), covar=tensor([0.5614, 0.3351, 0.4823, 0.6719], device='cuda:1'), in_proj_covar=tensor([0.1770, 0.1672, 0.1609, 0.1438], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:28:49,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6187, 1.7524, 1.7292, 1.4713], device='cuda:1'), covar=tensor([0.2848, 0.2613, 0.2184, 0.2775], device='cuda:1'), in_proj_covar=tensor([0.1991, 0.1931, 0.1855, 0.2002], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 06:28:53,454 INFO [optim.py:369] (1/2) 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 06:28:55,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1049973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:29:05,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2080, 1.4672, 1.5990, 1.2889], device='cuda:1'), covar=tensor([0.2227, 0.1956, 0.2461, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0752, 0.0721, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 06:29:20,821 INFO [train.py:968] (1/2) Epoch 24, batch 1100, giga_loss[loss=0.3092, simple_loss=0.3759, pruned_loss=0.1212, over 28843.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.336, pruned_loss=0.09073, over 5660337.48 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3367, pruned_loss=0.08639, over 2742464.75 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3341, pruned_loss=0.09047, over 5630372.02 frames. ], batch size: 199, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:29:28,380 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 06:29:36,905 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 06:29:57,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3439, 1.4325, 1.3732, 1.3212], device='cuda:1'), covar=tensor([0.2247, 0.2288, 0.1717, 0.1968], device='cuda:1'), in_proj_covar=tensor([0.1988, 0.1931, 0.1851, 0.2000], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 06:29:58,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 06:30:04,928 INFO [train.py:968] (1/2) Epoch 24, batch 1150, libri_loss[loss=0.2606, simple_loss=0.345, pruned_loss=0.08812, over 29541.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3382, pruned_loss=0.09184, over 5667728.67 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3358, pruned_loss=0.08591, over 2820796.55 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3372, pruned_loss=0.09198, over 5638180.89 frames. ], batch size: 89, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:30:12,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5814, 1.7664, 1.5059, 1.6461], device='cuda:1'), covar=tensor([0.2314, 0.2315, 0.2485, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.1540, 0.1111, 0.1363, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 06:30:24,235 INFO [optim.py:369] (1/2) 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 06:30:48,369 INFO [train.py:968] (1/2) Epoch 24, batch 1200, giga_loss[loss=0.2825, simple_loss=0.3558, pruned_loss=0.1046, over 28843.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.34, pruned_loss=0.09342, over 5680205.85 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3363, pruned_loss=0.08606, over 2942479.60 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3391, pruned_loss=0.09375, over 5647511.62 frames. ], batch size: 199, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:30:48,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:31:25,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1474, 1.5597, 0.9917, 1.1116], device='cuda:1'), covar=tensor([0.1361, 0.0633, 0.1561, 0.1303], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0450, 0.0523, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 06:31:33,594 INFO [train.py:968] (1/2) Epoch 24, batch 1250, giga_loss[loss=0.2662, simple_loss=0.3465, pruned_loss=0.09298, over 28841.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3445, pruned_loss=0.09643, over 5681692.04 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.337, pruned_loss=0.08636, over 3028271.80 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3436, pruned_loss=0.09683, over 5651071.39 frames. ], batch size: 199, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:31:52,207 INFO [optim.py:369] (1/2) 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 06:32:12,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3153, 3.1373, 1.4774, 1.4290], device='cuda:1'), covar=tensor([0.1093, 0.0262, 0.0941, 0.1475], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0553, 0.0392, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 06:32:14,654 INFO [train.py:968] (1/2) Epoch 24, batch 1300, giga_loss[loss=0.2891, simple_loss=0.3694, pruned_loss=0.1044, over 28995.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3463, pruned_loss=0.09681, over 5691790.91 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3367, pruned_loss=0.0861, over 3097675.93 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.346, pruned_loss=0.09751, over 5663989.89 frames. ], batch size: 106, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:32:28,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4973, 2.0635, 1.5614, 0.8716], device='cuda:1'), covar=tensor([0.6693, 0.3633, 0.3965, 0.6503], device='cuda:1'), in_proj_covar=tensor([0.1776, 0.1675, 0.1613, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:32:36,732 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050220.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:32:51,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:32:53,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:32:57,580 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1050244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:33:00,904 INFO [train.py:968] (1/2) Epoch 24, batch 1350, giga_loss[loss=0.2941, simple_loss=0.364, pruned_loss=0.1121, over 27614.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3492, pruned_loss=0.09785, over 5690297.10 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3366, pruned_loss=0.08598, over 3138841.86 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3491, pruned_loss=0.09861, over 5666246.70 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:33:18,831 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1050273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:33:45,671 INFO [train.py:968] (1/2) Epoch 24, batch 1400, giga_loss[loss=0.2554, simple_loss=0.3402, pruned_loss=0.08536, over 28683.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3502, pruned_loss=0.09751, over 5693662.10 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3363, pruned_loss=0.08575, over 3180159.65 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3505, pruned_loss=0.09839, over 5671683.69 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:34:20,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.88 vs. limit=2.0
+2023-03-12 06:34:28,438 INFO [train.py:968] (1/2) Epoch 24, batch 1450, giga_loss[loss=0.2644, simple_loss=0.3514, pruned_loss=0.08869, over 28912.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3499, pruned_loss=0.09608, over 5700472.91 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3355, pruned_loss=0.08516, over 3246071.65 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3508, pruned_loss=0.09733, over 5679719.49 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:34:38,205 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-12 06:34:42,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1098, 3.9160, 3.7204, 1.7668], device='cuda:1'), covar=tensor([0.0611, 0.0768, 0.0760, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.1236, 0.1146, 0.0964, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 06:34:47,328 INFO [optim.py:369] (1/2) 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 06:34:57,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:34:59,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1050385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:35:03,438 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050390.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:35:08,466 INFO [train.py:968] (1/2) Epoch 24, batch 1500, giga_loss[loss=0.258, simple_loss=0.3473, pruned_loss=0.08432, over 28709.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.0949, over 5704849.49 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3361, pruned_loss=0.08565, over 3323249.45 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3501, pruned_loss=0.09594, over 5684294.29 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:35:22,516 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1050414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:35:50,043 INFO [train.py:968] (1/2) Epoch 24, batch 1550, libri_loss[loss=0.2445, simple_loss=0.344, pruned_loss=0.07251, over 29124.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3477, pruned_loss=0.09313, over 5705135.09 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3364, pruned_loss=0.08584, over 3386836.35 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3485, pruned_loss=0.0941, over 5695456.93 frames. ], batch size: 101, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:36:09,624 INFO [optim.py:369] (1/2) 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 06:36:33,145 INFO [train.py:968] (1/2) Epoch 24, batch 1600, giga_loss[loss=0.2868, simple_loss=0.36, pruned_loss=0.1068, over 28863.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3481, pruned_loss=0.09385, over 5701705.42 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3363, pruned_loss=0.08572, over 3424394.73 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.349, pruned_loss=0.09479, over 5691360.45 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:36:50,270 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:37:00,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4239, 1.3943, 1.1689, 1.5042], device='cuda:1'), covar=tensor([0.0739, 0.0348, 0.0350, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 06:37:19,881 INFO [train.py:968] (1/2) Epoch 24, batch 1650, giga_loss[loss=0.377, simple_loss=0.4177, pruned_loss=0.1681, over 28264.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3508, pruned_loss=0.09745, over 5697865.08 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3365, pruned_loss=0.08559, over 3469072.36 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3516, pruned_loss=0.09843, over 5689990.27 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:37:31,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 06:37:41,009 INFO [optim.py:369] (1/2) 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 06:38:03,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050595.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:38:06,016 INFO [train.py:968] (1/2) Epoch 24, batch 1700, giga_loss[loss=0.2662, simple_loss=0.3422, pruned_loss=0.09513, over 29079.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3534, pruned_loss=0.1012, over 5701754.09 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3371, pruned_loss=0.08599, over 3533023.14 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3542, pruned_loss=0.1022, over 5696671.67 frames. ], batch size: 128, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:38:13,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5283, 1.1915, 4.3064, 3.5373], device='cuda:1'), covar=tensor([0.1617, 0.3024, 0.0427, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0656, 0.0973, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 06:38:22,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3552, 3.3131, 1.5612, 1.4951], device='cuda:1'), covar=tensor([0.1073, 0.0314, 0.0906, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0557, 0.0393, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 06:38:41,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4603, 2.7892, 1.6433, 1.5533], device='cuda:1'), covar=tensor([0.0828, 0.0325, 0.0697, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0557, 0.0393, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 06:38:49,421 INFO [train.py:968] (1/2) Epoch 24, batch 1750, giga_loss[loss=0.2804, simple_loss=0.3467, pruned_loss=0.107, over 27665.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3524, pruned_loss=0.102, over 5704623.94 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3362, pruned_loss=0.08555, over 3590336.05 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3539, pruned_loss=0.1034, over 5697207.76 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:39:11,616 INFO [optim.py:369] (1/2) 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 06:39:33,353 INFO [train.py:968] (1/2) Epoch 24, batch 1800, giga_loss[loss=0.3137, simple_loss=0.3732, pruned_loss=0.1271, over 27934.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3503, pruned_loss=0.1013, over 5696206.48 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3364, pruned_loss=0.08551, over 3659183.42 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3518, pruned_loss=0.1028, over 5684721.78 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:40:09,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:40:11,519 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 1850, giga_loss[loss=0.2383, simple_loss=0.3221, pruned_loss=0.07728, over 28844.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3491, pruned_loss=0.1005, over 5699543.02 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3368, pruned_loss=0.08551, over 3725205.51 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3505, pruned_loss=0.1021, over 5684992.29 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:40:30,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050765.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:40:35,580 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:40:59,266 INFO [train.py:968] (1/2) Epoch 24, batch 1900, giga_loss[loss=0.2534, simple_loss=0.3315, pruned_loss=0.08765, over 28847.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.348, pruned_loss=0.09939, over 5692963.02 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.337, pruned_loss=0.08558, over 3786303.33 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3492, pruned_loss=0.1011, over 5678480.42 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:41:47,874 INFO [train.py:968] (1/2) Epoch 24, batch 1950, giga_loss[loss=0.2473, simple_loss=0.3228, pruned_loss=0.08587, over 27975.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3454, pruned_loss=0.09721, over 5685467.79 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3372, pruned_loss=0.08556, over 3840178.38 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3466, pruned_loss=0.09891, over 5676906.96 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:41:52,507 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1050854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:42:07,219 INFO [optim.py:369] (1/2) 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 06:42:23,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1050889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:42:23,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7256, 1.8464, 1.3464, 1.4181], device='cuda:1'), covar=tensor([0.0965, 0.0562, 0.1033, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0447, 0.0521, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 06:42:31,026 INFO [train.py:968] (1/2) Epoch 24, batch 2000, giga_loss[loss=0.242, simple_loss=0.3099, pruned_loss=0.08708, over 28689.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3403, pruned_loss=0.09427, over 5687920.18 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3374, pruned_loss=0.08556, over 3911477.47 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3412, pruned_loss=0.09597, over 5674199.57 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:42:33,602 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-12 06:42:39,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1050908.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:42:41,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1050911.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 06:43:07,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1050940.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 06:43:14,886 INFO [train.py:968] (1/2) Epoch 24, batch 2050, giga_loss[loss=0.2269, simple_loss=0.3096, pruned_loss=0.07211, over 29040.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3352, pruned_loss=0.09162, over 5677682.28 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3368, pruned_loss=0.08517, over 3990379.50 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3364, pruned_loss=0.09355, over 5665378.11 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:43:23,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3051, 1.2682, 3.9332, 3.3215], device='cuda:1'), covar=tensor([0.1701, 0.2893, 0.0407, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0654, 0.0969, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 06:43:29,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 06:43:38,007 INFO [optim.py:369] (1/2) 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 06:44:02,550 INFO [train.py:968] (1/2) Epoch 24, batch 2100, libri_loss[loss=0.2184, simple_loss=0.3043, pruned_loss=0.06622, over 29616.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3301, pruned_loss=0.08886, over 5673576.50 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3371, pruned_loss=0.08536, over 4055270.45 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3308, pruned_loss=0.09049, over 5657627.24 frames. ], batch size: 73, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:44:35,739 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:44:38,198 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 2150, libri_loss[loss=0.2775, simple_loss=0.3507, pruned_loss=0.1022, over 29547.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3292, pruned_loss=0.08819, over 5673817.88 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3363, pruned_loss=0.08496, over 4110233.63 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3301, pruned_loss=0.08983, over 5654596.99 frames. ], batch size: 79, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:44:53,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1325, 1.2170, 1.1748, 1.0877], device='cuda:1'), covar=tensor([0.1991, 0.2032, 0.1554, 0.2044], device='cuda:1'), in_proj_covar=tensor([0.1977, 0.1915, 0.1843, 0.1990], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 06:45:00,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:45:08,743 INFO [optim.py:369] (1/2) 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 06:45:16,570 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 06:45:27,872 INFO [train.py:968] (1/2) Epoch 24, batch 2200, giga_loss[loss=0.262, simple_loss=0.3393, pruned_loss=0.09233, over 28515.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3303, pruned_loss=0.08811, over 5686078.78 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3363, pruned_loss=0.08485, over 4134514.68 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3309, pruned_loss=0.08952, over 5670104.88 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:46:08,954 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4696, 5.3060, 5.0408, 2.6507], device='cuda:1'), covar=tensor([0.0410, 0.0550, 0.0580, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.1237, 0.1146, 0.0961, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 06:46:09,349 INFO [train.py:968] (1/2) Epoch 24, batch 2250, giga_loss[loss=0.2199, simple_loss=0.3016, pruned_loss=0.06906, over 29065.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3294, pruned_loss=0.08791, over 5688917.09 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3363, pruned_loss=0.08476, over 4159873.82 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3298, pruned_loss=0.08912, over 5674577.88 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:46:31,066 INFO [optim.py:369] (1/2) 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 06:46:32,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 06:46:52,135 INFO [train.py:968] (1/2) Epoch 24, batch 2300, giga_loss[loss=0.2374, simple_loss=0.311, pruned_loss=0.08192, over 28904.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3275, pruned_loss=0.08693, over 5699757.27 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3364, pruned_loss=0.08462, over 4218412.42 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3275, pruned_loss=0.0881, over 5683177.63 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:46:53,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1051199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:47:18,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 06:47:20,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1051229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:47:21,521 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7344, 1.9249, 2.0757, 1.6278], device='cuda:1'), covar=tensor([0.3227, 0.2647, 0.2277, 0.3201], device='cuda:1'), in_proj_covar=tensor([0.1981, 0.1918, 0.1843, 0.1998], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 06:47:35,556 INFO [train.py:968] (1/2) Epoch 24, batch 2350, giga_loss[loss=0.2253, simple_loss=0.3034, pruned_loss=0.07362, over 28885.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3256, pruned_loss=0.08617, over 5706606.18 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3372, pruned_loss=0.08491, over 4242350.73 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3249, pruned_loss=0.08693, over 5691964.24 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:47:58,083 INFO [optim.py:369] (1/2) 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 06:48:16,532 INFO [train.py:968] (1/2) Epoch 24, batch 2400, giga_loss[loss=0.2198, simple_loss=0.3009, pruned_loss=0.06933, over 28316.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3244, pruned_loss=0.08573, over 5711470.88 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3379, pruned_loss=0.08513, over 4306226.28 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3229, pruned_loss=0.08621, over 5693653.95 frames. ], batch size: 65, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:48:55,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1051346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:48:56,428 INFO [train.py:968] (1/2) Epoch 24, batch 2450, giga_loss[loss=0.2283, simple_loss=0.3076, pruned_loss=0.07444, over 29015.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3236, pruned_loss=0.08539, over 5703460.94 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3384, pruned_loss=0.08531, over 4375927.23 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3213, pruned_loss=0.08568, over 5693492.58 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:49:16,326 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:49:18,179 INFO [optim.py:369] (1/2) 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 06:49:18,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1051375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:49:24,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2200, 1.7736, 1.4717, 0.4699], device='cuda:1'), covar=tensor([0.4915, 0.2626, 0.4700, 0.6562], device='cuda:1'), in_proj_covar=tensor([0.1761, 0.1655, 0.1597, 0.1428], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:49:36,914 INFO [train.py:968] (1/2) Epoch 24, batch 2500, giga_loss[loss=0.2991, simple_loss=0.3586, pruned_loss=0.1198, over 26828.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3219, pruned_loss=0.08471, over 5709503.87 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3386, pruned_loss=0.08535, over 4417770.76 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3195, pruned_loss=0.08491, over 5699356.69 frames. ], batch size: 555, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:49:42,045 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:49:50,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1766, 1.6808, 1.7793, 1.4260], device='cuda:1'), covar=tensor([0.2025, 0.1325, 0.2006, 0.1730], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0753, 0.0722, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 06:50:15,763 INFO [train.py:968] (1/2) Epoch 24, batch 2550, giga_loss[loss=0.2223, simple_loss=0.2991, pruned_loss=0.07278, over 29115.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3194, pruned_loss=0.08336, over 5710118.78 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3385, pruned_loss=0.08506, over 4458108.67 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.317, pruned_loss=0.08368, over 5706465.33 frames. ], batch size: 128, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:50:38,024 INFO [optim.py:369] (1/2) 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 06:50:57,819 INFO [train.py:968] (1/2) Epoch 24, batch 2600, giga_loss[loss=0.2145, simple_loss=0.29, pruned_loss=0.06947, over 28941.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3171, pruned_loss=0.08217, over 5717973.48 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3387, pruned_loss=0.08508, over 4478011.93 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3148, pruned_loss=0.08239, over 5713913.85 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:51:02,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 06:51:05,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2829, 1.9124, 1.3911, 0.5722], device='cuda:1'), covar=tensor([0.5712, 0.2862, 0.4758, 0.6558], device='cuda:1'), in_proj_covar=tensor([0.1763, 0.1655, 0.1599, 0.1430], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 06:51:37,873 INFO [train.py:968] (1/2) Epoch 24, batch 2650, libri_loss[loss=0.2911, simple_loss=0.3816, pruned_loss=0.1003, over 29187.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3164, pruned_loss=0.08162, over 5720470.17 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3392, pruned_loss=0.0851, over 4524866.21 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3135, pruned_loss=0.08166, over 5713295.59 frames. ], batch size: 97, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:51:58,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:52:16,398 INFO [train.py:968] (1/2) Epoch 24, batch 2700, libri_loss[loss=0.2578, simple_loss=0.3448, pruned_loss=0.08542, over 29554.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.3167, pruned_loss=0.08184, over 5726610.05 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3394, pruned_loss=0.08513, over 4577510.29 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3133, pruned_loss=0.08175, over 5715738.37 frames. ], batch size: 77, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:52:33,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-12 06:52:39,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4416, 1.3638, 1.4836, 1.1305], device='cuda:1'), covar=tensor([0.2240, 0.3713, 0.1773, 0.1962], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0708, 0.0964, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 06:52:43,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2907, 0.8020, 0.8609, 1.3505], device='cuda:1'), covar=tensor([0.0787, 0.0399, 0.0384, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 06:53:04,551 INFO [train.py:968] (1/2) Epoch 24, batch 2750, giga_loss[loss=0.2335, simple_loss=0.3109, pruned_loss=0.07805, over 28558.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3207, pruned_loss=0.08451, over 5722700.43 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.34, pruned_loss=0.08542, over 4590612.62 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3174, pruned_loss=0.08422, over 5712759.29 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:53:05,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 06:53:26,087 INFO [optim.py:369] (1/2) 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 06:53:47,709 INFO [train.py:968] (1/2) Epoch 24, batch 2800, giga_loss[loss=0.3241, simple_loss=0.3849, pruned_loss=0.1316, over 28938.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3257, pruned_loss=0.08782, over 5718465.12 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3401, pruned_loss=0.08548, over 4616451.40 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3227, pruned_loss=0.08757, over 5707536.37 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:53:54,206 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1051704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:06,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:08,509 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1051720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:09,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1051721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:20,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2650, 2.9690, 1.4975, 1.4312], device='cuda:1'), covar=tensor([0.1020, 0.0329, 0.0863, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0554, 0.0393, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 06:54:22,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4851, 1.7496, 1.4771, 1.5041], device='cuda:1'), covar=tensor([0.2462, 0.2473, 0.2738, 0.2179], device='cuda:1'), in_proj_covar=tensor([0.1540, 0.1111, 0.1361, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 06:54:34,110 INFO [train.py:968] (1/2) Epoch 24, batch 2850, giga_loss[loss=0.3013, simple_loss=0.3767, pruned_loss=0.113, over 28260.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3334, pruned_loss=0.09264, over 5716224.56 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3406, pruned_loss=0.08577, over 4648451.19 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3304, pruned_loss=0.09235, over 5703254.68 frames. ], batch size: 65, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:54:36,113 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:54:59,338 INFO [optim.py:369] (1/2) 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 06:55:06,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1171, 1.3428, 1.3338, 1.3589], device='cuda:1'), covar=tensor([0.0858, 0.0369, 0.0316, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 06:55:19,826 INFO [train.py:968] (1/2) Epoch 24, batch 2900, giga_loss[loss=0.2872, simple_loss=0.3626, pruned_loss=0.1059, over 28345.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3402, pruned_loss=0.09692, over 5711080.22 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3407, pruned_loss=0.0858, over 4679006.73 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3377, pruned_loss=0.0969, over 5696823.37 frames. ], batch size: 368, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 06:56:10,836 INFO [train.py:968] (1/2) Epoch 24, batch 2950, giga_loss[loss=0.28, simple_loss=0.3603, pruned_loss=0.09987, over 28982.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3445, pruned_loss=0.09806, over 5713601.56 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3408, pruned_loss=0.08599, over 4691220.84 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3425, pruned_loss=0.09801, over 5700742.56 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:56:22,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7395, 1.9939, 1.3990, 1.6191], device='cuda:1'), covar=tensor([0.1114, 0.0717, 0.1068, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0447, 0.0521, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 06:56:24,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1051864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:56:29,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8818, 3.6880, 3.4977, 2.0025], device='cuda:1'), covar=tensor([0.0694, 0.0893, 0.0829, 0.2324], device='cuda:1'), in_proj_covar=tensor([0.1241, 0.1151, 0.0966, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 06:56:29,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1051867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:56:36,502 INFO [optim.py:369] (1/2) 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 06:56:53,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1051896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:56:54,819 INFO [train.py:968] (1/2) Epoch 24, batch 3000, giga_loss[loss=0.3045, simple_loss=0.3707, pruned_loss=0.1192, over 28884.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3492, pruned_loss=0.1004, over 5712304.51 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3408, pruned_loss=0.08591, over 4719892.86 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3477, pruned_loss=0.1007, over 5698890.14 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:56:54,819 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 06:57:04,256 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 06:57:52,022 INFO [train.py:968] (1/2) Epoch 24, batch 3050, giga_loss[loss=0.2733, simple_loss=0.3543, pruned_loss=0.09613, over 29022.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3541, pruned_loss=0.1038, over 5691760.24 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3409, pruned_loss=0.08621, over 4759953.45 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3532, pruned_loss=0.1044, over 5674872.29 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 06:58:18,297 INFO [optim.py:369] (1/2) 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 06:58:18,888 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-12 06:58:35,422 INFO [train.py:968] (1/2) Epoch 24, batch 3100, giga_loss[loss=0.2523, simple_loss=0.3384, pruned_loss=0.08303, over 28859.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3501, pruned_loss=0.1005, over 5702075.81 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3402, pruned_loss=0.08598, over 4788184.37 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3502, pruned_loss=0.1016, over 5683952.28 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 06:58:39,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1052001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 06:59:19,694 INFO [train.py:968] (1/2) Epoch 24, batch 3150, giga_loss[loss=0.2269, simple_loss=0.3183, pruned_loss=0.06771, over 29067.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.09779, over 5707216.47 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.34, pruned_loss=0.08593, over 4799383.58 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3473, pruned_loss=0.09875, over 5691243.54 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 06:59:44,816 INFO [optim.py:369] (1/2) 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 06:59:46,932 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 3200, giga_loss[loss=0.2879, simple_loss=0.3601, pruned_loss=0.1079, over 28538.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.345, pruned_loss=0.09584, over 5709316.06 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3394, pruned_loss=0.0857, over 4821517.02 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3458, pruned_loss=0.09705, over 5701206.52 frames. ], batch size: 336, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:00:43,859 INFO [train.py:968] (1/2) Epoch 24, batch 3250, giga_loss[loss=0.2678, simple_loss=0.3522, pruned_loss=0.09167, over 28936.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3454, pruned_loss=0.09533, over 5711268.34 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3391, pruned_loss=0.08555, over 4843162.39 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3464, pruned_loss=0.09658, over 5701404.19 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:01:10,268 INFO [optim.py:369] (1/2) 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 07:01:26,059 INFO [train.py:968] (1/2) Epoch 24, batch 3300, giga_loss[loss=0.2851, simple_loss=0.3601, pruned_loss=0.1051, over 28756.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3473, pruned_loss=0.09633, over 5718219.26 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.339, pruned_loss=0.08544, over 4884814.66 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3485, pruned_loss=0.09787, over 5703096.70 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:01:49,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:01:51,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:01:52,568 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 3350, giga_loss[loss=0.2661, simple_loss=0.3455, pruned_loss=0.09339, over 28839.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3497, pruned_loss=0.09847, over 5712505.99 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3391, pruned_loss=0.08551, over 4905142.65 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3508, pruned_loss=0.09986, over 5697116.87 frames. ], batch size: 227, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:02:17,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:02:38,041 INFO [optim.py:369] (1/2) 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 07:02:43,876 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 07:02:49,912 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1413, 1.3067, 1.1150, 0.8831], device='cuda:1'), covar=tensor([0.1069, 0.0546, 0.1095, 0.1139], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0445, 0.0519, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 07:02:54,986 INFO [train.py:968] (1/2) Epoch 24, batch 3400, giga_loss[loss=0.2774, simple_loss=0.3536, pruned_loss=0.1006, over 29085.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3518, pruned_loss=0.1006, over 5703681.60 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3396, pruned_loss=0.08581, over 4910433.43 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3524, pruned_loss=0.1016, over 5697416.51 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:03:05,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8088, 2.0275, 1.9734, 1.6143], device='cuda:1'), covar=tensor([0.2644, 0.2432, 0.2565, 0.2636], device='cuda:1'), in_proj_covar=tensor([0.1984, 0.1924, 0.1854, 0.2006], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 07:03:12,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 07:03:37,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1052345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:03:39,417 INFO [train.py:968] (1/2) Epoch 24, batch 3450, giga_loss[loss=0.2407, simple_loss=0.3204, pruned_loss=0.08055, over 28516.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3512, pruned_loss=0.1001, over 5715390.29 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3393, pruned_loss=0.08564, over 4933207.62 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3522, pruned_loss=0.1014, over 5707687.73 frames. ], batch size: 71, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:04:04,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1052376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:04:04,562 INFO [optim.py:369] (1/2) 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 07:04:21,529 INFO [train.py:968] (1/2) Epoch 24, batch 3500, giga_loss[loss=0.2555, simple_loss=0.3368, pruned_loss=0.0871, over 28685.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3514, pruned_loss=0.1, over 5722511.45 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3395, pruned_loss=0.08566, over 4951216.53 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1013, over 5714100.05 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:04:58,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0771, 2.3141, 2.1962, 1.7924], device='cuda:1'), covar=tensor([0.2978, 0.2401, 0.2506, 0.2910], device='cuda:1'), in_proj_covar=tensor([0.1982, 0.1925, 0.1857, 0.2004], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 07:05:05,000 INFO [train.py:968] (1/2) Epoch 24, batch 3550, giga_loss[loss=0.2563, simple_loss=0.3387, pruned_loss=0.08693, over 28852.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3515, pruned_loss=0.09951, over 5721474.79 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.34, pruned_loss=0.08592, over 4974200.00 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3521, pruned_loss=0.1007, over 5711375.79 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:05:06,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-12 07:05:27,733 INFO [optim.py:369] (1/2) 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 07:05:44,216 INFO [train.py:968] (1/2) Epoch 24, batch 3600, libri_loss[loss=0.2062, simple_loss=0.2939, pruned_loss=0.05929, over 29504.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3515, pruned_loss=0.0987, over 5725089.10 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3398, pruned_loss=0.08587, over 5009814.03 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3526, pruned_loss=0.1002, over 5711203.06 frames. ], batch size: 70, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:05:44,637 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 07:06:02,988 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:06:05,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:06:27,154 INFO [train.py:968] (1/2) Epoch 24, batch 3650, giga_loss[loss=0.2741, simple_loss=0.3529, pruned_loss=0.09759, over 28719.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.351, pruned_loss=0.09796, over 5727011.34 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3394, pruned_loss=0.08573, over 5036751.52 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3526, pruned_loss=0.09963, over 5710788.92 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:06:31,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2606, 1.2431, 3.8497, 3.2282], device='cuda:1'), covar=tensor([0.1702, 0.2889, 0.0428, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0653, 0.0967, 0.0925], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 07:06:31,320 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:06:51,997 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.34 vs. limit=5.0
+2023-03-12 07:06:52,066 INFO [optim.py:369] (1/2) 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 07:07:11,530 INFO [train.py:968] (1/2) Epoch 24, batch 3700, giga_loss[loss=0.2655, simple_loss=0.3371, pruned_loss=0.09701, over 28825.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.349, pruned_loss=0.09697, over 5727517.24 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3393, pruned_loss=0.08567, over 5044099.71 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3504, pruned_loss=0.09848, over 5717254.33 frames. ], batch size: 99, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:07:14,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1052602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:07:55,136 INFO [train.py:968] (1/2) Epoch 24, batch 3750, giga_loss[loss=0.2427, simple_loss=0.3321, pruned_loss=0.07668, over 28994.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3485, pruned_loss=0.09724, over 5724740.45 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3405, pruned_loss=0.08633, over 5071390.42 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.349, pruned_loss=0.09831, over 5712917.07 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:08:00,209 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2200, 1.5079, 1.5055, 1.1006], device='cuda:1'), covar=tensor([0.1746, 0.2718, 0.1465, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0709, 0.0961, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 07:08:17,288 INFO [optim.py:369] (1/2) 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 07:08:31,486 INFO [train.py:968] (1/2) Epoch 24, batch 3800, giga_loss[loss=0.2429, simple_loss=0.3238, pruned_loss=0.08097, over 28957.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3462, pruned_loss=0.09607, over 5731214.74 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3403, pruned_loss=0.08628, over 5092205.82 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3469, pruned_loss=0.09718, over 5717733.13 frames. ], batch size: 106, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:08:52,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1052720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:08:58,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8926, 2.2634, 1.4962, 1.6675], device='cuda:1'), covar=tensor([0.1017, 0.0637, 0.1012, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0445, 0.0519, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 07:09:12,909 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:09:14,732 INFO [train.py:968] (1/2) Epoch 24, batch 3850, libri_loss[loss=0.2406, simple_loss=0.3177, pruned_loss=0.08177, over 29469.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3465, pruned_loss=0.09639, over 5738889.27 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.34, pruned_loss=0.08607, over 5104356.38 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09758, over 5725847.39 frames. ], batch size: 70, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:09:15,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:09:39,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:09:40,210 INFO [optim.py:369] (1/2) 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 07:09:56,356 INFO [train.py:968] (1/2) Epoch 24, batch 3900, giga_loss[loss=0.2426, simple_loss=0.3283, pruned_loss=0.07844, over 28830.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3475, pruned_loss=0.09719, over 5731063.22 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3401, pruned_loss=0.08617, over 5120398.83 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3484, pruned_loss=0.09838, over 5722684.02 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:10:18,803 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4003, 1.6741, 1.3635, 1.5683], device='cuda:1'), covar=tensor([0.0785, 0.0320, 0.0336, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 07:10:36,465 INFO [train.py:968] (1/2) Epoch 24, batch 3950, giga_loss[loss=0.2488, simple_loss=0.3359, pruned_loss=0.08091, over 29088.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3474, pruned_loss=0.09633, over 5729098.97 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3402, pruned_loss=0.08617, over 5134194.85 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3481, pruned_loss=0.09747, over 5720968.70 frames. ], batch size: 128, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:10:51,367 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1052863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:10:54,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1052866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:11:04,561 INFO [optim.py:369] (1/2) 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 07:11:16,554 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1052895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:11:18,303 INFO [train.py:968] (1/2) Epoch 24, batch 4000, giga_loss[loss=0.2455, simple_loss=0.3318, pruned_loss=0.0796, over 29034.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.345, pruned_loss=0.09432, over 5721389.88 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3403, pruned_loss=0.08624, over 5154413.80 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3457, pruned_loss=0.09551, over 5717219.41 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:11:59,938 INFO [train.py:968] (1/2) Epoch 24, batch 4050, giga_loss[loss=0.2812, simple_loss=0.3317, pruned_loss=0.1154, over 23631.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3448, pruned_loss=0.09475, over 5712040.46 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3409, pruned_loss=0.08671, over 5162180.03 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.345, pruned_loss=0.09548, over 5714045.06 frames. ], batch size: 705, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:12:03,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2468, 1.5021, 1.5345, 1.1211], device='cuda:1'), covar=tensor([0.1684, 0.2763, 0.1408, 0.1703], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0710, 0.0962, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 07:12:08,852 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1052960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:12:24,479 INFO [optim.py:369] (1/2) 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 07:12:40,077 INFO [train.py:968] (1/2) Epoch 24, batch 4100, giga_loss[loss=0.2396, simple_loss=0.3205, pruned_loss=0.07933, over 28517.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3438, pruned_loss=0.09501, over 5702781.30 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3412, pruned_loss=0.08702, over 5167579.10 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3438, pruned_loss=0.09548, over 5709647.77 frames. ], batch size: 71, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:13:19,920 INFO [train.py:968] (1/2) Epoch 24, batch 4150, giga_loss[loss=0.25, simple_loss=0.322, pruned_loss=0.08902, over 29010.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3402, pruned_loss=0.09321, over 5703704.81 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3415, pruned_loss=0.08721, over 5181872.60 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.34, pruned_loss=0.09355, over 5706047.23 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:13:42,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2701, 1.6058, 1.5804, 1.3954], device='cuda:1'), covar=tensor([0.2147, 0.1632, 0.2368, 0.1994], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0748, 0.0716, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 07:13:47,357 INFO [optim.py:369] (1/2) 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 07:14:03,064 INFO [train.py:968] (1/2) Epoch 24, batch 4200, giga_loss[loss=0.2681, simple_loss=0.3398, pruned_loss=0.09823, over 28699.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.0928, over 5695016.86 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3409, pruned_loss=0.08691, over 5188946.32 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3388, pruned_loss=0.09341, over 5702926.32 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:14:21,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:14:43,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6849, 4.3926, 1.9108, 1.7773], device='cuda:1'), covar=tensor([0.0932, 0.0479, 0.0889, 0.1270], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0554, 0.0392, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 07:14:43,879 INFO [train.py:968] (1/2) Epoch 24, batch 4250, giga_loss[loss=0.227, simple_loss=0.3132, pruned_loss=0.07044, over 28813.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3383, pruned_loss=0.09318, over 5691565.40 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3408, pruned_loss=0.08689, over 5190407.61 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3385, pruned_loss=0.09374, over 5703180.52 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:15:09,542 INFO [optim.py:369] (1/2) 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 07:15:27,080 INFO [train.py:968] (1/2) Epoch 24, batch 4300, giga_loss[loss=0.2261, simple_loss=0.3021, pruned_loss=0.07505, over 28825.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3367, pruned_loss=0.09303, over 5696260.65 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3409, pruned_loss=0.08714, over 5210673.99 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3367, pruned_loss=0.09346, over 5700665.61 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:16:06,329 INFO [train.py:968] (1/2) Epoch 24, batch 4350, giga_loss[loss=0.2653, simple_loss=0.3406, pruned_loss=0.095, over 28797.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3344, pruned_loss=0.09189, over 5696668.28 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3412, pruned_loss=0.08732, over 5211289.66 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3342, pruned_loss=0.09218, over 5707618.92 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:16:30,855 INFO [optim.py:369] (1/2) 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 07:16:46,187 INFO [train.py:968] (1/2) Epoch 24, batch 4400, giga_loss[loss=0.2259, simple_loss=0.297, pruned_loss=0.07739, over 28601.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3321, pruned_loss=0.091, over 5698725.73 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3411, pruned_loss=0.08726, over 5231799.82 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3318, pruned_loss=0.09142, over 5702002.83 frames. ], batch size: 92, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:16:46,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4539, 4.2938, 4.1316, 1.9495], device='cuda:1'), covar=tensor([0.0665, 0.0796, 0.0847, 0.2024], device='cuda:1'), in_proj_covar=tensor([0.1243, 0.1148, 0.0967, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 07:17:10,222 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 07:17:17,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1053335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:17:18,625 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8121, 2.7499, 1.6382, 0.8317], device='cuda:1'), covar=tensor([0.7903, 0.3463, 0.4295, 0.7620], device='cuda:1'), in_proj_covar=tensor([0.1773, 0.1667, 0.1615, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 07:17:26,643 INFO [train.py:968] (1/2) Epoch 24, batch 4450, giga_loss[loss=0.2083, simple_loss=0.2883, pruned_loss=0.06415, over 28931.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3294, pruned_loss=0.08934, over 5703114.38 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.341, pruned_loss=0.08728, over 5241206.54 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3289, pruned_loss=0.08974, over 5709772.52 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:17:50,300 INFO [optim.py:369] (1/2) 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 07:18:07,548 INFO [train.py:968] (1/2) Epoch 24, batch 4500, giga_loss[loss=0.2521, simple_loss=0.321, pruned_loss=0.09166, over 28431.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3298, pruned_loss=0.08909, over 5707362.19 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.341, pruned_loss=0.08745, over 5261071.30 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.329, pruned_loss=0.0893, over 5707272.73 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:18:55,145 INFO [train.py:968] (1/2) Epoch 24, batch 4550, giga_loss[loss=0.243, simple_loss=0.3254, pruned_loss=0.08028, over 28986.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3334, pruned_loss=0.09095, over 5701103.09 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3412, pruned_loss=0.08763, over 5270489.16 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3325, pruned_loss=0.091, over 5698520.24 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:19:19,490 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1053478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:20,663 INFO [optim.py:369] (1/2) 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 07:19:22,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1053481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:36,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1053495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:37,963 INFO [train.py:968] (1/2) Epoch 24, batch 4600, giga_loss[loss=0.2696, simple_loss=0.3556, pruned_loss=0.09175, over 29056.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3368, pruned_loss=0.09219, over 5708646.61 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3411, pruned_loss=0.08769, over 5289265.73 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.336, pruned_loss=0.09229, over 5701483.82 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:19:47,442 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:19:48,115 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1053510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:20:10,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6990, 1.9656, 1.3651, 1.4712], device='cuda:1'), covar=tensor([0.1011, 0.0676, 0.1073, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0447, 0.0521, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 07:20:21,606 INFO [train.py:968] (1/2) Epoch 24, batch 4650, libri_loss[loss=0.2718, simple_loss=0.3602, pruned_loss=0.09168, over 29524.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3398, pruned_loss=0.09326, over 5702939.44 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3413, pruned_loss=0.08782, over 5311885.28 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3389, pruned_loss=0.09344, over 5691268.51 frames. ], batch size: 82, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:20:31,647 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-12 07:20:52,204 INFO [optim.py:369] (1/2) 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 07:21:07,298 INFO [train.py:968] (1/2) Epoch 24, batch 4700, giga_loss[loss=0.2276, simple_loss=0.3221, pruned_loss=0.06658, over 28637.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3393, pruned_loss=0.09252, over 5701452.42 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3413, pruned_loss=0.08786, over 5317823.68 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3386, pruned_loss=0.09268, over 5690492.08 frames. ], batch size: 336, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:21:38,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1053638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:21:41,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1053641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:21:46,430 INFO [train.py:968] (1/2) Epoch 24, batch 4750, giga_loss[loss=0.2332, simple_loss=0.3048, pruned_loss=0.08079, over 28647.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3388, pruned_loss=0.09195, over 5710821.57 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3418, pruned_loss=0.08814, over 5343270.85 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3377, pruned_loss=0.09206, over 5697612.22 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:21:48,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4629, 2.0647, 1.4903, 1.7148], device='cuda:1'), covar=tensor([0.0758, 0.0268, 0.0328, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0118, 0.0117, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0063, 0.0110], device='cuda:1')
+2023-03-12 07:22:03,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1053670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:22:10,481 INFO [optim.py:369] (1/2) 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 07:22:25,315 INFO [train.py:968] (1/2) Epoch 24, batch 4800, giga_loss[loss=0.2487, simple_loss=0.3343, pruned_loss=0.08155, over 29015.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3385, pruned_loss=0.09173, over 5712418.77 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.08818, over 5367850.75 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3376, pruned_loss=0.0919, over 5694304.97 frames. ], batch size: 155, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:23:05,949 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:23:07,826 INFO [train.py:968] (1/2) Epoch 24, batch 4850, giga_loss[loss=0.2261, simple_loss=0.3157, pruned_loss=0.06826, over 29045.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3392, pruned_loss=0.09264, over 5704697.09 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3415, pruned_loss=0.08838, over 5374209.72 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3384, pruned_loss=0.09274, over 5694488.89 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:23:20,390 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-12 07:23:32,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:23:35,064 INFO [optim.py:369] (1/2) 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 07:23:50,947 INFO [train.py:968] (1/2) Epoch 24, batch 4900, giga_loss[loss=0.3256, simple_loss=0.3838, pruned_loss=0.1337, over 28796.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3416, pruned_loss=0.09394, over 5704032.36 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.08831, over 5392750.47 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.341, pruned_loss=0.09423, over 5689463.95 frames. ], batch size: 119, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:24:14,975 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-12 07:24:32,490 INFO [train.py:968] (1/2) Epoch 24, batch 4950, giga_loss[loss=0.2572, simple_loss=0.3377, pruned_loss=0.08833, over 28852.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3456, pruned_loss=0.09594, over 5713142.65 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08843, over 5407791.38 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3449, pruned_loss=0.09623, over 5696443.31 frames. ], batch size: 112, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:24:47,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3517, 1.5415, 1.5693, 1.1812], device='cuda:1'), covar=tensor([0.1742, 0.2430, 0.1455, 0.1683], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0707, 0.0961, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 07:24:51,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:25:01,752 INFO [optim.py:369] (1/2) 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 07:25:04,939 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1053884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:25:10,241 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1053890.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:25:15,839 INFO [train.py:968] (1/2) Epoch 24, batch 5000, libri_loss[loss=0.2462, simple_loss=0.3394, pruned_loss=0.0765, over 29521.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3478, pruned_loss=0.0969, over 5718856.32 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3424, pruned_loss=0.08866, over 5417372.48 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3468, pruned_loss=0.09711, over 5702345.63 frames. ], batch size: 84, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:25:48,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1311, 1.3922, 1.3162, 1.0615], device='cuda:1'), covar=tensor([0.2864, 0.2660, 0.1933, 0.2541], device='cuda:1'), in_proj_covar=tensor([0.1982, 0.1926, 0.1867, 0.2002], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 07:25:57,857 INFO [train.py:968] (1/2) Epoch 24, batch 5050, giga_loss[loss=0.2877, simple_loss=0.3641, pruned_loss=0.1056, over 28859.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3493, pruned_loss=0.09755, over 5721214.69 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3426, pruned_loss=0.08879, over 5424783.02 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3484, pruned_loss=0.09771, over 5705536.74 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:26:08,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-12 07:26:26,883 INFO [optim.py:369] (1/2) 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 07:26:37,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7706, 1.7987, 1.4744, 1.3497], device='cuda:1'), covar=tensor([0.0949, 0.0692, 0.1008, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0446, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 07:26:41,868 INFO [train.py:968] (1/2) Epoch 24, batch 5100, giga_loss[loss=0.2745, simple_loss=0.3554, pruned_loss=0.09675, over 28889.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3491, pruned_loss=0.09742, over 5726064.05 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3425, pruned_loss=0.08868, over 5428030.75 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3485, pruned_loss=0.0977, over 5713509.31 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:27:07,055 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:27:09,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:27:23,139 INFO [train.py:968] (1/2) Epoch 24, batch 5150, libri_loss[loss=0.2358, simple_loss=0.3198, pruned_loss=0.07594, over 29558.00 frames. ], tot_loss[loss=0.271, simple_loss=0.348, pruned_loss=0.09696, over 5729467.61 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3426, pruned_loss=0.08867, over 5440315.22 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3476, pruned_loss=0.09744, over 5716003.70 frames. ], batch size: 77, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:27:26,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-12 07:27:33,034 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:27:50,334 INFO [optim.py:369] (1/2) 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 07:27:54,633 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1054086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:28:04,144 INFO [train.py:968] (1/2) Epoch 24, batch 5200, giga_loss[loss=0.2214, simple_loss=0.3116, pruned_loss=0.06559, over 28886.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3462, pruned_loss=0.09626, over 5728960.72 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3427, pruned_loss=0.08867, over 5449630.71 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3459, pruned_loss=0.09678, over 5715033.44 frames. ], batch size: 174, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:28:24,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:28:46,610 INFO [train.py:968] (1/2) Epoch 24, batch 5250, giga_loss[loss=0.2337, simple_loss=0.3119, pruned_loss=0.07774, over 28881.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3423, pruned_loss=0.0944, over 5731407.59 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3427, pruned_loss=0.08864, over 5461063.32 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3422, pruned_loss=0.09499, over 5716191.20 frames. ], batch size: 99, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:28:50,345 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054152.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:29:12,825 INFO [optim.py:369] (1/2) 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 07:29:28,211 INFO [train.py:968] (1/2) Epoch 24, batch 5300, giga_loss[loss=0.2349, simple_loss=0.3143, pruned_loss=0.07772, over 28491.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3412, pruned_loss=0.09384, over 5727606.01 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.343, pruned_loss=0.0889, over 5468148.68 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3407, pruned_loss=0.09419, over 5714019.30 frames. ], batch size: 60, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:29:54,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-12 07:30:08,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:10,373 INFO [train.py:968] (1/2) Epoch 24, batch 5350, giga_loss[loss=0.2882, simple_loss=0.3613, pruned_loss=0.1075, over 26633.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3412, pruned_loss=0.09254, over 5724137.30 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3424, pruned_loss=0.08854, over 5481093.72 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3413, pruned_loss=0.09325, over 5708582.60 frames. ], batch size: 555, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:30:24,830 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:26,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054265.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:30:27,253 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:42,344 INFO [optim.py:369] (1/2) 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 07:30:48,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 07:30:52,904 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:30:52,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054295.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:30:55,831 INFO [train.py:968] (1/2) Epoch 24, batch 5400, giga_loss[loss=0.3257, simple_loss=0.3956, pruned_loss=0.1279, over 27671.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3421, pruned_loss=0.09239, over 5717773.01 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3425, pruned_loss=0.08866, over 5487529.82 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3421, pruned_loss=0.09288, over 5702732.77 frames. ], batch size: 472, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:30:56,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054298.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:31:20,271 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054327.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 07:31:38,484 INFO [train.py:968] (1/2) Epoch 24, batch 5450, giga_loss[loss=0.2583, simple_loss=0.3331, pruned_loss=0.09174, over 28987.00 frames. ], tot_loss[loss=0.264, simple_loss=0.342, pruned_loss=0.09304, over 5706455.14 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3429, pruned_loss=0.08891, over 5482811.39 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3417, pruned_loss=0.09329, over 5702319.84 frames. ], batch size: 145, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:32:09,266 INFO [optim.py:369] (1/2) 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 07:32:14,162 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:32:16,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:32:21,258 INFO [train.py:968] (1/2) Epoch 24, batch 5500, giga_loss[loss=0.3262, simple_loss=0.3859, pruned_loss=0.1332, over 27974.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3415, pruned_loss=0.09452, over 5697792.05 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3427, pruned_loss=0.08884, over 5476971.21 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3414, pruned_loss=0.09479, over 5702224.40 frames. ], batch size: 412, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:32:32,305 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054408.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:32:34,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054411.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:32:43,482 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:32:59,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1054440.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 07:33:06,664 INFO [train.py:968] (1/2) Epoch 24, batch 5550, giga_loss[loss=0.2796, simple_loss=0.3355, pruned_loss=0.1119, over 28516.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3402, pruned_loss=0.09503, over 5693221.94 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3427, pruned_loss=0.08891, over 5475289.01 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3401, pruned_loss=0.0953, over 5701763.90 frames. ], batch size: 85, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:33:18,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1054461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:33:34,826 INFO [optim.py:369] (1/2) 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 07:33:48,024 INFO [train.py:968] (1/2) Epoch 24, batch 5600, giga_loss[loss=0.2287, simple_loss=0.3103, pruned_loss=0.0736, over 28982.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3384, pruned_loss=0.09482, over 5695687.93 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3423, pruned_loss=0.0887, over 5486382.81 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3385, pruned_loss=0.0954, over 5698318.32 frames. ], batch size: 136, lr: 1.34e-03, grad_scale: 8.0
+2023-03-12 07:34:30,736 INFO [train.py:968] (1/2) Epoch 24, batch 5650, giga_loss[loss=0.2286, simple_loss=0.3102, pruned_loss=0.07352, over 28929.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3367, pruned_loss=0.09384, over 5700905.96 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08851, over 5495261.71 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3369, pruned_loss=0.09471, over 5703874.43 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:34:58,920 INFO [optim.py:369] (1/2) 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 07:35:09,652 INFO [train.py:968] (1/2) Epoch 24, batch 5700, giga_loss[loss=0.2483, simple_loss=0.3209, pruned_loss=0.08788, over 28676.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3338, pruned_loss=0.09222, over 5708803.42 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3416, pruned_loss=0.08834, over 5503704.94 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3342, pruned_loss=0.09318, over 5708194.70 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:35:17,164 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1054604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:35:19,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1054607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:35:24,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4666, 3.4761, 1.5799, 1.6520], device='cuda:1'), covar=tensor([0.0976, 0.0350, 0.0940, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0557, 0.0393, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 07:35:43,046 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 5750, giga_loss[loss=0.2196, simple_loss=0.3036, pruned_loss=0.06777, over 28786.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3289, pruned_loss=0.08958, over 5716363.19 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08838, over 5505912.75 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.329, pruned_loss=0.09032, over 5715911.61 frames. ], batch size: 284, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:36:22,886 INFO [optim.py:369] (1/2) 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 07:36:32,200 INFO [train.py:968] (1/2) Epoch 24, batch 5800, giga_loss[loss=0.2548, simple_loss=0.3365, pruned_loss=0.08655, over 28703.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3271, pruned_loss=0.08871, over 5721930.48 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3423, pruned_loss=0.08883, over 5519123.05 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3261, pruned_loss=0.08893, over 5716122.28 frames. ], batch size: 242, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:37:14,984 INFO [train.py:968] (1/2) Epoch 24, batch 5850, giga_loss[loss=0.2571, simple_loss=0.3432, pruned_loss=0.08546, over 28605.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3288, pruned_loss=0.08946, over 5722738.97 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.089, over 5523013.99 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3278, pruned_loss=0.08947, over 5716386.62 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:37:16,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5396, 1.8379, 1.4574, 1.7397], device='cuda:1'), covar=tensor([0.2667, 0.2789, 0.3226, 0.2375], device='cuda:1'), in_proj_covar=tensor([0.1536, 0.1107, 0.1354, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 07:37:45,223 INFO [optim.py:369] (1/2) 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 07:37:53,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4909, 1.6257, 1.2221, 1.2621], device='cuda:1'), covar=tensor([0.0950, 0.0599, 0.1042, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0447, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 07:37:56,219 INFO [train.py:968] (1/2) Epoch 24, batch 5900, giga_loss[loss=0.3811, simple_loss=0.419, pruned_loss=0.1716, over 26810.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3328, pruned_loss=0.09134, over 5715429.92 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08906, over 5521542.11 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3315, pruned_loss=0.09132, over 5715679.66 frames. ], batch size: 555, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:38:02,211 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4971, 2.2409, 1.6591, 0.7828], device='cuda:1'), covar=tensor([0.6219, 0.3007, 0.4237, 0.6491], device='cuda:1'), in_proj_covar=tensor([0.1780, 0.1670, 0.1617, 0.1445], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 07:38:05,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8374, 2.0377, 1.6366, 2.1106], device='cuda:1'), covar=tensor([0.2537, 0.2694, 0.3071, 0.2515], device='cuda:1'), in_proj_covar=tensor([0.1533, 0.1105, 0.1351, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 07:38:23,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2548, 3.0583, 1.4158, 1.4268], device='cuda:1'), covar=tensor([0.1015, 0.0366, 0.0960, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0557, 0.0393, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 07:38:39,542 INFO [train.py:968] (1/2) Epoch 24, batch 5950, giga_loss[loss=0.2899, simple_loss=0.3663, pruned_loss=0.1067, over 28896.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.336, pruned_loss=0.09226, over 5716426.57 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08921, over 5527032.98 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3346, pruned_loss=0.09215, over 5714402.81 frames. ], batch size: 213, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:38:43,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6970, 2.5708, 1.7098, 1.0204], device='cuda:1'), covar=tensor([0.8943, 0.4020, 0.3938, 0.6973], device='cuda:1'), in_proj_covar=tensor([0.1783, 0.1673, 0.1620, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 07:39:00,544 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1054872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:39:01,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4139, 1.6242, 1.3897, 1.5904], device='cuda:1'), covar=tensor([0.0744, 0.0324, 0.0333, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 07:39:10,101 INFO [optim.py:369] (1/2) 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 07:39:13,425 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 07:39:20,971 INFO [train.py:968] (1/2) Epoch 24, batch 6000, giga_loss[loss=0.2594, simple_loss=0.3408, pruned_loss=0.08904, over 28961.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3393, pruned_loss=0.09363, over 5705085.26 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08949, over 5521530.16 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3377, pruned_loss=0.09335, over 5711602.31 frames. ], batch size: 164, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:39:20,971 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 07:39:25,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0939, 1.2296, 3.4482, 3.1217], device='cuda:1'), covar=tensor([0.1921, 0.3078, 0.0528, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0652, 0.0971, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 07:39:29,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0853, 1.5722, 1.6052, 1.3965], device='cuda:1'), covar=tensor([0.2082, 0.1413, 0.1904, 0.1616], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0751, 0.0720, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 07:39:30,027 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 07:39:50,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3164, 2.6008, 2.3599, 1.8801], device='cuda:1'), covar=tensor([0.3197, 0.2358, 0.2641, 0.3275], device='cuda:1'), in_proj_covar=tensor([0.1979, 0.1927, 0.1864, 0.1993], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 07:40:13,550 INFO [train.py:968] (1/2) Epoch 24, batch 6050, giga_loss[loss=0.3399, simple_loss=0.3911, pruned_loss=0.1444, over 26622.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3417, pruned_loss=0.09513, over 5707051.71 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3437, pruned_loss=0.08955, over 5529064.11 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3402, pruned_loss=0.09496, over 5708691.17 frames. ], batch size: 555, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:40:47,543 INFO [optim.py:369] (1/2) 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 07:40:59,683 INFO [train.py:968] (1/2) Epoch 24, batch 6100, libri_loss[loss=0.2608, simple_loss=0.3256, pruned_loss=0.09795, over 29678.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3455, pruned_loss=0.09787, over 5692677.46 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3438, pruned_loss=0.08958, over 5522964.09 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3442, pruned_loss=0.09791, over 5706279.77 frames. ], batch size: 73, lr: 1.34e-03, grad_scale: 4.0
+2023-03-12 07:41:40,358 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-12 07:41:45,011 INFO [train.py:968] (1/2) Epoch 24, batch 6150, giga_loss[loss=0.3002, simple_loss=0.3756, pruned_loss=0.1124, over 28867.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3524, pruned_loss=0.1037, over 5698063.21 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08985, over 5534183.64 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3513, pruned_loss=0.1039, over 5703274.77 frames. ], batch size: 186, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:41:55,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:42:20,990 INFO [optim.py:369] (1/2) 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 07:42:35,113 INFO [train.py:968] (1/2) Epoch 24, batch 6200, giga_loss[loss=0.2937, simple_loss=0.3747, pruned_loss=0.1064, over 28594.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3578, pruned_loss=0.1081, over 5675998.69 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3434, pruned_loss=0.08955, over 5547693.02 frames. ], giga_tot_loss[loss=0.288, simple_loss=0.3579, pruned_loss=0.1091, over 5673100.20 frames. ], batch size: 307, lr: 1.34e-03, grad_scale: 2.0
+2023-03-12 07:42:46,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9482, 3.7617, 3.5888, 1.5701], device='cuda:1'), covar=tensor([0.0893, 0.1052, 0.1131, 0.2343], device='cuda:1'), in_proj_covar=tensor([0.1255, 0.1160, 0.0978, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 07:43:05,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2175, 1.2415, 1.1099, 0.8470], device='cuda:1'), covar=tensor([0.0792, 0.0407, 0.0841, 0.1100], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0449, 0.0520, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 07:43:11,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3627, 3.0753, 1.5277, 1.5203], device='cuda:1'), covar=tensor([0.0973, 0.0362, 0.0891, 0.1278], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0557, 0.0393, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 07:43:20,513 INFO [train.py:968] (1/2) Epoch 24, batch 6250, giga_loss[loss=0.313, simple_loss=0.3845, pruned_loss=0.1207, over 28767.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3645, pruned_loss=0.1126, over 5684809.23 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08978, over 5558492.28 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3646, pruned_loss=0.1138, over 5677006.15 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:43:58,349 INFO [optim.py:369] (1/2) 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 07:44:08,700 INFO [train.py:968] (1/2) Epoch 24, batch 6300, giga_loss[loss=0.301, simple_loss=0.3745, pruned_loss=0.1138, over 28963.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.37, pruned_loss=0.1176, over 5671780.94 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3444, pruned_loss=0.09, over 5555806.40 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3704, pruned_loss=0.1192, over 5672149.45 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:44:51,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:44:51,942 INFO [train.py:968] (1/2) Epoch 24, batch 6350, giga_loss[loss=0.4057, simple_loss=0.4245, pruned_loss=0.1935, over 23703.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.374, pruned_loss=0.1208, over 5667645.74 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08978, over 5559015.92 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.376, pruned_loss=0.1237, over 5670943.37 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:45:07,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-12 07:45:30,777 INFO [optim.py:369] (1/2) 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 07:45:46,504 INFO [train.py:968] (1/2) Epoch 24, batch 6400, giga_loss[loss=0.2965, simple_loss=0.3619, pruned_loss=0.1156, over 28373.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3778, pruned_loss=0.1246, over 5647888.43 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.344, pruned_loss=0.08972, over 5562782.34 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.38, pruned_loss=0.1275, over 5648910.49 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:46:39,516 INFO [train.py:968] (1/2) Epoch 24, batch 6450, giga_loss[loss=0.368, simple_loss=0.4124, pruned_loss=0.1618, over 28700.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3798, pruned_loss=0.1275, over 5644618.96 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08965, over 5565837.42 frames. ], giga_tot_loss[loss=0.3216, simple_loss=0.3823, pruned_loss=0.1304, over 5643816.49 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:46:52,945 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:47:21,968 INFO [optim.py:369] (1/2) 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 07:47:28,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1055390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:47:31,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1055393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:47:37,642 INFO [train.py:968] (1/2) Epoch 24, batch 6500, giga_loss[loss=0.3335, simple_loss=0.3914, pruned_loss=0.1378, over 28874.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3845, pruned_loss=0.1332, over 5609578.83 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3437, pruned_loss=0.08969, over 5559273.49 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3868, pruned_loss=0.1358, over 5616113.70 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:48:04,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1055422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:48:16,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:48:23,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8408, 4.6973, 4.4215, 2.1576], device='cuda:1'), covar=tensor([0.0694, 0.0793, 0.1050, 0.2066], device='cuda:1'), in_proj_covar=tensor([0.1263, 0.1168, 0.0988, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 07:48:31,522 INFO [train.py:968] (1/2) Epoch 24, batch 6550, giga_loss[loss=0.4296, simple_loss=0.4568, pruned_loss=0.2012, over 27924.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3879, pruned_loss=0.1365, over 5596686.72 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08968, over 5557358.55 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3912, pruned_loss=0.14, over 5605082.99 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:49:06,528 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-12 07:49:10,429 INFO [optim.py:369] (1/2) 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 07:49:17,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3190, 1.5399, 1.6054, 1.2008], device='cuda:1'), covar=tensor([0.1202, 0.2192, 0.1026, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0703, 0.0955, 0.0854], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 07:49:23,170 INFO [train.py:968] (1/2) Epoch 24, batch 6600, giga_loss[loss=0.318, simple_loss=0.3768, pruned_loss=0.1296, over 29004.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3885, pruned_loss=0.1369, over 5611647.98 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08929, over 5562452.25 frames. ], giga_tot_loss[loss=0.3375, simple_loss=0.3927, pruned_loss=0.1412, over 5614808.89 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:50:02,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:50:07,190 INFO [train.py:968] (1/2) Epoch 24, batch 6650, giga_loss[loss=0.3638, simple_loss=0.412, pruned_loss=0.1577, over 28083.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3863, pruned_loss=0.1354, over 5634717.94 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08925, over 5576900.04 frames. ], giga_tot_loss[loss=0.3374, simple_loss=0.3921, pruned_loss=0.1414, over 5626777.65 frames. ], batch size: 77, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:50:35,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1055575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:50:39,232 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:51:00,828 INFO [train.py:968] (1/2) Epoch 24, batch 6700, giga_loss[loss=0.3096, simple_loss=0.3689, pruned_loss=0.1252, over 28489.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3866, pruned_loss=0.1362, over 5634916.82 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3431, pruned_loss=0.08927, over 5579776.75 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3914, pruned_loss=0.1413, over 5626843.47 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:51:11,972 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1055607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:51:12,013 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4739, 1.8260, 1.5499, 1.5543], device='cuda:1'), covar=tensor([0.0639, 0.0278, 0.0280, 0.0667], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0118, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0071, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 07:51:17,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2144, 1.4541, 1.4950, 1.0930], device='cuda:1'), covar=tensor([0.1646, 0.2594, 0.1344, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.0907, 0.0704, 0.0954, 0.0853], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 07:51:43,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:51:53,401 INFO [train.py:968] (1/2) Epoch 24, batch 6750, giga_loss[loss=0.4667, simple_loss=0.4647, pruned_loss=0.2343, over 26540.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3852, pruned_loss=0.134, over 5639148.89 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3423, pruned_loss=0.08882, over 5587866.32 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.391, pruned_loss=0.1398, over 5626496.85 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 07:51:53,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1055648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:52:32,298 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 6800, giga_loss[loss=0.3108, simple_loss=0.3815, pruned_loss=0.1201, over 29049.00 frames. ], tot_loss[loss=0.3254, simple_loss=0.3849, pruned_loss=0.1329, over 5638308.60 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.0891, over 5594833.77 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3903, pruned_loss=0.1384, over 5622849.72 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:53:07,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-12 07:53:21,407 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:53:35,809 INFO [train.py:968] (1/2) Epoch 24, batch 6850, giga_loss[loss=0.2598, simple_loss=0.3398, pruned_loss=0.08993, over 28879.00 frames. ], tot_loss[loss=0.3255, simple_loss=0.3853, pruned_loss=0.1329, over 5620622.69 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08915, over 5597415.64 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3902, pruned_loss=0.1379, over 5606540.37 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:54:01,343 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 07:54:17,503 INFO [optim.py:369] (1/2) 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 07:54:29,995 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-12 07:54:32,646 INFO [train.py:968] (1/2) Epoch 24, batch 6900, giga_loss[loss=0.2988, simple_loss=0.3659, pruned_loss=0.1158, over 28007.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3829, pruned_loss=0.1305, over 5613456.84 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08923, over 5589703.57 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3868, pruned_loss=0.1346, over 5608610.29 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:54:38,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-12 07:55:10,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5724, 1.6147, 1.7879, 1.3655], device='cuda:1'), covar=tensor([0.1610, 0.2488, 0.1379, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.0905, 0.0702, 0.0952, 0.0851], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 07:55:23,672 INFO [train.py:968] (1/2) Epoch 24, batch 6950, giga_loss[loss=0.3063, simple_loss=0.3758, pruned_loss=0.1184, over 29081.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3811, pruned_loss=0.1279, over 5625879.28 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08921, over 5592881.63 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.3846, pruned_loss=0.1314, over 5619495.22 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:55:36,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3516, 1.3373, 3.6566, 3.2714], device='cuda:1'), covar=tensor([0.1556, 0.2720, 0.0451, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0653, 0.0974, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 07:55:53,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1055878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:55:56,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1390, 3.9902, 3.8006, 1.9370], device='cuda:1'), covar=tensor([0.0642, 0.0731, 0.0749, 0.2077], device='cuda:1'), in_proj_covar=tensor([0.1261, 0.1167, 0.0985, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 07:55:56,794 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:56:16,057 INFO [train.py:968] (1/2) Epoch 24, batch 7000, giga_loss[loss=0.2883, simple_loss=0.3632, pruned_loss=0.1067, over 28024.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3777, pruned_loss=0.1251, over 5642385.38 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3428, pruned_loss=0.08913, over 5597203.81 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.381, pruned_loss=0.1285, over 5634205.52 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:56:28,037 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1055910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:56:33,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1055916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:57:01,918 INFO [train.py:968] (1/2) Epoch 24, batch 7050, giga_loss[loss=0.335, simple_loss=0.3963, pruned_loss=0.1369, over 28943.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3747, pruned_loss=0.1227, over 5638465.22 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08904, over 5596587.98 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3788, pruned_loss=0.1266, over 5633709.55 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:57:02,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9957, 3.5084, 2.1949, 1.0796], device='cuda:1'), covar=tensor([0.7937, 0.3091, 0.4242, 0.7958], device='cuda:1'), in_proj_covar=tensor([0.1779, 0.1672, 0.1617, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 07:57:39,498 INFO [optim.py:369] (1/2) 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 07:57:50,453 INFO [train.py:968] (1/2) Epoch 24, batch 7100, giga_loss[loss=0.3448, simple_loss=0.3775, pruned_loss=0.156, over 23685.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3734, pruned_loss=0.1222, over 5646626.92 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3417, pruned_loss=0.0887, over 5604251.88 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3779, pruned_loss=0.1263, over 5637132.32 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:58:04,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:58:16,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:58:21,395 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-12 07:58:44,823 INFO [train.py:968] (1/2) Epoch 24, batch 7150, giga_loss[loss=0.2942, simple_loss=0.3778, pruned_loss=0.1053, over 28917.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3729, pruned_loss=0.1213, over 5659942.54 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.0888, over 5608682.91 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3769, pruned_loss=0.1251, over 5649330.93 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 07:58:58,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5369, 4.3895, 4.1551, 2.0504], device='cuda:1'), covar=tensor([0.0566, 0.0692, 0.0723, 0.1958], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1174, 0.0991, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 07:58:58,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5836, 2.2084, 1.5229, 0.8426], device='cuda:1'), covar=tensor([0.6516, 0.3164, 0.4471, 0.6890], device='cuda:1'), in_proj_covar=tensor([0.1781, 0.1675, 0.1617, 0.1450], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 07:58:58,483 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:59:01,183 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:59:31,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 07:59:32,025 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 7200, giga_loss[loss=0.2684, simple_loss=0.3364, pruned_loss=0.1001, over 28598.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3707, pruned_loss=0.119, over 5661102.29 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08913, over 5605687.96 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3739, pruned_loss=0.1223, over 5656093.20 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:00:30,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2432, 4.0766, 3.9123, 1.9473], device='cuda:1'), covar=tensor([0.0575, 0.0688, 0.0796, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.1268, 0.1175, 0.0992, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 08:00:41,104 INFO [train.py:968] (1/2) Epoch 24, batch 7250, libri_loss[loss=0.2633, simple_loss=0.3465, pruned_loss=0.09009, over 29651.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.37, pruned_loss=0.1168, over 5660645.14 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3422, pruned_loss=0.08896, over 5611980.75 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3733, pruned_loss=0.1201, over 5651797.72 frames. ], batch size: 91, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:00:45,816 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-12 08:00:47,707 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:00:51,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:01:01,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:01:06,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:01:21,618 INFO [optim.py:369] (1/2) 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 08:01:32,266 INFO [train.py:968] (1/2) Epoch 24, batch 7300, giga_loss[loss=0.3409, simple_loss=0.4071, pruned_loss=0.1374, over 28720.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3727, pruned_loss=0.1173, over 5666050.25 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3421, pruned_loss=0.08889, over 5615509.56 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.376, pruned_loss=0.1205, over 5656887.12 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:01:33,539 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:02:02,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6028, 1.7222, 1.6281, 1.3929], device='cuda:1'), covar=tensor([0.2774, 0.2543, 0.2154, 0.2587], device='cuda:1'), in_proj_covar=tensor([0.1997, 0.1949, 0.1876, 0.2010], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 08:02:24,416 INFO [train.py:968] (1/2) Epoch 24, batch 7350, giga_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 28948.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3732, pruned_loss=0.118, over 5673637.65 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08913, over 5625391.48 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3765, pruned_loss=0.1213, over 5658756.76 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:03:01,797 INFO [optim.py:369] (1/2) 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 08:03:12,841 INFO [train.py:968] (1/2) Epoch 24, batch 7400, giga_loss[loss=0.2853, simple_loss=0.3576, pruned_loss=0.1065, over 28258.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3712, pruned_loss=0.1169, over 5681264.33 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3423, pruned_loss=0.08914, over 5632973.11 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3748, pruned_loss=0.1203, over 5663960.31 frames. ], batch size: 77, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:03:13,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6767, 1.8856, 1.5889, 1.7421], device='cuda:1'), covar=tensor([0.2083, 0.2235, 0.2294, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.1538, 0.1110, 0.1359, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 08:04:01,575 INFO [train.py:968] (1/2) Epoch 24, batch 7450, giga_loss[loss=0.2605, simple_loss=0.3371, pruned_loss=0.09191, over 28836.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3694, pruned_loss=0.1163, over 5676350.39 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08911, over 5643463.29 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3735, pruned_loss=0.1204, over 5654303.57 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:04:36,784 INFO [optim.py:369] (1/2) 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 08:04:45,475 INFO [train.py:968] (1/2) Epoch 24, batch 7500, giga_loss[loss=0.3201, simple_loss=0.3832, pruned_loss=0.1285, over 29022.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3697, pruned_loss=0.1175, over 5670052.77 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08948, over 5633993.57 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3736, pruned_loss=0.1216, over 5660889.19 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:05:00,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:05:27,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9741, 1.4114, 2.6291, 2.7085], device='cuda:1'), covar=tensor([0.1730, 0.2415, 0.1166, 0.1419], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0655, 0.0974, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 08:05:34,503 INFO [train.py:968] (1/2) Epoch 24, batch 7550, giga_loss[loss=0.3042, simple_loss=0.3801, pruned_loss=0.1141, over 28947.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3685, pruned_loss=0.1169, over 5676714.84 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08936, over 5638745.77 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3724, pruned_loss=0.1208, over 5665981.03 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:05:45,044 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 08:05:51,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:05:55,782 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 08:06:12,796 INFO [optim.py:369] (1/2) 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 08:06:22,402 INFO [train.py:968] (1/2) Epoch 24, batch 7600, giga_loss[loss=0.3155, simple_loss=0.3595, pruned_loss=0.1358, over 23643.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3678, pruned_loss=0.1147, over 5691839.79 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08933, over 5644912.29 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3714, pruned_loss=0.1186, over 5679090.04 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:07:12,137 INFO [train.py:968] (1/2) Epoch 24, batch 7650, giga_loss[loss=0.2809, simple_loss=0.3567, pruned_loss=0.1026, over 28758.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3675, pruned_loss=0.1139, over 5693076.31 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08933, over 5639671.01 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.371, pruned_loss=0.1175, over 5688490.02 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:07:47,868 INFO [optim.py:369] (1/2) 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 08:07:54,268 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-12 08:07:56,325 INFO [train.py:968] (1/2) Epoch 24, batch 7700, giga_loss[loss=0.2871, simple_loss=0.3621, pruned_loss=0.1061, over 28824.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.367, pruned_loss=0.1142, over 5691814.29 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3427, pruned_loss=0.0893, over 5644567.13 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3705, pruned_loss=0.1176, over 5684972.70 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:08:06,508 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:08:09,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:08:39,496 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:08:46,096 INFO [train.py:968] (1/2) Epoch 24, batch 7750, libri_loss[loss=0.2746, simple_loss=0.3643, pruned_loss=0.09246, over 29381.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3652, pruned_loss=0.1132, over 5699232.69 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08928, over 5654709.41 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3686, pruned_loss=0.1169, over 5686165.49 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:09:26,839 INFO [optim.py:369] (1/2) 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 08:09:38,418 INFO [train.py:968] (1/2) Epoch 24, batch 7800, giga_loss[loss=0.3224, simple_loss=0.3824, pruned_loss=0.1312, over 28652.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3657, pruned_loss=0.1144, over 5681815.38 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08941, over 5648529.07 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3685, pruned_loss=0.1176, over 5677135.26 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:09:52,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5435, 2.0533, 1.8418, 1.7037], device='cuda:1'), covar=tensor([0.0753, 0.0271, 0.0274, 0.0811], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 08:10:23,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0698, 2.1387, 1.8725, 1.9342], device='cuda:1'), covar=tensor([0.1980, 0.2488, 0.2519, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0752, 0.0718, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 08:10:28,467 INFO [train.py:968] (1/2) Epoch 24, batch 7850, libri_loss[loss=0.2658, simple_loss=0.3215, pruned_loss=0.1051, over 29639.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3639, pruned_loss=0.1139, over 5693389.72 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3427, pruned_loss=0.08944, over 5657610.22 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3677, pruned_loss=0.1176, over 5682560.40 frames. ], batch size: 69, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:10:46,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8123, 2.0777, 1.7893, 1.9257], device='cuda:1'), covar=tensor([0.2470, 0.2465, 0.2749, 0.2087], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1114, 0.1362, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 08:11:07,420 INFO [optim.py:369] (1/2) 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 08:11:08,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1056790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:11:16,254 INFO [train.py:968] (1/2) Epoch 24, batch 7900, giga_loss[loss=0.2614, simple_loss=0.3425, pruned_loss=0.09018, over 28939.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3631, pruned_loss=0.1141, over 5692754.11 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3425, pruned_loss=0.08943, over 5654308.62 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3667, pruned_loss=0.1175, over 5687734.99 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:12:05,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-12 08:12:06,865 INFO [train.py:968] (1/2) Epoch 24, batch 7950, giga_loss[loss=0.2868, simple_loss=0.3469, pruned_loss=0.1133, over 28948.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3614, pruned_loss=0.1135, over 5688723.92 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08953, over 5648636.84 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3642, pruned_loss=0.1164, over 5690423.20 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:12:10,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5650, 1.8217, 1.4847, 1.8060], device='cuda:1'), covar=tensor([0.3099, 0.3109, 0.3499, 0.2475], device='cuda:1'), in_proj_covar=tensor([0.1542, 0.1112, 0.1360, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 08:12:18,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7369, 2.5671, 1.6064, 0.9794], device='cuda:1'), covar=tensor([0.8686, 0.3871, 0.4440, 0.7476], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1682, 0.1620, 0.1454], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 08:12:39,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3716, 1.7166, 1.3158, 1.6280], device='cuda:1'), covar=tensor([0.0793, 0.0324, 0.0347, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 08:12:40,011 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:12:46,243 INFO [optim.py:369] (1/2) 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 08:12:53,891 INFO [train.py:968] (1/2) Epoch 24, batch 8000, giga_loss[loss=0.2945, simple_loss=0.3534, pruned_loss=0.1178, over 28182.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3606, pruned_loss=0.113, over 5691182.63 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.08942, over 5651634.42 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3633, pruned_loss=0.1161, over 5691035.43 frames. ], batch size: 77, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:13:12,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:13:30,393 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1056933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:13:33,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1056936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:13:44,738 INFO [train.py:968] (1/2) Epoch 24, batch 8050, giga_loss[loss=0.341, simple_loss=0.3989, pruned_loss=0.1416, over 27579.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3624, pruned_loss=0.1144, over 5682617.22 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.08963, over 5654142.71 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3645, pruned_loss=0.117, over 5680456.52 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:14:00,213 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1056964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:14:01,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1056965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:14:25,417 INFO [optim.py:369] (1/2) 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 08:14:29,666 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.37 vs. limit=5.0
+2023-03-12 08:14:34,227 INFO [train.py:968] (1/2) Epoch 24, batch 8100, giga_loss[loss=0.2859, simple_loss=0.36, pruned_loss=0.1059, over 28936.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.363, pruned_loss=0.114, over 5681248.31 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08964, over 5659525.82 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3649, pruned_loss=0.1164, over 5675146.43 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:15:23,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3415, 1.4436, 1.2404, 1.5725], device='cuda:1'), covar=tensor([0.0791, 0.0353, 0.0363, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 08:15:23,699 INFO [train.py:968] (1/2) Epoch 24, batch 8150, giga_loss[loss=0.284, simple_loss=0.3651, pruned_loss=0.1014, over 28838.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3634, pruned_loss=0.1137, over 5678196.48 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.0896, over 5665303.45 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3655, pruned_loss=0.1163, over 5668301.45 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:15:54,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3355, 3.1613, 3.0158, 1.4493], device='cuda:1'), covar=tensor([0.0979, 0.1140, 0.1030, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.1274, 0.1178, 0.0994, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 08:16:05,690 INFO [optim.py:369] (1/2) 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 08:16:13,516 INFO [train.py:968] (1/2) Epoch 24, batch 8200, giga_loss[loss=0.2739, simple_loss=0.3455, pruned_loss=0.1011, over 28791.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3635, pruned_loss=0.1138, over 5686713.69 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3432, pruned_loss=0.08973, over 5671344.40 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3656, pruned_loss=0.1163, over 5673657.59 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:17:05,643 INFO [train.py:968] (1/2) Epoch 24, batch 8250, giga_loss[loss=0.2847, simple_loss=0.3518, pruned_loss=0.1088, over 28865.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3671, pruned_loss=0.1169, over 5685805.37 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08992, over 5674539.62 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.369, pruned_loss=0.1192, over 5672858.17 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:17:47,589 INFO [optim.py:369] (1/2) 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 08:17:49,813 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:17:55,128 INFO [train.py:968] (1/2) Epoch 24, batch 8300, giga_loss[loss=0.3091, simple_loss=0.3627, pruned_loss=0.1277, over 28442.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3683, pruned_loss=0.1185, over 5692436.57 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.344, pruned_loss=0.09031, over 5679659.42 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.121, over 5677942.40 frames. ], batch size: 65, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:18:19,672 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-12 08:18:45,939 INFO [train.py:968] (1/2) Epoch 24, batch 8350, libri_loss[loss=0.2469, simple_loss=0.337, pruned_loss=0.07837, over 29515.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3699, pruned_loss=0.121, over 5665316.16 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.09, over 5667216.79 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.373, pruned_loss=0.1246, over 5664200.49 frames. ], batch size: 89, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:18:53,856 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:27,367 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:27,740 INFO [optim.py:369] (1/2) 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 08:19:29,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:35,635 INFO [train.py:968] (1/2) Epoch 24, batch 8400, libri_loss[loss=0.2699, simple_loss=0.3547, pruned_loss=0.09254, over 29539.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3706, pruned_loss=0.122, over 5669653.86 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3436, pruned_loss=0.09012, over 5672943.53 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3735, pruned_loss=0.1255, over 5663278.40 frames. ], batch size: 82, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:19:35,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:19:51,458 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.07 vs. limit=2.0
+2023-03-12 08:20:14,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:20:21,732 INFO [train.py:968] (1/2) Epoch 24, batch 8450, giga_loss[loss=0.3291, simple_loss=0.3893, pruned_loss=0.1345, over 29055.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3699, pruned_loss=0.1216, over 5670247.20 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3434, pruned_loss=0.09006, over 5679719.71 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3732, pruned_loss=0.1254, over 5658944.95 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:20:58,859 INFO [optim.py:369] (1/2) 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 08:21:04,866 INFO [train.py:968] (1/2) Epoch 24, batch 8500, giga_loss[loss=0.3171, simple_loss=0.3893, pruned_loss=0.1225, over 28732.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3681, pruned_loss=0.1194, over 5681350.07 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08991, over 5687504.73 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3718, pruned_loss=0.1235, over 5665082.60 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:21:07,177 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:11,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:15,086 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5156, 2.2061, 1.6066, 0.7639], device='cuda:1'), covar=tensor([0.6205, 0.3302, 0.4071, 0.7086], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1685, 0.1614, 0.1451], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 08:21:39,952 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:41,591 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:45,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:21:55,518 INFO [train.py:968] (1/2) Epoch 24, batch 8550, giga_loss[loss=0.3088, simple_loss=0.3815, pruned_loss=0.118, over 28711.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3667, pruned_loss=0.1176, over 5673974.94 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3431, pruned_loss=0.08999, over 5691153.26 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.37, pruned_loss=0.1212, over 5657551.74 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:22:12,048 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:22:25,883 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:22:28,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:22:31,861 INFO [optim.py:369] (1/2) 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 08:22:40,159 INFO [train.py:968] (1/2) Epoch 24, batch 8600, giga_loss[loss=0.2908, simple_loss=0.354, pruned_loss=0.1138, over 28179.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3659, pruned_loss=0.117, over 5664950.81 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3436, pruned_loss=0.09023, over 5677152.94 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3685, pruned_loss=0.1202, over 5664863.29 frames. ], batch size: 77, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:22:57,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:23:26,707 INFO [train.py:968] (1/2) Epoch 24, batch 8650, giga_loss[loss=0.3343, simple_loss=0.3794, pruned_loss=0.1446, over 28277.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1152, over 5667939.92 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3436, pruned_loss=0.09034, over 5674379.87 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3653, pruned_loss=0.1185, over 5669548.39 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:23:44,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:24:08,555 INFO [optim.py:369] (1/2) 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 08:24:15,744 INFO [train.py:968] (1/2) Epoch 24, batch 8700, giga_loss[loss=0.3145, simple_loss=0.3728, pruned_loss=0.1281, over 28684.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3633, pruned_loss=0.1165, over 5655053.73 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3436, pruned_loss=0.09032, over 5672032.22 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3663, pruned_loss=0.1201, over 5658294.85 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:24:25,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7130, 1.8665, 1.8286, 1.7205], device='cuda:1'), covar=tensor([0.2132, 0.1901, 0.1584, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.1999, 0.1955, 0.1882, 0.2020], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 08:24:44,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2962, 1.4927, 1.3927, 1.2069], device='cuda:1'), covar=tensor([0.2451, 0.2623, 0.1820, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.1998, 0.1956, 0.1882, 0.2021], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 08:24:46,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 08:25:08,780 INFO [train.py:968] (1/2) Epoch 24, batch 8750, giga_loss[loss=0.3425, simple_loss=0.4008, pruned_loss=0.1421, over 27888.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3649, pruned_loss=0.1177, over 5643839.24 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3435, pruned_loss=0.09035, over 5666990.19 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3677, pruned_loss=0.121, over 5650010.23 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:25:22,407 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:25:33,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1057673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:25:48,239 INFO [optim.py:369] (1/2) 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 08:25:57,774 INFO [train.py:968] (1/2) Epoch 24, batch 8800, libri_loss[loss=0.2608, simple_loss=0.3351, pruned_loss=0.0933, over 29558.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3685, pruned_loss=0.1177, over 5657401.61 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3438, pruned_loss=0.09047, over 5671963.55 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3713, pruned_loss=0.1212, over 5657194.17 frames. ], batch size: 77, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:26:09,477 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:26:12,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:26:40,876 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:26:44,909 INFO [train.py:968] (1/2) Epoch 24, batch 8850, giga_loss[loss=0.2795, simple_loss=0.3571, pruned_loss=0.1009, over 28599.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3711, pruned_loss=0.1176, over 5667597.59 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3438, pruned_loss=0.09051, over 5674284.83 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3736, pruned_loss=0.1207, over 5665090.44 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:26:56,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6330, 1.7529, 1.3208, 1.3542], device='cuda:1'), covar=tensor([0.0995, 0.0630, 0.1017, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0448, 0.0519, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 08:27:19,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-12 08:27:21,158 INFO [optim.py:369] (1/2) 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 08:27:29,691 INFO [train.py:968] (1/2) Epoch 24, batch 8900, giga_loss[loss=0.3683, simple_loss=0.4169, pruned_loss=0.1599, over 28489.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3713, pruned_loss=0.1179, over 5668796.18 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.343, pruned_loss=0.09005, over 5681046.60 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3751, pruned_loss=0.1218, over 5660614.35 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:27:36,501 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:36,568 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:38,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:40,856 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1057812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:44,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1057816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:47,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1057819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:27:55,635 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2985, 0.8509, 0.8784, 1.3503], device='cuda:1'), covar=tensor([0.0763, 0.0402, 0.0382, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 08:28:04,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:28:17,249 INFO [train.py:968] (1/2) Epoch 24, batch 8950, giga_loss[loss=0.2681, simple_loss=0.3479, pruned_loss=0.09415, over 28673.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.373, pruned_loss=0.1198, over 5662070.61 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3429, pruned_loss=0.09017, over 5685646.76 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3768, pruned_loss=0.1234, over 5651170.60 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:28:17,482 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1057848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:28:35,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1316, 1.3063, 1.1511, 0.8458], device='cuda:1'), covar=tensor([0.1087, 0.0527, 0.1100, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0449, 0.0520, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 08:28:56,202 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 9000, giga_loss[loss=0.3224, simple_loss=0.375, pruned_loss=0.1349, over 28273.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3723, pruned_loss=0.1199, over 5664233.08 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3425, pruned_loss=0.09007, over 5687011.88 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3764, pruned_loss=0.1237, over 5653671.86 frames. ], batch size: 65, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:29:03,340 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 08:29:12,611 INFO [train.py:1012] (1/2) Epoch 24, validation: loss=0.2047, simple_loss=0.3121, pruned_loss=0.04866, over 944034.00 frames. 
+2023-03-12 08:29:12,612 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 08:29:15,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 08:30:00,340 INFO [train.py:968] (1/2) Epoch 24, batch 9050, giga_loss[loss=0.3528, simple_loss=0.4038, pruned_loss=0.1509, over 28624.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3703, pruned_loss=0.1195, over 5660575.89 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3425, pruned_loss=0.09006, over 5694505.99 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3747, pruned_loss=0.1236, over 5644252.39 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:30:41,787 INFO [optim.py:369] (1/2) 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 08:30:47,956 INFO [train.py:968] (1/2) Epoch 24, batch 9100, giga_loss[loss=0.2781, simple_loss=0.3473, pruned_loss=0.1045, over 28880.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3676, pruned_loss=0.1178, over 5662702.15 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3422, pruned_loss=0.08982, over 5698885.33 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3719, pruned_loss=0.1219, over 5645248.34 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:31:40,502 INFO [train.py:968] (1/2) Epoch 24, batch 9150, giga_loss[loss=0.3285, simple_loss=0.3845, pruned_loss=0.1363, over 28534.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3679, pruned_loss=0.1192, over 5663062.92 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3424, pruned_loss=0.08985, over 5700804.00 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3713, pruned_loss=0.1226, over 5647337.20 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:32:25,185 INFO [optim.py:369] (1/2) 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 08:32:26,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2214, 1.8004, 1.3864, 0.4189], device='cuda:1'), covar=tensor([0.4869, 0.2997, 0.4222, 0.6518], device='cuda:1'), in_proj_covar=tensor([0.1775, 0.1675, 0.1612, 0.1445], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 08:32:30,479 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 08:32:32,031 INFO [train.py:968] (1/2) Epoch 24, batch 9200, giga_loss[loss=0.3506, simple_loss=0.4104, pruned_loss=0.1454, over 28939.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1198, over 5663889.48 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3421, pruned_loss=0.08963, over 5704794.00 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.372, pruned_loss=0.1233, over 5647085.09 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:33:23,437 INFO [train.py:968] (1/2) Epoch 24, batch 9250, giga_loss[loss=0.3111, simple_loss=0.3752, pruned_loss=0.1234, over 28990.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3685, pruned_loss=0.1205, over 5658520.54 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3418, pruned_loss=0.08965, over 5708739.24 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3721, pruned_loss=0.1238, over 5640878.14 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:33:50,683 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 08:33:56,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1058181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:34:04,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1058187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:34:08,447 INFO [optim.py:369] (1/2) 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 08:34:13,321 INFO [train.py:968] (1/2) Epoch 24, batch 9300, giga_loss[loss=0.31, simple_loss=0.3724, pruned_loss=0.1238, over 28617.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3663, pruned_loss=0.1196, over 5666851.03 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3416, pruned_loss=0.08949, over 5711734.96 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3696, pruned_loss=0.1228, over 5649612.43 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:34:27,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9241, 1.9638, 1.6787, 2.2857], device='cuda:1'), covar=tensor([0.2418, 0.2789, 0.3052, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.1542, 0.1113, 0.1363, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 08:34:44,132 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5125, 1.7533, 1.3964, 1.4394], device='cuda:1'), covar=tensor([0.2605, 0.2777, 0.3126, 0.2484], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1114, 0.1363, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 08:35:04,906 INFO [train.py:968] (1/2) Epoch 24, batch 9350, libri_loss[loss=0.2599, simple_loss=0.335, pruned_loss=0.09246, over 29559.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3651, pruned_loss=0.1184, over 5662825.18 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3416, pruned_loss=0.08942, over 5716989.24 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3683, pruned_loss=0.1218, over 5642716.87 frames. ], batch size: 79, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:35:06,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5387, 2.1797, 1.6158, 0.8304], device='cuda:1'), covar=tensor([0.6290, 0.3022, 0.4168, 0.6631], device='cuda:1'), in_proj_covar=tensor([0.1774, 0.1675, 0.1609, 0.1444], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 08:35:44,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-12 08:35:47,649 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 9400, giga_loss[loss=0.2756, simple_loss=0.357, pruned_loss=0.09712, over 28973.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3657, pruned_loss=0.1174, over 5668279.82 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.342, pruned_loss=0.08956, over 5717118.05 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3682, pruned_loss=0.1204, over 5651417.40 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:36:18,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1058324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:21,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1058327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:23,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1058330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:25,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1058333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:40,116 INFO [train.py:968] (1/2) Epoch 24, batch 9450, giga_loss[loss=0.2899, simple_loss=0.3526, pruned_loss=0.1136, over 28997.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3674, pruned_loss=0.1184, over 5670783.95 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08969, over 5718414.63 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3695, pruned_loss=0.1212, over 5655308.45 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:36:52,925 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1058356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:36:58,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9235, 1.1038, 1.1071, 0.8906], device='cuda:1'), covar=tensor([0.2275, 0.2507, 0.1458, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.2004, 0.1952, 0.1886, 0.2026], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 08:36:58,719 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1058362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:37:29,023 INFO [optim.py:369] (1/2) 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 08:37:33,242 INFO [train.py:968] (1/2) Epoch 24, batch 9500, giga_loss[loss=0.281, simple_loss=0.3623, pruned_loss=0.09987, over 28913.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3688, pruned_loss=0.12, over 5658701.74 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08965, over 5718470.56 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3705, pruned_loss=0.1224, over 5646292.91 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:38:23,275 INFO [train.py:968] (1/2) Epoch 24, batch 9550, libri_loss[loss=0.3508, simple_loss=0.4138, pruned_loss=0.1439, over 19440.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3702, pruned_loss=0.1179, over 5657949.31 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3426, pruned_loss=0.0897, over 5711371.19 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3719, pruned_loss=0.1202, over 5654485.83 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:38:42,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:38:54,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-12 08:39:00,191 INFO [optim.py:369] (1/2) 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 08:39:04,898 INFO [train.py:968] (1/2) Epoch 24, batch 9600, giga_loss[loss=0.3725, simple_loss=0.4269, pruned_loss=0.1591, over 28758.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3714, pruned_loss=0.117, over 5658887.19 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08956, over 5707056.67 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3738, pruned_loss=0.12, over 5658517.29 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:39:52,238 INFO [train.py:968] (1/2) Epoch 24, batch 9650, giga_loss[loss=0.2515, simple_loss=0.3441, pruned_loss=0.07945, over 28872.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3727, pruned_loss=0.1169, over 5663035.91 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3426, pruned_loss=0.08968, over 5703272.87 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3756, pruned_loss=0.1202, over 5664156.70 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:40:32,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1983, 4.0236, 3.8395, 1.8432], device='cuda:1'), covar=tensor([0.0656, 0.0784, 0.0812, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1179, 0.0995, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 08:40:37,989 INFO [optim.py:369] (1/2) 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 08:40:43,876 INFO [train.py:968] (1/2) Epoch 24, batch 9700, giga_loss[loss=0.4141, simple_loss=0.436, pruned_loss=0.1961, over 26539.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3765, pruned_loss=0.1209, over 5666920.54 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08957, over 5706165.24 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3794, pruned_loss=0.124, over 5664676.72 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:41:32,710 INFO [train.py:968] (1/2) Epoch 24, batch 9750, libri_loss[loss=0.21, simple_loss=0.2977, pruned_loss=0.06115, over 29581.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3781, pruned_loss=0.1234, over 5655613.96 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.08944, over 5700348.53 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3812, pruned_loss=0.1265, over 5657685.84 frames. ], batch size: 76, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:42:16,164 INFO [optim.py:369] (1/2) 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 08:42:20,487 INFO [train.py:968] (1/2) Epoch 24, batch 9800, giga_loss[loss=0.3151, simple_loss=0.3806, pruned_loss=0.1248, over 28722.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3783, pruned_loss=0.124, over 5656762.81 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08954, over 5702554.31 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3817, pruned_loss=0.1275, over 5654972.04 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:42:23,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3339, 1.1834, 3.8175, 3.3221], device='cuda:1'), covar=tensor([0.1602, 0.2850, 0.0481, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0658, 0.0974, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 08:43:05,226 INFO [train.py:968] (1/2) Epoch 24, batch 9850, giga_loss[loss=0.2613, simple_loss=0.3481, pruned_loss=0.08731, over 28569.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3758, pruned_loss=0.122, over 5658289.41 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.08949, over 5697582.52 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3794, pruned_loss=0.1255, over 5659768.14 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:43:24,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5223, 1.6646, 1.7206, 1.3278], device='cuda:1'), covar=tensor([0.2171, 0.2872, 0.1792, 0.2044], device='cuda:1'), in_proj_covar=tensor([0.0909, 0.0709, 0.0957, 0.0854], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 08:43:30,627 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:43:49,501 INFO [optim.py:369] (1/2) 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 08:43:52,932 INFO [train.py:968] (1/2) Epoch 24, batch 9900, giga_loss[loss=0.2999, simple_loss=0.3734, pruned_loss=0.1131, over 28726.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.374, pruned_loss=0.1187, over 5663677.20 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3419, pruned_loss=0.08929, over 5697867.27 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3777, pruned_loss=0.1222, over 5664303.10 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:44:05,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:44:20,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:44:39,033 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 9950, giga_loss[loss=0.3594, simple_loss=0.4142, pruned_loss=0.1523, over 28594.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3747, pruned_loss=0.1181, over 5669255.54 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3422, pruned_loss=0.0894, over 5701115.68 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3779, pruned_loss=0.1212, over 5666262.13 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:45:26,773 INFO [optim.py:369] (1/2) 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 08:45:30,086 INFO [train.py:968] (1/2) Epoch 24, batch 10000, giga_loss[loss=0.2661, simple_loss=0.3419, pruned_loss=0.09519, over 29022.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3767, pruned_loss=0.1205, over 5663299.72 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.0892, over 5697654.58 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3803, pruned_loss=0.124, over 5662272.80 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:46:23,678 INFO [train.py:968] (1/2) Epoch 24, batch 10050, giga_loss[loss=0.2723, simple_loss=0.3501, pruned_loss=0.09727, over 28901.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3756, pruned_loss=0.1203, over 5660283.97 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.08897, over 5701921.30 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3794, pruned_loss=0.1238, over 5655128.90 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:46:38,579 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1058963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:46:54,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-12 08:47:06,443 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 08:47:07,980 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1058990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:10,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1058993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:10,629 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-12 08:47:10,691 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 10100, giga_loss[loss=0.3158, simple_loss=0.3769, pruned_loss=0.1274, over 29061.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3729, pruned_loss=0.1198, over 5648728.97 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3417, pruned_loss=0.08894, over 5697661.81 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3767, pruned_loss=0.1233, over 5647366.84 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:47:41,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:42,359 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:47:44,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6265, 1.9053, 1.2978, 1.3926], device='cuda:1'), covar=tensor([0.0992, 0.0534, 0.1092, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0449, 0.0521, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 08:47:47,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4387, 1.2145, 4.8767, 3.6244], device='cuda:1'), covar=tensor([0.1790, 0.2969, 0.0409, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0660, 0.0978, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 08:48:06,871 INFO [train.py:968] (1/2) Epoch 24, batch 10150, giga_loss[loss=0.3626, simple_loss=0.4095, pruned_loss=0.1579, over 28281.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3707, pruned_loss=0.1189, over 5651625.02 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3417, pruned_loss=0.08893, over 5690674.58 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.374, pruned_loss=0.122, over 5655358.48 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:48:55,043 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 10200, giga_loss[loss=0.2856, simple_loss=0.3594, pruned_loss=0.1059, over 28582.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.369, pruned_loss=0.1188, over 5644249.45 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3419, pruned_loss=0.08901, over 5692565.18 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3717, pruned_loss=0.1215, over 5645266.77 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:49:18,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-12 08:49:48,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:49:49,071 INFO [train.py:968] (1/2) Epoch 24, batch 10250, giga_loss[loss=0.3036, simple_loss=0.3709, pruned_loss=0.1181, over 28978.00 frames. ], tot_loss[loss=0.304, simple_loss=0.369, pruned_loss=0.1194, over 5643964.48 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3422, pruned_loss=0.08909, over 5688824.63 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3716, pruned_loss=0.1223, over 5646652.65 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:50:21,548 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-12 08:50:23,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:50:31,887 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:50:32,894 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 10300, giga_loss[loss=0.32, simple_loss=0.3795, pruned_loss=0.1303, over 27470.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3682, pruned_loss=0.1188, over 5652921.79 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08933, over 5696984.60 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3707, pruned_loss=0.122, over 5646195.10 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:50:36,615 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:51:19,275 INFO [train.py:968] (1/2) Epoch 24, batch 10350, giga_loss[loss=0.2715, simple_loss=0.3566, pruned_loss=0.09317, over 28906.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3644, pruned_loss=0.1142, over 5674972.67 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08909, over 5703274.43 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3674, pruned_loss=0.1178, over 5662845.98 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 08:51:59,988 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:00,574 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:02,925 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:04,296 INFO [optim.py:369] (1/2) 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 08:52:06,459 INFO [train.py:968] (1/2) Epoch 24, batch 10400, giga_loss[loss=0.2354, simple_loss=0.3182, pruned_loss=0.07625, over 29010.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3602, pruned_loss=0.1107, over 5668258.17 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08893, over 5709397.68 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3634, pruned_loss=0.1145, over 5651628.93 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:52:30,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:37,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:39,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:44,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:45,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3011, 1.3967, 1.3024, 1.4586], device='cuda:1'), covar=tensor([0.0738, 0.0384, 0.0341, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0120, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 08:52:48,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:50,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5616, 1.7743, 1.5922, 1.6443], device='cuda:1'), covar=tensor([0.2001, 0.2202, 0.2560, 0.2142], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0753, 0.0719, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 08:52:50,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:52:52,918 INFO [train.py:968] (1/2) Epoch 24, batch 10450, giga_loss[loss=0.2604, simple_loss=0.3344, pruned_loss=0.09317, over 28990.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3599, pruned_loss=0.1099, over 5675046.27 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08903, over 5711443.41 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3628, pruned_loss=0.1135, over 5658643.01 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:53:01,022 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-12 08:53:06,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:53:11,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4605, 1.6263, 1.6767, 1.2717], device='cuda:1'), covar=tensor([0.1778, 0.2629, 0.1459, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0711, 0.0962, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 08:53:18,652 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:53:40,096 INFO [optim.py:369] (1/2) 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 08:53:45,330 INFO [train.py:968] (1/2) Epoch 24, batch 10500, giga_loss[loss=0.3621, simple_loss=0.4043, pruned_loss=0.16, over 28941.00 frames. ], tot_loss[loss=0.2898, simple_loss=0.3588, pruned_loss=0.1104, over 5667469.00 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08938, over 5706529.77 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3612, pruned_loss=0.1135, over 5657497.14 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:53:46,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:54:31,959 INFO [train.py:968] (1/2) Epoch 24, batch 10550, giga_loss[loss=0.2602, simple_loss=0.3344, pruned_loss=0.09305, over 28499.00 frames. ], tot_loss[loss=0.2876, simple_loss=0.3564, pruned_loss=0.1094, over 5669736.92 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3427, pruned_loss=0.08912, over 5712414.37 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3591, pruned_loss=0.1129, over 5655168.81 frames. ], batch size: 65, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:55:02,136 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:55:05,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:55:15,277 INFO [optim.py:369] (1/2) 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 08:55:18,359 INFO [train.py:968] (1/2) Epoch 24, batch 10600, giga_loss[loss=0.3204, simple_loss=0.3782, pruned_loss=0.1313, over 27878.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3588, pruned_loss=0.1108, over 5673847.44 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08942, over 5716224.79 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3609, pruned_loss=0.1138, over 5657767.82 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:55:31,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:00,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:02,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:05,147 INFO [train.py:968] (1/2) Epoch 24, batch 10650, giga_loss[loss=0.2821, simple_loss=0.3563, pruned_loss=0.104, over 28461.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3603, pruned_loss=0.1109, over 5674628.97 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08938, over 5722663.63 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3626, pruned_loss=0.1141, over 5654388.52 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:56:08,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2487, 1.4301, 1.4683, 1.2789], device='cuda:1'), covar=tensor([0.1984, 0.1884, 0.2445, 0.2038], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0752, 0.0719, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 08:56:23,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 08:56:26,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:56:33,927 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:56:57,276 INFO [train.py:968] (1/2) Epoch 24, batch 10700, giga_loss[loss=0.2776, simple_loss=0.3406, pruned_loss=0.1073, over 28683.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3618, pruned_loss=0.1126, over 5668961.23 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.0892, over 5726406.09 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3642, pruned_loss=0.1158, over 5648394.73 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:57:45,422 INFO [train.py:968] (1/2) Epoch 24, batch 10750, giga_loss[loss=0.2858, simple_loss=0.3569, pruned_loss=0.1074, over 28896.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3618, pruned_loss=0.1133, over 5663161.12 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08919, over 5728411.51 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5644674.67 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 08:58:02,452 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1059666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:58:29,632 INFO [optim.py:369] (1/2) 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 08:58:33,985 INFO [train.py:968] (1/2) Epoch 24, batch 10800, giga_loss[loss=0.291, simple_loss=0.3541, pruned_loss=0.114, over 28668.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3632, pruned_loss=0.1143, over 5668174.56 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3433, pruned_loss=0.08927, over 5729549.98 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3649, pruned_loss=0.1168, over 5651384.23 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 08:58:48,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:58:50,324 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:59:21,247 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 08:59:24,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6192, 1.8304, 1.3044, 1.3415], device='cuda:1'), covar=tensor([0.1076, 0.0661, 0.1105, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0449, 0.0519, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 08:59:24,935 INFO [train.py:968] (1/2) Epoch 24, batch 10850, giga_loss[loss=0.2917, simple_loss=0.3619, pruned_loss=0.1108, over 28705.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.366, pruned_loss=0.1159, over 5667724.10 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.343, pruned_loss=0.08907, over 5733103.07 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3681, pruned_loss=0.1188, over 5649635.00 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:00:01,602 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 24, batch 10900, giga_loss[loss=0.3863, simple_loss=0.4245, pruned_loss=0.174, over 27661.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3678, pruned_loss=0.1173, over 5667752.43 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3432, pruned_loss=0.08911, over 5733367.86 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3697, pruned_loss=0.1199, over 5652306.55 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:00:23,204 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1059809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:23,685 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:27,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1059812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:00:56,350 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1059841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:01:02,337 INFO [train.py:968] (1/2) Epoch 24, batch 10950, giga_loss[loss=0.2774, simple_loss=0.3432, pruned_loss=0.1058, over 28536.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3694, pruned_loss=0.119, over 5676372.31 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3434, pruned_loss=0.08906, over 5736659.96 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3712, pruned_loss=0.1216, over 5660132.59 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:01:12,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3725, 1.3991, 1.2650, 1.4821], device='cuda:1'), covar=tensor([0.0755, 0.0355, 0.0341, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 09:01:29,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3070, 1.3917, 1.3705, 1.2568], device='cuda:1'), covar=tensor([0.2288, 0.2009, 0.1829, 0.2167], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1962, 0.1887, 0.2024], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 09:01:50,952 INFO [optim.py:369] (1/2) 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 09:01:53,659 INFO [train.py:968] (1/2) Epoch 24, batch 11000, giga_loss[loss=0.3003, simple_loss=0.3804, pruned_loss=0.1101, over 28282.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3715, pruned_loss=0.1198, over 5677754.92 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3437, pruned_loss=0.08924, over 5739051.10 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3731, pruned_loss=0.1222, over 5662002.56 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:02:35,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4877, 1.6582, 1.3710, 1.6223], device='cuda:1'), covar=tensor([0.0759, 0.0322, 0.0327, 0.0816], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 09:02:47,640 INFO [train.py:968] (1/2) Epoch 24, batch 11050, giga_loss[loss=0.3281, simple_loss=0.3966, pruned_loss=0.1298, over 28773.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3717, pruned_loss=0.119, over 5665066.89 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.344, pruned_loss=0.08947, over 5733156.13 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3731, pruned_loss=0.1212, over 5656311.41 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:03:19,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1059978.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:03:39,299 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 11100, giga_loss[loss=0.3389, simple_loss=0.4025, pruned_loss=0.1377, over 27943.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3722, pruned_loss=0.1201, over 5656821.81 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08964, over 5736037.95 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3736, pruned_loss=0.1221, over 5646564.57 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:04:00,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-12 09:04:37,548 INFO [train.py:968] (1/2) Epoch 24, batch 11150, libri_loss[loss=0.269, simple_loss=0.353, pruned_loss=0.09249, over 29558.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3717, pruned_loss=0.1207, over 5645172.70 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.08972, over 5739625.09 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3733, pruned_loss=0.1229, over 5632019.74 frames. ], batch size: 79, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:05:28,473 INFO [optim.py:369] (1/2) 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 09:05:31,556 INFO [train.py:968] (1/2) Epoch 24, batch 11200, giga_loss[loss=0.2908, simple_loss=0.3585, pruned_loss=0.1115, over 28825.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3697, pruned_loss=0.1198, over 5650308.43 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08962, over 5741429.55 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5637154.03 frames. ], batch size: 285, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:05:38,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:06:20,682 INFO [train.py:968] (1/2) Epoch 24, batch 11250, libri_loss[loss=0.2491, simple_loss=0.3411, pruned_loss=0.07855, over 27962.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3677, pruned_loss=0.1185, over 5653490.04 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3441, pruned_loss=0.08948, over 5744988.69 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.37, pruned_loss=0.1215, over 5636538.21 frames. ], batch size: 116, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:06:28,961 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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 09:07:04,320 INFO [train.py:968] (1/2) Epoch 24, batch 11300, giga_loss[loss=0.2927, simple_loss=0.365, pruned_loss=0.1102, over 28937.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3674, pruned_loss=0.1185, over 5663284.82 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3442, pruned_loss=0.08952, over 5739965.18 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3697, pruned_loss=0.1216, over 5651292.03 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:07:55,370 INFO [train.py:968] (1/2) Epoch 24, batch 11350, giga_loss[loss=0.2861, simple_loss=0.3556, pruned_loss=0.1083, over 28826.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3674, pruned_loss=0.1193, over 5662537.32 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3442, pruned_loss=0.08947, over 5743044.32 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3697, pruned_loss=0.1224, over 5648823.47 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:08:40,520 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:08:45,554 INFO [train.py:968] (1/2) Epoch 24, batch 11400, giga_loss[loss=0.3297, simple_loss=0.3918, pruned_loss=0.1338, over 28839.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3678, pruned_loss=0.1202, over 5646441.36 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3446, pruned_loss=0.08972, over 5736536.93 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3699, pruned_loss=0.1231, over 5639815.34 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:08:50,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:08:50,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4757, 1.8255, 1.4680, 1.4419], device='cuda:1'), covar=tensor([0.2495, 0.2442, 0.2777, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.1545, 0.1114, 0.1361, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 09:08:52,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:09:14,274 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:09:18,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:09:20,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:09:31,064 INFO [train.py:968] (1/2) Epoch 24, batch 11450, libri_loss[loss=0.2098, simple_loss=0.2929, pruned_loss=0.06332, over 29646.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1214, over 5649146.68 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.08963, over 5726441.93 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3725, pruned_loss=0.1251, over 5649555.19 frames. ], batch size: 69, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:09:36,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060353.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:09:42,995 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:10:25,176 INFO [optim.py:369] (1/2) 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 09:10:25,189 INFO [train.py:968] (1/2) Epoch 24, batch 11500, giga_loss[loss=0.3687, simple_loss=0.4022, pruned_loss=0.1676, over 23361.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3718, pruned_loss=0.1241, over 5632351.57 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08961, over 5727364.72 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3744, pruned_loss=0.1273, over 5631305.08 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 09:11:08,551 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 11550, giga_loss[loss=0.2843, simple_loss=0.3609, pruned_loss=0.1038, over 28981.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3714, pruned_loss=0.1237, over 5645483.21 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08964, over 5725288.50 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3736, pruned_loss=0.1264, over 5645909.42 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 09:11:46,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:11:48,768 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:12:01,092 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060496.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:12:02,094 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 11600, libri_loss[loss=0.2632, simple_loss=0.3476, pruned_loss=0.0894, over 25925.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3708, pruned_loss=0.1229, over 5652685.33 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08969, over 5727517.58 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3734, pruned_loss=0.126, over 5649013.47 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:12:04,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060499.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:12:21,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3001, 1.3586, 1.2557, 1.2706], device='cuda:1'), covar=tensor([0.1910, 0.1814, 0.1631, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.2013, 0.1971, 0.1896, 0.2036], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 09:12:35,610 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060528.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:12:54,299 INFO [train.py:968] (1/2) Epoch 24, batch 11650, giga_loss[loss=0.2949, simple_loss=0.3601, pruned_loss=0.1148, over 28560.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3708, pruned_loss=0.1226, over 5652785.98 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3438, pruned_loss=0.08945, over 5731016.71 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3738, pruned_loss=0.126, over 5645276.69 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:13:42,807 INFO [optim.py:369] (1/2) 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 09:13:42,820 INFO [train.py:968] (1/2) Epoch 24, batch 11700, giga_loss[loss=0.2814, simple_loss=0.3571, pruned_loss=0.1028, over 28802.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3695, pruned_loss=0.1204, over 5665017.85 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3439, pruned_loss=0.08941, over 5730492.36 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.1239, over 5658022.25 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:14:07,445 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:14:12,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:14:31,467 INFO [train.py:968] (1/2) Epoch 24, batch 11750, giga_loss[loss=0.3251, simple_loss=0.3859, pruned_loss=0.1322, over 28987.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3715, pruned_loss=0.1224, over 5662558.91 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3439, pruned_loss=0.08925, over 5736358.54 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3748, pruned_loss=0.1265, over 5648935.80 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:14:38,588 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1060655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:14:54,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:15:19,868 INFO [optim.py:369] (1/2) 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 09:15:19,880 INFO [train.py:968] (1/2) Epoch 24, batch 11800, giga_loss[loss=0.304, simple_loss=0.3691, pruned_loss=0.1194, over 28941.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3733, pruned_loss=0.1238, over 5662097.19 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3443, pruned_loss=0.08948, over 5738415.95 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3766, pruned_loss=0.1281, over 5646808.22 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:16:11,743 INFO [train.py:968] (1/2) Epoch 24, batch 11850, giga_loss[loss=0.3243, simple_loss=0.3849, pruned_loss=0.1319, over 28003.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3741, pruned_loss=0.1251, over 5650612.47 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3443, pruned_loss=0.08948, over 5738415.95 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3767, pruned_loss=0.1285, over 5638712.87 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:16:24,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 09:16:40,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:16:47,686 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1060783.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:17:01,251 INFO [optim.py:369] (1/2) 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 09:17:01,265 INFO [train.py:968] (1/2) Epoch 24, batch 11900, libri_loss[loss=0.2588, simple_loss=0.3389, pruned_loss=0.08931, over 29547.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3751, pruned_loss=0.1243, over 5656275.38 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3442, pruned_loss=0.08947, over 5741148.97 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3779, pruned_loss=0.1279, over 5642277.69 frames. ], batch size: 79, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:17:16,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:17:18,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1060814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:17:18,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:17:22,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-12 09:17:48,370 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 11950, giga_loss[loss=0.2759, simple_loss=0.3518, pruned_loss=0.1, over 28965.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.375, pruned_loss=0.1236, over 5650644.21 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3446, pruned_loss=0.08963, over 5733691.59 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3774, pruned_loss=0.1267, over 5644808.23 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:18:01,002 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 24, batch 12000, giga_loss[loss=0.3096, simple_loss=0.379, pruned_loss=0.1201, over 28627.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3732, pruned_loss=0.1221, over 5648706.37 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3444, pruned_loss=0.08944, over 5738121.37 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3761, pruned_loss=0.1257, over 5637575.22 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:18:41,172 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 09:18:50,078 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 09:19:36,409 INFO [train.py:968] (1/2) Epoch 24, batch 12050, giga_loss[loss=0.417, simple_loss=0.4483, pruned_loss=0.1929, over 27569.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3727, pruned_loss=0.1219, over 5644473.41 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3448, pruned_loss=0.08971, over 5720373.26 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3749, pruned_loss=0.1249, over 5651509.75 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:19:41,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-12 09:19:43,625 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1060957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:19:45,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1060960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:19:53,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-12 09:20:16,186 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 12100, giga_loss[loss=0.3387, simple_loss=0.3721, pruned_loss=0.1527, over 23600.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3737, pruned_loss=0.1229, over 5628177.73 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3449, pruned_loss=0.08969, over 5716794.51 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3762, pruned_loss=0.1263, over 5634035.35 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:20:25,251 INFO [optim.py:369] (1/2) 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 09:20:27,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:20:29,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:20:50,273 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061024.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:20:59,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:21:10,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4224, 1.5749, 1.6596, 1.3579], device='cuda:1'), covar=tensor([0.2141, 0.2073, 0.2233, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.2011, 0.1965, 0.1885, 0.2028], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 09:21:15,844 INFO [train.py:968] (1/2) Epoch 24, batch 12150, giga_loss[loss=0.2683, simple_loss=0.3441, pruned_loss=0.09618, over 28522.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3746, pruned_loss=0.1238, over 5644616.67 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3449, pruned_loss=0.08971, over 5718701.04 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3768, pruned_loss=0.1268, over 5646879.58 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:22:03,121 INFO [train.py:968] (1/2) Epoch 24, batch 12200, giga_loss[loss=0.308, simple_loss=0.3719, pruned_loss=0.122, over 28652.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.373, pruned_loss=0.1229, over 5657819.79 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3452, pruned_loss=0.08988, over 5713878.88 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3751, pruned_loss=0.1258, over 5663281.69 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:22:04,023 INFO [optim.py:369] (1/2) 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 09:22:54,482 INFO [train.py:968] (1/2) Epoch 24, batch 12250, giga_loss[loss=0.307, simple_loss=0.3771, pruned_loss=0.1184, over 28914.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3728, pruned_loss=0.1234, over 5659413.47 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3449, pruned_loss=0.08984, over 5717598.43 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3754, pruned_loss=0.1263, over 5659369.06 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:22:58,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:23:03,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061158.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:23:40,817 INFO [train.py:968] (1/2) Epoch 24, batch 12300, giga_loss[loss=0.3045, simple_loss=0.3678, pruned_loss=0.1206, over 28454.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5667652.62 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3446, pruned_loss=0.08981, over 5722781.25 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3762, pruned_loss=0.1266, over 5661422.31 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:23:41,820 INFO [optim.py:369] (1/2) 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 09:23:43,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3016, 1.8676, 1.3351, 0.5660], device='cuda:1'), covar=tensor([0.4658, 0.2526, 0.3978, 0.6100], device='cuda:1'), in_proj_covar=tensor([0.1781, 0.1683, 0.1620, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 09:24:28,838 INFO [train.py:968] (1/2) Epoch 24, batch 12350, giga_loss[loss=0.3076, simple_loss=0.3794, pruned_loss=0.118, over 28923.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3743, pruned_loss=0.1245, over 5657179.43 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3447, pruned_loss=0.08998, over 5716687.25 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3774, pruned_loss=0.1279, over 5655975.00 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:24:51,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6854, 1.7583, 1.8826, 1.4783], device='cuda:1'), covar=tensor([0.1960, 0.2586, 0.1594, 0.1905], device='cuda:1'), in_proj_covar=tensor([0.0913, 0.0710, 0.0961, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 09:25:16,297 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:18,671 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:22,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:22,915 INFO [train.py:968] (1/2) Epoch 24, batch 12400, giga_loss[loss=0.2785, simple_loss=0.3509, pruned_loss=0.103, over 28705.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3718, pruned_loss=0.1216, over 5668715.47 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3449, pruned_loss=0.09007, over 5718237.46 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3745, pruned_loss=0.1246, over 5665803.22 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:25:23,495 INFO [optim.py:369] (1/2) 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 09:25:26,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061301.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:25:28,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:25:46,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8385, 3.6729, 3.4966, 1.7220], device='cuda:1'), covar=tensor([0.0747, 0.0876, 0.0840, 0.2154], device='cuda:1'), in_proj_covar=tensor([0.1277, 0.1179, 0.1000, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 09:25:51,084 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:25:52,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2984, 1.3320, 3.9643, 3.2512], device='cuda:1'), covar=tensor([0.1723, 0.2793, 0.0443, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0658, 0.0975, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 09:25:56,945 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:26:10,548 INFO [train.py:968] (1/2) Epoch 24, batch 12450, giga_loss[loss=0.283, simple_loss=0.3585, pruned_loss=0.1038, over 28544.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3717, pruned_loss=0.1205, over 5671726.20 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3451, pruned_loss=0.09025, over 5722931.17 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1235, over 5664085.83 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:26:36,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4715, 2.1913, 1.7084, 0.6503], device='cuda:1'), covar=tensor([0.5802, 0.3002, 0.4374, 0.6638], device='cuda:1'), in_proj_covar=tensor([0.1787, 0.1686, 0.1624, 0.1450], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 09:26:57,459 INFO [train.py:968] (1/2) Epoch 24, batch 12500, libri_loss[loss=0.2525, simple_loss=0.3422, pruned_loss=0.08143, over 29283.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3702, pruned_loss=0.119, over 5678949.59 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3451, pruned_loss=0.09022, over 5724619.73 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3725, pruned_loss=0.1216, over 5670899.93 frames. ], batch size: 94, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:26:58,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061399.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:26:58,898 INFO [optim.py:369] (1/2) 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 09:27:35,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0824, 2.2007, 1.8368, 2.1693], device='cuda:1'), covar=tensor([0.2326, 0.2617, 0.2895, 0.2408], device='cuda:1'), in_proj_covar=tensor([0.1543, 0.1113, 0.1359, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 09:27:45,450 INFO [train.py:968] (1/2) Epoch 24, batch 12550, giga_loss[loss=0.2771, simple_loss=0.3464, pruned_loss=0.1039, over 28885.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.368, pruned_loss=0.1178, over 5671062.67 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3449, pruned_loss=0.09015, over 5725539.81 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3706, pruned_loss=0.1207, over 5662557.64 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:28:23,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:28:36,543 INFO [train.py:968] (1/2) Epoch 24, batch 12600, giga_loss[loss=0.286, simple_loss=0.3509, pruned_loss=0.1105, over 28533.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1182, over 5663938.85 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.345, pruned_loss=0.0901, over 5726636.92 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3706, pruned_loss=0.1215, over 5654508.53 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:28:37,828 INFO [optim.py:369] (1/2) 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 09:29:15,302 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061542.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:29:17,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061545.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:29:19,131 INFO [train.py:968] (1/2) Epoch 24, batch 12650, giga_loss[loss=0.284, simple_loss=0.3549, pruned_loss=0.1065, over 28972.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.366, pruned_loss=0.1171, over 5669269.56 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3454, pruned_loss=0.0901, over 5720863.74 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3688, pruned_loss=0.1209, over 5663751.46 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:29:27,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 09:29:45,860 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061574.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:30:08,237 INFO [train.py:968] (1/2) Epoch 24, batch 12700, libri_loss[loss=0.2302, simple_loss=0.313, pruned_loss=0.07372, over 29485.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.364, pruned_loss=0.1166, over 5686651.54 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3455, pruned_loss=0.09025, over 5725161.00 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3666, pruned_loss=0.1201, over 5677214.12 frames. ], batch size: 70, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:30:10,252 INFO [optim.py:369] (1/2) 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 09:30:30,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8633, 1.7970, 2.0395, 1.6413], device='cuda:1'), covar=tensor([0.1808, 0.2539, 0.1446, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0708, 0.0959, 0.0855], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 09:30:55,580 INFO [train.py:968] (1/2) Epoch 24, batch 12750, giga_loss[loss=0.3124, simple_loss=0.3695, pruned_loss=0.1277, over 28901.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.363, pruned_loss=0.1167, over 5688394.66 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3456, pruned_loss=0.09034, over 5727592.00 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3651, pruned_loss=0.1198, over 5678160.92 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:31:12,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:31:33,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 09:31:42,169 INFO [train.py:968] (1/2) Epoch 24, batch 12800, giga_loss[loss=0.2795, simple_loss=0.3464, pruned_loss=0.1063, over 28863.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3622, pruned_loss=0.1159, over 5689477.02 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3456, pruned_loss=0.09042, over 5734771.87 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3646, pruned_loss=0.1193, over 5673296.23 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:31:43,835 INFO [optim.py:369] (1/2) 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 09:32:29,060 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 12850, giga_loss[loss=0.2763, simple_loss=0.3567, pruned_loss=0.0979, over 28764.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.361, pruned_loss=0.1132, over 5688681.54 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3452, pruned_loss=0.09022, over 5737537.25 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3634, pruned_loss=0.1163, over 5673102.41 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:32:48,121 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 09:33:28,360 INFO [train.py:968] (1/2) Epoch 24, batch 12900, giga_loss[loss=0.2778, simple_loss=0.352, pruned_loss=0.1017, over 28857.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3584, pruned_loss=0.1097, over 5682887.29 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3446, pruned_loss=0.09007, over 5741278.86 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3611, pruned_loss=0.1128, over 5666044.68 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:33:31,507 INFO [optim.py:369] (1/2) 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 09:33:40,081 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1061810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:33:42,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1061813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:33:44,493 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5507, 4.3794, 4.1621, 2.1023], device='cuda:1'), covar=tensor([0.0568, 0.0728, 0.0866, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1174, 0.0996, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 09:33:51,618 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1061822.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:34:09,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1061842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:34:16,837 INFO [train.py:968] (1/2) Epoch 24, batch 12950, giga_loss[loss=0.2884, simple_loss=0.3478, pruned_loss=0.1146, over 26889.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3553, pruned_loss=0.1067, over 5671900.13 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3438, pruned_loss=0.08981, over 5736517.78 frames. ], giga_tot_loss[loss=0.2896, simple_loss=0.3588, pruned_loss=0.1102, over 5659714.38 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:34:31,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1061862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:35:05,477 INFO [train.py:968] (1/2) Epoch 24, batch 13000, giga_loss[loss=0.2619, simple_loss=0.3392, pruned_loss=0.09233, over 28710.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3514, pruned_loss=0.1031, over 5674502.13 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.08947, over 5741051.37 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3555, pruned_loss=0.1067, over 5658143.07 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:35:07,467 INFO [optim.py:369] (1/2) 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 09:35:22,000 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 13050, giga_loss[loss=0.2928, simple_loss=0.3575, pruned_loss=0.1141, over 26685.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3491, pruned_loss=0.1002, over 5676904.85 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08937, over 5741708.40 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3525, pruned_loss=0.1034, over 5662498.32 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:35:58,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 09:36:49,898 INFO [train.py:968] (1/2) Epoch 24, batch 13100, giga_loss[loss=0.28, simple_loss=0.3556, pruned_loss=0.1022, over 28508.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3486, pruned_loss=0.09808, over 5672749.02 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.08954, over 5741847.04 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3515, pruned_loss=0.1006, over 5659512.35 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:36:53,281 INFO [optim.py:369] (1/2) 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 09:37:00,186 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:37:02,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:37:20,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-12 09:37:35,380 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:37:44,628 INFO [train.py:968] (1/2) Epoch 24, batch 13150, giga_loss[loss=0.2601, simple_loss=0.3403, pruned_loss=0.08995, over 28616.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.35, pruned_loss=0.0994, over 5656827.98 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3427, pruned_loss=0.08975, over 5733717.86 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1013, over 5653513.33 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:38:31,593 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-12 09:38:33,550 INFO [train.py:968] (1/2) Epoch 24, batch 13200, giga_loss[loss=0.2523, simple_loss=0.3373, pruned_loss=0.08363, over 28259.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3476, pruned_loss=0.09774, over 5655266.08 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3427, pruned_loss=0.08993, over 5729619.58 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3495, pruned_loss=0.09937, over 5653765.13 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:38:36,700 INFO [optim.py:369] (1/2) 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 09:38:53,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:38:58,521 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062124.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:39:05,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2015, 1.6938, 1.7286, 1.3801], device='cuda:1'), covar=tensor([0.2186, 0.1765, 0.1957, 0.2010], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0739, 0.0709, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 09:39:06,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-12 09:39:17,190 INFO [train.py:968] (1/2) Epoch 24, batch 13250, giga_loss[loss=0.2261, simple_loss=0.3171, pruned_loss=0.06751, over 28992.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3441, pruned_loss=0.09563, over 5656485.79 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3416, pruned_loss=0.08967, over 5720780.58 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.347, pruned_loss=0.0976, over 5659344.56 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:40:00,514 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 09:40:08,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062197.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:40:08,440 INFO [train.py:968] (1/2) Epoch 24, batch 13300, giga_loss[loss=0.254, simple_loss=0.3414, pruned_loss=0.08333, over 28989.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.343, pruned_loss=0.0948, over 5661741.13 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3415, pruned_loss=0.0897, over 5724543.13 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3455, pruned_loss=0.09647, over 5659318.57 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:40:13,331 INFO [optim.py:369] (1/2) 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 09:40:59,458 INFO [train.py:968] (1/2) Epoch 24, batch 13350, giga_loss[loss=0.246, simple_loss=0.329, pruned_loss=0.0815, over 28935.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3424, pruned_loss=0.09382, over 5666464.74 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3408, pruned_loss=0.08937, over 5728220.37 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.345, pruned_loss=0.09556, over 5659966.09 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:41:08,429 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 09:41:11,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:14,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:20,253 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 09:41:40,364 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:42,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 09:41:42,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:41:49,890 INFO [train.py:968] (1/2) Epoch 24, batch 13400, giga_loss[loss=0.2393, simple_loss=0.3293, pruned_loss=0.07469, over 28945.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3408, pruned_loss=0.09247, over 5670825.69 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3406, pruned_loss=0.08927, over 5732710.10 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3431, pruned_loss=0.09404, over 5660515.90 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:41:53,244 INFO [optim.py:369] (1/2) 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 09:42:06,822 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062312.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:42:22,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8272, 2.3163, 2.1847, 1.8704], device='cuda:1'), covar=tensor([0.2160, 0.2101, 0.1865, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0739, 0.0708, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 09:42:34,580 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062340.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:42:35,180 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:42:36,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062343.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:42:41,318 INFO [train.py:968] (1/2) Epoch 24, batch 13450, giga_loss[loss=0.2876, simple_loss=0.3587, pruned_loss=0.1082, over 28019.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3389, pruned_loss=0.09093, over 5671638.87 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3407, pruned_loss=0.0894, over 5734677.93 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3406, pruned_loss=0.0921, over 5660559.02 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:42:57,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 09:43:11,545 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:43:12,884 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 13500, libri_loss[loss=0.2545, simple_loss=0.3334, pruned_loss=0.08782, over 27923.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3342, pruned_loss=0.08851, over 5661948.67 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3401, pruned_loss=0.08918, over 5735315.23 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3361, pruned_loss=0.08963, over 5651644.04 frames. ], batch size: 116, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:43:45,485 INFO [optim.py:369] (1/2) 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 09:44:22,462 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:44:26,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:44:37,302 INFO [train.py:968] (1/2) Epoch 24, batch 13550, giga_loss[loss=0.2348, simple_loss=0.3158, pruned_loss=0.07685, over 28646.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3332, pruned_loss=0.08878, over 5650712.02 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3401, pruned_loss=0.08931, over 5738656.29 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3346, pruned_loss=0.08954, over 5638186.16 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:44:40,127 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-12 09:44:54,010 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:45:07,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1718, 1.6556, 1.5059, 1.4339], device='cuda:1'), covar=tensor([0.2209, 0.1795, 0.1867, 0.1882], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0737, 0.0705, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 09:45:23,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-12 09:45:30,393 INFO [train.py:968] (1/2) Epoch 24, batch 13600, giga_loss[loss=0.2367, simple_loss=0.321, pruned_loss=0.07618, over 28393.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.334, pruned_loss=0.08988, over 5650035.06 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3399, pruned_loss=0.08932, over 5738976.60 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3351, pruned_loss=0.09047, over 5638092.92 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:45:31,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062499.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:45:34,597 INFO [optim.py:369] (1/2) 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 09:46:17,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9593, 3.7662, 3.5814, 1.8381], device='cuda:1'), covar=tensor([0.0705, 0.0892, 0.0923, 0.2319], device='cuda:1'), in_proj_covar=tensor([0.1248, 0.1156, 0.0975, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 09:46:28,263 INFO [train.py:968] (1/2) Epoch 24, batch 13650, giga_loss[loss=0.2697, simple_loss=0.3548, pruned_loss=0.09237, over 28337.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3357, pruned_loss=0.09028, over 5644475.08 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3393, pruned_loss=0.08913, over 5734959.02 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.337, pruned_loss=0.09096, over 5635767.61 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:47:28,658 INFO [train.py:968] (1/2) Epoch 24, batch 13700, libri_loss[loss=0.276, simple_loss=0.3506, pruned_loss=0.1007, over 29484.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3383, pruned_loss=0.09083, over 5639097.83 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3398, pruned_loss=0.08962, over 5726965.80 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3389, pruned_loss=0.09096, over 5636326.95 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:47:36,412 INFO [optim.py:369] (1/2) 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 09:48:21,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062642.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:48:24,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062645.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:48:27,215 INFO [train.py:968] (1/2) Epoch 24, batch 13750, giga_loss[loss=0.2539, simple_loss=0.3385, pruned_loss=0.0846, over 28917.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3389, pruned_loss=0.09139, over 5634486.87 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3392, pruned_loss=0.08946, over 5721641.18 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3399, pruned_loss=0.09169, over 5634572.31 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:49:05,695 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062674.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:49:22,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062687.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:49:35,858 INFO [train.py:968] (1/2) Epoch 24, batch 13800, giga_loss[loss=0.2793, simple_loss=0.3408, pruned_loss=0.1089, over 26963.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3373, pruned_loss=0.09068, over 5640168.69 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3389, pruned_loss=0.08939, over 5722384.97 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3382, pruned_loss=0.09098, over 5639201.79 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:49:40,490 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1062705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:49:56,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:50:35,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1785, 1.4843, 1.4753, 1.0776], device='cuda:1'), covar=tensor([0.1825, 0.2909, 0.1529, 0.1851], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0700, 0.0957, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 09:50:39,271 INFO [train.py:968] (1/2) Epoch 24, batch 13850, giga_loss[loss=0.2694, simple_loss=0.3482, pruned_loss=0.09523, over 27648.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3348, pruned_loss=0.08844, over 5635318.82 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3381, pruned_loss=0.08907, over 5716933.34 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3363, pruned_loss=0.08898, over 5637018.06 frames. ], batch size: 474, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:50:41,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1062749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:51:41,181 INFO [train.py:968] (1/2) Epoch 24, batch 13900, giga_loss[loss=0.2437, simple_loss=0.3227, pruned_loss=0.08241, over 28914.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3325, pruned_loss=0.08649, over 5636774.67 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3373, pruned_loss=0.08866, over 5718131.63 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3344, pruned_loss=0.08728, over 5635583.45 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:51:47,395 INFO [optim.py:369] (1/2) 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 09:52:23,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 09:52:23,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 09:52:28,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062833.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:52:48,633 INFO [train.py:968] (1/2) Epoch 24, batch 13950, giga_loss[loss=0.2166, simple_loss=0.2983, pruned_loss=0.06747, over 28874.00 frames. ], tot_loss[loss=0.251, simple_loss=0.33, pruned_loss=0.08604, over 5647165.81 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3371, pruned_loss=0.08854, over 5720055.50 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3316, pruned_loss=0.08674, over 5644042.15 frames. ], batch size: 120, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:53:00,707 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:05,252 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062862.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 09:53:05,276 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1062862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:39,993 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1062891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:41,673 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1062892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:53:45,691 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 14000, giga_loss[loss=0.2394, simple_loss=0.3214, pruned_loss=0.07867, over 28683.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3308, pruned_loss=0.08706, over 5647727.12 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3376, pruned_loss=0.08898, over 5718896.80 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3314, pruned_loss=0.08719, over 5644960.96 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:53:54,580 INFO [optim.py:369] (1/2) 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 09:54:23,434 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 14050, giga_loss[loss=0.3277, simple_loss=0.3864, pruned_loss=0.1345, over 28160.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3316, pruned_loss=0.08694, over 5661377.20 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3377, pruned_loss=0.08912, over 5722207.94 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3318, pruned_loss=0.08685, over 5654962.98 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:55:49,902 INFO [train.py:968] (1/2) Epoch 24, batch 14100, libri_loss[loss=0.2575, simple_loss=0.3382, pruned_loss=0.0884, over 29663.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3337, pruned_loss=0.08698, over 5674565.04 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3374, pruned_loss=0.08902, over 5726821.45 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3341, pruned_loss=0.08695, over 5663996.01 frames. ], batch size: 88, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 09:55:55,552 INFO [optim.py:369] (1/2) 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 09:56:35,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4254, 1.8599, 1.0217, 1.3571], device='cuda:1'), covar=tensor([0.1307, 0.0687, 0.1563, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0397, 0.0443, 0.0514, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 09:56:50,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6362, 1.4726, 4.3708, 3.4406], device='cuda:1'), covar=tensor([0.1520, 0.2721, 0.0391, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0660, 0.0971, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 09:56:54,013 INFO [train.py:968] (1/2) Epoch 24, batch 14150, giga_loss[loss=0.2145, simple_loss=0.2914, pruned_loss=0.06879, over 27581.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3323, pruned_loss=0.08598, over 5676827.69 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3368, pruned_loss=0.08881, over 5731295.44 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3331, pruned_loss=0.08609, over 5662985.61 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:57:35,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1063080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 09:58:03,657 INFO [train.py:968] (1/2) Epoch 24, batch 14200, giga_loss[loss=0.2614, simple_loss=0.3375, pruned_loss=0.09266, over 28842.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3305, pruned_loss=0.08512, over 5687629.13 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3364, pruned_loss=0.08867, over 5735644.64 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3314, pruned_loss=0.08524, over 5671776.81 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 09:58:11,628 INFO [optim.py:369] (1/2) 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 09:59:14,660 INFO [train.py:968] (1/2) Epoch 24, batch 14250, giga_loss[loss=0.2904, simple_loss=0.3799, pruned_loss=0.1004, over 28480.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3339, pruned_loss=0.08727, over 5669943.84 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3363, pruned_loss=0.08866, over 5735545.73 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3345, pruned_loss=0.08733, over 5656234.96 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:00:15,481 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1063192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:00:23,565 INFO [train.py:968] (1/2) Epoch 24, batch 14300, giga_loss[loss=0.2483, simple_loss=0.3415, pruned_loss=0.07759, over 27594.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3369, pruned_loss=0.0865, over 5661941.33 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3361, pruned_loss=0.08847, over 5738520.29 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3376, pruned_loss=0.0867, over 5646815.73 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:00:31,734 INFO [optim.py:369] (1/2) 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 10:00:40,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1988, 0.9011, 1.0005, 1.3189], device='cuda:1'), covar=tensor([0.0812, 0.0415, 0.0375, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 10:00:57,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1063223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:01:00,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1063226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:01:29,608 INFO [train.py:968] (1/2) Epoch 24, batch 14350, libri_loss[loss=0.2688, simple_loss=0.3491, pruned_loss=0.0942, over 29301.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3376, pruned_loss=0.0856, over 5652032.24 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3361, pruned_loss=0.08851, over 5740824.70 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3382, pruned_loss=0.08566, over 5636556.59 frames. ], batch size: 94, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:01:39,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8021, 3.6352, 3.4731, 1.5899], device='cuda:1'), covar=tensor([0.0683, 0.0845, 0.0845, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.1234, 0.1139, 0.0962, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 10:01:39,819 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1063255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:02:29,288 INFO [train.py:968] (1/2) Epoch 24, batch 14400, giga_loss[loss=0.2583, simple_loss=0.3443, pruned_loss=0.08618, over 28718.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.338, pruned_loss=0.0846, over 5663255.87 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3357, pruned_loss=0.08834, over 5744115.75 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3388, pruned_loss=0.08473, over 5646762.90 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:02:35,297 INFO [optim.py:369] (1/2) 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 10:03:36,752 INFO [train.py:968] (1/2) Epoch 24, batch 14450, giga_loss[loss=0.2943, simple_loss=0.3613, pruned_loss=0.1137, over 27760.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3395, pruned_loss=0.08637, over 5670636.19 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3357, pruned_loss=0.08833, over 5747258.15 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3403, pruned_loss=0.08643, over 5653264.93 frames. ], batch size: 474, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:04:38,543 INFO [train.py:968] (1/2) Epoch 24, batch 14500, giga_loss[loss=0.2412, simple_loss=0.3218, pruned_loss=0.08026, over 28874.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3382, pruned_loss=0.08679, over 5662988.24 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3359, pruned_loss=0.08844, over 5736073.99 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3387, pruned_loss=0.08671, over 5656661.59 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:04:48,399 INFO [optim.py:369] (1/2) 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 10:05:59,939 INFO [train.py:968] (1/2) Epoch 24, batch 14550, giga_loss[loss=0.2575, simple_loss=0.3378, pruned_loss=0.08863, over 28521.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3394, pruned_loss=0.08857, over 5662079.13 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3355, pruned_loss=0.0883, over 5738244.62 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3402, pruned_loss=0.08861, over 5654177.26 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:06:53,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4034, 1.7069, 1.6214, 1.3748], device='cuda:1'), covar=tensor([0.3276, 0.2270, 0.1630, 0.2441], device='cuda:1'), in_proj_covar=tensor([0.1978, 0.1915, 0.1831, 0.1976], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:06:59,706 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1063485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:07:16,457 INFO [train.py:968] (1/2) Epoch 24, batch 14600, libri_loss[loss=0.198, simple_loss=0.2765, pruned_loss=0.05978, over 29361.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3347, pruned_loss=0.08577, over 5674486.48 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.335, pruned_loss=0.08815, over 5740397.25 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3358, pruned_loss=0.08591, over 5664336.61 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:07:31,544 INFO [optim.py:369] (1/2) 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 10:08:29,880 INFO [train.py:968] (1/2) Epoch 24, batch 14650, libri_loss[loss=0.2323, simple_loss=0.316, pruned_loss=0.07425, over 29538.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3304, pruned_loss=0.08333, over 5674568.58 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.334, pruned_loss=0.08748, over 5745176.61 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3322, pruned_loss=0.08393, over 5659944.83 frames. ], batch size: 82, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:08:48,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-12 10:08:57,932 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 14700, giga_loss[loss=0.2364, simple_loss=0.3195, pruned_loss=0.07669, over 28689.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3269, pruned_loss=0.08179, over 5676566.36 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3334, pruned_loss=0.08719, over 5748172.01 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3287, pruned_loss=0.08243, over 5660878.05 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:09:47,552 INFO [optim.py:369] (1/2) 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 10:09:56,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2652, 1.7619, 1.4261, 1.3807], device='cuda:1'), covar=tensor([0.0788, 0.0302, 0.0311, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 10:10:39,119 INFO [train.py:968] (1/2) Epoch 24, batch 14750, giga_loss[loss=0.323, simple_loss=0.3991, pruned_loss=0.1235, over 28990.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3312, pruned_loss=0.08449, over 5678089.94 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.333, pruned_loss=0.08704, over 5740557.03 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3329, pruned_loss=0.08505, over 5670494.09 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:11:36,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6204, 1.7592, 1.4686, 1.6664], device='cuda:1'), covar=tensor([0.2714, 0.2794, 0.3194, 0.2418], device='cuda:1'), in_proj_covar=tensor([0.1548, 0.1113, 0.1366, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 10:11:43,260 INFO [train.py:968] (1/2) Epoch 24, batch 14800, giga_loss[loss=0.2483, simple_loss=0.3231, pruned_loss=0.08678, over 28918.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3332, pruned_loss=0.08572, over 5677257.95 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.333, pruned_loss=0.08718, over 5740414.66 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3345, pruned_loss=0.08601, over 5670286.14 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:11:53,189 INFO [optim.py:369] (1/2) 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 10:11:59,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1063710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:12:05,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1063713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:12:41,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1063742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:12:46,272 INFO [train.py:968] (1/2) Epoch 24, batch 14850, giga_loss[loss=0.2363, simple_loss=0.3133, pruned_loss=0.0797, over 28653.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3314, pruned_loss=0.08596, over 5683314.57 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3326, pruned_loss=0.0871, over 5744564.47 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3329, pruned_loss=0.08624, over 5672126.47 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:13:36,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5820, 2.0227, 1.9601, 1.5891], device='cuda:1'), covar=tensor([0.2867, 0.2069, 0.2247, 0.2446], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1903, 0.1821, 0.1964], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:13:39,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3692, 1.7661, 1.7425, 1.4377], device='cuda:1'), covar=tensor([0.2862, 0.2009, 0.2171, 0.2405], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1903, 0.1821, 0.1964], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:13:53,535 INFO [train.py:968] (1/2) Epoch 24, batch 14900, giga_loss[loss=0.2169, simple_loss=0.3001, pruned_loss=0.06687, over 28911.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.333, pruned_loss=0.08761, over 5680893.77 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08687, over 5747832.23 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3345, pruned_loss=0.08805, over 5667931.23 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:14:02,682 INFO [optim.py:369] (1/2) 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 10:14:51,812 INFO [train.py:968] (1/2) Epoch 24, batch 14950, libri_loss[loss=0.2193, simple_loss=0.3028, pruned_loss=0.06785, over 29661.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.333, pruned_loss=0.08761, over 5674570.09 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3316, pruned_loss=0.08665, over 5744908.35 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3348, pruned_loss=0.08821, over 5663198.46 frames. ], batch size: 91, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:15:08,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1063860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:15:13,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2846, 3.1135, 2.9607, 1.4386], device='cuda:1'), covar=tensor([0.0955, 0.1046, 0.0980, 0.2277], device='cuda:1'), in_proj_covar=tensor([0.1228, 0.1133, 0.0957, 0.0718], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 10:15:59,834 INFO [train.py:968] (1/2) Epoch 24, batch 15000, giga_loss[loss=0.3205, simple_loss=0.382, pruned_loss=0.1295, over 28831.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3361, pruned_loss=0.08828, over 5676338.94 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08696, over 5748313.59 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3373, pruned_loss=0.08852, over 5662452.54 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:15:59,835 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 10:16:08,944 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 10:16:17,706 INFO [optim.py:369] (1/2) 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 10:17:23,721 INFO [train.py:968] (1/2) Epoch 24, batch 15050, giga_loss[loss=0.2449, simple_loss=0.3283, pruned_loss=0.08073, over 28964.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3368, pruned_loss=0.0884, over 5673319.22 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3323, pruned_loss=0.08711, over 5747686.26 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3376, pruned_loss=0.08849, over 5662077.15 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:18:37,280 INFO [train.py:968] (1/2) Epoch 24, batch 15100, giga_loss[loss=0.2167, simple_loss=0.3, pruned_loss=0.06664, over 28770.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3322, pruned_loss=0.08645, over 5692256.37 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3316, pruned_loss=0.08693, over 5751283.77 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3336, pruned_loss=0.08668, over 5678516.41 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:18:38,426 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1063998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:18:44,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:18:46,316 INFO [optim.py:369] (1/2) 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 10:18:46,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7019, 1.9686, 1.3738, 1.5058], device='cuda:1'), covar=tensor([0.0993, 0.0546, 0.0996, 0.1051], device='cuda:1'), in_proj_covar=tensor([0.0394, 0.0440, 0.0511, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 10:18:46,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:19:33,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:19:37,044 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1064039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:19:46,372 INFO [train.py:968] (1/2) Epoch 24, batch 15150, giga_loss[loss=0.2295, simple_loss=0.3132, pruned_loss=0.07294, over 28409.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3266, pruned_loss=0.08431, over 5690588.50 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3311, pruned_loss=0.08665, over 5751511.42 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3281, pruned_loss=0.08472, over 5677905.99 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:20:38,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3005, 1.6312, 1.6999, 1.4458], device='cuda:1'), covar=tensor([0.2160, 0.2045, 0.2152, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0735, 0.0706, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 10:20:50,802 INFO [train.py:968] (1/2) Epoch 24, batch 15200, giga_loss[loss=0.2289, simple_loss=0.3142, pruned_loss=0.0718, over 28930.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3262, pruned_loss=0.08429, over 5676562.66 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3309, pruned_loss=0.08663, over 5744481.08 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3274, pruned_loss=0.08458, over 5671575.89 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:20:59,809 INFO [optim.py:369] (1/2) 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 10:21:46,579 INFO [train.py:968] (1/2) Epoch 24, batch 15250, libri_loss[loss=0.267, simple_loss=0.348, pruned_loss=0.09298, over 29274.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3289, pruned_loss=0.08636, over 5679686.56 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3311, pruned_loss=0.08668, over 5748573.73 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3296, pruned_loss=0.08653, over 5670117.01 frames. ], batch size: 94, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:22:44,791 INFO [train.py:968] (1/2) Epoch 24, batch 15300, giga_loss[loss=0.2335, simple_loss=0.316, pruned_loss=0.07552, over 28696.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3264, pruned_loss=0.08475, over 5664430.01 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3307, pruned_loss=0.08641, over 5750361.86 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3273, pruned_loss=0.0851, over 5652684.53 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:22:53,212 INFO [optim.py:369] (1/2) 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 10:23:34,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1774, 1.3818, 1.2635, 1.1579], device='cuda:1'), covar=tensor([0.2213, 0.1894, 0.1430, 0.1880], device='cuda:1'), in_proj_covar=tensor([0.1962, 0.1890, 0.1811, 0.1954], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:23:38,009 INFO [train.py:968] (1/2) Epoch 24, batch 15350, giga_loss[loss=0.2748, simple_loss=0.3403, pruned_loss=0.1047, over 26818.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3249, pruned_loss=0.083, over 5661750.58 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3304, pruned_loss=0.08629, over 5736828.02 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3255, pruned_loss=0.08325, over 5659626.00 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:24:45,488 INFO [train.py:968] (1/2) Epoch 24, batch 15400, giga_loss[loss=0.241, simple_loss=0.322, pruned_loss=0.08003, over 28631.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3239, pruned_loss=0.08242, over 5656873.74 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3302, pruned_loss=0.08618, over 5739384.75 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3245, pruned_loss=0.08265, over 5651830.22 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:24:59,186 INFO [optim.py:369] (1/2) 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 10:25:22,725 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 15450, giga_loss[loss=0.2566, simple_loss=0.3356, pruned_loss=0.08879, over 28708.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.08253, over 5671946.93 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3303, pruned_loss=0.08626, over 5741292.64 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3245, pruned_loss=0.0826, over 5665619.09 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:26:34,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1064373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:26:59,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9704, 1.2056, 0.9387, 0.7654], device='cuda:1'), covar=tensor([0.1201, 0.0579, 0.1554, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0395, 0.0440, 0.0513, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 10:27:01,018 INFO [train.py:968] (1/2) Epoch 24, batch 15500, giga_loss[loss=0.2726, simple_loss=0.3419, pruned_loss=0.1017, over 26838.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3255, pruned_loss=0.08274, over 5685400.64 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3305, pruned_loss=0.08639, over 5743924.56 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3254, pruned_loss=0.08257, over 5676316.30 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:27:12,734 INFO [optim.py:369] (1/2) 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 10:27:25,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1064414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:27:36,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7941, 1.9741, 1.4192, 1.5349], device='cuda:1'), covar=tensor([0.0971, 0.0636, 0.0971, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0441, 0.0515, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 10:27:40,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6765, 1.8780, 1.3905, 1.3643], device='cuda:1'), covar=tensor([0.0997, 0.0579, 0.0981, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0396, 0.0441, 0.0515, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 10:28:08,399 INFO [train.py:968] (1/2) Epoch 24, batch 15550, giga_loss[loss=0.2619, simple_loss=0.3329, pruned_loss=0.09544, over 28928.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3265, pruned_loss=0.08383, over 5683213.38 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3302, pruned_loss=0.08638, over 5741886.91 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3266, pruned_loss=0.08365, over 5676983.33 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:29:14,533 INFO [train.py:968] (1/2) Epoch 24, batch 15600, giga_loss[loss=0.2324, simple_loss=0.3155, pruned_loss=0.07469, over 27605.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3252, pruned_loss=0.08349, over 5685568.60 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3302, pruned_loss=0.0865, over 5744995.98 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3252, pruned_loss=0.08321, over 5676872.08 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:29:27,691 INFO [optim.py:369] (1/2) 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 10:29:39,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:29:42,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:29:53,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3486, 1.4638, 1.3539, 1.3507], device='cuda:1'), covar=tensor([0.2202, 0.1921, 0.1916, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.1967, 0.1890, 0.1811, 0.1956], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:30:18,020 INFO [train.py:968] (1/2) Epoch 24, batch 15650, giga_loss[loss=0.2465, simple_loss=0.3375, pruned_loss=0.0778, over 28918.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3247, pruned_loss=0.08172, over 5666728.82 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3304, pruned_loss=0.08653, over 5742831.47 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3245, pruned_loss=0.08143, over 5661390.15 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:30:18,339 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:30:31,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:30:34,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:31:10,407 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:31:23,510 INFO [train.py:968] (1/2) Epoch 24, batch 15700, giga_loss[loss=0.301, simple_loss=0.3596, pruned_loss=0.1212, over 26864.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3288, pruned_loss=0.0837, over 5663869.70 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3303, pruned_loss=0.08646, over 5744469.98 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3288, pruned_loss=0.0835, over 5657641.25 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:31:37,728 INFO [optim.py:369] (1/2) 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 10:32:27,197 INFO [train.py:968] (1/2) Epoch 24, batch 15750, giga_loss[loss=0.2592, simple_loss=0.3317, pruned_loss=0.09332, over 26914.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3304, pruned_loss=0.08393, over 5664346.29 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3302, pruned_loss=0.08636, over 5746893.48 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3304, pruned_loss=0.08384, over 5655925.05 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:32:39,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1064657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:33:06,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1284, 1.4566, 1.3835, 1.0337], device='cuda:1'), covar=tensor([0.1543, 0.2369, 0.1295, 0.1691], device='cuda:1'), in_proj_covar=tensor([0.0910, 0.0697, 0.0957, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 10:33:26,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4733, 1.9005, 1.7677, 1.6601], device='cuda:1'), covar=tensor([0.2028, 0.2050, 0.1811, 0.1956], device='cuda:1'), in_proj_covar=tensor([0.0469, 0.0733, 0.0703, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 10:33:27,437 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 15800, giga_loss[loss=0.2762, simple_loss=0.3541, pruned_loss=0.0991, over 28320.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3304, pruned_loss=0.08454, over 5647915.16 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.33, pruned_loss=0.08636, over 5738145.63 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3306, pruned_loss=0.08445, over 5647293.83 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:33:42,057 INFO [optim.py:369] (1/2) 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 10:34:27,233 INFO [train.py:968] (1/2) Epoch 24, batch 15850, giga_loss[loss=0.2135, simple_loss=0.2999, pruned_loss=0.06354, over 28487.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3278, pruned_loss=0.08324, over 5657718.27 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3292, pruned_loss=0.08604, over 5741700.27 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3288, pruned_loss=0.08336, over 5650335.56 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:35:29,294 INFO [train.py:968] (1/2) Epoch 24, batch 15900, giga_loss[loss=0.2459, simple_loss=0.3314, pruned_loss=0.08025, over 28904.00 frames. ], tot_loss[loss=0.246, simple_loss=0.3267, pruned_loss=0.08264, over 5651177.40 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3294, pruned_loss=0.08611, over 5734389.13 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3272, pruned_loss=0.08255, over 5648144.87 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:35:42,893 INFO [optim.py:369] (1/2) 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 10:36:21,665 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1064839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:36:27,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1064842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:36:27,371 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-12 10:36:31,641 INFO [train.py:968] (1/2) Epoch 24, batch 15950, giga_loss[loss=0.2601, simple_loss=0.3404, pruned_loss=0.08984, over 28732.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3255, pruned_loss=0.08275, over 5663415.23 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3291, pruned_loss=0.08593, over 5736514.83 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3261, pruned_loss=0.0828, over 5658043.26 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:36:58,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6435, 1.9994, 1.3286, 1.4828], device='cuda:1'), covar=tensor([0.1017, 0.0540, 0.1000, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0443, 0.0517, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 10:37:03,093 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1064871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:37:32,480 INFO [train.py:968] (1/2) Epoch 24, batch 16000, giga_loss[loss=0.3199, simple_loss=0.3737, pruned_loss=0.133, over 26854.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3258, pruned_loss=0.08288, over 5670860.77 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3284, pruned_loss=0.08559, over 5741466.70 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3269, pruned_loss=0.08316, over 5659946.98 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:37:44,411 INFO [optim.py:369] (1/2) 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 10:38:33,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-12 10:38:33,965 INFO [train.py:968] (1/2) Epoch 24, batch 16050, giga_loss[loss=0.267, simple_loss=0.3472, pruned_loss=0.09336, over 28704.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3273, pruned_loss=0.08315, over 5674815.64 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.328, pruned_loss=0.08542, over 5743076.81 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3285, pruned_loss=0.08347, over 5662641.80 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:39:43,817 INFO [train.py:968] (1/2) Epoch 24, batch 16100, giga_loss[loss=0.2721, simple_loss=0.3447, pruned_loss=0.0998, over 27597.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3286, pruned_loss=0.08477, over 5664122.10 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.328, pruned_loss=0.08533, over 5745540.39 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3295, pruned_loss=0.0851, over 5651107.42 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:39:51,199 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1539, 2.5792, 1.5829, 2.1626], device='cuda:1'), covar=tensor([0.0987, 0.0566, 0.1037, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0444, 0.0518, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 10:39:57,373 INFO [optim.py:369] (1/2) 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 10:39:58,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2164, 1.4364, 1.3065, 1.1991], device='cuda:1'), covar=tensor([0.2296, 0.1805, 0.1622, 0.1889], device='cuda:1'), in_proj_covar=tensor([0.1958, 0.1885, 0.1802, 0.1951], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:40:27,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:40:35,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 10:40:47,728 INFO [train.py:968] (1/2) Epoch 24, batch 16150, giga_loss[loss=0.2521, simple_loss=0.3463, pruned_loss=0.07898, over 28713.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3313, pruned_loss=0.08587, over 5662821.75 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3279, pruned_loss=0.08528, over 5747114.18 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3321, pruned_loss=0.08618, over 5650456.69 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:40:49,901 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3690, 1.2712, 4.2013, 3.3991], device='cuda:1'), covar=tensor([0.1731, 0.3021, 0.0422, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0656, 0.0965, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 10:41:45,268 INFO [train.py:968] (1/2) Epoch 24, batch 16200, giga_loss[loss=0.2363, simple_loss=0.337, pruned_loss=0.06775, over 28867.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3334, pruned_loss=0.0864, over 5663590.33 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3274, pruned_loss=0.08504, over 5751014.03 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3348, pruned_loss=0.08691, over 5647715.42 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:41:57,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:41:58,036 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:42:54,350 INFO [train.py:968] (1/2) Epoch 24, batch 16250, giga_loss[loss=0.2696, simple_loss=0.359, pruned_loss=0.09009, over 29059.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.335, pruned_loss=0.08705, over 5659592.44 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3273, pruned_loss=0.085, over 5752157.05 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3361, pruned_loss=0.0875, over 5645472.93 frames. ], batch size: 200, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:43:36,439 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:43:40,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:44:04,046 INFO [train.py:968] (1/2) Epoch 24, batch 16300, giga_loss[loss=0.2291, simple_loss=0.3097, pruned_loss=0.07426, over 28714.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3336, pruned_loss=0.08676, over 5663829.36 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.327, pruned_loss=0.08486, over 5754297.00 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.335, pruned_loss=0.0873, over 5648400.02 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:44:18,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:44:19,321 INFO [optim.py:369] (1/2) 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 10:44:37,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6072, 1.6677, 1.8461, 1.3748], device='cuda:1'), covar=tensor([0.2022, 0.2709, 0.1595, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0700, 0.0960, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 10:45:05,761 INFO [train.py:968] (1/2) Epoch 24, batch 16350, giga_loss[loss=0.2316, simple_loss=0.3259, pruned_loss=0.06862, over 28978.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3309, pruned_loss=0.0851, over 5672170.06 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3269, pruned_loss=0.08472, over 5757176.65 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3323, pruned_loss=0.08568, over 5654326.71 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:46:10,732 INFO [train.py:968] (1/2) Epoch 24, batch 16400, giga_loss[loss=0.2998, simple_loss=0.3591, pruned_loss=0.1202, over 28170.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3319, pruned_loss=0.0862, over 5675170.94 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3268, pruned_loss=0.0847, over 5758825.51 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3331, pruned_loss=0.08668, over 5659019.24 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:46:27,711 INFO [optim.py:369] (1/2) 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 10:46:29,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:46:29,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 10:46:57,540 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:47:14,580 INFO [train.py:968] (1/2) Epoch 24, batch 16450, giga_loss[loss=0.2399, simple_loss=0.3131, pruned_loss=0.08335, over 28722.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3309, pruned_loss=0.08696, over 5655677.09 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3266, pruned_loss=0.08461, over 5752135.11 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3322, pruned_loss=0.08751, over 5645916.60 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:47:34,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-12 10:48:14,101 INFO [train.py:968] (1/2) Epoch 24, batch 16500, giga_loss[loss=0.2117, simple_loss=0.2899, pruned_loss=0.06675, over 28466.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3288, pruned_loss=0.08589, over 5658236.78 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3265, pruned_loss=0.08472, over 5751315.64 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.33, pruned_loss=0.08628, over 5649139.50 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:48:20,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 10:48:32,966 INFO [optim.py:369] (1/2) 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 10:48:45,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6710, 1.7921, 1.5124, 1.7763], device='cuda:1'), covar=tensor([0.2855, 0.2769, 0.3108, 0.2545], device='cuda:1'), in_proj_covar=tensor([0.1545, 0.1109, 0.1364, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 10:49:13,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3906, 1.6552, 1.6752, 1.4367], device='cuda:1'), covar=tensor([0.2568, 0.2341, 0.1629, 0.2013], device='cuda:1'), in_proj_covar=tensor([0.1971, 0.1898, 0.1814, 0.1961], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:49:19,873 INFO [train.py:968] (1/2) Epoch 24, batch 16550, giga_loss[loss=0.2161, simple_loss=0.3105, pruned_loss=0.06088, over 28955.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3287, pruned_loss=0.08471, over 5668118.30 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3267, pruned_loss=0.08484, over 5753084.48 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3295, pruned_loss=0.08492, over 5658606.75 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:50:03,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:50:13,780 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 16600, libri_loss[loss=0.2629, simple_loss=0.3485, pruned_loss=0.08868, over 29662.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.329, pruned_loss=0.08312, over 5675449.71 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3266, pruned_loss=0.08475, over 5756526.31 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3297, pruned_loss=0.08333, over 5663372.51 frames. ], batch size: 88, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:50:34,107 INFO [optim.py:369] (1/2) 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 10:51:19,585 INFO [train.py:968] (1/2) Epoch 24, batch 16650, giga_loss[loss=0.2292, simple_loss=0.3221, pruned_loss=0.06814, over 28977.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3307, pruned_loss=0.082, over 5685333.78 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3267, pruned_loss=0.0848, over 5756335.68 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3311, pruned_loss=0.0821, over 5675722.41 frames. ], batch size: 285, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:51:55,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2800, 1.7178, 1.2745, 0.6735], device='cuda:1'), covar=tensor([0.5564, 0.2874, 0.3739, 0.6263], device='cuda:1'), in_proj_covar=tensor([0.1781, 0.1676, 0.1615, 0.1450], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 10:52:19,636 INFO [train.py:968] (1/2) Epoch 24, batch 16700, giga_loss[loss=0.2303, simple_loss=0.3177, pruned_loss=0.07143, over 28046.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3308, pruned_loss=0.08196, over 5675116.28 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3268, pruned_loss=0.08475, over 5757903.24 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3312, pruned_loss=0.08205, over 5664510.28 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:52:34,833 INFO [optim.py:369] (1/2) 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 10:52:39,538 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:52:59,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:02,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:10,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:16,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:30,475 INFO [train.py:968] (1/2) Epoch 24, batch 16750, giga_loss[loss=0.2846, simple_loss=0.3636, pruned_loss=0.1028, over 29091.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3308, pruned_loss=0.0824, over 5667679.32 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3267, pruned_loss=0.08471, over 5759361.58 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3313, pruned_loss=0.08249, over 5657220.48 frames. ], batch size: 200, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:53:42,313 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4266, 1.5867, 1.1917, 1.2163], device='cuda:1'), covar=tensor([0.1011, 0.0507, 0.1048, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0442, 0.0515, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 10:53:45,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:53:57,855 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:54:22,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:54:41,999 INFO [train.py:968] (1/2) Epoch 24, batch 16800, giga_loss[loss=0.2392, simple_loss=0.3269, pruned_loss=0.0757, over 29046.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.33, pruned_loss=0.08221, over 5660931.14 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3263, pruned_loss=0.08447, over 5762266.90 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3309, pruned_loss=0.08246, over 5648002.71 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:54:59,560 INFO [optim.py:369] (1/2) 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 10:54:59,994 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:55:52,998 INFO [train.py:968] (1/2) Epoch 24, batch 16850, libri_loss[loss=0.2119, simple_loss=0.2909, pruned_loss=0.06646, over 29657.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3301, pruned_loss=0.08134, over 5670466.86 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3262, pruned_loss=0.08434, over 5766013.35 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.331, pruned_loss=0.0816, over 5653858.93 frames. ], batch size: 73, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 10:56:21,370 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2149, 1.4801, 1.4040, 1.1580], device='cuda:1'), covar=tensor([0.2596, 0.2340, 0.1644, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.1962, 0.1893, 0.1804, 0.1952], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:56:31,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065777.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 10:56:58,501 INFO [train.py:968] (1/2) Epoch 24, batch 16900, giga_loss[loss=0.2901, simple_loss=0.3696, pruned_loss=0.1053, over 28433.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3313, pruned_loss=0.08223, over 5663430.47 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3268, pruned_loss=0.08483, over 5762961.34 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3316, pruned_loss=0.08184, over 5647986.87 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:57:15,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7761, 2.2289, 2.0853, 1.6428], device='cuda:1'), covar=tensor([0.2842, 0.1989, 0.2256, 0.2552], device='cuda:1'), in_proj_covar=tensor([0.1961, 0.1891, 0.1801, 0.1950], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 10:57:15,804 INFO [optim.py:369] (1/2) 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 10:57:41,627 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:57:44,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:08,341 INFO [train.py:968] (1/2) Epoch 24, batch 16950, giga_loss[loss=0.2485, simple_loss=0.3417, pruned_loss=0.07764, over 28675.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3345, pruned_loss=0.08359, over 5667606.90 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3269, pruned_loss=0.08491, over 5765356.30 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3348, pruned_loss=0.08316, over 5650561.54 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:58:17,177 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:19,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:24,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 10:58:55,364 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 17000, giga_loss[loss=0.3168, simple_loss=0.3802, pruned_loss=0.1268, over 27646.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3343, pruned_loss=0.08347, over 5683012.92 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3265, pruned_loss=0.08461, over 5768968.64 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3352, pruned_loss=0.08334, over 5663845.44 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 10:59:29,672 INFO [optim.py:369] (1/2) 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 10:59:42,394 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1065920.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 10:59:46,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1065923.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:00:11,252 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1065938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:00:22,135 INFO [train.py:968] (1/2) Epoch 24, batch 17050, giga_loss[loss=0.2391, simple_loss=0.3205, pruned_loss=0.07888, over 28839.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3332, pruned_loss=0.08367, over 5682412.04 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3264, pruned_loss=0.08451, over 5770491.05 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.334, pruned_loss=0.08362, over 5664452.44 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:00:33,026 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1065952.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:01:19,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1065986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:01:34,738 INFO [train.py:968] (1/2) Epoch 24, batch 17100, giga_loss[loss=0.2137, simple_loss=0.3113, pruned_loss=0.05808, over 29020.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3327, pruned_loss=0.08337, over 5688793.57 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3267, pruned_loss=0.08465, over 5773933.27 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3332, pruned_loss=0.08318, over 5669340.95 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:01:57,414 INFO [optim.py:369] (1/2) 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 11:01:59,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1066014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:02:43,804 INFO [train.py:968] (1/2) Epoch 24, batch 17150, giga_loss[loss=0.236, simple_loss=0.3001, pruned_loss=0.08597, over 24356.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3314, pruned_loss=0.08261, over 5672897.52 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3273, pruned_loss=0.08509, over 5765969.31 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3313, pruned_loss=0.08196, over 5661626.74 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:03:42,528 INFO [train.py:968] (1/2) Epoch 24, batch 17200, giga_loss[loss=0.2463, simple_loss=0.3326, pruned_loss=0.08004, over 28651.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3305, pruned_loss=0.08217, over 5678043.48 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3267, pruned_loss=0.0848, over 5767323.65 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3311, pruned_loss=0.08185, over 5665382.32 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:03:57,529 INFO [optim.py:369] (1/2) 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 11:04:21,892 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:04:26,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:04:43,024 INFO [train.py:968] (1/2) Epoch 24, batch 17250, giga_loss[loss=0.2261, simple_loss=0.3187, pruned_loss=0.06677, over 29013.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3331, pruned_loss=0.08359, over 5671381.15 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3265, pruned_loss=0.08472, over 5761870.92 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3339, pruned_loss=0.08337, over 5663353.71 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:04:59,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:05:39,287 INFO [train.py:968] (1/2) Epoch 24, batch 17300, giga_loss[loss=0.2326, simple_loss=0.3141, pruned_loss=0.07554, over 29075.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3328, pruned_loss=0.08374, over 5678554.17 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3264, pruned_loss=0.08462, over 5765066.32 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3337, pruned_loss=0.08365, over 5667632.08 frames. ], batch size: 285, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:05:53,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.41 vs. limit=5.0
+2023-03-12 11:05:53,841 INFO [optim.py:369] (1/2) 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 11:06:35,811 INFO [train.py:968] (1/2) Epoch 24, batch 17350, giga_loss[loss=0.2471, simple_loss=0.3247, pruned_loss=0.08477, over 28881.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3296, pruned_loss=0.08316, over 5674613.49 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3265, pruned_loss=0.08467, over 5767818.66 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3302, pruned_loss=0.083, over 5661485.55 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:07:37,581 INFO [train.py:968] (1/2) Epoch 24, batch 17400, giga_loss[loss=0.2493, simple_loss=0.3281, pruned_loss=0.08527, over 28876.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3299, pruned_loss=0.08429, over 5654390.65 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3266, pruned_loss=0.08474, over 5759205.48 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3304, pruned_loss=0.0841, over 5651675.58 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:07:57,081 INFO [optim.py:369] (1/2) 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 11:07:58,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1066313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:08:34,647 INFO [train.py:968] (1/2) Epoch 24, batch 17450, giga_loss[loss=0.2987, simple_loss=0.3788, pruned_loss=0.1093, over 28936.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3347, pruned_loss=0.08712, over 5659311.74 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3263, pruned_loss=0.08452, over 5763846.25 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3355, pruned_loss=0.08718, over 5650064.53 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:08:58,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4114, 1.7687, 1.5693, 1.5289], device='cuda:1'), covar=tensor([0.0794, 0.0323, 0.0314, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 11:09:16,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1066389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:09:23,954 INFO [train.py:968] (1/2) Epoch 24, batch 17500, giga_loss[loss=0.2849, simple_loss=0.3726, pruned_loss=0.09858, over 29031.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3435, pruned_loss=0.09172, over 5659860.96 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3263, pruned_loss=0.0845, over 5756838.38 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3444, pruned_loss=0.09188, over 5657448.84 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:09:34,624 INFO [optim.py:369] (1/2) 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 11:09:55,255 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1066436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:04,432 INFO [train.py:968] (1/2) Epoch 24, batch 17550, giga_loss[loss=0.2951, simple_loss=0.3678, pruned_loss=0.1112, over 29016.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3469, pruned_loss=0.09369, over 5675626.98 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3259, pruned_loss=0.08429, over 5762060.76 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3486, pruned_loss=0.0943, over 5666154.17 frames. ], batch size: 128, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:10:12,903 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:15,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:43,798 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:10:47,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-12 11:10:51,114 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 17600, giga_loss[loss=0.2189, simple_loss=0.2986, pruned_loss=0.06959, over 28797.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.344, pruned_loss=0.09309, over 5677398.28 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3258, pruned_loss=0.08421, over 5764794.51 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3458, pruned_loss=0.09381, over 5666096.54 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:11:08,220 INFO [optim.py:369] (1/2) 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 11:11:21,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8566, 2.0390, 1.6313, 1.9964], device='cuda:1'), covar=tensor([0.2672, 0.2831, 0.3285, 0.2648], device='cuda:1'), in_proj_covar=tensor([0.1538, 0.1105, 0.1359, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 11:11:27,178 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:11:29,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:11:41,264 INFO [train.py:968] (1/2) Epoch 24, batch 17650, giga_loss[loss=0.217, simple_loss=0.2928, pruned_loss=0.07057, over 28464.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3378, pruned_loss=0.09073, over 5679663.15 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.326, pruned_loss=0.08427, over 5763410.19 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3392, pruned_loss=0.09132, over 5670914.98 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:11:57,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:12:26,619 INFO [train.py:968] (1/2) Epoch 24, batch 17700, giga_loss[loss=0.2922, simple_loss=0.3605, pruned_loss=0.1119, over 28204.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.331, pruned_loss=0.08788, over 5689776.80 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3262, pruned_loss=0.08442, over 5766412.38 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3321, pruned_loss=0.08829, over 5678789.90 frames. ], batch size: 368, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:12:36,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-12 11:12:37,151 INFO [optim.py:369] (1/2) 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 11:13:02,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 11:13:09,063 INFO [train.py:968] (1/2) Epoch 24, batch 17750, giga_loss[loss=0.2263, simple_loss=0.2936, pruned_loss=0.07954, over 27765.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3247, pruned_loss=0.08549, over 5691508.36 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3271, pruned_loss=0.08474, over 5767921.56 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3248, pruned_loss=0.08559, over 5678431.53 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:13:38,113 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 11:13:50,411 INFO [train.py:968] (1/2) Epoch 24, batch 17800, libri_loss[loss=0.3104, simple_loss=0.3741, pruned_loss=0.1233, over 19731.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.319, pruned_loss=0.08304, over 5683900.46 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3275, pruned_loss=0.08493, over 5753306.34 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.3183, pruned_loss=0.08289, over 5683993.16 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:14:03,415 INFO [optim.py:369] (1/2) 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 11:14:12,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8644, 0.9669, 0.8493, 0.8314], device='cuda:1'), covar=tensor([0.1828, 0.2152, 0.1528, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.1961, 0.1885, 0.1804, 0.1951], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 11:14:32,589 INFO [train.py:968] (1/2) Epoch 24, batch 17850, giga_loss[loss=0.2159, simple_loss=0.2942, pruned_loss=0.06877, over 28493.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.3144, pruned_loss=0.08101, over 5688001.88 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.328, pruned_loss=0.08505, over 5753784.54 frames. ], giga_tot_loss[loss=0.2373, simple_loss=0.3132, pruned_loss=0.08072, over 5686514.66 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:15:04,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9116, 3.0319, 2.0537, 1.0738], device='cuda:1'), covar=tensor([0.9763, 0.3496, 0.4369, 0.7995], device='cuda:1'), in_proj_covar=tensor([0.1773, 0.1677, 0.1616, 0.1444], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 11:15:12,813 INFO [train.py:968] (1/2) Epoch 24, batch 17900, giga_loss[loss=0.2071, simple_loss=0.2801, pruned_loss=0.06708, over 28520.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3115, pruned_loss=0.07953, over 5688535.91 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3282, pruned_loss=0.08503, over 5747437.01 frames. ], giga_tot_loss[loss=0.2341, simple_loss=0.3099, pruned_loss=0.07917, over 5690795.39 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:15:25,607 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:15:49,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-12 11:15:58,486 INFO [train.py:968] (1/2) Epoch 24, batch 17950, giga_loss[loss=0.211, simple_loss=0.2806, pruned_loss=0.07067, over 28818.00 frames. ], tot_loss[loss=0.2321, simple_loss=0.308, pruned_loss=0.07809, over 5690502.85 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3285, pruned_loss=0.08528, over 5751125.55 frames. ], giga_tot_loss[loss=0.2306, simple_loss=0.3062, pruned_loss=0.07749, over 5688018.05 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:16:17,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1066871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:16:21,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1066875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:16:41,332 INFO [train.py:968] (1/2) Epoch 24, batch 18000, giga_loss[loss=0.1804, simple_loss=0.2669, pruned_loss=0.04696, over 28947.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3054, pruned_loss=0.07744, over 5688305.24 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3285, pruned_loss=0.08522, over 5752803.36 frames. ], giga_tot_loss[loss=0.2288, simple_loss=0.3038, pruned_loss=0.07694, over 5684424.56 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:16:41,332 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 11:16:50,592 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 11:16:58,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6156, 1.9504, 2.0354, 1.5671], device='cuda:1'), covar=tensor([0.3350, 0.2476, 0.2278, 0.3023], device='cuda:1'), in_proj_covar=tensor([0.1966, 0.1889, 0.1807, 0.1955], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 11:17:03,463 INFO [optim.py:369] (1/2) 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 11:17:33,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0986, 2.8844, 1.3163, 1.3292], device='cuda:1'), covar=tensor([0.1298, 0.0454, 0.1091, 0.1629], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0552, 0.0391, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 11:17:34,233 INFO [train.py:968] (1/2) Epoch 24, batch 18050, giga_loss[loss=0.2099, simple_loss=0.2888, pruned_loss=0.06548, over 28833.00 frames. ], tot_loss[loss=0.2272, simple_loss=0.3024, pruned_loss=0.076, over 5700517.38 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3286, pruned_loss=0.08531, over 5756515.35 frames. ], giga_tot_loss[loss=0.2255, simple_loss=0.3004, pruned_loss=0.07532, over 5692734.20 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:17:40,831 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1066954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:17:42,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1066957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:18:09,403 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1066986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:18:20,072 INFO [train.py:968] (1/2) Epoch 24, batch 18100, giga_loss[loss=0.1935, simple_loss=0.2718, pruned_loss=0.05756, over 28719.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2987, pruned_loss=0.07425, over 5691911.54 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3285, pruned_loss=0.08512, over 5757669.50 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.2969, pruned_loss=0.07374, over 5683741.90 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:18:27,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2915, 1.9580, 1.5626, 0.5042], device='cuda:1'), covar=tensor([0.6602, 0.3491, 0.4472, 0.7408], device='cuda:1'), in_proj_covar=tensor([0.1772, 0.1675, 0.1611, 0.1443], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 11:18:30,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4795, 1.7901, 1.5531, 1.6193], device='cuda:1'), covar=tensor([0.1983, 0.2178, 0.2257, 0.2131], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0734, 0.0706, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 11:18:34,400 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1067014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:18:34,734 INFO [optim.py:369] (1/2) 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 11:18:36,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1067017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:18:39,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1067022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:19:00,784 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1067046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:19:02,569 INFO [train.py:968] (1/2) Epoch 24, batch 18150, giga_loss[loss=0.2049, simple_loss=0.2852, pruned_loss=0.06234, over 28739.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2964, pruned_loss=0.07299, over 5698710.51 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3287, pruned_loss=0.0852, over 5761686.91 frames. ], giga_tot_loss[loss=0.2192, simple_loss=0.2939, pruned_loss=0.07222, over 5686938.84 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:19:49,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5510, 3.9866, 1.8586, 1.7507], device='cuda:1'), covar=tensor([0.0980, 0.0341, 0.0859, 0.1261], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0552, 0.0392, 0.0430], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 11:19:50,308 INFO [train.py:968] (1/2) Epoch 24, batch 18200, giga_loss[loss=0.1849, simple_loss=0.2651, pruned_loss=0.05239, over 28773.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2939, pruned_loss=0.07196, over 5705704.29 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3284, pruned_loss=0.08507, over 5764020.33 frames. ], giga_tot_loss[loss=0.217, simple_loss=0.2916, pruned_loss=0.07123, over 5693201.62 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:20:04,676 INFO [optim.py:369] (1/2) 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 11:20:34,341 INFO [train.py:968] (1/2) Epoch 24, batch 18250, giga_loss[loss=0.21, simple_loss=0.2817, pruned_loss=0.06913, over 28892.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2938, pruned_loss=0.07264, over 5704659.02 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3291, pruned_loss=0.08546, over 5766350.88 frames. ], giga_tot_loss[loss=0.2168, simple_loss=0.2906, pruned_loss=0.07146, over 5691552.76 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:21:25,933 INFO [train.py:968] (1/2) Epoch 24, batch 18300, giga_loss[loss=0.303, simple_loss=0.368, pruned_loss=0.119, over 27645.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3029, pruned_loss=0.07736, over 5706777.22 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3287, pruned_loss=0.08511, over 5768938.10 frames. ], giga_tot_loss[loss=0.2266, simple_loss=0.3002, pruned_loss=0.07653, over 5693027.11 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:21:40,238 INFO [optim.py:369] (1/2) 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 11:21:41,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-12 11:22:08,186 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.52 vs. limit=5.0
+2023-03-12 11:22:09,634 INFO [train.py:968] (1/2) Epoch 24, batch 18350, giga_loss[loss=0.353, simple_loss=0.4176, pruned_loss=0.1442, over 28776.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.3162, pruned_loss=0.08382, over 5707776.18 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3289, pruned_loss=0.08522, over 5772739.65 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3132, pruned_loss=0.08292, over 5691140.87 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:22:11,970 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1067250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:22:50,964 INFO [train.py:968] (1/2) Epoch 24, batch 18400, giga_loss[loss=0.2734, simple_loss=0.3549, pruned_loss=0.096, over 28793.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3276, pruned_loss=0.08953, over 5712797.82 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.329, pruned_loss=0.08513, over 5776168.00 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.325, pruned_loss=0.08895, over 5694899.20 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:23:06,699 INFO [optim.py:369] (1/2) 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 11:23:34,457 INFO [train.py:968] (1/2) Epoch 24, batch 18450, giga_loss[loss=0.2717, simple_loss=0.3511, pruned_loss=0.09617, over 28846.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3343, pruned_loss=0.09188, over 5702292.35 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3287, pruned_loss=0.08499, over 5777064.89 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3325, pruned_loss=0.09163, over 5686498.25 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:24:05,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4376, 1.7381, 1.6808, 1.2477], device='cuda:1'), covar=tensor([0.1951, 0.2833, 0.1714, 0.1999], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0705, 0.0965, 0.0864], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 11:24:13,323 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1067393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:24:15,682 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1067397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:24:16,669 INFO [train.py:968] (1/2) Epoch 24, batch 18500, libri_loss[loss=0.2478, simple_loss=0.3296, pruned_loss=0.08301, over 29513.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3376, pruned_loss=0.09222, over 5695305.35 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3293, pruned_loss=0.08515, over 5770904.35 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3358, pruned_loss=0.0921, over 5686371.84 frames. ], batch size: 81, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:24:30,285 INFO [optim.py:369] (1/2) 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 11:24:40,090 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1067425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:25:03,502 INFO [train.py:968] (1/2) Epoch 24, batch 18550, libri_loss[loss=0.2761, simple_loss=0.3575, pruned_loss=0.09737, over 29393.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3401, pruned_loss=0.09269, over 5691564.60 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3294, pruned_loss=0.0851, over 5773566.02 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3388, pruned_loss=0.09277, over 5680840.78 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:25:25,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6591, 1.1452, 4.7139, 3.5238], device='cuda:1'), covar=tensor([0.1787, 0.3232, 0.0395, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0657, 0.0967, 0.0926], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 11:25:47,300 INFO [train.py:968] (1/2) Epoch 24, batch 18600, libri_loss[loss=0.223, simple_loss=0.3005, pruned_loss=0.07276, over 28574.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3421, pruned_loss=0.09407, over 5695335.78 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3295, pruned_loss=0.08507, over 5772518.03 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3414, pruned_loss=0.09438, over 5685486.63 frames. ], batch size: 63, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:25:56,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-12 11:26:01,289 INFO [optim.py:369] (1/2) 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 11:26:23,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1067540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:26:25,294 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 18650, giga_loss[loss=0.3098, simple_loss=0.3795, pruned_loss=0.12, over 28663.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3445, pruned_loss=0.09582, over 5694460.23 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3298, pruned_loss=0.0852, over 5771253.08 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3439, pruned_loss=0.09621, over 5685516.80 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:26:43,915 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.38 vs. limit=5.0
+2023-03-12 11:26:51,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2735, 2.5060, 1.2634, 1.4890], device='cuda:1'), covar=tensor([0.1073, 0.0344, 0.0966, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0552, 0.0393, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 11:26:55,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1067572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:27:17,888 INFO [train.py:968] (1/2) Epoch 24, batch 18700, giga_loss[loss=0.2729, simple_loss=0.3511, pruned_loss=0.09732, over 29077.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3482, pruned_loss=0.09849, over 5698561.87 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3298, pruned_loss=0.0852, over 5771253.08 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3478, pruned_loss=0.09879, over 5691601.09 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:27:18,336 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 11:27:33,496 INFO [optim.py:369] (1/2) 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 11:27:57,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7936, 2.2413, 2.0042, 1.8582], device='cuda:1'), covar=tensor([0.0765, 0.0259, 0.0287, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 11:28:03,331 INFO [train.py:968] (1/2) Epoch 24, batch 18750, libri_loss[loss=0.2892, simple_loss=0.3643, pruned_loss=0.1071, over 29401.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3505, pruned_loss=0.09866, over 5694834.78 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3302, pruned_loss=0.08539, over 5762960.24 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3501, pruned_loss=0.09894, over 5696165.07 frames. ], batch size: 92, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:28:22,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3526, 1.1925, 1.1698, 1.5063], device='cuda:1'), covar=tensor([0.0823, 0.0377, 0.0366, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 11:28:24,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.72 vs. limit=5.0
+2023-03-12 11:28:46,465 INFO [train.py:968] (1/2) Epoch 24, batch 18800, giga_loss[loss=0.2972, simple_loss=0.3686, pruned_loss=0.1129, over 28906.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3531, pruned_loss=0.09968, over 5702217.60 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3304, pruned_loss=0.08542, over 5764932.75 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.353, pruned_loss=0.1001, over 5700572.62 frames. ], batch size: 186, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:29:00,678 INFO [optim.py:369] (1/2) 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 11:29:10,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6549, 1.8223, 1.7543, 1.6105], device='cuda:1'), covar=tensor([0.2108, 0.2487, 0.2457, 0.2368], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0741, 0.0712, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 11:29:30,416 INFO [train.py:968] (1/2) Epoch 24, batch 18850, giga_loss[loss=0.2982, simple_loss=0.3779, pruned_loss=0.1093, over 28975.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3545, pruned_loss=0.09976, over 5692030.85 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3308, pruned_loss=0.08556, over 5758416.76 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3544, pruned_loss=0.1002, over 5695975.78 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:29:31,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6987, 2.1164, 1.7672, 1.8134], device='cuda:1'), covar=tensor([0.0762, 0.0265, 0.0301, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 11:29:59,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2196, 1.1079, 0.9200, 1.4449], device='cuda:1'), covar=tensor([0.0783, 0.0383, 0.0390, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0118, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 11:30:11,246 INFO [train.py:968] (1/2) Epoch 24, batch 18900, giga_loss[loss=0.3142, simple_loss=0.3749, pruned_loss=0.1267, over 27568.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3529, pruned_loss=0.09766, over 5696425.83 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3311, pruned_loss=0.08555, over 5761867.04 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3531, pruned_loss=0.09831, over 5695100.00 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:30:26,855 INFO [optim.py:369] (1/2) 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 11:30:52,838 INFO [train.py:968] (1/2) Epoch 24, batch 18950, giga_loss[loss=0.2618, simple_loss=0.3473, pruned_loss=0.08815, over 28824.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3507, pruned_loss=0.09549, over 5701623.77 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3308, pruned_loss=0.08531, over 5763527.70 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3514, pruned_loss=0.09636, over 5698262.84 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:30:59,976 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 11:31:00,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4376, 1.4009, 1.6250, 1.4152], device='cuda:1'), covar=tensor([0.3365, 0.3147, 0.2283, 0.2778], device='cuda:1'), in_proj_covar=tensor([0.1988, 0.1918, 0.1835, 0.1985], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 11:31:31,398 INFO [train.py:968] (1/2) Epoch 24, batch 19000, giga_loss[loss=0.2695, simple_loss=0.3516, pruned_loss=0.09371, over 28627.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3489, pruned_loss=0.09411, over 5703208.47 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3309, pruned_loss=0.08532, over 5761904.82 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3495, pruned_loss=0.09491, over 5701420.17 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:31:44,390 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1067912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:31:46,028 INFO [optim.py:369] (1/2) 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 11:32:14,824 INFO [train.py:968] (1/2) Epoch 24, batch 19050, giga_loss[loss=0.3521, simple_loss=0.3792, pruned_loss=0.1626, over 23522.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3517, pruned_loss=0.09845, over 5685087.20 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3314, pruned_loss=0.08564, over 5758030.64 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3526, pruned_loss=0.09918, over 5685574.00 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:33:00,644 INFO [train.py:968] (1/2) Epoch 24, batch 19100, giga_loss[loss=0.2594, simple_loss=0.3329, pruned_loss=0.09297, over 28892.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3557, pruned_loss=0.1035, over 5683218.68 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3319, pruned_loss=0.08598, over 5759726.08 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3564, pruned_loss=0.1042, over 5680264.41 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:33:08,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5567, 1.7946, 1.2987, 1.3204], device='cuda:1'), covar=tensor([0.1149, 0.0674, 0.1194, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0444, 0.0519, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 11:33:15,773 INFO [optim.py:369] (1/2) 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 11:33:15,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1068016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:33:41,471 INFO [train.py:968] (1/2) Epoch 24, batch 19150, giga_loss[loss=0.2518, simple_loss=0.3269, pruned_loss=0.08838, over 28803.00 frames. ], tot_loss[loss=0.28, simple_loss=0.354, pruned_loss=0.1029, over 5694143.98 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3318, pruned_loss=0.08579, over 5761000.58 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3549, pruned_loss=0.1038, over 5689832.19 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:34:23,614 INFO [train.py:968] (1/2) Epoch 24, batch 19200, giga_loss[loss=0.3051, simple_loss=0.3698, pruned_loss=0.1202, over 29017.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3522, pruned_loss=0.1023, over 5699495.83 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3325, pruned_loss=0.08585, over 5763783.48 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3531, pruned_loss=0.1036, over 5691111.32 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:34:40,501 INFO [optim.py:369] (1/2) 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 11:35:03,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5710, 3.4188, 3.2489, 1.7900], device='cuda:1'), covar=tensor([0.0802, 0.0866, 0.0767, 0.2339], device='cuda:1'), in_proj_covar=tensor([0.1223, 0.1132, 0.0955, 0.0720], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 11:35:07,234 INFO [train.py:968] (1/2) Epoch 24, batch 19250, giga_loss[loss=0.2522, simple_loss=0.3329, pruned_loss=0.08575, over 28775.00 frames. ], tot_loss[loss=0.277, simple_loss=0.351, pruned_loss=0.1016, over 5693205.93 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3325, pruned_loss=0.08585, over 5758198.95 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.1029, over 5690902.03 frames. ], batch size: 284, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:35:53,833 INFO [train.py:968] (1/2) Epoch 24, batch 19300, giga_loss[loss=0.2885, simple_loss=0.3518, pruned_loss=0.1126, over 26658.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.35, pruned_loss=0.1004, over 5687432.76 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3326, pruned_loss=0.08579, over 5760188.32 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3511, pruned_loss=0.1018, over 5682751.88 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:36:08,284 INFO [optim.py:369] (1/2) 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 11:36:36,406 INFO [train.py:968] (1/2) Epoch 24, batch 19350, giga_loss[loss=0.2244, simple_loss=0.3122, pruned_loss=0.0683, over 28921.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3475, pruned_loss=0.09835, over 5695622.23 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3324, pruned_loss=0.08554, over 5765156.73 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3491, pruned_loss=0.1001, over 5685266.49 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:36:49,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-12 11:37:15,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1068287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:37:23,036 INFO [train.py:968] (1/2) Epoch 24, batch 19400, giga_loss[loss=0.23, simple_loss=0.3098, pruned_loss=0.0751, over 28608.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3417, pruned_loss=0.09504, over 5689797.05 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3322, pruned_loss=0.08534, over 5766653.09 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3432, pruned_loss=0.0967, over 5679811.67 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:37:27,475 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1068301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:37:43,687 INFO [optim.py:369] (1/2) 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 11:38:09,669 INFO [train.py:968] (1/2) Epoch 24, batch 19450, giga_loss[loss=0.2416, simple_loss=0.3188, pruned_loss=0.0822, over 28925.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3368, pruned_loss=0.09246, over 5690170.53 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3324, pruned_loss=0.08535, over 5768258.25 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.338, pruned_loss=0.09395, over 5678825.22 frames. ], batch size: 213, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:38:51,334 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 19500, giga_loss[loss=0.2113, simple_loss=0.2951, pruned_loss=0.06372, over 28691.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3327, pruned_loss=0.09077, over 5679094.68 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3328, pruned_loss=0.08559, over 5755647.84 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3333, pruned_loss=0.09187, over 5679075.33 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:39:15,622 INFO [optim.py:369] (1/2) 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 11:39:28,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1068430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:39:32,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1068433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:39:44,679 INFO [train.py:968] (1/2) Epoch 24, batch 19550, giga_loss[loss=0.2475, simple_loss=0.3364, pruned_loss=0.07932, over 28552.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.332, pruned_loss=0.08974, over 5674285.77 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3331, pruned_loss=0.08575, over 5748367.70 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3321, pruned_loss=0.09054, over 5680045.91 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 11:39:56,296 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1068462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:40:02,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 11:40:22,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4935, 1.7526, 1.4190, 1.5417], device='cuda:1'), covar=tensor([0.2719, 0.2858, 0.3192, 0.2503], device='cuda:1'), in_proj_covar=tensor([0.1537, 0.1106, 0.1355, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 11:40:22,761 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 11:40:25,602 INFO [train.py:968] (1/2) Epoch 24, batch 19600, giga_loss[loss=0.261, simple_loss=0.3381, pruned_loss=0.09193, over 29003.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3328, pruned_loss=0.08932, over 5694135.90 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3336, pruned_loss=0.08586, over 5752327.21 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3324, pruned_loss=0.08997, over 5693111.35 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:40:43,252 INFO [optim.py:369] (1/2) 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 11:40:55,519 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1068534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:40:57,405 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1068537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:41:06,331 INFO [train.py:968] (1/2) Epoch 24, batch 19650, giga_loss[loss=0.2318, simple_loss=0.3171, pruned_loss=0.07329, over 29024.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3327, pruned_loss=0.08922, over 5697548.27 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.334, pruned_loss=0.08589, over 5755425.99 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3321, pruned_loss=0.08987, over 5691850.94 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:41:16,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1068561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:41:20,996 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1068566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:41:46,067 INFO [train.py:968] (1/2) Epoch 24, batch 19700, libri_loss[loss=0.2445, simple_loss=0.3259, pruned_loss=0.08159, over 29331.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3301, pruned_loss=0.08761, over 5709131.97 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.334, pruned_loss=0.08569, over 5755202.05 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3294, pruned_loss=0.08838, over 5703345.73 frames. ], batch size: 67, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:41:51,766 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6283, 1.8932, 1.5785, 1.7367], device='cuda:1'), covar=tensor([0.2401, 0.2525, 0.2801, 0.2266], device='cuda:1'), in_proj_covar=tensor([0.1540, 0.1108, 0.1356, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 11:42:01,489 INFO [optim.py:369] (1/2) 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 11:42:13,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3045, 3.0672, 1.4202, 1.4630], device='cuda:1'), covar=tensor([0.1060, 0.0373, 0.0924, 0.1395], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0551, 0.0392, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 11:42:25,427 INFO [train.py:968] (1/2) Epoch 24, batch 19750, giga_loss[loss=0.2466, simple_loss=0.319, pruned_loss=0.08708, over 28938.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3272, pruned_loss=0.08629, over 5710648.91 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3341, pruned_loss=0.08559, over 5748502.96 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3264, pruned_loss=0.08702, over 5711202.07 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:42:51,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1068676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:42:57,383 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-12 11:43:08,675 INFO [train.py:968] (1/2) Epoch 24, batch 19800, giga_loss[loss=0.2247, simple_loss=0.3033, pruned_loss=0.07302, over 29038.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3246, pruned_loss=0.0854, over 5714334.94 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3342, pruned_loss=0.08565, over 5750239.24 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3238, pruned_loss=0.08591, over 5712970.71 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:43:14,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 11:43:24,579 INFO [optim.py:369] (1/2) 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 11:43:50,230 INFO [train.py:968] (1/2) Epoch 24, batch 19850, giga_loss[loss=0.2219, simple_loss=0.2919, pruned_loss=0.07593, over 28670.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.324, pruned_loss=0.08549, over 5719824.99 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3347, pruned_loss=0.08573, over 5753599.48 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3227, pruned_loss=0.08583, over 5714833.61 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:43:51,966 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-12 11:44:01,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8477, 1.7909, 2.0307, 1.5978], device='cuda:1'), covar=tensor([0.1765, 0.2647, 0.1398, 0.1697], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0704, 0.0962, 0.0860], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 11:44:33,186 INFO [train.py:968] (1/2) Epoch 24, batch 19900, giga_loss[loss=0.2245, simple_loss=0.3056, pruned_loss=0.0717, over 28297.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3219, pruned_loss=0.08479, over 5719395.37 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3346, pruned_loss=0.08571, over 5755891.33 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3207, pruned_loss=0.08508, over 5712891.69 frames. ], batch size: 65, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:44:49,137 INFO [optim.py:369] (1/2) 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 11:44:50,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1068819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:44:52,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1068822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:45:12,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3460, 3.2881, 1.5531, 1.5154], device='cuda:1'), covar=tensor([0.1065, 0.0357, 0.0904, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0550, 0.0392, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 11:45:14,014 INFO [train.py:968] (1/2) Epoch 24, batch 19950, giga_loss[loss=0.2241, simple_loss=0.2982, pruned_loss=0.07501, over 28798.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3203, pruned_loss=0.08411, over 5713729.51 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3351, pruned_loss=0.08579, over 5748792.94 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3186, pruned_loss=0.08425, over 5713576.47 frames. ], batch size: 99, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:45:17,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1068851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:45:35,076 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 11:45:50,699 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1912, 1.1490, 3.6990, 3.0083], device='cuda:1'), covar=tensor([0.1720, 0.2909, 0.0442, 0.1023], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0657, 0.0971, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 11:45:55,750 INFO [train.py:968] (1/2) Epoch 24, batch 20000, libri_loss[loss=0.2611, simple_loss=0.3337, pruned_loss=0.09419, over 29667.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3194, pruned_loss=0.08359, over 5712610.11 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3358, pruned_loss=0.08596, over 5743471.16 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.3171, pruned_loss=0.08348, over 5716916.36 frames. ], batch size: 69, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:46:11,794 INFO [optim.py:369] (1/2) 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 11:46:25,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1068936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:46:34,597 INFO [train.py:968] (1/2) Epoch 24, batch 20050, giga_loss[loss=0.2541, simple_loss=0.327, pruned_loss=0.09058, over 27935.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3181, pruned_loss=0.08298, over 5722355.64 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3368, pruned_loss=0.08646, over 5746162.95 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3151, pruned_loss=0.08241, over 5722900.38 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:46:58,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5628, 1.7628, 1.2476, 1.3137], device='cuda:1'), covar=tensor([0.1046, 0.0636, 0.1083, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0447, 0.0522, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 11:47:12,653 INFO [train.py:968] (1/2) Epoch 24, batch 20100, giga_loss[loss=0.2559, simple_loss=0.3285, pruned_loss=0.09172, over 28951.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3168, pruned_loss=0.08231, over 5729590.50 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3374, pruned_loss=0.08664, over 5748065.91 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3136, pruned_loss=0.08164, over 5728032.69 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:47:27,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 11:47:29,396 INFO [optim.py:369] (1/2) 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 11:47:57,574 INFO [train.py:968] (1/2) Epoch 24, batch 20150, giga_loss[loss=0.28, simple_loss=0.351, pruned_loss=0.1045, over 28804.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3198, pruned_loss=0.08428, over 5719633.99 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3379, pruned_loss=0.08682, over 5740328.44 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3165, pruned_loss=0.0835, over 5724733.15 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:48:26,800 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:48:29,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:48:37,598 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 20200, giga_loss[loss=0.328, simple_loss=0.3858, pruned_loss=0.1351, over 27592.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3262, pruned_loss=0.08869, over 5713208.38 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.338, pruned_loss=0.08688, over 5741968.25 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3234, pruned_loss=0.08803, over 5715537.03 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:48:58,721 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:49:08,155 INFO [optim.py:369] (1/2) 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 11:49:12,475 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:49:37,777 INFO [train.py:968] (1/2) Epoch 24, batch 20250, giga_loss[loss=0.3281, simple_loss=0.3944, pruned_loss=0.1309, over 28676.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.336, pruned_loss=0.09508, over 5696046.69 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3381, pruned_loss=0.08691, over 5746170.26 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3334, pruned_loss=0.09464, over 5693186.40 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:50:23,936 INFO [train.py:968] (1/2) Epoch 24, batch 20300, giga_loss[loss=0.2707, simple_loss=0.3516, pruned_loss=0.09487, over 28966.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3408, pruned_loss=0.09732, over 5692822.14 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3381, pruned_loss=0.0868, over 5746129.10 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3388, pruned_loss=0.09726, over 5689802.97 frames. ], batch size: 164, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:50:44,134 INFO [optim.py:369] (1/2) 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 11:51:10,953 INFO [train.py:968] (1/2) Epoch 24, batch 20350, giga_loss[loss=0.2805, simple_loss=0.3569, pruned_loss=0.102, over 28924.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3461, pruned_loss=0.09968, over 5686382.66 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3385, pruned_loss=0.08682, over 5749990.98 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3444, pruned_loss=0.09993, over 5678894.62 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:51:11,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 11:51:14,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4342, 3.5259, 1.6030, 1.5076], device='cuda:1'), covar=tensor([0.0964, 0.0281, 0.0901, 0.1406], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0553, 0.0393, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 11:51:15,791 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 11:51:56,100 INFO [train.py:968] (1/2) Epoch 24, batch 20400, giga_loss[loss=0.2811, simple_loss=0.3559, pruned_loss=0.1032, over 28452.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3507, pruned_loss=0.1019, over 5689567.70 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3391, pruned_loss=0.08717, over 5754936.39 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3491, pruned_loss=0.1022, over 5677145.82 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:52:05,292 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:52:16,971 INFO [optim.py:369] (1/2) 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 11:52:43,659 INFO [train.py:968] (1/2) Epoch 24, batch 20450, giga_loss[loss=0.3337, simple_loss=0.3744, pruned_loss=0.1465, over 23455.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3567, pruned_loss=0.1059, over 5676754.52 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3392, pruned_loss=0.08722, over 5756071.12 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3555, pruned_loss=0.1063, over 5665337.90 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:52:49,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:52:57,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7562, 1.0048, 2.8236, 2.7915], device='cuda:1'), covar=tensor([0.1881, 0.3020, 0.0506, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0651, 0.0963, 0.0922], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 11:53:27,648 INFO [train.py:968] (1/2) Epoch 24, batch 20500, libri_loss[loss=0.2681, simple_loss=0.3519, pruned_loss=0.09216, over 29753.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3522, pruned_loss=0.1021, over 5679125.96 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08749, over 5750640.71 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3513, pruned_loss=0.1025, over 5672730.98 frames. ], batch size: 87, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:53:45,750 INFO [optim.py:369] (1/2) 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 11:54:08,849 INFO [train.py:968] (1/2) Epoch 24, batch 20550, giga_loss[loss=0.2602, simple_loss=0.3393, pruned_loss=0.09052, over 28916.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3499, pruned_loss=0.09992, over 5691926.20 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08796, over 5752221.60 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3489, pruned_loss=0.1001, over 5684311.72 frames. ], batch size: 106, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:54:22,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069463.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:54:48,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0847, 2.1331, 2.2709, 1.8163], device='cuda:1'), covar=tensor([0.1889, 0.2360, 0.1510, 0.1754], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0706, 0.0965, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 11:54:48,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1069494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:54:51,821 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:54:52,123 INFO [train.py:968] (1/2) Epoch 24, batch 20600, giga_loss[loss=0.2983, simple_loss=0.3692, pruned_loss=0.1137, over 27595.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3487, pruned_loss=0.09866, over 5694404.08 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08794, over 5754482.38 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3481, pruned_loss=0.09911, over 5684986.45 frames. ], batch size: 472, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:55:09,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4517, 1.7690, 1.7729, 1.3488], device='cuda:1'), covar=tensor([0.2998, 0.2277, 0.2508, 0.2865], device='cuda:1'), in_proj_covar=tensor([0.2004, 0.1932, 0.1857, 0.2009], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 11:55:10,623 INFO [optim.py:369] (1/2) 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 11:55:33,074 INFO [train.py:968] (1/2) Epoch 24, batch 20650, giga_loss[loss=0.2976, simple_loss=0.37, pruned_loss=0.1126, over 28550.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3499, pruned_loss=0.0987, over 5697172.83 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3406, pruned_loss=0.08831, over 5758780.48 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3492, pruned_loss=0.0991, over 5683481.42 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:55:46,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 11:56:15,759 INFO [train.py:968] (1/2) Epoch 24, batch 20700, giga_loss[loss=0.2714, simple_loss=0.3509, pruned_loss=0.09598, over 28681.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3516, pruned_loss=0.1003, over 5701909.29 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3406, pruned_loss=0.08837, over 5761707.54 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3514, pruned_loss=0.1008, over 5687240.62 frames. ], batch size: 242, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:56:24,932 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069606.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 11:56:27,059 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069609.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 11:56:37,919 INFO [optim.py:369] (1/2) 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 11:56:51,948 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069638.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 11:56:53,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:56:55,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:56:58,952 INFO [train.py:968] (1/2) Epoch 24, batch 20750, giga_loss[loss=0.2655, simple_loss=0.3419, pruned_loss=0.09456, over 28887.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3538, pruned_loss=0.1019, over 5715160.69 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3405, pruned_loss=0.08842, over 5766674.23 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3541, pruned_loss=0.1027, over 5697003.87 frames. ], batch size: 112, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:57:23,461 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:57:33,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:57:49,131 INFO [train.py:968] (1/2) Epoch 24, batch 20800, giga_loss[loss=0.2559, simple_loss=0.3376, pruned_loss=0.08704, over 28720.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3544, pruned_loss=0.1028, over 5696146.67 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3412, pruned_loss=0.0888, over 5769025.17 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1034, over 5678489.70 frames. ], batch size: 66, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:58:09,514 INFO [optim.py:369] (1/2) 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 11:58:11,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-12 11:58:18,499 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 20850, giga_loss[loss=0.2814, simple_loss=0.352, pruned_loss=0.1054, over 28638.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3558, pruned_loss=0.1044, over 5700406.86 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3416, pruned_loss=0.08917, over 5772047.67 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3556, pruned_loss=0.1049, over 5681896.85 frames. ], batch size: 336, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 11:58:40,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9501, 1.2997, 1.0654, 0.2470], device='cuda:1'), covar=tensor([0.4225, 0.3135, 0.4941, 0.6773], device='cuda:1'), in_proj_covar=tensor([0.1770, 0.1674, 0.1617, 0.1442], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 11:59:13,453 INFO [train.py:968] (1/2) Epoch 24, batch 20900, giga_loss[loss=0.2569, simple_loss=0.3401, pruned_loss=0.08681, over 28667.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3562, pruned_loss=0.1045, over 5706613.94 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3422, pruned_loss=0.08962, over 5772250.03 frames. ], giga_tot_loss[loss=0.2827, simple_loss=0.3558, pruned_loss=0.1048, over 5689633.14 frames. ], batch size: 71, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 11:59:31,769 INFO [optim.py:369] (1/2) 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 11:59:34,245 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 11:59:37,116 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 20950, giga_loss[loss=0.2987, simple_loss=0.3756, pruned_loss=0.1109, over 28881.00 frames. ], tot_loss[loss=0.2816, simple_loss=0.3559, pruned_loss=0.1037, over 5706553.79 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3423, pruned_loss=0.08974, over 5774328.76 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3557, pruned_loss=0.104, over 5690334.86 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:00:01,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1069856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:12,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1069869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:16,044 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1069873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:18,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1069876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:00:35,793 INFO [train.py:968] (1/2) Epoch 24, batch 21000, giga_loss[loss=0.2507, simple_loss=0.3409, pruned_loss=0.08025, over 28932.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3555, pruned_loss=0.102, over 5709968.68 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3425, pruned_loss=0.08983, over 5776225.18 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3556, pruned_loss=0.1025, over 5693858.95 frames. ], batch size: 174, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:00:35,793 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 12:00:44,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4600, 1.8297, 1.4291, 1.3768], device='cuda:1'), covar=tensor([0.3068, 0.3194, 0.3477, 0.2891], device='cuda:1'), in_proj_covar=tensor([0.1541, 0.1112, 0.1359, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 12:00:44,741 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 12:00:51,126 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:01:10,668 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1733, 2.9372, 1.3296, 1.2690], device='cuda:1'), covar=tensor([0.1088, 0.0355, 0.0998, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0551, 0.0391, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 12:01:26,417 INFO [train.py:968] (1/2) Epoch 24, batch 21050, giga_loss[loss=0.2367, simple_loss=0.3238, pruned_loss=0.07485, over 28653.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3546, pruned_loss=0.1012, over 5706977.44 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3428, pruned_loss=0.08994, over 5776197.83 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3547, pruned_loss=0.1018, over 5692661.80 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:01:34,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3921, 1.5079, 1.3018, 1.5686], device='cuda:1'), covar=tensor([0.0812, 0.0333, 0.0340, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 12:01:48,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-12 12:02:05,368 INFO [train.py:968] (1/2) Epoch 24, batch 21100, giga_loss[loss=0.255, simple_loss=0.34, pruned_loss=0.08505, over 28940.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3512, pruned_loss=0.09927, over 5716692.91 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3424, pruned_loss=0.08998, over 5777602.84 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3518, pruned_loss=0.09989, over 5702513.38 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:02:15,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1070012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:02:17,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1070015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:02:17,826 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1070016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 12:02:21,009 INFO [optim.py:369] (1/2) 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 12:02:36,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5672, 1.5848, 1.8140, 1.3855], device='cuda:1'), covar=tensor([0.1698, 0.2544, 0.1366, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0705, 0.0961, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:02:40,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8372, 2.5524, 1.6207, 1.0245], device='cuda:1'), covar=tensor([0.7617, 0.3261, 0.4348, 0.6907], device='cuda:1'), in_proj_covar=tensor([0.1758, 0.1659, 0.1605, 0.1430], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 12:02:40,758 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1070044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:02:43,620 INFO [train.py:968] (1/2) Epoch 24, batch 21150, giga_loss[loss=0.2531, simple_loss=0.3325, pruned_loss=0.08681, over 28939.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3495, pruned_loss=0.0985, over 5714739.53 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3432, pruned_loss=0.09071, over 5773705.17 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3495, pruned_loss=0.09864, over 5704817.12 frames. ], batch size: 227, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:03:22,757 INFO [train.py:968] (1/2) Epoch 24, batch 21200, giga_loss[loss=0.2925, simple_loss=0.3382, pruned_loss=0.1234, over 23705.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.09823, over 5707166.47 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3437, pruned_loss=0.09131, over 5766455.24 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3477, pruned_loss=0.09797, over 5704227.35 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:03:26,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5929, 1.8578, 1.5023, 1.6428], device='cuda:1'), covar=tensor([0.2667, 0.2635, 0.2985, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1539, 0.1112, 0.1357, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 12:03:42,697 INFO [optim.py:369] (1/2) 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 12:04:05,963 INFO [train.py:968] (1/2) Epoch 24, batch 21250, giga_loss[loss=0.2792, simple_loss=0.3544, pruned_loss=0.102, over 28915.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3491, pruned_loss=0.0995, over 5705561.28 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3438, pruned_loss=0.09143, over 5766130.05 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3487, pruned_loss=0.09925, over 5703056.82 frames. ], batch size: 199, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:04:08,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1070151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:04:20,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5941, 1.9108, 1.2631, 1.4666], device='cuda:1'), covar=tensor([0.0962, 0.0552, 0.1031, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0445, 0.0521, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 12:04:37,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3400, 1.6795, 1.3237, 0.9920], device='cuda:1'), covar=tensor([0.2704, 0.2655, 0.3107, 0.2454], device='cuda:1'), in_proj_covar=tensor([0.1541, 0.1114, 0.1357, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 12:04:39,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2206, 4.0345, 3.8318, 1.7592], device='cuda:1'), covar=tensor([0.0582, 0.0759, 0.0744, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.1235, 0.1137, 0.0962, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0015, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 12:04:43,010 INFO [train.py:968] (1/2) Epoch 24, batch 21300, giga_loss[loss=0.269, simple_loss=0.3459, pruned_loss=0.09601, over 28764.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3491, pruned_loss=0.09937, over 5716712.64 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3442, pruned_loss=0.09197, over 5767091.95 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3487, pruned_loss=0.09898, over 5711610.85 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:04:48,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5163, 1.7555, 1.6401, 1.5337], device='cuda:1'), covar=tensor([0.2184, 0.2190, 0.2504, 0.2379], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0745, 0.0715, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 12:05:03,047 INFO [optim.py:369] (1/2) 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 12:05:26,082 INFO [train.py:968] (1/2) Epoch 24, batch 21350, giga_loss[loss=0.2619, simple_loss=0.3447, pruned_loss=0.08955, over 28705.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3481, pruned_loss=0.09813, over 5708449.65 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3446, pruned_loss=0.09235, over 5770206.97 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3476, pruned_loss=0.09769, over 5700359.36 frames. ], batch size: 60, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:05:27,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1148, 1.4424, 0.9518, 0.9673], device='cuda:1'), covar=tensor([0.1437, 0.0718, 0.1696, 0.1435], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0444, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 12:06:07,321 INFO [train.py:968] (1/2) Epoch 24, batch 21400, giga_loss[loss=0.2564, simple_loss=0.3352, pruned_loss=0.08878, over 28650.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3474, pruned_loss=0.09698, over 5720615.40 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3451, pruned_loss=0.09274, over 5772417.54 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3466, pruned_loss=0.09641, over 5711018.30 frames. ], batch size: 262, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:06:26,666 INFO [optim.py:369] (1/2) 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 12:06:45,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5023, 1.1720, 4.8157, 3.5282], device='cuda:1'), covar=tensor([0.1834, 0.3092, 0.0372, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0655, 0.0970, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 12:06:46,750 INFO [train.py:968] (1/2) Epoch 24, batch 21450, giga_loss[loss=0.2406, simple_loss=0.3208, pruned_loss=0.08019, over 28980.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3465, pruned_loss=0.09654, over 5726510.86 frames. ], libri_tot_loss[loss=0.2655, simple_loss=0.3451, pruned_loss=0.09291, over 5773740.71 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3458, pruned_loss=0.09597, over 5717253.72 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:07:21,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-12 12:07:22,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1070391.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 12:07:26,981 INFO [train.py:968] (1/2) Epoch 24, batch 21500, libri_loss[loss=0.3252, simple_loss=0.3805, pruned_loss=0.1349, over 29535.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3456, pruned_loss=0.09663, over 5734151.73 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3456, pruned_loss=0.09347, over 5779266.04 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3446, pruned_loss=0.09577, over 5719271.46 frames. ], batch size: 78, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 12:07:35,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 12:07:50,227 INFO [optim.py:369] (1/2) 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 12:07:57,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3551, 4.9883, 1.7054, 1.7333], device='cuda:1'), covar=tensor([0.1126, 0.0442, 0.0947, 0.1464], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0553, 0.0393, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 12:08:10,112 INFO [train.py:968] (1/2) Epoch 24, batch 21550, giga_loss[loss=0.2444, simple_loss=0.3217, pruned_loss=0.08355, over 28993.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3417, pruned_loss=0.09446, over 5727483.97 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3458, pruned_loss=0.09383, over 5779082.11 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3406, pruned_loss=0.09347, over 5714385.64 frames. ], batch size: 145, lr: 1.33e-03, grad_scale: 2.0
+2023-03-12 12:08:49,467 INFO [train.py:968] (1/2) Epoch 24, batch 21600, giga_loss[loss=0.2375, simple_loss=0.3213, pruned_loss=0.07683, over 29055.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3413, pruned_loss=0.09453, over 5724257.37 frames. ], libri_tot_loss[loss=0.268, simple_loss=0.3468, pruned_loss=0.09461, over 5772700.66 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3394, pruned_loss=0.09305, over 5717835.39 frames. ], batch size: 155, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:09:07,739 INFO [optim.py:369] (1/2) 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 12:09:11,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1070526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:09:17,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1070534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 12:09:20,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1070537.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 12:09:30,522 INFO [train.py:968] (1/2) Epoch 24, batch 21650, giga_loss[loss=0.2508, simple_loss=0.3236, pruned_loss=0.08898, over 28707.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3417, pruned_loss=0.09549, over 5717798.57 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.3475, pruned_loss=0.09509, over 5770395.04 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3395, pruned_loss=0.0939, over 5714040.85 frames. ], batch size: 119, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:09:46,287 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1070566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 12:10:05,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1070588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:10:13,498 INFO [train.py:968] (1/2) Epoch 24, batch 21700, libri_loss[loss=0.2951, simple_loss=0.3549, pruned_loss=0.1177, over 29586.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3389, pruned_loss=0.09424, over 5722742.83 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3475, pruned_loss=0.09527, over 5774375.04 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.337, pruned_loss=0.09277, over 5714546.86 frames. ], batch size: 74, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:10:34,902 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-12 12:10:35,807 INFO [optim.py:369] (1/2) 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 12:10:54,795 INFO [train.py:968] (1/2) Epoch 24, batch 21750, giga_loss[loss=0.3446, simple_loss=0.3884, pruned_loss=0.1505, over 26667.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3367, pruned_loss=0.09323, over 5707286.50 frames. ], libri_tot_loss[loss=0.269, simple_loss=0.3475, pruned_loss=0.09526, over 5759537.58 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3349, pruned_loss=0.09203, over 5713773.35 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:11:11,519 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1070669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:11:13,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1070672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:11:34,252 INFO [train.py:968] (1/2) Epoch 24, batch 21800, giga_loss[loss=0.2558, simple_loss=0.3309, pruned_loss=0.09037, over 28008.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.335, pruned_loss=0.09295, over 5707545.78 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3478, pruned_loss=0.09568, over 5760879.76 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3328, pruned_loss=0.09151, over 5709748.40 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:11:37,831 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:12:12,958 INFO [train.py:968] (1/2) Epoch 24, batch 21850, giga_loss[loss=0.2159, simple_loss=0.2983, pruned_loss=0.06675, over 28931.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3319, pruned_loss=0.09128, over 5713256.19 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3482, pruned_loss=0.09593, over 5763709.80 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3296, pruned_loss=0.08986, over 5711569.32 frames. ], batch size: 136, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:12:31,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3718, 1.6306, 1.3440, 1.0375], device='cuda:1'), covar=tensor([0.2734, 0.2832, 0.3240, 0.2487], device='cuda:1'), in_proj_covar=tensor([0.1541, 0.1111, 0.1355, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 12:12:42,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3905, 5.2148, 4.9869, 2.4560], device='cuda:1'), covar=tensor([0.0403, 0.0585, 0.0658, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.1238, 0.1143, 0.0965, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 12:12:55,903 INFO [train.py:968] (1/2) Epoch 24, batch 21900, libri_loss[loss=0.2756, simple_loss=0.3392, pruned_loss=0.106, over 29677.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3314, pruned_loss=0.09101, over 5713931.74 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3475, pruned_loss=0.09582, over 5768632.53 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3296, pruned_loss=0.0898, over 5706128.53 frames. ], batch size: 73, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:13:18,866 INFO [optim.py:369] (1/2) 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 12:13:39,365 INFO [train.py:968] (1/2) Epoch 24, batch 21950, giga_loss[loss=0.2553, simple_loss=0.3225, pruned_loss=0.09411, over 28305.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3351, pruned_loss=0.09296, over 5700091.67 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.3481, pruned_loss=0.09628, over 5757613.64 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3329, pruned_loss=0.09151, over 5703133.93 frames. ], batch size: 77, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:13:46,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-12 12:14:25,017 INFO [train.py:968] (1/2) Epoch 24, batch 22000, giga_loss[loss=0.3016, simple_loss=0.372, pruned_loss=0.1156, over 27939.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3382, pruned_loss=0.094, over 5709006.85 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3486, pruned_loss=0.09677, over 5761206.43 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3357, pruned_loss=0.09231, over 5706840.19 frames. ], batch size: 412, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:14:29,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9774, 1.3074, 1.0960, 0.2626], device='cuda:1'), covar=tensor([0.3696, 0.3280, 0.4287, 0.5847], device='cuda:1'), in_proj_covar=tensor([0.1762, 0.1657, 0.1607, 0.1430], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 12:14:45,581 INFO [optim.py:369] (1/2) 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 12:15:03,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2025, 1.7016, 1.3211, 0.4544], device='cuda:1'), covar=tensor([0.4821, 0.2828, 0.4182, 0.6386], device='cuda:1'), in_proj_covar=tensor([0.1765, 0.1660, 0.1609, 0.1432], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 12:15:07,672 INFO [train.py:968] (1/2) Epoch 24, batch 22050, giga_loss[loss=0.3122, simple_loss=0.3651, pruned_loss=0.1297, over 23651.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3409, pruned_loss=0.09489, over 5702761.15 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3491, pruned_loss=0.0972, over 5762881.27 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3383, pruned_loss=0.09309, over 5698320.67 frames. ], batch size: 705, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:15:22,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1070963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:15:29,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0947, 1.1046, 3.5173, 2.9855], device='cuda:1'), covar=tensor([0.1583, 0.2529, 0.0484, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0655, 0.0968, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 12:15:52,866 INFO [train.py:968] (1/2) Epoch 24, batch 22100, giga_loss[loss=0.2571, simple_loss=0.3507, pruned_loss=0.08171, over 28610.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.09418, over 5697753.30 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3491, pruned_loss=0.09734, over 5765056.00 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3385, pruned_loss=0.09258, over 5691420.19 frames. ], batch size: 307, lr: 1.33e-03, grad_scale: 8.0
+2023-03-12 12:16:15,506 INFO [optim.py:369] (1/2) 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 12:16:18,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8159, 1.8991, 2.0235, 1.5967], device='cuda:1'), covar=tensor([0.2005, 0.2455, 0.1581, 0.1816], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0703, 0.0960, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:16:35,458 INFO [train.py:968] (1/2) Epoch 24, batch 22150, giga_loss[loss=0.2598, simple_loss=0.3348, pruned_loss=0.09247, over 28658.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3392, pruned_loss=0.09318, over 5700133.91 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3488, pruned_loss=0.09744, over 5765155.14 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3375, pruned_loss=0.09174, over 5693567.40 frames. ], batch size: 85, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:17:20,373 INFO [train.py:968] (1/2) Epoch 24, batch 22200, giga_loss[loss=0.3458, simple_loss=0.3957, pruned_loss=0.1479, over 26700.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3399, pruned_loss=0.09386, over 5700919.13 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3487, pruned_loss=0.09743, over 5764817.56 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3386, pruned_loss=0.0927, over 5695413.22 frames. ], batch size: 555, lr: 1.33e-03, grad_scale: 4.0
+2023-03-12 12:17:27,209 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1071106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:17:29,287 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:17:50,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6083, 1.7908, 1.4915, 1.8675], device='cuda:1'), covar=tensor([0.2651, 0.2774, 0.3046, 0.2583], device='cuda:1'), in_proj_covar=tensor([0.1541, 0.1109, 0.1357, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 12:17:53,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1071138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:18:01,168 INFO [train.py:968] (1/2) Epoch 24, batch 22250, giga_loss[loss=0.2795, simple_loss=0.356, pruned_loss=0.1014, over 28718.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3408, pruned_loss=0.09461, over 5707225.16 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3492, pruned_loss=0.09771, over 5767538.19 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3391, pruned_loss=0.09336, over 5699181.01 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:18:04,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6643, 1.8231, 1.5176, 1.6885], device='cuda:1'), covar=tensor([0.2630, 0.2715, 0.3021, 0.2500], device='cuda:1'), in_proj_covar=tensor([0.1541, 0.1109, 0.1357, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 12:18:13,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9065, 1.2388, 1.3263, 1.0637], device='cuda:1'), covar=tensor([0.2002, 0.1458, 0.2407, 0.1860], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0743, 0.0714, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 12:18:43,688 INFO [train.py:968] (1/2) Epoch 24, batch 22300, giga_loss[loss=0.2514, simple_loss=0.3304, pruned_loss=0.08619, over 28598.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3436, pruned_loss=0.09623, over 5706529.18 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3496, pruned_loss=0.0982, over 5769679.50 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3417, pruned_loss=0.09473, over 5696858.55 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:19:05,481 INFO [optim.py:369] (1/2) 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 12:19:09,688 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1071230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:19:13,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4879, 1.7263, 1.4652, 1.3943], device='cuda:1'), covar=tensor([0.2647, 0.2661, 0.3088, 0.2415], device='cuda:1'), in_proj_covar=tensor([0.1538, 0.1108, 0.1355, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0012, 0.0009], device='cuda:1')
+2023-03-12 12:19:24,464 INFO [train.py:968] (1/2) Epoch 24, batch 22350, giga_loss[loss=0.2748, simple_loss=0.3569, pruned_loss=0.0964, over 28684.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3469, pruned_loss=0.09794, over 5715357.87 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3504, pruned_loss=0.09881, over 5773130.99 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3445, pruned_loss=0.09611, over 5702754.85 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:20:05,978 INFO [train.py:968] (1/2) Epoch 24, batch 22400, giga_loss[loss=0.3219, simple_loss=0.3911, pruned_loss=0.1263, over 28715.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3483, pruned_loss=0.09848, over 5715240.32 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3509, pruned_loss=0.09928, over 5773609.54 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3459, pruned_loss=0.09659, over 5703262.76 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:20:27,812 INFO [optim.py:369] (1/2) 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 12:20:41,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5335, 1.5030, 1.7217, 1.3431], device='cuda:1'), covar=tensor([0.1819, 0.2499, 0.1480, 0.1708], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0704, 0.0960, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:20:49,725 INFO [train.py:968] (1/2) Epoch 24, batch 22450, libri_loss[loss=0.3906, simple_loss=0.4337, pruned_loss=0.1737, over 29541.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.349, pruned_loss=0.09867, over 5704654.91 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3514, pruned_loss=0.09975, over 5762937.14 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3465, pruned_loss=0.09667, over 5704530.30 frames. ], batch size: 79, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:20:52,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-12 12:21:32,967 INFO [train.py:968] (1/2) Epoch 24, batch 22500, giga_loss[loss=0.2669, simple_loss=0.3461, pruned_loss=0.09387, over 28725.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09933, over 5709652.09 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3516, pruned_loss=0.09993, over 5764234.32 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3478, pruned_loss=0.09755, over 5707448.04 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:21:52,952 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 22550, giga_loss[loss=0.251, simple_loss=0.3261, pruned_loss=0.08792, over 28861.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3492, pruned_loss=0.0989, over 5708242.17 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.3517, pruned_loss=0.1, over 5765157.40 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3473, pruned_loss=0.09743, over 5705014.58 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:22:25,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4458, 2.3987, 1.9712, 2.1195], device='cuda:1'), covar=tensor([0.0779, 0.0569, 0.0832, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0446, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 12:22:59,218 INFO [train.py:968] (1/2) Epoch 24, batch 22600, giga_loss[loss=0.2621, simple_loss=0.3428, pruned_loss=0.09064, over 29003.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3469, pruned_loss=0.09797, over 5714001.79 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3519, pruned_loss=0.1004, over 5767780.88 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3452, pruned_loss=0.09638, over 5707732.20 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:23:20,156 INFO [optim.py:369] (1/2) 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 12:23:22,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3770, 2.0276, 1.5285, 0.5332], device='cuda:1'), covar=tensor([0.4810, 0.2503, 0.3948, 0.6276], device='cuda:1'), in_proj_covar=tensor([0.1768, 0.1666, 0.1614, 0.1437], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 12:23:39,864 INFO [train.py:968] (1/2) Epoch 24, batch 22650, giga_loss[loss=0.2489, simple_loss=0.3309, pruned_loss=0.08347, over 28743.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3433, pruned_loss=0.09622, over 5718039.65 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3517, pruned_loss=0.1004, over 5771382.62 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3418, pruned_loss=0.0949, over 5708788.97 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:24:18,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 12:24:20,530 INFO [train.py:968] (1/2) Epoch 24, batch 22700, giga_loss[loss=0.2284, simple_loss=0.3067, pruned_loss=0.07498, over 28654.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3419, pruned_loss=0.09548, over 5704387.73 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3521, pruned_loss=0.1007, over 5764254.66 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3403, pruned_loss=0.09408, over 5703319.44 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:24:26,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1071605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:24:43,753 INFO [optim.py:369] (1/2) 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 12:25:02,453 INFO [train.py:968] (1/2) Epoch 24, batch 22750, giga_loss[loss=0.2873, simple_loss=0.3691, pruned_loss=0.1028, over 28692.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3436, pruned_loss=0.09486, over 5696162.17 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3521, pruned_loss=0.1007, over 5758511.11 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3421, pruned_loss=0.09355, over 5698959.35 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:25:05,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 12:25:22,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2298, 1.5363, 1.5073, 1.0843], device='cuda:1'), covar=tensor([0.1613, 0.2681, 0.1496, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0704, 0.0960, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:25:28,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-12 12:25:43,992 INFO [train.py:968] (1/2) Epoch 24, batch 22800, giga_loss[loss=0.2813, simple_loss=0.3437, pruned_loss=0.1094, over 24083.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3456, pruned_loss=0.09573, over 5700204.26 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3522, pruned_loss=0.101, over 5763503.89 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.344, pruned_loss=0.09419, over 5695727.07 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:25:45,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5452, 1.5983, 1.7398, 1.3226], device='cuda:1'), covar=tensor([0.1853, 0.2487, 0.1564, 0.1789], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0704, 0.0960, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:26:03,489 INFO [optim.py:369] (1/2) 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 12:26:23,187 INFO [train.py:968] (1/2) Epoch 24, batch 22850, giga_loss[loss=0.2656, simple_loss=0.3342, pruned_loss=0.09844, over 29042.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3438, pruned_loss=0.09583, over 5698789.09 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3523, pruned_loss=0.1013, over 5757201.46 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3422, pruned_loss=0.09413, over 5698533.40 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:26:23,505 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1071748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:26:25,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1071751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:26:49,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1071780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:26:50,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5796, 1.6689, 1.7687, 1.3751], device='cuda:1'), covar=tensor([0.1892, 0.2577, 0.1607, 0.1753], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0705, 0.0960, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:27:04,183 INFO [train.py:968] (1/2) Epoch 24, batch 22900, libri_loss[loss=0.2907, simple_loss=0.3628, pruned_loss=0.1093, over 29517.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3431, pruned_loss=0.09701, over 5702838.03 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3528, pruned_loss=0.1018, over 5758533.45 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3411, pruned_loss=0.09511, over 5699806.00 frames. ], batch size: 84, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:27:25,142 INFO [optim.py:369] (1/2) 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 12:27:43,611 INFO [train.py:968] (1/2) Epoch 24, batch 22950, giga_loss[loss=0.2645, simple_loss=0.3241, pruned_loss=0.1025, over 28781.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3422, pruned_loss=0.09778, over 5710840.60 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3533, pruned_loss=0.1022, over 5761680.91 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.34, pruned_loss=0.09579, over 5704302.26 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:28:22,535 INFO [train.py:968] (1/2) Epoch 24, batch 23000, giga_loss[loss=0.2759, simple_loss=0.3351, pruned_loss=0.1084, over 28790.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3405, pruned_loss=0.09768, over 5717063.76 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.353, pruned_loss=0.1023, over 5766471.86 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3385, pruned_loss=0.09575, over 5705617.96 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:28:25,229 INFO [scaling.py:679] (1/2) 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] (1/2) 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 12:29:02,872 INFO [train.py:968] (1/2) Epoch 24, batch 23050, giga_loss[loss=0.2539, simple_loss=0.3369, pruned_loss=0.08543, over 28765.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3399, pruned_loss=0.09705, over 5708587.58 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3532, pruned_loss=0.1025, over 5754862.74 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3378, pruned_loss=0.09523, over 5709026.87 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:29:20,591 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1071970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:29:43,936 INFO [train.py:968] (1/2) Epoch 24, batch 23100, giga_loss[loss=0.277, simple_loss=0.333, pruned_loss=0.1105, over 23918.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3363, pruned_loss=0.0951, over 5708188.07 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3536, pruned_loss=0.1028, over 5756504.73 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3341, pruned_loss=0.0933, over 5706644.92 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:29:58,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 12:30:05,970 INFO [optim.py:369] (1/2) 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 12:30:10,106 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1072031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:30:23,853 INFO [train.py:968] (1/2) Epoch 24, batch 23150, giga_loss[loss=0.2588, simple_loss=0.3276, pruned_loss=0.095, over 28916.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3324, pruned_loss=0.0933, over 5710775.45 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3537, pruned_loss=0.1031, over 5758032.16 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3301, pruned_loss=0.0914, over 5707088.86 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:31:07,229 INFO [train.py:968] (1/2) Epoch 24, batch 23200, giga_loss[loss=0.2222, simple_loss=0.3048, pruned_loss=0.06982, over 29053.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3306, pruned_loss=0.09211, over 5711301.95 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.354, pruned_loss=0.1033, over 5758763.00 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3284, pruned_loss=0.09044, over 5707638.05 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:31:30,081 INFO [optim.py:369] (1/2) 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 12:31:48,437 INFO [train.py:968] (1/2) Epoch 24, batch 23250, giga_loss[loss=0.2276, simple_loss=0.307, pruned_loss=0.07408, over 28780.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3319, pruned_loss=0.09223, over 5717100.37 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3545, pruned_loss=0.1036, over 5761490.65 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3294, pruned_loss=0.09041, over 5710988.66 frames. ], batch size: 66, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:32:30,972 INFO [train.py:968] (1/2) Epoch 24, batch 23300, giga_loss[loss=0.2577, simple_loss=0.3426, pruned_loss=0.08638, over 28567.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3352, pruned_loss=0.09365, over 5702524.92 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.355, pruned_loss=0.104, over 5746014.56 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3323, pruned_loss=0.09156, over 5710115.67 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:32:53,237 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 23350, libri_loss[loss=0.2717, simple_loss=0.3466, pruned_loss=0.0984, over 29515.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3389, pruned_loss=0.09522, over 5705332.69 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3547, pruned_loss=0.104, over 5749613.37 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3363, pruned_loss=0.09332, over 5707044.79 frames. ], batch size: 80, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:33:10,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9128, 5.7221, 5.4319, 2.8512], device='cuda:1'), covar=tensor([0.0384, 0.0560, 0.0641, 0.1624], device='cuda:1'), in_proj_covar=tensor([0.1247, 0.1152, 0.0973, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 12:33:44,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4114, 1.7482, 1.4760, 1.5353], device='cuda:1'), covar=tensor([0.0748, 0.0299, 0.0324, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 12:33:52,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4106, 1.6909, 1.4486, 1.5953], device='cuda:1'), covar=tensor([0.0757, 0.0309, 0.0330, 0.0873], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0100, 0.0072, 0.0064, 0.0110], device='cuda:1')
+2023-03-12 12:33:54,699 INFO [train.py:968] (1/2) Epoch 24, batch 23400, libri_loss[loss=0.3342, simple_loss=0.3953, pruned_loss=0.1366, over 29381.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3434, pruned_loss=0.09738, over 5703208.73 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3551, pruned_loss=0.1043, over 5753426.14 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3407, pruned_loss=0.09544, over 5700114.81 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:34:19,561 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 12:34:19,696 INFO [optim.py:369] (1/2) 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 12:34:36,659 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 23450, giga_loss[loss=0.2961, simple_loss=0.367, pruned_loss=0.1126, over 28780.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3463, pruned_loss=0.0989, over 5702567.90 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3553, pruned_loss=0.1045, over 5756298.22 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3438, pruned_loss=0.09707, over 5696607.17 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:35:25,191 INFO [train.py:968] (1/2) Epoch 24, batch 23500, giga_loss[loss=0.343, simple_loss=0.4002, pruned_loss=0.1429, over 28656.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.35, pruned_loss=0.1028, over 5693883.26 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3549, pruned_loss=0.1046, over 5756373.41 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3482, pruned_loss=0.1012, over 5687503.50 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:35:32,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1072406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:35:50,414 INFO [optim.py:369] (1/2) 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 12:36:11,711 INFO [train.py:968] (1/2) Epoch 24, batch 23550, giga_loss[loss=0.2868, simple_loss=0.3616, pruned_loss=0.1061, over 28603.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.355, pruned_loss=0.1069, over 5680872.97 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3552, pruned_loss=0.1051, over 5747013.69 frames. ], giga_tot_loss[loss=0.2817, simple_loss=0.3531, pruned_loss=0.1051, over 5681395.96 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:36:27,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 12:36:55,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1072488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:36:57,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1072491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:37:01,439 INFO [train.py:968] (1/2) Epoch 24, batch 23600, libri_loss[loss=0.2946, simple_loss=0.3615, pruned_loss=0.1138, over 29543.00 frames. ], tot_loss[loss=0.29, simple_loss=0.36, pruned_loss=0.1101, over 5693779.49 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3551, pruned_loss=0.1052, over 5751945.25 frames. ], giga_tot_loss[loss=0.2879, simple_loss=0.3587, pruned_loss=0.1086, over 5687773.79 frames. ], batch size: 89, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:37:23,829 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1072520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:37:28,954 INFO [optim.py:369] (1/2) 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 12:37:47,831 INFO [train.py:968] (1/2) Epoch 24, batch 23650, giga_loss[loss=0.3347, simple_loss=0.3972, pruned_loss=0.1361, over 28855.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3663, pruned_loss=0.1151, over 5679183.12 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.356, pruned_loss=0.106, over 5744719.58 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3648, pruned_loss=0.1135, over 5678519.55 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:37:48,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6748, 1.5370, 1.8224, 1.4138], device='cuda:1'), covar=tensor([0.1611, 0.2453, 0.1335, 0.1666], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0704, 0.0959, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:37:48,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1072549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:37:52,909 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 24, batch 23700, giga_loss[loss=0.3331, simple_loss=0.4001, pruned_loss=0.133, over 29000.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3729, pruned_loss=0.1216, over 5669275.42 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3555, pruned_loss=0.106, over 5741996.99 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3726, pruned_loss=0.1208, over 5667960.03 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:39:07,826 INFO [optim.py:369] (1/2) 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 12:39:28,536 INFO [train.py:968] (1/2) Epoch 24, batch 23750, giga_loss[loss=0.316, simple_loss=0.3883, pruned_loss=0.1219, over 29010.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3788, pruned_loss=0.1262, over 5663717.27 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3557, pruned_loss=0.1062, over 5743199.13 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3787, pruned_loss=0.1256, over 5660624.67 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:40:11,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-12 12:40:16,060 INFO [train.py:968] (1/2) Epoch 24, batch 23800, giga_loss[loss=0.2999, simple_loss=0.3643, pruned_loss=0.1177, over 28895.00 frames. ], tot_loss[loss=0.3184, simple_loss=0.3805, pruned_loss=0.1282, over 5662434.37 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3559, pruned_loss=0.1067, over 5738778.56 frames. ], giga_tot_loss[loss=0.3183, simple_loss=0.3809, pruned_loss=0.1278, over 5661974.36 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:40:21,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3604, 3.1891, 3.0396, 1.5328], device='cuda:1'), covar=tensor([0.0932, 0.1067, 0.0981, 0.2164], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.1156, 0.0977, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 12:40:39,296 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:41:07,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.71 vs. limit=2.0
+2023-03-12 12:41:07,268 INFO [train.py:968] (1/2) Epoch 24, batch 23850, giga_loss[loss=0.3035, simple_loss=0.3756, pruned_loss=0.1157, over 29104.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.382, pruned_loss=0.1302, over 5660373.22 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3557, pruned_loss=0.1066, over 5739119.99 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.383, pruned_loss=0.1304, over 5658323.71 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:41:11,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2888, 1.5092, 1.3983, 1.1948], device='cuda:1'), covar=tensor([0.2163, 0.2189, 0.1734, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.2011, 0.1950, 0.1871, 0.2017], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 12:41:24,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5256, 1.8546, 1.2705, 0.8985], device='cuda:1'), covar=tensor([0.4310, 0.2844, 0.2247, 0.4692], device='cuda:1'), in_proj_covar=tensor([0.1776, 0.1677, 0.1623, 0.1446], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 12:42:02,230 INFO [train.py:968] (1/2) Epoch 24, batch 23900, giga_loss[loss=0.3387, simple_loss=0.3905, pruned_loss=0.1434, over 28919.00 frames. ], tot_loss[loss=0.3283, simple_loss=0.3865, pruned_loss=0.135, over 5642151.60 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3559, pruned_loss=0.1067, over 5738508.16 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3873, pruned_loss=0.1354, over 5640143.99 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:42:02,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4163, 1.5958, 1.6494, 1.2460], device='cuda:1'), covar=tensor([0.1600, 0.2365, 0.1322, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0911, 0.0704, 0.0957, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:42:34,666 INFO [optim.py:369] (1/2) 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 12:43:01,568 INFO [train.py:968] (1/2) Epoch 24, batch 23950, libri_loss[loss=0.276, simple_loss=0.3304, pruned_loss=0.1107, over 29475.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3911, pruned_loss=0.1394, over 5633043.34 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3561, pruned_loss=0.107, over 5741698.91 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3926, pruned_loss=0.1402, over 5626067.83 frames. ], batch size: 70, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:43:56,373 INFO [train.py:968] (1/2) Epoch 24, batch 24000, giga_loss[loss=0.3321, simple_loss=0.382, pruned_loss=0.1411, over 28532.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3912, pruned_loss=0.1408, over 5600947.36 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3562, pruned_loss=0.1074, over 5726876.91 frames. ], giga_tot_loss[loss=0.3388, simple_loss=0.3933, pruned_loss=0.1421, over 5604328.93 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:43:56,374 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 12:44:05,374 INFO [train.py:1012] (1/2) Epoch 24, validation: loss=0.2022, simple_loss=0.3102, pruned_loss=0.04706, over 944034.00 frames. 
+2023-03-12 12:44:05,374 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 12:44:08,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4309, 3.3819, 1.5113, 1.5855], device='cuda:1'), covar=tensor([0.1001, 0.0318, 0.0843, 0.1350], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0558, 0.0395, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 12:44:37,897 INFO [optim.py:369] (1/2) 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 12:44:56,137 INFO [train.py:968] (1/2) Epoch 24, batch 24050, giga_loss[loss=0.301, simple_loss=0.3675, pruned_loss=0.1172, over 28680.00 frames. ], tot_loss[loss=0.3344, simple_loss=0.3891, pruned_loss=0.1399, over 5618301.55 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.356, pruned_loss=0.1073, over 5726774.53 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3911, pruned_loss=0.1411, over 5620495.83 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:45:05,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3551, 1.6301, 1.5389, 1.4645], device='cuda:1'), covar=tensor([0.2187, 0.2074, 0.2557, 0.2216], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0755, 0.0724, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 12:45:44,123 INFO [train.py:968] (1/2) Epoch 24, batch 24100, giga_loss[loss=0.3779, simple_loss=0.4159, pruned_loss=0.17, over 26503.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3877, pruned_loss=0.1385, over 5624470.96 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3566, pruned_loss=0.1078, over 5725318.98 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3893, pruned_loss=0.1396, over 5625346.34 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:46:11,800 INFO [optim.py:369] (1/2) 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 12:46:21,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3980, 1.5049, 3.4615, 3.2642], device='cuda:1'), covar=tensor([0.1347, 0.2536, 0.0485, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0661, 0.0979, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 12:46:32,655 INFO [train.py:968] (1/2) Epoch 24, batch 24150, giga_loss[loss=0.3559, simple_loss=0.4048, pruned_loss=0.1536, over 27589.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3863, pruned_loss=0.1364, over 5616610.78 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3568, pruned_loss=0.1081, over 5728505.44 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3885, pruned_loss=0.138, over 5610883.05 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:47:03,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4996, 1.8565, 1.4402, 1.5612], device='cuda:1'), covar=tensor([0.0742, 0.0321, 0.0328, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 12:47:20,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1073096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:47:22,161 INFO [train.py:968] (1/2) Epoch 24, batch 24200, giga_loss[loss=0.4753, simple_loss=0.4825, pruned_loss=0.234, over 28237.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3887, pruned_loss=0.1376, over 5625498.92 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3566, pruned_loss=0.1081, over 5730564.58 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3911, pruned_loss=0.1394, over 5617326.36 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:47:40,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7190, 1.6365, 1.9365, 1.5116], device='cuda:1'), covar=tensor([0.1503, 0.2141, 0.1243, 0.1695], device='cuda:1'), in_proj_covar=tensor([0.0908, 0.0702, 0.0954, 0.0854], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 12:47:55,812 INFO [optim.py:369] (1/2) 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 12:48:17,072 INFO [train.py:968] (1/2) Epoch 24, batch 24250, giga_loss[loss=0.2628, simple_loss=0.3416, pruned_loss=0.09205, over 28950.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3883, pruned_loss=0.1371, over 5618644.62 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3573, pruned_loss=0.1088, over 5721778.09 frames. ], giga_tot_loss[loss=0.3335, simple_loss=0.3901, pruned_loss=0.1384, over 5617807.71 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:49:08,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4985, 1.6888, 1.2790, 1.2798], device='cuda:1'), covar=tensor([0.0970, 0.0524, 0.0990, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0451, 0.0522, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 12:49:09,210 INFO [train.py:968] (1/2) Epoch 24, batch 24300, giga_loss[loss=0.3163, simple_loss=0.386, pruned_loss=0.1233, over 28248.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3856, pruned_loss=0.134, over 5614438.11 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3578, pruned_loss=0.1091, over 5716552.94 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3874, pruned_loss=0.1354, over 5614949.97 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:49:22,986 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1073211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:49:40,288 INFO [optim.py:369] (1/2) 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 12:49:54,201 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1073239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:49:56,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1073242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:49:59,311 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1073244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:50:02,446 INFO [train.py:968] (1/2) Epoch 24, batch 24350, giga_loss[loss=0.2944, simple_loss=0.3694, pruned_loss=0.1097, over 28902.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3826, pruned_loss=0.1303, over 5625673.36 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3577, pruned_loss=0.1091, over 5716026.89 frames. ], giga_tot_loss[loss=0.324, simple_loss=0.3845, pruned_loss=0.1318, over 5625234.56 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:50:26,454 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1073271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:50:43,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4676, 3.5703, 1.6025, 1.5310], device='cuda:1'), covar=tensor([0.0989, 0.0392, 0.0896, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0560, 0.0395, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 12:50:51,261 INFO [train.py:968] (1/2) Epoch 24, batch 24400, giga_loss[loss=0.2827, simple_loss=0.3535, pruned_loss=0.1059, over 28888.00 frames. ], tot_loss[loss=0.3152, simple_loss=0.3777, pruned_loss=0.1263, over 5628691.97 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3575, pruned_loss=0.109, over 5720737.70 frames. ], giga_tot_loss[loss=0.3181, simple_loss=0.3799, pruned_loss=0.1281, over 5622150.96 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:51:22,532 INFO [optim.py:369] (1/2) 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 12:51:41,247 INFO [train.py:968] (1/2) Epoch 24, batch 24450, giga_loss[loss=0.296, simple_loss=0.3665, pruned_loss=0.1128, over 28661.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3756, pruned_loss=0.1247, over 5636604.67 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3574, pruned_loss=0.1089, over 5723979.75 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3777, pruned_loss=0.1264, over 5627537.27 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:52:30,567 INFO [train.py:968] (1/2) Epoch 24, batch 24500, libri_loss[loss=0.2995, simple_loss=0.3669, pruned_loss=0.116, over 29203.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3745, pruned_loss=0.1243, over 5631247.96 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.357, pruned_loss=0.1088, over 5719270.22 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3771, pruned_loss=0.1263, over 5625780.91 frames. ], batch size: 97, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:53:02,714 INFO [optim.py:369] (1/2) 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 12:53:25,194 INFO [train.py:968] (1/2) Epoch 24, batch 24550, giga_loss[loss=0.3306, simple_loss=0.3902, pruned_loss=0.1356, over 28593.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3747, pruned_loss=0.1243, over 5627583.92 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3573, pruned_loss=0.1091, over 5714014.77 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3769, pruned_loss=0.1259, over 5625777.65 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 12:53:26,389 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-12 12:54:15,996 INFO [train.py:968] (1/2) Epoch 24, batch 24600, giga_loss[loss=0.2975, simple_loss=0.3666, pruned_loss=0.1142, over 28957.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3731, pruned_loss=0.1227, over 5648573.44 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3574, pruned_loss=0.1092, over 5717001.72 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.375, pruned_loss=0.1241, over 5643472.70 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:54:48,237 INFO [optim.py:369] (1/2) 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 12:54:52,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4849, 1.6589, 1.2146, 1.2035], device='cuda:1'), covar=tensor([0.0964, 0.0538, 0.1005, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0452, 0.0521, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 12:55:08,746 INFO [train.py:968] (1/2) Epoch 24, batch 24650, giga_loss[loss=0.3419, simple_loss=0.4045, pruned_loss=0.1396, over 28651.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3734, pruned_loss=0.1208, over 5654513.46 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3573, pruned_loss=0.1092, over 5721322.53 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3755, pruned_loss=0.1222, over 5644670.09 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:55:45,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1073586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:55:56,986 INFO [train.py:968] (1/2) Epoch 24, batch 24700, giga_loss[loss=0.3091, simple_loss=0.3682, pruned_loss=0.125, over 28382.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3739, pruned_loss=0.1195, over 5658688.10 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3568, pruned_loss=0.1091, over 5716668.64 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3766, pruned_loss=0.1211, over 5653689.89 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:56:20,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1073619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:56:29,190 INFO [optim.py:369] (1/2) 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 12:56:47,929 INFO [train.py:968] (1/2) Epoch 24, batch 24750, giga_loss[loss=0.3851, simple_loss=0.4251, pruned_loss=0.1726, over 27625.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3744, pruned_loss=0.1205, over 5657820.72 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3566, pruned_loss=0.109, over 5719549.40 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3771, pruned_loss=0.1221, over 5650076.40 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 12:57:34,227 INFO [train.py:968] (1/2) Epoch 24, batch 24800, giga_loss[loss=0.3069, simple_loss=0.3775, pruned_loss=0.1181, over 28591.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3748, pruned_loss=0.1207, over 5675152.61 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3566, pruned_loss=0.1091, over 5721641.56 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3771, pruned_loss=0.1221, over 5666436.84 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:57:59,368 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 12:58:05,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1073729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:06,467 INFO [optim.py:369] (1/2) 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 12:58:08,434 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1073732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:22,488 INFO [train.py:968] (1/2) Epoch 24, batch 24850, libri_loss[loss=0.3411, simple_loss=0.3939, pruned_loss=0.1442, over 19233.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.372, pruned_loss=0.1194, over 5668568.47 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3567, pruned_loss=0.1092, over 5707194.80 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3743, pruned_loss=0.1207, over 5674821.02 frames. ], batch size: 187, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:58:34,605 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1073761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:37,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1073762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:58:39,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1073765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:59:06,578 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1073794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 12:59:09,076 INFO [train.py:968] (1/2) Epoch 24, batch 24900, libri_loss[loss=0.3217, simple_loss=0.3822, pruned_loss=0.1306, over 29545.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3699, pruned_loss=0.1192, over 5668813.66 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3569, pruned_loss=0.1093, over 5709588.43 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3719, pruned_loss=0.1204, over 5670756.27 frames. ], batch size: 89, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 12:59:39,980 INFO [optim.py:369] (1/2) 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 12:59:56,791 INFO [train.py:968] (1/2) Epoch 24, batch 24950, giga_loss[loss=0.3051, simple_loss=0.3965, pruned_loss=0.1069, over 29049.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3688, pruned_loss=0.118, over 5659693.25 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3567, pruned_loss=0.1092, over 5699514.19 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3707, pruned_loss=0.1191, over 5669192.44 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:00:43,101 INFO [train.py:968] (1/2) Epoch 24, batch 25000, giga_loss[loss=0.2852, simple_loss=0.3551, pruned_loss=0.1077, over 28254.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3693, pruned_loss=0.1165, over 5673892.03 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3569, pruned_loss=0.1093, over 5701910.30 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3707, pruned_loss=0.1174, over 5678839.50 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:00:46,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3500, 1.5764, 1.2236, 1.1779], device='cuda:1'), covar=tensor([0.1033, 0.0565, 0.1072, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0449, 0.0519, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 13:01:14,916 INFO [optim.py:369] (1/2) 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 13:01:27,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4734, 2.1930, 1.6512, 0.7587], device='cuda:1'), covar=tensor([0.5847, 0.3044, 0.4284, 0.6459], device='cuda:1'), in_proj_covar=tensor([0.1775, 0.1675, 0.1619, 0.1439], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 13:01:32,376 INFO [train.py:968] (1/2) Epoch 24, batch 25050, giga_loss[loss=0.2557, simple_loss=0.3343, pruned_loss=0.08854, over 28535.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3702, pruned_loss=0.1174, over 5672601.86 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3576, pruned_loss=0.1098, over 5706484.43 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.371, pruned_loss=0.1179, over 5672009.25 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:01:50,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-12 13:02:22,617 INFO [train.py:968] (1/2) Epoch 24, batch 25100, giga_loss[loss=0.2841, simple_loss=0.3531, pruned_loss=0.1075, over 28874.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3681, pruned_loss=0.1163, over 5678396.96 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3575, pruned_loss=0.1098, over 5709127.03 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3691, pruned_loss=0.1168, over 5674848.98 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:02:53,949 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 25150, giga_loss[loss=0.342, simple_loss=0.3773, pruned_loss=0.1534, over 23781.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3678, pruned_loss=0.1171, over 5665530.13 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3575, pruned_loss=0.1099, over 5703059.48 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3688, pruned_loss=0.1176, over 5667231.12 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:03:22,098 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-12 13:03:31,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6722, 2.6472, 2.4558, 2.6224], device='cuda:1'), covar=tensor([0.1870, 0.2287, 0.2158, 0.2071], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0758, 0.0726, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 13:04:05,555 INFO [train.py:968] (1/2) Epoch 24, batch 25200, giga_loss[loss=0.3329, simple_loss=0.3923, pruned_loss=0.1368, over 28916.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3672, pruned_loss=0.1176, over 5659757.43 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3574, pruned_loss=0.1099, over 5701282.22 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3682, pruned_loss=0.1181, over 5662320.26 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:04:40,399 INFO [optim.py:369] (1/2) 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 13:04:44,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2058, 3.2975, 1.4318, 1.3398], device='cuda:1'), covar=tensor([0.1096, 0.0454, 0.0923, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0562, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 13:04:55,238 INFO [train.py:968] (1/2) Epoch 24, batch 25250, giga_loss[loss=0.347, simple_loss=0.399, pruned_loss=0.1475, over 28340.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3671, pruned_loss=0.1184, over 5667571.85 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3573, pruned_loss=0.1098, over 5703538.05 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3682, pruned_loss=0.119, over 5666721.07 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:05:16,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 13:05:46,770 INFO [train.py:968] (1/2) Epoch 24, batch 25300, giga_loss[loss=0.2677, simple_loss=0.3408, pruned_loss=0.0973, over 28879.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3646, pruned_loss=0.1169, over 5666186.23 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3576, pruned_loss=0.1099, over 5702775.31 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3653, pruned_loss=0.1174, over 5665947.37 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:06:08,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6033, 1.7893, 1.4650, 1.6603], device='cuda:1'), covar=tensor([0.2532, 0.2583, 0.2794, 0.2474], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1116, 0.1364, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 13:06:19,486 INFO [optim.py:369] (1/2) 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 13:06:37,511 INFO [train.py:968] (1/2) Epoch 24, batch 25350, giga_loss[loss=0.299, simple_loss=0.3624, pruned_loss=0.1178, over 28310.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.365, pruned_loss=0.1179, over 5655805.05 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.358, pruned_loss=0.1103, over 5696404.87 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3653, pruned_loss=0.118, over 5661248.64 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:07:24,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4487, 4.0956, 1.6176, 1.6121], device='cuda:1'), covar=tensor([0.1008, 0.0396, 0.0917, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0562, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 13:07:24,995 INFO [train.py:968] (1/2) Epoch 24, batch 25400, giga_loss[loss=0.3053, simple_loss=0.3772, pruned_loss=0.1167, over 28699.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3655, pruned_loss=0.1179, over 5651409.01 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3584, pruned_loss=0.1106, over 5690518.79 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3655, pruned_loss=0.1179, over 5660701.02 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:07:37,844 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1074306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:07:59,706 INFO [optim.py:369] (1/2) 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 13:08:13,638 INFO [train.py:968] (1/2) Epoch 24, batch 25450, giga_loss[loss=0.3017, simple_loss=0.371, pruned_loss=0.1162, over 28740.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.1181, over 5657862.50 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3582, pruned_loss=0.1105, over 5694557.71 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3672, pruned_loss=0.1183, over 5660732.19 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:08:19,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8547, 2.9274, 1.9862, 1.0221], device='cuda:1'), covar=tensor([0.9000, 0.4011, 0.3886, 0.7936], device='cuda:1'), in_proj_covar=tensor([0.1787, 0.1683, 0.1623, 0.1444], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 13:08:35,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2379, 4.0831, 3.9075, 2.0846], device='cuda:1'), covar=tensor([0.0581, 0.0708, 0.0760, 0.2319], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1177, 0.0993, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 13:09:01,991 INFO [train.py:968] (1/2) Epoch 24, batch 25500, giga_loss[loss=0.2888, simple_loss=0.3599, pruned_loss=0.1089, over 28724.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3671, pruned_loss=0.1177, over 5654716.02 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3581, pruned_loss=0.1104, over 5694117.65 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3675, pruned_loss=0.118, over 5656930.17 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:09:04,757 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1074400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:09:08,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-12 13:09:35,929 INFO [optim.py:369] (1/2) 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 13:09:47,761 INFO [train.py:968] (1/2) Epoch 24, batch 25550, giga_loss[loss=0.3108, simple_loss=0.3747, pruned_loss=0.1235, over 28924.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3675, pruned_loss=0.1179, over 5658205.46 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3584, pruned_loss=0.1107, over 5694888.54 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3679, pruned_loss=0.1181, over 5658079.49 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:10:35,435 INFO [train.py:968] (1/2) Epoch 24, batch 25600, giga_loss[loss=0.4151, simple_loss=0.4465, pruned_loss=0.1919, over 28012.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1194, over 5659350.08 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3581, pruned_loss=0.1105, over 5699808.06 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3696, pruned_loss=0.12, over 5653604.19 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:10:49,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9863, 1.3380, 1.1009, 0.2128], device='cuda:1'), covar=tensor([0.4413, 0.3526, 0.4524, 0.6671], device='cuda:1'), in_proj_covar=tensor([0.1785, 0.1685, 0.1621, 0.1444], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 13:11:10,518 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 13:11:13,020 INFO [optim.py:369] (1/2) 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 13:11:26,613 INFO [train.py:968] (1/2) Epoch 24, batch 25650, giga_loss[loss=0.3173, simple_loss=0.3737, pruned_loss=0.1304, over 28019.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3703, pruned_loss=0.1214, over 5662835.50 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3582, pruned_loss=0.1106, over 5703201.55 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3712, pruned_loss=0.1221, over 5654298.19 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:12:16,956 INFO [train.py:968] (1/2) Epoch 24, batch 25700, giga_loss[loss=0.3056, simple_loss=0.3693, pruned_loss=0.1209, over 28657.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3709, pruned_loss=0.1227, over 5669208.03 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3581, pruned_loss=0.1105, over 5707414.14 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.372, pruned_loss=0.1235, over 5657467.06 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:12:55,406 INFO [optim.py:369] (1/2) 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 13:13:09,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3558, 1.7863, 1.4218, 1.6398], device='cuda:1'), covar=tensor([0.0770, 0.0314, 0.0333, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0224], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 13:13:11,534 INFO [train.py:968] (1/2) Epoch 24, batch 25750, giga_loss[loss=0.432, simple_loss=0.4524, pruned_loss=0.2058, over 27464.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3729, pruned_loss=0.1249, over 5655242.17 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3584, pruned_loss=0.1108, over 5711363.03 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3739, pruned_loss=0.1258, over 5641259.66 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:13:37,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1074681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:13:41,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1074686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:13:52,775 INFO [train.py:968] (1/2) Epoch 24, batch 25800, giga_loss[loss=0.2952, simple_loss=0.3655, pruned_loss=0.1125, over 28963.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3726, pruned_loss=0.1246, over 5653195.10 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3587, pruned_loss=0.1111, over 5700685.10 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3736, pruned_loss=0.1256, over 5648499.89 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:14:25,952 INFO [optim.py:369] (1/2) 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 13:14:41,637 INFO [train.py:968] (1/2) Epoch 24, batch 25850, giga_loss[loss=0.2732, simple_loss=0.3489, pruned_loss=0.09875, over 28923.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1239, over 5657373.38 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3588, pruned_loss=0.111, over 5707045.50 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3731, pruned_loss=0.125, over 5646841.94 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:15:03,994 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 25900, giga_loss[loss=0.3103, simple_loss=0.3662, pruned_loss=0.1272, over 28332.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3708, pruned_loss=0.121, over 5663956.81 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3591, pruned_loss=0.1114, over 5697318.64 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3715, pruned_loss=0.1217, over 5663195.09 frames. ], batch size: 369, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:15:52,852 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1074824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:15:56,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1074827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:16:03,400 INFO [optim.py:369] (1/2) 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 13:16:15,583 INFO [train.py:968] (1/2) Epoch 24, batch 25950, giga_loss[loss=0.317, simple_loss=0.3749, pruned_loss=0.1295, over 28888.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3688, pruned_loss=0.1197, over 5659792.97 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3595, pruned_loss=0.1117, over 5699811.14 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3692, pruned_loss=0.1201, over 5656221.52 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:16:23,497 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1074856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:16:56,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4170, 1.5877, 1.3733, 1.5885], device='cuda:1'), covar=tensor([0.0750, 0.0342, 0.0334, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 13:17:00,907 INFO [train.py:968] (1/2) Epoch 24, batch 26000, giga_loss[loss=0.3102, simple_loss=0.3729, pruned_loss=0.1237, over 28968.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3673, pruned_loss=0.1192, over 5672874.43 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3596, pruned_loss=0.1118, over 5707422.27 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1197, over 5661527.26 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:17:19,973 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1074918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:17:22,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1074921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:17:31,571 INFO [optim.py:369] (1/2) 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 13:17:46,186 INFO [train.py:968] (1/2) Epoch 24, batch 26050, giga_loss[loss=0.287, simple_loss=0.3552, pruned_loss=0.1093, over 28949.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3651, pruned_loss=0.1178, over 5669664.53 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3599, pruned_loss=0.1121, over 5694768.99 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3656, pruned_loss=0.1183, over 5669314.21 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:17:50,102 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1074950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:18:38,044 INFO [train.py:968] (1/2) Epoch 24, batch 26100, giga_loss[loss=0.3547, simple_loss=0.4066, pruned_loss=0.1514, over 28668.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 5674929.89 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3603, pruned_loss=0.1124, over 5698723.96 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.366, pruned_loss=0.1188, over 5670472.73 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:19:10,489 INFO [optim.py:369] (1/2) 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 13:19:12,005 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-12 13:19:22,913 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-12 13:19:25,490 INFO [train.py:968] (1/2) Epoch 24, batch 26150, giga_loss[loss=0.3797, simple_loss=0.4314, pruned_loss=0.164, over 27572.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3686, pruned_loss=0.1194, over 5680908.26 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3601, pruned_loss=0.1122, over 5703691.05 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.12, over 5672544.24 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:19:27,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6577, 1.8477, 1.4850, 2.0718], device='cuda:1'), covar=tensor([0.2743, 0.2951, 0.3314, 0.2529], device='cuda:1'), in_proj_covar=tensor([0.1544, 0.1112, 0.1358, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 13:19:33,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4549, 4.2957, 4.0057, 2.0403], device='cuda:1'), covar=tensor([0.0718, 0.0988, 0.1070, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1178, 0.0993, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 13:19:37,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:19:37,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4368, 1.7832, 1.1578, 1.3686], device='cuda:1'), covar=tensor([0.1121, 0.0619, 0.1185, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0451, 0.0521, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 13:20:13,562 INFO [train.py:968] (1/2) Epoch 24, batch 26200, giga_loss[loss=0.3007, simple_loss=0.3774, pruned_loss=0.112, over 28864.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3709, pruned_loss=0.1176, over 5674133.60 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3608, pruned_loss=0.1128, over 5696646.90 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3708, pruned_loss=0.1178, over 5673941.94 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:20:17,458 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 13:20:48,834 INFO [optim.py:369] (1/2) 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 13:21:03,698 INFO [train.py:968] (1/2) Epoch 24, batch 26250, giga_loss[loss=0.2712, simple_loss=0.3467, pruned_loss=0.09787, over 28442.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3722, pruned_loss=0.1184, over 5670782.90 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3607, pruned_loss=0.1128, over 5690820.54 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3725, pruned_loss=0.1186, over 5675429.38 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:21:21,089 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:21:51,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.54 vs. limit=5.0
+2023-03-12 13:21:53,782 INFO [train.py:968] (1/2) Epoch 24, batch 26300, giga_loss[loss=0.3191, simple_loss=0.3795, pruned_loss=0.1294, over 28901.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.375, pruned_loss=0.1213, over 5674234.32 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3607, pruned_loss=0.1127, over 5695146.58 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3755, pruned_loss=0.1216, over 5673545.37 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:21:59,997 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:22:02,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:22:22,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4801, 1.5992, 1.6746, 1.2829], device='cuda:1'), covar=tensor([0.1847, 0.2676, 0.1536, 0.1833], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0709, 0.0962, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 13:22:23,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6617, 1.7952, 1.7193, 1.5394], device='cuda:1'), covar=tensor([0.3148, 0.2689, 0.2394, 0.2720], device='cuda:1'), in_proj_covar=tensor([0.2018, 0.1959, 0.1875, 0.2021], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 13:22:24,457 INFO [optim.py:369] (1/2) 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 13:22:27,130 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1075236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:22:39,309 INFO [train.py:968] (1/2) Epoch 24, batch 26350, giga_loss[loss=0.3365, simple_loss=0.3943, pruned_loss=0.1393, over 27499.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3753, pruned_loss=0.1221, over 5672121.04 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3611, pruned_loss=0.1131, over 5694310.49 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3757, pruned_loss=0.1222, over 5672092.51 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:23:14,001 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:23:32,203 INFO [train.py:968] (1/2) Epoch 24, batch 26400, giga_loss[loss=0.4305, simple_loss=0.4449, pruned_loss=0.208, over 26494.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3748, pruned_loss=0.1227, over 5675341.36 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3611, pruned_loss=0.1131, over 5695517.93 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3752, pruned_loss=0.1229, over 5674287.76 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:24:06,677 INFO [optim.py:369] (1/2) 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 13:24:19,784 INFO [train.py:968] (1/2) Epoch 24, batch 26450, libri_loss[loss=0.411, simple_loss=0.442, pruned_loss=0.1901, over 19014.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3727, pruned_loss=0.1218, over 5676214.65 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3613, pruned_loss=0.1134, over 5691180.26 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3731, pruned_loss=0.1219, over 5679881.77 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:24:54,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8858, 3.7130, 3.5285, 1.6606], device='cuda:1'), covar=tensor([0.0693, 0.0825, 0.0829, 0.2178], device='cuda:1'), in_proj_covar=tensor([0.1269, 0.1177, 0.0993, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 13:25:05,570 INFO [train.py:968] (1/2) Epoch 24, batch 26500, giga_loss[loss=0.3488, simple_loss=0.3948, pruned_loss=0.1514, over 28736.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3718, pruned_loss=0.1221, over 5684096.36 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3613, pruned_loss=0.1132, over 5697202.20 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3725, pruned_loss=0.1225, over 5681372.02 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:25:20,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3125, 1.3315, 3.7443, 3.2130], device='cuda:1'), covar=tensor([0.1646, 0.2775, 0.0489, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0667, 0.0985, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 13:25:39,947 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:25:42,271 INFO [optim.py:369] (1/2) 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 13:25:56,948 INFO [train.py:968] (1/2) Epoch 24, batch 26550, giga_loss[loss=0.2949, simple_loss=0.3625, pruned_loss=0.1137, over 28924.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3699, pruned_loss=0.121, over 5682295.29 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3612, pruned_loss=0.113, over 5700018.68 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3707, pruned_loss=0.1217, over 5677678.68 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:26:44,668 INFO [train.py:968] (1/2) Epoch 24, batch 26600, giga_loss[loss=0.3138, simple_loss=0.3785, pruned_loss=0.1245, over 28334.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3705, pruned_loss=0.1215, over 5673575.20 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3611, pruned_loss=0.113, over 5693472.66 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3712, pruned_loss=0.1221, over 5675945.59 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:27:18,998 INFO [optim.py:369] (1/2) 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 13:27:23,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:27:26,593 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 26650, giga_loss[loss=0.314, simple_loss=0.3745, pruned_loss=0.1267, over 28982.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3694, pruned_loss=0.1219, over 5660991.34 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3612, pruned_loss=0.113, over 5692037.70 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3701, pruned_loss=0.1226, over 5663689.86 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:28:19,323 INFO [train.py:968] (1/2) Epoch 24, batch 26700, giga_loss[loss=0.3312, simple_loss=0.3939, pruned_loss=0.1342, over 28879.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.367, pruned_loss=0.1207, over 5660007.60 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3605, pruned_loss=0.1126, over 5697256.59 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3684, pruned_loss=0.1218, over 5656522.55 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:28:46,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4749, 1.7096, 1.4175, 1.6923], device='cuda:1'), covar=tensor([0.0772, 0.0318, 0.0328, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0120, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 13:28:54,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3902, 1.6024, 1.4225, 1.5244], device='cuda:1'), covar=tensor([0.0810, 0.0347, 0.0343, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0120, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 13:28:54,517 INFO [optim.py:369] (1/2) 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 13:29:05,976 INFO [train.py:968] (1/2) Epoch 24, batch 26750, giga_loss[loss=0.2782, simple_loss=0.3533, pruned_loss=0.1015, over 28829.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.367, pruned_loss=0.12, over 5664607.88 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3603, pruned_loss=0.1126, over 5701342.10 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3684, pruned_loss=0.121, over 5657427.57 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:29:14,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:29:40,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:29:44,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:29:53,977 INFO [train.py:968] (1/2) Epoch 24, batch 26800, giga_loss[loss=0.3298, simple_loss=0.3882, pruned_loss=0.1357, over 28654.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3698, pruned_loss=0.1212, over 5660965.03 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3605, pruned_loss=0.1128, over 5694992.79 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3709, pruned_loss=0.122, over 5659555.88 frames. ], batch size: 92, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:30:11,835 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:30:14,379 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1075715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:30:33,849 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 26850, libri_loss[loss=0.2484, simple_loss=0.3215, pruned_loss=0.08767, over 29582.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3707, pruned_loss=0.1224, over 5653509.78 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3603, pruned_loss=0.1128, over 5690606.72 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.372, pruned_loss=0.1234, over 5655279.26 frames. ], batch size: 74, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:30:47,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1075751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:30:52,800 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.56 vs. limit=5.0
+2023-03-12 13:31:31,111 INFO [train.py:968] (1/2) Epoch 24, batch 26900, giga_loss[loss=0.343, simple_loss=0.3807, pruned_loss=0.1527, over 23616.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3729, pruned_loss=0.1218, over 5666531.93 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3604, pruned_loss=0.1128, over 5691795.68 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3739, pruned_loss=0.1225, over 5666812.01 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:31:31,396 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:31:33,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:31:39,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:32:00,628 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1075830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:32:07,067 INFO [optim.py:369] (1/2) 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 13:32:08,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4588, 1.6090, 1.5938, 1.4058], device='cuda:1'), covar=tensor([0.2960, 0.2357, 0.2176, 0.2513], device='cuda:1'), in_proj_covar=tensor([0.2011, 0.1947, 0.1865, 0.2012], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 13:32:17,757 INFO [train.py:968] (1/2) Epoch 24, batch 26950, giga_loss[loss=0.2848, simple_loss=0.3723, pruned_loss=0.09862, over 29064.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3734, pruned_loss=0.1196, over 5667403.14 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3604, pruned_loss=0.1127, over 5682571.92 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3747, pruned_loss=0.1205, over 5675082.81 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:33:02,647 INFO [train.py:968] (1/2) Epoch 24, batch 27000, giga_loss[loss=0.2781, simple_loss=0.3582, pruned_loss=0.09903, over 28814.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3757, pruned_loss=0.1202, over 5673614.82 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3604, pruned_loss=0.1129, over 5684832.15 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3771, pruned_loss=0.1211, over 5677188.06 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:33:02,647 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 13:33:11,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3570, 3.0188, 1.4286, 1.5056], device='cuda:1'), covar=tensor([0.1064, 0.0349, 0.1004, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0563, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 13:33:11,742 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 13:33:12,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4569, 1.8309, 1.8108, 1.5523], device='cuda:1'), covar=tensor([0.1815, 0.1463, 0.1948, 0.1698], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0753, 0.0723, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 13:33:26,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1075916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:33:46,786 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 27050, giga_loss[loss=0.3272, simple_loss=0.3965, pruned_loss=0.129, over 28524.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3789, pruned_loss=0.123, over 5675066.30 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3601, pruned_loss=0.1127, over 5687197.67 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3805, pruned_loss=0.1239, over 5675765.20 frames. ], batch size: 60, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:34:03,099 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1075950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:34:05,560 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1075953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:34:32,888 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 27100, giga_loss[loss=0.3675, simple_loss=0.4238, pruned_loss=0.1556, over 28682.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3813, pruned_loss=0.1267, over 5665985.48 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3598, pruned_loss=0.1126, over 5689955.69 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3834, pruned_loss=0.1278, over 5663589.70 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:35:30,523 INFO [optim.py:369] (1/2) 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 13:35:40,661 INFO [train.py:968] (1/2) Epoch 24, batch 27150, giga_loss[loss=0.2803, simple_loss=0.3492, pruned_loss=0.1057, over 28908.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3808, pruned_loss=0.1276, over 5648962.27 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.113, over 5686712.72 frames. ], giga_tot_loss[loss=0.3198, simple_loss=0.3826, pruned_loss=0.1285, over 5650005.37 frames. ], batch size: 99, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:35:43,420 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:35:51,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076059.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:35:54,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076062.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:36:04,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:36:17,799 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076085.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:36:20,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:36:23,213 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 27200, giga_loss[loss=0.3689, simple_loss=0.3994, pruned_loss=0.1692, over 23535.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3797, pruned_loss=0.1271, over 5650657.36 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1126, over 5690619.50 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3826, pruned_loss=0.1287, over 5647257.90 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:36:40,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:36:55,494 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:37:05,987 INFO [optim.py:369] (1/2) 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 13:37:15,978 INFO [train.py:968] (1/2) Epoch 24, batch 27250, giga_loss[loss=0.3092, simple_loss=0.3773, pruned_loss=0.1205, over 28694.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3779, pruned_loss=0.1244, over 5650809.25 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5687421.80 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3807, pruned_loss=0.1259, over 5649719.79 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:37:54,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5745, 1.5380, 1.7423, 1.3508], device='cuda:1'), covar=tensor([0.1868, 0.2652, 0.1582, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0709, 0.0961, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 13:37:59,504 INFO [train.py:968] (1/2) Epoch 24, batch 27300, giga_loss[loss=0.2713, simple_loss=0.3577, pruned_loss=0.0925, over 28964.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3769, pruned_loss=0.1224, over 5654706.54 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3587, pruned_loss=0.1124, over 5684694.65 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3804, pruned_loss=0.1243, over 5654502.90 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:38:32,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:38:34,115 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:38:40,249 INFO [optim.py:369] (1/2) 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 13:38:47,885 INFO [train.py:968] (1/2) Epoch 24, batch 27350, giga_loss[loss=0.3366, simple_loss=0.4109, pruned_loss=0.1311, over 28800.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3775, pruned_loss=0.1226, over 5662679.31 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1125, over 5691114.31 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3808, pruned_loss=0.1243, over 5655410.56 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:38:58,015 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076258.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:39:03,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:39:10,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:39:14,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 27400, giga_loss[loss=0.2987, simple_loss=0.3733, pruned_loss=0.1121, over 28856.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3776, pruned_loss=0.123, over 5671835.67 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1127, over 5698684.87 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3809, pruned_loss=0.1246, over 5658053.90 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:39:40,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:39:57,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4346, 1.5771, 1.1757, 1.1443], device='cuda:1'), covar=tensor([0.0797, 0.0390, 0.0873, 0.1093], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0450, 0.0519, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 13:40:12,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-12 13:40:13,768 INFO [optim.py:369] (1/2) 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 13:40:26,294 INFO [train.py:968] (1/2) Epoch 24, batch 27450, giga_loss[loss=0.3054, simple_loss=0.3809, pruned_loss=0.1149, over 28913.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3771, pruned_loss=0.1227, over 5678375.93 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.359, pruned_loss=0.1126, over 5699918.01 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3798, pruned_loss=0.1242, over 5666219.64 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:40:52,052 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-12 13:41:19,945 INFO [train.py:968] (1/2) Epoch 24, batch 27500, giga_loss[loss=0.2772, simple_loss=0.3509, pruned_loss=0.1017, over 28918.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3756, pruned_loss=0.1239, over 5662513.29 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1127, over 5707166.36 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3784, pruned_loss=0.1254, over 5645011.60 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:41:46,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:41:58,777 INFO [optim.py:369] (1/2) 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 13:42:00,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1076440.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:42:11,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:42:11,611 INFO [train.py:968] (1/2) Epoch 24, batch 27550, giga_loss[loss=0.3502, simple_loss=0.399, pruned_loss=0.1507, over 27554.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3737, pruned_loss=0.1232, over 5655185.24 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3591, pruned_loss=0.1126, over 5711401.60 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3762, pruned_loss=0.1247, over 5636381.06 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:42:22,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6769, 1.8377, 1.4960, 1.8512], device='cuda:1'), covar=tensor([0.2650, 0.2822, 0.3136, 0.2649], device='cuda:1'), in_proj_covar=tensor([0.1546, 0.1115, 0.1363, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 13:42:22,994 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076460.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:42:48,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:43:01,564 INFO [train.py:968] (1/2) Epoch 24, batch 27600, giga_loss[loss=0.2681, simple_loss=0.3365, pruned_loss=0.09982, over 28455.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3724, pruned_loss=0.1229, over 5663207.58 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1128, over 5712508.52 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3744, pruned_loss=0.1241, over 5646451.13 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:43:40,276 INFO [optim.py:369] (1/2) 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 13:43:49,103 INFO [train.py:968] (1/2) Epoch 24, batch 27650, giga_loss[loss=0.2854, simple_loss=0.3536, pruned_loss=0.1086, over 28909.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3721, pruned_loss=0.1231, over 5660584.13 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3598, pruned_loss=0.1129, over 5713952.92 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3738, pruned_loss=0.1243, over 5644484.44 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:44:00,490 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-12 13:44:05,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5069, 1.6835, 1.4166, 1.4543], device='cuda:1'), covar=tensor([0.2856, 0.2719, 0.2943, 0.2421], device='cuda:1'), in_proj_covar=tensor([0.1550, 0.1119, 0.1365, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 13:44:12,120 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:14,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:31,608 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:37,249 INFO [train.py:968] (1/2) Epoch 24, batch 27700, giga_loss[loss=0.296, simple_loss=0.3627, pruned_loss=0.1146, over 28697.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3707, pruned_loss=0.1217, over 5662359.71 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3604, pruned_loss=0.1133, over 5717388.44 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3716, pruned_loss=0.1224, over 5645164.00 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:44:37,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8193, 4.6913, 4.4359, 2.1126], device='cuda:1'), covar=tensor([0.0496, 0.0594, 0.0679, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.1279, 0.1187, 0.1000, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 13:44:38,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:44:40,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076603.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:44:43,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076606.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:44:59,213 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:04,022 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:06,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:09,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076633.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:45:11,234 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076635.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:45:15,104 INFO [optim.py:369] (1/2) 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 13:45:24,615 INFO [train.py:968] (1/2) Epoch 24, batch 27750, giga_loss[loss=0.2639, simple_loss=0.3318, pruned_loss=0.09805, over 28586.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3673, pruned_loss=0.1179, over 5668798.30 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3605, pruned_loss=0.1135, over 5712018.73 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3682, pruned_loss=0.1184, over 5658179.22 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:45:34,523 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:45:35,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5404, 1.5520, 1.1734, 1.1563], device='cuda:1'), covar=tensor([0.0855, 0.0545, 0.0999, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0451, 0.0520, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 13:46:14,701 INFO [train.py:968] (1/2) Epoch 24, batch 27800, giga_loss[loss=0.3221, simple_loss=0.3652, pruned_loss=0.1395, over 23730.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1155, over 5664456.06 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3607, pruned_loss=0.1137, over 5714111.06 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3654, pruned_loss=0.1158, over 5653370.94 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:46:52,031 INFO [optim.py:369] (1/2) 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 13:46:56,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9489, 1.2250, 1.3308, 1.0297], device='cuda:1'), covar=tensor([0.1697, 0.1227, 0.2011, 0.1502], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0754, 0.0721, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 13:47:02,733 INFO [train.py:968] (1/2) Epoch 24, batch 27850, giga_loss[loss=0.2446, simple_loss=0.3221, pruned_loss=0.0835, over 28908.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3652, pruned_loss=0.1164, over 5661202.50 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1134, over 5721806.41 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1171, over 5643016.87 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:47:37,221 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076776.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:47:40,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076779.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:48:00,537 INFO [train.py:968] (1/2) Epoch 24, batch 27900, giga_loss[loss=0.2625, simple_loss=0.3362, pruned_loss=0.09443, over 28644.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3625, pruned_loss=0.1154, over 5676149.97 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3599, pruned_loss=0.1132, over 5723836.19 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3636, pruned_loss=0.1161, over 5659399.63 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:48:11,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076808.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:48:17,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1076815.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:48:42,713 INFO [optim.py:369] (1/2) 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 13:48:49,556 INFO [train.py:968] (1/2) Epoch 24, batch 27950, giga_loss[loss=0.2436, simple_loss=0.3181, pruned_loss=0.08449, over 28939.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3618, pruned_loss=0.1151, over 5667495.29 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5719478.88 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3627, pruned_loss=0.1159, over 5655898.72 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 13:49:39,891 INFO [train.py:968] (1/2) Epoch 24, batch 28000, libri_loss[loss=0.2993, simple_loss=0.3678, pruned_loss=0.1154, over 26109.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3652, pruned_loss=0.1168, over 5651445.11 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 5710640.68 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.366, pruned_loss=0.1175, over 5648564.38 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:49:46,638 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:50:29,152 INFO [train.py:968] (1/2) Epoch 24, batch 28050, giga_loss[loss=0.2905, simple_loss=0.3581, pruned_loss=0.1115, over 28998.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3675, pruned_loss=0.1184, over 5640810.73 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5705233.28 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.368, pruned_loss=0.1189, over 5642595.84 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:50:38,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1076958.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:50:40,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1076961.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:50:56,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3991, 1.4901, 1.1677, 1.0664], device='cuda:1'), covar=tensor([0.1056, 0.0586, 0.1102, 0.1163], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0451, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 13:51:07,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1076990.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 13:51:14,190 INFO [train.py:968] (1/2) Epoch 24, batch 28100, giga_loss[loss=0.3012, simple_loss=0.3699, pruned_loss=0.1162, over 28537.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3678, pruned_loss=0.1185, over 5653119.61 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5709189.83 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 5649396.29 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:51:34,008 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:51:34,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9572, 1.2189, 2.8476, 2.8111], device='cuda:1'), covar=tensor([0.1555, 0.2539, 0.0563, 0.1371], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0661, 0.0978, 0.0941], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 13:51:53,519 INFO [optim.py:369] (1/2) 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 13:52:00,141 INFO [train.py:968] (1/2) Epoch 24, batch 28150, giga_loss[loss=0.3359, simple_loss=0.3867, pruned_loss=0.1425, over 28593.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3681, pruned_loss=0.1192, over 5650923.01 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3606, pruned_loss=0.1133, over 5707713.38 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3685, pruned_loss=0.1197, over 5648363.79 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:52:47,516 INFO [train.py:968] (1/2) Epoch 24, batch 28200, giga_loss[loss=0.3501, simple_loss=0.4077, pruned_loss=0.1463, over 28223.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3706, pruned_loss=0.1211, over 5661684.12 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1136, over 5711705.44 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.371, pruned_loss=0.1214, over 5655036.34 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:53:29,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5424, 2.0803, 1.3115, 0.7744], device='cuda:1'), covar=tensor([0.6861, 0.3593, 0.3273, 0.7008], device='cuda:1'), in_proj_covar=tensor([0.1787, 0.1691, 0.1627, 0.1453], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 13:53:30,235 INFO [optim.py:369] (1/2) 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 13:53:39,310 INFO [train.py:968] (1/2) Epoch 24, batch 28250, giga_loss[loss=0.3066, simple_loss=0.3736, pruned_loss=0.1198, over 28584.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3722, pruned_loss=0.1218, over 5660212.31 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1137, over 5705001.46 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3725, pruned_loss=0.1221, over 5660283.59 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:53:47,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077157.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 13:54:00,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-12 13:54:03,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9627, 1.1459, 1.1329, 0.9402], device='cuda:1'), covar=tensor([0.2561, 0.3013, 0.1742, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.2012, 0.1959, 0.1872, 0.2013], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 13:54:29,842 INFO [train.py:968] (1/2) Epoch 24, batch 28300, giga_loss[loss=0.3511, simple_loss=0.3972, pruned_loss=0.1525, over 27680.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3732, pruned_loss=0.1227, over 5660101.48 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5710015.11 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3736, pruned_loss=0.123, over 5654061.67 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:54:52,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2859, 1.3451, 4.0186, 3.2289], device='cuda:1'), covar=tensor([0.1843, 0.2843, 0.0483, 0.0929], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0664, 0.0982, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 13:55:10,720 INFO [optim.py:369] (1/2) 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 13:55:20,006 INFO [train.py:968] (1/2) Epoch 24, batch 28350, giga_loss[loss=0.2831, simple_loss=0.3645, pruned_loss=0.1009, over 28811.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3732, pruned_loss=0.1232, over 5649281.75 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3612, pruned_loss=0.1137, over 5702231.14 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3737, pruned_loss=0.1236, over 5650120.84 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:55:31,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.9123, 4.7387, 4.5115, 2.9086], device='cuda:1'), covar=tensor([0.0621, 0.0836, 0.0938, 0.1578], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1186, 0.1002, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 13:55:54,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:56:13,441 INFO [train.py:968] (1/2) Epoch 24, batch 28400, giga_loss[loss=0.2931, simple_loss=0.3693, pruned_loss=0.1084, over 28972.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3747, pruned_loss=0.1226, over 5639706.07 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1139, over 5687929.63 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3753, pruned_loss=0.123, over 5651261.19 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 13:56:55,452 INFO [optim.py:369] (1/2) 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 13:56:59,403 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-12 13:57:03,945 INFO [train.py:968] (1/2) Epoch 24, batch 28450, giga_loss[loss=0.3455, simple_loss=0.3989, pruned_loss=0.146, over 28807.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3744, pruned_loss=0.1218, over 5655175.61 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1137, over 5693132.88 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3756, pruned_loss=0.1225, over 5658757.56 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:57:47,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:57:53,491 INFO [train.py:968] (1/2) Epoch 24, batch 28500, giga_loss[loss=0.3522, simple_loss=0.3871, pruned_loss=0.1587, over 23532.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3731, pruned_loss=0.122, over 5656456.10 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3611, pruned_loss=0.1139, over 5694878.34 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.374, pruned_loss=0.1226, over 5657329.66 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:58:22,673 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:58:25,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:58:30,488 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-12 13:58:41,143 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 28550, giga_loss[loss=0.2751, simple_loss=0.3444, pruned_loss=0.1029, over 28910.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3726, pruned_loss=0.1221, over 5666383.19 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3613, pruned_loss=0.1141, over 5696755.09 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3732, pruned_loss=0.1225, over 5665154.20 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:59:02,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1077455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:59:15,977 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 13:59:50,244 INFO [train.py:968] (1/2) Epoch 24, batch 28600, giga_loss[loss=0.2921, simple_loss=0.3577, pruned_loss=0.1133, over 27564.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3701, pruned_loss=0.1205, over 5671947.85 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3615, pruned_loss=0.114, over 5697794.17 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 5669606.07 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 13:59:53,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8030, 1.9454, 1.5486, 1.6617], device='cuda:1'), covar=tensor([0.0866, 0.0506, 0.0917, 0.0968], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0451, 0.0521, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 14:00:17,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9078, 2.1841, 1.7244, 1.9360], device='cuda:1'), covar=tensor([0.2719, 0.2785, 0.3199, 0.2578], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1118, 0.1364, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 14:00:23,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077532.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 14:00:24,580 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:00:26,707 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 14:00:27,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:00:28,953 INFO [optim.py:369] (1/2) 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 14:00:41,313 INFO [train.py:968] (1/2) Epoch 24, batch 28650, giga_loss[loss=0.381, simple_loss=0.4211, pruned_loss=0.1705, over 27587.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.37, pruned_loss=0.1211, over 5674807.91 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3611, pruned_loss=0.1137, over 5701885.93 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3709, pruned_loss=0.1218, over 5669065.08 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:00:58,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1077566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:01:12,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3387, 2.2845, 2.1596, 2.0078], device='cuda:1'), covar=tensor([0.1930, 0.2559, 0.2405, 0.2549], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0759, 0.0727, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 14:01:15,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5461, 2.2168, 1.6530, 0.7831], device='cuda:1'), covar=tensor([0.6634, 0.3475, 0.4417, 0.7312], device='cuda:1'), in_proj_covar=tensor([0.1792, 0.1693, 0.1628, 0.1459], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 14:01:24,098 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 14:01:28,839 INFO [train.py:968] (1/2) Epoch 24, batch 28700, giga_loss[loss=0.3235, simple_loss=0.3794, pruned_loss=0.1338, over 28255.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3709, pruned_loss=0.1224, over 5656582.15 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3615, pruned_loss=0.1139, over 5695170.95 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1229, over 5656898.15 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:02:14,144 INFO [optim.py:369] (1/2) 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 14:02:20,359 INFO [train.py:968] (1/2) Epoch 24, batch 28750, giga_loss[loss=0.3055, simple_loss=0.3745, pruned_loss=0.1182, over 28716.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5655353.08 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3618, pruned_loss=0.1141, over 5696408.07 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3722, pruned_loss=0.1237, over 5653877.42 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:02:25,957 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 14:02:27,450 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:02:51,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077675.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 14:02:51,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4906, 1.7428, 1.4014, 1.7006], device='cuda:1'), covar=tensor([0.2664, 0.2787, 0.3156, 0.2300], device='cuda:1'), in_proj_covar=tensor([0.1547, 0.1118, 0.1363, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 14:02:53,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077678.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 14:03:14,655 INFO [train.py:968] (1/2) Epoch 24, batch 28800, giga_loss[loss=0.2803, simple_loss=0.3564, pruned_loss=0.1021, over 28888.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3751, pruned_loss=0.1271, over 5640027.63 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3615, pruned_loss=0.1141, over 5687221.65 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3758, pruned_loss=0.1276, over 5646387.66 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:03:20,987 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1077707.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 14:03:33,333 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1077720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:03:51,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5495, 1.8352, 1.4365, 1.7393], device='cuda:1'), covar=tensor([0.2594, 0.2611, 0.3019, 0.2251], device='cuda:1'), in_proj_covar=tensor([0.1545, 0.1116, 0.1360, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 14:03:57,440 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 28850, giga_loss[loss=0.3193, simple_loss=0.3796, pruned_loss=0.1295, over 28774.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3748, pruned_loss=0.1268, over 5645177.36 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5692692.42 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3758, pruned_loss=0.1275, over 5644368.75 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:04:51,848 INFO [train.py:968] (1/2) Epoch 24, batch 28900, giga_loss[loss=0.3565, simple_loss=0.4028, pruned_loss=0.1551, over 29016.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3739, pruned_loss=0.1266, over 5649286.81 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5698119.00 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3747, pruned_loss=0.1273, over 5642654.39 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:05:33,604 INFO [optim.py:369] (1/2) 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 14:05:34,378 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1077841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:05:39,691 INFO [train.py:968] (1/2) Epoch 24, batch 28950, giga_loss[loss=0.3019, simple_loss=0.3662, pruned_loss=0.1188, over 28666.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.3742, pruned_loss=0.127, over 5655782.36 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5700645.43 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3751, pruned_loss=0.1279, over 5647550.28 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:06:27,245 INFO [train.py:968] (1/2) Epoch 24, batch 29000, giga_loss[loss=0.2916, simple_loss=0.3573, pruned_loss=0.113, over 28814.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3734, pruned_loss=0.1258, over 5649845.66 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3618, pruned_loss=0.1142, over 5703463.21 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3744, pruned_loss=0.1268, over 5638991.01 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:06:27,469 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2728, 3.5876, 1.4441, 1.5652], device='cuda:1'), covar=tensor([0.1059, 0.0554, 0.0930, 0.1432], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0561, 0.0394, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 14:06:36,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 14:06:53,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3975, 2.0856, 1.5080, 0.6181], device='cuda:1'), covar=tensor([0.5486, 0.2772, 0.3873, 0.6622], device='cuda:1'), in_proj_covar=tensor([0.1803, 0.1704, 0.1639, 0.1467], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 14:07:03,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4398, 1.7318, 1.3817, 1.2843], device='cuda:1'), covar=tensor([0.2888, 0.2807, 0.3304, 0.2527], device='cuda:1'), in_proj_covar=tensor([0.1548, 0.1118, 0.1364, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 14:07:08,646 INFO [optim.py:369] (1/2) 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 14:07:16,338 INFO [train.py:968] (1/2) Epoch 24, batch 29050, giga_loss[loss=0.2916, simple_loss=0.3623, pruned_loss=0.1105, over 29013.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.373, pruned_loss=0.1246, over 5654049.98 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1144, over 5705916.09 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.374, pruned_loss=0.1256, over 5642227.58 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:07:52,966 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1077984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:07:54,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1077987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:08:05,231 INFO [train.py:968] (1/2) Epoch 24, batch 29100, giga_loss[loss=0.4285, simple_loss=0.446, pruned_loss=0.2055, over 26697.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3745, pruned_loss=0.1254, over 5662299.27 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5709589.57 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3756, pruned_loss=0.1265, over 5648789.94 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:08:23,051 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:08:33,886 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:08:51,866 INFO [train.py:968] (1/2) Epoch 24, batch 29150, giga_loss[loss=0.3247, simple_loss=0.3916, pruned_loss=0.1289, over 29087.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3756, pruned_loss=0.1263, over 5675377.25 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3619, pruned_loss=0.1144, over 5714244.15 frames. ], giga_tot_loss[loss=0.3157, simple_loss=0.3767, pruned_loss=0.1273, over 5659286.67 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:09:22,816 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:09:31,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:09:33,530 INFO [train.py:968] (1/2) Epoch 24, batch 29200, giga_loss[loss=0.2949, simple_loss=0.3642, pruned_loss=0.1128, over 28948.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1268, over 5667774.38 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1148, over 5698247.45 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3776, pruned_loss=0.1278, over 5667440.57 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:10:13,373 INFO [optim.py:369] (1/2) 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 14:10:17,766 INFO [train.py:968] (1/2) Epoch 24, batch 29250, libri_loss[loss=0.2612, simple_loss=0.3308, pruned_loss=0.09579, over 29500.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3763, pruned_loss=0.1265, over 5665505.12 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3616, pruned_loss=0.1143, over 5702352.80 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3785, pruned_loss=0.1283, over 5660093.33 frames. ], batch size: 70, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:10:31,502 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-12 14:10:40,196 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:10:43,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:11:08,181 INFO [train.py:968] (1/2) Epoch 24, batch 29300, giga_loss[loss=0.2735, simple_loss=0.3629, pruned_loss=0.09203, over 28988.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3775, pruned_loss=0.1259, over 5658977.91 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3619, pruned_loss=0.1146, over 5694280.75 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3794, pruned_loss=0.1273, over 5661296.56 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:11:12,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:11:48,804 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:11:50,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:11:51,177 INFO [optim.py:369] (1/2) 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 14:11:57,023 INFO [train.py:968] (1/2) Epoch 24, batch 29350, giga_loss[loss=0.2816, simple_loss=0.3526, pruned_loss=0.1053, over 28509.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3759, pruned_loss=0.1243, over 5658702.25 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.362, pruned_loss=0.1146, over 5697431.86 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3775, pruned_loss=0.1256, over 5657069.28 frames. ], batch size: 60, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:12:15,287 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:12:27,953 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:12:40,427 INFO [train.py:968] (1/2) Epoch 24, batch 29400, giga_loss[loss=0.2809, simple_loss=0.3462, pruned_loss=0.1078, over 29010.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3735, pruned_loss=0.1229, over 5657800.64 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.362, pruned_loss=0.1147, over 5693356.76 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3752, pruned_loss=0.1242, over 5658800.44 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:13:17,329 INFO [optim.py:369] (1/2) 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 14:13:22,577 INFO [train.py:968] (1/2) Epoch 24, batch 29450, giga_loss[loss=0.3033, simple_loss=0.3662, pruned_loss=0.1202, over 28604.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.373, pruned_loss=0.1229, over 5655001.41 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1146, over 5692157.39 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3751, pruned_loss=0.1243, over 5654878.20 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:13:47,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 14:13:51,399 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:14:10,845 INFO [train.py:968] (1/2) Epoch 24, batch 29500, giga_loss[loss=0.277, simple_loss=0.3525, pruned_loss=0.1008, over 28624.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3745, pruned_loss=0.1238, over 5665296.39 frames. ], libri_tot_loss[loss=0.2954, simple_loss=0.3616, pruned_loss=0.1146, over 5698934.70 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3767, pruned_loss=0.1253, over 5657945.40 frames. ], batch size: 60, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:14:12,430 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 14:14:53,636 INFO [optim.py:369] (1/2) 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 14:15:01,668 INFO [train.py:968] (1/2) Epoch 24, batch 29550, giga_loss[loss=0.299, simple_loss=0.3601, pruned_loss=0.119, over 28613.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3752, pruned_loss=0.1249, over 5662779.53 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1146, over 5705084.21 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3773, pruned_loss=0.1263, over 5650089.76 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:15:12,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:15:50,207 INFO [train.py:968] (1/2) Epoch 24, batch 29600, giga_loss[loss=0.3063, simple_loss=0.3762, pruned_loss=0.1182, over 28514.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1247, over 5657325.45 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1146, over 5697119.03 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3762, pruned_loss=0.1262, over 5653529.43 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:16:33,657 INFO [optim.py:369] (1/2) 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 14:16:37,496 INFO [train.py:968] (1/2) Epoch 24, batch 29650, giga_loss[loss=0.2936, simple_loss=0.3635, pruned_loss=0.1119, over 28923.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3751, pruned_loss=0.1261, over 5656402.93 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3613, pruned_loss=0.1143, over 5700648.19 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3773, pruned_loss=0.1277, over 5649461.32 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:17:25,758 INFO [train.py:968] (1/2) Epoch 24, batch 29700, giga_loss[loss=0.3342, simple_loss=0.3896, pruned_loss=0.1394, over 27649.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3758, pruned_loss=0.1262, over 5662821.52 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3614, pruned_loss=0.1143, over 5704945.93 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3777, pruned_loss=0.1277, over 5652853.20 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:17:32,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:17:37,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:18:08,498 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:18:15,252 INFO [optim.py:369] (1/2) 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 14:18:18,629 INFO [train.py:968] (1/2) Epoch 24, batch 29750, giga_loss[loss=0.2621, simple_loss=0.3424, pruned_loss=0.09091, over 28947.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3757, pruned_loss=0.1259, over 5658412.17 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3613, pruned_loss=0.1143, over 5706160.20 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3773, pruned_loss=0.1272, over 5649431.34 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:18:27,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:18:27,587 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2750, 3.1064, 1.4417, 1.4595], device='cuda:1'), covar=tensor([0.1075, 0.0414, 0.0910, 0.1428], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0563, 0.0395, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 14:18:40,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3335, 1.5296, 1.5277, 1.2726], device='cuda:1'), covar=tensor([0.2798, 0.2408, 0.1946, 0.2467], device='cuda:1'), in_proj_covar=tensor([0.2022, 0.1969, 0.1879, 0.2029], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 14:18:52,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8645, 1.9890, 1.6262, 2.2011], device='cuda:1'), covar=tensor([0.2565, 0.2799, 0.3131, 0.2385], device='cuda:1'), in_proj_covar=tensor([0.1545, 0.1115, 0.1361, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 14:19:06,790 INFO [train.py:968] (1/2) Epoch 24, batch 29800, giga_loss[loss=0.312, simple_loss=0.3764, pruned_loss=0.1238, over 28507.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3736, pruned_loss=0.1235, over 5675838.60 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1144, over 5709047.58 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3749, pruned_loss=0.1245, over 5665641.41 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:19:50,309 INFO [optim.py:369] (1/2) 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 14:19:55,266 INFO [train.py:968] (1/2) Epoch 24, batch 29850, giga_loss[loss=0.295, simple_loss=0.3634, pruned_loss=0.1134, over 28736.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3724, pruned_loss=0.1224, over 5656102.96 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1142, over 5700654.70 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.374, pruned_loss=0.1236, over 5655138.51 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:19:59,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1078752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:20:22,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4078, 3.2093, 1.4546, 1.5461], device='cuda:1'), covar=tensor([0.0980, 0.0377, 0.0902, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0564, 0.0395, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 14:20:45,044 INFO [train.py:968] (1/2) Epoch 24, batch 29900, giga_loss[loss=0.295, simple_loss=0.3619, pruned_loss=0.114, over 28335.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1218, over 5661220.64 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3608, pruned_loss=0.1139, over 5704748.06 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3739, pruned_loss=0.1231, over 5656218.52 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:20:47,818 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:20:50,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:21:05,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-12 14:21:11,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1941, 4.0341, 3.8247, 2.0230], device='cuda:1'), covar=tensor([0.0634, 0.0757, 0.0762, 0.1994], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.1194, 0.1007, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 14:21:17,152 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:21:30,583 INFO [optim.py:369] (1/2) 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 14:21:35,792 INFO [train.py:968] (1/2) Epoch 24, batch 29950, giga_loss[loss=0.38, simple_loss=0.4166, pruned_loss=0.1716, over 26622.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3706, pruned_loss=0.121, over 5656594.32 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3613, pruned_loss=0.1142, over 5699612.70 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3717, pruned_loss=0.1218, over 5656912.75 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:22:18,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3263, 1.8414, 1.3476, 0.6334], device='cuda:1'), covar=tensor([0.4767, 0.2474, 0.3182, 0.5812], device='cuda:1'), in_proj_covar=tensor([0.1802, 0.1702, 0.1636, 0.1463], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 14:22:18,964 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1078895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:22:20,588 INFO [train.py:968] (1/2) Epoch 24, batch 30000, giga_loss[loss=0.2652, simple_loss=0.34, pruned_loss=0.0952, over 28981.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3694, pruned_loss=0.1208, over 5662474.70 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5705782.84 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3703, pruned_loss=0.1213, over 5655882.65 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:22:20,589 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 14:22:30,352 INFO [train.py:1012] (1/2) Epoch 24, validation: loss=0.2045, simple_loss=0.313, pruned_loss=0.04803, over 944034.00 frames. 
+2023-03-12 14:22:30,353 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 14:22:30,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1078898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:22:55,147 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1078927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:23:03,210 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1078937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:23:09,621 INFO [optim.py:369] (1/2) 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 14:23:12,989 INFO [train.py:968] (1/2) Epoch 24, batch 30050, giga_loss[loss=0.3142, simple_loss=0.3716, pruned_loss=0.1284, over 28267.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.368, pruned_loss=0.1203, over 5666372.43 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1149, over 5706685.58 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3683, pruned_loss=0.1206, over 5658837.36 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:23:58,368 INFO [train.py:968] (1/2) Epoch 24, batch 30100, giga_loss[loss=0.3085, simple_loss=0.3626, pruned_loss=0.1272, over 29001.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3648, pruned_loss=0.1188, over 5681281.45 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3617, pruned_loss=0.1145, over 5709422.48 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3657, pruned_loss=0.1196, over 5672042.29 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:24:15,183 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3878, 2.2812, 2.2921, 2.1167], device='cuda:1'), covar=tensor([0.1963, 0.2681, 0.2214, 0.2473], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0756, 0.0725, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 14:24:38,876 INFO [optim.py:369] (1/2) 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 14:24:42,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0499, 5.8869, 5.6413, 2.9152], device='cuda:1'), covar=tensor([0.0436, 0.0610, 0.0694, 0.1626], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.1193, 0.1006, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 14:24:43,301 INFO [train.py:968] (1/2) Epoch 24, batch 30150, giga_loss[loss=0.2857, simple_loss=0.3493, pruned_loss=0.111, over 28691.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3646, pruned_loss=0.1193, over 5688339.23 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1144, over 5707894.49 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3654, pruned_loss=0.1202, over 5681656.80 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:25:05,853 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 14:25:25,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4388, 1.7837, 1.3946, 1.3138], device='cuda:1'), covar=tensor([0.2765, 0.2746, 0.3223, 0.2403], device='cuda:1'), in_proj_covar=tensor([0.1550, 0.1119, 0.1366, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 14:25:33,145 INFO [train.py:968] (1/2) Epoch 24, batch 30200, giga_loss[loss=0.2593, simple_loss=0.3333, pruned_loss=0.09268, over 29105.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3645, pruned_loss=0.1191, over 5675992.32 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1149, over 5697993.56 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3647, pruned_loss=0.1195, over 5679821.73 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:25:34,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:26:14,844 INFO [optim.py:369] (1/2) 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 14:26:20,797 INFO [train.py:968] (1/2) Epoch 24, batch 30250, giga_loss[loss=0.2725, simple_loss=0.3529, pruned_loss=0.09611, over 28720.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1174, over 5670564.13 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3621, pruned_loss=0.1147, over 5693972.01 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3645, pruned_loss=0.1179, over 5676916.40 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:27:11,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 14:27:14,103 INFO [train.py:968] (1/2) Epoch 24, batch 30300, giga_loss[loss=0.2674, simple_loss=0.3476, pruned_loss=0.09357, over 28867.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3613, pruned_loss=0.1136, over 5654656.12 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3622, pruned_loss=0.1149, over 5686323.87 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3616, pruned_loss=0.1139, over 5666402.79 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:28:00,650 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 30350, giga_loss[loss=0.2574, simple_loss=0.3425, pruned_loss=0.08611, over 28580.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3572, pruned_loss=0.1098, over 5647652.04 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5681902.03 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3579, pruned_loss=0.1102, over 5660353.83 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:28:15,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:28:18,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4154, 1.6191, 1.6607, 1.2236], device='cuda:1'), covar=tensor([0.1839, 0.2912, 0.1572, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.0913, 0.0707, 0.0959, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 14:28:53,848 INFO [train.py:968] (1/2) Epoch 24, batch 30400, giga_loss[loss=0.2329, simple_loss=0.3205, pruned_loss=0.07262, over 28850.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3539, pruned_loss=0.1067, over 5644275.16 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5677105.55 frames. ], giga_tot_loss[loss=0.284, simple_loss=0.3544, pruned_loss=0.1068, over 5656709.77 frames. ], batch size: 285, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:29:08,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:29:24,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:29:32,073 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:29:41,369 INFO [optim.py:369] (1/2) 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 14:29:43,835 INFO [train.py:968] (1/2) Epoch 24, batch 30450, giga_loss[loss=0.2421, simple_loss=0.3391, pruned_loss=0.0726, over 28668.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3508, pruned_loss=0.1032, over 5645819.03 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3613, pruned_loss=0.1149, over 5676975.71 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3512, pruned_loss=0.103, over 5655140.22 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:30:26,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4407, 1.7713, 1.7164, 1.4080], device='cuda:1'), covar=tensor([0.2454, 0.1928, 0.1573, 0.2015], device='cuda:1'), in_proj_covar=tensor([0.2004, 0.1954, 0.1866, 0.2015], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 14:30:38,946 INFO [train.py:968] (1/2) Epoch 24, batch 30500, giga_loss[loss=0.3403, simple_loss=0.3912, pruned_loss=0.1447, over 27616.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3487, pruned_loss=0.1002, over 5634986.36 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3609, pruned_loss=0.1148, over 5680810.99 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3491, pruned_loss=0.09991, over 5638324.37 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:31:31,678 INFO [optim.py:369] (1/2) 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 14:31:34,366 INFO [train.py:968] (1/2) Epoch 24, batch 30550, giga_loss[loss=0.2674, simple_loss=0.3463, pruned_loss=0.09421, over 29015.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3503, pruned_loss=0.1013, over 5638665.33 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3604, pruned_loss=0.1145, over 5684611.56 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.351, pruned_loss=0.101, over 5637240.45 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:31:42,617 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:31:47,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:32:02,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:32:05,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-12 14:32:14,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:32:25,002 INFO [train.py:968] (1/2) Epoch 24, batch 30600, libri_loss[loss=0.3417, simple_loss=0.3893, pruned_loss=0.1471, over 29180.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3487, pruned_loss=0.1006, over 5639647.68 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3599, pruned_loss=0.1146, over 5687859.90 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3491, pruned_loss=0.09978, over 5633624.93 frames. ], batch size: 97, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:32:35,318 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:33:11,102 INFO [optim.py:369] (1/2) 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 14:33:14,101 INFO [train.py:968] (1/2) Epoch 24, batch 30650, giga_loss[loss=0.2453, simple_loss=0.3305, pruned_loss=0.08008, over 28733.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3448, pruned_loss=0.09795, over 5645275.04 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3588, pruned_loss=0.1141, over 5691878.15 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3459, pruned_loss=0.09745, over 5636256.97 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:34:05,465 INFO [train.py:968] (1/2) Epoch 24, batch 30700, giga_loss[loss=0.2669, simple_loss=0.3405, pruned_loss=0.09662, over 27681.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3452, pruned_loss=0.09844, over 5631786.91 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.359, pruned_loss=0.1144, over 5677447.21 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3456, pruned_loss=0.0974, over 5635668.85 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:34:15,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-12 14:34:25,821 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:34:29,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:34:41,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:34:53,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2247, 1.3258, 3.5884, 3.0884], device='cuda:1'), covar=tensor([0.1686, 0.2761, 0.0468, 0.1096], device='cuda:1'), in_proj_covar=tensor([0.0778, 0.0658, 0.0978, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 14:34:55,412 INFO [train.py:968] (1/2) Epoch 24, batch 30750, giga_loss[loss=0.3185, simple_loss=0.37, pruned_loss=0.1335, over 26765.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3458, pruned_loss=0.09864, over 5636570.52 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3584, pruned_loss=0.1141, over 5679148.96 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3464, pruned_loss=0.09781, over 5637374.80 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:34:57,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:35:27,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3518, 1.6145, 1.5812, 1.2142], device='cuda:1'), covar=tensor([0.1879, 0.2724, 0.1579, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.0913, 0.0704, 0.0958, 0.0858], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 14:35:45,959 INFO [train.py:968] (1/2) Epoch 24, batch 30800, giga_loss[loss=0.2574, simple_loss=0.3453, pruned_loss=0.08475, over 28571.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3434, pruned_loss=0.09605, over 5647839.81 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3584, pruned_loss=0.1142, over 5680298.22 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3437, pruned_loss=0.09498, over 5646463.25 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:35:49,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:36:01,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:36:32,843 INFO [optim.py:369] (1/2) 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 14:36:37,479 INFO [train.py:968] (1/2) Epoch 24, batch 30850, giga_loss[loss=0.2318, simple_loss=0.317, pruned_loss=0.07324, over 28567.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.34, pruned_loss=0.09344, over 5642896.36 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3578, pruned_loss=0.1139, over 5682195.31 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3404, pruned_loss=0.09237, over 5639254.96 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:37:04,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-12 14:37:04,863 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:37:07,368 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:37:10,807 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 30900, giga_loss[loss=0.2229, simple_loss=0.3075, pruned_loss=0.06919, over 29027.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3356, pruned_loss=0.09114, over 5642005.89 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3575, pruned_loss=0.1138, over 5684956.16 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3359, pruned_loss=0.09012, over 5636319.58 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:37:43,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:17,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:18,401 INFO [optim.py:369] (1/2) 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 14:38:20,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:20,695 INFO [train.py:968] (1/2) Epoch 24, batch 30950, giga_loss[loss=0.2361, simple_loss=0.3131, pruned_loss=0.07957, over 28956.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09133, over 5648557.32 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3574, pruned_loss=0.114, over 5689894.47 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3347, pruned_loss=0.08979, over 5638719.99 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:38:27,403 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1079853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:31,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1079856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:49,029 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1079875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:50,042 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:38:59,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1079884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:39:03,315 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1079885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:39:16,932 INFO [train.py:968] (1/2) Epoch 24, batch 31000, giga_loss[loss=0.2184, simple_loss=0.2837, pruned_loss=0.07658, over 24138.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3348, pruned_loss=0.09197, over 5633465.60 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3572, pruned_loss=0.1138, over 5692796.49 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3345, pruned_loss=0.09062, over 5622703.09 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:40:07,961 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 31050, giga_loss[loss=0.2926, simple_loss=0.3695, pruned_loss=0.1079, over 28788.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3371, pruned_loss=0.09302, over 5633360.64 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3572, pruned_loss=0.1142, over 5692361.57 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3363, pruned_loss=0.09102, over 5623635.79 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:40:20,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-12 14:41:06,532 INFO [train.py:968] (1/2) Epoch 24, batch 31100, libri_loss[loss=0.2783, simple_loss=0.3482, pruned_loss=0.1042, over 29541.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3401, pruned_loss=0.09344, over 5639829.90 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3573, pruned_loss=0.1142, over 5687614.44 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3387, pruned_loss=0.09104, over 5634232.55 frames. ], batch size: 80, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:41:41,003 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:41:43,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:41:54,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9797, 1.5276, 1.3515, 1.2854], device='cuda:1'), covar=tensor([0.2401, 0.1761, 0.2224, 0.2006], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0743, 0.0713, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 14:41:59,887 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:42:06,880 INFO [optim.py:369] (1/2) 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 14:42:07,685 INFO [train.py:968] (1/2) Epoch 24, batch 31150, giga_loss[loss=0.2962, simple_loss=0.3617, pruned_loss=0.1153, over 28987.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.341, pruned_loss=0.09361, over 5660767.65 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3568, pruned_loss=0.114, over 5693016.59 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3399, pruned_loss=0.09142, over 5650461.35 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:42:21,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:42:51,989 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:43:07,742 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 14:43:09,639 INFO [train.py:968] (1/2) Epoch 24, batch 31200, giga_loss[loss=0.2488, simple_loss=0.3288, pruned_loss=0.08438, over 28026.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.339, pruned_loss=0.09255, over 5669122.69 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3562, pruned_loss=0.1137, over 5697424.95 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3382, pruned_loss=0.09037, over 5655679.73 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:43:45,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2922, 4.1399, 3.8928, 1.9857], device='cuda:1'), covar=tensor([0.0601, 0.0723, 0.0778, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.1254, 0.1161, 0.0976, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 14:44:06,111 INFO [optim.py:369] (1/2) 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 14:44:07,581 INFO [train.py:968] (1/2) Epoch 24, batch 31250, giga_loss[loss=0.2241, simple_loss=0.3158, pruned_loss=0.06621, over 28424.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.338, pruned_loss=0.09152, over 5659818.43 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3559, pruned_loss=0.1137, over 5693708.28 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3371, pruned_loss=0.08915, over 5651796.90 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:44:11,838 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:45:13,254 INFO [train.py:968] (1/2) Epoch 24, batch 31300, giga_loss[loss=0.2812, simple_loss=0.3494, pruned_loss=0.1065, over 27633.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3369, pruned_loss=0.09019, over 5660931.47 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3559, pruned_loss=0.1138, over 5693063.00 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3358, pruned_loss=0.08776, over 5654236.12 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:45:46,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7673, 1.4734, 4.8304, 3.5499], device='cuda:1'), covar=tensor([0.1552, 0.2880, 0.0367, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0655, 0.0969, 0.0928], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 14:46:09,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7177, 1.9221, 1.5595, 1.8837], device='cuda:1'), covar=tensor([0.2738, 0.2659, 0.3014, 0.2613], device='cuda:1'), in_proj_covar=tensor([0.1558, 0.1120, 0.1374, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 14:46:17,995 INFO [optim.py:369] (1/2) 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 14:46:18,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5323, 1.7356, 1.7744, 1.5453], device='cuda:1'), covar=tensor([0.2738, 0.2094, 0.1762, 0.2149], device='cuda:1'), in_proj_covar=tensor([0.1969, 0.1913, 0.1824, 0.1970], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 14:46:18,977 INFO [train.py:968] (1/2) Epoch 24, batch 31350, giga_loss[loss=0.2426, simple_loss=0.3285, pruned_loss=0.07836, over 28989.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3331, pruned_loss=0.08905, over 5669384.70 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3554, pruned_loss=0.1135, over 5695715.74 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3325, pruned_loss=0.08705, over 5661211.35 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:46:24,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:47:16,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2717, 1.7010, 1.6865, 1.3926], device='cuda:1'), covar=tensor([0.2190, 0.1766, 0.2122, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.0481, 0.0742, 0.0712, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 14:47:16,740 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:47:19,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:47:21,565 INFO [train.py:968] (1/2) Epoch 24, batch 31400, giga_loss[loss=0.2513, simple_loss=0.3371, pruned_loss=0.08277, over 29098.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3324, pruned_loss=0.08853, over 5659415.24 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3552, pruned_loss=0.1134, over 5688306.26 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3317, pruned_loss=0.08675, over 5659257.11 frames. ], batch size: 285, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:47:43,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:47:57,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:48:01,186 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-12 14:48:24,066 INFO [optim.py:369] (1/2) 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 14:48:26,088 INFO [train.py:968] (1/2) Epoch 24, batch 31450, giga_loss[loss=0.2641, simple_loss=0.3489, pruned_loss=0.08962, over 28590.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3331, pruned_loss=0.08903, over 5660629.29 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3551, pruned_loss=0.1135, over 5688138.12 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3324, pruned_loss=0.0872, over 5660370.56 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:48:47,088 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:49:21,638 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:49:24,640 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 31500, giga_loss[loss=0.2913, simple_loss=0.3565, pruned_loss=0.113, over 26778.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3344, pruned_loss=0.08873, over 5657540.05 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3553, pruned_loss=0.1137, over 5687954.08 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3334, pruned_loss=0.08685, over 5657156.78 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:49:37,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2795, 2.5801, 1.2742, 1.4954], device='cuda:1'), covar=tensor([0.0980, 0.0364, 0.0948, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0560, 0.0394, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 14:49:42,803 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-12 14:49:49,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:49:54,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:50:04,096 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:50:30,982 INFO [optim.py:369] (1/2) 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 14:50:33,175 INFO [train.py:968] (1/2) Epoch 24, batch 31550, giga_loss[loss=0.2428, simple_loss=0.3102, pruned_loss=0.08768, over 26952.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.333, pruned_loss=0.08765, over 5663309.48 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3547, pruned_loss=0.1134, over 5691531.87 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3325, pruned_loss=0.08611, over 5659524.13 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:50:50,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:51:02,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0581, 2.2906, 1.6301, 2.0135], device='cuda:1'), covar=tensor([0.0973, 0.0669, 0.0998, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0447, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 14:51:40,596 INFO [train.py:968] (1/2) Epoch 24, batch 31600, giga_loss[loss=0.277, simple_loss=0.358, pruned_loss=0.09798, over 28889.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3324, pruned_loss=0.08777, over 5674820.89 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3542, pruned_loss=0.1132, over 5699214.35 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3317, pruned_loss=0.08583, over 5664018.53 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:51:41,490 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4393, 4.2822, 4.0997, 2.0031], device='cuda:1'), covar=tensor([0.0542, 0.0701, 0.0747, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.1158, 0.0974, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 14:52:28,052 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:52:46,726 INFO [train.py:968] (1/2) Epoch 24, batch 31650, giga_loss[loss=0.2405, simple_loss=0.3367, pruned_loss=0.07211, over 28975.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3338, pruned_loss=0.08848, over 5673375.48 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3535, pruned_loss=0.1129, over 5705315.95 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3333, pruned_loss=0.08656, over 5658250.00 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 14:53:00,051 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:00,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:04,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:40,874 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:46,369 INFO [train.py:968] (1/2) Epoch 24, batch 31700, giga_loss[loss=0.2522, simple_loss=0.3514, pruned_loss=0.07652, over 28838.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3378, pruned_loss=0.08903, over 5663245.41 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.353, pruned_loss=0.1125, over 5703715.31 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3371, pruned_loss=0.08677, over 5651506.03 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:53:52,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:53:54,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:54:35,221 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:54:51,527 INFO [optim.py:369] (1/2) 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 14:54:51,540 INFO [train.py:968] (1/2) Epoch 24, batch 31750, giga_loss[loss=0.242, simple_loss=0.3377, pruned_loss=0.07316, over 28675.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3401, pruned_loss=0.08817, over 5659851.85 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3533, pruned_loss=0.1128, over 5695764.94 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3392, pruned_loss=0.08597, over 5657103.90 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:55:45,261 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 31800, giga_loss[loss=0.2368, simple_loss=0.3218, pruned_loss=0.07591, over 27712.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3399, pruned_loss=0.08772, over 5651493.40 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3532, pruned_loss=0.1127, over 5694047.74 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3392, pruned_loss=0.08563, over 5649834.11 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 14:56:49,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:56:57,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 14:56:58,918 INFO [train.py:968] (1/2) Epoch 24, batch 31850, giga_loss[loss=0.2977, simple_loss=0.3678, pruned_loss=0.1138, over 28342.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3398, pruned_loss=0.08762, over 5661216.81 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.353, pruned_loss=0.1127, over 5698422.69 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.339, pruned_loss=0.08547, over 5655239.70 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:56:59,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1369, 3.9582, 3.7836, 1.9522], device='cuda:1'), covar=tensor([0.0580, 0.0738, 0.0772, 0.2232], device='cuda:1'), in_proj_covar=tensor([0.1246, 0.1152, 0.0968, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 14:57:00,186 INFO [optim.py:369] (1/2) 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 14:57:56,453 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 31900, giga_loss[loss=0.2531, simple_loss=0.3298, pruned_loss=0.08817, over 28987.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.34, pruned_loss=0.08929, over 5659401.90 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3538, pruned_loss=0.1133, over 5701490.30 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3383, pruned_loss=0.08645, over 5650900.35 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:58:36,602 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:58:57,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 14:59:06,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:59:09,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 14:59:26,597 INFO [train.py:968] (1/2) Epoch 24, batch 31950, giga_loss[loss=0.2477, simple_loss=0.3323, pruned_loss=0.08162, over 29070.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3404, pruned_loss=0.0899, over 5668742.10 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3539, pruned_loss=0.1134, over 5703685.70 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3388, pruned_loss=0.08709, over 5659312.27 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 14:59:27,715 INFO [optim.py:369] (1/2) 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 14:59:56,234 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:19,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:22,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:39,947 INFO [train.py:968] (1/2) Epoch 24, batch 32000, giga_loss[loss=0.2169, simple_loss=0.3038, pruned_loss=0.06499, over 28983.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3374, pruned_loss=0.08933, over 5678290.29 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3529, pruned_loss=0.1131, over 5708677.87 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3364, pruned_loss=0.08639, over 5664609.29 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:00:45,522 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1080902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:00:54,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:05,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:31,304 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1080932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:34,417 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1080935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:35,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1080936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:01:51,651 INFO [train.py:968] (1/2) Epoch 24, batch 32050, giga_loss[loss=0.3531, simple_loss=0.3965, pruned_loss=0.1549, over 26884.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3343, pruned_loss=0.08722, over 5673819.54 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3528, pruned_loss=0.113, over 5710731.75 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3335, pruned_loss=0.08475, over 5660955.40 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:01:52,605 INFO [optim.py:369] (1/2) 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 15:02:12,508 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1080964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:02:57,658 INFO [train.py:968] (1/2) Epoch 24, batch 32100, giga_loss[loss=0.2583, simple_loss=0.3311, pruned_loss=0.09272, over 28898.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3318, pruned_loss=0.08599, over 5671395.29 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3527, pruned_loss=0.113, over 5707220.34 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3306, pruned_loss=0.08332, over 5662631.92 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:03:16,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:01,845 INFO [train.py:968] (1/2) Epoch 24, batch 32150, giga_loss[loss=0.2395, simple_loss=0.3355, pruned_loss=0.07179, over 28973.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3345, pruned_loss=0.08776, over 5674242.38 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.352, pruned_loss=0.1125, over 5709666.80 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3337, pruned_loss=0.08519, over 5664233.80 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:04:02,570 INFO [optim.py:369] (1/2) 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 15:04:06,374 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:10,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:13,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1681, 4.0100, 3.8091, 2.0218], device='cuda:1'), covar=tensor([0.0604, 0.0744, 0.0856, 0.2090], device='cuda:1'), in_proj_covar=tensor([0.1251, 0.1156, 0.0975, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 15:04:23,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5538, 1.7827, 1.4925, 1.6132], device='cuda:1'), covar=tensor([0.2601, 0.2412, 0.2644, 0.2341], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1117, 0.1370, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 15:04:40,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:43,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:45,847 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:04:45,863 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 32200, giga_loss[loss=0.2934, simple_loss=0.3573, pruned_loss=0.1147, over 27647.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3373, pruned_loss=0.08901, over 5676271.15 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3514, pruned_loss=0.1121, over 5712802.86 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3368, pruned_loss=0.087, over 5665137.52 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:05:13,220 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:05:18,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:06:10,887 INFO [train.py:968] (1/2) Epoch 24, batch 32250, giga_loss[loss=0.2479, simple_loss=0.3291, pruned_loss=0.08333, over 29023.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3352, pruned_loss=0.08913, over 5665655.19 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3513, pruned_loss=0.1121, over 5708272.31 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3347, pruned_loss=0.08697, over 5659050.43 frames. ], batch size: 165, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:06:12,000 INFO [optim.py:369] (1/2) 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 15:07:07,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:07:11,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5790, 3.4365, 3.2337, 2.0442], device='cuda:1'), covar=tensor([0.0753, 0.0921, 0.0967, 0.1861], device='cuda:1'), in_proj_covar=tensor([0.1243, 0.1151, 0.0970, 0.0721], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 15:07:14,902 INFO [train.py:968] (1/2) Epoch 24, batch 32300, giga_loss[loss=0.2384, simple_loss=0.3212, pruned_loss=0.07775, over 28678.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3367, pruned_loss=0.09072, over 5667815.10 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3511, pruned_loss=0.112, over 5711043.60 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3362, pruned_loss=0.08877, over 5659491.82 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:07:49,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2236, 1.6162, 1.4985, 1.3405], device='cuda:1'), covar=tensor([0.1996, 0.1839, 0.2207, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0744, 0.0713, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 15:08:26,185 INFO [train.py:968] (1/2) Epoch 24, batch 32350, giga_loss[loss=0.2569, simple_loss=0.3407, pruned_loss=0.08652, over 28083.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3389, pruned_loss=0.09183, over 5669692.53 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3513, pruned_loss=0.1122, over 5711501.55 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3381, pruned_loss=0.08974, over 5661601.13 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:08:26,903 INFO [optim.py:369] (1/2) 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 15:08:42,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2987, 1.2664, 3.6283, 3.1568], device='cuda:1'), covar=tensor([0.1591, 0.2854, 0.0457, 0.1201], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0660, 0.0973, 0.0928], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 15:09:14,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:09:38,771 INFO [train.py:968] (1/2) Epoch 24, batch 32400, giga_loss[loss=0.2429, simple_loss=0.34, pruned_loss=0.07292, over 28871.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.34, pruned_loss=0.09172, over 5672163.66 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3507, pruned_loss=0.112, over 5711708.36 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3395, pruned_loss=0.08958, over 5664409.47 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 15:09:53,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8459, 1.2496, 1.2629, 1.0406], device='cuda:1'), covar=tensor([0.2151, 0.1365, 0.2401, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.0479, 0.0745, 0.0713, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 15:10:35,273 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:10:39,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:10:52,251 INFO [train.py:968] (1/2) Epoch 24, batch 32450, giga_loss[loss=0.256, simple_loss=0.3376, pruned_loss=0.08715, over 28139.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3395, pruned_loss=0.0911, over 5666380.47 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3508, pruned_loss=0.1121, over 5706764.90 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3389, pruned_loss=0.08885, over 5663091.53 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:10:56,025 INFO [optim.py:369] (1/2) 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 15:11:21,686 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:11:49,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:12:01,020 INFO [train.py:968] (1/2) Epoch 24, batch 32500, giga_loss[loss=0.2432, simple_loss=0.3149, pruned_loss=0.08573, over 28145.00 frames. ], tot_loss[loss=0.257, simple_loss=0.335, pruned_loss=0.08953, over 5675686.38 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3498, pruned_loss=0.1117, over 5710717.55 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.335, pruned_loss=0.08746, over 5668395.08 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:12:34,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:12:37,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:12:51,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3234, 0.8420, 0.8768, 1.5090], device='cuda:1'), covar=tensor([0.0754, 0.0417, 0.0381, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 15:13:08,672 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3736, 3.2788, 1.5307, 1.5622], device='cuda:1'), covar=tensor([0.1002, 0.0360, 0.0938, 0.1354], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0560, 0.0395, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 15:13:10,907 INFO [train.py:968] (1/2) Epoch 24, batch 32550, giga_loss[loss=0.2512, simple_loss=0.3285, pruned_loss=0.087, over 27678.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3298, pruned_loss=0.08761, over 5663043.22 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.35, pruned_loss=0.1119, over 5700972.23 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3292, pruned_loss=0.08525, over 5665360.54 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:13:12,936 INFO [optim.py:369] (1/2) 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 15:13:14,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:13:19,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:13:34,143 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4807, 1.5919, 3.2187, 3.2541], device='cuda:1'), covar=tensor([0.1243, 0.2506, 0.0434, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0769, 0.0656, 0.0969, 0.0924], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 15:13:53,994 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1081481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:14:10,776 INFO [train.py:968] (1/2) Epoch 24, batch 32600, giga_loss[loss=0.275, simple_loss=0.3507, pruned_loss=0.09965, over 28938.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3276, pruned_loss=0.08683, over 5655147.78 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3494, pruned_loss=0.1115, over 5697385.08 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3267, pruned_loss=0.08416, over 5658945.60 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:14:51,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:14:56,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:15:04,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2908, 1.6811, 1.5815, 1.5128], device='cuda:1'), covar=tensor([0.1840, 0.1679, 0.1902, 0.1659], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0740, 0.0709, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 15:15:12,096 INFO [train.py:968] (1/2) Epoch 24, batch 32650, giga_loss[loss=0.2484, simple_loss=0.33, pruned_loss=0.08339, over 28865.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3296, pruned_loss=0.08848, over 5653350.46 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3494, pruned_loss=0.1114, over 5700323.52 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3287, pruned_loss=0.08607, over 5653168.72 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:15:14,474 INFO [optim.py:369] (1/2) 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 15:15:27,043 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:10,883 INFO [train.py:968] (1/2) Epoch 24, batch 32700, giga_loss[loss=0.2433, simple_loss=0.3274, pruned_loss=0.07959, over 28631.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3286, pruned_loss=0.08786, over 5646334.89 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3489, pruned_loss=0.1113, over 5694102.20 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3275, pruned_loss=0.08515, over 5650813.63 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:16:15,967 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1081601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:19,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:32,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5287, 1.6053, 1.7511, 1.3584], device='cuda:1'), covar=tensor([0.1797, 0.2606, 0.1506, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.0912, 0.0700, 0.0958, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 15:16:45,620 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1081627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:16:56,333 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:17:14,716 INFO [train.py:968] (1/2) Epoch 24, batch 32750, giga_loss[loss=0.2542, simple_loss=0.3323, pruned_loss=0.08801, over 28928.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3275, pruned_loss=0.08619, over 5650645.53 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3485, pruned_loss=0.1111, over 5696606.89 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3267, pruned_loss=0.08387, over 5651331.91 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:17:20,983 INFO [optim.py:369] (1/2) 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 15:17:26,529 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1081656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:17:31,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7214, 1.9140, 1.3031, 1.5292], device='cuda:1'), covar=tensor([0.1109, 0.0656, 0.1046, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0447, 0.0522, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 15:18:14,264 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 15:18:17,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3500, 1.4982, 1.3274, 1.5649], device='cuda:1'), covar=tensor([0.0729, 0.0393, 0.0347, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 15:18:20,714 INFO [train.py:968] (1/2) Epoch 24, batch 32800, giga_loss[loss=0.2521, simple_loss=0.326, pruned_loss=0.08911, over 28919.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3271, pruned_loss=0.08633, over 5658724.97 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3481, pruned_loss=0.1109, over 5700708.76 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3263, pruned_loss=0.08409, over 5654637.39 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:18:55,447 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:19:00,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:19:07,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4122, 1.5361, 1.6850, 1.2811], device='cuda:1'), covar=tensor([0.1684, 0.2796, 0.1496, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.0912, 0.0700, 0.0958, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 15:19:12,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 15:19:30,554 INFO [train.py:968] (1/2) Epoch 24, batch 32850, giga_loss[loss=0.2219, simple_loss=0.2908, pruned_loss=0.07652, over 24713.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3262, pruned_loss=0.08545, over 5651437.41 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3482, pruned_loss=0.111, over 5691262.63 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3252, pruned_loss=0.08319, over 5655301.65 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:19:34,103 INFO [optim.py:369] (1/2) 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 15:20:36,401 INFO [train.py:968] (1/2) Epoch 24, batch 32900, giga_loss[loss=0.2707, simple_loss=0.3527, pruned_loss=0.09434, over 28459.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3268, pruned_loss=0.08566, over 5633169.90 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3478, pruned_loss=0.1109, over 5674967.94 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3255, pruned_loss=0.083, over 5649423.66 frames. ], batch size: 369, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:20:46,564 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:21:39,935 INFO [train.py:968] (1/2) Epoch 24, batch 32950, giga_loss[loss=0.2433, simple_loss=0.326, pruned_loss=0.08033, over 29046.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3275, pruned_loss=0.08648, over 5642296.28 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3476, pruned_loss=0.1108, over 5677795.82 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3262, pruned_loss=0.08396, over 5652090.00 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:21:45,878 INFO [optim.py:369] (1/2) 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 15:22:28,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:22:38,830 INFO [train.py:968] (1/2) Epoch 24, batch 33000, giga_loss[loss=0.232, simple_loss=0.3153, pruned_loss=0.07439, over 28633.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3275, pruned_loss=0.08723, over 5654622.41 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3472, pruned_loss=0.1106, over 5685077.85 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3263, pruned_loss=0.08449, over 5654877.93 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:22:38,831 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 15:22:47,606 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 15:23:46,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3830, 1.8246, 1.6317, 1.4934], device='cuda:1'), covar=tensor([0.0816, 0.0318, 0.0332, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:1')
+2023-03-12 15:23:46,877 INFO [train.py:968] (1/2) Epoch 24, batch 33050, giga_loss[loss=0.2502, simple_loss=0.3492, pruned_loss=0.07564, over 28664.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3269, pruned_loss=0.08502, over 5662640.45 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.347, pruned_loss=0.1104, over 5690437.37 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3256, pruned_loss=0.08249, over 5657368.64 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:23:51,293 INFO [optim.py:369] (1/2) 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 15:23:53,184 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1081953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:24:09,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 15:24:51,220 INFO [train.py:968] (1/2) Epoch 24, batch 33100, giga_loss[loss=0.2432, simple_loss=0.3307, pruned_loss=0.07788, over 29104.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.33, pruned_loss=0.08545, over 5659914.54 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.347, pruned_loss=0.1103, over 5690985.65 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3287, pruned_loss=0.08328, over 5654890.86 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:25:48,621 INFO [train.py:968] (1/2) Epoch 24, batch 33150, giga_loss[loss=0.2483, simple_loss=0.329, pruned_loss=0.08382, over 28006.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3325, pruned_loss=0.08719, over 5651180.29 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3466, pruned_loss=0.11, over 5689362.39 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3313, pruned_loss=0.08474, over 5647220.98 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:25:51,856 INFO [optim.py:369] (1/2) 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 15:26:53,755 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:26:54,218 INFO [train.py:968] (1/2) Epoch 24, batch 33200, libri_loss[loss=0.2744, simple_loss=0.3471, pruned_loss=0.1009, over 29647.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3333, pruned_loss=0.08767, over 5656607.12 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3471, pruned_loss=0.1105, over 5691480.93 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3316, pruned_loss=0.08486, over 5650716.81 frames. ], batch size: 88, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:26:57,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:27:12,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 15:27:56,382 INFO [train.py:968] (1/2) Epoch 24, batch 33250, giga_loss[loss=0.2286, simple_loss=0.3165, pruned_loss=0.07032, over 28667.00 frames. ], tot_loss[loss=0.254, simple_loss=0.333, pruned_loss=0.08746, over 5660471.78 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3475, pruned_loss=0.1107, over 5695408.95 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.331, pruned_loss=0.08455, over 5651383.92 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:28:02,168 INFO [optim.py:369] (1/2) 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 15:28:35,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:28:58,016 INFO [train.py:968] (1/2) Epoch 24, batch 33300, giga_loss[loss=0.2211, simple_loss=0.2906, pruned_loss=0.07581, over 24727.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3308, pruned_loss=0.08574, over 5666031.54 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3475, pruned_loss=0.1106, over 5700088.83 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3287, pruned_loss=0.08284, over 5653678.85 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:28:58,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 15:29:47,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:49,325 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:50,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:51,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:29:53,542 INFO [train.py:968] (1/2) Epoch 24, batch 33350, libri_loss[loss=0.2448, simple_loss=0.299, pruned_loss=0.09527, over 29652.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3293, pruned_loss=0.08607, over 5664070.35 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3472, pruned_loss=0.1105, over 5697243.08 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3272, pruned_loss=0.08281, over 5655597.57 frames. ], batch size: 69, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:29:57,750 INFO [optim.py:369] (1/2) 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 15:30:14,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:30:24,980 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:30:29,424 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:30:55,364 INFO [train.py:968] (1/2) Epoch 24, batch 33400, giga_loss[loss=0.2571, simple_loss=0.333, pruned_loss=0.09065, over 26963.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3273, pruned_loss=0.08465, over 5662477.72 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3468, pruned_loss=0.1103, over 5690350.03 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3258, pruned_loss=0.08199, over 5661990.80 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:31:26,616 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:31:31,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:31:33,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:32:03,348 INFO [train.py:968] (1/2) Epoch 24, batch 33450, giga_loss[loss=0.2261, simple_loss=0.3119, pruned_loss=0.07012, over 28866.00 frames. ], tot_loss[loss=0.251, simple_loss=0.33, pruned_loss=0.08601, over 5655766.29 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3462, pruned_loss=0.11, over 5684394.57 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.329, pruned_loss=0.08368, over 5659968.60 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:32:10,777 INFO [optim.py:369] (1/2) 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 15:32:13,459 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:33:02,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6566, 2.3264, 1.6953, 0.8375], device='cuda:1'), covar=tensor([0.6583, 0.3317, 0.4391, 0.6720], device='cuda:1'), in_proj_covar=tensor([0.1782, 0.1682, 0.1616, 0.1452], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 15:33:14,295 INFO [train.py:968] (1/2) Epoch 24, batch 33500, libri_loss[loss=0.2964, simple_loss=0.3465, pruned_loss=0.1231, over 29560.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.33, pruned_loss=0.08608, over 5657065.66 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.346, pruned_loss=0.11, over 5687860.44 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3292, pruned_loss=0.08388, over 5656658.41 frames. ], batch size: 78, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:33:26,281 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:33:31,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:34:11,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:34:21,034 INFO [train.py:968] (1/2) Epoch 24, batch 33550, giga_loss[loss=0.2782, simple_loss=0.3614, pruned_loss=0.09754, over 28980.00 frames. ], tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08704, over 5670287.60 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3458, pruned_loss=0.11, over 5688758.06 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3313, pruned_loss=0.08487, over 5668733.07 frames. ], batch size: 285, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:34:26,485 INFO [optim.py:369] (1/2) 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 15:34:30,817 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-12 15:34:50,788 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1082471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:34:54,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1082474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:35:16,449 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 15:35:18,552 INFO [train.py:968] (1/2) Epoch 24, batch 33600, giga_loss[loss=0.265, simple_loss=0.3529, pruned_loss=0.08856, over 28599.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3347, pruned_loss=0.08773, over 5665517.24 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3456, pruned_loss=0.11, over 5688075.36 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3341, pruned_loss=0.08576, over 5664760.62 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:35:27,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1082503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:36:21,823 INFO [train.py:968] (1/2) Epoch 24, batch 33650, giga_loss[loss=0.2592, simple_loss=0.3371, pruned_loss=0.09063, over 29107.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3353, pruned_loss=0.08778, over 5654921.66 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3458, pruned_loss=0.11, over 5674884.94 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3342, pruned_loss=0.08523, over 5664108.60 frames. ], batch size: 200, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:36:28,995 INFO [optim.py:369] (1/2) 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 15:37:39,927 INFO [train.py:968] (1/2) Epoch 24, batch 33700, giga_loss[loss=0.2337, simple_loss=0.319, pruned_loss=0.0742, over 28943.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.334, pruned_loss=0.08727, over 5649839.57 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3458, pruned_loss=0.11, over 5674884.94 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3331, pruned_loss=0.08528, over 5656989.88 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:37:53,926 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1082607.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 15:38:51,790 INFO [train.py:968] (1/2) Epoch 24, batch 33750, giga_loss[loss=0.23, simple_loss=0.3152, pruned_loss=0.07235, over 28881.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3331, pruned_loss=0.08756, over 5646369.70 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3463, pruned_loss=0.1103, over 5668747.60 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3319, pruned_loss=0.08541, over 5656682.92 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:39:00,876 INFO [optim.py:369] (1/2) 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 15:40:03,934 INFO [train.py:968] (1/2) Epoch 24, batch 33800, giga_loss[loss=0.2255, simple_loss=0.3186, pruned_loss=0.06618, over 28676.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3314, pruned_loss=0.08624, over 5640437.72 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3462, pruned_loss=0.1103, over 5665453.52 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3304, pruned_loss=0.08439, over 5651064.03 frames. ], batch size: 307, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:40:35,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1082723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:41:05,548 INFO [train.py:968] (1/2) Epoch 24, batch 33850, libri_loss[loss=0.2086, simple_loss=0.2791, pruned_loss=0.06903, over 29643.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3307, pruned_loss=0.08699, over 5656145.79 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3454, pruned_loss=0.1099, over 5674911.33 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.33, pruned_loss=0.08495, over 5655413.46 frames. ], batch size: 69, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:41:11,056 INFO [optim.py:369] (1/2) 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 15:42:10,385 INFO [train.py:968] (1/2) Epoch 24, batch 33900, giga_loss[loss=0.2046, simple_loss=0.2983, pruned_loss=0.05547, over 28887.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3284, pruned_loss=0.08644, over 5640455.87 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3456, pruned_loss=0.1103, over 5669345.82 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3274, pruned_loss=0.08405, over 5644887.75 frames. ], batch size: 145, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:42:43,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3960, 1.7093, 1.2329, 1.2613], device='cuda:1'), covar=tensor([0.1058, 0.0485, 0.1091, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0444, 0.0520, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 15:43:05,720 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-12 15:43:14,063 INFO [train.py:968] (1/2) Epoch 24, batch 33950, giga_loss[loss=0.2347, simple_loss=0.319, pruned_loss=0.07516, over 28422.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3284, pruned_loss=0.08542, over 5646346.33 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3456, pruned_loss=0.1102, over 5671814.88 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3274, pruned_loss=0.08334, over 5647390.02 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:43:19,965 INFO [optim.py:369] (1/2) 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 15:44:15,408 INFO [train.py:968] (1/2) Epoch 24, batch 34000, giga_loss[loss=0.2077, simple_loss=0.3076, pruned_loss=0.0539, over 28912.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3292, pruned_loss=0.08494, over 5663657.67 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3456, pruned_loss=0.1102, over 5677017.66 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.328, pruned_loss=0.08268, over 5659317.25 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 15:44:59,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-03-12 15:45:08,804 INFO [train.py:968] (1/2) Epoch 24, batch 34050, giga_loss[loss=0.2832, simple_loss=0.3428, pruned_loss=0.1118, over 24531.00 frames. ], tot_loss[loss=0.25, simple_loss=0.331, pruned_loss=0.08447, over 5669295.10 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.345, pruned_loss=0.1099, over 5680279.87 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.33, pruned_loss=0.08195, over 5662457.98 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:45:17,611 INFO [optim.py:369] (1/2) 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 15:45:17,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1082954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:45:50,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1082982.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 15:46:05,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0704, 2.5302, 2.4198, 1.9661], device='cuda:1'), covar=tensor([0.3181, 0.2028, 0.2176, 0.2470], device='cuda:1'), in_proj_covar=tensor([0.1978, 0.1906, 0.1818, 0.1969], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 15:46:12,111 INFO [train.py:968] (1/2) Epoch 24, batch 34100, giga_loss[loss=0.229, simple_loss=0.3188, pruned_loss=0.06963, over 28857.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3313, pruned_loss=0.0839, over 5670649.85 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3448, pruned_loss=0.1098, over 5684390.95 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3305, pruned_loss=0.08157, over 5661283.85 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:47:26,182 INFO [train.py:968] (1/2) Epoch 24, batch 34150, libri_loss[loss=0.2955, simple_loss=0.3596, pruned_loss=0.1157, over 29663.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3314, pruned_loss=0.08366, over 5659348.10 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3449, pruned_loss=0.1099, over 5676635.25 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3305, pruned_loss=0.08149, over 5659566.51 frames. ], batch size: 91, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:47:32,497 INFO [optim.py:369] (1/2) 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 15:48:24,988 INFO [train.py:968] (1/2) Epoch 24, batch 34200, giga_loss[loss=0.2893, simple_loss=0.374, pruned_loss=0.1022, over 28672.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3327, pruned_loss=0.08502, over 5671144.56 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3451, pruned_loss=0.1103, over 5683273.33 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3314, pruned_loss=0.08202, over 5664871.74 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:48:25,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:49:04,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083125.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 15:49:07,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083128.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 15:49:35,694 INFO [train.py:968] (1/2) Epoch 24, batch 34250, giga_loss[loss=0.227, simple_loss=0.3203, pruned_loss=0.06686, over 28818.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3323, pruned_loss=0.08443, over 5669734.85 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3451, pruned_loss=0.1103, over 5682813.83 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.331, pruned_loss=0.08144, over 5665136.00 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:49:45,639 INFO [optim.py:369] (1/2) 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 15:49:48,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083157.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 15:49:52,409 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:50:47,549 INFO [train.py:968] (1/2) Epoch 24, batch 34300, giga_loss[loss=0.2636, simple_loss=0.3476, pruned_loss=0.08984, over 28927.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3325, pruned_loss=0.08421, over 5660330.77 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3452, pruned_loss=0.1105, over 5678140.18 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.331, pruned_loss=0.08109, over 5661047.85 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:51:45,753 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:51:49,582 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:51:53,714 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8256, 3.6661, 3.4746, 1.6725], device='cuda:1'), covar=tensor([0.0783, 0.0884, 0.0914, 0.2268], device='cuda:1'), in_proj_covar=tensor([0.1243, 0.1143, 0.0964, 0.0719], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 15:51:54,067 INFO [train.py:968] (1/2) Epoch 24, batch 34350, giga_loss[loss=0.2796, simple_loss=0.3661, pruned_loss=0.09649, over 29094.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3357, pruned_loss=0.08605, over 5645955.38 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3451, pruned_loss=0.1106, over 5663448.88 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3344, pruned_loss=0.08284, over 5658325.14 frames. ], batch size: 285, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 15:52:02,459 INFO [optim.py:369] (1/2) 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 15:52:20,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4682, 1.6550, 1.2664, 1.2963], device='cuda:1'), covar=tensor([0.0913, 0.0477, 0.1003, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0443, 0.0518, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 15:52:26,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:52:44,408 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 34400, giga_loss[loss=0.2627, simple_loss=0.3451, pruned_loss=0.09008, over 28952.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3366, pruned_loss=0.08601, over 5666319.74 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3452, pruned_loss=0.1107, over 5670751.46 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3352, pruned_loss=0.08252, over 5669365.46 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 15:53:41,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:54:09,759 INFO [train.py:968] (1/2) Epoch 24, batch 34450, giga_loss[loss=0.2413, simple_loss=0.325, pruned_loss=0.07874, over 28659.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3363, pruned_loss=0.08658, over 5678344.62 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3453, pruned_loss=0.1108, over 5674419.46 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.335, pruned_loss=0.08336, over 5677734.04 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:54:20,432 INFO [optim.py:369] (1/2) 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 15:54:32,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.46 vs. limit=5.0
+2023-03-12 15:55:22,106 INFO [train.py:968] (1/2) Epoch 24, batch 34500, libri_loss[loss=0.284, simple_loss=0.349, pruned_loss=0.1095, over 29650.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3343, pruned_loss=0.08579, over 5682172.35 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3451, pruned_loss=0.1106, over 5679366.01 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3332, pruned_loss=0.08292, over 5677336.57 frames. ], batch size: 88, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:56:32,372 INFO [train.py:968] (1/2) Epoch 24, batch 34550, giga_loss[loss=0.2254, simple_loss=0.3259, pruned_loss=0.06238, over 28885.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3328, pruned_loss=0.08392, over 5684200.21 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3455, pruned_loss=0.1109, over 5674107.47 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3314, pruned_loss=0.08101, over 5684876.87 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:56:42,392 INFO [optim.py:369] (1/2) 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 15:57:00,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:57:06,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:57:34,557 INFO [train.py:968] (1/2) Epoch 24, batch 34600, giga_loss[loss=0.2942, simple_loss=0.3555, pruned_loss=0.1165, over 28808.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3323, pruned_loss=0.08431, over 5679642.04 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3453, pruned_loss=0.1108, over 5667068.65 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3309, pruned_loss=0.08126, over 5686192.47 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:57:41,437 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:57:52,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-12 15:58:21,490 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:58:21,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:58:28,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-12 15:58:35,652 INFO [train.py:968] (1/2) Epoch 24, batch 34650, giga_loss[loss=0.228, simple_loss=0.3203, pruned_loss=0.06788, over 28925.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3331, pruned_loss=0.08455, over 5676613.88 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3451, pruned_loss=0.1105, over 5668181.85 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3319, pruned_loss=0.08175, over 5681064.18 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 15:58:49,263 INFO [optim.py:369] (1/2) 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 15:59:35,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3454, 1.7904, 1.4012, 0.9407], device='cuda:1'), covar=tensor([0.2345, 0.2357, 0.2653, 0.2246], device='cuda:1'), in_proj_covar=tensor([0.1547, 0.1113, 0.1365, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 15:59:36,161 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 15:59:43,792 INFO [train.py:968] (1/2) Epoch 24, batch 34700, giga_loss[loss=0.2351, simple_loss=0.3037, pruned_loss=0.08322, over 24497.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3357, pruned_loss=0.08645, over 5667005.55 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3452, pruned_loss=0.1107, over 5672112.51 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3345, pruned_loss=0.08372, over 5666925.09 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:00:37,530 INFO [train.py:968] (1/2) Epoch 24, batch 34750, giga_loss[loss=0.2349, simple_loss=0.3166, pruned_loss=0.07653, over 28908.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3337, pruned_loss=0.08684, over 5678563.15 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3451, pruned_loss=0.1107, over 5678974.99 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3323, pruned_loss=0.08344, over 5672008.71 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:00:49,581 INFO [optim.py:369] (1/2) 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 16:00:54,977 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:02,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:12,185 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:16,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:01:33,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-12 16:01:34,130 INFO [train.py:968] (1/2) Epoch 24, batch 34800, giga_loss[loss=0.2451, simple_loss=0.3291, pruned_loss=0.08061, over 28637.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.332, pruned_loss=0.08652, over 5675945.12 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3447, pruned_loss=0.1104, over 5684004.99 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3309, pruned_loss=0.08336, over 5665856.88 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:01:48,072 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:02:01,493 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4982, 1.7028, 1.3975, 1.6461], device='cuda:1'), covar=tensor([0.0783, 0.0308, 0.0350, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0119, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:1')
+2023-03-12 16:02:30,614 INFO [train.py:968] (1/2) Epoch 24, batch 34850, libri_loss[loss=0.2596, simple_loss=0.3222, pruned_loss=0.09851, over 28592.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3341, pruned_loss=0.08841, over 5677112.01 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3447, pruned_loss=0.1105, over 5689589.68 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3328, pruned_loss=0.08488, over 5663233.44 frames. ], batch size: 63, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:02:39,379 INFO [optim.py:369] (1/2) 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 16:02:44,195 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:03:12,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2648, 1.8841, 1.4269, 0.4933], device='cuda:1'), covar=tensor([0.5773, 0.3039, 0.4274, 0.6813], device='cuda:1'), in_proj_covar=tensor([0.1790, 0.1690, 0.1629, 0.1459], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:03:19,386 INFO [train.py:968] (1/2) Epoch 24, batch 34900, giga_loss[loss=0.2941, simple_loss=0.3737, pruned_loss=0.1073, over 28864.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3414, pruned_loss=0.09262, over 5674818.22 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3446, pruned_loss=0.1105, over 5688972.22 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3402, pruned_loss=0.08926, over 5663919.73 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:03:24,280 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1083804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:03:26,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1083807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:03:54,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1083836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:04:00,537 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 34950, giga_loss[loss=0.3206, simple_loss=0.3923, pruned_loss=0.1245, over 27898.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3493, pruned_loss=0.09648, over 5685143.55 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.345, pruned_loss=0.1106, over 5692297.68 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.348, pruned_loss=0.09332, over 5673329.27 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:04:13,782 INFO [optim.py:369] (1/2) 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 16:04:15,644 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2501, 3.4435, 2.3585, 1.2867], device='cuda:1'), covar=tensor([0.7502, 0.2703, 0.3701, 0.6589], device='cuda:1'), in_proj_covar=tensor([0.1789, 0.1690, 0.1628, 0.1459], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:04:48,446 INFO [train.py:968] (1/2) Epoch 24, batch 35000, giga_loss[loss=0.2288, simple_loss=0.3167, pruned_loss=0.0704, over 29050.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3505, pruned_loss=0.09782, over 5683653.89 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.345, pruned_loss=0.1104, over 5690889.42 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3498, pruned_loss=0.09506, over 5674717.19 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:05:00,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:05:21,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-12 16:05:26,210 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-12 16:05:31,218 INFO [train.py:968] (1/2) Epoch 24, batch 35050, giga_loss[loss=0.2469, simple_loss=0.2988, pruned_loss=0.09748, over 24135.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3457, pruned_loss=0.09611, over 5689667.65 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3449, pruned_loss=0.1103, over 5700507.85 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3452, pruned_loss=0.09333, over 5673097.77 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:05:37,165 INFO [optim.py:369] (1/2) 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 16:05:46,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2162, 2.4131, 1.3290, 1.3596], device='cuda:1'), covar=tensor([0.1017, 0.0395, 0.0942, 0.1363], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0560, 0.0395, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 16:05:47,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1083968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:05:49,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-12 16:06:05,230 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1083988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:06:14,097 INFO [train.py:968] (1/2) Epoch 24, batch 35100, giga_loss[loss=0.2116, simple_loss=0.2973, pruned_loss=0.06299, over 28868.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.339, pruned_loss=0.0933, over 5688030.41 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3453, pruned_loss=0.1105, over 5702769.94 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3382, pruned_loss=0.09063, over 5672731.65 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:06:33,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-12 16:06:54,283 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1084044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:06:54,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:06:56,722 INFO [train.py:968] (1/2) Epoch 24, batch 35150, giga_loss[loss=0.2701, simple_loss=0.3378, pruned_loss=0.1012, over 28741.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3312, pruned_loss=0.08961, over 5694669.80 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3453, pruned_loss=0.1104, over 5702938.05 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3303, pruned_loss=0.0872, over 5682344.94 frames. ], batch size: 243, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:07:01,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:03,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:03,437 INFO [optim.py:369] (1/2) 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 16:07:14,583 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1084068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:28,428 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:37,911 INFO [train.py:968] (1/2) Epoch 24, batch 35200, giga_loss[loss=0.2217, simple_loss=0.2907, pruned_loss=0.07638, over 28749.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3238, pruned_loss=0.08661, over 5682688.93 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3455, pruned_loss=0.1104, over 5695779.47 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3225, pruned_loss=0.08413, over 5678521.70 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:07:49,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:07:51,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:10,450 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:17,937 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:21,115 INFO [train.py:968] (1/2) Epoch 24, batch 35250, libri_loss[loss=0.3104, simple_loss=0.3697, pruned_loss=0.1255, over 25707.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3189, pruned_loss=0.08445, over 5682583.13 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3452, pruned_loss=0.11, over 5697323.52 frames. ], giga_tot_loss[loss=0.2405, simple_loss=0.3172, pruned_loss=0.08194, over 5677726.36 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:08:27,115 INFO [optim.py:369] (1/2) 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 16:08:44,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-12 16:08:55,150 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:08:57,143 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:03,609 INFO [train.py:968] (1/2) Epoch 24, batch 35300, giga_loss[loss=0.293, simple_loss=0.3535, pruned_loss=0.1163, over 26731.00 frames. ], tot_loss[loss=0.24, simple_loss=0.315, pruned_loss=0.08249, over 5695436.64 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3452, pruned_loss=0.1097, over 5701960.77 frames. ], giga_tot_loss[loss=0.2366, simple_loss=0.313, pruned_loss=0.08009, over 5687243.52 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:09:21,218 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:22,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6799, 2.4400, 1.6344, 0.6945], device='cuda:1'), covar=tensor([0.5990, 0.3402, 0.4599, 0.6864], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1690, 0.1625, 0.1453], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:09:23,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:33,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1084229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:09:42,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5970, 2.2122, 1.6339, 0.8523], device='cuda:1'), covar=tensor([0.6632, 0.3171, 0.4582, 0.7213], device='cuda:1'), in_proj_covar=tensor([0.1787, 0.1692, 0.1627, 0.1455], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:09:50,611 INFO [train.py:968] (1/2) Epoch 24, batch 35350, giga_loss[loss=0.195, simple_loss=0.2726, pruned_loss=0.05875, over 28839.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.311, pruned_loss=0.08072, over 5689608.92 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3449, pruned_loss=0.1094, over 5696952.01 frames. ], giga_tot_loss[loss=0.2332, simple_loss=0.3092, pruned_loss=0.07862, over 5686770.25 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:09:57,498 INFO [optim.py:369] (1/2) 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 16:10:19,764 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:10:21,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:10:31,491 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 35400, giga_loss[loss=0.2275, simple_loss=0.3001, pruned_loss=0.07749, over 28744.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3075, pruned_loss=0.07905, over 5687641.02 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3448, pruned_loss=0.1093, over 5698872.14 frames. ], giga_tot_loss[loss=0.2301, simple_loss=0.3058, pruned_loss=0.07721, over 5683463.54 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:10:46,617 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:11:00,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3106, 1.8502, 1.4320, 0.6506], device='cuda:1'), covar=tensor([0.5916, 0.2738, 0.3704, 0.6678], device='cuda:1'), in_proj_covar=tensor([0.1788, 0.1691, 0.1628, 0.1454], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:11:20,407 INFO [train.py:968] (1/2) Epoch 24, batch 35450, giga_loss[loss=0.1947, simple_loss=0.277, pruned_loss=0.05621, over 28871.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.3051, pruned_loss=0.07828, over 5675146.09 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3445, pruned_loss=0.1089, over 5694949.61 frames. ], giga_tot_loss[loss=0.2277, simple_loss=0.3028, pruned_loss=0.07623, over 5674492.77 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:11:28,312 INFO [optim.py:369] (1/2) 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 16:11:30,092 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:11:33,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:11:33,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:12:00,049 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 35500, giga_loss[loss=0.2236, simple_loss=0.2997, pruned_loss=0.07378, over 28960.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3018, pruned_loss=0.07674, over 5682820.22 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3446, pruned_loss=0.109, over 5696007.86 frames. ], giga_tot_loss[loss=0.2247, simple_loss=0.2997, pruned_loss=0.07485, over 5681234.91 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:12:24,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:12:35,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7852, 2.1199, 1.4098, 1.5902], device='cuda:1'), covar=tensor([0.1053, 0.0622, 0.1182, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0398, 0.0442, 0.0517, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 16:12:45,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:12:48,397 INFO [train.py:968] (1/2) Epoch 24, batch 35550, giga_loss[loss=0.1857, simple_loss=0.2614, pruned_loss=0.05499, over 28622.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3015, pruned_loss=0.07724, over 5677946.26 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3458, pruned_loss=0.1098, over 5686474.91 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2975, pruned_loss=0.07409, over 5685406.42 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:12:55,389 INFO [optim.py:369] (1/2) 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 16:13:32,239 INFO [train.py:968] (1/2) Epoch 24, batch 35600, giga_loss[loss=0.2423, simple_loss=0.313, pruned_loss=0.0858, over 28228.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.2987, pruned_loss=0.07574, over 5675196.36 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3459, pruned_loss=0.1097, over 5686208.95 frames. ], giga_tot_loss[loss=0.22, simple_loss=0.2946, pruned_loss=0.07269, over 5681078.83 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:13:33,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6932, 2.0743, 1.7907, 1.7761], device='cuda:1'), covar=tensor([0.1741, 0.1616, 0.2044, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0740, 0.0711, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 16:13:38,809 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:13:41,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:09,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:15,819 INFO [train.py:968] (1/2) Epoch 24, batch 35650, libri_loss[loss=0.3427, simple_loss=0.3935, pruned_loss=0.146, over 19429.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.2974, pruned_loss=0.07551, over 5657325.82 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3465, pruned_loss=0.11, over 5672984.29 frames. ], giga_tot_loss[loss=0.2175, simple_loss=0.2919, pruned_loss=0.07161, over 5675089.91 frames. ], batch size: 187, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:14:25,788 INFO [optim.py:369] (1/2) 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 16:14:29,826 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:33,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:35,776 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2121, 1.3365, 1.3302, 1.2573], device='cuda:1'), covar=tensor([0.2661, 0.2077, 0.1844, 0.2201], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1927, 0.1838, 0.1998], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 16:14:50,194 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:54,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:14:59,245 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:15:02,984 INFO [train.py:968] (1/2) Epoch 24, batch 35700, giga_loss[loss=0.243, simple_loss=0.3161, pruned_loss=0.08491, over 27569.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.2989, pruned_loss=0.077, over 5658423.10 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3473, pruned_loss=0.1104, over 5674509.38 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.2931, pruned_loss=0.07303, over 5670935.48 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:15:08,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:15:22,209 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:15:30,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 16:15:37,064 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 16:15:47,449 INFO [train.py:968] (1/2) Epoch 24, batch 35750, libri_loss[loss=0.3092, simple_loss=0.3677, pruned_loss=0.1253, over 25698.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3102, pruned_loss=0.08249, over 5666527.49 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3476, pruned_loss=0.1106, over 5671033.13 frames. ], giga_tot_loss[loss=0.2304, simple_loss=0.3041, pruned_loss=0.07832, over 5679659.02 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:15:56,979 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1084668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:16:33,362 INFO [train.py:968] (1/2) Epoch 24, batch 35800, giga_loss[loss=0.2753, simple_loss=0.3589, pruned_loss=0.09587, over 28896.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3232, pruned_loss=0.08931, over 5674456.50 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3472, pruned_loss=0.1102, over 5677550.51 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3179, pruned_loss=0.08564, over 5678761.23 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:17:15,660 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:17:15,914 INFO [train.py:968] (1/2) Epoch 24, batch 35850, giga_loss[loss=0.324, simple_loss=0.397, pruned_loss=0.1255, over 27501.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3327, pruned_loss=0.09375, over 5681015.29 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3475, pruned_loss=0.1101, over 5683769.10 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3276, pruned_loss=0.09039, over 5678814.20 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:17:18,867 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:17:24,275 INFO [optim.py:369] (1/2) 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 16:17:42,851 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 35900, giga_loss[loss=0.2325, simple_loss=0.3283, pruned_loss=0.06836, over 28887.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3392, pruned_loss=0.09598, over 5665060.55 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3477, pruned_loss=0.1102, over 5669496.89 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3346, pruned_loss=0.09281, over 5676638.14 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:18:06,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2272, 1.4467, 1.2953, 1.1386], device='cuda:1'), covar=tensor([0.3162, 0.2871, 0.1954, 0.2683], device='cuda:1'), in_proj_covar=tensor([0.2001, 0.1924, 0.1837, 0.1994], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 16:18:11,110 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1084811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:18:14,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1084814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:18:44,860 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1084843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:18:48,034 INFO [train.py:968] (1/2) Epoch 24, batch 35950, giga_loss[loss=0.2726, simple_loss=0.3527, pruned_loss=0.09629, over 28939.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3407, pruned_loss=0.0951, over 5662284.24 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3478, pruned_loss=0.1102, over 5670710.41 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.337, pruned_loss=0.09254, over 5670373.91 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:18:51,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6828, 1.8800, 1.5654, 1.7790], device='cuda:1'), covar=tensor([0.2759, 0.2840, 0.3177, 0.2493], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1118, 0.1365, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 16:18:57,625 INFO [optim.py:369] (1/2) 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 16:19:35,675 INFO [train.py:968] (1/2) Epoch 24, batch 36000, giga_loss[loss=0.2698, simple_loss=0.3573, pruned_loss=0.09119, over 28602.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3422, pruned_loss=0.09537, over 5662452.08 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3481, pruned_loss=0.1102, over 5674061.95 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3388, pruned_loss=0.09295, over 5665889.95 frames. ], batch size: 242, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:19:35,675 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 16:19:44,690 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 16:20:27,026 INFO [train.py:968] (1/2) Epoch 24, batch 36050, giga_loss[loss=0.2644, simple_loss=0.3377, pruned_loss=0.0956, over 28787.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3445, pruned_loss=0.09665, over 5675916.96 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3486, pruned_loss=0.1101, over 5677786.79 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3412, pruned_loss=0.09428, over 5675259.54 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:20:35,019 INFO [optim.py:369] (1/2) 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 16:20:37,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1825, 2.4173, 2.2811, 1.8943], device='cuda:1'), covar=tensor([0.2735, 0.2366, 0.2500, 0.2655], device='cuda:1'), in_proj_covar=tensor([0.2007, 0.1932, 0.1846, 0.2004], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 16:20:46,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5406, 1.7208, 1.7434, 1.6446], device='cuda:1'), covar=tensor([0.2150, 0.2241, 0.2416, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0742, 0.0712, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 16:21:09,719 INFO [train.py:968] (1/2) Epoch 24, batch 36100, giga_loss[loss=0.2645, simple_loss=0.3454, pruned_loss=0.09181, over 28903.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3474, pruned_loss=0.09889, over 5678747.17 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3489, pruned_loss=0.1102, over 5681288.70 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3444, pruned_loss=0.09662, over 5674939.98 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:21:21,698 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 16:21:51,086 INFO [train.py:968] (1/2) Epoch 24, batch 36150, giga_loss[loss=0.2936, simple_loss=0.3696, pruned_loss=0.1088, over 28726.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3503, pruned_loss=0.1007, over 5671034.80 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.3494, pruned_loss=0.1106, over 5666698.95 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3474, pruned_loss=0.09829, over 5680891.57 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:22:01,469 INFO [optim.py:369] (1/2) 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 16:22:31,784 INFO [train.py:968] (1/2) Epoch 24, batch 36200, giga_loss[loss=0.2818, simple_loss=0.3681, pruned_loss=0.09772, over 28764.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3538, pruned_loss=0.1012, over 5690419.04 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3497, pruned_loss=0.1107, over 5669373.19 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3513, pruned_loss=0.09912, over 5696007.53 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:23:15,534 INFO [train.py:968] (1/2) Epoch 24, batch 36250, giga_loss[loss=0.2819, simple_loss=0.3657, pruned_loss=0.0991, over 28787.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3557, pruned_loss=0.1022, over 5683191.40 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3499, pruned_loss=0.1108, over 5668062.62 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3536, pruned_loss=0.1002, over 5689812.62 frames. ], batch size: 243, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:23:25,885 INFO [optim.py:369] (1/2) 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 16:23:56,745 INFO [train.py:968] (1/2) Epoch 24, batch 36300, giga_loss[loss=0.2656, simple_loss=0.3397, pruned_loss=0.09571, over 27590.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.355, pruned_loss=0.1008, over 5689361.93 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3498, pruned_loss=0.1107, over 5671094.07 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3535, pruned_loss=0.09908, over 5692383.22 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:24:28,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:24:38,508 INFO [train.py:968] (1/2) Epoch 24, batch 36350, giga_loss[loss=0.259, simple_loss=0.3515, pruned_loss=0.08328, over 28644.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3542, pruned_loss=0.09904, over 5691055.73 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3502, pruned_loss=0.1109, over 5669651.33 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3528, pruned_loss=0.09747, over 5694730.25 frames. ], batch size: 60, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:24:47,638 INFO [optim.py:369] (1/2) 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 16:25:18,120 INFO [train.py:968] (1/2) Epoch 24, batch 36400, giga_loss[loss=0.2433, simple_loss=0.331, pruned_loss=0.07775, over 28843.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3518, pruned_loss=0.09687, over 5696915.42 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3503, pruned_loss=0.1107, over 5677387.16 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3507, pruned_loss=0.09524, over 5693923.92 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:25:39,795 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:25:44,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2152, 1.6266, 1.1968, 0.8608], device='cuda:1'), covar=tensor([0.5526, 0.2635, 0.3325, 0.6046], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1682, 0.1628, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:26:01,919 INFO [train.py:968] (1/2) Epoch 24, batch 36450, giga_loss[loss=0.2683, simple_loss=0.3451, pruned_loss=0.0957, over 28707.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3509, pruned_loss=0.09656, over 5691506.16 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3499, pruned_loss=0.1105, over 5680625.14 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3504, pruned_loss=0.09533, over 5686663.05 frames. ], batch size: 284, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:26:10,149 INFO [optim.py:369] (1/2) 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 16:26:46,587 INFO [train.py:968] (1/2) Epoch 24, batch 36500, giga_loss[loss=0.364, simple_loss=0.3901, pruned_loss=0.169, over 26514.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3524, pruned_loss=0.09961, over 5682834.73 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3496, pruned_loss=0.11, over 5679583.23 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3522, pruned_loss=0.09839, over 5680904.42 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:27:28,741 INFO [train.py:968] (1/2) Epoch 24, batch 36550, libri_loss[loss=0.3061, simple_loss=0.3762, pruned_loss=0.118, over 29269.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3547, pruned_loss=0.1027, over 5679566.35 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3502, pruned_loss=0.1104, over 5666744.70 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3541, pruned_loss=0.1011, over 5690416.98 frames. ], batch size: 94, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:27:39,369 INFO [optim.py:369] (1/2) 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 16:28:06,464 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:28:14,080 INFO [train.py:968] (1/2) Epoch 24, batch 36600, giga_loss[loss=0.3271, simple_loss=0.3751, pruned_loss=0.1395, over 26401.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3555, pruned_loss=0.1048, over 5662458.96 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3506, pruned_loss=0.1106, over 5649925.87 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3547, pruned_loss=0.1033, over 5686105.08 frames. ], batch size: 555, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:29:03,866 INFO [train.py:968] (1/2) Epoch 24, batch 36650, giga_loss[loss=0.2856, simple_loss=0.3501, pruned_loss=0.1105, over 28945.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3535, pruned_loss=0.1044, over 5674019.16 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3505, pruned_loss=0.1105, over 5654737.66 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.353, pruned_loss=0.1032, over 5688736.40 frames. ], batch size: 213, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:29:12,023 INFO [optim.py:369] (1/2) 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 16:29:24,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3719, 1.5224, 1.5053, 1.4070], device='cuda:1'), covar=tensor([0.2011, 0.2067, 0.2452, 0.2020], device='cuda:1'), in_proj_covar=tensor([0.0482, 0.0746, 0.0716, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 16:29:26,848 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 16:29:43,185 INFO [train.py:968] (1/2) Epoch 24, batch 36700, giga_loss[loss=0.2642, simple_loss=0.3412, pruned_loss=0.09361, over 28906.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.352, pruned_loss=0.1041, over 5677190.95 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3508, pruned_loss=0.1106, over 5650096.78 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3515, pruned_loss=0.1028, over 5694566.37 frames. ], batch size: 199, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:29:46,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8285, 1.9919, 1.6393, 1.8645], device='cuda:1'), covar=tensor([0.2722, 0.2876, 0.3233, 0.2797], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1122, 0.1366, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 16:30:29,677 INFO [train.py:968] (1/2) Epoch 24, batch 36750, giga_loss[loss=0.2946, simple_loss=0.3694, pruned_loss=0.1099, over 29067.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.351, pruned_loss=0.1029, over 5678916.71 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3508, pruned_loss=0.1106, over 5652721.20 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3505, pruned_loss=0.1019, over 5690459.01 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:30:40,983 INFO [optim.py:369] (1/2) 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 16:31:09,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2892, 4.1463, 3.9402, 1.9392], device='cuda:1'), covar=tensor([0.0596, 0.0729, 0.0691, 0.2069], device='cuda:1'), in_proj_covar=tensor([0.1250, 0.1156, 0.0974, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 16:31:14,632 INFO [train.py:968] (1/2) Epoch 24, batch 36800, giga_loss[loss=0.2774, simple_loss=0.3441, pruned_loss=0.1054, over 27648.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3497, pruned_loss=0.1014, over 5687474.86 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3517, pruned_loss=0.111, over 5655795.94 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3485, pruned_loss=0.09987, over 5695122.16 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:31:15,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1085699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:31:25,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 16:31:43,519 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085726.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:32:01,180 INFO [train.py:968] (1/2) Epoch 24, batch 36850, giga_loss[loss=0.2154, simple_loss=0.3059, pruned_loss=0.06245, over 28930.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3454, pruned_loss=0.09874, over 5684251.02 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.352, pruned_loss=0.111, over 5657467.06 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3442, pruned_loss=0.09732, over 5689058.27 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:32:13,619 INFO [optim.py:369] (1/2) 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 16:32:45,461 INFO [train.py:968] (1/2) Epoch 24, batch 36900, giga_loss[loss=0.2187, simple_loss=0.2856, pruned_loss=0.07585, over 23396.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3395, pruned_loss=0.09529, over 5695954.44 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3526, pruned_loss=0.1112, over 5665101.00 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3377, pruned_loss=0.09355, over 5694050.95 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:33:13,961 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:33:36,254 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1085842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:33:40,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1085845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:33:42,492 INFO [train.py:968] (1/2) Epoch 24, batch 36950, giga_loss[loss=0.2609, simple_loss=0.3316, pruned_loss=0.09516, over 29060.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3329, pruned_loss=0.09184, over 5680031.32 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3526, pruned_loss=0.1111, over 5666348.49 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3315, pruned_loss=0.09045, over 5677672.73 frames. ], batch size: 136, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:33:50,230 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:34:00,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1085865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:34:07,031 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 37000, giga_loss[loss=0.2716, simple_loss=0.3434, pruned_loss=0.09993, over 28585.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3328, pruned_loss=0.09151, over 5673681.01 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3524, pruned_loss=0.111, over 5662582.77 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3314, pruned_loss=0.0901, over 5675277.19 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:34:36,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5036, 1.1915, 4.7304, 3.6117], device='cuda:1'), covar=tensor([0.1853, 0.3063, 0.0377, 0.0849], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0656, 0.0969, 0.0927], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 16:34:43,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:35:03,879 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1085936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:35:11,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:35:13,280 INFO [train.py:968] (1/2) Epoch 24, batch 37050, giga_loss[loss=0.2458, simple_loss=0.3196, pruned_loss=0.08604, over 28788.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3349, pruned_loss=0.09289, over 5676122.31 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3533, pruned_loss=0.1114, over 5658969.70 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3324, pruned_loss=0.09087, over 5680190.77 frames. ], batch size: 119, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:35:17,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5548, 1.7804, 1.2219, 1.3468], device='cuda:1'), covar=tensor([0.1065, 0.0635, 0.1145, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0446, 0.0523, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 16:35:23,118 INFO [optim.py:369] (1/2) 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 16:35:56,799 INFO [train.py:968] (1/2) Epoch 24, batch 37100, giga_loss[loss=0.2899, simple_loss=0.3598, pruned_loss=0.11, over 27624.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3347, pruned_loss=0.09259, over 5675413.09 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3536, pruned_loss=0.1114, over 5652412.57 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3319, pruned_loss=0.09044, over 5684660.16 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:36:04,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:08,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:22,803 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:33,511 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:36:38,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-12 16:36:40,412 INFO [train.py:968] (1/2) Epoch 24, batch 37150, giga_loss[loss=0.2463, simple_loss=0.3237, pruned_loss=0.08446, over 29062.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3322, pruned_loss=0.09133, over 5682339.58 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3538, pruned_loss=0.1115, over 5656229.02 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3296, pruned_loss=0.08935, over 5686689.03 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:36:50,650 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 37200, giga_loss[loss=0.2956, simple_loss=0.3605, pruned_loss=0.1154, over 28006.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3324, pruned_loss=0.09195, over 5696525.54 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3547, pruned_loss=0.1118, over 5663257.73 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3286, pruned_loss=0.08931, over 5694993.83 frames. ], batch size: 412, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:37:22,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086101.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:37:35,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:37:54,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5442, 1.7997, 1.2602, 1.3870], device='cuda:1'), covar=tensor([0.1079, 0.0702, 0.1090, 0.1378], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0447, 0.0524, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 16:38:02,506 INFO [train.py:968] (1/2) Epoch 24, batch 37250, giga_loss[loss=0.2627, simple_loss=0.3332, pruned_loss=0.09609, over 27606.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3293, pruned_loss=0.09031, over 5704350.48 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3545, pruned_loss=0.1116, over 5664312.44 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3264, pruned_loss=0.0883, over 5702422.22 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:38:11,174 INFO [optim.py:369] (1/2) 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 16:38:40,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 16:38:41,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:38:41,708 INFO [train.py:968] (1/2) Epoch 24, batch 37300, giga_loss[loss=0.293, simple_loss=0.3451, pruned_loss=0.1204, over 24169.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3275, pruned_loss=0.08976, over 5705459.10 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3548, pruned_loss=0.1117, over 5666832.40 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3247, pruned_loss=0.08792, over 5702121.71 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:38:58,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0412, 1.1827, 3.3409, 2.9231], device='cuda:1'), covar=tensor([0.1836, 0.2932, 0.0501, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0656, 0.0969, 0.0928], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 16:39:10,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4941, 1.7704, 1.2984, 1.5650], device='cuda:1'), covar=tensor([0.0794, 0.0305, 0.0355, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0065, 0.0112], device='cuda:1')
+2023-03-12 16:39:10,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:39:21,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086244.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:39:24,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086247.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:39:24,618 INFO [train.py:968] (1/2) Epoch 24, batch 37350, giga_loss[loss=0.2133, simple_loss=0.2887, pruned_loss=0.06893, over 28549.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3258, pruned_loss=0.08902, over 5706850.12 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3555, pruned_loss=0.1117, over 5669504.27 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3225, pruned_loss=0.08708, over 5702552.28 frames. ], batch size: 85, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:39:35,879 INFO [optim.py:369] (1/2) 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 16:39:47,029 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086276.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:39:55,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:04,090 INFO [train.py:968] (1/2) Epoch 24, batch 37400, libri_loss[loss=0.2481, simple_loss=0.3294, pruned_loss=0.08338, over 29673.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3256, pruned_loss=0.08903, over 5721339.86 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3562, pruned_loss=0.1119, over 5680396.72 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3209, pruned_loss=0.08633, over 5709402.64 frames. ], batch size: 73, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:40:13,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:30,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2170, 1.7272, 1.2962, 0.4519], device='cuda:1'), covar=tensor([0.4759, 0.2843, 0.4784, 0.6467], device='cuda:1'), in_proj_covar=tensor([0.1774, 0.1671, 0.1616, 0.1442], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:40:36,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:38,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:40:41,133 INFO [train.py:968] (1/2) Epoch 24, batch 37450, giga_loss[loss=0.2219, simple_loss=0.3028, pruned_loss=0.0705, over 28990.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3245, pruned_loss=0.08863, over 5718968.27 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3573, pruned_loss=0.1124, over 5674262.11 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.319, pruned_loss=0.08544, over 5715209.35 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:40:55,157 INFO [optim.py:369] (1/2) 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 16:41:02,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:03,208 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:05,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:22,946 INFO [train.py:968] (1/2) Epoch 24, batch 37500, libri_loss[loss=0.3912, simple_loss=0.4312, pruned_loss=0.1757, over 18763.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3234, pruned_loss=0.08818, over 5717823.63 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3577, pruned_loss=0.1125, over 5669812.92 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3177, pruned_loss=0.08486, over 5721032.96 frames. ], batch size: 187, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:41:28,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:28,760 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:48,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3636, 1.6296, 1.7041, 1.4645], device='cuda:1'), covar=tensor([0.2293, 0.2155, 0.2541, 0.2361], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0750, 0.0721, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 16:41:48,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:41:50,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:06,382 INFO [train.py:968] (1/2) Epoch 24, batch 37550, giga_loss[loss=0.2974, simple_loss=0.3553, pruned_loss=0.1197, over 29045.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3223, pruned_loss=0.08754, over 5719459.34 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3579, pruned_loss=0.1126, over 5673470.35 frames. ], giga_tot_loss[loss=0.2431, simple_loss=0.3171, pruned_loss=0.08453, over 5719385.74 frames. ], batch size: 128, lr: 1.32e-03, grad_scale: 2.0
+2023-03-12 16:42:10,801 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:12,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:15,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:19,093 INFO [optim.py:369] (1/2) 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 16:42:40,501 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:45,185 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:42:50,287 INFO [train.py:968] (1/2) Epoch 24, batch 37600, giga_loss[loss=0.2636, simple_loss=0.3369, pruned_loss=0.09513, over 28957.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3258, pruned_loss=0.08956, over 5704797.36 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3582, pruned_loss=0.1127, over 5667718.40 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3208, pruned_loss=0.08666, over 5711509.74 frames. ], batch size: 106, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:43:00,058 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:43:21,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2163, 2.3046, 2.2011, 2.0692], device='cuda:1'), covar=tensor([0.1923, 0.2208, 0.1982, 0.2074], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0750, 0.0722, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 16:43:34,989 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:43:35,403 INFO [train.py:968] (1/2) Epoch 24, batch 37650, giga_loss[loss=0.2361, simple_loss=0.3129, pruned_loss=0.07967, over 28561.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3304, pruned_loss=0.09229, over 5694655.08 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3588, pruned_loss=0.1128, over 5662508.63 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.325, pruned_loss=0.08926, over 5705762.07 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:43:38,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:43:48,556 INFO [optim.py:369] (1/2) 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 16:44:05,898 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:44:26,156 INFO [train.py:968] (1/2) Epoch 24, batch 37700, giga_loss[loss=0.3102, simple_loss=0.379, pruned_loss=0.1207, over 28575.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3386, pruned_loss=0.09757, over 5688718.79 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3589, pruned_loss=0.1128, over 5662973.44 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3339, pruned_loss=0.09492, over 5697407.16 frames. ], batch size: 336, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:44:43,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-12 16:44:58,619 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:05,027 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1086635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:08,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1086638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:19,118 INFO [train.py:968] (1/2) Epoch 24, batch 37750, giga_loss[loss=0.3113, simple_loss=0.3812, pruned_loss=0.1206, over 29066.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.344, pruned_loss=0.1007, over 5677155.59 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.359, pruned_loss=0.1129, over 5665558.99 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3402, pruned_loss=0.0984, over 5681741.71 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:45:34,891 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1086667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:45:59,115 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 16:46:02,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 16:46:03,633 INFO [train.py:968] (1/2) Epoch 24, batch 37800, giga_loss[loss=0.2935, simple_loss=0.3733, pruned_loss=0.1068, over 29077.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3485, pruned_loss=0.1024, over 5673567.81 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.359, pruned_loss=0.1128, over 5660626.17 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3448, pruned_loss=0.1002, over 5681459.36 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:46:13,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3416, 1.6346, 1.3265, 1.0341], device='cuda:1'), covar=tensor([0.2613, 0.2624, 0.2963, 0.2347], device='cuda:1'), in_proj_covar=tensor([0.1550, 0.1119, 0.1364, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 16:46:55,503 INFO [train.py:968] (1/2) Epoch 24, batch 37850, giga_loss[loss=0.4188, simple_loss=0.4376, pruned_loss=0.2, over 23466.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3538, pruned_loss=0.1052, over 5664267.61 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1128, over 5661844.04 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3509, pruned_loss=0.1035, over 5669366.05 frames. ], batch size: 705, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:47:10,085 INFO [optim.py:369] (1/2) 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 16:47:13,191 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:47:19,250 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086774.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:47:20,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9296, 3.7181, 3.4897, 1.8366], device='cuda:1'), covar=tensor([0.0683, 0.0896, 0.0854, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.1240, 0.1147, 0.0967, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 16:47:25,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.96 vs. limit=5.0
+2023-03-12 16:47:29,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:47:39,678 INFO [train.py:968] (1/2) Epoch 24, batch 37900, giga_loss[loss=0.29, simple_loss=0.3587, pruned_loss=0.1107, over 28221.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.358, pruned_loss=0.1077, over 5668951.56 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3592, pruned_loss=0.1129, over 5666256.54 frames. ], giga_tot_loss[loss=0.2839, simple_loss=0.3555, pruned_loss=0.1061, over 5668785.94 frames. ], batch size: 368, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:48:17,547 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1086842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:48:22,007 INFO [train.py:968] (1/2) Epoch 24, batch 37950, giga_loss[loss=0.2788, simple_loss=0.3661, pruned_loss=0.09577, over 28876.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3527, pruned_loss=0.1032, over 5676510.90 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3591, pruned_loss=0.1128, over 5667740.51 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3506, pruned_loss=0.102, over 5675245.37 frames. ], batch size: 112, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:48:34,793 INFO [optim.py:369] (1/2) 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 16:48:52,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1086882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:49:06,456 INFO [train.py:968] (1/2) Epoch 24, batch 38000, giga_loss[loss=0.31, simple_loss=0.3791, pruned_loss=0.1204, over 27649.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3505, pruned_loss=0.1009, over 5686103.89 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3592, pruned_loss=0.1129, over 5671390.16 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3487, pruned_loss=0.09954, over 5681960.26 frames. ], batch size: 472, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:49:35,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4181, 3.7493, 1.5454, 1.5777], device='cuda:1'), covar=tensor([0.1083, 0.0290, 0.0914, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0555, 0.0392, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 16:49:46,210 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4087, 4.4352, 1.5982, 1.6002], device='cuda:1'), covar=tensor([0.1085, 0.0276, 0.0938, 0.1426], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0555, 0.0392, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 16:49:47,811 INFO [train.py:968] (1/2) Epoch 24, batch 38050, giga_loss[loss=0.2469, simple_loss=0.3372, pruned_loss=0.0783, over 28841.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3499, pruned_loss=0.1007, over 5688391.64 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5679133.50 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3476, pruned_loss=0.09863, over 5678426.71 frames. ], batch size: 174, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:49:58,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8959, 2.0968, 1.3786, 1.6805], device='cuda:1'), covar=tensor([0.1038, 0.0691, 0.1073, 0.1206], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0448, 0.0524, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 16:49:59,847 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 16:50:02,952 INFO [optim.py:369] (1/2) 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 16:50:04,744 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-12 16:50:14,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2930, 1.4021, 1.4940, 1.0876], device='cuda:1'), covar=tensor([0.1907, 0.3203, 0.1527, 0.1627], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0707, 0.0964, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 16:50:16,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7213, 4.5597, 4.3307, 2.1136], device='cuda:1'), covar=tensor([0.0555, 0.0698, 0.0720, 0.1954], device='cuda:1'), in_proj_covar=tensor([0.1243, 0.1151, 0.0969, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 16:50:32,108 INFO [train.py:968] (1/2) Epoch 24, batch 38100, libri_loss[loss=0.3275, simple_loss=0.3948, pruned_loss=0.1301, over 29241.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3505, pruned_loss=0.1006, over 5686042.30 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1139, over 5673277.52 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3481, pruned_loss=0.09843, over 5684094.71 frames. ], batch size: 97, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:50:37,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:50:51,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 16:50:57,038 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:50:58,872 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:51:20,143 INFO [train.py:968] (1/2) Epoch 24, batch 38150, giga_loss[loss=0.2406, simple_loss=0.3179, pruned_loss=0.08167, over 28387.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.3527, pruned_loss=0.102, over 5684672.95 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3601, pruned_loss=0.1138, over 5673062.13 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3509, pruned_loss=0.1003, over 5683383.14 frames. ], batch size: 65, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:51:27,237 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:51:58,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3890, 3.6620, 1.5782, 1.6912], device='cuda:1'), covar=tensor([0.1065, 0.0263, 0.0865, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0557, 0.0393, 0.0432], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 16:52:00,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2366, 1.2249, 1.1713, 1.3832], device='cuda:1'), covar=tensor([0.0817, 0.0380, 0.0360, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 16:52:06,134 INFO [train.py:968] (1/2) Epoch 24, batch 38200, giga_loss[loss=0.2508, simple_loss=0.334, pruned_loss=0.08379, over 28683.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3544, pruned_loss=0.1033, over 5691300.37 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3602, pruned_loss=0.1139, over 5675620.57 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3528, pruned_loss=0.1019, over 5688220.71 frames. ], batch size: 262, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:52:09,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3533, 2.1324, 1.5795, 0.5684], device='cuda:1'), covar=tensor([0.4566, 0.3082, 0.3806, 0.5786], device='cuda:1'), in_proj_covar=tensor([0.1775, 0.1675, 0.1621, 0.1445], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 16:52:48,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:52:52,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:52:52,866 INFO [train.py:968] (1/2) Epoch 24, batch 38250, giga_loss[loss=0.2662, simple_loss=0.3395, pruned_loss=0.09648, over 28596.00 frames. ], tot_loss[loss=0.2819, simple_loss=0.3552, pruned_loss=0.1043, over 5695156.88 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1142, over 5679315.05 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3535, pruned_loss=0.1028, over 5689586.36 frames. ], batch size: 71, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:52:54,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087149.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:52:55,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:53:02,838 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:53:06,365 INFO [optim.py:369] (1/2) 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 16:53:11,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-12 16:53:20,126 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:53:28,827 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4973, 1.5753, 1.4512, 1.3020], device='cuda:1'), covar=tensor([0.2514, 0.2640, 0.2367, 0.2684], device='cuda:1'), in_proj_covar=tensor([0.2013, 0.1946, 0.1862, 0.2019], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 16:53:35,092 INFO [train.py:968] (1/2) Epoch 24, batch 38300, giga_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1104, over 28793.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3558, pruned_loss=0.105, over 5698546.54 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3608, pruned_loss=0.1143, over 5682523.33 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3542, pruned_loss=0.1035, over 5691612.39 frames. ], batch size: 186, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:53:37,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5391, 1.6481, 1.7556, 1.3490], device='cuda:1'), covar=tensor([0.1839, 0.2588, 0.1484, 0.1729], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0705, 0.0963, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 16:53:52,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:19,099 INFO [train.py:968] (1/2) Epoch 24, batch 38350, giga_loss[loss=0.2532, simple_loss=0.3428, pruned_loss=0.08179, over 28987.00 frames. ], tot_loss[loss=0.282, simple_loss=0.3556, pruned_loss=0.1041, over 5703927.58 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.361, pruned_loss=0.114, over 5687658.60 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3541, pruned_loss=0.1029, over 5694166.93 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:54:32,234 INFO [optim.py:369] (1/2) 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 16:54:34,900 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-12 16:54:36,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:49,520 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:51,369 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:54:54,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087292.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 16:54:58,165 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 38400, giga_loss[loss=0.2803, simple_loss=0.3618, pruned_loss=0.09941, over 28926.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3551, pruned_loss=0.1029, over 5709966.17 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.361, pruned_loss=0.1141, over 5689962.97 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3538, pruned_loss=0.1018, over 5700321.66 frames. ], batch size: 227, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:55:05,665 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:07,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:09,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 16:55:18,207 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:25,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087324.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 16:55:32,815 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:42,254 INFO [train.py:968] (1/2) Epoch 24, batch 38450, giga_loss[loss=0.2791, simple_loss=0.3573, pruned_loss=0.1004, over 29004.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3556, pruned_loss=0.1026, over 5707156.46 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3613, pruned_loss=0.1142, over 5689477.83 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3543, pruned_loss=0.1014, over 5700130.31 frames. ], batch size: 155, lr: 1.32e-03, grad_scale: 8.0
+2023-03-12 16:55:50,722 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:52,733 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:55,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:55:55,393 INFO [optim.py:369] (1/2) 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 16:56:19,747 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 38500, giga_loss[loss=0.2514, simple_loss=0.3327, pruned_loss=0.08509, over 28188.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3537, pruned_loss=0.1013, over 5696703.53 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1146, over 5679452.17 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3519, pruned_loss=0.09958, over 5700302.43 frames. ], batch size: 77, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:56:30,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1967, 1.5298, 1.6257, 1.3645], device='cuda:1'), covar=tensor([0.1628, 0.1323, 0.1824, 0.1416], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0747, 0.0717, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 16:57:03,486 INFO [train.py:968] (1/2) Epoch 24, batch 38550, giga_loss[loss=0.2661, simple_loss=0.3432, pruned_loss=0.09449, over 29027.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3519, pruned_loss=0.1006, over 5703021.14 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5680839.75 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3506, pruned_loss=0.099, over 5705293.30 frames. ], batch size: 164, lr: 1.32e-03, grad_scale: 4.0
+2023-03-12 16:57:16,471 INFO [optim.py:369] (1/2) 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 16:57:34,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1629, 1.4978, 1.4837, 1.3407], device='cuda:1'), covar=tensor([0.2015, 0.1699, 0.2479, 0.1941], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0746, 0.0717, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 16:57:42,723 INFO [train.py:968] (1/2) Epoch 24, batch 38600, giga_loss[loss=0.2656, simple_loss=0.3461, pruned_loss=0.0925, over 28882.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3505, pruned_loss=0.1002, over 5696804.76 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3621, pruned_loss=0.1149, over 5668594.20 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3485, pruned_loss=0.09813, over 5711169.38 frames. ], batch size: 174, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:58:20,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:58:25,059 INFO [train.py:968] (1/2) Epoch 24, batch 38650, giga_loss[loss=0.276, simple_loss=0.3472, pruned_loss=0.1024, over 28646.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3496, pruned_loss=0.1, over 5698289.96 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3622, pruned_loss=0.1149, over 5669631.76 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3478, pruned_loss=0.09807, over 5709419.33 frames. ], batch size: 307, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:58:34,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-12 16:58:39,848 INFO [optim.py:369] (1/2) 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 16:59:06,247 INFO [train.py:968] (1/2) Epoch 24, batch 38700, giga_loss[loss=0.2461, simple_loss=0.3341, pruned_loss=0.07898, over 29003.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3508, pruned_loss=0.1013, over 5697759.83 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5664832.43 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3488, pruned_loss=0.09914, over 5711635.15 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:59:26,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4246, 1.9283, 1.4538, 1.3190], device='cuda:1'), covar=tensor([0.2649, 0.2598, 0.3030, 0.2427], device='cuda:1'), in_proj_covar=tensor([0.1552, 0.1121, 0.1365, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 16:59:41,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:59:42,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 16:59:44,591 INFO [train.py:968] (1/2) Epoch 24, batch 38750, giga_loss[loss=0.2598, simple_loss=0.3526, pruned_loss=0.08352, over 28956.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3514, pruned_loss=0.1011, over 5705619.72 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3624, pruned_loss=0.1151, over 5668575.93 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3497, pruned_loss=0.09927, over 5713638.99 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 16:59:49,830 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 16:59:57,174 INFO [optim.py:369] (1/2) 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 17:00:12,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1718, 5.9894, 5.7127, 3.3692], device='cuda:1'), covar=tensor([0.0418, 0.0570, 0.0653, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.1241, 0.1148, 0.0969, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 17:00:21,537 INFO [train.py:968] (1/2) Epoch 24, batch 38800, giga_loss[loss=0.2346, simple_loss=0.3197, pruned_loss=0.07474, over 29025.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3493, pruned_loss=0.09921, over 5698394.70 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3614, pruned_loss=0.1145, over 5665540.49 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3484, pruned_loss=0.09782, over 5708817.00 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:00:48,655 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 38850, giga_loss[loss=0.2842, simple_loss=0.3593, pruned_loss=0.1046, over 28739.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3487, pruned_loss=0.09902, over 5709570.28 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3613, pruned_loss=0.1146, over 5671263.94 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3477, pruned_loss=0.09733, over 5713888.13 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:01:15,742 INFO [optim.py:369] (1/2) 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 17:01:33,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:01:35,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:01:43,319 INFO [train.py:968] (1/2) Epoch 24, batch 38900, giga_loss[loss=0.2457, simple_loss=0.3264, pruned_loss=0.08249, over 29087.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09868, over 5708023.26 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5674726.27 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3467, pruned_loss=0.09726, over 5708959.88 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:01:58,947 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:02:21,282 INFO [train.py:968] (1/2) Epoch 24, batch 38950, giga_loss[loss=0.2677, simple_loss=0.3427, pruned_loss=0.09636, over 28897.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.345, pruned_loss=0.09765, over 5700193.18 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3612, pruned_loss=0.1145, over 5671238.74 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3438, pruned_loss=0.09592, over 5705442.00 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:02:36,725 INFO [optim.py:369] (1/2) 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 17:02:43,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1087875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:02:45,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1087878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:00,507 INFO [train.py:968] (1/2) Epoch 24, batch 39000, giga_loss[loss=0.2532, simple_loss=0.3256, pruned_loss=0.09042, over 28549.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3424, pruned_loss=0.09633, over 5708873.33 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3616, pruned_loss=0.1147, over 5679756.60 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3405, pruned_loss=0.09414, over 5706479.59 frames. ], batch size: 85, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:03:00,507 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 17:03:09,471 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 17:03:16,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1087907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:24,329 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1087918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:25,236 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1087919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:03:49,281 INFO [train.py:968] (1/2) Epoch 24, batch 39050, giga_loss[loss=0.2815, simple_loss=0.3509, pruned_loss=0.1061, over 28792.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3407, pruned_loss=0.09552, over 5712147.87 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 5685560.66 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3395, pruned_loss=0.09378, over 5705607.91 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:04:07,699 INFO [optim.py:369] (1/2) 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 17:04:34,259 INFO [train.py:968] (1/2) Epoch 24, batch 39100, libri_loss[loss=0.3124, simple_loss=0.3845, pruned_loss=0.1202, over 29271.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.341, pruned_loss=0.09634, over 5708056.94 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3611, pruned_loss=0.1143, over 5689505.63 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3392, pruned_loss=0.09438, over 5699797.49 frames. ], batch size: 94, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:04:51,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:15,021 INFO [train.py:968] (1/2) Epoch 24, batch 39150, giga_loss[loss=0.221, simple_loss=0.2995, pruned_loss=0.07124, over 28444.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3381, pruned_loss=0.09481, over 5701185.90 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1146, over 5680064.57 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3359, pruned_loss=0.09274, over 5702956.95 frames. ], batch size: 65, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:05:16,467 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9471, 2.0054, 2.0963, 1.6832], device='cuda:1'), covar=tensor([0.1974, 0.2484, 0.1569, 0.1759], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0706, 0.0963, 0.0860], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 17:05:27,334 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:30,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:31,293 INFO [optim.py:369] (1/2) 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 17:05:34,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5078, 1.8526, 1.4520, 1.3518], device='cuda:1'), covar=tensor([0.2571, 0.2632, 0.2995, 0.2296], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1118, 0.1361, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 17:05:52,130 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:05:57,133 INFO [train.py:968] (1/2) Epoch 24, batch 39200, giga_loss[loss=0.2609, simple_loss=0.3421, pruned_loss=0.08979, over 28644.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3356, pruned_loss=0.09371, over 5708070.79 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1146, over 5682326.33 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3337, pruned_loss=0.09189, over 5707698.64 frames. ], batch size: 307, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:06:31,706 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-12 17:06:38,743 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:06:42,413 INFO [train.py:968] (1/2) Epoch 24, batch 39250, giga_loss[loss=0.3162, simple_loss=0.3654, pruned_loss=0.1335, over 23665.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3338, pruned_loss=0.09315, over 5699462.83 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1149, over 5683054.37 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3315, pruned_loss=0.09108, over 5699056.79 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:06:52,093 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:06:56,072 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:07:06,072 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:07:08,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1956, 1.3077, 1.1480, 0.9260], device='cuda:1'), covar=tensor([0.0987, 0.0525, 0.1078, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0448, 0.0523, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 17:07:21,732 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:07:25,108 INFO [train.py:968] (1/2) Epoch 24, batch 39300, giga_loss[loss=0.2771, simple_loss=0.3589, pruned_loss=0.09762, over 28292.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3326, pruned_loss=0.09211, over 5706289.80 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5684627.89 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3306, pruned_loss=0.09026, over 5704797.22 frames. ], batch size: 368, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:08:10,522 INFO [train.py:968] (1/2) Epoch 24, batch 39350, giga_loss[loss=0.243, simple_loss=0.3167, pruned_loss=0.08465, over 28881.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3361, pruned_loss=0.09355, over 5703400.23 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1146, over 5685739.57 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3338, pruned_loss=0.09164, over 5701506.15 frames. ], batch size: 106, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:08:29,539 INFO [optim.py:369] (1/2) 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 17:08:48,019 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 17:08:57,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:08:59,891 INFO [train.py:968] (1/2) Epoch 24, batch 39400, libri_loss[loss=0.3549, simple_loss=0.4114, pruned_loss=0.1492, over 29193.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3403, pruned_loss=0.09597, over 5683674.02 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3623, pruned_loss=0.1151, over 5676638.14 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3375, pruned_loss=0.09372, over 5689562.38 frames. ], batch size: 97, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:09:12,013 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:09:30,691 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:09:44,317 INFO [train.py:968] (1/2) Epoch 24, batch 39450, giga_loss[loss=0.3051, simple_loss=0.3718, pruned_loss=0.1192, over 29054.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3417, pruned_loss=0.09575, over 5693265.56 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.362, pruned_loss=0.1149, over 5681328.96 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3394, pruned_loss=0.09373, over 5694026.89 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:10:02,614 INFO [optim.py:369] (1/2) 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 17:10:28,097 INFO [train.py:968] (1/2) Epoch 24, batch 39500, giga_loss[loss=0.25, simple_loss=0.3354, pruned_loss=0.08231, over 28629.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3434, pruned_loss=0.0964, over 5681089.12 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1154, over 5671939.31 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3407, pruned_loss=0.09402, over 5691255.98 frames. ], batch size: 336, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:10:41,103 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:10:52,431 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-12 17:11:02,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:11:04,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:11:10,330 INFO [train.py:968] (1/2) Epoch 24, batch 39550, giga_loss[loss=0.2536, simple_loss=0.3229, pruned_loss=0.09219, over 28588.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3423, pruned_loss=0.09579, over 5692591.78 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1151, over 5677240.90 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3402, pruned_loss=0.09378, over 5696477.32 frames. ], batch size: 85, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:11:27,038 INFO [optim.py:369] (1/2) 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 17:11:28,600 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:11:55,436 INFO [train.py:968] (1/2) Epoch 24, batch 39600, giga_loss[loss=0.3291, simple_loss=0.3964, pruned_loss=0.1309, over 28312.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3411, pruned_loss=0.09488, over 5696405.98 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1151, over 5677240.90 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3394, pruned_loss=0.09332, over 5699430.14 frames. ], batch size: 368, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:12:12,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:12:19,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4754, 1.9621, 5.6084, 4.2307], device='cuda:1'), covar=tensor([0.1479, 0.2532, 0.0673, 0.0730], device='cuda:1'), in_proj_covar=tensor([0.0770, 0.0655, 0.0967, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 17:12:21,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5190, 1.7735, 1.4342, 1.7230], device='cuda:1'), covar=tensor([0.2688, 0.2773, 0.3187, 0.2502], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1119, 0.1364, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 17:12:38,709 INFO [train.py:968] (1/2) Epoch 24, batch 39650, giga_loss[loss=0.2714, simple_loss=0.339, pruned_loss=0.1019, over 28835.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3423, pruned_loss=0.0961, over 5711861.54 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3625, pruned_loss=0.1153, over 5680980.94 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3402, pruned_loss=0.09436, over 5711348.85 frames. ], batch size: 112, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:12:41,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:12:55,504 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 39700, giga_loss[loss=0.3109, simple_loss=0.3777, pruned_loss=0.1221, over 29053.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3452, pruned_loss=0.09745, over 5710904.25 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3623, pruned_loss=0.1152, over 5684620.37 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3435, pruned_loss=0.09595, over 5707655.87 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:14:04,074 INFO [train.py:968] (1/2) Epoch 24, batch 39750, giga_loss[loss=0.2422, simple_loss=0.3253, pruned_loss=0.07954, over 28826.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.349, pruned_loss=0.09936, over 5693970.25 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5670618.55 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3472, pruned_loss=0.09765, over 5705176.43 frames. ], batch size: 112, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:14:16,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:18,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:19,542 INFO [optim.py:369] (1/2) 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 17:14:33,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:34,675 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:43,271 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:43,845 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:45,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:14:45,608 INFO [train.py:968] (1/2) Epoch 24, batch 39800, giga_loss[loss=0.296, simple_loss=0.3693, pruned_loss=0.1113, over 28622.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5702024.65 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3622, pruned_loss=0.115, over 5676908.27 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3491, pruned_loss=0.09877, over 5705917.55 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:14:52,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:15:00,073 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:15:07,199 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:15:19,651 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-12 17:15:27,563 INFO [train.py:968] (1/2) Epoch 24, batch 39850, giga_loss[loss=0.2709, simple_loss=0.3503, pruned_loss=0.09572, over 29067.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3518, pruned_loss=0.1007, over 5705997.26 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.1151, over 5681674.49 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3504, pruned_loss=0.09918, over 5705525.02 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:15:44,727 INFO [optim.py:369] (1/2) 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 17:16:03,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1088790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:09,493 INFO [train.py:968] (1/2) Epoch 24, batch 39900, libri_loss[loss=0.3065, simple_loss=0.3774, pruned_loss=0.1178, over 29201.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3539, pruned_loss=0.1023, over 5711029.02 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5688201.33 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3521, pruned_loss=0.1004, over 5705567.98 frames. ], batch size: 97, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 17:16:22,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2521, 1.5576, 1.5420, 1.2062], device='cuda:1'), covar=tensor([0.2345, 0.2028, 0.1401, 0.1910], device='cuda:1'), in_proj_covar=tensor([0.2010, 0.1948, 0.1862, 0.2014], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 17:16:29,881 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7883, 4.5965, 4.3954, 2.1285], device='cuda:1'), covar=tensor([0.0575, 0.0772, 0.0752, 0.2006], device='cuda:1'), in_proj_covar=tensor([0.1249, 0.1155, 0.0973, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 17:16:33,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:36,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:37,741 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:48,304 INFO [train.py:968] (1/2) Epoch 24, batch 39950, giga_loss[loss=0.2689, simple_loss=0.3378, pruned_loss=0.1, over 28562.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3537, pruned_loss=0.102, over 5716214.86 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5693670.71 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3515, pruned_loss=0.09991, over 5707544.12 frames. ], batch size: 78, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 17:16:50,021 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:52,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:16:58,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:04,478 INFO [optim.py:369] (1/2) 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 17:17:14,242 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:25,919 INFO [train.py:968] (1/2) Epoch 24, batch 40000, giga_loss[loss=0.2507, simple_loss=0.3353, pruned_loss=0.083, over 28852.00 frames. ], tot_loss[loss=0.2801, simple_loss=0.3541, pruned_loss=0.103, over 5713773.34 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5691222.70 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.351, pruned_loss=0.1002, over 5709819.33 frames. ], batch size: 174, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:17:26,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088898.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:54,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1088933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:17:57,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1088936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:18:04,952 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1088945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:18:06,423 INFO [train.py:968] (1/2) Epoch 24, batch 40050, giga_loss[loss=0.2878, simple_loss=0.3544, pruned_loss=0.1106, over 28720.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3511, pruned_loss=0.1019, over 5715429.04 frames. ], libri_tot_loss[loss=0.2983, simple_loss=0.3643, pruned_loss=0.1162, over 5695687.77 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.348, pruned_loss=0.09892, over 5708935.62 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:18:19,736 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1088965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:18:22,762 INFO [optim.py:369] (1/2) 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 17:18:25,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2419, 4.0971, 1.5423, 1.4435], device='cuda:1'), covar=tensor([0.1056, 0.0303, 0.0969, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0556, 0.0394, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 17:18:47,827 INFO [train.py:968] (1/2) Epoch 24, batch 40100, giga_loss[loss=0.2419, simple_loss=0.3146, pruned_loss=0.08463, over 28702.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3467, pruned_loss=0.09908, over 5719860.91 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.364, pruned_loss=0.1159, over 5700592.82 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.344, pruned_loss=0.09664, over 5710770.11 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:19:07,508 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9955, 1.1609, 1.1358, 0.9485], device='cuda:1'), covar=tensor([0.2585, 0.2768, 0.1540, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.2007, 0.1943, 0.1860, 0.2011], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 17:19:31,281 INFO [train.py:968] (1/2) Epoch 24, batch 40150, giga_loss[loss=0.2782, simple_loss=0.3637, pruned_loss=0.09638, over 28926.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3468, pruned_loss=0.09752, over 5723819.21 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3637, pruned_loss=0.1157, over 5702462.47 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3449, pruned_loss=0.09562, over 5715302.82 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:19:40,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:19:46,958 INFO [optim.py:369] (1/2) 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 17:20:08,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:20:08,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3801, 1.4936, 1.1740, 1.1263], device='cuda:1'), covar=tensor([0.0968, 0.0574, 0.1076, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0447, 0.0521, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 17:20:13,152 INFO [train.py:968] (1/2) Epoch 24, batch 40200, giga_loss[loss=0.2661, simple_loss=0.3432, pruned_loss=0.09449, over 28791.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3481, pruned_loss=0.09713, over 5715781.39 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3641, pruned_loss=0.1159, over 5706351.75 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3459, pruned_loss=0.09507, over 5705835.33 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:20:53,742 INFO [train.py:968] (1/2) Epoch 24, batch 40250, giga_loss[loss=0.2616, simple_loss=0.3375, pruned_loss=0.09281, over 28820.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3465, pruned_loss=0.09644, over 5715190.53 frames. ], libri_tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5706271.75 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3442, pruned_loss=0.0944, over 5707884.84 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:20:55,218 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:11,454 INFO [optim.py:369] (1/2) 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 17:21:21,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-12 17:21:35,454 INFO [train.py:968] (1/2) Epoch 24, batch 40300, giga_loss[loss=0.2786, simple_loss=0.3542, pruned_loss=0.1014, over 28555.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3468, pruned_loss=0.09802, over 5716074.93 frames. ], libri_tot_loss[loss=0.2985, simple_loss=0.3647, pruned_loss=0.1161, over 5712468.49 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3441, pruned_loss=0.09566, over 5704989.36 frames. ], batch size: 336, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:21:38,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4568, 1.5540, 1.2126, 1.1830], device='cuda:1'), covar=tensor([0.0915, 0.0540, 0.1075, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0399, 0.0446, 0.0521, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 17:21:39,592 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:41,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:44,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:21:50,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5079, 1.6133, 1.7429, 1.3097], device='cuda:1'), covar=tensor([0.1850, 0.2668, 0.1592, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0705, 0.0960, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 17:21:55,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8492, 2.7907, 1.7067, 1.0156], device='cuda:1'), covar=tensor([0.8529, 0.3680, 0.4527, 0.7747], device='cuda:1'), in_proj_covar=tensor([0.1785, 0.1673, 0.1620, 0.1449], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:22:07,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:22:08,032 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:22:11,210 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 40350, giga_loss[loss=0.251, simple_loss=0.3252, pruned_loss=0.08837, over 28807.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3445, pruned_loss=0.09795, over 5712947.20 frames. ], libri_tot_loss[loss=0.2981, simple_loss=0.3644, pruned_loss=0.1159, over 5714397.05 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3424, pruned_loss=0.09616, over 5702518.52 frames. ], batch size: 284, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:22:28,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4227, 1.2946, 4.5469, 3.4498], device='cuda:1'), covar=tensor([0.1727, 0.2926, 0.0388, 0.0996], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0657, 0.0969, 0.0932], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 17:22:37,968 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:22:38,852 INFO [optim.py:369] (1/2) 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 17:22:42,965 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:01,229 INFO [train.py:968] (1/2) Epoch 24, batch 40400, giga_loss[loss=0.2394, simple_loss=0.3128, pruned_loss=0.08302, over 29041.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3439, pruned_loss=0.09901, over 5722008.96 frames. ], libri_tot_loss[loss=0.2984, simple_loss=0.3646, pruned_loss=0.1161, over 5718930.90 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3414, pruned_loss=0.09686, over 5709587.83 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:23:20,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:42,198 INFO [train.py:968] (1/2) Epoch 24, batch 40450, giga_loss[loss=0.2597, simple_loss=0.3239, pruned_loss=0.09776, over 28666.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3426, pruned_loss=0.09854, over 5727989.75 frames. ], libri_tot_loss[loss=0.2988, simple_loss=0.365, pruned_loss=0.1163, over 5722864.07 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.34, pruned_loss=0.09633, over 5714711.53 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:23:46,120 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:23:49,046 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:24:10,928 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 40500, giga_loss[loss=0.2328, simple_loss=0.3113, pruned_loss=0.07718, over 29096.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3394, pruned_loss=0.0963, over 5734567.58 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3643, pruned_loss=0.1156, over 5728822.59 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3371, pruned_loss=0.09446, over 5718642.21 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:24:36,438 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:24:38,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:01,967 INFO [train.py:968] (1/2) Epoch 24, batch 40550, giga_loss[loss=0.2083, simple_loss=0.2876, pruned_loss=0.06454, over 28945.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3349, pruned_loss=0.09412, over 5729605.53 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3642, pruned_loss=0.1156, over 5729854.92 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3325, pruned_loss=0.09223, over 5715891.40 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:25:04,080 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:11,442 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:14,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:17,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:20,108 INFO [optim.py:369] (1/2) 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 17:25:33,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3114, 4.1621, 3.9127, 1.9840], device='cuda:1'), covar=tensor([0.0576, 0.0692, 0.0647, 0.2099], device='cuda:1'), in_proj_covar=tensor([0.1255, 0.1162, 0.0977, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 17:25:40,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2546, 4.1112, 3.8561, 1.8601], device='cuda:1'), covar=tensor([0.0593, 0.0708, 0.0654, 0.2026], device='cuda:1'), in_proj_covar=tensor([0.1256, 0.1163, 0.0977, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 17:25:40,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3236, 2.0235, 1.5454, 0.6395], device='cuda:1'), covar=tensor([0.6023, 0.3146, 0.4869, 0.6961], device='cuda:1'), in_proj_covar=tensor([0.1788, 0.1676, 0.1623, 0.1453], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:25:42,295 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:25:43,938 INFO [train.py:968] (1/2) Epoch 24, batch 40600, giga_loss[loss=0.2537, simple_loss=0.3272, pruned_loss=0.09008, over 29010.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3312, pruned_loss=0.09221, over 5730417.22 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.364, pruned_loss=0.1156, over 5733124.76 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3284, pruned_loss=0.0899, over 5716008.73 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:25:44,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5366, 1.7279, 1.7064, 1.4317], device='cuda:1'), covar=tensor([0.2323, 0.2126, 0.1536, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.2017, 0.1951, 0.1872, 0.2017], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 17:26:05,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:26:08,788 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:26:23,147 INFO [train.py:968] (1/2) Epoch 24, batch 40650, giga_loss[loss=0.2797, simple_loss=0.3498, pruned_loss=0.1048, over 27561.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3311, pruned_loss=0.09197, over 5726473.44 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3639, pruned_loss=0.1157, over 5734402.60 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3279, pruned_loss=0.08929, over 5713876.81 frames. ], batch size: 472, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:26:42,705 INFO [optim.py:369] (1/2) 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 17:27:05,530 INFO [train.py:968] (1/2) Epoch 24, batch 40700, giga_loss[loss=0.2947, simple_loss=0.3674, pruned_loss=0.111, over 28295.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3333, pruned_loss=0.09267, over 5715973.32 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3636, pruned_loss=0.1156, over 5727807.72 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3302, pruned_loss=0.09006, over 5711650.60 frames. ], batch size: 368, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:27:46,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:27:49,373 INFO [train.py:968] (1/2) Epoch 24, batch 40750, giga_loss[loss=0.3094, simple_loss=0.3784, pruned_loss=0.1202, over 28979.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.336, pruned_loss=0.0936, over 5714169.83 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5733034.03 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3338, pruned_loss=0.0916, over 5705337.11 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:28:04,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:06,581 INFO [optim.py:369] (1/2) 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 17:28:06,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:29,501 INFO [train.py:968] (1/2) Epoch 24, batch 40800, giga_loss[loss=0.2765, simple_loss=0.3594, pruned_loss=0.09681, over 28692.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3386, pruned_loss=0.09392, over 5721223.26 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3624, pruned_loss=0.1147, over 5733892.08 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3368, pruned_loss=0.0923, over 5713421.34 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:28:30,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:55,599 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:28:57,257 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.05 vs. limit=5.0
+2023-03-12 17:29:12,334 INFO [train.py:968] (1/2) Epoch 24, batch 40850, giga_loss[loss=0.2552, simple_loss=0.3403, pruned_loss=0.08504, over 28920.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3412, pruned_loss=0.0953, over 5721652.33 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1143, over 5736164.33 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3396, pruned_loss=0.09393, over 5713149.47 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:29:28,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6546, 1.7539, 1.8509, 1.4184], device='cuda:1'), covar=tensor([0.1708, 0.2460, 0.1425, 0.1617], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0707, 0.0962, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 17:29:32,262 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 40900, giga_loss[loss=0.2508, simple_loss=0.3329, pruned_loss=0.08431, over 28871.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3434, pruned_loss=0.09688, over 5719197.99 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3617, pruned_loss=0.114, over 5741987.40 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3415, pruned_loss=0.09517, over 5706234.88 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:30:34,061 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:30:51,758 INFO [train.py:968] (1/2) Epoch 24, batch 40950, giga_loss[loss=0.3238, simple_loss=0.3846, pruned_loss=0.1315, over 28603.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3515, pruned_loss=0.1043, over 5696786.49 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3617, pruned_loss=0.114, over 5741987.40 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.35, pruned_loss=0.103, over 5686697.13 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:31:14,328 INFO [optim.py:369] (1/2) 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 17:31:16,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:20,510 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6977, 1.6500, 1.9218, 1.5096], device='cuda:1'), covar=tensor([0.1460, 0.1959, 0.1193, 0.1588], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0706, 0.0961, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 17:31:21,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:41,167 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:41,562 INFO [train.py:968] (1/2) Epoch 24, batch 41000, giga_loss[loss=0.3469, simple_loss=0.4126, pruned_loss=0.1406, over 29040.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3577, pruned_loss=0.109, over 5682060.22 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3614, pruned_loss=0.1138, over 5733745.53 frames. ], giga_tot_loss[loss=0.2864, simple_loss=0.3567, pruned_loss=0.1081, over 5679976.67 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:31:45,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1089903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:31:46,840 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1089905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:32:25,618 INFO [train.py:968] (1/2) Epoch 24, batch 41050, giga_loss[loss=0.3161, simple_loss=0.3846, pruned_loss=0.1238, over 28949.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3635, pruned_loss=0.1133, over 5683752.95 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.361, pruned_loss=0.1137, over 5737275.97 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.363, pruned_loss=0.1126, over 5677245.17 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:32:46,781 INFO [optim.py:369] (1/2) 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 17:32:52,580 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1089976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:32:54,361 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:32:54,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1089979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:03,215 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1089989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:10,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5389, 2.1886, 1.5948, 0.7902], device='cuda:1'), covar=tensor([0.5483, 0.2920, 0.4125, 0.6163], device='cuda:1'), in_proj_covar=tensor([0.1791, 0.1687, 0.1627, 0.1454], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:33:12,427 INFO [train.py:968] (1/2) Epoch 24, batch 41100, giga_loss[loss=0.3369, simple_loss=0.3999, pruned_loss=0.137, over 28919.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3695, pruned_loss=0.1184, over 5664742.29 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3611, pruned_loss=0.1137, over 5721822.72 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3692, pruned_loss=0.1179, over 5671015.89 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:33:20,719 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:31,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:33:55,694 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 41150, giga_loss[loss=0.3855, simple_loss=0.4273, pruned_loss=0.1719, over 27894.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3764, pruned_loss=0.124, over 5671431.49 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3612, pruned_loss=0.1136, over 5726877.78 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3763, pruned_loss=0.1239, over 5670600.43 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:33:57,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:34:21,513 INFO [optim.py:369] (1/2) 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 17:34:30,124 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:34:48,759 INFO [train.py:968] (1/2) Epoch 24, batch 41200, giga_loss[loss=0.3076, simple_loss=0.3757, pruned_loss=0.1198, over 28989.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3794, pruned_loss=0.1271, over 5654344.52 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.361, pruned_loss=0.1134, over 5728309.17 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.38, pruned_loss=0.1275, over 5650371.74 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:35:03,763 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090111.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 17:35:09,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 17:35:41,793 INFO [train.py:968] (1/2) Epoch 24, batch 41250, giga_loss[loss=0.2843, simple_loss=0.3599, pruned_loss=0.1043, over 28996.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3809, pruned_loss=0.129, over 5652191.79 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3606, pruned_loss=0.1132, over 5727794.58 frames. ], giga_tot_loss[loss=0.3211, simple_loss=0.3823, pruned_loss=0.13, over 5647651.36 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:35:59,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:36:03,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:36:07,663 INFO [optim.py:369] (1/2) 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 17:36:32,700 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 41300, giga_loss[loss=0.3039, simple_loss=0.3727, pruned_loss=0.1175, over 28875.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3831, pruned_loss=0.1319, over 5635589.02 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5729531.00 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.3856, pruned_loss=0.1336, over 5627081.06 frames. ], batch size: 112, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:36:38,338 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8301, 2.1191, 2.0266, 1.8263], device='cuda:1'), covar=tensor([0.1613, 0.1619, 0.1791, 0.1686], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0750, 0.0720, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 17:37:27,062 INFO [train.py:968] (1/2) Epoch 24, batch 41350, giga_loss[loss=0.4838, simple_loss=0.471, pruned_loss=0.2483, over 23540.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3866, pruned_loss=0.1358, over 5632672.45 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3595, pruned_loss=0.1125, over 5730495.99 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3896, pruned_loss=0.1379, over 5622762.70 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:37:50,342 INFO [optim.py:369] (1/2) 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 17:37:53,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:37:57,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9657, 1.3163, 1.0976, 0.2089], device='cuda:1'), covar=tensor([0.4183, 0.3257, 0.4450, 0.6687], device='cuda:1'), in_proj_covar=tensor([0.1793, 0.1690, 0.1629, 0.1454], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:38:17,364 INFO [train.py:968] (1/2) Epoch 24, batch 41400, giga_loss[loss=0.3719, simple_loss=0.4099, pruned_loss=0.1669, over 27928.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3897, pruned_loss=0.1378, over 5633035.03 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3595, pruned_loss=0.1124, over 5724202.46 frames. ], giga_tot_loss[loss=0.3368, simple_loss=0.393, pruned_loss=0.1403, over 5627171.80 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:38:56,693 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:39:11,557 INFO [train.py:968] (1/2) Epoch 24, batch 41450, giga_loss[loss=0.3182, simple_loss=0.3783, pruned_loss=0.1291, over 28646.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3892, pruned_loss=0.1382, over 5636132.72 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1123, over 5725155.72 frames. ], giga_tot_loss[loss=0.3371, simple_loss=0.3927, pruned_loss=0.1408, over 5628502.86 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:39:14,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-12 17:39:20,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:39:29,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:39:31,340 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4100, 1.5996, 3.1368, 2.9974], device='cuda:1'), covar=tensor([0.1185, 0.2112, 0.0480, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0660, 0.0978, 0.0937], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 17:39:35,534 INFO [optim.py:369] (1/2) 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 17:40:02,495 INFO [train.py:968] (1/2) Epoch 24, batch 41500, giga_loss[loss=0.4222, simple_loss=0.4465, pruned_loss=0.199, over 26657.00 frames. ], tot_loss[loss=0.3311, simple_loss=0.3873, pruned_loss=0.1374, over 5624616.07 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1125, over 5718449.79 frames. ], giga_tot_loss[loss=0.3353, simple_loss=0.3907, pruned_loss=0.14, over 5622724.77 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:40:21,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:40:24,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:40:53,520 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:40:53,851 INFO [train.py:968] (1/2) Epoch 24, batch 41550, giga_loss[loss=0.3525, simple_loss=0.3879, pruned_loss=0.1585, over 23494.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3876, pruned_loss=0.1369, over 5619326.15 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5711727.49 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3906, pruned_loss=0.1393, over 5622109.86 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 17:41:21,079 INFO [optim.py:369] (1/2) 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 17:41:36,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090486.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 17:41:49,435 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:41:49,840 INFO [train.py:968] (1/2) Epoch 24, batch 41600, libri_loss[loss=0.2449, simple_loss=0.32, pruned_loss=0.08491, over 29652.00 frames. ], tot_loss[loss=0.331, simple_loss=0.3879, pruned_loss=0.137, over 5609025.93 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5715873.16 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3912, pruned_loss=0.1396, over 5605788.02 frames. ], batch size: 73, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:41:51,464 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:41:57,679 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:02,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:16,732 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9673, 2.5311, 0.9744, 1.2979], device='cuda:1'), covar=tensor([0.1302, 0.0485, 0.1041, 0.1626], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0561, 0.0395, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 17:42:22,848 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:36,892 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:42:44,243 INFO [train.py:968] (1/2) Epoch 24, batch 41650, giga_loss[loss=0.3236, simple_loss=0.3865, pruned_loss=0.1303, over 28736.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3885, pruned_loss=0.1374, over 5596173.19 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5717657.96 frames. ], giga_tot_loss[loss=0.3356, simple_loss=0.3916, pruned_loss=0.1398, over 5590430.22 frames. ], batch size: 284, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:42:57,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:43:10,698 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 41700, giga_loss[loss=0.3105, simple_loss=0.3837, pruned_loss=0.1187, over 28840.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3854, pruned_loss=0.1342, over 5595291.04 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1127, over 5700238.59 frames. ], giga_tot_loss[loss=0.3304, simple_loss=0.3881, pruned_loss=0.1364, over 5603582.03 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:43:46,932 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1090608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:43:57,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6040, 1.7794, 1.7481, 1.5254], device='cuda:1'), covar=tensor([0.3093, 0.2528, 0.2164, 0.2476], device='cuda:1'), in_proj_covar=tensor([0.2025, 0.1956, 0.1869, 0.2018], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 17:44:05,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090629.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 17:44:10,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090632.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 17:44:24,668 INFO [train.py:968] (1/2) Epoch 24, batch 41750, giga_loss[loss=0.2916, simple_loss=0.3672, pruned_loss=0.1081, over 28948.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3832, pruned_loss=0.1312, over 5598409.56 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3594, pruned_loss=0.1128, over 5684257.22 frames. ], giga_tot_loss[loss=0.3264, simple_loss=0.386, pruned_loss=0.1334, over 5617902.27 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:44:37,374 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090661.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 17:44:40,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2607, 1.8903, 1.4742, 0.4590], device='cuda:1'), covar=tensor([0.4656, 0.3186, 0.4271, 0.6470], device='cuda:1'), in_proj_covar=tensor([0.1788, 0.1684, 0.1625, 0.1452], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:44:49,226 INFO [optim.py:369] (1/2) 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 17:44:52,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6122, 1.7944, 1.8335, 1.3672], device='cuda:1'), covar=tensor([0.1702, 0.2863, 0.1550, 0.2003], device='cuda:1'), in_proj_covar=tensor([0.0910, 0.0705, 0.0955, 0.0856], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 17:45:03,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6729, 2.3931, 2.0197, 1.8884], device='cuda:1'), covar=tensor([0.0784, 0.0241, 0.0259, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 17:45:16,986 INFO [train.py:968] (1/2) Epoch 24, batch 41800, giga_loss[loss=0.2915, simple_loss=0.3592, pruned_loss=0.1119, over 29002.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.38, pruned_loss=0.1283, over 5609180.44 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1128, over 5688704.78 frames. ], giga_tot_loss[loss=0.3217, simple_loss=0.3827, pruned_loss=0.1303, over 5619367.79 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:45:26,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:46:09,194 INFO [train.py:968] (1/2) Epoch 24, batch 41850, libri_loss[loss=0.2494, simple_loss=0.3234, pruned_loss=0.08774, over 29542.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.378, pruned_loss=0.1267, over 5600667.11 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1126, over 5686862.23 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3809, pruned_loss=0.1288, over 5608653.50 frames. ], batch size: 77, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:46:30,081 INFO [optim.py:369] (1/2) 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 17:46:55,376 INFO [train.py:968] (1/2) Epoch 24, batch 41900, giga_loss[loss=0.2592, simple_loss=0.3348, pruned_loss=0.09177, over 28783.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3739, pruned_loss=0.1234, over 5628148.66 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3588, pruned_loss=0.1125, over 5689643.97 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3771, pruned_loss=0.1256, over 5629574.92 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:47:19,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5000, 4.2952, 4.1157, 1.9892], device='cuda:1'), covar=tensor([0.0731, 0.0939, 0.1046, 0.2087], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1176, 0.0989, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 17:47:44,258 INFO [train.py:968] (1/2) Epoch 24, batch 41950, giga_loss[loss=0.3253, simple_loss=0.3909, pruned_loss=0.1299, over 28951.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3728, pruned_loss=0.1224, over 5629625.40 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3584, pruned_loss=0.1123, over 5685899.30 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3761, pruned_loss=0.1246, over 5633416.41 frames. ], batch size: 106, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:47:46,694 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1090851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:47:48,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1090854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:48:06,796 INFO [optim.py:369] (1/2) 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 17:48:18,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1090883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:48:36,299 INFO [train.py:968] (1/2) Epoch 24, batch 42000, giga_loss[loss=0.321, simple_loss=0.3873, pruned_loss=0.1273, over 28995.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3712, pruned_loss=0.1213, over 5635473.01 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3578, pruned_loss=0.112, over 5691515.23 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3747, pruned_loss=0.1236, over 5632288.16 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:48:36,299 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 17:48:45,309 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 17:48:55,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3063, 1.8552, 1.3965, 0.5583], device='cuda:1'), covar=tensor([0.5765, 0.2627, 0.3388, 0.7029], device='cuda:1'), in_proj_covar=tensor([0.1788, 0.1680, 0.1620, 0.1451], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:49:28,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:49:35,485 INFO [train.py:968] (1/2) Epoch 24, batch 42050, giga_loss[loss=0.2532, simple_loss=0.3337, pruned_loss=0.08635, over 28937.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3679, pruned_loss=0.1184, over 5644560.80 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3568, pruned_loss=0.1115, over 5697824.96 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3721, pruned_loss=0.121, over 5634387.85 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:50:00,031 INFO [optim.py:369] (1/2) 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 17:50:09,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1090983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:50:25,097 INFO [train.py:968] (1/2) Epoch 24, batch 42100, giga_loss[loss=0.2997, simple_loss=0.3783, pruned_loss=0.1106, over 28806.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3703, pruned_loss=0.1178, over 5652956.27 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3573, pruned_loss=0.1119, over 5702832.53 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3738, pruned_loss=0.1199, over 5637752.42 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:50:57,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3410, 1.6444, 1.6010, 1.1498], device='cuda:1'), covar=tensor([0.1685, 0.2716, 0.1495, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.0912, 0.0706, 0.0955, 0.0857], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 17:51:07,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:51:18,749 INFO [train.py:968] (1/2) Epoch 24, batch 42150, giga_loss[loss=0.3207, simple_loss=0.3678, pruned_loss=0.1369, over 23549.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3728, pruned_loss=0.1183, over 5661086.63 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3573, pruned_loss=0.1119, over 5705741.05 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3757, pruned_loss=0.1201, over 5645823.79 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:51:39,993 INFO [optim.py:369] (1/2) 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 17:51:49,473 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:51:52,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:05,025 INFO [train.py:968] (1/2) Epoch 24, batch 42200, giga_loss[loss=0.3284, simple_loss=0.3863, pruned_loss=0.1353, over 28716.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3725, pruned_loss=0.1186, over 5671692.34 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3566, pruned_loss=0.1115, over 5707073.58 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.376, pruned_loss=0.1206, over 5657100.83 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:52:07,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1637, 1.5399, 1.1933, 0.6919], device='cuda:1'), covar=tensor([0.3422, 0.1991, 0.2440, 0.4884], device='cuda:1'), in_proj_covar=tensor([0.1779, 0.1673, 0.1617, 0.1447], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:52:22,439 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:33,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:36,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:52:48,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5445, 1.8214, 1.4141, 1.7445], device='cuda:1'), covar=tensor([0.2632, 0.2748, 0.3111, 0.2403], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1119, 0.1369, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 17:52:53,600 INFO [train.py:968] (1/2) Epoch 24, batch 42250, giga_loss[loss=0.3056, simple_loss=0.3717, pruned_loss=0.1198, over 29064.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.372, pruned_loss=0.1186, over 5670325.54 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3568, pruned_loss=0.1116, over 5711326.14 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3749, pruned_loss=0.1201, over 5654048.78 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:53:02,742 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:53:12,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6036, 1.8227, 1.3006, 1.3449], device='cuda:1'), covar=tensor([0.0963, 0.0569, 0.1014, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0453, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 17:53:15,951 INFO [optim.py:369] (1/2) 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 17:53:39,570 INFO [train.py:968] (1/2) Epoch 24, batch 42300, giga_loss[loss=0.273, simple_loss=0.3399, pruned_loss=0.103, over 28887.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3707, pruned_loss=0.1191, over 5680220.69 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.357, pruned_loss=0.1118, over 5715094.09 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3731, pruned_loss=0.1203, over 5663371.02 frames. ], batch size: 106, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:54:06,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:54:28,328 INFO [train.py:968] (1/2) Epoch 24, batch 42350, giga_loss[loss=0.3287, simple_loss=0.3982, pruned_loss=0.1296, over 28274.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3701, pruned_loss=0.1196, over 5673467.30 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3575, pruned_loss=0.112, over 5716975.83 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.372, pruned_loss=0.1207, over 5656399.49 frames. ], batch size: 368, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:54:32,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5024, 4.3543, 4.1632, 1.8208], device='cuda:1'), covar=tensor([0.0584, 0.0694, 0.0713, 0.2106], device='cuda:1'), in_proj_covar=tensor([0.1274, 0.1181, 0.0992, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 17:54:53,676 INFO [optim.py:369] (1/2) 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 17:55:15,391 INFO [train.py:968] (1/2) Epoch 24, batch 42400, giga_loss[loss=0.2633, simple_loss=0.3428, pruned_loss=0.09188, over 28657.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3685, pruned_loss=0.1177, over 5673867.98 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3575, pruned_loss=0.1122, over 5713820.10 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3705, pruned_loss=0.1187, over 5660118.07 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 17:55:58,964 INFO [train.py:968] (1/2) Epoch 24, batch 42450, giga_loss[loss=0.2762, simple_loss=0.3563, pruned_loss=0.09804, over 28826.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3677, pruned_loss=0.1161, over 5679570.00 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3572, pruned_loss=0.1121, over 5711849.08 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3701, pruned_loss=0.1172, over 5668322.40 frames. ], batch size: 284, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:56:25,564 INFO [optim.py:369] (1/2) 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 17:56:50,343 INFO [train.py:968] (1/2) Epoch 24, batch 42500, giga_loss[loss=0.2984, simple_loss=0.3698, pruned_loss=0.1135, over 29009.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3683, pruned_loss=0.1165, over 5676352.10 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3568, pruned_loss=0.1118, over 5711439.97 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3708, pruned_loss=0.1178, over 5666685.52 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:57:02,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1091411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:57:41,099 INFO [train.py:968] (1/2) Epoch 24, batch 42550, giga_loss[loss=0.3358, simple_loss=0.3907, pruned_loss=0.1404, over 28935.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3689, pruned_loss=0.1177, over 5673928.13 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3567, pruned_loss=0.1118, over 5713214.51 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3711, pruned_loss=0.1187, over 5664625.49 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:58:03,014 INFO [optim.py:369] (1/2) 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 17:58:24,614 INFO [train.py:968] (1/2) Epoch 24, batch 42600, giga_loss[loss=0.3396, simple_loss=0.4, pruned_loss=0.1396, over 28926.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3673, pruned_loss=0.1169, over 5684472.82 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3571, pruned_loss=0.1121, over 5714668.14 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.369, pruned_loss=0.1177, over 5674490.03 frames. ], batch size: 285, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:58:25,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:58:41,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3413, 1.6327, 1.3119, 0.9351], device='cuda:1'), covar=tensor([0.2565, 0.2665, 0.2990, 0.2453], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1119, 0.1369, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 17:59:10,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9775, 1.3367, 1.0941, 0.1637], device='cuda:1'), covar=tensor([0.4124, 0.3169, 0.4524, 0.6982], device='cuda:1'), in_proj_covar=tensor([0.1781, 0.1674, 0.1617, 0.1445], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 17:59:12,597 INFO [train.py:968] (1/2) Epoch 24, batch 42650, giga_loss[loss=0.2816, simple_loss=0.3554, pruned_loss=0.1039, over 29041.00 frames. ], tot_loss[loss=0.301, simple_loss=0.367, pruned_loss=0.1175, over 5681028.57 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.357, pruned_loss=0.1118, over 5720288.54 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3688, pruned_loss=0.1185, over 5666589.13 frames. ], batch size: 155, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 17:59:21,200 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:59:22,908 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 17:59:28,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-12 17:59:30,856 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 17:59:33,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-12 17:59:40,414 INFO [optim.py:369] (1/2) 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 17:59:51,013 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:00:02,966 INFO [train.py:968] (1/2) Epoch 24, batch 42700, giga_loss[loss=0.2618, simple_loss=0.3366, pruned_loss=0.09352, over 28755.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3656, pruned_loss=0.117, over 5684755.51 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3569, pruned_loss=0.1118, over 5721992.54 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3671, pruned_loss=0.1178, over 5671473.72 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:00:04,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1091599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:00:29,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8299, 1.1502, 2.8929, 2.7725], device='cuda:1'), covar=tensor([0.1683, 0.2521, 0.0621, 0.0922], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0659, 0.0977, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 18:00:55,241 INFO [train.py:968] (1/2) Epoch 24, batch 42750, giga_loss[loss=0.2568, simple_loss=0.338, pruned_loss=0.08779, over 28973.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3643, pruned_loss=0.1168, over 5687671.68 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3569, pruned_loss=0.1119, over 5724889.50 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3656, pruned_loss=0.1175, over 5674199.49 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:01:23,811 INFO [optim.py:369] (1/2) 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 18:01:47,507 INFO [train.py:968] (1/2) Epoch 24, batch 42800, libri_loss[loss=0.3019, simple_loss=0.3733, pruned_loss=0.1153, over 29760.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3639, pruned_loss=0.117, over 5670043.92 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3574, pruned_loss=0.112, over 5727659.59 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3647, pruned_loss=0.1175, over 5655207.13 frames. ], batch size: 87, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 18:02:29,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1091742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:02:31,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1091745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:02:33,469 INFO [train.py:968] (1/2) Epoch 24, batch 42850, giga_loss[loss=0.2851, simple_loss=0.3629, pruned_loss=0.1037, over 29049.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3631, pruned_loss=0.1163, over 5674659.87 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3567, pruned_loss=0.1115, over 5734761.37 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3646, pruned_loss=0.1174, over 5653706.98 frames. ], batch size: 128, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:02:45,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4836, 1.8101, 1.4245, 1.7093], device='cuda:1'), covar=tensor([0.2752, 0.2736, 0.3235, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1120, 0.1369, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 18:02:57,791 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:02:59,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1091774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:03:00,051 INFO [optim.py:369] (1/2) 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 18:03:22,215 INFO [train.py:968] (1/2) Epoch 24, batch 42900, giga_loss[loss=0.2843, simple_loss=0.3581, pruned_loss=0.1053, over 28669.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3637, pruned_loss=0.1158, over 5673001.24 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3565, pruned_loss=0.1115, over 5726241.61 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3651, pruned_loss=0.1168, over 5662725.10 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:04:07,936 INFO [train.py:968] (1/2) Epoch 24, batch 42950, libri_loss[loss=0.3414, simple_loss=0.3914, pruned_loss=0.1457, over 29653.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3643, pruned_loss=0.1156, over 5679581.65 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3565, pruned_loss=0.1114, over 5730831.24 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3657, pruned_loss=0.1165, over 5665551.07 frames. ], batch size: 88, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:04:18,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 18:04:30,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1091874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:04:31,344 INFO [optim.py:369] (1/2) 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 18:04:35,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4213, 1.6435, 1.5891, 1.4902], device='cuda:1'), covar=tensor([0.1836, 0.2015, 0.2130, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0751, 0.0722, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 18:04:51,284 INFO [train.py:968] (1/2) Epoch 24, batch 43000, giga_loss[loss=0.3125, simple_loss=0.3749, pruned_loss=0.1251, over 28759.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3638, pruned_loss=0.1146, over 5688307.58 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3565, pruned_loss=0.1114, over 5733991.37 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3652, pruned_loss=0.1156, over 5672017.80 frames. ], batch size: 242, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:05:08,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5097, 1.6461, 1.5817, 1.3846], device='cuda:1'), covar=tensor([0.3036, 0.2586, 0.2124, 0.2623], device='cuda:1'), in_proj_covar=tensor([0.2008, 0.1940, 0.1864, 0.2007], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 18:05:30,975 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3538, 3.3298, 1.4609, 1.5378], device='cuda:1'), covar=tensor([0.1018, 0.0345, 0.0918, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0565, 0.0396, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 18:05:34,546 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 43050, giga_loss[loss=0.3755, simple_loss=0.4149, pruned_loss=0.1681, over 28000.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3647, pruned_loss=0.1156, over 5691896.09 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.356, pruned_loss=0.1111, over 5737673.21 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3667, pruned_loss=0.1168, over 5674141.48 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:06:09,501 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 43100, giga_loss[loss=0.286, simple_loss=0.3531, pruned_loss=0.1094, over 28671.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3678, pruned_loss=0.1184, over 5683294.79 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3561, pruned_loss=0.1112, over 5728011.84 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3692, pruned_loss=0.1192, over 5677057.45 frames. ], batch size: 262, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:06:52,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:06:57,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1092020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:07:27,346 INFO [train.py:968] (1/2) Epoch 24, batch 43150, giga_loss[loss=0.3135, simple_loss=0.3663, pruned_loss=0.1304, over 28685.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1214, over 5672807.87 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3566, pruned_loss=0.1115, over 5727479.88 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3706, pruned_loss=0.122, over 5667682.24 frames. ], batch size: 78, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:07:28,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1092049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:07:30,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8078, 1.9801, 2.0377, 1.5896], device='cuda:1'), covar=tensor([0.1823, 0.2602, 0.1527, 0.1849], device='cuda:1'), in_proj_covar=tensor([0.0914, 0.0707, 0.0958, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 18:07:36,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8934, 5.7301, 5.4321, 2.8185], device='cuda:1'), covar=tensor([0.0493, 0.0636, 0.0775, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.1286, 0.1189, 0.1000, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 18:07:40,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4648, 1.8439, 1.4460, 1.4464], device='cuda:1'), covar=tensor([0.2574, 0.2589, 0.2938, 0.2356], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1119, 0.1368, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 18:07:54,185 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092084.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:08:04,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1092087.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:08:17,691 INFO [train.py:968] (1/2) Epoch 24, batch 43200, giga_loss[loss=0.4477, simple_loss=0.4813, pruned_loss=0.207, over 24194.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3716, pruned_loss=0.1234, over 5665078.34 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3566, pruned_loss=0.1114, over 5722673.65 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3726, pruned_loss=0.1242, over 5664187.50 frames. ], batch size: 705, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:08:36,525 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1092116.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:08:42,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-12 18:09:07,415 INFO [train.py:968] (1/2) Epoch 24, batch 43250, giga_loss[loss=0.3011, simple_loss=0.3666, pruned_loss=0.1178, over 28852.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.1231, over 5664867.33 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3564, pruned_loss=0.1113, over 5725349.29 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.372, pruned_loss=0.124, over 5660986.89 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:09:07,917 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:09:20,516 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:09:37,850 INFO [optim.py:369] (1/2) 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 18:09:40,534 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-12 18:09:56,801 INFO [train.py:968] (1/2) Epoch 24, batch 43300, giga_loss[loss=0.256, simple_loss=0.3487, pruned_loss=0.08166, over 28801.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3696, pruned_loss=0.1221, over 5663224.32 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3564, pruned_loss=0.1112, over 5727139.39 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3708, pruned_loss=0.123, over 5657922.22 frames. ], batch size: 174, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:10:07,148 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2611, 1.8496, 1.3623, 0.4177], device='cuda:1'), covar=tensor([0.5301, 0.3400, 0.4816, 0.7311], device='cuda:1'), in_proj_covar=tensor([0.1805, 0.1700, 0.1634, 0.1467], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 18:10:42,589 INFO [train.py:968] (1/2) Epoch 24, batch 43350, giga_loss[loss=0.2713, simple_loss=0.348, pruned_loss=0.09732, over 28835.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.371, pruned_loss=0.1217, over 5660069.27 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3574, pruned_loss=0.1119, over 5721570.43 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3714, pruned_loss=0.122, over 5659087.37 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:11:04,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9353, 1.3839, 1.4628, 1.1624], device='cuda:1'), covar=tensor([0.1766, 0.1056, 0.1861, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0753, 0.0722, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 18:11:09,039 INFO [optim.py:369] (1/2) 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 18:11:22,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:11:24,886 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 43400, libri_loss[loss=0.2677, simple_loss=0.3368, pruned_loss=0.09936, over 29554.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3686, pruned_loss=0.1195, over 5666762.84 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3573, pruned_loss=0.1118, over 5729092.70 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3695, pruned_loss=0.1203, over 5656677.21 frames. ], batch size: 76, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:11:33,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5673, 1.7010, 1.6958, 1.4520], device='cuda:1'), covar=tensor([0.2791, 0.2529, 0.2152, 0.2503], device='cuda:1'), in_proj_covar=tensor([0.2017, 0.1952, 0.1869, 0.2015], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 18:11:51,288 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 24, batch 43450, giga_loss[loss=0.3603, simple_loss=0.3994, pruned_loss=0.1606, over 26534.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.367, pruned_loss=0.1184, over 5665026.99 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3581, pruned_loss=0.1122, over 5720007.32 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3673, pruned_loss=0.1188, over 5663599.62 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:12:16,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5844, 1.8155, 1.4596, 1.6468], device='cuda:1'), covar=tensor([0.2616, 0.2685, 0.3056, 0.2359], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1119, 0.1368, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 18:12:33,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5262, 3.5922, 1.6952, 1.5739], device='cuda:1'), covar=tensor([0.1000, 0.0332, 0.0860, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0566, 0.0396, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 18:12:41,236 INFO [optim.py:369] (1/2) 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 18:13:00,437 INFO [train.py:968] (1/2) Epoch 24, batch 43500, giga_loss[loss=0.2941, simple_loss=0.359, pruned_loss=0.1145, over 28447.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.367, pruned_loss=0.1193, over 5662187.58 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1124, over 5719771.87 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3669, pruned_loss=0.1196, over 5660554.14 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:13:22,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3632, 1.6326, 1.6648, 1.4513], device='cuda:1'), covar=tensor([0.1828, 0.1754, 0.2079, 0.1922], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0754, 0.0722, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 18:13:30,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5460, 1.6167, 1.7402, 1.3333], device='cuda:1'), covar=tensor([0.1720, 0.2588, 0.1441, 0.1685], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0709, 0.0960, 0.0860], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 18:13:50,102 INFO [train.py:968] (1/2) Epoch 24, batch 43550, giga_loss[loss=0.3624, simple_loss=0.413, pruned_loss=0.1559, over 28993.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3682, pruned_loss=0.1205, over 5662642.18 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3583, pruned_loss=0.1122, over 5711975.78 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3685, pruned_loss=0.1209, over 5668437.15 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:14:23,932 INFO [optim.py:369] (1/2) 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 18:14:34,276 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092490.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:14:41,580 INFO [train.py:968] (1/2) Epoch 24, batch 43600, giga_loss[loss=0.3132, simple_loss=0.3871, pruned_loss=0.1196, over 28933.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3714, pruned_loss=0.1213, over 5658146.02 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3582, pruned_loss=0.1121, over 5715544.99 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3719, pruned_loss=0.1219, over 5658631.19 frames. ], batch size: 213, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:15:17,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:15:32,842 INFO [train.py:968] (1/2) Epoch 24, batch 43650, giga_loss[loss=0.3477, simple_loss=0.4026, pruned_loss=0.1463, over 28976.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3731, pruned_loss=0.1199, over 5666954.57 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3579, pruned_loss=0.1121, over 5716342.22 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3741, pruned_loss=0.1206, over 5665636.63 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:16:03,191 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 24, batch 43700, libri_loss[loss=0.2338, simple_loss=0.3078, pruned_loss=0.07988, over 29368.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3745, pruned_loss=0.1211, over 5665486.96 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.358, pruned_loss=0.1121, over 5718086.39 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3758, pruned_loss=0.1219, over 5661254.08 frames. ], batch size: 71, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:16:33,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:17:11,412 INFO [train.py:968] (1/2) Epoch 24, batch 43750, giga_loss[loss=0.3189, simple_loss=0.3832, pruned_loss=0.1273, over 28765.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3766, pruned_loss=0.1227, over 5657938.58 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3581, pruned_loss=0.1121, over 5712042.04 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3779, pruned_loss=0.1236, over 5658143.46 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:17:42,181 INFO [optim.py:369] (1/2) 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 18:17:44,683 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1092678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:17:46,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1092681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:18:02,265 INFO [train.py:968] (1/2) Epoch 24, batch 43800, giga_loss[loss=0.3972, simple_loss=0.4354, pruned_loss=0.1795, over 27580.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3771, pruned_loss=0.1242, over 5659453.60 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3581, pruned_loss=0.1121, over 5714911.36 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3784, pruned_loss=0.125, over 5656447.13 frames. ], batch size: 472, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:18:16,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1092710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:18:18,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 18:18:32,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8566, 2.0126, 1.7470, 1.7701], device='cuda:1'), covar=tensor([0.2115, 0.2664, 0.2483, 0.2458], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0750, 0.0719, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 18:18:53,321 INFO [train.py:968] (1/2) Epoch 24, batch 43850, giga_loss[loss=0.2782, simple_loss=0.3519, pruned_loss=0.1022, over 29003.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3763, pruned_loss=0.1246, over 5647543.03 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3583, pruned_loss=0.1122, over 5706409.83 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3772, pruned_loss=0.1253, over 5652300.00 frames. ], batch size: 164, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:19:20,710 INFO [optim.py:369] (1/2) 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 18:19:39,119 INFO [train.py:968] (1/2) Epoch 24, batch 43900, giga_loss[loss=0.2741, simple_loss=0.3441, pruned_loss=0.102, over 28956.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3734, pruned_loss=0.1231, over 5641358.29 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1123, over 5692121.28 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3745, pruned_loss=0.1239, over 5657194.38 frames. ], batch size: 136, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:20:28,007 INFO [train.py:968] (1/2) Epoch 24, batch 43950, giga_loss[loss=0.3092, simple_loss=0.3694, pruned_loss=0.1245, over 28957.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3714, pruned_loss=0.1221, over 5652373.12 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3584, pruned_loss=0.1124, over 5687135.66 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3726, pruned_loss=0.1229, over 5667846.02 frames. ], batch size: 145, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:20:39,246 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1092858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:20:44,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092865.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:21:02,274 INFO [optim.py:369] (1/2) 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 18:21:04,835 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-12 18:21:21,187 INFO [train.py:968] (1/2) Epoch 24, batch 44000, giga_loss[loss=0.2949, simple_loss=0.363, pruned_loss=0.1134, over 28861.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3707, pruned_loss=0.122, over 5657401.31 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1124, over 5689555.41 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3718, pruned_loss=0.1228, over 5666877.65 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 8.0
+2023-03-12 18:22:10,648 INFO [train.py:968] (1/2) Epoch 24, batch 44050, giga_loss[loss=0.3967, simple_loss=0.4311, pruned_loss=0.1812, over 26601.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3724, pruned_loss=0.1239, over 5650276.53 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5684635.23 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5661176.58 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:22:39,491 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1092983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:22:55,163 INFO [train.py:968] (1/2) Epoch 24, batch 44100, giga_loss[loss=0.252, simple_loss=0.322, pruned_loss=0.09097, over 28430.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3709, pruned_loss=0.1231, over 5661587.95 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.359, pruned_loss=0.1126, over 5690491.03 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3719, pruned_loss=0.124, over 5664163.03 frames. ], batch size: 60, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:23:05,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093008.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:23:08,702 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093011.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:23:39,278 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093040.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:23:46,250 INFO [train.py:968] (1/2) Epoch 24, batch 44150, giga_loss[loss=0.3641, simple_loss=0.4102, pruned_loss=0.159, over 27912.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3699, pruned_loss=0.1227, over 5667565.97 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3591, pruned_loss=0.1127, over 5691512.58 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1234, over 5668464.18 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:23:55,834 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:24:16,348 INFO [optim.py:369] (1/2) 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 18:24:28,646 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:24:34,690 INFO [train.py:968] (1/2) Epoch 24, batch 44200, giga_loss[loss=0.2906, simple_loss=0.3668, pruned_loss=0.1072, over 28838.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3708, pruned_loss=0.1225, over 5664498.01 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1127, over 5693959.91 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3715, pruned_loss=0.1232, over 5662592.75 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:25:05,734 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:09,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:17,336 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-12 18:25:20,649 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5372, 1.9653, 1.4708, 1.7329], device='cuda:1'), covar=tensor([0.2641, 0.2680, 0.3093, 0.2488], device='cuda:1'), in_proj_covar=tensor([0.1552, 0.1121, 0.1371, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 18:25:22,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0079, 1.2242, 0.9321, 0.3563], device='cuda:1'), covar=tensor([0.3215, 0.2639, 0.3115, 0.5728], device='cuda:1'), in_proj_covar=tensor([0.1791, 0.1689, 0.1620, 0.1454], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 18:25:25,992 INFO [train.py:968] (1/2) Epoch 24, batch 44250, giga_loss[loss=0.2841, simple_loss=0.3559, pruned_loss=0.1062, over 28728.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3721, pruned_loss=0.123, over 5658291.37 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3589, pruned_loss=0.1125, over 5684963.90 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3733, pruned_loss=0.124, over 5662984.88 frames. ], batch size: 92, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:25:35,112 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:37,213 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:25:55,400 INFO [optim.py:369] (1/2) 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 18:26:15,991 INFO [train.py:968] (1/2) Epoch 24, batch 44300, giga_loss[loss=0.3085, simple_loss=0.3675, pruned_loss=0.1248, over 28510.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3723, pruned_loss=0.1238, over 5663797.04 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5690537.03 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3736, pruned_loss=0.1249, over 5662154.37 frames. ], batch size: 60, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:26:32,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7377, 2.3105, 1.4303, 1.0453], device='cuda:1'), covar=tensor([0.7649, 0.4883, 0.3954, 0.6505], device='cuda:1'), in_proj_covar=tensor([0.1789, 0.1687, 0.1620, 0.1453], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 18:26:48,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:27:04,249 INFO [train.py:968] (1/2) Epoch 24, batch 44350, giga_loss[loss=0.3526, simple_loss=0.402, pruned_loss=0.1516, over 27975.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3733, pruned_loss=0.123, over 5655154.08 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5681179.40 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3746, pruned_loss=0.124, over 5661610.97 frames. ], batch size: 412, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:27:31,317 INFO [optim.py:369] (1/2) 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 18:27:39,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8942, 1.7451, 4.9860, 3.7398], device='cuda:1'), covar=tensor([0.1549, 0.2564, 0.0494, 0.0793], device='cuda:1'), in_proj_covar=tensor([0.0780, 0.0661, 0.0982, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 18:27:43,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4468, 1.5601, 1.6506, 1.2095], device='cuda:1'), covar=tensor([0.1993, 0.3022, 0.1757, 0.2103], device='cuda:1'), in_proj_covar=tensor([0.0915, 0.0711, 0.0961, 0.0861], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 18:27:46,046 INFO [train.py:968] (1/2) Epoch 24, batch 44400, giga_loss[loss=0.3716, simple_loss=0.4133, pruned_loss=0.165, over 28348.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3727, pruned_loss=0.1201, over 5664816.26 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1127, over 5677769.34 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3741, pruned_loss=0.1211, over 5672101.80 frames. ], batch size: 368, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:27:46,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 18:28:37,195 INFO [train.py:968] (1/2) Epoch 24, batch 44450, giga_loss[loss=0.3674, simple_loss=0.4262, pruned_loss=0.1543, over 28549.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3745, pruned_loss=0.1197, over 5672820.74 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3587, pruned_loss=0.1125, over 5682061.86 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3762, pruned_loss=0.1209, over 5674685.69 frames. ], batch size: 336, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:28:39,134 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-12 18:29:04,386 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:29:07,769 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:29:08,137 INFO [optim.py:369] (1/2) 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 18:29:27,060 INFO [train.py:968] (1/2) Epoch 24, batch 44500, giga_loss[loss=0.333, simple_loss=0.3931, pruned_loss=0.1365, over 28899.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3774, pruned_loss=0.1224, over 5672468.28 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3583, pruned_loss=0.1123, over 5682822.30 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3793, pruned_loss=0.1236, over 5673413.16 frames. ], batch size: 186, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:29:29,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0219, 1.3497, 1.0824, 0.1859], device='cuda:1'), covar=tensor([0.3544, 0.2927, 0.3971, 0.6165], device='cuda:1'), in_proj_covar=tensor([0.1787, 0.1682, 0.1617, 0.1449], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 18:29:35,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-12 18:29:37,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:30:01,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:30:17,040 INFO [train.py:968] (1/2) Epoch 24, batch 44550, giga_loss[loss=0.3153, simple_loss=0.3789, pruned_loss=0.1258, over 28807.00 frames. ], tot_loss[loss=0.314, simple_loss=0.379, pruned_loss=0.1245, over 5672318.74 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1123, over 5685808.72 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3807, pruned_loss=0.1256, over 5670232.09 frames. ], batch size: 119, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:30:30,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8561, 0.9345, 0.8677, 0.8538], device='cuda:1'), covar=tensor([0.1617, 0.2067, 0.1429, 0.1687], device='cuda:1'), in_proj_covar=tensor([0.2008, 0.1940, 0.1863, 0.2004], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 18:30:39,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:30:48,849 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2454, 4.0939, 3.8898, 1.8883], device='cuda:1'), covar=tensor([0.0608, 0.0723, 0.0737, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.1277, 0.1186, 0.0993, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 18:30:50,265 INFO [optim.py:369] (1/2) 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 18:31:08,243 INFO [train.py:968] (1/2) Epoch 24, batch 44600, giga_loss[loss=0.3425, simple_loss=0.3991, pruned_loss=0.1429, over 28862.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3805, pruned_loss=0.1268, over 5639659.12 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3587, pruned_loss=0.1126, over 5672916.35 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3823, pruned_loss=0.1279, over 5649336.65 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:31:32,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:31:42,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:31:53,391 INFO [train.py:968] (1/2) Epoch 24, batch 44650, libri_loss[loss=0.2658, simple_loss=0.3407, pruned_loss=0.09546, over 29542.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3795, pruned_loss=0.1257, over 5647891.68 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3587, pruned_loss=0.1125, over 5669020.79 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3814, pruned_loss=0.1268, over 5658171.58 frames. ], batch size: 80, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:32:18,694 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:21,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:22,933 INFO [optim.py:369] (1/2) 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 18:32:39,415 INFO [train.py:968] (1/2) Epoch 24, batch 44700, giga_loss[loss=0.3447, simple_loss=0.3849, pruned_loss=0.1523, over 26606.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3767, pruned_loss=0.1221, over 5657648.16 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3582, pruned_loss=0.1122, over 5673005.99 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.379, pruned_loss=0.1236, over 5661833.63 frames. ], batch size: 555, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:32:49,389 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:52,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:32:54,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:33:25,309 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:33:26,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4834, 1.7483, 1.4165, 1.4197], device='cuda:1'), covar=tensor([0.2679, 0.2666, 0.3056, 0.2341], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1121, 0.1372, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 18:33:29,116 INFO [train.py:968] (1/2) Epoch 24, batch 44750, giga_loss[loss=0.2895, simple_loss=0.365, pruned_loss=0.107, over 28939.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3777, pruned_loss=0.1214, over 5655709.11 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3583, pruned_loss=0.1123, over 5672769.71 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3797, pruned_loss=0.1226, over 5658836.59 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 2.0
+2023-03-12 18:33:30,650 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-12 18:33:57,252 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1093679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:33:58,264 INFO [optim.py:369] (1/2) 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 18:34:00,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1093682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:34:10,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4853, 1.6656, 1.1762, 1.2168], device='cuda:1'), covar=tensor([0.0932, 0.0495, 0.1019, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0451, 0.0521, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 18:34:15,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6199, 1.8224, 1.8209, 1.5188], device='cuda:1'), covar=tensor([0.2540, 0.2334, 0.2475, 0.2547], device='cuda:1'), in_proj_covar=tensor([0.2008, 0.1943, 0.1868, 0.2006], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 18:34:16,329 INFO [train.py:968] (1/2) Epoch 24, batch 44800, giga_loss[loss=0.3124, simple_loss=0.3843, pruned_loss=0.1203, over 28854.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3794, pruned_loss=0.1229, over 5618498.77 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3593, pruned_loss=0.1133, over 5621335.97 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3806, pruned_loss=0.1232, over 5666733.58 frames. ], batch size: 199, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:34:28,807 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1093711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:34:40,316 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093723.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:34:42,634 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1093726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:35:05,668 INFO [train.py:968] (1/2) Epoch 24, batch 44850, giga_loss[loss=0.3192, simple_loss=0.3891, pruned_loss=0.1246, over 28958.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3796, pruned_loss=0.1238, over 5600445.74 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3596, pruned_loss=0.1136, over 5585711.87 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3805, pruned_loss=0.1239, over 5669242.96 frames. ], batch size: 227, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:35:37,114 INFO [optim.py:369] (1/2) 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 18:35:55,956 INFO [train.py:968] (1/2) Epoch 24, batch 44900, giga_loss[loss=0.2832, simple_loss=0.3536, pruned_loss=0.1064, over 28706.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3777, pruned_loss=0.1235, over 5592376.23 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.36, pruned_loss=0.1139, over 5568050.71 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3782, pruned_loss=0.1233, over 5661670.14 frames. ], batch size: 243, lr: 1.31e-03, grad_scale: 4.0
+2023-03-12 18:36:39,420 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-12 18:37:51,180 INFO [optim.py:369] (1/2) 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 18:38:00,131 INFO [train.py:968] (1/2) Epoch 25, batch 50, giga_loss[loss=0.2515, simple_loss=0.3447, pruned_loss=0.07915, over 28453.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3675, pruned_loss=0.1035, over 1267070.24 frames. ], libri_tot_loss[loss=0.2661, simple_loss=0.3502, pruned_loss=0.09095, over 145671.75 frames. ], giga_tot_loss[loss=0.2895, simple_loss=0.3694, pruned_loss=0.1048, over 1150694.68 frames. ], batch size: 71, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:38:09,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1093901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:38:49,205 INFO [train.py:968] (1/2) Epoch 25, batch 100, giga_loss[loss=0.2385, simple_loss=0.322, pruned_loss=0.07746, over 28870.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3576, pruned_loss=0.09966, over 2241580.45 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3352, pruned_loss=0.08445, over 314588.92 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3608, pruned_loss=0.1017, over 2039002.44 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:39:02,237 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-12 18:39:26,012 INFO [optim.py:369] (1/2) 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 18:39:34,303 INFO [train.py:968] (1/2) Epoch 25, batch 150, giga_loss[loss=0.2107, simple_loss=0.2875, pruned_loss=0.0669, over 28494.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.341, pruned_loss=0.09164, over 2999674.21 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3324, pruned_loss=0.08405, over 388565.09 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3425, pruned_loss=0.09259, over 2805090.12 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:39:51,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:39:52,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9737, 3.7950, 3.6054, 1.6214], device='cuda:1'), covar=tensor([0.0640, 0.0860, 0.0820, 0.2166], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.1184, 0.0992, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 18:40:18,872 INFO [train.py:968] (1/2) Epoch 25, batch 200, giga_loss[loss=0.2084, simple_loss=0.2886, pruned_loss=0.06409, over 28842.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3284, pruned_loss=0.08569, over 3602715.27 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3317, pruned_loss=0.08272, over 550266.68 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3288, pruned_loss=0.0864, over 3378558.25 frames. ], batch size: 66, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:40:22,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:24,586 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:36,416 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:47,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:40:51,418 INFO [optim.py:369] (1/2) 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 18:41:01,336 INFO [train.py:968] (1/2) Epoch 25, batch 250, giga_loss[loss=0.2085, simple_loss=0.2828, pruned_loss=0.06712, over 28590.00 frames. ], tot_loss[loss=0.2413, simple_loss=0.3193, pruned_loss=0.08158, over 4062996.35 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3321, pruned_loss=0.08181, over 653475.28 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.3189, pruned_loss=0.08202, over 3851184.53 frames. ], batch size: 71, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:41:10,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094098.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 18:41:13,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:41:23,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 18:41:38,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:41:45,973 INFO [train.py:968] (1/2) Epoch 25, batch 300, giga_loss[loss=0.1949, simple_loss=0.27, pruned_loss=0.05993, over 28892.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.3119, pruned_loss=0.07856, over 4422624.58 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3329, pruned_loss=0.08238, over 805381.55 frames. ], giga_tot_loss[loss=0.2339, simple_loss=0.3105, pruned_loss=0.07859, over 4211503.81 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:42:07,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:42:25,888 INFO [optim.py:369] (1/2) 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 18:42:32,912 INFO [train.py:968] (1/2) Epoch 25, batch 350, giga_loss[loss=0.1826, simple_loss=0.2592, pruned_loss=0.05304, over 28934.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.306, pruned_loss=0.0758, over 4701277.07 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3326, pruned_loss=0.082, over 954229.86 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.3039, pruned_loss=0.07559, over 4498008.10 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:43:13,533 INFO [train.py:968] (1/2) Epoch 25, batch 400, giga_loss[loss=0.244, simple_loss=0.316, pruned_loss=0.08601, over 28328.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3031, pruned_loss=0.07475, over 4929261.62 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3316, pruned_loss=0.08261, over 1100593.87 frames. ], giga_tot_loss[loss=0.2245, simple_loss=0.3006, pruned_loss=0.07421, over 4737374.18 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:43:14,435 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:43:17,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094244.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:43:17,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:43:19,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:43:41,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094273.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 18:43:43,582 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:43:53,694 INFO [train.py:968] (1/2) Epoch 25, batch 450, libri_loss[loss=0.2608, simple_loss=0.351, pruned_loss=0.08531, over 29368.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3023, pruned_loss=0.07467, over 5104864.94 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3333, pruned_loss=0.08307, over 1242141.53 frames. ], giga_tot_loss[loss=0.2234, simple_loss=0.2991, pruned_loss=0.07388, over 4927738.42 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:44:00,926 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-12 18:44:08,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3236, 1.4782, 1.4676, 1.3353], device='cuda:1'), covar=tensor([0.3275, 0.2571, 0.2324, 0.2711], device='cuda:1'), in_proj_covar=tensor([0.2010, 0.1938, 0.1863, 0.2004], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 18:44:11,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5828, 1.8268, 1.4954, 1.5499], device='cuda:1'), covar=tensor([0.2696, 0.2795, 0.3164, 0.2597], device='cuda:1'), in_proj_covar=tensor([0.1560, 0.1123, 0.1377, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 18:44:38,662 INFO [train.py:968] (1/2) Epoch 25, batch 500, giga_loss[loss=0.2081, simple_loss=0.2834, pruned_loss=0.06635, over 28658.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2999, pruned_loss=0.07352, over 5232747.35 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3334, pruned_loss=0.08301, over 1378650.63 frames. ], giga_tot_loss[loss=0.2208, simple_loss=0.2963, pruned_loss=0.07264, over 5072407.48 frames. ], batch size: 78, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:44:47,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.89 vs. limit=2.0
+2023-03-12 18:45:17,180 INFO [optim.py:369] (1/2) 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 18:45:17,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:45:25,657 INFO [train.py:968] (1/2) Epoch 25, batch 550, giga_loss[loss=0.2145, simple_loss=0.2943, pruned_loss=0.0674, over 28500.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2984, pruned_loss=0.07295, over 5335441.89 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3364, pruned_loss=0.08477, over 1488797.39 frames. ], giga_tot_loss[loss=0.2186, simple_loss=0.294, pruned_loss=0.07158, over 5195412.07 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:45:28,114 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:46:09,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:46:10,707 INFO [train.py:968] (1/2) Epoch 25, batch 600, giga_loss[loss=0.214, simple_loss=0.294, pruned_loss=0.06694, over 29058.00 frames. ], tot_loss[loss=0.2206, simple_loss=0.2963, pruned_loss=0.07242, over 5415081.12 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3353, pruned_loss=0.0844, over 1576450.72 frames. ], giga_tot_loss[loss=0.2174, simple_loss=0.2923, pruned_loss=0.07119, over 5294265.24 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:46:34,563 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3426, 1.6776, 1.5448, 1.5490], device='cuda:1'), covar=tensor([0.0777, 0.0330, 0.0322, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 18:46:50,253 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 650, giga_loss[loss=0.1997, simple_loss=0.2791, pruned_loss=0.06011, over 28911.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2944, pruned_loss=0.07153, over 5479265.34 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3356, pruned_loss=0.08441, over 1659282.09 frames. ], giga_tot_loss[loss=0.2155, simple_loss=0.2904, pruned_loss=0.07032, over 5377806.11 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:47:11,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:15,180 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2458, 1.5100, 1.3633, 1.1676], device='cuda:1'), covar=tensor([0.3113, 0.2710, 0.1893, 0.2594], device='cuda:1'), in_proj_covar=tensor([0.2010, 0.1936, 0.1859, 0.2002], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 18:47:28,803 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:31,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:42,159 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:47:43,159 INFO [train.py:968] (1/2) Epoch 25, batch 700, giga_loss[loss=0.1924, simple_loss=0.2713, pruned_loss=0.05672, over 28682.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.292, pruned_loss=0.07024, over 5527951.60 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3358, pruned_loss=0.08493, over 1781529.39 frames. ], giga_tot_loss[loss=0.2126, simple_loss=0.2876, pruned_loss=0.06876, over 5439059.62 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:47:59,629 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:21,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:21,293 INFO [optim.py:369] (1/2) 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 18:48:24,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:27,552 INFO [train.py:968] (1/2) Epoch 25, batch 750, giga_loss[loss=0.2006, simple_loss=0.2792, pruned_loss=0.06099, over 28984.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2902, pruned_loss=0.06931, over 5562558.97 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3373, pruned_loss=0.08579, over 1884103.24 frames. ], giga_tot_loss[loss=0.2101, simple_loss=0.2852, pruned_loss=0.06751, over 5483353.82 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:48:47,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:48:59,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4687, 1.6491, 1.1917, 1.2877], device='cuda:1'), covar=tensor([0.0985, 0.0625, 0.1091, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0449, 0.0521, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 18:49:09,949 INFO [train.py:968] (1/2) Epoch 25, batch 800, giga_loss[loss=0.1932, simple_loss=0.2699, pruned_loss=0.05824, over 28885.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2889, pruned_loss=0.06901, over 5593281.66 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3372, pruned_loss=0.08573, over 1981867.60 frames. ], giga_tot_loss[loss=0.2091, simple_loss=0.2839, pruned_loss=0.06721, over 5524301.72 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:49:18,911 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:21,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:49,880 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:50,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:52,529 INFO [optim.py:369] (1/2) 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 18:49:53,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:49:57,315 INFO [train.py:968] (1/2) Epoch 25, batch 850, giga_loss[loss=0.258, simple_loss=0.3386, pruned_loss=0.0887, over 28947.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2972, pruned_loss=0.07349, over 5607543.91 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.338, pruned_loss=0.08622, over 2069197.12 frames. ], giga_tot_loss[loss=0.2176, simple_loss=0.292, pruned_loss=0.07162, over 5553389.70 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:50:21,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:50:45,654 INFO [train.py:968] (1/2) Epoch 25, batch 900, giga_loss[loss=0.2955, simple_loss=0.3671, pruned_loss=0.1119, over 27947.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3104, pruned_loss=0.08013, over 5627200.21 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3377, pruned_loss=0.08611, over 2107576.30 frames. ], giga_tot_loss[loss=0.2318, simple_loss=0.3062, pruned_loss=0.07868, over 5582003.58 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:51:11,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1094768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:51:23,277 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 950, giga_loss[loss=0.3385, simple_loss=0.3908, pruned_loss=0.1431, over 23535.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3225, pruned_loss=0.0865, over 5615878.95 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3381, pruned_loss=0.08628, over 2162980.89 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3188, pruned_loss=0.08528, over 5593040.93 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:51:44,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3447, 1.4041, 1.2434, 1.4804], device='cuda:1'), covar=tensor([0.0747, 0.0429, 0.0368, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 18:52:09,406 INFO [train.py:968] (1/2) Epoch 25, batch 1000, giga_loss[loss=0.3193, simple_loss=0.3882, pruned_loss=0.1252, over 28286.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3292, pruned_loss=0.08864, over 5634811.28 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.338, pruned_loss=0.086, over 2252297.47 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3261, pruned_loss=0.08782, over 5613215.65 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:52:41,193 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 1050, giga_loss[loss=0.2605, simple_loss=0.3436, pruned_loss=0.08869, over 28810.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3328, pruned_loss=0.08879, over 5656175.67 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3384, pruned_loss=0.08614, over 2341208.42 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.33, pruned_loss=0.08815, over 5634544.20 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:53:08,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1094911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:53:11,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1094914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:53:12,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3830, 1.6744, 1.3278, 1.3523], device='cuda:1'), covar=tensor([0.2888, 0.2889, 0.3307, 0.2528], device='cuda:1'), in_proj_covar=tensor([0.1560, 0.1123, 0.1377, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 18:53:16,315 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 18:53:33,469 INFO [train.py:968] (1/2) Epoch 25, batch 1100, giga_loss[loss=0.2784, simple_loss=0.3672, pruned_loss=0.09477, over 28791.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.334, pruned_loss=0.08868, over 5657750.79 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3381, pruned_loss=0.08615, over 2411457.93 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3318, pruned_loss=0.08822, over 5636739.00 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:53:37,611 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1094943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:53:51,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-12 18:53:55,416 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1094964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:54:09,725 INFO [optim.py:369] (1/2) 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 18:54:16,053 INFO [train.py:968] (1/2) Epoch 25, batch 1150, giga_loss[loss=0.2637, simple_loss=0.3363, pruned_loss=0.09553, over 28848.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3379, pruned_loss=0.09149, over 5661307.68 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3389, pruned_loss=0.08662, over 2479212.01 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3358, pruned_loss=0.09103, over 5641485.48 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:54:24,292 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2750, 4.0854, 3.8968, 1.7762], device='cuda:1'), covar=tensor([0.0706, 0.0890, 0.0821, 0.2230], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.1165, 0.0978, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 18:54:58,305 INFO [train.py:968] (1/2) Epoch 25, batch 1200, giga_loss[loss=0.3466, simple_loss=0.3917, pruned_loss=0.1508, over 26564.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3398, pruned_loss=0.09279, over 5674572.85 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3373, pruned_loss=0.08551, over 2628983.22 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3388, pruned_loss=0.09313, over 5649873.89 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:55:26,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3884, 5.2191, 4.9079, 2.3794], device='cuda:1'), covar=tensor([0.0408, 0.0544, 0.0632, 0.1900], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.1163, 0.0977, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 18:55:36,442 INFO [optim.py:369] (1/2) 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 18:55:41,782 INFO [train.py:968] (1/2) Epoch 25, batch 1250, giga_loss[loss=0.282, simple_loss=0.3564, pruned_loss=0.1038, over 28711.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3434, pruned_loss=0.09497, over 5682041.56 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.338, pruned_loss=0.08584, over 2707993.80 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3424, pruned_loss=0.0953, over 5659059.49 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 18:56:23,472 INFO [train.py:968] (1/2) Epoch 25, batch 1300, giga_loss[loss=0.3908, simple_loss=0.4483, pruned_loss=0.1667, over 28616.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3465, pruned_loss=0.09592, over 5692617.56 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3378, pruned_loss=0.08592, over 2817936.49 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3461, pruned_loss=0.09647, over 5667535.27 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:56:56,147 INFO [optim.py:369] (1/2) 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 18:57:00,911 INFO [train.py:968] (1/2) Epoch 25, batch 1350, giga_loss[loss=0.256, simple_loss=0.3424, pruned_loss=0.08485, over 28591.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3488, pruned_loss=0.09674, over 5693580.46 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3379, pruned_loss=0.08591, over 2879491.59 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3485, pruned_loss=0.09737, over 5669864.34 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:57:41,815 INFO [train.py:968] (1/2) Epoch 25, batch 1400, giga_loss[loss=0.2708, simple_loss=0.3546, pruned_loss=0.09348, over 28675.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3499, pruned_loss=0.09655, over 5699290.26 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3376, pruned_loss=0.08548, over 2968915.31 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3501, pruned_loss=0.09756, over 5674904.32 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:58:13,356 INFO [optim.py:369] (1/2) 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 18:58:17,472 INFO [train.py:968] (1/2) Epoch 25, batch 1450, giga_loss[loss=0.2395, simple_loss=0.3391, pruned_loss=0.06996, over 28701.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3493, pruned_loss=0.0948, over 5700822.68 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3382, pruned_loss=0.08589, over 3101826.23 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3497, pruned_loss=0.09588, over 5682189.68 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:58:54,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1095339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 18:58:54,612 INFO [train.py:968] (1/2) Epoch 25, batch 1500, giga_loss[loss=0.2413, simple_loss=0.3281, pruned_loss=0.07725, over 28842.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3482, pruned_loss=0.0934, over 5706269.08 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3378, pruned_loss=0.08536, over 3197833.33 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3491, pruned_loss=0.0948, over 5685639.35 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 18:59:28,591 INFO [optim.py:369] (1/2) 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 18:59:30,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8334, 5.6002, 5.3420, 3.1017], device='cuda:1'), covar=tensor([0.0388, 0.0595, 0.0645, 0.1550], device='cuda:1'), in_proj_covar=tensor([0.1242, 0.1153, 0.0966, 0.0723], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 18:59:33,503 INFO [train.py:968] (1/2) Epoch 25, batch 1550, giga_loss[loss=0.2451, simple_loss=0.3334, pruned_loss=0.07838, over 28587.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3465, pruned_loss=0.09144, over 5714930.89 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3384, pruned_loss=0.08564, over 3274280.10 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3471, pruned_loss=0.0926, over 5696723.25 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:00:17,329 INFO [train.py:968] (1/2) Epoch 25, batch 1600, giga_loss[loss=0.2882, simple_loss=0.3579, pruned_loss=0.1093, over 28731.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3482, pruned_loss=0.09363, over 5694176.63 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3384, pruned_loss=0.0857, over 3314166.18 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3488, pruned_loss=0.09462, over 5686750.85 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:00:53,622 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1095482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:00:55,208 INFO [optim.py:369] (1/2) 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 19:00:55,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1095485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:01:00,910 INFO [train.py:968] (1/2) Epoch 25, batch 1650, giga_loss[loss=0.2904, simple_loss=0.3565, pruned_loss=0.1121, over 28655.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3502, pruned_loss=0.09734, over 5697507.59 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3381, pruned_loss=0.08561, over 3364303.88 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.351, pruned_loss=0.09837, over 5688938.48 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:01:21,977 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1095514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:01:41,134 INFO [train.py:968] (1/2) Epoch 25, batch 1700, libri_loss[loss=0.272, simple_loss=0.3563, pruned_loss=0.09385, over 29271.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3512, pruned_loss=0.09969, over 5709422.38 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3379, pruned_loss=0.08554, over 3451084.27 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3525, pruned_loss=0.101, over 5696845.13 frames. ], batch size: 97, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:02:16,126 INFO [optim.py:369] (1/2) 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 19:02:22,468 INFO [train.py:968] (1/2) Epoch 25, batch 1750, libri_loss[loss=0.232, simple_loss=0.3127, pruned_loss=0.07564, over 29566.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3504, pruned_loss=0.1003, over 5709566.15 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3378, pruned_loss=0.0856, over 3557395.54 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.352, pruned_loss=0.1019, over 5693413.03 frames. ], batch size: 76, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:02:42,012 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1095612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:03:04,193 INFO [train.py:968] (1/2) Epoch 25, batch 1800, giga_loss[loss=0.274, simple_loss=0.3465, pruned_loss=0.1007, over 28252.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.348, pruned_loss=0.09958, over 5702280.65 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3377, pruned_loss=0.08538, over 3603593.96 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3496, pruned_loss=0.1013, over 5686266.34 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:03:17,949 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1095655.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:03:41,813 INFO [optim.py:369] (1/2) 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 19:03:46,404 INFO [train.py:968] (1/2) Epoch 25, batch 1850, giga_loss[loss=0.2727, simple_loss=0.3414, pruned_loss=0.102, over 23634.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3458, pruned_loss=0.09841, over 5696969.60 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.337, pruned_loss=0.08522, over 3657960.51 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3477, pruned_loss=0.1002, over 5681927.48 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:03:58,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4181, 4.3411, 1.5742, 1.5850], device='cuda:1'), covar=tensor([0.1025, 0.0239, 0.0926, 0.1366], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0558, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 19:04:28,193 INFO [train.py:968] (1/2) Epoch 25, batch 1900, libri_loss[loss=0.2004, simple_loss=0.2879, pruned_loss=0.05642, over 29475.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3446, pruned_loss=0.09702, over 5697194.78 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3365, pruned_loss=0.08488, over 3691588.42 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3466, pruned_loss=0.09882, over 5682154.79 frames. ], batch size: 70, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:05:00,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-12 19:05:10,823 INFO [optim.py:369] (1/2) 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 19:05:14,748 INFO [train.py:968] (1/2) Epoch 25, batch 1950, libri_loss[loss=0.2577, simple_loss=0.3436, pruned_loss=0.08593, over 29544.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3415, pruned_loss=0.09476, over 5695619.00 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3367, pruned_loss=0.08486, over 3757430.19 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3431, pruned_loss=0.09653, over 5677599.80 frames. ], batch size: 89, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:05:58,248 INFO [train.py:968] (1/2) Epoch 25, batch 2000, giga_loss[loss=0.2842, simple_loss=0.3402, pruned_loss=0.1141, over 26555.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3366, pruned_loss=0.09189, over 5685108.08 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3368, pruned_loss=0.08501, over 3827797.24 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3379, pruned_loss=0.09352, over 5675376.78 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:06:07,796 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5139, 1.6540, 1.6772, 1.3649], device='cuda:1'), covar=tensor([0.3135, 0.2890, 0.2198, 0.3040], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1933, 0.1856, 0.2003], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 19:06:15,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5706, 1.5455, 1.2200, 1.1591], device='cuda:1'), covar=tensor([0.0821, 0.0422, 0.0913, 0.1166], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0450, 0.0525, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 19:06:38,522 INFO [optim.py:369] (1/2) 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 19:06:43,931 INFO [train.py:968] (1/2) Epoch 25, batch 2050, giga_loss[loss=0.2386, simple_loss=0.3126, pruned_loss=0.08229, over 28218.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3318, pruned_loss=0.08976, over 5666689.39 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3371, pruned_loss=0.08514, over 3859747.02 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3327, pruned_loss=0.09107, over 5664189.63 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:07:27,112 INFO [train.py:968] (1/2) Epoch 25, batch 2100, giga_loss[loss=0.2635, simple_loss=0.3432, pruned_loss=0.09193, over 28836.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.328, pruned_loss=0.08796, over 5663285.67 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3366, pruned_loss=0.08501, over 3929453.66 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3286, pruned_loss=0.08921, over 5654347.30 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:08:04,229 INFO [optim.py:369] (1/2) 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 19:08:05,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1095987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:08:07,594 INFO [train.py:968] (1/2) Epoch 25, batch 2150, giga_loss[loss=0.2671, simple_loss=0.3461, pruned_loss=0.094, over 28701.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3291, pruned_loss=0.08812, over 5672907.81 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3359, pruned_loss=0.08447, over 3987999.58 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3299, pruned_loss=0.08954, over 5659328.64 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:08:39,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1096030.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:08:46,221 INFO [train.py:968] (1/2) Epoch 25, batch 2200, giga_loss[loss=0.237, simple_loss=0.3158, pruned_loss=0.0791, over 29023.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3301, pruned_loss=0.08809, over 5680372.37 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.336, pruned_loss=0.08459, over 4034499.10 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3305, pruned_loss=0.08924, over 5672939.86 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:09:24,193 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 2250, libri_loss[loss=0.2493, simple_loss=0.3418, pruned_loss=0.07842, over 28609.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3294, pruned_loss=0.08777, over 5680622.93 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3368, pruned_loss=0.08476, over 4105503.80 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3289, pruned_loss=0.08874, over 5675926.15 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:09:41,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3541, 1.6084, 1.3257, 0.9228], device='cuda:1'), covar=tensor([0.2828, 0.2870, 0.3366, 0.2651], device='cuda:1'), in_proj_covar=tensor([0.1556, 0.1122, 0.1373, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:09:55,990 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1096130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:09:59,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1096133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:10:03,913 INFO [train.py:968] (1/2) Epoch 25, batch 2300, giga_loss[loss=0.2465, simple_loss=0.3172, pruned_loss=0.08787, over 29099.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3275, pruned_loss=0.08648, over 5693355.80 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3371, pruned_loss=0.08468, over 4148950.54 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3266, pruned_loss=0.08735, over 5685696.46 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:10:25,210 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1096162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:10:33,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1096173.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:10:35,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1096176.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:10:43,487 INFO [optim.py:369] (1/2) 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 19:10:46,266 INFO [train.py:968] (1/2) Epoch 25, batch 2350, giga_loss[loss=0.2444, simple_loss=0.321, pruned_loss=0.0839, over 27944.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3248, pruned_loss=0.08535, over 5700032.47 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3372, pruned_loss=0.08467, over 4172613.16 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3239, pruned_loss=0.08605, over 5693546.51 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:10:52,041 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1096196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:10:59,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1096205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:10:59,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6009, 1.5059, 1.7752, 1.3506], device='cuda:1'), covar=tensor([0.1657, 0.2454, 0.1359, 0.1632], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0709, 0.0967, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 19:11:02,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0425, 1.6278, 5.2342, 3.7864], device='cuda:1'), covar=tensor([0.1581, 0.2642, 0.0378, 0.0820], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0658, 0.0973, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 19:11:27,481 INFO [train.py:968] (1/2) Epoch 25, batch 2400, giga_loss[loss=0.2165, simple_loss=0.2918, pruned_loss=0.07061, over 28589.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3235, pruned_loss=0.08497, over 5702134.55 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3375, pruned_loss=0.08461, over 4215632.10 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3222, pruned_loss=0.08559, over 5692100.17 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:12:06,179 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 2450, giga_loss[loss=0.2458, simple_loss=0.3221, pruned_loss=0.08474, over 28705.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3216, pruned_loss=0.08402, over 5699351.92 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3384, pruned_loss=0.08492, over 4246140.49 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3196, pruned_loss=0.08427, over 5697639.05 frames. ], batch size: 66, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:12:17,721 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0158, 2.3689, 1.8714, 2.3652], device='cuda:1'), covar=tensor([0.2676, 0.2687, 0.3062, 0.2308], device='cuda:1'), in_proj_covar=tensor([0.1556, 0.1122, 0.1372, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:12:45,457 INFO [train.py:968] (1/2) Epoch 25, batch 2500, libri_loss[loss=0.2721, simple_loss=0.3601, pruned_loss=0.09205, over 29681.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3203, pruned_loss=0.08333, over 5696199.93 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3394, pruned_loss=0.0853, over 4288600.06 frames. ], giga_tot_loss[loss=0.242, simple_loss=0.3175, pruned_loss=0.08322, over 5701018.80 frames. ], batch size: 88, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:13:21,422 INFO [optim.py:369] (1/2) 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 19:13:23,139 INFO [train.py:968] (1/2) Epoch 25, batch 2550, giga_loss[loss=0.2164, simple_loss=0.3034, pruned_loss=0.06468, over 28621.00 frames. ], tot_loss[loss=0.2401, simple_loss=0.317, pruned_loss=0.08157, over 5706605.51 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3396, pruned_loss=0.08527, over 4310170.46 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3144, pruned_loss=0.08147, over 5709985.41 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:13:31,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3708, 2.0474, 1.5979, 0.5570], device='cuda:1'), covar=tensor([0.5388, 0.2719, 0.4641, 0.6787], device='cuda:1'), in_proj_covar=tensor([0.1783, 0.1676, 0.1614, 0.1445], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 19:14:01,816 INFO [train.py:968] (1/2) Epoch 25, batch 2600, giga_loss[loss=0.2096, simple_loss=0.292, pruned_loss=0.06358, over 28158.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3158, pruned_loss=0.08072, over 5718485.33 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3398, pruned_loss=0.08521, over 4365460.37 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3128, pruned_loss=0.08056, over 5714901.64 frames. ], batch size: 77, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:14:38,072 INFO [optim.py:369] (1/2) 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 19:14:39,495 INFO [train.py:968] (1/2) Epoch 25, batch 2650, giga_loss[loss=0.2113, simple_loss=0.2844, pruned_loss=0.06913, over 28591.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3157, pruned_loss=0.08088, over 5720286.08 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3408, pruned_loss=0.08569, over 4413986.30 frames. ], giga_tot_loss[loss=0.2362, simple_loss=0.3118, pruned_loss=0.08029, over 5715755.96 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:15:19,907 INFO [train.py:968] (1/2) Epoch 25, batch 2700, giga_loss[loss=0.2576, simple_loss=0.3203, pruned_loss=0.09749, over 28682.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.3176, pruned_loss=0.0824, over 5724560.50 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.341, pruned_loss=0.08562, over 4458193.16 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3137, pruned_loss=0.08189, over 5716040.02 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:15:51,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1096571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:16:05,672 INFO [optim.py:369] (1/2) 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 19:16:07,536 INFO [train.py:968] (1/2) Epoch 25, batch 2750, giga_loss[loss=0.322, simple_loss=0.3844, pruned_loss=0.1297, over 27713.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3222, pruned_loss=0.08568, over 5714930.03 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3412, pruned_loss=0.08575, over 4455530.20 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.319, pruned_loss=0.08518, over 5715336.85 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:16:13,413 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-12 19:16:51,486 INFO [train.py:968] (1/2) Epoch 25, batch 2800, libri_loss[loss=0.2152, simple_loss=0.2996, pruned_loss=0.06535, over 28486.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3278, pruned_loss=0.08902, over 5713854.65 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3411, pruned_loss=0.08561, over 4502590.61 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3248, pruned_loss=0.08877, over 5710987.61 frames. ], batch size: 63, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:17:05,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2490, 2.4221, 1.3098, 1.4265], device='cuda:1'), covar=tensor([0.0976, 0.0367, 0.0852, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0557, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 19:17:33,903 INFO [optim.py:369] (1/2) 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 19:17:37,453 INFO [train.py:968] (1/2) Epoch 25, batch 2850, giga_loss[loss=0.3145, simple_loss=0.3679, pruned_loss=0.1305, over 23597.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3355, pruned_loss=0.09429, over 5698214.99 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3409, pruned_loss=0.08546, over 4530684.15 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3331, pruned_loss=0.09429, over 5692339.12 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:17:57,636 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1096714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:17:59,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1096717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:18:20,073 INFO [train.py:968] (1/2) Epoch 25, batch 2900, giga_loss[loss=0.2971, simple_loss=0.371, pruned_loss=0.1115, over 28726.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3396, pruned_loss=0.09593, over 5700145.09 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3404, pruned_loss=0.08535, over 4555486.68 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3379, pruned_loss=0.09623, over 5699464.57 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:18:26,033 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1096746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:19:06,116 INFO [optim.py:369] (1/2) 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 19:19:07,428 INFO [train.py:968] (1/2) Epoch 25, batch 2950, giga_loss[loss=0.2756, simple_loss=0.3622, pruned_loss=0.09445, over 28950.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3439, pruned_loss=0.09707, over 5704695.20 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3408, pruned_loss=0.08544, over 4581459.50 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3424, pruned_loss=0.09745, over 5701117.48 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:19:52,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7976, 1.8807, 1.6945, 1.8811], device='cuda:1'), covar=tensor([0.2173, 0.2160, 0.2151, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1119, 0.1369, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:19:53,273 INFO [train.py:968] (1/2) Epoch 25, batch 3000, giga_loss[loss=0.2775, simple_loss=0.3541, pruned_loss=0.1005, over 28819.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3498, pruned_loss=0.1008, over 5696731.55 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3407, pruned_loss=0.08538, over 4607492.85 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3487, pruned_loss=0.1014, over 5690426.01 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:19:53,273 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 19:20:00,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4428, 1.7589, 1.4049, 1.3017], device='cuda:1'), covar=tensor([0.2825, 0.2826, 0.3330, 0.2503], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1119, 0.1369, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:20:02,036 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 19:20:27,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1096868.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:20:42,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3584, 3.3447, 1.5497, 1.4989], device='cuda:1'), covar=tensor([0.1073, 0.0271, 0.0966, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0558, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 19:20:43,754 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 3050, libri_loss[loss=0.3118, simple_loss=0.3863, pruned_loss=0.1187, over 29520.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3527, pruned_loss=0.1022, over 5684443.17 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3409, pruned_loss=0.08554, over 4623621.03 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3519, pruned_loss=0.1028, over 5679188.97 frames. ], batch size: 81, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:20:46,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2699, 3.2523, 1.4974, 1.4567], device='cuda:1'), covar=tensor([0.1081, 0.0324, 0.0929, 0.1485], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0558, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 19:21:10,785 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1096920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:21:25,424 INFO [train.py:968] (1/2) Epoch 25, batch 3100, giga_loss[loss=0.2498, simple_loss=0.3339, pruned_loss=0.08283, over 29087.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3479, pruned_loss=0.09857, over 5696743.19 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3404, pruned_loss=0.08536, over 4661827.76 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3478, pruned_loss=0.09959, over 5686480.21 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:21:51,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4598, 1.5778, 1.6274, 1.2319], device='cuda:1'), covar=tensor([0.1836, 0.2829, 0.1576, 0.1971], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0711, 0.0968, 0.0864], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 19:22:06,748 INFO [optim.py:369] (1/2) 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 19:22:09,040 INFO [train.py:968] (1/2) Epoch 25, batch 3150, giga_loss[loss=0.2661, simple_loss=0.3527, pruned_loss=0.08969, over 28831.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3462, pruned_loss=0.09677, over 5707686.54 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3405, pruned_loss=0.08574, over 4715094.82 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3464, pruned_loss=0.09782, over 5692418.47 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:22:50,780 INFO [train.py:968] (1/2) Epoch 25, batch 3200, giga_loss[loss=0.3521, simple_loss=0.4015, pruned_loss=0.1513, over 26618.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3455, pruned_loss=0.09585, over 5709763.52 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3403, pruned_loss=0.08569, over 4737546.97 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3459, pruned_loss=0.0969, over 5695229.91 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:23:18,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-12 19:23:22,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2375, 1.5617, 1.2528, 1.0551], device='cuda:1'), covar=tensor([0.2699, 0.2849, 0.3166, 0.2451], device='cuda:1'), in_proj_covar=tensor([0.1556, 0.1121, 0.1373, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:23:27,183 INFO [optim.py:369] (1/2) 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 19:23:30,263 INFO [train.py:968] (1/2) Epoch 25, batch 3250, giga_loss[loss=0.2631, simple_loss=0.3435, pruned_loss=0.0913, over 28881.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3464, pruned_loss=0.09579, over 5717242.46 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.34, pruned_loss=0.08574, over 4785033.83 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3471, pruned_loss=0.09702, over 5701493.88 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:23:55,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8770, 1.5158, 5.2467, 3.6250], device='cuda:1'), covar=tensor([0.1617, 0.2831, 0.0376, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0659, 0.0974, 0.0938], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 19:24:10,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6855, 1.8215, 1.6124, 1.7061], device='cuda:1'), covar=tensor([0.2213, 0.2177, 0.2175, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1557, 0.1121, 0.1372, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:24:11,896 INFO [train.py:968] (1/2) Epoch 25, batch 3300, giga_loss[loss=0.261, simple_loss=0.3403, pruned_loss=0.09089, over 28886.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3471, pruned_loss=0.0958, over 5713781.53 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3397, pruned_loss=0.08556, over 4817612.03 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3481, pruned_loss=0.09724, over 5700421.23 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:24:52,941 INFO [optim.py:369] (1/2) 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 19:24:54,509 INFO [train.py:968] (1/2) Epoch 25, batch 3350, giga_loss[loss=0.2991, simple_loss=0.3644, pruned_loss=0.1169, over 28680.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3492, pruned_loss=0.09784, over 5702781.91 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3399, pruned_loss=0.08561, over 4829466.09 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.35, pruned_loss=0.09914, over 5697870.52 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:25:01,051 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1097196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:25:25,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5371, 1.4894, 1.1933, 1.0856], device='cuda:1'), covar=tensor([0.0768, 0.0440, 0.0893, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0450, 0.0525, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 19:25:36,234 INFO [train.py:968] (1/2) Epoch 25, batch 3400, giga_loss[loss=0.2637, simple_loss=0.3398, pruned_loss=0.09377, over 28853.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3513, pruned_loss=0.09996, over 5704874.10 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3405, pruned_loss=0.08604, over 4839957.37 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3516, pruned_loss=0.1008, over 5699447.82 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:25:39,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097243.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:26:19,909 INFO [optim.py:369] (1/2) 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 19:26:21,111 INFO [train.py:968] (1/2) Epoch 25, batch 3450, giga_loss[loss=0.3255, simple_loss=0.3843, pruned_loss=0.1333, over 28859.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3519, pruned_loss=0.1007, over 5715500.01 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3402, pruned_loss=0.08586, over 4854097.86 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3524, pruned_loss=0.1017, over 5710173.57 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:26:22,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4672, 2.1538, 1.6197, 0.7830], device='cuda:1'), covar=tensor([0.6950, 0.3323, 0.4413, 0.6849], device='cuda:1'), in_proj_covar=tensor([0.1785, 0.1677, 0.1619, 0.1446], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 19:26:25,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:26:27,642 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-12 19:26:28,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.4078, 3.2732, 3.1486, 1.2040], device='cuda:1'), covar=tensor([0.1045, 0.1116, 0.1039, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1255, 0.1162, 0.0977, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 19:26:46,420 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1097322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:27:01,369 INFO [train.py:968] (1/2) Epoch 25, batch 3500, giga_loss[loss=0.2515, simple_loss=0.3398, pruned_loss=0.08161, over 28437.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3523, pruned_loss=0.1008, over 5713117.09 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3402, pruned_loss=0.08587, over 4859762.91 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3529, pruned_loss=0.1018, over 5714590.71 frames. ], batch size: 65, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:27:11,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2574, 1.7613, 1.2625, 0.5214], device='cuda:1'), covar=tensor([0.4960, 0.2535, 0.3893, 0.6690], device='cuda:1'), in_proj_covar=tensor([0.1784, 0.1674, 0.1617, 0.1444], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 19:27:26,626 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5619, 1.7826, 1.4182, 1.6088], device='cuda:1'), covar=tensor([0.2730, 0.2729, 0.3092, 0.2523], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1118, 0.1369, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:27:37,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097386.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:27:38,572 INFO [optim.py:369] (1/2) 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 19:27:39,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097389.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:27:39,998 INFO [train.py:968] (1/2) Epoch 25, batch 3550, giga_loss[loss=0.2501, simple_loss=0.3387, pruned_loss=0.08075, over 28606.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3517, pruned_loss=0.09963, over 5714492.50 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3399, pruned_loss=0.08577, over 4895002.61 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3527, pruned_loss=0.1009, over 5710488.45 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:28:04,808 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1097418.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:28:22,133 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:28:23,116 INFO [train.py:968] (1/2) Epoch 25, batch 3600, giga_loss[loss=0.2699, simple_loss=0.3545, pruned_loss=0.09261, over 28689.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3522, pruned_loss=0.09908, over 5708402.63 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3401, pruned_loss=0.08588, over 4904008.75 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3531, pruned_loss=0.1002, over 5711701.83 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:28:24,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:28:40,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9192, 1.3659, 1.2556, 1.1637], device='cuda:1'), covar=tensor([0.2351, 0.1716, 0.2360, 0.1967], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0750, 0.0721, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 19:28:48,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1097470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:28:59,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1097482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:29:04,880 INFO [optim.py:369] (1/2) 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 19:29:04,893 INFO [train.py:968] (1/2) Epoch 25, batch 3650, giga_loss[loss=0.2622, simple_loss=0.3512, pruned_loss=0.08662, over 28927.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3518, pruned_loss=0.09851, over 5712753.82 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3405, pruned_loss=0.08612, over 4917987.92 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3525, pruned_loss=0.09944, over 5713682.03 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:29:45,594 INFO [train.py:968] (1/2) Epoch 25, batch 3700, giga_loss[loss=0.2782, simple_loss=0.3485, pruned_loss=0.1039, over 27935.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3499, pruned_loss=0.09782, over 5707371.87 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3407, pruned_loss=0.08633, over 4923985.30 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3504, pruned_loss=0.09861, over 5716142.97 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:29:52,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-12 19:30:10,583 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:30:22,907 INFO [train.py:968] (1/2) Epoch 25, batch 3750, giga_loss[loss=0.2218, simple_loss=0.305, pruned_loss=0.06935, over 28337.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3474, pruned_loss=0.09657, over 5710375.60 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3405, pruned_loss=0.08625, over 4950131.12 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3481, pruned_loss=0.09757, over 5714902.95 frames. ], batch size: 65, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:30:23,510 INFO [optim.py:369] (1/2) 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 19:31:00,444 INFO [train.py:968] (1/2) Epoch 25, batch 3800, giga_loss[loss=0.2465, simple_loss=0.3279, pruned_loss=0.08254, over 28961.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3458, pruned_loss=0.09559, over 5723656.59 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3405, pruned_loss=0.08622, over 4988004.27 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3466, pruned_loss=0.09678, over 5720133.24 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:31:19,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-12 19:31:42,779 INFO [train.py:968] (1/2) Epoch 25, batch 3850, giga_loss[loss=0.2563, simple_loss=0.335, pruned_loss=0.08878, over 29069.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3466, pruned_loss=0.09651, over 5724741.47 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3405, pruned_loss=0.08612, over 5005923.31 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3474, pruned_loss=0.09771, over 5719079.26 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:31:45,408 INFO [optim.py:369] (1/2) 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 19:31:49,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:01,197 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:03,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:15,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5955, 1.8500, 1.5084, 1.5471], device='cuda:1'), covar=tensor([0.2845, 0.2806, 0.3227, 0.2505], device='cuda:1'), in_proj_covar=tensor([0.1560, 0.1124, 0.1374, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:32:17,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4491, 2.0740, 1.4899, 0.7657], device='cuda:1'), covar=tensor([0.6099, 0.2844, 0.4334, 0.6309], device='cuda:1'), in_proj_covar=tensor([0.1777, 0.1664, 0.1607, 0.1437], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 19:32:21,996 INFO [train.py:968] (1/2) Epoch 25, batch 3900, giga_loss[loss=0.2446, simple_loss=0.3314, pruned_loss=0.07891, over 28659.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3481, pruned_loss=0.09738, over 5731309.54 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3404, pruned_loss=0.08612, over 5023048.26 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3489, pruned_loss=0.09852, over 5724374.30 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:32:28,465 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1097746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:32:34,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-12 19:33:01,517 INFO [train.py:968] (1/2) Epoch 25, batch 3950, giga_loss[loss=0.2828, simple_loss=0.3522, pruned_loss=0.1067, over 28744.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3476, pruned_loss=0.0967, over 5728702.27 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.34, pruned_loss=0.08606, over 5053475.29 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3488, pruned_loss=0.09803, over 5717375.58 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:33:02,401 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 4000, libri_loss[loss=0.2399, simple_loss=0.3148, pruned_loss=0.08248, over 29578.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3472, pruned_loss=0.09632, over 5728961.47 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3402, pruned_loss=0.08634, over 5069356.59 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3481, pruned_loss=0.09733, over 5717592.54 frames. ], batch size: 74, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:33:44,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1097840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:33:46,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1097843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:33:55,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1097857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:34:08,591 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 4050, giga_loss[loss=0.2643, simple_loss=0.343, pruned_loss=0.09282, over 28578.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3447, pruned_loss=0.09517, over 5729382.12 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3395, pruned_loss=0.08599, over 5102206.22 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3463, pruned_loss=0.09669, over 5716536.51 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:34:22,327 INFO [optim.py:369] (1/2) 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 19:34:50,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1887, 4.0015, 3.8017, 1.7264], device='cuda:1'), covar=tensor([0.0613, 0.0801, 0.0819, 0.2234], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.1156, 0.0973, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 19:35:01,520 INFO [train.py:968] (1/2) Epoch 25, batch 4100, giga_loss[loss=0.2572, simple_loss=0.3292, pruned_loss=0.09259, over 28858.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3423, pruned_loss=0.09417, over 5723787.06 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3397, pruned_loss=0.08609, over 5118171.35 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3435, pruned_loss=0.09546, over 5710172.77 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:35:07,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5051, 2.2395, 1.6386, 0.8523], device='cuda:1'), covar=tensor([0.5491, 0.2550, 0.3815, 0.4924], device='cuda:1'), in_proj_covar=tensor([0.1775, 0.1661, 0.1605, 0.1436], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 19:35:08,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5611, 1.8575, 1.9349, 1.4595], device='cuda:1'), covar=tensor([0.3822, 0.2866, 0.2896, 0.3470], device='cuda:1'), in_proj_covar=tensor([0.2004, 0.1942, 0.1860, 0.2009], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 19:35:09,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4155, 1.3489, 1.1961, 1.5216], device='cuda:1'), covar=tensor([0.0798, 0.0360, 0.0358, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0119, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 19:35:35,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1398, 3.9635, 3.7758, 1.8550], device='cuda:1'), covar=tensor([0.0609, 0.0749, 0.0724, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.1156, 0.0973, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 19:35:39,702 INFO [train.py:968] (1/2) Epoch 25, batch 4150, giga_loss[loss=0.2259, simple_loss=0.3094, pruned_loss=0.07117, over 28753.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3388, pruned_loss=0.0922, over 5720192.14 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3393, pruned_loss=0.08588, over 5131465.50 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3401, pruned_loss=0.09351, over 5708178.26 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:35:41,285 INFO [optim.py:369] (1/2) 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 19:35:48,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:35:48,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1098000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:35:50,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1098003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:36:08,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7433, 4.5813, 4.3875, 1.7679], device='cuda:1'), covar=tensor([0.0613, 0.0743, 0.0860, 0.2123], device='cuda:1'), in_proj_covar=tensor([0.1255, 0.1158, 0.0975, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 19:36:13,593 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1098032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:36:15,570 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:36:18,738 INFO [train.py:968] (1/2) Epoch 25, batch 4200, giga_loss[loss=0.2392, simple_loss=0.3112, pruned_loss=0.08361, over 28528.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3391, pruned_loss=0.09325, over 5714612.45 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3395, pruned_loss=0.0861, over 5149046.77 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3399, pruned_loss=0.0943, over 5704829.63 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:37:01,350 INFO [train.py:968] (1/2) Epoch 25, batch 4250, giga_loss[loss=0.2368, simple_loss=0.3143, pruned_loss=0.07965, over 28746.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3376, pruned_loss=0.09299, over 5709294.19 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3396, pruned_loss=0.08616, over 5152777.53 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3381, pruned_loss=0.09381, over 5700818.67 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:37:02,192 INFO [optim.py:369] (1/2) 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 19:37:42,107 INFO [train.py:968] (1/2) Epoch 25, batch 4300, libri_loss[loss=0.2464, simple_loss=0.3244, pruned_loss=0.08422, over 29641.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3354, pruned_loss=0.09206, over 5708557.85 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3398, pruned_loss=0.08633, over 5168610.63 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3356, pruned_loss=0.09273, over 5699878.55 frames. ], batch size: 73, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:38:21,264 INFO [train.py:968] (1/2) Epoch 25, batch 4350, libri_loss[loss=0.2178, simple_loss=0.3025, pruned_loss=0.06655, over 29631.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3341, pruned_loss=0.09212, over 5717029.81 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.34, pruned_loss=0.08647, over 5186854.52 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.334, pruned_loss=0.09267, over 5705832.51 frames. ], batch size: 69, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:38:22,397 INFO [optim.py:369] (1/2) 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 19:39:00,231 INFO [train.py:968] (1/2) Epoch 25, batch 4400, giga_loss[loss=0.2719, simple_loss=0.3453, pruned_loss=0.09925, over 28215.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3332, pruned_loss=0.09212, over 5716090.34 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.34, pruned_loss=0.08652, over 5201036.03 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.333, pruned_loss=0.09262, over 5703873.26 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:39:09,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5502, 1.7151, 1.7539, 1.5174], device='cuda:1'), covar=tensor([0.3699, 0.2830, 0.2222, 0.2965], device='cuda:1'), in_proj_covar=tensor([0.2015, 0.1952, 0.1873, 0.2018], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 19:39:34,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5691, 1.6395, 1.7606, 1.3419], device='cuda:1'), covar=tensor([0.1976, 0.2673, 0.1643, 0.1896], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0713, 0.0970, 0.0866], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 19:39:37,433 INFO [train.py:968] (1/2) Epoch 25, batch 4450, giga_loss[loss=0.2302, simple_loss=0.3046, pruned_loss=0.07795, over 28763.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3313, pruned_loss=0.09089, over 5721559.09 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3399, pruned_loss=0.08656, over 5221939.83 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.331, pruned_loss=0.09139, over 5706838.28 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:39:38,988 INFO [optim.py:369] (1/2) 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 19:39:57,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-12 19:40:19,349 INFO [train.py:968] (1/2) Epoch 25, batch 4500, giga_loss[loss=0.308, simple_loss=0.3774, pruned_loss=0.1192, over 28671.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3341, pruned_loss=0.09158, over 5719713.66 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3401, pruned_loss=0.08658, over 5227146.42 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3337, pruned_loss=0.092, over 5707978.37 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:40:49,714 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 4550, libri_loss[loss=0.2317, simple_loss=0.3139, pruned_loss=0.07471, over 29493.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3371, pruned_loss=0.09277, over 5713495.17 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3405, pruned_loss=0.08694, over 5249558.98 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3363, pruned_loss=0.09301, over 5699054.82 frames. ], batch size: 70, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:41:02,939 INFO [optim.py:369] (1/2) 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 19:41:18,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1098410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:41:43,522 INFO [train.py:968] (1/2) Epoch 25, batch 4600, giga_loss[loss=0.2453, simple_loss=0.3204, pruned_loss=0.08509, over 28871.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3394, pruned_loss=0.09336, over 5708294.35 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3406, pruned_loss=0.08707, over 5255593.34 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3386, pruned_loss=0.09359, over 5701034.13 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:42:03,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6383, 1.7229, 1.4635, 1.7619], device='cuda:1'), covar=tensor([0.2975, 0.3044, 0.3327, 0.2666], device='cuda:1'), in_proj_covar=tensor([0.1558, 0.1124, 0.1374, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 19:42:26,532 INFO [train.py:968] (1/2) Epoch 25, batch 4650, giga_loss[loss=0.2768, simple_loss=0.3505, pruned_loss=0.1015, over 28686.00 frames. ], tot_loss[loss=0.264, simple_loss=0.341, pruned_loss=0.09343, over 5705201.92 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3407, pruned_loss=0.08719, over 5279409.69 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3403, pruned_loss=0.09375, over 5693424.71 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:42:27,888 INFO [optim.py:369] (1/2) 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 19:42:48,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5386, 1.8285, 1.3165, 1.4198], device='cuda:1'), covar=tensor([0.0981, 0.0571, 0.1014, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0449, 0.0523, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 19:42:52,376 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1098518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:42:54,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1098521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:08,199 INFO [train.py:968] (1/2) Epoch 25, batch 4700, giga_loss[loss=0.2572, simple_loss=0.3259, pruned_loss=0.09422, over 28812.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3402, pruned_loss=0.09251, over 5703001.11 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3408, pruned_loss=0.08727, over 5294419.33 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3395, pruned_loss=0.09285, over 5691072.88 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:43:17,884 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1098550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:20,417 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1098553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:21,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4067, 3.4660, 1.6053, 1.4764], device='cuda:1'), covar=tensor([0.0952, 0.0323, 0.0919, 0.1304], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0555, 0.0393, 0.0431], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 19:43:22,471 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:44,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1098585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:43:48,211 INFO [train.py:968] (1/2) Epoch 25, batch 4750, giga_loss[loss=0.2289, simple_loss=0.3109, pruned_loss=0.07347, over 28867.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.339, pruned_loss=0.09236, over 5704211.78 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3412, pruned_loss=0.08767, over 5301451.17 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3381, pruned_loss=0.09239, over 5695635.78 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:43:49,537 INFO [optim.py:369] (1/2) 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 19:44:14,830 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:44:27,005 INFO [train.py:968] (1/2) Epoch 25, batch 4800, libri_loss[loss=0.2718, simple_loss=0.3594, pruned_loss=0.09206, over 29286.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3388, pruned_loss=0.09192, over 5704208.16 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3406, pruned_loss=0.08713, over 5323662.09 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3385, pruned_loss=0.09256, over 5693017.32 frames. ], batch size: 94, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:45:08,283 INFO [train.py:968] (1/2) Epoch 25, batch 4850, giga_loss[loss=0.2704, simple_loss=0.3591, pruned_loss=0.09083, over 28760.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3408, pruned_loss=0.09334, over 5702424.26 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3407, pruned_loss=0.08723, over 5329491.97 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3404, pruned_loss=0.09382, over 5691820.40 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:45:10,794 INFO [optim.py:369] (1/2) 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 19:45:51,788 INFO [train.py:968] (1/2) Epoch 25, batch 4900, giga_loss[loss=0.3035, simple_loss=0.3657, pruned_loss=0.1207, over 28930.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3454, pruned_loss=0.09613, over 5703620.47 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3409, pruned_loss=0.08728, over 5335052.75 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.345, pruned_loss=0.09654, over 5693557.22 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:46:18,326 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.40 vs. limit=5.0
+2023-03-12 19:46:32,080 INFO [train.py:968] (1/2) Epoch 25, batch 4950, giga_loss[loss=0.2939, simple_loss=0.3711, pruned_loss=0.1083, over 28324.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3471, pruned_loss=0.09623, over 5714362.46 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3408, pruned_loss=0.08721, over 5350184.85 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3469, pruned_loss=0.09685, over 5702536.42 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:46:33,408 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 5000, giga_loss[loss=0.2824, simple_loss=0.3586, pruned_loss=0.1031, over 28631.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3471, pruned_loss=0.09592, over 5719558.49 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3409, pruned_loss=0.08716, over 5368351.73 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3472, pruned_loss=0.09676, over 5704959.59 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:47:51,257 INFO [train.py:968] (1/2) Epoch 25, batch 5050, giga_loss[loss=0.2883, simple_loss=0.3552, pruned_loss=0.1107, over 28880.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3467, pruned_loss=0.09555, over 5724091.23 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3411, pruned_loss=0.08735, over 5370739.97 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3466, pruned_loss=0.0961, over 5712136.68 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:47:52,757 INFO [optim.py:369] (1/2) 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 19:48:29,331 INFO [train.py:968] (1/2) Epoch 25, batch 5100, libri_loss[loss=0.2358, simple_loss=0.3174, pruned_loss=0.07708, over 29646.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3475, pruned_loss=0.0964, over 5722266.48 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3421, pruned_loss=0.08812, over 5377541.07 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3467, pruned_loss=0.0964, over 5716389.19 frames. ], batch size: 73, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:48:34,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1098947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:49:09,269 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1098989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:49:09,746 INFO [train.py:968] (1/2) Epoch 25, batch 5150, libri_loss[loss=0.298, simple_loss=0.3717, pruned_loss=0.1121, over 29644.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3465, pruned_loss=0.09613, over 5723558.39 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3424, pruned_loss=0.08828, over 5393560.69 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3457, pruned_loss=0.09622, over 5713330.94 frames. ], batch size: 88, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:49:11,702 INFO [optim.py:369] (1/2) 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 19:49:16,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:49:49,711 INFO [train.py:968] (1/2) Epoch 25, batch 5200, giga_loss[loss=0.2369, simple_loss=0.3187, pruned_loss=0.07753, over 29005.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3433, pruned_loss=0.09455, over 5727300.47 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3428, pruned_loss=0.08847, over 5400091.86 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3423, pruned_loss=0.09454, over 5717826.87 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:49:50,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 19:50:28,585 INFO [train.py:968] (1/2) Epoch 25, batch 5250, giga_loss[loss=0.2286, simple_loss=0.3196, pruned_loss=0.06879, over 28955.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3404, pruned_loss=0.09273, over 5731223.62 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3425, pruned_loss=0.0882, over 5416755.14 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3399, pruned_loss=0.09315, over 5720057.96 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:50:28,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:50:30,701 INFO [optim.py:369] (1/2) 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 19:50:31,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:50:32,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4911, 1.5701, 1.7254, 1.5199], device='cuda:1'), covar=tensor([0.3656, 0.3153, 0.2592, 0.3080], device='cuda:1'), in_proj_covar=tensor([0.2014, 0.1954, 0.1870, 0.2012], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 19:50:54,469 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:01,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4890, 1.5902, 1.2309, 1.1988], device='cuda:1'), covar=tensor([0.0886, 0.0582, 0.0933, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0450, 0.0523, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 19:51:08,420 INFO [train.py:968] (1/2) Epoch 25, batch 5300, giga_loss[loss=0.2602, simple_loss=0.3464, pruned_loss=0.08696, over 28254.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3396, pruned_loss=0.09189, over 5724297.68 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3425, pruned_loss=0.08811, over 5428924.24 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3392, pruned_loss=0.09239, over 5711280.30 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:51:11,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:14,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:15,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6666, 1.8847, 1.3157, 1.5390], device='cuda:1'), covar=tensor([0.1029, 0.0763, 0.1187, 0.1280], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0450, 0.0523, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 19:51:40,250 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:51:52,483 INFO [train.py:968] (1/2) Epoch 25, batch 5350, giga_loss[loss=0.2681, simple_loss=0.3517, pruned_loss=0.09223, over 28566.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.341, pruned_loss=0.09124, over 5717820.88 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3427, pruned_loss=0.08823, over 5433727.90 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3405, pruned_loss=0.09157, over 5706086.96 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:51:54,534 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 5400, giga_loss[loss=0.2385, simple_loss=0.3209, pruned_loss=0.07802, over 28915.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3417, pruned_loss=0.09249, over 5710171.97 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3427, pruned_loss=0.08818, over 5436819.19 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3413, pruned_loss=0.09282, over 5700589.39 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:52:47,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1435, 1.5087, 1.5770, 1.3194], device='cuda:1'), covar=tensor([0.2106, 0.1697, 0.2282, 0.2005], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0748, 0.0722, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 19:52:52,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:52:59,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:53:06,558 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099280.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 19:53:12,718 INFO [train.py:968] (1/2) Epoch 25, batch 5450, giga_loss[loss=0.288, simple_loss=0.3536, pruned_loss=0.1112, over 28966.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3404, pruned_loss=0.09244, over 5709075.16 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3425, pruned_loss=0.08818, over 5444609.74 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3402, pruned_loss=0.09286, over 5704134.41 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:53:16,525 INFO [optim.py:369] (1/2) 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 19:53:36,621 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:53:56,002 INFO [train.py:968] (1/2) Epoch 25, batch 5500, giga_loss[loss=0.3099, simple_loss=0.3683, pruned_loss=0.1258, over 27698.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.34, pruned_loss=0.0939, over 5690016.92 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.343, pruned_loss=0.08847, over 5439101.86 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3394, pruned_loss=0.0941, over 5696704.58 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:54:13,050 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 5550, giga_loss[loss=0.2481, simple_loss=0.3162, pruned_loss=0.09002, over 28258.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3388, pruned_loss=0.0943, over 5695009.27 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3433, pruned_loss=0.08865, over 5450448.06 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.338, pruned_loss=0.09443, over 5695952.28 frames. ], batch size: 65, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:54:36,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-12 19:54:37,083 INFO [optim.py:369] (1/2) 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 19:55:00,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:55:14,525 INFO [train.py:968] (1/2) Epoch 25, batch 5600, giga_loss[loss=0.2879, simple_loss=0.3609, pruned_loss=0.1074, over 28850.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3366, pruned_loss=0.09345, over 5700219.41 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3431, pruned_loss=0.08857, over 5458309.29 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3359, pruned_loss=0.09376, over 5699562.03 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 19:55:26,280 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:55:55,494 INFO [train.py:968] (1/2) Epoch 25, batch 5650, giga_loss[loss=0.2451, simple_loss=0.3296, pruned_loss=0.08031, over 28873.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3365, pruned_loss=0.09325, over 5710985.57 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3436, pruned_loss=0.08875, over 5471234.50 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3353, pruned_loss=0.09353, over 5707218.01 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:55:58,879 INFO [optim.py:369] (1/2) 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 19:56:08,169 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:56:11,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:56:24,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1525, 3.0287, 2.9165, 1.7492], device='cuda:1'), covar=tensor([0.0748, 0.0894, 0.0819, 0.1634], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1165, 0.0981, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 19:56:31,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1650, 1.2509, 1.1291, 0.8626], device='cuda:1'), covar=tensor([0.1003, 0.0557, 0.1130, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0449, 0.0522, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 19:56:34,317 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:56:34,770 INFO [train.py:968] (1/2) Epoch 25, batch 5700, giga_loss[loss=0.2035, simple_loss=0.2758, pruned_loss=0.06558, over 28539.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3335, pruned_loss=0.09219, over 5709988.13 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.344, pruned_loss=0.08904, over 5471421.42 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3321, pruned_loss=0.09226, over 5712678.03 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:56:36,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-12 19:57:08,360 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:57:14,121 INFO [train.py:968] (1/2) Epoch 25, batch 5750, libri_loss[loss=0.2254, simple_loss=0.3113, pruned_loss=0.06974, over 29565.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3282, pruned_loss=0.08945, over 5714561.28 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08869, over 5477881.60 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3273, pruned_loss=0.08984, over 5713908.59 frames. ], batch size: 74, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 19:57:17,270 INFO [optim.py:369] (1/2) 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 19:57:35,558 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099618.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:57:51,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:57:53,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:57:55,283 INFO [train.py:968] (1/2) Epoch 25, batch 5800, giga_loss[loss=0.2557, simple_loss=0.3194, pruned_loss=0.09597, over 29031.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3282, pruned_loss=0.08943, over 5716264.11 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08886, over 5487768.47 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.327, pruned_loss=0.0896, over 5712300.80 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:58:00,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:58:06,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099655.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 19:58:28,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 19:58:31,191 INFO [train.py:968] (1/2) Epoch 25, batch 5850, giga_loss[loss=0.2591, simple_loss=0.3311, pruned_loss=0.09351, over 28641.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.33, pruned_loss=0.08969, over 5728135.77 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08894, over 5506484.52 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3284, pruned_loss=0.08978, over 5717308.11 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:58:34,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:58:36,176 INFO [optim.py:369] (1/2) 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 19:59:12,737 INFO [train.py:968] (1/2) Epoch 25, batch 5900, giga_loss[loss=0.2679, simple_loss=0.3502, pruned_loss=0.0928, over 28831.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3337, pruned_loss=0.0908, over 5732102.83 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08892, over 5516569.97 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3321, pruned_loss=0.09093, over 5719354.75 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:59:43,077 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:45,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:48,246 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:53,039 INFO [train.py:968] (1/2) Epoch 25, batch 5950, giga_loss[loss=0.2539, simple_loss=0.335, pruned_loss=0.08642, over 28751.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3378, pruned_loss=0.09265, over 5720808.79 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3441, pruned_loss=0.08916, over 5519556.08 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3359, pruned_loss=0.09263, over 5716034.50 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 19:59:53,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:55,919 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:58,327 INFO [optim.py:369] (1/2) 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 19:59:59,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 19:59:59,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099798.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:00:01,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099801.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:00:11,806 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:19,588 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:23,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:26,364 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:00:33,216 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:35,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:00:35,591 INFO [train.py:968] (1/2) Epoch 25, batch 6000, giga_loss[loss=0.2591, simple_loss=0.334, pruned_loss=0.09211, over 28808.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3402, pruned_loss=0.09371, over 5719848.44 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3442, pruned_loss=0.08921, over 5527184.42 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3385, pruned_loss=0.09373, over 5712924.16 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:00:35,591 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 20:00:44,414 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 20:01:08,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:01:24,591 INFO [train.py:968] (1/2) Epoch 25, batch 6050, giga_loss[loss=0.2653, simple_loss=0.3451, pruned_loss=0.09279, over 28887.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3421, pruned_loss=0.09474, over 5720132.29 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3441, pruned_loss=0.0892, over 5538154.73 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3407, pruned_loss=0.09491, over 5709528.99 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:01:28,579 INFO [optim.py:369] (1/2) 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 20:01:54,134 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-12 20:02:08,598 INFO [train.py:968] (1/2) Epoch 25, batch 6100, giga_loss[loss=0.3455, simple_loss=0.4011, pruned_loss=0.1449, over 28714.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3476, pruned_loss=0.0996, over 5705700.03 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08897, over 5536820.15 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3468, pruned_loss=0.1001, over 5702016.92 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:02:09,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:13,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:21,059 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:34,257 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1099970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:36,554 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1099973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:36,594 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1099973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:02:52,861 INFO [train.py:968] (1/2) Epoch 25, batch 6150, giga_loss[loss=0.3271, simple_loss=0.3933, pruned_loss=0.1304, over 28652.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3521, pruned_loss=0.1034, over 5701240.25 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3431, pruned_loss=0.08866, over 5547717.02 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3522, pruned_loss=0.1044, over 5693839.18 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:02:56,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1099993.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:02:58,013 INFO [optim.py:369] (1/2) 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 20:03:00,393 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1099999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:03:02,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:03:05,645 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4018, 1.7749, 1.0780, 1.3505], device='cuda:1'), covar=tensor([0.1319, 0.0732, 0.1604, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0452, 0.0524, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 20:03:10,956 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:03:39,563 INFO [train.py:968] (1/2) Epoch 25, batch 6200, giga_loss[loss=0.3404, simple_loss=0.4061, pruned_loss=0.1374, over 28321.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3593, pruned_loss=0.1087, over 5679111.26 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3427, pruned_loss=0.08859, over 5547020.85 frames. ], giga_tot_loss[loss=0.2901, simple_loss=0.36, pruned_loss=0.11, over 5676144.03 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:04:29,205 INFO [train.py:968] (1/2) Epoch 25, batch 6250, giga_loss[loss=0.3215, simple_loss=0.3821, pruned_loss=0.1304, over 28302.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3661, pruned_loss=0.1144, over 5679011.02 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.343, pruned_loss=0.08855, over 5555739.24 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3669, pruned_loss=0.116, over 5671833.64 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:04:33,987 INFO [optim.py:369] (1/2) 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 20:04:36,172 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:04:39,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:04:49,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5342, 1.9836, 2.0316, 1.6815], device='cuda:1'), covar=tensor([0.2022, 0.1715, 0.1839, 0.1930], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0750, 0.0723, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 20:05:00,792 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:06,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100136.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:05:08,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100139.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:05:09,180 INFO [train.py:968] (1/2) Epoch 25, batch 6300, giga_loss[loss=0.5063, simple_loss=0.5011, pruned_loss=0.2557, over 26680.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3703, pruned_loss=0.118, over 5684846.80 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.343, pruned_loss=0.08858, over 5567406.53 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.372, pruned_loss=0.1204, over 5673724.87 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:05:21,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:21,919 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:24,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:34,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9236, 3.7284, 3.5420, 1.6109], device='cuda:1'), covar=tensor([0.0730, 0.0934, 0.0878, 0.2141], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.1168, 0.0985, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 20:05:34,793 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100168.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:05:53,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:05:56,786 INFO [train.py:968] (1/2) Epoch 25, batch 6350, giga_loss[loss=0.2985, simple_loss=0.3675, pruned_loss=0.1148, over 28948.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3766, pruned_loss=0.1235, over 5670761.15 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3431, pruned_loss=0.08879, over 5565048.48 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3783, pruned_loss=0.1257, over 5665980.27 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:06:01,368 INFO [optim.py:369] (1/2) 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 20:06:48,376 INFO [train.py:968] (1/2) Epoch 25, batch 6400, giga_loss[loss=0.3384, simple_loss=0.3729, pruned_loss=0.1519, over 23669.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3778, pruned_loss=0.1255, over 5648698.91 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3429, pruned_loss=0.08868, over 5567932.96 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.38, pruned_loss=0.128, over 5643806.21 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:07:13,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:07:24,088 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:07:39,425 INFO [train.py:968] (1/2) Epoch 25, batch 6450, giga_loss[loss=0.3365, simple_loss=0.3892, pruned_loss=0.1419, over 28703.00 frames. ], tot_loss[loss=0.3204, simple_loss=0.3811, pruned_loss=0.1299, over 5623959.92 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3432, pruned_loss=0.08879, over 5562791.25 frames. ], giga_tot_loss[loss=0.3239, simple_loss=0.3831, pruned_loss=0.1324, over 5626048.97 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:07:46,848 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:07:48,187 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:07:49,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:08:05,658 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:08:21,352 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:08:35,333 INFO [train.py:968] (1/2) Epoch 25, batch 6500, giga_loss[loss=0.4207, simple_loss=0.4472, pruned_loss=0.1971, over 27519.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3864, pruned_loss=0.1354, over 5611137.96 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.08879, over 5565842.79 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3882, pruned_loss=0.1378, over 5610576.04 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:09:10,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:09:25,473 INFO [train.py:968] (1/2) Epoch 25, batch 6550, giga_loss[loss=0.331, simple_loss=0.3869, pruned_loss=0.1375, over 28702.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3879, pruned_loss=0.1362, over 5613227.60 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3432, pruned_loss=0.08887, over 5573309.88 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3904, pruned_loss=0.1391, over 5607218.19 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:09:33,469 INFO [optim.py:369] (1/2) 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 20:09:42,365 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:09:46,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:10:01,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4368, 1.8360, 1.4112, 1.5645], device='cuda:1'), covar=tensor([0.2086, 0.1908, 0.2333, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0749, 0.0723, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 20:10:15,186 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:10:15,536 INFO [train.py:968] (1/2) Epoch 25, batch 6600, giga_loss[loss=0.3264, simple_loss=0.38, pruned_loss=0.1363, over 28966.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3879, pruned_loss=0.1364, over 5627476.30 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3436, pruned_loss=0.089, over 5578515.93 frames. ], giga_tot_loss[loss=0.3344, simple_loss=0.3902, pruned_loss=0.1393, over 5618621.16 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:11:01,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-12 20:11:04,780 INFO [train.py:968] (1/2) Epoch 25, batch 6650, giga_loss[loss=0.3304, simple_loss=0.3869, pruned_loss=0.137, over 28645.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3862, pruned_loss=0.1359, over 5629265.72 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08893, over 5576321.65 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3886, pruned_loss=0.1388, over 5624689.15 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:11:16,112 INFO [optim.py:369] (1/2) 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 20:11:32,061 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:11:32,628 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:11:35,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 6700, giga_loss[loss=0.3915, simple_loss=0.4229, pruned_loss=0.18, over 26545.00 frames. ], tot_loss[loss=0.3286, simple_loss=0.3862, pruned_loss=0.1355, over 5631958.49 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3442, pruned_loss=0.08934, over 5585822.38 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3889, pruned_loss=0.1389, over 5621484.52 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:11:59,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-12 20:12:03,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:12:08,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6554, 4.4897, 4.2582, 1.9256], device='cuda:1'), covar=tensor([0.0530, 0.0662, 0.0716, 0.2291], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1182, 0.0998, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 20:12:43,299 INFO [train.py:968] (1/2) Epoch 25, batch 6750, giga_loss[loss=0.2997, simple_loss=0.3762, pruned_loss=0.1116, over 28921.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3855, pruned_loss=0.1334, over 5647584.18 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.344, pruned_loss=0.08927, over 5593939.23 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.3888, pruned_loss=0.1373, over 5632965.88 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 20:12:50,121 INFO [optim.py:369] (1/2) 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 20:13:01,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5875, 3.4507, 3.2815, 1.9679], device='cuda:1'), covar=tensor([0.0719, 0.0857, 0.0814, 0.1761], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1183, 0.0998, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 20:13:34,295 INFO [train.py:968] (1/2) Epoch 25, batch 6800, giga_loss[loss=0.3281, simple_loss=0.3977, pruned_loss=0.1292, over 28883.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.387, pruned_loss=0.1347, over 5623612.77 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3441, pruned_loss=0.08924, over 5595103.54 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3898, pruned_loss=0.1379, over 5611441.46 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:13:42,987 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:07,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:10,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:11,660 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100676.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:14:24,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1100689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:14:24,432 INFO [train.py:968] (1/2) Epoch 25, batch 6850, giga_loss[loss=0.2794, simple_loss=0.3556, pruned_loss=0.1016, over 28985.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3847, pruned_loss=0.1326, over 5622675.48 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3439, pruned_loss=0.08913, over 5598206.51 frames. ], giga_tot_loss[loss=0.3294, simple_loss=0.3874, pruned_loss=0.1357, over 5610603.71 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:14:36,934 INFO [optim.py:369] (1/2) 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 20:14:55,443 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-12 20:15:03,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100729.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:15:13,909 INFO [train.py:968] (1/2) Epoch 25, batch 6900, libri_loss[loss=0.2652, simple_loss=0.3451, pruned_loss=0.09263, over 29559.00 frames. ], tot_loss[loss=0.3195, simple_loss=0.3815, pruned_loss=0.1288, over 5626104.57 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3438, pruned_loss=0.08926, over 5604092.30 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3853, pruned_loss=0.1327, over 5611606.00 frames. ], batch size: 78, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:15:47,775 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:00,676 INFO [train.py:968] (1/2) Epoch 25, batch 6950, giga_loss[loss=0.3232, simple_loss=0.3855, pruned_loss=0.1305, over 27542.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3776, pruned_loss=0.1249, over 5641920.85 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3437, pruned_loss=0.08924, over 5608499.84 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3812, pruned_loss=0.1285, over 5626836.55 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:16:01,664 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:05,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:09,608 INFO [optim.py:369] (1/2) 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 20:16:25,342 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:27,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:32,870 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:41,453 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1100832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:44,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1100835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:16:46,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5329, 5.3796, 5.1141, 2.4859], device='cuda:1'), covar=tensor([0.0444, 0.0544, 0.0644, 0.1745], device='cuda:1'), in_proj_covar=tensor([0.1283, 0.1186, 0.1000, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 20:16:48,847 INFO [train.py:968] (1/2) Epoch 25, batch 7000, giga_loss[loss=0.3057, simple_loss=0.3567, pruned_loss=0.1273, over 23759.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3753, pruned_loss=0.123, over 5641593.53 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.344, pruned_loss=0.08946, over 5610913.55 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3786, pruned_loss=0.1265, over 5628244.58 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:16:56,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3705, 1.7402, 1.5222, 1.4256], device='cuda:1'), covar=tensor([0.0725, 0.0389, 0.0325, 0.0824], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 20:16:56,970 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:17:12,617 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1100864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:17:13,352 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:17:38,206 INFO [train.py:968] (1/2) Epoch 25, batch 7050, giga_loss[loss=0.2635, simple_loss=0.3389, pruned_loss=0.09403, over 29007.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.1219, over 5647867.56 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3438, pruned_loss=0.08934, over 5612004.86 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3764, pruned_loss=0.1251, over 5636655.14 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:17:40,357 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:17:47,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1100900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:18:25,400 INFO [train.py:968] (1/2) Epoch 25, batch 7100, giga_loss[loss=0.267, simple_loss=0.3467, pruned_loss=0.09366, over 28841.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3717, pruned_loss=0.1205, over 5660496.28 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3433, pruned_loss=0.08913, over 5616396.80 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3751, pruned_loss=0.1237, over 5648542.43 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:18:40,422 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 20:19:21,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0617, 5.8549, 5.5616, 2.9972], device='cuda:1'), covar=tensor([0.0455, 0.0618, 0.0679, 0.1604], device='cuda:1'), in_proj_covar=tensor([0.1287, 0.1192, 0.1004, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 20:19:22,173 INFO [train.py:968] (1/2) Epoch 25, batch 7150, giga_loss[loss=0.2471, simple_loss=0.3328, pruned_loss=0.08071, over 28656.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3707, pruned_loss=0.1193, over 5661508.10 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3432, pruned_loss=0.08904, over 5620146.70 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3739, pruned_loss=0.1224, over 5649272.53 frames. ], batch size: 71, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:19:31,925 INFO [optim.py:369] (1/2) 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 20:20:10,372 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:20:13,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:20:14,346 INFO [train.py:968] (1/2) Epoch 25, batch 7200, libri_loss[loss=0.2902, simple_loss=0.3725, pruned_loss=0.104, over 29477.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.37, pruned_loss=0.1172, over 5673563.87 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3433, pruned_loss=0.08909, over 5622803.80 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3726, pruned_loss=0.1199, over 5661851.57 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:20:25,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:20:28,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101051.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:20:34,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3299, 1.3206, 3.7031, 3.2288], device='cuda:1'), covar=tensor([0.1635, 0.2825, 0.0513, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0663, 0.0985, 0.0950], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 20:20:43,704 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:21:05,785 INFO [train.py:968] (1/2) Epoch 25, batch 7250, giga_loss[loss=0.2986, simple_loss=0.3786, pruned_loss=0.1093, over 28520.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.37, pruned_loss=0.1155, over 5670466.39 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3427, pruned_loss=0.08881, over 5632876.66 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3737, pruned_loss=0.1189, over 5653618.81 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:21:10,072 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1101094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:21:13,982 INFO [optim.py:369] (1/2) 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 20:21:17,917 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101104.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:21:52,032 INFO [train.py:968] (1/2) Epoch 25, batch 7300, giga_loss[loss=0.3004, simple_loss=0.3699, pruned_loss=0.1155, over 28523.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3706, pruned_loss=0.1156, over 5670826.33 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3422, pruned_loss=0.08872, over 5642417.03 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3749, pruned_loss=0.1193, over 5649918.65 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:22:05,607 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:22:40,719 INFO [train.py:968] (1/2) Epoch 25, batch 7350, giga_loss[loss=0.2945, simple_loss=0.3632, pruned_loss=0.1129, over 28686.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3703, pruned_loss=0.1158, over 5680731.25 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3423, pruned_loss=0.08874, over 5645096.46 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.374, pruned_loss=0.1191, over 5662364.25 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:22:43,688 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:22:47,335 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:22:48,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:22:50,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101197.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:22:51,277 INFO [optim.py:369] (1/2) 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 20:23:14,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:23:16,127 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 20:23:29,383 INFO [train.py:968] (1/2) Epoch 25, batch 7400, libri_loss[loss=0.3335, simple_loss=0.3963, pruned_loss=0.1354, over 29173.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3701, pruned_loss=0.1166, over 5677565.08 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08898, over 5647460.38 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3731, pruned_loss=0.1192, over 5661324.32 frames. ], batch size: 97, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:23:29,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:23:39,655 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101247.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:23:41,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101250.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:24:01,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:24:07,014 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101279.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 20:24:15,011 INFO [train.py:968] (1/2) Epoch 25, batch 7450, libri_loss[loss=0.2543, simple_loss=0.3477, pruned_loss=0.08045, over 27599.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3691, pruned_loss=0.1173, over 5672481.04 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3429, pruned_loss=0.08922, over 5651841.04 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3723, pruned_loss=0.1203, over 5656526.33 frames. ], batch size: 115, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:24:19,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:24:20,982 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:24:46,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:24:57,919 INFO [train.py:968] (1/2) Epoch 25, batch 7500, giga_loss[loss=0.2791, simple_loss=0.3574, pruned_loss=0.1004, over 28839.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3665, pruned_loss=0.1153, over 5679035.21 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3431, pruned_loss=0.08929, over 5648669.63 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3695, pruned_loss=0.1184, over 5669138.52 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:25:42,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:25:44,417 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:25:47,582 INFO [train.py:968] (1/2) Epoch 25, batch 7550, giga_loss[loss=0.3103, simple_loss=0.3744, pruned_loss=0.1231, over 28266.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3666, pruned_loss=0.1142, over 5693368.29 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.343, pruned_loss=0.08916, over 5655542.85 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3698, pruned_loss=0.1175, over 5680075.49 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:25:54,064 INFO [optim.py:369] (1/2) 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 20:26:09,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:12,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5100, 2.1828, 1.5538, 0.7498], device='cuda:1'), covar=tensor([0.6530, 0.3261, 0.4429, 0.7151], device='cuda:1'), in_proj_covar=tensor([0.1802, 0.1695, 0.1633, 0.1457], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 20:26:13,071 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:17,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:29,493 INFO [train.py:968] (1/2) Epoch 25, batch 7600, giga_loss[loss=0.2807, simple_loss=0.3653, pruned_loss=0.09801, over 28972.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3652, pruned_loss=0.112, over 5690854.05 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3434, pruned_loss=0.08949, over 5650826.74 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3682, pruned_loss=0.1152, over 5685870.28 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:26:41,820 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:26:57,984 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:27:00,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0941, 2.2122, 2.2140, 1.8145], device='cuda:1'), covar=tensor([0.1856, 0.2387, 0.1505, 0.1717], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0710, 0.0963, 0.0862], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 20:27:14,021 INFO [train.py:968] (1/2) Epoch 25, batch 7650, giga_loss[loss=0.3124, simple_loss=0.3767, pruned_loss=0.1241, over 28627.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3651, pruned_loss=0.1126, over 5690587.27 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3434, pruned_loss=0.08955, over 5655182.19 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.368, pruned_loss=0.1156, over 5683361.37 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:27:21,134 INFO [optim.py:369] (1/2) 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 20:27:46,335 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 7700, libri_loss[loss=0.2577, simple_loss=0.3363, pruned_loss=0.08952, over 29584.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3637, pruned_loss=0.1119, over 5693416.34 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.0899, over 5659873.69 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3665, pruned_loss=0.1149, over 5684726.15 frames. ], batch size: 76, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:28:46,761 INFO [train.py:968] (1/2) Epoch 25, batch 7750, giga_loss[loss=0.3381, simple_loss=0.3873, pruned_loss=0.1445, over 26664.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3616, pruned_loss=0.1111, over 5692226.07 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08964, over 5661426.60 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3647, pruned_loss=0.1144, over 5684976.86 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:28:55,774 INFO [optim.py:369] (1/2) 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 20:29:08,477 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1101612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:29:11,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1101615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:29:12,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-12 20:29:25,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4929, 1.6874, 1.5089, 1.5560], device='cuda:1'), covar=tensor([0.0762, 0.0334, 0.0313, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 20:29:34,765 INFO [train.py:968] (1/2) Epoch 25, batch 7800, libri_loss[loss=0.2633, simple_loss=0.3484, pruned_loss=0.08912, over 29540.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.362, pruned_loss=0.1124, over 5688331.48 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3434, pruned_loss=0.08959, over 5667357.94 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.365, pruned_loss=0.1155, over 5677600.29 frames. ], batch size: 83, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:29:38,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1101644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:29:48,526 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1101654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:30:24,959 INFO [train.py:968] (1/2) Epoch 25, batch 7850, giga_loss[loss=0.3155, simple_loss=0.3771, pruned_loss=0.127, over 28816.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3616, pruned_loss=0.1125, over 5700813.76 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3435, pruned_loss=0.08963, over 5669660.67 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3639, pruned_loss=0.1151, over 5690592.74 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:30:33,183 INFO [optim.py:369] (1/2) 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 20:31:10,709 INFO [train.py:968] (1/2) Epoch 25, batch 7900, giga_loss[loss=0.2591, simple_loss=0.3378, pruned_loss=0.09026, over 28761.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3599, pruned_loss=0.1118, over 5703115.39 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3441, pruned_loss=0.08984, over 5675782.50 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5690361.55 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:31:40,356 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1101774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:31:52,974 INFO [train.py:968] (1/2) Epoch 25, batch 7950, giga_loss[loss=0.3072, simple_loss=0.3687, pruned_loss=0.1229, over 28759.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3599, pruned_loss=0.112, over 5704904.88 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3441, pruned_loss=0.0898, over 5680685.36 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.362, pruned_loss=0.1151, over 5691281.89 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:32:01,561 INFO [optim.py:369] (1/2) 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 20:32:35,542 INFO [train.py:968] (1/2) Epoch 25, batch 8000, giga_loss[loss=0.2924, simple_loss=0.3624, pruned_loss=0.1112, over 28420.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3601, pruned_loss=0.1123, over 5689942.52 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3434, pruned_loss=0.08946, over 5676551.73 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3634, pruned_loss=0.1163, over 5683195.79 frames. ], batch size: 65, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:32:37,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7484, 2.1025, 1.4230, 1.7569], device='cuda:1'), covar=tensor([0.0995, 0.0556, 0.1042, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0452, 0.0523, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 20:33:20,108 INFO [train.py:968] (1/2) Epoch 25, batch 8050, giga_loss[loss=0.3184, simple_loss=0.3817, pruned_loss=0.1275, over 28860.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3608, pruned_loss=0.1118, over 5687367.60 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3435, pruned_loss=0.08942, over 5677509.78 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3641, pruned_loss=0.116, over 5681309.15 frames. ], batch size: 285, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:33:29,470 INFO [optim.py:369] (1/2) 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 20:33:32,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1101903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:33:41,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4875, 1.7573, 1.4241, 1.3054], device='cuda:1'), covar=tensor([0.2852, 0.2856, 0.3439, 0.2536], device='cuda:1'), in_proj_covar=tensor([0.1554, 0.1120, 0.1372, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 20:33:58,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4181, 2.0231, 1.5596, 0.7002], device='cuda:1'), covar=tensor([0.6709, 0.3129, 0.3631, 0.7076], device='cuda:1'), in_proj_covar=tensor([0.1805, 0.1702, 0.1632, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 20:34:03,003 INFO [train.py:968] (1/2) Epoch 25, batch 8100, giga_loss[loss=0.331, simple_loss=0.3943, pruned_loss=0.1339, over 28628.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3609, pruned_loss=0.1112, over 5666297.43 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3441, pruned_loss=0.08981, over 5665069.10 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3636, pruned_loss=0.115, over 5672990.94 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:34:43,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5453, 1.7565, 1.2709, 1.3243], device='cuda:1'), covar=tensor([0.1091, 0.0633, 0.1161, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0452, 0.0522, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 20:34:47,636 INFO [train.py:968] (1/2) Epoch 25, batch 8150, giga_loss[loss=0.2914, simple_loss=0.3643, pruned_loss=0.1093, over 28971.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3616, pruned_loss=0.1116, over 5664275.80 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3434, pruned_loss=0.08946, over 5668988.79 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3649, pruned_loss=0.1156, over 5665979.29 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:34:57,430 INFO [optim.py:369] (1/2) 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 20:35:05,432 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-12 20:35:23,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1102029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:35:34,304 INFO [train.py:968] (1/2) Epoch 25, batch 8200, giga_loss[loss=0.2945, simple_loss=0.3577, pruned_loss=0.1157, over 28666.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3634, pruned_loss=0.113, over 5679261.98 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08919, over 5674403.29 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3667, pruned_loss=0.117, over 5675908.07 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:35:36,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4422, 1.2396, 4.4846, 3.6943], device='cuda:1'), covar=tensor([0.1751, 0.2961, 0.0416, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0665, 0.0988, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 20:35:40,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1102046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:35:44,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1102049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:36:13,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1102078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:36:15,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 20:36:22,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5465, 1.7532, 1.7492, 1.5571], device='cuda:1'), covar=tensor([0.1843, 0.2119, 0.2117, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0754, 0.0727, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 20:36:27,711 INFO [train.py:968] (1/2) Epoch 25, batch 8250, giga_loss[loss=0.2804, simple_loss=0.3459, pruned_loss=0.1075, over 29121.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3666, pruned_loss=0.1165, over 5673671.40 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08938, over 5676653.76 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3693, pruned_loss=0.1197, over 5669057.41 frames. ], batch size: 113, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:36:36,981 INFO [optim.py:369] (1/2) 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 20:36:55,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6733, 1.6488, 1.8623, 1.4434], device='cuda:1'), covar=tensor([0.1740, 0.2521, 0.1429, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0712, 0.0965, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 20:37:16,021 INFO [train.py:968] (1/2) Epoch 25, batch 8300, giga_loss[loss=0.3361, simple_loss=0.3896, pruned_loss=0.1413, over 28838.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3682, pruned_loss=0.1188, over 5670384.08 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3435, pruned_loss=0.08959, over 5672634.38 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3707, pruned_loss=0.1219, over 5670855.26 frames. ], batch size: 285, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:37:25,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1102149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:37:27,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5693, 1.7637, 1.7379, 1.4140], device='cuda:1'), covar=tensor([0.3502, 0.2785, 0.2426, 0.3064], device='cuda:1'), in_proj_covar=tensor([0.2028, 0.1963, 0.1891, 0.2031], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 20:37:49,273 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1102172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:37:51,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1102175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:38:04,336 INFO [train.py:968] (1/2) Epoch 25, batch 8350, libri_loss[loss=0.2384, simple_loss=0.3273, pruned_loss=0.07477, over 29570.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3696, pruned_loss=0.1206, over 5669493.09 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08916, over 5680876.80 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3732, pruned_loss=0.1247, over 5661956.17 frames. ], batch size: 76, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:38:07,216 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 20:38:15,483 INFO [optim.py:369] (1/2) 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 20:38:17,940 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 8400, libri_loss[loss=0.2168, simple_loss=0.2967, pruned_loss=0.06847, over 29475.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3691, pruned_loss=0.1209, over 5645930.76 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08934, over 5659827.86 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3727, pruned_loss=0.1248, over 5657119.15 frames. ], batch size: 70, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:39:35,030 INFO [train.py:968] (1/2) Epoch 25, batch 8450, giga_loss[loss=0.305, simple_loss=0.3655, pruned_loss=0.1222, over 28609.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3672, pruned_loss=0.1198, over 5651414.02 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3421, pruned_loss=0.08907, over 5661217.77 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.372, pruned_loss=0.1246, over 5658914.04 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:39:36,475 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1102292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:39:39,302 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:39:52,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3831, 1.7541, 1.4947, 1.5344], device='cuda:1'), covar=tensor([0.0701, 0.0412, 0.0336, 0.0735], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0120, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0101, 0.0072, 0.0064, 0.0111], device='cuda:1')
+2023-03-12 20:40:03,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1102324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:40:17,632 INFO [train.py:968] (1/2) Epoch 25, batch 8500, libri_loss[loss=0.2741, simple_loss=0.3583, pruned_loss=0.09497, over 29767.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3676, pruned_loss=0.1186, over 5664222.37 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3424, pruned_loss=0.08922, over 5665097.36 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3714, pruned_loss=0.1227, over 5666542.87 frames. ], batch size: 87, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:40:41,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-12 20:41:03,342 INFO [train.py:968] (1/2) Epoch 25, batch 8550, libri_loss[loss=0.2201, simple_loss=0.3099, pruned_loss=0.06512, over 29595.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3651, pruned_loss=0.1164, over 5662159.33 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08911, over 5670030.76 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3689, pruned_loss=0.1203, over 5659381.48 frames. ], batch size: 75, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:41:12,168 INFO [optim.py:369] (1/2) 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 20:41:46,222 INFO [train.py:968] (1/2) Epoch 25, batch 8600, giga_loss[loss=0.2785, simple_loss=0.3537, pruned_loss=0.1016, over 29090.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3633, pruned_loss=0.1155, over 5672395.73 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3416, pruned_loss=0.08868, over 5675366.80 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3674, pruned_loss=0.1198, over 5665244.04 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:42:27,321 INFO [train.py:968] (1/2) Epoch 25, batch 8650, giga_loss[loss=0.3027, simple_loss=0.3637, pruned_loss=0.1208, over 28844.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3614, pruned_loss=0.1147, over 5676480.40 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08883, over 5679745.11 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3649, pruned_loss=0.1185, over 5667093.95 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:42:43,600 INFO [optim.py:369] (1/2) 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 20:43:19,220 INFO [train.py:968] (1/2) Epoch 25, batch 8700, giga_loss[loss=0.2798, simple_loss=0.3588, pruned_loss=0.1004, over 29017.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3633, pruned_loss=0.1169, over 5651381.82 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08918, over 5672835.77 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3658, pruned_loss=0.12, over 5649522.09 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:44:07,253 INFO [train.py:968] (1/2) Epoch 25, batch 8750, giga_loss[loss=0.2835, simple_loss=0.3766, pruned_loss=0.09523, over 28976.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1167, over 5660373.36 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3422, pruned_loss=0.08905, over 5677385.03 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5654342.55 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:44:18,752 INFO [optim.py:369] (1/2) 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 20:44:53,562 INFO [train.py:968] (1/2) Epoch 25, batch 8800, giga_loss[loss=0.2954, simple_loss=0.3704, pruned_loss=0.1102, over 28563.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3684, pruned_loss=0.1167, over 5662930.35 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3422, pruned_loss=0.08901, over 5680687.93 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3708, pruned_loss=0.1196, over 5655237.78 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:45:43,074 INFO [train.py:968] (1/2) Epoch 25, batch 8850, giga_loss[loss=0.3323, simple_loss=0.3768, pruned_loss=0.1439, over 23923.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3705, pruned_loss=0.1172, over 5665999.63 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.08893, over 5681863.49 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3727, pruned_loss=0.1197, over 5658831.56 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:45:52,508 INFO [optim.py:369] (1/2) 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 20:46:24,178 INFO [train.py:968] (1/2) Epoch 25, batch 8900, giga_loss[loss=0.2918, simple_loss=0.3543, pruned_loss=0.1147, over 28455.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3722, pruned_loss=0.1189, over 5671495.69 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3418, pruned_loss=0.08869, over 5686687.26 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3748, pruned_loss=0.1217, over 5661410.68 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:46:59,556 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1102775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:47:01,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9376, 1.3200, 1.0872, 0.1690], device='cuda:1'), covar=tensor([0.4365, 0.3274, 0.4478, 0.6923], device='cuda:1'), in_proj_covar=tensor([0.1785, 0.1690, 0.1616, 0.1449], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 20:47:14,641 INFO [train.py:968] (1/2) Epoch 25, batch 8950, giga_loss[loss=0.3198, simple_loss=0.3769, pruned_loss=0.1314, over 28774.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3739, pruned_loss=0.121, over 5659564.63 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3415, pruned_loss=0.08852, over 5688729.72 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3765, pruned_loss=0.1236, over 5649741.86 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:47:25,578 INFO [optim.py:369] (1/2) 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 20:48:03,302 INFO [train.py:968] (1/2) Epoch 25, batch 9000, giga_loss[loss=0.3257, simple_loss=0.3681, pruned_loss=0.1417, over 23570.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3738, pruned_loss=0.1222, over 5649905.91 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.08847, over 5689822.08 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.376, pruned_loss=0.1245, over 5641195.43 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:48:03,303 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 20:48:11,869 INFO [train.py:1012] (1/2) Epoch 25, validation: loss=0.2069, simple_loss=0.3152, pruned_loss=0.0493, over 944034.00 frames. 
+2023-03-12 20:48:11,869 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 20:48:21,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1460, 1.6434, 1.6674, 1.3242], device='cuda:1'), covar=tensor([0.1849, 0.1232, 0.1926, 0.1608], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0753, 0.0724, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 20:48:54,311 INFO [train.py:968] (1/2) Epoch 25, batch 9050, giga_loss[loss=0.3603, simple_loss=0.4011, pruned_loss=0.1598, over 28601.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3713, pruned_loss=0.1207, over 5655663.28 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3417, pruned_loss=0.08854, over 5697446.87 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3739, pruned_loss=0.1236, over 5640604.37 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:49:06,168 INFO [optim.py:369] (1/2) 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 20:49:35,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6261, 2.1402, 1.9030, 1.6598], device='cuda:1'), covar=tensor([0.0750, 0.0268, 0.0286, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 20:49:41,921 INFO [train.py:968] (1/2) Epoch 25, batch 9100, libri_loss[loss=0.2915, simple_loss=0.3789, pruned_loss=0.102, over 25785.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3697, pruned_loss=0.1202, over 5660533.34 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08861, over 5695761.30 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3717, pruned_loss=0.1227, over 5650088.83 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:49:45,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6292, 1.8539, 1.5584, 1.7837], device='cuda:1'), covar=tensor([0.2330, 0.2443, 0.2491, 0.2455], device='cuda:1'), in_proj_covar=tensor([0.1553, 0.1118, 0.1369, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 20:50:28,151 INFO [train.py:968] (1/2) Epoch 25, batch 9150, giga_loss[loss=0.3487, simple_loss=0.3732, pruned_loss=0.1621, over 23662.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3702, pruned_loss=0.1212, over 5659016.07 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3424, pruned_loss=0.08884, over 5700479.54 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3722, pruned_loss=0.1239, over 5645306.49 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:50:38,098 INFO [optim.py:369] (1/2) 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 20:50:47,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4198, 3.2865, 1.5063, 1.6894], device='cuda:1'), covar=tensor([0.0991, 0.0324, 0.0882, 0.1258], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0567, 0.0399, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 20:51:13,244 INFO [train.py:968] (1/2) Epoch 25, batch 9200, giga_loss[loss=0.2459, simple_loss=0.3275, pruned_loss=0.08213, over 28965.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3699, pruned_loss=0.1212, over 5640486.46 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08934, over 5692221.11 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.372, pruned_loss=0.124, over 5634779.64 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 20:51:56,073 INFO [train.py:968] (1/2) Epoch 25, batch 9250, giga_loss[loss=0.2499, simple_loss=0.3255, pruned_loss=0.08716, over 28989.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3667, pruned_loss=0.1192, over 5663926.51 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08901, over 5699330.99 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3696, pruned_loss=0.1228, over 5651493.83 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:52:03,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1961, 1.5938, 1.1780, 0.4159], device='cuda:1'), covar=tensor([0.3176, 0.1828, 0.2522, 0.5628], device='cuda:1'), in_proj_covar=tensor([0.1797, 0.1700, 0.1625, 0.1457], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 20:52:09,380 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1103108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:52:43,602 INFO [train.py:968] (1/2) Epoch 25, batch 9300, giga_loss[loss=0.2593, simple_loss=0.3365, pruned_loss=0.09103, over 28618.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.365, pruned_loss=0.1184, over 5656131.00 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3428, pruned_loss=0.08912, over 5701949.77 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3678, pruned_loss=0.122, over 5642733.06 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:52:48,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-12 20:52:51,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1103150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:53:07,543 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-12 20:53:23,778 INFO [train.py:968] (1/2) Epoch 25, batch 9350, giga_loss[loss=0.4061, simple_loss=0.4318, pruned_loss=0.1902, over 26591.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3652, pruned_loss=0.1174, over 5660557.88 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3432, pruned_loss=0.08929, over 5702385.97 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3679, pruned_loss=0.1212, over 5648089.18 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:53:34,506 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-12 20:53:38,966 INFO [optim.py:369] (1/2) 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 20:54:14,001 INFO [train.py:968] (1/2) Epoch 25, batch 9400, giga_loss[loss=0.2838, simple_loss=0.3538, pruned_loss=0.1069, over 28598.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3689, pruned_loss=0.1193, over 5660561.04 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3437, pruned_loss=0.08961, over 5700269.58 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5651916.88 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:54:17,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3438, 1.4515, 3.3603, 3.2057], device='cuda:1'), covar=tensor([0.1389, 0.2516, 0.0499, 0.1060], device='cuda:1'), in_proj_covar=tensor([0.0787, 0.0667, 0.0990, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 20:55:00,519 INFO [train.py:968] (1/2) Epoch 25, batch 9450, giga_loss[loss=0.3099, simple_loss=0.3726, pruned_loss=0.1236, over 28949.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3682, pruned_loss=0.1194, over 5657846.01 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08938, over 5705357.47 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3708, pruned_loss=0.1227, over 5645395.65 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:55:02,998 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1103293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:55:05,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1103296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:55:12,739 INFO [optim.py:369] (1/2) 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 20:55:32,986 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1103325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:55:45,126 INFO [train.py:968] (1/2) Epoch 25, batch 9500, giga_loss[loss=0.3082, simple_loss=0.3754, pruned_loss=0.1205, over 28698.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.37, pruned_loss=0.1193, over 5662218.57 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08942, over 5708080.39 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3723, pruned_loss=0.1223, over 5649387.00 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:56:28,981 INFO [train.py:968] (1/2) Epoch 25, batch 9550, giga_loss[loss=0.3579, simple_loss=0.4189, pruned_loss=0.1485, over 28729.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3713, pruned_loss=0.118, over 5673368.66 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3433, pruned_loss=0.08932, over 5712670.79 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 5658162.68 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 20:56:40,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4785, 1.7633, 1.3517, 1.7457], device='cuda:1'), covar=tensor([0.2796, 0.2791, 0.3305, 0.2354], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1115, 0.1366, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 20:56:42,389 INFO [optim.py:369] (1/2) 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 20:57:10,452 INFO [train.py:968] (1/2) Epoch 25, batch 9600, giga_loss[loss=0.2951, simple_loss=0.3718, pruned_loss=0.1092, over 28761.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.373, pruned_loss=0.1176, over 5690135.72 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3434, pruned_loss=0.08957, over 5721388.10 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3762, pruned_loss=0.1211, over 5668273.10 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:57:52,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1103483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:57:59,536 INFO [train.py:968] (1/2) Epoch 25, batch 9650, giga_loss[loss=0.3334, simple_loss=0.392, pruned_loss=0.1375, over 28987.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3755, pruned_loss=0.1197, over 5682660.57 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08912, over 5726455.71 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3797, pruned_loss=0.1238, over 5659751.41 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:58:01,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6018, 1.7703, 1.8439, 1.3627], device='cuda:1'), covar=tensor([0.1964, 0.2948, 0.1710, 0.1993], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0710, 0.0965, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 20:58:09,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1103501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 20:58:13,348 INFO [optim.py:369] (1/2) 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 20:58:40,633 INFO [train.py:968] (1/2) Epoch 25, batch 9700, giga_loss[loss=0.3087, simple_loss=0.3701, pruned_loss=0.1236, over 28783.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3778, pruned_loss=0.1224, over 5691214.72 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3424, pruned_loss=0.08905, over 5730198.05 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3822, pruned_loss=0.1266, over 5668508.62 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:59:29,175 INFO [train.py:968] (1/2) Epoch 25, batch 9750, giga_loss[loss=0.3006, simple_loss=0.3692, pruned_loss=0.116, over 28609.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3791, pruned_loss=0.1245, over 5676968.96 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08897, over 5732090.11 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3829, pruned_loss=0.1282, over 5657262.92 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 20:59:43,500 INFO [optim.py:369] (1/2) 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 21:00:04,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1103626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:00:06,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1103629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:00:16,867 INFO [train.py:968] (1/2) Epoch 25, batch 9800, giga_loss[loss=0.2815, simple_loss=0.3652, pruned_loss=0.0989, over 29078.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3778, pruned_loss=0.1237, over 5672997.85 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.0889, over 5733109.66 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3811, pruned_loss=0.1268, over 5656565.60 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:00:31,884 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1103658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:00:38,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3440, 1.6143, 1.3418, 1.5187], device='cuda:1'), covar=tensor([0.0762, 0.0343, 0.0337, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 21:00:48,790 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-12 21:01:00,177 INFO [train.py:968] (1/2) Epoch 25, batch 9850, giga_loss[loss=0.3012, simple_loss=0.3775, pruned_loss=0.1125, over 28600.00 frames. ], tot_loss[loss=0.309, simple_loss=0.376, pruned_loss=0.121, over 5677147.26 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3426, pruned_loss=0.08928, over 5733913.83 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.379, pruned_loss=0.1239, over 5662102.38 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:01:12,473 INFO [optim.py:369] (1/2) 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 21:01:41,278 INFO [train.py:968] (1/2) Epoch 25, batch 9900, giga_loss[loss=0.2981, simple_loss=0.3792, pruned_loss=0.1085, over 28881.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3756, pruned_loss=0.1193, over 5679004.54 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.08944, over 5737654.31 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3787, pruned_loss=0.1224, over 5661896.65 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:02:27,675 INFO [train.py:968] (1/2) Epoch 25, batch 9950, giga_loss[loss=0.328, simple_loss=0.3935, pruned_loss=0.1312, over 28726.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3771, pruned_loss=0.1205, over 5680791.10 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08965, over 5739709.48 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3798, pruned_loss=0.1231, over 5664506.85 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 1.0
+2023-03-12 21:02:44,406 INFO [optim.py:369] (1/2) 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 21:03:00,889 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 21:03:13,691 INFO [train.py:968] (1/2) Epoch 25, batch 10000, libri_loss[loss=0.2757, simple_loss=0.3595, pruned_loss=0.09592, over 29474.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3767, pruned_loss=0.1209, over 5677478.15 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3431, pruned_loss=0.08952, over 5744475.35 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3798, pruned_loss=0.1241, over 5657886.65 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:03:45,875 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 10050, giga_loss[loss=0.3095, simple_loss=0.3717, pruned_loss=0.1236, over 28850.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3742, pruned_loss=0.1197, over 5675807.33 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3427, pruned_loss=0.08925, over 5749886.65 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3783, pruned_loss=0.1237, over 5652434.59 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:04:11,049 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:04:47,635 INFO [train.py:968] (1/2) Epoch 25, batch 10100, giga_loss[loss=0.3499, simple_loss=0.3979, pruned_loss=0.1509, over 28675.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3737, pruned_loss=0.1208, over 5672033.18 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3428, pruned_loss=0.08927, over 5753529.14 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3779, pruned_loss=0.1248, over 5647891.70 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:05:32,290 INFO [train.py:968] (1/2) Epoch 25, batch 10150, giga_loss[loss=0.2788, simple_loss=0.3466, pruned_loss=0.1055, over 28974.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3702, pruned_loss=0.1188, over 5676657.24 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3428, pruned_loss=0.08924, over 5754894.96 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.374, pruned_loss=0.1226, over 5654935.19 frames. ], batch size: 200, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:05:50,407 INFO [optim.py:369] (1/2) 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 21:06:01,450 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:06:05,236 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 10200, giga_loss[loss=0.3372, simple_loss=0.3957, pruned_loss=0.1394, over 29018.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3698, pruned_loss=0.1199, over 5661507.44 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.343, pruned_loss=0.08929, over 5755599.13 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.373, pruned_loss=0.1232, over 5642611.33 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:06:37,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:07:02,117 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:07:11,729 INFO [train.py:968] (1/2) Epoch 25, batch 10250, giga_loss[loss=0.269, simple_loss=0.3444, pruned_loss=0.09678, over 28639.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3696, pruned_loss=0.1202, over 5665402.42 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08936, over 5749191.06 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3727, pruned_loss=0.1234, over 5654497.99 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:07:25,417 INFO [optim.py:369] (1/2) 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 21:07:55,480 INFO [train.py:968] (1/2) Epoch 25, batch 10300, giga_loss[loss=0.3112, simple_loss=0.3832, pruned_loss=0.1196, over 28712.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3671, pruned_loss=0.1177, over 5664954.36 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08949, over 5752886.18 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3699, pruned_loss=0.1208, over 5651358.50 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:08:20,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4033, 1.2735, 1.2174, 1.4567], device='cuda:1'), covar=tensor([0.0782, 0.0369, 0.0353, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 21:08:39,028 INFO [train.py:968] (1/2) Epoch 25, batch 10350, giga_loss[loss=0.2699, simple_loss=0.3449, pruned_loss=0.09749, over 28943.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3639, pruned_loss=0.1141, over 5646319.54 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3437, pruned_loss=0.08998, over 5737095.73 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3662, pruned_loss=0.1167, over 5648045.48 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 21:08:55,847 INFO [optim.py:369] (1/2) 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 21:09:28,563 INFO [train.py:968] (1/2) Epoch 25, batch 10400, giga_loss[loss=0.2901, simple_loss=0.3711, pruned_loss=0.1045, over 28927.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3617, pruned_loss=0.112, over 5658507.67 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.0895, over 5741166.42 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3646, pruned_loss=0.115, over 5654294.06 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:09:32,393 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:09:51,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5005, 1.8308, 1.5352, 1.5884], device='cuda:1'), covar=tensor([0.0773, 0.0302, 0.0323, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 21:10:03,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:10:14,329 INFO [train.py:968] (1/2) Epoch 25, batch 10450, libri_loss[loss=0.3015, simple_loss=0.3817, pruned_loss=0.1107, over 29488.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.361, pruned_loss=0.1119, over 5660252.72 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.08947, over 5735414.46 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3638, pruned_loss=0.115, over 5659762.97 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:10:29,859 INFO [optim.py:369] (1/2) 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 21:10:50,485 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1537, 1.2716, 1.1319, 0.9301], device='cuda:1'), covar=tensor([0.1097, 0.0559, 0.1137, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0449, 0.0520, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 21:10:55,194 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:11:01,798 INFO [train.py:968] (1/2) Epoch 25, batch 10500, giga_loss[loss=0.2842, simple_loss=0.3519, pruned_loss=0.1083, over 28770.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3587, pruned_loss=0.1116, over 5662252.42 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.08939, over 5736853.93 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3612, pruned_loss=0.1146, over 5658886.41 frames. ], batch size: 243, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:11:13,144 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104349.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:11:36,945 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:11:47,662 INFO [train.py:968] (1/2) Epoch 25, batch 10550, giga_loss[loss=0.3218, simple_loss=0.3918, pruned_loss=0.1259, over 28669.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.1122, over 5661287.40 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08969, over 5737995.85 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3614, pruned_loss=0.1149, over 5655610.26 frames. ], batch size: 242, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:12:04,062 INFO [optim.py:369] (1/2) 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 21:12:14,909 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:12:17,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:12:32,173 INFO [train.py:968] (1/2) Epoch 25, batch 10600, giga_loss[loss=0.308, simple_loss=0.3786, pruned_loss=0.1187, over 28281.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3615, pruned_loss=0.1128, over 5666374.80 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08948, over 5740747.12 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3637, pruned_loss=0.1157, over 5657581.32 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:12:42,748 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:12:46,059 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:13:16,526 INFO [train.py:968] (1/2) Epoch 25, batch 10650, giga_loss[loss=0.277, simple_loss=0.3468, pruned_loss=0.1036, over 29077.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3643, pruned_loss=0.1149, over 5641459.59 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3437, pruned_loss=0.08979, over 5725731.25 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3663, pruned_loss=0.1177, over 5645841.43 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:13:16,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1949, 1.7056, 1.2902, 0.4140], device='cuda:1'), covar=tensor([0.4700, 0.2616, 0.3777, 0.6203], device='cuda:1'), in_proj_covar=tensor([0.1789, 0.1686, 0.1620, 0.1449], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 21:13:19,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 21:13:35,010 INFO [optim.py:369] (1/2) 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 21:13:39,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2455, 3.0536, 1.3490, 1.5055], device='cuda:1'), covar=tensor([0.1047, 0.0282, 0.0947, 0.1391], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0564, 0.0399, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 21:14:07,249 INFO [train.py:968] (1/2) Epoch 25, batch 10700, giga_loss[loss=0.2843, simple_loss=0.3537, pruned_loss=0.1074, over 29033.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3626, pruned_loss=0.1139, over 5650803.93 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08959, over 5727524.12 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3649, pruned_loss=0.1166, over 5651379.98 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:14:34,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5933, 1.7579, 1.8483, 1.3754], device='cuda:1'), covar=tensor([0.1749, 0.2492, 0.1352, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0715, 0.0970, 0.0866], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 21:14:37,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9772, 1.2751, 1.0138, 0.3364], device='cuda:1'), covar=tensor([0.3866, 0.2742, 0.4053, 0.6079], device='cuda:1'), in_proj_covar=tensor([0.1785, 0.1683, 0.1619, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 21:14:52,416 INFO [train.py:968] (1/2) Epoch 25, batch 10750, giga_loss[loss=0.3126, simple_loss=0.3814, pruned_loss=0.122, over 28247.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3644, pruned_loss=0.1159, over 5648556.72 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08951, over 5729369.98 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3664, pruned_loss=0.1185, over 5646231.86 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:14:59,223 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:15:04,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:15:09,645 INFO [optim.py:369] (1/2) 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 21:15:24,416 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:15:35,522 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 10800, giga_loss[loss=0.2877, simple_loss=0.3637, pruned_loss=0.1059, over 28822.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3666, pruned_loss=0.1173, over 5648973.27 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3432, pruned_loss=0.08939, over 5732291.80 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3686, pruned_loss=0.1198, over 5643445.50 frames. ], batch size: 199, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:16:34,021 INFO [train.py:968] (1/2) Epoch 25, batch 10850, giga_loss[loss=0.367, simple_loss=0.4086, pruned_loss=0.1626, over 27473.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3677, pruned_loss=0.1175, over 5659883.01 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3434, pruned_loss=0.08941, over 5731904.84 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3697, pruned_loss=0.1201, over 5654022.53 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:16:50,987 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:16:55,601 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104713.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:17:05,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:17:20,041 INFO [train.py:968] (1/2) Epoch 25, batch 10900, giga_loss[loss=0.3436, simple_loss=0.4025, pruned_loss=0.1423, over 28556.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3704, pruned_loss=0.1195, over 5669519.65 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3436, pruned_loss=0.08946, over 5732447.27 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.372, pruned_loss=0.1217, over 5663814.16 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:17:33,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1104753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:17:42,954 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:17:45,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:17:50,055 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-12 21:17:56,127 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-12 21:18:07,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 21:18:08,034 INFO [train.py:968] (1/2) Epoch 25, batch 10950, giga_loss[loss=0.305, simple_loss=0.3733, pruned_loss=0.1183, over 28902.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1203, over 5674617.22 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3435, pruned_loss=0.08949, over 5734110.77 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.373, pruned_loss=0.1227, over 5667135.22 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:18:14,048 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:18:24,462 INFO [optim.py:369] (1/2) 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 21:18:29,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3361, 1.5485, 1.2295, 1.4651], device='cuda:1'), covar=tensor([0.0759, 0.0368, 0.0356, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 21:18:55,093 INFO [train.py:968] (1/2) Epoch 25, batch 11000, libri_loss[loss=0.306, simple_loss=0.3757, pruned_loss=0.1181, over 29154.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3716, pruned_loss=0.1188, over 5667341.76 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3436, pruned_loss=0.08952, over 5736842.78 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3741, pruned_loss=0.1219, over 5656300.69 frames. ], batch size: 101, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:19:05,644 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:08,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:22,698 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104867.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:19:24,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:19:37,106 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:45,953 INFO [train.py:968] (1/2) Epoch 25, batch 11050, giga_loss[loss=0.2942, simple_loss=0.3701, pruned_loss=0.1091, over 28840.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3712, pruned_loss=0.119, over 5660758.73 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3434, pruned_loss=0.08949, over 5739944.34 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3739, pruned_loss=0.1219, over 5647993.31 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:19:51,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1104896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:19:53,189 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:19:53,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1104899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:20:01,406 INFO [optim.py:369] (1/2) 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 21:20:21,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1104928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:20:33,279 INFO [train.py:968] (1/2) Epoch 25, batch 11100, giga_loss[loss=0.2715, simple_loss=0.3457, pruned_loss=0.09871, over 28948.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3715, pruned_loss=0.1199, over 5663853.96 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3437, pruned_loss=0.08953, over 5742129.29 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3743, pruned_loss=0.1232, over 5648704.93 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:21:03,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1104964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:21:19,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 21:21:28,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-12 21:21:31,781 INFO [train.py:968] (1/2) Epoch 25, batch 11150, giga_loss[loss=0.2525, simple_loss=0.3317, pruned_loss=0.08669, over 28887.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3713, pruned_loss=0.1207, over 5650174.38 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3438, pruned_loss=0.08957, over 5743706.19 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1236, over 5635891.70 frames. ], batch size: 112, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:21:49,298 INFO [optim.py:369] (1/2) 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 21:22:08,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7110, 3.3717, 1.6498, 1.7059], device='cuda:1'), covar=tensor([0.0854, 0.0314, 0.0789, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0566, 0.0400, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 21:22:19,796 INFO [train.py:968] (1/2) Epoch 25, batch 11200, giga_loss[loss=0.3003, simple_loss=0.3612, pruned_loss=0.1197, over 28995.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3694, pruned_loss=0.1202, over 5651896.06 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3438, pruned_loss=0.08961, over 5745934.36 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3716, pruned_loss=0.1229, over 5637354.20 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:22:53,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8649, 5.0436, 2.0293, 1.9285], device='cuda:1'), covar=tensor([0.0918, 0.0224, 0.0864, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0566, 0.0399, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 21:23:04,938 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:23:05,809 INFO [train.py:968] (1/2) Epoch 25, batch 11250, giga_loss[loss=0.2963, simple_loss=0.3601, pruned_loss=0.1162, over 28303.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3679, pruned_loss=0.1194, over 5655273.86 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3438, pruned_loss=0.08956, over 5746594.68 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1217, over 5642867.87 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:23:21,984 INFO [optim.py:369] (1/2) 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 21:23:30,942 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-12 21:23:33,522 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-12 21:23:35,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 21:23:49,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0441, 2.1073, 1.9087, 1.7705], device='cuda:1'), covar=tensor([0.2018, 0.2552, 0.2450, 0.2592], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0756, 0.0726, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 21:23:52,922 INFO [train.py:968] (1/2) Epoch 25, batch 11300, giga_loss[loss=0.2706, simple_loss=0.3419, pruned_loss=0.09966, over 29050.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5660943.59 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3438, pruned_loss=0.08952, over 5750085.09 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3694, pruned_loss=0.1221, over 5646528.31 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:24:39,973 INFO [train.py:968] (1/2) Epoch 25, batch 11350, giga_loss[loss=0.3758, simple_loss=0.4176, pruned_loss=0.167, over 28317.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.367, pruned_loss=0.1192, over 5665410.94 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3438, pruned_loss=0.08946, over 5754767.09 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3691, pruned_loss=0.122, over 5647601.84 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:24:57,653 INFO [optim.py:369] (1/2) 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 21:25:20,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105231.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 21:25:22,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:25:27,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:25:27,888 INFO [train.py:968] (1/2) Epoch 25, batch 11400, giga_loss[loss=0.2723, simple_loss=0.3469, pruned_loss=0.09888, over 29032.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3701, pruned_loss=0.1219, over 5662989.29 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08961, over 5756570.06 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3718, pruned_loss=0.1245, over 5645980.40 frames. ], batch size: 155, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:25:49,820 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105263.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 21:26:13,251 INFO [train.py:968] (1/2) Epoch 25, batch 11450, giga_loss[loss=0.3273, simple_loss=0.3772, pruned_loss=0.1387, over 28873.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3718, pruned_loss=0.1236, over 5651036.08 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3443, pruned_loss=0.08976, over 5750644.27 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3739, pruned_loss=0.1266, over 5638694.24 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:26:34,086 INFO [optim.py:369] (1/2) 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 21:27:03,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:27:03,516 INFO [train.py:968] (1/2) Epoch 25, batch 11500, giga_loss[loss=0.3187, simple_loss=0.3748, pruned_loss=0.1313, over 28778.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.371, pruned_loss=0.1232, over 5650890.94 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3444, pruned_loss=0.08971, over 5753297.53 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3732, pruned_loss=0.1263, over 5636945.09 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:27:25,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:27:47,824 INFO [train.py:968] (1/2) Epoch 25, batch 11550, giga_loss[loss=0.3088, simple_loss=0.3785, pruned_loss=0.1196, over 28242.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1217, over 5665415.08 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.08966, over 5756910.52 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.372, pruned_loss=0.1249, over 5649103.29 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:28:06,621 INFO [optim.py:369] (1/2) 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 21:28:28,085 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 21:28:29,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9095, 3.0360, 2.1451, 0.9771], device='cuda:1'), covar=tensor([0.8864, 0.3739, 0.3888, 0.7989], device='cuda:1'), in_proj_covar=tensor([0.1806, 0.1700, 0.1635, 0.1464], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 21:28:37,146 INFO [train.py:968] (1/2) Epoch 25, batch 11600, giga_loss[loss=0.3837, simple_loss=0.4118, pruned_loss=0.1778, over 26476.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1234, over 5657961.27 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.08959, over 5759930.58 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3743, pruned_loss=0.1266, over 5640450.98 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:28:38,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4850, 1.6246, 1.6210, 1.3894], device='cuda:1'), covar=tensor([0.2752, 0.2746, 0.1858, 0.2336], device='cuda:1'), in_proj_covar=tensor([0.2024, 0.1963, 0.1887, 0.2032], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 21:28:43,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4039, 5.2399, 4.9962, 2.3927], device='cuda:1'), covar=tensor([0.0451, 0.0601, 0.0677, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.1191, 0.1004, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 21:28:54,019 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-12 21:29:15,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:29:18,001 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:29:21,992 INFO [train.py:968] (1/2) Epoch 25, batch 11650, giga_loss[loss=0.3706, simple_loss=0.4138, pruned_loss=0.1637, over 27831.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3715, pruned_loss=0.1224, over 5673847.69 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.08956, over 5762288.18 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3739, pruned_loss=0.1254, over 5656885.08 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:29:41,303 INFO [optim.py:369] (1/2) 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 21:29:46,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:30:00,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7166, 3.5807, 3.4007, 2.0917], device='cuda:1'), covar=tensor([0.0692, 0.0833, 0.0835, 0.1659], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.1191, 0.1005, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 21:30:13,001 INFO [train.py:968] (1/2) Epoch 25, batch 11700, giga_loss[loss=0.2736, simple_loss=0.3588, pruned_loss=0.09422, over 28741.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3727, pruned_loss=0.1233, over 5658630.62 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3442, pruned_loss=0.08951, over 5761911.19 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3753, pruned_loss=0.1267, over 5643076.80 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:30:13,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6419, 1.9425, 1.4049, 1.8353], device='cuda:1'), covar=tensor([0.0769, 0.0293, 0.0350, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0064, 0.0112], device='cuda:1')
+2023-03-12 21:30:48,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5556, 2.2858, 1.7209, 0.7040], device='cuda:1'), covar=tensor([0.6296, 0.3100, 0.3810, 0.6884], device='cuda:1'), in_proj_covar=tensor([0.1808, 0.1702, 0.1640, 0.1466], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 21:30:49,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4997, 1.7298, 1.7518, 1.5667], device='cuda:1'), covar=tensor([0.2218, 0.2164, 0.2395, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0758, 0.0728, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 21:31:01,670 INFO [train.py:968] (1/2) Epoch 25, batch 11750, giga_loss[loss=0.2567, simple_loss=0.3355, pruned_loss=0.08894, over 28772.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3744, pruned_loss=0.1244, over 5658830.00 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3442, pruned_loss=0.08947, over 5763145.86 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3769, pruned_loss=0.1278, over 5643460.22 frames. ], batch size: 99, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:31:17,803 INFO [optim.py:369] (1/2) 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 21:31:25,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:31:48,414 INFO [train.py:968] (1/2) Epoch 25, batch 11800, giga_loss[loss=0.3238, simple_loss=0.3838, pruned_loss=0.1319, over 28701.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.375, pruned_loss=0.1257, over 5658670.67 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.344, pruned_loss=0.08935, over 5765478.55 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3776, pruned_loss=0.1289, over 5643263.60 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:32:35,011 INFO [train.py:968] (1/2) Epoch 25, batch 11850, giga_loss[loss=0.3197, simple_loss=0.3918, pruned_loss=0.1238, over 28893.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3744, pruned_loss=0.1237, over 5660473.05 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3439, pruned_loss=0.08928, over 5768167.89 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3772, pruned_loss=0.127, over 5644102.93 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:32:41,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6030, 2.2169, 1.6422, 0.8961], device='cuda:1'), covar=tensor([0.6239, 0.3409, 0.4137, 0.6391], device='cuda:1'), in_proj_covar=tensor([0.1805, 0.1700, 0.1638, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 21:32:54,193 INFO [optim.py:369] (1/2) 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 21:33:06,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4826, 1.5758, 1.5323, 1.3768], device='cuda:1'), covar=tensor([0.2606, 0.2518, 0.2124, 0.2482], device='cuda:1'), in_proj_covar=tensor([0.2026, 0.1964, 0.1886, 0.2032], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 21:33:13,904 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:33:22,571 INFO [train.py:968] (1/2) Epoch 25, batch 11900, giga_loss[loss=0.3087, simple_loss=0.3771, pruned_loss=0.1201, over 28881.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3736, pruned_loss=0.1223, over 5665878.60 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3435, pruned_loss=0.0891, over 5771243.85 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3768, pruned_loss=0.1258, over 5647609.40 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:33:23,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1105740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:33:39,306 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:33:41,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:33:57,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4286, 1.9388, 1.4146, 0.7368], device='cuda:1'), covar=tensor([0.5929, 0.3120, 0.3857, 0.6626], device='cuda:1'), in_proj_covar=tensor([0.1802, 0.1698, 0.1634, 0.1458], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 21:34:03,389 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-12 21:34:10,123 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:34:10,451 INFO [train.py:968] (1/2) Epoch 25, batch 11950, giga_loss[loss=0.3184, simple_loss=0.3817, pruned_loss=0.1275, over 27947.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3735, pruned_loss=0.1225, over 5653736.03 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3436, pruned_loss=0.0891, over 5770143.15 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3765, pruned_loss=0.1259, over 5638142.36 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:34:28,423 INFO [optim.py:369] (1/2) 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 21:34:53,059 INFO [train.py:968] (1/2) Epoch 25, batch 12000, libri_loss[loss=0.2775, simple_loss=0.3559, pruned_loss=0.09953, over 29526.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3719, pruned_loss=0.1214, over 5665052.99 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08957, over 5773339.04 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3745, pruned_loss=0.1246, over 5646633.05 frames. ], batch size: 80, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:34:53,059 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 21:35:01,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1421, 1.5397, 1.6030, 1.3638], device='cuda:1'), covar=tensor([0.1852, 0.1566, 0.2033, 0.1657], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0761, 0.0730, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 21:35:01,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2914, 1.8650, 1.4370, 0.4767], device='cuda:1'), covar=tensor([0.5410, 0.4221, 0.5008, 0.6818], device='cuda:1'), in_proj_covar=tensor([0.1802, 0.1699, 0.1635, 0.1458], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 21:35:02,176 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 21:35:41,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1105883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:35:43,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1105886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:35:46,221 INFO [train.py:968] (1/2) Epoch 25, batch 12050, libri_loss[loss=0.2627, simple_loss=0.3517, pruned_loss=0.08687, over 29231.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3728, pruned_loss=0.1217, over 5662056.96 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3445, pruned_loss=0.08973, over 5764764.81 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3758, pruned_loss=0.1254, over 5650325.85 frames. ], batch size: 94, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:35:48,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 21:36:06,298 INFO [optim.py:369] (1/2) 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 21:36:09,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1105915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:36:25,843 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:36:28,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6494, 3.4710, 3.3083, 1.6949], device='cuda:1'), covar=tensor([0.0833, 0.0970, 0.0886, 0.2408], device='cuda:1'), in_proj_covar=tensor([0.1291, 0.1193, 0.1006, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 21:36:33,543 INFO [train.py:968] (1/2) Epoch 25, batch 12100, giga_loss[loss=0.2995, simple_loss=0.3649, pruned_loss=0.1171, over 28601.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3725, pruned_loss=0.1215, over 5654110.89 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3444, pruned_loss=0.08964, over 5765937.90 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3752, pruned_loss=0.1247, over 5642755.79 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:36:54,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1541, 2.1975, 1.8186, 1.7697], device='cuda:1'), covar=tensor([0.0941, 0.0688, 0.0913, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0453, 0.0525, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 21:37:20,628 INFO [train.py:968] (1/2) Epoch 25, batch 12150, giga_loss[loss=0.3313, simple_loss=0.3859, pruned_loss=0.1384, over 28175.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3707, pruned_loss=0.1206, over 5672223.89 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3446, pruned_loss=0.08969, over 5770567.91 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 5656408.27 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:37:28,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1105999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:37:41,169 INFO [optim.py:369] (1/2) 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 21:37:42,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-12 21:38:07,322 INFO [train.py:968] (1/2) Epoch 25, batch 12200, giga_loss[loss=0.3708, simple_loss=0.42, pruned_loss=0.1608, over 28494.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3711, pruned_loss=0.1212, over 5676508.75 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3443, pruned_loss=0.08947, over 5773199.39 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.374, pruned_loss=0.1247, over 5659896.41 frames. ], batch size: 60, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:38:08,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7582, 1.9282, 1.5205, 1.3969], device='cuda:1'), covar=tensor([0.0990, 0.0596, 0.0938, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0453, 0.0526, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 21:38:15,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3287, 3.1614, 3.0224, 1.3447], device='cuda:1'), covar=tensor([0.0936, 0.1047, 0.0900, 0.2298], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1194, 0.1008, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 21:38:57,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2347, 1.5285, 1.5058, 1.1312], device='cuda:1'), covar=tensor([0.1587, 0.2484, 0.1336, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0711, 0.0966, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 21:38:58,401 INFO [train.py:968] (1/2) Epoch 25, batch 12250, libri_loss[loss=0.2082, simple_loss=0.2915, pruned_loss=0.06246, over 29366.00 frames. ], tot_loss[loss=0.308, simple_loss=0.372, pruned_loss=0.122, over 5675412.92 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.344, pruned_loss=0.08934, over 5773699.40 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3747, pruned_loss=0.125, over 5661424.45 frames. ], batch size: 67, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:39:15,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:39:18,362 INFO [optim.py:369] (1/2) 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 21:39:21,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4113, 1.3829, 4.4048, 3.6020], device='cuda:1'), covar=tensor([0.1706, 0.2890, 0.0433, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0663, 0.0980, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 21:39:45,758 INFO [train.py:968] (1/2) Epoch 25, batch 12300, giga_loss[loss=0.3059, simple_loss=0.3615, pruned_loss=0.1252, over 28588.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3725, pruned_loss=0.1226, over 5668322.70 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3444, pruned_loss=0.08962, over 5774122.80 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3747, pruned_loss=0.1252, over 5655433.46 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:40:06,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1106163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:40:32,432 INFO [train.py:968] (1/2) Epoch 25, batch 12350, giga_loss[loss=0.2906, simple_loss=0.3646, pruned_loss=0.1083, over 28643.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3727, pruned_loss=0.1218, over 5686330.85 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3445, pruned_loss=0.08963, over 5776612.98 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3749, pruned_loss=0.1247, over 5671588.37 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:40:52,957 INFO [optim.py:369] (1/2) 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 21:41:03,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1106220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:17,684 INFO [train.py:968] (1/2) Epoch 25, batch 12400, giga_loss[loss=0.3305, simple_loss=0.3977, pruned_loss=0.1316, over 28723.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3722, pruned_loss=0.1209, over 5662000.39 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3453, pruned_loss=0.09016, over 5764748.34 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.374, pruned_loss=0.1235, over 5658896.88 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:41:26,284 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:28,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:52,454 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:41:59,064 INFO [train.py:968] (1/2) Epoch 25, batch 12450, giga_loss[loss=0.2956, simple_loss=0.3646, pruned_loss=0.1133, over 28933.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3712, pruned_loss=0.1197, over 5673070.53 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3452, pruned_loss=0.09008, over 5766205.79 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3735, pruned_loss=0.1228, over 5666662.96 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:42:14,173 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:42:45,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6574, 3.5120, 3.3342, 2.1771], device='cuda:1'), covar=tensor([0.0668, 0.0809, 0.0800, 0.1616], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1199, 0.1011, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 21:42:48,388 INFO [train.py:968] (1/2) Epoch 25, batch 12500, giga_loss[loss=0.3245, simple_loss=0.3649, pruned_loss=0.142, over 23690.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3707, pruned_loss=0.1196, over 5660890.99 frames. ], libri_tot_loss[loss=0.2632, simple_loss=0.3456, pruned_loss=0.09046, over 5760860.62 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3729, pruned_loss=0.1225, over 5657648.20 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:43:17,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:43:29,469 INFO [train.py:968] (1/2) Epoch 25, batch 12550, giga_loss[loss=0.2796, simple_loss=0.3499, pruned_loss=0.1046, over 28914.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3688, pruned_loss=0.1185, over 5660114.95 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3456, pruned_loss=0.09053, over 5745827.05 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3712, pruned_loss=0.1215, over 5669174.49 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:43:48,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4551, 1.6842, 1.7296, 1.2587], device='cuda:1'), covar=tensor([0.1719, 0.2543, 0.1428, 0.1735], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0712, 0.0967, 0.0864], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 21:43:52,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3721, 1.8367, 1.4381, 1.5821], device='cuda:1'), covar=tensor([0.0746, 0.0288, 0.0315, 0.0821], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0121, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-12 21:43:52,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5471, 1.9465, 1.9390, 1.4077], device='cuda:1'), covar=tensor([0.3588, 0.2452, 0.2566, 0.3055], device='cuda:1'), in_proj_covar=tensor([0.2025, 0.1963, 0.1882, 0.2028], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 21:43:53,981 INFO [optim.py:369] (1/2) 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 21:44:00,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3651, 1.5877, 1.1933, 1.1602], device='cuda:1'), covar=tensor([0.1030, 0.0511, 0.1032, 0.1116], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0452, 0.0526, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 21:44:18,939 INFO [train.py:968] (1/2) Epoch 25, batch 12600, giga_loss[loss=0.2736, simple_loss=0.3437, pruned_loss=0.1018, over 28868.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3654, pruned_loss=0.117, over 5658451.92 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3452, pruned_loss=0.09026, over 5746373.25 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.368, pruned_loss=0.1202, over 5663361.41 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:44:31,634 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:44:33,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:00,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:06,128 INFO [train.py:968] (1/2) Epoch 25, batch 12650, giga_loss[loss=0.2735, simple_loss=0.3382, pruned_loss=0.1044, over 28617.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3612, pruned_loss=0.1149, over 5675142.40 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3448, pruned_loss=0.09002, over 5749837.66 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3641, pruned_loss=0.1181, over 5674409.83 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:45:28,137 INFO [optim.py:369] (1/2) 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 21:45:33,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:35,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:52,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:45:54,315 INFO [train.py:968] (1/2) Epoch 25, batch 12700, giga_loss[loss=0.3129, simple_loss=0.3809, pruned_loss=0.1224, over 28645.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3598, pruned_loss=0.1146, over 5684834.45 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3447, pruned_loss=0.09018, over 5753108.35 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3624, pruned_loss=0.1175, over 5680070.94 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:46:02,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:46:15,402 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-12 21:46:40,315 INFO [train.py:968] (1/2) Epoch 25, batch 12750, giga_loss[loss=0.3095, simple_loss=0.3787, pruned_loss=0.1201, over 28750.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3597, pruned_loss=0.1145, over 5685641.20 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3445, pruned_loss=0.09011, over 5756616.47 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3623, pruned_loss=0.1175, over 5677185.11 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:46:45,344 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 21:46:46,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1106595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:47:00,568 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 12800, giga_loss[loss=0.2702, simple_loss=0.3536, pruned_loss=0.09345, over 28201.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1124, over 5683102.41 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3445, pruned_loss=0.0901, over 5757869.32 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3616, pruned_loss=0.1151, over 5674131.30 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:48:09,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:48:11,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:48:16,767 INFO [train.py:968] (1/2) Epoch 25, batch 12850, giga_loss[loss=0.3251, simple_loss=0.389, pruned_loss=0.1306, over 28039.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3577, pruned_loss=0.1093, over 5668373.95 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3443, pruned_loss=0.09016, over 5752973.13 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3601, pruned_loss=0.112, over 5664117.06 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:48:41,383 INFO [optim.py:369] (1/2) 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 21:48:42,225 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:03,376 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-12 21:49:05,962 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1106738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:07,833 INFO [train.py:968] (1/2) Epoch 25, batch 12900, giga_loss[loss=0.2597, simple_loss=0.341, pruned_loss=0.08924, over 28926.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3556, pruned_loss=0.1069, over 5659961.98 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3441, pruned_loss=0.09015, over 5746322.07 frames. ], giga_tot_loss[loss=0.2884, simple_loss=0.358, pruned_loss=0.1094, over 5661552.95 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:49:08,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1106741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:36,264 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1106770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:49:57,307 INFO [train.py:968] (1/2) Epoch 25, batch 12950, giga_loss[loss=0.296, simple_loss=0.3621, pruned_loss=0.1149, over 28953.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3525, pruned_loss=0.1037, over 5660141.51 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3438, pruned_loss=0.08997, over 5749499.47 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3549, pruned_loss=0.1061, over 5657279.08 frames. ], batch size: 106, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:50:07,004 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5699, 1.7425, 1.4314, 1.7079], device='cuda:1'), covar=tensor([0.2878, 0.2859, 0.3347, 0.2495], device='cuda:1'), in_proj_covar=tensor([0.1564, 0.1126, 0.1381, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 21:50:08,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6681, 3.1367, 2.8419, 2.3753], device='cuda:1'), covar=tensor([0.2557, 0.1648, 0.1693, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.2011, 0.1949, 0.1867, 0.2016], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 21:50:21,567 INFO [optim.py:369] (1/2) 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 21:50:34,048 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1106826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:50:45,880 INFO [train.py:968] (1/2) Epoch 25, batch 13000, giga_loss[loss=0.3176, simple_loss=0.3857, pruned_loss=0.1247, over 28943.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3497, pruned_loss=0.1008, over 5669772.52 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3429, pruned_loss=0.08967, over 5752272.97 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3525, pruned_loss=0.1033, over 5663028.54 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:51:33,251 INFO [train.py:968] (1/2) Epoch 25, batch 13050, giga_loss[loss=0.2626, simple_loss=0.355, pruned_loss=0.08512, over 28735.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3485, pruned_loss=0.09837, over 5674318.82 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08967, over 5757387.65 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3516, pruned_loss=0.1008, over 5660685.81 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:51:35,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-12 21:51:50,011 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 21:51:56,537 INFO [optim.py:369] (1/2) 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 21:52:25,642 INFO [train.py:968] (1/2) Epoch 25, batch 13100, giga_loss[loss=0.2695, simple_loss=0.3329, pruned_loss=0.103, over 23945.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3496, pruned_loss=0.0994, over 5660611.28 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3422, pruned_loss=0.08961, over 5756686.59 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3523, pruned_loss=0.1015, over 5648766.15 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:53:14,266 INFO [train.py:968] (1/2) Epoch 25, batch 13150, giga_loss[loss=0.2492, simple_loss=0.3327, pruned_loss=0.08288, over 28831.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3474, pruned_loss=0.09789, over 5664453.62 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3416, pruned_loss=0.08938, over 5758780.86 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3502, pruned_loss=0.09992, over 5652032.30 frames. ], batch size: 119, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:53:27,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7072, 3.5600, 3.3597, 2.0575], device='cuda:1'), covar=tensor([0.0706, 0.0853, 0.0862, 0.2273], device='cuda:1'), in_proj_covar=tensor([0.1277, 0.1180, 0.0995, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 21:53:36,154 INFO [optim.py:369] (1/2) 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 21:53:39,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-12 21:54:00,355 INFO [train.py:968] (1/2) Epoch 25, batch 13200, giga_loss[loss=0.256, simple_loss=0.3342, pruned_loss=0.08889, over 28942.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3448, pruned_loss=0.09579, over 5669990.14 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3419, pruned_loss=0.08966, over 5757050.20 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3469, pruned_loss=0.09744, over 5658239.99 frames. ], batch size: 227, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:54:09,214 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-12 21:54:21,273 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:54:46,101 INFO [train.py:968] (1/2) Epoch 25, batch 13250, libri_loss[loss=0.3077, simple_loss=0.3718, pruned_loss=0.1218, over 29546.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3443, pruned_loss=0.09571, over 5669805.70 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3418, pruned_loss=0.08977, over 5757115.17 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3461, pruned_loss=0.09704, over 5658359.48 frames. ], batch size: 84, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:55:08,520 INFO [optim.py:369] (1/2) 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 21:55:08,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:55:33,022 INFO [train.py:968] (1/2) Epoch 25, batch 13300, giga_loss[loss=0.2189, simple_loss=0.3097, pruned_loss=0.06406, over 28861.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3435, pruned_loss=0.09481, over 5672479.52 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3412, pruned_loss=0.08961, over 5760504.14 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3455, pruned_loss=0.09614, over 5658404.31 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 21:56:24,688 INFO [train.py:968] (1/2) Epoch 25, batch 13350, giga_loss[loss=0.248, simple_loss=0.3385, pruned_loss=0.07872, over 28945.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3423, pruned_loss=0.09369, over 5673111.71 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3413, pruned_loss=0.08978, over 5761352.85 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3438, pruned_loss=0.09462, over 5660515.90 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:56:35,628 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:57:05,789 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-12 21:57:13,414 INFO [train.py:968] (1/2) Epoch 25, batch 13400, giga_loss[loss=0.2674, simple_loss=0.3413, pruned_loss=0.09672, over 28019.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3404, pruned_loss=0.09214, over 5669954.31 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3415, pruned_loss=0.08997, over 5758839.40 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3414, pruned_loss=0.09279, over 5660559.02 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:58:08,565 INFO [train.py:968] (1/2) Epoch 25, batch 13450, giga_loss[loss=0.2525, simple_loss=0.3298, pruned_loss=0.08761, over 28518.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3367, pruned_loss=0.09049, over 5653847.94 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3413, pruned_loss=0.08997, over 5751328.89 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3377, pruned_loss=0.091, over 5651862.35 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:58:32,709 INFO [optim.py:369] (1/2) 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 21:58:59,515 INFO [train.py:968] (1/2) Epoch 25, batch 13500, giga_loss[loss=0.2285, simple_loss=0.3109, pruned_loss=0.07305, over 29051.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3344, pruned_loss=0.0899, over 5637078.12 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3403, pruned_loss=0.08957, over 5747839.63 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3359, pruned_loss=0.09068, over 5635490.18 frames. ], batch size: 136, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 21:59:03,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1107344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:59:05,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1107347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:59:10,307 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6812, 1.9558, 1.3494, 1.6742], device='cuda:1'), covar=tensor([0.0970, 0.0579, 0.0968, 0.1147], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0444, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 21:59:33,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1107376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 21:59:46,461 INFO [train.py:968] (1/2) Epoch 25, batch 13550, libri_loss[loss=0.2838, simple_loss=0.3567, pruned_loss=0.1054, over 29533.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3351, pruned_loss=0.09105, over 5652964.03 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3405, pruned_loss=0.0899, over 5753253.32 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3359, pruned_loss=0.09142, over 5643205.48 frames. ], batch size: 89, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:00:10,838 INFO [optim.py:369] (1/2) 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 22:00:42,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1107438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:00:43,907 INFO [train.py:968] (1/2) Epoch 25, batch 13600, giga_loss[loss=0.2741, simple_loss=0.3591, pruned_loss=0.09453, over 28639.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3377, pruned_loss=0.09254, over 5644292.75 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3401, pruned_loss=0.08974, over 5756434.32 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3386, pruned_loss=0.09299, over 5632244.45 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:01:37,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1107488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:01:38,161 INFO [train.py:968] (1/2) Epoch 25, batch 13650, giga_loss[loss=0.2789, simple_loss=0.3544, pruned_loss=0.1017, over 28947.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3403, pruned_loss=0.09276, over 5650963.08 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3402, pruned_loss=0.08991, over 5758775.38 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.341, pruned_loss=0.09308, over 5636040.32 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:01:41,635 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-12 22:02:07,290 INFO [optim.py:369] (1/2) 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 22:02:17,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3103, 3.1297, 3.0111, 1.3948], device='cuda:1'), covar=tensor([0.0996, 0.1148, 0.1081, 0.2327], device='cuda:1'), in_proj_covar=tensor([0.1263, 0.1170, 0.0983, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 22:02:37,028 INFO [train.py:968] (1/2) Epoch 25, batch 13700, giga_loss[loss=0.2605, simple_loss=0.3433, pruned_loss=0.08888, over 28293.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3405, pruned_loss=0.09274, over 5638145.56 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3399, pruned_loss=0.08995, over 5750151.96 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3412, pruned_loss=0.09299, over 5632506.86 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:02:54,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2810, 1.6714, 1.0112, 1.2461], device='cuda:1'), covar=tensor([0.1326, 0.0751, 0.1558, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0443, 0.0517, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 22:03:31,668 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1107581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:03:35,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1107584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:03:42,654 INFO [train.py:968] (1/2) Epoch 25, batch 13750, giga_loss[loss=0.2385, simple_loss=0.3262, pruned_loss=0.07541, over 28637.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09163, over 5645571.34 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3399, pruned_loss=0.08996, over 5751156.20 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3394, pruned_loss=0.09183, over 5639846.92 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:04:08,284 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1107613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:04:09,455 INFO [optim.py:369] (1/2) 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 22:04:28,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1107631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:04:32,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1107634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:04:37,857 INFO [train.py:968] (1/2) Epoch 25, batch 13800, giga_loss[loss=0.2746, simple_loss=0.351, pruned_loss=0.09908, over 27610.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3375, pruned_loss=0.09036, over 5643616.86 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3396, pruned_loss=0.08996, over 5749463.40 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3381, pruned_loss=0.09056, over 5636741.01 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:05:06,813 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:05:35,611 INFO [train.py:968] (1/2) Epoch 25, batch 13850, giga_loss[loss=0.2425, simple_loss=0.3252, pruned_loss=0.07988, over 28719.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3353, pruned_loss=0.088, over 5652237.83 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3392, pruned_loss=0.08981, over 5753219.34 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3362, pruned_loss=0.08828, over 5640900.87 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:06:06,178 INFO [optim.py:369] (1/2) 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 22:06:39,716 INFO [train.py:968] (1/2) Epoch 25, batch 13900, giga_loss[loss=0.2337, simple_loss=0.3091, pruned_loss=0.07918, over 28965.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3328, pruned_loss=0.08783, over 5651882.11 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3392, pruned_loss=0.08988, over 5753572.14 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3335, pruned_loss=0.08798, over 5641802.61 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:07:06,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:07:38,555 INFO [train.py:968] (1/2) Epoch 25, batch 13950, libri_loss[loss=0.2462, simple_loss=0.3192, pruned_loss=0.08659, over 29593.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.333, pruned_loss=0.08838, over 5659040.40 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3386, pruned_loss=0.08967, over 5756871.65 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3339, pruned_loss=0.08866, over 5646030.60 frames. ], batch size: 74, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:07:41,337 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:08:09,403 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:08:10,651 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:08:31,129 INFO [train.py:968] (1/2) Epoch 25, batch 14000, giga_loss[loss=0.2407, simple_loss=0.3226, pruned_loss=0.07944, over 28527.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3327, pruned_loss=0.08785, over 5670840.10 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3382, pruned_loss=0.08941, over 5760209.01 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3336, pruned_loss=0.08828, over 5654410.44 frames. ], batch size: 336, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:08:40,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1107848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:09:16,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3399, 1.3824, 3.7418, 3.3120], device='cuda:1'), covar=tensor([0.1578, 0.2903, 0.0430, 0.1947], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0659, 0.0971, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 22:09:17,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5511, 2.3752, 1.7428, 0.7777], device='cuda:1'), covar=tensor([0.6004, 0.2991, 0.3820, 0.5719], device='cuda:1'), in_proj_covar=tensor([0.1779, 0.1666, 0.1612, 0.1445], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 22:09:32,940 INFO [train.py:968] (1/2) Epoch 25, batch 14050, giga_loss[loss=0.3017, simple_loss=0.3767, pruned_loss=0.1134, over 27627.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3353, pruned_loss=0.08842, over 5675249.37 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.338, pruned_loss=0.08926, over 5761346.28 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3362, pruned_loss=0.08887, over 5660666.37 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:10:01,604 INFO [optim.py:369] (1/2) 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 22:10:28,831 INFO [train.py:968] (1/2) Epoch 25, batch 14100, giga_loss[loss=0.2381, simple_loss=0.3192, pruned_loss=0.07847, over 27541.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3355, pruned_loss=0.08781, over 5673730.30 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3378, pruned_loss=0.08917, over 5755133.32 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3363, pruned_loss=0.08819, over 5664896.40 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:11:22,158 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-12 22:11:30,698 INFO [train.py:968] (1/2) Epoch 25, batch 14150, giga_loss[loss=0.2002, simple_loss=0.2838, pruned_loss=0.05824, over 28791.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3335, pruned_loss=0.08732, over 5683003.31 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.337, pruned_loss=0.0888, over 5757885.62 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3348, pruned_loss=0.0879, over 5670846.69 frames. ], batch size: 92, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:12:06,445 INFO [optim.py:369] (1/2) 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 22:12:20,253 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-12 22:12:36,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 22:12:36,591 INFO [train.py:968] (1/2) Epoch 25, batch 14200, giga_loss[loss=0.2692, simple_loss=0.3591, pruned_loss=0.08963, over 29005.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3356, pruned_loss=0.08872, over 5674064.63 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3368, pruned_loss=0.08869, over 5760024.05 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3367, pruned_loss=0.08928, over 5661388.65 frames. ], batch size: 120, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:12:44,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:13:42,405 INFO [train.py:968] (1/2) Epoch 25, batch 14250, libri_loss[loss=0.244, simple_loss=0.3285, pruned_loss=0.07976, over 29480.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3388, pruned_loss=0.08847, over 5666163.04 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3365, pruned_loss=0.08845, over 5763491.51 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.34, pruned_loss=0.08914, over 5650921.67 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:14:13,606 INFO [optim.py:369] (1/2) 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 22:14:38,220 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:14:41,500 INFO [train.py:968] (1/2) Epoch 25, batch 14300, giga_loss[loss=0.2015, simple_loss=0.2765, pruned_loss=0.06325, over 24349.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3386, pruned_loss=0.08629, over 5650599.13 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.336, pruned_loss=0.08823, over 5757369.78 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.34, pruned_loss=0.08699, over 5642025.74 frames. ], batch size: 705, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:15:08,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2849, 5.1054, 4.8872, 2.2660], device='cuda:1'), covar=tensor([0.0401, 0.0559, 0.0597, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.1258, 0.1162, 0.0980, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 22:15:15,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:37,016 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 14350, giga_loss[loss=0.2153, simple_loss=0.3116, pruned_loss=0.05953, over 28153.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3377, pruned_loss=0.08442, over 5657154.09 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3357, pruned_loss=0.08808, over 5759989.41 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3391, pruned_loss=0.08507, over 5645986.11 frames. ], batch size: 412, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:15:39,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:41,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6748, 2.3420, 1.6606, 0.9596], device='cuda:1'), covar=tensor([0.6581, 0.3234, 0.4778, 0.6632], device='cuda:1'), in_proj_covar=tensor([0.1791, 0.1677, 0.1620, 0.1452], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 22:15:43,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:15:44,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:16:08,235 INFO [optim.py:369] (1/2) 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 22:16:11,134 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:16:14,937 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:16:34,874 INFO [train.py:968] (1/2) Epoch 25, batch 14400, giga_loss[loss=0.2872, simple_loss=0.3637, pruned_loss=0.1053, over 28385.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3395, pruned_loss=0.08609, over 5653977.89 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3355, pruned_loss=0.08814, over 5745953.49 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3409, pruned_loss=0.08643, over 5654162.10 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:16:40,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1108242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:17:31,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:17:33,420 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 14450, giga_loss[loss=0.2241, simple_loss=0.3034, pruned_loss=0.07243, over 28442.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3392, pruned_loss=0.08741, over 5656785.29 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3355, pruned_loss=0.08824, over 5746587.72 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08757, over 5655848.51 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:18:07,293 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:08,688 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:11,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:13,794 INFO [optim.py:369] (1/2) 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 22:18:45,037 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:45,714 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:47,068 INFO [train.py:968] (1/2) Epoch 25, batch 14500, giga_loss[loss=0.2306, simple_loss=0.3232, pruned_loss=0.06902, over 28906.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.339, pruned_loss=0.08838, over 5662811.25 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.335, pruned_loss=0.08814, over 5750146.45 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3405, pruned_loss=0.0886, over 5656628.28 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:18:47,918 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:49,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:18:51,186 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:29,911 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:32,144 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:32,176 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:19:33,014 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:20:02,950 INFO [train.py:968] (1/2) Epoch 25, batch 14550, giga_loss[loss=0.2413, simple_loss=0.32, pruned_loss=0.08126, over 29004.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3366, pruned_loss=0.08688, over 5668801.86 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3348, pruned_loss=0.08807, over 5744220.33 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.338, pruned_loss=0.08712, over 5667448.76 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:20:20,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:20:46,981 INFO [optim.py:369] (1/2) 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 22:21:18,874 INFO [train.py:968] (1/2) Epoch 25, batch 14600, giga_loss[loss=0.2301, simple_loss=0.3229, pruned_loss=0.06862, over 28659.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3322, pruned_loss=0.08445, over 5661615.84 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3349, pruned_loss=0.08812, over 5746414.33 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3332, pruned_loss=0.08454, over 5657237.39 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:21:48,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4006, 1.2537, 4.0125, 3.3313], device='cuda:1'), covar=tensor([0.1615, 0.2800, 0.0470, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0771, 0.0655, 0.0967, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 22:22:13,690 INFO [train.py:968] (1/2) Epoch 25, batch 14650, giga_loss[loss=0.2193, simple_loss=0.3012, pruned_loss=0.06873, over 28909.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3299, pruned_loss=0.08336, over 5670816.15 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3341, pruned_loss=0.08765, over 5750175.93 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3313, pruned_loss=0.08363, over 5659226.89 frames. ], batch size: 145, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:22:47,896 INFO [optim.py:369] (1/2) 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 22:23:04,166 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1108528.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 22:23:11,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1792, 1.4630, 1.3857, 1.1245], device='cuda:1'), covar=tensor([0.2354, 0.2017, 0.1567, 0.2034], device='cuda:1'), in_proj_covar=tensor([0.1987, 0.1919, 0.1829, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 22:23:16,133 INFO [train.py:968] (1/2) Epoch 25, batch 14700, giga_loss[loss=0.2787, simple_loss=0.3641, pruned_loss=0.09669, over 28699.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3313, pruned_loss=0.08443, over 5679743.23 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3341, pruned_loss=0.08771, over 5749994.05 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3323, pruned_loss=0.08457, over 5670503.15 frames. ], batch size: 307, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:23:16,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1912, 1.4891, 1.4873, 1.3446], device='cuda:1'), covar=tensor([0.1941, 0.1785, 0.2131, 0.1839], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0740, 0.0712, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 22:24:02,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1710, 2.5414, 1.2243, 1.3736], device='cuda:1'), covar=tensor([0.1041, 0.0432, 0.0959, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0561, 0.0398, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 22:24:13,567 INFO [train.py:968] (1/2) Epoch 25, batch 14750, giga_loss[loss=0.2943, simple_loss=0.3583, pruned_loss=0.1152, over 26951.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3348, pruned_loss=0.08647, over 5673736.61 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3334, pruned_loss=0.08734, over 5744814.60 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3363, pruned_loss=0.08687, over 5668348.47 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:24:28,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4671, 1.8988, 1.7170, 1.6325], device='cuda:1'), covar=tensor([0.2046, 0.2025, 0.2184, 0.2153], device='cuda:1'), in_proj_covar=tensor([0.0480, 0.0739, 0.0711, 0.0681], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-12 22:24:44,357 INFO [optim.py:369] (1/2) 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 22:24:44,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:25:06,652 INFO [train.py:968] (1/2) Epoch 25, batch 14800, giga_loss[loss=0.221, simple_loss=0.3068, pruned_loss=0.06762, over 28826.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3331, pruned_loss=0.08631, over 5682036.97 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3329, pruned_loss=0.08706, over 5746587.96 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3348, pruned_loss=0.08686, over 5672206.96 frames. ], batch size: 164, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:25:33,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6106, 1.9054, 1.2908, 1.5287], device='cuda:1'), covar=tensor([0.1028, 0.0664, 0.1021, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0444, 0.0519, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 22:26:12,047 INFO [train.py:968] (1/2) Epoch 25, batch 14850, giga_loss[loss=0.2409, simple_loss=0.3214, pruned_loss=0.08021, over 29050.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.334, pruned_loss=0.0881, over 5673192.52 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3331, pruned_loss=0.08731, over 5749102.54 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3352, pruned_loss=0.08829, over 5662160.57 frames. ], batch size: 128, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:26:42,478 INFO [optim.py:369] (1/2) 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 22:27:09,818 INFO [train.py:968] (1/2) Epoch 25, batch 14900, giga_loss[loss=0.289, simple_loss=0.3637, pruned_loss=0.1072, over 28447.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3336, pruned_loss=0.08777, over 5677545.18 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3326, pruned_loss=0.08708, over 5754173.98 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.335, pruned_loss=0.08816, over 5662189.45 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:27:37,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1108760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:27:41,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1108763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:28:13,585 INFO [train.py:968] (1/2) Epoch 25, batch 14950, giga_loss[loss=0.27, simple_loss=0.3514, pruned_loss=0.09432, over 28389.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3351, pruned_loss=0.0877, over 5668380.40 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3321, pruned_loss=0.08684, over 5746098.47 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3367, pruned_loss=0.08824, over 5662225.55 frames. ], batch size: 368, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:28:17,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1108792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 22:28:34,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2644, 1.2863, 3.4024, 3.0886], device='cuda:1'), covar=tensor([0.1524, 0.2850, 0.0451, 0.1955], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0661, 0.0972, 0.0934], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 22:28:54,763 INFO [optim.py:369] (1/2) 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 22:29:27,259 INFO [train.py:968] (1/2) Epoch 25, batch 15000, giga_loss[loss=0.2308, simple_loss=0.3184, pruned_loss=0.07158, over 28957.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3371, pruned_loss=0.08826, over 5666370.43 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.332, pruned_loss=0.08682, over 5740661.04 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3386, pruned_loss=0.08875, over 5665459.63 frames. ], batch size: 186, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:29:27,259 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 22:29:36,905 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 22:30:47,131 INFO [train.py:968] (1/2) Epoch 25, batch 15050, giga_loss[loss=0.2753, simple_loss=0.3426, pruned_loss=0.104, over 28748.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3337, pruned_loss=0.08656, over 5676432.31 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3314, pruned_loss=0.08654, over 5744822.25 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3356, pruned_loss=0.08723, over 5670119.79 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:30:52,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 22:31:06,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1108903.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 22:31:21,884 INFO [optim.py:369] (1/2) 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 22:31:49,358 INFO [train.py:968] (1/2) Epoch 25, batch 15100, giga_loss[loss=0.2246, simple_loss=0.309, pruned_loss=0.07011, over 27734.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3291, pruned_loss=0.08486, over 5689575.65 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3311, pruned_loss=0.08636, over 5748550.55 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3308, pruned_loss=0.08552, over 5679567.80 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 22:32:56,048 INFO [train.py:968] (1/2) Epoch 25, batch 15150, giga_loss[loss=0.2452, simple_loss=0.3174, pruned_loss=0.08652, over 28880.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3259, pruned_loss=0.08371, over 5684680.67 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3313, pruned_loss=0.08651, over 5749332.80 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.327, pruned_loss=0.08407, over 5675787.89 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 2.0
+2023-03-12 22:33:29,656 INFO [optim.py:369] (1/2) 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 22:33:55,050 INFO [train.py:968] (1/2) Epoch 25, batch 15200, giga_loss[loss=0.2691, simple_loss=0.3496, pruned_loss=0.0943, over 28825.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3276, pruned_loss=0.08544, over 5681789.47 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.331, pruned_loss=0.08638, over 5752191.60 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3287, pruned_loss=0.08582, over 5670996.18 frames. ], batch size: 243, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:34:02,202 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1109046.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 22:34:04,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1109049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 22:34:23,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2688, 4.1020, 3.8904, 1.7568], device='cuda:1'), covar=tensor([0.0609, 0.0762, 0.0891, 0.2254], device='cuda:1'), in_proj_covar=tensor([0.1254, 0.1158, 0.0976, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 22:34:34,579 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1109078.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 22:34:48,357 INFO [train.py:968] (1/2) Epoch 25, batch 15250, giga_loss[loss=0.228, simple_loss=0.3135, pruned_loss=0.07121, over 28770.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3268, pruned_loss=0.08489, over 5677461.38 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3308, pruned_loss=0.08639, over 5755322.95 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3278, pruned_loss=0.08517, over 5664807.62 frames. ], batch size: 263, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:35:24,949 INFO [optim.py:369] (1/2) 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 22:35:49,185 INFO [train.py:968] (1/2) Epoch 25, batch 15300, giga_loss[loss=0.2424, simple_loss=0.321, pruned_loss=0.0819, over 27501.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3247, pruned_loss=0.08298, over 5672141.83 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3307, pruned_loss=0.08644, over 5755067.93 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3256, pruned_loss=0.0831, over 5660736.25 frames. ], batch size: 472, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:36:50,817 INFO [train.py:968] (1/2) Epoch 25, batch 15350, giga_loss[loss=0.2466, simple_loss=0.3256, pruned_loss=0.08382, over 28713.00 frames. ], tot_loss[loss=0.2434, simple_loss=0.3229, pruned_loss=0.08194, over 5663069.31 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3302, pruned_loss=0.08628, over 5755712.47 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3238, pruned_loss=0.08206, over 5650926.55 frames. ], batch size: 262, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:36:52,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5985, 1.7685, 1.2419, 1.4343], device='cuda:1'), covar=tensor([0.0948, 0.0522, 0.0953, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0442, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 22:37:29,480 INFO [optim.py:369] (1/2) 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 22:37:55,265 INFO [train.py:968] (1/2) Epoch 25, batch 15400, libri_loss[loss=0.2365, simple_loss=0.3166, pruned_loss=0.07818, over 29550.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3225, pruned_loss=0.0816, over 5665201.11 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3299, pruned_loss=0.08618, over 5745251.95 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3233, pruned_loss=0.08167, over 5662852.41 frames. ], batch size: 79, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:38:37,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3811, 1.1169, 4.0065, 3.3137], device='cuda:1'), covar=tensor([0.1610, 0.2938, 0.0434, 0.0917], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0658, 0.0967, 0.0929], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 22:38:45,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8270, 2.0998, 1.9869, 1.8021], device='cuda:1'), covar=tensor([0.2167, 0.2705, 0.2185, 0.2536], device='cuda:1'), in_proj_covar=tensor([0.0476, 0.0734, 0.0707, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-12 22:38:50,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1380, 1.2809, 1.1263, 0.9564], device='cuda:1'), covar=tensor([0.0935, 0.0457, 0.0984, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0441, 0.0517, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 22:38:55,116 INFO [train.py:968] (1/2) Epoch 25, batch 15450, giga_loss[loss=0.2175, simple_loss=0.2978, pruned_loss=0.06866, over 28565.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3237, pruned_loss=0.08166, over 5678083.28 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3298, pruned_loss=0.08616, over 5745526.20 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3243, pruned_loss=0.08162, over 5674339.87 frames. ], batch size: 85, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:39:30,054 INFO [optim.py:369] (1/2) 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 22:39:52,982 INFO [train.py:968] (1/2) Epoch 25, batch 15500, libri_loss[loss=0.2818, simple_loss=0.3567, pruned_loss=0.1035, over 28838.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3259, pruned_loss=0.0833, over 5690293.79 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3302, pruned_loss=0.08638, over 5749146.87 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3257, pruned_loss=0.08292, over 5682117.03 frames. ], batch size: 107, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:40:56,836 INFO [train.py:968] (1/2) Epoch 25, batch 15550, giga_loss[loss=0.2529, simple_loss=0.3281, pruned_loss=0.0888, over 29022.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3263, pruned_loss=0.08431, over 5681922.14 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3301, pruned_loss=0.08632, over 5743191.62 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3262, pruned_loss=0.08402, over 5680152.61 frames. ], batch size: 213, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:41:26,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3542, 1.6272, 1.5371, 1.2698], device='cuda:1'), covar=tensor([0.3085, 0.2459, 0.1927, 0.2544], device='cuda:1'), in_proj_covar=tensor([0.1973, 0.1899, 0.1814, 0.1965], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 22:41:33,285 INFO [optim.py:369] (1/2) 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 22:41:58,379 INFO [train.py:968] (1/2) Epoch 25, batch 15600, giga_loss[loss=0.2365, simple_loss=0.309, pruned_loss=0.08198, over 26846.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3257, pruned_loss=0.08301, over 5671310.20 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3304, pruned_loss=0.08652, over 5745095.90 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3254, pruned_loss=0.0826, over 5667758.90 frames. ], batch size: 555, lr: 1.28e-03, grad_scale: 8.0
+2023-03-12 22:42:58,350 INFO [train.py:968] (1/2) Epoch 25, batch 15650, libri_loss[loss=0.2656, simple_loss=0.3364, pruned_loss=0.09737, over 29548.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3284, pruned_loss=0.08344, over 5663159.45 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3305, pruned_loss=0.08677, over 5744568.16 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3279, pruned_loss=0.08282, over 5659596.73 frames. ], batch size: 79, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:43:27,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5085, 1.6533, 1.7339, 1.3227], device='cuda:1'), covar=tensor([0.1774, 0.2587, 0.1516, 0.1979], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0702, 0.0965, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 22:43:28,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5588, 2.1111, 1.6598, 1.6331], device='cuda:1'), covar=tensor([0.0789, 0.0257, 0.0319, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-12 22:43:33,789 INFO [optim.py:369] (1/2) 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 22:43:57,183 INFO [train.py:968] (1/2) Epoch 25, batch 15700, giga_loss[loss=0.2244, simple_loss=0.3213, pruned_loss=0.06378, over 28880.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3307, pruned_loss=0.08449, over 5664365.16 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3301, pruned_loss=0.08661, over 5749425.15 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3306, pruned_loss=0.08407, over 5654826.02 frames. ], batch size: 174, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:44:05,624 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2428, 4.0815, 3.8741, 1.8440], device='cuda:1'), covar=tensor([0.0626, 0.0735, 0.0806, 0.2124], device='cuda:1'), in_proj_covar=tensor([0.1250, 0.1153, 0.0970, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 22:44:55,524 INFO [train.py:968] (1/2) Epoch 25, batch 15750, giga_loss[loss=0.2569, simple_loss=0.3418, pruned_loss=0.08603, over 28879.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3322, pruned_loss=0.08544, over 5663882.86 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3302, pruned_loss=0.08657, over 5752380.68 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3322, pruned_loss=0.08513, over 5651697.12 frames. ], batch size: 284, lr: 1.28e-03, grad_scale: 4.0
+2023-03-12 22:45:28,520 INFO [optim.py:369] (1/2) 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 22:45:50,475 INFO [train.py:968] (1/2) Epoch 25, batch 15800, giga_loss[loss=0.2016, simple_loss=0.2897, pruned_loss=0.05677, over 28820.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3304, pruned_loss=0.08477, over 5661847.67 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3295, pruned_loss=0.08633, over 5755642.64 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3311, pruned_loss=0.0847, over 5646459.31 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:46:50,578 INFO [train.py:968] (1/2) Epoch 25, batch 15850, giga_loss[loss=0.2359, simple_loss=0.3162, pruned_loss=0.07779, over 28651.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.327, pruned_loss=0.08288, over 5659569.41 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3291, pruned_loss=0.08603, over 5754881.52 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3279, pruned_loss=0.08302, over 5645142.79 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:47:22,699 INFO [optim.py:369] (1/2) 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 22:47:43,960 INFO [train.py:968] (1/2) Epoch 25, batch 15900, giga_loss[loss=0.2346, simple_loss=0.3131, pruned_loss=0.07803, over 28770.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3261, pruned_loss=0.08268, over 5664066.56 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3289, pruned_loss=0.08597, over 5747360.75 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3269, pruned_loss=0.08274, over 5654899.63 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:48:08,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-12 22:48:32,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9058, 1.2542, 0.9754, 0.2396], device='cuda:1'), covar=tensor([0.3151, 0.2340, 0.3491, 0.5210], device='cuda:1'), in_proj_covar=tensor([0.1792, 0.1683, 0.1624, 0.1457], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 22:48:39,756 INFO [train.py:968] (1/2) Epoch 25, batch 15950, libri_loss[loss=0.2326, simple_loss=0.32, pruned_loss=0.07257, over 29149.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3253, pruned_loss=0.08243, over 5664823.00 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3284, pruned_loss=0.08569, over 5742570.19 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3263, pruned_loss=0.08262, over 5658248.29 frames. ], batch size: 101, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:49:00,919 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-12 22:49:15,390 INFO [optim.py:369] (1/2) 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 22:49:39,650 INFO [train.py:968] (1/2) Epoch 25, batch 16000, giga_loss[loss=0.2761, simple_loss=0.3574, pruned_loss=0.09734, over 28978.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3279, pruned_loss=0.08344, over 5673534.69 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3283, pruned_loss=0.08565, over 5746719.18 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3288, pruned_loss=0.08356, over 5662818.26 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 22:50:43,149 INFO [train.py:968] (1/2) Epoch 25, batch 16050, giga_loss[loss=0.3028, simple_loss=0.3608, pruned_loss=0.1223, over 26904.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3294, pruned_loss=0.08494, over 5662166.07 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3279, pruned_loss=0.08547, over 5748951.09 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3305, pruned_loss=0.08517, over 5649915.50 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:51:18,829 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 16100, giga_loss[loss=0.2558, simple_loss=0.3377, pruned_loss=0.08696, over 28731.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3308, pruned_loss=0.08586, over 5674372.76 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3274, pruned_loss=0.08517, over 5753645.21 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3321, pruned_loss=0.08629, over 5658010.39 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:52:19,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6140, 1.8762, 1.5578, 1.8118], device='cuda:1'), covar=tensor([0.2883, 0.2744, 0.3048, 0.2562], device='cuda:1'), in_proj_covar=tensor([0.1557, 0.1117, 0.1374, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 22:52:37,675 INFO [train.py:968] (1/2) Epoch 25, batch 16150, libri_loss[loss=0.2463, simple_loss=0.3297, pruned_loss=0.0814, over 29520.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3345, pruned_loss=0.08752, over 5662583.77 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3275, pruned_loss=0.08508, over 5756159.75 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3357, pruned_loss=0.08803, over 5644248.12 frames. ], batch size: 84, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:52:46,488 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0351, 3.8747, 3.6840, 1.9398], device='cuda:1'), covar=tensor([0.0618, 0.0789, 0.0855, 0.2323], device='cuda:1'), in_proj_covar=tensor([0.1251, 0.1149, 0.0971, 0.0724], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 22:53:09,710 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 16200, giga_loss[loss=0.256, simple_loss=0.3436, pruned_loss=0.08423, over 28770.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3358, pruned_loss=0.08764, over 5665473.24 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3276, pruned_loss=0.08534, over 5759733.91 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3368, pruned_loss=0.08787, over 5645951.23 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:54:41,078 INFO [train.py:968] (1/2) Epoch 25, batch 16250, giga_loss[loss=0.336, simple_loss=0.3773, pruned_loss=0.1474, over 26879.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3361, pruned_loss=0.08821, over 5658865.28 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3272, pruned_loss=0.08515, over 5761953.70 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3374, pruned_loss=0.08859, over 5640101.88 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:55:25,025 INFO [optim.py:369] (1/2) 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 22:55:42,252 INFO [train.py:968] (1/2) Epoch 25, batch 16300, giga_loss[loss=0.2138, simple_loss=0.2979, pruned_loss=0.06489, over 28956.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3328, pruned_loss=0.08641, over 5670529.79 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.327, pruned_loss=0.08508, over 5764076.60 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3341, pruned_loss=0.0868, over 5652296.41 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:56:46,397 INFO [train.py:968] (1/2) Epoch 25, batch 16350, giga_loss[loss=0.2834, simple_loss=0.3587, pruned_loss=0.104, over 28043.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3315, pruned_loss=0.08549, over 5671633.24 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3266, pruned_loss=0.0849, over 5762129.57 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3329, pruned_loss=0.08598, over 5656646.93 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 2.0
+2023-03-12 22:57:02,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 22:57:21,970 INFO [optim.py:369] (1/2) 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 22:57:42,016 INFO [train.py:968] (1/2) Epoch 25, batch 16400, giga_loss[loss=0.2358, simple_loss=0.3108, pruned_loss=0.08043, over 29145.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3309, pruned_loss=0.08609, over 5679231.31 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3257, pruned_loss=0.08449, over 5767760.55 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3331, pruned_loss=0.08689, over 5658667.96 frames. ], batch size: 100, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:57:51,285 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5606, 1.9302, 1.2500, 1.5474], device='cuda:1'), covar=tensor([0.1068, 0.0582, 0.1193, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0443, 0.0519, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 22:58:43,591 INFO [train.py:968] (1/2) Epoch 25, batch 16450, giga_loss[loss=0.2746, simple_loss=0.3378, pruned_loss=0.1057, over 26852.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.329, pruned_loss=0.08597, over 5667245.34 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3259, pruned_loss=0.08462, over 5766894.01 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3307, pruned_loss=0.08651, over 5649880.59 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:59:19,580 INFO [optim.py:369] (1/2) 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 22:59:34,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2027, 1.6199, 1.4892, 1.0859], device='cuda:1'), covar=tensor([0.1751, 0.2936, 0.1589, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0701, 0.0965, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-12 22:59:42,580 INFO [train.py:968] (1/2) Epoch 25, batch 16500, giga_loss[loss=0.2184, simple_loss=0.3049, pruned_loss=0.06597, over 28769.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3267, pruned_loss=0.08438, over 5666575.41 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3255, pruned_loss=0.08448, over 5768496.93 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3284, pruned_loss=0.08495, over 5648263.84 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 22:59:50,762 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-12 22:59:58,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2291, 5.0485, 4.8407, 2.3540], device='cuda:1'), covar=tensor([0.0510, 0.0631, 0.0887, 0.1899], device='cuda:1'), in_proj_covar=tensor([0.1253, 0.1150, 0.0974, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 23:00:41,538 INFO [train.py:968] (1/2) Epoch 25, batch 16550, giga_loss[loss=0.2421, simple_loss=0.3387, pruned_loss=0.07271, over 28372.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3267, pruned_loss=0.08294, over 5673027.04 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3254, pruned_loss=0.08441, over 5761236.93 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3282, pruned_loss=0.08343, over 5663397.44 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:01:17,951 INFO [optim.py:369] (1/2) 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 23:01:38,254 INFO [train.py:968] (1/2) Epoch 25, batch 16600, giga_loss[loss=0.2456, simple_loss=0.3346, pruned_loss=0.07833, over 28889.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3274, pruned_loss=0.08099, over 5681255.72 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3252, pruned_loss=0.08426, over 5763583.89 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3288, pruned_loss=0.08148, over 5670360.28 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:02:35,298 INFO [train.py:968] (1/2) Epoch 25, batch 16650, giga_loss[loss=0.2408, simple_loss=0.3311, pruned_loss=0.07522, over 28474.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3291, pruned_loss=0.081, over 5676925.27 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3252, pruned_loss=0.08423, over 5762677.67 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3303, pruned_loss=0.08138, over 5668016.55 frames. ], batch size: 369, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:03:09,462 INFO [optim.py:369] (1/2) 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 23:03:31,645 INFO [train.py:968] (1/2) Epoch 25, batch 16700, giga_loss[loss=0.2532, simple_loss=0.3399, pruned_loss=0.08325, over 29024.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3297, pruned_loss=0.08123, over 5674491.09 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3252, pruned_loss=0.08421, over 5763654.07 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3307, pruned_loss=0.08146, over 5664126.37 frames. ], batch size: 214, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:04:33,369 INFO [train.py:968] (1/2) Epoch 25, batch 16750, giga_loss[loss=0.2365, simple_loss=0.3189, pruned_loss=0.077, over 28168.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3289, pruned_loss=0.08143, over 5670445.12 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3246, pruned_loss=0.08404, over 5766556.01 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3304, pruned_loss=0.0817, over 5657318.22 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:05:16,294 INFO [optim.py:369] (1/2) 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 23:05:41,676 INFO [train.py:968] (1/2) Epoch 25, batch 16800, giga_loss[loss=0.2641, simple_loss=0.3406, pruned_loss=0.09376, over 28720.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.328, pruned_loss=0.08073, over 5662665.59 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3247, pruned_loss=0.08401, over 5766580.48 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3292, pruned_loss=0.08092, over 5650936.98 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:06:01,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3548, 1.8720, 1.4027, 1.5444], device='cuda:1'), covar=tensor([0.0781, 0.0320, 0.0353, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-12 23:06:29,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6241, 2.3599, 1.7106, 0.7692], device='cuda:1'), covar=tensor([0.7342, 0.3913, 0.4999, 0.7666], device='cuda:1'), in_proj_covar=tensor([0.1787, 0.1679, 0.1622, 0.1453], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 23:06:35,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8293, 2.6800, 1.7011, 0.9016], device='cuda:1'), covar=tensor([0.8347, 0.3998, 0.4849, 0.7621], device='cuda:1'), in_proj_covar=tensor([0.1787, 0.1678, 0.1622, 0.1453], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 23:06:51,642 INFO [train.py:968] (1/2) Epoch 25, batch 16850, giga_loss[loss=0.2231, simple_loss=0.3129, pruned_loss=0.06663, over 28928.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3281, pruned_loss=0.07994, over 5655703.52 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3248, pruned_loss=0.0841, over 5760821.56 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.329, pruned_loss=0.07991, over 5648381.65 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:07:02,282 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 23:07:34,243 INFO [optim.py:369] (1/2) 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 23:07:39,595 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1110725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:07:59,741 INFO [train.py:968] (1/2) Epoch 25, batch 16900, giga_loss[loss=0.2872, simple_loss=0.3719, pruned_loss=0.1013, over 28815.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3316, pruned_loss=0.0817, over 5658467.02 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.325, pruned_loss=0.08413, over 5761734.55 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3322, pruned_loss=0.08159, over 5650251.78 frames. ], batch size: 263, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:09:08,501 INFO [train.py:968] (1/2) Epoch 25, batch 16950, giga_loss[loss=0.2574, simple_loss=0.337, pruned_loss=0.0889, over 27611.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3353, pruned_loss=0.08385, over 5664558.87 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3256, pruned_loss=0.0845, over 5763369.91 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3354, pruned_loss=0.08339, over 5654828.69 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:09:53,320 INFO [optim.py:369] (1/2) 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 23:10:09,125 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1110835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:10:16,060 INFO [train.py:968] (1/2) Epoch 25, batch 17000, giga_loss[loss=0.2243, simple_loss=0.3105, pruned_loss=0.06908, over 29033.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3348, pruned_loss=0.08411, over 5674029.71 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3253, pruned_loss=0.08435, over 5763427.67 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3352, pruned_loss=0.0839, over 5665431.49 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:10:44,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-12 23:11:21,407 INFO [train.py:968] (1/2) Epoch 25, batch 17050, giga_loss[loss=0.2256, simple_loss=0.3134, pruned_loss=0.06896, over 28227.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.333, pruned_loss=0.08359, over 5676825.32 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3257, pruned_loss=0.08442, over 5764354.59 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3333, pruned_loss=0.08335, over 5666170.45 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:11:36,558 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-12 23:12:05,091 INFO [optim.py:369] (1/2) 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 23:12:23,831 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1110932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:12:31,102 INFO [train.py:968] (1/2) Epoch 25, batch 17100, giga_loss[loss=0.2396, simple_loss=0.3276, pruned_loss=0.07582, over 28073.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.332, pruned_loss=0.0826, over 5678250.26 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3256, pruned_loss=0.08443, over 5767355.99 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3324, pruned_loss=0.08239, over 5665617.99 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:13:29,555 INFO [train.py:968] (1/2) Epoch 25, batch 17150, libri_loss[loss=0.2767, simple_loss=0.353, pruned_loss=0.1002, over 29276.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3312, pruned_loss=0.0825, over 5667483.34 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3256, pruned_loss=0.08459, over 5753655.19 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3318, pruned_loss=0.08207, over 5665181.12 frames. ], batch size: 94, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:13:30,211 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-12 23:13:41,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6389, 1.8519, 1.2300, 1.4635], device='cuda:1'), covar=tensor([0.1004, 0.0569, 0.1077, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0441, 0.0516, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 23:14:06,282 INFO [optim.py:369] (1/2) 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 23:14:07,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1111024.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:14:25,371 INFO [train.py:968] (1/2) Epoch 25, batch 17200, giga_loss[loss=0.256, simple_loss=0.3428, pruned_loss=0.08464, over 28912.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3318, pruned_loss=0.08307, over 5673558.94 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3251, pruned_loss=0.08436, over 5757162.35 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3329, pruned_loss=0.0829, over 5666615.85 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:14:39,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5473, 1.8814, 2.1097, 1.5789], device='cuda:1'), covar=tensor([0.3466, 0.2545, 0.2332, 0.2836], device='cuda:1'), in_proj_covar=tensor([0.1981, 0.1905, 0.1816, 0.1966], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 23:15:19,562 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4276, 1.6665, 1.2492, 1.2037], device='cuda:1'), covar=tensor([0.1022, 0.0472, 0.1031, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0441, 0.0516, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 23:15:20,769 INFO [train.py:968] (1/2) Epoch 25, batch 17250, giga_loss[loss=0.2632, simple_loss=0.3411, pruned_loss=0.0927, over 29021.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3343, pruned_loss=0.08457, over 5655049.06 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3253, pruned_loss=0.08456, over 5740203.77 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3352, pruned_loss=0.08425, over 5662513.09 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:15:33,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1111100.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:15:59,040 INFO [optim.py:369] (1/2) 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 23:16:14,770 INFO [train.py:968] (1/2) Epoch 25, batch 17300, giga_loss[loss=0.2874, simple_loss=0.354, pruned_loss=0.1104, over 28888.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3332, pruned_loss=0.08538, over 5654485.26 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3249, pruned_loss=0.08444, over 5734386.18 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3344, pruned_loss=0.08522, over 5664443.21 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:17:15,390 INFO [train.py:968] (1/2) Epoch 25, batch 17350, giga_loss[loss=0.2299, simple_loss=0.3188, pruned_loss=0.07052, over 28808.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3304, pruned_loss=0.08472, over 5655158.08 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3243, pruned_loss=0.08411, over 5738167.49 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.332, pruned_loss=0.0849, over 5658162.24 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:17:39,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1111210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:17:50,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-12 23:17:53,277 INFO [optim.py:369] (1/2) 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 23:18:11,574 INFO [train.py:968] (1/2) Epoch 25, batch 17400, giga_loss[loss=0.3141, simple_loss=0.3676, pruned_loss=0.1303, over 26837.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3308, pruned_loss=0.08561, over 5648220.76 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3245, pruned_loss=0.08414, over 5738553.87 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3321, pruned_loss=0.08575, over 5647842.26 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:18:16,752 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111243.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:18:19,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111246.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:18:45,690 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 17450, libri_loss[loss=0.2379, simple_loss=0.3173, pruned_loss=0.07921, over 29341.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.337, pruned_loss=0.08886, over 5652473.05 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3244, pruned_loss=0.08401, over 5732949.88 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3385, pruned_loss=0.08921, over 5653435.18 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:19:19,801 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:19:44,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4174, 1.6286, 1.2410, 1.1168], device='cuda:1'), covar=tensor([0.1118, 0.0578, 0.1106, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0441, 0.0517, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 23:19:48,534 INFO [train.py:968] (1/2) Epoch 25, batch 17500, giga_loss[loss=0.2907, simple_loss=0.3675, pruned_loss=0.1069, over 28787.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3465, pruned_loss=0.09408, over 5663133.75 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3241, pruned_loss=0.08388, over 5735506.77 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3482, pruned_loss=0.09456, over 5660717.73 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:20:01,583 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:20:03,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:20:30,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1111385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:20:33,340 INFO [train.py:968] (1/2) Epoch 25, batch 17550, giga_loss[loss=0.2782, simple_loss=0.3447, pruned_loss=0.1058, over 27914.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3479, pruned_loss=0.09524, over 5665428.02 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3239, pruned_loss=0.08375, over 5737856.91 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3498, pruned_loss=0.09598, over 5659891.34 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:20:42,014 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:21:19,292 INFO [train.py:968] (1/2) Epoch 25, batch 17600, giga_loss[loss=0.2234, simple_loss=0.304, pruned_loss=0.07144, over 28996.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.343, pruned_loss=0.0933, over 5674960.57 frames. ], libri_tot_loss[loss=0.2455, simple_loss=0.3238, pruned_loss=0.08364, over 5741787.71 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3452, pruned_loss=0.09424, over 5665552.38 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:21:27,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:21:29,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:21:55,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1111482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:22:05,224 INFO [train.py:968] (1/2) Epoch 25, batch 17650, giga_loss[loss=0.2578, simple_loss=0.3345, pruned_loss=0.09054, over 28471.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3364, pruned_loss=0.09071, over 5684631.83 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.324, pruned_loss=0.08364, over 5743283.69 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3381, pruned_loss=0.09152, over 5675370.37 frames. ], batch size: 65, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:22:35,156 INFO [optim.py:369] (1/2) 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 23:22:43,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0024, 5.0275, 2.1317, 2.2677], device='cuda:1'), covar=tensor([0.0925, 0.0219, 0.0842, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0559, 0.0399, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 23:22:45,722 INFO [train.py:968] (1/2) Epoch 25, batch 17700, giga_loss[loss=0.1965, simple_loss=0.2802, pruned_loss=0.05636, over 29005.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3297, pruned_loss=0.08759, over 5694233.79 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3247, pruned_loss=0.0839, over 5746639.85 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3306, pruned_loss=0.08812, over 5682514.07 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:22:48,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1111542.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:22:52,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1111545.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:23:07,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.23 vs. limit=5.0
+2023-03-12 23:23:19,929 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1111574.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:23:32,880 INFO [train.py:968] (1/2) Epoch 25, batch 17750, giga_loss[loss=0.2138, simple_loss=0.2928, pruned_loss=0.06741, over 28818.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3224, pruned_loss=0.08494, over 5694127.87 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3249, pruned_loss=0.08396, over 5747968.78 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3229, pruned_loss=0.08533, over 5683260.47 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:23:49,707 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3766, 1.6194, 1.3970, 1.5398], device='cuda:1'), covar=tensor([0.0786, 0.0332, 0.0335, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-12 23:23:56,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3242, 3.1432, 3.0088, 1.4726], device='cuda:1'), covar=tensor([0.1113, 0.1290, 0.1155, 0.2288], device='cuda:1'), in_proj_covar=tensor([0.1249, 0.1149, 0.0969, 0.0725], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 23:24:02,366 INFO [optim.py:369] (1/2) 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 23:24:15,395 INFO [train.py:968] (1/2) Epoch 25, batch 17800, giga_loss[loss=0.2277, simple_loss=0.3037, pruned_loss=0.07581, over 28258.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3173, pruned_loss=0.08288, over 5688623.60 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3257, pruned_loss=0.08437, over 5747235.42 frames. ], giga_tot_loss[loss=0.2412, simple_loss=0.3168, pruned_loss=0.08277, over 5679537.63 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:24:16,878 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1111642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:24:46,705 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-12 23:24:54,797 INFO [train.py:968] (1/2) Epoch 25, batch 17850, giga_loss[loss=0.2292, simple_loss=0.3007, pruned_loss=0.07888, over 28927.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.314, pruned_loss=0.08153, over 5698458.62 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.326, pruned_loss=0.08445, over 5751110.03 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.3129, pruned_loss=0.08128, over 5685938.74 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:25:21,817 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 17900, giga_loss[loss=0.2037, simple_loss=0.2845, pruned_loss=0.06146, over 28868.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3119, pruned_loss=0.08038, over 5694890.96 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3267, pruned_loss=0.08482, over 5744737.83 frames. ], giga_tot_loss[loss=0.2348, simple_loss=0.31, pruned_loss=0.07976, over 5688957.03 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:25:55,930 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4906, 2.2047, 1.7749, 0.7396], device='cuda:1'), covar=tensor([0.6592, 0.3314, 0.4552, 0.6918], device='cuda:1'), in_proj_covar=tensor([0.1788, 0.1683, 0.1621, 0.1450], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 23:26:14,498 INFO [train.py:968] (1/2) Epoch 25, batch 17950, libri_loss[loss=0.2895, simple_loss=0.3716, pruned_loss=0.1037, over 29766.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3091, pruned_loss=0.07886, over 5694500.79 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3276, pruned_loss=0.08502, over 5745537.71 frames. ], giga_tot_loss[loss=0.2311, simple_loss=0.3063, pruned_loss=0.07799, over 5687129.89 frames. ], batch size: 87, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:26:40,950 INFO [optim.py:369] (1/2) 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 23:26:51,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1155, 3.3142, 2.3532, 1.1509], device='cuda:1'), covar=tensor([0.8894, 0.3069, 0.4115, 0.7682], device='cuda:1'), in_proj_covar=tensor([0.1786, 0.1682, 0.1620, 0.1448], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 23:26:54,197 INFO [train.py:968] (1/2) Epoch 25, batch 18000, giga_loss[loss=0.1995, simple_loss=0.2706, pruned_loss=0.06415, over 28729.00 frames. ], tot_loss[loss=0.2301, simple_loss=0.3057, pruned_loss=0.07724, over 5693085.80 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3277, pruned_loss=0.08504, over 5740959.65 frames. ], giga_tot_loss[loss=0.2277, simple_loss=0.3027, pruned_loss=0.07632, over 5689199.85 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:26:54,198 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-12 23:27:03,034 INFO [train.py:1012] (1/2) Epoch 25, validation: loss=0.2015, simple_loss=0.3074, pruned_loss=0.04783, over 944034.00 frames. 
+2023-03-12 23:27:03,035 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-12 23:27:42,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7250, 2.0389, 1.9815, 1.5880], device='cuda:1'), covar=tensor([0.3717, 0.2730, 0.2674, 0.3314], device='cuda:1'), in_proj_covar=tensor([0.1998, 0.1923, 0.1827, 0.1985], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 23:27:44,233 INFO [train.py:968] (1/2) Epoch 25, batch 18050, giga_loss[loss=0.2004, simple_loss=0.2796, pruned_loss=0.06062, over 28596.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.3031, pruned_loss=0.07585, over 5698507.05 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3276, pruned_loss=0.08489, over 5743297.96 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.3, pruned_loss=0.07499, over 5691915.12 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:28:15,316 INFO [optim.py:369] (1/2) 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 23:28:17,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3965, 3.0281, 1.5830, 1.5037], device='cuda:1'), covar=tensor([0.0936, 0.0369, 0.0848, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0556, 0.0397, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-12 23:28:30,115 INFO [train.py:968] (1/2) Epoch 25, batch 18100, giga_loss[loss=0.2184, simple_loss=0.2861, pruned_loss=0.07533, over 28661.00 frames. ], tot_loss[loss=0.2253, simple_loss=0.3005, pruned_loss=0.07507, over 5683603.48 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3276, pruned_loss=0.08487, over 5735585.75 frames. ], giga_tot_loss[loss=0.2232, simple_loss=0.2978, pruned_loss=0.07432, over 5683889.94 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:28:44,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6514, 1.9148, 1.5288, 1.8516], device='cuda:1'), covar=tensor([0.2646, 0.2727, 0.3247, 0.2490], device='cuda:1'), in_proj_covar=tensor([0.1559, 0.1120, 0.1377, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 23:29:09,562 INFO [train.py:968] (1/2) Epoch 25, batch 18150, giga_loss[loss=0.2065, simple_loss=0.2811, pruned_loss=0.06597, over 29026.00 frames. ], tot_loss[loss=0.2228, simple_loss=0.2981, pruned_loss=0.07375, over 5686483.81 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3278, pruned_loss=0.08481, over 5735823.71 frames. ], giga_tot_loss[loss=0.2204, simple_loss=0.295, pruned_loss=0.07289, over 5685137.76 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:29:23,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:29:35,966 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:29:55,049 INFO [train.py:968] (1/2) Epoch 25, batch 18200, giga_loss[loss=0.2452, simple_loss=0.3033, pruned_loss=0.09352, over 26704.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2959, pruned_loss=0.07287, over 5688383.72 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3279, pruned_loss=0.08474, over 5728417.70 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2925, pruned_loss=0.072, over 5693425.38 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:30:37,280 INFO [train.py:968] (1/2) Epoch 25, batch 18250, giga_loss[loss=0.2764, simple_loss=0.3494, pruned_loss=0.1017, over 28698.00 frames. ], tot_loss[loss=0.2215, simple_loss=0.2962, pruned_loss=0.07336, over 5681780.35 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3279, pruned_loss=0.08474, over 5719394.87 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2929, pruned_loss=0.0724, over 5693272.15 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:30:38,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-12 23:31:14,412 INFO [optim.py:369] (1/2) 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 23:31:17,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3184, 1.2990, 1.3472, 1.5725], device='cuda:1'), covar=tensor([0.0799, 0.0383, 0.0345, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-12 23:31:26,725 INFO [train.py:968] (1/2) Epoch 25, batch 18300, giga_loss[loss=0.283, simple_loss=0.3598, pruned_loss=0.1031, over 28775.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3083, pruned_loss=0.07985, over 5683903.06 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3282, pruned_loss=0.08478, over 5718956.75 frames. ], giga_tot_loss[loss=0.2314, simple_loss=0.3051, pruned_loss=0.07891, over 5692840.52 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:31:39,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-12 23:31:46,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4372, 1.7095, 1.2839, 1.2969], device='cuda:1'), covar=tensor([0.1003, 0.0468, 0.1015, 0.0990], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0443, 0.0519, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 23:31:46,787 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:31:50,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1112163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:32:10,093 INFO [train.py:968] (1/2) Epoch 25, batch 18350, giga_loss[loss=0.2668, simple_loss=0.3384, pruned_loss=0.09757, over 28615.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3217, pruned_loss=0.08673, over 5687618.85 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3284, pruned_loss=0.08479, over 5722829.85 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3187, pruned_loss=0.08596, over 5690514.56 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:32:12,845 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:32:37,860 INFO [optim.py:369] (1/2) 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 23:32:50,110 INFO [train.py:968] (1/2) Epoch 25, batch 18400, giga_loss[loss=0.2925, simple_loss=0.3666, pruned_loss=0.1092, over 28860.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3303, pruned_loss=0.09033, over 5695189.46 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3284, pruned_loss=0.0848, over 5726509.81 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3278, pruned_loss=0.0898, over 5693496.88 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:33:24,484 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:33:29,316 INFO [train.py:968] (1/2) Epoch 25, batch 18450, giga_loss[loss=0.3001, simple_loss=0.3818, pruned_loss=0.1092, over 27597.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3352, pruned_loss=0.0914, over 5698374.59 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3286, pruned_loss=0.08484, over 5732053.91 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3332, pruned_loss=0.0911, over 5691083.69 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:33:58,093 INFO [optim.py:369] (1/2) 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 23:34:06,842 INFO [train.py:968] (1/2) Epoch 25, batch 18500, giga_loss[loss=0.239, simple_loss=0.3325, pruned_loss=0.0728, over 28697.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3382, pruned_loss=0.09157, over 5700495.03 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3293, pruned_loss=0.08516, over 5735723.55 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3362, pruned_loss=0.09122, over 5690495.35 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:34:19,281 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:34:46,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1112379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:34:56,224 INFO [train.py:968] (1/2) Epoch 25, batch 18550, giga_loss[loss=0.2951, simple_loss=0.3472, pruned_loss=0.1215, over 23615.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3391, pruned_loss=0.09181, over 5690313.04 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3293, pruned_loss=0.08509, over 5736583.93 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3375, pruned_loss=0.09162, over 5681663.16 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:35:20,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3255, 3.1273, 2.9769, 1.6264], device='cuda:1'), covar=tensor([0.0943, 0.1152, 0.0937, 0.2138], device='cuda:1'), in_proj_covar=tensor([0.1245, 0.1151, 0.0970, 0.0728], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-12 23:35:28,855 INFO [optim.py:369] (1/2) 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 23:35:38,852 INFO [train.py:968] (1/2) Epoch 25, batch 18600, libri_loss[loss=0.2387, simple_loss=0.3331, pruned_loss=0.07221, over 29524.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3408, pruned_loss=0.09333, over 5688704.96 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.08494, over 5731581.45 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.34, pruned_loss=0.09357, over 5685358.90 frames. ], batch size: 82, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:35:42,616 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-12 23:36:22,939 INFO [train.py:968] (1/2) Epoch 25, batch 18650, giga_loss[loss=0.295, simple_loss=0.3636, pruned_loss=0.1132, over 29023.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3451, pruned_loss=0.0964, over 5688920.51 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3295, pruned_loss=0.085, over 5730864.76 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3444, pruned_loss=0.09662, over 5686463.78 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:36:50,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:36:52,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6188, 1.8342, 1.3033, 1.3117], device='cuda:1'), covar=tensor([0.1129, 0.0669, 0.1133, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0445, 0.0522, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-12 23:36:52,056 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 25, batch 18700, giga_loss[loss=0.2706, simple_loss=0.3528, pruned_loss=0.09418, over 28723.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3479, pruned_loss=0.09746, over 5695552.99 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3299, pruned_loss=0.08502, over 5734377.62 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3474, pruned_loss=0.0979, over 5689570.08 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:37:17,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:37:19,606 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1734, 1.3432, 5.2534, 3.8906], device='cuda:1'), covar=tensor([0.1424, 0.2863, 0.0367, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0773, 0.0657, 0.0965, 0.0928], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 23:37:43,089 INFO [train.py:968] (1/2) Epoch 25, batch 18750, giga_loss[loss=0.2799, simple_loss=0.358, pruned_loss=0.1009, over 28263.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3499, pruned_loss=0.09772, over 5704092.69 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3296, pruned_loss=0.08484, over 5735742.68 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3502, pruned_loss=0.09859, over 5697375.27 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:38:13,494 INFO [optim.py:369] (1/2) 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 23:38:25,995 INFO [train.py:968] (1/2) Epoch 25, batch 18800, giga_loss[loss=0.2735, simple_loss=0.3553, pruned_loss=0.09586, over 28862.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3511, pruned_loss=0.09762, over 5703506.53 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3298, pruned_loss=0.08484, over 5735360.18 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3514, pruned_loss=0.0984, over 5698223.37 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:38:39,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1112659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:39:05,625 INFO [train.py:968] (1/2) Epoch 25, batch 18850, giga_loss[loss=0.2784, simple_loss=0.3574, pruned_loss=0.09967, over 28615.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3527, pruned_loss=0.09814, over 5703786.78 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.33, pruned_loss=0.08488, over 5740477.48 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3533, pruned_loss=0.09911, over 5694171.24 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:39:12,927 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-12 23:39:35,429 INFO [optim.py:369] (1/2) 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 23:39:35,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1112727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:39:45,872 INFO [train.py:968] (1/2) Epoch 25, batch 18900, giga_loss[loss=0.2669, simple_loss=0.3487, pruned_loss=0.09257, over 28219.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3522, pruned_loss=0.09709, over 5701155.93 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3302, pruned_loss=0.08487, over 5741270.43 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3531, pruned_loss=0.09815, over 5692065.32 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:40:02,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5789, 1.7902, 1.4611, 1.6497], device='cuda:1'), covar=tensor([0.2581, 0.2665, 0.2873, 0.2537], device='cuda:1'), in_proj_covar=tensor([0.1557, 0.1121, 0.1374, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-12 23:40:11,318 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:40:23,378 INFO [train.py:968] (1/2) Epoch 25, batch 18950, giga_loss[loss=0.253, simple_loss=0.3441, pruned_loss=0.08097, over 28584.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3498, pruned_loss=0.09457, over 5710806.26 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3307, pruned_loss=0.08512, over 5743174.10 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3503, pruned_loss=0.09532, over 5701456.11 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:40:35,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:40:37,535 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1112805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:40:54,216 INFO [optim.py:369] (1/2) 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 23:41:00,452 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:03,857 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:04,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-12 23:41:05,803 INFO [train.py:968] (1/2) Epoch 25, batch 19000, giga_loss[loss=0.2912, simple_loss=0.3687, pruned_loss=0.1069, over 29016.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3503, pruned_loss=0.09513, over 5705964.42 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.331, pruned_loss=0.08527, over 5744645.34 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3505, pruned_loss=0.09568, over 5696993.08 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:41:30,570 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1112867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:35,026 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1112870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:37,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1112873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:41:53,431 INFO [train.py:968] (1/2) Epoch 25, batch 19050, giga_loss[loss=0.2929, simple_loss=0.3595, pruned_loss=0.1131, over 28714.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3535, pruned_loss=0.1003, over 5692052.59 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3312, pruned_loss=0.08532, over 5743597.12 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3537, pruned_loss=0.1007, over 5685789.01 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:42:06,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1112902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:42:09,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-12 23:42:25,816 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:42:37,196 INFO [train.py:968] (1/2) Epoch 25, batch 19100, giga_loss[loss=0.3538, simple_loss=0.3984, pruned_loss=0.1546, over 28194.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3543, pruned_loss=0.1022, over 5690875.35 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3314, pruned_loss=0.08536, over 5747769.62 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3549, pruned_loss=0.1029, over 5680720.16 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:43:12,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3320, 2.9661, 1.4003, 1.4604], device='cuda:1'), covar=tensor([0.1027, 0.0346, 0.0904, 0.1360], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0556, 0.0398, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 23:43:17,412 INFO [train.py:968] (1/2) Epoch 25, batch 19150, giga_loss[loss=0.3013, simple_loss=0.364, pruned_loss=0.1193, over 27705.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3539, pruned_loss=0.1029, over 5698800.64 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.08552, over 5750483.72 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3544, pruned_loss=0.1037, over 5687236.53 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:43:48,891 INFO [optim.py:369] (1/2) 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 23:43:58,900 INFO [train.py:968] (1/2) Epoch 25, batch 19200, giga_loss[loss=0.427, simple_loss=0.4418, pruned_loss=0.2061, over 26575.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3515, pruned_loss=0.1019, over 5702645.56 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3321, pruned_loss=0.08553, over 5751071.48 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3519, pruned_loss=0.1028, over 5692295.03 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:44:40,382 INFO [train.py:968] (1/2) Epoch 25, batch 19250, giga_loss[loss=0.2403, simple_loss=0.3238, pruned_loss=0.0784, over 28595.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3521, pruned_loss=0.1021, over 5687129.26 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3326, pruned_loss=0.08572, over 5745069.38 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3525, pruned_loss=0.103, over 5682977.76 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:45:09,670 INFO [optim.py:369] (1/2) 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 23:45:17,237 INFO [train.py:968] (1/2) Epoch 25, batch 19300, giga_loss[loss=0.2322, simple_loss=0.3255, pruned_loss=0.06943, over 28954.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.351, pruned_loss=0.1007, over 5698895.46 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3333, pruned_loss=0.08594, over 5751116.97 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3515, pruned_loss=0.1018, over 5688136.35 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:45:24,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:46:00,838 INFO [train.py:968] (1/2) Epoch 25, batch 19350, giga_loss[loss=0.2352, simple_loss=0.3181, pruned_loss=0.07613, over 28881.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3466, pruned_loss=0.09767, over 5691169.67 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3331, pruned_loss=0.08587, over 5745972.92 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3475, pruned_loss=0.09901, over 5685299.52 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:46:22,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:46:35,037 INFO [optim.py:369] (1/2) 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 23:46:43,074 INFO [train.py:968] (1/2) Epoch 25, batch 19400, giga_loss[loss=0.2394, simple_loss=0.3177, pruned_loss=0.0805, over 28282.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3418, pruned_loss=0.09489, over 5690810.25 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3337, pruned_loss=0.08602, over 5751198.23 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3423, pruned_loss=0.0962, over 5678963.07 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:46:45,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:14,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-12 23:47:22,790 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:47:25,264 INFO [train.py:968] (1/2) Epoch 25, batch 19450, giga_loss[loss=0.239, simple_loss=0.318, pruned_loss=0.08, over 28963.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3367, pruned_loss=0.09222, over 5693597.66 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3339, pruned_loss=0.08594, over 5752440.67 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3371, pruned_loss=0.0935, over 5681675.25 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:47:26,857 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:30,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:37,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:47:59,033 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:03,850 INFO [optim.py:369] (1/2) 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 23:48:15,400 INFO [train.py:968] (1/2) Epoch 25, batch 19500, giga_loss[loss=0.2083, simple_loss=0.282, pruned_loss=0.06729, over 28944.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3321, pruned_loss=0.09017, over 5694794.96 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3338, pruned_loss=0.08589, over 5753093.16 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3325, pruned_loss=0.0913, over 5683808.35 frames. ], batch size: 66, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:48:28,260 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:31,547 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:55,232 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:56,544 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:48:57,972 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 19550, giga_loss[loss=0.2602, simple_loss=0.3396, pruned_loss=0.0904, over 28595.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3333, pruned_loss=0.09044, over 5693516.00 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3341, pruned_loss=0.08592, over 5754461.41 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3334, pruned_loss=0.09135, over 5682863.81 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:49:00,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9795, 1.2986, 1.1065, 0.2040], device='cuda:1'), covar=tensor([0.4497, 0.3361, 0.4993, 0.6940], device='cuda:1'), in_proj_covar=tensor([0.1779, 0.1676, 0.1612, 0.1441], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-12 23:49:21,854 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:49:23,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7736, 1.9305, 1.9884, 1.6728], device='cuda:1'), covar=tensor([0.3544, 0.2902, 0.2679, 0.3276], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1933, 0.1843, 0.2001], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-12 23:49:31,394 INFO [optim.py:369] (1/2) 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 23:49:35,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3210, 3.0285, 1.4546, 1.4746], device='cuda:1'), covar=tensor([0.1104, 0.0348, 0.0970, 0.1458], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0555, 0.0397, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 23:49:43,035 INFO [train.py:968] (1/2) Epoch 25, batch 19600, giga_loss[loss=0.2777, simple_loss=0.3478, pruned_loss=0.1038, over 28813.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3325, pruned_loss=0.08909, over 5699297.06 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.334, pruned_loss=0.08576, over 5756594.84 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3325, pruned_loss=0.09001, over 5688350.12 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:49:46,780 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:49:48,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:49:55,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-12 23:50:11,586 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:50:17,317 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:50:22,287 INFO [train.py:968] (1/2) Epoch 25, batch 19650, giga_loss[loss=0.2012, simple_loss=0.2851, pruned_loss=0.05864, over 28898.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3312, pruned_loss=0.08868, over 5710696.50 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3339, pruned_loss=0.08554, over 5759447.60 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3314, pruned_loss=0.08965, over 5698590.91 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:50:54,434 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 19700, giga_loss[loss=0.2449, simple_loss=0.3199, pruned_loss=0.08493, over 29005.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.329, pruned_loss=0.08774, over 5719262.37 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3342, pruned_loss=0.08557, over 5761712.20 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3288, pruned_loss=0.08851, over 5707165.81 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:51:43,793 INFO [train.py:968] (1/2) Epoch 25, batch 19750, giga_loss[loss=0.2368, simple_loss=0.3071, pruned_loss=0.08325, over 28876.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3266, pruned_loss=0.08681, over 5721558.29 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3345, pruned_loss=0.08558, over 5761893.79 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.326, pruned_loss=0.08743, over 5711345.45 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:52:01,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:52:14,337 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 19800, giga_loss[loss=0.2548, simple_loss=0.3317, pruned_loss=0.08895, over 29013.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3265, pruned_loss=0.08681, over 5725625.23 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3353, pruned_loss=0.0858, over 5767140.71 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3251, pruned_loss=0.08715, over 5711505.73 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:52:39,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:52:40,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113661.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:53:03,289 INFO [train.py:968] (1/2) Epoch 25, batch 19850, giga_loss[loss=0.23, simple_loss=0.3093, pruned_loss=0.07535, over 28965.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3247, pruned_loss=0.08626, over 5727610.81 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3356, pruned_loss=0.08585, over 5769417.43 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.323, pruned_loss=0.08651, over 5713388.03 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:53:06,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2603, 2.8999, 1.4092, 1.4242], device='cuda:1'), covar=tensor([0.1105, 0.0387, 0.0941, 0.1439], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0556, 0.0399, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0026, 0.0030], device='cuda:1')
+2023-03-12 23:53:33,791 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 19900, giga_loss[loss=0.2401, simple_loss=0.3169, pruned_loss=0.08166, over 28678.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3225, pruned_loss=0.08488, over 5731340.83 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3363, pruned_loss=0.08601, over 5771332.52 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3203, pruned_loss=0.08492, over 5717387.06 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:54:23,305 INFO [train.py:968] (1/2) Epoch 25, batch 19950, giga_loss[loss=0.2291, simple_loss=0.301, pruned_loss=0.07866, over 28870.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3208, pruned_loss=0.08421, over 5725165.17 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3368, pruned_loss=0.08617, over 5770656.77 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3182, pruned_loss=0.08406, over 5714147.11 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:54:33,306 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1113804.0, num_to_drop=1, layers_to_drop={1}
+2023-03-12 23:54:37,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1113807.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:54:53,609 INFO [optim.py:369] (1/2) 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 23:54:57,830 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1113836.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:55:00,661 INFO [train.py:968] (1/2) Epoch 25, batch 20000, giga_loss[loss=0.2247, simple_loss=0.3054, pruned_loss=0.07195, over 28872.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3193, pruned_loss=0.08308, over 5732719.00 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3375, pruned_loss=0.08621, over 5771090.86 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3162, pruned_loss=0.08285, over 5722424.82 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:55:15,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:55:27,013 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:55:32,503 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-12 23:55:40,683 INFO [train.py:968] (1/2) Epoch 25, batch 20050, giga_loss[loss=0.252, simple_loss=0.3214, pruned_loss=0.09127, over 28913.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3179, pruned_loss=0.08256, over 5724609.27 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3379, pruned_loss=0.0863, over 5762719.78 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3148, pruned_loss=0.08224, over 5724158.79 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:55:56,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-12 23:56:11,445 INFO [optim.py:369] (1/2) 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 23:56:20,981 INFO [train.py:968] (1/2) Epoch 25, batch 20100, libri_loss[loss=0.2228, simple_loss=0.3134, pruned_loss=0.06608, over 29582.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3184, pruned_loss=0.08281, over 5731033.38 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3383, pruned_loss=0.08632, over 5766384.60 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.315, pruned_loss=0.08245, over 5726427.94 frames. ], batch size: 74, lr: 1.27e-03, grad_scale: 8.0
+2023-03-12 23:56:24,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0608, 1.2118, 3.4141, 3.0376], device='cuda:1'), covar=tensor([0.1803, 0.2825, 0.0510, 0.0931], device='cuda:1'), in_proj_covar=tensor([0.0774, 0.0657, 0.0967, 0.0930], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-12 23:56:57,647 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1113982.0, num_to_drop=1, layers_to_drop={0}
+2023-03-12 23:57:05,301 INFO [train.py:968] (1/2) Epoch 25, batch 20150, giga_loss[loss=0.2983, simple_loss=0.3593, pruned_loss=0.1186, over 28380.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.323, pruned_loss=0.08627, over 5715854.36 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3384, pruned_loss=0.08634, over 5764371.64 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3201, pruned_loss=0.08594, over 5713551.85 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:57:06,023 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1113990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:16,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:18,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:45,387 INFO [optim.py:369] (1/2) 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 23:57:46,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:57:47,660 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 20200, libri_loss[loss=0.262, simple_loss=0.3518, pruned_loss=0.08612, over 25802.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3303, pruned_loss=0.09072, over 5694770.66 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3387, pruned_loss=0.08649, over 5744801.24 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3274, pruned_loss=0.09035, over 5709418.17 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:58:38,013 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:58:43,191 INFO [train.py:968] (1/2) Epoch 25, batch 20250, giga_loss[loss=0.2541, simple_loss=0.3295, pruned_loss=0.08936, over 28877.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.339, pruned_loss=0.09632, over 5679267.24 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3392, pruned_loss=0.08669, over 5745287.78 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3362, pruned_loss=0.09594, over 5689559.49 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:59:07,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3401, 1.7852, 1.5418, 1.4708], device='cuda:1'), covar=tensor([0.0794, 0.0342, 0.0328, 0.0881], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-12 23:59:23,847 INFO [optim.py:369] (1/2) 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 23:59:25,494 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:59:27,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-12 23:59:30,232 INFO [train.py:968] (1/2) Epoch 25, batch 20300, giga_loss[loss=0.3586, simple_loss=0.4161, pruned_loss=0.1506, over 27968.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.344, pruned_loss=0.09843, over 5678188.72 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3394, pruned_loss=0.08687, over 5746101.34 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3417, pruned_loss=0.09801, over 5685119.41 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-12 23:59:52,056 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:02,405 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:04,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:13,519 INFO [train.py:968] (1/2) Epoch 25, batch 20350, giga_loss[loss=0.2604, simple_loss=0.3383, pruned_loss=0.09123, over 28886.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3477, pruned_loss=0.09973, over 5678113.48 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3392, pruned_loss=0.08663, over 5750970.92 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3462, pruned_loss=0.09996, over 5677250.54 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:00:18,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5984, 1.7970, 1.7846, 1.6094], device='cuda:1'), covar=tensor([0.2650, 0.2337, 0.1800, 0.2181], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1929, 0.1844, 0.2002], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 00:00:28,725 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:47,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:00:48,162 INFO [optim.py:369] (1/2) 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 00:00:55,155 INFO [train.py:968] (1/2) Epoch 25, batch 20400, giga_loss[loss=0.3213, simple_loss=0.3913, pruned_loss=0.1256, over 28551.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3523, pruned_loss=0.1024, over 5672512.72 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3394, pruned_loss=0.08675, over 5746767.10 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3513, pruned_loss=0.103, over 5671995.74 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:01:00,560 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:01:04,311 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:01:37,148 INFO [train.py:968] (1/2) Epoch 25, batch 20450, libri_loss[loss=0.2424, simple_loss=0.3253, pruned_loss=0.07973, over 29562.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3513, pruned_loss=0.1012, over 5679045.75 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3392, pruned_loss=0.08676, over 5751280.50 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3511, pruned_loss=0.1022, over 5672219.77 frames. ], batch size: 79, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:01:48,497 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7984, 3.0181, 1.7611, 0.9755], device='cuda:1'), covar=tensor([0.8007, 0.2681, 0.4352, 0.6107], device='cuda:1'), in_proj_covar=tensor([0.1788, 0.1680, 0.1622, 0.1452], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 00:02:13,124 INFO [optim.py:369] (1/2) 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 00:02:19,374 INFO [train.py:968] (1/2) Epoch 25, batch 20500, giga_loss[loss=0.2428, simple_loss=0.324, pruned_loss=0.08082, over 28874.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3469, pruned_loss=0.09792, over 5682633.04 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3389, pruned_loss=0.08663, over 5752882.69 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3471, pruned_loss=0.09894, over 5675212.24 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:02:33,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114357.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:02:48,059 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:03:01,199 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 20550, libri_loss[loss=0.2914, simple_loss=0.368, pruned_loss=0.1074, over 29675.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3453, pruned_loss=0.09603, over 5694053.18 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3392, pruned_loss=0.08682, over 5754556.03 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3452, pruned_loss=0.09674, over 5686099.01 frames. ], batch size: 91, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:03:03,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:03:27,799 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:03:38,527 INFO [optim.py:369] (1/2) 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 00:03:44,075 INFO [train.py:968] (1/2) Epoch 25, batch 20600, giga_loss[loss=0.2872, simple_loss=0.3621, pruned_loss=0.1062, over 28353.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3453, pruned_loss=0.09564, over 5696306.19 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.339, pruned_loss=0.08664, over 5758931.70 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3457, pruned_loss=0.09675, over 5683223.85 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:04:00,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:04:18,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5888, 1.7605, 1.7095, 1.5394], device='cuda:1'), covar=tensor([0.2958, 0.2408, 0.2224, 0.2461], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1931, 0.1845, 0.2002], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 00:04:26,724 INFO [train.py:968] (1/2) Epoch 25, batch 20650, giga_loss[loss=0.2803, simple_loss=0.3526, pruned_loss=0.104, over 28799.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3475, pruned_loss=0.09686, over 5693771.08 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3392, pruned_loss=0.08667, over 5759154.60 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3477, pruned_loss=0.0978, over 5682682.35 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:04:37,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114500.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:04:39,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114503.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:05:03,168 INFO [optim.py:369] (1/2) 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 00:05:03,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114532.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:05:10,565 INFO [train.py:968] (1/2) Epoch 25, batch 20700, giga_loss[loss=0.3697, simple_loss=0.4066, pruned_loss=0.1664, over 26560.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3502, pruned_loss=0.09903, over 5705641.91 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3397, pruned_loss=0.08688, over 5761897.95 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.35, pruned_loss=0.0998, over 5693271.00 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:05:55,750 INFO [train.py:968] (1/2) Epoch 25, batch 20750, giga_loss[loss=0.3824, simple_loss=0.4188, pruned_loss=0.173, over 26581.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3521, pruned_loss=0.1007, over 5693644.66 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3398, pruned_loss=0.08683, over 5764621.88 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3521, pruned_loss=0.1016, over 5680082.00 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:05:56,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:05,008 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:08,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:08,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:13,748 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-13 00:06:24,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:28,795 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:06:31,329 INFO [optim.py:369] (1/2) 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 00:06:32,953 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 20800, giga_loss[loss=0.2742, simple_loss=0.3514, pruned_loss=0.0985, over 28924.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3533, pruned_loss=0.1019, over 5698440.39 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3397, pruned_loss=0.08669, over 5769072.19 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3539, pruned_loss=0.1033, over 5681062.59 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:07:04,807 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1114675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:07:16,511 INFO [train.py:968] (1/2) Epoch 25, batch 20850, giga_loss[loss=0.3416, simple_loss=0.4168, pruned_loss=0.1332, over 29063.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3534, pruned_loss=0.1025, over 5703890.57 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3393, pruned_loss=0.08648, over 5771749.54 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.104, over 5686569.85 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:07:22,775 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5704, 1.4857, 1.7556, 1.4031], device='cuda:1'), covar=tensor([0.1640, 0.2528, 0.1363, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0709, 0.0971, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 00:07:48,781 INFO [optim.py:369] (1/2) 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 00:07:55,649 INFO [train.py:968] (1/2) Epoch 25, batch 20900, giga_loss[loss=0.2825, simple_loss=0.3521, pruned_loss=0.1065, over 28848.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3531, pruned_loss=0.1017, over 5711318.55 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3394, pruned_loss=0.08651, over 5772943.37 frames. ], giga_tot_loss[loss=0.2805, simple_loss=0.3543, pruned_loss=0.1034, over 5694750.55 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:08:01,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:02,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:06,407 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:21,112 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:23,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:30,496 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:08:38,047 INFO [train.py:968] (1/2) Epoch 25, batch 20950, giga_loss[loss=0.2967, simple_loss=0.3686, pruned_loss=0.1124, over 28830.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3523, pruned_loss=0.09979, over 5708537.35 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3395, pruned_loss=0.08653, over 5773619.65 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3531, pruned_loss=0.1011, over 5694601.49 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:08:44,981 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:09:12,167 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 21000, giga_loss[loss=0.2712, simple_loss=0.3543, pruned_loss=0.09403, over 28323.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3529, pruned_loss=0.09988, over 5695540.80 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3399, pruned_loss=0.08678, over 5766874.40 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3535, pruned_loss=0.1009, over 5688935.46 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:09:19,041 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 00:09:23,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4543, 1.8098, 1.4332, 1.4438], device='cuda:1'), covar=tensor([0.3288, 0.3214, 0.3707, 0.2745], device='cuda:1'), in_proj_covar=tensor([0.1555, 0.1123, 0.1374, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 00:09:25,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4804, 1.7620, 1.3084, 1.3491], device='cuda:1'), covar=tensor([0.1027, 0.0468, 0.1032, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0444, 0.0521, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 00:09:26,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2344, 1.5535, 1.5054, 1.0736], device='cuda:1'), covar=tensor([0.1641, 0.2880, 0.1539, 0.2015], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0711, 0.0972, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 00:09:28,061 INFO [train.py:1012] (1/2) Epoch 25, validation: loss=0.2071, simple_loss=0.3152, pruned_loss=0.04952, over 944034.00 frames. 
+2023-03-13 00:09:28,061 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 00:09:35,948 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9832, 1.1504, 1.0529, 0.9732], device='cuda:1'), covar=tensor([0.1994, 0.1864, 0.1422, 0.1749], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1941, 0.1850, 0.2008], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 00:09:59,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6137, 3.6590, 1.7327, 1.7643], device='cuda:1'), covar=tensor([0.0998, 0.0273, 0.0892, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0555, 0.0396, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-13 00:10:05,358 INFO [train.py:968] (1/2) Epoch 25, batch 21050, giga_loss[loss=0.2853, simple_loss=0.3483, pruned_loss=0.1111, over 27618.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3518, pruned_loss=0.0993, over 5697020.35 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3404, pruned_loss=0.08718, over 5756354.92 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3521, pruned_loss=0.1001, over 5699270.62 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:10:05,626 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1114890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:10:07,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1114893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:10:28,191 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1114922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:10:37,196 INFO [optim.py:369] (1/2) 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 00:10:41,836 INFO [train.py:968] (1/2) Epoch 25, batch 21100, giga_loss[loss=0.2859, simple_loss=0.3543, pruned_loss=0.1088, over 28876.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3494, pruned_loss=0.09816, over 5706299.16 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3407, pruned_loss=0.08742, over 5761597.30 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3496, pruned_loss=0.09897, over 5701657.54 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:10:49,691 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3798, 1.6893, 1.5743, 1.4699], device='cuda:1'), covar=tensor([0.0829, 0.0328, 0.0318, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0064, 0.0112], device='cuda:1')
+2023-03-13 00:11:02,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1114966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:11:20,118 INFO [train.py:968] (1/2) Epoch 25, batch 21150, giga_loss[loss=0.2715, simple_loss=0.3411, pruned_loss=0.101, over 28967.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3469, pruned_loss=0.09697, over 5714201.38 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3406, pruned_loss=0.0874, over 5763176.17 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3473, pruned_loss=0.09771, over 5708827.79 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:11:31,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1115003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:11:31,319 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5549, 1.8032, 1.6671, 1.6329], device='cuda:1'), covar=tensor([0.2177, 0.2173, 0.2501, 0.2172], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0747, 0.0720, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 00:11:57,459 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 21200, giga_loss[loss=0.34, simple_loss=0.4003, pruned_loss=0.1398, over 28827.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.347, pruned_loss=0.0977, over 5710177.56 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3405, pruned_loss=0.08739, over 5765316.02 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3474, pruned_loss=0.09842, over 5703300.64 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:12:10,282 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1115050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:12:41,339 INFO [train.py:968] (1/2) Epoch 25, batch 21250, giga_loss[loss=0.2406, simple_loss=0.3301, pruned_loss=0.0755, over 29022.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3467, pruned_loss=0.09717, over 5718430.70 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3404, pruned_loss=0.08749, over 5766604.04 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09789, over 5710582.66 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:12:56,066 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:12:58,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:02,405 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-13 00:13:15,510 INFO [optim.py:369] (1/2) 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 00:13:21,362 INFO [train.py:968] (1/2) Epoch 25, batch 21300, giga_loss[loss=0.2883, simple_loss=0.368, pruned_loss=0.1043, over 28626.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3464, pruned_loss=0.0969, over 5708746.75 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3403, pruned_loss=0.08749, over 5768592.12 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.347, pruned_loss=0.09764, over 5699999.42 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:13:22,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:26,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:28,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:50,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:13:54,437 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-13 00:13:59,657 INFO [train.py:968] (1/2) Epoch 25, batch 21350, giga_loss[loss=0.2582, simple_loss=0.3368, pruned_loss=0.08983, over 29076.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3451, pruned_loss=0.09523, over 5721759.57 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3412, pruned_loss=0.08821, over 5772631.21 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.345, pruned_loss=0.0954, over 5709657.56 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:14:02,791 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:14:04,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:14:12,262 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115206.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:14:28,373 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:14:32,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0595, 2.2255, 2.2708, 1.7653], device='cuda:1'), covar=tensor([0.1652, 0.2561, 0.1352, 0.1803], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0711, 0.0972, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 00:14:32,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-13 00:14:34,629 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 21400, giga_loss[loss=0.2659, simple_loss=0.3443, pruned_loss=0.09378, over 28929.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3449, pruned_loss=0.0951, over 5727046.92 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3413, pruned_loss=0.08847, over 5772760.56 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3447, pruned_loss=0.0951, over 5716743.94 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:15:12,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0545, 3.1903, 2.1542, 1.2425], device='cuda:1'), covar=tensor([0.8457, 0.2759, 0.4098, 0.6516], device='cuda:1'), in_proj_covar=tensor([0.1789, 0.1670, 0.1619, 0.1449], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 00:15:18,815 INFO [train.py:968] (1/2) Epoch 25, batch 21450, giga_loss[loss=0.2497, simple_loss=0.3209, pruned_loss=0.08923, over 28855.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3415, pruned_loss=0.09354, over 5729105.50 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3413, pruned_loss=0.08859, over 5773885.10 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3415, pruned_loss=0.09354, over 5719083.88 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:15:49,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1715, 1.4417, 5.2603, 3.7516], device='cuda:1'), covar=tensor([0.1502, 0.2913, 0.0352, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0772, 0.0653, 0.0964, 0.0931], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 00:15:51,848 INFO [optim.py:369] (1/2) 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 00:15:53,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-13 00:15:57,421 INFO [train.py:968] (1/2) Epoch 25, batch 21500, giga_loss[loss=0.2661, simple_loss=0.3377, pruned_loss=0.0973, over 28866.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3389, pruned_loss=0.09227, over 5726121.32 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3412, pruned_loss=0.08888, over 5777189.00 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3388, pruned_loss=0.09216, over 5712905.89 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:16:08,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5732, 1.7325, 1.8301, 1.5963], device='cuda:1'), covar=tensor([0.2111, 0.2212, 0.2251, 0.2096], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0747, 0.0720, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 00:16:36,956 INFO [train.py:968] (1/2) Epoch 25, batch 21550, libri_loss[loss=0.3116, simple_loss=0.3854, pruned_loss=0.119, over 29202.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3388, pruned_loss=0.09251, over 5725169.51 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3415, pruned_loss=0.08918, over 5772472.53 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3384, pruned_loss=0.09222, over 5717405.72 frames. ], batch size: 97, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:17:05,864 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.06 vs. limit=5.0
+2023-03-13 00:17:08,916 INFO [optim.py:369] (1/2) 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 00:17:13,795 INFO [train.py:968] (1/2) Epoch 25, batch 21600, libri_loss[loss=0.3014, simple_loss=0.3765, pruned_loss=0.1132, over 29530.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3386, pruned_loss=0.09276, over 5726922.19 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3417, pruned_loss=0.08944, over 5774563.93 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.338, pruned_loss=0.09243, over 5716907.06 frames. ], batch size: 83, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:17:53,128 INFO [train.py:968] (1/2) Epoch 25, batch 21650, giga_loss[loss=0.2392, simple_loss=0.3106, pruned_loss=0.08388, over 28809.00 frames. ], tot_loss[loss=0.262, simple_loss=0.338, pruned_loss=0.093, over 5726594.72 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3426, pruned_loss=0.09015, over 5777956.62 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3367, pruned_loss=0.09219, over 5714645.20 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:18:29,069 INFO [optim.py:369] (1/2) 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 00:18:31,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4818, 4.3205, 4.1135, 2.0191], device='cuda:1'), covar=tensor([0.0594, 0.0758, 0.0788, 0.2036], device='cuda:1'), in_proj_covar=tensor([0.1247, 0.1153, 0.0971, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 00:18:32,317 INFO [train.py:968] (1/2) Epoch 25, batch 21700, giga_loss[loss=0.2541, simple_loss=0.3209, pruned_loss=0.09367, over 28671.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3379, pruned_loss=0.09385, over 5725932.02 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3431, pruned_loss=0.091, over 5780979.80 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3361, pruned_loss=0.09252, over 5710625.20 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:19:03,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1115581.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:19:09,838 INFO [train.py:968] (1/2) Epoch 25, batch 21750, giga_loss[loss=0.3324, simple_loss=0.3764, pruned_loss=0.1442, over 23707.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3356, pruned_loss=0.09294, over 5714929.81 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3435, pruned_loss=0.09133, over 5773051.26 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3336, pruned_loss=0.09161, over 5709094.45 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:19:23,016 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-13 00:19:32,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4670, 1.6729, 1.7469, 1.2988], device='cuda:1'), covar=tensor([0.1818, 0.2562, 0.1494, 0.1684], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0710, 0.0972, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 00:19:47,954 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 21800, libri_loss[loss=0.3001, simple_loss=0.3782, pruned_loss=0.111, over 28627.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3333, pruned_loss=0.09196, over 5714060.63 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.344, pruned_loss=0.09164, over 5773735.59 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3311, pruned_loss=0.09064, over 5707888.34 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:20:02,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3306, 1.6092, 1.5609, 1.4528], device='cuda:1'), covar=tensor([0.1835, 0.1735, 0.2201, 0.1853], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0746, 0.0720, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 00:20:30,537 INFO [train.py:968] (1/2) Epoch 25, batch 21850, giga_loss[loss=0.2413, simple_loss=0.3189, pruned_loss=0.08179, over 29075.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3326, pruned_loss=0.09182, over 5713083.29 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3437, pruned_loss=0.09179, over 5776186.52 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3308, pruned_loss=0.09065, over 5705044.44 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:20:59,044 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1115724.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:21:03,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1115727.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:21:05,374 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:21:09,908 INFO [optim.py:369] (1/2) 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 00:21:14,021 INFO [train.py:968] (1/2) Epoch 25, batch 21900, giga_loss[loss=0.2896, simple_loss=0.359, pruned_loss=0.1101, over 28890.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3345, pruned_loss=0.09277, over 5714660.82 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3439, pruned_loss=0.0919, over 5778029.12 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3329, pruned_loss=0.09176, over 5705904.45 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:21:27,624 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1115756.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:21:38,562 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:21:51,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2118, 2.1300, 1.9817, 2.0686], device='cuda:1'), covar=tensor([0.2142, 0.2896, 0.2505, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0746, 0.0719, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 00:21:55,078 INFO [train.py:968] (1/2) Epoch 25, batch 21950, giga_loss[loss=0.2937, simple_loss=0.3678, pruned_loss=0.1098, over 29028.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3374, pruned_loss=0.09385, over 5715095.33 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3446, pruned_loss=0.09248, over 5779301.34 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3353, pruned_loss=0.09257, over 5705735.00 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:22:00,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1115794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:22:32,381 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 22000, giga_loss[loss=0.3503, simple_loss=0.4147, pruned_loss=0.1429, over 27537.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3389, pruned_loss=0.0934, over 5712827.24 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3447, pruned_loss=0.09266, over 5780357.19 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3371, pruned_loss=0.09226, over 5703625.74 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:23:09,547 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-13 00:23:19,089 INFO [train.py:968] (1/2) Epoch 25, batch 22050, giga_loss[loss=0.2351, simple_loss=0.316, pruned_loss=0.07713, over 28868.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.339, pruned_loss=0.09259, over 5701646.38 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3453, pruned_loss=0.09314, over 5772370.21 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3369, pruned_loss=0.09124, over 5700167.76 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:24:00,291 INFO [optim.py:369] (1/2) 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 00:24:02,782 INFO [train.py:968] (1/2) Epoch 25, batch 22100, giga_loss[loss=0.2677, simple_loss=0.3379, pruned_loss=0.09876, over 23633.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3381, pruned_loss=0.09195, over 5696193.41 frames. ], libri_tot_loss[loss=0.2663, simple_loss=0.3456, pruned_loss=0.09345, over 5775735.77 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3359, pruned_loss=0.09056, over 5690222.58 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:24:25,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5460, 1.7692, 1.6252, 1.5184], device='cuda:1'), covar=tensor([0.1956, 0.2306, 0.2412, 0.2443], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0746, 0.0719, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 00:24:26,095 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-13 00:24:41,181 INFO [train.py:968] (1/2) Epoch 25, batch 22150, libri_loss[loss=0.296, simple_loss=0.3722, pruned_loss=0.1099, over 29148.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3389, pruned_loss=0.09276, over 5712082.84 frames. ], libri_tot_loss[loss=0.2667, simple_loss=0.3458, pruned_loss=0.09382, over 5781557.45 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3366, pruned_loss=0.09121, over 5698548.58 frames. ], batch size: 101, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:24:57,708 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:25:04,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:25:08,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5910, 1.1835, 4.3804, 3.6369], device='cuda:1'), covar=tensor([0.1499, 0.2771, 0.0412, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0658, 0.0971, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 00:25:18,796 INFO [optim.py:369] (1/2) 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 00:25:22,877 INFO [train.py:968] (1/2) Epoch 25, batch 22200, giga_loss[loss=0.261, simple_loss=0.3347, pruned_loss=0.09363, over 28940.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.341, pruned_loss=0.09467, over 5712699.53 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3464, pruned_loss=0.0944, over 5785096.87 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3385, pruned_loss=0.09291, over 5696830.48 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:26:03,502 INFO [train.py:968] (1/2) Epoch 25, batch 22250, giga_loss[loss=0.2484, simple_loss=0.3244, pruned_loss=0.08615, over 28854.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.343, pruned_loss=0.09578, over 5713141.08 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.347, pruned_loss=0.09494, over 5784438.33 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09394, over 5700167.41 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:26:17,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:26:40,679 INFO [optim.py:369] (1/2) 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 00:26:43,996 INFO [train.py:968] (1/2) Epoch 25, batch 22300, giga_loss[loss=0.2688, simple_loss=0.3519, pruned_loss=0.09281, over 28947.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3461, pruned_loss=0.09741, over 5713315.76 frames. ], libri_tot_loss[loss=0.2691, simple_loss=0.3475, pruned_loss=0.09538, over 5786734.32 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3436, pruned_loss=0.09558, over 5699777.66 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:26:47,998 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:27:09,235 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:27:12,002 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:27:23,897 INFO [train.py:968] (1/2) Epoch 25, batch 22350, giga_loss[loss=0.3086, simple_loss=0.3786, pruned_loss=0.1193, over 27590.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.349, pruned_loss=0.09887, over 5715311.35 frames. ], libri_tot_loss[loss=0.2698, simple_loss=0.3479, pruned_loss=0.09586, over 5784500.87 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3466, pruned_loss=0.09704, over 5703538.14 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:27:33,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1699, 1.4179, 1.3145, 1.0728], device='cuda:1'), covar=tensor([0.2913, 0.2715, 0.1843, 0.2603], device='cuda:1'), in_proj_covar=tensor([0.2018, 0.1959, 0.1871, 0.2024], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 00:27:55,564 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3924, 1.5518, 1.6162, 1.2183], device='cuda:1'), covar=tensor([0.1855, 0.2547, 0.1548, 0.1756], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0709, 0.0970, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 00:28:01,354 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 22400, giga_loss[loss=0.3391, simple_loss=0.3951, pruned_loss=0.1415, over 28741.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3502, pruned_loss=0.09947, over 5718872.39 frames. ], libri_tot_loss[loss=0.2699, simple_loss=0.3479, pruned_loss=0.09596, over 5785528.53 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3483, pruned_loss=0.09798, over 5707978.19 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:28:11,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:13,582 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:39,521 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:45,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:46,609 INFO [train.py:968] (1/2) Epoch 25, batch 22450, giga_loss[loss=0.2849, simple_loss=0.3614, pruned_loss=0.1041, over 28864.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 5718996.92 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3482, pruned_loss=0.09636, over 5786950.48 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3493, pruned_loss=0.09885, over 5708161.03 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:28:47,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:28:55,747 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-13 00:29:05,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:29:07,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:29:08,503 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116316.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:29:11,103 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:29:28,237 INFO [train.py:968] (1/2) Epoch 25, batch 22500, giga_loss[loss=0.3752, simple_loss=0.4157, pruned_loss=0.1673, over 26718.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3511, pruned_loss=0.1007, over 5713320.61 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.3482, pruned_loss=0.09653, over 5788542.40 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3498, pruned_loss=0.0994, over 5702186.24 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:29:31,898 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:29:43,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2089, 1.1311, 3.7163, 3.2010], device='cuda:1'), covar=tensor([0.1704, 0.2997, 0.0436, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0778, 0.0658, 0.0973, 0.0940], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 00:30:00,875 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:30:05,520 INFO [train.py:968] (1/2) Epoch 25, batch 22550, giga_loss[loss=0.2649, simple_loss=0.3432, pruned_loss=0.09332, over 28792.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3486, pruned_loss=0.09898, over 5723743.23 frames. ], libri_tot_loss[loss=0.2714, simple_loss=0.3488, pruned_loss=0.09702, over 5790816.02 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.347, pruned_loss=0.09759, over 5710783.25 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:30:10,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:30:33,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4072, 4.2363, 4.0340, 1.8152], device='cuda:1'), covar=tensor([0.0592, 0.0742, 0.0758, 0.2145], device='cuda:1'), in_proj_covar=tensor([0.1249, 0.1155, 0.0977, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 00:30:36,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3690, 2.1646, 1.7268, 1.4863], device='cuda:1'), covar=tensor([0.0780, 0.0252, 0.0318, 0.0942], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0119, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 00:30:47,091 INFO [optim.py:369] (1/2) 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 00:30:50,488 INFO [train.py:968] (1/2) Epoch 25, batch 22600, giga_loss[loss=0.2363, simple_loss=0.3187, pruned_loss=0.07698, over 28958.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3459, pruned_loss=0.09769, over 5713258.87 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3496, pruned_loss=0.09777, over 5787983.59 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3437, pruned_loss=0.09586, over 5703725.92 frames. ], batch size: 164, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:31:28,147 INFO [train.py:968] (1/2) Epoch 25, batch 22650, giga_loss[loss=0.2133, simple_loss=0.2964, pruned_loss=0.06512, over 28668.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3436, pruned_loss=0.0965, over 5707250.67 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3504, pruned_loss=0.09837, over 5779045.61 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.341, pruned_loss=0.09445, over 5705422.01 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:31:57,575 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:31:59,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:05,392 INFO [optim.py:369] (1/2) 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 00:32:06,410 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:08,558 INFO [train.py:968] (1/2) Epoch 25, batch 22700, giga_loss[loss=0.2701, simple_loss=0.3607, pruned_loss=0.0897, over 28760.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3442, pruned_loss=0.09588, over 5703756.77 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3505, pruned_loss=0.09874, over 5778270.49 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3418, pruned_loss=0.09384, over 5700790.68 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:32:09,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:15,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:17,186 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116550.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:32:22,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:32:33,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 22750, libri_loss[loss=0.3504, simple_loss=0.4037, pruned_loss=0.1486, over 29656.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3458, pruned_loss=0.09562, over 5696179.01 frames. ], libri_tot_loss[loss=0.2746, simple_loss=0.3509, pruned_loss=0.09918, over 5767983.19 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3435, pruned_loss=0.0935, over 5701843.23 frames. ], batch size: 91, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:33:03,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3338, 1.5794, 1.3961, 1.2322], device='cuda:1'), covar=tensor([0.3319, 0.2672, 0.2204, 0.2919], device='cuda:1'), in_proj_covar=tensor([0.2022, 0.1968, 0.1875, 0.2026], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 00:33:28,463 INFO [optim.py:369] (1/2) 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 00:33:28,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7772, 1.8698, 1.4675, 1.4445], device='cuda:1'), covar=tensor([0.1049, 0.0736, 0.1075, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0445, 0.0519, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 00:33:30,471 INFO [train.py:968] (1/2) Epoch 25, batch 22800, libri_loss[loss=0.3235, simple_loss=0.3919, pruned_loss=0.1276, over 29119.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3451, pruned_loss=0.09574, over 5693091.90 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3512, pruned_loss=0.09948, over 5767932.96 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3428, pruned_loss=0.09369, over 5695938.70 frames. ], batch size: 101, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:34:12,325 INFO [train.py:968] (1/2) Epoch 25, batch 22850, giga_loss[loss=0.236, simple_loss=0.3088, pruned_loss=0.08157, over 28781.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3437, pruned_loss=0.09632, over 5689939.57 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3512, pruned_loss=0.09955, over 5761507.01 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3417, pruned_loss=0.09455, over 5696195.15 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:34:12,611 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:13,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116691.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:34:15,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1116693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:32,590 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:38,094 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:34:50,152 INFO [optim.py:369] (1/2) 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 00:34:52,189 INFO [train.py:968] (1/2) Epoch 25, batch 22900, giga_loss[loss=0.339, simple_loss=0.3906, pruned_loss=0.1438, over 28372.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3424, pruned_loss=0.09681, over 5699407.76 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3514, pruned_loss=0.09987, over 5764078.25 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3404, pruned_loss=0.09502, over 5700396.42 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:35:30,073 INFO [train.py:968] (1/2) Epoch 25, batch 22950, libri_loss[loss=0.3053, simple_loss=0.3691, pruned_loss=0.1208, over 25927.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3403, pruned_loss=0.09614, over 5710567.72 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3517, pruned_loss=0.1002, over 5766086.60 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.338, pruned_loss=0.09419, over 5707411.77 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:35:43,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3730, 1.6402, 1.3881, 1.5399], device='cuda:1'), covar=tensor([0.0718, 0.0360, 0.0345, 0.0848], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0072, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 00:36:06,340 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1116834.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:36:08,068 INFO [optim.py:369] (1/2) 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 00:36:09,272 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 23000, giga_loss[loss=0.3023, simple_loss=0.3695, pruned_loss=0.1176, over 28595.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3405, pruned_loss=0.09716, over 5712764.22 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3519, pruned_loss=0.1006, over 5769394.63 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3381, pruned_loss=0.09517, over 5705730.45 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:36:11,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 00:36:30,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1116866.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:36:49,432 INFO [train.py:968] (1/2) Epoch 25, batch 23050, giga_loss[loss=0.2502, simple_loss=0.3262, pruned_loss=0.08711, over 28862.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3379, pruned_loss=0.09553, over 5718926.38 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3523, pruned_loss=0.1009, over 5772167.00 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3354, pruned_loss=0.09357, over 5709768.85 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:37:18,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1116925.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:37:29,093 INFO [optim.py:369] (1/2) 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 00:37:30,435 INFO [train.py:968] (1/2) Epoch 25, batch 23100, giga_loss[loss=0.2093, simple_loss=0.2822, pruned_loss=0.06818, over 28345.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3334, pruned_loss=0.09338, over 5712296.92 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3525, pruned_loss=0.1011, over 5764714.74 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3311, pruned_loss=0.09158, over 5710447.71 frames. ], batch size: 65, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:37:50,979 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1116981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:38:11,052 INFO [train.py:968] (1/2) Epoch 25, batch 23150, giga_loss[loss=0.2172, simple_loss=0.3004, pruned_loss=0.06698, over 28943.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3302, pruned_loss=0.09183, over 5702860.74 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3531, pruned_loss=0.1016, over 5757955.83 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3274, pruned_loss=0.08981, over 5705799.99 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:38:43,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6995, 1.9337, 1.6112, 1.6493], device='cuda:1'), covar=tensor([0.2628, 0.2732, 0.3116, 0.2504], device='cuda:1'), in_proj_covar=tensor([0.1559, 0.1124, 0.1375, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 00:38:46,338 INFO [optim.py:369] (1/2) 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 00:38:48,783 INFO [train.py:968] (1/2) Epoch 25, batch 23200, giga_loss[loss=0.2439, simple_loss=0.3304, pruned_loss=0.0787, over 28960.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3298, pruned_loss=0.09102, over 5713702.12 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3531, pruned_loss=0.1017, over 5762397.76 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3269, pruned_loss=0.08905, over 5710721.96 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:39:11,337 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117068.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:39:13,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117071.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:39:18,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7191, 1.8955, 1.8800, 1.6529], device='cuda:1'), covar=tensor([0.2067, 0.1981, 0.1592, 0.1842], device='cuda:1'), in_proj_covar=tensor([0.2025, 0.1966, 0.1879, 0.2030], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 00:39:27,841 INFO [train.py:968] (1/2) Epoch 25, batch 23250, giga_loss[loss=0.2915, simple_loss=0.3554, pruned_loss=0.1138, over 24051.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3319, pruned_loss=0.09195, over 5714911.47 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3529, pruned_loss=0.1019, over 5767818.45 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.329, pruned_loss=0.0898, over 5706236.08 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:39:28,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:39:36,184 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117100.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:39:49,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:40:03,998 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 00:40:08,617 INFO [optim.py:369] (1/2) 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 00:40:09,272 INFO [train.py:968] (1/2) Epoch 25, batch 23300, giga_loss[loss=0.2519, simple_loss=0.3278, pruned_loss=0.08799, over 28584.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3358, pruned_loss=0.09375, over 5711739.02 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3531, pruned_loss=0.1024, over 5766747.83 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3329, pruned_loss=0.09145, over 5704916.06 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:40:16,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:40:47,505 INFO [train.py:968] (1/2) Epoch 25, batch 23350, giga_loss[loss=0.2838, simple_loss=0.3642, pruned_loss=0.1017, over 28489.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3396, pruned_loss=0.09537, over 5707513.82 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3528, pruned_loss=0.1023, over 5761689.93 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3371, pruned_loss=0.09333, over 5704192.80 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:41:26,044 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:41:28,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:41:29,854 INFO [optim.py:369] (1/2) 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 00:41:30,487 INFO [train.py:968] (1/2) Epoch 25, batch 23400, giga_loss[loss=0.2968, simple_loss=0.3717, pruned_loss=0.1109, over 28826.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3432, pruned_loss=0.09724, over 5704338.64 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3529, pruned_loss=0.1025, over 5764132.28 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3409, pruned_loss=0.09536, over 5698748.22 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:41:39,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3984, 3.1540, 1.5478, 1.5223], device='cuda:1'), covar=tensor([0.0963, 0.0332, 0.0965, 0.1294], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0558, 0.0397, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 00:41:53,500 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:42:01,985 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117278.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:42:08,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2005, 0.7983, 0.9368, 1.3857], device='cuda:1'), covar=tensor([0.0764, 0.0389, 0.0367, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 00:42:09,739 INFO [train.py:968] (1/2) Epoch 25, batch 23450, giga_loss[loss=0.2664, simple_loss=0.3339, pruned_loss=0.09944, over 28503.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3438, pruned_loss=0.09744, over 5705678.71 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3527, pruned_loss=0.1026, over 5766040.63 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3419, pruned_loss=0.09569, over 5697766.58 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:42:23,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3058, 1.6209, 1.3086, 0.9456], device='cuda:1'), covar=tensor([0.2276, 0.2315, 0.2611, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.1560, 0.1125, 0.1375, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 00:42:52,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4459, 1.7518, 1.4010, 1.5081], device='cuda:1'), covar=tensor([0.2287, 0.2264, 0.2521, 0.2060], device='cuda:1'), in_proj_covar=tensor([0.1559, 0.1124, 0.1374, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 00:42:57,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117338.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:42:58,402 INFO [optim.py:369] (1/2) 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 00:42:58,415 INFO [train.py:968] (1/2) Epoch 25, batch 23500, libri_loss[loss=0.2577, simple_loss=0.3379, pruned_loss=0.08875, over 29515.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3508, pruned_loss=0.1036, over 5690228.87 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3525, pruned_loss=0.1025, over 5757590.45 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3493, pruned_loss=0.1022, over 5689902.67 frames. ], batch size: 81, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 00:43:13,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:43:49,174 INFO [train.py:968] (1/2) Epoch 25, batch 23550, giga_loss[loss=0.369, simple_loss=0.4073, pruned_loss=0.1653, over 26712.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.357, pruned_loss=0.1083, over 5688652.04 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3528, pruned_loss=0.1027, over 5759071.16 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3556, pruned_loss=0.1072, over 5686159.93 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 00:43:55,559 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1117397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:44:22,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6741, 1.7085, 1.3009, 1.3794], device='cuda:1'), covar=tensor([0.0858, 0.0568, 0.0948, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0446, 0.0519, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 00:44:32,582 INFO [optim.py:369] (1/2) 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 00:44:32,596 INFO [train.py:968] (1/2) Epoch 25, batch 23600, giga_loss[loss=0.2914, simple_loss=0.368, pruned_loss=0.1074, over 29018.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3633, pruned_loss=0.1131, over 5684971.84 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.353, pruned_loss=0.1031, over 5759154.68 frames. ], giga_tot_loss[loss=0.2932, simple_loss=0.3623, pruned_loss=0.112, over 5679903.15 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:45:12,444 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117481.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:45:15,410 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117484.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:45:19,997 INFO [train.py:968] (1/2) Epoch 25, batch 23650, giga_loss[loss=0.3708, simple_loss=0.4222, pruned_loss=0.1597, over 28538.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3714, pruned_loss=0.1205, over 5681466.52 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3531, pruned_loss=0.1033, over 5761092.14 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3707, pruned_loss=0.1197, over 5674623.90 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:45:20,253 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:45:31,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:45:34,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:45:46,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117513.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 00:45:55,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3091, 1.7193, 1.4496, 1.4882], device='cuda:1'), covar=tensor([0.0718, 0.0355, 0.0318, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0190, 0.0120, 0.0119, 0.0225], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 00:45:56,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:46:03,576 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 25, batch 23700, giga_loss[loss=0.331, simple_loss=0.3896, pruned_loss=0.1362, over 28967.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3787, pruned_loss=0.1267, over 5668557.94 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3533, pruned_loss=0.1035, over 5762669.08 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3782, pruned_loss=0.1261, over 5660987.06 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:46:59,077 INFO [train.py:968] (1/2) Epoch 25, batch 23750, giga_loss[loss=0.2932, simple_loss=0.3643, pruned_loss=0.1111, over 28924.00 frames. ], tot_loss[loss=0.3188, simple_loss=0.3811, pruned_loss=0.1283, over 5659460.24 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3533, pruned_loss=0.1036, over 5747322.26 frames. ], giga_tot_loss[loss=0.3191, simple_loss=0.3815, pruned_loss=0.1283, over 5664691.25 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:47:17,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4791, 1.6592, 1.6432, 1.5927], device='cuda:1'), covar=tensor([0.1481, 0.1449, 0.1606, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0752, 0.0722, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 00:47:38,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:47:42,027 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:47:46,259 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 23800, giga_loss[loss=0.4807, simple_loss=0.4833, pruned_loss=0.239, over 27600.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3833, pruned_loss=0.1309, over 5659814.41 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3536, pruned_loss=0.1039, over 5751855.55 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3841, pruned_loss=0.1313, over 5657618.57 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:47:59,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:48:10,979 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:48:13,159 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:48:18,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:48:37,938 INFO [train.py:968] (1/2) Epoch 25, batch 23850, giga_loss[loss=0.3196, simple_loss=0.3713, pruned_loss=0.134, over 28664.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.3851, pruned_loss=0.1335, over 5643083.12 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3535, pruned_loss=0.104, over 5753455.00 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.386, pruned_loss=0.1341, over 5639278.97 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:48:47,071 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:49:27,560 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 23900, libri_loss[loss=0.3031, simple_loss=0.369, pruned_loss=0.1186, over 29548.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3895, pruned_loss=0.1376, over 5648053.70 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3541, pruned_loss=0.1044, over 5756951.61 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3904, pruned_loss=0.1382, over 5639916.12 frames. ], batch size: 83, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:50:03,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1117772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:50:22,026 INFO [train.py:968] (1/2) Epoch 25, batch 23950, giga_loss[loss=0.3249, simple_loss=0.3867, pruned_loss=0.1315, over 28885.00 frames. ], tot_loss[loss=0.3358, simple_loss=0.3914, pruned_loss=0.1401, over 5612696.22 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.354, pruned_loss=0.1045, over 5752048.03 frames. ], giga_tot_loss[loss=0.3383, simple_loss=0.3933, pruned_loss=0.1416, over 5607775.66 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:50:24,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 00:50:33,119 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117796.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:50:36,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117799.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:50:38,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4587, 3.4224, 1.6007, 1.6157], device='cuda:1'), covar=tensor([0.0992, 0.0302, 0.0857, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0561, 0.0399, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 00:51:01,683 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117828.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 00:51:14,969 INFO [optim.py:369] (1/2) 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 00:51:14,981 INFO [train.py:968] (1/2) Epoch 25, batch 24000, giga_loss[loss=0.3606, simple_loss=0.4051, pruned_loss=0.158, over 29009.00 frames. ], tot_loss[loss=0.3332, simple_loss=0.3886, pruned_loss=0.1389, over 5619586.39 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3534, pruned_loss=0.1043, over 5754742.05 frames. ], giga_tot_loss[loss=0.3372, simple_loss=0.3918, pruned_loss=0.1413, over 5609224.00 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:51:14,982 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 00:51:23,955 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 00:51:50,557 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.55 vs. limit=5.0
+2023-03-13 00:52:04,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3706, 1.3056, 3.9714, 3.4734], device='cuda:1'), covar=tensor([0.1615, 0.2791, 0.0447, 0.1469], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0660, 0.0977, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 00:52:06,954 INFO [train.py:968] (1/2) Epoch 25, batch 24050, giga_loss[loss=0.3714, simple_loss=0.4158, pruned_loss=0.1634, over 28862.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.387, pruned_loss=0.1379, over 5641879.74 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3537, pruned_loss=0.1046, over 5759803.33 frames. ], giga_tot_loss[loss=0.3358, simple_loss=0.3904, pruned_loss=0.1406, over 5624995.43 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:52:30,356 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1117915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:52:33,747 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1117918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:52:49,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3927, 1.5205, 1.6189, 1.2287], device='cuda:1'), covar=tensor([0.1638, 0.2462, 0.1352, 0.1653], device='cuda:1'), in_proj_covar=tensor([0.0913, 0.0706, 0.0961, 0.0859], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 00:52:50,573 INFO [train.py:968] (1/2) Epoch 25, batch 24100, giga_loss[loss=0.3168, simple_loss=0.3897, pruned_loss=0.1219, over 28558.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.3861, pruned_loss=0.1365, over 5628523.15 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3537, pruned_loss=0.1049, over 5748708.51 frames. ], giga_tot_loss[loss=0.3343, simple_loss=0.3898, pruned_loss=0.1395, over 5620660.76 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:52:51,235 INFO [optim.py:369] (1/2) 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 00:52:59,048 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1117947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:53:46,813 INFO [train.py:968] (1/2) Epoch 25, batch 24150, giga_loss[loss=0.3795, simple_loss=0.4279, pruned_loss=0.1655, over 28692.00 frames. ], tot_loss[loss=0.3315, simple_loss=0.3884, pruned_loss=0.1373, over 5618284.20 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3538, pruned_loss=0.1049, over 5750265.49 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3916, pruned_loss=0.1399, over 5609431.68 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:54:05,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1836, 2.4579, 1.2488, 1.4169], device='cuda:1'), covar=tensor([0.1008, 0.0336, 0.0892, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0561, 0.0399, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 00:54:32,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3178, 3.1616, 3.0060, 1.3205], device='cuda:1'), covar=tensor([0.0995, 0.1110, 0.0977, 0.2367], device='cuda:1'), in_proj_covar=tensor([0.1277, 0.1181, 0.0994, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 00:54:34,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5273, 1.7169, 1.2502, 1.2973], device='cuda:1'), covar=tensor([0.0949, 0.0488, 0.0924, 0.1197], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0451, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 00:54:38,274 INFO [train.py:968] (1/2) Epoch 25, batch 24200, giga_loss[loss=0.2938, simple_loss=0.3706, pruned_loss=0.1085, over 29038.00 frames. ], tot_loss[loss=0.3326, simple_loss=0.3897, pruned_loss=0.1378, over 5627617.63 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3541, pruned_loss=0.1055, over 5753660.06 frames. ], giga_tot_loss[loss=0.3365, simple_loss=0.3927, pruned_loss=0.1401, over 5614682.00 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:54:38,953 INFO [optim.py:369] (1/2) 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 00:55:27,302 INFO [train.py:968] (1/2) Epoch 25, batch 24250, giga_loss[loss=0.3148, simple_loss=0.3879, pruned_loss=0.1209, over 28888.00 frames. ], tot_loss[loss=0.3291, simple_loss=0.3874, pruned_loss=0.1354, over 5627361.07 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3546, pruned_loss=0.1057, over 5754909.61 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3904, pruned_loss=0.1379, over 5612658.24 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:55:58,313 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4936, 3.6800, 1.6260, 1.7106], device='cuda:1'), covar=tensor([0.1002, 0.0393, 0.0939, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0563, 0.0400, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 00:56:14,259 INFO [train.py:968] (1/2) Epoch 25, batch 24300, giga_loss[loss=0.3254, simple_loss=0.3902, pruned_loss=0.1303, over 28960.00 frames. ], tot_loss[loss=0.3243, simple_loss=0.3843, pruned_loss=0.1322, over 5630389.96 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3541, pruned_loss=0.1056, over 5749076.41 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3879, pruned_loss=0.135, over 5620498.88 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:56:17,148 INFO [optim.py:369] (1/2) 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 00:56:30,192 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8157, 1.3458, 4.9701, 3.6354], device='cuda:1'), covar=tensor([0.1552, 0.2940, 0.0442, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0661, 0.0980, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 00:57:03,879 INFO [train.py:968] (1/2) Epoch 25, batch 24350, giga_loss[loss=0.3862, simple_loss=0.4246, pruned_loss=0.1739, over 28560.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.3803, pruned_loss=0.1285, over 5630113.62 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3541, pruned_loss=0.1057, over 5751876.40 frames. ], giga_tot_loss[loss=0.3233, simple_loss=0.3839, pruned_loss=0.1313, over 5616453.04 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:57:19,333 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 00:57:49,895 INFO [train.py:968] (1/2) Epoch 25, batch 24400, giga_loss[loss=0.2677, simple_loss=0.3359, pruned_loss=0.09972, over 28502.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3758, pruned_loss=0.1249, over 5639816.28 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3537, pruned_loss=0.1055, over 5754958.04 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3794, pruned_loss=0.1277, over 5624352.45 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:57:50,508 INFO [optim.py:369] (1/2) 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 00:58:38,437 INFO [train.py:968] (1/2) Epoch 25, batch 24450, giga_loss[loss=0.2849, simple_loss=0.3505, pruned_loss=0.1097, over 28562.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.374, pruned_loss=0.1241, over 5635823.58 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3538, pruned_loss=0.1058, over 5754198.36 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3772, pruned_loss=0.1264, over 5622366.16 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 00:59:10,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9847, 2.1696, 1.8699, 2.3272], device='cuda:1'), covar=tensor([0.2400, 0.2646, 0.2851, 0.2223], device='cuda:1'), in_proj_covar=tensor([0.1549, 0.1119, 0.1368, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 00:59:27,620 INFO [train.py:968] (1/2) Epoch 25, batch 24500, libri_loss[loss=0.3003, simple_loss=0.3687, pruned_loss=0.1159, over 29535.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1246, over 5641467.07 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3544, pruned_loss=0.1062, over 5753559.66 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3772, pruned_loss=0.1264, over 5629275.50 frames. ], batch size: 83, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 00:59:29,948 INFO [optim.py:369] (1/2) 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 00:59:35,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8581, 1.9440, 1.6353, 1.9267], device='cuda:1'), covar=tensor([0.2528, 0.2771, 0.3073, 0.2581], device='cuda:1'), in_proj_covar=tensor([0.1551, 0.1119, 0.1369, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 01:00:19,308 INFO [train.py:968] (1/2) Epoch 25, batch 24550, giga_loss[loss=0.2962, simple_loss=0.3642, pruned_loss=0.1141, over 28701.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3737, pruned_loss=0.1237, over 5645660.80 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3543, pruned_loss=0.1064, over 5752031.27 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3763, pruned_loss=0.1255, over 5634203.51 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:01:08,320 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:01:09,504 INFO [train.py:968] (1/2) Epoch 25, batch 24600, giga_loss[loss=0.3513, simple_loss=0.4163, pruned_loss=0.1432, over 28126.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3723, pruned_loss=0.1213, over 5652777.63 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3545, pruned_loss=0.1065, over 5746470.91 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3746, pruned_loss=0.1229, over 5646532.27 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:01:11,631 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:02:01,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2694, 1.4973, 1.4088, 1.1891], device='cuda:1'), covar=tensor([0.2933, 0.2704, 0.1920, 0.2530], device='cuda:1'), in_proj_covar=tensor([0.2026, 0.1968, 0.1881, 0.2028], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 01:02:03,053 INFO [train.py:968] (1/2) Epoch 25, batch 24650, giga_loss[loss=0.3931, simple_loss=0.4233, pruned_loss=0.1814, over 26661.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3734, pruned_loss=0.1197, over 5663165.73 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3542, pruned_loss=0.1065, over 5749334.36 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3758, pruned_loss=0.1213, over 5654034.85 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:02:30,101 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:02:34,731 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:02:53,316 INFO [train.py:968] (1/2) Epoch 25, batch 24700, giga_loss[loss=0.3692, simple_loss=0.411, pruned_loss=0.1636, over 28787.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3748, pruned_loss=0.1207, over 5651067.49 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3546, pruned_loss=0.1069, over 5746215.46 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3768, pruned_loss=0.122, over 5644734.74 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:02:54,777 INFO [optim.py:369] (1/2) 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 01:03:06,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:03:28,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9511, 3.7809, 3.6176, 1.7362], device='cuda:1'), covar=tensor([0.0717, 0.0831, 0.0809, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.1280, 0.1189, 0.0998, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 01:03:36,341 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 24750, giga_loss[loss=0.3234, simple_loss=0.3938, pruned_loss=0.1265, over 28633.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3743, pruned_loss=0.1204, over 5669337.76 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3543, pruned_loss=0.1067, over 5748499.50 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3764, pruned_loss=0.1217, over 5661163.31 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:03:56,513 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 24800, giga_loss[loss=0.2982, simple_loss=0.3622, pruned_loss=0.1171, over 28870.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3734, pruned_loss=0.1202, over 5684780.67 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3547, pruned_loss=0.1072, over 5751520.25 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3751, pruned_loss=0.1211, over 5674204.50 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:04:29,799 INFO [optim.py:369] (1/2) 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 01:04:30,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2557, 1.5013, 1.3358, 1.1599], device='cuda:1'), covar=tensor([0.2649, 0.2519, 0.1923, 0.2435], device='cuda:1'), in_proj_covar=tensor([0.2028, 0.1971, 0.1882, 0.2029], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 01:05:14,213 INFO [train.py:968] (1/2) Epoch 25, batch 24850, giga_loss[loss=0.2876, simple_loss=0.3556, pruned_loss=0.1098, over 28685.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3717, pruned_loss=0.1201, over 5675911.31 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3551, pruned_loss=0.1075, over 5744273.84 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.373, pruned_loss=0.1207, over 5671957.51 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:05:21,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 01:05:36,518 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1118717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:05:41,039 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4897, 2.1424, 1.5244, 0.7926], device='cuda:1'), covar=tensor([0.6608, 0.3192, 0.4377, 0.7215], device='cuda:1'), in_proj_covar=tensor([0.1813, 0.1702, 0.1641, 0.1471], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 01:05:45,150 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1118726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:05:50,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1118729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:05:57,147 INFO [train.py:968] (1/2) Epoch 25, batch 24900, giga_loss[loss=0.3093, simple_loss=0.3684, pruned_loss=0.1251, over 28751.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3693, pruned_loss=0.1188, over 5675091.98 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3548, pruned_loss=0.1073, over 5746564.16 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3711, pruned_loss=0.1199, over 5668235.56 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:05:59,518 INFO [optim.py:369] (1/2) 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 01:06:14,864 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1118758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:06:38,505 INFO [train.py:968] (1/2) Epoch 25, batch 24950, giga_loss[loss=0.2852, simple_loss=0.3625, pruned_loss=0.1039, over 28917.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3687, pruned_loss=0.117, over 5684263.94 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3552, pruned_loss=0.1076, over 5749710.91 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3701, pruned_loss=0.1178, over 5673966.34 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:06:57,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4062, 1.6092, 1.6307, 1.3669], device='cuda:1'), covar=tensor([0.3169, 0.2498, 0.1954, 0.2594], device='cuda:1'), in_proj_covar=tensor([0.2030, 0.1974, 0.1886, 0.2032], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 01:07:01,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:07:14,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:07:28,764 INFO [train.py:968] (1/2) Epoch 25, batch 25000, giga_loss[loss=0.3154, simple_loss=0.384, pruned_loss=0.1234, over 28873.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3701, pruned_loss=0.1172, over 5677674.98 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3563, pruned_loss=0.1084, over 5742633.18 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3706, pruned_loss=0.1174, over 5674760.29 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:07:30,800 INFO [optim.py:369] (1/2) 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 01:08:14,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:08:15,543 INFO [train.py:968] (1/2) Epoch 25, batch 25050, giga_loss[loss=0.2902, simple_loss=0.3597, pruned_loss=0.1104, over 29080.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3701, pruned_loss=0.1174, over 5671628.50 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3567, pruned_loss=0.1085, over 5741072.47 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3702, pruned_loss=0.1175, over 5669417.97 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:08:18,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:08:21,474 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-13 01:08:28,945 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 01:08:47,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:08:56,604 INFO [train.py:968] (1/2) Epoch 25, batch 25100, giga_loss[loss=0.2823, simple_loss=0.3545, pruned_loss=0.1051, over 28952.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3683, pruned_loss=0.1167, over 5678962.92 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.357, pruned_loss=0.1091, over 5736019.74 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3687, pruned_loss=0.1166, over 5679543.45 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:09:00,075 INFO [optim.py:369] (1/2) 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 01:09:13,996 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1118956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:17,129 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1118959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:17,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3511, 1.5360, 1.5742, 1.1853], device='cuda:1'), covar=tensor([0.1633, 0.2536, 0.1368, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0710, 0.0965, 0.0864], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 01:09:28,949 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1118970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:30,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1118973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:37,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1118981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:43,411 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1118988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:09:45,330 INFO [train.py:968] (1/2) Epoch 25, batch 25150, giga_loss[loss=0.3234, simple_loss=0.3794, pruned_loss=0.1337, over 27589.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3673, pruned_loss=0.1169, over 5667393.71 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3575, pruned_loss=0.1094, over 5738873.79 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3675, pruned_loss=0.1168, over 5663072.79 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:09:57,900 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:00,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5496, 1.7432, 1.7761, 1.3093], device='cuda:1'), covar=tensor([0.1655, 0.2832, 0.1498, 0.1863], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0710, 0.0967, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 01:10:23,091 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:25,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:27,398 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 25200, giga_loss[loss=0.3468, simple_loss=0.3966, pruned_loss=0.1485, over 28865.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3674, pruned_loss=0.1178, over 5667738.00 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3581, pruned_loss=0.1101, over 5735781.66 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3675, pruned_loss=0.1174, over 5664063.22 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:10:30,032 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:10:52,457 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:56,231 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:57,008 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:10:59,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:16,618 INFO [train.py:968] (1/2) Epoch 25, batch 25250, giga_loss[loss=0.2826, simple_loss=0.351, pruned_loss=0.1071, over 28771.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.367, pruned_loss=0.1182, over 5663853.38 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3582, pruned_loss=0.1103, over 5729883.93 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3672, pruned_loss=0.118, over 5663767.41 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:11:18,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1119092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:19,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1119093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:28,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:51,901 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:11:54,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:12:04,191 INFO [train.py:968] (1/2) Epoch 25, batch 25300, giga_loss[loss=0.264, simple_loss=0.3375, pruned_loss=0.09528, over 28800.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3654, pruned_loss=0.1175, over 5668219.99 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3587, pruned_loss=0.1106, over 5731293.69 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.1171, over 5666318.63 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:12:07,614 INFO [optim.py:369] (1/2) 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 01:12:19,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:12:24,648 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2096, 2.5713, 2.3866, 2.0570], device='cuda:1'), covar=tensor([0.2207, 0.1601, 0.1719, 0.1791], device='cuda:1'), in_proj_covar=tensor([0.2027, 0.1965, 0.1880, 0.2028], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 01:12:53,754 INFO [train.py:968] (1/2) Epoch 25, batch 25350, giga_loss[loss=0.2729, simple_loss=0.3447, pruned_loss=0.1005, over 28960.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3655, pruned_loss=0.1181, over 5647741.57 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3594, pruned_loss=0.1111, over 5718853.61 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3648, pruned_loss=0.1175, over 5655812.35 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:13:35,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:13:37,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:13:38,959 INFO [train.py:968] (1/2) Epoch 25, batch 25400, giga_loss[loss=0.2825, simple_loss=0.3565, pruned_loss=0.1043, over 28949.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.365, pruned_loss=0.1171, over 5653197.26 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3593, pruned_loss=0.111, over 5713635.19 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3648, pruned_loss=0.1169, over 5662278.78 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:13:45,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1882, 5.0153, 4.7701, 2.4527], device='cuda:1'), covar=tensor([0.0502, 0.0626, 0.0678, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.1192, 0.1003, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 01:13:46,208 INFO [optim.py:369] (1/2) 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 01:14:05,396 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:14:05,572 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 01:14:26,064 INFO [train.py:968] (1/2) Epoch 25, batch 25450, giga_loss[loss=0.2742, simple_loss=0.357, pruned_loss=0.09566, over 28703.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3646, pruned_loss=0.1161, over 5655233.65 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3585, pruned_loss=0.1106, over 5717783.38 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3652, pruned_loss=0.1164, over 5657770.38 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:14:50,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3841, 2.1401, 1.7159, 1.7885], device='cuda:1'), covar=tensor([0.0813, 0.0273, 0.0322, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 01:15:05,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2514, 3.0977, 1.3024, 1.5658], device='cuda:1'), covar=tensor([0.1061, 0.0547, 0.0990, 0.1372], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0565, 0.0400, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 01:15:13,249 INFO [train.py:968] (1/2) Epoch 25, batch 25500, giga_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.103, over 28593.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3651, pruned_loss=0.1159, over 5656519.08 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3589, pruned_loss=0.1109, over 5715961.02 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3652, pruned_loss=0.1159, over 5659280.00 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:15:17,507 INFO [optim.py:369] (1/2) 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 01:15:41,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6882, 1.6270, 1.9062, 1.4714], device='cuda:1'), covar=tensor([0.1771, 0.2484, 0.1423, 0.1659], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0711, 0.0966, 0.0866], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 01:15:58,652 INFO [train.py:968] (1/2) Epoch 25, batch 25550, giga_loss[loss=0.4205, simple_loss=0.4336, pruned_loss=0.2037, over 26637.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3657, pruned_loss=0.1167, over 5648357.03 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3591, pruned_loss=0.111, over 5707190.29 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3659, pruned_loss=0.1167, over 5657282.08 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:16:44,793 INFO [train.py:968] (1/2) Epoch 25, batch 25600, giga_loss[loss=0.28, simple_loss=0.3467, pruned_loss=0.1067, over 28852.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 5650467.52 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3588, pruned_loss=0.1109, over 5707652.62 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3684, pruned_loss=0.1193, over 5655692.42 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:16:51,044 INFO [optim.py:369] (1/2) 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 01:17:12,477 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1119468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:17:16,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0549, 1.5234, 1.5012, 1.2865], device='cuda:1'), covar=tensor([0.2262, 0.1597, 0.2316, 0.1937], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0755, 0.0724, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 01:17:33,075 INFO [train.py:968] (1/2) Epoch 25, batch 25650, giga_loss[loss=0.2772, simple_loss=0.3448, pruned_loss=0.1047, over 28926.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3706, pruned_loss=0.1224, over 5645592.42 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.359, pruned_loss=0.1111, over 5709473.60 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3711, pruned_loss=0.1227, over 5646616.47 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:18:22,189 INFO [train.py:968] (1/2) Epoch 25, batch 25700, giga_loss[loss=0.2859, simple_loss=0.351, pruned_loss=0.1104, over 29082.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3715, pruned_loss=0.1238, over 5658092.91 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3592, pruned_loss=0.1113, over 5704603.37 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3719, pruned_loss=0.1241, over 5661748.20 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:18:27,742 INFO [optim.py:369] (1/2) 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 01:19:03,894 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-13 01:19:10,616 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1119587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:12,410 INFO [train.py:968] (1/2) Epoch 25, batch 25750, giga_loss[loss=0.2971, simple_loss=0.3649, pruned_loss=0.1147, over 28548.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3735, pruned_loss=0.1259, over 5642376.79 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3592, pruned_loss=0.1112, over 5706608.45 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.374, pruned_loss=0.1263, over 5642814.95 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:19:28,977 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1119611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:31,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1119614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:54,929 INFO [train.py:968] (1/2) Epoch 25, batch 25800, giga_loss[loss=0.3208, simple_loss=0.3785, pruned_loss=0.1316, over 28844.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.37, pruned_loss=0.1229, over 5661209.89 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3584, pruned_loss=0.1107, over 5712460.53 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3714, pruned_loss=0.124, over 5655037.30 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:19:57,747 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1119643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:19:58,705 INFO [optim.py:369] (1/2) 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 01:20:33,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6216, 5.4128, 5.1758, 2.6763], device='cuda:1'), covar=tensor([0.0554, 0.0727, 0.0736, 0.1762], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.1189, 0.1003, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 01:20:43,263 INFO [train.py:968] (1/2) Epoch 25, batch 25850, giga_loss[loss=0.2963, simple_loss=0.3782, pruned_loss=0.1072, over 28408.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3706, pruned_loss=0.1226, over 5652137.27 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3584, pruned_loss=0.1106, over 5705825.90 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3719, pruned_loss=0.1238, over 5652537.67 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:21:19,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.02 vs. limit=5.0
+2023-03-13 01:21:28,205 INFO [train.py:968] (1/2) Epoch 25, batch 25900, giga_loss[loss=0.3109, simple_loss=0.3824, pruned_loss=0.1197, over 28539.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3684, pruned_loss=0.1195, over 5658447.08 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3581, pruned_loss=0.1104, over 5707799.62 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3699, pruned_loss=0.1207, over 5656655.35 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:21:33,438 INFO [optim.py:369] (1/2) 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 01:21:39,318 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-13 01:22:12,658 INFO [train.py:968] (1/2) Epoch 25, batch 25950, giga_loss[loss=0.2548, simple_loss=0.3349, pruned_loss=0.08735, over 28974.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3671, pruned_loss=0.1188, over 5654347.78 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3585, pruned_loss=0.1106, over 5706020.90 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3683, pruned_loss=0.1199, over 5652366.81 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:22:28,760 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 01:22:35,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8060, 2.0601, 1.4314, 1.6773], device='cuda:1'), covar=tensor([0.1087, 0.0745, 0.1148, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0450, 0.0520, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 01:22:49,134 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-13 01:22:56,601 INFO [train.py:968] (1/2) Epoch 25, batch 26000, giga_loss[loss=0.2638, simple_loss=0.339, pruned_loss=0.09431, over 28904.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3661, pruned_loss=0.1187, over 5666115.55 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3591, pruned_loss=0.1111, over 5712276.46 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3668, pruned_loss=0.1195, over 5657534.40 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:23:00,531 INFO [optim.py:369] (1/2) 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 01:23:42,786 INFO [train.py:968] (1/2) Epoch 25, batch 26050, giga_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1133, over 28705.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3644, pruned_loss=0.1178, over 5684083.60 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3586, pruned_loss=0.1109, over 5717945.28 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3656, pruned_loss=0.1188, over 5671022.41 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:24:30,031 INFO [train.py:968] (1/2) Epoch 25, batch 26100, giga_loss[loss=0.3085, simple_loss=0.3794, pruned_loss=0.1188, over 28889.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3669, pruned_loss=0.1191, over 5685726.08 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3582, pruned_loss=0.1107, over 5719826.67 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3683, pruned_loss=0.1201, over 5673355.55 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:24:35,753 INFO [optim.py:369] (1/2) 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 01:24:38,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-13 01:24:45,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 01:24:49,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1119962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:25:14,231 INFO [train.py:968] (1/2) Epoch 25, batch 26150, giga_loss[loss=0.2823, simple_loss=0.3465, pruned_loss=0.1091, over 23854.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3703, pruned_loss=0.1194, over 5687218.67 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3579, pruned_loss=0.1107, over 5724740.88 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3721, pruned_loss=0.1205, over 5671906.83 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:25:49,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9142, 2.5358, 1.0480, 1.1836], device='cuda:1'), covar=tensor([0.1395, 0.0603, 0.1153, 0.1781], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0565, 0.0400, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 01:25:56,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4156, 1.1256, 3.9706, 3.2786], device='cuda:1'), covar=tensor([0.1608, 0.2974, 0.0494, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0667, 0.0989, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 01:25:59,402 INFO [train.py:968] (1/2) Epoch 25, batch 26200, giga_loss[loss=0.3252, simple_loss=0.3901, pruned_loss=0.1302, over 28661.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3727, pruned_loss=0.1189, over 5673223.19 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3583, pruned_loss=0.1109, over 5706588.66 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3741, pruned_loss=0.1197, over 5675687.10 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:26:05,446 INFO [optim.py:369] (1/2) 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 01:26:43,966 INFO [train.py:968] (1/2) Epoch 25, batch 26250, giga_loss[loss=0.2613, simple_loss=0.3417, pruned_loss=0.0905, over 28650.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3727, pruned_loss=0.1193, over 5686923.91 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3571, pruned_loss=0.1104, over 5713106.03 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3753, pruned_loss=0.1208, over 5681928.27 frames. ], batch size: 60, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:26:50,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5539, 1.7570, 1.6561, 1.3929], device='cuda:1'), covar=tensor([0.3073, 0.2797, 0.2346, 0.2732], device='cuda:1'), in_proj_covar=tensor([0.2034, 0.1977, 0.1888, 0.2035], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 01:27:00,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1120105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:27:02,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1120108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:27:12,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2519, 1.4682, 1.4649, 1.3636], device='cuda:1'), covar=tensor([0.1537, 0.1419, 0.1965, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0753, 0.0722, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 01:27:26,809 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 26300, libri_loss[loss=0.3016, simple_loss=0.3721, pruned_loss=0.1156, over 29517.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3749, pruned_loss=0.1217, over 5681997.65 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3575, pruned_loss=0.1106, over 5713598.78 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3772, pruned_loss=0.123, over 5676802.91 frames. ], batch size: 84, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:27:33,126 INFO [optim.py:369] (1/2) 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 01:27:33,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9023, 1.9749, 1.5426, 1.5303], device='cuda:1'), covar=tensor([0.1066, 0.0782, 0.1062, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0452, 0.0522, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 01:28:16,028 INFO [train.py:968] (1/2) Epoch 25, batch 26350, giga_loss[loss=0.3284, simple_loss=0.3832, pruned_loss=0.1368, over 28473.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3754, pruned_loss=0.1228, over 5681640.68 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3577, pruned_loss=0.1106, over 5717087.61 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3774, pruned_loss=0.124, over 5673811.15 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:29:04,945 INFO [train.py:968] (1/2) Epoch 25, batch 26400, giga_loss[loss=0.2741, simple_loss=0.3461, pruned_loss=0.1011, over 29025.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3754, pruned_loss=0.1236, over 5662468.99 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3582, pruned_loss=0.1112, over 5690529.18 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3767, pruned_loss=0.1243, over 5678536.47 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:29:10,211 INFO [optim.py:369] (1/2) 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 01:29:40,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2905, 3.1340, 1.5231, 1.5232], device='cuda:1'), covar=tensor([0.1027, 0.0421, 0.0921, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0564, 0.0400, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 01:29:50,288 INFO [train.py:968] (1/2) Epoch 25, batch 26450, giga_loss[loss=0.3134, simple_loss=0.3728, pruned_loss=0.127, over 28652.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3728, pruned_loss=0.1224, over 5668961.97 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.358, pruned_loss=0.111, over 5694326.70 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3743, pruned_loss=0.1234, over 5677627.25 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:30:37,030 INFO [train.py:968] (1/2) Epoch 25, batch 26500, giga_loss[loss=0.2639, simple_loss=0.3315, pruned_loss=0.09809, over 28991.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3698, pruned_loss=0.1209, over 5686219.34 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3575, pruned_loss=0.1107, over 5700246.32 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3719, pruned_loss=0.1223, over 5687087.72 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:30:43,197 INFO [optim.py:369] (1/2) 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 01:31:24,325 INFO [train.py:968] (1/2) Epoch 25, batch 26550, giga_loss[loss=0.2893, simple_loss=0.3623, pruned_loss=0.1082, over 28990.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3701, pruned_loss=0.1215, over 5671275.06 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3574, pruned_loss=0.1106, over 5693894.93 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3721, pruned_loss=0.1228, over 5677316.63 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:31:48,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1120417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:32:08,675 INFO [train.py:968] (1/2) Epoch 25, batch 26600, giga_loss[loss=0.4041, simple_loss=0.4281, pruned_loss=0.1901, over 26746.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3703, pruned_loss=0.122, over 5675874.62 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3576, pruned_loss=0.1108, over 5698354.22 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3719, pruned_loss=0.1232, over 5676240.94 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:32:13,942 INFO [optim.py:369] (1/2) 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 01:32:52,567 INFO [train.py:968] (1/2) Epoch 25, batch 26650, giga_loss[loss=0.3005, simple_loss=0.344, pruned_loss=0.1285, over 23691.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3687, pruned_loss=0.1218, over 5660035.77 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.1109, over 5700385.60 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.37, pruned_loss=0.1227, over 5658141.84 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:33:38,559 INFO [train.py:968] (1/2) Epoch 25, batch 26700, libri_loss[loss=0.3097, simple_loss=0.3699, pruned_loss=0.1248, over 19041.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3684, pruned_loss=0.122, over 5652912.44 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.1109, over 5697734.39 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.37, pruned_loss=0.1232, over 5653255.99 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:33:44,132 INFO [optim.py:369] (1/2) 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 01:34:25,390 INFO [train.py:968] (1/2) Epoch 25, batch 26750, giga_loss[loss=0.3045, simple_loss=0.3786, pruned_loss=0.1152, over 28863.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1209, over 5660100.12 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3577, pruned_loss=0.1109, over 5696365.29 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3702, pruned_loss=0.1219, over 5660956.89 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:35:10,310 INFO [train.py:968] (1/2) Epoch 25, batch 26800, libri_loss[loss=0.2565, simple_loss=0.326, pruned_loss=0.09354, over 29567.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3708, pruned_loss=0.1221, over 5656066.41 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3572, pruned_loss=0.1107, over 5698883.24 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3727, pruned_loss=0.1235, over 5653271.31 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:35:18,147 INFO [optim.py:369] (1/2) 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 01:35:26,647 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 01:35:56,826 INFO [train.py:968] (1/2) Epoch 25, batch 26850, giga_loss[loss=0.303, simple_loss=0.3662, pruned_loss=0.1199, over 28689.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3716, pruned_loss=0.1229, over 5655219.72 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3576, pruned_loss=0.1109, over 5693656.71 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3731, pruned_loss=0.1241, over 5656522.79 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:36:40,835 INFO [train.py:968] (1/2) Epoch 25, batch 26900, giga_loss[loss=0.3008, simple_loss=0.3891, pruned_loss=0.1062, over 28913.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 5656589.71 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1112, over 5685360.37 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3729, pruned_loss=0.121, over 5664541.75 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:36:48,023 INFO [optim.py:369] (1/2) 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 01:37:28,038 INFO [train.py:968] (1/2) Epoch 25, batch 26950, giga_loss[loss=0.3485, simple_loss=0.4076, pruned_loss=0.1447, over 28682.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3742, pruned_loss=0.1196, over 5673861.08 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3582, pruned_loss=0.1115, over 5689912.15 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3752, pruned_loss=0.1202, over 5675808.94 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:37:30,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1120792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:38:10,838 INFO [train.py:968] (1/2) Epoch 25, batch 27000, giga_loss[loss=0.3772, simple_loss=0.4363, pruned_loss=0.1591, over 28918.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3766, pruned_loss=0.1202, over 5684257.90 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3583, pruned_loss=0.1116, over 5692745.41 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3775, pruned_loss=0.1207, over 5683158.11 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:38:10,838 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 01:38:19,495 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 01:38:25,722 INFO [optim.py:369] (1/2) 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 01:39:06,920 INFO [train.py:968] (1/2) Epoch 25, batch 27050, giga_loss[loss=0.3254, simple_loss=0.3864, pruned_loss=0.1322, over 28809.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3791, pruned_loss=0.1234, over 5677115.49 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3582, pruned_loss=0.1116, over 5691538.75 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3803, pruned_loss=0.124, over 5676818.41 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:39:43,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7305, 1.6823, 1.9253, 1.4896], device='cuda:1'), covar=tensor([0.1756, 0.2525, 0.1385, 0.1696], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0711, 0.0966, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 01:39:46,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9159, 3.7372, 3.5772, 1.6562], device='cuda:1'), covar=tensor([0.0721, 0.0869, 0.0846, 0.2207], device='cuda:1'), in_proj_covar=tensor([0.1291, 0.1193, 0.1005, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 01:39:53,225 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1120935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:39:56,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1120938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:39:57,463 INFO [train.py:968] (1/2) Epoch 25, batch 27100, giga_loss[loss=0.35, simple_loss=0.399, pruned_loss=0.1505, over 28798.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3818, pruned_loss=0.1268, over 5664068.30 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3585, pruned_loss=0.1117, over 5693854.84 frames. ], giga_tot_loss[loss=0.3187, simple_loss=0.3828, pruned_loss=0.1274, over 5661578.13 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:40:06,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3891, 1.8946, 1.6610, 1.4112], device='cuda:1'), covar=tensor([0.0645, 0.0268, 0.0259, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 01:40:09,213 INFO [optim.py:369] (1/2) 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 01:40:19,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9610, 3.7918, 3.6399, 1.9518], device='cuda:1'), covar=tensor([0.0703, 0.0815, 0.0778, 0.2076], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1193, 0.1005, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 01:40:26,156 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1120967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:40:36,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-13 01:40:50,750 INFO [train.py:968] (1/2) Epoch 25, batch 27150, giga_loss[loss=0.2858, simple_loss=0.3605, pruned_loss=0.1055, over 28644.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3804, pruned_loss=0.1265, over 5659292.08 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1117, over 5695040.57 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3812, pruned_loss=0.127, over 5656272.90 frames. ], batch size: 242, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:41:35,692 INFO [train.py:968] (1/2) Epoch 25, batch 27200, giga_loss[loss=0.3103, simple_loss=0.379, pruned_loss=0.1208, over 28714.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3787, pruned_loss=0.1251, over 5657307.12 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3587, pruned_loss=0.1118, over 5702584.17 frames. ], giga_tot_loss[loss=0.3158, simple_loss=0.3799, pruned_loss=0.1259, over 5646738.48 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 01:41:40,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:41:45,047 INFO [optim.py:369] (1/2) 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 01:41:55,730 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-13 01:42:21,102 INFO [train.py:968] (1/2) Epoch 25, batch 27250, giga_loss[loss=0.2736, simple_loss=0.3573, pruned_loss=0.0949, over 28814.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3772, pruned_loss=0.1226, over 5667282.05 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3585, pruned_loss=0.1118, over 5708598.30 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3788, pruned_loss=0.1235, over 5651985.29 frames. ], batch size: 66, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:42:22,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6140, 1.8051, 1.8399, 1.3714], device='cuda:1'), covar=tensor([0.1931, 0.2736, 0.1662, 0.1997], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0712, 0.0967, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 01:43:06,127 INFO [train.py:968] (1/2) Epoch 25, batch 27300, libri_loss[loss=0.2957, simple_loss=0.3656, pruned_loss=0.1129, over 29215.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.377, pruned_loss=0.1211, over 5670050.27 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3587, pruned_loss=0.112, over 5699517.00 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3786, pruned_loss=0.1219, over 5665547.66 frames. ], batch size: 97, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:43:14,342 INFO [optim.py:369] (1/2) 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 01:43:24,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-13 01:43:31,679 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 01:43:53,792 INFO [train.py:968] (1/2) Epoch 25, batch 27350, giga_loss[loss=0.3735, simple_loss=0.4105, pruned_loss=0.1683, over 27949.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3781, pruned_loss=0.1227, over 5665709.94 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5703362.09 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3797, pruned_loss=0.1234, over 5657988.24 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:44:15,607 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 01:44:33,631 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 01:44:39,302 INFO [train.py:968] (1/2) Epoch 25, batch 27400, giga_loss[loss=0.2824, simple_loss=0.353, pruned_loss=0.1059, over 28982.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3765, pruned_loss=0.1217, over 5667175.08 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.359, pruned_loss=0.1124, over 5696115.51 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3779, pruned_loss=0.1223, over 5666407.77 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:44:46,710 INFO [optim.py:369] (1/2) 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 01:45:21,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:45:28,756 INFO [train.py:968] (1/2) Epoch 25, batch 27450, giga_loss[loss=0.2846, simple_loss=0.3546, pruned_loss=0.1074, over 28671.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3752, pruned_loss=0.1224, over 5642261.92 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1126, over 5681403.37 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3764, pruned_loss=0.1228, over 5655492.44 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:46:13,207 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:46:13,663 INFO [train.py:968] (1/2) Epoch 25, batch 27500, giga_loss[loss=0.2832, simple_loss=0.3464, pruned_loss=0.11, over 28758.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3734, pruned_loss=0.1223, over 5637202.09 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5687928.21 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3746, pruned_loss=0.1229, over 5640296.83 frames. ], batch size: 99, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:46:18,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7363, 1.9071, 1.6698, 1.7142], device='cuda:1'), covar=tensor([0.1956, 0.2571, 0.2472, 0.2331], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0756, 0.0725, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 01:46:19,403 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:46:23,312 INFO [optim.py:369] (1/2) 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 01:47:01,321 INFO [train.py:968] (1/2) Epoch 25, batch 27550, giga_loss[loss=0.3647, simple_loss=0.4088, pruned_loss=0.1603, over 27541.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3719, pruned_loss=0.1218, over 5650119.42 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1128, over 5693836.75 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3732, pruned_loss=0.1224, over 5645893.63 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:47:33,917 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 27600, giga_loss[loss=0.3221, simple_loss=0.3888, pruned_loss=0.1277, over 28303.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1237, over 5645862.80 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5694893.56 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3732, pruned_loss=0.1242, over 5641449.48 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:48:03,099 INFO [optim.py:369] (1/2) 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 01:48:05,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5599, 4.3915, 4.1869, 2.2039], device='cuda:1'), covar=tensor([0.0556, 0.0676, 0.0706, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.1203, 0.1012, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 01:48:38,278 INFO [train.py:968] (1/2) Epoch 25, batch 27650, giga_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 28904.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3711, pruned_loss=0.1227, over 5650331.38 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.113, over 5696543.31 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 5644477.50 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:49:21,541 INFO [train.py:968] (1/2) Epoch 25, batch 27700, giga_loss[loss=0.2857, simple_loss=0.3608, pruned_loss=0.1053, over 28830.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 5652725.73 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1131, over 5692312.27 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3685, pruned_loss=0.1194, over 5650442.90 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:49:32,178 INFO [optim.py:369] (1/2) 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 01:49:41,930 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:49:44,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:50:02,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121587.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 01:50:05,426 INFO [train.py:968] (1/2) Epoch 25, batch 27750, giga_loss[loss=0.2775, simple_loss=0.349, pruned_loss=0.1029, over 28741.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3639, pruned_loss=0.1149, over 5667010.32 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.113, over 5698142.99 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3647, pruned_loss=0.1155, over 5658440.00 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:50:08,181 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:50:57,204 INFO [train.py:968] (1/2) Epoch 25, batch 27800, giga_loss[loss=0.3218, simple_loss=0.3832, pruned_loss=0.1302, over 28833.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3639, pruned_loss=0.115, over 5643598.61 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1132, over 5682625.58 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3644, pruned_loss=0.1154, over 5648543.54 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:51:06,716 INFO [optim.py:369] (1/2) 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 01:51:12,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:51:14,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5397, 1.8678, 1.4969, 1.5132], device='cuda:1'), covar=tensor([0.2659, 0.2699, 0.3109, 0.2465], device='cuda:1'), in_proj_covar=tensor([0.1562, 0.1125, 0.1378, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 01:51:47,500 INFO [train.py:968] (1/2) Epoch 25, batch 27850, giga_loss[loss=0.2473, simple_loss=0.3165, pruned_loss=0.0891, over 28638.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3612, pruned_loss=0.1137, over 5652921.12 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3606, pruned_loss=0.1134, over 5686134.95 frames. ], giga_tot_loss[loss=0.2945, simple_loss=0.3614, pruned_loss=0.1139, over 5652986.20 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:51:52,210 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:52:13,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:52:19,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1121721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:52:29,053 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.91 vs. limit=2.0
+2023-03-13 01:52:30,654 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121730.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 01:52:33,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121733.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 01:52:38,906 INFO [train.py:968] (1/2) Epoch 25, batch 27900, giga_loss[loss=0.3304, simple_loss=0.3901, pruned_loss=0.1353, over 28729.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5656158.55 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1132, over 5690388.29 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.361, pruned_loss=0.1144, over 5651723.68 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:52:49,318 INFO [optim.py:369] (1/2) 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 01:52:57,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121762.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 01:53:23,950 INFO [train.py:968] (1/2) Epoch 25, batch 27950, giga_loss[loss=0.2858, simple_loss=0.3635, pruned_loss=0.104, over 28933.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3637, pruned_loss=0.1153, over 5665795.15 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3608, pruned_loss=0.1136, over 5689414.28 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3637, pruned_loss=0.1152, over 5662559.46 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:53:36,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:53:39,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:02,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:10,464 INFO [train.py:968] (1/2) Epoch 25, batch 28000, giga_loss[loss=0.3574, simple_loss=0.3979, pruned_loss=0.1584, over 23467.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3659, pruned_loss=0.1168, over 5657964.86 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5693338.08 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3666, pruned_loss=0.1171, over 5650790.75 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:54:22,168 INFO [optim.py:369] (1/2) 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 01:54:23,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1121851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:24,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9030, 1.1062, 1.0876, 0.8539], device='cuda:1'), covar=tensor([0.2446, 0.3162, 0.2069, 0.2584], device='cuda:1'), in_proj_covar=tensor([0.2037, 0.1982, 0.1897, 0.2040], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 01:54:28,204 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:29,044 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 01:54:31,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:35,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1121864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:37,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1121867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:46,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.18 vs. limit=5.0
+2023-03-13 01:54:58,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:54:58,801 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-13 01:54:59,026 INFO [train.py:968] (1/2) Epoch 25, batch 28050, giga_loss[loss=0.2813, simple_loss=0.3543, pruned_loss=0.1042, over 29076.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3668, pruned_loss=0.1175, over 5653110.10 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5690785.39 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3672, pruned_loss=0.1177, over 5649763.58 frames. ], batch size: 155, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:55:04,905 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1121896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:55:45,346 INFO [train.py:968] (1/2) Epoch 25, batch 28100, giga_loss[loss=0.3143, simple_loss=0.3788, pruned_loss=0.1249, over 28237.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3682, pruned_loss=0.119, over 5647146.25 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3609, pruned_loss=0.1137, over 5692077.02 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3681, pruned_loss=0.1189, over 5642613.58 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 01:55:52,444 INFO [optim.py:369] (1/2) 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 01:56:26,036 INFO [train.py:968] (1/2) Epoch 25, batch 28150, libri_loss[loss=0.2103, simple_loss=0.2916, pruned_loss=0.06453, over 29640.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3704, pruned_loss=0.1205, over 5663297.95 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5689348.98 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3703, pruned_loss=0.1204, over 5661660.45 frames. ], batch size: 69, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:56:36,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9401, 3.7944, 3.6349, 1.8875], device='cuda:1'), covar=tensor([0.0640, 0.0733, 0.0766, 0.2363], device='cuda:1'), in_proj_covar=tensor([0.1294, 0.1194, 0.1007, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 01:56:59,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2454, 1.5581, 1.3728, 1.2967], device='cuda:1'), covar=tensor([0.2207, 0.2005, 0.2317, 0.2272], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0755, 0.0724, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 01:57:11,604 INFO [train.py:968] (1/2) Epoch 25, batch 28200, giga_loss[loss=0.2874, simple_loss=0.3621, pruned_loss=0.1064, over 28902.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5657019.77 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3614, pruned_loss=0.1141, over 5688567.71 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5656362.62 frames. ], batch size: 119, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:57:24,192 INFO [optim.py:369] (1/2) 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 01:57:41,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:58:02,231 INFO [train.py:968] (1/2) Epoch 25, batch 28250, giga_loss[loss=0.3228, simple_loss=0.3923, pruned_loss=0.1267, over 28726.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3738, pruned_loss=0.1233, over 5653097.38 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3615, pruned_loss=0.1142, over 5689751.87 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3737, pruned_loss=0.1232, over 5651234.52 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:58:18,752 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-13 01:58:38,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 01:58:49,363 INFO [train.py:968] (1/2) Epoch 25, batch 28300, giga_loss[loss=0.3353, simple_loss=0.392, pruned_loss=0.1393, over 28589.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1235, over 5655496.16 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3619, pruned_loss=0.1143, over 5693062.61 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5650353.73 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:59:03,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.36 vs. limit=5.0
+2023-03-13 01:59:03,198 INFO [optim.py:369] (1/2) 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 01:59:19,398 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2358, 0.8850, 0.9689, 1.3407], device='cuda:1'), covar=tensor([0.0747, 0.0400, 0.0355, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 01:59:39,815 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-13 01:59:40,830 INFO [train.py:968] (1/2) Epoch 25, batch 28350, giga_loss[loss=0.3437, simple_loss=0.411, pruned_loss=0.1382, over 28647.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3752, pruned_loss=0.1232, over 5657351.54 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3619, pruned_loss=0.1142, over 5695989.22 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3753, pruned_loss=0.1235, over 5650131.08 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 01:59:42,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3795, 1.7198, 1.2927, 1.4239], device='cuda:1'), covar=tensor([0.0704, 0.0413, 0.0365, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 02:00:05,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:08,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:19,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:32,466 INFO [train.py:968] (1/2) Epoch 25, batch 28400, giga_loss[loss=0.3838, simple_loss=0.4285, pruned_loss=0.1695, over 27627.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3747, pruned_loss=0.1217, over 5665067.88 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.362, pruned_loss=0.1142, over 5698075.79 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3748, pruned_loss=0.122, over 5657203.99 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:00:37,046 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:00:41,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3985, 1.6699, 1.4602, 1.6500], device='cuda:1'), covar=tensor([0.0792, 0.0330, 0.0326, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0121, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 02:00:42,520 INFO [optim.py:369] (1/2) 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 02:01:21,395 INFO [train.py:968] (1/2) Epoch 25, batch 28450, giga_loss[loss=0.2649, simple_loss=0.3313, pruned_loss=0.09923, over 28595.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3731, pruned_loss=0.122, over 5670360.65 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3622, pruned_loss=0.1144, over 5700491.68 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3732, pruned_loss=0.1221, over 5661632.36 frames. ], batch size: 92, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:02:14,073 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:02:14,403 INFO [train.py:968] (1/2) Epoch 25, batch 28500, giga_loss[loss=0.3961, simple_loss=0.4212, pruned_loss=0.1855, over 26463.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3719, pruned_loss=0.122, over 5676692.10 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3617, pruned_loss=0.1142, over 5706918.60 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3727, pruned_loss=0.1225, over 5662522.60 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:02:24,194 INFO [optim.py:369] (1/2) 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 02:02:41,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 02:02:46,383 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:02:49,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3959, 1.5546, 1.4602, 1.5265], device='cuda:1'), covar=tensor([0.0815, 0.0332, 0.0326, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 02:02:49,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:03:10,514 INFO [train.py:968] (1/2) Epoch 25, batch 28550, giga_loss[loss=0.2798, simple_loss=0.3503, pruned_loss=0.1046, over 28667.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.12, over 5681743.39 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3617, pruned_loss=0.114, over 5710122.87 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3697, pruned_loss=0.1206, over 5667598.50 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:03:20,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:03:45,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1759, 1.6467, 1.2747, 0.4212], device='cuda:1'), covar=tensor([0.4444, 0.2530, 0.3750, 0.6381], device='cuda:1'), in_proj_covar=tensor([0.1808, 0.1696, 0.1635, 0.1468], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 02:03:54,037 INFO [train.py:968] (1/2) Epoch 25, batch 28600, giga_loss[loss=0.2891, simple_loss=0.357, pruned_loss=0.1106, over 28868.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3682, pruned_loss=0.1197, over 5675755.65 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3614, pruned_loss=0.1139, over 5699629.19 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3693, pruned_loss=0.1206, over 5672316.05 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:03:59,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 02:04:07,077 INFO [optim.py:369] (1/2) 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 02:04:20,402 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-13 02:04:32,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3698, 1.5270, 1.4993, 1.3608], device='cuda:1'), covar=tensor([0.2757, 0.2590, 0.1955, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.2037, 0.1987, 0.1897, 0.2039], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 02:04:38,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.04 vs. limit=5.0
+2023-03-13 02:04:42,648 INFO [train.py:968] (1/2) Epoch 25, batch 28650, giga_loss[loss=0.3022, simple_loss=0.3751, pruned_loss=0.1146, over 28838.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3686, pruned_loss=0.1208, over 5649350.95 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3613, pruned_loss=0.1138, over 5692396.58 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1217, over 5652891.72 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:04:55,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:05:26,195 INFO [train.py:968] (1/2) Epoch 25, batch 28700, giga_loss[loss=0.3231, simple_loss=0.3848, pruned_loss=0.1307, over 28529.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1209, over 5658152.80 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1139, over 5696275.97 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3699, pruned_loss=0.1218, over 5655952.97 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:05:34,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:05:37,184 INFO [optim.py:369] (1/2) 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 02:05:48,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3575, 2.0240, 1.5481, 0.6473], device='cuda:1'), covar=tensor([0.6259, 0.3184, 0.4052, 0.6755], device='cuda:1'), in_proj_covar=tensor([0.1810, 0.1698, 0.1637, 0.1468], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 02:06:16,261 INFO [train.py:968] (1/2) Epoch 25, batch 28750, giga_loss[loss=0.2796, simple_loss=0.3427, pruned_loss=0.1082, over 28841.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1227, over 5659019.28 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.361, pruned_loss=0.1137, over 5698517.33 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3716, pruned_loss=0.1235, over 5655189.05 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:06:50,023 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:01,633 INFO [train.py:968] (1/2) Epoch 25, batch 28800, giga_loss[loss=0.3548, simple_loss=0.4066, pruned_loss=0.1515, over 28997.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3718, pruned_loss=0.1238, over 5655871.14 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.361, pruned_loss=0.1139, over 5705593.50 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3733, pruned_loss=0.1248, over 5644367.54 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:07:04,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 02:07:08,535 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:11,559 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:13,714 INFO [optim.py:369] (1/2) 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 02:07:41,322 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:07:50,717 INFO [train.py:968] (1/2) Epoch 25, batch 28850, giga_loss[loss=0.2692, simple_loss=0.3387, pruned_loss=0.09984, over 28818.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3712, pruned_loss=0.1237, over 5650877.12 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.361, pruned_loss=0.1139, over 5706043.35 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3724, pruned_loss=0.1246, over 5641091.99 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:08:12,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1122714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:08:36,356 INFO [train.py:968] (1/2) Epoch 25, batch 28900, giga_loss[loss=0.2988, simple_loss=0.3721, pruned_loss=0.1128, over 28857.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3711, pruned_loss=0.1239, over 5656521.69 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.361, pruned_loss=0.1138, over 5706644.31 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3724, pruned_loss=0.1249, over 5647168.26 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:08:38,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4532, 3.6476, 1.6024, 1.6646], device='cuda:1'), covar=tensor([0.1061, 0.0353, 0.0914, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0564, 0.0400, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 02:08:49,111 INFO [optim.py:369] (1/2) 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 02:09:22,524 INFO [train.py:968] (1/2) Epoch 25, batch 28950, giga_loss[loss=0.3406, simple_loss=0.3917, pruned_loss=0.1448, over 29101.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3722, pruned_loss=0.1249, over 5640566.07 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5700091.22 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3729, pruned_loss=0.1256, over 5637859.27 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:09:30,363 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-13 02:10:10,802 INFO [train.py:968] (1/2) Epoch 25, batch 29000, giga_loss[loss=0.3014, simple_loss=0.368, pruned_loss=0.1174, over 28949.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3716, pruned_loss=0.1234, over 5645792.17 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3616, pruned_loss=0.1142, over 5700297.22 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3722, pruned_loss=0.124, over 5642642.17 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:10:22,634 INFO [optim.py:369] (1/2) 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 02:10:27,197 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1122857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:29,329 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1122860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:50,629 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1122882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:56,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1122889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:10:56,687 INFO [train.py:968] (1/2) Epoch 25, batch 29050, libri_loss[loss=0.3178, simple_loss=0.3845, pruned_loss=0.1255, over 29755.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1228, over 5651816.05 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5702876.23 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3723, pruned_loss=0.1235, over 5645858.86 frames. ], batch size: 87, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:11:09,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-13 02:11:29,080 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 29100, libri_loss[loss=0.311, simple_loss=0.3798, pruned_loss=0.1211, over 26388.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3725, pruned_loss=0.1235, over 5664103.66 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3615, pruned_loss=0.1142, over 5703907.97 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3733, pruned_loss=0.1242, over 5657739.03 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:11:42,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2196, 2.5220, 2.2828, 1.9151], device='cuda:1'), covar=tensor([0.3124, 0.2422, 0.2728, 0.3022], device='cuda:1'), in_proj_covar=tensor([0.2035, 0.1984, 0.1894, 0.2035], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 02:11:51,417 INFO [optim.py:369] (1/2) 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 02:12:24,625 INFO [train.py:968] (1/2) Epoch 25, batch 29150, giga_loss[loss=0.3112, simple_loss=0.3763, pruned_loss=0.123, over 28795.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3735, pruned_loss=0.1244, over 5664933.68 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3615, pruned_loss=0.1142, over 5696995.55 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3744, pruned_loss=0.1251, over 5665872.13 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:12:36,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:13:10,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:13:11,348 INFO [train.py:968] (1/2) Epoch 25, batch 29200, giga_loss[loss=0.2827, simple_loss=0.3651, pruned_loss=0.1001, over 28974.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3747, pruned_loss=0.1254, over 5652840.93 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3617, pruned_loss=0.1144, over 5689745.75 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.126, over 5659056.00 frames. ], batch size: 106, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:13:25,342 INFO [optim.py:369] (1/2) 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 02:13:38,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:13:43,878 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:14:02,260 INFO [train.py:968] (1/2) Epoch 25, batch 29250, giga_loss[loss=0.3244, simple_loss=0.3634, pruned_loss=0.1427, over 23677.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3746, pruned_loss=0.1238, over 5657123.77 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.362, pruned_loss=0.1147, over 5693854.96 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.375, pruned_loss=0.1242, over 5657731.71 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:14:10,083 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:14:14,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-13 02:14:29,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9759, 2.1770, 1.4856, 1.7809], device='cuda:1'), covar=tensor([0.1024, 0.0693, 0.1062, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0451, 0.0519, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 02:14:38,732 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:14:47,548 INFO [train.py:968] (1/2) Epoch 25, batch 29300, giga_loss[loss=0.2766, simple_loss=0.3569, pruned_loss=0.09816, over 28924.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3735, pruned_loss=0.1225, over 5661987.59 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1149, over 5698120.82 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3739, pruned_loss=0.1228, over 5657707.93 frames. ], batch size: 174, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:14:53,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:14:55,375 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:15:00,694 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:15:34,573 INFO [train.py:968] (1/2) Epoch 25, batch 29350, giga_loss[loss=0.3051, simple_loss=0.3683, pruned_loss=0.121, over 28803.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3714, pruned_loss=0.1214, over 5661306.51 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1148, over 5699295.21 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3718, pruned_loss=0.1218, over 5656643.05 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:16:10,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:16:19,173 INFO [train.py:968] (1/2) Epoch 25, batch 29400, giga_loss[loss=0.2979, simple_loss=0.3647, pruned_loss=0.1155, over 28889.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3731, pruned_loss=0.123, over 5649285.46 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3625, pruned_loss=0.1149, over 5688849.42 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3734, pruned_loss=0.1234, over 5653365.88 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:16:35,842 INFO [optim.py:369] (1/2) 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 02:16:37,586 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:17:07,052 INFO [train.py:968] (1/2) Epoch 25, batch 29450, giga_loss[loss=0.2822, simple_loss=0.3568, pruned_loss=0.1039, over 28170.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3742, pruned_loss=0.1236, over 5649243.76 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5690778.69 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3747, pruned_loss=0.1242, over 5649296.83 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:17:43,835 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 02:17:45,973 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:17:54,120 INFO [train.py:968] (1/2) Epoch 25, batch 29500, giga_loss[loss=0.3183, simple_loss=0.3755, pruned_loss=0.1305, over 29000.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3756, pruned_loss=0.1253, over 5659946.75 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3625, pruned_loss=0.1149, over 5694851.25 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3764, pruned_loss=0.126, over 5655527.55 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:18:06,721 INFO [optim.py:369] (1/2) 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 02:18:26,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6811, 1.6106, 1.9127, 1.4780], device='cuda:1'), covar=tensor([0.1762, 0.2525, 0.1411, 0.1728], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0714, 0.0967, 0.0864], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 02:18:29,042 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 02:18:35,982 INFO [train.py:968] (1/2) Epoch 25, batch 29550, giga_loss[loss=0.2613, simple_loss=0.3393, pruned_loss=0.09165, over 28314.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3742, pruned_loss=0.1246, over 5664867.75 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3622, pruned_loss=0.1146, over 5697325.51 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3756, pruned_loss=0.1258, over 5657944.12 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:18:46,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:18:48,838 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:18:57,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:19:09,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8689, 5.7082, 5.3950, 2.9240], device='cuda:1'), covar=tensor([0.0414, 0.0515, 0.0647, 0.1585], device='cuda:1'), in_proj_covar=tensor([0.1295, 0.1197, 0.1009, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 02:19:13,386 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:19:19,961 INFO [train.py:968] (1/2) Epoch 25, batch 29600, giga_loss[loss=0.3084, simple_loss=0.3767, pruned_loss=0.1201, over 28552.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3753, pruned_loss=0.1256, over 5643463.94 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3628, pruned_loss=0.1151, over 5679048.15 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3763, pruned_loss=0.1265, over 5652089.89 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:19:33,674 INFO [optim.py:369] (1/2) 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 02:19:37,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3019, 1.1729, 3.7427, 3.2228], device='cuda:1'), covar=tensor([0.1669, 0.2897, 0.0476, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0670, 0.0995, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 02:20:05,428 INFO [train.py:968] (1/2) Epoch 25, batch 29650, giga_loss[loss=0.3035, simple_loss=0.3689, pruned_loss=0.1191, over 28687.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.377, pruned_loss=0.1268, over 5646758.59 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.363, pruned_loss=0.115, over 5682268.62 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.378, pruned_loss=0.1278, over 5650144.47 frames. ], batch size: 78, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:20:19,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:20:54,789 INFO [train.py:968] (1/2) Epoch 25, batch 29700, giga_loss[loss=0.3602, simple_loss=0.4149, pruned_loss=0.1527, over 28721.00 frames. ], tot_loss[loss=0.3156, simple_loss=0.3773, pruned_loss=0.127, over 5653729.77 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1148, over 5684657.23 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3784, pruned_loss=0.128, over 5653951.99 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:21:01,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2665, 1.3409, 3.9345, 3.2359], device='cuda:1'), covar=tensor([0.1759, 0.2820, 0.0485, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0670, 0.0993, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 02:21:07,419 INFO [optim.py:369] (1/2) 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 02:21:09,154 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:21:12,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:21:41,703 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:21:42,081 INFO [train.py:968] (1/2) Epoch 25, batch 29750, giga_loss[loss=0.3071, simple_loss=0.3714, pruned_loss=0.1213, over 28921.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3759, pruned_loss=0.1249, over 5671507.51 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3625, pruned_loss=0.1147, over 5686878.04 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.377, pruned_loss=0.126, over 5669644.53 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:21:45,596 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2272, 1.5181, 1.5617, 1.3399], device='cuda:1'), covar=tensor([0.2211, 0.1906, 0.2515, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0763, 0.0731, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 02:21:54,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:22:17,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1884, 5.0338, 4.7819, 2.1345], device='cuda:1'), covar=tensor([0.0438, 0.0549, 0.0642, 0.2018], device='cuda:1'), in_proj_covar=tensor([0.1299, 0.1199, 0.1012, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 02:22:27,361 INFO [train.py:968] (1/2) Epoch 25, batch 29800, giga_loss[loss=0.3378, simple_loss=0.3918, pruned_loss=0.1419, over 26571.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3751, pruned_loss=0.1241, over 5662184.74 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3621, pruned_loss=0.1142, over 5690970.71 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3769, pruned_loss=0.1258, over 5655870.42 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:22:35,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:22:37,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:22:40,188 INFO [optim.py:369] (1/2) 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 02:22:49,570 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:23:03,975 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:23:08,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4969, 2.0212, 1.6900, 1.5883], device='cuda:1'), covar=tensor([0.0645, 0.0251, 0.0272, 0.0691], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 02:23:12,913 INFO [train.py:968] (1/2) Epoch 25, batch 29850, giga_loss[loss=0.2676, simple_loss=0.3424, pruned_loss=0.09646, over 29023.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3745, pruned_loss=0.1236, over 5667513.12 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3619, pruned_loss=0.1142, over 5696189.77 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3765, pruned_loss=0.1253, over 5656981.86 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:23:26,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1123706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:23:37,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4061, 1.1918, 3.9539, 3.3253], device='cuda:1'), covar=tensor([0.1627, 0.2856, 0.0459, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0669, 0.0994, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 02:23:38,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 02:23:58,921 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 02:24:00,922 INFO [train.py:968] (1/2) Epoch 25, batch 29900, giga_loss[loss=0.3328, simple_loss=0.3893, pruned_loss=0.1382, over 28769.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3722, pruned_loss=0.1224, over 5668809.77 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3618, pruned_loss=0.1141, over 5697767.79 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3739, pruned_loss=0.1239, over 5658797.87 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:24:06,831 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:24:09,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:24:13,352 INFO [optim.py:369] (1/2) 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 02:24:30,730 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4021, 1.5898, 1.5272, 1.4343], device='cuda:1'), covar=tensor([0.1660, 0.1867, 0.2157, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0760, 0.0729, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 02:24:34,733 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:24:41,139 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-13 02:24:46,194 INFO [train.py:968] (1/2) Epoch 25, batch 29950, giga_loss[loss=0.3431, simple_loss=0.4011, pruned_loss=0.1426, over 28318.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3713, pruned_loss=0.1226, over 5665352.26 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3615, pruned_loss=0.1139, over 5703207.77 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3732, pruned_loss=0.1241, over 5652275.17 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 02:25:34,032 INFO [train.py:968] (1/2) Epoch 25, batch 30000, libri_loss[loss=0.237, simple_loss=0.3108, pruned_loss=0.08164, over 28552.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5674414.36 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3611, pruned_loss=0.1136, over 5706321.24 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3701, pruned_loss=0.1227, over 5660105.81 frames. ], batch size: 63, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:25:34,032 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 02:25:42,600 INFO [train.py:1012] (1/2) Epoch 25, validation: loss=0.2049, simple_loss=0.3135, pruned_loss=0.04814, over 944034.00 frames. 
+2023-03-13 02:25:42,601 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 02:25:51,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1123849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:25:53,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1123852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:25:56,812 INFO [optim.py:369] (1/2) 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 02:26:17,920 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1123881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:26:24,956 INFO [train.py:968] (1/2) Epoch 25, batch 30050, giga_loss[loss=0.2624, simple_loss=0.3345, pruned_loss=0.09516, over 28826.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3656, pruned_loss=0.1198, over 5682277.03 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3615, pruned_loss=0.1139, over 5701720.58 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.367, pruned_loss=0.1211, over 5675231.27 frames. ], batch size: 199, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:27:10,488 INFO [train.py:968] (1/2) Epoch 25, batch 30100, giga_loss[loss=0.3162, simple_loss=0.3791, pruned_loss=0.1267, over 29073.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3628, pruned_loss=0.1178, over 5695204.99 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3612, pruned_loss=0.1136, over 5706200.94 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3643, pruned_loss=0.1192, over 5684813.41 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:27:25,683 INFO [optim.py:369] (1/2) 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 02:27:34,109 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1123965.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:27:56,534 INFO [train.py:968] (1/2) Epoch 25, batch 30150, giga_loss[loss=0.2649, simple_loss=0.335, pruned_loss=0.09735, over 28434.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3623, pruned_loss=0.117, over 5692857.96 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3613, pruned_loss=0.1138, over 5708791.76 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3635, pruned_loss=0.118, over 5682148.54 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:28:44,259 INFO [train.py:968] (1/2) Epoch 25, batch 30200, giga_loss[loss=0.2866, simple_loss=0.3646, pruned_loss=0.1042, over 28619.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5685214.56 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5710481.49 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1155, over 5674570.99 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:28:45,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:29:00,681 INFO [optim.py:369] (1/2) 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 02:29:36,533 INFO [train.py:968] (1/2) Epoch 25, batch 30250, giga_loss[loss=0.2683, simple_loss=0.3246, pruned_loss=0.106, over 24548.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3589, pruned_loss=0.1114, over 5671101.73 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3609, pruned_loss=0.1137, over 5711930.74 frames. ], giga_tot_loss[loss=0.2921, simple_loss=0.3599, pruned_loss=0.1121, over 5660517.16 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:29:47,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6216, 5.4629, 5.1911, 2.6619], device='cuda:1'), covar=tensor([0.0457, 0.0612, 0.0734, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1194, 0.1006, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 02:29:52,481 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:30:22,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3473, 4.1819, 3.9948, 1.9175], device='cuda:1'), covar=tensor([0.0597, 0.0761, 0.0837, 0.2174], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1192, 0.1005, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 02:30:22,891 INFO [train.py:968] (1/2) Epoch 25, batch 30300, giga_loss[loss=0.2605, simple_loss=0.3469, pruned_loss=0.08701, over 28745.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3548, pruned_loss=0.1075, over 5668138.43 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3604, pruned_loss=0.1136, over 5713304.67 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.356, pruned_loss=0.1081, over 5657601.68 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:30:42,216 INFO [optim.py:369] (1/2) 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 02:31:09,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1124184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:31:11,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1124187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:31:12,714 INFO [train.py:968] (1/2) Epoch 25, batch 30350, giga_loss[loss=0.2744, simple_loss=0.3494, pruned_loss=0.09971, over 28563.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3511, pruned_loss=0.104, over 5664264.60 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3601, pruned_loss=0.1134, over 5715780.35 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3522, pruned_loss=0.1045, over 5652812.29 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:31:38,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1124216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:31:57,368 INFO [train.py:968] (1/2) Epoch 25, batch 30400, giga_loss[loss=0.2588, simple_loss=0.3482, pruned_loss=0.0847, over 28738.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.35, pruned_loss=0.1017, over 5657039.86 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3597, pruned_loss=0.1134, over 5712197.53 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3508, pruned_loss=0.1018, over 5649852.63 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:32:12,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5064, 1.7559, 1.4525, 1.5104], device='cuda:1'), covar=tensor([0.2760, 0.2664, 0.2979, 0.2435], device='cuda:1'), in_proj_covar=tensor([0.1567, 0.1128, 0.1385, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 02:32:15,878 INFO [optim.py:369] (1/2) 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 02:32:50,131 INFO [train.py:968] (1/2) Epoch 25, batch 30450, libri_loss[loss=0.2209, simple_loss=0.2921, pruned_loss=0.07481, over 29609.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3483, pruned_loss=0.09964, over 5643146.09 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3588, pruned_loss=0.1129, over 5707223.54 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3496, pruned_loss=0.09984, over 5639428.74 frames. ], batch size: 74, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:32:57,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 02:33:05,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-13 02:33:13,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3272, 1.5700, 1.4548, 1.3277], device='cuda:1'), covar=tensor([0.2202, 0.1839, 0.1509, 0.1904], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1952, 0.1870, 0.2011], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 02:33:45,501 INFO [train.py:968] (1/2) Epoch 25, batch 30500, giga_loss[loss=0.304, simple_loss=0.37, pruned_loss=0.119, over 27621.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3496, pruned_loss=0.1005, over 5635666.68 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3588, pruned_loss=0.1129, over 5707223.54 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3507, pruned_loss=0.1007, over 5632773.42 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:33:45,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124340.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:34:01,840 INFO [optim.py:369] (1/2) 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 02:34:30,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 02:34:36,659 INFO [train.py:968] (1/2) Epoch 25, batch 30550, giga_loss[loss=0.2452, simple_loss=0.3282, pruned_loss=0.0811, over 28984.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3475, pruned_loss=0.09946, over 5627392.93 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3586, pruned_loss=0.1131, over 5699282.43 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3482, pruned_loss=0.09901, over 5630379.71 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:35:05,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5406, 1.9312, 1.9396, 1.5502], device='cuda:1'), covar=tensor([0.2956, 0.2083, 0.2070, 0.2556], device='cuda:1'), in_proj_covar=tensor([0.2000, 0.1947, 0.1861, 0.2005], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 02:35:20,938 INFO [train.py:968] (1/2) Epoch 25, batch 30600, libri_loss[loss=0.2889, simple_loss=0.3411, pruned_loss=0.1184, over 29545.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3455, pruned_loss=0.09835, over 5642593.84 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.358, pruned_loss=0.1129, over 5705700.61 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3461, pruned_loss=0.09755, over 5636768.93 frames. ], batch size: 77, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:35:38,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2063, 2.4740, 2.3769, 2.0991], device='cuda:1'), covar=tensor([0.2084, 0.2174, 0.1817, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0749, 0.0719, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 02:35:39,453 INFO [optim.py:369] (1/2) 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 02:36:02,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:36:04,830 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1124483.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:36:07,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1124486.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:36:09,729 INFO [train.py:968] (1/2) Epoch 25, batch 30650, libri_loss[loss=0.2562, simple_loss=0.3262, pruned_loss=0.09307, over 29538.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3444, pruned_loss=0.0973, over 5633824.47 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3575, pruned_loss=0.1127, over 5699446.68 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3451, pruned_loss=0.09663, over 5633183.32 frames. ], batch size: 80, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:36:32,766 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:36:32,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1124515.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:36:57,930 INFO [train.py:968] (1/2) Epoch 25, batch 30700, giga_loss[loss=0.2513, simple_loss=0.3329, pruned_loss=0.0849, over 28590.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3435, pruned_loss=0.09608, over 5634047.41 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3573, pruned_loss=0.1126, over 5691771.42 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3442, pruned_loss=0.09548, over 5640053.79 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:37:00,045 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:37:15,143 INFO [optim.py:369] (1/2) 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 02:37:47,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4604, 1.0018, 4.2753, 3.3870], device='cuda:1'), covar=tensor([0.1636, 0.3218, 0.0410, 0.1203], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0663, 0.0982, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 02:37:47,713 INFO [train.py:968] (1/2) Epoch 25, batch 30750, giga_loss[loss=0.2446, simple_loss=0.3339, pruned_loss=0.07763, over 28783.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3413, pruned_loss=0.09395, over 5641190.97 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3573, pruned_loss=0.1128, over 5693772.99 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3415, pruned_loss=0.09295, over 5643013.94 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:38:22,003 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1124625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:38:24,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1124628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:38:36,197 INFO [train.py:968] (1/2) Epoch 25, batch 30800, libri_loss[loss=0.3676, simple_loss=0.4077, pruned_loss=0.1637, over 25906.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3378, pruned_loss=0.09164, over 5628927.66 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.357, pruned_loss=0.1127, over 5685932.40 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3378, pruned_loss=0.09034, over 5636195.02 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:38:53,790 INFO [optim.py:369] (1/2) 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 02:38:54,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1124657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:39:02,282 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:39:24,756 INFO [train.py:968] (1/2) Epoch 25, batch 30850, libri_loss[loss=0.2677, simple_loss=0.3226, pruned_loss=0.1064, over 29342.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3345, pruned_loss=0.09052, over 5638793.45 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3563, pruned_loss=0.1126, over 5692347.02 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3344, pruned_loss=0.08884, over 5636870.24 frames. ], batch size: 71, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:40:02,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5997, 1.7687, 1.2989, 1.3119], device='cuda:1'), covar=tensor([0.0982, 0.0526, 0.0980, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0447, 0.0518, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 02:40:06,680 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 30900, libri_loss[loss=0.3204, simple_loss=0.3754, pruned_loss=0.1326, over 27547.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3346, pruned_loss=0.09162, over 5649607.82 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3563, pruned_loss=0.1129, over 5697924.28 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3337, pruned_loss=0.08911, over 5641037.03 frames. ], batch size: 115, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:40:24,821 INFO [optim.py:369] (1/2) 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 02:40:57,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2221, 3.8321, 1.4294, 1.4655], device='cuda:1'), covar=tensor([0.1275, 0.0453, 0.1105, 0.1604], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0561, 0.0399, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 02:41:01,988 INFO [train.py:968] (1/2) Epoch 25, batch 30950, giga_loss[loss=0.2772, simple_loss=0.3526, pruned_loss=0.1009, over 27977.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3341, pruned_loss=0.09175, over 5631818.47 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3559, pruned_loss=0.1128, over 5701768.04 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3333, pruned_loss=0.08945, over 5620730.52 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:41:18,811 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4607, 1.6605, 1.7039, 1.3044], device='cuda:1'), covar=tensor([0.1958, 0.2920, 0.1682, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0707, 0.0965, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 02:41:51,074 INFO [train.py:968] (1/2) Epoch 25, batch 31000, giga_loss[loss=0.2314, simple_loss=0.3218, pruned_loss=0.07046, over 28897.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3359, pruned_loss=0.09214, over 5631959.23 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.355, pruned_loss=0.1126, over 5695731.74 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3352, pruned_loss=0.08959, over 5626092.69 frames. ], batch size: 112, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:42:05,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5566, 1.8279, 1.8198, 1.4672], device='cuda:1'), covar=tensor([0.3123, 0.2316, 0.1991, 0.2575], device='cuda:1'), in_proj_covar=tensor([0.1984, 0.1929, 0.1842, 0.1986], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 02:42:11,201 INFO [optim.py:369] (1/2) 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 02:42:44,413 INFO [train.py:968] (1/2) Epoch 25, batch 31050, libri_loss[loss=0.3073, simple_loss=0.3642, pruned_loss=0.1252, over 25697.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09269, over 5642579.30 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3551, pruned_loss=0.1127, over 5697908.17 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3378, pruned_loss=0.08993, over 5634459.99 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:42:45,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:43:15,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1124917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:43:20,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1124921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:43:41,997 INFO [train.py:968] (1/2) Epoch 25, batch 31100, giga_loss[loss=0.254, simple_loss=0.337, pruned_loss=0.08553, over 28930.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3383, pruned_loss=0.09187, over 5666099.73 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3538, pruned_loss=0.1121, over 5704153.40 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3381, pruned_loss=0.08965, over 5652681.77 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:44:04,316 INFO [optim.py:369] (1/2) 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 02:44:15,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4700, 1.7423, 1.7515, 1.4714], device='cuda:1'), covar=tensor([0.3384, 0.2425, 0.1979, 0.2640], device='cuda:1'), in_proj_covar=tensor([0.1988, 0.1932, 0.1846, 0.1989], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 02:44:43,655 INFO [train.py:968] (1/2) Epoch 25, batch 31150, libri_loss[loss=0.2842, simple_loss=0.3469, pruned_loss=0.1107, over 29526.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3361, pruned_loss=0.09047, over 5660566.16 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3536, pruned_loss=0.112, over 5697937.71 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3359, pruned_loss=0.08842, over 5655170.23 frames. ], batch size: 82, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:45:10,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:45:32,802 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:45:35,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:45:37,676 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:45:40,713 INFO [train.py:968] (1/2) Epoch 25, batch 31200, giga_loss[loss=0.2583, simple_loss=0.3438, pruned_loss=0.08638, over 29002.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3355, pruned_loss=0.0894, over 5655672.62 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3527, pruned_loss=0.1116, over 5693069.37 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3356, pruned_loss=0.08749, over 5655048.99 frames. ], batch size: 120, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:45:41,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:02,196 INFO [optim.py:369] (1/2) 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 02:46:04,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:06,642 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:08,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:09,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:37,204 INFO [train.py:968] (1/2) Epoch 25, batch 31250, giga_loss[loss=0.2541, simple_loss=0.3279, pruned_loss=0.0901, over 28740.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3341, pruned_loss=0.08822, over 5662143.46 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3524, pruned_loss=0.1114, over 5698142.21 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.334, pruned_loss=0.08631, over 5656237.93 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:46:41,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:46:51,861 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 02:47:01,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125111.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 02:47:32,476 INFO [train.py:968] (1/2) Epoch 25, batch 31300, giga_loss[loss=0.2449, simple_loss=0.3282, pruned_loss=0.08078, over 28989.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3321, pruned_loss=0.08863, over 5672352.50 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3529, pruned_loss=0.112, over 5703838.04 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3309, pruned_loss=0.08568, over 5661211.35 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:47:57,511 INFO [optim.py:369] (1/2) 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 02:48:22,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:48:26,811 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 31350, libri_loss[loss=0.2857, simple_loss=0.3508, pruned_loss=0.1103, over 26085.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3314, pruned_loss=0.08858, over 5669449.05 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3522, pruned_loss=0.1117, over 5706425.68 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3305, pruned_loss=0.08584, over 5657342.22 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:48:55,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1288, 1.5799, 1.7075, 1.3163], device='cuda:1'), covar=tensor([0.2145, 0.1561, 0.1873, 0.1825], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0744, 0.0715, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 02:49:00,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:49:23,725 INFO [train.py:968] (1/2) Epoch 25, batch 31400, libri_loss[loss=0.3047, simple_loss=0.3621, pruned_loss=0.1237, over 19997.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3316, pruned_loss=0.08913, over 5664408.69 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3515, pruned_loss=0.1115, over 5700508.66 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3309, pruned_loss=0.08638, over 5659550.63 frames. ], batch size: 187, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:49:40,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125254.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:49:44,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125257.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:49:45,202 INFO [optim.py:369] (1/2) 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 02:50:10,043 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:50:16,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125286.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:50:20,557 INFO [train.py:968] (1/2) Epoch 25, batch 31450, giga_loss[loss=0.2406, simple_loss=0.3283, pruned_loss=0.07647, over 28535.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3334, pruned_loss=0.089, over 5666160.97 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3509, pruned_loss=0.1113, over 5705250.98 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3329, pruned_loss=0.08655, over 5657368.45 frames. ], batch size: 85, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:50:29,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:50:41,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2576, 2.5589, 1.2276, 1.3825], device='cuda:1'), covar=tensor([0.1009, 0.0362, 0.0992, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0562, 0.0400, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 02:51:20,826 INFO [train.py:968] (1/2) Epoch 25, batch 31500, giga_loss[loss=0.2619, simple_loss=0.3275, pruned_loss=0.09811, over 27561.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3339, pruned_loss=0.08889, over 5670117.72 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3507, pruned_loss=0.1112, over 5709196.47 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3332, pruned_loss=0.08628, over 5658483.92 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:51:41,843 INFO [optim.py:369] (1/2) 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 02:51:46,493 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:52:18,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9841, 1.1615, 2.8730, 2.8331], device='cuda:1'), covar=tensor([0.1618, 0.2627, 0.0579, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0666, 0.0981, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 02:52:25,165 INFO [train.py:968] (1/2) Epoch 25, batch 31550, giga_loss[loss=0.3046, simple_loss=0.3667, pruned_loss=0.1213, over 26934.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3317, pruned_loss=0.08762, over 5669791.74 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3511, pruned_loss=0.1114, over 5709129.48 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3304, pruned_loss=0.08482, over 5659752.28 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:52:25,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125390.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:52:54,717 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:53:25,479 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:53:28,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2316, 1.6978, 1.5513, 1.1324], device='cuda:1'), covar=tensor([0.1556, 0.2489, 0.1387, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0705, 0.0966, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 02:53:29,407 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 31600, giga_loss[loss=0.2427, simple_loss=0.3318, pruned_loss=0.07676, over 28716.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3343, pruned_loss=0.08887, over 5666320.83 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3511, pruned_loss=0.1115, over 5697687.91 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.333, pruned_loss=0.08631, over 5666928.14 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 02:53:33,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:53:54,068 INFO [optim.py:369] (1/2) 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 02:54:10,059 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:54:34,700 INFO [train.py:968] (1/2) Epoch 25, batch 31650, giga_loss[loss=0.2695, simple_loss=0.3643, pruned_loss=0.08734, over 28982.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3365, pruned_loss=0.0879, over 5652494.30 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3503, pruned_loss=0.111, over 5699166.64 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.336, pruned_loss=0.08603, over 5650901.41 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:55:29,772 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125533.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:55:31,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125536.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:55:34,806 INFO [train.py:968] (1/2) Epoch 25, batch 31700, giga_loss[loss=0.2421, simple_loss=0.3366, pruned_loss=0.07383, over 28639.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3388, pruned_loss=0.08764, over 5661801.98 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3505, pruned_loss=0.1113, over 5705389.22 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.338, pruned_loss=0.08521, over 5653554.72 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:55:52,678 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:55:56,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:55:56,957 INFO [optim.py:369] (1/2) 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 02:56:04,236 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125565.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 02:56:20,390 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:56:23,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:56:30,996 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:56:32,274 INFO [train.py:968] (1/2) Epoch 25, batch 31750, giga_loss[loss=0.2643, simple_loss=0.3444, pruned_loss=0.0921, over 28948.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3394, pruned_loss=0.08692, over 5661651.02 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3506, pruned_loss=0.1113, over 5703581.99 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3384, pruned_loss=0.08452, over 5656017.51 frames. ], batch size: 213, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:56:56,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:57:30,677 INFO [train.py:968] (1/2) Epoch 25, batch 31800, giga_loss[loss=0.2822, simple_loss=0.3611, pruned_loss=0.1016, over 28954.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3384, pruned_loss=0.08666, over 5660847.42 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3501, pruned_loss=0.111, over 5707529.66 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3377, pruned_loss=0.08421, over 5651347.53 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:57:52,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1125656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:57:54,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2580, 1.1466, 3.7164, 3.1835], device='cuda:1'), covar=tensor([0.1895, 0.3119, 0.0773, 0.1473], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0665, 0.0979, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 02:57:55,257 INFO [optim.py:369] (1/2) 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 02:58:33,658 INFO [train.py:968] (1/2) Epoch 25, batch 31850, giga_loss[loss=0.2408, simple_loss=0.3272, pruned_loss=0.07725, over 28729.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3395, pruned_loss=0.08882, over 5654724.35 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3501, pruned_loss=0.1111, over 5707320.20 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3388, pruned_loss=0.08636, over 5646134.87 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:59:30,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 02:59:41,342 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 31900, giga_loss[loss=0.2665, simple_loss=0.3457, pruned_loss=0.09369, over 28883.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3397, pruned_loss=0.08966, over 5668426.68 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3498, pruned_loss=0.1108, over 5709991.05 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3391, pruned_loss=0.08749, over 5658159.58 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 02:59:57,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5056, 2.1689, 1.7652, 1.7750], device='cuda:1'), covar=tensor([0.0773, 0.0251, 0.0317, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 03:00:13,440 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-13 03:00:16,493 INFO [optim.py:369] (1/2) 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 03:00:51,235 INFO [train.py:968] (1/2) Epoch 25, batch 31950, giga_loss[loss=0.2139, simple_loss=0.303, pruned_loss=0.06243, over 28714.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3382, pruned_loss=0.08952, over 5663516.51 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.35, pruned_loss=0.1113, over 5693611.32 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3373, pruned_loss=0.08682, over 5669363.17 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:01:07,521 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:01:11,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:01:29,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7921, 1.8832, 1.6393, 1.8948], device='cuda:1'), covar=tensor([0.0546, 0.0285, 0.0267, 0.0615], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 03:01:45,179 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:01:55,332 INFO [train.py:968] (1/2) Epoch 25, batch 32000, giga_loss[loss=0.2668, simple_loss=0.3432, pruned_loss=0.09519, over 28031.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3346, pruned_loss=0.08738, over 5660730.04 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.35, pruned_loss=0.1113, over 5689386.01 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3336, pruned_loss=0.08462, over 5668327.16 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:02:18,941 INFO [optim.py:369] (1/2) 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 03:02:46,763 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1125881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:02:52,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1125884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:03:01,038 INFO [train.py:968] (1/2) Epoch 25, batch 32050, giga_loss[loss=0.2571, simple_loss=0.3333, pruned_loss=0.09049, over 28333.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.332, pruned_loss=0.08631, over 5657127.85 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3492, pruned_loss=0.1108, over 5694421.24 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3315, pruned_loss=0.08399, over 5658028.36 frames. ], batch size: 368, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:03:30,197 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1125913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:04:02,938 INFO [train.py:968] (1/2) Epoch 25, batch 32100, giga_loss[loss=0.2435, simple_loss=0.329, pruned_loss=0.07903, over 28982.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3334, pruned_loss=0.08738, over 5654303.52 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3491, pruned_loss=0.1107, over 5686545.23 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3328, pruned_loss=0.08507, over 5661031.83 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:04:26,472 INFO [optim.py:369] (1/2) 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 03:04:56,725 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1125985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:05:01,479 INFO [train.py:968] (1/2) Epoch 25, batch 32150, giga_loss[loss=0.3093, simple_loss=0.3788, pruned_loss=0.1199, over 27694.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.336, pruned_loss=0.08826, over 5663957.24 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3486, pruned_loss=0.1105, over 5691913.81 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3356, pruned_loss=0.08609, over 5664111.88 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:06:02,492 INFO [train.py:968] (1/2) Epoch 25, batch 32200, giga_loss[loss=0.3102, simple_loss=0.3634, pruned_loss=0.1285, over 26808.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3351, pruned_loss=0.08905, over 5661158.22 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3483, pruned_loss=0.1104, over 5685495.59 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3347, pruned_loss=0.08668, over 5665248.63 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:06:05,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3682, 3.1616, 1.5523, 1.5614], device='cuda:1'), covar=tensor([0.1008, 0.0384, 0.0954, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0558, 0.0399, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 03:06:25,331 INFO [optim.py:369] (1/2) 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 03:07:00,526 INFO [train.py:968] (1/2) Epoch 25, batch 32250, giga_loss[loss=0.2908, simple_loss=0.3558, pruned_loss=0.1129, over 26893.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3354, pruned_loss=0.09028, over 5666728.47 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3481, pruned_loss=0.1106, over 5690997.02 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3349, pruned_loss=0.08778, over 5664683.90 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:07:18,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:07:38,394 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-13 03:07:56,055 INFO [train.py:968] (1/2) Epoch 25, batch 32300, giga_loss[loss=0.3069, simple_loss=0.371, pruned_loss=0.1214, over 26914.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3364, pruned_loss=0.09135, over 5669379.32 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3481, pruned_loss=0.1108, over 5696168.90 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3355, pruned_loss=0.08832, over 5662180.13 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:08:27,328 INFO [optim.py:369] (1/2) 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 03:08:43,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 03:09:07,394 INFO [train.py:968] (1/2) Epoch 25, batch 32350, giga_loss[loss=0.2952, simple_loss=0.3723, pruned_loss=0.109, over 28973.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3383, pruned_loss=0.09148, over 5668406.42 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3479, pruned_loss=0.1106, over 5699872.50 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3375, pruned_loss=0.08874, over 5658640.95 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:09:09,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-13 03:10:18,083 INFO [train.py:968] (1/2) Epoch 25, batch 32400, giga_loss[loss=0.2136, simple_loss=0.2898, pruned_loss=0.06875, over 24307.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3393, pruned_loss=0.09121, over 5664913.31 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.348, pruned_loss=0.1107, over 5694319.19 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3384, pruned_loss=0.08853, over 5660719.96 frames. ], batch size: 705, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:10:30,770 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1126248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:10:34,228 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:10:53,452 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:11:17,020 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1126280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:11:26,576 INFO [train.py:968] (1/2) Epoch 25, batch 32450, giga_loss[loss=0.2464, simple_loss=0.3225, pruned_loss=0.08519, over 29062.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3371, pruned_loss=0.09027, over 5663011.62 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.348, pruned_loss=0.1107, over 5687810.14 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3363, pruned_loss=0.08789, over 5665818.45 frames. ], batch size: 285, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:12:09,076 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-13 03:12:11,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-13 03:12:31,979 INFO [train.py:968] (1/2) Epoch 25, batch 32500, giga_loss[loss=0.2033, simple_loss=0.2891, pruned_loss=0.05872, over 28036.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3318, pruned_loss=0.08829, over 5668050.47 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3477, pruned_loss=0.1106, over 5689386.49 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3311, pruned_loss=0.08608, over 5668117.73 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:12:59,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:13:02,590 INFO [optim.py:369] (1/2) 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 03:13:34,574 INFO [train.py:968] (1/2) Epoch 25, batch 32550, giga_loss[loss=0.2504, simple_loss=0.3294, pruned_loss=0.08574, over 27720.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3287, pruned_loss=0.08746, over 5652057.99 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3475, pruned_loss=0.1105, over 5683002.22 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3276, pruned_loss=0.0848, over 5657617.50 frames. ], batch size: 474, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:14:28,524 INFO [train.py:968] (1/2) Epoch 25, batch 32600, giga_loss[loss=0.2455, simple_loss=0.3238, pruned_loss=0.08359, over 28059.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3295, pruned_loss=0.08802, over 5658358.97 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3472, pruned_loss=0.1103, over 5690523.56 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3282, pruned_loss=0.08534, over 5654960.83 frames. ], batch size: 412, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:14:52,811 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 32650, giga_loss[loss=0.25, simple_loss=0.3337, pruned_loss=0.08313, over 28843.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3295, pruned_loss=0.08806, over 5651658.20 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3468, pruned_loss=0.1102, over 5684521.63 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3286, pruned_loss=0.08565, over 5653415.68 frames. ], batch size: 284, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:15:43,142 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1126503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:15:43,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:15:46,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1126506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:15:53,574 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 03:16:26,682 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1126535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:16:32,057 INFO [train.py:968] (1/2) Epoch 25, batch 32700, giga_loss[loss=0.2206, simple_loss=0.3109, pruned_loss=0.06514, over 28891.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.327, pruned_loss=0.08576, over 5647986.15 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3464, pruned_loss=0.11, over 5686904.52 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3264, pruned_loss=0.08381, over 5646832.26 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:16:36,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3715, 3.2956, 1.4543, 1.5714], device='cuda:1'), covar=tensor([0.1022, 0.0351, 0.0993, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0558, 0.0398, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 03:16:57,128 INFO [optim.py:369] (1/2) 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 03:17:31,954 INFO [train.py:968] (1/2) Epoch 25, batch 32750, giga_loss[loss=0.2537, simple_loss=0.3383, pruned_loss=0.08457, over 28956.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3266, pruned_loss=0.08546, over 5648645.46 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.347, pruned_loss=0.1106, over 5673247.47 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3252, pruned_loss=0.08291, over 5658028.21 frames. ], batch size: 186, lr: 1.27e-03, grad_scale: 2.0
+2023-03-13 03:18:36,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:18:36,627 INFO [train.py:968] (1/2) Epoch 25, batch 32800, giga_loss[loss=0.2374, simple_loss=0.3259, pruned_loss=0.07447, over 28683.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3259, pruned_loss=0.08551, over 5650738.30 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3467, pruned_loss=0.1105, over 5677959.12 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3245, pruned_loss=0.08282, over 5653545.36 frames. ], batch size: 262, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:19:08,053 INFO [optim.py:369] (1/2) 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 03:19:41,353 INFO [train.py:968] (1/2) Epoch 25, batch 32850, giga_loss[loss=0.2699, simple_loss=0.3483, pruned_loss=0.09569, over 28894.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3262, pruned_loss=0.08508, over 5633042.87 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3466, pruned_loss=0.1106, over 5662488.36 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3249, pruned_loss=0.08245, over 5648277.31 frames. ], batch size: 227, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:20:08,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-13 03:20:32,184 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:20:43,525 INFO [train.py:968] (1/2) Epoch 25, batch 32900, giga_loss[loss=0.2268, simple_loss=0.3112, pruned_loss=0.07119, over 27668.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.327, pruned_loss=0.08587, over 5637196.58 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3472, pruned_loss=0.1112, over 5656166.39 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3252, pruned_loss=0.08292, over 5655174.62 frames. ], batch size: 472, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:21:01,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6697, 2.1129, 2.1994, 1.5849], device='cuda:1'), covar=tensor([0.3436, 0.2245, 0.2271, 0.2858], device='cuda:1'), in_proj_covar=tensor([0.1983, 0.1916, 0.1828, 0.1980], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 03:21:08,102 INFO [optim.py:369] (1/2) 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 03:21:27,519 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1126782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:21:30,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1126785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:21:35,128 INFO [train.py:968] (1/2) Epoch 25, batch 32950, giga_loss[loss=0.2349, simple_loss=0.32, pruned_loss=0.07495, over 28641.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3269, pruned_loss=0.08648, over 5654585.29 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3463, pruned_loss=0.1107, over 5667294.42 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3253, pruned_loss=0.08341, over 5658340.19 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:22:03,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1126814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:22:31,253 INFO [train.py:968] (1/2) Epoch 25, batch 33000, giga_loss[loss=0.2217, simple_loss=0.309, pruned_loss=0.06725, over 28422.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3262, pruned_loss=0.08518, over 5656711.61 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3462, pruned_loss=0.1107, over 5672858.59 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3243, pruned_loss=0.08194, over 5654077.85 frames. ], batch size: 369, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:22:31,253 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 03:22:39,891 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 03:22:57,044 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:23:05,325 INFO [optim.py:369] (1/2) 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 03:23:06,215 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:23:24,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1126879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:23:34,671 INFO [train.py:968] (1/2) Epoch 25, batch 33050, giga_loss[loss=0.2221, simple_loss=0.3201, pruned_loss=0.06198, over 28952.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3284, pruned_loss=0.08483, over 5660383.46 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.346, pruned_loss=0.1106, over 5677319.48 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3268, pruned_loss=0.08187, over 5653843.51 frames. ], batch size: 136, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:23:47,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-13 03:24:31,965 INFO [train.py:968] (1/2) Epoch 25, batch 33100, giga_loss[loss=0.2345, simple_loss=0.3262, pruned_loss=0.0714, over 28866.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3306, pruned_loss=0.08565, over 5658801.59 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3455, pruned_loss=0.1102, over 5682524.17 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3294, pruned_loss=0.08307, over 5648425.05 frames. ], batch size: 145, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:24:36,637 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1126944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:24:59,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-13 03:24:59,271 INFO [optim.py:369] (1/2) 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 03:25:36,946 INFO [train.py:968] (1/2) Epoch 25, batch 33150, giga_loss[loss=0.246, simple_loss=0.3323, pruned_loss=0.07981, over 28608.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3313, pruned_loss=0.08561, over 5659641.91 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3454, pruned_loss=0.11, over 5684669.00 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3302, pruned_loss=0.08337, over 5649251.26 frames. ], batch size: 307, lr: 1.27e-03, grad_scale: 4.0
+2023-03-13 03:25:48,128 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-13 03:26:16,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:19,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:33,547 INFO [train.py:968] (1/2) Epoch 25, batch 33200, giga_loss[loss=0.2208, simple_loss=0.3111, pruned_loss=0.06524, over 28777.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3317, pruned_loss=0.08656, over 5650188.84 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3451, pruned_loss=0.1099, over 5671981.11 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3305, pruned_loss=0.08403, over 5652108.08 frames. ], batch size: 243, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:26:47,030 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:50,425 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:26:58,388 INFO [optim.py:369] (1/2) 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 03:27:22,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4103, 1.9162, 1.8440, 1.6230], device='cuda:1'), covar=tensor([0.2279, 0.2076, 0.2164, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.0472, 0.0733, 0.0708, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 03:27:34,277 INFO [train.py:968] (1/2) Epoch 25, batch 33250, giga_loss[loss=0.2279, simple_loss=0.3153, pruned_loss=0.07025, over 28512.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3293, pruned_loss=0.08495, over 5656643.49 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3448, pruned_loss=0.1097, over 5676501.32 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3284, pruned_loss=0.08268, over 5653983.88 frames. ], batch size: 336, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:27:52,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:28:13,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2504, 1.5504, 1.4364, 1.2397], device='cuda:1'), covar=tensor([0.2833, 0.2291, 0.1547, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1980, 0.1912, 0.1820, 0.1976], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 03:28:31,761 INFO [train.py:968] (1/2) Epoch 25, batch 33300, giga_loss[loss=0.2231, simple_loss=0.3106, pruned_loss=0.06782, over 29141.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3277, pruned_loss=0.08468, over 5659517.10 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3444, pruned_loss=0.1094, over 5679141.36 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.327, pruned_loss=0.08267, over 5654525.70 frames. ], batch size: 128, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:28:55,478 INFO [optim.py:369] (1/2) 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 03:29:29,867 INFO [train.py:968] (1/2) Epoch 25, batch 33350, giga_loss[loss=0.2801, simple_loss=0.349, pruned_loss=0.1056, over 26838.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3269, pruned_loss=0.08422, over 5667457.85 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3442, pruned_loss=0.1091, over 5679942.66 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3263, pruned_loss=0.08243, over 5662377.09 frames. ], batch size: 555, lr: 1.27e-03, grad_scale: 8.0
+2023-03-13 03:30:16,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:26,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:28,246 INFO [train.py:968] (1/2) Epoch 25, batch 33400, giga_loss[loss=0.2616, simple_loss=0.3439, pruned_loss=0.08964, over 28674.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.33, pruned_loss=0.08584, over 5666141.45 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3434, pruned_loss=0.1087, over 5683047.39 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3296, pruned_loss=0.08392, over 5658965.81 frames. ], batch size: 242, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:30:29,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0424, 3.8812, 3.6980, 1.7970], device='cuda:1'), covar=tensor([0.0576, 0.0750, 0.0753, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.1259, 0.1155, 0.0974, 0.0726], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 03:30:37,817 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:43,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:30:56,132 INFO [optim.py:369] (1/2) 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 03:31:15,996 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:31:26,259 INFO [train.py:968] (1/2) Epoch 25, batch 33450, giga_loss[loss=0.2641, simple_loss=0.3406, pruned_loss=0.09382, over 28768.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3308, pruned_loss=0.08682, over 5672091.43 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3434, pruned_loss=0.1089, over 5687102.51 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.33, pruned_loss=0.08432, over 5661880.76 frames. ], batch size: 243, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:31:53,052 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:32:28,187 INFO [train.py:968] (1/2) Epoch 25, batch 33500, giga_loss[loss=0.2911, simple_loss=0.3683, pruned_loss=0.1069, over 28690.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3318, pruned_loss=0.08751, over 5677732.70 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3436, pruned_loss=0.1091, over 5691567.63 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3308, pruned_loss=0.08497, over 5665231.60 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:32:33,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1179, 1.2535, 1.0878, 0.8425], device='cuda:1'), covar=tensor([0.1046, 0.0497, 0.1081, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0444, 0.0518, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 03:32:58,038 INFO [optim.py:369] (1/2) 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 03:33:09,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:11,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:17,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:19,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:23,580 INFO [train.py:968] (1/2) Epoch 25, batch 33550, giga_loss[loss=0.2535, simple_loss=0.3429, pruned_loss=0.08204, over 28868.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3344, pruned_loss=0.08875, over 5662451.86 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3432, pruned_loss=0.1088, over 5681858.01 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3337, pruned_loss=0.08619, over 5659891.61 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:33:39,019 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:33:48,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:03,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:16,719 INFO [train.py:968] (1/2) Epoch 25, batch 33600, giga_loss[loss=0.2531, simple_loss=0.3336, pruned_loss=0.08623, over 28522.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3355, pruned_loss=0.0885, over 5664310.64 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3431, pruned_loss=0.1088, over 5685989.60 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3349, pruned_loss=0.08606, over 5658021.90 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:34:41,151 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:47,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:34:51,065 INFO [optim.py:369] (1/2) 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 03:34:54,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:35:03,541 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3768, 1.5474, 1.3481, 1.5029], device='cuda:1'), covar=tensor([0.0780, 0.0348, 0.0360, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 03:35:26,388 INFO [train.py:968] (1/2) Epoch 25, batch 33650, giga_loss[loss=0.2758, simple_loss=0.3413, pruned_loss=0.1051, over 26807.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3337, pruned_loss=0.0876, over 5664482.47 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3426, pruned_loss=0.1085, over 5690252.47 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3335, pruned_loss=0.08555, over 5655347.19 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:35:32,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:36:19,306 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:36:20,146 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:36:30,152 INFO [train.py:968] (1/2) Epoch 25, batch 33700, giga_loss[loss=0.2524, simple_loss=0.3206, pruned_loss=0.09212, over 26903.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3316, pruned_loss=0.08692, over 5655346.27 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3421, pruned_loss=0.1083, over 5676654.07 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3317, pruned_loss=0.08495, over 5659623.84 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:36:41,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4096, 3.3957, 1.4422, 1.5224], device='cuda:1'), covar=tensor([0.1033, 0.0387, 0.0993, 0.1444], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0559, 0.0398, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 03:36:49,202 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-13 03:36:59,735 INFO [optim.py:369] (1/2) 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 03:37:05,395 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:37:07,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:37:23,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9650, 1.3379, 1.1369, 0.1781], device='cuda:1'), covar=tensor([0.4488, 0.3585, 0.5317, 0.7387], device='cuda:1'), in_proj_covar=tensor([0.1801, 0.1691, 0.1629, 0.1467], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 03:37:29,220 INFO [train.py:968] (1/2) Epoch 25, batch 33750, giga_loss[loss=0.2421, simple_loss=0.3263, pruned_loss=0.07895, over 28572.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.331, pruned_loss=0.08638, over 5653386.35 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3418, pruned_loss=0.108, over 5677770.28 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.331, pruned_loss=0.08445, over 5655017.13 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:37:44,094 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:38:05,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4779, 1.7374, 1.3772, 1.6588], device='cuda:1'), covar=tensor([0.0771, 0.0308, 0.0348, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 03:38:06,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3854, 1.6896, 1.3349, 1.0103], device='cuda:1'), covar=tensor([0.2883, 0.2830, 0.3412, 0.2491], device='cuda:1'), in_proj_covar=tensor([0.1561, 0.1120, 0.1379, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 03:38:30,894 INFO [train.py:968] (1/2) Epoch 25, batch 33800, giga_loss[loss=0.2821, simple_loss=0.3536, pruned_loss=0.1053, over 28030.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3311, pruned_loss=0.0872, over 5650211.48 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3417, pruned_loss=0.1081, over 5683319.33 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3309, pruned_loss=0.08503, over 5646118.97 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:39:03,765 INFO [optim.py:369] (1/2) 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 03:39:33,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:39:35,991 INFO [train.py:968] (1/2) Epoch 25, batch 33850, giga_loss[loss=0.2017, simple_loss=0.2814, pruned_loss=0.061, over 28746.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3297, pruned_loss=0.08732, over 5646454.70 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.342, pruned_loss=0.1084, over 5676718.89 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3291, pruned_loss=0.08503, over 5648640.55 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:40:33,148 INFO [train.py:968] (1/2) Epoch 25, batch 33900, giga_loss[loss=0.2322, simple_loss=0.3194, pruned_loss=0.07246, over 28492.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3294, pruned_loss=0.08679, over 5637086.84 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3418, pruned_loss=0.1082, over 5671045.82 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3288, pruned_loss=0.0846, over 5643567.29 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:41:02,647 INFO [optim.py:369] (1/2) 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 03:41:33,078 INFO [train.py:968] (1/2) Epoch 25, batch 33950, giga_loss[loss=0.2394, simple_loss=0.3239, pruned_loss=0.07744, over 28609.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3281, pruned_loss=0.08489, over 5657056.44 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3414, pruned_loss=0.1079, over 5676996.43 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3276, pruned_loss=0.08294, over 5656337.85 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:41:43,998 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:19,245 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:20,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:21,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:29,299 INFO [train.py:968] (1/2) Epoch 25, batch 34000, giga_loss[loss=0.2388, simple_loss=0.3318, pruned_loss=0.07285, over 28599.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3309, pruned_loss=0.08436, over 5667053.48 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3418, pruned_loss=0.108, over 5676755.03 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3301, pruned_loss=0.08233, over 5666444.27 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:42:51,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1127862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:42:54,591 INFO [optim.py:369] (1/2) 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 03:43:23,671 INFO [train.py:968] (1/2) Epoch 25, batch 34050, giga_loss[loss=0.2169, simple_loss=0.3028, pruned_loss=0.06548, over 28669.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3326, pruned_loss=0.08504, over 5668811.31 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3415, pruned_loss=0.1078, over 5685249.68 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3319, pruned_loss=0.08276, over 5660296.41 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:43:42,476 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:43:43,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1127906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:44:06,993 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1127929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:44:21,378 INFO [train.py:968] (1/2) Epoch 25, batch 34100, giga_loss[loss=0.2135, simple_loss=0.3124, pruned_loss=0.05725, over 29028.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3314, pruned_loss=0.08432, over 5669531.37 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3407, pruned_loss=0.1074, over 5689847.25 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3311, pruned_loss=0.08182, over 5657904.14 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:44:58,595 INFO [optim.py:369] (1/2) 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 03:45:13,541 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1127975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:45:16,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1127978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:45:31,381 INFO [train.py:968] (1/2) Epoch 25, batch 34150, giga_loss[loss=0.275, simple_loss=0.3545, pruned_loss=0.09776, over 28863.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3313, pruned_loss=0.08364, over 5672729.56 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3408, pruned_loss=0.1074, over 5690435.02 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.331, pruned_loss=0.08159, over 5663085.31 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:45:53,324 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:37,347 INFO [train.py:968] (1/2) Epoch 25, batch 34200, giga_loss[loss=0.2448, simple_loss=0.3352, pruned_loss=0.07718, over 28476.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3314, pruned_loss=0.08381, over 5662325.00 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3413, pruned_loss=0.1078, over 5683355.35 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3305, pruned_loss=0.08132, over 5661149.96 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:46:48,632 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:50,135 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:51,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:46:53,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:47:03,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4660, 1.5577, 1.6847, 1.2693], device='cuda:1'), covar=tensor([0.2063, 0.2893, 0.1678, 0.2058], device='cuda:1'), in_proj_covar=tensor([0.0916, 0.0700, 0.0963, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 03:47:14,402 INFO [optim.py:369] (1/2) 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 03:47:34,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:47:35,616 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:47:50,487 INFO [train.py:968] (1/2) Epoch 25, batch 34250, giga_loss[loss=0.2333, simple_loss=0.3288, pruned_loss=0.06893, over 28783.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3309, pruned_loss=0.08262, over 5661447.58 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3411, pruned_loss=0.1077, over 5685205.42 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3303, pruned_loss=0.08053, over 5658749.98 frames. ], batch size: 243, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:48:52,705 INFO [train.py:968] (1/2) Epoch 25, batch 34300, giga_loss[loss=0.2517, simple_loss=0.3369, pruned_loss=0.08329, over 29039.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3322, pruned_loss=0.08356, over 5659632.86 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3408, pruned_loss=0.1075, over 5691742.04 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3317, pruned_loss=0.08139, over 5650821.15 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:49:26,487 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:49:57,546 INFO [train.py:968] (1/2) Epoch 25, batch 34350, giga_loss[loss=0.2407, simple_loss=0.3277, pruned_loss=0.07687, over 29007.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3351, pruned_loss=0.08461, over 5675458.01 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3405, pruned_loss=0.1073, over 5696272.27 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3348, pruned_loss=0.08257, over 5663772.47 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:50:41,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3653, 1.7556, 1.7170, 1.5510], device='cuda:1'), covar=tensor([0.1961, 0.2072, 0.2069, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.0468, 0.0730, 0.0702, 0.0674], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-13 03:51:00,192 INFO [train.py:968] (1/2) Epoch 25, batch 34400, giga_loss[loss=0.2393, simple_loss=0.3279, pruned_loss=0.07537, over 28620.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3354, pruned_loss=0.08529, over 5674413.37 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3403, pruned_loss=0.1071, over 5688789.25 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3354, pruned_loss=0.08327, over 5671226.93 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:51:33,507 INFO [optim.py:369] (1/2) 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 03:52:09,082 INFO [train.py:968] (1/2) Epoch 25, batch 34450, giga_loss[loss=0.2584, simple_loss=0.3498, pruned_loss=0.08355, over 29025.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3331, pruned_loss=0.08418, over 5675162.50 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3405, pruned_loss=0.1072, over 5684636.27 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3327, pruned_loss=0.08225, over 5676174.72 frames. ], batch size: 285, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 03:52:28,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:52:45,154 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:52:49,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:53:05,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2759, 1.7030, 1.3573, 1.4709], device='cuda:1'), covar=tensor([0.0783, 0.0302, 0.0321, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 03:53:18,257 INFO [train.py:968] (1/2) Epoch 25, batch 34500, giga_loss[loss=0.2576, simple_loss=0.3357, pruned_loss=0.08972, over 26973.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3316, pruned_loss=0.08277, over 5682557.25 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3406, pruned_loss=0.1073, over 5687496.13 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.331, pruned_loss=0.08061, over 5680612.35 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:53:28,251 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:54:22,467 INFO [train.py:968] (1/2) Epoch 25, batch 34550, giga_loss[loss=0.2817, simple_loss=0.3605, pruned_loss=0.1015, over 28795.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3308, pruned_loss=0.08222, over 5682590.00 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3407, pruned_loss=0.1074, over 5680317.68 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3302, pruned_loss=0.0803, over 5688024.12 frames. ], batch size: 243, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:54:45,773 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 03:55:20,043 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 34600, giga_loss[loss=0.3002, simple_loss=0.3722, pruned_loss=0.1141, over 28086.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3313, pruned_loss=0.08272, over 5671174.24 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3407, pruned_loss=0.1073, over 5674303.01 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3308, pruned_loss=0.08083, over 5680013.48 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:55:20,727 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1128440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:55:30,532 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 03:55:32,990 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:56:07,951 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 34650, giga_loss[loss=0.2485, simple_loss=0.3222, pruned_loss=0.08738, over 24626.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3337, pruned_loss=0.0847, over 5669207.02 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3404, pruned_loss=0.1072, over 5681957.91 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3332, pruned_loss=0.0824, over 5668804.46 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:56:31,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4586, 1.9335, 1.7962, 1.7045], device='cuda:1'), covar=tensor([0.2138, 0.1965, 0.2132, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0470, 0.0732, 0.0706, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-13 03:57:10,132 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-13 03:57:13,986 INFO [train.py:968] (1/2) Epoch 25, batch 34700, giga_loss[loss=0.3082, simple_loss=0.363, pruned_loss=0.1267, over 28965.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3325, pruned_loss=0.08495, over 5671943.16 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3401, pruned_loss=0.107, over 5686649.87 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3322, pruned_loss=0.08274, over 5667183.83 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:57:42,535 INFO [optim.py:369] (1/2) 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 03:57:44,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-13 03:58:04,408 INFO [train.py:968] (1/2) Epoch 25, batch 34750, giga_loss[loss=0.2676, simple_loss=0.3336, pruned_loss=0.1008, over 26819.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3305, pruned_loss=0.08507, over 5666332.51 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3401, pruned_loss=0.1071, over 5681929.94 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.33, pruned_loss=0.08254, over 5666939.68 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 03:58:19,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5515, 3.5417, 1.5321, 1.6992], device='cuda:1'), covar=tensor([0.0958, 0.0437, 0.0968, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0559, 0.0400, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 03:58:26,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4765, 1.6584, 1.3789, 1.5408], device='cuda:1'), covar=tensor([0.0765, 0.0328, 0.0353, 0.0901], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 03:59:00,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9030, 1.9570, 2.1246, 1.6216], device='cuda:1'), covar=tensor([0.1856, 0.2634, 0.1495, 0.1866], device='cuda:1'), in_proj_covar=tensor([0.0913, 0.0697, 0.0960, 0.0863], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 03:59:02,883 INFO [train.py:968] (1/2) Epoch 25, batch 34800, giga_loss[loss=0.2803, simple_loss=0.362, pruned_loss=0.09932, over 28888.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3306, pruned_loss=0.08517, over 5665656.20 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3401, pruned_loss=0.107, over 5683091.29 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3301, pruned_loss=0.0831, over 5664975.61 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 03:59:32,666 INFO [optim.py:369] (1/2) 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 03:59:52,777 INFO [train.py:968] (1/2) Epoch 25, batch 34850, giga_loss[loss=0.2936, simple_loss=0.3764, pruned_loss=0.1054, over 28577.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3381, pruned_loss=0.08952, over 5661337.90 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.34, pruned_loss=0.107, over 5680762.16 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3377, pruned_loss=0.08761, over 5662528.67 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:00:40,393 INFO [train.py:968] (1/2) Epoch 25, batch 34900, giga_loss[loss=0.3106, simple_loss=0.3651, pruned_loss=0.1281, over 23746.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3468, pruned_loss=0.09392, over 5668817.82 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3398, pruned_loss=0.1069, over 5681920.35 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3467, pruned_loss=0.09251, over 5668708.62 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:00:51,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1128755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:01:01,924 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 04:01:03,580 INFO [optim.py:369] (1/2) 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 04:01:21,639 INFO [train.py:968] (1/2) Epoch 25, batch 34950, giga_loss[loss=0.2529, simple_loss=0.3332, pruned_loss=0.08625, over 28457.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3483, pruned_loss=0.09578, over 5674726.17 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3398, pruned_loss=0.1069, over 5685315.70 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3483, pruned_loss=0.09443, over 5671491.60 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:01:43,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128814.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:01:44,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1128815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:02:03,238 INFO [train.py:968] (1/2) Epoch 25, batch 35000, giga_loss[loss=0.2356, simple_loss=0.3222, pruned_loss=0.07447, over 28983.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3433, pruned_loss=0.094, over 5674223.97 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3403, pruned_loss=0.1071, over 5680797.36 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3431, pruned_loss=0.09238, over 5676161.90 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:02:12,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2241, 3.0671, 2.8882, 1.3839], device='cuda:1'), covar=tensor([0.0992, 0.1096, 0.0980, 0.2537], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1159, 0.0978, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 04:02:29,324 INFO [optim.py:369] (1/2) 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 04:02:50,078 INFO [train.py:968] (1/2) Epoch 25, batch 35050, giga_loss[loss=0.2564, simple_loss=0.3309, pruned_loss=0.09095, over 28542.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3354, pruned_loss=0.0906, over 5673880.49 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3401, pruned_loss=0.107, over 5681974.37 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3354, pruned_loss=0.0894, over 5674378.85 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:03:30,526 INFO [train.py:968] (1/2) Epoch 25, batch 35100, giga_loss[loss=0.2427, simple_loss=0.3149, pruned_loss=0.08529, over 28729.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3287, pruned_loss=0.08758, over 5683874.85 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3407, pruned_loss=0.1071, over 5687465.46 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3279, pruned_loss=0.0861, over 5679250.26 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:03:44,725 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128957.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:03:45,278 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1128958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:03:46,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1128960.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:03:47,134 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:04:08,365 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128989.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:04:08,644 INFO [train.py:968] (1/2) Epoch 25, batch 35150, giga_loss[loss=0.2498, simple_loss=0.302, pruned_loss=0.09879, over 23583.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3219, pruned_loss=0.085, over 5684159.41 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3407, pruned_loss=0.107, over 5691834.72 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3207, pruned_loss=0.08321, over 5676441.77 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:04:08,841 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1128990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:04:51,260 INFO [train.py:968] (1/2) Epoch 25, batch 35200, giga_loss[loss=0.2605, simple_loss=0.3245, pruned_loss=0.09824, over 28631.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3165, pruned_loss=0.08278, over 5691540.24 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.341, pruned_loss=0.1071, over 5696939.69 frames. ], giga_tot_loss[loss=0.2381, simple_loss=0.3148, pruned_loss=0.08069, over 5680489.01 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:04:56,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6356, 1.7106, 1.8472, 1.4521], device='cuda:1'), covar=tensor([0.1915, 0.2521, 0.1536, 0.1775], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0705, 0.0970, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 04:05:14,871 INFO [optim.py:369] (1/2) 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 04:05:19,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2353, 1.6607, 1.2071, 0.5428], device='cuda:1'), covar=tensor([0.4658, 0.2331, 0.3384, 0.6619], device='cuda:1'), in_proj_covar=tensor([0.1805, 0.1701, 0.1629, 0.1468], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 04:05:30,072 INFO [train.py:968] (1/2) Epoch 25, batch 35250, giga_loss[loss=0.2413, simple_loss=0.3139, pruned_loss=0.08434, over 28643.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3137, pruned_loss=0.08176, over 5699282.76 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3409, pruned_loss=0.107, over 5697296.42 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3119, pruned_loss=0.07969, over 5690374.48 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:06:07,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:06:14,751 INFO [train.py:968] (1/2) Epoch 25, batch 35300, giga_loss[loss=0.2348, simple_loss=0.3125, pruned_loss=0.0785, over 28637.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.31, pruned_loss=0.08025, over 5685491.40 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3411, pruned_loss=0.1072, over 5688276.94 frames. ], giga_tot_loss[loss=0.2326, simple_loss=0.3084, pruned_loss=0.07841, over 5687004.53 frames. ], batch size: 242, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:06:38,166 INFO [optim.py:369] (1/2) 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 04:06:54,051 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:06:54,436 INFO [train.py:968] (1/2) Epoch 25, batch 35350, giga_loss[loss=0.2184, simple_loss=0.2972, pruned_loss=0.06977, over 29010.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.3081, pruned_loss=0.07968, over 5680180.98 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3418, pruned_loss=0.1074, over 5694492.63 frames. ], giga_tot_loss[loss=0.2297, simple_loss=0.3051, pruned_loss=0.07715, over 5675778.28 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:07:26,466 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5798, 1.9004, 1.5557, 1.3358], device='cuda:1'), covar=tensor([0.2654, 0.2744, 0.3096, 0.2525], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1125, 0.1385, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 04:07:29,293 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:07:39,854 INFO [train.py:968] (1/2) Epoch 25, batch 35400, giga_loss[loss=0.2149, simple_loss=0.2883, pruned_loss=0.07073, over 28821.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.3042, pruned_loss=0.07766, over 5684731.02 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3418, pruned_loss=0.1072, over 5697123.08 frames. ], giga_tot_loss[loss=0.2261, simple_loss=0.3014, pruned_loss=0.07544, over 5678798.09 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:07:43,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0707, 1.2448, 1.1650, 1.0721], device='cuda:1'), covar=tensor([0.2381, 0.2294, 0.1634, 0.2153], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1929, 0.1836, 0.1996], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:07:52,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 04:07:53,158 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:07:53,294 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2622, 1.5526, 1.3544, 1.1624], device='cuda:1'), covar=tensor([0.3100, 0.2771, 0.1948, 0.2638], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1929, 0.1835, 0.1996], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:08:01,835 INFO [optim.py:369] (1/2) 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 04:08:06,067 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:08:08,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:08:18,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1771, 1.2798, 1.1828, 0.8251], device='cuda:1'), covar=tensor([0.1108, 0.0574, 0.1138, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0442, 0.0515, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 04:08:20,958 INFO [train.py:968] (1/2) Epoch 25, batch 35450, giga_loss[loss=0.2282, simple_loss=0.3011, pruned_loss=0.07766, over 28601.00 frames. ], tot_loss[loss=0.2296, simple_loss=0.3035, pruned_loss=0.07782, over 5688643.05 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3426, pruned_loss=0.1077, over 5698633.46 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.2997, pruned_loss=0.07499, over 5682305.89 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:08:36,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:08:59,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7134, 1.9665, 1.6064, 1.8813], device='cuda:1'), covar=tensor([0.2727, 0.2792, 0.3090, 0.2710], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1126, 0.1385, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 04:09:01,648 INFO [train.py:968] (1/2) Epoch 25, batch 35500, giga_loss[loss=0.196, simple_loss=0.2665, pruned_loss=0.0628, over 28550.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.3011, pruned_loss=0.07692, over 5689319.70 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3429, pruned_loss=0.1077, over 5699794.67 frames. ], giga_tot_loss[loss=0.2224, simple_loss=0.2969, pruned_loss=0.07391, over 5683132.07 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:09:25,122 INFO [optim.py:369] (1/2) 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 04:09:44,245 INFO [train.py:968] (1/2) Epoch 25, batch 35550, giga_loss[loss=0.2018, simple_loss=0.2735, pruned_loss=0.06503, over 28959.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.2982, pruned_loss=0.07545, over 5682847.48 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3431, pruned_loss=0.1078, over 5692383.75 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2942, pruned_loss=0.07262, over 5684065.06 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:10:11,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:10:32,719 INFO [train.py:968] (1/2) Epoch 25, batch 35600, giga_loss[loss=0.2134, simple_loss=0.2813, pruned_loss=0.07276, over 28708.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2956, pruned_loss=0.07451, over 5675607.92 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3434, pruned_loss=0.1078, over 5693919.09 frames. ], giga_tot_loss[loss=0.2176, simple_loss=0.2916, pruned_loss=0.0718, over 5674912.38 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:10:58,310 INFO [optim.py:369] (1/2) 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 04:11:13,548 INFO [train.py:968] (1/2) Epoch 25, batch 35650, giga_loss[loss=0.2518, simple_loss=0.3322, pruned_loss=0.08568, over 28852.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3007, pruned_loss=0.07768, over 5635659.80 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3438, pruned_loss=0.1083, over 5653121.06 frames. ], giga_tot_loss[loss=0.2219, simple_loss=0.2956, pruned_loss=0.0741, over 5672229.08 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:11:56,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-13 04:12:03,056 INFO [train.py:968] (1/2) Epoch 25, batch 35700, giga_loss[loss=0.3317, simple_loss=0.3949, pruned_loss=0.1343, over 27890.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.3146, pruned_loss=0.08485, over 5650935.27 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.344, pruned_loss=0.1084, over 5654195.45 frames. ], giga_tot_loss[loss=0.2369, simple_loss=0.3102, pruned_loss=0.0818, over 5678587.93 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:12:25,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:12:26,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6753, 1.8544, 1.5313, 1.7433], device='cuda:1'), covar=tensor([0.2572, 0.2741, 0.3030, 0.2391], device='cuda:1'), in_proj_covar=tensor([0.1561, 0.1122, 0.1380, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 04:12:30,680 INFO [optim.py:369] (1/2) 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 04:12:48,083 INFO [train.py:968] (1/2) Epoch 25, batch 35750, giga_loss[loss=0.3244, simple_loss=0.3839, pruned_loss=0.1325, over 27639.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3258, pruned_loss=0.09041, over 5655183.47 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3435, pruned_loss=0.108, over 5659043.50 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3222, pruned_loss=0.08782, over 5672884.38 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:12:58,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:13:12,754 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 04:13:25,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:13:26,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-13 04:13:30,895 INFO [train.py:968] (1/2) Epoch 25, batch 35800, giga_loss[loss=0.2619, simple_loss=0.346, pruned_loss=0.08888, over 28949.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3345, pruned_loss=0.09381, over 5666315.02 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3433, pruned_loss=0.1078, over 5660503.57 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3315, pruned_loss=0.09175, over 5679000.50 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:13:56,870 INFO [optim.py:369] (1/2) 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 04:14:14,300 INFO [train.py:968] (1/2) Epoch 25, batch 35850, giga_loss[loss=0.3029, simple_loss=0.375, pruned_loss=0.1154, over 27911.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3382, pruned_loss=0.09443, over 5673927.36 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3434, pruned_loss=0.1077, over 5662574.95 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3358, pruned_loss=0.09264, over 5682350.07 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:14:23,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4496, 1.5151, 1.6480, 1.2672], device='cuda:1'), covar=tensor([0.1706, 0.2496, 0.1455, 0.1765], device='cuda:1'), in_proj_covar=tensor([0.0926, 0.0708, 0.0974, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 04:14:24,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4622, 1.6480, 1.6177, 1.4103], device='cuda:1'), covar=tensor([0.3354, 0.2955, 0.2211, 0.2781], device='cuda:1'), in_proj_covar=tensor([0.2009, 0.1937, 0.1840, 0.2004], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:14:27,888 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:14:33,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:14:42,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6933, 1.8980, 1.6207, 1.8178], device='cuda:1'), covar=tensor([0.2783, 0.2908, 0.3229, 0.2505], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1125, 0.1383, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 04:14:58,953 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:14:59,419 INFO [train.py:968] (1/2) Epoch 25, batch 35900, giga_loss[loss=0.2735, simple_loss=0.3515, pruned_loss=0.09776, over 28577.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3392, pruned_loss=0.0936, over 5664985.93 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3434, pruned_loss=0.1077, over 5667304.83 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.337, pruned_loss=0.09189, over 5667834.97 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:14:59,607 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1129740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:04,060 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:06,914 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:15:32,634 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:33,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:34,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:37,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5584, 1.9194, 1.0802, 1.5088], device='cuda:1'), covar=tensor([0.1271, 0.0814, 0.1635, 0.1410], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0444, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 04:15:44,208 INFO [train.py:968] (1/2) Epoch 25, batch 35950, giga_loss[loss=0.2743, simple_loss=0.3503, pruned_loss=0.09912, over 28943.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.342, pruned_loss=0.09536, over 5671341.88 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3434, pruned_loss=0.1077, over 5670009.21 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3402, pruned_loss=0.09385, over 5671237.31 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:15:45,113 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1129791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:15:58,622 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:16:05,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-13 04:16:26,788 INFO [train.py:968] (1/2) Epoch 25, batch 36000, giga_loss[loss=0.28, simple_loss=0.3575, pruned_loss=0.1012, over 28930.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3442, pruned_loss=0.09668, over 5681446.18 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3435, pruned_loss=0.1076, over 5672775.28 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3427, pruned_loss=0.09538, over 5678841.74 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:16:26,789 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 04:16:34,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2434, 1.8164, 1.4287, 0.4611], device='cuda:1'), covar=tensor([0.5581, 0.3454, 0.4526, 0.6764], device='cuda:1'), in_proj_covar=tensor([0.1810, 0.1703, 0.1636, 0.1469], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 04:16:35,859 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 04:16:53,789 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-13 04:16:59,131 INFO [optim.py:369] (1/2) 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 04:17:16,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3880, 3.3538, 1.4315, 1.6148], device='cuda:1'), covar=tensor([0.1059, 0.0257, 0.0965, 0.1368], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0554, 0.0397, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-13 04:17:16,799 INFO [train.py:968] (1/2) Epoch 25, batch 36050, giga_loss[loss=0.2606, simple_loss=0.3454, pruned_loss=0.08787, over 28954.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3477, pruned_loss=0.0994, over 5682617.39 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3438, pruned_loss=0.1076, over 5676395.69 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3464, pruned_loss=0.09805, over 5677163.01 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:17:46,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-13 04:17:50,404 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2226, 1.0656, 3.7997, 3.1431], device='cuda:1'), covar=tensor([0.1896, 0.3198, 0.0466, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0779, 0.0660, 0.0973, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 04:17:52,736 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1129934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:17:54,775 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1129937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:17:56,434 INFO [train.py:968] (1/2) Epoch 25, batch 36100, giga_loss[loss=0.2758, simple_loss=0.3596, pruned_loss=0.09602, over 28977.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3507, pruned_loss=0.09996, over 5701498.53 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3439, pruned_loss=0.1075, over 5682238.99 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3497, pruned_loss=0.09885, over 5692261.36 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:18:16,483 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1129966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:18:21,267 INFO [optim.py:369] (1/2) 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 04:18:35,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2948, 1.4429, 1.3318, 1.2082], device='cuda:1'), covar=tensor([0.2488, 0.2893, 0.1952, 0.2374], device='cuda:1'), in_proj_covar=tensor([0.1998, 0.1928, 0.1834, 0.1999], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:18:37,567 INFO [train.py:968] (1/2) Epoch 25, batch 36150, giga_loss[loss=0.2817, simple_loss=0.3651, pruned_loss=0.09918, over 28943.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3531, pruned_loss=0.1003, over 5692530.57 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3442, pruned_loss=0.1076, over 5674503.14 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3521, pruned_loss=0.09929, over 5691839.49 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:19:15,269 INFO [train.py:968] (1/2) Epoch 25, batch 36200, giga_loss[loss=0.2618, simple_loss=0.3509, pruned_loss=0.08634, over 29044.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3531, pruned_loss=0.09955, over 5693127.15 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3442, pruned_loss=0.1074, over 5674460.23 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3527, pruned_loss=0.09858, over 5693716.34 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:19:40,197 INFO [optim.py:369] (1/2) 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 04:19:54,459 INFO [train.py:968] (1/2) Epoch 25, batch 36250, giga_loss[loss=0.2995, simple_loss=0.3823, pruned_loss=0.1083, over 28556.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.354, pruned_loss=0.09929, over 5691367.03 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3448, pruned_loss=0.1076, over 5675129.90 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3534, pruned_loss=0.09819, over 5691140.62 frames. ], batch size: 78, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:20:13,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:20:24,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:20:26,256 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:20:33,200 INFO [train.py:968] (1/2) Epoch 25, batch 36300, giga_loss[loss=0.238, simple_loss=0.3035, pruned_loss=0.08625, over 23775.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3528, pruned_loss=0.09747, over 5687773.97 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3453, pruned_loss=0.1078, over 5668545.47 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.352, pruned_loss=0.0962, over 5694597.25 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:20:58,340 INFO [optim.py:369] (1/2) 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 04:21:04,785 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:21:11,556 INFO [train.py:968] (1/2) Epoch 25, batch 36350, giga_loss[loss=0.273, simple_loss=0.3504, pruned_loss=0.09774, over 28159.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3512, pruned_loss=0.09643, over 5676914.52 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3455, pruned_loss=0.1079, over 5655946.00 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3506, pruned_loss=0.09511, over 5693476.88 frames. ], batch size: 77, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:21:23,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-13 04:21:50,688 INFO [train.py:968] (1/2) Epoch 25, batch 36400, giga_loss[loss=0.3442, simple_loss=0.3983, pruned_loss=0.145, over 28999.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.354, pruned_loss=0.09927, over 5678359.30 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3469, pruned_loss=0.1084, over 5660031.33 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3525, pruned_loss=0.09729, over 5688979.32 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:22:07,396 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:22:09,302 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:22:20,346 INFO [optim.py:369] (1/2) 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 04:22:36,620 INFO [train.py:968] (1/2) Epoch 25, batch 36450, giga_loss[loss=0.3747, simple_loss=0.4084, pruned_loss=0.1705, over 27635.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3559, pruned_loss=0.1029, over 5666451.79 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3472, pruned_loss=0.1084, over 5649411.11 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3547, pruned_loss=0.101, over 5684420.67 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:22:36,873 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 36500, libri_loss[loss=0.3443, simple_loss=0.4005, pruned_loss=0.144, over 20159.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3567, pruned_loss=0.1052, over 5664250.64 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3475, pruned_loss=0.1086, over 5642833.41 frames. ], giga_tot_loss[loss=0.2813, simple_loss=0.3556, pruned_loss=0.1035, over 5685282.02 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:23:47,752 INFO [optim.py:369] (1/2) 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 04:24:03,305 INFO [train.py:968] (1/2) Epoch 25, batch 36550, giga_loss[loss=0.2585, simple_loss=0.3373, pruned_loss=0.08989, over 28956.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.356, pruned_loss=0.106, over 5674975.89 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3479, pruned_loss=0.1087, over 5652109.12 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.355, pruned_loss=0.1044, over 5684241.87 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:24:44,746 INFO [train.py:968] (1/2) Epoch 25, batch 36600, giga_loss[loss=0.2782, simple_loss=0.3571, pruned_loss=0.09963, over 28599.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3536, pruned_loss=0.1049, over 5687303.12 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3482, pruned_loss=0.1089, over 5652341.79 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3526, pruned_loss=0.1034, over 5694757.39 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:25:08,765 INFO [optim.py:369] (1/2) 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 04:25:25,893 INFO [train.py:968] (1/2) Epoch 25, batch 36650, giga_loss[loss=0.2776, simple_loss=0.3545, pruned_loss=0.1004, over 29073.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3512, pruned_loss=0.1033, over 5684274.40 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3483, pruned_loss=0.109, over 5645480.86 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3504, pruned_loss=0.1019, over 5697086.13 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:25:26,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3694, 1.5020, 1.4879, 1.3184], device='cuda:1'), covar=tensor([0.3357, 0.2850, 0.2390, 0.2868], device='cuda:1'), in_proj_covar=tensor([0.2012, 0.1943, 0.1853, 0.2017], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:25:37,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:25:40,311 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:26:07,230 INFO [train.py:968] (1/2) Epoch 25, batch 36700, giga_loss[loss=0.2675, simple_loss=0.3465, pruned_loss=0.09423, over 28928.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3505, pruned_loss=0.102, over 5683117.09 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3492, pruned_loss=0.1094, over 5648790.55 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3492, pruned_loss=0.1004, over 5691211.59 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:26:22,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:26:35,841 INFO [optim.py:369] (1/2) 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 04:26:49,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3291, 1.5149, 1.5839, 1.4605], device='cuda:1'), covar=tensor([0.1897, 0.1547, 0.1902, 0.1591], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0752, 0.0725, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 04:26:51,422 INFO [train.py:968] (1/2) Epoch 25, batch 36750, giga_loss[loss=0.2444, simple_loss=0.3213, pruned_loss=0.08378, over 28624.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3479, pruned_loss=0.1003, over 5670404.47 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3496, pruned_loss=0.1095, over 5637402.19 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3463, pruned_loss=0.09852, over 5688788.88 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:27:17,121 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:35,506 INFO [train.py:968] (1/2) Epoch 25, batch 36800, giga_loss[loss=0.2173, simple_loss=0.2972, pruned_loss=0.06869, over 28996.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3424, pruned_loss=0.09708, over 5682653.98 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.35, pruned_loss=0.1098, over 5641301.75 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3409, pruned_loss=0.09534, over 5694191.68 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:27:43,424 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:45,225 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:45,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:45,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:27:47,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:04,797 INFO [optim.py:369] (1/2) 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 04:28:05,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7168, 3.5421, 3.3317, 1.8724], device='cuda:1'), covar=tensor([0.0624, 0.0774, 0.0726, 0.1941], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.1160, 0.0972, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 04:28:09,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3775, 1.8332, 1.2745, 0.9234], device='cuda:1'), covar=tensor([0.6083, 0.3032, 0.3142, 0.5808], device='cuda:1'), in_proj_covar=tensor([0.1808, 0.1700, 0.1629, 0.1463], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 04:28:13,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1130678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:16,853 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1130681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:26,960 INFO [train.py:968] (1/2) Epoch 25, batch 36850, giga_loss[loss=0.2481, simple_loss=0.3237, pruned_loss=0.08626, over 29109.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3359, pruned_loss=0.09376, over 5672141.39 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.35, pruned_loss=0.1098, over 5643996.12 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3346, pruned_loss=0.09224, over 5679269.75 frames. ], batch size: 113, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:28:38,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1130699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:28:41,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1130702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:29:06,624 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1130731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:29:13,148 INFO [train.py:968] (1/2) Epoch 25, batch 36900, giga_loss[loss=0.2261, simple_loss=0.3096, pruned_loss=0.07127, over 28460.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3326, pruned_loss=0.09207, over 5667748.84 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3505, pruned_loss=0.1099, over 5645568.13 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3305, pruned_loss=0.09014, over 5672887.01 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:29:20,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0617, 3.9283, 3.7096, 1.7324], device='cuda:1'), covar=tensor([0.0593, 0.0687, 0.0635, 0.2233], device='cuda:1'), in_proj_covar=tensor([0.1246, 0.1154, 0.0967, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 04:29:44,823 INFO [optim.py:369] (1/2) 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 04:29:57,947 INFO [train.py:968] (1/2) Epoch 25, batch 36950, giga_loss[loss=0.2276, simple_loss=0.3149, pruned_loss=0.07017, over 28725.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3339, pruned_loss=0.0922, over 5666433.85 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3507, pruned_loss=0.1101, over 5638417.52 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3319, pruned_loss=0.09042, over 5676314.36 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:30:35,370 INFO [train.py:968] (1/2) Epoch 25, batch 37000, giga_loss[loss=0.2393, simple_loss=0.3216, pruned_loss=0.07853, over 28886.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3345, pruned_loss=0.09201, over 5685289.23 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3515, pruned_loss=0.1103, over 5645831.14 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3317, pruned_loss=0.08975, over 5687892.81 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:31:04,244 INFO [optim.py:369] (1/2) 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 04:31:17,794 INFO [train.py:968] (1/2) Epoch 25, batch 37050, giga_loss[loss=0.2157, simple_loss=0.2993, pruned_loss=0.06601, over 29022.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3332, pruned_loss=0.09175, over 5686813.77 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3514, pruned_loss=0.1101, over 5648547.70 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3309, pruned_loss=0.08995, over 5686704.36 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:31:45,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1729, 1.5715, 1.4778, 1.4231], device='cuda:1'), covar=tensor([0.2238, 0.1769, 0.2623, 0.2132], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0751, 0.0723, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 04:31:53,401 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:31:54,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5091, 3.5425, 1.5681, 1.6602], device='cuda:1'), covar=tensor([0.1051, 0.0346, 0.0934, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0554, 0.0396, 0.0433], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0025, 0.0030], device='cuda:1')
+2023-03-13 04:31:58,848 INFO [train.py:968] (1/2) Epoch 25, batch 37100, giga_loss[loss=0.2831, simple_loss=0.341, pruned_loss=0.1125, over 28958.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3309, pruned_loss=0.09064, over 5684997.32 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3517, pruned_loss=0.1102, over 5642023.72 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3287, pruned_loss=0.08899, over 5691698.54 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:32:05,527 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1130947.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:32:23,351 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 37150, giga_loss[loss=0.2459, simple_loss=0.3215, pruned_loss=0.0851, over 28752.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3297, pruned_loss=0.09001, over 5704871.76 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.352, pruned_loss=0.11, over 5654823.62 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3265, pruned_loss=0.08791, over 5700978.62 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:32:37,976 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-13 04:32:39,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3277, 1.4717, 1.3758, 1.5118], device='cuda:1'), covar=tensor([0.0774, 0.0374, 0.0341, 0.0841], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 04:32:39,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1130994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:33:03,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:33:14,639 INFO [train.py:968] (1/2) Epoch 25, batch 37200, giga_loss[loss=0.2523, simple_loss=0.3267, pruned_loss=0.08892, over 28595.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3269, pruned_loss=0.08871, over 5705505.42 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.352, pruned_loss=0.1099, over 5651543.92 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3239, pruned_loss=0.08667, over 5706418.41 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:33:16,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1739, 1.4573, 1.4987, 1.3368], device='cuda:1'), covar=tensor([0.2236, 0.1834, 0.2473, 0.2147], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0751, 0.0723, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 04:33:40,099 INFO [optim.py:369] (1/2) 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 04:33:49,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1131083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:33:53,965 INFO [train.py:968] (1/2) Epoch 25, batch 37250, giga_loss[loss=0.2713, simple_loss=0.3383, pruned_loss=0.1022, over 28895.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.326, pruned_loss=0.08863, over 5707485.80 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3521, pruned_loss=0.1098, over 5656698.47 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3231, pruned_loss=0.08677, over 5704542.74 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:33:54,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1131091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:23,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-13 04:34:34,308 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:35,915 INFO [train.py:968] (1/2) Epoch 25, batch 37300, giga_loss[loss=0.236, simple_loss=0.3115, pruned_loss=0.0802, over 28827.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3244, pruned_loss=0.08805, over 5709839.79 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3523, pruned_loss=0.1098, over 5659473.63 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3214, pruned_loss=0.08615, over 5705889.01 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:34:36,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:57,622 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:58,882 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:34:59,584 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:35:03,010 INFO [optim.py:369] (1/2) 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 04:35:14,350 INFO [train.py:968] (1/2) Epoch 25, batch 37350, giga_loss[loss=0.2438, simple_loss=0.3022, pruned_loss=0.09268, over 28611.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3231, pruned_loss=0.08746, over 5706752.05 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3528, pruned_loss=0.1099, over 5653743.29 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3195, pruned_loss=0.08529, over 5710260.20 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:35:19,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4470, 2.2821, 1.6215, 0.7407], device='cuda:1'), covar=tensor([0.5617, 0.3288, 0.4410, 0.5883], device='cuda:1'), in_proj_covar=tensor([0.1804, 0.1690, 0.1628, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 04:35:21,140 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:35:53,299 INFO [train.py:968] (1/2) Epoch 25, batch 37400, giga_loss[loss=0.2349, simple_loss=0.3023, pruned_loss=0.08378, over 28845.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3225, pruned_loss=0.08719, over 5702609.73 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.354, pruned_loss=0.1104, over 5640089.34 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3181, pruned_loss=0.08458, over 5718486.38 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:36:19,584 INFO [optim.py:369] (1/2) 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 04:36:33,231 INFO [train.py:968] (1/2) Epoch 25, batch 37450, giga_loss[loss=0.2319, simple_loss=0.3017, pruned_loss=0.08101, over 28532.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3224, pruned_loss=0.08706, over 5713030.21 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.355, pruned_loss=0.1109, over 5646081.46 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3168, pruned_loss=0.08377, over 5722420.51 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:36:43,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4759, 2.1742, 1.6539, 0.7346], device='cuda:1'), covar=tensor([0.7282, 0.3170, 0.4496, 0.7471], device='cuda:1'), in_proj_covar=tensor([0.1797, 0.1684, 0.1625, 0.1455], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 04:36:45,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:36:58,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131322.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:37:02,414 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-13 04:37:12,210 INFO [train.py:968] (1/2) Epoch 25, batch 37500, giga_loss[loss=0.2569, simple_loss=0.3246, pruned_loss=0.09464, over 28794.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3231, pruned_loss=0.08763, over 5714823.59 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3553, pruned_loss=0.1108, over 5652041.21 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3176, pruned_loss=0.08448, over 5718719.75 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:37:40,812 INFO [optim.py:369] (1/2) 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 04:37:53,212 INFO [train.py:968] (1/2) Epoch 25, batch 37550, giga_loss[loss=0.2669, simple_loss=0.3351, pruned_loss=0.09929, over 28545.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3279, pruned_loss=0.09047, over 5714322.06 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.356, pruned_loss=0.1109, over 5659971.13 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3218, pruned_loss=0.08704, over 5712700.12 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:38:08,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9002, 2.9569, 1.9321, 0.9814], device='cuda:1'), covar=tensor([0.8771, 0.2964, 0.4397, 0.7898], device='cuda:1'), in_proj_covar=tensor([0.1797, 0.1686, 0.1622, 0.1455], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 04:38:37,486 INFO [train.py:968] (1/2) Epoch 25, batch 37600, giga_loss[loss=0.317, simple_loss=0.3791, pruned_loss=0.1275, over 28588.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3327, pruned_loss=0.09343, over 5711993.37 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3556, pruned_loss=0.1107, over 5664493.32 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3277, pruned_loss=0.0906, over 5707496.27 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:38:49,030 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:38:51,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:38:55,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:39:03,288 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131465.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:39:03,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1131466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:39:06,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131468.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:39:10,344 INFO [optim.py:369] (1/2) 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 04:39:18,588 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:39:25,053 INFO [train.py:968] (1/2) Epoch 25, batch 37650, giga_loss[loss=0.3019, simple_loss=0.3689, pruned_loss=0.1174, over 27913.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3407, pruned_loss=0.09872, over 5698686.14 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3556, pruned_loss=0.1106, over 5668173.69 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3365, pruned_loss=0.09636, over 5692484.41 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:39:33,076 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131497.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:40:14,181 INFO [train.py:968] (1/2) Epoch 25, batch 37700, giga_loss[loss=0.2916, simple_loss=0.3655, pruned_loss=0.1089, over 29091.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3449, pruned_loss=0.1004, over 5674303.44 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3561, pruned_loss=0.1109, over 5659605.72 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3409, pruned_loss=0.09805, over 5677032.96 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:40:39,677 INFO [optim.py:369] (1/2) 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 04:40:48,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4265, 1.5952, 1.6819, 1.4007], device='cuda:1'), covar=tensor([0.2530, 0.2478, 0.1827, 0.2375], device='cuda:1'), in_proj_covar=tensor([0.2014, 0.1943, 0.1857, 0.2025], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:40:55,247 INFO [train.py:968] (1/2) Epoch 25, batch 37750, giga_loss[loss=0.2537, simple_loss=0.3389, pruned_loss=0.08427, over 28124.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3506, pruned_loss=0.1034, over 5672196.18 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3572, pruned_loss=0.1117, over 5657466.47 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3461, pruned_loss=0.1005, over 5676643.41 frames. ], batch size: 77, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:41:05,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:08,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:13,008 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1131609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:15,110 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1131612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:34,714 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:41:39,345 INFO [train.py:968] (1/2) Epoch 25, batch 37800, giga_loss[loss=0.3031, simple_loss=0.3757, pruned_loss=0.1152, over 28882.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3558, pruned_loss=0.1064, over 5671013.46 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3572, pruned_loss=0.1117, over 5660162.20 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3522, pruned_loss=0.1039, over 5672236.37 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:41:40,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1131641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:42:09,676 INFO [optim.py:369] (1/2) 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 04:42:14,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-13 04:42:20,851 INFO [train.py:968] (1/2) Epoch 25, batch 37850, giga_loss[loss=0.2504, simple_loss=0.3321, pruned_loss=0.08434, over 28804.00 frames. ], tot_loss[loss=0.2859, simple_loss=0.3574, pruned_loss=0.1071, over 5670766.02 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3575, pruned_loss=0.1119, over 5663393.42 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3543, pruned_loss=0.1049, over 5668932.66 frames. ], batch size: 112, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:42:52,250 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0896, 0.9930, 3.6865, 3.2621], device='cuda:1'), covar=tensor([0.1836, 0.3153, 0.0451, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0775, 0.0656, 0.0966, 0.0933], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 04:42:58,442 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4445, 1.8654, 1.5749, 1.5473], device='cuda:1'), covar=tensor([0.2268, 0.2403, 0.2427, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0753, 0.0725, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 04:43:02,930 INFO [train.py:968] (1/2) Epoch 25, batch 37900, giga_loss[loss=0.2525, simple_loss=0.3334, pruned_loss=0.08579, over 28950.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3526, pruned_loss=0.1028, over 5682671.61 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3573, pruned_loss=0.1119, over 5665866.39 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3502, pruned_loss=0.101, over 5679280.20 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:43:32,207 INFO [optim.py:369] (1/2) 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 04:43:41,694 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-13 04:43:45,068 INFO [train.py:968] (1/2) Epoch 25, batch 37950, giga_loss[loss=0.2677, simple_loss=0.3479, pruned_loss=0.09375, over 29039.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3516, pruned_loss=0.1017, over 5676492.73 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.358, pruned_loss=0.1124, over 5660223.39 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.349, pruned_loss=0.09958, over 5679960.32 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:44:23,990 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 38000, giga_loss[loss=0.2524, simple_loss=0.3387, pruned_loss=0.0831, over 28753.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3511, pruned_loss=0.1009, over 5680042.26 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3582, pruned_loss=0.1125, over 5660277.64 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3487, pruned_loss=0.09901, over 5683162.43 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:44:57,085 INFO [optim.py:369] (1/2) 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 04:45:00,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4206, 1.5399, 1.5513, 1.3557], device='cuda:1'), covar=tensor([0.3023, 0.2858, 0.2574, 0.2918], device='cuda:1'), in_proj_covar=tensor([0.2018, 0.1946, 0.1859, 0.2026], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:45:08,150 INFO [train.py:968] (1/2) Epoch 25, batch 38050, giga_loss[loss=0.3159, simple_loss=0.3819, pruned_loss=0.125, over 28680.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3523, pruned_loss=0.1016, over 5684439.23 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.358, pruned_loss=0.1125, over 5664977.38 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3504, pruned_loss=0.09988, over 5683364.57 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:45:24,611 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 38100, giga_loss[loss=0.2778, simple_loss=0.356, pruned_loss=0.09978, over 29093.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3551, pruned_loss=0.1038, over 5684630.67 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3589, pruned_loss=0.1131, over 5666664.70 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3527, pruned_loss=0.1018, over 5682610.12 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:46:20,098 INFO [optim.py:369] (1/2) 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 04:46:33,046 INFO [train.py:968] (1/2) Epoch 25, batch 38150, giga_loss[loss=0.295, simple_loss=0.3688, pruned_loss=0.1106, over 29003.00 frames. ], tot_loss[loss=0.2824, simple_loss=0.3559, pruned_loss=0.1045, over 5680359.89 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3592, pruned_loss=0.1132, over 5657014.37 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3536, pruned_loss=0.1025, over 5688895.68 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:47:19,305 INFO [train.py:968] (1/2) Epoch 25, batch 38200, giga_loss[loss=0.2924, simple_loss=0.358, pruned_loss=0.1134, over 28851.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3557, pruned_loss=0.1049, over 5677030.43 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3594, pruned_loss=0.1133, over 5654266.30 frames. ], giga_tot_loss[loss=0.28, simple_loss=0.3538, pruned_loss=0.1032, over 5686153.73 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:47:35,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4986, 1.8015, 1.4890, 1.4177], device='cuda:1'), covar=tensor([0.2430, 0.2363, 0.2552, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1127, 0.1382, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 04:47:47,953 INFO [optim.py:369] (1/2) 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 04:47:59,756 INFO [train.py:968] (1/2) Epoch 25, batch 38250, giga_loss[loss=0.2724, simple_loss=0.3467, pruned_loss=0.099, over 28536.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3568, pruned_loss=0.1058, over 5689221.58 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5656888.30 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3548, pruned_loss=0.1041, over 5694337.23 frames. ], batch size: 78, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:48:41,236 INFO [train.py:968] (1/2) Epoch 25, batch 38300, giga_loss[loss=0.2478, simple_loss=0.3344, pruned_loss=0.0806, over 28808.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3558, pruned_loss=0.1042, over 5697642.81 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3594, pruned_loss=0.1134, over 5660668.43 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3546, pruned_loss=0.103, over 5698690.95 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:49:08,569 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:49:10,817 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 04:49:11,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4745, 1.2088, 4.2909, 3.4419], device='cuda:1'), covar=tensor([0.1471, 0.2687, 0.0458, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0655, 0.0969, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 04:49:11,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-13 04:49:17,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6640, 1.4818, 4.8818, 3.6690], device='cuda:1'), covar=tensor([0.1628, 0.2887, 0.0355, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0655, 0.0969, 0.0936], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 04:49:21,524 INFO [train.py:968] (1/2) Epoch 25, batch 38350, giga_loss[loss=0.2449, simple_loss=0.3311, pruned_loss=0.07937, over 28564.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.356, pruned_loss=0.1032, over 5705250.39 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1139, over 5665619.01 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3541, pruned_loss=0.1014, over 5702555.50 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:49:39,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:50:01,782 INFO [train.py:968] (1/2) Epoch 25, batch 38400, giga_loss[loss=0.2495, simple_loss=0.3336, pruned_loss=0.08266, over 29009.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3555, pruned_loss=0.1021, over 5705626.13 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3602, pruned_loss=0.1138, over 5666906.92 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3541, pruned_loss=0.1007, over 5702598.14 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 04:50:07,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4724, 4.3117, 4.0832, 1.8942], device='cuda:1'), covar=tensor([0.0536, 0.0717, 0.0699, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.1159, 0.0972, 0.0731], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 04:50:24,075 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0392, 3.2114, 1.9010, 1.1002], device='cuda:1'), covar=tensor([0.8176, 0.3601, 0.4611, 0.6950], device='cuda:1'), in_proj_covar=tensor([0.1804, 0.1687, 0.1627, 0.1463], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 04:50:28,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3506, 1.5334, 1.5516, 1.3989], device='cuda:1'), covar=tensor([0.1691, 0.1783, 0.2090, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0490, 0.0753, 0.0725, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 04:50:32,229 INFO [optim.py:369] (1/2) 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 04:50:39,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132285.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:50:41,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3027, 2.7863, 1.8633, 2.2166], device='cuda:1'), covar=tensor([0.0974, 0.0608, 0.1033, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0449, 0.0523, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 04:50:42,168 INFO [train.py:968] (1/2) Epoch 25, batch 38450, giga_loss[loss=0.2358, simple_loss=0.3207, pruned_loss=0.07548, over 28466.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3532, pruned_loss=0.1012, over 5688134.79 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1136, over 5651637.50 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.352, pruned_loss=0.09964, over 5701150.12 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:51:14,245 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 04:51:18,419 INFO [train.py:968] (1/2) Epoch 25, batch 38500, giga_loss[loss=0.2764, simple_loss=0.3481, pruned_loss=0.1024, over 28014.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3528, pruned_loss=0.1016, over 5689523.29 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3608, pruned_loss=0.1144, over 5647440.14 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3507, pruned_loss=0.09911, over 5705628.41 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:51:27,269 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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 04:51:52,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1132385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:51:55,611 INFO [train.py:968] (1/2) Epoch 25, batch 38550, libri_loss[loss=0.3505, simple_loss=0.3969, pruned_loss=0.152, over 19606.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1005, over 5686209.55 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.361, pruned_loss=0.1146, over 5636092.51 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3485, pruned_loss=0.09789, over 5711654.60 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:52:09,866 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6057, 1.4975, 1.5608, 1.2225], device='cuda:1'), covar=tensor([0.1999, 0.3345, 0.1585, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0712, 0.0971, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 04:52:14,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2752, 2.3121, 2.4329, 1.9030], device='cuda:1'), covar=tensor([0.2977, 0.2800, 0.2561, 0.3272], device='cuda:1'), in_proj_covar=tensor([0.2008, 0.1943, 0.1852, 0.2023], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 04:52:24,716 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1132428.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:52:28,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1132431.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 04:52:33,003 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:52:35,522 INFO [train.py:968] (1/2) Epoch 25, batch 38600, giga_loss[loss=0.3165, simple_loss=0.3808, pruned_loss=0.1261, over 28646.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3505, pruned_loss=0.1009, over 5694230.88 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.361, pruned_loss=0.1145, over 5643987.79 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3484, pruned_loss=0.09841, over 5709419.33 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:52:45,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4201, 1.6476, 1.2518, 1.2865], device='cuda:1'), covar=tensor([0.0918, 0.0476, 0.1017, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0449, 0.0523, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 04:52:51,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1132460.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 04:53:06,694 INFO [optim.py:369] (1/2) 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 04:53:15,868 INFO [train.py:968] (1/2) Epoch 25, batch 38650, giga_loss[loss=0.2876, simple_loss=0.3579, pruned_loss=0.1087, over 28604.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3525, pruned_loss=0.1027, over 5702232.47 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3618, pruned_loss=0.115, over 5648400.71 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3499, pruned_loss=0.1001, over 5711680.98 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:53:53,690 INFO [train.py:968] (1/2) Epoch 25, batch 38700, giga_loss[loss=0.2565, simple_loss=0.3424, pruned_loss=0.0853, over 28875.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3512, pruned_loss=0.1011, over 5707210.24 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3618, pruned_loss=0.1149, over 5649817.09 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3491, pruned_loss=0.09889, over 5713829.19 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:53:59,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-13 04:54:00,026 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:54:23,399 INFO [optim.py:369] (1/2) 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 04:54:31,175 INFO [train.py:968] (1/2) Epoch 25, batch 38750, giga_loss[loss=0.2579, simple_loss=0.3326, pruned_loss=0.09166, over 28627.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3498, pruned_loss=0.09954, over 5701727.75 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3616, pruned_loss=0.1148, over 5649080.67 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.348, pruned_loss=0.09749, over 5708903.81 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 04:55:09,818 INFO [train.py:968] (1/2) Epoch 25, batch 38800, giga_loss[loss=0.2531, simple_loss=0.3343, pruned_loss=0.08593, over 28952.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3483, pruned_loss=0.09838, over 5703298.65 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3615, pruned_loss=0.1147, over 5643283.58 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3466, pruned_loss=0.0964, over 5714746.10 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:55:19,482 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-13 04:55:41,358 INFO [optim.py:369] (1/2) 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 04:55:51,470 INFO [train.py:968] (1/2) Epoch 25, batch 38850, giga_loss[loss=0.2363, simple_loss=0.3201, pruned_loss=0.07627, over 29087.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3463, pruned_loss=0.09748, over 5705399.33 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3614, pruned_loss=0.1146, over 5650922.51 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3447, pruned_loss=0.09548, over 5708959.88 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:55:52,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1132691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:55:54,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1132694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:56:15,884 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1132723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:56:29,211 INFO [train.py:968] (1/2) Epoch 25, batch 38900, giga_loss[loss=0.2407, simple_loss=0.3217, pruned_loss=0.07982, over 28979.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3438, pruned_loss=0.09638, over 5705796.45 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3612, pruned_loss=0.1144, over 5655534.42 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3423, pruned_loss=0.09456, over 5705926.42 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:57:00,067 INFO [optim.py:369] (1/2) 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 04:57:11,459 INFO [train.py:968] (1/2) Epoch 25, batch 38950, giga_loss[loss=0.2242, simple_loss=0.3057, pruned_loss=0.07131, over 28881.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3412, pruned_loss=0.09529, over 5707018.01 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3613, pruned_loss=0.1145, over 5658154.82 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3397, pruned_loss=0.0936, over 5705234.68 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:57:26,407 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1132811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:57:49,963 INFO [train.py:968] (1/2) Epoch 25, batch 39000, giga_loss[loss=0.2254, simple_loss=0.3083, pruned_loss=0.07126, over 28640.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3405, pruned_loss=0.09469, over 5710564.41 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3611, pruned_loss=0.1143, over 5661578.59 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3392, pruned_loss=0.09321, over 5706931.92 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:57:49,964 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 04:57:58,920 INFO [train.py:1012] (1/2) Epoch 25, validation: loss=0.2055, simple_loss=0.3136, pruned_loss=0.0487, over 944034.00 frames. 
+2023-03-13 04:57:58,920 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 04:58:24,514 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:58:26,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 04:58:27,041 INFO [optim.py:369] (1/2) 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 04:58:36,682 INFO [train.py:968] (1/2) Epoch 25, batch 39050, giga_loss[loss=0.2611, simple_loss=0.3335, pruned_loss=0.09435, over 28588.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3416, pruned_loss=0.09644, over 5705553.47 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3618, pruned_loss=0.1149, over 5665538.83 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3392, pruned_loss=0.09414, over 5700282.05 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:58:39,813 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:58:47,092 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1132902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:59:15,244 INFO [train.py:968] (1/2) Epoch 25, batch 39100, giga_loss[loss=0.3017, simple_loss=0.3604, pruned_loss=0.1215, over 26797.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3385, pruned_loss=0.09467, over 5706126.01 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3615, pruned_loss=0.1145, over 5666444.93 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3365, pruned_loss=0.0928, over 5701948.66 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 04:59:28,091 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1132954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:59:30,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1132957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 04:59:44,580 INFO [optim.py:369] (1/2) 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 04:59:51,987 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 39150, giga_loss[loss=0.2366, simple_loss=0.3148, pruned_loss=0.07919, over 28911.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3354, pruned_loss=0.09329, over 5705657.17 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1147, over 5660463.05 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3333, pruned_loss=0.09145, over 5709290.83 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:00:34,178 INFO [train.py:968] (1/2) Epoch 25, batch 39200, giga_loss[loss=0.3091, simple_loss=0.3842, pruned_loss=0.117, over 28591.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3346, pruned_loss=0.09318, over 5700727.15 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.3622, pruned_loss=0.1148, over 5665351.32 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3319, pruned_loss=0.09104, over 5700072.69 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:00:34,743 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 05:01:05,758 INFO [optim.py:369] (1/2) 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 05:01:14,597 INFO [train.py:968] (1/2) Epoch 25, batch 39250, giga_loss[loss=0.2193, simple_loss=0.2917, pruned_loss=0.07347, over 28576.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3333, pruned_loss=0.09207, over 5709311.44 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3622, pruned_loss=0.1146, over 5670198.38 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3302, pruned_loss=0.08983, over 5705457.24 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:01:25,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6814, 4.5001, 4.3172, 2.0139], device='cuda:1'), covar=tensor([0.0660, 0.0800, 0.1002, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1162, 0.0977, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 05:01:43,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.00 vs. limit=5.0
+2023-03-13 05:01:52,784 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133137.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:01:54,605 INFO [train.py:968] (1/2) Epoch 25, batch 39300, giga_loss[loss=0.2417, simple_loss=0.327, pruned_loss=0.07819, over 28894.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3353, pruned_loss=0.09271, over 5710141.04 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5678548.60 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3321, pruned_loss=0.09038, over 5701130.81 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:02:03,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6316, 1.9855, 1.6533, 1.9816], device='cuda:1'), covar=tensor([0.0730, 0.0265, 0.0321, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 05:02:14,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2726, 3.2647, 1.4434, 1.3726], device='cuda:1'), covar=tensor([0.1038, 0.0283, 0.1004, 0.1436], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0555, 0.0396, 0.0434], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0025, 0.0030], device='cuda:1')
+2023-03-13 05:02:27,102 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 39350, giga_loss[loss=0.2939, simple_loss=0.3706, pruned_loss=0.1086, over 28646.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3395, pruned_loss=0.09551, over 5686072.92 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3623, pruned_loss=0.1149, over 5662697.49 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.336, pruned_loss=0.09268, over 5693684.24 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:03:17,941 INFO [train.py:968] (1/2) Epoch 25, batch 39400, giga_loss[loss=0.246, simple_loss=0.3258, pruned_loss=0.08312, over 28697.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.341, pruned_loss=0.09539, over 5691219.97 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3618, pruned_loss=0.1146, over 5668993.36 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3381, pruned_loss=0.09292, over 5692139.00 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:03:26,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:03:42,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:03:49,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:03:50,569 INFO [optim.py:369] (1/2) 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 05:03:59,514 INFO [train.py:968] (1/2) Epoch 25, batch 39450, libri_loss[loss=0.2781, simple_loss=0.3526, pruned_loss=0.1018, over 29538.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3417, pruned_loss=0.09513, over 5694287.79 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1144, over 5674556.86 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.339, pruned_loss=0.09277, over 5690529.26 frames. ], batch size: 83, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:04:39,763 INFO [train.py:968] (1/2) Epoch 25, batch 39500, giga_loss[loss=0.2157, simple_loss=0.2964, pruned_loss=0.06751, over 28638.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3415, pruned_loss=0.09487, over 5700133.37 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.362, pruned_loss=0.1147, over 5676157.20 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3381, pruned_loss=0.09191, over 5696477.69 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:04:56,598 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9311, 1.9672, 1.7193, 2.1702], device='cuda:1'), covar=tensor([0.2593, 0.2851, 0.3207, 0.2559], device='cuda:1'), in_proj_covar=tensor([0.1564, 0.1127, 0.1381, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 05:05:08,691 INFO [optim.py:369] (1/2) 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 05:05:08,950 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 39550, giga_loss[loss=0.2553, simple_loss=0.3301, pruned_loss=0.09027, over 28456.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3416, pruned_loss=0.09534, over 5694588.94 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3622, pruned_loss=0.1149, over 5673186.33 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3377, pruned_loss=0.09191, over 5694825.73 frames. ], batch size: 65, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:05:17,791 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2710, 1.1248, 3.5951, 3.1266], device='cuda:1'), covar=tensor([0.1643, 0.2947, 0.0480, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0776, 0.0656, 0.0969, 0.0939], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 05:05:19,188 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:21,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:34,419 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:36,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:39,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:41,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:43,891 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:05:56,394 INFO [train.py:968] (1/2) Epoch 25, batch 39600, giga_loss[loss=0.2468, simple_loss=0.3235, pruned_loss=0.08501, over 28965.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3415, pruned_loss=0.09552, over 5702695.00 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3617, pruned_loss=0.1147, over 5670159.15 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3384, pruned_loss=0.09241, over 5706203.47 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:05:59,438 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:06:04,474 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:06:07,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4697, 1.8056, 1.4304, 1.5392], device='cuda:1'), covar=tensor([0.2756, 0.2760, 0.3251, 0.2493], device='cuda:1'), in_proj_covar=tensor([0.1563, 0.1127, 0.1381, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 05:06:28,029 INFO [optim.py:369] (1/2) 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 05:06:36,755 INFO [train.py:968] (1/2) Epoch 25, batch 39650, giga_loss[loss=0.2852, simple_loss=0.3645, pruned_loss=0.103, over 28917.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3424, pruned_loss=0.09587, over 5701666.78 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3619, pruned_loss=0.1149, over 5664310.32 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3395, pruned_loss=0.09306, over 5709547.40 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:06:43,499 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7204, 1.5919, 1.8915, 1.4159], device='cuda:1'), covar=tensor([0.2039, 0.3329, 0.1576, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0712, 0.0972, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 05:06:54,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133512.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:07:17,473 INFO [train.py:968] (1/2) Epoch 25, batch 39700, giga_loss[loss=0.2482, simple_loss=0.3346, pruned_loss=0.08088, over 29063.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3481, pruned_loss=0.09901, over 5710252.45 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3626, pruned_loss=0.1153, over 5673958.43 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3443, pruned_loss=0.0956, over 5709399.64 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:07:46,858 INFO [optim.py:369] (1/2) 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 05:07:53,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2420, 1.4861, 1.4243, 1.1949], device='cuda:1'), covar=tensor([0.3002, 0.2689, 0.1767, 0.2643], device='cuda:1'), in_proj_covar=tensor([0.2031, 0.1961, 0.1881, 0.2040], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 05:07:55,613 INFO [train.py:968] (1/2) Epoch 25, batch 39750, giga_loss[loss=0.272, simple_loss=0.3494, pruned_loss=0.09731, over 29022.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3498, pruned_loss=0.0998, over 5701047.61 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3629, pruned_loss=0.1156, over 5664194.45 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3463, pruned_loss=0.09657, over 5710252.62 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:08:32,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3927, 1.6618, 1.4056, 1.6512], device='cuda:1'), covar=tensor([0.0759, 0.0309, 0.0334, 0.0855], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 05:08:36,327 INFO [train.py:968] (1/2) Epoch 25, batch 39800, giga_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1012, over 28968.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.09965, over 5703474.89 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1155, over 5667591.71 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3472, pruned_loss=0.09697, over 5708120.34 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:08:48,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133655.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:08:51,137 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133658.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:09:10,071 INFO [optim.py:369] (1/2) 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 05:09:15,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133687.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:09:17,032 INFO [train.py:968] (1/2) Epoch 25, batch 39850, libri_loss[loss=0.3259, simple_loss=0.3878, pruned_loss=0.132, over 29632.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3527, pruned_loss=0.1013, over 5708277.87 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3627, pruned_loss=0.1153, over 5673371.93 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3503, pruned_loss=0.09903, over 5707464.34 frames. ], batch size: 91, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:09:56,742 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 39900, giga_loss[loss=0.2628, simple_loss=0.3391, pruned_loss=0.09329, over 28504.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3529, pruned_loss=0.1017, over 5696963.05 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1152, over 5666723.56 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3508, pruned_loss=0.0995, over 5702821.89 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:10:09,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1133754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:10:19,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6173, 1.7082, 1.8343, 1.3956], device='cuda:1'), covar=tensor([0.1747, 0.2542, 0.1505, 0.1685], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0710, 0.0969, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 05:10:27,780 INFO [optim.py:369] (1/2) 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 05:10:28,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-13 05:10:35,161 INFO [train.py:968] (1/2) Epoch 25, batch 39950, giga_loss[loss=0.2453, simple_loss=0.3332, pruned_loss=0.07872, over 29117.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3514, pruned_loss=0.1008, over 5707421.62 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3625, pruned_loss=0.1151, over 5669145.36 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3497, pruned_loss=0.09902, over 5710105.44 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:10:36,295 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:11:12,755 INFO [train.py:968] (1/2) Epoch 25, batch 40000, giga_loss[loss=0.2511, simple_loss=0.3322, pruned_loss=0.08501, over 29119.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3484, pruned_loss=0.09957, over 5704977.75 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1153, over 5662074.25 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3463, pruned_loss=0.0974, over 5713952.13 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:11:45,690 INFO [optim.py:369] (1/2) 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 05:11:53,320 INFO [train.py:968] (1/2) Epoch 25, batch 40050, giga_loss[loss=0.2333, simple_loss=0.3136, pruned_loss=0.07644, over 29006.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.346, pruned_loss=0.09881, over 5687117.83 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5648612.49 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3434, pruned_loss=0.09614, over 5708006.74 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:11:59,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6331, 1.9169, 1.2482, 1.4880], device='cuda:1'), covar=tensor([0.1043, 0.0656, 0.1137, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0449, 0.0519, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 05:12:00,381 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1133897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:02,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1133900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:02,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2726, 1.0915, 3.7259, 3.2659], device='cuda:1'), covar=tensor([0.1571, 0.2715, 0.0507, 0.0976], device='cuda:1'), in_proj_covar=tensor([0.0777, 0.0656, 0.0970, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 05:12:20,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:23,082 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1133929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:30,814 INFO [train.py:968] (1/2) Epoch 25, batch 40100, giga_loss[loss=0.2555, simple_loss=0.3451, pruned_loss=0.0829, over 28884.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3461, pruned_loss=0.09849, over 5698548.64 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5652879.05 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3427, pruned_loss=0.09539, over 5713356.01 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:12:32,260 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:12:35,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4671, 1.8461, 1.6125, 1.5839], device='cuda:1'), covar=tensor([0.2308, 0.2653, 0.2542, 0.2658], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0756, 0.0730, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 05:12:40,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 05:13:02,131 INFO [optim.py:369] (1/2) 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 05:13:11,699 INFO [train.py:968] (1/2) Epoch 25, batch 40150, giga_loss[loss=0.2894, simple_loss=0.3716, pruned_loss=0.1037, over 28300.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.09781, over 5700204.53 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3636, pruned_loss=0.1159, over 5652538.59 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3456, pruned_loss=0.0952, over 5713067.49 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:13:20,560 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1133998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:13:32,067 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9680, 1.3227, 1.0743, 0.1967], device='cuda:1'), covar=tensor([0.4226, 0.3005, 0.4395, 0.7017], device='cuda:1'), in_proj_covar=tensor([0.1796, 0.1683, 0.1626, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 05:13:53,232 INFO [train.py:968] (1/2) Epoch 25, batch 40200, giga_loss[loss=0.2492, simple_loss=0.3207, pruned_loss=0.08889, over 28424.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.09778, over 5688915.76 frames. ], libri_tot_loss[loss=0.2979, simple_loss=0.3637, pruned_loss=0.116, over 5647335.89 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3457, pruned_loss=0.09529, over 5704933.96 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:14:26,177 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 40250, libri_loss[loss=0.3045, simple_loss=0.3752, pruned_loss=0.1169, over 28682.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3471, pruned_loss=0.09828, over 5700054.95 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3636, pruned_loss=0.1159, over 5652777.25 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3449, pruned_loss=0.09602, over 5709039.83 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:14:53,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:15:13,192 INFO [train.py:968] (1/2) Epoch 25, batch 40300, giga_loss[loss=0.2417, simple_loss=0.3209, pruned_loss=0.0812, over 28931.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3454, pruned_loss=0.09827, over 5708515.86 frames. ], libri_tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1159, over 5660108.47 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3432, pruned_loss=0.09595, over 5710289.39 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:15:34,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:15:46,941 INFO [optim.py:369] (1/2) 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 05:15:53,604 INFO [train.py:968] (1/2) Epoch 25, batch 40350, giga_loss[loss=0.2985, simple_loss=0.3577, pruned_loss=0.1197, over 28744.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3451, pruned_loss=0.09959, over 5691701.85 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3637, pruned_loss=0.1162, over 5648860.64 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3426, pruned_loss=0.0969, over 5704334.57 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:16:17,544 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7173, 1.9162, 1.9276, 1.6793], device='cuda:1'), covar=tensor([0.1759, 0.1986, 0.1984, 0.1903], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0754, 0.0725, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 05:16:26,896 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6706, 1.9050, 1.4026, 1.4963], device='cuda:1'), covar=tensor([0.0883, 0.0461, 0.0940, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0449, 0.0520, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 05:16:33,904 INFO [train.py:968] (1/2) Epoch 25, batch 40400, giga_loss[loss=0.3225, simple_loss=0.3794, pruned_loss=0.1328, over 27578.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3434, pruned_loss=0.09925, over 5704807.41 frames. ], libri_tot_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.1161, over 5653570.29 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3413, pruned_loss=0.0969, over 5711352.89 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:16:47,675 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:16:49,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:16:57,771 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3445, 3.1828, 3.0500, 1.2614], device='cuda:1'), covar=tensor([0.0966, 0.1114, 0.0946, 0.2352], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1160, 0.0977, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 05:17:07,471 INFO [optim.py:369] (1/2) 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 05:17:14,760 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:15,084 INFO [train.py:968] (1/2) Epoch 25, batch 40450, giga_loss[loss=0.2471, simple_loss=0.3289, pruned_loss=0.08266, over 28947.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3419, pruned_loss=0.09838, over 5713827.45 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3636, pruned_loss=0.1162, over 5658270.55 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3397, pruned_loss=0.0961, over 5715918.58 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:17:22,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:24,205 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1134302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:29,370 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:31,273 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:37,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:17:41,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5583, 1.7966, 1.4203, 1.7061], device='cuda:1'), covar=tensor([0.0682, 0.0313, 0.0321, 0.0789], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 05:17:53,387 INFO [train.py:968] (1/2) Epoch 25, batch 40500, libri_loss[loss=0.33, simple_loss=0.3959, pruned_loss=0.1321, over 29280.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3385, pruned_loss=0.09656, over 5715523.75 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3637, pruned_loss=0.1162, over 5661197.07 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3359, pruned_loss=0.0941, over 5716318.29 frames. ], batch size: 94, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:17:54,367 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:18:05,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-13 05:18:12,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2799, 0.9178, 1.0300, 1.3030], device='cuda:1'), covar=tensor([0.0704, 0.0355, 0.0322, 0.0830], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 05:18:18,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:18:24,441 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1134381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:18:24,811 INFO [optim.py:369] (1/2) 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 05:18:29,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4331, 2.0553, 1.5150, 0.7335], device='cuda:1'), covar=tensor([0.6521, 0.2900, 0.4724, 0.7429], device='cuda:1'), in_proj_covar=tensor([0.1791, 0.1678, 0.1623, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 05:18:32,408 INFO [train.py:968] (1/2) Epoch 25, batch 40550, giga_loss[loss=0.2443, simple_loss=0.3204, pruned_loss=0.08406, over 28910.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3339, pruned_loss=0.09409, over 5722607.37 frames. ], libri_tot_loss[loss=0.2986, simple_loss=0.3642, pruned_loss=0.1165, over 5666291.06 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3307, pruned_loss=0.09139, over 5719954.97 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:19:13,311 INFO [train.py:968] (1/2) Epoch 25, batch 40600, giga_loss[loss=0.2745, simple_loss=0.3513, pruned_loss=0.09879, over 28926.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3322, pruned_loss=0.09316, over 5716524.39 frames. ], libri_tot_loss[loss=0.2989, simple_loss=0.3645, pruned_loss=0.1167, over 5668480.56 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3292, pruned_loss=0.09066, over 5712914.76 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:19:16,055 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:18,003 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:29,061 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:31,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:19:41,234 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 25, batch 40650, giga_loss[loss=0.3255, simple_loss=0.387, pruned_loss=0.132, over 28853.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3346, pruned_loss=0.09405, over 5718063.27 frames. ], libri_tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5675376.15 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3319, pruned_loss=0.09189, over 5710226.03 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:19:54,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:20:15,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:20:18,379 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:20:35,115 INFO [train.py:968] (1/2) Epoch 25, batch 40700, giga_loss[loss=0.2759, simple_loss=0.3455, pruned_loss=0.1031, over 28458.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3368, pruned_loss=0.09444, over 5720563.95 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3633, pruned_loss=0.1158, over 5679830.95 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3346, pruned_loss=0.09261, over 5710897.53 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:20:35,472 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3009, 1.4018, 1.2851, 1.4744], device='cuda:1'), covar=tensor([0.0762, 0.0359, 0.0353, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 05:20:42,607 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:21:07,848 INFO [optim.py:369] (1/2) 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 05:21:13,681 INFO [train.py:968] (1/2) Epoch 25, batch 40750, giga_loss[loss=0.2749, simple_loss=0.3504, pruned_loss=0.0997, over 28687.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3399, pruned_loss=0.09552, over 5723037.53 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3633, pruned_loss=0.1158, over 5686168.32 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3372, pruned_loss=0.09329, over 5710743.54 frames. ], batch size: 99, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:21:37,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2963, 2.4967, 1.3181, 1.4263], device='cuda:1'), covar=tensor([0.0948, 0.0393, 0.0947, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0558, 0.0398, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 05:21:51,029 INFO [train.py:968] (1/2) Epoch 25, batch 40800, giga_loss[loss=0.2719, simple_loss=0.3488, pruned_loss=0.09753, over 28964.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3421, pruned_loss=0.09623, over 5730551.18 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3629, pruned_loss=0.1155, over 5690458.97 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3396, pruned_loss=0.09403, over 5717770.49 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:22:12,565 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 05:22:20,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1134677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:22:25,026 INFO [optim.py:369] (1/2) 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 05:22:33,178 INFO [train.py:968] (1/2) Epoch 25, batch 40850, giga_loss[loss=0.252, simple_loss=0.3365, pruned_loss=0.08374, over 28917.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3449, pruned_loss=0.09798, over 5723376.31 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.363, pruned_loss=0.1157, over 5691254.04 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3423, pruned_loss=0.09558, over 5713124.21 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:22:37,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.64 vs. limit=2.0
+2023-03-13 05:23:18,187 INFO [train.py:968] (1/2) Epoch 25, batch 40900, giga_loss[loss=0.4487, simple_loss=0.473, pruned_loss=0.2122, over 27613.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.348, pruned_loss=0.1006, over 5708429.18 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5684819.09 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3455, pruned_loss=0.09845, over 5705743.90 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:23:37,474 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:24:02,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6858, 1.9140, 1.6243, 1.7979], device='cuda:1'), covar=tensor([0.2176, 0.2231, 0.2297, 0.2198], device='cuda:1'), in_proj_covar=tensor([0.1568, 0.1132, 0.1385, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 05:24:04,713 INFO [train.py:968] (1/2) Epoch 25, batch 40950, giga_loss[loss=0.3375, simple_loss=0.3956, pruned_loss=0.1397, over 28742.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3551, pruned_loss=0.1068, over 5692213.03 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3622, pruned_loss=0.115, over 5693098.07 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3535, pruned_loss=0.1053, over 5682960.39 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:24:09,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9562, 1.1662, 1.0553, 0.9672], device='cuda:1'), covar=tensor([0.2260, 0.2658, 0.1707, 0.2043], device='cuda:1'), in_proj_covar=tensor([0.2040, 0.1970, 0.1890, 0.2044], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 05:24:30,236 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:24:32,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:24:46,678 INFO [train.py:968] (1/2) Epoch 25, batch 41000, giga_loss[loss=0.3678, simple_loss=0.3982, pruned_loss=0.1687, over 23729.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3601, pruned_loss=0.1105, over 5686025.64 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3617, pruned_loss=0.1147, over 5690940.10 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.3592, pruned_loss=0.1093, over 5680420.09 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:24:57,119 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:25:22,094 INFO [optim.py:369] (1/2) 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 05:25:27,508 INFO [train.py:968] (1/2) Epoch 25, batch 41050, giga_loss[loss=0.3925, simple_loss=0.4368, pruned_loss=0.1741, over 28622.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3658, pruned_loss=0.1149, over 5681281.39 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3616, pruned_loss=0.1145, over 5697120.81 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3652, pruned_loss=0.114, over 5670748.52 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:25:36,976 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1134899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:25:38,862 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1134902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:26:03,666 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1134931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:26:08,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3599, 2.8512, 1.4101, 1.4999], device='cuda:1'), covar=tensor([0.0963, 0.0396, 0.0878, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0558, 0.0397, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 05:26:10,623 INFO [train.py:968] (1/2) Epoch 25, batch 41100, giga_loss[loss=0.433, simple_loss=0.4638, pruned_loss=0.2011, over 28231.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3729, pruned_loss=0.1209, over 5669252.01 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5686801.11 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3726, pruned_loss=0.1204, over 5670487.68 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:26:48,674 INFO [optim.py:369] (1/2) 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 05:26:54,769 INFO [train.py:968] (1/2) Epoch 25, batch 41150, giga_loss[loss=0.3144, simple_loss=0.3782, pruned_loss=0.1253, over 29062.00 frames. ], tot_loss[loss=0.314, simple_loss=0.378, pruned_loss=0.125, over 5671532.90 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3612, pruned_loss=0.1141, over 5690350.24 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3784, pruned_loss=0.1249, over 5669037.65 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:27:44,630 INFO [train.py:968] (1/2) Epoch 25, batch 41200, giga_loss[loss=0.2912, simple_loss=0.3636, pruned_loss=0.1094, over 28859.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3794, pruned_loss=0.1269, over 5666133.87 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3608, pruned_loss=0.1136, over 5696210.40 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.3806, pruned_loss=0.1277, over 5658012.01 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:27:47,600 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1135042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:28:01,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1529, 1.3133, 1.0904, 0.9237], device='cuda:1'), covar=tensor([0.0923, 0.0438, 0.0987, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0452, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 05:28:33,269 INFO [optim.py:369] (1/2) 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 05:28:39,251 INFO [train.py:968] (1/2) Epoch 25, batch 41250, giga_loss[loss=0.4047, simple_loss=0.418, pruned_loss=0.1958, over 23426.00 frames. ], tot_loss[loss=0.3207, simple_loss=0.3817, pruned_loss=0.1298, over 5637647.51 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.361, pruned_loss=0.1137, over 5692938.73 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3827, pruned_loss=0.1306, over 5633534.68 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:28:46,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-13 05:29:27,781 INFO [train.py:968] (1/2) Epoch 25, batch 41300, giga_loss[loss=0.4453, simple_loss=0.4705, pruned_loss=0.2101, over 27491.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3848, pruned_loss=0.1332, over 5631787.53 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3611, pruned_loss=0.1137, over 5696156.70 frames. ], giga_tot_loss[loss=0.3272, simple_loss=0.3861, pruned_loss=0.1342, over 5624002.80 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:29:40,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5714, 1.8238, 1.4214, 1.9181], device='cuda:1'), covar=tensor([0.2424, 0.2614, 0.2860, 0.2307], device='cuda:1'), in_proj_covar=tensor([0.1560, 0.1125, 0.1380, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 05:30:13,612 INFO [optim.py:369] (1/2) 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 05:30:22,915 INFO [train.py:968] (1/2) Epoch 25, batch 41350, libri_loss[loss=0.2698, simple_loss=0.3478, pruned_loss=0.09588, over 29540.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.389, pruned_loss=0.1367, over 5633075.90 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3608, pruned_loss=0.1135, over 5698312.62 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3906, pruned_loss=0.138, over 5624349.69 frames. ], batch size: 82, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:31:09,759 INFO [train.py:968] (1/2) Epoch 25, batch 41400, giga_loss[loss=0.3158, simple_loss=0.3897, pruned_loss=0.121, over 29049.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3901, pruned_loss=0.1381, over 5625260.14 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5683538.73 frames. ], giga_tot_loss[loss=0.3364, simple_loss=0.3926, pruned_loss=0.1401, over 5630034.18 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:31:36,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-13 05:31:44,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6875, 1.7723, 1.8901, 1.4418], device='cuda:1'), covar=tensor([0.1714, 0.2427, 0.1399, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0710, 0.0965, 0.0865], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 05:31:53,851 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 41450, giga_loss[loss=0.3586, simple_loss=0.4152, pruned_loss=0.151, over 29006.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3885, pruned_loss=0.138, over 5628153.77 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.1129, over 5688194.00 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.3913, pruned_loss=0.1403, over 5626840.22 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:32:49,705 INFO [train.py:968] (1/2) Epoch 25, batch 41500, giga_loss[loss=0.3666, simple_loss=0.4082, pruned_loss=0.1625, over 26730.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3873, pruned_loss=0.137, over 5626460.47 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5690124.85 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3898, pruned_loss=0.1392, over 5623123.91 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:33:36,510 INFO [optim.py:369] (1/2) 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 05:33:39,817 INFO [train.py:968] (1/2) Epoch 25, batch 41550, giga_loss[loss=0.3208, simple_loss=0.3893, pruned_loss=0.1261, over 28606.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3856, pruned_loss=0.1346, over 5613133.93 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1126, over 5682835.34 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3882, pruned_loss=0.1369, over 5614947.08 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 05:34:08,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1135417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:34:20,101 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-13 05:34:25,350 INFO [train.py:968] (1/2) Epoch 25, batch 41600, giga_loss[loss=0.3425, simple_loss=0.4088, pruned_loss=0.1381, over 28938.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.388, pruned_loss=0.1368, over 5616439.40 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5683208.57 frames. ], giga_tot_loss[loss=0.336, simple_loss=0.392, pruned_loss=0.14, over 5612842.52 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:35:07,951 INFO [optim.py:369] (1/2) 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 05:35:12,771 INFO [train.py:968] (1/2) Epoch 25, batch 41650, giga_loss[loss=0.3113, simple_loss=0.3756, pruned_loss=0.1235, over 28697.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3865, pruned_loss=0.1358, over 5591415.23 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5670419.82 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3908, pruned_loss=0.1392, over 5595796.50 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:35:23,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6876, 1.8260, 1.7650, 1.6022], device='cuda:1'), covar=tensor([0.3011, 0.2579, 0.2224, 0.2670], device='cuda:1'), in_proj_covar=tensor([0.2039, 0.1969, 0.1888, 0.2043], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 05:36:01,549 INFO [train.py:968] (1/2) Epoch 25, batch 41700, giga_loss[loss=0.326, simple_loss=0.3856, pruned_loss=0.1332, over 27883.00 frames. ], tot_loss[loss=0.3238, simple_loss=0.3834, pruned_loss=0.1321, over 5610888.87 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5675844.65 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3876, pruned_loss=0.1355, over 5607637.97 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:36:20,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1135560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:36:24,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1135563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:36:45,518 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 41750, giga_loss[loss=0.2621, simple_loss=0.342, pruned_loss=0.09108, over 28485.00 frames. ], tot_loss[loss=0.3208, simple_loss=0.3822, pruned_loss=0.1297, over 5621617.17 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5669601.06 frames. ], giga_tot_loss[loss=0.3254, simple_loss=0.3858, pruned_loss=0.1325, over 5623206.06 frames. ], batch size: 65, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:36:53,111 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1135592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:36:57,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2620, 2.5682, 1.2882, 1.3811], device='cuda:1'), covar=tensor([0.1027, 0.0460, 0.0942, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0563, 0.0400, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 05:37:37,228 INFO [train.py:968] (1/2) Epoch 25, batch 41800, giga_loss[loss=0.3048, simple_loss=0.3677, pruned_loss=0.121, over 27666.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3789, pruned_loss=0.1271, over 5617157.12 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5664726.83 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3824, pruned_loss=0.13, over 5620401.74 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:38:00,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 05:38:19,995 INFO [optim.py:369] (1/2) 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 05:38:23,316 INFO [train.py:968] (1/2) Epoch 25, batch 41850, giga_loss[loss=0.2937, simple_loss=0.3633, pruned_loss=0.1121, over 28776.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3757, pruned_loss=0.1245, over 5604454.25 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1126, over 5654237.75 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.3795, pruned_loss=0.1274, over 5615539.09 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:39:07,138 INFO [train.py:968] (1/2) Epoch 25, batch 41900, libri_loss[loss=0.2653, simple_loss=0.3356, pruned_loss=0.09746, over 29646.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.373, pruned_loss=0.1224, over 5636144.78 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3587, pruned_loss=0.1121, over 5664284.70 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3773, pruned_loss=0.1258, over 5634315.15 frames. ], batch size: 73, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:39:49,125 INFO [optim.py:369] (1/2) 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 05:39:51,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1907, 1.7963, 1.3765, 0.4420], device='cuda:1'), covar=tensor([0.5289, 0.3285, 0.4416, 0.7098], device='cuda:1'), in_proj_covar=tensor([0.1811, 0.1700, 0.1636, 0.1476], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 05:39:53,426 INFO [train.py:968] (1/2) Epoch 25, batch 41950, giga_loss[loss=0.2866, simple_loss=0.3603, pruned_loss=0.1065, over 29082.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3733, pruned_loss=0.1227, over 5643699.57 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5667294.76 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3769, pruned_loss=0.1255, over 5638870.22 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:40:23,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4140, 1.6896, 1.4523, 1.5259], device='cuda:1'), covar=tensor([0.0734, 0.0386, 0.0334, 0.0825], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 05:40:42,463 INFO [train.py:968] (1/2) Epoch 25, batch 42000, giga_loss[loss=0.3065, simple_loss=0.3777, pruned_loss=0.1176, over 28861.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.372, pruned_loss=0.1217, over 5617200.44 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1125, over 5641030.68 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3751, pruned_loss=0.1239, over 5635155.83 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:40:42,463 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 05:40:51,195 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 05:41:38,755 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1135884.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:41:39,890 INFO [optim.py:369] (1/2) 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 05:41:43,188 INFO [train.py:968] (1/2) Epoch 25, batch 42050, giga_loss[loss=0.313, simple_loss=0.3877, pruned_loss=0.1192, over 27984.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3703, pruned_loss=0.1194, over 5622195.38 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1125, over 5642598.67 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3729, pruned_loss=0.1213, over 5634325.62 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:41:53,191 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1135896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:42:34,533 INFO [train.py:968] (1/2) Epoch 25, batch 42100, giga_loss[loss=0.2639, simple_loss=0.3514, pruned_loss=0.0882, over 28829.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3722, pruned_loss=0.1185, over 5633978.83 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1124, over 5644902.58 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3748, pruned_loss=0.1202, over 5641224.49 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:42:44,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 05:43:11,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5089, 1.7598, 1.4416, 1.5998], device='cuda:1'), covar=tensor([0.2562, 0.2607, 0.2884, 0.2320], device='cuda:1'), in_proj_covar=tensor([0.1563, 0.1126, 0.1381, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 05:43:18,313 INFO [optim.py:369] (1/2) 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 05:43:23,414 INFO [train.py:968] (1/2) Epoch 25, batch 42150, giga_loss[loss=0.2888, simple_loss=0.3598, pruned_loss=0.1089, over 28703.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3736, pruned_loss=0.1189, over 5647078.74 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3589, pruned_loss=0.1124, over 5645590.71 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3758, pruned_loss=0.1204, over 5652086.82 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:43:52,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9398, 1.0958, 1.0972, 0.9244], device='cuda:1'), covar=tensor([0.2487, 0.2768, 0.1676, 0.2206], device='cuda:1'), in_proj_covar=tensor([0.2021, 0.1956, 0.1873, 0.2029], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 05:44:07,961 INFO [train.py:968] (1/2) Epoch 25, batch 42200, giga_loss[loss=0.3156, simple_loss=0.3776, pruned_loss=0.1268, over 27942.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3733, pruned_loss=0.1194, over 5642969.09 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3585, pruned_loss=0.1121, over 5643136.98 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.376, pruned_loss=0.1211, over 5649827.32 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:44:48,255 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:44:48,625 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 42250, giga_loss[loss=0.2664, simple_loss=0.3426, pruned_loss=0.09508, over 28935.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3721, pruned_loss=0.119, over 5638521.35 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3578, pruned_loss=0.1117, over 5630228.48 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3753, pruned_loss=0.1209, over 5655327.67 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:45:08,663 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8932, 2.8071, 1.9092, 1.0632], device='cuda:1'), covar=tensor([0.8273, 0.3745, 0.4104, 0.7539], device='cuda:1'), in_proj_covar=tensor([0.1810, 0.1695, 0.1637, 0.1472], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 05:45:37,620 INFO [train.py:968] (1/2) Epoch 25, batch 42300, giga_loss[loss=0.4201, simple_loss=0.4371, pruned_loss=0.2016, over 26604.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3712, pruned_loss=0.1198, over 5650653.61 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.358, pruned_loss=0.1117, over 5635950.72 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3739, pruned_loss=0.1215, over 5659077.96 frames. ], batch size: 555, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:46:21,603 INFO [optim.py:369] (1/2) 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 05:46:26,116 INFO [train.py:968] (1/2) Epoch 25, batch 42350, giga_loss[loss=0.3117, simple_loss=0.3775, pruned_loss=0.123, over 28914.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3694, pruned_loss=0.1191, over 5653977.91 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3579, pruned_loss=0.1116, over 5641191.57 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3719, pruned_loss=0.1206, over 5656479.20 frames. ], batch size: 199, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:46:26,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4112, 3.3919, 1.5544, 1.5704], device='cuda:1'), covar=tensor([0.1031, 0.0374, 0.0910, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0565, 0.0400, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 05:46:59,967 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:47:16,831 INFO [train.py:968] (1/2) Epoch 25, batch 42400, giga_loss[loss=0.2604, simple_loss=0.3507, pruned_loss=0.08507, over 28884.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3694, pruned_loss=0.1177, over 5663288.49 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5643894.88 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3717, pruned_loss=0.1192, over 5662837.31 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:47:34,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136259.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:47:45,399 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:48:03,450 INFO [train.py:968] (1/2) Epoch 25, batch 42450, giga_loss[loss=0.2793, simple_loss=0.3514, pruned_loss=0.1036, over 28665.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3699, pruned_loss=0.1176, over 5674116.62 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3572, pruned_loss=0.1111, over 5646929.05 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3725, pruned_loss=0.1192, over 5671747.49 frames. ], batch size: 119, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:48:50,053 INFO [train.py:968] (1/2) Epoch 25, batch 42500, giga_loss[loss=0.2933, simple_loss=0.3648, pruned_loss=0.1109, over 29080.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3715, pruned_loss=0.1191, over 5660942.73 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3579, pruned_loss=0.1116, over 5642182.41 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3732, pruned_loss=0.1201, over 5664245.90 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:48:51,356 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:49:29,277 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 05:49:32,304 INFO [optim.py:369] (1/2) 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 05:49:35,265 INFO [train.py:968] (1/2) Epoch 25, batch 42550, giga_loss[loss=0.389, simple_loss=0.4031, pruned_loss=0.1875, over 23715.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3694, pruned_loss=0.1183, over 5660374.69 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1118, over 5636006.74 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3709, pruned_loss=0.119, over 5668367.03 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:49:47,811 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136402.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 05:49:50,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136405.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:49:58,347 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:50:01,098 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:50:19,656 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 42600, giga_loss[loss=0.2904, simple_loss=0.3637, pruned_loss=0.1085, over 28896.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3678, pruned_loss=0.1176, over 5660381.39 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3576, pruned_loss=0.1116, over 5638174.74 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3694, pruned_loss=0.1185, over 5665114.65 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:50:29,331 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:50:43,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:50:49,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2120, 1.5409, 1.1862, 0.7473], device='cuda:1'), covar=tensor([0.3577, 0.2486, 0.2940, 0.5491], device='cuda:1'), in_proj_covar=tensor([0.1816, 0.1703, 0.1640, 0.1478], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 05:51:06,109 INFO [optim.py:369] (1/2) 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 05:51:10,682 INFO [train.py:968] (1/2) Epoch 25, batch 42650, giga_loss[loss=0.2698, simple_loss=0.3491, pruned_loss=0.09521, over 28854.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3672, pruned_loss=0.118, over 5668428.62 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.358, pruned_loss=0.1117, over 5641944.19 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3684, pruned_loss=0.1188, over 5669295.11 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:51:55,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3965, 1.1314, 4.0981, 3.4450], device='cuda:1'), covar=tensor([0.1577, 0.2838, 0.0467, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0665, 0.0982, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 05:51:58,216 INFO [train.py:968] (1/2) Epoch 25, batch 42700, giga_loss[loss=0.2612, simple_loss=0.3375, pruned_loss=0.09245, over 28838.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3655, pruned_loss=0.117, over 5676990.40 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3583, pruned_loss=0.1118, over 5645826.27 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3664, pruned_loss=0.1176, over 5674838.46 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:52:30,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4101, 1.5530, 1.2446, 1.1217], device='cuda:1'), covar=tensor([0.1061, 0.0577, 0.1078, 0.1180], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0449, 0.0520, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 05:52:36,027 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 25, batch 42750, giga_loss[loss=0.2775, simple_loss=0.3412, pruned_loss=0.1069, over 28918.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3644, pruned_loss=0.1171, over 5661132.58 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3585, pruned_loss=0.1121, over 5634625.70 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1174, over 5670444.11 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:52:50,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:53:00,044 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:53:05,061 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:53:05,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.31 vs. limit=5.0
+2023-03-13 05:53:33,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:53:39,243 INFO [train.py:968] (1/2) Epoch 25, batch 42800, giga_loss[loss=0.3125, simple_loss=0.374, pruned_loss=0.1255, over 29005.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3646, pruned_loss=0.1182, over 5646506.52 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3587, pruned_loss=0.1123, over 5632576.57 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.365, pruned_loss=0.1183, over 5655551.03 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:54:19,837 INFO [optim.py:369] (1/2) 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 05:54:23,284 INFO [train.py:968] (1/2) Epoch 25, batch 42850, giga_loss[loss=0.2874, simple_loss=0.3627, pruned_loss=0.106, over 28999.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1176, over 5655855.79 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5639619.02 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3653, pruned_loss=0.1179, over 5657701.69 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 05:54:44,746 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:04,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:05,444 INFO [train.py:968] (1/2) Epoch 25, batch 42900, giga_loss[loss=0.2646, simple_loss=0.3488, pruned_loss=0.09017, over 28946.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3641, pruned_loss=0.1157, over 5670439.09 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1122, over 5647371.06 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3649, pruned_loss=0.1163, over 5665617.55 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:55:06,479 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:32,725 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:55:47,227 INFO [optim.py:369] (1/2) 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 05:55:49,784 INFO [train.py:968] (1/2) Epoch 25, batch 42950, libri_loss[loss=0.3167, simple_loss=0.3737, pruned_loss=0.1298, over 29228.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3647, pruned_loss=0.1154, over 5672844.05 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3584, pruned_loss=0.1119, over 5650566.72 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3658, pruned_loss=0.1161, over 5666415.68 frames. ], batch size: 101, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:56:38,464 INFO [train.py:968] (1/2) Epoch 25, batch 43000, giga_loss[loss=0.2486, simple_loss=0.3226, pruned_loss=0.08729, over 28491.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3655, pruned_loss=0.1155, over 5683775.97 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5655585.20 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3661, pruned_loss=0.116, over 5674626.80 frames. ], batch size: 60, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:56:50,938 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 05:56:58,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1136859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:56:58,520 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 05:57:00,972 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1136862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:57:25,052 INFO [optim.py:369] (1/2) 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 05:57:26,705 INFO [train.py:968] (1/2) Epoch 25, batch 43050, giga_loss[loss=0.3166, simple_loss=0.3919, pruned_loss=0.1207, over 28999.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3688, pruned_loss=0.1184, over 5676478.61 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3586, pruned_loss=0.1121, over 5649955.27 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3695, pruned_loss=0.1189, over 5675435.05 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:57:27,620 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1136891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:57:54,943 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:58:02,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1136927.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 05:58:12,600 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-13 05:58:12,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-13 05:58:16,429 INFO [train.py:968] (1/2) Epoch 25, batch 43100, giga_loss[loss=0.3197, simple_loss=0.3774, pruned_loss=0.131, over 28761.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3694, pruned_loss=0.1198, over 5676044.45 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3591, pruned_loss=0.1125, over 5643593.75 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3698, pruned_loss=0.1201, over 5681394.41 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:58:34,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1136954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 05:59:02,659 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.29 vs. limit=5.0
+2023-03-13 05:59:08,725 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 43150, giga_loss[loss=0.3517, simple_loss=0.3988, pruned_loss=0.1523, over 28191.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3712, pruned_loss=0.1227, over 5672414.39 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3589, pruned_loss=0.1124, over 5644735.78 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3717, pruned_loss=0.123, over 5675720.12 frames. ], batch size: 368, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 05:59:59,443 INFO [train.py:968] (1/2) Epoch 25, batch 43200, giga_loss[loss=0.3201, simple_loss=0.3878, pruned_loss=0.1262, over 28488.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3743, pruned_loss=0.1256, over 5652223.65 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3591, pruned_loss=0.1125, over 5641233.82 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3749, pruned_loss=0.126, over 5658162.69 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:00:05,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 06:00:42,004 INFO [optim.py:369] (1/2) 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 06:00:42,598 INFO [train.py:968] (1/2) Epoch 25, batch 43250, giga_loss[loss=0.2575, simple_loss=0.3342, pruned_loss=0.09038, over 28979.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3714, pruned_loss=0.1239, over 5664388.31 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3583, pruned_loss=0.1121, over 5650666.70 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3731, pruned_loss=0.1251, over 5660805.70 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:00:48,421 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:00:52,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:01:16,866 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:01:27,848 INFO [train.py:968] (1/2) Epoch 25, batch 43300, giga_loss[loss=0.2944, simple_loss=0.3658, pruned_loss=0.1115, over 28650.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3719, pruned_loss=0.1236, over 5663475.22 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3588, pruned_loss=0.1125, over 5648450.55 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3732, pruned_loss=0.1245, over 5662955.31 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:02:08,033 INFO [optim.py:369] (1/2) 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 06:02:08,579 INFO [train.py:968] (1/2) Epoch 25, batch 43350, giga_loss[loss=0.2937, simple_loss=0.3652, pruned_loss=0.1111, over 29023.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3711, pruned_loss=0.1218, over 5655381.16 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3587, pruned_loss=0.1124, over 5648070.49 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3727, pruned_loss=0.123, over 5655746.24 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:02:55,178 INFO [train.py:968] (1/2) Epoch 25, batch 43400, giga_loss[loss=0.3018, simple_loss=0.3643, pruned_loss=0.1197, over 28938.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3697, pruned_loss=0.121, over 5648914.36 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3592, pruned_loss=0.1127, over 5641871.43 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1218, over 5654374.36 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:03:40,552 INFO [optim.py:369] (1/2) 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 06:03:41,973 INFO [train.py:968] (1/2) Epoch 25, batch 43450, giga_loss[loss=0.3053, simple_loss=0.3675, pruned_loss=0.1216, over 28522.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5661113.82 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5644288.99 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3687, pruned_loss=0.1208, over 5663424.71 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:03:45,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1137293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:03:52,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1137302.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 06:04:00,596 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:04:25,020 INFO [train.py:968] (1/2) Epoch 25, batch 43500, giga_loss[loss=0.2862, simple_loss=0.3496, pruned_loss=0.1114, over 28949.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3647, pruned_loss=0.1181, over 5668088.64 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3587, pruned_loss=0.1124, over 5648376.57 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3661, pruned_loss=0.1192, over 5666915.96 frames. ], batch size: 106, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:04:59,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-13 06:05:13,251 INFO [optim.py:369] (1/2) 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 06:05:14,158 INFO [train.py:968] (1/2) Epoch 25, batch 43550, giga_loss[loss=0.2882, simple_loss=0.368, pruned_loss=0.1042, over 28938.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1207, over 5665667.87 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3584, pruned_loss=0.1123, over 5652237.21 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3697, pruned_loss=0.1217, over 5661333.52 frames. ], batch size: 112, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:05:50,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3787, 1.2541, 3.9189, 3.3401], device='cuda:1'), covar=tensor([0.1609, 0.2848, 0.0458, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0664, 0.0982, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 06:05:57,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:05:59,698 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:06:00,131 INFO [train.py:968] (1/2) Epoch 25, batch 43600, giga_loss[loss=0.297, simple_loss=0.3821, pruned_loss=0.106, over 28999.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3705, pruned_loss=0.1196, over 5668269.97 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.358, pruned_loss=0.112, over 5657079.29 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3724, pruned_loss=0.1211, over 5660564.81 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:06:05,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137445.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:06:10,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137448.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 06:06:14,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4694, 1.9612, 1.8493, 1.6475], device='cuda:1'), covar=tensor([0.2347, 0.2264, 0.2565, 0.2707], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0753, 0.0723, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 06:06:28,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:06:35,066 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137477.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:06:50,797 INFO [optim.py:369] (1/2) 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 06:06:50,810 INFO [train.py:968] (1/2) Epoch 25, batch 43650, giga_loss[loss=0.2846, simple_loss=0.3644, pruned_loss=0.1024, over 28360.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.372, pruned_loss=0.1189, over 5671926.04 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1119, over 5659813.04 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3739, pruned_loss=0.1202, over 5663644.89 frames. ], batch size: 71, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:07:38,278 INFO [train.py:968] (1/2) Epoch 25, batch 43700, giga_loss[loss=0.302, simple_loss=0.3712, pruned_loss=0.1163, over 27975.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3748, pruned_loss=0.1213, over 5668630.39 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1119, over 5663694.30 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3765, pruned_loss=0.1224, over 5658792.61 frames. ], batch size: 412, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:08:23,983 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 25, batch 43750, giga_loss[loss=0.2793, simple_loss=0.3541, pruned_loss=0.1022, over 29094.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3742, pruned_loss=0.121, over 5675455.74 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1114, over 5667191.63 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3764, pruned_loss=0.1225, over 5664455.18 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:09:09,534 INFO [train.py:968] (1/2) Epoch 25, batch 43800, giga_loss[loss=0.2964, simple_loss=0.3638, pruned_loss=0.1145, over 28632.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3748, pruned_loss=0.1227, over 5667144.06 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3569, pruned_loss=0.1114, over 5659645.26 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3771, pruned_loss=0.1241, over 5664797.41 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:09:49,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:09:55,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1137687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:09:57,212 INFO [optim.py:369] (1/2) 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 06:09:57,225 INFO [train.py:968] (1/2) Epoch 25, batch 43850, libri_loss[loss=0.3181, simple_loss=0.379, pruned_loss=0.1286, over 29643.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3739, pruned_loss=0.1231, over 5657908.26 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3577, pruned_loss=0.1119, over 5655725.63 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3754, pruned_loss=0.1239, over 5659378.88 frames. ], batch size: 88, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:09:58,756 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:10:30,336 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1137729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:10:42,953 INFO [train.py:968] (1/2) Epoch 25, batch 43900, giga_loss[loss=0.3097, simple_loss=0.3748, pruned_loss=0.1223, over 28979.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3727, pruned_loss=0.1227, over 5665097.15 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3582, pruned_loss=0.1122, over 5659481.33 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3738, pruned_loss=0.1233, over 5662949.29 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:11:10,925 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7026, 1.8081, 1.8020, 1.6803], device='cuda:1'), covar=tensor([0.1974, 0.2277, 0.2408, 0.2153], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0758, 0.0725, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 06:11:28,551 INFO [optim.py:369] (1/2) 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 06:11:28,563 INFO [train.py:968] (1/2) Epoch 25, batch 43950, giga_loss[loss=0.3208, simple_loss=0.3806, pruned_loss=0.1305, over 28489.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3718, pruned_loss=0.1229, over 5668308.14 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3581, pruned_loss=0.1121, over 5664319.26 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3731, pruned_loss=0.1237, over 5662566.30 frames. ], batch size: 336, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:11:47,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 06:11:59,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1624, 1.2545, 3.5169, 3.1202], device='cuda:1'), covar=tensor([0.1669, 0.2759, 0.0504, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0667, 0.0985, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 06:12:09,260 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1137830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:12:12,148 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1137833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:12:15,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4756, 1.5640, 1.6705, 1.2751], device='cuda:1'), covar=tensor([0.1660, 0.2412, 0.1366, 0.1644], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0714, 0.0969, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 06:12:17,538 INFO [train.py:968] (1/2) Epoch 25, batch 44000, giga_loss[loss=0.3735, simple_loss=0.415, pruned_loss=0.166, over 27569.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1234, over 5677123.76 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3578, pruned_loss=0.1119, over 5666129.40 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5671350.20 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:12:40,438 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1137862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:13:06,242 INFO [train.py:968] (1/2) Epoch 25, batch 44050, giga_loss[loss=0.3347, simple_loss=0.3885, pruned_loss=0.1404, over 28831.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3717, pruned_loss=0.1235, over 5670665.66 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3574, pruned_loss=0.1117, over 5667500.44 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3734, pruned_loss=0.1246, over 5665001.19 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:13:07,734 INFO [optim.py:369] (1/2) 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 06:13:53,567 INFO [train.py:968] (1/2) Epoch 25, batch 44100, giga_loss[loss=0.2962, simple_loss=0.3573, pruned_loss=0.1176, over 29107.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3689, pruned_loss=0.1219, over 5675412.31 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3572, pruned_loss=0.1115, over 5669933.38 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.123, over 5668726.17 frames. ], batch size: 128, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:14:18,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3575, 3.1651, 1.4707, 1.4843], device='cuda:1'), covar=tensor([0.1043, 0.0377, 0.0942, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0566, 0.0401, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 06:14:34,900 INFO [train.py:968] (1/2) Epoch 25, batch 44150, libri_loss[loss=0.3263, simple_loss=0.382, pruned_loss=0.1353, over 29534.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3672, pruned_loss=0.1198, over 5680581.85 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3572, pruned_loss=0.1113, over 5677824.69 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3689, pruned_loss=0.1213, over 5667750.27 frames. ], batch size: 79, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:14:35,506 INFO [optim.py:369] (1/2) 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 06:15:26,938 INFO [train.py:968] (1/2) Epoch 25, batch 44200, giga_loss[loss=0.3115, simple_loss=0.3815, pruned_loss=0.1207, over 28542.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3688, pruned_loss=0.1201, over 5677457.04 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.357, pruned_loss=0.1111, over 5679608.66 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3705, pruned_loss=0.1216, over 5665525.56 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:15:28,035 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:15:31,721 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3182, 1.6347, 1.3290, 1.0441], device='cuda:1'), covar=tensor([0.2417, 0.2445, 0.2728, 0.2326], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1130, 0.1385, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 06:15:34,681 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5726, 1.8650, 1.5216, 1.7058], device='cuda:1'), covar=tensor([0.2301, 0.2278, 0.2434, 0.2204], device='cuda:1'), in_proj_covar=tensor([0.1565, 0.1130, 0.1385, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 06:15:41,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:15:43,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2331, 2.6124, 1.3371, 1.3286], device='cuda:1'), covar=tensor([0.1037, 0.0394, 0.0894, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0566, 0.0401, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 06:15:51,652 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:16:02,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 06:16:12,188 INFO [train.py:968] (1/2) Epoch 25, batch 44250, giga_loss[loss=0.2743, simple_loss=0.3496, pruned_loss=0.09949, over 29063.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 5668463.80 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3574, pruned_loss=0.1114, over 5666020.39 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3715, pruned_loss=0.1222, over 5670897.01 frames. ], batch size: 155, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:16:13,703 INFO [optim.py:369] (1/2) 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 06:16:28,858 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:17:01,906 INFO [train.py:968] (1/2) Epoch 25, batch 44300, giga_loss[loss=0.3043, simple_loss=0.397, pruned_loss=0.1058, over 28997.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3706, pruned_loss=0.1217, over 5661133.07 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3572, pruned_loss=0.1112, over 5666701.18 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3722, pruned_loss=0.123, over 5662396.80 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:17:11,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4839, 1.6586, 1.5369, 1.5375], device='cuda:1'), covar=tensor([0.0629, 0.0301, 0.0286, 0.0677], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 06:17:11,995 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:17:45,442 INFO [train.py:968] (1/2) Epoch 25, batch 44350, giga_loss[loss=0.2679, simple_loss=0.3492, pruned_loss=0.09331, over 28681.00 frames. ], tot_loss[loss=0.306, simple_loss=0.372, pruned_loss=0.12, over 5668895.58 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3571, pruned_loss=0.1111, over 5670874.28 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3737, pruned_loss=0.1213, over 5666048.64 frames. ], batch size: 92, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:17:45,997 INFO [optim.py:369] (1/2) 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 06:17:53,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:17:56,034 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:04,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:06,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:21,598 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 44400, giga_loss[loss=0.2698, simple_loss=0.362, pruned_loss=0.08874, over 28843.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3729, pruned_loss=0.1182, over 5680421.09 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3569, pruned_loss=0.1108, over 5673386.04 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3748, pruned_loss=0.1197, over 5676034.65 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:18:31,634 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:36,693 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:38,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:18:54,379 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:19:08,423 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:19:18,610 INFO [train.py:968] (1/2) Epoch 25, batch 44450, giga_loss[loss=0.3131, simple_loss=0.3871, pruned_loss=0.1196, over 28904.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3763, pruned_loss=0.1202, over 5671971.41 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3575, pruned_loss=0.1112, over 5657645.31 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3777, pruned_loss=0.1213, over 5682528.58 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:19:19,911 INFO [optim.py:369] (1/2) 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 06:19:28,335 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:19:47,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-13 06:20:03,077 INFO [train.py:968] (1/2) Epoch 25, batch 44500, giga_loss[loss=0.3525, simple_loss=0.3845, pruned_loss=0.1602, over 23340.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3782, pruned_loss=0.1229, over 5670361.39 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3575, pruned_loss=0.1111, over 5667571.33 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.38, pruned_loss=0.1242, over 5670397.44 frames. ], batch size: 705, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:20:27,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:20:53,501 INFO [train.py:968] (1/2) Epoch 25, batch 44550, giga_loss[loss=0.2976, simple_loss=0.3713, pruned_loss=0.1119, over 28899.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.3788, pruned_loss=0.1247, over 5640902.15 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3575, pruned_loss=0.1112, over 5652119.06 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3805, pruned_loss=0.1259, over 5654216.77 frames. ], batch size: 174, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:20:55,081 INFO [optim.py:369] (1/2) 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 06:21:15,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:21:38,302 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-13 06:21:38,406 INFO [train.py:968] (1/2) Epoch 25, batch 44600, giga_loss[loss=0.3044, simple_loss=0.3758, pruned_loss=0.1165, over 28997.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3781, pruned_loss=0.1247, over 5653482.67 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3571, pruned_loss=0.111, over 5657621.36 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3803, pruned_loss=0.1262, over 5659078.30 frames. ], batch size: 164, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:22:23,578 INFO [train.py:968] (1/2) Epoch 25, batch 44650, giga_loss[loss=0.2815, simple_loss=0.3582, pruned_loss=0.1024, over 28994.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3771, pruned_loss=0.1237, over 5658444.21 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3567, pruned_loss=0.1107, over 5661368.83 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3794, pruned_loss=0.1253, over 5659692.78 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:22:26,109 INFO [optim.py:369] (1/2) 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 06:22:45,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3320, 1.2571, 3.5839, 3.1754], device='cuda:1'), covar=tensor([0.1558, 0.2818, 0.0542, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0668, 0.0985, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 06:22:58,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:11,968 INFO [train.py:968] (1/2) Epoch 25, batch 44700, giga_loss[loss=0.3542, simple_loss=0.4067, pruned_loss=0.1508, over 27672.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3768, pruned_loss=0.1215, over 5666219.79 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3564, pruned_loss=0.1106, over 5665278.35 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3792, pruned_loss=0.123, over 5663717.69 frames. ], batch size: 472, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:23:26,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:30,968 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:32,814 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 44750, giga_loss[loss=0.3038, simple_loss=0.3737, pruned_loss=0.1169, over 28967.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3779, pruned_loss=0.1217, over 5669207.52 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3568, pruned_loss=0.111, over 5670175.20 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3799, pruned_loss=0.1228, over 5662673.02 frames. ], batch size: 213, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:23:57,725 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:23:58,696 INFO [optim.py:369] (1/2) 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 06:24:10,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-13 06:24:43,358 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 44800, giga_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 28903.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3786, pruned_loss=0.1229, over 5660612.19 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.357, pruned_loss=0.1112, over 5662186.85 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3804, pruned_loss=0.1238, over 5663573.54 frames. ], batch size: 227, lr: 1.26e-03, grad_scale: 8.0
+2023-03-13 06:24:46,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:00,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:10,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:12,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:17,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:25:30,070 INFO [train.py:968] (1/2) Epoch 25, batch 44850, giga_loss[loss=0.3162, simple_loss=0.3803, pruned_loss=0.126, over 28622.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3775, pruned_loss=0.1226, over 5667359.67 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3571, pruned_loss=0.1112, over 5657087.40 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3792, pruned_loss=0.1235, over 5674181.36 frames. ], batch size: 307, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:25:33,418 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:26:18,075 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:26:18,404 INFO [train.py:968] (1/2) Epoch 25, batch 44900, giga_loss[loss=0.2906, simple_loss=0.3607, pruned_loss=0.1103, over 28902.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3755, pruned_loss=0.1227, over 5654332.49 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.357, pruned_loss=0.1112, over 5661726.60 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3772, pruned_loss=0.1236, over 5655341.73 frames. ], batch size: 186, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:26:58,960 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:03,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1138786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:06,077 INFO [train.py:968] (1/2) Epoch 25, batch 44950, giga_loss[loss=0.2783, simple_loss=0.3547, pruned_loss=0.1009, over 28943.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3732, pruned_loss=0.122, over 5653333.12 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.357, pruned_loss=0.1111, over 5662929.81 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3749, pruned_loss=0.1229, over 5652859.43 frames. ], batch size: 136, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:27:07,843 INFO [optim.py:369] (1/2) 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 06:27:29,425 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:32,604 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:27:36,245 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 45000, giga_loss[loss=0.2835, simple_loss=0.3557, pruned_loss=0.1057, over 28768.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3706, pruned_loss=0.1205, over 5662005.01 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3571, pruned_loss=0.1111, over 5669965.73 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3723, pruned_loss=0.1216, over 5655279.01 frames. ], batch size: 284, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:27:51,422 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 06:28:00,388 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 06:28:11,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:28:23,130 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138866.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:28:26,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6333, 1.7116, 1.8347, 1.4072], device='cuda:1'), covar=tensor([0.1873, 0.2583, 0.1497, 0.1763], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0715, 0.0971, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 06:28:36,424 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1138882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:28:38,444 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 25, batch 45050, giga_loss[loss=0.2878, simple_loss=0.3641, pruned_loss=0.1057, over 28933.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3701, pruned_loss=0.1213, over 5647218.46 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1115, over 5654575.83 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.1221, over 5656286.53 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:28:45,772 INFO [optim.py:369] (1/2) 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 06:28:57,678 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1138908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:29:04,247 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1138914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:29:15,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4170, 1.6384, 1.6632, 1.2279], device='cuda:1'), covar=tensor([0.1923, 0.2769, 0.1618, 0.1866], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0714, 0.0971, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 06:29:16,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1138927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:29:28,466 INFO [train.py:968] (1/2) Epoch 25, batch 45100, giga_loss[loss=0.268, simple_loss=0.335, pruned_loss=0.1005, over 28599.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3691, pruned_loss=0.1204, over 5632922.72 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3578, pruned_loss=0.112, over 5630777.38 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3701, pruned_loss=0.1208, over 5660130.72 frames. ], batch size: 85, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 06:30:05,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5883, 5.4550, 5.1624, 2.4858], device='cuda:1'), covar=tensor([0.0435, 0.0535, 0.0608, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.1199, 0.1009, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 06:30:10,865 INFO [train.py:968] (1/2) Epoch 25, batch 45150, giga_loss[loss=0.2609, simple_loss=0.3449, pruned_loss=0.08849, over 28957.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3675, pruned_loss=0.1183, over 5571909.02 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.359, pruned_loss=0.113, over 5562128.41 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3674, pruned_loss=0.1178, over 5655323.37 frames. ], batch size: 145, lr: 1.26e-03, grad_scale: 2.0
+2023-03-13 06:30:13,893 INFO [optim.py:369] (1/2) 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 06:30:31,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:30:38,954 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:30:47,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:30:54,049 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-13 06:30:55,380 INFO [train.py:968] (1/2) Epoch 25, batch 45200, giga_loss[loss=0.3041, simple_loss=0.3745, pruned_loss=0.1169, over 28755.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3659, pruned_loss=0.1165, over 5578924.33 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3595, pruned_loss=0.1135, over 5537663.90 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3656, pruned_loss=0.1158, over 5664999.29 frames. ], batch size: 262, lr: 1.26e-03, grad_scale: 4.0
+2023-03-13 06:31:12,945 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-13 06:31:55,534 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:31:57,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:32:15,343 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.5439, 1.8597, 1.4368, 1.9049], device='cuda:1'), covar=tensor([0.2859, 0.2900, 0.3308, 0.2609], device='cuda:1'), in_proj_covar=tensor([0.1572, 0.1134, 0.1390, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 06:32:17,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0641, 1.2349, 1.0409, 0.8853], device='cuda:1'), covar=tensor([0.1134, 0.0505, 0.1162, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0451, 0.0521, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 06:32:21,611 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:32:25,296 INFO [train.py:968] (1/2) Epoch 26, batch 50, giga_loss[loss=0.2662, simple_loss=0.3554, pruned_loss=0.08847, over 28692.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.3657, pruned_loss=0.1043, over 1259928.96 frames. ], libri_tot_loss[loss=0.2415, simple_loss=0.3226, pruned_loss=0.08019, over 192639.08 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.3724, pruned_loss=0.108, over 1104715.67 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:32:34,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4531, 4.1418, 1.6111, 1.7054], device='cuda:1'), covar=tensor([0.1046, 0.0240, 0.0943, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0562, 0.0399, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 06:33:09,644 INFO [train.py:968] (1/2) Epoch 26, batch 100, giga_loss[loss=0.2481, simple_loss=0.3299, pruned_loss=0.08313, over 28643.00 frames. ], tot_loss[loss=0.2808, simple_loss=0.3592, pruned_loss=0.1012, over 2238256.76 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3293, pruned_loss=0.08325, over 388835.75 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3641, pruned_loss=0.1042, over 1983987.80 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:33:10,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:13,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:24,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:26,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:37,282 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:33:40,397 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 150, giga_loss[loss=0.2558, simple_loss=0.3297, pruned_loss=0.09095, over 28301.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3453, pruned_loss=0.09478, over 3011538.84 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3297, pruned_loss=0.08374, over 525358.84 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3479, pruned_loss=0.09655, over 2735699.96 frames. ], batch size: 369, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:33:52,647 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:34:29,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6236, 1.8027, 1.5233, 1.6333], device='cuda:1'), covar=tensor([0.2681, 0.2816, 0.3112, 0.2698], device='cuda:1'), in_proj_covar=tensor([0.1574, 0.1137, 0.1393, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 06:34:31,231 INFO [train.py:968] (1/2) Epoch 26, batch 200, giga_loss[loss=0.2225, simple_loss=0.3069, pruned_loss=0.06905, over 28958.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3308, pruned_loss=0.08768, over 3606952.33 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3298, pruned_loss=0.08438, over 605487.71 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3316, pruned_loss=0.08839, over 3354731.65 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:34:52,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:34:59,892 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 250, giga_loss[loss=0.2329, simple_loss=0.3058, pruned_loss=0.08003, over 28224.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3208, pruned_loss=0.08273, over 4067886.79 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3328, pruned_loss=0.08499, over 709597.74 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3202, pruned_loss=0.08286, over 3831947.28 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:35:31,299 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 06:35:54,546 INFO [train.py:968] (1/2) Epoch 26, batch 300, giga_loss[loss=0.1975, simple_loss=0.277, pruned_loss=0.05901, over 28874.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.3133, pruned_loss=0.07974, over 4426833.66 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3342, pruned_loss=0.08469, over 836055.47 frames. ], giga_tot_loss[loss=0.2356, simple_loss=0.3118, pruned_loss=0.07964, over 4203629.96 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:36:16,547 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139384.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:36:22,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139387.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 06:36:26,625 INFO [optim.py:369] (1/2) 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 06:36:30,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:36:38,359 INFO [train.py:968] (1/2) Epoch 26, batch 350, giga_loss[loss=0.2066, simple_loss=0.2848, pruned_loss=0.06417, over 28738.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.307, pruned_loss=0.07708, over 4703463.13 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3325, pruned_loss=0.08387, over 982925.24 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3051, pruned_loss=0.07684, over 4491531.76 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:36:46,103 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1139416.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 06:36:54,224 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:36:56,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:37:19,516 INFO [train.py:968] (1/2) Epoch 26, batch 400, libri_loss[loss=0.2527, simple_loss=0.3381, pruned_loss=0.08363, over 29531.00 frames. ], tot_loss[loss=0.226, simple_loss=0.3022, pruned_loss=0.07493, over 4928940.20 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3322, pruned_loss=0.08367, over 1031890.63 frames. ], giga_tot_loss[loss=0.2248, simple_loss=0.3003, pruned_loss=0.07463, over 4752767.29 frames. ], batch size: 83, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:37:21,895 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:38:00,149 INFO [train.py:968] (1/2) Epoch 26, batch 450, giga_loss[loss=0.202, simple_loss=0.283, pruned_loss=0.06048, over 28903.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3007, pruned_loss=0.07437, over 5105357.90 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3319, pruned_loss=0.08397, over 1176300.36 frames. ], giga_tot_loss[loss=0.2229, simple_loss=0.2983, pruned_loss=0.07377, over 4940086.54 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:38:29,439 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1139540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:38:31,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1139543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:38:42,381 INFO [train.py:968] (1/2) Epoch 26, batch 500, giga_loss[loss=0.2023, simple_loss=0.2809, pruned_loss=0.06187, over 28730.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2992, pruned_loss=0.074, over 5231682.36 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.332, pruned_loss=0.08355, over 1269816.52 frames. ], giga_tot_loss[loss=0.2218, simple_loss=0.2967, pruned_loss=0.07345, over 5087731.26 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:38:45,161 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9392, 1.2103, 1.2783, 1.1361], device='cuda:1'), covar=tensor([0.1822, 0.1131, 0.1931, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0751, 0.0721, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 06:38:55,919 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:39:16,482 INFO [optim.py:369] (1/2) 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 06:39:26,947 INFO [train.py:968] (1/2) Epoch 26, batch 550, giga_loss[loss=0.2047, simple_loss=0.2761, pruned_loss=0.06664, over 28405.00 frames. ], tot_loss[loss=0.222, simple_loss=0.2973, pruned_loss=0.07339, over 5327651.22 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3333, pruned_loss=0.08427, over 1329163.77 frames. ], giga_tot_loss[loss=0.2199, simple_loss=0.2945, pruned_loss=0.07266, over 5213775.17 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:40:12,448 INFO [train.py:968] (1/2) Epoch 26, batch 600, giga_loss[loss=0.2002, simple_loss=0.2819, pruned_loss=0.05927, over 28980.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2949, pruned_loss=0.07183, over 5410638.13 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3343, pruned_loss=0.08421, over 1441161.59 frames. ], giga_tot_loss[loss=0.2167, simple_loss=0.2915, pruned_loss=0.07094, over 5308727.61 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:40:50,627 INFO [optim.py:369] (1/2) 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 06:40:59,659 INFO [train.py:968] (1/2) Epoch 26, batch 650, giga_loss[loss=0.1879, simple_loss=0.2633, pruned_loss=0.05623, over 29069.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.293, pruned_loss=0.07102, over 5477322.46 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.336, pruned_loss=0.08547, over 1507008.90 frames. ], giga_tot_loss[loss=0.2144, simple_loss=0.2892, pruned_loss=0.06978, over 5391362.86 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 06:41:43,255 INFO [train.py:968] (1/2) Epoch 26, batch 700, giga_loss[loss=0.1992, simple_loss=0.2713, pruned_loss=0.06355, over 28822.00 frames. ], tot_loss[loss=0.2159, simple_loss=0.2911, pruned_loss=0.07035, over 5526688.18 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3357, pruned_loss=0.08555, over 1592902.62 frames. ], giga_tot_loss[loss=0.2127, simple_loss=0.2873, pruned_loss=0.06905, over 5452404.80 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 06:42:19,137 INFO [optim.py:369] (1/2) 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 06:42:23,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3969, 1.5874, 1.6431, 1.3115], device='cuda:1'), covar=tensor([0.1460, 0.2016, 0.1217, 0.1523], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0720, 0.0985, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 06:42:29,943 INFO [train.py:968] (1/2) Epoch 26, batch 750, giga_loss[loss=0.1859, simple_loss=0.2648, pruned_loss=0.05349, over 28959.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2894, pruned_loss=0.06987, over 5553416.46 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3351, pruned_loss=0.08514, over 1657595.21 frames. ], giga_tot_loss[loss=0.2117, simple_loss=0.2859, pruned_loss=0.06872, over 5489543.13 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 06:43:03,210 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:43:12,333 INFO [train.py:968] (1/2) Epoch 26, batch 800, giga_loss[loss=0.2496, simple_loss=0.3252, pruned_loss=0.087, over 27993.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2895, pruned_loss=0.06983, over 5586861.56 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3355, pruned_loss=0.08475, over 1742128.09 frames. ], giga_tot_loss[loss=0.2116, simple_loss=0.2857, pruned_loss=0.06874, over 5530055.19 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:43:31,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0973, 1.2301, 3.6095, 3.1504], device='cuda:1'), covar=tensor([0.1822, 0.2919, 0.0477, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0667, 0.0986, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 06:43:37,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:43:49,219 INFO [optim.py:369] (1/2) 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 06:43:50,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5007, 1.7670, 1.4865, 1.3366], device='cuda:1'), covar=tensor([0.2645, 0.2677, 0.3000, 0.2429], device='cuda:1'), in_proj_covar=tensor([0.1577, 0.1135, 0.1395, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 06:44:02,161 INFO [train.py:968] (1/2) Epoch 26, batch 850, giga_loss[loss=0.2386, simple_loss=0.3249, pruned_loss=0.07615, over 28898.00 frames. ], tot_loss[loss=0.2262, simple_loss=0.3008, pruned_loss=0.0758, over 5606684.78 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3355, pruned_loss=0.08483, over 1803453.70 frames. ], giga_tot_loss[loss=0.2235, simple_loss=0.2974, pruned_loss=0.07481, over 5558641.95 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:44:41,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1139951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:44:48,288 INFO [train.py:968] (1/2) Epoch 26, batch 900, giga_loss[loss=0.2934, simple_loss=0.3706, pruned_loss=0.1081, over 28968.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.3135, pruned_loss=0.0816, over 5629908.37 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3337, pruned_loss=0.08374, over 1883249.31 frames. ], giga_tot_loss[loss=0.2365, simple_loss=0.3109, pruned_loss=0.08107, over 5588148.70 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:45:13,271 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1139985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:45:16,032 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.1253, 5.9902, 5.6423, 3.0591], device='cuda:1'), covar=tensor([0.0529, 0.0650, 0.0873, 0.1558], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1179, 0.0991, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 06:45:22,204 INFO [optim.py:369] (1/2) 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 06:45:30,440 INFO [train.py:968] (1/2) Epoch 26, batch 950, giga_loss[loss=0.2899, simple_loss=0.3713, pruned_loss=0.1043, over 28970.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3243, pruned_loss=0.08718, over 5632615.74 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3335, pruned_loss=0.08358, over 1991885.10 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3221, pruned_loss=0.08692, over 5600785.98 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:45:34,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:45:41,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:45:56,369 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 1000, giga_loss[loss=0.2401, simple_loss=0.3316, pruned_loss=0.07429, over 28907.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3301, pruned_loss=0.08875, over 5641421.77 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3341, pruned_loss=0.08398, over 2065769.14 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.328, pruned_loss=0.08851, over 5614844.24 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:46:15,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:46:40,894 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:46:44,531 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 1050, giga_loss[loss=0.2674, simple_loss=0.3521, pruned_loss=0.09132, over 28990.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3336, pruned_loss=0.08908, over 5652368.07 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3352, pruned_loss=0.08462, over 2100220.54 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3317, pruned_loss=0.08873, over 5632559.01 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:47:10,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2365, 1.5929, 1.6005, 1.4468], device='cuda:1'), covar=tensor([0.2319, 0.2043, 0.2464, 0.2048], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0754, 0.0724, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 06:47:15,298 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 1100, giga_loss[loss=0.315, simple_loss=0.3825, pruned_loss=0.1238, over 27986.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3359, pruned_loss=0.08992, over 5658273.95 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3357, pruned_loss=0.08484, over 2212023.50 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3341, pruned_loss=0.0897, over 5636128.75 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:47:38,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 06:47:49,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4488, 1.5998, 1.1836, 1.1731], device='cuda:1'), covar=tensor([0.1091, 0.0620, 0.1103, 0.1342], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0450, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 06:48:12,360 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 1150, giga_loss[loss=0.2535, simple_loss=0.3357, pruned_loss=0.08564, over 28224.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.338, pruned_loss=0.09157, over 5665683.64 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3356, pruned_loss=0.08485, over 2285235.61 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3367, pruned_loss=0.0915, over 5643918.80 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:48:36,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:49:06,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:49:07,310 INFO [train.py:968] (1/2) Epoch 26, batch 1200, giga_loss[loss=0.262, simple_loss=0.3411, pruned_loss=0.09141, over 28987.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3416, pruned_loss=0.09411, over 5670583.19 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3353, pruned_loss=0.08469, over 2336986.76 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3407, pruned_loss=0.0942, over 5652471.50 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:49:15,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:49:41,434 INFO [optim.py:369] (1/2) 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 06:49:49,655 INFO [train.py:968] (1/2) Epoch 26, batch 1250, giga_loss[loss=0.2441, simple_loss=0.3384, pruned_loss=0.07488, over 28848.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3453, pruned_loss=0.09598, over 5680092.46 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3355, pruned_loss=0.08461, over 2390098.50 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3446, pruned_loss=0.09623, over 5662829.02 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:50:35,392 INFO [train.py:968] (1/2) Epoch 26, batch 1300, giga_loss[loss=0.2652, simple_loss=0.3436, pruned_loss=0.09344, over 28581.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3482, pruned_loss=0.09694, over 5682345.65 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.335, pruned_loss=0.08436, over 2442380.44 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.348, pruned_loss=0.0974, over 5666067.91 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:50:35,578 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:38,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:42,938 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:44,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:50:56,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:06,342 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:07,604 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:51:09,458 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:11,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:14,406 INFO [train.py:968] (1/2) Epoch 26, batch 1350, giga_loss[loss=0.278, simple_loss=0.3583, pruned_loss=0.09887, over 28873.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3498, pruned_loss=0.09702, over 5686482.72 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3357, pruned_loss=0.08426, over 2574702.15 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3499, pruned_loss=0.0979, over 5668818.40 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:51:20,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:32,859 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:36,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:51:52,664 INFO [train.py:968] (1/2) Epoch 26, batch 1400, giga_loss[loss=0.2545, simple_loss=0.3223, pruned_loss=0.09334, over 23577.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3501, pruned_loss=0.09608, over 5684571.39 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3362, pruned_loss=0.08449, over 2629321.44 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3501, pruned_loss=0.09687, over 5676818.76 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:52:22,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-13 06:52:26,628 INFO [optim.py:369] (1/2) 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 06:52:31,758 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:34,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:35,019 INFO [train.py:968] (1/2) Epoch 26, batch 1450, libri_loss[loss=0.2662, simple_loss=0.3522, pruned_loss=0.0901, over 29515.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3491, pruned_loss=0.09458, over 5689862.14 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3361, pruned_loss=0.08424, over 2707072.46 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3495, pruned_loss=0.09556, over 5681836.46 frames. ], batch size: 81, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:52:49,958 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:54,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:57,091 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:52:58,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:01,421 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:13,239 INFO [train.py:968] (1/2) Epoch 26, batch 1500, giga_loss[loss=0.242, simple_loss=0.3277, pruned_loss=0.07819, over 28571.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3479, pruned_loss=0.09293, over 5701706.36 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3367, pruned_loss=0.08446, over 2782384.01 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3483, pruned_loss=0.09384, over 5693823.66 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:53:14,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:15,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:16,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:23,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:31,402 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:33,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:40,351 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:53:46,313 INFO [optim.py:369] (1/2) 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 06:53:55,089 INFO [train.py:968] (1/2) Epoch 26, batch 1550, giga_loss[loss=0.3506, simple_loss=0.4044, pruned_loss=0.1484, over 28726.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3478, pruned_loss=0.09303, over 5697756.93 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3369, pruned_loss=0.08448, over 2813593.52 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3481, pruned_loss=0.09382, over 5689614.32 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:53:57,685 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:54:25,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:54:25,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 06:54:38,314 INFO [train.py:968] (1/2) Epoch 26, batch 1600, giga_loss[loss=0.336, simple_loss=0.3875, pruned_loss=0.1422, over 28826.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3496, pruned_loss=0.09615, over 5690870.31 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3375, pruned_loss=0.08504, over 2849632.57 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3497, pruned_loss=0.09665, over 5691181.57 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:55:15,752 INFO [optim.py:369] (1/2) 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 06:55:24,121 INFO [train.py:968] (1/2) Epoch 26, batch 1650, giga_loss[loss=0.2993, simple_loss=0.3649, pruned_loss=0.1168, over 28626.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3513, pruned_loss=0.09927, over 5698969.96 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3374, pruned_loss=0.08481, over 2965502.11 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.352, pruned_loss=0.1003, over 5695405.58 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:55:32,900 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-13 06:55:44,346 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140730.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 06:55:46,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1140732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:55:52,208 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.54 vs. limit=5.0
+2023-03-13 06:56:06,243 INFO [train.py:968] (1/2) Epoch 26, batch 1700, giga_loss[loss=0.2588, simple_loss=0.3399, pruned_loss=0.08888, over 28842.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1006, over 5702881.34 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3377, pruned_loss=0.08524, over 3023227.99 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3517, pruned_loss=0.1015, over 5696876.45 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:56:31,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:56:34,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:56:34,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.73 vs. limit=5.0
+2023-03-13 06:56:41,671 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2221, 5.0267, 4.7891, 2.3871], device='cuda:1'), covar=tensor([0.0484, 0.0668, 0.0621, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.1257, 0.1168, 0.0980, 0.0735], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 06:56:42,809 INFO [optim.py:369] (1/2) 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 06:56:50,917 INFO [train.py:968] (1/2) Epoch 26, batch 1750, giga_loss[loss=0.2448, simple_loss=0.3176, pruned_loss=0.086, over 28642.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3483, pruned_loss=0.09975, over 5691670.90 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3376, pruned_loss=0.08526, over 3078269.64 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1007, over 5685291.30 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 06:56:59,702 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:57:11,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3884, 1.9670, 1.4817, 0.7766], device='cuda:1'), covar=tensor([0.6824, 0.2918, 0.3702, 0.6721], device='cuda:1'), in_proj_covar=tensor([0.1802, 0.1693, 0.1630, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 06:57:16,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:57:33,053 INFO [train.py:968] (1/2) Epoch 26, batch 1800, giga_loss[loss=0.2416, simple_loss=0.3182, pruned_loss=0.08243, over 28613.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3469, pruned_loss=0.09914, over 5692826.19 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3372, pruned_loss=0.08514, over 3120169.98 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3478, pruned_loss=0.1002, over 5685180.32 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:57:45,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0664, 3.2454, 2.2743, 1.0641], device='cuda:1'), covar=tensor([0.8937, 0.3046, 0.4077, 0.7926], device='cuda:1'), in_proj_covar=tensor([0.1802, 0.1693, 0.1628, 0.1460], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 06:57:47,241 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1140875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:57:50,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1140878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:58:07,404 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 1850, giga_loss[loss=0.2632, simple_loss=0.3379, pruned_loss=0.09422, over 28888.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3459, pruned_loss=0.09819, over 5690091.87 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3372, pruned_loss=0.0851, over 3201860.73 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3469, pruned_loss=0.09941, over 5678976.02 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:58:15,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1140907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 06:58:19,447 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.81 vs. limit=5.0
+2023-03-13 06:58:32,500 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 06:58:55,026 INFO [train.py:968] (1/2) Epoch 26, batch 1900, libri_loss[loss=0.2512, simple_loss=0.3424, pruned_loss=0.08005, over 28672.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3431, pruned_loss=0.09543, over 5700633.40 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3375, pruned_loss=0.08506, over 3356151.64 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3441, pruned_loss=0.09711, over 5683028.98 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 06:59:01,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4765, 1.4904, 1.5278, 1.4432], device='cuda:1'), covar=tensor([0.2742, 0.2600, 0.2380, 0.2627], device='cuda:1'), in_proj_covar=tensor([0.2019, 0.1961, 0.1883, 0.2023], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 06:59:14,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2673, 1.5492, 1.5438, 1.1302], device='cuda:1'), covar=tensor([0.1831, 0.2767, 0.1506, 0.1737], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0716, 0.0978, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:1')
+2023-03-13 06:59:14,862 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1140977.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 06:59:34,221 INFO [optim.py:369] (1/2) 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 06:59:40,976 INFO [train.py:968] (1/2) Epoch 26, batch 1950, giga_loss[loss=0.2645, simple_loss=0.3329, pruned_loss=0.09808, over 28848.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3393, pruned_loss=0.09343, over 5691129.35 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3382, pruned_loss=0.08531, over 3406861.22 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3399, pruned_loss=0.0948, over 5673271.32 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:00:29,883 INFO [train.py:968] (1/2) Epoch 26, batch 2000, giga_loss[loss=0.2135, simple_loss=0.274, pruned_loss=0.07649, over 23223.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3338, pruned_loss=0.09108, over 5678662.55 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3378, pruned_loss=0.08518, over 3443707.01 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3344, pruned_loss=0.09233, over 5662052.65 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:01:05,149 INFO [optim.py:369] (1/2) 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 07:01:14,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141105.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:01:17,442 INFO [train.py:968] (1/2) Epoch 26, batch 2050, giga_loss[loss=0.2482, simple_loss=0.3234, pruned_loss=0.08649, over 28592.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3301, pruned_loss=0.08943, over 5674399.32 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.338, pruned_loss=0.08515, over 3516774.80 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3303, pruned_loss=0.09058, over 5655104.39 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:01:57,233 INFO [train.py:968] (1/2) Epoch 26, batch 2100, giga_loss[loss=0.2511, simple_loss=0.3298, pruned_loss=0.08625, over 28707.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3314, pruned_loss=0.08981, over 5668523.86 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3389, pruned_loss=0.08597, over 3575280.26 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3308, pruned_loss=0.09034, over 5657841.44 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:02:07,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2730, 3.1997, 1.3960, 1.5820], device='cuda:1'), covar=tensor([0.1124, 0.0328, 0.0966, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0560, 0.0400, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 07:02:09,449 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9304, 5.7368, 5.4378, 3.5105], device='cuda:1'), covar=tensor([0.0418, 0.0591, 0.0562, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.1252, 0.1166, 0.0978, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 07:02:22,322 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1141188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:02:31,071 INFO [optim.py:369] (1/2) 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 07:02:36,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9169, 2.1592, 1.9377, 1.8415], device='cuda:1'), covar=tensor([0.2251, 0.2586, 0.2514, 0.2636], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0755, 0.0726, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 07:02:37,366 INFO [train.py:968] (1/2) Epoch 26, batch 2150, giga_loss[loss=0.3563, simple_loss=0.3983, pruned_loss=0.1571, over 26653.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3328, pruned_loss=0.09012, over 5672128.70 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3395, pruned_loss=0.08616, over 3630722.12 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3317, pruned_loss=0.09053, over 5669004.83 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:02:41,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:03:10,104 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141248.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:03:12,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141251.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:03:16,637 INFO [train.py:968] (1/2) Epoch 26, batch 2200, giga_loss[loss=0.2352, simple_loss=0.323, pruned_loss=0.07373, over 29065.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3309, pruned_loss=0.08877, over 5679451.58 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3396, pruned_loss=0.08593, over 3684104.53 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3299, pruned_loss=0.08929, over 5674577.88 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:03:24,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-13 07:03:35,537 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141280.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:03:50,596 INFO [optim.py:369] (1/2) 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 07:03:55,468 INFO [train.py:968] (1/2) Epoch 26, batch 2250, giga_loss[loss=0.2079, simple_loss=0.2854, pruned_loss=0.06524, over 28181.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.33, pruned_loss=0.08829, over 5681607.45 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3409, pruned_loss=0.08646, over 3766594.06 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.328, pruned_loss=0.08849, over 5682559.86 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:04:13,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8034, 1.9064, 1.6290, 1.9462], device='cuda:1'), covar=tensor([0.2645, 0.2936, 0.3207, 0.2644], device='cuda:1'), in_proj_covar=tensor([0.1572, 0.1130, 0.1387, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 07:04:36,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141352.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:04:38,382 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:04:39,186 INFO [train.py:968] (1/2) Epoch 26, batch 2300, giga_loss[loss=0.2324, simple_loss=0.3101, pruned_loss=0.07733, over 28885.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3265, pruned_loss=0.08657, over 5692389.38 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3412, pruned_loss=0.08655, over 3777421.09 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3247, pruned_loss=0.08668, over 5691964.24 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:04:40,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:04:45,320 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1141364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:05:03,673 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 26, batch 2350, giga_loss[loss=0.2218, simple_loss=0.2992, pruned_loss=0.07221, over 28856.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3241, pruned_loss=0.08556, over 5693929.45 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.341, pruned_loss=0.08641, over 3819563.82 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3225, pruned_loss=0.08572, over 5690207.29 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:06:00,847 INFO [train.py:968] (1/2) Epoch 26, batch 2400, giga_loss[loss=0.2272, simple_loss=0.3039, pruned_loss=0.07523, over 28804.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3225, pruned_loss=0.08491, over 5693305.26 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3412, pruned_loss=0.08642, over 3878739.33 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3205, pruned_loss=0.08498, over 5695443.33 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:06:29,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4752, 1.9530, 1.3953, 1.6571], device='cuda:1'), covar=tensor([0.0761, 0.0289, 0.0346, 0.0857], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 07:06:30,833 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141495.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:06:31,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1212, 1.3166, 1.2564, 1.0623], device='cuda:1'), covar=tensor([0.2315, 0.2015, 0.1712, 0.2296], device='cuda:1'), in_proj_covar=tensor([0.2013, 0.1948, 0.1876, 0.2021], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 07:06:32,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141498.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 07:06:33,690 INFO [optim.py:369] (1/2) 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 07:06:38,252 INFO [train.py:968] (1/2) Epoch 26, batch 2450, giga_loss[loss=0.241, simple_loss=0.318, pruned_loss=0.082, over 29076.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3207, pruned_loss=0.08438, over 5702970.62 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3411, pruned_loss=0.08645, over 3918794.85 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3188, pruned_loss=0.08438, over 5701066.82 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:06:53,782 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141527.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:07:11,712 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1141550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:07:16,108 INFO [train.py:968] (1/2) Epoch 26, batch 2500, giga_loss[loss=0.2351, simple_loss=0.3084, pruned_loss=0.0809, over 28627.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3197, pruned_loss=0.08348, over 5714931.50 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3412, pruned_loss=0.08598, over 4003303.66 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3172, pruned_loss=0.08368, over 5707881.51 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:07:20,557 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:07:54,047 INFO [train.py:968] (1/2) Epoch 26, batch 2550, giga_loss[loss=0.2083, simple_loss=0.2843, pruned_loss=0.06611, over 28767.00 frames. ], tot_loss[loss=0.243, simple_loss=0.3195, pruned_loss=0.0833, over 5726662.97 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3421, pruned_loss=0.08633, over 4077705.48 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3159, pruned_loss=0.08312, over 5714411.34 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:08:34,114 INFO [train.py:968] (1/2) Epoch 26, batch 2600, giga_loss[loss=0.271, simple_loss=0.3488, pruned_loss=0.09661, over 28731.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.3177, pruned_loss=0.08263, over 5725730.62 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3422, pruned_loss=0.08623, over 4095977.61 frames. ], giga_tot_loss[loss=0.2398, simple_loss=0.3146, pruned_loss=0.08252, over 5714291.78 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:08:39,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2311, 1.5685, 1.5324, 1.1288], device='cuda:1'), covar=tensor([0.1848, 0.2691, 0.1538, 0.1689], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0718, 0.0981, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 07:08:54,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2935, 1.2366, 3.5437, 3.1570], device='cuda:1'), covar=tensor([0.1655, 0.2887, 0.0452, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0662, 0.0975, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:09:09,767 INFO [optim.py:369] (1/2) 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 07:09:11,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9209, 1.1380, 1.0512, 0.8296], device='cuda:1'), covar=tensor([0.2460, 0.2750, 0.1718, 0.2659], device='cuda:1'), in_proj_covar=tensor([0.2010, 0.1946, 0.1872, 0.2017], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 07:09:13,293 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:09:13,712 INFO [train.py:968] (1/2) Epoch 26, batch 2650, libri_loss[loss=0.229, simple_loss=0.3128, pruned_loss=0.07256, over 29666.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3185, pruned_loss=0.0834, over 5729301.37 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3424, pruned_loss=0.08628, over 4140746.62 frames. ], giga_tot_loss[loss=0.2409, simple_loss=0.3154, pruned_loss=0.08322, over 5716319.65 frames. ], batch size: 69, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:09:15,387 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1141709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:09:22,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8655, 2.0673, 1.9959, 1.6592], device='cuda:1'), covar=tensor([0.2215, 0.2671, 0.2330, 0.2717], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0759, 0.0728, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 07:09:39,634 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:09:41,140 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 2700, giga_loss[loss=0.2609, simple_loss=0.3376, pruned_loss=0.09204, over 28538.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3224, pruned_loss=0.08572, over 5718775.39 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3432, pruned_loss=0.08661, over 4174284.72 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3189, pruned_loss=0.08532, over 5713492.01 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:10:22,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6213, 1.8348, 1.8208, 1.4398], device='cuda:1'), covar=tensor([0.1718, 0.2363, 0.1420, 0.1625], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0716, 0.0981, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 07:10:34,650 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 2750, giga_loss[loss=0.267, simple_loss=0.3473, pruned_loss=0.09335, over 28846.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3272, pruned_loss=0.08877, over 5706494.06 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3433, pruned_loss=0.08685, over 4197914.67 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.324, pruned_loss=0.0883, over 5709454.83 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 07:11:27,110 INFO [train.py:968] (1/2) Epoch 26, batch 2800, giga_loss[loss=0.2881, simple_loss=0.3632, pruned_loss=0.1065, over 28851.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.336, pruned_loss=0.0949, over 5701300.12 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3435, pruned_loss=0.08684, over 4206071.10 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3334, pruned_loss=0.09456, over 5703112.09 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:11:35,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1509, 1.2525, 3.7051, 3.2177], device='cuda:1'), covar=tensor([0.2054, 0.3027, 0.0782, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0662, 0.0974, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:11:51,007 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1141882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:11:53,194 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:12:10,659 INFO [train.py:968] (1/2) Epoch 26, batch 2850, giga_loss[loss=0.2693, simple_loss=0.3518, pruned_loss=0.09343, over 28721.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3403, pruned_loss=0.09677, over 5700208.94 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3431, pruned_loss=0.08676, over 4246731.20 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3384, pruned_loss=0.09677, over 5698267.53 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:12:17,859 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1141914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:12:27,248 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1141925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:12:57,836 INFO [train.py:968] (1/2) Epoch 26, batch 2900, giga_loss[loss=0.255, simple_loss=0.3325, pruned_loss=0.08876, over 28592.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3439, pruned_loss=0.09751, over 5688177.14 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.343, pruned_loss=0.08683, over 4268344.84 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3423, pruned_loss=0.09765, over 5699465.51 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:13:34,792 INFO [optim.py:369] (1/2) 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 07:13:39,760 INFO [train.py:968] (1/2) Epoch 26, batch 2950, giga_loss[loss=0.3657, simple_loss=0.4124, pruned_loss=0.1596, over 27516.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3492, pruned_loss=0.1006, over 5688188.19 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3423, pruned_loss=0.08649, over 4311175.77 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3485, pruned_loss=0.1012, over 5696740.11 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:14:27,128 INFO [train.py:968] (1/2) Epoch 26, batch 3000, giga_loss[loss=0.2746, simple_loss=0.3533, pruned_loss=0.09791, over 28764.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3533, pruned_loss=0.1032, over 5667889.48 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3423, pruned_loss=0.08664, over 4333706.17 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3529, pruned_loss=0.1038, over 5678046.40 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:14:27,128 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 07:14:35,949 INFO [train.py:1012] (1/2) Epoch 26, validation: loss=0.2114, simple_loss=0.3175, pruned_loss=0.0527, over 944034.00 frames. 
+2023-03-13 07:14:35,950 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 07:14:45,269 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1142068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:14:47,058 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1142071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:15:12,220 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1142100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:15:12,751 INFO [optim.py:369] (1/2) 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 07:15:16,712 INFO [train.py:968] (1/2) Epoch 26, batch 3050, giga_loss[loss=0.2613, simple_loss=0.3378, pruned_loss=0.09238, over 28876.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3482, pruned_loss=0.09922, over 5680413.01 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.342, pruned_loss=0.08655, over 4357254.76 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3482, pruned_loss=0.1, over 5685289.47 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:15:59,217 INFO [train.py:968] (1/2) Epoch 26, batch 3100, giga_loss[loss=0.2507, simple_loss=0.3376, pruned_loss=0.08186, over 28831.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3467, pruned_loss=0.09715, over 5692575.39 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3424, pruned_loss=0.08674, over 4388389.98 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3465, pruned_loss=0.09787, over 5692418.47 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:16:18,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8097, 1.0261, 2.8286, 2.6689], device='cuda:1'), covar=tensor([0.1883, 0.2894, 0.0651, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0662, 0.0973, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:16:20,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6482, 5.4740, 5.2379, 2.4662], device='cuda:1'), covar=tensor([0.0424, 0.0599, 0.0658, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.1172, 0.0983, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 07:16:26,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1702, 1.3863, 1.2839, 1.0987], device='cuda:1'), covar=tensor([0.2898, 0.2913, 0.2054, 0.2799], device='cuda:1'), in_proj_covar=tensor([0.2018, 0.1956, 0.1886, 0.2030], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 07:16:36,400 INFO [optim.py:369] (1/2) 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 07:16:37,555 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-13 07:16:41,766 INFO [train.py:968] (1/2) Epoch 26, batch 3150, giga_loss[loss=0.2947, simple_loss=0.3643, pruned_loss=0.1126, over 26618.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3454, pruned_loss=0.09596, over 5699420.94 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3421, pruned_loss=0.08662, over 4418896.79 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3455, pruned_loss=0.09678, over 5695229.91 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:17:23,848 INFO [train.py:968] (1/2) Epoch 26, batch 3200, giga_loss[loss=0.2887, simple_loss=0.3652, pruned_loss=0.1062, over 27932.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3456, pruned_loss=0.09551, over 5702711.71 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3416, pruned_loss=0.08643, over 4454109.59 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3461, pruned_loss=0.09651, over 5696487.75 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:17:27,053 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1142260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:17:41,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3510, 1.3177, 3.9353, 3.3041], device='cuda:1'), covar=tensor([0.1678, 0.2756, 0.0437, 0.0845], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0661, 0.0973, 0.0945], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:17:42,710 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6720, 1.6534, 1.8542, 1.4534], device='cuda:1'), covar=tensor([0.1832, 0.2590, 0.1483, 0.1732], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0715, 0.0977, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:1')
+2023-03-13 07:17:55,446 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4016, 1.4569, 1.4588, 1.6492], device='cuda:1'), covar=tensor([0.0658, 0.0317, 0.0294, 0.0685], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 07:17:59,338 INFO [optim.py:369] (1/2) 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 07:17:59,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2757, 1.2181, 3.7455, 3.2233], device='cuda:1'), covar=tensor([0.1605, 0.2682, 0.0489, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0660, 0.0972, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:18:03,601 INFO [train.py:968] (1/2) Epoch 26, batch 3250, giga_loss[loss=0.3114, simple_loss=0.3744, pruned_loss=0.1241, over 27607.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3474, pruned_loss=0.09637, over 5706231.94 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3415, pruned_loss=0.08635, over 4486015.90 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.348, pruned_loss=0.09745, over 5700297.70 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:18:14,222 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-13 07:18:17,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6761, 4.5346, 4.2662, 2.1358], device='cuda:1'), covar=tensor([0.0518, 0.0630, 0.0650, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.1260, 0.1171, 0.0983, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 07:18:45,631 INFO [train.py:968] (1/2) Epoch 26, batch 3300, giga_loss[loss=0.29, simple_loss=0.3581, pruned_loss=0.1109, over 28555.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3491, pruned_loss=0.09787, over 5708710.09 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3419, pruned_loss=0.0868, over 4527238.36 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3494, pruned_loss=0.09872, over 5699166.51 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:19:20,530 INFO [optim.py:369] (1/2) 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 07:19:24,768 INFO [train.py:968] (1/2) Epoch 26, batch 3350, giga_loss[loss=0.2743, simple_loss=0.3535, pruned_loss=0.09751, over 28853.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3506, pruned_loss=0.09928, over 5712085.37 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3422, pruned_loss=0.08692, over 4567836.37 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3509, pruned_loss=0.1002, over 5699447.82 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:19:55,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.88 vs. limit=2.0
+2023-03-13 07:20:02,091 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-13 07:20:06,567 INFO [train.py:968] (1/2) Epoch 26, batch 3400, libri_loss[loss=0.2243, simple_loss=0.3134, pruned_loss=0.06767, over 29558.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3508, pruned_loss=0.09951, over 5724622.82 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3422, pruned_loss=0.08685, over 4607327.69 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3513, pruned_loss=0.1007, over 5709452.07 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:20:41,878 INFO [optim.py:369] (1/2) 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 07:20:47,413 INFO [train.py:968] (1/2) Epoch 26, batch 3450, giga_loss[loss=0.3039, simple_loss=0.3797, pruned_loss=0.1141, over 28845.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.351, pruned_loss=0.09973, over 5728925.62 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3416, pruned_loss=0.08632, over 4638723.53 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3521, pruned_loss=0.1013, over 5713511.13 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:21:27,048 INFO [train.py:968] (1/2) Epoch 26, batch 3500, giga_loss[loss=0.2446, simple_loss=0.3241, pruned_loss=0.08253, over 28672.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3513, pruned_loss=0.0992, over 5717000.59 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3415, pruned_loss=0.08629, over 4648426.12 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3523, pruned_loss=0.1007, over 5710434.63 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:21:43,931 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6010, 1.8918, 1.8725, 1.6679], device='cuda:1'), covar=tensor([0.2229, 0.2248, 0.2308, 0.2202], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0758, 0.0729, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 07:22:02,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2882, 1.5511, 1.3704, 1.6238], device='cuda:1'), covar=tensor([0.0823, 0.0336, 0.0341, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 07:22:03,195 INFO [optim.py:369] (1/2) 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 07:22:04,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 07:22:07,248 INFO [train.py:968] (1/2) Epoch 26, batch 3550, giga_loss[loss=0.2616, simple_loss=0.345, pruned_loss=0.08913, over 28741.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3523, pruned_loss=0.09902, over 5717345.92 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3416, pruned_loss=0.08633, over 4681154.87 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3534, pruned_loss=0.1005, over 5710723.48 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:22:27,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5752, 1.8050, 1.4534, 1.6294], device='cuda:1'), covar=tensor([0.2661, 0.2693, 0.3125, 0.2370], device='cuda:1'), in_proj_covar=tensor([0.1575, 0.1135, 0.1392, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 07:22:30,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1142635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:22:49,340 INFO [train.py:968] (1/2) Epoch 26, batch 3600, libri_loss[loss=0.2458, simple_loss=0.3163, pruned_loss=0.08767, over 29502.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3511, pruned_loss=0.09794, over 5720216.35 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3413, pruned_loss=0.08622, over 4693374.55 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3523, pruned_loss=0.09931, over 5713323.26 frames. ], batch size: 70, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:23:25,786 INFO [optim.py:369] (1/2) 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 07:23:29,552 INFO [train.py:968] (1/2) Epoch 26, batch 3650, libri_loss[loss=0.2656, simple_loss=0.3485, pruned_loss=0.09138, over 29529.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3492, pruned_loss=0.09702, over 5726808.89 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3415, pruned_loss=0.08631, over 4722870.26 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3502, pruned_loss=0.09831, over 5717462.96 frames. ], batch size: 84, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:23:55,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5107, 1.8181, 1.4606, 1.4714], device='cuda:1'), covar=tensor([0.2737, 0.2699, 0.3115, 0.2419], device='cuda:1'), in_proj_covar=tensor([0.1574, 0.1134, 0.1389, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 07:24:06,204 INFO [train.py:968] (1/2) Epoch 26, batch 3700, giga_loss[loss=0.251, simple_loss=0.3363, pruned_loss=0.08292, over 28886.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3469, pruned_loss=0.09587, over 5730259.68 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3408, pruned_loss=0.08582, over 4780695.09 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3486, pruned_loss=0.09778, over 5714502.57 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:24:22,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1142778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:24:24,346 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:24:43,325 INFO [train.py:968] (1/2) Epoch 26, batch 3750, giga_loss[loss=0.2588, simple_loss=0.331, pruned_loss=0.09333, over 28478.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3448, pruned_loss=0.09475, over 5734209.20 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3406, pruned_loss=0.08563, over 4801924.13 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3464, pruned_loss=0.0966, over 5719475.85 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:24:45,680 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1142810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:25:01,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8596, 1.4118, 5.1242, 3.7415], device='cuda:1'), covar=tensor([0.1637, 0.2978, 0.0371, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0786, 0.0663, 0.0975, 0.0948], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:25:28,072 INFO [train.py:968] (1/2) Epoch 26, batch 3800, giga_loss[loss=0.3078, simple_loss=0.3726, pruned_loss=0.1215, over 27573.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3457, pruned_loss=0.09593, over 5731707.48 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3405, pruned_loss=0.08566, over 4812980.15 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3471, pruned_loss=0.09745, over 5718603.27 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:26:01,987 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 07:26:02,165 INFO [optim.py:369] (1/2) 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 07:26:06,508 INFO [train.py:968] (1/2) Epoch 26, batch 3850, giga_loss[loss=0.2783, simple_loss=0.3536, pruned_loss=0.1015, over 28437.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3466, pruned_loss=0.09601, over 5736227.47 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3405, pruned_loss=0.08558, over 4837717.94 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3478, pruned_loss=0.09751, over 5724055.26 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:26:15,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4734, 4.3296, 4.0439, 1.9817], device='cuda:1'), covar=tensor([0.0553, 0.0687, 0.0727, 0.2127], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1174, 0.0989, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 07:26:45,464 INFO [train.py:968] (1/2) Epoch 26, batch 3900, giga_loss[loss=0.2332, simple_loss=0.3238, pruned_loss=0.07132, over 29121.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3457, pruned_loss=0.09493, over 5731707.61 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3406, pruned_loss=0.08577, over 4869712.77 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3467, pruned_loss=0.09627, over 5717217.67 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:27:09,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1142986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:27:23,350 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 3950, giga_loss[loss=0.2734, simple_loss=0.3458, pruned_loss=0.1005, over 28904.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3457, pruned_loss=0.09473, over 5724280.55 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3409, pruned_loss=0.08588, over 4880731.86 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3464, pruned_loss=0.09586, over 5717619.64 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:27:37,740 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-13 07:28:06,532 INFO [train.py:968] (1/2) Epoch 26, batch 4000, giga_loss[loss=0.2671, simple_loss=0.3359, pruned_loss=0.09913, over 28854.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3438, pruned_loss=0.09427, over 5717099.19 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3409, pruned_loss=0.08584, over 4901798.20 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3444, pruned_loss=0.0954, over 5711636.01 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:28:28,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6002, 4.4137, 1.7331, 1.7382], device='cuda:1'), covar=tensor([0.1020, 0.0198, 0.0904, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0554, 0.0397, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 07:28:44,936 INFO [optim.py:369] (1/2) 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 07:28:47,582 INFO [train.py:968] (1/2) Epoch 26, batch 4050, giga_loss[loss=0.2419, simple_loss=0.3152, pruned_loss=0.08425, over 28843.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3411, pruned_loss=0.09309, over 5718024.73 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3406, pruned_loss=0.08576, over 4921575.37 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3418, pruned_loss=0.09419, over 5710581.48 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:29:05,849 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 07:29:24,945 INFO [train.py:968] (1/2) Epoch 26, batch 4100, giga_loss[loss=0.2492, simple_loss=0.3283, pruned_loss=0.08505, over 29036.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3396, pruned_loss=0.09249, over 5719230.89 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3413, pruned_loss=0.08617, over 4950354.18 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3396, pruned_loss=0.09324, over 5708673.37 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:29:37,232 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2728, 0.8629, 1.0107, 1.4001], device='cuda:1'), covar=tensor([0.0810, 0.0383, 0.0356, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 07:29:51,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4789, 1.3122, 4.0631, 3.2027], device='cuda:1'), covar=tensor([0.1676, 0.3015, 0.0462, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0661, 0.0974, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:30:00,762 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 4150, giga_loss[loss=0.2527, simple_loss=0.3236, pruned_loss=0.09086, over 28528.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.339, pruned_loss=0.09266, over 5718007.50 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3411, pruned_loss=0.08613, over 4981678.21 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3391, pruned_loss=0.09351, over 5704829.63 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:30:15,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-13 07:30:42,206 INFO [train.py:968] (1/2) Epoch 26, batch 4200, giga_loss[loss=0.2696, simple_loss=0.3469, pruned_loss=0.09613, over 28420.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3381, pruned_loss=0.09274, over 5716189.80 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3414, pruned_loss=0.08644, over 5016727.81 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3379, pruned_loss=0.09342, over 5699785.92 frames. ], batch size: 369, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:30:53,644 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3956, 1.5650, 1.4656, 1.5232], device='cuda:1'), covar=tensor([0.0733, 0.0364, 0.0332, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 07:31:21,449 INFO [optim.py:369] (1/2) 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 07:31:23,614 INFO [train.py:968] (1/2) Epoch 26, batch 4250, giga_loss[loss=0.2229, simple_loss=0.3022, pruned_loss=0.07182, over 28884.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3359, pruned_loss=0.09203, over 5713813.60 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3411, pruned_loss=0.08626, over 5029711.03 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3359, pruned_loss=0.09279, over 5698512.81 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:31:33,734 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3764, 3.3108, 1.4776, 1.6311], device='cuda:1'), covar=tensor([0.1016, 0.0332, 0.0905, 0.1299], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0556, 0.0397, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 07:32:00,457 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:32:03,474 INFO [train.py:968] (1/2) Epoch 26, batch 4300, giga_loss[loss=0.2407, simple_loss=0.3191, pruned_loss=0.08117, over 28152.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3329, pruned_loss=0.09093, over 5719449.25 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3407, pruned_loss=0.08605, over 5038519.81 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3331, pruned_loss=0.09174, over 5705718.93 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:32:09,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:32:26,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5956, 1.8298, 1.5333, 1.7308], device='cuda:1'), covar=tensor([0.2669, 0.2869, 0.3220, 0.2486], device='cuda:1'), in_proj_covar=tensor([0.1573, 0.1132, 0.1388, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 07:32:41,356 INFO [optim.py:369] (1/2) 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 07:32:43,293 INFO [train.py:968] (1/2) Epoch 26, batch 4350, giga_loss[loss=0.2506, simple_loss=0.3248, pruned_loss=0.08819, over 28870.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3304, pruned_loss=0.08994, over 5717762.32 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3407, pruned_loss=0.0861, over 5055358.09 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3304, pruned_loss=0.09062, over 5703797.14 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:33:12,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1829, 2.3992, 2.4362, 2.1667], device='cuda:1'), covar=tensor([0.1573, 0.1646, 0.1418, 0.1481], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0754, 0.0725, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 07:33:20,441 INFO [train.py:968] (1/2) Epoch 26, batch 4400, giga_loss[loss=0.2305, simple_loss=0.3133, pruned_loss=0.07385, over 28982.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3285, pruned_loss=0.08856, over 5704536.07 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3409, pruned_loss=0.08629, over 5061175.69 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3281, pruned_loss=0.08901, over 5706236.51 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:33:51,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:33:58,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1143504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:33:59,278 INFO [optim.py:369] (1/2) 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 07:34:00,496 INFO [train.py:968] (1/2) Epoch 26, batch 4450, giga_loss[loss=0.3221, simple_loss=0.3943, pruned_loss=0.125, over 28334.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3315, pruned_loss=0.08978, over 5703615.84 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.341, pruned_loss=0.08638, over 5081339.55 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3307, pruned_loss=0.09015, over 5706352.24 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:34:00,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1143507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:34:26,348 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1143536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:34:26,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3715, 1.6399, 1.6024, 1.2177], device='cuda:1'), covar=tensor([0.1962, 0.2617, 0.1691, 0.1936], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0713, 0.0976, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:1')
+2023-03-13 07:34:45,209 INFO [train.py:968] (1/2) Epoch 26, batch 4500, giga_loss[loss=0.2687, simple_loss=0.344, pruned_loss=0.09669, over 28995.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3347, pruned_loss=0.09115, over 5699441.47 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.341, pruned_loss=0.08647, over 5097862.13 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3338, pruned_loss=0.09144, over 5698165.43 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:35:08,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:35:10,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2614, 1.6619, 0.9984, 1.1832], device='cuda:1'), covar=tensor([0.1346, 0.0809, 0.1700, 0.1504], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0449, 0.0522, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 07:35:23,130 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 4550, giga_loss[loss=0.2852, simple_loss=0.3575, pruned_loss=0.1065, over 27631.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3368, pruned_loss=0.0915, over 5702672.17 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3405, pruned_loss=0.08619, over 5108708.31 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3365, pruned_loss=0.092, over 5700111.01 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:36:07,755 INFO [train.py:968] (1/2) Epoch 26, batch 4600, giga_loss[loss=0.2922, simple_loss=0.3668, pruned_loss=0.1088, over 28668.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3394, pruned_loss=0.09225, over 5697097.46 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3413, pruned_loss=0.08669, over 5130841.60 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3385, pruned_loss=0.09243, over 5693204.74 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:36:11,456 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:36:16,748 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:36:46,659 INFO [train.py:968] (1/2) Epoch 26, batch 4650, giga_loss[loss=0.2337, simple_loss=0.3194, pruned_loss=0.07401, over 28902.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3391, pruned_loss=0.09188, over 5699744.23 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3416, pruned_loss=0.08708, over 5157252.95 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.338, pruned_loss=0.09188, over 5690314.02 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:37:03,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:37:03,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:37:18,441 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:37:25,737 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143756.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:37:26,142 INFO [train.py:968] (1/2) Epoch 26, batch 4700, giga_loss[loss=0.2505, simple_loss=0.3206, pruned_loss=0.09022, over 28412.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3387, pruned_loss=0.09212, over 5706929.38 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3419, pruned_loss=0.08727, over 5175842.34 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3375, pruned_loss=0.09207, over 5694724.43 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:37:53,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1143790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:38:06,602 INFO [optim.py:369] (1/2) 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 07:38:07,901 INFO [train.py:968] (1/2) Epoch 26, batch 4750, giga_loss[loss=0.2909, simple_loss=0.3665, pruned_loss=0.1077, over 28773.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.34, pruned_loss=0.09327, over 5705541.95 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3423, pruned_loss=0.08753, over 5186156.90 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3388, pruned_loss=0.0931, over 5693682.33 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:38:38,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5112, 1.4807, 1.2312, 1.1079], device='cuda:1'), covar=tensor([0.0775, 0.0370, 0.0833, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0447, 0.0519, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 07:38:49,180 INFO [train.py:968] (1/2) Epoch 26, batch 4800, giga_loss[loss=0.2615, simple_loss=0.3422, pruned_loss=0.09037, over 28765.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3407, pruned_loss=0.09344, over 5695260.66 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3427, pruned_loss=0.08779, over 5188028.85 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3393, pruned_loss=0.09317, over 5691665.51 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:38:55,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5715, 1.6934, 1.2163, 1.3045], device='cuda:1'), covar=tensor([0.0949, 0.0594, 0.0994, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0446, 0.0519, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 07:39:00,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:39:00,915 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1143870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:39:03,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1143873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:39:27,465 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:39:30,369 INFO [train.py:968] (1/2) Epoch 26, batch 4850, giga_loss[loss=0.2847, simple_loss=0.3548, pruned_loss=0.1073, over 28443.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3443, pruned_loss=0.0954, over 5681743.06 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3432, pruned_loss=0.08813, over 5184920.82 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3428, pruned_loss=0.09505, over 5693092.69 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:40:01,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9648, 2.3018, 2.2084, 1.6638], device='cuda:1'), covar=tensor([0.3282, 0.2380, 0.2575, 0.3144], device='cuda:1'), in_proj_covar=tensor([0.2034, 0.1976, 0.1909, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 07:40:10,610 INFO [train.py:968] (1/2) Epoch 26, batch 4900, libri_loss[loss=0.2784, simple_loss=0.3645, pruned_loss=0.09617, over 29658.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3465, pruned_loss=0.09607, over 5701287.84 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3431, pruned_loss=0.08821, over 5213510.95 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3454, pruned_loss=0.09602, over 5702309.10 frames. ], batch size: 88, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:40:14,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1143962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:40:15,638 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 07:40:51,091 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 4950, giga_loss[loss=0.2709, simple_loss=0.352, pruned_loss=0.09484, over 28631.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3489, pruned_loss=0.0977, over 5707533.97 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3435, pruned_loss=0.08851, over 5227084.62 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3479, pruned_loss=0.09758, over 5704959.59 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:40:56,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:40:57,569 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:40:58,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:14,149 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:18,856 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:20,214 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:41:30,951 INFO [train.py:968] (1/2) Epoch 26, batch 5000, giga_loss[loss=0.2752, simple_loss=0.3469, pruned_loss=0.1018, over 28205.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3486, pruned_loss=0.09707, over 5716887.67 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3432, pruned_loss=0.08834, over 5240528.09 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3481, pruned_loss=0.09728, over 5711402.99 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:41:34,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9970, 4.8507, 2.0570, 2.0341], device='cuda:1'), covar=tensor([0.0827, 0.0346, 0.0785, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0559, 0.0399, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 07:41:38,291 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:05,488 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:42:08,249 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:09,143 INFO [train.py:968] (1/2) Epoch 26, batch 5050, giga_loss[loss=0.2686, simple_loss=0.3521, pruned_loss=0.09262, over 28632.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3485, pruned_loss=0.09738, over 5725236.49 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3433, pruned_loss=0.08839, over 5260082.44 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3482, pruned_loss=0.09776, over 5715930.11 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:42:10,107 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:19,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:21,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:30,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144131.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:42:34,895 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:42:50,408 INFO [train.py:968] (1/2) Epoch 26, batch 5100, giga_loss[loss=0.242, simple_loss=0.3191, pruned_loss=0.08251, over 28756.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3467, pruned_loss=0.09669, over 5720778.10 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3435, pruned_loss=0.08864, over 5267045.24 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3463, pruned_loss=0.09689, over 5713330.94 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:42:57,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:08,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:12,116 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:15,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:18,344 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:43:32,070 INFO [train.py:968] (1/2) Epoch 26, batch 5150, giga_loss[loss=0.2259, simple_loss=0.3028, pruned_loss=0.07443, over 28684.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3438, pruned_loss=0.09533, over 5718541.22 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3438, pruned_loss=0.08884, over 5270077.71 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3434, pruned_loss=0.09549, over 5717408.92 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:43:36,279 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:43:40,177 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:00,939 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:03,035 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 5200, giga_loss[loss=0.2546, simple_loss=0.331, pruned_loss=0.08914, over 28564.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3407, pruned_loss=0.09352, over 5723806.89 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3435, pruned_loss=0.08874, over 5291380.53 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09394, over 5719701.47 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:44:17,758 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:20,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:25,821 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144274.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:44:27,929 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:27,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144277.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:44:42,140 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:44:51,161 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144306.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 07:44:51,490 INFO [train.py:968] (1/2) Epoch 26, batch 5250, giga_loss[loss=0.2966, simple_loss=0.3739, pruned_loss=0.1097, over 28254.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3408, pruned_loss=0.09322, over 5709350.18 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3437, pruned_loss=0.08888, over 5293537.21 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3405, pruned_loss=0.09351, over 5711280.30 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:44:52,289 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:44:55,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:45:04,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7450, 1.9400, 1.8877, 1.6655], device='cuda:1'), covar=tensor([0.2422, 0.2023, 0.1806, 0.2092], device='cuda:1'), in_proj_covar=tensor([0.2028, 0.1976, 0.1904, 0.2035], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 07:45:18,222 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:45:32,815 INFO [train.py:968] (1/2) Epoch 26, batch 5300, giga_loss[loss=0.2846, simple_loss=0.37, pruned_loss=0.09963, over 28727.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3429, pruned_loss=0.09282, over 5708478.39 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3444, pruned_loss=0.08937, over 5308694.51 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3419, pruned_loss=0.09274, over 5705684.02 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:45:59,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:46:14,603 INFO [optim.py:369] (1/2) 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 07:46:14,616 INFO [train.py:968] (1/2) Epoch 26, batch 5350, giga_loss[loss=0.2329, simple_loss=0.3108, pruned_loss=0.07746, over 29075.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3422, pruned_loss=0.0926, over 5707465.65 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3438, pruned_loss=0.08904, over 5323532.16 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3419, pruned_loss=0.09291, over 5700615.12 frames. ], batch size: 113, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:46:26,074 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1749, 1.1214, 1.1935, 1.2987], device='cuda:1'), covar=tensor([0.0747, 0.0381, 0.0312, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 07:46:35,227 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-13 07:46:40,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:46:40,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:46:52,267 INFO [train.py:968] (1/2) Epoch 26, batch 5400, giga_loss[loss=0.2624, simple_loss=0.3376, pruned_loss=0.09363, over 28602.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3406, pruned_loss=0.09241, over 5712723.53 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3435, pruned_loss=0.08889, over 5337196.49 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3405, pruned_loss=0.09286, over 5703444.23 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:47:05,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1204, 1.3217, 1.0758, 0.8695], device='cuda:1'), covar=tensor([0.1027, 0.0506, 0.1188, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0400, 0.0444, 0.0516, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 07:47:06,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1099, 1.2122, 3.3757, 3.0097], device='cuda:1'), covar=tensor([0.1667, 0.2710, 0.0494, 0.1008], device='cuda:1'), in_proj_covar=tensor([0.0783, 0.0661, 0.0975, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 07:47:24,340 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:47:36,335 INFO [train.py:968] (1/2) Epoch 26, batch 5450, giga_loss[loss=0.2887, simple_loss=0.3576, pruned_loss=0.1099, over 27698.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.34, pruned_loss=0.0938, over 5705759.95 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3436, pruned_loss=0.08893, over 5342999.32 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3398, pruned_loss=0.09416, over 5696704.58 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:47:54,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:47:58,810 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:17,313 INFO [train.py:968] (1/2) Epoch 26, batch 5500, giga_loss[loss=0.3011, simple_loss=0.3647, pruned_loss=0.1187, over 28371.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3392, pruned_loss=0.09466, over 5705185.79 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3437, pruned_loss=0.0889, over 5348285.16 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.339, pruned_loss=0.09501, over 5696695.14 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:48:23,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:39,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:41,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:41,976 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:48:59,656 INFO [optim.py:369] (1/2) 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 07:48:59,670 INFO [train.py:968] (1/2) Epoch 26, batch 5550, giga_loss[loss=0.306, simple_loss=0.3701, pruned_loss=0.1209, over 28616.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3374, pruned_loss=0.09448, over 5707307.84 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08887, over 5351104.73 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3373, pruned_loss=0.0948, over 5699792.57 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:49:06,010 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:49:24,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:49:26,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1144640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:49:39,030 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-13 07:49:39,292 INFO [train.py:968] (1/2) Epoch 26, batch 5600, libri_loss[loss=0.2935, simple_loss=0.3708, pruned_loss=0.1081, over 26156.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3365, pruned_loss=0.09381, over 5715687.27 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3438, pruned_loss=0.08896, over 5367020.26 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.336, pruned_loss=0.09414, over 5706891.48 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:49:49,174 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:50:15,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:50:19,319 INFO [train.py:968] (1/2) Epoch 26, batch 5650, giga_loss[loss=0.2277, simple_loss=0.3034, pruned_loss=0.07604, over 28882.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3317, pruned_loss=0.09164, over 5722294.00 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3439, pruned_loss=0.08897, over 5372460.34 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3312, pruned_loss=0.09193, over 5713657.00 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:50:19,993 INFO [optim.py:369] (1/2) 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 07:50:59,616 INFO [train.py:968] (1/2) Epoch 26, batch 5700, giga_loss[loss=0.2542, simple_loss=0.327, pruned_loss=0.09067, over 28906.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3278, pruned_loss=0.08947, over 5724390.73 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3437, pruned_loss=0.08902, over 5381544.44 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3272, pruned_loss=0.08965, over 5715861.74 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:51:33,124 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:51:38,198 INFO [train.py:968] (1/2) Epoch 26, batch 5750, giga_loss[loss=0.2645, simple_loss=0.3375, pruned_loss=0.09576, over 28830.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3278, pruned_loss=0.08913, over 5726808.49 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3439, pruned_loss=0.08916, over 5397097.92 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3268, pruned_loss=0.08919, over 5715098.79 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:51:38,358 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1144807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:51:38,732 INFO [optim.py:369] (1/2) 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 07:51:41,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 07:51:43,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:52:18,270 INFO [train.py:968] (1/2) Epoch 26, batch 5800, giga_loss[loss=0.2408, simple_loss=0.3282, pruned_loss=0.07667, over 29061.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3312, pruned_loss=0.09007, over 5732321.14 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3445, pruned_loss=0.08945, over 5411748.45 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3294, pruned_loss=0.08989, over 5718490.98 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:52:28,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4746, 2.1980, 1.5796, 0.7809], device='cuda:1'), covar=tensor([0.6583, 0.3091, 0.4604, 0.6979], device='cuda:1'), in_proj_covar=tensor([0.1797, 0.1677, 0.1622, 0.1462], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 07:52:57,939 INFO [train.py:968] (1/2) Epoch 26, batch 5850, giga_loss[loss=0.2342, simple_loss=0.3146, pruned_loss=0.07692, over 28902.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3337, pruned_loss=0.09082, over 5731208.17 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3442, pruned_loss=0.08927, over 5424752.09 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3323, pruned_loss=0.09087, over 5717013.16 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:52:58,514 INFO [optim.py:369] (1/2) 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 07:53:02,567 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 07:53:36,979 INFO [train.py:968] (1/2) Epoch 26, batch 5900, giga_loss[loss=0.2999, simple_loss=0.3855, pruned_loss=0.1072, over 28541.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3377, pruned_loss=0.09238, over 5725658.39 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3446, pruned_loss=0.08956, over 5433409.71 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.336, pruned_loss=0.09223, over 5713379.22 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:53:38,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1144958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:53:40,984 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1144963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:54:07,622 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1144990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:54:22,681 INFO [train.py:968] (1/2) Epoch 26, batch 5950, giga_loss[loss=0.2656, simple_loss=0.3416, pruned_loss=0.0948, over 28991.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3399, pruned_loss=0.09336, over 5721157.45 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3449, pruned_loss=0.08969, over 5440815.05 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3382, pruned_loss=0.0932, over 5709761.10 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:54:23,409 INFO [optim.py:369] (1/2) 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 07:54:38,879 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 07:54:48,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-13 07:55:03,110 INFO [train.py:968] (1/2) Epoch 26, batch 6000, giga_loss[loss=0.2728, simple_loss=0.3516, pruned_loss=0.09703, over 28602.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3422, pruned_loss=0.09485, over 5722329.80 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3452, pruned_loss=0.08993, over 5459180.65 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3404, pruned_loss=0.09468, over 5706173.47 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:55:03,110 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 07:55:11,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2812, 1.8794, 1.4609, 0.4614], device='cuda:1'), covar=tensor([0.5235, 0.4026, 0.4834, 0.6743], device='cuda:1'), in_proj_covar=tensor([0.1795, 0.1676, 0.1621, 0.1461], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 07:55:11,781 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 07:55:29,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:55:55,167 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:55:55,470 INFO [train.py:968] (1/2) Epoch 26, batch 6050, giga_loss[loss=0.3058, simple_loss=0.3707, pruned_loss=0.1205, over 28895.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3498, pruned_loss=0.1011, over 5718626.42 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3455, pruned_loss=0.09007, over 5464692.76 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3481, pruned_loss=0.101, over 5705636.76 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:55:56,278 INFO [optim.py:369] (1/2) 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 07:55:58,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:56:26,868 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:56:43,958 INFO [train.py:968] (1/2) Epoch 26, batch 6100, giga_loss[loss=0.2657, simple_loss=0.3436, pruned_loss=0.09392, over 28662.00 frames. ], tot_loss[loss=0.2832, simple_loss=0.355, pruned_loss=0.1057, over 5700207.43 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3454, pruned_loss=0.0901, over 5471519.42 frames. ], giga_tot_loss[loss=0.2828, simple_loss=0.354, pruned_loss=0.1059, over 5686881.31 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:57:03,382 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:57:07,522 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 6150, giga_loss[loss=0.3508, simple_loss=0.4149, pruned_loss=0.1433, over 28662.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3628, pruned_loss=0.1113, over 5692367.95 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3454, pruned_loss=0.09013, over 5478024.12 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3622, pruned_loss=0.1116, over 5680001.20 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:57:33,566 INFO [optim.py:369] (1/2) 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 07:57:44,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:57:47,666 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:58:17,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:58:20,024 INFO [train.py:968] (1/2) Epoch 26, batch 6200, giga_loss[loss=0.357, simple_loss=0.4055, pruned_loss=0.1542, over 28850.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3681, pruned_loss=0.1164, over 5683124.77 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3451, pruned_loss=0.09002, over 5488420.60 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3684, pruned_loss=0.1173, over 5668538.00 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 07:58:56,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-13 07:59:03,674 INFO [train.py:968] (1/2) Epoch 26, batch 6250, libri_loss[loss=0.2514, simple_loss=0.3275, pruned_loss=0.08768, over 29343.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3726, pruned_loss=0.1201, over 5681484.98 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3447, pruned_loss=0.08984, over 5486829.20 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3737, pruned_loss=0.1217, over 5675129.46 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:59:05,159 INFO [optim.py:369] (1/2) 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 07:59:15,587 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:17,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:19,858 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1145325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:22,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1145328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:46,241 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 07:59:52,032 INFO [train.py:968] (1/2) Epoch 26, batch 6300, giga_loss[loss=0.3414, simple_loss=0.41, pruned_loss=0.1364, over 28978.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3763, pruned_loss=0.1236, over 5662140.77 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3444, pruned_loss=0.08951, over 5499092.63 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3783, pruned_loss=0.126, over 5651167.89 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 07:59:53,261 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1145357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:00:43,459 INFO [train.py:968] (1/2) Epoch 26, batch 6350, giga_loss[loss=0.3996, simple_loss=0.4207, pruned_loss=0.1892, over 23433.00 frames. ], tot_loss[loss=0.315, simple_loss=0.378, pruned_loss=0.1261, over 5655580.86 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3439, pruned_loss=0.0892, over 5504458.09 frames. ], giga_tot_loss[loss=0.3189, simple_loss=0.3804, pruned_loss=0.1287, over 5644205.75 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:00:44,794 INFO [optim.py:369] (1/2) 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 08:00:57,831 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 08:01:36,522 INFO [train.py:968] (1/2) Epoch 26, batch 6400, giga_loss[loss=0.2822, simple_loss=0.3518, pruned_loss=0.1063, over 28704.00 frames. ], tot_loss[loss=0.3205, simple_loss=0.3812, pruned_loss=0.1299, over 5639353.15 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3438, pruned_loss=0.089, over 5510408.08 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.384, pruned_loss=0.133, over 5627406.48 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:02:00,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3147, 1.2257, 1.1708, 1.5486], device='cuda:1'), covar=tensor([0.0780, 0.0369, 0.0349, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 08:02:30,406 INFO [train.py:968] (1/2) Epoch 26, batch 6450, giga_loss[loss=0.3912, simple_loss=0.4109, pruned_loss=0.1858, over 23431.00 frames. ], tot_loss[loss=0.3267, simple_loss=0.3852, pruned_loss=0.1341, over 5610325.15 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08899, over 5514761.20 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3883, pruned_loss=0.1375, over 5599816.35 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:02:32,303 INFO [optim.py:369] (1/2) 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 08:02:32,489 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1145510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:03:02,367 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1145541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:03:19,809 INFO [train.py:968] (1/2) Epoch 26, batch 6500, libri_loss[loss=0.2658, simple_loss=0.3486, pruned_loss=0.09148, over 29546.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3871, pruned_loss=0.1352, over 5619015.40 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08898, over 5518869.46 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3903, pruned_loss=0.1387, over 5608850.92 frames. ], batch size: 83, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:03:22,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5477, 1.3619, 1.6194, 1.1824], device='cuda:1'), covar=tensor([0.2002, 0.3591, 0.1603, 0.1966], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0711, 0.0969, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 08:03:58,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5163, 1.8799, 1.8750, 1.5959], device='cuda:1'), covar=tensor([0.2149, 0.2041, 0.2141, 0.2180], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0757, 0.0727, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 08:04:07,347 INFO [train.py:968] (1/2) Epoch 26, batch 6550, giga_loss[loss=0.3007, simple_loss=0.3625, pruned_loss=0.1195, over 29009.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3858, pruned_loss=0.1347, over 5639701.82 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3437, pruned_loss=0.08898, over 5526755.00 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3894, pruned_loss=0.1386, over 5626828.79 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:04:09,497 INFO [optim.py:369] (1/2) 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 08:04:29,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2622, 1.4632, 1.4333, 1.2183], device='cuda:1'), covar=tensor([0.2411, 0.2389, 0.1931, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.2033, 0.1979, 0.1908, 0.2038], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 08:04:57,368 INFO [train.py:968] (1/2) Epoch 26, batch 6600, giga_loss[loss=0.3269, simple_loss=0.3894, pruned_loss=0.1322, over 28648.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3845, pruned_loss=0.1341, over 5634386.72 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3434, pruned_loss=0.08897, over 5526311.03 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3886, pruned_loss=0.1383, over 5626474.07 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:05:28,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9221, 1.3209, 1.0839, 0.1525], device='cuda:1'), covar=tensor([0.4626, 0.3290, 0.4726, 0.7283], device='cuda:1'), in_proj_covar=tensor([0.1807, 0.1694, 0.1631, 0.1470], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 08:05:48,571 INFO [train.py:968] (1/2) Epoch 26, batch 6650, giga_loss[loss=0.3964, simple_loss=0.4235, pruned_loss=0.1847, over 26540.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3861, pruned_loss=0.1347, over 5634269.65 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3435, pruned_loss=0.08912, over 5529137.20 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3896, pruned_loss=0.1383, over 5626496.85 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:05:53,826 INFO [optim.py:369] (1/2) 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 08:06:34,832 INFO [train.py:968] (1/2) Epoch 26, batch 6700, giga_loss[loss=0.4016, simple_loss=0.417, pruned_loss=0.1931, over 23317.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.386, pruned_loss=0.1338, over 5631987.24 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3437, pruned_loss=0.08915, over 5536840.78 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3899, pruned_loss=0.1379, over 5621910.27 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:06:42,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8472, 2.0229, 1.7869, 1.8065], device='cuda:1'), covar=tensor([0.2053, 0.2328, 0.2355, 0.2058], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0758, 0.0728, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 08:07:21,351 INFO [train.py:968] (1/2) Epoch 26, batch 6750, giga_loss[loss=0.4396, simple_loss=0.4552, pruned_loss=0.212, over 24117.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.385, pruned_loss=0.1323, over 5618286.36 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3433, pruned_loss=0.0888, over 5541242.62 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3897, pruned_loss=0.1373, over 5608718.31 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 1.0
+2023-03-13 08:07:25,958 INFO [optim.py:369] (1/2) 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 08:07:27,415 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7103, 1.9133, 1.6357, 1.6909], device='cuda:1'), covar=tensor([0.2171, 0.2039, 0.2117, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.1571, 0.1133, 0.1388, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 08:08:12,515 INFO [train.py:968] (1/2) Epoch 26, batch 6800, giga_loss[loss=0.3364, simple_loss=0.3732, pruned_loss=0.1498, over 23893.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3818, pruned_loss=0.1292, over 5610948.90 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3436, pruned_loss=0.08896, over 5542101.03 frames. ], giga_tot_loss[loss=0.3271, simple_loss=0.3862, pruned_loss=0.134, over 5604371.69 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:08:39,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:08:43,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0885, 2.1919, 1.6439, 1.7579], device='cuda:1'), covar=tensor([0.0982, 0.0729, 0.1020, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0451, 0.0522, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 08:08:57,701 INFO [train.py:968] (1/2) Epoch 26, batch 6850, libri_loss[loss=0.2791, simple_loss=0.3534, pruned_loss=0.1024, over 29565.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3786, pruned_loss=0.1251, over 5630703.97 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3432, pruned_loss=0.08875, over 5552819.00 frames. ], giga_tot_loss[loss=0.322, simple_loss=0.3836, pruned_loss=0.1303, over 5617429.10 frames. ], batch size: 74, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:09:02,915 INFO [optim.py:369] (1/2) 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 08:09:05,949 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1145916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:09:19,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-13 08:09:21,259 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-13 08:09:25,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.12 vs. limit=5.0
+2023-03-13 08:09:44,816 INFO [train.py:968] (1/2) Epoch 26, batch 6900, giga_loss[loss=0.2833, simple_loss=0.3553, pruned_loss=0.1057, over 28544.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3745, pruned_loss=0.1215, over 5645216.03 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.343, pruned_loss=0.08867, over 5562995.94 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3797, pruned_loss=0.1268, over 5627970.49 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:10:07,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2600, 3.1071, 2.9665, 1.5145], device='cuda:1'), covar=tensor([0.1028, 0.1059, 0.0938, 0.2235], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.1191, 0.1005, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 08:10:32,237 INFO [train.py:968] (1/2) Epoch 26, batch 6950, giga_loss[loss=0.2946, simple_loss=0.3628, pruned_loss=0.1132, over 29120.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.373, pruned_loss=0.1207, over 5650213.12 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3429, pruned_loss=0.08857, over 5569645.12 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3778, pruned_loss=0.1255, over 5632261.17 frames. ], batch size: 113, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:10:37,566 INFO [optim.py:369] (1/2) 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 08:10:37,773 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:10:50,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7400, 1.9712, 1.3866, 1.5739], device='cuda:1'), covar=tensor([0.1025, 0.0694, 0.1103, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0451, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 08:10:52,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:10:54,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:03,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9014, 3.7366, 3.5518, 1.7574], device='cuda:1'), covar=tensor([0.0693, 0.0801, 0.0741, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.1189, 0.1004, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 08:11:19,090 INFO [train.py:968] (1/2) Epoch 26, batch 7000, giga_loss[loss=0.2949, simple_loss=0.3571, pruned_loss=0.1163, over 28617.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3707, pruned_loss=0.1194, over 5651199.00 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3429, pruned_loss=0.08873, over 5569065.18 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3753, pruned_loss=0.1239, over 5639325.63 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:11:19,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1933, 1.4435, 1.4010, 1.1599], device='cuda:1'), covar=tensor([0.2897, 0.2680, 0.1807, 0.2484], device='cuda:1'), in_proj_covar=tensor([0.2038, 0.1980, 0.1914, 0.2043], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 08:11:21,161 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:21,812 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:25,112 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:11:50,058 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:12:06,078 INFO [train.py:968] (1/2) Epoch 26, batch 7050, giga_loss[loss=0.3181, simple_loss=0.3702, pruned_loss=0.133, over 28944.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3705, pruned_loss=0.1192, over 5656131.53 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.08871, over 5570430.95 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3747, pruned_loss=0.1233, over 5646597.59 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:12:07,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5611, 1.6261, 1.2426, 1.2584], device='cuda:1'), covar=tensor([0.0897, 0.0538, 0.0960, 0.1196], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0451, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 08:12:12,693 INFO [optim.py:369] (1/2) 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 08:13:03,605 INFO [train.py:968] (1/2) Epoch 26, batch 7100, giga_loss[loss=0.2881, simple_loss=0.3535, pruned_loss=0.1114, over 28677.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3695, pruned_loss=0.1183, over 5660623.60 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08843, over 5575278.53 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3738, pruned_loss=0.1221, over 5650041.93 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:13:17,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9439, 3.7191, 3.5561, 1.5995], device='cuda:1'), covar=tensor([0.0766, 0.0953, 0.0918, 0.2270], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1192, 0.1006, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 08:13:48,953 INFO [train.py:968] (1/2) Epoch 26, batch 7150, libri_loss[loss=0.2361, simple_loss=0.305, pruned_loss=0.08364, over 29396.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3673, pruned_loss=0.1153, over 5678454.56 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.08828, over 5588481.08 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3722, pruned_loss=0.1196, over 5661171.42 frames. ], batch size: 67, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:13:56,302 INFO [optim.py:369] (1/2) 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 08:14:17,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5052, 2.1153, 1.5278, 0.8321], device='cuda:1'), covar=tensor([0.5477, 0.3089, 0.4179, 0.6578], device='cuda:1'), in_proj_covar=tensor([0.1813, 0.1696, 0.1637, 0.1473], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 08:14:19,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 08:14:43,174 INFO [train.py:968] (1/2) Epoch 26, batch 7200, giga_loss[loss=0.2895, simple_loss=0.37, pruned_loss=0.1045, over 29038.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3682, pruned_loss=0.1143, over 5674321.75 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3408, pruned_loss=0.0879, over 5596390.70 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3734, pruned_loss=0.1187, over 5655151.79 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:15:25,973 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5002, 1.6155, 1.7226, 1.3089], device='cuda:1'), covar=tensor([0.1715, 0.2500, 0.1412, 0.1693], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0710, 0.0968, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 08:15:30,700 INFO [train.py:968] (1/2) Epoch 26, batch 7250, giga_loss[loss=0.28, simple_loss=0.3521, pruned_loss=0.1039, over 28759.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3702, pruned_loss=0.1157, over 5663369.73 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.088, over 5596850.22 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3754, pruned_loss=0.1202, over 5650428.06 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:15:34,062 INFO [optim.py:369] (1/2) 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 08:16:17,546 INFO [train.py:968] (1/2) Epoch 26, batch 7300, giga_loss[loss=0.2714, simple_loss=0.3541, pruned_loss=0.09432, over 28813.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3693, pruned_loss=0.1154, over 5675143.75 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3404, pruned_loss=0.08771, over 5601904.38 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3746, pruned_loss=0.1199, over 5661988.19 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:16:24,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8556, 1.1369, 2.8345, 2.7126], device='cuda:1'), covar=tensor([0.1659, 0.2614, 0.0589, 0.1723], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0668, 0.0986, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 08:16:49,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1146387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:17:05,357 INFO [train.py:968] (1/2) Epoch 26, batch 7350, giga_loss[loss=0.348, simple_loss=0.382, pruned_loss=0.157, over 23456.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3678, pruned_loss=0.1148, over 5676320.50 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08758, over 5609054.77 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3727, pruned_loss=0.1191, over 5660740.15 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:17:10,733 INFO [optim.py:369] (1/2) 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 08:17:25,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:17:55,596 INFO [train.py:968] (1/2) Epoch 26, batch 7400, giga_loss[loss=0.3214, simple_loss=0.3778, pruned_loss=0.1325, over 28721.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3664, pruned_loss=0.1152, over 5671869.67 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08771, over 5611646.66 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1188, over 5658014.64 frames. ], batch size: 243, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:18:25,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4037, 1.8109, 1.5179, 1.5012], device='cuda:1'), covar=tensor([0.0754, 0.0321, 0.0318, 0.0859], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 08:18:44,603 INFO [train.py:968] (1/2) Epoch 26, batch 7450, giga_loss[loss=0.3153, simple_loss=0.3886, pruned_loss=0.121, over 28901.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3666, pruned_loss=0.1157, over 5676144.37 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08771, over 5611646.66 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3697, pruned_loss=0.1185, over 5665360.83 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:18:52,373 INFO [optim.py:369] (1/2) 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 08:19:08,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:19:11,168 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:19:27,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5462, 1.9464, 1.6368, 1.6226], device='cuda:1'), covar=tensor([0.2467, 0.2440, 0.2480, 0.2508], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0758, 0.0726, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 08:19:32,714 INFO [train.py:968] (1/2) Epoch 26, batch 7500, giga_loss[loss=0.3429, simple_loss=0.3749, pruned_loss=0.1555, over 23643.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3667, pruned_loss=0.114, over 5688831.47 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3408, pruned_loss=0.08784, over 5614415.18 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3695, pruned_loss=0.1168, over 5679090.04 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:19:37,335 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:20:18,661 INFO [train.py:968] (1/2) Epoch 26, batch 7550, giga_loss[loss=0.2993, simple_loss=0.3719, pruned_loss=0.1134, over 28896.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3673, pruned_loss=0.1137, over 5699582.38 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3411, pruned_loss=0.08803, over 5618864.55 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3696, pruned_loss=0.116, over 5688943.57 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:20:21,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7642, 1.9695, 1.6341, 1.8115], device='cuda:1'), covar=tensor([0.3304, 0.3257, 0.3717, 0.3009], device='cuda:1'), in_proj_covar=tensor([0.1572, 0.1135, 0.1388, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 08:20:23,699 INFO [optim.py:369] (1/2) 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 08:20:37,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3197, 1.6829, 1.2761, 0.8779], device='cuda:1'), covar=tensor([0.4738, 0.2697, 0.2788, 0.5874], device='cuda:1'), in_proj_covar=tensor([0.1816, 0.1701, 0.1642, 0.1475], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 08:20:58,074 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146651.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 08:21:01,657 INFO [train.py:968] (1/2) Epoch 26, batch 7600, giga_loss[loss=0.3426, simple_loss=0.4088, pruned_loss=0.1382, over 28824.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3671, pruned_loss=0.1138, over 5697891.68 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.341, pruned_loss=0.08791, over 5626071.21 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3697, pruned_loss=0.1165, over 5684972.70 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:21:44,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-13 08:21:51,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2653, 1.3138, 3.4084, 2.9741], device='cuda:1'), covar=tensor([0.1484, 0.2666, 0.0479, 0.1808], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0669, 0.0990, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 08:21:52,111 INFO [train.py:968] (1/2) Epoch 26, batch 7650, giga_loss[loss=0.3671, simple_loss=0.422, pruned_loss=0.1561, over 27536.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.3661, pruned_loss=0.1142, over 5699400.58 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08764, over 5630324.15 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.369, pruned_loss=0.1169, over 5686383.60 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:21:56,473 INFO [optim.py:369] (1/2) 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 08:22:42,540 INFO [train.py:968] (1/2) Epoch 26, batch 7700, giga_loss[loss=0.3439, simple_loss=0.3981, pruned_loss=0.1449, over 28762.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3666, pruned_loss=0.1152, over 5688766.67 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3408, pruned_loss=0.08787, over 5631308.27 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3691, pruned_loss=0.1177, over 5678751.70 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:23:24,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1146800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:23:30,151 INFO [train.py:968] (1/2) Epoch 26, batch 7750, giga_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1025, over 28965.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3649, pruned_loss=0.1145, over 5695583.32 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08772, over 5634178.79 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3676, pruned_loss=0.1172, over 5686215.62 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:23:36,428 INFO [optim.py:369] (1/2) 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 08:24:17,632 INFO [train.py:968] (1/2) Epoch 26, batch 7800, giga_loss[loss=0.3096, simple_loss=0.3688, pruned_loss=0.1252, over 28599.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3635, pruned_loss=0.1143, over 5690515.70 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08776, over 5629525.44 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.366, pruned_loss=0.1168, over 5689159.20 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:25:03,929 INFO [train.py:968] (1/2) Epoch 26, batch 7850, giga_loss[loss=0.279, simple_loss=0.3505, pruned_loss=0.1037, over 28809.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3625, pruned_loss=0.1144, over 5696729.58 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.341, pruned_loss=0.08791, over 5634628.78 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3646, pruned_loss=0.1168, over 5692245.64 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:25:07,792 INFO [optim.py:369] (1/2) 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 08:25:35,548 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1146943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:25:38,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1146946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:25:48,738 INFO [train.py:968] (1/2) Epoch 26, batch 7900, giga_loss[loss=0.2927, simple_loss=0.3627, pruned_loss=0.1113, over 28632.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1143, over 5698064.92 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08767, over 5639664.68 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3642, pruned_loss=0.1169, over 5691255.54 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:26:01,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1146969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:26:07,953 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1146975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:26:35,874 INFO [train.py:968] (1/2) Epoch 26, batch 7950, giga_loss[loss=0.2759, simple_loss=0.3584, pruned_loss=0.09671, over 28904.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3627, pruned_loss=0.1144, over 5683397.29 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08767, over 5635409.16 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3651, pruned_loss=0.117, over 5682104.94 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:26:42,117 INFO [optim.py:369] (1/2) 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 08:26:55,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147026.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:27:13,307 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3203, 1.8399, 1.5938, 1.6128], device='cuda:1'), covar=tensor([0.2271, 0.1855, 0.2662, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0756, 0.0724, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 08:27:22,355 INFO [train.py:968] (1/2) Epoch 26, batch 8000, giga_loss[loss=0.275, simple_loss=0.35, pruned_loss=0.1, over 28744.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3631, pruned_loss=0.1142, over 5673292.10 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08758, over 5635918.42 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3654, pruned_loss=0.1166, over 5672366.32 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:28:06,922 INFO [train.py:968] (1/2) Epoch 26, batch 8050, giga_loss[loss=0.2854, simple_loss=0.3555, pruned_loss=0.1077, over 29032.00 frames. ], tot_loss[loss=0.295, simple_loss=0.363, pruned_loss=0.1135, over 5666987.84 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3407, pruned_loss=0.0878, over 5639877.50 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3653, pruned_loss=0.1161, over 5664327.30 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:28:14,202 INFO [optim.py:369] (1/2) 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 08:28:51,633 INFO [train.py:968] (1/2) Epoch 26, batch 8100, giga_loss[loss=0.313, simple_loss=0.3804, pruned_loss=0.1228, over 29009.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3635, pruned_loss=0.1136, over 5683417.97 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.341, pruned_loss=0.08796, over 5648501.22 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3658, pruned_loss=0.1163, over 5674614.14 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:29:03,452 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1147169.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 08:29:08,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1147172.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:29:35,626 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1147201.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:29:36,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1147203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:29:40,471 INFO [train.py:968] (1/2) Epoch 26, batch 8150, giga_loss[loss=0.3217, simple_loss=0.3817, pruned_loss=0.1309, over 28953.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3663, pruned_loss=0.116, over 5678772.36 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.088, over 5651780.93 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3685, pruned_loss=0.1189, over 5669472.64 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:29:49,283 INFO [optim.py:369] (1/2) 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 08:30:29,965 INFO [train.py:968] (1/2) Epoch 26, batch 8200, giga_loss[loss=0.2734, simple_loss=0.3494, pruned_loss=0.09875, over 28855.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1186, over 5690093.21 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3408, pruned_loss=0.08776, over 5659254.94 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3708, pruned_loss=0.1219, over 5676772.92 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:31:18,289 INFO [train.py:968] (1/2) Epoch 26, batch 8250, giga_loss[loss=0.2853, simple_loss=0.3518, pruned_loss=0.1094, over 28929.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1203, over 5682554.06 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08775, over 5666810.74 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3719, pruned_loss=0.1238, over 5665861.15 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:31:24,039 INFO [optim.py:369] (1/2) 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 08:31:29,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3253, 1.4974, 1.2970, 1.2823], device='cuda:1'), covar=tensor([0.2140, 0.2274, 0.2071, 0.1973], device='cuda:1'), in_proj_covar=tensor([0.2044, 0.1987, 0.1918, 0.2048], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 08:31:55,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:32:07,317 INFO [train.py:968] (1/2) Epoch 26, batch 8300, giga_loss[loss=0.3133, simple_loss=0.3767, pruned_loss=0.1249, over 28989.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3691, pruned_loss=0.1211, over 5676261.44 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08775, over 5670452.68 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3722, pruned_loss=0.1245, over 5660068.45 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:32:51,363 INFO [train.py:968] (1/2) Epoch 26, batch 8350, giga_loss[loss=0.2306, simple_loss=0.3125, pruned_loss=0.07441, over 28485.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3659, pruned_loss=0.1189, over 5675944.80 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3398, pruned_loss=0.08736, over 5675745.42 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.37, pruned_loss=0.1229, over 5657969.88 frames. ], batch size: 60, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:32:56,870 INFO [optim.py:369] (1/2) 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 08:33:31,262 INFO [train.py:968] (1/2) Epoch 26, batch 8400, giga_loss[loss=0.2745, simple_loss=0.3532, pruned_loss=0.09795, over 28845.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3672, pruned_loss=0.1185, over 5677303.33 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3402, pruned_loss=0.08753, over 5672800.06 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3708, pruned_loss=0.1224, over 5666352.99 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:33:53,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3675, 1.7601, 1.6791, 1.3996], device='cuda:1'), covar=tensor([0.2223, 0.1902, 0.2402, 0.2360], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0751, 0.0721, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 08:34:00,199 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1147485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:34:02,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4190, 3.2248, 1.5275, 1.4877], device='cuda:1'), covar=tensor([0.1040, 0.0359, 0.0913, 0.1367], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0568, 0.0402, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 08:34:03,022 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1147487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:34:05,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1147490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:34:19,807 INFO [train.py:968] (1/2) Epoch 26, batch 8450, libri_loss[loss=0.2402, simple_loss=0.3249, pruned_loss=0.0778, over 29540.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3655, pruned_loss=0.1168, over 5669180.74 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3401, pruned_loss=0.08746, over 5676296.42 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3688, pruned_loss=0.1204, over 5657265.78 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:34:22,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-13 08:34:26,228 INFO [optim.py:369] (1/2) 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 08:34:29,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1147519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:35:01,293 INFO [train.py:968] (1/2) Epoch 26, batch 8500, giga_loss[loss=0.2984, simple_loss=0.3448, pruned_loss=0.1259, over 23675.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3632, pruned_loss=0.115, over 5661764.33 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3398, pruned_loss=0.08744, over 5665710.93 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3674, pruned_loss=0.1192, over 5662284.97 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:35:16,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7127, 3.5647, 3.3994, 1.8149], device='cuda:1'), covar=tensor([0.0840, 0.0909, 0.0864, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1193, 0.1007, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 08:35:21,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:35:47,141 INFO [train.py:968] (1/2) Epoch 26, batch 8550, giga_loss[loss=0.2694, simple_loss=0.3394, pruned_loss=0.09972, over 28905.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3623, pruned_loss=0.1153, over 5673593.88 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3397, pruned_loss=0.08731, over 5670754.81 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3662, pruned_loss=0.1194, over 5669416.77 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:35:54,444 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 8600, giga_loss[loss=0.4794, simple_loss=0.4808, pruned_loss=0.2389, over 26731.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3625, pruned_loss=0.1163, over 5663336.76 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3395, pruned_loss=0.08729, over 5674399.96 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.366, pruned_loss=0.1199, over 5656615.07 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:36:54,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2287, 1.4105, 1.3795, 1.1683], device='cuda:1'), covar=tensor([0.2639, 0.2602, 0.1807, 0.2359], device='cuda:1'), in_proj_covar=tensor([0.2036, 0.1982, 0.1913, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 08:37:21,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-13 08:37:26,566 INFO [train.py:968] (1/2) Epoch 26, batch 8650, giga_loss[loss=0.2992, simple_loss=0.3695, pruned_loss=0.1145, over 28952.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.364, pruned_loss=0.1165, over 5660124.96 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3393, pruned_loss=0.08726, over 5671894.87 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3676, pruned_loss=0.1202, over 5655788.68 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:37:29,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1009, 0.9491, 0.9936, 1.2831], device='cuda:1'), covar=tensor([0.0832, 0.0365, 0.0356, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 08:37:32,514 INFO [optim.py:369] (1/2) 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 08:37:40,951 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1147721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:37:43,771 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1147724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:38:12,358 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1147753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:38:15,723 INFO [train.py:968] (1/2) Epoch 26, batch 8700, giga_loss[loss=0.3057, simple_loss=0.375, pruned_loss=0.1182, over 28944.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3678, pruned_loss=0.1161, over 5654110.20 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3398, pruned_loss=0.0875, over 5662110.94 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3706, pruned_loss=0.1191, over 5658549.17 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:39:02,070 INFO [train.py:968] (1/2) Epoch 26, batch 8750, giga_loss[loss=0.3345, simple_loss=0.3952, pruned_loss=0.1369, over 28237.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3705, pruned_loss=0.1169, over 5665217.37 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3398, pruned_loss=0.08753, over 5666363.02 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3731, pruned_loss=0.1198, over 5664990.23 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:39:09,247 INFO [optim.py:369] (1/2) 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 08:39:34,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 08:39:43,760 INFO [train.py:968] (1/2) Epoch 26, batch 8800, giga_loss[loss=0.3668, simple_loss=0.4143, pruned_loss=0.1596, over 28867.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3716, pruned_loss=0.1181, over 5672479.81 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3397, pruned_loss=0.08754, over 5670826.83 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3745, pruned_loss=0.121, over 5668167.27 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:39:48,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1147860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:40:29,416 INFO [train.py:968] (1/2) Epoch 26, batch 8850, giga_loss[loss=0.3029, simple_loss=0.3674, pruned_loss=0.1192, over 28941.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3726, pruned_loss=0.1195, over 5663438.12 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08741, over 5675580.55 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3756, pruned_loss=0.1225, over 5655681.88 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:40:39,324 INFO [optim.py:369] (1/2) 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 08:40:56,372 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1147935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:41:05,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.85 vs. limit=5.0
+2023-03-13 08:41:08,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4950, 1.6299, 1.5966, 1.4014], device='cuda:1'), covar=tensor([0.2914, 0.2765, 0.2240, 0.2687], device='cuda:1'), in_proj_covar=tensor([0.2035, 0.1981, 0.1915, 0.2044], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 08:41:14,862 INFO [train.py:968] (1/2) Epoch 26, batch 8900, giga_loss[loss=0.3264, simple_loss=0.3817, pruned_loss=0.1355, over 28640.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3717, pruned_loss=0.1199, over 5664274.63 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3392, pruned_loss=0.08731, over 5681758.82 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3753, pruned_loss=0.1233, over 5652052.50 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:41:51,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3012, 1.6864, 1.4540, 1.4326], device='cuda:1'), covar=tensor([0.0735, 0.0348, 0.0317, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 08:42:01,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:42:03,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1148006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:42:04,192 INFO [train.py:968] (1/2) Epoch 26, batch 8950, giga_loss[loss=0.2797, simple_loss=0.3549, pruned_loss=0.1023, over 28705.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3706, pruned_loss=0.12, over 5642691.71 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3394, pruned_loss=0.08749, over 5675563.65 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.374, pruned_loss=0.1232, over 5637418.77 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:42:11,179 INFO [optim.py:369] (1/2) 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 08:42:28,419 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1148035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:42:49,616 INFO [train.py:968] (1/2) Epoch 26, batch 9000, giga_loss[loss=0.2988, simple_loss=0.3641, pruned_loss=0.1168, over 28809.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3682, pruned_loss=0.1185, over 5648324.45 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3395, pruned_loss=0.08748, over 5671717.98 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3717, pruned_loss=0.1221, over 5647668.53 frames. ], batch size: 112, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:42:49,617 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 08:42:53,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1645, 1.2499, 3.3782, 3.1268], device='cuda:1'), covar=tensor([0.1921, 0.3140, 0.0566, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0671, 0.0989, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 08:42:58,105 INFO [train.py:1012] (1/2) Epoch 26, validation: loss=0.2061, simple_loss=0.3142, pruned_loss=0.04894, over 944034.00 frames. 
+2023-03-13 08:42:58,105 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 08:43:05,331 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5875, 1.8406, 1.5210, 1.7462], device='cuda:1'), covar=tensor([0.2309, 0.2381, 0.2497, 0.2198], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1135, 0.1393, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 08:43:24,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6081, 2.4963, 1.5604, 0.8543], device='cuda:1'), covar=tensor([0.7604, 0.3242, 0.3573, 0.5946], device='cuda:1'), in_proj_covar=tensor([0.1813, 0.1704, 0.1636, 0.1474], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 08:43:41,560 INFO [train.py:968] (1/2) Epoch 26, batch 9050, giga_loss[loss=0.3153, simple_loss=0.3724, pruned_loss=0.1291, over 28521.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.367, pruned_loss=0.1181, over 5653845.71 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3389, pruned_loss=0.08699, over 5676242.00 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3716, pruned_loss=0.1227, over 5648078.02 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:43:48,747 INFO [optim.py:369] (1/2) 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 08:43:51,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-13 08:44:28,461 INFO [train.py:968] (1/2) Epoch 26, batch 9100, giga_loss[loss=0.276, simple_loss=0.353, pruned_loss=0.0995, over 28844.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1183, over 5654854.57 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3393, pruned_loss=0.08722, over 5681663.80 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3714, pruned_loss=0.1229, over 5644434.79 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:45:15,678 INFO [train.py:968] (1/2) Epoch 26, batch 9150, giga_loss[loss=0.2974, simple_loss=0.363, pruned_loss=0.1159, over 28836.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5656163.31 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3395, pruned_loss=0.08735, over 5683405.83 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3717, pruned_loss=0.1238, over 5645109.06 frames. ], batch size: 285, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 08:45:24,474 INFO [optim.py:369] (1/2) 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 08:45:25,325 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:45:49,373 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-13 08:45:58,072 INFO [train.py:968] (1/2) Epoch 26, batch 9200, giga_loss[loss=0.3071, simple_loss=0.3675, pruned_loss=0.1233, over 28704.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3648, pruned_loss=0.1174, over 5667006.66 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3398, pruned_loss=0.08736, over 5691215.32 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3687, pruned_loss=0.122, over 5649879.38 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:46:47,857 INFO [train.py:968] (1/2) Epoch 26, batch 9250, giga_loss[loss=0.3138, simple_loss=0.3593, pruned_loss=0.1342, over 23619.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.365, pruned_loss=0.1179, over 5658169.67 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.08726, over 5695533.58 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3687, pruned_loss=0.1222, over 5640284.10 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:46:50,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1148310.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:46:54,949 INFO [optim.py:369] (1/2) 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 08:47:17,971 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:47:33,916 INFO [train.py:968] (1/2) Epoch 26, batch 9300, libri_loss[loss=0.2396, simple_loss=0.3172, pruned_loss=0.08097, over 29565.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3657, pruned_loss=0.1171, over 5673133.98 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3392, pruned_loss=0.08701, over 5701844.47 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.37, pruned_loss=0.1219, over 5651747.32 frames. ], batch size: 74, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:48:03,928 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-13 08:48:21,001 INFO [train.py:968] (1/2) Epoch 26, batch 9350, libri_loss[loss=0.3503, simple_loss=0.4183, pruned_loss=0.1412, over 28643.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3686, pruned_loss=0.1197, over 5662964.24 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3396, pruned_loss=0.08728, over 5701978.25 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 5645905.26 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:48:30,927 INFO [optim.py:369] (1/2) 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 08:49:02,015 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:49:04,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1148456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:49:04,703 INFO [train.py:968] (1/2) Epoch 26, batch 9400, giga_loss[loss=0.2735, simple_loss=0.3568, pruned_loss=0.09512, over 28913.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3686, pruned_loss=0.1199, over 5664531.20 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3394, pruned_loss=0.08706, over 5706095.32 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.372, pruned_loss=0.1238, over 5646292.91 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:49:28,572 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1148485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:49:50,066 INFO [train.py:968] (1/2) Epoch 26, batch 9450, giga_loss[loss=0.2976, simple_loss=0.3722, pruned_loss=0.1115, over 29098.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3706, pruned_loss=0.1184, over 5672850.67 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3399, pruned_loss=0.08736, over 5709705.26 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3736, pruned_loss=0.122, over 5654332.50 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:49:59,680 INFO [optim.py:369] (1/2) 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 08:50:34,428 INFO [train.py:968] (1/2) Epoch 26, batch 9500, giga_loss[loss=0.3054, simple_loss=0.3755, pruned_loss=0.1176, over 28966.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3719, pruned_loss=0.1175, over 5679064.40 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3398, pruned_loss=0.08742, over 5713612.05 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3749, pruned_loss=0.1208, over 5660098.82 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:50:46,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-13 08:51:11,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1148593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:51:20,947 INFO [train.py:968] (1/2) Epoch 26, batch 9550, giga_loss[loss=0.3023, simple_loss=0.3701, pruned_loss=0.1172, over 28246.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.374, pruned_loss=0.1178, over 5686508.61 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3399, pruned_loss=0.08744, over 5717324.57 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.377, pruned_loss=0.1211, over 5667404.11 frames. ], batch size: 77, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:51:31,441 INFO [optim.py:369] (1/2) 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 08:51:36,160 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 08:52:06,554 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 08:52:08,930 INFO [train.py:968] (1/2) Epoch 26, batch 9600, giga_loss[loss=0.3332, simple_loss=0.3918, pruned_loss=0.1373, over 28997.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3777, pruned_loss=0.122, over 5685612.09 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.34, pruned_loss=0.08745, over 5718413.94 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3801, pruned_loss=0.1247, over 5669590.08 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:52:56,126 INFO [train.py:968] (1/2) Epoch 26, batch 9650, giga_loss[loss=0.268, simple_loss=0.3444, pruned_loss=0.09579, over 28877.00 frames. ], tot_loss[loss=0.3146, simple_loss=0.3797, pruned_loss=0.1247, over 5670922.23 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3403, pruned_loss=0.08783, over 5722804.70 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3822, pruned_loss=0.1274, over 5653235.19 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:53:06,212 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:53:26,562 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:53:28,891 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148744.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:53:42,819 INFO [train.py:968] (1/2) Epoch 26, batch 9700, giga_loss[loss=0.341, simple_loss=0.3913, pruned_loss=0.1454, over 27551.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3792, pruned_loss=0.1244, over 5677138.77 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08813, over 5725960.62 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3817, pruned_loss=0.1271, over 5658754.70 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:53:51,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1148768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:54:24,242 INFO [train.py:968] (1/2) Epoch 26, batch 9750, giga_loss[loss=0.4028, simple_loss=0.4518, pruned_loss=0.177, over 28905.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3778, pruned_loss=0.1228, over 5679751.14 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3411, pruned_loss=0.0885, over 5726700.65 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3804, pruned_loss=0.1254, over 5663108.31 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:54:34,302 INFO [optim.py:369] (1/2) 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 08:55:08,356 INFO [train.py:968] (1/2) Epoch 26, batch 9800, giga_loss[loss=0.2848, simple_loss=0.3684, pruned_loss=0.1006, over 29105.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3765, pruned_loss=0.1201, over 5685811.61 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3406, pruned_loss=0.08822, over 5731176.97 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3801, pruned_loss=0.1234, over 5667035.78 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:55:12,232 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1148860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:55:14,495 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1148863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:55:39,211 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 9850, giga_loss[loss=0.3334, simple_loss=0.3957, pruned_loss=0.1355, over 28629.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3761, pruned_loss=0.1193, over 5688354.89 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3406, pruned_loss=0.08818, over 5734677.17 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.38, pruned_loss=0.1229, over 5668700.42 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:56:00,416 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:56:24,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1116, 1.3151, 1.2553, 1.0223], device='cuda:1'), covar=tensor([0.3020, 0.2753, 0.2038, 0.2756], device='cuda:1'), in_proj_covar=tensor([0.2038, 0.1982, 0.1914, 0.2048], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 08:56:37,800 INFO [train.py:968] (1/2) Epoch 26, batch 9900, giga_loss[loss=0.3016, simple_loss=0.3722, pruned_loss=0.1155, over 28581.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.377, pruned_loss=0.1206, over 5677115.32 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3407, pruned_loss=0.08832, over 5729775.22 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3816, pruned_loss=0.1247, over 5662873.40 frames. ], batch size: 60, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:57:00,721 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1148980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 08:57:10,572 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 08:57:25,129 INFO [train.py:968] (1/2) Epoch 26, batch 9950, giga_loss[loss=0.2792, simple_loss=0.354, pruned_loss=0.1022, over 28633.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3763, pruned_loss=0.1207, over 5660214.42 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3411, pruned_loss=0.08848, over 5721494.62 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3803, pruned_loss=0.1245, over 5655486.26 frames. ], batch size: 242, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:57:35,041 INFO [optim.py:369] (1/2) 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 08:57:44,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149028.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 08:58:11,942 INFO [train.py:968] (1/2) Epoch 26, batch 10000, giga_loss[loss=0.3015, simple_loss=0.3733, pruned_loss=0.1148, over 28819.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3738, pruned_loss=0.1204, over 5658236.51 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3406, pruned_loss=0.08822, over 5728323.98 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3786, pruned_loss=0.1248, over 5646538.53 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 08:58:53,590 INFO [train.py:968] (1/2) Epoch 26, batch 10050, giga_loss[loss=0.2895, simple_loss=0.3656, pruned_loss=0.1067, over 28891.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3717, pruned_loss=0.1192, over 5671639.24 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08848, over 5730166.09 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3766, pruned_loss=0.1239, over 5657841.49 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:59:06,940 INFO [optim.py:369] (1/2) 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 08:59:07,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149119.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 08:59:18,287 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-13 08:59:22,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-13 08:59:41,192 INFO [train.py:968] (1/2) Epoch 26, batch 10100, giga_loss[loss=0.2865, simple_loss=0.3579, pruned_loss=0.1075, over 28732.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3701, pruned_loss=0.1191, over 5657611.86 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3412, pruned_loss=0.08859, over 5725547.47 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3748, pruned_loss=0.1237, over 5649069.46 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 08:59:53,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149171.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 08:59:57,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149174.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:00:15,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2327, 3.4342, 1.3016, 1.6465], device='cuda:1'), covar=tensor([0.1253, 0.0456, 0.1002, 0.1472], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0568, 0.0403, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 09:00:25,625 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149203.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:00:28,827 INFO [train.py:968] (1/2) Epoch 26, batch 10150, giga_loss[loss=0.2921, simple_loss=0.358, pruned_loss=0.1131, over 28897.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3692, pruned_loss=0.1194, over 5662466.82 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3409, pruned_loss=0.0886, over 5729891.63 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.374, pruned_loss=0.1239, over 5649970.83 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:00:30,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-13 09:00:38,906 INFO [optim.py:369] (1/2) 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 09:00:52,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 09:01:15,759 INFO [train.py:968] (1/2) Epoch 26, batch 10200, giga_loss[loss=0.2615, simple_loss=0.3413, pruned_loss=0.09087, over 28976.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 5660584.91 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3409, pruned_loss=0.0886, over 5729891.63 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3719, pruned_loss=0.1227, over 5650859.12 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:01:22,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-13 09:01:22,650 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149262.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:01:24,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149265.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:01:47,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4893, 1.5895, 1.7254, 1.5263], device='cuda:1'), covar=tensor([0.3233, 0.2962, 0.2392, 0.2703], device='cuda:1'), in_proj_covar=tensor([0.2049, 0.1994, 0.1929, 0.2060], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 09:01:49,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:01:51,163 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149294.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:01:55,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149299.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:02:01,808 INFO [train.py:968] (1/2) Epoch 26, batch 10250, giga_loss[loss=0.276, simple_loss=0.3535, pruned_loss=0.09926, over 28916.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3657, pruned_loss=0.1157, over 5667640.66 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3413, pruned_loss=0.08874, over 5726466.70 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3691, pruned_loss=0.1193, over 5660372.48 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:02:12,671 INFO [optim.py:369] (1/2) 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 09:02:39,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5958, 1.8795, 1.4706, 1.9135], device='cuda:1'), covar=tensor([0.2699, 0.2899, 0.3228, 0.2730], device='cuda:1'), in_proj_covar=tensor([0.1570, 0.1133, 0.1387, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 09:02:49,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:02:50,265 INFO [train.py:968] (1/2) Epoch 26, batch 10300, giga_loss[loss=0.2914, simple_loss=0.3676, pruned_loss=0.1076, over 28870.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.362, pruned_loss=0.1125, over 5659572.04 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3411, pruned_loss=0.08866, over 5728880.28 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3653, pruned_loss=0.1158, over 5650873.30 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:02:54,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3199, 1.4980, 1.2447, 1.4891], device='cuda:1'), covar=tensor([0.0737, 0.0408, 0.0360, 0.0817], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0121, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 09:02:57,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6291, 1.7807, 1.4965, 1.6502], device='cuda:1'), covar=tensor([0.2581, 0.2803, 0.3102, 0.2528], device='cuda:1'), in_proj_covar=tensor([0.1570, 0.1132, 0.1387, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 09:03:20,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5358, 1.9704, 1.7412, 1.7031], device='cuda:1'), covar=tensor([0.0791, 0.0292, 0.0315, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0121, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 09:03:32,448 INFO [train.py:968] (1/2) Epoch 26, batch 10350, giga_loss[loss=0.3054, simple_loss=0.3691, pruned_loss=0.1209, over 28905.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3609, pruned_loss=0.1108, over 5674398.67 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3412, pruned_loss=0.08866, over 5738187.91 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3645, pruned_loss=0.1147, over 5655713.75 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:03:42,506 INFO [optim.py:369] (1/2) 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 09:03:48,033 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:03:52,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 09:03:54,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5460, 1.8455, 1.4575, 1.6138], device='cuda:1'), covar=tensor([0.2723, 0.2713, 0.3155, 0.2464], device='cuda:1'), in_proj_covar=tensor([0.1571, 0.1133, 0.1389, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 09:03:57,744 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:04:11,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5446, 1.6986, 1.6979, 1.4760], device='cuda:1'), covar=tensor([0.2905, 0.2567, 0.2207, 0.2674], device='cuda:1'), in_proj_covar=tensor([0.2046, 0.1993, 0.1924, 0.2058], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 09:04:19,990 INFO [train.py:968] (1/2) Epoch 26, batch 10400, giga_loss[loss=0.2549, simple_loss=0.3243, pruned_loss=0.09274, over 28452.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3586, pruned_loss=0.1099, over 5675376.71 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3409, pruned_loss=0.08862, over 5742347.29 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3626, pruned_loss=0.114, over 5654155.81 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:04:32,045 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:03,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:07,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:12,706 INFO [train.py:968] (1/2) Epoch 26, batch 10450, giga_loss[loss=0.3395, simple_loss=0.3969, pruned_loss=0.1411, over 28681.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1111, over 5671461.02 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3409, pruned_loss=0.08862, over 5742347.29 frames. ], giga_tot_loss[loss=0.2948, simple_loss=0.361, pruned_loss=0.1143, over 5654944.53 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:05:26,421 INFO [optim.py:369] (1/2) 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 09:05:33,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:05:55,907 INFO [train.py:968] (1/2) Epoch 26, batch 10500, giga_loss[loss=0.272, simple_loss=0.3547, pruned_loss=0.09467, over 28905.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3596, pruned_loss=0.1117, over 5670848.70 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3415, pruned_loss=0.08886, over 5737689.41 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3624, pruned_loss=0.115, over 5658393.40 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:06:41,676 INFO [train.py:968] (1/2) Epoch 26, batch 10550, libri_loss[loss=0.2556, simple_loss=0.3472, pruned_loss=0.08196, over 29746.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3613, pruned_loss=0.1118, over 5673509.88 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3419, pruned_loss=0.08897, over 5741017.67 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3637, pruned_loss=0.115, over 5658778.72 frames. ], batch size: 87, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:06:52,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4224, 1.6062, 1.4881, 1.5662], device='cuda:1'), covar=tensor([0.0654, 0.0310, 0.0292, 0.0705], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 09:06:53,128 INFO [optim.py:369] (1/2) 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 09:07:16,904 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:07:29,433 INFO [train.py:968] (1/2) Epoch 26, batch 10600, giga_loss[loss=0.2838, simple_loss=0.36, pruned_loss=0.1038, over 28678.00 frames. ], tot_loss[loss=0.2949, simple_loss=0.3631, pruned_loss=0.1134, over 5660065.35 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.08887, over 5741573.45 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3655, pruned_loss=0.1165, over 5646261.55 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:07:32,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:07:46,472 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1149674.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:08:17,255 INFO [train.py:968] (1/2) Epoch 26, batch 10650, giga_loss[loss=0.2865, simple_loss=0.3561, pruned_loss=0.1085, over 28887.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3628, pruned_loss=0.1134, over 5662300.19 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3416, pruned_loss=0.08867, over 5744493.00 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3655, pruned_loss=0.1167, over 5646553.82 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:08:29,177 INFO [optim.py:369] (1/2) 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 09:09:05,342 INFO [train.py:968] (1/2) Epoch 26, batch 10700, giga_loss[loss=0.3258, simple_loss=0.391, pruned_loss=0.1303, over 28685.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3642, pruned_loss=0.1148, over 5665903.93 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.0889, over 5746845.62 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3661, pruned_loss=0.1175, over 5650386.40 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:09:29,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1811, 1.3040, 1.0965, 0.9333], device='cuda:1'), covar=tensor([0.0848, 0.0404, 0.0898, 0.1110], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0450, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 09:09:34,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.09 vs. limit=5.0
+2023-03-13 09:09:39,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3716, 1.4988, 3.5011, 3.3024], device='cuda:1'), covar=tensor([0.1398, 0.2520, 0.0448, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0669, 0.0989, 0.0960], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 09:09:49,193 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 10750, giga_loss[loss=0.3062, simple_loss=0.3823, pruned_loss=0.115, over 28296.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3666, pruned_loss=0.1164, over 5653206.21 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08904, over 5738262.83 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3685, pruned_loss=0.1191, over 5646168.03 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:10:05,532 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149817.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:10:08,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149820.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:10:08,541 INFO [optim.py:369] (1/2) 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 09:10:36,527 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149849.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:10:40,167 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.33 vs. limit=2.0
+2023-03-13 09:10:42,950 INFO [train.py:968] (1/2) Epoch 26, batch 10800, giga_loss[loss=0.3772, simple_loss=0.42, pruned_loss=0.1672, over 28914.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3687, pruned_loss=0.1178, over 5663637.16 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08905, over 5739691.27 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3703, pruned_loss=0.1201, over 5656162.27 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:10:45,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6287, 1.8283, 1.8388, 1.3926], device='cuda:1'), covar=tensor([0.2021, 0.3040, 0.1704, 0.2094], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0716, 0.0974, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 09:10:54,102 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1149868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:11:32,054 INFO [train.py:968] (1/2) Epoch 26, batch 10850, giga_loss[loss=0.3017, simple_loss=0.3654, pruned_loss=0.119, over 28870.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3709, pruned_loss=0.1202, over 5659074.04 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08894, over 5732249.86 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1225, over 5658547.41 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:11:45,666 INFO [optim.py:369] (1/2) 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 09:12:08,184 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1149943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:12:10,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1149946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:12:20,450 INFO [train.py:968] (1/2) Epoch 26, batch 10900, giga_loss[loss=0.2902, simple_loss=0.3593, pruned_loss=0.1105, over 28700.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3714, pruned_loss=0.1201, over 5669799.59 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3419, pruned_loss=0.08871, over 5734358.67 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3736, pruned_loss=0.1226, over 5666475.87 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:12:26,681 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 09:12:37,096 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1149975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:13:08,237 INFO [train.py:968] (1/2) Epoch 26, batch 10950, giga_loss[loss=0.2738, simple_loss=0.3499, pruned_loss=0.09881, over 28810.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3704, pruned_loss=0.1185, over 5666005.77 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3416, pruned_loss=0.08876, over 5738469.25 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3731, pruned_loss=0.1212, over 5657927.11 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:13:21,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1150020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:13:22,329 INFO [optim.py:369] (1/2) 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 09:13:33,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1150035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:13:56,813 INFO [train.py:968] (1/2) Epoch 26, batch 11000, libri_loss[loss=0.2593, simple_loss=0.3442, pruned_loss=0.0872, over 29308.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.371, pruned_loss=0.1196, over 5661210.03 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3417, pruned_loss=0.08884, over 5742524.57 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3738, pruned_loss=0.1224, over 5649519.55 frames. ], batch size: 94, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:14:48,130 INFO [train.py:968] (1/2) Epoch 26, batch 11050, giga_loss[loss=0.276, simple_loss=0.3583, pruned_loss=0.09687, over 29045.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3696, pruned_loss=0.1194, over 5646540.35 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3415, pruned_loss=0.08882, over 5744273.76 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3725, pruned_loss=0.1222, over 5634362.09 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:15:06,808 INFO [optim.py:369] (1/2) 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 09:15:43,676 INFO [train.py:968] (1/2) Epoch 26, batch 11100, giga_loss[loss=0.2866, simple_loss=0.3562, pruned_loss=0.1085, over 28856.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3681, pruned_loss=0.1188, over 5643195.55 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3417, pruned_loss=0.08891, over 5737479.47 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3708, pruned_loss=0.1215, over 5638254.22 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:15:49,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1150163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:15:52,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1150166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:15:52,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6533, 1.7201, 1.8907, 1.4155], device='cuda:1'), covar=tensor([0.1808, 0.2556, 0.1418, 0.1687], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0718, 0.0975, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:1')
+2023-03-13 09:16:00,423 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1150178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:03,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1150181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:16,384 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1150195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:16:28,141 INFO [train.py:968] (1/2) Epoch 26, batch 11150, giga_loss[loss=0.3417, simple_loss=0.3942, pruned_loss=0.1446, over 27618.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3657, pruned_loss=0.1178, over 5619708.09 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3416, pruned_loss=0.089, over 5713446.50 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3686, pruned_loss=0.1208, over 5632984.98 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 2.0
+2023-03-13 09:16:31,174 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:16:48,091 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3980, 1.5362, 1.2278, 1.1002], device='cuda:1'), covar=tensor([0.0951, 0.0516, 0.0988, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0451, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 09:16:51,475 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7120, 4.5140, 1.7300, 1.9535], device='cuda:1'), covar=tensor([0.0961, 0.0331, 0.0896, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0568, 0.0402, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 09:17:01,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1150243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:17:12,540 INFO [train.py:968] (1/2) Epoch 26, batch 11200, giga_loss[loss=0.301, simple_loss=0.3687, pruned_loss=0.1166, over 29095.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3664, pruned_loss=0.1184, over 5643321.48 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.342, pruned_loss=0.0893, over 5718314.75 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3689, pruned_loss=0.1211, over 5647670.61 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:17:56,439 INFO [train.py:968] (1/2) Epoch 26, batch 11250, giga_loss[loss=0.3507, simple_loss=0.3926, pruned_loss=0.1545, over 26682.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.365, pruned_loss=0.1172, over 5645979.91 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3419, pruned_loss=0.08905, over 5720213.04 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1209, over 5644692.64 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:18:00,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3001, 2.0105, 1.5544, 0.5307], device='cuda:1'), covar=tensor([0.5546, 0.3192, 0.4332, 0.6792], device='cuda:1'), in_proj_covar=tensor([0.1817, 0.1704, 0.1636, 0.1474], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 09:18:08,495 INFO [optim.py:369] (1/2) 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 09:18:11,071 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9475, 2.0786, 1.6024, 1.6148], device='cuda:1'), covar=tensor([0.1116, 0.0750, 0.1026, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0452, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 09:18:13,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3172, 1.4576, 1.6120, 1.3598], device='cuda:1'), covar=tensor([0.1604, 0.1247, 0.1556, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0759, 0.0727, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 09:18:43,437 INFO [train.py:968] (1/2) Epoch 26, batch 11300, giga_loss[loss=0.2562, simple_loss=0.3345, pruned_loss=0.08895, over 28988.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3646, pruned_loss=0.117, over 5649979.58 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.08901, over 5723092.40 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3674, pruned_loss=0.1203, over 5645477.61 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:19:11,608 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1150386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:19:15,919 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1150389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:19:21,110 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-13 09:19:29,237 INFO [train.py:968] (1/2) Epoch 26, batch 11350, giga_loss[loss=0.3202, simple_loss=0.3912, pruned_loss=0.1246, over 29020.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.3673, pruned_loss=0.1191, over 5657280.69 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3419, pruned_loss=0.08905, over 5727702.37 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.37, pruned_loss=0.1224, over 5647926.98 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:19:41,104 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:20:00,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4628, 2.2117, 1.5838, 0.7410], device='cuda:1'), covar=tensor([0.4963, 0.2462, 0.3033, 0.5581], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1708, 0.1642, 0.1477], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 09:20:17,077 INFO [train.py:968] (1/2) Epoch 26, batch 11400, giga_loss[loss=0.3342, simple_loss=0.387, pruned_loss=0.1407, over 28603.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3699, pruned_loss=0.1216, over 5647752.34 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08909, over 5730075.49 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3724, pruned_loss=0.1247, over 5637017.14 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:20:30,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9118, 1.2473, 2.8863, 2.8465], device='cuda:1'), covar=tensor([0.1655, 0.2582, 0.0636, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0672, 0.0994, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 09:21:06,636 INFO [train.py:968] (1/2) Epoch 26, batch 11450, giga_loss[loss=0.2651, simple_loss=0.3412, pruned_loss=0.09447, over 28798.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3689, pruned_loss=0.1212, over 5654973.94 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3414, pruned_loss=0.08859, over 5733944.50 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3721, pruned_loss=0.1249, over 5641059.04 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:21:19,929 INFO [optim.py:369] (1/2) 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 09:21:29,214 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 09:21:36,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-13 09:21:53,310 INFO [train.py:968] (1/2) Epoch 26, batch 11500, giga_loss[loss=0.2895, simple_loss=0.3571, pruned_loss=0.1109, over 28936.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3682, pruned_loss=0.1207, over 5652227.49 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3415, pruned_loss=0.08859, over 5724804.41 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3712, pruned_loss=0.1241, over 5647444.48 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:22:42,164 INFO [train.py:968] (1/2) Epoch 26, batch 11550, giga_loss[loss=0.2636, simple_loss=0.3439, pruned_loss=0.0917, over 28983.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3701, pruned_loss=0.1219, over 5642778.22 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3415, pruned_loss=0.08867, over 5718752.28 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3731, pruned_loss=0.1253, over 5642832.83 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:22:48,964 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3685, 1.5756, 1.5840, 1.3444], device='cuda:1'), covar=tensor([0.2528, 0.2346, 0.1495, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.2044, 0.1998, 0.1926, 0.2058], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 09:22:55,419 INFO [optim.py:369] (1/2) 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 09:23:03,230 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1150629.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:23:17,241 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-13 09:23:26,516 INFO [train.py:968] (1/2) Epoch 26, batch 11600, giga_loss[loss=0.2954, simple_loss=0.3659, pruned_loss=0.1124, over 29057.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3703, pruned_loss=0.1211, over 5661894.15 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08888, over 5721906.73 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3733, pruned_loss=0.1247, over 5657308.98 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:23:46,480 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1150677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:23:59,806 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1150691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:24:17,721 INFO [train.py:968] (1/2) Epoch 26, batch 11650, giga_loss[loss=0.3395, simple_loss=0.3911, pruned_loss=0.144, over 27445.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3716, pruned_loss=0.1227, over 5652959.97 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3415, pruned_loss=0.08868, over 5726663.03 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.375, pruned_loss=0.1266, over 5643593.55 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:24:22,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0286, 5.1236, 2.1127, 2.3850], device='cuda:1'), covar=tensor([0.0894, 0.0345, 0.0800, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0566, 0.0401, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 09:24:24,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 09:24:30,889 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 11700, giga_loss[loss=0.3056, simple_loss=0.3698, pruned_loss=0.1207, over 28939.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3743, pruned_loss=0.1248, over 5656814.23 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.08888, over 5729043.91 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3771, pruned_loss=0.1281, over 5646230.85 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:25:50,625 INFO [train.py:968] (1/2) Epoch 26, batch 11750, giga_loss[loss=0.3, simple_loss=0.3688, pruned_loss=0.1156, over 28912.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3732, pruned_loss=0.1243, over 5658143.00 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3415, pruned_loss=0.0886, over 5733312.71 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3767, pruned_loss=0.1282, over 5643959.28 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:26:06,452 INFO [optim.py:369] (1/2) 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 09:26:06,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8574, 1.2045, 1.3252, 1.0295], device='cuda:1'), covar=tensor([0.2121, 0.1472, 0.2421, 0.1812], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0757, 0.0725, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 09:26:15,448 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.87 vs. limit=5.0
+2023-03-13 09:26:38,879 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6706, 1.8429, 1.4012, 1.3837], device='cuda:1'), covar=tensor([0.1037, 0.0632, 0.1022, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0451, 0.0522, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 09:26:39,236 INFO [train.py:968] (1/2) Epoch 26, batch 11800, giga_loss[loss=0.3277, simple_loss=0.3925, pruned_loss=0.1315, over 28639.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3734, pruned_loss=0.123, over 5660065.74 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3416, pruned_loss=0.08875, over 5736513.53 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3766, pruned_loss=0.1267, over 5644521.41 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:27:26,801 INFO [train.py:968] (1/2) Epoch 26, batch 11850, giga_loss[loss=0.3478, simple_loss=0.3923, pruned_loss=0.1517, over 26602.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.373, pruned_loss=0.1221, over 5657369.25 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3418, pruned_loss=0.08884, over 5739271.69 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3758, pruned_loss=0.1253, over 5641540.72 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:27:39,426 INFO [optim.py:369] (1/2) 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 09:28:12,771 INFO [train.py:968] (1/2) Epoch 26, batch 11900, libri_loss[loss=0.2765, simple_loss=0.3644, pruned_loss=0.09429, over 29411.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3717, pruned_loss=0.1212, over 5657229.36 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3416, pruned_loss=0.08875, over 5743497.55 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1247, over 5639004.85 frames. ], batch size: 92, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:28:48,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1743, 1.2998, 1.1854, 0.9283], device='cuda:1'), covar=tensor([0.1085, 0.0527, 0.1047, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0453, 0.0523, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 09:28:50,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151004.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:28:53,548 INFO [train.py:968] (1/2) Epoch 26, batch 11950, libri_loss[loss=0.2403, simple_loss=0.3291, pruned_loss=0.07574, over 29537.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3679, pruned_loss=0.1179, over 5664470.59 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08882, over 5741593.53 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3717, pruned_loss=0.1222, over 5646847.55 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:29:07,766 INFO [optim.py:369] (1/2) 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 09:29:33,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:29:39,504 INFO [train.py:968] (1/2) Epoch 26, batch 12000, giga_loss[loss=0.2832, simple_loss=0.358, pruned_loss=0.1042, over 28877.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3695, pruned_loss=0.1191, over 5671951.45 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3417, pruned_loss=0.08872, over 5747094.27 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3735, pruned_loss=0.1235, over 5650977.74 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:29:39,504 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 09:29:48,477 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 09:29:56,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:30:13,003 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1151083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:30:33,052 INFO [train.py:968] (1/2) Epoch 26, batch 12050, libri_loss[loss=0.2527, simple_loss=0.3419, pruned_loss=0.08174, over 29555.00 frames. ], tot_loss[loss=0.304, simple_loss=0.37, pruned_loss=0.119, over 5670107.05 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08847, over 5751435.28 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3743, pruned_loss=0.124, over 5646005.08 frames. ], batch size: 83, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:30:48,600 INFO [optim.py:369] (1/2) 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 09:31:09,284 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-13 09:31:11,026 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151147.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:31:14,334 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151150.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:31:19,967 INFO [train.py:968] (1/2) Epoch 26, batch 12100, giga_loss[loss=0.2728, simple_loss=0.3383, pruned_loss=0.1036, over 28579.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3689, pruned_loss=0.1189, over 5683587.31 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3415, pruned_loss=0.08833, over 5754586.13 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.373, pruned_loss=0.1235, over 5660265.73 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:31:39,220 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151179.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:31:55,600 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:31:59,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:06,838 INFO [train.py:968] (1/2) Epoch 26, batch 12150, giga_loss[loss=0.3446, simple_loss=0.3985, pruned_loss=0.1453, over 28307.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3708, pruned_loss=0.121, over 5681333.74 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3413, pruned_loss=0.0882, over 5757753.16 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3747, pruned_loss=0.1253, over 5658996.51 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:32:08,680 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:10,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:20,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4859, 2.0167, 1.6014, 1.6165], device='cuda:1'), covar=tensor([0.0764, 0.0287, 0.0314, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0112], device='cuda:1')
+2023-03-13 09:32:25,519 INFO [optim.py:369] (1/2) 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 09:32:30,234 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:43,135 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:32:57,777 INFO [train.py:968] (1/2) Epoch 26, batch 12200, giga_loss[loss=0.2783, simple_loss=0.3587, pruned_loss=0.09891, over 28919.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3731, pruned_loss=0.1228, over 5680019.23 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3413, pruned_loss=0.08817, over 5759255.69 frames. ], giga_tot_loss[loss=0.3148, simple_loss=0.3765, pruned_loss=0.1266, over 5660476.57 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:33:02,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3146, 1.5927, 1.5386, 1.3784], device='cuda:1'), covar=tensor([0.2023, 0.1925, 0.2421, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0759, 0.0727, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 09:33:15,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3093, 1.1935, 3.6985, 3.3143], device='cuda:1'), covar=tensor([0.1651, 0.3046, 0.0494, 0.2232], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0671, 0.0992, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 09:33:45,257 INFO [train.py:968] (1/2) Epoch 26, batch 12250, giga_loss[loss=0.2928, simple_loss=0.3661, pruned_loss=0.1097, over 28954.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3739, pruned_loss=0.1235, over 5670755.58 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3415, pruned_loss=0.08821, over 5760785.54 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.377, pruned_loss=0.1271, over 5652514.54 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:33:59,220 INFO [optim.py:369] (1/2) 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 09:34:32,457 INFO [train.py:968] (1/2) Epoch 26, batch 12300, giga_loss[loss=0.2947, simple_loss=0.3592, pruned_loss=0.115, over 28997.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3729, pruned_loss=0.1223, over 5677922.50 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08834, over 5752796.70 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3757, pruned_loss=0.1256, over 5668312.88 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:35:18,263 INFO [train.py:968] (1/2) Epoch 26, batch 12350, giga_loss[loss=0.2927, simple_loss=0.3676, pruned_loss=0.1089, over 29074.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3723, pruned_loss=0.1215, over 5660511.59 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08858, over 5746241.14 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3753, pruned_loss=0.125, over 5656037.22 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:35:33,659 INFO [optim.py:369] (1/2) 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 09:36:01,486 INFO [train.py:968] (1/2) Epoch 26, batch 12400, libri_loss[loss=0.2557, simple_loss=0.3442, pruned_loss=0.08357, over 29480.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3714, pruned_loss=0.12, over 5674435.31 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08864, over 5749983.13 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3744, pruned_loss=0.1235, over 5665701.37 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:36:02,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:36:38,870 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1151497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:36:49,679 INFO [train.py:968] (1/2) Epoch 26, batch 12450, giga_loss[loss=0.3972, simple_loss=0.4353, pruned_loss=0.1796, over 27572.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3701, pruned_loss=0.1192, over 5673421.22 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08866, over 5748671.65 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3727, pruned_loss=0.1223, over 5666957.33 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:37:03,450 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 12500, giga_loss[loss=0.2656, simple_loss=0.3403, pruned_loss=0.09541, over 29085.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3684, pruned_loss=0.1185, over 5681419.24 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3421, pruned_loss=0.0888, over 5753157.85 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3714, pruned_loss=0.1218, over 5669860.26 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:37:37,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1151562.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:38:13,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1151601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:38:16,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1151604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:38:19,011 INFO [train.py:968] (1/2) Epoch 26, batch 12550, giga_loss[loss=0.2563, simple_loss=0.3277, pruned_loss=0.09247, over 28850.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3663, pruned_loss=0.118, over 5677427.49 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3417, pruned_loss=0.0886, over 5755849.32 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3698, pruned_loss=0.1217, over 5663400.05 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:38:36,159 INFO [optim.py:369] (1/2) 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 09:38:45,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1151633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:39:06,346 INFO [train.py:968] (1/2) Epoch 26, batch 12600, giga_loss[loss=0.2766, simple_loss=0.3462, pruned_loss=0.1035, over 28905.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3624, pruned_loss=0.1159, over 5686854.08 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08881, over 5758133.80 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3653, pruned_loss=0.1191, over 5672675.59 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:39:54,582 INFO [train.py:968] (1/2) Epoch 26, batch 12650, giga_loss[loss=0.2846, simple_loss=0.343, pruned_loss=0.1131, over 28403.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3606, pruned_loss=0.115, over 5692944.93 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.342, pruned_loss=0.08869, over 5760514.75 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3632, pruned_loss=0.1181, over 5678349.39 frames. ], batch size: 78, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:40:09,463 INFO [optim.py:369] (1/2) 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 09:40:40,911 INFO [train.py:968] (1/2) Epoch 26, batch 12700, libri_loss[loss=0.2363, simple_loss=0.319, pruned_loss=0.07679, over 29572.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3601, pruned_loss=0.1149, over 5694751.40 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.08843, over 5764984.46 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3631, pruned_loss=0.1184, over 5677291.57 frames. ], batch size: 76, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:41:24,893 INFO [train.py:968] (1/2) Epoch 26, batch 12750, giga_loss[loss=0.2721, simple_loss=0.3506, pruned_loss=0.09678, over 28581.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5695911.58 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.08855, over 5767918.97 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3631, pruned_loss=0.1168, over 5676909.86 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:41:44,593 INFO [optim.py:369] (1/2) 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 09:41:52,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1555, 1.6849, 1.7365, 1.3487], device='cuda:1'), covar=tensor([0.2070, 0.1332, 0.2083, 0.1767], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0755, 0.0722, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 09:42:06,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8176, 2.1233, 1.4702, 1.6369], device='cuda:1'), covar=tensor([0.1022, 0.0551, 0.0982, 0.1089], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0451, 0.0522, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 09:42:07,156 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-13 09:42:14,233 INFO [train.py:968] (1/2) Epoch 26, batch 12800, giga_loss[loss=0.2725, simple_loss=0.3401, pruned_loss=0.1024, over 28593.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3589, pruned_loss=0.1105, over 5689747.06 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.08848, over 5766773.57 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3617, pruned_loss=0.1138, over 5674049.74 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:42:32,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:43:04,786 INFO [train.py:968] (1/2) Epoch 26, batch 12850, giga_loss[loss=0.2856, simple_loss=0.3597, pruned_loss=0.1057, over 28589.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3563, pruned_loss=0.1076, over 5672987.22 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3411, pruned_loss=0.08845, over 5760830.57 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3595, pruned_loss=0.1109, over 5663254.40 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:43:25,034 INFO [optim.py:369] (1/2) 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 09:43:37,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1151937.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:43:57,099 INFO [train.py:968] (1/2) Epoch 26, batch 12900, giga_loss[loss=0.2551, simple_loss=0.3283, pruned_loss=0.0909, over 28551.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3532, pruned_loss=0.1045, over 5667198.61 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3409, pruned_loss=0.08834, over 5760806.68 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.356, pruned_loss=0.1074, over 5658638.78 frames. ], batch size: 85, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:44:32,457 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.46 vs. limit=5.0
+2023-03-13 09:44:42,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5998, 1.9152, 1.4836, 1.8229], device='cuda:1'), covar=tensor([0.2880, 0.2703, 0.3095, 0.2601], device='cuda:1'), in_proj_covar=tensor([0.1570, 0.1130, 0.1389, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 09:44:44,195 INFO [train.py:968] (1/2) Epoch 26, batch 12950, giga_loss[loss=0.2358, simple_loss=0.3243, pruned_loss=0.07371, over 28988.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3489, pruned_loss=0.1005, over 5676438.86 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3399, pruned_loss=0.08802, over 5764763.57 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3525, pruned_loss=0.1036, over 5662924.93 frames. ], batch size: 128, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:44:47,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4511, 1.6184, 1.6802, 1.2886], device='cuda:1'), covar=tensor([0.1885, 0.2947, 0.1626, 0.1948], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0710, 0.0968, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 09:44:51,620 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:44:53,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:44:57,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 09:45:02,174 INFO [optim.py:369] (1/2) 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 09:45:21,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.55 vs. limit=5.0
+2023-03-13 09:45:22,953 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:45:34,620 INFO [train.py:968] (1/2) Epoch 26, batch 13000, giga_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1117, over 27669.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.348, pruned_loss=0.09808, over 5673074.46 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3396, pruned_loss=0.08803, over 5766560.30 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3511, pruned_loss=0.1007, over 5659670.07 frames. ], batch size: 472, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:45:43,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8022, 1.0160, 2.8366, 2.7978], device='cuda:1'), covar=tensor([0.2074, 0.3043, 0.1112, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0671, 0.0990, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 09:45:58,595 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152080.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 09:46:00,847 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152083.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:46:22,657 INFO [train.py:968] (1/2) Epoch 26, batch 13050, giga_loss[loss=0.258, simple_loss=0.3434, pruned_loss=0.08624, over 28988.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3492, pruned_loss=0.09917, over 5654388.65 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3392, pruned_loss=0.08802, over 5755549.33 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3524, pruned_loss=0.1016, over 5650664.17 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:46:26,975 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152112.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 09:46:38,743 INFO [optim.py:369] (1/2) 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 09:46:39,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:46:51,964 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:47:09,012 INFO [train.py:968] (1/2) Epoch 26, batch 13100, giga_loss[loss=0.2466, simple_loss=0.3285, pruned_loss=0.08238, over 28796.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3482, pruned_loss=0.0985, over 5662444.34 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3389, pruned_loss=0.08812, over 5758109.22 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3513, pruned_loss=0.1007, over 5654595.43 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:47:53,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:47:57,607 INFO [train.py:968] (1/2) Epoch 26, batch 13150, giga_loss[loss=0.2445, simple_loss=0.3259, pruned_loss=0.0815, over 28569.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3445, pruned_loss=0.09583, over 5664851.93 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3384, pruned_loss=0.08789, over 5758178.65 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3476, pruned_loss=0.09794, over 5656791.02 frames. ], batch size: 336, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:48:16,036 INFO [optim.py:369] (1/2) 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 09:48:23,487 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:48:42,149 INFO [train.py:968] (1/2) Epoch 26, batch 13200, giga_loss[loss=0.242, simple_loss=0.3298, pruned_loss=0.07707, over 28593.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3437, pruned_loss=0.09525, over 5666909.04 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3386, pruned_loss=0.08815, over 5759497.66 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.346, pruned_loss=0.0969, over 5656576.52 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:48:53,058 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 09:49:27,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 09:49:28,100 INFO [train.py:968] (1/2) Epoch 26, batch 13250, giga_loss[loss=0.2452, simple_loss=0.3281, pruned_loss=0.08115, over 28822.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3428, pruned_loss=0.09433, over 5670664.74 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3383, pruned_loss=0.08813, over 5758583.90 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3452, pruned_loss=0.09584, over 5660561.40 frames. ], batch size: 145, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:49:39,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2856, 1.4988, 1.3004, 1.5262], device='cuda:1'), covar=tensor([0.0810, 0.0338, 0.0361, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 09:49:48,038 INFO [optim.py:369] (1/2) 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 09:50:16,900 INFO [train.py:968] (1/2) Epoch 26, batch 13300, giga_loss[loss=0.2469, simple_loss=0.3308, pruned_loss=0.08151, over 28314.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3419, pruned_loss=0.09336, over 5657983.01 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3381, pruned_loss=0.08809, over 5750177.40 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3439, pruned_loss=0.09464, over 5657617.38 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:51:06,280 INFO [train.py:968] (1/2) Epoch 26, batch 13350, giga_loss[loss=0.2339, simple_loss=0.3164, pruned_loss=0.07574, over 29020.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3399, pruned_loss=0.09188, over 5656710.25 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3385, pruned_loss=0.08854, over 5743976.29 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3412, pruned_loss=0.09256, over 5659907.26 frames. ], batch size: 106, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:51:27,739 INFO [optim.py:369] (1/2) 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 09:51:42,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1929, 0.8511, 0.9705, 1.4172], device='cuda:1'), covar=tensor([0.0794, 0.0409, 0.0383, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 09:51:54,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.27 vs. limit=5.0
+2023-03-13 09:51:59,252 INFO [train.py:968] (1/2) Epoch 26, batch 13400, giga_loss[loss=0.2031, simple_loss=0.2985, pruned_loss=0.05388, over 28940.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3357, pruned_loss=0.08942, over 5648862.84 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3388, pruned_loss=0.08885, over 5740501.88 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3365, pruned_loss=0.08972, over 5653125.75 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:52:00,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3800, 1.9935, 1.3388, 0.6157], device='cuda:1'), covar=tensor([0.6059, 0.3089, 0.4715, 0.6895], device='cuda:1'), in_proj_covar=tensor([0.1806, 0.1694, 0.1636, 0.1469], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 09:52:07,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:52:48,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:52:53,519 INFO [train.py:968] (1/2) Epoch 26, batch 13450, giga_loss[loss=0.2826, simple_loss=0.3565, pruned_loss=0.1043, over 28896.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3347, pruned_loss=0.08956, over 5638577.80 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3387, pruned_loss=0.08875, over 5742510.03 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3354, pruned_loss=0.0899, over 5638006.59 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:53:01,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:53:12,465 INFO [optim.py:369] (1/2) 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 09:53:41,748 INFO [train.py:968] (1/2) Epoch 26, batch 13500, giga_loss[loss=0.274, simple_loss=0.3387, pruned_loss=0.1047, over 26676.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3333, pruned_loss=0.08944, over 5630615.90 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3378, pruned_loss=0.08843, over 5727496.65 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3344, pruned_loss=0.09003, over 5640572.72 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:53:47,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4163, 1.6243, 1.2970, 1.1920], device='cuda:1'), covar=tensor([0.1003, 0.0478, 0.0931, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0446, 0.0517, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 09:54:02,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:54:35,951 INFO [train.py:968] (1/2) Epoch 26, batch 13550, giga_loss[loss=0.3159, simple_loss=0.3983, pruned_loss=0.1168, over 28747.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3343, pruned_loss=0.08977, over 5616377.24 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3379, pruned_loss=0.08856, over 5719807.39 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3351, pruned_loss=0.09014, over 5628534.95 frames. ], batch size: 284, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:54:39,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:54:48,355 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6780, 2.2911, 1.4634, 0.8765], device='cuda:1'), covar=tensor([0.7872, 0.4193, 0.4453, 0.7209], device='cuda:1'), in_proj_covar=tensor([0.1809, 0.1697, 0.1639, 0.1472], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 09:54:55,596 INFO [optim.py:369] (1/2) 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 09:55:14,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:14,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-13 09:55:20,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:31,059 INFO [train.py:968] (1/2) Epoch 26, batch 13600, giga_loss[loss=0.3027, simple_loss=0.3832, pruned_loss=0.1111, over 27994.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.337, pruned_loss=0.0901, over 5630941.68 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.337, pruned_loss=0.0883, over 5722796.24 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3383, pruned_loss=0.09066, over 5635912.80 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 09:55:32,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:35,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:55:56,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:10,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:31,077 INFO [train.py:968] (1/2) Epoch 26, batch 13650, giga_loss[loss=0.2973, simple_loss=0.3671, pruned_loss=0.1137, over 28593.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3386, pruned_loss=0.09128, over 5633806.36 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3365, pruned_loss=0.08803, over 5725358.37 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3401, pruned_loss=0.09198, over 5634323.22 frames. ], batch size: 262, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:56:47,422 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:54,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:56:58,508 INFO [optim.py:369] (1/2) 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 09:57:25,355 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:27,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:27,783 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:29,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:57:31,684 INFO [train.py:968] (1/2) Epoch 26, batch 13700, giga_loss[loss=0.2739, simple_loss=0.339, pruned_loss=0.1044, over 26703.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3371, pruned_loss=0.09057, over 5643514.41 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3363, pruned_loss=0.08801, over 5731480.78 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3386, pruned_loss=0.09127, over 5635457.88 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:57:53,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4020, 1.5039, 1.2376, 1.5348], device='cuda:1'), covar=tensor([0.0755, 0.0345, 0.0355, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0118, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 09:57:59,925 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1152783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:58:32,072 INFO [train.py:968] (1/2) Epoch 26, batch 13750, giga_loss[loss=0.2311, simple_loss=0.3187, pruned_loss=0.07171, over 29176.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3355, pruned_loss=0.08902, over 5641417.72 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.336, pruned_loss=0.08794, over 5725494.65 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3369, pruned_loss=0.08967, over 5638722.35 frames. ], batch size: 113, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:58:57,148 INFO [optim.py:369] (1/2) 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 09:59:12,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1152839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 09:59:21,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3429, 4.1921, 3.9647, 1.9331], device='cuda:1'), covar=tensor([0.0582, 0.0726, 0.0771, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.1264, 0.1166, 0.0982, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 09:59:29,225 INFO [train.py:968] (1/2) Epoch 26, batch 13800, libri_loss[loss=0.2651, simple_loss=0.3442, pruned_loss=0.09295, over 29529.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3352, pruned_loss=0.08795, over 5645646.55 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3361, pruned_loss=0.08806, over 5727401.70 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3362, pruned_loss=0.08836, over 5639178.92 frames. ], batch size: 81, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 09:59:52,874 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:00:09,127 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:00:31,672 INFO [train.py:968] (1/2) Epoch 26, batch 13850, giga_loss[loss=0.2224, simple_loss=0.299, pruned_loss=0.07293, over 28804.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3321, pruned_loss=0.08691, over 5650237.17 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3356, pruned_loss=0.08785, over 5730101.15 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3333, pruned_loss=0.0874, over 5641825.50 frames. ], batch size: 243, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:00:57,091 INFO [optim.py:369] (1/2) 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 10:01:00,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4345, 2.0060, 1.4067, 0.7495], device='cuda:1'), covar=tensor([0.6501, 0.3258, 0.4371, 0.6835], device='cuda:1'), in_proj_covar=tensor([0.1798, 0.1686, 0.1631, 0.1467], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 10:01:31,087 INFO [train.py:968] (1/2) Epoch 26, batch 13900, libri_loss[loss=0.2322, simple_loss=0.3082, pruned_loss=0.07807, over 29357.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.0876, over 5656415.92 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3352, pruned_loss=0.08779, over 5732443.75 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3332, pruned_loss=0.08804, over 5646216.30 frames. ], batch size: 71, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:01:37,398 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1152961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:03,172 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1152982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:06,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1152985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:10,898 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8758, 2.2268, 1.8075, 2.2867], device='cuda:1'), covar=tensor([0.2796, 0.2792, 0.3158, 0.2407], device='cuda:1'), in_proj_covar=tensor([0.1571, 0.1129, 0.1389, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 10:02:29,401 INFO [train.py:968] (1/2) Epoch 26, batch 13950, giga_loss[loss=0.3261, simple_loss=0.3911, pruned_loss=0.1306, over 28886.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.08758, over 5667986.35 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3349, pruned_loss=0.08763, over 5736205.98 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3332, pruned_loss=0.08808, over 5654722.32 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:02:32,030 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:36,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:02:52,127 INFO [optim.py:369] (1/2) 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 10:03:24,339 INFO [train.py:968] (1/2) Epoch 26, batch 14000, giga_loss[loss=0.3197, simple_loss=0.3749, pruned_loss=0.1322, over 26863.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3322, pruned_loss=0.08693, over 5670693.67 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3341, pruned_loss=0.0873, over 5735744.64 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3338, pruned_loss=0.08761, over 5658834.84 frames. ], batch size: 555, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:03:47,130 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 10:04:26,626 INFO [train.py:968] (1/2) Epoch 26, batch 14050, giga_loss[loss=0.2343, simple_loss=0.3037, pruned_loss=0.08247, over 24764.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3343, pruned_loss=0.0873, over 5677623.00 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3342, pruned_loss=0.08738, over 5737838.66 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3355, pruned_loss=0.08776, over 5665414.58 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:04:51,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:04:53,569 INFO [optim.py:369] (1/2) 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 10:04:54,552 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3710, 1.5541, 1.1528, 1.1347], device='cuda:1'), covar=tensor([0.0992, 0.0459, 0.1001, 0.1080], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0446, 0.0519, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 10:05:34,667 INFO [train.py:968] (1/2) Epoch 26, batch 14100, giga_loss[loss=0.3335, simple_loss=0.3905, pruned_loss=0.1382, over 28484.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3314, pruned_loss=0.08587, over 5681496.41 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.334, pruned_loss=0.08733, over 5737896.82 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3325, pruned_loss=0.08627, over 5670519.17 frames. ], batch size: 370, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:06:11,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3995, 1.7678, 1.7414, 1.5181], device='cuda:1'), covar=tensor([0.2156, 0.2162, 0.2176, 0.2075], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0746, 0.0713, 0.0685], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 10:06:39,111 INFO [train.py:968] (1/2) Epoch 26, batch 14150, giga_loss[loss=0.2787, simple_loss=0.3391, pruned_loss=0.1091, over 24444.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3331, pruned_loss=0.08736, over 5670886.74 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3337, pruned_loss=0.08722, over 5739683.75 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3342, pruned_loss=0.08776, over 5659489.81 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:07:05,358 INFO [optim.py:369] (1/2) 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 10:07:25,129 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0001, 3.8245, 3.6632, 2.0442], device='cuda:1'), covar=tensor([0.0589, 0.0783, 0.0727, 0.2183], device='cuda:1'), in_proj_covar=tensor([0.1263, 0.1168, 0.0983, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 10:07:33,783 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:07:37,332 INFO [train.py:968] (1/2) Epoch 26, batch 14200, giga_loss[loss=0.2657, simple_loss=0.3586, pruned_loss=0.08639, over 28179.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3367, pruned_loss=0.08745, over 5671659.34 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3335, pruned_loss=0.08719, over 5746348.29 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3378, pruned_loss=0.08782, over 5653373.48 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:07:44,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:07:51,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:07:57,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:08:31,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:08:36,937 INFO [train.py:968] (1/2) Epoch 26, batch 14250, giga_loss[loss=0.2594, simple_loss=0.3518, pruned_loss=0.08357, over 28726.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3398, pruned_loss=0.08727, over 5668406.48 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.3337, pruned_loss=0.08733, over 5748061.14 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3407, pruned_loss=0.08745, over 5650918.42 frames. ], batch size: 99, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:08:52,202 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:09:00,767 INFO [optim.py:369] (1/2) 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 10:09:12,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:09:34,803 INFO [train.py:968] (1/2) Epoch 26, batch 14300, giga_loss[loss=0.2571, simple_loss=0.3516, pruned_loss=0.08128, over 28931.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3389, pruned_loss=0.08571, over 5659335.34 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3335, pruned_loss=0.08717, over 5751036.68 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3398, pruned_loss=0.08598, over 5641208.51 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:10:09,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:20,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:26,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:36,798 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:10:37,108 INFO [train.py:968] (1/2) Epoch 26, batch 14350, giga_loss[loss=0.2716, simple_loss=0.353, pruned_loss=0.09514, over 28921.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3401, pruned_loss=0.08604, over 5663637.01 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3335, pruned_loss=0.08718, over 5745930.34 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3409, pruned_loss=0.08622, over 5652604.83 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:10:40,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:11:05,943 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:11:07,206 INFO [optim.py:369] (1/2) 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 10:11:19,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:11:42,236 INFO [train.py:968] (1/2) Epoch 26, batch 14400, giga_loss[loss=0.2447, simple_loss=0.3247, pruned_loss=0.08237, over 28919.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08759, over 5666487.58 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3335, pruned_loss=0.08718, over 5745930.34 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3411, pruned_loss=0.08773, over 5657901.10 frames. ], batch size: 186, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:11:54,791 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2342, 2.5327, 1.3057, 1.4088], device='cuda:1'), covar=tensor([0.1001, 0.0373, 0.0928, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0565, 0.0403, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 10:12:05,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:12:13,663 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:12:17,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:12:52,999 INFO [train.py:968] (1/2) Epoch 26, batch 14450, giga_loss[loss=0.262, simple_loss=0.3448, pruned_loss=0.08954, over 28906.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3403, pruned_loss=0.08844, over 5665287.42 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3334, pruned_loss=0.08713, over 5747182.91 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.341, pruned_loss=0.08861, over 5656628.28 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:12:58,469 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:13:23,391 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:13:25,775 INFO [optim.py:369] (1/2) 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 10:13:27,914 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:14:09,032 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-13 10:14:09,218 INFO [train.py:968] (1/2) Epoch 26, batch 14500, libri_loss[loss=0.2592, simple_loss=0.3421, pruned_loss=0.08821, over 29655.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3376, pruned_loss=0.08687, over 5678405.07 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3335, pruned_loss=0.08709, over 5749610.00 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3382, pruned_loss=0.08705, over 5667448.76 frames. ], batch size: 88, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:14:15,714 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:15:19,555 INFO [train.py:968] (1/2) Epoch 26, batch 14550, giga_loss[loss=0.2308, simple_loss=0.3169, pruned_loss=0.07238, over 28959.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3325, pruned_loss=0.08408, over 5672675.94 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3331, pruned_loss=0.087, over 5754016.55 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3334, pruned_loss=0.08425, over 5657596.88 frames. ], batch size: 285, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:15:49,846 INFO [optim.py:369] (1/2) 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 10:16:23,545 INFO [train.py:968] (1/2) Epoch 26, batch 14600, giga_loss[loss=0.1983, simple_loss=0.2858, pruned_loss=0.05539, over 28752.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3302, pruned_loss=0.08315, over 5677314.74 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3327, pruned_loss=0.0869, over 5756815.47 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3311, pruned_loss=0.0833, over 5661245.04 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:17:11,214 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 14650, giga_loss[loss=0.2635, simple_loss=0.3513, pruned_loss=0.08789, over 28852.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.332, pruned_loss=0.08483, over 5678072.96 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3323, pruned_loss=0.0868, over 5750223.54 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3331, pruned_loss=0.08498, over 5669556.39 frames. ], batch size: 174, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:17:47,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4625, 4.2944, 4.1042, 1.9644], device='cuda:1'), covar=tensor([0.0538, 0.0711, 0.0780, 0.1901], device='cuda:1'), in_proj_covar=tensor([0.1265, 0.1165, 0.0984, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 10:17:50,393 INFO [optim.py:369] (1/2) 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 10:18:25,693 INFO [train.py:968] (1/2) Epoch 26, batch 14700, giga_loss[loss=0.2344, simple_loss=0.2979, pruned_loss=0.0854, over 24381.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3351, pruned_loss=0.08624, over 5677932.21 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3319, pruned_loss=0.08657, over 5754224.58 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3364, pruned_loss=0.08653, over 5666120.86 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:19:22,845 INFO [train.py:968] (1/2) Epoch 26, batch 14750, giga_loss[loss=0.2451, simple_loss=0.3212, pruned_loss=0.08446, over 28787.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3338, pruned_loss=0.0869, over 5688457.43 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3319, pruned_loss=0.0867, over 5757064.39 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3349, pruned_loss=0.08703, over 5675091.35 frames. ], batch size: 119, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:19:46,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1153825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:19:52,183 INFO [optim.py:369] (1/2) 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 10:20:02,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 10:20:06,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:20:08,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:20:19,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1153850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:20:26,563 INFO [train.py:968] (1/2) Epoch 26, batch 14800, giga_loss[loss=0.2613, simple_loss=0.3436, pruned_loss=0.08947, over 28966.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.335, pruned_loss=0.08867, over 5679560.55 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3312, pruned_loss=0.0864, over 5760521.17 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3365, pruned_loss=0.0891, over 5664044.29 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:20:42,401 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1153871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:21:25,720 INFO [train.py:968] (1/2) Epoch 26, batch 14850, giga_loss[loss=0.2706, simple_loss=0.3624, pruned_loss=0.08939, over 28360.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3343, pruned_loss=0.08793, over 5676997.42 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3307, pruned_loss=0.08608, over 5762716.15 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3361, pruned_loss=0.08861, over 5661013.91 frames. ], batch size: 368, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:21:55,100 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 14900, giga_loss[loss=0.2299, simple_loss=0.3215, pruned_loss=0.06914, over 29033.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3362, pruned_loss=0.08786, over 5678803.23 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3303, pruned_loss=0.08597, over 5764505.48 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.338, pruned_loss=0.08852, over 5663406.23 frames. ], batch size: 285, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:23:29,080 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1153993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:23:33,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:23:46,794 INFO [train.py:968] (1/2) Epoch 26, batch 14950, giga_loss[loss=0.2359, simple_loss=0.3271, pruned_loss=0.07235, over 28934.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3369, pruned_loss=0.08808, over 5675544.41 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3305, pruned_loss=0.08611, over 5765693.20 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3382, pruned_loss=0.08848, over 5661420.25 frames. ], batch size: 164, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:24:15,983 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:24:20,463 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 15000, giga_loss[loss=0.2527, simple_loss=0.331, pruned_loss=0.08719, over 28737.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3329, pruned_loss=0.08627, over 5685122.40 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3307, pruned_loss=0.08628, over 5759118.33 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3339, pruned_loss=0.08647, over 5677654.37 frames. ], batch size: 243, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:24:55,775 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 10:25:05,779 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 10:25:42,888 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154088.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:25:53,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3291, 1.7343, 1.6307, 1.5459], device='cuda:1'), covar=tensor([0.2092, 0.2090, 0.2117, 0.1939], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0740, 0.0711, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 10:26:11,075 INFO [train.py:968] (1/2) Epoch 26, batch 15050, giga_loss[loss=0.2113, simple_loss=0.2892, pruned_loss=0.06672, over 28665.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3269, pruned_loss=0.08397, over 5687213.19 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3303, pruned_loss=0.086, over 5761053.79 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3281, pruned_loss=0.08435, over 5677581.83 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:26:29,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3930, 1.5987, 1.5747, 1.5742], device='cuda:1'), covar=tensor([0.0776, 0.0325, 0.0316, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 10:26:39,050 INFO [optim.py:369] (1/2) 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 10:27:11,300 INFO [train.py:968] (1/2) Epoch 26, batch 15100, giga_loss[loss=0.2092, simple_loss=0.2814, pruned_loss=0.06847, over 24080.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3266, pruned_loss=0.08411, over 5680891.06 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3307, pruned_loss=0.08627, over 5763025.94 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3271, pruned_loss=0.08413, over 5670500.40 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:28:03,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:28:08,941 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-13 10:28:09,061 INFO [train.py:968] (1/2) Epoch 26, batch 15150, giga_loss[loss=0.2679, simple_loss=0.3427, pruned_loss=0.09657, over 29002.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.329, pruned_loss=0.08604, over 5681438.37 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3305, pruned_loss=0.08626, over 5764333.87 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3295, pruned_loss=0.08605, over 5671378.77 frames. ], batch size: 213, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:28:33,532 INFO [optim.py:369] (1/2) 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 10:29:05,249 INFO [train.py:968] (1/2) Epoch 26, batch 15200, giga_loss[loss=0.2424, simple_loss=0.3254, pruned_loss=0.07965, over 28925.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3267, pruned_loss=0.08478, over 5666582.09 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3306, pruned_loss=0.08637, over 5766531.61 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.327, pruned_loss=0.08467, over 5654426.71 frames. ], batch size: 136, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:29:47,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4772, 1.6433, 1.6522, 1.4067], device='cuda:1'), covar=tensor([0.2656, 0.2315, 0.1945, 0.2521], device='cuda:1'), in_proj_covar=tensor([0.1994, 0.1933, 0.1849, 0.1995], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 10:30:04,167 INFO [train.py:968] (1/2) Epoch 26, batch 15250, giga_loss[loss=0.1999, simple_loss=0.2773, pruned_loss=0.06125, over 24354.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3258, pruned_loss=0.0834, over 5674108.85 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3302, pruned_loss=0.08612, over 5770213.65 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3263, pruned_loss=0.08349, over 5658478.84 frames. ], batch size: 705, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:30:30,589 INFO [optim.py:369] (1/2) 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 10:30:49,042 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1154343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:30:52,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1154346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:31:03,049 INFO [train.py:968] (1/2) Epoch 26, batch 15300, giga_loss[loss=0.2285, simple_loss=0.3133, pruned_loss=0.07184, over 29036.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.325, pruned_loss=0.08335, over 5674709.64 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3301, pruned_loss=0.08607, over 5774294.22 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3253, pruned_loss=0.08338, over 5654757.42 frames. ], batch size: 100, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:31:27,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:31:33,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 10:31:43,838 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 15350, giga_loss[loss=0.2934, simple_loss=0.3661, pruned_loss=0.1103, over 28097.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.08245, over 5685235.49 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.33, pruned_loss=0.08595, over 5775686.36 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3245, pruned_loss=0.0825, over 5666036.64 frames. ], batch size: 412, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:32:11,337 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 26, batch 15400, giga_loss[loss=0.2281, simple_loss=0.3132, pruned_loss=0.0715, over 28868.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3248, pruned_loss=0.08216, over 5691715.12 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3304, pruned_loss=0.08607, over 5774077.65 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3246, pruned_loss=0.08202, over 5676390.11 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:33:16,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154463.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:34:10,931 INFO [train.py:968] (1/2) Epoch 26, batch 15450, giga_loss[loss=0.2318, simple_loss=0.3174, pruned_loss=0.07312, over 28967.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3259, pruned_loss=0.08298, over 5692513.52 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3305, pruned_loss=0.08606, over 5774697.62 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3256, pruned_loss=0.08283, over 5678297.01 frames. ], batch size: 155, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:34:41,179 INFO [optim.py:369] (1/2) 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 10:35:01,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2722, 1.5168, 1.4237, 1.1850], device='cuda:1'), covar=tensor([0.2580, 0.2400, 0.1722, 0.2448], device='cuda:1'), in_proj_covar=tensor([0.1997, 0.1937, 0.1855, 0.1998], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 10:35:11,566 INFO [train.py:968] (1/2) Epoch 26, batch 15500, giga_loss[loss=0.2178, simple_loss=0.3073, pruned_loss=0.06411, over 28481.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.08251, over 5694805.47 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3296, pruned_loss=0.08583, over 5778894.45 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3245, pruned_loss=0.0825, over 5676968.72 frames. ], batch size: 370, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:36:09,897 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1154606.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:36:10,159 INFO [train.py:968] (1/2) Epoch 26, batch 15550, giga_loss[loss=0.2701, simple_loss=0.3622, pruned_loss=0.08904, over 28918.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3244, pruned_loss=0.08142, over 5670161.44 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3297, pruned_loss=0.08588, over 5771393.16 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3246, pruned_loss=0.08131, over 5661390.15 frames. ], batch size: 227, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:36:12,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1154609.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:36:36,934 INFO [optim.py:369] (1/2) 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 10:36:46,328 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154638.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:37:04,790 INFO [train.py:968] (1/2) Epoch 26, batch 15600, giga_loss[loss=0.2582, simple_loss=0.345, pruned_loss=0.08565, over 28769.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3272, pruned_loss=0.08221, over 5675574.45 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.329, pruned_loss=0.08558, over 5775826.89 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3278, pruned_loss=0.08227, over 5661251.48 frames. ], batch size: 243, lr: 1.23e-03, grad_scale: 8.0
+2023-03-13 10:37:19,667 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2170, 1.2005, 3.5996, 3.1783], device='cuda:1'), covar=tensor([0.1702, 0.2894, 0.0472, 0.1045], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0663, 0.0973, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 10:38:00,320 INFO [train.py:968] (1/2) Epoch 26, batch 15650, giga_loss[loss=0.2515, simple_loss=0.3367, pruned_loss=0.08314, over 28541.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3302, pruned_loss=0.08407, over 5659209.63 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3289, pruned_loss=0.08555, over 5761325.77 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3308, pruned_loss=0.08404, over 5656247.98 frames. ], batch size: 307, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:38:09,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2673, 1.5702, 1.5388, 1.2059], device='cuda:1'), covar=tensor([0.2943, 0.2340, 0.1460, 0.2364], device='cuda:1'), in_proj_covar=tensor([0.1999, 0.1938, 0.1854, 0.1999], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 10:38:32,438 INFO [optim.py:369] (1/2) 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 10:38:55,416 INFO [train.py:968] (1/2) Epoch 26, batch 15700, libri_loss[loss=0.2546, simple_loss=0.3257, pruned_loss=0.09173, over 29563.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3315, pruned_loss=0.08531, over 5662840.48 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3293, pruned_loss=0.08612, over 5766093.52 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3316, pruned_loss=0.08469, over 5652269.64 frames. ], batch size: 76, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:38:59,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:39:12,625 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-13 10:39:15,391 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1154773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:39:29,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1154784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:39:41,359 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 15750, giga_loss[loss=0.2221, simple_loss=0.3063, pruned_loss=0.06893, over 28936.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3306, pruned_loss=0.08511, over 5662615.10 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3291, pruned_loss=0.08594, over 5769524.17 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.331, pruned_loss=0.08477, over 5648488.34 frames. ], batch size: 199, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:40:09,806 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5709, 2.0394, 1.3180, 1.0092], device='cuda:1'), covar=tensor([0.8186, 0.4860, 0.4238, 0.6996], device='cuda:1'), in_proj_covar=tensor([0.1811, 0.1700, 0.1637, 0.1475], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 10:40:22,863 INFO [optim.py:369] (1/2) 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 10:40:30,313 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 15800, libri_loss[loss=0.269, simple_loss=0.3493, pruned_loss=0.09432, over 29188.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3283, pruned_loss=0.08407, over 5662824.96 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3286, pruned_loss=0.08584, over 5770941.31 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3291, pruned_loss=0.08382, over 5644715.37 frames. ], batch size: 97, lr: 1.23e-03, grad_scale: 4.0
+2023-03-13 10:41:48,429 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 15850, giga_loss[loss=0.2225, simple_loss=0.3029, pruned_loss=0.07108, over 29000.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3271, pruned_loss=0.08374, over 5669752.66 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3286, pruned_loss=0.08584, over 5770941.31 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3277, pruned_loss=0.08354, over 5655657.74 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:41:51,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1154907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:42:15,041 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1154927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:42:19,553 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:42:25,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1154936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:42:48,854 INFO [train.py:968] (1/2) Epoch 26, batch 15900, giga_loss[loss=0.2655, simple_loss=0.3559, pruned_loss=0.08756, over 28831.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3272, pruned_loss=0.08355, over 5677523.60 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3283, pruned_loss=0.08561, over 5774933.96 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3279, pruned_loss=0.08354, over 5659977.47 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:42:51,110 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 15950, libri_loss[loss=0.2445, simple_loss=0.3247, pruned_loss=0.08219, over 29651.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3296, pruned_loss=0.08432, over 5680043.58 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3282, pruned_loss=0.08555, over 5774723.08 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3302, pruned_loss=0.08434, over 5664915.90 frames. ], batch size: 88, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:44:10,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 10:44:28,235 INFO [optim.py:369] (1/2) 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 10:44:56,562 INFO [train.py:968] (1/2) Epoch 26, batch 16000, giga_loss[loss=0.2312, simple_loss=0.2934, pruned_loss=0.08456, over 24461.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3294, pruned_loss=0.08487, over 5666623.08 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3282, pruned_loss=0.08551, over 5777570.87 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.33, pruned_loss=0.0849, over 5650145.90 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:45:09,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2905, 1.4823, 1.4963, 1.1434], device='cuda:1'), covar=tensor([0.1746, 0.2632, 0.1464, 0.1797], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0701, 0.0965, 0.0866], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 10:45:51,674 INFO [train.py:968] (1/2) Epoch 26, batch 16050, giga_loss[loss=0.3011, simple_loss=0.3836, pruned_loss=0.1093, over 28430.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3318, pruned_loss=0.08608, over 5673105.30 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.328, pruned_loss=0.08544, over 5781184.65 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3324, pruned_loss=0.08618, over 5654466.07 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:46:22,901 INFO [optim.py:369] (1/2) 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 10:46:38,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:46:50,007 INFO [train.py:968] (1/2) Epoch 26, batch 16100, giga_loss[loss=0.2854, simple_loss=0.3639, pruned_loss=0.1034, over 28034.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3351, pruned_loss=0.08779, over 5652908.34 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.328, pruned_loss=0.08536, over 5773990.49 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3358, pruned_loss=0.08797, over 5642634.14 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:47:00,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:47:27,465 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155192.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:47:38,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-13 10:47:45,470 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:47:45,894 INFO [train.py:968] (1/2) Epoch 26, batch 16150, giga_loss[loss=0.2481, simple_loss=0.3319, pruned_loss=0.08216, over 29056.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3356, pruned_loss=0.0878, over 5654935.22 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3274, pruned_loss=0.08531, over 5776095.53 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3368, pruned_loss=0.08804, over 5641196.25 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:47:55,851 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:48:55,299 INFO [train.py:968] (1/2) Epoch 26, batch 16200, giga_loss[loss=0.2304, simple_loss=0.3167, pruned_loss=0.07203, over 28402.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3363, pruned_loss=0.08874, over 5658136.73 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08544, over 5778335.77 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3373, pruned_loss=0.08888, over 5643092.89 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:49:06,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6434, 1.9559, 1.3227, 1.5287], device='cuda:1'), covar=tensor([0.1024, 0.0582, 0.1049, 0.1143], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0444, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 10:49:38,341 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:49:43,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:49:43,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8582, 1.1271, 2.8105, 2.6729], device='cuda:1'), covar=tensor([0.1716, 0.2718, 0.0671, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0666, 0.0978, 0.0944], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 10:49:58,134 INFO [train.py:968] (1/2) Epoch 26, batch 16250, giga_loss[loss=0.2288, simple_loss=0.3242, pruned_loss=0.06674, over 28856.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3339, pruned_loss=0.08758, over 5655407.05 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3276, pruned_loss=0.08553, over 5769970.78 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3348, pruned_loss=0.08765, over 5649075.64 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:50:06,004 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:08,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:18,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:32,324 INFO [optim.py:369] (1/2) 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 10:50:43,234 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:50:57,142 INFO [train.py:968] (1/2) Epoch 26, batch 16300, giga_loss[loss=0.2591, simple_loss=0.3394, pruned_loss=0.08939, over 28705.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3326, pruned_loss=0.08648, over 5668361.53 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3273, pruned_loss=0.08545, over 5768889.61 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3338, pruned_loss=0.08665, over 5661540.34 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:50:57,684 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:51:02,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:51:37,161 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 16350, libri_loss[loss=0.2535, simple_loss=0.3326, pruned_loss=0.08725, over 29500.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.332, pruned_loss=0.08726, over 5666312.98 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.327, pruned_loss=0.08546, over 5769796.06 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3334, pruned_loss=0.08746, over 5656209.10 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:52:30,034 INFO [optim.py:369] (1/2) 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 10:52:40,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5931, 2.3354, 1.7581, 0.8305], device='cuda:1'), covar=tensor([0.7057, 0.3475, 0.4158, 0.7037], device='cuda:1'), in_proj_covar=tensor([0.1811, 0.1696, 0.1639, 0.1478], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 10:52:59,865 INFO [train.py:968] (1/2) Epoch 26, batch 16400, giga_loss[loss=0.2424, simple_loss=0.3271, pruned_loss=0.07883, over 28917.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3287, pruned_loss=0.08602, over 5659737.25 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3267, pruned_loss=0.08519, over 5771935.35 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3302, pruned_loss=0.08644, over 5648332.09 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:53:51,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4144, 1.3260, 3.9672, 3.3465], device='cuda:1'), covar=tensor([0.1684, 0.2857, 0.0446, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0665, 0.0975, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 10:53:57,545 INFO [train.py:968] (1/2) Epoch 26, batch 16450, giga_loss[loss=0.2059, simple_loss=0.3006, pruned_loss=0.05564, over 28704.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3278, pruned_loss=0.0842, over 5665482.24 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3268, pruned_loss=0.08526, over 5772617.02 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3288, pruned_loss=0.08446, over 5653417.16 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:54:30,115 INFO [optim.py:369] (1/2) 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 10:54:57,208 INFO [train.py:968] (1/2) Epoch 26, batch 16500, giga_loss[loss=0.2516, simple_loss=0.3338, pruned_loss=0.08471, over 28683.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3272, pruned_loss=0.08274, over 5672040.88 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3269, pruned_loss=0.08532, over 5771426.40 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3279, pruned_loss=0.08285, over 5661405.22 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 10:55:01,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2006, 1.5073, 1.3891, 1.1779], device='cuda:1'), covar=tensor([0.2868, 0.2462, 0.1766, 0.2425], device='cuda:1'), in_proj_covar=tensor([0.1996, 0.1932, 0.1851, 0.1993], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 10:55:05,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155567.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:55:09,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4822, 1.6978, 1.3885, 1.4537], device='cuda:1'), covar=tensor([0.3103, 0.3018, 0.3529, 0.2754], device='cuda:1'), in_proj_covar=tensor([0.1567, 0.1126, 0.1386, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 10:55:19,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:55:43,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:55:46,580 INFO [train.py:968] (1/2) Epoch 26, batch 16550, giga_loss[loss=0.2655, simple_loss=0.3487, pruned_loss=0.09116, over 28902.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.3276, pruned_loss=0.08059, over 5690847.10 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3259, pruned_loss=0.08464, over 5777198.83 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3291, pruned_loss=0.08116, over 5673517.70 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:56:16,224 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 16600, giga_loss[loss=0.2626, simple_loss=0.3448, pruned_loss=0.09016, over 29001.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.329, pruned_loss=0.08073, over 5688960.39 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3257, pruned_loss=0.08449, over 5780680.56 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3305, pruned_loss=0.08123, over 5670119.21 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:57:09,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5850, 2.3444, 1.5877, 0.6822], device='cuda:1'), covar=tensor([0.6740, 0.3589, 0.5040, 0.7680], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1698, 0.1639, 0.1479], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 10:57:43,083 INFO [train.py:968] (1/2) Epoch 26, batch 16650, giga_loss[loss=0.2438, simple_loss=0.3287, pruned_loss=0.07943, over 28997.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3295, pruned_loss=0.08145, over 5671350.04 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3256, pruned_loss=0.08448, over 5774013.76 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3308, pruned_loss=0.08177, over 5660887.09 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:57:47,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155710.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:57:50,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155713.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 10:58:03,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1155724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:08,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1155727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:18,840 INFO [optim.py:369] (1/2) 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 10:58:25,397 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:27,552 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155742.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 10:58:39,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3452, 1.5169, 1.3172, 1.3848], device='cuda:1'), covar=tensor([0.0743, 0.0331, 0.0334, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0073, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 10:58:42,733 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1155756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 10:58:43,035 INFO [train.py:968] (1/2) Epoch 26, batch 16700, giga_loss[loss=0.2367, simple_loss=0.3045, pruned_loss=0.08451, over 24895.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3296, pruned_loss=0.08187, over 5669574.95 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3258, pruned_loss=0.0847, over 5777004.15 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3307, pruned_loss=0.08181, over 5654530.76 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 10:59:52,702 INFO [train.py:968] (1/2) Epoch 26, batch 16750, giga_loss[loss=0.2561, simple_loss=0.3441, pruned_loss=0.08405, over 29054.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3299, pruned_loss=0.08188, over 5671282.83 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.326, pruned_loss=0.08481, over 5780590.18 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3307, pruned_loss=0.08166, over 5653086.93 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:00:05,669 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 11:00:24,878 INFO [optim.py:369] (1/2) 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 11:00:36,475 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:00:59,405 INFO [train.py:968] (1/2) Epoch 26, batch 16800, giga_loss[loss=0.224, simple_loss=0.3175, pruned_loss=0.06522, over 28014.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3291, pruned_loss=0.08102, over 5670309.28 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3252, pruned_loss=0.08434, over 5785323.47 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3306, pruned_loss=0.08116, over 5648144.22 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:01:21,306 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1155873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:02:09,778 INFO [train.py:968] (1/2) Epoch 26, batch 16850, giga_loss[loss=0.2705, simple_loss=0.3415, pruned_loss=0.09974, over 28144.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.332, pruned_loss=0.08247, over 5672753.32 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3252, pruned_loss=0.08431, over 5786907.73 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3333, pruned_loss=0.08255, over 5651782.31 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:02:52,944 INFO [optim.py:369] (1/2) 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 11:03:17,253 INFO [train.py:968] (1/2) Epoch 26, batch 16900, libri_loss[loss=0.2442, simple_loss=0.325, pruned_loss=0.08168, over 29510.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3345, pruned_loss=0.08334, over 5675476.31 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3253, pruned_loss=0.08444, over 5786909.21 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3355, pruned_loss=0.08329, over 5657181.50 frames. ], batch size: 81, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:03:47,309 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-13 11:03:48,253 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1155978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:04:08,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2608, 2.5888, 1.2422, 1.4756], device='cuda:1'), covar=tensor([0.0986, 0.0450, 0.0964, 0.1290], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0559, 0.0401, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 11:04:30,269 INFO [train.py:968] (1/2) Epoch 26, batch 16950, giga_loss[loss=0.267, simple_loss=0.3453, pruned_loss=0.0943, over 28393.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3336, pruned_loss=0.08329, over 5676988.69 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3252, pruned_loss=0.0844, over 5787531.66 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3345, pruned_loss=0.08328, over 5661740.65 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:04:47,750 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 11:05:14,154 INFO [optim.py:369] (1/2) 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 11:05:41,914 INFO [train.py:968] (1/2) Epoch 26, batch 17000, giga_loss[loss=0.2539, simple_loss=0.3387, pruned_loss=0.08458, over 29048.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3325, pruned_loss=0.08302, over 5678774.95 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3252, pruned_loss=0.08431, over 5785036.95 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3334, pruned_loss=0.08308, over 5668014.37 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:06:47,872 INFO [train.py:968] (1/2) Epoch 26, batch 17050, giga_loss[loss=0.2468, simple_loss=0.3298, pruned_loss=0.08192, over 28107.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3307, pruned_loss=0.08162, over 5668892.27 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3252, pruned_loss=0.08434, over 5778239.77 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3316, pruned_loss=0.08158, over 5662532.23 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 1.0
+2023-03-13 11:06:59,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:07:05,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:07:09,380 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:07:39,626 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156149.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 11:07:43,838 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:07:47,140 INFO [train.py:968] (1/2) Epoch 26, batch 17100, giga_loss[loss=0.2753, simple_loss=0.3516, pruned_loss=0.09952, over 28635.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3305, pruned_loss=0.08182, over 5674301.45 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3252, pruned_loss=0.08438, over 5780953.47 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3312, pruned_loss=0.0817, over 5665052.31 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 1.0
+2023-03-13 11:08:47,739 INFO [train.py:968] (1/2) Epoch 26, batch 17150, giga_loss[loss=0.3021, simple_loss=0.3768, pruned_loss=0.1137, over 28405.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.333, pruned_loss=0.08328, over 5671931.47 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.325, pruned_loss=0.08435, over 5782696.57 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3338, pruned_loss=0.08318, over 5661569.34 frames. ], batch size: 369, lr: 1.22e-03, grad_scale: 1.0
+2023-03-13 11:09:00,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:09:22,705 INFO [optim.py:369] (1/2) 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 11:09:34,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:09:34,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6308, 1.9771, 1.3459, 1.4471], device='cuda:1'), covar=tensor([0.1051, 0.0524, 0.1040, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0402, 0.0441, 0.0517, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 11:09:41,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2200, 1.2331, 3.2138, 2.8855], device='cuda:1'), covar=tensor([0.1535, 0.2799, 0.0546, 0.1779], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0665, 0.0973, 0.0943], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 11:09:44,553 INFO [train.py:968] (1/2) Epoch 26, batch 17200, libri_loss[loss=0.2888, simple_loss=0.3617, pruned_loss=0.1079, over 29519.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3342, pruned_loss=0.08417, over 5674755.83 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3253, pruned_loss=0.08446, over 5784596.01 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3349, pruned_loss=0.08397, over 5661705.94 frames. ], batch size: 81, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:09:44,962 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:09:46,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:09:53,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5212, 1.6566, 1.7135, 1.3133], device='cuda:1'), covar=tensor([0.2016, 0.2806, 0.1701, 0.1967], device='cuda:1'), in_proj_covar=tensor([0.0917, 0.0701, 0.0966, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 11:10:16,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7199, 1.9007, 1.8065, 1.6438], device='cuda:1'), covar=tensor([0.2802, 0.2388, 0.2051, 0.2266], device='cuda:1'), in_proj_covar=tensor([0.1995, 0.1933, 0.1844, 0.1992], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 11:10:18,669 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:10:41,151 INFO [train.py:968] (1/2) Epoch 26, batch 17250, giga_loss[loss=0.2574, simple_loss=0.3215, pruned_loss=0.09666, over 26781.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3315, pruned_loss=0.08415, over 5671017.72 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.325, pruned_loss=0.0843, over 5786450.65 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3324, pruned_loss=0.08413, over 5657516.52 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:10:58,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4450, 1.8437, 1.5725, 1.6947], device='cuda:1'), covar=tensor([0.0687, 0.0283, 0.0316, 0.0736], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0120, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0073, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 11:11:15,580 INFO [optim.py:369] (1/2) 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 11:11:34,338 INFO [train.py:968] (1/2) Epoch 26, batch 17300, giga_loss[loss=0.2534, simple_loss=0.3302, pruned_loss=0.08826, over 28645.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3312, pruned_loss=0.08482, over 5666724.92 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3248, pruned_loss=0.08432, over 5781047.72 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3323, pruned_loss=0.08478, over 5656957.16 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:11:38,697 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:11:42,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:11:59,437 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:11,797 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:11,867 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:15,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:12:30,070 INFO [train.py:968] (1/2) Epoch 26, batch 17350, giga_loss[loss=0.2923, simple_loss=0.3613, pruned_loss=0.1117, over 27603.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3327, pruned_loss=0.08647, over 5657249.24 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3244, pruned_loss=0.08408, over 5781191.42 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3341, pruned_loss=0.08668, over 5646069.02 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:12:48,409 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:13:02,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4794, 1.6280, 1.7255, 1.2879], device='cuda:1'), covar=tensor([0.1767, 0.2789, 0.1549, 0.1948], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0701, 0.0967, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 11:13:04,848 INFO [optim.py:369] (1/2) 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 11:13:22,625 INFO [train.py:968] (1/2) Epoch 26, batch 17400, giga_loss[loss=0.3073, simple_loss=0.389, pruned_loss=0.1128, over 28571.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3414, pruned_loss=0.09099, over 5657559.75 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3244, pruned_loss=0.08404, over 5769821.17 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3428, pruned_loss=0.09128, over 5656785.20 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:14:08,027 INFO [train.py:968] (1/2) Epoch 26, batch 17450, giga_loss[loss=0.329, simple_loss=0.3807, pruned_loss=0.1386, over 23760.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3485, pruned_loss=0.09515, over 5663197.59 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3244, pruned_loss=0.08401, over 5772090.61 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3498, pruned_loss=0.0955, over 5659371.66 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:14:23,421 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156524.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 11:14:33,985 INFO [optim.py:369] (1/2) 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 11:14:44,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4797, 1.2571, 4.4066, 3.4386], device='cuda:1'), covar=tensor([0.1734, 0.3037, 0.0409, 0.1097], device='cuda:1'), in_proj_covar=tensor([0.0785, 0.0667, 0.0976, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 11:14:48,574 INFO [train.py:968] (1/2) Epoch 26, batch 17500, giga_loss[loss=0.292, simple_loss=0.3612, pruned_loss=0.1114, over 29056.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3466, pruned_loss=0.09431, over 5675595.21 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3244, pruned_loss=0.08371, over 5776806.01 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3486, pruned_loss=0.09535, over 5663784.27 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:14:57,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5115, 2.2192, 1.6861, 0.7765], device='cuda:1'), covar=tensor([0.7777, 0.3495, 0.4550, 0.7665], device='cuda:1'), in_proj_covar=tensor([0.1810, 0.1701, 0.1643, 0.1478], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 11:15:03,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:15:32,080 INFO [train.py:968] (1/2) Epoch 26, batch 17550, libri_loss[loss=0.2748, simple_loss=0.3515, pruned_loss=0.09906, over 29245.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3413, pruned_loss=0.09218, over 5683450.13 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3246, pruned_loss=0.08368, over 5779433.71 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3434, pruned_loss=0.09338, over 5668178.39 frames. ], batch size: 97, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:15:51,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1428, 2.1569, 2.2799, 1.8745], device='cuda:1'), covar=tensor([0.1872, 0.2366, 0.1471, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0702, 0.0968, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 11:15:56,905 INFO [optim.py:369] (1/2) 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 11:16:14,569 INFO [train.py:968] (1/2) Epoch 26, batch 17600, libri_loss[loss=0.2195, simple_loss=0.2994, pruned_loss=0.06983, over 28056.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3342, pruned_loss=0.08902, over 5683301.78 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3249, pruned_loss=0.08383, over 5768867.53 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3359, pruned_loss=0.09, over 5678740.71 frames. ], batch size: 62, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:16:24,748 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156667.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 11:16:28,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1156670.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 11:16:50,420 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1156699.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 11:16:58,548 INFO [train.py:968] (1/2) Epoch 26, batch 17650, giga_loss[loss=0.2201, simple_loss=0.2962, pruned_loss=0.072, over 28923.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3268, pruned_loss=0.08611, over 5682365.35 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3251, pruned_loss=0.08407, over 5761816.30 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.328, pruned_loss=0.08668, over 5684295.78 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:17:27,361 INFO [optim.py:369] (1/2) 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 11:17:38,710 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 17700, giga_loss[loss=0.2397, simple_loss=0.3141, pruned_loss=0.08263, over 27981.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3205, pruned_loss=0.08361, over 5686087.14 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3255, pruned_loss=0.08415, over 5763334.07 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3211, pruned_loss=0.08402, over 5684043.67 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:17:50,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2169, 1.4449, 1.3534, 1.1282], device='cuda:1'), covar=tensor([0.3227, 0.2846, 0.1883, 0.2602], device='cuda:1'), in_proj_covar=tensor([0.2008, 0.1940, 0.1858, 0.2005], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 11:18:11,477 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5042, 1.6728, 1.6333, 1.4226], device='cuda:1'), covar=tensor([0.3165, 0.2894, 0.2204, 0.2814], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1939, 0.1857, 0.2004], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 11:18:22,171 INFO [train.py:968] (1/2) Epoch 26, batch 17750, giga_loss[loss=0.2591, simple_loss=0.3308, pruned_loss=0.09368, over 28769.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3159, pruned_loss=0.08173, over 5687663.15 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.326, pruned_loss=0.08436, over 5766924.28 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3156, pruned_loss=0.08182, over 5680817.62 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:18:47,621 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 17800, giga_loss[loss=0.2194, simple_loss=0.292, pruned_loss=0.07342, over 28756.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.3129, pruned_loss=0.08042, over 5687005.11 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.327, pruned_loss=0.08471, over 5757596.95 frames. ], giga_tot_loss[loss=0.2358, simple_loss=0.3114, pruned_loss=0.08007, over 5687122.47 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:19:35,627 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1156897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:19:37,197 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 17850, giga_loss[loss=0.1978, simple_loss=0.2845, pruned_loss=0.05554, over 28871.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.3104, pruned_loss=0.0786, over 5695258.09 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3269, pruned_loss=0.08442, over 5761312.99 frames. ], giga_tot_loss[loss=0.2328, simple_loss=0.3088, pruned_loss=0.07844, over 5689830.21 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:19:43,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1156908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:20:03,334 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:20:17,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1156946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:20:28,842 INFO [train.py:968] (1/2) Epoch 26, batch 17900, giga_loss[loss=0.1859, simple_loss=0.2529, pruned_loss=0.05945, over 24060.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.3072, pruned_loss=0.07765, over 5685788.37 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3273, pruned_loss=0.08461, over 5762645.02 frames. ], giga_tot_loss[loss=0.23, simple_loss=0.3054, pruned_loss=0.07727, over 5679545.16 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:21:09,108 INFO [train.py:968] (1/2) Epoch 26, batch 17950, giga_loss[loss=0.1998, simple_loss=0.2822, pruned_loss=0.05871, over 28843.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.304, pruned_loss=0.0761, over 5696235.41 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3275, pruned_loss=0.08453, over 5760756.31 frames. ], giga_tot_loss[loss=0.2266, simple_loss=0.3019, pruned_loss=0.0757, over 5691411.41 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:21:35,551 INFO [optim.py:369] (1/2) 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 11:21:50,269 INFO [train.py:968] (1/2) Epoch 26, batch 18000, giga_loss[loss=0.1993, simple_loss=0.2677, pruned_loss=0.06545, over 28068.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.3017, pruned_loss=0.07499, over 5689280.19 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3277, pruned_loss=0.08453, over 5756060.92 frames. ], giga_tot_loss[loss=0.2239, simple_loss=0.299, pruned_loss=0.07438, over 5686621.62 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:21:50,269 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 11:21:58,990 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 11:22:26,729 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1157089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:22:31,013 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1157092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:22:42,118 INFO [train.py:968] (1/2) Epoch 26, batch 18050, giga_loss[loss=0.1909, simple_loss=0.2549, pruned_loss=0.06339, over 23799.00 frames. ], tot_loss[loss=0.2238, simple_loss=0.2994, pruned_loss=0.0741, over 5686760.26 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3282, pruned_loss=0.08472, over 5758215.57 frames. ], giga_tot_loss[loss=0.2213, simple_loss=0.2962, pruned_loss=0.07323, over 5681340.65 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:22:53,432 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1157121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:23:08,091 INFO [optim.py:369] (1/2) 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 11:23:23,473 INFO [train.py:968] (1/2) Epoch 26, batch 18100, giga_loss[loss=0.22, simple_loss=0.29, pruned_loss=0.07496, over 28995.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2972, pruned_loss=0.07311, over 5698517.99 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3282, pruned_loss=0.08465, over 5762902.08 frames. ], giga_tot_loss[loss=0.219, simple_loss=0.2937, pruned_loss=0.0721, over 5687771.16 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:23:39,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-13 11:23:51,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4569, 1.7009, 1.3140, 1.2179], device='cuda:1'), covar=tensor([0.1076, 0.0558, 0.1137, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0441, 0.0518, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 11:24:06,015 INFO [train.py:968] (1/2) Epoch 26, batch 18150, giga_loss[loss=0.1929, simple_loss=0.2721, pruned_loss=0.05684, over 28988.00 frames. ], tot_loss[loss=0.2202, simple_loss=0.2956, pruned_loss=0.07242, over 5711670.39 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3282, pruned_loss=0.08448, over 5769529.65 frames. ], giga_tot_loss[loss=0.2167, simple_loss=0.2911, pruned_loss=0.07117, over 5693860.56 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:24:29,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9140, 2.1145, 1.9602, 1.8739], device='cuda:1'), covar=tensor([0.2174, 0.2499, 0.2357, 0.2386], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0743, 0.0713, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 11:24:32,471 INFO [optim.py:369] (1/2) 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 11:24:43,275 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4468, 1.9858, 1.4594, 0.8723], device='cuda:1'), covar=tensor([0.6731, 0.3380, 0.3891, 0.6530], device='cuda:1'), in_proj_covar=tensor([0.1806, 0.1700, 0.1636, 0.1473], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 11:24:47,976 INFO [train.py:968] (1/2) Epoch 26, batch 18200, giga_loss[loss=0.286, simple_loss=0.3545, pruned_loss=0.1087, over 28698.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2971, pruned_loss=0.07362, over 5713565.39 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3284, pruned_loss=0.08447, over 5773781.25 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2922, pruned_loss=0.07221, over 5693272.15 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:25:13,966 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1157283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:25:16,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5315, 3.3696, 3.1824, 1.7740], device='cuda:1'), covar=tensor([0.0808, 0.0885, 0.0825, 0.2366], device='cuda:1'), in_proj_covar=tensor([0.1251, 0.1152, 0.0972, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 11:25:36,111 INFO [train.py:968] (1/2) Epoch 26, batch 18250, libri_loss[loss=0.2982, simple_loss=0.3646, pruned_loss=0.1159, over 29553.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3071, pruned_loss=0.07871, over 5713268.97 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3283, pruned_loss=0.08455, over 5777359.45 frames. ], giga_tot_loss[loss=0.2287, simple_loss=0.3027, pruned_loss=0.07733, over 5692316.74 frames. ], batch size: 79, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:25:57,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2312, 1.3819, 1.3893, 1.1646], device='cuda:1'), covar=tensor([0.2387, 0.2589, 0.1651, 0.2252], device='cuda:1'), in_proj_covar=tensor([0.2007, 0.1943, 0.1861, 0.2003], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 11:26:04,082 INFO [optim.py:369] (1/2) 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 11:26:17,793 INFO [train.py:968] (1/2) Epoch 26, batch 18300, giga_loss[loss=0.2597, simple_loss=0.3436, pruned_loss=0.08786, over 29033.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3217, pruned_loss=0.08632, over 5709143.63 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3284, pruned_loss=0.08436, over 5778537.48 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3177, pruned_loss=0.08537, over 5689748.28 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:26:25,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2548, 0.8832, 1.0560, 1.3861], device='cuda:1'), covar=tensor([0.0785, 0.0413, 0.0358, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 11:26:55,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5645, 3.5051, 1.5529, 1.5847], device='cuda:1'), covar=tensor([0.1003, 0.0309, 0.0933, 0.1371], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0559, 0.0401, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 11:27:00,661 INFO [train.py:968] (1/2) Epoch 26, batch 18350, giga_loss[loss=0.2927, simple_loss=0.3643, pruned_loss=0.1105, over 28813.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3311, pruned_loss=0.09077, over 5711739.77 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3283, pruned_loss=0.08419, over 5780795.02 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3281, pruned_loss=0.09027, over 5693086.83 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:27:15,459 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1157426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:27:18,636 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1157429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:27:26,701 INFO [optim.py:369] (1/2) 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 11:27:41,640 INFO [train.py:968] (1/2) Epoch 26, batch 18400, giga_loss[loss=0.3092, simple_loss=0.3931, pruned_loss=0.1126, over 27597.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3365, pruned_loss=0.09221, over 5708535.41 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3283, pruned_loss=0.08418, over 5782320.33 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3342, pruned_loss=0.09197, over 5691083.69 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:27:43,373 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1157458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:28:23,005 INFO [train.py:968] (1/2) Epoch 26, batch 18450, giga_loss[loss=0.2545, simple_loss=0.3348, pruned_loss=0.08713, over 28817.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3406, pruned_loss=0.09333, over 5704926.06 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.329, pruned_loss=0.08454, over 5780777.61 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3384, pruned_loss=0.09299, over 5690859.87 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:28:44,956 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6601, 1.8743, 1.6279, 1.6894], device='cuda:1'), covar=tensor([0.2404, 0.2270, 0.2326, 0.2073], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1127, 0.1384, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 11:28:51,883 INFO [optim.py:369] (1/2) 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 11:29:05,475 INFO [train.py:968] (1/2) Epoch 26, batch 18500, giga_loss[loss=0.2623, simple_loss=0.349, pruned_loss=0.08776, over 28922.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3412, pruned_loss=0.09323, over 5691152.30 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3293, pruned_loss=0.08476, over 5772634.93 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3394, pruned_loss=0.09302, over 5685192.11 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:29:11,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-13 11:29:47,776 INFO [train.py:968] (1/2) Epoch 26, batch 18550, giga_loss[loss=0.2976, simple_loss=0.3694, pruned_loss=0.1129, over 28869.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3424, pruned_loss=0.09437, over 5693790.21 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3292, pruned_loss=0.08449, over 5774845.81 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3417, pruned_loss=0.09484, over 5683806.48 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:30:15,972 INFO [optim.py:369] (1/2) 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 11:30:31,909 INFO [train.py:968] (1/2) Epoch 26, batch 18600, libri_loss[loss=0.2072, simple_loss=0.2893, pruned_loss=0.06257, over 29366.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3448, pruned_loss=0.09646, over 5702628.79 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3286, pruned_loss=0.08416, over 5778739.09 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3452, pruned_loss=0.09746, over 5689044.44 frames. ], batch size: 67, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:31:00,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-13 11:31:13,272 INFO [train.py:968] (1/2) Epoch 26, batch 18650, giga_loss[loss=0.2854, simple_loss=0.3659, pruned_loss=0.1024, over 29008.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3479, pruned_loss=0.09808, over 5702064.51 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3287, pruned_loss=0.08414, over 5780091.37 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3483, pruned_loss=0.09908, over 5689162.19 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:31:40,567 INFO [optim.py:369] (1/2) 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 11:31:53,611 INFO [train.py:968] (1/2) Epoch 26, batch 18700, giga_loss[loss=0.3168, simple_loss=0.3877, pruned_loss=0.123, over 29058.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3507, pruned_loss=0.09848, over 5706628.17 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3291, pruned_loss=0.08422, over 5779001.40 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.351, pruned_loss=0.09944, over 5696246.90 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:32:20,494 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1157788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:32:35,374 INFO [train.py:968] (1/2) Epoch 26, batch 18750, libri_loss[loss=0.2632, simple_loss=0.3463, pruned_loss=0.09008, over 29514.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3517, pruned_loss=0.09832, over 5709483.64 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3291, pruned_loss=0.08415, over 5782119.27 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3525, pruned_loss=0.09951, over 5696532.28 frames. ], batch size: 84, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:32:58,921 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 18800, giga_loss[loss=0.3689, simple_loss=0.4272, pruned_loss=0.1554, over 28975.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3529, pruned_loss=0.09823, over 5712603.92 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3293, pruned_loss=0.0842, over 5785584.29 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3544, pruned_loss=0.09979, over 5695975.78 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:33:53,025 INFO [train.py:968] (1/2) Epoch 26, batch 18850, giga_loss[loss=0.264, simple_loss=0.3543, pruned_loss=0.08685, over 28922.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3525, pruned_loss=0.09707, over 5711518.28 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3299, pruned_loss=0.08451, over 5787195.90 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3537, pruned_loss=0.09826, over 5695546.50 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:34:15,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4672, 1.7800, 1.4171, 1.3253], device='cuda:1'), covar=tensor([0.2879, 0.2861, 0.3247, 0.2521], device='cuda:1'), in_proj_covar=tensor([0.1564, 0.1128, 0.1381, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 11:34:21,668 INFO [optim.py:369] (1/2) 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 11:34:34,902 INFO [train.py:968] (1/2) Epoch 26, batch 18900, giga_loss[loss=0.2711, simple_loss=0.3517, pruned_loss=0.09522, over 28856.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3509, pruned_loss=0.09508, over 5713158.12 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3301, pruned_loss=0.08462, over 5788409.97 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3518, pruned_loss=0.09602, over 5698793.37 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:35:14,961 INFO [train.py:968] (1/2) Epoch 26, batch 18950, giga_loss[loss=0.2578, simple_loss=0.3413, pruned_loss=0.08719, over 28717.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3502, pruned_loss=0.09444, over 5715057.55 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3302, pruned_loss=0.08454, over 5789852.47 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3511, pruned_loss=0.09538, over 5701368.28 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:35:44,118 INFO [optim.py:369] (1/2) 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 11:35:58,665 INFO [train.py:968] (1/2) Epoch 26, batch 19000, giga_loss[loss=0.2869, simple_loss=0.3691, pruned_loss=0.1024, over 28229.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.353, pruned_loss=0.09885, over 5701168.23 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3302, pruned_loss=0.08441, over 5791233.48 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3541, pruned_loss=0.09993, over 5687732.46 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:36:11,936 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:36:42,575 INFO [train.py:968] (1/2) Epoch 26, batch 19050, giga_loss[loss=0.2729, simple_loss=0.3444, pruned_loss=0.1007, over 28613.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3563, pruned_loss=0.1033, over 5689191.17 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3309, pruned_loss=0.08474, over 5784414.07 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3572, pruned_loss=0.1043, over 5681914.14 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:37:08,077 INFO [optim.py:369] (1/2) 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 11:37:20,121 INFO [train.py:968] (1/2) Epoch 26, batch 19100, giga_loss[loss=0.3424, simple_loss=0.4068, pruned_loss=0.139, over 29051.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3553, pruned_loss=0.1033, over 5700345.74 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3307, pruned_loss=0.08478, over 5786953.24 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3568, pruned_loss=0.1045, over 5690009.01 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:37:27,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:38:03,287 INFO [train.py:968] (1/2) Epoch 26, batch 19150, giga_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1213, over 27935.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3526, pruned_loss=0.1023, over 5705078.46 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3304, pruned_loss=0.08453, over 5790562.18 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3546, pruned_loss=0.104, over 5691382.10 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:38:30,298 INFO [optim.py:369] (1/2) 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 11:38:45,749 INFO [train.py:968] (1/2) Epoch 26, batch 19200, giga_loss[loss=0.2792, simple_loss=0.3429, pruned_loss=0.1078, over 23606.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3508, pruned_loss=0.1014, over 5702704.82 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3306, pruned_loss=0.08448, over 5794215.88 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3529, pruned_loss=0.1033, over 5686237.52 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:39:28,515 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 19250, giga_loss[loss=0.2399, simple_loss=0.3222, pruned_loss=0.07877, over 28916.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3505, pruned_loss=0.1008, over 5700158.94 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3308, pruned_loss=0.0846, over 5795810.68 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3522, pruned_loss=0.1024, over 5684313.58 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:39:30,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:39:40,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-13 11:39:53,729 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:39:56,536 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:39:57,738 INFO [optim.py:369] (1/2) 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 11:40:11,873 INFO [train.py:968] (1/2) Epoch 26, batch 19300, libri_loss[loss=0.2305, simple_loss=0.315, pruned_loss=0.07296, over 29584.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3475, pruned_loss=0.09821, over 5691195.82 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3313, pruned_loss=0.08477, over 5787292.21 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3491, pruned_loss=0.09987, over 5683618.76 frames. ], batch size: 74, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:40:39,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:40:51,446 INFO [train.py:968] (1/2) Epoch 26, batch 19350, giga_loss[loss=0.2726, simple_loss=0.3396, pruned_loss=0.1028, over 28330.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3426, pruned_loss=0.09525, over 5678296.21 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3316, pruned_loss=0.08479, over 5776350.24 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.344, pruned_loss=0.09692, over 5679385.23 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:41:26,478 INFO [optim.py:369] (1/2) 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 11:41:28,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:41:37,887 INFO [train.py:968] (1/2) Epoch 26, batch 19400, giga_loss[loss=0.2372, simple_loss=0.3154, pruned_loss=0.07948, over 28245.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3365, pruned_loss=0.09208, over 5680448.10 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3315, pruned_loss=0.08475, over 5778437.60 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3378, pruned_loss=0.09359, over 5677718.27 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:42:23,078 INFO [train.py:968] (1/2) Epoch 26, batch 19450, giga_loss[loss=0.231, simple_loss=0.3082, pruned_loss=0.07684, over 28680.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3325, pruned_loss=0.09014, over 5680556.39 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3318, pruned_loss=0.08492, over 5774751.40 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3335, pruned_loss=0.0915, over 5677265.62 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:42:50,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:42:55,180 INFO [optim.py:369] (1/2) 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 11:43:07,436 INFO [train.py:968] (1/2) Epoch 26, batch 19500, giga_loss[loss=0.2453, simple_loss=0.3274, pruned_loss=0.0816, over 28758.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3306, pruned_loss=0.08861, over 5687508.66 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3319, pruned_loss=0.08494, over 5777054.51 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3313, pruned_loss=0.08975, over 5681224.68 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:43:35,600 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1158588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:43:37,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:43:51,108 INFO [train.py:968] (1/2) Epoch 26, batch 19550, giga_loss[loss=0.2624, simple_loss=0.3387, pruned_loss=0.09301, over 28886.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3312, pruned_loss=0.08829, over 5691107.14 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.332, pruned_loss=0.085, over 5769553.28 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3316, pruned_loss=0.08918, over 5691751.44 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:44:00,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6725, 1.7957, 1.5308, 1.6138], device='cuda:1'), covar=tensor([0.2741, 0.2929, 0.3276, 0.2471], device='cuda:1'), in_proj_covar=tensor([0.1562, 0.1126, 0.1381, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 11:44:04,864 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:44:15,065 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:44:32,229 INFO [train.py:968] (1/2) Epoch 26, batch 19600, giga_loss[loss=0.2332, simple_loss=0.3087, pruned_loss=0.07888, over 28402.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3315, pruned_loss=0.08845, over 5697989.42 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3316, pruned_loss=0.08461, over 5774532.52 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3321, pruned_loss=0.08965, over 5691283.36 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:45:13,977 INFO [train.py:968] (1/2) Epoch 26, batch 19650, giga_loss[loss=0.2208, simple_loss=0.3058, pruned_loss=0.06789, over 29050.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3295, pruned_loss=0.08774, over 5707525.61 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3319, pruned_loss=0.08473, over 5772329.64 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3298, pruned_loss=0.0886, over 5703761.43 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:45:20,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:45:42,683 INFO [optim.py:369] (1/2) 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 11:45:54,290 INFO [train.py:968] (1/2) Epoch 26, batch 19700, giga_loss[loss=0.2353, simple_loss=0.3152, pruned_loss=0.07775, over 28843.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.327, pruned_loss=0.08679, over 5713293.86 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3321, pruned_loss=0.08481, over 5771212.99 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.327, pruned_loss=0.08738, over 5711137.55 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:46:00,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:46:34,775 INFO [train.py:968] (1/2) Epoch 26, batch 19750, giga_loss[loss=0.2221, simple_loss=0.2974, pruned_loss=0.07342, over 28468.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3248, pruned_loss=0.08603, over 5714907.62 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3322, pruned_loss=0.08476, over 5771684.90 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3246, pruned_loss=0.08656, over 5712277.41 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:47:05,836 INFO [optim.py:369] (1/2) 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 11:47:15,784 INFO [train.py:968] (1/2) Epoch 26, batch 19800, giga_loss[loss=0.219, simple_loss=0.2927, pruned_loss=0.07259, over 28749.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3229, pruned_loss=0.08555, over 5713630.47 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3328, pruned_loss=0.08506, over 5764101.91 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3221, pruned_loss=0.08572, over 5717344.55 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:47:16,833 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1158858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:19,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:42,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:56,069 INFO [train.py:968] (1/2) Epoch 26, batch 19850, giga_loss[loss=0.1982, simple_loss=0.2837, pruned_loss=0.05639, over 28994.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3217, pruned_loss=0.08488, over 5715544.69 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3333, pruned_loss=0.08512, over 5766935.68 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3204, pruned_loss=0.08495, over 5715065.25 frames. ], batch size: 175, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:47:56,993 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1158908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:59,264 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1158911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:47:59,278 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1158911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:48:01,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5584, 1.6382, 1.7851, 1.3743], device='cuda:1'), covar=tensor([0.1915, 0.2584, 0.1515, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0707, 0.0971, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 11:48:14,097 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1158929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:48:23,641 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1158940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:48:25,472 INFO [optim.py:369] (1/2) 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 11:48:36,382 INFO [train.py:968] (1/2) Epoch 26, batch 19900, giga_loss[loss=0.2231, simple_loss=0.3033, pruned_loss=0.07145, over 28978.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3198, pruned_loss=0.08387, over 5708581.53 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3337, pruned_loss=0.08515, over 5760323.69 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3181, pruned_loss=0.08388, over 5713594.26 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:48:55,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 11:49:16,635 INFO [train.py:968] (1/2) Epoch 26, batch 19950, giga_loss[loss=0.2225, simple_loss=0.2988, pruned_loss=0.07313, over 28798.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3183, pruned_loss=0.0829, over 5710801.99 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3343, pruned_loss=0.08542, over 5753335.76 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3161, pruned_loss=0.08263, over 5719623.71 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:49:16,887 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:49:46,022 INFO [optim.py:369] (1/2) 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 11:49:54,981 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:49:56,654 INFO [train.py:968] (1/2) Epoch 26, batch 20000, giga_loss[loss=0.2508, simple_loss=0.3185, pruned_loss=0.09156, over 28881.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.3168, pruned_loss=0.08218, over 5717343.63 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3348, pruned_loss=0.08543, over 5755750.98 frames. ], giga_tot_loss[loss=0.2391, simple_loss=0.3143, pruned_loss=0.0819, over 5721457.23 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:49:56,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:50:17,773 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:50:34,051 INFO [train.py:968] (1/2) Epoch 26, batch 20050, giga_loss[loss=0.2263, simple_loss=0.3042, pruned_loss=0.07422, over 28931.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3159, pruned_loss=0.08172, over 5728583.28 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3351, pruned_loss=0.0856, over 5757580.15 frames. ], giga_tot_loss[loss=0.238, simple_loss=0.3134, pruned_loss=0.08131, over 5729756.96 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:51:06,432 INFO [optim.py:369] (1/2) 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 11:51:11,148 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:51:17,600 INFO [train.py:968] (1/2) Epoch 26, batch 20100, giga_loss[loss=0.2528, simple_loss=0.3276, pruned_loss=0.08894, over 28841.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3205, pruned_loss=0.08441, over 5710953.13 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3359, pruned_loss=0.08584, over 5751081.44 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3174, pruned_loss=0.08382, over 5716733.58 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:51:18,487 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5001, 1.9923, 1.0432, 1.4868], device='cuda:1'), covar=tensor([0.1299, 0.0701, 0.1559, 0.1411], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0448, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 11:51:38,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:51:40,819 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 20150, giga_loss[loss=0.2642, simple_loss=0.3438, pruned_loss=0.09227, over 28981.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.327, pruned_loss=0.08829, over 5711283.39 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3359, pruned_loss=0.08574, over 5751839.33 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3244, pruned_loss=0.08789, over 5714752.02 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 11:52:16,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0691, 1.0929, 3.4021, 2.8918], device='cuda:1'), covar=tensor([0.1733, 0.2826, 0.0503, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0781, 0.0660, 0.0973, 0.0946], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 11:52:37,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2175, 1.5479, 1.4922, 1.4805], device='cuda:1'), covar=tensor([0.2004, 0.1517, 0.2258, 0.1549], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0754, 0.0723, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 11:52:41,996 INFO [optim.py:369] (1/2) 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 11:52:55,186 INFO [train.py:968] (1/2) Epoch 26, batch 20200, giga_loss[loss=0.3086, simple_loss=0.3739, pruned_loss=0.1216, over 28540.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3375, pruned_loss=0.09569, over 5690307.96 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3364, pruned_loss=0.08586, over 5746085.78 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.335, pruned_loss=0.09537, over 5696611.96 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:53:36,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:53:38,958 INFO [train.py:968] (1/2) Epoch 26, batch 20250, giga_loss[loss=0.247, simple_loss=0.3346, pruned_loss=0.07965, over 28890.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3412, pruned_loss=0.0966, over 5691946.70 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3364, pruned_loss=0.08576, over 5748800.18 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3392, pruned_loss=0.09664, over 5693418.38 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:54:13,177 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 20300, giga_loss[loss=0.3605, simple_loss=0.3967, pruned_loss=0.1621, over 23506.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3464, pruned_loss=0.09911, over 5673457.26 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3365, pruned_loss=0.08569, over 5748793.37 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3449, pruned_loss=0.09948, over 5673220.35 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:54:34,430 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:54:39,665 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:55:07,201 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6210, 1.7989, 1.4698, 1.6464], device='cuda:1'), covar=tensor([0.2625, 0.2665, 0.3008, 0.2380], device='cuda:1'), in_proj_covar=tensor([0.1566, 0.1129, 0.1385, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 11:55:07,500 INFO [train.py:968] (1/2) Epoch 26, batch 20350, giga_loss[loss=0.3137, simple_loss=0.3886, pruned_loss=0.1194, over 28768.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3509, pruned_loss=0.1018, over 5666646.62 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3365, pruned_loss=0.08565, over 5744351.60 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3501, pruned_loss=0.1026, over 5668646.99 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:55:40,232 INFO [optim.py:369] (1/2) 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 11:55:42,135 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:55:43,447 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159449.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 11:55:44,396 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:55:48,945 INFO [train.py:968] (1/2) Epoch 26, batch 20400, libri_loss[loss=0.2525, simple_loss=0.3269, pruned_loss=0.08905, over 29647.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3539, pruned_loss=0.1034, over 5670751.57 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3364, pruned_loss=0.08563, over 5746101.94 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3535, pruned_loss=0.1042, over 5669827.77 frames. ], batch size: 73, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:56:07,154 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:56:10,953 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1159484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:56:30,617 INFO [train.py:968] (1/2) Epoch 26, batch 20450, giga_loss[loss=0.2516, simple_loss=0.3297, pruned_loss=0.08669, over 28637.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3486, pruned_loss=0.09912, over 5677050.38 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3367, pruned_loss=0.08573, over 5748291.55 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3486, pruned_loss=0.1002, over 5672047.79 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:56:46,951 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8230, 3.6561, 3.4358, 1.6263], device='cuda:1'), covar=tensor([0.0774, 0.0893, 0.0798, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.1255, 0.1160, 0.0976, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 11:56:58,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2822, 1.3377, 4.1395, 3.3572], device='cuda:1'), covar=tensor([0.1863, 0.2907, 0.0437, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0782, 0.0662, 0.0974, 0.0947], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 11:56:59,318 INFO [optim.py:369] (1/2) 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 11:57:06,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6434, 1.7232, 1.8462, 1.4123], device='cuda:1'), covar=tensor([0.1940, 0.2639, 0.1602, 0.1878], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0708, 0.0972, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 11:57:07,636 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:57:09,976 INFO [train.py:968] (1/2) Epoch 26, batch 20500, giga_loss[loss=0.2699, simple_loss=0.3554, pruned_loss=0.09222, over 28971.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3472, pruned_loss=0.09751, over 5698007.54 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3367, pruned_loss=0.08573, over 5752741.80 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3473, pruned_loss=0.09866, over 5688575.25 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 11:57:28,583 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 11:57:50,897 INFO [train.py:968] (1/2) Epoch 26, batch 20550, libri_loss[loss=0.284, simple_loss=0.3612, pruned_loss=0.1034, over 29287.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3483, pruned_loss=0.09818, over 5682263.50 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3374, pruned_loss=0.08626, over 5739081.74 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.348, pruned_loss=0.09893, over 5685674.18 frames. ], batch size: 97, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:58:23,313 INFO [optim.py:369] (1/2) 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 11:58:31,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4748, 3.7573, 1.6194, 1.7033], device='cuda:1'), covar=tensor([0.0987, 0.0310, 0.0853, 0.1260], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0557, 0.0398, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 11:58:32,081 INFO [train.py:968] (1/2) Epoch 26, batch 20600, giga_loss[loss=0.2786, simple_loss=0.3486, pruned_loss=0.1043, over 28674.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3484, pruned_loss=0.09758, over 5690438.66 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3371, pruned_loss=0.08599, over 5743257.45 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3488, pruned_loss=0.09878, over 5687661.69 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:59:07,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:09,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:14,310 INFO [train.py:968] (1/2) Epoch 26, batch 20650, giga_loss[loss=0.3018, simple_loss=0.3795, pruned_loss=0.1121, over 28850.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.35, pruned_loss=0.0989, over 5692149.72 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3375, pruned_loss=0.08612, over 5741555.94 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3502, pruned_loss=0.09996, over 5690769.42 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 11:59:33,491 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:43,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:46,290 INFO [optim.py:369] (1/2) 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 11:59:48,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 11:59:52,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4097, 1.8562, 1.1902, 0.8135], device='cuda:1'), covar=tensor([0.6696, 0.3073, 0.3524, 0.6117], device='cuda:1'), in_proj_covar=tensor([0.1805, 0.1693, 0.1635, 0.1471], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 11:59:59,910 INFO [train.py:968] (1/2) Epoch 26, batch 20700, giga_loss[loss=0.3831, simple_loss=0.4177, pruned_loss=0.1743, over 26720.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3524, pruned_loss=0.101, over 5684880.05 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3379, pruned_loss=0.08627, over 5741859.61 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3523, pruned_loss=0.1018, over 5683103.81 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:00:45,389 INFO [train.py:968] (1/2) Epoch 26, batch 20750, giga_loss[loss=0.3734, simple_loss=0.4141, pruned_loss=0.1663, over 26521.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3549, pruned_loss=0.1039, over 5681217.25 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3378, pruned_loss=0.08621, over 5743673.66 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3551, pruned_loss=0.1047, over 5677911.54 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:00:58,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159824.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:01:08,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7449, 1.8968, 1.8773, 1.6677], device='cuda:1'), covar=tensor([0.2808, 0.2229, 0.1873, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.2031, 0.1973, 0.1891, 0.2035], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 12:01:15,282 INFO [optim.py:369] (1/2) 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 12:01:23,952 INFO [train.py:968] (1/2) Epoch 26, batch 20800, giga_loss[loss=0.2705, simple_loss=0.3513, pruned_loss=0.09491, over 28830.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3538, pruned_loss=0.103, over 5694032.33 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.338, pruned_loss=0.08617, over 5748917.21 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3545, pruned_loss=0.1045, over 5683956.15 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:01:25,543 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1159859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:43,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:45,662 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:47,574 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:01:50,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:02:00,711 INFO [train.py:968] (1/2) Epoch 26, batch 20850, giga_loss[loss=0.258, simple_loss=0.3406, pruned_loss=0.08766, over 28595.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.353, pruned_loss=0.1017, over 5707439.80 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3381, pruned_loss=0.08606, over 5754435.96 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.354, pruned_loss=0.1036, over 5692794.15 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:02:08,114 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:02:24,308 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:02:37,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4399, 1.4352, 4.5138, 3.4098], device='cuda:1'), covar=tensor([0.1687, 0.2843, 0.0390, 0.1027], device='cuda:1'), in_proj_covar=tensor([0.0784, 0.0663, 0.0974, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 12:02:42,281 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 26, batch 20900, giga_loss[loss=0.259, simple_loss=0.3519, pruned_loss=0.08311, over 28943.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3522, pruned_loss=0.1003, over 5691109.94 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3385, pruned_loss=0.08635, over 5738210.10 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.353, pruned_loss=0.1019, over 5692810.84 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:03:00,998 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1159967.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:03:02,943 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1159970.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:03:25,788 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1159999.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:03:28,578 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:03:31,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:03:32,522 INFO [train.py:968] (1/2) Epoch 26, batch 20950, giga_loss[loss=0.2829, simple_loss=0.3657, pruned_loss=0.1001, over 28714.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3523, pruned_loss=0.09918, over 5699441.22 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3382, pruned_loss=0.08626, over 5742367.92 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3535, pruned_loss=0.1008, over 5695984.86 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:03:54,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:04:04,111 INFO [optim.py:369] (1/2) 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 12:04:12,379 INFO [train.py:968] (1/2) Epoch 26, batch 21000, giga_loss[loss=0.256, simple_loss=0.3377, pruned_loss=0.08716, over 28523.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3501, pruned_loss=0.09795, over 5691835.26 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3378, pruned_loss=0.08605, over 5735970.51 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3517, pruned_loss=0.09965, over 5694831.65 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:04:12,379 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 12:04:20,545 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 12:04:21,723 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.65 vs. limit=2.0
+2023-03-13 12:04:53,065 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4389, 4.2750, 4.0743, 1.7758], device='cuda:1'), covar=tensor([0.0633, 0.0786, 0.0777, 0.2192], device='cuda:1'), in_proj_covar=tensor([0.1257, 0.1163, 0.0980, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 12:04:56,803 INFO [train.py:968] (1/2) Epoch 26, batch 21050, giga_loss[loss=0.264, simple_loss=0.3342, pruned_loss=0.0969, over 28790.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3487, pruned_loss=0.09754, over 5705716.14 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3377, pruned_loss=0.08609, over 5738290.00 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3502, pruned_loss=0.09908, over 5705383.51 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:05:30,400 INFO [optim.py:369] (1/2) 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 12:05:39,282 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-13 12:05:39,524 INFO [train.py:968] (1/2) Epoch 26, batch 21100, giga_loss[loss=0.2614, simple_loss=0.3325, pruned_loss=0.09511, over 28253.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.347, pruned_loss=0.09682, over 5707011.07 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3379, pruned_loss=0.08604, over 5740174.13 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3482, pruned_loss=0.09824, over 5704567.58 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:05:43,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6187, 1.6924, 1.8115, 1.4124], device='cuda:1'), covar=tensor([0.1771, 0.2542, 0.1451, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0709, 0.0972, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 12:06:15,956 INFO [train.py:968] (1/2) Epoch 26, batch 21150, giga_loss[loss=0.2766, simple_loss=0.3549, pruned_loss=0.09918, over 28997.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.09616, over 5710482.56 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3385, pruned_loss=0.0862, over 5740712.12 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3465, pruned_loss=0.09759, over 5706360.67 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:06:49,175 INFO [optim.py:369] (1/2) 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 12:06:56,550 INFO [train.py:968] (1/2) Epoch 26, batch 21200, giga_loss[loss=0.2959, simple_loss=0.3676, pruned_loss=0.1121, over 27933.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.09821, over 5711447.95 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3386, pruned_loss=0.08629, over 5743387.57 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3485, pruned_loss=0.09942, over 5705212.81 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:07:07,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1160266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:07:13,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4037, 4.2468, 4.0160, 1.7467], device='cuda:1'), covar=tensor([0.0600, 0.0754, 0.0784, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.1259, 0.1161, 0.0980, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 12:07:19,313 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 12:07:28,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1160293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:07:40,213 INFO [train.py:968] (1/2) Epoch 26, batch 21250, giga_loss[loss=0.2564, simple_loss=0.3437, pruned_loss=0.08452, over 28706.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3482, pruned_loss=0.09811, over 5710967.29 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3386, pruned_loss=0.08638, over 5743194.63 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3486, pruned_loss=0.09903, over 5706135.57 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:07:51,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:07:55,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4196, 1.5053, 1.5374, 1.3601], device='cuda:1'), covar=tensor([0.2931, 0.3026, 0.2125, 0.2598], device='cuda:1'), in_proj_covar=tensor([0.2027, 0.1975, 0.1892, 0.2032], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 12:08:07,851 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1160341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:08:12,370 INFO [optim.py:369] (1/2) 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 12:08:19,135 INFO [train.py:968] (1/2) Epoch 26, batch 21300, giga_loss[loss=0.2834, simple_loss=0.3661, pruned_loss=0.1004, over 28557.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3475, pruned_loss=0.09698, over 5711301.95 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3393, pruned_loss=0.08686, over 5744869.66 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3475, pruned_loss=0.09769, over 5704422.80 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:09:00,158 INFO [train.py:968] (1/2) Epoch 26, batch 21350, giga_loss[loss=0.2761, simple_loss=0.3532, pruned_loss=0.09957, over 28899.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3455, pruned_loss=0.09546, over 5720315.44 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3388, pruned_loss=0.08658, over 5746166.83 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3462, pruned_loss=0.09645, over 5713164.65 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:09:12,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2683, 1.4121, 1.4902, 1.0837], device='cuda:1'), covar=tensor([0.1644, 0.2900, 0.1340, 0.1510], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0712, 0.0975, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:1')
+2023-03-13 12:09:34,176 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 21400, giga_loss[loss=0.2582, simple_loss=0.3336, pruned_loss=0.09137, over 28872.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3439, pruned_loss=0.09468, over 5726352.66 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3388, pruned_loss=0.08661, over 5746386.08 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3445, pruned_loss=0.09552, over 5720335.95 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:09:48,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8825, 1.0974, 1.0382, 0.8522], device='cuda:1'), covar=tensor([0.2857, 0.3016, 0.1797, 0.2701], device='cuda:1'), in_proj_covar=tensor([0.2029, 0.1979, 0.1891, 0.2034], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 12:09:48,561 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:09:50,526 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:10:12,073 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:10:21,654 INFO [train.py:968] (1/2) Epoch 26, batch 21450, giga_loss[loss=0.2739, simple_loss=0.3464, pruned_loss=0.1007, over 28582.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3401, pruned_loss=0.09271, over 5717417.09 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3388, pruned_loss=0.08656, over 5747458.52 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3407, pruned_loss=0.09351, over 5711475.68 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:10:46,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.17 vs. limit=5.0
+2023-03-13 12:10:52,295 INFO [optim.py:369] (1/2) 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 12:10:59,799 INFO [train.py:968] (1/2) Epoch 26, batch 21500, giga_loss[loss=0.2499, simple_loss=0.3257, pruned_loss=0.08705, over 28960.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3387, pruned_loss=0.09188, over 5716181.92 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3393, pruned_loss=0.08681, over 5741140.60 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3386, pruned_loss=0.09244, over 5716496.23 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:11:16,109 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4047, 3.3035, 1.5790, 1.5525], device='cuda:1'), covar=tensor([0.1004, 0.0346, 0.0905, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0555, 0.0398, 0.0435], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0033, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 12:11:19,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0535, 3.8824, 3.6762, 1.8671], device='cuda:1'), covar=tensor([0.0696, 0.0856, 0.0808, 0.2210], device='cuda:1'), in_proj_covar=tensor([0.1262, 0.1164, 0.0981, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 12:11:36,484 INFO [train.py:968] (1/2) Epoch 26, batch 21550, giga_loss[loss=0.3403, simple_loss=0.4011, pruned_loss=0.1397, over 28574.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.339, pruned_loss=0.09259, over 5725698.41 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3392, pruned_loss=0.08705, over 5747324.59 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.339, pruned_loss=0.093, over 5719547.91 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:12:05,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:12:09,639 INFO [optim.py:369] (1/2) 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 12:12:18,390 INFO [train.py:968] (1/2) Epoch 26, batch 21600, giga_loss[loss=0.2862, simple_loss=0.3476, pruned_loss=0.1123, over 26705.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3382, pruned_loss=0.09299, over 5711592.60 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3392, pruned_loss=0.08709, over 5739509.12 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3382, pruned_loss=0.09339, over 5713874.16 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:12:27,033 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:12:55,353 INFO [train.py:968] (1/2) Epoch 26, batch 21650, giga_loss[loss=0.2144, simple_loss=0.2949, pruned_loss=0.06693, over 28832.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3375, pruned_loss=0.0931, over 5705313.04 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3398, pruned_loss=0.08774, over 5733424.57 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3369, pruned_loss=0.09308, over 5711028.54 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:13:03,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1160716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:13:28,227 INFO [optim.py:369] (1/2) 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 12:13:34,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6091, 1.6941, 1.8275, 1.4105], device='cuda:1'), covar=tensor([0.1928, 0.2567, 0.1590, 0.1825], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0709, 0.0972, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 12:13:36,281 INFO [train.py:968] (1/2) Epoch 26, batch 21700, giga_loss[loss=0.2492, simple_loss=0.322, pruned_loss=0.08823, over 28895.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3349, pruned_loss=0.09205, over 5709244.63 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3398, pruned_loss=0.08778, over 5735076.06 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3344, pruned_loss=0.09205, over 5711995.10 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:13:50,287 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-13 12:13:58,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160784.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:14:00,917 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160787.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:14:02,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1339, 2.4092, 2.3025, 1.8305], device='cuda:1'), covar=tensor([0.3605, 0.2361, 0.2624, 0.3159], device='cuda:1'), in_proj_covar=tensor([0.2039, 0.1988, 0.1903, 0.2043], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 12:14:18,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3358, 2.5593, 2.5253, 2.0208], device='cuda:1'), covar=tensor([0.3802, 0.2594, 0.2571, 0.3278], device='cuda:1'), in_proj_covar=tensor([0.2040, 0.1988, 0.1904, 0.2044], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 12:14:18,560 INFO [train.py:968] (1/2) Epoch 26, batch 21750, giga_loss[loss=0.2013, simple_loss=0.2877, pruned_loss=0.05743, over 28380.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3325, pruned_loss=0.09133, over 5698997.92 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3401, pruned_loss=0.08802, over 5726881.79 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3319, pruned_loss=0.09115, over 5707555.89 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:14:23,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:14:25,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:14:26,220 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160816.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:14:34,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6425, 1.8022, 1.5407, 1.8961], device='cuda:1'), covar=tensor([0.2808, 0.3008, 0.3339, 0.2580], device='cuda:1'), in_proj_covar=tensor([0.1567, 0.1130, 0.1383, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 12:14:46,611 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:14:59,382 INFO [train.py:968] (1/2) Epoch 26, batch 21800, giga_loss[loss=0.2523, simple_loss=0.3264, pruned_loss=0.08914, over 28779.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3322, pruned_loss=0.09129, over 5696200.57 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3405, pruned_loss=0.08841, over 5726558.36 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3311, pruned_loss=0.09086, over 5702558.16 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:15:01,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1160859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:15:03,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1160862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:15:24,553 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1160891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:15:31,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2185, 3.3844, 2.1994, 1.5424], device='cuda:1'), covar=tensor([0.8820, 0.2737, 0.4393, 0.6802], device='cuda:1'), in_proj_covar=tensor([0.1807, 0.1692, 0.1637, 0.1474], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 12:15:37,478 INFO [train.py:968] (1/2) Epoch 26, batch 21850, giga_loss[loss=0.3027, simple_loss=0.3696, pruned_loss=0.1179, over 27571.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3327, pruned_loss=0.09096, over 5700810.26 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3412, pruned_loss=0.08895, over 5728223.09 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3309, pruned_loss=0.09021, over 5703347.05 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:16:13,332 INFO [optim.py:369] (1/2) 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 12:16:21,541 INFO [train.py:968] (1/2) Epoch 26, batch 21900, giga_loss[loss=0.2604, simple_loss=0.3281, pruned_loss=0.09632, over 28805.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3359, pruned_loss=0.09233, over 5696925.25 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3414, pruned_loss=0.08914, over 5721546.21 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3341, pruned_loss=0.0916, over 5704513.43 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:16:57,605 INFO [train.py:968] (1/2) Epoch 26, batch 21950, giga_loss[loss=0.2594, simple_loss=0.3447, pruned_loss=0.08702, over 28675.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3371, pruned_loss=0.09223, over 5705028.48 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3415, pruned_loss=0.08958, over 5727533.43 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3353, pruned_loss=0.09141, over 5704802.95 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:17:35,797 INFO [optim.py:369] (1/2) 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 12:17:42,458 INFO [train.py:968] (1/2) Epoch 26, batch 22000, giga_loss[loss=0.2343, simple_loss=0.3236, pruned_loss=0.07255, over 28991.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3397, pruned_loss=0.0931, over 5699588.76 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3416, pruned_loss=0.08966, over 5728979.38 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3381, pruned_loss=0.09241, over 5697673.13 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:18:24,598 INFO [train.py:968] (1/2) Epoch 26, batch 22050, giga_loss[loss=0.2814, simple_loss=0.3631, pruned_loss=0.09988, over 28719.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3396, pruned_loss=0.09283, over 5695011.32 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.342, pruned_loss=0.09014, over 5725668.76 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3378, pruned_loss=0.09191, over 5695204.89 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:18:27,916 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.39 vs. limit=2.0
+2023-03-13 12:18:34,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3646, 1.4531, 1.3707, 1.4815], device='cuda:1'), covar=tensor([0.0744, 0.0347, 0.0339, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 12:19:01,725 INFO [optim.py:369] (1/2) 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 12:19:07,822 INFO [train.py:968] (1/2) Epoch 26, batch 22100, giga_loss[loss=0.2298, simple_loss=0.314, pruned_loss=0.0728, over 28823.00 frames. ], tot_loss[loss=0.262, simple_loss=0.339, pruned_loss=0.09253, over 5702838.79 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3424, pruned_loss=0.09047, over 5729857.28 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3372, pruned_loss=0.09154, over 5698450.20 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:19:46,989 INFO [train.py:968] (1/2) Epoch 26, batch 22150, giga_loss[loss=0.2649, simple_loss=0.3431, pruned_loss=0.09331, over 29001.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3408, pruned_loss=0.09388, over 5696710.94 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.343, pruned_loss=0.09093, over 5723604.01 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09275, over 5698196.04 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:20:14,925 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:20:23,155 INFO [optim.py:369] (1/2) 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 12:20:30,066 INFO [train.py:968] (1/2) Epoch 26, batch 22200, giga_loss[loss=0.2775, simple_loss=0.3467, pruned_loss=0.1042, over 28388.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3422, pruned_loss=0.09484, over 5700841.77 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.343, pruned_loss=0.09096, over 5723558.99 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09394, over 5701923.48 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:21:11,298 INFO [train.py:968] (1/2) Epoch 26, batch 22250, giga_loss[loss=0.2825, simple_loss=0.3573, pruned_loss=0.1039, over 28703.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3439, pruned_loss=0.09582, over 5698413.31 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3427, pruned_loss=0.09094, over 5727049.03 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3428, pruned_loss=0.0952, over 5695587.57 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:21:46,272 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 22300, giga_loss[loss=0.2476, simple_loss=0.328, pruned_loss=0.08358, over 29039.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3462, pruned_loss=0.09675, over 5709492.24 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3429, pruned_loss=0.09125, over 5731736.04 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3452, pruned_loss=0.09607, over 5702370.88 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:21:53,341 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 12:22:31,792 INFO [train.py:968] (1/2) Epoch 26, batch 22350, giga_loss[loss=0.2672, simple_loss=0.3451, pruned_loss=0.09462, over 28987.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3489, pruned_loss=0.09852, over 5711653.00 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3434, pruned_loss=0.09163, over 5731210.26 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.09776, over 5706175.93 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:22:32,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 12:22:32,911 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-13 12:22:52,009 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-13 12:23:04,630 INFO [optim.py:369] (1/2) 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 12:23:07,457 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:23:12,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8152, 1.1153, 2.8720, 2.7609], device='cuda:1'), covar=tensor([0.2159, 0.3008, 0.1106, 0.1254], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0667, 0.0985, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 12:23:13,013 INFO [train.py:968] (1/2) Epoch 26, batch 22400, giga_loss[loss=0.2816, simple_loss=0.3593, pruned_loss=0.102, over 28992.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3489, pruned_loss=0.09808, over 5716714.80 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3438, pruned_loss=0.09182, over 5734096.55 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3478, pruned_loss=0.09746, over 5709353.31 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:23:52,509 INFO [train.py:968] (1/2) Epoch 26, batch 22450, giga_loss[loss=0.2801, simple_loss=0.3551, pruned_loss=0.1025, over 28895.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.351, pruned_loss=0.09977, over 5722578.13 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.345, pruned_loss=0.09264, over 5739250.41 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.09875, over 5711399.66 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:24:27,636 INFO [optim.py:369] (1/2) 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 12:24:29,129 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:24:34,053 INFO [train.py:968] (1/2) Epoch 26, batch 22500, libri_loss[loss=0.2563, simple_loss=0.3312, pruned_loss=0.09073, over 29504.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3497, pruned_loss=0.09935, over 5721370.52 frames. ], libri_tot_loss[loss=0.2658, simple_loss=0.3452, pruned_loss=0.09318, over 5743836.24 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3484, pruned_loss=0.09835, over 5706877.51 frames. ], batch size: 81, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:25:14,124 INFO [train.py:968] (1/2) Epoch 26, batch 22550, giga_loss[loss=0.2396, simple_loss=0.3112, pruned_loss=0.08406, over 28839.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3473, pruned_loss=0.09817, over 5716294.04 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3459, pruned_loss=0.09389, over 5738542.56 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3456, pruned_loss=0.09687, over 5708889.13 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:25:21,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1161614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:25:29,600 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9287, 2.2827, 1.8661, 2.1617], device='cuda:1'), covar=tensor([0.2452, 0.2533, 0.2924, 0.2340], device='cuda:1'), in_proj_covar=tensor([0.1564, 0.1126, 0.1380, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 12:25:49,828 INFO [optim.py:369] (1/2) 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 12:25:56,002 INFO [train.py:968] (1/2) Epoch 26, batch 22600, giga_loss[loss=0.2828, simple_loss=0.3409, pruned_loss=0.1124, over 23828.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3435, pruned_loss=0.09653, over 5711375.23 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3465, pruned_loss=0.09445, over 5741235.48 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3416, pruned_loss=0.09503, over 5702831.93 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:26:21,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-13 12:26:36,948 INFO [train.py:968] (1/2) Epoch 26, batch 22650, giga_loss[loss=0.2852, simple_loss=0.3671, pruned_loss=0.1016, over 28670.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3426, pruned_loss=0.09562, over 5715099.63 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3472, pruned_loss=0.09524, over 5744511.45 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3404, pruned_loss=0.09374, over 5704342.05 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:27:09,215 INFO [optim.py:369] (1/2) 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 12:27:14,423 INFO [train.py:968] (1/2) Epoch 26, batch 22700, giga_loss[loss=0.2587, simple_loss=0.3504, pruned_loss=0.08354, over 28726.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3446, pruned_loss=0.09549, over 5696164.10 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3479, pruned_loss=0.0961, over 5729759.65 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3419, pruned_loss=0.09313, over 5699731.47 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:27:15,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1161757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:27:18,904 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1161760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:27:43,083 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1161789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:27:58,324 INFO [train.py:968] (1/2) Epoch 26, batch 22750, giga_loss[loss=0.2627, simple_loss=0.3289, pruned_loss=0.09823, over 28154.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3462, pruned_loss=0.09623, over 5696363.36 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3483, pruned_loss=0.0966, over 5732297.22 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3437, pruned_loss=0.09391, over 5696394.27 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:28:13,784 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1161825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:28:18,257 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161832.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:28:22,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6002, 1.7570, 1.7760, 1.3958], device='cuda:1'), covar=tensor([0.1823, 0.2509, 0.1547, 0.1722], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0708, 0.0971, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 12:28:22,590 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:28:33,612 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 22800, giga_loss[loss=0.2817, simple_loss=0.3525, pruned_loss=0.1054, over 28590.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3429, pruned_loss=0.09545, over 5695149.68 frames. ], libri_tot_loss[loss=0.2706, simple_loss=0.348, pruned_loss=0.09657, over 5731834.79 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3411, pruned_loss=0.09365, over 5695320.57 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:28:40,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3236, 1.6114, 1.2973, 1.4278], device='cuda:1'), covar=tensor([0.0730, 0.0393, 0.0364, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 12:29:01,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9622, 3.0666, 2.0820, 1.2051], device='cuda:1'), covar=tensor([0.9884, 0.3418, 0.4314, 0.8326], device='cuda:1'), in_proj_covar=tensor([0.1811, 0.1699, 0.1643, 0.1478], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 12:29:04,478 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1161888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:29:19,420 INFO [train.py:968] (1/2) Epoch 26, batch 22850, giga_loss[loss=0.2553, simple_loss=0.3301, pruned_loss=0.09018, over 29062.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3423, pruned_loss=0.09641, over 5703228.47 frames. ], libri_tot_loss[loss=0.271, simple_loss=0.3483, pruned_loss=0.09683, over 5734496.65 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3405, pruned_loss=0.09474, over 5700500.94 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:29:36,523 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:29:58,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3061, 4.1502, 3.9531, 1.7606], device='cuda:1'), covar=tensor([0.0746, 0.0884, 0.0783, 0.2206], device='cuda:1'), in_proj_covar=tensor([0.1267, 0.1168, 0.0987, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 12:30:03,253 INFO [train.py:968] (1/2) Epoch 26, batch 22900, giga_loss[loss=0.2503, simple_loss=0.3181, pruned_loss=0.09123, over 28774.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3397, pruned_loss=0.09606, over 5705495.27 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3483, pruned_loss=0.09698, over 5728189.18 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3382, pruned_loss=0.09461, over 5708769.15 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:30:10,540 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1161968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:30:12,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1161971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:30:21,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3829, 1.4131, 1.2528, 1.5983], device='cuda:1'), covar=tensor([0.0742, 0.0358, 0.0351, 0.0846], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0102, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 12:30:37,623 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:30:41,880 INFO [train.py:968] (1/2) Epoch 26, batch 22950, giga_loss[loss=0.2571, simple_loss=0.336, pruned_loss=0.08908, over 28878.00 frames. ], tot_loss[loss=0.266, simple_loss=0.339, pruned_loss=0.09645, over 5705238.41 frames. ], libri_tot_loss[loss=0.2716, simple_loss=0.3486, pruned_loss=0.09728, over 5731785.37 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3374, pruned_loss=0.095, over 5704048.57 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:30:58,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8529, 2.9137, 1.7997, 1.1234], device='cuda:1'), covar=tensor([0.9955, 0.3321, 0.4912, 0.8206], device='cuda:1'), in_proj_covar=tensor([0.1805, 0.1691, 0.1637, 0.1473], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 12:31:16,662 INFO [optim.py:369] (1/2) 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 12:31:22,114 INFO [train.py:968] (1/2) Epoch 26, batch 23000, giga_loss[loss=0.3021, simple_loss=0.3712, pruned_loss=0.1165, over 28538.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3386, pruned_loss=0.09591, over 5716298.20 frames. ], libri_tot_loss[loss=0.2726, simple_loss=0.3494, pruned_loss=0.09794, over 5735563.10 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3364, pruned_loss=0.09416, over 5711478.09 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:31:32,363 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:31:35,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:31:58,353 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:32:02,791 INFO [train.py:968] (1/2) Epoch 26, batch 23050, giga_loss[loss=0.2068, simple_loss=0.2905, pruned_loss=0.0616, over 28992.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3345, pruned_loss=0.09394, over 5706693.52 frames. ], libri_tot_loss[loss=0.2733, simple_loss=0.3498, pruned_loss=0.09835, over 5729136.04 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3321, pruned_loss=0.09212, over 5708417.73 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:32:14,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4271, 1.6493, 1.6865, 1.2849], device='cuda:1'), covar=tensor([0.1944, 0.2528, 0.1573, 0.1710], device='cuda:1'), in_proj_covar=tensor([0.0926, 0.0709, 0.0971, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 12:32:38,311 INFO [optim.py:369] (1/2) 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 12:32:42,421 INFO [train.py:968] (1/2) Epoch 26, batch 23100, libri_loss[loss=0.3004, simple_loss=0.374, pruned_loss=0.1134, over 29370.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3314, pruned_loss=0.09264, over 5689587.48 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3508, pruned_loss=0.09927, over 5711702.00 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3278, pruned_loss=0.09008, over 5706298.03 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:32:46,954 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1162162.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:33:21,448 INFO [train.py:968] (1/2) Epoch 26, batch 23150, libri_loss[loss=0.249, simple_loss=0.3194, pruned_loss=0.08934, over 29469.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3305, pruned_loss=0.09188, over 5696086.05 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3512, pruned_loss=0.09976, over 5713621.52 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3269, pruned_loss=0.0892, over 5707221.25 frames. ], batch size: 70, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:33:21,694 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162207.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:33:25,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6899, 1.7733, 1.8752, 1.4788], device='cuda:1'), covar=tensor([0.1987, 0.2649, 0.1633, 0.1869], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0707, 0.0969, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 12:33:25,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:33:29,693 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1162219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:33:56,154 INFO [optim.py:369] (1/2) 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 12:34:01,317 INFO [train.py:968] (1/2) Epoch 26, batch 23200, giga_loss[loss=0.2786, simple_loss=0.3582, pruned_loss=0.09954, over 29144.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.332, pruned_loss=0.09211, over 5701521.46 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.3508, pruned_loss=0.09968, over 5716093.52 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3288, pruned_loss=0.08976, over 5707987.43 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:34:05,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:34:38,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4295, 1.6790, 1.3577, 1.5160], device='cuda:1'), covar=tensor([0.2720, 0.2899, 0.3312, 0.2496], device='cuda:1'), in_proj_covar=tensor([0.1572, 0.1132, 0.1386, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 12:34:40,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3234, 3.8054, 1.5395, 1.4712], device='cuda:1'), covar=tensor([0.1009, 0.0306, 0.0964, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0560, 0.0400, 0.0437], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 12:34:43,597 INFO [train.py:968] (1/2) Epoch 26, batch 23250, libri_loss[loss=0.3188, simple_loss=0.3914, pruned_loss=0.1231, over 29230.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3361, pruned_loss=0.09404, over 5704720.23 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3513, pruned_loss=0.1002, over 5716710.01 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3325, pruned_loss=0.09151, over 5709191.16 frames. ], batch size: 97, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:34:43,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9023, 1.1504, 1.3626, 0.9520], device='cuda:1'), covar=tensor([0.2165, 0.1642, 0.2480, 0.1945], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0757, 0.0725, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 12:35:20,098 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162350.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:35:20,446 INFO [optim.py:369] (1/2) 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 12:35:23,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162353.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:35:25,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:35:25,903 INFO [train.py:968] (1/2) Epoch 26, batch 23300, giga_loss[loss=0.2673, simple_loss=0.3401, pruned_loss=0.09718, over 28802.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3398, pruned_loss=0.09556, over 5706237.59 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.351, pruned_loss=0.1001, over 5719539.63 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3371, pruned_loss=0.09358, over 5706831.01 frames. ], batch size: 119, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:35:27,507 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:35:47,014 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162382.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:35:52,057 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:36:07,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 23350, giga_loss[loss=0.2937, simple_loss=0.3696, pruned_loss=0.109, over 28261.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3435, pruned_loss=0.09741, over 5704840.92 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3509, pruned_loss=0.1001, over 5721417.24 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3413, pruned_loss=0.09577, over 5703362.88 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:36:10,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1162409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:36:29,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5038, 1.7890, 1.4302, 1.5963], device='cuda:1'), covar=tensor([0.2699, 0.2686, 0.3032, 0.2570], device='cuda:1'), in_proj_covar=tensor([0.1571, 0.1132, 0.1386, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 12:36:35,415 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:36:45,816 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 23400, giga_loss[loss=0.2239, simple_loss=0.3084, pruned_loss=0.06964, over 29028.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3458, pruned_loss=0.09875, over 5700988.50 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3513, pruned_loss=0.1006, over 5728369.76 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3434, pruned_loss=0.09684, over 5692540.72 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:37:34,962 INFO [train.py:968] (1/2) Epoch 26, batch 23450, giga_loss[loss=0.3072, simple_loss=0.374, pruned_loss=0.1203, over 28122.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3511, pruned_loss=0.1033, over 5699475.91 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3514, pruned_loss=0.1008, over 5732013.08 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.349, pruned_loss=0.1017, over 5688653.91 frames. ], batch size: 77, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:37:57,198 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-13 12:38:04,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162537.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:38:18,971 INFO [optim.py:369] (1/2) 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 12:38:24,312 INFO [train.py:968] (1/2) Epoch 26, batch 23500, giga_loss[loss=0.2943, simple_loss=0.3627, pruned_loss=0.1129, over 28867.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3562, pruned_loss=0.1076, over 5681701.26 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3519, pruned_loss=0.1013, over 5720869.55 frames. ], giga_tot_loss[loss=0.2829, simple_loss=0.3541, pruned_loss=0.1059, over 5682488.46 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:38:44,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 12:38:53,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3269, 1.4868, 1.3382, 1.5373], device='cuda:1'), covar=tensor([0.0729, 0.0438, 0.0353, 0.0810], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 12:39:01,253 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1162594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:39:15,332 INFO [train.py:968] (1/2) Epoch 26, batch 23550, giga_loss[loss=0.2899, simple_loss=0.3617, pruned_loss=0.109, over 28611.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3626, pruned_loss=0.1122, over 5681045.41 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1012, over 5723646.33 frames. ], giga_tot_loss[loss=0.2917, simple_loss=0.3613, pruned_loss=0.111, over 5678826.58 frames. ], batch size: 92, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:39:47,008 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-13 12:39:58,683 INFO [optim.py:369] (1/2) 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 12:40:03,194 INFO [train.py:968] (1/2) Epoch 26, batch 23600, giga_loss[loss=0.3513, simple_loss=0.4045, pruned_loss=0.149, over 28352.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3683, pruned_loss=0.1171, over 5678455.17 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3515, pruned_loss=0.1012, over 5722334.65 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3675, pruned_loss=0.1164, over 5677156.08 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:40:27,286 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162680.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 12:40:29,784 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 23650, giga_loss[loss=0.3796, simple_loss=0.4267, pruned_loss=0.1662, over 27887.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3749, pruned_loss=0.1225, over 5659137.62 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3517, pruned_loss=0.1014, over 5715330.32 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3744, pruned_loss=0.1221, over 5662677.04 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:41:02,057 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162712.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 12:41:26,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1162737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:41:30,042 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 26, batch 23700, giga_loss[loss=0.3291, simple_loss=0.3912, pruned_loss=0.1335, over 28575.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3784, pruned_loss=0.1252, over 5663223.12 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3517, pruned_loss=0.1014, over 5718018.89 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3785, pruned_loss=0.1253, over 5662447.31 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:41:55,731 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1162769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:41:56,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.93 vs. limit=5.0
+2023-03-13 12:42:31,014 INFO [train.py:968] (1/2) Epoch 26, batch 23750, giga_loss[loss=0.3105, simple_loss=0.3898, pruned_loss=0.1155, over 28878.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3803, pruned_loss=0.1278, over 5662829.77 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3517, pruned_loss=0.1016, over 5723502.40 frames. ], giga_tot_loss[loss=0.319, simple_loss=0.3812, pruned_loss=0.1284, over 5655854.56 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:43:13,858 INFO [optim.py:369] (1/2) 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 12:43:16,741 INFO [train.py:968] (1/2) Epoch 26, batch 23800, giga_loss[loss=0.2989, simple_loss=0.3673, pruned_loss=0.1152, over 28958.00 frames. ], tot_loss[loss=0.3213, simple_loss=0.3819, pruned_loss=0.1303, over 5645366.58 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3524, pruned_loss=0.1023, over 5716740.32 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.383, pruned_loss=0.1311, over 5643489.70 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:43:57,026 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-13 12:44:04,619 INFO [train.py:968] (1/2) Epoch 26, batch 23850, giga_loss[loss=0.3228, simple_loss=0.3819, pruned_loss=0.1318, over 28847.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3841, pruned_loss=0.1326, over 5643329.77 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3524, pruned_loss=0.1024, over 5716506.08 frames. ], giga_tot_loss[loss=0.3269, simple_loss=0.3859, pruned_loss=0.1339, over 5640048.87 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:44:10,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.35 vs. limit=5.0
+2023-03-13 12:44:44,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7690, 1.7024, 1.9434, 1.5066], device='cuda:1'), covar=tensor([0.1684, 0.2438, 0.1382, 0.1748], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0707, 0.0966, 0.0866], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 12:44:54,349 INFO [optim.py:369] (1/2) 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 12:44:58,033 INFO [train.py:968] (1/2) Epoch 26, batch 23900, giga_loss[loss=0.3051, simple_loss=0.3681, pruned_loss=0.121, over 28824.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3862, pruned_loss=0.1345, over 5630691.38 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3525, pruned_loss=0.1028, over 5710558.14 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3892, pruned_loss=0.1369, over 5629064.51 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:45:24,573 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 23950, libri_loss[loss=0.2867, simple_loss=0.3476, pruned_loss=0.1129, over 29309.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3874, pruned_loss=0.1372, over 5609382.52 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3524, pruned_loss=0.1028, over 5713377.81 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3905, pruned_loss=0.1396, over 5603865.12 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:46:24,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 26, batch 24000, giga_loss[loss=0.346, simple_loss=0.3947, pruned_loss=0.1487, over 27557.00 frames. ], tot_loss[loss=0.3287, simple_loss=0.3853, pruned_loss=0.1361, over 5626154.62 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3529, pruned_loss=0.1032, over 5715377.17 frames. ], giga_tot_loss[loss=0.3326, simple_loss=0.3882, pruned_loss=0.1385, over 5617219.11 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 12:46:40,581 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 12:46:48,698 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 12:47:02,126 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:47:32,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4593, 1.6558, 1.4672, 1.6475], device='cuda:1'), covar=tensor([0.0639, 0.0295, 0.0286, 0.0655], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 12:47:33,113 INFO [train.py:968] (1/2) Epoch 26, batch 24050, giga_loss[loss=0.3881, simple_loss=0.4094, pruned_loss=0.1834, over 23610.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3837, pruned_loss=0.1345, over 5633523.85 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.353, pruned_loss=0.1033, over 5717700.49 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3868, pruned_loss=0.1373, over 5622120.54 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:47:57,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3321, 1.5909, 1.3743, 1.5529], device='cuda:1'), covar=tensor([0.0754, 0.0382, 0.0336, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 12:48:22,168 INFO [optim.py:369] (1/2) 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 12:48:25,660 INFO [train.py:968] (1/2) Epoch 26, batch 24100, giga_loss[loss=0.3296, simple_loss=0.403, pruned_loss=0.1282, over 29030.00 frames. ], tot_loss[loss=0.3246, simple_loss=0.3833, pruned_loss=0.133, over 5625960.39 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3529, pruned_loss=0.1033, over 5717995.30 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3863, pruned_loss=0.1356, over 5615497.00 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:48:30,926 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5367, 1.7915, 1.6844, 1.6149], device='cuda:1'), covar=tensor([0.2297, 0.2248, 0.2408, 0.2170], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0764, 0.0730, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 12:49:03,201 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 24150, giga_loss[loss=0.3349, simple_loss=0.3892, pruned_loss=0.1403, over 27535.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3859, pruned_loss=0.1347, over 5617556.55 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.353, pruned_loss=0.1036, over 5711767.98 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3889, pruned_loss=0.1371, over 5613729.53 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:49:59,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1615, 5.0140, 4.7938, 2.4098], device='cuda:1'), covar=tensor([0.0517, 0.0645, 0.0699, 0.1776], device='cuda:1'), in_proj_covar=tensor([0.1284, 0.1183, 0.1002, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 12:50:06,582 INFO [optim.py:369] (1/2) 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 12:50:08,191 INFO [train.py:968] (1/2) Epoch 26, batch 24200, giga_loss[loss=0.2734, simple_loss=0.3497, pruned_loss=0.09851, over 28915.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3842, pruned_loss=0.1328, over 5627961.48 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3532, pruned_loss=0.1038, over 5714777.10 frames. ], giga_tot_loss[loss=0.3286, simple_loss=0.3869, pruned_loss=0.1351, over 5620661.18 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:50:59,206 INFO [train.py:968] (1/2) Epoch 26, batch 24250, giga_loss[loss=0.314, simple_loss=0.3793, pruned_loss=0.1243, over 28899.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3809, pruned_loss=0.1294, over 5630254.43 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3527, pruned_loss=0.1037, over 5717336.68 frames. ], giga_tot_loss[loss=0.3237, simple_loss=0.384, pruned_loss=0.1317, over 5620647.18 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:51:45,684 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:51:47,377 INFO [train.py:968] (1/2) Epoch 26, batch 24300, giga_loss[loss=0.279, simple_loss=0.3555, pruned_loss=0.1013, over 28898.00 frames. ], tot_loss[loss=0.3172, simple_loss=0.3794, pruned_loss=0.1275, over 5640079.99 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.353, pruned_loss=0.1043, over 5721344.45 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3823, pruned_loss=0.1296, over 5626747.16 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:51:58,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:52:03,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4950, 1.8614, 1.9125, 1.4393], device='cuda:1'), covar=tensor([0.3543, 0.2754, 0.2825, 0.3170], device='cuda:1'), in_proj_covar=tensor([0.2046, 0.1990, 0.1921, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 12:52:05,332 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2893, 1.3804, 1.2940, 1.5352], device='cuda:1'), covar=tensor([0.0725, 0.0404, 0.0343, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 12:52:17,494 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:52:37,559 INFO [train.py:968] (1/2) Epoch 26, batch 24350, giga_loss[loss=0.3052, simple_loss=0.3561, pruned_loss=0.1272, over 23819.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3755, pruned_loss=0.1244, over 5635088.27 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3531, pruned_loss=0.1043, over 5724150.08 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3781, pruned_loss=0.1264, over 5621015.67 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 12:52:44,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:53:17,290 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:53:24,058 INFO [optim.py:369] (1/2) 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 12:53:27,391 INFO [train.py:968] (1/2) Epoch 26, batch 24400, giga_loss[loss=0.2837, simple_loss=0.3569, pruned_loss=0.1053, over 28577.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3728, pruned_loss=0.1226, over 5638299.65 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3535, pruned_loss=0.1046, over 5723474.22 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3749, pruned_loss=0.1242, over 5626422.12 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:54:05,840 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:54:07,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:54:14,798 INFO [train.py:968] (1/2) Epoch 26, batch 24450, giga_loss[loss=0.2745, simple_loss=0.3532, pruned_loss=0.09791, over 29082.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3747, pruned_loss=0.1245, over 5638721.96 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3545, pruned_loss=0.1055, over 5724149.80 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.376, pruned_loss=0.1254, over 5626267.69 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:54:40,592 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:54:45,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8010, 2.0103, 1.4271, 1.6458], device='cuda:1'), covar=tensor([0.1046, 0.0696, 0.1028, 0.1252], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0454, 0.0525, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 12:55:08,601 INFO [optim.py:369] (1/2) 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 12:55:10,952 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:11,224 INFO [train.py:968] (1/2) Epoch 26, batch 24500, giga_loss[loss=0.2868, simple_loss=0.3567, pruned_loss=0.1085, over 28528.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3737, pruned_loss=0.1235, over 5639401.49 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3544, pruned_loss=0.1056, over 5722477.43 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3751, pruned_loss=0.1245, over 5629500.00 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:55:12,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:20,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:40,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:42,462 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:45,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:55:55,522 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 12:55:59,390 INFO [train.py:968] (1/2) Epoch 26, batch 24550, giga_loss[loss=0.3208, simple_loss=0.3637, pruned_loss=0.139, over 23454.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1217, over 5647059.96 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3547, pruned_loss=0.106, over 5718454.82 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3735, pruned_loss=0.1226, over 5640783.23 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:56:09,478 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 12:56:13,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:56:32,518 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:56:48,792 INFO [train.py:968] (1/2) Epoch 26, batch 24600, giga_loss[loss=0.3117, simple_loss=0.3886, pruned_loss=0.1174, over 28901.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3721, pruned_loss=0.1191, over 5650789.17 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3544, pruned_loss=0.1059, over 5713717.70 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3739, pruned_loss=0.1203, over 5648398.23 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:57:04,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1163671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:57:17,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5300, 1.9349, 1.6789, 1.4841], device='cuda:1'), covar=tensor([0.2539, 0.2578, 0.2649, 0.2867], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0761, 0.0729, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 12:57:39,752 INFO [train.py:968] (1/2) Epoch 26, batch 24650, giga_loss[loss=0.3276, simple_loss=0.383, pruned_loss=0.1361, over 27578.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.373, pruned_loss=0.1187, over 5653675.11 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3548, pruned_loss=0.1062, over 5713650.24 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3746, pruned_loss=0.1196, over 5649894.67 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:57:44,630 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:57:47,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:58:15,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:58:17,520 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:58:29,951 INFO [optim.py:369] (1/2) 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 12:58:31,523 INFO [train.py:968] (1/2) Epoch 26, batch 24700, giga_loss[loss=0.3007, simple_loss=0.3691, pruned_loss=0.1162, over 28465.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.373, pruned_loss=0.1191, over 5659168.10 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3546, pruned_loss=0.1061, over 5714664.99 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3745, pruned_loss=0.12, over 5655020.67 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 12:58:36,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1163761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 12:59:20,193 INFO [train.py:968] (1/2) Epoch 26, batch 24750, giga_loss[loss=0.2803, simple_loss=0.3587, pruned_loss=0.1009, over 28945.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3723, pruned_loss=0.1187, over 5676692.42 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3547, pruned_loss=0.1063, over 5717170.72 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3737, pruned_loss=0.1194, over 5670437.29 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:00:08,850 INFO [optim.py:369] (1/2) 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 13:00:10,100 INFO [train.py:968] (1/2) Epoch 26, batch 24800, giga_loss[loss=0.3358, simple_loss=0.3864, pruned_loss=0.1426, over 28780.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3709, pruned_loss=0.1192, over 5671315.52 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3547, pruned_loss=0.1062, over 5715569.16 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3721, pruned_loss=0.1199, over 5667604.80 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:00:33,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:00:35,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:00:50,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1163904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:00:54,888 INFO [train.py:968] (1/2) Epoch 26, batch 24850, libri_loss[loss=0.2843, simple_loss=0.3611, pruned_loss=0.1037, over 29493.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3694, pruned_loss=0.1191, over 5670464.00 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3547, pruned_loss=0.1061, over 5718484.70 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3709, pruned_loss=0.1202, over 5663567.13 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:00:55,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1163907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:01:01,800 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:01:19,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1163936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:01:36,777 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 24900, giga_loss[loss=0.2905, simple_loss=0.3714, pruned_loss=0.1048, over 28860.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3686, pruned_loss=0.1173, over 5676802.17 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.355, pruned_loss=0.1066, over 5713813.82 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3699, pruned_loss=0.1182, over 5674159.14 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:02:08,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3986, 4.2477, 1.6202, 1.6678], device='cuda:1'), covar=tensor([0.1069, 0.0489, 0.0977, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0568, 0.0403, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 13:02:18,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9329, 1.1324, 1.0769, 0.9000], device='cuda:1'), covar=tensor([0.2367, 0.2732, 0.1777, 0.2165], device='cuda:1'), in_proj_covar=tensor([0.2041, 0.1989, 0.1918, 0.2041], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 13:02:24,738 INFO [train.py:968] (1/2) Epoch 26, batch 24950, giga_loss[loss=0.3364, simple_loss=0.3996, pruned_loss=0.1366, over 28384.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.369, pruned_loss=0.1163, over 5686399.22 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3556, pruned_loss=0.1071, over 5716244.59 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3696, pruned_loss=0.1166, over 5681537.89 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:02:32,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1164014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:02:41,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 13:03:00,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1164046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:03:11,113 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 25000, giga_loss[loss=0.2726, simple_loss=0.3547, pruned_loss=0.09524, over 28624.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3688, pruned_loss=0.1164, over 5679221.42 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3556, pruned_loss=0.1071, over 5720073.61 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3696, pruned_loss=0.1168, over 5671370.41 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:03:59,606 INFO [train.py:968] (1/2) Epoch 26, batch 25050, giga_loss[loss=0.2931, simple_loss=0.362, pruned_loss=0.1121, over 28951.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3665, pruned_loss=0.1153, over 5686024.94 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3556, pruned_loss=0.1071, over 5721026.63 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3672, pruned_loss=0.1156, over 5678760.15 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:04:50,257 INFO [optim.py:369] (1/2) 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 13:04:50,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2927, 1.3916, 1.3781, 1.2500], device='cuda:1'), covar=tensor([0.2724, 0.2772, 0.2041, 0.2427], device='cuda:1'), in_proj_covar=tensor([0.2040, 0.1989, 0.1918, 0.2041], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 13:04:50,919 INFO [train.py:968] (1/2) Epoch 26, batch 25100, giga_loss[loss=0.306, simple_loss=0.3759, pruned_loss=0.118, over 28323.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3662, pruned_loss=0.1158, over 5675741.45 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3558, pruned_loss=0.1072, over 5726352.43 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3669, pruned_loss=0.1163, over 5663802.81 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:04:51,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1164157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:04:53,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1164160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:04:56,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4432, 1.8458, 1.2593, 0.9916], device='cuda:1'), covar=tensor([0.6316, 0.3637, 0.3185, 0.5801], device='cuda:1'), in_proj_covar=tensor([0.1822, 0.1719, 0.1652, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 13:05:20,945 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1164189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:05:20,999 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1164189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:05:23,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1164192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:05:36,295 INFO [train.py:968] (1/2) Epoch 26, batch 25150, giga_loss[loss=0.3286, simple_loss=0.3863, pruned_loss=0.1354, over 28592.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3668, pruned_loss=0.117, over 5676590.84 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3564, pruned_loss=0.1077, over 5727374.52 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3671, pruned_loss=0.1172, over 5664334.90 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:05:43,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 13:05:51,974 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:06:27,280 INFO [train.py:968] (1/2) Epoch 26, batch 25200, giga_loss[loss=0.3029, simple_loss=0.3617, pruned_loss=0.122, over 28758.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3661, pruned_loss=0.1174, over 5674988.64 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3564, pruned_loss=0.1077, over 5729183.81 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3665, pruned_loss=0.1177, over 5662952.73 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:06:46,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-13 13:07:10,728 INFO [train.py:968] (1/2) Epoch 26, batch 25250, giga_loss[loss=0.3128, simple_loss=0.377, pruned_loss=0.1243, over 28724.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3633, pruned_loss=0.1155, over 5662699.65 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3563, pruned_loss=0.1078, over 5713118.02 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5665847.87 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:08:02,341 INFO [optim.py:369] (1/2) 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 13:08:02,354 INFO [train.py:968] (1/2) Epoch 26, batch 25300, libri_loss[loss=0.287, simple_loss=0.3593, pruned_loss=0.1073, over 29242.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3632, pruned_loss=0.1163, over 5656533.27 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3565, pruned_loss=0.108, over 5714658.84 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3637, pruned_loss=0.1166, over 5656552.29 frames. ], batch size: 97, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:08:52,108 INFO [train.py:968] (1/2) Epoch 26, batch 25350, libri_loss[loss=0.2848, simple_loss=0.3486, pruned_loss=0.1105, over 29555.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3638, pruned_loss=0.1161, over 5657476.75 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3566, pruned_loss=0.1081, over 5713654.85 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3643, pruned_loss=0.1164, over 5657161.84 frames. ], batch size: 76, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:09:35,532 INFO [optim.py:369] (1/2) 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 13:09:35,545 INFO [train.py:968] (1/2) Epoch 26, batch 25400, giga_loss[loss=0.2935, simple_loss=0.3649, pruned_loss=0.111, over 29054.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.365, pruned_loss=0.1161, over 5664983.14 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3562, pruned_loss=0.1079, over 5717833.23 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3658, pruned_loss=0.1168, over 5660003.78 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:09:57,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3021, 1.2568, 1.2457, 1.4366], device='cuda:1'), covar=tensor([0.0807, 0.0392, 0.0359, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 13:10:18,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2718, 1.3319, 3.7076, 3.2776], device='cuda:1'), covar=tensor([0.1711, 0.2795, 0.0502, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0671, 0.0992, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 13:10:21,311 INFO [train.py:968] (1/2) Epoch 26, batch 25450, giga_loss[loss=0.296, simple_loss=0.3646, pruned_loss=0.1137, over 28814.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3642, pruned_loss=0.1151, over 5668253.30 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.356, pruned_loss=0.1079, over 5721775.22 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3653, pruned_loss=0.1157, over 5659797.60 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:10:49,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5485, 1.7560, 1.8634, 1.4846], device='cuda:1'), covar=tensor([0.3023, 0.2628, 0.2586, 0.2860], device='cuda:1'), in_proj_covar=tensor([0.2045, 0.1996, 0.1923, 0.2047], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 13:11:08,353 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 25500, giga_loss[loss=0.2799, simple_loss=0.3517, pruned_loss=0.1041, over 28887.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3655, pruned_loss=0.1162, over 5668245.39 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3564, pruned_loss=0.1081, over 5724171.30 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3662, pruned_loss=0.1167, over 5657882.67 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:11:56,411 INFO [train.py:968] (1/2) Epoch 26, batch 25550, giga_loss[loss=0.3009, simple_loss=0.3741, pruned_loss=0.1139, over 29013.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3676, pruned_loss=0.1181, over 5661505.03 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3565, pruned_loss=0.1081, over 5718200.82 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3682, pruned_loss=0.1187, over 5656426.96 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:12:44,307 INFO [optim.py:369] (1/2) 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 13:12:44,319 INFO [train.py:968] (1/2) Epoch 26, batch 25600, giga_loss[loss=0.2931, simple_loss=0.3616, pruned_loss=0.1123, over 28982.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3699, pruned_loss=0.1213, over 5647921.99 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3563, pruned_loss=0.1081, over 5712858.27 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.122, over 5647365.39 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:13:32,429 INFO [train.py:968] (1/2) Epoch 26, batch 25650, giga_loss[loss=0.2854, simple_loss=0.3515, pruned_loss=0.1097, over 29082.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3707, pruned_loss=0.1227, over 5663412.28 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3569, pruned_loss=0.1087, over 5712593.92 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3713, pruned_loss=0.1231, over 5661748.20 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:13:33,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2913, 1.1733, 1.1515, 1.5535], device='cuda:1'), covar=tensor([0.0746, 0.0367, 0.0357, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 13:14:23,222 INFO [train.py:968] (1/2) Epoch 26, batch 25700, giga_loss[loss=0.3471, simple_loss=0.3992, pruned_loss=0.1475, over 28269.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3734, pruned_loss=0.1253, over 5651996.04 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3572, pruned_loss=0.1089, over 5716531.90 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3738, pruned_loss=0.1257, over 5646333.01 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:14:23,824 INFO [optim.py:369] (1/2) 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 13:15:04,836 INFO [train.py:968] (1/2) Epoch 26, batch 25750, giga_loss[loss=0.2942, simple_loss=0.3624, pruned_loss=0.113, over 28977.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3714, pruned_loss=0.1236, over 5662218.38 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3574, pruned_loss=0.109, over 5719317.84 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3722, pruned_loss=0.1244, over 5652345.38 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:15:51,611 INFO [train.py:968] (1/2) Epoch 26, batch 25800, giga_loss[loss=0.2897, simple_loss=0.3619, pruned_loss=0.1088, over 28925.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3717, pruned_loss=0.1235, over 5650665.98 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3578, pruned_loss=0.1093, over 5710041.32 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3722, pruned_loss=0.1241, over 5650245.98 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:15:52,727 INFO [optim.py:369] (1/2) 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 13:15:54,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2312, 1.8244, 1.4274, 0.4565], device='cuda:1'), covar=tensor([0.5920, 0.4076, 0.5109, 0.7607], device='cuda:1'), in_proj_covar=tensor([0.1824, 0.1724, 0.1652, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 13:16:13,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3610, 2.9048, 2.7148, 2.0240], device='cuda:1'), covar=tensor([0.3557, 0.2186, 0.2535, 0.3148], device='cuda:1'), in_proj_covar=tensor([0.2047, 0.1997, 0.1927, 0.2049], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 13:16:26,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 13:16:33,038 INFO [train.py:968] (1/2) Epoch 26, batch 25850, giga_loss[loss=0.2794, simple_loss=0.3544, pruned_loss=0.1022, over 28924.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3698, pruned_loss=0.1206, over 5664595.45 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3578, pruned_loss=0.1095, over 5712433.88 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3705, pruned_loss=0.1212, over 5661066.36 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:17:00,115 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4002, 1.2911, 1.3095, 1.5138], device='cuda:1'), covar=tensor([0.0734, 0.0357, 0.0340, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 13:17:10,161 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4434, 2.0654, 1.5056, 0.6922], device='cuda:1'), covar=tensor([0.5501, 0.3188, 0.3810, 0.6598], device='cuda:1'), in_proj_covar=tensor([0.1815, 0.1720, 0.1650, 0.1485], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 13:17:19,091 INFO [train.py:968] (1/2) Epoch 26, batch 25900, giga_loss[loss=0.278, simple_loss=0.3521, pruned_loss=0.1019, over 28998.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3675, pruned_loss=0.119, over 5663221.30 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3579, pruned_loss=0.1096, over 5717092.92 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3682, pruned_loss=0.1196, over 5654954.32 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:17:20,453 INFO [optim.py:369] (1/2) 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 13:17:57,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3755, 2.6553, 2.4316, 1.8740], device='cuda:1'), covar=tensor([0.2773, 0.2055, 0.2436, 0.2964], device='cuda:1'), in_proj_covar=tensor([0.2045, 0.1997, 0.1925, 0.2047], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 13:18:05,580 INFO [train.py:968] (1/2) Epoch 26, batch 25950, giga_loss[loss=0.3393, simple_loss=0.3845, pruned_loss=0.147, over 27556.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3665, pruned_loss=0.1192, over 5658973.69 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3587, pruned_loss=0.1103, over 5711327.75 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3666, pruned_loss=0.1193, over 5656591.36 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:18:20,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8124, 1.9780, 1.4724, 1.7068], device='cuda:1'), covar=tensor([0.0893, 0.0499, 0.0946, 0.0940], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0452, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 13:18:52,735 INFO [train.py:968] (1/2) Epoch 26, batch 26000, giga_loss[loss=0.3067, simple_loss=0.3663, pruned_loss=0.1235, over 29086.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3633, pruned_loss=0.1167, over 5682120.72 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3583, pruned_loss=0.1102, over 5715719.39 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3639, pruned_loss=0.1171, over 5674998.56 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:18:56,151 INFO [optim.py:369] (1/2) 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 13:19:13,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4503, 2.0175, 1.4345, 0.8622], device='cuda:1'), covar=tensor([0.6698, 0.3272, 0.3932, 0.6727], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1725, 0.1654, 0.1490], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 13:19:24,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3166, 4.1679, 3.9324, 1.9092], device='cuda:1'), covar=tensor([0.0651, 0.0770, 0.0767, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.1295, 0.1190, 0.1007, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 13:19:39,804 INFO [train.py:968] (1/2) Epoch 26, batch 26050, giga_loss[loss=0.342, simple_loss=0.4035, pruned_loss=0.1402, over 28622.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3654, pruned_loss=0.118, over 5682804.65 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3584, pruned_loss=0.1104, over 5720303.16 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3659, pruned_loss=0.1184, over 5671571.05 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:19:50,669 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-13 13:19:55,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-13 13:20:06,196 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0887, 3.9347, 3.6866, 2.0021], device='cuda:1'), covar=tensor([0.0743, 0.0925, 0.1000, 0.1997], device='cuda:1'), in_proj_covar=tensor([0.1294, 0.1191, 0.1008, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 13:20:10,089 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-13 13:20:25,708 INFO [train.py:968] (1/2) Epoch 26, batch 26100, giga_loss[loss=0.2828, simple_loss=0.3725, pruned_loss=0.09655, over 29027.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3692, pruned_loss=0.1184, over 5692623.65 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3585, pruned_loss=0.1106, over 5723928.15 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3697, pruned_loss=0.1187, over 5679732.19 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:20:27,611 INFO [optim.py:369] (1/2) 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 13:21:13,549 INFO [train.py:968] (1/2) Epoch 26, batch 26150, giga_loss[loss=0.3145, simple_loss=0.3907, pruned_loss=0.1192, over 28668.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3704, pruned_loss=0.1173, over 5689458.01 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.358, pruned_loss=0.1104, over 5728556.24 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3715, pruned_loss=0.1179, over 5674224.16 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:21:28,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5599, 4.4044, 4.1823, 1.9290], device='cuda:1'), covar=tensor([0.0686, 0.0777, 0.0877, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1191, 0.1007, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 13:21:46,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 13:22:01,480 INFO [train.py:968] (1/2) Epoch 26, batch 26200, giga_loss[loss=0.2721, simple_loss=0.3488, pruned_loss=0.09768, over 28650.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3718, pruned_loss=0.1185, over 5696431.43 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3578, pruned_loss=0.1104, over 5730721.88 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.373, pruned_loss=0.119, over 5681928.27 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:22:04,671 INFO [optim.py:369] (1/2) 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 13:22:10,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3441, 1.6403, 1.5883, 1.4684], device='cuda:1'), covar=tensor([0.2155, 0.2075, 0.2447, 0.2117], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0762, 0.0728, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 13:22:48,294 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:22:49,327 INFO [train.py:968] (1/2) Epoch 26, batch 26250, giga_loss[loss=0.3232, simple_loss=0.3896, pruned_loss=0.1284, over 28809.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3736, pruned_loss=0.1203, over 5693512.29 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3579, pruned_loss=0.1105, over 5735274.72 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3748, pruned_loss=0.1208, over 5677227.89 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:23:11,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6185, 1.5890, 1.8152, 1.4576], device='cuda:1'), covar=tensor([0.1339, 0.1955, 0.1122, 0.1422], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0712, 0.0970, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 13:23:37,592 INFO [train.py:968] (1/2) Epoch 26, batch 26300, giga_loss[loss=0.3048, simple_loss=0.3694, pruned_loss=0.1201, over 28702.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3738, pruned_loss=0.1214, over 5680802.19 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3577, pruned_loss=0.1104, over 5727941.17 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3752, pruned_loss=0.1221, over 5674144.09 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:23:41,808 INFO [optim.py:369] (1/2) 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 13:24:12,658 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:24:23,566 INFO [train.py:968] (1/2) Epoch 26, batch 26350, giga_loss[loss=0.3052, simple_loss=0.3685, pruned_loss=0.1209, over 28854.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3732, pruned_loss=0.1215, over 5683350.07 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.358, pruned_loss=0.1106, over 5726755.06 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3747, pruned_loss=0.1223, over 5677075.20 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:25:08,610 INFO [train.py:968] (1/2) Epoch 26, batch 26400, giga_loss[loss=0.3085, simple_loss=0.3688, pruned_loss=0.1241, over 28649.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3711, pruned_loss=0.1208, over 5685565.38 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.358, pruned_loss=0.1105, over 5729210.63 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3724, pruned_loss=0.1216, over 5678052.33 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:25:11,403 INFO [optim.py:369] (1/2) 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 13:25:59,148 INFO [train.py:968] (1/2) Epoch 26, batch 26450, giga_loss[loss=0.2672, simple_loss=0.3332, pruned_loss=0.1006, over 28991.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3694, pruned_loss=0.1203, over 5688151.56 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.358, pruned_loss=0.1104, over 5722930.66 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3706, pruned_loss=0.1212, over 5687087.72 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:26:45,722 INFO [train.py:968] (1/2) Epoch 26, batch 26500, libri_loss[loss=0.302, simple_loss=0.3643, pruned_loss=0.1199, over 20269.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3698, pruned_loss=0.121, over 5671579.00 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3581, pruned_loss=0.1105, over 5716199.64 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3709, pruned_loss=0.1219, over 5676710.18 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:26:47,763 INFO [optim.py:369] (1/2) 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 13:27:03,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-13 13:27:25,507 INFO [train.py:968] (1/2) Epoch 26, batch 26550, giga_loss[loss=0.292, simple_loss=0.3599, pruned_loss=0.112, over 28849.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3686, pruned_loss=0.1204, over 5682477.73 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3579, pruned_loss=0.1106, over 5725444.98 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3702, pruned_loss=0.1215, over 5676401.31 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:27:39,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4003, 3.1706, 1.5057, 1.4696], device='cuda:1'), covar=tensor([0.0979, 0.0395, 0.0882, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0567, 0.0403, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 13:27:52,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 13:28:01,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1566, 1.4712, 1.1222, 0.5592], device='cuda:1'), covar=tensor([0.3249, 0.2177, 0.2831, 0.5457], device='cuda:1'), in_proj_covar=tensor([0.1817, 0.1717, 0.1647, 0.1483], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 13:28:08,885 INFO [train.py:968] (1/2) Epoch 26, batch 26600, giga_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.09642, over 28857.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3678, pruned_loss=0.1209, over 5666223.76 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3582, pruned_loss=0.1109, over 5719776.56 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3691, pruned_loss=0.1217, over 5664578.77 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:28:11,044 INFO [optim.py:369] (1/2) 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 13:28:34,124 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1165680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:29:00,308 INFO [train.py:968] (1/2) Epoch 26, batch 26650, giga_loss[loss=0.3007, simple_loss=0.3713, pruned_loss=0.1151, over 28838.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3665, pruned_loss=0.1205, over 5654536.30 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3582, pruned_loss=0.1109, over 5719776.56 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3675, pruned_loss=0.1211, over 5653255.99 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:29:10,944 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:29:46,116 INFO [train.py:968] (1/2) Epoch 26, batch 26700, giga_loss[loss=0.2713, simple_loss=0.3517, pruned_loss=0.09547, over 28731.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5650133.71 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1114, over 5704381.20 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3679, pruned_loss=0.1201, over 5660418.96 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:29:48,237 INFO [optim.py:369] (1/2) 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 13:30:00,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1165774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:30:31,232 INFO [train.py:968] (1/2) Epoch 26, batch 26750, giga_loss[loss=0.2952, simple_loss=0.3663, pruned_loss=0.1121, over 28591.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.37, pruned_loss=0.1216, over 5647577.82 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3588, pruned_loss=0.1115, over 5706191.63 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3706, pruned_loss=0.122, over 5652760.11 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:30:31,417 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:30:48,084 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1165823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:30:50,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1165826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:31:05,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0621, 4.9174, 4.7052, 2.2809], device='cuda:1'), covar=tensor([0.0628, 0.0739, 0.0755, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.1194, 0.1010, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 13:31:16,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1165855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:31:17,393 INFO [train.py:968] (1/2) Epoch 26, batch 26800, giga_loss[loss=0.2906, simple_loss=0.362, pruned_loss=0.1096, over 29032.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3698, pruned_loss=0.1215, over 5654643.07 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3593, pruned_loss=0.1118, over 5701372.11 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1218, over 5660990.32 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:31:20,148 INFO [optim.py:369] (1/2) 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 13:31:43,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165886.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:32:01,861 INFO [train.py:968] (1/2) Epoch 26, batch 26850, giga_loss[loss=0.2825, simple_loss=0.3662, pruned_loss=0.09942, over 28913.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3697, pruned_loss=0.1187, over 5659426.06 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3593, pruned_loss=0.1119, over 5700643.43 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3701, pruned_loss=0.119, over 5664541.75 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:32:13,003 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1165917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:32:16,137 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1165949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:32:48,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3599, 1.1575, 3.6816, 3.2319], device='cuda:1'), covar=tensor([0.1631, 0.2853, 0.0548, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0672, 0.0995, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 13:32:48,435 INFO [train.py:968] (1/2) Epoch 26, batch 26900, libri_loss[loss=0.2891, simple_loss=0.3622, pruned_loss=0.108, over 29523.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3705, pruned_loss=0.1167, over 5679541.02 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3591, pruned_loss=0.1117, over 5706741.49 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3714, pruned_loss=0.1172, over 5677215.54 frames. ], batch size: 84, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:32:51,363 INFO [optim.py:369] (1/2) 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 13:32:52,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1165961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:33:31,096 INFO [train.py:968] (1/2) Epoch 26, batch 26950, giga_loss[loss=0.3224, simple_loss=0.3914, pruned_loss=0.1267, over 28856.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3738, pruned_loss=0.1187, over 5689586.28 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3591, pruned_loss=0.1119, over 5714051.81 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3749, pruned_loss=0.1191, over 5680272.29 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:34:19,228 INFO [train.py:968] (1/2) Epoch 26, batch 27000, giga_loss[loss=0.3235, simple_loss=0.3916, pruned_loss=0.1277, over 29065.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3755, pruned_loss=0.1205, over 5684819.21 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3588, pruned_loss=0.1118, over 5714557.96 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3769, pruned_loss=0.1212, over 5676216.42 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:34:19,229 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 13:34:27,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0156, 1.0983, 3.0505, 2.8059], device='cuda:1'), covar=tensor([0.1669, 0.2768, 0.0562, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0670, 0.0993, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 13:34:27,998 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 13:34:30,724 INFO [optim.py:369] (1/2) 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 13:34:48,178 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3287, 1.8288, 1.3101, 0.7077], device='cuda:1'), covar=tensor([0.4445, 0.2442, 0.3120, 0.5938], device='cuda:1'), in_proj_covar=tensor([0.1827, 0.1723, 0.1652, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 13:35:03,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:35:04,897 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:35:14,193 INFO [train.py:968] (1/2) Epoch 26, batch 27050, libri_loss[loss=0.2629, simple_loss=0.3423, pruned_loss=0.09172, over 29566.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3779, pruned_loss=0.1239, over 5676497.22 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5718614.02 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3801, pruned_loss=0.1251, over 5664541.06 frames. ], batch size: 75, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:35:16,881 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:36:05,180 INFO [train.py:968] (1/2) Epoch 26, batch 27100, giga_loss[loss=0.3283, simple_loss=0.387, pruned_loss=0.1348, over 28663.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.378, pruned_loss=0.1252, over 5656020.71 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3582, pruned_loss=0.1114, over 5712287.38 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3801, pruned_loss=0.1264, over 5652104.60 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:36:09,184 INFO [optim.py:369] (1/2) 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 13:36:30,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:36:57,010 INFO [train.py:968] (1/2) Epoch 26, batch 27150, giga_loss[loss=0.3038, simple_loss=0.3743, pruned_loss=0.1166, over 28358.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3772, pruned_loss=0.1246, over 5649292.66 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5716160.03 frames. ], giga_tot_loss[loss=0.3153, simple_loss=0.379, pruned_loss=0.1258, over 5641866.41 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:37:25,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:37:27,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:37:41,185 INFO [train.py:968] (1/2) Epoch 26, batch 27200, giga_loss[loss=0.2984, simple_loss=0.3718, pruned_loss=0.1125, over 28594.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3764, pruned_loss=0.1222, over 5658433.54 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3588, pruned_loss=0.1116, over 5717318.73 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.378, pruned_loss=0.1232, over 5650307.26 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:37:47,150 INFO [optim.py:369] (1/2) 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 13:37:47,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166261.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:37:54,262 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:38:30,353 INFO [train.py:968] (1/2) Epoch 26, batch 27250, giga_loss[loss=0.283, simple_loss=0.3633, pruned_loss=0.1013, over 29051.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3758, pruned_loss=0.1201, over 5673021.18 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3585, pruned_loss=0.1115, over 5720033.36 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3775, pruned_loss=0.1211, over 5663271.28 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 13:38:47,689 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:38:51,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:38:58,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:39:15,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9373, 1.0679, 1.1002, 0.9574], device='cuda:1'), covar=tensor([0.2249, 0.2765, 0.1693, 0.2041], device='cuda:1'), in_proj_covar=tensor([0.2046, 0.1994, 0.1924, 0.2046], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 13:39:19,034 INFO [train.py:968] (1/2) Epoch 26, batch 27300, giga_loss[loss=0.3414, simple_loss=0.3972, pruned_loss=0.1428, over 28575.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3766, pruned_loss=0.121, over 5653009.44 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3585, pruned_loss=0.1115, over 5701774.77 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3784, pruned_loss=0.1219, over 5660363.45 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:39:19,296 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:39:24,852 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166404.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:40:05,239 INFO [train.py:968] (1/2) Epoch 26, batch 27350, giga_loss[loss=0.3885, simple_loss=0.4251, pruned_loss=0.176, over 26586.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3774, pruned_loss=0.1226, over 5652307.18 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3584, pruned_loss=0.1117, over 5694565.50 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3796, pruned_loss=0.1236, over 5662787.59 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:40:06,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166407.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:40:31,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:40:31,513 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166436.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:40:52,183 INFO [train.py:968] (1/2) Epoch 26, batch 27400, giga_loss[loss=0.2943, simple_loss=0.3535, pruned_loss=0.1175, over 28616.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.375, pruned_loss=0.122, over 5647447.74 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3587, pruned_loss=0.112, over 5686714.72 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3766, pruned_loss=0.1226, over 5662820.58 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:41:00,407 INFO [optim.py:369] (1/2) 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 13:41:08,436 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:41:15,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:41:20,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:41:23,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166485.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:41:44,372 INFO [train.py:968] (1/2) Epoch 26, batch 27450, giga_loss[loss=0.2786, simple_loss=0.3533, pruned_loss=0.1019, over 28902.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3733, pruned_loss=0.1222, over 5629971.40 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3591, pruned_loss=0.1122, over 5685714.17 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3745, pruned_loss=0.1227, over 5642036.76 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:41:50,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:42:22,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6849, 1.8402, 1.5229, 1.8076], device='cuda:1'), covar=tensor([0.2601, 0.2782, 0.3007, 0.2661], device='cuda:1'), in_proj_covar=tensor([0.1573, 0.1131, 0.1391, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 13:42:34,488 INFO [train.py:968] (1/2) Epoch 26, batch 27500, giga_loss[loss=0.2968, simple_loss=0.3638, pruned_loss=0.1149, over 28890.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3716, pruned_loss=0.1212, over 5644646.99 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3597, pruned_loss=0.1124, over 5693333.47 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3725, pruned_loss=0.1218, over 5645613.36 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:42:38,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4826, 1.5742, 1.2135, 1.1514], device='cuda:1'), covar=tensor([0.0947, 0.0553, 0.1014, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0456, 0.0527, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 13:42:38,611 INFO [optim.py:369] (1/2) 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 13:43:16,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4988, 1.7589, 1.3888, 1.3348], device='cuda:1'), covar=tensor([0.1137, 0.0655, 0.1052, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0455, 0.0526, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 13:43:21,629 INFO [train.py:968] (1/2) Epoch 26, batch 27550, giga_loss[loss=0.3427, simple_loss=0.3939, pruned_loss=0.1458, over 28618.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3721, pruned_loss=0.1232, over 5646249.55 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3596, pruned_loss=0.1122, over 5698965.70 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3731, pruned_loss=0.124, over 5640839.80 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:43:22,028 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5585, 1.6980, 1.3058, 1.2642], device='cuda:1'), covar=tensor([0.1093, 0.0695, 0.1094, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0456, 0.0526, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 13:43:29,765 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:43:32,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:43:42,810 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1166628.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:43:44,637 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166631.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 13:43:59,976 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:44:07,235 INFO [train.py:968] (1/2) Epoch 26, batch 27600, giga_loss[loss=0.3138, simple_loss=0.3688, pruned_loss=0.1294, over 28930.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3717, pruned_loss=0.1233, over 5635647.30 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3597, pruned_loss=0.1123, over 5684822.02 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3727, pruned_loss=0.1241, over 5643662.89 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:44:09,313 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166660.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:44:11,611 INFO [optim.py:369] (1/2) 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 13:44:40,905 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:44:47,096 INFO [train.py:968] (1/2) Epoch 26, batch 27650, giga_loss[loss=0.3119, simple_loss=0.3847, pruned_loss=0.1195, over 28695.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3689, pruned_loss=0.1201, over 5656011.00 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3598, pruned_loss=0.1123, over 5693635.15 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3701, pruned_loss=0.1211, over 5652648.15 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:45:27,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3963, 1.5056, 3.2827, 3.1835], device='cuda:1'), covar=tensor([0.1292, 0.2486, 0.0436, 0.1026], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0673, 0.0997, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 13:45:32,093 INFO [train.py:968] (1/2) Epoch 26, batch 27700, giga_loss[loss=0.2664, simple_loss=0.3471, pruned_loss=0.09281, over 29076.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3647, pruned_loss=0.1157, over 5658173.51 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1123, over 5695676.05 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3658, pruned_loss=0.1166, over 5652518.43 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:45:35,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9254, 1.1527, 1.0993, 0.9133], device='cuda:1'), covar=tensor([0.2494, 0.2825, 0.1811, 0.2253], device='cuda:1'), in_proj_covar=tensor([0.2047, 0.1994, 0.1924, 0.2045], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 13:45:38,995 INFO [optim.py:369] (1/2) 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 13:45:58,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6123, 1.8784, 1.4728, 2.0425], device='cuda:1'), covar=tensor([0.2730, 0.2873, 0.3261, 0.2520], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1137, 0.1398, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 13:46:23,781 INFO [train.py:968] (1/2) Epoch 26, batch 27750, giga_loss[loss=0.2662, simple_loss=0.3438, pruned_loss=0.09426, over 28985.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.363, pruned_loss=0.1141, over 5657962.34 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3595, pruned_loss=0.1121, over 5698029.39 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3643, pruned_loss=0.1151, over 5650758.46 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:46:28,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1166811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:46:52,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 13:47:12,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-13 13:47:13,027 INFO [train.py:968] (1/2) Epoch 26, batch 27800, giga_loss[loss=0.2782, simple_loss=0.3458, pruned_loss=0.1053, over 28919.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3622, pruned_loss=0.1144, over 5645025.50 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3601, pruned_loss=0.1127, over 5690487.42 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3627, pruned_loss=0.1147, over 5644849.14 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:47:19,011 INFO [optim.py:369] (1/2) 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 13:47:34,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1166877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:48:05,624 INFO [train.py:968] (1/2) Epoch 26, batch 27850, giga_loss[loss=0.3196, simple_loss=0.3776, pruned_loss=0.1308, over 28293.00 frames. ], tot_loss[loss=0.292, simple_loss=0.3589, pruned_loss=0.1126, over 5663544.03 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1125, over 5694962.02 frames. ], giga_tot_loss[loss=0.2927, simple_loss=0.3595, pruned_loss=0.113, over 5658845.27 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:48:52,083 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 27900, giga_loss[loss=0.2857, simple_loss=0.3592, pruned_loss=0.1061, over 29018.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1145, over 5661302.49 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 5687928.98 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1144, over 5662292.66 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:48:54,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1166957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:48:59,203 INFO [optim.py:369] (1/2) 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 13:49:21,374 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-13 13:49:25,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1166986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:49:45,132 INFO [train.py:968] (1/2) Epoch 26, batch 27950, giga_loss[loss=0.2822, simple_loss=0.3574, pruned_loss=0.1035, over 28951.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3644, pruned_loss=0.1158, over 5653299.75 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3605, pruned_loss=0.113, over 5689165.33 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3643, pruned_loss=0.1158, over 5652744.41 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:50:00,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6642, 1.8992, 1.7584, 1.6312], device='cuda:1'), covar=tensor([0.2134, 0.2302, 0.2558, 0.2356], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0759, 0.0724, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 13:50:16,170 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:50:34,701 INFO [train.py:968] (1/2) Epoch 26, batch 28000, giga_loss[loss=0.3415, simple_loss=0.379, pruned_loss=0.152, over 23125.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3654, pruned_loss=0.1165, over 5648364.86 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5691264.29 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3651, pruned_loss=0.1163, over 5645697.62 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:50:40,700 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:51:20,403 INFO [train.py:968] (1/2) Epoch 26, batch 28050, giga_loss[loss=0.2726, simple_loss=0.3473, pruned_loss=0.09898, over 28950.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.367, pruned_loss=0.1179, over 5653930.87 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.361, pruned_loss=0.1133, over 5696332.43 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3667, pruned_loss=0.1178, over 5646113.54 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:52:01,364 INFO [train.py:968] (1/2) Epoch 26, batch 28100, giga_loss[loss=0.3607, simple_loss=0.4187, pruned_loss=0.1514, over 28659.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3685, pruned_loss=0.119, over 5657069.92 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3617, pruned_loss=0.1138, over 5688388.74 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1187, over 5656925.17 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:52:07,844 INFO [optim.py:369] (1/2) 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 13:52:43,759 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-13 13:52:49,629 INFO [train.py:968] (1/2) Epoch 26, batch 28150, giga_loss[loss=0.3203, simple_loss=0.3864, pruned_loss=0.127, over 28929.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3702, pruned_loss=0.1201, over 5661957.42 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3616, pruned_loss=0.1136, over 5691694.62 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.37, pruned_loss=0.12, over 5658407.35 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:53:01,482 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:53:04,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:53:32,966 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:53:34,241 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 28200, libri_loss[loss=0.2995, simple_loss=0.3716, pruned_loss=0.1137, over 29232.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3714, pruned_loss=0.1206, over 5656132.14 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.362, pruned_loss=0.114, over 5684767.44 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.371, pruned_loss=0.1204, over 5658337.38 frames. ], batch size: 97, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:53:38,816 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 13:53:47,402 INFO [optim.py:369] (1/2) 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 13:54:26,871 INFO [train.py:968] (1/2) Epoch 26, batch 28250, giga_loss[loss=0.2918, simple_loss=0.3713, pruned_loss=0.1062, over 28553.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3715, pruned_loss=0.1212, over 5648715.87 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3619, pruned_loss=0.1139, over 5689452.43 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3716, pruned_loss=0.1213, over 5644769.26 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:54:33,220 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:54:33,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:54:58,926 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:55:04,239 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 13:55:19,305 INFO [train.py:968] (1/2) Epoch 26, batch 28300, giga_loss[loss=0.3068, simple_loss=0.3874, pruned_loss=0.1131, over 28299.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3729, pruned_loss=0.1221, over 5649287.82 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3621, pruned_loss=0.114, over 5692956.08 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.373, pruned_loss=0.1222, over 5642344.67 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:55:26,468 INFO [optim.py:369] (1/2) 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 13:55:58,445 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:00,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:08,524 INFO [train.py:968] (1/2) Epoch 26, batch 28350, giga_loss[loss=0.2931, simple_loss=0.3727, pruned_loss=0.1068, over 28871.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3734, pruned_loss=0.1209, over 5661545.00 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3622, pruned_loss=0.1141, over 5694900.57 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3736, pruned_loss=0.1211, over 5653551.55 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 13:56:16,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2468, 1.5219, 1.5538, 1.1077], device='cuda:1'), covar=tensor([0.1745, 0.2669, 0.1460, 0.1744], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0715, 0.0973, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:1')
+2023-03-13 13:56:17,043 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:27,882 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:56:29,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7506, 1.7730, 1.9780, 1.5249], device='cuda:1'), covar=tensor([0.1911, 0.2548, 0.1536, 0.1828], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0715, 0.0973, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0014], device='cuda:1')
+2023-03-13 13:56:33,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5809, 1.8785, 1.5290, 1.7497], device='cuda:1'), covar=tensor([0.2558, 0.2626, 0.2806, 0.2608], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1138, 0.1396, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 13:56:43,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167444.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 13:56:55,841 INFO [train.py:968] (1/2) Epoch 26, batch 28400, giga_loss[loss=0.2665, simple_loss=0.3393, pruned_loss=0.09687, over 28569.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1202, over 5669457.24 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3615, pruned_loss=0.1139, over 5702096.59 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3729, pruned_loss=0.1208, over 5655635.12 frames. ], batch size: 60, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:56:57,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4138, 1.6850, 1.3565, 1.2101], device='cuda:1'), covar=tensor([0.2860, 0.2691, 0.3177, 0.2342], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1138, 0.1396, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 13:57:04,218 INFO [optim.py:369] (1/2) 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 13:57:05,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0043, 1.3368, 1.1247, 0.2276], device='cuda:1'), covar=tensor([0.4583, 0.3625, 0.4972, 0.7612], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1725, 0.1654, 0.1488], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 13:57:43,447 INFO [train.py:968] (1/2) Epoch 26, batch 28450, giga_loss[loss=0.2862, simple_loss=0.3516, pruned_loss=0.1104, over 28909.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.371, pruned_loss=0.1204, over 5675394.59 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3617, pruned_loss=0.1138, over 5707801.72 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.372, pruned_loss=0.1211, over 5658133.45 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:57:57,780 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:58:27,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-13 13:58:44,609 INFO [train.py:968] (1/2) Epoch 26, batch 28500, giga_loss[loss=0.2876, simple_loss=0.3563, pruned_loss=0.1094, over 28305.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3696, pruned_loss=0.1199, over 5683189.15 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3612, pruned_loss=0.1135, over 5710013.93 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.371, pruned_loss=0.1209, over 5666984.18 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 13:58:45,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:58:48,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:58:53,050 INFO [optim.py:369] (1/2) 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 13:59:12,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4744, 1.9248, 1.5351, 1.7779], device='cuda:1'), covar=tensor([0.0801, 0.0297, 0.0323, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0114], device='cuda:1')
+2023-03-13 13:59:18,702 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 13:59:34,211 INFO [train.py:968] (1/2) Epoch 26, batch 28550, giga_loss[loss=0.2869, simple_loss=0.347, pruned_loss=0.1134, over 28880.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3667, pruned_loss=0.1182, over 5678628.82 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5704797.04 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3684, pruned_loss=0.1193, over 5669889.32 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:00:20,800 INFO [train.py:968] (1/2) Epoch 26, batch 28600, giga_loss[loss=0.2691, simple_loss=0.3455, pruned_loss=0.09632, over 28878.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3673, pruned_loss=0.1194, over 5681556.18 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3609, pruned_loss=0.1136, over 5707634.20 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3686, pruned_loss=0.1201, over 5671833.84 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:00:21,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8380, 1.8646, 2.0020, 1.5843], device='cuda:1'), covar=tensor([0.1943, 0.2511, 0.1503, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0712, 0.0970, 0.0868], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 14:00:24,646 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167659.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:00:29,874 INFO [optim.py:369] (1/2) 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 14:00:35,762 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:00:50,229 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:00:54,078 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 26, batch 28650, giga_loss[loss=0.3024, simple_loss=0.37, pruned_loss=0.1175, over 28906.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3673, pruned_loss=0.1198, over 5654701.15 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3611, pruned_loss=0.1137, over 5700093.98 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3682, pruned_loss=0.1203, over 5653436.57 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:01:20,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:01:23,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5936, 1.6582, 1.7718, 1.3591], device='cuda:1'), covar=tensor([0.1846, 0.2651, 0.1560, 0.1796], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0713, 0.0971, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 14:01:59,645 INFO [train.py:968] (1/2) Epoch 26, batch 28700, giga_loss[loss=0.3182, simple_loss=0.3831, pruned_loss=0.1266, over 28721.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3678, pruned_loss=0.1205, over 5653817.01 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1132, over 5701492.19 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3693, pruned_loss=0.1215, over 5650527.76 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:02:07,894 INFO [optim.py:369] (1/2) 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 14:02:28,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:02:46,832 INFO [train.py:968] (1/2) Epoch 26, batch 28750, giga_loss[loss=0.2951, simple_loss=0.3587, pruned_loss=0.1158, over 28972.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3694, pruned_loss=0.1217, over 5658380.26 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3608, pruned_loss=0.1134, over 5703309.42 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3705, pruned_loss=0.1226, over 5653003.53 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:02:51,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-13 14:03:00,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167819.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:03:13,918 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:14,599 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:16,841 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:17,446 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:37,878 INFO [train.py:968] (1/2) Epoch 26, batch 28800, giga_loss[loss=0.2647, simple_loss=0.3428, pruned_loss=0.09325, over 28957.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3697, pruned_loss=0.1225, over 5647112.17 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5706742.02 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3707, pruned_loss=0.1232, over 5639087.21 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:03:38,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:40,744 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:44,845 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:03:45,498 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:03:57,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 14:04:01,270 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:02,920 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1167885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:05,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:09,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1167894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:04:20,272 INFO [train.py:968] (1/2) Epoch 26, batch 28850, giga_loss[loss=0.2794, simple_loss=0.351, pruned_loss=0.1039, over 29023.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3704, pruned_loss=0.1234, over 5657566.67 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3605, pruned_loss=0.113, over 5713442.88 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3717, pruned_loss=0.1247, over 5643399.51 frames. ], batch size: 155, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:04:33,353 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 14:04:37,359 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1905, 1.4982, 1.4683, 1.2829], device='cuda:1'), covar=tensor([0.2073, 0.1683, 0.2343, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0765, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 14:05:07,213 INFO [train.py:968] (1/2) Epoch 26, batch 28900, giga_loss[loss=0.2876, simple_loss=0.3648, pruned_loss=0.1052, over 28928.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.37, pruned_loss=0.1232, over 5660582.43 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3604, pruned_loss=0.1128, over 5716741.40 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3713, pruned_loss=0.1246, over 5645635.69 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:05:12,837 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1167962.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:05:16,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1167965.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:05:17,109 INFO [optim.py:369] (1/2) 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 14:05:22,137 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1640, 1.3191, 1.2439, 1.0967], device='cuda:1'), covar=tensor([0.2305, 0.2190, 0.1795, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.2052, 0.2001, 0.1929, 0.2053], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 14:05:43,859 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1167994.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:05:44,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8757, 1.2230, 1.2847, 1.0171], device='cuda:1'), covar=tensor([0.2282, 0.1563, 0.2683, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0762, 0.0729, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 14:05:58,902 INFO [train.py:968] (1/2) Epoch 26, batch 28950, giga_loss[loss=0.2691, simple_loss=0.3514, pruned_loss=0.09341, over 28825.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3702, pruned_loss=0.1227, over 5649359.71 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3605, pruned_loss=0.1128, over 5716255.66 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3713, pruned_loss=0.1239, over 5637121.95 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:06:14,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-13 14:06:17,738 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:23,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:27,227 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:29,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:32,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9762, 2.0801, 1.5047, 1.7068], device='cuda:1'), covar=tensor([0.1033, 0.0785, 0.1102, 0.1214], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0456, 0.0528, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 14:06:33,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:43,097 INFO [train.py:968] (1/2) Epoch 26, batch 29000, giga_loss[loss=0.3327, simple_loss=0.3902, pruned_loss=0.1376, over 28268.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3707, pruned_loss=0.1224, over 5639116.16 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3605, pruned_loss=0.113, over 5696419.79 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1234, over 5646687.22 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:06:44,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:06:52,023 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:06:56,222 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:07:19,679 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:07:29,581 INFO [train.py:968] (1/2) Epoch 26, batch 29050, giga_loss[loss=0.2784, simple_loss=0.3483, pruned_loss=0.1043, over 28835.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3707, pruned_loss=0.1223, over 5649341.64 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.36, pruned_loss=0.1126, over 5700763.65 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1237, over 5649764.13 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:08:04,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9754, 3.8054, 3.6444, 2.0088], device='cuda:1'), covar=tensor([0.0707, 0.0836, 0.0802, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1314, 0.1210, 0.1023, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 14:08:15,504 INFO [train.py:968] (1/2) Epoch 26, batch 29100, libri_loss[loss=0.2497, simple_loss=0.3168, pruned_loss=0.09128, over 28554.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3712, pruned_loss=0.1226, over 5664332.06 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3599, pruned_loss=0.1125, over 5700663.53 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.373, pruned_loss=0.1241, over 5663704.36 frames. ], batch size: 63, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:08:17,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:24,897 INFO [optim.py:369] (1/2) 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 14:08:34,722 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:38,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:44,345 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:44,440 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:46,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:57,191 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:08:58,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0847, 1.2303, 1.2166, 1.1154], device='cuda:1'), covar=tensor([0.2050, 0.2302, 0.1352, 0.1654], device='cuda:1'), in_proj_covar=tensor([0.2048, 0.2000, 0.1925, 0.2050], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 14:08:59,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:01,443 INFO [train.py:968] (1/2) Epoch 26, batch 29150, giga_loss[loss=0.3892, simple_loss=0.4358, pruned_loss=0.1713, over 28276.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3719, pruned_loss=0.1234, over 5669982.57 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1126, over 5704412.13 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3735, pruned_loss=0.1247, over 5665444.20 frames. ], batch size: 369, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:09:04,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:13,639 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:25,752 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:27,832 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:38,821 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 29200, giga_loss[loss=0.284, simple_loss=0.3573, pruned_loss=0.1054, over 28779.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1236, over 5671150.95 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1126, over 5709578.52 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3748, pruned_loss=0.1249, over 5662018.76 frames. ], batch size: 99, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:09:52,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:55,063 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:09:59,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1154, 1.2696, 3.3625, 2.9800], device='cuda:1'), covar=tensor([0.1726, 0.2845, 0.0565, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0669, 0.0996, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 14:10:01,682 INFO [optim.py:369] (1/2) 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 14:10:04,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5046, 1.6068, 1.6946, 1.2927], device='cuda:1'), covar=tensor([0.1859, 0.2774, 0.1558, 0.1835], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0714, 0.0971, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 14:10:35,380 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:10:38,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 29250, giga_loss[loss=0.2596, simple_loss=0.3434, pruned_loss=0.08786, over 28981.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3725, pruned_loss=0.1222, over 5663403.43 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3597, pruned_loss=0.1126, over 5713951.74 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3743, pruned_loss=0.1236, over 5650877.57 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:10:47,674 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:11:03,509 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:11:22,513 INFO [train.py:968] (1/2) Epoch 26, batch 29300, giga_loss[loss=0.2635, simple_loss=0.3426, pruned_loss=0.09223, over 28929.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3705, pruned_loss=0.1204, over 5678628.26 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1127, over 5716614.67 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3721, pruned_loss=0.1215, over 5665749.19 frames. ], batch size: 164, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:11:32,081 INFO [optim.py:369] (1/2) 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 14:12:03,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:04,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:04,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:05,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:06,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:07,128 INFO [train.py:968] (1/2) Epoch 26, batch 29350, giga_loss[loss=0.2819, simple_loss=0.3516, pruned_loss=0.1061, over 28924.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3697, pruned_loss=0.1205, over 5667254.84 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5722735.34 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3716, pruned_loss=0.1218, over 5649976.31 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:12:19,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-13 14:12:30,612 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:32,648 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:12:37,655 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 29400, libri_loss[loss=0.3052, simple_loss=0.3716, pruned_loss=0.1194, over 29288.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3714, pruned_loss=0.121, over 5677852.91 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1127, over 5725234.71 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3729, pruned_loss=0.1221, over 5661009.49 frames. ], batch size: 94, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:13:04,209 INFO [optim.py:369] (1/2) 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 14:13:07,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:13:09,762 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:13:14,129 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:13:43,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2391, 1.7943, 1.3318, 0.4470], device='cuda:1'), covar=tensor([0.4838, 0.3264, 0.4775, 0.6619], device='cuda:1'), in_proj_covar=tensor([0.1836, 0.1739, 0.1662, 0.1502], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 14:13:44,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3127, 0.8877, 0.9951, 1.5216], device='cuda:1'), covar=tensor([0.0731, 0.0392, 0.0339, 0.0809], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 14:13:45,295 INFO [train.py:968] (1/2) Epoch 26, batch 29450, giga_loss[loss=0.273, simple_loss=0.3477, pruned_loss=0.0992, over 29057.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3729, pruned_loss=0.1226, over 5666632.22 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5725721.02 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3745, pruned_loss=0.1238, over 5651904.32 frames. ], batch size: 128, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:14:22,295 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:25,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:33,076 INFO [train.py:968] (1/2) Epoch 26, batch 29500, giga_loss[loss=0.2935, simple_loss=0.3599, pruned_loss=0.1136, over 28937.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3722, pruned_loss=0.1229, over 5670013.01 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5730246.90 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3738, pruned_loss=0.124, over 5652966.29 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:14:37,229 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:42,978 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:54,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:14:56,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:19,881 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 29550, giga_loss[loss=0.4243, simple_loss=0.4546, pruned_loss=0.197, over 27837.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3731, pruned_loss=0.1241, over 5662422.63 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3595, pruned_loss=0.1126, over 5733687.84 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3747, pruned_loss=0.1252, over 5644894.84 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 14:15:24,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:25,097 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:27,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:31,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:15:52,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:16:03,790 INFO [train.py:968] (1/2) Epoch 26, batch 29600, giga_loss[loss=0.2917, simple_loss=0.3626, pruned_loss=0.1104, over 28551.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3725, pruned_loss=0.1231, over 5677460.30 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1124, over 5737059.07 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3744, pruned_loss=0.1245, over 5658756.04 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:16:11,629 INFO [optim.py:369] (1/2) 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 14:16:37,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:16:49,494 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:16:50,320 INFO [train.py:968] (1/2) Epoch 26, batch 29650, giga_loss[loss=0.2857, simple_loss=0.3649, pruned_loss=0.1033, over 28899.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1243, over 5645876.82 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5718275.56 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3756, pruned_loss=0.1256, over 5645371.23 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:16:53,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:16,969 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:26,819 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:28,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:34,005 INFO [train.py:968] (1/2) Epoch 26, batch 29700, giga_loss[loss=0.3578, simple_loss=0.3989, pruned_loss=0.1584, over 26484.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3726, pruned_loss=0.1228, over 5664963.32 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3591, pruned_loss=0.1123, over 5721808.53 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3746, pruned_loss=0.1243, over 5659678.53 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:17:40,186 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:42,644 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:17:44,217 INFO [optim.py:369] (1/2) 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 14:17:59,667 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1168781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:18:10,607 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 29750, giga_loss[loss=0.286, simple_loss=0.3581, pruned_loss=0.1069, over 28905.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3715, pruned_loss=0.1217, over 5667471.33 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1121, over 5725815.92 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3739, pruned_loss=0.1233, over 5658131.96 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:18:33,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7805, 1.5887, 4.9298, 3.7269], device='cuda:1'), covar=tensor([0.1650, 0.2806, 0.0415, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0670, 0.0996, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 14:18:46,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2806, 3.4692, 1.4977, 1.4480], device='cuda:1'), covar=tensor([0.1107, 0.0422, 0.0944, 0.1488], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0568, 0.0404, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 14:18:49,613 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:18:52,408 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:18:59,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:19:04,181 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1168853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:19:07,125 INFO [train.py:968] (1/2) Epoch 26, batch 29800, libri_loss[loss=0.2577, simple_loss=0.3245, pruned_loss=0.09543, over 29569.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1224, over 5671016.78 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3588, pruned_loss=0.1122, over 5732174.82 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3751, pruned_loss=0.1241, over 5655796.44 frames. ], batch size: 75, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:19:17,100 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:19:52,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3075, 1.6152, 1.3777, 1.5571], device='cuda:1'), covar=tensor([0.0804, 0.0344, 0.0339, 0.0887], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 14:19:53,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 14:19:54,923 INFO [train.py:968] (1/2) Epoch 26, batch 29850, libri_loss[loss=0.2386, simple_loss=0.3092, pruned_loss=0.08403, over 28182.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3717, pruned_loss=0.1219, over 5670760.04 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1123, over 5733895.50 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3736, pruned_loss=0.1233, over 5656078.50 frames. ], batch size: 62, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:20:42,402 INFO [train.py:968] (1/2) Epoch 26, batch 29900, giga_loss[loss=0.352, simple_loss=0.4012, pruned_loss=0.1514, over 27875.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3699, pruned_loss=0.1207, over 5675743.66 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3589, pruned_loss=0.1122, over 5737573.47 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3717, pruned_loss=0.1221, over 5659591.37 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:20:50,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5458, 2.2073, 1.5231, 0.7707], device='cuda:1'), covar=tensor([0.5452, 0.3231, 0.4735, 0.6957], device='cuda:1'), in_proj_covar=tensor([0.1838, 0.1740, 0.1663, 0.1499], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 14:20:52,293 INFO [optim.py:369] (1/2) 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 14:21:14,172 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:16,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:20,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1168996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:22,497 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1168999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:22,539 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1168999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:28,083 INFO [train.py:968] (1/2) Epoch 26, batch 29950, giga_loss[loss=0.261, simple_loss=0.3329, pruned_loss=0.09459, over 28928.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3687, pruned_loss=0.1207, over 5672426.87 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5737642.81 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3701, pruned_loss=0.1218, over 5658435.08 frames. ], batch size: 66, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:21:43,105 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:21:52,171 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:22:14,328 INFO [train.py:968] (1/2) Epoch 26, batch 30000, giga_loss[loss=0.3133, simple_loss=0.371, pruned_loss=0.1278, over 28597.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3649, pruned_loss=0.1187, over 5665825.19 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5722134.15 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3659, pruned_loss=0.1194, over 5666226.08 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:22:14,329 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 14:22:22,588 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 14:22:27,185 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-13 14:22:33,237 INFO [optim.py:369] (1/2) 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 14:23:03,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2254, 4.0751, 3.8598, 2.0037], device='cuda:1'), covar=tensor([0.0578, 0.0699, 0.0687, 0.2273], device='cuda:1'), in_proj_covar=tensor([0.1316, 0.1214, 0.1026, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-13 14:23:05,938 INFO [train.py:968] (1/2) Epoch 26, batch 30050, giga_loss[loss=0.3038, simple_loss=0.3688, pruned_loss=0.1194, over 28250.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3637, pruned_loss=0.1184, over 5677589.84 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3599, pruned_loss=0.1131, over 5718912.40 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3644, pruned_loss=0.1189, over 5678783.43 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:23:54,749 INFO [train.py:968] (1/2) Epoch 26, batch 30100, giga_loss[loss=0.3174, simple_loss=0.3719, pruned_loss=0.1315, over 28919.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3622, pruned_loss=0.1176, over 5680182.90 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3594, pruned_loss=0.1129, over 5710985.12 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3632, pruned_loss=0.1183, over 5686920.61 frames. ], batch size: 112, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:24:08,416 INFO [optim.py:369] (1/2) 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 14:24:42,834 INFO [train.py:968] (1/2) Epoch 26, batch 30150, giga_loss[loss=0.3119, simple_loss=0.3814, pruned_loss=0.1212, over 27983.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.362, pruned_loss=0.1161, over 5670423.10 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3594, pruned_loss=0.1129, over 5704789.52 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3629, pruned_loss=0.1168, over 5680270.19 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:25:18,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2526, 2.6033, 1.2218, 1.4838], device='cuda:1'), covar=tensor([0.1038, 0.0391, 0.1019, 0.1360], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0571, 0.0405, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 14:25:36,040 INFO [train.py:968] (1/2) Epoch 26, batch 30200, giga_loss[loss=0.2957, simple_loss=0.3759, pruned_loss=0.1078, over 28876.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3604, pruned_loss=0.113, over 5664855.62 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3591, pruned_loss=0.1128, over 5697827.28 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.3614, pruned_loss=0.1137, over 5678428.10 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:25:36,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6647, 4.2344, 1.6083, 1.7686], device='cuda:1'), covar=tensor([0.0951, 0.0298, 0.0973, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0570, 0.0405, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 14:25:49,512 INFO [optim.py:369] (1/2) 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 14:26:23,487 INFO [train.py:968] (1/2) Epoch 26, batch 30250, giga_loss[loss=0.3039, simple_loss=0.3666, pruned_loss=0.1206, over 27540.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3575, pruned_loss=0.1104, over 5652458.21 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3579, pruned_loss=0.1122, over 5698425.96 frames. ], giga_tot_loss[loss=0.2912, simple_loss=0.3594, pruned_loss=0.1115, over 5660369.96 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:26:27,803 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:27:03,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3517, 1.4134, 3.4552, 3.1116], device='cuda:1'), covar=tensor([0.1612, 0.2802, 0.0520, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0671, 0.0998, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 14:27:14,590 INFO [train.py:968] (1/2) Epoch 26, batch 30300, giga_loss[loss=0.2818, simple_loss=0.3534, pruned_loss=0.1051, over 28657.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3541, pruned_loss=0.1069, over 5656318.20 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3578, pruned_loss=0.1123, over 5703924.72 frames. ], giga_tot_loss[loss=0.2854, simple_loss=0.3557, pruned_loss=0.1075, over 5656488.80 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:27:24,466 INFO [optim.py:369] (1/2) 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 14:27:28,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1169374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:27:56,785 INFO [train.py:968] (1/2) Epoch 26, batch 30350, giga_loss[loss=0.2587, simple_loss=0.342, pruned_loss=0.08771, over 28962.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3511, pruned_loss=0.1042, over 5664904.87 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.357, pruned_loss=0.1122, over 5712833.02 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3528, pruned_loss=0.1045, over 5655083.40 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:28:48,394 INFO [train.py:968] (1/2) Epoch 26, batch 30400, giga_loss[loss=0.2673, simple_loss=0.3418, pruned_loss=0.0964, over 27660.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3485, pruned_loss=0.1004, over 5649935.90 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3571, pruned_loss=0.1123, over 5710792.17 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3498, pruned_loss=0.1004, over 5643285.43 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:29:01,501 INFO [optim.py:369] (1/2) 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 14:29:03,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1948, 3.0537, 2.8955, 1.4701], device='cuda:1'), covar=tensor([0.1133, 0.1172, 0.1116, 0.2620], device='cuda:1'), in_proj_covar=tensor([0.1304, 0.1206, 0.1018, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 14:29:03,889 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:29:38,887 INFO [train.py:968] (1/2) Epoch 26, batch 30450, libri_loss[loss=0.2965, simple_loss=0.3515, pruned_loss=0.1208, over 29535.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1001, over 5650364.31 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3563, pruned_loss=0.112, over 5718600.49 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3496, pruned_loss=0.0998, over 5634816.22 frames. ], batch size: 80, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:29:46,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1169517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:29:49,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1169520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:30:16,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3277, 1.5217, 1.5485, 1.1504], device='cuda:1'), covar=tensor([0.1811, 0.2860, 0.1555, 0.1930], device='cuda:1'), in_proj_covar=tensor([0.0918, 0.0707, 0.0965, 0.0867], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 14:30:20,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:30:28,359 INFO [train.py:968] (1/2) Epoch 26, batch 30500, giga_loss[loss=0.2632, simple_loss=0.3388, pruned_loss=0.09378, over 27617.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3486, pruned_loss=0.1002, over 5648860.61 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3559, pruned_loss=0.1119, over 5720364.66 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3499, pruned_loss=0.09978, over 5632976.92 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:30:41,883 INFO [optim.py:369] (1/2) 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 14:30:54,717 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 14:31:21,119 INFO [train.py:968] (1/2) Epoch 26, batch 30550, giga_loss[loss=0.2348, simple_loss=0.3152, pruned_loss=0.07719, over 27923.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3464, pruned_loss=0.0985, over 5644527.27 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3558, pruned_loss=0.1118, over 5719750.83 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3474, pruned_loss=0.09815, over 5632333.83 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:31:32,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-13 14:31:39,554 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6604, 5.1335, 1.8660, 2.0328], device='cuda:1'), covar=tensor([0.0949, 0.0384, 0.0904, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0568, 0.0404, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 14:31:39,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2318, 2.2822, 1.7825, 2.1677], device='cuda:1'), covar=tensor([0.0897, 0.0630, 0.0909, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0449, 0.0520, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 14:31:58,411 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-13 14:32:13,712 INFO [train.py:968] (1/2) Epoch 26, batch 30600, giga_loss[loss=0.2797, simple_loss=0.3536, pruned_loss=0.1029, over 28587.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3447, pruned_loss=0.09761, over 5645533.47 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3559, pruned_loss=0.1119, over 5723642.98 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3451, pruned_loss=0.09688, over 5630371.24 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:32:23,183 INFO [optim.py:369] (1/2) 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 14:32:39,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1169687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:32:56,832 INFO [train.py:968] (1/2) Epoch 26, batch 30650, giga_loss[loss=0.2466, simple_loss=0.3337, pruned_loss=0.07972, over 28836.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3441, pruned_loss=0.09686, over 5649438.41 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3551, pruned_loss=0.1116, over 5721692.78 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3448, pruned_loss=0.09612, over 5636582.60 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:33:44,645 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:33:51,197 INFO [train.py:968] (1/2) Epoch 26, batch 30700, libri_loss[loss=0.2682, simple_loss=0.3295, pruned_loss=0.1035, over 29599.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3429, pruned_loss=0.09542, over 5641541.97 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3551, pruned_loss=0.1116, over 5713797.51 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3434, pruned_loss=0.09465, over 5638602.59 frames. ], batch size: 74, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:34:00,555 INFO [optim.py:369] (1/2) 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 14:34:38,616 INFO [train.py:968] (1/2) Epoch 26, batch 30750, giga_loss[loss=0.248, simple_loss=0.3256, pruned_loss=0.0852, over 27589.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3411, pruned_loss=0.09396, over 5654486.77 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3546, pruned_loss=0.1114, over 5720955.31 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3415, pruned_loss=0.09298, over 5643639.31 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:34:40,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3269, 1.5541, 1.4142, 1.6967], device='cuda:1'), covar=tensor([0.0823, 0.0333, 0.0350, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 14:35:02,191 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1169830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:04,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1169833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:16,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1169846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:27,349 INFO [train.py:968] (1/2) Epoch 26, batch 30800, giga_loss[loss=0.2413, simple_loss=0.3202, pruned_loss=0.08114, over 28891.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.338, pruned_loss=0.09213, over 5641846.03 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3543, pruned_loss=0.1115, over 5713740.38 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3383, pruned_loss=0.09089, over 5637418.41 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:35:32,029 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1169862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:35:40,504 INFO [optim.py:369] (1/2) 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 14:36:12,573 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1169900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:36:13,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5565, 1.9164, 1.9386, 1.5094], device='cuda:1'), covar=tensor([0.2920, 0.2183, 0.2123, 0.2699], device='cuda:1'), in_proj_covar=tensor([0.2019, 0.1969, 0.1892, 0.2018], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 14:36:16,826 INFO [train.py:968] (1/2) Epoch 26, batch 30850, giga_loss[loss=0.3382, simple_loss=0.3839, pruned_loss=0.1463, over 26714.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.336, pruned_loss=0.09176, over 5647431.82 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3546, pruned_loss=0.1119, over 5717559.45 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3354, pruned_loss=0.0898, over 5638583.34 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:36:17,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4133, 1.6708, 1.4347, 1.6287], device='cuda:1'), covar=tensor([0.0743, 0.0346, 0.0341, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 14:36:26,064 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 30900, libri_loss[loss=0.2479, simple_loss=0.3175, pruned_loss=0.0892, over 29589.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3348, pruned_loss=0.09167, over 5652252.51 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.354, pruned_loss=0.1117, over 5722109.61 frames. ], giga_tot_loss[loss=0.257, simple_loss=0.3344, pruned_loss=0.08978, over 5639166.05 frames. ], batch size: 74, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:37:17,792 INFO [optim.py:369] (1/2) 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 14:37:40,213 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1169989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:37:44,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1169992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:37:58,114 INFO [train.py:968] (1/2) Epoch 26, batch 30950, giga_loss[loss=0.2712, simple_loss=0.336, pruned_loss=0.1032, over 26669.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3345, pruned_loss=0.09142, over 5628009.45 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3536, pruned_loss=0.1115, over 5712389.53 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3341, pruned_loss=0.08967, over 5624467.87 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:38:14,495 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:38:54,540 INFO [train.py:968] (1/2) Epoch 26, batch 31000, giga_loss[loss=0.2593, simple_loss=0.3414, pruned_loss=0.08863, over 28970.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3374, pruned_loss=0.09211, over 5615689.80 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3541, pruned_loss=0.112, over 5694336.21 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3364, pruned_loss=0.08993, over 5627117.30 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:38:57,774 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:39:53,596 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1170106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:39:53,970 INFO [train.py:968] (1/2) Epoch 26, batch 31050, giga_loss[loss=0.2502, simple_loss=0.3329, pruned_loss=0.08375, over 28715.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3381, pruned_loss=0.09145, over 5636241.67 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3539, pruned_loss=0.1121, over 5697452.06 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3373, pruned_loss=0.08943, over 5641539.03 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:40:21,971 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 31100, giga_loss[loss=0.2317, simple_loss=0.3171, pruned_loss=0.07316, over 28840.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3379, pruned_loss=0.09113, over 5657317.22 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3535, pruned_loss=0.1119, over 5700590.61 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3374, pruned_loss=0.08938, over 5658069.05 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:41:15,165 INFO [optim.py:369] (1/2) 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 14:41:59,894 INFO [train.py:968] (1/2) Epoch 26, batch 31150, giga_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08974, over 28624.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3352, pruned_loss=0.08931, over 5647905.37 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3532, pruned_loss=0.1118, over 5700512.02 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3347, pruned_loss=0.08758, over 5647746.83 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:43:03,507 INFO [train.py:968] (1/2) Epoch 26, batch 31200, giga_loss[loss=0.2102, simple_loss=0.3011, pruned_loss=0.05965, over 29080.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3334, pruned_loss=0.08724, over 5656380.87 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3525, pruned_loss=0.1114, over 5704771.25 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3333, pruned_loss=0.08574, over 5651659.31 frames. ], batch size: 113, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:43:16,709 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:43:16,993 INFO [optim.py:369] (1/2) 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 14:43:21,338 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:43:23,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:43:32,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3229, 1.7531, 1.8076, 1.4784], device='cuda:1'), covar=tensor([0.2194, 0.1894, 0.2132, 0.2166], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0748, 0.0717, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 14:43:44,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170290.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:43:58,108 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:44:01,932 INFO [train.py:968] (1/2) Epoch 26, batch 31250, libri_loss[loss=0.3317, simple_loss=0.3801, pruned_loss=0.1417, over 19474.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3331, pruned_loss=0.08788, over 5657335.79 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3527, pruned_loss=0.1117, over 5699086.47 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3323, pruned_loss=0.08577, over 5658173.59 frames. ], batch size: 187, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:44:34,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3994, 3.2366, 3.0696, 1.9135], device='cuda:1'), covar=tensor([0.0857, 0.1014, 0.0944, 0.1908], device='cuda:1'), in_proj_covar=tensor([0.1283, 0.1181, 0.0997, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 14:45:06,031 INFO [train.py:968] (1/2) Epoch 26, batch 31300, giga_loss[loss=0.2393, simple_loss=0.3255, pruned_loss=0.0765, over 28704.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3321, pruned_loss=0.08802, over 5656825.10 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3527, pruned_loss=0.1117, over 5701541.27 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3314, pruned_loss=0.08616, over 5655117.83 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:45:12,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 14:45:17,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 14:45:20,337 INFO [optim.py:369] (1/2) 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 14:45:38,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0131, 3.8646, 3.6628, 1.9104], device='cuda:1'), covar=tensor([0.0669, 0.0794, 0.0904, 0.2317], device='cuda:1'), in_proj_covar=tensor([0.1284, 0.1182, 0.0998, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 14:46:03,069 INFO [train.py:968] (1/2) Epoch 26, batch 31350, giga_loss[loss=0.3197, simple_loss=0.3778, pruned_loss=0.1308, over 27662.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3313, pruned_loss=0.08821, over 5665587.56 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3519, pruned_loss=0.1113, over 5706462.59 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3309, pruned_loss=0.0865, over 5658844.47 frames. ], batch size: 472, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:46:17,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:46:20,210 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:46:35,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170433.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:46:35,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170433.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:46:38,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170436.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:46:54,332 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:47:01,218 INFO [train.py:968] (1/2) Epoch 26, batch 31400, giga_loss[loss=0.2669, simple_loss=0.3516, pruned_loss=0.09107, over 28985.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3336, pruned_loss=0.08859, over 5664615.59 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3517, pruned_loss=0.1111, over 5708498.25 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3333, pruned_loss=0.08721, over 5657064.76 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:47:13,653 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170465.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 14:47:19,665 INFO [optim.py:369] (1/2) 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 14:47:34,706 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1170481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:48:08,873 INFO [train.py:968] (1/2) Epoch 26, batch 31450, giga_loss[loss=0.2248, simple_loss=0.2954, pruned_loss=0.07704, over 24650.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3355, pruned_loss=0.08912, over 5653338.70 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3515, pruned_loss=0.1111, over 5697797.79 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3352, pruned_loss=0.08784, over 5655741.93 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:49:03,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1249, 1.3255, 3.3255, 2.9699], device='cuda:1'), covar=tensor([0.1812, 0.2752, 0.0914, 0.1436], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0668, 0.0990, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 14:49:14,166 INFO [train.py:968] (1/2) Epoch 26, batch 31500, libri_loss[loss=0.2587, simple_loss=0.327, pruned_loss=0.09521, over 29573.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3314, pruned_loss=0.08679, over 5663360.93 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3509, pruned_loss=0.1107, over 5701955.07 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3314, pruned_loss=0.08563, over 5660526.90 frames. ], batch size: 75, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:49:29,953 INFO [optim.py:369] (1/2) 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 14:49:38,612 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:49:42,374 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:50:13,131 INFO [train.py:968] (1/2) Epoch 26, batch 31550, giga_loss[loss=0.2367, simple_loss=0.3205, pruned_loss=0.07645, over 28952.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3326, pruned_loss=0.08806, over 5665530.11 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3499, pruned_loss=0.1103, over 5699478.05 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3329, pruned_loss=0.08666, over 5664072.05 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:50:16,486 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:50:38,668 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1170624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:50:40,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1170627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:51:14,462 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1170656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 14:51:14,790 INFO [train.py:968] (1/2) Epoch 26, batch 31600, giga_loss[loss=0.2505, simple_loss=0.3342, pruned_loss=0.08346, over 26862.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3358, pruned_loss=0.08833, over 5656157.16 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3497, pruned_loss=0.1102, over 5699847.98 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3359, pruned_loss=0.08681, over 5653896.05 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 14:51:32,396 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:51:58,825 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1170691.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 14:52:16,762 INFO [train.py:968] (1/2) Epoch 26, batch 31650, giga_loss[loss=0.1932, simple_loss=0.2673, pruned_loss=0.05958, over 24493.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3392, pruned_loss=0.08795, over 5651830.72 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3497, pruned_loss=0.1102, over 5694270.43 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3389, pruned_loss=0.08619, over 5654746.98 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:53:17,499 INFO [train.py:968] (1/2) Epoch 26, batch 31700, giga_loss[loss=0.2345, simple_loss=0.3284, pruned_loss=0.07034, over 28347.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3393, pruned_loss=0.08706, over 5647146.57 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3494, pruned_loss=0.1101, over 5692767.45 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3392, pruned_loss=0.08542, over 5650333.99 frames. ], batch size: 369, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:53:36,440 INFO [optim.py:369] (1/2) 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 14:53:47,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 14:54:20,494 INFO [train.py:968] (1/2) Epoch 26, batch 31750, giga_loss[loss=0.2482, simple_loss=0.3341, pruned_loss=0.08118, over 28428.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3371, pruned_loss=0.08502, over 5651340.58 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3493, pruned_loss=0.1101, over 5694657.65 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.337, pruned_loss=0.08348, over 5651809.73 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:55:28,146 INFO [train.py:968] (1/2) Epoch 26, batch 31800, giga_loss[loss=0.2439, simple_loss=0.3294, pruned_loss=0.0792, over 28880.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3386, pruned_loss=0.087, over 5652421.73 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3492, pruned_loss=0.11, over 5697489.70 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3384, pruned_loss=0.08552, over 5649626.39 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:55:49,680 INFO [optim.py:369] (1/2) 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 14:56:42,295 INFO [train.py:968] (1/2) Epoch 26, batch 31850, giga_loss[loss=0.2506, simple_loss=0.3336, pruned_loss=0.08383, over 28472.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3395, pruned_loss=0.08897, over 5654758.78 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3495, pruned_loss=0.1104, over 5692364.46 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.339, pruned_loss=0.08714, over 5655794.47 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:56:52,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9049, 1.1477, 1.1404, 0.9427], device='cuda:1'), covar=tensor([0.2448, 0.2539, 0.1429, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1956, 0.1875, 0.2006], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 14:57:55,071 INFO [train.py:968] (1/2) Epoch 26, batch 31900, giga_loss[loss=0.2628, simple_loss=0.3402, pruned_loss=0.0927, over 28697.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3395, pruned_loss=0.08977, over 5662535.49 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3492, pruned_loss=0.1104, over 5687947.64 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3391, pruned_loss=0.08784, over 5667082.70 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:58:20,464 INFO [optim.py:369] (1/2) 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 14:58:56,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8514, 3.7013, 3.5382, 2.0003], device='cuda:1'), covar=tensor([0.0709, 0.0825, 0.0878, 0.2271], device='cuda:1'), in_proj_covar=tensor([0.1280, 0.1177, 0.0994, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 14:59:03,956 INFO [train.py:968] (1/2) Epoch 26, batch 31950, giga_loss[loss=0.2575, simple_loss=0.3376, pruned_loss=0.08866, over 28201.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3353, pruned_loss=0.08747, over 5668364.02 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3492, pruned_loss=0.1105, over 5692684.17 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3347, pruned_loss=0.08527, over 5667027.50 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 14:59:53,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1171045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:00:13,441 INFO [train.py:968] (1/2) Epoch 26, batch 32000, giga_loss[loss=0.2116, simple_loss=0.2848, pruned_loss=0.06921, over 24190.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3326, pruned_loss=0.08635, over 5651847.81 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3494, pruned_loss=0.1107, over 5684266.66 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3318, pruned_loss=0.08413, over 5657190.23 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:00:23,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1171066.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:00:31,233 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 32050, giga_loss[loss=0.2808, simple_loss=0.3648, pruned_loss=0.09843, over 28923.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3326, pruned_loss=0.08651, over 5657901.74 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3487, pruned_loss=0.1102, over 5686461.74 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3321, pruned_loss=0.08459, over 5659680.99 frames. ], batch size: 227, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:02:16,523 INFO [train.py:968] (1/2) Epoch 26, batch 32100, giga_loss[loss=0.3012, simple_loss=0.3777, pruned_loss=0.1123, over 28119.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3366, pruned_loss=0.08857, over 5662066.80 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3486, pruned_loss=0.1101, over 5680615.73 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.336, pruned_loss=0.08643, over 5667547.48 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:02:36,777 INFO [optim.py:369] (1/2) 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 15:02:50,379 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1171188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:02:54,381 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1171191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:03:16,166 INFO [train.py:968] (1/2) Epoch 26, batch 32150, libri_loss[loss=0.2549, simple_loss=0.3293, pruned_loss=0.09019, over 29523.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3349, pruned_loss=0.0886, over 5659723.77 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3479, pruned_loss=0.1098, over 5678269.53 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3346, pruned_loss=0.0865, over 5665901.35 frames. ], batch size: 83, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:03:18,556 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1171209.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:03:23,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1171212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:03:35,579 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:04:00,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171241.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:04:12,537 INFO [train.py:968] (1/2) Epoch 26, batch 32200, giga_loss[loss=0.2693, simple_loss=0.3461, pruned_loss=0.09623, over 28926.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3352, pruned_loss=0.08989, over 5664140.61 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.348, pruned_loss=0.1099, over 5681774.62 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3344, pruned_loss=0.08741, over 5664935.15 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:04:32,808 INFO [optim.py:369] (1/2) 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 15:04:41,307 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-13 15:04:42,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-13 15:04:51,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5846, 1.6804, 1.7962, 1.3886], device='cuda:1'), covar=tensor([0.1990, 0.2720, 0.1627, 0.1994], device='cuda:1'), in_proj_covar=tensor([0.0919, 0.0704, 0.0967, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 15:05:12,974 INFO [train.py:968] (1/2) Epoch 26, batch 32250, libri_loss[loss=0.2854, simple_loss=0.3475, pruned_loss=0.1117, over 29493.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3356, pruned_loss=0.09022, over 5666387.78 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.348, pruned_loss=0.11, over 5685173.02 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3348, pruned_loss=0.0878, over 5663584.05 frames. ], batch size: 85, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:05:49,335 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 15:06:19,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2543, 3.1142, 2.9342, 1.5241], device='cuda:1'), covar=tensor([0.1013, 0.1048, 0.0987, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.1277, 0.1175, 0.0992, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 15:06:23,846 INFO [train.py:968] (1/2) Epoch 26, batch 32300, giga_loss[loss=0.2839, simple_loss=0.3673, pruned_loss=0.1003, over 28737.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3381, pruned_loss=0.09111, over 5665813.11 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3483, pruned_loss=0.1102, over 5686290.96 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3369, pruned_loss=0.08848, over 5661801.74 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:06:50,793 INFO [optim.py:369] (1/2) 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 15:06:58,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7107, 1.8328, 1.3077, 1.4931], device='cuda:1'), covar=tensor([0.0942, 0.0572, 0.0986, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0401, 0.0445, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 15:07:39,662 INFO [train.py:968] (1/2) Epoch 26, batch 32350, giga_loss[loss=0.2305, simple_loss=0.3005, pruned_loss=0.08026, over 24307.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3376, pruned_loss=0.08962, over 5667434.71 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3477, pruned_loss=0.1099, over 5689723.99 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3371, pruned_loss=0.08757, over 5660719.96 frames. ], batch size: 705, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:07:53,087 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1171415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:08:46,103 INFO [train.py:968] (1/2) Epoch 26, batch 32400, giga_loss[loss=0.2261, simple_loss=0.3025, pruned_loss=0.0749, over 28903.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3355, pruned_loss=0.08889, over 5675237.24 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3476, pruned_loss=0.1101, over 5694522.07 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3349, pruned_loss=0.08643, over 5664901.84 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:09:09,942 INFO [optim.py:369] (1/2) 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 15:09:50,040 INFO [train.py:968] (1/2) Epoch 26, batch 32450, libri_loss[loss=0.2965, simple_loss=0.3611, pruned_loss=0.116, over 29662.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3301, pruned_loss=0.08682, over 5683327.21 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3471, pruned_loss=0.1099, over 5700893.62 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3296, pruned_loss=0.08439, over 5668347.28 frames. ], batch size: 88, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:10:36,529 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4346, 1.3242, 3.3954, 3.2634], device='cuda:1'), covar=tensor([0.1343, 0.2852, 0.0479, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0669, 0.0990, 0.0952], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 15:10:50,282 INFO [train.py:968] (1/2) Epoch 26, batch 32500, libri_loss[loss=0.2432, simple_loss=0.3065, pruned_loss=0.08997, over 29358.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.327, pruned_loss=0.08628, over 5676471.83 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3469, pruned_loss=0.11, over 5706304.62 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.326, pruned_loss=0.08331, over 5658628.08 frames. ], batch size: 71, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:11:11,494 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 32550, giga_loss[loss=0.2442, simple_loss=0.3219, pruned_loss=0.08323, over 28012.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.328, pruned_loss=0.08705, over 5664379.91 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3468, pruned_loss=0.1099, over 5699684.66 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3268, pruned_loss=0.08403, over 5654201.22 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:11:52,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5460, 1.8876, 1.8898, 1.4880], device='cuda:1'), covar=tensor([0.2858, 0.2110, 0.2292, 0.2701], device='cuda:1'), in_proj_covar=tensor([0.2002, 0.1943, 0.1861, 0.1998], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 15:12:45,003 INFO [train.py:968] (1/2) Epoch 26, batch 32600, giga_loss[loss=0.2358, simple_loss=0.3201, pruned_loss=0.07576, over 28459.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3287, pruned_loss=0.0874, over 5664287.46 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3467, pruned_loss=0.1098, over 5704354.01 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3274, pruned_loss=0.08461, over 5651068.52 frames. ], batch size: 369, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:13:05,702 INFO [optim.py:369] (1/2) 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 15:13:45,128 INFO [train.py:968] (1/2) Epoch 26, batch 32650, giga_loss[loss=0.2523, simple_loss=0.3358, pruned_loss=0.08443, over 28945.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3274, pruned_loss=0.08596, over 5666070.51 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3467, pruned_loss=0.1097, over 5708339.94 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.326, pruned_loss=0.08334, over 5651221.23 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:14:43,258 INFO [train.py:968] (1/2) Epoch 26, batch 32700, giga_loss[loss=0.2466, simple_loss=0.3231, pruned_loss=0.085, over 28946.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3263, pruned_loss=0.08505, over 5676280.31 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3463, pruned_loss=0.1095, over 5714686.35 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3248, pruned_loss=0.08217, over 5656962.32 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:14:48,166 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 15:14:51,252 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2873, 1.2414, 1.2386, 1.5039], device='cuda:1'), covar=tensor([0.0780, 0.0396, 0.0360, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0073, 0.0065, 0.0114], device='cuda:1')
+2023-03-13 15:15:04,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1898, 1.1119, 1.1528, 1.3252], device='cuda:1'), covar=tensor([0.0768, 0.0420, 0.0328, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0073, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 15:15:09,130 INFO [optim.py:369] (1/2) 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 15:15:29,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1171790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:15:51,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4014, 4.2825, 4.0157, 1.9788], device='cuda:1'), covar=tensor([0.0560, 0.0652, 0.0716, 0.2030], device='cuda:1'), in_proj_covar=tensor([0.1273, 0.1169, 0.0987, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 15:15:51,375 INFO [train.py:968] (1/2) Epoch 26, batch 32750, giga_loss[loss=0.2249, simple_loss=0.3127, pruned_loss=0.06854, over 28673.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3255, pruned_loss=0.08499, over 5669458.65 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3461, pruned_loss=0.1095, over 5716613.16 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.324, pruned_loss=0.08229, over 5651489.55 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:16:50,667 INFO [train.py:968] (1/2) Epoch 26, batch 32800, giga_loss[loss=0.2859, simple_loss=0.3688, pruned_loss=0.1015, over 28443.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3274, pruned_loss=0.08581, over 5672105.12 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3469, pruned_loss=0.1099, over 5720140.56 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3246, pruned_loss=0.08213, over 5651986.19 frames. ], batch size: 369, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:17:10,460 INFO [optim.py:369] (1/2) 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 15:17:56,305 INFO [train.py:968] (1/2) Epoch 26, batch 32850, giga_loss[loss=0.2174, simple_loss=0.2966, pruned_loss=0.06912, over 28947.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3263, pruned_loss=0.08527, over 5672846.59 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3465, pruned_loss=0.1098, over 5724781.19 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.324, pruned_loss=0.0819, over 5651965.32 frames. ], batch size: 106, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:18:29,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1171933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:18:31,869 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1171936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:18:55,508 INFO [train.py:968] (1/2) Epoch 26, batch 32900, giga_loss[loss=0.2169, simple_loss=0.3036, pruned_loss=0.06509, over 28646.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3274, pruned_loss=0.08671, over 5666466.38 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3465, pruned_loss=0.11, over 5718471.83 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.325, pruned_loss=0.08317, over 5654777.04 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:19:01,029 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1171962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:19:04,930 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1171965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:19:16,405 INFO [optim.py:369] (1/2) 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 15:19:35,073 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 15:19:58,785 INFO [train.py:968] (1/2) Epoch 26, batch 32950, giga_loss[loss=0.2556, simple_loss=0.3524, pruned_loss=0.07944, over 28956.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3245, pruned_loss=0.08403, over 5662710.66 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3462, pruned_loss=0.1099, over 5720402.73 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3226, pruned_loss=0.08101, over 5651225.21 frames. ], batch size: 186, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:20:53,276 INFO [train.py:968] (1/2) Epoch 26, batch 33000, giga_loss[loss=0.2488, simple_loss=0.3503, pruned_loss=0.07367, over 28905.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3269, pruned_loss=0.08366, over 5660885.79 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3457, pruned_loss=0.1095, over 5714723.42 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3252, pruned_loss=0.08086, over 5655692.69 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:20:53,276 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 15:21:02,129 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 15:21:22,346 INFO [optim.py:369] (1/2) 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 15:21:48,647 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 15:21:54,824 INFO [train.py:968] (1/2) Epoch 26, batch 33050, giga_loss[loss=0.2716, simple_loss=0.3481, pruned_loss=0.09753, over 28946.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.08497, over 5655704.35 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3453, pruned_loss=0.1094, over 5711037.70 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3273, pruned_loss=0.08142, over 5652812.88 frames. ], batch size: 285, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:22:48,967 INFO [train.py:968] (1/2) Epoch 26, batch 33100, giga_loss[loss=0.2644, simple_loss=0.3409, pruned_loss=0.09394, over 27836.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3312, pruned_loss=0.08652, over 5640901.42 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3453, pruned_loss=0.1095, over 5699443.77 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3289, pruned_loss=0.08246, over 5646077.57 frames. ], batch size: 476, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:23:16,695 INFO [optim.py:369] (1/2) 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 15:23:52,729 INFO [train.py:968] (1/2) Epoch 26, batch 33150, giga_loss[loss=0.2317, simple_loss=0.3152, pruned_loss=0.07415, over 28608.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3317, pruned_loss=0.08714, over 5652767.57 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3457, pruned_loss=0.1099, over 5704291.25 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3291, pruned_loss=0.08294, over 5651165.78 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:24:01,551 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7299, 2.0798, 2.0643, 1.6759], device='cuda:1'), covar=tensor([0.2239, 0.2518, 0.2016, 0.2546], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0740, 0.0710, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 15:24:46,126 INFO [train.py:968] (1/2) Epoch 26, batch 33200, giga_loss[loss=0.2504, simple_loss=0.3343, pruned_loss=0.08319, over 28094.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3306, pruned_loss=0.08644, over 5660100.07 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3455, pruned_loss=0.1098, over 5707198.89 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3282, pruned_loss=0.0824, over 5655029.54 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:25:08,230 INFO [optim.py:369] (1/2) 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 15:25:46,704 INFO [train.py:968] (1/2) Epoch 26, batch 33250, giga_loss[loss=0.2269, simple_loss=0.3142, pruned_loss=0.0698, over 28960.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3285, pruned_loss=0.08503, over 5659528.58 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3455, pruned_loss=0.1098, over 5711823.07 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3262, pruned_loss=0.08123, over 5650360.99 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:26:19,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1172337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:26:26,143 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0962, 1.5188, 1.4959, 1.3638], device='cuda:1'), covar=tensor([0.2075, 0.1618, 0.2078, 0.1767], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0741, 0.0711, 0.0683], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 15:26:40,759 INFO [train.py:968] (1/2) Epoch 26, batch 33300, libri_loss[loss=0.2521, simple_loss=0.3149, pruned_loss=0.09468, over 29587.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3259, pruned_loss=0.08443, over 5667367.18 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.345, pruned_loss=0.1095, over 5715938.70 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3241, pruned_loss=0.081, over 5655029.19 frames. ], batch size: 74, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:26:50,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8157, 2.1294, 2.2076, 1.7985], device='cuda:1'), covar=tensor([0.3177, 0.2248, 0.2245, 0.2663], device='cuda:1'), in_proj_covar=tensor([0.2003, 0.1943, 0.1857, 0.1991], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 15:27:06,594 INFO [optim.py:369] (1/2) 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 15:27:40,572 INFO [train.py:968] (1/2) Epoch 26, batch 33350, libri_loss[loss=0.2818, simple_loss=0.3389, pruned_loss=0.1124, over 29596.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3274, pruned_loss=0.0848, over 5675951.01 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3449, pruned_loss=0.1094, over 5717128.56 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3256, pruned_loss=0.08155, over 5663933.53 frames. ], batch size: 73, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:28:04,975 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1172426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:28:11,628 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1172431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:28:36,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9297, 1.1950, 1.1012, 0.9138], device='cuda:1'), covar=tensor([0.2486, 0.2634, 0.1666, 0.2232], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1947, 0.1857, 0.1994], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 15:28:43,530 INFO [train.py:968] (1/2) Epoch 26, batch 33400, giga_loss[loss=0.2408, simple_loss=0.3233, pruned_loss=0.07916, over 28633.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3298, pruned_loss=0.08578, over 5676010.68 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3445, pruned_loss=0.1091, over 5720933.54 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3285, pruned_loss=0.083, over 5662316.38 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:29:06,308 INFO [optim.py:369] (1/2) 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 15:29:11,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1172480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:29:18,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1172483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:29:45,024 INFO [train.py:968] (1/2) Epoch 26, batch 33450, giga_loss[loss=0.263, simple_loss=0.3413, pruned_loss=0.09235, over 28925.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3295, pruned_loss=0.08599, over 5671712.47 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.344, pruned_loss=0.1088, over 5723628.13 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3285, pruned_loss=0.08349, over 5657286.69 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:29:54,818 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1172512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:30:27,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9083, 2.0287, 1.4753, 1.6467], device='cuda:1'), covar=tensor([0.1035, 0.0699, 0.1007, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0445, 0.0519, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 15:30:47,402 INFO [train.py:968] (1/2) Epoch 26, batch 33500, giga_loss[loss=0.2954, simple_loss=0.3758, pruned_loss=0.1075, over 28702.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3318, pruned_loss=0.08715, over 5669086.32 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3437, pruned_loss=0.1086, over 5717670.36 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3309, pruned_loss=0.08449, over 5661739.85 frames. ], batch size: 262, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:31:10,640 INFO [optim.py:369] (1/2) 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 15:31:40,361 INFO [train.py:968] (1/2) Epoch 26, batch 33550, giga_loss[loss=0.2549, simple_loss=0.3447, pruned_loss=0.08252, over 28645.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3339, pruned_loss=0.08766, over 5664966.68 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3438, pruned_loss=0.1088, over 5718519.50 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3327, pruned_loss=0.08464, over 5656831.97 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:31:48,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5403, 1.7248, 1.7395, 1.3399], device='cuda:1'), covar=tensor([0.1695, 0.2713, 0.1484, 0.1776], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0703, 0.0969, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 15:32:39,919 INFO [train.py:968] (1/2) Epoch 26, batch 33600, giga_loss[loss=0.3032, simple_loss=0.3794, pruned_loss=0.1135, over 28020.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3341, pruned_loss=0.08748, over 5660165.87 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3434, pruned_loss=0.1085, over 5711911.20 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3332, pruned_loss=0.08467, over 5657713.84 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:32:59,880 INFO [optim.py:369] (1/2) 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 15:33:42,513 INFO [train.py:968] (1/2) Epoch 26, batch 33650, giga_loss[loss=0.2541, simple_loss=0.3426, pruned_loss=0.08283, over 28943.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.08725, over 5666443.17 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3422, pruned_loss=0.1077, over 5718737.74 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.333, pruned_loss=0.08506, over 5656989.88 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:33:51,998 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1172712.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:34:03,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3632, 1.4448, 1.3323, 1.6183], device='cuda:1'), covar=tensor([0.0658, 0.0311, 0.0327, 0.0737], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0120, 0.0120, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0073, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 15:34:25,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3635, 1.2960, 3.8852, 3.2404], device='cuda:1'), covar=tensor([0.1688, 0.2975, 0.0476, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0672, 0.0993, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 15:34:25,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4680, 1.6775, 1.6923, 1.4245], device='cuda:1'), covar=tensor([0.3237, 0.2636, 0.2224, 0.2828], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1947, 0.1855, 0.1996], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 15:34:49,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6433, 2.1358, 2.0850, 1.5793], device='cuda:1'), covar=tensor([0.3220, 0.2231, 0.2372, 0.2963], device='cuda:1'), in_proj_covar=tensor([0.2007, 0.1948, 0.1856, 0.1997], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 15:34:51,908 INFO [train.py:968] (1/2) Epoch 26, batch 33700, giga_loss[loss=0.2099, simple_loss=0.3011, pruned_loss=0.05935, over 28881.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3332, pruned_loss=0.08826, over 5666936.77 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3425, pruned_loss=0.1079, over 5720714.58 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3329, pruned_loss=0.08597, over 5656682.92 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:35:15,659 INFO [optim.py:369] (1/2) 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 15:35:46,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1172801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:35:53,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1172806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:35:53,397 INFO [train.py:968] (1/2) Epoch 26, batch 33750, giga_loss[loss=0.2328, simple_loss=0.3145, pruned_loss=0.0755, over 28939.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3321, pruned_loss=0.08756, over 5668335.41 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3422, pruned_loss=0.108, over 5727273.98 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3316, pruned_loss=0.08494, over 5652534.43 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:36:51,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-13 15:36:58,337 INFO [train.py:968] (1/2) Epoch 26, batch 33800, giga_loss[loss=0.2469, simple_loss=0.3227, pruned_loss=0.08554, over 29000.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3316, pruned_loss=0.08799, over 5672402.62 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3424, pruned_loss=0.1081, over 5731147.20 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.331, pruned_loss=0.08545, over 5655142.17 frames. ], batch size: 199, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:37:28,515 INFO [optim.py:369] (1/2) 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 15:38:03,434 INFO [train.py:968] (1/2) Epoch 26, batch 33850, giga_loss[loss=0.2227, simple_loss=0.3115, pruned_loss=0.06696, over 28887.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.331, pruned_loss=0.08871, over 5650637.51 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3429, pruned_loss=0.1085, over 5720445.48 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3298, pruned_loss=0.08597, over 5644887.75 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:38:47,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1172944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:50,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1172947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:53,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1172949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:55,609 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1172952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:38:59,771 INFO [train.py:968] (1/2) Epoch 26, batch 33900, giga_loss[loss=0.2298, simple_loss=0.3167, pruned_loss=0.07146, over 28818.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3302, pruned_loss=0.08741, over 5648012.81 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.343, pruned_loss=0.1087, over 5715520.57 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3288, pruned_loss=0.08453, over 5645971.70 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:39:23,368 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1172976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:39:25,751 INFO [optim.py:369] (1/2) 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 15:39:29,835 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1172981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:39:52,269 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:40:00,048 INFO [train.py:968] (1/2) Epoch 26, batch 33950, giga_loss[loss=0.2466, simple_loss=0.3401, pruned_loss=0.07651, over 29042.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3289, pruned_loss=0.08543, over 5652378.14 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.343, pruned_loss=0.1088, over 5705629.38 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3274, pruned_loss=0.08237, over 5657672.74 frames. ], batch size: 200, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:40:36,865 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-13 15:40:55,927 INFO [train.py:968] (1/2) Epoch 26, batch 34000, giga_loss[loss=0.2348, simple_loss=0.3213, pruned_loss=0.0742, over 29007.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3313, pruned_loss=0.08455, over 5662574.94 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3431, pruned_loss=0.1088, over 5705756.16 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3298, pruned_loss=0.08168, over 5666110.82 frames. ], batch size: 145, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:41:08,215 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:41:21,023 INFO [optim.py:369] (1/2) 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 15:41:28,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1173087.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:41:49,986 INFO [train.py:968] (1/2) Epoch 26, batch 34050, giga_loss[loss=0.2197, simple_loss=0.3087, pruned_loss=0.06535, over 29038.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3313, pruned_loss=0.08409, over 5663813.27 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3428, pruned_loss=0.1086, over 5711686.97 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3301, pruned_loss=0.08126, over 5659951.23 frames. ], batch size: 136, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:42:18,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-13 15:42:44,570 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9818, 1.2098, 2.8143, 2.7309], device='cuda:1'), covar=tensor([0.1574, 0.2624, 0.0589, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0668, 0.0984, 0.0951], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 15:42:52,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5029, 1.8467, 1.4370, 1.5096], device='cuda:1'), covar=tensor([0.2851, 0.2808, 0.3299, 0.2515], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1135, 0.1400, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 15:42:56,201 INFO [train.py:968] (1/2) Epoch 26, batch 34100, giga_loss[loss=0.2568, simple_loss=0.3381, pruned_loss=0.0877, over 28143.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3313, pruned_loss=0.08396, over 5669058.94 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3423, pruned_loss=0.1083, over 5715277.50 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3304, pruned_loss=0.0813, over 5661449.56 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:43:02,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4351, 1.6289, 1.6487, 1.2716], device='cuda:1'), covar=tensor([0.1916, 0.2802, 0.1601, 0.1916], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0705, 0.0971, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:1')
+2023-03-13 15:43:22,392 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.34 vs. limit=2.0
+2023-03-13 15:43:25,115 INFO [optim.py:369] (1/2) 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 15:44:00,113 INFO [train.py:968] (1/2) Epoch 26, batch 34150, giga_loss[loss=0.2606, simple_loss=0.3395, pruned_loss=0.0908, over 28073.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3319, pruned_loss=0.08423, over 5676220.19 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3424, pruned_loss=0.1083, over 5718750.61 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3308, pruned_loss=0.08146, over 5665785.90 frames. ], batch size: 412, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:44:30,464 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1173230.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:44:33,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1173233.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 15:45:03,595 INFO [train.py:968] (1/2) Epoch 26, batch 34200, giga_loss[loss=0.2418, simple_loss=0.3281, pruned_loss=0.07773, over 28306.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3324, pruned_loss=0.08501, over 5679413.15 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3423, pruned_loss=0.1082, over 5724614.30 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3311, pruned_loss=0.08175, over 5663727.32 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:45:11,714 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1173262.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 15:45:11,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-13 15:45:34,453 INFO [optim.py:369] (1/2) 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 15:46:14,188 INFO [train.py:968] (1/2) Epoch 26, batch 34250, giga_loss[loss=0.2342, simple_loss=0.3259, pruned_loss=0.07132, over 28543.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3323, pruned_loss=0.08439, over 5679071.28 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3421, pruned_loss=0.108, over 5729762.36 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3313, pruned_loss=0.08132, over 5660436.89 frames. ], batch size: 336, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:47:00,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-13 15:47:11,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8540, 1.1565, 1.1350, 0.8319], device='cuda:1'), covar=tensor([0.2405, 0.2300, 0.1528, 0.2222], device='cuda:1'), in_proj_covar=tensor([0.2002, 0.1943, 0.1848, 0.1993], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 15:47:15,703 INFO [train.py:968] (1/2) Epoch 26, batch 34300, giga_loss[loss=0.2556, simple_loss=0.3469, pruned_loss=0.0822, over 28899.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3335, pruned_loss=0.08515, over 5673385.78 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.342, pruned_loss=0.108, over 5732340.91 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3324, pruned_loss=0.08196, over 5654680.33 frames. ], batch size: 174, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:47:37,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1173376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:47:40,637 INFO [optim.py:369] (1/2) 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 15:48:19,519 INFO [train.py:968] (1/2) Epoch 26, batch 34350, giga_loss[loss=0.2424, simple_loss=0.3316, pruned_loss=0.0766, over 28976.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3368, pruned_loss=0.08615, over 5682880.16 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.342, pruned_loss=0.1079, over 5734689.16 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3358, pruned_loss=0.08322, over 5664919.65 frames. ], batch size: 213, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:49:06,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1173442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:49:23,356 INFO [train.py:968] (1/2) Epoch 26, batch 34400, giga_loss[loss=0.2504, simple_loss=0.3334, pruned_loss=0.08368, over 28687.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3373, pruned_loss=0.087, over 5690123.53 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3423, pruned_loss=0.1081, over 5736166.56 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3361, pruned_loss=0.0839, over 5673137.35 frames. ], batch size: 307, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:49:50,949 INFO [optim.py:369] (1/2) 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 15:50:31,411 INFO [train.py:968] (1/2) Epoch 26, batch 34450, libri_loss[loss=0.2714, simple_loss=0.3434, pruned_loss=0.09975, over 28733.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3349, pruned_loss=0.08609, over 5682408.45 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3425, pruned_loss=0.1082, over 5727218.73 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3337, pruned_loss=0.0832, over 5675392.16 frames. ], batch size: 107, lr: 1.22e-03, grad_scale: 8.0
+2023-03-13 15:50:48,579 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1173519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:50:52,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1173522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:51:31,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1173551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:51:36,509 INFO [train.py:968] (1/2) Epoch 26, batch 34500, giga_loss[loss=0.2249, simple_loss=0.3166, pruned_loss=0.0666, over 28830.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.334, pruned_loss=0.08512, over 5679175.51 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3426, pruned_loss=0.1081, over 5717815.59 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3326, pruned_loss=0.08212, over 5681160.62 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:52:03,637 INFO [optim.py:369] (1/2) 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 15:52:08,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1173585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:52:11,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1173588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:52:27,337 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 15:52:37,971 INFO [train.py:968] (1/2) Epoch 26, batch 34550, giga_loss[loss=0.2453, simple_loss=0.3297, pruned_loss=0.08043, over 28795.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.332, pruned_loss=0.08387, over 5690085.34 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3424, pruned_loss=0.1081, over 5721305.15 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3309, pruned_loss=0.08106, over 5688024.12 frames. ], batch size: 243, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:52:48,724 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1173617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:53:02,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3813, 4.2199, 4.0429, 1.8524], device='cuda:1'), covar=tensor([0.0638, 0.0768, 0.0849, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1165, 0.0985, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 15:53:26,657 INFO [train.py:968] (1/2) Epoch 26, batch 34600, giga_loss[loss=0.2892, simple_loss=0.3752, pruned_loss=0.1016, over 28716.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3324, pruned_loss=0.08475, over 5676441.46 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3421, pruned_loss=0.1079, over 5712829.57 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3312, pruned_loss=0.08125, over 5681424.87 frames. ], batch size: 242, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:53:39,396 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:53:54,861 INFO [optim.py:369] (1/2) 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 15:54:09,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4048, 1.6005, 1.6292, 1.4757], device='cuda:1'), covar=tensor([0.1386, 0.1143, 0.1470, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0475, 0.0728, 0.0701, 0.0672], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0009, 0.0009, 0.0009], device='cuda:1')
+2023-03-13 15:54:26,930 INFO [train.py:968] (1/2) Epoch 26, batch 34650, libri_loss[loss=0.2965, simple_loss=0.344, pruned_loss=0.1245, over 29577.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3349, pruned_loss=0.08604, over 5666090.30 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3421, pruned_loss=0.108, over 5707807.49 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3337, pruned_loss=0.08267, over 5674037.12 frames. ], batch size: 76, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:54:57,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:55:26,869 INFO [train.py:968] (1/2) Epoch 26, batch 34700, giga_loss[loss=0.219, simple_loss=0.2999, pruned_loss=0.06898, over 28334.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3348, pruned_loss=0.08666, over 5653883.06 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3421, pruned_loss=0.1079, over 5700031.91 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3337, pruned_loss=0.08366, over 5666154.67 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:55:55,254 INFO [optim.py:369] (1/2) 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 15:56:25,218 INFO [train.py:968] (1/2) Epoch 26, batch 34750, giga_loss[loss=0.2656, simple_loss=0.3347, pruned_loss=0.09819, over 26819.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3319, pruned_loss=0.0858, over 5659608.74 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3419, pruned_loss=0.1078, over 5702289.40 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3311, pruned_loss=0.08334, over 5666939.68 frames. ], batch size: 555, lr: 1.22e-03, grad_scale: 2.0
+2023-03-13 15:56:28,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 15:56:44,478 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1173823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 15:57:14,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.84 vs. limit=5.0
+2023-03-13 15:57:22,399 INFO [train.py:968] (1/2) Epoch 26, batch 34800, giga_loss[loss=0.2628, simple_loss=0.3442, pruned_loss=0.09075, over 28338.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3324, pruned_loss=0.08671, over 5653408.54 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3422, pruned_loss=0.108, over 5696407.71 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3313, pruned_loss=0.08396, over 5663152.73 frames. ], batch size: 368, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:57:46,820 INFO [optim.py:369] (1/2) 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 15:58:11,219 INFO [train.py:968] (1/2) Epoch 26, batch 34850, giga_loss[loss=0.2638, simple_loss=0.3495, pruned_loss=0.08912, over 28217.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3395, pruned_loss=0.09113, over 5646099.40 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3424, pruned_loss=0.1083, over 5688997.23 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3382, pruned_loss=0.08823, over 5660410.80 frames. ], batch size: 65, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:58:54,455 INFO [train.py:968] (1/2) Epoch 26, batch 34900, giga_loss[loss=0.2974, simple_loss=0.3824, pruned_loss=0.1063, over 28796.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3486, pruned_loss=0.09666, over 5646321.98 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3424, pruned_loss=0.1084, over 5675920.14 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3476, pruned_loss=0.09384, over 5669855.66 frames. ], batch size: 284, lr: 1.22e-03, grad_scale: 4.0
+2023-03-13 15:59:14,311 INFO [optim.py:369] (1/2) 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 15:59:26,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5447, 1.7325, 1.7359, 1.3268], device='cuda:1'), covar=tensor([0.1562, 0.2483, 0.1424, 0.1681], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0700, 0.0968, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 15:59:37,622 INFO [train.py:968] (1/2) Epoch 26, batch 34950, giga_loss[loss=0.2285, simple_loss=0.3073, pruned_loss=0.07491, over 28753.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3505, pruned_loss=0.09773, over 5656312.26 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3426, pruned_loss=0.1084, over 5677781.13 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3497, pruned_loss=0.0951, over 5672533.05 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 15:59:44,077 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2091, 1.2997, 1.1898, 0.9423], device='cuda:1'), covar=tensor([0.1069, 0.0553, 0.1091, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0403, 0.0444, 0.0518, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 16:00:07,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:00:19,328 INFO [train.py:968] (1/2) Epoch 26, batch 35000, giga_loss[loss=0.2426, simple_loss=0.3205, pruned_loss=0.08241, over 28479.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.347, pruned_loss=0.09644, over 5667753.97 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3425, pruned_loss=0.1082, over 5682684.84 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3466, pruned_loss=0.09433, over 5675865.87 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:00:38,353 INFO [optim.py:369] (1/2) 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 16:00:55,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:00:57,567 INFO [train.py:968] (1/2) Epoch 26, batch 35050, giga_loss[loss=0.2571, simple_loss=0.3304, pruned_loss=0.0919, over 28250.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3414, pruned_loss=0.0946, over 5672003.46 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3435, pruned_loss=0.1087, over 5687059.74 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3402, pruned_loss=0.0919, over 5674134.07 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:01:40,144 INFO [train.py:968] (1/2) Epoch 26, batch 35100, giga_loss[loss=0.2155, simple_loss=0.298, pruned_loss=0.06648, over 28733.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3339, pruned_loss=0.09129, over 5683842.50 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3432, pruned_loss=0.1084, over 5693376.50 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3331, pruned_loss=0.08891, over 5679565.75 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:01:55,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-13 16:02:01,622 INFO [optim.py:369] (1/2) 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 16:02:04,109 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:06,973 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:15,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:24,187 INFO [train.py:968] (1/2) Epoch 26, batch 35150, giga_loss[loss=0.2152, simple_loss=0.2892, pruned_loss=0.07055, over 28948.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3259, pruned_loss=0.08752, over 5681396.01 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3436, pruned_loss=0.1086, over 5694811.42 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3248, pruned_loss=0.08525, over 5676490.71 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:02:30,164 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:32,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:38,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3028, 1.1758, 3.9640, 3.2407], device='cuda:1'), covar=tensor([0.1626, 0.2809, 0.0498, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0671, 0.0991, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 16:02:53,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:57,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:02:59,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4949, 1.6945, 1.7147, 1.5737], device='cuda:1'), covar=tensor([0.2217, 0.2332, 0.2595, 0.2263], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0734, 0.0706, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0009, 0.0009], device='cuda:1')
+2023-03-13 16:03:02,933 INFO [train.py:968] (1/2) Epoch 26, batch 35200, giga_loss[loss=0.2076, simple_loss=0.2941, pruned_loss=0.06053, over 29076.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.32, pruned_loss=0.08478, over 5684908.61 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3439, pruned_loss=0.1086, over 5694997.58 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3182, pruned_loss=0.08234, over 5680349.96 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:03:22,252 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 26, batch 35250, giga_loss[loss=0.2067, simple_loss=0.2822, pruned_loss=0.06563, over 28996.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3163, pruned_loss=0.08345, over 5694494.56 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3439, pruned_loss=0.1085, over 5699244.48 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3145, pruned_loss=0.08116, over 5687027.29 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:04:00,196 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:04:16,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:04:18,195 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 35300, giga_loss[loss=0.2626, simple_loss=0.3177, pruned_loss=0.1037, over 26530.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3134, pruned_loss=0.08244, over 5684551.48 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3441, pruned_loss=0.1086, over 5692171.68 frames. ], giga_tot_loss[loss=0.2357, simple_loss=0.3112, pruned_loss=0.08007, over 5684662.94 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:04:42,502 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:04:43,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4208, 1.6601, 1.6674, 1.5205], device='cuda:1'), covar=tensor([0.2181, 0.2076, 0.2527, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.0478, 0.0734, 0.0706, 0.0676], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 16:04:49,795 INFO [optim.py:369] (1/2) 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 16:04:56,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:05:12,119 INFO [train.py:968] (1/2) Epoch 26, batch 35350, giga_loss[loss=0.2091, simple_loss=0.2781, pruned_loss=0.07003, over 28629.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.31, pruned_loss=0.08064, over 5684278.96 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3442, pruned_loss=0.1085, over 5696581.97 frames. ], giga_tot_loss[loss=0.2322, simple_loss=0.3076, pruned_loss=0.0784, over 5680428.52 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:05:14,527 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 16:05:53,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4680, 1.9091, 1.4863, 1.3993], device='cuda:1'), covar=tensor([0.2675, 0.2671, 0.3247, 0.2442], device='cuda:1'), in_proj_covar=tensor([0.1577, 0.1133, 0.1394, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 16:05:56,451 INFO [train.py:968] (1/2) Epoch 26, batch 35400, giga_loss[loss=0.2123, simple_loss=0.2889, pruned_loss=0.06784, over 28891.00 frames. ], tot_loss[loss=0.2347, simple_loss=0.3086, pruned_loss=0.08036, over 5679747.94 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3447, pruned_loss=0.1086, over 5701397.00 frames. ], giga_tot_loss[loss=0.2303, simple_loss=0.3053, pruned_loss=0.07769, over 5671859.06 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:06:15,398 INFO [optim.py:369] (1/2) 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 16:06:36,842 INFO [train.py:968] (1/2) Epoch 26, batch 35450, giga_loss[loss=0.1834, simple_loss=0.2685, pruned_loss=0.04914, over 28901.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3053, pruned_loss=0.07871, over 5686591.64 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3445, pruned_loss=0.1083, over 5701957.32 frames. ], giga_tot_loss[loss=0.2272, simple_loss=0.302, pruned_loss=0.07618, over 5679481.50 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:07:02,785 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:07:19,124 INFO [train.py:968] (1/2) Epoch 26, batch 35500, giga_loss[loss=0.2486, simple_loss=0.3052, pruned_loss=0.09598, over 27631.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.3032, pruned_loss=0.07791, over 5691629.42 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3445, pruned_loss=0.1083, over 5705907.42 frames. ], giga_tot_loss[loss=0.2254, simple_loss=0.2999, pruned_loss=0.07543, over 5682345.28 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:07:39,670 INFO [optim.py:369] (1/2) 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 16:07:42,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1579, 1.2792, 3.3188, 2.8767], device='cuda:1'), covar=tensor([0.1650, 0.2758, 0.0531, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0669, 0.0994, 0.0956], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 16:07:52,119 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:08:01,132 INFO [train.py:968] (1/2) Epoch 26, batch 35550, giga_loss[loss=0.1866, simple_loss=0.2645, pruned_loss=0.05432, over 28821.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.3004, pruned_loss=0.0766, over 5692305.18 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.345, pruned_loss=0.1085, over 5704408.72 frames. ], giga_tot_loss[loss=0.2222, simple_loss=0.2965, pruned_loss=0.07392, over 5686212.46 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:08:11,967 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:08:12,775 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:08:30,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3610, 1.6982, 1.3669, 0.9043], device='cuda:1'), covar=tensor([0.2681, 0.2776, 0.3211, 0.2549], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1135, 0.1396, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 16:08:47,410 INFO [train.py:968] (1/2) Epoch 26, batch 35600, giga_loss[loss=0.1894, simple_loss=0.266, pruned_loss=0.05634, over 28569.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2976, pruned_loss=0.0755, over 5680438.99 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3448, pruned_loss=0.1081, over 5705857.82 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2936, pruned_loss=0.07292, over 5673833.90 frames. ], batch size: 60, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:09:01,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-13 16:09:06,479 INFO [optim.py:369] (1/2) 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 16:09:12,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5865, 4.4017, 4.1727, 2.0442], device='cuda:1'), covar=tensor([0.0502, 0.0725, 0.0727, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.1175, 0.0989, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 16:09:15,117 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-13 16:09:21,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:09:27,647 INFO [train.py:968] (1/2) Epoch 26, batch 35650, giga_loss[loss=0.214, simple_loss=0.2842, pruned_loss=0.07185, over 29064.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2983, pruned_loss=0.07654, over 5685110.53 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3452, pruned_loss=0.1084, over 5707669.41 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2932, pruned_loss=0.07319, over 5677352.30 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:09:53,581 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 16:09:56,974 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:09:58,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8053, 1.9963, 2.2468, 1.6851], device='cuda:1'), covar=tensor([0.3645, 0.2877, 0.2524, 0.3016], device='cuda:1'), in_proj_covar=tensor([0.2025, 0.1962, 0.1867, 0.2014], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:10:00,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:10:15,739 INFO [train.py:968] (1/2) Epoch 26, batch 35700, giga_loss[loss=0.3512, simple_loss=0.4003, pruned_loss=0.151, over 28959.00 frames. ], tot_loss[loss=0.236, simple_loss=0.3088, pruned_loss=0.08164, over 5688159.09 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3453, pruned_loss=0.1084, over 5708704.06 frames. ], giga_tot_loss[loss=0.2311, simple_loss=0.3045, pruned_loss=0.07886, over 5681097.50 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:10:23,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:10:26,770 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:10:41,801 INFO [optim.py:369] (1/2) 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 16:11:03,585 INFO [train.py:968] (1/2) Epoch 26, batch 35750, giga_loss[loss=0.3006, simple_loss=0.3731, pruned_loss=0.1141, over 28897.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3233, pruned_loss=0.08939, over 5686858.30 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3454, pruned_loss=0.1083, over 5711733.60 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3194, pruned_loss=0.08699, over 5678256.52 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:11:06,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:11:17,502 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1174822.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:11:32,975 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:11:34,811 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:11:46,970 INFO [train.py:968] (1/2) Epoch 26, batch 35800, giga_loss[loss=0.2706, simple_loss=0.355, pruned_loss=0.09314, over 28715.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.333, pruned_loss=0.09388, over 5685922.08 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3458, pruned_loss=0.1085, over 5710130.91 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3295, pruned_loss=0.09168, over 5680115.52 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:11:51,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9302, 1.1638, 1.0357, 0.8243], device='cuda:1'), covar=tensor([0.2282, 0.2522, 0.1731, 0.2307], device='cuda:1'), in_proj_covar=tensor([0.2025, 0.1964, 0.1872, 0.2022], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:12:02,792 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:09,252 INFO [optim.py:369] (1/2) 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 16:12:29,096 INFO [train.py:968] (1/2) Epoch 26, batch 35850, giga_loss[loss=0.2507, simple_loss=0.3312, pruned_loss=0.08509, over 28597.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3386, pruned_loss=0.09556, over 5678157.39 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3462, pruned_loss=0.1087, over 5705053.65 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3352, pruned_loss=0.0933, over 5677601.67 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:12:30,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1174908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:32,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1174911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:32,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:54,559 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1174940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:12:56,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5631, 1.5492, 1.7756, 1.3887], device='cuda:1'), covar=tensor([0.1693, 0.2361, 0.1413, 0.1699], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0709, 0.0977, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 16:13:14,997 INFO [train.py:968] (1/2) Epoch 26, batch 35900, giga_loss[loss=0.2945, simple_loss=0.3728, pruned_loss=0.1081, over 28918.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3409, pruned_loss=0.0958, over 5656581.98 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3465, pruned_loss=0.1089, over 5688433.20 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3378, pruned_loss=0.09352, over 5669783.82 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:13:35,946 INFO [optim.py:369] (1/2) 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 16:13:45,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:13:45,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-13 16:13:46,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1174994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:13:58,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9806, 1.3025, 1.3223, 1.1370], device='cuda:1'), covar=tensor([0.2036, 0.1354, 0.2289, 0.1644], device='cuda:1'), in_proj_covar=tensor([0.0485, 0.0742, 0.0715, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 16:13:59,063 INFO [train.py:968] (1/2) Epoch 26, batch 35950, giga_loss[loss=0.3223, simple_loss=0.3911, pruned_loss=0.1268, over 28811.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3422, pruned_loss=0.096, over 5658828.09 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3467, pruned_loss=0.109, over 5691726.11 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3394, pruned_loss=0.09368, over 5665616.03 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:14:09,788 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4615, 1.7811, 1.6965, 1.3889], device='cuda:1'), covar=tensor([0.2889, 0.2354, 0.1775, 0.2419], device='cuda:1'), in_proj_covar=tensor([0.2022, 0.1964, 0.1870, 0.2020], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:14:19,976 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1175033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:14:28,047 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.98 vs. limit=5.0
+2023-03-13 16:14:37,891 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 36000, giga_loss[loss=0.272, simple_loss=0.3533, pruned_loss=0.09535, over 28628.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3453, pruned_loss=0.09795, over 5668040.97 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3473, pruned_loss=0.1093, over 5689431.98 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3423, pruned_loss=0.09544, over 5674846.78 frames. ], batch size: 242, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:14:38,808 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 16:14:44,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3463, 3.1168, 1.4223, 1.5238], device='cuda:1'), covar=tensor([0.1151, 0.0382, 0.1043, 0.1537], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0562, 0.0403, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 16:14:47,089 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 16:14:48,132 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:01,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3504, 1.4227, 1.3932, 1.2441], device='cuda:1'), covar=tensor([0.2700, 0.2939, 0.2143, 0.2635], device='cuda:1'), in_proj_covar=tensor([0.2024, 0.1966, 0.1872, 0.2023], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:15:09,584 INFO [optim.py:369] (1/2) 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 16:15:11,806 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:28,685 INFO [train.py:968] (1/2) Epoch 26, batch 36050, giga_loss[loss=0.3172, simple_loss=0.3824, pruned_loss=0.126, over 28335.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3478, pruned_loss=0.0998, over 5673492.37 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3473, pruned_loss=0.1092, over 5694013.86 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3454, pruned_loss=0.09764, over 5674409.03 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:15:43,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.09 vs. limit=5.0
+2023-03-13 16:15:52,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:53,003 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:55,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:15:56,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:05,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9130, 1.9555, 1.4649, 1.5261], device='cuda:1'), covar=tensor([0.1056, 0.0717, 0.1076, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0447, 0.0525, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 16:16:08,292 INFO [train.py:968] (1/2) Epoch 26, batch 36100, giga_loss[loss=0.2797, simple_loss=0.359, pruned_loss=0.1002, over 28679.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3505, pruned_loss=0.1014, over 5676999.98 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3477, pruned_loss=0.1096, over 5689995.66 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3483, pruned_loss=0.09911, over 5680568.41 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:16:19,027 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:19,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:20,571 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4633, 1.5229, 1.5620, 1.3484], device='cuda:1'), covar=tensor([0.3108, 0.3073, 0.2390, 0.3158], device='cuda:1'), in_proj_covar=tensor([0.2021, 0.1962, 0.1869, 0.2021], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:16:31,676 INFO [optim.py:369] (1/2) 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 16:16:33,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:16:41,687 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175197.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:16:48,143 INFO [train.py:968] (1/2) Epoch 26, batch 36150, giga_loss[loss=0.2747, simple_loss=0.3585, pruned_loss=0.09547, over 29085.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3543, pruned_loss=0.1023, over 5695236.72 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3479, pruned_loss=0.1096, over 5692577.17 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3524, pruned_loss=0.1003, over 5695674.51 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:17:31,472 INFO [train.py:968] (1/2) Epoch 26, batch 36200, giga_loss[loss=0.2452, simple_loss=0.3323, pruned_loss=0.07902, over 28554.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3559, pruned_loss=0.1025, over 5693620.10 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3483, pruned_loss=0.1098, over 5697070.57 frames. ], giga_tot_loss[loss=0.2778, simple_loss=0.3542, pruned_loss=0.1007, over 5689917.56 frames. ], batch size: 65, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:17:53,968 INFO [optim.py:369] (1/2) 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 16:18:10,861 INFO [train.py:968] (1/2) Epoch 26, batch 36250, giga_loss[loss=0.2846, simple_loss=0.3466, pruned_loss=0.1113, over 23667.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3568, pruned_loss=0.1022, over 5701556.81 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3487, pruned_loss=0.1099, over 5703538.21 frames. ], giga_tot_loss[loss=0.278, simple_loss=0.3553, pruned_loss=0.1003, over 5692780.42 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:18:29,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:18:31,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:18:38,227 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175340.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:18:40,054 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175343.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:18:51,492 INFO [train.py:968] (1/2) Epoch 26, batch 36300, giga_loss[loss=0.2236, simple_loss=0.3172, pruned_loss=0.06504, over 28630.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3566, pruned_loss=0.1012, over 5699722.25 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3495, pruned_loss=0.1103, over 5703309.92 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3549, pruned_loss=0.09924, over 5692694.57 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:18:54,063 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:19:01,503 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:19:03,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-13 16:19:10,338 INFO [optim.py:369] (1/2) 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 16:19:20,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-13 16:19:29,580 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 36350, libri_loss[loss=0.275, simple_loss=0.3593, pruned_loss=0.09536, over 29381.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3545, pruned_loss=0.09893, over 5703082.06 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3498, pruned_loss=0.1103, over 5708567.68 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.353, pruned_loss=0.09705, over 5692712.83 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:19:30,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:20:10,663 INFO [train.py:968] (1/2) Epoch 26, batch 36400, giga_loss[loss=0.2963, simple_loss=0.3737, pruned_loss=0.1094, over 28564.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3538, pruned_loss=0.09847, over 5688916.41 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3501, pruned_loss=0.1104, over 5702862.91 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3525, pruned_loss=0.09667, over 5685467.31 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:20:32,916 INFO [optim.py:369] (1/2) 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 16:20:46,234 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4578, 3.2230, 1.5082, 1.6448], device='cuda:1'), covar=tensor([0.1078, 0.0300, 0.0923, 0.1412], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0560, 0.0402, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 16:20:56,075 INFO [train.py:968] (1/2) Epoch 26, batch 36450, giga_loss[loss=0.3495, simple_loss=0.3823, pruned_loss=0.1583, over 26514.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3568, pruned_loss=0.1025, over 5668268.34 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3505, pruned_loss=0.1107, over 5685737.50 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3554, pruned_loss=0.1006, over 5680904.42 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:21:24,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2598, 4.0286, 3.8848, 1.9563], device='cuda:1'), covar=tensor([0.0633, 0.0880, 0.0820, 0.2040], device='cuda:1'), in_proj_covar=tensor([0.1259, 0.1161, 0.0979, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 16:21:33,498 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:21:37,016 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 36500, giga_loss[loss=0.27, simple_loss=0.3364, pruned_loss=0.1018, over 28830.00 frames. ], tot_loss[loss=0.2862, simple_loss=0.3594, pruned_loss=0.1065, over 5679101.70 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3509, pruned_loss=0.1109, over 5690043.33 frames. ], giga_tot_loss[loss=0.2837, simple_loss=0.3582, pruned_loss=0.1046, over 5685098.43 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:21:59,310 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:22:22,094 INFO [train.py:968] (1/2) Epoch 26, batch 36550, giga_loss[loss=0.2786, simple_loss=0.3327, pruned_loss=0.1123, over 23995.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3589, pruned_loss=0.1076, over 5679112.66 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3509, pruned_loss=0.1108, over 5691409.65 frames. ], giga_tot_loss[loss=0.2851, simple_loss=0.3579, pruned_loss=0.1061, over 5682497.17 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:23:04,250 INFO [train.py:968] (1/2) Epoch 26, batch 36600, giga_loss[loss=0.2477, simple_loss=0.3262, pruned_loss=0.08458, over 28542.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3569, pruned_loss=0.1072, over 5691054.59 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3509, pruned_loss=0.1108, over 5695233.86 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.3563, pruned_loss=0.106, over 5690201.15 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:23:26,644 INFO [optim.py:369] (1/2) 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 16:23:46,284 INFO [train.py:968] (1/2) Epoch 26, batch 36650, giga_loss[loss=0.2625, simple_loss=0.3384, pruned_loss=0.09324, over 28509.00 frames. ], tot_loss[loss=0.2848, simple_loss=0.3559, pruned_loss=0.1069, over 5694773.70 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3514, pruned_loss=0.111, over 5697588.80 frames. ], giga_tot_loss[loss=0.2832, simple_loss=0.355, pruned_loss=0.1057, over 5692143.67 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:24:18,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3281, 1.5932, 1.3364, 1.0700], device='cuda:1'), covar=tensor([0.2220, 0.2188, 0.2288, 0.2235], device='cuda:1'), in_proj_covar=tensor([0.1573, 0.1132, 0.1390, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 16:24:27,756 INFO [train.py:968] (1/2) Epoch 26, batch 36700, giga_loss[loss=0.2586, simple_loss=0.3356, pruned_loss=0.09082, over 28803.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3541, pruned_loss=0.1051, over 5689530.27 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3521, pruned_loss=0.1114, over 5691546.36 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3529, pruned_loss=0.1037, over 5692438.31 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:24:50,948 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 16:24:53,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1175781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:24:55,991 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 16:24:56,148 INFO [optim.py:369] (1/2) 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 16:25:15,583 INFO [train.py:968] (1/2) Epoch 26, batch 36750, giga_loss[loss=0.2504, simple_loss=0.3252, pruned_loss=0.08781, over 28235.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3506, pruned_loss=0.1019, over 5696387.46 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3522, pruned_loss=0.1114, over 5692640.62 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3495, pruned_loss=0.1007, over 5697650.04 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:26:04,422 INFO [train.py:968] (1/2) Epoch 26, batch 36800, giga_loss[loss=0.2193, simple_loss=0.3039, pruned_loss=0.06735, over 29018.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3445, pruned_loss=0.09816, over 5694349.77 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3525, pruned_loss=0.1116, over 5697176.66 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3434, pruned_loss=0.09696, over 5691420.73 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:26:30,450 INFO [optim.py:369] (1/2) 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 16:26:51,496 INFO [train.py:968] (1/2) Epoch 26, batch 36850, giga_loss[loss=0.2196, simple_loss=0.2969, pruned_loss=0.07109, over 28189.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3386, pruned_loss=0.09502, over 5693677.33 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3528, pruned_loss=0.1117, over 5700942.77 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3371, pruned_loss=0.09355, over 5687977.16 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:27:08,963 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1175924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:27:12,017 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1175927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:27:21,757 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0695, 4.9187, 4.6328, 2.1630], device='cuda:1'), covar=tensor([0.0422, 0.0529, 0.0582, 0.1963], device='cuda:1'), in_proj_covar=tensor([0.1261, 0.1163, 0.0981, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 16:27:40,532 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1175956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:27:40,926 INFO [train.py:968] (1/2) Epoch 26, batch 36900, giga_loss[loss=0.231, simple_loss=0.3159, pruned_loss=0.0731, over 28973.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3332, pruned_loss=0.09233, over 5681337.33 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3528, pruned_loss=0.1116, over 5702102.71 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3316, pruned_loss=0.09086, over 5675639.83 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:28:05,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7263, 1.9199, 1.4162, 1.4514], device='cuda:1'), covar=tensor([0.1121, 0.0710, 0.1090, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0446, 0.0522, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 16:28:07,462 INFO [optim.py:369] (1/2) 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 16:28:26,332 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 16:28:26,548 INFO [train.py:968] (1/2) Epoch 26, batch 36950, giga_loss[loss=0.2309, simple_loss=0.3171, pruned_loss=0.07232, over 29063.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3339, pruned_loss=0.09219, over 5682523.55 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3528, pruned_loss=0.1114, over 5707264.86 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3321, pruned_loss=0.09064, over 5672729.99 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:28:41,081 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1882, 1.7249, 5.4492, 3.7956], device='cuda:1'), covar=tensor([0.1375, 0.2500, 0.0369, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0790, 0.0667, 0.0987, 0.0953], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 16:29:07,410 INFO [train.py:968] (1/2) Epoch 26, batch 37000, libri_loss[loss=0.2541, simple_loss=0.3273, pruned_loss=0.09046, over 29364.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3349, pruned_loss=0.09239, over 5697309.15 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3526, pruned_loss=0.1112, over 5712127.65 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3332, pruned_loss=0.0909, over 5684700.49 frames. ], batch size: 67, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:29:08,953 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5071, 1.1867, 4.3149, 3.5699], device='cuda:1'), covar=tensor([0.1518, 0.2758, 0.0423, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0668, 0.0989, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 16:29:29,997 INFO [optim.py:369] (1/2) 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 16:29:49,365 INFO [train.py:968] (1/2) Epoch 26, batch 37050, giga_loss[loss=0.3405, simple_loss=0.3891, pruned_loss=0.146, over 26795.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3348, pruned_loss=0.09259, over 5701197.87 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3523, pruned_loss=0.1108, over 5717781.55 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3333, pruned_loss=0.09126, over 5685797.87 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:30:01,893 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-13 16:30:29,571 INFO [train.py:968] (1/2) Epoch 26, batch 37100, giga_loss[loss=0.2107, simple_loss=0.2878, pruned_loss=0.06678, over 28480.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3327, pruned_loss=0.09157, over 5702655.93 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3526, pruned_loss=0.1107, over 5719396.05 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3307, pruned_loss=0.0901, over 5688583.89 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:30:53,896 INFO [optim.py:369] (1/2) 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 16:31:11,573 INFO [train.py:968] (1/2) Epoch 26, batch 37150, giga_loss[loss=0.2107, simple_loss=0.2924, pruned_loss=0.06455, over 28915.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3299, pruned_loss=0.09023, over 5711875.68 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3525, pruned_loss=0.1106, over 5721273.16 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3282, pruned_loss=0.08887, over 5698880.23 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:31:50,278 INFO [train.py:968] (1/2) Epoch 26, batch 37200, libri_loss[loss=0.2995, simple_loss=0.3779, pruned_loss=0.1106, over 27743.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3287, pruned_loss=0.08988, over 5716768.38 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3533, pruned_loss=0.1109, over 5722875.73 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3262, pruned_loss=0.08818, over 5704983.80 frames. ], batch size: 115, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:31:51,331 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 16:32:12,284 INFO [optim.py:369] (1/2) 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 16:32:24,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2015, 2.5904, 1.2591, 1.3726], device='cuda:1'), covar=tensor([0.1101, 0.0391, 0.0963, 0.1456], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0559, 0.0401, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 16:32:30,471 INFO [train.py:968] (1/2) Epoch 26, batch 37250, giga_loss[loss=0.2395, simple_loss=0.3097, pruned_loss=0.08459, over 28818.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3276, pruned_loss=0.08953, over 5714404.67 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3534, pruned_loss=0.1108, over 5725416.78 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.325, pruned_loss=0.08795, over 5702546.71 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:32:35,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7033, 1.9708, 1.6494, 1.6845], device='cuda:1'), covar=tensor([0.2640, 0.2804, 0.3150, 0.2565], device='cuda:1'), in_proj_covar=tensor([0.1575, 0.1134, 0.1390, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 16:33:09,568 INFO [train.py:968] (1/2) Epoch 26, batch 37300, giga_loss[loss=0.2218, simple_loss=0.3051, pruned_loss=0.06924, over 28853.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3267, pruned_loss=0.08905, over 5714127.31 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3543, pruned_loss=0.111, over 5727038.08 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3231, pruned_loss=0.08698, over 5702587.67 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:33:23,589 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:33:31,018 INFO [optim.py:369] (1/2) 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 16:33:47,871 INFO [train.py:968] (1/2) Epoch 26, batch 37350, giga_loss[loss=0.256, simple_loss=0.3277, pruned_loss=0.09215, over 28629.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3242, pruned_loss=0.08756, over 5724441.38 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3549, pruned_loss=0.1112, over 5730960.31 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3201, pruned_loss=0.08522, over 5711520.01 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:34:28,506 INFO [train.py:968] (1/2) Epoch 26, batch 37400, libri_loss[loss=0.2879, simple_loss=0.3526, pruned_loss=0.1117, over 29644.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.322, pruned_loss=0.08646, over 5721544.30 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3556, pruned_loss=0.1117, over 5724972.70 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.3176, pruned_loss=0.08372, over 5717024.48 frames. ], batch size: 69, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:34:53,510 INFO [optim.py:369] (1/2) 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 16:35:08,703 INFO [train.py:968] (1/2) Epoch 26, batch 37450, giga_loss[loss=0.2129, simple_loss=0.2952, pruned_loss=0.06529, over 28621.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3197, pruned_loss=0.085, over 5718151.97 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3557, pruned_loss=0.1117, over 5716937.24 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3159, pruned_loss=0.08262, over 5722526.81 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:35:51,429 INFO [train.py:968] (1/2) Epoch 26, batch 37500, giga_loss[loss=0.3063, simple_loss=0.3636, pruned_loss=0.1245, over 28895.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3214, pruned_loss=0.08654, over 5716756.76 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.356, pruned_loss=0.1117, over 5719179.30 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3174, pruned_loss=0.08417, over 5718122.95 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:36:17,663 INFO [optim.py:369] (1/2) 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 16:36:33,230 INFO [train.py:968] (1/2) Epoch 26, batch 37550, giga_loss[loss=0.2588, simple_loss=0.3334, pruned_loss=0.09213, over 28757.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3251, pruned_loss=0.08855, over 5714699.24 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3565, pruned_loss=0.1118, over 5722328.16 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3207, pruned_loss=0.08594, over 5712893.62 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:37:06,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9711, 2.2259, 2.2176, 1.7718], device='cuda:1'), covar=tensor([0.1456, 0.2086, 0.1290, 0.1686], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0712, 0.0978, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 16:37:17,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:37:17,702 INFO [train.py:968] (1/2) Epoch 26, batch 37600, giga_loss[loss=0.3021, simple_loss=0.3718, pruned_loss=0.1161, over 28588.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3315, pruned_loss=0.09254, over 5703551.87 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3572, pruned_loss=0.1121, over 5716598.98 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3267, pruned_loss=0.08971, over 5707496.27 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:37:25,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:38:07,570 INFO [train.py:968] (1/2) Epoch 26, batch 37650, giga_loss[loss=0.3498, simple_loss=0.3998, pruned_loss=0.1499, over 27818.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3402, pruned_loss=0.09868, over 5684025.86 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3572, pruned_loss=0.1122, over 5709783.87 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3359, pruned_loss=0.09607, over 5693036.59 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:38:15,629 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:38:52,365 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1176751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:38:55,805 INFO [train.py:968] (1/2) Epoch 26, batch 37700, giga_loss[loss=0.2787, simple_loss=0.3491, pruned_loss=0.1042, over 28722.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3444, pruned_loss=0.1006, over 5666692.39 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3577, pruned_loss=0.1125, over 5705198.24 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.34, pruned_loss=0.09779, over 5677834.64 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:39:09,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:39:15,471 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1819, 1.3066, 1.2792, 1.1638], device='cuda:1'), covar=tensor([0.2219, 0.2509, 0.1803, 0.2180], device='cuda:1'), in_proj_covar=tensor([0.2040, 0.1972, 0.1890, 0.2045], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:39:19,462 INFO [optim.py:369] (1/2) 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 16:39:23,068 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:39:31,963 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9201, 1.2178, 1.2492, 1.0822], device='cuda:1'), covar=tensor([0.1966, 0.1388, 0.2309, 0.1613], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0754, 0.0725, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 16:39:39,311 INFO [train.py:968] (1/2) Epoch 26, batch 37750, giga_loss[loss=0.2549, simple_loss=0.3406, pruned_loss=0.08456, over 28890.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3486, pruned_loss=0.1019, over 5669800.74 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3581, pruned_loss=0.1127, over 5706056.30 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3445, pruned_loss=0.09935, over 5677150.19 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:40:26,530 INFO [train.py:968] (1/2) Epoch 26, batch 37800, giga_loss[loss=0.3506, simple_loss=0.4146, pruned_loss=0.1433, over 27984.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3545, pruned_loss=0.1053, over 5668707.12 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3583, pruned_loss=0.1129, over 5708290.22 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3511, pruned_loss=0.103, over 5672225.17 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:40:52,132 INFO [optim.py:369] (1/2) 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 16:40:57,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1176894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:40:59,245 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1176897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:41:04,198 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1176904.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:41:05,786 INFO [train.py:968] (1/2) Epoch 26, batch 37850, giga_loss[loss=0.2414, simple_loss=0.3212, pruned_loss=0.08076, over 28804.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3554, pruned_loss=0.1057, over 5659450.18 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3586, pruned_loss=0.1131, over 5700634.14 frames. ], giga_tot_loss[loss=0.2796, simple_loss=0.3523, pruned_loss=0.1035, over 5668932.66 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:41:20,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2935, 1.5028, 1.2717, 1.5407], device='cuda:1'), covar=tensor([0.0811, 0.0340, 0.0347, 0.0910], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0114], device='cuda:1')
+2023-03-13 16:41:21,457 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1176926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:41:46,686 INFO [train.py:968] (1/2) Epoch 26, batch 37900, giga_loss[loss=0.2681, simple_loss=0.3505, pruned_loss=0.0929, over 28634.00 frames. ], tot_loss[loss=0.2773, simple_loss=0.3509, pruned_loss=0.1019, over 5674457.75 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3587, pruned_loss=0.1132, over 5704155.81 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3482, pruned_loss=0.09983, over 5678397.81 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:41:55,455 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:42:27,089 INFO [train.py:968] (1/2) Epoch 26, batch 37950, giga_loss[loss=0.2737, simple_loss=0.3494, pruned_loss=0.09898, over 28694.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3495, pruned_loss=0.1004, over 5675723.40 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3586, pruned_loss=0.1131, over 5698403.82 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3472, pruned_loss=0.09855, over 5683815.98 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:42:33,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2180, 1.4467, 1.4110, 1.1427], device='cuda:1'), covar=tensor([0.3224, 0.2798, 0.1807, 0.2664], device='cuda:1'), in_proj_covar=tensor([0.2043, 0.1978, 0.1898, 0.2053], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:42:48,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:42:54,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:42:57,936 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-13 16:43:10,787 INFO [train.py:968] (1/2) Epoch 26, batch 38000, giga_loss[loss=0.2806, simple_loss=0.3594, pruned_loss=0.1009, over 28920.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3481, pruned_loss=0.09902, over 5676439.32 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3587, pruned_loss=0.1131, over 5699389.61 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.346, pruned_loss=0.0974, over 5681731.74 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:43:36,379 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3737, 1.9433, 1.4897, 0.6453], device='cuda:1'), covar=tensor([0.5986, 0.3176, 0.4596, 0.6814], device='cuda:1'), in_proj_covar=tensor([0.1815, 0.1711, 0.1651, 0.1486], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 16:43:36,904 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:43:37,248 INFO [optim.py:369] (1/2) 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 16:43:42,821 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:43:54,398 INFO [train.py:968] (1/2) Epoch 26, batch 38050, giga_loss[loss=0.3224, simple_loss=0.3928, pruned_loss=0.126, over 28996.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3499, pruned_loss=0.09992, over 5682190.05 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.359, pruned_loss=0.1133, over 5702556.95 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.348, pruned_loss=0.09829, over 5683100.07 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:44:15,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8671, 1.0816, 2.8127, 2.6497], device='cuda:1'), covar=tensor([0.1689, 0.2729, 0.0619, 0.1564], device='cuda:1'), in_proj_covar=tensor([0.0788, 0.0664, 0.0983, 0.0949], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 16:44:27,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:35,353 INFO [train.py:968] (1/2) Epoch 26, batch 38100, giga_loss[loss=0.2487, simple_loss=0.3299, pruned_loss=0.08375, over 28558.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3518, pruned_loss=0.1012, over 5683323.75 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.359, pruned_loss=0.1131, over 5700931.30 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3499, pruned_loss=0.09954, over 5683923.04 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:44:41,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:50,636 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:55,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:44:57,354 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:00,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:05,543 INFO [optim.py:369] (1/2) 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 16:45:19,110 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 38150, giga_loss[loss=0.3282, simple_loss=0.3902, pruned_loss=0.1331, over 28280.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3532, pruned_loss=0.1023, over 5691500.42 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3589, pruned_loss=0.1129, over 5704221.88 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3516, pruned_loss=0.1009, over 5688783.93 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:45:24,725 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:35,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177225.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:45:41,844 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:45:44,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:02,974 INFO [train.py:968] (1/2) Epoch 26, batch 38200, giga_loss[loss=0.3076, simple_loss=0.3737, pruned_loss=0.1207, over 27586.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3546, pruned_loss=0.1039, over 5690218.76 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3588, pruned_loss=0.1127, over 5708504.03 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3534, pruned_loss=0.1028, over 5684048.00 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:46:11,542 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:23,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177279.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:46:31,577 INFO [optim.py:369] (1/2) 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 16:46:31,870 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:34,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:34,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2238, 1.5619, 1.4865, 1.0949], device='cuda:1'), covar=tensor([0.1447, 0.2533, 0.1326, 0.1612], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0710, 0.0975, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 16:46:38,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4111, 1.4712, 1.5121, 1.3098], device='cuda:1'), covar=tensor([0.2641, 0.2991, 0.2252, 0.2525], device='cuda:1'), in_proj_covar=tensor([0.2048, 0.1981, 0.1903, 0.2054], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:46:43,921 INFO [train.py:968] (1/2) Epoch 26, batch 38250, giga_loss[loss=0.313, simple_loss=0.3757, pruned_loss=0.1251, over 28793.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3562, pruned_loss=0.1054, over 5700413.20 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3593, pruned_loss=0.1129, over 5712323.51 frames. ], giga_tot_loss[loss=0.2815, simple_loss=0.3547, pruned_loss=0.1041, over 5691612.39 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:46:46,943 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:49,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:57,140 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:46:57,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2741, 2.6051, 1.3171, 1.3598], device='cuda:1'), covar=tensor([0.1069, 0.0350, 0.0940, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0560, 0.0402, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 16:47:12,969 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:47:15,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:47:26,870 INFO [train.py:968] (1/2) Epoch 26, batch 38300, giga_loss[loss=0.2705, simple_loss=0.3464, pruned_loss=0.0973, over 28735.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.356, pruned_loss=0.1046, over 5696795.93 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3598, pruned_loss=0.1131, over 5707123.68 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3542, pruned_loss=0.1032, over 5694431.10 frames. ], batch size: 242, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:47:53,541 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 38350, libri_loss[loss=0.2895, simple_loss=0.3512, pruned_loss=0.1139, over 29589.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.3552, pruned_loss=0.103, over 5702732.64 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3599, pruned_loss=0.1132, over 5708235.54 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3536, pruned_loss=0.1016, over 5699399.13 frames. ], batch size: 74, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 16:48:18,243 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177422.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:48:21,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177425.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:48:43,781 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 38400, giga_loss[loss=0.2526, simple_loss=0.338, pruned_loss=0.08359, over 28914.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3557, pruned_loss=0.1027, over 5706115.72 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3599, pruned_loss=0.1133, over 5712324.82 frames. ], giga_tot_loss[loss=0.2783, simple_loss=0.3543, pruned_loss=0.1012, over 5699960.57 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:48:50,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:49:10,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:49:12,783 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:49:27,752 INFO [train.py:968] (1/2) Epoch 26, batch 38450, giga_loss[loss=0.2916, simple_loss=0.3628, pruned_loss=0.1102, over 28802.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3538, pruned_loss=0.1012, over 5708031.15 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3601, pruned_loss=0.1134, over 5715095.22 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3524, pruned_loss=0.09976, over 5700617.51 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:49:35,651 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:49:49,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 16:49:57,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8936, 1.0078, 5.0529, 3.5992], device='cuda:1'), covar=tensor([0.1414, 0.3041, 0.0440, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0791, 0.0667, 0.0988, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 16:50:05,089 INFO [train.py:968] (1/2) Epoch 26, batch 38500, giga_loss[loss=0.2787, simple_loss=0.3507, pruned_loss=0.1033, over 28883.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3516, pruned_loss=0.1001, over 5706891.02 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1133, over 5709927.14 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3503, pruned_loss=0.09876, over 5705649.83 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:50:32,118 INFO [optim.py:369] (1/2) 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 16:50:39,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1177600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:50:43,762 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 38550, giga_loss[loss=0.2394, simple_loss=0.325, pruned_loss=0.07686, over 28931.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3488, pruned_loss=0.0987, over 5716957.53 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 5714497.47 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.09742, over 5711680.00 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:50:46,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:50:57,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 16:51:07,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:51:24,078 INFO [train.py:968] (1/2) Epoch 26, batch 38600, libri_loss[loss=0.3123, simple_loss=0.3678, pruned_loss=0.1284, over 29396.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3489, pruned_loss=0.09952, over 5717822.70 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3599, pruned_loss=0.113, over 5719079.37 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3477, pruned_loss=0.09819, over 5709419.33 frames. ], batch size: 67, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:51:49,617 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 38650, giga_loss[loss=0.277, simple_loss=0.3524, pruned_loss=0.1008, over 28604.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3496, pruned_loss=0.1005, over 5722166.78 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3601, pruned_loss=0.1132, over 5723400.99 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3483, pruned_loss=0.09899, over 5711680.98 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:52:07,869 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:52:33,007 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1177743.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 16:52:35,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3921, 1.5708, 1.3596, 1.5758], device='cuda:1'), covar=tensor([0.0840, 0.0337, 0.0347, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 16:52:35,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1177746.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:52:40,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 16:52:42,295 INFO [train.py:968] (1/2) Epoch 26, batch 38700, giga_loss[loss=0.2689, simple_loss=0.3465, pruned_loss=0.09564, over 28775.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3493, pruned_loss=0.09976, over 5726556.08 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3598, pruned_loss=0.113, over 5729001.69 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3482, pruned_loss=0.09832, over 5712826.37 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:52:57,110 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1177775.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 16:53:07,523 INFO [optim.py:369] (1/2) 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 16:53:21,342 INFO [train.py:968] (1/2) Epoch 26, batch 38750, giga_loss[loss=0.276, simple_loss=0.3497, pruned_loss=0.1012, over 28611.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3487, pruned_loss=0.0983, over 5721495.02 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3598, pruned_loss=0.1129, over 5730878.35 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3477, pruned_loss=0.09713, over 5708970.29 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:53:57,523 INFO [train.py:968] (1/2) Epoch 26, batch 38800, giga_loss[loss=0.2813, simple_loss=0.3515, pruned_loss=0.1056, over 28984.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3479, pruned_loss=0.09809, over 5721573.76 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3596, pruned_loss=0.1129, over 5726924.06 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3469, pruned_loss=0.09664, over 5714365.93 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:54:24,501 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1177888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:54:26,456 INFO [optim.py:369] (1/2) 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 16:54:38,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1464, 3.9740, 3.7708, 1.8803], device='cuda:1'), covar=tensor([0.0683, 0.0790, 0.0755, 0.2106], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1170, 0.0985, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 16:54:38,906 INFO [train.py:968] (1/2) Epoch 26, batch 38850, giga_loss[loss=0.2512, simple_loss=0.3312, pruned_loss=0.08563, over 29087.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3473, pruned_loss=0.09854, over 5717585.76 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5729704.28 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3459, pruned_loss=0.09673, over 5708959.88 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:55:18,205 INFO [train.py:968] (1/2) Epoch 26, batch 38900, giga_loss[loss=0.2634, simple_loss=0.3338, pruned_loss=0.09656, over 28560.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3439, pruned_loss=0.09669, over 5713275.69 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5730443.59 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3427, pruned_loss=0.09524, over 5705474.79 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:55:28,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2436, 1.4063, 1.5727, 1.3659], device='cuda:1'), covar=tensor([0.2607, 0.2339, 0.1716, 0.2053], device='cuda:1'), in_proj_covar=tensor([0.2040, 0.1983, 0.1903, 0.2047], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 16:55:44,924 INFO [optim.py:369] (1/2) 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 16:55:59,402 INFO [train.py:968] (1/2) Epoch 26, batch 38950, giga_loss[loss=0.3138, simple_loss=0.3791, pruned_loss=0.1243, over 26759.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.341, pruned_loss=0.09531, over 5710537.01 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3597, pruned_loss=0.1129, over 5729939.54 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3398, pruned_loss=0.09384, over 5704421.17 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:56:22,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4526, 1.7700, 1.4678, 1.4854], device='cuda:1'), covar=tensor([0.2521, 0.2471, 0.2649, 0.2498], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0757, 0.0728, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 16:56:39,432 INFO [train.py:968] (1/2) Epoch 26, batch 39000, giga_loss[loss=0.3492, simple_loss=0.4002, pruned_loss=0.1491, over 28307.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3402, pruned_loss=0.09494, over 5706070.20 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3593, pruned_loss=0.1128, over 5723918.92 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3392, pruned_loss=0.09356, over 5706179.28 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:56:39,432 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 16:56:48,185 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 16:57:13,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1178087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:57:16,100 INFO [optim.py:369] (1/2) 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 16:57:27,661 INFO [train.py:968] (1/2) Epoch 26, batch 39050, giga_loss[loss=0.2953, simple_loss=0.3594, pruned_loss=0.1156, over 27665.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3407, pruned_loss=0.09602, over 5695060.45 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3597, pruned_loss=0.1131, over 5718587.85 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3386, pruned_loss=0.09379, over 5698963.50 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:58:04,663 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 16:58:07,483 INFO [train.py:968] (1/2) Epoch 26, batch 39100, giga_loss[loss=0.2631, simple_loss=0.3258, pruned_loss=0.1002, over 28794.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3384, pruned_loss=0.09502, over 5698811.73 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3597, pruned_loss=0.1131, over 5716660.58 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3365, pruned_loss=0.09302, over 5703347.82 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:58:33,790 INFO [optim.py:369] (1/2) 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 16:58:41,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3810, 1.8000, 1.0901, 1.3033], device='cuda:1'), covar=tensor([0.1324, 0.0789, 0.1614, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0448, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 16:58:47,264 INFO [train.py:968] (1/2) Epoch 26, batch 39150, giga_loss[loss=0.2497, simple_loss=0.3262, pruned_loss=0.08663, over 28644.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3371, pruned_loss=0.09475, over 5706014.05 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3596, pruned_loss=0.1131, over 5718578.91 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3353, pruned_loss=0.09285, over 5707698.64 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 16:59:01,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.49 vs. limit=5.0
+2023-03-13 16:59:05,869 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1178230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:07,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1178233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:20,211 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-13 16:59:26,062 INFO [train.py:968] (1/2) Epoch 26, batch 39200, giga_loss[loss=0.2342, simple_loss=0.3172, pruned_loss=0.07556, over 28982.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3347, pruned_loss=0.09372, over 5698781.03 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3599, pruned_loss=0.1133, over 5713028.39 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3325, pruned_loss=0.0916, over 5703911.35 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 16:59:32,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1178262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:32,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1178263.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 16:59:42,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2609, 3.1193, 1.5044, 1.3860], device='cuda:1'), covar=tensor([0.1024, 0.0429, 0.0916, 0.1427], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0558, 0.0400, 0.0436], device='cuda:1'), out_proj_covar=tensor([0.0027, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 16:59:57,660 INFO [optim.py:369] (1/2) 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 17:00:06,323 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 17:00:10,397 INFO [train.py:968] (1/2) Epoch 26, batch 39250, giga_loss[loss=0.2323, simple_loss=0.3092, pruned_loss=0.07767, over 28745.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3326, pruned_loss=0.09208, over 5703473.85 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3597, pruned_loss=0.1134, over 5715177.64 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3308, pruned_loss=0.09016, over 5705408.28 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:00:51,887 INFO [train.py:968] (1/2) Epoch 26, batch 39300, giga_loss[loss=0.2766, simple_loss=0.3431, pruned_loss=0.105, over 28881.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3363, pruned_loss=0.09375, over 5697601.77 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3602, pruned_loss=0.1136, over 5711694.57 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3338, pruned_loss=0.09148, over 5701506.15 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:01:07,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6520, 1.9452, 1.5644, 1.7204], device='cuda:1'), covar=tensor([0.2690, 0.2807, 0.3231, 0.2581], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1137, 0.1390, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 17:01:21,127 INFO [optim.py:369] (1/2) 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 17:01:36,522 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1178406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:01:36,865 INFO [train.py:968] (1/2) Epoch 26, batch 39350, giga_loss[loss=0.2426, simple_loss=0.3302, pruned_loss=0.07745, over 28921.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3398, pruned_loss=0.09549, over 5687367.67 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3601, pruned_loss=0.1137, over 5711951.47 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3374, pruned_loss=0.09334, over 5689562.38 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:01:39,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1178409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:01:53,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1178428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:02:01,017 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1178438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:02:16,260 INFO [train.py:968] (1/2) Epoch 26, batch 39400, giga_loss[loss=0.2609, simple_loss=0.3321, pruned_loss=0.09486, over 28697.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.343, pruned_loss=0.0969, over 5676348.16 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3607, pruned_loss=0.1143, over 5696439.78 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3399, pruned_loss=0.09401, over 5692139.00 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:02:45,897 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 39450, giga_loss[loss=0.2879, simple_loss=0.3661, pruned_loss=0.1048, over 28717.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3439, pruned_loss=0.09682, over 5684776.69 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3606, pruned_loss=0.1142, over 5703870.06 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3409, pruned_loss=0.094, over 5690128.91 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:02:59,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6839, 2.2224, 1.3438, 1.0717], device='cuda:1'), covar=tensor([0.8228, 0.4129, 0.3994, 0.7197], device='cuda:1'), in_proj_covar=tensor([0.1798, 0.1693, 0.1635, 0.1473], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 17:03:38,754 INFO [train.py:968] (1/2) Epoch 26, batch 39500, giga_loss[loss=0.3039, simple_loss=0.3731, pruned_loss=0.1174, over 28026.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09571, over 5696963.59 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3607, pruned_loss=0.1143, over 5708621.25 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3396, pruned_loss=0.09288, over 5696508.46 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:04:06,606 INFO [optim.py:369] (1/2) 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 17:04:17,031 INFO [train.py:968] (1/2) Epoch 26, batch 39550, giga_loss[loss=0.2655, simple_loss=0.3367, pruned_loss=0.09717, over 28907.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3428, pruned_loss=0.09648, over 5690321.42 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.361, pruned_loss=0.1146, over 5701503.56 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3399, pruned_loss=0.09357, over 5696642.19 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:04:58,958 INFO [train.py:968] (1/2) Epoch 26, batch 39600, libri_loss[loss=0.3177, simple_loss=0.3854, pruned_loss=0.125, over 25701.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.344, pruned_loss=0.09752, over 5696088.14 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3615, pruned_loss=0.1149, over 5694773.72 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3407, pruned_loss=0.09437, over 5706844.29 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:05:02,382 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-13 17:05:27,815 INFO [optim.py:369] (1/2) 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 17:05:38,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4649, 1.7991, 1.9158, 1.4063], device='cuda:1'), covar=tensor([0.3441, 0.2476, 0.2480, 0.2907], device='cuda:1'), in_proj_covar=tensor([0.2036, 0.1975, 0.1899, 0.2038], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 17:05:39,207 INFO [train.py:968] (1/2) Epoch 26, batch 39650, giga_loss[loss=0.2729, simple_loss=0.3567, pruned_loss=0.09457, over 28696.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3448, pruned_loss=0.0979, over 5696811.02 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3613, pruned_loss=0.1149, over 5689037.03 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3419, pruned_loss=0.09499, over 5710005.98 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:06:22,173 INFO [train.py:968] (1/2) Epoch 26, batch 39700, giga_loss[loss=0.2575, simple_loss=0.3435, pruned_loss=0.08575, over 29068.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3479, pruned_loss=0.0989, over 5698121.19 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3614, pruned_loss=0.1148, over 5691057.56 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3454, pruned_loss=0.09651, over 5706692.99 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:06:49,373 INFO [optim.py:369] (1/2) 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 17:06:57,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1178803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:06:59,700 INFO [train.py:968] (1/2) Epoch 26, batch 39750, giga_loss[loss=0.2791, simple_loss=0.349, pruned_loss=0.1046, over 23752.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3503, pruned_loss=0.1003, over 5699095.15 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.361, pruned_loss=0.1143, over 5690351.22 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3481, pruned_loss=0.09826, over 5706451.07 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:07:27,079 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1178843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:07:38,168 INFO [train.py:968] (1/2) Epoch 26, batch 39800, giga_loss[loss=0.2698, simple_loss=0.3539, pruned_loss=0.09281, over 28800.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3512, pruned_loss=0.1008, over 5705621.93 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3608, pruned_loss=0.1141, over 5694206.95 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3493, pruned_loss=0.09887, over 5708379.73 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:07:48,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4435, 1.3663, 3.9892, 3.2890], device='cuda:1'), covar=tensor([0.1528, 0.2753, 0.0438, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0789, 0.0664, 0.0986, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 17:08:10,416 INFO [optim.py:369] (1/2) 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 17:08:19,895 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5125, 1.6733, 1.6482, 1.3793], device='cuda:1'), covar=tensor([0.3311, 0.2796, 0.2512, 0.3069], device='cuda:1'), in_proj_covar=tensor([0.2042, 0.1985, 0.1906, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 17:08:22,205 INFO [train.py:968] (1/2) Epoch 26, batch 39850, libri_loss[loss=0.3318, simple_loss=0.3998, pruned_loss=0.1319, over 29523.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3529, pruned_loss=0.1013, over 5707534.73 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 5695374.10 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3511, pruned_loss=0.09965, over 5708754.79 frames. ], batch size: 82, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:08:25,171 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-13 17:08:40,169 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 17:08:53,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1178946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:08:55,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1178949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:08:58,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6260, 3.9387, 1.7955, 1.7169], device='cuda:1'), covar=tensor([0.0868, 0.0367, 0.0825, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0560, 0.0402, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 17:09:01,552 INFO [train.py:968] (1/2) Epoch 26, batch 39900, giga_loss[loss=0.2782, simple_loss=0.3593, pruned_loss=0.09856, over 28966.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3528, pruned_loss=0.1016, over 5707457.84 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3607, pruned_loss=0.1139, over 5698884.28 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.3516, pruned_loss=0.1003, over 5705432.17 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:09:13,122 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 17:09:19,892 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1178978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:09:22,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-13 17:09:30,824 INFO [optim.py:369] (1/2) 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 17:09:31,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4181, 1.5378, 1.5678, 1.4136], device='cuda:1'), covar=tensor([0.3001, 0.2779, 0.2283, 0.2681], device='cuda:1'), in_proj_covar=tensor([0.2045, 0.1989, 0.1909, 0.2046], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 17:09:41,013 INFO [train.py:968] (1/2) Epoch 26, batch 39950, giga_loss[loss=0.3211, simple_loss=0.3745, pruned_loss=0.1339, over 28779.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3513, pruned_loss=0.1008, over 5708906.96 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3609, pruned_loss=0.1139, over 5696200.32 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3498, pruned_loss=0.09935, over 5709883.52 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:09:43,085 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:09:45,343 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-13 17:09:47,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-13 17:10:21,005 INFO [train.py:968] (1/2) Epoch 26, batch 40000, giga_loss[loss=0.2414, simple_loss=0.3217, pruned_loss=0.08057, over 29102.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3484, pruned_loss=0.09942, over 5715729.44 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3614, pruned_loss=0.1143, over 5699932.76 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3465, pruned_loss=0.09771, over 5713552.81 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:10:35,916 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-13 17:10:40,256 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 17:10:50,335 INFO [optim.py:369] (1/2) 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 17:11:01,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:11:03,562 INFO [train.py:968] (1/2) Epoch 26, batch 40050, giga_loss[loss=0.2371, simple_loss=0.307, pruned_loss=0.08366, over 28648.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3457, pruned_loss=0.09828, over 5707464.31 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.3622, pruned_loss=0.1149, over 5695357.43 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3434, pruned_loss=0.09612, over 5709723.29 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:11:39,917 INFO [train.py:968] (1/2) Epoch 26, batch 40100, giga_loss[loss=0.3576, simple_loss=0.4267, pruned_loss=0.1443, over 28731.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3458, pruned_loss=0.09753, over 5706332.19 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3615, pruned_loss=0.1144, over 5689094.33 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.344, pruned_loss=0.09569, over 5713522.57 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:12:15,597 INFO [optim.py:369] (1/2) 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 17:12:15,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7248, 4.6679, 1.8815, 1.8970], device='cuda:1'), covar=tensor([0.0927, 0.0283, 0.0930, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0561, 0.0403, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 17:12:18,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-13 17:12:22,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4920, 2.1561, 1.6550, 0.7966], device='cuda:1'), covar=tensor([0.6201, 0.3091, 0.4420, 0.7220], device='cuda:1'), in_proj_covar=tensor([0.1811, 0.1702, 0.1644, 0.1477], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 17:12:26,639 INFO [train.py:968] (1/2) Epoch 26, batch 40150, giga_loss[loss=0.2603, simple_loss=0.3373, pruned_loss=0.09159, over 28793.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3468, pruned_loss=0.09647, over 5703858.50 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.3615, pruned_loss=0.1144, over 5694697.86 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3451, pruned_loss=0.09458, over 5704935.23 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:12:30,789 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1476, 1.3224, 1.1417, 0.9542], device='cuda:1'), covar=tensor([0.1027, 0.0520, 0.1102, 0.1126], device='cuda:1'), in_proj_covar=tensor([0.0404, 0.0447, 0.0522, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 17:12:35,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:13:07,424 INFO [train.py:968] (1/2) Epoch 26, batch 40200, giga_loss[loss=0.2366, simple_loss=0.3159, pruned_loss=0.07859, over 28494.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3459, pruned_loss=0.09629, over 5710538.55 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3612, pruned_loss=0.1142, over 5699908.35 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3445, pruned_loss=0.09452, over 5707031.25 frames. ], batch size: 60, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:13:16,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2649, 3.1114, 2.9434, 1.4198], device='cuda:1'), covar=tensor([0.1045, 0.1110, 0.0944, 0.2395], device='cuda:1'), in_proj_covar=tensor([0.1265, 0.1167, 0.0984, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 17:13:38,106 INFO [optim.py:369] (1/2) 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 17:13:48,578 INFO [train.py:968] (1/2) Epoch 26, batch 40250, giga_loss[loss=0.2724, simple_loss=0.3336, pruned_loss=0.1056, over 28768.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3446, pruned_loss=0.0967, over 5709988.26 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3608, pruned_loss=0.114, over 5702848.68 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3435, pruned_loss=0.09526, over 5704728.65 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:14:10,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8616, 2.1532, 2.0626, 1.6769], device='cuda:1'), covar=tensor([0.3676, 0.2609, 0.2853, 0.3176], device='cuda:1'), in_proj_covar=tensor([0.2041, 0.1979, 0.1905, 0.2040], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 17:14:23,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:14:31,986 INFO [train.py:968] (1/2) Epoch 26, batch 40300, giga_loss[loss=0.2225, simple_loss=0.3029, pruned_loss=0.07107, over 28934.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3437, pruned_loss=0.09744, over 5706048.51 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.114, over 5703552.44 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3426, pruned_loss=0.09622, over 5701302.76 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:14:35,293 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:14:37,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:14:39,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6175, 1.6721, 1.8044, 1.4154], device='cuda:1'), covar=tensor([0.2062, 0.2514, 0.1663, 0.1830], device='cuda:1'), in_proj_covar=tensor([0.0926, 0.0711, 0.0973, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:1')
+2023-03-13 17:14:44,078 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-13 17:14:56,432 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:15:05,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:15:05,540 INFO [optim.py:369] (1/2) 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 17:15:14,084 INFO [train.py:968] (1/2) Epoch 26, batch 40350, giga_loss[loss=0.2637, simple_loss=0.328, pruned_loss=0.09973, over 29041.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3411, pruned_loss=0.09705, over 5713281.50 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.114, over 5703552.44 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3403, pruned_loss=0.0961, over 5709587.83 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:15:28,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 17:15:54,274 INFO [train.py:968] (1/2) Epoch 26, batch 40400, giga_loss[loss=0.2529, simple_loss=0.3365, pruned_loss=0.08461, over 29033.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3401, pruned_loss=0.09637, over 5716936.60 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3613, pruned_loss=0.1141, over 5703269.62 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3387, pruned_loss=0.09515, over 5714618.63 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:16:13,445 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:16:34,391 INFO [train.py:968] (1/2) Epoch 26, batch 40450, giga_loss[loss=0.2319, simple_loss=0.3088, pruned_loss=0.07754, over 29096.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3378, pruned_loss=0.09499, over 5725563.86 frames. ], libri_tot_loss[loss=0.2945, simple_loss=0.3612, pruned_loss=0.1139, over 5709359.90 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3362, pruned_loss=0.09371, over 5718642.21 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:16:52,211 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:16:54,821 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:15,251 INFO [train.py:968] (1/2) Epoch 26, batch 40500, giga_loss[loss=0.2339, simple_loss=0.3083, pruned_loss=0.07969, over 28972.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3335, pruned_loss=0.09318, over 5721489.47 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3605, pruned_loss=0.1136, over 5710158.16 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3322, pruned_loss=0.09192, over 5715524.02 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:17:17,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:32,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6851, 5.5257, 5.2224, 2.9875], device='cuda:1'), covar=tensor([0.0408, 0.0541, 0.0588, 0.1427], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1175, 0.0989, 0.0739], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 17:17:36,602 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:43,915 INFO [optim.py:369] (1/2) 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 17:17:46,050 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:17:52,728 INFO [train.py:968] (1/2) Epoch 26, batch 40550, libri_loss[loss=0.2945, simple_loss=0.3653, pruned_loss=0.1119, over 29519.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3304, pruned_loss=0.09132, over 5722533.55 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3605, pruned_loss=0.1134, over 5712930.17 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3283, pruned_loss=0.08979, over 5715576.61 frames. ], batch size: 89, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:17:59,068 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1179615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:04,622 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:06,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:12,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4125, 2.2242, 1.6482, 1.5749], device='cuda:1'), covar=tensor([0.0799, 0.0246, 0.0305, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 17:18:29,562 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:18:31,997 INFO [train.py:968] (1/2) Epoch 26, batch 40600, giga_loss[loss=0.2347, simple_loss=0.3213, pruned_loss=0.074, over 28945.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3305, pruned_loss=0.09116, over 5723837.36 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1134, over 5717592.72 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3281, pruned_loss=0.08941, over 5714252.43 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:19:01,630 INFO [optim.py:369] (1/2) 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 17:19:11,894 INFO [train.py:968] (1/2) Epoch 26, batch 40650, giga_loss[loss=0.2732, simple_loss=0.3516, pruned_loss=0.09735, over 28543.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3335, pruned_loss=0.09266, over 5724394.23 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1134, over 5723842.25 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3305, pruned_loss=0.09039, over 5711459.04 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:19:26,421 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:19:54,353 INFO [train.py:968] (1/2) Epoch 26, batch 40700, giga_loss[loss=0.2165, simple_loss=0.301, pruned_loss=0.06601, over 28324.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3363, pruned_loss=0.09356, over 5716100.57 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3606, pruned_loss=0.1135, over 5724374.04 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3338, pruned_loss=0.0916, over 5705476.02 frames. ], batch size: 60, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:20:22,833 INFO [optim.py:369] (1/2) 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 17:20:32,129 INFO [train.py:968] (1/2) Epoch 26, batch 40750, giga_loss[loss=0.3029, simple_loss=0.3665, pruned_loss=0.1196, over 27645.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3401, pruned_loss=0.09508, over 5724182.31 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.361, pruned_loss=0.1138, over 5727043.84 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3372, pruned_loss=0.09279, over 5713295.82 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:21:04,786 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-13 17:21:14,082 INFO [train.py:968] (1/2) Epoch 26, batch 40800, giga_loss[loss=0.302, simple_loss=0.3754, pruned_loss=0.1142, over 28618.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3431, pruned_loss=0.0967, over 5722378.59 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.361, pruned_loss=0.1138, over 5727760.05 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3405, pruned_loss=0.09463, over 5713201.72 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:21:21,995 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1179867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:21:24,794 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1179870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:21:26,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7782, 2.0193, 1.4759, 1.7047], device='cuda:1'), covar=tensor([0.0860, 0.0491, 0.0887, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0449, 0.0523, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 17:21:46,117 INFO [optim.py:369] (1/2) 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 17:21:48,248 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1179899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:21:52,017 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 17:21:54,161 INFO [train.py:968] (1/2) Epoch 26, batch 40850, giga_loss[loss=0.2782, simple_loss=0.3492, pruned_loss=0.1036, over 28647.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.344, pruned_loss=0.09767, over 5717915.98 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3604, pruned_loss=0.1135, over 5731664.66 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.342, pruned_loss=0.09579, over 5706843.85 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:21:57,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4960, 1.6977, 1.4301, 1.5601], device='cuda:1'), covar=tensor([0.0745, 0.0310, 0.0331, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 17:22:43,093 INFO [train.py:968] (1/2) Epoch 26, batch 40900, giga_loss[loss=0.458, simple_loss=0.4533, pruned_loss=0.2314, over 23709.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3504, pruned_loss=0.1034, over 5691973.48 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5727851.08 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3484, pruned_loss=0.1018, over 5685762.93 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:22:45,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:22:57,883 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1179972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:23:16,119 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:23:32,026 INFO [train.py:968] (1/2) Epoch 26, batch 40950, giga_loss[loss=0.3017, simple_loss=0.3743, pruned_loss=0.1145, over 28901.00 frames. ], tot_loss[loss=0.287, simple_loss=0.357, pruned_loss=0.1085, over 5685145.21 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5728860.82 frames. ], giga_tot_loss[loss=0.285, simple_loss=0.3555, pruned_loss=0.1072, over 5679333.34 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:24:14,850 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:24:15,575 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:24:18,760 INFO [train.py:968] (1/2) Epoch 26, batch 41000, giga_loss[loss=0.3353, simple_loss=0.3936, pruned_loss=0.1385, over 28809.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3621, pruned_loss=0.1118, over 5684020.16 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5728979.10 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3611, pruned_loss=0.1108, over 5678265.71 frames. ], batch size: 119, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:24:45,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 17:24:52,533 INFO [optim.py:369] (1/2) 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 17:24:59,529 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:01,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:01,642 INFO [train.py:968] (1/2) Epoch 26, batch 41050, giga_loss[loss=0.3049, simple_loss=0.3733, pruned_loss=0.1182, over 28990.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3682, pruned_loss=0.1173, over 5680523.03 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3596, pruned_loss=0.1127, over 5732321.64 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3678, pruned_loss=0.1168, over 5671873.62 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:25:09,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3092, 1.8536, 1.2873, 0.5328], device='cuda:1'), covar=tensor([0.4424, 0.2738, 0.4646, 0.5848], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1712, 0.1650, 0.1484], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 17:25:09,945 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:12,049 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:25,267 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:28,484 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:39,283 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:47,888 INFO [train.py:968] (1/2) Epoch 26, batch 41100, giga_loss[loss=0.3255, simple_loss=0.3877, pruned_loss=0.1317, over 29086.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3741, pruned_loss=0.1221, over 5682126.82 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5735319.96 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3738, pruned_loss=0.1217, over 5671687.04 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:25:54,958 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:55,643 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:25:58,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:26:12,931 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.85 vs. limit=5.0
+2023-03-13 17:26:31,496 INFO [optim.py:369] (1/2) 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 17:26:42,637 INFO [train.py:968] (1/2) Epoch 26, batch 41150, giga_loss[loss=0.3125, simple_loss=0.3814, pruned_loss=0.1218, over 28973.00 frames. ], tot_loss[loss=0.3149, simple_loss=0.378, pruned_loss=0.126, over 5665633.36 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3598, pruned_loss=0.1129, over 5737168.49 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.378, pruned_loss=0.1257, over 5655079.31 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:27:14,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-13 17:27:20,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-13 17:27:34,688 INFO [train.py:968] (1/2) Epoch 26, batch 41200, giga_loss[loss=0.4098, simple_loss=0.4417, pruned_loss=0.189, over 28671.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3822, pruned_loss=0.1304, over 5653599.66 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.113, over 5740473.07 frames. ], giga_tot_loss[loss=0.3218, simple_loss=0.3826, pruned_loss=0.1305, over 5640706.50 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:28:17,165 INFO [optim.py:369] (1/2) 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 17:28:27,677 INFO [train.py:968] (1/2) Epoch 26, batch 41250, libri_loss[loss=0.2835, simple_loss=0.3555, pruned_loss=0.1058, over 29516.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3859, pruned_loss=0.1343, over 5633935.95 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5740259.81 frames. ], giga_tot_loss[loss=0.328, simple_loss=0.3865, pruned_loss=0.1348, over 5621934.37 frames. ], batch size: 82, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:28:38,717 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 17:28:52,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2032, 1.2852, 1.2221, 0.8647], device='cuda:1'), covar=tensor([0.1068, 0.0533, 0.1062, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0453, 0.0527, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 17:29:19,053 INFO [train.py:968] (1/2) Epoch 26, batch 41300, giga_loss[loss=0.3613, simple_loss=0.4192, pruned_loss=0.1516, over 29076.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3905, pruned_loss=0.1389, over 5621666.48 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.36, pruned_loss=0.113, over 5731150.94 frames. ], giga_tot_loss[loss=0.3355, simple_loss=0.3915, pruned_loss=0.1397, over 5619158.19 frames. ], batch size: 155, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:30:00,175 INFO [optim.py:369] (1/2) 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 17:30:09,694 INFO [train.py:968] (1/2) Epoch 26, batch 41350, giga_loss[loss=0.2842, simple_loss=0.3559, pruned_loss=0.1062, over 28419.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3923, pruned_loss=0.1401, over 5628941.52 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 5729316.83 frames. ], giga_tot_loss[loss=0.3385, simple_loss=0.394, pruned_loss=0.1415, over 5626801.78 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:30:32,232 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:30:33,128 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:30:40,126 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 17:30:57,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3718, 1.1256, 4.0673, 3.2952], device='cuda:1'), covar=tensor([0.1616, 0.2854, 0.0471, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0666, 0.0993, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 17:31:00,933 INFO [train.py:968] (1/2) Epoch 26, batch 41400, giga_loss[loss=0.2912, simple_loss=0.3599, pruned_loss=0.1112, over 28869.00 frames. ], tot_loss[loss=0.335, simple_loss=0.3904, pruned_loss=0.1398, over 5629479.77 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5732067.96 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3922, pruned_loss=0.1412, over 5623997.41 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:31:08,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:31:23,760 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:31:39,710 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 41450, libri_loss[loss=0.3781, simple_loss=0.423, pruned_loss=0.1666, over 19170.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3894, pruned_loss=0.1394, over 5624762.50 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5725640.43 frames. ], giga_tot_loss[loss=0.337, simple_loss=0.3917, pruned_loss=0.1412, over 5624802.93 frames. ], batch size: 187, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:31:52,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4355, 1.5095, 1.4277, 1.3779], device='cuda:1'), covar=tensor([0.2352, 0.2339, 0.2089, 0.2205], device='cuda:1'), in_proj_covar=tensor([0.2051, 0.1991, 0.1913, 0.2048], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 17:32:21,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:22,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:34,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:35,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1698, 1.4837, 1.1389, 0.5825], device='cuda:1'), covar=tensor([0.2927, 0.1997, 0.2589, 0.5169], device='cuda:1'), in_proj_covar=tensor([0.1817, 0.1712, 0.1649, 0.1486], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 17:32:42,272 INFO [train.py:968] (1/2) Epoch 26, batch 41500, giga_loss[loss=0.3431, simple_loss=0.3966, pruned_loss=0.1448, over 27491.00 frames. ], tot_loss[loss=0.3304, simple_loss=0.3873, pruned_loss=0.1367, over 5620817.97 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3591, pruned_loss=0.1124, over 5729003.65 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.39, pruned_loss=0.1389, over 5615894.36 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:32:53,731 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:54,387 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:57,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:32:59,962 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:33:20,595 INFO [optim.py:369] (1/2) 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 17:33:23,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6069, 1.1547, 4.2562, 3.5617], device='cuda:1'), covar=tensor([0.1567, 0.2942, 0.0484, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0666, 0.0993, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 17:33:23,565 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:33:24,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9051, 3.7286, 3.5340, 1.6987], device='cuda:1'), covar=tensor([0.0775, 0.0929, 0.0868, 0.2171], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1192, 0.1004, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 17:33:24,230 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:33:28,236 INFO [train.py:968] (1/2) Epoch 26, batch 41550, giga_loss[loss=0.3748, simple_loss=0.4209, pruned_loss=0.1644, over 27525.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3888, pruned_loss=0.1375, over 5613091.93 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1127, over 5723263.23 frames. ], giga_tot_loss[loss=0.3359, simple_loss=0.3918, pruned_loss=0.14, over 5610099.02 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:34:17,666 INFO [train.py:968] (1/2) Epoch 26, batch 41600, giga_loss[loss=0.3431, simple_loss=0.3953, pruned_loss=0.1455, over 27896.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3879, pruned_loss=0.1369, over 5602611.55 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3598, pruned_loss=0.113, over 5726413.93 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3913, pruned_loss=0.1396, over 5592980.61 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:34:43,753 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:44,377 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:45,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:47,111 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:34:55,299 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:35:06,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9552, 5.0891, 2.0765, 2.2677], device='cuda:1'), covar=tensor([0.0907, 0.0309, 0.0875, 0.1172], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0566, 0.0404, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 17:35:07,237 INFO [train.py:968] (1/2) Epoch 26, batch 41650, giga_loss[loss=0.2755, simple_loss=0.3574, pruned_loss=0.0968, over 28833.00 frames. ], tot_loss[loss=0.3253, simple_loss=0.3846, pruned_loss=0.133, over 5614866.54 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5723097.77 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3875, pruned_loss=0.1353, over 5607397.65 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:35:11,521 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:35:12,682 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:35:53,682 INFO [train.py:968] (1/2) Epoch 26, batch 41700, giga_loss[loss=0.2842, simple_loss=0.3597, pruned_loss=0.1044, over 28550.00 frames. ], tot_loss[loss=0.3217, simple_loss=0.3827, pruned_loss=0.1303, over 5625145.54 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.36, pruned_loss=0.1133, over 5719397.89 frames. ], giga_tot_loss[loss=0.326, simple_loss=0.3861, pruned_loss=0.1329, over 5619227.58 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:36:15,508 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-13 17:36:30,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 17:36:35,201 INFO [optim.py:369] (1/2) 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 17:36:44,646 INFO [train.py:968] (1/2) Epoch 26, batch 41750, giga_loss[loss=0.3227, simple_loss=0.3948, pruned_loss=0.1252, over 28939.00 frames. ], tot_loss[loss=0.3168, simple_loss=0.3788, pruned_loss=0.1274, over 5625999.74 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3596, pruned_loss=0.1131, over 5721143.87 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3821, pruned_loss=0.1298, over 5618640.26 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:36:50,420 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:37:01,514 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8774, 1.8864, 2.1154, 1.6423], device='cuda:1'), covar=tensor([0.1738, 0.2598, 0.1406, 0.1705], device='cuda:1'), in_proj_covar=tensor([0.0920, 0.0710, 0.0967, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 17:37:11,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180836.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 17:37:14,096 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1180839.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 17:37:15,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:37:30,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6156, 1.7069, 1.8008, 1.3994], device='cuda:1'), covar=tensor([0.1845, 0.2572, 0.1509, 0.1799], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0711, 0.0968, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 17:37:31,832 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:37:32,759 INFO [train.py:968] (1/2) Epoch 26, batch 41800, giga_loss[loss=0.2919, simple_loss=0.3637, pruned_loss=0.1101, over 28934.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3769, pruned_loss=0.1258, over 5611526.90 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3601, pruned_loss=0.1138, over 5707157.08 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3796, pruned_loss=0.1275, over 5614837.28 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:37:42,049 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1180868.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 17:38:13,087 INFO [optim.py:369] (1/2) 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 17:38:16,832 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-13 17:38:19,135 INFO [train.py:968] (1/2) Epoch 26, batch 41850, giga_loss[loss=0.296, simple_loss=0.383, pruned_loss=0.1045, over 28870.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3747, pruned_loss=0.124, over 5630617.86 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.114, over 5701302.84 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3769, pruned_loss=0.1253, over 5636187.79 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:38:36,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1180925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:38:50,979 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1180938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:04,814 INFO [train.py:968] (1/2) Epoch 26, batch 41900, giga_loss[loss=0.3487, simple_loss=0.3966, pruned_loss=0.1504, over 27586.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3738, pruned_loss=0.1236, over 5641213.31 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.36, pruned_loss=0.1138, over 5705629.05 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3762, pruned_loss=0.1251, over 5639862.24 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:39:26,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3179, 1.2200, 1.1766, 1.4757], device='cuda:1'), covar=tensor([0.0746, 0.0408, 0.0354, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 17:39:34,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:38,755 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:39:50,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1180998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:51,945 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:39:57,056 INFO [train.py:968] (1/2) Epoch 26, batch 41950, giga_loss[loss=0.2985, simple_loss=0.3718, pruned_loss=0.1126, over 28649.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3739, pruned_loss=0.1235, over 5641454.22 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.1141, over 5708050.85 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3757, pruned_loss=0.1247, over 5636713.98 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:40:10,267 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:40:27,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:40:30,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2854, 1.1634, 3.5889, 3.0898], device='cuda:1'), covar=tensor([0.1715, 0.2956, 0.0542, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0667, 0.0997, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 17:40:34,700 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-13 17:40:46,519 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.96 vs. limit=2.0
+2023-03-13 17:40:50,099 INFO [train.py:968] (1/2) Epoch 26, batch 42000, giga_loss[loss=0.2922, simple_loss=0.3782, pruned_loss=0.1031, over 28697.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3729, pruned_loss=0.1214, over 5631869.33 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3605, pruned_loss=0.1141, over 5702825.50 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3746, pruned_loss=0.1225, over 5631309.31 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:40:50,099 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 17:40:58,163 INFO [train.py:1012] (1/2) Epoch 26, validation: loss=0.202, simple_loss=0.3083, pruned_loss=0.04787, over 944034.00 frames. 
+2023-03-13 17:40:58,164 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 17:41:08,755 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:41:10,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:41:36,509 INFO [optim.py:369] (1/2) 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 17:41:39,424 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:41:47,770 INFO [train.py:968] (1/2) Epoch 26, batch 42050, giga_loss[loss=0.2843, simple_loss=0.3598, pruned_loss=0.1044, over 28476.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3734, pruned_loss=0.1193, over 5632917.08 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1144, over 5688891.95 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3749, pruned_loss=0.1201, over 5643124.12 frames. ], batch size: 85, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:42:29,206 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.63 vs. limit=5.0
+2023-03-13 17:42:33,513 INFO [train.py:968] (1/2) Epoch 26, batch 42100, giga_loss[loss=0.2709, simple_loss=0.3559, pruned_loss=0.09299, over 28967.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3744, pruned_loss=0.1196, over 5640878.54 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3609, pruned_loss=0.1146, over 5682977.13 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3757, pruned_loss=0.1202, over 5653336.23 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 17:43:05,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:43:13,902 INFO [optim.py:369] (1/2) 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 17:43:21,244 INFO [train.py:968] (1/2) Epoch 26, batch 42150, giga_loss[loss=0.2924, simple_loss=0.3664, pruned_loss=0.1092, over 28632.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3743, pruned_loss=0.1199, over 5644637.57 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 5683736.24 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3753, pruned_loss=0.1203, over 5653433.86 frames. ], batch size: 60, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:44:07,486 INFO [train.py:968] (1/2) Epoch 26, batch 42200, libri_loss[loss=0.3081, simple_loss=0.3718, pruned_loss=0.1222, over 29513.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3728, pruned_loss=0.1198, over 5656898.08 frames. ], libri_tot_loss[loss=0.2951, simple_loss=0.361, pruned_loss=0.1147, over 5687103.44 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3739, pruned_loss=0.1202, over 5660249.26 frames. ], batch size: 84, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:44:36,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5503, 2.0700, 1.3317, 0.9321], device='cuda:1'), covar=tensor([0.7296, 0.3771, 0.3352, 0.6675], device='cuda:1'), in_proj_covar=tensor([0.1819, 0.1715, 0.1652, 0.1488], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 17:44:49,867 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 42250, giga_loss[loss=0.3129, simple_loss=0.3716, pruned_loss=0.1271, over 28992.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3723, pruned_loss=0.1212, over 5654319.21 frames. ], libri_tot_loss[loss=0.2949, simple_loss=0.3608, pruned_loss=0.1145, over 5689078.92 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3734, pruned_loss=0.1217, over 5655037.03 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:45:02,643 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:05,516 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:06,869 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:20,690 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:23,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:45:43,819 INFO [train.py:968] (1/2) Epoch 26, batch 42300, giga_loss[loss=0.2593, simple_loss=0.3331, pruned_loss=0.09278, over 29063.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3721, pruned_loss=0.121, over 5649038.40 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3614, pruned_loss=0.115, over 5680433.29 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3727, pruned_loss=0.1211, over 5656209.65 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:45:52,706 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:46:25,015 INFO [optim.py:369] (1/2) 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 17:46:30,472 INFO [train.py:968] (1/2) Epoch 26, batch 42350, giga_loss[loss=0.2835, simple_loss=0.3612, pruned_loss=0.1029, over 28828.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3709, pruned_loss=0.1184, over 5654792.07 frames. ], libri_tot_loss[loss=0.2955, simple_loss=0.3612, pruned_loss=0.1149, over 5673797.21 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3717, pruned_loss=0.1186, over 5666039.26 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:47:16,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:47:16,806 INFO [train.py:968] (1/2) Epoch 26, batch 42400, giga_loss[loss=0.3912, simple_loss=0.414, pruned_loss=0.1842, over 23611.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3715, pruned_loss=0.1188, over 5662916.50 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3604, pruned_loss=0.1144, over 5676495.90 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3731, pruned_loss=0.1195, over 5669186.48 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:47:20,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:47:48,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:48:00,468 INFO [optim.py:369] (1/2) 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 17:48:07,334 INFO [train.py:968] (1/2) Epoch 26, batch 42450, giga_loss[loss=0.3034, simple_loss=0.3691, pruned_loss=0.1188, over 28934.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3722, pruned_loss=0.1198, over 5661456.48 frames. ], libri_tot_loss[loss=0.2952, simple_loss=0.3609, pruned_loss=0.1147, over 5679227.74 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3731, pruned_loss=0.1201, over 5663806.46 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:48:50,775 INFO [train.py:968] (1/2) Epoch 26, batch 42500, giga_loss[loss=0.3164, simple_loss=0.3783, pruned_loss=0.1272, over 28940.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3693, pruned_loss=0.1179, over 5677346.14 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3604, pruned_loss=0.1141, over 5687036.57 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3709, pruned_loss=0.1188, over 5671788.64 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:49:29,708 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 42550, giga_loss[loss=0.3057, simple_loss=0.372, pruned_loss=0.1197, over 28604.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3684, pruned_loss=0.118, over 5664589.63 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3602, pruned_loss=0.114, over 5679416.61 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.37, pruned_loss=0.1191, over 5666750.69 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:49:50,194 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-13 17:50:20,551 INFO [train.py:968] (1/2) Epoch 26, batch 42600, giga_loss[loss=0.3098, simple_loss=0.3739, pruned_loss=0.1229, over 28520.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.366, pruned_loss=0.1168, over 5672035.80 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5681677.40 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3679, pruned_loss=0.1181, over 5671289.59 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:50:21,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:50:50,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:50:53,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1181694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:50:56,929 INFO [optim.py:369] (1/2) 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 17:51:05,662 INFO [train.py:968] (1/2) Epoch 26, batch 42650, giga_loss[loss=0.3567, simple_loss=0.3962, pruned_loss=0.1586, over 26874.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3646, pruned_loss=0.1163, over 5679261.39 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5687565.58 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3664, pruned_loss=0.1175, over 5673200.66 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:51:15,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6514, 1.8986, 1.5980, 1.7032], device='cuda:1'), covar=tensor([0.2038, 0.2499, 0.2593, 0.2398], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0758, 0.0729, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 17:51:46,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2085, 1.4610, 1.3799, 1.1550], device='cuda:1'), covar=tensor([0.2975, 0.2902, 0.1962, 0.2582], device='cuda:1'), in_proj_covar=tensor([0.2057, 0.1995, 0.1921, 0.2051], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 17:51:51,057 INFO [train.py:968] (1/2) Epoch 26, batch 42700, giga_loss[loss=0.3124, simple_loss=0.3789, pruned_loss=0.1229, over 28530.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3634, pruned_loss=0.1162, over 5678260.39 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3595, pruned_loss=0.1133, over 5694098.94 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3652, pruned_loss=0.1175, over 5667278.81 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:51:51,383 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6130, 1.7243, 1.7456, 1.5548], device='cuda:1'), covar=tensor([0.2156, 0.2488, 0.2605, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0759, 0.0729, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 17:52:34,448 INFO [optim.py:369] (1/2) 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 17:52:40,244 INFO [train.py:968] (1/2) Epoch 26, batch 42750, libri_loss[loss=0.3154, simple_loss=0.3865, pruned_loss=0.1221, over 29382.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3642, pruned_loss=0.1177, over 5658713.11 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3603, pruned_loss=0.1139, over 5686621.72 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.365, pruned_loss=0.1183, over 5655551.03 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:52:57,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5312, 1.7214, 1.7390, 1.3092], device='cuda:1'), covar=tensor([0.1715, 0.2863, 0.1544, 0.1882], device='cuda:1'), in_proj_covar=tensor([0.0921, 0.0712, 0.0970, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 17:53:05,481 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:07,384 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1181837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:08,104 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:09,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1181840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:25,799 INFO [train.py:968] (1/2) Epoch 26, batch 42800, giga_loss[loss=0.2951, simple_loss=0.3683, pruned_loss=0.111, over 28869.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3641, pruned_loss=0.1166, over 5667620.76 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.36, pruned_loss=0.1135, over 5692362.74 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1175, over 5659490.86 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:53:35,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:53:37,210 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1181869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:54:08,004 INFO [optim.py:369] (1/2) 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 17:54:08,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2936, 1.2725, 3.5627, 3.1471], device='cuda:1'), covar=tensor([0.1597, 0.2843, 0.0493, 0.1715], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0668, 0.0998, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 17:54:12,045 INFO [train.py:968] (1/2) Epoch 26, batch 42850, giga_loss[loss=0.3321, simple_loss=0.3887, pruned_loss=0.1377, over 28568.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3646, pruned_loss=0.116, over 5674466.40 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3599, pruned_loss=0.1134, over 5694490.20 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3656, pruned_loss=0.1169, over 5665711.53 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:54:50,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8329, 3.6648, 3.5131, 1.7932], device='cuda:1'), covar=tensor([0.0816, 0.0932, 0.0915, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.1198, 0.1008, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 17:54:55,887 INFO [train.py:968] (1/2) Epoch 26, batch 42900, giga_loss[loss=0.2741, simple_loss=0.3575, pruned_loss=0.09538, over 28951.00 frames. ], tot_loss[loss=0.297, simple_loss=0.364, pruned_loss=0.115, over 5679926.70 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3594, pruned_loss=0.113, over 5699488.90 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3654, pruned_loss=0.1161, over 5668056.80 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:55:17,386 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1181979.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 17:55:36,665 INFO [optim.py:369] (1/2) 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 17:55:40,977 INFO [train.py:968] (1/2) Epoch 26, batch 42950, giga_loss[loss=0.2665, simple_loss=0.3469, pruned_loss=0.09304, over 28956.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3647, pruned_loss=0.1156, over 5694277.27 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1123, over 5708998.57 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3668, pruned_loss=0.1173, over 5674975.57 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:56:08,084 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:56:32,048 INFO [train.py:968] (1/2) Epoch 26, batch 43000, giga_loss[loss=0.3144, simple_loss=0.3729, pruned_loss=0.128, over 28927.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3691, pruned_loss=0.1195, over 5694208.45 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1124, over 5710749.31 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3708, pruned_loss=0.1208, over 5677287.72 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:56:47,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1182073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:57:14,800 INFO [optim.py:369] (1/2) 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 17:57:19,664 INFO [train.py:968] (1/2) Epoch 26, batch 43050, giga_loss[loss=0.3873, simple_loss=0.417, pruned_loss=0.1789, over 26540.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3695, pruned_loss=0.1208, over 5686249.60 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1127, over 5702640.62 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 5679531.45 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:58:01,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4741, 1.6493, 1.5731, 1.3777], device='cuda:1'), covar=tensor([0.3067, 0.2561, 0.2109, 0.2776], device='cuda:1'), in_proj_covar=tensor([0.2050, 0.1993, 0.1915, 0.2046], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 17:58:07,937 INFO [train.py:968] (1/2) Epoch 26, batch 43100, giga_loss[loss=0.3331, simple_loss=0.386, pruned_loss=0.1401, over 28326.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3698, pruned_loss=0.1219, over 5675797.84 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.359, pruned_loss=0.1126, over 5696255.09 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3714, pruned_loss=0.1232, over 5674513.38 frames. ], batch size: 368, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:58:28,513 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:58:32,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:58:53,395 INFO [optim.py:369] (1/2) 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 17:58:58,067 INFO [train.py:968] (1/2) Epoch 26, batch 43150, giga_loss[loss=0.2941, simple_loss=0.3663, pruned_loss=0.1109, over 28488.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3725, pruned_loss=0.1243, over 5661181.29 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3591, pruned_loss=0.1125, over 5698106.14 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3738, pruned_loss=0.1256, over 5658162.69 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 17:58:59,624 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 17:59:07,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5460, 1.7148, 1.7555, 1.3410], device='cuda:1'), covar=tensor([0.1699, 0.2416, 0.1408, 0.1683], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0713, 0.0970, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 17:59:41,702 INFO [train.py:968] (1/2) Epoch 26, batch 43200, libri_loss[loss=0.2833, simple_loss=0.3594, pruned_loss=0.1036, over 28980.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3709, pruned_loss=0.1234, over 5670344.30 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3591, pruned_loss=0.1124, over 5704644.61 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3724, pruned_loss=0.1248, over 5660805.70 frames. ], batch size: 107, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 17:59:51,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5976, 1.7765, 1.7149, 1.4967], device='cuda:1'), covar=tensor([0.2070, 0.2406, 0.2532, 0.2517], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0763, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 18:00:21,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4135, 1.6510, 1.6590, 1.4684], device='cuda:1'), covar=tensor([0.2197, 0.2377, 0.2516, 0.2388], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0763, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 18:00:25,639 INFO [optim.py:369] (1/2) 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 18:00:29,601 INFO [train.py:968] (1/2) Epoch 26, batch 43250, giga_loss[loss=0.2681, simple_loss=0.3482, pruned_loss=0.09402, over 28645.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3715, pruned_loss=0.1229, over 5671497.97 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1124, over 5706719.97 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3728, pruned_loss=0.1241, over 5662052.21 frames. ], batch size: 92, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:00:56,745 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6078, 1.6002, 1.7975, 1.3511], device='cuda:1'), covar=tensor([0.1883, 0.2721, 0.1566, 0.1852], device='cuda:1'), in_proj_covar=tensor([0.0923, 0.0713, 0.0971, 0.0871], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 18:00:57,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-13 18:01:13,269 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182354.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:01:15,209 INFO [train.py:968] (1/2) Epoch 26, batch 43300, giga_loss[loss=0.2646, simple_loss=0.3463, pruned_loss=0.09147, over 28558.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3703, pruned_loss=0.1205, over 5672247.58 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1125, over 5711519.98 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3713, pruned_loss=0.1216, over 5659206.54 frames. ], batch size: 78, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:01:24,112 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1182365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:01:59,618 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 18:01:59,776 INFO [optim.py:369] (1/2) 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 18:02:03,399 INFO [train.py:968] (1/2) Epoch 26, batch 43350, giga_loss[loss=0.3079, simple_loss=0.3683, pruned_loss=0.1237, over 28895.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3681, pruned_loss=0.1193, over 5670124.72 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3591, pruned_loss=0.1123, over 5712254.87 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3693, pruned_loss=0.1206, over 5658277.08 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:02:40,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:02:47,774 INFO [train.py:968] (1/2) Epoch 26, batch 43400, giga_loss[loss=0.3034, simple_loss=0.3711, pruned_loss=0.1178, over 28937.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3658, pruned_loss=0.1181, over 5678943.06 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1125, over 5716974.19 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3667, pruned_loss=0.1191, over 5664649.61 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:03:10,773 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5149, 1.6776, 1.6341, 1.4376], device='cuda:1'), covar=tensor([0.1826, 0.2151, 0.2283, 0.2231], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0763, 0.0733, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 18:03:23,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182497.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:03:26,924 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182500.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:03:29,787 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 43450, giga_loss[loss=0.3114, simple_loss=0.377, pruned_loss=0.1229, over 28870.00 frames. ], tot_loss[loss=0.302, simple_loss=0.366, pruned_loss=0.119, over 5672503.60 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1124, over 5709749.62 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3669, pruned_loss=0.1199, over 5667308.13 frames. ], batch size: 186, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:03:54,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182529.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:04:22,665 INFO [train.py:968] (1/2) Epoch 26, batch 43500, giga_loss[loss=0.3043, simple_loss=0.3806, pruned_loss=0.114, over 28542.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.122, over 5659722.51 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1125, over 5703863.82 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3714, pruned_loss=0.1228, over 5660282.90 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:04:44,792 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8163, 3.6594, 3.4448, 1.7603], device='cuda:1'), covar=tensor([0.0835, 0.0955, 0.1006, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1199, 0.1011, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 18:04:53,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:04:57,141 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3155, 1.3709, 3.7396, 3.2961], device='cuda:1'), covar=tensor([0.1662, 0.2714, 0.0503, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0671, 0.0999, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 18:04:57,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:05:05,887 INFO [optim.py:369] (1/2) 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 18:05:11,296 INFO [train.py:968] (1/2) Epoch 26, batch 43550, giga_loss[loss=0.3068, simple_loss=0.3895, pruned_loss=0.1121, over 29082.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3729, pruned_loss=0.1205, over 5666541.11 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1123, over 5705849.46 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3739, pruned_loss=0.1213, over 5665007.94 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:05:28,874 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:05:38,194 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3960, 1.6949, 1.6520, 1.5389], device='cuda:1'), covar=tensor([0.1855, 0.1631, 0.1760, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0760, 0.0730, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 18:05:59,052 INFO [train.py:968] (1/2) Epoch 26, batch 43600, giga_loss[loss=0.3243, simple_loss=0.395, pruned_loss=0.1268, over 28782.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3725, pruned_loss=0.1195, over 5661411.45 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3587, pruned_loss=0.1123, over 5702909.17 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.374, pruned_loss=0.1204, over 5661104.40 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:06:14,893 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6033, 1.5722, 1.8058, 1.4183], device='cuda:1'), covar=tensor([0.1579, 0.2405, 0.1319, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0714, 0.0973, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 18:06:39,956 INFO [optim.py:369] (1/2) 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 18:06:43,276 INFO [train.py:968] (1/2) Epoch 26, batch 43650, giga_loss[loss=0.3049, simple_loss=0.377, pruned_loss=0.1164, over 29000.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3746, pruned_loss=0.121, over 5666666.87 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3586, pruned_loss=0.1122, over 5707779.00 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3765, pruned_loss=0.1221, over 5660573.20 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:07:14,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1182740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:07:30,979 INFO [train.py:968] (1/2) Epoch 26, batch 43700, giga_loss[loss=0.3332, simple_loss=0.3971, pruned_loss=0.1346, over 28615.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3747, pruned_loss=0.1214, over 5664092.05 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3583, pruned_loss=0.112, over 5702891.68 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3767, pruned_loss=0.1226, over 5662245.56 frames. ], batch size: 262, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:07:53,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2278, 1.8037, 1.4130, 0.4682], device='cuda:1'), covar=tensor([0.5214, 0.3326, 0.4769, 0.6822], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1719, 0.1652, 0.1488], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 18:08:09,056 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 43750, libri_loss[loss=0.3171, simple_loss=0.3782, pruned_loss=0.128, over 29532.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3741, pruned_loss=0.1221, over 5673818.27 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3577, pruned_loss=0.1117, over 5710573.03 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3769, pruned_loss=0.1237, over 5663902.36 frames. ], batch size: 84, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:08:28,066 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2634, 1.4790, 1.3994, 1.1698], device='cuda:1'), covar=tensor([0.2868, 0.2883, 0.1916, 0.2564], device='cuda:1'), in_proj_covar=tensor([0.2048, 0.1997, 0.1919, 0.2052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 18:08:58,418 INFO [train.py:968] (1/2) Epoch 26, batch 43800, giga_loss[loss=0.3059, simple_loss=0.3687, pruned_loss=0.1215, over 27958.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.373, pruned_loss=0.1221, over 5666338.96 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1118, over 5709482.74 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3753, pruned_loss=0.1235, over 5658131.16 frames. ], batch size: 412, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:09:05,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 18:09:23,136 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1182883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:09:25,328 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1182886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:09:29,413 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-13 18:09:39,589 INFO [optim.py:369] (1/2) 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 18:09:41,553 INFO [train.py:968] (1/2) Epoch 26, batch 43850, giga_loss[loss=0.2804, simple_loss=0.3523, pruned_loss=0.1042, over 28895.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3705, pruned_loss=0.1211, over 5662619.37 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3578, pruned_loss=0.1117, over 5702915.26 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.373, pruned_loss=0.1228, over 5660215.22 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:09:50,001 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1182915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:10:27,351 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-13 18:10:30,014 INFO [train.py:968] (1/2) Epoch 26, batch 43900, giga_loss[loss=0.2928, simple_loss=0.3664, pruned_loss=0.1096, over 28919.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1211, over 5669076.04 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3579, pruned_loss=0.1117, over 5703456.01 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3716, pruned_loss=0.1226, over 5666081.96 frames. ], batch size: 145, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:10:49,759 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1182975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:11:06,867 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-13 18:11:16,216 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1403, 1.2707, 1.1445, 0.8648], device='cuda:1'), covar=tensor([0.1179, 0.0598, 0.1158, 0.1175], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0455, 0.0528, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 18:11:17,930 INFO [optim.py:369] (1/2) 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 18:11:20,834 INFO [train.py:968] (1/2) Epoch 26, batch 43950, giga_loss[loss=0.2678, simple_loss=0.3487, pruned_loss=0.09344, over 28978.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3708, pruned_loss=0.1226, over 5678480.48 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1118, over 5709230.02 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3729, pruned_loss=0.124, over 5670044.86 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:12:11,460 INFO [train.py:968] (1/2) Epoch 26, batch 44000, giga_loss[loss=0.2931, simple_loss=0.3619, pruned_loss=0.1122, over 28688.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3712, pruned_loss=0.1233, over 5669745.65 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1118, over 5710271.87 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3729, pruned_loss=0.1244, over 5662200.88 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:12:12,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-13 18:12:40,027 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-13 18:12:53,330 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 44050, giga_loss[loss=0.2738, simple_loss=0.3421, pruned_loss=0.1027, over 28898.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3682, pruned_loss=0.1213, over 5676146.71 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3578, pruned_loss=0.1117, over 5714099.53 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3698, pruned_loss=0.1225, over 5666163.79 frames. ], batch size: 199, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:13:19,965 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 44100, giga_loss[loss=0.3095, simple_loss=0.3769, pruned_loss=0.1211, over 28549.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3672, pruned_loss=0.1202, over 5669824.19 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3577, pruned_loss=0.1116, over 5708508.69 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3686, pruned_loss=0.1213, over 5666745.05 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:14:24,722 INFO [optim.py:369] (1/2) 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 18:14:27,754 INFO [train.py:968] (1/2) Epoch 26, batch 44150, giga_loss[loss=0.2846, simple_loss=0.3624, pruned_loss=0.1034, over 28902.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3687, pruned_loss=0.1204, over 5672284.02 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3577, pruned_loss=0.1115, over 5714093.73 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3702, pruned_loss=0.1217, over 5663358.05 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:14:38,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1183217.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:15:14,001 INFO [train.py:968] (1/2) Epoch 26, batch 44200, giga_loss[loss=0.29, simple_loss=0.3564, pruned_loss=0.1118, over 28891.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3693, pruned_loss=0.1208, over 5677034.09 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5714397.22 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3708, pruned_loss=0.1221, over 5669076.60 frames. ], batch size: 227, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:15:56,093 INFO [optim.py:369] (1/2) 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 18:15:59,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1313, 1.4794, 1.1845, 0.5204], device='cuda:1'), covar=tensor([0.3081, 0.1890, 0.2325, 0.5179], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1726, 0.1653, 0.1493], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 18:15:59,382 INFO [train.py:968] (1/2) Epoch 26, batch 44250, giga_loss[loss=0.3613, simple_loss=0.3896, pruned_loss=0.1665, over 23811.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3689, pruned_loss=0.1208, over 5668864.64 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3576, pruned_loss=0.1112, over 5718073.04 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3707, pruned_loss=0.1224, over 5658141.98 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:16:38,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:16:42,961 INFO [train.py:968] (1/2) Epoch 26, batch 44300, giga_loss[loss=0.3014, simple_loss=0.3839, pruned_loss=0.1095, over 28983.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3714, pruned_loss=0.12, over 5673124.16 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3578, pruned_loss=0.1113, over 5719102.31 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3729, pruned_loss=0.1214, over 5662165.58 frames. ], batch size: 136, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:17:24,446 INFO [optim.py:369] (1/2) 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 18:17:27,341 INFO [train.py:968] (1/2) Epoch 26, batch 44350, giga_loss[loss=0.2877, simple_loss=0.3761, pruned_loss=0.09967, over 28963.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3728, pruned_loss=0.1184, over 5687259.59 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3578, pruned_loss=0.1114, over 5718941.51 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3742, pruned_loss=0.1196, over 5678295.21 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:17:32,448 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6928, 5.0380, 1.7702, 2.2232], device='cuda:1'), covar=tensor([0.0994, 0.0444, 0.0967, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0569, 0.0405, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 18:17:51,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-13 18:18:18,528 INFO [train.py:968] (1/2) Epoch 26, batch 44400, giga_loss[loss=0.2919, simple_loss=0.3686, pruned_loss=0.1076, over 28797.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3769, pruned_loss=0.121, over 5690333.69 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3577, pruned_loss=0.1114, over 5721768.96 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3784, pruned_loss=0.1221, over 5680160.94 frames. ], batch size: 99, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 18:18:18,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5287, 1.8516, 1.4850, 1.5416], device='cuda:1'), covar=tensor([0.2823, 0.2762, 0.3203, 0.2405], device='cuda:1'), in_proj_covar=tensor([0.1575, 0.1136, 0.1392, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 18:18:24,778 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1183462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:18:54,689 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1183493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:18:56,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:19:05,177 INFO [optim.py:369] (1/2) 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 18:19:07,169 INFO [train.py:968] (1/2) Epoch 26, batch 44450, giga_loss[loss=0.3349, simple_loss=0.3995, pruned_loss=0.1351, over 28605.00 frames. ], tot_loss[loss=0.3151, simple_loss=0.3805, pruned_loss=0.1248, over 5688082.17 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5726352.94 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3823, pruned_loss=0.1259, over 5674844.91 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:19:10,380 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:19:26,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1183525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:19:30,803 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6746, 1.9258, 1.9920, 1.5916], device='cuda:1'), covar=tensor([0.2991, 0.2631, 0.2579, 0.2912], device='cuda:1'), in_proj_covar=tensor([0.2050, 0.1995, 0.1921, 0.2047], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 18:19:58,060 INFO [train.py:968] (1/2) Epoch 26, batch 44500, giga_loss[loss=0.2882, simple_loss=0.3661, pruned_loss=0.1051, over 28884.00 frames. ], tot_loss[loss=0.3182, simple_loss=0.3817, pruned_loss=0.1274, over 5659354.75 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3578, pruned_loss=0.1116, over 5720800.46 frames. ], giga_tot_loss[loss=0.3202, simple_loss=0.3835, pruned_loss=0.1285, over 5653384.48 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:20:00,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1183559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:20:03,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.10 vs. limit=5.0
+2023-03-13 18:20:31,365 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-13 18:20:32,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183592.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:20:36,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5481, 1.7460, 1.2217, 1.3382], device='cuda:1'), covar=tensor([0.1014, 0.0590, 0.1059, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0452, 0.0525, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 18:20:42,042 INFO [optim.py:369] (1/2) 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 18:20:44,435 INFO [train.py:968] (1/2) Epoch 26, batch 44550, giga_loss[loss=0.3198, simple_loss=0.3857, pruned_loss=0.127, over 29015.00 frames. ], tot_loss[loss=0.318, simple_loss=0.3811, pruned_loss=0.1274, over 5659762.89 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1117, over 5716465.64 frames. ], giga_tot_loss[loss=0.32, simple_loss=0.3829, pruned_loss=0.1286, over 5657488.03 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:21:26,570 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:21:30,364 INFO [train.py:968] (1/2) Epoch 26, batch 44600, giga_loss[loss=0.353, simple_loss=0.3951, pruned_loss=0.1554, over 26548.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3786, pruned_loss=0.125, over 5662329.86 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3578, pruned_loss=0.1116, over 5719512.68 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3805, pruned_loss=0.1262, over 5657116.11 frames. ], batch size: 555, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:21:54,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1183684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:22:14,036 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 44650, giga_loss[loss=0.3308, simple_loss=0.3987, pruned_loss=0.1314, over 28458.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.378, pruned_loss=0.1224, over 5670680.93 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3579, pruned_loss=0.1116, over 5720223.49 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3798, pruned_loss=0.1235, over 5664530.68 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:22:41,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8175, 2.2025, 2.1315, 1.6171], device='cuda:1'), covar=tensor([0.3388, 0.2596, 0.2551, 0.3146], device='cuda:1'), in_proj_covar=tensor([0.2049, 0.1996, 0.1916, 0.2048], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 18:22:43,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1183735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 18:22:45,495 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183738.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 18:23:00,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-13 18:23:02,534 INFO [train.py:968] (1/2) Epoch 26, batch 44700, giga_loss[loss=0.3037, simple_loss=0.3745, pruned_loss=0.1165, over 28704.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3779, pruned_loss=0.1216, over 5667985.77 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3579, pruned_loss=0.1116, over 5722358.89 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3799, pruned_loss=0.1228, over 5660020.43 frames. ], batch size: 242, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:23:14,804 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1183767.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 18:23:48,392 INFO [optim.py:369] (1/2) 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 18:23:49,655 INFO [train.py:968] (1/2) Epoch 26, batch 44750, giga_loss[loss=0.2976, simple_loss=0.3701, pruned_loss=0.1125, over 29053.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3777, pruned_loss=0.1221, over 5673648.09 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3578, pruned_loss=0.1115, over 5725347.77 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3798, pruned_loss=0.1234, over 5663310.85 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:23:51,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4083, 1.7049, 1.4694, 1.5675], device='cuda:1'), covar=tensor([0.0790, 0.0335, 0.0327, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 18:24:20,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:24:28,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0779, 3.9171, 3.7367, 1.7742], device='cuda:1'), covar=tensor([0.0704, 0.0817, 0.0777, 0.2156], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.1213, 0.1023, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 18:24:37,676 INFO [train.py:968] (1/2) Epoch 26, batch 44800, giga_loss[loss=0.2723, simple_loss=0.3473, pruned_loss=0.09864, over 28737.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3765, pruned_loss=0.1215, over 5684257.57 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1116, over 5726821.16 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3781, pruned_loss=0.1225, over 5674357.53 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 8.0
+2023-03-13 18:25:26,156 INFO [optim.py:369] (1/2) 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 18:25:26,829 INFO [train.py:968] (1/2) Epoch 26, batch 44850, libri_loss[loss=0.3329, simple_loss=0.39, pruned_loss=0.1379, over 20514.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3761, pruned_loss=0.123, over 5643626.35 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3584, pruned_loss=0.1119, over 5708372.08 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3776, pruned_loss=0.1238, over 5650605.08 frames. ], batch size: 187, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:25:55,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1183934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:26:15,629 INFO [train.py:968] (1/2) Epoch 26, batch 44900, giga_loss[loss=0.2808, simple_loss=0.3507, pruned_loss=0.1054, over 28700.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.374, pruned_loss=0.1224, over 5651804.30 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3584, pruned_loss=0.1119, over 5710480.82 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3753, pruned_loss=0.1232, over 5654882.95 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:26:36,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1183980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:26:38,487 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1183983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:27:03,442 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 26, batch 44950, libri_loss[loss=0.2878, simple_loss=0.3595, pruned_loss=0.1081, over 29538.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3713, pruned_loss=0.1211, over 5655614.19 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5715103.08 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3723, pruned_loss=0.1218, over 5652762.84 frames. ], batch size: 83, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:27:07,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1184012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:27:15,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-13 18:27:50,363 INFO [train.py:968] (1/2) Epoch 26, batch 45000, libri_loss[loss=0.3048, simple_loss=0.37, pruned_loss=0.1198, over 28686.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3699, pruned_loss=0.1208, over 5660306.74 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3591, pruned_loss=0.1123, over 5716518.59 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1213, over 5655387.22 frames. ], batch size: 106, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:27:50,363 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 18:27:53,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3068, 1.2757, 1.1278, 1.4766], device='cuda:1'), covar=tensor([0.0865, 0.0385, 0.0388, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 18:27:58,374 INFO [train.py:1012] (1/2) Epoch 26, validation: loss=0.2047, simple_loss=0.3142, pruned_loss=0.04765, over 944034.00 frames. 
+2023-03-13 18:27:58,374 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 18:28:17,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1184077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:28:21,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1184080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:28:34,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.39 vs. limit=5.0
+2023-03-13 18:28:42,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5590, 2.0821, 1.3293, 0.8579], device='cuda:1'), covar=tensor([0.7715, 0.3574, 0.3411, 0.6767], device='cuda:1'), in_proj_covar=tensor([0.1831, 0.1731, 0.1658, 0.1494], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 18:28:43,766 INFO [optim.py:369] (1/2) 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 18:28:45,013 INFO [train.py:968] (1/2) Epoch 26, batch 45050, giga_loss[loss=0.3222, simple_loss=0.3823, pruned_loss=0.1311, over 28616.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3689, pruned_loss=0.1201, over 5664302.15 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3591, pruned_loss=0.1123, over 5717465.70 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3697, pruned_loss=0.1206, over 5659196.86 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:28:47,327 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1184109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:29:01,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-13 18:29:31,346 INFO [train.py:968] (1/2) Epoch 26, batch 45100, giga_loss[loss=0.2525, simple_loss=0.3405, pruned_loss=0.0823, over 28668.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.365, pruned_loss=0.116, over 5665226.63 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3592, pruned_loss=0.1122, over 5722256.27 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3657, pruned_loss=0.1166, over 5655714.75 frames. ], batch size: 307, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:30:09,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3043, 3.2296, 1.4872, 1.4148], device='cuda:1'), covar=tensor([0.1046, 0.0377, 0.0939, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0568, 0.0406, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 18:30:10,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7015, 1.8545, 1.3593, 1.4876], device='cuda:1'), covar=tensor([0.1017, 0.0693, 0.0997, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0450, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 18:30:10,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7240, 1.1303, 1.1053, 0.9562], device='cuda:1'), covar=tensor([0.2010, 0.1377, 0.2205, 0.1612], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0763, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 18:30:13,281 INFO [optim.py:369] (1/2) 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 18:30:13,936 INFO [train.py:968] (1/2) Epoch 26, batch 45150, libri_loss[loss=0.2394, simple_loss=0.3075, pruned_loss=0.08565, over 29350.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3631, pruned_loss=0.114, over 5674269.94 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3589, pruned_loss=0.1122, over 5722826.48 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3641, pruned_loss=0.1146, over 5664455.04 frames. ], batch size: 71, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:30:44,694 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 18:31:04,706 INFO [train.py:968] (1/2) Epoch 26, batch 45200, libri_loss[loss=0.3144, simple_loss=0.3765, pruned_loss=0.1261, over 19700.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3628, pruned_loss=0.1141, over 5656670.72 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5717760.46 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3635, pruned_loss=0.1144, over 5653272.30 frames. ], batch size: 187, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:31:10,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:31:20,999 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-13 18:32:00,335 INFO [train.py:968] (1/2) Epoch 26, batch 45250, giga_loss[loss=0.2917, simple_loss=0.3565, pruned_loss=0.1134, over 28581.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3598, pruned_loss=0.1135, over 5635816.02 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5718682.65 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3604, pruned_loss=0.1138, over 5631940.51 frames. ], batch size: 336, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:32:00,896 INFO [optim.py:369] (1/2) 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 18:32:45,800 INFO [train.py:968] (1/2) Epoch 26, batch 45300, giga_loss[loss=0.2509, simple_loss=0.3295, pruned_loss=0.08614, over 28878.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.36, pruned_loss=0.1135, over 5643145.21 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5716889.07 frames. ], giga_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1136, over 5640654.61 frames. ], batch size: 174, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:33:18,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3158, 1.3603, 3.7324, 3.2074], device='cuda:1'), covar=tensor([0.1646, 0.2797, 0.0453, 0.1122], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0675, 0.1008, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 18:33:19,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7675, 1.2348, 4.8952, 3.6023], device='cuda:1'), covar=tensor([0.1651, 0.3107, 0.0400, 0.0957], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0675, 0.1008, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 18:33:32,754 INFO [train.py:968] (1/2) Epoch 26, batch 45350, giga_loss[loss=0.3401, simple_loss=0.3939, pruned_loss=0.1431, over 27580.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.3634, pruned_loss=0.1151, over 5647957.14 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.36, pruned_loss=0.1129, over 5718082.35 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3631, pruned_loss=0.1148, over 5643767.32 frames. ], batch size: 472, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:33:33,345 INFO [optim.py:369] (1/2) 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 18:33:37,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2836, 1.3226, 1.2505, 1.2497], device='cuda:1'), covar=tensor([0.1797, 0.1998, 0.1574, 0.1679], device='cuda:1'), in_proj_covar=tensor([0.2052, 0.2004, 0.1922, 0.2060], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 18:34:22,460 INFO [train.py:968] (1/2) Epoch 26, batch 45400, giga_loss[loss=0.2438, simple_loss=0.3219, pruned_loss=0.08288, over 28934.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3663, pruned_loss=0.1174, over 5627980.08 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3604, pruned_loss=0.1133, over 5708125.52 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3657, pruned_loss=0.1168, over 5631665.83 frames. ], batch size: 213, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:34:38,080 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4685, 1.6162, 1.5381, 1.3993], device='cuda:1'), covar=tensor([0.2004, 0.2297, 0.2367, 0.2262], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0758, 0.0727, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 18:35:08,812 INFO [train.py:968] (1/2) Epoch 26, batch 45450, giga_loss[loss=0.2918, simple_loss=0.3631, pruned_loss=0.1103, over 29031.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3663, pruned_loss=0.1175, over 5626461.09 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3606, pruned_loss=0.1132, over 5712851.27 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3658, pruned_loss=0.1172, over 5623481.66 frames. ], batch size: 128, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:35:09,515 INFO [optim.py:369] (1/2) 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 18:35:56,129 INFO [train.py:968] (1/2) Epoch 26, batch 45500, giga_loss[loss=0.2964, simple_loss=0.3458, pruned_loss=0.1235, over 23432.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3667, pruned_loss=0.1181, over 5624840.08 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3609, pruned_loss=0.1134, over 5705020.31 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.366, pruned_loss=0.1177, over 5629120.69 frames. ], batch size: 705, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:36:19,027 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 26, batch 45550, giga_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1217, over 28856.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3698, pruned_loss=0.1199, over 5642663.76 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3609, pruned_loss=0.1132, over 5708633.68 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3694, pruned_loss=0.12, over 5641616.66 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:36:43,688 INFO [optim.py:369] (1/2) 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 18:37:09,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1184637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:37:28,212 INFO [train.py:968] (1/2) Epoch 26, batch 45600, giga_loss[loss=0.305, simple_loss=0.3754, pruned_loss=0.1173, over 28729.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3706, pruned_loss=0.1202, over 5642530.02 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3605, pruned_loss=0.1129, over 5701277.10 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3708, pruned_loss=0.1206, over 5647251.24 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:37:47,335 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:37:54,265 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-13 18:38:15,962 INFO [train.py:968] (1/2) Epoch 26, batch 45650, giga_loss[loss=0.2629, simple_loss=0.3466, pruned_loss=0.08959, over 28996.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3733, pruned_loss=0.1227, over 5605709.01 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3613, pruned_loss=0.1135, over 5656162.57 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.373, pruned_loss=0.1227, over 5646492.26 frames. ], batch size: 164, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:38:16,526 INFO [optim.py:369] (1/2) 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 18:38:22,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8244, 3.6615, 3.5094, 1.6906], device='cuda:1'), covar=tensor([0.0804, 0.0896, 0.0835, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.1307, 0.1212, 0.1021, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 18:39:06,105 INFO [train.py:968] (1/2) Epoch 26, batch 45700, giga_loss[loss=0.3042, simple_loss=0.3808, pruned_loss=0.1138, over 28741.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3754, pruned_loss=0.1248, over 5585331.19 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.3618, pruned_loss=0.1139, over 5620493.19 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3749, pruned_loss=0.1245, over 5648234.37 frames. ], batch size: 284, lr: 1.21e-03, grad_scale: 4.0
+2023-03-13 18:39:30,052 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1184780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:39:32,224 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1184783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:39:57,181 INFO [train.py:968] (1/2) Epoch 26, batch 45750, giga_loss[loss=0.3286, simple_loss=0.3909, pruned_loss=0.1331, over 28835.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3763, pruned_loss=0.1238, over 5552843.03 frames. ], libri_tot_loss[loss=0.2956, simple_loss=0.3623, pruned_loss=0.1145, over 5570227.20 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3756, pruned_loss=0.1232, over 5648691.83 frames. ], batch size: 112, lr: 1.21e-03, grad_scale: 2.0
+2023-03-13 18:39:59,019 INFO [optim.py:369] (1/2) 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 18:40:03,587 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1184812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:40:48,298 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-13 18:42:03,684 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:42:08,747 INFO [train.py:968] (1/2) Epoch 27, batch 50, giga_loss[loss=0.2962, simple_loss=0.3684, pruned_loss=0.112, over 28969.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.369, pruned_loss=0.1053, over 1266423.53 frames. ], libri_tot_loss[loss=0.2413, simple_loss=0.3242, pruned_loss=0.07926, over 116006.31 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.373, pruned_loss=0.1076, over 1173910.21 frames. ], batch size: 106, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:42:13,758 INFO [optim.py:369] (1/2) 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 18:42:59,631 INFO [train.py:968] (1/2) Epoch 27, batch 100, giga_loss[loss=0.2561, simple_loss=0.3427, pruned_loss=0.08475, over 28909.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.364, pruned_loss=0.1046, over 2240211.41 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3287, pruned_loss=0.08166, over 173474.53 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3663, pruned_loss=0.1061, over 2131274.50 frames. ], batch size: 145, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:43:03,999 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1184959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:43:04,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1184960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:43:43,587 INFO [train.py:968] (1/2) Epoch 27, batch 150, giga_loss[loss=0.1846, simple_loss=0.2707, pruned_loss=0.04926, over 28432.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3471, pruned_loss=0.09644, over 3006950.30 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3347, pruned_loss=0.08725, over 367730.77 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3487, pruned_loss=0.09749, over 2819496.67 frames. ], batch size: 71, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:43:46,431 INFO [optim.py:369] (1/2) 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 18:43:59,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6412, 1.5373, 4.1163, 3.3738], device='cuda:1'), covar=tensor([0.1506, 0.2632, 0.0464, 0.0959], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0671, 0.1001, 0.0969], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 18:44:17,123 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 18:44:22,456 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:44:23,503 INFO [train.py:968] (1/2) Epoch 27, batch 200, giga_loss[loss=0.1969, simple_loss=0.2736, pruned_loss=0.06006, over 28481.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3325, pruned_loss=0.08942, over 3608505.16 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3311, pruned_loss=0.08605, over 477439.99 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3334, pruned_loss=0.0901, over 3413894.40 frames. ], batch size: 78, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:44:36,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6537, 1.8791, 1.5983, 1.5859], device='cuda:1'), covar=tensor([0.2722, 0.2714, 0.2980, 0.2671], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1141, 0.1399, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 18:45:07,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:45:08,251 INFO [train.py:968] (1/2) Epoch 27, batch 250, giga_loss[loss=0.2116, simple_loss=0.2891, pruned_loss=0.06706, over 28968.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3226, pruned_loss=0.08496, over 4059613.09 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3341, pruned_loss=0.08761, over 504148.79 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3225, pruned_loss=0.08509, over 3898717.92 frames. ], batch size: 164, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:45:09,115 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:45:29,627 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 300, giga_loss[loss=0.2175, simple_loss=0.2921, pruned_loss=0.07147, over 28249.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.315, pruned_loss=0.08139, over 4420354.36 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3369, pruned_loss=0.08784, over 680651.42 frames. ], giga_tot_loss[loss=0.2379, simple_loss=0.3136, pruned_loss=0.08114, over 4247709.88 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:46:22,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2239, 0.8008, 0.8886, 1.4161], device='cuda:1'), covar=tensor([0.0780, 0.0423, 0.0395, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 18:46:22,625 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:46:25,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1185199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:46:29,463 INFO [train.py:968] (1/2) Epoch 27, batch 350, giga_loss[loss=0.2116, simple_loss=0.2825, pruned_loss=0.07036, over 28672.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.3091, pruned_loss=0.07871, over 4691462.46 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3396, pruned_loss=0.08971, over 821665.49 frames. ], giga_tot_loss[loss=0.2311, simple_loss=0.3065, pruned_loss=0.07784, over 4529499.84 frames. ], batch size: 92, lr: 1.19e-03, grad_scale: 2.0
+2023-03-13 18:46:32,756 INFO [optim.py:369] (1/2) 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 18:46:49,815 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 400, giga_loss[loss=0.2844, simple_loss=0.3395, pruned_loss=0.1147, over 26681.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3052, pruned_loss=0.0767, over 4918935.32 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3384, pruned_loss=0.08891, over 971290.96 frames. ], giga_tot_loss[loss=0.2268, simple_loss=0.3021, pruned_loss=0.07569, over 4760336.24 frames. ], batch size: 555, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:47:26,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:47:50,697 INFO [train.py:968] (1/2) Epoch 27, batch 450, giga_loss[loss=0.2274, simple_loss=0.2848, pruned_loss=0.08494, over 24003.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3027, pruned_loss=0.07552, over 5089205.93 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3387, pruned_loss=0.08889, over 1090915.78 frames. ], giga_tot_loss[loss=0.224, simple_loss=0.2993, pruned_loss=0.07436, over 4944881.70 frames. ], batch size: 705, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:47:55,278 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:48:19,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4650, 4.2908, 4.0560, 2.1067], device='cuda:1'), covar=tensor([0.0572, 0.0783, 0.0779, 0.1936], device='cuda:1'), in_proj_covar=tensor([0.1295, 0.1199, 0.1010, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 18:48:33,729 INFO [train.py:968] (1/2) Epoch 27, batch 500, giga_loss[loss=0.2296, simple_loss=0.2975, pruned_loss=0.08082, over 27728.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.3003, pruned_loss=0.0745, over 5215674.89 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3382, pruned_loss=0.0884, over 1182387.69 frames. ], giga_tot_loss[loss=0.2219, simple_loss=0.297, pruned_loss=0.0734, over 5091621.98 frames. ], batch size: 472, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:49:13,824 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 27, batch 550, giga_loss[loss=0.1993, simple_loss=0.2598, pruned_loss=0.06942, over 23897.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2977, pruned_loss=0.07333, over 5323483.09 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3382, pruned_loss=0.08807, over 1275231.57 frames. ], giga_tot_loss[loss=0.2194, simple_loss=0.2943, pruned_loss=0.07222, over 5214447.12 frames. ], batch size: 705, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:49:22,187 INFO [optim.py:369] (1/2) 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 18:49:29,568 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:49:31,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1185420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:49:56,401 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1185449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:01,281 INFO [train.py:968] (1/2) Epoch 27, batch 600, giga_loss[loss=0.2051, simple_loss=0.2781, pruned_loss=0.06606, over 28819.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2952, pruned_loss=0.072, over 5410985.61 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3386, pruned_loss=0.08822, over 1343742.70 frames. ], giga_tot_loss[loss=0.2167, simple_loss=0.2918, pruned_loss=0.07085, over 5318110.07 frames. ], batch size: 186, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:50:22,546 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:23,803 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1185479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:24,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1185480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:48,406 INFO [train.py:968] (1/2) Epoch 27, batch 650, giga_loss[loss=0.2275, simple_loss=0.3006, pruned_loss=0.07721, over 28858.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2937, pruned_loss=0.07127, over 5477722.36 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3391, pruned_loss=0.08846, over 1454643.47 frames. ], giga_tot_loss[loss=0.2149, simple_loss=0.2899, pruned_loss=0.06993, over 5395115.92 frames. ], batch size: 186, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:50:52,001 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1185509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:50:58,395 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4734, 3.5490, 1.6727, 1.6706], device='cuda:1'), covar=tensor([0.1018, 0.0336, 0.0914, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0564, 0.0404, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 18:51:20,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3608, 1.1712, 3.9577, 3.3082], device='cuda:1'), covar=tensor([0.1718, 0.3009, 0.0473, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0670, 0.1001, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 18:51:31,291 INFO [train.py:968] (1/2) Epoch 27, batch 700, giga_loss[loss=0.2239, simple_loss=0.2987, pruned_loss=0.07456, over 29113.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2912, pruned_loss=0.07027, over 5522288.47 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3399, pruned_loss=0.08858, over 1498952.84 frames. ], giga_tot_loss[loss=0.2128, simple_loss=0.2875, pruned_loss=0.06903, over 5453860.67 frames. ], batch size: 155, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:51:38,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 18:51:46,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8341, 2.2568, 1.9869, 1.9214], device='cuda:1'), covar=tensor([0.2449, 0.2519, 0.2494, 0.2569], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0760, 0.0727, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 18:52:17,471 INFO [train.py:968] (1/2) Epoch 27, batch 750, giga_loss[loss=0.2077, simple_loss=0.2829, pruned_loss=0.06622, over 28629.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2891, pruned_loss=0.06947, over 5554976.53 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3383, pruned_loss=0.08771, over 1607740.40 frames. ], giga_tot_loss[loss=0.2109, simple_loss=0.2854, pruned_loss=0.06823, over 5492167.44 frames. ], batch size: 262, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:52:21,143 INFO [optim.py:369] (1/2) 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 18:53:00,261 INFO [train.py:968] (1/2) Epoch 27, batch 800, giga_loss[loss=0.1962, simple_loss=0.2754, pruned_loss=0.05853, over 29058.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2902, pruned_loss=0.07036, over 5586267.72 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3371, pruned_loss=0.08717, over 1711018.39 frames. ], giga_tot_loss[loss=0.2123, simple_loss=0.2864, pruned_loss=0.06911, over 5531462.91 frames. ], batch size: 128, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 18:53:47,920 INFO [train.py:968] (1/2) Epoch 27, batch 850, giga_loss[loss=0.3113, simple_loss=0.3824, pruned_loss=0.1201, over 28316.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.3002, pruned_loss=0.07557, over 5607332.98 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3354, pruned_loss=0.08623, over 1833265.94 frames. ], giga_tot_loss[loss=0.2228, simple_loss=0.2966, pruned_loss=0.07449, over 5556739.30 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:53:52,191 INFO [optim.py:369] (1/2) 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 18:54:27,787 INFO [train.py:968] (1/2) Epoch 27, batch 900, giga_loss[loss=0.2871, simple_loss=0.3641, pruned_loss=0.105, over 28973.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3114, pruned_loss=0.08061, over 5633005.88 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.335, pruned_loss=0.08615, over 2024687.51 frames. ], giga_tot_loss[loss=0.2331, simple_loss=0.3073, pruned_loss=0.0794, over 5585739.43 frames. ], batch size: 136, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:54:32,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5765, 1.8216, 1.5231, 1.6091], device='cuda:1'), covar=tensor([0.2788, 0.2841, 0.3201, 0.2436], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1141, 0.1402, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 18:54:38,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3583, 1.6543, 1.5992, 1.1829], device='cuda:1'), covar=tensor([0.1854, 0.2880, 0.1622, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0721, 0.0985, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 18:55:10,279 INFO [train.py:968] (1/2) Epoch 27, batch 950, giga_loss[loss=0.2587, simple_loss=0.327, pruned_loss=0.09519, over 23535.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3222, pruned_loss=0.08606, over 5631793.50 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3353, pruned_loss=0.08629, over 2167299.23 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3183, pruned_loss=0.08501, over 5593040.93 frames. ], batch size: 705, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:55:14,472 INFO [optim.py:369] (1/2) 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 18:55:51,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1185854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:55:51,456 INFO [train.py:968] (1/2) Epoch 27, batch 1000, giga_loss[loss=0.2831, simple_loss=0.3625, pruned_loss=0.1019, over 28286.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3294, pruned_loss=0.08855, over 5650074.25 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3353, pruned_loss=0.0861, over 2259719.81 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3261, pruned_loss=0.08784, over 5613215.65 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:56:29,715 INFO [train.py:968] (1/2) Epoch 27, batch 1050, giga_loss[loss=0.2524, simple_loss=0.3316, pruned_loss=0.08659, over 28673.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3332, pruned_loss=0.08903, over 5661423.37 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3357, pruned_loss=0.08652, over 2373689.49 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3303, pruned_loss=0.08838, over 5634425.85 frames. ], batch size: 92, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:56:34,360 INFO [optim.py:369] (1/2) 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 18:57:14,024 INFO [train.py:968] (1/2) Epoch 27, batch 1100, giga_loss[loss=0.2458, simple_loss=0.3414, pruned_loss=0.07505, over 28971.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3353, pruned_loss=0.08954, over 5663017.67 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3363, pruned_loss=0.08698, over 2478647.31 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3327, pruned_loss=0.08891, over 5634414.71 frames. ], batch size: 174, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:57:27,825 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1185971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:57:51,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1185997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:57:54,677 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:57:57,661 INFO [train.py:968] (1/2) Epoch 27, batch 1150, giga_loss[loss=0.2554, simple_loss=0.3347, pruned_loss=0.08808, over 28886.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3388, pruned_loss=0.09184, over 5657433.50 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3377, pruned_loss=0.08771, over 2568283.28 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3362, pruned_loss=0.09115, over 5637831.23 frames. ], batch size: 145, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 18:58:00,819 INFO [optim.py:369] (1/2) 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 18:58:06,402 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5871, 1.8195, 1.5476, 1.4502], device='cuda:1'), covar=tensor([0.2299, 0.2269, 0.2423, 0.2288], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1142, 0.1403, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 18:58:20,635 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 1200, giga_loss[loss=0.3825, simple_loss=0.4142, pruned_loss=0.1754, over 26564.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.342, pruned_loss=0.09437, over 5668376.15 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3372, pruned_loss=0.0874, over 2617730.70 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3402, pruned_loss=0.09404, over 5649873.89 frames. ], batch size: 555, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 18:58:42,128 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 18:59:24,473 INFO [train.py:968] (1/2) Epoch 27, batch 1250, giga_loss[loss=0.2929, simple_loss=0.3593, pruned_loss=0.1133, over 28711.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3459, pruned_loss=0.09707, over 5675846.65 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3377, pruned_loss=0.08753, over 2695459.48 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3444, pruned_loss=0.09697, over 5659059.49 frames. ], batch size: 92, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 18:59:28,413 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 1300, giga_loss[loss=0.3226, simple_loss=0.3994, pruned_loss=0.1229, over 28616.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3486, pruned_loss=0.09769, over 5688086.48 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3385, pruned_loss=0.08816, over 2806192.61 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09763, over 5667535.27 frames. ], batch size: 307, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:00:27,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-13 19:00:44,004 INFO [train.py:968] (1/2) Epoch 27, batch 1350, giga_loss[loss=0.2762, simple_loss=0.3589, pruned_loss=0.09677, over 28246.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3494, pruned_loss=0.09748, over 5690864.31 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3382, pruned_loss=0.08796, over 2882803.64 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3489, pruned_loss=0.09776, over 5669621.45 frames. ], batch size: 368, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:00:47,692 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 1400, giga_loss[loss=0.2767, simple_loss=0.3584, pruned_loss=0.09746, over 28675.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3506, pruned_loss=0.09717, over 5693145.52 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3388, pruned_loss=0.08815, over 2954080.49 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3502, pruned_loss=0.09751, over 5674904.32 frames. ], batch size: 262, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:02:06,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:02:07,871 INFO [train.py:968] (1/2) Epoch 27, batch 1450, giga_loss[loss=0.2403, simple_loss=0.3297, pruned_loss=0.07548, over 28372.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3502, pruned_loss=0.09551, over 5694146.20 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3391, pruned_loss=0.08818, over 2992547.37 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3498, pruned_loss=0.09588, over 5682082.75 frames. ], batch size: 77, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:02:11,735 INFO [optim.py:369] (1/2) 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 19:02:38,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:02:45,292 INFO [train.py:968] (1/2) Epoch 27, batch 1500, giga_loss[loss=0.2504, simple_loss=0.3371, pruned_loss=0.08189, over 28763.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3482, pruned_loss=0.09372, over 5694787.99 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3388, pruned_loss=0.08796, over 3097425.95 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3484, pruned_loss=0.09434, over 5686604.54 frames. ], batch size: 92, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:02:46,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3671, 1.6930, 1.5577, 1.5054], device='cuda:1'), covar=tensor([0.0784, 0.0379, 0.0331, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 19:02:51,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9550, 1.3028, 1.1211, 0.2102], device='cuda:1'), covar=tensor([0.4718, 0.3611, 0.4789, 0.7240], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1701, 0.1643, 0.1477], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 19:02:58,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-13 19:03:24,338 INFO [train.py:968] (1/2) Epoch 27, batch 1550, giga_loss[loss=0.2432, simple_loss=0.328, pruned_loss=0.07921, over 28819.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3461, pruned_loss=0.0913, over 5707898.61 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3385, pruned_loss=0.08756, over 3152327.65 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3465, pruned_loss=0.09206, over 5698643.56 frames. ], batch size: 66, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:03:29,269 INFO [optim.py:369] (1/2) 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 19:03:41,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186425.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:03:45,244 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:04:07,740 INFO [train.py:968] (1/2) Epoch 27, batch 1600, giga_loss[loss=0.2295, simple_loss=0.3191, pruned_loss=0.06999, over 28472.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.348, pruned_loss=0.0941, over 5691954.48 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3381, pruned_loss=0.08736, over 3216491.73 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3488, pruned_loss=0.09493, over 5683438.16 frames. ], batch size: 71, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:04:34,674 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:04:38,368 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 1650, libri_loss[loss=0.2682, simple_loss=0.3519, pruned_loss=0.09226, over 29519.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3504, pruned_loss=0.09779, over 5701039.56 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.338, pruned_loss=0.08733, over 3294318.18 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3513, pruned_loss=0.09871, over 5690256.03 frames. ], batch size: 84, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:04:53,207 INFO [optim.py:369] (1/2) 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 19:05:04,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:05:06,653 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1186524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:05:18,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8897, 4.8335, 2.2228, 1.9037], device='cuda:1'), covar=tensor([0.0987, 0.0222, 0.0813, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0563, 0.0404, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 19:05:30,699 INFO [train.py:968] (1/2) Epoch 27, batch 1700, giga_loss[loss=0.2986, simple_loss=0.3613, pruned_loss=0.1179, over 28914.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3514, pruned_loss=0.09979, over 5711841.03 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3383, pruned_loss=0.08742, over 3371421.56 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3524, pruned_loss=0.1008, over 5698190.81 frames. ], batch size: 213, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:05:45,105 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:05:47,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:06:12,099 INFO [train.py:968] (1/2) Epoch 27, batch 1750, giga_loss[loss=0.3457, simple_loss=0.3976, pruned_loss=0.1469, over 28566.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3499, pruned_loss=0.1, over 5704825.16 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3375, pruned_loss=0.0869, over 3421524.83 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3513, pruned_loss=0.1013, over 5690648.08 frames. ], batch size: 71, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:06:12,289 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:06:16,066 INFO [optim.py:369] (1/2) 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 19:06:48,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 19:06:51,608 INFO [train.py:968] (1/2) Epoch 27, batch 1800, giga_loss[loss=0.2462, simple_loss=0.3296, pruned_loss=0.0814, over 28840.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1, over 5697065.57 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3375, pruned_loss=0.08681, over 3470110.24 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3497, pruned_loss=0.1014, over 5682739.74 frames. ], batch size: 174, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:06:55,049 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3279, 1.1217, 3.7603, 3.2320], device='cuda:1'), covar=tensor([0.1613, 0.2935, 0.0455, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0666, 0.0991, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 19:07:10,606 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:07:32,917 INFO [train.py:968] (1/2) Epoch 27, batch 1850, giga_loss[loss=0.2871, simple_loss=0.3519, pruned_loss=0.1111, over 28665.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.348, pruned_loss=0.09947, over 5694250.73 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3377, pruned_loss=0.08678, over 3506584.67 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3491, pruned_loss=0.1007, over 5679964.87 frames. ], batch size: 92, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:07:36,779 INFO [optim.py:369] (1/2) 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 19:08:10,398 INFO [train.py:968] (1/2) Epoch 27, batch 1900, giga_loss[loss=0.2604, simple_loss=0.3401, pruned_loss=0.09038, over 28677.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3461, pruned_loss=0.09732, over 5701444.50 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3385, pruned_loss=0.08699, over 3621673.79 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.347, pruned_loss=0.09875, over 5682508.19 frames. ], batch size: 242, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:08:54,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1186800.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:08:58,095 INFO [train.py:968] (1/2) Epoch 27, batch 1950, giga_loss[loss=0.2289, simple_loss=0.3107, pruned_loss=0.07349, over 28897.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.342, pruned_loss=0.09462, over 5694189.41 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3389, pruned_loss=0.08706, over 3683551.80 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3426, pruned_loss=0.09595, over 5678875.95 frames. ], batch size: 145, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:09:01,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6141, 2.0718, 2.0356, 1.8017], device='cuda:1'), covar=tensor([0.2134, 0.1878, 0.1997, 0.2006], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0760, 0.0729, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 19:09:02,126 INFO [optim.py:369] (1/2) 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 19:09:10,472 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:09:13,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:09:20,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-13 19:09:42,616 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:09:44,236 INFO [train.py:968] (1/2) Epoch 27, batch 2000, libri_loss[loss=0.2807, simple_loss=0.3672, pruned_loss=0.09708, over 29259.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3368, pruned_loss=0.09199, over 5686899.66 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3391, pruned_loss=0.08711, over 3738595.29 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3371, pruned_loss=0.09316, over 5670098.19 frames. ], batch size: 94, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:10:23,871 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 2050, giga_loss[loss=0.258, simple_loss=0.3294, pruned_loss=0.09332, over 29046.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3313, pruned_loss=0.0893, over 5685004.80 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3386, pruned_loss=0.08679, over 3771071.49 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3317, pruned_loss=0.09045, over 5668895.49 frames. ], batch size: 113, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:10:34,437 INFO [optim.py:369] (1/2) 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 19:11:07,885 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1186943.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:11:09,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1186946.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:11:16,393 INFO [train.py:968] (1/2) Epoch 27, batch 2100, libri_loss[loss=0.2733, simple_loss=0.3467, pruned_loss=0.09996, over 29555.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3283, pruned_loss=0.08826, over 5666879.83 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3383, pruned_loss=0.08659, over 3832749.15 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3286, pruned_loss=0.08935, over 5649038.89 frames. ], batch size: 78, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:11:31,568 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1186975.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:11:57,007 INFO [train.py:968] (1/2) Epoch 27, batch 2150, giga_loss[loss=0.2536, simple_loss=0.3339, pruned_loss=0.08662, over 27659.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3305, pruned_loss=0.08882, over 5675205.58 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3388, pruned_loss=0.08691, over 3841298.40 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3304, pruned_loss=0.08949, over 5661812.99 frames. ], batch size: 472, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:12:02,074 INFO [optim.py:369] (1/2) 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 19:12:24,979 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1187042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:12:27,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-13 19:12:28,430 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 2200, libri_loss[loss=0.2832, simple_loss=0.3763, pruned_loss=0.09508, over 29504.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.33, pruned_loss=0.08812, over 5688865.29 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3395, pruned_loss=0.08698, over 3888643.03 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3292, pruned_loss=0.08865, over 5676491.42 frames. ], batch size: 85, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:12:51,508 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1187074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:13:05,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9702, 1.1379, 1.0729, 0.9266], device='cuda:1'), covar=tensor([0.2638, 0.2944, 0.1714, 0.2544], device='cuda:1'), in_proj_covar=tensor([0.2018, 0.1962, 0.1885, 0.2025], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 19:13:17,971 INFO [train.py:968] (1/2) Epoch 27, batch 2250, giga_loss[loss=0.2187, simple_loss=0.3013, pruned_loss=0.068, over 28740.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3278, pruned_loss=0.08705, over 5693450.81 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3396, pruned_loss=0.08678, over 3908717.20 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3271, pruned_loss=0.08758, over 5681993.21 frames. ], batch size: 284, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:13:24,313 INFO [optim.py:369] (1/2) 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 19:13:25,200 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1751, 4.0246, 3.8111, 1.7870], device='cuda:1'), covar=tensor([0.0608, 0.0730, 0.0669, 0.2407], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1176, 0.0992, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 19:14:00,902 INFO [train.py:968] (1/2) Epoch 27, batch 2300, giga_loss[loss=0.2352, simple_loss=0.3181, pruned_loss=0.07612, over 29095.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3267, pruned_loss=0.0868, over 5706177.26 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3395, pruned_loss=0.08676, over 3958130.01 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.326, pruned_loss=0.08726, over 5692298.91 frames. ], batch size: 155, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:14:42,149 INFO [train.py:968] (1/2) Epoch 27, batch 2350, libri_loss[loss=0.2761, simple_loss=0.3695, pruned_loss=0.09138, over 29221.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3243, pruned_loss=0.08555, over 5689540.42 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3401, pruned_loss=0.08696, over 3976586.65 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3231, pruned_loss=0.08577, over 5692689.89 frames. ], batch size: 97, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:14:47,589 INFO [optim.py:369] (1/2) 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 19:15:23,783 INFO [train.py:968] (1/2) Epoch 27, batch 2400, giga_loss[loss=0.2234, simple_loss=0.2969, pruned_loss=0.0749, over 28970.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3222, pruned_loss=0.08514, over 5691252.93 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3402, pruned_loss=0.08693, over 3986455.71 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3211, pruned_loss=0.08532, over 5692543.00 frames. ], batch size: 106, lr: 1.19e-03, grad_scale: 8.0
+2023-03-13 19:15:25,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8132, 1.7991, 2.0418, 1.6017], device='cuda:1'), covar=tensor([0.2098, 0.2569, 0.1577, 0.1806], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0716, 0.0979, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 19:15:55,080 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-13 19:15:59,300 INFO [train.py:968] (1/2) Epoch 27, batch 2450, giga_loss[loss=0.2351, simple_loss=0.3167, pruned_loss=0.07675, over 28656.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3205, pruned_loss=0.08387, over 5701449.10 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3411, pruned_loss=0.08701, over 4047936.86 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3186, pruned_loss=0.08391, over 5700229.85 frames. ], batch size: 307, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:16:04,806 INFO [optim.py:369] (1/2) 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 19:16:09,266 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 2500, giga_loss[loss=0.2311, simple_loss=0.3068, pruned_loss=0.07771, over 29048.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.319, pruned_loss=0.08332, over 5711528.16 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3412, pruned_loss=0.08687, over 4102565.84 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3168, pruned_loss=0.08335, over 5705734.40 frames. ], batch size: 136, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:17:01,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3154, 1.6646, 1.3951, 1.5062], device='cuda:1'), covar=tensor([0.0751, 0.0433, 0.0361, 0.0783], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 19:17:01,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4994, 2.0533, 1.3417, 0.9077], device='cuda:1'), covar=tensor([0.8206, 0.3771, 0.3746, 0.7522], device='cuda:1'), in_proj_covar=tensor([0.1810, 0.1697, 0.1638, 0.1477], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 19:17:05,960 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 2550, giga_loss[loss=0.2213, simple_loss=0.2963, pruned_loss=0.07311, over 28809.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3162, pruned_loss=0.08148, over 5720067.39 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3415, pruned_loss=0.08676, over 4128919.09 frames. ], giga_tot_loss[loss=0.2384, simple_loss=0.3138, pruned_loss=0.08151, over 5713338.24 frames. ], batch size: 199, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:17:21,524 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 2600, giga_loss[loss=0.2347, simple_loss=0.3053, pruned_loss=0.08205, over 28812.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.3158, pruned_loss=0.08117, over 5721141.19 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.342, pruned_loss=0.08676, over 4195476.62 frames. ], giga_tot_loss[loss=0.2373, simple_loss=0.3126, pruned_loss=0.08098, over 5712130.02 frames. ], batch size: 99, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:18:28,541 INFO [train.py:968] (1/2) Epoch 27, batch 2650, giga_loss[loss=0.2182, simple_loss=0.2912, pruned_loss=0.07267, over 28502.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3158, pruned_loss=0.08131, over 5718725.94 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3425, pruned_loss=0.08697, over 4236262.79 frames. ], giga_tot_loss[loss=0.237, simple_loss=0.3122, pruned_loss=0.08087, over 5715463.69 frames. ], batch size: 85, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:18:35,943 INFO [optim.py:369] (1/2) 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 19:19:06,798 INFO [train.py:968] (1/2) Epoch 27, batch 2700, giga_loss[loss=0.2385, simple_loss=0.3073, pruned_loss=0.08482, over 28539.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.3192, pruned_loss=0.08334, over 5724960.57 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3429, pruned_loss=0.08686, over 4285828.95 frames. ], giga_tot_loss[loss=0.2406, simple_loss=0.3154, pruned_loss=0.08293, over 5717248.77 frames. ], batch size: 85, lr: 1.19e-03, grad_scale: 4.0
+2023-03-13 19:19:18,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 19:19:49,212 INFO [train.py:968] (1/2) Epoch 27, batch 2750, giga_loss[loss=0.2696, simple_loss=0.3407, pruned_loss=0.09922, over 28173.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3229, pruned_loss=0.08549, over 5720696.38 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3423, pruned_loss=0.08643, over 4317901.60 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3199, pruned_loss=0.0854, over 5711323.15 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:19:55,799 INFO [optim.py:369] (1/2) 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 19:20:31,682 INFO [train.py:968] (1/2) Epoch 27, batch 2800, giga_loss[loss=0.3477, simple_loss=0.389, pruned_loss=0.1532, over 23711.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3302, pruned_loss=0.09009, over 5716969.54 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.342, pruned_loss=0.08646, over 4377301.70 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3274, pruned_loss=0.09003, over 5705794.84 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:20:46,221 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-13 19:21:07,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1187693.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:21:16,632 INFO [train.py:968] (1/2) Epoch 27, batch 2850, giga_loss[loss=0.2757, simple_loss=0.3596, pruned_loss=0.09589, over 28737.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3372, pruned_loss=0.09471, over 5706123.07 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3421, pruned_loss=0.08646, over 4384952.19 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3348, pruned_loss=0.0947, over 5696425.11 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:21:18,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6171, 1.7867, 1.8224, 1.5186], device='cuda:1'), covar=tensor([0.2168, 0.2497, 0.2480, 0.2600], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0758, 0.0729, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 19:21:24,578 INFO [optim.py:369] (1/2) 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 19:21:47,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4126, 2.1051, 1.5372, 0.6678], device='cuda:1'), covar=tensor([0.7207, 0.3412, 0.4356, 0.7363], device='cuda:1'), in_proj_covar=tensor([0.1817, 0.1705, 0.1644, 0.1483], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 19:22:04,654 INFO [train.py:968] (1/2) Epoch 27, batch 2900, giga_loss[loss=0.2936, simple_loss=0.3726, pruned_loss=0.1073, over 28945.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3413, pruned_loss=0.0956, over 5713363.74 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3423, pruned_loss=0.08649, over 4399943.14 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3393, pruned_loss=0.09566, over 5704217.71 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:22:08,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5413, 1.7674, 1.7365, 1.4649], device='cuda:1'), covar=tensor([0.2625, 0.2331, 0.2726, 0.2643], device='cuda:1'), in_proj_covar=tensor([0.2021, 0.1965, 0.1896, 0.2030], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 19:22:13,868 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1187767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:22:34,551 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-13 19:22:43,747 INFO [train.py:968] (1/2) Epoch 27, batch 2950, giga_loss[loss=0.278, simple_loss=0.3613, pruned_loss=0.09733, over 28283.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3474, pruned_loss=0.09898, over 5709994.89 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3422, pruned_loss=0.08634, over 4422009.31 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.346, pruned_loss=0.0993, over 5700494.10 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:22:52,604 INFO [optim.py:369] (1/2) 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 19:22:58,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1187821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:23:11,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1187836.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:23:14,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1187839.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:23:27,252 INFO [train.py:968] (1/2) Epoch 27, batch 3000, giga_loss[loss=0.3147, simple_loss=0.3635, pruned_loss=0.133, over 23311.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3526, pruned_loss=0.1024, over 5684456.33 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3426, pruned_loss=0.08664, over 4488756.33 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3514, pruned_loss=0.1031, over 5672668.46 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:23:27,252 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 19:23:35,708 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5582, 1.7829, 1.3670, 1.3959], device='cuda:1'), covar=tensor([0.0852, 0.0372, 0.0891, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0449, 0.0524, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 19:23:36,364 INFO [train.py:1012] (1/2) Epoch 27, validation: loss=0.2096, simple_loss=0.3155, pruned_loss=0.05187, over 944034.00 frames. 
+2023-03-13 19:23:36,365 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 19:23:37,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6248, 1.9346, 1.3625, 1.5017], device='cuda:1'), covar=tensor([0.1034, 0.0498, 0.1004, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0449, 0.0524, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 19:23:45,087 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1187868.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:24:03,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0842, 3.2031, 1.2661, 1.4037], device='cuda:1'), covar=tensor([0.1357, 0.0332, 0.1119, 0.1757], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0562, 0.0404, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 19:24:15,180 INFO [train.py:968] (1/2) Epoch 27, batch 3050, giga_loss[loss=0.2423, simple_loss=0.3204, pruned_loss=0.08207, over 28910.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3497, pruned_loss=0.09948, over 5697086.06 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3421, pruned_loss=0.08636, over 4527005.38 frames. ], giga_tot_loss[loss=0.2753, simple_loss=0.3494, pruned_loss=0.1006, over 5685045.82 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:24:21,052 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1187910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:24:22,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1187913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:24:23,359 INFO [optim.py:369] (1/2) 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 19:24:47,182 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1187942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:24:49,443 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.99 vs. limit=5.0
+2023-03-13 19:24:57,006 INFO [train.py:968] (1/2) Epoch 27, batch 3100, libri_loss[loss=0.2782, simple_loss=0.3609, pruned_loss=0.09778, over 28662.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3468, pruned_loss=0.09693, over 5699611.71 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3427, pruned_loss=0.08675, over 4558140.04 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3462, pruned_loss=0.09782, over 5688101.12 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:25:37,084 INFO [train.py:968] (1/2) Epoch 27, batch 3150, giga_loss[loss=0.2363, simple_loss=0.3236, pruned_loss=0.07448, over 28608.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3451, pruned_loss=0.09528, over 5710582.91 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3427, pruned_loss=0.08692, over 4590018.72 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3448, pruned_loss=0.09608, over 5698362.15 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:25:43,000 INFO [optim.py:369] (1/2) 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 19:26:00,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3391, 1.3215, 3.7198, 3.3064], device='cuda:1'), covar=tensor([0.1613, 0.2785, 0.0455, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0668, 0.0994, 0.0964], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 19:26:15,180 INFO [train.py:968] (1/2) Epoch 27, batch 3200, giga_loss[loss=0.258, simple_loss=0.337, pruned_loss=0.08944, over 28678.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3459, pruned_loss=0.0951, over 5716569.26 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3438, pruned_loss=0.08768, over 4642426.68 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3449, pruned_loss=0.09549, over 5699207.23 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:26:42,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-13 19:26:53,440 INFO [train.py:968] (1/2) Epoch 27, batch 3250, giga_loss[loss=0.2864, simple_loss=0.3588, pruned_loss=0.107, over 28435.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3475, pruned_loss=0.09585, over 5708554.75 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3437, pruned_loss=0.08782, over 4655551.70 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3468, pruned_loss=0.09618, over 5702526.55 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:26:55,725 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.2709, 1.4738, 1.3006, 1.4949], device='cuda:1'), covar=tensor([0.0819, 0.0355, 0.0348, 0.0898], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 19:27:27,701 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1188148.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:27:31,987 INFO [train.py:968] (1/2) Epoch 27, batch 3300, giga_loss[loss=0.2918, simple_loss=0.3578, pruned_loss=0.1129, over 28546.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3498, pruned_loss=0.09754, over 5703080.94 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3441, pruned_loss=0.08788, over 4679914.08 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3491, pruned_loss=0.09794, over 5699983.51 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:28:04,827 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 3350, libri_loss[loss=0.2997, simple_loss=0.3778, pruned_loss=0.1108, over 29373.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3509, pruned_loss=0.09906, over 5706590.97 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3436, pruned_loss=0.08768, over 4715526.54 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3508, pruned_loss=0.09978, over 5699524.13 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:28:18,852 INFO [optim.py:369] (1/2) 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 19:28:51,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7527, 3.5885, 3.3685, 1.6121], device='cuda:1'), covar=tensor([0.0783, 0.0850, 0.0740, 0.2345], device='cuda:1'), in_proj_covar=tensor([0.1278, 0.1184, 0.0997, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 19:28:53,562 INFO [train.py:968] (1/2) Epoch 27, batch 3400, giga_loss[loss=0.2615, simple_loss=0.3498, pruned_loss=0.08659, over 28798.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3514, pruned_loss=0.09971, over 5713313.39 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3434, pruned_loss=0.08746, over 4727435.40 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3516, pruned_loss=0.1005, over 5705941.54 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:29:32,767 INFO [train.py:968] (1/2) Epoch 27, batch 3450, giga_loss[loss=0.2606, simple_loss=0.3578, pruned_loss=0.08169, over 28929.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3518, pruned_loss=0.09975, over 5719911.31 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3438, pruned_loss=0.08752, over 4753902.16 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3518, pruned_loss=0.1007, over 5712435.00 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:29:39,126 INFO [optim.py:369] (1/2) 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 19:29:58,583 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1188339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:30:00,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1188342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:30:11,660 INFO [train.py:968] (1/2) Epoch 27, batch 3500, giga_loss[loss=0.3188, simple_loss=0.3875, pruned_loss=0.1251, over 28929.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3525, pruned_loss=0.1001, over 5711675.19 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3439, pruned_loss=0.08776, over 4772736.49 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3526, pruned_loss=0.1009, over 5709837.82 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:30:24,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1188371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:30:51,710 INFO [train.py:968] (1/2) Epoch 27, batch 3550, giga_loss[loss=0.2394, simple_loss=0.3228, pruned_loss=0.07795, over 28628.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.352, pruned_loss=0.09887, over 5714388.37 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.344, pruned_loss=0.08779, over 4784258.66 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3521, pruned_loss=0.09959, over 5711094.44 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:31:00,311 INFO [optim.py:369] (1/2) 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 19:31:32,913 INFO [train.py:968] (1/2) Epoch 27, batch 3600, giga_loss[loss=0.3162, simple_loss=0.3747, pruned_loss=0.1289, over 28993.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3524, pruned_loss=0.09857, over 5718831.43 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.344, pruned_loss=0.08786, over 4812764.84 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3527, pruned_loss=0.09939, over 5715145.91 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:31:54,597 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 3650, giga_loss[loss=0.2835, simple_loss=0.3564, pruned_loss=0.1053, over 28940.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3501, pruned_loss=0.09714, over 5722815.43 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3441, pruned_loss=0.08801, over 4852993.56 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3505, pruned_loss=0.09807, over 5715494.22 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:32:10,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.58 vs. limit=5.0
+2023-03-13 19:32:17,271 INFO [optim.py:369] (1/2) 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 19:32:24,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1188523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:32:36,237 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1188537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:32:48,088 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3898, 1.4400, 1.2055, 1.5524], device='cuda:1'), covar=tensor([0.0824, 0.0357, 0.0363, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 19:32:50,698 INFO [train.py:968] (1/2) Epoch 27, batch 3700, giga_loss[loss=0.2507, simple_loss=0.3342, pruned_loss=0.08359, over 28906.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3482, pruned_loss=0.09653, over 5719186.33 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3444, pruned_loss=0.08815, over 4864888.06 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3484, pruned_loss=0.09727, over 5714187.52 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:32:56,274 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9853, 1.3893, 1.0742, 0.4352], device='cuda:1'), covar=tensor([0.3941, 0.3023, 0.3684, 0.5325], device='cuda:1'), in_proj_covar=tensor([0.1808, 0.1692, 0.1633, 0.1471], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 19:33:27,245 INFO [train.py:968] (1/2) Epoch 27, batch 3750, libri_loss[loss=0.2751, simple_loss=0.3527, pruned_loss=0.09876, over 29553.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3462, pruned_loss=0.09562, over 5716116.79 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3442, pruned_loss=0.08827, over 4889239.09 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3467, pruned_loss=0.09638, over 5715920.56 frames. ], batch size: 79, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:33:34,611 INFO [optim.py:369] (1/2) 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 19:33:42,428 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1188626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:33:44,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1188629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:34:07,247 INFO [train.py:968] (1/2) Epoch 27, batch 3800, giga_loss[loss=0.2607, simple_loss=0.3352, pruned_loss=0.09312, over 28907.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3464, pruned_loss=0.09575, over 5728542.00 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3446, pruned_loss=0.08849, over 4923632.14 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3465, pruned_loss=0.09642, over 5723443.44 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:34:09,334 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1188658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:34:14,728 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1188666.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:34:16,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1188669.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 19:34:39,333 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1188698.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 19:34:43,753 INFO [train.py:968] (1/2) Epoch 27, batch 3850, giga_loss[loss=0.2621, simple_loss=0.3415, pruned_loss=0.0914, over 28966.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3471, pruned_loss=0.09648, over 5723213.88 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3445, pruned_loss=0.08843, over 4947180.86 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3473, pruned_loss=0.09729, over 5722498.75 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:34:53,505 INFO [optim.py:369] (1/2) 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 19:35:25,154 INFO [train.py:968] (1/2) Epoch 27, batch 3900, giga_loss[loss=0.2635, simple_loss=0.3433, pruned_loss=0.09188, over 29057.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3484, pruned_loss=0.0972, over 5720391.83 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3448, pruned_loss=0.08864, over 4959918.18 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3484, pruned_loss=0.09779, over 5718773.54 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:35:35,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2369, 1.6381, 1.2665, 0.7692], device='cuda:1'), covar=tensor([0.5065, 0.2534, 0.3085, 0.6199], device='cuda:1'), in_proj_covar=tensor([0.1807, 0.1689, 0.1630, 0.1470], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 19:36:08,462 INFO [train.py:968] (1/2) Epoch 27, batch 3950, giga_loss[loss=0.245, simple_loss=0.3304, pruned_loss=0.07974, over 28889.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3467, pruned_loss=0.09557, over 5718581.12 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3449, pruned_loss=0.08877, over 4964670.59 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3466, pruned_loss=0.09596, over 5716371.22 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:36:16,985 INFO [optim.py:369] (1/2) 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 19:36:36,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2339, 1.5230, 1.5224, 1.1036], device='cuda:1'), covar=tensor([0.1828, 0.2795, 0.1556, 0.1766], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0718, 0.0980, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 19:36:47,727 INFO [train.py:968] (1/2) Epoch 27, batch 4000, giga_loss[loss=0.263, simple_loss=0.334, pruned_loss=0.096, over 28243.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3461, pruned_loss=0.09559, over 5724183.60 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3445, pruned_loss=0.08844, over 4987890.18 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3464, pruned_loss=0.09631, over 5718461.70 frames. ], batch size: 65, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:37:12,506 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5165, 1.6628, 1.7640, 1.3199], device='cuda:1'), covar=tensor([0.1966, 0.2733, 0.1606, 0.1855], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0718, 0.0981, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 19:37:28,132 INFO [train.py:968] (1/2) Epoch 27, batch 4050, giga_loss[loss=0.2459, simple_loss=0.3205, pruned_loss=0.08566, over 29086.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3438, pruned_loss=0.09478, over 5716825.92 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3445, pruned_loss=0.08854, over 5008585.34 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3441, pruned_loss=0.09544, over 5709817.39 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:37:33,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1188912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:37:35,918 INFO [optim.py:369] (1/2) 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 19:37:58,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.64 vs. limit=5.0
+2023-03-13 19:38:05,566 INFO [train.py:968] (1/2) Epoch 27, batch 4100, giga_loss[loss=0.2348, simple_loss=0.3161, pruned_loss=0.07678, over 28661.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3407, pruned_loss=0.09327, over 5715152.99 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3444, pruned_loss=0.08843, over 5030162.20 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3409, pruned_loss=0.09401, over 5705801.05 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:38:18,631 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1188973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:38:43,662 INFO [train.py:968] (1/2) Epoch 27, batch 4150, libri_loss[loss=0.3171, simple_loss=0.3968, pruned_loss=0.1187, over 27667.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3396, pruned_loss=0.09294, over 5709561.87 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3448, pruned_loss=0.08861, over 5048157.92 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3393, pruned_loss=0.09347, over 5701072.97 frames. ], batch size: 116, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:38:44,656 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2012, 1.2964, 1.1655, 0.8953], device='cuda:1'), covar=tensor([0.1006, 0.0516, 0.1066, 0.1061], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0446, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 19:38:53,816 INFO [scaling.py:679] (1/2) 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] (1/2) 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 19:39:23,395 INFO [train.py:968] (1/2) Epoch 27, batch 4200, giga_loss[loss=0.2472, simple_loss=0.3152, pruned_loss=0.08964, over 28506.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3384, pruned_loss=0.09314, over 5703389.30 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3446, pruned_loss=0.08857, over 5055788.99 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3382, pruned_loss=0.09369, over 5701158.09 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:39:23,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:39:25,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:39:49,836 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:39:51,350 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5949, 1.7924, 1.3187, 1.3062], device='cuda:1'), covar=tensor([0.1033, 0.0630, 0.1099, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0447, 0.0526, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 19:40:03,044 INFO [train.py:968] (1/2) Epoch 27, batch 4250, giga_loss[loss=0.2849, simple_loss=0.3565, pruned_loss=0.1067, over 28659.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3362, pruned_loss=0.09233, over 5698985.21 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3446, pruned_loss=0.08853, over 5059495.70 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.336, pruned_loss=0.09283, over 5699859.55 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:40:13,884 INFO [optim.py:369] (1/2) 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 19:40:42,298 INFO [train.py:968] (1/2) Epoch 27, batch 4300, giga_loss[loss=0.2654, simple_loss=0.3285, pruned_loss=0.1011, over 28724.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3337, pruned_loss=0.09114, over 5707994.76 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.345, pruned_loss=0.08887, over 5080861.83 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3329, pruned_loss=0.09133, over 5706672.07 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:41:20,049 INFO [train.py:968] (1/2) Epoch 27, batch 4350, giga_loss[loss=0.2331, simple_loss=0.3119, pruned_loss=0.07722, over 28908.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3322, pruned_loss=0.09079, over 5703420.43 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3447, pruned_loss=0.08881, over 5099122.72 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3314, pruned_loss=0.09107, over 5705048.34 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:41:25,225 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-13 19:41:28,795 INFO [optim.py:369] (1/2) 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 19:41:56,827 INFO [train.py:968] (1/2) Epoch 27, batch 4400, giga_loss[loss=0.2596, simple_loss=0.3324, pruned_loss=0.09343, over 28568.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3293, pruned_loss=0.08897, over 5711136.70 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3442, pruned_loss=0.08859, over 5122168.63 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3288, pruned_loss=0.08941, over 5708011.45 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:42:03,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4082, 1.5801, 1.6282, 1.2341], device='cuda:1'), covar=tensor([0.1935, 0.2641, 0.1603, 0.1813], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0718, 0.0982, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 19:42:08,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2354, 1.4124, 1.5643, 1.3180], device='cuda:1'), covar=tensor([0.1602, 0.1121, 0.1616, 0.1325], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0759, 0.0727, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 19:42:37,401 INFO [train.py:968] (1/2) Epoch 27, batch 4450, giga_loss[loss=0.2999, simple_loss=0.3737, pruned_loss=0.1131, over 28704.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3294, pruned_loss=0.08854, over 5710754.75 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3442, pruned_loss=0.08859, over 5126111.78 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3289, pruned_loss=0.08888, over 5707378.62 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:42:38,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1189306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:42:48,128 INFO [optim.py:369] (1/2) 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 19:42:49,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3052, 1.4947, 1.5219, 1.1539], device='cuda:1'), covar=tensor([0.1797, 0.2579, 0.1566, 0.1815], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0717, 0.0980, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 19:43:11,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1261, 1.4472, 1.5440, 1.2368], device='cuda:1'), covar=tensor([0.2162, 0.1759, 0.2249, 0.2043], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0758, 0.0726, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 19:43:13,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1189348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:43:17,608 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 4500, libri_loss[loss=0.2616, simple_loss=0.3469, pruned_loss=0.08821, over 29171.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3334, pruned_loss=0.09085, over 5695167.85 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3444, pruned_loss=0.08881, over 5127607.54 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3327, pruned_loss=0.09096, over 5698024.36 frames. ], batch size: 101, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:44:03,124 INFO [train.py:968] (1/2) Epoch 27, batch 4550, giga_loss[loss=0.2451, simple_loss=0.3328, pruned_loss=0.07875, over 28836.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.336, pruned_loss=0.09139, over 5700308.10 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3441, pruned_loss=0.08869, over 5131451.51 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3357, pruned_loss=0.09159, over 5701664.56 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:44:14,702 INFO [optim.py:369] (1/2) 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 19:44:49,776 INFO [train.py:968] (1/2) Epoch 27, batch 4600, giga_loss[loss=0.2558, simple_loss=0.3364, pruned_loss=0.08758, over 28318.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3382, pruned_loss=0.09203, over 5692170.97 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3438, pruned_loss=0.08848, over 5139373.88 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3381, pruned_loss=0.09239, over 5691175.82 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:45:21,176 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:45:23,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:45:30,893 INFO [train.py:968] (1/2) Epoch 27, batch 4650, giga_loss[loss=0.279, simple_loss=0.3475, pruned_loss=0.1053, over 27883.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3374, pruned_loss=0.09108, over 5693290.82 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3436, pruned_loss=0.08844, over 5151121.72 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3374, pruned_loss=0.09143, over 5689290.78 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:45:32,310 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 19:45:35,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 19:45:41,138 INFO [optim.py:369] (1/2) 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 19:45:46,669 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:45:51,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6801, 1.9748, 1.8865, 1.7160], device='cuda:1'), covar=tensor([0.2125, 0.2288, 0.2322, 0.2321], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0758, 0.0726, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 19:46:10,981 INFO [train.py:968] (1/2) Epoch 27, batch 4700, giga_loss[loss=0.2748, simple_loss=0.3379, pruned_loss=0.1058, over 28812.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3365, pruned_loss=0.09044, over 5702760.47 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3435, pruned_loss=0.08838, over 5173559.59 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3363, pruned_loss=0.09085, over 5696170.72 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:46:49,486 INFO [train.py:968] (1/2) Epoch 27, batch 4750, giga_loss[loss=0.272, simple_loss=0.3409, pruned_loss=0.1016, over 28642.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3377, pruned_loss=0.09163, over 5701370.75 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3434, pruned_loss=0.08844, over 5188359.59 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3375, pruned_loss=0.09196, over 5692042.15 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 19:47:00,604 INFO [optim.py:369] (1/2) 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 19:47:03,358 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-13 19:47:28,335 INFO [train.py:968] (1/2) Epoch 27, batch 4800, giga_loss[loss=0.247, simple_loss=0.3258, pruned_loss=0.08416, over 28888.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3391, pruned_loss=0.09309, over 5701702.79 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3436, pruned_loss=0.08881, over 5202187.67 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3387, pruned_loss=0.09313, over 5691111.92 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:47:50,092 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-13 19:47:51,838 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2739, 1.3587, 1.2178, 1.4902], device='cuda:1'), covar=tensor([0.0754, 0.0372, 0.0366, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0073, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 19:47:53,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1189681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:47:55,885 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 19:48:11,864 INFO [train.py:968] (1/2) Epoch 27, batch 4850, giga_loss[loss=0.2771, simple_loss=0.3558, pruned_loss=0.09921, over 27712.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3438, pruned_loss=0.09586, over 5702221.63 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3438, pruned_loss=0.08895, over 5208928.80 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3434, pruned_loss=0.09584, over 5692321.02 frames. ], batch size: 474, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:48:12,418 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-13 19:48:22,461 INFO [optim.py:369] (1/2) 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 19:48:28,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1189725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:48:51,424 INFO [train.py:968] (1/2) Epoch 27, batch 4900, giga_loss[loss=0.2678, simple_loss=0.3488, pruned_loss=0.09339, over 28324.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.346, pruned_loss=0.09652, over 5712298.30 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3433, pruned_loss=0.08869, over 5216099.39 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.346, pruned_loss=0.09678, over 5702536.42 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:49:29,890 INFO [train.py:968] (1/2) Epoch 27, batch 4950, giga_loss[loss=0.2629, simple_loss=0.3485, pruned_loss=0.08861, over 28760.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3466, pruned_loss=0.0965, over 5711049.16 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3434, pruned_loss=0.08864, over 5228432.38 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3466, pruned_loss=0.09696, over 5705194.79 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:49:40,343 INFO [optim.py:369] (1/2) 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 19:49:45,341 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:49:47,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:08,539 INFO [train.py:968] (1/2) Epoch 27, batch 5000, giga_loss[loss=0.2882, simple_loss=0.3654, pruned_loss=0.1055, over 28997.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3472, pruned_loss=0.09644, over 5720120.77 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3434, pruned_loss=0.08864, over 5242119.64 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3473, pruned_loss=0.09695, over 5711842.98 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:50:09,531 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:18,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1189868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:20,600 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1189871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:41,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3449, 3.1637, 3.0120, 1.3762], device='cuda:1'), covar=tensor([0.1006, 0.1138, 0.0944, 0.2354], device='cuda:1'), in_proj_covar=tensor([0.1272, 0.1178, 0.0994, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 19:50:43,999 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1189900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:50:48,108 INFO [train.py:968] (1/2) Epoch 27, batch 5050, giga_loss[loss=0.2514, simple_loss=0.3278, pruned_loss=0.08748, over 28852.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3476, pruned_loss=0.09724, over 5724831.61 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08867, over 5248895.86 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.09769, over 5716515.02 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:50:51,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 19:50:57,796 INFO [optim.py:369] (1/2) 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 19:51:30,788 INFO [train.py:968] (1/2) Epoch 27, batch 5100, giga_loss[loss=0.2741, simple_loss=0.35, pruned_loss=0.09907, over 29021.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3453, pruned_loss=0.09621, over 5718251.67 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08867, over 5248895.86 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3454, pruned_loss=0.09656, over 5711778.76 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:52:11,104 INFO [train.py:968] (1/2) Epoch 27, batch 5150, giga_loss[loss=0.2214, simple_loss=0.3027, pruned_loss=0.07009, over 28916.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3415, pruned_loss=0.09417, over 5725423.30 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3434, pruned_loss=0.08866, over 5255500.28 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3416, pruned_loss=0.09452, over 5718743.27 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:52:22,574 INFO [optim.py:369] (1/2) 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 19:52:54,620 INFO [train.py:968] (1/2) Epoch 27, batch 5200, giga_loss[loss=0.2819, simple_loss=0.3417, pruned_loss=0.1111, over 23831.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3397, pruned_loss=0.09325, over 5719152.94 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3434, pruned_loss=0.08862, over 5261729.35 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3398, pruned_loss=0.0936, over 5712541.29 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:52:57,409 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5107, 2.1075, 1.6489, 1.9026], device='cuda:1'), covar=tensor([0.0734, 0.0253, 0.0330, 0.0866], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 19:53:06,609 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-13 19:53:31,891 INFO [train.py:968] (1/2) Epoch 27, batch 5250, giga_loss[loss=0.2483, simple_loss=0.3344, pruned_loss=0.08107, over 28773.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3408, pruned_loss=0.09299, over 5714681.17 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3432, pruned_loss=0.08859, over 5270568.73 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3409, pruned_loss=0.09337, over 5709551.81 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:53:41,235 INFO [optim.py:369] (1/2) 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 19:53:43,367 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.41 vs. limit=5.0
+2023-03-13 19:54:11,313 INFO [train.py:968] (1/2) Epoch 27, batch 5300, giga_loss[loss=0.2511, simple_loss=0.3355, pruned_loss=0.08337, over 28958.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3425, pruned_loss=0.0931, over 5714971.75 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3434, pruned_loss=0.08858, over 5286066.82 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3424, pruned_loss=0.09351, over 5706860.97 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:54:13,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4661, 1.7420, 1.4039, 1.5341], device='cuda:1'), covar=tensor([0.2588, 0.2654, 0.2943, 0.2413], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1136, 0.1393, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 19:54:49,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:54:49,490 INFO [train.py:968] (1/2) Epoch 27, batch 5350, giga_loss[loss=0.2145, simple_loss=0.2884, pruned_loss=0.07025, over 28540.00 frames. ], tot_loss[loss=0.265, simple_loss=0.343, pruned_loss=0.09347, over 5706835.37 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3435, pruned_loss=0.08867, over 5300521.18 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3428, pruned_loss=0.09388, over 5700582.74 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:54:59,685 INFO [optim.py:369] (1/2) 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 19:55:27,131 INFO [train.py:968] (1/2) Epoch 27, batch 5400, giga_loss[loss=0.2443, simple_loss=0.3193, pruned_loss=0.08469, over 28782.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3427, pruned_loss=0.09455, over 5710011.24 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3441, pruned_loss=0.08909, over 5320506.05 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.342, pruned_loss=0.09467, over 5699656.80 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:56:06,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1174, 2.1471, 1.6827, 1.8347], device='cuda:1'), covar=tensor([0.0944, 0.0757, 0.1027, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0448, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 19:56:07,227 INFO [train.py:968] (1/2) Epoch 27, batch 5450, giga_loss[loss=0.2673, simple_loss=0.3315, pruned_loss=0.1016, over 28470.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.342, pruned_loss=0.09563, over 5708232.01 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3446, pruned_loss=0.0894, over 5328463.99 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3411, pruned_loss=0.09557, over 5698182.12 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:56:13,406 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 19:56:17,482 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 5500, giga_loss[loss=0.2598, simple_loss=0.3349, pruned_loss=0.09233, over 28753.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3409, pruned_loss=0.09577, over 5703602.23 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.345, pruned_loss=0.08966, over 5338402.78 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3397, pruned_loss=0.0957, over 5698273.42 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:57:29,166 INFO [train.py:968] (1/2) Epoch 27, batch 5550, giga_loss[loss=0.2309, simple_loss=0.3191, pruned_loss=0.07139, over 28948.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3377, pruned_loss=0.09429, over 5707192.73 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3448, pruned_loss=0.08957, over 5341224.77 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3369, pruned_loss=0.09434, over 5702044.30 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:57:29,450 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:57:32,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.25 vs. limit=5.0
+2023-03-13 19:57:39,421 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 19:57:39,587 INFO [optim.py:369] (1/2) 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 19:57:47,069 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 19:57:51,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3417, 1.4016, 3.8454, 3.3982], device='cuda:1'), covar=tensor([0.1628, 0.2788, 0.0424, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0664, 0.0987, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 19:58:10,757 INFO [train.py:968] (1/2) Epoch 27, batch 5600, giga_loss[loss=0.2248, simple_loss=0.2936, pruned_loss=0.07805, over 28485.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3379, pruned_loss=0.09509, over 5712251.69 frames. ], libri_tot_loss[loss=0.2629, simple_loss=0.3456, pruned_loss=0.09014, over 5348148.11 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3366, pruned_loss=0.09474, over 5706856.85 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:58:43,781 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2562, 0.8087, 0.8874, 1.4203], device='cuda:1'), covar=tensor([0.0760, 0.0390, 0.0384, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 19:58:48,867 INFO [train.py:968] (1/2) Epoch 27, batch 5650, libri_loss[loss=0.2757, simple_loss=0.3619, pruned_loss=0.09478, over 27732.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3332, pruned_loss=0.09237, over 5719198.86 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3457, pruned_loss=0.0902, over 5357919.05 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3317, pruned_loss=0.09209, over 5714322.11 frames. ], batch size: 116, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:59:00,519 INFO [optim.py:369] (1/2) 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 19:59:25,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2994, 4.1529, 3.9217, 1.8754], device='cuda:1'), covar=tensor([0.0638, 0.0791, 0.0735, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1273, 0.1178, 0.0993, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 19:59:27,483 INFO [train.py:968] (1/2) Epoch 27, batch 5700, giga_loss[loss=0.235, simple_loss=0.3123, pruned_loss=0.07887, over 28900.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3285, pruned_loss=0.08971, over 5725001.09 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3453, pruned_loss=0.09012, over 5374435.33 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3272, pruned_loss=0.08957, over 5716076.48 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 19:59:48,511 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 5750, giga_loss[loss=0.2194, simple_loss=0.3039, pruned_loss=0.06748, over 28912.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3284, pruned_loss=0.0893, over 5726494.75 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3448, pruned_loss=0.08971, over 5389286.24 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3275, pruned_loss=0.08955, over 5715020.30 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:00:16,379 INFO [optim.py:369] (1/2) 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 20:00:29,318 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1190636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:00:46,022 INFO [train.py:968] (1/2) Epoch 27, batch 5800, giga_loss[loss=0.243, simple_loss=0.3242, pruned_loss=0.08089, over 28751.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3318, pruned_loss=0.09053, over 5720422.54 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3448, pruned_loss=0.08984, over 5395162.80 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3307, pruned_loss=0.09064, over 5714573.16 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:01:19,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5747, 1.8650, 1.5255, 1.6369], device='cuda:1'), covar=tensor([0.2697, 0.2772, 0.3284, 0.2452], device='cuda:1'), in_proj_covar=tensor([0.1582, 0.1138, 0.1395, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 20:01:23,657 INFO [train.py:968] (1/2) Epoch 27, batch 5850, giga_loss[loss=0.2553, simple_loss=0.3308, pruned_loss=0.08986, over 28589.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3342, pruned_loss=0.09098, over 5721610.52 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3447, pruned_loss=0.08974, over 5401674.84 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3332, pruned_loss=0.09117, over 5716952.38 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:01:36,273 INFO [optim.py:369] (1/2) 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 20:01:37,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1190722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:01:39,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1190725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:01:59,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3790, 1.5676, 1.4737, 1.2619], device='cuda:1'), covar=tensor([0.3107, 0.2855, 0.2251, 0.2692], device='cuda:1'), in_proj_covar=tensor([0.2043, 0.1997, 0.1913, 0.2046], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 20:02:05,006 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1190754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:02:05,435 INFO [train.py:968] (1/2) Epoch 27, batch 5900, giga_loss[loss=0.2442, simple_loss=0.3204, pruned_loss=0.08398, over 28839.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3368, pruned_loss=0.09176, over 5716191.88 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3446, pruned_loss=0.08961, over 5411375.47 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3359, pruned_loss=0.09207, over 5709461.98 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:02:26,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1190780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:02:41,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1190800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:02:44,338 INFO [train.py:968] (1/2) Epoch 27, batch 5950, giga_loss[loss=0.2773, simple_loss=0.3512, pruned_loss=0.1017, over 28863.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3406, pruned_loss=0.09359, over 5712138.89 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3451, pruned_loss=0.08974, over 5418784.70 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3392, pruned_loss=0.09382, over 5710369.18 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:02:58,016 INFO [optim.py:369] (1/2) 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 20:03:06,836 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-13 20:03:30,663 INFO [train.py:968] (1/2) Epoch 27, batch 6000, giga_loss[loss=0.28, simple_loss=0.3522, pruned_loss=0.1039, over 29041.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3451, pruned_loss=0.09754, over 5700204.97 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3454, pruned_loss=0.08993, over 5414025.94 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3438, pruned_loss=0.0976, over 5705630.26 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:03:30,663 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 20:03:39,122 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 20:04:24,927 INFO [train.py:968] (1/2) Epoch 27, batch 6050, giga_loss[loss=0.3869, simple_loss=0.4268, pruned_loss=0.1735, over 28282.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3512, pruned_loss=0.1027, over 5697879.08 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3452, pruned_loss=0.08983, over 5420225.94 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3503, pruned_loss=0.103, over 5701625.76 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:04:42,077 INFO [optim.py:369] (1/2) 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 20:04:43,692 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1190923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:04:45,736 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1190926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:04:58,063 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1190943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:00,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1190946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:09,759 INFO [train.py:968] (1/2) Epoch 27, batch 6100, giga_loss[loss=0.3601, simple_loss=0.3942, pruned_loss=0.163, over 23784.00 frames. ], tot_loss[loss=0.2843, simple_loss=0.3558, pruned_loss=0.1064, over 5681549.07 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3451, pruned_loss=0.08996, over 5434519.34 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.3555, pruned_loss=0.1071, over 5682180.47 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:05:10,009 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1190955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:10,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-13 20:05:28,322 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1190975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:05:53,470 INFO [train.py:968] (1/2) Epoch 27, batch 6150, giga_loss[loss=0.3034, simple_loss=0.3747, pruned_loss=0.116, over 28628.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3624, pruned_loss=0.1113, over 5682057.43 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3447, pruned_loss=0.08974, over 5450561.36 frames. ], giga_tot_loss[loss=0.2942, simple_loss=0.363, pruned_loss=0.1127, over 5675699.95 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:05:54,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2379, 3.0289, 1.4220, 1.3366], device='cuda:1'), covar=tensor([0.1036, 0.0385, 0.0930, 0.1424], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0561, 0.0403, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 20:06:00,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:06:10,555 INFO [optim.py:369] (1/2) 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 20:06:21,367 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7190, 2.6741, 2.5338, 2.3880], device='cuda:1'), covar=tensor([0.1853, 0.2252, 0.2106, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0760, 0.0730, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 20:06:43,934 INFO [train.py:968] (1/2) Epoch 27, batch 6200, giga_loss[loss=0.3371, simple_loss=0.3861, pruned_loss=0.144, over 28827.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3685, pruned_loss=0.117, over 5674780.43 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3447, pruned_loss=0.08974, over 5450561.36 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.369, pruned_loss=0.1181, over 5669832.33 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:06:51,290 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:07:27,980 INFO [train.py:968] (1/2) Epoch 27, batch 6250, giga_loss[loss=0.3692, simple_loss=0.424, pruned_loss=0.1572, over 28781.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3745, pruned_loss=0.1218, over 5679843.34 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3445, pruned_loss=0.08971, over 5459577.10 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3755, pruned_loss=0.1232, over 5671983.85 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:07:29,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2720, 1.4816, 1.4207, 1.2107], device='cuda:1'), covar=tensor([0.2614, 0.2517, 0.2045, 0.2390], device='cuda:1'), in_proj_covar=tensor([0.2037, 0.1994, 0.1911, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 20:07:44,552 INFO [optim.py:369] (1/2) 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 20:08:16,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 6300, giga_loss[loss=0.3295, simple_loss=0.3863, pruned_loss=0.1363, over 28277.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3776, pruned_loss=0.125, over 5656252.80 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3445, pruned_loss=0.08972, over 5463999.77 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.379, pruned_loss=0.1267, over 5649689.63 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:08:19,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1191157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:08:47,072 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:08:47,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4069, 4.0398, 1.4918, 1.7132], device='cuda:1'), covar=tensor([0.0988, 0.0299, 0.0895, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0564, 0.0404, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 20:09:04,613 INFO [train.py:968] (1/2) Epoch 27, batch 6350, giga_loss[loss=0.3179, simple_loss=0.3808, pruned_loss=0.1275, over 28700.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3788, pruned_loss=0.1271, over 5654506.76 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3442, pruned_loss=0.08959, over 5474999.37 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.381, pruned_loss=0.1294, over 5643816.49 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:09:11,874 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:09:21,886 INFO [optim.py:369] (1/2) 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 20:09:56,833 INFO [train.py:968] (1/2) Epoch 27, batch 6400, giga_loss[loss=0.3185, simple_loss=0.3814, pruned_loss=0.1277, over 28874.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3841, pruned_loss=0.1329, over 5617620.01 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3444, pruned_loss=0.08971, over 5469522.29 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.386, pruned_loss=0.135, over 5616113.70 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:10:45,214 INFO [train.py:968] (1/2) Epoch 27, batch 6450, giga_loss[loss=0.3354, simple_loss=0.4042, pruned_loss=0.1333, over 28965.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3881, pruned_loss=0.1363, over 5613810.98 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3442, pruned_loss=0.08968, over 5478296.61 frames. ], giga_tot_loss[loss=0.3341, simple_loss=0.3905, pruned_loss=0.1388, over 5607152.00 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:11:01,152 INFO [optim.py:369] (1/2) 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 20:11:28,176 INFO [train.py:968] (1/2) Epoch 27, batch 6500, giga_loss[loss=0.3343, simple_loss=0.3908, pruned_loss=0.1389, over 29004.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3877, pruned_loss=0.1356, over 5621449.55 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3447, pruned_loss=0.09002, over 5485236.87 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3909, pruned_loss=0.1391, over 5614808.89 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:12:12,813 INFO [train.py:968] (1/2) Epoch 27, batch 6550, libri_loss[loss=0.2979, simple_loss=0.3714, pruned_loss=0.1122, over 19624.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3875, pruned_loss=0.136, over 5632765.34 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3448, pruned_loss=0.08999, over 5486952.57 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3912, pruned_loss=0.1401, over 5630552.67 frames. ], batch size: 187, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:12:17,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191410.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:12:28,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6287, 1.5631, 1.8076, 1.4288], device='cuda:1'), covar=tensor([0.1544, 0.2334, 0.1323, 0.1674], device='cuda:1'), in_proj_covar=tensor([0.0922, 0.0710, 0.0970, 0.0869], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 20:12:29,626 INFO [optim.py:369] (1/2) 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 20:12:40,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:12:56,065 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 20:12:57,028 INFO [train.py:968] (1/2) Epoch 27, batch 6600, giga_loss[loss=0.2894, simple_loss=0.3563, pruned_loss=0.1112, over 28819.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3864, pruned_loss=0.1353, over 5637067.61 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3446, pruned_loss=0.09003, over 5502006.55 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.3912, pruned_loss=0.1404, over 5627920.51 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:13:17,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7100, 1.9311, 1.6345, 1.6458], device='cuda:1'), covar=tensor([0.2331, 0.2389, 0.2516, 0.2297], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1144, 0.1400, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 20:13:25,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7639, 2.0109, 1.6583, 1.7932], device='cuda:1'), covar=tensor([0.2753, 0.2829, 0.3166, 0.2641], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1144, 0.1401, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 20:13:41,643 INFO [train.py:968] (1/2) Epoch 27, batch 6650, giga_loss[loss=0.2973, simple_loss=0.3753, pruned_loss=0.1097, over 28981.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3863, pruned_loss=0.1342, over 5641039.16 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3449, pruned_loss=0.09016, over 5511469.45 frames. ], giga_tot_loss[loss=0.3347, simple_loss=0.3909, pruned_loss=0.1392, over 5628097.34 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:13:59,538 INFO [optim.py:369] (1/2) 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 20:14:27,110 INFO [train.py:968] (1/2) Epoch 27, batch 6700, giga_loss[loss=0.389, simple_loss=0.4345, pruned_loss=0.1718, over 27981.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3856, pruned_loss=0.1329, over 5648173.49 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3444, pruned_loss=0.08984, over 5524267.53 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3913, pruned_loss=0.1389, over 5630714.41 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:14:51,992 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1191584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:14:56,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:15:12,252 INFO [train.py:968] (1/2) Epoch 27, batch 6750, giga_loss[loss=0.3023, simple_loss=0.3708, pruned_loss=0.1168, over 29006.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3857, pruned_loss=0.1328, over 5633283.43 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.344, pruned_loss=0.08963, over 5535196.16 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3918, pruned_loss=0.1391, over 5612221.74 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:15:18,909 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1790, 1.7879, 1.3751, 0.3982], device='cuda:1'), covar=tensor([0.5028, 0.3002, 0.4171, 0.6598], device='cuda:1'), in_proj_covar=tensor([0.1807, 0.1700, 0.1640, 0.1477], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 20:15:19,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:15:22,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-13 20:15:26,652 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:16:00,050 INFO [train.py:968] (1/2) Epoch 27, batch 6800, giga_loss[loss=0.2891, simple_loss=0.3639, pruned_loss=0.1072, over 28665.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3826, pruned_loss=0.1301, over 5613547.74 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3445, pruned_loss=0.09006, over 5526856.41 frames. ], giga_tot_loss[loss=0.3298, simple_loss=0.388, pruned_loss=0.1358, over 5607150.71 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:16:25,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3641, 1.6738, 1.3767, 0.9619], device='cuda:1'), covar=tensor([0.2250, 0.2261, 0.2497, 0.2316], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1144, 0.1402, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 20:16:44,134 INFO [train.py:968] (1/2) Epoch 27, batch 6850, giga_loss[loss=0.2797, simple_loss=0.3615, pruned_loss=0.09893, over 29000.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3809, pruned_loss=0.1273, over 5628582.76 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.3444, pruned_loss=0.09002, over 5537928.93 frames. ], giga_tot_loss[loss=0.3261, simple_loss=0.3864, pruned_loss=0.133, over 5615739.33 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:17:00,683 INFO [optim.py:369] (1/2) 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 20:17:07,797 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1191729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:17:10,353 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 6900, giga_loss[loss=0.2798, simple_loss=0.3441, pruned_loss=0.1077, over 28778.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3771, pruned_loss=0.124, over 5633944.91 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3443, pruned_loss=0.09, over 5532268.64 frames. ], giga_tot_loss[loss=0.3196, simple_loss=0.3818, pruned_loss=0.1287, over 5629159.02 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:17:39,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:18:06,453 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:18:13,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3656, 1.3922, 3.7372, 3.1655], device='cuda:1'), covar=tensor([0.1584, 0.2702, 0.0442, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0666, 0.0992, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 20:18:24,346 INFO [train.py:968] (1/2) Epoch 27, batch 6950, giga_loss[loss=0.2971, simple_loss=0.3661, pruned_loss=0.114, over 28274.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3744, pruned_loss=0.1216, over 5640085.97 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3442, pruned_loss=0.08995, over 5536934.39 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3787, pruned_loss=0.1259, over 5633355.32 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:18:38,458 INFO [optim.py:369] (1/2) 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 20:19:05,331 INFO [train.py:968] (1/2) Epoch 27, batch 7000, giga_loss[loss=0.2561, simple_loss=0.3361, pruned_loss=0.08811, over 28867.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3716, pruned_loss=0.1196, over 5655948.17 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3442, pruned_loss=0.08992, over 5555766.14 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3766, pruned_loss=0.1247, over 5638111.71 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:19:42,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6518, 2.3692, 1.7024, 0.9605], device='cuda:1'), covar=tensor([0.6265, 0.3034, 0.4386, 0.6707], device='cuda:1'), in_proj_covar=tensor([0.1809, 0.1699, 0.1637, 0.1476], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 20:19:46,820 INFO [train.py:968] (1/2) Epoch 27, batch 7050, giga_loss[loss=0.2902, simple_loss=0.3656, pruned_loss=0.1074, over 29005.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3714, pruned_loss=0.1193, over 5669260.54 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3439, pruned_loss=0.08979, over 5567404.10 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3766, pruned_loss=0.1244, over 5647941.13 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:20:04,511 INFO [optim.py:369] (1/2) 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 20:20:10,995 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1191928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:20:13,695 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1191931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:20:41,807 INFO [train.py:968] (1/2) Epoch 27, batch 7100, giga_loss[loss=0.2664, simple_loss=0.3541, pruned_loss=0.08933, over 28656.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3714, pruned_loss=0.1188, over 5668301.65 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3441, pruned_loss=0.08985, over 5570928.04 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3756, pruned_loss=0.1231, over 5649272.53 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:20:47,641 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1191960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:20:50,980 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1191963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:21:22,714 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1191996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:21:29,358 INFO [train.py:968] (1/2) Epoch 27, batch 7150, libri_loss[loss=0.2859, simple_loss=0.3631, pruned_loss=0.1043, over 29549.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3696, pruned_loss=0.1164, over 5672512.52 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08981, over 5568872.67 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3737, pruned_loss=0.1205, over 5661220.76 frames. ], batch size: 84, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:21:51,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-13 20:21:54,114 INFO [optim.py:369] (1/2) 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 20:22:00,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5384, 1.9397, 1.4938, 1.6373], device='cuda:1'), covar=tensor([0.0781, 0.0301, 0.0339, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0121, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0113], device='cuda:1')
+2023-03-13 20:22:10,996 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-13 20:22:26,220 INFO [train.py:968] (1/2) Epoch 27, batch 7200, giga_loss[loss=0.2928, simple_loss=0.3714, pruned_loss=0.1071, over 28952.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3706, pruned_loss=0.1152, over 5666838.69 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3438, pruned_loss=0.08973, over 5572230.59 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3741, pruned_loss=0.1188, over 5655828.03 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:23:16,054 INFO [train.py:968] (1/2) Epoch 27, batch 7250, giga_loss[loss=0.2732, simple_loss=0.3547, pruned_loss=0.09584, over 28950.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3713, pruned_loss=0.1156, over 5667024.51 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.344, pruned_loss=0.08984, over 5581311.77 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3747, pruned_loss=0.1191, over 5652469.61 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:23:30,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4584, 1.5377, 1.3407, 1.1528], device='cuda:1'), covar=tensor([0.1056, 0.0615, 0.1034, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0452, 0.0525, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 20:23:37,660 INFO [optim.py:369] (1/2) 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 20:23:49,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:23:50,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:23:53,763 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 7300, giga_loss[loss=0.275, simple_loss=0.3482, pruned_loss=0.1009, over 28686.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.37, pruned_loss=0.1148, over 5677581.07 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3441, pruned_loss=0.08985, over 5588349.75 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3735, pruned_loss=0.1184, over 5662364.25 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:24:18,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:24:43,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3384, 1.4235, 1.3250, 1.2760], device='cuda:1'), covar=tensor([0.2458, 0.2330, 0.2007, 0.2203], device='cuda:1'), in_proj_covar=tensor([0.2047, 0.2002, 0.1919, 0.2054], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 20:24:49,150 INFO [train.py:968] (1/2) Epoch 27, batch 7350, giga_loss[loss=0.3331, simple_loss=0.3921, pruned_loss=0.137, over 27865.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3679, pruned_loss=0.1144, over 5676935.43 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3435, pruned_loss=0.08963, over 5594503.15 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3718, pruned_loss=0.1179, over 5660953.48 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:25:09,599 INFO [optim.py:369] (1/2) 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 20:25:34,083 INFO [train.py:968] (1/2) Epoch 27, batch 7400, giga_loss[loss=0.2432, simple_loss=0.3223, pruned_loss=0.08207, over 28587.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3674, pruned_loss=0.1153, over 5675803.09 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3437, pruned_loss=0.08977, over 5604013.76 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.371, pruned_loss=0.1187, over 5656830.70 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:26:16,216 INFO [train.py:968] (1/2) Epoch 27, batch 7450, giga_loss[loss=0.2812, simple_loss=0.357, pruned_loss=0.1027, over 28568.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3666, pruned_loss=0.1151, over 5684288.62 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3438, pruned_loss=0.08994, over 5606197.01 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3699, pruned_loss=0.1183, over 5669360.83 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:26:37,216 INFO [optim.py:369] (1/2) 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 20:26:49,109 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1192338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:27:04,516 INFO [train.py:968] (1/2) Epoch 27, batch 7500, giga_loss[loss=0.3175, simple_loss=0.3792, pruned_loss=0.1279, over 28266.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3658, pruned_loss=0.1134, over 5699243.88 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3437, pruned_loss=0.09001, over 5616806.84 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3694, pruned_loss=0.1167, over 5680075.49 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:27:07,640 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:27:10,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3181, 3.1122, 1.4467, 1.5598], device='cuda:1'), covar=tensor([0.1065, 0.0458, 0.0960, 0.1415], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0567, 0.0405, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 20:27:10,581 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-13 20:27:44,477 INFO [train.py:968] (1/2) Epoch 27, batch 7550, giga_loss[loss=0.2508, simple_loss=0.3382, pruned_loss=0.08164, over 29008.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3647, pruned_loss=0.1114, over 5707150.82 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3433, pruned_loss=0.08978, over 5625446.49 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3686, pruned_loss=0.115, over 5686448.93 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:28:04,469 INFO [optim.py:369] (1/2) 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 20:28:29,387 INFO [train.py:968] (1/2) Epoch 27, batch 7600, giga_loss[loss=0.2945, simple_loss=0.365, pruned_loss=0.112, over 28754.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3647, pruned_loss=0.1119, over 5692652.90 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.09, over 5619243.22 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3679, pruned_loss=0.1151, over 5683698.56 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:28:52,844 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:28:57,346 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:29:19,562 INFO [train.py:968] (1/2) Epoch 27, batch 7650, giga_loss[loss=0.2688, simple_loss=0.3431, pruned_loss=0.09729, over 29067.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.364, pruned_loss=0.1122, over 5694884.09 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.0902, over 5621955.49 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3665, pruned_loss=0.1147, over 5686217.10 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:29:25,846 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:29:31,793 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1192521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:29:34,152 INFO [optim.py:369] (1/2) 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 20:30:05,276 INFO [train.py:968] (1/2) Epoch 27, batch 7700, giga_loss[loss=0.2695, simple_loss=0.3491, pruned_loss=0.09492, over 28808.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3626, pruned_loss=0.1121, over 5685092.62 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09014, over 5628281.33 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3652, pruned_loss=0.1146, over 5674337.91 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:30:49,382 INFO [train.py:968] (1/2) Epoch 27, batch 7750, giga_loss[loss=0.2835, simple_loss=0.3529, pruned_loss=0.107, over 28850.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3621, pruned_loss=0.1121, over 5683774.93 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09011, over 5622240.12 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3646, pruned_loss=0.1146, over 5681913.99 frames. ], batch size: 285, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:31:08,178 INFO [optim.py:369] (1/2) 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 20:31:21,440 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-13 20:31:34,930 INFO [train.py:968] (1/2) Epoch 27, batch 7800, giga_loss[loss=0.2895, simple_loss=0.3583, pruned_loss=0.1104, over 29104.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3613, pruned_loss=0.1116, over 5699093.96 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3436, pruned_loss=0.08986, over 5631907.48 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3642, pruned_loss=0.1147, over 5691243.77 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:31:44,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:31:48,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:32:13,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:32:21,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0329, 2.4473, 1.7747, 2.1339], device='cuda:1'), covar=tensor([0.1011, 0.0575, 0.0963, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0451, 0.0524, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 20:32:21,470 INFO [train.py:968] (1/2) Epoch 27, batch 7850, libri_loss[loss=0.256, simple_loss=0.3448, pruned_loss=0.08361, over 29278.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3586, pruned_loss=0.1103, over 5702427.38 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3432, pruned_loss=0.08961, over 5639682.48 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3618, pruned_loss=0.1137, over 5690655.74 frames. ], batch size: 94, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:32:29,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192716.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:32:36,225 INFO [optim.py:369] (1/2) 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 20:32:43,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1192734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:32:49,796 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192743.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:33:00,458 INFO [train.py:968] (1/2) Epoch 27, batch 7900, giga_loss[loss=0.2926, simple_loss=0.3634, pruned_loss=0.111, over 28665.00 frames. ], tot_loss[loss=0.2899, simple_loss=0.3584, pruned_loss=0.1107, over 5700238.28 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.343, pruned_loss=0.0895, over 5640130.12 frames. ], giga_tot_loss[loss=0.2952, simple_loss=0.3618, pruned_loss=0.1143, over 5692192.55 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:33:46,796 INFO [train.py:968] (1/2) Epoch 27, batch 7950, giga_loss[loss=0.3583, simple_loss=0.4004, pruned_loss=0.1581, over 26676.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3598, pruned_loss=0.1119, over 5694187.36 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3429, pruned_loss=0.08943, over 5644435.79 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3629, pruned_loss=0.1152, over 5684885.29 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:33:49,273 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4293, 4.4225, 1.5736, 1.6918], device='cuda:1'), covar=tensor([0.1071, 0.0340, 0.0944, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0569, 0.0405, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 20:34:05,364 INFO [optim.py:369] (1/2) 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 20:34:32,200 INFO [train.py:968] (1/2) Epoch 27, batch 8000, giga_loss[loss=0.2959, simple_loss=0.3645, pruned_loss=0.1137, over 28819.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3613, pruned_loss=0.1123, over 5680769.65 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3429, pruned_loss=0.0896, over 5637783.41 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3641, pruned_loss=0.1154, over 5679636.07 frames. ], batch size: 243, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:34:49,876 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1192877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:34:52,118 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1192880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:35:10,312 INFO [train.py:968] (1/2) Epoch 27, batch 8050, giga_loss[loss=0.2693, simple_loss=0.3454, pruned_loss=0.09658, over 28942.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3599, pruned_loss=0.1104, over 5668194.56 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3425, pruned_loss=0.08952, over 5628820.52 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3632, pruned_loss=0.1137, over 5676873.38 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:35:16,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1192909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:35:30,155 INFO [optim.py:369] (1/2) 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 20:35:34,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9776, 2.3681, 1.5943, 1.8797], device='cuda:1'), covar=tensor([0.1086, 0.0644, 0.1062, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0452, 0.0525, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 20:35:55,511 INFO [train.py:968] (1/2) Epoch 27, batch 8100, giga_loss[loss=0.2998, simple_loss=0.3702, pruned_loss=0.1146, over 29055.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3619, pruned_loss=0.112, over 5666043.12 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08957, over 5632940.20 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3649, pruned_loss=0.1152, over 5669938.46 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:36:01,220 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1192960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:36:07,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0549, 2.2345, 2.2844, 1.9008], device='cuda:1'), covar=tensor([0.3277, 0.2705, 0.2694, 0.2962], device='cuda:1'), in_proj_covar=tensor([0.2052, 0.2006, 0.1922, 0.2059], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 20:36:23,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-13 20:36:43,461 INFO [train.py:968] (1/2) Epoch 27, batch 8150, giga_loss[loss=0.3398, simple_loss=0.399, pruned_loss=0.1403, over 28601.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.363, pruned_loss=0.1129, over 5668469.04 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3424, pruned_loss=0.0895, over 5626847.81 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3659, pruned_loss=0.1158, over 5677142.28 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:37:04,008 INFO [optim.py:369] (1/2) 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 20:37:22,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-13 20:37:33,666 INFO [train.py:968] (1/2) Epoch 27, batch 8200, giga_loss[loss=0.3254, simple_loss=0.3832, pruned_loss=0.1338, over 28619.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3646, pruned_loss=0.1152, over 5666623.07 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3424, pruned_loss=0.08956, over 5630696.14 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3674, pruned_loss=0.1179, over 5670718.06 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:38:10,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193091.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 20:38:21,690 INFO [train.py:968] (1/2) Epoch 27, batch 8250, giga_loss[loss=0.3142, simple_loss=0.3723, pruned_loss=0.1281, over 28653.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3673, pruned_loss=0.1184, over 5664615.47 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3426, pruned_loss=0.0896, over 5628407.94 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3694, pruned_loss=0.1207, over 5669980.32 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:38:37,531 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193118.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:38:44,269 INFO [optim.py:369] (1/2) 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 20:39:01,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3751, 1.7659, 1.6438, 1.5276], device='cuda:1'), covar=tensor([0.2380, 0.2111, 0.2296, 0.2159], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0762, 0.0728, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 20:39:10,538 INFO [train.py:968] (1/2) Epoch 27, batch 8300, giga_loss[loss=0.2816, simple_loss=0.3493, pruned_loss=0.107, over 28889.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3688, pruned_loss=0.1203, over 5665235.63 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3428, pruned_loss=0.08963, over 5634555.95 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3711, pruned_loss=0.123, over 5664535.09 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:39:50,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8103, 2.0856, 2.1259, 1.6618], device='cuda:1'), covar=tensor([0.3600, 0.2957, 0.3080, 0.3374], device='cuda:1'), in_proj_covar=tensor([0.2057, 0.2015, 0.1928, 0.2066], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 20:39:53,341 INFO [train.py:968] (1/2) Epoch 27, batch 8350, giga_loss[loss=0.2996, simple_loss=0.3737, pruned_loss=0.1128, over 29102.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3689, pruned_loss=0.1206, over 5665860.94 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3428, pruned_loss=0.08959, over 5637338.88 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3711, pruned_loss=0.1233, over 5663055.50 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:40:09,892 INFO [optim.py:369] (1/2) 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 20:40:16,421 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193234.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:40:17,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2415, 1.5227, 1.6313, 1.3071], device='cuda:1'), covar=tensor([0.2038, 0.1766, 0.2336, 0.2080], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0763, 0.0728, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 20:40:18,371 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193237.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:40:32,964 INFO [train.py:968] (1/2) Epoch 27, batch 8400, giga_loss[loss=0.2664, simple_loss=0.3485, pruned_loss=0.09214, over 29023.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3688, pruned_loss=0.1199, over 5671086.45 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3429, pruned_loss=0.08955, over 5645550.53 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3712, pruned_loss=0.1229, over 5662357.32 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:40:38,760 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193261.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:40:42,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193264.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 20:40:43,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1193266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:40:43,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 20:41:05,581 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193293.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 20:41:17,292 INFO [train.py:968] (1/2) Epoch 27, batch 8450, giga_loss[loss=0.2968, simple_loss=0.3611, pruned_loss=0.1162, over 28857.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3689, pruned_loss=0.1193, over 5666245.95 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3428, pruned_loss=0.08952, over 5649749.56 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1222, over 5656102.70 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:41:33,170 INFO [optim.py:369] (1/2) 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 20:41:40,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:41:40,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.49 vs. limit=5.0
+2023-03-13 20:41:55,783 INFO [train.py:968] (1/2) Epoch 27, batch 8500, giga_loss[loss=0.3223, simple_loss=0.3805, pruned_loss=0.1321, over 28743.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3679, pruned_loss=0.1176, over 5683047.42 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3436, pruned_loss=0.08974, over 5659418.23 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3701, pruned_loss=0.1208, over 5666949.74 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:41:58,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 20:42:42,302 INFO [train.py:968] (1/2) Epoch 27, batch 8550, giga_loss[loss=0.2874, simple_loss=0.3504, pruned_loss=0.1123, over 28800.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3656, pruned_loss=0.1172, over 5689119.83 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3434, pruned_loss=0.08965, over 5665099.87 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.368, pruned_loss=0.1204, over 5671890.28 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:42:57,921 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-13 20:43:01,890 INFO [optim.py:369] (1/2) 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 20:43:29,982 INFO [train.py:968] (1/2) Epoch 27, batch 8600, libri_loss[loss=0.225, simple_loss=0.301, pruned_loss=0.07451, over 29678.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3634, pruned_loss=0.1162, over 5682553.43 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08948, over 5671284.16 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3661, pruned_loss=0.1196, over 5663417.55 frames. ], batch size: 73, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:43:53,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:43:58,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:44:01,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5674, 1.6931, 1.7324, 1.5108], device='cuda:1'), covar=tensor([0.3181, 0.2957, 0.2451, 0.2965], device='cuda:1'), in_proj_covar=tensor([0.2059, 0.2016, 0.1930, 0.2066], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 20:44:18,591 INFO [train.py:968] (1/2) Epoch 27, batch 8650, giga_loss[loss=0.313, simple_loss=0.3856, pruned_loss=0.1202, over 28956.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3656, pruned_loss=0.1179, over 5658384.17 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3437, pruned_loss=0.08983, over 5663812.32 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3676, pruned_loss=0.1208, over 5649871.81 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:44:22,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:44:35,993 INFO [optim.py:369] (1/2) 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 20:44:37,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-13 20:45:01,904 INFO [train.py:968] (1/2) Epoch 27, batch 8700, giga_loss[loss=0.321, simple_loss=0.3915, pruned_loss=0.1253, over 27524.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3678, pruned_loss=0.1173, over 5663415.26 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08957, over 5663868.17 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3708, pruned_loss=0.1209, over 5656615.77 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:45:46,458 INFO [train.py:968] (1/2) Epoch 27, batch 8750, libri_loss[loss=0.2616, simple_loss=0.3478, pruned_loss=0.08769, over 29541.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3691, pruned_loss=0.1159, over 5672416.86 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3431, pruned_loss=0.08952, over 5668677.79 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.372, pruned_loss=0.1193, over 5662461.44 frames. ], batch size: 84, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:45:48,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6406, 1.8450, 1.2771, 1.4534], device='cuda:1'), covar=tensor([0.1105, 0.0648, 0.1107, 0.1150], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0451, 0.0523, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 20:46:07,266 INFO [optim.py:369] (1/2) 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 20:46:21,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1193641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:46:32,889 INFO [train.py:968] (1/2) Epoch 27, batch 8800, libri_loss[loss=0.282, simple_loss=0.3715, pruned_loss=0.09628, over 29782.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3733, pruned_loss=0.119, over 5670511.28 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3433, pruned_loss=0.08956, over 5670116.41 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3755, pruned_loss=0.1218, over 5661287.28 frames. ], batch size: 87, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:47:02,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6877, 1.5856, 1.8416, 1.4202], device='cuda:1'), covar=tensor([0.1776, 0.2644, 0.1452, 0.1744], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0715, 0.0973, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 20:47:16,250 INFO [train.py:968] (1/2) Epoch 27, batch 8850, giga_loss[loss=0.3414, simple_loss=0.3944, pruned_loss=0.1443, over 27838.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3749, pruned_loss=0.1211, over 5658651.06 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08946, over 5670005.04 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3774, pruned_loss=0.1239, over 5651601.63 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:47:35,486 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 8900, giga_loss[loss=0.302, simple_loss=0.3653, pruned_loss=0.1194, over 28273.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3728, pruned_loss=0.1201, over 5667172.06 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3425, pruned_loss=0.08912, over 5677768.23 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3766, pruned_loss=0.124, over 5653671.86 frames. ], batch size: 65, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:48:24,381 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1193784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:48:25,668 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1193786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:48:26,443 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1193787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:48:42,941 INFO [train.py:968] (1/2) Epoch 27, batch 8950, giga_loss[loss=0.295, simple_loss=0.3577, pruned_loss=0.1161, over 28954.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.371, pruned_loss=0.1199, over 5653510.80 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3423, pruned_loss=0.08902, over 5679608.32 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3749, pruned_loss=0.1238, over 5640985.37 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:48:54,598 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1193816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:49:01,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-13 20:49:04,843 INFO [optim.py:369] (1/2) 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 20:49:24,333 INFO [train.py:968] (1/2) Epoch 27, batch 9000, giga_loss[loss=0.2793, simple_loss=0.3497, pruned_loss=0.1045, over 28928.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3684, pruned_loss=0.1182, over 5662814.53 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3421, pruned_loss=0.08897, over 5688261.19 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3732, pruned_loss=0.123, over 5643204.33 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:49:24,333 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 20:49:32,819 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 20:50:03,326 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-13 20:50:14,508 INFO [train.py:968] (1/2) Epoch 27, batch 9050, giga_loss[loss=0.3769, simple_loss=0.4105, pruned_loss=0.1716, over 26706.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3665, pruned_loss=0.1174, over 5671762.28 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3418, pruned_loss=0.08877, over 5694142.14 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3714, pruned_loss=0.1223, over 5650088.83 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:50:36,806 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 9100, giga_loss[loss=0.2795, simple_loss=0.3541, pruned_loss=0.1024, over 28555.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3671, pruned_loss=0.1189, over 5665562.44 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3418, pruned_loss=0.08881, over 5697378.07 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3713, pruned_loss=0.1231, over 5645279.16 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:51:53,225 INFO [train.py:968] (1/2) Epoch 27, batch 9150, libri_loss[loss=0.2396, simple_loss=0.3162, pruned_loss=0.08156, over 28562.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1197, over 5657476.77 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08883, over 5700421.04 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3717, pruned_loss=0.1237, over 5637627.27 frames. ], batch size: 63, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:51:57,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4119, 1.2973, 4.0419, 3.2626], device='cuda:1'), covar=tensor([0.1664, 0.2766, 0.0477, 0.0947], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0674, 0.1004, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 20:52:11,187 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1194024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:52:13,655 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 9200, giga_loss[loss=0.3092, simple_loss=0.3759, pruned_loss=0.1212, over 28957.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3657, pruned_loss=0.1192, over 5665018.70 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3414, pruned_loss=0.08859, over 5703512.60 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3695, pruned_loss=0.123, over 5645941.08 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 20:53:14,476 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-13 20:53:19,720 INFO [train.py:968] (1/2) Epoch 27, batch 9250, giga_loss[loss=0.2461, simple_loss=0.3274, pruned_loss=0.08242, over 28981.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3654, pruned_loss=0.1186, over 5662390.22 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08883, over 5707371.46 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3686, pruned_loss=0.1222, over 5642690.68 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:53:40,082 INFO [optim.py:369] (1/2) 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 20:53:58,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9535, 5.7945, 5.5255, 3.2012], device='cuda:1'), covar=tensor([0.0458, 0.0574, 0.0608, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.1203, 0.1016, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 20:54:07,775 INFO [train.py:968] (1/2) Epoch 27, batch 9300, giga_loss[loss=0.2964, simple_loss=0.3662, pruned_loss=0.1133, over 28706.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3666, pruned_loss=0.1183, over 5665915.49 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.342, pruned_loss=0.08885, over 5708363.60 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3692, pruned_loss=0.1212, over 5649542.71 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:54:14,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:54:49,661 INFO [train.py:968] (1/2) Epoch 27, batch 9350, giga_loss[loss=0.2764, simple_loss=0.3464, pruned_loss=0.1032, over 29002.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3678, pruned_loss=0.1186, over 5673659.12 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3417, pruned_loss=0.08862, over 5714325.93 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.1221, over 5653819.11 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:55:00,959 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4577, 1.5283, 3.4773, 3.2908], device='cuda:1'), covar=tensor([0.1307, 0.2506, 0.0485, 0.1064], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0672, 0.1003, 0.0974], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 20:55:11,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2416, 0.8650, 0.9299, 1.3239], device='cuda:1'), covar=tensor([0.0790, 0.0373, 0.0365, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0113], device='cuda:1')
+2023-03-13 20:55:12,273 INFO [optim.py:369] (1/2) 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 20:55:35,226 INFO [train.py:968] (1/2) Epoch 27, batch 9400, giga_loss[loss=0.3079, simple_loss=0.3788, pruned_loss=0.1184, over 28835.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3688, pruned_loss=0.12, over 5663198.13 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3418, pruned_loss=0.08861, over 5713327.50 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3716, pruned_loss=0.1232, over 5647830.07 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:55:57,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5400, 1.7591, 1.3069, 1.3071], device='cuda:1'), covar=tensor([0.1176, 0.0684, 0.1153, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0453, 0.0526, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 20:56:23,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1194304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:56:23,758 INFO [train.py:968] (1/2) Epoch 27, batch 9450, giga_loss[loss=0.4003, simple_loss=0.419, pruned_loss=0.1908, over 23482.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.371, pruned_loss=0.119, over 5664426.96 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3417, pruned_loss=0.08861, over 5714299.86 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3733, pruned_loss=0.1217, over 5651423.37 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:56:25,289 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:56:47,890 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1194336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:56:57,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6196, 2.2423, 1.5579, 0.8754], device='cuda:1'), covar=tensor([0.6908, 0.3375, 0.4641, 0.7332], device='cuda:1'), in_proj_covar=tensor([0.1816, 0.1713, 0.1638, 0.1478], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 20:57:02,500 INFO [train.py:968] (1/2) Epoch 27, batch 9500, giga_loss[loss=0.356, simple_loss=0.4237, pruned_loss=0.1441, over 28695.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3713, pruned_loss=0.1173, over 5674110.38 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3419, pruned_loss=0.08884, over 5718400.87 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3742, pruned_loss=0.1204, over 5657700.92 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:57:22,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5772, 1.7658, 1.7890, 1.3410], device='cuda:1'), covar=tensor([0.2027, 0.2921, 0.1764, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0718, 0.0976, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 20:57:38,754 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:57:41,137 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 20:57:43,075 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 9550, giga_loss[loss=0.3064, simple_loss=0.3766, pruned_loss=0.1181, over 28581.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3737, pruned_loss=0.1183, over 5673037.30 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.08897, over 5715314.47 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3772, pruned_loss=0.1216, over 5661795.77 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 20:58:03,620 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-13 20:58:05,075 INFO [optim.py:369] (1/2) 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 20:58:28,326 INFO [train.py:968] (1/2) Epoch 27, batch 9600, giga_loss[loss=0.2909, simple_loss=0.3679, pruned_loss=0.1069, over 28394.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3765, pruned_loss=0.1208, over 5668507.58 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3424, pruned_loss=0.08928, over 5709445.70 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3798, pruned_loss=0.1241, over 5662798.78 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:58:46,984 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-13 20:58:52,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 20:59:08,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5863, 1.7701, 1.7813, 1.5525], device='cuda:1'), covar=tensor([0.2171, 0.2268, 0.2375, 0.2327], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0762, 0.0728, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 20:59:08,600 INFO [train.py:968] (1/2) Epoch 27, batch 9650, giga_loss[loss=0.2779, simple_loss=0.3529, pruned_loss=0.1014, over 28844.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3769, pruned_loss=0.1215, over 5682714.06 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3419, pruned_loss=0.08901, over 5717239.95 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3813, pruned_loss=0.1256, over 5669704.44 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 20:59:19,014 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1194514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:59:32,204 INFO [optim.py:369] (1/2) 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 20:59:43,456 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1194542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:59:44,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.19 vs. limit=5.0
+2023-03-13 20:59:47,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1194545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 20:59:57,035 INFO [train.py:968] (1/2) Epoch 27, batch 9700, giga_loss[loss=0.2958, simple_loss=0.3713, pruned_loss=0.1101, over 29079.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.377, pruned_loss=0.1228, over 5661461.08 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3423, pruned_loss=0.0893, over 5715361.20 frames. ], giga_tot_loss[loss=0.3163, simple_loss=0.3805, pruned_loss=0.1261, over 5652468.99 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:00:13,162 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1194574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:00:39,725 INFO [train.py:968] (1/2) Epoch 27, batch 9750, giga_loss[loss=0.3404, simple_loss=0.4104, pruned_loss=0.1353, over 28923.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3764, pruned_loss=0.1222, over 5668785.42 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3423, pruned_loss=0.08928, over 5718471.37 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3799, pruned_loss=0.1256, over 5657787.06 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:00:44,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-13 21:00:51,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0945, 2.3785, 2.1098, 2.1320], device='cuda:1'), covar=tensor([0.2132, 0.2301, 0.2208, 0.2323], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0761, 0.0727, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 21:01:01,202 INFO [optim.py:369] (1/2) 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 21:01:22,420 INFO [train.py:968] (1/2) Epoch 27, batch 9800, giga_loss[loss=0.3025, simple_loss=0.372, pruned_loss=0.1165, over 28009.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3749, pruned_loss=0.1193, over 5673216.57 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3422, pruned_loss=0.08923, over 5720579.77 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.378, pruned_loss=0.1224, over 5662227.31 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:01:26,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1998, 1.8799, 1.4179, 0.4323], device='cuda:1'), covar=tensor([0.5669, 0.3647, 0.5227, 0.7351], device='cuda:1'), in_proj_covar=tensor([0.1816, 0.1713, 0.1642, 0.1479], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 21:01:30,918 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-13 21:02:03,888 INFO [train.py:968] (1/2) Epoch 27, batch 9850, giga_loss[loss=0.2664, simple_loss=0.3444, pruned_loss=0.09424, over 28715.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3739, pruned_loss=0.1178, over 5677270.02 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.089, over 5725044.92 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3778, pruned_loss=0.1213, over 5663339.26 frames. ], batch size: 66, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:02:13,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-13 21:02:25,565 INFO [optim.py:369] (1/2) 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 21:02:50,877 INFO [train.py:968] (1/2) Epoch 27, batch 9900, giga_loss[loss=0.3141, simple_loss=0.3861, pruned_loss=0.121, over 28849.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3751, pruned_loss=0.1187, over 5675937.72 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08915, over 5728557.45 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3785, pruned_loss=0.1219, over 5661256.64 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:02:53,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5232, 4.3476, 4.1649, 1.9765], device='cuda:1'), covar=tensor([0.0680, 0.0849, 0.0830, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.1209, 0.1020, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 21:02:58,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9698, 2.2342, 2.1364, 1.7919], device='cuda:1'), covar=tensor([0.2978, 0.2724, 0.3035, 0.2996], device='cuda:1'), in_proj_covar=tensor([0.2051, 0.2004, 0.1929, 0.2060], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 21:03:13,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:03:40,100 INFO [train.py:968] (1/2) Epoch 27, batch 9950, giga_loss[loss=0.2806, simple_loss=0.3478, pruned_loss=0.1066, over 28682.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3751, pruned_loss=0.1196, over 5672301.42 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3421, pruned_loss=0.08907, over 5730772.80 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3784, pruned_loss=0.1227, over 5657517.91 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:04:02,734 INFO [optim.py:369] (1/2) 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 21:04:15,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3463, 4.2294, 1.6562, 1.7149], device='cuda:1'), covar=tensor([0.1238, 0.0437, 0.0934, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0570, 0.0407, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 21:04:26,192 INFO [train.py:968] (1/2) Epoch 27, batch 10000, giga_loss[loss=0.3332, simple_loss=0.3895, pruned_loss=0.1385, over 27880.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3741, pruned_loss=0.1206, over 5665283.31 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3421, pruned_loss=0.08908, over 5734205.60 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3774, pruned_loss=0.1237, over 5649450.06 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:04:52,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8397, 1.9270, 1.6514, 1.8675], device='cuda:1'), covar=tensor([0.2809, 0.2950, 0.3278, 0.2597], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1145, 0.1402, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 21:04:58,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1194889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:05:12,536 INFO [train.py:968] (1/2) Epoch 27, batch 10050, giga_loss[loss=0.3042, simple_loss=0.3814, pruned_loss=0.1134, over 29039.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3736, pruned_loss=0.1213, over 5670137.99 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3424, pruned_loss=0.08921, over 5734209.79 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3765, pruned_loss=0.1242, over 5656587.81 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:05:30,035 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1194922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:05:33,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1194925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:05:38,556 INFO [optim.py:369] (1/2) 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 21:06:01,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1194954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:06:01,991 INFO [train.py:968] (1/2) Epoch 27, batch 10100, giga_loss[loss=0.2845, simple_loss=0.3568, pruned_loss=0.1062, over 28829.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.371, pruned_loss=0.1204, over 5662762.07 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3425, pruned_loss=0.08927, over 5735873.24 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3736, pruned_loss=0.1231, over 5649653.11 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:06:30,693 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-13 21:06:44,750 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2558, 1.3465, 3.7074, 3.1474], device='cuda:1'), covar=tensor([0.1729, 0.2730, 0.0502, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0673, 0.1007, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 21:06:49,545 INFO [train.py:968] (1/2) Epoch 27, batch 10150, libri_loss[loss=0.2475, simple_loss=0.3291, pruned_loss=0.08297, over 29569.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3695, pruned_loss=0.1202, over 5660066.25 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3422, pruned_loss=0.089, over 5738470.72 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3723, pruned_loss=0.1231, over 5646000.66 frames. ], batch size: 75, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:07:12,452 INFO [optim.py:369] (1/2) 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 21:07:13,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1195032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:07:16,166 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1195035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:07:16,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1195036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:07:31,975 INFO [train.py:968] (1/2) Epoch 27, batch 10200, giga_loss[loss=0.2575, simple_loss=0.339, pruned_loss=0.088, over 28842.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.1189, over 5668672.97 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3426, pruned_loss=0.08914, over 5742818.58 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3705, pruned_loss=0.1219, over 5651867.13 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:07:42,702 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 21:07:43,157 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1195064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:08:21,727 INFO [train.py:968] (1/2) Epoch 27, batch 10250, giga_loss[loss=0.2668, simple_loss=0.3479, pruned_loss=0.09287, over 29060.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3649, pruned_loss=0.1151, over 5674358.59 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08907, over 5741263.24 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.118, over 5661130.62 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:08:44,526 INFO [optim.py:369] (1/2) 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 21:09:07,210 INFO [train.py:968] (1/2) Epoch 27, batch 10300, giga_loss[loss=0.2907, simple_loss=0.3596, pruned_loss=0.1109, over 28622.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3609, pruned_loss=0.1117, over 5667331.22 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.342, pruned_loss=0.08873, over 5746530.01 frames. ], giga_tot_loss[loss=0.297, simple_loss=0.3639, pruned_loss=0.1151, over 5649792.53 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:09:21,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.43 vs. limit=2.0
+2023-03-13 21:09:51,830 INFO [train.py:968] (1/2) Epoch 27, batch 10350, giga_loss[loss=0.3046, simple_loss=0.3732, pruned_loss=0.118, over 28761.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3604, pruned_loss=0.1105, over 5674322.55 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3424, pruned_loss=0.08879, over 5747697.98 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3632, pruned_loss=0.114, over 5656196.19 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:10:15,251 INFO [optim.py:369] (1/2) 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 21:10:38,156 INFO [train.py:968] (1/2) Epoch 27, batch 10400, giga_loss[loss=0.289, simple_loss=0.349, pruned_loss=0.1145, over 28347.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.36, pruned_loss=0.1112, over 5668426.01 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08893, over 5743610.49 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3625, pruned_loss=0.1145, over 5655489.77 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:11:24,418 INFO [train.py:968] (1/2) Epoch 27, batch 10450, giga_loss[loss=0.2658, simple_loss=0.3365, pruned_loss=0.09753, over 28716.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3577, pruned_loss=0.1108, over 5667252.13 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3427, pruned_loss=0.08893, over 5744433.44 frames. ], giga_tot_loss[loss=0.2933, simple_loss=0.3597, pruned_loss=0.1134, over 5655950.40 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:11:45,319 INFO [optim.py:369] (1/2) 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 21:12:05,307 INFO [train.py:968] (1/2) Epoch 27, batch 10500, giga_loss[loss=0.3534, simple_loss=0.4022, pruned_loss=0.1523, over 27878.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3595, pruned_loss=0.1113, over 5670266.07 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3429, pruned_loss=0.08891, over 5745365.15 frames. ], giga_tot_loss[loss=0.2949, simple_loss=0.3614, pruned_loss=0.1142, over 5657767.82 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:12:15,619 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5981, 1.7921, 1.8216, 1.5530], device='cuda:1'), covar=tensor([0.3384, 0.2378, 0.1973, 0.2540], device='cuda:1'), in_proj_covar=tensor([0.2052, 0.2005, 0.1933, 0.2061], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 21:12:50,258 INFO [train.py:968] (1/2) Epoch 27, batch 10550, giga_loss[loss=0.3629, simple_loss=0.4157, pruned_loss=0.155, over 28325.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3631, pruned_loss=0.113, over 5662930.75 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3436, pruned_loss=0.08937, over 5740313.38 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3645, pruned_loss=0.1155, over 5654441.57 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:12:55,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1195411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:12:55,902 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1361, 1.2692, 1.0374, 0.9032], device='cuda:1'), covar=tensor([0.0904, 0.0409, 0.0876, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0453, 0.0526, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 21:13:13,155 INFO [optim.py:369] (1/2) 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 21:13:35,550 INFO [train.py:968] (1/2) Epoch 27, batch 10600, giga_loss[loss=0.2754, simple_loss=0.343, pruned_loss=0.104, over 28683.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3638, pruned_loss=0.114, over 5660336.82 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3431, pruned_loss=0.08906, over 5744317.75 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3657, pruned_loss=0.1168, over 5648394.73 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:14:21,281 INFO [train.py:968] (1/2) Epoch 27, batch 10650, giga_loss[loss=0.3092, simple_loss=0.369, pruned_loss=0.1247, over 27927.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3635, pruned_loss=0.1142, over 5638856.36 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3437, pruned_loss=0.08943, over 5726584.10 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3648, pruned_loss=0.1165, over 5643365.49 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:14:26,002 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-13 21:14:42,449 INFO [optim.py:369] (1/2) 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 21:15:01,531 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 10700, giga_loss[loss=0.3628, simple_loss=0.4132, pruned_loss=0.1562, over 27989.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.366, pruned_loss=0.1162, over 5651823.53 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3443, pruned_loss=0.08969, over 5728778.02 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3671, pruned_loss=0.1184, over 5651557.85 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:15:04,920 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1195557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:15:35,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1195586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:15:50,701 INFO [train.py:968] (1/2) Epoch 27, batch 10750, giga_loss[loss=0.2528, simple_loss=0.3292, pruned_loss=0.0882, over 28934.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3677, pruned_loss=0.1169, over 5654004.93 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3441, pruned_loss=0.08949, over 5732641.29 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3693, pruned_loss=0.1196, over 5648141.80 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:16:03,770 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.67 vs. limit=2.0
+2023-03-13 21:16:15,355 INFO [optim.py:369] (1/2) 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 21:16:33,793 INFO [train.py:968] (1/2) Epoch 27, batch 10800, giga_loss[loss=0.3328, simple_loss=0.3908, pruned_loss=0.1374, over 28789.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3697, pruned_loss=0.1183, over 5665259.40 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3443, pruned_loss=0.08959, over 5735577.02 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3712, pruned_loss=0.1208, over 5656670.46 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:17:18,428 INFO [train.py:968] (1/2) Epoch 27, batch 10850, giga_loss[loss=0.3333, simple_loss=0.3849, pruned_loss=0.1409, over 27508.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3693, pruned_loss=0.118, over 5672197.13 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3446, pruned_loss=0.08968, over 5734520.29 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3712, pruned_loss=0.121, over 5663867.88 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:17:44,474 INFO [optim.py:369] (1/2) 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 21:18:06,785 INFO [train.py:968] (1/2) Epoch 27, batch 10900, giga_loss[loss=0.2928, simple_loss=0.3742, pruned_loss=0.1057, over 28803.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3701, pruned_loss=0.1187, over 5666336.47 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3444, pruned_loss=0.08955, over 5727849.44 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3723, pruned_loss=0.1218, over 5663755.67 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:18:15,352 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.64 vs. limit=5.0
+2023-03-13 21:18:55,011 INFO [train.py:968] (1/2) Epoch 27, batch 10950, giga_loss[loss=0.3116, simple_loss=0.3762, pruned_loss=0.1235, over 28909.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3708, pruned_loss=0.118, over 5661114.74 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3444, pruned_loss=0.08956, over 5729806.25 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3729, pruned_loss=0.1209, over 5656446.41 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:19:17,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 21:19:20,569 INFO [optim.py:369] (1/2) 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 21:19:42,515 INFO [train.py:968] (1/2) Epoch 27, batch 11000, giga_loss[loss=0.2748, simple_loss=0.3415, pruned_loss=0.1041, over 28526.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3705, pruned_loss=0.1187, over 5659472.52 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3445, pruned_loss=0.08959, over 5734409.59 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3729, pruned_loss=0.1217, over 5650040.29 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:20:06,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3373, 3.2418, 1.5206, 1.5366], device='cuda:1'), covar=tensor([0.1007, 0.0377, 0.0867, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0570, 0.0406, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 21:20:36,946 INFO [train.py:968] (1/2) Epoch 27, batch 11050, giga_loss[loss=0.3335, simple_loss=0.3912, pruned_loss=0.1379, over 28846.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3711, pruned_loss=0.1203, over 5642535.72 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3443, pruned_loss=0.0896, over 5735784.87 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3734, pruned_loss=0.123, over 5633027.66 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:21:01,979 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 21:21:12,067 INFO [optim.py:369] (1/2) 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 21:21:26,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6254, 1.9838, 1.5644, 1.9130], device='cuda:1'), covar=tensor([0.3024, 0.3039, 0.3425, 0.2499], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1146, 0.1400, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 21:21:32,608 INFO [train.py:968] (1/2) Epoch 27, batch 11100, giga_loss[loss=0.2912, simple_loss=0.3634, pruned_loss=0.1095, over 28973.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3687, pruned_loss=0.1192, over 5647833.66 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3438, pruned_loss=0.08932, over 5737275.10 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3713, pruned_loss=0.1221, over 5637586.15 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:22:17,918 INFO [train.py:968] (1/2) Epoch 27, batch 11150, libri_loss[loss=0.2316, simple_loss=0.3152, pruned_loss=0.07394, over 29455.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3661, pruned_loss=0.1177, over 5648393.12 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3433, pruned_loss=0.08892, over 5742489.10 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1212, over 5632984.98 frames. ], batch size: 70, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:22:39,216 INFO [optim.py:369] (1/2) 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 21:22:41,131 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5419, 1.5840, 1.2798, 1.1603], device='cuda:1'), covar=tensor([0.0851, 0.0469, 0.0857, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0453, 0.0526, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 21:22:57,852 INFO [train.py:968] (1/2) Epoch 27, batch 11200, giga_loss[loss=0.2586, simple_loss=0.3334, pruned_loss=0.09189, over 28917.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3661, pruned_loss=0.1174, over 5658792.64 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3439, pruned_loss=0.0892, over 5739841.57 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1213, over 5644653.46 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:23:17,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3970, 1.3861, 3.5031, 3.3656], device='cuda:1'), covar=tensor([0.1392, 0.2606, 0.0474, 0.1063], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0670, 0.1001, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 21:23:42,004 INFO [train.py:968] (1/2) Epoch 27, batch 11250, libri_loss[loss=0.2098, simple_loss=0.2937, pruned_loss=0.06299, over 29634.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3639, pruned_loss=0.1163, over 5656895.54 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3429, pruned_loss=0.08875, over 5736102.28 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3682, pruned_loss=0.1208, over 5645706.85 frames. ], batch size: 74, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:24:07,747 INFO [optim.py:369] (1/2) 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 21:24:27,777 INFO [train.py:968] (1/2) Epoch 27, batch 11300, giga_loss[loss=0.3292, simple_loss=0.3905, pruned_loss=0.1339, over 28665.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3646, pruned_loss=0.1168, over 5661053.69 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3429, pruned_loss=0.08883, over 5739316.89 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3687, pruned_loss=0.1213, over 5646899.95 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:25:16,110 INFO [train.py:968] (1/2) Epoch 27, batch 11350, giga_loss[loss=0.4023, simple_loss=0.4311, pruned_loss=0.1867, over 26600.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3672, pruned_loss=0.1195, over 5650030.31 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3429, pruned_loss=0.08881, over 5731976.79 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3707, pruned_loss=0.1234, over 5644821.38 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:25:17,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6510, 2.0742, 1.5664, 1.9416], device='cuda:1'), covar=tensor([0.3024, 0.2991, 0.3563, 0.2715], device='cuda:1'), in_proj_covar=tensor([0.1587, 0.1147, 0.1401, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 21:25:41,650 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-13 21:25:42,398 INFO [optim.py:369] (1/2) 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 21:26:01,688 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7809, 1.9029, 1.5303, 1.4117], device='cuda:1'), covar=tensor([0.1054, 0.0664, 0.1007, 0.1190], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0453, 0.0525, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 21:26:02,813 INFO [train.py:968] (1/2) Epoch 27, batch 11400, giga_loss[loss=0.2407, simple_loss=0.3155, pruned_loss=0.0829, over 28624.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3684, pruned_loss=0.1204, over 5647857.43 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.08871, over 5735543.37 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3718, pruned_loss=0.1242, over 5639124.76 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:26:51,040 INFO [train.py:968] (1/2) Epoch 27, batch 11450, giga_loss[loss=0.3754, simple_loss=0.4055, pruned_loss=0.1726, over 26766.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3686, pruned_loss=0.1212, over 5642073.10 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3424, pruned_loss=0.08876, over 5731407.16 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3725, pruned_loss=0.1252, over 5636634.00 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:27:17,254 INFO [optim.py:369] (1/2) 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 21:27:35,402 INFO [train.py:968] (1/2) Epoch 27, batch 11500, giga_loss[loss=0.3394, simple_loss=0.3941, pruned_loss=0.1424, over 28616.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5659881.40 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3428, pruned_loss=0.0887, over 5738401.56 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.373, pruned_loss=0.1258, over 5646141.17 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:27:55,082 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1196374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:27:55,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-13 21:28:23,898 INFO [train.py:968] (1/2) Epoch 27, batch 11550, giga_loss[loss=0.3165, simple_loss=0.3752, pruned_loss=0.1289, over 28677.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3702, pruned_loss=0.122, over 5655252.92 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3424, pruned_loss=0.08846, over 5741097.40 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3742, pruned_loss=0.1266, over 5639977.78 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:28:47,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7147, 4.5534, 4.3561, 2.1494], device='cuda:1'), covar=tensor([0.0591, 0.0738, 0.0781, 0.1755], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.1213, 0.1024, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 21:28:51,977 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 11600, giga_loss[loss=0.3508, simple_loss=0.4095, pruned_loss=0.1461, over 28270.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1218, over 5670177.04 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3425, pruned_loss=0.08853, over 5742606.99 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3742, pruned_loss=0.1256, over 5656037.97 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:29:59,095 INFO [train.py:968] (1/2) Epoch 27, batch 11650, giga_loss[loss=0.3804, simple_loss=0.4116, pruned_loss=0.1746, over 26420.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.372, pruned_loss=0.1228, over 5653884.00 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3426, pruned_loss=0.08878, over 5736853.97 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3756, pruned_loss=0.1269, over 5644357.73 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:30:11,378 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1196518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:30:28,837 INFO [optim.py:369] (1/2) 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 21:30:48,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3068, 1.7084, 1.4371, 1.4507], device='cuda:1'), covar=tensor([0.0775, 0.0308, 0.0329, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 21:30:48,334 INFO [train.py:968] (1/2) Epoch 27, batch 11700, giga_loss[loss=0.3216, simple_loss=0.3885, pruned_loss=0.1273, over 28989.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3737, pruned_loss=0.1243, over 5654740.59 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3426, pruned_loss=0.08877, over 5740003.98 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3772, pruned_loss=0.1282, over 5642782.22 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:31:20,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2299, 1.5472, 1.4901, 1.3725], device='cuda:1'), covar=tensor([0.1876, 0.1457, 0.2269, 0.1799], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0764, 0.0729, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 21:31:32,558 INFO [train.py:968] (1/2) Epoch 27, batch 11750, giga_loss[loss=0.2734, simple_loss=0.3435, pruned_loss=0.1017, over 28682.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3734, pruned_loss=0.1244, over 5660256.67 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3423, pruned_loss=0.08869, over 5741562.69 frames. ], giga_tot_loss[loss=0.3176, simple_loss=0.3775, pruned_loss=0.1288, over 5646395.84 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:31:59,628 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 11800, giga_loss[loss=0.3632, simple_loss=0.4036, pruned_loss=0.1614, over 26533.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3748, pruned_loss=0.1246, over 5659280.64 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3421, pruned_loss=0.08858, over 5744716.66 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.3789, pruned_loss=0.1289, over 5643948.01 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:33:05,970 INFO [train.py:968] (1/2) Epoch 27, batch 11850, giga_loss[loss=0.2695, simple_loss=0.3431, pruned_loss=0.09794, over 28599.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3744, pruned_loss=0.1233, over 5658712.18 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3422, pruned_loss=0.08863, over 5747960.99 frames. ], giga_tot_loss[loss=0.3164, simple_loss=0.3782, pruned_loss=0.1274, over 5642181.41 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:33:34,527 INFO [optim.py:369] (1/2) 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 21:33:48,233 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1196749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:33:53,747 INFO [train.py:968] (1/2) Epoch 27, batch 11900, giga_loss[loss=0.3503, simple_loss=0.3964, pruned_loss=0.1521, over 26506.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3736, pruned_loss=0.1227, over 5660649.11 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3421, pruned_loss=0.08867, over 5749571.19 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.377, pruned_loss=0.1262, over 5645274.52 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:34:36,389 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1196801.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:34:38,772 INFO [train.py:968] (1/2) Epoch 27, batch 11950, giga_loss[loss=0.2739, simple_loss=0.3503, pruned_loss=0.09874, over 28877.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3706, pruned_loss=0.1206, over 5660133.58 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3422, pruned_loss=0.08869, over 5751104.88 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3734, pruned_loss=0.1236, over 5646106.08 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:35:09,254 INFO [optim.py:369] (1/2) 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 21:35:16,628 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-13 21:35:28,180 INFO [train.py:968] (1/2) Epoch 27, batch 12000, giga_loss[loss=0.3328, simple_loss=0.3876, pruned_loss=0.139, over 28877.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3711, pruned_loss=0.1206, over 5657731.25 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08862, over 5743890.82 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3742, pruned_loss=0.1238, over 5650977.74 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:35:28,180 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 21:35:36,334 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 21:35:59,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 21:36:08,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1196891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:09,997 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1196892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:10,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1196893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:12,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1196895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:22,749 INFO [train.py:968] (1/2) Epoch 27, batch 12050, giga_loss[loss=0.2969, simple_loss=0.3709, pruned_loss=0.1115, over 28925.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3728, pruned_loss=0.1219, over 5648227.09 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3423, pruned_loss=0.08884, over 5737534.12 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3756, pruned_loss=0.125, over 5646537.06 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:36:39,591 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1196924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:36:45,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4384, 1.9240, 1.4830, 1.4817], device='cuda:1'), covar=tensor([0.0745, 0.0306, 0.0329, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0113], device='cuda:1')
+2023-03-13 21:36:49,007 INFO [optim.py:369] (1/2) 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 21:36:51,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 21:37:04,779 INFO [train.py:968] (1/2) Epoch 27, batch 12100, giga_loss[loss=0.3127, simple_loss=0.3672, pruned_loss=0.1291, over 28628.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3699, pruned_loss=0.1197, over 5671209.74 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3423, pruned_loss=0.08881, over 5745232.93 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3737, pruned_loss=0.1239, over 5658459.13 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:37:21,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5549, 1.8850, 1.4709, 1.7114], device='cuda:1'), covar=tensor([0.2760, 0.2855, 0.3152, 0.2473], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1147, 0.1399, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 21:37:33,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1328, 2.5758, 1.7816, 2.2268], device='cuda:1'), covar=tensor([0.1037, 0.0599, 0.1032, 0.1123], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0454, 0.0527, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 21:37:52,448 INFO [train.py:968] (1/2) Epoch 27, batch 12150, libri_loss[loss=0.229, simple_loss=0.3068, pruned_loss=0.07557, over 29369.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3705, pruned_loss=0.1208, over 5671818.60 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08858, over 5748391.51 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3744, pruned_loss=0.1251, over 5657127.28 frames. ], batch size: 67, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:38:16,875 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:17,809 INFO [optim.py:369] (1/2) 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 21:38:18,217 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:18,840 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:19,707 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-13 21:38:20,836 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:36,080 INFO [train.py:968] (1/2) Epoch 27, batch 12200, giga_loss[loss=0.2965, simple_loss=0.3633, pruned_loss=0.1149, over 28951.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3721, pruned_loss=0.1218, over 5676895.42 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3426, pruned_loss=0.08875, over 5752822.58 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3757, pruned_loss=0.1261, over 5659060.40 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:38:45,299 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:47,252 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:38:49,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:39:11,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4269, 1.8041, 1.5971, 1.5320], device='cuda:1'), covar=tensor([0.0768, 0.0337, 0.0330, 0.0878], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0113], device='cuda:1')
+2023-03-13 21:39:16,082 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2660, 1.5070, 1.3781, 1.1397], device='cuda:1'), covar=tensor([0.2674, 0.2635, 0.1922, 0.2428], device='cuda:1'), in_proj_covar=tensor([0.2051, 0.2010, 0.1931, 0.2062], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 21:39:24,752 INFO [train.py:968] (1/2) Epoch 27, batch 12250, giga_loss[loss=0.4416, simple_loss=0.4641, pruned_loss=0.2096, over 27823.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.374, pruned_loss=0.1234, over 5671322.38 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3427, pruned_loss=0.08875, over 5754461.55 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.377, pruned_loss=0.1271, over 5655108.73 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:39:32,976 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:39:46,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2946, 0.8409, 0.9933, 1.4021], device='cuda:1'), covar=tensor([0.0781, 0.0400, 0.0364, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0121, 0.0120, 0.0227], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0066, 0.0113], device='cuda:1')
+2023-03-13 21:39:54,839 INFO [optim.py:369] (1/2) 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 21:40:14,311 INFO [train.py:968] (1/2) Epoch 27, batch 12300, giga_loss[loss=0.3065, simple_loss=0.3709, pruned_loss=0.1211, over 28226.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3739, pruned_loss=0.1229, over 5681277.86 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3428, pruned_loss=0.08883, over 5755854.88 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3765, pruned_loss=0.1261, over 5666484.44 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:40:15,096 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:40:33,170 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197176.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:40:47,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4256, 3.4245, 1.5374, 1.6180], device='cuda:1'), covar=tensor([0.0989, 0.0303, 0.0921, 0.1307], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0572, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 21:40:58,326 INFO [train.py:968] (1/2) Epoch 27, batch 12350, giga_loss[loss=0.2834, simple_loss=0.3602, pruned_loss=0.1033, over 28656.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3722, pruned_loss=0.1215, over 5672477.45 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3423, pruned_loss=0.08858, over 5761970.69 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3758, pruned_loss=0.1256, over 5652382.19 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:41:19,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6282, 1.7024, 1.4615, 1.6288], device='cuda:1'), covar=tensor([0.2664, 0.2842, 0.3118, 0.2405], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1147, 0.1401, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 21:41:27,160 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 12400, giga_loss[loss=0.2853, simple_loss=0.3556, pruned_loss=0.1075, over 28958.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3712, pruned_loss=0.1198, over 5683838.83 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3423, pruned_loss=0.08845, over 5764859.65 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3748, pruned_loss=0.1239, over 5663255.23 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:42:05,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5760, 2.8349, 1.6106, 1.7009], device='cuda:1'), covar=tensor([0.0780, 0.0381, 0.0749, 0.1109], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0572, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 21:42:26,133 INFO [train.py:968] (1/2) Epoch 27, batch 12450, giga_loss[loss=0.2776, simple_loss=0.3451, pruned_loss=0.105, over 28886.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3689, pruned_loss=0.1176, over 5686926.79 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3427, pruned_loss=0.08861, over 5756983.61 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3725, pruned_loss=0.1218, over 5674734.59 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:42:37,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 21:42:42,091 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197319.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:42:44,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197322.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:42:52,679 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:43:05,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-13 21:43:11,213 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197351.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:43:12,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 21:43:14,092 INFO [train.py:968] (1/2) Epoch 27, batch 12500, giga_loss[loss=0.2776, simple_loss=0.3518, pruned_loss=0.1017, over 28757.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3675, pruned_loss=0.1173, over 5677817.73 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3425, pruned_loss=0.08856, over 5758085.28 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3706, pruned_loss=0.1209, over 5666624.87 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:43:24,525 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-13 21:43:56,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2147, 0.8194, 0.9091, 1.5312], device='cuda:1'), covar=tensor([0.0766, 0.0375, 0.0351, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 21:44:02,538 INFO [train.py:968] (1/2) Epoch 27, batch 12550, giga_loss[loss=0.4194, simple_loss=0.4335, pruned_loss=0.2027, over 26642.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.366, pruned_loss=0.1171, over 5674669.35 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3427, pruned_loss=0.08854, over 5760924.75 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5661518.72 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:44:10,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-13 21:44:34,375 INFO [optim.py:369] (1/2) 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 21:44:34,594 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 12600, giga_loss[loss=0.2764, simple_loss=0.3444, pruned_loss=0.1043, over 28662.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3626, pruned_loss=0.1158, over 5683720.94 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3423, pruned_loss=0.08835, over 5762479.07 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3654, pruned_loss=0.1189, over 5671316.11 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:44:53,507 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:45:06,592 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-13 21:45:20,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:45:28,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-13 21:45:38,467 INFO [train.py:968] (1/2) Epoch 27, batch 12650, giga_loss[loss=0.2547, simple_loss=0.3165, pruned_loss=0.09645, over 28563.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.36, pruned_loss=0.1144, over 5683857.86 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3423, pruned_loss=0.08835, over 5756604.63 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3626, pruned_loss=0.1175, over 5676831.84 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:46:03,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:46:09,661 INFO [optim.py:369] (1/2) 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 21:46:17,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5363, 1.7510, 1.6922, 1.5573], device='cuda:1'), covar=tensor([0.1889, 0.2219, 0.2208, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0765, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 21:46:26,819 INFO [train.py:968] (1/2) Epoch 27, batch 12700, giga_loss[loss=0.3528, simple_loss=0.3979, pruned_loss=0.1538, over 26616.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3595, pruned_loss=0.1143, over 5691306.02 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3427, pruned_loss=0.08852, over 5759535.00 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3615, pruned_loss=0.117, over 5682056.42 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:46:52,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:46:54,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5473, 3.1433, 1.6443, 1.6466], device='cuda:1'), covar=tensor([0.0869, 0.0366, 0.0807, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0572, 0.0407, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 21:46:54,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:47:14,782 INFO [train.py:968] (1/2) Epoch 27, batch 12750, giga_loss[loss=0.2999, simple_loss=0.3677, pruned_loss=0.116, over 27964.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 5687779.78 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3425, pruned_loss=0.08843, over 5762436.99 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3618, pruned_loss=0.1155, over 5676489.31 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 21:47:24,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:47:38,954 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:47:49,319 INFO [optim.py:369] (1/2) 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 21:48:05,788 INFO [train.py:968] (1/2) Epoch 27, batch 12800, giga_loss[loss=0.3123, simple_loss=0.3856, pruned_loss=0.1195, over 28593.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3589, pruned_loss=0.1102, over 5679419.58 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3424, pruned_loss=0.08844, over 5763933.98 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3608, pruned_loss=0.1126, over 5668545.79 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:48:08,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.28 vs. limit=5.0
+2023-03-13 21:48:13,572 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:25,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:28,615 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:54,624 INFO [train.py:968] (1/2) Epoch 27, batch 12850, giga_loss[loss=0.2568, simple_loss=0.3428, pruned_loss=0.08541, over 28874.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3572, pruned_loss=0.108, over 5669696.07 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3424, pruned_loss=0.08864, over 5756786.40 frames. ], giga_tot_loss[loss=0.2899, simple_loss=0.3592, pruned_loss=0.1103, over 5664846.93 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:48:56,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:48:56,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:49:08,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4985, 1.7786, 1.7467, 1.5041], device='cuda:1'), covar=tensor([0.2897, 0.2189, 0.1831, 0.2346], device='cuda:1'), in_proj_covar=tensor([0.2048, 0.2009, 0.1932, 0.2061], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 21:49:20,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-13 21:49:29,161 INFO [optim.py:369] (1/2) 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 21:49:45,681 INFO [train.py:968] (1/2) Epoch 27, batch 12900, giga_loss[loss=0.2601, simple_loss=0.3398, pruned_loss=0.09022, over 28938.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3542, pruned_loss=0.105, over 5661333.80 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3419, pruned_loss=0.08843, over 5753035.89 frames. ], giga_tot_loss[loss=0.2857, simple_loss=0.3565, pruned_loss=0.1074, over 5659283.58 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:49:56,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1197766.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:50:32,324 INFO [train.py:968] (1/2) Epoch 27, batch 12950, giga_loss[loss=0.253, simple_loss=0.339, pruned_loss=0.08346, over 28056.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3507, pruned_loss=0.1013, over 5673231.81 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08857, over 5759234.51 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.353, pruned_loss=0.1037, over 5662666.31 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:50:34,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1679, 3.9884, 3.7842, 1.7691], device='cuda:1'), covar=tensor([0.0707, 0.0835, 0.0919, 0.2257], device='cuda:1'), in_proj_covar=tensor([0.1304, 0.1204, 0.1015, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 21:50:57,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1197831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:51:03,626 INFO [optim.py:369] (1/2) 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 21:51:13,463 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:51:16,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:51:19,823 INFO [train.py:968] (1/2) Epoch 27, batch 13000, giga_loss[loss=0.2368, simple_loss=0.3316, pruned_loss=0.07104, over 28987.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3486, pruned_loss=0.0981, over 5665486.72 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3412, pruned_loss=0.08836, over 5753309.03 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3512, pruned_loss=0.1004, over 5660302.59 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:51:47,415 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1197881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:52:10,199 INFO [train.py:968] (1/2) Epoch 27, batch 13050, giga_loss[loss=0.2899, simple_loss=0.3648, pruned_loss=0.1075, over 28866.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3489, pruned_loss=0.0985, over 5653314.68 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08854, over 5748753.00 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3514, pruned_loss=0.1005, over 5649925.80 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:52:36,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3919, 1.5748, 1.6141, 1.2188], device='cuda:1'), covar=tensor([0.1856, 0.2812, 0.1584, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.0924, 0.0712, 0.0969, 0.0870], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0014], device='cuda:1')
+2023-03-13 21:52:38,746 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 13100, giga_loss[loss=0.2675, simple_loss=0.3429, pruned_loss=0.0961, over 28598.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3477, pruned_loss=0.0975, over 5667608.25 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3408, pruned_loss=0.08866, over 5752475.57 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3502, pruned_loss=0.09944, over 5658149.27 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:53:14,624 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1197974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:53:17,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1197977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:53:20,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3769, 1.8089, 1.6466, 1.6006], device='cuda:1'), covar=tensor([0.2033, 0.2021, 0.1925, 0.1929], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0752, 0.0723, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 21:53:45,066 INFO [train.py:968] (1/2) Epoch 27, batch 13150, giga_loss[loss=0.245, simple_loss=0.3399, pruned_loss=0.07502, over 28963.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3447, pruned_loss=0.09556, over 5667337.47 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3409, pruned_loss=0.08866, over 5755302.12 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3468, pruned_loss=0.09725, over 5655609.98 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:53:46,963 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:54:27,740 INFO [train.py:968] (1/2) Epoch 27, batch 13200, giga_loss[loss=0.2521, simple_loss=0.3289, pruned_loss=0.08765, over 27815.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3428, pruned_loss=0.09441, over 5669587.37 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3406, pruned_loss=0.08868, over 5755147.90 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3448, pruned_loss=0.09597, over 5657125.16 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 21:54:32,163 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 21:55:16,076 INFO [train.py:968] (1/2) Epoch 27, batch 13250, giga_loss[loss=0.2585, simple_loss=0.3381, pruned_loss=0.08942, over 28314.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3431, pruned_loss=0.09423, over 5663413.01 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3407, pruned_loss=0.08883, over 5747172.32 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3446, pruned_loss=0.09543, over 5659961.05 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:55:46,477 INFO [optim.py:369] (1/2) 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 21:55:50,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198141.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:56:03,520 INFO [train.py:968] (1/2) Epoch 27, batch 13300, giga_loss[loss=0.2287, simple_loss=0.3141, pruned_loss=0.07166, over 28637.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3416, pruned_loss=0.09295, over 5657900.84 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3408, pruned_loss=0.08894, over 5742607.97 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3428, pruned_loss=0.09395, over 5656751.41 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:56:22,057 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-13 21:56:23,289 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2683, 1.3034, 3.1847, 2.8700], device='cuda:1'), covar=tensor([0.1484, 0.2660, 0.0521, 0.2174], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0669, 0.0996, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 21:56:50,818 INFO [train.py:968] (1/2) Epoch 27, batch 13350, libri_loss[loss=0.2586, simple_loss=0.3358, pruned_loss=0.09071, over 29537.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3396, pruned_loss=0.09135, over 5668554.32 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3403, pruned_loss=0.08889, over 5748199.23 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.341, pruned_loss=0.0923, over 5659436.10 frames. ], batch size: 89, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:57:25,118 INFO [optim.py:369] (1/2) 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 21:57:32,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 21:57:40,227 INFO [train.py:968] (1/2) Epoch 27, batch 13400, giga_loss[loss=0.2544, simple_loss=0.3323, pruned_loss=0.08832, over 28640.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3365, pruned_loss=0.08953, over 5670304.51 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.34, pruned_loss=0.08894, over 5752073.78 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3378, pruned_loss=0.09027, over 5657814.88 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:58:12,961 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1198284.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:58:14,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1198287.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 21:58:37,061 INFO [train.py:968] (1/2) Epoch 27, batch 13450, giga_loss[loss=0.2356, simple_loss=0.3152, pruned_loss=0.07802, over 27965.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3336, pruned_loss=0.08856, over 5655313.63 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3396, pruned_loss=0.08873, over 5751413.93 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3349, pruned_loss=0.08933, over 5644448.66 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 21:58:48,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1198316.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 21:59:00,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5308, 1.7322, 1.4814, 1.6030], device='cuda:1'), covar=tensor([0.0754, 0.0310, 0.0339, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 21:59:08,672 INFO [optim.py:369] (1/2) 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 21:59:19,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-13 21:59:23,613 INFO [train.py:968] (1/2) Epoch 27, batch 13500, giga_loss[loss=0.2564, simple_loss=0.3304, pruned_loss=0.09121, over 28234.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3325, pruned_loss=0.08885, over 5650596.85 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3387, pruned_loss=0.08842, over 5748865.57 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3341, pruned_loss=0.08973, over 5640048.45 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:00:23,371 INFO [train.py:968] (1/2) Epoch 27, batch 13550, giga_loss[loss=0.2811, simple_loss=0.3578, pruned_loss=0.1022, over 28947.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3343, pruned_loss=0.09041, over 5634266.24 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3387, pruned_loss=0.08849, over 5750213.16 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3356, pruned_loss=0.09106, over 5624084.65 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:00:25,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:00:52,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:00:59,377 INFO [optim.py:369] (1/2) 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 22:01:18,293 INFO [train.py:968] (1/2) Epoch 27, batch 13600, giga_loss[loss=0.2448, simple_loss=0.335, pruned_loss=0.07733, over 28902.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3364, pruned_loss=0.09024, over 5651001.72 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.338, pruned_loss=0.08828, over 5754439.18 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3378, pruned_loss=0.09098, over 5636913.31 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:02:08,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3663, 1.8484, 1.2727, 0.7538], device='cuda:1'), covar=tensor([0.5588, 0.2882, 0.4101, 0.6548], device='cuda:1'), in_proj_covar=tensor([0.1818, 0.1713, 0.1643, 0.1481], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 22:02:17,232 INFO [train.py:968] (1/2) Epoch 27, batch 13650, libri_loss[loss=0.2556, simple_loss=0.3381, pruned_loss=0.08652, over 29544.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3365, pruned_loss=0.08983, over 5649746.64 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3377, pruned_loss=0.0882, over 5757547.93 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.338, pruned_loss=0.09054, over 5633420.30 frames. ], batch size: 81, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:02:36,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:03:00,033 INFO [optim.py:369] (1/2) 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 22:03:21,335 INFO [train.py:968] (1/2) Epoch 27, batch 13700, giga_loss[loss=0.2557, simple_loss=0.3341, pruned_loss=0.08864, over 28736.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3365, pruned_loss=0.08971, over 5649815.61 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3377, pruned_loss=0.08818, over 5758713.83 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3377, pruned_loss=0.0903, over 5634890.86 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:03:23,399 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.91 vs. limit=5.0
+2023-03-13 22:03:23,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5459, 1.6410, 1.7583, 1.3510], device='cuda:1'), covar=tensor([0.2023, 0.2881, 0.1680, 0.1994], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0712, 0.0974, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 22:03:46,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1198577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:03:48,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1198580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:04:16,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4746, 1.9946, 1.5088, 1.6596], device='cuda:1'), covar=tensor([0.0780, 0.0273, 0.0339, 0.0876], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 22:04:18,924 INFO [train.py:968] (1/2) Epoch 27, batch 13750, giga_loss[loss=0.2559, simple_loss=0.3243, pruned_loss=0.09374, over 24613.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3351, pruned_loss=0.08847, over 5649165.71 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3377, pruned_loss=0.08834, over 5756218.20 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3359, pruned_loss=0.0888, over 5637310.65 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:04:24,380 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1198609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:05:02,922 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1198641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:05:18,276 INFO [train.py:968] (1/2) Epoch 27, batch 13800, libri_loss[loss=0.2456, simple_loss=0.3224, pruned_loss=0.08441, over 29522.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3342, pruned_loss=0.08669, over 5651235.18 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3378, pruned_loss=0.08843, over 5755657.56 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3348, pruned_loss=0.08685, over 5640149.32 frames. ], batch size: 80, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:06:15,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 22:06:22,538 INFO [train.py:968] (1/2) Epoch 27, batch 13850, giga_loss[loss=0.2305, simple_loss=0.3062, pruned_loss=0.07745, over 29093.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3313, pruned_loss=0.08563, over 5640238.91 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3381, pruned_loss=0.08871, over 5743649.96 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3314, pruned_loss=0.0855, over 5641227.64 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:07:01,563 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 13900, giga_loss[loss=0.2463, simple_loss=0.3262, pruned_loss=0.08318, over 28920.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3304, pruned_loss=0.08568, over 5639022.60 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3379, pruned_loss=0.0886, over 5735323.47 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3305, pruned_loss=0.0856, over 5644686.75 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:07:55,450 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:08:20,635 INFO [train.py:968] (1/2) Epoch 27, batch 13950, giga_loss[loss=0.2207, simple_loss=0.3034, pruned_loss=0.06894, over 28861.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3314, pruned_loss=0.08645, over 5644945.54 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.338, pruned_loss=0.08864, over 5727034.74 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3313, pruned_loss=0.08634, over 5655309.71 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:08:56,785 INFO [optim.py:369] (1/2) 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 22:09:15,407 INFO [train.py:968] (1/2) Epoch 27, batch 14000, giga_loss[loss=0.249, simple_loss=0.3261, pruned_loss=0.08598, over 28378.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3326, pruned_loss=0.08647, over 5656903.94 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3378, pruned_loss=0.08855, over 5731805.95 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3327, pruned_loss=0.08641, over 5658918.66 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:09:27,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1438, 1.3543, 1.2356, 0.9497], device='cuda:1'), covar=tensor([0.1070, 0.0462, 0.0959, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0444, 0.0521, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 22:10:00,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1198895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:10:11,298 INFO [train.py:968] (1/2) Epoch 27, batch 14050, giga_loss[loss=0.2188, simple_loss=0.3036, pruned_loss=0.06705, over 28020.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3344, pruned_loss=0.08692, over 5673743.57 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3368, pruned_loss=0.08822, over 5738095.42 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3352, pruned_loss=0.08713, over 5666761.38 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:10:27,427 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-13 22:10:39,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1198925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:10:42,088 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1198928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:10:55,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-13 22:10:59,170 INFO [optim.py:369] (1/2) 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 22:11:15,451 INFO [train.py:968] (1/2) Epoch 27, batch 14100, giga_loss[loss=0.2411, simple_loss=0.3177, pruned_loss=0.0823, over 29055.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3304, pruned_loss=0.08491, over 5679767.27 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3355, pruned_loss=0.0876, over 5743673.47 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3321, pruned_loss=0.08555, over 5667005.49 frames. ], batch size: 214, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:11:18,308 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1198957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:12:22,870 INFO [train.py:968] (1/2) Epoch 27, batch 14150, giga_loss[loss=0.2376, simple_loss=0.3237, pruned_loss=0.07573, over 28888.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3332, pruned_loss=0.08721, over 5669142.96 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3356, pruned_loss=0.08763, over 5734692.85 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3344, pruned_loss=0.08768, over 5665665.35 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:12:32,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-13 22:12:37,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:13:11,071 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:13:14,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:13:30,833 INFO [train.py:968] (1/2) Epoch 27, batch 14200, giga_loss[loss=0.2521, simple_loss=0.3468, pruned_loss=0.07873, over 28179.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3364, pruned_loss=0.08748, over 5659072.94 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3355, pruned_loss=0.08773, over 5737246.21 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3374, pruned_loss=0.08778, over 5653373.48 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:13:50,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:14:32,080 INFO [train.py:968] (1/2) Epoch 27, batch 14250, giga_loss[loss=0.2769, simple_loss=0.3657, pruned_loss=0.09403, over 28726.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3382, pruned_loss=0.08608, over 5651687.54 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3358, pruned_loss=0.08798, over 5730793.12 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3389, pruned_loss=0.0861, over 5650918.42 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:14:56,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4188, 1.5540, 1.6408, 1.2629], device='cuda:1'), covar=tensor([0.1935, 0.2868, 0.1669, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0713, 0.0978, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 22:15:18,372 INFO [optim.py:369] (1/2) 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 22:15:33,200 INFO [train.py:968] (1/2) Epoch 27, batch 14300, giga_loss[loss=0.2199, simple_loss=0.3245, pruned_loss=0.05759, over 28843.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3379, pruned_loss=0.08487, over 5637153.25 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3358, pruned_loss=0.08804, over 5723912.41 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3383, pruned_loss=0.08477, over 5641845.47 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:15:38,435 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:15:41,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:15:51,644 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 22:16:15,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:16:20,019 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-13 22:16:33,996 INFO [train.py:968] (1/2) Epoch 27, batch 14350, giga_loss[loss=0.2789, simple_loss=0.3566, pruned_loss=0.1006, over 28921.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3375, pruned_loss=0.08419, over 5641352.43 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3359, pruned_loss=0.08822, over 5714278.83 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3378, pruned_loss=0.08389, over 5652604.83 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:16:51,753 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:17:20,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:17:38,606 INFO [train.py:968] (1/2) Epoch 27, batch 14400, giga_loss[loss=0.2967, simple_loss=0.3685, pruned_loss=0.1124, over 27647.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3378, pruned_loss=0.08592, over 5655617.83 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3352, pruned_loss=0.08802, over 5720445.53 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3388, pruned_loss=0.08576, over 5657003.40 frames. ], batch size: 474, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:18:01,229 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:18:43,660 INFO [train.py:968] (1/2) Epoch 27, batch 14450, giga_loss[loss=0.2703, simple_loss=0.3504, pruned_loss=0.09515, over 28457.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3376, pruned_loss=0.08691, over 5655784.58 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3352, pruned_loss=0.08797, over 5722804.40 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3384, pruned_loss=0.0868, over 5654077.33 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:19:36,403 INFO [optim.py:369] (1/2) 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 22:20:00,276 INFO [train.py:968] (1/2) Epoch 27, batch 14500, giga_loss[loss=0.2241, simple_loss=0.3082, pruned_loss=0.07005, over 28228.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.337, pruned_loss=0.08677, over 5669537.67 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.335, pruned_loss=0.08791, over 5726552.54 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3379, pruned_loss=0.08672, over 5663853.27 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:20:21,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4952, 2.1801, 1.5769, 0.6790], device='cuda:1'), covar=tensor([0.7476, 0.3224, 0.5175, 0.7870], device='cuda:1'), in_proj_covar=tensor([0.1822, 0.1715, 0.1648, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 22:20:35,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3680, 3.0996, 1.3822, 1.6687], device='cuda:1'), covar=tensor([0.1008, 0.0317, 0.1011, 0.1276], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0563, 0.0405, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 22:20:40,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2734, 1.7170, 1.2834, 0.6278], device='cuda:1'), covar=tensor([0.6091, 0.2992, 0.3996, 0.6782], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1714, 0.1646, 0.1486], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 22:21:04,641 INFO [train.py:968] (1/2) Epoch 27, batch 14550, libri_loss[loss=0.2566, simple_loss=0.3351, pruned_loss=0.089, over 29464.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3322, pruned_loss=0.08418, over 5669056.35 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3347, pruned_loss=0.08792, over 5726304.29 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3332, pruned_loss=0.08397, over 5661956.51 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:21:51,356 INFO [optim.py:369] (1/2) 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 22:22:08,099 INFO [train.py:968] (1/2) Epoch 27, batch 14600, giga_loss[loss=0.1906, simple_loss=0.2794, pruned_loss=0.05084, over 28959.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3301, pruned_loss=0.08296, over 5668318.57 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3344, pruned_loss=0.08783, over 5730447.87 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.331, pruned_loss=0.08278, over 5657652.90 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:23:00,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.61 vs. limit=5.0
+2023-03-13 22:23:13,855 INFO [train.py:968] (1/2) Epoch 27, batch 14650, giga_loss[loss=0.2754, simple_loss=0.3622, pruned_loss=0.09436, over 28886.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3293, pruned_loss=0.08305, over 5678616.36 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3339, pruned_loss=0.08758, over 5733856.41 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3303, pruned_loss=0.08303, over 5666021.85 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:23:36,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-13 22:23:54,721 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 14700, giga_loss[loss=0.2506, simple_loss=0.3412, pruned_loss=0.08005, over 28421.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3344, pruned_loss=0.08564, over 5688070.76 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.334, pruned_loss=0.08766, over 5739511.88 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3351, pruned_loss=0.08545, over 5671128.59 frames. ], batch size: 369, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:24:57,311 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:25:12,449 INFO [train.py:968] (1/2) Epoch 27, batch 14750, giga_loss[loss=0.261, simple_loss=0.3285, pruned_loss=0.09676, over 28592.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3339, pruned_loss=0.08641, over 5686329.34 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3337, pruned_loss=0.08749, over 5741232.40 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3347, pruned_loss=0.0864, over 5670917.25 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:25:29,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:25:58,150 INFO [optim.py:369] (1/2) 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 22:26:09,584 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1199649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:26:16,732 INFO [train.py:968] (1/2) Epoch 27, batch 14800, giga_loss[loss=0.2701, simple_loss=0.3465, pruned_loss=0.09682, over 28565.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3352, pruned_loss=0.08842, over 5678276.01 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3342, pruned_loss=0.08777, over 5744455.52 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.3354, pruned_loss=0.08817, over 5662080.73 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:27:07,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3893, 1.7381, 1.4296, 1.5803], device='cuda:1'), covar=tensor([0.0788, 0.0318, 0.0341, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 22:27:09,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199700.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:27:15,434 INFO [train.py:968] (1/2) Epoch 27, batch 14850, giga_loss[loss=0.2243, simple_loss=0.3094, pruned_loss=0.06954, over 28608.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3343, pruned_loss=0.08799, over 5679365.78 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3342, pruned_loss=0.08779, over 5747782.96 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3346, pruned_loss=0.08777, over 5662052.71 frames. ], batch size: 65, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:27:54,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:27:59,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:03,216 INFO [optim.py:369] (1/2) 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 22:28:16,696 INFO [train.py:968] (1/2) Epoch 27, batch 14900, giga_loss[loss=0.2373, simple_loss=0.331, pruned_loss=0.0718, over 28855.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3353, pruned_loss=0.08758, over 5684414.61 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3328, pruned_loss=0.0871, over 5750789.75 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3367, pruned_loss=0.08802, over 5666017.37 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:28:26,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:30,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:36,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:28:52,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1199782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:08,783 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1199792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:09,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:12,012 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1199795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:29:23,686 INFO [train.py:968] (1/2) Epoch 27, batch 14950, giga_loss[loss=0.2373, simple_loss=0.3231, pruned_loss=0.07581, over 28925.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3365, pruned_loss=0.08774, over 5686569.01 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3325, pruned_loss=0.08695, over 5755394.96 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.338, pruned_loss=0.08829, over 5664946.29 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:29:27,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4133, 1.6492, 1.3962, 1.5903], device='cuda:1'), covar=tensor([0.0744, 0.0397, 0.0361, 0.0827], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 22:29:27,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5880, 1.7547, 1.2188, 1.3791], device='cuda:1'), covar=tensor([0.0976, 0.0539, 0.0960, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0444, 0.0520, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 22:29:33,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3256, 1.4595, 3.2593, 2.9210], device='cuda:1'), covar=tensor([0.1277, 0.2264, 0.0488, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0794, 0.0666, 0.0984, 0.0955], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 22:29:49,464 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1199824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:30:12,781 INFO [optim.py:369] (1/2) 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 22:30:29,810 INFO [train.py:968] (1/2) Epoch 27, batch 15000, libri_loss[loss=0.2479, simple_loss=0.3277, pruned_loss=0.08407, over 29589.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3338, pruned_loss=0.08652, over 5683040.42 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3319, pruned_loss=0.087, over 5752313.58 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3359, pruned_loss=0.08694, over 5663541.57 frames. ], batch size: 74, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:30:29,811 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 22:30:38,196 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 22:31:42,046 INFO [train.py:968] (1/2) Epoch 27, batch 15050, giga_loss[loss=0.2547, simple_loss=0.3288, pruned_loss=0.09032, over 29090.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3291, pruned_loss=0.08477, over 5688829.21 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3317, pruned_loss=0.08695, over 5745353.71 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3309, pruned_loss=0.08513, over 5678026.73 frames. ], batch size: 113, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 22:32:32,475 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 15100, libri_loss[loss=0.2546, simple_loss=0.3376, pruned_loss=0.0858, over 29247.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3244, pruned_loss=0.0827, over 5688498.80 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08699, over 5749470.15 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3256, pruned_loss=0.08285, over 5674500.64 frames. ], batch size: 94, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 22:32:44,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4512, 1.7790, 1.4505, 1.3242], device='cuda:1'), covar=tensor([0.2520, 0.2477, 0.2901, 0.2379], device='cuda:1'), in_proj_covar=tensor([0.1587, 0.1141, 0.1404, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 22:33:41,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5583, 3.6502, 1.6983, 1.6517], device='cuda:1'), covar=tensor([0.0920, 0.0341, 0.0903, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0563, 0.0405, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 22:33:46,149 INFO [train.py:968] (1/2) Epoch 27, batch 15150, giga_loss[loss=0.2359, simple_loss=0.3251, pruned_loss=0.07335, over 28957.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3267, pruned_loss=0.08448, over 5685784.01 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08703, over 5750839.64 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3274, pruned_loss=0.08454, over 5673044.08 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 22:34:09,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-13 22:34:28,205 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 15200, giga_loss[loss=0.2464, simple_loss=0.3283, pruned_loss=0.08221, over 28515.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3276, pruned_loss=0.08515, over 5680046.04 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3322, pruned_loss=0.08718, over 5750629.85 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3279, pruned_loss=0.08503, over 5668995.87 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:35:03,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200075.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:35:15,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5143, 1.3493, 4.6360, 3.5459], device='cuda:1'), covar=tensor([0.1720, 0.2901, 0.0417, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0666, 0.0984, 0.0954], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 22:35:37,075 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1200103.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:35:40,888 INFO [train.py:968] (1/2) Epoch 27, batch 15250, giga_loss[loss=0.2228, simple_loss=0.3139, pruned_loss=0.06587, over 28904.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3249, pruned_loss=0.08305, over 5673987.36 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3319, pruned_loss=0.087, over 5754335.21 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3253, pruned_loss=0.08304, over 5660401.41 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:36:23,658 INFO [optim.py:369] (1/2) 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 22:36:27,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6321, 1.9656, 1.3686, 1.5379], device='cuda:1'), covar=tensor([0.1061, 0.0543, 0.0899, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0445, 0.0522, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 22:36:36,433 INFO [train.py:968] (1/2) Epoch 27, batch 15300, giga_loss[loss=0.2325, simple_loss=0.3126, pruned_loss=0.07617, over 29030.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3244, pruned_loss=0.08208, over 5679220.09 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08703, over 5759689.79 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3245, pruned_loss=0.08187, over 5660711.35 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:36:40,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:37:45,312 INFO [train.py:968] (1/2) Epoch 27, batch 15350, giga_loss[loss=0.248, simple_loss=0.3351, pruned_loss=0.08045, over 28434.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3239, pruned_loss=0.08257, over 5676107.24 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.332, pruned_loss=0.08723, over 5759492.78 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3237, pruned_loss=0.08217, over 5660487.65 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:38:01,408 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200218.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:38:05,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1200221.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:38:06,595 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3282, 4.1951, 3.9512, 2.0948], device='cuda:1'), covar=tensor([0.0579, 0.0705, 0.0731, 0.1996], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1172, 0.0988, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 22:38:24,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-13 22:38:31,289 INFO [optim.py:369] (1/2) 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 22:38:40,691 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200250.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:38:42,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6484, 4.5009, 4.2682, 1.9629], device='cuda:1'), covar=tensor([0.0565, 0.0695, 0.0773, 0.1980], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1172, 0.0988, 0.0732], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 22:38:45,006 INFO [train.py:968] (1/2) Epoch 27, batch 15400, giga_loss[loss=0.2484, simple_loss=0.3319, pruned_loss=0.08243, over 28732.00 frames. ], tot_loss[loss=0.245, simple_loss=0.325, pruned_loss=0.08252, over 5692927.15 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3315, pruned_loss=0.08692, over 5763214.89 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3251, pruned_loss=0.0823, over 5673950.36 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:39:43,406 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:39:46,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1200303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:39:48,008 INFO [train.py:968] (1/2) Epoch 27, batch 15450, giga_loss[loss=0.2726, simple_loss=0.3486, pruned_loss=0.09831, over 28672.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3256, pruned_loss=0.08279, over 5698328.92 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3314, pruned_loss=0.08696, over 5764470.08 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3256, pruned_loss=0.08255, over 5681780.58 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:40:21,793 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:40:36,317 INFO [optim.py:369] (1/2) 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 22:40:48,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8313, 3.7156, 3.4869, 1.6850], device='cuda:1'), covar=tensor([0.0735, 0.0769, 0.0789, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.1273, 0.1171, 0.0989, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 22:40:51,148 INFO [train.py:968] (1/2) Epoch 27, batch 15500, giga_loss[loss=0.2378, simple_loss=0.3045, pruned_loss=0.08553, over 26912.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3268, pruned_loss=0.08432, over 5697722.02 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3315, pruned_loss=0.08706, over 5768052.88 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3266, pruned_loss=0.08395, over 5679998.28 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:41:15,444 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 15550, giga_loss[loss=0.2333, simple_loss=0.3188, pruned_loss=0.0739, over 28644.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3257, pruned_loss=0.08336, over 5690823.11 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3312, pruned_loss=0.08705, over 5770758.67 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3257, pruned_loss=0.08302, over 5673000.21 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:42:33,639 INFO [optim.py:369] (1/2) 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 22:42:47,836 INFO [train.py:968] (1/2) Epoch 27, batch 15600, giga_loss[loss=0.2253, simple_loss=0.3263, pruned_loss=0.06209, over 28850.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3273, pruned_loss=0.08288, over 5677306.35 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3314, pruned_loss=0.08715, over 5773806.10 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3271, pruned_loss=0.08241, over 5658134.13 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:43:06,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4684, 1.8576, 1.7645, 1.6241], device='cuda:1'), covar=tensor([0.2179, 0.1995, 0.2072, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0739, 0.0708, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 22:43:15,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200478.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:43:46,830 INFO [train.py:968] (1/2) Epoch 27, batch 15650, giga_loss[loss=0.2438, simple_loss=0.3307, pruned_loss=0.07842, over 28934.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3305, pruned_loss=0.08422, over 5677915.59 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3309, pruned_loss=0.0869, over 5776032.70 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3307, pruned_loss=0.08404, over 5658765.95 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:44:28,661 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-13 22:44:32,092 INFO [optim.py:369] (1/2) 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 22:44:42,305 INFO [train.py:968] (1/2) Epoch 27, batch 15700, giga_loss[loss=0.2948, simple_loss=0.3547, pruned_loss=0.1175, over 26801.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.331, pruned_loss=0.08444, over 5669681.67 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3306, pruned_loss=0.08655, over 5772913.91 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3315, pruned_loss=0.08451, over 5653074.08 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:45:37,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4745, 3.7031, 1.6858, 1.5501], device='cuda:1'), covar=tensor([0.1010, 0.0459, 0.0941, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0563, 0.0405, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 22:45:39,920 INFO [train.py:968] (1/2) Epoch 27, batch 15750, giga_loss[loss=0.2508, simple_loss=0.3346, pruned_loss=0.08348, over 28319.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3305, pruned_loss=0.08457, over 5666355.71 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3299, pruned_loss=0.08625, over 5776248.02 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3315, pruned_loss=0.08483, over 5646212.10 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:45:56,865 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200621.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 22:45:59,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1200624.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:46:22,930 INFO [optim.py:369] (1/2) 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 22:46:29,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6267, 1.6156, 1.8490, 1.4415], device='cuda:1'), covar=tensor([0.2064, 0.2723, 0.1696, 0.2072], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0708, 0.0975, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-13 22:46:35,004 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200653.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 22:46:36,386 INFO [train.py:968] (1/2) Epoch 27, batch 15800, giga_loss[loss=0.2073, simple_loss=0.2916, pruned_loss=0.06151, over 27680.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3277, pruned_loss=0.0831, over 5664891.36 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3298, pruned_loss=0.08615, over 5777136.86 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3287, pruned_loss=0.08338, over 5646209.43 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:47:17,483 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-13 22:47:34,504 INFO [train.py:968] (1/2) Epoch 27, batch 15850, giga_loss[loss=0.2405, simple_loss=0.3284, pruned_loss=0.07632, over 28985.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3269, pruned_loss=0.08275, over 5668747.30 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.3299, pruned_loss=0.08636, over 5778024.74 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3275, pruned_loss=0.08265, over 5649134.35 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:47:56,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-13 22:48:15,790 INFO [optim.py:369] (1/2) 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 22:48:19,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1200749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:48:26,920 INFO [train.py:968] (1/2) Epoch 27, batch 15900, giga_loss[loss=0.2323, simple_loss=0.3129, pruned_loss=0.07581, over 28561.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3255, pruned_loss=0.08256, over 5675640.68 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3293, pruned_loss=0.08608, over 5776695.02 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3264, pruned_loss=0.08258, over 5657270.58 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:49:09,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2389, 3.1130, 1.3500, 1.4274], device='cuda:1'), covar=tensor([0.1091, 0.0314, 0.1006, 0.1515], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0563, 0.0405, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 22:49:10,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4118, 1.4420, 1.3136, 1.5836], device='cuda:1'), covar=tensor([0.0761, 0.0359, 0.0345, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-13 22:49:24,781 INFO [train.py:968] (1/2) Epoch 27, batch 15950, giga_loss[loss=0.2527, simple_loss=0.3398, pruned_loss=0.08285, over 28631.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3274, pruned_loss=0.08327, over 5683607.56 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3293, pruned_loss=0.08603, over 5780164.75 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.328, pruned_loss=0.08324, over 5662340.73 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:49:29,220 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2077, 1.5758, 0.9722, 1.0953], device='cuda:1'), covar=tensor([0.1345, 0.0653, 0.1574, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0446, 0.0523, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 22:50:12,212 INFO [optim.py:369] (1/2) 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 22:50:23,839 INFO [train.py:968] (1/2) Epoch 27, batch 16000, giga_loss[loss=0.2464, simple_loss=0.3285, pruned_loss=0.08221, over 28922.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3296, pruned_loss=0.08429, over 5673977.12 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3299, pruned_loss=0.0864, over 5772485.09 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3294, pruned_loss=0.08383, over 5661252.02 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:51:13,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1200892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:51:16,458 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 16050, giga_loss[loss=0.2806, simple_loss=0.3561, pruned_loss=0.1026, over 28913.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3312, pruned_loss=0.08619, over 5665357.70 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3296, pruned_loss=0.08628, over 5772126.52 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3313, pruned_loss=0.0859, over 5653138.82 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:51:48,372 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1200924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:52:11,006 INFO [optim.py:369] (1/2) 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 22:52:15,869 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1200950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 22:52:20,392 INFO [train.py:968] (1/2) Epoch 27, batch 16100, giga_loss[loss=0.2895, simple_loss=0.3709, pruned_loss=0.1041, over 28815.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3333, pruned_loss=0.08719, over 5667271.04 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3288, pruned_loss=0.0858, over 5776386.79 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3342, pruned_loss=0.08743, over 5649335.33 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:53:17,401 INFO [train.py:968] (1/2) Epoch 27, batch 16150, giga_loss[loss=0.2919, simple_loss=0.3725, pruned_loss=0.1056, over 28609.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3351, pruned_loss=0.08722, over 5656353.36 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3288, pruned_loss=0.08583, over 5767745.84 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3359, pruned_loss=0.0874, over 5648085.84 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:54:08,476 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 16200, giga_loss[loss=0.2665, simple_loss=0.3495, pruned_loss=0.09175, over 28816.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3352, pruned_loss=0.08704, over 5657684.77 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3283, pruned_loss=0.08553, over 5771514.37 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3366, pruned_loss=0.08751, over 5644556.30 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:55:30,437 INFO [train.py:968] (1/2) Epoch 27, batch 16250, giga_loss[loss=0.2391, simple_loss=0.319, pruned_loss=0.07962, over 28583.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3338, pruned_loss=0.08653, over 5660095.12 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3281, pruned_loss=0.08543, over 5771919.92 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3351, pruned_loss=0.087, over 5647925.65 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:55:46,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4306, 2.0624, 1.4613, 0.6838], device='cuda:1'), covar=tensor([0.6620, 0.3173, 0.4793, 0.7096], device='cuda:1'), in_proj_covar=tensor([0.1822, 0.1718, 0.1644, 0.1489], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 22:56:09,185 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-13 22:56:19,571 INFO [optim.py:369] (1/2) 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 22:56:31,316 INFO [train.py:968] (1/2) Epoch 27, batch 16300, giga_loss[loss=0.2721, simple_loss=0.3426, pruned_loss=0.1008, over 27829.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3309, pruned_loss=0.08493, over 5667992.17 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3277, pruned_loss=0.08517, over 5774226.32 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3324, pruned_loss=0.08557, over 5653884.07 frames. ], batch size: 476, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:56:42,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-13 22:56:56,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1407, 1.2654, 3.4444, 3.0397], device='cuda:1'), covar=tensor([0.1661, 0.2676, 0.0546, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0670, 0.0988, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 22:56:58,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2117, 1.6426, 1.5728, 1.3672], device='cuda:1'), covar=tensor([0.2148, 0.1709, 0.2129, 0.1926], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0737, 0.0707, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 22:57:31,336 INFO [train.py:968] (1/2) Epoch 27, batch 16350, giga_loss[loss=0.2451, simple_loss=0.3254, pruned_loss=0.0824, over 28946.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3308, pruned_loss=0.08517, over 5664351.72 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3278, pruned_loss=0.08529, over 5765011.35 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.332, pruned_loss=0.08556, over 5658514.54 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:57:59,730 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-13 22:58:22,734 INFO [optim.py:369] (1/2) 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 22:58:31,353 INFO [train.py:968] (1/2) Epoch 27, batch 16400, giga_loss[loss=0.2272, simple_loss=0.2924, pruned_loss=0.08101, over 24450.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3294, pruned_loss=0.08558, over 5656731.62 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3273, pruned_loss=0.08499, over 5768653.41 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3309, pruned_loss=0.0862, over 5645188.65 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 22:59:30,901 INFO [train.py:968] (1/2) Epoch 27, batch 16450, giga_loss[loss=0.2237, simple_loss=0.3121, pruned_loss=0.06763, over 28931.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3275, pruned_loss=0.08453, over 5643628.86 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3273, pruned_loss=0.08492, over 5750137.25 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3287, pruned_loss=0.08509, over 5648393.06 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 22:59:56,716 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:00:28,135 INFO [train.py:968] (1/2) Epoch 27, batch 16500, giga_loss[loss=0.2243, simple_loss=0.3143, pruned_loss=0.06709, over 28926.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3267, pruned_loss=0.08309, over 5658403.67 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3269, pruned_loss=0.08462, over 5753599.92 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.328, pruned_loss=0.08379, over 5656293.84 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:01:25,000 INFO [train.py:968] (1/2) Epoch 27, batch 16550, giga_loss[loss=0.2099, simple_loss=0.3146, pruned_loss=0.05259, over 29027.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.3254, pruned_loss=0.08088, over 5672284.37 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3265, pruned_loss=0.08442, over 5758488.79 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3269, pruned_loss=0.08154, over 5664013.91 frames. ], batch size: 165, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:02:13,388 INFO [optim.py:369] (1/2) 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 23:02:23,890 INFO [train.py:968] (1/2) Epoch 27, batch 16600, giga_loss[loss=0.2478, simple_loss=0.3356, pruned_loss=0.08001, over 28673.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3288, pruned_loss=0.08073, over 5681141.87 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3265, pruned_loss=0.08446, over 5758886.35 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3298, pruned_loss=0.0812, over 5673987.06 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:02:27,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3332, 4.1932, 3.9498, 2.0930], device='cuda:1'), covar=tensor([0.0577, 0.0694, 0.0769, 0.1997], device='cuda:1'), in_proj_covar=tensor([0.1273, 0.1169, 0.0989, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 23:02:35,251 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1201468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:02:40,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1201471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:02:54,332 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:03:01,227 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-13 23:03:03,692 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9184, 1.0965, 1.1535, 0.9362], device='cuda:1'), covar=tensor([0.2269, 0.2429, 0.1457, 0.2026], device='cuda:1'), in_proj_covar=tensor([0.2011, 0.1954, 0.1859, 0.2001], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 23:03:13,191 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1201500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:03:16,820 INFO [train.py:968] (1/2) Epoch 27, batch 16650, giga_loss[loss=0.2732, simple_loss=0.3559, pruned_loss=0.09531, over 28516.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3294, pruned_loss=0.08101, over 5677662.97 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3261, pruned_loss=0.08421, over 5761636.69 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3307, pruned_loss=0.08151, over 5667363.71 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:04:08,707 INFO [optim.py:369] (1/2) 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 23:04:17,498 INFO [train.py:968] (1/2) Epoch 27, batch 16700, giga_loss[loss=0.2618, simple_loss=0.3229, pruned_loss=0.1003, over 24329.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3285, pruned_loss=0.08067, over 5668712.17 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3262, pruned_loss=0.08422, over 5763183.83 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3296, pruned_loss=0.08099, over 5657160.57 frames. ], batch size: 705, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:04:31,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7407, 2.1634, 2.1009, 1.6645], device='cuda:1'), covar=tensor([0.3405, 0.2240, 0.2467, 0.2941], device='cuda:1'), in_proj_covar=tensor([0.2014, 0.1956, 0.1863, 0.2003], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 23:05:10,281 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3126, 1.4110, 1.4262, 1.3291], device='cuda:1'), covar=tensor([0.2421, 0.2207, 0.1652, 0.2096], device='cuda:1'), in_proj_covar=tensor([0.2009, 0.1952, 0.1858, 0.1999], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 23:05:22,312 INFO [train.py:968] (1/2) Epoch 27, batch 16750, giga_loss[loss=0.2315, simple_loss=0.3159, pruned_loss=0.07357, over 27660.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.3287, pruned_loss=0.08127, over 5660756.41 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3256, pruned_loss=0.08396, over 5764667.73 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3301, pruned_loss=0.08169, over 5648203.00 frames. ], batch size: 474, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:06:21,408 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 16800, giga_loss[loss=0.2409, simple_loss=0.3336, pruned_loss=0.07405, over 28686.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.3287, pruned_loss=0.08055, over 5665156.24 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3256, pruned_loss=0.08391, over 5765857.81 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3299, pruned_loss=0.08089, over 5652583.64 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:06:52,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:07:05,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9129, 2.1434, 1.4972, 1.8584], device='cuda:1'), covar=tensor([0.1115, 0.0736, 0.1109, 0.1198], device='cuda:1'), in_proj_covar=tensor([0.0405, 0.0442, 0.0520, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 23:07:40,346 INFO [train.py:968] (1/2) Epoch 27, batch 16850, giga_loss[loss=0.295, simple_loss=0.3799, pruned_loss=0.1051, over 29051.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3298, pruned_loss=0.08092, over 5657387.96 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3255, pruned_loss=0.08389, over 5760268.67 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3309, pruned_loss=0.08113, over 5650056.96 frames. ], batch size: 285, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:08:08,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3549, 1.5386, 1.1752, 1.1059], device='cuda:1'), covar=tensor([0.1044, 0.0525, 0.1050, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0442, 0.0520, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 23:08:38,776 INFO [optim.py:369] (1/2) 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 23:08:45,057 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.30 vs. limit=5.0
+2023-03-13 23:08:49,209 INFO [train.py:968] (1/2) Epoch 27, batch 16900, giga_loss[loss=0.2922, simple_loss=0.3834, pruned_loss=0.1005, over 28417.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3337, pruned_loss=0.08254, over 5666678.29 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3255, pruned_loss=0.08391, over 5762044.80 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3347, pruned_loss=0.08267, over 5657879.00 frames. ], batch size: 369, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:09:54,694 INFO [train.py:968] (1/2) Epoch 27, batch 16950, giga_loss[loss=0.2278, simple_loss=0.3219, pruned_loss=0.06687, over 28437.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3334, pruned_loss=0.08238, over 5673721.68 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3255, pruned_loss=0.08389, over 5762713.24 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3345, pruned_loss=0.08247, over 5663484.83 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:10:19,761 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-13 23:10:46,429 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201842.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 23:10:52,361 INFO [optim.py:369] (1/2) 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 23:11:01,861 INFO [train.py:968] (1/2) Epoch 27, batch 17000, giga_loss[loss=0.1823, simple_loss=0.2742, pruned_loss=0.04522, over 28571.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3326, pruned_loss=0.08295, over 5667407.90 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3251, pruned_loss=0.08365, over 5756582.75 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.334, pruned_loss=0.08323, over 5661885.48 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:11:02,102 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1716, 1.3065, 3.4382, 3.0972], device='cuda:1'), covar=tensor([0.1917, 0.2886, 0.0919, 0.1144], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0670, 0.0987, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 23:11:06,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1201860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:12:01,191 INFO [train.py:968] (1/2) Epoch 27, batch 17050, giga_loss[loss=0.2482, simple_loss=0.3373, pruned_loss=0.07952, over 28615.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3304, pruned_loss=0.08174, over 5677592.10 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3247, pruned_loss=0.08337, over 5759165.29 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3321, pruned_loss=0.08217, over 5667209.17 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:13:01,002 INFO [optim.py:369] (1/2) 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 23:13:10,136 INFO [train.py:968] (1/2) Epoch 27, batch 17100, giga_loss[loss=0.2355, simple_loss=0.3193, pruned_loss=0.07583, over 29245.00 frames. ], tot_loss[loss=0.244, simple_loss=0.3281, pruned_loss=0.07998, over 5676206.94 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3243, pruned_loss=0.08307, over 5763268.87 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.33, pruned_loss=0.08052, over 5661883.57 frames. ], batch size: 113, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:13:44,229 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1201985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:05,323 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 17150, giga_loss[loss=0.2241, simple_loss=0.3107, pruned_loss=0.06872, over 28975.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3285, pruned_loss=0.08039, over 5679525.69 frames. ], libri_tot_loss[loss=0.2452, simple_loss=0.3243, pruned_loss=0.08306, over 5765971.37 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3301, pruned_loss=0.08078, over 5663838.82 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:14:09,285 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:29,495 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:49,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:14:51,288 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 17200, giga_loss[loss=0.2643, simple_loss=0.3514, pruned_loss=0.08862, over 28405.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3299, pruned_loss=0.08152, over 5675002.01 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.3236, pruned_loss=0.08278, over 5761703.94 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3321, pruned_loss=0.08201, over 5661569.34 frames. ], batch size: 369, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:15:08,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2071, 4.2209, 1.4430, 1.5198], device='cuda:1'), covar=tensor([0.1256, 0.0373, 0.1072, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0561, 0.0405, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 23:15:37,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4578, 5.3063, 5.0404, 2.4218], device='cuda:1'), covar=tensor([0.0439, 0.0562, 0.0715, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.1266, 0.1161, 0.0983, 0.0729], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 23:15:54,796 INFO [train.py:968] (1/2) Epoch 27, batch 17250, giga_loss[loss=0.2504, simple_loss=0.3324, pruned_loss=0.08419, over 28899.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3312, pruned_loss=0.08246, over 5668422.54 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.3236, pruned_loss=0.08283, over 5755092.75 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3331, pruned_loss=0.0828, over 5662296.41 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:16:41,326 INFO [optim.py:369] (1/2) 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 23:16:49,134 INFO [train.py:968] (1/2) Epoch 27, batch 17300, giga_loss[loss=0.236, simple_loss=0.3121, pruned_loss=0.07997, over 28892.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3283, pruned_loss=0.08203, over 5660739.89 frames. ], libri_tot_loss[loss=0.2445, simple_loss=0.3235, pruned_loss=0.08271, over 5749171.87 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.33, pruned_loss=0.0824, over 5659030.67 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:17:03,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6281, 1.9511, 1.5586, 1.7562], device='cuda:1'), covar=tensor([0.2808, 0.2597, 0.3098, 0.2399], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1135, 0.1400, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 23:17:28,356 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:17:29,612 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:17:31,693 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 17350, giga_loss[loss=0.2311, simple_loss=0.3139, pruned_loss=0.0742, over 28645.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3278, pruned_loss=0.08241, over 5659013.81 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3234, pruned_loss=0.08262, over 5747653.91 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3295, pruned_loss=0.08277, over 5656957.16 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:18:01,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202217.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 23:18:04,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:18:36,754 INFO [optim.py:369] (1/2) 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 23:18:43,559 INFO [train.py:968] (1/2) Epoch 27, batch 17400, giga_loss[loss=0.2269, simple_loss=0.3111, pruned_loss=0.07131, over 28443.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3296, pruned_loss=0.08385, over 5652192.00 frames. ], libri_tot_loss[loss=0.2442, simple_loss=0.3232, pruned_loss=0.08256, over 5748903.23 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3311, pruned_loss=0.0842, over 5648595.09 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:19:29,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6199, 1.8754, 1.5566, 1.6427], device='cuda:1'), covar=tensor([0.2800, 0.2887, 0.3315, 0.2520], device='cuda:1'), in_proj_covar=tensor([0.1577, 0.1134, 0.1397, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 23:19:35,508 INFO [train.py:968] (1/2) Epoch 27, batch 17450, giga_loss[loss=0.2879, simple_loss=0.3687, pruned_loss=0.1035, over 28900.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3398, pruned_loss=0.08944, over 5663442.52 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3233, pruned_loss=0.08261, over 5750433.23 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.341, pruned_loss=0.08971, over 5658498.15 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:20:15,634 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 17500, giga_loss[loss=0.2413, simple_loss=0.3272, pruned_loss=0.07772, over 28848.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3463, pruned_loss=0.09316, over 5668701.84 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3233, pruned_loss=0.08259, over 5751779.27 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3474, pruned_loss=0.0935, over 5662748.43 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:20:24,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:20:24,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202360.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 23:20:25,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4733, 2.0235, 1.4999, 1.5428], device='cuda:1'), covar=tensor([0.0802, 0.0291, 0.0337, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-13 23:20:27,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202363.0, num_to_drop=1, layers_to_drop={1}
+2023-03-13 23:20:51,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3557, 1.7023, 1.6231, 1.5347], device='cuda:1'), covar=tensor([0.2256, 0.1996, 0.2487, 0.1997], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0744, 0.0715, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 23:20:52,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202392.0, num_to_drop=1, layers_to_drop={0}
+2023-03-13 23:20:52,529 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-13 23:21:02,309 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:21:02,946 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1725, 4.0045, 3.7904, 1.8600], device='cuda:1'), covar=tensor([0.0625, 0.0782, 0.0767, 0.2147], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1163, 0.0985, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 23:21:04,665 INFO [train.py:968] (1/2) Epoch 27, batch 17550, giga_loss[loss=0.2215, simple_loss=0.3122, pruned_loss=0.06539, over 28839.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3444, pruned_loss=0.09327, over 5672392.62 frames. ], libri_tot_loss[loss=0.2438, simple_loss=0.3228, pruned_loss=0.08238, over 5755184.01 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3464, pruned_loss=0.09402, over 5662388.31 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:21:29,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4013, 2.4891, 2.3418, 2.2635], device='cuda:1'), covar=tensor([0.2097, 0.2340, 0.2339, 0.2247], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0744, 0.0715, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 23:21:43,881 INFO [optim.py:369] (1/2) 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 23:21:49,344 INFO [train.py:968] (1/2) Epoch 27, batch 17600, giga_loss[loss=0.3191, simple_loss=0.3622, pruned_loss=0.138, over 26583.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3376, pruned_loss=0.09029, over 5681361.86 frames. ], libri_tot_loss[loss=0.2438, simple_loss=0.3228, pruned_loss=0.0824, over 5757204.59 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.3394, pruned_loss=0.09098, over 5670583.13 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:22:27,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3858, 1.8039, 1.5686, 1.5688], device='cuda:1'), covar=tensor([0.0748, 0.0386, 0.0333, 0.0838], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 23:22:29,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:22:30,146 INFO [train.py:968] (1/2) Epoch 27, batch 17650, libri_loss[loss=0.2268, simple_loss=0.307, pruned_loss=0.07326, over 29541.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3311, pruned_loss=0.08718, over 5692671.82 frames. ], libri_tot_loss[loss=0.2439, simple_loss=0.3231, pruned_loss=0.08234, over 5758863.06 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3327, pruned_loss=0.08805, over 5679125.11 frames. ], batch size: 79, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:22:31,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:22:55,823 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:08,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:09,630 INFO [optim.py:369] (1/2) 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 23:23:14,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7341, 1.8530, 1.8731, 1.6386], device='cuda:1'), covar=tensor([0.2230, 0.2433, 0.2537, 0.2458], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0745, 0.0716, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 23:23:14,739 INFO [train.py:968] (1/2) Epoch 27, batch 17700, giga_loss[loss=0.2188, simple_loss=0.2919, pruned_loss=0.07285, over 28655.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.324, pruned_loss=0.08431, over 5696563.28 frames. ], libri_tot_loss[loss=0.2444, simple_loss=0.3237, pruned_loss=0.08254, over 5759000.15 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3249, pruned_loss=0.08488, over 5684008.86 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:23:23,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1202564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:34,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:23:56,330 INFO [train.py:968] (1/2) Epoch 27, batch 17750, giga_loss[loss=0.2008, simple_loss=0.2819, pruned_loss=0.05986, over 28937.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.3165, pruned_loss=0.08096, over 5700130.89 frames. ], libri_tot_loss[loss=0.2444, simple_loss=0.3238, pruned_loss=0.08249, over 5761824.64 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.317, pruned_loss=0.08145, over 5686505.88 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:23:56,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4450, 2.9385, 1.5545, 1.5353], device='cuda:1'), covar=tensor([0.0993, 0.0367, 0.0897, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0561, 0.0403, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 23:24:13,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-13 23:24:31,464 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 17800, giga_loss[loss=0.2351, simple_loss=0.3036, pruned_loss=0.0833, over 28751.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.3128, pruned_loss=0.0798, over 5696940.39 frames. ], libri_tot_loss[loss=0.2453, simple_loss=0.3246, pruned_loss=0.08302, over 5761741.34 frames. ], giga_tot_loss[loss=0.2358, simple_loss=0.3123, pruned_loss=0.07969, over 5684684.46 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:24:47,718 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 17850, libri_loss[loss=0.274, simple_loss=0.3546, pruned_loss=0.09667, over 29139.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.31, pruned_loss=0.07829, over 5694934.80 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3256, pruned_loss=0.08342, over 5753624.96 frames. ], giga_tot_loss[loss=0.2319, simple_loss=0.3083, pruned_loss=0.07772, over 5690030.97 frames. ], batch size: 101, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:25:17,155 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1202707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:19,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1202710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:24,226 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:32,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:45,811 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1202739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:25:55,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1181, 1.1575, 3.6182, 3.0876], device='cuda:1'), covar=tensor([0.2119, 0.3247, 0.0811, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0666, 0.0987, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 23:25:55,715 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 17900, giga_loss[loss=0.2052, simple_loss=0.2843, pruned_loss=0.06307, over 28999.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3071, pruned_loss=0.07721, over 5689354.36 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3256, pruned_loss=0.08334, over 5750738.91 frames. ], giga_tot_loss[loss=0.2295, simple_loss=0.3055, pruned_loss=0.07675, over 5687365.14 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:26:39,985 INFO [train.py:968] (1/2) Epoch 27, batch 17950, giga_loss[loss=0.1952, simple_loss=0.2742, pruned_loss=0.05814, over 29072.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3055, pruned_loss=0.0765, over 5687737.84 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3262, pruned_loss=0.08357, over 5753364.18 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.3029, pruned_loss=0.07565, over 5680861.05 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:26:41,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 23:26:47,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3975, 4.2349, 4.0018, 1.9006], device='cuda:1'), covar=tensor([0.0530, 0.0727, 0.0746, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.1278, 0.1173, 0.0992, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-13 23:27:06,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3659, 3.0401, 1.5096, 1.5584], device='cuda:1'), covar=tensor([0.1018, 0.0344, 0.0919, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0563, 0.0404, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 23:27:14,628 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:27:16,361 INFO [optim.py:369] (1/2) 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 23:27:23,565 INFO [train.py:968] (1/2) Epoch 27, batch 18000, giga_loss[loss=0.2098, simple_loss=0.2839, pruned_loss=0.06788, over 28838.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3024, pruned_loss=0.07513, over 5700124.08 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3262, pruned_loss=0.08359, over 5755509.25 frames. ], giga_tot_loss[loss=0.2243, simple_loss=0.2999, pruned_loss=0.07433, over 5692125.53 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:27:23,565 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-13 23:27:29,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1951, 1.2667, 3.3869, 3.0807], device='cuda:1'), covar=tensor([0.1854, 0.3072, 0.0562, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0666, 0.0987, 0.0958], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 23:27:31,990 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-13 23:27:47,705 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1202874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:27:54,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4902, 2.1088, 1.6343, 0.8145], device='cuda:1'), covar=tensor([0.6762, 0.3912, 0.5285, 0.7277], device='cuda:1'), in_proj_covar=tensor([0.1820, 0.1720, 0.1646, 0.1488], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 23:28:14,541 INFO [train.py:968] (1/2) Epoch 27, batch 18050, giga_loss[loss=0.2085, simple_loss=0.2866, pruned_loss=0.06522, over 28717.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2992, pruned_loss=0.07375, over 5690027.44 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3265, pruned_loss=0.08367, over 5757134.43 frames. ], giga_tot_loss[loss=0.2214, simple_loss=0.2968, pruned_loss=0.07297, over 5681820.52 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:28:52,695 INFO [optim.py:369] (1/2) 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 23:28:55,897 INFO [train.py:968] (1/2) Epoch 27, batch 18100, giga_loss[loss=0.2045, simple_loss=0.2859, pruned_loss=0.06154, over 28991.00 frames. ], tot_loss[loss=0.2226, simple_loss=0.2979, pruned_loss=0.07368, over 5696717.60 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3273, pruned_loss=0.08402, over 5760654.11 frames. ], giga_tot_loss[loss=0.2196, simple_loss=0.2944, pruned_loss=0.0724, over 5684836.03 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:29:40,228 INFO [train.py:968] (1/2) Epoch 27, batch 18150, giga_loss[loss=0.1861, simple_loss=0.2627, pruned_loss=0.0547, over 28606.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2948, pruned_loss=0.07211, over 5704755.27 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3272, pruned_loss=0.0839, over 5761884.31 frames. ], giga_tot_loss[loss=0.2169, simple_loss=0.2917, pruned_loss=0.07106, over 5693561.49 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:30:17,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:30:21,103 INFO [optim.py:369] (1/2) 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 23:30:25,595 INFO [train.py:968] (1/2) Epoch 27, batch 18200, giga_loss[loss=0.2365, simple_loss=0.3135, pruned_loss=0.07973, over 28587.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2913, pruned_loss=0.07074, over 5703321.18 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.3268, pruned_loss=0.08372, over 5763311.12 frames. ], giga_tot_loss[loss=0.2144, simple_loss=0.2888, pruned_loss=0.06997, over 5692718.01 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:31:00,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:31:07,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:31:12,408 INFO [train.py:968] (1/2) Epoch 27, batch 18250, giga_loss[loss=0.3069, simple_loss=0.3671, pruned_loss=0.1233, over 26674.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2997, pruned_loss=0.07499, over 5697780.59 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3276, pruned_loss=0.08405, over 5757646.41 frames. ], giga_tot_loss[loss=0.2217, simple_loss=0.296, pruned_loss=0.07372, over 5692159.63 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:31:49,066 INFO [optim.py:369] (1/2) 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 23:31:52,789 INFO [train.py:968] (1/2) Epoch 27, batch 18300, giga_loss[loss=0.2655, simple_loss=0.3523, pruned_loss=0.0893, over 28816.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.3125, pruned_loss=0.08112, over 5701288.47 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3275, pruned_loss=0.08398, over 5759614.53 frames. ], giga_tot_loss[loss=0.2347, simple_loss=0.3092, pruned_loss=0.08007, over 5693648.04 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:32:09,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-13 23:32:16,682 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:18,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:30,878 INFO [train.py:968] (1/2) Epoch 27, batch 18350, giga_loss[loss=0.3233, simple_loss=0.3934, pruned_loss=0.1266, over 27937.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3243, pruned_loss=0.08725, over 5704372.94 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3277, pruned_loss=0.08401, over 5763072.14 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3213, pruned_loss=0.08639, over 5693944.64 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:32:41,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:45,867 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:51,471 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-13 23:32:57,023 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:32:58,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:02,896 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:04,763 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:07,586 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:08,602 INFO [optim.py:369] (1/2) 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 23:33:13,865 INFO [train.py:968] (1/2) Epoch 27, batch 18400, giga_loss[loss=0.2863, simple_loss=0.3666, pruned_loss=0.103, over 27874.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3316, pruned_loss=0.09016, over 5687567.50 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3279, pruned_loss=0.08409, over 5755590.64 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3291, pruned_loss=0.08947, over 5685178.08 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:33:23,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:30,038 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:33:37,018 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-13 23:33:55,958 INFO [train.py:968] (1/2) Epoch 27, batch 18450, giga_loss[loss=0.2615, simple_loss=0.35, pruned_loss=0.08649, over 27867.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3356, pruned_loss=0.09088, over 5689659.33 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3276, pruned_loss=0.08385, over 5757174.25 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3339, pruned_loss=0.0906, over 5685876.22 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:34:19,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-13 23:34:37,423 INFO [optim.py:369] (1/2) 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 23:34:39,397 INFO [train.py:968] (1/2) Epoch 27, batch 18500, giga_loss[loss=0.3123, simple_loss=0.3818, pruned_loss=0.1214, over 28636.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3383, pruned_loss=0.0914, over 5683978.25 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3283, pruned_loss=0.08395, over 5752328.96 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3367, pruned_loss=0.09134, over 5682446.27 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:34:41,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-13 23:34:50,663 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:34:52,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:14,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1203392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:16,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1203395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:18,025 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:22,057 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3939, 1.7751, 1.4686, 1.4826], device='cuda:1'), covar=tensor([0.0830, 0.0339, 0.0352, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-13 23:35:23,753 INFO [train.py:968] (1/2) Epoch 27, batch 18550, giga_loss[loss=0.2882, simple_loss=0.3724, pruned_loss=0.102, over 29011.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3404, pruned_loss=0.09283, over 5687262.55 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3283, pruned_loss=0.08395, over 5752328.96 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3391, pruned_loss=0.09279, over 5686070.20 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:35:41,112 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:35:49,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5845, 2.8489, 1.6525, 1.6810], device='cuda:1'), covar=tensor([0.0834, 0.0304, 0.0727, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0560, 0.0403, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 23:36:04,501 INFO [optim.py:369] (1/2) 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 23:36:07,121 INFO [train.py:968] (1/2) Epoch 27, batch 18600, giga_loss[loss=0.272, simple_loss=0.3543, pruned_loss=0.09485, over 28877.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3423, pruned_loss=0.09416, over 5692391.48 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3284, pruned_loss=0.08396, over 5756615.89 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3415, pruned_loss=0.09436, over 5685966.21 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:36:40,093 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-13 23:36:48,745 INFO [train.py:968] (1/2) Epoch 27, batch 18650, giga_loss[loss=0.3082, simple_loss=0.3784, pruned_loss=0.119, over 28903.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3457, pruned_loss=0.09646, over 5699856.10 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3286, pruned_loss=0.08405, over 5757731.54 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3451, pruned_loss=0.09673, over 5692624.22 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:37:23,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1203547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:37:27,806 INFO [optim.py:369] (1/2) 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 23:37:29,838 INFO [train.py:968] (1/2) Epoch 27, batch 18700, giga_loss[loss=0.2632, simple_loss=0.3495, pruned_loss=0.08842, over 28576.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.349, pruned_loss=0.09765, over 5703935.89 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3294, pruned_loss=0.08433, over 5759704.06 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09797, over 5694981.21 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 23:37:36,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7196, 1.9443, 1.3346, 1.4813], device='cuda:1'), covar=tensor([0.1123, 0.0663, 0.1171, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0443, 0.0524, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 23:37:53,329 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5364, 1.6061, 1.6383, 1.5137], device='cuda:1'), covar=tensor([0.2782, 0.2836, 0.2181, 0.2499], device='cuda:1'), in_proj_covar=tensor([0.2038, 0.1973, 0.1876, 0.2026], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-13 23:38:07,765 INFO [train.py:968] (1/2) Epoch 27, batch 18750, giga_loss[loss=0.2632, simple_loss=0.3449, pruned_loss=0.09076, over 27951.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3493, pruned_loss=0.09695, over 5699477.99 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3292, pruned_loss=0.08435, over 5752331.33 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3496, pruned_loss=0.09756, over 5697573.29 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 2.0
+2023-03-13 23:38:12,072 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5976, 1.8935, 1.5276, 1.6199], device='cuda:1'), covar=tensor([0.2771, 0.2841, 0.3223, 0.2529], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1138, 0.1397, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 23:38:23,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3937, 1.7002, 1.3452, 1.3605], device='cuda:1'), covar=tensor([0.2817, 0.2846, 0.3210, 0.2395], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1138, 0.1397, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 23:38:46,110 INFO [optim.py:369] (1/2) 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 23:38:47,344 INFO [train.py:968] (1/2) Epoch 27, batch 18800, giga_loss[loss=0.266, simple_loss=0.3516, pruned_loss=0.09019, over 28611.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3515, pruned_loss=0.09752, over 5706640.69 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3292, pruned_loss=0.08439, over 5754383.42 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3519, pruned_loss=0.09813, over 5702799.50 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:39:20,036 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9515, 1.2844, 1.0987, 0.2062], device='cuda:1'), covar=tensor([0.5268, 0.3809, 0.5543, 0.7823], device='cuda:1'), in_proj_covar=tensor([0.1823, 0.1720, 0.1650, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 23:39:26,350 INFO [train.py:968] (1/2) Epoch 27, batch 18850, giga_loss[loss=0.2599, simple_loss=0.3447, pruned_loss=0.08753, over 28784.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3512, pruned_loss=0.09645, over 5690310.87 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3299, pruned_loss=0.0847, over 5747037.86 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3518, pruned_loss=0.09714, over 5692234.47 frames. ], batch size: 66, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:39:48,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2689, 1.6804, 1.3544, 1.4203], device='cuda:1'), covar=tensor([0.2264, 0.1986, 0.2271, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0748, 0.0718, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 23:40:08,222 INFO [optim.py:369] (1/2) 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 23:40:09,661 INFO [train.py:968] (1/2) Epoch 27, batch 18900, giga_loss[loss=0.2643, simple_loss=0.3518, pruned_loss=0.08838, over 28723.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3505, pruned_loss=0.09506, over 5696097.99 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3299, pruned_loss=0.0847, over 5747037.86 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.351, pruned_loss=0.0956, over 5697595.15 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:40:20,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3350, 3.3930, 1.5763, 1.4615], device='cuda:1'), covar=tensor([0.1018, 0.0292, 0.0908, 0.1410], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0560, 0.0403, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-13 23:40:46,607 INFO [train.py:968] (1/2) Epoch 27, batch 18950, giga_loss[loss=0.2502, simple_loss=0.3371, pruned_loss=0.08163, over 28592.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3489, pruned_loss=0.09349, over 5703551.20 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3302, pruned_loss=0.0846, over 5751624.53 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3495, pruned_loss=0.09427, over 5699501.39 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:41:01,491 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-13 23:41:25,440 INFO [optim.py:369] (1/2) 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 23:41:26,732 INFO [train.py:968] (1/2) Epoch 27, batch 19000, giga_loss[loss=0.277, simple_loss=0.3587, pruned_loss=0.09763, over 28910.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.351, pruned_loss=0.09615, over 5697844.21 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3307, pruned_loss=0.08468, over 5756250.01 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09702, over 5689046.00 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:41:37,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2391, 0.8835, 0.9601, 1.4166], device='cuda:1'), covar=tensor([0.0798, 0.0370, 0.0358, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 23:42:07,909 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-13 23:42:11,489 INFO [train.py:968] (1/2) Epoch 27, batch 19050, giga_loss[loss=0.3632, simple_loss=0.4021, pruned_loss=0.1621, over 26594.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3546, pruned_loss=0.1013, over 5688311.09 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3312, pruned_loss=0.08503, over 5759904.45 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.3552, pruned_loss=0.1021, over 5676532.55 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:42:25,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1203922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:42:26,488 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1203923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:42:42,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5908, 2.1709, 1.5103, 0.8630], device='cuda:1'), covar=tensor([0.5642, 0.3158, 0.4226, 0.6146], device='cuda:1'), in_proj_covar=tensor([0.1818, 0.1711, 0.1644, 0.1482], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 23:42:47,955 INFO [optim.py:369] (1/2) 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 23:42:50,144 INFO [train.py:968] (1/2) Epoch 27, batch 19100, giga_loss[loss=0.2507, simple_loss=0.3279, pruned_loss=0.08673, over 28495.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3538, pruned_loss=0.1018, over 5690962.91 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3314, pruned_loss=0.08508, over 5754838.90 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3548, pruned_loss=0.1029, over 5684659.91 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:42:58,474 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1203966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:43:30,846 INFO [train.py:968] (1/2) Epoch 27, batch 19150, giga_loss[loss=0.284, simple_loss=0.3501, pruned_loss=0.109, over 28846.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3521, pruned_loss=0.1017, over 5695434.32 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3315, pruned_loss=0.08506, over 5756800.73 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.353, pruned_loss=0.1028, over 5688062.70 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:44:08,819 INFO [optim.py:369] (1/2) 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 23:44:10,018 INFO [train.py:968] (1/2) Epoch 27, batch 19200, giga_loss[loss=0.2433, simple_loss=0.3208, pruned_loss=0.08293, over 28902.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3491, pruned_loss=0.09972, over 5705866.22 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3316, pruned_loss=0.08501, over 5760325.78 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3507, pruned_loss=0.1014, over 5693953.83 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:44:20,402 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:44:22,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:44:32,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5146, 1.7254, 1.4370, 1.4716], device='cuda:1'), covar=tensor([0.2656, 0.2754, 0.3148, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1135, 0.1391, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 23:44:50,266 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:44:56,008 INFO [train.py:968] (1/2) Epoch 27, batch 19250, giga_loss[loss=0.2642, simple_loss=0.3393, pruned_loss=0.09457, over 27973.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3493, pruned_loss=0.09943, over 5694028.96 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3318, pruned_loss=0.08496, over 5763355.22 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3508, pruned_loss=0.1011, over 5680328.76 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:45:34,143 INFO [optim.py:369] (1/2) 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 23:45:35,376 INFO [train.py:968] (1/2) Epoch 27, batch 19300, giga_loss[loss=0.2541, simple_loss=0.3369, pruned_loss=0.08569, over 28629.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.0973, over 5697409.16 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3321, pruned_loss=0.08506, over 5762319.07 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3483, pruned_loss=0.09874, over 5686559.40 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:46:22,436 INFO [train.py:968] (1/2) Epoch 27, batch 19350, libri_loss[loss=0.2141, simple_loss=0.2932, pruned_loss=0.06755, over 29497.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3421, pruned_loss=0.0943, over 5691661.42 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3316, pruned_loss=0.08476, over 5765314.87 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3437, pruned_loss=0.09593, over 5679122.66 frames. ], batch size: 70, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:47:05,145 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 19400, giga_loss[loss=0.2334, simple_loss=0.3103, pruned_loss=0.07829, over 28313.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3367, pruned_loss=0.09143, over 5692159.40 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3312, pruned_loss=0.08438, over 5769608.00 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3385, pruned_loss=0.0933, over 5676447.99 frames. ], batch size: 368, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:47:42,251 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1204298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:47:48,698 INFO [train.py:968] (1/2) Epoch 27, batch 19450, giga_loss[loss=0.2156, simple_loss=0.2981, pruned_loss=0.06656, over 28986.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3321, pruned_loss=0.08944, over 5684080.65 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3316, pruned_loss=0.0846, over 5764816.14 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3333, pruned_loss=0.09092, over 5673029.13 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:47:53,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1313, 1.3133, 1.1450, 0.9409], device='cuda:1'), covar=tensor([0.1145, 0.0555, 0.1102, 0.1164], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0444, 0.0524, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 23:48:11,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9476, 1.4561, 1.3203, 1.2861], device='cuda:1'), covar=tensor([0.2738, 0.1921, 0.2642, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0751, 0.0722, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-13 23:48:13,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5878, 1.8045, 1.3120, 1.3308], device='cuda:1'), covar=tensor([0.1152, 0.0673, 0.1124, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0444, 0.0523, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 23:48:24,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1204341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:48:35,969 INFO [optim.py:369] (1/2) 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 23:48:36,687 INFO [train.py:968] (1/2) Epoch 27, batch 19500, libri_loss[loss=0.2239, simple_loss=0.306, pruned_loss=0.07092, over 29339.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3291, pruned_loss=0.08759, over 5695184.89 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3315, pruned_loss=0.08448, over 5766717.68 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3301, pruned_loss=0.08892, over 5683756.69 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:49:14,543 INFO [train.py:968] (1/2) Epoch 27, batch 19550, giga_loss[loss=0.268, simple_loss=0.3393, pruned_loss=0.09839, over 28956.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.331, pruned_loss=0.08801, over 5692730.14 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3325, pruned_loss=0.08499, over 5760350.97 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3308, pruned_loss=0.08873, over 5686640.02 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:49:15,346 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1204406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:49:33,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9445, 2.9527, 1.9712, 1.1805], device='cuda:1'), covar=tensor([0.9040, 0.4277, 0.4627, 0.7453], device='cuda:1'), in_proj_covar=tensor([0.1819, 0.1718, 0.1642, 0.1484], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 23:49:48,583 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:49:50,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:49:58,776 INFO [optim.py:369] (1/2) 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 23:49:59,602 INFO [train.py:968] (1/2) Epoch 27, batch 19600, giga_loss[loss=0.2139, simple_loss=0.2962, pruned_loss=0.06582, over 28443.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3308, pruned_loss=0.08735, over 5700318.94 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.333, pruned_loss=0.08509, over 5764422.46 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3302, pruned_loss=0.0879, over 5690071.71 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:50:12,657 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:22,892 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:26,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:39,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5811, 1.7237, 1.3456, 1.2983], device='cuda:1'), covar=tensor([0.1058, 0.0618, 0.1067, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0444, 0.0523, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-13 23:50:40,974 INFO [train.py:968] (1/2) Epoch 27, batch 19650, giga_loss[loss=0.2285, simple_loss=0.3088, pruned_loss=0.07406, over 28750.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3289, pruned_loss=0.08668, over 5709228.17 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3331, pruned_loss=0.085, over 5763534.62 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3283, pruned_loss=0.08722, over 5700886.13 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:50:49,737 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:50:50,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7200, 1.9573, 1.6154, 1.7444], device='cuda:1'), covar=tensor([0.2691, 0.2837, 0.3206, 0.2524], device='cuda:1'), in_proj_covar=tensor([0.1576, 0.1136, 0.1394, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 23:51:18,570 INFO [optim.py:369] (1/2) 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 23:51:19,265 INFO [train.py:968] (1/2) Epoch 27, batch 19700, giga_loss[loss=0.2531, simple_loss=0.3204, pruned_loss=0.09289, over 28800.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3274, pruned_loss=0.08597, over 5717483.93 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3338, pruned_loss=0.08516, over 5765237.00 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3262, pruned_loss=0.08628, over 5708070.04 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:52:00,473 INFO [train.py:968] (1/2) Epoch 27, batch 19750, giga_loss[loss=0.2736, simple_loss=0.346, pruned_loss=0.1006, over 28789.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.326, pruned_loss=0.08545, over 5723272.97 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3348, pruned_loss=0.08539, over 5767889.62 frames. ], giga_tot_loss[loss=0.2475, simple_loss=0.3239, pruned_loss=0.08551, over 5712143.05 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:52:21,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1204631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:52:40,145 INFO [optim.py:369] (1/2) 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 23:52:40,601 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-13 23:52:40,819 INFO [train.py:968] (1/2) Epoch 27, batch 19800, giga_loss[loss=0.2254, simple_loss=0.3014, pruned_loss=0.07475, over 28811.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3245, pruned_loss=0.08506, over 5724359.80 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3351, pruned_loss=0.08541, over 5769892.17 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3224, pruned_loss=0.08509, over 5713170.87 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:52:41,776 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7233, 1.9625, 1.6398, 1.7503], device='cuda:1'), covar=tensor([0.2656, 0.2832, 0.3079, 0.2558], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1138, 0.1397, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-13 23:53:00,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3556, 3.0274, 1.3732, 1.5768], device='cuda:1'), covar=tensor([0.1041, 0.0361, 0.0940, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0558, 0.0402, 0.0438], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-13 23:53:17,271 INFO [train.py:968] (1/2) Epoch 27, batch 19850, giga_loss[loss=0.2298, simple_loss=0.3114, pruned_loss=0.07414, over 28740.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3224, pruned_loss=0.08423, over 5728218.03 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.335, pruned_loss=0.0852, over 5773267.90 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3203, pruned_loss=0.08441, over 5714195.43 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:53:35,430 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6803, 2.3403, 1.6440, 0.9345], device='cuda:1'), covar=tensor([0.5643, 0.3289, 0.4088, 0.6253], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1711, 0.1638, 0.1481], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-13 23:53:58,383 INFO [optim.py:369] (1/2) 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 23:53:58,395 INFO [train.py:968] (1/2) Epoch 27, batch 19900, giga_loss[loss=0.2412, simple_loss=0.3188, pruned_loss=0.08183, over 28729.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3209, pruned_loss=0.08403, over 5721614.43 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.335, pruned_loss=0.08519, over 5771546.56 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.319, pruned_loss=0.08416, over 5711530.17 frames. ], batch size: 242, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:54:08,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5136, 1.8592, 1.5224, 1.7043], device='cuda:1'), covar=tensor([0.0798, 0.0300, 0.0334, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0121, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 23:54:17,776 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1204781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:54:36,417 INFO [train.py:968] (1/2) Epoch 27, batch 19950, giga_loss[loss=0.2379, simple_loss=0.3167, pruned_loss=0.07957, over 28727.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3194, pruned_loss=0.08339, over 5725849.46 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3353, pruned_loss=0.0852, over 5772867.53 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3175, pruned_loss=0.08346, over 5715983.97 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:54:57,281 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.42 vs. limit=5.0
+2023-03-13 23:55:16,774 INFO [optim.py:369] (1/2) 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 23:55:16,786 INFO [train.py:968] (1/2) Epoch 27, batch 20000, giga_loss[loss=0.2026, simple_loss=0.2829, pruned_loss=0.06115, over 28913.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3169, pruned_loss=0.08226, over 5731467.71 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3356, pruned_loss=0.08535, over 5773116.02 frames. ], giga_tot_loss[loss=0.2397, simple_loss=0.315, pruned_loss=0.08218, over 5723109.81 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:55:55,435 INFO [train.py:968] (1/2) Epoch 27, batch 20050, giga_loss[loss=0.3074, simple_loss=0.3651, pruned_loss=0.1248, over 27944.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3158, pruned_loss=0.08175, over 5730850.73 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3357, pruned_loss=0.08528, over 5771647.23 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.314, pruned_loss=0.08172, over 5724769.06 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:56:09,888 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1204924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:56:12,141 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1204927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:56:33,885 INFO [optim.py:369] (1/2) 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 23:56:33,897 INFO [train.py:968] (1/2) Epoch 27, batch 20100, libri_loss[loss=0.3012, simple_loss=0.3805, pruned_loss=0.1109, over 29513.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.3182, pruned_loss=0.083, over 5736054.75 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3367, pruned_loss=0.0856, over 5775216.15 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.3154, pruned_loss=0.08259, over 5726934.35 frames. ], batch size: 81, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:56:34,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1204956.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:57:17,345 INFO [train.py:968] (1/2) Epoch 27, batch 20150, giga_loss[loss=0.3413, simple_loss=0.394, pruned_loss=0.1443, over 26619.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3241, pruned_loss=0.08652, over 5724047.64 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.337, pruned_loss=0.08552, over 5777724.06 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3209, pruned_loss=0.08623, over 5712854.89 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:57:19,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:57:53,475 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-13 23:57:57,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6211, 0.9304, 4.7828, 3.5519], device='cuda:1'), covar=tensor([0.1666, 0.3198, 0.0390, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0662, 0.0983, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 23:57:59,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3432, 1.3956, 3.7267, 3.2462], device='cuda:1'), covar=tensor([0.1518, 0.2712, 0.0425, 0.2003], device='cuda:1'), in_proj_covar=tensor([0.0792, 0.0662, 0.0983, 0.0957], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-13 23:58:04,016 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-13 23:58:04,166 INFO [optim.py:369] (1/2) 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 23:58:04,179 INFO [train.py:968] (1/2) Epoch 27, batch 20200, giga_loss[loss=0.324, simple_loss=0.3996, pruned_loss=0.1243, over 28581.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3313, pruned_loss=0.09087, over 5715491.63 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3369, pruned_loss=0.08533, over 5781170.23 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3286, pruned_loss=0.09087, over 5702163.92 frames. ], batch size: 307, lr: 1.18e-03, grad_scale: 8.0
+2023-03-13 23:58:24,860 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:58:47,603 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2652, 1.4778, 1.3241, 1.5396], device='cuda:1'), covar=tensor([0.0814, 0.0343, 0.0346, 0.0898], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-13 23:58:50,025 INFO [train.py:968] (1/2) Epoch 27, batch 20250, giga_loss[loss=0.2467, simple_loss=0.3323, pruned_loss=0.08053, over 28894.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3381, pruned_loss=0.09498, over 5700276.18 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3371, pruned_loss=0.08545, over 5777957.60 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3357, pruned_loss=0.09513, over 5690629.31 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:59:29,931 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:31,367 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:33,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:35,184 INFO [train.py:968] (1/2) Epoch 27, batch 20300, giga_loss[loss=0.2698, simple_loss=0.355, pruned_loss=0.09229, over 28861.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.342, pruned_loss=0.09626, over 5695782.53 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3371, pruned_loss=0.08548, over 5778686.20 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3402, pruned_loss=0.09658, over 5685579.30 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-13 23:59:35,824 INFO [optim.py:369] (1/2) 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 23:59:50,063 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-13 23:59:58,933 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:00:19,143 INFO [train.py:968] (1/2) Epoch 27, batch 20350, libri_loss[loss=0.2308, simple_loss=0.3159, pruned_loss=0.0729, over 29571.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3467, pruned_loss=0.0983, over 5690392.07 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3371, pruned_loss=0.08542, over 5781843.94 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3455, pruned_loss=0.0989, over 5677250.54 frames. ], batch size: 77, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:01:04,753 INFO [train.py:968] (1/2) Epoch 27, batch 20400, giga_loss[loss=0.2765, simple_loss=0.3605, pruned_loss=0.09621, over 28935.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3527, pruned_loss=0.1023, over 5684684.85 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3365, pruned_loss=0.08513, over 5785107.57 frames. ], giga_tot_loss[loss=0.2798, simple_loss=0.3526, pruned_loss=0.1035, over 5668617.93 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:01:05,423 INFO [optim.py:369] (1/2) 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 00:01:45,215 INFO [train.py:968] (1/2) Epoch 27, batch 20450, giga_loss[loss=0.2585, simple_loss=0.3388, pruned_loss=0.08916, over 28665.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3522, pruned_loss=0.1017, over 5690944.30 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3366, pruned_loss=0.08543, over 5787806.94 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3525, pruned_loss=0.1029, over 5672572.30 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:02:26,020 INFO [train.py:968] (1/2) Epoch 27, batch 20500, giga_loss[loss=0.2449, simple_loss=0.3347, pruned_loss=0.07757, over 28924.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3493, pruned_loss=0.09954, over 5694453.94 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3373, pruned_loss=0.08604, over 5790361.54 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3493, pruned_loss=0.1004, over 5674711.80 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:02:27,378 INFO [optim.py:369] (1/2) 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 00:02:41,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3882, 1.7058, 1.6171, 1.5440], device='cuda:1'), covar=tensor([0.2129, 0.1813, 0.2352, 0.1936], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0752, 0.0723, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 00:03:07,103 INFO [train.py:968] (1/2) Epoch 27, batch 20550, giga_loss[loss=0.2351, simple_loss=0.3186, pruned_loss=0.0758, over 28709.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3471, pruned_loss=0.0973, over 5707175.57 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3377, pruned_loss=0.08629, over 5789786.00 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3468, pruned_loss=0.09788, over 5690819.65 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:03:45,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:03:50,384 INFO [train.py:968] (1/2) Epoch 27, batch 20600, giga_loss[loss=0.2405, simple_loss=0.329, pruned_loss=0.07604, over 28353.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3468, pruned_loss=0.09681, over 5691628.29 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.338, pruned_loss=0.08645, over 5781607.69 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3466, pruned_loss=0.09742, over 5683223.85 frames. ], batch size: 60, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:03:52,814 INFO [optim.py:369] (1/2) 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 00:04:00,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:04:20,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3588, 1.7192, 1.4083, 1.5088], device='cuda:1'), covar=tensor([0.0744, 0.0413, 0.0341, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 00:04:24,557 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:04:30,588 INFO [train.py:968] (1/2) Epoch 27, batch 20650, giga_loss[loss=0.2817, simple_loss=0.3553, pruned_loss=0.1041, over 28674.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3496, pruned_loss=0.09905, over 5676942.84 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3381, pruned_loss=0.08669, over 5764094.22 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3496, pruned_loss=0.09961, over 5682688.81 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:04:45,144 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:04:51,810 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-14 00:05:05,087 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:05:12,293 INFO [train.py:968] (1/2) Epoch 27, batch 20700, giga_loss[loss=0.2526, simple_loss=0.3351, pruned_loss=0.08508, over 29086.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3513, pruned_loss=0.1, over 5694780.54 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3386, pruned_loss=0.08689, over 5767321.38 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3512, pruned_loss=0.1006, over 5694257.76 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:05:13,550 INFO [optim.py:369] (1/2) 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 00:05:34,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3949, 1.6723, 1.6424, 1.1824], device='cuda:1'), covar=tensor([0.1860, 0.3096, 0.1701, 0.2075], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0712, 0.0977, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 00:05:39,795 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:05:42,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:05:50,555 INFO [train.py:968] (1/2) Epoch 27, batch 20750, giga_loss[loss=0.2489, simple_loss=0.3317, pruned_loss=0.08305, over 28859.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3519, pruned_loss=0.1006, over 5688653.92 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3387, pruned_loss=0.08693, over 5768170.54 frames. ], giga_tot_loss[loss=0.2777, simple_loss=0.3522, pruned_loss=0.1016, over 5684592.12 frames. ], batch size: 145, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:05:53,234 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:06:08,395 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:06:23,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205641.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:06:25,045 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8984, 1.1672, 2.8547, 2.8123], device='cuda:1'), covar=tensor([0.1664, 0.2661, 0.0591, 0.1262], device='cuda:1'), in_proj_covar=tensor([0.0793, 0.0662, 0.0987, 0.0961], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 00:06:34,590 INFO [train.py:968] (1/2) Epoch 27, batch 20800, giga_loss[loss=0.2921, simple_loss=0.3623, pruned_loss=0.1109, over 28760.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3548, pruned_loss=0.1033, over 5685345.67 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3388, pruned_loss=0.08693, over 5770476.23 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3552, pruned_loss=0.1044, over 5678912.70 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:06:36,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4386, 1.7843, 1.4278, 1.4161], device='cuda:1'), covar=tensor([0.2570, 0.2583, 0.2955, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.1577, 0.1138, 0.1392, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 00:06:37,714 INFO [optim.py:369] (1/2) 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 00:06:44,609 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:06:48,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:06:51,898 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 00:07:02,404 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:05,447 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:09,266 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:14,551 INFO [train.py:968] (1/2) Epoch 27, batch 20850, giga_loss[loss=0.244, simple_loss=0.3233, pruned_loss=0.08233, over 28368.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3541, pruned_loss=0.1032, over 5687908.18 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3388, pruned_loss=0.08698, over 5765471.67 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3549, pruned_loss=0.1044, over 5685110.28 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:07:26,786 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1205720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:34,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:07:52,916 INFO [train.py:968] (1/2) Epoch 27, batch 20900, giga_loss[loss=0.2525, simple_loss=0.3345, pruned_loss=0.08521, over 29103.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3539, pruned_loss=0.1023, over 5690729.45 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3395, pruned_loss=0.08741, over 5755528.97 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3544, pruned_loss=0.1034, over 5694945.67 frames. ], batch size: 113, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:07:55,068 INFO [optim.py:369] (1/2) 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 00:08:32,406 INFO [train.py:968] (1/2) Epoch 27, batch 20950, giga_loss[loss=0.2732, simple_loss=0.3446, pruned_loss=0.1009, over 28641.00 frames. ], tot_loss[loss=0.279, simple_loss=0.3541, pruned_loss=0.102, over 5689461.99 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3396, pruned_loss=0.08772, over 5757194.37 frames. ], giga_tot_loss[loss=0.2807, simple_loss=0.3549, pruned_loss=0.1032, over 5688778.20 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:09:02,185 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:09:13,271 INFO [train.py:968] (1/2) Epoch 27, batch 21000, giga_loss[loss=0.2642, simple_loss=0.3478, pruned_loss=0.09033, over 28868.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3552, pruned_loss=0.1015, over 5696597.30 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08792, over 5759617.47 frames. ], giga_tot_loss[loss=0.2803, simple_loss=0.3557, pruned_loss=0.1025, over 5692979.45 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:09:13,272 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 00:09:21,777 INFO [train.py:1012] (1/2) Epoch 27, validation: loss=0.2045, simple_loss=0.3123, pruned_loss=0.04838, over 944034.00 frames. 
+2023-03-14 00:09:21,777 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 00:09:23,893 INFO [optim.py:369] (1/2) 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 00:09:36,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:09:45,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4681, 1.7609, 1.1486, 1.3083], device='cuda:1'), covar=tensor([0.0964, 0.0469, 0.1050, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0444, 0.0522, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 00:10:01,708 INFO [train.py:968] (1/2) Epoch 27, batch 21050, giga_loss[loss=0.2832, simple_loss=0.3555, pruned_loss=0.1054, over 28877.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3539, pruned_loss=0.1011, over 5700248.26 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3401, pruned_loss=0.08799, over 5761190.01 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3544, pruned_loss=0.1019, over 5695388.56 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:10:11,525 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1205918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:10:16,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5457, 2.1792, 1.7471, 0.7984], device='cuda:1'), covar=tensor([0.6111, 0.3098, 0.4342, 0.6614], device='cuda:1'), in_proj_covar=tensor([0.1806, 0.1696, 0.1627, 0.1473], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 00:10:39,122 INFO [train.py:968] (1/2) Epoch 27, batch 21100, giga_loss[loss=0.2577, simple_loss=0.3316, pruned_loss=0.09189, over 28839.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3499, pruned_loss=0.09862, over 5713213.92 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3399, pruned_loss=0.08798, over 5762849.20 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3508, pruned_loss=0.09959, over 5706618.95 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 2.0
+2023-03-14 00:10:41,950 INFO [optim.py:369] (1/2) 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 00:11:01,181 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 00:11:02,568 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1205983.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:03,180 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1205984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:05,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1205987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:19,381 INFO [train.py:968] (1/2) Epoch 27, batch 21150, giga_loss[loss=0.309, simple_loss=0.3754, pruned_loss=0.1213, over 28677.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3485, pruned_loss=0.09808, over 5713581.08 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3404, pruned_loss=0.08841, over 5764354.09 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3488, pruned_loss=0.09862, over 5705934.62 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 2.0
+2023-03-14 00:11:24,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2161, 1.4833, 1.5266, 1.1002], device='cuda:1'), covar=tensor([0.1577, 0.2928, 0.1340, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0713, 0.0976, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 00:11:27,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:27,707 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:27,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206016.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:11:29,122 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:11:36,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-14 00:11:52,347 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 21200, giga_loss[loss=0.3142, simple_loss=0.3805, pruned_loss=0.1239, over 28724.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3492, pruned_loss=0.09938, over 5710162.45 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3408, pruned_loss=0.08874, over 5767050.19 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3493, pruned_loss=0.09971, over 5700589.78 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:12:03,221 INFO [optim.py:369] (1/2) 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 00:12:16,865 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 00:12:39,631 INFO [train.py:968] (1/2) Epoch 27, batch 21250, giga_loss[loss=0.2604, simple_loss=0.3483, pruned_loss=0.08624, over 29090.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3499, pruned_loss=0.09959, over 5718427.25 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3414, pruned_loss=0.08919, over 5769390.18 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3496, pruned_loss=0.09974, over 5706961.62 frames. ], batch size: 155, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:12:40,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:12:54,730 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:12:56,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:13:18,378 INFO [train.py:968] (1/2) Epoch 27, batch 21300, giga_loss[loss=0.2836, simple_loss=0.3494, pruned_loss=0.1089, over 28590.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.0988, over 5712185.94 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3415, pruned_loss=0.08938, over 5763174.10 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3491, pruned_loss=0.09896, over 5706868.11 frames. ], batch size: 78, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:13:21,770 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:13:22,390 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206159.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:13:24,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206162.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:13:26,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1085, 2.3306, 2.2284, 2.0166], device='cuda:1'), covar=tensor([0.2283, 0.2490, 0.2136, 0.2434], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0750, 0.0719, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 00:13:46,229 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206191.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:13:57,055 INFO [train.py:968] (1/2) Epoch 27, batch 21350, giga_loss[loss=0.2897, simple_loss=0.3597, pruned_loss=0.1098, over 28666.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3477, pruned_loss=0.09707, over 5704562.39 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3414, pruned_loss=0.08949, over 5755079.76 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3478, pruned_loss=0.09726, over 5705638.95 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:14:31,131 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:14:33,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:14:36,111 INFO [train.py:968] (1/2) Epoch 27, batch 21400, giga_loss[loss=0.2424, simple_loss=0.3183, pruned_loss=0.08325, over 28586.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3463, pruned_loss=0.09592, over 5715375.94 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3415, pruned_loss=0.08954, over 5756981.23 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3465, pruned_loss=0.09616, over 5713495.57 frames. ], batch size: 71, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:14:39,946 INFO [optim.py:369] (1/2) 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 00:14:57,578 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:15:07,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:15:16,009 INFO [train.py:968] (1/2) Epoch 27, batch 21450, giga_loss[loss=0.2375, simple_loss=0.3318, pruned_loss=0.07161, over 28975.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.345, pruned_loss=0.09541, over 5717899.77 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.342, pruned_loss=0.08992, over 5757346.17 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3447, pruned_loss=0.09533, over 5715620.18 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:15:54,359 INFO [train.py:968] (1/2) Epoch 27, batch 21500, giga_loss[loss=0.269, simple_loss=0.3413, pruned_loss=0.09836, over 28840.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3426, pruned_loss=0.09483, over 5716002.95 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3418, pruned_loss=0.09015, over 5760059.54 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3425, pruned_loss=0.0947, over 5710812.68 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:15:57,169 INFO [optim.py:369] (1/2) 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 00:16:31,622 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6081, 1.6618, 1.7915, 1.3999], device='cuda:1'), covar=tensor([0.1887, 0.2722, 0.1550, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0716, 0.0980, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 00:16:32,699 INFO [train.py:968] (1/2) Epoch 27, batch 21550, giga_loss[loss=0.25, simple_loss=0.3348, pruned_loss=0.08254, over 28942.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3406, pruned_loss=0.09377, over 5717880.99 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3422, pruned_loss=0.09041, over 5754647.50 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3401, pruned_loss=0.0935, over 5717345.03 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:16:43,421 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:16:57,376 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:16:57,487 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:17:11,997 INFO [train.py:968] (1/2) Epoch 27, batch 21600, giga_loss[loss=0.2429, simple_loss=0.3195, pruned_loss=0.08313, over 29086.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3405, pruned_loss=0.09411, over 5723670.56 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3428, pruned_loss=0.09088, over 5756902.95 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3397, pruned_loss=0.09353, over 5720605.80 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:17:14,860 INFO [optim.py:369] (1/2) 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 00:17:23,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:17:24,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:17:51,513 INFO [train.py:968] (1/2) Epoch 27, batch 21650, giga_loss[loss=0.341, simple_loss=0.3874, pruned_loss=0.1473, over 26792.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3397, pruned_loss=0.09481, over 5709763.13 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3429, pruned_loss=0.09118, over 5749556.22 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3389, pruned_loss=0.09414, over 5713280.83 frames. ], batch size: 555, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:18:09,308 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:18:32,994 INFO [train.py:968] (1/2) Epoch 27, batch 21700, giga_loss[loss=0.2631, simple_loss=0.3339, pruned_loss=0.09619, over 28735.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3376, pruned_loss=0.09394, over 5701310.44 frames. ], libri_tot_loss[loss=0.2631, simple_loss=0.3433, pruned_loss=0.0915, over 5742384.49 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3365, pruned_loss=0.09315, over 5710807.07 frames. ], batch size: 119, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:18:38,006 INFO [optim.py:369] (1/2) 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 00:18:58,644 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-14 00:19:15,415 INFO [train.py:968] (1/2) Epoch 27, batch 21750, giga_loss[loss=0.2407, simple_loss=0.3153, pruned_loss=0.08308, over 28678.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3336, pruned_loss=0.09209, over 5701977.37 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3431, pruned_loss=0.09147, over 5743165.57 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3328, pruned_loss=0.09151, over 5708426.59 frames. ], batch size: 92, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:19:45,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5164, 1.8987, 1.4857, 1.6212], device='cuda:1'), covar=tensor([0.2795, 0.2845, 0.3289, 0.2561], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1140, 0.1396, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 00:19:51,868 INFO [train.py:968] (1/2) Epoch 27, batch 21800, giga_loss[loss=0.2228, simple_loss=0.31, pruned_loss=0.0678, over 28925.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3316, pruned_loss=0.09128, over 5709738.09 frames. ], libri_tot_loss[loss=0.2638, simple_loss=0.3435, pruned_loss=0.09204, over 5748566.36 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3303, pruned_loss=0.0903, over 5708772.95 frames. ], batch size: 174, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:19:55,336 INFO [optim.py:369] (1/2) 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 00:20:24,870 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1206694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:20:33,490 INFO [train.py:968] (1/2) Epoch 27, batch 21850, giga_loss[loss=0.244, simple_loss=0.3215, pruned_loss=0.0832, over 28844.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3312, pruned_loss=0.09091, over 5706129.46 frames. ], libri_tot_loss[loss=0.2642, simple_loss=0.3438, pruned_loss=0.09225, over 5749007.14 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3297, pruned_loss=0.08992, over 5704498.75 frames. ], batch size: 112, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:20:50,150 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3967, 3.1546, 1.4564, 1.4779], device='cuda:1'), covar=tensor([0.0962, 0.0405, 0.0935, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0561, 0.0403, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-14 00:21:14,267 INFO [train.py:968] (1/2) Epoch 27, batch 21900, giga_loss[loss=0.2773, simple_loss=0.357, pruned_loss=0.09881, over 28868.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3331, pruned_loss=0.09164, over 5712080.74 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3439, pruned_loss=0.09248, over 5752999.46 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3315, pruned_loss=0.09062, over 5706565.79 frames. ], batch size: 199, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:21:18,882 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:21:56,886 INFO [train.py:968] (1/2) Epoch 27, batch 21950, giga_loss[loss=0.2394, simple_loss=0.325, pruned_loss=0.07688, over 28981.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.337, pruned_loss=0.0933, over 5706587.57 frames. ], libri_tot_loss[loss=0.2654, simple_loss=0.3447, pruned_loss=0.09303, over 5747657.61 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3348, pruned_loss=0.09196, over 5706577.68 frames. ], batch size: 164, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:22:02,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:22:09,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3187, 1.2803, 1.2216, 1.5270], device='cuda:1'), covar=tensor([0.0763, 0.0344, 0.0355, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0120, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0103, 0.0074, 0.0065, 0.0114], device='cuda:1')
+2023-03-14 00:22:13,648 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-14 00:22:30,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206844.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:22:39,300 INFO [train.py:968] (1/2) Epoch 27, batch 22000, giga_loss[loss=0.276, simple_loss=0.3432, pruned_loss=0.1044, over 28968.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3395, pruned_loss=0.09377, over 5706542.08 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3449, pruned_loss=0.09325, over 5749278.06 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3376, pruned_loss=0.09252, over 5704750.15 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:22:43,306 INFO [optim.py:369] (1/2) 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 00:22:46,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.08 vs. limit=5.0
+2023-03-14 00:23:15,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1206900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:23:19,979 INFO [train.py:968] (1/2) Epoch 27, batch 22050, giga_loss[loss=0.2592, simple_loss=0.3413, pruned_loss=0.08856, over 27556.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3397, pruned_loss=0.09338, over 5698861.88 frames. ], libri_tot_loss[loss=0.2659, simple_loss=0.3449, pruned_loss=0.09346, over 5744683.68 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.338, pruned_loss=0.09221, over 5699520.59 frames. ], batch size: 472, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:23:43,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:23:45,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:00,120 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:00,394 INFO [train.py:968] (1/2) Epoch 27, batch 22100, giga_loss[loss=0.2701, simple_loss=0.3517, pruned_loss=0.0942, over 28689.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3388, pruned_loss=0.09283, over 5701528.41 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.345, pruned_loss=0.09394, over 5750468.07 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3371, pruned_loss=0.09142, over 5694712.46 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:24:02,918 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:13,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7749, 2.0219, 1.6801, 2.2049], device='cuda:1'), covar=tensor([0.2669, 0.2862, 0.3210, 0.2468], device='cuda:1'), in_proj_covar=tensor([0.1578, 0.1137, 0.1394, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 00:24:27,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1206986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:27,966 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1206987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:30,684 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1206990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:24:43,152 INFO [train.py:968] (1/2) Epoch 27, batch 22150, giga_loss[loss=0.252, simple_loss=0.3305, pruned_loss=0.08669, over 28954.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3398, pruned_loss=0.09366, over 5710250.99 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3449, pruned_loss=0.09393, over 5755109.02 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3383, pruned_loss=0.09251, over 5699243.92 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:24:54,852 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1207019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:12,522 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:14,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1207046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:22,459 INFO [train.py:968] (1/2) Epoch 27, batch 22200, giga_loss[loss=0.2429, simple_loss=0.3254, pruned_loss=0.08021, over 28940.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3417, pruned_loss=0.09519, over 5706316.97 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3458, pruned_loss=0.09468, over 5754017.82 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3395, pruned_loss=0.09356, over 5696830.48 frames. ], batch size: 136, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:25:25,969 INFO [optim.py:369] (1/2) 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 00:25:33,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1207069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:39,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1207075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:25:49,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4164, 1.8915, 1.3300, 0.8229], device='cuda:1'), covar=tensor([0.6664, 0.3179, 0.4001, 0.6883], device='cuda:1'), in_proj_covar=tensor([0.1819, 0.1703, 0.1643, 0.1482], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 00:26:03,880 INFO [train.py:968] (1/2) Epoch 27, batch 22250, giga_loss[loss=0.2611, simple_loss=0.3463, pruned_loss=0.08794, over 28756.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3429, pruned_loss=0.09577, over 5702382.33 frames. ], libri_tot_loss[loss=0.2677, simple_loss=0.3457, pruned_loss=0.09481, over 5747803.65 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3411, pruned_loss=0.09437, over 5699812.47 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:26:32,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1754, 1.3540, 1.4064, 1.0338], device='cuda:1'), covar=tensor([0.1491, 0.2283, 0.1266, 0.1505], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0709, 0.0975, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 00:26:45,744 INFO [train.py:968] (1/2) Epoch 27, batch 22300, giga_loss[loss=0.2548, simple_loss=0.3353, pruned_loss=0.08712, over 29052.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3456, pruned_loss=0.09717, over 5697184.84 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3463, pruned_loss=0.09522, over 5741184.78 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3436, pruned_loss=0.0957, over 5699327.30 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:26:49,454 INFO [optim.py:369] (1/2) 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 00:27:26,356 INFO [train.py:968] (1/2) Epoch 27, batch 22350, giga_loss[loss=0.2797, simple_loss=0.3403, pruned_loss=0.1095, over 28805.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3478, pruned_loss=0.09795, over 5706464.33 frames. ], libri_tot_loss[loss=0.2685, simple_loss=0.3464, pruned_loss=0.0953, over 5745074.71 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09676, over 5703842.67 frames. ], batch size: 99, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:27:31,976 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:27:34,524 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1207215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:27:56,728 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1207244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:28:03,764 INFO [train.py:968] (1/2) Epoch 27, batch 22400, giga_loss[loss=0.3197, simple_loss=0.3858, pruned_loss=0.1268, over 28890.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3501, pruned_loss=0.0992, over 5703464.52 frames. ], libri_tot_loss[loss=0.2697, simple_loss=0.3474, pruned_loss=0.09595, over 5735981.37 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3479, pruned_loss=0.09771, over 5707776.07 frames. ], batch size: 227, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:28:09,324 INFO [optim.py:369] (1/2) 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 00:28:47,170 INFO [train.py:968] (1/2) Epoch 27, batch 22450, giga_loss[loss=0.2758, simple_loss=0.3503, pruned_loss=0.1007, over 28878.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3517, pruned_loss=0.1003, over 5690543.57 frames. ], libri_tot_loss[loss=0.2707, simple_loss=0.3483, pruned_loss=0.09656, over 5720333.72 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3492, pruned_loss=0.09862, over 5708360.34 frames. ], batch size: 186, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:29:28,266 INFO [train.py:968] (1/2) Epoch 27, batch 22500, giga_loss[loss=0.2608, simple_loss=0.3304, pruned_loss=0.09563, over 28912.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.3529, pruned_loss=0.1016, over 5688001.66 frames. ], libri_tot_loss[loss=0.2717, simple_loss=0.3489, pruned_loss=0.09727, over 5713738.19 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.09972, over 5708016.06 frames. ], batch size: 106, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:29:35,744 INFO [optim.py:369] (1/2) 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 00:30:11,002 INFO [train.py:968] (1/2) Epoch 27, batch 22550, giga_loss[loss=0.3355, simple_loss=0.3984, pruned_loss=0.1363, over 28792.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3497, pruned_loss=0.09978, over 5696904.55 frames. ], libri_tot_loss[loss=0.272, simple_loss=0.3492, pruned_loss=0.09739, over 5715754.50 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3475, pruned_loss=0.09823, over 5710783.25 frames. ], batch size: 284, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:30:53,791 INFO [train.py:968] (1/2) Epoch 27, batch 22600, giga_loss[loss=0.2562, simple_loss=0.3332, pruned_loss=0.08954, over 28924.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3465, pruned_loss=0.09807, over 5697109.42 frames. ], libri_tot_loss[loss=0.2723, simple_loss=0.3494, pruned_loss=0.09756, over 5719597.02 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3445, pruned_loss=0.0967, over 5704131.29 frames. ], batch size: 213, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:30:56,678 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1207459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:30:58,303 INFO [optim.py:369] (1/2) 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 00:31:31,130 INFO [train.py:968] (1/2) Epoch 27, batch 22650, libri_loss[loss=0.3066, simple_loss=0.3743, pruned_loss=0.1195, over 27885.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3451, pruned_loss=0.09777, over 5692774.81 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3506, pruned_loss=0.0986, over 5710353.04 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3424, pruned_loss=0.09573, over 5706170.07 frames. ], batch size: 116, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:32:09,863 INFO [train.py:968] (1/2) Epoch 27, batch 22700, giga_loss[loss=0.2904, simple_loss=0.3854, pruned_loss=0.09769, over 28566.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3451, pruned_loss=0.09653, over 5694225.79 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3503, pruned_loss=0.09873, over 5713924.47 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3429, pruned_loss=0.09468, over 5700852.73 frames. ], batch size: 336, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:32:17,213 INFO [optim.py:369] (1/2) 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 00:32:50,866 INFO [train.py:968] (1/2) Epoch 27, batch 22750, giga_loss[loss=0.2351, simple_loss=0.3093, pruned_loss=0.08049, over 28441.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3481, pruned_loss=0.09742, over 5698181.01 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3515, pruned_loss=0.09969, over 5715162.41 frames. ], giga_tot_loss[loss=0.2675, simple_loss=0.3451, pruned_loss=0.09496, over 5701843.23 frames. ], batch size: 85, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:32:59,483 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1207616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:33:26,713 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-14 00:33:29,103 INFO [train.py:968] (1/2) Epoch 27, batch 22800, giga_loss[loss=0.2631, simple_loss=0.3444, pruned_loss=0.09096, over 28721.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3474, pruned_loss=0.09726, over 5697407.77 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.352, pruned_loss=0.1003, over 5720578.08 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3443, pruned_loss=0.09456, over 5694553.49 frames. ], batch size: 262, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:33:33,706 INFO [optim.py:369] (1/2) 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 00:34:04,172 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0719, 5.1103, 2.0697, 2.1689], device='cuda:1'), covar=tensor([0.0828, 0.0346, 0.0823, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0563, 0.0405, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 00:34:10,768 INFO [train.py:968] (1/2) Epoch 27, batch 22850, libri_loss[loss=0.2493, simple_loss=0.3283, pruned_loss=0.08519, over 29560.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3454, pruned_loss=0.09769, over 5700786.14 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3519, pruned_loss=0.1004, over 5724412.39 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.343, pruned_loss=0.09538, over 5694628.79 frames. ], batch size: 79, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:34:19,691 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-14 00:34:50,013 INFO [train.py:968] (1/2) Epoch 27, batch 22900, giga_loss[loss=0.3483, simple_loss=0.4, pruned_loss=0.1482, over 27863.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3436, pruned_loss=0.09757, over 5707451.29 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3518, pruned_loss=0.1005, over 5726820.80 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3416, pruned_loss=0.09567, over 5700337.74 frames. ], batch size: 412, lr: 1.18e-03, grad_scale: 8.0
+2023-03-14 00:34:54,930 INFO [optim.py:369] (1/2) 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 00:35:29,608 INFO [train.py:968] (1/2) Epoch 27, batch 22950, giga_loss[loss=0.3346, simple_loss=0.3763, pruned_loss=0.1464, over 29033.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3419, pruned_loss=0.09784, over 5718062.63 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3518, pruned_loss=0.1008, over 5731185.49 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3401, pruned_loss=0.09596, over 5707907.71 frames. ], batch size: 128, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:35:52,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1207834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:35:58,327 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.73 vs. limit=5.0
+2023-03-14 00:36:07,421 INFO [train.py:968] (1/2) Epoch 27, batch 23000, libri_loss[loss=0.299, simple_loss=0.3694, pruned_loss=0.1144, over 29528.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3421, pruned_loss=0.09851, over 5717341.57 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3521, pruned_loss=0.1012, over 5735131.15 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3398, pruned_loss=0.09649, over 5704939.92 frames. ], batch size: 89, lr: 1.18e-03, grad_scale: 4.0
+2023-03-14 00:36:13,166 INFO [optim.py:369] (1/2) 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 00:36:43,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 00:36:45,669 INFO [train.py:968] (1/2) Epoch 27, batch 23050, giga_loss[loss=0.2578, simple_loss=0.3313, pruned_loss=0.09218, over 28583.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3397, pruned_loss=0.0971, over 5722072.02 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.352, pruned_loss=0.1013, over 5738494.68 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3378, pruned_loss=0.09534, over 5709024.24 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:37:24,605 INFO [train.py:968] (1/2) Epoch 27, batch 23100, giga_loss[loss=0.2207, simple_loss=0.3016, pruned_loss=0.06992, over 28713.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3364, pruned_loss=0.09557, over 5714162.40 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3528, pruned_loss=0.1019, over 5730978.52 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3336, pruned_loss=0.09335, over 5709373.51 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 00:37:31,581 INFO [optim.py:369] (1/2) 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 00:37:43,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1207977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:37:44,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1207980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:37:54,059 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1207991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:37:56,775 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1207995.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:38:04,277 INFO [train.py:968] (1/2) Epoch 27, batch 23150, libri_loss[loss=0.3736, simple_loss=0.4223, pruned_loss=0.1624, over 29151.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3332, pruned_loss=0.0942, over 5709931.60 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3532, pruned_loss=0.1026, over 5733262.63 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3301, pruned_loss=0.09159, over 5703080.70 frames. ], batch size: 101, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 00:38:07,111 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:38:42,973 INFO [train.py:968] (1/2) Epoch 27, batch 23200, giga_loss[loss=0.2423, simple_loss=0.3236, pruned_loss=0.08052, over 28801.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3327, pruned_loss=0.09389, over 5699228.77 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3529, pruned_loss=0.1027, over 5717203.30 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3298, pruned_loss=0.09136, over 5708377.40 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:38:50,310 INFO [optim.py:369] (1/2) 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 00:39:18,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8470, 3.6788, 3.4950, 1.7505], device='cuda:1'), covar=tensor([0.0702, 0.0815, 0.0741, 0.2069], device='cuda:1'), in_proj_covar=tensor([0.1274, 0.1170, 0.0990, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 00:39:22,623 INFO [train.py:968] (1/2) Epoch 27, batch 23250, giga_loss[loss=0.2376, simple_loss=0.3203, pruned_loss=0.07748, over 28965.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3341, pruned_loss=0.09398, over 5698022.39 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3526, pruned_loss=0.1028, over 5712154.33 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3314, pruned_loss=0.09158, over 5708829.94 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:39:24,131 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:39:40,216 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:39:46,442 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:39:50,674 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:40:04,995 INFO [train.py:968] (1/2) Epoch 27, batch 23300, giga_loss[loss=0.2361, simple_loss=0.3226, pruned_loss=0.07484, over 28870.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3372, pruned_loss=0.09487, over 5697142.27 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3532, pruned_loss=0.1033, over 5711234.75 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3343, pruned_loss=0.09244, over 5706626.58 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:40:12,039 INFO [optim.py:369] (1/2) 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 00:40:13,596 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:40:29,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:40:43,821 INFO [train.py:968] (1/2) Epoch 27, batch 23350, giga_loss[loss=0.2578, simple_loss=0.3395, pruned_loss=0.08808, over 28676.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.341, pruned_loss=0.09644, over 5705288.70 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3538, pruned_loss=0.1037, over 5715955.72 frames. ], giga_tot_loss[loss=0.2627, simple_loss=0.3377, pruned_loss=0.09389, over 5708401.64 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:41:26,538 INFO [train.py:968] (1/2) Epoch 27, batch 23400, giga_loss[loss=0.282, simple_loss=0.3701, pruned_loss=0.09692, over 28595.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3436, pruned_loss=0.09727, over 5696346.88 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3537, pruned_loss=0.1037, over 5715257.97 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3409, pruned_loss=0.09517, over 5699086.33 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:41:33,978 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 23450, giga_loss[loss=0.2843, simple_loss=0.3584, pruned_loss=0.1051, over 28872.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3455, pruned_loss=0.09847, over 5698506.58 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3538, pruned_loss=0.1039, over 5719910.04 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3431, pruned_loss=0.09649, over 5695978.75 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:42:53,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3216, 1.2618, 3.4604, 3.0458], device='cuda:1'), covar=tensor([0.1554, 0.2810, 0.0535, 0.1647], device='cuda:1'), in_proj_covar=tensor([0.0795, 0.0664, 0.0993, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 00:42:57,007 INFO [train.py:968] (1/2) Epoch 27, batch 23500, giga_loss[loss=0.3836, simple_loss=0.4299, pruned_loss=0.1687, over 28011.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3521, pruned_loss=0.1044, over 5683869.54 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3541, pruned_loss=0.1043, over 5709521.82 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3497, pruned_loss=0.1024, over 5690691.60 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:43:03,891 INFO [optim.py:369] (1/2) 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 00:43:12,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1208370.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:43:31,362 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-14 00:43:45,563 INFO [train.py:968] (1/2) Epoch 27, batch 23550, libri_loss[loss=0.2772, simple_loss=0.3501, pruned_loss=0.1021, over 29466.00 frames. ], tot_loss[loss=0.287, simple_loss=0.3573, pruned_loss=0.1084, over 5684088.05 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3542, pruned_loss=0.1044, over 5712507.30 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.3553, pruned_loss=0.1067, over 5686201.77 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:44:31,557 INFO [train.py:968] (1/2) Epoch 27, batch 23600, giga_loss[loss=0.3531, simple_loss=0.4123, pruned_loss=0.147, over 28548.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3638, pruned_loss=0.1131, over 5673261.48 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3542, pruned_loss=0.1045, over 5706805.83 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3624, pruned_loss=0.1119, over 5678394.30 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:44:39,903 INFO [optim.py:369] (1/2) 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 00:44:57,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1208482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:45:01,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5521, 1.7912, 1.4384, 1.8419], device='cuda:1'), covar=tensor([0.2374, 0.2425, 0.2648, 0.2101], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1140, 0.1395, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 00:45:15,697 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1208501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:45:18,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3238, 2.7304, 1.4070, 1.4263], device='cuda:1'), covar=tensor([0.0943, 0.0420, 0.0897, 0.1292], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0565, 0.0406, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 00:45:19,050 INFO [train.py:968] (1/2) Epoch 27, batch 23650, giga_loss[loss=0.3852, simple_loss=0.4256, pruned_loss=0.1724, over 27568.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3702, pruned_loss=0.1189, over 5657053.16 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3541, pruned_loss=0.1046, over 5690984.61 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3693, pruned_loss=0.118, over 5675440.30 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:45:28,979 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6527, 1.7085, 1.3057, 1.2882], device='cuda:1'), covar=tensor([0.0746, 0.0443, 0.0804, 0.1154], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0447, 0.0521, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 00:45:29,006 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208513.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:45:31,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208516.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 00:45:59,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208545.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 00:46:07,737 INFO [train.py:968] (1/2) Epoch 27, batch 23700, giga_loss[loss=0.4712, simple_loss=0.4676, pruned_loss=0.2374, over 23448.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3757, pruned_loss=0.1234, over 5648225.06 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3541, pruned_loss=0.1045, over 5693898.89 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3754, pruned_loss=0.1231, over 5659403.53 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:46:15,571 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:46:54,383 INFO [train.py:968] (1/2) Epoch 27, batch 23750, giga_loss[loss=0.2766, simple_loss=0.346, pruned_loss=0.1036, over 28489.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.378, pruned_loss=0.1251, over 5661582.80 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3538, pruned_loss=0.1045, over 5700976.43 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3789, pruned_loss=0.1256, over 5662624.07 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:47:12,797 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:14,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:30,671 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:38,504 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-14 00:47:39,335 INFO [train.py:968] (1/2) Epoch 27, batch 23800, giga_loss[loss=0.2888, simple_loss=0.3641, pruned_loss=0.1068, over 28854.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3796, pruned_loss=0.1272, over 5651336.70 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3538, pruned_loss=0.1047, over 5693324.26 frames. ], giga_tot_loss[loss=0.3184, simple_loss=0.381, pruned_loss=0.1279, over 5657032.98 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:47:42,328 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:47:51,090 INFO [optim.py:369] (1/2) 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 00:48:03,640 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:48:33,218 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1208704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:48:33,503 INFO [train.py:968] (1/2) Epoch 27, batch 23850, giga_loss[loss=0.4262, simple_loss=0.4417, pruned_loss=0.2054, over 23432.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3824, pruned_loss=0.1309, over 5637959.99 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3537, pruned_loss=0.1047, over 5694446.64 frames. ], giga_tot_loss[loss=0.3235, simple_loss=0.3837, pruned_loss=0.1317, over 5641222.63 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:48:36,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1208707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:48:55,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2871, 0.8806, 0.9414, 1.4610], device='cuda:1'), covar=tensor([0.0735, 0.0406, 0.0357, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 00:49:05,378 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1208736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:49:24,279 INFO [train.py:968] (1/2) Epoch 27, batch 23900, libri_loss[loss=0.2547, simple_loss=0.3195, pruned_loss=0.09495, over 29414.00 frames. ], tot_loss[loss=0.3261, simple_loss=0.385, pruned_loss=0.1337, over 5641536.58 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3536, pruned_loss=0.1047, over 5694944.78 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3866, pruned_loss=0.1347, over 5642816.66 frames. ], batch size: 67, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:49:36,543 INFO [optim.py:369] (1/2) 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 00:49:40,596 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 00:50:22,870 INFO [train.py:968] (1/2) Epoch 27, batch 23950, giga_loss[loss=0.318, simple_loss=0.3769, pruned_loss=0.1296, over 28800.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.3886, pruned_loss=0.1371, over 5620029.09 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3541, pruned_loss=0.105, over 5699061.24 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3901, pruned_loss=0.1383, over 5616238.11 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:51:14,836 INFO [train.py:968] (1/2) Epoch 27, batch 24000, giga_loss[loss=0.388, simple_loss=0.4083, pruned_loss=0.1838, over 23570.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3883, pruned_loss=0.1381, over 5617000.72 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.354, pruned_loss=0.105, over 5704948.48 frames. ], giga_tot_loss[loss=0.3352, simple_loss=0.3905, pruned_loss=0.1399, over 5606522.14 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:51:14,837 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 00:51:21,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9557, 3.6660, 3.5736, 1.7779], device='cuda:1'), covar=tensor([0.0882, 0.1069, 0.1314, 0.2544], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.1186, 0.1004, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 00:51:23,047 INFO [train.py:1012] (1/2) Epoch 27, validation: loss=0.2026, simple_loss=0.3107, pruned_loss=0.04727, over 944034.00 frames. 
+2023-03-14 00:51:23,048 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 00:51:30,375 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 24050, giga_loss[loss=0.2828, simple_loss=0.3534, pruned_loss=0.1061, over 28714.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3864, pruned_loss=0.137, over 5639327.51 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3539, pruned_loss=0.105, over 5710365.92 frames. ], giga_tot_loss[loss=0.3339, simple_loss=0.3891, pruned_loss=0.1393, over 5624254.70 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:52:28,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5549, 1.2894, 4.1588, 3.5027], device='cuda:1'), covar=tensor([0.1525, 0.2635, 0.0505, 0.0960], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0664, 0.0994, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 00:52:52,374 INFO [train.py:968] (1/2) Epoch 27, batch 24100, giga_loss[loss=0.3138, simple_loss=0.3751, pruned_loss=0.1262, over 28934.00 frames. ], tot_loss[loss=0.328, simple_loss=0.385, pruned_loss=0.1355, over 5631208.33 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3533, pruned_loss=0.1048, over 5707349.28 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3888, pruned_loss=0.1386, over 5619578.68 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:53:03,563 INFO [optim.py:369] (1/2) 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 00:53:10,009 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1208971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:53:22,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7034, 4.5630, 4.3173, 2.1090], device='cuda:1'), covar=tensor([0.0548, 0.0672, 0.0742, 0.1943], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1189, 0.1006, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 00:53:45,324 INFO [train.py:968] (1/2) Epoch 27, batch 24150, giga_loss[loss=0.3275, simple_loss=0.3853, pruned_loss=0.1348, over 27564.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3867, pruned_loss=0.136, over 5615590.29 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3538, pruned_loss=0.1052, over 5704659.50 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3896, pruned_loss=0.1385, over 5607630.69 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:54:18,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4200, 1.5348, 1.5067, 1.2775], device='cuda:1'), covar=tensor([0.3222, 0.3040, 0.2396, 0.2953], device='cuda:1'), in_proj_covar=tensor([0.2063, 0.2017, 0.1932, 0.2064], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 00:54:36,900 INFO [train.py:968] (1/2) Epoch 27, batch 24200, giga_loss[loss=0.3065, simple_loss=0.3724, pruned_loss=0.1203, over 28855.00 frames. ], tot_loss[loss=0.3309, simple_loss=0.3884, pruned_loss=0.1367, over 5624157.94 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3539, pruned_loss=0.1054, over 5708833.90 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3913, pruned_loss=0.1391, over 5612913.13 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:54:47,470 INFO [optim.py:369] (1/2) 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 00:55:27,517 INFO [train.py:968] (1/2) Epoch 27, batch 24250, giga_loss[loss=0.3125, simple_loss=0.3739, pruned_loss=0.1255, over 27909.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3857, pruned_loss=0.1342, over 5623355.78 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3542, pruned_loss=0.1056, over 5709469.64 frames. ], giga_tot_loss[loss=0.3309, simple_loss=0.3885, pruned_loss=0.1366, over 5611829.38 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:56:15,459 INFO [train.py:968] (1/2) Epoch 27, batch 24300, libri_loss[loss=0.2919, simple_loss=0.3671, pruned_loss=0.1083, over 25787.00 frames. ], tot_loss[loss=0.3216, simple_loss=0.3825, pruned_loss=0.1304, over 5631921.47 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3542, pruned_loss=0.1056, over 5710223.77 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3853, pruned_loss=0.1328, over 5620498.88 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:56:26,678 INFO [optim.py:369] (1/2) 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 00:56:58,439 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:57:06,999 INFO [train.py:968] (1/2) Epoch 27, batch 24350, giga_loss[loss=0.2862, simple_loss=0.3592, pruned_loss=0.1066, over 28596.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3793, pruned_loss=0.1273, over 5626871.58 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3542, pruned_loss=0.1057, over 5709541.18 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.3819, pruned_loss=0.1295, over 5617670.54 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:57:54,386 INFO [train.py:968] (1/2) Epoch 27, batch 24400, giga_loss[loss=0.2883, simple_loss=0.3662, pruned_loss=0.1053, over 28835.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3759, pruned_loss=0.1246, over 5636465.27 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3542, pruned_loss=0.1058, over 5712481.34 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3783, pruned_loss=0.1265, over 5625513.92 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 00:58:08,480 INFO [optim.py:369] (1/2) 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 00:58:10,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6612, 1.9772, 1.5895, 1.8429], device='cuda:1'), covar=tensor([0.2595, 0.2739, 0.3114, 0.2332], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1142, 0.1397, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 00:58:24,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2682, 1.3700, 1.2016, 1.1913], device='cuda:1'), covar=tensor([0.1958, 0.1822, 0.1681, 0.1792], device='cuda:1'), in_proj_covar=tensor([0.2071, 0.2022, 0.1941, 0.2071], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 00:58:30,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2812, 4.1304, 3.9444, 1.9342], device='cuda:1'), covar=tensor([0.0660, 0.0806, 0.0791, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1197, 0.1010, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 00:58:39,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2759, 1.6043, 1.4503, 1.3209], device='cuda:1'), covar=tensor([0.0741, 0.0334, 0.0291, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 00:58:44,107 INFO [train.py:968] (1/2) Epoch 27, batch 24450, giga_loss[loss=0.3462, simple_loss=0.4118, pruned_loss=0.1403, over 28814.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3752, pruned_loss=0.1247, over 5634773.42 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3546, pruned_loss=0.1061, over 5714212.76 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3768, pruned_loss=0.1261, over 5623912.64 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:58:48,520 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4769, 1.4368, 3.9484, 3.3220], device='cuda:1'), covar=tensor([0.1532, 0.2715, 0.0457, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0670, 0.1001, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 00:58:54,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-14 00:59:20,605 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1209346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 00:59:22,067 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-14 00:59:31,394 INFO [train.py:968] (1/2) Epoch 27, batch 24500, giga_loss[loss=0.2831, simple_loss=0.3574, pruned_loss=0.1045, over 28447.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3745, pruned_loss=0.1243, over 5638196.26 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3545, pruned_loss=0.1062, over 5710452.13 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3767, pruned_loss=0.126, over 5630992.13 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 00:59:42,598 INFO [optim.py:369] (1/2) 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 01:00:20,880 INFO [train.py:968] (1/2) Epoch 27, batch 24550, giga_loss[loss=0.3106, simple_loss=0.3726, pruned_loss=0.1243, over 28979.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3732, pruned_loss=0.1232, over 5646867.92 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3541, pruned_loss=0.106, over 5715403.14 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3759, pruned_loss=0.1252, over 5634411.49 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:01:01,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2325, 3.0718, 2.9300, 1.4175], device='cuda:1'), covar=tensor([0.1045, 0.1067, 0.1021, 0.2399], device='cuda:1'), in_proj_covar=tensor([0.1301, 0.1199, 0.1011, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 01:01:12,786 INFO [train.py:968] (1/2) Epoch 27, batch 24600, giga_loss[loss=0.2992, simple_loss=0.3832, pruned_loss=0.1076, over 27983.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3711, pruned_loss=0.1199, over 5658341.87 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3546, pruned_loss=0.1064, over 5715788.59 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3731, pruned_loss=0.1215, over 5647043.77 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:01:16,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7974, 1.9533, 1.9700, 1.5516], device='cuda:1'), covar=tensor([0.1999, 0.2559, 0.1639, 0.1957], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0711, 0.0971, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0016, 0.0015], device='cuda:1')
+2023-03-14 01:01:24,257 INFO [optim.py:369] (1/2) 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 01:01:47,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1209489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:01:49,012 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 24650, giga_loss[loss=0.3419, simple_loss=0.4121, pruned_loss=0.1358, over 28976.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3731, pruned_loss=0.1191, over 5666486.12 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3543, pruned_loss=0.1064, over 5716926.29 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3753, pruned_loss=0.1206, over 5655505.66 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:02:20,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1209521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:02:27,710 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.26 vs. limit=5.0
+2023-03-14 01:02:51,043 INFO [train.py:968] (1/2) Epoch 27, batch 24700, giga_loss[loss=0.317, simple_loss=0.3845, pruned_loss=0.1247, over 28725.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3739, pruned_loss=0.1197, over 5658667.86 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3544, pruned_loss=0.1065, over 5720802.48 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3763, pruned_loss=0.1213, over 5644168.58 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:02:58,056 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-14 01:03:00,790 INFO [optim.py:369] (1/2) 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 01:03:04,602 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1209571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:03:13,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3204, 1.7214, 1.4794, 1.4641], device='cuda:1'), covar=tensor([0.0736, 0.0377, 0.0321, 0.0815], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 01:03:35,604 INFO [train.py:968] (1/2) Epoch 27, batch 24750, giga_loss[loss=0.3422, simple_loss=0.4001, pruned_loss=0.1422, over 28583.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3736, pruned_loss=0.1198, over 5676099.96 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3542, pruned_loss=0.1064, over 5725271.62 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3764, pruned_loss=0.1216, over 5658821.11 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:03:46,753 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-14 01:04:23,912 INFO [train.py:968] (1/2) Epoch 27, batch 24800, giga_loss[loss=0.3223, simple_loss=0.382, pruned_loss=0.1314, over 29041.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3735, pruned_loss=0.1201, over 5683720.91 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3548, pruned_loss=0.1069, over 5720769.51 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3757, pruned_loss=0.1215, over 5672410.53 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:04:31,842 INFO [optim.py:369] (1/2) 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 01:04:39,603 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.40 vs. limit=5.0
+2023-03-14 01:05:08,074 INFO [train.py:968] (1/2) Epoch 27, batch 24850, giga_loss[loss=0.3275, simple_loss=0.3857, pruned_loss=0.1346, over 28315.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3726, pruned_loss=0.1206, over 5673188.19 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3559, pruned_loss=0.1078, over 5714914.26 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.374, pruned_loss=0.1213, over 5667581.78 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:05:15,411 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1209714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:05:17,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1209717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:05:21,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2286, 4.0806, 3.8985, 1.9590], device='cuda:1'), covar=tensor([0.0569, 0.0681, 0.0678, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1299, 0.1197, 0.1010, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 01:05:35,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-14 01:05:38,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4376, 1.5981, 1.6459, 1.2531], device='cuda:1'), covar=tensor([0.1783, 0.2647, 0.1484, 0.1717], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0713, 0.0973, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 01:05:45,199 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 24900, giga_loss[loss=0.252, simple_loss=0.3382, pruned_loss=0.08288, over 28883.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3704, pruned_loss=0.1197, over 5670496.25 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3558, pruned_loss=0.1077, over 5714960.18 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3718, pruned_loss=0.1205, over 5665541.02 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:06:03,071 INFO [optim.py:369] (1/2) 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 01:06:29,765 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1209798.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:06:29,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9662, 2.0680, 2.1249, 1.7117], device='cuda:1'), covar=tensor([0.1937, 0.2466, 0.1528, 0.1801], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0713, 0.0973, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 01:06:34,462 INFO [train.py:968] (1/2) Epoch 27, batch 24950, giga_loss[loss=0.3215, simple_loss=0.3923, pruned_loss=0.1254, over 28984.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3706, pruned_loss=0.1187, over 5681859.69 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3559, pruned_loss=0.1078, over 5719777.38 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3719, pruned_loss=0.1195, over 5672541.16 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:07:21,760 INFO [train.py:968] (1/2) Epoch 27, batch 25000, giga_loss[loss=0.3299, simple_loss=0.3882, pruned_loss=0.1358, over 28018.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3707, pruned_loss=0.1179, over 5690166.49 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.356, pruned_loss=0.1081, over 5724682.58 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.372, pruned_loss=0.1185, over 5677270.41 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:07:34,091 INFO [optim.py:369] (1/2) 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 01:07:44,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4991, 3.7374, 1.7615, 1.5973], device='cuda:1'), covar=tensor([0.1020, 0.0424, 0.0889, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0570, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 01:08:09,252 INFO [train.py:968] (1/2) Epoch 27, batch 25050, giga_loss[loss=0.2777, simple_loss=0.3528, pruned_loss=0.1013, over 28775.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3708, pruned_loss=0.1183, over 5676545.18 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3564, pruned_loss=0.1083, over 5716036.74 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3717, pruned_loss=0.1188, over 5672986.10 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:08:57,109 INFO [train.py:968] (1/2) Epoch 27, batch 25100, giga_loss[loss=0.2811, simple_loss=0.3471, pruned_loss=0.1076, over 28584.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3694, pruned_loss=0.1178, over 5680005.34 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3564, pruned_loss=0.1084, over 5717951.49 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3702, pruned_loss=0.1182, over 5675028.64 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:09:05,326 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:09:48,695 INFO [train.py:968] (1/2) Epoch 27, batch 25150, giga_loss[loss=0.2868, simple_loss=0.3558, pruned_loss=0.1089, over 28864.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3681, pruned_loss=0.1178, over 5667515.45 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3565, pruned_loss=0.1085, over 5721994.40 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3689, pruned_loss=0.1182, over 5659534.00 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:10:34,546 INFO [train.py:968] (1/2) Epoch 27, batch 25200, giga_loss[loss=0.2712, simple_loss=0.3355, pruned_loss=0.1034, over 28479.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3679, pruned_loss=0.1184, over 5674107.91 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3567, pruned_loss=0.1086, over 5727071.34 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3686, pruned_loss=0.1188, over 5661915.50 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 01:10:49,784 INFO [optim.py:369] (1/2) 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 01:11:04,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4747, 1.6011, 3.5003, 3.2471], device='cuda:1'), covar=tensor([0.1280, 0.2356, 0.0455, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0670, 0.1002, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 01:11:21,372 INFO [train.py:968] (1/2) Epoch 27, batch 25250, giga_loss[loss=0.3447, simple_loss=0.3954, pruned_loss=0.147, over 28914.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.367, pruned_loss=0.1183, over 5670545.67 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3567, pruned_loss=0.1087, over 5720431.25 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3678, pruned_loss=0.1187, over 5665597.50 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:11:47,833 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 01:11:56,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9589, 3.1180, 1.9683, 0.9927], device='cuda:1'), covar=tensor([0.9371, 0.3453, 0.4498, 0.8570], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1733, 0.1661, 0.1499], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 01:12:06,289 INFO [train.py:968] (1/2) Epoch 27, batch 25300, giga_loss[loss=0.3127, simple_loss=0.3826, pruned_loss=0.1214, over 28900.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3654, pruned_loss=0.1175, over 5670921.74 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3571, pruned_loss=0.109, over 5715828.25 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3659, pruned_loss=0.1177, over 5669731.47 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:12:21,143 INFO [optim.py:369] (1/2) 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 01:12:25,786 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1210173.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:12:59,172 INFO [train.py:968] (1/2) Epoch 27, batch 25350, giga_loss[loss=0.3009, simple_loss=0.3765, pruned_loss=0.1127, over 28578.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3658, pruned_loss=0.1185, over 5665294.66 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3571, pruned_loss=0.1091, over 5718612.16 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3663, pruned_loss=0.1187, over 5660993.71 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:13:05,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 01:13:44,192 INFO [train.py:968] (1/2) Epoch 27, batch 25400, libri_loss[loss=0.2717, simple_loss=0.3492, pruned_loss=0.09712, over 28556.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3666, pruned_loss=0.1185, over 5667015.53 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3569, pruned_loss=0.109, over 5721714.83 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3674, pruned_loss=0.1191, over 5659298.05 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:13:56,961 INFO [optim.py:369] (1/2) 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 01:14:29,015 INFO [train.py:968] (1/2) Epoch 27, batch 25450, giga_loss[loss=0.2933, simple_loss=0.359, pruned_loss=0.1138, over 28488.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3671, pruned_loss=0.1181, over 5665919.45 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3575, pruned_loss=0.1094, over 5716226.71 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3675, pruned_loss=0.1184, over 5663290.93 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:14:42,302 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1210316.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:14:44,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1210319.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:15:00,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1210335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:15:12,689 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1210348.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:15:18,067 INFO [train.py:968] (1/2) Epoch 27, batch 25500, giga_loss[loss=0.3086, simple_loss=0.3766, pruned_loss=0.1203, over 28665.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.367, pruned_loss=0.1176, over 5663436.27 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3574, pruned_loss=0.1094, over 5719072.95 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3676, pruned_loss=0.118, over 5658045.07 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:15:20,292 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.75 vs. limit=2.0
+2023-03-14 01:15:33,073 INFO [optim.py:369] (1/2) 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 01:15:56,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 01:16:01,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-14 01:16:03,463 INFO [train.py:968] (1/2) Epoch 27, batch 25550, giga_loss[loss=0.2978, simple_loss=0.3649, pruned_loss=0.1154, over 28960.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3692, pruned_loss=0.1195, over 5658826.42 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3585, pruned_loss=0.1103, over 5709762.88 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3687, pruned_loss=0.1191, over 5662613.80 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:16:54,484 INFO [train.py:968] (1/2) Epoch 27, batch 25600, giga_loss[loss=0.3611, simple_loss=0.4164, pruned_loss=0.1529, over 28318.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3723, pruned_loss=0.1224, over 5648544.73 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3584, pruned_loss=0.1101, over 5713303.56 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3723, pruned_loss=0.1224, over 5647764.11 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:17:07,165 INFO [optim.py:369] (1/2) 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 01:17:16,009 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1210478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:17:18,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1210481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:17:42,968 INFO [train.py:968] (1/2) Epoch 27, batch 25650, giga_loss[loss=0.31, simple_loss=0.371, pruned_loss=0.1245, over 28934.00 frames. ], tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.123, over 5657994.43 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3586, pruned_loss=0.1102, over 5717671.75 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.372, pruned_loss=0.1232, over 5652071.03 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:17:48,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1210510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:18:32,140 INFO [train.py:968] (1/2) Epoch 27, batch 25700, giga_loss[loss=0.3694, simple_loss=0.4073, pruned_loss=0.1657, over 27884.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.374, pruned_loss=0.1256, over 5648372.75 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3588, pruned_loss=0.1103, over 5702675.45 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3741, pruned_loss=0.126, over 5655750.70 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:18:49,916 INFO [optim.py:369] (1/2) 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 01:19:22,737 INFO [train.py:968] (1/2) Epoch 27, batch 25750, libri_loss[loss=0.2348, simple_loss=0.3079, pruned_loss=0.0809, over 28623.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3745, pruned_loss=0.1264, over 5643393.80 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3587, pruned_loss=0.1103, over 5702991.10 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3749, pruned_loss=0.1269, over 5648036.29 frames. ], batch size: 63, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:19:39,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4515, 2.1510, 1.7162, 0.7742], device='cuda:1'), covar=tensor([0.6477, 0.3628, 0.3963, 0.7020], device='cuda:1'), in_proj_covar=tensor([0.1838, 0.1731, 0.1659, 0.1499], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 01:20:05,857 INFO [train.py:968] (1/2) Epoch 27, batch 25800, giga_loss[loss=0.3215, simple_loss=0.3852, pruned_loss=0.129, over 28561.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3739, pruned_loss=0.126, over 5657320.53 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3591, pruned_loss=0.1107, over 5707718.02 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3744, pruned_loss=0.1265, over 5654901.91 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:20:20,674 INFO [optim.py:369] (1/2) 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 01:20:48,978 INFO [train.py:968] (1/2) Epoch 27, batch 25850, giga_loss[loss=0.3028, simple_loss=0.3724, pruned_loss=0.1166, over 28964.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3738, pruned_loss=0.1248, over 5664077.87 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3597, pruned_loss=0.111, over 5713246.28 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3741, pruned_loss=0.1254, over 5655832.28 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:21:38,076 INFO [train.py:968] (1/2) Epoch 27, batch 25900, giga_loss[loss=0.256, simple_loss=0.3058, pruned_loss=0.1031, over 23904.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3708, pruned_loss=0.1215, over 5661288.63 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3598, pruned_loss=0.1111, over 5714248.07 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.371, pruned_loss=0.122, over 5653605.22 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:21:54,835 INFO [optim.py:369] (1/2) 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 01:22:26,763 INFO [train.py:968] (1/2) Epoch 27, batch 25950, giga_loss[loss=0.2646, simple_loss=0.3353, pruned_loss=0.09692, over 28803.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3696, pruned_loss=0.121, over 5663753.39 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3598, pruned_loss=0.1112, over 5715231.83 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3698, pruned_loss=0.1213, over 5656621.64 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:22:46,071 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-14 01:22:55,387 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1210836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:23:10,607 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5575, 1.8751, 1.4892, 1.5112], device='cuda:1'), covar=tensor([0.2605, 0.2620, 0.2959, 0.2430], device='cuda:1'), in_proj_covar=tensor([0.1584, 0.1142, 0.1399, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 01:23:11,581 INFO [train.py:968] (1/2) Epoch 27, batch 26000, libri_loss[loss=0.2527, simple_loss=0.331, pruned_loss=0.08716, over 29556.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3665, pruned_loss=0.1192, over 5666769.22 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3602, pruned_loss=0.1115, over 5708479.45 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3666, pruned_loss=0.1194, over 5665469.77 frames. ], batch size: 74, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:23:14,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6670, 1.9086, 1.4681, 1.5895], device='cuda:1'), covar=tensor([0.2307, 0.2312, 0.2541, 0.2399], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0760, 0.0731, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 01:23:26,954 INFO [optim.py:369] (1/2) 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 01:23:41,626 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-14 01:23:51,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4600, 1.7105, 1.2472, 1.2952], device='cuda:1'), covar=tensor([0.1060, 0.0603, 0.1088, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0449, 0.0521, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 01:23:57,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6011, 1.5894, 1.7748, 1.3755], device='cuda:1'), covar=tensor([0.1604, 0.2422, 0.1333, 0.1605], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0716, 0.0975, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 01:24:01,433 INFO [train.py:968] (1/2) Epoch 27, batch 26050, giga_loss[loss=0.3103, simple_loss=0.3744, pruned_loss=0.1231, over 28253.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1183, over 5660658.77 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3602, pruned_loss=0.1117, over 5698141.59 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3657, pruned_loss=0.1185, over 5667488.23 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:24:19,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7828, 1.9733, 1.5874, 2.1635], device='cuda:1'), covar=tensor([0.2703, 0.2872, 0.3269, 0.2444], device='cuda:1'), in_proj_covar=tensor([0.1582, 0.1140, 0.1398, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 01:24:48,747 INFO [train.py:968] (1/2) Epoch 27, batch 26100, giga_loss[loss=0.2796, simple_loss=0.3567, pruned_loss=0.1012, over 28792.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3679, pruned_loss=0.1187, over 5668298.88 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3601, pruned_loss=0.1116, over 5699205.88 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3681, pruned_loss=0.119, over 5672621.11 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:25:05,043 INFO [optim.py:369] (1/2) 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 01:25:37,283 INFO [train.py:968] (1/2) Epoch 27, batch 26150, giga_loss[loss=0.2957, simple_loss=0.3739, pruned_loss=0.1087, over 28819.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3704, pruned_loss=0.1171, over 5672209.65 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3599, pruned_loss=0.1115, over 5702069.23 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3709, pruned_loss=0.1175, over 5672939.87 frames. ], batch size: 243, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:25:46,843 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-14 01:25:54,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-14 01:26:19,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0381, 1.7310, 1.5442, 1.3616], device='cuda:1'), covar=tensor([0.1755, 0.1207, 0.1772, 0.1588], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0758, 0.0728, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 01:26:22,520 INFO [train.py:968] (1/2) Epoch 27, batch 26200, giga_loss[loss=0.3151, simple_loss=0.3815, pruned_loss=0.1244, over 28284.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3712, pruned_loss=0.1174, over 5660665.42 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.36, pruned_loss=0.1118, over 5688381.67 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3719, pruned_loss=0.1176, over 5673397.13 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:26:27,450 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:27:10,362 INFO [train.py:968] (1/2) Epoch 27, batch 26250, giga_loss[loss=0.3476, simple_loss=0.4018, pruned_loss=0.1467, over 28291.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3743, pruned_loss=0.1204, over 5670507.97 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.36, pruned_loss=0.1119, over 5691027.47 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3752, pruned_loss=0.1207, over 5677952.05 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:27:20,698 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:27:24,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-14 01:27:51,042 INFO [train.py:968] (1/2) Epoch 27, batch 26300, giga_loss[loss=0.371, simple_loss=0.4161, pruned_loss=0.163, over 27537.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3747, pruned_loss=0.1211, over 5682765.22 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3601, pruned_loss=0.112, over 5697393.43 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3757, pruned_loss=0.1214, over 5682753.69 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:28:09,360 INFO [optim.py:369] (1/2) 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 01:28:21,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5246, 3.0675, 1.6411, 1.6289], device='cuda:1'), covar=tensor([0.0802, 0.0336, 0.0718, 0.1118], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0571, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 01:28:41,140 INFO [train.py:968] (1/2) Epoch 27, batch 26350, giga_loss[loss=0.2815, simple_loss=0.3538, pruned_loss=0.1046, over 28682.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3753, pruned_loss=0.1229, over 5675191.44 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3603, pruned_loss=0.1121, over 5697686.26 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3761, pruned_loss=0.1232, over 5674825.33 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:28:46,952 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:29:01,270 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2846, 1.5983, 1.2726, 0.9922], device='cuda:1'), covar=tensor([0.2427, 0.2422, 0.2789, 0.2229], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1142, 0.1400, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 01:29:26,243 INFO [train.py:968] (1/2) Epoch 27, batch 26400, giga_loss[loss=0.2809, simple_loss=0.3602, pruned_loss=0.1008, over 28979.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3735, pruned_loss=0.1221, over 5688455.29 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3601, pruned_loss=0.112, over 5702101.23 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3747, pruned_loss=0.1228, over 5683693.54 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:29:41,255 INFO [optim.py:369] (1/2) 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 01:29:52,187 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 01:30:04,800 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 01:30:12,320 INFO [train.py:968] (1/2) Epoch 27, batch 26450, giga_loss[loss=0.3188, simple_loss=0.3761, pruned_loss=0.1307, over 28739.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3713, pruned_loss=0.1214, over 5681662.08 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3608, pruned_loss=0.1125, over 5696778.30 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3719, pruned_loss=0.1216, over 5682654.61 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:30:17,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-14 01:30:22,626 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:30:31,780 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 01:30:32,429 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 01:30:47,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2580, 1.2145, 3.9801, 3.3540], device='cuda:1'), covar=tensor([0.1766, 0.2891, 0.0458, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0673, 0.1005, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 01:30:59,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:30:59,924 INFO [train.py:968] (1/2) Epoch 27, batch 26500, libri_loss[loss=0.3067, simple_loss=0.3698, pruned_loss=0.1218, over 28980.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3699, pruned_loss=0.1211, over 5673455.21 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3608, pruned_loss=0.1127, over 5693493.52 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3707, pruned_loss=0.1214, over 5677489.97 frames. ], batch size: 107, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:31:01,620 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:31:17,870 INFO [optim.py:369] (1/2) 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 01:31:30,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:31:31,515 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:31:37,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5087, 1.6441, 1.1725, 1.2648], device='cuda:1'), covar=tensor([0.1012, 0.0594, 0.1091, 0.1157], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0449, 0.0521, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 01:31:44,401 INFO [train.py:968] (1/2) Epoch 27, batch 26550, giga_loss[loss=0.2594, simple_loss=0.3376, pruned_loss=0.0906, over 28790.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.371, pruned_loss=0.122, over 5669113.05 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.361, pruned_loss=0.1128, over 5686845.14 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3718, pruned_loss=0.1225, over 5676912.73 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:32:12,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:32:30,670 INFO [train.py:968] (1/2) Epoch 27, batch 26600, giga_loss[loss=0.2802, simple_loss=0.3509, pruned_loss=0.1048, over 28789.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3699, pruned_loss=0.1218, over 5675822.87 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3605, pruned_loss=0.1125, over 5691434.58 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3713, pruned_loss=0.1227, over 5677409.77 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:32:31,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6655, 1.5863, 1.8826, 1.4727], device='cuda:1'), covar=tensor([0.1537, 0.2086, 0.1233, 0.1563], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0715, 0.0975, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 01:32:47,265 INFO [optim.py:369] (1/2) 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 01:33:07,623 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 26650, giga_loss[loss=0.286, simple_loss=0.3514, pruned_loss=0.1103, over 29014.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3673, pruned_loss=0.121, over 5654165.45 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3606, pruned_loss=0.1127, over 5692146.41 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3682, pruned_loss=0.1216, over 5654555.87 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:33:23,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6148, 1.9778, 1.5153, 1.5860], device='cuda:1'), covar=tensor([0.2395, 0.2282, 0.2540, 0.2480], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0761, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 01:34:06,150 INFO [train.py:968] (1/2) Epoch 27, batch 26700, giga_loss[loss=0.2886, simple_loss=0.3628, pruned_loss=0.1072, over 28613.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3672, pruned_loss=0.1204, over 5650689.85 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3608, pruned_loss=0.1128, over 5686480.92 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.368, pruned_loss=0.121, over 5655348.65 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:34:23,034 INFO [optim.py:369] (1/2) 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 01:34:27,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1736, 4.0492, 3.8144, 1.7608], device='cuda:1'), covar=tensor([0.0632, 0.0705, 0.0709, 0.2112], device='cuda:1'), in_proj_covar=tensor([0.1306, 0.1203, 0.1014, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 01:34:29,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:34:31,935 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211600.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:34:53,358 INFO [train.py:968] (1/2) Epoch 27, batch 26750, giga_loss[loss=0.3063, simple_loss=0.3812, pruned_loss=0.1157, over 28599.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3681, pruned_loss=0.12, over 5657901.17 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3606, pruned_loss=0.1127, over 5687797.31 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.369, pruned_loss=0.1206, over 5659951.64 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:34:59,321 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 01:34:59,714 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:04,407 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:04,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8985, 1.2854, 1.3252, 1.0918], device='cuda:1'), covar=tensor([0.2372, 0.1531, 0.2692, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0762, 0.0730, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 01:35:14,523 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:24,662 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-14 01:35:27,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:30,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:35:42,868 INFO [train.py:968] (1/2) Epoch 27, batch 26800, giga_loss[loss=0.3367, simple_loss=0.3994, pruned_loss=0.137, over 28767.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.37, pruned_loss=0.1217, over 5660425.05 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3608, pruned_loss=0.113, over 5694354.16 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3708, pruned_loss=0.1222, over 5654982.32 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:35:56,971 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:36:00,615 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 27, batch 26850, giga_loss[loss=0.3007, simple_loss=0.3829, pruned_loss=0.1093, over 28937.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3695, pruned_loss=0.1206, over 5671368.12 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3603, pruned_loss=0.1127, over 5696352.01 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3707, pruned_loss=0.1213, over 5664932.49 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:36:48,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:36:50,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6150, 1.8992, 1.5604, 1.6300], device='cuda:1'), covar=tensor([0.2476, 0.2376, 0.2601, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1140, 0.1398, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 01:37:13,373 INFO [train.py:968] (1/2) Epoch 27, batch 26900, giga_loss[loss=0.3299, simple_loss=0.3995, pruned_loss=0.1302, over 28693.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3701, pruned_loss=0.1181, over 5668842.36 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1124, over 5690111.58 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3717, pruned_loss=0.1191, over 5668665.77 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:37:21,827 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:37:24,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9119, 1.1402, 1.1155, 0.9100], device='cuda:1'), covar=tensor([0.2789, 0.3168, 0.1941, 0.2604], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2015, 0.1939, 0.2069], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 01:37:30,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1211770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:37:32,817 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 26950, giga_loss[loss=0.2688, simple_loss=0.3563, pruned_loss=0.09062, over 28551.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3723, pruned_loss=0.1179, over 5672073.44 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.36, pruned_loss=0.1124, over 5686425.01 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3738, pruned_loss=0.1188, over 5674662.64 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:38:14,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3484, 1.8218, 1.3790, 1.5260], device='cuda:1'), covar=tensor([0.0781, 0.0302, 0.0349, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 01:38:14,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 01:38:21,920 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:38:24,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:38:43,778 INFO [train.py:968] (1/2) Epoch 27, batch 27000, giga_loss[loss=0.2888, simple_loss=0.358, pruned_loss=0.1098, over 29011.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3756, pruned_loss=0.1201, over 5676895.81 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3602, pruned_loss=0.1126, over 5688229.22 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3767, pruned_loss=0.1207, over 5677350.55 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:38:43,778 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 01:38:47,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8386, 3.6220, 3.4773, 1.6913], device='cuda:1'), covar=tensor([0.0848, 0.0998, 0.0970, 0.2545], device='cuda:1'), in_proj_covar=tensor([0.1314, 0.1208, 0.1018, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 01:38:52,019 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 01:38:59,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:39:02,632 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 01:39:09,459 INFO [optim.py:369] (1/2) 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 01:39:12,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3394, 1.3071, 3.5204, 3.0970], device='cuda:1'), covar=tensor([0.1500, 0.2711, 0.0497, 0.1567], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0673, 0.1005, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 01:39:40,227 INFO [train.py:968] (1/2) Epoch 27, batch 27050, giga_loss[loss=0.3069, simple_loss=0.3789, pruned_loss=0.1174, over 28375.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3789, pruned_loss=0.124, over 5674273.54 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3603, pruned_loss=0.1128, over 5692526.99 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3799, pruned_loss=0.1245, over 5670737.71 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:39:40,504 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1211905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:39:43,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1211908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:40:17,892 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1211937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:40:34,362 INFO [train.py:968] (1/2) Epoch 27, batch 27100, giga_loss[loss=0.3463, simple_loss=0.3932, pruned_loss=0.1497, over 28540.00 frames. ], tot_loss[loss=0.317, simple_loss=0.3805, pruned_loss=0.1268, over 5651770.19 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3605, pruned_loss=0.113, over 5694456.83 frames. ], giga_tot_loss[loss=0.3177, simple_loss=0.3813, pruned_loss=0.1271, over 5646647.94 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:40:51,609 INFO [optim.py:369] (1/2) 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 01:40:53,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211975.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:41:05,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1211989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:41:17,136 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:41:19,904 INFO [train.py:968] (1/2) Epoch 27, batch 27150, giga_loss[loss=0.29, simple_loss=0.3631, pruned_loss=0.1085, over 28907.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3774, pruned_loss=0.1244, over 5656493.95 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5688774.59 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3783, pruned_loss=0.1247, over 5656367.54 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:42:05,284 INFO [train.py:968] (1/2) Epoch 27, batch 27200, giga_loss[loss=0.2954, simple_loss=0.3782, pruned_loss=0.1063, over 28892.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3765, pruned_loss=0.1232, over 5634229.34 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3608, pruned_loss=0.1136, over 5672831.64 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3776, pruned_loss=0.1234, over 5648120.01 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:42:25,158 INFO [optim.py:369] (1/2) 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 01:42:51,969 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 27250, giga_loss[loss=0.2744, simple_loss=0.3549, pruned_loss=0.09694, over 28720.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3759, pruned_loss=0.1209, over 5637176.59 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3607, pruned_loss=0.1136, over 5664538.70 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3769, pruned_loss=0.1211, over 5655626.47 frames. ], batch size: 66, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:43:07,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212118.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:43:08,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212121.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 01:43:17,720 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:22,160 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:32,877 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:32,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:36,068 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:43:37,476 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212150.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 01:43:42,075 INFO [train.py:968] (1/2) Epoch 27, batch 27300, giga_loss[loss=0.3054, simple_loss=0.3757, pruned_loss=0.1176, over 28864.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3762, pruned_loss=0.1209, over 5643512.21 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3607, pruned_loss=0.1136, over 5667342.20 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3773, pruned_loss=0.1212, over 5655214.47 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:43:50,813 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 01:44:03,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:44:18,776 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 27350, giga_loss[loss=0.2869, simple_loss=0.3569, pruned_loss=0.1084, over 28710.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3759, pruned_loss=0.1209, over 5656873.05 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3605, pruned_loss=0.1135, over 5671972.37 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3772, pruned_loss=0.1214, over 5661923.82 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:44:44,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3492, 3.3537, 1.4054, 1.5103], device='cuda:1'), covar=tensor([0.1117, 0.0498, 0.0997, 0.1486], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0572, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 01:45:07,991 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:11,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:17,901 INFO [train.py:968] (1/2) Epoch 27, batch 27400, giga_loss[loss=0.2628, simple_loss=0.336, pruned_loss=0.09483, over 29020.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3742, pruned_loss=0.1202, over 5668340.11 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3606, pruned_loss=0.1135, over 5674746.98 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3753, pruned_loss=0.1207, over 5669770.37 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:45:37,127 INFO [optim.py:369] (1/2) 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 01:45:41,188 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:52,237 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:45:54,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:46:05,826 INFO [train.py:968] (1/2) Epoch 27, batch 27450, giga_loss[loss=0.3269, simple_loss=0.3881, pruned_loss=0.1328, over 28682.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3726, pruned_loss=0.1212, over 5652365.28 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5679339.07 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3739, pruned_loss=0.1219, over 5649095.64 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:46:20,951 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1212320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:46:20,964 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:46:39,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8721, 2.3545, 1.7969, 2.2047], device='cuda:1'), covar=tensor([0.2626, 0.2653, 0.3045, 0.2395], device='cuda:1'), in_proj_covar=tensor([0.1583, 0.1142, 0.1401, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 01:46:55,981 INFO [train.py:968] (1/2) Epoch 27, batch 27500, giga_loss[loss=0.2594, simple_loss=0.3389, pruned_loss=0.08994, over 28933.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3707, pruned_loss=0.1205, over 5647053.10 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3604, pruned_loss=0.1133, over 5682687.72 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3722, pruned_loss=0.1214, over 5640762.14 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:47:14,898 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 27550, giga_loss[loss=0.3928, simple_loss=0.4163, pruned_loss=0.1846, over 26539.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.369, pruned_loss=0.1201, over 5653838.61 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3604, pruned_loss=0.1134, over 5682376.14 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.1209, over 5648019.86 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:48:22,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6116, 1.8520, 1.5057, 1.6778], device='cuda:1'), covar=tensor([0.2580, 0.2683, 0.3040, 0.2440], device='cuda:1'), in_proj_covar=tensor([0.1580, 0.1140, 0.1399, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 01:48:25,325 INFO [train.py:968] (1/2) Epoch 27, batch 27600, giga_loss[loss=0.3157, simple_loss=0.3779, pruned_loss=0.1267, over 27922.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3691, pruned_loss=0.1206, over 5658043.82 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1135, over 5689605.20 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3704, pruned_loss=0.1214, over 5645951.07 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:48:45,259 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 27650, giga_loss[loss=0.2674, simple_loss=0.3466, pruned_loss=0.0941, over 28678.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3671, pruned_loss=0.1186, over 5660990.33 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1135, over 5690788.17 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3681, pruned_loss=0.1193, over 5650419.33 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:49:14,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2052, 1.4117, 1.3767, 1.1336], device='cuda:1'), covar=tensor([0.3117, 0.3015, 0.2090, 0.2721], device='cuda:1'), in_proj_covar=tensor([0.2064, 0.2024, 0.1947, 0.2077], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 01:49:25,589 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 01:49:26,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3381, 1.5623, 1.2725, 1.4741], device='cuda:1'), covar=tensor([0.0769, 0.0389, 0.0369, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 01:49:40,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7094, 1.8747, 1.3178, 1.5043], device='cuda:1'), covar=tensor([0.1012, 0.0630, 0.1066, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0449, 0.0519, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 01:49:55,583 INFO [train.py:968] (1/2) Epoch 27, batch 27700, giga_loss[loss=0.2974, simple_loss=0.3757, pruned_loss=0.1095, over 28642.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3637, pruned_loss=0.1148, over 5663477.18 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5693311.09 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3648, pruned_loss=0.1156, over 5652361.74 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:50:07,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:50:09,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4610, 2.0680, 1.4723, 0.7729], device='cuda:1'), covar=tensor([0.7080, 0.3206, 0.4517, 0.7208], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1736, 0.1663, 0.1499], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 01:50:14,768 INFO [optim.py:369] (1/2) 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 01:50:21,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3685, 1.6655, 1.5204, 1.4842], device='cuda:1'), covar=tensor([0.2162, 0.1838, 0.2481, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0760, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 01:50:32,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8089, 3.0203, 1.8847, 1.0177], device='cuda:1'), covar=tensor([1.0866, 0.3795, 0.4962, 0.8800], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1737, 0.1663, 0.1500], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 01:50:45,180 INFO [train.py:968] (1/2) Epoch 27, batch 27750, giga_loss[loss=0.2852, simple_loss=0.362, pruned_loss=0.1042, over 28793.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3624, pruned_loss=0.1135, over 5667547.08 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.113, over 5697079.00 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3637, pruned_loss=0.1144, over 5654790.53 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:50:54,237 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4518, 1.5841, 1.4980, 1.4317], device='cuda:1'), covar=tensor([0.2468, 0.2228, 0.2297, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.2064, 0.2023, 0.1944, 0.2075], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 01:51:32,729 INFO [train.py:968] (1/2) Epoch 27, batch 27800, giga_loss[loss=0.2876, simple_loss=0.3509, pruned_loss=0.1121, over 29050.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.361, pruned_loss=0.1134, over 5658438.39 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1126, over 5700349.81 frames. ], giga_tot_loss[loss=0.2956, simple_loss=0.3624, pruned_loss=0.1144, over 5644150.89 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:51:54,610 INFO [optim.py:369] (1/2) 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 01:52:13,762 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1212695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:52:21,594 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5387, 1.9759, 1.5380, 1.6225], device='cuda:1'), covar=tensor([0.0793, 0.0291, 0.0327, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 01:52:23,183 INFO [train.py:968] (1/2) Epoch 27, batch 27850, giga_loss[loss=0.2705, simple_loss=0.3382, pruned_loss=0.1015, over 28855.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3576, pruned_loss=0.1117, over 5658782.32 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.1131, over 5686864.83 frames. ], giga_tot_loss[loss=0.2913, simple_loss=0.3583, pruned_loss=0.1122, over 5658513.85 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:52:29,337 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:52:32,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:52:53,381 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 01:52:58,756 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1212743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:53:00,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.48 vs. limit=2.0
+2023-03-14 01:53:07,622 INFO [train.py:968] (1/2) Epoch 27, batch 27900, libri_loss[loss=0.3122, simple_loss=0.3831, pruned_loss=0.1207, over 28622.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3599, pruned_loss=0.1133, over 5658896.51 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.113, over 5684240.78 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3605, pruned_loss=0.1137, over 5659345.08 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:53:16,063 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-14 01:53:24,692 INFO [optim.py:369] (1/2) 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 01:53:52,912 INFO [train.py:968] (1/2) Epoch 27, batch 27950, giga_loss[loss=0.2873, simple_loss=0.3608, pruned_loss=0.1069, over 28693.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3635, pruned_loss=0.1156, over 5647280.64 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 5681149.17 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3639, pruned_loss=0.1158, over 5649738.22 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 01:54:16,115 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.32 vs. limit=2.0
+2023-03-14 01:54:23,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1212838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:54:25,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1212841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:54:40,123 INFO [train.py:968] (1/2) Epoch 27, batch 28000, giga_loss[loss=0.3091, simple_loss=0.3767, pruned_loss=0.1208, over 28694.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3652, pruned_loss=0.1164, over 5648075.87 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.36, pruned_loss=0.1129, over 5686645.39 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3657, pruned_loss=0.1168, over 5644527.29 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:54:55,327 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 27, batch 28050, giga_loss[loss=0.2735, simple_loss=0.3449, pruned_loss=0.1011, over 28947.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3657, pruned_loss=0.1172, over 5654321.01 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3595, pruned_loss=0.1125, over 5691040.71 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3668, pruned_loss=0.118, over 5646096.13 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:56:11,126 INFO [train.py:968] (1/2) Epoch 27, batch 28100, giga_loss[loss=0.2988, simple_loss=0.366, pruned_loss=0.1157, over 28941.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3663, pruned_loss=0.118, over 5660531.04 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5695198.39 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3674, pruned_loss=0.1189, over 5649323.45 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:56:29,937 INFO [optim.py:369] (1/2) 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 01:56:45,787 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 28150, libri_loss[loss=0.2516, simple_loss=0.3203, pruned_loss=0.09144, over 29353.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3677, pruned_loss=0.1188, over 5665183.39 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5698420.81 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.369, pruned_loss=0.1197, over 5652760.69 frames. ], batch size: 67, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:57:06,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6303, 1.7487, 1.7960, 1.3932], device='cuda:1'), covar=tensor([0.1982, 0.2650, 0.1623, 0.1872], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0717, 0.0977, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 01:57:36,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2004, 0.9023, 0.9227, 1.3795], device='cuda:1'), covar=tensor([0.0724, 0.0419, 0.0361, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0122, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 01:57:43,228 INFO [train.py:968] (1/2) Epoch 27, batch 28200, giga_loss[loss=0.3175, simple_loss=0.3847, pruned_loss=0.1251, over 28974.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3701, pruned_loss=0.1196, over 5667539.37 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 5697988.42 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3708, pruned_loss=0.12, over 5657576.41 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:58:00,473 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 01:58:05,015 INFO [optim.py:369] (1/2) 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 01:58:33,163 INFO [train.py:968] (1/2) Epoch 27, batch 28250, giga_loss[loss=0.3258, simple_loss=0.3843, pruned_loss=0.1336, over 28681.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.372, pruned_loss=0.1217, over 5659690.64 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3598, pruned_loss=0.1127, over 5698438.10 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3729, pruned_loss=0.1223, over 5650576.80 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:58:56,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4890, 3.5830, 1.6676, 1.5601], device='cuda:1'), covar=tensor([0.0975, 0.0357, 0.0865, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0571, 0.0407, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 01:59:21,296 INFO [train.py:968] (1/2) Epoch 27, batch 28300, giga_loss[loss=0.2913, simple_loss=0.3553, pruned_loss=0.1137, over 28689.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3738, pruned_loss=0.1239, over 5640946.45 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1131, over 5687413.13 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3741, pruned_loss=0.1242, over 5642842.00 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 01:59:41,102 INFO [optim.py:369] (1/2) 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 02:00:03,955 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 28350, giga_loss[loss=0.3076, simple_loss=0.3748, pruned_loss=0.1202, over 28600.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3735, pruned_loss=0.1218, over 5647825.54 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3599, pruned_loss=0.1129, over 5682993.07 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3747, pruned_loss=0.1225, over 5652322.13 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:00:31,145 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2127, 1.5241, 1.5564, 1.3233], device='cuda:1'), covar=tensor([0.2158, 0.1752, 0.2417, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0756, 0.0727, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 02:00:58,482 INFO [train.py:968] (1/2) Epoch 27, batch 28400, giga_loss[loss=0.2746, simple_loss=0.3399, pruned_loss=0.1047, over 28611.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3721, pruned_loss=0.1199, over 5655296.79 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1128, over 5683479.69 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3734, pruned_loss=0.1207, over 5658046.96 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:01:17,796 INFO [optim.py:369] (1/2) 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 02:01:45,488 INFO [train.py:968] (1/2) Epoch 27, batch 28450, giga_loss[loss=0.2952, simple_loss=0.3664, pruned_loss=0.112, over 28644.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3715, pruned_loss=0.1207, over 5643698.88 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1131, over 5669176.91 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3724, pruned_loss=0.1212, over 5659248.10 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:02:39,914 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 28500, giga_loss[loss=0.3104, simple_loss=0.3671, pruned_loss=0.1269, over 28900.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3703, pruned_loss=0.1205, over 5659208.79 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3602, pruned_loss=0.1133, over 5676391.88 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3713, pruned_loss=0.121, over 5664567.03 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:02:57,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:03:07,130 INFO [optim.py:369] (1/2) 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 02:03:36,076 INFO [train.py:968] (1/2) Epoch 27, batch 28550, giga_loss[loss=0.3011, simple_loss=0.3673, pruned_loss=0.1175, over 28591.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3675, pruned_loss=0.1185, over 5668691.04 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3602, pruned_loss=0.1132, over 5678913.97 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3683, pruned_loss=0.1191, over 5670394.04 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:04:15,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:04:16,166 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 02:04:17,802 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4514, 4.1516, 1.6505, 1.6037], device='cuda:1'), covar=tensor([0.1054, 0.0333, 0.0908, 0.1364], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0571, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 02:04:25,914 INFO [train.py:968] (1/2) Epoch 27, batch 28600, giga_loss[loss=0.3847, simple_loss=0.4222, pruned_loss=0.1736, over 27587.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3684, pruned_loss=0.1199, over 5661549.97 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3605, pruned_loss=0.1133, over 5671843.15 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.369, pruned_loss=0.1202, over 5669065.08 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:04:45,173 INFO [optim.py:369] (1/2) 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 02:05:12,482 INFO [train.py:968] (1/2) Epoch 27, batch 28650, giga_loss[loss=0.3109, simple_loss=0.3756, pruned_loss=0.1232, over 28544.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1205, over 5656034.20 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3606, pruned_loss=0.1134, over 5677105.86 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3691, pruned_loss=0.1209, over 5656927.79 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:05:18,769 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:05:21,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:05:27,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3537, 1.5383, 1.6193, 1.3622], device='cuda:1'), covar=tensor([0.1965, 0.1934, 0.2183, 0.2093], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0758, 0.0728, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 02:05:47,413 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:05:59,715 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:06:00,941 INFO [train.py:968] (1/2) Epoch 27, batch 28700, giga_loss[loss=0.2868, simple_loss=0.3599, pruned_loss=0.1068, over 28716.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3689, pruned_loss=0.1209, over 5651955.29 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3607, pruned_loss=0.1134, over 5675621.81 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3693, pruned_loss=0.1213, over 5653877.42 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:06:19,684 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:06:30,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4252, 1.8166, 1.3943, 1.4357], device='cuda:1'), covar=tensor([0.2723, 0.2794, 0.3305, 0.2516], device='cuda:1'), in_proj_covar=tensor([0.1584, 0.1142, 0.1403, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 02:06:35,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:06:37,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:06:49,235 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 02:06:50,693 INFO [train.py:968] (1/2) Epoch 27, batch 28750, giga_loss[loss=0.2891, simple_loss=0.3638, pruned_loss=0.1072, over 28888.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3717, pruned_loss=0.1238, over 5641043.72 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3612, pruned_loss=0.1138, over 5670500.64 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3717, pruned_loss=0.1239, over 5646387.66 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:07:06,799 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:07:38,847 INFO [train.py:968] (1/2) Epoch 27, batch 28800, giga_loss[loss=0.3034, simple_loss=0.3662, pruned_loss=0.1203, over 28342.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3728, pruned_loss=0.1245, over 5644614.39 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3614, pruned_loss=0.114, over 5674218.79 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3728, pruned_loss=0.1246, over 5645170.63 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:08:02,059 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 28850, giga_loss[loss=0.2672, simple_loss=0.3355, pruned_loss=0.09949, over 28530.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3736, pruned_loss=0.1256, over 5639544.38 frames. ], libri_tot_loss[loss=0.2962, simple_loss=0.3626, pruned_loss=0.1149, over 5672252.02 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3729, pruned_loss=0.1253, over 5641216.71 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:08:35,334 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:08:37,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:08:44,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:09:03,840 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:09:08,846 INFO [train.py:968] (1/2) Epoch 27, batch 28900, giga_loss[loss=0.2755, simple_loss=0.3498, pruned_loss=0.1006, over 28257.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3735, pruned_loss=0.1263, over 5650432.46 frames. ], libri_tot_loss[loss=0.2958, simple_loss=0.3622, pruned_loss=0.1147, over 5678034.14 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3737, pruned_loss=0.1265, over 5645956.56 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:09:16,820 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3004, 0.8056, 0.9509, 1.5127], device='cuda:1'), covar=tensor([0.0794, 0.0387, 0.0368, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0122, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 02:09:28,867 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 28950, giga_loss[loss=0.3342, simple_loss=0.3964, pruned_loss=0.136, over 28716.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3729, pruned_loss=0.1255, over 5639093.44 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3624, pruned_loss=0.1149, over 5674571.09 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.373, pruned_loss=0.1257, over 5637553.21 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:09:56,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0345, 2.0290, 2.2115, 1.7915], device='cuda:1'), covar=tensor([0.1909, 0.2479, 0.1510, 0.1710], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0719, 0.0977, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 02:10:04,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.74 vs. limit=2.0
+2023-03-14 02:10:23,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2212, 2.5601, 2.0462, 1.9835], device='cuda:1'), covar=tensor([0.1774, 0.1520, 0.1857, 0.1770], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0759, 0.0730, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 02:10:38,660 INFO [train.py:968] (1/2) Epoch 27, batch 29000, giga_loss[loss=0.2896, simple_loss=0.3583, pruned_loss=0.1105, over 28651.00 frames. ], tot_loss[loss=0.3117, simple_loss=0.3735, pruned_loss=0.1249, over 5636459.88 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3631, pruned_loss=0.1154, over 5668769.86 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3734, pruned_loss=0.125, over 5639534.81 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:10:50,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4072, 3.3475, 1.5537, 1.4743], device='cuda:1'), covar=tensor([0.0944, 0.0402, 0.0895, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0571, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 02:10:51,274 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1213871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:10:52,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9102, 1.1566, 2.8367, 2.7246], device='cuda:1'), covar=tensor([0.1586, 0.2542, 0.0597, 0.1845], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0673, 0.1007, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 02:10:54,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1213874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:10:56,966 INFO [optim.py:369] (1/2) 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 02:11:20,357 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1213903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:11:21,506 INFO [train.py:968] (1/2) Epoch 27, batch 29050, giga_loss[loss=0.2644, simple_loss=0.336, pruned_loss=0.09643, over 28527.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.373, pruned_loss=0.1243, over 5650942.87 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1152, over 5675719.42 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3739, pruned_loss=0.1249, over 5646156.57 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:11:31,739 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1213916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:11:42,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1213928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:11:45,613 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 29100, giga_loss[loss=0.3043, simple_loss=0.3723, pruned_loss=0.1182, over 28741.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3734, pruned_loss=0.1244, over 5657439.66 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1152, over 5670379.08 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3742, pruned_loss=0.1251, over 5658191.33 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:12:25,206 INFO [optim.py:369] (1/2) 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 02:12:49,943 INFO [train.py:968] (1/2) Epoch 27, batch 29150, giga_loss[loss=0.2882, simple_loss=0.3674, pruned_loss=0.1045, over 28671.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3758, pruned_loss=0.1261, over 5668372.57 frames. ], libri_tot_loss[loss=0.2969, simple_loss=0.363, pruned_loss=0.1154, over 5673364.60 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3762, pruned_loss=0.1266, over 5666213.77 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:13:33,169 INFO [train.py:968] (1/2) Epoch 27, batch 29200, libri_loss[loss=0.3294, simple_loss=0.3818, pruned_loss=0.1385, over 19154.00 frames. ], tot_loss[loss=0.3165, simple_loss=0.3777, pruned_loss=0.1277, over 5651559.48 frames. ], libri_tot_loss[loss=0.2973, simple_loss=0.3633, pruned_loss=0.1156, over 5663068.06 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3783, pruned_loss=0.1284, over 5659114.72 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:13:48,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:13:50,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:13:53,005 INFO [optim.py:369] (1/2) 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 02:13:54,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3777, 1.6933, 1.6315, 1.5401], device='cuda:1'), covar=tensor([0.2020, 0.1954, 0.2213, 0.1995], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0760, 0.0731, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 02:14:18,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1214103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:14:20,716 INFO [train.py:968] (1/2) Epoch 27, batch 29250, giga_loss[loss=0.3491, simple_loss=0.4, pruned_loss=0.1491, over 27607.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3765, pruned_loss=0.1252, over 5661050.80 frames. ], libri_tot_loss[loss=0.2972, simple_loss=0.3635, pruned_loss=0.1155, over 5671030.03 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3774, pruned_loss=0.1262, over 5659858.68 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:15:05,126 INFO [train.py:968] (1/2) Epoch 27, batch 29300, giga_loss[loss=0.2687, simple_loss=0.3438, pruned_loss=0.09678, over 28696.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3748, pruned_loss=0.1233, over 5665774.66 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.363, pruned_loss=0.1152, over 5675909.27 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3763, pruned_loss=0.1246, over 5660131.85 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:15:05,529 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:15:41,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.04 vs. limit=2.0
+2023-03-14 02:15:49,683 INFO [train.py:968] (1/2) Epoch 27, batch 29350, giga_loss[loss=0.3284, simple_loss=0.3881, pruned_loss=0.1344, over 28533.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3726, pruned_loss=0.1223, over 5653021.76 frames. ], libri_tot_loss[loss=0.2975, simple_loss=0.3633, pruned_loss=0.1158, over 5666988.67 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3737, pruned_loss=0.123, over 5656962.35 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:16:32,475 INFO [train.py:968] (1/2) Epoch 27, batch 29400, giga_loss[loss=0.2665, simple_loss=0.3448, pruned_loss=0.09414, over 28993.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3729, pruned_loss=0.1227, over 5647673.57 frames. ], libri_tot_loss[loss=0.2977, simple_loss=0.3635, pruned_loss=0.116, over 5663016.01 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3738, pruned_loss=0.1233, over 5654546.94 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:16:56,005 INFO [optim.py:369] (1/2) 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 02:17:07,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:17:20,330 INFO [train.py:968] (1/2) Epoch 27, batch 29450, giga_loss[loss=0.2806, simple_loss=0.3539, pruned_loss=0.1037, over 28624.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3731, pruned_loss=0.1226, over 5648082.12 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5656922.95 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3745, pruned_loss=0.1235, over 5657945.40 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:17:20,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214305.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 02:17:51,470 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 02:18:07,752 INFO [train.py:968] (1/2) Epoch 27, batch 29500, giga_loss[loss=0.2723, simple_loss=0.3512, pruned_loss=0.09666, over 28889.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5648048.95 frames. ], libri_tot_loss[loss=0.2974, simple_loss=0.3632, pruned_loss=0.1158, over 5661766.24 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3751, pruned_loss=0.1243, over 5651318.58 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:18:29,917 INFO [optim.py:369] (1/2) 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 02:18:53,525 INFO [train.py:968] (1/2) Epoch 27, batch 29550, giga_loss[loss=0.296, simple_loss=0.3647, pruned_loss=0.1136, over 28975.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3731, pruned_loss=0.1237, over 5656705.29 frames. ], libri_tot_loss[loss=0.297, simple_loss=0.3629, pruned_loss=0.1155, over 5666491.03 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3746, pruned_loss=0.1247, over 5654934.19 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:18:58,728 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2681, 4.1183, 3.9014, 1.9925], device='cuda:1'), covar=tensor([0.0589, 0.0708, 0.0694, 0.2008], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.1214, 0.1023, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 02:18:58,776 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4123, 1.5606, 1.3839, 1.4804], device='cuda:1'), covar=tensor([0.0747, 0.0367, 0.0342, 0.0880], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 02:19:01,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2650, 1.2614, 1.2838, 1.5074], device='cuda:1'), covar=tensor([0.0776, 0.0376, 0.0338, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 02:19:21,500 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:19:24,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:19:35,817 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214448.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 02:19:37,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214451.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 02:19:40,914 INFO [train.py:968] (1/2) Epoch 27, batch 29600, giga_loss[loss=0.3909, simple_loss=0.4163, pruned_loss=0.1828, over 26695.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3743, pruned_loss=0.125, over 5655322.75 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3624, pruned_loss=0.1153, over 5670008.74 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.376, pruned_loss=0.1262, over 5650528.83 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:19:52,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1214466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:20:04,865 INFO [optim.py:369] (1/2) 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 02:20:05,069 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 29650, libri_loss[loss=0.2983, simple_loss=0.366, pruned_loss=0.1154, over 29526.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.124, over 5664534.42 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3622, pruned_loss=0.1152, over 5677491.98 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3755, pruned_loss=0.1254, over 5653249.30 frames. ], batch size: 81, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:20:27,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1214508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:20:49,052 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:21:12,827 INFO [train.py:968] (1/2) Epoch 27, batch 29700, libri_loss[loss=0.2765, simple_loss=0.339, pruned_loss=0.107, over 29352.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.374, pruned_loss=0.1245, over 5661625.26 frames. ], libri_tot_loss[loss=0.296, simple_loss=0.362, pruned_loss=0.115, over 5680767.10 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.376, pruned_loss=0.1259, over 5649431.34 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:21:33,436 INFO [optim.py:369] (1/2) 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 02:21:40,656 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 29750, giga_loss[loss=0.2984, simple_loss=0.369, pruned_loss=0.1139, over 28498.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.373, pruned_loss=0.1229, over 5678734.34 frames. ], libri_tot_loss[loss=0.2961, simple_loss=0.3621, pruned_loss=0.1151, over 5684019.43 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3748, pruned_loss=0.1242, over 5665880.71 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:22:46,464 INFO [train.py:968] (1/2) Epoch 27, batch 29800, giga_loss[loss=0.359, simple_loss=0.3928, pruned_loss=0.1626, over 23932.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3744, pruned_loss=0.1237, over 5667122.18 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3624, pruned_loss=0.1152, over 5686334.35 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3757, pruned_loss=0.1247, over 5654675.89 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:23:04,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:23:08,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1214676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:23:10,375 INFO [optim.py:369] (1/2) 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 02:23:37,069 INFO [train.py:968] (1/2) Epoch 27, batch 29850, giga_loss[loss=0.2909, simple_loss=0.36, pruned_loss=0.1109, over 29023.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3735, pruned_loss=0.1227, over 5666668.96 frames. ], libri_tot_loss[loss=0.2964, simple_loss=0.3624, pruned_loss=0.1152, over 5686334.35 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3746, pruned_loss=0.1235, over 5656981.86 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:23:37,337 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1214705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:23:59,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 02:24:06,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6046, 1.6246, 1.7941, 1.3967], device='cuda:1'), covar=tensor([0.1771, 0.2593, 0.1488, 0.1733], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0719, 0.0977, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 02:24:21,828 INFO [train.py:968] (1/2) Epoch 27, batch 29900, libri_loss[loss=0.3641, simple_loss=0.4147, pruned_loss=0.1567, over 29486.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3714, pruned_loss=0.1219, over 5666053.23 frames. ], libri_tot_loss[loss=0.2965, simple_loss=0.3625, pruned_loss=0.1153, over 5686784.25 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3723, pruned_loss=0.1226, over 5657543.19 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:24:29,973 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 02:24:45,895 INFO [optim.py:369] (1/2) 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 02:24:54,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1214790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:24:59,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.27 vs. limit=5.0
+2023-03-14 02:25:06,863 INFO [train.py:968] (1/2) Epoch 27, batch 29950, giga_loss[loss=0.3341, simple_loss=0.3756, pruned_loss=0.1463, over 23497.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3686, pruned_loss=0.1202, over 5663371.16 frames. ], libri_tot_loss[loss=0.2966, simple_loss=0.3627, pruned_loss=0.1153, over 5691130.34 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3693, pruned_loss=0.1208, over 5652218.26 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:25:54,472 INFO [train.py:968] (1/2) Epoch 27, batch 30000, giga_loss[loss=0.2687, simple_loss=0.3415, pruned_loss=0.09796, over 29037.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3648, pruned_loss=0.118, over 5667148.29 frames. ], libri_tot_loss[loss=0.2963, simple_loss=0.3625, pruned_loss=0.115, over 5685696.48 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3657, pruned_loss=0.1188, over 5661978.70 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:25:54,472 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 02:25:58,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2868, 1.4973, 1.4616, 1.2655], device='cuda:1'), covar=tensor([0.2723, 0.2775, 0.1806, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.2057, 0.2012, 0.1936, 0.2067], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 02:26:02,679 INFO [train.py:1012] (1/2) Epoch 27, validation: loss=0.2043, simple_loss=0.313, pruned_loss=0.04778, over 944034.00 frames. 
+2023-03-14 02:26:02,679 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 02:26:23,330 INFO [optim.py:369] (1/2) 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 02:26:26,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1214883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:26:44,446 INFO [train.py:968] (1/2) Epoch 27, batch 30050, libri_loss[loss=0.3185, simple_loss=0.3878, pruned_loss=0.1245, over 29278.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3631, pruned_loss=0.1173, over 5683620.87 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.363, pruned_loss=0.1153, over 5690579.91 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3633, pruned_loss=0.1178, over 5674779.16 frames. ], batch size: 94, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:26:46,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3523, 2.0138, 1.5130, 0.5821], device='cuda:1'), covar=tensor([0.5543, 0.3194, 0.4468, 0.7211], device='cuda:1'), in_proj_covar=tensor([0.1849, 0.1749, 0.1672, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 02:27:30,954 INFO [train.py:968] (1/2) Epoch 27, batch 30100, giga_loss[loss=0.3298, simple_loss=0.391, pruned_loss=0.1343, over 28661.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3616, pruned_loss=0.1166, over 5695924.65 frames. ], libri_tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1152, over 5694625.11 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3619, pruned_loss=0.1171, over 5685050.35 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:27:37,595 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 27, batch 30150, giga_loss[loss=0.283, simple_loss=0.3528, pruned_loss=0.1066, over 28835.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3618, pruned_loss=0.1162, over 5685194.42 frames. ], libri_tot_loss[loss=0.2968, simple_loss=0.3629, pruned_loss=0.1154, over 5687303.81 frames. ], giga_tot_loss[loss=0.2974, simple_loss=0.3619, pruned_loss=0.1165, over 5682572.80 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:28:32,689 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:28:33,239 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3502, 2.3350, 2.0659, 1.7108], device='cuda:1'), covar=tensor([0.1832, 0.1740, 0.1784, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0761, 0.0731, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 02:28:34,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:28:48,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 02:28:58,276 INFO [train.py:968] (1/2) Epoch 27, batch 30200, giga_loss[loss=0.2755, simple_loss=0.3539, pruned_loss=0.09853, over 27906.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3604, pruned_loss=0.1138, over 5684469.04 frames. ], libri_tot_loss[loss=0.2959, simple_loss=0.362, pruned_loss=0.1149, over 5696818.91 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3613, pruned_loss=0.1144, over 5673568.83 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:29:02,642 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:29:50,310 INFO [train.py:968] (1/2) Epoch 27, batch 30250, libri_loss[loss=0.2747, simple_loss=0.3455, pruned_loss=0.102, over 27647.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3581, pruned_loss=0.1106, over 5677702.61 frames. ], libri_tot_loss[loss=0.2957, simple_loss=0.3617, pruned_loss=0.1148, over 5697735.42 frames. ], giga_tot_loss[loss=0.2906, simple_loss=0.3591, pruned_loss=0.1111, over 5667599.18 frames. ], batch size: 116, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:29:51,343 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:29:54,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:30:16,333 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-14 02:30:21,519 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:30:38,799 INFO [train.py:968] (1/2) Epoch 27, batch 30300, giga_loss[loss=0.295, simple_loss=0.3697, pruned_loss=0.1102, over 28555.00 frames. ], tot_loss[loss=0.2852, simple_loss=0.3553, pruned_loss=0.1075, over 5664504.59 frames. ], libri_tot_loss[loss=0.2953, simple_loss=0.3613, pruned_loss=0.1147, over 5698407.77 frames. ], giga_tot_loss[loss=0.2862, simple_loss=0.3564, pruned_loss=0.108, over 5655733.91 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:30:51,179 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:31:06,701 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:31:24,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:31:25,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-14 02:31:28,552 INFO [train.py:968] (1/2) Epoch 27, batch 30350, giga_loss[loss=0.2372, simple_loss=0.323, pruned_loss=0.07573, over 29104.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3515, pruned_loss=0.1039, over 5661093.25 frames. ], libri_tot_loss[loss=0.2947, simple_loss=0.3607, pruned_loss=0.1143, over 5700600.01 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3528, pruned_loss=0.1045, over 5652085.28 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:31:32,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4929, 1.6539, 1.7083, 1.2839], device='cuda:1'), covar=tensor([0.1985, 0.2927, 0.1677, 0.2005], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0716, 0.0974, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 02:32:16,301 INFO [train.py:968] (1/2) Epoch 27, batch 30400, giga_loss[loss=0.244, simple_loss=0.3304, pruned_loss=0.07879, over 27941.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3499, pruned_loss=0.1012, over 5661229.40 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3599, pruned_loss=0.114, over 5704615.80 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3515, pruned_loss=0.1018, over 5649961.01 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:32:43,417 INFO [optim.py:369] (1/2) 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 02:33:07,744 INFO [train.py:968] (1/2) Epoch 27, batch 30450, giga_loss[loss=0.2609, simple_loss=0.3424, pruned_loss=0.08971, over 27939.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3495, pruned_loss=0.1003, over 5641319.18 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3595, pruned_loss=0.1139, over 5700022.72 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.3508, pruned_loss=0.1005, over 5635069.06 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:33:11,242 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:33:15,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:33:41,270 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:33:53,010 INFO [train.py:968] (1/2) Epoch 27, batch 30500, libri_loss[loss=0.2609, simple_loss=0.3212, pruned_loss=0.1003, over 29679.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.35, pruned_loss=0.1012, over 5654927.18 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.358, pruned_loss=0.1133, over 5706921.78 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3521, pruned_loss=0.1013, over 5639994.35 frames. ], batch size: 73, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:34:03,121 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2323, 1.6242, 1.5734, 1.4282], device='cuda:1'), covar=tensor([0.1901, 0.1518, 0.1861, 0.1686], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0753, 0.0724, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 02:34:15,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 02:34:17,825 INFO [optim.py:369] (1/2) 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 02:34:35,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0289, 1.3538, 5.0431, 3.6181], device='cuda:1'), covar=tensor([0.1475, 0.2859, 0.0494, 0.1097], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0675, 0.1007, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 02:34:40,551 INFO [train.py:968] (1/2) Epoch 27, batch 30550, giga_loss[loss=0.2667, simple_loss=0.3281, pruned_loss=0.1026, over 24145.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3482, pruned_loss=0.1005, over 5641780.30 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3574, pruned_loss=0.1129, over 5708273.77 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3502, pruned_loss=0.1005, over 5627110.89 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:35:09,720 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:35:31,486 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-14 02:35:31,602 INFO [train.py:968] (1/2) Epoch 27, batch 30600, giga_loss[loss=0.2545, simple_loss=0.3308, pruned_loss=0.08911, over 28252.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3451, pruned_loss=0.0979, over 5650098.38 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3573, pruned_loss=0.1129, over 5710494.00 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3466, pruned_loss=0.09778, over 5636085.90 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:35:59,314 INFO [optim.py:369] (1/2) 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 02:36:21,178 INFO [train.py:968] (1/2) Epoch 27, batch 30650, libri_loss[loss=0.2832, simple_loss=0.3316, pruned_loss=0.1174, over 28620.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3436, pruned_loss=0.09702, over 5649906.33 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3568, pruned_loss=0.1128, over 5715724.33 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3449, pruned_loss=0.09659, over 5631875.11 frames. ], batch size: 63, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:37:09,018 INFO [train.py:968] (1/2) Epoch 27, batch 30700, giga_loss[loss=0.2336, simple_loss=0.3043, pruned_loss=0.08148, over 24204.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3441, pruned_loss=0.09657, over 5653793.83 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3569, pruned_loss=0.1129, over 5717587.56 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3449, pruned_loss=0.09595, over 5637188.42 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:37:11,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215558.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:37:20,522 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:37:28,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:37:34,622 INFO [optim.py:369] (1/2) 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 02:37:43,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4262, 3.7164, 1.5710, 1.6409], device='cuda:1'), covar=tensor([0.1027, 0.0273, 0.0974, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0568, 0.0407, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 02:37:58,380 INFO [train.py:968] (1/2) Epoch 27, batch 30750, giga_loss[loss=0.2375, simple_loss=0.3112, pruned_loss=0.08185, over 26685.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3412, pruned_loss=0.09396, over 5656598.73 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3565, pruned_loss=0.1128, over 5717738.32 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3419, pruned_loss=0.09322, over 5642204.63 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:38:47,220 INFO [train.py:968] (1/2) Epoch 27, batch 30800, giga_loss[loss=0.2084, simple_loss=0.2798, pruned_loss=0.06848, over 24059.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09265, over 5646919.99 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3563, pruned_loss=0.1128, over 5717366.82 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3394, pruned_loss=0.0918, over 5635037.13 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:39:16,105 INFO [optim.py:369] (1/2) 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 02:39:24,422 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6816, 1.7925, 1.9282, 1.4909], device='cuda:1'), covar=tensor([0.2155, 0.2904, 0.1749, 0.2100], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0713, 0.0973, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 02:39:31,999 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:39:35,844 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 30850, giga_loss[loss=0.3092, simple_loss=0.3696, pruned_loss=0.1243, over 28661.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3348, pruned_loss=0.09053, over 5650693.48 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3558, pruned_loss=0.1125, over 5720134.33 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3354, pruned_loss=0.08977, over 5637883.84 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:39:49,181 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:39:51,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:40:00,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 02:40:02,484 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:40:18,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:40:22,730 INFO [train.py:968] (1/2) Epoch 27, batch 30900, giga_loss[loss=0.2425, simple_loss=0.3238, pruned_loss=0.08062, over 28621.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3338, pruned_loss=0.09043, over 5652568.92 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3554, pruned_loss=0.1124, over 5719647.80 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3342, pruned_loss=0.08956, over 5641612.95 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:40:23,257 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:41:12,589 INFO [train.py:968] (1/2) Epoch 27, batch 30950, giga_loss[loss=0.2437, simple_loss=0.3321, pruned_loss=0.07763, over 28647.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3341, pruned_loss=0.09097, over 5638970.21 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3555, pruned_loss=0.1125, over 5724978.38 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3337, pruned_loss=0.08946, over 5622935.99 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:41:18,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:42:02,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6935, 1.7841, 1.4584, 1.7887], device='cuda:1'), covar=tensor([0.2859, 0.2953, 0.3440, 0.2732], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1143, 0.1411, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 02:42:10,847 INFO [train.py:968] (1/2) Epoch 27, batch 31000, giga_loss[loss=0.2789, simple_loss=0.364, pruned_loss=0.09693, over 28355.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3367, pruned_loss=0.0914, over 5638798.71 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3554, pruned_loss=0.1124, over 5726464.03 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3363, pruned_loss=0.09007, over 5623933.12 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:42:44,392 INFO [optim.py:369] (1/2) 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 02:42:52,996 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1215891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:43:06,424 INFO [train.py:968] (1/2) Epoch 27, batch 31050, giga_loss[loss=0.2727, simple_loss=0.3516, pruned_loss=0.09693, over 28670.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3388, pruned_loss=0.09164, over 5652297.57 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3549, pruned_loss=0.1121, over 5726797.95 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3386, pruned_loss=0.09037, over 5638118.05 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:43:52,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1215943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:44:03,126 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1215952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:44:07,383 INFO [train.py:968] (1/2) Epoch 27, batch 31100, giga_loss[loss=0.2232, simple_loss=0.3067, pruned_loss=0.06987, over 28968.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3398, pruned_loss=0.09227, over 5672853.48 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3553, pruned_loss=0.1126, over 5731076.61 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3389, pruned_loss=0.09043, over 5656481.90 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:44:08,304 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1215955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:44:18,186 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2849, 1.3600, 1.3262, 1.2859], device='cuda:1'), covar=tensor([0.2099, 0.1924, 0.1682, 0.1834], device='cuda:1'), in_proj_covar=tensor([0.2019, 0.1975, 0.1894, 0.2025], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 02:44:43,219 INFO [optim.py:369] (1/2) 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 02:44:44,821 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1215984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:45:07,302 INFO [train.py:968] (1/2) Epoch 27, batch 31150, giga_loss[loss=0.2216, simple_loss=0.3136, pruned_loss=0.06474, over 28869.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3365, pruned_loss=0.09063, over 5663016.99 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3545, pruned_loss=0.1122, over 5732724.53 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3362, pruned_loss=0.0891, over 5647451.67 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:45:11,989 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-14 02:45:38,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3261, 1.5691, 1.5909, 1.1802], device='cuda:1'), covar=tensor([0.1944, 0.2810, 0.1622, 0.1911], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0712, 0.0973, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 02:46:07,548 INFO [train.py:968] (1/2) Epoch 27, batch 31200, giga_loss[loss=0.2036, simple_loss=0.28, pruned_loss=0.06365, over 24262.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3365, pruned_loss=0.08982, over 5658049.83 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3542, pruned_loss=0.1122, over 5726118.20 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3359, pruned_loss=0.08774, over 5649514.32 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:46:19,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-14 02:46:41,932 INFO [optim.py:369] (1/2) 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 02:46:44,353 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-14 02:46:46,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:46:49,937 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1216089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:47:07,293 INFO [train.py:968] (1/2) Epoch 27, batch 31250, giga_loss[loss=0.2364, simple_loss=0.3176, pruned_loss=0.07759, over 28167.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3342, pruned_loss=0.08831, over 5659231.02 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.354, pruned_loss=0.1121, over 5718936.09 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3337, pruned_loss=0.08646, over 5658516.74 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:47:24,941 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1216118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:47:36,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:48:07,727 INFO [train.py:968] (1/2) Epoch 27, batch 31300, giga_loss[loss=0.2637, simple_loss=0.3414, pruned_loss=0.09302, over 28976.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3321, pruned_loss=0.08809, over 5655361.30 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3536, pruned_loss=0.112, over 5719471.25 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3317, pruned_loss=0.08642, over 5653617.07 frames. ], batch size: 285, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:48:21,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4419, 1.4406, 3.8386, 3.2991], device='cuda:1'), covar=tensor([0.1640, 0.2838, 0.0499, 0.1066], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0670, 0.0998, 0.0966], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 02:48:39,954 INFO [optim.py:369] (1/2) 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 02:49:00,778 INFO [train.py:968] (1/2) Epoch 27, batch 31350, giga_loss[loss=0.2315, simple_loss=0.2973, pruned_loss=0.08289, over 24549.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3315, pruned_loss=0.08849, over 5653016.15 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.353, pruned_loss=0.1119, over 5710679.50 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3309, pruned_loss=0.08621, over 5656624.40 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:49:58,400 INFO [train.py:968] (1/2) Epoch 27, batch 31400, giga_loss[loss=0.2495, simple_loss=0.3362, pruned_loss=0.08136, over 28129.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3325, pruned_loss=0.08826, over 5657002.03 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3527, pruned_loss=0.1117, over 5710904.29 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3321, pruned_loss=0.08641, over 5659112.37 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:50:10,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:50:18,121 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:50:21,577 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:50:45,940 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1216295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:50:55,872 INFO [train.py:968] (1/2) Epoch 27, batch 31450, giga_loss[loss=0.2393, simple_loss=0.327, pruned_loss=0.0758, over 29038.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3338, pruned_loss=0.08811, over 5660274.79 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3523, pruned_loss=0.1116, over 5713502.69 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3333, pruned_loss=0.08625, over 5658371.13 frames. ], batch size: 100, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:50:57,642 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1216305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:51:42,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-14 02:51:54,456 INFO [train.py:968] (1/2) Epoch 27, batch 31500, giga_loss[loss=0.2347, simple_loss=0.3194, pruned_loss=0.07499, over 28950.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3324, pruned_loss=0.08747, over 5662940.51 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3522, pruned_loss=0.1116, over 5710537.43 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3316, pruned_loss=0.08517, over 5662508.98 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:52:31,055 INFO [optim.py:369] (1/2) 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 02:52:50,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6279, 1.8811, 1.3305, 1.4911], device='cuda:1'), covar=tensor([0.1054, 0.0563, 0.0994, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0448, 0.0523, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 02:52:59,706 INFO [train.py:968] (1/2) Epoch 27, batch 31550, giga_loss[loss=0.2366, simple_loss=0.3126, pruned_loss=0.0803, over 28826.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3319, pruned_loss=0.08727, over 5672296.59 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3516, pruned_loss=0.1113, over 5716039.11 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3312, pruned_loss=0.08509, over 5666016.86 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:53:04,756 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:53:10,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1216412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:53:28,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3553, 1.6520, 1.6949, 1.4382], device='cuda:1'), covar=tensor([0.2097, 0.2105, 0.2121, 0.2100], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0747, 0.0720, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 02:53:44,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7481, 1.9532, 1.6820, 1.6972], device='cuda:1'), covar=tensor([0.2492, 0.2408, 0.2510, 0.2336], device='cuda:1'), in_proj_covar=tensor([0.1587, 0.1143, 0.1408, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 02:53:46,017 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1216441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:54:00,135 INFO [train.py:968] (1/2) Epoch 27, batch 31600, giga_loss[loss=0.2699, simple_loss=0.3599, pruned_loss=0.08997, over 28893.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3342, pruned_loss=0.08765, over 5667221.74 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3514, pruned_loss=0.1112, over 5715661.79 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3336, pruned_loss=0.08569, over 5661927.66 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:54:16,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4011, 2.0623, 1.6948, 1.6424], device='cuda:1'), covar=tensor([0.0823, 0.0275, 0.0328, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 02:54:40,251 INFO [optim.py:369] (1/2) 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 02:54:51,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 02:54:53,635 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 02:55:07,015 INFO [train.py:968] (1/2) Epoch 27, batch 31650, giga_loss[loss=0.252, simple_loss=0.3521, pruned_loss=0.07591, over 28684.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3378, pruned_loss=0.08691, over 5655969.94 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3514, pruned_loss=0.1113, over 5708645.95 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3371, pruned_loss=0.08491, over 5657106.98 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 02:56:06,650 INFO [train.py:968] (1/2) Epoch 27, batch 31700, giga_loss[loss=0.1872, simple_loss=0.2623, pruned_loss=0.05607, over 24247.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3393, pruned_loss=0.08662, over 5652260.09 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3516, pruned_loss=0.1114, over 5707331.54 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3383, pruned_loss=0.0845, over 5653269.34 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:56:20,059 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:56:54,660 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3731, 3.2366, 1.4719, 1.4548], device='cuda:1'), covar=tensor([0.1046, 0.0300, 0.0951, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0565, 0.0407, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 02:57:06,699 INFO [train.py:968] (1/2) Epoch 27, batch 31750, giga_loss[loss=0.2488, simple_loss=0.3429, pruned_loss=0.07734, over 28044.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.338, pruned_loss=0.08553, over 5652539.19 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3511, pruned_loss=0.1113, over 5712212.06 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3374, pruned_loss=0.08337, over 5647594.17 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:57:27,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1216622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:58:07,838 INFO [train.py:968] (1/2) Epoch 27, batch 31800, giga_loss[loss=0.2392, simple_loss=0.3262, pruned_loss=0.07611, over 28401.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.339, pruned_loss=0.08683, over 5648679.24 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.351, pruned_loss=0.1113, over 5705983.00 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3384, pruned_loss=0.08468, over 5648854.37 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 02:58:29,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:58:46,241 INFO [optim.py:369] (1/2) 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 02:58:46,431 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1216685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 02:59:15,635 INFO [train.py:968] (1/2) Epoch 27, batch 31850, giga_loss[loss=0.2502, simple_loss=0.3348, pruned_loss=0.0828, over 29085.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3377, pruned_loss=0.08724, over 5657990.22 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3512, pruned_loss=0.1115, over 5706654.22 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3368, pruned_loss=0.08491, over 5656400.57 frames. ], batch size: 165, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:00:30,560 INFO [train.py:968] (1/2) Epoch 27, batch 31900, giga_loss[loss=0.2499, simple_loss=0.3266, pruned_loss=0.08664, over 28242.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3383, pruned_loss=0.0883, over 5669507.56 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3507, pruned_loss=0.1112, over 5709887.19 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3378, pruned_loss=0.0862, over 5664219.51 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:00:37,568 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4069, 1.7165, 1.3963, 1.2909], device='cuda:1'), covar=tensor([0.2544, 0.2529, 0.2888, 0.2404], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1140, 0.1404, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 03:00:49,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.6944, 5.5188, 5.2663, 2.9255], device='cuda:1'), covar=tensor([0.0451, 0.0567, 0.0730, 0.1489], device='cuda:1'), in_proj_covar=tensor([0.1278, 0.1179, 0.0993, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 03:01:11,481 INFO [optim.py:369] (1/2) 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 03:01:37,650 INFO [train.py:968] (1/2) Epoch 27, batch 31950, giga_loss[loss=0.186, simple_loss=0.2858, pruned_loss=0.04312, over 28846.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3339, pruned_loss=0.08638, over 5670868.36 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3502, pruned_loss=0.111, over 5714141.83 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3334, pruned_loss=0.08406, over 5661621.14 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:01:47,190 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1216813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:01:51,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1216816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:02:29,460 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 32000, giga_loss[loss=0.2182, simple_loss=0.3071, pruned_loss=0.06461, over 28433.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3322, pruned_loss=0.08564, over 5675358.64 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3497, pruned_loss=0.1107, over 5719519.09 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3317, pruned_loss=0.08329, over 5662009.97 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:03:14,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6060, 2.2886, 1.6945, 0.9403], device='cuda:1'), covar=tensor([0.7044, 0.3664, 0.4506, 0.6788], device='cuda:1'), in_proj_covar=tensor([0.1822, 0.1715, 0.1648, 0.1493], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:03:16,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3496, 1.6141, 1.4379, 1.5499], device='cuda:1'), covar=tensor([0.0754, 0.0368, 0.0347, 0.0851], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 03:03:17,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7770, 1.7239, 1.9063, 1.4715], device='cuda:1'), covar=tensor([0.2307, 0.3308, 0.1718, 0.1913], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0712, 0.0974, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 03:03:19,244 INFO [optim.py:369] (1/2) 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 03:03:45,100 INFO [train.py:968] (1/2) Epoch 27, batch 32050, giga_loss[loss=0.292, simple_loss=0.375, pruned_loss=0.1045, over 28668.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3303, pruned_loss=0.08521, over 5678263.84 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3492, pruned_loss=0.1105, over 5723339.55 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.33, pruned_loss=0.08311, over 5663883.00 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:04:28,062 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1216941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:04:44,299 INFO [train.py:968] (1/2) Epoch 27, batch 32100, giga_loss[loss=0.2422, simple_loss=0.3213, pruned_loss=0.08154, over 26890.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.334, pruned_loss=0.08724, over 5676050.16 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3489, pruned_loss=0.1103, over 5724182.70 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3336, pruned_loss=0.08506, over 5662133.55 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:04:47,286 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 03:05:05,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-14 03:05:21,468 INFO [optim.py:369] (1/2) 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 03:05:27,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1528, 1.2202, 3.4462, 3.0691], device='cuda:1'), covar=tensor([0.1610, 0.2570, 0.0587, 0.0993], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0669, 0.0995, 0.0962], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 03:05:33,152 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 32150, libri_loss[loss=0.2804, simple_loss=0.346, pruned_loss=0.1073, over 29536.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3353, pruned_loss=0.08813, over 5680213.23 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3488, pruned_loss=0.1103, over 5727577.29 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3348, pruned_loss=0.08602, over 5665151.91 frames. ], batch size: 89, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:06:44,260 INFO [train.py:968] (1/2) Epoch 27, batch 32200, giga_loss[loss=0.3857, simple_loss=0.4229, pruned_loss=0.1742, over 26915.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3332, pruned_loss=0.08839, over 5676651.94 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.348, pruned_loss=0.1099, over 5732024.49 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3331, pruned_loss=0.08638, over 5658958.56 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:06:51,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1217060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:07:05,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4479, 1.6241, 1.3732, 1.4836], device='cuda:1'), covar=tensor([0.0775, 0.0327, 0.0353, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 03:07:18,853 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:07:19,011 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-14 03:07:20,125 INFO [optim.py:369] (1/2) 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 03:07:22,732 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 32250, giga_loss[loss=0.2128, simple_loss=0.3017, pruned_loss=0.06195, over 28792.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3344, pruned_loss=0.08964, over 5662291.12 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3484, pruned_loss=0.1104, over 5715696.00 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3338, pruned_loss=0.08723, over 5662257.58 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:07:56,386 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1217116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:08:29,089 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:08:32,676 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1217143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:08:48,684 INFO [train.py:968] (1/2) Epoch 27, batch 32300, giga_loss[loss=0.2843, simple_loss=0.3659, pruned_loss=0.1013, over 28955.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3357, pruned_loss=0.08963, over 5649155.37 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3483, pruned_loss=0.1106, over 5699207.76 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3351, pruned_loss=0.08728, over 5662103.71 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:09:11,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1217172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:09:20,829 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8152, 2.6438, 1.6517, 0.9368], device='cuda:1'), covar=tensor([0.9508, 0.4238, 0.4999, 0.8017], device='cuda:1'), in_proj_covar=tensor([0.1823, 0.1717, 0.1644, 0.1493], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:09:26,033 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:09:56,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:09:57,857 INFO [train.py:968] (1/2) Epoch 27, batch 32350, giga_loss[loss=0.2313, simple_loss=0.3226, pruned_loss=0.06996, over 28729.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3378, pruned_loss=0.08987, over 5663824.42 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3485, pruned_loss=0.1108, over 5702625.75 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3369, pruned_loss=0.08736, over 5670077.50 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:10:00,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1217206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:10:39,058 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-14 03:10:43,171 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 32400, giga_loss[loss=0.2649, simple_loss=0.3278, pruned_loss=0.101, over 26923.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3364, pruned_loss=0.08886, over 5666853.99 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3486, pruned_loss=0.1109, over 5708254.57 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3352, pruned_loss=0.08606, over 5665364.40 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:11:48,265 INFO [optim.py:369] (1/2) 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 03:11:52,059 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 03:12:12,979 INFO [train.py:968] (1/2) Epoch 27, batch 32450, libri_loss[loss=0.2999, simple_loss=0.365, pruned_loss=0.1174, over 29537.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3322, pruned_loss=0.08781, over 5669402.79 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3484, pruned_loss=0.1109, over 5712268.64 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3311, pruned_loss=0.08513, over 5663744.72 frames. ], batch size: 84, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:12:31,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3427, 3.1846, 3.0446, 1.5316], device='cuda:1'), covar=tensor([0.0944, 0.1092, 0.0982, 0.2055], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.1179, 0.0993, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 03:13:08,950 INFO [train.py:968] (1/2) Epoch 27, batch 32500, giga_loss[loss=0.2255, simple_loss=0.3048, pruned_loss=0.07315, over 28408.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3263, pruned_loss=0.08518, over 5681927.09 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3473, pruned_loss=0.1102, over 5717728.26 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3258, pruned_loss=0.08278, over 5670880.88 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:13:21,964 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.86 vs. limit=5.0
+2023-03-14 03:13:21,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 03:13:26,314 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8433, 2.4068, 1.6092, 1.9075], device='cuda:1'), covar=tensor([0.1034, 0.0538, 0.0948, 0.1056], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0445, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 03:13:51,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 03:13:51,916 INFO [optim.py:369] (1/2) 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 03:14:14,386 INFO [train.py:968] (1/2) Epoch 27, batch 32550, giga_loss[loss=0.236, simple_loss=0.3191, pruned_loss=0.07639, over 28876.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.327, pruned_loss=0.08561, over 5667349.04 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3473, pruned_loss=0.1101, over 5719591.30 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3264, pruned_loss=0.08359, over 5656663.34 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:14:31,255 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 03:15:12,702 INFO [train.py:968] (1/2) Epoch 27, batch 32600, giga_loss[loss=0.2433, simple_loss=0.3312, pruned_loss=0.07769, over 28701.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3287, pruned_loss=0.08689, over 5664305.69 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.347, pruned_loss=0.1099, over 5722243.98 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3282, pruned_loss=0.08506, over 5652536.10 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:15:50,820 INFO [optim.py:369] (1/2) 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 03:16:08,513 INFO [train.py:968] (1/2) Epoch 27, batch 32650, libri_loss[loss=0.3354, simple_loss=0.3846, pruned_loss=0.1431, over 29384.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3267, pruned_loss=0.08565, over 5667858.36 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3465, pruned_loss=0.1097, over 5727667.87 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.326, pruned_loss=0.08345, over 5651162.43 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:16:20,240 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:16:52,665 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3978, 1.6440, 1.6359, 1.2065], device='cuda:1'), covar=tensor([0.1758, 0.2793, 0.1544, 0.1938], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0711, 0.0976, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 03:17:09,651 INFO [train.py:968] (1/2) Epoch 27, batch 32700, giga_loss[loss=0.2232, simple_loss=0.3008, pruned_loss=0.07281, over 29016.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3259, pruned_loss=0.08443, over 5671292.86 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3465, pruned_loss=0.1098, over 5726667.49 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3251, pruned_loss=0.08227, over 5658005.86 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:17:16,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1217558.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:17:18,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8660, 1.0420, 0.9657, 0.8372], device='cuda:1'), covar=tensor([0.2367, 0.2270, 0.1550, 0.2176], device='cuda:1'), in_proj_covar=tensor([0.2005, 0.1956, 0.1870, 0.2007], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 03:17:18,685 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 03:17:50,377 INFO [optim.py:369] (1/2) 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 03:18:11,135 INFO [train.py:968] (1/2) Epoch 27, batch 32750, giga_loss[loss=0.2712, simple_loss=0.342, pruned_loss=0.1002, over 28121.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3262, pruned_loss=0.08532, over 5676142.96 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3465, pruned_loss=0.1098, over 5731829.91 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3249, pruned_loss=0.08281, over 5658911.93 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:19:18,458 INFO [train.py:968] (1/2) Epoch 27, batch 32800, libri_loss[loss=0.2845, simple_loss=0.3515, pruned_loss=0.1088, over 29145.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3255, pruned_loss=0.08439, over 5666354.25 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3463, pruned_loss=0.1095, over 5733025.77 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3241, pruned_loss=0.08199, over 5650078.32 frames. ], batch size: 101, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:19:26,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5743, 2.8871, 2.6736, 2.2498], device='cuda:1'), covar=tensor([0.2190, 0.1657, 0.1852, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.2006, 0.1959, 0.1871, 0.2010], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 03:19:49,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 03:19:58,551 INFO [optim.py:369] (1/2) 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 03:20:16,698 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1217701.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:20:21,989 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 32850, giga_loss[loss=0.2613, simple_loss=0.3334, pruned_loss=0.0946, over 29012.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.325, pruned_loss=0.08393, over 5668254.18 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3458, pruned_loss=0.1092, over 5737386.55 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3239, pruned_loss=0.08172, over 5650096.80 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:20:37,804 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5441, 4.3713, 4.1444, 1.9184], device='cuda:1'), covar=tensor([0.0661, 0.0817, 0.0933, 0.2084], device='cuda:1'), in_proj_covar=tensor([0.1278, 0.1180, 0.0993, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 03:20:47,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5854, 1.8248, 1.4603, 1.4999], device='cuda:1'), covar=tensor([0.2810, 0.2838, 0.3286, 0.2436], device='cuda:1'), in_proj_covar=tensor([0.1589, 0.1142, 0.1407, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 03:20:51,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3648, 2.9816, 1.5604, 1.4741], device='cuda:1'), covar=tensor([0.0927, 0.0512, 0.0891, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0562, 0.0405, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:20:56,599 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1217733.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:21:13,624 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 32900, giga_loss[loss=0.222, simple_loss=0.3062, pruned_loss=0.06894, over 28602.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3267, pruned_loss=0.08584, over 5661159.16 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3454, pruned_loss=0.1091, over 5726965.03 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3257, pruned_loss=0.08384, over 5654449.02 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:22:01,843 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:22:21,985 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-14 03:22:24,429 INFO [train.py:968] (1/2) Epoch 27, batch 32950, giga_loss[loss=0.2232, simple_loss=0.3229, pruned_loss=0.06172, over 28863.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3245, pruned_loss=0.08403, over 5649272.80 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3456, pruned_loss=0.1092, over 5719895.38 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3234, pruned_loss=0.082, over 5649678.81 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:22:25,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2192, 1.4329, 1.4103, 1.1780], device='cuda:1'), covar=tensor([0.2804, 0.2852, 0.1789, 0.2580], device='cuda:1'), in_proj_covar=tensor([0.2016, 0.1967, 0.1879, 0.2018], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 03:23:06,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3197, 3.8263, 1.5117, 1.4864], device='cuda:1'), covar=tensor([0.1044, 0.0304, 0.0991, 0.1390], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0563, 0.0406, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:23:23,103 INFO [train.py:968] (1/2) Epoch 27, batch 33000, libri_loss[loss=0.3383, simple_loss=0.3877, pruned_loss=0.1445, over 19219.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3273, pruned_loss=0.08411, over 5648826.32 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3455, pruned_loss=0.1092, over 5712886.58 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3261, pruned_loss=0.08205, over 5655063.01 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:23:23,103 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 03:23:30,726 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2577, 1.6030, 1.5104, 1.0947], device='cuda:1'), covar=tensor([0.1637, 0.2800, 0.1550, 0.2002], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0709, 0.0975, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 03:23:31,647 INFO [train.py:1012] (1/2) Epoch 27, validation: loss=0.1936, simple_loss=0.295, pruned_loss=0.04608, over 944034.00 frames. 
+2023-03-14 03:23:31,647 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 03:23:50,565 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 03:24:09,326 INFO [optim.py:369] (1/2) 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 03:24:09,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1217888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:24:27,102 INFO [train.py:968] (1/2) Epoch 27, batch 33050, giga_loss[loss=0.2462, simple_loss=0.3415, pruned_loss=0.07547, over 28812.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3294, pruned_loss=0.08477, over 5655827.43 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3453, pruned_loss=0.1091, over 5718166.96 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.328, pruned_loss=0.0823, over 5653937.85 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:24:35,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5748, 5.1807, 1.9018, 1.9675], device='cuda:1'), covar=tensor([0.1025, 0.0284, 0.0926, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0563, 0.0406, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:25:13,611 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1217943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:25:29,696 INFO [train.py:968] (1/2) Epoch 27, batch 33100, giga_loss[loss=0.2382, simple_loss=0.3267, pruned_loss=0.07484, over 28373.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3295, pruned_loss=0.08488, over 5648317.07 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3449, pruned_loss=0.1089, over 5723038.36 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3285, pruned_loss=0.08249, over 5641017.50 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:26:08,319 INFO [optim.py:369] (1/2) 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 03:26:27,312 INFO [train.py:968] (1/2) Epoch 27, batch 33150, giga_loss[loss=0.263, simple_loss=0.3337, pruned_loss=0.09619, over 26929.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.33, pruned_loss=0.08557, over 5644495.84 frames. ], libri_tot_loss[loss=0.2814, simple_loss=0.3447, pruned_loss=0.1091, over 5706865.86 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3288, pruned_loss=0.08282, over 5649790.08 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:26:32,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9036, 1.0884, 0.8702, 0.2914], device='cuda:1'), covar=tensor([0.4478, 0.3149, 0.3512, 0.6426], device='cuda:1'), in_proj_covar=tensor([0.1824, 0.1722, 0.1645, 0.1493], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:26:34,959 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:26:52,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2335, 1.9019, 1.5938, 1.4031], device='cuda:1'), covar=tensor([0.0849, 0.0304, 0.0316, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 03:26:54,454 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:26:58,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:27:22,476 INFO [train.py:968] (1/2) Epoch 27, batch 33200, giga_loss[loss=0.2028, simple_loss=0.2879, pruned_loss=0.05883, over 28972.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.329, pruned_loss=0.08491, over 5656433.48 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3446, pruned_loss=0.109, over 5711821.76 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3277, pruned_loss=0.0821, over 5654759.82 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:27:22,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3886, 2.8684, 1.4749, 1.4792], device='cuda:1'), covar=tensor([0.0979, 0.0358, 0.0945, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0564, 0.0406, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:27:31,857 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:28:03,610 INFO [optim.py:369] (1/2) 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 03:28:05,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4461, 3.3734, 1.6204, 1.5332], device='cuda:1'), covar=tensor([0.0951, 0.0357, 0.0929, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0565, 0.0407, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:28:25,909 INFO [train.py:968] (1/2) Epoch 27, batch 33250, giga_loss[loss=0.2494, simple_loss=0.33, pruned_loss=0.08438, over 27686.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3274, pruned_loss=0.08373, over 5657187.24 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3446, pruned_loss=0.109, over 5711821.76 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3264, pruned_loss=0.08154, over 5655884.62 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:28:45,857 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:29:25,141 INFO [train.py:968] (1/2) Epoch 27, batch 33300, giga_loss[loss=0.2877, simple_loss=0.3704, pruned_loss=0.1025, over 28747.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.325, pruned_loss=0.08298, over 5665148.27 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3444, pruned_loss=0.1088, over 5715759.68 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3241, pruned_loss=0.08096, over 5659851.07 frames. ], batch size: 243, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:29:33,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:29:47,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:30:22,987 INFO [train.py:968] (1/2) Epoch 27, batch 33350, giga_loss[loss=0.2523, simple_loss=0.3424, pruned_loss=0.08113, over 29042.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3277, pruned_loss=0.08407, over 5672861.10 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3439, pruned_loss=0.1084, over 5718657.38 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.327, pruned_loss=0.0822, over 5664866.10 frames. ], batch size: 285, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:31:06,528 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 03:31:28,676 INFO [train.py:968] (1/2) Epoch 27, batch 33400, giga_loss[loss=0.296, simple_loss=0.3684, pruned_loss=0.1118, over 28948.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3303, pruned_loss=0.08542, over 5669741.26 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3442, pruned_loss=0.1086, over 5717384.89 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3292, pruned_loss=0.08346, over 5663906.51 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:31:45,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:31:49,311 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:31:49,365 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4852, 1.7811, 1.4274, 1.4497], device='cuda:1'), covar=tensor([0.2814, 0.2681, 0.3107, 0.2419], device='cuda:1'), in_proj_covar=tensor([0.1591, 0.1142, 0.1408, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 03:31:53,960 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7292, 1.8495, 1.3481, 1.4207], device='cuda:1'), covar=tensor([0.0984, 0.0582, 0.1025, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0445, 0.0521, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 03:32:01,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2083, 1.6481, 1.2352, 0.5353], device='cuda:1'), covar=tensor([0.4324, 0.2713, 0.3968, 0.6119], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1727, 0.1649, 0.1495], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:32:13,625 INFO [optim.py:369] (1/2) 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 03:32:25,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:31,001 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 33450, giga_loss[loss=0.2344, simple_loss=0.3215, pruned_loss=0.0737, over 29053.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3301, pruned_loss=0.08539, over 5672001.51 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3438, pruned_loss=0.1084, over 5720531.03 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3294, pruned_loss=0.08365, over 5663875.60 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:32:33,584 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-14 03:32:33,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:44,833 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:46,005 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:32:46,699 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:33:12,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5966, 1.8599, 2.0226, 1.6222], device='cuda:1'), covar=tensor([0.2990, 0.2389, 0.2334, 0.2560], device='cuda:1'), in_proj_covar=tensor([0.2013, 0.1962, 0.1872, 0.2012], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 03:33:15,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1803, 2.6157, 1.2315, 1.3672], device='cuda:1'), covar=tensor([0.1082, 0.0428, 0.1035, 0.1493], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0564, 0.0407, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:33:21,764 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 33500, giga_loss[loss=0.2235, simple_loss=0.3126, pruned_loss=0.06718, over 29127.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3323, pruned_loss=0.08653, over 5673882.12 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.343, pruned_loss=0.1079, over 5722995.57 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3318, pruned_loss=0.08445, over 5662832.97 frames. ], batch size: 113, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:33:38,563 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218365.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:34:02,502 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:34:04,223 INFO [optim.py:369] (1/2) 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 03:34:04,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-14 03:34:05,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1687, 1.3066, 1.1078, 0.9484], device='cuda:1'), covar=tensor([0.1064, 0.0506, 0.1148, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0444, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 03:34:21,671 INFO [train.py:968] (1/2) Epoch 27, batch 33550, libri_loss[loss=0.2268, simple_loss=0.2903, pruned_loss=0.08167, over 29500.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3346, pruned_loss=0.08733, over 5674364.68 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3425, pruned_loss=0.1076, over 5728541.11 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3345, pruned_loss=0.08534, over 5658932.05 frames. ], batch size: 70, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 03:34:40,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:35:23,460 INFO [train.py:968] (1/2) Epoch 27, batch 33600, giga_loss[loss=0.2292, simple_loss=0.3151, pruned_loss=0.07167, over 29081.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3346, pruned_loss=0.08745, over 5658977.56 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3429, pruned_loss=0.1079, over 5714770.27 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3338, pruned_loss=0.08513, over 5658438.41 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:35:36,836 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:35:40,137 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:35:41,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-14 03:35:55,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4310, 1.4029, 3.7800, 3.2217], device='cuda:1'), covar=tensor([0.1589, 0.2722, 0.0494, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0669, 0.0993, 0.0959], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 03:36:13,206 INFO [optim.py:369] (1/2) 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 03:36:15,872 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:36:30,760 INFO [train.py:968] (1/2) Epoch 27, batch 33650, giga_loss[loss=0.2308, simple_loss=0.3172, pruned_loss=0.07218, over 28703.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3322, pruned_loss=0.0868, over 5653536.44 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3429, pruned_loss=0.108, over 5709532.55 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3314, pruned_loss=0.08426, over 5656783.05 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:37:03,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:37:07,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:37:16,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4965, 1.6882, 1.2277, 1.3104], device='cuda:1'), covar=tensor([0.1077, 0.0609, 0.1087, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0406, 0.0443, 0.0520, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 03:37:30,287 INFO [train.py:968] (1/2) Epoch 27, batch 33700, giga_loss[loss=0.2249, simple_loss=0.3138, pruned_loss=0.06796, over 28961.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3328, pruned_loss=0.08751, over 5651832.63 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.343, pruned_loss=0.1081, over 5702443.21 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3318, pruned_loss=0.08492, over 5659696.73 frames. ], batch size: 120, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:37:39,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:37:47,279 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4201, 3.1599, 1.5017, 1.5520], device='cuda:1'), covar=tensor([0.0994, 0.0356, 0.0938, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0565, 0.0407, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:38:17,107 INFO [optim.py:369] (1/2) 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 03:38:33,539 INFO [train.py:968] (1/2) Epoch 27, batch 33750, giga_loss[loss=0.2963, simple_loss=0.3604, pruned_loss=0.1161, over 28495.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3332, pruned_loss=0.08838, over 5647711.57 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3433, pruned_loss=0.1083, over 5702720.14 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.0855, over 5652337.89 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:39:37,832 INFO [train.py:968] (1/2) Epoch 27, batch 33800, giga_loss[loss=0.2525, simple_loss=0.33, pruned_loss=0.08752, over 27590.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3314, pruned_loss=0.08825, over 5652806.42 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3434, pruned_loss=0.1084, over 5707307.81 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3299, pruned_loss=0.08529, over 5650906.52 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:40:21,877 INFO [optim.py:369] (1/2) 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 03:40:40,931 INFO [train.py:968] (1/2) Epoch 27, batch 33850, giga_loss[loss=0.2367, simple_loss=0.323, pruned_loss=0.07516, over 28505.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3296, pruned_loss=0.0876, over 5646211.34 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3426, pruned_loss=0.108, over 5711630.34 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3288, pruned_loss=0.08509, over 5639245.88 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:40:48,877 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1218712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:41:20,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218740.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:41:39,037 INFO [train.py:968] (1/2) Epoch 27, batch 33900, giga_loss[loss=0.236, simple_loss=0.3228, pruned_loss=0.07464, over 28697.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3287, pruned_loss=0.08552, over 5662696.72 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3422, pruned_loss=0.1078, over 5716773.40 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3281, pruned_loss=0.08326, over 5651000.83 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:42:21,043 INFO [optim.py:369] (1/2) 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 03:42:26,936 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:42:37,113 INFO [train.py:968] (1/2) Epoch 27, batch 33950, giga_loss[loss=0.268, simple_loss=0.3621, pruned_loss=0.08697, over 28622.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3288, pruned_loss=0.08371, over 5675158.58 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3422, pruned_loss=0.1077, over 5718399.95 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3282, pruned_loss=0.08154, over 5663652.03 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:43:13,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5039, 1.8528, 1.5427, 1.3618], device='cuda:1'), covar=tensor([0.2357, 0.2156, 0.2414, 0.2241], device='cuda:1'), in_proj_covar=tensor([0.1592, 0.1143, 0.1408, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 03:43:36,730 INFO [train.py:968] (1/2) Epoch 27, batch 34000, giga_loss[loss=0.2441, simple_loss=0.3359, pruned_loss=0.07616, over 28644.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3314, pruned_loss=0.08372, over 5670412.62 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3425, pruned_loss=0.1079, over 5719674.21 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3305, pruned_loss=0.08151, over 5659499.22 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:44:05,523 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218883.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:44:08,791 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:44:23,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-14 03:44:28,345 INFO [train.py:968] (1/2) Epoch 27, batch 34050, giga_loss[loss=0.2369, simple_loss=0.3268, pruned_loss=0.07348, over 28325.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3305, pruned_loss=0.083, over 5667382.05 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3423, pruned_loss=0.1079, over 5714138.17 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3295, pruned_loss=0.08028, over 5661057.82 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:44:43,906 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218915.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:44:58,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5495, 1.6979, 1.7864, 1.3757], device='cuda:1'), covar=tensor([0.2025, 0.2902, 0.1679, 0.1946], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0709, 0.0977, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 03:45:19,272 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1218939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:45:21,849 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1218942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:45:36,374 INFO [train.py:968] (1/2) Epoch 27, batch 34100, giga_loss[loss=0.2472, simple_loss=0.3328, pruned_loss=0.08079, over 28467.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3316, pruned_loss=0.08405, over 5662533.16 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3426, pruned_loss=0.1084, over 5707692.12 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3303, pruned_loss=0.08107, over 5662891.90 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:45:46,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5532, 2.1473, 1.5596, 0.9147], device='cuda:1'), covar=tensor([0.6829, 0.3235, 0.4943, 0.6934], device='cuda:1'), in_proj_covar=tensor([0.1836, 0.1729, 0.1658, 0.1502], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:45:58,272 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1218971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:46:22,763 INFO [optim.py:369] (1/2) 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 03:46:44,260 INFO [train.py:968] (1/2) Epoch 27, batch 34150, libri_loss[loss=0.2748, simple_loss=0.3383, pruned_loss=0.1056, over 29549.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3311, pruned_loss=0.08318, over 5663823.38 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3426, pruned_loss=0.1084, over 5708702.66 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3301, pruned_loss=0.08073, over 5662867.93 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:47:13,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219025.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:47:18,142 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-14 03:47:32,193 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7217, 2.4749, 1.5436, 1.0205], device='cuda:1'), covar=tensor([1.0002, 0.4951, 0.5416, 0.7770], device='cuda:1'), in_proj_covar=tensor([0.1836, 0.1730, 0.1659, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:47:57,609 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:47:57,965 INFO [train.py:968] (1/2) Epoch 27, batch 34200, giga_loss[loss=0.2221, simple_loss=0.3156, pruned_loss=0.06435, over 28148.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3302, pruned_loss=0.08205, over 5659671.78 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3424, pruned_loss=0.1083, over 5710490.14 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3294, pruned_loss=0.07993, over 5656935.31 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:48:43,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:48:49,619 INFO [optim.py:369] (1/2) 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 03:49:06,522 INFO [train.py:968] (1/2) Epoch 27, batch 34250, giga_loss[loss=0.2945, simple_loss=0.3722, pruned_loss=0.1084, over 28952.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3322, pruned_loss=0.0833, over 5658305.57 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3425, pruned_loss=0.1083, over 5709656.34 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3312, pruned_loss=0.08084, over 5655730.09 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:49:40,779 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:50:09,284 INFO [train.py:968] (1/2) Epoch 27, batch 34300, giga_loss[loss=0.2284, simple_loss=0.3163, pruned_loss=0.07027, over 28405.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3356, pruned_loss=0.08488, over 5657489.99 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3428, pruned_loss=0.1086, over 5703495.31 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3343, pruned_loss=0.08213, over 5659263.94 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:50:55,077 INFO [optim.py:369] (1/2) 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 03:51:14,702 INFO [train.py:968] (1/2) Epoch 27, batch 34350, giga_loss[loss=0.2588, simple_loss=0.3484, pruned_loss=0.08462, over 28918.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3349, pruned_loss=0.08447, over 5670632.81 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.342, pruned_loss=0.1082, over 5707495.75 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3345, pruned_loss=0.08215, over 5667587.77 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:51:49,925 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:51:53,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:52:22,095 INFO [train.py:968] (1/2) Epoch 27, batch 34400, giga_loss[loss=0.2379, simple_loss=0.3236, pruned_loss=0.07612, over 28737.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3336, pruned_loss=0.08438, over 5681027.45 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3425, pruned_loss=0.1085, over 5706992.24 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3327, pruned_loss=0.08197, over 5678709.71 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 03:52:34,693 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1219262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:53:05,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 03:53:19,687 INFO [optim.py:369] (1/2) 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 03:53:35,290 INFO [train.py:968] (1/2) Epoch 27, batch 34450, giga_loss[loss=0.2171, simple_loss=0.3097, pruned_loss=0.06222, over 28694.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3316, pruned_loss=0.08251, over 5685094.97 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3425, pruned_loss=0.1084, over 5709449.49 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3308, pruned_loss=0.08037, over 5680783.02 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:54:35,549 INFO [train.py:968] (1/2) Epoch 27, batch 34500, giga_loss[loss=0.2489, simple_loss=0.333, pruned_loss=0.08246, over 28515.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.331, pruned_loss=0.08223, over 5698885.13 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3426, pruned_loss=0.1084, over 5713986.37 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3298, pruned_loss=0.07961, over 5690845.92 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:54:46,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3778, 1.8156, 1.3157, 0.7921], device='cuda:1'), covar=tensor([0.6826, 0.3534, 0.3559, 0.7036], device='cuda:1'), in_proj_covar=tensor([0.1826, 0.1724, 0.1650, 0.1496], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:54:49,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4318, 3.3582, 1.5153, 1.6195], device='cuda:1'), covar=tensor([0.1006, 0.0357, 0.0958, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0565, 0.0408, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 03:55:27,115 INFO [optim.py:369] (1/2) 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 03:55:37,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219400.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 03:55:38,693 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8385, 2.6897, 1.5746, 1.1249], device='cuda:1'), covar=tensor([0.7082, 0.3262, 0.4296, 0.5365], device='cuda:1'), in_proj_covar=tensor([0.1826, 0.1723, 0.1650, 0.1496], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 03:55:41,039 INFO [train.py:968] (1/2) Epoch 27, batch 34550, giga_loss[loss=0.2173, simple_loss=0.3089, pruned_loss=0.06287, over 29056.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3313, pruned_loss=0.08229, over 5680452.23 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3427, pruned_loss=0.1085, over 5704761.44 frames. ], giga_tot_loss[loss=0.2452, simple_loss=0.3303, pruned_loss=0.08009, over 5683075.94 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:56:12,722 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:56:45,783 INFO [train.py:968] (1/2) Epoch 27, batch 34600, giga_loss[loss=0.2573, simple_loss=0.339, pruned_loss=0.08776, over 28247.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3339, pruned_loss=0.08376, over 5667534.97 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3426, pruned_loss=0.1084, over 5705520.63 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3331, pruned_loss=0.082, over 5668707.44 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:56:59,045 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-14 03:57:28,215 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 34650, giga_loss[loss=0.233, simple_loss=0.313, pruned_loss=0.0765, over 28546.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3326, pruned_loss=0.08396, over 5668808.81 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3429, pruned_loss=0.1087, over 5706342.69 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3316, pruned_loss=0.082, over 5668499.23 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:57:48,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:58:19,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2864, 4.1405, 3.9102, 2.0552], device='cuda:1'), covar=tensor([0.0575, 0.0686, 0.0739, 0.1985], device='cuda:1'), in_proj_covar=tensor([0.1270, 0.1168, 0.0984, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 03:58:22,926 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219543.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:58:25,517 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219546.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:58:28,487 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4470, 1.7825, 1.4076, 1.3081], device='cuda:1'), covar=tensor([0.2811, 0.2640, 0.3158, 0.2469], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1141, 0.1409, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 03:58:32,161 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 34700, giga_loss[loss=0.2597, simple_loss=0.3379, pruned_loss=0.09077, over 28882.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3314, pruned_loss=0.08448, over 5659566.39 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3433, pruned_loss=0.1091, over 5695105.78 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.33, pruned_loss=0.082, over 5667948.76 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 03:58:57,300 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:58:58,102 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:59:02,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1219575.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 03:59:02,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 03:59:19,141 INFO [optim.py:369] (1/2) 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 03:59:31,190 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-14 03:59:34,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 34750, giga_loss[loss=0.2807, simple_loss=0.3615, pruned_loss=0.09999, over 28926.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3311, pruned_loss=0.08467, over 5662646.03 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3429, pruned_loss=0.1088, over 5698466.49 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3302, pruned_loss=0.08259, over 5665968.88 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:00:01,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0205, 3.8751, 3.6824, 1.8468], device='cuda:1'), covar=tensor([0.0658, 0.0757, 0.0810, 0.2280], device='cuda:1'), in_proj_covar=tensor([0.1269, 0.1168, 0.0984, 0.0727], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 04:00:18,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1219650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:00:21,960 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1219653.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:00:21,996 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1219653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:00:22,992 INFO [train.py:968] (1/2) Epoch 27, batch 34800, giga_loss[loss=0.2927, simple_loss=0.374, pruned_loss=0.1057, over 29041.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3395, pruned_loss=0.09062, over 5653416.86 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3429, pruned_loss=0.1091, over 5690919.50 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3385, pruned_loss=0.08813, over 5662262.98 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:00:49,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1219682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:00:51,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6038, 3.4943, 1.5794, 1.6849], device='cuda:1'), covar=tensor([0.0900, 0.0394, 0.0917, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0565, 0.0408, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 04:00:57,443 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 34850, giga_loss[loss=0.268, simple_loss=0.3595, pruned_loss=0.08826, over 28923.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3471, pruned_loss=0.09473, over 5670974.95 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3424, pruned_loss=0.1088, over 5695694.40 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09263, over 5672961.40 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:01:12,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4923, 1.8287, 1.6840, 1.6529], device='cuda:1'), covar=tensor([0.2244, 0.2048, 0.2353, 0.2157], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0740, 0.0713, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 04:01:17,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8490, 1.4430, 5.2738, 3.6583], device='cuda:1'), covar=tensor([0.1598, 0.2748, 0.0470, 0.1009], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0669, 0.0995, 0.0963], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 04:01:49,665 INFO [train.py:968] (1/2) Epoch 27, batch 34900, giga_loss[loss=0.262, simple_loss=0.3393, pruned_loss=0.09234, over 28899.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3489, pruned_loss=0.09622, over 5676798.17 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3422, pruned_loss=0.1086, over 5699995.01 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.349, pruned_loss=0.09449, over 5674037.38 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:02:01,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6209, 1.8667, 1.8384, 1.6472], device='cuda:1'), covar=tensor([0.2193, 0.2239, 0.2212, 0.2263], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0740, 0.0714, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 04:02:21,218 INFO [optim.py:369] (1/2) 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 04:02:30,307 INFO [train.py:968] (1/2) Epoch 27, batch 34950, giga_loss[loss=0.214, simple_loss=0.2917, pruned_loss=0.06809, over 28608.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3434, pruned_loss=0.09385, over 5680323.82 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3424, pruned_loss=0.1087, over 5703074.66 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3434, pruned_loss=0.09224, over 5675076.06 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:03:11,983 INFO [train.py:968] (1/2) Epoch 27, batch 35000, giga_loss[loss=0.2041, simple_loss=0.2804, pruned_loss=0.06386, over 28865.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3354, pruned_loss=0.0902, over 5680604.04 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3422, pruned_loss=0.1084, over 5706292.58 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3355, pruned_loss=0.08891, over 5673463.62 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:03:40,552 INFO [optim.py:369] (1/2) 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 04:03:52,007 INFO [train.py:968] (1/2) Epoch 27, batch 35050, giga_loss[loss=0.2195, simple_loss=0.3008, pruned_loss=0.06913, over 28475.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3289, pruned_loss=0.08732, over 5689206.32 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3425, pruned_loss=0.1083, over 5710694.78 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3284, pruned_loss=0.08587, over 5679128.07 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:04:08,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:04:26,074 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1219946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:04:32,068 INFO [train.py:968] (1/2) Epoch 27, batch 35100, giga_loss[loss=0.2118, simple_loss=0.2856, pruned_loss=0.06898, over 28869.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.321, pruned_loss=0.08388, over 5693645.42 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3428, pruned_loss=0.1085, over 5714262.60 frames. ], giga_tot_loss[loss=0.2421, simple_loss=0.32, pruned_loss=0.08214, over 5682006.13 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:05:05,371 INFO [optim.py:369] (1/2) 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 04:05:07,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5451, 3.4668, 1.5784, 1.7627], device='cuda:1'), covar=tensor([0.0998, 0.0329, 0.0938, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0565, 0.0406, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 04:05:15,769 INFO [train.py:968] (1/2) Epoch 27, batch 35150, giga_loss[loss=0.2223, simple_loss=0.2961, pruned_loss=0.07429, over 28631.00 frames. ], tot_loss[loss=0.2387, simple_loss=0.3149, pruned_loss=0.08125, over 5684048.83 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3428, pruned_loss=0.1084, over 5707332.05 frames. ], giga_tot_loss[loss=0.2365, simple_loss=0.3138, pruned_loss=0.07963, over 5680489.01 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:05:26,451 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6411, 1.7802, 1.6958, 1.4903], device='cuda:1'), covar=tensor([0.3853, 0.3089, 0.2580, 0.3502], device='cuda:1'), in_proj_covar=tensor([0.2022, 0.1970, 0.1880, 0.2025], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:05:36,668 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220028.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:05:57,781 INFO [train.py:968] (1/2) Epoch 27, batch 35200, giga_loss[loss=0.2434, simple_loss=0.3133, pruned_loss=0.08675, over 28579.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.3118, pruned_loss=0.08003, over 5694443.99 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3428, pruned_loss=0.1083, over 5709295.02 frames. ], giga_tot_loss[loss=0.234, simple_loss=0.3107, pruned_loss=0.0786, over 5689660.96 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:06:07,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7182, 1.9450, 1.4917, 1.4260], device='cuda:1'), covar=tensor([0.1080, 0.0632, 0.1116, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0443, 0.0520, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 04:06:09,365 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:11,200 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:26,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:29,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:29,387 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:06:38,641 INFO [train.py:968] (1/2) Epoch 27, batch 35250, giga_loss[loss=0.2318, simple_loss=0.3085, pruned_loss=0.07752, over 28637.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.3075, pruned_loss=0.07804, over 5692743.78 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3425, pruned_loss=0.1081, over 5711246.33 frames. ], giga_tot_loss[loss=0.2299, simple_loss=0.3064, pruned_loss=0.07671, over 5687004.53 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:06:48,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3616, 1.2449, 1.1973, 1.4798], device='cuda:1'), covar=tensor([0.0727, 0.0422, 0.0376, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 04:06:51,993 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:07:19,586 INFO [train.py:968] (1/2) Epoch 27, batch 35300, giga_loss[loss=0.2058, simple_loss=0.2854, pruned_loss=0.06307, over 28799.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.3069, pruned_loss=0.07827, over 5676045.16 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.343, pruned_loss=0.108, over 5706513.37 frames. ], giga_tot_loss[loss=0.228, simple_loss=0.304, pruned_loss=0.07602, over 5674289.66 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:07:35,471 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220171.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:07:37,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220174.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:07:54,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6915, 1.8904, 1.6042, 1.6202], device='cuda:1'), covar=tensor([0.2651, 0.2738, 0.2967, 0.2751], device='cuda:1'), in_proj_covar=tensor([0.1594, 0.1145, 0.1412, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 04:07:55,085 INFO [optim.py:369] (1/2) 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 04:08:02,228 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220203.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:08:03,237 INFO [train.py:968] (1/2) Epoch 27, batch 35350, libri_loss[loss=0.2852, simple_loss=0.3614, pruned_loss=0.1045, over 29661.00 frames. ], tot_loss[loss=0.228, simple_loss=0.3032, pruned_loss=0.07642, over 5672964.23 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.343, pruned_loss=0.1078, over 5699707.47 frames. ], giga_tot_loss[loss=0.2244, simple_loss=0.3003, pruned_loss=0.07422, over 5677487.95 frames. ], batch size: 91, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:08:36,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 04:08:43,916 INFO [train.py:968] (1/2) Epoch 27, batch 35400, giga_loss[loss=0.2132, simple_loss=0.2875, pruned_loss=0.06951, over 28736.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.3026, pruned_loss=0.07659, over 5685957.33 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3435, pruned_loss=0.108, over 5706259.18 frames. ], giga_tot_loss[loss=0.2225, simple_loss=0.2982, pruned_loss=0.07341, over 5682820.24 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:09:12,813 INFO [optim.py:369] (1/2) 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 04:09:20,947 INFO [train.py:968] (1/2) Epoch 27, batch 35450, libri_loss[loss=0.3092, simple_loss=0.3718, pruned_loss=0.1233, over 28541.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.3026, pruned_loss=0.07706, over 5693490.30 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3441, pruned_loss=0.1082, over 5710943.96 frames. ], giga_tot_loss[loss=0.2212, simple_loss=0.2966, pruned_loss=0.07288, over 5685723.39 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:09:46,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:09:53,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7042, 2.0076, 1.5165, 1.4903], device='cuda:1'), covar=tensor([0.0992, 0.0569, 0.1030, 0.1182], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0442, 0.0520, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 04:10:01,030 INFO [train.py:968] (1/2) Epoch 27, batch 35500, libri_loss[loss=0.297, simple_loss=0.3536, pruned_loss=0.1202, over 29339.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.3006, pruned_loss=0.07645, over 5675005.30 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3447, pruned_loss=0.1087, over 5693397.72 frames. ], giga_tot_loss[loss=0.2187, simple_loss=0.2939, pruned_loss=0.0717, over 5683066.45 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:10:05,343 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:10:36,276 INFO [optim.py:369] (1/2) 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 04:10:46,593 INFO [train.py:968] (1/2) Epoch 27, batch 35550, giga_loss[loss=0.2086, simple_loss=0.2675, pruned_loss=0.07481, over 23829.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2967, pruned_loss=0.07479, over 5666348.96 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3447, pruned_loss=0.1087, over 5694483.73 frames. ], giga_tot_loss[loss=0.2166, simple_loss=0.2912, pruned_loss=0.07095, over 5671605.81 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:10:58,090 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220416.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:11:02,879 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:11:28,555 INFO [train.py:968] (1/2) Epoch 27, batch 35600, giga_loss[loss=0.2431, simple_loss=0.3231, pruned_loss=0.0816, over 29055.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.2992, pruned_loss=0.07676, over 5671033.97 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.345, pruned_loss=0.1087, over 5697499.95 frames. ], giga_tot_loss[loss=0.2193, simple_loss=0.2933, pruned_loss=0.07268, over 5671732.64 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:12:00,870 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-14 04:12:04,001 INFO [optim.py:369] (1/2) 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 04:12:13,515 INFO [train.py:968] (1/2) Epoch 27, batch 35650, giga_loss[loss=0.3231, simple_loss=0.3866, pruned_loss=0.1298, over 28942.00 frames. ], tot_loss[loss=0.2363, simple_loss=0.3096, pruned_loss=0.0815, over 5681588.76 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3447, pruned_loss=0.1085, over 5698021.81 frames. ], giga_tot_loss[loss=0.2301, simple_loss=0.3044, pruned_loss=0.07791, over 5681497.04 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:12:21,809 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-14 04:12:52,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7802, 2.6337, 1.7064, 0.9478], device='cuda:1'), covar=tensor([0.8824, 0.4003, 0.4380, 0.7791], device='cuda:1'), in_proj_covar=tensor([0.1837, 0.1739, 0.1661, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 04:12:57,516 INFO [train.py:968] (1/2) Epoch 27, batch 35700, giga_loss[loss=0.3065, simple_loss=0.3787, pruned_loss=0.1172, over 28670.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3232, pruned_loss=0.08884, over 5685454.45 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.345, pruned_loss=0.1087, over 5703363.85 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.318, pruned_loss=0.08521, over 5680212.27 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:13:12,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7232, 1.0403, 2.9417, 2.8049], device='cuda:1'), covar=tensor([0.1789, 0.2635, 0.0663, 0.1101], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0669, 0.1000, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 04:13:28,153 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 35750, giga_loss[loss=0.3177, simple_loss=0.3866, pruned_loss=0.1244, over 28917.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3333, pruned_loss=0.09384, over 5667848.73 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3459, pruned_loss=0.1093, over 5685161.40 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.328, pruned_loss=0.09008, over 5681006.48 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:14:08,975 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1220642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:14:18,645 INFO [train.py:968] (1/2) Epoch 27, batch 35800, giga_loss[loss=0.2816, simple_loss=0.3682, pruned_loss=0.09744, over 28784.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3394, pruned_loss=0.09566, over 5654655.67 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3462, pruned_loss=0.1094, over 5671654.94 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3346, pruned_loss=0.09217, over 5677266.33 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:14:31,431 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5505, 1.0568, 4.8521, 3.4885], device='cuda:1'), covar=tensor([0.1827, 0.3208, 0.0402, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0670, 0.1001, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 04:14:50,765 INFO [optim.py:369] (1/2) 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 04:14:57,525 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6533, 4.4680, 4.2271, 1.8736], device='cuda:1'), covar=tensor([0.0540, 0.0745, 0.0740, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.1273, 0.1177, 0.0986, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 04:14:59,346 INFO [train.py:968] (1/2) Epoch 27, batch 35850, giga_loss[loss=0.2537, simple_loss=0.3396, pruned_loss=0.08392, over 29041.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3408, pruned_loss=0.0954, over 5656298.84 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3468, pruned_loss=0.1096, over 5676826.30 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3361, pruned_loss=0.0918, over 5669211.88 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:15:02,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 04:15:03,870 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:15:26,844 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:15:39,800 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 04:15:46,701 INFO [train.py:968] (1/2) Epoch 27, batch 35900, giga_loss[loss=0.2668, simple_loss=0.3532, pruned_loss=0.09021, over 28891.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3416, pruned_loss=0.09477, over 5658905.10 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3468, pruned_loss=0.1096, over 5679019.16 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.338, pruned_loss=0.09186, over 5667143.58 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:16:16,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1220791.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:16:19,192 INFO [optim.py:369] (1/2) 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 04:16:22,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 35950, giga_loss[loss=0.2845, simple_loss=0.3576, pruned_loss=0.1057, over 28770.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3437, pruned_loss=0.09629, over 5673554.41 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3467, pruned_loss=0.1094, over 5681394.01 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3409, pruned_loss=0.09394, over 5677737.41 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:17:08,201 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-14 04:17:09,458 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:10,453 INFO [train.py:968] (1/2) Epoch 27, batch 36000, giga_loss[loss=0.2684, simple_loss=0.3447, pruned_loss=0.09607, over 28960.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3473, pruned_loss=0.09901, over 5667993.20 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.347, pruned_loss=0.1095, over 5676610.92 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3446, pruned_loss=0.09674, over 5676199.47 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:17:10,453 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 04:17:20,036 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 04:17:21,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:38,385 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:39,970 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:43,586 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:17:50,739 INFO [optim.py:369] (1/2) 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 04:17:57,480 INFO [train.py:968] (1/2) Epoch 27, batch 36050, giga_loss[loss=0.2737, simple_loss=0.3577, pruned_loss=0.09482, over 28679.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3502, pruned_loss=0.1012, over 5666034.96 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3478, pruned_loss=0.1102, over 5667468.76 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3474, pruned_loss=0.09817, over 5680568.41 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:18:02,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:18:07,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 04:18:21,034 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220934.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 04:18:23,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220937.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:18:25,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1220940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:18:27,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5608, 4.3550, 4.2246, 1.9175], device='cuda:1'), covar=tensor([0.0748, 0.0884, 0.1016, 0.1968], device='cuda:1'), in_proj_covar=tensor([0.1264, 0.1167, 0.0978, 0.0730], device='cuda:1'), out_proj_covar=tensor([0.0018, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 04:18:27,867 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1220943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:18:36,583 INFO [train.py:968] (1/2) Epoch 27, batch 36100, giga_loss[loss=0.2696, simple_loss=0.3541, pruned_loss=0.09253, over 29085.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.354, pruned_loss=0.1018, over 5686860.07 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3478, pruned_loss=0.1101, over 5670649.20 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3518, pruned_loss=0.09945, over 5695674.51 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:18:48,304 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220966.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 04:18:51,867 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1220972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:19:13,235 INFO [optim.py:369] (1/2) 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 04:19:18,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2504, 2.9173, 1.4162, 1.4046], device='cuda:1'), covar=tensor([0.1102, 0.0311, 0.0966, 0.1454], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0564, 0.0406, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 04:19:19,221 INFO [train.py:968] (1/2) Epoch 27, batch 36150, giga_loss[loss=0.2796, simple_loss=0.3613, pruned_loss=0.09894, over 28877.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.355, pruned_loss=0.1019, over 5688685.63 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3477, pruned_loss=0.1099, over 5677840.63 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3536, pruned_loss=0.09998, over 5689472.99 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:19:31,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1221017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:20:01,773 INFO [train.py:968] (1/2) Epoch 27, batch 36200, giga_loss[loss=0.2918, simple_loss=0.367, pruned_loss=0.1083, over 28774.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3555, pruned_loss=0.1009, over 5691101.04 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3475, pruned_loss=0.1097, over 5676584.43 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3545, pruned_loss=0.09946, over 5692754.09 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:20:02,216 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-14 04:20:34,462 INFO [optim.py:369] (1/2) 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 04:20:41,310 INFO [train.py:968] (1/2) Epoch 27, batch 36250, giga_loss[loss=0.2456, simple_loss=0.3382, pruned_loss=0.0765, over 28644.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3543, pruned_loss=0.09898, over 5696834.57 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3477, pruned_loss=0.1097, over 5680678.44 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3536, pruned_loss=0.09766, over 5694730.25 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:21:02,960 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.33 vs. limit=2.0
+2023-03-14 04:21:12,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1510, 1.4342, 1.5082, 1.3133], device='cuda:1'), covar=tensor([0.2512, 0.2036, 0.2723, 0.2281], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0752, 0.0723, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 04:21:21,825 INFO [train.py:968] (1/2) Epoch 27, batch 36300, giga_loss[loss=0.2555, simple_loss=0.3419, pruned_loss=0.08454, over 28672.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3533, pruned_loss=0.09788, over 5689343.94 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3482, pruned_loss=0.11, over 5673912.59 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3524, pruned_loss=0.09636, over 5694760.26 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:21:26,669 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1221160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:21:28,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1221163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:21:48,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3775, 1.4871, 1.4779, 1.3450], device='cuda:1'), covar=tensor([0.3128, 0.2846, 0.2576, 0.2825], device='cuda:1'), in_proj_covar=tensor([0.2041, 0.1983, 0.1899, 0.2043], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:21:51,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8741, 2.1726, 2.1069, 1.6624], device='cuda:1'), covar=tensor([0.3749, 0.2889, 0.3076, 0.3605], device='cuda:1'), in_proj_covar=tensor([0.2041, 0.1984, 0.1899, 0.2043], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:21:53,193 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1221192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:21:56,633 INFO [optim.py:369] (1/2) 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 04:21:58,971 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-14 04:22:03,144 INFO [train.py:968] (1/2) Epoch 27, batch 36350, giga_loss[loss=0.2268, simple_loss=0.3172, pruned_loss=0.06819, over 28508.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3517, pruned_loss=0.09699, over 5684515.74 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.348, pruned_loss=0.1098, over 5675594.50 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3511, pruned_loss=0.09579, over 5687512.33 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:22:43,583 INFO [train.py:968] (1/2) Epoch 27, batch 36400, giga_loss[loss=0.2796, simple_loss=0.3533, pruned_loss=0.103, over 28882.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3539, pruned_loss=0.1007, over 5683984.17 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3481, pruned_loss=0.1097, over 5680957.07 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3535, pruned_loss=0.09946, over 5682065.57 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:23:19,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6317, 1.9246, 1.5538, 1.6462], device='cuda:1'), covar=tensor([0.2542, 0.2567, 0.2968, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1143, 0.1401, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 04:23:20,926 INFO [optim.py:369] (1/2) 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 04:23:26,141 INFO [train.py:968] (1/2) Epoch 27, batch 36450, giga_loss[loss=0.2665, simple_loss=0.3413, pruned_loss=0.09591, over 28801.00 frames. ], tot_loss[loss=0.2823, simple_loss=0.3563, pruned_loss=0.1042, over 5693917.39 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3484, pruned_loss=0.1098, over 5686413.03 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3559, pruned_loss=0.1029, over 5687350.23 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:24:08,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4136, 2.0665, 1.5850, 0.7435], device='cuda:1'), covar=tensor([0.7201, 0.3296, 0.4355, 0.7200], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1727, 0.1656, 0.1498], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 04:24:09,293 INFO [train.py:968] (1/2) Epoch 27, batch 36500, giga_loss[loss=0.2625, simple_loss=0.3368, pruned_loss=0.09411, over 28982.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3556, pruned_loss=0.1052, over 5681721.90 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3487, pruned_loss=0.11, over 5680059.85 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3551, pruned_loss=0.104, over 5683099.81 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:24:19,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3961, 1.7307, 1.4001, 1.0571], device='cuda:1'), covar=tensor([0.2511, 0.2639, 0.2945, 0.2411], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1144, 0.1402, 0.1008], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 04:24:32,046 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1221382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:24:35,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-14 04:24:45,970 INFO [optim.py:369] (1/2) 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 04:24:48,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5549, 2.9968, 2.5655, 2.1737], device='cuda:1'), covar=tensor([0.2478, 0.1663, 0.2040, 0.2328], device='cuda:1'), in_proj_covar=tensor([0.2044, 0.1989, 0.1903, 0.2048], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:24:51,479 INFO [train.py:968] (1/2) Epoch 27, batch 36550, giga_loss[loss=0.2711, simple_loss=0.3427, pruned_loss=0.09981, over 28896.00 frames. ], tot_loss[loss=0.2811, simple_loss=0.3535, pruned_loss=0.1043, over 5692196.48 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.349, pruned_loss=0.11, over 5681910.31 frames. ], giga_tot_loss[loss=0.2797, simple_loss=0.353, pruned_loss=0.1032, over 5691677.99 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:25:32,757 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2136, 1.2633, 3.7082, 3.1921], device='cuda:1'), covar=tensor([0.1650, 0.2822, 0.0459, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0671, 0.1003, 0.0970], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 04:25:33,081 INFO [train.py:968] (1/2) Epoch 27, batch 36600, giga_loss[loss=0.2536, simple_loss=0.3348, pruned_loss=0.08616, over 28758.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3519, pruned_loss=0.1039, over 5697639.10 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3488, pruned_loss=0.1098, over 5687227.09 frames. ], giga_tot_loss[loss=0.279, simple_loss=0.3518, pruned_loss=0.1031, over 5692641.23 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:26:07,889 INFO [optim.py:369] (1/2) 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 04:26:11,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3836, 1.5195, 1.5507, 1.3888], device='cuda:1'), covar=tensor([0.3771, 0.3305, 0.2453, 0.2873], device='cuda:1'), in_proj_covar=tensor([0.2047, 0.1992, 0.1906, 0.2049], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:26:16,758 INFO [train.py:968] (1/2) Epoch 27, batch 36650, giga_loss[loss=0.2594, simple_loss=0.343, pruned_loss=0.08792, over 29043.00 frames. ], tot_loss[loss=0.2781, simple_loss=0.351, pruned_loss=0.1026, over 5695574.89 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.349, pruned_loss=0.1099, over 5690346.89 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3508, pruned_loss=0.1018, over 5689030.77 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:26:39,490 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-14 04:26:58,330 INFO [train.py:968] (1/2) Epoch 27, batch 36700, giga_loss[loss=0.2488, simple_loss=0.3231, pruned_loss=0.08722, over 28551.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3481, pruned_loss=0.1001, over 5707610.80 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3487, pruned_loss=0.1096, over 5695863.26 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3482, pruned_loss=0.09945, over 5697601.57 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:27:03,910 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6493, 1.7811, 1.5019, 1.8728], device='cuda:1'), covar=tensor([0.2756, 0.2895, 0.3131, 0.2604], device='cuda:1'), in_proj_covar=tensor([0.1585, 0.1142, 0.1403, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 04:27:37,586 INFO [optim.py:369] (1/2) 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 04:27:42,735 INFO [train.py:968] (1/2) Epoch 27, batch 36750, giga_loss[loss=0.2315, simple_loss=0.3104, pruned_loss=0.0763, over 28848.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3436, pruned_loss=0.09752, over 5704399.40 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3493, pruned_loss=0.1098, over 5701256.46 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.343, pruned_loss=0.09642, over 5691317.71 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:28:31,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4011, 1.5640, 1.3204, 1.5218], device='cuda:1'), covar=tensor([0.0809, 0.0360, 0.0361, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 04:28:34,069 INFO [train.py:968] (1/2) Epoch 27, batch 36800, giga_loss[loss=0.2605, simple_loss=0.3304, pruned_loss=0.09524, over 28839.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3377, pruned_loss=0.09432, over 5699299.34 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3497, pruned_loss=0.11, over 5702625.78 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3367, pruned_loss=0.09313, over 5687865.64 frames. ], batch size: 243, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:29:19,516 INFO [optim.py:369] (1/2) 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 04:29:28,444 INFO [train.py:968] (1/2) Epoch 27, batch 36850, giga_loss[loss=0.24, simple_loss=0.3231, pruned_loss=0.07842, over 28973.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3328, pruned_loss=0.09185, over 5683973.05 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3498, pruned_loss=0.1099, over 5701851.52 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3317, pruned_loss=0.09076, over 5675639.83 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:29:44,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2221, 3.0810, 2.9013, 1.4769], device='cuda:1'), covar=tensor([0.1019, 0.1046, 0.0854, 0.2441], device='cuda:1'), in_proj_covar=tensor([0.1271, 0.1177, 0.0987, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 04:29:56,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6613, 1.8516, 1.5443, 1.7565], device='cuda:1'), covar=tensor([0.2761, 0.2870, 0.3163, 0.2460], device='cuda:1'), in_proj_covar=tensor([0.1581, 0.1141, 0.1401, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 04:30:08,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2868, 0.8077, 0.9431, 1.4960], device='cuda:1'), covar=tensor([0.0818, 0.0410, 0.0380, 0.0893], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 04:30:13,434 INFO [train.py:968] (1/2) Epoch 27, batch 36900, giga_loss[loss=0.2308, simple_loss=0.3121, pruned_loss=0.07471, over 28758.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3337, pruned_loss=0.092, over 5684234.58 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3498, pruned_loss=0.1098, over 5705858.54 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3326, pruned_loss=0.09088, over 5673627.16 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:30:15,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1221757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:30:48,002 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 36950, giga_loss[loss=0.2768, simple_loss=0.3475, pruned_loss=0.1031, over 28902.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3341, pruned_loss=0.09165, over 5695665.47 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3497, pruned_loss=0.1097, over 5707018.47 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.333, pruned_loss=0.09056, over 5686030.26 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:31:06,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-14 04:31:33,326 INFO [train.py:968] (1/2) Epoch 27, batch 37000, giga_loss[loss=0.2579, simple_loss=0.3262, pruned_loss=0.09479, over 28981.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3341, pruned_loss=0.09189, over 5698689.30 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3501, pruned_loss=0.1097, over 5710368.04 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3326, pruned_loss=0.0906, over 5687876.19 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:31:39,405 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1221864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:07,520 INFO [optim.py:369] (1/2) 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 04:32:08,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1221900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:12,411 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1221903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:13,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5100, 3.5291, 1.6905, 1.5512], device='cuda:1'), covar=tensor([0.1061, 0.0342, 0.0892, 0.1443], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0561, 0.0404, 0.0440], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 04:32:13,494 INFO [train.py:968] (1/2) Epoch 27, batch 37050, giga_loss[loss=0.2221, simple_loss=0.2979, pruned_loss=0.07311, over 28483.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3326, pruned_loss=0.09152, over 5698836.52 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3506, pruned_loss=0.1099, over 5713173.78 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3307, pruned_loss=0.09006, over 5687659.45 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:32:33,489 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1221932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:32:50,361 INFO [train.py:968] (1/2) Epoch 27, batch 37100, libri_loss[loss=0.2462, simple_loss=0.3215, pruned_loss=0.08549, over 29500.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3302, pruned_loss=0.09018, over 5709959.79 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3508, pruned_loss=0.1098, over 5715046.72 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.328, pruned_loss=0.08861, over 5699327.65 frames. ], batch size: 70, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:33:10,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5148, 2.3404, 1.7538, 0.6270], device='cuda:1'), covar=tensor([0.5977, 0.2585, 0.4112, 0.6032], device='cuda:1'), in_proj_covar=tensor([0.1817, 0.1719, 0.1649, 0.1492], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 04:33:22,507 INFO [optim.py:369] (1/2) 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 04:33:27,894 INFO [train.py:968] (1/2) Epoch 27, batch 37150, giga_loss[loss=0.2323, simple_loss=0.3127, pruned_loss=0.0759, over 28961.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3289, pruned_loss=0.08977, over 5718499.07 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3518, pruned_loss=0.1105, over 5719608.91 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3255, pruned_loss=0.08732, over 5705759.25 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:33:33,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 04:34:07,670 INFO [train.py:968] (1/2) Epoch 27, batch 37200, giga_loss[loss=0.2131, simple_loss=0.2955, pruned_loss=0.06531, over 28875.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3278, pruned_loss=0.08941, over 5713106.21 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3523, pruned_loss=0.1105, over 5719628.79 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3243, pruned_loss=0.08701, over 5702908.97 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:34:16,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2671, 1.3720, 1.3665, 1.2397], device='cuda:1'), covar=tensor([0.2740, 0.2634, 0.1965, 0.2619], device='cuda:1'), in_proj_covar=tensor([0.2034, 0.1978, 0.1891, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:34:18,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4522, 3.4795, 1.5645, 1.6517], device='cuda:1'), covar=tensor([0.1047, 0.0308, 0.0892, 0.1318], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0562, 0.0404, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 04:34:41,605 INFO [optim.py:369] (1/2) 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 04:34:46,521 INFO [train.py:968] (1/2) Epoch 27, batch 37250, giga_loss[loss=0.2204, simple_loss=0.2995, pruned_loss=0.07062, over 29044.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3252, pruned_loss=0.08823, over 5705984.42 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3525, pruned_loss=0.1106, over 5714384.48 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3217, pruned_loss=0.08582, over 5703210.33 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:35:23,315 INFO [train.py:968] (1/2) Epoch 27, batch 37300, libri_loss[loss=0.2943, simple_loss=0.368, pruned_loss=0.1103, over 29537.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3257, pruned_loss=0.08829, over 5714050.12 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3533, pruned_loss=0.1105, over 5714015.75 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3206, pruned_loss=0.0852, over 5711224.74 frames. ], batch size: 84, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:35:56,705 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 37350, giga_loss[loss=0.2373, simple_loss=0.3152, pruned_loss=0.0797, over 28941.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.324, pruned_loss=0.08752, over 5715857.43 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3537, pruned_loss=0.1106, over 5710317.33 frames. ], giga_tot_loss[loss=0.2434, simple_loss=0.3185, pruned_loss=0.08417, over 5716852.31 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:36:12,963 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:36:25,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4194, 1.7077, 1.4371, 1.5512], device='cuda:1'), covar=tensor([0.0805, 0.0331, 0.0339, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 04:36:31,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3850, 1.4214, 1.3487, 1.5138], device='cuda:1'), covar=tensor([0.0810, 0.0366, 0.0344, 0.0895], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 04:36:31,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1222239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:36:42,383 INFO [train.py:968] (1/2) Epoch 27, batch 37400, giga_loss[loss=0.2352, simple_loss=0.3086, pruned_loss=0.08083, over 29084.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3217, pruned_loss=0.08614, over 5714043.14 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.354, pruned_loss=0.1107, over 5702569.15 frames. ], giga_tot_loss[loss=0.2417, simple_loss=0.3169, pruned_loss=0.08322, over 5721975.41 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:37:01,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-14 04:37:09,492 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:37:19,665 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 37450, giga_loss[loss=0.3614, simple_loss=0.4049, pruned_loss=0.1589, over 26495.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3218, pruned_loss=0.08647, over 5713445.72 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3544, pruned_loss=0.1108, over 5703817.77 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3173, pruned_loss=0.08392, over 5718724.74 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:37:35,197 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.62 vs. limit=5.0
+2023-03-14 04:37:51,431 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-14 04:38:05,943 INFO [train.py:968] (1/2) Epoch 27, batch 37500, giga_loss[loss=0.2753, simple_loss=0.3474, pruned_loss=0.1016, over 29006.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3257, pruned_loss=0.08883, over 5699304.91 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3548, pruned_loss=0.1109, over 5693222.21 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3215, pruned_loss=0.08645, over 5712395.50 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:38:26,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5162, 1.7494, 1.3044, 1.2748], device='cuda:1'), covar=tensor([0.1161, 0.0620, 0.1108, 0.1199], device='cuda:1'), in_proj_covar=tensor([0.0407, 0.0444, 0.0520, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 04:38:28,555 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1222382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:38:31,079 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 27, batch 37550, giga_loss[loss=0.3104, simple_loss=0.3759, pruned_loss=0.1224, over 28860.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3318, pruned_loss=0.09276, over 5697703.46 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3553, pruned_loss=0.1111, over 5695804.59 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3272, pruned_loss=0.09005, over 5706185.66 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:38:57,050 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1222414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:39:14,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 04:39:34,365 INFO [train.py:968] (1/2) Epoch 27, batch 37600, giga_loss[loss=0.2784, simple_loss=0.3464, pruned_loss=0.1052, over 28749.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3402, pruned_loss=0.09844, over 5685488.73 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3556, pruned_loss=0.1113, over 5690447.56 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.336, pruned_loss=0.09584, over 5697597.29 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:40:17,198 INFO [optim.py:369] (1/2) 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 04:40:21,916 INFO [train.py:968] (1/2) Epoch 27, batch 37650, giga_loss[loss=0.2659, simple_loss=0.3498, pruned_loss=0.09106, over 28637.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3442, pruned_loss=0.1001, over 5654310.43 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3559, pruned_loss=0.1116, over 5675558.90 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3403, pruned_loss=0.09763, over 5677448.94 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:40:23,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6085, 1.9636, 1.9417, 1.6396], device='cuda:1'), covar=tensor([0.2727, 0.1936, 0.1637, 0.2117], device='cuda:1'), in_proj_covar=tensor([0.2042, 0.1988, 0.1904, 0.2052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:40:35,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4824, 1.9559, 1.7140, 1.5899], device='cuda:1'), covar=tensor([0.0681, 0.0266, 0.0280, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 04:40:42,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7190, 2.0759, 1.3823, 1.5627], device='cuda:1'), covar=tensor([0.1168, 0.0614, 0.1124, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0446, 0.0523, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 04:41:01,221 INFO [train.py:968] (1/2) Epoch 27, batch 37700, giga_loss[loss=0.2764, simple_loss=0.3565, pruned_loss=0.09818, over 28867.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3492, pruned_loss=0.1026, over 5669554.64 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3558, pruned_loss=0.1116, over 5682803.70 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3458, pruned_loss=0.1002, over 5681172.44 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:41:17,602 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:41:34,940 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1222592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:41:37,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7521, 1.8995, 1.8385, 1.6740], device='cuda:1'), covar=tensor([0.2187, 0.2708, 0.2333, 0.2385], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0753, 0.0721, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 04:41:43,343 INFO [optim.py:369] (1/2) 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 04:41:44,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3480, 3.4723, 1.5408, 1.4792], device='cuda:1'), covar=tensor([0.1110, 0.0296, 0.0973, 0.1462], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0563, 0.0404, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 04:41:47,429 INFO [train.py:968] (1/2) Epoch 27, batch 37750, giga_loss[loss=0.3188, simple_loss=0.385, pruned_loss=0.1263, over 28696.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3527, pruned_loss=0.1041, over 5662992.22 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3556, pruned_loss=0.1113, over 5685027.39 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3502, pruned_loss=0.1023, over 5670009.65 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:42:00,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2508, 1.8507, 1.0422, 1.3476], device='cuda:1'), covar=tensor([0.1322, 0.0550, 0.1497, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0446, 0.0523, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 04:42:16,099 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 04:42:25,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3616, 1.8535, 1.0738, 1.2629], device='cuda:1'), covar=tensor([0.1275, 0.0513, 0.1369, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0447, 0.0524, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 04:42:28,295 INFO [train.py:968] (1/2) Epoch 27, batch 37800, giga_loss[loss=0.2513, simple_loss=0.3284, pruned_loss=0.08712, over 28944.00 frames. ], tot_loss[loss=0.2834, simple_loss=0.3554, pruned_loss=0.1057, over 5664185.65 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3555, pruned_loss=0.1114, over 5687116.36 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3534, pruned_loss=0.1042, over 5667570.68 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:42:34,577 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1222661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:42:39,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9826, 2.9136, 1.8466, 1.0869], device='cuda:1'), covar=tensor([0.8685, 0.3869, 0.5010, 0.7917], device='cuda:1'), in_proj_covar=tensor([0.1837, 0.1735, 0.1665, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 04:42:56,428 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-14 04:43:03,318 INFO [optim.py:369] (1/2) 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 04:43:06,647 INFO [train.py:968] (1/2) Epoch 27, batch 37850, libri_loss[loss=0.3013, simple_loss=0.3684, pruned_loss=0.1171, over 29514.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3513, pruned_loss=0.1025, over 5674576.59 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.356, pruned_loss=0.1117, over 5687876.24 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3492, pruned_loss=0.1006, over 5675655.20 frames. ], batch size: 81, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:43:31,875 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1222735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:34,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1222738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:47,374 INFO [train.py:968] (1/2) Epoch 27, batch 37900, giga_loss[loss=0.256, simple_loss=0.3369, pruned_loss=0.08752, over 28745.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3494, pruned_loss=0.1004, over 5672684.15 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3564, pruned_loss=0.1121, over 5679039.24 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3472, pruned_loss=0.09833, over 5682295.46 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:43:48,535 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:58,834 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1222767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:43:58,849 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1222767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:44:07,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6766, 1.7422, 1.8850, 1.4478], device='cuda:1'), covar=tensor([0.2041, 0.2748, 0.1655, 0.1899], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0713, 0.0979, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 04:44:26,702 INFO [optim.py:369] (1/2) 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 04:44:28,958 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 37950, giga_loss[loss=0.2426, simple_loss=0.321, pruned_loss=0.08206, over 28479.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3474, pruned_loss=0.09866, over 5674240.17 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3565, pruned_loss=0.1123, over 5682301.05 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3454, pruned_loss=0.09651, over 5678736.50 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:44:31,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1222807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:44:33,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4676, 1.9618, 1.4236, 0.7301], device='cuda:1'), covar=tensor([0.5944, 0.2830, 0.4165, 0.7525], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1727, 0.1659, 0.1500], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 04:44:53,050 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 38000, libri_loss[loss=0.3172, simple_loss=0.3706, pruned_loss=0.1319, over 29544.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3489, pruned_loss=0.09936, over 5687992.56 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3567, pruned_loss=0.1123, over 5690866.30 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3467, pruned_loss=0.09715, over 5683815.78 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:45:46,717 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 38050, giga_loss[loss=0.2514, simple_loss=0.3363, pruned_loss=0.0833, over 28607.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3512, pruned_loss=0.1009, over 5687862.50 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3569, pruned_loss=0.1124, over 5692756.67 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3493, pruned_loss=0.09894, over 5682913.77 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:46:27,118 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 38100, giga_loss[loss=0.2729, simple_loss=0.3488, pruned_loss=0.0985, over 28921.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3517, pruned_loss=0.1016, over 5695097.80 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3562, pruned_loss=0.112, over 5696674.91 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3508, pruned_loss=0.1002, over 5687465.25 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:46:49,419 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 04:47:11,753 INFO [optim.py:369] (1/2) 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 04:47:15,430 INFO [train.py:968] (1/2) Epoch 27, batch 38150, giga_loss[loss=0.3422, simple_loss=0.3884, pruned_loss=0.148, over 26763.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3534, pruned_loss=0.1031, over 5690506.41 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3567, pruned_loss=0.1123, over 5690485.42 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3521, pruned_loss=0.1016, over 5689227.13 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:47:50,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1347, 1.6206, 1.5807, 1.4110], device='cuda:1'), covar=tensor([0.2025, 0.1400, 0.2090, 0.1675], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0758, 0.0727, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 04:47:58,674 INFO [train.py:968] (1/2) Epoch 27, batch 38200, giga_loss[loss=0.2796, simple_loss=0.3515, pruned_loss=0.1038, over 28894.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3534, pruned_loss=0.1032, over 5693969.11 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3568, pruned_loss=0.1122, over 5692740.59 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3523, pruned_loss=0.102, over 5690810.68 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:48:22,881 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:48:29,749 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:48:31,800 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:48:38,556 INFO [optim.py:369] (1/2) 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 04:48:41,231 INFO [train.py:968] (1/2) Epoch 27, batch 38250, giga_loss[loss=0.268, simple_loss=0.3531, pruned_loss=0.09147, over 28987.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3535, pruned_loss=0.1025, over 5698383.21 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3571, pruned_loss=0.1124, over 5694525.39 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1013, over 5694166.93 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:48:55,007 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:01,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6955, 2.1306, 1.7064, 1.8952], device='cuda:1'), covar=tensor([0.2687, 0.2534, 0.2957, 0.2202], device='cuda:1'), in_proj_covar=tensor([0.1590, 0.1147, 0.1406, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 04:49:02,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:08,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:10,533 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:49:16,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5772, 4.4176, 4.1566, 2.0357], device='cuda:1'), covar=tensor([0.0585, 0.0746, 0.0738, 0.2138], device='cuda:1'), in_proj_covar=tensor([0.1282, 0.1187, 0.0995, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 04:49:19,188 INFO [train.py:968] (1/2) Epoch 27, batch 38300, libri_loss[loss=0.2915, simple_loss=0.3659, pruned_loss=0.1086, over 29668.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3524, pruned_loss=0.1009, over 5707542.54 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3564, pruned_loss=0.1118, over 5698457.22 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3519, pruned_loss=0.1001, over 5700765.90 frames. ], batch size: 88, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:49:24,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5420, 1.5732, 1.7747, 1.3394], device='cuda:1'), covar=tensor([0.1964, 0.2836, 0.1613, 0.1948], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0713, 0.0978, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 04:49:32,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5740, 1.7392, 1.7400, 1.3427], device='cuda:1'), covar=tensor([0.1710, 0.2905, 0.1545, 0.1874], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0713, 0.0978, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 04:49:49,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5536, 2.2811, 1.6492, 0.8729], device='cuda:1'), covar=tensor([0.4871, 0.3215, 0.3892, 0.5691], device='cuda:1'), in_proj_covar=tensor([0.1818, 0.1715, 0.1647, 0.1491], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 04:49:57,555 INFO [optim.py:369] (1/2) 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 04:50:00,273 INFO [train.py:968] (1/2) Epoch 27, batch 38350, giga_loss[loss=0.3257, simple_loss=0.3719, pruned_loss=0.1397, over 26703.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3537, pruned_loss=0.1011, over 5708522.97 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3569, pruned_loss=0.1121, over 5704171.34 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3527, pruned_loss=0.09982, over 5698099.43 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:50:27,874 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-14 04:50:40,498 INFO [train.py:968] (1/2) Epoch 27, batch 38400, giga_loss[loss=0.271, simple_loss=0.3449, pruned_loss=0.09849, over 27941.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3525, pruned_loss=0.1005, over 5716887.23 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3573, pruned_loss=0.1124, over 5708438.08 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3513, pruned_loss=0.09894, over 5704799.10 frames. ], batch size: 412, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:50:56,549 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:50:58,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:04,799 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:07,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:13,935 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:51:17,937 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 38450, giga_loss[loss=0.2733, simple_loss=0.3551, pruned_loss=0.09579, over 28516.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.351, pruned_loss=0.1001, over 5705265.14 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3574, pruned_loss=0.1126, over 5701207.39 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3497, pruned_loss=0.09825, over 5702761.09 frames. ], batch size: 336, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:51:22,512 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:52:01,091 INFO [train.py:968] (1/2) Epoch 27, batch 38500, giga_loss[loss=0.2529, simple_loss=0.3328, pruned_loss=0.08648, over 28839.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3486, pruned_loss=0.09886, over 5716072.10 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3573, pruned_loss=0.1125, over 5705315.22 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3475, pruned_loss=0.09723, over 5710359.09 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 04:52:17,283 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-14 04:52:35,819 INFO [optim.py:369] (1/2) 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 04:52:38,525 INFO [train.py:968] (1/2) Epoch 27, batch 38550, giga_loss[loss=0.3082, simple_loss=0.3536, pruned_loss=0.1314, over 23560.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3481, pruned_loss=0.09888, over 5696841.63 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.358, pruned_loss=0.1128, over 5688788.22 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3464, pruned_loss=0.09687, over 5707279.77 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:53:18,328 INFO [train.py:968] (1/2) Epoch 27, batch 38600, giga_loss[loss=0.2541, simple_loss=0.3363, pruned_loss=0.08591, over 29018.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3495, pruned_loss=0.1004, over 5704246.45 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3577, pruned_loss=0.1126, over 5690849.46 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3482, pruned_loss=0.09876, over 5710663.04 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:53:19,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:53:55,395 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 38650, giga_loss[loss=0.2648, simple_loss=0.3456, pruned_loss=0.09204, over 28858.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.35, pruned_loss=0.1001, over 5710577.57 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.358, pruned_loss=0.1128, over 5694874.04 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3485, pruned_loss=0.09828, over 5712401.92 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:53:57,691 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4469, 3.5205, 1.5106, 1.6441], device='cuda:1'), covar=tensor([0.1057, 0.0314, 0.0924, 0.1369], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0560, 0.0404, 0.0439], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0030], device='cuda:1')
+2023-03-14 04:54:03,548 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:54:35,483 INFO [train.py:968] (1/2) Epoch 27, batch 38700, giga_loss[loss=0.2443, simple_loss=0.3332, pruned_loss=0.07771, over 28779.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3489, pruned_loss=0.09879, over 5690849.76 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3582, pruned_loss=0.113, over 5679905.29 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3474, pruned_loss=0.09696, over 5706663.57 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:54:36,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6703, 1.9318, 2.0062, 1.6064], device='cuda:1'), covar=tensor([0.3892, 0.2960, 0.2753, 0.3410], device='cuda:1'), in_proj_covar=tensor([0.2047, 0.2001, 0.1915, 0.2059], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:55:09,254 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:55:12,125 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:55:13,636 INFO [train.py:968] (1/2) Epoch 27, batch 38750, giga_loss[loss=0.2716, simple_loss=0.347, pruned_loss=0.09814, over 28992.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3479, pruned_loss=0.09837, over 5700997.38 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3578, pruned_loss=0.1128, over 5681897.89 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3468, pruned_loss=0.0967, over 5712525.59 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:55:34,120 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:55:52,298 INFO [train.py:968] (1/2) Epoch 27, batch 38800, giga_loss[loss=0.2542, simple_loss=0.331, pruned_loss=0.08868, over 28733.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3468, pruned_loss=0.09819, over 5698785.33 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3581, pruned_loss=0.1131, over 5680439.50 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3456, pruned_loss=0.0964, over 5709520.13 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:55:56,795 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:55:59,758 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:56:06,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1223670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:56:07,007 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-14 04:56:20,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 04:56:21,108 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:56:34,128 INFO [train.py:968] (1/2) Epoch 27, batch 38850, giga_loss[loss=0.2373, simple_loss=0.3199, pruned_loss=0.07733, over 28897.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3437, pruned_loss=0.09635, over 5699218.34 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3579, pruned_loss=0.1129, over 5682966.74 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3426, pruned_loss=0.09486, over 5705442.00 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 04:57:11,840 INFO [train.py:968] (1/2) Epoch 27, batch 38900, giga_loss[loss=0.2406, simple_loss=0.3176, pruned_loss=0.08181, over 28231.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3405, pruned_loss=0.09476, over 5695905.39 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3586, pruned_loss=0.1135, over 5676313.46 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09275, over 5706589.58 frames. ], batch size: 77, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:57:38,645 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-14 04:57:49,853 INFO [optim.py:369] (1/2) 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 04:57:50,830 INFO [train.py:968] (1/2) Epoch 27, batch 38950, giga_loss[loss=0.2475, simple_loss=0.3265, pruned_loss=0.08422, over 29104.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3395, pruned_loss=0.09435, over 5694137.58 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.358, pruned_loss=0.1132, over 5672679.32 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3381, pruned_loss=0.0925, over 5705645.54 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:57:59,787 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1223813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:02,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1223816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:26,955 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1223845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:26,969 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1223845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 04:58:33,322 INFO [train.py:968] (1/2) Epoch 27, batch 39000, giga_loss[loss=0.3371, simple_loss=0.3861, pruned_loss=0.144, over 27665.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3394, pruned_loss=0.0948, over 5690124.61 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3581, pruned_loss=0.113, over 5673434.85 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3377, pruned_loss=0.09292, over 5698963.50 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:58:33,322 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 04:58:41,968 INFO [train.py:1012] (1/2) Epoch 27, validation: loss=0.2028, simple_loss=0.3112, pruned_loss=0.04724, over 944034.00 frames. 
+2023-03-14 04:58:41,968 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 04:58:58,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8427, 2.1104, 1.9644, 1.7326], device='cuda:1'), covar=tensor([0.2358, 0.1980, 0.1820, 0.2049], device='cuda:1'), in_proj_covar=tensor([0.2039, 0.1991, 0.1904, 0.2051], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 04:59:18,641 INFO [optim.py:369] (1/2) 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 04:59:19,300 INFO [train.py:968] (1/2) Epoch 27, batch 39050, giga_loss[loss=0.2468, simple_loss=0.3229, pruned_loss=0.08535, over 29059.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3374, pruned_loss=0.09393, over 5698555.47 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3582, pruned_loss=0.1131, over 5677298.69 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3354, pruned_loss=0.09174, over 5702870.45 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 04:59:57,716 INFO [train.py:968] (1/2) Epoch 27, batch 39100, giga_loss[loss=0.2185, simple_loss=0.2995, pruned_loss=0.06877, over 28642.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3345, pruned_loss=0.09248, over 5707400.76 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.358, pruned_loss=0.1128, over 5682070.38 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3326, pruned_loss=0.09063, over 5707141.44 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:00:35,951 INFO [optim.py:369] (1/2) 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 05:00:36,527 INFO [train.py:968] (1/2) Epoch 27, batch 39150, giga_loss[loss=0.2551, simple_loss=0.3215, pruned_loss=0.09432, over 28874.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3331, pruned_loss=0.09202, over 5705847.74 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3581, pruned_loss=0.1127, over 5684552.41 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3308, pruned_loss=0.09008, over 5704199.86 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:00:58,573 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 39200, giga_loss[loss=0.2735, simple_loss=0.3449, pruned_loss=0.1011, over 28951.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3316, pruned_loss=0.09138, over 5708329.79 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3575, pruned_loss=0.1124, over 5688322.20 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3298, pruned_loss=0.08966, over 5704059.04 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:02:00,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4594, 1.7024, 1.2919, 1.2677], device='cuda:1'), covar=tensor([0.1070, 0.0625, 0.1067, 0.1202], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0446, 0.0523, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 05:02:00,410 INFO [optim.py:369] (1/2) 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 05:02:01,035 INFO [train.py:968] (1/2) Epoch 27, batch 39250, libri_loss[loss=0.3248, simple_loss=0.3667, pruned_loss=0.1414, over 29372.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3335, pruned_loss=0.09185, over 5706800.86 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3574, pruned_loss=0.1125, over 5692257.14 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3316, pruned_loss=0.08997, over 5699851.35 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:02:29,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3830, 1.5298, 1.2458, 1.1744], device='cuda:1'), covar=tensor([0.1087, 0.0635, 0.1124, 0.1194], device='cuda:1'), in_proj_covar=tensor([0.0408, 0.0445, 0.0522, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 05:02:42,896 INFO [train.py:968] (1/2) Epoch 27, batch 39300, giga_loss[loss=0.2516, simple_loss=0.3347, pruned_loss=0.08428, over 28867.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3391, pruned_loss=0.09512, over 5701085.58 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3577, pruned_loss=0.1127, over 5696549.68 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3368, pruned_loss=0.09301, over 5692085.95 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:02:49,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5415, 1.8609, 1.6870, 1.6065], device='cuda:1'), covar=tensor([0.2217, 0.2496, 0.2345, 0.2523], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0752, 0.0723, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 05:03:24,538 INFO [optim.py:369] (1/2) 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 05:03:25,107 INFO [train.py:968] (1/2) Epoch 27, batch 39350, giga_loss[loss=0.2639, simple_loss=0.3466, pruned_loss=0.0906, over 28670.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3412, pruned_loss=0.09524, over 5693666.46 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3581, pruned_loss=0.113, over 5692678.02 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.09281, over 5690661.54 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:03:36,451 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:03:38,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3766, 3.0011, 1.5180, 1.4441], device='cuda:1'), covar=tensor([0.0899, 0.0309, 0.0923, 0.1282], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0563, 0.0405, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 05:03:54,250 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:04:05,705 INFO [train.py:968] (1/2) Epoch 27, batch 39400, giga_loss[loss=0.2621, simple_loss=0.348, pruned_loss=0.08814, over 28923.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.343, pruned_loss=0.09601, over 5673036.97 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3585, pruned_loss=0.1134, over 5670145.77 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3401, pruned_loss=0.09315, over 5690684.01 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:04:48,866 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 27, batch 39450, giga_loss[loss=0.2263, simple_loss=0.3138, pruned_loss=0.06938, over 28929.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3418, pruned_loss=0.09503, over 5686056.67 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3585, pruned_loss=0.1134, over 5675018.00 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3391, pruned_loss=0.09235, over 5695848.54 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:05:05,412 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2202, 1.7614, 1.3650, 0.4747], device='cuda:1'), covar=tensor([0.5341, 0.2972, 0.4557, 0.7113], device='cuda:1'), in_proj_covar=tensor([0.1823, 0.1712, 0.1647, 0.1493], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 05:05:07,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4336, 1.7509, 1.4473, 1.4542], device='cuda:1'), covar=tensor([0.0754, 0.0313, 0.0341, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0120, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:05:28,402 INFO [train.py:968] (1/2) Epoch 27, batch 39500, giga_loss[loss=0.2591, simple_loss=0.3422, pruned_loss=0.08798, over 28922.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3402, pruned_loss=0.0943, over 5692341.09 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3581, pruned_loss=0.1132, over 5680517.11 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.338, pruned_loss=0.09192, over 5695415.76 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:05:36,411 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1224363.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:05:38,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1224366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:06:03,573 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1224395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:06:10,186 INFO [optim.py:369] (1/2) 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 05:06:10,201 INFO [train.py:968] (1/2) Epoch 27, batch 39550, giga_loss[loss=0.3241, simple_loss=0.3884, pruned_loss=0.1299, over 28657.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3419, pruned_loss=0.09536, over 5706156.63 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3585, pruned_loss=0.1134, over 5683749.29 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3392, pruned_loss=0.09273, over 5706329.45 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:06:11,268 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:06:53,876 INFO [train.py:968] (1/2) Epoch 27, batch 39600, giga_loss[loss=0.2819, simple_loss=0.3671, pruned_loss=0.0984, over 28931.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3426, pruned_loss=0.09565, over 5711022.52 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3588, pruned_loss=0.1135, over 5685535.34 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3402, pruned_loss=0.09326, over 5709859.28 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:07:36,409 INFO [train.py:968] (1/2) Epoch 27, batch 39650, libri_loss[loss=0.3157, simple_loss=0.3832, pruned_loss=0.1241, over 29481.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3472, pruned_loss=0.09823, over 5709124.68 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3591, pruned_loss=0.1137, over 5689784.88 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3446, pruned_loss=0.09575, over 5704984.17 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:07:37,139 INFO [optim.py:369] (1/2) 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 05:07:57,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4850, 1.7008, 1.1970, 1.3312], device='cuda:1'), covar=tensor([0.1043, 0.0695, 0.1092, 0.1271], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0448, 0.0524, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 05:08:13,193 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1224549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:08:15,787 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1224552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:08:17,588 INFO [train.py:968] (1/2) Epoch 27, batch 39700, giga_loss[loss=0.2975, simple_loss=0.3596, pruned_loss=0.1177, over 28705.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3495, pruned_loss=0.09917, over 5712492.35 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3592, pruned_loss=0.1137, over 5692093.52 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3471, pruned_loss=0.09706, over 5707263.35 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:08:38,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1224581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:08:43,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2525, 0.7644, 0.8355, 1.3722], device='cuda:1'), covar=tensor([0.0763, 0.0387, 0.0384, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:08:56,308 INFO [train.py:968] (1/2) Epoch 27, batch 39750, libri_loss[loss=0.3308, simple_loss=0.3899, pruned_loss=0.1359, over 29530.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3507, pruned_loss=0.09974, over 5716764.13 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3597, pruned_loss=0.1137, over 5696551.12 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3482, pruned_loss=0.09767, over 5709036.56 frames. ], batch size: 84, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:08:56,886 INFO [optim.py:369] (1/2) 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 05:09:06,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:09:07,633 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:09:38,387 INFO [train.py:968] (1/2) Epoch 27, batch 39800, giga_loss[loss=0.2602, simple_loss=0.3454, pruned_loss=0.08751, over 28849.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3528, pruned_loss=0.1008, over 5714934.02 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3599, pruned_loss=0.1137, over 5697410.10 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3505, pruned_loss=0.09889, over 5708382.34 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:10:17,956 INFO [train.py:968] (1/2) Epoch 27, batch 39850, giga_loss[loss=0.2893, simple_loss=0.3609, pruned_loss=0.1089, over 28769.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3522, pruned_loss=0.1007, over 5712611.51 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5699613.12 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3502, pruned_loss=0.09912, over 5705617.03 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:10:18,650 INFO [optim.py:369] (1/2) 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 05:10:56,376 INFO [train.py:968] (1/2) Epoch 27, batch 39900, giga_loss[loss=0.2482, simple_loss=0.3233, pruned_loss=0.08652, over 28909.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3508, pruned_loss=0.1004, over 5705997.78 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3603, pruned_loss=0.1138, over 5689517.30 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3489, pruned_loss=0.09881, over 5709301.41 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:10:59,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1224758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:01,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1224761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:25,599 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1224790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:37,088 INFO [train.py:968] (1/2) Epoch 27, batch 39950, giga_loss[loss=0.2794, simple_loss=0.352, pruned_loss=0.1035, over 27631.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3482, pruned_loss=0.09954, over 5704271.14 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3602, pruned_loss=0.1136, over 5688449.20 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3464, pruned_loss=0.09803, over 5708759.42 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:11:37,766 INFO [optim.py:369] (1/2) 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 05:11:46,507 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1224817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:11:57,161 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 40000, giga_loss[loss=0.2987, simple_loss=0.3722, pruned_loss=0.1126, over 28858.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3448, pruned_loss=0.09761, over 5710238.04 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3603, pruned_loss=0.1137, over 5691838.13 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3431, pruned_loss=0.09616, over 5711074.26 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:13:00,990 INFO [train.py:968] (1/2) Epoch 27, batch 40050, giga_loss[loss=0.2413, simple_loss=0.3334, pruned_loss=0.07455, over 28273.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3462, pruned_loss=0.09718, over 5715297.41 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3605, pruned_loss=0.1139, over 5693212.84 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3445, pruned_loss=0.09578, over 5714999.31 frames. ], batch size: 368, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:13:01,496 INFO [optim.py:369] (1/2) 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 05:13:42,656 INFO [train.py:968] (1/2) Epoch 27, batch 40100, giga_loss[loss=0.2651, simple_loss=0.3446, pruned_loss=0.09284, over 28883.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3459, pruned_loss=0.09569, over 5708250.91 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3601, pruned_loss=0.1137, over 5695281.24 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3448, pruned_loss=0.09461, over 5706530.72 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:14:09,580 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:14:09,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8639, 2.1984, 1.9713, 1.8679], device='cuda:1'), covar=tensor([0.2410, 0.2669, 0.2459, 0.2729], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0760, 0.0729, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 05:14:19,382 INFO [train.py:968] (1/2) Epoch 27, batch 40150, giga_loss[loss=0.2628, simple_loss=0.328, pruned_loss=0.09877, over 28851.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3451, pruned_loss=0.09622, over 5688323.75 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3599, pruned_loss=0.1138, over 5671230.87 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3437, pruned_loss=0.09442, over 5708899.07 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:14:20,762 INFO [optim.py:369] (1/2) 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 05:14:26,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2925, 1.2400, 1.1418, 1.4703], device='cuda:1'), covar=tensor([0.0748, 0.0388, 0.0368, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:14:39,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5261, 1.4407, 3.7918, 3.3330], device='cuda:1'), covar=tensor([0.1458, 0.2408, 0.0861, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0667, 0.0994, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 05:14:57,507 INFO [train.py:968] (1/2) Epoch 27, batch 40200, giga_loss[loss=0.3639, simple_loss=0.4032, pruned_loss=0.1623, over 26722.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3459, pruned_loss=0.09815, over 5689198.81 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.114, over 5674583.96 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3441, pruned_loss=0.09601, over 5703465.73 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:15:36,432 INFO [train.py:968] (1/2) Epoch 27, batch 40250, giga_loss[loss=0.2969, simple_loss=0.3642, pruned_loss=0.1148, over 27691.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3432, pruned_loss=0.09753, over 5694296.71 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3597, pruned_loss=0.1135, over 5680531.93 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3419, pruned_loss=0.09578, over 5700873.13 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:15:37,697 INFO [optim.py:369] (1/2) 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 05:16:02,047 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:04,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:18,286 INFO [train.py:968] (1/2) Epoch 27, batch 40300, giga_loss[loss=0.2598, simple_loss=0.3291, pruned_loss=0.09523, over 28818.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3413, pruned_loss=0.09706, over 5697610.04 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3599, pruned_loss=0.1137, over 5674461.09 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3397, pruned_loss=0.09524, over 5707645.77 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:16:27,737 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:36,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2131, 1.0860, 3.8263, 3.1741], device='cuda:1'), covar=tensor([0.1806, 0.3079, 0.0464, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0796, 0.0667, 0.0993, 0.0968], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 05:16:48,133 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1225192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:55,983 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1225204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:16:56,400 INFO [train.py:968] (1/2) Epoch 27, batch 40350, giga_loss[loss=0.2217, simple_loss=0.3048, pruned_loss=0.06936, over 28922.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3402, pruned_loss=0.09682, over 5708281.05 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5677873.60 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3385, pruned_loss=0.09504, over 5713766.73 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:16:57,719 INFO [optim.py:369] (1/2) 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 05:17:30,997 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 40400, giga_loss[loss=0.2499, simple_loss=0.3133, pruned_loss=0.09326, over 28673.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3367, pruned_loss=0.09467, over 5714940.42 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3592, pruned_loss=0.1132, over 5681271.32 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3354, pruned_loss=0.09313, over 5717284.58 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:18:15,451 INFO [train.py:968] (1/2) Epoch 27, batch 40450, giga_loss[loss=0.2855, simple_loss=0.3405, pruned_loss=0.1153, over 23849.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3337, pruned_loss=0.09328, over 5712623.56 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3592, pruned_loss=0.1131, over 5682213.02 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3322, pruned_loss=0.09177, over 5714327.11 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:18:17,104 INFO [optim.py:369] (1/2) 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 05:18:31,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1075, 4.9503, 4.7015, 2.4889], device='cuda:1'), covar=tensor([0.0441, 0.0550, 0.0616, 0.1682], device='cuda:1'), in_proj_covar=tensor([0.1290, 0.1194, 0.1002, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 05:18:34,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9548, 1.1561, 1.0898, 0.9152], device='cuda:1'), covar=tensor([0.2823, 0.3036, 0.1863, 0.2546], device='cuda:1'), in_proj_covar=tensor([0.2052, 0.2003, 0.1919, 0.2054], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 05:18:38,557 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:40,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:48,204 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:50,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:18:55,522 INFO [train.py:968] (1/2) Epoch 27, batch 40500, giga_loss[loss=0.2119, simple_loss=0.2955, pruned_loss=0.06413, over 28735.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3301, pruned_loss=0.09151, over 5701192.06 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3592, pruned_loss=0.1132, over 5669167.05 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.328, pruned_loss=0.08965, over 5714949.91 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:19:04,401 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:19:13,577 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:19:26,763 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9354, 2.9307, 1.8104, 1.0247], device='cuda:1'), covar=tensor([1.0090, 0.3676, 0.4972, 0.8620], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1716, 0.1650, 0.1495], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 05:19:32,928 INFO [train.py:968] (1/2) Epoch 27, batch 40550, giga_loss[loss=0.2558, simple_loss=0.3343, pruned_loss=0.08867, over 28740.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3292, pruned_loss=0.09074, over 5706776.66 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3591, pruned_loss=0.1131, over 5673512.40 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3271, pruned_loss=0.08897, over 5714568.11 frames. ], batch size: 262, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:19:35,123 INFO [optim.py:369] (1/2) 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 05:19:57,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2109, 1.7884, 1.3857, 0.4239], device='cuda:1'), covar=tensor([0.5432, 0.3301, 0.5127, 0.7327], device='cuda:1'), in_proj_covar=tensor([0.1826, 0.1716, 0.1650, 0.1494], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 05:20:04,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0398, 1.5161, 1.4674, 1.2600], device='cuda:1'), covar=tensor([0.2332, 0.1627, 0.2356, 0.2007], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0760, 0.0727, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 05:20:15,924 INFO [train.py:968] (1/2) Epoch 27, batch 40600, libri_loss[loss=0.2608, simple_loss=0.3329, pruned_loss=0.09436, over 29574.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3334, pruned_loss=0.09293, over 5706739.62 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.359, pruned_loss=0.1129, over 5675741.17 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3312, pruned_loss=0.09115, over 5711514.75 frames. ], batch size: 75, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:20:58,492 INFO [train.py:968] (1/2) Epoch 27, batch 40650, giga_loss[loss=0.2594, simple_loss=0.3427, pruned_loss=0.08805, over 28996.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3357, pruned_loss=0.09336, over 5709311.99 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3582, pruned_loss=0.1123, over 5681011.92 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.334, pruned_loss=0.09195, over 5709430.67 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:21:00,755 INFO [optim.py:369] (1/2) 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 05:21:29,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2871, 1.1703, 1.1239, 1.4879], device='cuda:1'), covar=tensor([0.0780, 0.0376, 0.0380, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:21:35,842 INFO [train.py:968] (1/2) Epoch 27, batch 40700, giga_loss[loss=0.2808, simple_loss=0.3593, pruned_loss=0.1011, over 28890.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3397, pruned_loss=0.09506, over 5707590.80 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3581, pruned_loss=0.1123, over 5674849.43 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3378, pruned_loss=0.09338, over 5714110.56 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:21:42,751 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9753, 2.3180, 2.2158, 1.9209], device='cuda:1'), covar=tensor([0.2211, 0.1536, 0.1631, 0.1829], device='cuda:1'), in_proj_covar=tensor([0.2051, 0.2008, 0.1921, 0.2052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 05:22:14,701 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 05:22:16,638 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2874, 3.1468, 1.4953, 1.3986], device='cuda:1'), covar=tensor([0.1043, 0.0353, 0.0981, 0.1398], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0565, 0.0406, 0.0442], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 05:22:17,010 INFO [train.py:968] (1/2) Epoch 27, batch 40750, giga_loss[loss=0.2834, simple_loss=0.3667, pruned_loss=0.1001, over 28779.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3412, pruned_loss=0.09539, over 5707957.73 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3581, pruned_loss=0.1124, over 5672626.86 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3391, pruned_loss=0.09342, over 5715494.56 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:22:18,258 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1225623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:22:52,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9384, 1.9646, 2.1369, 1.6592], device='cuda:1'), covar=tensor([0.1859, 0.2449, 0.1527, 0.1740], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0711, 0.0974, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 05:22:56,806 INFO [train.py:968] (1/2) Epoch 27, batch 40800, giga_loss[loss=0.2684, simple_loss=0.3423, pruned_loss=0.09728, over 29063.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3433, pruned_loss=0.09676, over 5702316.44 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3582, pruned_loss=0.1123, over 5677051.06 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3411, pruned_loss=0.09472, over 5705545.47 frames. ], batch size: 155, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:22:59,690 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 05:23:30,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4095, 1.5905, 1.6417, 1.2688], device='cuda:1'), covar=tensor([0.1337, 0.2068, 0.1129, 0.1390], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0711, 0.0975, 0.0875], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 05:23:44,487 INFO [train.py:968] (1/2) Epoch 27, batch 40850, giga_loss[loss=0.2933, simple_loss=0.3661, pruned_loss=0.1102, over 28952.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3484, pruned_loss=0.1012, over 5697436.49 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3579, pruned_loss=0.1121, over 5679840.64 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3467, pruned_loss=0.09963, over 5697837.88 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:23:49,049 INFO [optim.py:369] (1/2) 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 05:24:09,990 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-14 05:24:26,677 INFO [train.py:968] (1/2) Epoch 27, batch 40900, giga_loss[loss=0.3352, simple_loss=0.4035, pruned_loss=0.1335, over 28852.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3538, pruned_loss=0.1061, over 5688992.50 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3567, pruned_loss=0.1113, over 5689636.06 frames. ], giga_tot_loss[loss=0.2818, simple_loss=0.3533, pruned_loss=0.1052, over 5680577.90 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:24:27,767 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 05:24:37,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1225766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:24:39,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1225769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:25:06,358 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1225798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:25:11,526 INFO [train.py:968] (1/2) Epoch 27, batch 40950, giga_loss[loss=0.3225, simple_loss=0.3936, pruned_loss=0.1257, over 29013.00 frames. ], tot_loss[loss=0.2904, simple_loss=0.36, pruned_loss=0.1103, over 5694398.49 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3567, pruned_loss=0.1113, over 5694767.44 frames. ], giga_tot_loss[loss=0.2894, simple_loss=0.3597, pruned_loss=0.1095, over 5683043.58 frames. ], batch size: 145, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:25:14,251 INFO [optim.py:369] (1/2) 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 05:25:56,684 INFO [train.py:968] (1/2) Epoch 27, batch 41000, giga_loss[loss=0.3452, simple_loss=0.4041, pruned_loss=0.1431, over 28862.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3672, pruned_loss=0.1163, over 5672680.06 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3565, pruned_loss=0.1112, over 5686919.29 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3671, pruned_loss=0.1158, over 5670854.78 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:26:42,334 INFO [train.py:968] (1/2) Epoch 27, batch 41050, giga_loss[loss=0.4382, simple_loss=0.4477, pruned_loss=0.2144, over 23548.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3735, pruned_loss=0.1219, over 5673704.87 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3563, pruned_loss=0.111, over 5691142.19 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3739, pruned_loss=0.1218, over 5668298.85 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:26:44,308 INFO [optim.py:369] (1/2) 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 05:26:52,151 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 41100, giga_loss[loss=0.2913, simple_loss=0.354, pruned_loss=0.1143, over 28411.00 frames. ], tot_loss[loss=0.3145, simple_loss=0.3775, pruned_loss=0.1257, over 5660873.14 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.356, pruned_loss=0.111, over 5692787.65 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3785, pruned_loss=0.1259, over 5654539.85 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:27:56,996 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 05:28:19,119 INFO [train.py:968] (1/2) Epoch 27, batch 41150, giga_loss[loss=0.3178, simple_loss=0.3832, pruned_loss=0.1262, over 28891.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3769, pruned_loss=0.1253, over 5677523.36 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3558, pruned_loss=0.1108, over 5700527.30 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3786, pruned_loss=0.1262, over 5664562.15 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:28:20,950 INFO [optim.py:369] (1/2) 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 05:28:59,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5644, 1.7206, 1.6483, 1.5158], device='cuda:1'), covar=tensor([0.1833, 0.1926, 0.2207, 0.1981], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0760, 0.0730, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 05:29:13,930 INFO [train.py:968] (1/2) Epoch 27, batch 41200, giga_loss[loss=0.4341, simple_loss=0.4443, pruned_loss=0.2119, over 23519.00 frames. ], tot_loss[loss=0.3209, simple_loss=0.3812, pruned_loss=0.1303, over 5641854.85 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3558, pruned_loss=0.1109, over 5703274.58 frames. ], giga_tot_loss[loss=0.3226, simple_loss=0.3829, pruned_loss=0.1312, over 5628402.13 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:29:31,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0134, 3.2742, 2.1064, 1.1023], device='cuda:1'), covar=tensor([0.9226, 0.3238, 0.4376, 0.7762], device='cuda:1'), in_proj_covar=tensor([0.1827, 0.1728, 0.1651, 0.1493], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 05:29:42,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3774, 1.9929, 1.5500, 0.6106], device='cuda:1'), covar=tensor([0.5101, 0.2945, 0.3689, 0.6478], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1728, 0.1652, 0.1494], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 05:29:54,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:30:03,605 INFO [train.py:968] (1/2) Epoch 27, batch 41250, giga_loss[loss=0.2915, simple_loss=0.3569, pruned_loss=0.1131, over 28901.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3844, pruned_loss=0.1342, over 5625998.09 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3562, pruned_loss=0.1112, over 5688813.84 frames. ], giga_tot_loss[loss=0.3284, simple_loss=0.3864, pruned_loss=0.1353, over 5625660.58 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:30:07,601 INFO [optim.py:369] (1/2) 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 05:30:53,356 INFO [train.py:968] (1/2) Epoch 27, batch 41300, giga_loss[loss=0.2871, simple_loss=0.3606, pruned_loss=0.1068, over 28929.00 frames. ], tot_loss[loss=0.3323, simple_loss=0.3891, pruned_loss=0.1377, over 5625312.03 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3566, pruned_loss=0.1115, over 5685034.30 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.3912, pruned_loss=0.139, over 5626510.88 frames. ], batch size: 106, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:31:44,224 INFO [train.py:968] (1/2) Epoch 27, batch 41350, libri_loss[loss=0.2574, simple_loss=0.3333, pruned_loss=0.09074, over 29546.00 frames. ], tot_loss[loss=0.3329, simple_loss=0.3892, pruned_loss=0.1383, over 5636223.22 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3562, pruned_loss=0.1111, over 5691419.87 frames. ], giga_tot_loss[loss=0.3367, simple_loss=0.3922, pruned_loss=0.1406, over 5629602.40 frames. ], batch size: 78, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:31:47,634 INFO [optim.py:369] (1/2) 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 05:32:26,024 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1723, 1.3054, 3.3074, 3.0012], device='cuda:1'), covar=tensor([0.1550, 0.2566, 0.0535, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0670, 0.1001, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 05:32:30,354 INFO [train.py:968] (1/2) Epoch 27, batch 41400, giga_loss[loss=0.3173, simple_loss=0.3846, pruned_loss=0.125, over 29013.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3871, pruned_loss=0.1372, over 5632935.27 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3561, pruned_loss=0.111, over 5691908.01 frames. ], giga_tot_loss[loss=0.3342, simple_loss=0.3898, pruned_loss=0.1393, over 5626839.24 frames. ], batch size: 213, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:33:10,270 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 41450, libri_loss[loss=0.2952, simple_loss=0.3633, pruned_loss=0.1136, over 29565.00 frames. ], tot_loss[loss=0.3316, simple_loss=0.388, pruned_loss=0.1376, over 5627921.24 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3565, pruned_loss=0.1112, over 5689434.75 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3905, pruned_loss=0.1397, over 5623924.06 frames. ], batch size: 75, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:33:26,077 INFO [optim.py:369] (1/2) 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 05:34:15,734 INFO [train.py:968] (1/2) Epoch 27, batch 41500, giga_loss[loss=0.3814, simple_loss=0.4234, pruned_loss=0.1697, over 27532.00 frames. ], tot_loss[loss=0.3319, simple_loss=0.3885, pruned_loss=0.1376, over 5609731.58 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3565, pruned_loss=0.1112, over 5691656.93 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3907, pruned_loss=0.1395, over 5604019.02 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:34:19,182 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4015, 3.5136, 1.5428, 1.5386], device='cuda:1'), covar=tensor([0.1028, 0.0331, 0.0906, 0.1418], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0570, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 05:34:24,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 05:34:36,751 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 05:34:38,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4122, 1.7151, 1.3956, 1.3317], device='cuda:1'), covar=tensor([0.2753, 0.2765, 0.3125, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1148, 0.1408, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 05:35:05,835 INFO [train.py:968] (1/2) Epoch 27, batch 41550, giga_loss[loss=0.4297, simple_loss=0.4557, pruned_loss=0.2019, over 26231.00 frames. ], tot_loss[loss=0.3336, simple_loss=0.3898, pruned_loss=0.1387, over 5599802.19 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3564, pruned_loss=0.1113, over 5694746.79 frames. ], giga_tot_loss[loss=0.3373, simple_loss=0.3927, pruned_loss=0.1409, over 5590407.16 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:35:10,303 INFO [optim.py:369] (1/2) 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 05:35:39,139 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1226434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:35:41,283 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1226437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:35:49,033 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3245, 1.5171, 1.5216, 1.2452], device='cuda:1'), covar=tensor([0.3626, 0.3209, 0.2498, 0.3078], device='cuda:1'), in_proj_covar=tensor([0.2072, 0.2026, 0.1935, 0.2064], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 05:35:56,299 INFO [train.py:968] (1/2) Epoch 27, batch 41600, giga_loss[loss=0.2961, simple_loss=0.3676, pruned_loss=0.1123, over 28743.00 frames. ], tot_loss[loss=0.3299, simple_loss=0.3874, pruned_loss=0.1362, over 5585643.22 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3569, pruned_loss=0.1116, over 5671339.17 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3898, pruned_loss=0.1382, over 5598129.86 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:36:08,962 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1226466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:36:11,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1226469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:36:24,076 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 05:36:45,872 INFO [train.py:968] (1/2) Epoch 27, batch 41650, giga_loss[loss=0.279, simple_loss=0.3552, pruned_loss=0.1014, over 28685.00 frames. ], tot_loss[loss=0.325, simple_loss=0.3849, pruned_loss=0.1325, over 5609876.17 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3568, pruned_loss=0.1116, over 5674611.85 frames. ], giga_tot_loss[loss=0.3281, simple_loss=0.3873, pruned_loss=0.1344, over 5615836.84 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:36:51,625 INFO [optim.py:369] (1/2) 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 05:37:38,700 INFO [train.py:968] (1/2) Epoch 27, batch 41700, giga_loss[loss=0.2729, simple_loss=0.3544, pruned_loss=0.09573, over 28961.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3807, pruned_loss=0.1287, over 5614695.78 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3568, pruned_loss=0.1116, over 5675010.79 frames. ], giga_tot_loss[loss=0.3219, simple_loss=0.3829, pruned_loss=0.1304, over 5618458.08 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:38:01,173 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 05:38:21,763 INFO [train.py:968] (1/2) Epoch 27, batch 41750, libri_loss[loss=0.2875, simple_loss=0.3668, pruned_loss=0.1041, over 29206.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3781, pruned_loss=0.127, over 5622658.58 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3567, pruned_loss=0.1117, over 5688115.49 frames. ], giga_tot_loss[loss=0.3201, simple_loss=0.3815, pruned_loss=0.1294, over 5610402.98 frames. ], batch size: 97, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:38:23,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3812, 1.8403, 1.5859, 1.5720], device='cuda:1'), covar=tensor([0.0671, 0.0278, 0.0286, 0.0713], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:38:27,474 INFO [optim.py:369] (1/2) 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 05:38:28,955 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4375, 3.1235, 1.5236, 1.5338], device='cuda:1'), covar=tensor([0.0956, 0.0359, 0.0915, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0570, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 05:38:28,987 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1226612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:30,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1226615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:39,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:41,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.90 vs. limit=5.0
+2023-03-14 05:38:43,326 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:38:55,902 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1226644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:39:05,831 INFO [train.py:968] (1/2) Epoch 27, batch 41800, libri_loss[loss=0.256, simple_loss=0.3213, pruned_loss=0.09534, over 29576.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.125, over 5634405.26 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3568, pruned_loss=0.1119, over 5685222.12 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3792, pruned_loss=0.1272, over 5624914.36 frames. ], batch size: 74, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:39:57,050 INFO [train.py:968] (1/2) Epoch 27, batch 41850, giga_loss[loss=0.3611, simple_loss=0.3991, pruned_loss=0.1615, over 26692.00 frames. ], tot_loss[loss=0.3129, simple_loss=0.3758, pruned_loss=0.1251, over 5638833.66 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3568, pruned_loss=0.1119, over 5686059.01 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3785, pruned_loss=0.1269, over 5630544.04 frames. ], batch size: 555, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:40:01,300 INFO [optim.py:369] (1/2) 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 05:40:01,816 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-14 05:40:28,170 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-14 05:40:45,043 INFO [train.py:968] (1/2) Epoch 27, batch 41900, giga_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 28682.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3742, pruned_loss=0.1237, over 5626578.84 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3568, pruned_loss=0.1119, over 5673316.08 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3769, pruned_loss=0.1254, over 5629968.65 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:41:17,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:41:21,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2486, 1.5028, 1.5086, 1.1020], device='cuda:1'), covar=tensor([0.1909, 0.2747, 0.1626, 0.1867], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0713, 0.0972, 0.0873], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 05:41:29,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.77 vs. limit=5.0
+2023-03-14 05:41:33,782 INFO [train.py:968] (1/2) Epoch 27, batch 41950, giga_loss[loss=0.2473, simple_loss=0.3351, pruned_loss=0.07979, over 28396.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3714, pruned_loss=0.1212, over 5633969.75 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3567, pruned_loss=0.112, over 5678763.33 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3743, pruned_loss=0.123, over 5629941.42 frames. ], batch size: 65, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:41:38,996 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:41:39,377 INFO [optim.py:369] (1/2) 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 05:42:24,664 INFO [train.py:968] (1/2) Epoch 27, batch 42000, giga_loss[loss=0.2917, simple_loss=0.3661, pruned_loss=0.1087, over 28879.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.372, pruned_loss=0.1188, over 5647950.57 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3567, pruned_loss=0.112, over 5684431.24 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3746, pruned_loss=0.1203, over 5638672.44 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:42:24,665 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 05:42:33,252 INFO [train.py:1012] (1/2) Epoch 27, validation: loss=0.1998, simple_loss=0.3058, pruned_loss=0.04693, over 944034.00 frames. 
+2023-03-14 05:42:33,252 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 05:43:17,778 INFO [train.py:968] (1/2) Epoch 27, batch 42050, giga_loss[loss=0.337, simple_loss=0.407, pruned_loss=0.1335, over 29060.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3723, pruned_loss=0.1178, over 5660031.88 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3567, pruned_loss=0.112, over 5687435.50 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3753, pruned_loss=0.1194, over 5647632.75 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:43:20,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-14 05:43:21,719 INFO [optim.py:369] (1/2) 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 05:43:35,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5417, 1.8361, 1.4390, 1.4145], device='cuda:1'), covar=tensor([0.2639, 0.2749, 0.3169, 0.2488], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1146, 0.1404, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 05:43:45,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1226934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:44:05,843 INFO [train.py:968] (1/2) Epoch 27, batch 42100, libri_loss[loss=0.2729, simple_loss=0.3517, pruned_loss=0.09708, over 29504.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3741, pruned_loss=0.1192, over 5669406.54 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3568, pruned_loss=0.1119, over 5692837.78 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3768, pruned_loss=0.1207, over 5653880.30 frames. ], batch size: 82, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:44:38,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4685, 1.7488, 1.3711, 1.6457], device='cuda:1'), covar=tensor([0.0798, 0.0308, 0.0347, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:44:43,067 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:44:47,186 INFO [train.py:968] (1/2) Epoch 27, batch 42150, giga_loss[loss=0.2888, simple_loss=0.364, pruned_loss=0.1068, over 29015.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3731, pruned_loss=0.1191, over 5665114.02 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3565, pruned_loss=0.1118, over 5692030.28 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3763, pruned_loss=0.1208, over 5652034.12 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:44:49,616 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:44:53,613 INFO [optim.py:369] (1/2) 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 05:45:31,629 INFO [train.py:968] (1/2) Epoch 27, batch 42200, giga_loss[loss=0.3126, simple_loss=0.3722, pruned_loss=0.1265, over 28780.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3725, pruned_loss=0.1197, over 5676282.72 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3567, pruned_loss=0.1118, over 5695400.68 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3751, pruned_loss=0.1211, over 5662466.94 frames. ], batch size: 284, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:46:20,819 INFO [train.py:968] (1/2) Epoch 27, batch 42250, giga_loss[loss=0.3134, simple_loss=0.3849, pruned_loss=0.1209, over 28831.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3712, pruned_loss=0.1202, over 5666019.94 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3567, pruned_loss=0.1117, over 5691952.05 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3738, pruned_loss=0.1217, over 5657007.57 frames. ], batch size: 174, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:46:25,510 INFO [optim.py:369] (1/2) 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 05:46:30,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6570, 4.7229, 1.7490, 1.8252], device='cuda:1'), covar=tensor([0.0992, 0.0363, 0.0905, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0573, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 05:46:59,219 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:02,299 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:07,045 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:09,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:12,291 INFO [train.py:968] (1/2) Epoch 27, batch 42300, giga_loss[loss=0.2783, simple_loss=0.3675, pruned_loss=0.09452, over 29055.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3702, pruned_loss=0.119, over 5657032.60 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3568, pruned_loss=0.1118, over 5683046.28 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3721, pruned_loss=0.12, over 5658355.99 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:47:19,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:30,104 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:36,831 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:41,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:47:57,121 INFO [train.py:968] (1/2) Epoch 27, batch 42350, giga_loss[loss=0.2793, simple_loss=0.3579, pruned_loss=0.1004, over 29067.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3701, pruned_loss=0.1174, over 5662080.08 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3568, pruned_loss=0.1119, over 5676620.11 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3718, pruned_loss=0.1184, over 5669047.79 frames. ], batch size: 136, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 05:48:06,068 INFO [optim.py:369] (1/2) 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 05:48:20,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 05:48:45,651 INFO [train.py:968] (1/2) Epoch 27, batch 42400, giga_loss[loss=0.2499, simple_loss=0.3251, pruned_loss=0.08736, over 28658.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3691, pruned_loss=0.1166, over 5665399.29 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3566, pruned_loss=0.1115, over 5680102.84 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.371, pruned_loss=0.1177, over 5667615.62 frames. ], batch size: 60, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:48:50,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-14 05:49:35,038 INFO [train.py:968] (1/2) Epoch 27, batch 42450, giga_loss[loss=0.2706, simple_loss=0.3451, pruned_loss=0.09808, over 28811.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3697, pruned_loss=0.1175, over 5665444.12 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3566, pruned_loss=0.1115, over 5681215.32 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3712, pruned_loss=0.1184, over 5666078.07 frames. ], batch size: 119, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:49:35,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:38,424 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227308.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:39,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1227309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:40,772 INFO [optim.py:369] (1/2) 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 05:49:54,501 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1227327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:49:57,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:50:02,714 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:50:17,201 INFO [train.py:968] (1/2) Epoch 27, batch 42500, giga_loss[loss=0.3404, simple_loss=0.3923, pruned_loss=0.1442, over 27578.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3671, pruned_loss=0.116, over 5679595.47 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3571, pruned_loss=0.1117, over 5687426.34 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3683, pruned_loss=0.1168, over 5673695.58 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:50:20,830 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:51:05,457 INFO [train.py:968] (1/2) Epoch 27, batch 42550, giga_loss[loss=0.2871, simple_loss=0.3363, pruned_loss=0.1189, over 23576.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3664, pruned_loss=0.1165, over 5658549.54 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3572, pruned_loss=0.1117, over 5679371.00 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3674, pruned_loss=0.1172, over 5661660.41 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:51:11,719 INFO [optim.py:369] (1/2) 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 05:51:27,517 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4027, 1.7852, 1.3948, 1.2443], device='cuda:1'), covar=tensor([0.2686, 0.2681, 0.3171, 0.2557], device='cuda:1'), in_proj_covar=tensor([0.1588, 0.1144, 0.1401, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 05:51:49,888 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 42600, giga_loss[loss=0.2801, simple_loss=0.3485, pruned_loss=0.1058, over 28922.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3645, pruned_loss=0.1158, over 5675190.28 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3567, pruned_loss=0.1114, over 5684694.20 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.366, pruned_loss=0.1167, over 5672655.88 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:51:51,860 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1227455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:52:21,953 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1227484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 05:52:39,869 INFO [train.py:968] (1/2) Epoch 27, batch 42650, giga_loss[loss=0.2663, simple_loss=0.3443, pruned_loss=0.09419, over 28973.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3631, pruned_loss=0.1154, over 5677844.75 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3568, pruned_loss=0.1113, over 5686239.72 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3644, pruned_loss=0.1164, over 5674199.49 frames. ], batch size: 164, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:52:45,281 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6928, 1.6701, 1.8406, 1.4314], device='cuda:1'), covar=tensor([0.1849, 0.2614, 0.1520, 0.1748], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0715, 0.0975, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 05:52:45,536 INFO [optim.py:369] (1/2) 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 05:53:08,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4251, 1.9936, 1.6442, 1.6414], device='cuda:1'), covar=tensor([0.0791, 0.0284, 0.0318, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 05:53:26,088 INFO [train.py:968] (1/2) Epoch 27, batch 42700, giga_loss[loss=0.276, simple_loss=0.3441, pruned_loss=0.1039, over 28948.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3633, pruned_loss=0.1166, over 5652792.75 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3569, pruned_loss=0.1115, over 5676590.22 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3645, pruned_loss=0.1174, over 5657689.05 frames. ], batch size: 227, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:53:42,500 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5760, 1.6481, 1.7775, 1.3511], device='cuda:1'), covar=tensor([0.1769, 0.2765, 0.1501, 0.1849], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0715, 0.0975, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 05:54:13,297 INFO [train.py:968] (1/2) Epoch 27, batch 42750, giga_loss[loss=0.2759, simple_loss=0.3416, pruned_loss=0.1051, over 28615.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3637, pruned_loss=0.1172, over 5646628.84 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.357, pruned_loss=0.1115, over 5674423.49 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3648, pruned_loss=0.118, over 5652189.54 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:54:20,110 INFO [optim.py:369] (1/2) 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 05:54:58,098 INFO [train.py:968] (1/2) Epoch 27, batch 42800, giga_loss[loss=0.3284, simple_loss=0.3815, pruned_loss=0.1377, over 27458.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3644, pruned_loss=0.1169, over 5657100.42 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3574, pruned_loss=0.1117, over 5675521.43 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3651, pruned_loss=0.1175, over 5660004.55 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 05:55:42,750 INFO [train.py:968] (1/2) Epoch 27, batch 42850, giga_loss[loss=0.3227, simple_loss=0.3855, pruned_loss=0.1299, over 28743.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3646, pruned_loss=0.1159, over 5665262.67 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3573, pruned_loss=0.1115, over 5678686.50 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3653, pruned_loss=0.1166, over 5664382.73 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:55:48,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5286, 1.7541, 1.4464, 1.6059], device='cuda:1'), covar=tensor([0.2873, 0.2898, 0.3157, 0.2520], device='cuda:1'), in_proj_covar=tensor([0.1590, 0.1145, 0.1403, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 05:55:50,219 INFO [optim.py:369] (1/2) 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 05:56:25,535 INFO [train.py:968] (1/2) Epoch 27, batch 42900, giga_loss[loss=0.3026, simple_loss=0.365, pruned_loss=0.1201, over 27524.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3642, pruned_loss=0.1149, over 5673737.56 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3572, pruned_loss=0.1114, over 5679726.74 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3653, pruned_loss=0.1157, over 5671616.88 frames. ], batch size: 472, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:57:18,403 INFO [train.py:968] (1/2) Epoch 27, batch 42950, giga_loss[loss=0.3081, simple_loss=0.3738, pruned_loss=0.1212, over 28942.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3657, pruned_loss=0.1162, over 5680058.47 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3577, pruned_loss=0.112, over 5683866.54 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3662, pruned_loss=0.1164, over 5674712.78 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:57:23,691 INFO [optim.py:369] (1/2) 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 05:57:37,175 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4072, 1.4744, 1.4077, 1.4920], device='cuda:1'), covar=tensor([0.0766, 0.0343, 0.0320, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:57:42,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5616, 1.9176, 1.4671, 1.7227], device='cuda:1'), covar=tensor([0.2684, 0.2724, 0.3124, 0.2378], device='cuda:1'), in_proj_covar=tensor([0.1586, 0.1143, 0.1400, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 05:57:46,542 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 05:58:02,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3631, 1.6612, 1.5942, 1.5320], device='cuda:1'), covar=tensor([0.2071, 0.1868, 0.2340, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0760, 0.0730, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 05:58:02,519 INFO [train.py:968] (1/2) Epoch 27, batch 43000, giga_loss[loss=0.3166, simple_loss=0.3765, pruned_loss=0.1283, over 28670.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3678, pruned_loss=0.118, over 5684091.99 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.358, pruned_loss=0.1121, over 5681856.40 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3682, pruned_loss=0.1182, over 5680619.19 frames. ], batch size: 242, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:58:34,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-14 05:58:56,852 INFO [train.py:968] (1/2) Epoch 27, batch 43050, giga_loss[loss=0.2893, simple_loss=0.354, pruned_loss=0.1123, over 28826.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3692, pruned_loss=0.1205, over 5677690.35 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3581, pruned_loss=0.1122, over 5685269.55 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3696, pruned_loss=0.1208, over 5671971.69 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 05:59:05,374 INFO [optim.py:369] (1/2) 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 05:59:14,299 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4558, 1.9949, 1.6366, 1.5837], device='cuda:1'), covar=tensor([0.0768, 0.0289, 0.0317, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 05:59:46,566 INFO [train.py:968] (1/2) Epoch 27, batch 43100, giga_loss[loss=0.2783, simple_loss=0.3476, pruned_loss=0.1045, over 29014.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3719, pruned_loss=0.1235, over 5677426.89 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3586, pruned_loss=0.1126, over 5689745.83 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.372, pruned_loss=0.1236, over 5668451.43 frames. ], batch size: 128, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:00:25,827 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1227996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:00:31,895 INFO [train.py:968] (1/2) Epoch 27, batch 43150, giga_loss[loss=0.3198, simple_loss=0.3708, pruned_loss=0.1344, over 28756.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3727, pruned_loss=0.1244, over 5673736.75 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3588, pruned_loss=0.1127, over 5694809.88 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3728, pruned_loss=0.1246, over 5661397.27 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:00:40,371 INFO [optim.py:369] (1/2) 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 06:01:15,151 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6209, 1.9066, 1.2961, 1.5563], device='cuda:1'), covar=tensor([0.1044, 0.0693, 0.1093, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0409, 0.0449, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 06:01:20,990 INFO [train.py:968] (1/2) Epoch 27, batch 43200, giga_loss[loss=0.2788, simple_loss=0.3508, pruned_loss=0.1034, over 28508.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3717, pruned_loss=0.1237, over 5664482.47 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.359, pruned_loss=0.1128, over 5688731.71 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3719, pruned_loss=0.124, over 5659167.63 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 8.0
+2023-03-14 06:01:29,152 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 06:01:49,289 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:02:06,832 INFO [train.py:968] (1/2) Epoch 27, batch 43250, giga_loss[loss=0.2818, simple_loss=0.3643, pruned_loss=0.09967, over 28854.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3713, pruned_loss=0.1217, over 5666854.98 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3591, pruned_loss=0.1128, over 5694328.46 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3717, pruned_loss=0.1222, over 5657274.33 frames. ], batch size: 199, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:02:13,644 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3587, 1.4103, 4.1421, 3.4901], device='cuda:1'), covar=tensor([0.2054, 0.2981, 0.0924, 0.1036], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0676, 0.1010, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 06:02:14,031 INFO [optim.py:369] (1/2) 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 06:02:37,878 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-14 06:02:52,276 INFO [train.py:968] (1/2) Epoch 27, batch 43300, giga_loss[loss=0.301, simple_loss=0.3656, pruned_loss=0.1182, over 28616.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3688, pruned_loss=0.1193, over 5667244.62 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5697648.19 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3696, pruned_loss=0.1199, over 5656097.70 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:03:14,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5525, 2.0954, 1.6460, 1.7797], device='cuda:1'), covar=tensor([0.0793, 0.0276, 0.0313, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 06:03:18,595 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 06:03:21,363 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:03:32,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:03:36,223 INFO [train.py:968] (1/2) Epoch 27, batch 43350, libri_loss[loss=0.3813, simple_loss=0.4186, pruned_loss=0.172, over 18770.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3675, pruned_loss=0.1188, over 5666481.94 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3595, pruned_loss=0.1132, over 5690858.58 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3677, pruned_loss=0.1189, over 5664155.63 frames. ], batch size: 187, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:03:41,932 INFO [optim.py:369] (1/2) 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 06:04:20,714 INFO [train.py:968] (1/2) Epoch 27, batch 43400, giga_loss[loss=0.2927, simple_loss=0.3486, pruned_loss=0.1184, over 28649.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3665, pruned_loss=0.1192, over 5664029.45 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3597, pruned_loss=0.1133, over 5692592.95 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3667, pruned_loss=0.1193, over 5660409.19 frames. ], batch size: 92, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:05:05,625 INFO [train.py:968] (1/2) Epoch 27, batch 43450, giga_loss[loss=0.2828, simple_loss=0.3503, pruned_loss=0.1076, over 28705.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3669, pruned_loss=0.1198, over 5675319.41 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.359, pruned_loss=0.113, over 5698492.63 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3678, pruned_loss=0.1203, over 5666227.01 frames. ], batch size: 99, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:05:12,037 INFO [optim.py:369] (1/2) 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 06:05:49,566 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:05:49,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3038, 1.4831, 1.3564, 1.5841], device='cuda:1'), covar=tensor([0.0788, 0.0399, 0.0370, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 06:05:52,628 INFO [train.py:968] (1/2) Epoch 27, batch 43500, giga_loss[loss=0.2689, simple_loss=0.3587, pruned_loss=0.08958, over 28487.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3705, pruned_loss=0.121, over 5671662.91 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.359, pruned_loss=0.113, over 5700888.87 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3713, pruned_loss=0.1215, over 5662110.60 frames. ], batch size: 71, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:06:07,495 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:06:43,538 INFO [train.py:968] (1/2) Epoch 27, batch 43550, giga_loss[loss=0.2918, simple_loss=0.3474, pruned_loss=0.1181, over 23741.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3713, pruned_loss=0.1185, over 5675814.60 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3589, pruned_loss=0.1129, over 5704190.14 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3723, pruned_loss=0.1191, over 5664985.56 frames. ], batch size: 705, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:06:52,354 INFO [optim.py:369] (1/2) 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 06:07:32,816 INFO [train.py:968] (1/2) Epoch 27, batch 43600, giga_loss[loss=0.3433, simple_loss=0.4047, pruned_loss=0.141, over 28872.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3732, pruned_loss=0.1199, over 5674836.59 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3589, pruned_loss=0.113, over 5708310.24 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3742, pruned_loss=0.1204, over 5662045.41 frames. ], batch size: 186, lr: 1.17e-03, grad_scale: 4.0
+2023-03-14 06:07:41,910 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:07:41,941 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:08:15,900 INFO [train.py:968] (1/2) Epoch 27, batch 43650, libri_loss[loss=0.2606, simple_loss=0.3304, pruned_loss=0.09539, over 29563.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3747, pruned_loss=0.1215, over 5658026.39 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3586, pruned_loss=0.113, over 5695644.58 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3763, pruned_loss=0.1222, over 5657666.79 frames. ], batch size: 76, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:08:25,193 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:08:26,013 INFO [optim.py:369] (1/2) 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 06:08:28,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:08:56,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:04,274 INFO [train.py:968] (1/2) Epoch 27, batch 43700, giga_loss[loss=0.2946, simple_loss=0.3607, pruned_loss=0.1143, over 28834.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3752, pruned_loss=0.1227, over 5660825.77 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3585, pruned_loss=0.113, over 5699456.30 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3768, pruned_loss=0.1234, over 5656532.56 frames. ], batch size: 112, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:09:10,046 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1228562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:10,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:20,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:41,716 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 43750, giga_loss[loss=0.313, simple_loss=0.3722, pruned_loss=0.1269, over 28569.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3745, pruned_loss=0.123, over 5666489.34 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3589, pruned_loss=0.1131, over 5694798.37 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.376, pruned_loss=0.1239, over 5665362.09 frames. ], batch size: 307, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:09:47,104 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:50,881 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:09:56,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4807, 1.5800, 1.7122, 1.2800], device='cuda:1'), covar=tensor([0.1797, 0.2653, 0.1451, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0714, 0.0973, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 06:09:57,155 INFO [optim.py:369] (1/2) 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 06:10:17,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:10:33,650 INFO [train.py:968] (1/2) Epoch 27, batch 43800, giga_loss[loss=0.2753, simple_loss=0.3424, pruned_loss=0.1041, over 28746.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3719, pruned_loss=0.1221, over 5658991.54 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3586, pruned_loss=0.1129, over 5696195.70 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3736, pruned_loss=0.123, over 5656284.10 frames. ], batch size: 66, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:11:21,181 INFO [train.py:968] (1/2) Epoch 27, batch 43850, giga_loss[loss=0.2834, simple_loss=0.3467, pruned_loss=0.11, over 28621.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.37, pruned_loss=0.1212, over 5672675.19 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.359, pruned_loss=0.1132, over 5700772.00 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3713, pruned_loss=0.1218, over 5665890.34 frames. ], batch size: 85, lr: 1.17e-03, grad_scale: 2.0
+2023-03-14 06:11:21,552 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:24,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:32,742 INFO [optim.py:369] (1/2) 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 06:11:35,524 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:37,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:43,541 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:11:56,192 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:12:07,235 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 43900, giga_loss[loss=0.4476, simple_loss=0.4743, pruned_loss=0.2104, over 27701.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.371, pruned_loss=0.1226, over 5662855.33 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3588, pruned_loss=0.1131, over 5692500.69 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3724, pruned_loss=0.1235, over 5664797.13 frames. ], batch size: 472, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:13:01,111 INFO [train.py:968] (1/2) Epoch 27, batch 43950, giga_loss[loss=0.4069, simple_loss=0.4326, pruned_loss=0.1906, over 26824.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3723, pruned_loss=0.1238, over 5651847.02 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3593, pruned_loss=0.1135, over 5681788.97 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3732, pruned_loss=0.1243, over 5662890.03 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:13:10,773 INFO [optim.py:369] (1/2) 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 06:13:28,943 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:13:39,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7031, 3.5480, 3.3884, 1.8427], device='cuda:1'), covar=tensor([0.0850, 0.0918, 0.0863, 0.2167], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.1215, 0.1025, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 06:13:41,225 INFO [train.py:968] (1/2) Epoch 27, batch 44000, giga_loss[loss=0.2844, simple_loss=0.348, pruned_loss=0.1104, over 28917.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3714, pruned_loss=0.1235, over 5655959.80 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1137, over 5677476.51 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3723, pruned_loss=0.1241, over 5668371.56 frames. ], batch size: 227, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:13:56,875 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:14:00,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1228872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:14:06,206 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5433, 1.7778, 1.7601, 1.6132], device='cuda:1'), covar=tensor([0.2299, 0.2441, 0.2635, 0.2474], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0764, 0.0732, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 06:14:26,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1228901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:14:30,281 INFO [train.py:968] (1/2) Epoch 27, batch 44050, giga_loss[loss=0.2665, simple_loss=0.3363, pruned_loss=0.09836, over 28911.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3689, pruned_loss=0.1219, over 5662609.71 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3595, pruned_loss=0.1135, over 5681115.14 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3698, pruned_loss=0.1227, over 5669098.10 frames. ], batch size: 112, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:14:40,093 INFO [optim.py:369] (1/2) 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 06:14:45,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4665, 3.4722, 1.5640, 1.5215], device='cuda:1'), covar=tensor([0.1020, 0.0356, 0.0938, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0573, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 06:14:49,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5701, 1.6409, 1.7000, 1.5269], device='cuda:1'), covar=tensor([0.3087, 0.2935, 0.2396, 0.2440], device='cuda:1'), in_proj_covar=tensor([0.2067, 0.2022, 0.1935, 0.2071], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 06:14:56,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:15:12,378 INFO [train.py:968] (1/2) Epoch 27, batch 44100, giga_loss[loss=0.3445, simple_loss=0.3995, pruned_loss=0.1448, over 27554.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3688, pruned_loss=0.1213, over 5668547.72 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.359, pruned_loss=0.1131, over 5686957.60 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3703, pruned_loss=0.1226, over 5668051.43 frames. ], batch size: 472, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:15:34,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1228976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:15:37,585 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1228981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:15:40,220 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 44150, giga_loss[loss=0.3162, simple_loss=0.3835, pruned_loss=0.1245, over 28626.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3701, pruned_loss=0.1213, over 5657635.89 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3587, pruned_loss=0.1127, over 5683122.49 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3719, pruned_loss=0.1229, over 5660307.64 frames. ], batch size: 307, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:16:09,004 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:16:10,529 INFO [optim.py:369] (1/2) 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 06:16:10,776 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229016.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:16:21,317 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:16:41,425 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3295, 1.5799, 1.6635, 1.2691], device='cuda:1'), covar=tensor([0.3001, 0.2398, 0.1773, 0.2390], device='cuda:1'), in_proj_covar=tensor([0.2067, 0.2022, 0.1935, 0.2071], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 06:16:45,986 INFO [train.py:968] (1/2) Epoch 27, batch 44200, giga_loss[loss=0.282, simple_loss=0.3534, pruned_loss=0.1053, over 29017.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3712, pruned_loss=0.1224, over 5663856.44 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3592, pruned_loss=0.1132, over 5678316.09 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3724, pruned_loss=0.1234, over 5670807.28 frames. ], batch size: 164, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:17:11,049 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:14,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:33,689 INFO [train.py:968] (1/2) Epoch 27, batch 44250, giga_loss[loss=0.3262, simple_loss=0.3659, pruned_loss=0.1433, over 23835.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3715, pruned_loss=0.1221, over 5655765.00 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3588, pruned_loss=0.1128, over 5679767.29 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3731, pruned_loss=0.1234, over 5659949.63 frames. ], batch size: 705, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:17:40,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:43,988 INFO [optim.py:369] (1/2) 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 06:17:44,866 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:47,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:17:50,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:18:12,870 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:18:18,083 INFO [train.py:968] (1/2) Epoch 27, batch 44300, giga_loss[loss=0.2458, simple_loss=0.3397, pruned_loss=0.07595, over 28980.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3717, pruned_loss=0.1197, over 5667349.87 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3589, pruned_loss=0.113, over 5682172.83 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3732, pruned_loss=0.1208, over 5668125.10 frames. ], batch size: 155, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:18:45,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4680, 1.6949, 1.5685, 1.5977], device='cuda:1'), covar=tensor([0.0621, 0.0284, 0.0275, 0.0668], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 06:19:00,286 INFO [train.py:968] (1/2) Epoch 27, batch 44350, giga_loss[loss=0.3251, simple_loss=0.3923, pruned_loss=0.1289, over 28640.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.373, pruned_loss=0.1187, over 5667171.88 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3591, pruned_loss=0.1132, over 5676531.42 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3743, pruned_loss=0.1195, over 5673268.76 frames. ], batch size: 336, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:19:15,784 INFO [optim.py:369] (1/2) 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 06:19:31,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-14 06:19:47,620 INFO [train.py:968] (1/2) Epoch 27, batch 44400, giga_loss[loss=0.4834, simple_loss=0.4809, pruned_loss=0.243, over 26510.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.376, pruned_loss=0.1206, over 5683280.90 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3592, pruned_loss=0.1133, over 5683958.79 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3775, pruned_loss=0.1213, over 5681179.16 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:19:53,871 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:20:37,411 INFO [train.py:968] (1/2) Epoch 27, batch 44450, giga_loss[loss=0.3106, simple_loss=0.3774, pruned_loss=0.1219, over 28668.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3781, pruned_loss=0.1235, over 5675503.43 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.359, pruned_loss=0.1131, over 5688689.28 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3797, pruned_loss=0.1244, over 5669540.87 frames. ], batch size: 242, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:20:46,843 INFO [optim.py:369] (1/2) 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 06:20:51,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4103, 1.6618, 1.6885, 1.4854], device='cuda:1'), covar=tensor([0.1457, 0.1133, 0.1482, 0.1274], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0761, 0.0730, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 06:21:10,725 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:21:22,255 INFO [train.py:968] (1/2) Epoch 27, batch 44500, giga_loss[loss=0.2805, simple_loss=0.3506, pruned_loss=0.1052, over 28929.00 frames. ], tot_loss[loss=0.315, simple_loss=0.379, pruned_loss=0.1255, over 5645284.52 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3593, pruned_loss=0.1134, over 5673663.34 frames. ], giga_tot_loss[loss=0.3167, simple_loss=0.3808, pruned_loss=0.1263, over 5652580.16 frames. ], batch size: 145, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:21:24,774 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:21:55,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229391.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 06:22:06,675 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 44550, giga_loss[loss=0.3551, simple_loss=0.3994, pruned_loss=0.1554, over 26444.00 frames. ], tot_loss[loss=0.3157, simple_loss=0.3794, pruned_loss=0.126, over 5651057.98 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3595, pruned_loss=0.1135, over 5675711.92 frames. ], giga_tot_loss[loss=0.3171, simple_loss=0.3808, pruned_loss=0.1267, over 5654816.27 frames. ], batch size: 555, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:22:15,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1812, 1.5618, 1.3580, 1.3240], device='cuda:1'), covar=tensor([0.2138, 0.1902, 0.2226, 0.2063], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0762, 0.0730, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 06:22:19,512 INFO [optim.py:369] (1/2) 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 06:22:53,561 INFO [train.py:968] (1/2) Epoch 27, batch 44600, giga_loss[loss=0.2884, simple_loss=0.3726, pruned_loss=0.1021, over 29050.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3781, pruned_loss=0.1237, over 5657516.92 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3596, pruned_loss=0.1136, over 5676566.36 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3791, pruned_loss=0.1243, over 5659520.90 frames. ], batch size: 136, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:22:55,035 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:23:29,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:23:42,738 INFO [train.py:968] (1/2) Epoch 27, batch 44650, giga_loss[loss=0.3434, simple_loss=0.3823, pruned_loss=0.1523, over 23469.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.3788, pruned_loss=0.123, over 5654648.51 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3595, pruned_loss=0.1137, over 5676502.68 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3799, pruned_loss=0.1234, over 5656371.76 frames. ], batch size: 705, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:23:52,663 INFO [optim.py:369] (1/2) 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 06:24:06,773 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:24:10,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229537.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 06:24:19,999 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:24:24,331 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:24:25,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5453, 1.7164, 1.2659, 1.3207], device='cuda:1'), covar=tensor([0.0953, 0.0494, 0.0933, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0451, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 06:24:28,942 INFO [train.py:968] (1/2) Epoch 27, batch 44700, giga_loss[loss=0.3736, simple_loss=0.4231, pruned_loss=0.162, over 28519.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3793, pruned_loss=0.1229, over 5667154.96 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3598, pruned_loss=0.1138, over 5678464.92 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3801, pruned_loss=0.1231, over 5666615.71 frames. ], batch size: 336, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:24:40,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:24:51,032 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:25:00,841 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-14 06:25:15,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3275, 1.3900, 1.4165, 1.2954], device='cuda:1'), covar=tensor([0.2202, 0.2361, 0.1799, 0.2081], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2022, 0.1931, 0.2070], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 06:25:16,416 INFO [train.py:968] (1/2) Epoch 27, batch 44750, giga_loss[loss=0.3427, simple_loss=0.4041, pruned_loss=0.1406, over 28915.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3787, pruned_loss=0.1232, over 5670147.69 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3593, pruned_loss=0.1135, over 5680234.13 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3802, pruned_loss=0.124, over 5667700.40 frames. ], batch size: 164, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:25:27,316 INFO [optim.py:369] (1/2) 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 06:25:44,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:25:44,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:25:46,469 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:25:52,360 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5855, 1.5867, 1.7647, 1.3407], device='cuda:1'), covar=tensor([0.1838, 0.2695, 0.1493, 0.1781], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0719, 0.0979, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 06:26:02,509 INFO [train.py:968] (1/2) Epoch 27, batch 44800, giga_loss[loss=0.3484, simple_loss=0.3979, pruned_loss=0.1495, over 27956.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3785, pruned_loss=0.124, over 5664835.25 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3594, pruned_loss=0.1135, over 5683134.40 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.38, pruned_loss=0.1249, over 5660018.77 frames. ], batch size: 412, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:26:14,741 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:26:32,165 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6988, 1.9430, 1.5470, 1.8588], device='cuda:1'), covar=tensor([0.2888, 0.2987, 0.3354, 0.2837], device='cuda:1'), in_proj_covar=tensor([0.1591, 0.1147, 0.1406, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 06:26:55,571 INFO [train.py:968] (1/2) Epoch 27, batch 44850, giga_loss[loss=0.2862, simple_loss=0.349, pruned_loss=0.1117, over 28688.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3757, pruned_loss=0.1235, over 5654291.58 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3593, pruned_loss=0.1133, over 5684263.73 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.377, pruned_loss=0.1243, over 5649493.01 frames. ], batch size: 92, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:27:05,308 INFO [optim.py:369] (1/2) 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 06:27:06,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:27:07,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5560, 1.7452, 1.6244, 1.5693], device='cuda:1'), covar=tensor([0.1918, 0.2173, 0.2359, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0764, 0.0732, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 06:27:20,620 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 44900, giga_loss[loss=0.3103, simple_loss=0.3695, pruned_loss=0.1255, over 28886.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3735, pruned_loss=0.1225, over 5661525.46 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3592, pruned_loss=0.1132, over 5688828.86 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.375, pruned_loss=0.1235, over 5652936.92 frames. ], batch size: 99, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:27:40,084 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1229755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:28:02,936 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:28:05,208 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:28:23,862 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 06:28:26,115 INFO [train.py:968] (1/2) Epoch 27, batch 44950, giga_loss[loss=0.3311, simple_loss=0.3847, pruned_loss=0.1388, over 28697.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3718, pruned_loss=0.1219, over 5659599.60 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3591, pruned_loss=0.1131, over 5688765.27 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3733, pruned_loss=0.1229, over 5652352.86 frames. ], batch size: 119, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:28:30,530 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:28:50,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1229832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:12,117 INFO [train.py:968] (1/2) Epoch 27, batch 45000, giga_loss[loss=0.2968, simple_loss=0.3633, pruned_loss=0.1151, over 27872.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3714, pruned_loss=0.1216, over 5673471.50 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3588, pruned_loss=0.1128, over 5693216.75 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3731, pruned_loss=0.1229, over 5663307.79 frames. ], batch size: 412, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:29:12,117 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 06:29:20,636 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 06:29:25,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:28,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:39,945 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:43,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:29:44,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8422, 2.0047, 1.9118, 1.7374], device='cuda:1'), covar=tensor([0.2167, 0.2768, 0.2449, 0.2578], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0766, 0.0733, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 06:29:54,046 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 45050, giga_loss[loss=0.2569, simple_loss=0.3313, pruned_loss=0.09121, over 28442.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3688, pruned_loss=0.119, over 5667590.51 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3588, pruned_loss=0.1128, over 5694970.30 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3703, pruned_loss=0.1201, over 5657746.44 frames. ], batch size: 85, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:30:08,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1229907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:30:19,861 INFO [optim.py:369] (1/2) 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 06:30:27,075 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 06:30:49,182 INFO [train.py:968] (1/2) Epoch 27, batch 45100, giga_loss[loss=0.2832, simple_loss=0.3682, pruned_loss=0.09907, over 29001.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3652, pruned_loss=0.1156, over 5667323.67 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5690887.67 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3657, pruned_loss=0.1158, over 5661468.87 frames. ], batch size: 164, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:31:05,582 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1229975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:31:09,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1229978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:31:36,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-14 06:31:36,814 INFO [train.py:968] (1/2) Epoch 27, batch 45150, giga_loss[loss=0.3051, simple_loss=0.3692, pruned_loss=0.1205, over 28584.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3651, pruned_loss=0.1156, over 5666242.08 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3594, pruned_loss=0.1135, over 5693352.04 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3659, pruned_loss=0.1159, over 5659133.02 frames. ], batch size: 336, lr: 1.16e-03, grad_scale: 2.0
+2023-03-14 06:31:40,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1230007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:31:51,269 INFO [optim.py:369] (1/2) 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 06:31:57,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5378, 1.8594, 1.4436, 1.3417], device='cuda:1'), covar=tensor([0.1117, 0.0576, 0.1026, 0.1218], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0456, 0.0528, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 06:32:23,343 INFO [train.py:968] (1/2) Epoch 27, batch 45200, giga_loss[loss=0.2834, simple_loss=0.3523, pruned_loss=0.1072, over 28992.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3626, pruned_loss=0.1148, over 5659634.38 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3589, pruned_loss=0.1132, over 5697792.43 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3638, pruned_loss=0.1154, over 5649373.65 frames. ], batch size: 155, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:33:14,168 INFO [train.py:968] (1/2) Epoch 27, batch 45250, giga_loss[loss=0.2851, simple_loss=0.3557, pruned_loss=0.1072, over 28924.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3614, pruned_loss=0.1153, over 5652008.20 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3593, pruned_loss=0.1135, over 5700783.16 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3621, pruned_loss=0.1155, over 5640416.32 frames. ], batch size: 199, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:33:25,771 INFO [optim.py:369] (1/2) 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 06:33:34,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1230130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:33:54,328 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 45300, giga_loss[loss=0.3457, simple_loss=0.4016, pruned_loss=0.1449, over 27918.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3625, pruned_loss=0.1155, over 5653107.14 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3595, pruned_loss=0.1134, over 5708238.07 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.363, pruned_loss=0.1158, over 5635126.02 frames. ], batch size: 412, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:34:35,628 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6360, 1.7844, 1.7134, 1.5357], device='cuda:1'), covar=tensor([0.2170, 0.2264, 0.2624, 0.2438], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0766, 0.0734, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 06:34:42,374 INFO [train.py:968] (1/2) Epoch 27, batch 45350, giga_loss[loss=0.2929, simple_loss=0.3673, pruned_loss=0.1093, over 28678.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3652, pruned_loss=0.1164, over 5655907.19 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5706364.15 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3653, pruned_loss=0.1164, over 5642712.85 frames. ], batch size: 262, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:34:56,943 INFO [optim.py:369] (1/2) 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 06:35:12,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6205, 1.8264, 1.4824, 1.7917], device='cuda:1'), covar=tensor([0.2985, 0.2967, 0.3357, 0.2641], device='cuda:1'), in_proj_covar=tensor([0.1599, 0.1151, 0.1409, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 06:35:32,288 INFO [train.py:968] (1/2) Epoch 27, batch 45400, giga_loss[loss=0.2766, simple_loss=0.3451, pruned_loss=0.104, over 28807.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3657, pruned_loss=0.1169, over 5637147.14 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.36, pruned_loss=0.1137, over 5704580.37 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3657, pruned_loss=0.1169, over 5627852.35 frames. ], batch size: 112, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:35:47,907 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1230273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:35:50,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1230276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:36:16,939 INFO [train.py:968] (1/2) Epoch 27, batch 45450, giga_loss[loss=0.3271, simple_loss=0.3784, pruned_loss=0.1379, over 28692.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3652, pruned_loss=0.1169, over 5640171.93 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3599, pruned_loss=0.1136, over 5704817.07 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.117, over 5631426.12 frames. ], batch size: 92, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:36:17,195 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:36:48,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2504, 4.1121, 3.9167, 1.7609], device='cuda:1'), covar=tensor([0.0598, 0.0683, 0.0733, 0.2145], device='cuda:1'), in_proj_covar=tensor([0.1318, 0.1218, 0.1026, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 06:37:00,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 06:37:01,158 INFO [train.py:968] (1/2) Epoch 27, batch 45500, giga_loss[loss=0.2829, simple_loss=0.3582, pruned_loss=0.1038, over 29066.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3664, pruned_loss=0.1177, over 5637611.69 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3603, pruned_loss=0.1139, over 5696824.92 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3662, pruned_loss=0.1176, over 5637507.34 frames. ], batch size: 128, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:37:50,163 INFO [train.py:968] (1/2) Epoch 27, batch 45550, giga_loss[loss=0.3044, simple_loss=0.3835, pruned_loss=0.1127, over 28921.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.369, pruned_loss=0.119, over 5642938.86 frames. ], libri_tot_loss[loss=0.2948, simple_loss=0.361, pruned_loss=0.1143, over 5691288.73 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3684, pruned_loss=0.1186, over 5647141.91 frames. ], batch size: 199, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:38:03,542 INFO [optim.py:369] (1/2) 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 06:38:35,213 INFO [train.py:968] (1/2) Epoch 27, batch 45600, giga_loss[loss=0.3353, simple_loss=0.394, pruned_loss=0.1384, over 28535.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3703, pruned_loss=0.1203, over 5651012.14 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3608, pruned_loss=0.1142, over 5695661.12 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3702, pruned_loss=0.1201, over 5649347.18 frames. ], batch size: 336, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:39:26,327 INFO [train.py:968] (1/2) Epoch 27, batch 45650, giga_loss[loss=0.2992, simple_loss=0.3601, pruned_loss=0.1192, over 28794.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3718, pruned_loss=0.1222, over 5648113.20 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3603, pruned_loss=0.1141, over 5699192.22 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3723, pruned_loss=0.1224, over 5642532.39 frames. ], batch size: 119, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:39:40,335 INFO [optim.py:369] (1/2) 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 06:39:47,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1230528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:40:14,115 INFO [train.py:968] (1/2) Epoch 27, batch 45700, giga_loss[loss=0.2986, simple_loss=0.3795, pruned_loss=0.1088, over 29003.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3721, pruned_loss=0.1218, over 5655992.69 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.36, pruned_loss=0.1138, over 5704242.36 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3731, pruned_loss=0.1224, over 5645979.96 frames. ], batch size: 164, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:41:06,155 INFO [train.py:968] (1/2) Epoch 27, batch 45750, giga_loss[loss=0.2973, simple_loss=0.3561, pruned_loss=0.1192, over 28646.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3722, pruned_loss=0.1206, over 5659880.26 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3598, pruned_loss=0.1137, over 5706617.70 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3734, pruned_loss=0.1214, over 5649122.49 frames. ], batch size: 92, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:41:20,254 INFO [optim.py:369] (1/2) 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 06:41:24,366 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6234, 1.8067, 1.2816, 1.4160], device='cuda:1'), covar=tensor([0.1032, 0.0604, 0.1091, 0.1091], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0456, 0.0527, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 06:41:36,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4730, 2.5873, 1.5824, 1.5870], device='cuda:1'), covar=tensor([0.0828, 0.0369, 0.0723, 0.1149], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0574, 0.0410, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 06:41:36,675 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:41:56,115 INFO [train.py:968] (1/2) Epoch 27, batch 45800, giga_loss[loss=0.274, simple_loss=0.3437, pruned_loss=0.1021, over 28951.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3708, pruned_loss=0.1205, over 5643349.95 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3597, pruned_loss=0.1136, over 5696413.11 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3721, pruned_loss=0.1212, over 5643045.47 frames. ], batch size: 128, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:42:13,292 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1230671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:42:15,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1230674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:42:16,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:42:47,106 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 27, batch 45850, giga_loss[loss=0.3167, simple_loss=0.3703, pruned_loss=0.1316, over 28550.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3699, pruned_loss=0.1208, over 5640653.40 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3593, pruned_loss=0.1134, over 5697263.63 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3713, pruned_loss=0.1217, over 5639189.17 frames. ], batch size: 78, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:43:03,972 INFO [optim.py:369] (1/2) 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 06:43:08,146 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:43:26,761 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1230739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:43:43,340 INFO [train.py:968] (1/2) Epoch 27, batch 45900, giga_loss[loss=0.2805, simple_loss=0.356, pruned_loss=0.1025, over 28716.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3694, pruned_loss=0.1212, over 5639742.32 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3591, pruned_loss=0.1132, over 5700160.83 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3709, pruned_loss=0.1221, over 5635152.73 frames. ], batch size: 262, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:44:31,237 INFO [train.py:968] (1/2) Epoch 27, batch 45950, giga_loss[loss=0.2971, simple_loss=0.3643, pruned_loss=0.1149, over 28749.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3696, pruned_loss=0.1221, over 5586998.70 frames. ], libri_tot_loss[loss=0.2943, simple_loss=0.3602, pruned_loss=0.1142, over 5644906.25 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.37, pruned_loss=0.1221, over 5631326.11 frames. ], batch size: 243, lr: 1.16e-03, grad_scale: 4.0
+2023-03-14 06:44:45,741 INFO [optim.py:369] (1/2) 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 06:45:14,382 INFO [train.py:968] (1/2) Epoch 27, batch 46000, giga_loss[loss=0.2705, simple_loss=0.3473, pruned_loss=0.09682, over 28848.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3688, pruned_loss=0.1219, over 5559013.67 frames. ], libri_tot_loss[loss=0.295, simple_loss=0.3608, pruned_loss=0.1146, over 5602698.96 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3687, pruned_loss=0.1216, over 5630068.26 frames. ], batch size: 199, lr: 1.16e-03, grad_scale: 8.0
+2023-03-14 06:45:53,412 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-14 06:46:46,778 INFO [optim.py:369] (1/2) 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 06:46:55,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3664, 1.4877, 1.5545, 1.1828], device='cuda:1'), covar=tensor([0.1760, 0.2723, 0.1568, 0.1888], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0717, 0.0977, 0.0876], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 06:47:13,127 INFO [train.py:968] (1/2) Epoch 28, batch 50, libri_loss[loss=0.2877, simple_loss=0.3631, pruned_loss=0.1061, over 27875.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3693, pruned_loss=0.1045, over 1265762.19 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3604, pruned_loss=0.09831, over 115446.08 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3701, pruned_loss=0.105, over 1173910.21 frames. ], batch size: 116, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:47:16,868 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 06:47:59,894 INFO [train.py:968] (1/2) Epoch 28, batch 100, giga_loss[loss=0.2556, simple_loss=0.3456, pruned_loss=0.08278, over 28864.00 frames. ], tot_loss[loss=0.2818, simple_loss=0.3609, pruned_loss=0.1014, over 2241281.44 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3411, pruned_loss=0.08779, over 230610.00 frames. ], giga_tot_loss[loss=0.2842, simple_loss=0.363, pruned_loss=0.1027, over 2094471.51 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:48:09,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:48:19,900 INFO [optim.py:369] (1/2) 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 06:48:21,040 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1231022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:48:23,053 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1231025.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:48:45,279 INFO [train.py:968] (1/2) Epoch 28, batch 150, giga_loss[loss=0.2431, simple_loss=0.3116, pruned_loss=0.08731, over 28870.00 frames. ], tot_loss[loss=0.2652, simple_loss=0.3436, pruned_loss=0.09342, over 3006269.17 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08932, over 285916.57 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3442, pruned_loss=0.09391, over 2863146.52 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:48:46,257 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:48:55,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2059, 1.3273, 3.2308, 2.9246], device='cuda:1'), covar=tensor([0.1561, 0.2705, 0.0514, 0.1697], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0674, 0.1009, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 06:49:14,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 06:49:24,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:49:25,209 INFO [train.py:968] (1/2) Epoch 28, batch 200, libri_loss[loss=0.2671, simple_loss=0.3531, pruned_loss=0.09056, over 29541.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3314, pruned_loss=0.08774, over 3607951.96 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3398, pruned_loss=0.08789, over 450533.95 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3313, pruned_loss=0.08805, over 3425338.93 frames. ], batch size: 83, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:49:36,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:49:38,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-14 06:49:40,345 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.5457, 4.4353, 1.6712, 1.8951], device='cuda:1'), covar=tensor([0.1173, 0.0275, 0.0995, 0.1378], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0570, 0.0407, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 06:50:04,667 INFO [train.py:968] (1/2) Epoch 28, batch 250, giga_loss[loss=0.2053, simple_loss=0.2785, pruned_loss=0.06608, over 28590.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3241, pruned_loss=0.08457, over 4061182.31 frames. ], libri_tot_loss[loss=0.2648, simple_loss=0.3458, pruned_loss=0.09192, over 651583.29 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3222, pruned_loss=0.08394, over 3851184.53 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:50:10,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:12,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:35,053 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:42,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:45,057 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:50:47,628 INFO [train.py:968] (1/2) Epoch 28, batch 300, giga_loss[loss=0.188, simple_loss=0.2723, pruned_loss=0.05191, over 28870.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3155, pruned_loss=0.08112, over 4420260.74 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3425, pruned_loss=0.08962, over 754889.00 frames. ], giga_tot_loss[loss=0.2375, simple_loss=0.3136, pruned_loss=0.08065, over 4224835.04 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:51:03,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3142, 3.1481, 2.9694, 1.3150], device='cuda:1'), covar=tensor([0.0934, 0.1070, 0.0966, 0.2323], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.1212, 0.1020, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 06:51:04,328 INFO [optim.py:369] (1/2) 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 06:51:08,214 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:23,849 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:25,987 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:29,962 INFO [train.py:968] (1/2) Epoch 28, batch 350, libri_loss[loss=0.2754, simple_loss=0.3623, pruned_loss=0.09424, over 29127.00 frames. ], tot_loss[loss=0.2325, simple_loss=0.3092, pruned_loss=0.07794, over 4699275.61 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08781, over 927152.24 frames. ], giga_tot_loss[loss=0.2306, simple_loss=0.3064, pruned_loss=0.07738, over 4504530.06 frames. ], batch size: 101, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:51:36,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:38,659 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:51:48,796 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:52:02,667 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:52:10,040 INFO [train.py:968] (1/2) Epoch 28, batch 400, giga_loss[loss=0.1948, simple_loss=0.2656, pruned_loss=0.06199, over 28622.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.3051, pruned_loss=0.07596, over 4919736.37 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08771, over 1041295.73 frames. ], giga_tot_loss[loss=0.2263, simple_loss=0.3022, pruned_loss=0.07522, over 4747656.57 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:52:27,383 INFO [optim.py:369] (1/2) 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 06:52:42,574 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 450, giga_loss[loss=0.2116, simple_loss=0.2832, pruned_loss=0.07, over 28573.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3048, pruned_loss=0.07623, over 5097388.97 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3396, pruned_loss=0.08686, over 1230362.42 frames. ], giga_tot_loss[loss=0.226, simple_loss=0.3013, pruned_loss=0.07536, over 4927738.42 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:53:31,329 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 500, giga_loss[loss=0.2032, simple_loss=0.2734, pruned_loss=0.06648, over 28429.00 frames. ], tot_loss[loss=0.226, simple_loss=0.302, pruned_loss=0.07501, over 5227238.58 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3402, pruned_loss=0.08741, over 1345757.09 frames. ], giga_tot_loss[loss=0.223, simple_loss=0.2982, pruned_loss=0.0739, over 5075474.44 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:53:33,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231400.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:53:54,134 INFO [optim.py:369] (1/2) 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 06:54:11,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1586, 1.5092, 1.5889, 1.3248], device='cuda:1'), covar=tensor([0.2429, 0.1930, 0.2626, 0.2184], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0761, 0.0729, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 06:54:17,826 INFO [train.py:968] (1/2) Epoch 28, batch 550, giga_loss[loss=0.2097, simple_loss=0.284, pruned_loss=0.06768, over 28802.00 frames. ], tot_loss[loss=0.225, simple_loss=0.3007, pruned_loss=0.07463, over 5328036.35 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.343, pruned_loss=0.08918, over 1431031.96 frames. ], giga_tot_loss[loss=0.2211, simple_loss=0.2962, pruned_loss=0.07305, over 5201253.79 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:54:35,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7813, 2.0612, 1.7011, 1.8732], device='cuda:1'), covar=tensor([0.2693, 0.2832, 0.3244, 0.2631], device='cuda:1'), in_proj_covar=tensor([0.1603, 0.1153, 0.1415, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 06:55:02,636 INFO [train.py:968] (1/2) Epoch 28, batch 600, giga_loss[loss=0.21, simple_loss=0.2827, pruned_loss=0.06862, over 28834.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.298, pruned_loss=0.07327, over 5408920.14 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3439, pruned_loss=0.08968, over 1519323.72 frames. ], giga_tot_loss[loss=0.2182, simple_loss=0.2933, pruned_loss=0.07158, over 5299079.27 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:55:23,734 INFO [optim.py:369] (1/2) 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 06:55:40,617 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:55:44,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:55:44,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231543.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:55:46,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231546.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 06:55:48,006 INFO [train.py:968] (1/2) Epoch 28, batch 650, giga_loss[loss=0.2056, simple_loss=0.2726, pruned_loss=0.06927, over 28516.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.2957, pruned_loss=0.07207, over 5477675.09 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3422, pruned_loss=0.08847, over 1628230.62 frames. ], giga_tot_loss[loss=0.2163, simple_loss=0.2913, pruned_loss=0.07064, over 5379433.08 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:56:05,211 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:56:08,779 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231575.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 06:56:30,627 INFO [train.py:968] (1/2) Epoch 28, batch 700, giga_loss[loss=0.1943, simple_loss=0.2688, pruned_loss=0.0599, over 28989.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2947, pruned_loss=0.07155, over 5520454.95 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3415, pruned_loss=0.08776, over 1783602.83 frames. ], giga_tot_loss[loss=0.2149, simple_loss=0.2897, pruned_loss=0.07003, over 5437565.45 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:56:45,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5578, 3.5301, 1.7327, 1.6402], device='cuda:1'), covar=tensor([0.1001, 0.0327, 0.0910, 0.1397], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0568, 0.0408, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 06:56:49,100 INFO [optim.py:369] (1/2) 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 06:57:12,646 INFO [train.py:968] (1/2) Epoch 28, batch 750, giga_loss[loss=0.1714, simple_loss=0.2532, pruned_loss=0.04483, over 28980.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2935, pruned_loss=0.07096, over 5548859.49 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3415, pruned_loss=0.08784, over 1935936.66 frames. ], giga_tot_loss[loss=0.2129, simple_loss=0.2877, pruned_loss=0.06906, over 5477756.56 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 06:57:56,039 INFO [train.py:968] (1/2) Epoch 28, batch 800, giga_loss[loss=0.1878, simple_loss=0.2604, pruned_loss=0.05757, over 29006.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2916, pruned_loss=0.07055, over 5574555.08 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3415, pruned_loss=0.08762, over 2004237.19 frames. ], giga_tot_loss[loss=0.2119, simple_loss=0.2861, pruned_loss=0.06883, over 5520289.01 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:57:57,785 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1231700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:58:09,711 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.15 vs. limit=5.0
+2023-03-14 06:58:10,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1231715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 06:58:16,508 INFO [optim.py:369] (1/2) 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 06:58:42,957 INFO [train.py:968] (1/2) Epoch 28, batch 850, giga_loss[loss=0.2767, simple_loss=0.3493, pruned_loss=0.102, over 27694.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2983, pruned_loss=0.074, over 5593728.08 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3429, pruned_loss=0.08855, over 2093913.03 frames. ], giga_tot_loss[loss=0.2182, simple_loss=0.2924, pruned_loss=0.072, over 5549780.38 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:58:44,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2950, 2.9775, 1.4074, 1.4554], device='cuda:1'), covar=tensor([0.1065, 0.0343, 0.0985, 0.1435], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0568, 0.0408, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 06:59:21,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-14 06:59:28,525 INFO [train.py:968] (1/2) Epoch 28, batch 900, libri_loss[loss=0.2587, simple_loss=0.3369, pruned_loss=0.0902, over 29522.00 frames. ], tot_loss[loss=0.2352, simple_loss=0.3105, pruned_loss=0.0799, over 5619846.07 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3422, pruned_loss=0.0881, over 2206567.39 frames. ], giga_tot_loss[loss=0.2308, simple_loss=0.3052, pruned_loss=0.07819, over 5576593.89 frames. ], batch size: 84, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 06:59:47,172 INFO [optim.py:369] (1/2) 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 07:00:11,758 INFO [train.py:968] (1/2) Epoch 28, batch 950, giga_loss[loss=0.2793, simple_loss=0.3479, pruned_loss=0.1053, over 28565.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3213, pruned_loss=0.08522, over 5632780.70 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3415, pruned_loss=0.08758, over 2261794.36 frames. ], giga_tot_loss[loss=0.2425, simple_loss=0.317, pruned_loss=0.084, over 5595022.10 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:00:18,555 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1231858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:00:21,867 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1231861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:00:26,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5130, 4.3067, 4.1341, 1.6934], device='cuda:1'), covar=tensor([0.0725, 0.0851, 0.1006, 0.2251], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.1199, 0.1010, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 07:00:42,801 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1231890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:00:51,246 INFO [train.py:968] (1/2) Epoch 28, batch 1000, giga_loss[loss=0.295, simple_loss=0.3755, pruned_loss=0.1073, over 27948.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3281, pruned_loss=0.08756, over 5655993.08 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3407, pruned_loss=0.08728, over 2422195.25 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3243, pruned_loss=0.08665, over 5615080.06 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:00:55,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3157, 1.1623, 1.0540, 1.4374], device='cuda:1'), covar=tensor([0.0841, 0.0390, 0.0389, 0.0967], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 07:01:11,392 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 1050, giga_loss[loss=0.2361, simple_loss=0.3271, pruned_loss=0.07257, over 28958.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3319, pruned_loss=0.08813, over 5664542.29 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3417, pruned_loss=0.08802, over 2536602.58 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3281, pruned_loss=0.08713, over 5628709.75 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:01:42,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 07:02:17,944 INFO [train.py:968] (1/2) Epoch 28, batch 1100, giga_loss[loss=0.2725, simple_loss=0.3358, pruned_loss=0.1047, over 23656.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3342, pruned_loss=0.08865, over 5662695.23 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3418, pruned_loss=0.08829, over 2634501.37 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3309, pruned_loss=0.08776, over 5629132.88 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:02:36,055 INFO [optim.py:369] (1/2) 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 07:02:53,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2443, 0.8007, 0.8405, 1.4835], device='cuda:1'), covar=tensor([0.0780, 0.0393, 0.0389, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 07:02:59,828 INFO [train.py:968] (1/2) Epoch 28, batch 1150, giga_loss[loss=0.3354, simple_loss=0.3851, pruned_loss=0.1429, over 26588.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3371, pruned_loss=0.0904, over 5662797.98 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.342, pruned_loss=0.08882, over 2705195.29 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3343, pruned_loss=0.08948, over 5639219.20 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:03:20,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1232075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:03:43,945 INFO [train.py:968] (1/2) Epoch 28, batch 1200, giga_loss[loss=0.3021, simple_loss=0.3695, pruned_loss=0.1174, over 28606.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3398, pruned_loss=0.09267, over 5670557.48 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3422, pruned_loss=0.08905, over 2768073.20 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3374, pruned_loss=0.09189, over 5648334.19 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:04:02,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5207, 1.7258, 1.6145, 1.5507], device='cuda:1'), covar=tensor([0.2037, 0.2117, 0.2246, 0.2140], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0761, 0.0731, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 07:04:04,365 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 1250, giga_loss[loss=0.3119, simple_loss=0.3778, pruned_loss=0.123, over 28863.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3435, pruned_loss=0.09545, over 5678664.84 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.342, pruned_loss=0.08889, over 2830481.26 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3417, pruned_loss=0.095, over 5656870.51 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:04:42,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1452, 2.1621, 2.2935, 1.8440], device='cuda:1'), covar=tensor([0.1765, 0.2542, 0.1413, 0.1863], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0720, 0.0989, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 07:04:56,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5283, 4.3581, 4.1366, 2.0767], device='cuda:1'), covar=tensor([0.0538, 0.0691, 0.0717, 0.2068], device='cuda:1'), in_proj_covar=tensor([0.1284, 0.1190, 0.1001, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 07:05:08,593 INFO [train.py:968] (1/2) Epoch 28, batch 1300, giga_loss[loss=0.2715, simple_loss=0.3575, pruned_loss=0.09276, over 28875.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3464, pruned_loss=0.09588, over 5691445.29 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.342, pruned_loss=0.08889, over 2906185.45 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3451, pruned_loss=0.09568, over 5669509.32 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:05:23,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-14 07:05:25,362 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:05:28,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:05:29,626 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 1350, giga_loss[loss=0.3113, simple_loss=0.3972, pruned_loss=0.1127, over 29015.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3489, pruned_loss=0.09697, over 5687905.46 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3426, pruned_loss=0.08927, over 2936089.50 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3476, pruned_loss=0.09671, over 5668701.17 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:05:51,328 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 1400, giga_loss[loss=0.2539, simple_loss=0.3381, pruned_loss=0.08486, over 28842.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3494, pruned_loss=0.09634, over 5690361.43 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08864, over 3018468.70 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3488, pruned_loss=0.09663, over 5674167.89 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:06:50,701 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 1450, giga_loss[loss=0.2546, simple_loss=0.3376, pruned_loss=0.08577, over 28620.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3493, pruned_loss=0.09494, over 5700574.21 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08859, over 3116688.83 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3491, pruned_loss=0.09541, over 5682087.61 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:07:27,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1232367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:07:52,894 INFO [train.py:968] (1/2) Epoch 28, batch 1500, libri_loss[loss=0.2755, simple_loss=0.3648, pruned_loss=0.09308, over 29528.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3476, pruned_loss=0.09268, over 5704268.35 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3419, pruned_loss=0.08828, over 3172262.08 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3475, pruned_loss=0.09328, over 5685799.15 frames. ], batch size: 84, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:07:54,491 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1232401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:08:11,976 INFO [optim.py:369] (1/2) 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 07:08:20,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5269, 1.8294, 1.4037, 1.7664], device='cuda:1'), covar=tensor([0.0830, 0.0306, 0.0352, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 07:08:29,824 INFO [train.py:968] (1/2) Epoch 28, batch 1550, giga_loss[loss=0.2496, simple_loss=0.3337, pruned_loss=0.08278, over 28925.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3457, pruned_loss=0.09093, over 5716639.91 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3412, pruned_loss=0.08789, over 3239791.93 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3462, pruned_loss=0.09169, over 5697678.43 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:08:41,645 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 1600, giga_loss[loss=0.2753, simple_loss=0.3548, pruned_loss=0.09785, over 28600.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3475, pruned_loss=0.09329, over 5699880.06 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3413, pruned_loss=0.08809, over 3279426.46 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3479, pruned_loss=0.09386, over 5682391.68 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:09:37,791 INFO [optim.py:369] (1/2) 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 07:10:00,148 INFO [train.py:968] (1/2) Epoch 28, batch 1650, giga_loss[loss=0.3095, simple_loss=0.3674, pruned_loss=0.1258, over 28785.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.3488, pruned_loss=0.0963, over 5706642.69 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3406, pruned_loss=0.08772, over 3318757.77 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3495, pruned_loss=0.09705, over 5689909.58 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:10:05,567 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5322, 2.1372, 1.6006, 0.8267], device='cuda:1'), covar=tensor([0.7124, 0.3499, 0.4507, 0.7139], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1731, 0.1657, 0.1504], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 07:10:23,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5729, 1.7403, 1.2523, 1.3164], device='cuda:1'), covar=tensor([0.1026, 0.0573, 0.1038, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0454, 0.0525, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 07:10:40,381 INFO [train.py:968] (1/2) Epoch 28, batch 1700, giga_loss[loss=0.2741, simple_loss=0.3454, pruned_loss=0.1014, over 28648.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.35, pruned_loss=0.09844, over 5707151.23 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3409, pruned_loss=0.08765, over 3397289.90 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3507, pruned_loss=0.09939, over 5697976.52 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:11:01,715 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 1750, giga_loss[loss=0.2746, simple_loss=0.3264, pruned_loss=0.1114, over 23717.00 frames. ], tot_loss[loss=0.2745, simple_loss=0.3497, pruned_loss=0.0996, over 5696498.08 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3413, pruned_loss=0.08777, over 3430330.59 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3502, pruned_loss=0.1004, over 5690534.76 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:11:30,575 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0984, 1.8245, 2.1915, 1.7391], device='cuda:1'), covar=tensor([0.2191, 0.3475, 0.1609, 0.1919], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0717, 0.0985, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 07:11:38,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1232662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:11:51,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7526, 2.0545, 1.6941, 1.8869], device='cuda:1'), covar=tensor([0.2638, 0.2666, 0.3065, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1151, 0.1407, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 07:12:09,052 INFO [train.py:968] (1/2) Epoch 28, batch 1800, giga_loss[loss=0.3007, simple_loss=0.3603, pruned_loss=0.1205, over 28764.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09977, over 5688297.00 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.342, pruned_loss=0.08826, over 3499037.96 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.3487, pruned_loss=0.1005, over 5681671.60 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:12:11,291 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 07:12:16,985 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3403, 1.4042, 1.1735, 1.6119], device='cuda:1'), covar=tensor([0.0815, 0.0353, 0.0360, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 07:12:31,809 INFO [optim.py:369] (1/2) 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 07:12:45,979 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1232742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:12:51,189 INFO [train.py:968] (1/2) Epoch 28, batch 1850, giga_loss[loss=0.2626, simple_loss=0.3367, pruned_loss=0.09421, over 28408.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3469, pruned_loss=0.09864, over 5692788.90 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3414, pruned_loss=0.08803, over 3546751.99 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3475, pruned_loss=0.09956, over 5683577.65 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:12:53,622 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-14 07:13:15,533 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 1900, giga_loss[loss=0.3024, simple_loss=0.3685, pruned_loss=0.1181, over 28677.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3452, pruned_loss=0.09705, over 5691255.99 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08761, over 3611392.41 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3463, pruned_loss=0.09832, over 5682508.19 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:13:56,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8057, 1.4496, 4.9449, 3.6944], device='cuda:1'), covar=tensor([0.1628, 0.2789, 0.0364, 0.0791], device='cuda:1'), in_proj_covar=tensor([0.0798, 0.0670, 0.1002, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 07:13:59,599 INFO [optim.py:369] (1/2) 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 07:14:12,265 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1232836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:14:22,541 INFO [train.py:968] (1/2) Epoch 28, batch 1950, libri_loss[loss=0.2169, simple_loss=0.2983, pruned_loss=0.06778, over 29390.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3413, pruned_loss=0.09475, over 5688961.90 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3403, pruned_loss=0.08737, over 3656564.97 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3425, pruned_loss=0.09606, over 5678284.21 frames. ], batch size: 67, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:14:54,894 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:14:56,988 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:07,474 INFO [train.py:968] (1/2) Epoch 28, batch 2000, giga_loss[loss=0.2618, simple_loss=0.3326, pruned_loss=0.09549, over 28837.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3351, pruned_loss=0.09138, over 5675647.52 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3396, pruned_loss=0.08707, over 3702399.66 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3364, pruned_loss=0.09272, over 5670783.21 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:15:25,088 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1232917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:26,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:28,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:15:29,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4932, 1.7072, 1.5506, 1.3555], device='cuda:1'), covar=tensor([0.2797, 0.2784, 0.2318, 0.2913], device='cuda:1'), in_proj_covar=tensor([0.2048, 0.2007, 0.1914, 0.2056], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 07:15:30,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4189, 1.6117, 1.6664, 1.2547], device='cuda:1'), covar=tensor([0.1908, 0.2698, 0.1552, 0.1831], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0717, 0.0986, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 07:15:30,938 INFO [optim.py:369] (1/2) 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 07:15:47,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5358, 1.8662, 1.5137, 1.3394], device='cuda:1'), covar=tensor([0.2854, 0.2804, 0.3259, 0.2608], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1150, 0.1407, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 07:15:51,910 INFO [train.py:968] (1/2) Epoch 28, batch 2050, giga_loss[loss=0.239, simple_loss=0.3144, pruned_loss=0.08181, over 28781.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3305, pruned_loss=0.08858, over 5668825.13 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3402, pruned_loss=0.0872, over 3766673.56 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3309, pruned_loss=0.08967, over 5667891.91 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:15:53,363 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1232951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:20,087 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1232979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:22,240 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1232982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:38,074 INFO [train.py:968] (1/2) Epoch 28, batch 2100, giga_loss[loss=0.2309, simple_loss=0.3124, pruned_loss=0.07472, over 28701.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3282, pruned_loss=0.08798, over 5648585.66 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3403, pruned_loss=0.08733, over 3820939.09 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3282, pruned_loss=0.08881, over 5649041.27 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:16:48,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1233011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:16:58,461 INFO [optim.py:369] (1/2) 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 07:17:06,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1233037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:17:11,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1772, 1.3797, 1.3154, 1.1391], device='cuda:1'), covar=tensor([0.3565, 0.3251, 0.2205, 0.2997], device='cuda:1'), in_proj_covar=tensor([0.2047, 0.2003, 0.1913, 0.2055], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 07:17:15,779 INFO [train.py:968] (1/2) Epoch 28, batch 2150, giga_loss[loss=0.2285, simple_loss=0.3128, pruned_loss=0.07213, over 28915.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3294, pruned_loss=0.08777, over 5667097.15 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3404, pruned_loss=0.08725, over 3872259.21 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.329, pruned_loss=0.08848, over 5662227.31 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:17:37,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7653, 2.1280, 2.0556, 1.5788], device='cuda:1'), covar=tensor([0.2076, 0.2706, 0.1734, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0718, 0.0986, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 07:17:52,979 INFO [train.py:968] (1/2) Epoch 28, batch 2200, giga_loss[loss=0.2994, simple_loss=0.3642, pruned_loss=0.1173, over 28669.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3292, pruned_loss=0.08739, over 5686020.18 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3398, pruned_loss=0.08675, over 3932541.22 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.329, pruned_loss=0.08829, over 5675992.83 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:18:16,406 INFO [optim.py:369] (1/2) 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 07:18:17,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7955, 1.8691, 1.5388, 2.0684], device='cuda:1'), covar=tensor([0.2736, 0.3114, 0.3381, 0.2617], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1150, 0.1407, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 07:18:35,999 INFO [train.py:968] (1/2) Epoch 28, batch 2250, libri_loss[loss=0.2954, simple_loss=0.3763, pruned_loss=0.1073, over 29548.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3286, pruned_loss=0.08722, over 5693833.13 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3402, pruned_loss=0.08677, over 3999143.44 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3278, pruned_loss=0.08794, over 5679983.09 frames. ], batch size: 83, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:18:47,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2663, 1.5119, 1.3107, 1.3976], device='cuda:1'), covar=tensor([0.0797, 0.0411, 0.0368, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 07:18:59,862 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1233180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:19:02,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1233183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:19:18,410 INFO [train.py:968] (1/2) Epoch 28, batch 2300, giga_loss[loss=0.233, simple_loss=0.3107, pruned_loss=0.07766, over 28825.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3265, pruned_loss=0.08628, over 5700397.39 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3404, pruned_loss=0.08678, over 4016972.01 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3256, pruned_loss=0.08684, over 5688717.66 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:19:31,327 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:19:50,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1233238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:19:56,742 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 07:19:59,661 INFO [train.py:968] (1/2) Epoch 28, batch 2350, giga_loss[loss=0.2459, simple_loss=0.3319, pruned_loss=0.07997, over 28268.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3239, pruned_loss=0.08515, over 5702488.00 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3401, pruned_loss=0.08653, over 4044858.67 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3232, pruned_loss=0.08573, over 5690934.26 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:20:10,422 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1233264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:20:41,328 INFO [train.py:968] (1/2) Epoch 28, batch 2400, giga_loss[loss=0.2427, simple_loss=0.3158, pruned_loss=0.0848, over 29065.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3224, pruned_loss=0.0849, over 5700840.73 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3404, pruned_loss=0.08665, over 4063069.40 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3215, pruned_loss=0.08527, over 5690398.97 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:21:01,656 INFO [optim.py:369] (1/2) 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 07:21:18,749 INFO [train.py:968] (1/2) Epoch 28, batch 2450, giga_loss[loss=0.2223, simple_loss=0.3017, pruned_loss=0.07149, over 28626.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3208, pruned_loss=0.08434, over 5709596.26 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3407, pruned_loss=0.08653, over 4099537.09 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3196, pruned_loss=0.08466, over 5697768.48 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:21:58,207 INFO [train.py:968] (1/2) Epoch 28, batch 2500, giga_loss[loss=0.2273, simple_loss=0.2947, pruned_loss=0.07992, over 28689.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3187, pruned_loss=0.08372, over 5713368.65 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3406, pruned_loss=0.08639, over 4114334.21 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3176, pruned_loss=0.08403, over 5705130.11 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:22:18,722 INFO [optim.py:369] (1/2) 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 07:22:36,664 INFO [train.py:968] (1/2) Epoch 28, batch 2550, giga_loss[loss=0.2368, simple_loss=0.3105, pruned_loss=0.08161, over 29042.00 frames. ], tot_loss[loss=0.242, simple_loss=0.318, pruned_loss=0.08299, over 5716999.39 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3412, pruned_loss=0.08648, over 4174998.25 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.3159, pruned_loss=0.08307, over 5712327.07 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:23:14,066 INFO [train.py:968] (1/2) Epoch 28, batch 2600, giga_loss[loss=0.235, simple_loss=0.3103, pruned_loss=0.07986, over 28967.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3173, pruned_loss=0.08236, over 5726426.45 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3412, pruned_loss=0.08655, over 4274768.85 frames. ], giga_tot_loss[loss=0.2393, simple_loss=0.3143, pruned_loss=0.08214, over 5713678.82 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:23:28,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3108, 1.9454, 1.5099, 0.5898], device='cuda:1'), covar=tensor([0.6326, 0.3102, 0.5186, 0.7392], device='cuda:1'), in_proj_covar=tensor([0.1824, 0.1715, 0.1644, 0.1492], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 07:23:37,266 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 2650, giga_loss[loss=0.2245, simple_loss=0.2944, pruned_loss=0.07734, over 28687.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.3168, pruned_loss=0.08201, over 5720814.92 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3415, pruned_loss=0.08661, over 4317683.30 frames. ], giga_tot_loss[loss=0.2383, simple_loss=0.3134, pruned_loss=0.08167, over 5716680.74 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:24:31,891 INFO [train.py:968] (1/2) Epoch 28, batch 2700, giga_loss[loss=0.301, simple_loss=0.3755, pruned_loss=0.1133, over 28329.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3192, pruned_loss=0.08354, over 5721336.95 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3414, pruned_loss=0.08649, over 4359568.28 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3158, pruned_loss=0.08323, over 5717877.32 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:24:36,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1233604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:24:45,974 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1233613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:24:56,922 INFO [optim.py:369] (1/2) 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 07:25:05,694 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 2750, libri_loss[loss=0.354, simple_loss=0.4135, pruned_loss=0.1472, over 29568.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3235, pruned_loss=0.08622, over 5719164.68 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3414, pruned_loss=0.08665, over 4404705.77 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3202, pruned_loss=0.0858, over 5711807.60 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:25:18,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3049, 4.1480, 3.9951, 1.8016], device='cuda:1'), covar=tensor([0.0790, 0.0891, 0.1095, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1280, 0.1182, 0.0998, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 07:25:46,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4266, 1.6226, 1.5902, 1.3276], device='cuda:1'), covar=tensor([0.3000, 0.2623, 0.2184, 0.2739], device='cuda:1'), in_proj_covar=tensor([0.2040, 0.1996, 0.1915, 0.2054], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 07:25:57,516 INFO [train.py:968] (1/2) Epoch 28, batch 2800, giga_loss[loss=0.2477, simple_loss=0.3206, pruned_loss=0.08747, over 28523.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3284, pruned_loss=0.08904, over 5719717.50 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3413, pruned_loss=0.08657, over 4424492.77 frames. ], giga_tot_loss[loss=0.2516, simple_loss=0.3257, pruned_loss=0.08876, over 5713783.87 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:26:23,485 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 2850, giga_loss[loss=0.2617, simple_loss=0.3464, pruned_loss=0.08849, over 28669.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3371, pruned_loss=0.09519, over 5699305.90 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3414, pruned_loss=0.08657, over 4431623.31 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3348, pruned_loss=0.095, over 5694135.66 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:26:50,258 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1233756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:26:52,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1233759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:13,228 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1233782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:15,234 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1233785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:16,896 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1233788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:27,734 INFO [train.py:968] (1/2) Epoch 28, batch 2900, giga_loss[loss=0.251, simple_loss=0.3378, pruned_loss=0.08208, over 28635.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3402, pruned_loss=0.09552, over 5711947.22 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3414, pruned_loss=0.08655, over 4482952.63 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3383, pruned_loss=0.09566, over 5701160.75 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:27:43,538 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1233814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:27:55,468 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:28:12,636 INFO [train.py:968] (1/2) Epoch 28, batch 2950, giga_loss[loss=0.2633, simple_loss=0.3454, pruned_loss=0.09057, over 28770.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.345, pruned_loss=0.0973, over 5710885.70 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3412, pruned_loss=0.08646, over 4510691.53 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3436, pruned_loss=0.09765, over 5699322.04 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:28:56,900 INFO [train.py:968] (1/2) Epoch 28, batch 3000, giga_loss[loss=0.35, simple_loss=0.4215, pruned_loss=0.1392, over 28556.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3506, pruned_loss=0.1011, over 5694119.72 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3406, pruned_loss=0.08621, over 4556632.75 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3501, pruned_loss=0.102, over 5680377.21 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:28:56,900 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 07:29:04,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2161, 1.8162, 1.3873, 0.4417], device='cuda:1'), covar=tensor([0.5378, 0.3859, 0.5040, 0.6656], device='cuda:1'), in_proj_covar=tensor([0.1829, 0.1720, 0.1647, 0.1496], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 07:29:05,375 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 07:29:07,662 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7711, 3.6012, 1.6739, 1.7150], device='cuda:1'), covar=tensor([0.0926, 0.0276, 0.0906, 0.1326], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0566, 0.0407, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 07:29:30,122 INFO [optim.py:369] (1/2) 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 07:29:47,538 INFO [train.py:968] (1/2) Epoch 28, batch 3050, giga_loss[loss=0.2697, simple_loss=0.3413, pruned_loss=0.09904, over 28838.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 5695913.06 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3412, pruned_loss=0.08668, over 4581774.83 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3509, pruned_loss=0.1016, over 5682549.74 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:30:11,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1233979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:30:27,024 INFO [train.py:968] (1/2) Epoch 28, batch 3100, giga_loss[loss=0.2695, simple_loss=0.3583, pruned_loss=0.09038, over 29079.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3473, pruned_loss=0.09765, over 5703710.64 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3404, pruned_loss=0.08648, over 4627762.67 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09871, over 5686295.88 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:30:50,405 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 3150, giga_loss[loss=0.2825, simple_loss=0.3559, pruned_loss=0.1045, over 28792.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3454, pruned_loss=0.09576, over 5710929.70 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3401, pruned_loss=0.08637, over 4639942.15 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3459, pruned_loss=0.09676, over 5696033.94 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:31:51,509 INFO [train.py:968] (1/2) Epoch 28, batch 3200, giga_loss[loss=0.2481, simple_loss=0.3306, pruned_loss=0.08282, over 28674.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3452, pruned_loss=0.09529, over 5712827.79 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.34, pruned_loss=0.08628, over 4664778.98 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3458, pruned_loss=0.09632, over 5697910.06 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:31:56,616 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:32:03,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1728, 0.8499, 0.9180, 1.4789], device='cuda:1'), covar=tensor([0.0826, 0.0401, 0.0369, 0.0879], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 07:32:07,767 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:32:11,401 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:32:29,963 INFO [train.py:968] (1/2) Epoch 28, batch 3250, giga_loss[loss=0.2529, simple_loss=0.3378, pruned_loss=0.08402, over 29015.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.346, pruned_loss=0.09543, over 5717866.54 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3385, pruned_loss=0.08564, over 4707859.81 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3478, pruned_loss=0.097, over 5699971.63 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:32:35,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:32:59,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9174, 1.2820, 1.3059, 1.1310], device='cuda:1'), covar=tensor([0.2181, 0.1361, 0.2369, 0.1631], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0761, 0.0732, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 07:33:11,483 INFO [train.py:968] (1/2) Epoch 28, batch 3300, giga_loss[loss=0.2854, simple_loss=0.3522, pruned_loss=0.1093, over 28555.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3474, pruned_loss=0.09651, over 5720373.64 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3384, pruned_loss=0.08551, over 4730601.65 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3491, pruned_loss=0.09806, over 5703644.36 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:33:14,561 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:33:50,635 INFO [train.py:968] (1/2) Epoch 28, batch 3350, giga_loss[loss=0.2914, simple_loss=0.3604, pruned_loss=0.1112, over 28895.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3491, pruned_loss=0.09818, over 5715803.15 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3385, pruned_loss=0.08558, over 4769624.54 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3506, pruned_loss=0.09974, over 5697864.37 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:34:02,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3916, 1.4505, 1.4612, 1.2974], device='cuda:1'), covar=tensor([0.2743, 0.2711, 0.2411, 0.2760], device='cuda:1'), in_proj_covar=tensor([0.2043, 0.2004, 0.1918, 0.2059], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 07:34:03,965 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234265.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 07:34:26,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2089, 1.5507, 1.5308, 1.4673], device='cuda:1'), covar=tensor([0.2161, 0.1666, 0.2284, 0.1624], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0761, 0.0731, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 07:34:33,854 INFO [train.py:968] (1/2) Epoch 28, batch 3400, giga_loss[loss=0.2756, simple_loss=0.351, pruned_loss=0.1002, over 29034.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3498, pruned_loss=0.09901, over 5720847.45 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3383, pruned_loss=0.08543, over 4786982.64 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3513, pruned_loss=0.1005, over 5703670.49 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:34:35,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8584, 3.6964, 3.4969, 1.6112], device='cuda:1'), covar=tensor([0.0757, 0.0891, 0.0790, 0.2269], device='cuda:1'), in_proj_covar=tensor([0.1279, 0.1185, 0.0998, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 07:34:58,081 INFO [optim.py:369] (1/2) 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 07:35:12,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:35:14,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1234348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:35:14,424 INFO [train.py:968] (1/2) Epoch 28, batch 3450, libri_loss[loss=0.2892, simple_loss=0.371, pruned_loss=0.1037, over 29371.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3499, pruned_loss=0.09904, over 5729245.63 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3389, pruned_loss=0.08571, over 4819628.78 frames. ], giga_tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 5710956.90 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:35:36,301 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:35:44,598 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 07:35:51,550 INFO [train.py:968] (1/2) Epoch 28, batch 3500, giga_loss[loss=0.3174, simple_loss=0.3838, pruned_loss=0.1255, over 27926.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.35, pruned_loss=0.09856, over 5730150.46 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.339, pruned_loss=0.08562, over 4848071.85 frames. ], giga_tot_loss[loss=0.2756, simple_loss=0.3511, pruned_loss=0.1001, over 5714225.04 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:35:59,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5671, 1.5705, 1.6270, 1.4156], device='cuda:1'), covar=tensor([0.3081, 0.2838, 0.2588, 0.3080], device='cuda:1'), in_proj_covar=tensor([0.2044, 0.2004, 0.1918, 0.2060], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 07:36:14,750 INFO [optim.py:369] (1/2) 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 07:36:30,201 INFO [train.py:968] (1/2) Epoch 28, batch 3550, giga_loss[loss=0.2491, simple_loss=0.3376, pruned_loss=0.08035, over 28578.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3489, pruned_loss=0.09695, over 5723048.57 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3382, pruned_loss=0.0852, over 4888545.52 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3509, pruned_loss=0.09909, over 5712373.53 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:36:57,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1234481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:37:11,873 INFO [train.py:968] (1/2) Epoch 28, batch 3600, giga_loss[loss=0.3189, simple_loss=0.3961, pruned_loss=0.1208, over 28977.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3496, pruned_loss=0.09667, over 5718944.46 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3382, pruned_loss=0.08526, over 4909689.28 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3515, pruned_loss=0.09864, over 5712784.19 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:37:12,098 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8003, 0.9921, 2.8624, 2.6710], device='cuda:1'), covar=tensor([0.1825, 0.2879, 0.0610, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0667, 0.0993, 0.0971], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 07:37:35,396 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 3650, giga_loss[loss=0.2611, simple_loss=0.3339, pruned_loss=0.09422, over 28875.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3489, pruned_loss=0.09621, over 5718195.45 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3382, pruned_loss=0.08519, over 4921861.27 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3505, pruned_loss=0.09797, over 5713134.16 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:38:17,164 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-14 07:38:29,244 INFO [train.py:968] (1/2) Epoch 28, batch 3700, giga_loss[loss=0.2619, simple_loss=0.3351, pruned_loss=0.09431, over 28959.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3473, pruned_loss=0.0958, over 5722871.98 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3379, pruned_loss=0.08507, over 4947452.85 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.349, pruned_loss=0.09761, over 5717118.94 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:38:49,187 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:38:49,757 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:38:51,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1234627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:38:51,661 INFO [optim.py:369] (1/2) 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 07:38:59,902 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1234640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 07:39:01,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:39:07,472 INFO [train.py:968] (1/2) Epoch 28, batch 3750, giga_loss[loss=0.289, simple_loss=0.3533, pruned_loss=0.1124, over 27600.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3451, pruned_loss=0.09497, over 5712442.12 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.338, pruned_loss=0.08532, over 4955132.29 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3466, pruned_loss=0.09639, over 5713967.77 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:39:13,347 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:39:40,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5949, 1.7488, 1.8134, 1.3970], device='cuda:1'), covar=tensor([0.1848, 0.2592, 0.1524, 0.1800], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0718, 0.0986, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 07:39:48,930 INFO [train.py:968] (1/2) Epoch 28, batch 3800, giga_loss[loss=0.278, simple_loss=0.3551, pruned_loss=0.1005, over 29027.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3445, pruned_loss=0.09493, over 5722876.89 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.338, pruned_loss=0.0853, over 4959952.63 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3457, pruned_loss=0.09609, over 5723163.99 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:40:10,585 INFO [optim.py:369] (1/2) 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 07:40:12,181 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:40:19,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2280, 1.7153, 1.3418, 0.4703], device='cuda:1'), covar=tensor([0.4988, 0.3199, 0.4385, 0.6971], device='cuda:1'), in_proj_covar=tensor([0.1821, 0.1708, 0.1638, 0.1489], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 07:40:27,393 INFO [train.py:968] (1/2) Epoch 28, batch 3850, libri_loss[loss=0.2391, simple_loss=0.3143, pruned_loss=0.08193, over 29657.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3449, pruned_loss=0.09528, over 5723063.86 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3383, pruned_loss=0.08542, over 4988291.32 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.09643, over 5721334.98 frames. ], batch size: 69, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:40:53,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1234783.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 07:40:55,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1234786.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 07:41:06,924 INFO [train.py:968] (1/2) Epoch 28, batch 3900, giga_loss[loss=0.2558, simple_loss=0.3409, pruned_loss=0.08532, over 28922.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3455, pruned_loss=0.09532, over 5713280.29 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3383, pruned_loss=0.08546, over 5001072.39 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3465, pruned_loss=0.09645, over 5720231.64 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:41:14,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5741, 1.8479, 1.5139, 1.6880], device='cuda:1'), covar=tensor([0.2849, 0.2841, 0.3159, 0.2478], device='cuda:1'), in_proj_covar=tensor([0.1594, 0.1150, 0.1405, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 07:41:20,794 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1234815.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 07:41:30,360 INFO [optim.py:369] (1/2) 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 07:41:49,212 INFO [train.py:968] (1/2) Epoch 28, batch 3950, giga_loss[loss=0.2964, simple_loss=0.3772, pruned_loss=0.1078, over 28595.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3443, pruned_loss=0.09383, over 5701008.76 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3385, pruned_loss=0.08555, over 5002198.87 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.345, pruned_loss=0.09477, over 5714785.68 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:42:13,876 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1234879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:42:14,808 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.56 vs. limit=5.0
+2023-03-14 07:42:28,996 INFO [train.py:968] (1/2) Epoch 28, batch 4000, giga_loss[loss=0.2398, simple_loss=0.3246, pruned_loss=0.0775, over 28136.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3439, pruned_loss=0.09388, over 5708748.62 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3385, pruned_loss=0.08553, over 5020160.58 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3445, pruned_loss=0.09479, over 5716220.40 frames. ], batch size: 77, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:42:51,641 INFO [optim.py:369] (1/2) 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 07:42:58,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7936, 1.9529, 1.3969, 1.5126], device='cuda:1'), covar=tensor([0.1038, 0.0605, 0.1084, 0.1106], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0452, 0.0525, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 07:42:59,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9877, 1.3026, 1.0998, 0.2605], device='cuda:1'), covar=tensor([0.4423, 0.3391, 0.5045, 0.7332], device='cuda:1'), in_proj_covar=tensor([0.1814, 0.1699, 0.1630, 0.1484], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0004, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 07:43:08,898 INFO [train.py:968] (1/2) Epoch 28, batch 4050, giga_loss[loss=0.2978, simple_loss=0.3635, pruned_loss=0.116, over 28867.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3426, pruned_loss=0.09372, over 5703565.76 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3388, pruned_loss=0.08563, over 5031314.51 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3429, pruned_loss=0.09447, over 5708631.39 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:43:26,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2189, 1.4411, 1.3915, 1.1684], device='cuda:1'), covar=tensor([0.2556, 0.2401, 0.1728, 0.2315], device='cuda:1'), in_proj_covar=tensor([0.2037, 0.2001, 0.1909, 0.2052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 07:43:33,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6565, 1.7155, 1.3425, 1.3114], device='cuda:1'), covar=tensor([0.1010, 0.0598, 0.1056, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0451, 0.0524, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 07:43:46,629 INFO [train.py:968] (1/2) Epoch 28, batch 4100, giga_loss[loss=0.2504, simple_loss=0.3265, pruned_loss=0.0872, over 28652.00 frames. ], tot_loss[loss=0.2619, simple_loss=0.3397, pruned_loss=0.09209, over 5708421.53 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3385, pruned_loss=0.0854, over 5066422.57 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3403, pruned_loss=0.0932, over 5707565.73 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:43:47,576 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:43:48,347 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3917, 1.5894, 1.6606, 1.2310], device='cuda:1'), covar=tensor([0.1897, 0.2787, 0.1578, 0.1855], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0720, 0.0987, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 07:44:00,127 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 28, batch 4150, giga_loss[loss=0.2622, simple_loss=0.3407, pruned_loss=0.09192, over 28725.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3376, pruned_loss=0.09131, over 5710882.55 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3383, pruned_loss=0.08541, over 5086712.26 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09232, over 5706352.01 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:45:06,759 INFO [train.py:968] (1/2) Epoch 28, batch 4200, giga_loss[loss=0.3321, simple_loss=0.3861, pruned_loss=0.139, over 26648.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3375, pruned_loss=0.09234, over 5697693.11 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3384, pruned_loss=0.08555, over 5086155.07 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3379, pruned_loss=0.09306, over 5700568.34 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:45:11,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:15,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2407, 3.5780, 1.4802, 1.6765], device='cuda:1'), covar=tensor([0.1223, 0.0429, 0.1004, 0.1392], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0563, 0.0406, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 07:45:30,090 INFO [optim.py:369] (1/2) 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 07:45:42,616 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:43,883 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3400, 3.0996, 1.3886, 1.5790], device='cuda:1'), covar=tensor([0.1006, 0.0499, 0.0965, 0.1320], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0564, 0.0406, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 07:45:44,820 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:47,257 INFO [train.py:968] (1/2) Epoch 28, batch 4250, giga_loss[loss=0.2756, simple_loss=0.35, pruned_loss=0.1006, over 27847.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.336, pruned_loss=0.09178, over 5706453.03 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3386, pruned_loss=0.08568, over 5110834.56 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3361, pruned_loss=0.09245, over 5702931.40 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:45:57,378 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:45:59,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:10,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-14 07:46:11,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:15,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8853, 5.7115, 5.4150, 3.0716], device='cuda:1'), covar=tensor([0.0449, 0.0620, 0.0668, 0.1611], device='cuda:1'), in_proj_covar=tensor([0.1285, 0.1189, 0.1000, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 07:46:24,787 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:30,923 INFO [train.py:968] (1/2) Epoch 28, batch 4300, giga_loss[loss=0.2294, simple_loss=0.3072, pruned_loss=0.07582, over 28985.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3329, pruned_loss=0.09057, over 5708427.45 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3384, pruned_loss=0.08555, over 5122756.60 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3331, pruned_loss=0.09128, over 5702931.90 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:46:40,673 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1235210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:46:54,321 INFO [optim.py:369] (1/2) 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 07:47:07,956 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.70 vs. limit=5.0
+2023-03-14 07:47:09,776 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:11,661 INFO [train.py:968] (1/2) Epoch 28, batch 4350, giga_loss[loss=0.2165, simple_loss=0.2978, pruned_loss=0.0676, over 28781.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3307, pruned_loss=0.09, over 5710502.56 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3385, pruned_loss=0.08575, over 5138294.21 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3306, pruned_loss=0.09048, over 5702594.81 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:47:11,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:15,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:23,433 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-14 07:47:34,023 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:36,158 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1235281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:47:50,658 INFO [train.py:968] (1/2) Epoch 28, batch 4400, giga_loss[loss=0.2163, simple_loss=0.2996, pruned_loss=0.06652, over 29056.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3277, pruned_loss=0.0881, over 5716437.87 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3386, pruned_loss=0.08584, over 5152146.35 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3274, pruned_loss=0.08847, over 5707894.93 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:47:57,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1235306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:48:09,856 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.56 vs. limit=2.0
+2023-03-14 07:48:15,326 INFO [optim.py:369] (1/2) 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 07:48:29,683 INFO [train.py:968] (1/2) Epoch 28, batch 4450, giga_loss[loss=0.2635, simple_loss=0.3484, pruned_loss=0.08928, over 28875.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3277, pruned_loss=0.08789, over 5703798.27 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.339, pruned_loss=0.08613, over 5155057.76 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3268, pruned_loss=0.08798, over 5710399.12 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:49:07,656 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:49:08,639 INFO [train.py:968] (1/2) Epoch 28, batch 4500, giga_loss[loss=0.2665, simple_loss=0.3302, pruned_loss=0.1014, over 28335.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3314, pruned_loss=0.08911, over 5708652.22 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3395, pruned_loss=0.08632, over 5187031.66 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.33, pruned_loss=0.08914, over 5706911.37 frames. ], batch size: 77, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:49:11,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:49:18,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5248, 1.7498, 1.4791, 1.5373], device='cuda:1'), covar=tensor([0.0726, 0.0301, 0.0342, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 07:49:23,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.76 vs. limit=5.0
+2023-03-14 07:49:36,142 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:49:51,833 INFO [train.py:968] (1/2) Epoch 28, batch 4550, giga_loss[loss=0.2252, simple_loss=0.3087, pruned_loss=0.07087, over 28936.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3348, pruned_loss=0.09093, over 5706379.43 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3396, pruned_loss=0.08649, over 5211517.87 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3333, pruned_loss=0.09095, over 5698782.87 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:49:55,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4292, 2.0575, 1.4600, 0.5847], device='cuda:1'), covar=tensor([0.7579, 0.3374, 0.5055, 0.8621], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1715, 0.1648, 0.1498], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 07:50:28,929 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 4600, libri_loss[loss=0.2919, simple_loss=0.3612, pruned_loss=0.1113, over 20353.00 frames. ], tot_loss[loss=0.261, simple_loss=0.338, pruned_loss=0.092, over 5705276.89 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3397, pruned_loss=0.08654, over 5225929.02 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3366, pruned_loss=0.09215, over 5702216.16 frames. ], batch size: 189, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:50:58,835 INFO [optim.py:369] (1/2) 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 07:51:12,504 INFO [train.py:968] (1/2) Epoch 28, batch 4650, giga_loss[loss=0.3354, simple_loss=0.3849, pruned_loss=0.143, over 26593.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3399, pruned_loss=0.0926, over 5699300.41 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3397, pruned_loss=0.08658, over 5248574.26 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3387, pruned_loss=0.09285, over 5691187.64 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:51:44,031 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:51:56,333 INFO [train.py:968] (1/2) Epoch 28, batch 4700, giga_loss[loss=0.2457, simple_loss=0.3224, pruned_loss=0.08446, over 28497.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3395, pruned_loss=0.0922, over 5702627.54 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3399, pruned_loss=0.08681, over 5257490.17 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3384, pruned_loss=0.09229, over 5694263.84 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:52:21,390 INFO [optim.py:369] (1/2) 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 07:52:35,783 INFO [train.py:968] (1/2) Epoch 28, batch 4750, giga_loss[loss=0.2825, simple_loss=0.3532, pruned_loss=0.1059, over 28865.00 frames. ], tot_loss[loss=0.261, simple_loss=0.3383, pruned_loss=0.09182, over 5708954.50 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3396, pruned_loss=0.08667, over 5273756.72 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3376, pruned_loss=0.09212, over 5697788.42 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 07:52:41,603 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:03,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1235681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:17,051 INFO [train.py:968] (1/2) Epoch 28, batch 4800, giga_loss[loss=0.2375, simple_loss=0.3252, pruned_loss=0.07484, over 27919.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3392, pruned_loss=0.09302, over 5704188.81 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3396, pruned_loss=0.08675, over 5277047.94 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3388, pruned_loss=0.09321, over 5694333.25 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:53:33,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4211, 1.7730, 1.3006, 1.4675], device='cuda:1'), covar=tensor([0.1015, 0.0420, 0.1030, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0449, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 07:53:41,044 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:43,427 INFO [optim.py:369] (1/2) 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 07:53:43,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:53:45,380 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2670, 3.2364, 1.5473, 1.3787], device='cuda:1'), covar=tensor([0.1001, 0.0330, 0.0927, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0563, 0.0406, 0.0441], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 07:54:00,755 INFO [train.py:968] (1/2) Epoch 28, batch 4850, giga_loss[loss=0.3053, simple_loss=0.3783, pruned_loss=0.1162, over 27677.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3421, pruned_loss=0.09478, over 5696427.79 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3398, pruned_loss=0.0869, over 5283668.21 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3416, pruned_loss=0.09493, over 5688480.74 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:54:07,799 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:54:39,117 INFO [train.py:968] (1/2) Epoch 28, batch 4900, giga_loss[loss=0.2623, simple_loss=0.3381, pruned_loss=0.09329, over 28852.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3445, pruned_loss=0.09566, over 5704419.10 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3398, pruned_loss=0.08699, over 5291906.63 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3442, pruned_loss=0.09581, over 5696443.31 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:54:39,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3114, 1.3561, 1.2344, 1.4081], device='cuda:1'), covar=tensor([0.0753, 0.0356, 0.0353, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 07:54:39,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:54:42,856 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:54:59,127 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1235824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:01,544 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1235827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:03,827 INFO [optim.py:369] (1/2) 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 07:55:04,112 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:14,343 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 28, batch 4950, giga_loss[loss=0.2888, simple_loss=0.3537, pruned_loss=0.112, over 23881.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3463, pruned_loss=0.09605, over 5714229.72 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3398, pruned_loss=0.08707, over 5314979.60 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3463, pruned_loss=0.09647, over 5701728.33 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:55:23,882 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1235856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:55:34,031 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 5000, giga_loss[loss=0.2946, simple_loss=0.3643, pruned_loss=0.1125, over 28579.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3479, pruned_loss=0.09696, over 5715463.41 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3394, pruned_loss=0.08682, over 5326827.89 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3484, pruned_loss=0.0977, over 5703770.61 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:55:58,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4256, 1.6853, 1.3547, 1.3689], device='cuda:1'), covar=tensor([0.2698, 0.2804, 0.3231, 0.2476], device='cuda:1'), in_proj_covar=tensor([0.1592, 0.1147, 0.1402, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 07:56:20,598 INFO [optim.py:369] (1/2) 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 07:56:35,021 INFO [train.py:968] (1/2) Epoch 28, batch 5050, giga_loss[loss=0.2905, simple_loss=0.3618, pruned_loss=0.1096, over 28934.00 frames. ], tot_loss[loss=0.2711, simple_loss=0.3482, pruned_loss=0.09698, over 5723595.57 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3398, pruned_loss=0.08703, over 5337507.09 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3484, pruned_loss=0.09759, over 5711864.35 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:57:16,678 INFO [train.py:968] (1/2) Epoch 28, batch 5100, giga_loss[loss=0.2532, simple_loss=0.3297, pruned_loss=0.08834, over 28870.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3474, pruned_loss=0.09656, over 5729377.01 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.34, pruned_loss=0.08721, over 5351396.95 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.3476, pruned_loss=0.09712, over 5716133.29 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:57:29,602 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:57:32,025 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:57:56,518 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 07:57:58,158 INFO [train.py:968] (1/2) Epoch 28, batch 5150, giga_loss[loss=0.2344, simple_loss=0.3206, pruned_loss=0.07415, over 28588.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3455, pruned_loss=0.09571, over 5726092.24 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3402, pruned_loss=0.08727, over 5359803.13 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3455, pruned_loss=0.09621, over 5713194.51 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:58:29,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3761, 1.6722, 1.4565, 1.5474], device='cuda:1'), covar=tensor([0.0724, 0.0356, 0.0340, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 07:58:35,261 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2404, 0.7857, 0.8347, 1.4464], device='cuda:1'), covar=tensor([0.0776, 0.0403, 0.0405, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0119, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 07:58:40,400 INFO [train.py:968] (1/2) Epoch 28, batch 5200, giga_loss[loss=0.3143, simple_loss=0.3655, pruned_loss=0.1315, over 24045.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3426, pruned_loss=0.09437, over 5727701.55 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3405, pruned_loss=0.08725, over 5367388.90 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3425, pruned_loss=0.09489, over 5715682.28 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 07:59:01,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2936, 3.4944, 1.5311, 1.4457], device='cuda:1'), covar=tensor([0.1048, 0.0322, 0.0978, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0564, 0.0408, 0.0443], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 07:59:05,121 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 5250, giga_loss[loss=0.262, simple_loss=0.3309, pruned_loss=0.09657, over 23831.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3405, pruned_loss=0.09314, over 5723351.96 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3406, pruned_loss=0.08734, over 5382875.10 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3403, pruned_loss=0.09368, over 5712541.29 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 07:59:56,540 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-14 07:59:56,578 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.80 vs. limit=2.0
+2023-03-14 07:59:59,219 INFO [train.py:968] (1/2) Epoch 28, batch 5300, giga_loss[loss=0.2429, simple_loss=0.3363, pruned_loss=0.07471, over 28877.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3412, pruned_loss=0.09239, over 5727066.72 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3411, pruned_loss=0.08757, over 5403737.39 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3405, pruned_loss=0.09287, over 5712921.14 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:00:02,460 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236203.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:00:27,906 INFO [optim.py:369] (1/2) 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 08:00:40,984 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:00:44,081 INFO [train.py:968] (1/2) Epoch 28, batch 5350, giga_loss[loss=0.2641, simple_loss=0.3465, pruned_loss=0.09091, over 29043.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3435, pruned_loss=0.09298, over 5718276.11 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3412, pruned_loss=0.08756, over 5406254.71 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3429, pruned_loss=0.09338, over 5706435.15 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:00:58,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5375, 1.5601, 1.7193, 1.4196], device='cuda:1'), covar=tensor([0.1597, 0.2112, 0.1388, 0.1594], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0715, 0.0981, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 08:01:24,079 INFO [train.py:968] (1/2) Epoch 28, batch 5400, giga_loss[loss=0.249, simple_loss=0.3146, pruned_loss=0.09173, over 28540.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3436, pruned_loss=0.09356, over 5712110.87 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3412, pruned_loss=0.08754, over 5417828.23 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3432, pruned_loss=0.09405, over 5700582.74 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:01:48,231 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.51 vs. limit=2.0
+2023-03-14 08:01:49,038 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-14 08:01:49,087 INFO [optim.py:369] (1/2) 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 08:01:59,017 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:02:03,783 INFO [train.py:968] (1/2) Epoch 28, batch 5450, giga_loss[loss=0.2369, simple_loss=0.3079, pruned_loss=0.08294, over 28866.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3423, pruned_loss=0.0941, over 5703512.68 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3411, pruned_loss=0.08749, over 5418892.96 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3422, pruned_loss=0.09467, over 5700477.95 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:02:32,206 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.60 vs. limit=5.0
+2023-03-14 08:02:41,984 INFO [train.py:968] (1/2) Epoch 28, batch 5500, libri_loss[loss=0.2754, simple_loss=0.3531, pruned_loss=0.09879, over 29557.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3413, pruned_loss=0.09485, over 5705632.75 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3409, pruned_loss=0.08746, over 5435881.89 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3414, pruned_loss=0.09559, over 5697500.07 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:03:07,747 INFO [optim.py:369] (1/2) 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 08:03:21,254 INFO [train.py:968] (1/2) Epoch 28, batch 5550, giga_loss[loss=0.237, simple_loss=0.3114, pruned_loss=0.08131, over 28557.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3389, pruned_loss=0.0945, over 5707497.98 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3406, pruned_loss=0.08746, over 5446184.25 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3393, pruned_loss=0.09529, over 5698010.48 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:04:03,526 INFO [train.py:968] (1/2) Epoch 28, batch 5600, libri_loss[loss=0.2361, simple_loss=0.3324, pruned_loss=0.06994, over 29479.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3366, pruned_loss=0.09369, over 5710691.33 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3405, pruned_loss=0.08739, over 5450670.16 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3369, pruned_loss=0.09445, over 5701487.76 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:04:16,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3379, 1.3949, 1.2873, 1.4445], device='cuda:1'), covar=tensor([0.0756, 0.0373, 0.0367, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0192, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0074, 0.0066, 0.0114], device='cuda:1')
+2023-03-14 08:04:32,711 INFO [optim.py:369] (1/2) 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 08:04:48,676 INFO [train.py:968] (1/2) Epoch 28, batch 5650, giga_loss[loss=0.2385, simple_loss=0.3025, pruned_loss=0.08726, over 28485.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3343, pruned_loss=0.09243, over 5708106.15 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3406, pruned_loss=0.08745, over 5444363.81 frames. ], giga_tot_loss[loss=0.2602, simple_loss=0.3344, pruned_loss=0.09299, over 5706856.85 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:04:55,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4955, 2.2145, 1.7912, 0.7793], device='cuda:1'), covar=tensor([0.6995, 0.3314, 0.5057, 0.7948], device='cuda:1'), in_proj_covar=tensor([0.1826, 0.1710, 0.1644, 0.1494], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 08:05:12,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236578.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:05:15,598 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:05:27,473 INFO [train.py:968] (1/2) Epoch 28, batch 5700, libri_loss[loss=0.2765, simple_loss=0.3561, pruned_loss=0.09841, over 26126.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3298, pruned_loss=0.09008, over 5716174.01 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3406, pruned_loss=0.08751, over 5452219.54 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3296, pruned_loss=0.09056, over 5714322.11 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:05:45,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:05:54,191 INFO [optim.py:369] (1/2) 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 08:06:07,869 INFO [train.py:968] (1/2) Epoch 28, batch 5750, giga_loss[loss=0.2268, simple_loss=0.3135, pruned_loss=0.07011, over 28948.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3265, pruned_loss=0.08818, over 5719996.55 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.341, pruned_loss=0.08769, over 5461136.31 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3258, pruned_loss=0.08847, over 5716444.10 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:06:14,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0836, 1.5006, 0.9457, 1.1119], device='cuda:1'), covar=tensor([0.1243, 0.0617, 0.1529, 0.1361], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0450, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 08:06:24,239 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1236670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:06:46,439 INFO [train.py:968] (1/2) Epoch 28, batch 5800, giga_loss[loss=0.2442, simple_loss=0.3274, pruned_loss=0.08049, over 28722.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3276, pruned_loss=0.08847, over 5711704.01 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3411, pruned_loss=0.08775, over 5461230.02 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3267, pruned_loss=0.08865, over 5714793.97 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:07:01,131 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:07:03,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236721.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:07:05,909 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1236724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:07:13,191 INFO [optim.py:369] (1/2) 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 08:07:27,570 INFO [train.py:968] (1/2) Epoch 28, batch 5850, giga_loss[loss=0.2686, simple_loss=0.358, pruned_loss=0.08962, over 28762.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3313, pruned_loss=0.09004, over 5715157.77 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3409, pruned_loss=0.08761, over 5466467.20 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3305, pruned_loss=0.0903, over 5716249.83 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:07:30,291 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236753.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:07:37,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:07:40,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1236766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:08:04,926 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:08:07,231 INFO [train.py:968] (1/2) Epoch 28, batch 5900, giga_loss[loss=0.2929, simple_loss=0.3656, pruned_loss=0.1101, over 28714.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3342, pruned_loss=0.09127, over 5721569.95 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3411, pruned_loss=0.08784, over 5480347.34 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3332, pruned_loss=0.09139, over 5717474.88 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:08:34,951 INFO [optim.py:369] (1/2) 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 08:08:48,560 INFO [train.py:968] (1/2) Epoch 28, batch 5950, giga_loss[loss=0.2851, simple_loss=0.366, pruned_loss=0.1022, over 28415.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3371, pruned_loss=0.09199, over 5716728.75 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3412, pruned_loss=0.08796, over 5490710.07 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.336, pruned_loss=0.09207, over 5709329.67 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:08:49,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5734, 1.7858, 1.2979, 1.4003], device='cuda:1'), covar=tensor([0.1060, 0.0720, 0.1042, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0451, 0.0523, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 08:08:58,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1236861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:09:01,390 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1236864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:09:26,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1236893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:09:30,180 INFO [train.py:968] (1/2) Epoch 28, batch 6000, giga_loss[loss=0.2888, simple_loss=0.3708, pruned_loss=0.1034, over 28309.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3391, pruned_loss=0.09277, over 5711481.88 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3409, pruned_loss=0.08778, over 5489526.58 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3385, pruned_loss=0.09307, over 5710902.26 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:09:30,181 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 08:09:38,311 INFO [train.py:1012] (1/2) Epoch 28, validation: loss=0.2075, simple_loss=0.3151, pruned_loss=0.04996, over 944034.00 frames. 
+2023-03-14 08:09:38,311 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 08:10:09,787 INFO [optim.py:369] (1/2) 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 08:10:24,669 INFO [train.py:968] (1/2) Epoch 28, batch 6050, giga_loss[loss=0.2534, simple_loss=0.327, pruned_loss=0.0899, over 28546.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3429, pruned_loss=0.09602, over 5706530.73 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08768, over 5492891.60 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3426, pruned_loss=0.09637, over 5705120.82 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:10:33,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1236958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:10:58,495 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.94 vs. limit=5.0
+2023-03-14 08:11:10,147 INFO [train.py:968] (1/2) Epoch 28, batch 6100, giga_loss[loss=0.3376, simple_loss=0.3924, pruned_loss=0.1414, over 28773.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3494, pruned_loss=0.1014, over 5704336.87 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3408, pruned_loss=0.08776, over 5497262.81 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3492, pruned_loss=0.1018, over 5702029.93 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:11:40,781 INFO [optim.py:369] (1/2) 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 08:11:53,317 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1237045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:11:56,132 INFO [train.py:968] (1/2) Epoch 28, batch 6150, giga_loss[loss=0.3613, simple_loss=0.4161, pruned_loss=0.1533, over 27596.00 frames. ], tot_loss[loss=0.2856, simple_loss=0.3565, pruned_loss=0.1074, over 5681639.76 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08757, over 5509466.31 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.357, pruned_loss=0.1083, over 5673996.75 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:12:31,142 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7992, 2.0042, 1.3884, 1.6245], device='cuda:1'), covar=tensor([0.0984, 0.0651, 0.1028, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0451, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 08:12:42,695 INFO [train.py:968] (1/2) Epoch 28, batch 6200, giga_loss[loss=0.4314, simple_loss=0.4537, pruned_loss=0.2046, over 27825.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3637, pruned_loss=0.1123, over 5687305.65 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3404, pruned_loss=0.08751, over 5518989.81 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3646, pruned_loss=0.1136, over 5676622.26 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:12:45,516 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1237101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:12:47,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1237104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:13:13,559 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1237133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:13:14,089 INFO [optim.py:369] (1/2) 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 08:13:21,468 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5487, 1.7865, 1.4393, 1.6814], device='cuda:1'), covar=tensor([0.2517, 0.2652, 0.2976, 0.2295], device='cuda:1'), in_proj_covar=tensor([0.1596, 0.1150, 0.1406, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 08:13:26,462 INFO [train.py:968] (1/2) Epoch 28, batch 6250, giga_loss[loss=0.2972, simple_loss=0.3721, pruned_loss=0.1112, over 28964.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3676, pruned_loss=0.1159, over 5687240.30 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08777, over 5533939.45 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.369, pruned_loss=0.1177, over 5671005.20 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:14:03,487 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1237188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:14:06,669 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1237191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:14:14,668 INFO [train.py:968] (1/2) Epoch 28, batch 6300, giga_loss[loss=0.2965, simple_loss=0.3649, pruned_loss=0.114, over 28703.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3732, pruned_loss=0.1202, over 5688311.13 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3406, pruned_loss=0.08772, over 5537711.71 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3745, pruned_loss=0.1219, over 5673761.64 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:14:18,666 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4473, 4.3121, 4.1051, 2.1687], device='cuda:1'), covar=tensor([0.0610, 0.0735, 0.0795, 0.1869], device='cuda:1'), in_proj_covar=tensor([0.1296, 0.1195, 0.1007, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 08:14:37,418 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1237220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:14:39,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6054, 1.8790, 1.4836, 1.8268], device='cuda:1'), covar=tensor([0.2519, 0.2619, 0.2893, 0.2519], device='cuda:1'), in_proj_covar=tensor([0.1598, 0.1151, 0.1409, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 08:14:44,813 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6285, 1.7988, 1.3422, 1.3382], device='cuda:1'), covar=tensor([0.1056, 0.0616, 0.1024, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0450, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 08:14:51,574 INFO [optim.py:369] (1/2) 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 08:14:53,994 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6475, 1.8570, 1.5262, 1.5997], device='cuda:1'), covar=tensor([0.2640, 0.2774, 0.3182, 0.2478], device='cuda:1'), in_proj_covar=tensor([0.1598, 0.1151, 0.1409, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 08:15:08,447 INFO [train.py:968] (1/2) Epoch 28, batch 6350, giga_loss[loss=0.334, simple_loss=0.3914, pruned_loss=0.1383, over 29054.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3773, pruned_loss=0.1246, over 5663818.96 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08768, over 5540523.12 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3787, pruned_loss=0.1262, over 5651205.16 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:15:51,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6286, 1.9846, 1.4362, 1.5891], device='cuda:1'), covar=tensor([0.1018, 0.0527, 0.0961, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0451, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 08:15:57,207 INFO [train.py:968] (1/2) Epoch 28, batch 6400, giga_loss[loss=0.2735, simple_loss=0.349, pruned_loss=0.09907, over 28957.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.379, pruned_loss=0.1269, over 5642655.37 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08767, over 5539867.00 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.381, pruned_loss=0.1292, over 5634841.88 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:16:31,860 INFO [optim.py:369] (1/2) 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 08:16:49,320 INFO [train.py:968] (1/2) Epoch 28, batch 6450, giga_loss[loss=0.3568, simple_loss=0.4088, pruned_loss=0.1524, over 28963.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.383, pruned_loss=0.1315, over 5624285.99 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.341, pruned_loss=0.08806, over 5548868.92 frames. ], giga_tot_loss[loss=0.3267, simple_loss=0.3853, pruned_loss=0.1341, over 5612465.33 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:17:39,202 INFO [train.py:968] (1/2) Epoch 28, batch 6500, giga_loss[loss=0.3319, simple_loss=0.4037, pruned_loss=0.13, over 28965.00 frames. ], tot_loss[loss=0.3277, simple_loss=0.3864, pruned_loss=0.1344, over 5616313.16 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3408, pruned_loss=0.08789, over 5549873.36 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3897, pruned_loss=0.1379, over 5607152.00 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:17:47,508 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1237407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:18:14,357 INFO [optim.py:369] (1/2) 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 08:18:26,602 INFO [train.py:968] (1/2) Epoch 28, batch 6550, giga_loss[loss=0.3395, simple_loss=0.395, pruned_loss=0.142, over 28775.00 frames. ], tot_loss[loss=0.3292, simple_loss=0.3876, pruned_loss=0.1354, over 5625307.81 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3408, pruned_loss=0.08784, over 5553935.15 frames. ], giga_tot_loss[loss=0.3346, simple_loss=0.391, pruned_loss=0.1391, over 5615500.91 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:19:15,864 INFO [train.py:968] (1/2) Epoch 28, batch 6600, giga_loss[loss=0.359, simple_loss=0.3928, pruned_loss=0.1626, over 23730.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3855, pruned_loss=0.1342, over 5640814.22 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3411, pruned_loss=0.08787, over 5560993.36 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3891, pruned_loss=0.1381, over 5628511.89 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:19:41,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.67 vs. limit=5.0
+2023-03-14 08:19:51,147 INFO [optim.py:369] (1/2) 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 08:20:03,007 INFO [train.py:968] (1/2) Epoch 28, batch 6650, giga_loss[loss=0.3592, simple_loss=0.4049, pruned_loss=0.1567, over 27552.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3845, pruned_loss=0.1337, over 5641004.90 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3409, pruned_loss=0.08765, over 5570545.69 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3888, pruned_loss=0.1383, over 5624844.52 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:20:13,331 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1237560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:20:46,585 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 6700, giga_loss[loss=0.3018, simple_loss=0.3743, pruned_loss=0.1147, over 28495.00 frames. ], tot_loss[loss=0.3245, simple_loss=0.3842, pruned_loss=0.1324, over 5649427.93 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3411, pruned_loss=0.08804, over 5583176.65 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.389, pruned_loss=0.1374, over 5627711.80 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:20:54,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3729, 2.8836, 2.7549, 2.0515], device='cuda:1'), covar=tensor([0.2984, 0.1820, 0.1950, 0.2632], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2031, 0.1938, 0.2072], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 08:21:17,890 INFO [optim.py:369] (1/2) 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 08:21:24,218 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5226, 1.5981, 1.7175, 1.3137], device='cuda:1'), covar=tensor([0.1628, 0.2679, 0.1377, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.0925, 0.0711, 0.0974, 0.0872], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 08:21:32,670 INFO [train.py:968] (1/2) Epoch 28, batch 6750, giga_loss[loss=0.3282, simple_loss=0.3957, pruned_loss=0.1303, over 29049.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3835, pruned_loss=0.1307, over 5649572.26 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3412, pruned_loss=0.08789, over 5596361.46 frames. ], giga_tot_loss[loss=0.3313, simple_loss=0.3891, pruned_loss=0.1367, over 5622849.72 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:22:19,915 INFO [train.py:968] (1/2) Epoch 28, batch 6800, giga_loss[loss=0.3127, simple_loss=0.38, pruned_loss=0.1227, over 28891.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3828, pruned_loss=0.13, over 5629195.94 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08766, over 5596660.00 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3891, pruned_loss=0.1364, over 5609565.72 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:22:58,769 INFO [optim.py:369] (1/2) 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 08:23:13,848 INFO [train.py:968] (1/2) Epoch 28, batch 6850, giga_loss[loss=0.2841, simple_loss=0.3647, pruned_loss=0.1018, over 28695.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3801, pruned_loss=0.1276, over 5624742.96 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3409, pruned_loss=0.08783, over 5600289.46 frames. ], giga_tot_loss[loss=0.3258, simple_loss=0.3855, pruned_loss=0.1331, over 5606295.55 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:23:47,062 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 6900, giga_loss[loss=0.3017, simple_loss=0.3744, pruned_loss=0.1145, over 28867.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3776, pruned_loss=0.1242, over 5638333.09 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.08813, over 5605029.48 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3828, pruned_loss=0.1294, over 5619690.20 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:24:37,547 INFO [optim.py:369] (1/2) 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 08:24:48,334 INFO [train.py:968] (1/2) Epoch 28, batch 6950, libri_loss[loss=0.2861, simple_loss=0.3643, pruned_loss=0.104, over 29511.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3743, pruned_loss=0.1212, over 5648261.01 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.0881, over 5607228.71 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.379, pruned_loss=0.126, over 5631932.55 frames. ], batch size: 89, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:25:34,734 INFO [train.py:968] (1/2) Epoch 28, batch 7000, giga_loss[loss=0.3345, simple_loss=0.3958, pruned_loss=0.1366, over 28599.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3722, pruned_loss=0.1199, over 5652698.31 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.341, pruned_loss=0.08809, over 5613129.77 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3767, pruned_loss=0.1243, over 5635051.79 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:25:58,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-14 08:25:59,677 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1237928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:09,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1237935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:11,237 INFO [optim.py:369] (1/2) 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 08:26:22,557 INFO [train.py:968] (1/2) Epoch 28, batch 7050, giga_loss[loss=0.3158, simple_loss=0.3772, pruned_loss=0.1272, over 29004.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.3721, pruned_loss=0.1206, over 5651694.52 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.341, pruned_loss=0.08817, over 5612004.02 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3761, pruned_loss=0.1245, over 5639422.88 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:26:30,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1237957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:44,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1237972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:26:53,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.74 vs. limit=5.0
+2023-03-14 08:27:00,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8143, 2.7867, 1.7798, 1.0199], device='cuda:1'), covar=tensor([0.8921, 0.3642, 0.4190, 0.8057], device='cuda:1'), in_proj_covar=tensor([0.1840, 0.1729, 0.1653, 0.1501], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 08:27:11,891 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 08:27:12,145 INFO [train.py:968] (1/2) Epoch 28, batch 7100, libri_loss[loss=0.2123, simple_loss=0.2953, pruned_loss=0.06467, over 29473.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3715, pruned_loss=0.12, over 5663100.94 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08806, over 5619041.86 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3757, pruned_loss=0.124, over 5647693.41 frames. ], batch size: 70, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:27:12,527 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1237999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:27:37,714 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:27:42,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-14 08:27:49,830 INFO [optim.py:369] (1/2) 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 08:28:00,859 INFO [train.py:968] (1/2) Epoch 28, batch 7150, libri_loss[loss=0.2249, simple_loss=0.3104, pruned_loss=0.06966, over 29572.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3691, pruned_loss=0.1178, over 5671121.03 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08809, over 5629280.68 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3738, pruned_loss=0.1223, over 5650835.72 frames. ], batch size: 75, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:28:29,661 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:28:33,206 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:28:53,159 INFO [train.py:968] (1/2) Epoch 28, batch 7200, giga_loss[loss=0.2882, simple_loss=0.3704, pruned_loss=0.103, over 28875.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3687, pruned_loss=0.1161, over 5675291.95 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3406, pruned_loss=0.08814, over 5633425.95 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3729, pruned_loss=0.12, over 5656060.47 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:28:54,097 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:28:57,538 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:05,838 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:11,411 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:14,172 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:29:29,930 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5341, 1.9408, 1.8670, 1.6893], device='cuda:1'), covar=tensor([0.2404, 0.2161, 0.2440, 0.2448], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0762, 0.0732, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 08:29:32,652 INFO [optim.py:369] (1/2) 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 08:29:41,516 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 7250, giga_loss[loss=0.3251, simple_loss=0.3892, pruned_loss=0.1306, over 27522.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3682, pruned_loss=0.1137, over 5677178.26 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3402, pruned_loss=0.08792, over 5640669.97 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3726, pruned_loss=0.1176, over 5656206.20 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:30:08,981 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:30:25,306 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3420, 1.5421, 1.3832, 1.2430], device='cuda:1'), covar=tensor([0.2882, 0.2604, 0.2202, 0.2506], device='cuda:1'), in_proj_covar=tensor([0.2051, 0.2019, 0.1927, 0.2059], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 08:30:37,136 INFO [train.py:968] (1/2) Epoch 28, batch 7300, giga_loss[loss=0.2819, simple_loss=0.3527, pruned_loss=0.1056, over 28902.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3688, pruned_loss=0.1141, over 5675150.49 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3402, pruned_loss=0.08792, over 5642129.62 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3723, pruned_loss=0.1173, over 5657616.46 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:31:13,241 INFO [optim.py:369] (1/2) 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 08:31:23,969 INFO [train.py:968] (1/2) Epoch 28, batch 7350, giga_loss[loss=0.317, simple_loss=0.3641, pruned_loss=0.135, over 23577.00 frames. ], tot_loss[loss=0.298, simple_loss=0.368, pruned_loss=0.114, over 5673662.29 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3404, pruned_loss=0.08799, over 5641697.24 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3716, pruned_loss=0.1174, over 5661307.05 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:31:59,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3591, 1.4505, 1.3244, 1.3374], device='cuda:1'), covar=tensor([0.2300, 0.2338, 0.2359, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.2057, 0.2026, 0.1932, 0.2067], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 08:32:01,404 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2988, 1.5923, 1.5147, 1.4262], device='cuda:1'), covar=tensor([0.1942, 0.1820, 0.2233, 0.1890], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0763, 0.0733, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 08:32:06,527 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:32:10,951 INFO [train.py:968] (1/2) Epoch 28, batch 7400, giga_loss[loss=0.2939, simple_loss=0.3645, pruned_loss=0.1117, over 28942.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3666, pruned_loss=0.1139, over 5670329.13 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3403, pruned_loss=0.08804, over 5644301.20 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3704, pruned_loss=0.1174, over 5658982.69 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:32:14,070 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:32:18,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2424, 2.3202, 2.3597, 1.8837], device='cuda:1'), covar=tensor([0.3100, 0.2571, 0.2263, 0.2884], device='cuda:1'), in_proj_covar=tensor([0.2057, 0.2026, 0.1932, 0.2068], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 08:32:20,329 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 08:32:30,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-14 08:32:38,962 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3515, 2.7848, 1.4629, 1.4549], device='cuda:1'), covar=tensor([0.0940, 0.0393, 0.0859, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0572, 0.0409, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 08:32:38,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1394, 1.3534, 1.2035, 1.0634], device='cuda:1'), covar=tensor([0.2492, 0.2409, 0.1987, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.2058, 0.2028, 0.1934, 0.2069], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 08:32:46,626 INFO [optim.py:369] (1/2) 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 08:32:55,146 INFO [train.py:968] (1/2) Epoch 28, batch 7450, giga_loss[loss=0.266, simple_loss=0.3384, pruned_loss=0.09683, over 28933.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3643, pruned_loss=0.1134, over 5676660.06 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3399, pruned_loss=0.08779, over 5653046.48 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3687, pruned_loss=0.1173, over 5660168.03 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:33:00,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 08:33:17,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:33:42,553 INFO [train.py:968] (1/2) Epoch 28, batch 7500, giga_loss[loss=0.2875, simple_loss=0.3648, pruned_loss=0.1051, over 28800.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3645, pruned_loss=0.1138, over 5681466.22 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3394, pruned_loss=0.08751, over 5655697.61 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3685, pruned_loss=0.1174, over 5666451.12 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:33:42,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:34:09,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8424, 3.6372, 3.4422, 1.6707], device='cuda:1'), covar=tensor([0.0755, 0.0906, 0.0893, 0.2175], device='cuda:1'), in_proj_covar=tensor([0.1307, 0.1207, 0.1015, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 08:34:21,697 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 7550, giga_loss[loss=0.2961, simple_loss=0.3791, pruned_loss=0.1065, over 29036.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3657, pruned_loss=0.1134, over 5694525.98 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3395, pruned_loss=0.08755, over 5658208.26 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3691, pruned_loss=0.1164, over 5680697.75 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:34:57,914 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:35:00,763 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 7600, giga_loss[loss=0.2878, simple_loss=0.36, pruned_loss=0.1078, over 28730.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.367, pruned_loss=0.1138, over 5698441.87 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3395, pruned_loss=0.08754, over 5660935.05 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.37, pruned_loss=0.1165, over 5685568.96 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:35:33,311 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4851, 2.6009, 1.5866, 1.6282], device='cuda:1'), covar=tensor([0.0824, 0.0349, 0.0701, 0.1107], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0574, 0.0410, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 08:35:38,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:35:40,737 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:35:56,669 INFO [optim.py:369] (1/2) 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 08:35:59,661 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:02,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:05,518 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:05,867 INFO [train.py:968] (1/2) Epoch 28, batch 7650, giga_loss[loss=0.3329, simple_loss=0.3848, pruned_loss=0.1404, over 27531.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3658, pruned_loss=0.1133, over 5700135.94 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3396, pruned_loss=0.08761, over 5663728.44 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3684, pruned_loss=0.1156, over 5687780.06 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:36:06,582 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:32,269 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:36:55,028 INFO [train.py:968] (1/2) Epoch 28, batch 7700, giga_loss[loss=0.2751, simple_loss=0.3299, pruned_loss=0.1102, over 23784.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3644, pruned_loss=0.1131, over 5698481.32 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3398, pruned_loss=0.08768, over 5669716.70 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3669, pruned_loss=0.1155, over 5683860.44 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:37:15,369 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:18,350 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:18,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:20,473 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:32,447 INFO [optim.py:369] (1/2) 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 08:37:41,752 INFO [train.py:968] (1/2) Epoch 28, batch 7750, libri_loss[loss=0.2749, simple_loss=0.36, pruned_loss=0.09493, over 29764.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3645, pruned_loss=0.1136, over 5697669.38 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3397, pruned_loss=0.08758, over 5677156.95 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3677, pruned_loss=0.1167, over 5679744.47 frames. ], batch size: 87, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:37:43,645 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:37:46,067 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:03,179 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:06,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1238674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:21,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:23,252 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:38:27,690 INFO [train.py:968] (1/2) Epoch 28, batch 7800, giga_loss[loss=0.2972, simple_loss=0.3648, pruned_loss=0.1148, over 28658.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3631, pruned_loss=0.113, over 5698636.95 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3399, pruned_loss=0.08769, over 5674463.69 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.366, pruned_loss=0.1161, over 5686442.54 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:38:49,238 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:39:05,053 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 7850, giga_loss[loss=0.2775, simple_loss=0.3489, pruned_loss=0.103, over 28972.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3608, pruned_loss=0.1118, over 5693818.46 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3394, pruned_loss=0.08741, over 5671242.14 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3643, pruned_loss=0.1153, over 5688427.36 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:39:15,998 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238752.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:39:16,611 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0049, 2.3524, 1.7619, 2.1166], device='cuda:1'), covar=tensor([0.0994, 0.0618, 0.0969, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0453, 0.0524, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 08:39:58,717 INFO [train.py:968] (1/2) Epoch 28, batch 7900, giga_loss[loss=0.2922, simple_loss=0.3575, pruned_loss=0.1135, over 28888.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3609, pruned_loss=0.1129, over 5700150.90 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3392, pruned_loss=0.08737, over 5677564.39 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3644, pruned_loss=0.1163, over 5690951.82 frames. ], batch size: 285, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:40:08,621 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:10,610 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1238816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:11,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1238817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:14,188 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:40:35,636 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:40:39,720 INFO [train.py:968] (1/2) Epoch 28, batch 7950, giga_loss[loss=0.2731, simple_loss=0.3498, pruned_loss=0.0982, over 28723.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3604, pruned_loss=0.1122, over 5702083.42 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3397, pruned_loss=0.08749, over 5680912.49 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3635, pruned_loss=0.1158, over 5692469.63 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:40:39,945 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1238849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:41:22,363 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3806, 3.2303, 1.4891, 1.5855], device='cuda:1'), covar=tensor([0.0991, 0.0358, 0.0886, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0572, 0.0409, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 08:41:29,331 INFO [train.py:968] (1/2) Epoch 28, batch 8000, libri_loss[loss=0.2109, simple_loss=0.2983, pruned_loss=0.06176, over 29497.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3613, pruned_loss=0.1129, over 5689758.90 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3394, pruned_loss=0.08727, over 5683867.47 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3644, pruned_loss=0.1163, over 5679449.87 frames. ], batch size: 70, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:41:38,223 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1238909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:41:53,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5970, 1.7923, 1.4738, 1.6839], device='cuda:1'), covar=tensor([0.2820, 0.2825, 0.3249, 0.2423], device='cuda:1'), in_proj_covar=tensor([0.1595, 0.1151, 0.1408, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 08:42:04,815 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 8050, libri_loss[loss=0.2484, simple_loss=0.3272, pruned_loss=0.08482, over 29666.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3619, pruned_loss=0.1124, over 5687205.55 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3394, pruned_loss=0.08722, over 5687327.62 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.365, pruned_loss=0.1159, over 5675724.70 frames. ], batch size: 73, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:42:34,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-14 08:42:59,800 INFO [train.py:968] (1/2) Epoch 28, batch 8100, giga_loss[loss=0.3725, simple_loss=0.4078, pruned_loss=0.1686, over 26674.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3633, pruned_loss=0.1131, over 5671727.22 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3399, pruned_loss=0.0876, over 5682854.41 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3658, pruned_loss=0.1161, over 5666633.95 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:43:24,311 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 28, batch 8150, giga_loss[loss=0.2495, simple_loss=0.3267, pruned_loss=0.08619, over 28481.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3629, pruned_loss=0.1123, over 5678911.64 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3404, pruned_loss=0.08778, over 5685296.73 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3654, pruned_loss=0.1156, over 5672089.36 frames. ], batch size: 60, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:44:34,722 INFO [train.py:968] (1/2) Epoch 28, batch 8200, giga_loss[loss=0.2889, simple_loss=0.3616, pruned_loss=0.1081, over 28963.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3656, pruned_loss=0.1148, over 5673098.76 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08775, over 5679963.23 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3679, pruned_loss=0.118, over 5671641.92 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:44:53,364 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:45:01,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239127.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:45:11,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5056, 1.2422, 4.4681, 3.4817], device='cuda:1'), covar=tensor([0.1676, 0.2870, 0.0448, 0.1082], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0677, 0.1012, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 08:45:15,385 INFO [optim.py:369] (1/2) 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 08:45:21,264 INFO [train.py:968] (1/2) Epoch 28, batch 8250, giga_loss[loss=0.3222, simple_loss=0.3779, pruned_loss=0.1332, over 28802.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3665, pruned_loss=0.1166, over 5673079.34 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08812, over 5674300.01 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3688, pruned_loss=0.1195, over 5677464.17 frames. ], batch size: 243, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:46:05,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2551, 1.6301, 1.0221, 1.2372], device='cuda:1'), covar=tensor([0.1432, 0.0827, 0.1622, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0452, 0.0524, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 08:46:10,887 INFO [train.py:968] (1/2) Epoch 28, batch 8300, giga_loss[loss=0.3454, simple_loss=0.3991, pruned_loss=0.1459, over 28592.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3685, pruned_loss=0.1196, over 5654587.30 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.341, pruned_loss=0.08837, over 5667853.03 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3707, pruned_loss=0.1223, over 5664278.37 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:46:12,326 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:46:58,880 INFO [train.py:968] (1/2) Epoch 28, batch 8350, giga_loss[loss=0.3484, simple_loss=0.3952, pruned_loss=0.1508, over 24052.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1201, over 5646024.40 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3412, pruned_loss=0.08853, over 5660858.77 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3706, pruned_loss=0.1228, over 5659914.61 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 08:47:20,712 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:47:21,429 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239270.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:47:23,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239273.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 08:47:33,076 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 8400, giga_loss[loss=0.315, simple_loss=0.3698, pruned_loss=0.1301, over 28704.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3677, pruned_loss=0.1199, over 5640913.77 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3414, pruned_loss=0.08879, over 5653624.70 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3697, pruned_loss=0.1223, over 5658492.82 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:47:48,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239302.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 08:48:17,884 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 8450, giga_loss[loss=0.3046, simple_loss=0.3783, pruned_loss=0.1154, over 28888.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3686, pruned_loss=0.1199, over 5656415.11 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3416, pruned_loss=0.08924, over 5661237.26 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3707, pruned_loss=0.1222, over 5663613.68 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:48:31,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 08:49:09,087 INFO [train.py:968] (1/2) Epoch 28, batch 8500, libri_loss[loss=0.2796, simple_loss=0.3579, pruned_loss=0.1006, over 25674.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3675, pruned_loss=0.118, over 5655237.41 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3418, pruned_loss=0.0892, over 5665483.39 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3701, pruned_loss=0.1211, over 5657551.74 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:49:13,202 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:49:17,878 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7951, 1.3230, 5.1904, 3.6376], device='cuda:1'), covar=tensor([0.1708, 0.2960, 0.0400, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0678, 0.1014, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 08:49:33,160 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:49:35,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:49:42,396 INFO [optim.py:369] (1/2) 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 08:49:50,849 INFO [train.py:968] (1/2) Epoch 28, batch 8550, giga_loss[loss=0.2952, simple_loss=0.3584, pruned_loss=0.116, over 28573.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3656, pruned_loss=0.1166, over 5663843.95 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3422, pruned_loss=0.08939, over 5665281.91 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3677, pruned_loss=0.1193, over 5665560.40 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:50:00,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3907, 3.4062, 2.3160, 1.4493], device='cuda:1'), covar=tensor([0.7659, 0.3316, 0.3814, 0.7129], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1743, 0.1659, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 08:50:01,292 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:50:30,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239493.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:50:34,732 INFO [train.py:968] (1/2) Epoch 28, batch 8600, giga_loss[loss=0.2569, simple_loss=0.3304, pruned_loss=0.09172, over 29011.00 frames. ], tot_loss[loss=0.2962, simple_loss=0.3627, pruned_loss=0.1149, over 5674705.13 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3421, pruned_loss=0.08935, over 5672705.26 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3655, pruned_loss=0.1183, over 5669619.48 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:51:14,246 INFO [optim.py:369] (1/2) 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 08:51:22,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:51:24,000 INFO [train.py:968] (1/2) Epoch 28, batch 8650, giga_loss[loss=0.2712, simple_loss=0.3421, pruned_loss=0.1002, over 28906.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3623, pruned_loss=0.1154, over 5666872.12 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3421, pruned_loss=0.08928, over 5676615.97 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3649, pruned_loss=0.1187, over 5659451.61 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:51:24,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:51:53,505 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1239575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:51:55,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:52:14,738 INFO [train.py:968] (1/2) Epoch 28, batch 8700, giga_loss[loss=0.3156, simple_loss=0.3892, pruned_loss=0.121, over 28781.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3636, pruned_loss=0.1161, over 5652474.94 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3421, pruned_loss=0.08936, over 5669745.22 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3659, pruned_loss=0.119, over 5652750.61 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:52:49,856 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:52:53,165 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:52:53,477 INFO [optim.py:369] (1/2) 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 08:52:59,097 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 8750, giga_loss[loss=0.308, simple_loss=0.3852, pruned_loss=0.1154, over 28946.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.366, pruned_loss=0.1153, over 5662175.07 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3422, pruned_loss=0.08941, over 5677327.61 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3684, pruned_loss=0.1183, over 5654992.02 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:53:13,388 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-14 08:53:20,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:53:40,509 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6136, 1.8498, 1.4642, 1.6212], device='cuda:1'), covar=tensor([0.0767, 0.0317, 0.0338, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 08:53:48,519 INFO [train.py:968] (1/2) Epoch 28, batch 8800, giga_loss[loss=0.3026, simple_loss=0.3777, pruned_loss=0.1138, over 28962.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.367, pruned_loss=0.1139, over 5676661.69 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3417, pruned_loss=0.08909, over 5682234.69 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.37, pruned_loss=0.1173, over 5666321.55 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 08:54:07,466 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:54:09,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:54:27,328 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 8850, giga_loss[loss=0.3455, simple_loss=0.4113, pruned_loss=0.1398, over 28911.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3694, pruned_loss=0.1158, over 5675209.09 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3413, pruned_loss=0.08885, over 5684647.46 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3724, pruned_loss=0.1189, over 5664843.13 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:54:34,577 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:05,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1239787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:09,795 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1239790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:17,416 INFO [train.py:968] (1/2) Epoch 28, batch 8900, giga_loss[loss=0.2831, simple_loss=0.3607, pruned_loss=0.1028, over 28982.00 frames. ], tot_loss[loss=0.303, simple_loss=0.371, pruned_loss=0.1175, over 5668474.12 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3418, pruned_loss=0.08916, over 5691499.78 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3741, pruned_loss=0.1208, over 5653475.84 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:55:34,949 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1239819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:55:53,523 INFO [optim.py:369] (1/2) 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 08:55:58,743 INFO [train.py:968] (1/2) Epoch 28, batch 8950, libri_loss[loss=0.2082, simple_loss=0.2926, pruned_loss=0.06191, over 28469.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3702, pruned_loss=0.1174, over 5674633.98 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.342, pruned_loss=0.08925, over 5698466.08 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3736, pruned_loss=0.121, over 5655203.50 frames. ], batch size: 63, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:56:07,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1239859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 08:56:27,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6404, 2.4523, 1.7911, 0.7395], device='cuda:1'), covar=tensor([0.6817, 0.3637, 0.5073, 0.7753], device='cuda:1'), in_proj_covar=tensor([0.1846, 0.1746, 0.1661, 0.1508], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 08:56:45,585 INFO [train.py:968] (1/2) Epoch 28, batch 9000, giga_loss[loss=0.3949, simple_loss=0.4223, pruned_loss=0.1838, over 23405.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.37, pruned_loss=0.1189, over 5657674.61 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3421, pruned_loss=0.08933, over 5703211.21 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3734, pruned_loss=0.1224, over 5637511.74 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:56:45,585 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 08:56:52,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3492, 1.6876, 1.6211, 1.1664], device='cuda:1'), covar=tensor([0.1945, 0.3072, 0.1679, 0.2231], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0717, 0.0980, 0.0878], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 08:56:53,879 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 08:57:33,254 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 9050, giga_loss[loss=0.3158, simple_loss=0.3612, pruned_loss=0.1352, over 23437.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3677, pruned_loss=0.1179, over 5662438.41 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3419, pruned_loss=0.08919, over 5707519.27 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.371, pruned_loss=0.1213, over 5642085.11 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:57:47,918 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5605, 1.8334, 1.4655, 1.7453], device='cuda:1'), covar=tensor([0.2544, 0.2619, 0.2916, 0.2404], device='cuda:1'), in_proj_covar=tensor([0.1593, 0.1150, 0.1408, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 08:58:24,788 INFO [train.py:968] (1/2) Epoch 28, batch 9100, giga_loss[loss=0.3229, simple_loss=0.3841, pruned_loss=0.1309, over 28884.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3659, pruned_loss=0.1172, over 5662839.94 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3416, pruned_loss=0.08898, over 5704106.12 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3695, pruned_loss=0.121, over 5647984.51 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 08:59:06,475 INFO [optim.py:369] (1/2) 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 08:59:15,963 INFO [train.py:968] (1/2) Epoch 28, batch 9150, giga_loss[loss=0.4118, simple_loss=0.4409, pruned_loss=0.1914, over 26509.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3683, pruned_loss=0.12, over 5659594.37 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3417, pruned_loss=0.08902, over 5706091.36 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3716, pruned_loss=0.1235, over 5645358.66 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:00:01,613 INFO [train.py:968] (1/2) Epoch 28, batch 9200, giga_loss[loss=0.2716, simple_loss=0.3327, pruned_loss=0.1052, over 29031.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3681, pruned_loss=0.1201, over 5657610.69 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3418, pruned_loss=0.0891, over 5710084.23 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3711, pruned_loss=0.1234, over 5641704.75 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:00:27,456 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 28, batch 9250, giga_loss[loss=0.2703, simple_loss=0.346, pruned_loss=0.0973, over 28899.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3655, pruned_loss=0.1189, over 5658273.63 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.342, pruned_loss=0.08921, over 5703968.44 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3683, pruned_loss=0.122, over 5649241.64 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:01:33,847 INFO [train.py:968] (1/2) Epoch 28, batch 9300, giga_loss[loss=0.2961, simple_loss=0.3649, pruned_loss=0.1137, over 29024.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3658, pruned_loss=0.119, over 5653888.48 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3419, pruned_loss=0.08906, over 5707644.67 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3685, pruned_loss=0.1222, over 5642716.87 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:01:49,410 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.63 vs. limit=5.0
+2023-03-14 09:02:00,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2636, 4.0846, 3.8638, 1.9406], device='cuda:1'), covar=tensor([0.0839, 0.1004, 0.1136, 0.2185], device='cuda:1'), in_proj_covar=tensor([0.1316, 0.1215, 0.1025, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 09:02:04,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:02:12,248 INFO [optim.py:369] (1/2) 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 09:02:14,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5250, 3.4761, 1.5439, 1.8244], device='cuda:1'), covar=tensor([0.0967, 0.0376, 0.0900, 0.1210], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0572, 0.0409, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 09:02:18,943 INFO [train.py:968] (1/2) Epoch 28, batch 9350, giga_loss[loss=0.3425, simple_loss=0.4031, pruned_loss=0.141, over 28514.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3661, pruned_loss=0.1175, over 5658446.18 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3421, pruned_loss=0.08904, over 5705134.20 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3689, pruned_loss=0.1212, over 5649809.00 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:02:37,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:02:39,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:02:44,756 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3261, 1.4153, 1.3241, 1.4787], device='cuda:1'), covar=tensor([0.0768, 0.0365, 0.0341, 0.0875], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 09:03:02,365 INFO [train.py:968] (1/2) Epoch 28, batch 9400, giga_loss[loss=0.3092, simple_loss=0.3694, pruned_loss=0.1245, over 28896.00 frames. ], tot_loss[loss=0.3028, simple_loss=0.368, pruned_loss=0.1188, over 5665141.75 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3423, pruned_loss=0.08917, over 5709767.07 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.122, over 5653476.44 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:03:06,897 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:03:30,225 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:03:43,110 INFO [optim.py:369] (1/2) 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 09:03:50,293 INFO [train.py:968] (1/2) Epoch 28, batch 9450, giga_loss[loss=0.3743, simple_loss=0.4082, pruned_loss=0.1702, over 26816.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3688, pruned_loss=0.1202, over 5659976.28 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08913, over 5712188.43 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3715, pruned_loss=0.1236, over 5647165.95 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:04:15,995 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:04:18,354 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240380.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:04:38,652 INFO [train.py:968] (1/2) Epoch 28, batch 9500, giga_loss[loss=0.3149, simple_loss=0.3748, pruned_loss=0.1275, over 27565.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3702, pruned_loss=0.1185, over 5662920.81 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08916, over 5714307.93 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3726, pruned_loss=0.1214, over 5650731.25 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:04:46,700 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:05:01,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6361, 1.8785, 1.4593, 2.0109], device='cuda:1'), covar=tensor([0.2804, 0.2913, 0.3383, 0.2541], device='cuda:1'), in_proj_covar=tensor([0.1595, 0.1150, 0.1410, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 09:05:08,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240431.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 09:05:14,563 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-14 09:05:15,363 INFO [optim.py:369] (1/2) 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 09:05:22,993 INFO [train.py:968] (1/2) Epoch 28, batch 9550, giga_loss[loss=0.2769, simple_loss=0.3653, pruned_loss=0.09423, over 28630.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3709, pruned_loss=0.117, over 5671503.82 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3417, pruned_loss=0.08888, over 5718054.89 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3737, pruned_loss=0.12, over 5657826.67 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:06:07,442 INFO [train.py:968] (1/2) Epoch 28, batch 9600, giga_loss[loss=0.304, simple_loss=0.3869, pruned_loss=0.1106, over 29057.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3738, pruned_loss=0.1177, over 5670755.89 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3421, pruned_loss=0.08903, over 5711294.02 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3762, pruned_loss=0.1205, over 5665600.58 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:06:21,779 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:06:53,550 INFO [train.py:968] (1/2) Epoch 28, batch 9650, giga_loss[loss=0.4648, simple_loss=0.4655, pruned_loss=0.2321, over 26539.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3762, pruned_loss=0.1206, over 5670954.58 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.342, pruned_loss=0.08903, over 5712516.65 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3789, pruned_loss=0.1235, over 5664676.72 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:07:00,144 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0076, 5.8014, 5.5245, 3.4259], device='cuda:1'), covar=tensor([0.0604, 0.0738, 0.1034, 0.1433], device='cuda:1'), in_proj_covar=tensor([0.1319, 0.1219, 0.1026, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 09:07:38,342 INFO [train.py:968] (1/2) Epoch 28, batch 9700, giga_loss[loss=0.3281, simple_loss=0.392, pruned_loss=0.1321, over 28906.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3789, pruned_loss=0.1238, over 5656969.58 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3424, pruned_loss=0.08921, over 5706484.39 frames. ], giga_tot_loss[loss=0.3175, simple_loss=0.3816, pruned_loss=0.1267, over 5657115.41 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:08:19,309 INFO [optim.py:369] (1/2) 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 09:08:24,849 INFO [train.py:968] (1/2) Epoch 28, batch 9750, giga_loss[loss=0.2854, simple_loss=0.3598, pruned_loss=0.1055, over 28282.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3785, pruned_loss=0.124, over 5659716.52 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3427, pruned_loss=0.08953, over 5710415.98 frames. ], giga_tot_loss[loss=0.3172, simple_loss=0.381, pruned_loss=0.1267, over 5655373.34 frames. ], batch size: 369, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:09:07,147 INFO [train.py:968] (1/2) Epoch 28, batch 9800, giga_loss[loss=0.2692, simple_loss=0.3618, pruned_loss=0.08829, over 28962.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3759, pruned_loss=0.1215, over 5662373.47 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3429, pruned_loss=0.08967, over 5707417.59 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3787, pruned_loss=0.1244, over 5660431.30 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:09:10,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:09:41,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5874, 1.6587, 1.7575, 1.3390], device='cuda:1'), covar=tensor([0.2066, 0.2719, 0.1745, 0.1972], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0719, 0.0982, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 09:09:46,383 INFO [optim.py:369] (1/2) 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 09:09:50,899 INFO [train.py:968] (1/2) Epoch 28, batch 9850, giga_loss[loss=0.2987, simple_loss=0.378, pruned_loss=0.1097, over 29025.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3753, pruned_loss=0.1191, over 5669478.46 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.343, pruned_loss=0.08966, over 5709473.70 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3777, pruned_loss=0.1217, over 5665807.02 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:10:37,241 INFO [train.py:968] (1/2) Epoch 28, batch 9900, giga_loss[loss=0.2935, simple_loss=0.3652, pruned_loss=0.1109, over 28874.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3738, pruned_loss=0.1174, over 5667268.86 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3427, pruned_loss=0.08964, over 5705281.42 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3766, pruned_loss=0.1201, over 5666804.82 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:10:45,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240806.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 09:10:58,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8334, 4.8375, 1.9626, 2.1727], device='cuda:1'), covar=tensor([0.0947, 0.0236, 0.0858, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0574, 0.0410, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 09:11:15,829 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1240839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:11:18,110 INFO [optim.py:369] (1/2) 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 09:11:20,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:11:22,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:11:23,068 INFO [train.py:968] (1/2) Epoch 28, batch 9950, libri_loss[loss=0.2321, simple_loss=0.3107, pruned_loss=0.07674, over 29335.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3751, pruned_loss=0.1188, over 5670083.35 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3424, pruned_loss=0.08948, over 5712279.39 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3791, pruned_loss=0.1224, over 5661438.78 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:11:50,121 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:12:02,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1240889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:12:11,713 INFO [train.py:968] (1/2) Epoch 28, batch 10000, giga_loss[loss=0.3382, simple_loss=0.3889, pruned_loss=0.1437, over 27614.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3738, pruned_loss=0.1184, over 5657157.96 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.08946, over 5706949.06 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3778, pruned_loss=0.1219, over 5654448.66 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:12:13,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3880, 1.6459, 1.3612, 1.4398], device='cuda:1'), covar=tensor([0.0710, 0.0402, 0.0346, 0.0803], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 09:12:37,436 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:12:56,566 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 09:12:59,843 INFO [train.py:968] (1/2) Epoch 28, batch 10050, giga_loss[loss=0.2985, simple_loss=0.3657, pruned_loss=0.1157, over 28909.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3715, pruned_loss=0.1179, over 5654945.75 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.342, pruned_loss=0.0892, over 5708091.80 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3757, pruned_loss=0.1217, over 5650579.17 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:13:00,404 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1240949.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 09:13:04,100 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1240952.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 09:13:34,845 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1240981.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 09:13:48,382 INFO [train.py:968] (1/2) Epoch 28, batch 10100, libri_loss[loss=0.2559, simple_loss=0.3393, pruned_loss=0.08628, over 29531.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3707, pruned_loss=0.1185, over 5660396.81 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08915, over 5710257.95 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3745, pruned_loss=0.1221, over 5653888.55 frames. ], batch size: 80, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:14:18,596 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:14:19,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3261, 3.1539, 3.0152, 1.4232], device='cuda:1'), covar=tensor([0.1049, 0.1112, 0.1031, 0.2242], device='cuda:1'), in_proj_covar=tensor([0.1325, 0.1223, 0.1032, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 09:14:20,556 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:14:28,509 INFO [optim.py:369] (1/2) 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 09:14:32,232 INFO [train.py:968] (1/2) Epoch 28, batch 10150, giga_loss[loss=0.2751, simple_loss=0.3497, pruned_loss=0.1003, over 28829.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3672, pruned_loss=0.1167, over 5662037.57 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3415, pruned_loss=0.08886, over 5715673.27 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.372, pruned_loss=0.1211, over 5649653.11 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:14:46,168 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:15:11,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2280, 1.5558, 1.5033, 1.3734], device='cuda:1'), covar=tensor([0.2125, 0.1723, 0.2435, 0.1920], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0762, 0.0730, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 09:15:21,263 INFO [train.py:968] (1/2) Epoch 28, batch 10200, giga_loss[loss=0.3391, simple_loss=0.3817, pruned_loss=0.1483, over 28508.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3676, pruned_loss=0.1179, over 5655604.53 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3418, pruned_loss=0.08903, over 5708904.91 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3718, pruned_loss=0.122, over 5650228.97 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:16:03,532 INFO [optim.py:369] (1/2) 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 09:16:06,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6499, 1.9396, 1.5292, 1.7425], device='cuda:1'), covar=tensor([0.2647, 0.2761, 0.3111, 0.2541], device='cuda:1'), in_proj_covar=tensor([0.1599, 0.1153, 0.1412, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 09:16:07,106 INFO [train.py:968] (1/2) Epoch 28, batch 10250, giga_loss[loss=0.2783, simple_loss=0.3477, pruned_loss=0.1044, over 28874.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3666, pruned_loss=0.1172, over 5661112.15 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3417, pruned_loss=0.0889, over 5714373.82 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3707, pruned_loss=0.1213, over 5650774.41 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:16:52,803 INFO [train.py:968] (1/2) Epoch 28, batch 10300, giga_loss[loss=0.2449, simple_loss=0.3316, pruned_loss=0.07912, over 28945.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3644, pruned_loss=0.1143, over 5669854.46 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3414, pruned_loss=0.08867, over 5715432.58 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3683, pruned_loss=0.1182, over 5659755.26 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:17:07,197 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:17:27,005 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-14 09:17:36,204 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 10350, giga_loss[loss=0.2552, simple_loss=0.3309, pruned_loss=0.08982, over 28688.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.3607, pruned_loss=0.1111, over 5653082.96 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3419, pruned_loss=0.08912, over 5708031.42 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3639, pruned_loss=0.1144, over 5649967.46 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:18:05,462 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:18:25,870 INFO [train.py:968] (1/2) Epoch 28, batch 10400, libri_loss[loss=0.2839, simple_loss=0.3745, pruned_loss=0.09666, over 29525.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3609, pruned_loss=0.1106, over 5661351.43 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.342, pruned_loss=0.08905, over 5709731.91 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3638, pruned_loss=0.114, over 5656196.19 frames. ], batch size: 82, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:18:29,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:19:06,479 INFO [optim.py:369] (1/2) 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 09:19:10,597 INFO [train.py:968] (1/2) Epoch 28, batch 10450, giga_loss[loss=0.2658, simple_loss=0.3357, pruned_loss=0.09793, over 28879.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3592, pruned_loss=0.1101, over 5657789.07 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3423, pruned_loss=0.08934, over 5706846.95 frames. ], giga_tot_loss[loss=0.2944, simple_loss=0.362, pruned_loss=0.1134, over 5654819.97 frames. ], batch size: 112, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:19:20,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:19:23,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:19:50,482 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 10500, giga_loss[loss=0.2573, simple_loss=0.33, pruned_loss=0.09228, over 28628.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3566, pruned_loss=0.1097, over 5660695.12 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3423, pruned_loss=0.0893, over 5710761.75 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3592, pruned_loss=0.1127, over 5653864.53 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:20:09,604 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3606, 2.0117, 1.5308, 0.5649], device='cuda:1'), covar=tensor([0.6042, 0.3196, 0.4745, 0.7300], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1736, 0.1655, 0.1501], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 09:20:30,207 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:20:44,762 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:20:45,131 INFO [optim.py:369] (1/2) 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 09:20:46,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:20:47,305 INFO [train.py:968] (1/2) Epoch 28, batch 10550, giga_loss[loss=0.308, simple_loss=0.3739, pruned_loss=0.1211, over 27564.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3589, pruned_loss=0.1111, over 5667043.85 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08946, over 5713526.50 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.361, pruned_loss=0.1138, over 5658180.72 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:21:04,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4052, 1.8117, 1.2404, 0.8594], device='cuda:1'), covar=tensor([0.5766, 0.2967, 0.3502, 0.6654], device='cuda:1'), in_proj_covar=tensor([0.1842, 0.1734, 0.1654, 0.1501], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 09:21:12,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1291, 1.1211, 5.4493, 3.7972], device='cuda:1'), covar=tensor([0.1532, 0.3064, 0.0395, 0.0943], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0677, 0.1013, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 09:21:13,927 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:21:32,952 INFO [train.py:968] (1/2) Epoch 28, batch 10600, giga_loss[loss=0.4162, simple_loss=0.4415, pruned_loss=0.1955, over 26599.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.3611, pruned_loss=0.1121, over 5662822.57 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3423, pruned_loss=0.0892, over 5715787.58 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3634, pruned_loss=0.1149, over 5652768.37 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:21:42,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0253, 2.6185, 1.8110, 2.2172], device='cuda:1'), covar=tensor([0.0962, 0.0496, 0.0896, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0410, 0.0450, 0.0519, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 09:22:17,610 INFO [optim.py:369] (1/2) 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 09:22:19,557 INFO [train.py:968] (1/2) Epoch 28, batch 10650, giga_loss[loss=0.3197, simple_loss=0.3658, pruned_loss=0.1369, over 23424.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.3626, pruned_loss=0.1133, over 5656511.15 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3433, pruned_loss=0.08984, over 5720211.62 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.364, pruned_loss=0.1155, over 5643576.75 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:22:20,360 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:22:56,719 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 09:23:00,300 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-14 09:23:02,835 INFO [train.py:968] (1/2) Epoch 28, batch 10700, giga_loss[loss=0.2809, simple_loss=0.3572, pruned_loss=0.1023, over 28879.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3622, pruned_loss=0.113, over 5662672.10 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08978, over 5722596.77 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.364, pruned_loss=0.1158, over 5647780.37 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:23:31,034 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.43 vs. limit=5.0
+2023-03-14 09:23:45,456 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 10750, giga_loss[loss=0.3151, simple_loss=0.3834, pruned_loss=0.1234, over 28685.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3634, pruned_loss=0.1142, over 5666209.21 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3438, pruned_loss=0.09009, over 5725334.17 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3649, pruned_loss=0.1167, over 5650386.40 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:23:50,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:24:15,215 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-14 09:24:38,858 INFO [train.py:968] (1/2) Epoch 28, batch 10800, giga_loss[loss=0.2991, simple_loss=0.3687, pruned_loss=0.1148, over 28895.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3665, pruned_loss=0.1163, over 5662652.14 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09017, over 5727809.30 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3679, pruned_loss=0.1186, over 5646832.19 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:25:21,354 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 10850, giga_loss[loss=0.2747, simple_loss=0.3474, pruned_loss=0.101, over 28969.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3681, pruned_loss=0.1174, over 5671733.55 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09017, over 5730817.66 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3698, pruned_loss=0.1198, over 5655525.90 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:25:31,772 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 09:25:37,859 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-14 09:26:02,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:05,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:08,488 INFO [train.py:968] (1/2) Epoch 28, batch 10900, giga_loss[loss=0.3071, simple_loss=0.3634, pruned_loss=0.1254, over 27508.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3685, pruned_loss=0.1178, over 5671285.25 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.08995, over 5722204.20 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3709, pruned_loss=0.1206, over 5663867.88 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:26:11,838 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1241800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:16,310 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:30,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-14 09:26:34,047 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1241825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:26:53,829 INFO [optim.py:369] (1/2) 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 09:26:57,352 INFO [train.py:968] (1/2) Epoch 28, batch 10950, giga_loss[loss=0.3087, simple_loss=0.3789, pruned_loss=0.1192, over 28803.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3692, pruned_loss=0.1183, over 5675334.14 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3429, pruned_loss=0.08965, over 5727126.74 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3721, pruned_loss=0.1215, over 5663755.67 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:27:35,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4503, 1.6192, 1.2385, 1.2573], device='cuda:1'), covar=tensor([0.1028, 0.0539, 0.1050, 0.1078], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0453, 0.0522, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 09:27:47,609 INFO [train.py:968] (1/2) Epoch 28, batch 11000, giga_loss[loss=0.2883, simple_loss=0.3633, pruned_loss=0.1067, over 28909.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3686, pruned_loss=0.1164, over 5667608.48 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3429, pruned_loss=0.08966, over 5728519.64 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3713, pruned_loss=0.1193, over 5656446.41 frames. ], batch size: 186, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:28:13,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1241925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:28:35,108 INFO [optim.py:369] (1/2) 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 09:28:39,086 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1241948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:28:39,365 INFO [train.py:968] (1/2) Epoch 28, batch 11050, giga_loss[loss=0.4393, simple_loss=0.4561, pruned_loss=0.2113, over 26394.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3698, pruned_loss=0.1181, over 5661373.22 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.343, pruned_loss=0.08972, over 5731536.52 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3723, pruned_loss=0.1207, over 5649024.67 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:28:43,204 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1241951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:29:09,507 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 11100, libri_loss[loss=0.2496, simple_loss=0.3367, pruned_loss=0.08126, over 29523.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3692, pruned_loss=0.1182, over 5648408.86 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3434, pruned_loss=0.08995, over 5735726.91 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3716, pruned_loss=0.121, over 5632343.37 frames. ], batch size: 84, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:30:10,293 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-14 09:30:17,665 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 11150, giga_loss[loss=0.2978, simple_loss=0.3652, pruned_loss=0.1152, over 28953.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3673, pruned_loss=0.1171, over 5660761.88 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.08969, over 5742598.14 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3704, pruned_loss=0.1206, over 5638608.15 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:30:35,138 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:30:37,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:30:53,427 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9321, 1.1383, 1.1224, 0.8855], device='cuda:1'), covar=tensor([0.2650, 0.3017, 0.1870, 0.2547], device='cuda:1'), in_proj_covar=tensor([0.2072, 0.2038, 0.1950, 0.2085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 09:31:02,448 INFO [train.py:968] (1/2) Epoch 28, batch 11200, giga_loss[loss=0.3153, simple_loss=0.3754, pruned_loss=0.1276, over 28880.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3657, pruned_loss=0.1164, over 5661552.95 frames. ], libri_tot_loss[loss=0.2612, simple_loss=0.3431, pruned_loss=0.08964, over 5743964.01 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.12, over 5640536.87 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:31:03,356 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1242100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:31:20,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-14 09:31:45,998 INFO [optim.py:369] (1/2) 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 09:31:47,328 INFO [train.py:968] (1/2) Epoch 28, batch 11250, libri_loss[loss=0.2764, simple_loss=0.3658, pruned_loss=0.09349, over 29484.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.366, pruned_loss=0.1171, over 5657197.50 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08964, over 5737709.66 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3688, pruned_loss=0.1206, over 5643256.53 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:32:14,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1242175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:32:14,951 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 09:32:24,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1242187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:32:36,958 INFO [train.py:968] (1/2) Epoch 28, batch 11300, giga_loss[loss=0.3768, simple_loss=0.4066, pruned_loss=0.1735, over 26476.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3661, pruned_loss=0.1179, over 5655255.58 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3433, pruned_loss=0.08961, over 5738790.11 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3684, pruned_loss=0.1208, over 5642945.18 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:33:22,689 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 11350, giga_loss[loss=0.2957, simple_loss=0.3614, pruned_loss=0.115, over 28533.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3659, pruned_loss=0.1175, over 5665962.57 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3436, pruned_loss=0.08984, over 5743827.74 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3681, pruned_loss=0.1203, over 5649311.73 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:33:41,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5922, 1.6564, 1.7761, 1.3575], device='cuda:1'), covar=tensor([0.1713, 0.2673, 0.1460, 0.1717], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0721, 0.0983, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 09:33:59,055 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5770, 1.6371, 1.7531, 1.3416], device='cuda:1'), covar=tensor([0.1753, 0.2622, 0.1462, 0.1742], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0721, 0.0983, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 09:34:10,303 INFO [train.py:968] (1/2) Epoch 28, batch 11400, giga_loss[loss=0.3438, simple_loss=0.4005, pruned_loss=0.1435, over 28678.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1198, over 5663927.07 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09006, over 5745906.23 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3702, pruned_loss=0.1225, over 5646903.21 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:34:28,003 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:34:31,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:34:56,058 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 11450, giga_loss[loss=0.3244, simple_loss=0.3876, pruned_loss=0.1306, over 29030.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3678, pruned_loss=0.1196, over 5649273.51 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3436, pruned_loss=0.08997, over 5737389.92 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3703, pruned_loss=0.1226, over 5640553.53 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:34:59,331 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1242350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:35:24,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1463, 3.9945, 3.7930, 1.8848], device='cuda:1'), covar=tensor([0.0686, 0.0767, 0.0822, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.1318, 0.1215, 0.1026, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 09:35:50,041 INFO [train.py:968] (1/2) Epoch 28, batch 11500, giga_loss[loss=0.311, simple_loss=0.3778, pruned_loss=0.1221, over 28780.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3694, pruned_loss=0.1217, over 5649386.68 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3434, pruned_loss=0.0899, over 5738752.29 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3717, pruned_loss=0.1245, over 5640463.35 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:35:55,850 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 09:35:58,643 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1242407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:36:00,353 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 09:36:34,545 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 11550, giga_loss[loss=0.2783, simple_loss=0.3499, pruned_loss=0.1033, over 28936.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3681, pruned_loss=0.1206, over 5658064.94 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3436, pruned_loss=0.09006, over 5741591.32 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3702, pruned_loss=0.123, over 5647444.48 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:37:25,133 INFO [train.py:968] (1/2) Epoch 28, batch 11600, giga_loss[loss=0.3212, simple_loss=0.3873, pruned_loss=0.1276, over 28938.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3705, pruned_loss=0.1222, over 5655585.43 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3438, pruned_loss=0.09016, over 5744036.74 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3724, pruned_loss=0.1246, over 5643556.66 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:38:11,768 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 11650, giga_loss[loss=0.27, simple_loss=0.3511, pruned_loss=0.09445, over 29022.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3704, pruned_loss=0.121, over 5673138.97 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09016, over 5746835.12 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3723, pruned_loss=0.1235, over 5659612.18 frames. ], batch size: 136, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:38:23,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4096, 1.4954, 1.6135, 1.2205], device='cuda:1'), covar=tensor([0.1610, 0.2652, 0.1365, 0.1635], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0722, 0.0984, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 09:38:25,181 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 11700, giga_loss[loss=0.3102, simple_loss=0.3822, pruned_loss=0.1191, over 28968.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3717, pruned_loss=0.1226, over 5662381.75 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09013, over 5751366.81 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.374, pruned_loss=0.1254, over 5645449.97 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:39:35,127 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-14 09:39:40,414 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9722, 1.4941, 5.4902, 3.9404], device='cuda:1'), covar=tensor([0.1677, 0.2928, 0.0434, 0.0801], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0677, 0.1013, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 09:39:49,913 INFO [train.py:968] (1/2) Epoch 28, batch 11750, giga_loss[loss=0.2759, simple_loss=0.3399, pruned_loss=0.106, over 28442.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3734, pruned_loss=0.1239, over 5661517.94 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09014, over 5751320.09 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3755, pruned_loss=0.1267, over 5646830.86 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:39:51,552 INFO [optim.py:369] (1/2) 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 09:39:56,408 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-14 09:40:35,934 INFO [train.py:968] (1/2) Epoch 28, batch 11800, giga_loss[loss=0.3226, simple_loss=0.3842, pruned_loss=0.1305, over 29072.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3713, pruned_loss=0.1227, over 5664932.12 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3434, pruned_loss=0.08982, over 5756322.32 frames. ], giga_tot_loss[loss=0.3139, simple_loss=0.3747, pruned_loss=0.1265, over 5644811.48 frames. ], batch size: 128, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:40:42,179 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:40:45,121 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:40:57,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5196, 1.7044, 1.3216, 1.3235], device='cuda:1'), covar=tensor([0.1040, 0.0571, 0.0991, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0452, 0.0521, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 09:41:13,577 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1242737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:41:25,652 INFO [train.py:968] (1/2) Epoch 28, batch 11850, giga_loss[loss=0.3027, simple_loss=0.3494, pruned_loss=0.128, over 23815.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3723, pruned_loss=0.1224, over 5658908.22 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3435, pruned_loss=0.08994, over 5757826.28 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3752, pruned_loss=0.1256, over 5640781.06 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:41:26,367 INFO [optim.py:369] (1/2) 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 09:41:57,348 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 11900, giga_loss[loss=0.3104, simple_loss=0.3753, pruned_loss=0.1228, over 28906.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3732, pruned_loss=0.1226, over 5659534.61 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3436, pruned_loss=0.09005, over 5760889.81 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3759, pruned_loss=0.1256, over 5640667.21 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:43:00,343 INFO [train.py:968] (1/2) Epoch 28, batch 11950, giga_loss[loss=0.2452, simple_loss=0.3253, pruned_loss=0.08252, over 28407.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3713, pruned_loss=0.1209, over 5660809.53 frames. ], libri_tot_loss[loss=0.2615, simple_loss=0.3433, pruned_loss=0.08989, over 5765359.48 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1246, over 5638358.65 frames. ], batch size: 65, lr: 1.14e-03, grad_scale: 1.0
+2023-03-14 09:43:01,042 INFO [optim.py:369] (1/2) 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 09:43:26,575 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1242877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:43:46,808 INFO [train.py:968] (1/2) Epoch 28, batch 12000, libri_loss[loss=0.2848, simple_loss=0.3696, pruned_loss=0.1, over 29203.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3696, pruned_loss=0.1197, over 5666111.35 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3433, pruned_loss=0.08993, over 5763732.79 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3726, pruned_loss=0.123, over 5648229.20 frames. ], batch size: 101, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:43:46,809 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 09:43:55,222 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 09:44:19,597 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1242925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:44:21,815 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1242928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:44:40,454 INFO [train.py:968] (1/2) Epoch 28, batch 12050, libri_loss[loss=0.2793, simple_loss=0.3693, pruned_loss=0.09468, over 28096.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3716, pruned_loss=0.121, over 5664652.31 frames. ], libri_tot_loss[loss=0.262, simple_loss=0.3437, pruned_loss=0.09018, over 5764338.02 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3744, pruned_loss=0.1241, over 5647631.60 frames. ], batch size: 116, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:44:41,075 INFO [optim.py:369] (1/2) 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 09:44:47,364 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 12100, libri_loss[loss=0.3176, simple_loss=0.3904, pruned_loss=0.1223, over 19932.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3728, pruned_loss=0.1218, over 5649580.35 frames. ], libri_tot_loss[loss=0.2622, simple_loss=0.344, pruned_loss=0.09023, over 5751792.60 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3759, pruned_loss=0.1256, over 5642755.79 frames. ], batch size: 188, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:46:14,284 INFO [train.py:968] (1/2) Epoch 28, batch 12150, giga_loss[loss=0.3029, simple_loss=0.3679, pruned_loss=0.119, over 28584.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3712, pruned_loss=0.1213, over 5666306.99 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.344, pruned_loss=0.09027, over 5753970.00 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3742, pruned_loss=0.1249, over 5657154.95 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:46:15,033 INFO [optim.py:369] (1/2) 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 09:46:42,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2404, 1.3765, 3.6096, 3.1487], device='cuda:1'), covar=tensor([0.1674, 0.2676, 0.0488, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0677, 0.1013, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 09:46:58,120 INFO [train.py:968] (1/2) Epoch 28, batch 12200, giga_loss[loss=0.3595, simple_loss=0.4074, pruned_loss=0.1558, over 27690.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3705, pruned_loss=0.1207, over 5671879.12 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3439, pruned_loss=0.09015, over 5757931.10 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3736, pruned_loss=0.1244, over 5659286.93 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:47:35,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4130, 1.6885, 1.1448, 1.2794], device='cuda:1'), covar=tensor([0.1113, 0.0572, 0.1121, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0453, 0.0524, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 09:47:49,931 INFO [train.py:968] (1/2) Epoch 28, batch 12250, giga_loss[loss=0.3603, simple_loss=0.3985, pruned_loss=0.161, over 26736.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3731, pruned_loss=0.1228, over 5663961.61 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.344, pruned_loss=0.09012, over 5748953.25 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3759, pruned_loss=0.1262, over 5661071.81 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:47:50,643 INFO [optim.py:369] (1/2) 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 09:48:16,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4650, 1.5348, 1.2728, 1.5475], device='cuda:1'), covar=tensor([0.0735, 0.0357, 0.0345, 0.0842], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 09:48:28,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:48:36,402 INFO [train.py:968] (1/2) Epoch 28, batch 12300, giga_loss[loss=0.3229, simple_loss=0.377, pruned_loss=0.1344, over 28762.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3726, pruned_loss=0.1227, over 5662513.43 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09005, over 5752665.59 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3757, pruned_loss=0.1262, over 5654739.19 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:49:04,848 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:49:15,739 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 12350, giga_loss[loss=0.3093, simple_loss=0.3727, pruned_loss=0.123, over 28643.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3721, pruned_loss=0.1217, over 5679326.27 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3437, pruned_loss=0.09005, over 5753712.81 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3748, pruned_loss=0.1247, over 5671588.37 frames. ], batch size: 284, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 09:49:26,490 INFO [optim.py:369] (1/2) 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 09:49:28,814 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:50:08,536 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 09:50:11,982 INFO [train.py:968] (1/2) Epoch 28, batch 12400, giga_loss[loss=0.3201, simple_loss=0.3891, pruned_loss=0.1256, over 28615.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3719, pruned_loss=0.1209, over 5663004.81 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3437, pruned_loss=0.08991, over 5749110.42 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3751, pruned_loss=0.1247, over 5658017.30 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:50:13,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:50:29,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6170, 1.5687, 1.7653, 1.4182], device='cuda:1'), covar=tensor([0.1706, 0.2569, 0.1424, 0.1692], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0720, 0.0982, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 09:50:55,950 INFO [train.py:968] (1/2) Epoch 28, batch 12450, giga_loss[loss=0.2734, simple_loss=0.3499, pruned_loss=0.09842, over 28916.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 5674311.83 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3438, pruned_loss=0.08993, over 5750651.23 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3744, pruned_loss=0.1233, over 5668326.70 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:50:57,498 INFO [optim.py:369] (1/2) 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 09:51:09,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2949, 2.2280, 2.3653, 2.0230], device='cuda:1'), covar=tensor([0.1432, 0.2208, 0.1214, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0721, 0.0983, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 09:51:10,424 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:51:43,175 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:51:45,232 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 12500, giga_loss[loss=0.2717, simple_loss=0.3364, pruned_loss=0.1036, over 28237.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3697, pruned_loss=0.119, over 5668543.53 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3439, pruned_loss=0.08987, over 5753879.64 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3722, pruned_loss=0.1221, over 5659624.17 frames. ], batch size: 77, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:51:49,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6553, 4.5012, 4.2823, 2.1324], device='cuda:1'), covar=tensor([0.0600, 0.0708, 0.0814, 0.1930], device='cuda:1'), in_proj_covar=tensor([0.1325, 0.1221, 0.1032, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 09:52:12,516 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:52:36,359 INFO [train.py:968] (1/2) Epoch 28, batch 12550, giga_loss[loss=0.3044, simple_loss=0.3564, pruned_loss=0.1262, over 28783.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3692, pruned_loss=0.1193, over 5662778.92 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.344, pruned_loss=0.08994, over 5754639.24 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3712, pruned_loss=0.1218, over 5654736.66 frames. ], batch size: 99, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:52:37,997 INFO [optim.py:369] (1/2) 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 09:53:20,266 INFO [train.py:968] (1/2) Epoch 28, batch 12600, giga_loss[loss=0.2384, simple_loss=0.3151, pruned_loss=0.08089, over 28922.00 frames. ], tot_loss[loss=0.302, simple_loss=0.367, pruned_loss=0.1185, over 5673412.61 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3439, pruned_loss=0.0899, over 5755858.40 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3694, pruned_loss=0.1215, over 5663064.23 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:54:04,421 INFO [train.py:968] (1/2) Epoch 28, batch 12650, giga_loss[loss=0.326, simple_loss=0.3761, pruned_loss=0.1379, over 27819.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3629, pruned_loss=0.1164, over 5688653.99 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3432, pruned_loss=0.08955, over 5758950.99 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3664, pruned_loss=0.1202, over 5674789.27 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:54:05,954 INFO [optim.py:369] (1/2) 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 09:54:18,934 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1243563.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 09:54:19,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:54:30,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.48 vs. limit=5.0
+2023-03-14 09:54:50,787 INFO [train.py:968] (1/2) Epoch 28, batch 12700, giga_loss[loss=0.2803, simple_loss=0.3438, pruned_loss=0.1085, over 28882.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3614, pruned_loss=0.1157, over 5689314.71 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08945, over 5753421.42 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3648, pruned_loss=0.1196, over 5680552.59 frames. ], batch size: 106, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:54:54,130 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:55:02,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:55:37,392 INFO [train.py:968] (1/2) Epoch 28, batch 12750, giga_loss[loss=0.3149, simple_loss=0.3735, pruned_loss=0.1282, over 28935.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3606, pruned_loss=0.1151, over 5690032.46 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08943, over 5757660.29 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3637, pruned_loss=0.1188, over 5677734.18 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:55:38,093 INFO [optim.py:369] (1/2) 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 09:55:59,023 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 09:56:05,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:56:26,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2693, 1.2302, 1.0777, 1.5548], device='cuda:1'), covar=tensor([0.0801, 0.0383, 0.0391, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 09:56:27,927 INFO [train.py:968] (1/2) Epoch 28, batch 12800, giga_loss[loss=0.2593, simple_loss=0.3398, pruned_loss=0.08946, over 28810.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1131, over 5686475.89 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3431, pruned_loss=0.0894, over 5758652.13 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.363, pruned_loss=0.1164, over 5675069.48 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:56:37,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:56:39,429 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:05,684 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:08,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:15,809 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:17,479 INFO [train.py:968] (1/2) Epoch 28, batch 12850, giga_loss[loss=0.2512, simple_loss=0.328, pruned_loss=0.08723, over 28809.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3588, pruned_loss=0.1102, over 5677037.61 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3431, pruned_loss=0.08941, over 5761640.60 frames. ], giga_tot_loss[loss=0.2941, simple_loss=0.3615, pruned_loss=0.1134, over 5663253.72 frames. ], batch size: 92, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:57:17,901 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:18,247 INFO [optim.py:369] (1/2) 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 09:57:26,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:28,591 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:48,504 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:57:57,050 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 12900, giga_loss[loss=0.3033, simple_loss=0.3572, pruned_loss=0.1247, over 26459.00 frames. ], tot_loss[loss=0.2863, simple_loss=0.3569, pruned_loss=0.1079, over 5665141.76 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.343, pruned_loss=0.08942, over 5752423.83 frames. ], giga_tot_loss[loss=0.2902, simple_loss=0.3592, pruned_loss=0.1106, over 5660697.53 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:58:24,005 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.81 vs. limit=5.0
+2023-03-14 09:58:28,647 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:58:30,944 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 12950, giga_loss[loss=0.2755, simple_loss=0.3467, pruned_loss=0.1022, over 28710.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3535, pruned_loss=0.1045, over 5665103.08 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3427, pruned_loss=0.08933, over 5753662.89 frames. ], giga_tot_loss[loss=0.2852, simple_loss=0.356, pruned_loss=0.1072, over 5658143.07 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 09:58:58,819 INFO [optim.py:369] (1/2) 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 09:58:59,809 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:59:34,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1243884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:59:38,438 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1243887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 09:59:49,567 INFO [train.py:968] (1/2) Epoch 28, batch 13000, giga_loss[loss=0.268, simple_loss=0.337, pruned_loss=0.09949, over 26685.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3511, pruned_loss=0.1015, over 5671060.18 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3426, pruned_loss=0.08939, over 5756067.22 frames. ], giga_tot_loss[loss=0.2804, simple_loss=0.3532, pruned_loss=0.1038, over 5662498.32 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 09:59:49,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0603, 3.2759, 2.3320, 1.0174], device='cuda:1'), covar=tensor([0.8287, 0.3273, 0.3601, 0.7548], device='cuda:1'), in_proj_covar=tensor([0.1842, 0.1729, 0.1659, 0.1500], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 10:00:06,475 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1243916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:00:12,400 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6139, 1.8154, 1.5156, 1.6811], device='cuda:1'), covar=tensor([0.2808, 0.2744, 0.3148, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.1602, 0.1152, 0.1417, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 10:00:28,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1243938.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:00:40,124 INFO [train.py:968] (1/2) Epoch 28, batch 13050, libri_loss[loss=0.238, simple_loss=0.3188, pruned_loss=0.07854, over 29539.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3506, pruned_loss=0.09935, over 5670703.36 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3426, pruned_loss=0.08953, over 5758534.96 frames. ], giga_tot_loss[loss=0.2776, simple_loss=0.3525, pruned_loss=0.1013, over 5659512.35 frames. ], batch size: 83, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:00:41,558 INFO [optim.py:369] (1/2) 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 10:01:30,640 INFO [train.py:968] (1/2) Epoch 28, batch 13100, libri_loss[loss=0.278, simple_loss=0.3468, pruned_loss=0.1046, over 19452.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3512, pruned_loss=0.09999, over 5655570.05 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3423, pruned_loss=0.08944, over 5751214.62 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3531, pruned_loss=0.1018, over 5652546.88 frames. ], batch size: 187, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:02:13,033 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 13150, giga_loss[loss=0.2664, simple_loss=0.3351, pruned_loss=0.09885, over 26555.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3492, pruned_loss=0.0986, over 5654987.14 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3423, pruned_loss=0.0894, over 5752000.55 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3508, pruned_loss=0.1001, over 5651602.32 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:02:22,895 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:02:24,342 INFO [optim.py:369] (1/2) 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 10:02:36,009 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:02:55,686 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244081.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:02:58,283 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 13200, giga_loss[loss=0.2618, simple_loss=0.3466, pruned_loss=0.08851, over 28858.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3458, pruned_loss=0.09616, over 5665416.23 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3419, pruned_loss=0.08928, over 5754445.62 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09756, over 5659395.29 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:03:29,409 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244113.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:04:02,385 INFO [train.py:968] (1/2) Epoch 28, batch 13250, giga_loss[loss=0.2934, simple_loss=0.3657, pruned_loss=0.1105, over 28735.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3459, pruned_loss=0.09628, over 5665641.23 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3418, pruned_loss=0.08922, over 5753657.53 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09757, over 5659947.60 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:04:04,480 INFO [optim.py:369] (1/2) 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 10:04:12,972 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:04:15,295 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-14 10:04:51,901 INFO [train.py:968] (1/2) Epoch 28, batch 13300, giga_loss[loss=0.298, simple_loss=0.3745, pruned_loss=0.1107, over 28701.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3458, pruned_loss=0.09608, over 5647523.22 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.342, pruned_loss=0.08955, over 5732563.71 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3468, pruned_loss=0.09696, over 5658919.81 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:05:11,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7489, 1.9728, 1.5205, 1.8496], device='cuda:1'), covar=tensor([0.2826, 0.2817, 0.3355, 0.2704], device='cuda:1'), in_proj_covar=tensor([0.1602, 0.1152, 0.1418, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 10:05:43,794 INFO [train.py:968] (1/2) Epoch 28, batch 13350, giga_loss[loss=0.2644, simple_loss=0.3432, pruned_loss=0.09278, over 28233.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3436, pruned_loss=0.09436, over 5653548.58 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3415, pruned_loss=0.08934, over 5736108.44 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3449, pruned_loss=0.09536, over 5657837.39 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:05:48,796 INFO [optim.py:369] (1/2) 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 10:06:10,618 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4688, 2.1520, 1.6891, 0.6778], device='cuda:1'), covar=tensor([0.5572, 0.3630, 0.4541, 0.6866], device='cuda:1'), in_proj_covar=tensor([0.1832, 0.1718, 0.1649, 0.1492], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 10:06:32,934 INFO [train.py:968] (1/2) Epoch 28, batch 13400, libri_loss[loss=0.2404, simple_loss=0.3267, pruned_loss=0.0771, over 29533.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3396, pruned_loss=0.09157, over 5667701.81 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3405, pruned_loss=0.08899, over 5742393.46 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3416, pruned_loss=0.09281, over 5663022.03 frames. ], batch size: 84, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:07:28,437 INFO [train.py:968] (1/2) Epoch 28, batch 13450, giga_loss[loss=0.2299, simple_loss=0.3117, pruned_loss=0.07402, over 28360.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3351, pruned_loss=0.08924, over 5658431.73 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.34, pruned_loss=0.08876, over 5744473.16 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.337, pruned_loss=0.09045, over 5652074.71 frames. ], batch size: 65, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:07:31,867 INFO [optim.py:369] (1/2) 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 10:07:38,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3647, 3.2135, 3.0422, 1.3831], device='cuda:1'), covar=tensor([0.0926, 0.1040, 0.0989, 0.2291], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.1199, 0.1011, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 10:07:55,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:08:01,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 10:08:19,306 INFO [train.py:968] (1/2) Epoch 28, batch 13500, giga_loss[loss=0.2816, simple_loss=0.3513, pruned_loss=0.1059, over 28421.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3355, pruned_loss=0.09038, over 5641957.01 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.34, pruned_loss=0.08897, over 5736010.01 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3369, pruned_loss=0.09121, over 5641059.44 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:08:38,191 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244416.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:08:46,638 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244424.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:08:59,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:09:12,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4454, 1.5588, 1.4522, 1.7143], device='cuda:1'), covar=tensor([0.0659, 0.0309, 0.0314, 0.0702], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 10:09:13,521 INFO [train.py:968] (1/2) Epoch 28, batch 13550, giga_loss[loss=0.249, simple_loss=0.3298, pruned_loss=0.08406, over 27940.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3362, pruned_loss=0.09157, over 5638612.20 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3397, pruned_loss=0.08887, over 5738644.98 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3375, pruned_loss=0.09233, over 5634341.49 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:09:17,686 INFO [optim.py:369] (1/2) 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 10:10:13,000 INFO [train.py:968] (1/2) Epoch 28, batch 13600, giga_loss[loss=0.2797, simple_loss=0.3643, pruned_loss=0.09758, over 28607.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3379, pruned_loss=0.09162, over 5630509.68 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3394, pruned_loss=0.08887, over 5731783.05 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3392, pruned_loss=0.09229, over 5631116.06 frames. ], batch size: 242, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:10:52,182 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:08,283 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4139, 1.5761, 1.1380, 1.2282], device='cuda:1'), covar=tensor([0.0908, 0.0463, 0.0928, 0.1162], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0450, 0.0522, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 10:11:12,587 INFO [train.py:968] (1/2) Epoch 28, batch 13650, giga_loss[loss=0.2242, simple_loss=0.3129, pruned_loss=0.06774, over 28780.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3395, pruned_loss=0.09166, over 5637927.19 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3391, pruned_loss=0.08872, over 5736179.74 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3408, pruned_loss=0.09238, over 5632619.83 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:11:16,516 INFO [optim.py:369] (1/2) 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 10:11:24,342 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244559.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:11:28,237 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244562.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:11:33,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:37,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:44,227 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:48,588 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:11:52,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:12:05,516 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 13700, giga_loss[loss=0.2312, simple_loss=0.321, pruned_loss=0.07068, over 28694.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3394, pruned_loss=0.0918, over 5639539.03 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3387, pruned_loss=0.08854, over 5737901.56 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3408, pruned_loss=0.0926, over 5632591.96 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:12:14,686 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244599.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:12:27,041 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4529, 3.3734, 1.5847, 1.6512], device='cuda:1'), covar=tensor([0.0992, 0.0346, 0.0925, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0571, 0.0410, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 10:12:30,434 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:13:11,897 INFO [train.py:968] (1/2) Epoch 28, batch 13750, giga_loss[loss=0.2537, simple_loss=0.3401, pruned_loss=0.08372, over 28050.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3361, pruned_loss=0.08961, over 5652348.34 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3381, pruned_loss=0.08841, over 5741663.02 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3377, pruned_loss=0.09039, over 5641842.36 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:13:15,711 INFO [optim.py:369] (1/2) 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 10:13:34,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:13:48,215 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:13:51,470 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:14:13,929 INFO [train.py:968] (1/2) Epoch 28, batch 13800, giga_loss[loss=0.2281, simple_loss=0.3229, pruned_loss=0.06671, over 28375.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.335, pruned_loss=0.08793, over 5646521.92 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3376, pruned_loss=0.08823, over 5743861.57 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3367, pruned_loss=0.08872, over 5634433.26 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:14:21,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2471, 1.8388, 1.7118, 1.5010], device='cuda:1'), covar=tensor([0.2533, 0.1796, 0.2355, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0754, 0.0723, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 10:14:23,488 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:15:10,271 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1244747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:15:11,774 INFO [train.py:968] (1/2) Epoch 28, batch 13850, giga_loss[loss=0.239, simple_loss=0.3044, pruned_loss=0.08676, over 24867.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3322, pruned_loss=0.08577, over 5649779.84 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3373, pruned_loss=0.08807, over 5743320.58 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3337, pruned_loss=0.08649, over 5637397.76 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:15:15,313 INFO [optim.py:369] (1/2) 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 10:15:29,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5426, 1.6756, 1.7699, 1.3850], device='cuda:1'), covar=tensor([0.1848, 0.2717, 0.1601, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0713, 0.0981, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 10:15:58,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4670, 1.5235, 1.3689, 1.7011], device='cuda:1'), covar=tensor([0.0747, 0.0326, 0.0346, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 10:16:08,637 INFO [train.py:968] (1/2) Epoch 28, batch 13900, giga_loss[loss=0.2444, simple_loss=0.3283, pruned_loss=0.08032, over 28124.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3298, pruned_loss=0.08559, over 5658181.30 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3365, pruned_loss=0.08765, over 5747540.04 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3315, pruned_loss=0.08646, over 5640294.34 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:16:09,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1479, 1.7686, 1.5622, 1.3623], device='cuda:1'), covar=tensor([0.2510, 0.1960, 0.2234, 0.2347], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0751, 0.0720, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 10:16:21,168 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1244810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:17:03,017 INFO [train.py:968] (1/2) Epoch 28, batch 13950, giga_loss[loss=0.2263, simple_loss=0.3046, pruned_loss=0.07399, over 28490.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3303, pruned_loss=0.08648, over 5655614.15 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.336, pruned_loss=0.08763, over 5740767.48 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.332, pruned_loss=0.08717, over 5645510.92 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:17:08,123 INFO [optim.py:369] (1/2) 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 10:17:48,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1244890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:17:52,101 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1244893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:17:58,121 INFO [train.py:968] (1/2) Epoch 28, batch 14000, giga_loss[loss=0.2948, simple_loss=0.3627, pruned_loss=0.1135, over 26658.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3307, pruned_loss=0.08636, over 5670628.34 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3356, pruned_loss=0.08748, over 5746218.81 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3323, pruned_loss=0.08701, over 5655078.37 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:18:02,542 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6306, 2.0192, 1.7451, 1.7125], device='cuda:1'), covar=tensor([0.0700, 0.0270, 0.0293, 0.0828], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 10:18:25,861 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1244922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:19:01,202 INFO [train.py:968] (1/2) Epoch 28, batch 14050, giga_loss[loss=0.3007, simple_loss=0.3724, pruned_loss=0.1144, over 28961.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3346, pruned_loss=0.08756, over 5679852.89 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3359, pruned_loss=0.08771, over 5748605.20 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3355, pruned_loss=0.08786, over 5664637.03 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:19:02,608 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:20:02,853 INFO [train.py:968] (1/2) Epoch 28, batch 14100, giga_loss[loss=0.2352, simple_loss=0.3183, pruned_loss=0.07609, over 28178.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3337, pruned_loss=0.08687, over 5679711.28 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3358, pruned_loss=0.08771, over 5751624.57 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3345, pruned_loss=0.0871, over 5662848.68 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:20:15,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 10:20:56,115 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:21:04,885 INFO [train.py:968] (1/2) Epoch 28, batch 14150, giga_loss[loss=0.2446, simple_loss=0.3364, pruned_loss=0.07638, over 28889.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3319, pruned_loss=0.08605, over 5682875.86 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3355, pruned_loss=0.08771, over 5747325.82 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3327, pruned_loss=0.08619, over 5670759.06 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:21:11,360 INFO [optim.py:369] (1/2) 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 10:21:32,724 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.89 vs. limit=2.0
+2023-03-14 10:21:55,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:21:58,767 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 14200, giga_loss[loss=0.2583, simple_loss=0.3377, pruned_loss=0.08941, over 28392.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3337, pruned_loss=0.08733, over 5675566.38 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3356, pruned_loss=0.08782, over 5748137.05 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3342, pruned_loss=0.08732, over 5661473.71 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:22:05,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2464, 1.0795, 1.1074, 1.3924], device='cuda:1'), covar=tensor([0.0693, 0.0359, 0.0338, 0.0726], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 10:22:39,085 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:22:39,100 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:22:40,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3532, 4.1850, 4.0265, 1.8250], device='cuda:1'), covar=tensor([0.0549, 0.0682, 0.0694, 0.2127], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1192, 0.1004, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 10:23:05,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1767, 1.6446, 1.1678, 0.5109], device='cuda:1'), covar=tensor([0.4356, 0.2630, 0.4153, 0.5886], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1725, 0.1653, 0.1496], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 10:23:09,695 INFO [train.py:968] (1/2) Epoch 28, batch 14250, giga_loss[loss=0.258, simple_loss=0.3552, pruned_loss=0.08044, over 28621.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3381, pruned_loss=0.08778, over 5662821.87 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3353, pruned_loss=0.0877, over 5747164.32 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3387, pruned_loss=0.08789, over 5651358.30 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:23:17,944 INFO [optim.py:369] (1/2) 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 10:23:47,354 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:23:49,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:23:50,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:24:06,938 INFO [train.py:968] (1/2) Epoch 28, batch 14300, giga_loss[loss=0.2476, simple_loss=0.3206, pruned_loss=0.08725, over 26972.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3383, pruned_loss=0.08652, over 5657624.47 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3346, pruned_loss=0.08756, over 5746567.81 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3395, pruned_loss=0.08672, over 5645906.27 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:24:29,406 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:24:42,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:25:05,659 INFO [train.py:968] (1/2) Epoch 28, batch 14350, giga_loss[loss=0.2524, simple_loss=0.3357, pruned_loss=0.08454, over 27601.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3381, pruned_loss=0.08521, over 5659508.90 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3345, pruned_loss=0.08752, over 5749426.17 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3392, pruned_loss=0.08536, over 5645376.31 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:25:12,166 INFO [optim.py:369] (1/2) 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 10:25:22,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 10:26:04,840 INFO [train.py:968] (1/2) Epoch 28, batch 14400, giga_loss[loss=0.2545, simple_loss=0.3424, pruned_loss=0.08332, over 28617.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.338, pruned_loss=0.08491, over 5668767.63 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3345, pruned_loss=0.08752, over 5750233.06 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.339, pruned_loss=0.08494, over 5654047.34 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:26:17,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5528, 1.7882, 1.2270, 1.3476], device='cuda:1'), covar=tensor([0.1097, 0.0560, 0.1057, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0449, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 10:26:23,081 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:26:45,145 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:26:47,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:27:08,577 INFO [train.py:968] (1/2) Epoch 28, batch 14450, giga_loss[loss=0.2321, simple_loss=0.3142, pruned_loss=0.07495, over 28780.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3375, pruned_loss=0.08557, over 5671333.59 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3345, pruned_loss=0.08749, over 5750654.52 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3384, pruned_loss=0.08559, over 5658391.59 frames. ], batch size: 119, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:27:15,335 INFO [optim.py:369] (1/2) 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 10:27:23,465 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:27:40,652 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245372.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:28:21,897 INFO [train.py:968] (1/2) Epoch 28, batch 14500, giga_loss[loss=0.2703, simple_loss=0.3488, pruned_loss=0.09594, over 27714.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3379, pruned_loss=0.08683, over 5668236.81 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3345, pruned_loss=0.08752, over 5752003.26 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3387, pruned_loss=0.08681, over 5656059.14 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:29:14,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 10:29:34,235 INFO [train.py:968] (1/2) Epoch 28, batch 14550, giga_loss[loss=0.2145, simple_loss=0.3052, pruned_loss=0.06194, over 28658.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3354, pruned_loss=0.0855, over 5679791.42 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.334, pruned_loss=0.08727, over 5752474.32 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3366, pruned_loss=0.08568, over 5666282.81 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:29:42,072 INFO [optim.py:369] (1/2) 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 10:30:22,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:30:34,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6296, 1.5409, 1.8704, 1.4490], device='cuda:1'), covar=tensor([0.1777, 0.2658, 0.1490, 0.1879], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0713, 0.0983, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 10:30:46,242 INFO [train.py:968] (1/2) Epoch 28, batch 14600, giga_loss[loss=0.2666, simple_loss=0.3485, pruned_loss=0.09239, over 27977.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3318, pruned_loss=0.08373, over 5673242.79 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3343, pruned_loss=0.08747, over 5755856.64 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3325, pruned_loss=0.08362, over 5657679.89 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:30:48,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:31:16,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3606, 3.2993, 1.5731, 1.4964], device='cuda:1'), covar=tensor([0.1015, 0.0331, 0.0951, 0.1403], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0569, 0.0409, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 10:31:55,253 INFO [train.py:968] (1/2) Epoch 28, batch 14650, giga_loss[loss=0.2307, simple_loss=0.3072, pruned_loss=0.07711, over 28901.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3299, pruned_loss=0.08289, over 5674863.55 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3342, pruned_loss=0.08744, over 5756627.88 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3305, pruned_loss=0.08277, over 5660797.02 frames. ], batch size: 93, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:32:04,102 INFO [optim.py:369] (1/2) 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 10:32:27,914 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:32:57,837 INFO [train.py:968] (1/2) Epoch 28, batch 14700, giga_loss[loss=0.2425, simple_loss=0.3351, pruned_loss=0.07491, over 28852.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3311, pruned_loss=0.08396, over 5683575.63 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3342, pruned_loss=0.0875, over 5758777.32 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3315, pruned_loss=0.08376, over 5669556.39 frames. ], batch size: 174, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:32:59,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:10,740 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:30,783 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9226, 3.7450, 3.5948, 1.6210], device='cuda:1'), covar=tensor([0.0743, 0.0882, 0.0907, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1183, 0.0997, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 10:33:42,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5077, 4.0608, 1.6155, 1.6199], device='cuda:1'), covar=tensor([0.0982, 0.0304, 0.0919, 0.1316], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0568, 0.0410, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 10:33:53,489 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:57,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:33:59,941 INFO [train.py:968] (1/2) Epoch 28, batch 14750, giga_loss[loss=0.2341, simple_loss=0.3254, pruned_loss=0.07144, over 28891.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3345, pruned_loss=0.0857, over 5679140.03 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3339, pruned_loss=0.08748, over 5753501.68 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.335, pruned_loss=0.0855, over 5670090.31 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:34:06,918 INFO [optim.py:369] (1/2) 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 10:34:29,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:34:34,678 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 10:34:43,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:34:58,504 INFO [train.py:968] (1/2) Epoch 28, batch 14800, giga_loss[loss=0.2903, simple_loss=0.3622, pruned_loss=0.1091, over 28703.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.333, pruned_loss=0.08616, over 5685400.15 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.334, pruned_loss=0.08754, over 5756052.13 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3333, pruned_loss=0.08589, over 5674677.73 frames. ], batch size: 262, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:35:15,425 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1245714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:35:57,707 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:35:57,873 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.44 vs. limit=2.0
+2023-03-14 10:36:00,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245747.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:36:00,120 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:36:01,293 INFO [train.py:968] (1/2) Epoch 28, batch 14850, giga_loss[loss=0.2898, simple_loss=0.3539, pruned_loss=0.1128, over 27639.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3341, pruned_loss=0.08829, over 5677281.81 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3331, pruned_loss=0.08713, over 5760530.33 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3353, pruned_loss=0.08846, over 5662098.61 frames. ], batch size: 474, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:36:06,248 INFO [optim.py:369] (1/2) 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 10:36:32,599 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:36:58,101 INFO [train.py:968] (1/2) Epoch 28, batch 14900, giga_loss[loss=0.27, simple_loss=0.3507, pruned_loss=0.09467, over 27509.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3333, pruned_loss=0.0874, over 5674809.61 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3329, pruned_loss=0.08695, over 5760803.37 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3344, pruned_loss=0.08771, over 5660143.04 frames. ], batch size: 472, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:37:37,555 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:37:40,554 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:38:06,272 INFO [train.py:968] (1/2) Epoch 28, batch 14950, libri_loss[loss=0.2231, simple_loss=0.296, pruned_loss=0.07514, over 29464.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3345, pruned_loss=0.08684, over 5677436.38 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3324, pruned_loss=0.08671, over 5763682.77 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3359, pruned_loss=0.08732, over 5661449.38 frames. ], batch size: 70, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:38:13,738 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:38:20,142 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:38:43,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4313, 2.9340, 1.5273, 1.5854], device='cuda:1'), covar=tensor([0.0910, 0.0373, 0.0925, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0566, 0.0408, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 10:38:46,650 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3789, 1.5591, 1.6159, 1.2240], device='cuda:1'), covar=tensor([0.1891, 0.2894, 0.1657, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.0927, 0.0711, 0.0980, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 10:39:06,899 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1245890.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 10:39:11,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1245893.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:39:18,972 INFO [train.py:968] (1/2) Epoch 28, batch 15000, giga_loss[loss=0.2573, simple_loss=0.3406, pruned_loss=0.08701, over 28937.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3348, pruned_loss=0.08665, over 5680581.87 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.332, pruned_loss=0.08652, over 5767303.30 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3363, pruned_loss=0.08719, over 5662587.32 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:39:18,972 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 10:39:27,332 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 10:40:06,111 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1245922.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 10:40:15,739 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 15050, giga_loss[loss=0.267, simple_loss=0.3431, pruned_loss=0.09546, over 28529.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3313, pruned_loss=0.08477, over 5691288.07 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3319, pruned_loss=0.08643, over 5769091.56 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3325, pruned_loss=0.08525, over 5674291.35 frames. ], batch size: 78, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:40:43,775 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:41:29,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1245987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:41:47,795 INFO [train.py:968] (1/2) Epoch 28, batch 15100, giga_loss[loss=0.2108, simple_loss=0.2913, pruned_loss=0.06519, over 28952.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3262, pruned_loss=0.08335, over 5694711.96 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3312, pruned_loss=0.08629, over 5773280.56 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3278, pruned_loss=0.0838, over 5675269.84 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:41:50,530 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:41:54,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:42:27,822 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 15150, giga_loss[loss=0.2421, simple_loss=0.3261, pruned_loss=0.07903, over 28905.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3255, pruned_loss=0.08349, over 5693082.98 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3316, pruned_loss=0.08667, over 5774301.50 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3263, pruned_loss=0.08348, over 5676279.72 frames. ], batch size: 199, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:42:58,293 INFO [optim.py:369] (1/2) 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 10:43:33,934 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:43:38,527 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:43:41,801 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 15200, giga_loss[loss=0.2488, simple_loss=0.3302, pruned_loss=0.08375, over 28819.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3271, pruned_loss=0.08494, over 5686237.33 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3315, pruned_loss=0.08655, over 5776608.94 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3277, pruned_loss=0.08499, over 5668542.19 frames. ], batch size: 227, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:43:49,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8208, 1.1798, 2.8662, 2.7908], device='cuda:1'), covar=tensor([0.1801, 0.2702, 0.0594, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0674, 0.1001, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 10:44:01,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3142, 1.0632, 1.0813, 1.4410], device='cuda:1'), covar=tensor([0.0679, 0.0363, 0.0339, 0.0738], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 10:44:12,889 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:44:19,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3140, 1.3374, 3.4508, 3.2146], device='cuda:1'), covar=tensor([0.1621, 0.2970, 0.0507, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0673, 0.1000, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 10:44:20,184 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:44:24,926 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:44:47,235 INFO [train.py:968] (1/2) Epoch 28, batch 15250, giga_loss[loss=0.2261, simple_loss=0.309, pruned_loss=0.07163, over 28053.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3261, pruned_loss=0.08433, over 5666240.66 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3318, pruned_loss=0.08677, over 5776341.89 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3263, pruned_loss=0.08416, over 5652249.78 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:44:54,933 INFO [optim.py:369] (1/2) 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 10:45:02,849 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:45:46,681 INFO [train.py:968] (1/2) Epoch 28, batch 15300, giga_loss[loss=0.24, simple_loss=0.329, pruned_loss=0.07551, over 28048.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3241, pruned_loss=0.08199, over 5676881.92 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3316, pruned_loss=0.0867, over 5778239.78 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3244, pruned_loss=0.08185, over 5662247.48 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:46:26,479 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:46:31,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:46:47,664 INFO [train.py:968] (1/2) Epoch 28, batch 15350, libri_loss[loss=0.2173, simple_loss=0.2944, pruned_loss=0.07007, over 29342.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3241, pruned_loss=0.08249, over 5671640.23 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3317, pruned_loss=0.08674, over 5781447.30 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.3239, pruned_loss=0.08218, over 5653032.65 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:46:58,170 INFO [optim.py:369] (1/2) 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 10:47:10,440 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:47:38,994 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-14 10:47:55,856 INFO [train.py:968] (1/2) Epoch 28, batch 15400, giga_loss[loss=0.2465, simple_loss=0.3273, pruned_loss=0.08278, over 28980.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3228, pruned_loss=0.08139, over 5682310.79 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.3312, pruned_loss=0.08646, over 5781517.07 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.323, pruned_loss=0.08133, over 5666271.00 frames. ], batch size: 120, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:48:00,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6743, 1.9641, 1.3734, 1.5453], device='cuda:1'), covar=tensor([0.1101, 0.0595, 0.1052, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0447, 0.0520, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 10:48:04,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:48:18,879 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:48:57,648 INFO [train.py:968] (1/2) Epoch 28, batch 15450, giga_loss[loss=0.1965, simple_loss=0.2891, pruned_loss=0.05198, over 28850.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3238, pruned_loss=0.08155, over 5686499.43 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3308, pruned_loss=0.08636, over 5775939.01 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3241, pruned_loss=0.08147, over 5676784.72 frames. ], batch size: 164, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:49:08,892 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 15500, giga_loss[loss=0.2421, simple_loss=0.3262, pruned_loss=0.07902, over 28892.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3249, pruned_loss=0.08269, over 5687785.87 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3305, pruned_loss=0.08621, over 5778022.73 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3254, pruned_loss=0.08268, over 5677234.78 frames. ], batch size: 213, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:51:11,408 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 15550, giga_loss[loss=0.2514, simple_loss=0.303, pruned_loss=0.09993, over 24381.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3242, pruned_loss=0.08273, over 5683883.76 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3304, pruned_loss=0.08619, over 5779331.32 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3246, pruned_loss=0.08271, over 5673670.69 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:51:14,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1246451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:51:22,312 INFO [optim.py:369] (1/2) 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 10:51:48,441 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 15600, giga_loss[loss=0.2731, simple_loss=0.3624, pruned_loss=0.09185, over 28618.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.3238, pruned_loss=0.08096, over 5673825.04 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3297, pruned_loss=0.08582, over 5781372.87 frames. ], giga_tot_loss[loss=0.2435, simple_loss=0.3246, pruned_loss=0.08118, over 5662283.32 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 10:52:47,228 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1832, 1.3842, 1.3636, 1.1772], device='cuda:1'), covar=tensor([0.2325, 0.2344, 0.1649, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.2014, 0.1974, 0.1875, 0.2019], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 10:52:48,749 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:53:08,079 INFO [train.py:968] (1/2) Epoch 28, batch 15650, giga_loss[loss=0.2929, simple_loss=0.3589, pruned_loss=0.1135, over 26795.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3271, pruned_loss=0.08221, over 5674210.63 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3292, pruned_loss=0.08567, over 5783111.59 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.328, pruned_loss=0.08237, over 5659890.21 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 10:53:19,649 INFO [optim.py:369] (1/2) 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 10:54:08,068 INFO [train.py:968] (1/2) Epoch 28, batch 15700, giga_loss[loss=0.2272, simple_loss=0.3213, pruned_loss=0.06651, over 28537.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3296, pruned_loss=0.0832, over 5659234.85 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3294, pruned_loss=0.08583, over 5771838.15 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3301, pruned_loss=0.08312, over 5655753.41 frames. ], batch size: 71, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:55:06,427 INFO [train.py:968] (1/2) Epoch 28, batch 15750, giga_loss[loss=0.2883, simple_loss=0.3383, pruned_loss=0.1192, over 24551.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3296, pruned_loss=0.08368, over 5653072.53 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.329, pruned_loss=0.08564, over 5773264.80 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3304, pruned_loss=0.08377, over 5646377.15 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:55:13,533 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:55:21,458 INFO [optim.py:369] (1/2) 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 10:56:01,335 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:56:06,537 INFO [train.py:968] (1/2) Epoch 28, batch 15800, giga_loss[loss=0.2452, simple_loss=0.3219, pruned_loss=0.08428, over 28487.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3285, pruned_loss=0.08326, over 5656566.56 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3292, pruned_loss=0.08569, over 5772467.99 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3289, pruned_loss=0.08326, over 5650335.56 frames. ], batch size: 368, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:56:13,345 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-14 10:57:08,256 INFO [train.py:968] (1/2) Epoch 28, batch 15850, giga_loss[loss=0.2093, simple_loss=0.2936, pruned_loss=0.06245, over 28654.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3271, pruned_loss=0.0824, over 5661556.78 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3293, pruned_loss=0.0856, over 5776399.25 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3273, pruned_loss=0.08239, over 5649908.10 frames. ], batch size: 85, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:57:20,567 INFO [optim.py:369] (1/2) 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 10:58:02,791 INFO [train.py:968] (1/2) Epoch 28, batch 15900, giga_loss[loss=0.2456, simple_loss=0.3264, pruned_loss=0.08237, over 28454.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3257, pruned_loss=0.08264, over 5672110.96 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.3292, pruned_loss=0.08561, over 5778105.98 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3259, pruned_loss=0.08252, over 5657599.26 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:58:16,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8717, 4.6984, 4.5040, 2.4735], device='cuda:1'), covar=tensor([0.0504, 0.0650, 0.0769, 0.1821], device='cuda:1'), in_proj_covar=tensor([0.1282, 0.1181, 0.0996, 0.0737], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 10:58:47,288 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1246836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:58:52,135 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 15950, giga_loss[loss=0.293, simple_loss=0.3532, pruned_loss=0.1164, over 26854.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3266, pruned_loss=0.08307, over 5667727.27 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3292, pruned_loss=0.08567, over 5771216.53 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3267, pruned_loss=0.08286, over 5659946.98 frames. ], batch size: 555, lr: 1.14e-03, grad_scale: 2.0
+2023-03-14 10:59:15,467 INFO [optim.py:369] (1/2) 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 10:59:28,649 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:59:28,670 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1246868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 10:59:33,479 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1246872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:00:02,233 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5478, 1.6615, 1.7308, 1.3484], device='cuda:1'), covar=tensor([0.1884, 0.2759, 0.1603, 0.1971], device='cuda:1'), in_proj_covar=tensor([0.0929, 0.0709, 0.0979, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 11:00:06,097 INFO [train.py:968] (1/2) Epoch 28, batch 16000, giga_loss[loss=0.2264, simple_loss=0.2973, pruned_loss=0.07772, over 24681.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3278, pruned_loss=0.08338, over 5668935.50 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3288, pruned_loss=0.08561, over 5772623.84 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3283, pruned_loss=0.08322, over 5659570.56 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:00:18,629 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:00:28,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9869, 1.2688, 0.9832, 0.2895], device='cuda:1'), covar=tensor([0.4061, 0.3242, 0.4642, 0.6602], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1731, 0.1654, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 11:01:08,645 INFO [train.py:968] (1/2) Epoch 28, batch 16050, giga_loss[loss=0.2593, simple_loss=0.34, pruned_loss=0.08933, over 28122.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3286, pruned_loss=0.08437, over 5666744.25 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3286, pruned_loss=0.08537, over 5776854.86 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3291, pruned_loss=0.08441, over 5652008.31 frames. ], batch size: 412, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:01:18,376 INFO [optim.py:369] (1/2) 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 11:01:23,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-14 11:02:07,884 INFO [train.py:968] (1/2) Epoch 28, batch 16100, giga_loss[loss=0.2619, simple_loss=0.349, pruned_loss=0.08743, over 28927.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.332, pruned_loss=0.08624, over 5668840.03 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3285, pruned_loss=0.0853, over 5779591.64 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3326, pruned_loss=0.08634, over 5652841.36 frames. ], batch size: 145, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:02:30,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2045, 1.4892, 1.4363, 1.1781], device='cuda:1'), covar=tensor([0.3033, 0.2578, 0.1829, 0.2433], device='cuda:1'), in_proj_covar=tensor([0.2016, 0.1971, 0.1872, 0.2018], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 11:02:39,837 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247029.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:03:01,750 INFO [train.py:968] (1/2) Epoch 28, batch 16150, giga_loss[loss=0.236, simple_loss=0.3314, pruned_loss=0.07034, over 28911.00 frames. ], tot_loss[loss=0.255, simple_loss=0.335, pruned_loss=0.08748, over 5665082.77 frames. ], libri_tot_loss[loss=0.2498, simple_loss=0.3288, pruned_loss=0.08537, over 5783461.24 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3353, pruned_loss=0.08756, over 5643832.29 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:03:03,524 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:03:07,144 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:03:40,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:04:03,412 INFO [train.py:968] (1/2) Epoch 28, batch 16200, giga_loss[loss=0.2686, simple_loss=0.3439, pruned_loss=0.09669, over 28583.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3356, pruned_loss=0.08739, over 5664534.63 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3286, pruned_loss=0.08521, over 5787326.74 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3362, pruned_loss=0.08764, over 5641214.22 frames. ], batch size: 307, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:04:14,969 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4005, 1.2421, 3.9238, 3.4119], device='cuda:1'), covar=tensor([0.1686, 0.2934, 0.0476, 0.0831], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0670, 0.0994, 0.0967], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 11:04:19,910 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1247113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:05:05,325 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3787, 3.0713, 1.5143, 1.5088], device='cuda:1'), covar=tensor([0.0985, 0.0428, 0.0965, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0566, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 11:05:10,263 INFO [train.py:968] (1/2) Epoch 28, batch 16250, giga_loss[loss=0.2558, simple_loss=0.3409, pruned_loss=0.08538, over 28402.00 frames. ], tot_loss[loss=0.255, simple_loss=0.335, pruned_loss=0.08747, over 5668800.92 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3283, pruned_loss=0.08501, over 5791142.26 frames. ], giga_tot_loss[loss=0.2559, simple_loss=0.336, pruned_loss=0.08792, over 5643092.89 frames. ], batch size: 336, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:05:23,043 INFO [optim.py:369] (1/2) 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 11:05:23,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1247159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:05:23,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4261, 5.2922, 5.0398, 2.2191], device='cuda:1'), covar=tensor([0.0485, 0.0660, 0.0716, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.1280, 0.1178, 0.0995, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 11:05:40,188 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:05:43,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:06:10,360 INFO [train.py:968] (1/2) Epoch 28, batch 16300, giga_loss[loss=0.2218, simple_loss=0.2897, pruned_loss=0.07693, over 24614.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3321, pruned_loss=0.08615, over 5673898.63 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3278, pruned_loss=0.08477, over 5793598.80 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3335, pruned_loss=0.08679, over 5647856.59 frames. ], batch size: 705, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:06:18,660 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:07:07,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:07:13,095 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:07:14,047 INFO [train.py:968] (1/2) Epoch 28, batch 16350, giga_loss[loss=0.2689, simple_loss=0.3439, pruned_loss=0.09694, over 29078.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.331, pruned_loss=0.08539, over 5685695.33 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3273, pruned_loss=0.08468, over 5795414.08 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3326, pruned_loss=0.086, over 5661141.05 frames. ], batch size: 285, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:07:25,778 INFO [optim.py:369] (1/2) 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 11:07:44,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3271, 1.8028, 1.3129, 0.7659], device='cuda:1'), covar=tensor([0.6814, 0.3353, 0.3771, 0.6941], device='cuda:1'), in_proj_covar=tensor([0.1837, 0.1728, 0.1651, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 11:08:19,746 INFO [train.py:968] (1/2) Epoch 28, batch 16400, giga_loss[loss=0.2298, simple_loss=0.3128, pruned_loss=0.07341, over 29084.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3308, pruned_loss=0.0863, over 5677017.75 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3273, pruned_loss=0.08465, over 5796921.29 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3321, pruned_loss=0.08682, over 5654723.37 frames. ], batch size: 200, lr: 1.14e-03, grad_scale: 8.0
+2023-03-14 11:08:31,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6746, 3.5400, 3.3469, 2.0160], device='cuda:1'), covar=tensor([0.0669, 0.0848, 0.0932, 0.1795], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1177, 0.0995, 0.0736], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 11:08:34,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3839, 1.6721, 1.5831, 1.3278], device='cuda:1'), covar=tensor([0.3675, 0.2712, 0.1967, 0.2780], device='cuda:1'), in_proj_covar=tensor([0.2024, 0.1980, 0.1878, 0.2029], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 11:09:16,158 INFO [train.py:968] (1/2) Epoch 28, batch 16450, giga_loss[loss=0.2408, simple_loss=0.3249, pruned_loss=0.07839, over 29023.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3283, pruned_loss=0.08524, over 5680945.57 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3268, pruned_loss=0.08439, over 5801867.19 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.33, pruned_loss=0.08597, over 5651673.97 frames. ], batch size: 155, lr: 1.14e-03, grad_scale: 4.0
+2023-03-14 11:09:28,623 INFO [optim.py:369] (1/2) 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 11:09:57,776 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:01,867 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:02,528 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:05,295 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:08,789 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.07 vs. limit=5.0
+2023-03-14 11:10:09,550 INFO [train.py:968] (1/2) Epoch 28, batch 16500, giga_loss[loss=0.2687, simple_loss=0.3645, pruned_loss=0.08642, over 28989.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3274, pruned_loss=0.08418, over 5665272.40 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.327, pruned_loss=0.08465, over 5784924.41 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3286, pruned_loss=0.08455, over 5651117.33 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:10:26,647 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5613, 1.8592, 1.5039, 1.7163], device='cuda:1'), covar=tensor([0.2668, 0.2640, 0.2946, 0.2621], device='cuda:1'), in_proj_covar=tensor([0.1599, 0.1148, 0.1412, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 11:10:33,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:36,173 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:10:37,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4740, 1.4233, 1.3404, 1.6449], device='cuda:1'), covar=tensor([0.0797, 0.0353, 0.0362, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 11:11:05,987 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 11:11:11,287 INFO [train.py:968] (1/2) Epoch 28, batch 16550, giga_loss[loss=0.2521, simple_loss=0.3404, pruned_loss=0.08188, over 28751.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3263, pruned_loss=0.0824, over 5678071.80 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3264, pruned_loss=0.08431, over 5787542.73 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3278, pruned_loss=0.08296, over 5662503.42 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:11:23,722 INFO [optim.py:369] (1/2) 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 11:11:55,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:12:06,880 INFO [train.py:968] (1/2) Epoch 28, batch 16600, giga_loss[loss=0.2605, simple_loss=0.3526, pruned_loss=0.08416, over 28637.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3281, pruned_loss=0.08137, over 5679856.94 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3267, pruned_loss=0.08455, over 5780161.93 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3291, pruned_loss=0.08153, over 5671572.57 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:12:52,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1247534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:13:05,982 INFO [train.py:968] (1/2) Epoch 28, batch 16650, giga_loss[loss=0.2357, simple_loss=0.3252, pruned_loss=0.07307, over 28828.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3309, pruned_loss=0.08213, over 5676388.24 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3268, pruned_loss=0.08455, over 5780782.12 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3316, pruned_loss=0.08224, over 5668845.76 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:13:18,727 INFO [optim.py:369] (1/2) 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 11:14:00,388 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8705, 2.2572, 2.2324, 1.7567], device='cuda:1'), covar=tensor([0.3298, 0.2334, 0.2454, 0.2922], device='cuda:1'), in_proj_covar=tensor([0.2026, 0.1982, 0.1881, 0.2028], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 11:14:12,634 INFO [train.py:968] (1/2) Epoch 28, batch 16700, giga_loss[loss=0.2809, simple_loss=0.3591, pruned_loss=0.1013, over 27586.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3316, pruned_loss=0.08294, over 5664200.95 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3267, pruned_loss=0.08454, over 5777560.90 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3323, pruned_loss=0.08302, over 5660168.65 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:14:16,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2434, 1.6069, 1.5377, 1.1161], device='cuda:1'), covar=tensor([0.1700, 0.2685, 0.1472, 0.1779], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0709, 0.0981, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 11:14:30,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4388, 1.5184, 1.3753, 1.5572], device='cuda:1'), covar=tensor([0.0777, 0.0369, 0.0359, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 11:14:57,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:15:00,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1247634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:15:17,496 INFO [train.py:968] (1/2) Epoch 28, batch 16750, libri_loss[loss=0.2574, simple_loss=0.3353, pruned_loss=0.08977, over 29513.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3311, pruned_loss=0.08288, over 5664971.05 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3265, pruned_loss=0.08461, over 5780508.90 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3319, pruned_loss=0.08284, over 5655742.82 frames. ], batch size: 89, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:15:30,967 INFO [optim.py:369] (1/2) 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 11:15:35,519 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:15:57,603 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1247677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:16:01,410 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 16800, giga_loss[loss=0.2089, simple_loss=0.2771, pruned_loss=0.07036, over 24624.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3307, pruned_loss=0.08249, over 5662532.83 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3267, pruned_loss=0.08469, over 5783068.40 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3313, pruned_loss=0.08235, over 5650870.68 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:16:40,990 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1247709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:17:12,980 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3810, 1.5615, 1.1847, 1.1573], device='cuda:1'), covar=tensor([0.1047, 0.0495, 0.0997, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0447, 0.0522, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 11:17:38,165 INFO [train.py:968] (1/2) Epoch 28, batch 16850, giga_loss[loss=0.2521, simple_loss=0.3319, pruned_loss=0.08616, over 27701.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3307, pruned_loss=0.08185, over 5663867.93 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3263, pruned_loss=0.08451, over 5784757.65 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3317, pruned_loss=0.08186, over 5651028.81 frames. ], batch size: 474, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:17:54,349 INFO [optim.py:369] (1/2) 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 11:18:46,172 INFO [train.py:968] (1/2) Epoch 28, batch 16900, giga_loss[loss=0.2551, simple_loss=0.3473, pruned_loss=0.08149, over 28719.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3337, pruned_loss=0.08352, over 5661466.91 frames. ], libri_tot_loss[loss=0.2479, simple_loss=0.3263, pruned_loss=0.0847, over 5778162.83 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3346, pruned_loss=0.08331, over 5652982.52 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:19:54,438 INFO [train.py:968] (1/2) Epoch 28, batch 16950, giga_loss[loss=0.292, simple_loss=0.3722, pruned_loss=0.1059, over 28199.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3354, pruned_loss=0.08403, over 5665905.28 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3259, pruned_loss=0.0845, over 5777357.35 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3365, pruned_loss=0.08401, over 5658454.89 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:19:54,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-14 11:20:08,646 INFO [optim.py:369] (1/2) 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 11:21:01,377 INFO [train.py:968] (1/2) Epoch 28, batch 17000, giga_loss[loss=0.2445, simple_loss=0.332, pruned_loss=0.07849, over 28393.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.334, pruned_loss=0.08413, over 5673545.29 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3252, pruned_loss=0.08426, over 5780890.70 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3358, pruned_loss=0.08433, over 5661740.65 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:21:33,771 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1247921.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 11:22:11,379 INFO [train.py:968] (1/2) Epoch 28, batch 17050, giga_loss[loss=0.2476, simple_loss=0.3305, pruned_loss=0.08236, over 28441.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3339, pruned_loss=0.08498, over 5681689.96 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3252, pruned_loss=0.08428, over 5784059.54 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3355, pruned_loss=0.08513, over 5667029.62 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:22:24,411 INFO [optim.py:369] (1/2) 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 11:23:22,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7008, 1.9289, 1.6156, 1.7390], device='cuda:1'), covar=tensor([0.2785, 0.2721, 0.3260, 0.2461], device='cuda:1'), in_proj_covar=tensor([0.1599, 0.1149, 0.1414, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 11:23:25,008 INFO [train.py:968] (1/2) Epoch 28, batch 17100, giga_loss[loss=0.2193, simple_loss=0.3141, pruned_loss=0.06222, over 28446.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3322, pruned_loss=0.08343, over 5678523.03 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3249, pruned_loss=0.08422, over 5786862.46 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3339, pruned_loss=0.08357, over 5662665.57 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:23:27,690 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1005, 1.2820, 5.2929, 3.8408], device='cuda:1'), covar=tensor([0.1495, 0.3072, 0.0419, 0.0870], device='cuda:1'), in_proj_covar=tensor([0.0800, 0.0670, 0.0991, 0.0965], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 11:23:34,403 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1508, 2.4419, 2.3054, 2.0540], device='cuda:1'), covar=tensor([0.2058, 0.2381, 0.2172, 0.2282], device='cuda:1'), in_proj_covar=tensor([0.0483, 0.0739, 0.0711, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 11:23:41,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 11:24:02,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4010, 1.8829, 1.6452, 1.6432], device='cuda:1'), covar=tensor([0.2162, 0.2209, 0.2234, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.0484, 0.0739, 0.0711, 0.0682], device='cuda:1'), out_proj_covar=tensor([0.0006, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 11:24:22,726 INFO [train.py:968] (1/2) Epoch 28, batch 17150, giga_loss[loss=0.2816, simple_loss=0.3691, pruned_loss=0.09704, over 28516.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3322, pruned_loss=0.08347, over 5674063.80 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3255, pruned_loss=0.08454, over 5779909.01 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3332, pruned_loss=0.08326, over 5664504.54 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:24:36,763 INFO [optim.py:369] (1/2) 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 11:25:14,009 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 17200, giga_loss[loss=0.2523, simple_loss=0.339, pruned_loss=0.0828, over 28881.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3328, pruned_loss=0.08347, over 5673283.76 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3257, pruned_loss=0.0845, over 5780402.42 frames. ], giga_tot_loss[loss=0.2502, simple_loss=0.3337, pruned_loss=0.08332, over 5662098.66 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:26:18,175 INFO [train.py:968] (1/2) Epoch 28, batch 17250, giga_loss[loss=0.2647, simple_loss=0.3432, pruned_loss=0.09315, over 28882.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3352, pruned_loss=0.08485, over 5677907.79 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3254, pruned_loss=0.0843, over 5781452.05 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3363, pruned_loss=0.08491, over 5666128.86 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:26:32,045 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 17300, giga_loss[loss=0.2093, simple_loss=0.2789, pruned_loss=0.06985, over 24386.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3311, pruned_loss=0.08361, over 5670377.73 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3251, pruned_loss=0.08417, over 5783739.69 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3324, pruned_loss=0.08377, over 5657171.26 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:27:24,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5333, 3.0996, 1.5370, 1.6367], device='cuda:1'), covar=tensor([0.0929, 0.0536, 0.0946, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0566, 0.0410, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 11:28:03,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5727, 4.4177, 4.2128, 1.7938], device='cuda:1'), covar=tensor([0.0712, 0.0835, 0.1015, 0.2224], device='cuda:1'), in_proj_covar=tensor([0.1275, 0.1171, 0.0988, 0.0734], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 11:28:13,969 INFO [train.py:968] (1/2) Epoch 28, batch 17350, giga_loss[loss=0.2471, simple_loss=0.32, pruned_loss=0.08717, over 27653.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3303, pruned_loss=0.08394, over 5655693.42 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.325, pruned_loss=0.08417, over 5770078.73 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3315, pruned_loss=0.08406, over 5655576.79 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:28:27,325 INFO [optim.py:369] (1/2) 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 11:28:34,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8755, 2.2916, 1.4455, 1.7732], device='cuda:1'), covar=tensor([0.1087, 0.0609, 0.1027, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0445, 0.0520, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 11:29:08,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248296.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 11:29:11,794 INFO [train.py:968] (1/2) Epoch 28, batch 17400, giga_loss[loss=0.2505, simple_loss=0.3408, pruned_loss=0.08008, over 28491.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3316, pruned_loss=0.08543, over 5651426.38 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3247, pruned_loss=0.08398, over 5772776.95 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3329, pruned_loss=0.0857, over 5646393.49 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:29:35,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1351, 3.9841, 3.7607, 1.7999], device='cuda:1'), covar=tensor([0.0641, 0.0751, 0.0801, 0.2144], device='cuda:1'), in_proj_covar=tensor([0.1276, 0.1170, 0.0988, 0.0733], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 11:30:03,870 INFO [train.py:968] (1/2) Epoch 28, batch 17450, giga_loss[loss=0.3393, simple_loss=0.3966, pruned_loss=0.141, over 26841.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3398, pruned_loss=0.08986, over 5665496.72 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3245, pruned_loss=0.08384, over 5776378.30 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3414, pruned_loss=0.09034, over 5655040.06 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:30:15,536 INFO [optim.py:369] (1/2) 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 11:30:46,916 INFO [train.py:968] (1/2) Epoch 28, batch 17500, libri_loss[loss=0.2561, simple_loss=0.3416, pruned_loss=0.08531, over 29520.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3465, pruned_loss=0.09353, over 5676117.58 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.3241, pruned_loss=0.08356, over 5778270.75 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.349, pruned_loss=0.09451, over 5662199.15 frames. ], batch size: 82, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:30:48,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 11:31:08,503 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:31:15,627 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9327, 1.2444, 1.3024, 1.0103], device='cuda:1'), covar=tensor([0.2080, 0.1423, 0.2528, 0.1875], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0745, 0.0716, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 11:31:21,342 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248439.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 11:31:24,285 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 17550, giga_loss[loss=0.295, simple_loss=0.3746, pruned_loss=0.1076, over 28985.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3472, pruned_loss=0.09436, over 5677623.57 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3243, pruned_loss=0.08362, over 5781428.55 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3496, pruned_loss=0.09539, over 5661079.27 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:31:42,604 INFO [optim.py:369] (1/2) 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 11:31:48,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:31:50,251 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1248471.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 11:31:56,579 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:32:16,193 INFO [train.py:968] (1/2) Epoch 28, batch 17600, libri_loss[loss=0.2206, simple_loss=0.2932, pruned_loss=0.07397, over 29646.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3431, pruned_loss=0.09346, over 5681926.12 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3244, pruned_loss=0.08372, over 5781661.09 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3452, pruned_loss=0.09432, over 5667087.87 frames. ], batch size: 73, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:32:28,707 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 11:32:42,596 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:32:58,323 INFO [train.py:968] (1/2) Epoch 28, batch 17650, giga_loss[loss=0.2496, simple_loss=0.3219, pruned_loss=0.08864, over 28943.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3359, pruned_loss=0.09007, over 5692698.35 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3245, pruned_loss=0.08355, over 5782285.29 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.338, pruned_loss=0.09114, over 5677735.99 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:33:11,698 INFO [optim.py:369] (1/2) 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 11:33:45,408 INFO [train.py:968] (1/2) Epoch 28, batch 17700, giga_loss[loss=0.2383, simple_loss=0.3012, pruned_loss=0.08773, over 26810.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3271, pruned_loss=0.08599, over 5696291.34 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.3244, pruned_loss=0.08344, over 5782495.71 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3288, pruned_loss=0.08696, over 5683515.32 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:33:55,888 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:33:58,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1248615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:34:25,231 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1248644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:34:28,023 INFO [train.py:968] (1/2) Epoch 28, batch 17750, giga_loss[loss=0.1933, simple_loss=0.2735, pruned_loss=0.05657, over 29079.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3202, pruned_loss=0.08318, over 5690798.30 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3249, pruned_loss=0.08368, over 5775838.42 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3211, pruned_loss=0.08378, over 5684163.92 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:34:32,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7336, 1.8858, 1.9357, 1.7652], device='cuda:1'), covar=tensor([0.2878, 0.2414, 0.2039, 0.2262], device='cuda:1'), in_proj_covar=tensor([0.2032, 0.1982, 0.1883, 0.2034], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 11:34:40,272 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 17800, giga_loss[loss=0.1974, simple_loss=0.2791, pruned_loss=0.05787, over 29048.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3143, pruned_loss=0.08058, over 5686975.82 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3252, pruned_loss=0.08378, over 5774658.08 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3146, pruned_loss=0.08092, over 5681461.53 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:35:23,341 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4276, 1.7819, 1.6536, 1.5116], device='cuda:1'), covar=tensor([0.2241, 0.2097, 0.2475, 0.2324], device='cuda:1'), in_proj_covar=tensor([0.0489, 0.0747, 0.0719, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 11:35:50,565 INFO [train.py:968] (1/2) Epoch 28, batch 17850, giga_loss[loss=0.2578, simple_loss=0.3366, pruned_loss=0.08947, over 28711.00 frames. ], tot_loss[loss=0.234, simple_loss=0.3104, pruned_loss=0.0788, over 5694702.52 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3252, pruned_loss=0.08369, over 5775920.89 frames. ], giga_tot_loss[loss=0.2342, simple_loss=0.3103, pruned_loss=0.07906, over 5687340.34 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:35:50,876 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8040, 2.0419, 1.9562, 1.7235], device='cuda:1'), covar=tensor([0.2940, 0.2263, 0.2072, 0.2491], device='cuda:1'), in_proj_covar=tensor([0.2033, 0.1981, 0.1882, 0.2034], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 11:36:00,740 INFO [optim.py:369] (1/2) 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 11:36:25,163 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4697, 1.6549, 1.6240, 1.4623], device='cuda:1'), covar=tensor([0.3358, 0.2902, 0.2480, 0.2926], device='cuda:1'), in_proj_covar=tensor([0.2029, 0.1977, 0.1877, 0.2029], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 11:36:33,760 INFO [train.py:968] (1/2) Epoch 28, batch 17900, giga_loss[loss=0.2839, simple_loss=0.3347, pruned_loss=0.1166, over 26631.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3081, pruned_loss=0.0779, over 5690715.19 frames. ], libri_tot_loss[loss=0.2463, simple_loss=0.3252, pruned_loss=0.08374, over 5774784.95 frames. ], giga_tot_loss[loss=0.2318, simple_loss=0.3077, pruned_loss=0.07796, over 5684159.96 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:36:35,403 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:37:02,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6641, 1.9672, 1.5913, 1.6321], device='cuda:1'), covar=tensor([0.2818, 0.2814, 0.3342, 0.2628], device='cuda:1'), in_proj_covar=tensor([0.1597, 0.1149, 0.1411, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 11:37:15,636 INFO [train.py:968] (1/2) Epoch 28, batch 17950, libri_loss[loss=0.2784, simple_loss=0.3675, pruned_loss=0.09467, over 29504.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.3061, pruned_loss=0.07718, over 5691692.88 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3255, pruned_loss=0.08381, over 5779532.05 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3048, pruned_loss=0.07694, over 5679545.16 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:37:18,736 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1248851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:37:18,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:37:26,638 INFO [optim.py:369] (1/2) 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 11:37:57,387 INFO [train.py:968] (1/2) Epoch 28, batch 18000, giga_loss[loss=0.2236, simple_loss=0.3025, pruned_loss=0.07235, over 28847.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3023, pruned_loss=0.0752, over 5697144.86 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3259, pruned_loss=0.08396, over 5772398.94 frames. ], giga_tot_loss[loss=0.2251, simple_loss=0.3006, pruned_loss=0.07476, over 5691801.35 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:37:57,388 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 11:38:05,501 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 11:38:11,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1248905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:38:44,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:38:46,969 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1248947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:38:47,931 INFO [train.py:968] (1/2) Epoch 28, batch 18050, giga_loss[loss=0.1915, simple_loss=0.2666, pruned_loss=0.05818, over 27673.00 frames. ], tot_loss[loss=0.2231, simple_loss=0.2991, pruned_loss=0.07357, over 5692893.34 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3256, pruned_loss=0.0837, over 5773059.38 frames. ], giga_tot_loss[loss=0.2221, simple_loss=0.2976, pruned_loss=0.07329, over 5686658.38 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:39:02,039 INFO [optim.py:369] (1/2) 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 11:39:04,750 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.80 vs. limit=5.0
+2023-03-14 11:39:13,477 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1248976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:39:30,365 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1248994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:39:33,200 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 18100, giga_loss[loss=0.1902, simple_loss=0.2681, pruned_loss=0.05621, over 29004.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2968, pruned_loss=0.07293, over 5689693.61 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3258, pruned_loss=0.08382, over 5773762.84 frames. ], giga_tot_loss[loss=0.2199, simple_loss=0.295, pruned_loss=0.07244, over 5682826.74 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:39:56,074 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:40:16,610 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1249048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:40:16,902 INFO [train.py:968] (1/2) Epoch 28, batch 18150, giga_loss[loss=0.2117, simple_loss=0.287, pruned_loss=0.06822, over 28960.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2956, pruned_loss=0.07227, over 5702002.35 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3264, pruned_loss=0.08405, over 5777916.72 frames. ], giga_tot_loss[loss=0.2176, simple_loss=0.2927, pruned_loss=0.07132, over 5690272.59 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:40:19,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1249051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:40:28,897 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-14 11:40:30,470 INFO [optim.py:369] (1/2) 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 11:40:46,550 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:40:46,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5256, 3.7144, 1.8158, 1.6988], device='cuda:1'), covar=tensor([0.1015, 0.0339, 0.0861, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0564, 0.0408, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 11:41:00,035 INFO [train.py:968] (1/2) Epoch 28, batch 18200, libri_loss[loss=0.2363, simple_loss=0.318, pruned_loss=0.07736, over 29552.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2937, pruned_loss=0.07131, over 5708108.75 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3267, pruned_loss=0.08396, over 5778187.64 frames. ], giga_tot_loss[loss=0.2147, simple_loss=0.2894, pruned_loss=0.07001, over 5694565.77 frames. ], batch size: 79, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:41:27,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4489, 2.0693, 1.6086, 0.7833], device='cuda:1'), covar=tensor([0.7895, 0.3638, 0.4329, 0.7511], device='cuda:1'), in_proj_covar=tensor([0.1845, 0.1741, 0.1653, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 11:41:47,368 INFO [train.py:968] (1/2) Epoch 28, batch 18250, libri_loss[loss=0.2492, simple_loss=0.3347, pruned_loss=0.08191, over 29670.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2964, pruned_loss=0.0732, over 5703350.48 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3271, pruned_loss=0.08405, over 5780252.39 frames. ], giga_tot_loss[loss=0.2179, simple_loss=0.2921, pruned_loss=0.07187, over 5689267.28 frames. ], batch size: 88, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:42:01,794 INFO [optim.py:369] (1/2) 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 11:42:36,074 INFO [train.py:968] (1/2) Epoch 28, batch 18300, giga_loss[loss=0.2702, simple_loss=0.3528, pruned_loss=0.09386, over 29013.00 frames. ], tot_loss[loss=0.2323, simple_loss=0.3077, pruned_loss=0.07851, over 5706173.92 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3273, pruned_loss=0.08399, over 5783948.15 frames. ], giga_tot_loss[loss=0.2289, simple_loss=0.3033, pruned_loss=0.07728, over 5689656.49 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:43:00,340 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1249226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:43:12,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3309, 3.1423, 2.9823, 1.3932], device='cuda:1'), covar=tensor([0.1012, 0.1174, 0.1081, 0.2434], device='cuda:1'), in_proj_covar=tensor([0.1284, 0.1183, 0.0998, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 11:43:20,583 INFO [train.py:968] (1/2) Epoch 28, batch 18350, giga_loss[loss=0.2681, simple_loss=0.346, pruned_loss=0.09509, over 28892.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3207, pruned_loss=0.08533, over 5705380.98 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3272, pruned_loss=0.08386, over 5784735.13 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.3172, pruned_loss=0.08444, over 5690659.96 frames. ], batch size: 66, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:43:32,644 INFO [optim.py:369] (1/2) 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 11:43:48,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5027, 2.0825, 1.6756, 0.8074], device='cuda:1'), covar=tensor([0.6339, 0.3501, 0.4487, 0.6850], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1737, 0.1652, 0.1500], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 11:43:56,743 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4291, 2.6578, 2.5868, 2.2085], device='cuda:1'), covar=tensor([0.2061, 0.2122, 0.1834, 0.2106], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0749, 0.0721, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 11:44:02,096 INFO [train.py:968] (1/2) Epoch 28, batch 18400, giga_loss[loss=0.2871, simple_loss=0.3643, pruned_loss=0.105, over 27919.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3301, pruned_loss=0.0897, over 5707903.37 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3267, pruned_loss=0.08349, over 5787514.06 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3277, pruned_loss=0.08942, over 5692294.11 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:44:45,836 INFO [train.py:968] (1/2) Epoch 28, batch 18450, giga_loss[loss=0.2429, simple_loss=0.3318, pruned_loss=0.07702, over 28246.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3352, pruned_loss=0.09102, over 5702573.27 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.327, pruned_loss=0.08357, over 5782909.01 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3332, pruned_loss=0.09086, over 5692663.34 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:44:58,749 INFO [optim.py:369] (1/2) 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 11:45:03,086 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1249369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:45:05,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1249372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:45:27,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.45 vs. limit=2.0
+2023-03-14 11:45:27,784 INFO [train.py:968] (1/2) Epoch 28, batch 18500, giga_loss[loss=0.2503, simple_loss=0.3176, pruned_loss=0.09155, over 23478.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3373, pruned_loss=0.0909, over 5701048.84 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3275, pruned_loss=0.08364, over 5785816.05 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3355, pruned_loss=0.09092, over 5688196.50 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:45:31,671 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:45:36,815 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1249407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:46:16,036 INFO [train.py:968] (1/2) Epoch 28, batch 18550, giga_loss[loss=0.2857, simple_loss=0.3544, pruned_loss=0.1085, over 28892.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.339, pruned_loss=0.09166, over 5694549.86 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3276, pruned_loss=0.08363, over 5786772.80 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3376, pruned_loss=0.09179, over 5682393.01 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:46:29,739 INFO [optim.py:369] (1/2) 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 11:46:48,408 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2543, 1.5106, 1.5323, 1.3297], device='cuda:1'), covar=tensor([0.2240, 0.1846, 0.2506, 0.1951], device='cuda:1'), in_proj_covar=tensor([0.0492, 0.0748, 0.0719, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 11:46:50,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2029, 0.9010, 1.0672, 1.3721], device='cuda:1'), covar=tensor([0.0786, 0.0484, 0.0365, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 11:46:57,673 INFO [train.py:968] (1/2) Epoch 28, batch 18600, giga_loss[loss=0.2827, simple_loss=0.3508, pruned_loss=0.1074, over 28902.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.3416, pruned_loss=0.0936, over 5694100.98 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3283, pruned_loss=0.08391, over 5779769.93 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3402, pruned_loss=0.09369, over 5687920.49 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:47:39,291 INFO [train.py:968] (1/2) Epoch 28, batch 18650, giga_loss[loss=0.3138, simple_loss=0.3784, pruned_loss=0.1246, over 27626.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3441, pruned_loss=0.09537, over 5700506.06 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3289, pruned_loss=0.08426, over 5785451.74 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3433, pruned_loss=0.09567, over 5685870.13 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:47:53,504 INFO [optim.py:369] (1/2) 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 11:48:20,853 INFO [train.py:968] (1/2) Epoch 28, batch 18700, giga_loss[loss=0.3149, simple_loss=0.3865, pruned_loss=0.1216, over 27899.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3464, pruned_loss=0.09622, over 5705922.88 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3288, pruned_loss=0.08398, over 5788170.44 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3463, pruned_loss=0.09708, over 5689293.54 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:49:01,023 INFO [train.py:968] (1/2) Epoch 28, batch 18750, giga_loss[loss=0.295, simple_loss=0.3701, pruned_loss=0.1099, over 28883.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.35, pruned_loss=0.09756, over 5714650.80 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.329, pruned_loss=0.08405, over 5790573.27 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3501, pruned_loss=0.09844, over 5697771.40 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:49:03,610 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 28, batch 18800, giga_loss[loss=0.2736, simple_loss=0.3572, pruned_loss=0.09494, over 28723.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3505, pruned_loss=0.09691, over 5712497.81 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3289, pruned_loss=0.08384, over 5790667.85 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3512, pruned_loss=0.09815, over 5696953.57 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:50:14,764 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8381, 1.0837, 2.9502, 2.7737], device='cuda:1'), covar=tensor([0.1705, 0.2639, 0.0623, 0.1019], device='cuda:1'), in_proj_covar=tensor([0.0799, 0.0671, 0.0995, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 11:50:18,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1711, 1.2973, 1.0832, 0.9482], device='cuda:1'), covar=tensor([0.1196, 0.0579, 0.1224, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0446, 0.0520, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 11:50:27,194 INFO [train.py:968] (1/2) Epoch 28, batch 18850, giga_loss[loss=0.2634, simple_loss=0.3423, pruned_loss=0.09224, over 28826.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3513, pruned_loss=0.09671, over 5709722.46 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3288, pruned_loss=0.08378, over 5792773.65 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3524, pruned_loss=0.09799, over 5694026.37 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:50:31,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1249753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:50:42,649 INFO [optim.py:369] (1/2) 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 11:50:54,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1249782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:50:59,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3167, 4.1494, 3.9180, 2.0093], device='cuda:1'), covar=tensor([0.0587, 0.0722, 0.0729, 0.1980], device='cuda:1'), in_proj_covar=tensor([0.1273, 0.1174, 0.0991, 0.0738], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 11:51:09,389 INFO [train.py:968] (1/2) Epoch 28, batch 18900, giga_loss[loss=0.2639, simple_loss=0.339, pruned_loss=0.09438, over 27499.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3503, pruned_loss=0.09503, over 5709471.75 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3288, pruned_loss=0.08378, over 5795771.81 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3516, pruned_loss=0.09632, over 5692307.86 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:51:48,832 INFO [train.py:968] (1/2) Epoch 28, batch 18950, libri_loss[loss=0.2628, simple_loss=0.3468, pruned_loss=0.08938, over 29535.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.349, pruned_loss=0.0934, over 5718641.87 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3292, pruned_loss=0.08383, over 5798488.28 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3503, pruned_loss=0.09462, over 5700515.55 frames. ], batch size: 83, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:52:05,016 INFO [optim.py:369] (1/2) 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 11:52:29,903 INFO [train.py:968] (1/2) Epoch 28, batch 19000, giga_loss[loss=0.2908, simple_loss=0.3704, pruned_loss=0.1056, over 29013.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.35, pruned_loss=0.09436, over 5713602.70 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.3294, pruned_loss=0.08382, over 5800374.13 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3512, pruned_loss=0.09553, over 5695922.19 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:52:54,649 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1249925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:52:57,343 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1249928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:53:16,625 INFO [train.py:968] (1/2) Epoch 28, batch 19050, giga_loss[loss=0.2896, simple_loss=0.3654, pruned_loss=0.1069, over 28362.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3522, pruned_loss=0.09875, over 5700179.17 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3292, pruned_loss=0.08372, over 5800672.69 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3537, pruned_loss=0.09999, over 5684606.99 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:53:20,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1313, 1.1694, 3.4206, 2.9443], device='cuda:1'), covar=tensor([0.1834, 0.2986, 0.0588, 0.1097], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0672, 0.0999, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 11:53:22,800 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1249957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:53:32,196 INFO [optim.py:369] (1/2) 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 11:54:00,207 INFO [train.py:968] (1/2) Epoch 28, batch 19100, giga_loss[loss=0.2486, simple_loss=0.3344, pruned_loss=0.08142, over 28206.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.353, pruned_loss=0.1011, over 5700415.25 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.329, pruned_loss=0.08357, over 5802711.13 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3548, pruned_loss=0.1025, over 5684404.96 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:54:20,498 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 19150, giga_loss[loss=0.2904, simple_loss=0.3684, pruned_loss=0.1062, over 29013.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3527, pruned_loss=0.102, over 5706345.96 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3291, pruned_loss=0.08353, over 5805073.26 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3547, pruned_loss=0.1036, over 5688900.37 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:54:54,360 INFO [optim.py:369] (1/2) 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 11:55:01,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7249, 1.7740, 1.8897, 1.5082], device='cuda:1'), covar=tensor([0.1791, 0.2458, 0.1433, 0.1673], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0712, 0.0984, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 11:55:23,884 INFO [train.py:968] (1/2) Epoch 28, batch 19200, giga_loss[loss=0.2417, simple_loss=0.3139, pruned_loss=0.08475, over 28597.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3501, pruned_loss=0.1009, over 5700310.53 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3292, pruned_loss=0.08352, over 5794352.25 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3521, pruned_loss=0.1027, over 5693248.79 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 11:55:28,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4477, 1.7117, 1.4270, 1.3200], device='cuda:1'), covar=tensor([0.2433, 0.2458, 0.2659, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.1594, 0.1148, 0.1408, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 11:55:29,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4365, 1.6789, 1.2157, 1.2803], device='cuda:1'), covar=tensor([0.1149, 0.0595, 0.1186, 0.1132], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0447, 0.0521, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 11:55:31,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.19 vs. limit=5.0
+2023-03-14 11:55:33,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3450, 1.3574, 1.2153, 1.6050], device='cuda:1'), covar=tensor([0.0785, 0.0364, 0.0351, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 11:55:49,314 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1250128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:55:52,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9479, 1.1375, 1.1044, 0.8809], device='cuda:1'), covar=tensor([0.2471, 0.2927, 0.1755, 0.2500], device='cuda:1'), in_proj_covar=tensor([0.2045, 0.2002, 0.1901, 0.2056], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 11:56:06,199 INFO [train.py:968] (1/2) Epoch 28, batch 19250, giga_loss[loss=0.2901, simple_loss=0.3548, pruned_loss=0.1127, over 28083.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3492, pruned_loss=0.09981, over 5685170.24 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3296, pruned_loss=0.08356, over 5785718.32 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.351, pruned_loss=0.1017, over 5684938.18 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:56:18,801 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5590, 4.1095, 1.7605, 1.6015], device='cuda:1'), covar=tensor([0.0975, 0.0296, 0.0858, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0565, 0.0407, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 11:56:21,183 INFO [optim.py:369] (1/2) 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 11:56:23,175 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1250169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:56:25,332 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1250172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:56:46,474 INFO [train.py:968] (1/2) Epoch 28, batch 19300, libri_loss[loss=0.2111, simple_loss=0.2937, pruned_loss=0.06424, over 29497.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3475, pruned_loss=0.09812, over 5685662.93 frames. ], libri_tot_loss[loss=0.2482, simple_loss=0.3297, pruned_loss=0.08341, over 5780331.81 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3496, pruned_loss=0.1003, over 5686675.60 frames. ], batch size: 70, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:56:47,991 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1250201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:57:07,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4883, 1.6054, 1.1788, 1.1598], device='cuda:1'), covar=tensor([0.1068, 0.0616, 0.1081, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0447, 0.0521, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 11:57:20,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 11:57:31,161 INFO [train.py:968] (1/2) Epoch 28, batch 19350, giga_loss[loss=0.2489, simple_loss=0.3134, pruned_loss=0.09218, over 23659.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3447, pruned_loss=0.09607, over 5678579.76 frames. ], libri_tot_loss[loss=0.2485, simple_loss=0.33, pruned_loss=0.08351, over 5775551.60 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3465, pruned_loss=0.09812, over 5681058.12 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 11:57:45,598 INFO [optim.py:369] (1/2) 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 11:57:49,055 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1250271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:57:52,300 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1250274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:58:09,805 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1666, 1.6168, 1.1906, 0.4550], device='cuda:1'), covar=tensor([0.4586, 0.2541, 0.3894, 0.6499], device='cuda:1'), in_proj_covar=tensor([0.1834, 0.1725, 0.1642, 0.1492], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 11:58:14,476 INFO [train.py:968] (1/2) Epoch 28, batch 19400, giga_loss[loss=0.2416, simple_loss=0.3165, pruned_loss=0.08335, over 28779.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3395, pruned_loss=0.0932, over 5682703.59 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.33, pruned_loss=0.08342, over 5778188.03 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3413, pruned_loss=0.09511, over 5680791.23 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:58:17,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-14 11:58:19,255 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1250303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 11:58:58,751 INFO [train.py:968] (1/2) Epoch 28, batch 19450, giga_loss[loss=0.2221, simple_loss=0.3005, pruned_loss=0.07188, over 28799.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3353, pruned_loss=0.09131, over 5683545.21 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3304, pruned_loss=0.08366, over 5777690.53 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3365, pruned_loss=0.09287, over 5679822.83 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 11:59:17,433 INFO [optim.py:369] (1/2) 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 11:59:18,997 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-14 11:59:46,120 INFO [train.py:968] (1/2) Epoch 28, batch 19500, giga_loss[loss=0.2479, simple_loss=0.3228, pruned_loss=0.08653, over 28849.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3313, pruned_loss=0.08928, over 5690149.21 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3312, pruned_loss=0.084, over 5781578.90 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3316, pruned_loss=0.0904, over 5681017.91 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:00:04,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5381, 1.7074, 1.2395, 1.3338], device='cuda:1'), covar=tensor([0.1070, 0.0616, 0.1126, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0447, 0.0521, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 12:00:31,252 INFO [train.py:968] (1/2) Epoch 28, batch 19550, giga_loss[loss=0.2399, simple_loss=0.3248, pruned_loss=0.07748, over 28552.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.333, pruned_loss=0.09, over 5683408.23 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3319, pruned_loss=0.08434, over 5772504.06 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.3327, pruned_loss=0.0908, over 5680045.91 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:00:47,272 INFO [optim.py:369] (1/2) 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 12:01:05,601 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4263, 1.6523, 1.6955, 1.4897], device='cuda:1'), covar=tensor([0.2350, 0.2207, 0.2700, 0.2304], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0753, 0.0726, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 12:01:14,650 INFO [train.py:968] (1/2) Epoch 28, batch 19600, giga_loss[loss=0.2342, simple_loss=0.3227, pruned_loss=0.07282, over 28943.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3332, pruned_loss=0.08933, over 5699410.25 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3321, pruned_loss=0.08439, over 5773695.25 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3327, pruned_loss=0.09001, over 5694136.47 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:01:58,437 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3765, 3.0478, 1.4279, 1.5845], device='cuda:1'), covar=tensor([0.1085, 0.0347, 0.0934, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0563, 0.0407, 0.0444], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 12:01:58,773 INFO [train.py:968] (1/2) Epoch 28, batch 19650, giga_loss[loss=0.2419, simple_loss=0.321, pruned_loss=0.08135, over 29079.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.332, pruned_loss=0.0889, over 5699932.94 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3324, pruned_loss=0.08451, over 5772672.77 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3314, pruned_loss=0.08934, over 5696383.80 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:02:02,293 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-14 12:02:15,317 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 19700, giga_loss[loss=0.2413, simple_loss=0.3141, pruned_loss=0.08418, over 28723.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.331, pruned_loss=0.08846, over 5711254.60 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3333, pruned_loss=0.08474, over 5773962.10 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3296, pruned_loss=0.08877, over 5704626.36 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:03:19,924 INFO [train.py:968] (1/2) Epoch 28, batch 19750, giga_loss[loss=0.2447, simple_loss=0.3103, pruned_loss=0.08957, over 28528.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3285, pruned_loss=0.08751, over 5715222.94 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3336, pruned_loss=0.08477, over 5770876.09 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3271, pruned_loss=0.08778, over 5711820.06 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:03:29,972 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5708, 2.2004, 1.6469, 0.8070], device='cuda:1'), covar=tensor([0.7378, 0.3186, 0.4788, 0.7774], device='cuda:1'), in_proj_covar=tensor([0.1825, 0.1715, 0.1639, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 12:03:34,218 INFO [optim.py:369] (1/2) 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 12:03:40,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0292, 3.0767, 2.0361, 1.1259], device='cuda:1'), covar=tensor([0.9477, 0.3420, 0.4687, 0.8186], device='cuda:1'), in_proj_covar=tensor([0.1825, 0.1715, 0.1638, 0.1486], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 12:03:58,517 INFO [train.py:968] (1/2) Epoch 28, batch 19800, libri_loss[loss=0.3137, simple_loss=0.3933, pruned_loss=0.1171, over 29653.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3265, pruned_loss=0.08668, over 5720062.22 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.334, pruned_loss=0.08486, over 5775010.54 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3248, pruned_loss=0.08685, over 5712277.41 frames. ], batch size: 91, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:04:00,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2310, 1.6980, 1.2262, 0.6475], device='cuda:1'), covar=tensor([0.5617, 0.2364, 0.3503, 0.6997], device='cuda:1'), in_proj_covar=tensor([0.1823, 0.1712, 0.1635, 0.1484], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 12:04:38,665 INFO [train.py:968] (1/2) Epoch 28, batch 19850, giga_loss[loss=0.2239, simple_loss=0.3013, pruned_loss=0.07328, over 28749.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.325, pruned_loss=0.08621, over 5717803.13 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3345, pruned_loss=0.08496, over 5767466.28 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3232, pruned_loss=0.08628, over 5717344.55 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:04:54,947 INFO [optim.py:369] (1/2) 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 12:05:20,640 INFO [train.py:968] (1/2) Epoch 28, batch 19900, giga_loss[loss=0.2256, simple_loss=0.3075, pruned_loss=0.07185, over 28896.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3226, pruned_loss=0.08506, over 5717744.51 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3345, pruned_loss=0.08487, over 5768824.42 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3211, pruned_loss=0.08519, over 5715904.99 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:05:58,576 INFO [train.py:968] (1/2) Epoch 28, batch 19950, giga_loss[loss=0.2226, simple_loss=0.2973, pruned_loss=0.07399, over 28470.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3219, pruned_loss=0.08482, over 5718916.84 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3351, pruned_loss=0.08482, over 5771530.35 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3197, pruned_loss=0.08497, over 5713496.29 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:06:14,322 INFO [optim.py:369] (1/2) 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 12:06:40,206 INFO [train.py:968] (1/2) Epoch 28, batch 20000, giga_loss[loss=0.2353, simple_loss=0.3048, pruned_loss=0.08294, over 28974.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.3202, pruned_loss=0.08385, over 5722359.23 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3355, pruned_loss=0.08486, over 5767997.17 frames. ], giga_tot_loss[loss=0.243, simple_loss=0.318, pruned_loss=0.08394, over 5720403.05 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:07:18,843 INFO [train.py:968] (1/2) Epoch 28, batch 20050, libri_loss[loss=0.3072, simple_loss=0.3895, pruned_loss=0.1124, over 20631.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3187, pruned_loss=0.08312, over 5718277.46 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3361, pruned_loss=0.08501, over 5762180.14 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.316, pruned_loss=0.08302, over 5721703.94 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:07:32,101 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 20100, giga_loss[loss=0.2313, simple_loss=0.3178, pruned_loss=0.07237, over 28931.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3187, pruned_loss=0.08315, over 5727372.40 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3364, pruned_loss=0.08501, over 5761639.43 frames. ], giga_tot_loss[loss=0.241, simple_loss=0.316, pruned_loss=0.08303, over 5729756.96 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:08:40,250 INFO [train.py:968] (1/2) Epoch 28, batch 20150, giga_loss[loss=0.2786, simple_loss=0.355, pruned_loss=0.1011, over 28543.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3233, pruned_loss=0.08595, over 5722042.81 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3373, pruned_loss=0.08519, over 5763861.03 frames. ], giga_tot_loss[loss=0.2456, simple_loss=0.3198, pruned_loss=0.08567, over 5721004.43 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:08:55,554 INFO [optim.py:369] (1/2) 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 12:09:23,055 INFO [train.py:968] (1/2) Epoch 28, batch 20200, giga_loss[loss=0.3008, simple_loss=0.3683, pruned_loss=0.1166, over 27638.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3287, pruned_loss=0.08918, over 5703141.37 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3381, pruned_loss=0.08562, over 5747434.71 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3248, pruned_loss=0.08864, over 5714726.02 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:09:41,780 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:10:12,453 INFO [train.py:968] (1/2) Epoch 28, batch 20250, giga_loss[loss=0.2658, simple_loss=0.3455, pruned_loss=0.09309, over 28989.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3373, pruned_loss=0.09523, over 5688752.76 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.338, pruned_loss=0.08554, over 5750529.20 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3342, pruned_loss=0.09502, over 5694142.36 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:10:30,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3922, 1.6546, 1.5601, 1.4360], device='cuda:1'), covar=tensor([0.1997, 0.2098, 0.2379, 0.2124], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0758, 0.0730, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 12:10:31,922 INFO [optim.py:369] (1/2) 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 12:10:55,252 INFO [train.py:968] (1/2) Epoch 28, batch 20300, giga_loss[loss=0.3135, simple_loss=0.3852, pruned_loss=0.121, over 28966.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3417, pruned_loss=0.0974, over 5678640.18 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3381, pruned_loss=0.08567, over 5741350.40 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3391, pruned_loss=0.09738, over 5689802.97 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:11:16,070 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:11:25,839 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:11:42,991 INFO [train.py:968] (1/2) Epoch 28, batch 20350, giga_loss[loss=0.2976, simple_loss=0.3551, pruned_loss=0.12, over 23367.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3452, pruned_loss=0.0987, over 5668265.52 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3377, pruned_loss=0.08551, over 5743640.20 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3435, pruned_loss=0.09904, over 5674131.34 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:11:59,362 INFO [optim.py:369] (1/2) 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 12:12:28,632 INFO [train.py:968] (1/2) Epoch 28, batch 20400, giga_loss[loss=0.2645, simple_loss=0.354, pruned_loss=0.08752, over 28747.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.349, pruned_loss=0.1002, over 5668076.13 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.338, pruned_loss=0.08562, over 5746975.45 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3476, pruned_loss=0.1007, over 5668524.29 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:12:43,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3939, 3.1837, 1.5489, 1.4999], device='cuda:1'), covar=tensor([0.0992, 0.0359, 0.0922, 0.1398], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0566, 0.0409, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 12:13:12,160 INFO [train.py:968] (1/2) Epoch 28, batch 20450, giga_loss[loss=0.2655, simple_loss=0.3466, pruned_loss=0.09222, over 28717.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3536, pruned_loss=0.1029, over 5669506.95 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.338, pruned_loss=0.08557, over 5749721.69 frames. ], giga_tot_loss[loss=0.2799, simple_loss=0.3527, pruned_loss=0.1035, over 5666332.93 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:13:31,900 INFO [optim.py:369] (1/2) 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 12:13:38,752 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1696, 1.7646, 1.4723, 1.5279], device='cuda:1'), covar=tensor([0.0800, 0.0281, 0.0286, 0.1004], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 12:13:52,353 INFO [train.py:968] (1/2) Epoch 28, batch 20500, giga_loss[loss=0.231, simple_loss=0.3128, pruned_loss=0.07461, over 28618.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3487, pruned_loss=0.09916, over 5683612.57 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3378, pruned_loss=0.08566, over 5754936.64 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3487, pruned_loss=0.1002, over 5672985.32 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:14:34,290 INFO [train.py:968] (1/2) Epoch 28, batch 20550, giga_loss[loss=0.238, simple_loss=0.3189, pruned_loss=0.0785, over 28351.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.09739, over 5687766.21 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3377, pruned_loss=0.0857, over 5747301.29 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3473, pruned_loss=0.09837, over 5684565.98 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:14:46,379 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 12:14:53,186 INFO [optim.py:369] (1/2) 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 12:14:57,722 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-14 12:15:06,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.66 vs. limit=2.0
+2023-03-14 12:15:10,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:15:17,245 INFO [train.py:968] (1/2) Epoch 28, batch 20600, giga_loss[loss=0.2585, simple_loss=0.3437, pruned_loss=0.0867, over 28788.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3469, pruned_loss=0.09689, over 5686028.32 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.338, pruned_loss=0.08584, over 5743129.68 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3471, pruned_loss=0.09786, over 5685349.52 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:15:25,175 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:15:48,705 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:15:57,896 INFO [train.py:968] (1/2) Epoch 28, batch 20650, giga_loss[loss=0.2777, simple_loss=0.356, pruned_loss=0.09967, over 28760.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3479, pruned_loss=0.09708, over 5674879.45 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.338, pruned_loss=0.08587, over 5729002.14 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3483, pruned_loss=0.09811, over 5684899.00 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:16:17,323 INFO [optim.py:369] (1/2) 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 12:16:39,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:16:42,932 INFO [train.py:968] (1/2) Epoch 28, batch 20700, giga_loss[loss=0.2779, simple_loss=0.3564, pruned_loss=0.09971, over 29063.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3513, pruned_loss=0.09988, over 5681957.23 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3388, pruned_loss=0.08658, over 5729307.83 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3512, pruned_loss=0.1004, over 5688308.00 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:16:47,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:17:16,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1251635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:17:18,724 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1251638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:17:27,213 INFO [train.py:968] (1/2) Epoch 28, batch 20750, giga_loss[loss=0.2561, simple_loss=0.3291, pruned_loss=0.09161, over 28591.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.353, pruned_loss=0.1012, over 5685863.04 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3387, pruned_loss=0.08654, over 5719636.84 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3532, pruned_loss=0.1019, over 5698035.09 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 12:17:47,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:18:16,215 INFO [train.py:968] (1/2) Epoch 28, batch 20800, giga_loss[loss=0.2525, simple_loss=0.3356, pruned_loss=0.08466, over 28947.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3546, pruned_loss=0.103, over 5665173.49 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3391, pruned_loss=0.08661, over 5713177.70 frames. ], giga_tot_loss[loss=0.2814, simple_loss=0.3549, pruned_loss=0.104, over 5679044.25 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:18:51,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1251737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:18:53,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1251740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:18:59,862 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1251748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:00,173 INFO [train.py:968] (1/2) Epoch 28, batch 20850, giga_loss[loss=0.2911, simple_loss=0.3621, pruned_loss=0.1101, over 28874.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3561, pruned_loss=0.1049, over 5674432.27 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3391, pruned_loss=0.08661, over 5715179.79 frames. ], giga_tot_loss[loss=0.2841, simple_loss=0.3565, pruned_loss=0.1058, over 5683145.75 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:19:02,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1251751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:17,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-14 12:19:19,455 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1251769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:20,560 INFO [optim.py:369] (1/2) 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 12:19:28,006 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1251780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:19:38,032 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 12:19:43,401 INFO [train.py:968] (1/2) Epoch 28, batch 20900, giga_loss[loss=0.264, simple_loss=0.346, pruned_loss=0.09099, over 28868.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.356, pruned_loss=0.1047, over 5686701.58 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3394, pruned_loss=0.08676, over 5716772.07 frames. ], giga_tot_loss[loss=0.2836, simple_loss=0.3562, pruned_loss=0.1055, over 5691799.83 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:20:12,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4911, 3.5744, 1.6071, 1.6208], device='cuda:1'), covar=tensor([0.1073, 0.0262, 0.0926, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0566, 0.0409, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 12:20:21,510 INFO [train.py:968] (1/2) Epoch 28, batch 20950, giga_loss[loss=0.287, simple_loss=0.3635, pruned_loss=0.1052, over 28489.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3541, pruned_loss=0.1025, over 5691107.96 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3397, pruned_loss=0.08706, over 5720214.35 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3545, pruned_loss=0.1034, over 5691387.20 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:20:38,680 INFO [optim.py:369] (1/2) 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 12:20:44,994 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1251877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:20:49,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:20:56,191 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-14 12:21:01,999 INFO [train.py:968] (1/2) Epoch 28, batch 21000, giga_loss[loss=0.2654, simple_loss=0.3463, pruned_loss=0.0922, over 28695.00 frames. ], tot_loss[loss=0.2794, simple_loss=0.3552, pruned_loss=0.1018, over 5690140.45 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3403, pruned_loss=0.08747, over 5714577.04 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3553, pruned_loss=0.1025, over 5694672.35 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:21:01,999 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 12:21:11,421 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 12:21:21,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1251911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:21:51,704 INFO [train.py:968] (1/2) Epoch 28, batch 21050, giga_loss[loss=0.2709, simple_loss=0.3405, pruned_loss=0.1007, over 28649.00 frames. ], tot_loss[loss=0.2776, simple_loss=0.3536, pruned_loss=0.1008, over 5693955.43 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.0876, over 5719583.23 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3539, pruned_loss=0.1016, over 5692338.27 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:22:07,882 INFO [optim.py:369] (1/2) 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 12:22:21,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-14 12:22:29,155 INFO [train.py:968] (1/2) Epoch 28, batch 21100, giga_loss[loss=0.2918, simple_loss=0.3631, pruned_loss=0.1102, over 28588.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3513, pruned_loss=0.0996, over 5708432.87 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08808, over 5722742.65 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3514, pruned_loss=0.1001, over 5703848.19 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:22:36,052 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252008.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:22:53,144 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:22:54,963 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:01,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0177, 2.2199, 1.9525, 1.8665], device='cuda:1'), covar=tensor([0.2280, 0.2457, 0.2493, 0.2547], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0757, 0.0728, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 12:23:10,036 INFO [train.py:968] (1/2) Epoch 28, batch 21150, giga_loss[loss=0.2528, simple_loss=0.333, pruned_loss=0.08627, over 28633.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3498, pruned_loss=0.09905, over 5712947.18 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3416, pruned_loss=0.08878, over 5726454.71 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3494, pruned_loss=0.09906, over 5705832.40 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:23:15,053 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:17,306 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:18,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:26,819 INFO [optim.py:369] (1/2) 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 12:23:37,782 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:23:48,775 INFO [train.py:968] (1/2) Epoch 28, batch 21200, giga_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 28692.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3473, pruned_loss=0.09798, over 5704896.36 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3416, pruned_loss=0.08905, over 5720300.80 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3473, pruned_loss=0.09813, over 5704459.29 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:24:28,108 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252147.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:24:29,287 INFO [train.py:968] (1/2) Epoch 28, batch 21250, giga_loss[loss=0.2799, simple_loss=0.3618, pruned_loss=0.09905, over 28710.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3483, pruned_loss=0.09926, over 5707412.59 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3417, pruned_loss=0.08936, over 5725072.54 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3484, pruned_loss=0.09942, over 5702236.27 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:24:51,915 INFO [optim.py:369] (1/2) 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 12:25:12,825 INFO [train.py:968] (1/2) Epoch 28, batch 21300, giga_loss[loss=0.2982, simple_loss=0.3658, pruned_loss=0.1153, over 27537.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3488, pruned_loss=0.09901, over 5712898.64 frames. ], libri_tot_loss[loss=0.261, simple_loss=0.3423, pruned_loss=0.08982, over 5727811.50 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3484, pruned_loss=0.09886, over 5706170.36 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:25:34,538 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-14 12:25:36,313 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:25:55,051 INFO [train.py:968] (1/2) Epoch 28, batch 21350, giga_loss[loss=0.263, simple_loss=0.3367, pruned_loss=0.09462, over 28835.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3473, pruned_loss=0.09737, over 5710242.66 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3421, pruned_loss=0.08975, over 5730546.14 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3474, pruned_loss=0.09747, over 5702241.79 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:25:57,140 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:26:10,287 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:26:31,753 INFO [train.py:968] (1/2) Epoch 28, batch 21400, giga_loss[loss=0.2301, simple_loss=0.3103, pruned_loss=0.07493, over 28919.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3459, pruned_loss=0.09608, over 5723398.54 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3419, pruned_loss=0.08998, over 5735721.40 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3463, pruned_loss=0.09617, over 5711625.80 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:26:53,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5379, 2.5882, 2.3982, 2.2241], device='cuda:1'), covar=tensor([0.2037, 0.2522, 0.2286, 0.2363], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0755, 0.0727, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 12:27:15,755 INFO [train.py:968] (1/2) Epoch 28, batch 21450, giga_loss[loss=0.2556, simple_loss=0.3365, pruned_loss=0.0874, over 29041.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3457, pruned_loss=0.09605, over 5728291.04 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3421, pruned_loss=0.09011, over 5736278.79 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3458, pruned_loss=0.09604, over 5718530.68 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:27:33,432 INFO [optim.py:369] (1/2) 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 12:27:41,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252383.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:27:51,326 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:27:54,022 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:27:54,443 INFO [train.py:968] (1/2) Epoch 28, batch 21500, giga_loss[loss=0.2621, simple_loss=0.34, pruned_loss=0.09216, over 28987.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3437, pruned_loss=0.0953, over 5730933.46 frames. ], libri_tot_loss[loss=0.2619, simple_loss=0.3426, pruned_loss=0.09056, over 5738620.80 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3434, pruned_loss=0.09502, over 5720753.03 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:28:16,426 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:28:17,721 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:28:34,459 INFO [train.py:968] (1/2) Epoch 28, batch 21550, libri_loss[loss=0.2962, simple_loss=0.366, pruned_loss=0.1132, over 28731.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.341, pruned_loss=0.0941, over 5725865.12 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3431, pruned_loss=0.09099, over 5742203.90 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3403, pruned_loss=0.09356, over 5713925.37 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:28:50,395 INFO [optim.py:369] (1/2) 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 12:29:11,631 INFO [train.py:968] (1/2) Epoch 28, batch 21600, giga_loss[loss=0.288, simple_loss=0.3622, pruned_loss=0.1069, over 28238.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3403, pruned_loss=0.09352, over 5724228.12 frames. ], libri_tot_loss[loss=0.2627, simple_loss=0.3432, pruned_loss=0.09113, over 5737710.35 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3395, pruned_loss=0.09304, over 5718271.00 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:29:32,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:29:36,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252526.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:29:38,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252529.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 12:29:52,434 INFO [train.py:968] (1/2) Epoch 28, batch 21650, giga_loss[loss=0.268, simple_loss=0.3465, pruned_loss=0.09472, over 28640.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09368, over 5719707.30 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3439, pruned_loss=0.09165, over 5736736.27 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3383, pruned_loss=0.09287, over 5715710.84 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:30:00,141 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252558.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:30:11,587 INFO [optim.py:369] (1/2) 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 12:30:16,642 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2036, 1.1964, 3.8407, 3.2864], device='cuda:1'), covar=tensor([0.2089, 0.3307, 0.0759, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0797, 0.0666, 0.0991, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0010, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 12:30:21,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3827, 1.5455, 1.5669, 1.3873], device='cuda:1'), covar=tensor([0.1909, 0.2136, 0.2256, 0.2186], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0755, 0.0726, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 12:30:23,176 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:30:36,172 INFO [train.py:968] (1/2) Epoch 28, batch 21700, giga_loss[loss=0.261, simple_loss=0.3392, pruned_loss=0.0914, over 28809.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3386, pruned_loss=0.09407, over 5719745.81 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.344, pruned_loss=0.09183, over 5739323.18 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3374, pruned_loss=0.09332, over 5714147.85 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:30:40,648 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:31:16,102 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-14 12:31:16,926 INFO [train.py:968] (1/2) Epoch 28, batch 21750, giga_loss[loss=0.224, simple_loss=0.3052, pruned_loss=0.07139, over 28911.00 frames. ], tot_loss[loss=0.261, simple_loss=0.336, pruned_loss=0.09297, over 5712561.39 frames. ], libri_tot_loss[loss=0.2644, simple_loss=0.3444, pruned_loss=0.09217, over 5732161.39 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3346, pruned_loss=0.0921, over 5714603.08 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:31:19,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:31:29,898 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:31:33,038 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:31:57,482 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 21800, giga_loss[loss=0.2239, simple_loss=0.2941, pruned_loss=0.07684, over 28166.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3324, pruned_loss=0.09154, over 5704981.63 frames. ], libri_tot_loss[loss=0.2645, simple_loss=0.3445, pruned_loss=0.09225, over 5732417.05 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3312, pruned_loss=0.0908, over 5706050.27 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:32:36,129 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:32:37,411 INFO [train.py:968] (1/2) Epoch 28, batch 21850, giga_loss[loss=0.3181, simple_loss=0.3774, pruned_loss=0.1294, over 26792.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3306, pruned_loss=0.09041, over 5708123.96 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3444, pruned_loss=0.09243, over 5735167.69 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3295, pruned_loss=0.08964, over 5706123.56 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:32:39,174 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:32:57,704 INFO [optim.py:369] (1/2) 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 12:33:03,027 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:16,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:19,056 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:19,432 INFO [train.py:968] (1/2) Epoch 28, batch 21900, giga_loss[loss=0.2563, simple_loss=0.3309, pruned_loss=0.09082, over 28893.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.331, pruned_loss=0.09029, over 5708265.04 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3443, pruned_loss=0.09243, over 5737786.23 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3301, pruned_loss=0.08965, over 5703998.81 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:33:22,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:33:42,930 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 21950, giga_loss[loss=0.2792, simple_loss=0.3577, pruned_loss=0.1004, over 28888.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.335, pruned_loss=0.09223, over 5708871.85 frames. ], libri_tot_loss[loss=0.265, simple_loss=0.3445, pruned_loss=0.09272, over 5738182.06 frames. ], giga_tot_loss[loss=0.2583, simple_loss=0.3338, pruned_loss=0.09142, over 5704212.17 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:34:19,225 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:34:25,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1558, 2.0743, 1.7233, 1.6431], device='cuda:1'), covar=tensor([0.0950, 0.0746, 0.0974, 0.1257], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0448, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 12:34:45,023 INFO [train.py:968] (1/2) Epoch 28, batch 22000, giga_loss[loss=0.31, simple_loss=0.3728, pruned_loss=0.1236, over 26734.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3377, pruned_loss=0.09341, over 5711048.34 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.345, pruned_loss=0.09321, over 5742704.56 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3362, pruned_loss=0.09229, over 5702617.15 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:35:07,913 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1522, 3.2939, 2.0266, 0.9968], device='cuda:1'), covar=tensor([0.7287, 0.3145, 0.4232, 0.7392], device='cuda:1'), in_proj_covar=tensor([0.1827, 0.1709, 0.1645, 0.1487], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 12:35:22,520 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1252943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:25,696 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1252946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:27,313 INFO [train.py:968] (1/2) Epoch 28, batch 22050, giga_loss[loss=0.2533, simple_loss=0.3399, pruned_loss=0.08339, over 28649.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3397, pruned_loss=0.09356, over 5708730.30 frames. ], libri_tot_loss[loss=0.2662, simple_loss=0.3453, pruned_loss=0.0935, over 5746343.25 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.338, pruned_loss=0.09242, over 5697833.76 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:35:35,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1252960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:36,860 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1252962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:35:43,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-14 12:35:46,046 INFO [optim.py:369] (1/2) 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 12:35:48,720 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1252975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:36:04,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4539, 1.5268, 1.6362, 1.2634], device='cuda:1'), covar=tensor([0.1919, 0.2743, 0.1683, 0.2038], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0715, 0.0983, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 12:36:09,157 INFO [train.py:968] (1/2) Epoch 28, batch 22100, giga_loss[loss=0.213, simple_loss=0.3004, pruned_loss=0.06276, over 28991.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3383, pruned_loss=0.0922, over 5707214.00 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3454, pruned_loss=0.09368, over 5748297.66 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3368, pruned_loss=0.09112, over 5696171.55 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:36:50,610 INFO [train.py:968] (1/2) Epoch 28, batch 22150, giga_loss[loss=0.2612, simple_loss=0.3421, pruned_loss=0.09022, over 28626.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.3372, pruned_loss=0.09165, over 5707932.71 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3453, pruned_loss=0.09381, over 5749411.26 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.336, pruned_loss=0.09064, over 5697643.52 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:36:53,830 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6789, 1.8577, 1.8899, 1.4524], device='cuda:1'), covar=tensor([0.1817, 0.2750, 0.1591, 0.1926], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0715, 0.0984, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 12:37:08,649 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1253070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:37:11,536 INFO [optim.py:369] (1/2) 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 12:37:32,346 INFO [train.py:968] (1/2) Epoch 28, batch 22200, giga_loss[loss=0.2559, simple_loss=0.3269, pruned_loss=0.09247, over 28726.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.339, pruned_loss=0.09328, over 5706055.56 frames. ], libri_tot_loss[loss=0.2665, simple_loss=0.3452, pruned_loss=0.09385, over 5751593.48 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.338, pruned_loss=0.09241, over 5695256.57 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:37:36,225 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:37:38,125 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1253106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:38:01,622 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:38:12,236 INFO [train.py:968] (1/2) Epoch 28, batch 22250, giga_loss[loss=0.2961, simple_loss=0.3707, pruned_loss=0.1107, over 28289.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3412, pruned_loss=0.09437, over 5715450.85 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3457, pruned_loss=0.09412, over 5755204.15 frames. ], giga_tot_loss[loss=0.2633, simple_loss=0.3398, pruned_loss=0.09344, over 5702707.80 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:38:33,758 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 22300, giga_loss[loss=0.2565, simple_loss=0.3454, pruned_loss=0.08386, over 28951.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3439, pruned_loss=0.09587, over 5708844.62 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3458, pruned_loss=0.09421, over 5756011.13 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3427, pruned_loss=0.09507, over 5697863.83 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:39:26,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4863, 2.1575, 1.6021, 0.7701], device='cuda:1'), covar=tensor([0.6091, 0.2913, 0.4057, 0.6623], device='cuda:1'), in_proj_covar=tensor([0.1831, 0.1714, 0.1649, 0.1490], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 12:39:29,712 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 22350, giga_loss[loss=0.3035, simple_loss=0.3793, pruned_loss=0.1139, over 27590.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3464, pruned_loss=0.09696, over 5714089.47 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3456, pruned_loss=0.09418, over 5757143.39 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3457, pruned_loss=0.09638, over 5703538.14 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:39:52,914 INFO [optim.py:369] (1/2) 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 12:40:10,447 INFO [train.py:968] (1/2) Epoch 28, batch 22400, giga_loss[loss=0.2656, simple_loss=0.3426, pruned_loss=0.09424, over 28503.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.347, pruned_loss=0.0971, over 5717446.70 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3456, pruned_loss=0.09417, over 5758220.55 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3465, pruned_loss=0.09673, over 5706671.06 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:40:41,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1253337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:40:51,141 INFO [train.py:968] (1/2) Epoch 28, batch 22450, libri_loss[loss=0.3108, simple_loss=0.3886, pruned_loss=0.1165, over 26148.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3478, pruned_loss=0.09707, over 5716723.67 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3461, pruned_loss=0.09457, over 5755732.70 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3469, pruned_loss=0.0965, over 5709361.30 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:41:06,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3650, 1.3778, 1.2441, 1.4974], device='cuda:1'), covar=tensor([0.0703, 0.0422, 0.0369, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 12:41:10,631 INFO [optim.py:369] (1/2) 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 12:41:21,057 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:41:23,057 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 22500, giga_loss[loss=0.3057, simple_loss=0.3737, pruned_loss=0.1188, over 28766.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3499, pruned_loss=0.09912, over 5715606.42 frames. ], libri_tot_loss[loss=0.2682, simple_loss=0.3465, pruned_loss=0.09497, over 5758167.98 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.349, pruned_loss=0.09837, over 5707088.42 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:41:47,967 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:41:58,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-14 12:42:10,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1253445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:42:12,081 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 12:42:12,841 INFO [train.py:968] (1/2) Epoch 28, batch 22550, giga_loss[loss=0.2735, simple_loss=0.3452, pruned_loss=0.1009, over 27940.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.347, pruned_loss=0.09744, over 5717840.92 frames. ], libri_tot_loss[loss=0.2684, simple_loss=0.3466, pruned_loss=0.09507, over 5758906.14 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09678, over 5710312.53 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:42:35,772 INFO [optim.py:369] (1/2) 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 12:42:41,321 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:42:43,401 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 22600, giga_loss[loss=0.2295, simple_loss=0.3157, pruned_loss=0.0717, over 29064.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3445, pruned_loss=0.09641, over 5695134.33 frames. ], libri_tot_loss[loss=0.2689, simple_loss=0.347, pruned_loss=0.09542, over 5743091.95 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3434, pruned_loss=0.0956, over 5702942.05 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:43:02,606 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:43:31,458 INFO [train.py:968] (1/2) Epoch 28, batch 22650, giga_loss[loss=0.2662, simple_loss=0.351, pruned_loss=0.09066, over 28774.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3425, pruned_loss=0.09544, over 5701990.88 frames. ], libri_tot_loss[loss=0.27, simple_loss=0.3477, pruned_loss=0.09616, over 5745360.81 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3408, pruned_loss=0.09412, over 5705081.82 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:43:50,634 INFO [optim.py:369] (1/2) 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 12:44:00,003 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2430, 1.5997, 1.6623, 1.3707], device='cuda:1'), covar=tensor([0.2234, 0.1992, 0.2350, 0.2389], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0754, 0.0726, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 12:44:02,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1253588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:44:05,463 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1253591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:44:07,407 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 12:44:10,920 INFO [train.py:968] (1/2) Epoch 28, batch 22700, giga_loss[loss=0.2638, simple_loss=0.35, pruned_loss=0.08878, over 28566.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3425, pruned_loss=0.09434, over 5701214.22 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3481, pruned_loss=0.09645, over 5747599.63 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3407, pruned_loss=0.09302, over 5700852.73 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:44:31,396 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1253620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:44:31,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6371, 1.8350, 1.5262, 1.6693], device='cuda:1'), covar=tensor([0.2797, 0.2915, 0.3306, 0.2623], device='cuda:1'), in_proj_covar=tensor([0.1598, 0.1151, 0.1411, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 12:44:53,315 INFO [train.py:968] (1/2) Epoch 28, batch 22750, giga_loss[loss=0.2552, simple_loss=0.3469, pruned_loss=0.08181, over 28934.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3456, pruned_loss=0.09517, over 5700675.39 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3486, pruned_loss=0.0968, over 5744325.80 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3437, pruned_loss=0.09375, over 5702268.01 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:45:13,809 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 22800, giga_loss[loss=0.2828, simple_loss=0.3568, pruned_loss=0.1044, over 27866.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3455, pruned_loss=0.09573, over 5697598.45 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3491, pruned_loss=0.09734, over 5747363.55 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3435, pruned_loss=0.09409, over 5695325.01 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:45:51,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4116, 1.6476, 1.5303, 1.3864], device='cuda:1'), covar=tensor([0.3612, 0.2850, 0.2358, 0.3073], device='cuda:1'), in_proj_covar=tensor([0.2040, 0.2003, 0.1907, 0.2052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 12:46:04,542 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 22850, giga_loss[loss=0.2847, simple_loss=0.3543, pruned_loss=0.1076, over 28858.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3447, pruned_loss=0.09695, over 5697470.52 frames. ], libri_tot_loss[loss=0.2724, simple_loss=0.3494, pruned_loss=0.09775, over 5745144.41 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3427, pruned_loss=0.09524, over 5696572.17 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:46:32,597 INFO [optim.py:369] (1/2) 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 12:46:48,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-14 12:46:51,301 INFO [train.py:968] (1/2) Epoch 28, batch 22900, giga_loss[loss=0.2866, simple_loss=0.3582, pruned_loss=0.1074, over 28372.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3433, pruned_loss=0.09739, over 5699174.34 frames. ], libri_tot_loss[loss=0.2728, simple_loss=0.3496, pruned_loss=0.09799, over 5741016.87 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3413, pruned_loss=0.09578, over 5700396.42 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:47:30,844 INFO [train.py:968] (1/2) Epoch 28, batch 22950, giga_loss[loss=0.2721, simple_loss=0.3461, pruned_loss=0.09903, over 28877.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3415, pruned_loss=0.09736, over 5704822.64 frames. ], libri_tot_loss[loss=0.2736, simple_loss=0.3502, pruned_loss=0.09851, over 5736809.53 frames. ], giga_tot_loss[loss=0.2651, simple_loss=0.3391, pruned_loss=0.09556, over 5708245.17 frames. ], batch size: 285, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:47:46,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3054, 1.4302, 1.2179, 1.5199], device='cuda:1'), covar=tensor([0.0731, 0.0414, 0.0373, 0.0833], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0104, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 12:47:51,949 INFO [optim.py:369] (1/2) 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 12:48:10,279 INFO [train.py:968] (1/2) Epoch 28, batch 23000, giga_loss[loss=0.2429, simple_loss=0.3328, pruned_loss=0.07655, over 28817.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3407, pruned_loss=0.09742, over 5695589.42 frames. ], libri_tot_loss[loss=0.2742, simple_loss=0.3506, pruned_loss=0.09889, over 5727041.73 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3382, pruned_loss=0.09564, over 5706297.91 frames. ], batch size: 285, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:48:21,861 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 12:48:48,998 INFO [train.py:968] (1/2) Epoch 28, batch 23050, giga_loss[loss=0.3, simple_loss=0.3624, pruned_loss=0.1188, over 28891.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3382, pruned_loss=0.09579, over 5707567.82 frames. ], libri_tot_loss[loss=0.2743, simple_loss=0.3506, pruned_loss=0.09898, over 5732829.89 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.336, pruned_loss=0.09419, over 5710111.00 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:49:07,889 INFO [optim.py:369] (1/2) 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 12:49:24,955 INFO [train.py:968] (1/2) Epoch 28, batch 23100, giga_loss[loss=0.2322, simple_loss=0.3086, pruned_loss=0.07792, over 28898.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3346, pruned_loss=0.0944, over 5712605.69 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3512, pruned_loss=0.09981, over 5737225.83 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3317, pruned_loss=0.09218, over 5709840.03 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:49:45,539 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1254023.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:49:59,720 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8641, 2.0615, 1.6742, 2.0498], device='cuda:1'), covar=tensor([0.2723, 0.2880, 0.3326, 0.2704], device='cuda:1'), in_proj_covar=tensor([0.1600, 0.1152, 0.1411, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 12:50:07,089 INFO [train.py:968] (1/2) Epoch 28, batch 23150, giga_loss[loss=0.221, simple_loss=0.3062, pruned_loss=0.06794, over 28643.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3308, pruned_loss=0.09234, over 5711000.97 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3516, pruned_loss=0.1002, over 5740862.70 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3278, pruned_loss=0.09007, over 5704933.58 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:50:27,323 INFO [optim.py:369] (1/2) 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 12:50:44,616 INFO [train.py:968] (1/2) Epoch 28, batch 23200, giga_loss[loss=0.2549, simple_loss=0.3368, pruned_loss=0.08652, over 28899.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3306, pruned_loss=0.0918, over 5701615.05 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3518, pruned_loss=0.1006, over 5725761.89 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3275, pruned_loss=0.0895, over 5710601.53 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:50:56,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1254112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:51:12,294 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 12:51:27,203 INFO [train.py:968] (1/2) Epoch 28, batch 23250, giga_loss[loss=0.2379, simple_loss=0.3147, pruned_loss=0.08052, over 28789.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.333, pruned_loss=0.09269, over 5708137.99 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3521, pruned_loss=0.101, over 5730163.86 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3297, pruned_loss=0.09023, over 5710738.78 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:51:49,345 INFO [optim.py:369] (1/2) 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 12:52:08,246 INFO [train.py:968] (1/2) Epoch 28, batch 23300, giga_loss[loss=0.3224, simple_loss=0.3864, pruned_loss=0.1292, over 27717.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3372, pruned_loss=0.09447, over 5709327.37 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.353, pruned_loss=0.1017, over 5734808.41 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3333, pruned_loss=0.09171, over 5706760.15 frames. ], batch size: 474, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:52:43,025 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1254242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:52:49,726 INFO [train.py:968] (1/2) Epoch 28, batch 23350, giga_loss[loss=0.3029, simple_loss=0.3746, pruned_loss=0.1156, over 27598.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3431, pruned_loss=0.09803, over 5706549.63 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3543, pruned_loss=0.1029, over 5735308.04 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3386, pruned_loss=0.09458, over 5703269.83 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:52:54,599 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1254255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:52:56,704 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:53:19,585 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1254287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:53:27,537 INFO [train.py:968] (1/2) Epoch 28, batch 23400, giga_loss[loss=0.2624, simple_loss=0.3448, pruned_loss=0.08993, over 28683.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3455, pruned_loss=0.09886, over 5698924.78 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3544, pruned_loss=0.1032, over 5730433.68 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3414, pruned_loss=0.09565, over 5700222.62 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:53:57,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5178, 1.6284, 1.3511, 1.5651], device='cuda:1'), covar=tensor([0.0747, 0.0316, 0.0339, 0.0888], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0074, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 12:54:10,804 INFO [train.py:968] (1/2) Epoch 28, batch 23450, giga_loss[loss=0.2398, simple_loss=0.3168, pruned_loss=0.08141, over 28503.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3476, pruned_loss=0.1002, over 5699638.51 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3545, pruned_loss=0.1034, over 5733142.40 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3441, pruned_loss=0.09739, over 5697766.58 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:54:38,231 INFO [optim.py:369] (1/2) 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 12:54:49,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2340, 3.0687, 2.9593, 1.5309], device='cuda:1'), covar=tensor([0.1052, 0.1157, 0.1009, 0.2348], device='cuda:1'), in_proj_covar=tensor([0.1289, 0.1190, 0.1003, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 12:54:56,036 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 12:54:58,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1254398.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:54:58,448 INFO [train.py:968] (1/2) Epoch 28, batch 23500, libri_loss[loss=0.3431, simple_loss=0.4035, pruned_loss=0.1413, over 28561.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3546, pruned_loss=0.1063, over 5696261.73 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3548, pruned_loss=0.1037, over 5737345.42 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3514, pruned_loss=0.1037, over 5689902.67 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:55:49,319 INFO [train.py:968] (1/2) Epoch 28, batch 23550, libri_loss[loss=0.2317, simple_loss=0.3082, pruned_loss=0.07755, over 29578.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3584, pruned_loss=0.1089, over 5693642.21 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3548, pruned_loss=0.1038, over 5739217.04 frames. ], giga_tot_loss[loss=0.2849, simple_loss=0.356, pruned_loss=0.1069, over 5686489.13 frames. ], batch size: 74, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:56:15,855 INFO [optim.py:369] (1/2) 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 12:56:21,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4146, 1.6371, 1.6210, 1.5028], device='cuda:1'), covar=tensor([0.1913, 0.2059, 0.2124, 0.1987], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0759, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 12:56:39,263 INFO [train.py:968] (1/2) Epoch 28, batch 23600, giga_loss[loss=0.3038, simple_loss=0.3707, pruned_loss=0.1184, over 28856.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3653, pruned_loss=0.114, over 5691356.08 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3552, pruned_loss=0.1041, over 5742258.28 frames. ], giga_tot_loss[loss=0.2938, simple_loss=0.3631, pruned_loss=0.1123, over 5681982.10 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 12:56:58,460 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 12:57:07,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1254531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 12:57:15,815 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1254541.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:57:19,953 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1254544.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:57:23,507 INFO [train.py:968] (1/2) Epoch 28, batch 23650, giga_loss[loss=0.2861, simple_loss=0.3598, pruned_loss=0.1062, over 28999.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3712, pruned_loss=0.1192, over 5670629.99 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3556, pruned_loss=0.1045, over 5725856.33 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3695, pruned_loss=0.1179, over 5675249.78 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:57:44,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1254573.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 12:57:50,192 INFO [optim.py:369] (1/2) 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 12:58:10,499 INFO [train.py:968] (1/2) Epoch 28, batch 23700, giga_loss[loss=0.4703, simple_loss=0.481, pruned_loss=0.2298, over 27543.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3768, pruned_loss=0.1244, over 5651442.96 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3555, pruned_loss=0.1046, over 5718714.13 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3761, pruned_loss=0.1237, over 5659132.07 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:58:29,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4014, 1.5860, 3.1522, 2.9683], device='cuda:1'), covar=tensor([0.1196, 0.2130, 0.0466, 0.1075], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0672, 0.1005, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 12:58:29,223 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 23750, giga_loss[loss=0.2946, simple_loss=0.3667, pruned_loss=0.1112, over 28924.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3786, pruned_loss=0.1254, over 5661005.88 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3552, pruned_loss=0.1044, over 5720562.61 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3786, pruned_loss=0.1253, over 5664691.25 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 12:59:29,319 INFO [optim.py:369] (1/2) 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 12:59:38,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9235, 2.1136, 2.1967, 1.8051], device='cuda:1'), covar=tensor([0.3346, 0.2694, 0.2483, 0.2754], device='cuda:1'), in_proj_covar=tensor([0.2057, 0.2017, 0.1922, 0.2068], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 12:59:49,405 INFO [train.py:968] (1/2) Epoch 28, batch 23800, giga_loss[loss=0.2982, simple_loss=0.3628, pruned_loss=0.1167, over 28881.00 frames. ], tot_loss[loss=0.3199, simple_loss=0.3817, pruned_loss=0.129, over 5656466.04 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3556, pruned_loss=0.1048, over 5719877.46 frames. ], giga_tot_loss[loss=0.3197, simple_loss=0.3817, pruned_loss=0.1288, over 5659031.54 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:00:41,678 INFO [train.py:968] (1/2) Epoch 28, batch 23850, giga_loss[loss=0.3389, simple_loss=0.3966, pruned_loss=0.1406, over 28661.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.383, pruned_loss=0.1312, over 5637349.19 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3554, pruned_loss=0.1048, over 5719084.76 frames. ], giga_tot_loss[loss=0.3232, simple_loss=0.3835, pruned_loss=0.1315, over 5639030.76 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:00:53,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1254760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:00:56,937 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:01:26,313 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1254792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:01:33,527 INFO [train.py:968] (1/2) Epoch 28, batch 23900, giga_loss[loss=0.3993, simple_loss=0.4167, pruned_loss=0.191, over 23553.00 frames. ], tot_loss[loss=0.3281, simple_loss=0.3866, pruned_loss=0.1348, over 5639593.69 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3557, pruned_loss=0.1051, over 5723578.52 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.3874, pruned_loss=0.1353, over 5635269.54 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:01:48,528 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.55 vs. limit=2.0
+2023-03-14 13:02:19,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0848, 1.4338, 1.1291, 0.5119], device='cuda:1'), covar=tensor([0.3017, 0.2553, 0.2608, 0.4629], device='cuda:1'), in_proj_covar=tensor([0.1846, 0.1738, 0.1660, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 13:02:29,429 INFO [train.py:968] (1/2) Epoch 28, batch 23950, giga_loss[loss=0.332, simple_loss=0.3872, pruned_loss=0.1384, over 28690.00 frames. ], tot_loss[loss=0.3328, simple_loss=0.3895, pruned_loss=0.1381, over 5605497.63 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3562, pruned_loss=0.1056, over 5713010.56 frames. ], giga_tot_loss[loss=0.3337, simple_loss=0.3902, pruned_loss=0.1386, over 5610448.82 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:02:47,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3826, 3.0205, 1.5011, 1.4892], device='cuda:1'), covar=tensor([0.0960, 0.0338, 0.0842, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0571, 0.0411, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 13:03:02,731 INFO [optim.py:369] (1/2) 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 13:03:22,316 INFO [train.py:968] (1/2) Epoch 28, batch 24000, giga_loss[loss=0.2859, simple_loss=0.3602, pruned_loss=0.1058, over 28716.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.387, pruned_loss=0.1371, over 5613324.22 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3561, pruned_loss=0.1056, over 5716504.56 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3884, pruned_loss=0.1381, over 5612037.85 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:03:22,316 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 13:03:29,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2522, 2.8757, 1.3168, 1.4023], device='cuda:1'), covar=tensor([0.1015, 0.0347, 0.0960, 0.1377], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0571, 0.0411, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 13:03:29,689 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7318, 3.5055, 3.3933, 1.7874], device='cuda:1'), covar=tensor([0.0849, 0.0979, 0.0996, 0.2163], device='cuda:1'), in_proj_covar=tensor([0.1303, 0.1202, 0.1013, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 13:03:30,586 INFO [train.py:1012] (1/2) Epoch 28, validation: loss=0.2029, simple_loss=0.3112, pruned_loss=0.04725, over 944034.00 frames. 
+2023-03-14 13:03:30,587 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 13:03:33,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1365, 1.3219, 1.3439, 1.0698], device='cuda:1'), covar=tensor([0.3312, 0.2886, 0.1945, 0.2820], device='cuda:1'), in_proj_covar=tensor([0.2051, 0.2016, 0.1920, 0.2068], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 13:03:36,373 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1254906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:04:15,861 INFO [train.py:968] (1/2) Epoch 28, batch 24050, giga_loss[loss=0.3587, simple_loss=0.3891, pruned_loss=0.1641, over 23671.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3846, pruned_loss=0.1355, over 5619854.27 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3564, pruned_loss=0.1059, over 5709016.50 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3859, pruned_loss=0.1366, over 5624690.94 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:04:23,451 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-14 13:04:44,760 INFO [optim.py:369] (1/2) 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 13:05:02,356 INFO [train.py:968] (1/2) Epoch 28, batch 24100, giga_loss[loss=0.365, simple_loss=0.4298, pruned_loss=0.1501, over 28588.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.384, pruned_loss=0.1339, over 5611217.49 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3562, pruned_loss=0.1059, over 5701224.18 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.3855, pruned_loss=0.1351, over 5621183.66 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:05:36,881 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-14 13:05:38,995 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-14 13:05:56,155 INFO [train.py:968] (1/2) Epoch 28, batch 24150, giga_loss[loss=0.3314, simple_loss=0.3981, pruned_loss=0.1324, over 28640.00 frames. ], tot_loss[loss=0.328, simple_loss=0.3861, pruned_loss=0.135, over 5609335.17 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.356, pruned_loss=0.1058, over 5703506.26 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3877, pruned_loss=0.1363, over 5614289.11 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:05:56,502 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1255049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:05:59,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:06:25,943 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:06:27,843 INFO [optim.py:369] (1/2) 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 13:06:29,481 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1255081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:06:51,608 INFO [train.py:968] (1/2) Epoch 28, batch 24200, giga_loss[loss=0.2936, simple_loss=0.3643, pruned_loss=0.1114, over 28697.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3856, pruned_loss=0.1341, over 5608482.21 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3559, pruned_loss=0.1059, over 5698183.65 frames. ], giga_tot_loss[loss=0.3295, simple_loss=0.3876, pruned_loss=0.1357, over 5614912.41 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:07:40,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3842, 1.6040, 1.5331, 1.3376], device='cuda:1'), covar=tensor([0.2910, 0.2590, 0.1949, 0.2504], device='cuda:1'), in_proj_covar=tensor([0.2052, 0.2016, 0.1918, 0.2066], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 13:07:41,829 INFO [train.py:968] (1/2) Epoch 28, batch 24250, giga_loss[loss=0.293, simple_loss=0.367, pruned_loss=0.1095, over 28893.00 frames. ], tot_loss[loss=0.3229, simple_loss=0.3829, pruned_loss=0.1315, over 5613660.90 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3563, pruned_loss=0.1063, over 5702497.46 frames. ], giga_tot_loss[loss=0.3252, simple_loss=0.3846, pruned_loss=0.1328, over 5613419.89 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:08:10,919 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 24300, giga_loss[loss=0.2913, simple_loss=0.3607, pruned_loss=0.111, over 28817.00 frames. ], tot_loss[loss=0.3187, simple_loss=0.3806, pruned_loss=0.1285, over 5625000.69 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3562, pruned_loss=0.1062, over 5704560.38 frames. ], giga_tot_loss[loss=0.321, simple_loss=0.3823, pruned_loss=0.1299, over 5622241.63 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:09:04,759 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0046, 1.2492, 1.0481, 0.3179], device='cuda:1'), covar=tensor([0.3905, 0.3311, 0.3920, 0.6771], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1741, 0.1665, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 13:09:16,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6236, 1.9022, 1.6193, 1.5911], device='cuda:1'), covar=tensor([0.2053, 0.2091, 0.2283, 0.1958], device='cuda:1'), in_proj_covar=tensor([0.1598, 0.1152, 0.1411, 0.1011], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 13:09:16,697 INFO [train.py:968] (1/2) Epoch 28, batch 24350, libri_loss[loss=0.3507, simple_loss=0.399, pruned_loss=0.1511, over 19614.00 frames. ], tot_loss[loss=0.3136, simple_loss=0.3766, pruned_loss=0.1253, over 5617122.94 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3564, pruned_loss=0.1065, over 5698640.77 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3786, pruned_loss=0.1268, over 5618604.27 frames. ], batch size: 187, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:09:47,368 INFO [optim.py:369] (1/2) 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 13:09:48,954 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:09:51,752 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:10:08,653 INFO [train.py:968] (1/2) Epoch 28, batch 24400, giga_loss[loss=0.3608, simple_loss=0.3958, pruned_loss=0.1629, over 26515.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3741, pruned_loss=0.1234, over 5625314.35 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3566, pruned_loss=0.1068, over 5699551.56 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3755, pruned_loss=0.1244, over 5625215.16 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:10:15,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-14 13:10:17,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4692, 1.7269, 1.6738, 1.5133], device='cuda:1'), covar=tensor([0.1999, 0.2076, 0.2208, 0.2150], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0760, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 13:10:42,412 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-14 13:10:54,622 INFO [train.py:968] (1/2) Epoch 28, batch 24450, giga_loss[loss=0.3053, simple_loss=0.3705, pruned_loss=0.12, over 27605.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5631118.15 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3569, pruned_loss=0.1073, over 5704036.70 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3744, pruned_loss=0.124, over 5624716.24 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:11:26,596 INFO [optim.py:369] (1/2) 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 13:11:50,075 INFO [train.py:968] (1/2) Epoch 28, batch 24500, giga_loss[loss=0.3299, simple_loss=0.3986, pruned_loss=0.1306, over 28945.00 frames. ], tot_loss[loss=0.3099, simple_loss=0.3732, pruned_loss=0.1233, over 5624730.88 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3571, pruned_loss=0.1075, over 5696173.33 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3742, pruned_loss=0.1238, over 5625839.41 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:11:51,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 13:12:01,615 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5310, 1.6909, 1.7566, 1.3305], device='cuda:1'), covar=tensor([0.1945, 0.2643, 0.1583, 0.1837], device='cuda:1'), in_proj_covar=tensor([0.0926, 0.0712, 0.0974, 0.0874], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 13:12:12,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4470, 2.5986, 2.6337, 2.0533], device='cuda:1'), covar=tensor([0.3451, 0.2623, 0.2415, 0.3316], device='cuda:1'), in_proj_covar=tensor([0.2058, 0.2018, 0.1921, 0.2071], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 13:12:41,261 INFO [train.py:968] (1/2) Epoch 28, batch 24550, giga_loss[loss=0.2891, simple_loss=0.3578, pruned_loss=0.1102, over 28865.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3717, pruned_loss=0.1213, over 5644341.49 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3574, pruned_loss=0.1076, over 5698580.05 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3725, pruned_loss=0.1218, over 5642339.05 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:12:44,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1255451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:12:55,833 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4618, 1.6834, 1.2317, 1.2880], device='cuda:1'), covar=tensor([0.1102, 0.0629, 0.1084, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0450, 0.0522, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 13:13:13,748 INFO [optim.py:369] (1/2) 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 13:13:35,680 INFO [train.py:968] (1/2) Epoch 28, batch 24600, giga_loss[loss=0.2675, simple_loss=0.3549, pruned_loss=0.09007, over 28880.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.3716, pruned_loss=0.1189, over 5650982.58 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3571, pruned_loss=0.1076, over 5701940.75 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3726, pruned_loss=0.1196, over 5645638.82 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:14:27,871 INFO [train.py:968] (1/2) Epoch 28, batch 24650, giga_loss[loss=0.3056, simple_loss=0.3522, pruned_loss=0.1295, over 23530.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3737, pruned_loss=0.1191, over 5649763.07 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3573, pruned_loss=0.1078, over 5695664.29 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3748, pruned_loss=0.1198, over 5649539.04 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:14:55,652 INFO [optim.py:369] (1/2) 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 13:15:12,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1255594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:15:15,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:15:16,095 INFO [train.py:968] (1/2) Epoch 28, batch 24700, giga_loss[loss=0.2926, simple_loss=0.3587, pruned_loss=0.1132, over 28935.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3741, pruned_loss=0.1199, over 5653765.19 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3576, pruned_loss=0.108, over 5696613.46 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.375, pruned_loss=0.1204, over 5651663.49 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:15:40,847 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 24750, giga_loss[loss=0.3663, simple_loss=0.4122, pruned_loss=0.1602, over 28719.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3722, pruned_loss=0.1187, over 5657047.98 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3571, pruned_loss=0.108, over 5682566.48 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.374, pruned_loss=0.1196, over 5666176.72 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:16:07,028 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1255654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:16:13,621 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1255658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:16:34,740 INFO [optim.py:369] (1/2) 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 13:16:53,430 INFO [train.py:968] (1/2) Epoch 28, batch 24800, giga_loss[loss=0.2761, simple_loss=0.3476, pruned_loss=0.1023, over 28676.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3713, pruned_loss=0.1188, over 5668507.12 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3571, pruned_loss=0.108, over 5683356.65 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3727, pruned_loss=0.1195, over 5674870.75 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:17:25,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4750, 1.6874, 1.5071, 1.5586], device='cuda:1'), covar=tensor([0.0813, 0.0333, 0.0327, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 13:17:35,190 INFO [train.py:968] (1/2) Epoch 28, batch 24850, giga_loss[loss=0.3115, simple_loss=0.3813, pruned_loss=0.1208, over 28618.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3713, pruned_loss=0.1202, over 5671485.01 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3577, pruned_loss=0.1085, over 5687874.89 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3722, pruned_loss=0.1206, over 5672247.78 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:18:01,916 INFO [optim.py:369] (1/2) 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 13:18:17,680 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 24900, giga_loss[loss=0.321, simple_loss=0.3894, pruned_loss=0.1263, over 28603.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3702, pruned_loss=0.1191, over 5663438.18 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3577, pruned_loss=0.1087, over 5680509.53 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3712, pruned_loss=0.1195, over 5669968.27 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:18:20,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:21,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1255801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:23,581 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1255804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:30,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4826, 1.7030, 1.3841, 1.3442], device='cuda:1'), covar=tensor([0.0920, 0.0461, 0.0940, 0.0938], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0453, 0.0525, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 13:18:44,398 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1255829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:18:47,276 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1255833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:19:01,720 INFO [train.py:968] (1/2) Epoch 28, batch 24950, giga_loss[loss=0.3084, simple_loss=0.383, pruned_loss=0.1169, over 28960.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3691, pruned_loss=0.1165, over 5675873.08 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3582, pruned_loss=0.1089, over 5683131.15 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3698, pruned_loss=0.1169, over 5678538.10 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:19:04,962 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1255852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:19:15,054 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8402, 1.2037, 1.2723, 0.9879], device='cuda:1'), covar=tensor([0.2290, 0.1581, 0.2716, 0.2050], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0765, 0.0736, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-14 13:19:31,085 INFO [optim.py:369] (1/2) 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 13:19:49,626 INFO [train.py:968] (1/2) Epoch 28, batch 25000, giga_loss[loss=0.3092, simple_loss=0.3777, pruned_loss=0.1204, over 28753.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3699, pruned_loss=0.1175, over 5674007.49 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3583, pruned_loss=0.1091, over 5688174.51 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3708, pruned_loss=0.1178, over 5671424.98 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:20:33,535 INFO [train.py:968] (1/2) Epoch 28, batch 25050, giga_loss[loss=0.2921, simple_loss=0.3635, pruned_loss=0.1104, over 28709.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3691, pruned_loss=0.1173, over 5682434.26 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3588, pruned_loss=0.1095, over 5694265.63 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3698, pruned_loss=0.1175, over 5674287.96 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:21:03,228 INFO [optim.py:369] (1/2) 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 13:21:21,971 INFO [train.py:968] (1/2) Epoch 28, batch 25100, giga_loss[loss=0.2629, simple_loss=0.3392, pruned_loss=0.09324, over 28933.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3685, pruned_loss=0.1175, over 5685764.68 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3588, pruned_loss=0.1095, over 5696875.99 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3692, pruned_loss=0.1179, over 5676630.12 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:22:03,961 INFO [train.py:968] (1/2) Epoch 28, batch 25150, giga_loss[loss=0.2676, simple_loss=0.344, pruned_loss=0.09559, over 28740.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3682, pruned_loss=0.1185, over 5666372.28 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3586, pruned_loss=0.1098, over 5696032.77 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3694, pruned_loss=0.119, over 5658902.83 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:22:24,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:22:36,096 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 25200, giga_loss[loss=0.2746, simple_loss=0.3486, pruned_loss=0.1003, over 28897.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3682, pruned_loss=0.1193, over 5670813.32 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3586, pruned_loss=0.1098, over 5696032.77 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3692, pruned_loss=0.1197, over 5664999.76 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:23:40,151 INFO [train.py:968] (1/2) Epoch 28, batch 25250, giga_loss[loss=0.2594, simple_loss=0.3336, pruned_loss=0.09259, over 29031.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3666, pruned_loss=0.1187, over 5668814.11 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.359, pruned_loss=0.11, over 5698572.99 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3673, pruned_loss=0.119, over 5661528.00 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:23:51,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6744, 4.5350, 4.2901, 2.1289], device='cuda:1'), covar=tensor([0.0620, 0.0697, 0.0796, 0.1919], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.1211, 0.1021, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 13:23:54,804 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256164.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:24:09,612 INFO [optim.py:369] (1/2) 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 13:24:15,587 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:24:26,917 INFO [train.py:968] (1/2) Epoch 28, batch 25300, giga_loss[loss=0.3624, simple_loss=0.4066, pruned_loss=0.1591, over 27546.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3661, pruned_loss=0.1185, over 5675368.77 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3591, pruned_loss=0.1101, over 5698407.03 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3666, pruned_loss=0.1188, over 5669544.89 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:24:56,046 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:25:12,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7775, 2.0054, 2.0136, 1.5366], device='cuda:1'), covar=tensor([0.1952, 0.2783, 0.1695, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0715, 0.0979, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 13:25:15,794 INFO [train.py:968] (1/2) Epoch 28, batch 25350, giga_loss[loss=0.3013, simple_loss=0.3738, pruned_loss=0.1143, over 28943.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3661, pruned_loss=0.1188, over 5665375.81 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3589, pruned_loss=0.1099, over 5699947.48 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3669, pruned_loss=0.1194, over 5658147.37 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:25:44,572 INFO [optim.py:369] (1/2) 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 13:25:59,172 INFO [train.py:968] (1/2) Epoch 28, batch 25400, giga_loss[loss=0.2735, simple_loss=0.3571, pruned_loss=0.09495, over 28956.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3661, pruned_loss=0.1176, over 5662304.02 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.359, pruned_loss=0.1101, over 5695608.29 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3668, pruned_loss=0.118, over 5659500.56 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:26:40,644 INFO [train.py:968] (1/2) Epoch 28, batch 25450, libri_loss[loss=0.3195, simple_loss=0.3712, pruned_loss=0.1339, over 19406.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3665, pruned_loss=0.1172, over 5646656.41 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3597, pruned_loss=0.1108, over 5675609.80 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3667, pruned_loss=0.1173, over 5661569.76 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:26:59,056 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:00,871 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:08,908 INFO [optim.py:369] (1/2) 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 13:27:11,201 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:19,003 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-14 13:27:25,279 INFO [train.py:968] (1/2) Epoch 28, batch 25500, giga_loss[loss=0.2579, simple_loss=0.345, pruned_loss=0.08538, over 28857.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3665, pruned_loss=0.1171, over 5650479.83 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3598, pruned_loss=0.1112, over 5680133.15 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3668, pruned_loss=0.117, over 5657758.03 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:27:29,459 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:27:44,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.99 vs. limit=2.0
+2023-03-14 13:28:07,398 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:28:10,884 INFO [train.py:968] (1/2) Epoch 28, batch 25550, giga_loss[loss=0.2583, simple_loss=0.3349, pruned_loss=0.09083, over 28462.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.367, pruned_loss=0.1181, over 5652986.03 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3597, pruned_loss=0.1111, over 5684208.95 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3674, pruned_loss=0.1181, over 5654645.82 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:28:26,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2516, 0.9749, 1.0862, 1.3799], device='cuda:1'), covar=tensor([0.0694, 0.0402, 0.0328, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 13:28:38,476 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3372, 3.5321, 1.5128, 1.5437], device='cuda:1'), covar=tensor([0.1079, 0.0570, 0.0917, 0.1487], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0574, 0.0411, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 13:28:39,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-14 13:28:44,460 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 25600, giga_loss[loss=0.3642, simple_loss=0.4061, pruned_loss=0.1611, over 28866.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1218, over 5649150.34 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3599, pruned_loss=0.1112, over 5688482.27 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3713, pruned_loss=0.1219, over 5645747.93 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:29:13,531 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 13:29:39,316 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256539.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:29:49,420 INFO [train.py:968] (1/2) Epoch 28, batch 25650, giga_loss[loss=0.2887, simple_loss=0.3536, pruned_loss=0.1119, over 28608.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.371, pruned_loss=0.1224, over 5658964.26 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3601, pruned_loss=0.1114, over 5687290.05 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3713, pruned_loss=0.1224, over 5656808.48 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:30:01,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1256562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:30:21,137 INFO [optim.py:369] (1/2) 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 13:30:26,664 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:30:28,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:30:37,469 INFO [train.py:968] (1/2) Epoch 28, batch 25700, giga_loss[loss=0.2828, simple_loss=0.3515, pruned_loss=0.1071, over 28867.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3712, pruned_loss=0.1239, over 5660855.69 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3593, pruned_loss=0.1109, over 5695207.85 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3726, pruned_loss=0.1249, over 5651116.36 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:30:49,112 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 13:30:58,572 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:31:23,575 INFO [train.py:968] (1/2) Epoch 28, batch 25750, giga_loss[loss=0.2952, simple_loss=0.3614, pruned_loss=0.1145, over 28566.00 frames. ], tot_loss[loss=0.3105, simple_loss=0.3719, pruned_loss=0.1245, over 5661411.02 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3591, pruned_loss=0.1109, over 5698430.42 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3734, pruned_loss=0.1255, over 5650272.62 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:31:54,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256682.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:31:55,093 INFO [optim.py:369] (1/2) 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 13:31:59,467 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256685.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:32:07,119 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4307, 1.6023, 1.5243, 1.6089], device='cuda:1'), covar=tensor([0.0798, 0.0349, 0.0329, 0.0877], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 13:32:11,530 INFO [train.py:968] (1/2) Epoch 28, batch 25800, giga_loss[loss=0.3102, simple_loss=0.371, pruned_loss=0.1247, over 28322.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.371, pruned_loss=0.1239, over 5657338.04 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3587, pruned_loss=0.1106, over 5700319.19 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3728, pruned_loss=0.1252, over 5646244.96 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:32:16,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:32:19,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:32:26,165 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256714.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:32:27,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256716.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:32:42,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1256737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:32:53,044 INFO [train.py:968] (1/2) Epoch 28, batch 25850, giga_loss[loss=0.2427, simple_loss=0.3318, pruned_loss=0.07682, over 28254.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3723, pruned_loss=0.1231, over 5674287.81 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3594, pruned_loss=0.1111, over 5704380.09 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3733, pruned_loss=0.1239, over 5661235.07 frames. ], batch size: 65, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:32:54,304 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5253, 1.8482, 1.8289, 1.5760], device='cuda:1'), covar=tensor([0.2241, 0.2284, 0.2301, 0.2452], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0762, 0.0734, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 13:33:05,001 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:33:41,816 INFO [train.py:968] (1/2) Epoch 28, batch 25900, giga_loss[loss=0.3441, simple_loss=0.3899, pruned_loss=0.1492, over 27583.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.37, pruned_loss=0.1212, over 5654231.76 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3597, pruned_loss=0.1112, over 5695743.88 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3708, pruned_loss=0.1219, over 5651071.12 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:34:16,081 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1256838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:34:26,098 INFO [train.py:968] (1/2) Epoch 28, batch 25950, giga_loss[loss=0.2771, simple_loss=0.3443, pruned_loss=0.1049, over 28794.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3666, pruned_loss=0.1186, over 5667061.07 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.36, pruned_loss=0.1116, over 5699169.84 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.367, pruned_loss=0.119, over 5660831.57 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:34:57,752 INFO [optim.py:369] (1/2) 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 13:35:12,931 INFO [train.py:968] (1/2) Epoch 28, batch 26000, giga_loss[loss=0.2917, simple_loss=0.3551, pruned_loss=0.1142, over 28666.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 5678817.97 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3604, pruned_loss=0.1117, over 5703046.27 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3665, pruned_loss=0.1193, over 5669765.41 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:35:18,662 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1256903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:35:22,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1256906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:35:53,550 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 26050, giga_loss[loss=0.3221, simple_loss=0.3808, pruned_loss=0.1317, over 28233.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3667, pruned_loss=0.119, over 5671205.67 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3608, pruned_loss=0.112, over 5694061.00 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3666, pruned_loss=0.1191, over 5671352.72 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:36:34,224 INFO [optim.py:369] (1/2) 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 13:36:40,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5077, 2.0527, 1.2722, 0.8014], device='cuda:1'), covar=tensor([0.7915, 0.4742, 0.3684, 0.7734], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1751, 0.1666, 0.1509], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 13:36:46,328 INFO [train.py:968] (1/2) Epoch 28, batch 26100, giga_loss[loss=0.2844, simple_loss=0.358, pruned_loss=0.1054, over 28846.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3698, pruned_loss=0.1201, over 5671944.46 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3603, pruned_loss=0.1118, over 5692401.60 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3705, pruned_loss=0.1208, over 5673338.93 frames. ], batch size: 66, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:37:32,326 INFO [train.py:968] (1/2) Epoch 28, batch 26150, giga_loss[loss=0.334, simple_loss=0.3971, pruned_loss=0.1355, over 28700.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3716, pruned_loss=0.1186, over 5676949.66 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3598, pruned_loss=0.1116, over 5697737.54 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.373, pruned_loss=0.1195, over 5672483.28 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:38:04,672 INFO [optim.py:369] (1/2) 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 13:38:13,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257091.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:38:19,805 INFO [train.py:968] (1/2) Epoch 28, batch 26200, giga_loss[loss=0.3334, simple_loss=0.3877, pruned_loss=0.1395, over 28921.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3734, pruned_loss=0.1194, over 5679779.07 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3601, pruned_loss=0.1118, over 5699093.39 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3745, pruned_loss=0.1201, over 5674572.24 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:38:24,658 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:39:01,323 INFO [train.py:968] (1/2) Epoch 28, batch 26250, giga_loss[loss=0.3214, simple_loss=0.3928, pruned_loss=0.125, over 29056.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3755, pruned_loss=0.1216, over 5680029.25 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.36, pruned_loss=0.1121, over 5696012.25 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.377, pruned_loss=0.1222, over 5678188.25 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:39:31,833 INFO [scaling.py:679] (1/2) 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] (1/2) 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 13:39:47,533 INFO [train.py:968] (1/2) Epoch 28, batch 26300, giga_loss[loss=0.3032, simple_loss=0.3657, pruned_loss=0.1203, over 28925.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3768, pruned_loss=0.123, over 5676140.56 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3602, pruned_loss=0.1122, over 5697090.19 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3779, pruned_loss=0.1235, over 5673709.75 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:40:03,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:40:21,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5905, 1.8766, 1.7104, 1.6344], device='cuda:1'), covar=tensor([0.2295, 0.2164, 0.2579, 0.2321], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0758, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 13:40:23,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3245, 1.0932, 3.7065, 3.3101], device='cuda:1'), covar=tensor([0.1895, 0.3119, 0.0978, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0677, 0.1011, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 13:40:24,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1257234.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:40:26,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1257237.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:40:27,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5450, 4.4039, 4.2046, 2.0934], device='cuda:1'), covar=tensor([0.0598, 0.0734, 0.0714, 0.1940], device='cuda:1'), in_proj_covar=tensor([0.1319, 0.1216, 0.1028, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 13:40:37,048 INFO [train.py:968] (1/2) Epoch 28, batch 26350, giga_loss[loss=0.399, simple_loss=0.438, pruned_loss=0.18, over 28923.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3771, pruned_loss=0.1245, over 5680658.92 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3605, pruned_loss=0.1124, over 5700308.41 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.378, pruned_loss=0.1249, over 5675654.67 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:40:52,838 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1257266.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:41:05,904 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5121, 1.8891, 1.5077, 1.5396], device='cuda:1'), covar=tensor([0.2631, 0.2663, 0.3048, 0.2401], device='cuda:1'), in_proj_covar=tensor([0.1600, 0.1155, 0.1415, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0014, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 13:41:08,020 INFO [optim.py:369] (1/2) 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 13:41:12,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4638, 4.3134, 4.0984, 2.0948], device='cuda:1'), covar=tensor([0.0604, 0.0727, 0.0720, 0.1973], device='cuda:1'), in_proj_covar=tensor([0.1318, 0.1216, 0.1028, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 13:41:20,568 INFO [train.py:968] (1/2) Epoch 28, batch 26400, giga_loss[loss=0.302, simple_loss=0.3664, pruned_loss=0.1188, over 28791.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3743, pruned_loss=0.1228, over 5678350.57 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3603, pruned_loss=0.1123, over 5688767.43 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3756, pruned_loss=0.1236, over 5684066.07 frames. ], batch size: 243, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:42:05,672 INFO [train.py:968] (1/2) Epoch 28, batch 26450, giga_loss[loss=0.3041, simple_loss=0.3756, pruned_loss=0.1163, over 29048.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3716, pruned_loss=0.1216, over 5675669.81 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3601, pruned_loss=0.1121, over 5685852.52 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3733, pruned_loss=0.1227, over 5682327.25 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:42:13,150 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1257356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:42:15,601 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1257359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:42:41,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1810, 1.4492, 1.4906, 1.2639], device='cuda:1'), covar=tensor([0.3129, 0.2636, 0.1738, 0.2221], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2026, 0.1932, 0.2076], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 13:42:43,291 INFO [optim.py:369] (1/2) 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 13:42:46,658 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1257388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:42:55,230 INFO [train.py:968] (1/2) Epoch 28, batch 26500, libri_loss[loss=0.317, simple_loss=0.3827, pruned_loss=0.1256, over 29210.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3702, pruned_loss=0.1208, over 5676463.53 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3606, pruned_loss=0.1124, over 5690366.68 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3714, pruned_loss=0.1217, over 5677270.42 frames. ], batch size: 97, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:43:37,828 INFO [train.py:968] (1/2) Epoch 28, batch 26550, giga_loss[loss=0.2701, simple_loss=0.3404, pruned_loss=0.09987, over 28830.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3704, pruned_loss=0.1209, over 5678194.92 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3607, pruned_loss=0.1124, over 5691126.57 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3716, pruned_loss=0.1219, over 5677658.10 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:44:06,564 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:44:11,930 INFO [optim.py:369] (1/2) 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 13:44:12,773 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:44:24,795 INFO [train.py:968] (1/2) Epoch 28, batch 26600, giga_loss[loss=0.2364, simple_loss=0.3046, pruned_loss=0.08412, over 28543.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3705, pruned_loss=0.1221, over 5675677.96 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.361, pruned_loss=0.1125, over 5693147.89 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3713, pruned_loss=0.1228, over 5673383.54 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:44:41,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9411, 2.1301, 1.5440, 1.7823], device='cuda:1'), covar=tensor([0.1104, 0.0755, 0.1124, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 13:44:52,046 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2750, 1.9381, 1.4887, 0.4109], device='cuda:1'), covar=tensor([0.4973, 0.3564, 0.5123, 0.6948], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1748, 0.1664, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 13:45:06,995 INFO [train.py:968] (1/2) Epoch 28, batch 26650, giga_loss[loss=0.3325, simple_loss=0.3709, pruned_loss=0.147, over 23771.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3681, pruned_loss=0.1216, over 5660173.54 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.36, pruned_loss=0.1122, over 5698114.69 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3699, pruned_loss=0.1229, over 5652485.01 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:45:33,727 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 13:45:42,904 INFO [optim.py:369] (1/2) 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 13:45:54,639 INFO [train.py:968] (1/2) Epoch 28, batch 26700, libri_loss[loss=0.2203, simple_loss=0.2962, pruned_loss=0.0722, over 29668.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3671, pruned_loss=0.1203, over 5656742.21 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3599, pruned_loss=0.1121, over 5693446.40 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3689, pruned_loss=0.1217, over 5654520.30 frames. ], batch size: 69, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:46:18,911 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1257624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:46:21,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.8019, 2.0438, 1.7752, 1.8283], device='cuda:1'), covar=tensor([0.2262, 0.2524, 0.2631, 0.2359], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0764, 0.0735, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 13:46:39,754 INFO [train.py:968] (1/2) Epoch 28, batch 26750, giga_loss[loss=0.3253, simple_loss=0.39, pruned_loss=0.1303, over 28683.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3684, pruned_loss=0.1198, over 5654478.02 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3602, pruned_loss=0.1123, over 5683772.55 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3697, pruned_loss=0.1208, over 5661060.90 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:46:48,082 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1257656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:47:18,700 INFO [optim.py:369] (1/2) 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 13:47:30,728 INFO [train.py:968] (1/2) Epoch 28, batch 26800, libri_loss[loss=0.3061, simple_loss=0.3741, pruned_loss=0.1191, over 29544.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3705, pruned_loss=0.1217, over 5644140.21 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3601, pruned_loss=0.1122, over 5677399.71 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3718, pruned_loss=0.1227, over 5653544.02 frames. ], batch size: 82, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:47:42,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5915, 2.2894, 1.7285, 0.8896], device='cuda:1'), covar=tensor([0.6538, 0.3217, 0.4460, 0.7102], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1745, 0.1658, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 13:48:03,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9077, 2.1104, 1.3899, 1.8421], device='cuda:1'), covar=tensor([0.1025, 0.0681, 0.1098, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0455, 0.0526, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 13:48:12,271 INFO [train.py:968] (1/2) Epoch 28, batch 26850, giga_loss[loss=0.266, simple_loss=0.3357, pruned_loss=0.09821, over 28945.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3697, pruned_loss=0.1209, over 5659987.21 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3605, pruned_loss=0.1124, over 5681956.30 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3708, pruned_loss=0.1219, over 5662550.47 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:48:22,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2827, 1.9111, 1.4081, 0.5274], device='cuda:1'), covar=tensor([0.6187, 0.2982, 0.4203, 0.7238], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1746, 0.1659, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 13:48:46,547 INFO [optim.py:369] (1/2) 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 13:48:56,470 INFO [train.py:968] (1/2) Epoch 28, batch 26900, giga_loss[loss=0.3506, simple_loss=0.4116, pruned_loss=0.1448, over 28790.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3709, pruned_loss=0.1187, over 5665138.09 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3607, pruned_loss=0.1126, over 5683278.88 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3716, pruned_loss=0.1194, over 5665700.64 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:49:43,451 INFO [train.py:968] (1/2) Epoch 28, batch 26950, libri_loss[loss=0.3572, simple_loss=0.4045, pruned_loss=0.1549, over 18201.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3727, pruned_loss=0.1182, over 5660662.29 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3607, pruned_loss=0.1127, over 5668354.83 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3735, pruned_loss=0.1187, over 5674884.31 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:49:53,018 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1257861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:49:56,230 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9495, 1.1410, 1.1005, 0.9335], device='cuda:1'), covar=tensor([0.2702, 0.2921, 0.1758, 0.2327], device='cuda:1'), in_proj_covar=tensor([0.2066, 0.2026, 0.1940, 0.2082], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 13:50:13,803 INFO [optim.py:369] (1/2) 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 13:50:25,411 INFO [train.py:968] (1/2) Epoch 28, batch 27000, giga_loss[loss=0.319, simple_loss=0.3859, pruned_loss=0.1261, over 28902.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3762, pruned_loss=0.1204, over 5665514.16 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3609, pruned_loss=0.1128, over 5672232.32 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3769, pruned_loss=0.1208, over 5673376.79 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:50:25,411 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 13:50:34,436 INFO [train.py:1012] (1/2) Epoch 28, validation: loss=0.2021, simple_loss=0.3093, pruned_loss=0.04742, over 944034.00 frames. 
+2023-03-14 13:50:34,437 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 13:51:01,325 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257927.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:51:13,315 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1257938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:51:24,080 INFO [train.py:968] (1/2) Epoch 28, batch 27050, giga_loss[loss=0.3327, simple_loss=0.3898, pruned_loss=0.1378, over 28953.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3785, pruned_loss=0.1233, over 5668750.48 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3608, pruned_loss=0.1128, over 5673436.16 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3792, pruned_loss=0.1237, over 5673844.36 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:51:54,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.83 vs. limit=5.0
+2023-03-14 13:51:57,817 INFO [optim.py:369] (1/2) 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 13:52:12,871 INFO [train.py:968] (1/2) Epoch 28, batch 27100, giga_loss[loss=0.3466, simple_loss=0.3996, pruned_loss=0.1468, over 28543.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3795, pruned_loss=0.1256, over 5649007.59 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3607, pruned_loss=0.1128, over 5676618.03 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3807, pruned_loss=0.1262, over 5649766.86 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:52:19,520 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:52:21,847 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:52:44,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5666, 1.7197, 1.2405, 1.3306], device='cuda:1'), covar=tensor([0.1090, 0.0652, 0.1145, 0.1184], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0455, 0.0526, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 13:52:48,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:53:00,336 INFO [train.py:968] (1/2) Epoch 28, batch 27150, giga_loss[loss=0.2754, simple_loss=0.3576, pruned_loss=0.09663, over 28889.00 frames. ], tot_loss[loss=0.3131, simple_loss=0.3775, pruned_loss=0.1243, over 5663750.92 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3606, pruned_loss=0.1128, over 5681965.33 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3789, pruned_loss=0.1251, over 5659055.15 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 13:53:17,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.02 vs. limit=5.0
+2023-03-14 13:53:36,283 INFO [optim.py:369] (1/2) 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 13:53:40,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3619, 1.4727, 3.4715, 3.2881], device='cuda:1'), covar=tensor([0.1386, 0.2447, 0.0471, 0.1073], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0675, 0.1011, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 13:53:46,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3039, 1.3659, 1.2197, 1.5332], device='cuda:1'), covar=tensor([0.0803, 0.0379, 0.0372, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 13:53:49,348 INFO [train.py:968] (1/2) Epoch 28, batch 27200, giga_loss[loss=0.3461, simple_loss=0.3891, pruned_loss=0.1516, over 26601.00 frames. ], tot_loss[loss=0.3116, simple_loss=0.3763, pruned_loss=0.1235, over 5650252.71 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.36, pruned_loss=0.1125, over 5685777.38 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3783, pruned_loss=0.1247, over 5642570.54 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 13:54:29,440 INFO [train.py:968] (1/2) Epoch 28, batch 27250, giga_loss[loss=0.2779, simple_loss=0.3618, pruned_loss=0.09706, over 29004.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.3749, pruned_loss=0.1208, over 5662939.05 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5684667.78 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.377, pruned_loss=0.1218, over 5657518.62 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:54:49,089 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:54:52,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8687, 3.6899, 3.5237, 1.8320], device='cuda:1'), covar=tensor([0.0804, 0.0923, 0.0895, 0.2048], device='cuda:1'), in_proj_covar=tensor([0.1321, 0.1218, 0.1028, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 13:55:07,106 INFO [optim.py:369] (1/2) 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 13:55:16,355 INFO [train.py:968] (1/2) Epoch 28, batch 27300, giga_loss[loss=0.3218, simple_loss=0.3909, pruned_loss=0.1263, over 28862.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.374, pruned_loss=0.1193, over 5659952.08 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3599, pruned_loss=0.1128, over 5681386.76 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3762, pruned_loss=0.1202, over 5657472.68 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:55:28,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9116, 2.2213, 1.5289, 1.7121], device='cuda:1'), covar=tensor([0.1149, 0.0705, 0.1075, 0.1217], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0455, 0.0524, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 13:56:07,051 INFO [train.py:968] (1/2) Epoch 28, batch 27350, giga_loss[loss=0.2741, simple_loss=0.3502, pruned_loss=0.099, over 28970.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3737, pruned_loss=0.1194, over 5658348.69 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5677483.02 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3762, pruned_loss=0.1205, over 5658971.69 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:56:24,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2358, 1.4803, 1.4955, 1.1147], device='cuda:1'), covar=tensor([0.1479, 0.2562, 0.1305, 0.1567], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0720, 0.0981, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 13:56:33,196 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:56:53,998 INFO [train.py:968] (1/2) Epoch 28, batch 27400, giga_loss[loss=0.3078, simple_loss=0.3741, pruned_loss=0.1208, over 28958.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3746, pruned_loss=0.1203, over 5666073.35 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3599, pruned_loss=0.1129, over 5680651.16 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3764, pruned_loss=0.1211, over 5663489.21 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:56:56,989 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258302.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:57:07,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:57:46,183 INFO [train.py:968] (1/2) Epoch 28, batch 27450, giga_loss[loss=0.2868, simple_loss=0.3601, pruned_loss=0.1067, over 28806.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3722, pruned_loss=0.1203, over 5658612.36 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3594, pruned_loss=0.1126, over 5684113.20 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3742, pruned_loss=0.1213, over 5653223.76 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:58:23,748 INFO [optim.py:369] (1/2) 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 13:58:35,936 INFO [train.py:968] (1/2) Epoch 28, batch 27500, libri_loss[loss=0.3156, simple_loss=0.3901, pruned_loss=0.1205, over 29546.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3717, pruned_loss=0.1212, over 5648753.02 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1126, over 5687598.51 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3733, pruned_loss=0.122, over 5640811.37 frames. ], batch size: 83, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:59:16,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258440.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 13:59:20,745 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258445.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:59:26,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258448.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 13:59:26,637 INFO [train.py:968] (1/2) Epoch 28, batch 27550, giga_loss[loss=0.4172, simple_loss=0.4348, pruned_loss=0.1998, over 26600.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.371, pruned_loss=0.1216, over 5656539.68 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5693520.21 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3725, pruned_loss=0.1224, over 5643818.64 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 13:59:31,707 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:33,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:45,258 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:49,159 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 13:59:56,324 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 13:59:56,511 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.50 vs. limit=5.0
+2023-03-14 14:00:06,086 INFO [train.py:968] (1/2) Epoch 28, batch 27600, giga_loss[loss=0.3333, simple_loss=0.3681, pruned_loss=0.1493, over 23539.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3692, pruned_loss=0.121, over 5658492.22 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5700600.99 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3709, pruned_loss=0.122, over 5639894.00 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:00:26,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6211, 2.0667, 1.3159, 0.8505], device='cuda:1'), covar=tensor([0.7856, 0.3962, 0.3405, 0.7208], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1749, 0.1662, 0.1508], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 14:00:41,699 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 27650, giga_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.0957, over 29037.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3676, pruned_loss=0.1199, over 5662979.69 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3593, pruned_loss=0.1126, over 5703204.01 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3697, pruned_loss=0.1212, over 5644217.20 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:01:02,491 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3213, 3.1663, 3.0014, 1.4437], device='cuda:1'), covar=tensor([0.0947, 0.1052, 0.0929, 0.2281], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1225, 0.1035, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 14:01:21,585 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1258592.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:01:29,898 INFO [train.py:968] (1/2) Epoch 28, batch 27700, giga_loss[loss=0.2572, simple_loss=0.3421, pruned_loss=0.08618, over 28298.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3638, pruned_loss=0.1156, over 5668844.72 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5705504.90 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3659, pruned_loss=0.1169, over 5650488.69 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:02:09,705 INFO [train.py:968] (1/2) Epoch 28, batch 27750, giga_loss[loss=0.2829, simple_loss=0.3565, pruned_loss=0.1047, over 28679.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3608, pruned_loss=0.1122, over 5669101.53 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5699839.52 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3625, pruned_loss=0.1134, over 5657032.98 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:02:10,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:02:46,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:02:49,838 INFO [optim.py:369] (1/2) 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 14:02:50,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:03:01,017 INFO [train.py:968] (1/2) Epoch 28, batch 27800, giga_loss[loss=0.2838, simple_loss=0.3557, pruned_loss=0.1059, over 28581.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3611, pruned_loss=0.1127, over 5662968.45 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5704114.12 frames. ], giga_tot_loss[loss=0.2951, simple_loss=0.3627, pruned_loss=0.1138, over 5648942.52 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:03:18,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:03:53,077 INFO [train.py:968] (1/2) Epoch 28, batch 27850, giga_loss[loss=0.2684, simple_loss=0.3467, pruned_loss=0.09502, over 28945.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.359, pruned_loss=0.1118, over 5666933.65 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3595, pruned_loss=0.1125, over 5707546.66 frames. ], giga_tot_loss[loss=0.2924, simple_loss=0.3599, pruned_loss=0.1125, over 5652005.28 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:04:08,858 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-14 14:04:32,549 INFO [optim.py:369] (1/2) 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 14:04:39,427 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:04:42,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:04:43,722 INFO [train.py:968] (1/2) Epoch 28, batch 27900, giga_loss[loss=0.3083, simple_loss=0.3752, pruned_loss=0.1208, over 28654.00 frames. ], tot_loss[loss=0.2909, simple_loss=0.3578, pruned_loss=0.112, over 5667831.74 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1123, over 5713621.05 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3586, pruned_loss=0.1127, over 5649172.21 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:04:59,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258815.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:05:08,523 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:05:29,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:05:29,886 INFO [train.py:968] (1/2) Epoch 28, batch 27950, giga_loss[loss=0.3096, simple_loss=0.376, pruned_loss=0.1216, over 28289.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3608, pruned_loss=0.1131, over 5683856.33 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3594, pruned_loss=0.1123, over 5716677.58 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3614, pruned_loss=0.1136, over 5665632.63 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:05:42,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3251, 1.4557, 1.6259, 1.3955], device='cuda:1'), covar=tensor([0.1700, 0.1348, 0.1647, 0.1402], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0761, 0.0734, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 14:06:05,336 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 28000, giga_loss[loss=0.2771, simple_loss=0.3563, pruned_loss=0.09893, over 28768.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3628, pruned_loss=0.1144, over 5669517.77 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5715631.45 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3633, pruned_loss=0.1145, over 5653304.88 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:06:58,591 INFO [train.py:968] (1/2) Epoch 28, batch 28050, giga_loss[loss=0.402, simple_loss=0.4309, pruned_loss=0.1865, over 26445.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3645, pruned_loss=0.1155, over 5670337.15 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5722235.63 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3652, pruned_loss=0.1159, over 5650045.98 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:07:07,151 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258958.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:07:10,030 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258961.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:07:14,480 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1258967.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:07:35,749 INFO [optim.py:369] (1/2) 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 14:07:37,123 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1258990.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:07:37,732 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1258991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:07:41,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1258994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:07:45,157 INFO [train.py:968] (1/2) Epoch 28, batch 28100, giga_loss[loss=0.2909, simple_loss=0.3513, pruned_loss=0.1153, over 28620.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3661, pruned_loss=0.1169, over 5654381.56 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3599, pruned_loss=0.1127, over 5712352.99 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1171, over 5645390.49 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:08:06,328 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 28150, giga_loss[loss=0.312, simple_loss=0.3802, pruned_loss=0.1219, over 28404.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3675, pruned_loss=0.1178, over 5661787.50 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3602, pruned_loss=0.1129, over 5709008.37 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3675, pruned_loss=0.1179, over 5656934.62 frames. ], batch size: 369, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:09:08,807 INFO [optim.py:369] (1/2) 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 14:09:13,005 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 14:09:15,840 INFO [train.py:968] (1/2) Epoch 28, batch 28200, libri_loss[loss=0.3068, simple_loss=0.376, pruned_loss=0.1187, over 29373.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3687, pruned_loss=0.1183, over 5668487.40 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3606, pruned_loss=0.1132, over 5714450.50 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3685, pruned_loss=0.1183, over 5658407.35 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:09:28,147 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1259110.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:09:32,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1259113.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:09:59,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1259142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:10:01,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 14:10:05,562 INFO [train.py:968] (1/2) Epoch 28, batch 28250, giga_loss[loss=0.2663, simple_loss=0.3438, pruned_loss=0.09444, over 28957.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3702, pruned_loss=0.119, over 5669631.63 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5717684.63 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3701, pruned_loss=0.119, over 5657806.41 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:10:40,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-14 14:10:45,779 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 28300, giga_loss[loss=0.305, simple_loss=0.373, pruned_loss=0.1185, over 29012.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3711, pruned_loss=0.1201, over 5655932.01 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.3605, pruned_loss=0.1129, over 5716655.91 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3715, pruned_loss=0.1207, over 5645824.23 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:10:54,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.79 vs. limit=5.0
+2023-03-14 14:11:26,581 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 28350, giga_loss[loss=0.2787, simple_loss=0.3516, pruned_loss=0.1029, over 28894.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3716, pruned_loss=0.1206, over 5655777.40 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3602, pruned_loss=0.1126, over 5719713.21 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3723, pruned_loss=0.1213, over 5644343.31 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:12:24,807 INFO [optim.py:369] (1/2) 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 14:12:30,597 INFO [train.py:968] (1/2) Epoch 28, batch 28400, giga_loss[loss=0.3596, simple_loss=0.4135, pruned_loss=0.1529, over 28715.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3709, pruned_loss=0.1184, over 5658726.28 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3598, pruned_loss=0.1125, over 5713306.48 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3722, pruned_loss=0.1194, over 5653998.80 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:13:08,758 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.87 vs. limit=5.0
+2023-03-14 14:13:19,208 INFO [train.py:968] (1/2) Epoch 28, batch 28450, giga_loss[loss=0.2887, simple_loss=0.3593, pruned_loss=0.1091, over 28516.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3708, pruned_loss=0.1187, over 5657296.02 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3601, pruned_loss=0.1126, over 5707196.69 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3719, pruned_loss=0.1195, over 5656866.67 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:13:25,005 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-14 14:13:58,840 INFO [optim.py:369] (1/2) 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 14:14:07,331 INFO [train.py:968] (1/2) Epoch 28, batch 28500, giga_loss[loss=0.2715, simple_loss=0.3492, pruned_loss=0.09689, over 28914.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3687, pruned_loss=0.1182, over 5656043.95 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3603, pruned_loss=0.1129, over 5702257.92 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3696, pruned_loss=0.1188, over 5659418.99 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:14:17,426 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-14 14:14:54,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 14:15:05,761 INFO [train.py:968] (1/2) Epoch 28, batch 28550, libri_loss[loss=0.2966, simple_loss=0.3647, pruned_loss=0.1142, over 29292.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3679, pruned_loss=0.1183, over 5668628.29 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.36, pruned_loss=0.1126, over 5704871.47 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3692, pruned_loss=0.1191, over 5667659.26 frames. ], batch size: 94, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:15:49,580 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 28600, libri_loss[loss=0.2342, simple_loss=0.3086, pruned_loss=0.07994, over 29399.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3666, pruned_loss=0.118, over 5669432.23 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.36, pruned_loss=0.1127, over 5706970.65 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3677, pruned_loss=0.1188, over 5665915.74 frames. ], batch size: 67, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:16:24,305 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 14:16:40,890 INFO [train.py:968] (1/2) Epoch 28, batch 28650, giga_loss[loss=0.4426, simple_loss=0.4425, pruned_loss=0.2214, over 23417.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.368, pruned_loss=0.1197, over 5665972.37 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1128, over 5701007.78 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3689, pruned_loss=0.1202, over 5667414.88 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:16:52,927 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:17:22,005 INFO [optim.py:369] (1/2) 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 14:17:31,281 INFO [train.py:968] (1/2) Epoch 28, batch 28700, giga_loss[loss=0.3515, simple_loss=0.4041, pruned_loss=0.1494, over 28519.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3677, pruned_loss=0.1201, over 5655818.59 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3594, pruned_loss=0.1124, over 5703963.09 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3691, pruned_loss=0.1211, over 5653688.38 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:17:40,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1259608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:18:18,289 INFO [train.py:968] (1/2) Epoch 28, batch 28750, giga_loss[loss=0.2765, simple_loss=0.3542, pruned_loss=0.09947, over 29067.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3675, pruned_loss=0.12, over 5654998.45 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3593, pruned_loss=0.1122, over 5706495.65 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.369, pruned_loss=0.121, over 5650304.91 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:18:41,424 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:18:46,402 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-14 14:18:57,209 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 28800, giga_loss[loss=0.3665, simple_loss=0.4194, pruned_loss=0.1568, over 28675.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3687, pruned_loss=0.121, over 5659578.31 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3587, pruned_loss=0.1118, over 5706985.46 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3706, pruned_loss=0.1225, over 5653822.44 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:19:10,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:19:13,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4523, 2.0709, 1.4857, 0.7626], device='cuda:1'), covar=tensor([0.6346, 0.3258, 0.4518, 0.6896], device='cuda:1'), in_proj_covar=tensor([0.1863, 0.1754, 0.1668, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 14:19:52,552 INFO [train.py:968] (1/2) Epoch 28, batch 28850, giga_loss[loss=0.29, simple_loss=0.3552, pruned_loss=0.1124, over 28843.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3685, pruned_loss=0.1215, over 5646191.14 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3582, pruned_loss=0.1116, over 5709266.73 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.1232, over 5638321.82 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:19:54,491 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1259751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:19:56,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1259754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:20:00,029 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1259756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:20:23,812 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1259783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:20:25,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3507, 1.8348, 1.4707, 1.5059], device='cuda:1'), covar=tensor([0.0755, 0.0355, 0.0334, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 14:20:31,106 INFO [optim.py:369] (1/2) 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 14:20:37,340 INFO [train.py:968] (1/2) Epoch 28, batch 28900, libri_loss[loss=0.2544, simple_loss=0.3313, pruned_loss=0.08874, over 29538.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3687, pruned_loss=0.1221, over 5641843.68 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3586, pruned_loss=0.1119, over 5700561.23 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3703, pruned_loss=0.1232, over 5642787.51 frames. ], batch size: 80, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:21:20,620 INFO [train.py:968] (1/2) Epoch 28, batch 28950, giga_loss[loss=0.2941, simple_loss=0.3671, pruned_loss=0.1106, over 28973.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.369, pruned_loss=0.1222, over 5650171.79 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5704587.43 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3705, pruned_loss=0.1234, over 5646380.51 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:22:00,893 INFO [optim.py:369] (1/2) 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 14:22:03,118 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4319, 3.3739, 1.5659, 1.5477], device='cuda:1'), covar=tensor([0.1008, 0.0405, 0.0906, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0574, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 14:22:11,590 INFO [train.py:968] (1/2) Epoch 28, batch 29000, giga_loss[loss=0.3431, simple_loss=0.3994, pruned_loss=0.1434, over 28825.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3701, pruned_loss=0.1228, over 5635466.82 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3587, pruned_loss=0.112, over 5698117.95 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3714, pruned_loss=0.1238, over 5637121.95 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:22:45,912 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 29050, giga_loss[loss=0.2516, simple_loss=0.3338, pruned_loss=0.08473, over 28495.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3704, pruned_loss=0.1224, over 5643717.14 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5700801.30 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3717, pruned_loss=0.1234, over 5642048.97 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:23:36,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1793, 5.0240, 4.7858, 2.4448], device='cuda:1'), covar=tensor([0.0513, 0.0634, 0.0730, 0.1814], device='cuda:1'), in_proj_covar=tensor([0.1330, 0.1225, 0.1033, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 14:23:37,436 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 29100, giga_loss[loss=0.2819, simple_loss=0.3578, pruned_loss=0.103, over 28831.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.3726, pruned_loss=0.1238, over 5653748.98 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3587, pruned_loss=0.1119, over 5699914.37 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3737, pruned_loss=0.1248, over 5652423.16 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:24:27,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:24:28,620 INFO [train.py:968] (1/2) Epoch 28, batch 29150, giga_loss[loss=0.3629, simple_loss=0.4048, pruned_loss=0.1605, over 27530.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3737, pruned_loss=0.1245, over 5659723.65 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3588, pruned_loss=0.1119, over 5692844.77 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3747, pruned_loss=0.1255, over 5665161.39 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:24:54,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:24:57,824 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:24:58,388 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:25:07,597 INFO [optim.py:369] (1/2) 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 14:25:13,364 INFO [train.py:968] (1/2) Epoch 28, batch 29200, giga_loss[loss=0.3966, simple_loss=0.4204, pruned_loss=0.1863, over 26547.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.374, pruned_loss=0.1248, over 5663753.16 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3586, pruned_loss=0.1117, over 5695750.65 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3751, pruned_loss=0.1259, over 5665089.22 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:25:23,179 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260109.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:25:44,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4854, 5.2785, 4.9997, 2.6169], device='cuda:1'), covar=tensor([0.0492, 0.0668, 0.0770, 0.1787], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1224, 0.1032, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 14:25:45,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:26:02,691 INFO [train.py:968] (1/2) Epoch 28, batch 29250, giga_loss[loss=0.3741, simple_loss=0.4107, pruned_loss=0.1688, over 27932.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3754, pruned_loss=0.1251, over 5646894.37 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3588, pruned_loss=0.112, over 5680881.05 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3765, pruned_loss=0.1261, over 5660619.32 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:26:38,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5843, 4.4111, 4.2029, 2.1064], device='cuda:1'), covar=tensor([0.0558, 0.0714, 0.0751, 0.1961], device='cuda:1'), in_proj_covar=tensor([0.1328, 0.1223, 0.1031, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 14:26:40,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:26:45,804 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:26:46,214 INFO [optim.py:369] (1/2) 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 14:26:50,345 INFO [train.py:968] (1/2) Epoch 28, batch 29300, giga_loss[loss=0.2759, simple_loss=0.3501, pruned_loss=0.1008, over 29007.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3735, pruned_loss=0.1228, over 5649848.58 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3587, pruned_loss=0.1119, over 5687669.72 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.375, pruned_loss=0.1239, over 5653819.73 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:26:58,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:08,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:11,341 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:13,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:33,744 INFO [train.py:968] (1/2) Epoch 28, batch 29350, giga_loss[loss=0.3163, simple_loss=0.3722, pruned_loss=0.1302, over 27867.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3711, pruned_loss=0.1207, over 5659523.74 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3586, pruned_loss=0.1119, over 5687931.86 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3726, pruned_loss=0.1219, over 5661569.85 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:27:42,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:56,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:27:59,185 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:28:14,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:28:17,953 INFO [train.py:968] (1/2) Epoch 28, batch 29400, giga_loss[loss=0.3359, simple_loss=0.3967, pruned_loss=0.1375, over 27990.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3705, pruned_loss=0.1208, over 5656650.81 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3588, pruned_loss=0.112, over 5695171.40 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3721, pruned_loss=0.1221, over 5650515.53 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:28:21,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2663, 2.4945, 1.2966, 1.3204], device='cuda:1'), covar=tensor([0.1008, 0.0465, 0.0937, 0.1410], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0575, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 14:28:23,328 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:29:00,498 INFO [train.py:968] (1/2) Epoch 28, batch 29450, giga_loss[loss=0.3231, simple_loss=0.3915, pruned_loss=0.1273, over 28930.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3711, pruned_loss=0.1206, over 5668867.90 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3588, pruned_loss=0.1121, over 5697381.61 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3726, pruned_loss=0.1217, over 5661009.49 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:29:47,321 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 29500, giga_loss[loss=0.2812, simple_loss=0.3476, pruned_loss=0.1074, over 28733.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3731, pruned_loss=0.1227, over 5661724.01 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3586, pruned_loss=0.112, over 5699932.79 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3749, pruned_loss=0.1238, over 5652377.80 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:30:09,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:30:41,057 INFO [train.py:968] (1/2) Epoch 28, batch 29550, giga_loss[loss=0.3036, simple_loss=0.3743, pruned_loss=0.1165, over 29035.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.3738, pruned_loss=0.1243, over 5664504.01 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3591, pruned_loss=0.1126, over 5702240.12 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3749, pruned_loss=0.1249, over 5654693.36 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:31:00,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9859, 2.2886, 1.7896, 2.3538], device='cuda:1'), covar=tensor([0.2562, 0.2632, 0.2995, 0.2312], device='cuda:1'), in_proj_covar=tensor([0.1609, 0.1157, 0.1419, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 14:31:24,209 INFO [optim.py:369] (1/2) 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 14:31:25,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6531, 2.3838, 1.7575, 0.8086], device='cuda:1'), covar=tensor([0.6475, 0.3231, 0.4545, 0.7301], device='cuda:1'), in_proj_covar=tensor([0.1858, 0.1751, 0.1665, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 14:31:28,485 INFO [train.py:968] (1/2) Epoch 28, batch 29600, giga_loss[loss=0.3315, simple_loss=0.3949, pruned_loss=0.1341, over 29002.00 frames. ], tot_loss[loss=0.3144, simple_loss=0.3759, pruned_loss=0.1264, over 5656124.45 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3591, pruned_loss=0.1126, over 5703298.92 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3769, pruned_loss=0.127, over 5647387.06 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:31:43,687 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:32:09,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.97 vs. limit=2.0
+2023-03-14 14:32:17,993 INFO [train.py:968] (1/2) Epoch 28, batch 29650, giga_loss[loss=0.2962, simple_loss=0.3698, pruned_loss=0.1112, over 28984.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3759, pruned_loss=0.1263, over 5653662.93 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1125, over 5694892.83 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.377, pruned_loss=0.127, over 5653059.86 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:32:54,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:32:59,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7113, 0.9342, 0.7633, 0.1727], device='cuda:1'), covar=tensor([0.3170, 0.2780, 0.2861, 0.4745], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1751, 0.1664, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 14:33:03,231 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 29700, giga_loss[loss=0.3021, simple_loss=0.3642, pruned_loss=0.12, over 28625.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3756, pruned_loss=0.1259, over 5641870.92 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1124, over 5689006.66 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3767, pruned_loss=0.1267, over 5646196.13 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:33:08,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:33:23,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:33:52,010 INFO [train.py:968] (1/2) Epoch 28, batch 29750, libri_loss[loss=0.3021, simple_loss=0.3743, pruned_loss=0.1149, over 29522.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3736, pruned_loss=0.1234, over 5666175.10 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3586, pruned_loss=0.1121, over 5694254.52 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3752, pruned_loss=0.1247, over 5663703.02 frames. ], batch size: 84, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:34:00,433 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:34:10,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:34:11,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1260669.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:34:12,479 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:34:19,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2431, 1.3506, 3.9314, 3.3683], device='cuda:1'), covar=tensor([0.1814, 0.2875, 0.0448, 0.0919], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0680, 0.1015, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 14:34:21,582 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0292, 3.8771, 3.6925, 1.8488], device='cuda:1'), covar=tensor([0.0658, 0.0770, 0.0742, 0.2250], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1224, 0.1032, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 14:34:30,620 INFO [optim.py:369] (1/2) 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 14:34:37,545 INFO [train.py:968] (1/2) Epoch 28, batch 29800, giga_loss[loss=0.3374, simple_loss=0.397, pruned_loss=0.1389, over 28957.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.373, pruned_loss=0.1226, over 5661870.29 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3589, pruned_loss=0.1123, over 5696945.05 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3744, pruned_loss=0.1236, over 5656925.44 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:34:54,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9230, 1.1106, 1.0512, 0.8754], device='cuda:1'), covar=tensor([0.2162, 0.2531, 0.1628, 0.2241], device='cuda:1'), in_proj_covar=tensor([0.2077, 0.2045, 0.1950, 0.2097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 14:35:05,629 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:35:08,503 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:35:28,180 INFO [train.py:968] (1/2) Epoch 28, batch 29850, giga_loss[loss=0.2938, simple_loss=0.3628, pruned_loss=0.1124, over 28910.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3708, pruned_loss=0.1203, over 5670250.82 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3589, pruned_loss=0.1122, over 5699502.40 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3721, pruned_loss=0.1213, over 5663631.58 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:35:38,825 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:08,943 INFO [optim.py:369] (1/2) 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 14:36:09,250 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:15,556 INFO [train.py:968] (1/2) Epoch 28, batch 29900, giga_loss[loss=0.3597, simple_loss=0.3894, pruned_loss=0.165, over 23513.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3683, pruned_loss=0.119, over 5658877.95 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3585, pruned_loss=0.1121, over 5696279.92 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.37, pruned_loss=0.1202, over 5655257.65 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:36:19,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-14 14:36:20,528 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-14 14:36:27,862 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:29,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:58,132 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:36:58,342 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-14 14:37:00,486 INFO [train.py:968] (1/2) Epoch 28, batch 29950, giga_loss[loss=0.2628, simple_loss=0.3386, pruned_loss=0.09347, over 28843.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3676, pruned_loss=0.1191, over 5655437.41 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3586, pruned_loss=0.1121, over 5688592.53 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.369, pruned_loss=0.1201, over 5658235.80 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:37:02,464 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-14 14:37:15,585 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5914, 1.6877, 1.7918, 1.3864], device='cuda:1'), covar=tensor([0.1834, 0.2598, 0.1521, 0.1762], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0722, 0.0986, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 14:37:38,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:37:41,378 INFO [optim.py:369] (1/2) 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 14:37:49,665 INFO [train.py:968] (1/2) Epoch 28, batch 30000, giga_loss[loss=0.2737, simple_loss=0.3381, pruned_loss=0.1046, over 28605.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3656, pruned_loss=0.1185, over 5658980.67 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3592, pruned_loss=0.1126, over 5690719.92 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3663, pruned_loss=0.1189, over 5658864.68 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:37:49,665 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 14:37:58,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5251, 1.9223, 1.5136, 1.3760], device='cuda:1'), covar=tensor([0.3186, 0.3065, 0.3412, 0.2727], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1155, 0.1419, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 14:37:58,751 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 14:38:29,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-14 14:38:30,322 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1260936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:38:32,351 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1260939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:38:42,398 INFO [train.py:968] (1/2) Epoch 28, batch 30050, giga_loss[loss=0.2914, simple_loss=0.3534, pruned_loss=0.1147, over 28920.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3641, pruned_loss=0.1182, over 5653028.78 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.36, pruned_loss=0.1132, over 5669133.22 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3641, pruned_loss=0.1183, over 5671325.68 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:38:47,539 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5721, 1.2897, 4.8421, 3.6517], device='cuda:1'), covar=tensor([0.1723, 0.2882, 0.0422, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0679, 0.1015, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 14:38:57,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1260968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:39:01,249 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.5154, 5.3626, 5.1023, 2.3500], device='cuda:1'), covar=tensor([0.0477, 0.0631, 0.0716, 0.1878], device='cuda:1'), in_proj_covar=tensor([0.1326, 0.1221, 0.1030, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 14:39:02,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.08 vs. limit=5.0
+2023-03-14 14:39:04,979 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1260975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:39:19,870 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 28, batch 30100, libri_loss[loss=0.3076, simple_loss=0.3759, pruned_loss=0.1197, over 29526.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3632, pruned_loss=0.1181, over 5669705.23 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3603, pruned_loss=0.1133, over 5672126.74 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.363, pruned_loss=0.1181, over 5681656.80 frames. ], batch size: 82, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:39:40,408 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261012.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:39:59,343 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:39:59,959 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:01,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:09,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:12,830 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261044.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:40:17,062 INFO [train.py:968] (1/2) Epoch 28, batch 30150, giga_loss[loss=0.2828, simple_loss=0.3573, pruned_loss=0.1041, over 27988.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3636, pruned_loss=0.1184, over 5673310.29 frames. ], libri_tot_loss[loss=0.294, simple_loss=0.3607, pruned_loss=0.1136, over 5676582.15 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3632, pruned_loss=0.1182, over 5679058.52 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:40:31,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:40:35,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2409, 2.4798, 1.2923, 1.4171], device='cuda:1'), covar=tensor([0.1046, 0.0456, 0.0992, 0.1402], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0576, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 14:40:56,798 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 30200, giga_loss[loss=0.2716, simple_loss=0.3522, pruned_loss=0.09557, over 28687.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3627, pruned_loss=0.1157, over 5678943.99 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3609, pruned_loss=0.1137, over 5683563.93 frames. ], giga_tot_loss[loss=0.2968, simple_loss=0.3623, pruned_loss=0.1157, over 5677218.81 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:41:18,956 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:21,768 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:34,639 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:37,923 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:41:51,774 INFO [train.py:968] (1/2) Epoch 28, batch 30250, giga_loss[loss=0.2517, simple_loss=0.3362, pruned_loss=0.08363, over 28867.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3598, pruned_loss=0.1119, over 5672817.87 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3606, pruned_loss=0.1136, over 5689136.31 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3598, pruned_loss=0.112, over 5666402.79 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:41:53,303 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:09,016 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:19,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:22,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:26,083 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:28,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261187.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:42:28,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:30,334 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261190.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:42:34,080 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 30300, giga_loss[loss=0.2726, simple_loss=0.3492, pruned_loss=0.09805, over 27580.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3576, pruned_loss=0.1098, over 5659629.48 frames. ], libri_tot_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1141, over 5677338.98 frames. ], giga_tot_loss[loss=0.2878, simple_loss=0.3572, pruned_loss=0.1092, over 5664259.50 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 1.0
+2023-03-14 14:42:47,402 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:56,839 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:42:58,673 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261219.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:43:24,954 INFO [train.py:968] (1/2) Epoch 28, batch 30350, giga_loss[loss=0.2602, simple_loss=0.343, pruned_loss=0.08867, over 28780.00 frames. ], tot_loss[loss=0.2826, simple_loss=0.3532, pruned_loss=0.106, over 5659237.47 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1136, over 5684599.31 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3537, pruned_loss=0.1058, over 5655475.86 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 1.0
+2023-03-14 14:43:28,656 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:43:32,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2098, 1.5513, 1.4810, 1.0839], device='cuda:1'), covar=tensor([0.1824, 0.2963, 0.1617, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0720, 0.0984, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 14:43:48,455 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:44:11,392 INFO [optim.py:369] (1/2) 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 14:44:15,586 INFO [train.py:968] (1/2) Epoch 28, batch 30400, giga_loss[loss=0.2688, simple_loss=0.3553, pruned_loss=0.09112, over 28559.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3495, pruned_loss=0.1021, over 5659097.97 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3591, pruned_loss=0.1133, over 5686545.17 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3504, pruned_loss=0.1021, over 5654208.99 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:44:25,547 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3725, 3.0548, 1.4542, 1.4924], device='cuda:1'), covar=tensor([0.0990, 0.0322, 0.0973, 0.1355], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0572, 0.0410, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 14:44:40,851 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.01 vs. limit=2.0
+2023-03-14 14:44:49,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2778, 2.6402, 1.2795, 1.5438], device='cuda:1'), covar=tensor([0.1044, 0.0396, 0.1011, 0.1373], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0573, 0.0410, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 14:45:00,635 INFO [train.py:968] (1/2) Epoch 28, batch 30450, giga_loss[loss=0.2644, simple_loss=0.3447, pruned_loss=0.09206, over 28599.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3483, pruned_loss=0.1, over 5647967.55 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3591, pruned_loss=0.1135, over 5689009.01 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3486, pruned_loss=0.09933, over 5640773.09 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:45:39,797 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261387.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:45:47,604 INFO [optim.py:369] (1/2) 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 14:45:50,602 INFO [train.py:968] (1/2) Epoch 28, batch 30500, giga_loss[loss=0.3039, simple_loss=0.3755, pruned_loss=0.1162, over 28619.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3501, pruned_loss=0.1012, over 5649209.85 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3585, pruned_loss=0.1133, over 5691867.92 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3506, pruned_loss=0.1004, over 5640137.83 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:46:14,854 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1261422.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:46:40,926 INFO [train.py:968] (1/2) Epoch 28, batch 30550, giga_loss[loss=0.2268, simple_loss=0.3132, pruned_loss=0.07017, over 28431.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3485, pruned_loss=0.1, over 5645847.38 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3581, pruned_loss=0.113, over 5695588.22 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3491, pruned_loss=0.09949, over 5634319.88 frames. ], batch size: 85, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:47:00,436 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6794, 1.9252, 1.5615, 1.9599], device='cuda:1'), covar=tensor([0.2839, 0.2766, 0.3136, 0.2538], device='cuda:1'), in_proj_covar=tensor([0.1606, 0.1152, 0.1422, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 14:47:27,344 INFO [optim.py:369] (1/2) 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 14:47:31,696 INFO [train.py:968] (1/2) Epoch 28, batch 30600, giga_loss[loss=0.2621, simple_loss=0.3328, pruned_loss=0.09572, over 27523.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3454, pruned_loss=0.09777, over 5643171.52 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3578, pruned_loss=0.1128, over 5694525.77 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.346, pruned_loss=0.09738, over 5634308.94 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:47:41,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5117, 1.7129, 1.2939, 1.3103], device='cuda:1'), covar=tensor([0.1026, 0.0520, 0.0975, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0454, 0.0526, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 14:48:03,355 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261530.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:48:06,022 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 30650, giga_loss[loss=0.2495, simple_loss=0.338, pruned_loss=0.08052, over 28978.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3446, pruned_loss=0.09726, over 5640180.23 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3572, pruned_loss=0.1125, over 5691511.39 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3452, pruned_loss=0.09676, over 5634343.27 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:48:31,502 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261562.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 14:49:04,339 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 30700, giga_loss[loss=0.2528, simple_loss=0.3322, pruned_loss=0.08672, over 27912.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3445, pruned_loss=0.09676, over 5637625.35 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3566, pruned_loss=0.1123, over 5685416.71 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3453, pruned_loss=0.09626, over 5636797.06 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:49:26,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8873, 3.7468, 3.5475, 1.7065], device='cuda:1'), covar=tensor([0.0739, 0.0835, 0.0842, 0.2245], device='cuda:1'), in_proj_covar=tensor([0.1310, 0.1210, 0.1018, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 14:49:31,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3801, 1.8125, 1.7621, 1.5951], device='cuda:1'), covar=tensor([0.2214, 0.2015, 0.2140, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.0505, 0.0756, 0.0729, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 14:49:33,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:49:50,900 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0384, 4.8975, 4.6236, 2.2285], device='cuda:1'), covar=tensor([0.0496, 0.0620, 0.0704, 0.2022], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.1208, 0.1016, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 14:49:53,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1261648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:49:53,986 INFO [train.py:968] (1/2) Epoch 28, batch 30750, giga_loss[loss=0.2416, simple_loss=0.3293, pruned_loss=0.07702, over 28545.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3424, pruned_loss=0.09513, over 5650545.64 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3559, pruned_loss=0.1122, over 5692880.59 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3431, pruned_loss=0.09422, over 5641395.66 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 14:50:40,042 INFO [optim.py:369] (1/2) 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 14:50:42,396 INFO [train.py:968] (1/2) Epoch 28, batch 30800, giga_loss[loss=0.2661, simple_loss=0.3448, pruned_loss=0.09371, over 28889.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3404, pruned_loss=0.09323, over 5652111.97 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3559, pruned_loss=0.1122, over 5688070.73 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3409, pruned_loss=0.09217, over 5647935.70 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:51:32,096 INFO [train.py:968] (1/2) Epoch 28, batch 30850, giga_loss[loss=0.2367, simple_loss=0.3245, pruned_loss=0.07439, over 28884.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3358, pruned_loss=0.09084, over 5642753.07 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3555, pruned_loss=0.1122, over 5693374.90 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3361, pruned_loss=0.0895, over 5633421.28 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:51:54,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:51:57,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:12,984 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:15,910 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:16,900 INFO [optim.py:369] (1/2) 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 14:52:17,995 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 30900, giga_loss[loss=0.2575, simple_loss=0.3407, pruned_loss=0.08709, over 28633.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3357, pruned_loss=0.09158, over 5645576.17 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.355, pruned_loss=0.1121, over 5688542.74 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3359, pruned_loss=0.09017, over 5641343.73 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:52:21,726 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:41,340 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:52:53,899 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 30950, giga_loss[loss=0.2605, simple_loss=0.3127, pruned_loss=0.1041, over 24188.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.334, pruned_loss=0.09124, over 5631397.53 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3547, pruned_loss=0.1121, over 5685324.24 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3338, pruned_loss=0.08949, over 5628889.93 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:53:58,636 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 31000, giga_loss[loss=0.279, simple_loss=0.3554, pruned_loss=0.1013, over 28903.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3361, pruned_loss=0.09209, over 5623884.55 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3547, pruned_loss=0.112, over 5686542.62 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3359, pruned_loss=0.09064, over 5620643.56 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:54:34,686 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 14:54:45,887 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1261940.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:54:47,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1261943.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:54:54,945 INFO [train.py:968] (1/2) Epoch 28, batch 31050, giga_loss[loss=0.3134, simple_loss=0.3729, pruned_loss=0.127, over 26773.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3392, pruned_loss=0.09286, over 5640578.10 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3546, pruned_loss=0.1121, over 5693315.22 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3383, pruned_loss=0.09072, over 5629414.62 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:54:59,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5366, 1.8338, 1.7694, 1.5700], device='cuda:1'), covar=tensor([0.3134, 0.2309, 0.1801, 0.2445], device='cuda:1'), in_proj_covar=tensor([0.2038, 0.2003, 0.1910, 0.2052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 14:55:09,121 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-14 14:55:19,366 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1261972.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 14:55:22,464 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4424, 1.8927, 1.7407, 1.6242], device='cuda:1'), covar=tensor([0.2214, 0.2361, 0.2069, 0.2239], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0751, 0.0724, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 14:55:47,132 INFO [optim.py:369] (1/2) 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 14:55:49,608 INFO [train.py:968] (1/2) Epoch 28, batch 31100, giga_loss[loss=0.2668, simple_loss=0.3543, pruned_loss=0.08966, over 28662.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3398, pruned_loss=0.09333, over 5643262.04 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3539, pruned_loss=0.1121, over 5680153.42 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3391, pruned_loss=0.09076, over 5643516.77 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:56:55,787 INFO [train.py:968] (1/2) Epoch 28, batch 31150, giga_loss[loss=0.2195, simple_loss=0.3063, pruned_loss=0.06635, over 29077.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3386, pruned_loss=0.09248, over 5655207.60 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3532, pruned_loss=0.1117, over 5682443.79 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3383, pruned_loss=0.0904, over 5653084.02 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:57:49,006 INFO [optim.py:369] (1/2) 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 14:57:53,332 INFO [train.py:968] (1/2) Epoch 28, batch 31200, giga_loss[loss=0.2505, simple_loss=0.3337, pruned_loss=0.08366, over 28911.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3358, pruned_loss=0.09023, over 5648770.12 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3531, pruned_loss=0.1118, over 5677328.48 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3354, pruned_loss=0.08802, over 5651135.10 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 14:58:50,517 INFO [train.py:968] (1/2) Epoch 28, batch 31250, giga_loss[loss=0.2327, simple_loss=0.3197, pruned_loss=0.07282, over 28895.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3349, pruned_loss=0.08859, over 5654613.30 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3524, pruned_loss=0.1113, over 5680462.24 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3345, pruned_loss=0.08648, over 5652639.99 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 14:59:43,702 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 14:59:50,592 INFO [optim.py:369] (1/2) 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 14:59:51,878 INFO [train.py:968] (1/2) Epoch 28, batch 31300, libri_loss[loss=0.2878, simple_loss=0.356, pruned_loss=0.1097, over 28670.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3326, pruned_loss=0.08781, over 5653724.03 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3522, pruned_loss=0.1112, over 5675321.62 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.332, pruned_loss=0.08571, over 5655217.30 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:00:01,608 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:00:49,748 INFO [train.py:968] (1/2) Epoch 28, batch 31350, giga_loss[loss=0.2671, simple_loss=0.3473, pruned_loss=0.09345, over 28457.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3313, pruned_loss=0.08759, over 5658274.19 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3523, pruned_loss=0.1114, over 5679515.58 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3304, pruned_loss=0.08531, over 5655299.24 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:01:47,547 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 31400, giga_loss[loss=0.2199, simple_loss=0.3141, pruned_loss=0.06286, over 28949.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3311, pruned_loss=0.08728, over 5664673.34 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3524, pruned_loss=0.1115, over 5680393.01 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3302, pruned_loss=0.08531, over 5661520.28 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:01:54,962 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 31450, giga_loss[loss=0.2333, simple_loss=0.3273, pruned_loss=0.06968, over 28521.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3329, pruned_loss=0.08802, over 5666182.58 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3518, pruned_loss=0.1111, over 5686819.00 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3322, pruned_loss=0.08611, over 5657300.43 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:02:47,983 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:02:51,454 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:03:00,393 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262362.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:03:25,169 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:03:46,538 INFO [train.py:968] (1/2) Epoch 28, batch 31500, giga_loss[loss=0.2765, simple_loss=0.3378, pruned_loss=0.1076, over 26927.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3342, pruned_loss=0.08798, over 5670143.50 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3517, pruned_loss=0.1111, over 5690563.05 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3333, pruned_loss=0.08596, over 5659388.87 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:04:13,032 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:04:49,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4918, 1.6705, 1.2226, 1.2623], device='cuda:1'), covar=tensor([0.1032, 0.0523, 0.0968, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0449, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 15:04:49,887 INFO [train.py:968] (1/2) Epoch 28, batch 31550, giga_loss[loss=0.2681, simple_loss=0.3436, pruned_loss=0.0963, over 28718.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3314, pruned_loss=0.08647, over 5669078.14 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.352, pruned_loss=0.1115, over 5689975.78 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3301, pruned_loss=0.08421, over 5660526.90 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:05:17,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5723, 1.7086, 1.7598, 1.6293], device='cuda:1'), covar=tensor([0.2591, 0.2344, 0.1753, 0.2278], device='cuda:1'), in_proj_covar=tensor([0.2041, 0.2005, 0.1911, 0.2059], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 15:05:46,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:05:54,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4501, 1.9925, 1.3795, 0.7989], device='cuda:1'), covar=tensor([0.7181, 0.3777, 0.4777, 0.7246], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1746, 0.1661, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 15:05:55,707 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 31600, giga_loss[loss=0.2788, simple_loss=0.351, pruned_loss=0.1033, over 28052.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3333, pruned_loss=0.08789, over 5674680.06 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3512, pruned_loss=0.1109, over 5693525.28 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3326, pruned_loss=0.08614, over 5664319.26 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:07:01,559 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:07:01,881 INFO [train.py:968] (1/2) Epoch 28, batch 31650, giga_loss[loss=0.2062, simple_loss=0.2891, pruned_loss=0.06168, over 24628.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3361, pruned_loss=0.08715, over 5663425.76 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.351, pruned_loss=0.1109, over 5696416.39 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3356, pruned_loss=0.08552, over 5652587.29 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:07:02,539 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:07:25,032 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:07:26,926 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:08:01,452 INFO [optim.py:369] (1/2) 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 15:08:04,420 INFO [train.py:968] (1/2) Epoch 28, batch 31700, giga_loss[loss=0.2507, simple_loss=0.3377, pruned_loss=0.0818, over 27819.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3382, pruned_loss=0.0864, over 5665853.30 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3509, pruned_loss=0.1108, over 5701593.03 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3376, pruned_loss=0.08455, over 5651847.86 frames. ], batch size: 474, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:08:04,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 15:08:07,829 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-14 15:08:13,997 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4568, 1.9783, 1.7838, 1.7748], device='cuda:1'), covar=tensor([0.2596, 0.2680, 0.2518, 0.2433], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0751, 0.0724, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 15:08:14,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7627, 2.4712, 1.5452, 1.1558], device='cuda:1'), covar=tensor([0.9663, 0.4283, 0.5268, 0.7215], device='cuda:1'), in_proj_covar=tensor([0.1851, 0.1739, 0.1658, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 15:09:00,901 INFO [train.py:968] (1/2) Epoch 28, batch 31750, giga_loss[loss=0.241, simple_loss=0.3345, pruned_loss=0.07379, over 28653.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.339, pruned_loss=0.08631, over 5649881.96 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3509, pruned_loss=0.111, over 5687547.30 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3382, pruned_loss=0.08408, over 5650622.58 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:09:17,345 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1262663.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 15:09:37,023 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:09:59,889 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9879, 1.2830, 1.0164, 0.2064], device='cuda:1'), covar=tensor([0.4272, 0.3447, 0.5093, 0.7446], device='cuda:1'), in_proj_covar=tensor([0.1849, 0.1738, 0.1658, 0.1504], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 15:10:00,156 INFO [train.py:968] (1/2) Epoch 28, batch 31800, giga_loss[loss=0.2686, simple_loss=0.3497, pruned_loss=0.09381, over 28428.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.3381, pruned_loss=0.08581, over 5654057.48 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3508, pruned_loss=0.1111, over 5689430.12 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3372, pruned_loss=0.08321, over 5651809.73 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:10:10,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4616, 1.7883, 1.7346, 1.4159], device='cuda:1'), covar=tensor([0.2924, 0.2081, 0.1958, 0.2623], device='cuda:1'), in_proj_covar=tensor([0.2028, 0.1992, 0.1898, 0.2048], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 15:10:13,074 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:10:16,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:10:45,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262737.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 15:10:54,860 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:11:00,507 INFO [train.py:968] (1/2) Epoch 28, batch 31850, libri_loss[loss=0.2541, simple_loss=0.3271, pruned_loss=0.09061, over 29691.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3398, pruned_loss=0.08826, over 5654891.88 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.351, pruned_loss=0.1112, over 5693369.06 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3386, pruned_loss=0.08548, over 5648362.16 frames. ], batch size: 88, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:11:07,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.58 vs. limit=2.0
+2023-03-14 15:12:03,217 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:12:06,490 INFO [optim.py:369] (1/2) 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 15:12:08,089 INFO [train.py:968] (1/2) Epoch 28, batch 31900, libri_loss[loss=0.2594, simple_loss=0.3286, pruned_loss=0.09508, over 29543.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3388, pruned_loss=0.08877, over 5644816.88 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3507, pruned_loss=0.1112, over 5675050.93 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3378, pruned_loss=0.08603, over 5653853.25 frames. ], batch size: 79, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:12:36,409 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:12:38,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:13:15,766 INFO [train.py:968] (1/2) Epoch 28, batch 31950, giga_loss[loss=0.2298, simple_loss=0.3107, pruned_loss=0.07447, over 28390.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3388, pruned_loss=0.08953, over 5667144.30 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3498, pruned_loss=0.1106, over 5685270.86 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3383, pruned_loss=0.08688, over 5664500.18 frames. ], batch size: 369, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:13:20,089 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:13:38,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:13:59,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262880.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:14:03,807 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:14:22,523 INFO [train.py:968] (1/2) Epoch 28, batch 32000, giga_loss[loss=0.2252, simple_loss=0.3112, pruned_loss=0.06961, over 28939.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3338, pruned_loss=0.08658, over 5670455.65 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3492, pruned_loss=0.1103, over 5686864.23 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3337, pruned_loss=0.08439, over 5666772.94 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:14:42,319 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1262912.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:14:55,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:14:55,655 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:14:55,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5455, 3.1953, 2.9220, 2.1433], device='cuda:1'), covar=tensor([0.2679, 0.1491, 0.1624, 0.2476], device='cuda:1'), in_proj_covar=tensor([0.2022, 0.1988, 0.1893, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 15:15:09,763 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1262938.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:15:12,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1262941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:15:15,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1262944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:15:20,997 INFO [train.py:968] (1/2) Epoch 28, batch 32050, giga_loss[loss=0.2201, simple_loss=0.3064, pruned_loss=0.06692, over 27727.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3315, pruned_loss=0.08557, over 5664941.92 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.349, pruned_loss=0.1102, over 5683067.64 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3309, pruned_loss=0.08268, over 5664548.39 frames. ], batch size: 474, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:15:49,096 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 28, batch 32100, giga_loss[loss=0.3378, simple_loss=0.3862, pruned_loss=0.1447, over 26806.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3304, pruned_loss=0.08527, over 5662752.66 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3492, pruned_loss=0.1105, over 5683891.52 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3294, pruned_loss=0.08233, over 5661224.27 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:16:38,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:16:42,186 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:16:46,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3258, 1.4148, 1.2484, 1.5493], device='cuda:1'), covar=tensor([0.0780, 0.0363, 0.0363, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 15:17:15,681 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263038.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 15:17:18,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 32150, giga_loss[loss=0.2569, simple_loss=0.3452, pruned_loss=0.08432, over 29042.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3354, pruned_loss=0.08795, over 5670552.26 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3493, pruned_loss=0.1107, over 5687231.11 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3343, pruned_loss=0.08507, over 5665890.24 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:17:48,777 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:17:49,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263067.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:17:52,323 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:17:52,908 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:10,431 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:14,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263090.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:28,257 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:28,727 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 32200, libri_loss[loss=0.2337, simple_loss=0.3039, pruned_loss=0.0817, over 29360.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3347, pruned_loss=0.08817, over 5668617.61 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3488, pruned_loss=0.1104, over 5688461.23 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.334, pruned_loss=0.0857, over 5663303.82 frames. ], batch size: 67, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:18:29,326 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:18:36,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2716, 0.9010, 0.9542, 1.4936], device='cuda:1'), covar=tensor([0.0749, 0.0418, 0.0379, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 15:18:51,289 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-14 15:18:57,657 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:19:28,629 INFO [train.py:968] (1/2) Epoch 28, batch 32250, giga_loss[loss=0.2446, simple_loss=0.3214, pruned_loss=0.08396, over 28926.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.335, pruned_loss=0.08972, over 5668566.06 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3489, pruned_loss=0.1106, over 5689513.46 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.334, pruned_loss=0.08708, over 5662430.55 frames. ], batch size: 120, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:20:02,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4330, 3.3383, 1.5341, 1.6284], device='cuda:1'), covar=tensor([0.0994, 0.0349, 0.0964, 0.1348], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0570, 0.0410, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 15:20:05,028 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263181.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:20:08,423 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263184.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:20:28,627 INFO [optim.py:369] (1/2) 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 15:20:28,639 INFO [train.py:968] (1/2) Epoch 28, batch 32300, giga_loss[loss=0.256, simple_loss=0.3409, pruned_loss=0.08551, over 28690.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3357, pruned_loss=0.09052, over 5672063.87 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3488, pruned_loss=0.1106, over 5694914.24 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3347, pruned_loss=0.08792, over 5662014.94 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:20:50,025 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1263213.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 15:21:23,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8697, 2.2607, 2.0709, 1.8609], device='cuda:1'), covar=tensor([0.2137, 0.2405, 0.2122, 0.2363], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0749, 0.0722, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 15:21:23,580 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-14 15:21:38,393 INFO [train.py:968] (1/2) Epoch 28, batch 32350, giga_loss[loss=0.2863, simple_loss=0.3723, pruned_loss=0.1001, over 28544.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3374, pruned_loss=0.08999, over 5672127.81 frames. ], libri_tot_loss[loss=0.285, simple_loss=0.3488, pruned_loss=0.1106, over 5697054.76 frames. ], giga_tot_loss[loss=0.256, simple_loss=0.3366, pruned_loss=0.08775, over 5661917.19 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:22:36,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4643, 1.6476, 1.6547, 1.4673], device='cuda:1'), covar=tensor([0.2708, 0.2188, 0.1969, 0.2396], device='cuda:1'), in_proj_covar=tensor([0.2027, 0.1986, 0.1889, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 15:22:50,279 INFO [optim.py:369] (1/2) 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 15:22:50,292 INFO [train.py:968] (1/2) Epoch 28, batch 32400, giga_loss[loss=0.2122, simple_loss=0.2826, pruned_loss=0.07092, over 24246.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3386, pruned_loss=0.08996, over 5667127.37 frames. ], libri_tot_loss[loss=0.2853, simple_loss=0.349, pruned_loss=0.1108, over 5686721.49 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3375, pruned_loss=0.08759, over 5667485.89 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:23:58,023 INFO [train.py:968] (1/2) Epoch 28, batch 32450, giga_loss[loss=0.2423, simple_loss=0.3148, pruned_loss=0.08492, over 27575.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3354, pruned_loss=0.08838, over 5668768.83 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3482, pruned_loss=0.1104, over 5693337.28 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3349, pruned_loss=0.08618, over 5662633.36 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:24:23,010 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-14 15:24:58,401 INFO [optim.py:369] (1/2) 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 15:24:58,413 INFO [train.py:968] (1/2) Epoch 28, batch 32500, giga_loss[loss=0.2143, simple_loss=0.2928, pruned_loss=0.06791, over 28709.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3315, pruned_loss=0.08757, over 5673220.78 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.348, pruned_loss=0.1104, over 5695183.98 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3308, pruned_loss=0.08513, over 5666347.85 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:26:02,381 INFO [train.py:968] (1/2) Epoch 28, batch 32550, giga_loss[loss=0.2197, simple_loss=0.3014, pruned_loss=0.069, over 28927.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3272, pruned_loss=0.08601, over 5666825.99 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3476, pruned_loss=0.1102, over 5697729.19 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3264, pruned_loss=0.08356, over 5658372.96 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:26:15,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:26:38,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:26:46,738 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6368, 1.8241, 1.8100, 1.3923], device='cuda:1'), covar=tensor([0.1584, 0.2580, 0.1421, 0.1729], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0712, 0.0982, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 15:26:53,419 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3320, 3.3668, 1.5369, 1.6274], device='cuda:1'), covar=tensor([0.1049, 0.0379, 0.0988, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0570, 0.0411, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 15:26:55,554 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 15:26:59,451 INFO [train.py:968] (1/2) Epoch 28, batch 32600, libri_loss[loss=0.2444, simple_loss=0.3104, pruned_loss=0.08916, over 29561.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3277, pruned_loss=0.08657, over 5653591.02 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3469, pruned_loss=0.1098, over 5687538.66 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3271, pruned_loss=0.08421, over 5655395.51 frames. ], batch size: 75, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:27:00,260 INFO [optim.py:369] (1/2) 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 15:27:14,952 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:27:41,525 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:27:55,931 INFO [train.py:968] (1/2) Epoch 28, batch 32650, giga_loss[loss=0.2775, simple_loss=0.3511, pruned_loss=0.102, over 28658.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3294, pruned_loss=0.08781, over 5651244.31 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3467, pruned_loss=0.1097, over 5689957.83 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3287, pruned_loss=0.08548, over 5649516.19 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:27:57,328 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 15:28:06,426 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6290, 1.7199, 1.3015, 1.6945], device='cuda:1'), covar=tensor([0.0762, 0.0316, 0.0367, 0.0921], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0120, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 15:28:58,475 INFO [train.py:968] (1/2) Epoch 28, batch 32700, giga_loss[loss=0.2325, simple_loss=0.3144, pruned_loss=0.07531, over 28667.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.3273, pruned_loss=0.0858, over 5653729.33 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3467, pruned_loss=0.1098, over 5692069.62 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3265, pruned_loss=0.08362, over 5650105.41 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:28:59,474 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 32750, giga_loss[loss=0.2467, simple_loss=0.3333, pruned_loss=0.08004, over 28603.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3265, pruned_loss=0.08486, over 5665329.09 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3464, pruned_loss=0.1095, over 5696438.87 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3258, pruned_loss=0.083, over 5657971.23 frames. ], batch size: 307, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:31:08,741 INFO [train.py:968] (1/2) Epoch 28, batch 32800, giga_loss[loss=0.2583, simple_loss=0.3391, pruned_loss=0.08871, over 28082.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3264, pruned_loss=0.08544, over 5661978.03 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3467, pruned_loss=0.1098, over 5699809.54 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3253, pruned_loss=0.08329, over 5652674.30 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:31:09,794 INFO [optim.py:369] (1/2) 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 15:32:12,380 INFO [train.py:968] (1/2) Epoch 28, batch 32850, giga_loss[loss=0.2428, simple_loss=0.3252, pruned_loss=0.08022, over 28933.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.326, pruned_loss=0.08474, over 5663088.27 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3464, pruned_loss=0.1096, over 5704557.00 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3247, pruned_loss=0.08242, over 5650500.05 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:32:22,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 15:33:13,772 INFO [train.py:968] (1/2) Epoch 28, batch 32900, giga_loss[loss=0.2232, simple_loss=0.2978, pruned_loss=0.07428, over 29090.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3252, pruned_loss=0.0842, over 5670338.56 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.346, pruned_loss=0.1093, over 5709091.71 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3241, pruned_loss=0.08201, over 5655177.10 frames. ], batch size: 113, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:33:16,197 INFO [optim.py:369] (1/2) 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 15:33:29,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0523, 1.6317, 5.0010, 3.8361], device='cuda:1'), covar=tensor([0.1448, 0.2698, 0.0487, 0.0963], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0674, 0.1002, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 15:33:53,742 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:34:03,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2099, 1.6322, 1.6538, 1.3503], device='cuda:1'), covar=tensor([0.2292, 0.1908, 0.2161, 0.2091], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0748, 0.0719, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 15:34:13,756 INFO [train.py:968] (1/2) Epoch 28, batch 32950, giga_loss[loss=0.2029, simple_loss=0.2905, pruned_loss=0.05766, over 28929.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3267, pruned_loss=0.08584, over 5672154.95 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3459, pruned_loss=0.1093, over 5712125.26 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3254, pruned_loss=0.08357, over 5656636.14 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:34:18,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9355, 1.1385, 1.0382, 0.8990], device='cuda:1'), covar=tensor([0.2108, 0.2370, 0.1630, 0.2161], device='cuda:1'), in_proj_covar=tensor([0.2026, 0.1981, 0.1888, 0.2041], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 15:34:19,015 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:34:58,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:35:00,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-14 15:35:11,400 INFO [train.py:968] (1/2) Epoch 28, batch 33000, giga_loss[loss=0.2644, simple_loss=0.3507, pruned_loss=0.08903, over 28961.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.325, pruned_loss=0.08428, over 5661752.09 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3456, pruned_loss=0.109, over 5707299.04 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.3236, pruned_loss=0.082, over 5651739.92 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:35:11,400 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 15:35:19,554 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 15:35:20,953 INFO [optim.py:369] (1/2) 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 15:35:36,089 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1263913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:36:16,368 INFO [train.py:968] (1/2) Epoch 28, batch 33050, giga_loss[loss=0.2729, simple_loss=0.3577, pruned_loss=0.09408, over 29031.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3269, pruned_loss=0.0837, over 5660024.18 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.3448, pruned_loss=0.1086, over 5699777.04 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.326, pruned_loss=0.08158, over 5656793.02 frames. ], batch size: 200, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:36:26,447 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1263958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:36:48,019 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1263976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:36:51,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:06,456 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 33100, giga_loss[loss=0.2623, simple_loss=0.3391, pruned_loss=0.09281, over 28282.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3293, pruned_loss=0.085, over 5656370.70 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3444, pruned_loss=0.1084, over 5699084.37 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3283, pruned_loss=0.08239, over 5652575.38 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:37:08,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1263999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:09,788 INFO [optim.py:369] (1/2) 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 15:37:18,607 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:43,446 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:48,256 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:37:52,277 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:38:09,359 INFO [train.py:968] (1/2) Epoch 28, batch 33150, giga_loss[loss=0.2547, simple_loss=0.3318, pruned_loss=0.08884, over 29010.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.33, pruned_loss=0.08542, over 5648264.53 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3444, pruned_loss=0.1085, over 5700654.14 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3288, pruned_loss=0.08279, over 5642665.09 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:38:18,269 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:38:22,625 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:38:30,023 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:39:00,470 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 33200, giga_loss[loss=0.2297, simple_loss=0.3152, pruned_loss=0.07208, over 28984.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3298, pruned_loss=0.08516, over 5660109.50 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3441, pruned_loss=0.1082, over 5703833.47 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3288, pruned_loss=0.08275, over 5651681.29 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:39:13,389 INFO [optim.py:369] (1/2) 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 15:39:39,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6570, 1.8618, 1.8677, 1.4129], device='cuda:1'), covar=tensor([0.1787, 0.2785, 0.1543, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0712, 0.0981, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 15:40:03,630 INFO [train.py:968] (1/2) Epoch 28, batch 33250, giga_loss[loss=0.2459, simple_loss=0.3334, pruned_loss=0.0792, over 28660.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3286, pruned_loss=0.08458, over 5658600.57 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3441, pruned_loss=0.1083, over 5698320.85 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3273, pruned_loss=0.08175, over 5655252.68 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:41:01,180 INFO [train.py:968] (1/2) Epoch 28, batch 33300, giga_loss[loss=0.2879, simple_loss=0.3605, pruned_loss=0.1076, over 28943.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3273, pruned_loss=0.08407, over 5658763.68 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3439, pruned_loss=0.108, over 5698998.29 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.326, pruned_loss=0.08134, over 5654282.03 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:41:04,176 INFO [optim.py:369] (1/2) 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 15:41:09,138 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7011, 2.1507, 1.3602, 1.6034], device='cuda:1'), covar=tensor([0.1045, 0.0482, 0.1020, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0446, 0.0523, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 15:42:01,807 INFO [train.py:968] (1/2) Epoch 28, batch 33350, giga_loss[loss=0.2525, simple_loss=0.3352, pruned_loss=0.08493, over 28769.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.325, pruned_loss=0.08332, over 5658759.72 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3438, pruned_loss=0.108, over 5692891.97 frames. ], giga_tot_loss[loss=0.2424, simple_loss=0.3236, pruned_loss=0.08059, over 5658748.42 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:42:09,480 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:42:28,382 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.4065, 5.2215, 4.9711, 2.3145], device='cuda:1'), covar=tensor([0.0482, 0.0628, 0.0770, 0.1920], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.1194, 0.1006, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 15:43:00,494 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 15:43:00,658 INFO [train.py:968] (1/2) Epoch 28, batch 33400, giga_loss[loss=0.2731, simple_loss=0.3529, pruned_loss=0.09671, over 27647.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3271, pruned_loss=0.0839, over 5666704.58 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3436, pruned_loss=0.1078, over 5696979.71 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3258, pruned_loss=0.08149, over 5662563.22 frames. ], batch size: 474, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:43:03,419 INFO [optim.py:369] (1/2) 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 15:43:10,455 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:43:45,073 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1264333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:43:45,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0256, 2.9376, 1.8205, 1.1070], device='cuda:1'), covar=tensor([0.8785, 0.4101, 0.5142, 0.7928], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1736, 0.1654, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 15:44:01,282 INFO [train.py:968] (1/2) Epoch 28, batch 33450, giga_loss[loss=0.2042, simple_loss=0.2847, pruned_loss=0.06187, over 28455.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3295, pruned_loss=0.08539, over 5665628.35 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3436, pruned_loss=0.1078, over 5697431.43 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.328, pruned_loss=0.08278, over 5660939.26 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:44:47,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2544, 2.5840, 1.3017, 1.4152], device='cuda:1'), covar=tensor([0.1038, 0.0434, 0.0989, 0.1451], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0567, 0.0410, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 15:45:08,967 INFO [train.py:968] (1/2) Epoch 28, batch 33500, giga_loss[loss=0.2453, simple_loss=0.3321, pruned_loss=0.07926, over 28913.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3302, pruned_loss=0.08595, over 5666631.46 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3435, pruned_loss=0.1077, over 5699766.74 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3289, pruned_loss=0.08368, over 5660434.09 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:45:11,669 INFO [optim.py:369] (1/2) 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 15:46:10,411 INFO [train.py:968] (1/2) Epoch 28, batch 33550, giga_loss[loss=0.231, simple_loss=0.3021, pruned_loss=0.07994, over 24412.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.334, pruned_loss=0.08772, over 5661521.41 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3436, pruned_loss=0.1078, over 5692905.41 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3326, pruned_loss=0.0854, over 5662016.05 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:46:37,788 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264476.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:46:40,761 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 33600, giga_loss[loss=0.2422, simple_loss=0.3289, pruned_loss=0.07773, over 29158.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3355, pruned_loss=0.08766, over 5652063.08 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3437, pruned_loss=0.1077, over 5685198.54 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3342, pruned_loss=0.08542, over 5658932.05 frames. ], batch size: 113, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:47:08,217 INFO [optim.py:369] (1/2) 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 15:47:15,573 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:47:38,552 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8356, 2.1618, 2.0037, 1.9117], device='cuda:1'), covar=tensor([0.2102, 0.2458, 0.2081, 0.2221], device='cuda:1'), in_proj_covar=tensor([0.0491, 0.0745, 0.0715, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 15:47:49,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2988, 1.2992, 3.1548, 2.9875], device='cuda:1'), covar=tensor([0.1428, 0.2676, 0.0504, 0.2041], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0674, 0.1003, 0.0972], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 15:48:04,583 INFO [train.py:968] (1/2) Epoch 28, batch 33650, giga_loss[loss=0.2426, simple_loss=0.3251, pruned_loss=0.08006, over 29060.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3348, pruned_loss=0.08782, over 5651158.98 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3431, pruned_loss=0.1076, over 5684637.84 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3339, pruned_loss=0.08531, over 5655889.37 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:49:06,387 INFO [train.py:968] (1/2) Epoch 28, batch 33700, giga_loss[loss=0.2271, simple_loss=0.31, pruned_loss=0.07207, over 29192.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3326, pruned_loss=0.0873, over 5654045.44 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3424, pruned_loss=0.1074, over 5681934.80 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3321, pruned_loss=0.08455, over 5658809.63 frames. ], batch size: 113, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:49:12,491 INFO [optim.py:369] (1/2) 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 15:49:37,823 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5808, 4.4436, 4.2142, 1.9973], device='cuda:1'), covar=tensor([0.0532, 0.0716, 0.0787, 0.2299], device='cuda:1'), in_proj_covar=tensor([0.1301, 0.1193, 0.1007, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 15:49:49,892 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 33750, giga_loss[loss=0.2345, simple_loss=0.3215, pruned_loss=0.0737, over 28097.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3312, pruned_loss=0.08696, over 5653597.85 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3422, pruned_loss=0.1073, over 5679941.22 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3306, pruned_loss=0.08413, over 5657862.79 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:50:28,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5060, 1.5664, 1.7408, 1.4148], device='cuda:1'), covar=tensor([0.1722, 0.2383, 0.1433, 0.1957], device='cuda:1'), in_proj_covar=tensor([0.0930, 0.0710, 0.0980, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 15:50:49,843 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 33800, giga_loss[loss=0.2869, simple_loss=0.3586, pruned_loss=0.1076, over 28962.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3297, pruned_loss=0.08574, over 5653285.69 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3419, pruned_loss=0.1072, over 5682153.41 frames. ], giga_tot_loss[loss=0.2482, simple_loss=0.3295, pruned_loss=0.08342, over 5654272.53 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:51:19,505 INFO [optim.py:369] (1/2) 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 15:52:20,609 INFO [train.py:968] (1/2) Epoch 28, batch 33850, giga_loss[loss=0.2821, simple_loss=0.344, pruned_loss=0.1101, over 28900.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3287, pruned_loss=0.08614, over 5651955.77 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3416, pruned_loss=0.1071, over 5676680.38 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3286, pruned_loss=0.08403, over 5656997.26 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:52:40,185 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:52:47,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1264769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:52:53,132 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:52:58,052 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:53:24,752 INFO [train.py:968] (1/2) Epoch 28, batch 33900, giga_loss[loss=0.2552, simple_loss=0.3309, pruned_loss=0.08973, over 28788.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3276, pruned_loss=0.08614, over 5632401.34 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3413, pruned_loss=0.1069, over 5669650.79 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3275, pruned_loss=0.08415, over 5641580.30 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:53:29,307 INFO [optim.py:369] (1/2) 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 15:53:33,377 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:53:52,553 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1264823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:53:55,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1264826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:54:25,742 INFO [train.py:968] (1/2) Epoch 28, batch 33950, libri_loss[loss=0.2854, simple_loss=0.3624, pruned_loss=0.1042, over 29514.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3275, pruned_loss=0.08475, over 5634036.88 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3415, pruned_loss=0.107, over 5658932.81 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.327, pruned_loss=0.08279, over 5649908.16 frames. ], batch size: 89, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:54:33,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1264855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 15:55:21,285 INFO [train.py:968] (1/2) Epoch 28, batch 34000, giga_loss[loss=0.2197, simple_loss=0.3149, pruned_loss=0.06221, over 28733.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3282, pruned_loss=0.08339, over 5655965.78 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3415, pruned_loss=0.1069, over 5665143.20 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3274, pruned_loss=0.08137, over 5662968.50 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:55:25,915 INFO [optim.py:369] (1/2) 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 15:55:59,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6554, 1.8545, 1.8344, 1.7193], device='cuda:1'), covar=tensor([0.3241, 0.2506, 0.2241, 0.2556], device='cuda:1'), in_proj_covar=tensor([0.2034, 0.1986, 0.1889, 0.2044], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 15:56:17,634 INFO [train.py:968] (1/2) Epoch 28, batch 34050, giga_loss[loss=0.2616, simple_loss=0.3357, pruned_loss=0.0937, over 26788.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3313, pruned_loss=0.08425, over 5657550.77 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3417, pruned_loss=0.107, over 5666924.37 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3302, pruned_loss=0.08173, over 5661563.92 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:56:39,307 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.63 vs. limit=5.0
+2023-03-14 15:57:12,979 INFO [train.py:968] (1/2) Epoch 28, batch 34100, giga_loss[loss=0.2404, simple_loss=0.3297, pruned_loss=0.07555, over 28675.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3316, pruned_loss=0.0839, over 5662471.87 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3414, pruned_loss=0.1068, over 5673873.50 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3306, pruned_loss=0.08146, over 5659327.94 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 15:57:17,963 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 34150, giga_loss[loss=0.2476, simple_loss=0.3303, pruned_loss=0.08248, over 28915.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.331, pruned_loss=0.08364, over 5668053.02 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3409, pruned_loss=0.1066, over 5679644.01 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3304, pruned_loss=0.08137, over 5660265.04 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:59:20,522 INFO [train.py:968] (1/2) Epoch 28, batch 34200, giga_loss[loss=0.2396, simple_loss=0.3231, pruned_loss=0.07805, over 28023.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3321, pruned_loss=0.08454, over 5677356.97 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3409, pruned_loss=0.1066, over 5684948.64 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3313, pruned_loss=0.08174, over 5665776.44 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 15:59:27,388 INFO [optim.py:369] (1/2) 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 16:00:14,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1265137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:00:17,882 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5792, 1.9715, 1.4810, 1.8394], device='cuda:1'), covar=tensor([0.2751, 0.2688, 0.3226, 0.2314], device='cuda:1'), in_proj_covar=tensor([0.1605, 0.1149, 0.1420, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 16:00:22,558 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 34250, giga_loss[loss=0.2034, simple_loss=0.2816, pruned_loss=0.06261, over 24954.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3321, pruned_loss=0.08448, over 5659174.76 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3409, pruned_loss=0.1066, over 5675840.42 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3312, pruned_loss=0.08176, over 5658744.06 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:01:27,312 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 34300, giga_loss[loss=0.2399, simple_loss=0.3334, pruned_loss=0.07315, over 28954.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3324, pruned_loss=0.08491, over 5668849.28 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3404, pruned_loss=0.1064, over 5684508.67 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3317, pruned_loss=0.08161, over 5659661.19 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:01:35,996 INFO [optim.py:369] (1/2) 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 16:02:17,280 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 16:02:36,862 INFO [train.py:968] (1/2) Epoch 28, batch 34350, giga_loss[loss=0.2523, simple_loss=0.3455, pruned_loss=0.07957, over 29094.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3355, pruned_loss=0.08588, over 5667797.32 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3409, pruned_loss=0.1069, over 5686742.94 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3343, pruned_loss=0.0826, over 5658325.14 frames. ], batch size: 285, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:02:45,482 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5923, 2.0209, 2.0771, 1.5816], device='cuda:1'), covar=tensor([0.3535, 0.2180, 0.2359, 0.2897], device='cuda:1'), in_proj_covar=tensor([0.2026, 0.1980, 0.1882, 0.2032], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 16:03:14,478 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1265280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:18,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1265283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:24,302 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1265287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:29,716 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1265290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:03:38,433 INFO [train.py:968] (1/2) Epoch 28, batch 34400, libri_loss[loss=0.2907, simple_loss=0.3508, pruned_loss=0.1154, over 29573.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3377, pruned_loss=0.08693, over 5670018.20 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3414, pruned_loss=0.1072, over 5678299.55 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3362, pruned_loss=0.08325, over 5668757.24 frames. ], batch size: 75, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:03:44,757 INFO [optim.py:369] (1/2) 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 16:03:56,700 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1265312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:04:08,107 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1265319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:04:47,124 INFO [train.py:968] (1/2) Epoch 28, batch 34450, giga_loss[loss=0.2679, simple_loss=0.345, pruned_loss=0.09544, over 28659.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3379, pruned_loss=0.0876, over 5678785.03 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3419, pruned_loss=0.1076, over 5678387.20 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3362, pruned_loss=0.08411, over 5677734.04 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:04:54,100 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 16:05:57,780 INFO [train.py:968] (1/2) Epoch 28, batch 34500, giga_loss[loss=0.2487, simple_loss=0.3323, pruned_loss=0.08255, over 28882.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.336, pruned_loss=0.08683, over 5680048.53 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3422, pruned_loss=0.1076, over 5680072.09 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3344, pruned_loss=0.08375, over 5677831.89 frames. ], batch size: 227, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:06:07,262 INFO [optim.py:369] (1/2) 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 16:07:05,462 INFO [train.py:968] (1/2) Epoch 28, batch 34550, giga_loss[loss=0.2049, simple_loss=0.3028, pruned_loss=0.05346, over 28885.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3328, pruned_loss=0.08398, over 5681167.38 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3415, pruned_loss=0.1074, over 5675000.20 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3319, pruned_loss=0.08124, over 5684876.87 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:08:09,243 INFO [train.py:968] (1/2) Epoch 28, batch 34600, giga_loss[loss=0.2384, simple_loss=0.3305, pruned_loss=0.07315, over 28316.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3321, pruned_loss=0.08381, over 5687947.78 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3413, pruned_loss=0.1073, over 5680506.16 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.3313, pruned_loss=0.08116, over 5686171.08 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:08:16,392 INFO [optim.py:369] (1/2) 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 16:09:07,573 INFO [train.py:968] (1/2) Epoch 28, batch 34650, giga_loss[loss=0.2452, simple_loss=0.3344, pruned_loss=0.07803, over 28104.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3337, pruned_loss=0.08494, over 5684800.23 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3419, pruned_loss=0.1077, over 5684439.33 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3324, pruned_loss=0.08187, over 5679897.89 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:09:23,979 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.69 vs. limit=5.0
+2023-03-14 16:09:33,154 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1265571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:10:00,337 INFO [train.py:968] (1/2) Epoch 28, batch 34700, libri_loss[loss=0.2926, simple_loss=0.3531, pruned_loss=0.1161, over 27995.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3363, pruned_loss=0.08756, over 5679112.92 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3412, pruned_loss=0.1073, over 5690945.79 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3356, pruned_loss=0.08438, over 5668988.26 frames. ], batch size: 116, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:10:05,640 INFO [optim.py:369] (1/2) 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 16:10:43,086 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1265638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:10:55,466 INFO [train.py:968] (1/2) Epoch 28, batch 34750, giga_loss[loss=0.2246, simple_loss=0.3134, pruned_loss=0.06795, over 28891.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3342, pruned_loss=0.08737, over 5656889.05 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3415, pruned_loss=0.1077, over 5666683.02 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3331, pruned_loss=0.08376, over 5670008.10 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:11:45,770 INFO [train.py:968] (1/2) Epoch 28, batch 34800, giga_loss[loss=0.2523, simple_loss=0.3378, pruned_loss=0.08343, over 28880.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3324, pruned_loss=0.08685, over 5657694.00 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3413, pruned_loss=0.1076, over 5664958.33 frames. ], giga_tot_loss[loss=0.249, simple_loss=0.3315, pruned_loss=0.08332, over 5669656.05 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:11:53,460 INFO [optim.py:369] (1/2) 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 16:12:05,542 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1265714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:12:08,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1265717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:12:36,870 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 34850, libri_loss[loss=0.2282, simple_loss=0.2968, pruned_loss=0.07983, over 29662.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3328, pruned_loss=0.08752, over 5658410.28 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3406, pruned_loss=0.1071, over 5666756.63 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3325, pruned_loss=0.08461, over 5665968.88 frames. ], batch size: 73, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:12:56,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4175, 1.7621, 1.4309, 1.5767], device='cuda:1'), covar=tensor([0.0722, 0.0301, 0.0322, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 16:13:26,700 INFO [train.py:968] (1/2) Epoch 28, batch 34900, giga_loss[loss=0.2843, simple_loss=0.3641, pruned_loss=0.1022, over 28856.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3401, pruned_loss=0.09196, over 5645773.43 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3404, pruned_loss=0.1069, over 5656506.99 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3398, pruned_loss=0.08921, over 5661529.63 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:13:31,916 INFO [optim.py:369] (1/2) 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 16:13:58,571 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1265833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:14:14,508 INFO [train.py:968] (1/2) Epoch 28, batch 34950, giga_loss[loss=0.3488, simple_loss=0.4194, pruned_loss=0.1392, over 29057.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3494, pruned_loss=0.09682, over 5661436.11 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.341, pruned_loss=0.1072, over 5658570.32 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3486, pruned_loss=0.09401, over 5672400.40 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:14:56,990 INFO [train.py:968] (1/2) Epoch 28, batch 35000, giga_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08543, over 28917.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3516, pruned_loss=0.09854, over 5650513.04 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3412, pruned_loss=0.1073, over 5644801.14 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.351, pruned_loss=0.09598, over 5672382.65 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:15:04,288 INFO [optim.py:369] (1/2) 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 16:15:29,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7088, 2.0108, 1.6125, 1.8601], device='cuda:1'), covar=tensor([0.2809, 0.2816, 0.3373, 0.2439], device='cuda:1'), in_proj_covar=tensor([0.1604, 0.1151, 0.1419, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 16:15:39,514 INFO [train.py:968] (1/2) Epoch 28, batch 35050, giga_loss[loss=0.2382, simple_loss=0.3168, pruned_loss=0.0798, over 28117.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3467, pruned_loss=0.09639, over 5658531.32 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3413, pruned_loss=0.1073, over 5647642.73 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3463, pruned_loss=0.09418, over 5673320.42 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:16:02,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0975, 2.1839, 2.1293, 1.9356], device='cuda:1'), covar=tensor([0.2235, 0.2910, 0.2559, 0.2794], device='cuda:1'), in_proj_covar=tensor([0.0494, 0.0751, 0.0723, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 16:16:18,838 INFO [train.py:968] (1/2) Epoch 28, batch 35100, libri_loss[loss=0.2686, simple_loss=0.3352, pruned_loss=0.1009, over 29540.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3396, pruned_loss=0.0936, over 5658316.83 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3414, pruned_loss=0.1073, over 5646356.66 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3393, pruned_loss=0.09146, over 5672224.78 frames. ], batch size: 79, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:16:24,300 INFO [optim.py:369] (1/2) 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 16:16:29,261 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-14 16:16:29,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1266013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:16:59,400 INFO [train.py:968] (1/2) Epoch 28, batch 35150, giga_loss[loss=0.2384, simple_loss=0.3096, pruned_loss=0.08361, over 27967.00 frames. ], tot_loss[loss=0.2567, simple_loss=0.3325, pruned_loss=0.09041, over 5676074.09 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3418, pruned_loss=0.1074, over 5654039.94 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.3319, pruned_loss=0.08825, over 5680983.71 frames. ], batch size: 412, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:17:14,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 16:17:30,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7658, 1.9538, 1.6950, 1.7584], device='cuda:1'), covar=tensor([0.2335, 0.2236, 0.2364, 0.2195], device='cuda:1'), in_proj_covar=tensor([0.1605, 0.1150, 0.1420, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 16:17:38,102 INFO [train.py:968] (1/2) Epoch 28, batch 35200, giga_loss[loss=0.1935, simple_loss=0.2719, pruned_loss=0.05751, over 28709.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3253, pruned_loss=0.08717, over 5681617.64 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3416, pruned_loss=0.1071, over 5663113.05 frames. ], giga_tot_loss[loss=0.2471, simple_loss=0.3244, pruned_loss=0.08488, over 5677827.99 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:17:40,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5318, 1.4751, 1.3217, 1.6273], device='cuda:1'), covar=tensor([0.0779, 0.0352, 0.0355, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 16:17:44,389 INFO [optim.py:369] (1/2) 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 16:18:20,723 INFO [train.py:968] (1/2) Epoch 28, batch 35250, giga_loss[loss=0.2155, simple_loss=0.2921, pruned_loss=0.06941, over 28957.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3189, pruned_loss=0.08427, over 5674978.75 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.342, pruned_loss=0.1072, over 5656207.64 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3175, pruned_loss=0.08199, over 5678055.63 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:18:26,723 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1266156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:18:28,682 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1266159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:18:56,016 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1266188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:19:03,648 INFO [train.py:968] (1/2) Epoch 28, batch 35300, giga_loss[loss=0.2392, simple_loss=0.3102, pruned_loss=0.08406, over 28640.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.3149, pruned_loss=0.08244, over 5689627.62 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3422, pruned_loss=0.1073, over 5660570.99 frames. ], giga_tot_loss[loss=0.2366, simple_loss=0.3131, pruned_loss=0.08008, over 5688957.31 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:19:09,726 INFO [optim.py:369] (1/2) 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 16:19:10,642 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 35350, giga_loss[loss=0.2551, simple_loss=0.3139, pruned_loss=0.09813, over 26530.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.3137, pruned_loss=0.08242, over 5688578.48 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3425, pruned_loss=0.1072, over 5665864.30 frames. ], giga_tot_loss[loss=0.235, simple_loss=0.3108, pruned_loss=0.07958, over 5684662.94 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:20:02,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-14 16:20:26,423 INFO [train.py:968] (1/2) Epoch 28, batch 35400, giga_loss[loss=0.2028, simple_loss=0.2724, pruned_loss=0.06661, over 28629.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.3098, pruned_loss=0.0804, over 5687202.78 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3425, pruned_loss=0.107, over 5670463.71 frames. ], giga_tot_loss[loss=0.2313, simple_loss=0.3069, pruned_loss=0.07781, over 5680428.52 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:20:31,661 INFO [optim.py:369] (1/2) 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 16:20:45,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:10,095 INFO [train.py:968] (1/2) Epoch 28, batch 35450, giga_loss[loss=0.209, simple_loss=0.2909, pruned_loss=0.06353, over 28631.00 frames. ], tot_loss[loss=0.2334, simple_loss=0.3074, pruned_loss=0.07964, over 5679109.28 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3431, pruned_loss=0.1073, over 5672303.58 frames. ], giga_tot_loss[loss=0.2286, simple_loss=0.3038, pruned_loss=0.07668, over 5672130.76 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:21:11,701 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1266351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:13,978 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1266354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:38,798 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1266383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:21:50,805 INFO [train.py:968] (1/2) Epoch 28, batch 35500, giga_loss[loss=0.248, simple_loss=0.319, pruned_loss=0.08846, over 28174.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.3047, pruned_loss=0.07839, over 5685271.24 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3431, pruned_loss=0.1072, over 5672739.18 frames. ], giga_tot_loss[loss=0.2265, simple_loss=0.3013, pruned_loss=0.07579, over 5679489.69 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:21:58,194 INFO [optim.py:369] (1/2) 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 16:22:31,701 INFO [train.py:968] (1/2) Epoch 28, batch 35550, giga_loss[loss=0.2209, simple_loss=0.2858, pruned_loss=0.07794, over 28970.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.3027, pruned_loss=0.07743, over 5683880.89 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.343, pruned_loss=0.1071, over 5669138.28 frames. ], giga_tot_loss[loss=0.2242, simple_loss=0.2991, pruned_loss=0.0747, over 5682903.55 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:22:35,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3832, 1.7445, 1.7436, 1.5683], device='cuda:1'), covar=tensor([0.2392, 0.1953, 0.2604, 0.2081], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0754, 0.0724, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 16:22:48,013 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-14 16:23:08,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6024, 1.7804, 1.5290, 1.5976], device='cuda:1'), covar=tensor([0.2949, 0.2964, 0.3219, 0.2944], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1152, 0.1422, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 16:23:13,404 INFO [train.py:968] (1/2) Epoch 28, batch 35600, giga_loss[loss=0.1942, simple_loss=0.2746, pruned_loss=0.05685, over 28953.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.3006, pruned_loss=0.07663, over 5673849.70 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3431, pruned_loss=0.1071, over 5656783.10 frames. ], giga_tot_loss[loss=0.2216, simple_loss=0.2962, pruned_loss=0.0735, over 5685201.57 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:23:20,519 INFO [optim.py:369] (1/2) 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 16:23:35,778 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5359, 3.7719, 1.8183, 1.6588], device='cuda:1'), covar=tensor([0.1029, 0.0315, 0.0883, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0566, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 16:23:55,361 INFO [train.py:968] (1/2) Epoch 28, batch 35650, giga_loss[loss=0.2126, simple_loss=0.2856, pruned_loss=0.06976, over 28886.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2989, pruned_loss=0.07598, over 5673308.81 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3435, pruned_loss=0.107, over 5662312.33 frames. ], giga_tot_loss[loss=0.2195, simple_loss=0.2938, pruned_loss=0.07265, over 5677584.25 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:24:38,340 INFO [train.py:968] (1/2) Epoch 28, batch 35700, giga_loss[loss=0.2027, simple_loss=0.2764, pruned_loss=0.06447, over 29110.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.2989, pruned_loss=0.07636, over 5674614.46 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3437, pruned_loss=0.1068, over 5664246.50 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2935, pruned_loss=0.07297, over 5676337.34 frames. ], batch size: 136, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:24:41,394 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:24:46,151 INFO [optim.py:369] (1/2) 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 16:25:07,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2705, 1.4261, 1.5170, 1.1224], device='cuda:1'), covar=tensor([0.1772, 0.2685, 0.1490, 0.1802], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0715, 0.0990, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 16:25:27,260 INFO [train.py:968] (1/2) Epoch 28, batch 35750, giga_loss[loss=0.2797, simple_loss=0.3566, pruned_loss=0.1014, over 28560.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.3082, pruned_loss=0.08077, over 5677685.67 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3439, pruned_loss=0.107, over 5663014.45 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3034, pruned_loss=0.07771, over 5680015.11 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:26:10,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1266697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:26:11,946 INFO [train.py:968] (1/2) Epoch 28, batch 35800, giga_loss[loss=0.288, simple_loss=0.3652, pruned_loss=0.1054, over 28996.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3226, pruned_loss=0.08866, over 5680671.35 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3441, pruned_loss=0.1071, over 5667525.31 frames. ], giga_tot_loss[loss=0.2448, simple_loss=0.3181, pruned_loss=0.0858, over 5678743.46 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:26:20,554 INFO [optim.py:369] (1/2) 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 16:26:55,644 INFO [train.py:968] (1/2) Epoch 28, batch 35850, libri_loss[loss=0.3214, simple_loss=0.383, pruned_loss=0.1299, over 20092.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3332, pruned_loss=0.09379, over 5674401.56 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3445, pruned_loss=0.1073, over 5661672.33 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.329, pruned_loss=0.09113, over 5679185.68 frames. ], batch size: 187, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:27:39,465 INFO [train.py:968] (1/2) Epoch 28, batch 35900, giga_loss[loss=0.2442, simple_loss=0.3371, pruned_loss=0.07568, over 28887.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3396, pruned_loss=0.09604, over 5675133.38 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3446, pruned_loss=0.1073, over 5664293.19 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.336, pruned_loss=0.09368, over 5676638.14 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:27:41,886 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7710, 4.6001, 4.3149, 2.1449], device='cuda:1'), covar=tensor([0.0559, 0.0737, 0.0756, 0.2011], device='cuda:1'), in_proj_covar=tensor([0.1283, 0.1186, 0.0996, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 16:27:46,479 INFO [optim.py:369] (1/2) 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 16:28:12,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1266840.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:28:15,907 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 35950, giga_loss[loss=0.2478, simple_loss=0.3333, pruned_loss=0.08114, over 28821.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3411, pruned_loss=0.09557, over 5666451.62 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3453, pruned_loss=0.1077, over 5659275.87 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3375, pruned_loss=0.09288, over 5672463.43 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:28:44,550 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 36000, giga_loss[loss=0.2963, simple_loss=0.362, pruned_loss=0.1153, over 28781.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3424, pruned_loss=0.09531, over 5667511.26 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3458, pruned_loss=0.108, over 5660590.40 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.339, pruned_loss=0.0927, over 5671239.46 frames. ], batch size: 99, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:29:08,719 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 16:29:16,932 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 16:29:23,132 INFO [optim.py:369] (1/2) 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 16:29:55,395 INFO [train.py:968] (1/2) Epoch 28, batch 36050, giga_loss[loss=0.2572, simple_loss=0.3304, pruned_loss=0.09207, over 28905.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3447, pruned_loss=0.09718, over 5677670.35 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.346, pruned_loss=0.1079, over 5667730.41 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3418, pruned_loss=0.09481, over 5674468.10 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:30:19,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1266977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:30:39,363 INFO [train.py:968] (1/2) Epoch 28, batch 36100, giga_loss[loss=0.387, simple_loss=0.4126, pruned_loss=0.1807, over 23287.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3467, pruned_loss=0.09865, over 5679512.60 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3459, pruned_loss=0.1078, over 5671874.04 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3444, pruned_loss=0.09667, over 5673741.87 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:30:46,492 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2042, 2.2740, 2.0131, 1.9698], device='cuda:1'), covar=tensor([0.2116, 0.2467, 0.2470, 0.2256], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0755, 0.0726, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 16:30:48,280 INFO [optim.py:369] (1/2) 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 16:31:19,531 INFO [train.py:968] (1/2) Epoch 28, batch 36150, giga_loss[loss=0.2674, simple_loss=0.3443, pruned_loss=0.09524, over 28955.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3495, pruned_loss=0.1003, over 5677952.97 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3465, pruned_loss=0.1082, over 5666037.93 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3471, pruned_loss=0.09802, over 5678937.39 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:31:58,417 INFO [train.py:968] (1/2) Epoch 28, batch 36200, libri_loss[loss=0.3078, simple_loss=0.3686, pruned_loss=0.1235, over 19183.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3521, pruned_loss=0.1002, over 5670024.75 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3469, pruned_loss=0.1084, over 5642069.09 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3499, pruned_loss=0.0981, over 5693125.37 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:31:58,884 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-14 16:32:05,645 INFO [optim.py:369] (1/2) 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 16:32:16,317 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1267120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:32:19,010 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1267123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:32:40,423 INFO [train.py:968] (1/2) Epoch 28, batch 36250, libri_loss[loss=0.3144, simple_loss=0.3782, pruned_loss=0.1253, over 29533.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3533, pruned_loss=0.1002, over 5673870.64 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3469, pruned_loss=0.1082, over 5648414.84 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3518, pruned_loss=0.09851, over 5687174.35 frames. ], batch size: 89, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:32:40,730 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9531, 1.3088, 1.1202, 0.1958], device='cuda:1'), covar=tensor([0.5397, 0.3932, 0.5394, 0.7871], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1744, 0.1663, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 16:32:42,542 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1267152.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:33:19,256 INFO [train.py:968] (1/2) Epoch 28, batch 36300, giga_loss[loss=0.2691, simple_loss=0.3529, pruned_loss=0.09268, over 28368.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.355, pruned_loss=0.1001, over 5692392.72 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3477, pruned_loss=0.1086, over 5655915.26 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3532, pruned_loss=0.09812, over 5697138.17 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:33:28,405 INFO [optim.py:369] (1/2) 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 16:33:59,668 INFO [train.py:968] (1/2) Epoch 28, batch 36350, giga_loss[loss=0.3349, simple_loss=0.3859, pruned_loss=0.142, over 26695.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.355, pruned_loss=0.09928, over 5677233.65 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3483, pruned_loss=0.1089, over 5643686.02 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3532, pruned_loss=0.09735, over 5692067.16 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:34:41,044 INFO [train.py:968] (1/2) Epoch 28, batch 36400, giga_loss[loss=0.2375, simple_loss=0.3276, pruned_loss=0.07367, over 29062.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.353, pruned_loss=0.09708, over 5685529.26 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3487, pruned_loss=0.1091, over 5645369.11 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3513, pruned_loss=0.09519, over 5696890.32 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:34:49,414 INFO [optim.py:369] (1/2) 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 16:34:52,422 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:35:07,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4680, 1.6151, 1.7123, 1.2801], device='cuda:1'), covar=tensor([0.1919, 0.2650, 0.1570, 0.1834], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0713, 0.0988, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 16:35:23,192 INFO [train.py:968] (1/2) Epoch 28, batch 36450, giga_loss[loss=0.2427, simple_loss=0.3293, pruned_loss=0.07801, over 28742.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3512, pruned_loss=0.09611, over 5678695.38 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3487, pruned_loss=0.109, over 5649138.70 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3499, pruned_loss=0.09452, over 5684874.33 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:36:06,318 INFO [train.py:968] (1/2) Epoch 28, batch 36500, libri_loss[loss=0.2792, simple_loss=0.3547, pruned_loss=0.1018, over 29389.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3536, pruned_loss=0.09939, over 5679938.21 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.349, pruned_loss=0.109, over 5652008.75 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3524, pruned_loss=0.09789, over 5682804.45 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:36:15,214 INFO [optim.py:369] (1/2) 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 16:36:30,983 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-14 16:36:46,784 INFO [train.py:968] (1/2) Epoch 28, batch 36550, giga_loss[loss=0.3667, simple_loss=0.4075, pruned_loss=0.163, over 27624.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3574, pruned_loss=0.1042, over 5687720.52 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3492, pruned_loss=0.109, over 5658325.11 frames. ], giga_tot_loss[loss=0.2811, simple_loss=0.3564, pruned_loss=0.1029, over 5684898.38 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:37:04,826 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 16:37:26,585 INFO [train.py:968] (1/2) Epoch 28, batch 36600, giga_loss[loss=0.3306, simple_loss=0.3875, pruned_loss=0.1369, over 27584.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3569, pruned_loss=0.1054, over 5689470.03 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3494, pruned_loss=0.1088, over 5661135.66 frames. ], giga_tot_loss[loss=0.2824, simple_loss=0.3563, pruned_loss=0.1042, over 5686395.24 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:37:37,239 INFO [optim.py:369] (1/2) 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 16:37:59,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-14 16:38:08,619 INFO [train.py:968] (1/2) Epoch 28, batch 36650, giga_loss[loss=0.2747, simple_loss=0.3482, pruned_loss=0.1006, over 28740.00 frames. ], tot_loss[loss=0.2835, simple_loss=0.3558, pruned_loss=0.1056, over 5694325.84 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3497, pruned_loss=0.1088, over 5667524.49 frames. ], giga_tot_loss[loss=0.2821, simple_loss=0.3552, pruned_loss=0.1045, over 5686771.39 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:38:22,242 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.11 vs. limit=5.0
+2023-03-14 16:38:50,654 INFO [train.py:968] (1/2) Epoch 28, batch 36700, giga_loss[loss=0.251, simple_loss=0.329, pruned_loss=0.0865, over 28418.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3526, pruned_loss=0.1039, over 5701821.81 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3498, pruned_loss=0.1088, over 5669468.57 frames. ], giga_tot_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.103, over 5694568.51 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:38:59,076 INFO [optim.py:369] (1/2) 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 16:39:32,293 INFO [train.py:968] (1/2) Epoch 28, batch 36750, libri_loss[loss=0.2828, simple_loss=0.3594, pruned_loss=0.1031, over 29666.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3511, pruned_loss=0.1026, over 5705162.04 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3502, pruned_loss=0.109, over 5672339.91 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3504, pruned_loss=0.1016, over 5697545.47 frames. ], batch size: 91, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:40:04,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1267688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:40:14,246 INFO [train.py:968] (1/2) Epoch 28, batch 36800, giga_loss[loss=0.2423, simple_loss=0.321, pruned_loss=0.08184, over 28936.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3486, pruned_loss=0.1001, over 5707983.04 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3496, pruned_loss=0.1085, over 5680510.28 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3485, pruned_loss=0.09942, over 5695359.49 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:40:23,863 INFO [optim.py:369] (1/2) 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 16:40:37,084 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:40:58,901 INFO [train.py:968] (1/2) Epoch 28, batch 36850, giga_loss[loss=0.2412, simple_loss=0.3119, pruned_loss=0.0853, over 28166.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.344, pruned_loss=0.09718, over 5701597.42 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3499, pruned_loss=0.1085, over 5684797.08 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3437, pruned_loss=0.09651, over 5688436.27 frames. ], batch size: 77, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:41:37,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3944, 1.6333, 1.3589, 1.4886], device='cuda:1'), covar=tensor([0.0843, 0.0342, 0.0361, 0.0961], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 16:41:41,663 INFO [train.py:968] (1/2) Epoch 28, batch 36900, giga_loss[loss=0.2158, simple_loss=0.2971, pruned_loss=0.06726, over 28735.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3381, pruned_loss=0.09388, over 5712144.51 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3501, pruned_loss=0.1086, over 5689760.24 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3375, pruned_loss=0.0929, over 5697707.11 frames. ], batch size: 242, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:41:53,362 INFO [optim.py:369] (1/2) 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 16:41:57,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9868, 1.3855, 1.1440, 0.1887], device='cuda:1'), covar=tensor([0.3869, 0.3239, 0.3987, 0.6594], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1735, 0.1654, 0.1498], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 16:42:04,046 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:42:17,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1267831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:42:19,538 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1267834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:42:33,814 INFO [train.py:968] (1/2) Epoch 28, batch 36950, giga_loss[loss=0.2294, simple_loss=0.31, pruned_loss=0.07446, over 28886.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.3334, pruned_loss=0.09182, over 5693196.05 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3507, pruned_loss=0.1089, over 5692438.64 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.332, pruned_loss=0.09044, over 5679472.10 frames. ], batch size: 145, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:42:49,381 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1267863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:43:19,389 INFO [train.py:968] (1/2) Epoch 28, batch 37000, giga_loss[loss=0.2557, simple_loss=0.3377, pruned_loss=0.08687, over 28860.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3321, pruned_loss=0.09093, over 5691287.10 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3505, pruned_loss=0.1087, over 5696458.37 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3307, pruned_loss=0.08957, over 5676603.43 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:43:29,349 INFO [optim.py:369] (1/2) 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 16:43:34,836 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1267916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:43:36,738 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.38 vs. limit=2.0
+2023-03-14 16:43:58,880 INFO [train.py:968] (1/2) Epoch 28, batch 37050, giga_loss[loss=0.254, simple_loss=0.3339, pruned_loss=0.08705, over 28741.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3338, pruned_loss=0.09182, over 5694489.50 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3508, pruned_loss=0.1087, over 5702521.59 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3319, pruned_loss=0.09013, over 5677071.79 frames. ], batch size: 284, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:44:39,786 INFO [train.py:968] (1/2) Epoch 28, batch 37100, giga_loss[loss=0.2786, simple_loss=0.3558, pruned_loss=0.1008, over 28376.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3337, pruned_loss=0.09145, over 5700129.59 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3511, pruned_loss=0.1089, over 5699281.98 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3316, pruned_loss=0.08965, over 5689153.72 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:44:51,276 INFO [optim.py:369] (1/2) 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 16:45:19,779 INFO [train.py:968] (1/2) Epoch 28, batch 37150, giga_loss[loss=0.3266, simple_loss=0.3938, pruned_loss=0.1297, over 26712.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3334, pruned_loss=0.09148, over 5702185.88 frames. ], libri_tot_loss[loss=0.2845, simple_loss=0.3515, pruned_loss=0.1088, over 5707196.28 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3307, pruned_loss=0.08951, over 5686011.13 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:45:43,667 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 16:45:59,989 INFO [train.py:968] (1/2) Epoch 28, batch 37200, giga_loss[loss=0.2162, simple_loss=0.2981, pruned_loss=0.06711, over 28999.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3317, pruned_loss=0.09093, over 5697036.64 frames. ], libri_tot_loss[loss=0.2856, simple_loss=0.3524, pruned_loss=0.1094, over 5698128.45 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3284, pruned_loss=0.08853, over 5693024.21 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:46:01,348 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:46:36,235 INFO [train.py:968] (1/2) Epoch 28, batch 37250, giga_loss[loss=0.2235, simple_loss=0.3003, pruned_loss=0.07341, over 28945.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.3299, pruned_loss=0.08997, over 5701401.62 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3527, pruned_loss=0.1095, over 5694721.85 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3263, pruned_loss=0.08742, over 5701463.83 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:47:13,898 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:47:15,689 INFO [train.py:968] (1/2) Epoch 28, batch 37300, giga_loss[loss=0.2216, simple_loss=0.2931, pruned_loss=0.07505, over 28614.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3268, pruned_loss=0.08849, over 5710194.38 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3528, pruned_loss=0.1095, over 5695923.24 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3238, pruned_loss=0.08637, over 5709171.44 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:47:27,938 INFO [optim.py:369] (1/2) 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 16:47:38,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1161, 1.2735, 1.1181, 0.8453], device='cuda:1'), covar=tensor([0.1244, 0.0581, 0.1159, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0449, 0.0526, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 16:47:53,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:47:55,556 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:47:56,607 INFO [train.py:968] (1/2) Epoch 28, batch 37350, giga_loss[loss=0.2252, simple_loss=0.2983, pruned_loss=0.07608, over 28892.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3255, pruned_loss=0.08805, over 5707155.03 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.353, pruned_loss=0.1095, over 5698072.48 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3227, pruned_loss=0.08619, over 5704419.32 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:48:17,463 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:48:29,856 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:48:35,404 INFO [train.py:968] (1/2) Epoch 28, batch 37400, libri_loss[loss=0.2931, simple_loss=0.3706, pruned_loss=0.1078, over 29515.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3246, pruned_loss=0.08747, over 5705628.75 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3537, pruned_loss=0.1096, over 5695056.26 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3207, pruned_loss=0.08508, over 5707033.40 frames. ], batch size: 82, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:48:46,836 INFO [optim.py:369] (1/2) 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 16:49:06,724 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:49:08,736 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 37450, giga_loss[loss=0.2534, simple_loss=0.3299, pruned_loss=0.08839, over 28730.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3232, pruned_loss=0.08682, over 5713444.21 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3547, pruned_loss=0.1098, over 5698712.13 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.318, pruned_loss=0.08378, over 5711785.79 frames. ], batch size: 262, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:49:29,917 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:49:51,771 INFO [train.py:968] (1/2) Epoch 28, batch 37500, giga_loss[loss=0.2382, simple_loss=0.3145, pruned_loss=0.08095, over 28909.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3219, pruned_loss=0.08616, over 5723199.36 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3549, pruned_loss=0.11, over 5702779.38 frames. ], giga_tot_loss[loss=0.2416, simple_loss=0.3169, pruned_loss=0.08313, over 5718511.89 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:50:03,751 INFO [optim.py:369] (1/2) 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 16:50:05,347 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268415.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:14,670 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:15,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5584, 1.8502, 1.4855, 1.5744], device='cuda:1'), covar=tensor([0.2729, 0.2821, 0.3120, 0.2554], device='cuda:1'), in_proj_covar=tensor([0.1609, 0.1157, 0.1421, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 16:50:21,566 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:23,447 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 28, batch 37550, giga_loss[loss=0.2213, simple_loss=0.2956, pruned_loss=0.0735, over 28767.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3195, pruned_loss=0.0848, over 5727861.09 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3548, pruned_loss=0.1099, over 5704866.92 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3153, pruned_loss=0.08229, over 5722514.61 frames. ], batch size: 119, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 16:50:47,655 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:50:52,985 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:51:15,701 INFO [train.py:968] (1/2) Epoch 28, batch 37600, giga_loss[loss=0.2435, simple_loss=0.3136, pruned_loss=0.0867, over 28702.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3223, pruned_loss=0.08671, over 5718186.07 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.355, pruned_loss=0.1099, over 5704093.59 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3181, pruned_loss=0.08418, over 5714899.98 frames. ], batch size: 92, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:51:28,044 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 37650, giga_loss[loss=0.2731, simple_loss=0.3484, pruned_loss=0.09888, over 28973.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3267, pruned_loss=0.08941, over 5715242.36 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3551, pruned_loss=0.1098, over 5707591.24 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.3229, pruned_loss=0.0872, over 5709716.56 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:52:12,880 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.84 vs. limit=2.0
+2023-03-14 16:52:48,293 INFO [train.py:968] (1/2) Epoch 28, batch 37700, giga_loss[loss=0.3361, simple_loss=0.3923, pruned_loss=0.1399, over 28833.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3347, pruned_loss=0.09442, over 5704843.03 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3556, pruned_loss=0.11, over 5710759.22 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3307, pruned_loss=0.09212, over 5697707.45 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:52:58,593 INFO [optim.py:369] (1/2) 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 16:53:34,323 INFO [train.py:968] (1/2) Epoch 28, batch 37750, giga_loss[loss=0.2543, simple_loss=0.3329, pruned_loss=0.08789, over 29017.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3407, pruned_loss=0.09797, over 5685424.70 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3557, pruned_loss=0.1101, over 5708307.75 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3373, pruned_loss=0.09597, over 5681598.54 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:54:22,489 INFO [train.py:968] (1/2) Epoch 28, batch 37800, libri_loss[loss=0.2213, simple_loss=0.3012, pruned_loss=0.07065, over 29356.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3446, pruned_loss=0.09902, over 5686795.71 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3558, pruned_loss=0.1099, over 5711311.98 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3416, pruned_loss=0.09735, over 5680398.74 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:54:33,479 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 37850, giga_loss[loss=0.2843, simple_loss=0.3626, pruned_loss=0.103, over 28873.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3488, pruned_loss=0.1012, over 5685597.39 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3554, pruned_loss=0.1096, over 5718778.12 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3465, pruned_loss=0.09982, over 5672586.53 frames. ], batch size: 199, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:55:41,247 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:55:49,114 INFO [train.py:968] (1/2) Epoch 28, batch 37900, giga_loss[loss=0.3725, simple_loss=0.4054, pruned_loss=0.1698, over 26796.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3557, pruned_loss=0.1062, over 5678755.65 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3556, pruned_loss=0.1098, over 5722394.60 frames. ], giga_tot_loss[loss=0.2816, simple_loss=0.3536, pruned_loss=0.1048, over 5664439.88 frames. ], batch size: 555, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:55:50,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:56:00,000 INFO [optim.py:369] (1/2) 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 16:56:02,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:56:27,504 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1268847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:56:28,499 INFO [train.py:968] (1/2) Epoch 28, batch 37950, giga_loss[loss=0.2344, simple_loss=0.3237, pruned_loss=0.07257, over 28931.00 frames. ], tot_loss[loss=0.2791, simple_loss=0.3521, pruned_loss=0.1031, over 5680866.89 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.356, pruned_loss=0.1101, over 5712724.57 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3501, pruned_loss=0.1015, over 5676648.32 frames. ], batch size: 164, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:57:08,425 INFO [train.py:968] (1/2) Epoch 28, batch 38000, giga_loss[loss=0.2732, simple_loss=0.3479, pruned_loss=0.09923, over 28530.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3497, pruned_loss=0.1008, over 5685873.78 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3557, pruned_loss=0.11, over 5715588.29 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3482, pruned_loss=0.09944, over 5679199.58 frames. ], batch size: 71, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:57:10,792 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1268901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:18,791 INFO [optim.py:369] (1/2) 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 16:57:35,082 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:38,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:44,895 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:47,711 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 38050, giga_loss[loss=0.2871, simple_loss=0.3506, pruned_loss=0.1118, over 23344.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3474, pruned_loss=0.09881, over 5686143.72 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3554, pruned_loss=0.1099, over 5719350.88 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3463, pruned_loss=0.09762, over 5676573.37 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 16:57:55,758 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:57:58,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:01,260 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:10,317 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1268990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:20,626 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1268990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:23,234 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1268993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:58:28,220 INFO [train.py:968] (1/2) Epoch 28, batch 38100, giga_loss[loss=0.2473, simple_loss=0.338, pruned_loss=0.07829, over 28862.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3477, pruned_loss=0.09833, over 5666791.78 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3559, pruned_loss=0.1102, over 5690582.98 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3462, pruned_loss=0.09679, over 5684834.05 frames. ], batch size: 174, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:58:28,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5254, 1.7957, 1.2631, 1.2794], device='cuda:1'), covar=tensor([0.1182, 0.0642, 0.1182, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0450, 0.0525, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 16:58:42,298 INFO [optim.py:369] (1/2) 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 16:58:50,158 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1269022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 16:59:10,924 INFO [train.py:968] (1/2) Epoch 28, batch 38150, giga_loss[loss=0.3165, simple_loss=0.3872, pruned_loss=0.1229, over 28566.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3498, pruned_loss=0.09973, over 5670433.64 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3557, pruned_loss=0.1101, over 5692637.71 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3486, pruned_loss=0.09839, over 5682517.64 frames. ], batch size: 336, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 16:59:31,714 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 16:59:57,471 INFO [train.py:968] (1/2) Epoch 28, batch 38200, giga_loss[loss=0.28, simple_loss=0.3522, pruned_loss=0.1039, over 29040.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3511, pruned_loss=0.1008, over 5676222.05 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3559, pruned_loss=0.1102, over 5696080.83 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3499, pruned_loss=0.09949, over 5682448.14 frames. ], batch size: 128, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:00:09,094 INFO [optim.py:369] (1/2) 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 17:00:41,176 INFO [train.py:968] (1/2) Epoch 28, batch 38250, giga_loss[loss=0.2921, simple_loss=0.365, pruned_loss=0.1096, over 28928.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3529, pruned_loss=0.1024, over 5685958.90 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3559, pruned_loss=0.1102, over 5698580.81 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3519, pruned_loss=0.1013, over 5688445.79 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:01:24,103 INFO [train.py:968] (1/2) Epoch 28, batch 38300, giga_loss[loss=0.2814, simple_loss=0.3523, pruned_loss=0.1053, over 29028.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3538, pruned_loss=0.1035, over 5688195.93 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3557, pruned_loss=0.11, over 5702590.30 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3531, pruned_loss=0.1027, over 5686303.61 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:01:35,641 INFO [optim.py:369] (1/2) 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 17:01:58,501 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.75 vs. limit=5.0
+2023-03-14 17:02:05,448 INFO [train.py:968] (1/2) Epoch 28, batch 38350, giga_loss[loss=0.3066, simple_loss=0.3739, pruned_loss=0.1197, over 28937.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.354, pruned_loss=0.1037, over 5699306.04 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3559, pruned_loss=0.1101, over 5707519.25 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3534, pruned_loss=0.1028, over 5693270.32 frames. ], batch size: 213, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:02:28,306 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 38400, giga_loss[loss=0.3048, simple_loss=0.3851, pruned_loss=0.1123, over 28885.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3544, pruned_loss=0.1027, over 5707602.27 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3561, pruned_loss=0.1103, over 5708648.65 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3536, pruned_loss=0.1018, over 5701844.34 frames. ], batch size: 186, lr: 1.13e-03, grad_scale: 8.0
+2023-03-14 17:02:58,016 INFO [optim.py:369] (1/2) 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 17:03:17,451 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:03:28,903 INFO [train.py:968] (1/2) Epoch 28, batch 38450, giga_loss[loss=0.3146, simple_loss=0.3644, pruned_loss=0.1324, over 23949.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3555, pruned_loss=0.1025, over 5703682.89 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3567, pruned_loss=0.1107, over 5710973.17 frames. ], giga_tot_loss[loss=0.2784, simple_loss=0.3543, pruned_loss=0.1013, over 5696979.98 frames. ], batch size: 705, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:04:08,349 INFO [train.py:968] (1/2) Epoch 28, batch 38500, giga_loss[loss=0.28, simple_loss=0.3615, pruned_loss=0.09924, over 28197.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3547, pruned_loss=0.1015, over 5710449.91 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3569, pruned_loss=0.1107, over 5711877.59 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3536, pruned_loss=0.1003, over 5704454.04 frames. ], batch size: 368, lr: 1.13e-03, grad_scale: 4.0
+2023-03-14 17:04:22,285 INFO [optim.py:369] (1/2) 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 17:04:25,725 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1269419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:04:28,007 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1269422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:04:39,769 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 17:04:42,393 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5077, 1.7003, 1.3542, 1.3441], device='cuda:1'), covar=tensor([0.0906, 0.0453, 0.0917, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0448, 0.0523, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:04:49,409 INFO [train.py:968] (1/2) Epoch 28, batch 38550, giga_loss[loss=0.2434, simple_loss=0.3211, pruned_loss=0.08285, over 28812.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3521, pruned_loss=0.1007, over 5701542.32 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.357, pruned_loss=0.1108, over 5705731.01 frames. ], giga_tot_loss[loss=0.2749, simple_loss=0.351, pruned_loss=0.09942, over 5702483.38 frames. ], batch size: 112, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:04:50,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1269451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:05:20,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1556, 1.2705, 1.0929, 0.9028], device='cuda:1'), covar=tensor([0.1200, 0.0599, 0.1183, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0448, 0.0524, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:05:28,131 INFO [train.py:968] (1/2) Epoch 28, batch 38600, giga_loss[loss=0.2194, simple_loss=0.3033, pruned_loss=0.06774, over 28497.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.09961, over 5709882.73 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3569, pruned_loss=0.1105, over 5708203.66 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3491, pruned_loss=0.09851, over 5708216.59 frames. ], batch size: 60, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:05:35,802 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:05:42,595 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 38650, giga_loss[loss=0.2679, simple_loss=0.3466, pruned_loss=0.09458, over 29034.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3472, pruned_loss=0.09793, over 5715009.79 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3571, pruned_loss=0.1106, over 5712072.04 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3461, pruned_loss=0.09677, over 5710464.91 frames. ], batch size: 155, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:06:42,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8680, 2.1720, 1.4676, 1.7246], device='cuda:1'), covar=tensor([0.1085, 0.0633, 0.1045, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0448, 0.0523, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:06:49,768 INFO [train.py:968] (1/2) Epoch 28, batch 38700, giga_loss[loss=0.2967, simple_loss=0.3604, pruned_loss=0.1164, over 27476.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3474, pruned_loss=0.09836, over 5710491.15 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.357, pruned_loss=0.1106, over 5709331.68 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3466, pruned_loss=0.09743, over 5709516.24 frames. ], batch size: 472, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:07:03,631 INFO [optim.py:369] (1/2) 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 17:07:10,917 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 17:07:28,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-14 17:07:28,996 INFO [train.py:968] (1/2) Epoch 28, batch 38750, giga_loss[loss=0.2585, simple_loss=0.3396, pruned_loss=0.08873, over 28929.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3483, pruned_loss=0.09855, over 5715379.43 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3571, pruned_loss=0.1106, over 5712541.30 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3475, pruned_loss=0.09765, over 5711573.63 frames. ], batch size: 106, lr: 1.13e-03, grad_scale: 2.0
+2023-03-14 17:07:59,677 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:08:08,493 INFO [train.py:968] (1/2) Epoch 28, batch 38800, giga_loss[loss=0.2983, simple_loss=0.3674, pruned_loss=0.1146, over 28859.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3482, pruned_loss=0.09777, over 5710821.23 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3571, pruned_loss=0.1104, over 5714747.43 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3475, pruned_loss=0.09707, over 5705922.26 frames. ], batch size: 106, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:08:15,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1269709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:08:21,569 INFO [optim.py:369] (1/2) 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 17:08:43,730 INFO [train.py:968] (1/2) Epoch 28, batch 38850, giga_loss[loss=0.2682, simple_loss=0.3457, pruned_loss=0.09533, over 28875.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3473, pruned_loss=0.09725, over 5710274.71 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3575, pruned_loss=0.1108, over 5709235.50 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3459, pruned_loss=0.09586, over 5711575.25 frames. ], batch size: 227, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:09:22,971 INFO [train.py:968] (1/2) Epoch 28, batch 38900, giga_loss[loss=0.2541, simple_loss=0.3353, pruned_loss=0.0865, over 28955.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3467, pruned_loss=0.09718, over 5710039.63 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3574, pruned_loss=0.1107, over 5711072.75 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3455, pruned_loss=0.09592, over 5709376.76 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:09:39,145 INFO [optim.py:369] (1/2) 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 17:10:03,157 INFO [train.py:968] (1/2) Epoch 28, batch 38950, giga_loss[loss=0.2275, simple_loss=0.3089, pruned_loss=0.07311, over 28300.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3454, pruned_loss=0.09722, over 5700463.87 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3582, pruned_loss=0.1114, over 5705207.68 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3437, pruned_loss=0.09541, over 5704839.86 frames. ], batch size: 65, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:10:05,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1269852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:07,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1269855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:16,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6143, 1.9249, 1.5810, 1.6776], device='cuda:1'), covar=tensor([0.3008, 0.2895, 0.3305, 0.2700], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1157, 0.1420, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 17:10:26,748 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1269882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:28,775 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1269884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:10:33,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 17:10:40,095 INFO [train.py:968] (1/2) Epoch 28, batch 39000, giga_loss[loss=0.2831, simple_loss=0.3584, pruned_loss=0.1039, over 28306.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3413, pruned_loss=0.09511, over 5709899.98 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3573, pruned_loss=0.1108, over 5713294.36 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3401, pruned_loss=0.09361, over 5705951.24 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:10:40,095 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 17:10:48,943 INFO [train.py:1012] (1/2) Epoch 28, validation: loss=0.2032, simple_loss=0.3111, pruned_loss=0.0476, over 944034.00 frames. 
+2023-03-14 17:10:48,943 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 17:11:02,226 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 39050, giga_loss[loss=0.2443, simple_loss=0.3209, pruned_loss=0.0838, over 28797.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3401, pruned_loss=0.09478, over 5711460.79 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3568, pruned_loss=0.1106, over 5718231.40 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3389, pruned_loss=0.09309, over 5703847.25 frames. ], batch size: 92, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:11:28,709 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-14 17:11:48,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9943, 2.1156, 2.1447, 1.7523], device='cuda:1'), covar=tensor([0.1885, 0.2440, 0.1508, 0.1761], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0720, 0.0988, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:12:00,373 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1269990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:07,257 INFO [train.py:968] (1/2) Epoch 28, batch 39100, giga_loss[loss=0.2568, simple_loss=0.32, pruned_loss=0.0968, over 28150.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3404, pruned_loss=0.0955, over 5705575.21 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 5721999.54 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3392, pruned_loss=0.09401, over 5695979.75 frames. ], batch size: 77, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:12:22,565 INFO [optim.py:369] (1/2) 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 17:12:28,276 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:30,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:44,949 INFO [train.py:968] (1/2) Epoch 28, batch 39150, libri_loss[loss=0.2902, simple_loss=0.3591, pruned_loss=0.1107, over 29521.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3372, pruned_loss=0.09389, over 5709148.49 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3565, pruned_loss=0.1104, over 5722842.61 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.336, pruned_loss=0.09228, over 5700293.59 frames. ], batch size: 84, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 17:12:51,256 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:12:56,142 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1270063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:13:02,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 17:13:09,018 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:13:23,485 INFO [train.py:968] (1/2) Epoch 28, batch 39200, giga_loss[loss=0.2318, simple_loss=0.3055, pruned_loss=0.07905, over 28996.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3353, pruned_loss=0.0931, over 5720029.86 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3566, pruned_loss=0.1105, over 5729432.52 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3334, pruned_loss=0.09115, over 5706467.28 frames. ], batch size: 106, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:13:38,407 INFO [optim.py:369] (1/2) 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 17:14:02,040 INFO [train.py:968] (1/2) Epoch 28, batch 39250, giga_loss[loss=0.2984, simple_loss=0.3579, pruned_loss=0.1194, over 23831.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3327, pruned_loss=0.09188, over 5714352.88 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3567, pruned_loss=0.1104, over 5730930.05 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3307, pruned_loss=0.09001, over 5702050.89 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:14:13,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2313, 1.6813, 1.2107, 0.6027], device='cuda:1'), covar=tensor([0.4617, 0.2233, 0.3170, 0.6568], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1720, 0.1654, 0.1495], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 17:14:43,306 INFO [train.py:968] (1/2) Epoch 28, batch 39300, giga_loss[loss=0.2497, simple_loss=0.3292, pruned_loss=0.08508, over 28738.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3306, pruned_loss=0.0904, over 5719079.45 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3564, pruned_loss=0.1102, over 5733135.61 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3288, pruned_loss=0.08877, over 5706971.94 frames. ], batch size: 262, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:14:51,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:14:52,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8562, 3.6934, 3.4781, 2.0303], device='cuda:1'), covar=tensor([0.0590, 0.0739, 0.0740, 0.1827], device='cuda:1'), in_proj_covar=tensor([0.1286, 0.1187, 0.0999, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 17:14:53,037 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:14:59,355 INFO [optim.py:369] (1/2) 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 17:15:17,799 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270238.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:15:26,972 INFO [train.py:968] (1/2) Epoch 28, batch 39350, giga_loss[loss=0.237, simple_loss=0.3246, pruned_loss=0.07473, over 28718.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3327, pruned_loss=0.0913, over 5715660.89 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3565, pruned_loss=0.1103, over 5735885.04 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3307, pruned_loss=0.08945, over 5703121.25 frames. ], batch size: 262, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:15:36,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5189, 1.7571, 1.2291, 1.2633], device='cuda:1'), covar=tensor([0.1056, 0.0599, 0.1148, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0449, 0.0524, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:16:05,651 INFO [train.py:968] (1/2) Epoch 28, batch 39400, giga_loss[loss=0.2522, simple_loss=0.3389, pruned_loss=0.08276, over 28682.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3359, pruned_loss=0.09277, over 5701845.86 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3566, pruned_loss=0.1103, over 5727354.28 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3334, pruned_loss=0.09064, over 5697561.02 frames. ], batch size: 262, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:16:22,633 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 39450, giga_loss[loss=0.2319, simple_loss=0.3208, pruned_loss=0.0715, over 28994.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3394, pruned_loss=0.09414, over 5702962.71 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3564, pruned_loss=0.1101, over 5732277.33 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3368, pruned_loss=0.09198, over 5693883.57 frames. ], batch size: 213, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:17:02,355 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1270365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:17:29,453 INFO [train.py:968] (1/2) Epoch 28, batch 39500, giga_loss[loss=0.2332, simple_loss=0.3173, pruned_loss=0.07452, over 28891.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3406, pruned_loss=0.09457, over 5689085.96 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3562, pruned_loss=0.1102, over 5723491.49 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09229, over 5688946.39 frames. ], batch size: 112, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:17:45,101 INFO [optim.py:369] (1/2) 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 17:17:49,108 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3082, 1.4280, 1.4217, 1.3359], device='cuda:1'), covar=tensor([0.3279, 0.2647, 0.2468, 0.2727], device='cuda:1'), in_proj_covar=tensor([0.2060, 0.2022, 0.1931, 0.2078], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 17:18:09,125 INFO [train.py:968] (1/2) Epoch 28, batch 39550, giga_loss[loss=0.2403, simple_loss=0.3165, pruned_loss=0.08202, over 28317.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3397, pruned_loss=0.09317, over 5690167.53 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3563, pruned_loss=0.1102, over 5717777.77 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3372, pruned_loss=0.09082, over 5695150.44 frames. ], batch size: 77, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:18:16,563 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1270454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:18:38,148 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.47 vs. limit=5.0
+2023-03-14 17:18:51,637 INFO [train.py:968] (1/2) Epoch 28, batch 39600, giga_loss[loss=0.2559, simple_loss=0.3242, pruned_loss=0.09381, over 28697.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3387, pruned_loss=0.0927, over 5695717.72 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3561, pruned_loss=0.1102, over 5719532.03 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3367, pruned_loss=0.09074, over 5697773.23 frames. ], batch size: 85, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:18:59,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:19:02,723 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:19:07,838 INFO [optim.py:369] (1/2) 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 17:19:09,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9103, 1.9516, 2.0729, 1.6207], device='cuda:1'), covar=tensor([0.1860, 0.2496, 0.1548, 0.1813], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0718, 0.0986, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:19:18,794 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3430, 1.1421, 3.8273, 3.2351], device='cuda:1'), covar=tensor([0.1537, 0.2664, 0.0501, 0.1024], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0669, 0.0998, 0.0973], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 17:19:26,453 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:19:39,984 INFO [train.py:968] (1/2) Epoch 28, batch 39650, giga_loss[loss=0.26, simple_loss=0.3399, pruned_loss=0.0901, over 28916.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3392, pruned_loss=0.09338, over 5692281.92 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3562, pruned_loss=0.1102, over 5709638.33 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3373, pruned_loss=0.09149, over 5702722.19 frames. ], batch size: 213, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:19:49,434 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-14 17:19:51,639 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6280, 1.6462, 1.8042, 1.3923], device='cuda:1'), covar=tensor([0.1920, 0.2621, 0.1579, 0.1766], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0717, 0.0985, 0.0884], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:20:21,493 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1270597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:20:22,367 INFO [train.py:968] (1/2) Epoch 28, batch 39700, giga_loss[loss=0.3177, simple_loss=0.3894, pruned_loss=0.1229, over 28948.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3399, pruned_loss=0.09372, over 5705023.46 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3563, pruned_loss=0.1103, over 5710466.61 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.338, pruned_loss=0.09195, over 5712343.67 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:20:23,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1270600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:20:37,427 INFO [optim.py:369] (1/2) 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 17:20:48,130 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1270629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:20:50,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:21:02,614 INFO [train.py:968] (1/2) Epoch 28, batch 39750, giga_loss[loss=0.2802, simple_loss=0.362, pruned_loss=0.09919, over 28864.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3454, pruned_loss=0.09712, over 5697700.92 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3572, pruned_loss=0.1108, over 5704228.34 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3426, pruned_loss=0.09478, over 5709138.99 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:21:42,501 INFO [train.py:968] (1/2) Epoch 28, batch 39800, libri_loss[loss=0.2767, simple_loss=0.3372, pruned_loss=0.1081, over 29488.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3473, pruned_loss=0.09784, over 5694094.99 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3568, pruned_loss=0.1103, over 5700009.35 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3451, pruned_loss=0.09596, over 5707869.81 frames. ], batch size: 70, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:21:56,829 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 39850, giga_loss[loss=0.2326, simple_loss=0.3166, pruned_loss=0.07434, over 29136.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3488, pruned_loss=0.09852, over 5694431.14 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1105, over 5696334.93 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3466, pruned_loss=0.0965, over 5708822.98 frames. ], batch size: 113, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:22:36,335 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270767.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:22:59,505 INFO [train.py:968] (1/2) Epoch 28, batch 39900, libri_loss[loss=0.2535, simple_loss=0.3315, pruned_loss=0.08772, over 29553.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.351, pruned_loss=0.09984, over 5697114.68 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3576, pruned_loss=0.1106, over 5698528.56 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3485, pruned_loss=0.09769, over 5706547.53 frames. ], batch size: 75, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:23:01,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5290, 1.7352, 1.4715, 1.5479], device='cuda:1'), covar=tensor([0.2318, 0.2464, 0.2602, 0.2395], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1156, 0.1417, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 17:23:05,210 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:23:09,897 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5355, 1.5575, 1.7446, 1.3493], device='cuda:1'), covar=tensor([0.1655, 0.2370, 0.1358, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0716, 0.0983, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:23:16,891 INFO [optim.py:369] (1/2) 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 17:23:41,567 INFO [train.py:968] (1/2) Epoch 28, batch 39950, giga_loss[loss=0.2667, simple_loss=0.3417, pruned_loss=0.09578, over 28295.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3519, pruned_loss=0.101, over 5698022.42 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3574, pruned_loss=0.1105, over 5698147.18 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09925, over 5705871.09 frames. ], batch size: 77, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:23:48,158 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1846, 0.9008, 0.9395, 1.3902], device='cuda:1'), covar=tensor([0.0732, 0.0419, 0.0382, 0.0858], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 17:23:51,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5437, 1.5235, 4.2373, 3.4146], device='cuda:1'), covar=tensor([0.1601, 0.2603, 0.0430, 0.0805], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0670, 0.0999, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 17:24:20,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3846, 3.5684, 1.5411, 1.6291], device='cuda:1'), covar=tensor([0.1006, 0.0365, 0.0933, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0567, 0.0409, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 17:24:20,356 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-14 17:24:21,408 INFO [train.py:968] (1/2) Epoch 28, batch 40000, giga_loss[loss=0.2618, simple_loss=0.3429, pruned_loss=0.09035, over 28368.00 frames. ], tot_loss[loss=0.2754, simple_loss=0.3508, pruned_loss=0.1, over 5706938.97 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.357, pruned_loss=0.1101, over 5704410.57 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3494, pruned_loss=0.09868, over 5707951.14 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:24:37,378 INFO [optim.py:369] (1/2) 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 17:24:40,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5691, 1.5124, 1.7808, 1.3901], device='cuda:1'), covar=tensor([0.1602, 0.2469, 0.1390, 0.1691], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0716, 0.0984, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:25:02,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5911, 1.8515, 1.2926, 1.4145], device='cuda:1'), covar=tensor([0.1038, 0.0661, 0.1167, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0448, 0.0523, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:25:02,908 INFO [train.py:968] (1/2) Epoch 28, batch 40050, giga_loss[loss=0.2375, simple_loss=0.3134, pruned_loss=0.08074, over 28950.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3481, pruned_loss=0.09888, over 5704604.72 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3572, pruned_loss=0.1102, over 5697398.21 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3466, pruned_loss=0.09754, over 5712529.57 frames. ], batch size: 106, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:25:09,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1270959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:25:26,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9139, 1.1924, 1.0610, 0.8979], device='cuda:1'), covar=tensor([0.2744, 0.2996, 0.1850, 0.2568], device='cuda:1'), in_proj_covar=tensor([0.2072, 0.2041, 0.1947, 0.2094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 17:25:40,425 INFO [train.py:968] (1/2) Epoch 28, batch 40100, giga_loss[loss=0.3044, simple_loss=0.3678, pruned_loss=0.1204, over 27956.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3452, pruned_loss=0.09762, over 5692044.24 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3579, pruned_loss=0.1107, over 5682181.22 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3434, pruned_loss=0.09596, over 5712170.94 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:25:48,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:25:58,106 INFO [optim.py:369] (1/2) 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 17:26:14,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6577, 1.9380, 1.3477, 1.5070], device='cuda:1'), covar=tensor([0.1099, 0.0723, 0.1178, 0.1265], device='cuda:1'), in_proj_covar=tensor([0.0412, 0.0449, 0.0523, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:26:22,689 INFO [train.py:968] (1/2) Epoch 28, batch 40150, giga_loss[loss=0.2363, simple_loss=0.3163, pruned_loss=0.07822, over 28857.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3441, pruned_loss=0.09672, over 5690009.59 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3583, pruned_loss=0.1109, over 5674359.64 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.342, pruned_loss=0.09498, over 5712199.03 frames. ], batch size: 112, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:26:53,217 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-14 17:26:55,535 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4013, 3.8945, 1.6481, 1.6111], device='cuda:1'), covar=tensor([0.0988, 0.0427, 0.0884, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0423, 0.0567, 0.0409, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 17:26:59,425 INFO [train.py:968] (1/2) Epoch 28, batch 40200, giga_loss[loss=0.2904, simple_loss=0.3728, pruned_loss=0.104, over 28971.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.346, pruned_loss=0.09617, over 5700114.58 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3584, pruned_loss=0.111, over 5675093.18 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3439, pruned_loss=0.09434, over 5717810.78 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:27:18,523 INFO [optim.py:369] (1/2) 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 17:27:38,223 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271142.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:27:43,602 INFO [train.py:968] (1/2) Epoch 28, batch 40250, giga_loss[loss=0.2937, simple_loss=0.3665, pruned_loss=0.1105, over 28942.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.346, pruned_loss=0.09529, over 5695696.12 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3586, pruned_loss=0.111, over 5675992.72 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3441, pruned_loss=0.09374, over 5708823.11 frames. ], batch size: 213, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:27:45,413 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:27:48,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5266, 1.7610, 1.5031, 1.4198], device='cuda:1'), covar=tensor([0.2227, 0.2320, 0.2406, 0.2236], device='cuda:1'), in_proj_covar=tensor([0.1609, 0.1158, 0.1420, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 17:27:48,961 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1271154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:28:04,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271178.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:28:08,561 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1271183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:28:15,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2903, 3.1428, 2.9521, 1.3528], device='cuda:1'), covar=tensor([0.0963, 0.1083, 0.0949, 0.2400], device='cuda:1'), in_proj_covar=tensor([0.1286, 0.1189, 0.1000, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 17:28:17,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-14 17:28:22,066 INFO [train.py:968] (1/2) Epoch 28, batch 40300, giga_loss[loss=0.2602, simple_loss=0.3291, pruned_loss=0.09571, over 29107.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3455, pruned_loss=0.09636, over 5703671.32 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3584, pruned_loss=0.1109, over 5682316.06 frames. ], giga_tot_loss[loss=0.2667, simple_loss=0.3438, pruned_loss=0.09478, over 5709043.99 frames. ], batch size: 128, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:28:37,797 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 40350, giga_loss[loss=0.2715, simple_loss=0.3437, pruned_loss=0.09968, over 28737.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3448, pruned_loss=0.09734, over 5706848.80 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3583, pruned_loss=0.111, over 5684772.83 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3435, pruned_loss=0.09588, over 5709109.59 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:29:03,903 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5173, 4.4862, 1.6351, 1.8291], device='cuda:1'), covar=tensor([0.0969, 0.0351, 0.0947, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0565, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 17:29:31,940 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271285.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:29:34,601 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 40400, giga_loss[loss=0.2697, simple_loss=0.3448, pruned_loss=0.09735, over 28744.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.344, pruned_loss=0.09827, over 5688907.50 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3589, pruned_loss=0.1114, over 5669015.11 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3422, pruned_loss=0.09655, over 5704334.57 frames. ], batch size: 262, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:29:58,967 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:30:01,524 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:30:03,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1271324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:30:07,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3676, 3.7280, 1.6090, 1.5045], device='cuda:1'), covar=tensor([0.0973, 0.0388, 0.0952, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0566, 0.0408, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 17:30:12,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1271334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:30:25,203 INFO [train.py:968] (1/2) Epoch 28, batch 40450, giga_loss[loss=0.299, simple_loss=0.3682, pruned_loss=0.1149, over 27578.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3433, pruned_loss=0.09843, over 5699057.52 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3593, pruned_loss=0.1116, over 5669292.79 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3412, pruned_loss=0.09664, over 5711352.89 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:30:27,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1271353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:30:32,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2289, 1.6937, 1.3353, 0.4989], device='cuda:1'), covar=tensor([0.5224, 0.2950, 0.4452, 0.7288], device='cuda:1'), in_proj_covar=tensor([0.1842, 0.1722, 0.1655, 0.1495], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 17:30:57,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4577, 2.9517, 1.6024, 1.6291], device='cuda:1'), covar=tensor([0.0804, 0.0341, 0.0775, 0.1105], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0566, 0.0409, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0026, 0.0031], device='cuda:1')
+2023-03-14 17:31:02,896 INFO [train.py:968] (1/2) Epoch 28, batch 40500, giga_loss[loss=0.2684, simple_loss=0.3428, pruned_loss=0.097, over 28884.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.342, pruned_loss=0.09809, over 5706353.94 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3592, pruned_loss=0.1117, over 5671517.45 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3399, pruned_loss=0.09622, over 5715155.10 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:31:19,854 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 40550, giga_loss[loss=0.2504, simple_loss=0.318, pruned_loss=0.09143, over 28815.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3385, pruned_loss=0.09624, over 5713733.57 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3596, pruned_loss=0.112, over 5677844.70 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3359, pruned_loss=0.09413, over 5715956.07 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:31:49,377 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7354, 1.8989, 1.6006, 1.8596], device='cuda:1'), covar=tensor([0.2743, 0.2965, 0.3404, 0.2589], device='cuda:1'), in_proj_covar=tensor([0.1611, 0.1157, 0.1422, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 17:32:04,985 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1271477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:32:06,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1271480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:32:07,697 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-14 17:32:21,026 INFO [train.py:968] (1/2) Epoch 28, batch 40600, giga_loss[loss=0.2314, simple_loss=0.3129, pruned_loss=0.075, over 28570.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3323, pruned_loss=0.09291, over 5720166.53 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3596, pruned_loss=0.112, over 5680264.87 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3301, pruned_loss=0.09104, over 5720131.38 frames. ], batch size: 336, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:32:30,625 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1271509.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:32:38,496 INFO [optim.py:369] (1/2) 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 17:32:51,880 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1271538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:33:00,786 INFO [train.py:968] (1/2) Epoch 28, batch 40650, giga_loss[loss=0.2916, simple_loss=0.3451, pruned_loss=0.1191, over 23874.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3303, pruned_loss=0.09168, over 5696728.15 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3596, pruned_loss=0.112, over 5663878.30 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3278, pruned_loss=0.08961, over 5712144.73 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:33:27,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5324, 4.3841, 4.1410, 2.0343], device='cuda:1'), covar=tensor([0.0553, 0.0699, 0.0669, 0.1982], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1194, 0.1004, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 17:33:41,914 INFO [train.py:968] (1/2) Epoch 28, batch 40700, libri_loss[loss=0.2858, simple_loss=0.3591, pruned_loss=0.1062, over 29511.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3327, pruned_loss=0.09244, over 5701574.09 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3597, pruned_loss=0.1122, over 5668476.72 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.33, pruned_loss=0.09022, over 5709922.65 frames. ], batch size: 84, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:33:47,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-14 17:33:59,128 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 40750, giga_loss[loss=0.2493, simple_loss=0.3323, pruned_loss=0.08318, over 29059.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3366, pruned_loss=0.09413, over 5699547.45 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3601, pruned_loss=0.1125, over 5664448.15 frames. ], giga_tot_loss[loss=0.2584, simple_loss=0.3335, pruned_loss=0.09168, over 5710727.14 frames. ], batch size: 155, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:35:00,477 INFO [train.py:968] (1/2) Epoch 28, batch 40800, giga_loss[loss=0.2873, simple_loss=0.3589, pruned_loss=0.1078, over 28625.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3388, pruned_loss=0.09445, over 5702914.98 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3602, pruned_loss=0.1127, over 5665742.79 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3359, pruned_loss=0.09204, over 5711178.28 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:35:12,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.32 vs. limit=5.0
+2023-03-14 17:35:17,132 INFO [optim.py:369] (1/2) 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 17:35:21,619 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1271726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:35:41,449 INFO [train.py:968] (1/2) Epoch 28, batch 40850, giga_loss[loss=0.2949, simple_loss=0.375, pruned_loss=0.1075, over 28574.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3409, pruned_loss=0.09496, over 5713326.92 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3604, pruned_loss=0.1127, over 5668907.61 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3378, pruned_loss=0.09252, over 5718034.36 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:36:18,459 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2572, 1.5057, 1.3298, 1.1154], device='cuda:1'), covar=tensor([0.3056, 0.3002, 0.2075, 0.2746], device='cuda:1'), in_proj_covar=tensor([0.2076, 0.2045, 0.1953, 0.2096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 17:36:22,031 INFO [train.py:968] (1/2) Epoch 28, batch 40900, giga_loss[loss=0.278, simple_loss=0.3453, pruned_loss=0.1053, over 24006.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3444, pruned_loss=0.09701, over 5697452.35 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3612, pruned_loss=0.1132, over 5660803.90 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.341, pruned_loss=0.09435, over 5708802.29 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:36:39,810 INFO [optim.py:369] (1/2) 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 17:36:59,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6969, 2.0174, 1.4714, 1.6184], device='cuda:1'), covar=tensor([0.0872, 0.0475, 0.0915, 0.0955], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0450, 0.0525, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:37:06,247 INFO [train.py:968] (1/2) Epoch 28, batch 40950, giga_loss[loss=0.2704, simple_loss=0.3455, pruned_loss=0.09761, over 28904.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3482, pruned_loss=0.1006, over 5703564.79 frames. ], libri_tot_loss[loss=0.2942, simple_loss=0.3614, pruned_loss=0.1135, over 5669313.77 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3448, pruned_loss=0.09767, over 5706119.18 frames. ], batch size: 213, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:37:40,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6311, 1.6653, 1.8328, 1.4056], device='cuda:1'), covar=tensor([0.1870, 0.2672, 0.1517, 0.1769], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0714, 0.0981, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:37:51,134 INFO [train.py:968] (1/2) Epoch 28, batch 41000, giga_loss[loss=0.3191, simple_loss=0.3876, pruned_loss=0.1252, over 28960.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3552, pruned_loss=0.1068, over 5689069.17 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.113, over 5678674.82 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.3529, pruned_loss=0.1045, over 5683505.59 frames. ], batch size: 285, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:37:55,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=2.03 vs. limit=2.0
+2023-03-14 17:38:04,925 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:38:34,511 INFO [train.py:968] (1/2) Epoch 28, batch 41050, giga_loss[loss=0.2966, simple_loss=0.3736, pruned_loss=0.1097, over 29069.00 frames. ], tot_loss[loss=0.2925, simple_loss=0.3618, pruned_loss=0.1116, over 5684377.37 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3607, pruned_loss=0.1132, over 5679284.76 frames. ], giga_tot_loss[loss=0.2893, simple_loss=0.3597, pruned_loss=0.1095, over 5679585.38 frames. ], batch size: 128, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:38:45,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8028, 1.8132, 1.9694, 1.5471], device='cuda:1'), covar=tensor([0.1853, 0.2524, 0.1475, 0.1796], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0714, 0.0980, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:39:20,023 INFO [train.py:968] (1/2) Epoch 28, batch 41100, giga_loss[loss=0.2998, simple_loss=0.3702, pruned_loss=0.1147, over 28602.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3686, pruned_loss=0.1172, over 5663183.96 frames. ], libri_tot_loss[loss=0.2941, simple_loss=0.3611, pruned_loss=0.1135, over 5672416.24 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3667, pruned_loss=0.1152, over 5666281.31 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:39:36,705 INFO [optim.py:369] (1/2) 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 17:40:00,676 INFO [train.py:968] (1/2) Epoch 28, batch 41150, giga_loss[loss=0.3578, simple_loss=0.4183, pruned_loss=0.1487, over 28990.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3744, pruned_loss=0.1224, over 5664691.18 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3608, pruned_loss=0.1134, over 5667269.80 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3735, pruned_loss=0.1211, over 5671717.59 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:40:08,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:40:11,009 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:40:27,549 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6100, 1.9044, 1.8284, 1.7211], device='cuda:1'), covar=tensor([0.2179, 0.2122, 0.2317, 0.2063], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0763, 0.0732, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 17:40:38,134 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 41200, giga_loss[loss=0.3499, simple_loss=0.4068, pruned_loss=0.1465, over 28861.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3787, pruned_loss=0.1257, over 5659524.34 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.113, over 5670251.40 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3786, pruned_loss=0.1252, over 5662375.10 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:40:53,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1272101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:41:12,077 INFO [optim.py:369] (1/2) 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 17:41:14,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:41:35,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5669, 1.8006, 1.2739, 1.4159], device='cuda:1'), covar=tensor([0.0938, 0.0567, 0.0919, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0452, 0.0527, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 17:41:39,655 INFO [train.py:968] (1/2) Epoch 28, batch 41250, giga_loss[loss=0.3191, simple_loss=0.3837, pruned_loss=0.1272, over 28816.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3801, pruned_loss=0.1273, over 5663868.83 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.113, over 5674845.02 frames. ], giga_tot_loss[loss=0.3174, simple_loss=0.3804, pruned_loss=0.1272, over 5661562.43 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:42:36,628 INFO [train.py:968] (1/2) Epoch 28, batch 41300, giga_loss[loss=0.3102, simple_loss=0.3802, pruned_loss=0.1202, over 28912.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3828, pruned_loss=0.1311, over 5628343.15 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1129, over 5669078.96 frames. ], giga_tot_loss[loss=0.3231, simple_loss=0.3836, pruned_loss=0.1313, over 5630605.81 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:42:37,945 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5950, 1.8650, 1.5342, 1.6891], device='cuda:1'), covar=tensor([0.2224, 0.2175, 0.2404, 0.2102], device='cuda:1'), in_proj_covar=tensor([0.1605, 0.1155, 0.1419, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 17:43:01,719 INFO [optim.py:369] (1/2) 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 17:43:21,902 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:43:23,978 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 41350, giga_loss[loss=0.3855, simple_loss=0.4268, pruned_loss=0.1721, over 28709.00 frames. ], tot_loss[loss=0.3279, simple_loss=0.3862, pruned_loss=0.1348, over 5626007.03 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1132, over 5674374.57 frames. ], giga_tot_loss[loss=0.3289, simple_loss=0.3871, pruned_loss=0.1353, over 5621617.17 frames. ], batch size: 262, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:43:52,962 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:44:15,659 INFO [train.py:968] (1/2) Epoch 28, batch 41400, libri_loss[loss=0.291, simple_loss=0.367, pruned_loss=0.1075, over 29533.00 frames. ], tot_loss[loss=0.3317, simple_loss=0.3893, pruned_loss=0.1371, over 5630636.35 frames. ], libri_tot_loss[loss=0.2938, simple_loss=0.3609, pruned_loss=0.1134, over 5676915.20 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3905, pruned_loss=0.1378, over 5623617.73 frames. ], batch size: 89, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:44:19,716 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6956, 1.9356, 1.5983, 2.0247], device='cuda:1'), covar=tensor([0.2591, 0.2765, 0.3023, 0.2454], device='cuda:1'), in_proj_covar=tensor([0.1605, 0.1156, 0.1419, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 17:44:36,598 INFO [optim.py:369] (1/2) 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 17:45:03,857 INFO [train.py:968] (1/2) Epoch 28, batch 41450, giga_loss[loss=0.3717, simple_loss=0.4189, pruned_loss=0.1622, over 28282.00 frames. ], tot_loss[loss=0.3322, simple_loss=0.3893, pruned_loss=0.1375, over 5634836.86 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5674183.21 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3916, pruned_loss=0.1392, over 5629196.03 frames. ], batch size: 369, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:45:10,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 17:45:18,922 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272360.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:45:34,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-14 17:45:53,145 INFO [train.py:968] (1/2) Epoch 28, batch 41500, giga_loss[loss=0.3328, simple_loss=0.3879, pruned_loss=0.1389, over 29009.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3871, pruned_loss=0.1367, over 5635058.82 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3605, pruned_loss=0.1133, over 5677875.39 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3896, pruned_loss=0.1385, over 5626284.17 frames. ], batch size: 128, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:45:57,593 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:46:21,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4556, 3.3334, 1.5423, 1.6094], device='cuda:1'), covar=tensor([0.1012, 0.0342, 0.0904, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0571, 0.0411, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 17:46:39,320 INFO [train.py:968] (1/2) Epoch 28, batch 41550, giga_loss[loss=0.3271, simple_loss=0.3827, pruned_loss=0.1357, over 28925.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3871, pruned_loss=0.1368, over 5631315.33 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3607, pruned_loss=0.1134, over 5675803.21 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3897, pruned_loss=0.1389, over 5623933.29 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:47:08,923 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.91 vs. limit=5.0
+2023-03-14 17:47:24,127 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0933, 2.1872, 2.2280, 1.7930], device='cuda:1'), covar=tensor([0.1836, 0.2345, 0.1458, 0.1722], device='cuda:1'), in_proj_covar=tensor([0.0928, 0.0713, 0.0977, 0.0877], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:47:26,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1272497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:47:27,693 INFO [train.py:968] (1/2) Epoch 28, batch 41600, libri_loss[loss=0.2739, simple_loss=0.3443, pruned_loss=0.1017, over 29505.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3853, pruned_loss=0.1342, over 5625193.80 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3606, pruned_loss=0.1132, over 5679646.20 frames. ], giga_tot_loss[loss=0.3306, simple_loss=0.3881, pruned_loss=0.1366, over 5614947.08 frames. ], batch size: 81, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:47:31,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5232, 3.6521, 1.6833, 1.7051], device='cuda:1'), covar=tensor([0.1014, 0.0332, 0.0873, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0573, 0.0412, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 17:47:51,923 INFO [optim.py:369] (1/2) 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 17:48:03,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2945, 1.6727, 1.4662, 1.5081], device='cuda:1'), covar=tensor([0.0752, 0.0379, 0.0330, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 17:48:14,517 INFO [train.py:968] (1/2) Epoch 28, batch 41650, giga_loss[loss=0.3157, simple_loss=0.3761, pruned_loss=0.1277, over 28629.00 frames. ], tot_loss[loss=0.3312, simple_loss=0.3887, pruned_loss=0.1368, over 5619428.12 frames. ], libri_tot_loss[loss=0.2934, simple_loss=0.3606, pruned_loss=0.1131, over 5679723.30 frames. ], giga_tot_loss[loss=0.3351, simple_loss=0.3916, pruned_loss=0.1393, over 5610100.57 frames. ], batch size: 85, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:49:05,643 INFO [train.py:968] (1/2) Epoch 28, batch 41700, giga_loss[loss=0.314, simple_loss=0.3747, pruned_loss=0.1266, over 27532.00 frames. ], tot_loss[loss=0.3284, simple_loss=0.3862, pruned_loss=0.1353, over 5613576.61 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3603, pruned_loss=0.1131, over 5687976.71 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.3897, pruned_loss=0.1381, over 5596326.33 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:49:23,334 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1861, 0.7914, 0.8684, 1.3903], device='cuda:1'), covar=tensor([0.0790, 0.0404, 0.0379, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 17:49:27,504 INFO [optim.py:369] (1/2) 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 17:49:45,391 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:49:47,678 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:49:52,333 INFO [train.py:968] (1/2) Epoch 28, batch 41750, giga_loss[loss=0.31, simple_loss=0.382, pruned_loss=0.119, over 28768.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3822, pruned_loss=0.1307, over 5618009.11 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3603, pruned_loss=0.1131, over 5682152.07 frames. ], giga_tot_loss[loss=0.3265, simple_loss=0.3858, pruned_loss=0.1336, over 5606915.30 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:50:16,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:50:21,058 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1783, 1.8888, 5.0493, 3.6435], device='cuda:1'), covar=tensor([0.1344, 0.2464, 0.0439, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0675, 0.1012, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 17:50:42,538 INFO [train.py:968] (1/2) Epoch 28, batch 41800, giga_loss[loss=0.3002, simple_loss=0.3648, pruned_loss=0.1179, over 28970.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3798, pruned_loss=0.1274, over 5633997.23 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3602, pruned_loss=0.1131, over 5684762.79 frames. ], giga_tot_loss[loss=0.3212, simple_loss=0.3829, pruned_loss=0.1298, over 5622235.27 frames. ], batch size: 106, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:51:06,433 INFO [optim.py:369] (1/2) 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 17:51:14,014 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 17:51:16,564 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 41850, giga_loss[loss=0.2677, simple_loss=0.3452, pruned_loss=0.09514, over 28876.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3759, pruned_loss=0.1243, over 5632596.54 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3598, pruned_loss=0.1128, over 5686644.74 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3789, pruned_loss=0.1267, over 5620706.09 frames. ], batch size: 112, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:51:47,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:51:55,845 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272773.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:52:00,530 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1272778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:52:16,624 INFO [train.py:968] (1/2) Epoch 28, batch 41900, giga_loss[loss=0.346, simple_loss=0.4001, pruned_loss=0.146, over 27940.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3743, pruned_loss=0.1232, over 5624739.55 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 5677208.64 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3772, pruned_loss=0.1254, over 5621957.88 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:52:25,546 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 17:52:37,859 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5314, 1.9226, 1.4962, 1.5450], device='cuda:1'), covar=tensor([0.2572, 0.2586, 0.2949, 0.2424], device='cuda:1'), in_proj_covar=tensor([0.1604, 0.1155, 0.1418, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 17:52:41,613 INFO [optim.py:369] (1/2) 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 17:53:01,936 INFO [train.py:968] (1/2) Epoch 28, batch 41950, giga_loss[loss=0.3244, simple_loss=0.388, pruned_loss=0.1304, over 28737.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3722, pruned_loss=0.1215, over 5642229.84 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1129, over 5684885.50 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3751, pruned_loss=0.1236, over 5631118.44 frames. ], batch size: 284, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:53:26,883 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272878.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 17:53:29,490 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1272881.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:53:32,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2718, 1.2400, 3.7107, 3.0865], device='cuda:1'), covar=tensor([0.1742, 0.2887, 0.0489, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0676, 0.1012, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 17:53:47,931 INFO [train.py:968] (1/2) Epoch 28, batch 42000, giga_loss[loss=0.2736, simple_loss=0.3545, pruned_loss=0.09636, over 29024.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3715, pruned_loss=0.1212, over 5653883.16 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3593, pruned_loss=0.1127, over 5690501.65 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3744, pruned_loss=0.1232, over 5639496.69 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:53:47,931 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 17:53:56,247 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 17:54:08,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272910.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:54:20,663 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1272921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:54:21,711 INFO [optim.py:369] (1/2) 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 17:54:22,733 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 42050, libri_loss[loss=0.243, simple_loss=0.3072, pruned_loss=0.08943, over 29370.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3701, pruned_loss=0.1201, over 5644277.80 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1124, over 5694643.73 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3731, pruned_loss=0.1222, over 5627732.42 frames. ], batch size: 67, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:54:48,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5985, 1.6784, 1.7845, 1.3766], device='cuda:1'), covar=tensor([0.1999, 0.2844, 0.1705, 0.1988], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0716, 0.0979, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:54:50,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1272953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:54:56,107 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4123, 1.1394, 4.3690, 3.4851], device='cuda:1'), covar=tensor([0.1761, 0.3090, 0.0452, 0.1169], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0677, 0.1015, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 17:55:00,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272962.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:55:02,453 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8004, 2.5425, 1.5365, 0.9255], device='cuda:1'), covar=tensor([0.9084, 0.4529, 0.4796, 0.8598], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1742, 0.1664, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 17:55:03,261 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:55:16,905 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3426, 1.6632, 1.6367, 1.5009], device='cuda:1'), covar=tensor([0.2099, 0.2039, 0.2149, 0.2088], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0767, 0.0736, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-14 17:55:34,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2803, 3.5697, 1.5142, 1.4620], device='cuda:1'), covar=tensor([0.1095, 0.0453, 0.0967, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0574, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 17:55:39,158 INFO [train.py:968] (1/2) Epoch 28, batch 42100, giga_loss[loss=0.2821, simple_loss=0.3736, pruned_loss=0.09529, over 28968.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.369, pruned_loss=0.1168, over 5650820.15 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.359, pruned_loss=0.1126, over 5697450.88 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3715, pruned_loss=0.1184, over 5634629.49 frames. ], batch size: 164, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:56:00,597 INFO [optim.py:369] (1/2) 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 17:56:26,202 INFO [train.py:968] (1/2) Epoch 28, batch 42150, giga_loss[loss=0.3091, simple_loss=0.378, pruned_loss=0.1201, over 29044.00 frames. ], tot_loss[loss=0.302, simple_loss=0.371, pruned_loss=0.1165, over 5655752.08 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3591, pruned_loss=0.1128, over 5688948.04 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3731, pruned_loss=0.1176, over 5649924.57 frames. ], batch size: 128, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 17:57:13,353 INFO [train.py:968] (1/2) Epoch 28, batch 42200, libri_loss[loss=0.3011, simple_loss=0.3729, pruned_loss=0.1147, over 29022.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3729, pruned_loss=0.1185, over 5662575.81 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3588, pruned_loss=0.1126, over 5693555.29 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3752, pruned_loss=0.1197, over 5652595.77 frames. ], batch size: 101, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 17:57:36,740 INFO [optim.py:369] (1/2) 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 17:57:49,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:57:56,569 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273148.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 17:57:56,937 INFO [train.py:968] (1/2) Epoch 28, batch 42250, giga_loss[loss=0.2845, simple_loss=0.3635, pruned_loss=0.1027, over 28822.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3722, pruned_loss=0.1184, over 5656725.71 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.359, pruned_loss=0.1127, over 5683583.48 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3743, pruned_loss=0.1194, over 5656090.95 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:58:39,272 INFO [train.py:968] (1/2) Epoch 28, batch 42300, giga_loss[loss=0.2707, simple_loss=0.3472, pruned_loss=0.09706, over 28857.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3703, pruned_loss=0.118, over 5664568.84 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1124, over 5685543.40 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3725, pruned_loss=0.1192, over 5662131.39 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:58:51,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:58:56,817 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 17:59:04,154 INFO [optim.py:369] (1/2) 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 17:59:24,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4924, 1.6172, 1.7186, 1.3486], device='cuda:1'), covar=tensor([0.1578, 0.2294, 0.1336, 0.1543], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0717, 0.0981, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 17:59:25,796 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-14 17:59:25,851 INFO [train.py:968] (1/2) Epoch 28, batch 42350, giga_loss[loss=0.2834, simple_loss=0.3563, pruned_loss=0.1053, over 28558.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3699, pruned_loss=0.1194, over 5655024.77 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3586, pruned_loss=0.1124, over 5681113.51 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3721, pruned_loss=0.1206, over 5655582.37 frames. ], batch size: 78, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 17:59:56,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-14 17:59:57,122 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 17:59:59,112 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:00:01,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-14 18:00:05,415 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273291.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:00:08,425 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 42400, giga_loss[loss=0.3072, simple_loss=0.3696, pruned_loss=0.1223, over 29034.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3694, pruned_loss=0.1191, over 5661128.73 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3587, pruned_loss=0.1125, over 5686382.87 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3713, pruned_loss=0.1201, over 5656074.09 frames. ], batch size: 113, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:00:27,744 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:00:35,595 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273323.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:00:36,488 INFO [optim.py:369] (1/2) 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 18:00:47,829 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273337.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:00:51,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:00:56,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2873, 1.1221, 1.1193, 1.4565], device='cuda:1'), covar=tensor([0.0808, 0.0392, 0.0375, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 18:00:56,791 INFO [train.py:968] (1/2) Epoch 28, batch 42450, giga_loss[loss=0.2923, simple_loss=0.3654, pruned_loss=0.1096, over 28970.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3687, pruned_loss=0.1169, over 5663675.40 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3587, pruned_loss=0.1125, over 5678665.14 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3704, pruned_loss=0.1178, over 5666679.07 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:01:13,343 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5500, 2.2512, 1.7068, 0.7211], device='cuda:1'), covar=tensor([0.7773, 0.3958, 0.4552, 0.7936], device='cuda:1'), in_proj_covar=tensor([0.1852, 0.1741, 0.1663, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 18:01:45,251 INFO [train.py:968] (1/2) Epoch 28, batch 42500, giga_loss[loss=0.3305, simple_loss=0.3704, pruned_loss=0.1453, over 23644.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3687, pruned_loss=0.1164, over 5656882.17 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3587, pruned_loss=0.1125, over 5671023.73 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3702, pruned_loss=0.1172, over 5664888.58 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:02:01,983 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 18:02:10,821 INFO [optim.py:369] (1/2) 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 18:02:27,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7627, 1.6359, 1.9457, 1.5419], device='cuda:1'), covar=tensor([0.1596, 0.2237, 0.1309, 0.1544], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0716, 0.0981, 0.0880], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 18:02:31,517 INFO [train.py:968] (1/2) Epoch 28, batch 42550, giga_loss[loss=0.3105, simple_loss=0.3743, pruned_loss=0.1234, over 28003.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3693, pruned_loss=0.117, over 5661176.78 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3588, pruned_loss=0.1125, over 5675098.69 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3707, pruned_loss=0.1178, over 5663490.43 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:02:57,757 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273480.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:03:00,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273483.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:03:01,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:03:03,859 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1273487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:03:08,771 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-14 18:03:15,698 INFO [train.py:968] (1/2) Epoch 28, batch 42600, giga_loss[loss=0.3095, simple_loss=0.374, pruned_loss=0.1225, over 28071.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3663, pruned_loss=0.1153, over 5676354.18 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3583, pruned_loss=0.1122, over 5681904.96 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3681, pruned_loss=0.1163, over 5671989.30 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:03:28,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273512.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:03:33,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:03:40,947 INFO [optim.py:369] (1/2) 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 18:04:00,856 INFO [train.py:968] (1/2) Epoch 28, batch 42650, giga_loss[loss=0.2296, simple_loss=0.3094, pruned_loss=0.07484, over 28478.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.3649, pruned_loss=0.115, over 5677315.83 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3579, pruned_loss=0.1117, over 5688679.94 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3669, pruned_loss=0.1163, over 5667311.10 frames. ], batch size: 71, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:04:37,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:04:45,510 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:04:49,844 INFO [train.py:968] (1/2) Epoch 28, batch 42700, giga_loss[loss=0.3116, simple_loss=0.3721, pruned_loss=0.1255, over 28670.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3644, pruned_loss=0.1153, over 5684879.10 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.358, pruned_loss=0.1116, over 5691896.58 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.366, pruned_loss=0.1165, over 5673908.81 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:05:06,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8400, 2.8813, 1.8668, 0.8498], device='cuda:1'), covar=tensor([0.9341, 0.3611, 0.4423, 0.8801], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1741, 0.1662, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 18:05:14,857 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 42750, libri_loss[loss=0.3446, simple_loss=0.3946, pruned_loss=0.1473, over 19374.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3636, pruned_loss=0.1153, over 5675402.85 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3578, pruned_loss=0.1115, over 5684618.13 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3652, pruned_loss=0.1163, over 5674006.84 frames. ], batch size: 187, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:06:05,687 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 18:06:28,254 INFO [train.py:968] (1/2) Epoch 28, batch 42800, giga_loss[loss=0.2906, simple_loss=0.3659, pruned_loss=0.1077, over 28689.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3641, pruned_loss=0.1168, over 5658035.56 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3577, pruned_loss=0.1114, over 5685746.04 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3654, pruned_loss=0.1178, over 5656012.91 frames. ], batch size: 262, lr: 1.12e-03, grad_scale: 8.0
+2023-03-14 18:06:52,791 INFO [optim.py:369] (1/2) 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 18:06:57,021 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1273730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:07:00,278 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 28, batch 42850, giga_loss[loss=0.2837, simple_loss=0.3617, pruned_loss=0.1028, over 28994.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.364, pruned_loss=0.1171, over 5650133.67 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3581, pruned_loss=0.1117, over 5679826.44 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3648, pruned_loss=0.1177, over 5652922.59 frames. ], batch size: 136, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:07:26,101 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1273762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:07:29,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3737, 1.6538, 1.1695, 1.3268], device='cuda:1'), covar=tensor([0.0955, 0.0460, 0.1022, 0.0889], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0454, 0.0527, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 18:07:55,934 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5252, 1.5632, 1.7030, 1.3027], device='cuda:1'), covar=tensor([0.1845, 0.2611, 0.1562, 0.1836], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0720, 0.0983, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 18:07:56,198 INFO [train.py:968] (1/2) Epoch 28, batch 42900, giga_loss[loss=0.2848, simple_loss=0.3621, pruned_loss=0.1037, over 28895.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3652, pruned_loss=0.1173, over 5652286.25 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3587, pruned_loss=0.1123, over 5673347.29 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3656, pruned_loss=0.1174, over 5659741.32 frames. ], batch size: 213, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:08:15,793 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:08:21,163 INFO [optim.py:369] (1/2) 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 18:08:38,945 INFO [train.py:968] (1/2) Epoch 28, batch 42950, giga_loss[loss=0.2753, simple_loss=0.3558, pruned_loss=0.09737, over 28828.00 frames. ], tot_loss[loss=0.299, simple_loss=0.3652, pruned_loss=0.1164, over 5660705.53 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3587, pruned_loss=0.1123, over 5675459.96 frames. ], giga_tot_loss[loss=0.2995, simple_loss=0.3657, pruned_loss=0.1166, over 5664389.48 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:08:42,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4607, 2.1475, 1.5241, 0.7478], device='cuda:1'), covar=tensor([0.6576, 0.3485, 0.4873, 0.7232], device='cuda:1'), in_proj_covar=tensor([0.1855, 0.1742, 0.1662, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 18:09:23,339 INFO [train.py:968] (1/2) Epoch 28, batch 43000, giga_loss[loss=0.3167, simple_loss=0.3723, pruned_loss=0.1305, over 27524.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3652, pruned_loss=0.1159, over 5675230.64 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3584, pruned_loss=0.1122, over 5682452.68 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.366, pruned_loss=0.1163, over 5671616.88 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:09:47,982 INFO [optim.py:369] (1/2) 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 18:10:11,957 INFO [train.py:968] (1/2) Epoch 28, batch 43050, libri_loss[loss=0.3046, simple_loss=0.3724, pruned_loss=0.1184, over 25696.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3659, pruned_loss=0.1164, over 5678492.06 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3587, pruned_loss=0.1123, over 5684307.19 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3664, pruned_loss=0.1168, over 5673981.96 frames. ], batch size: 136, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:10:26,720 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1273967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:10:30,666 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1273971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:10:39,980 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-14 18:10:55,904 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.71 vs. limit=5.0
+2023-03-14 18:10:56,117 INFO [train.py:968] (1/2) Epoch 28, batch 43100, giga_loss[loss=0.2972, simple_loss=0.3665, pruned_loss=0.114, over 28766.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3683, pruned_loss=0.1189, over 5672971.38 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3586, pruned_loss=0.1123, over 5671441.00 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3691, pruned_loss=0.1192, over 5680350.95 frames. ], batch size: 119, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:11:25,677 INFO [optim.py:369] (1/2) 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 18:11:48,019 INFO [train.py:968] (1/2) Epoch 28, batch 43150, giga_loss[loss=0.3172, simple_loss=0.3799, pruned_loss=0.1273, over 28948.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3696, pruned_loss=0.1209, over 5664611.23 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1125, over 5665957.22 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3701, pruned_loss=0.1211, over 5676086.64 frames. ], batch size: 145, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:12:09,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-14 18:12:26,284 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5704, 1.7932, 1.3079, 1.3587], device='cuda:1'), covar=tensor([0.1038, 0.0580, 0.1037, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0454, 0.0527, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 18:12:41,875 INFO [train.py:968] (1/2) Epoch 28, batch 43200, giga_loss[loss=0.2735, simple_loss=0.3413, pruned_loss=0.1029, over 28226.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3704, pruned_loss=0.1222, over 5660347.20 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3588, pruned_loss=0.1124, over 5667095.43 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3709, pruned_loss=0.1225, over 5668335.17 frames. ], batch size: 77, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:12:49,733 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-14 18:12:52,649 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:12:56,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:13:08,385 INFO [optim.py:369] (1/2) 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 18:13:23,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:13:27,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4483, 3.4916, 1.5367, 1.6510], device='cuda:1'), covar=tensor([0.1061, 0.0386, 0.0946, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0579, 0.0415, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 18:13:28,160 INFO [train.py:968] (1/2) Epoch 28, batch 43250, giga_loss[loss=0.2798, simple_loss=0.3494, pruned_loss=0.1051, over 28852.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.371, pruned_loss=0.1229, over 5656600.79 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.359, pruned_loss=0.1124, over 5669151.30 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3713, pruned_loss=0.1232, over 5660986.89 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:13:44,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5423, 2.3184, 1.7501, 0.7496], device='cuda:1'), covar=tensor([0.6203, 0.3134, 0.4076, 0.7336], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1746, 0.1664, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 18:14:12,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:14:14,630 INFO [train.py:968] (1/2) Epoch 28, batch 43300, giga_loss[loss=0.2806, simple_loss=0.3655, pruned_loss=0.09784, over 28801.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3707, pruned_loss=0.1226, over 5656476.68 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1125, over 5671196.98 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3708, pruned_loss=0.1228, over 5657922.22 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:14:15,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6416, 1.6632, 1.8222, 1.3954], device='cuda:1'), covar=tensor([0.1927, 0.2634, 0.1591, 0.1844], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0720, 0.0983, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 18:14:29,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2891, 1.6078, 1.3020, 0.9743], device='cuda:1'), covar=tensor([0.2841, 0.2844, 0.3325, 0.2662], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1158, 0.1421, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 18:14:39,767 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 43350, giga_loss[loss=0.2903, simple_loss=0.3571, pruned_loss=0.1118, over 28191.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3712, pruned_loss=0.1216, over 5660086.51 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1125, over 5673312.67 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.122, over 5658982.94 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:15:41,386 INFO [train.py:968] (1/2) Epoch 28, batch 43400, giga_loss[loss=0.2795, simple_loss=0.3589, pruned_loss=0.1, over 28969.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3691, pruned_loss=0.1193, over 5664736.66 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5680446.85 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3699, pruned_loss=0.1202, over 5657362.82 frames. ], batch size: 136, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:16:05,282 INFO [optim.py:369] (1/2) 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 18:16:15,771 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:18,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:22,709 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:24,986 INFO [train.py:968] (1/2) Epoch 28, batch 43450, giga_loss[loss=0.3018, simple_loss=0.3683, pruned_loss=0.1177, over 28644.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3669, pruned_loss=0.1182, over 5672933.91 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3593, pruned_loss=0.1124, over 5683121.87 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3676, pruned_loss=0.1189, over 5664430.31 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:16:46,054 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:16:49,607 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-14 18:17:09,573 INFO [train.py:968] (1/2) Epoch 28, batch 43500, libri_loss[loss=0.3229, simple_loss=0.3874, pruned_loss=0.1292, over 29115.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3657, pruned_loss=0.1182, over 5673021.92 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3593, pruned_loss=0.1123, over 5689014.58 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3664, pruned_loss=0.119, over 5660554.14 frames. ], batch size: 101, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:17:09,884 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1274399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:17:16,483 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4926, 1.8400, 1.7330, 1.6248], device='cuda:1'), covar=tensor([0.2146, 0.2159, 0.2277, 0.2148], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0768, 0.0737, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-14 18:17:21,722 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:17:54,480 INFO [train.py:968] (1/2) Epoch 28, batch 43550, giga_loss[loss=0.3513, simple_loss=0.4008, pruned_loss=0.1509, over 28580.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3661, pruned_loss=0.1185, over 5679183.68 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5690888.26 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3675, pruned_loss=0.1198, over 5667087.67 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:18:25,004 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-14 18:18:30,890 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:18:33,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:18:40,337 INFO [train.py:968] (1/2) Epoch 28, batch 43600, giga_loss[loss=0.2781, simple_loss=0.3641, pruned_loss=0.0961, over 28628.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3697, pruned_loss=0.12, over 5668533.50 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3585, pruned_loss=0.1117, over 5686610.48 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3711, pruned_loss=0.1212, over 5661933.27 frames. ], batch size: 242, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:18:51,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1840, 1.4932, 1.1598, 0.6088], device='cuda:1'), covar=tensor([0.2513, 0.1882, 0.2136, 0.4211], device='cuda:1'), in_proj_covar=tensor([0.1850, 0.1743, 0.1660, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 18:19:01,310 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:19:06,763 INFO [optim.py:369] (1/2) 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 18:19:10,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9213, 1.5994, 5.1269, 3.8756], device='cuda:1'), covar=tensor([0.1519, 0.2788, 0.0426, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0675, 0.1014, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 18:19:30,423 INFO [train.py:968] (1/2) Epoch 28, batch 43650, giga_loss[loss=0.2931, simple_loss=0.3501, pruned_loss=0.118, over 23741.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3717, pruned_loss=0.1183, over 5672504.55 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3588, pruned_loss=0.1119, over 5689170.26 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3727, pruned_loss=0.1192, over 5664985.56 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:19:41,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.52 vs. limit=2.0
+2023-03-14 18:20:19,712 INFO [train.py:968] (1/2) Epoch 28, batch 43700, giga_loss[loss=0.3413, simple_loss=0.4035, pruned_loss=0.1395, over 28933.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3739, pruned_loss=0.1199, over 5658213.71 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3592, pruned_loss=0.1123, over 5678489.31 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3745, pruned_loss=0.1203, over 5660973.47 frames. ], batch size: 199, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:20:39,323 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3174, 2.9135, 1.4066, 1.5047], device='cuda:1'), covar=tensor([0.1002, 0.0407, 0.0915, 0.1297], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0577, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 18:20:49,465 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 28, batch 43750, giga_loss[loss=0.339, simple_loss=0.3917, pruned_loss=0.1431, over 27980.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3753, pruned_loss=0.1215, over 5646046.21 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5669711.87 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3757, pruned_loss=0.1218, over 5655482.89 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:21:55,803 INFO [train.py:968] (1/2) Epoch 28, batch 43800, giga_loss[loss=0.3686, simple_loss=0.4094, pruned_loss=0.1639, over 28268.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3753, pruned_loss=0.1227, over 5654849.44 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1121, over 5674471.18 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3764, pruned_loss=0.1233, over 5657624.61 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:22:14,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5321, 4.3738, 4.1852, 1.8781], device='cuda:1'), covar=tensor([0.0579, 0.0755, 0.0727, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1225, 0.1032, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 18:22:18,356 INFO [optim.py:369] (1/2) 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 18:22:32,176 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1274741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:22:38,640 INFO [train.py:968] (1/2) Epoch 28, batch 43850, giga_loss[loss=0.2907, simple_loss=0.3614, pruned_loss=0.11, over 28135.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3744, pruned_loss=0.123, over 5655057.01 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3587, pruned_loss=0.1122, over 5678867.76 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3758, pruned_loss=0.1237, over 5652977.96 frames. ], batch size: 77, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:23:02,435 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:23:12,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1274785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:23:24,283 INFO [train.py:968] (1/2) Epoch 28, batch 43900, giga_loss[loss=0.2974, simple_loss=0.3618, pruned_loss=0.1165, over 28834.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3733, pruned_loss=0.1232, over 5651113.17 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5674693.20 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3745, pruned_loss=0.1237, over 5653251.76 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 18:23:52,729 INFO [optim.py:369] (1/2) 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 18:23:53,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.2955, 6.1618, 5.8748, 3.2712], device='cuda:1'), covar=tensor([0.0438, 0.0514, 0.0638, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.1331, 0.1227, 0.1034, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 18:23:55,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2704, 2.0271, 1.5378, 0.5518], device='cuda:1'), covar=tensor([0.6112, 0.3417, 0.4714, 0.7348], device='cuda:1'), in_proj_covar=tensor([0.1851, 0.1743, 0.1662, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 18:23:55,729 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6084, 1.6190, 1.7880, 1.3813], device='cuda:1'), covar=tensor([0.1824, 0.2612, 0.1484, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0720, 0.0984, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 18:24:01,813 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.62 vs. limit=2.0
+2023-03-14 18:24:08,826 INFO [train.py:968] (1/2) Epoch 28, batch 43950, giga_loss[loss=0.2752, simple_loss=0.3504, pruned_loss=0.09998, over 28987.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3709, pruned_loss=0.1219, over 5658373.23 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3585, pruned_loss=0.1123, over 5670061.61 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3726, pruned_loss=0.1227, over 5664012.23 frames. ], batch size: 136, lr: 1.12e-03, grad_scale: 1.0
+2023-03-14 18:24:21,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1274861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:24:58,441 INFO [train.py:968] (1/2) Epoch 28, batch 44000, giga_loss[loss=0.3474, simple_loss=0.4089, pruned_loss=0.143, over 27926.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.371, pruned_loss=0.1222, over 5656313.24 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3585, pruned_loss=0.1122, over 5663429.81 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3727, pruned_loss=0.1233, over 5666469.26 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:25:02,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.69 vs. limit=5.0
+2023-03-14 18:25:17,096 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1274917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:19,230 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:26,471 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:25:30,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1274931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:47,212 INFO [train.py:968] (1/2) Epoch 28, batch 44050, giga_loss[loss=0.3062, simple_loss=0.3652, pruned_loss=0.1236, over 28832.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3716, pruned_loss=0.123, over 5658191.67 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3582, pruned_loss=0.1121, over 5666895.78 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3734, pruned_loss=0.1241, over 5663070.57 frames. ], batch size: 186, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:25:47,526 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:25:57,697 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1274960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:26:29,756 INFO [train.py:968] (1/2) Epoch 28, batch 44100, giga_loss[loss=0.3228, simple_loss=0.3872, pruned_loss=0.1293, over 28212.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3699, pruned_loss=0.1221, over 5663940.26 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3586, pruned_loss=0.1123, over 5668953.00 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3712, pruned_loss=0.1231, over 5665847.85 frames. ], batch size: 368, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:26:46,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7098, 1.9548, 1.4116, 1.5107], device='cuda:1'), covar=tensor([0.1112, 0.0676, 0.1124, 0.1186], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 18:26:56,398 INFO [optim.py:369] (1/2) 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 18:27:12,352 INFO [train.py:968] (1/2) Epoch 28, batch 44150, giga_loss[loss=0.3578, simple_loss=0.4084, pruned_loss=0.1536, over 27912.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3674, pruned_loss=0.1201, over 5673490.26 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3583, pruned_loss=0.1119, over 5676205.62 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3691, pruned_loss=0.1215, over 5668464.18 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:27:33,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6119, 1.5697, 1.8033, 1.4088], device='cuda:1'), covar=tensor([0.1784, 0.2597, 0.1501, 0.1740], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0718, 0.0983, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 18:27:57,366 INFO [train.py:968] (1/2) Epoch 28, batch 44200, giga_loss[loss=0.3024, simple_loss=0.3723, pruned_loss=0.1163, over 29016.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3684, pruned_loss=0.1201, over 5667129.22 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3587, pruned_loss=0.112, over 5677329.28 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3696, pruned_loss=0.1213, over 5662065.08 frames. ], batch size: 106, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:28:09,566 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3513, 2.8799, 1.4201, 1.5502], device='cuda:1'), covar=tensor([0.1004, 0.0389, 0.0924, 0.1334], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0577, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 18:28:17,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:28:31,492 INFO [optim.py:369] (1/2) 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 18:28:34,070 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:28:48,290 INFO [train.py:968] (1/2) Epoch 28, batch 44250, giga_loss[loss=0.282, simple_loss=0.3516, pruned_loss=0.1062, over 28931.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3705, pruned_loss=0.1209, over 5669241.70 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3588, pruned_loss=0.112, over 5677329.00 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3716, pruned_loss=0.1221, over 5664777.19 frames. ], batch size: 136, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:29:35,608 INFO [train.py:968] (1/2) Epoch 28, batch 44300, giga_loss[loss=0.2824, simple_loss=0.3507, pruned_loss=0.1071, over 28805.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3703, pruned_loss=0.1216, over 5671622.99 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.359, pruned_loss=0.1122, over 5681715.24 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3712, pruned_loss=0.1225, over 5663926.28 frames. ], batch size: 99, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:30:03,547 INFO [optim.py:369] (1/2) 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 18:30:08,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:30:20,122 INFO [train.py:968] (1/2) Epoch 28, batch 44350, giga_loss[loss=0.2841, simple_loss=0.3694, pruned_loss=0.0994, over 28918.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3726, pruned_loss=0.1216, over 5673574.95 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3591, pruned_loss=0.1122, over 5686903.68 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3735, pruned_loss=0.1225, over 5662818.85 frames. ], batch size: 112, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:30:31,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:30:33,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:30:57,461 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:31:03,326 INFO [train.py:968] (1/2) Epoch 28, batch 44400, giga_loss[loss=0.2997, simple_loss=0.3734, pruned_loss=0.113, over 28952.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3724, pruned_loss=0.1188, over 5677024.16 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3588, pruned_loss=0.1119, over 5679649.11 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3738, pruned_loss=0.12, over 5674182.05 frames. ], batch size: 227, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:31:32,502 INFO [optim.py:369] (1/2) 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 18:31:44,112 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275340.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:31:44,184 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8174, 2.2904, 1.9638, 2.1458], device='cuda:1'), covar=tensor([0.0737, 0.0254, 0.0300, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 18:31:53,328 INFO [train.py:968] (1/2) Epoch 28, batch 44450, giga_loss[loss=0.4284, simple_loss=0.4678, pruned_loss=0.1945, over 28609.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3758, pruned_loss=0.1199, over 5680926.91 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3587, pruned_loss=0.1119, over 5681294.24 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3772, pruned_loss=0.121, over 5677097.26 frames. ], batch size: 307, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:32:11,231 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:32:22,247 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:32:25,332 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:32:44,357 INFO [train.py:968] (1/2) Epoch 28, batch 44500, giga_loss[loss=0.4577, simple_loss=0.4661, pruned_loss=0.2247, over 26572.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3781, pruned_loss=0.1224, over 5681660.88 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3588, pruned_loss=0.112, over 5686071.74 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.3795, pruned_loss=0.1233, over 5674208.61 frames. ], batch size: 555, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:32:55,228 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:33:10,051 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-14 18:33:12,370 INFO [optim.py:369] (1/2) 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 18:33:18,550 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4731, 1.8440, 1.3609, 1.7470], device='cuda:1'), covar=tensor([0.2766, 0.2894, 0.3315, 0.2494], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1157, 0.1422, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 18:33:30,310 INFO [train.py:968] (1/2) Epoch 28, batch 44550, giga_loss[loss=0.3486, simple_loss=0.3871, pruned_loss=0.155, over 23402.00 frames. ], tot_loss[loss=0.3148, simple_loss=0.3797, pruned_loss=0.125, over 5667680.61 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3587, pruned_loss=0.1119, over 5686506.20 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3812, pruned_loss=0.126, over 5661172.65 frames. ], batch size: 705, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:33:58,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0931, 1.3646, 1.0516, 0.8995], device='cuda:1'), covar=tensor([0.0933, 0.0366, 0.0944, 0.1065], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0455, 0.0527, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 18:34:18,644 INFO [train.py:968] (1/2) Epoch 28, batch 44600, giga_loss[loss=0.3358, simple_loss=0.3972, pruned_loss=0.1372, over 28869.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3811, pruned_loss=0.1272, over 5660994.34 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.359, pruned_loss=0.1122, over 5689982.73 frames. ], giga_tot_loss[loss=0.3192, simple_loss=0.3824, pruned_loss=0.128, over 5652157.11 frames. ], batch size: 174, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:34:25,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:34:27,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5637, 1.8520, 1.3713, 1.3847], device='cuda:1'), covar=tensor([0.1128, 0.0617, 0.1079, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 18:34:45,417 INFO [optim.py:369] (1/2) 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 18:35:01,507 INFO [train.py:968] (1/2) Epoch 28, batch 44650, giga_loss[loss=0.291, simple_loss=0.3607, pruned_loss=0.1107, over 27527.00 frames. ], tot_loss[loss=0.3147, simple_loss=0.3788, pruned_loss=0.1253, over 5658874.58 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3588, pruned_loss=0.1123, over 5686882.72 frames. ], giga_tot_loss[loss=0.3165, simple_loss=0.3806, pruned_loss=0.1262, over 5654395.25 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:35:40,810 INFO [train.py:968] (1/2) Epoch 28, batch 44700, giga_loss[loss=0.2545, simple_loss=0.348, pruned_loss=0.0805, over 29098.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3762, pruned_loss=0.1217, over 5673684.14 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3584, pruned_loss=0.112, over 5694059.91 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3788, pruned_loss=0.1232, over 5662622.46 frames. ], batch size: 128, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:35:56,844 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-14 18:36:09,234 INFO [optim.py:369] (1/2) 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 18:36:26,390 INFO [train.py:968] (1/2) Epoch 28, batch 44750, giga_loss[loss=0.3805, simple_loss=0.4312, pruned_loss=0.1649, over 27892.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3769, pruned_loss=0.1209, over 5673336.42 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3584, pruned_loss=0.1119, over 5700589.09 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3797, pruned_loss=0.1225, over 5657887.97 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 2.0
+2023-03-14 18:36:26,630 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:36:29,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:36:51,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:37:04,290 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9012, 1.3118, 2.8388, 2.7364], device='cuda:1'), covar=tensor([0.1731, 0.2590, 0.0648, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0682, 0.1022, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-14 18:37:09,608 INFO [train.py:968] (1/2) Epoch 28, batch 44800, giga_loss[loss=0.3294, simple_loss=0.3935, pruned_loss=0.1326, over 28933.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3781, pruned_loss=0.1216, over 5615336.43 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3595, pruned_loss=0.113, over 5628877.71 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3798, pruned_loss=0.1222, over 5666428.86 frames. ], batch size: 136, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:37:23,894 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275715.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 18:37:37,677 INFO [optim.py:369] (1/2) 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 18:37:41,934 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-14 18:37:50,476 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1275742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:37:56,488 INFO [train.py:968] (1/2) Epoch 28, batch 44850, giga_loss[loss=0.3574, simple_loss=0.409, pruned_loss=0.1529, over 27586.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3796, pruned_loss=0.1239, over 5588185.29 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3604, pruned_loss=0.1137, over 5584697.89 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3804, pruned_loss=0.1238, over 5667121.01 frames. ], batch size: 472, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:38:42,349 INFO [train.py:968] (1/2) Epoch 28, batch 44900, giga_loss[loss=0.37, simple_loss=0.4145, pruned_loss=0.1628, over 27956.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3789, pruned_loss=0.1242, over 5576410.64 frames. ], libri_tot_loss[loss=0.2944, simple_loss=0.3608, pruned_loss=0.114, over 5560626.59 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3794, pruned_loss=0.124, over 5660018.77 frames. ], batch size: 412, lr: 1.12e-03, grad_scale: 4.0
+2023-03-14 18:39:03,306 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-14 18:39:45,794 INFO [optim.py:369] (1/2) 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 18:40:08,273 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275858.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:40:11,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275861.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:40:20,510 INFO [train.py:968] (1/2) Epoch 29, batch 50, giga_loss[loss=0.3304, simple_loss=0.392, pruned_loss=0.1344, over 28969.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3719, pruned_loss=0.108, over 1267197.51 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3408, pruned_loss=0.09025, over 117183.83 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3746, pruned_loss=0.1095, over 1173910.21 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:40:34,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1275885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:40:36,537 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1275888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:40:37,699 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275890.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 18:41:02,705 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1275917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:41:05,394 INFO [train.py:968] (1/2) Epoch 29, batch 100, giga_loss[loss=0.26, simple_loss=0.3444, pruned_loss=0.08776, over 28990.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3631, pruned_loss=0.1039, over 2242285.12 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3413, pruned_loss=0.08988, over 287427.12 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3656, pruned_loss=0.1055, over 2057797.43 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:41:07,568 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1275923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:41:13,625 INFO [optim.py:369] (1/2) 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 18:41:49,602 INFO [train.py:968] (1/2) Epoch 29, batch 150, giga_loss[loss=0.2475, simple_loss=0.3265, pruned_loss=0.08427, over 28640.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.348, pruned_loss=0.09662, over 2999565.15 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3442, pruned_loss=0.0916, over 414665.84 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3486, pruned_loss=0.09728, over 2790548.87 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:42:28,654 INFO [train.py:968] (1/2) Epoch 29, batch 200, giga_loss[loss=0.2146, simple_loss=0.3024, pruned_loss=0.06346, over 28958.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.3338, pruned_loss=0.0896, over 3603828.63 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08735, over 602953.62 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3344, pruned_loss=0.09039, over 3354731.65 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:42:36,063 INFO [optim.py:369] (1/2) 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 18:42:57,701 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:43:10,200 INFO [train.py:968] (1/2) Epoch 29, batch 250, giga_loss[loss=0.2001, simple_loss=0.2807, pruned_loss=0.0598, over 28224.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3233, pruned_loss=0.08454, over 4056946.15 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3377, pruned_loss=0.0871, over 697880.53 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3228, pruned_loss=0.08484, over 3831947.28 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:43:19,824 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:43:29,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5027, 1.2455, 4.6364, 3.6695], device='cuda:1'), covar=tensor([0.2093, 0.3408, 0.0711, 0.1135], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0680, 0.1021, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 18:43:46,278 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4790, 4.3262, 4.0603, 1.9487], device='cuda:1'), covar=tensor([0.0531, 0.0712, 0.0748, 0.2017], device='cuda:1'), in_proj_covar=tensor([0.1321, 0.1219, 0.1028, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 18:43:47,817 INFO [train.py:968] (1/2) Epoch 29, batch 300, giga_loss[loss=0.1998, simple_loss=0.2789, pruned_loss=0.06032, over 28727.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.3154, pruned_loss=0.08064, over 4424946.76 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3333, pruned_loss=0.08446, over 972974.95 frames. ], giga_tot_loss[loss=0.238, simple_loss=0.3142, pruned_loss=0.08092, over 4160939.24 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:43:55,581 INFO [optim.py:369] (1/2) 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 18:44:25,115 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:44:30,405 INFO [train.py:968] (1/2) Epoch 29, batch 350, giga_loss[loss=0.1797, simple_loss=0.2642, pruned_loss=0.04754, over 29031.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3079, pruned_loss=0.07745, over 4699394.46 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3306, pruned_loss=0.08338, over 1069926.81 frames. ], giga_tot_loss[loss=0.231, simple_loss=0.3067, pruned_loss=0.07761, over 4466788.65 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:44:44,820 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:44:56,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5681, 1.6887, 1.7998, 1.3525], device='cuda:1'), covar=tensor([0.2059, 0.2846, 0.1703, 0.1873], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0722, 0.0990, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 18:45:08,116 INFO [train.py:968] (1/2) Epoch 29, batch 400, giga_loss[loss=0.2388, simple_loss=0.3069, pruned_loss=0.08535, over 28640.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.3042, pruned_loss=0.07598, over 4925756.86 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3324, pruned_loss=0.08529, over 1188750.83 frames. ], giga_tot_loss[loss=0.2264, simple_loss=0.3021, pruned_loss=0.0754, over 4716621.80 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:45:17,739 INFO [optim.py:369] (1/2) 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 18:45:44,922 INFO [train.py:968] (1/2) Epoch 29, batch 450, giga_loss[loss=0.2234, simple_loss=0.2992, pruned_loss=0.07375, over 28274.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.3038, pruned_loss=0.07576, over 5102633.80 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3328, pruned_loss=0.08547, over 1415954.24 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.3005, pruned_loss=0.07476, over 4894507.28 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:45:51,634 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7084, 4.5472, 4.3161, 2.0111], device='cuda:1'), covar=tensor([0.0642, 0.0736, 0.0923, 0.1996], device='cuda:1'), in_proj_covar=tensor([0.1314, 0.1215, 0.1024, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 18:45:59,830 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:46:26,527 INFO [train.py:968] (1/2) Epoch 29, batch 500, giga_loss[loss=0.2116, simple_loss=0.2831, pruned_loss=0.07, over 28468.00 frames. ], tot_loss[loss=0.2263, simple_loss=0.3021, pruned_loss=0.07527, over 5230527.07 frames. ], libri_tot_loss[loss=0.252, simple_loss=0.333, pruned_loss=0.08547, over 1570910.97 frames. ], giga_tot_loss[loss=0.2233, simple_loss=0.2983, pruned_loss=0.07413, over 5039946.95 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:46:35,585 INFO [optim.py:369] (1/2) 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 18:47:05,746 INFO [train.py:968] (1/2) Epoch 29, batch 550, giga_loss[loss=0.2286, simple_loss=0.3, pruned_loss=0.07863, over 28489.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2997, pruned_loss=0.07411, over 5331924.87 frames. ], libri_tot_loss[loss=0.2524, simple_loss=0.3339, pruned_loss=0.08541, over 1674673.02 frames. ], giga_tot_loss[loss=0.2207, simple_loss=0.2956, pruned_loss=0.07292, over 5168993.07 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:47:47,967 INFO [train.py:968] (1/2) Epoch 29, batch 600, giga_loss[loss=0.2602, simple_loss=0.3257, pruned_loss=0.09738, over 28532.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.299, pruned_loss=0.07388, over 5413119.45 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3348, pruned_loss=0.08562, over 1799985.72 frames. ], giga_tot_loss[loss=0.2197, simple_loss=0.2944, pruned_loss=0.07248, over 5268222.94 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:47:53,229 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8768, 1.9565, 1.7011, 2.1101], device='cuda:1'), covar=tensor([0.2820, 0.3068, 0.3389, 0.2703], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1161, 0.1427, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 18:47:55,746 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7542, 2.0681, 2.0680, 1.6697], device='cuda:1'), covar=tensor([0.3657, 0.2751, 0.2730, 0.3402], device='cuda:1'), in_proj_covar=tensor([0.2079, 0.2040, 0.1955, 0.2097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 18:47:57,374 INFO [optim.py:369] (1/2) 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 18:47:59,270 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:48:06,698 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:48:12,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:48:22,539 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 650, giga_loss[loss=0.243, simple_loss=0.3088, pruned_loss=0.08862, over 28770.00 frames. ], tot_loss[loss=0.2205, simple_loss=0.2959, pruned_loss=0.07256, over 5477439.60 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3349, pruned_loss=0.08552, over 1820204.79 frames. ], giga_tot_loss[loss=0.2174, simple_loss=0.292, pruned_loss=0.07142, over 5362299.29 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:48:37,826 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276473.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:49:07,002 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5123, 2.1217, 1.5772, 0.7848], device='cuda:1'), covar=tensor([0.6002, 0.3371, 0.4573, 0.7177], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1736, 0.1657, 0.1501], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 18:49:22,068 INFO [train.py:968] (1/2) Epoch 29, batch 700, giga_loss[loss=0.1824, simple_loss=0.2624, pruned_loss=0.05116, over 28992.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2933, pruned_loss=0.07113, over 5526749.10 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3353, pruned_loss=0.08549, over 1879985.97 frames. ], giga_tot_loss[loss=0.2148, simple_loss=0.2895, pruned_loss=0.07002, over 5431367.93 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:49:32,985 INFO [optim.py:369] (1/2) 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 18:49:39,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:49:53,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:05,293 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 750, giga_loss[loss=0.221, simple_loss=0.286, pruned_loss=0.07796, over 28919.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.292, pruned_loss=0.07043, over 5559945.95 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3359, pruned_loss=0.08546, over 1978915.41 frames. ], giga_tot_loss[loss=0.213, simple_loss=0.2877, pruned_loss=0.06916, over 5476157.35 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:50:11,240 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:14,384 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:32,668 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:34,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:38,098 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:41,147 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:48,587 INFO [train.py:968] (1/2) Epoch 29, batch 800, libri_loss[loss=0.238, simple_loss=0.3124, pruned_loss=0.08175, over 29656.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2904, pruned_loss=0.07006, over 5591128.60 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3346, pruned_loss=0.08483, over 2110964.60 frames. ], giga_tot_loss[loss=0.2116, simple_loss=0.2858, pruned_loss=0.06873, over 5516016.23 frames. ], batch size: 69, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:50:50,343 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1276622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:50:56,558 INFO [optim.py:369] (1/2) 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 18:50:56,751 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:51:24,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:51:29,110 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-14 18:51:29,327 INFO [train.py:968] (1/2) Epoch 29, batch 850, libri_loss[loss=0.2368, simple_loss=0.3123, pruned_loss=0.08072, over 29489.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2953, pruned_loss=0.07232, over 5619671.61 frames. ], libri_tot_loss[loss=0.2503, simple_loss=0.3329, pruned_loss=0.08381, over 2312698.50 frames. ], giga_tot_loss[loss=0.216, simple_loss=0.29, pruned_loss=0.07093, over 5544590.26 frames. ], batch size: 70, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:51:32,006 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 18:51:33,188 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.67 vs. limit=5.0
+2023-03-14 18:51:39,024 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:51:40,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:02,900 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:06,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:07,781 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:12,520 INFO [train.py:968] (1/2) Epoch 29, batch 900, giga_loss[loss=0.3048, simple_loss=0.3578, pruned_loss=0.1259, over 23737.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.3085, pruned_loss=0.07899, over 5634300.02 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3344, pruned_loss=0.08459, over 2399488.21 frames. ], giga_tot_loss[loss=0.2291, simple_loss=0.3032, pruned_loss=0.07746, over 5569901.63 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:52:24,693 INFO [optim.py:369] (1/2) 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 18:52:32,612 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:52:33,787 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2419, 3.0184, 1.4003, 1.3973], device='cuda:1'), covar=tensor([0.1107, 0.0348, 0.0997, 0.1494], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0573, 0.0413, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 18:52:37,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6402, 1.8715, 1.5235, 1.9155], device='cuda:1'), covar=tensor([0.2764, 0.2959, 0.3234, 0.2556], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1162, 0.1426, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 18:52:57,105 INFO [train.py:968] (1/2) Epoch 29, batch 950, giga_loss[loss=0.3557, simple_loss=0.4086, pruned_loss=0.1514, over 28849.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3211, pruned_loss=0.08564, over 5648849.21 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.334, pruned_loss=0.08441, over 2417138.77 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3172, pruned_loss=0.08457, over 5597264.47 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:53:29,726 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1276807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:53:31,642 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1276810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:53:40,502 INFO [train.py:968] (1/2) Epoch 29, batch 1000, giga_loss[loss=0.2275, simple_loss=0.316, pruned_loss=0.06952, over 28721.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.329, pruned_loss=0.089, over 5659354.31 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.334, pruned_loss=0.08455, over 2469268.90 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3257, pruned_loss=0.08816, over 5615208.10 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:53:49,625 INFO [optim.py:369] (1/2) 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 18:53:55,944 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1276839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:53:56,750 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 18:54:11,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8452, 2.0620, 2.0652, 1.6691], device='cuda:1'), covar=tensor([0.3472, 0.3099, 0.3197, 0.3335], device='cuda:1'), in_proj_covar=tensor([0.2081, 0.2042, 0.1958, 0.2097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 18:54:19,358 INFO [train.py:968] (1/2) Epoch 29, batch 1050, giga_loss[loss=0.2444, simple_loss=0.3295, pruned_loss=0.07962, over 29010.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3327, pruned_loss=0.08919, over 5665259.93 frames. ], libri_tot_loss[loss=0.2516, simple_loss=0.3342, pruned_loss=0.08452, over 2533145.22 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3301, pruned_loss=0.08864, over 5629699.04 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:55:04,335 INFO [train.py:968] (1/2) Epoch 29, batch 1100, giga_loss[loss=0.2495, simple_loss=0.3296, pruned_loss=0.08474, over 28792.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3343, pruned_loss=0.08908, over 5663870.22 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3342, pruned_loss=0.0844, over 2600261.78 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3322, pruned_loss=0.08878, over 5631476.98 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:55:04,886 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-14 18:55:11,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:55:14,780 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 1150, giga_loss[loss=0.2416, simple_loss=0.3265, pruned_loss=0.07829, over 28886.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3368, pruned_loss=0.09072, over 5663043.29 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3342, pruned_loss=0.0844, over 2600261.78 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3352, pruned_loss=0.0905, over 5637831.23 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 18:55:52,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6064, 4.3925, 1.7460, 2.0995], device='cuda:1'), covar=tensor([0.1034, 0.0240, 0.0926, 0.1240], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0571, 0.0413, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 18:56:00,262 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:56:10,997 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1276997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:56:12,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3604, 1.5250, 1.1053, 1.1276], device='cuda:1'), covar=tensor([0.1050, 0.0515, 0.1145, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0451, 0.0526, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 18:56:18,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 18:56:29,601 INFO [train.py:968] (1/2) Epoch 29, batch 1200, giga_loss[loss=0.2625, simple_loss=0.3468, pruned_loss=0.08916, over 28694.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.339, pruned_loss=0.09218, over 5674537.58 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3342, pruned_loss=0.0841, over 2664422.21 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3378, pruned_loss=0.09226, over 5651567.73 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:56:39,298 INFO [optim.py:369] (1/2) 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 18:56:55,156 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7142, 4.3170, 1.7214, 1.9867], device='cuda:1'), covar=tensor([0.0995, 0.0221, 0.0921, 0.1252], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0571, 0.0413, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 18:57:08,357 INFO [train.py:968] (1/2) Epoch 29, batch 1250, giga_loss[loss=0.2714, simple_loss=0.3425, pruned_loss=0.1001, over 28711.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3426, pruned_loss=0.09426, over 5679580.57 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3353, pruned_loss=0.08455, over 2726693.48 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3413, pruned_loss=0.09431, over 5659059.49 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:57:10,274 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:57:12,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:57:39,458 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:57:51,850 INFO [train.py:968] (1/2) Epoch 29, batch 1300, giga_loss[loss=0.3086, simple_loss=0.3864, pruned_loss=0.1154, over 28742.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.346, pruned_loss=0.09514, over 5688432.22 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3364, pruned_loss=0.08508, over 2789346.21 frames. ], giga_tot_loss[loss=0.2674, simple_loss=0.3446, pruned_loss=0.09513, over 5668466.88 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:57:58,207 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:01,025 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:58:06,256 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:08,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:22,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:28,770 INFO [train.py:968] (1/2) Epoch 29, batch 1350, giga_loss[loss=0.2716, simple_loss=0.3513, pruned_loss=0.09596, over 28204.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.3482, pruned_loss=0.096, over 5691304.54 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3367, pruned_loss=0.08507, over 2866514.17 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3473, pruned_loss=0.09625, over 5670247.05 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:58:29,624 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 18:58:44,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2757, 1.0858, 1.1404, 1.3438], device='cuda:1'), covar=tensor([0.0849, 0.0366, 0.0333, 0.1129], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 18:59:10,092 INFO [train.py:968] (1/2) Epoch 29, batch 1400, giga_loss[loss=0.2786, simple_loss=0.3624, pruned_loss=0.09745, over 29057.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3498, pruned_loss=0.09639, over 5694480.38 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3369, pruned_loss=0.0853, over 2938051.51 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3492, pruned_loss=0.09673, over 5676159.85 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:59:19,550 INFO [optim.py:369] (1/2) 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 18:59:26,613 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7303, 1.3392, 4.9556, 3.6581], device='cuda:1'), covar=tensor([0.1863, 0.3221, 0.0396, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0673, 0.1006, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 18:59:26,677 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5715, 2.5051, 2.6172, 2.2616], device='cuda:1'), covar=tensor([0.2853, 0.2920, 0.2620, 0.2864], device='cuda:1'), in_proj_covar=tensor([0.2071, 0.2031, 0.1944, 0.2088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 18:59:51,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-14 18:59:52,002 INFO [train.py:968] (1/2) Epoch 29, batch 1450, giga_loss[loss=0.2569, simple_loss=0.3447, pruned_loss=0.08457, over 28916.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3498, pruned_loss=0.09557, over 5696114.61 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3369, pruned_loss=0.08535, over 2997085.47 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3496, pruned_loss=0.09602, over 5677933.31 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 18:59:55,278 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-14 19:00:27,152 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:00:29,685 INFO [train.py:968] (1/2) Epoch 29, batch 1500, giga_loss[loss=0.2711, simple_loss=0.3513, pruned_loss=0.09549, over 28091.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3481, pruned_loss=0.09356, over 5704577.62 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3367, pruned_loss=0.08513, over 3053600.42 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3483, pruned_loss=0.09418, over 5687626.78 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:00:37,888 INFO [optim.py:369] (1/2) 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 19:00:47,570 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:01:03,006 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:01:06,510 INFO [train.py:968] (1/2) Epoch 29, batch 1550, giga_loss[loss=0.2344, simple_loss=0.3234, pruned_loss=0.07272, over 28652.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3464, pruned_loss=0.09184, over 5713982.01 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3367, pruned_loss=0.08518, over 3137853.38 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3467, pruned_loss=0.09249, over 5695389.26 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:01:21,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4625, 1.6229, 1.3650, 1.6861], device='cuda:1'), covar=tensor([0.0768, 0.0331, 0.0355, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 19:01:45,117 INFO [train.py:968] (1/2) Epoch 29, batch 1600, giga_loss[loss=0.2633, simple_loss=0.3408, pruned_loss=0.09293, over 28472.00 frames. ], tot_loss[loss=0.268, simple_loss=0.3475, pruned_loss=0.09419, over 5698836.84 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3355, pruned_loss=0.08479, over 3251147.50 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3489, pruned_loss=0.09524, over 5683438.16 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:01:54,957 INFO [optim.py:369] (1/2) 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 19:02:10,168 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-14 19:02:23,211 INFO [train.py:968] (1/2) Epoch 29, batch 1650, giga_loss[loss=0.2815, simple_loss=0.356, pruned_loss=0.1036, over 28796.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3494, pruned_loss=0.0971, over 5705309.09 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3353, pruned_loss=0.08445, over 3326925.60 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3509, pruned_loss=0.09842, over 5690256.03 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:03:04,360 INFO [train.py:968] (1/2) Epoch 29, batch 1700, giga_loss[loss=0.2623, simple_loss=0.3335, pruned_loss=0.09552, over 28279.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3507, pruned_loss=0.09946, over 5715986.39 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3358, pruned_loss=0.08467, over 3415673.45 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3522, pruned_loss=0.1009, over 5697765.64 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:03:13,741 INFO [optim.py:369] (1/2) 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 19:03:40,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3680, 3.1424, 1.5005, 1.4814], device='cuda:1'), covar=tensor([0.1022, 0.0318, 0.0905, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0571, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 19:03:44,464 INFO [train.py:968] (1/2) Epoch 29, batch 1750, giga_loss[loss=0.2687, simple_loss=0.3436, pruned_loss=0.09686, over 28608.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3495, pruned_loss=0.09962, over 5713611.55 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3353, pruned_loss=0.08442, over 3476560.91 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3512, pruned_loss=0.1013, over 5695294.23 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:03:56,529 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 29, batch 1800, giga_loss[loss=0.2433, simple_loss=0.3203, pruned_loss=0.08318, over 29031.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.347, pruned_loss=0.09885, over 5700838.48 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3361, pruned_loss=0.08487, over 3547572.69 frames. ], giga_tot_loss[loss=0.2745, simple_loss=0.3483, pruned_loss=0.1003, over 5681314.17 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:04:36,778 INFO [optim.py:369] (1/2) 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 19:05:04,914 INFO [train.py:968] (1/2) Epoch 29, batch 1850, giga_loss[loss=0.2769, simple_loss=0.3522, pruned_loss=0.1007, over 27899.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3458, pruned_loss=0.09757, over 5700580.79 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.337, pruned_loss=0.08527, over 3602700.09 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3465, pruned_loss=0.09882, over 5683149.34 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:05:21,835 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1277692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:05:44,088 INFO [train.py:968] (1/2) Epoch 29, batch 1900, giga_loss[loss=0.2606, simple_loss=0.3441, pruned_loss=0.08855, over 27924.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3443, pruned_loss=0.09615, over 5700772.29 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3364, pruned_loss=0.08502, over 3670029.74 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3455, pruned_loss=0.09762, over 5682062.72 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:05:44,288 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1277721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:05:53,842 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277731.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:05:55,808 INFO [optim.py:369] (1/2) 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 19:06:01,443 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 1950, giga_loss[loss=0.2164, simple_loss=0.2979, pruned_loss=0.06749, over 28873.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.341, pruned_loss=0.09416, over 5696991.09 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3368, pruned_loss=0.08524, over 3724430.33 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3419, pruned_loss=0.09544, over 5678370.81 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:07:03,251 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277808.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:07:08,285 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:14,505 INFO [train.py:968] (1/2) Epoch 29, batch 2000, giga_loss[loss=0.2266, simple_loss=0.3036, pruned_loss=0.07478, over 28837.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3353, pruned_loss=0.09141, over 5686780.61 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3368, pruned_loss=0.08519, over 3745722.30 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3359, pruned_loss=0.09251, over 5670783.21 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:07:26,534 INFO [optim.py:369] (1/2) 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 19:07:27,082 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-14 19:07:27,504 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:29,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:51,011 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:53,597 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:53,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:07:56,795 INFO [train.py:968] (1/2) Epoch 29, batch 2050, giga_loss[loss=0.242, simple_loss=0.3194, pruned_loss=0.0823, over 29046.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3306, pruned_loss=0.08904, over 5674217.79 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3365, pruned_loss=0.08509, over 3786217.23 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3312, pruned_loss=0.09006, over 5668895.49 frames. ], batch size: 113, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:08:06,702 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1277883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:09,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1277886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:19,750 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:34,679 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1277915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:08:38,668 INFO [train.py:968] (1/2) Epoch 29, batch 2100, giga_loss[loss=0.2781, simple_loss=0.3615, pruned_loss=0.09734, over 28899.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.3287, pruned_loss=0.0881, over 5661994.43 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3372, pruned_loss=0.08525, over 3892060.25 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3283, pruned_loss=0.089, over 5653103.04 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:08:49,836 INFO [optim.py:369] (1/2) 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 19:09:19,259 INFO [train.py:968] (1/2) Epoch 29, batch 2150, giga_loss[loss=0.2319, simple_loss=0.324, pruned_loss=0.06991, over 28878.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3295, pruned_loss=0.08777, over 5676080.33 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3366, pruned_loss=0.08496, over 3961163.03 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3293, pruned_loss=0.08875, over 5661817.24 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:09:22,010 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7752, 1.8053, 1.9776, 1.5361], device='cuda:1'), covar=tensor([0.1934, 0.2610, 0.1554, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0722, 0.0993, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 19:09:25,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1277977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:09:25,572 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6355, 2.0217, 1.8847, 1.5649], device='cuda:1'), covar=tensor([0.3439, 0.2443, 0.2062, 0.2844], device='cuda:1'), in_proj_covar=tensor([0.2075, 0.2037, 0.1949, 0.2097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 19:09:59,728 INFO [train.py:968] (1/2) Epoch 29, batch 2200, libri_loss[loss=0.2381, simple_loss=0.3326, pruned_loss=0.07185, over 29541.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3297, pruned_loss=0.08705, over 5683767.71 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3363, pruned_loss=0.08455, over 4034713.34 frames. ], giga_tot_loss[loss=0.253, simple_loss=0.3295, pruned_loss=0.08823, over 5673813.96 frames. ], batch size: 80, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:10:07,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4834, 1.7898, 1.3778, 1.3349], device='cuda:1'), covar=tensor([0.1185, 0.0599, 0.1071, 0.1208], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0449, 0.0525, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 19:10:10,722 INFO [optim.py:369] (1/2) 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 19:10:42,019 INFO [train.py:968] (1/2) Epoch 29, batch 2250, giga_loss[loss=0.2438, simple_loss=0.3199, pruned_loss=0.08388, over 28715.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3283, pruned_loss=0.08666, over 5690337.87 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3366, pruned_loss=0.08454, over 4062521.80 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3278, pruned_loss=0.08762, over 5679699.40 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:10:51,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5113, 2.1720, 1.7442, 0.7954], device='cuda:1'), covar=tensor([0.6432, 0.2928, 0.4858, 0.7332], device='cuda:1'), in_proj_covar=tensor([0.1835, 0.1719, 0.1647, 0.1490], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 19:11:09,630 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 2300, giga_loss[loss=0.2655, simple_loss=0.3331, pruned_loss=0.0989, over 29034.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3264, pruned_loss=0.08587, over 5698004.74 frames. ], libri_tot_loss[loss=0.2528, simple_loss=0.3366, pruned_loss=0.08452, over 4098258.75 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3259, pruned_loss=0.08668, over 5686617.33 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:11:33,671 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-14 19:11:34,625 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 2350, giga_loss[loss=0.2227, simple_loss=0.3011, pruned_loss=0.07212, over 28593.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3237, pruned_loss=0.08471, over 5702094.57 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.337, pruned_loss=0.08471, over 4107289.46 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3229, pruned_loss=0.08523, over 5692291.18 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:12:15,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1278183.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:12:20,740 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 2400, giga_loss[loss=0.2298, simple_loss=0.3089, pruned_loss=0.07535, over 28714.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3229, pruned_loss=0.08472, over 5694927.27 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3373, pruned_loss=0.08473, over 4141974.09 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3218, pruned_loss=0.08512, over 5692045.46 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:12:55,995 INFO [optim.py:369] (1/2) 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 19:13:07,233 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:13:09,226 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1278252.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:13:24,374 INFO [train.py:968] (1/2) Epoch 29, batch 2450, giga_loss[loss=0.251, simple_loss=0.3159, pruned_loss=0.09302, over 28978.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3207, pruned_loss=0.08379, over 5701633.92 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3372, pruned_loss=0.08448, over 4185186.21 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3195, pruned_loss=0.08426, over 5694977.00 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:13:31,325 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1278281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:13:37,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 19:13:50,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4728, 1.7449, 1.4194, 1.4072], device='cuda:1'), covar=tensor([0.2843, 0.3024, 0.3427, 0.2653], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1163, 0.1428, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 19:13:58,867 INFO [train.py:968] (1/2) Epoch 29, batch 2500, giga_loss[loss=0.2135, simple_loss=0.296, pruned_loss=0.06548, over 28921.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.3195, pruned_loss=0.08296, over 5707199.03 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3383, pruned_loss=0.08486, over 4257600.05 frames. ], giga_tot_loss[loss=0.2415, simple_loss=0.317, pruned_loss=0.08303, over 5703377.64 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:14:03,355 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278326.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:14:05,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1278329.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:14:08,281 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:10,279 INFO [optim.py:369] (1/2) 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 19:14:11,106 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1278335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:12,927 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3623, 1.4785, 1.5087, 1.3197], device='cuda:1'), covar=tensor([0.4010, 0.3302, 0.2766, 0.3626], device='cuda:1'), in_proj_covar=tensor([0.2065, 0.2027, 0.1938, 0.2091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 19:14:25,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1278352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:27,024 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 19:14:29,457 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2095, 1.2791, 3.8002, 3.1548], device='cuda:1'), covar=tensor([0.1791, 0.2898, 0.0444, 0.1028], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0669, 0.1001, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 19:14:29,992 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1278358.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:14:35,423 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1278364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:14:40,433 INFO [train.py:968] (1/2) Epoch 29, batch 2550, giga_loss[loss=0.2206, simple_loss=0.2975, pruned_loss=0.07189, over 28103.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.317, pruned_loss=0.08194, over 5716020.09 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3381, pruned_loss=0.08479, over 4264840.05 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3151, pruned_loss=0.08203, over 5713065.54 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:14:45,450 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-14 19:15:20,351 INFO [train.py:968] (1/2) Epoch 29, batch 2600, giga_loss[loss=0.2206, simple_loss=0.2962, pruned_loss=0.07253, over 28422.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3156, pruned_loss=0.08107, over 5715848.26 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3389, pruned_loss=0.08508, over 4293980.35 frames. ], giga_tot_loss[loss=0.2374, simple_loss=0.3131, pruned_loss=0.08086, over 5713278.21 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:15:20,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5699, 4.0164, 1.6029, 1.6620], device='cuda:1'), covar=tensor([0.0932, 0.0240, 0.0909, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0568, 0.0412, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 19:15:32,399 INFO [optim.py:369] (1/2) 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 19:15:43,784 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3368, 2.1173, 1.5128, 0.6340], device='cuda:1'), covar=tensor([0.5946, 0.2540, 0.4209, 0.5778], device='cuda:1'), in_proj_covar=tensor([0.1830, 0.1714, 0.1645, 0.1489], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 19:15:58,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-14 19:15:59,715 INFO [train.py:968] (1/2) Epoch 29, batch 2650, giga_loss[loss=0.2083, simple_loss=0.2911, pruned_loss=0.06279, over 28643.00 frames. ], tot_loss[loss=0.2388, simple_loss=0.3154, pruned_loss=0.08115, over 5720345.00 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3391, pruned_loss=0.08505, over 4330762.00 frames. ], giga_tot_loss[loss=0.2373, simple_loss=0.3127, pruned_loss=0.08091, over 5716740.34 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:16:08,680 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6912, 1.9884, 1.9302, 1.6554], device='cuda:1'), covar=tensor([0.2784, 0.2197, 0.1992, 0.2383], device='cuda:1'), in_proj_covar=tensor([0.2060, 0.2020, 0.1932, 0.2087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 19:16:19,244 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1278495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:16:21,207 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 2700, giga_loss[loss=0.252, simple_loss=0.3277, pruned_loss=0.08818, over 28905.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3178, pruned_loss=0.08268, over 5726471.82 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.339, pruned_loss=0.08485, over 4391335.88 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3149, pruned_loss=0.08251, over 5718139.02 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:16:43,807 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:17:17,530 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7882, 4.9899, 2.0087, 2.1450], device='cuda:1'), covar=tensor([0.0987, 0.0258, 0.0834, 0.1203], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0569, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 19:17:21,709 INFO [train.py:968] (1/2) Epoch 29, batch 2750, giga_loss[loss=0.3007, simple_loss=0.368, pruned_loss=0.1167, over 28590.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.323, pruned_loss=0.08564, over 5728267.73 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3397, pruned_loss=0.08514, over 4436424.11 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3196, pruned_loss=0.08527, over 5716709.15 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:18:04,356 INFO [train.py:968] (1/2) Epoch 29, batch 2800, libri_loss[loss=0.2391, simple_loss=0.3315, pruned_loss=0.0733, over 29250.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3286, pruned_loss=0.08866, over 5722457.21 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3399, pruned_loss=0.08501, over 4487877.39 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3252, pruned_loss=0.08856, over 5712102.01 frames. ], batch size: 94, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:18:17,461 INFO [optim.py:369] (1/2) 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 19:18:50,486 INFO [train.py:968] (1/2) Epoch 29, batch 2850, giga_loss[loss=0.3673, simple_loss=0.4097, pruned_loss=0.1624, over 26622.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.337, pruned_loss=0.09478, over 5687600.65 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3399, pruned_loss=0.0852, over 4496916.09 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3342, pruned_loss=0.09466, over 5692636.19 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:19:14,328 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:19:24,219 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3672, 1.1725, 4.1452, 3.1758], device='cuda:1'), covar=tensor([0.1635, 0.2841, 0.0475, 0.1199], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0671, 0.1001, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 19:19:30,815 INFO [train.py:968] (1/2) Epoch 29, batch 2900, giga_loss[loss=0.2634, simple_loss=0.3498, pruned_loss=0.08849, over 28909.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3404, pruned_loss=0.09568, over 5696954.20 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3397, pruned_loss=0.08506, over 4561686.67 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3382, pruned_loss=0.09623, over 5699876.24 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:19:41,629 INFO [optim.py:369] (1/2) 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 19:20:17,013 INFO [train.py:968] (1/2) Epoch 29, batch 2950, giga_loss[loss=0.2907, simple_loss=0.3706, pruned_loss=0.1054, over 28933.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3441, pruned_loss=0.09659, over 5688007.35 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3399, pruned_loss=0.0852, over 4584767.84 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3423, pruned_loss=0.09721, over 5700670.83 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:20:39,167 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.73 vs. limit=5.0
+2023-03-14 19:20:57,943 INFO [train.py:968] (1/2) Epoch 29, batch 3000, giga_loss[loss=0.3113, simple_loss=0.3757, pruned_loss=0.1235, over 28522.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3497, pruned_loss=0.1002, over 5680407.00 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3399, pruned_loss=0.0854, over 4617896.37 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3484, pruned_loss=0.1009, over 5695056.07 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:20:57,943 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 19:21:07,153 INFO [train.py:1012] (1/2) Epoch 29, validation: loss=0.2057, simple_loss=0.3132, pruned_loss=0.04906, over 944034.00 frames. 
+2023-03-14 19:21:07,153 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 19:21:09,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:21:20,077 INFO [optim.py:369] (1/2) 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 19:21:47,350 INFO [train.py:968] (1/2) Epoch 29, batch 3050, giga_loss[loss=0.2369, simple_loss=0.3289, pruned_loss=0.07243, over 28961.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3523, pruned_loss=0.1013, over 5661583.26 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3406, pruned_loss=0.08583, over 4642829.34 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.3511, pruned_loss=0.102, over 5678617.17 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:22:16,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:22:17,153 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:22:29,987 INFO [train.py:968] (1/2) Epoch 29, batch 3100, giga_loss[loss=0.2409, simple_loss=0.3223, pruned_loss=0.07971, over 28930.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3478, pruned_loss=0.09803, over 5675448.35 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3401, pruned_loss=0.08557, over 4661614.97 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3474, pruned_loss=0.09894, over 5685793.03 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:22:42,813 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 3150, giga_loss[loss=0.2284, simple_loss=0.3154, pruned_loss=0.07066, over 28916.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3454, pruned_loss=0.09568, over 5688782.44 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3397, pruned_loss=0.08549, over 4686558.06 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3454, pruned_loss=0.09667, over 5692872.38 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 19:23:17,948 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1278977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:23:55,486 INFO [train.py:968] (1/2) Epoch 29, batch 3200, giga_loss[loss=0.2712, simple_loss=0.3535, pruned_loss=0.09446, over 29045.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3443, pruned_loss=0.09474, over 5696495.26 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3391, pruned_loss=0.08519, over 4710820.05 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3449, pruned_loss=0.09596, over 5695798.77 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:24:13,045 INFO [optim.py:369] (1/2) 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 19:24:39,802 INFO [train.py:968] (1/2) Epoch 29, batch 3250, giga_loss[loss=0.2917, simple_loss=0.3674, pruned_loss=0.108, over 28902.00 frames. ], tot_loss[loss=0.2686, simple_loss=0.3463, pruned_loss=0.09541, over 5698668.96 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3397, pruned_loss=0.08562, over 4736698.66 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3465, pruned_loss=0.09627, over 5696907.32 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:24:43,135 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 3300, giga_loss[loss=0.2298, simple_loss=0.3173, pruned_loss=0.07114, over 28542.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3481, pruned_loss=0.09632, over 5704345.91 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3399, pruned_loss=0.08568, over 4748249.27 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3482, pruned_loss=0.09706, over 5701143.29 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:25:35,011 INFO [optim.py:369] (1/2) 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 19:26:05,131 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 19:26:05,283 INFO [train.py:968] (1/2) Epoch 29, batch 3350, giga_loss[loss=0.2849, simple_loss=0.3473, pruned_loss=0.1113, over 28907.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3489, pruned_loss=0.09711, over 5706467.62 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3403, pruned_loss=0.08577, over 4770647.57 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3489, pruned_loss=0.09783, over 5700882.85 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:26:27,780 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:26:42,277 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:26:43,961 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 3400, libri_loss[loss=0.2646, simple_loss=0.3484, pruned_loss=0.0904, over 29532.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3509, pruned_loss=0.09891, over 5711009.67 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3406, pruned_loss=0.08582, over 4798737.27 frames. ], giga_tot_loss[loss=0.2752, simple_loss=0.3509, pruned_loss=0.09977, over 5701806.82 frames. ], batch size: 80, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:26:57,965 INFO [optim.py:369] (1/2) 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 19:27:07,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-14 19:27:08,632 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:27:28,018 INFO [train.py:968] (1/2) Epoch 29, batch 3450, libri_loss[loss=0.2976, simple_loss=0.3815, pruned_loss=0.1069, over 27749.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3512, pruned_loss=0.09926, over 5713735.21 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3407, pruned_loss=0.08589, over 4815502.83 frames. ], giga_tot_loss[loss=0.2758, simple_loss=0.3513, pruned_loss=0.1002, over 5711469.33 frames. ], batch size: 116, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:27:28,898 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279272.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:27:34,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:27:35,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:27:36,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2843, 1.4123, 1.2976, 1.4303], device='cuda:1'), covar=tensor([0.0809, 0.0368, 0.0348, 0.0894], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 19:28:07,939 INFO [train.py:968] (1/2) Epoch 29, batch 3500, giga_loss[loss=0.2469, simple_loss=0.3299, pruned_loss=0.08196, over 28444.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3504, pruned_loss=0.09848, over 5717869.48 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3406, pruned_loss=0.08583, over 4826098.26 frames. ], giga_tot_loss[loss=0.2747, simple_loss=0.3506, pruned_loss=0.09935, over 5714687.24 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:28:13,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2862, 1.3004, 3.4287, 3.0624], device='cuda:1'), covar=tensor([0.1510, 0.2763, 0.0469, 0.2071], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0671, 0.1001, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 19:28:22,411 INFO [optim.py:369] (1/2) 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 19:28:26,703 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:28:28,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:28:34,786 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 3550, giga_loss[loss=0.2828, simple_loss=0.3393, pruned_loss=0.1131, over 23891.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3504, pruned_loss=0.09798, over 5711409.96 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3406, pruned_loss=0.08582, over 4831495.77 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3506, pruned_loss=0.09873, over 5708001.30 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:28:50,919 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:12,454 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:30,840 INFO [train.py:968] (1/2) Epoch 29, batch 3600, giga_loss[loss=0.2456, simple_loss=0.3325, pruned_loss=0.0793, over 28914.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3507, pruned_loss=0.09719, over 5719469.95 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3403, pruned_loss=0.08558, over 4851478.49 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3513, pruned_loss=0.09818, over 5714418.83 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:29:31,924 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:32,666 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:35,427 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:36,042 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279426.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:42,836 INFO [optim.py:369] (1/2) 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 19:29:49,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3500, 2.9804, 1.5479, 1.4730], device='cuda:1'), covar=tensor([0.1035, 0.0314, 0.0930, 0.1403], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0566, 0.0410, 0.0447], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 19:29:55,474 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279453.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:29:55,998 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:56,711 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:29:58,384 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-14 19:30:00,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2941, 4.1379, 3.9053, 1.8551], device='cuda:1'), covar=tensor([0.0629, 0.0772, 0.0826, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.1287, 0.1192, 0.1003, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 19:30:01,584 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4334, 1.6875, 1.6494, 1.5326], device='cuda:1'), covar=tensor([0.2392, 0.2442, 0.2604, 0.2349], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0762, 0.0735, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 19:30:08,504 INFO [train.py:968] (1/2) Epoch 29, batch 3650, giga_loss[loss=0.2785, simple_loss=0.3559, pruned_loss=0.1005, over 28584.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.349, pruned_loss=0.09621, over 5720907.68 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.3396, pruned_loss=0.08521, over 4876575.39 frames. ], giga_tot_loss[loss=0.2726, simple_loss=0.3502, pruned_loss=0.09755, over 5713545.25 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:30:27,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:29,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:46,019 INFO [train.py:968] (1/2) Epoch 29, batch 3700, giga_loss[loss=0.2747, simple_loss=0.3492, pruned_loss=0.1002, over 28708.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3475, pruned_loss=0.09578, over 5725356.14 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3402, pruned_loss=0.08564, over 4899732.73 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3481, pruned_loss=0.09675, over 5717552.15 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:30:51,390 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:53,561 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:30:55,800 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-14 19:30:57,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1599, 1.4261, 1.4100, 1.1033], device='cuda:1'), covar=tensor([0.1298, 0.2098, 0.1093, 0.1413], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0721, 0.0990, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 19:30:59,843 INFO [optim.py:369] (1/2) 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 19:31:23,501 INFO [train.py:968] (1/2) Epoch 29, batch 3750, giga_loss[loss=0.233, simple_loss=0.3182, pruned_loss=0.07388, over 28782.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3455, pruned_loss=0.09507, over 5724100.59 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3395, pruned_loss=0.08518, over 4919374.50 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3467, pruned_loss=0.0964, over 5714790.27 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:31:43,537 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1279600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:32:03,292 INFO [train.py:968] (1/2) Epoch 29, batch 3800, giga_loss[loss=0.2392, simple_loss=0.3233, pruned_loss=0.07751, over 28481.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3449, pruned_loss=0.09467, over 5731999.92 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3397, pruned_loss=0.08531, over 4949083.42 frames. ], giga_tot_loss[loss=0.2689, simple_loss=0.3459, pruned_loss=0.09592, over 5722750.74 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:32:16,096 INFO [optim.py:369] (1/2) 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 19:32:22,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279647.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:32:42,254 INFO [train.py:968] (1/2) Epoch 29, batch 3850, giga_loss[loss=0.2747, simple_loss=0.3531, pruned_loss=0.09809, over 28786.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.345, pruned_loss=0.09501, over 5734166.91 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3394, pruned_loss=0.08504, over 4982032.75 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3462, pruned_loss=0.09656, over 5721187.91 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:33:21,443 INFO [train.py:968] (1/2) Epoch 29, batch 3900, giga_loss[loss=0.2708, simple_loss=0.3503, pruned_loss=0.09567, over 28642.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3454, pruned_loss=0.09479, over 5734831.62 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3392, pruned_loss=0.08493, over 5009531.76 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3466, pruned_loss=0.09636, over 5719909.18 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:33:34,882 INFO [optim.py:369] (1/2) 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 19:33:59,293 INFO [train.py:968] (1/2) Epoch 29, batch 3950, giga_loss[loss=0.2735, simple_loss=0.3458, pruned_loss=0.1006, over 28976.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3453, pruned_loss=0.09437, over 5727573.84 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3392, pruned_loss=0.08492, over 5032469.34 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3465, pruned_loss=0.09591, over 5713998.72 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:34:02,734 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:34:16,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279790.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:34:18,178 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279793.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:34:38,498 INFO [train.py:968] (1/2) Epoch 29, batch 4000, libri_loss[loss=0.205, simple_loss=0.2891, pruned_loss=0.06049, over 29384.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3441, pruned_loss=0.09357, over 5733384.73 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3389, pruned_loss=0.08493, over 5073473.15 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3456, pruned_loss=0.09525, over 5715631.05 frames. ], batch size: 67, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:34:39,436 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1279822.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:34:43,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279828.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:34:50,744 INFO [optim.py:369] (1/2) 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 19:34:59,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.0320, 4.8655, 4.5931, 2.2865], device='cuda:1'), covar=tensor([0.0493, 0.0648, 0.0740, 0.1983], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.1192, 0.1005, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 19:35:15,758 INFO [train.py:968] (1/2) Epoch 29, batch 4050, giga_loss[loss=0.2674, simple_loss=0.3417, pruned_loss=0.09657, over 28419.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3428, pruned_loss=0.09371, over 5729409.16 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3384, pruned_loss=0.08481, over 5093216.22 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3445, pruned_loss=0.09533, over 5712368.21 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:35:26,199 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-14 19:35:39,253 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:35:49,834 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:35:52,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:35:53,252 INFO [train.py:968] (1/2) Epoch 29, batch 4100, giga_loss[loss=0.2775, simple_loss=0.3303, pruned_loss=0.1123, over 23236.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3401, pruned_loss=0.0925, over 5721066.30 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3382, pruned_loss=0.08464, over 5107600.75 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3417, pruned_loss=0.09407, over 5706133.75 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:36:04,781 INFO [optim.py:369] (1/2) 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 19:36:12,319 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 4150, libri_loss[loss=0.278, simple_loss=0.3485, pruned_loss=0.1038, over 29547.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3375, pruned_loss=0.09132, over 5715059.85 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3381, pruned_loss=0.08476, over 5121502.15 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3388, pruned_loss=0.09265, over 5707171.93 frames. ], batch size: 79, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:36:27,838 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1279971.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 19:36:30,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1279974.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:36:30,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-14 19:36:30,893 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1279975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:36:44,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4605, 1.7794, 1.6768, 1.6077], device='cuda:1'), covar=tensor([0.2339, 0.2196, 0.2573, 0.2261], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0762, 0.0735, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 19:36:52,356 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280003.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 19:37:06,570 INFO [train.py:968] (1/2) Epoch 29, batch 4200, giga_loss[loss=0.2341, simple_loss=0.3186, pruned_loss=0.0748, over 28854.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3375, pruned_loss=0.09195, over 5706856.92 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3383, pruned_loss=0.085, over 5129940.08 frames. ], giga_tot_loss[loss=0.2622, simple_loss=0.3385, pruned_loss=0.09298, over 5705778.72 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:37:19,627 INFO [optim.py:369] (1/2) 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 19:37:25,578 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:37:27,492 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:37:42,096 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.9215, 1.5638, 5.2610, 3.8310], device='cuda:1'), covar=tensor([0.1511, 0.2688, 0.0411, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0802, 0.0669, 0.0997, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 19:37:43,132 INFO [train.py:968] (1/2) Epoch 29, batch 4250, giga_loss[loss=0.2754, simple_loss=0.3497, pruned_loss=0.1006, over 28727.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3376, pruned_loss=0.0928, over 5705043.26 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3384, pruned_loss=0.08512, over 5144182.57 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3382, pruned_loss=0.09367, over 5701462.37 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:37:48,993 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:38:13,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-14 19:38:14,236 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-14 19:38:18,928 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:38:20,605 INFO [train.py:968] (1/2) Epoch 29, batch 4300, libri_loss[loss=0.2743, simple_loss=0.3517, pruned_loss=0.09847, over 29515.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3348, pruned_loss=0.09153, over 5706746.99 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3385, pruned_loss=0.08528, over 5155511.94 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3352, pruned_loss=0.09217, over 5701046.25 frames. ], batch size: 84, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:38:22,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:38:36,340 INFO [optim.py:369] (1/2) 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 19:38:44,467 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:39:00,111 INFO [train.py:968] (1/2) Epoch 29, batch 4350, libri_loss[loss=0.2633, simple_loss=0.3437, pruned_loss=0.09145, over 29529.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3331, pruned_loss=0.09091, over 5717270.73 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3387, pruned_loss=0.08537, over 5187167.64 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3329, pruned_loss=0.09159, over 5705832.51 frames. ], batch size: 81, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:39:15,517 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:39:36,556 INFO [train.py:968] (1/2) Epoch 29, batch 4400, giga_loss[loss=0.2404, simple_loss=0.3175, pruned_loss=0.08168, over 28870.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.331, pruned_loss=0.0903, over 5717829.65 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3385, pruned_loss=0.08541, over 5208081.49 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3309, pruned_loss=0.09098, over 5703797.14 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:39:42,230 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:40:16,264 INFO [train.py:968] (1/2) Epoch 29, batch 4450, libri_loss[loss=0.2903, simple_loss=0.3715, pruned_loss=0.1045, over 26200.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3294, pruned_loss=0.08927, over 5719587.87 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3384, pruned_loss=0.08535, over 5228909.91 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3292, pruned_loss=0.08999, over 5705783.42 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:40:55,558 INFO [train.py:968] (1/2) Epoch 29, batch 4500, libri_loss[loss=0.2776, simple_loss=0.355, pruned_loss=0.1001, over 29545.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3317, pruned_loss=0.08988, over 5724646.74 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3384, pruned_loss=0.08537, over 5255523.81 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3311, pruned_loss=0.0906, over 5706820.09 frames. ], batch size: 82, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:41:11,060 INFO [optim.py:369] (1/2) 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 19:41:35,748 INFO [train.py:968] (1/2) Epoch 29, batch 4550, giga_loss[loss=0.2479, simple_loss=0.338, pruned_loss=0.07893, over 28632.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3352, pruned_loss=0.0915, over 5717852.34 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.339, pruned_loss=0.08577, over 5271843.83 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3342, pruned_loss=0.09188, over 5699131.27 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:42:01,430 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:42:16,959 INFO [train.py:968] (1/2) Epoch 29, batch 4600, giga_loss[loss=0.3197, simple_loss=0.3788, pruned_loss=0.1303, over 26609.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.0928, over 5706856.28 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.3393, pruned_loss=0.08595, over 5271256.23 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3375, pruned_loss=0.09307, over 5699967.64 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:42:23,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2436, 4.0633, 3.8696, 1.8023], device='cuda:1'), covar=tensor([0.0581, 0.0720, 0.0716, 0.2155], device='cuda:1'), in_proj_covar=tensor([0.1288, 0.1190, 0.1002, 0.0746], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 19:42:31,752 INFO [optim.py:369] (1/2) 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 19:42:57,977 INFO [train.py:968] (1/2) Epoch 29, batch 4650, giga_loss[loss=0.3508, simple_loss=0.3972, pruned_loss=0.1522, over 26680.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3401, pruned_loss=0.09313, over 5699812.07 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3391, pruned_loss=0.08601, over 5297844.49 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3395, pruned_loss=0.09368, over 5690932.05 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:43:26,648 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 4700, giga_loss[loss=0.2205, simple_loss=0.2988, pruned_loss=0.07112, over 28558.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3391, pruned_loss=0.09185, over 5700548.82 frames. ], libri_tot_loss[loss=0.2556, simple_loss=0.339, pruned_loss=0.08612, over 5314208.64 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3386, pruned_loss=0.09241, over 5690945.84 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:43:54,238 INFO [optim.py:369] (1/2) 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 19:44:13,183 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-14 19:44:14,215 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:44:16,295 INFO [train.py:968] (1/2) Epoch 29, batch 4750, giga_loss[loss=0.2605, simple_loss=0.3379, pruned_loss=0.09159, over 28922.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.3386, pruned_loss=0.09155, over 5709136.39 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3393, pruned_loss=0.08618, over 5327920.73 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.338, pruned_loss=0.09211, over 5699423.33 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:44:39,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:44:54,261 INFO [train.py:968] (1/2) Epoch 29, batch 4800, giga_loss[loss=0.33, simple_loss=0.3849, pruned_loss=0.1375, over 23901.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3403, pruned_loss=0.09319, over 5704283.49 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3397, pruned_loss=0.08637, over 5344431.82 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3394, pruned_loss=0.09364, over 5691544.30 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:45:08,231 INFO [optim.py:369] (1/2) 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 19:45:23,835 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2276, 1.7102, 1.2435, 0.6714], device='cuda:1'), covar=tensor([0.5147, 0.2624, 0.3214, 0.6814], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1723, 0.1659, 0.1502], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 19:45:33,037 INFO [train.py:968] (1/2) Epoch 29, batch 4850, giga_loss[loss=0.2503, simple_loss=0.3296, pruned_loss=0.08546, over 29084.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3408, pruned_loss=0.09346, over 5707929.12 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3397, pruned_loss=0.08633, over 5354027.17 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3401, pruned_loss=0.09397, over 5695560.19 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:46:04,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:07,638 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:09,587 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280714.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:14,597 INFO [train.py:968] (1/2) Epoch 29, batch 4900, giga_loss[loss=0.2786, simple_loss=0.359, pruned_loss=0.09906, over 28990.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.344, pruned_loss=0.09524, over 5702149.74 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3401, pruned_loss=0.08653, over 5361496.16 frames. ], giga_tot_loss[loss=0.2671, simple_loss=0.3431, pruned_loss=0.09558, over 5690410.04 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:46:29,224 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7238, 1.8478, 1.6657, 1.5359], device='cuda:1'), covar=tensor([0.2107, 0.2747, 0.2573, 0.2804], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0759, 0.0732, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 19:46:29,476 INFO [optim.py:369] (1/2) 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 19:46:31,124 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:31,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:33,653 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1280747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:36,580 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1280750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:41,059 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3439, 2.0179, 1.4422, 0.6137], device='cuda:1'), covar=tensor([0.4738, 0.2662, 0.3824, 0.5932], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1724, 0.1662, 0.1501], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 19:46:51,823 INFO [train.py:968] (1/2) Epoch 29, batch 4950, giga_loss[loss=0.2668, simple_loss=0.3467, pruned_loss=0.09346, over 29041.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3473, pruned_loss=0.09654, over 5719165.85 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3408, pruned_loss=0.08689, over 5384978.76 frames. ], giga_tot_loss[loss=0.27, simple_loss=0.3462, pruned_loss=0.09687, over 5702117.31 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:46:55,876 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:57,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280777.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:46:57,573 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6103, 2.2883, 1.6027, 0.7966], device='cuda:1'), covar=tensor([0.7300, 0.3539, 0.5205, 0.7782], device='cuda:1'), in_proj_covar=tensor([0.1841, 0.1725, 0.1663, 0.1502], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 19:47:29,490 INFO [train.py:968] (1/2) Epoch 29, batch 5000, libri_loss[loss=0.2396, simple_loss=0.3227, pruned_loss=0.07829, over 29576.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3488, pruned_loss=0.09736, over 5714274.04 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3407, pruned_loss=0.08686, over 5395535.85 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3484, pruned_loss=0.09804, over 5702586.45 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:47:44,406 INFO [optim.py:369] (1/2) 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 19:47:49,492 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 19:48:06,150 INFO [train.py:968] (1/2) Epoch 29, batch 5050, giga_loss[loss=0.2986, simple_loss=0.3658, pruned_loss=0.1157, over 29086.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3501, pruned_loss=0.09797, over 5717899.24 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.341, pruned_loss=0.08727, over 5404382.36 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3496, pruned_loss=0.09842, over 5707783.92 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:48:17,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1280882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:48:44,480 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1280920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:48:44,886 INFO [train.py:968] (1/2) Epoch 29, batch 5100, giga_loss[loss=0.2306, simple_loss=0.3117, pruned_loss=0.07479, over 28372.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3501, pruned_loss=0.09804, over 5726890.39 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3415, pruned_loss=0.0876, over 5414492.27 frames. ], giga_tot_loss[loss=0.2731, simple_loss=0.3496, pruned_loss=0.09832, over 5715240.86 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:48:46,309 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:49:08,870 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1280952.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:49:23,127 INFO [train.py:968] (1/2) Epoch 29, batch 5150, libri_loss[loss=0.2772, simple_loss=0.3675, pruned_loss=0.09344, over 26154.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3487, pruned_loss=0.09745, over 5717249.11 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3413, pruned_loss=0.08753, over 5413107.67 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3486, pruned_loss=0.09796, over 5716944.99 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:49:38,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2972, 1.0429, 3.9547, 3.2722], device='cuda:1'), covar=tensor([0.1619, 0.2991, 0.0491, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0801, 0.0669, 0.1001, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 19:49:57,559 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2616, 0.8079, 0.8972, 1.4951], device='cuda:1'), covar=tensor([0.0765, 0.0381, 0.0385, 0.0892], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 19:50:02,707 INFO [train.py:968] (1/2) Epoch 29, batch 5200, giga_loss[loss=0.2581, simple_loss=0.3279, pruned_loss=0.09418, over 28774.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3466, pruned_loss=0.09631, over 5719835.49 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3417, pruned_loss=0.08777, over 5427434.97 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3463, pruned_loss=0.09676, over 5714787.02 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:50:06,335 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1281025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:07,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4290, 1.7078, 1.3851, 1.2699], device='cuda:1'), covar=tensor([0.2682, 0.2817, 0.3168, 0.2624], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1161, 0.1422, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 19:50:08,482 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1281028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:19,308 INFO [optim.py:369] (1/2) 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 19:50:20,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5890, 1.8842, 1.5143, 1.7648], device='cuda:1'), covar=tensor([0.2864, 0.2974, 0.3329, 0.2577], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1162, 0.1423, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 19:50:31,053 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281057.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:40,815 INFO [train.py:968] (1/2) Epoch 29, batch 5250, giga_loss[loss=0.257, simple_loss=0.3375, pruned_loss=0.08823, over 28630.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3438, pruned_loss=0.09517, over 5723966.48 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.342, pruned_loss=0.08801, over 5437618.62 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3434, pruned_loss=0.09549, over 5717292.06 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:50:51,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1281083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:50:51,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-14 19:51:20,990 INFO [train.py:968] (1/2) Epoch 29, batch 5300, giga_loss[loss=0.2712, simple_loss=0.3465, pruned_loss=0.09792, over 28528.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3425, pruned_loss=0.09436, over 5721407.55 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3419, pruned_loss=0.08798, over 5440051.52 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3423, pruned_loss=0.09466, over 5715203.18 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:51:24,512 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 19:52:01,129 INFO [train.py:968] (1/2) Epoch 29, batch 5350, giga_loss[loss=0.2901, simple_loss=0.369, pruned_loss=0.1056, over 28731.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3434, pruned_loss=0.09362, over 5715849.01 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3419, pruned_loss=0.08799, over 5451517.91 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3432, pruned_loss=0.09399, over 5706395.18 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:52:39,159 INFO [train.py:968] (1/2) Epoch 29, batch 5400, giga_loss[loss=0.2774, simple_loss=0.3603, pruned_loss=0.09724, over 28691.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3433, pruned_loss=0.0928, over 5717928.57 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3424, pruned_loss=0.0882, over 5469321.79 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3428, pruned_loss=0.09311, over 5703844.72 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:52:45,001 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1281226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:52:47,023 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1281229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:52:57,639 INFO [optim.py:369] (1/2) 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 19:53:00,087 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.92 vs. limit=5.0
+2023-03-14 19:53:04,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4722, 1.6881, 1.6980, 1.2392], device='cuda:1'), covar=tensor([0.1781, 0.2838, 0.1597, 0.1950], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0717, 0.0986, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 19:53:08,901 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:53:15,362 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8655, 1.9042, 1.5750, 1.4022], device='cuda:1'), covar=tensor([0.0934, 0.0649, 0.0954, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0411, 0.0446, 0.0522, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 19:53:17,087 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 5450, giga_loss[loss=0.2652, simple_loss=0.3388, pruned_loss=0.0958, over 28931.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3424, pruned_loss=0.09305, over 5716317.69 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3424, pruned_loss=0.08818, over 5476314.93 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3419, pruned_loss=0.0934, over 5704201.45 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:53:19,541 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1281272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:53:36,621 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5986, 1.6137, 1.7748, 1.4206], device='cuda:1'), covar=tensor([0.1410, 0.2062, 0.1210, 0.1467], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0717, 0.0985, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 19:53:44,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:54:02,404 INFO [train.py:968] (1/2) Epoch 29, batch 5500, giga_loss[loss=0.258, simple_loss=0.3283, pruned_loss=0.09383, over 28545.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3417, pruned_loss=0.09419, over 5714253.86 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3423, pruned_loss=0.08817, over 5480423.00 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3414, pruned_loss=0.09453, over 5703157.94 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:54:18,618 INFO [optim.py:369] (1/2) 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 19:54:24,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2026, 0.8482, 0.9086, 1.4154], device='cuda:1'), covar=tensor([0.0754, 0.0395, 0.0382, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0121, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 19:54:41,198 INFO [train.py:968] (1/2) Epoch 29, batch 5550, giga_loss[loss=0.2427, simple_loss=0.3204, pruned_loss=0.08253, over 29053.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3404, pruned_loss=0.09487, over 5715458.93 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3422, pruned_loss=0.08823, over 5490008.29 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3403, pruned_loss=0.09524, over 5703121.97 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:55:18,042 INFO [train.py:968] (1/2) Epoch 29, batch 5600, giga_loss[loss=0.2781, simple_loss=0.3485, pruned_loss=0.1038, over 28862.00 frames. ], tot_loss[loss=0.2637, simple_loss=0.3389, pruned_loss=0.09421, over 5720034.25 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3422, pruned_loss=0.08802, over 5510536.65 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3387, pruned_loss=0.09504, over 5702441.66 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:55:33,937 INFO [optim.py:369] (1/2) 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 19:56:00,815 INFO [train.py:968] (1/2) Epoch 29, batch 5650, giga_loss[loss=0.2561, simple_loss=0.3374, pruned_loss=0.08737, over 28794.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3377, pruned_loss=0.09393, over 5721898.84 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3423, pruned_loss=0.08806, over 5518449.87 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.3374, pruned_loss=0.09467, over 5704543.20 frames. ], batch size: 243, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:56:39,670 INFO [train.py:968] (1/2) Epoch 29, batch 5700, libri_loss[loss=0.26, simple_loss=0.3403, pruned_loss=0.08981, over 29562.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3346, pruned_loss=0.09231, over 5727945.08 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3417, pruned_loss=0.0879, over 5532606.51 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3346, pruned_loss=0.09324, over 5708194.70 frames. ], batch size: 74, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 19:56:58,153 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 5750, giga_loss[loss=0.2096, simple_loss=0.284, pruned_loss=0.06762, over 28607.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3287, pruned_loss=0.08916, over 5730417.37 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3416, pruned_loss=0.08787, over 5534532.06 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3287, pruned_loss=0.08994, over 5715704.13 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:57:58,015 INFO [train.py:968] (1/2) Epoch 29, batch 5800, giga_loss[loss=0.264, simple_loss=0.3356, pruned_loss=0.09618, over 28843.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3268, pruned_loss=0.08816, over 5731065.98 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3419, pruned_loss=0.08801, over 5544763.60 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.326, pruned_loss=0.08866, over 5715999.28 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:58:17,389 INFO [optim.py:369] (1/2) 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 19:58:20,319 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1281647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:58:36,546 INFO [train.py:968] (1/2) Epoch 29, batch 5850, libri_loss[loss=0.2648, simple_loss=0.3479, pruned_loss=0.09081, over 25813.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3282, pruned_loss=0.08891, over 5726712.75 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3417, pruned_loss=0.08786, over 5546488.66 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3276, pruned_loss=0.08946, over 5716016.52 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:58:41,831 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1281677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 19:58:55,166 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4431, 1.4868, 1.2329, 1.1336], device='cuda:1'), covar=tensor([0.0934, 0.0566, 0.1027, 0.1151], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0450, 0.0525, 0.0460], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 19:59:09,825 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-14 19:59:13,936 INFO [train.py:968] (1/2) Epoch 29, batch 5900, giga_loss[loss=0.2888, simple_loss=0.3724, pruned_loss=0.1025, over 28719.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.332, pruned_loss=0.09055, over 5722148.42 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3417, pruned_loss=0.08805, over 5548023.49 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3312, pruned_loss=0.09089, over 5717326.06 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:59:30,426 INFO [optim.py:369] (1/2) 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 19:59:52,883 INFO [train.py:968] (1/2) Epoch 29, batch 5950, libri_loss[loss=0.2695, simple_loss=0.3505, pruned_loss=0.09426, over 28587.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3348, pruned_loss=0.09138, over 5718919.03 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3416, pruned_loss=0.08808, over 5549952.79 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.334, pruned_loss=0.09171, over 5718678.18 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 19:59:56,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6812, 2.2177, 1.3984, 0.9506], device='cuda:1'), covar=tensor([0.8071, 0.4037, 0.3716, 0.7356], device='cuda:1'), in_proj_covar=tensor([0.1845, 0.1726, 0.1663, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 20:00:07,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1281790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:00:09,885 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1281793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:00:32,738 INFO [train.py:968] (1/2) Epoch 29, batch 6000, giga_loss[loss=0.2926, simple_loss=0.3689, pruned_loss=0.1082, over 28415.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3383, pruned_loss=0.09295, over 5716777.46 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3416, pruned_loss=0.08806, over 5565271.68 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3374, pruned_loss=0.09341, over 5709329.67 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:00:32,739 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 20:00:40,809 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 20:00:41,685 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1281822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:01:00,582 INFO [optim.py:369] (1/2) 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 20:01:21,532 INFO [train.py:968] (1/2) Epoch 29, batch 6050, giga_loss[loss=0.2466, simple_loss=0.3332, pruned_loss=0.07997, over 29002.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.3402, pruned_loss=0.09374, over 5719831.42 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3411, pruned_loss=0.08777, over 5572163.46 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3399, pruned_loss=0.09449, over 5711149.00 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:02:06,309 INFO [train.py:968] (1/2) Epoch 29, batch 6100, giga_loss[loss=0.3199, simple_loss=0.3822, pruned_loss=0.1288, over 28223.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3434, pruned_loss=0.09665, over 5715065.09 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3407, pruned_loss=0.08762, over 5576954.04 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3435, pruned_loss=0.09751, over 5705625.45 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:02:25,382 INFO [optim.py:369] (1/2) 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 20:02:43,946 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.80 vs. limit=5.0
+2023-03-14 20:02:49,703 INFO [train.py:968] (1/2) Epoch 29, batch 6150, giga_loss[loss=0.3017, simple_loss=0.3694, pruned_loss=0.117, over 28956.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3498, pruned_loss=0.1015, over 5713796.91 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3414, pruned_loss=0.08804, over 5586324.22 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3494, pruned_loss=0.1023, over 5701853.02 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:03:37,126 INFO [train.py:968] (1/2) Epoch 29, batch 6200, libri_loss[loss=0.2839, simple_loss=0.3644, pruned_loss=0.1017, over 28588.00 frames. ], tot_loss[loss=0.285, simple_loss=0.3562, pruned_loss=0.1069, over 5689750.10 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3411, pruned_loss=0.08793, over 5590371.46 frames. ], giga_tot_loss[loss=0.2861, simple_loss=0.3564, pruned_loss=0.1079, over 5678800.59 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:03:56,623 INFO [optim.py:369] (1/2) 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 20:04:03,899 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 6250, giga_loss[loss=0.4512, simple_loss=0.4633, pruned_loss=0.2196, over 26558.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3633, pruned_loss=0.1121, over 5689476.47 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3409, pruned_loss=0.08775, over 5599242.17 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3643, pruned_loss=0.1137, over 5675122.02 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:04:41,076 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6066, 1.6702, 1.2647, 1.2218], device='cuda:1'), covar=tensor([0.0921, 0.0577, 0.0993, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0450, 0.0526, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 20:05:04,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2622, 0.8197, 0.9990, 1.4390], device='cuda:1'), covar=tensor([0.0767, 0.0392, 0.0351, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0066, 0.0115], device='cuda:1')
+2023-03-14 20:05:05,313 INFO [train.py:968] (1/2) Epoch 29, batch 6300, giga_loss[loss=0.3333, simple_loss=0.385, pruned_loss=0.1408, over 28827.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3682, pruned_loss=0.1167, over 5678062.94 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.341, pruned_loss=0.08781, over 5597114.87 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3694, pruned_loss=0.1185, over 5669832.33 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:05:27,332 INFO [optim.py:369] (1/2) 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 20:05:40,329 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-14 20:05:49,719 INFO [train.py:968] (1/2) Epoch 29, batch 6350, giga_loss[loss=0.3716, simple_loss=0.4276, pruned_loss=0.1578, over 28781.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3746, pruned_loss=0.1217, over 5682710.30 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3409, pruned_loss=0.08786, over 5603477.59 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3762, pruned_loss=0.1237, over 5671983.85 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:06:12,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1282195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:06:15,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1282198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:06:37,663 INFO [train.py:968] (1/2) Epoch 29, batch 6400, giga_loss[loss=0.4651, simple_loss=0.4645, pruned_loss=0.2329, over 26553.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3778, pruned_loss=0.1248, over 5652894.07 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3413, pruned_loss=0.08809, over 5601110.18 frames. ], giga_tot_loss[loss=0.3173, simple_loss=0.3799, pruned_loss=0.1274, over 5648781.40 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:06:44,054 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:07:21,267 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2523, 4.0938, 3.8934, 1.6449], device='cuda:1'), covar=tensor([0.0624, 0.0743, 0.0782, 0.2189], device='cuda:1'), in_proj_covar=tensor([0.1301, 0.1201, 0.1012, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 20:07:27,387 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-14 20:07:27,544 INFO [train.py:968] (1/2) Epoch 29, batch 6450, giga_loss[loss=0.325, simple_loss=0.3861, pruned_loss=0.1319, over 28825.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.379, pruned_loss=0.1268, over 5650866.30 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3411, pruned_loss=0.088, over 5607426.31 frames. ], giga_tot_loss[loss=0.3204, simple_loss=0.3815, pruned_loss=0.1297, over 5642780.05 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:08:16,194 INFO [train.py:968] (1/2) Epoch 29, batch 6500, giga_loss[loss=0.4061, simple_loss=0.4276, pruned_loss=0.1923, over 23528.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3829, pruned_loss=0.1313, over 5638659.01 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08812, over 5617189.58 frames. ], giga_tot_loss[loss=0.3279, simple_loss=0.386, pruned_loss=0.1349, over 5624372.93 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:08:42,407 INFO [optim.py:369] (1/2) 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 20:09:11,427 INFO [train.py:968] (1/2) Epoch 29, batch 6550, libri_loss[loss=0.251, simple_loss=0.324, pruned_loss=0.08898, over 29457.00 frames. ], tot_loss[loss=0.3288, simple_loss=0.3869, pruned_loss=0.1354, over 5620140.67 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.08809, over 5623983.57 frames. ], giga_tot_loss[loss=0.3349, simple_loss=0.3908, pruned_loss=0.1395, over 5602057.61 frames. ], batch size: 70, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:09:13,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9009, 1.5037, 1.4306, 1.2642], device='cuda:1'), covar=tensor([0.2559, 0.1594, 0.2316, 0.1991], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0762, 0.0735, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 20:10:01,077 INFO [train.py:968] (1/2) Epoch 29, batch 6600, giga_loss[loss=0.3572, simple_loss=0.4073, pruned_loss=0.1535, over 28964.00 frames. ], tot_loss[loss=0.3306, simple_loss=0.3887, pruned_loss=0.1363, over 5629671.28 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3419, pruned_loss=0.08845, over 5627958.40 frames. ], giga_tot_loss[loss=0.3361, simple_loss=0.392, pruned_loss=0.1401, over 5611607.56 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:10:23,462 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 6650, giga_loss[loss=0.3033, simple_loss=0.3747, pruned_loss=0.1159, over 29085.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3877, pruned_loss=0.1364, over 5647043.59 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3421, pruned_loss=0.08854, over 5630781.77 frames. ], giga_tot_loss[loss=0.335, simple_loss=0.3906, pruned_loss=0.1397, over 5630552.67 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:11:27,632 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7478, 1.9842, 1.9362, 1.7808], device='cuda:1'), covar=tensor([0.3369, 0.2846, 0.2191, 0.2496], device='cuda:1'), in_proj_covar=tensor([0.2073, 0.2049, 0.1945, 0.2089], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 20:11:42,104 INFO [train.py:968] (1/2) Epoch 29, batch 6700, giga_loss[loss=0.331, simple_loss=0.3957, pruned_loss=0.1332, over 28930.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3851, pruned_loss=0.1345, over 5642617.06 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3415, pruned_loss=0.08817, over 5637842.53 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.389, pruned_loss=0.1387, over 5623089.90 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:11:52,411 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4666, 1.6717, 1.6607, 1.5891], device='cuda:1'), covar=tensor([0.1709, 0.1719, 0.1795, 0.1652], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0763, 0.0736, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-14 20:11:58,284 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1282538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:12:05,934 INFO [optim.py:369] (1/2) 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 20:12:11,828 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.60 vs. limit=2.0
+2023-03-14 20:12:28,765 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3638, 1.6065, 1.5879, 1.1685], device='cuda:1'), covar=tensor([0.1855, 0.3008, 0.1667, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.0932, 0.0715, 0.0979, 0.0881], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 20:12:30,353 INFO [train.py:968] (1/2) Epoch 29, batch 6750, libri_loss[loss=0.3502, simple_loss=0.4051, pruned_loss=0.1476, over 29609.00 frames. ], tot_loss[loss=0.3265, simple_loss=0.3855, pruned_loss=0.1337, over 5645205.99 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.0884, over 5641392.50 frames. ], giga_tot_loss[loss=0.3328, simple_loss=0.3896, pruned_loss=0.138, over 5626550.32 frames. ], batch size: 91, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:12:54,182 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 20:13:07,696 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3057, 1.3196, 4.0145, 3.4429], device='cuda:1'), covar=tensor([0.1788, 0.2878, 0.0464, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0675, 0.1013, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 20:13:17,345 INFO [train.py:968] (1/2) Epoch 29, batch 6800, giga_loss[loss=0.3212, simple_loss=0.3792, pruned_loss=0.1317, over 28491.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3856, pruned_loss=0.133, over 5643456.44 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08834, over 5645371.12 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3895, pruned_loss=0.1371, over 5625175.54 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:13:22,992 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4802, 2.0285, 1.6631, 1.5103], device='cuda:1'), covar=tensor([0.0777, 0.0293, 0.0314, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 20:13:41,478 INFO [optim.py:369] (1/2) 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 20:14:02,507 INFO [train.py:968] (1/2) Epoch 29, batch 6850, giga_loss[loss=0.3947, simple_loss=0.4242, pruned_loss=0.1826, over 23839.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3852, pruned_loss=0.1325, over 5624062.25 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08832, over 5652060.89 frames. ], giga_tot_loss[loss=0.3323, simple_loss=0.3898, pruned_loss=0.1374, over 5602346.67 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:14:47,794 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 6900, giga_loss[loss=0.3646, simple_loss=0.4107, pruned_loss=0.1593, over 27640.00 frames. ], tot_loss[loss=0.3201, simple_loss=0.3818, pruned_loss=0.1293, over 5619418.35 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08816, over 5646058.87 frames. ], giga_tot_loss[loss=0.3274, simple_loss=0.3866, pruned_loss=0.1342, over 5606890.64 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:15:12,385 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1282739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:15:12,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3781, 1.5724, 1.3873, 1.5938], device='cuda:1'), covar=tensor([0.0702, 0.0403, 0.0346, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0228], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 20:15:13,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3181, 2.2066, 1.3583, 1.4459], device='cuda:1'), covar=tensor([0.0823, 0.0417, 0.0728, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0573, 0.0413, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 20:15:18,868 INFO [optim.py:369] (1/2) 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 20:15:42,738 INFO [train.py:968] (1/2) Epoch 29, batch 6950, giga_loss[loss=0.324, simple_loss=0.385, pruned_loss=0.1315, over 29071.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.379, pruned_loss=0.1257, over 5629433.83 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3415, pruned_loss=0.08819, over 5648815.69 frames. ], giga_tot_loss[loss=0.3214, simple_loss=0.3831, pruned_loss=0.1299, over 5617024.88 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:16:30,716 INFO [train.py:968] (1/2) Epoch 29, batch 7000, giga_loss[loss=0.3212, simple_loss=0.3767, pruned_loss=0.1329, over 26630.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.376, pruned_loss=0.1232, over 5649239.66 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3412, pruned_loss=0.08822, over 5656409.41 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3804, pruned_loss=0.1275, over 5632228.16 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:16:54,977 INFO [optim.py:369] (1/2) 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 20:17:16,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-14 20:17:16,789 INFO [train.py:968] (1/2) Epoch 29, batch 7050, giga_loss[loss=0.3329, simple_loss=0.3834, pruned_loss=0.1412, over 28691.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3737, pruned_loss=0.1216, over 5646166.36 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08797, over 5656675.24 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3781, pruned_loss=0.1257, over 5632240.40 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:17:39,852 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1738, 1.1786, 3.4694, 3.2335], device='cuda:1'), covar=tensor([0.1902, 0.3116, 0.0873, 0.2371], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0678, 0.1013, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 20:17:59,933 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1282913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:18:05,924 INFO [train.py:968] (1/2) Epoch 29, batch 7100, giga_loss[loss=0.3233, simple_loss=0.3637, pruned_loss=0.1414, over 23685.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3718, pruned_loss=0.1206, over 5651590.10 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3404, pruned_loss=0.08778, over 5660337.55 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3762, pruned_loss=0.1245, over 5637132.32 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:18:28,449 INFO [optim.py:369] (1/2) 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 20:18:29,547 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 20:18:32,737 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3080, 2.9005, 1.4377, 1.4274], device='cuda:1'), covar=tensor([0.0996, 0.0417, 0.0902, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0574, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 20:18:43,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5827, 1.8191, 1.2688, 1.3587], device='cuda:1'), covar=tensor([0.1058, 0.0670, 0.1048, 0.1353], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0453, 0.0527, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 20:18:51,788 INFO [train.py:968] (1/2) Epoch 29, batch 7150, giga_loss[loss=0.3147, simple_loss=0.3797, pruned_loss=0.1248, over 28902.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3712, pruned_loss=0.1193, over 5658015.17 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3407, pruned_loss=0.08784, over 5659503.76 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3757, pruned_loss=0.1237, over 5646921.69 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:19:18,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1282998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:19:38,074 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 20:19:43,710 INFO [train.py:968] (1/2) Epoch 29, batch 7200, giga_loss[loss=0.2699, simple_loss=0.3522, pruned_loss=0.09381, over 28855.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3694, pruned_loss=0.1176, over 5664929.50 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08797, over 5660833.45 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3731, pruned_loss=0.1213, over 5655061.70 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:20:06,946 INFO [optim.py:369] (1/2) 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 20:20:16,097 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:20:19,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:20:36,393 INFO [train.py:968] (1/2) Epoch 29, batch 7250, giga_loss[loss=0.3065, simple_loss=0.3899, pruned_loss=0.1116, over 29003.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3694, pruned_loss=0.1159, over 5667586.95 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08807, over 5666657.49 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.373, pruned_loss=0.1193, over 5654968.39 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:20:56,093 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:20:56,793 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4810, 3.4021, 1.4551, 1.5390], device='cuda:1'), covar=tensor([0.0957, 0.0354, 0.0931, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0573, 0.0412, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 20:20:56,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:21:19,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:21:25,179 INFO [train.py:968] (1/2) Epoch 29, batch 7300, giga_loss[loss=0.3332, simple_loss=0.3987, pruned_loss=0.1339, over 28060.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3718, pruned_loss=0.1162, over 5667517.46 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.08808, over 5667892.20 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3747, pruned_loss=0.119, over 5656449.37 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:21:32,972 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-14 20:21:52,561 INFO [optim.py:369] (1/2) 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 20:21:53,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5272, 1.7362, 1.2470, 1.2700], device='cuda:1'), covar=tensor([0.1038, 0.0591, 0.1048, 0.1220], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0452, 0.0525, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 20:21:55,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:22:16,236 INFO [train.py:968] (1/2) Epoch 29, batch 7350, giga_loss[loss=0.2777, simple_loss=0.357, pruned_loss=0.09915, over 28877.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3709, pruned_loss=0.1157, over 5673084.08 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3406, pruned_loss=0.08798, over 5673247.23 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3742, pruned_loss=0.1187, over 5659339.98 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:22:52,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-14 20:23:00,780 INFO [train.py:968] (1/2) Epoch 29, batch 7400, giga_loss[loss=0.3834, simple_loss=0.4146, pruned_loss=0.1761, over 26575.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3713, pruned_loss=0.1168, over 5677925.55 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3403, pruned_loss=0.08779, over 5678498.52 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.375, pruned_loss=0.1201, over 5662215.51 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:23:09,482 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:12,332 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:22,588 INFO [optim.py:369] (1/2) 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 20:23:32,891 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:37,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:42,517 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:23:48,512 INFO [train.py:968] (1/2) Epoch 29, batch 7450, giga_loss[loss=0.3633, simple_loss=0.4065, pruned_loss=0.16, over 28560.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.369, pruned_loss=0.1162, over 5671748.32 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3403, pruned_loss=0.08785, over 5682735.06 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3725, pruned_loss=0.1194, over 5655618.13 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:24:01,372 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-14 20:24:04,428 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1283289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:24:32,083 INFO [train.py:968] (1/2) Epoch 29, batch 7500, giga_loss[loss=0.304, simple_loss=0.3491, pruned_loss=0.1294, over 23669.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3679, pruned_loss=0.116, over 5679700.33 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08808, over 5689470.01 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3711, pruned_loss=0.1191, over 5660314.09 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:24:54,523 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 7550, giga_loss[loss=0.2804, simple_loss=0.353, pruned_loss=0.1038, over 28907.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3675, pruned_loss=0.1153, over 5685686.16 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3411, pruned_loss=0.08831, over 5689240.33 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.37, pruned_loss=0.118, over 5670621.87 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:25:24,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1283373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:25:35,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9045, 1.2193, 1.2679, 1.0485], device='cuda:1'), covar=tensor([0.2081, 0.1524, 0.2584, 0.1918], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0764, 0.0738, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-14 20:26:08,689 INFO [train.py:968] (1/2) Epoch 29, batch 7600, giga_loss[loss=0.3403, simple_loss=0.3934, pruned_loss=0.1436, over 28831.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3671, pruned_loss=0.1135, over 5699304.83 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3411, pruned_loss=0.08826, over 5693637.22 frames. ], giga_tot_loss[loss=0.3009, simple_loss=0.3695, pruned_loss=0.1161, over 5683482.98 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:26:33,962 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 7650, giga_loss[loss=0.3172, simple_loss=0.3773, pruned_loss=0.1286, over 28608.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3669, pruned_loss=0.1133, over 5702930.72 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08802, over 5696701.74 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3695, pruned_loss=0.1159, over 5687688.43 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:27:16,082 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:27:26,741 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6465, 2.3989, 1.5818, 0.9547], device='cuda:1'), covar=tensor([0.9939, 0.4506, 0.4359, 0.8306], device='cuda:1'), in_proj_covar=tensor([0.1852, 0.1740, 0.1666, 0.1509], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 20:27:36,556 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283516.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:27:41,040 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:27:42,100 INFO [train.py:968] (1/2) Epoch 29, batch 7700, giga_loss[loss=0.2588, simple_loss=0.3257, pruned_loss=0.09593, over 28619.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3667, pruned_loss=0.1141, over 5694809.07 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.088, over 5697785.23 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3689, pruned_loss=0.1163, over 5681993.16 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:27:45,665 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 20:28:06,987 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 7750, giga_loss[loss=0.2791, simple_loss=0.353, pruned_loss=0.1026, over 28579.00 frames. ], tot_loss[loss=0.2958, simple_loss=0.3649, pruned_loss=0.1133, over 5700016.97 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08774, over 5703474.71 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3677, pruned_loss=0.116, over 5684630.92 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:28:36,916 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5514, 1.6201, 1.7363, 1.3399], device='cuda:1'), covar=tensor([0.1816, 0.2599, 0.1531, 0.1782], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0719, 0.0985, 0.0884], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 20:29:19,010 INFO [train.py:968] (1/2) Epoch 29, batch 7800, giga_loss[loss=0.2995, simple_loss=0.3756, pruned_loss=0.1117, over 29060.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3656, pruned_loss=0.1148, over 5692192.00 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3407, pruned_loss=0.08785, over 5703155.08 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3678, pruned_loss=0.1172, over 5680207.58 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:29:38,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1841, 1.2881, 3.4332, 3.1199], device='cuda:1'), covar=tensor([0.1749, 0.2811, 0.0567, 0.1072], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0677, 0.1015, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 20:29:44,075 INFO [optim.py:369] (1/2) 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 20:29:58,153 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8947, 5.6806, 5.3882, 3.0255], device='cuda:1'), covar=tensor([0.0579, 0.0737, 0.0950, 0.1723], device='cuda:1'), in_proj_covar=tensor([0.1316, 0.1217, 0.1022, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 20:29:59,496 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1283667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:30:01,785 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1283670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:30:02,105 INFO [train.py:968] (1/2) Epoch 29, batch 7850, giga_loss[loss=0.2814, simple_loss=0.3522, pruned_loss=0.1053, over 27992.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3644, pruned_loss=0.1141, over 5700686.03 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3409, pruned_loss=0.08773, over 5708107.89 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3669, pruned_loss=0.1171, over 5686179.91 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:30:27,813 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 7900, giga_loss[loss=0.3219, simple_loss=0.3789, pruned_loss=0.1325, over 28562.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3622, pruned_loss=0.1131, over 5701737.30 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3409, pruned_loss=0.08771, over 5710698.55 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3644, pruned_loss=0.1158, over 5687914.16 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:31:12,467 INFO [optim.py:369] (1/2) 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 20:31:31,064 INFO [train.py:968] (1/2) Epoch 29, batch 7950, giga_loss[loss=0.3339, simple_loss=0.3874, pruned_loss=0.1402, over 28044.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3616, pruned_loss=0.1129, over 5695102.79 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3413, pruned_loss=0.08785, over 5699969.61 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3637, pruned_loss=0.1157, over 5693929.58 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:32:15,034 INFO [train.py:968] (1/2) Epoch 29, batch 8000, giga_loss[loss=0.2682, simple_loss=0.3412, pruned_loss=0.09766, over 28993.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.361, pruned_loss=0.1126, over 5688994.04 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3409, pruned_loss=0.08762, over 5703877.91 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3635, pruned_loss=0.1158, over 5684501.50 frames. ], batch size: 106, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:32:42,037 INFO [optim.py:369] (1/2) 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 20:32:56,731 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283868.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:32:58,483 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 8050, giga_loss[loss=0.3093, simple_loss=0.374, pruned_loss=0.1223, over 28860.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3617, pruned_loss=0.1127, over 5689501.11 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3408, pruned_loss=0.08754, over 5708410.64 frames. ], giga_tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5681309.15 frames. ], batch size: 285, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:33:09,516 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1283881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:33:10,168 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 8100, giga_loss[loss=0.2811, simple_loss=0.3558, pruned_loss=0.1032, over 28848.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3622, pruned_loss=0.1124, over 5684292.63 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3406, pruned_loss=0.08755, over 5712231.85 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3649, pruned_loss=0.1155, over 5673719.53 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:34:06,140 INFO [optim.py:369] (1/2) 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 20:34:21,322 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4342, 3.2714, 1.5250, 1.5607], device='cuda:1'), covar=tensor([0.0959, 0.0338, 0.0897, 0.1279], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0572, 0.0411, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 20:34:25,517 INFO [train.py:968] (1/2) Epoch 29, batch 8150, giga_loss[loss=0.2883, simple_loss=0.3627, pruned_loss=0.107, over 28672.00 frames. ], tot_loss[loss=0.293, simple_loss=0.362, pruned_loss=0.112, over 5677398.19 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08777, over 5713285.49 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3647, pruned_loss=0.1148, over 5667268.37 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:35:04,404 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:35:07,846 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:35:12,592 INFO [train.py:968] (1/2) Epoch 29, batch 8200, giga_loss[loss=0.3154, simple_loss=0.3833, pruned_loss=0.1238, over 28954.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.363, pruned_loss=0.1129, over 5677718.30 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3408, pruned_loss=0.08797, over 5702728.04 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3652, pruned_loss=0.1153, over 5679012.23 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:35:38,286 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:35:42,420 INFO [optim.py:369] (1/2) 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 20:36:03,152 INFO [train.py:968] (1/2) Epoch 29, batch 8250, giga_loss[loss=0.3076, simple_loss=0.3691, pruned_loss=0.123, over 28836.00 frames. ], tot_loss[loss=0.299, simple_loss=0.366, pruned_loss=0.116, over 5672099.19 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.0881, over 5704162.81 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.368, pruned_loss=0.1184, over 5670829.74 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:36:09,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3807, 1.5331, 1.6230, 1.2232], device='cuda:1'), covar=tensor([0.1748, 0.2739, 0.1449, 0.1779], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0719, 0.0983, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 20:36:48,921 INFO [train.py:968] (1/2) Epoch 29, batch 8300, giga_loss[loss=0.2975, simple_loss=0.3691, pruned_loss=0.1129, over 28681.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3673, pruned_loss=0.1182, over 5676624.28 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.341, pruned_loss=0.08804, over 5709406.66 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3697, pruned_loss=0.1209, over 5670089.84 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:37:15,402 INFO [optim.py:369] (1/2) 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 20:37:33,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2967, 1.5782, 1.4634, 1.5377], device='cuda:1'), covar=tensor([0.0724, 0.0399, 0.0329, 0.0794], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 20:37:34,113 INFO [train.py:968] (1/2) Epoch 29, batch 8350, giga_loss[loss=0.3602, simple_loss=0.4065, pruned_loss=0.157, over 27513.00 frames. ], tot_loss[loss=0.305, simple_loss=0.369, pruned_loss=0.1205, over 5660828.97 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3404, pruned_loss=0.08783, over 5705144.75 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3724, pruned_loss=0.1241, over 5658621.14 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 20:38:19,174 INFO [train.py:968] (1/2) Epoch 29, batch 8400, libri_loss[loss=0.2335, simple_loss=0.3198, pruned_loss=0.07365, over 29569.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3698, pruned_loss=0.1216, over 5667787.50 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.08807, over 5711553.20 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3734, pruned_loss=0.1254, over 5658677.07 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:38:32,070 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0278, 2.0854, 1.5435, 1.6874], device='cuda:1'), covar=tensor([0.1056, 0.0764, 0.1123, 0.1263], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0454, 0.0528, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 20:38:36,312 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284243.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:38:41,069 INFO [optim.py:369] (1/2) 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 20:38:46,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:38:47,212 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:38:56,516 INFO [train.py:968] (1/2) Epoch 29, batch 8450, giga_loss[loss=0.2622, simple_loss=0.3326, pruned_loss=0.09589, over 28655.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3669, pruned_loss=0.1192, over 5666248.42 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3406, pruned_loss=0.08809, over 5708512.75 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3708, pruned_loss=0.1234, over 5659435.70 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:39:34,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-14 20:39:34,940 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8305, 2.5831, 1.6357, 1.0867], device='cuda:1'), covar=tensor([0.9284, 0.5465, 0.4501, 0.8056], device='cuda:1'), in_proj_covar=tensor([0.1852, 0.1741, 0.1663, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 20:39:35,221 INFO [train.py:968] (1/2) Epoch 29, batch 8500, giga_loss[loss=0.2625, simple_loss=0.3461, pruned_loss=0.08946, over 28700.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3666, pruned_loss=0.1173, over 5680891.48 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08808, over 5716573.48 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3707, pruned_loss=0.1221, over 5666542.87 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:39:59,611 INFO [optim.py:369] (1/2) 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 20:40:14,030 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:16,405 INFO [train.py:968] (1/2) Epoch 29, batch 8550, giga_loss[loss=0.273, simple_loss=0.3503, pruned_loss=0.09783, over 28906.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3649, pruned_loss=0.1156, over 5676439.59 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3406, pruned_loss=0.08793, over 5722908.08 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3693, pruned_loss=0.1207, over 5657265.78 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:40:27,994 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284386.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:40:30,850 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284389.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:40:38,328 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:39,005 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:40,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:40:40,857 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284418.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:40:57,975 INFO [train.py:968] (1/2) Epoch 29, batch 8600, giga_loss[loss=0.2758, simple_loss=0.3387, pruned_loss=0.1064, over 28458.00 frames. ], tot_loss[loss=0.2974, simple_loss=0.364, pruned_loss=0.1154, over 5686009.02 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.08815, over 5728838.61 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3682, pruned_loss=0.12, over 5664435.77 frames. ], batch size: 78, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:41:08,989 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:41:09,765 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:41:23,856 INFO [optim.py:369] (1/2) 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 20:41:37,417 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 20:41:42,734 INFO [train.py:968] (1/2) Epoch 29, batch 8650, giga_loss[loss=0.2873, simple_loss=0.353, pruned_loss=0.1109, over 28844.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3627, pruned_loss=0.1157, over 5684110.94 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.08817, over 5728806.41 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.366, pruned_loss=0.1194, over 5667093.95 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:41:49,894 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5727, 1.5898, 1.7542, 1.4061], device='cuda:1'), covar=tensor([0.1482, 0.2205, 0.1216, 0.1556], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0719, 0.0983, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 20:42:29,031 INFO [train.py:968] (1/2) Epoch 29, batch 8700, giga_loss[loss=0.2649, simple_loss=0.3383, pruned_loss=0.0958, over 28914.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3623, pruned_loss=0.116, over 5664462.36 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3405, pruned_loss=0.08802, over 5729734.92 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3657, pruned_loss=0.1197, over 5649092.29 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:42:57,315 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 8750, giga_loss[loss=0.3534, simple_loss=0.3889, pruned_loss=0.159, over 23704.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.1171, over 5662240.99 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3403, pruned_loss=0.08797, over 5723965.76 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3687, pruned_loss=0.1208, over 5653018.82 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:43:59,961 INFO [train.py:968] (1/2) Epoch 29, batch 8800, giga_loss[loss=0.3161, simple_loss=0.3811, pruned_loss=0.1256, over 28855.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3685, pruned_loss=0.1166, over 5673052.63 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3405, pruned_loss=0.08812, over 5728197.86 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3718, pruned_loss=0.12, over 5660657.12 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:44:28,130 INFO [optim.py:369] (1/2) 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 20:44:44,500 INFO [train.py:968] (1/2) Epoch 29, batch 8850, giga_loss[loss=0.2875, simple_loss=0.3625, pruned_loss=0.1063, over 28945.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3704, pruned_loss=0.117, over 5667266.50 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3405, pruned_loss=0.08825, over 5718260.45 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3733, pruned_loss=0.12, over 5665789.59 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:44:49,781 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:44:53,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3981, 1.1446, 4.5219, 3.4196], device='cuda:1'), covar=tensor([0.1759, 0.3077, 0.0454, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0679, 0.1015, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 20:45:20,189 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-14 20:45:26,683 INFO [train.py:968] (1/2) Epoch 29, batch 8900, giga_loss[loss=0.3887, simple_loss=0.4134, pruned_loss=0.182, over 23465.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3716, pruned_loss=0.1183, over 5665213.31 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3401, pruned_loss=0.08803, over 5722322.38 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3749, pruned_loss=0.1214, over 5659268.44 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:45:40,868 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:45:45,604 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1284742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:45:52,178 INFO [optim.py:369] (1/2) 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 20:45:53,287 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 29, batch 8950, libri_loss[loss=0.267, simple_loss=0.3571, pruned_loss=0.08844, over 29524.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3727, pruned_loss=0.1194, over 5659744.13 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3403, pruned_loss=0.08799, over 5723907.76 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.376, pruned_loss=0.1228, over 5651700.32 frames. ], batch size: 82, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:46:19,337 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284783.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:46:29,066 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1284795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:46:48,723 INFO [train.py:968] (1/2) Epoch 29, batch 9000, giga_loss[loss=0.3011, simple_loss=0.372, pruned_loss=0.1151, over 28613.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3697, pruned_loss=0.1179, over 5652588.73 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3399, pruned_loss=0.0878, over 5712718.38 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3741, pruned_loss=0.1222, over 5652405.24 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:46:48,723 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 20:46:57,132 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 20:47:25,889 INFO [optim.py:369] (1/2) 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 20:47:40,282 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8490, 2.0174, 2.1197, 1.6318], device='cuda:1'), covar=tensor([0.1770, 0.2198, 0.1359, 0.1617], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0718, 0.0982, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 20:47:45,177 INFO [train.py:968] (1/2) Epoch 29, batch 9050, giga_loss[loss=0.2873, simple_loss=0.3524, pruned_loss=0.1111, over 28902.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3682, pruned_loss=0.1182, over 5637559.32 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3396, pruned_loss=0.08775, over 5715307.47 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3724, pruned_loss=0.122, over 5634156.82 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:47:55,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1284885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:47:57,631 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1284888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:48:15,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5241, 1.8490, 1.5078, 1.6017], device='cuda:1'), covar=tensor([0.2170, 0.2167, 0.2272, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.1609, 0.1159, 0.1421, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 20:48:25,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1284917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:48:28,196 INFO [train.py:968] (1/2) Epoch 29, batch 9100, giga_loss[loss=0.3133, simple_loss=0.386, pruned_loss=0.1203, over 29084.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.367, pruned_loss=0.1178, over 5653420.84 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3397, pruned_loss=0.08775, over 5719130.34 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3709, pruned_loss=0.1215, over 5645836.63 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:48:55,122 INFO [optim.py:369] (1/2) 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 20:49:11,065 INFO [train.py:968] (1/2) Epoch 29, batch 9150, giga_loss[loss=0.282, simple_loss=0.3554, pruned_loss=0.1043, over 29010.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3678, pruned_loss=0.1188, over 5661953.47 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3402, pruned_loss=0.08793, over 5722663.05 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3711, pruned_loss=0.1224, over 5651374.29 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:49:57,266 INFO [train.py:968] (1/2) Epoch 29, batch 9200, giga_loss[loss=0.2993, simple_loss=0.3707, pruned_loss=0.1139, over 28723.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3681, pruned_loss=0.1195, over 5648065.20 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3406, pruned_loss=0.08813, over 5722389.15 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.123, over 5637861.44 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:50:00,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0200, 1.5870, 5.1273, 4.0169], device='cuda:1'), covar=tensor([0.1433, 0.2571, 0.0395, 0.0681], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0678, 0.1016, 0.0994], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 20:50:22,874 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:50:26,856 INFO [optim.py:369] (1/2) 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 20:50:36,978 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-14 20:50:41,801 INFO [train.py:968] (1/2) Epoch 29, batch 9250, libri_loss[loss=0.2908, simple_loss=0.3669, pruned_loss=0.1073, over 29525.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3666, pruned_loss=0.1188, over 5655125.97 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.341, pruned_loss=0.08841, over 5720396.45 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3691, pruned_loss=0.1219, over 5647506.76 frames. ], batch size: 82, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:51:15,955 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:51:16,623 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285112.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:51:25,360 INFO [train.py:968] (1/2) Epoch 29, batch 9300, giga_loss[loss=0.2977, simple_loss=0.3651, pruned_loss=0.1152, over 28668.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3642, pruned_loss=0.1177, over 5660581.17 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.341, pruned_loss=0.0884, over 5724779.37 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3668, pruned_loss=0.1209, over 5649329.60 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:51:26,998 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:51:51,527 INFO [optim.py:369] (1/2) 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 20:51:54,310 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7327, 1.3120, 4.7339, 3.7390], device='cuda:1'), covar=tensor([0.1620, 0.2897, 0.0421, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0680, 0.1019, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 20:51:54,906 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:51:56,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:52:05,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:05,968 INFO [train.py:968] (1/2) Epoch 29, batch 9350, giga_loss[loss=0.2493, simple_loss=0.3317, pruned_loss=0.08351, over 28433.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.363, pruned_loss=0.1158, over 5661533.91 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3409, pruned_loss=0.08831, over 5730471.94 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.366, pruned_loss=0.1195, over 5644770.50 frames. ], batch size: 65, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:52:19,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:28,278 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:31,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:52:53,244 INFO [train.py:968] (1/2) Epoch 29, batch 9400, giga_loss[loss=0.3851, simple_loss=0.4165, pruned_loss=0.1768, over 26488.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3647, pruned_loss=0.1159, over 5666169.58 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.341, pruned_loss=0.08835, over 5733336.28 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3673, pruned_loss=0.1192, over 5649463.23 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:52:56,397 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:19,527 INFO [optim.py:369] (1/2) 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 20:53:21,765 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:23,661 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:36,602 INFO [train.py:968] (1/2) Epoch 29, batch 9450, giga_loss[loss=0.3253, simple_loss=0.3816, pruned_loss=0.1345, over 27966.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3673, pruned_loss=0.1183, over 5649694.29 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.341, pruned_loss=0.08842, over 5726450.37 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5642069.61 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:53:47,566 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:53:48,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0075, 3.0022, 1.9080, 1.2289], device='cuda:1'), covar=tensor([0.8659, 0.3747, 0.4516, 0.7371], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1742, 0.1665, 0.1508], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 20:54:00,990 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285301.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:54:03,817 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:54:12,507 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:54:15,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:54:20,375 INFO [train.py:968] (1/2) Epoch 29, batch 9500, giga_loss[loss=0.2621, simple_loss=0.3528, pruned_loss=0.08575, over 28638.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3672, pruned_loss=0.1172, over 5664248.78 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.08813, over 5732608.61 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3701, pruned_loss=0.1207, over 5650786.68 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:54:30,655 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285333.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:54:40,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0049, 1.5408, 1.3401, 1.2934], device='cuda:1'), covar=tensor([0.2625, 0.1999, 0.2612, 0.2369], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0763, 0.0735, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 20:54:41,856 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:54:47,990 INFO [optim.py:369] (1/2) 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 20:55:02,210 INFO [train.py:968] (1/2) Epoch 29, batch 9550, giga_loss[loss=0.3122, simple_loss=0.3885, pruned_loss=0.1179, over 28881.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3681, pruned_loss=0.1157, over 5669139.93 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3405, pruned_loss=0.08806, over 5736524.91 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3713, pruned_loss=0.1192, over 5653228.35 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:55:42,557 INFO [train.py:968] (1/2) Epoch 29, batch 9600, giga_loss[loss=0.289, simple_loss=0.3708, pruned_loss=0.1036, over 29138.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3712, pruned_loss=0.116, over 5681690.96 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3407, pruned_loss=0.0881, over 5739519.73 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3742, pruned_loss=0.1193, over 5665353.95 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:56:10,954 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 9650, giga_loss[loss=0.3169, simple_loss=0.3775, pruned_loss=0.1281, over 28677.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3743, pruned_loss=0.1181, over 5678131.27 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3408, pruned_loss=0.08815, over 5740785.88 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3771, pruned_loss=0.1212, over 5662952.36 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 20:56:45,439 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285487.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:56:54,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:56:58,559 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:57:13,015 INFO [train.py:968] (1/2) Epoch 29, batch 9700, giga_loss[loss=0.3137, simple_loss=0.3697, pruned_loss=0.1288, over 28849.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3765, pruned_loss=0.1208, over 5678046.80 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.0881, over 5738308.91 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3797, pruned_loss=0.1242, over 5666698.55 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:57:23,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:57:43,357 INFO [optim.py:369] (1/2) 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 20:57:49,774 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:57:59,211 INFO [train.py:968] (1/2) Epoch 29, batch 9750, giga_loss[loss=0.3202, simple_loss=0.3825, pruned_loss=0.1289, over 28962.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3787, pruned_loss=0.124, over 5664365.05 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3409, pruned_loss=0.08818, over 5738048.48 frames. ], giga_tot_loss[loss=0.3178, simple_loss=0.3816, pruned_loss=0.1271, over 5654597.47 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:58:26,656 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-14 20:58:41,423 INFO [train.py:968] (1/2) Epoch 29, batch 9800, giga_loss[loss=0.2845, simple_loss=0.3559, pruned_loss=0.1065, over 28854.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3765, pruned_loss=0.1224, over 5671030.38 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08845, over 5743322.65 frames. ], giga_tot_loss[loss=0.3161, simple_loss=0.3802, pruned_loss=0.126, over 5655931.79 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:58:50,589 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285630.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 20:58:52,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285633.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 20:58:57,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3716, 1.4032, 1.3659, 1.5478], device='cuda:1'), covar=tensor([0.0799, 0.0351, 0.0326, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 20:59:00,018 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:01,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:07,673 INFO [optim.py:369] (1/2) 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 20:59:10,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6668, 1.9717, 1.8104, 1.6700], device='cuda:1'), covar=tensor([0.2514, 0.2907, 0.2599, 0.2988], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0759, 0.0733, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 20:59:16,192 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 9850, giga_loss[loss=0.2938, simple_loss=0.3654, pruned_loss=0.1111, over 28672.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3751, pruned_loss=0.1204, over 5675595.86 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3414, pruned_loss=0.08866, over 5741786.56 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3781, pruned_loss=0.1235, over 5663921.46 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 20:59:25,863 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:27,318 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:30,061 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1285678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:45,636 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4487, 1.7688, 1.4214, 1.2819], device='cuda:1'), covar=tensor([0.1189, 0.0674, 0.1078, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 20:59:51,927 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1285703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 20:59:53,754 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:00:05,046 INFO [train.py:968] (1/2) Epoch 29, batch 9900, giga_loss[loss=0.282, simple_loss=0.3554, pruned_loss=0.1043, over 28579.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3729, pruned_loss=0.1172, over 5665241.84 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3412, pruned_loss=0.08847, over 5731463.96 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3767, pruned_loss=0.1208, over 5663382.10 frames. ], batch size: 71, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:00:19,562 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1285735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:00:29,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9410, 3.7509, 3.5709, 1.7733], device='cuda:1'), covar=tensor([0.0807, 0.0967, 0.0930, 0.2040], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1227, 0.1031, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 21:00:33,342 INFO [optim.py:369] (1/2) 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 21:00:48,178 INFO [train.py:968] (1/2) Epoch 29, batch 9950, giga_loss[loss=0.2978, simple_loss=0.3757, pruned_loss=0.1099, over 28971.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3731, pruned_loss=0.1168, over 5672565.35 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3412, pruned_loss=0.08845, over 5735208.07 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3766, pruned_loss=0.1202, over 5666683.26 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:01:10,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3249, 1.6363, 1.5748, 1.1583], device='cuda:1'), covar=tensor([0.1930, 0.3061, 0.1754, 0.2100], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0721, 0.0986, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 21:01:35,785 INFO [train.py:968] (1/2) Epoch 29, batch 10000, giga_loss[loss=0.3051, simple_loss=0.3692, pruned_loss=0.1206, over 28537.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3746, pruned_loss=0.1187, over 5669007.75 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3413, pruned_loss=0.08844, over 5740279.41 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3783, pruned_loss=0.1224, over 5657696.32 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:02:06,363 INFO [optim.py:369] (1/2) 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 21:02:06,643 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:02:20,265 INFO [train.py:968] (1/2) Epoch 29, batch 10050, giga_loss[loss=0.3307, simple_loss=0.3856, pruned_loss=0.1379, over 28345.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3727, pruned_loss=0.1178, over 5666997.80 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3414, pruned_loss=0.08842, over 5743855.52 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3764, pruned_loss=0.1216, over 5652992.13 frames. ], batch size: 369, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:02:27,327 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1285878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:02:46,739 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.22 vs. limit=5.0
+2023-03-14 21:03:11,497 INFO [train.py:968] (1/2) Epoch 29, batch 10100, libri_loss[loss=0.2838, simple_loss=0.36, pruned_loss=0.1038, over 19858.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3729, pruned_loss=0.1199, over 5647323.87 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.08847, over 5734994.24 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3759, pruned_loss=0.1229, over 5644789.02 frames. ], batch size: 187, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:03:34,594 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285947.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:03:39,108 INFO [optim.py:369] (1/2) 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 21:03:41,042 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-14 21:03:46,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1285957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:03:57,309 INFO [train.py:968] (1/2) Epoch 29, batch 10150, giga_loss[loss=0.2798, simple_loss=0.3524, pruned_loss=0.1036, over 28848.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3708, pruned_loss=0.1191, over 5663666.82 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3412, pruned_loss=0.0883, over 5738553.29 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3743, pruned_loss=0.1225, over 5656393.58 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:04:49,866 INFO [train.py:968] (1/2) Epoch 29, batch 10200, giga_loss[loss=0.3356, simple_loss=0.3817, pruned_loss=0.1448, over 27516.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3688, pruned_loss=0.1186, over 5654299.75 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3412, pruned_loss=0.08823, over 5741396.62 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3722, pruned_loss=0.122, over 5644479.44 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:04:50,216 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:04:53,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:05:19,958 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:05:20,288 INFO [optim.py:369] (1/2) 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 21:05:38,593 INFO [train.py:968] (1/2) Epoch 29, batch 10250, giga_loss[loss=0.2767, simple_loss=0.3543, pruned_loss=0.0995, over 28735.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3681, pruned_loss=0.1188, over 5655286.22 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08836, over 5735359.23 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3712, pruned_loss=0.122, over 5651438.06 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:05:59,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6750, 1.5653, 1.8870, 1.5164], device='cuda:1'), covar=tensor([0.1368, 0.2152, 0.1163, 0.1482], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0722, 0.0986, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 21:06:23,046 INFO [train.py:968] (1/2) Epoch 29, batch 10300, giga_loss[loss=0.2884, simple_loss=0.3572, pruned_loss=0.1098, over 27646.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3665, pruned_loss=0.1178, over 5644504.08 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3414, pruned_loss=0.0884, over 5727444.05 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3691, pruned_loss=0.1206, over 5647070.74 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:06:29,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1286126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:06:53,412 INFO [optim.py:369] (1/2) 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 21:07:05,569 INFO [train.py:968] (1/2) Epoch 29, batch 10350, giga_loss[loss=0.2509, simple_loss=0.329, pruned_loss=0.08642, over 28819.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3631, pruned_loss=0.1134, over 5662884.97 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3413, pruned_loss=0.0882, over 5734788.70 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3665, pruned_loss=0.117, over 5655081.14 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:07:43,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-14 21:07:53,345 INFO [train.py:968] (1/2) Epoch 29, batch 10400, giga_loss[loss=0.3055, simple_loss=0.3747, pruned_loss=0.1181, over 28728.00 frames. ], tot_loss[loss=0.2913, simple_loss=0.3604, pruned_loss=0.1111, over 5662700.00 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3414, pruned_loss=0.0884, over 5738220.86 frames. ], giga_tot_loss[loss=0.2961, simple_loss=0.3634, pruned_loss=0.1144, over 5651917.08 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:07:56,627 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-14 21:08:01,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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 21:08:40,435 INFO [train.py:968] (1/2) Epoch 29, batch 10450, giga_loss[loss=0.2852, simple_loss=0.3605, pruned_loss=0.1049, over 29077.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3604, pruned_loss=0.1109, over 5669931.70 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08831, over 5738044.82 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3631, pruned_loss=0.114, over 5660525.91 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:09:29,980 INFO [train.py:968] (1/2) Epoch 29, batch 10500, giga_loss[loss=0.2973, simple_loss=0.3596, pruned_loss=0.1175, over 27997.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3576, pruned_loss=0.1103, over 5667137.75 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3417, pruned_loss=0.08853, over 5740080.24 frames. ], giga_tot_loss[loss=0.293, simple_loss=0.3599, pruned_loss=0.113, over 5656448.56 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:09:30,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:09:31,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-14 21:09:39,652 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:10:17,876 INFO [train.py:968] (1/2) Epoch 29, batch 10550, giga_loss[loss=0.2827, simple_loss=0.364, pruned_loss=0.1006, over 28946.00 frames. ], tot_loss[loss=0.2884, simple_loss=0.3565, pruned_loss=0.1101, over 5669714.39 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08834, over 5742564.41 frames. ], giga_tot_loss[loss=0.2922, simple_loss=0.3588, pruned_loss=0.1128, over 5657867.95 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:10:18,916 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:10:20,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:10:43,615 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:10:59,063 INFO [train.py:968] (1/2) Epoch 29, batch 10600, giga_loss[loss=0.3263, simple_loss=0.3888, pruned_loss=0.1319, over 29061.00 frames. ], tot_loss[loss=0.2912, simple_loss=0.3596, pruned_loss=0.1115, over 5673472.33 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.08834, over 5747540.84 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3619, pruned_loss=0.1143, over 5657501.68 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:11:33,501 INFO [optim.py:369] (1/2) 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 21:11:41,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:11:43,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:11:44,881 INFO [train.py:968] (1/2) Epoch 29, batch 10650, giga_loss[loss=0.3312, simple_loss=0.3922, pruned_loss=0.1351, over 28263.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3631, pruned_loss=0.1136, over 5649436.42 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3418, pruned_loss=0.08865, over 5730606.26 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3653, pruned_loss=0.1164, over 5649987.07 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:11:49,743 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:11:52,442 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:10,909 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:14,333 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:19,649 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:12:32,299 INFO [train.py:968] (1/2) Epoch 29, batch 10700, giga_loss[loss=0.2952, simple_loss=0.3635, pruned_loss=0.1134, over 28656.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3628, pruned_loss=0.1137, over 5646922.80 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3419, pruned_loss=0.08867, over 5730256.25 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3648, pruned_loss=0.1163, over 5646162.13 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:13:04,681 INFO [optim.py:369] (1/2) 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 21:13:04,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4632, 1.8780, 1.4406, 1.6082], device='cuda:1'), covar=tensor([0.2608, 0.2626, 0.3033, 0.2391], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1165, 0.1428, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 21:13:19,302 INFO [train.py:968] (1/2) Epoch 29, batch 10750, giga_loss[loss=0.2946, simple_loss=0.3632, pruned_loss=0.113, over 28595.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3622, pruned_loss=0.1137, over 5648140.70 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3417, pruned_loss=0.08844, over 5731114.59 frames. ], giga_tot_loss[loss=0.2984, simple_loss=0.3642, pruned_loss=0.1163, over 5645885.33 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:13:31,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5154, 4.3600, 4.1684, 2.0121], device='cuda:1'), covar=tensor([0.0608, 0.0720, 0.0744, 0.1979], device='cuda:1'), in_proj_covar=tensor([0.1325, 0.1223, 0.1028, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 21:14:13,048 INFO [train.py:968] (1/2) Epoch 29, batch 10800, giga_loss[loss=0.3487, simple_loss=0.3968, pruned_loss=0.1503, over 27543.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3654, pruned_loss=0.1161, over 5656032.09 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08836, over 5733690.27 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3674, pruned_loss=0.1186, over 5651004.20 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:14:19,759 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1286627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:14:33,677 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:14:37,153 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 29, batch 10850, giga_loss[loss=0.283, simple_loss=0.3593, pruned_loss=0.1033, over 28815.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3659, pruned_loss=0.1156, over 5653656.99 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3416, pruned_loss=0.08825, over 5727170.26 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3682, pruned_loss=0.1186, over 5653060.95 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:15:02,807 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1286676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:15:43,306 INFO [train.py:968] (1/2) Epoch 29, batch 10900, libri_loss[loss=0.2656, simple_loss=0.3391, pruned_loss=0.09611, over 29562.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3677, pruned_loss=0.1168, over 5664173.27 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3418, pruned_loss=0.0883, over 5730438.76 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3701, pruned_loss=0.1199, over 5658670.74 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:16:06,037 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.80 vs. limit=5.0
+2023-03-14 21:16:13,473 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 21:16:27,351 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1286770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:16:27,737 INFO [train.py:968] (1/2) Epoch 29, batch 10950, libri_loss[loss=0.278, simple_loss=0.3553, pruned_loss=0.1003, over 19321.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3698, pruned_loss=0.1189, over 5659611.57 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3422, pruned_loss=0.08862, over 5721258.64 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3723, pruned_loss=0.122, over 5662524.93 frames. ], batch size: 187, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:16:29,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1286773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:16:40,413 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 21:16:49,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4582, 1.6090, 1.4395, 1.4860], device='cuda:1'), covar=tensor([0.0767, 0.0338, 0.0335, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 21:16:54,997 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 11000, giga_loss[loss=0.2795, simple_loss=0.3628, pruned_loss=0.09809, over 28713.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3722, pruned_loss=0.1201, over 5650400.14 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3421, pruned_loss=0.08884, over 5714875.01 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3746, pruned_loss=0.1229, over 5657633.56 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:17:52,403 INFO [optim.py:369] (1/2) 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 21:18:05,900 INFO [train.py:968] (1/2) Epoch 29, batch 11050, giga_loss[loss=0.2795, simple_loss=0.357, pruned_loss=0.101, over 29034.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3711, pruned_loss=0.1186, over 5656345.97 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3424, pruned_loss=0.08896, over 5717719.93 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3732, pruned_loss=0.1211, over 5658574.78 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:18:55,201 INFO [train.py:968] (1/2) Epoch 29, batch 11100, libri_loss[loss=0.2866, simple_loss=0.3697, pruned_loss=0.1018, over 25985.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1202, over 5640461.56 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08903, over 5709480.58 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.374, pruned_loss=0.1228, over 5648495.98 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:19:09,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8024, 2.0424, 2.0486, 1.6844], device='cuda:1'), covar=tensor([0.3631, 0.3074, 0.3402, 0.3223], device='cuda:1'), in_proj_covar=tensor([0.2083, 0.2063, 0.1960, 0.2109], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 21:19:28,539 INFO [optim.py:369] (1/2) 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 21:19:51,322 INFO [train.py:968] (1/2) Epoch 29, batch 11150, giga_loss[loss=0.2694, simple_loss=0.3394, pruned_loss=0.09975, over 28434.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.371, pruned_loss=0.1204, over 5629165.71 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08904, over 5705253.53 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3733, pruned_loss=0.1233, over 5637170.41 frames. ], batch size: 60, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:20:21,042 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 11200, libri_loss[loss=0.2659, simple_loss=0.3486, pruned_loss=0.09164, over 29527.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3679, pruned_loss=0.1184, over 5639186.28 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3422, pruned_loss=0.08885, over 5711080.29 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.3709, pruned_loss=0.1217, over 5638392.81 frames. ], batch size: 83, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:20:45,746 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-14 21:21:11,281 INFO [optim.py:369] (1/2) 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 21:21:24,848 INFO [train.py:968] (1/2) Epoch 29, batch 11250, giga_loss[loss=0.28, simple_loss=0.3538, pruned_loss=0.1031, over 28823.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3668, pruned_loss=0.1182, over 5646084.43 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3426, pruned_loss=0.08903, over 5713694.59 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3692, pruned_loss=0.121, over 5642406.29 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:22:12,405 INFO [train.py:968] (1/2) Epoch 29, batch 11300, giga_loss[loss=0.3209, simple_loss=0.3591, pruned_loss=0.1414, over 23461.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3653, pruned_loss=0.117, over 5653617.98 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08904, over 5718587.36 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3677, pruned_loss=0.1199, over 5644906.31 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:22:20,486 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1287130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:22:27,099 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1287139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:22:34,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.59 vs. limit=2.0
+2023-03-14 21:22:44,019 INFO [optim.py:369] (1/2) 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 21:22:58,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2723, 2.2035, 2.1899, 1.8476], device='cuda:1'), covar=tensor([0.2160, 0.2641, 0.2278, 0.2646], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0761, 0.0734, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 21:22:59,188 INFO [train.py:968] (1/2) Epoch 29, batch 11350, giga_loss[loss=0.2578, simple_loss=0.3314, pruned_loss=0.0921, over 29185.00 frames. ], tot_loss[loss=0.298, simple_loss=0.364, pruned_loss=0.116, over 5649259.04 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3427, pruned_loss=0.08888, over 5715339.04 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.3667, pruned_loss=0.1195, over 5642401.18 frames. ], batch size: 113, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:23:39,327 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3205, 1.6769, 1.2451, 0.8467], device='cuda:1'), covar=tensor([0.4645, 0.2655, 0.2718, 0.5605], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1747, 0.1669, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 21:23:43,591 INFO [train.py:968] (1/2) Epoch 29, batch 11400, libri_loss[loss=0.2407, simple_loss=0.3311, pruned_loss=0.07512, over 29451.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3644, pruned_loss=0.1166, over 5647199.58 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08887, over 5712848.69 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3675, pruned_loss=0.1206, over 5641319.37 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:24:13,002 INFO [optim.py:369] (1/2) 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 21:24:27,611 INFO [train.py:968] (1/2) Epoch 29, batch 11450, giga_loss[loss=0.2845, simple_loss=0.3572, pruned_loss=0.1059, over 29013.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3672, pruned_loss=0.1187, over 5661616.93 frames. ], libri_tot_loss[loss=0.2603, simple_loss=0.3425, pruned_loss=0.08904, over 5719082.53 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3705, pruned_loss=0.1227, over 5649680.47 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:24:29,302 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1287273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:24:32,552 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1287276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:24:55,015 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 21:25:00,222 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1287305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:25:15,405 INFO [train.py:968] (1/2) Epoch 29, batch 11500, giga_loss[loss=0.2975, simple_loss=0.3603, pruned_loss=0.1173, over 28960.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3686, pruned_loss=0.1202, over 5645807.69 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3431, pruned_loss=0.08916, over 5722025.43 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3712, pruned_loss=0.1238, over 5632627.42 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:25:28,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6371, 1.8524, 1.6479, 1.4679], device='cuda:1'), covar=tensor([0.3054, 0.2689, 0.2812, 0.2887], device='cuda:1'), in_proj_covar=tensor([0.2082, 0.2062, 0.1958, 0.2109], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 21:25:31,888 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2442, 3.0916, 2.9516, 1.4939], device='cuda:1'), covar=tensor([0.1087, 0.1124, 0.1027, 0.2325], device='cuda:1'), in_proj_covar=tensor([0.1332, 0.1227, 0.1031, 0.0762], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 21:25:51,783 INFO [optim.py:369] (1/2) 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 21:26:05,310 INFO [train.py:968] (1/2) Epoch 29, batch 11550, giga_loss[loss=0.3241, simple_loss=0.381, pruned_loss=0.1336, over 28619.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3697, pruned_loss=0.1214, over 5658801.58 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.343, pruned_loss=0.08913, over 5723077.30 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3719, pruned_loss=0.1244, over 5647337.56 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:26:10,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1287376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:26:33,649 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-14 21:26:52,951 INFO [train.py:968] (1/2) Epoch 29, batch 11600, giga_loss[loss=0.2967, simple_loss=0.3722, pruned_loss=0.1106, over 29049.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3693, pruned_loss=0.1212, over 5654883.49 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08887, over 5725624.97 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3717, pruned_loss=0.1242, over 5642754.08 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:27:23,560 INFO [optim.py:369] (1/2) 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 21:27:24,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-14 21:27:36,940 INFO [train.py:968] (1/2) Epoch 29, batch 11650, giga_loss[loss=0.2715, simple_loss=0.3457, pruned_loss=0.09861, over 28560.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3684, pruned_loss=0.1197, over 5664335.95 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3428, pruned_loss=0.08909, over 5731488.71 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3716, pruned_loss=0.1236, over 5645276.69 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:28:17,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1287514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:22,897 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1287519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:24,008 INFO [train.py:968] (1/2) Epoch 29, batch 11700, giga_loss[loss=0.3164, simple_loss=0.3801, pruned_loss=0.1263, over 28601.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3684, pruned_loss=0.119, over 5665914.53 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.3427, pruned_loss=0.08903, over 5721512.48 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3714, pruned_loss=0.1227, over 5658333.14 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:28:25,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1287522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:52,498 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1287551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:28:52,675 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 21:28:58,354 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 11750, giga_loss[loss=0.2907, simple_loss=0.3603, pruned_loss=0.1105, over 28987.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3703, pruned_loss=0.1206, over 5657675.40 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08887, over 5721813.56 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3739, pruned_loss=0.1249, over 5648935.80 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:29:58,743 INFO [train.py:968] (1/2) Epoch 29, batch 11800, giga_loss[loss=0.2665, simple_loss=0.3408, pruned_loss=0.09612, over 28823.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3717, pruned_loss=0.1221, over 5657166.32 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3422, pruned_loss=0.08857, over 5723070.43 frames. ], giga_tot_loss[loss=0.3145, simple_loss=0.3757, pruned_loss=0.1267, over 5647397.18 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:30:33,260 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1287660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:30:45,641 INFO [train.py:968] (1/2) Epoch 29, batch 11850, libri_loss[loss=0.2956, simple_loss=0.3757, pruned_loss=0.1078, over 29194.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3719, pruned_loss=0.1228, over 5654405.09 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3423, pruned_loss=0.08857, over 5725402.30 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3755, pruned_loss=0.1269, over 5643664.71 frames. ], batch size: 101, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:31:01,859 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1287689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:31:33,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5129, 3.7150, 1.5645, 1.6141], device='cuda:1'), covar=tensor([0.1012, 0.0289, 0.0967, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0577, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 21:31:34,205 INFO [train.py:968] (1/2) Epoch 29, batch 11900, libri_loss[loss=0.2646, simple_loss=0.3484, pruned_loss=0.09038, over 29268.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3734, pruned_loss=0.1226, over 5653080.23 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3423, pruned_loss=0.08859, over 5726943.75 frames. ], giga_tot_loss[loss=0.3144, simple_loss=0.3764, pruned_loss=0.1262, over 5642579.30 frames. ], batch size: 97, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:31:56,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.8771, 1.4752, 4.8428, 3.6851], device='cuda:1'), covar=tensor([0.1563, 0.2799, 0.0421, 0.0914], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0677, 0.1016, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-14 21:32:11,230 INFO [optim.py:369] (1/2) 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 21:32:22,521 INFO [train.py:968] (1/2) Epoch 29, batch 11950, giga_loss[loss=0.2826, simple_loss=0.3561, pruned_loss=0.1046, over 28612.00 frames. ], tot_loss[loss=0.31, simple_loss=0.3742, pruned_loss=0.1229, over 5656230.64 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3425, pruned_loss=0.0887, over 5728422.02 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3768, pruned_loss=0.1259, over 5645995.13 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:32:23,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5481, 1.6453, 1.7356, 1.3450], device='cuda:1'), covar=tensor([0.1701, 0.2519, 0.1412, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0721, 0.0986, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 21:32:55,419 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1287804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:33:08,932 INFO [train.py:968] (1/2) Epoch 29, batch 12000, giga_loss[loss=0.3046, simple_loss=0.3642, pruned_loss=0.1225, over 28744.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3724, pruned_loss=0.1219, over 5642238.43 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3424, pruned_loss=0.08865, over 5722199.18 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3752, pruned_loss=0.1251, over 5638667.68 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:33:08,933 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 21:33:17,848 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 21:33:48,478 INFO [optim.py:369] (1/2) 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 21:33:53,966 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-14 21:34:01,787 INFO [train.py:968] (1/2) Epoch 29, batch 12050, giga_loss[loss=0.3139, simple_loss=0.3721, pruned_loss=0.1279, over 28716.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3705, pruned_loss=0.1206, over 5662174.69 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.0884, over 5727784.29 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3743, pruned_loss=0.1244, over 5652201.76 frames. ], batch size: 92, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:34:50,223 INFO [train.py:968] (1/2) Epoch 29, batch 12100, giga_loss[loss=0.3156, simple_loss=0.3817, pruned_loss=0.1247, over 28888.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3732, pruned_loss=0.1227, over 5643578.25 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.0885, over 5726765.91 frames. ], giga_tot_loss[loss=0.3142, simple_loss=0.3763, pruned_loss=0.1261, over 5635535.83 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:34:56,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2748, 1.8483, 1.6241, 1.5181], device='cuda:1'), covar=tensor([0.2286, 0.1609, 0.2322, 0.2003], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0764, 0.0736, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-14 21:35:11,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6797, 1.8364, 1.3046, 1.4204], device='cuda:1'), covar=tensor([0.1028, 0.0641, 0.1039, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 21:35:26,338 INFO [optim.py:369] (1/2) 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 21:35:36,116 INFO [train.py:968] (1/2) Epoch 29, batch 12150, giga_loss[loss=0.3064, simple_loss=0.3705, pruned_loss=0.1212, over 28225.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3721, pruned_loss=0.1214, over 5658013.60 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3424, pruned_loss=0.08854, over 5729812.89 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3754, pruned_loss=0.1253, over 5645772.06 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:36:13,168 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-14 21:36:21,252 INFO [train.py:968] (1/2) Epoch 29, batch 12200, giga_loss[loss=0.2849, simple_loss=0.3586, pruned_loss=0.1056, over 28496.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3703, pruned_loss=0.1202, over 5668802.86 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3422, pruned_loss=0.08845, over 5725715.66 frames. ], giga_tot_loss[loss=0.3114, simple_loss=0.3739, pruned_loss=0.1245, over 5661428.17 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:36:21,571 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3866, 1.7268, 1.4069, 0.9524], device='cuda:1'), covar=tensor([0.2428, 0.2454, 0.2815, 0.2345], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1163, 0.1426, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 21:36:24,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4282, 3.6061, 1.5584, 1.5920], device='cuda:1'), covar=tensor([0.1035, 0.0362, 0.0927, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0578, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 21:36:27,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5371, 1.8732, 1.4354, 1.5498], device='cuda:1'), covar=tensor([0.2608, 0.2610, 0.3006, 0.2388], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1164, 0.1427, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 21:36:56,344 INFO [optim.py:369] (1/2) 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 21:36:58,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3761, 3.6087, 1.4348, 1.5670], device='cuda:1'), covar=tensor([0.1226, 0.0434, 0.1044, 0.1572], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0577, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 21:37:09,506 INFO [train.py:968] (1/2) Epoch 29, batch 12250, giga_loss[loss=0.2805, simple_loss=0.355, pruned_loss=0.103, over 29113.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3713, pruned_loss=0.1215, over 5661183.45 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3423, pruned_loss=0.08851, over 5721808.76 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3746, pruned_loss=0.1254, over 5657896.70 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:37:19,917 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9074, 2.9382, 1.8994, 1.0615], device='cuda:1'), covar=tensor([0.8514, 0.3493, 0.4114, 0.7563], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1752, 0.1671, 0.1515], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 21:37:36,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5943, 1.8560, 1.4491, 1.7846], device='cuda:1'), covar=tensor([0.2803, 0.2927, 0.3279, 0.2746], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1164, 0.1426, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 21:37:41,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3150, 1.6279, 1.3120, 0.9474], device='cuda:1'), covar=tensor([0.2295, 0.2297, 0.2560, 0.2239], device='cuda:1'), in_proj_covar=tensor([0.1613, 0.1163, 0.1425, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 21:37:57,424 INFO [train.py:968] (1/2) Epoch 29, batch 12300, giga_loss[loss=0.2701, simple_loss=0.346, pruned_loss=0.09707, over 28864.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3724, pruned_loss=0.1218, over 5669135.82 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3425, pruned_loss=0.08858, over 5727718.89 frames. ], giga_tot_loss[loss=0.3137, simple_loss=0.3757, pruned_loss=0.1258, over 5659574.33 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:38:32,754 INFO [optim.py:369] (1/2) 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 21:38:43,528 INFO [train.py:968] (1/2) Epoch 29, batch 12350, giga_loss[loss=0.3132, simple_loss=0.3765, pruned_loss=0.125, over 28867.00 frames. ], tot_loss[loss=0.3107, simple_loss=0.3742, pruned_loss=0.1236, over 5662681.11 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3427, pruned_loss=0.0887, over 5726872.79 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3769, pruned_loss=0.1269, over 5655328.65 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:38:52,417 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 12400, giga_loss[loss=0.2822, simple_loss=0.3519, pruned_loss=0.1062, over 28705.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3724, pruned_loss=0.1213, over 5673850.31 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3427, pruned_loss=0.08867, over 5728750.35 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3748, pruned_loss=0.1243, over 5665803.22 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:40:04,519 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1288250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:40:12,001 INFO [optim.py:369] (1/2) 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 21:40:21,510 INFO [train.py:968] (1/2) Epoch 29, batch 12450, giga_loss[loss=0.2736, simple_loss=0.3544, pruned_loss=0.09637, over 28924.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3722, pruned_loss=0.1201, over 5667335.97 frames. ], libri_tot_loss[loss=0.2607, simple_loss=0.3434, pruned_loss=0.08903, over 5724000.02 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3744, pruned_loss=0.1232, over 5663543.86 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:40:46,254 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 12500, giga_loss[loss=0.3042, simple_loss=0.3753, pruned_loss=0.1166, over 29028.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3708, pruned_loss=0.1185, over 5677763.17 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3439, pruned_loss=0.08934, over 5728664.98 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3735, pruned_loss=0.1221, over 5667929.41 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:41:04,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1288322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:41:05,992 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288325.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:41:36,918 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:41:41,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3446, 3.1052, 1.4945, 1.4600], device='cuda:1'), covar=tensor([0.0980, 0.0389, 0.0889, 0.1307], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0415, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 21:41:42,324 INFO [optim.py:369] (1/2) 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 21:41:53,830 INFO [train.py:968] (1/2) Epoch 29, batch 12550, giga_loss[loss=0.3104, simple_loss=0.3752, pruned_loss=0.1228, over 29070.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3701, pruned_loss=0.1186, over 5670710.17 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.344, pruned_loss=0.08943, over 5730459.81 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3724, pruned_loss=0.1217, over 5661122.29 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:42:01,351 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1288379.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 21:42:21,091 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6338, 1.9101, 1.7423, 1.6771], device='cuda:1'), covar=tensor([0.2223, 0.2249, 0.2510, 0.2333], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0767, 0.0739, 0.0708], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-14 21:42:43,594 INFO [train.py:968] (1/2) Epoch 29, batch 12600, giga_loss[loss=0.3908, simple_loss=0.4173, pruned_loss=0.1822, over 26629.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.37, pruned_loss=0.1197, over 5653647.69 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3442, pruned_loss=0.08953, over 5719758.68 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.372, pruned_loss=0.1224, over 5653930.41 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:43:11,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3333, 1.6188, 1.4820, 1.5640], device='cuda:1'), covar=tensor([0.0729, 0.0404, 0.0328, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 21:43:16,844 INFO [optim.py:369] (1/2) 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 21:43:27,089 INFO [train.py:968] (1/2) Epoch 29, batch 12650, giga_loss[loss=0.2962, simple_loss=0.357, pruned_loss=0.1177, over 28264.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3658, pruned_loss=0.1173, over 5667273.46 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3439, pruned_loss=0.08933, over 5724793.94 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3682, pruned_loss=0.1202, over 5661914.90 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:44:15,584 INFO [train.py:968] (1/2) Epoch 29, batch 12700, giga_loss[loss=0.2918, simple_loss=0.3486, pruned_loss=0.1175, over 28991.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.362, pruned_loss=0.1154, over 5680072.03 frames. ], libri_tot_loss[loss=0.2616, simple_loss=0.3441, pruned_loss=0.08955, over 5727462.63 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3641, pruned_loss=0.118, over 5672489.81 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:44:52,575 INFO [optim.py:369] (1/2) 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 21:45:01,898 INFO [train.py:968] (1/2) Epoch 29, batch 12750, giga_loss[loss=0.2722, simple_loss=0.3381, pruned_loss=0.1032, over 28987.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.36, pruned_loss=0.1143, over 5687983.90 frames. ], libri_tot_loss[loss=0.2617, simple_loss=0.3443, pruned_loss=0.08961, over 5729714.39 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3617, pruned_loss=0.1167, over 5679402.73 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 21:45:28,701 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4027, 1.3576, 1.1280, 1.5608], device='cuda:1'), covar=tensor([0.0726, 0.0414, 0.0379, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 21:45:50,266 INFO [train.py:968] (1/2) Epoch 29, batch 12800, giga_loss[loss=0.3352, simple_loss=0.3874, pruned_loss=0.1416, over 28773.00 frames. ], tot_loss[loss=0.2927, simple_loss=0.359, pruned_loss=0.1133, over 5687288.12 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3438, pruned_loss=0.08937, over 5733987.69 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3612, pruned_loss=0.116, over 5675641.05 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:45:53,999 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1288625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:46:06,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4349, 1.8325, 1.3847, 1.5340], device='cuda:1'), covar=tensor([0.2922, 0.2844, 0.3351, 0.2495], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1165, 0.1431, 0.1022], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 21:46:27,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2493, 4.0977, 3.8691, 1.8607], device='cuda:1'), covar=tensor([0.0687, 0.0823, 0.0897, 0.2130], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.1233, 0.1035, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-14 21:46:27,406 INFO [optim.py:369] (1/2) 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 21:46:35,325 INFO [train.py:968] (1/2) Epoch 29, batch 12850, giga_loss[loss=0.2823, simple_loss=0.3583, pruned_loss=0.1032, over 28991.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3578, pruned_loss=0.1103, over 5688304.17 frames. ], libri_tot_loss[loss=0.2614, simple_loss=0.3438, pruned_loss=0.08945, over 5737769.03 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 5674207.98 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:46:41,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1288677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:47:27,016 INFO [train.py:968] (1/2) Epoch 29, batch 12900, giga_loss[loss=0.2844, simple_loss=0.3626, pruned_loss=0.1032, over 28673.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3557, pruned_loss=0.1072, over 5682282.87 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3435, pruned_loss=0.08935, over 5741659.66 frames. ], giga_tot_loss[loss=0.2889, simple_loss=0.358, pruned_loss=0.1099, over 5666713.60 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:48:01,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1288754.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 21:48:07,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6520, 1.9089, 1.6036, 1.6625], device='cuda:1'), covar=tensor([0.2655, 0.2527, 0.2775, 0.2413], device='cuda:1'), in_proj_covar=tensor([0.1616, 0.1163, 0.1430, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 21:48:07,996 INFO [optim.py:369] (1/2) 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 21:48:17,335 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 12950, giga_loss[loss=0.2263, simple_loss=0.3222, pruned_loss=0.06525, over 28894.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3535, pruned_loss=0.1048, over 5676270.15 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3435, pruned_loss=0.08939, over 5742490.37 frames. ], giga_tot_loss[loss=0.2846, simple_loss=0.3553, pruned_loss=0.1069, over 5663098.34 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:48:21,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:48:48,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7585, 2.1006, 2.1653, 1.7907], device='cuda:1'), covar=tensor([0.2360, 0.2480, 0.1853, 0.2310], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0760, 0.0731, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 21:48:51,006 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:10,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1288820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:10,800 INFO [train.py:968] (1/2) Epoch 29, batch 13000, giga_loss[loss=0.2753, simple_loss=0.3344, pruned_loss=0.1081, over 26578.00 frames. ], tot_loss[loss=0.2775, simple_loss=0.3506, pruned_loss=0.1022, over 5663084.21 frames. ], libri_tot_loss[loss=0.2613, simple_loss=0.3435, pruned_loss=0.08959, over 5736905.46 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3523, pruned_loss=0.104, over 5655814.62 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:49:12,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:40,756 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:49:49,519 INFO [optim.py:369] (1/2) 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 21:50:01,427 INFO [train.py:968] (1/2) Epoch 29, batch 13050, giga_loss[loss=0.2539, simple_loss=0.3462, pruned_loss=0.08079, over 28595.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3491, pruned_loss=0.09945, over 5665179.25 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3433, pruned_loss=0.08949, over 5739789.55 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3507, pruned_loss=0.1011, over 5656175.50 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:50:25,987 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1288897.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 21:50:28,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1288900.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 21:50:47,633 INFO [train.py:968] (1/2) Epoch 29, batch 13100, giga_loss[loss=0.2687, simple_loss=0.3468, pruned_loss=0.09531, over 28628.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3488, pruned_loss=0.09803, over 5666001.73 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3428, pruned_loss=0.0894, over 5742005.98 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3506, pruned_loss=0.09974, over 5654482.46 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:50:56,567 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1288929.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 21:51:30,294 INFO [optim.py:369] (1/2) 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 21:51:39,938 INFO [train.py:968] (1/2) Epoch 29, batch 13150, giga_loss[loss=0.26, simple_loss=0.344, pruned_loss=0.08807, over 28620.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3483, pruned_loss=0.09783, over 5663519.17 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3424, pruned_loss=0.08928, over 5741665.85 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3502, pruned_loss=0.09936, over 5653954.95 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:51:40,786 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-14 21:52:02,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3608, 3.1981, 3.0352, 1.4547], device='cuda:1'), covar=tensor([0.1023, 0.1181, 0.1096, 0.2319], device='cuda:1'), in_proj_covar=tensor([0.1316, 0.1217, 0.1021, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 21:52:14,318 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:52:31,998 INFO [train.py:968] (1/2) Epoch 29, batch 13200, giga_loss[loss=0.2419, simple_loss=0.3226, pruned_loss=0.08058, over 28767.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3462, pruned_loss=0.09659, over 5659091.10 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3424, pruned_loss=0.08929, over 5738698.52 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3477, pruned_loss=0.0978, over 5653706.32 frames. ], batch size: 243, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:53:12,635 INFO [optim.py:369] (1/2) 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 21:53:16,970 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-14 21:53:22,271 INFO [train.py:968] (1/2) Epoch 29, batch 13250, giga_loss[loss=0.3108, simple_loss=0.3609, pruned_loss=0.1303, over 26624.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3438, pruned_loss=0.09527, over 5669852.89 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3414, pruned_loss=0.08898, over 5742951.80 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3459, pruned_loss=0.09668, over 5660144.64 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:53:25,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3355, 3.0059, 1.4503, 1.5296], device='cuda:1'), covar=tensor([0.0992, 0.0386, 0.0961, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0576, 0.0413, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 21:53:48,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4900, 1.7198, 1.2176, 1.3016], device='cuda:1'), covar=tensor([0.1059, 0.0548, 0.1063, 0.1156], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0450, 0.0522, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 21:54:13,738 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289120.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:54:14,121 INFO [train.py:968] (1/2) Epoch 29, batch 13300, giga_loss[loss=0.2518, simple_loss=0.3338, pruned_loss=0.08495, over 28697.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3437, pruned_loss=0.09466, over 5666670.61 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3414, pruned_loss=0.08898, over 5742951.80 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3453, pruned_loss=0.09575, over 5659114.55 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:54:54,274 INFO [optim.py:369] (1/2) 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 21:54:58,378 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3235, 3.2910, 1.4727, 1.4790], device='cuda:1'), covar=tensor([0.1075, 0.0422, 0.1025, 0.1465], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0575, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 21:55:03,308 INFO [train.py:968] (1/2) Epoch 29, batch 13350, giga_loss[loss=0.2626, simple_loss=0.3409, pruned_loss=0.0922, over 28773.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3429, pruned_loss=0.09402, over 5658475.50 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3413, pruned_loss=0.08897, over 5736507.49 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3444, pruned_loss=0.09502, over 5655769.24 frames. ], batch size: 243, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:55:47,450 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9476, 1.2018, 1.1482, 0.9893], device='cuda:1'), covar=tensor([0.2381, 0.2448, 0.1574, 0.1998], device='cuda:1'), in_proj_covar=tensor([0.2055, 0.2026, 0.1920, 0.2068], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 21:55:52,059 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:55:55,735 INFO [train.py:968] (1/2) Epoch 29, batch 13400, giga_loss[loss=0.2362, simple_loss=0.3197, pruned_loss=0.07638, over 28948.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.341, pruned_loss=0.09209, over 5666498.01 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3411, pruned_loss=0.08891, over 5739248.05 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3423, pruned_loss=0.09299, over 5661169.57 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:55:58,467 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:56:15,446 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-14 21:56:18,837 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3160, 2.5807, 1.3009, 1.4252], device='cuda:1'), covar=tensor([0.1026, 0.0392, 0.1004, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0575, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 21:56:21,429 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4448, 2.0222, 1.8469, 1.6611], device='cuda:1'), covar=tensor([0.2449, 0.2121, 0.2175, 0.2228], device='cuda:1'), in_proj_covar=tensor([0.0506, 0.0754, 0.0725, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 21:56:37,455 INFO [optim.py:369] (1/2) 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 21:56:44,436 INFO [train.py:968] (1/2) Epoch 29, batch 13450, libri_loss[loss=0.2491, simple_loss=0.3256, pruned_loss=0.08624, over 20382.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3377, pruned_loss=0.09, over 5660979.01 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3407, pruned_loss=0.08883, over 5735245.56 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3391, pruned_loss=0.09085, over 5659132.10 frames. ], batch size: 187, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:57:22,531 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3827, 1.2325, 1.2031, 1.6001], device='cuda:1'), covar=tensor([0.0745, 0.0410, 0.0377, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 21:57:34,166 INFO [train.py:968] (1/2) Epoch 29, batch 13500, giga_loss[loss=0.2928, simple_loss=0.349, pruned_loss=0.1183, over 26621.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3355, pruned_loss=0.0897, over 5648974.19 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3402, pruned_loss=0.08874, over 5735827.88 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.337, pruned_loss=0.09053, over 5643651.89 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:58:07,779 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:58:14,839 INFO [optim.py:369] (1/2) 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 21:58:24,230 INFO [train.py:968] (1/2) Epoch 29, batch 13550, giga_loss[loss=0.2621, simple_loss=0.3432, pruned_loss=0.09047, over 28699.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3341, pruned_loss=0.08971, over 5646152.10 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3395, pruned_loss=0.08843, over 5737435.29 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3358, pruned_loss=0.09067, over 5639100.23 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 21:58:33,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:59:14,397 INFO [train.py:968] (1/2) Epoch 29, batch 13600, giga_loss[loss=0.3181, simple_loss=0.3858, pruned_loss=0.1252, over 28307.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3346, pruned_loss=0.0905, over 5644156.66 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3381, pruned_loss=0.08788, over 5744841.49 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.337, pruned_loss=0.09188, over 5627923.27 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 21:59:14,577 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 21:59:47,509 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.47 vs. limit=2.0
+2023-03-14 21:59:59,920 INFO [optim.py:369] (1/2) 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 22:00:10,105 INFO [train.py:968] (1/2) Epoch 29, batch 13650, giga_loss[loss=0.2808, simple_loss=0.3468, pruned_loss=0.1074, over 26900.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3369, pruned_loss=0.09059, over 5649725.24 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3377, pruned_loss=0.08765, over 5746226.43 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3391, pruned_loss=0.09194, over 5633827.68 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:00:36,651 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4294, 4.2849, 4.0806, 2.0153], device='cuda:1'), covar=tensor([0.0599, 0.0685, 0.0790, 0.1941], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1206, 0.1010, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 22:00:38,905 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:00:39,784 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:06,546 INFO [train.py:968] (1/2) Epoch 29, batch 13700, giga_loss[loss=0.2544, simple_loss=0.337, pruned_loss=0.0859, over 27639.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.339, pruned_loss=0.09129, over 5654432.71 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3379, pruned_loss=0.08792, over 5749888.77 frames. ], giga_tot_loss[loss=0.2625, simple_loss=0.3407, pruned_loss=0.0922, over 5635574.21 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:01:08,851 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:10,818 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:44,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:01:54,589 INFO [optim.py:369] (1/2) 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 22:02:02,175 INFO [train.py:968] (1/2) Epoch 29, batch 13750, giga_loss[loss=0.2458, simple_loss=0.3279, pruned_loss=0.08189, over 28727.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3396, pruned_loss=0.09179, over 5645990.56 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3378, pruned_loss=0.08796, over 5748216.48 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3411, pruned_loss=0.09263, over 5629060.77 frames. ], batch size: 243, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:02:25,205 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:02:32,155 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:02:59,020 INFO [train.py:968] (1/2) Epoch 29, batch 13800, giga_loss[loss=0.2667, simple_loss=0.3478, pruned_loss=0.09283, over 28865.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3374, pruned_loss=0.09033, over 5652362.75 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3374, pruned_loss=0.08783, over 5742158.64 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.339, pruned_loss=0.09117, over 5642479.37 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:03:22,017 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:03:24,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2757, 1.2469, 1.1629, 1.5411], device='cuda:1'), covar=tensor([0.0800, 0.0364, 0.0376, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-14 22:03:26,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:03:49,192 INFO [optim.py:369] (1/2) 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 22:03:59,275 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:03:59,592 INFO [train.py:968] (1/2) Epoch 29, batch 13850, giga_loss[loss=0.2623, simple_loss=0.3334, pruned_loss=0.09562, over 26910.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3373, pruned_loss=0.08953, over 5647421.32 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3372, pruned_loss=0.08783, over 5746400.49 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3386, pruned_loss=0.09024, over 5633354.01 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:04:59,338 INFO [train.py:968] (1/2) Epoch 29, batch 13900, giga_loss[loss=0.224, simple_loss=0.3026, pruned_loss=0.07271, over 28984.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3351, pruned_loss=0.08828, over 5656172.06 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3367, pruned_loss=0.08763, over 5749030.06 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.3367, pruned_loss=0.08906, over 5640862.67 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:05:10,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:16,330 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:19,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:21,816 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1289738.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:05:25,100 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:29,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:50,442 INFO [optim.py:369] (1/2) 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 22:05:55,639 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:05:59,085 INFO [train.py:968] (1/2) Epoch 29, batch 13950, giga_loss[loss=0.254, simple_loss=0.3293, pruned_loss=0.08938, over 27791.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.333, pruned_loss=0.08799, over 5658904.39 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3369, pruned_loss=0.08785, over 5750906.42 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.334, pruned_loss=0.08839, over 5643884.57 frames. ], batch size: 474, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:06:01,633 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:06:15,718 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1935, 1.7022, 1.6563, 1.4706], device='cuda:1'), covar=tensor([0.2299, 0.1858, 0.2090, 0.1944], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0753, 0.0724, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 22:06:27,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:06:53,249 INFO [train.py:968] (1/2) Epoch 29, batch 14000, giga_loss[loss=0.2772, simple_loss=0.3543, pruned_loss=0.1, over 28325.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3321, pruned_loss=0.08769, over 5664351.65 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3364, pruned_loss=0.08776, over 5753684.37 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3332, pruned_loss=0.08809, over 5646950.28 frames. ], batch size: 368, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:07:29,416 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6442, 2.0491, 1.3262, 1.0227], device='cuda:1'), covar=tensor([0.7818, 0.4445, 0.4121, 0.7061], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1741, 0.1666, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 22:07:40,190 INFO [optim.py:369] (1/2) 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 22:07:50,132 INFO [train.py:968] (1/2) Epoch 29, batch 14050, giga_loss[loss=0.264, simple_loss=0.3534, pruned_loss=0.08737, over 28693.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3325, pruned_loss=0.08691, over 5675754.40 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3363, pruned_loss=0.08773, over 5755957.76 frames. ], giga_tot_loss[loss=0.2539, simple_loss=0.3334, pruned_loss=0.08725, over 5658830.07 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:07:50,562 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1289871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:07:53,424 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1289873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:07:57,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:08:31,102 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:08:48,802 INFO [train.py:968] (1/2) Epoch 29, batch 14100, giga_loss[loss=0.2504, simple_loss=0.3375, pruned_loss=0.08165, over 28450.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3349, pruned_loss=0.08704, over 5683754.28 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3363, pruned_loss=0.08771, over 5759220.02 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3356, pruned_loss=0.08731, over 5665529.48 frames. ], batch size: 85, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:09:11,100 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1289942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:09:33,586 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 22:09:36,638 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 14150, giga_loss[loss=0.2219, simple_loss=0.3056, pruned_loss=0.06917, over 29041.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3316, pruned_loss=0.08525, over 5668099.14 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3356, pruned_loss=0.08749, over 5742313.24 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3327, pruned_loss=0.08563, over 5665287.01 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:09:47,604 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1289971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:10:44,327 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:10:46,897 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 14200, giga_loss[loss=0.2806, simple_loss=0.3632, pruned_loss=0.09898, over 28855.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3328, pruned_loss=0.08637, over 5676535.96 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3356, pruned_loss=0.08749, over 5744948.94 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3336, pruned_loss=0.08664, over 5670597.54 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:11:03,858 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290030.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:11:23,820 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:11:30,434 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 29, batch 14250, giga_loss[loss=0.255, simple_loss=0.3487, pruned_loss=0.08069, over 29087.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.3361, pruned_loss=0.08803, over 5656765.95 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3357, pruned_loss=0.08758, over 5742472.32 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3367, pruned_loss=0.08815, over 5652389.28 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:12:51,711 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 14300, giga_loss[loss=0.2827, simple_loss=0.3542, pruned_loss=0.1055, over 26800.00 frames. ], tot_loss[loss=0.256, simple_loss=0.3386, pruned_loss=0.08669, over 5657851.33 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3353, pruned_loss=0.08745, over 5745648.83 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3394, pruned_loss=0.0869, over 5650079.54 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:13:13,960 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:13:50,316 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5466, 1.7981, 1.4700, 1.5386], device='cuda:1'), covar=tensor([0.3191, 0.3167, 0.3612, 0.2907], device='cuda:1'), in_proj_covar=tensor([0.1616, 0.1164, 0.1432, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 22:13:55,257 INFO [optim.py:369] (1/2) 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 22:14:02,817 INFO [train.py:968] (1/2) Epoch 29, batch 14350, giga_loss[loss=0.2517, simple_loss=0.3446, pruned_loss=0.07942, over 28654.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3391, pruned_loss=0.08588, over 5645535.60 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3354, pruned_loss=0.08756, over 5746404.58 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3396, pruned_loss=0.08593, over 5638239.51 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:14:10,346 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6169, 1.8503, 1.5325, 1.4850], device='cuda:1'), covar=tensor([0.2771, 0.2810, 0.3176, 0.2631], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1162, 0.1430, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 22:14:29,401 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2812, 1.6547, 1.3114, 0.9046], device='cuda:1'), covar=tensor([0.2680, 0.2620, 0.3063, 0.2678], device='cuda:1'), in_proj_covar=tensor([0.1614, 0.1161, 0.1429, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 22:14:39,632 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2992, 2.9784, 1.4127, 1.4728], device='cuda:1'), covar=tensor([0.1024, 0.0313, 0.0965, 0.1362], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0572, 0.0413, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 22:14:46,191 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 22:14:58,830 INFO [train.py:968] (1/2) Epoch 29, batch 14400, giga_loss[loss=0.2771, simple_loss=0.356, pruned_loss=0.09913, over 28422.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3393, pruned_loss=0.08502, over 5663548.46 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.335, pruned_loss=0.08729, over 5751271.37 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3402, pruned_loss=0.08524, over 5650723.14 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:15:45,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290256.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:15:50,192 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:16:01,180 INFO [train.py:968] (1/2) Epoch 29, batch 14450, giga_loss[loss=0.2765, simple_loss=0.3521, pruned_loss=0.1005, over 28979.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3408, pruned_loss=0.08697, over 5670135.70 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3351, pruned_loss=0.08732, over 5753673.76 frames. ], giga_tot_loss[loss=0.2579, simple_loss=0.3416, pruned_loss=0.08712, over 5656165.77 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:16:23,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290288.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:16:36,295 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4631, 1.4223, 1.6544, 1.1967], device='cuda:1'), covar=tensor([0.1895, 0.3655, 0.1574, 0.1919], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0714, 0.0983, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 22:17:04,862 INFO [train.py:968] (1/2) Epoch 29, batch 14500, giga_loss[loss=0.2852, simple_loss=0.3598, pruned_loss=0.1053, over 28937.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3394, pruned_loss=0.08774, over 5668287.89 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3348, pruned_loss=0.08716, over 5754813.73 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3404, pruned_loss=0.08801, over 5654601.40 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:17:17,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:18:02,366 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.89 vs. limit=5.0
+2023-03-14 22:18:11,652 INFO [optim.py:369] (1/2) 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 22:18:22,366 INFO [train.py:968] (1/2) Epoch 29, batch 14550, giga_loss[loss=0.2234, simple_loss=0.3139, pruned_loss=0.06641, over 29037.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3388, pruned_loss=0.08763, over 5673210.04 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3346, pruned_loss=0.08705, over 5756866.07 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3398, pruned_loss=0.08794, over 5659569.29 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:18:45,087 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:19:16,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290405.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:19:35,653 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3646, 1.6719, 1.6075, 1.2000], device='cuda:1'), covar=tensor([0.1918, 0.2795, 0.1633, 0.1844], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0714, 0.0984, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 22:19:37,216 INFO [train.py:968] (1/2) Epoch 29, batch 14600, giga_loss[loss=0.2184, simple_loss=0.302, pruned_loss=0.06736, over 28840.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3326, pruned_loss=0.08397, over 5682780.05 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3341, pruned_loss=0.08688, over 5761119.77 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.334, pruned_loss=0.08434, over 5665937.92 frames. ], batch size: 112, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:19:42,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:20:30,622 INFO [optim.py:369] (1/2) 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 22:20:35,958 INFO [train.py:968] (1/2) Epoch 29, batch 14650, giga_loss[loss=0.2488, simple_loss=0.3299, pruned_loss=0.08384, over 28945.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3306, pruned_loss=0.08316, over 5672731.65 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3335, pruned_loss=0.08661, over 5758592.18 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.332, pruned_loss=0.08361, over 5659170.99 frames. ], batch size: 213, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:21:27,696 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:21:47,180 INFO [train.py:968] (1/2) Epoch 29, batch 14700, giga_loss[loss=0.2532, simple_loss=0.338, pruned_loss=0.08423, over 28050.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3289, pruned_loss=0.08273, over 5675813.89 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3335, pruned_loss=0.08661, over 5758592.18 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.33, pruned_loss=0.08308, over 5665259.46 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:22:16,248 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290548.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:22:18,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290551.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:22:36,646 INFO [optim.py:369] (1/2) 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 22:22:41,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:22:43,872 INFO [train.py:968] (1/2) Epoch 29, batch 14750, giga_loss[loss=0.2711, simple_loss=0.3494, pruned_loss=0.09635, over 28836.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.334, pruned_loss=0.08549, over 5691010.37 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3329, pruned_loss=0.08631, over 5763187.68 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3355, pruned_loss=0.08595, over 5675522.14 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:22:44,173 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:22:54,083 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290580.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:23:16,158 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:23:37,837 INFO [train.py:968] (1/2) Epoch 29, batch 14800, giga_loss[loss=0.2689, simple_loss=0.3449, pruned_loss=0.09644, over 28497.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3334, pruned_loss=0.08605, over 5682589.16 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3319, pruned_loss=0.08583, over 5761108.92 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3355, pruned_loss=0.08685, over 5669278.61 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 22:23:43,149 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4873, 3.4647, 1.5290, 1.6351], device='cuda:1'), covar=tensor([0.0995, 0.0378, 0.0953, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0573, 0.0415, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 22:23:54,703 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-14 22:24:11,497 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:24:15,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:24:29,255 INFO [optim.py:369] (1/2) 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 22:24:35,312 INFO [train.py:968] (1/2) Epoch 29, batch 14850, giga_loss[loss=0.2313, simple_loss=0.3158, pruned_loss=0.07342, over 28621.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3332, pruned_loss=0.08725, over 5680287.00 frames. ], libri_tot_loss[loss=0.2517, simple_loss=0.3317, pruned_loss=0.0858, over 5760602.99 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3351, pruned_loss=0.08797, over 5667402.17 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:24:47,982 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:25:14,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:25:16,955 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-14 22:25:31,866 INFO [train.py:968] (1/2) Epoch 29, batch 14900, giga_loss[loss=0.2603, simple_loss=0.3374, pruned_loss=0.09164, over 29034.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3339, pruned_loss=0.08831, over 5680388.31 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3313, pruned_loss=0.08568, over 5762351.25 frames. ], giga_tot_loss[loss=0.2571, simple_loss=0.336, pruned_loss=0.08911, over 5665033.42 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:26:22,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1290762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:26:28,569 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 14950, giga_loss[loss=0.2941, simple_loss=0.3786, pruned_loss=0.1048, over 29001.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3345, pruned_loss=0.08787, over 5681796.61 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3308, pruned_loss=0.0854, over 5765394.06 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3367, pruned_loss=0.08882, over 5664985.61 frames. ], batch size: 186, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:26:50,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290784.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:27:09,092 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:27:29,933 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:27:41,300 INFO [train.py:968] (1/2) Epoch 29, batch 15000, giga_loss[loss=0.3165, simple_loss=0.3694, pruned_loss=0.1319, over 26785.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3374, pruned_loss=0.08893, over 5679408.06 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3307, pruned_loss=0.08551, over 5765083.09 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3393, pruned_loss=0.08965, over 5664446.84 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:27:41,300 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 22:27:50,053 INFO [train.py:1012] (1/2) Epoch 29, validation: loss=0.1935, simple_loss=0.2949, pruned_loss=0.04606, over 944034.00 frames. 
+2023-03-14 22:27:50,054 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 22:28:30,026 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290847.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:28:34,227 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1290850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:29:00,820 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 15050, giga_loss[loss=0.235, simple_loss=0.3185, pruned_loss=0.07569, over 28606.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3363, pruned_loss=0.08818, over 5680319.89 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3308, pruned_loss=0.08554, over 5766681.00 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3378, pruned_loss=0.08874, over 5666446.29 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:29:19,102 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:29:55,416 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1290905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:29:59,063 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 15100, giga_loss[loss=0.2184, simple_loss=0.2996, pruned_loss=0.06865, over 28662.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3306, pruned_loss=0.08547, over 5692998.38 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3307, pruned_loss=0.08553, over 5766236.00 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.332, pruned_loss=0.08595, over 5680498.81 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:30:22,353 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1290927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:30:35,341 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1290937.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:30:41,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2967, 1.5981, 1.5181, 1.1953], device='cuda:1'), covar=tensor([0.1773, 0.2704, 0.1500, 0.1823], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0712, 0.0982, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 22:31:10,788 INFO [optim.py:369] (1/2) 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 22:31:16,681 INFO [train.py:968] (1/2) Epoch 29, batch 15150, giga_loss[loss=0.263, simple_loss=0.3354, pruned_loss=0.09533, over 27559.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3256, pruned_loss=0.08337, over 5687285.38 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3305, pruned_loss=0.08549, over 5767263.16 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3268, pruned_loss=0.08379, over 5674155.87 frames. ], batch size: 472, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:32:10,946 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:32:15,871 INFO [train.py:968] (1/2) Epoch 29, batch 15200, giga_loss[loss=0.2244, simple_loss=0.3049, pruned_loss=0.07193, over 28278.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3275, pruned_loss=0.08503, over 5682240.33 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3306, pruned_loss=0.08557, over 5766695.87 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3283, pruned_loss=0.08527, over 5671658.55 frames. ], batch size: 77, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:32:41,035 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2063, 2.5018, 1.2242, 1.3907], device='cuda:1'), covar=tensor([0.1006, 0.0420, 0.0962, 0.1436], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0572, 0.0413, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 22:33:05,630 INFO [optim.py:369] (1/2) 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 22:33:09,901 INFO [train.py:968] (1/2) Epoch 29, batch 15250, giga_loss[loss=0.2302, simple_loss=0.3163, pruned_loss=0.07209, over 28830.00 frames. ], tot_loss[loss=0.2507, simple_loss=0.3291, pruned_loss=0.0862, over 5673036.74 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3308, pruned_loss=0.08581, over 5760066.22 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3295, pruned_loss=0.08618, over 5667625.21 frames. ], batch size: 119, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:34:08,282 INFO [train.py:968] (1/2) Epoch 29, batch 15300, giga_loss[loss=0.244, simple_loss=0.3322, pruned_loss=0.07787, over 28932.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3263, pruned_loss=0.08416, over 5662488.12 frames. ], libri_tot_loss[loss=0.251, simple_loss=0.3306, pruned_loss=0.08569, over 5759719.79 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3268, pruned_loss=0.08424, over 5656912.75 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:34:35,068 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-14 22:34:56,260 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291159.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:35:04,784 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 15350, giga_loss[loss=0.2435, simple_loss=0.3232, pruned_loss=0.08195, over 28167.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.3259, pruned_loss=0.08329, over 5665925.53 frames. ], libri_tot_loss[loss=0.2504, simple_loss=0.3299, pruned_loss=0.08541, over 5761616.52 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3268, pruned_loss=0.08357, over 5658269.53 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:35:12,671 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:35:36,105 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:36:17,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6387, 1.8663, 1.5383, 1.6056], device='cuda:1'), covar=tensor([0.2773, 0.2805, 0.3221, 0.2521], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1162, 0.1433, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 22:36:17,509 INFO [train.py:968] (1/2) Epoch 29, batch 15400, giga_loss[loss=0.216, simple_loss=0.2988, pruned_loss=0.0666, over 28973.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3253, pruned_loss=0.08343, over 5666921.02 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3306, pruned_loss=0.08584, over 5761573.90 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3253, pruned_loss=0.08321, over 5659511.79 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:37:16,029 INFO [optim.py:369] (1/2) 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 22:37:23,344 INFO [train.py:968] (1/2) Epoch 29, batch 15450, giga_loss[loss=0.2571, simple_loss=0.3292, pruned_loss=0.09248, over 26901.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.326, pruned_loss=0.08308, over 5683676.11 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3302, pruned_loss=0.08563, over 5764722.63 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3262, pruned_loss=0.08303, over 5673586.29 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:37:33,187 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.13 vs. limit=5.0
+2023-03-14 22:37:57,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:37:57,571 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291302.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:38:00,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291305.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:38:10,499 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:38:18,473 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 15500, giga_loss[loss=0.2451, simple_loss=0.3257, pruned_loss=0.08227, over 28876.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.3263, pruned_loss=0.08326, over 5689438.83 frames. ], libri_tot_loss[loss=0.2505, simple_loss=0.3299, pruned_loss=0.08556, over 5762332.30 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3266, pruned_loss=0.08317, over 5679582.27 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:38:31,766 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:38:34,932 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291334.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:38:54,931 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:39:10,280 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:39:18,273 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 15550, giga_loss[loss=0.2236, simple_loss=0.3075, pruned_loss=0.06984, over 28446.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3272, pruned_loss=0.08425, over 5692881.40 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3299, pruned_loss=0.08562, over 5765847.23 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3273, pruned_loss=0.0841, over 5679965.24 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:39:27,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2127, 1.3130, 3.2197, 2.9449], device='cuda:1'), covar=tensor([0.1538, 0.2724, 0.0572, 0.1772], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0676, 0.1006, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 22:39:27,621 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.51 vs. limit=5.0
+2023-03-14 22:39:44,741 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:40:22,285 INFO [train.py:968] (1/2) Epoch 29, batch 15600, giga_loss[loss=0.2188, simple_loss=0.3073, pruned_loss=0.06515, over 28946.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.325, pruned_loss=0.0829, over 5678443.58 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.33, pruned_loss=0.08579, over 5760922.91 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3249, pruned_loss=0.08259, over 5670324.86 frames. ], batch size: 227, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:40:36,474 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291434.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 22:40:50,371 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:40:53,402 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:41:13,439 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-14 22:41:13,611 INFO [optim.py:369] (1/2) 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 22:41:17,259 INFO [train.py:968] (1/2) Epoch 29, batch 15650, giga_loss[loss=0.2619, simple_loss=0.3472, pruned_loss=0.08832, over 28521.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.326, pruned_loss=0.08213, over 5671722.63 frames. ], libri_tot_loss[loss=0.2507, simple_loss=0.3299, pruned_loss=0.08581, over 5764435.35 frames. ], giga_tot_loss[loss=0.2447, simple_loss=0.3259, pruned_loss=0.08173, over 5659389.51 frames. ], batch size: 336, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:41:17,680 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:41:24,346 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:41:30,396 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3608, 1.5027, 3.8446, 3.3430], device='cuda:1'), covar=tensor([0.1649, 0.2701, 0.0488, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0806, 0.0676, 0.1004, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 22:42:13,670 INFO [train.py:968] (1/2) Epoch 29, batch 15700, giga_loss[loss=0.2402, simple_loss=0.3297, pruned_loss=0.07531, over 28947.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3284, pruned_loss=0.08294, over 5668578.94 frames. ], libri_tot_loss[loss=0.2501, simple_loss=0.3292, pruned_loss=0.08545, over 5764829.48 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3288, pruned_loss=0.08284, over 5655327.27 frames. ], batch size: 285, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:42:25,822 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:42:28,745 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:42:43,098 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1291548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:42:59,822 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:43:03,532 INFO [optim.py:369] (1/2) 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 22:43:06,942 INFO [train.py:968] (1/2) Epoch 29, batch 15750, giga_loss[loss=0.259, simple_loss=0.3331, pruned_loss=0.09242, over 26914.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.33, pruned_loss=0.08365, over 5663796.61 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3286, pruned_loss=0.08523, over 5759019.80 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.331, pruned_loss=0.0837, over 5655925.05 frames. ], batch size: 555, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:44:00,634 INFO [train.py:968] (1/2) Epoch 29, batch 15800, giga_loss[loss=0.2362, simple_loss=0.3029, pruned_loss=0.0847, over 24551.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3304, pruned_loss=0.08478, over 5658873.56 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.328, pruned_loss=0.08511, over 5761599.21 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3318, pruned_loss=0.08488, over 5646377.15 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:44:16,025 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-14 22:44:44,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4634, 1.1377, 4.5596, 3.6725], device='cuda:1'), covar=tensor([0.1707, 0.3156, 0.0379, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0675, 0.1002, 0.0976], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 22:44:54,012 INFO [optim.py:369] (1/2) 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 22:44:55,867 INFO [train.py:968] (1/2) Epoch 29, batch 15850, giga_loss[loss=0.2416, simple_loss=0.3128, pruned_loss=0.0852, over 28780.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3291, pruned_loss=0.08416, over 5665476.77 frames. ], libri_tot_loss[loss=0.249, simple_loss=0.3277, pruned_loss=0.08511, over 5765176.18 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3305, pruned_loss=0.08424, over 5649860.36 frames. ], batch size: 99, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:45:55,290 INFO [train.py:968] (1/2) Epoch 29, batch 15900, giga_loss[loss=0.2257, simple_loss=0.3123, pruned_loss=0.06953, over 28906.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3272, pruned_loss=0.08311, over 5664656.99 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3278, pruned_loss=0.08524, over 5765149.79 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3282, pruned_loss=0.08293, over 5646924.66 frames. ], batch size: 284, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:46:14,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6734, 2.5584, 1.5734, 0.9023], device='cuda:1'), covar=tensor([0.9414, 0.4461, 0.4510, 0.7964], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1738, 0.1661, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 22:46:43,964 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:46:52,403 INFO [train.py:968] (1/2) Epoch 29, batch 15950, giga_loss[loss=0.2429, simple_loss=0.317, pruned_loss=0.08439, over 29109.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3257, pruned_loss=0.08286, over 5673082.60 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.328, pruned_loss=0.08533, over 5765605.20 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3263, pruned_loss=0.0826, over 5657414.50 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:47:34,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291809.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:47:50,548 INFO [train.py:968] (1/2) Epoch 29, batch 16000, giga_loss[loss=0.2369, simple_loss=0.3313, pruned_loss=0.07125, over 28804.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3263, pruned_loss=0.08269, over 5676627.81 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3281, pruned_loss=0.0854, over 5767060.24 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3266, pruned_loss=0.08236, over 5660879.38 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:47:59,243 INFO [scaling.py:679] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:48:46,702 INFO [optim.py:369] (1/2) 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 22:48:48,218 INFO [train.py:968] (1/2) Epoch 29, batch 16050, giga_loss[loss=0.2431, simple_loss=0.336, pruned_loss=0.0751, over 28682.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3279, pruned_loss=0.08306, over 5677532.05 frames. ], libri_tot_loss[loss=0.2488, simple_loss=0.3276, pruned_loss=0.08502, over 5767108.37 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3287, pruned_loss=0.08312, over 5662147.78 frames. ], batch size: 307, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:49:12,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2635, 2.3515, 1.3506, 1.3325], device='cuda:1'), covar=tensor([0.0965, 0.0425, 0.0950, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0570, 0.0413, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 22:49:49,692 INFO [train.py:968] (1/2) Epoch 29, batch 16100, giga_loss[loss=0.2243, simple_loss=0.3094, pruned_loss=0.06958, over 29085.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.3276, pruned_loss=0.08371, over 5661442.68 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3271, pruned_loss=0.08473, over 5761039.06 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3287, pruned_loss=0.08397, over 5650980.17 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:49:52,491 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1291923.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:50:22,832 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291952.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:50:24,848 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 29, batch 16150, giga_loss[loss=0.2566, simple_loss=0.3345, pruned_loss=0.0893, over 28944.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3299, pruned_loss=0.08497, over 5669668.82 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3263, pruned_loss=0.08436, over 5765580.95 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3315, pruned_loss=0.0855, over 5654307.60 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:50:47,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4443, 1.5771, 1.5855, 1.4258], device='cuda:1'), covar=tensor([0.2780, 0.2460, 0.2133, 0.2443], device='cuda:1'), in_proj_covar=tensor([0.2033, 0.1989, 0.1884, 0.2037], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 22:50:57,466 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1291984.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 22:50:57,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5189, 1.3319, 1.6260, 1.2050], device='cuda:1'), covar=tensor([0.2197, 0.3531, 0.1715, 0.1983], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0708, 0.0979, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 22:51:03,125 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1291989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:51:06,177 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1291992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:51:40,928 INFO [train.py:968] (1/2) Epoch 29, batch 16200, giga_loss[loss=0.2742, simple_loss=0.3617, pruned_loss=0.09328, over 28034.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3337, pruned_loss=0.08676, over 5658049.67 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3265, pruned_loss=0.08445, over 5767292.29 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.335, pruned_loss=0.08714, over 5642634.14 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:51:41,284 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1292021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:52:29,950 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:52:34,103 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1292069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:52:35,379 INFO [train.py:968] (1/2) Epoch 29, batch 16250, giga_loss[loss=0.2047, simple_loss=0.3025, pruned_loss=0.0535, over 29056.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3335, pruned_loss=0.0859, over 5651997.73 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3261, pruned_loss=0.08429, over 5761427.81 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3351, pruned_loss=0.08641, over 5641196.25 frames. ], batch size: 136, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:53:07,799 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1292098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:53:11,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1737, 1.6655, 1.2180, 0.5693], device='cuda:1'), covar=tensor([0.5273, 0.2729, 0.3877, 0.6703], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1737, 0.1661, 0.1504], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 22:53:39,111 INFO [train.py:968] (1/2) Epoch 29, batch 16300, giga_loss[loss=0.2261, simple_loss=0.3182, pruned_loss=0.06704, over 29000.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3335, pruned_loss=0.08656, over 5648371.44 frames. ], libri_tot_loss[loss=0.2476, simple_loss=0.3261, pruned_loss=0.08451, over 5756293.09 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3351, pruned_loss=0.08684, over 5640261.86 frames. ], batch size: 155, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:54:02,224 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 22:54:39,922 INFO [train.py:968] (1/2) Epoch 29, batch 16350, giga_loss[loss=0.2317, simple_loss=0.2926, pruned_loss=0.08536, over 24436.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3326, pruned_loss=0.08666, over 5660403.53 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3259, pruned_loss=0.0844, over 5758568.43 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3341, pruned_loss=0.08703, over 5650085.65 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:55:08,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8549, 2.2529, 1.8480, 2.0110], device='cuda:1'), covar=tensor([0.0689, 0.0264, 0.0295, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 22:55:19,839 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5563, 1.7595, 1.7675, 1.3547], device='cuda:1'), covar=tensor([0.2011, 0.2834, 0.1678, 0.2016], device='cuda:1'), in_proj_covar=tensor([0.0933, 0.0709, 0.0979, 0.0882], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 22:55:40,709 INFO [train.py:968] (1/2) Epoch 29, batch 16400, giga_loss[loss=0.2285, simple_loss=0.3132, pruned_loss=0.07188, over 29075.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3314, pruned_loss=0.08565, over 5669566.54 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3256, pruned_loss=0.08412, over 5758731.33 frames. ], giga_tot_loss[loss=0.2528, simple_loss=0.3331, pruned_loss=0.08626, over 5658147.94 frames. ], batch size: 128, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:55:41,268 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3212, 1.2646, 3.6173, 3.2471], device='cuda:1'), covar=tensor([0.1578, 0.2812, 0.0481, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0674, 0.1002, 0.0975], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 22:56:21,792 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-14 22:56:42,076 INFO [optim.py:369] (1/2) 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 22:56:42,919 INFO [train.py:968] (1/2) Epoch 29, batch 16450, giga_loss[loss=0.2451, simple_loss=0.3176, pruned_loss=0.08625, over 28982.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3309, pruned_loss=0.08636, over 5657458.97 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3255, pruned_loss=0.08412, over 5750967.49 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3325, pruned_loss=0.08684, over 5654276.59 frames. ], batch size: 120, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:56:54,708 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1292281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:56:58,995 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1292284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:57:03,885 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4288, 2.9249, 1.5257, 1.5660], device='cuda:1'), covar=tensor([0.0949, 0.0390, 0.0940, 0.1285], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0569, 0.0413, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 22:57:19,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:57:31,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1292313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 22:57:41,591 INFO [train.py:968] (1/2) Epoch 29, batch 16500, giga_loss[loss=0.2812, simple_loss=0.3517, pruned_loss=0.1054, over 28798.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3278, pruned_loss=0.08523, over 5659290.84 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3248, pruned_loss=0.08375, over 5754744.26 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3297, pruned_loss=0.086, over 5650905.49 frames. ], batch size: 243, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 22:58:08,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2880, 1.1177, 4.0264, 3.3031], device='cuda:1'), covar=tensor([0.1744, 0.3009, 0.0453, 0.0935], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0675, 0.1003, 0.0977], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 22:58:35,713 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0355, 3.1730, 2.0141, 1.1474], device='cuda:1'), covar=tensor([0.8761, 0.3626, 0.4627, 0.7685], device='cuda:1'), in_proj_covar=tensor([0.1845, 0.1737, 0.1662, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 22:58:43,387 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 16550, giga_loss[loss=0.2555, simple_loss=0.3434, pruned_loss=0.08382, over 28794.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3265, pruned_loss=0.08369, over 5663023.34 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3246, pruned_loss=0.0837, over 5758485.11 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3282, pruned_loss=0.08436, over 5650901.62 frames. ], batch size: 243, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 22:58:52,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5546, 1.7529, 1.2031, 1.3586], device='cuda:1'), covar=tensor([0.1053, 0.0556, 0.1088, 0.1124], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0447, 0.0523, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 22:59:34,511 INFO [train.py:968] (1/2) Epoch 29, batch 16600, giga_loss[loss=0.2034, simple_loss=0.3096, pruned_loss=0.04856, over 28872.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3255, pruned_loss=0.08207, over 5674533.02 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3244, pruned_loss=0.08357, over 5757448.22 frames. ], giga_tot_loss[loss=0.2463, simple_loss=0.3272, pruned_loss=0.08271, over 5661932.39 frames. ], batch size: 164, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:00:29,768 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 23:00:32,740 INFO [optim.py:369] (1/2) 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 23:00:32,752 INFO [train.py:968] (1/2) Epoch 29, batch 16650, giga_loss[loss=0.2449, simple_loss=0.3357, pruned_loss=0.07705, over 28719.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3272, pruned_loss=0.08053, over 5685008.62 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3243, pruned_loss=0.08356, over 5759728.28 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3287, pruned_loss=0.081, over 5671933.70 frames. ], batch size: 242, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:00:52,314 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-14 23:01:22,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-14 23:01:30,317 INFO [train.py:968] (1/2) Epoch 29, batch 16700, giga_loss[loss=0.2325, simple_loss=0.3219, pruned_loss=0.07157, over 28909.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3286, pruned_loss=0.08052, over 5682757.86 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3244, pruned_loss=0.08361, over 5762061.53 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3297, pruned_loss=0.0808, over 5669107.10 frames. ], batch size: 174, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:02:25,762 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 16750, giga_loss[loss=0.251, simple_loss=0.3335, pruned_loss=0.08427, over 29000.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3279, pruned_loss=0.0803, over 5682910.48 frames. ], libri_tot_loss[loss=0.2453, simple_loss=0.324, pruned_loss=0.08336, over 5767182.69 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3294, pruned_loss=0.08062, over 5664217.06 frames. ], batch size: 199, lr: 1.10e-03, grad_scale: 2.0
+2023-03-14 23:03:35,922 INFO [train.py:968] (1/2) Epoch 29, batch 16800, giga_loss[loss=0.2388, simple_loss=0.3277, pruned_loss=0.07501, over 28682.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3283, pruned_loss=0.08056, over 5669489.43 frames. ], libri_tot_loss[loss=0.2453, simple_loss=0.324, pruned_loss=0.08336, over 5767182.69 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3294, pruned_loss=0.08081, over 5654940.11 frames. ], batch size: 262, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:04:44,773 INFO [optim.py:369] (1/2) 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 23:04:44,785 INFO [train.py:968] (1/2) Epoch 29, batch 16850, giga_loss[loss=0.2343, simple_loss=0.3235, pruned_loss=0.07255, over 28454.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3288, pruned_loss=0.08088, over 5658608.87 frames. ], libri_tot_loss[loss=0.2453, simple_loss=0.3239, pruned_loss=0.08334, over 5756778.56 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3299, pruned_loss=0.08107, over 5653914.85 frames. ], batch size: 369, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:04:45,413 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7248, 2.0878, 1.4788, 1.6170], device='cuda:1'), covar=tensor([0.1133, 0.0589, 0.1020, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0446, 0.0522, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 23:04:55,195 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1292679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:05:46,546 INFO [train.py:968] (1/2) Epoch 29, batch 16900, giga_loss[loss=0.2368, simple_loss=0.3267, pruned_loss=0.07344, over 28111.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3289, pruned_loss=0.08053, over 5659789.87 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3237, pruned_loss=0.08312, over 5759933.54 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3302, pruned_loss=0.08079, over 5648372.08 frames. ], batch size: 412, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:06:03,472 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-14 23:06:50,713 INFO [optim.py:369] (1/2) 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 23:06:50,725 INFO [train.py:968] (1/2) Epoch 29, batch 16950, giga_loss[loss=0.2761, simple_loss=0.3653, pruned_loss=0.09342, over 28792.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3317, pruned_loss=0.08191, over 5665204.25 frames. ], libri_tot_loss[loss=0.2455, simple_loss=0.3241, pruned_loss=0.08341, over 5762680.51 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3325, pruned_loss=0.0818, over 5650792.52 frames. ], batch size: 243, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:07:04,872 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:07:32,307 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:07:57,237 INFO [train.py:968] (1/2) Epoch 29, batch 17000, libri_loss[loss=0.2459, simple_loss=0.3351, pruned_loss=0.07833, over 29522.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3346, pruned_loss=0.08327, over 5670349.82 frames. ], libri_tot_loss[loss=0.2453, simple_loss=0.3241, pruned_loss=0.08327, over 5764717.44 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3356, pruned_loss=0.0833, over 5654828.69 frames. ], batch size: 80, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:07:59,672 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1292822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:08:02,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1292825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:08:39,412 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) 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 23:08:59,372 INFO [train.py:968] (1/2) Epoch 29, batch 17050, giga_loss[loss=0.3402, simple_loss=0.381, pruned_loss=0.1497, over 24381.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3336, pruned_loss=0.08305, over 5681918.39 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3238, pruned_loss=0.08309, over 5766564.28 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3349, pruned_loss=0.08323, over 5665381.73 frames. ], batch size: 705, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:09:46,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4682, 1.5259, 1.2143, 1.1725], device='cuda:1'), covar=tensor([0.0788, 0.0352, 0.0841, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0446, 0.0523, 0.0457], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 23:10:07,910 INFO [train.py:968] (1/2) Epoch 29, batch 17100, giga_loss[loss=0.2127, simple_loss=0.3041, pruned_loss=0.06063, over 28931.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3321, pruned_loss=0.083, over 5676533.76 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3237, pruned_loss=0.08303, over 5760026.74 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3334, pruned_loss=0.08319, over 5667700.32 frames. ], batch size: 145, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:10:14,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4197, 1.8376, 1.4006, 1.5369], device='cuda:1'), covar=tensor([0.0778, 0.0322, 0.0357, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0121, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 23:10:26,870 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:10:49,010 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1292950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:11:16,655 INFO [optim.py:369] (1/2) 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 23:11:16,668 INFO [train.py:968] (1/2) Epoch 29, batch 17150, giga_loss[loss=0.2368, simple_loss=0.3246, pruned_loss=0.0745, over 28529.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3317, pruned_loss=0.08218, over 5675947.32 frames. ], libri_tot_loss[loss=0.2454, simple_loss=0.3242, pruned_loss=0.0833, over 5762679.26 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3324, pruned_loss=0.08208, over 5665178.25 frames. ], batch size: 370, lr: 1.10e-03, grad_scale: 4.0
+2023-03-14 23:11:20,252 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4924, 1.9636, 1.7451, 1.6274], device='cuda:1'), covar=tensor([0.2359, 0.2614, 0.2415, 0.2588], device='cuda:1'), in_proj_covar=tensor([0.0488, 0.0739, 0.0712, 0.0679], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 23:12:14,626 INFO [train.py:968] (1/2) Epoch 29, batch 17200, libri_loss[loss=0.241, simple_loss=0.3169, pruned_loss=0.08252, over 29553.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3303, pruned_loss=0.08158, over 5680252.67 frames. ], libri_tot_loss[loss=0.2451, simple_loss=0.324, pruned_loss=0.08308, over 5765709.29 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3315, pruned_loss=0.08165, over 5665168.96 frames. ], batch size: 79, lr: 1.10e-03, grad_scale: 8.0
+2023-03-14 23:13:10,001 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 17250, giga_loss[loss=0.2721, simple_loss=0.3588, pruned_loss=0.09269, over 29025.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.3318, pruned_loss=0.08268, over 5673783.09 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3238, pruned_loss=0.08298, over 5758783.38 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.333, pruned_loss=0.08281, over 5666034.02 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:13:31,560 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1293091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:14:04,957 INFO [train.py:968] (1/2) Epoch 29, batch 17300, giga_loss[loss=0.2212, simple_loss=0.3122, pruned_loss=0.06507, over 28870.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3343, pruned_loss=0.08413, over 5671984.88 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3238, pruned_loss=0.08309, over 5761138.72 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3355, pruned_loss=0.08415, over 5661801.10 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:14:20,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3649, 1.5993, 1.6258, 1.2208], device='cuda:1'), covar=tensor([0.1914, 0.2753, 0.1621, 0.1931], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0708, 0.0981, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 23:14:21,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3076, 1.6378, 1.6874, 1.4706], device='cuda:1'), covar=tensor([0.1903, 0.1422, 0.1556, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0486, 0.0737, 0.0710, 0.0677], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 23:14:41,072 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:14:51,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4674, 3.5019, 1.5782, 1.6101], device='cuda:1'), covar=tensor([0.1024, 0.0521, 0.0973, 0.1409], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0569, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 23:14:57,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2850, 1.7411, 1.7757, 1.5130], device='cuda:1'), covar=tensor([0.2361, 0.1944, 0.2104, 0.2008], device='cuda:1'), in_proj_covar=tensor([0.0487, 0.0737, 0.0711, 0.0678], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 23:14:58,157 INFO [train.py:968] (1/2) Epoch 29, batch 17350, giga_loss[loss=0.2244, simple_loss=0.3071, pruned_loss=0.07088, over 28921.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3325, pruned_loss=0.08411, over 5678144.54 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3237, pruned_loss=0.08307, over 5764718.90 frames. ], giga_tot_loss[loss=0.2511, simple_loss=0.3338, pruned_loss=0.08417, over 5664698.60 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:15:00,378 INFO [optim.py:369] (1/2) 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 23:15:04,770 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:15:52,300 INFO [train.py:968] (1/2) Epoch 29, batch 17400, giga_loss[loss=0.2699, simple_loss=0.3492, pruned_loss=0.09531, over 28719.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3301, pruned_loss=0.08406, over 5672656.06 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3235, pruned_loss=0.08304, over 5767822.69 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3315, pruned_loss=0.08417, over 5656047.33 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:16:18,120 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 17450, giga_loss[loss=0.278, simple_loss=0.3541, pruned_loss=0.101, over 28964.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3305, pruned_loss=0.08478, over 5662039.53 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3237, pruned_loss=0.08309, over 5761601.55 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3316, pruned_loss=0.08486, over 5651376.78 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:16:48,018 INFO [optim.py:369] (1/2) 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 23:17:19,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:21,955 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:44,379 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:45,264 INFO [train.py:968] (1/2) Epoch 29, batch 17500, giga_loss[loss=0.2898, simple_loss=0.3766, pruned_loss=0.1015, over 29092.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3386, pruned_loss=0.08929, over 5664373.53 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3238, pruned_loss=0.08313, over 5761458.81 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3395, pruned_loss=0.08936, over 5655098.61 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:17:46,921 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:17:49,376 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293331.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:18:11,352 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:18:29,225 INFO [train.py:968] (1/2) Epoch 29, batch 17550, giga_loss[loss=0.2637, simple_loss=0.3472, pruned_loss=0.09008, over 28787.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3454, pruned_loss=0.09264, over 5670088.32 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3238, pruned_loss=0.08308, over 5761313.75 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3466, pruned_loss=0.09291, over 5660717.73 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:18:29,824 INFO [optim.py:369] (1/2) 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 23:18:43,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:18:46,946 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:12,835 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:13,889 INFO [train.py:968] (1/2) Epoch 29, batch 17600, giga_loss[loss=0.2551, simple_loss=0.3276, pruned_loss=0.09125, over 28462.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3472, pruned_loss=0.09409, over 5669681.59 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3238, pruned_loss=0.08303, over 5762361.47 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3484, pruned_loss=0.0945, over 5660053.89 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:19:37,981 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4508, 1.5946, 1.2845, 1.1670], device='cuda:1'), covar=tensor([0.1162, 0.0616, 0.1139, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0445, 0.0522, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 23:19:46,475 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:49,263 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:56,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1293466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:58,166 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:19:59,867 INFO [train.py:968] (1/2) Epoch 29, batch 17650, giga_loss[loss=0.221, simple_loss=0.3029, pruned_loss=0.06959, over 28772.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.3427, pruned_loss=0.09269, over 5675560.56 frames. ], libri_tot_loss[loss=0.2449, simple_loss=0.3238, pruned_loss=0.08298, over 5764830.49 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.344, pruned_loss=0.09321, over 5664463.89 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:20:00,139 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:20:00,496 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:20:23,070 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1293500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:20:40,218 INFO [train.py:968] (1/2) Epoch 29, batch 17700, giga_loss[loss=0.2569, simple_loss=0.3254, pruned_loss=0.09422, over 27642.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.335, pruned_loss=0.08898, over 5687953.30 frames. ], libri_tot_loss[loss=0.2451, simple_loss=0.324, pruned_loss=0.08309, over 5766666.84 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3363, pruned_loss=0.08955, over 5674520.08 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:20:40,605 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0601, 2.4236, 2.0317, 2.3851], device='cuda:1'), covar=tensor([0.2545, 0.2592, 0.3111, 0.2485], device='cuda:1'), in_proj_covar=tensor([0.1604, 0.1155, 0.1423, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 23:21:19,250 INFO [train.py:968] (1/2) Epoch 29, batch 17750, giga_loss[loss=0.2053, simple_loss=0.2892, pruned_loss=0.06072, over 28854.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3283, pruned_loss=0.08601, over 5699229.24 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.3236, pruned_loss=0.08284, over 5771389.68 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3301, pruned_loss=0.08691, over 5680567.58 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:21:20,575 INFO [optim.py:369] (1/2) 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 23:21:52,543 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1293609.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:21:55,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1293612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:22:04,482 INFO [train.py:968] (1/2) Epoch 29, batch 17800, giga_loss[loss=0.2014, simple_loss=0.2764, pruned_loss=0.06326, over 28936.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.3216, pruned_loss=0.08342, over 5693085.30 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3237, pruned_loss=0.0828, over 5765708.04 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3229, pruned_loss=0.08421, over 5680761.20 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:22:15,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.24 vs. limit=5.0
+2023-03-14 23:22:18,654 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 17850, giga_loss[loss=0.1998, simple_loss=0.2769, pruned_loss=0.06137, over 29041.00 frames. ], tot_loss[loss=0.2376, simple_loss=0.3148, pruned_loss=0.08019, over 5695195.86 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3236, pruned_loss=0.08252, over 5770075.13 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3158, pruned_loss=0.08105, over 5679398.03 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:22:46,259 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 17900, giga_loss[loss=0.2532, simple_loss=0.3251, pruned_loss=0.09063, over 28949.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.3113, pruned_loss=0.07892, over 5696952.74 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3236, pruned_loss=0.08253, over 5769660.11 frames. ], giga_tot_loss[loss=0.2355, simple_loss=0.3119, pruned_loss=0.07952, over 5683443.62 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:24:08,446 INFO [train.py:968] (1/2) Epoch 29, batch 17950, giga_loss[loss=0.1978, simple_loss=0.2789, pruned_loss=0.05837, over 29062.00 frames. ], tot_loss[loss=0.2328, simple_loss=0.3091, pruned_loss=0.07819, over 5704015.63 frames. ], libri_tot_loss[loss=0.2444, simple_loss=0.3238, pruned_loss=0.08256, over 5771786.22 frames. ], giga_tot_loss[loss=0.2332, simple_loss=0.3092, pruned_loss=0.07858, over 5690543.96 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:24:09,652 INFO [optim.py:369] (1/2) 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 23:24:14,966 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5679, 2.1027, 1.5116, 0.8512], device='cuda:1'), covar=tensor([0.7471, 0.3329, 0.4764, 0.7823], device='cuda:1'), in_proj_covar=tensor([0.1842, 0.1743, 0.1665, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 23:24:42,611 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-14 23:24:49,889 INFO [train.py:968] (1/2) Epoch 29, batch 18000, giga_loss[loss=0.2285, simple_loss=0.3039, pruned_loss=0.07655, over 28953.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.3071, pruned_loss=0.07751, over 5701243.02 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3241, pruned_loss=0.08261, over 5775563.68 frames. ], giga_tot_loss[loss=0.2308, simple_loss=0.3064, pruned_loss=0.07764, over 5684651.77 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:24:49,889 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-14 23:24:56,593 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3245, 1.2642, 1.1086, 1.5183], device='cuda:1'), covar=tensor([0.0850, 0.0399, 0.0406, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 23:24:58,147 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-14 23:25:23,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7266, 2.0379, 1.8702, 1.9233], device='cuda:1'), covar=tensor([0.0738, 0.0286, 0.0304, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 23:25:39,576 INFO [train.py:968] (1/2) Epoch 29, batch 18050, libri_loss[loss=0.2666, simple_loss=0.3531, pruned_loss=0.09004, over 25984.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.3048, pruned_loss=0.07623, over 5700334.07 frames. ], libri_tot_loss[loss=0.2453, simple_loss=0.3248, pruned_loss=0.08289, over 5773301.62 frames. ], giga_tot_loss[loss=0.2275, simple_loss=0.3032, pruned_loss=0.07596, over 5687599.56 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:25:40,962 INFO [optim.py:369] (1/2) 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 23:25:59,493 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 18100, giga_loss[loss=0.1778, simple_loss=0.2551, pruned_loss=0.05029, over 28501.00 frames. ], tot_loss[loss=0.226, simple_loss=0.302, pruned_loss=0.07502, over 5694214.07 frames. ], libri_tot_loss[loss=0.2453, simple_loss=0.3247, pruned_loss=0.08289, over 5764649.01 frames. ], giga_tot_loss[loss=0.2245, simple_loss=0.3, pruned_loss=0.07453, over 5689023.21 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:26:55,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5130, 1.5078, 1.5639, 1.1093], device='cuda:1'), covar=tensor([0.2083, 0.3766, 0.1789, 0.2035], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0714, 0.0990, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 23:26:58,365 INFO [train.py:968] (1/2) Epoch 29, batch 18150, giga_loss[loss=0.2096, simple_loss=0.2821, pruned_loss=0.06851, over 28746.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2998, pruned_loss=0.07397, over 5681206.37 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3253, pruned_loss=0.08315, over 5755193.38 frames. ], giga_tot_loss[loss=0.2217, simple_loss=0.2971, pruned_loss=0.07312, over 5684166.00 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:27:00,415 INFO [optim.py:369] (1/2) 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 23:27:43,264 INFO [train.py:968] (1/2) Epoch 29, batch 18200, giga_loss[loss=0.1862, simple_loss=0.2515, pruned_loss=0.06045, over 23987.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2965, pruned_loss=0.0727, over 5683591.85 frames. ], libri_tot_loss[loss=0.2456, simple_loss=0.3251, pruned_loss=0.083, over 5757377.50 frames. ], giga_tot_loss[loss=0.2191, simple_loss=0.2941, pruned_loss=0.07203, over 5683035.09 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:28:09,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7288, 1.8372, 1.9326, 1.5282], device='cuda:1'), covar=tensor([0.1998, 0.2696, 0.1619, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0717, 0.0994, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 23:28:27,678 INFO [train.py:968] (1/2) Epoch 29, batch 18250, giga_loss[loss=0.205, simple_loss=0.2873, pruned_loss=0.06136, over 29040.00 frames. ], tot_loss[loss=0.2189, simple_loss=0.2946, pruned_loss=0.07154, over 5700003.41 frames. ], libri_tot_loss[loss=0.246, simple_loss=0.3258, pruned_loss=0.08313, over 5761433.40 frames. ], giga_tot_loss[loss=0.2161, simple_loss=0.2912, pruned_loss=0.07056, over 5693972.10 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:28:29,506 INFO [optim.py:369] (1/2) 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 23:28:49,145 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:29:08,121 INFO [train.py:968] (1/2) Epoch 29, batch 18300, giga_loss[loss=0.259, simple_loss=0.3312, pruned_loss=0.09338, over 27842.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2943, pruned_loss=0.0718, over 5705243.40 frames. ], libri_tot_loss[loss=0.2461, simple_loss=0.3259, pruned_loss=0.08312, over 5766443.80 frames. ], giga_tot_loss[loss=0.2155, simple_loss=0.2901, pruned_loss=0.07049, over 5692914.91 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:29:58,301 INFO [train.py:968] (1/2) Epoch 29, batch 18350, libri_loss[loss=0.2477, simple_loss=0.3167, pruned_loss=0.08941, over 29347.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.3033, pruned_loss=0.07665, over 5698818.18 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3262, pruned_loss=0.08334, over 5759515.15 frames. ], giga_tot_loss[loss=0.2249, simple_loss=0.2992, pruned_loss=0.07527, over 5693874.75 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:30:01,374 INFO [optim.py:369] (1/2) 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 23:30:31,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294206.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:30:32,020 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:30:42,262 INFO [train.py:968] (1/2) Epoch 29, batch 18400, giga_loss[loss=0.358, simple_loss=0.4046, pruned_loss=0.1557, over 26588.00 frames. ], tot_loss[loss=0.241, simple_loss=0.3161, pruned_loss=0.08294, over 5699115.67 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3265, pruned_loss=0.08335, over 5760114.42 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.312, pruned_loss=0.08169, over 5692133.42 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:30:47,991 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4867, 1.7372, 1.6673, 1.6013], device='cuda:1'), covar=tensor([0.1690, 0.1467, 0.1596, 0.1490], device='cuda:1'), in_proj_covar=tensor([0.0495, 0.0750, 0.0720, 0.0687], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 23:31:22,615 INFO [train.py:968] (1/2) Epoch 29, batch 18450, giga_loss[loss=0.2615, simple_loss=0.3488, pruned_loss=0.08709, over 28681.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3266, pruned_loss=0.08822, over 5695247.51 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3267, pruned_loss=0.08358, over 5752954.24 frames. ], giga_tot_loss[loss=0.2486, simple_loss=0.323, pruned_loss=0.08707, over 5694579.09 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:31:24,698 INFO [optim.py:369] (1/2) 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 23:31:25,767 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1294275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:32:01,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4276, 1.8604, 1.3985, 0.7980], device='cuda:1'), covar=tensor([0.6519, 0.3174, 0.3705, 0.6690], device='cuda:1'), in_proj_covar=tensor([0.1842, 0.1738, 0.1663, 0.1502], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-14 23:32:04,567 INFO [train.py:968] (1/2) Epoch 29, batch 18500, giga_loss[loss=0.2901, simple_loss=0.3639, pruned_loss=0.1081, over 28846.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3337, pruned_loss=0.09076, over 5688781.66 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3265, pruned_loss=0.08342, over 5754343.43 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3311, pruned_loss=0.09005, over 5686498.25 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:32:30,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9109, 1.3152, 1.3030, 1.1152], device='cuda:1'), covar=tensor([0.2135, 0.1438, 0.2509, 0.1826], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0752, 0.0722, 0.0688], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 23:32:46,975 INFO [train.py:968] (1/2) Epoch 29, batch 18550, libri_loss[loss=0.2408, simple_loss=0.3295, pruned_loss=0.07606, over 29660.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3368, pruned_loss=0.09097, over 5693316.41 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3266, pruned_loss=0.08328, over 5757669.72 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3349, pruned_loss=0.09073, over 5686717.35 frames. ], batch size: 88, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:32:48,968 INFO [optim.py:369] (1/2) 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 23:33:25,305 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:33:27,945 INFO [train.py:968] (1/2) Epoch 29, batch 18600, libri_loss[loss=0.2377, simple_loss=0.3273, pruned_loss=0.07411, over 29257.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3379, pruned_loss=0.09057, over 5694731.85 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3273, pruned_loss=0.08341, over 5762054.80 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3362, pruned_loss=0.09061, over 5682447.04 frames. ], batch size: 94, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:33:28,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:33:30,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2407, 4.0588, 3.8266, 1.8534], device='cuda:1'), covar=tensor([0.0645, 0.0814, 0.0769, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1281, 0.1184, 0.0995, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-14 23:33:55,095 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 18650, giga_loss[loss=0.259, simple_loss=0.3356, pruned_loss=0.09119, over 28813.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3399, pruned_loss=0.09201, over 5697215.03 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3272, pruned_loss=0.08333, over 5763626.60 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3388, pruned_loss=0.09219, over 5685486.63 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:34:12,922 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:34:55,367 INFO [train.py:968] (1/2) Epoch 29, batch 18700, giga_loss[loss=0.2721, simple_loss=0.3497, pruned_loss=0.0972, over 29035.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3435, pruned_loss=0.09477, over 5699018.56 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3274, pruned_loss=0.08333, over 5765835.61 frames. ], giga_tot_loss[loss=0.2664, simple_loss=0.3427, pruned_loss=0.09503, over 5687080.00 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:35:29,828 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4902, 3.5959, 1.6172, 1.6858], device='cuda:1'), covar=tensor([0.1099, 0.0256, 0.0927, 0.1412], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0566, 0.0412, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-14 23:35:38,538 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2010, 1.3174, 3.6829, 3.1854], device='cuda:1'), covar=tensor([0.1744, 0.2918, 0.0434, 0.1193], device='cuda:1'), in_proj_covar=tensor([0.0803, 0.0673, 0.1004, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 23:35:39,609 INFO [train.py:968] (1/2) Epoch 29, batch 18750, giga_loss[loss=0.283, simple_loss=0.3579, pruned_loss=0.1041, over 29077.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3463, pruned_loss=0.0965, over 5702545.64 frames. ], libri_tot_loss[loss=0.2473, simple_loss=0.3278, pruned_loss=0.08344, over 5766445.08 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3456, pruned_loss=0.0968, over 5691601.09 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:35:43,341 INFO [optim.py:369] (1/2) 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 23:35:47,256 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1294581.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:35:48,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1294582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:06,404 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7343, 2.0129, 1.4429, 1.5699], device='cuda:1'), covar=tensor([0.1155, 0.0647, 0.1069, 0.1158], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0447, 0.0526, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 23:36:10,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2786, 1.4897, 1.3792, 1.4381], device='cuda:1'), covar=tensor([0.0876, 0.0365, 0.0362, 0.0966], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 23:36:16,503 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:18,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:19,844 INFO [train.py:968] (1/2) Epoch 29, batch 18800, giga_loss[loss=0.2499, simple_loss=0.338, pruned_loss=0.08093, over 28620.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.349, pruned_loss=0.09698, over 5708857.29 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.328, pruned_loss=0.08353, over 5769813.05 frames. ], giga_tot_loss[loss=0.2718, simple_loss=0.3486, pruned_loss=0.09744, over 5695808.11 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:36:29,213 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-14 23:36:41,274 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:36:42,443 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1294648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:37:02,531 INFO [train.py:968] (1/2) Epoch 29, batch 18850, giga_loss[loss=0.2859, simple_loss=0.3615, pruned_loss=0.1052, over 29117.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3512, pruned_loss=0.09757, over 5712193.77 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.328, pruned_loss=0.08346, over 5770490.98 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.351, pruned_loss=0.09804, over 5701186.76 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:37:06,214 INFO [optim.py:369] (1/2) 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 23:37:10,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294681.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:37:42,992 INFO [train.py:968] (1/2) Epoch 29, batch 18900, giga_loss[loss=0.2998, simple_loss=0.3798, pruned_loss=0.1099, over 28975.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3525, pruned_loss=0.09761, over 5706285.84 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3286, pruned_loss=0.08368, over 5770668.26 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3523, pruned_loss=0.09811, over 5695975.78 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:37:45,620 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294724.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:37:47,033 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1294725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:37:47,044 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6775, 1.8492, 1.6453, 1.5885], device='cuda:1'), covar=tensor([0.2315, 0.2248, 0.2252, 0.2153], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1160, 0.1423, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 23:37:48,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294727.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:37:49,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1294728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:38:11,161 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1294756.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:38:11,682 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1294757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:38:21,542 INFO [train.py:968] (1/2) Epoch 29, batch 18950, giga_loss[loss=0.25, simple_loss=0.3396, pruned_loss=0.08018, over 28635.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3516, pruned_loss=0.09601, over 5706286.96 frames. ], libri_tot_loss[loss=0.2481, simple_loss=0.3287, pruned_loss=0.0837, over 5772085.09 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3515, pruned_loss=0.0965, over 5696180.64 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:38:24,370 INFO [optim.py:369] (1/2) 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 23:38:53,328 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:39:01,747 INFO [train.py:968] (1/2) Epoch 29, batch 19000, giga_loss[loss=0.2878, simple_loss=0.3647, pruned_loss=0.1054, over 28856.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3497, pruned_loss=0.09403, over 5709416.69 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3291, pruned_loss=0.0838, over 5773680.65 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3497, pruned_loss=0.09453, over 5698793.37 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:39:11,271 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9356, 2.3477, 1.5812, 1.8756], device='cuda:1'), covar=tensor([0.1200, 0.0687, 0.1134, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0447, 0.0525, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-14 23:39:38,914 INFO [train.py:968] (1/2) Epoch 29, batch 19050, giga_loss[loss=0.281, simple_loss=0.362, pruned_loss=0.1, over 28627.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3483, pruned_loss=0.09283, over 5713991.58 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3292, pruned_loss=0.08376, over 5776715.26 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3487, pruned_loss=0.09348, over 5701420.17 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:39:43,214 INFO [optim.py:369] (1/2) 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 23:40:19,196 INFO [train.py:968] (1/2) Epoch 29, batch 19100, libri_loss[loss=0.2689, simple_loss=0.3535, pruned_loss=0.09222, over 28704.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3507, pruned_loss=0.09684, over 5700979.08 frames. ], libri_tot_loss[loss=0.2483, simple_loss=0.3293, pruned_loss=0.08365, over 5779701.92 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3517, pruned_loss=0.09788, over 5685574.00 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:41:04,580 INFO [train.py:968] (1/2) Epoch 29, batch 19150, giga_loss[loss=0.283, simple_loss=0.3529, pruned_loss=0.1065, over 28916.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.354, pruned_loss=0.1014, over 5688608.04 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3299, pruned_loss=0.08396, over 5772340.04 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3546, pruned_loss=0.1022, over 5680779.09 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:41:08,496 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 19200, giga_loss[loss=0.2736, simple_loss=0.3454, pruned_loss=0.101, over 29037.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3524, pruned_loss=0.1008, over 5698247.50 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3301, pruned_loss=0.08405, over 5772609.59 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3533, pruned_loss=0.1018, over 5689259.88 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:41:41,502 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:41:41,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:42:11,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295056.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:42:11,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6197, 1.8258, 1.7005, 1.5571], device='cuda:1'), covar=tensor([0.2243, 0.2320, 0.2530, 0.2359], device='cuda:1'), in_proj_covar=tensor([0.0499, 0.0753, 0.0723, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 23:42:26,374 INFO [train.py:968] (1/2) Epoch 29, batch 19250, giga_loss[loss=0.2408, simple_loss=0.3349, pruned_loss=0.07329, over 28618.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3504, pruned_loss=0.1006, over 5682712.95 frames. ], libri_tot_loss[loss=0.2492, simple_loss=0.3302, pruned_loss=0.08414, over 5752314.93 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3515, pruned_loss=0.1016, over 5691273.77 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:42:30,074 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 19300, giga_loss[loss=0.2983, simple_loss=0.3628, pruned_loss=0.1169, over 28264.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.3488, pruned_loss=0.09981, over 5688702.95 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3303, pruned_loss=0.08425, over 5756022.14 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.35, pruned_loss=0.1009, over 5690711.52 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:43:26,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-14 23:43:26,252 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-14 23:43:35,368 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5912, 1.8711, 1.5104, 1.6100], device='cuda:1'), covar=tensor([0.2740, 0.2816, 0.3177, 0.2426], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1159, 0.1421, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 23:43:42,666 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:43:45,457 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:43:48,260 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 19350, giga_loss[loss=0.251, simple_loss=0.3342, pruned_loss=0.08392, over 28952.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3479, pruned_loss=0.09859, over 5676677.52 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3305, pruned_loss=0.08426, over 5748556.31 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3489, pruned_loss=0.09967, over 5683290.12 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:43:54,856 INFO [optim.py:369] (1/2) 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 23:44:01,864 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:44:10,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:44:14,000 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295199.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:44:15,277 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2297, 1.2338, 3.4661, 3.0171], device='cuda:1'), covar=tensor([0.1592, 0.2824, 0.0429, 0.1528], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0674, 0.1007, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 23:44:15,854 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295202.0, num_to_drop=1, layers_to_drop={0}
+2023-03-14 23:44:34,037 INFO [train.py:968] (1/2) Epoch 29, batch 19400, giga_loss[loss=0.2609, simple_loss=0.3335, pruned_loss=0.09411, over 28617.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3453, pruned_loss=0.09641, over 5683877.25 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3306, pruned_loss=0.08419, over 5752738.40 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3464, pruned_loss=0.09764, over 5683817.66 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:44:43,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-14 23:45:15,242 INFO [train.py:968] (1/2) Epoch 29, batch 19450, giga_loss[loss=0.2709, simple_loss=0.3466, pruned_loss=0.09757, over 28608.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.34, pruned_loss=0.09348, over 5684779.85 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3305, pruned_loss=0.08419, over 5756443.79 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3413, pruned_loss=0.09472, over 5679811.67 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:45:21,256 INFO [optim.py:369] (1/2) 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 23:46:00,055 INFO [train.py:968] (1/2) Epoch 29, batch 19500, giga_loss[loss=0.2163, simple_loss=0.2987, pruned_loss=0.06697, over 28925.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.335, pruned_loss=0.09082, over 5687861.49 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3307, pruned_loss=0.08419, over 5760023.32 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.336, pruned_loss=0.092, over 5678825.22 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:46:06,263 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:46:09,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4512, 1.7328, 1.6674, 1.5299], device='cuda:1'), covar=tensor([0.2365, 0.2272, 0.2665, 0.2243], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0754, 0.0726, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-14 23:46:11,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:46:39,314 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 19550, giga_loss[loss=0.2247, simple_loss=0.3026, pruned_loss=0.07343, over 28691.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3314, pruned_loss=0.08938, over 5682133.43 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3317, pruned_loss=0.08474, over 5753032.18 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3313, pruned_loss=0.08995, over 5679075.33 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 2.0
+2023-03-14 23:46:53,586 INFO [optim.py:369] (1/2) 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 23:47:10,228 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 19600, libri_loss[loss=0.2614, simple_loss=0.3471, pruned_loss=0.08791, over 28553.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3316, pruned_loss=0.08892, over 5691668.15 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3323, pruned_loss=0.085, over 5757823.81 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3309, pruned_loss=0.08932, over 5681829.87 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:47:47,836 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3328, 1.3282, 1.1768, 1.4996], device='cuda:1'), covar=tensor([0.0791, 0.0380, 0.0380, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-14 23:47:55,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-14 23:48:09,841 INFO [train.py:968] (1/2) Epoch 29, batch 19650, giga_loss[loss=0.2347, simple_loss=0.3263, pruned_loss=0.07157, over 28896.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3314, pruned_loss=0.08783, over 5702472.22 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3321, pruned_loss=0.08475, over 5759544.06 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.331, pruned_loss=0.08849, over 5690932.71 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:48:15,112 INFO [optim.py:369] (1/2) 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 23:48:53,513 INFO [train.py:968] (1/2) Epoch 29, batch 19700, giga_loss[loss=0.2365, simple_loss=0.3062, pruned_loss=0.08338, over 28694.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3317, pruned_loss=0.08839, over 5703648.39 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3324, pruned_loss=0.08485, over 5761050.96 frames. ], giga_tot_loss[loss=0.2544, simple_loss=0.3311, pruned_loss=0.08885, over 5692633.68 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:48:53,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:00,133 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-14 23:49:01,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:07,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:09,705 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:49:11,544 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 19750, giga_loss[loss=0.2482, simple_loss=0.3234, pruned_loss=0.08653, over 28705.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3292, pruned_loss=0.08696, over 5716209.73 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3325, pruned_loss=0.08483, over 5763879.90 frames. ], giga_tot_loss[loss=0.2517, simple_loss=0.3286, pruned_loss=0.08739, over 5703947.63 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:49:34,976 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:50:16,111 INFO [train.py:968] (1/2) Epoch 29, batch 19800, giga_loss[loss=0.2478, simple_loss=0.32, pruned_loss=0.08781, over 28796.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.327, pruned_loss=0.08632, over 5721523.94 frames. ], libri_tot_loss[loss=0.2512, simple_loss=0.3327, pruned_loss=0.08486, over 5764282.50 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3264, pruned_loss=0.08664, over 5711377.86 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:50:32,893 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1295640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:50:57,684 INFO [train.py:968] (1/2) Epoch 29, batch 19850, giga_loss[loss=0.2558, simple_loss=0.3163, pruned_loss=0.09768, over 28799.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3248, pruned_loss=0.0857, over 5721454.01 frames. ], libri_tot_loss[loss=0.2518, simple_loss=0.3332, pruned_loss=0.08516, over 5764825.52 frames. ], giga_tot_loss[loss=0.2476, simple_loss=0.3238, pruned_loss=0.08569, over 5712515.02 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:50:58,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7466, 1.8839, 1.5493, 1.9123], device='cuda:1'), covar=tensor([0.2659, 0.2931, 0.3213, 0.2689], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1160, 0.1424, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 23:51:02,347 INFO [optim.py:369] (1/2) 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 23:51:03,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1295679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:51:06,376 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1295682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:51:28,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1295711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:51:34,982 INFO [train.py:968] (1/2) Epoch 29, batch 19900, giga_loss[loss=0.2728, simple_loss=0.3458, pruned_loss=0.09987, over 28704.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3227, pruned_loss=0.08459, over 5728109.85 frames. ], libri_tot_loss[loss=0.2521, simple_loss=0.3338, pruned_loss=0.0852, over 5768243.84 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3212, pruned_loss=0.08455, over 5717181.46 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:52:13,668 INFO [train.py:968] (1/2) Epoch 29, batch 19950, giga_loss[loss=0.2134, simple_loss=0.2903, pruned_loss=0.06824, over 28810.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3226, pruned_loss=0.08464, over 5724006.75 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3348, pruned_loss=0.08555, over 5767857.15 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3201, pruned_loss=0.08426, over 5714130.92 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:52:18,438 INFO [optim.py:369] (1/2) 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 23:52:50,239 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 20000, giga_loss[loss=0.2183, simple_loss=0.2948, pruned_loss=0.07087, over 28579.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3207, pruned_loss=0.08391, over 5720217.92 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3349, pruned_loss=0.08543, over 5765675.06 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3183, pruned_loss=0.08366, over 5713099.20 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:53:32,600 INFO [train.py:968] (1/2) Epoch 29, batch 20050, giga_loss[loss=0.2432, simple_loss=0.3144, pruned_loss=0.08594, over 28877.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3199, pruned_loss=0.0833, over 5720436.21 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3356, pruned_loss=0.08584, over 5759452.86 frames. ], giga_tot_loss[loss=0.2411, simple_loss=0.3169, pruned_loss=0.08269, over 5718489.12 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:53:36,937 INFO [optim.py:369] (1/2) 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 23:53:39,528 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2826, 1.4544, 3.8413, 3.3873], device='cuda:1'), covar=tensor([0.1728, 0.2717, 0.0416, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0673, 0.1003, 0.0978], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-14 23:53:51,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:53:56,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1295905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:54:09,298 INFO [train.py:968] (1/2) Epoch 29, batch 20100, giga_loss[loss=0.3324, simple_loss=0.3806, pruned_loss=0.1421, over 26540.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.3184, pruned_loss=0.08247, over 5715588.72 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3365, pruned_loss=0.0862, over 5750314.65 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3148, pruned_loss=0.08157, over 5721197.39 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:54:10,245 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6221, 1.7296, 1.8239, 1.4090], device='cuda:1'), covar=tensor([0.1868, 0.2596, 0.1522, 0.1725], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0713, 0.0987, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-14 23:54:10,339 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-14 23:54:45,681 INFO [train.py:968] (1/2) Epoch 29, batch 20150, giga_loss[loss=0.2404, simple_loss=0.315, pruned_loss=0.08287, over 28639.00 frames. ], tot_loss[loss=0.2396, simple_loss=0.3164, pruned_loss=0.08141, over 5726036.74 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3368, pruned_loss=0.08624, over 5752185.13 frames. ], giga_tot_loss[loss=0.2372, simple_loss=0.3132, pruned_loss=0.08063, over 5728554.68 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:54:52,691 INFO [optim.py:369] (1/2) 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 23:55:03,753 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5916, 1.8242, 1.7629, 1.5061], device='cuda:1'), covar=tensor([0.2718, 0.2163, 0.1762, 0.2274], device='cuda:1'), in_proj_covar=tensor([0.2057, 0.2011, 0.1915, 0.2071], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 23:55:22,420 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 20200, giga_loss[loss=0.2523, simple_loss=0.3244, pruned_loss=0.09014, over 28790.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3215, pruned_loss=0.08472, over 5729835.11 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3376, pruned_loss=0.08669, over 5758294.96 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3173, pruned_loss=0.08353, over 5724899.48 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:55:43,069 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:55:45,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1296042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:55:49,449 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:55:51,369 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 20250, giga_loss[loss=0.291, simple_loss=0.3656, pruned_loss=0.1081, over 28836.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3266, pruned_loss=0.08784, over 5714857.24 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3378, pruned_loss=0.08679, over 5753057.86 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3226, pruned_loss=0.08676, over 5715411.23 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:56:13,323 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 23:56:20,561 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1296080.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:57:00,251 INFO [train.py:968] (1/2) Epoch 29, batch 20300, libri_loss[loss=0.3091, simple_loss=0.3762, pruned_loss=0.121, over 29752.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.3362, pruned_loss=0.09464, over 5700782.70 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.338, pruned_loss=0.08701, over 5755939.24 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3326, pruned_loss=0.09364, over 5697450.85 frames. ], batch size: 87, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:57:10,312 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-14 23:57:32,486 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:57:34,448 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 20350, giga_loss[loss=0.2719, simple_loss=0.354, pruned_loss=0.09488, over 28530.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3407, pruned_loss=0.09679, over 5694000.75 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3385, pruned_loss=0.08748, over 5750756.04 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3373, pruned_loss=0.09595, over 5692542.26 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:57:47,335 INFO [optim.py:369] (1/2) 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 23:57:56,215 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1296190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:57:57,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1296192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-14 23:58:04,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7386, 1.8502, 1.5645, 2.0518], device='cuda:1'), covar=tensor([0.2694, 0.2948, 0.3193, 0.2450], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1161, 0.1424, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-14 23:58:24,512 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4567, 1.6833, 1.6407, 1.3607], device='cuda:1'), covar=tensor([0.2592, 0.2393, 0.1829, 0.2256], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2019, 0.1922, 0.2073], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-14 23:58:24,884 INFO [train.py:968] (1/2) Epoch 29, batch 20400, libri_loss[loss=0.282, simple_loss=0.3613, pruned_loss=0.1014, over 29512.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3442, pruned_loss=0.09786, over 5673432.65 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.339, pruned_loss=0.08789, over 5744146.30 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.3411, pruned_loss=0.09709, over 5676205.12 frames. ], batch size: 81, lr: 1.09e-03, grad_scale: 8.0
+2023-03-14 23:58:32,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-14 23:59:05,531 INFO [train.py:968] (1/2) Epoch 29, batch 20450, giga_loss[loss=0.3999, simple_loss=0.4434, pruned_loss=0.1782, over 26633.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.348, pruned_loss=0.09946, over 5671793.26 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3389, pruned_loss=0.08792, over 5739811.47 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3458, pruned_loss=0.09917, over 5674832.13 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-14 23:59:12,480 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 20500, giga_loss[loss=0.2652, simple_loss=0.3461, pruned_loss=0.09213, over 28783.00 frames. ], tot_loss[loss=0.2799, simple_loss=0.3537, pruned_loss=0.1031, over 5669156.41 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3387, pruned_loss=0.08782, over 5740579.83 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3523, pruned_loss=0.1032, over 5669579.59 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:00:01,619 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1296338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:00:28,902 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 20550, giga_loss[loss=0.2536, simple_loss=0.3362, pruned_loss=0.08546, over 28861.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.3519, pruned_loss=0.1017, over 5675313.93 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3394, pruned_loss=0.08836, over 5739139.94 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3506, pruned_loss=0.1016, over 5675499.13 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:00:39,364 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 20600, giga_loss[loss=0.2564, simple_loss=0.3356, pruned_loss=0.08863, over 28689.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3489, pruned_loss=0.09892, over 5680901.90 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3397, pruned_loss=0.08868, over 5739106.89 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09866, over 5680500.31 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:01:48,337 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3205, 1.2813, 1.2834, 1.4966], device='cuda:1'), covar=tensor([0.0842, 0.0377, 0.0351, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0121, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 00:01:58,692 INFO [train.py:968] (1/2) Epoch 29, batch 20650, giga_loss[loss=0.3228, simple_loss=0.3786, pruned_loss=0.1334, over 26622.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3475, pruned_loss=0.09787, over 5685573.06 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3396, pruned_loss=0.08864, over 5740055.36 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3466, pruned_loss=0.09775, over 5684152.57 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:02:05,278 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1296479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:02:05,631 INFO [optim.py:369] (1/2) 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 00:02:41,500 INFO [train.py:968] (1/2) Epoch 29, batch 20700, giga_loss[loss=0.2554, simple_loss=0.3415, pruned_loss=0.08469, over 28550.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3473, pruned_loss=0.09698, over 5686285.77 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3394, pruned_loss=0.08863, over 5741590.75 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3469, pruned_loss=0.09696, over 5683481.42 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:03:15,190 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7313, 1.7306, 1.7355, 1.5656], device='cuda:1'), covar=tensor([0.2798, 0.2729, 0.2304, 0.2723], device='cuda:1'), in_proj_covar=tensor([0.2068, 0.2027, 0.1929, 0.2082], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 00:03:22,867 INFO [train.py:968] (1/2) Epoch 29, batch 20750, giga_loss[loss=0.2504, simple_loss=0.3299, pruned_loss=0.08541, over 28562.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3502, pruned_loss=0.0992, over 5682829.62 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3405, pruned_loss=0.08937, over 5732882.71 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.349, pruned_loss=0.09872, over 5686994.59 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:03:29,363 INFO [optim.py:369] (1/2) 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 00:03:33,794 INFO [scaling.py:679] (1/2) 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] (1/2) Epoch 29, batch 20800, giga_loss[loss=0.2268, simple_loss=0.3105, pruned_loss=0.07158, over 28591.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3521, pruned_loss=0.1005, over 5687691.56 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3409, pruned_loss=0.08965, over 5724176.93 frames. ], giga_tot_loss[loss=0.2755, simple_loss=0.3509, pruned_loss=0.1, over 5698035.09 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:04:53,299 INFO [train.py:968] (1/2) Epoch 29, batch 20850, giga_loss[loss=0.4552, simple_loss=0.4738, pruned_loss=0.2182, over 26619.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3545, pruned_loss=0.1031, over 5671144.09 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3411, pruned_loss=0.08971, over 5725734.57 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3535, pruned_loss=0.1028, over 5677557.03 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:05:01,110 INFO [optim.py:369] (1/2) 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 00:05:23,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-15 00:05:36,583 INFO [train.py:968] (1/2) Epoch 29, batch 20900, giga_loss[loss=0.2627, simple_loss=0.3438, pruned_loss=0.09077, over 28499.00 frames. ], tot_loss[loss=0.28, simple_loss=0.3539, pruned_loss=0.103, over 5682115.85 frames. ], libri_tot_loss[loss=0.2606, simple_loss=0.3412, pruned_loss=0.08995, over 5728445.42 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3532, pruned_loss=0.1028, over 5683788.48 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:06:00,715 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1131, 1.3135, 1.1142, 0.8453], device='cuda:1'), covar=tensor([0.1174, 0.0552, 0.1146, 0.1230], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0449, 0.0528, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 00:06:05,064 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0956, 1.0508, 4.0013, 3.3239], device='cuda:1'), covar=tensor([0.2433, 0.3521, 0.0888, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0672, 0.1003, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 00:06:13,697 INFO [train.py:968] (1/2) Epoch 29, batch 20950, giga_loss[loss=0.2885, simple_loss=0.3599, pruned_loss=0.1085, over 28962.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3531, pruned_loss=0.1021, over 5687844.53 frames. ], libri_tot_loss[loss=0.2611, simple_loss=0.3416, pruned_loss=0.09024, over 5724732.39 frames. ], giga_tot_loss[loss=0.2785, simple_loss=0.3527, pruned_loss=0.1021, over 5691260.82 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:06:21,311 INFO [optim.py:369] (1/2) 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 00:06:31,757 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-15 00:06:55,620 INFO [train.py:968] (1/2) Epoch 29, batch 21000, giga_loss[loss=0.237, simple_loss=0.3223, pruned_loss=0.0758, over 28634.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3529, pruned_loss=0.1013, over 5681959.87 frames. ], libri_tot_loss[loss=0.2618, simple_loss=0.3422, pruned_loss=0.09074, over 5716583.37 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3523, pruned_loss=0.1012, over 5690517.19 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:06:55,620 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 00:07:05,058 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 00:07:30,706 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 21050, giga_loss[loss=0.2453, simple_loss=0.3373, pruned_loss=0.07666, over 28932.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3534, pruned_loss=0.1006, over 5690010.88 frames. ], libri_tot_loss[loss=0.2621, simple_loss=0.3422, pruned_loss=0.091, over 5719045.02 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3533, pruned_loss=0.1006, over 5693858.95 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:07:51,009 INFO [optim.py:369] (1/2) 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 00:08:24,736 INFO [train.py:968] (1/2) Epoch 29, batch 21100, giga_loss[loss=0.3136, simple_loss=0.3689, pruned_loss=0.1291, over 27627.00 frames. ], tot_loss[loss=0.2757, simple_loss=0.3521, pruned_loss=0.09969, over 5691832.28 frames. ], libri_tot_loss[loss=0.2625, simple_loss=0.3425, pruned_loss=0.09129, over 5721630.77 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3518, pruned_loss=0.09951, over 5692128.66 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:08:48,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3848, 1.1929, 1.1107, 1.4587], device='cuda:1'), covar=tensor([0.0821, 0.0389, 0.0381, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0121, 0.0120, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 00:09:00,974 INFO [train.py:968] (1/2) Epoch 29, batch 21150, libri_loss[loss=0.2696, simple_loss=0.3542, pruned_loss=0.09247, over 29521.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3488, pruned_loss=0.09781, over 5706391.85 frames. ], libri_tot_loss[loss=0.2626, simple_loss=0.3425, pruned_loss=0.0914, over 5724509.00 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3487, pruned_loss=0.09773, over 5703438.27 frames. ], batch size: 82, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:09:08,370 INFO [optim.py:369] (1/2) 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 00:09:22,638 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1296997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:09:24,643 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 21200, giga_loss[loss=0.2396, simple_loss=0.3202, pruned_loss=0.07947, over 28633.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3468, pruned_loss=0.09687, over 5710940.00 frames. ], libri_tot_loss[loss=0.2628, simple_loss=0.3426, pruned_loss=0.09151, over 5727321.49 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3467, pruned_loss=0.09677, over 5705832.40 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:09:46,099 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:10:00,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3902, 1.3719, 4.0638, 3.4394], device='cuda:1'), covar=tensor([0.1710, 0.2905, 0.0459, 0.0837], device='cuda:1'), in_proj_covar=tensor([0.0805, 0.0672, 0.1003, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 00:10:08,592 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7230, 2.0405, 1.3259, 1.6623], device='cuda:1'), covar=tensor([0.0941, 0.0481, 0.1012, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0448, 0.0527, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 00:10:18,641 INFO [train.py:968] (1/2) Epoch 29, batch 21250, giga_loss[loss=0.2715, simple_loss=0.3434, pruned_loss=0.09978, over 28965.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3455, pruned_loss=0.09649, over 5713166.83 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3428, pruned_loss=0.09157, over 5730741.49 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3453, pruned_loss=0.09645, over 5705600.81 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:10:29,000 INFO [optim.py:369] (1/2) 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 00:10:34,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8995, 1.8149, 2.0056, 1.6162], device='cuda:1'), covar=tensor([0.1917, 0.2720, 0.1548, 0.1812], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0719, 0.0989, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 00:11:01,369 INFO [train.py:968] (1/2) Epoch 29, batch 21300, giga_loss[loss=0.2903, simple_loss=0.3572, pruned_loss=0.1117, over 28492.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3466, pruned_loss=0.09774, over 5706238.54 frames. ], libri_tot_loss[loss=0.263, simple_loss=0.3427, pruned_loss=0.09163, over 5725358.83 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3466, pruned_loss=0.09781, over 5704205.30 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:11:12,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 00:11:43,298 INFO [train.py:968] (1/2) Epoch 29, batch 21350, giga_loss[loss=0.2836, simple_loss=0.3619, pruned_loss=0.1027, over 28864.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3473, pruned_loss=0.09775, over 5711197.42 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3428, pruned_loss=0.0919, over 5727314.92 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3472, pruned_loss=0.0976, over 5707714.69 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:11:52,629 INFO [optim.py:369] (1/2) 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 00:12:24,109 INFO [train.py:968] (1/2) Epoch 29, batch 21400, giga_loss[loss=0.256, simple_loss=0.3355, pruned_loss=0.08825, over 28610.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3471, pruned_loss=0.09728, over 5700804.71 frames. ], libri_tot_loss[loss=0.2637, simple_loss=0.3431, pruned_loss=0.09212, over 5720034.52 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3468, pruned_loss=0.09701, over 5704041.95 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:12:39,785 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4006, 4.2551, 3.9822, 1.9031], device='cuda:1'), covar=tensor([0.0593, 0.0736, 0.0737, 0.2119], device='cuda:1'), in_proj_covar=tensor([0.1286, 0.1187, 0.1000, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 00:13:03,892 INFO [train.py:968] (1/2) Epoch 29, batch 21450, giga_loss[loss=0.2351, simple_loss=0.3232, pruned_loss=0.0735, over 28550.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3462, pruned_loss=0.09644, over 5717354.30 frames. ], libri_tot_loss[loss=0.2646, simple_loss=0.3437, pruned_loss=0.09278, over 5727996.19 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3456, pruned_loss=0.09578, over 5712109.67 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:13:11,699 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 21500, giga_loss[loss=0.273, simple_loss=0.3389, pruned_loss=0.1035, over 28977.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.0964, over 5718928.10 frames. ], libri_tot_loss[loss=0.2651, simple_loss=0.3441, pruned_loss=0.09309, over 5721163.67 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3451, pruned_loss=0.09567, over 5719853.05 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:13:48,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8828, 3.0594, 1.9171, 1.0035], device='cuda:1'), covar=tensor([0.9921, 0.3133, 0.4708, 0.8281], device='cuda:1'), in_proj_covar=tensor([0.1828, 0.1713, 0.1644, 0.1488], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:14:20,085 INFO [train.py:968] (1/2) Epoch 29, batch 21550, giga_loss[loss=0.2429, simple_loss=0.3144, pruned_loss=0.08564, over 28670.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3428, pruned_loss=0.09488, over 5719870.41 frames. ], libri_tot_loss[loss=0.2653, simple_loss=0.3441, pruned_loss=0.09327, over 5725873.95 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3421, pruned_loss=0.09419, over 5716128.13 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:14:29,334 INFO [optim.py:369] (1/2) 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 00:14:58,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1297418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:15:00,644 INFO [train.py:968] (1/2) Epoch 29, batch 21600, giga_loss[loss=0.2773, simple_loss=0.3548, pruned_loss=0.09985, over 27899.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3396, pruned_loss=0.093, over 5716318.61 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3444, pruned_loss=0.09351, over 5725147.69 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3388, pruned_loss=0.09226, over 5713983.11 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:15:24,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3569, 3.4680, 1.5156, 1.5411], device='cuda:1'), covar=tensor([0.1050, 0.0295, 0.0914, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0424, 0.0565, 0.0409, 0.0445], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 00:15:29,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:15:40,583 INFO [train.py:968] (1/2) Epoch 29, batch 21650, giga_loss[loss=0.274, simple_loss=0.3506, pruned_loss=0.09872, over 28681.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3391, pruned_loss=0.09281, over 5721389.51 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3444, pruned_loss=0.09351, over 5725147.69 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3385, pruned_loss=0.09223, over 5719571.76 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:15:50,367 INFO [optim.py:369] (1/2) 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 00:16:02,302 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9426, 1.3013, 1.0790, 0.2481], device='cuda:1'), covar=tensor([0.5115, 0.3494, 0.5079, 0.7242], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1724, 0.1656, 0.1497], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:16:20,473 INFO [train.py:968] (1/2) Epoch 29, batch 21700, libri_loss[loss=0.3027, simple_loss=0.3697, pruned_loss=0.1178, over 29731.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3386, pruned_loss=0.09326, over 5720144.02 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3448, pruned_loss=0.09401, over 5727421.80 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3374, pruned_loss=0.09231, over 5716307.80 frames. ], batch size: 87, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:16:52,338 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1297561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:16:55,151 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1297564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:16:59,501 INFO [train.py:968] (1/2) Epoch 29, batch 21750, libri_loss[loss=0.3024, simple_loss=0.3726, pruned_loss=0.1161, over 27766.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3371, pruned_loss=0.09301, over 5718404.97 frames. ], libri_tot_loss[loss=0.267, simple_loss=0.3453, pruned_loss=0.09439, over 5732061.74 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3355, pruned_loss=0.09183, over 5710750.29 frames. ], batch size: 116, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:17:08,883 INFO [optim.py:369] (1/2) 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 00:17:17,199 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1297593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:17:30,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1938, 1.2723, 3.2653, 2.9396], device='cuda:1'), covar=tensor([0.1511, 0.2631, 0.0452, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0804, 0.0672, 0.1003, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 00:17:38,127 INFO [train.py:968] (1/2) Epoch 29, batch 21800, giga_loss[loss=0.3426, simple_loss=0.388, pruned_loss=0.1486, over 23652.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3342, pruned_loss=0.0919, over 5720718.26 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3449, pruned_loss=0.09442, over 5735802.47 frames. ], giga_tot_loss[loss=0.2574, simple_loss=0.333, pruned_loss=0.0909, over 5711092.33 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:17:53,855 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-15 00:18:11,769 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 21850, giga_loss[loss=0.2935, simple_loss=0.3678, pruned_loss=0.1096, over 28613.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.332, pruned_loss=0.09117, over 5707702.14 frames. ], libri_tot_loss[loss=0.2673, simple_loss=0.3453, pruned_loss=0.09469, over 5728915.24 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3306, pruned_loss=0.09008, over 5706513.56 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:18:26,877 INFO [optim.py:369] (1/2) 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 00:18:30,189 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2943, 2.0434, 1.5723, 0.5658], device='cuda:1'), covar=tensor([0.6844, 0.3583, 0.5315, 0.7646], device='cuda:1'), in_proj_covar=tensor([0.1844, 0.1733, 0.1663, 0.1500], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:18:57,992 INFO [train.py:968] (1/2) Epoch 29, batch 21900, giga_loss[loss=0.3244, simple_loss=0.3807, pruned_loss=0.1341, over 23818.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.3316, pruned_loss=0.09107, over 5705486.65 frames. ], libri_tot_loss[loss=0.2676, simple_loss=0.3455, pruned_loss=0.09491, over 5731740.73 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3302, pruned_loss=0.08994, over 5701820.91 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:19:29,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4448, 1.7157, 1.3983, 1.3746], device='cuda:1'), covar=tensor([0.2885, 0.2937, 0.3396, 0.2533], device='cuda:1'), in_proj_covar=tensor([0.1608, 0.1160, 0.1420, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 00:19:38,780 INFO [train.py:968] (1/2) Epoch 29, batch 21950, libri_loss[loss=0.2662, simple_loss=0.3337, pruned_loss=0.09937, over 29552.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.332, pruned_loss=0.09106, over 5701263.42 frames. ], libri_tot_loss[loss=0.2681, simple_loss=0.3457, pruned_loss=0.09527, over 5724597.33 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3303, pruned_loss=0.08975, over 5704297.54 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:19:44,452 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:19:47,089 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 22000, giga_loss[loss=0.2692, simple_loss=0.3543, pruned_loss=0.09207, over 29034.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3358, pruned_loss=0.09277, over 5707271.79 frames. ], libri_tot_loss[loss=0.2686, simple_loss=0.3459, pruned_loss=0.09561, over 5729514.83 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3339, pruned_loss=0.09129, over 5704725.11 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:20:28,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1297831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:21:02,228 INFO [train.py:968] (1/2) Epoch 29, batch 22050, giga_loss[loss=0.2824, simple_loss=0.3698, pruned_loss=0.09755, over 28700.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3376, pruned_loss=0.09319, over 5706488.42 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3459, pruned_loss=0.09588, over 5730969.19 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3359, pruned_loss=0.09173, over 5702804.06 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:21:11,351 INFO [optim.py:369] (1/2) 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 00:21:28,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8985, 1.1517, 1.0980, 0.8989], device='cuda:1'), covar=tensor([0.2651, 0.2665, 0.1704, 0.2372], device='cuda:1'), in_proj_covar=tensor([0.2079, 0.2038, 0.1938, 0.2088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 00:21:33,978 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1297909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:21:43,223 INFO [train.py:968] (1/2) Epoch 29, batch 22100, giga_loss[loss=0.2559, simple_loss=0.3445, pruned_loss=0.08369, over 28362.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.339, pruned_loss=0.09299, over 5703296.39 frames. ], libri_tot_loss[loss=0.2692, simple_loss=0.3461, pruned_loss=0.09615, over 5732719.39 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3374, pruned_loss=0.09158, over 5698544.50 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:22:28,350 INFO [train.py:968] (1/2) Epoch 29, batch 22150, giga_loss[loss=0.2674, simple_loss=0.3348, pruned_loss=0.09999, over 24072.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3386, pruned_loss=0.09275, over 5690271.91 frames. ], libri_tot_loss[loss=0.2703, simple_loss=0.3467, pruned_loss=0.09697, over 5725157.95 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3367, pruned_loss=0.09088, over 5691849.40 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:22:31,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1297974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:22:33,641 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:22:55,294 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-15 00:22:57,784 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298006.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:23:06,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3935, 3.3618, 1.5441, 1.4531], device='cuda:1'), covar=tensor([0.0974, 0.0333, 0.0934, 0.1328], device='cuda:1'), in_proj_covar=tensor([0.0425, 0.0568, 0.0411, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 00:23:10,370 INFO [train.py:968] (1/2) Epoch 29, batch 22200, giga_loss[loss=0.2781, simple_loss=0.3551, pruned_loss=0.1005, over 28026.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.339, pruned_loss=0.09333, over 5689751.99 frames. ], libri_tot_loss[loss=0.2705, simple_loss=0.3468, pruned_loss=0.09711, over 5718573.23 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3374, pruned_loss=0.09165, over 5697181.30 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:23:23,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1298038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:23:24,439 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6731, 1.8234, 2.0354, 1.6350], device='cuda:1'), covar=tensor([0.3497, 0.2874, 0.2815, 0.3249], device='cuda:1'), in_proj_covar=tensor([0.2073, 0.2030, 0.1931, 0.2081], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 00:23:26,677 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.70 vs. limit=5.0
+2023-03-15 00:23:50,094 INFO [train.py:968] (1/2) Epoch 29, batch 22250, giga_loss[loss=0.2991, simple_loss=0.3566, pruned_loss=0.1208, over 23723.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3404, pruned_loss=0.09405, over 5683771.87 frames. ], libri_tot_loss[loss=0.2712, simple_loss=0.3475, pruned_loss=0.09748, over 5711342.99 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3382, pruned_loss=0.09231, over 5695005.59 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:23:55,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6429, 1.8114, 1.2996, 1.4874], device='cuda:1'), covar=tensor([0.1160, 0.0798, 0.1164, 0.1395], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0449, 0.0528, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 00:24:00,101 INFO [optim.py:369] (1/2) 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 00:24:30,057 INFO [train.py:968] (1/2) Epoch 29, batch 22300, libri_loss[loss=0.2676, simple_loss=0.3361, pruned_loss=0.09952, over 29581.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3427, pruned_loss=0.09558, over 5698932.39 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3479, pruned_loss=0.09784, over 5716507.10 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3405, pruned_loss=0.0938, over 5702853.74 frames. ], batch size: 74, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:24:57,289 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1298154.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:02,369 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6351, 1.7378, 1.8188, 1.3896], device='cuda:1'), covar=tensor([0.1976, 0.2609, 0.1628, 0.1832], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0718, 0.0989, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 00:25:09,224 INFO [train.py:968] (1/2) Epoch 29, batch 22350, giga_loss[loss=0.2851, simple_loss=0.3648, pruned_loss=0.1027, over 28894.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3445, pruned_loss=0.09667, over 5697404.33 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3481, pruned_loss=0.0982, over 5720837.30 frames. ], giga_tot_loss[loss=0.2661, simple_loss=0.3425, pruned_loss=0.09487, over 5696003.63 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:25:16,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7960, 2.9415, 1.8164, 1.0502], device='cuda:1'), covar=tensor([1.0111, 0.3644, 0.5292, 0.8510], device='cuda:1'), in_proj_covar=tensor([0.1843, 0.1727, 0.1664, 0.1499], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:25:16,774 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:20,761 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:21,177 INFO [optim.py:369] (1/2) 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 00:25:44,122 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:25:51,089 INFO [train.py:968] (1/2) Epoch 29, batch 22400, giga_loss[loss=0.2719, simple_loss=0.3565, pruned_loss=0.09367, over 28979.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3473, pruned_loss=0.0978, over 5707539.96 frames. ], libri_tot_loss[loss=0.2727, simple_loss=0.3484, pruned_loss=0.09848, over 5724241.39 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3453, pruned_loss=0.09611, over 5703196.37 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:26:01,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3132, 1.5558, 1.4532, 1.1806], device='cuda:1'), covar=tensor([0.2975, 0.2658, 0.2120, 0.2853], device='cuda:1'), in_proj_covar=tensor([0.2074, 0.2033, 0.1932, 0.2082], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 00:26:27,898 INFO [train.py:968] (1/2) Epoch 29, batch 22450, libri_loss[loss=0.269, simple_loss=0.3462, pruned_loss=0.09593, over 27876.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3496, pruned_loss=0.09907, over 5702236.02 frames. ], libri_tot_loss[loss=0.274, simple_loss=0.3495, pruned_loss=0.09923, over 5715052.00 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3471, pruned_loss=0.09701, over 5706175.93 frames. ], batch size: 116, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:26:38,263 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:26:47,784 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:26:49,660 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:27:05,450 INFO [train.py:968] (1/2) Epoch 29, batch 22500, giga_loss[loss=0.277, simple_loss=0.3583, pruned_loss=0.09791, over 28751.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3505, pruned_loss=0.09961, over 5711552.90 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3499, pruned_loss=0.09983, over 5721514.25 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.348, pruned_loss=0.09737, over 5708357.62 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:27:13,426 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:27:45,269 INFO [train.py:968] (1/2) Epoch 29, batch 22550, libri_loss[loss=0.3024, simple_loss=0.3695, pruned_loss=0.1177, over 29560.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3512, pruned_loss=0.1004, over 5701559.11 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3502, pruned_loss=0.1002, over 5707790.17 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.349, pruned_loss=0.09832, over 5710629.10 frames. ], batch size: 76, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:27:55,843 INFO [optim.py:369] (1/2) 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 00:28:26,928 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298420.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 00:28:27,269 INFO [train.py:968] (1/2) Epoch 29, batch 22600, giga_loss[loss=0.2753, simple_loss=0.347, pruned_loss=0.1018, over 28827.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3483, pruned_loss=0.09866, over 5702831.34 frames. ], libri_tot_loss[loss=0.2751, simple_loss=0.35, pruned_loss=0.1001, over 5709915.77 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3467, pruned_loss=0.09708, over 5708102.32 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:28:31,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:28:34,094 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:28:58,623 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:29:08,723 INFO [train.py:968] (1/2) Epoch 29, batch 22650, giga_loss[loss=0.2479, simple_loss=0.3292, pruned_loss=0.08335, over 28580.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3459, pruned_loss=0.09738, over 5707828.57 frames. ], libri_tot_loss[loss=0.2748, simple_loss=0.3498, pruned_loss=0.09991, over 5717234.82 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3447, pruned_loss=0.09619, over 5705439.41 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:29:18,492 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 22700, giga_loss[loss=0.2404, simple_loss=0.3194, pruned_loss=0.0807, over 28801.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3431, pruned_loss=0.09615, over 5707703.86 frames. ], libri_tot_loss[loss=0.2753, simple_loss=0.3502, pruned_loss=0.1002, over 5720299.88 frames. ], giga_tot_loss[loss=0.2658, simple_loss=0.3417, pruned_loss=0.0949, over 5702846.01 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:30:04,389 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3046, 1.4351, 1.3438, 1.4662], device='cuda:1'), covar=tensor([0.0746, 0.0376, 0.0363, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 00:30:27,484 INFO [train.py:968] (1/2) Epoch 29, batch 22750, giga_loss[loss=0.2607, simple_loss=0.339, pruned_loss=0.09115, over 28701.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.342, pruned_loss=0.09446, over 5699250.68 frames. ], libri_tot_loss[loss=0.2754, simple_loss=0.3503, pruned_loss=0.1002, over 5712082.38 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3408, pruned_loss=0.0934, over 5703487.84 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:30:41,874 INFO [optim.py:369] (1/2) 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 00:31:09,835 INFO [train.py:968] (1/2) Epoch 29, batch 22800, giga_loss[loss=0.2784, simple_loss=0.3443, pruned_loss=0.1063, over 28812.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3444, pruned_loss=0.09477, over 5700748.48 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3508, pruned_loss=0.1007, over 5714666.37 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3429, pruned_loss=0.09335, over 5701342.47 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:31:46,242 INFO [train.py:968] (1/2) Epoch 29, batch 22850, giga_loss[loss=0.2377, simple_loss=0.3241, pruned_loss=0.07562, over 28698.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3453, pruned_loss=0.09561, over 5694564.36 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3511, pruned_loss=0.1012, over 5710724.19 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.3436, pruned_loss=0.09383, over 5697544.20 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:31:57,965 INFO [optim.py:369] (1/2) 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 00:32:03,797 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3117, 2.0246, 1.5452, 0.5905], device='cuda:1'), covar=tensor([0.6620, 0.3401, 0.4970, 0.8060], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1730, 0.1664, 0.1502], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:32:28,195 INFO [train.py:968] (1/2) Epoch 29, batch 22900, giga_loss[loss=0.3158, simple_loss=0.3731, pruned_loss=0.1293, over 28694.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3435, pruned_loss=0.09579, over 5696719.45 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3513, pruned_loss=0.1014, over 5713709.36 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3419, pruned_loss=0.09408, over 5696127.12 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:32:38,640 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.88 vs. limit=5.0
+2023-03-15 00:33:05,004 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:33:07,397 INFO [train.py:968] (1/2) Epoch 29, batch 22950, giga_loss[loss=0.2658, simple_loss=0.3292, pruned_loss=0.1012, over 28855.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3427, pruned_loss=0.0968, over 5696291.93 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3517, pruned_loss=0.1017, over 5708282.45 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3408, pruned_loss=0.09504, over 5701175.47 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:33:18,617 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8752, 1.2770, 1.3910, 1.0306], device='cuda:1'), covar=tensor([0.1825, 0.1233, 0.2043, 0.1559], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0761, 0.0734, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 00:33:18,745 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.11 vs. limit=5.0
+2023-03-15 00:33:18,988 INFO [optim.py:369] (1/2) 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 00:33:26,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1298795.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 00:33:47,084 INFO [train.py:968] (1/2) Epoch 29, batch 23000, giga_loss[loss=0.2527, simple_loss=0.3198, pruned_loss=0.0928, over 28774.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3405, pruned_loss=0.09625, over 5701762.64 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.352, pruned_loss=0.102, over 5704408.06 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3382, pruned_loss=0.09437, over 5708769.15 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:33:54,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3781, 1.8215, 1.3637, 1.6935], device='cuda:1'), covar=tensor([0.0759, 0.0292, 0.0349, 0.0871], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-15 00:34:06,687 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3661, 1.6800, 1.3333, 1.2827], device='cuda:1'), covar=tensor([0.2958, 0.2913, 0.3475, 0.2651], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1159, 0.1421, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 00:34:23,989 INFO [train.py:968] (1/2) Epoch 29, batch 23050, giga_loss[loss=0.2911, simple_loss=0.3584, pruned_loss=0.1119, over 27987.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3407, pruned_loss=0.09736, over 5695380.56 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3526, pruned_loss=0.1027, over 5701788.57 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3379, pruned_loss=0.09504, over 5702992.69 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:34:34,285 INFO [optim.py:369] (1/2) 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 00:34:57,591 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298917.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:35:00,814 INFO [train.py:968] (1/2) Epoch 29, batch 23100, giga_loss[loss=0.2502, simple_loss=0.3259, pruned_loss=0.08726, over 28895.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3396, pruned_loss=0.09662, over 5712406.73 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3524, pruned_loss=0.1027, over 5708065.02 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3373, pruned_loss=0.09457, over 5712980.03 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:35:13,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1298938.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 00:35:16,414 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:35:16,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1298941.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 00:35:25,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3456, 1.3222, 3.9559, 3.3253], device='cuda:1'), covar=tensor([0.1611, 0.2571, 0.0480, 0.0981], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0673, 0.1009, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 00:35:37,515 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1298970.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 00:35:38,023 INFO [train.py:968] (1/2) Epoch 29, batch 23150, giga_loss[loss=0.2319, simple_loss=0.3049, pruned_loss=0.07946, over 28563.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3348, pruned_loss=0.09424, over 5709080.40 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.352, pruned_loss=0.1026, over 5710515.15 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.333, pruned_loss=0.09255, over 5707294.33 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:35:44,531 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1298981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:35:45,845 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1041, 2.3080, 2.2954, 1.8061], device='cuda:1'), covar=tensor([0.3864, 0.2810, 0.2738, 0.3693], device='cuda:1'), in_proj_covar=tensor([0.2083, 0.2039, 0.1936, 0.2084], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 00:35:48,401 INFO [optim.py:369] (1/2) 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 00:36:16,806 INFO [train.py:968] (1/2) Epoch 29, batch 23200, giga_loss[loss=0.2272, simple_loss=0.2976, pruned_loss=0.07836, over 28674.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.331, pruned_loss=0.09225, over 5712696.74 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.3526, pruned_loss=0.1032, over 5716148.68 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3282, pruned_loss=0.0899, over 5705745.33 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:36:39,461 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1299052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:36:54,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9070, 2.7560, 1.7365, 1.1156], device='cuda:1'), covar=tensor([0.9524, 0.3831, 0.4965, 0.8445], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1731, 0.1663, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:36:54,377 INFO [train.py:968] (1/2) Epoch 29, batch 23250, giga_loss[loss=0.2353, simple_loss=0.3193, pruned_loss=0.07563, over 28828.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3284, pruned_loss=0.09043, over 5714177.92 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3521, pruned_loss=0.1029, over 5717804.72 frames. ], giga_tot_loss[loss=0.2519, simple_loss=0.3264, pruned_loss=0.08872, over 5707221.25 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:37:06,312 INFO [optim.py:369] (1/2) 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 00:37:20,111 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 00:37:34,826 INFO [train.py:968] (1/2) Epoch 29, batch 23300, giga_loss[loss=0.2777, simple_loss=0.3568, pruned_loss=0.09927, over 29144.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3312, pruned_loss=0.09151, over 5719053.92 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3527, pruned_loss=0.1036, over 5724100.18 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3283, pruned_loss=0.0892, over 5707987.43 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:37:51,727 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:38:12,765 INFO [train.py:968] (1/2) Epoch 29, batch 23350, giga_loss[loss=0.2646, simple_loss=0.3459, pruned_loss=0.09169, over 28818.00 frames. ], tot_loss[loss=0.2607, simple_loss=0.3348, pruned_loss=0.09328, over 5711791.52 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3526, pruned_loss=0.1037, over 5717483.88 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3319, pruned_loss=0.09089, over 5708801.16 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:38:24,633 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 23400, libri_loss[loss=0.2745, simple_loss=0.3349, pruned_loss=0.107, over 29469.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3387, pruned_loss=0.09493, over 5712930.49 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3523, pruned_loss=0.1036, over 5722168.88 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3362, pruned_loss=0.09287, over 5706219.55 frames. ], batch size: 70, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:39:08,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5473, 1.7583, 1.5118, 1.5609], device='cuda:1'), covar=tensor([0.0743, 0.0304, 0.0323, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0193, 0.0122, 0.0120, 0.0229], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0115], device='cuda:1')
+2023-03-15 00:39:31,332 INFO [train.py:968] (1/2) Epoch 29, batch 23450, giga_loss[loss=0.2813, simple_loss=0.3644, pruned_loss=0.09909, over 28974.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3422, pruned_loss=0.09653, over 5703815.63 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.352, pruned_loss=0.1035, over 5716511.12 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3403, pruned_loss=0.0948, over 5702162.36 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:39:34,138 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-15 00:39:42,540 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:39:45,296 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:39:49,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:40:10,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:40:11,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:40:13,600 INFO [train.py:968] (1/2) Epoch 29, batch 23500, giga_loss[loss=0.2529, simple_loss=0.3262, pruned_loss=0.08984, over 28624.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3458, pruned_loss=0.09854, over 5694271.71 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3523, pruned_loss=0.1037, over 5711420.96 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3437, pruned_loss=0.0968, over 5696218.76 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:40:42,514 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 23550, giga_loss[loss=0.364, simple_loss=0.4179, pruned_loss=0.155, over 28614.00 frames. ], tot_loss[loss=0.2772, simple_loss=0.3499, pruned_loss=0.1022, over 5690467.04 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3525, pruned_loss=0.1039, over 5714553.07 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3479, pruned_loss=0.1006, over 5688976.80 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:41:13,006 INFO [optim.py:369] (1/2) 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 00:41:29,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6735, 1.8519, 1.5629, 1.8267], device='cuda:1'), covar=tensor([0.2713, 0.2822, 0.3206, 0.2413], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1164, 0.1425, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 00:41:34,293 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4768, 3.6552, 1.6004, 1.7044], device='cuda:1'), covar=tensor([0.1033, 0.0438, 0.0943, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0572, 0.0412, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 00:41:45,158 INFO [train.py:968] (1/2) Epoch 29, batch 23600, giga_loss[loss=0.3412, simple_loss=0.3967, pruned_loss=0.1428, over 27527.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3541, pruned_loss=0.1059, over 5675647.00 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3526, pruned_loss=0.1041, over 5706521.04 frames. ], giga_tot_loss[loss=0.2806, simple_loss=0.3524, pruned_loss=0.1044, over 5681767.45 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:41:54,080 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:41:59,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:01,749 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:22,275 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:24,851 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:28,330 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:42:32,866 INFO [train.py:968] (1/2) Epoch 29, batch 23650, giga_loss[loss=0.3854, simple_loss=0.4164, pruned_loss=0.1772, over 23936.00 frames. ], tot_loss[loss=0.291, simple_loss=0.3606, pruned_loss=0.1107, over 5676269.33 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3526, pruned_loss=0.1047, over 5702605.51 frames. ], giga_tot_loss[loss=0.2888, simple_loss=0.3594, pruned_loss=0.1091, over 5683552.34 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:42:51,145 INFO [optim.py:369] (1/2) 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 00:42:54,657 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:03,647 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:05,699 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:06,298 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1299503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:20,652 INFO [train.py:968] (1/2) Epoch 29, batch 23700, giga_loss[loss=0.3216, simple_loss=0.3879, pruned_loss=0.1276, over 28888.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.366, pruned_loss=0.1152, over 5663917.28 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3527, pruned_loss=0.1048, over 5694879.23 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3651, pruned_loss=0.114, over 5676175.93 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:43:31,304 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:43:48,001 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 00:44:07,336 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1299570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:44:07,740 INFO [train.py:968] (1/2) Epoch 29, batch 23750, giga_loss[loss=0.3203, simple_loss=0.3847, pruned_loss=0.128, over 28657.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3728, pruned_loss=0.1212, over 5649447.18 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3531, pruned_loss=0.105, over 5691106.32 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3723, pruned_loss=0.1205, over 5661946.83 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:44:10,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1299573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:44:26,244 INFO [optim.py:369] (1/2) 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 00:44:38,513 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1299602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:44:57,562 INFO [train.py:968] (1/2) Epoch 29, batch 23800, libri_loss[loss=0.328, simple_loss=0.3861, pruned_loss=0.1349, over 25954.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3795, pruned_loss=0.1263, over 5650102.87 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3536, pruned_loss=0.1053, over 5691070.37 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.379, pruned_loss=0.1257, over 5659552.98 frames. ], batch size: 137, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:45:43,284 INFO [train.py:968] (1/2) Epoch 29, batch 23850, giga_loss[loss=0.3828, simple_loss=0.4336, pruned_loss=0.166, over 28627.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3802, pruned_loss=0.1277, over 5652608.60 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3534, pruned_loss=0.1055, over 5696974.38 frames. ], giga_tot_loss[loss=0.318, simple_loss=0.3807, pruned_loss=0.1277, over 5654088.04 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:45:44,597 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 00:46:32,899 INFO [train.py:968] (1/2) Epoch 29, batch 23900, giga_loss[loss=0.4026, simple_loss=0.4252, pruned_loss=0.19, over 23516.00 frames. ], tot_loss[loss=0.32, simple_loss=0.3812, pruned_loss=0.1294, over 5648738.03 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3536, pruned_loss=0.1057, over 5698686.45 frames. ], giga_tot_loss[loss=0.3206, simple_loss=0.382, pruned_loss=0.1296, over 5647598.58 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:46:51,240 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 00:47:15,901 INFO [train.py:968] (1/2) Epoch 29, batch 23950, giga_loss[loss=0.2919, simple_loss=0.3573, pruned_loss=0.1133, over 28872.00 frames. ], tot_loss[loss=0.3215, simple_loss=0.3818, pruned_loss=0.1306, over 5637925.80 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3532, pruned_loss=0.1057, over 5691876.91 frames. ], giga_tot_loss[loss=0.3247, simple_loss=0.3846, pruned_loss=0.1324, over 5639422.10 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:47:29,709 INFO [optim.py:369] (1/2) 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 00:47:54,507 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6335, 1.6384, 1.8110, 1.4220], device='cuda:1'), covar=tensor([0.1590, 0.2477, 0.1316, 0.1667], device='cuda:1'), in_proj_covar=tensor([0.0931, 0.0714, 0.0979, 0.0879], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 00:48:07,152 INFO [train.py:968] (1/2) Epoch 29, batch 24000, giga_loss[loss=0.3028, simple_loss=0.3726, pruned_loss=0.1165, over 28624.00 frames. ], tot_loss[loss=0.3285, simple_loss=0.3867, pruned_loss=0.1351, over 5624302.65 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3532, pruned_loss=0.1059, over 5685742.23 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3899, pruned_loss=0.1372, over 5629490.80 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:48:07,152 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 00:48:15,435 INFO [train.py:1012] (1/2) Epoch 29, validation: loss=0.2002, simple_loss=0.3074, pruned_loss=0.04648, over 944034.00 frames. 
+2023-03-15 00:48:15,436 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 00:49:03,991 INFO [train.py:968] (1/2) Epoch 29, batch 24050, giga_loss[loss=0.3601, simple_loss=0.4092, pruned_loss=0.1555, over 28965.00 frames. ], tot_loss[loss=0.3325, simple_loss=0.3888, pruned_loss=0.1381, over 5602987.75 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3541, pruned_loss=0.1068, over 5682827.00 frames. ], giga_tot_loss[loss=0.3357, simple_loss=0.3916, pruned_loss=0.1399, over 5607940.88 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:49:11,907 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1299878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:49:20,124 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.72 vs. limit=5.0
+2023-03-15 00:49:23,239 INFO [optim.py:369] (1/2) 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 00:49:54,226 INFO [train.py:968] (1/2) Epoch 29, batch 24100, giga_loss[loss=0.4753, simple_loss=0.4846, pruned_loss=0.233, over 26645.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3867, pruned_loss=0.1369, over 5618074.95 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3541, pruned_loss=0.1068, over 5687062.80 frames. ], giga_tot_loss[loss=0.3336, simple_loss=0.3896, pruned_loss=0.1389, over 5616671.08 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:50:41,242 INFO [train.py:968] (1/2) Epoch 29, batch 24150, giga_loss[loss=0.3124, simple_loss=0.3742, pruned_loss=0.1253, over 28659.00 frames. ], tot_loss[loss=0.3274, simple_loss=0.3846, pruned_loss=0.1351, over 5633794.41 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.3543, pruned_loss=0.1068, over 5690436.36 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3874, pruned_loss=0.1373, over 5628468.31 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:50:59,199 INFO [optim.py:369] (1/2) 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 00:51:27,784 INFO [train.py:968] (1/2) Epoch 29, batch 24200, giga_loss[loss=0.3349, simple_loss=0.3721, pruned_loss=0.1489, over 23618.00 frames. ], tot_loss[loss=0.3249, simple_loss=0.3834, pruned_loss=0.1332, over 5622323.05 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3541, pruned_loss=0.1067, over 5692595.79 frames. ], giga_tot_loss[loss=0.3291, simple_loss=0.3867, pruned_loss=0.1358, over 5614459.02 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:51:28,050 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:51:33,440 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:51:56,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:51:59,790 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:52:06,706 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:52:17,011 INFO [train.py:968] (1/2) Epoch 29, batch 24250, libri_loss[loss=0.2621, simple_loss=0.3262, pruned_loss=0.09896, over 29593.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3841, pruned_loss=0.1332, over 5630488.06 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3539, pruned_loss=0.1068, over 5700950.37 frames. ], giga_tot_loss[loss=0.3303, simple_loss=0.3881, pruned_loss=0.1362, over 5613717.70 frames. ], batch size: 74, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:52:34,798 INFO [optim.py:369] (1/2) 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 00:53:11,715 INFO [train.py:968] (1/2) Epoch 29, batch 24300, giga_loss[loss=0.2869, simple_loss=0.3592, pruned_loss=0.1073, over 28576.00 frames. ], tot_loss[loss=0.3259, simple_loss=0.3849, pruned_loss=0.1334, over 5631525.56 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.3542, pruned_loss=0.1071, over 5703124.35 frames. ], giga_tot_loss[loss=0.3301, simple_loss=0.3883, pruned_loss=0.1359, over 5615413.85 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:53:21,156 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:53:27,077 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-15 00:53:35,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.63 vs. limit=2.0
+2023-03-15 00:53:55,650 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3654, 1.1835, 3.9685, 3.3418], device='cuda:1'), covar=tensor([0.1688, 0.3011, 0.0455, 0.1605], device='cuda:1'), in_proj_covar=tensor([0.0814, 0.0678, 0.1018, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 00:53:58,513 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4312, 1.9057, 1.7959, 1.5595], device='cuda:1'), covar=tensor([0.2369, 0.1937, 0.2367, 0.2425], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0761, 0.0733, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 00:54:02,079 INFO [train.py:968] (1/2) Epoch 29, batch 24350, giga_loss[loss=0.3565, simple_loss=0.4065, pruned_loss=0.1533, over 27611.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3829, pruned_loss=0.1314, over 5621184.25 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3545, pruned_loss=0.1074, over 5697125.66 frames. ], giga_tot_loss[loss=0.3263, simple_loss=0.3858, pruned_loss=0.1335, over 5612963.79 frames. ], batch size: 474, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:54:04,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6044, 2.0575, 1.9204, 1.6808], device='cuda:1'), covar=tensor([0.2338, 0.2069, 0.2277, 0.2462], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0761, 0.0733, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 00:54:05,863 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4765, 1.7318, 1.3979, 1.3594], device='cuda:1'), covar=tensor([0.2730, 0.2874, 0.3280, 0.2431], device='cuda:1'), in_proj_covar=tensor([0.1610, 0.1164, 0.1424, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 00:54:17,847 INFO [optim.py:369] (1/2) 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 00:54:18,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:54:23,322 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:54:28,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6822, 2.4598, 1.7663, 0.8984], device='cuda:1'), covar=tensor([0.7305, 0.3931, 0.4768, 0.7170], device='cuda:1'), in_proj_covar=tensor([0.1856, 0.1745, 0.1671, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:54:50,584 INFO [train.py:968] (1/2) Epoch 29, batch 24400, giga_loss[loss=0.2961, simple_loss=0.3678, pruned_loss=0.1121, over 28817.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.38, pruned_loss=0.1279, over 5632685.89 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3544, pruned_loss=0.1074, over 5700455.32 frames. ], giga_tot_loss[loss=0.3213, simple_loss=0.3828, pruned_loss=0.1299, over 5622241.63 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 00:54:51,249 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 00:54:52,853 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:54:56,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1963, 4.0563, 3.8520, 1.8016], device='cuda:1'), covar=tensor([0.0688, 0.0783, 0.0840, 0.2043], device='cuda:1'), in_proj_covar=tensor([0.1315, 0.1216, 0.1021, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 00:55:37,935 INFO [train.py:968] (1/2) Epoch 29, batch 24450, giga_loss[loss=0.2825, simple_loss=0.3566, pruned_loss=0.1042, over 28288.00 frames. ], tot_loss[loss=0.3134, simple_loss=0.3765, pruned_loss=0.1251, over 5629558.70 frames. ], libri_tot_loss[loss=0.2842, simple_loss=0.354, pruned_loss=0.1072, over 5701517.07 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3794, pruned_loss=0.1271, over 5619513.58 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 00:55:56,203 INFO [optim.py:369] (1/2) 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 00:56:10,315 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4508, 1.3825, 4.3706, 3.6052], device='cuda:1'), covar=tensor([0.1841, 0.2842, 0.0685, 0.1330], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0678, 0.1017, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 00:56:18,564 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:56:22,195 INFO [train.py:968] (1/2) Epoch 29, batch 24500, giga_loss[loss=0.2877, simple_loss=0.3564, pruned_loss=0.1095, over 28865.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3732, pruned_loss=0.1228, over 5630627.61 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3541, pruned_loss=0.1075, over 5694829.17 frames. ], giga_tot_loss[loss=0.3127, simple_loss=0.3761, pruned_loss=0.1247, over 5626580.19 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:56:33,309 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:57:06,533 INFO [train.py:968] (1/2) Epoch 29, batch 24550, giga_loss[loss=0.3165, simple_loss=0.3794, pruned_loss=0.1268, over 28771.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3726, pruned_loss=0.1227, over 5624640.03 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3545, pruned_loss=0.1079, over 5690872.20 frames. ], giga_tot_loss[loss=0.3118, simple_loss=0.3751, pruned_loss=0.1243, over 5623869.35 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:57:22,574 INFO [optim.py:369] (1/2) 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 00:57:54,603 INFO [train.py:968] (1/2) Epoch 29, batch 24600, giga_loss[loss=0.3101, simple_loss=0.3755, pruned_loss=0.1224, over 28741.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3726, pruned_loss=0.1229, over 5633207.60 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3539, pruned_loss=0.1077, over 5689687.14 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.3756, pruned_loss=0.1248, over 5631983.04 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:58:10,948 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:58:19,974 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:58:45,935 INFO [train.py:968] (1/2) Epoch 29, batch 24650, giga_loss[loss=0.2757, simple_loss=0.3464, pruned_loss=0.1025, over 28838.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3717, pruned_loss=0.1218, over 5642070.63 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3541, pruned_loss=0.1079, over 5689988.75 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3741, pruned_loss=0.1233, over 5640172.02 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:59:07,111 INFO [optim.py:369] (1/2) 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 00:59:23,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 00:59:36,916 INFO [train.py:968] (1/2) Epoch 29, batch 24700, giga_loss[loss=0.3163, simple_loss=0.385, pruned_loss=0.1237, over 27534.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3719, pruned_loss=0.1198, over 5646352.57 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3544, pruned_loss=0.108, over 5690813.60 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3738, pruned_loss=0.1211, over 5643548.65 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 00:59:43,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3525, 1.9763, 1.3561, 0.6685], device='cuda:1'), covar=tensor([0.6937, 0.3258, 0.4519, 0.7234], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1736, 0.1661, 0.1505], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 00:59:59,684 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2608, 1.6512, 1.2400, 0.9597], device='cuda:1'), covar=tensor([0.2644, 0.2593, 0.3133, 0.2342], device='cuda:1'), in_proj_covar=tensor([0.1607, 0.1162, 0.1423, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 01:00:27,717 INFO [train.py:968] (1/2) Epoch 29, batch 24750, giga_loss[loss=0.2904, simple_loss=0.3686, pruned_loss=0.1061, over 28984.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3725, pruned_loss=0.1182, over 5647936.84 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3545, pruned_loss=0.1082, over 5682840.50 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3743, pruned_loss=0.1193, over 5652713.91 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:00:29,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2924, 1.2465, 1.2134, 1.4294], device='cuda:1'), covar=tensor([0.0784, 0.0428, 0.0372, 0.0904], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 01:00:37,128 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:00:38,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:00:47,971 INFO [optim.py:369] (1/2) 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 01:01:03,439 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:01:04,370 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 01:01:12,462 INFO [train.py:968] (1/2) Epoch 29, batch 24800, giga_loss[loss=0.3032, simple_loss=0.3723, pruned_loss=0.1171, over 28993.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3714, pruned_loss=0.1181, over 5652336.82 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3542, pruned_loss=0.1081, over 5688095.73 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.374, pruned_loss=0.1195, over 5649582.94 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:01:20,301 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6965, 1.8201, 1.2620, 1.4560], device='cuda:1'), covar=tensor([0.1048, 0.0644, 0.1112, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0453, 0.0527, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 01:01:42,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:01:44,721 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 24850, giga_loss[loss=0.2963, simple_loss=0.3733, pruned_loss=0.1096, over 28984.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3712, pruned_loss=0.118, over 5661186.14 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.3543, pruned_loss=0.1083, over 5680402.68 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3736, pruned_loss=0.1192, over 5665106.86 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:02:11,898 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:21,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:21,757 INFO [optim.py:369] (1/2) 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 01:02:38,779 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:40,110 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1300710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:02:49,549 INFO [train.py:968] (1/2) Epoch 29, batch 24900, giga_loss[loss=0.281, simple_loss=0.3511, pruned_loss=0.1054, over 28814.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3694, pruned_loss=0.1172, over 5671018.18 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3546, pruned_loss=0.1085, over 5681407.50 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3714, pruned_loss=0.1181, over 5672946.44 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:03:24,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9814, 3.8054, 3.6493, 1.8758], device='cuda:1'), covar=tensor([0.0749, 0.0904, 0.0848, 0.1957], device='cuda:1'), in_proj_covar=tensor([0.1319, 0.1218, 0.1024, 0.0760], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 01:03:31,840 INFO [train.py:968] (1/2) Epoch 29, batch 24950, giga_loss[loss=0.3224, simple_loss=0.3828, pruned_loss=0.131, over 29145.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3696, pruned_loss=0.1187, over 5674844.31 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3548, pruned_loss=0.1086, over 5685853.26 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3715, pruned_loss=0.1197, over 5672117.57 frames. ], batch size: 113, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:03:48,455 INFO [optim.py:369] (1/2) 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 01:04:17,640 INFO [train.py:968] (1/2) Epoch 29, batch 25000, giga_loss[loss=0.2947, simple_loss=0.3643, pruned_loss=0.1126, over 28824.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3683, pruned_loss=0.118, over 5670482.94 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.355, pruned_loss=0.1088, over 5682807.60 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3702, pruned_loss=0.119, over 5669864.92 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:04:17,928 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1300821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:28,476 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:30,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:44,622 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:46,680 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1300854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:04:55,664 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:05:03,601 INFO [train.py:968] (1/2) Epoch 29, batch 25050, giga_loss[loss=0.2893, simple_loss=0.3659, pruned_loss=0.1064, over 28951.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3682, pruned_loss=0.1163, over 5668128.35 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3554, pruned_loss=0.1092, over 5675097.72 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3695, pruned_loss=0.1168, over 5673535.49 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:05:13,227 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:05:20,732 INFO [optim.py:369] (1/2) 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 01:05:24,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4256, 3.6635, 1.5427, 1.6566], device='cuda:1'), covar=tensor([0.1074, 0.0435, 0.0979, 0.1384], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0415, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 01:05:40,606 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 01:05:44,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6671, 1.8702, 1.5366, 1.8706], device='cuda:1'), covar=tensor([0.2740, 0.2925, 0.3221, 0.2706], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1164, 0.1428, 0.1016], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 01:05:50,605 INFO [train.py:968] (1/2) Epoch 29, batch 25100, giga_loss[loss=0.3275, simple_loss=0.3977, pruned_loss=0.1286, over 28679.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3681, pruned_loss=0.1155, over 5667825.39 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3554, pruned_loss=0.1092, over 5674742.61 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3692, pruned_loss=0.1159, over 5672270.48 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:06:31,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1300964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:06:33,032 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 25150, giga_loss[loss=0.3034, simple_loss=0.3571, pruned_loss=0.1248, over 28570.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3667, pruned_loss=0.1146, over 5673877.48 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3551, pruned_loss=0.109, over 5679783.30 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3682, pruned_loss=0.1153, over 5672655.99 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:06:50,417 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4251, 1.6274, 1.1536, 1.1680], device='cuda:1'), covar=tensor([0.1156, 0.0625, 0.1213, 0.1269], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0453, 0.0527, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 01:06:58,101 INFO [optim.py:369] (1/2) 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 01:07:02,902 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1300996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:07:22,571 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5988, 1.7137, 1.6471, 1.5409], device='cuda:1'), covar=tensor([0.3008, 0.2772, 0.2599, 0.2609], device='cuda:1'), in_proj_covar=tensor([0.2098, 0.2061, 0.1960, 0.2109], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 01:07:25,601 INFO [train.py:968] (1/2) Epoch 29, batch 25200, giga_loss[loss=0.3338, simple_loss=0.3922, pruned_loss=0.1378, over 27511.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3668, pruned_loss=0.1154, over 5676370.38 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3555, pruned_loss=0.1092, over 5680634.05 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3678, pruned_loss=0.1159, over 5674616.50 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:07:56,706 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 25250, giga_loss[loss=0.345, simple_loss=0.3997, pruned_loss=0.1452, over 28569.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3656, pruned_loss=0.1157, over 5666367.97 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3555, pruned_loss=0.1093, over 5687680.20 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3668, pruned_loss=0.1162, over 5658199.10 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:08:17,221 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4633, 3.7750, 1.6117, 1.6379], device='cuda:1'), covar=tensor([0.1033, 0.0310, 0.0888, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0415, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 01:08:22,944 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:08:28,553 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.37 vs. limit=2.0
+2023-03-15 01:08:28,649 INFO [optim.py:369] (1/2) 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 01:08:54,910 INFO [train.py:968] (1/2) Epoch 29, batch 25300, giga_loss[loss=0.3129, simple_loss=0.3761, pruned_loss=0.1249, over 28669.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3656, pruned_loss=0.1165, over 5676165.65 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3558, pruned_loss=0.1095, over 5695906.58 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3667, pruned_loss=0.1169, over 5661333.70 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:08:55,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5541, 2.4755, 2.6129, 2.2840], device='cuda:1'), covar=tensor([0.2173, 0.2660, 0.2040, 0.2433], device='cuda:1'), in_proj_covar=tensor([0.0510, 0.0761, 0.0736, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 01:09:11,565 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3500, 3.3793, 1.5089, 1.4406], device='cuda:1'), covar=tensor([0.0948, 0.0372, 0.0886, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0577, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 01:09:36,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2611, 0.8353, 0.9576, 1.4450], device='cuda:1'), covar=tensor([0.0745, 0.0406, 0.0364, 0.0826], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0105, 0.0075, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 01:09:39,619 INFO [train.py:968] (1/2) Epoch 29, batch 25350, giga_loss[loss=0.3973, simple_loss=0.4281, pruned_loss=0.1832, over 27979.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3653, pruned_loss=0.1169, over 5678687.54 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3558, pruned_loss=0.1096, over 5701374.30 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3664, pruned_loss=0.1174, over 5661648.25 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:09:48,779 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1301179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:10:00,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 01:10:01,023 INFO [optim.py:369] (1/2) 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 01:10:27,275 INFO [train.py:968] (1/2) Epoch 29, batch 25400, libri_loss[loss=0.2467, simple_loss=0.3237, pruned_loss=0.08489, over 29533.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.3641, pruned_loss=0.1168, over 5679029.51 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3559, pruned_loss=0.1097, over 5700050.80 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3651, pruned_loss=0.1172, over 5666119.94 frames. ], batch size: 80, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:10:34,194 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:10:35,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:11:04,774 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1301260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:11:15,017 INFO [train.py:968] (1/2) Epoch 29, batch 25450, libri_loss[loss=0.323, simple_loss=0.3874, pruned_loss=0.1293, over 26024.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3642, pruned_loss=0.1174, over 5668753.46 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3558, pruned_loss=0.1097, over 5701453.95 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3652, pruned_loss=0.1179, over 5657036.70 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:11:31,744 INFO [optim.py:369] (1/2) 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 01:11:57,105 INFO [train.py:968] (1/2) Epoch 29, batch 25500, libri_loss[loss=0.2503, simple_loss=0.3178, pruned_loss=0.09133, over 29396.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.364, pruned_loss=0.1167, over 5677658.28 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3555, pruned_loss=0.1095, over 5710521.61 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3656, pruned_loss=0.1177, over 5658177.90 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:12:39,848 INFO [train.py:968] (1/2) Epoch 29, batch 25550, giga_loss[loss=0.2799, simple_loss=0.3568, pruned_loss=0.1015, over 28971.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3633, pruned_loss=0.1148, over 5679290.74 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3553, pruned_loss=0.1094, over 5710054.15 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3648, pruned_loss=0.1157, over 5663945.47 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:12:58,048 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 25600, libri_loss[loss=0.258, simple_loss=0.3263, pruned_loss=0.09487, over 29523.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3637, pruned_loss=0.115, over 5663606.61 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3559, pruned_loss=0.11, over 5702616.33 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3647, pruned_loss=0.1154, over 5656231.24 frames. ], batch size: 70, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:13:29,534 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:14:06,461 INFO [train.py:968] (1/2) Epoch 29, batch 25650, giga_loss[loss=0.2941, simple_loss=0.3663, pruned_loss=0.111, over 28862.00 frames. ], tot_loss[loss=0.2973, simple_loss=0.364, pruned_loss=0.1153, over 5668096.46 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.356, pruned_loss=0.1101, over 5703621.69 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3649, pruned_loss=0.1157, over 5660739.47 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:14:28,513 INFO [optim.py:369] (1/2) 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 01:14:40,872 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5176, 1.6462, 1.7635, 1.3001], device='cuda:1'), covar=tensor([0.1712, 0.2572, 0.1430, 0.1745], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0719, 0.0984, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 01:14:57,225 INFO [train.py:968] (1/2) Epoch 29, batch 25700, giga_loss[loss=0.3012, simple_loss=0.3651, pruned_loss=0.1187, over 28449.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.367, pruned_loss=0.1182, over 5653564.95 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3557, pruned_loss=0.1099, over 5703641.30 frames. ], giga_tot_loss[loss=0.3027, simple_loss=0.3681, pruned_loss=0.1187, over 5647025.81 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:15:06,778 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1301529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:15:25,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-15 01:15:29,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:15:47,278 INFO [train.py:968] (1/2) Epoch 29, batch 25750, giga_loss[loss=0.2661, simple_loss=0.3416, pruned_loss=0.09533, over 28932.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3682, pruned_loss=0.12, over 5658825.63 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3561, pruned_loss=0.1101, over 5706414.62 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3689, pruned_loss=0.1204, over 5650281.96 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:15:48,168 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:15:50,297 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:16:06,283 INFO [optim.py:369] (1/2) 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 01:16:15,263 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1301604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:16:19,087 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4837, 1.7771, 1.6738, 1.5680], device='cuda:1'), covar=tensor([0.2184, 0.1995, 0.2401, 0.2126], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0761, 0.0734, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 01:16:32,880 INFO [train.py:968] (1/2) Epoch 29, batch 25800, giga_loss[loss=0.2749, simple_loss=0.3428, pruned_loss=0.1035, over 28439.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3692, pruned_loss=0.1213, over 5663520.38 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3565, pruned_loss=0.1104, over 5702163.39 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3697, pruned_loss=0.1217, over 5659529.76 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:17:02,583 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2286, 1.4290, 1.4697, 1.0965], device='cuda:1'), covar=tensor([0.1673, 0.2538, 0.1424, 0.1678], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0718, 0.0982, 0.0883], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 01:17:16,938 INFO [train.py:968] (1/2) Epoch 29, batch 25850, giga_loss[loss=0.4089, simple_loss=0.4349, pruned_loss=0.1915, over 28548.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3711, pruned_loss=0.1235, over 5654397.83 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3562, pruned_loss=0.11, over 5709085.56 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3725, pruned_loss=0.1247, over 5642814.95 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:17:36,117 INFO [optim.py:369] (1/2) 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 01:17:38,945 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1301697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:17:42,416 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1301700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:17:47,359 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.82 vs. limit=5.0
+2023-03-15 01:17:58,258 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6000, 1.8034, 1.2175, 1.3464], device='cuda:1'), covar=tensor([0.1003, 0.0599, 0.1089, 0.1181], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0455, 0.0528, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 01:18:00,422 INFO [train.py:968] (1/2) Epoch 29, batch 25900, giga_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1127, over 28844.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3693, pruned_loss=0.1223, over 5668993.65 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3558, pruned_loss=0.1099, over 5712676.07 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3712, pruned_loss=0.1237, over 5655037.30 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:18:06,965 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1301729.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:18:42,910 INFO [train.py:968] (1/2) Epoch 29, batch 25950, giga_loss[loss=0.3865, simple_loss=0.4259, pruned_loss=0.1735, over 26551.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.123, over 5671496.41 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.356, pruned_loss=0.11, over 5721833.67 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3725, pruned_loss=0.1247, over 5649568.83 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:18:51,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3266, 2.8201, 1.4312, 1.4225], device='cuda:1'), covar=tensor([0.1010, 0.0384, 0.0896, 0.1390], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0577, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 01:19:02,022 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 26000, giga_loss[loss=0.2555, simple_loss=0.3378, pruned_loss=0.08655, over 29022.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.369, pruned_loss=0.12, over 5678283.45 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3561, pruned_loss=0.11, over 5719559.38 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3706, pruned_loss=0.1214, over 5662513.03 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:20:16,755 INFO [train.py:968] (1/2) Epoch 29, batch 26050, giga_loss[loss=0.2819, simple_loss=0.3535, pruned_loss=0.1052, over 28578.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1191, over 5665875.18 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.356, pruned_loss=0.1099, over 5720447.58 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3688, pruned_loss=0.1203, over 5652682.97 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:20:40,662 INFO [optim.py:369] (1/2) 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 01:20:46,870 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6298, 1.8472, 1.7558, 1.5597], device='cuda:1'), covar=tensor([0.3377, 0.2825, 0.2659, 0.2921], device='cuda:1'), in_proj_covar=tensor([0.2094, 0.2066, 0.1959, 0.2105], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 01:20:48,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1301904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:21:04,935 INFO [train.py:968] (1/2) Epoch 29, batch 26100, giga_loss[loss=0.3146, simple_loss=0.3774, pruned_loss=0.126, over 28923.00 frames. ], tot_loss[loss=0.3008, simple_loss=0.3651, pruned_loss=0.1182, over 5672882.19 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3561, pruned_loss=0.11, over 5722359.44 frames. ], giga_tot_loss[loss=0.3023, simple_loss=0.3662, pruned_loss=0.1192, over 5660385.73 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:21:53,698 INFO [train.py:968] (1/2) Epoch 29, batch 26150, giga_loss[loss=0.3074, simple_loss=0.3841, pruned_loss=0.1154, over 28983.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.3638, pruned_loss=0.117, over 5680230.93 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3565, pruned_loss=0.1103, over 5716971.06 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3646, pruned_loss=0.1177, over 5673547.34 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:22:16,040 INFO [optim.py:369] (1/2) 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 01:22:41,362 INFO [train.py:968] (1/2) Epoch 29, batch 26200, giga_loss[loss=0.2865, simple_loss=0.3774, pruned_loss=0.0978, over 28548.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.367, pruned_loss=0.1189, over 5671344.73 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3568, pruned_loss=0.1105, over 5706008.68 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3675, pruned_loss=0.1194, over 5674689.81 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:23:03,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1302047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:23:07,180 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1302050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:23:09,906 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2845, 1.5725, 1.5536, 1.2991], device='cuda:1'), covar=tensor([0.3677, 0.3256, 0.2491, 0.3038], device='cuda:1'), in_proj_covar=tensor([0.2089, 0.2062, 0.1955, 0.2103], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 01:23:13,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1701, 1.6000, 0.9721, 1.1175], device='cuda:1'), covar=tensor([0.1468, 0.0724, 0.1624, 0.1479], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0527, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 01:23:25,426 INFO [train.py:968] (1/2) Epoch 29, batch 26250, giga_loss[loss=0.3116, simple_loss=0.3955, pruned_loss=0.1138, over 28996.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3708, pruned_loss=0.1188, over 5675541.33 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3573, pruned_loss=0.1109, over 5709939.13 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.371, pruned_loss=0.119, over 5673994.10 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:23:34,073 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1302079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:23:47,273 INFO [optim.py:369] (1/2) 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 01:24:00,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7218, 2.0000, 1.3896, 1.5634], device='cuda:1'), covar=tensor([0.1081, 0.0655, 0.1025, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0527, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 01:24:10,075 INFO [train.py:968] (1/2) Epoch 29, batch 26300, giga_loss[loss=0.2649, simple_loss=0.3466, pruned_loss=0.09162, over 28529.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3714, pruned_loss=0.1177, over 5685236.79 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3574, pruned_loss=0.1111, over 5715963.31 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3719, pruned_loss=0.118, over 5677595.93 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:24:31,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8654, 1.9274, 2.0089, 1.6166], device='cuda:1'), covar=tensor([0.1894, 0.2520, 0.1558, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0720, 0.0985, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 01:24:55,813 INFO [train.py:968] (1/2) Epoch 29, batch 26350, giga_loss[loss=0.2811, simple_loss=0.3481, pruned_loss=0.1071, over 28592.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3721, pruned_loss=0.119, over 5682114.72 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3572, pruned_loss=0.1113, over 5712274.75 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3735, pruned_loss=0.1194, over 5677388.71 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:25:03,417 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 01:25:07,021 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2424, 1.6860, 1.2498, 0.6032], device='cuda:1'), covar=tensor([0.4520, 0.2324, 0.2908, 0.6269], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1749, 0.1662, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 01:25:16,663 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 26400, giga_loss[loss=0.3082, simple_loss=0.3739, pruned_loss=0.1213, over 28853.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.3752, pruned_loss=0.1216, over 5675445.81 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3579, pruned_loss=0.1119, over 5703774.37 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.376, pruned_loss=0.1217, over 5678202.06 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:26:11,912 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1302255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:26:26,812 INFO [train.py:968] (1/2) Epoch 29, batch 26450, giga_loss[loss=0.3932, simple_loss=0.4145, pruned_loss=0.1859, over 23389.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3765, pruned_loss=0.1238, over 5669592.97 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3577, pruned_loss=0.1116, over 5705592.61 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3778, pruned_loss=0.1243, over 5669162.37 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:26:39,935 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 01:26:46,495 INFO [optim.py:369] (1/2) 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 01:27:12,831 INFO [train.py:968] (1/2) Epoch 29, batch 26500, giga_loss[loss=0.2626, simple_loss=0.3399, pruned_loss=0.09258, over 29025.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3754, pruned_loss=0.1234, over 5679710.08 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.358, pruned_loss=0.1119, over 5705451.24 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3766, pruned_loss=0.1239, over 5678536.47 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:27:19,742 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5405, 1.7850, 1.4440, 1.3955], device='cuda:1'), covar=tensor([0.2725, 0.2907, 0.3335, 0.2512], device='cuda:1'), in_proj_covar=tensor([0.1613, 0.1162, 0.1428, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 01:27:58,170 INFO [train.py:968] (1/2) Epoch 29, batch 26550, libri_loss[loss=0.2547, simple_loss=0.3305, pruned_loss=0.08951, over 29566.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3711, pruned_loss=0.1209, over 5680347.04 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1114, over 5707000.37 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.373, pruned_loss=0.1219, over 5677627.25 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:28:17,182 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1302394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:28:18,238 INFO [optim.py:369] (1/2) 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 01:28:48,219 INFO [train.py:968] (1/2) Epoch 29, batch 26600, giga_loss[loss=0.2682, simple_loss=0.3341, pruned_loss=0.1011, over 28991.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3685, pruned_loss=0.1194, over 5684479.92 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3572, pruned_loss=0.1113, over 5700409.31 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3704, pruned_loss=0.1206, over 5687087.72 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:29:32,238 INFO [train.py:968] (1/2) Epoch 29, batch 26650, libri_loss[loss=0.3428, simple_loss=0.3985, pruned_loss=0.1436, over 29518.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.368, pruned_loss=0.1195, over 5679346.74 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.357, pruned_loss=0.1111, over 5705891.09 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.37, pruned_loss=0.1209, over 5675920.16 frames. ], batch size: 89, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:29:54,394 INFO [optim.py:369] (1/2) 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 01:30:16,871 INFO [train.py:968] (1/2) Epoch 29, batch 26700, giga_loss[loss=0.258, simple_loss=0.3322, pruned_loss=0.09188, over 29029.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.368, pruned_loss=0.1197, over 5683605.37 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3575, pruned_loss=0.1115, over 5708707.38 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3694, pruned_loss=0.1205, over 5678040.57 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:31:00,217 INFO [train.py:968] (1/2) Epoch 29, batch 26750, giga_loss[loss=0.2663, simple_loss=0.3412, pruned_loss=0.09568, over 28895.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3677, pruned_loss=0.1204, over 5677266.29 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3575, pruned_loss=0.1115, over 5715649.36 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3691, pruned_loss=0.1213, over 5665901.05 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:31:25,525 INFO [optim.py:369] (1/2) 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 01:31:37,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2735, 2.5746, 1.2530, 1.4058], device='cuda:1'), covar=tensor([0.1047, 0.0373, 0.0930, 0.1443], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0576, 0.0413, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 01:31:48,311 INFO [train.py:968] (1/2) Epoch 29, batch 26800, giga_loss[loss=0.3214, simple_loss=0.3928, pruned_loss=0.1251, over 28240.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.367, pruned_loss=0.1201, over 5665370.90 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3577, pruned_loss=0.1115, over 5718660.95 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3681, pruned_loss=0.1209, over 5652681.40 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:31:58,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1302630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:32:34,355 INFO [train.py:968] (1/2) Epoch 29, batch 26850, giga_loss[loss=0.2986, simple_loss=0.3689, pruned_loss=0.1142, over 28731.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3676, pruned_loss=0.1193, over 5670436.10 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5717919.76 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1202, over 5660418.96 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:32:58,260 INFO [optim.py:369] (1/2) 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 01:33:09,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9752, 2.2097, 1.8118, 2.2009], device='cuda:1'), covar=tensor([0.2587, 0.2727, 0.3168, 0.2440], device='cuda:1'), in_proj_covar=tensor([0.1612, 0.1162, 0.1427, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 01:33:21,746 INFO [train.py:968] (1/2) Epoch 29, batch 26900, giga_loss[loss=0.3374, simple_loss=0.3879, pruned_loss=0.1435, over 28819.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3707, pruned_loss=0.1216, over 5663049.92 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3575, pruned_loss=0.1114, over 5718206.88 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3719, pruned_loss=0.1225, over 5653823.96 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:33:59,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3900, 1.7150, 1.3889, 1.5958], device='cuda:1'), covar=tensor([0.0818, 0.0313, 0.0345, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 01:34:07,982 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 26950, giga_loss[loss=0.3526, simple_loss=0.3958, pruned_loss=0.1547, over 28837.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3703, pruned_loss=0.1218, over 5652946.23 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3576, pruned_loss=0.1115, over 5702652.35 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3714, pruned_loss=0.1226, over 5657077.18 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:34:11,148 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1302773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:34:13,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1302776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:34:31,289 INFO [optim.py:369] (1/2) 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 01:34:34,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2070, 4.9865, 4.7138, 2.4717], device='cuda:1'), covar=tensor([0.0576, 0.0725, 0.0765, 0.1864], device='cuda:1'), in_proj_covar=tensor([0.1332, 0.1230, 0.1034, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 01:34:38,838 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1302805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:34:52,573 INFO [train.py:968] (1/2) Epoch 29, batch 27000, giga_loss[loss=0.3052, simple_loss=0.3722, pruned_loss=0.1192, over 28311.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3714, pruned_loss=0.1199, over 5663452.63 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3576, pruned_loss=0.1114, over 5705989.33 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3725, pruned_loss=0.1208, over 5663152.10 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:34:52,573 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 01:35:02,466 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 01:35:19,123 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7382, 2.0129, 2.0856, 1.7496], device='cuda:1'), covar=tensor([0.3747, 0.2772, 0.2787, 0.2930], device='cuda:1'), in_proj_covar=tensor([0.2083, 0.2055, 0.1955, 0.2102], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 01:35:51,331 INFO [train.py:968] (1/2) Epoch 29, batch 27050, giga_loss[loss=0.2884, simple_loss=0.3652, pruned_loss=0.1058, over 28711.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3721, pruned_loss=0.1178, over 5675037.32 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3574, pruned_loss=0.1114, over 5707838.36 frames. ], giga_tot_loss[loss=0.3052, simple_loss=0.3733, pruned_loss=0.1186, over 5672926.26 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:36:12,270 INFO [optim.py:369] (1/2) 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 01:36:25,117 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1302912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:36:27,942 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 27100, giga_loss[loss=0.3245, simple_loss=0.3877, pruned_loss=0.1306, over 28994.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3736, pruned_loss=0.1177, over 5689963.82 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3573, pruned_loss=0.1112, over 5711306.45 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3748, pruned_loss=0.1186, over 5684736.68 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:36:53,417 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1302944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:37:19,314 INFO [train.py:968] (1/2) Epoch 29, batch 27150, giga_loss[loss=0.3368, simple_loss=0.3965, pruned_loss=0.1386, over 28624.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3767, pruned_loss=0.1211, over 5685967.99 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3575, pruned_loss=0.1113, over 5714413.40 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.378, pruned_loss=0.122, over 5678270.90 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:37:45,747 INFO [optim.py:369] (1/2) 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 01:38:09,338 INFO [train.py:968] (1/2) Epoch 29, batch 27200, giga_loss[loss=0.2843, simple_loss=0.3599, pruned_loss=0.1043, over 28795.00 frames. ], tot_loss[loss=0.3133, simple_loss=0.3784, pruned_loss=0.1241, over 5665419.22 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3572, pruned_loss=0.1111, over 5716067.84 frames. ], giga_tot_loss[loss=0.3152, simple_loss=0.3801, pruned_loss=0.1252, over 5657351.04 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:38:32,920 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:38:51,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3225, 1.5369, 1.5861, 1.2957], device='cuda:1'), covar=tensor([0.3816, 0.3004, 0.2437, 0.3123], device='cuda:1'), in_proj_covar=tensor([0.2082, 0.2055, 0.1951, 0.2098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 01:38:59,018 INFO [train.py:968] (1/2) Epoch 29, batch 27250, giga_loss[loss=0.3074, simple_loss=0.374, pruned_loss=0.1204, over 28795.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.3763, pruned_loss=0.1232, over 5658738.16 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3575, pruned_loss=0.1114, over 5710634.69 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3777, pruned_loss=0.124, over 5656296.93 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:39:16,265 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 01:39:26,077 INFO [optim.py:369] (1/2) 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 01:39:45,816 INFO [train.py:968] (1/2) Epoch 29, batch 27300, giga_loss[loss=0.283, simple_loss=0.3625, pruned_loss=0.1017, over 28441.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.376, pruned_loss=0.1229, over 5643549.89 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3578, pruned_loss=0.1117, over 5702917.37 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3772, pruned_loss=0.1234, over 5646993.86 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:40:25,109 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 27350, libri_loss[loss=0.2497, simple_loss=0.3323, pruned_loss=0.08357, over 29531.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3747, pruned_loss=0.1202, over 5641473.35 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3578, pruned_loss=0.1118, over 5690446.88 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3761, pruned_loss=0.1208, over 5652581.36 frames. ], batch size: 81, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:40:56,414 INFO [optim.py:369] (1/2) 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 01:41:09,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:41:15,800 INFO [train.py:968] (1/2) Epoch 29, batch 27400, libri_loss[loss=0.2711, simple_loss=0.3401, pruned_loss=0.1011, over 29574.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3744, pruned_loss=0.1189, over 5654017.57 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3577, pruned_loss=0.1119, over 5687055.76 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3761, pruned_loss=0.1194, over 5665547.66 frames. ], batch size: 76, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:41:33,452 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:42:02,489 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-15 01:42:05,728 INFO [train.py:968] (1/2) Epoch 29, batch 27450, giga_loss[loss=0.317, simple_loss=0.3861, pruned_loss=0.1239, over 28783.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3753, pruned_loss=0.12, over 5654029.28 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3579, pruned_loss=0.1122, over 5691986.76 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3769, pruned_loss=0.1204, over 5657949.64 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:42:32,671 INFO [optim.py:369] (1/2) 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 01:42:50,584 INFO [train.py:968] (1/2) Epoch 29, batch 27500, giga_loss[loss=0.3138, simple_loss=0.3757, pruned_loss=0.1259, over 29102.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3746, pruned_loss=0.1198, over 5658745.90 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3584, pruned_loss=0.1126, over 5687540.40 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.3758, pruned_loss=0.12, over 5664533.33 frames. ], batch size: 113, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:43:30,084 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-15 01:43:40,824 INFO [train.py:968] (1/2) Epoch 29, batch 27550, giga_loss[loss=0.4068, simple_loss=0.4202, pruned_loss=0.1968, over 23329.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3719, pruned_loss=0.1195, over 5653428.70 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3581, pruned_loss=0.1124, over 5689937.44 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3734, pruned_loss=0.1198, over 5655444.15 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 01:43:45,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303376.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:44:05,745 INFO [optim.py:369] (1/2) 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 01:44:27,071 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:44:27,558 INFO [train.py:968] (1/2) Epoch 29, batch 27600, libri_loss[loss=0.2488, simple_loss=0.3163, pruned_loss=0.09065, over 29653.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3699, pruned_loss=0.1195, over 5636845.76 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3579, pruned_loss=0.1124, over 5686149.76 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3718, pruned_loss=0.1201, over 5639737.52 frames. ], batch size: 73, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:45:16,453 INFO [train.py:968] (1/2) Epoch 29, batch 27650, giga_loss[loss=0.29, simple_loss=0.3621, pruned_loss=0.1089, over 28712.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3675, pruned_loss=0.1181, over 5648215.78 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3576, pruned_loss=0.1122, over 5689763.05 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3695, pruned_loss=0.1189, over 5646376.16 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:45:26,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4123, 1.8474, 1.5882, 1.4935], device='cuda:1'), covar=tensor([0.0704, 0.0404, 0.0323, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 01:45:38,822 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:45:41,179 INFO [optim.py:369] (1/2) 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 01:46:02,384 INFO [train.py:968] (1/2) Epoch 29, batch 27700, libri_loss[loss=0.2725, simple_loss=0.3411, pruned_loss=0.1019, over 29593.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3673, pruned_loss=0.1196, over 5640445.02 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3574, pruned_loss=0.1119, over 5687988.54 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3695, pruned_loss=0.1207, over 5638226.49 frames. ], batch size: 76, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:46:19,897 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:46:38,590 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:46:41,833 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:46:46,169 INFO [train.py:968] (1/2) Epoch 29, batch 27750, giga_loss[loss=0.2551, simple_loss=0.3336, pruned_loss=0.08827, over 28918.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.366, pruned_loss=0.1189, over 5648323.38 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3571, pruned_loss=0.1117, over 5693865.64 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3684, pruned_loss=0.1202, over 5639791.24 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:46:59,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:47:06,904 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:47:08,548 INFO [optim.py:369] (1/2) 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 01:47:22,756 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:47:31,677 INFO [train.py:968] (1/2) Epoch 29, batch 27800, giga_loss[loss=0.3019, simple_loss=0.3638, pruned_loss=0.12, over 26573.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3637, pruned_loss=0.1162, over 5656569.19 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.357, pruned_loss=0.1117, over 5694489.96 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3658, pruned_loss=0.1175, over 5648335.09 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:47:38,043 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 01:47:44,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3246, 1.9260, 1.3398, 0.6696], device='cuda:1'), covar=tensor([0.7031, 0.3357, 0.4877, 0.7685], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1743, 0.1660, 0.1509], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 01:48:12,856 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 27850, giga_loss[loss=0.2578, simple_loss=0.3411, pruned_loss=0.08725, over 28679.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3598, pruned_loss=0.1119, over 5658230.77 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.357, pruned_loss=0.1117, over 5687790.42 frames. ], giga_tot_loss[loss=0.2937, simple_loss=0.3615, pruned_loss=0.1129, over 5657032.98 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:48:30,156 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:48:32,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:48:35,923 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9035, 1.1969, 1.3542, 1.0069], device='cuda:1'), covar=tensor([0.2211, 0.1565, 0.2458, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.0510, 0.0762, 0.0735, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 01:48:38,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303691.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:48:42,869 INFO [optim.py:369] (1/2) 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 01:48:49,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0436, 1.5216, 5.3073, 3.7559], device='cuda:1'), covar=tensor([0.1588, 0.2871, 0.0389, 0.0853], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0682, 0.1022, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 01:49:02,274 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303715.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:03,448 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 01:49:03,907 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.1887, 2.1508, 4.4720, 3.8928], device='cuda:1'), covar=tensor([0.1191, 0.2352, 0.0472, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0682, 0.1022, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 01:49:06,382 INFO [train.py:968] (1/2) Epoch 29, batch 27900, giga_loss[loss=0.2808, simple_loss=0.334, pruned_loss=0.1138, over 23610.00 frames. ], tot_loss[loss=0.2918, simple_loss=0.3596, pruned_loss=0.112, over 5646197.58 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3573, pruned_loss=0.1118, over 5685002.02 frames. ], giga_tot_loss[loss=0.2931, simple_loss=0.3609, pruned_loss=0.1127, over 5647492.44 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:49:15,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:18,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:36,363 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303751.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:49:43,179 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303757.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:46,931 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:48,614 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:49:56,763 INFO [train.py:968] (1/2) Epoch 29, batch 27950, giga_loss[loss=0.2421, simple_loss=0.3134, pruned_loss=0.08544, over 29032.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.3567, pruned_loss=0.1105, over 5653132.42 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3568, pruned_loss=0.1115, over 5688180.44 frames. ], giga_tot_loss[loss=0.2904, simple_loss=0.3581, pruned_loss=0.1114, over 5650537.64 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:50:10,470 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4275, 1.8147, 1.7420, 1.4626], device='cuda:1'), covar=tensor([0.0780, 0.0300, 0.0291, 0.0899], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 01:50:19,162 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:50:26,862 INFO [optim.py:369] (1/2) 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 01:50:27,696 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:50:51,237 INFO [train.py:968] (1/2) Epoch 29, batch 28000, giga_loss[loss=0.3509, simple_loss=0.3896, pruned_loss=0.1561, over 26716.00 frames. ], tot_loss[loss=0.2881, simple_loss=0.3554, pruned_loss=0.1104, over 5656742.68 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3569, pruned_loss=0.1115, over 5694948.47 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3564, pruned_loss=0.1111, over 5647897.88 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:51:33,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1303869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:51:34,145 INFO [train.py:968] (1/2) Epoch 29, batch 28050, libri_loss[loss=0.2307, simple_loss=0.3049, pruned_loss=0.0782, over 27213.00 frames. ], tot_loss[loss=0.2901, simple_loss=0.3576, pruned_loss=0.1113, over 5674524.20 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3561, pruned_loss=0.111, over 5697692.82 frames. ], giga_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5664611.89 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:51:59,354 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1303894.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 01:52:01,367 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1303897.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:52:01,740 INFO [optim.py:369] (1/2) 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 01:52:23,022 INFO [train.py:968] (1/2) Epoch 29, batch 28100, giga_loss[loss=0.3023, simple_loss=0.3677, pruned_loss=0.1184, over 28542.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.361, pruned_loss=0.1133, over 5665834.15 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3561, pruned_loss=0.1109, over 5702121.13 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3623, pruned_loss=0.1142, over 5653580.35 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:52:29,048 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1303926.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:52:52,863 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:53:07,666 INFO [train.py:968] (1/2) Epoch 29, batch 28150, giga_loss[loss=0.2695, simple_loss=0.3477, pruned_loss=0.0956, over 29031.00 frames. ], tot_loss[loss=0.2942, simple_loss=0.3617, pruned_loss=0.1134, over 5658947.01 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3563, pruned_loss=0.1106, over 5699291.53 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3627, pruned_loss=0.1144, over 5650826.75 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:53:12,091 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1303974.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:53:28,275 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 01:53:32,911 INFO [optim.py:369] (1/2) 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 01:53:45,921 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:53:49,098 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 28200, giga_loss[loss=0.3037, simple_loss=0.3695, pruned_loss=0.1189, over 28950.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.363, pruned_loss=0.1147, over 5656703.52 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3564, pruned_loss=0.1107, over 5702839.57 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3639, pruned_loss=0.1155, over 5646113.54 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:54:05,599 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:54:13,598 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:54:15,431 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:54:34,166 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304066.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 01:54:37,295 INFO [train.py:968] (1/2) Epoch 29, batch 28250, giga_loss[loss=0.3822, simple_loss=0.4369, pruned_loss=0.1638, over 28659.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3655, pruned_loss=0.1164, over 5674234.07 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3574, pruned_loss=0.1112, over 5710440.74 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3656, pruned_loss=0.1168, over 5656925.17 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:54:59,335 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304093.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:54:59,421 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7761, 1.8725, 1.6462, 1.9182], device='cuda:1'), covar=tensor([0.0716, 0.0283, 0.0308, 0.0813], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 01:55:03,220 INFO [optim.py:369] (1/2) 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 01:55:24,277 INFO [train.py:968] (1/2) Epoch 29, batch 28300, giga_loss[loss=0.3007, simple_loss=0.3733, pruned_loss=0.1141, over 28665.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.3673, pruned_loss=0.1175, over 5656170.69 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3577, pruned_loss=0.1115, over 5693714.50 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3673, pruned_loss=0.1177, over 5657346.92 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:55:32,308 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:12,414 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 28350, giga_loss[loss=0.3921, simple_loss=0.4345, pruned_loss=0.1748, over 27507.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3697, pruned_loss=0.1191, over 5654106.60 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3576, pruned_loss=0.1114, over 5690651.50 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3699, pruned_loss=0.1194, over 5657135.09 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:56:16,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8517, 1.8738, 2.0226, 1.5794], device='cuda:1'), covar=tensor([0.1881, 0.2653, 0.1546, 0.1884], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0725, 0.0990, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 01:56:17,241 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:21,410 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3189, 1.4062, 1.3031, 1.3435], device='cuda:1'), covar=tensor([0.2109, 0.2300, 0.2369, 0.2010], device='cuda:1'), in_proj_covar=tensor([0.2093, 0.2066, 0.1958, 0.2106], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 01:56:30,364 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:32,284 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:56:41,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3811, 3.5936, 1.5621, 1.5872], device='cuda:1'), covar=tensor([0.1030, 0.0308, 0.0854, 0.1333], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0414, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 01:56:44,555 INFO [optim.py:369] (1/2) 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 01:56:54,570 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304209.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:56:57,873 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304212.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:57:00,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:57:04,036 INFO [train.py:968] (1/2) Epoch 29, batch 28400, giga_loss[loss=0.2806, simple_loss=0.3548, pruned_loss=0.1031, over 28915.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3715, pruned_loss=0.1211, over 5646468.14 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3578, pruned_loss=0.1115, over 5691723.24 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3715, pruned_loss=0.1213, over 5647555.10 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 01:57:26,957 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:57:26,972 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304241.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:57:56,274 INFO [train.py:968] (1/2) Epoch 29, batch 28450, giga_loss[loss=0.3042, simple_loss=0.3718, pruned_loss=0.1183, over 28894.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3725, pruned_loss=0.1216, over 5648648.37 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3578, pruned_loss=0.1115, over 5695620.06 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.373, pruned_loss=0.1221, over 5644343.31 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:58:12,348 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2857, 3.1274, 2.9657, 1.3924], device='cuda:1'), covar=tensor([0.1052, 0.1098, 0.1108, 0.2382], device='cuda:1'), in_proj_covar=tensor([0.1335, 0.1235, 0.1039, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 01:58:24,110 INFO [optim.py:369] (1/2) 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 01:58:27,083 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 01:58:39,013 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:58:41,789 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:58:42,107 INFO [train.py:968] (1/2) Epoch 29, batch 28500, giga_loss[loss=0.2926, simple_loss=0.3714, pruned_loss=0.1069, over 28715.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3718, pruned_loss=0.1196, over 5642567.83 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3577, pruned_loss=0.1116, over 5677637.23 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3727, pruned_loss=0.1202, over 5653998.80 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:58:49,394 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:59:11,769 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 01:59:11,783 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304349.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 01:59:13,867 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4996, 1.8187, 1.4861, 1.5301], device='cuda:1'), covar=tensor([0.2379, 0.2324, 0.2571, 0.2195], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1165, 0.1433, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 01:59:31,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-15 01:59:31,953 INFO [train.py:968] (1/2) Epoch 29, batch 28550, giga_loss[loss=0.3262, simple_loss=0.3802, pruned_loss=0.1361, over 27917.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3713, pruned_loss=0.1198, over 5652122.03 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3576, pruned_loss=0.1116, over 5682221.31 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3724, pruned_loss=0.1205, over 5656499.34 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 01:59:57,214 INFO [optim.py:369] (1/2) 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 01:59:59,697 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:00:03,085 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:00:03,838 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 02:00:18,507 INFO [train.py:968] (1/2) Epoch 29, batch 28600, giga_loss[loss=0.2639, simple_loss=0.3396, pruned_loss=0.09413, over 28532.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.37, pruned_loss=0.1197, over 5650700.62 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3574, pruned_loss=0.1114, over 5675481.41 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3715, pruned_loss=0.1206, over 5658798.99 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:00:36,184 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.35 vs. limit=2.0
+2023-03-15 02:01:10,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2494, 1.0912, 4.0256, 3.4690], device='cuda:1'), covar=tensor([0.2073, 0.3119, 0.0872, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0685, 0.1026, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 02:01:12,874 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:14,772 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:15,598 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 28650, giga_loss[loss=0.2443, simple_loss=0.3196, pruned_loss=0.08451, over 28511.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3692, pruned_loss=0.1197, over 5665567.57 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3575, pruned_loss=0.1114, over 5679171.69 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3704, pruned_loss=0.1206, over 5668422.70 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:01:19,900 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:41,386 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304492.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:01:44,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304495.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:01:48,056 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:01:51,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:09,849 INFO [train.py:968] (1/2) Epoch 29, batch 28700, giga_loss[loss=0.3715, simple_loss=0.4061, pruned_loss=0.1684, over 26535.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3671, pruned_loss=0.1188, over 5655233.49 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3572, pruned_loss=0.1112, over 5672294.79 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3686, pruned_loss=0.1198, over 5664121.17 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:02:12,336 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304524.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 02:02:19,267 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-15 02:02:29,807 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:32,544 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:36,399 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:02:48,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6724, 1.1989, 2.8960, 2.8649], device='cuda:1'), covar=tensor([0.2216, 0.2706, 0.1118, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0683, 0.1023, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 02:02:55,758 INFO [train.py:968] (1/2) Epoch 29, batch 28750, giga_loss[loss=0.3523, simple_loss=0.3837, pruned_loss=0.1605, over 23552.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3673, pruned_loss=0.1192, over 5659526.38 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.357, pruned_loss=0.111, over 5675920.92 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3687, pruned_loss=0.1202, over 5663329.21 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:03:05,385 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:23,916 INFO [optim.py:369] (1/2) 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 02:03:34,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:37,175 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:38,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:42,416 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:03:43,437 INFO [train.py:968] (1/2) Epoch 29, batch 28800, giga_loss[loss=0.3378, simple_loss=0.3917, pruned_loss=0.1419, over 28519.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3683, pruned_loss=0.1207, over 5655856.17 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3573, pruned_loss=0.1113, over 5681244.53 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3695, pruned_loss=0.1215, over 5653688.38 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:04:03,216 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:07,220 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:10,753 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:29,557 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4637, 1.9175, 1.3783, 0.9021], device='cuda:1'), covar=tensor([0.6446, 0.3369, 0.4032, 0.6817], device='cuda:1'), in_proj_covar=tensor([0.1860, 0.1753, 0.1670, 0.1517], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 02:04:33,087 INFO [train.py:968] (1/2) Epoch 29, batch 28850, giga_loss[loss=0.345, simple_loss=0.4052, pruned_loss=0.1424, over 28576.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3686, pruned_loss=0.121, over 5654071.16 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3575, pruned_loss=0.1116, over 5683171.32 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3694, pruned_loss=0.1215, over 5650629.38 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:04:40,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:49,781 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:04:52,081 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:05:04,090 INFO [optim.py:369] (1/2) 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 02:05:20,408 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:05:20,778 INFO [train.py:968] (1/2) Epoch 29, batch 28900, giga_loss[loss=0.3332, simple_loss=0.3918, pruned_loss=0.1372, over 28927.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3709, pruned_loss=0.1229, over 5655632.98 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3574, pruned_loss=0.1115, over 5685007.38 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3717, pruned_loss=0.1235, over 5651037.33 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:05:59,430 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:01,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:09,275 INFO [train.py:968] (1/2) Epoch 29, batch 28950, giga_loss[loss=0.302, simple_loss=0.359, pruned_loss=0.1225, over 28416.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3705, pruned_loss=0.123, over 5649349.96 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3571, pruned_loss=0.1112, over 5689568.54 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.372, pruned_loss=0.1241, over 5640331.08 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:06:14,623 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:28,680 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:06:36,007 INFO [optim.py:369] (1/2) 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 02:06:44,254 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-15 02:06:55,260 INFO [train.py:968] (1/2) Epoch 29, batch 29000, giga_loss[loss=0.3228, simple_loss=0.3843, pruned_loss=0.1307, over 29034.00 frames. ], tot_loss[loss=0.3115, simple_loss=0.3725, pruned_loss=0.1253, over 5640228.56 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3575, pruned_loss=0.1115, over 5680613.52 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3735, pruned_loss=0.126, over 5639782.90 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:07:04,518 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-15 02:07:14,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:07:44,060 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:07:44,479 INFO [train.py:968] (1/2) Epoch 29, batch 29050, giga_loss[loss=0.3204, simple_loss=0.3866, pruned_loss=0.1271, over 28649.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3718, pruned_loss=0.1247, over 5640551.73 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3575, pruned_loss=0.1115, over 5681728.45 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3726, pruned_loss=0.1254, over 5639015.17 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:08:10,508 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.49 vs. limit=2.0
+2023-03-15 02:08:13,351 INFO [optim.py:369] (1/2) 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 02:08:22,982 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1304910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:08:31,197 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:08:32,139 INFO [train.py:968] (1/2) Epoch 29, batch 29100, giga_loss[loss=0.341, simple_loss=0.3967, pruned_loss=0.1427, over 28596.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3724, pruned_loss=0.1245, over 5639074.03 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3578, pruned_loss=0.1118, over 5677931.18 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3731, pruned_loss=0.1251, over 5639603.35 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:08:35,422 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:08:49,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5762, 3.8773, 1.7346, 1.8185], device='cuda:1'), covar=tensor([0.0947, 0.0357, 0.0891, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0575, 0.0413, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 02:09:01,239 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1304951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:12,787 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-15 02:09:19,445 INFO [train.py:968] (1/2) Epoch 29, batch 29150, giga_loss[loss=0.2668, simple_loss=0.349, pruned_loss=0.09229, over 29073.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3723, pruned_loss=0.1236, over 5648986.38 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3575, pruned_loss=0.1115, over 5682352.76 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3736, pruned_loss=0.1247, over 5644554.71 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:09:33,010 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1304984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:35,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1304987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:41,991 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1304994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:09:49,528 INFO [optim.py:369] (1/2) 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 02:09:56,441 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7165, 2.0084, 1.3472, 1.6260], device='cuda:1'), covar=tensor([0.1177, 0.0775, 0.1148, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0527, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 02:10:03,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:10:07,846 INFO [train.py:968] (1/2) Epoch 29, batch 29200, giga_loss[loss=0.2613, simple_loss=0.3403, pruned_loss=0.09116, over 28481.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3733, pruned_loss=0.124, over 5659642.40 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3581, pruned_loss=0.1119, over 5684245.12 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3739, pruned_loss=0.1246, over 5654104.01 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:10:53,611 INFO [train.py:968] (1/2) Epoch 29, batch 29250, libri_loss[loss=0.2373, simple_loss=0.3084, pruned_loss=0.08307, over 29625.00 frames. ], tot_loss[loss=0.3109, simple_loss=0.3734, pruned_loss=0.1242, over 5662972.47 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.358, pruned_loss=0.1118, over 5678011.69 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3742, pruned_loss=0.125, over 5663945.59 frames. ], batch size: 69, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:11:21,930 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 29300, giga_loss[loss=0.377, simple_loss=0.426, pruned_loss=0.164, over 28597.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3744, pruned_loss=0.1251, over 5649137.83 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3583, pruned_loss=0.1121, over 5661912.50 frames. ], giga_tot_loss[loss=0.3131, simple_loss=0.375, pruned_loss=0.1256, over 5664649.35 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:11:55,515 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:11:59,064 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:12:05,260 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4477, 1.8676, 1.1898, 1.4104], device='cuda:1'), covar=tensor([0.1276, 0.0734, 0.1318, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0455, 0.0527, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 02:12:28,923 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:12:30,002 INFO [train.py:968] (1/2) Epoch 29, batch 29350, giga_loss[loss=0.289, simple_loss=0.3718, pruned_loss=0.1031, over 28891.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3756, pruned_loss=0.1247, over 5653997.59 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3586, pruned_loss=0.1123, over 5665489.93 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3762, pruned_loss=0.1253, over 5662944.68 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:13:04,417 INFO [optim.py:369] (1/2) 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 02:13:18,921 INFO [train.py:968] (1/2) Epoch 29, batch 29400, giga_loss[loss=0.3279, simple_loss=0.387, pruned_loss=0.1344, over 28575.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3739, pruned_loss=0.1226, over 5655604.87 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3586, pruned_loss=0.1122, over 5670692.64 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3748, pruned_loss=0.1234, over 5657739.25 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:13:29,088 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.82 vs. limit=5.0
+2023-03-15 02:13:39,645 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:14:00,104 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1305266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:14:04,597 INFO [train.py:968] (1/2) Epoch 29, batch 29450, giga_loss[loss=0.4302, simple_loss=0.4422, pruned_loss=0.2091, over 26500.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3715, pruned_loss=0.1211, over 5656886.53 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3589, pruned_loss=0.1123, over 5674020.62 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3722, pruned_loss=0.1219, over 5655377.18 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:14:15,435 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 02:14:48,465 INFO [train.py:968] (1/2) Epoch 29, batch 29500, libri_loss[loss=0.3415, simple_loss=0.396, pruned_loss=0.1435, over 25540.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3729, pruned_loss=0.1224, over 5654656.31 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3593, pruned_loss=0.1126, over 5674608.08 frames. ], giga_tot_loss[loss=0.3097, simple_loss=0.3735, pruned_loss=0.123, over 5653184.98 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:15:05,659 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1305341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:15:32,369 INFO [train.py:968] (1/2) Epoch 29, batch 29550, giga_loss[loss=0.3707, simple_loss=0.4229, pruned_loss=0.1593, over 28774.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.3736, pruned_loss=0.1225, over 5667632.30 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3592, pruned_loss=0.1124, over 5682391.91 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3745, pruned_loss=0.1234, over 5658999.19 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:15:52,112 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:15:55,476 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:06,270 INFO [optim.py:369] (1/2) 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 02:16:19,779 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:20,088 INFO [train.py:968] (1/2) Epoch 29, batch 29600, giga_loss[loss=0.4006, simple_loss=0.4271, pruned_loss=0.1871, over 27901.00 frames. ], tot_loss[loss=0.3127, simple_loss=0.3759, pruned_loss=0.1248, over 5659820.12 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3596, pruned_loss=0.1126, over 5680100.28 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3768, pruned_loss=0.1257, over 5653725.12 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:16:28,606 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:30,762 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4048, 3.3957, 1.5181, 1.6642], device='cuda:1'), covar=tensor([0.1028, 0.0413, 0.0894, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0579, 0.0415, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 02:16:31,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:16:58,715 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:17:08,992 INFO [train.py:968] (1/2) Epoch 29, batch 29650, giga_loss[loss=0.3387, simple_loss=0.3686, pruned_loss=0.1544, over 23560.00 frames. ], tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1257, over 5661338.23 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3597, pruned_loss=0.1128, over 5683514.84 frames. ], giga_tot_loss[loss=0.3147, simple_loss=0.3765, pruned_loss=0.1264, over 5653281.84 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:17:09,481 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 02:17:38,148 INFO [optim.py:369] (1/2) 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 02:17:54,168 INFO [train.py:968] (1/2) Epoch 29, batch 29700, giga_loss[loss=0.2925, simple_loss=0.3622, pruned_loss=0.1114, over 28753.00 frames. ], tot_loss[loss=0.3137, simple_loss=0.3754, pruned_loss=0.126, over 5645147.26 frames. ], libri_tot_loss[loss=0.2931, simple_loss=0.36, pruned_loss=0.1131, over 5673291.07 frames. ], giga_tot_loss[loss=0.3151, simple_loss=0.3765, pruned_loss=0.1269, over 5646618.15 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:18:39,188 INFO [train.py:968] (1/2) Epoch 29, batch 29750, giga_loss[loss=0.2851, simple_loss=0.359, pruned_loss=0.1056, over 28837.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3749, pruned_loss=0.1252, over 5652623.91 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3596, pruned_loss=0.1129, over 5669117.32 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3764, pruned_loss=0.1264, over 5656150.81 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:19:07,295 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1305600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:19:09,175 INFO [optim.py:369] (1/2) 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 02:19:12,140 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5856, 1.5863, 1.7985, 1.4186], device='cuda:1'), covar=tensor([0.1586, 0.2414, 0.1353, 0.1701], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0724, 0.0990, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 02:19:27,675 INFO [train.py:968] (1/2) Epoch 29, batch 29800, giga_loss[loss=0.3072, simple_loss=0.3718, pruned_loss=0.1213, over 28964.00 frames. ], tot_loss[loss=0.3138, simple_loss=0.3757, pruned_loss=0.126, over 5641825.40 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3594, pruned_loss=0.1127, over 5669345.88 frames. ], giga_tot_loss[loss=0.316, simple_loss=0.3774, pruned_loss=0.1273, over 5643921.24 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:19:28,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4350, 1.7339, 1.5319, 1.5882], device='cuda:1'), covar=tensor([0.0784, 0.0316, 0.0325, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 02:19:44,542 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:20:10,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-15 02:20:10,836 INFO [train.py:968] (1/2) Epoch 29, batch 29850, giga_loss[loss=0.3702, simple_loss=0.411, pruned_loss=0.1647, over 27643.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.374, pruned_loss=0.1241, over 5666203.76 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3593, pruned_loss=0.1128, over 5675064.76 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3759, pruned_loss=0.1255, over 5662254.23 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:20:41,957 INFO [optim.py:369] (1/2) 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 02:20:53,504 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 29900, giga_loss[loss=0.2915, simple_loss=0.3575, pruned_loss=0.1127, over 28856.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3728, pruned_loss=0.1225, over 5667932.23 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3591, pruned_loss=0.1126, over 5680566.95 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3748, pruned_loss=0.1239, over 5659824.05 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:21:28,460 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6679, 1.6700, 1.8596, 1.4300], device='cuda:1'), covar=tensor([0.1906, 0.2727, 0.1584, 0.1818], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0724, 0.0991, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 02:21:31,693 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 02:21:48,751 INFO [train.py:968] (1/2) Epoch 29, batch 29950, giga_loss[loss=0.2847, simple_loss=0.3593, pruned_loss=0.1051, over 28710.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.3712, pruned_loss=0.1208, over 5669347.46 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3591, pruned_loss=0.1126, over 5680566.95 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3727, pruned_loss=0.1219, over 5663036.76 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:22:00,628 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1305784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:22:03,025 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:22:15,050 INFO [optim.py:369] (1/2) 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 02:22:28,501 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 02:22:28,890 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 30000, giga_loss[loss=0.3761, simple_loss=0.4243, pruned_loss=0.164, over 28595.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3691, pruned_loss=0.1198, over 5668906.47 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5683153.16 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3709, pruned_loss=0.1211, over 5661083.83 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:22:32,412 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 02:22:40,684 INFO [train.py:1012] (1/2) Epoch 29, validation: loss=0.2031, simple_loss=0.3116, pruned_loss=0.04732, over 944034.00 frames. 
+2023-03-15 02:22:40,685 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 02:23:19,740 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1305862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:23:27,769 INFO [train.py:968] (1/2) Epoch 29, batch 30050, giga_loss[loss=0.2606, simple_loss=0.3345, pruned_loss=0.09332, over 28843.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3685, pruned_loss=0.1199, over 5664324.26 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.359, pruned_loss=0.1125, over 5683153.16 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3698, pruned_loss=0.1209, over 5658235.80 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:23:48,590 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1305891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:23:58,985 INFO [optim.py:369] (1/2) 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 02:24:16,805 INFO [train.py:968] (1/2) Epoch 29, batch 30100, giga_loss[loss=0.2607, simple_loss=0.3284, pruned_loss=0.09651, over 29015.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3657, pruned_loss=0.1186, over 5667572.58 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.359, pruned_loss=0.1126, over 5688423.29 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.367, pruned_loss=0.1195, over 5657832.61 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:24:32,069 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6666, 2.0193, 1.6094, 1.6213], device='cuda:1'), covar=tensor([0.2806, 0.2925, 0.3300, 0.2638], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1163, 0.1431, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 02:25:03,925 INFO [train.py:968] (1/2) Epoch 29, batch 30150, giga_loss[loss=0.3069, simple_loss=0.3572, pruned_loss=0.1283, over 28836.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3633, pruned_loss=0.1175, over 5673642.38 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1125, over 5683000.13 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.3645, pruned_loss=0.1184, over 5670759.48 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:25:06,683 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1305975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:25:30,961 INFO [optim.py:369] (1/2) 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 02:25:48,651 INFO [train.py:968] (1/2) Epoch 29, batch 30200, giga_loss[loss=0.2955, simple_loss=0.3625, pruned_loss=0.1142, over 28776.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3629, pruned_loss=0.1177, over 5687344.44 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3596, pruned_loss=0.1128, over 5687073.67 frames. ], giga_tot_loss[loss=0.3, simple_loss=0.3633, pruned_loss=0.1183, over 5681372.66 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:26:05,440 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-15 02:26:37,990 INFO [train.py:968] (1/2) Epoch 29, batch 30250, giga_loss[loss=0.2905, simple_loss=0.3584, pruned_loss=0.1113, over 28929.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.362, pruned_loss=0.1166, over 5686432.05 frames. ], libri_tot_loss[loss=0.2932, simple_loss=0.3601, pruned_loss=0.1131, over 5688623.16 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3619, pruned_loss=0.1169, over 5680351.62 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:27:09,162 INFO [optim.py:369] (1/2) 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 02:27:25,243 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 30300, giga_loss[loss=0.2365, simple_loss=0.3152, pruned_loss=0.07897, over 29117.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3609, pruned_loss=0.1141, over 5685985.22 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3597, pruned_loss=0.1128, over 5691366.03 frames. ], giga_tot_loss[loss=0.2953, simple_loss=0.3612, pruned_loss=0.1147, over 5678562.04 frames. ], batch size: 113, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:27:27,646 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1306121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:27:46,932 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-15 02:27:53,068 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5364, 1.5809, 1.7439, 1.3485], device='cuda:1'), covar=tensor([0.1874, 0.2660, 0.1631, 0.1948], device='cuda:1'), in_proj_covar=tensor([0.0941, 0.0723, 0.0990, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 02:27:57,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1306150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:28:07,473 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2965, 1.5955, 1.1879, 0.5692], device='cuda:1'), covar=tensor([0.3759, 0.2565, 0.3667, 0.6227], device='cuda:1'), in_proj_covar=tensor([0.1860, 0.1751, 0.1673, 0.1518], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 02:28:20,271 INFO [train.py:968] (1/2) Epoch 29, batch 30350, giga_loss[loss=0.2734, simple_loss=0.3523, pruned_loss=0.09728, over 28710.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3581, pruned_loss=0.1101, over 5672482.92 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3593, pruned_loss=0.1126, over 5691123.40 frames. ], giga_tot_loss[loss=0.29, simple_loss=0.3586, pruned_loss=0.1107, over 5667030.34 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:28:42,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1435, 1.3573, 1.1325, 0.9745], device='cuda:1'), covar=tensor([0.1115, 0.0482, 0.1066, 0.1120], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0453, 0.0526, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 02:28:49,907 INFO [optim.py:369] (1/2) 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 02:28:55,933 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2436, 2.4228, 2.3256, 1.9862], device='cuda:1'), covar=tensor([0.1881, 0.2087, 0.1849, 0.2210], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0763, 0.0739, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 02:29:04,275 INFO [train.py:968] (1/2) Epoch 29, batch 30400, giga_loss[loss=0.2359, simple_loss=0.3241, pruned_loss=0.07389, over 28685.00 frames. ], tot_loss[loss=0.2847, simple_loss=0.3549, pruned_loss=0.1073, over 5666625.23 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3585, pruned_loss=0.1122, over 5690625.33 frames. ], giga_tot_loss[loss=0.286, simple_loss=0.3561, pruned_loss=0.108, over 5661224.05 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:29:50,741 INFO [train.py:968] (1/2) Epoch 29, batch 30450, giga_loss[loss=0.2735, simple_loss=0.3366, pruned_loss=0.1053, over 26763.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3524, pruned_loss=0.1052, over 5665836.47 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3582, pruned_loss=0.1126, over 5696184.34 frames. ], giga_tot_loss[loss=0.282, simple_loss=0.3534, pruned_loss=0.1053, over 5655848.36 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:30:21,315 INFO [optim.py:369] (1/2) 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 02:30:34,949 INFO [train.py:968] (1/2) Epoch 29, batch 30500, libri_loss[loss=0.2964, simple_loss=0.3608, pruned_loss=0.116, over 29512.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3495, pruned_loss=0.1021, over 5670468.57 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.358, pruned_loss=0.1125, over 5701991.74 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.3502, pruned_loss=0.1018, over 5655589.54 frames. ], batch size: 81, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:31:22,203 INFO [train.py:968] (1/2) Epoch 29, batch 30550, libri_loss[loss=0.2763, simple_loss=0.345, pruned_loss=0.1038, over 29189.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3475, pruned_loss=0.09911, over 5659914.71 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3575, pruned_loss=0.1124, over 5708536.58 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3482, pruned_loss=0.09864, over 5640773.09 frames. ], batch size: 97, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:31:57,257 INFO [optim.py:369] (1/2) 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 02:32:13,432 INFO [train.py:968] (1/2) Epoch 29, batch 30600, giga_loss[loss=0.3019, simple_loss=0.3629, pruned_loss=0.1205, over 26838.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3497, pruned_loss=0.1007, over 5649316.03 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3571, pruned_loss=0.1122, over 5701284.65 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3505, pruned_loss=0.1002, over 5638775.14 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:33:05,550 INFO [train.py:968] (1/2) Epoch 29, batch 30650, giga_loss[loss=0.2397, simple_loss=0.3188, pruned_loss=0.08027, over 28739.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09979, over 5641850.41 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.357, pruned_loss=0.1122, over 5701939.22 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3491, pruned_loss=0.09939, over 5632845.15 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:33:37,526 INFO [optim.py:369] (1/2) 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 02:33:38,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4077, 1.6730, 1.2155, 1.2284], device='cuda:1'), covar=tensor([0.1093, 0.0483, 0.1042, 0.1134], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0452, 0.0526, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 02:33:52,391 INFO [train.py:968] (1/2) Epoch 29, batch 30700, giga_loss[loss=0.2512, simple_loss=0.3285, pruned_loss=0.08701, over 28002.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3454, pruned_loss=0.09752, over 5640003.35 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3569, pruned_loss=0.1122, over 5697062.95 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3459, pruned_loss=0.09694, over 5635702.48 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:34:38,589 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8476, 3.6897, 3.5044, 1.8300], device='cuda:1'), covar=tensor([0.0796, 0.0900, 0.1010, 0.2121], device='cuda:1'), in_proj_covar=tensor([0.1321, 0.1224, 0.1025, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 02:34:39,569 INFO [train.py:968] (1/2) Epoch 29, batch 30750, giga_loss[loss=0.242, simple_loss=0.3267, pruned_loss=0.07869, over 28703.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3439, pruned_loss=0.09683, over 5634747.36 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3566, pruned_loss=0.1124, over 5688998.88 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3441, pruned_loss=0.09578, over 5636168.70 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:34:53,747 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1169, 3.9594, 3.7494, 1.9793], device='cuda:1'), covar=tensor([0.0670, 0.0792, 0.0907, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.1321, 0.1225, 0.1025, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 02:35:08,741 INFO [optim.py:369] (1/2) 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 02:35:21,887 INFO [train.py:968] (1/2) Epoch 29, batch 30800, libri_loss[loss=0.2964, simple_loss=0.3587, pruned_loss=0.117, over 29677.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3454, pruned_loss=0.09794, over 5643024.16 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3568, pruned_loss=0.1129, over 5694540.33 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3449, pruned_loss=0.09587, over 5636460.16 frames. ], batch size: 88, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:36:10,088 INFO [train.py:968] (1/2) Epoch 29, batch 30850, giga_loss[loss=0.2558, simple_loss=0.341, pruned_loss=0.08529, over 28642.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3434, pruned_loss=0.09588, over 5633960.08 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3569, pruned_loss=0.1131, over 5677165.96 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3428, pruned_loss=0.09391, over 5643024.71 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:36:15,784 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:36:46,314 INFO [optim.py:369] (1/2) 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 02:36:59,577 INFO [train.py:968] (1/2) Epoch 29, batch 30900, libri_loss[loss=0.3189, simple_loss=0.3791, pruned_loss=0.1293, over 29652.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3402, pruned_loss=0.09316, over 5643344.80 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3561, pruned_loss=0.1127, over 5680704.50 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.34, pruned_loss=0.09144, over 5646273.02 frames. ], batch size: 91, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:37:18,769 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3048, 1.8680, 1.8131, 1.5964], device='cuda:1'), covar=tensor([0.2446, 0.1878, 0.2013, 0.2013], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0754, 0.0732, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 02:37:25,465 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2978, 1.7225, 1.0582, 1.2360], device='cuda:1'), covar=tensor([0.1406, 0.0629, 0.1537, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0451, 0.0524, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 02:37:51,008 INFO [train.py:968] (1/2) Epoch 29, batch 30950, giga_loss[loss=0.2567, simple_loss=0.3355, pruned_loss=0.08896, over 28707.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3362, pruned_loss=0.09094, over 5636903.12 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3557, pruned_loss=0.1125, over 5684185.52 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3361, pruned_loss=0.08942, over 5635380.64 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:37:51,352 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:38:18,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306798.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:38:18,386 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.03 vs. limit=5.0
+2023-03-15 02:38:25,279 INFO [optim.py:369] (1/2) 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 02:38:38,526 INFO [train.py:968] (1/2) Epoch 29, batch 31000, giga_loss[loss=0.2297, simple_loss=0.2962, pruned_loss=0.08159, over 23814.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3356, pruned_loss=0.0912, over 5638840.60 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3555, pruned_loss=0.1125, over 5684796.96 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3353, pruned_loss=0.08962, over 5636451.95 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:38:43,228 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:39:21,392 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1306867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:39:26,620 INFO [train.py:968] (1/2) Epoch 29, batch 31050, giga_loss[loss=0.2647, simple_loss=0.3432, pruned_loss=0.09306, over 28523.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.3329, pruned_loss=0.08987, over 5634694.06 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3549, pruned_loss=0.1121, over 5691096.38 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3325, pruned_loss=0.08827, over 5625578.23 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:40:04,533 INFO [optim.py:369] (1/2) 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 02:40:19,454 INFO [train.py:968] (1/2) Epoch 29, batch 31100, giga_loss[loss=0.2905, simple_loss=0.3586, pruned_loss=0.1112, over 26954.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3345, pruned_loss=0.09057, over 5631040.87 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3547, pruned_loss=0.112, over 5694358.73 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3342, pruned_loss=0.08906, over 5620132.92 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:40:42,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4087, 1.2285, 3.7423, 3.3108], device='cuda:1'), covar=tensor([0.1523, 0.2941, 0.0426, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0677, 0.1012, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 02:40:42,485 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3972, 1.8315, 1.6801, 1.6408], device='cuda:1'), covar=tensor([0.1983, 0.1898, 0.1866, 0.1852], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0751, 0.0729, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 02:40:59,993 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.36 vs. limit=2.0
+2023-03-15 02:41:13,297 INFO [train.py:968] (1/2) Epoch 29, batch 31150, giga_loss[loss=0.256, simple_loss=0.3431, pruned_loss=0.08445, over 28956.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3383, pruned_loss=0.09177, over 5645638.17 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.355, pruned_loss=0.1125, over 5698861.53 frames. ], giga_tot_loss[loss=0.258, simple_loss=0.3371, pruned_loss=0.08943, over 5630992.21 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:41:44,723 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4863, 1.7839, 1.2401, 1.3288], device='cuda:1'), covar=tensor([0.1126, 0.0530, 0.1040, 0.1161], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0451, 0.0525, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 02:41:52,812 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 31200, giga_loss[loss=0.27, simple_loss=0.3498, pruned_loss=0.09514, over 28036.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.3392, pruned_loss=0.09226, over 5660368.31 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3547, pruned_loss=0.1125, over 5699955.32 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.338, pruned_loss=0.08979, over 5646319.06 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:42:50,522 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 31250, giga_loss[loss=0.2512, simple_loss=0.3286, pruned_loss=0.08687, over 28424.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3376, pruned_loss=0.09136, over 5670143.78 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3541, pruned_loss=0.1122, over 5702824.36 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3367, pruned_loss=0.08888, over 5654831.09 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:43:43,578 INFO [optim.py:369] (1/2) 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 02:44:01,823 INFO [train.py:968] (1/2) Epoch 29, batch 31300, giga_loss[loss=0.2695, simple_loss=0.3586, pruned_loss=0.09019, over 28958.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3356, pruned_loss=0.08994, over 5653176.78 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3537, pruned_loss=0.1121, over 5689132.08 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3345, pruned_loss=0.08722, over 5651837.42 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:44:16,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4677, 1.2466, 4.1100, 3.3604], device='cuda:1'), covar=tensor([0.1623, 0.3143, 0.0435, 0.1437], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0677, 0.1011, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 02:44:34,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:44:48,247 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 02:45:01,265 INFO [train.py:968] (1/2) Epoch 29, batch 31350, giga_loss[loss=0.2303, simple_loss=0.318, pruned_loss=0.07131, over 28443.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3344, pruned_loss=0.08821, over 5658677.26 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3533, pruned_loss=0.112, over 5692030.28 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3334, pruned_loss=0.08553, over 5654281.51 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:45:04,822 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:31,950 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:34,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:38,445 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:45:45,319 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 31400, giga_loss[loss=0.2128, simple_loss=0.2956, pruned_loss=0.06495, over 29107.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3317, pruned_loss=0.0874, over 5653782.21 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3532, pruned_loss=0.112, over 5683290.70 frames. ], giga_tot_loss[loss=0.2501, simple_loss=0.3306, pruned_loss=0.08482, over 5657927.84 frames. ], batch size: 146, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:46:07,734 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:46:28,214 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:46:29,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 02:46:59,211 INFO [train.py:968] (1/2) Epoch 29, batch 31450, libri_loss[loss=0.2705, simple_loss=0.3272, pruned_loss=0.1069, over 29488.00 frames. ], tot_loss[loss=0.253, simple_loss=0.3309, pruned_loss=0.08758, over 5659207.51 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3528, pruned_loss=0.1119, over 5688789.55 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.33, pruned_loss=0.08501, over 5656859.30 frames. ], batch size: 70, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:47:20,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:22,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:42,106 INFO [optim.py:369] (1/2) 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 02:47:51,638 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:54,573 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:47:55,620 INFO [train.py:968] (1/2) Epoch 29, batch 31500, giga_loss[loss=0.2523, simple_loss=0.3417, pruned_loss=0.08147, over 28952.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3298, pruned_loss=0.08662, over 5666603.66 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3527, pruned_loss=0.1118, over 5690622.43 frames. ], giga_tot_loss[loss=0.2489, simple_loss=0.3289, pruned_loss=0.08444, over 5662833.71 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:47:58,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:23,174 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307344.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:48:23,214 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:25,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:26,021 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:48:52,766 INFO [train.py:968] (1/2) Epoch 29, batch 31550, giga_loss[loss=0.2694, simple_loss=0.355, pruned_loss=0.09187, over 28902.00 frames. ], tot_loss[loss=0.253, simple_loss=0.332, pruned_loss=0.08703, over 5662081.04 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.353, pruned_loss=0.112, over 5689481.69 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3306, pruned_loss=0.08472, over 5659696.99 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:48:59,840 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:11,403 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:15,417 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:17,588 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 02:49:30,869 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2914, 3.9046, 1.4394, 1.5637], device='cuda:1'), covar=tensor([0.1073, 0.0370, 0.1017, 0.1395], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0575, 0.0415, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 02:49:37,968 INFO [optim.py:369] (1/2) 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 02:49:52,220 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:49:55,850 INFO [train.py:968] (1/2) Epoch 29, batch 31600, giga_loss[loss=0.2209, simple_loss=0.307, pruned_loss=0.06744, over 28832.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3333, pruned_loss=0.08748, over 5663425.12 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3528, pruned_loss=0.112, over 5692212.87 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3321, pruned_loss=0.08528, over 5658760.10 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:50:34,456 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2483, 2.6196, 1.2533, 1.4620], device='cuda:1'), covar=tensor([0.1019, 0.0352, 0.0975, 0.1382], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0574, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 02:51:01,183 INFO [train.py:968] (1/2) Epoch 29, batch 31650, giga_loss[loss=0.2111, simple_loss=0.2965, pruned_loss=0.06284, over 28088.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3304, pruned_loss=0.08578, over 5667834.57 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3529, pruned_loss=0.1121, over 5695309.13 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.3291, pruned_loss=0.08359, over 5661123.90 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:51:45,086 INFO [optim.py:369] (1/2) 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 02:52:01,543 INFO [train.py:968] (1/2) Epoch 29, batch 31700, giga_loss[loss=0.2668, simple_loss=0.353, pruned_loss=0.0903, over 28462.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3328, pruned_loss=0.08764, over 5665270.12 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3525, pruned_loss=0.112, over 5688017.68 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3314, pruned_loss=0.0852, over 5664974.06 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:52:19,763 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 02:52:41,961 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.64 vs. limit=5.0
+2023-03-15 02:52:59,864 INFO [train.py:968] (1/2) Epoch 29, batch 31750, giga_loss[loss=0.233, simple_loss=0.3352, pruned_loss=0.06534, over 28504.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3373, pruned_loss=0.08837, over 5657832.97 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3523, pruned_loss=0.1119, over 5692915.99 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.336, pruned_loss=0.08601, over 5652828.36 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:53:39,402 INFO [optim.py:369] (1/2) 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 02:53:56,297 INFO [train.py:968] (1/2) Epoch 29, batch 31800, giga_loss[loss=0.2736, simple_loss=0.3625, pruned_loss=0.09233, over 28649.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.3406, pruned_loss=0.08909, over 5655643.00 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3528, pruned_loss=0.1125, over 5691737.71 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3386, pruned_loss=0.0856, over 5650690.49 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:54:04,930 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:54:11,265 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6604, 2.1252, 1.1815, 1.6132], device='cuda:1'), covar=tensor([0.1249, 0.0679, 0.1251, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0452, 0.0526, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 02:54:33,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.74 vs. limit=5.0
+2023-03-15 02:54:33,884 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9261, 1.0808, 0.8819, 0.2889], device='cuda:1'), covar=tensor([0.4060, 0.3036, 0.4172, 0.6261], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1740, 0.1660, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 02:54:56,267 INFO [train.py:968] (1/2) Epoch 29, batch 31850, giga_loss[loss=0.2321, simple_loss=0.3341, pruned_loss=0.06506, over 28844.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3402, pruned_loss=0.08738, over 5657235.67 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3522, pruned_loss=0.1122, over 5693722.81 frames. ], giga_tot_loss[loss=0.2542, simple_loss=0.339, pruned_loss=0.08468, over 5651213.47 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:55:42,225 INFO [optim.py:369] (1/2) 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 02:55:53,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1307719.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 02:55:54,060 INFO [train.py:968] (1/2) Epoch 29, batch 31900, giga_loss[loss=0.2371, simple_loss=0.3195, pruned_loss=0.07733, over 27608.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3386, pruned_loss=0.08622, over 5660610.65 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.352, pruned_loss=0.1121, over 5697284.60 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.3374, pruned_loss=0.08355, over 5651565.58 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:56:05,820 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 02:56:55,540 INFO [train.py:968] (1/2) Epoch 29, batch 31950, giga_loss[loss=0.2678, simple_loss=0.3475, pruned_loss=0.09404, over 28356.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3393, pruned_loss=0.08779, over 5659028.33 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3516, pruned_loss=0.1119, over 5698793.10 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3386, pruned_loss=0.08554, over 5650124.16 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 02:57:25,905 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 29, batch 32000, giga_loss[loss=0.2497, simple_loss=0.3345, pruned_loss=0.08241, over 29005.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.3389, pruned_loss=0.0888, over 5665498.12 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3514, pruned_loss=0.1118, over 5700658.54 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3381, pruned_loss=0.08625, over 5655600.47 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 02:59:07,033 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1307862.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:59:11,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1307865.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 02:59:18,026 INFO [train.py:968] (1/2) Epoch 29, batch 32050, giga_loss[loss=0.2066, simple_loss=0.298, pruned_loss=0.05762, over 28869.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3368, pruned_loss=0.0875, over 5669322.79 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3514, pruned_loss=0.1118, over 5692554.27 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.336, pruned_loss=0.08532, over 5667588.02 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 02:59:45,825 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1307894.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 03:00:04,982 INFO [optim.py:369] (1/2) 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 03:00:17,896 INFO [train.py:968] (1/2) Epoch 29, batch 32100, giga_loss[loss=0.2693, simple_loss=0.3465, pruned_loss=0.09605, over 28791.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3338, pruned_loss=0.08665, over 5667889.18 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3509, pruned_loss=0.1118, over 5691236.04 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3331, pruned_loss=0.08391, over 5666471.67 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:00:28,748 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1710, 0.8559, 0.9527, 1.3697], device='cuda:1'), covar=tensor([0.0801, 0.0412, 0.0384, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 03:00:55,630 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1307949.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:01:12,326 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3017, 1.2041, 1.1587, 1.5543], device='cuda:1'), covar=tensor([0.0793, 0.0364, 0.0371, 0.0903], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 03:01:22,439 INFO [train.py:968] (1/2) Epoch 29, batch 32150, giga_loss[loss=0.2327, simple_loss=0.3135, pruned_loss=0.07592, over 27676.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3329, pruned_loss=0.08648, over 5656023.24 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3512, pruned_loss=0.1121, over 5685273.65 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3317, pruned_loss=0.08363, over 5660378.01 frames. ], batch size: 474, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:01:58,189 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:02:07,156 INFO [optim.py:369] (1/2) 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 03:02:22,184 INFO [train.py:968] (1/2) Epoch 29, batch 32200, giga_loss[loss=0.262, simple_loss=0.3441, pruned_loss=0.08996, over 28078.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3322, pruned_loss=0.08644, over 5657279.57 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3506, pruned_loss=0.1117, over 5686263.18 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3312, pruned_loss=0.08375, over 5659576.54 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:03:02,043 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9493, 2.1331, 1.7974, 2.2955], device='cuda:1'), covar=tensor([0.2803, 0.2873, 0.3281, 0.2417], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1163, 0.1434, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 03:03:14,740 INFO [train.py:968] (1/2) Epoch 29, batch 32250, giga_loss[loss=0.2886, simple_loss=0.3651, pruned_loss=0.1061, over 28525.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3364, pruned_loss=0.08876, over 5661732.00 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3502, pruned_loss=0.1116, over 5682067.35 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3353, pruned_loss=0.08574, over 5667084.07 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:03:46,342 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1308100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:03:49,980 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:03:54,801 INFO [optim.py:369] (1/2) 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 03:04:01,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-15 03:04:08,289 INFO [train.py:968] (1/2) Epoch 29, batch 32300, giga_loss[loss=0.2518, simple_loss=0.3337, pruned_loss=0.08495, over 28568.00 frames. ], tot_loss[loss=0.259, simple_loss=0.3373, pruned_loss=0.09034, over 5666343.97 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3502, pruned_loss=0.1117, over 5689716.55 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3359, pruned_loss=0.08687, over 5662609.87 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:04:12,670 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7835, 1.3180, 1.3709, 0.9901], device='cuda:1'), covar=tensor([0.2304, 0.1227, 0.2247, 0.1817], device='cuda:1'), in_proj_covar=tensor([0.0503, 0.0751, 0.0730, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 03:04:38,696 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:04:42,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:05:04,635 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 03:05:05,760 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:05:07,998 INFO [train.py:968] (1/2) Epoch 29, batch 32350, giga_loss[loss=0.2995, simple_loss=0.3685, pruned_loss=0.1153, over 28554.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3368, pruned_loss=0.0912, over 5668658.38 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3496, pruned_loss=0.1113, over 5692845.28 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.336, pruned_loss=0.08847, over 5662672.40 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:05:15,742 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:05:24,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5992, 1.9235, 1.9008, 1.5694], device='cuda:1'), covar=tensor([0.3580, 0.2530, 0.2112, 0.2931], device='cuda:1'), in_proj_covar=tensor([0.2056, 0.2025, 0.1907, 0.2062], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 03:05:54,799 INFO [optim.py:369] (1/2) 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 03:06:08,241 INFO [train.py:968] (1/2) Epoch 29, batch 32400, giga_loss[loss=0.288, simple_loss=0.3648, pruned_loss=0.1056, over 29054.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3374, pruned_loss=0.09184, over 5670182.06 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3495, pruned_loss=0.1113, over 5696237.47 frames. ], giga_tot_loss[loss=0.2577, simple_loss=0.3366, pruned_loss=0.08934, over 5662113.03 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:06:48,265 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:06:51,039 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:07:20,658 INFO [train.py:968] (1/2) Epoch 29, batch 32450, giga_loss[loss=0.2468, simple_loss=0.3183, pruned_loss=0.08767, over 24465.00 frames. ], tot_loss[loss=0.2602, simple_loss=0.3384, pruned_loss=0.09097, over 5665003.44 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3495, pruned_loss=0.1113, over 5696980.29 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3378, pruned_loss=0.08893, over 5657957.73 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:07:32,750 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:08:15,085 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:08:22,802 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:08:29,252 INFO [train.py:968] (1/2) Epoch 29, batch 32500, libri_loss[loss=0.3388, simple_loss=0.3919, pruned_loss=0.1429, over 29277.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3382, pruned_loss=0.09009, over 5676423.20 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3493, pruned_loss=0.1112, over 5704038.08 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3375, pruned_loss=0.08781, over 5663521.46 frames. ], batch size: 94, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:08:34,047 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308324.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 03:09:02,594 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 32550, giga_loss[loss=0.2209, simple_loss=0.3053, pruned_loss=0.06822, over 28117.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3346, pruned_loss=0.08784, over 5675707.55 frames. ], libri_tot_loss[loss=0.2852, simple_loss=0.3487, pruned_loss=0.1109, over 5704815.09 frames. ], giga_tot_loss[loss=0.2534, simple_loss=0.3344, pruned_loss=0.08617, over 5664694.33 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:10:23,941 INFO [optim.py:369] (1/2) 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 03:10:31,391 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6916, 4.7451, 1.9287, 2.0195], device='cuda:1'), covar=tensor([0.0975, 0.0345, 0.0871, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0573, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 03:10:35,511 INFO [train.py:968] (1/2) Epoch 29, batch 32600, giga_loss[loss=0.2582, simple_loss=0.3257, pruned_loss=0.09537, over 28444.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3302, pruned_loss=0.08665, over 5680729.16 frames. ], libri_tot_loss[loss=0.2848, simple_loss=0.3482, pruned_loss=0.1107, over 5708107.44 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.33, pruned_loss=0.08462, over 5667806.31 frames. ], batch size: 369, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:11:30,864 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308467.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:11:33,496 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308470.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:11:33,821 INFO [train.py:968] (1/2) Epoch 29, batch 32650, giga_loss[loss=0.2411, simple_loss=0.3195, pruned_loss=0.08136, over 28594.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3263, pruned_loss=0.08532, over 5672669.84 frames. ], libri_tot_loss[loss=0.2841, simple_loss=0.3476, pruned_loss=0.1103, over 5710616.37 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.326, pruned_loss=0.0831, over 5659051.34 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:11:38,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2686, 3.8133, 1.5183, 1.6707], device='cuda:1'), covar=tensor([0.1151, 0.0484, 0.0992, 0.1441], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0573, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 03:11:40,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:11:43,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0491, 2.4532, 1.6747, 2.1736], device='cuda:1'), covar=tensor([0.0968, 0.0573, 0.0972, 0.0990], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0451, 0.0527, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 03:12:01,474 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6582, 4.5024, 4.2886, 2.0693], device='cuda:1'), covar=tensor([0.0556, 0.0706, 0.0791, 0.1960], device='cuda:1'), in_proj_covar=tensor([0.1306, 0.1207, 0.1014, 0.0748], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 03:12:04,507 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-15 03:12:07,817 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308499.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 03:12:19,602 INFO [optim.py:369] (1/2) 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 03:12:32,272 INFO [train.py:968] (1/2) Epoch 29, batch 32700, giga_loss[loss=0.24, simple_loss=0.3197, pruned_loss=0.08018, over 28012.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3278, pruned_loss=0.08637, over 5670961.46 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3471, pruned_loss=0.11, over 5716302.73 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3274, pruned_loss=0.08421, over 5654201.22 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:13:02,020 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3043, 2.6777, 1.4066, 1.4554], device='cuda:1'), covar=tensor([0.0996, 0.0470, 0.0971, 0.1400], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0573, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 03:13:17,432 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-15 03:13:27,544 INFO [train.py:968] (1/2) Epoch 29, batch 32750, giga_loss[loss=0.2602, simple_loss=0.3421, pruned_loss=0.08912, over 29042.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3288, pruned_loss=0.08737, over 5671313.37 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3466, pruned_loss=0.1098, over 5722911.31 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3284, pruned_loss=0.08508, over 5650174.71 frames. ], batch size: 285, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:14:09,923 INFO [optim.py:369] (1/2) 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 03:14:12,401 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1308613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:16,641 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-15 03:14:19,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:23,508 INFO [train.py:968] (1/2) Epoch 29, batch 32800, giga_loss[loss=0.2357, simple_loss=0.3251, pruned_loss=0.0731, over 28925.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3272, pruned_loss=0.08603, over 5654270.89 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3469, pruned_loss=0.1102, over 5707058.19 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3259, pruned_loss=0.08313, over 5649177.04 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:14:24,108 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:30,914 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:14:50,269 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-15 03:14:55,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3396, 3.2000, 1.3785, 1.5294], device='cuda:1'), covar=tensor([0.1043, 0.0397, 0.1000, 0.1446], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0572, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 03:14:58,639 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:15:24,472 INFO [train.py:968] (1/2) Epoch 29, batch 32850, giga_loss[loss=0.2463, simple_loss=0.3193, pruned_loss=0.08662, over 28962.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3262, pruned_loss=0.08453, over 5665637.64 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3466, pruned_loss=0.1099, over 5710464.38 frames. ], giga_tot_loss[loss=0.2446, simple_loss=0.3252, pruned_loss=0.08203, over 5657774.69 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:15:45,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1333, 1.5708, 1.5309, 1.3427], device='cuda:1'), covar=tensor([0.2011, 0.1490, 0.2022, 0.1677], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0749, 0.0727, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 03:15:54,912 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.82 vs. limit=2.0
+2023-03-15 03:16:07,851 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3558, 1.7057, 1.6632, 1.4679], device='cuda:1'), covar=tensor([0.2106, 0.2124, 0.2143, 0.2039], device='cuda:1'), in_proj_covar=tensor([0.0501, 0.0750, 0.0727, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 03:16:08,714 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1308706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:16:15,402 INFO [optim.py:369] (1/2) 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 03:16:28,142 INFO [train.py:968] (1/2) Epoch 29, batch 32900, giga_loss[loss=0.2552, simple_loss=0.3114, pruned_loss=0.09952, over 24292.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.327, pruned_loss=0.086, over 5656477.13 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3466, pruned_loss=0.11, over 5705087.71 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3257, pruned_loss=0.08328, over 5653734.26 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:16:58,883 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.54 vs. limit=2.0
+2023-03-15 03:17:35,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1308769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:17:36,428 INFO [train.py:968] (1/2) Epoch 29, batch 32950, giga_loss[loss=0.2935, simple_loss=0.3712, pruned_loss=0.108, over 28748.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3267, pruned_loss=0.08476, over 5654243.66 frames. ], libri_tot_loss[loss=0.2836, simple_loss=0.3468, pruned_loss=0.1102, over 5702862.05 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.3252, pruned_loss=0.08228, over 5653474.51 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:17:40,689 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1308772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:18:14,064 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1308801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:18:24,521 INFO [optim.py:369] (1/2) 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 03:18:36,382 INFO [train.py:968] (1/2) Epoch 29, batch 33000, giga_loss[loss=0.2329, simple_loss=0.3158, pruned_loss=0.07503, over 29044.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3262, pruned_loss=0.08477, over 5651669.39 frames. ], libri_tot_loss[loss=0.2839, simple_loss=0.347, pruned_loss=0.1103, over 5699602.45 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3243, pruned_loss=0.08183, over 5652749.49 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:18:36,382 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 03:18:45,816 INFO [train.py:1012] (1/2) Epoch 29, validation: loss=0.1925, simple_loss=0.2942, pruned_loss=0.04546, over 944034.00 frames. 
+2023-03-15 03:18:45,816 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 03:19:39,123 INFO [train.py:968] (1/2) Epoch 29, batch 33050, giga_loss[loss=0.2195, simple_loss=0.3053, pruned_loss=0.06685, over 28646.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3272, pruned_loss=0.08626, over 5651467.32 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3466, pruned_loss=0.1102, over 5695426.00 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3255, pruned_loss=0.08327, over 5654777.04 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:20:31,160 INFO [optim.py:369] (1/2) 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 03:20:42,582 INFO [train.py:968] (1/2) Epoch 29, batch 33100, giga_loss[loss=0.2291, simple_loss=0.3156, pruned_loss=0.07128, over 28065.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3243, pruned_loss=0.08368, over 5650015.37 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3464, pruned_loss=0.1101, over 5696645.24 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3229, pruned_loss=0.08116, over 5651034.09 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:20:54,724 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3177, 3.2875, 1.4509, 1.5643], device='cuda:1'), covar=tensor([0.1125, 0.0437, 0.1021, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0574, 0.0415, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 03:21:29,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0661, 1.3920, 2.7241, 2.7072], device='cuda:1'), covar=tensor([0.1362, 0.2247, 0.0578, 0.1320], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0679, 0.1010, 0.0983], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 03:21:38,467 INFO [train.py:968] (1/2) Epoch 29, batch 33150, giga_loss[loss=0.2886, simple_loss=0.3672, pruned_loss=0.105, over 29031.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3267, pruned_loss=0.08345, over 5657822.60 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3464, pruned_loss=0.1103, over 5698166.65 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3251, pruned_loss=0.08072, over 5656793.02 frames. ], batch size: 200, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:21:56,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1308988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:22:25,893 INFO [optim.py:369] (1/2) 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 03:22:34,917 INFO [train.py:968] (1/2) Epoch 29, batch 33200, libri_loss[loss=0.3125, simple_loss=0.3773, pruned_loss=0.1238, over 29268.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3292, pruned_loss=0.08421, over 5660745.35 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.346, pruned_loss=0.1098, over 5704089.13 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3277, pruned_loss=0.08168, over 5653545.65 frames. ], batch size: 94, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:23:35,409 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 03:23:36,302 INFO [train.py:968] (1/2) Epoch 29, batch 33250, giga_loss[loss=0.222, simple_loss=0.3081, pruned_loss=0.06794, over 28743.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.33, pruned_loss=0.08468, over 5654820.10 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3464, pruned_loss=0.1101, over 5707944.95 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3281, pruned_loss=0.0819, over 5644636.31 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:23:52,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1309081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:24:12,629 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3156, 1.5626, 1.3849, 1.5484], device='cuda:1'), covar=tensor([0.0791, 0.0356, 0.0358, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0122, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 03:24:29,437 INFO [optim.py:369] (1/2) 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 03:24:37,692 INFO [train.py:968] (1/2) Epoch 29, batch 33300, libri_loss[loss=0.2343, simple_loss=0.2989, pruned_loss=0.08489, over 29651.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3304, pruned_loss=0.08539, over 5661610.80 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.346, pruned_loss=0.1099, over 5711060.94 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3289, pruned_loss=0.08276, over 5649369.20 frames. ], batch size: 73, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:24:47,255 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1309131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:24:50,593 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1309134.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:25:21,838 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 33350, giga_loss[loss=0.2245, simple_loss=0.3112, pruned_loss=0.06889, over 28897.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3282, pruned_loss=0.08411, over 5659323.29 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3453, pruned_loss=0.1095, over 5704730.44 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3271, pruned_loss=0.08171, over 5654305.81 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:25:47,780 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3779, 3.4502, 1.4870, 1.5366], device='cuda:1'), covar=tensor([0.1045, 0.0311, 0.0991, 0.1375], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0573, 0.0415, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 03:26:23,563 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 33400, giga_loss[loss=0.2239, simple_loss=0.298, pruned_loss=0.07491, over 29122.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3271, pruned_loss=0.08366, over 5663877.91 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.345, pruned_loss=0.1092, over 5706415.42 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3265, pruned_loss=0.08179, over 5657949.66 frames. ], batch size: 120, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:26:36,633 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1309224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:26:39,412 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1309227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:27:07,805 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1309256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:27:15,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5538, 2.1700, 1.5473, 0.7748], device='cuda:1'), covar=tensor([0.4573, 0.2569, 0.3870, 0.6063], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1744, 0.1661, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 03:27:27,048 INFO [train.py:968] (1/2) Epoch 29, batch 33450, giga_loss[loss=0.2583, simple_loss=0.3345, pruned_loss=0.09109, over 27719.00 frames. ], tot_loss[loss=0.2467, simple_loss=0.3259, pruned_loss=0.08372, over 5657879.07 frames. ], libri_tot_loss[loss=0.2815, simple_loss=0.3448, pruned_loss=0.1091, over 5696263.11 frames. ], giga_tot_loss[loss=0.2441, simple_loss=0.325, pruned_loss=0.08161, over 5659857.04 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:28:14,265 INFO [optim.py:369] (1/2) 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 03:28:25,183 INFO [train.py:968] (1/2) Epoch 29, batch 33500, giga_loss[loss=0.2644, simple_loss=0.3488, pruned_loss=0.09002, over 28941.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3285, pruned_loss=0.08485, over 5656517.97 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3443, pruned_loss=0.1089, over 5691418.40 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3277, pruned_loss=0.08261, over 5661646.72 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:29:27,715 INFO [train.py:968] (1/2) Epoch 29, batch 33550, giga_loss[loss=0.2375, simple_loss=0.3191, pruned_loss=0.07796, over 28846.00 frames. ], tot_loss[loss=0.2509, simple_loss=0.3302, pruned_loss=0.08582, over 5663931.50 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3444, pruned_loss=0.109, over 5694511.96 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3293, pruned_loss=0.08365, over 5664560.80 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:30:21,539 INFO [optim.py:369] (1/2) 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 03:30:30,562 INFO [train.py:968] (1/2) Epoch 29, batch 33600, giga_loss[loss=0.2371, simple_loss=0.3205, pruned_loss=0.07686, over 29053.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3306, pruned_loss=0.08619, over 5656959.78 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3443, pruned_loss=0.109, over 5685725.84 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3296, pruned_loss=0.084, over 5663875.60 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:31:09,186 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 33650, giga_loss[loss=0.2313, simple_loss=0.3187, pruned_loss=0.07196, over 28786.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3337, pruned_loss=0.08719, over 5645743.35 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3443, pruned_loss=0.109, over 5673979.67 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3329, pruned_loss=0.08531, over 5660378.15 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:32:19,247 INFO [optim.py:369] (1/2) 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 03:32:27,773 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 03:32:27,993 INFO [train.py:968] (1/2) Epoch 29, batch 33700, giga_loss[loss=0.2346, simple_loss=0.3191, pruned_loss=0.07504, over 28522.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3342, pruned_loss=0.08672, over 5652110.21 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3438, pruned_loss=0.1087, over 5679973.78 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3337, pruned_loss=0.0849, over 5658021.90 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:32:34,510 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.78 vs. limit=5.0
+2023-03-15 03:33:38,383 INFO [train.py:968] (1/2) Epoch 29, batch 33750, giga_loss[loss=0.2661, simple_loss=0.334, pruned_loss=0.09912, over 26807.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3331, pruned_loss=0.08631, over 5653578.94 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3437, pruned_loss=0.1086, over 5682834.30 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3326, pruned_loss=0.08455, over 5655347.19 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:34:32,735 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 03:34:32,934 INFO [optim.py:369] (1/2) 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 03:34:37,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2245, 1.2242, 3.5801, 3.1559], device='cuda:1'), covar=tensor([0.1715, 0.2872, 0.0536, 0.1051], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0679, 0.1010, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 03:34:41,879 INFO [train.py:968] (1/2) Epoch 29, batch 33800, giga_loss[loss=0.2852, simple_loss=0.343, pruned_loss=0.1137, over 26903.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3308, pruned_loss=0.08556, over 5663348.64 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.343, pruned_loss=0.1082, over 5687809.85 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3307, pruned_loss=0.08391, over 5659623.84 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:34:54,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9612, 2.4049, 2.2382, 1.9809], device='cuda:1'), covar=tensor([0.2184, 0.2232, 0.2108, 0.2366], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0747, 0.0725, 0.0689], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 03:35:00,405 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.76 vs. limit=5.0
+2023-03-15 03:35:27,484 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0662, 2.4864, 2.3922, 2.0720], device='cuda:1'), covar=tensor([0.2381, 0.2449, 0.2209, 0.2578], device='cuda:1'), in_proj_covar=tensor([0.0498, 0.0748, 0.0726, 0.0690], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 03:35:36,201 INFO [train.py:968] (1/2) Epoch 29, batch 33850, giga_loss[loss=0.2448, simple_loss=0.3117, pruned_loss=0.08897, over 24513.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.3308, pruned_loss=0.08595, over 5670122.42 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3424, pruned_loss=0.1078, over 5696932.68 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3308, pruned_loss=0.08402, over 5657975.38 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:36:32,731 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 03:36:32,859 INFO [optim.py:369] (1/2) 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 03:36:42,098 INFO [train.py:968] (1/2) Epoch 29, batch 33900, giga_loss[loss=0.2587, simple_loss=0.3351, pruned_loss=0.09116, over 28030.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3312, pruned_loss=0.08704, over 5655316.12 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3423, pruned_loss=0.1078, over 5696236.33 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3311, pruned_loss=0.08527, over 5646118.97 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:37:44,391 INFO [train.py:968] (1/2) Epoch 29, batch 33950, giga_loss[loss=0.2451, simple_loss=0.32, pruned_loss=0.08512, over 28972.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3291, pruned_loss=0.08673, over 5653338.70 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3419, pruned_loss=0.1076, over 5692740.21 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3292, pruned_loss=0.08506, over 5647714.80 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:38:34,229 INFO [optim.py:369] (1/2) 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 03:38:43,697 INFO [train.py:968] (1/2) Epoch 29, batch 34000, giga_loss[loss=0.2096, simple_loss=0.2847, pruned_loss=0.06723, over 24539.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3292, pruned_loss=0.0866, over 5647526.10 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3417, pruned_loss=0.1075, over 5696013.51 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3293, pruned_loss=0.08512, over 5639888.45 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 03:38:49,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1309824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:38:54,054 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 03:39:41,295 INFO [train.py:968] (1/2) Epoch 29, batch 34050, giga_loss[loss=0.2426, simple_loss=0.3259, pruned_loss=0.07967, over 28609.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.328, pruned_loss=0.08477, over 5667333.35 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.3416, pruned_loss=0.1074, over 5700400.61 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3278, pruned_loss=0.08299, over 5656337.85 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:40:24,236 INFO [optim.py:369] (1/2) 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 03:40:30,321 INFO [train.py:968] (1/2) Epoch 29, batch 34100, giga_loss[loss=0.3231, simple_loss=0.3858, pruned_loss=0.1302, over 26836.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3296, pruned_loss=0.08444, over 5672930.16 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3413, pruned_loss=0.1073, over 5698858.81 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3291, pruned_loss=0.08202, over 5663967.13 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:40:39,811 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 03:41:22,815 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.58 vs. limit=5.0
+2023-03-15 03:41:23,250 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1309967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:41:27,729 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 34150, giga_loss[loss=0.235, simple_loss=0.3243, pruned_loss=0.07287, over 28059.00 frames. ], tot_loss[loss=0.2499, simple_loss=0.3315, pruned_loss=0.08419, over 5669725.98 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.341, pruned_loss=0.1072, over 5700819.24 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.331, pruned_loss=0.0818, over 5660390.36 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:41:32,728 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.96 vs. limit=5.0
+2023-03-15 03:41:56,792 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1309999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:42:15,914 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 34200, giga_loss[loss=0.2326, simple_loss=0.3173, pruned_loss=0.07389, over 28596.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.331, pruned_loss=0.08365, over 5666595.22 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3413, pruned_loss=0.1074, over 5702033.95 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3301, pruned_loss=0.08103, over 5657161.95 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:42:42,683 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4824, 1.7812, 1.3870, 1.5228], device='cuda:1'), covar=tensor([0.2912, 0.2968, 0.3444, 0.2554], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1161, 0.1431, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 03:43:18,305 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7452, 3.5809, 3.3921, 1.6772], device='cuda:1'), covar=tensor([0.0778, 0.0894, 0.0830, 0.2341], device='cuda:1'), in_proj_covar=tensor([0.1293, 0.1197, 0.1004, 0.0740], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 03:43:31,834 INFO [train.py:968] (1/2) Epoch 29, batch 34250, giga_loss[loss=0.2654, simple_loss=0.3494, pruned_loss=0.09075, over 28460.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3312, pruned_loss=0.08349, over 5667359.70 frames. ], libri_tot_loss[loss=0.2785, simple_loss=0.3417, pruned_loss=0.1076, over 5696045.58 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.33, pruned_loss=0.08078, over 5665134.91 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:43:36,297 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4767, 1.4828, 1.5135, 1.1415], device='cuda:1'), covar=tensor([0.2131, 0.3395, 0.1736, 0.1936], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0711, 0.0986, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 03:44:31,949 INFO [optim.py:369] (1/2) 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 03:44:40,532 INFO [train.py:968] (1/2) Epoch 29, batch 34300, libri_loss[loss=0.3383, simple_loss=0.386, pruned_loss=0.1453, over 19931.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3308, pruned_loss=0.08283, over 5655394.36 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3416, pruned_loss=0.1076, over 5688688.14 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3298, pruned_loss=0.08047, over 5660978.85 frames. ], batch size: 187, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:44:41,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.71 vs. limit=5.0
+2023-03-15 03:45:41,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5295, 1.6145, 1.7613, 1.3396], device='cuda:1'), covar=tensor([0.1720, 0.2667, 0.1475, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0710, 0.0985, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 03:45:43,950 INFO [train.py:968] (1/2) Epoch 29, batch 34350, giga_loss[loss=0.2623, simple_loss=0.3458, pruned_loss=0.08939, over 28148.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3315, pruned_loss=0.08342, over 5659890.31 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3415, pruned_loss=0.1076, over 5695712.45 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3305, pruned_loss=0.08076, over 5656935.31 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 03:46:32,006 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3060, 1.2432, 4.0184, 3.3719], device='cuda:1'), covar=tensor([0.1696, 0.2756, 0.0505, 0.0975], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0677, 0.1008, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 03:46:42,903 INFO [optim.py:369] (1/2) 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 03:46:46,433 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 03:46:51,460 INFO [train.py:968] (1/2) Epoch 29, batch 34400, libri_loss[loss=0.2637, simple_loss=0.3299, pruned_loss=0.09875, over 29551.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.333, pruned_loss=0.08437, over 5655982.67 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3411, pruned_loss=0.1074, over 5691644.93 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3322, pruned_loss=0.08175, over 5655730.09 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:47:50,083 INFO [train.py:968] (1/2) Epoch 29, batch 34450, giga_loss[loss=0.2303, simple_loss=0.3161, pruned_loss=0.07222, over 28405.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3354, pruned_loss=0.08504, over 5665101.86 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3411, pruned_loss=0.1075, over 5697216.46 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3347, pruned_loss=0.08238, over 5659263.94 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:48:35,744 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:48:46,625 INFO [optim.py:369] (1/2) 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 03:48:54,057 INFO [train.py:968] (1/2) Epoch 29, batch 34500, giga_loss[loss=0.3092, simple_loss=0.3691, pruned_loss=0.1246, over 27662.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3357, pruned_loss=0.08546, over 5676765.72 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3405, pruned_loss=0.107, over 5702197.02 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3356, pruned_loss=0.08314, over 5667004.79 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:50:00,678 INFO [train.py:968] (1/2) Epoch 29, batch 34550, giga_loss[loss=0.2237, simple_loss=0.3048, pruned_loss=0.07129, over 28585.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3341, pruned_loss=0.08491, over 5688959.28 frames. ], libri_tot_loss[loss=0.2772, simple_loss=0.3405, pruned_loss=0.107, over 5705142.94 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3339, pruned_loss=0.08284, over 5678364.53 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:50:51,573 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:51:07,687 INFO [train.py:968] (1/2) Epoch 29, batch 34600, libri_loss[loss=0.2808, simple_loss=0.3522, pruned_loss=0.1047, over 29536.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.333, pruned_loss=0.08429, over 5684806.84 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3404, pruned_loss=0.1068, over 5701696.97 frames. ], giga_tot_loss[loss=0.2484, simple_loss=0.3327, pruned_loss=0.08203, over 5678543.82 frames. ], batch size: 84, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:52:05,428 INFO [train.py:968] (1/2) Epoch 29, batch 34650, giga_loss[loss=0.2399, simple_loss=0.3271, pruned_loss=0.07634, over 28506.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.3318, pruned_loss=0.08343, over 5683343.14 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3402, pruned_loss=0.1069, over 5688946.77 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3314, pruned_loss=0.08084, over 5689648.91 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:53:03,772 INFO [optim.py:369] (1/2) 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 03:53:07,919 INFO [train.py:968] (1/2) Epoch 29, batch 34700, giga_loss[loss=0.2494, simple_loss=0.3375, pruned_loss=0.08071, over 28425.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3314, pruned_loss=0.08317, over 5681872.26 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3398, pruned_loss=0.1066, over 5691949.78 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3313, pruned_loss=0.08093, over 5683721.53 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:53:11,875 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.90 vs. limit=5.0
+2023-03-15 03:54:02,235 INFO [train.py:968] (1/2) Epoch 29, batch 34750, giga_loss[loss=0.2424, simple_loss=0.3319, pruned_loss=0.07642, over 28980.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.3339, pruned_loss=0.0849, over 5665513.58 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3399, pruned_loss=0.1066, over 5680392.60 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3336, pruned_loss=0.08242, over 5676828.33 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:54:11,210 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:54:21,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 03:54:51,856 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4313, 1.9261, 1.7547, 1.5979], device='cuda:1'), covar=tensor([0.2278, 0.1964, 0.2235, 0.2181], device='cuda:1'), in_proj_covar=tensor([0.0493, 0.0740, 0.0720, 0.0684], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 03:54:53,650 INFO [optim.py:369] (1/2) 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 03:54:59,478 INFO [train.py:968] (1/2) Epoch 29, batch 34800, giga_loss[loss=0.2267, simple_loss=0.3137, pruned_loss=0.06983, over 28666.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3334, pruned_loss=0.08498, over 5646814.26 frames. ], libri_tot_loss[loss=0.2765, simple_loss=0.3399, pruned_loss=0.1066, over 5666797.33 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.333, pruned_loss=0.08262, over 5667630.41 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 03:55:13,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5592, 1.7808, 1.7463, 1.5023], device='cuda:1'), covar=tensor([0.3265, 0.2749, 0.2350, 0.2790], device='cuda:1'), in_proj_covar=tensor([0.2043, 0.2004, 0.1889, 0.2042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 03:55:24,453 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310640.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 03:55:54,952 INFO [train.py:968] (1/2) Epoch 29, batch 34850, giga_loss[loss=0.2378, simple_loss=0.3211, pruned_loss=0.07724, over 28528.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3305, pruned_loss=0.0845, over 5651074.71 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3395, pruned_loss=0.1063, over 5669127.78 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.3303, pruned_loss=0.08224, over 5664915.39 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 03:56:03,887 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 03:56:48,751 INFO [train.py:968] (1/2) Epoch 29, batch 34900, giga_loss[loss=0.2858, simple_loss=0.3571, pruned_loss=0.1073, over 28630.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3315, pruned_loss=0.08602, over 5650346.14 frames. ], libri_tot_loss[loss=0.2764, simple_loss=0.3398, pruned_loss=0.1065, over 5663230.21 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.331, pruned_loss=0.08358, over 5666141.69 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:57:38,473 INFO [train.py:968] (1/2) Epoch 29, batch 34950, giga_loss[loss=0.2385, simple_loss=0.3232, pruned_loss=0.07694, over 28060.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3357, pruned_loss=0.08859, over 5653321.12 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3398, pruned_loss=0.1064, over 5668090.38 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.335, pruned_loss=0.08607, over 5661317.16 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:57:50,657 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1310783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:58:21,962 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 35000, giga_loss[loss=0.2916, simple_loss=0.3734, pruned_loss=0.1049, over 29114.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3449, pruned_loss=0.09293, over 5663283.44 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3398, pruned_loss=0.1064, over 5669058.93 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3443, pruned_loss=0.09086, over 5668670.72 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:58:27,988 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:58:29,735 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1310824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:58:32,619 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1310827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:58:58,473 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1310856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:59:09,229 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 03:59:11,578 INFO [train.py:968] (1/2) Epoch 29, batch 35050, giga_loss[loss=0.2413, simple_loss=0.3307, pruned_loss=0.07597, over 28638.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3495, pruned_loss=0.09565, over 5652114.10 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.34, pruned_loss=0.1066, over 5653002.94 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3489, pruned_loss=0.09368, over 5670509.13 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:59:26,420 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-15 03:59:40,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 03:59:47,305 INFO [optim.py:369] (1/2) 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 03:59:52,171 INFO [train.py:968] (1/2) Epoch 29, batch 35100, giga_loss[loss=0.2432, simple_loss=0.3212, pruned_loss=0.08254, over 28250.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3458, pruned_loss=0.09458, over 5665487.99 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3396, pruned_loss=0.1063, over 5660403.83 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.3459, pruned_loss=0.09303, over 5673802.49 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 03:59:56,916 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1310926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 03:59:58,803 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1310929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:11,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1310945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:18,654 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1310953.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:23,507 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1310958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:00:34,982 INFO [train.py:968] (1/2) Epoch 29, batch 35150, giga_loss[loss=0.233, simple_loss=0.3129, pruned_loss=0.07655, over 28250.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3386, pruned_loss=0.09156, over 5660873.95 frames. ], libri_tot_loss[loss=0.2763, simple_loss=0.3398, pruned_loss=0.1064, over 5652974.75 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3385, pruned_loss=0.09005, over 5674134.07 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:01:03,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1706, 1.2713, 1.1784, 0.8545], device='cuda:1'), covar=tensor([0.1183, 0.0584, 0.1143, 0.1225], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0446, 0.0522, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 04:01:04,304 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 04:01:11,305 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311015.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:01:11,675 INFO [optim.py:369] (1/2) 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 04:01:15,004 INFO [train.py:968] (1/2) Epoch 29, batch 35200, giga_loss[loss=0.2256, simple_loss=0.3024, pruned_loss=0.0744, over 29075.00 frames. ], tot_loss[loss=0.2555, simple_loss=0.3324, pruned_loss=0.08936, over 5672974.43 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3403, pruned_loss=0.1067, over 5658552.31 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3317, pruned_loss=0.08734, over 5679228.90 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:01:54,646 INFO [train.py:968] (1/2) Epoch 29, batch 35250, giga_loss[loss=0.2134, simple_loss=0.2906, pruned_loss=0.06807, over 28641.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3244, pruned_loss=0.08579, over 5672498.68 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3405, pruned_loss=0.1067, over 5661397.42 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3235, pruned_loss=0.08374, over 5674916.25 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:02:13,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:02:15,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:02:23,947 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6481, 1.6945, 1.8526, 1.4369], device='cuda:1'), covar=tensor([0.2021, 0.2744, 0.1644, 0.1871], device='cuda:1'), in_proj_covar=tensor([0.0941, 0.0714, 0.0991, 0.0891], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 04:02:29,613 INFO [optim.py:369] (1/2) 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 04:02:32,141 INFO [train.py:968] (1/2) Epoch 29, batch 35300, giga_loss[loss=0.222, simple_loss=0.2963, pruned_loss=0.07382, over 28825.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3185, pruned_loss=0.08328, over 5682230.77 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3404, pruned_loss=0.1066, over 5667221.78 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3171, pruned_loss=0.08104, over 5679294.14 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:02:39,763 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:03:05,280 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311158.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:03:07,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311161.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:03:16,430 INFO [train.py:968] (1/2) Epoch 29, batch 35350, giga_loss[loss=0.1903, simple_loss=0.2658, pruned_loss=0.05739, over 28621.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.3143, pruned_loss=0.08147, over 5689361.86 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3406, pruned_loss=0.1066, over 5668655.76 frames. ], giga_tot_loss[loss=0.2355, simple_loss=0.3126, pruned_loss=0.07921, over 5686040.34 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:03:16,660 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:03:30,010 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:03:30,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311190.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:03:38,834 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:03:47,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-15 04:03:51,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.81 vs. limit=2.0
+2023-03-15 04:03:53,929 INFO [optim.py:369] (1/2) 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 04:03:58,087 INFO [train.py:968] (1/2) Epoch 29, batch 35400, giga_loss[loss=0.2579, simple_loss=0.3299, pruned_loss=0.09288, over 27930.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3114, pruned_loss=0.08038, over 5691592.73 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3405, pruned_loss=0.1063, over 5673083.40 frames. ], giga_tot_loss[loss=0.2332, simple_loss=0.3096, pruned_loss=0.07836, over 5685230.38 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:04:31,134 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3669, 1.6404, 1.4402, 1.5985], device='cuda:1'), covar=tensor([0.0779, 0.0382, 0.0359, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0122, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 04:04:38,174 INFO [train.py:968] (1/2) Epoch 29, batch 35450, giga_loss[loss=0.2031, simple_loss=0.2788, pruned_loss=0.06365, over 28494.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.3102, pruned_loss=0.08042, over 5678732.49 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3415, pruned_loss=0.1067, over 5666144.84 frames. ], giga_tot_loss[loss=0.2307, simple_loss=0.3067, pruned_loss=0.07739, over 5679877.06 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:04:57,155 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4987, 1.5297, 1.2233, 1.1383], device='cuda:1'), covar=tensor([0.0791, 0.0362, 0.0874, 0.1127], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0446, 0.0521, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 04:05:14,445 INFO [optim.py:369] (1/2) 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 04:05:17,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:05:17,878 INFO [train.py:968] (1/2) Epoch 29, batch 35500, giga_loss[loss=0.1993, simple_loss=0.2778, pruned_loss=0.06042, over 28938.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.3079, pruned_loss=0.07959, over 5678710.68 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3421, pruned_loss=0.1069, over 5672950.68 frames. ], giga_tot_loss[loss=0.2278, simple_loss=0.3035, pruned_loss=0.0761, over 5673646.52 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:05:33,727 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:05:35,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:06:00,952 INFO [train.py:968] (1/2) Epoch 29, batch 35550, giga_loss[loss=0.2027, simple_loss=0.2779, pruned_loss=0.06376, over 28722.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3046, pruned_loss=0.07806, over 5675745.99 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3422, pruned_loss=0.1068, over 5667147.55 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.3003, pruned_loss=0.07484, over 5678102.69 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:06:01,855 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311372.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:06:39,081 INFO [optim.py:369] (1/2) 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 04:06:39,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9162, 2.0858, 1.8279, 1.8074], device='cuda:1'), covar=tensor([0.2733, 0.2976, 0.3360, 0.2815], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1161, 0.1430, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 04:06:41,674 INFO [train.py:968] (1/2) Epoch 29, batch 35600, giga_loss[loss=0.2202, simple_loss=0.291, pruned_loss=0.07469, over 28558.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.303, pruned_loss=0.07724, over 5681286.51 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3428, pruned_loss=0.1069, over 5668232.43 frames. ], giga_tot_loss[loss=0.2231, simple_loss=0.2982, pruned_loss=0.07395, over 5682063.51 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:07:15,558 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:07:17,653 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 35650, giga_loss[loss=0.2206, simple_loss=0.29, pruned_loss=0.07565, over 28929.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.3005, pruned_loss=0.07619, over 5685624.66 frames. ], libri_tot_loss[loss=0.278, simple_loss=0.3427, pruned_loss=0.1066, over 5671570.29 frames. ], giga_tot_loss[loss=0.221, simple_loss=0.2958, pruned_loss=0.07314, over 5683597.04 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:07:43,290 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:07:54,543 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0129, 2.0316, 2.2653, 1.7828], device='cuda:1'), covar=tensor([0.1949, 0.2636, 0.1553, 0.1816], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.0718, 0.0996, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 04:07:59,043 INFO [optim.py:369] (1/2) 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 04:08:02,001 INFO [train.py:968] (1/2) Epoch 29, batch 35700, giga_loss[loss=0.2258, simple_loss=0.2908, pruned_loss=0.08042, over 27731.00 frames. ], tot_loss[loss=0.2265, simple_loss=0.3003, pruned_loss=0.0764, over 5683450.44 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.343, pruned_loss=0.1066, over 5672133.75 frames. ], giga_tot_loss[loss=0.22, simple_loss=0.2945, pruned_loss=0.0727, over 5681625.90 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:08:22,711 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:08:41,719 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:08:43,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.68 vs. limit=2.0
+2023-03-15 04:08:46,037 INFO [train.py:968] (1/2) Epoch 29, batch 35750, giga_loss[loss=0.2316, simple_loss=0.3086, pruned_loss=0.07732, over 28497.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.298, pruned_loss=0.07581, over 5670955.46 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3436, pruned_loss=0.1069, over 5664807.95 frames. ], giga_tot_loss[loss=0.2183, simple_loss=0.2923, pruned_loss=0.07213, over 5675692.71 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:09:17,450 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:09:26,163 INFO [optim.py:369] (1/2) 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 04:09:28,612 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 35800, giga_loss[loss=0.277, simple_loss=0.3447, pruned_loss=0.1047, over 28369.00 frames. ], tot_loss[loss=0.2319, simple_loss=0.3046, pruned_loss=0.07957, over 5660469.91 frames. ], libri_tot_loss[loss=0.2786, simple_loss=0.3434, pruned_loss=0.1069, over 5651786.21 frames. ], giga_tot_loss[loss=0.2258, simple_loss=0.2993, pruned_loss=0.07609, over 5674778.85 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:10:16,472 INFO [train.py:968] (1/2) Epoch 29, batch 35850, giga_loss[loss=0.293, simple_loss=0.3742, pruned_loss=0.1059, over 28579.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3172, pruned_loss=0.08575, over 5658406.02 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3437, pruned_loss=0.107, over 5645420.64 frames. ], giga_tot_loss[loss=0.2388, simple_loss=0.3123, pruned_loss=0.08261, over 5675675.57 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:10:19,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-15 04:10:24,136 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.1718, 4.0064, 3.7881, 1.7428], device='cuda:1'), covar=tensor([0.0670, 0.0818, 0.0808, 0.2338], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1201, 0.1008, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:10:34,104 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:10:38,015 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8962, 3.7265, 3.5011, 1.6137], device='cuda:1'), covar=tensor([0.0737, 0.0875, 0.0886, 0.2337], device='cuda:1'), in_proj_covar=tensor([0.1299, 0.1199, 0.1007, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:10:38,091 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:10:53,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1311708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:10:55,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1311711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:10:58,810 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2864, 1.6339, 1.3095, 0.9262], device='cuda:1'), covar=tensor([0.2638, 0.2625, 0.2944, 0.2598], device='cuda:1'), in_proj_covar=tensor([0.1615, 0.1160, 0.1427, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 04:11:00,626 INFO [optim.py:369] (1/2) 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 04:11:03,439 INFO [train.py:968] (1/2) Epoch 29, batch 35900, giga_loss[loss=0.2965, simple_loss=0.3781, pruned_loss=0.1075, over 28888.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3288, pruned_loss=0.09139, over 5664407.40 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3437, pruned_loss=0.107, over 5645420.64 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.325, pruned_loss=0.08893, over 5677848.51 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:11:03,656 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:11:17,906 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1311740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:11:43,653 INFO [train.py:968] (1/2) Epoch 29, batch 35950, giga_loss[loss=0.2673, simple_loss=0.3564, pruned_loss=0.08906, over 28681.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3359, pruned_loss=0.09381, over 5666646.00 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3442, pruned_loss=0.1073, over 5641857.39 frames. ], giga_tot_loss[loss=0.2572, simple_loss=0.3321, pruned_loss=0.09121, over 5681085.33 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:12:10,398 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:12:17,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:12:23,983 INFO [optim.py:369] (1/2) 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 04:12:24,903 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1311819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:12:26,176 INFO [train.py:968] (1/2) Epoch 29, batch 36000, giga_loss[loss=0.2406, simple_loss=0.3186, pruned_loss=0.08135, over 23588.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.3385, pruned_loss=0.09396, over 5662626.64 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3442, pruned_loss=0.1072, over 5638213.60 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3353, pruned_loss=0.09166, over 5677987.93 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:12:26,176 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 04:12:33,532 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2457, 1.6975, 1.3098, 0.4677], device='cuda:1'), covar=tensor([0.5939, 0.4380, 0.5680, 0.7797], device='cuda:1'), in_proj_covar=tensor([0.1854, 0.1744, 0.1669, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 04:12:36,686 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 04:13:18,353 INFO [train.py:968] (1/2) Epoch 29, batch 36050, giga_loss[loss=0.2754, simple_loss=0.3524, pruned_loss=0.09917, over 28583.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3397, pruned_loss=0.09343, over 5660645.52 frames. ], libri_tot_loss[loss=0.2791, simple_loss=0.3442, pruned_loss=0.107, over 5642888.28 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.337, pruned_loss=0.09139, over 5669445.49 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:13:54,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7510, 4.3044, 1.7886, 1.9076], device='cuda:1'), covar=tensor([0.1002, 0.0270, 0.0890, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0571, 0.0412, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 04:14:01,484 INFO [optim.py:369] (1/2) 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 04:14:04,298 INFO [train.py:968] (1/2) Epoch 29, batch 36100, giga_loss[loss=0.2703, simple_loss=0.3463, pruned_loss=0.09718, over 27891.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3412, pruned_loss=0.09412, over 5669498.69 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3439, pruned_loss=0.1068, over 5646797.23 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.3392, pruned_loss=0.09249, over 5673210.09 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:14:44,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3268, 2.9213, 1.3623, 1.4524], device='cuda:1'), covar=tensor([0.1110, 0.0317, 0.0959, 0.1460], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0571, 0.0412, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 04:14:46,584 INFO [train.py:968] (1/2) Epoch 29, batch 36150, giga_loss[loss=0.274, simple_loss=0.3485, pruned_loss=0.09973, over 28762.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3437, pruned_loss=0.09588, over 5673142.54 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3445, pruned_loss=0.1071, over 5642506.58 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3417, pruned_loss=0.09418, over 5680129.63 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:14:57,658 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:15:08,425 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1311995.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:15:21,185 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:15:28,193 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 36200, giga_loss[loss=0.3071, simple_loss=0.3818, pruned_loss=0.1162, over 27849.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3481, pruned_loss=0.09874, over 5674399.37 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3453, pruned_loss=0.1077, over 5644479.36 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3457, pruned_loss=0.09677, over 5678374.26 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:16:09,786 INFO [train.py:968] (1/2) Epoch 29, batch 36250, giga_loss[loss=0.2464, simple_loss=0.3415, pruned_loss=0.07565, over 28500.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3513, pruned_loss=0.09921, over 5690596.88 frames. ], libri_tot_loss[loss=0.281, simple_loss=0.346, pruned_loss=0.108, over 5646041.97 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.3488, pruned_loss=0.0972, over 5693125.37 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:16:10,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1971, 1.5246, 1.4544, 1.0695], device='cuda:1'), covar=tensor([0.1514, 0.2668, 0.1390, 0.1712], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0717, 0.0994, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 04:16:11,262 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2230, 2.2175, 2.0619, 2.0131], device='cuda:1'), covar=tensor([0.2134, 0.2442, 0.2297, 0.2280], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0752, 0.0728, 0.0692], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 04:16:17,616 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2515, 4.0674, 3.8386, 2.0194], device='cuda:1'), covar=tensor([0.0592, 0.0768, 0.0736, 0.2059], device='cuda:1'), in_proj_covar=tensor([0.1291, 0.1193, 0.1000, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:16:22,185 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7173, 1.0891, 2.8246, 2.6727], device='cuda:1'), covar=tensor([0.1919, 0.2887, 0.0647, 0.1090], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0676, 0.1011, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 04:16:51,645 INFO [optim.py:369] (1/2) 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 04:16:52,327 INFO [train.py:968] (1/2) Epoch 29, batch 36300, giga_loss[loss=0.2723, simple_loss=0.354, pruned_loss=0.09528, over 28659.00 frames. ], tot_loss[loss=0.2761, simple_loss=0.3531, pruned_loss=0.0996, over 5686605.48 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3462, pruned_loss=0.108, over 5647313.04 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.351, pruned_loss=0.09788, over 5687940.88 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:16:55,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4954, 1.5629, 1.7052, 1.2975], device='cuda:1'), covar=tensor([0.1841, 0.2829, 0.1552, 0.1932], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0716, 0.0993, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 04:16:56,586 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:16:58,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:09,399 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312138.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:17:12,188 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312141.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:17:14,661 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3111, 3.1202, 2.9725, 1.3607], device='cuda:1'), covar=tensor([0.0903, 0.1102, 0.0902, 0.2404], device='cuda:1'), in_proj_covar=tensor([0.1287, 0.1191, 0.0999, 0.0741], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:17:23,029 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:33,269 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312170.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:17:33,569 INFO [train.py:968] (1/2) Epoch 29, batch 36350, giga_loss[loss=0.2928, simple_loss=0.3498, pruned_loss=0.118, over 23667.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.353, pruned_loss=0.09808, over 5693721.77 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3464, pruned_loss=0.1081, over 5649946.41 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3514, pruned_loss=0.09658, over 5692780.42 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:17:37,450 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:44,044 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:17:52,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:18:12,930 INFO [optim.py:369] (1/2) 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 04:18:13,512 INFO [train.py:968] (1/2) Epoch 29, batch 36400, libri_loss[loss=0.3177, simple_loss=0.3683, pruned_loss=0.1336, over 29628.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3524, pruned_loss=0.09679, over 5698411.52 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3468, pruned_loss=0.1082, over 5655875.22 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3509, pruned_loss=0.09523, over 5693223.09 frames. ], batch size: 69, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:18:54,092 INFO [train.py:968] (1/2) Epoch 29, batch 36450, giga_loss[loss=0.2513, simple_loss=0.3416, pruned_loss=0.08049, over 28576.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3509, pruned_loss=0.09509, over 5686566.13 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3473, pruned_loss=0.1083, over 5646025.69 frames. ], giga_tot_loss[loss=0.2682, simple_loss=0.3494, pruned_loss=0.09351, over 5693215.74 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:19:32,390 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312318.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:33,351 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 36500, giga_loss[loss=0.2942, simple_loss=0.385, pruned_loss=0.1017, over 29030.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3502, pruned_loss=0.09521, over 5676440.97 frames. ], libri_tot_loss[loss=0.2822, simple_loss=0.3475, pruned_loss=0.1084, over 5642497.42 frames. ], giga_tot_loss[loss=0.2679, simple_loss=0.3489, pruned_loss=0.09349, over 5686069.97 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:19:34,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:38,538 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:41,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:47,652 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:49,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:19:56,147 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3016, 1.6364, 1.5246, 1.4557], device='cuda:1'), covar=tensor([0.2358, 0.2187, 0.2713, 0.2397], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0754, 0.0730, 0.0695], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 04:19:59,359 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312350.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:05,312 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:15,846 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 36550, giga_loss[loss=0.3569, simple_loss=0.385, pruned_loss=0.1644, over 26514.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.352, pruned_loss=0.09861, over 5675450.21 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3473, pruned_loss=0.1082, over 5646348.65 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3512, pruned_loss=0.09712, over 5680904.42 frames. ], batch size: 555, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:20:17,909 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:20:28,691 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 29, batch 36600, giga_loss[loss=0.2897, simple_loss=0.3495, pruned_loss=0.115, over 28830.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3544, pruned_loss=0.1024, over 5684977.27 frames. ], libri_tot_loss[loss=0.282, simple_loss=0.3474, pruned_loss=0.1083, over 5651628.00 frames. ], giga_tot_loss[loss=0.2779, simple_loss=0.3538, pruned_loss=0.101, over 5685098.43 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:21:24,562 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 04:21:40,783 INFO [train.py:968] (1/2) Epoch 29, batch 36650, giga_loss[loss=0.2536, simple_loss=0.3314, pruned_loss=0.08792, over 28548.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3539, pruned_loss=0.1037, over 5684480.51 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3471, pruned_loss=0.1081, over 5648879.81 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3539, pruned_loss=0.1026, over 5688144.80 frames. ], batch size: 60, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:22:24,559 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.79 vs. limit=2.0
+2023-03-15 04:22:26,643 INFO [optim.py:369] (1/2) 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 04:22:27,253 INFO [train.py:968] (1/2) Epoch 29, batch 36700, giga_loss[loss=0.2539, simple_loss=0.3248, pruned_loss=0.09149, over 28791.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.353, pruned_loss=0.1042, over 5686733.32 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3475, pruned_loss=0.1083, over 5649493.70 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3526, pruned_loss=0.1031, over 5689474.07 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:22:30,455 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.61 vs. limit=2.0
+2023-03-15 04:22:30,853 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5076, 1.7227, 1.4307, 1.6952], device='cuda:1'), covar=tensor([0.2696, 0.2848, 0.3152, 0.2485], device='cuda:1'), in_proj_covar=tensor([0.1616, 0.1162, 0.1426, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 04:22:32,320 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:22:34,360 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:22:52,209 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.26 vs. limit=5.0
+2023-03-15 04:22:59,995 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:23:11,968 INFO [train.py:968] (1/2) Epoch 29, batch 36750, giga_loss[loss=0.2888, simple_loss=0.3628, pruned_loss=0.1074, over 28509.00 frames. ], tot_loss[loss=0.2782, simple_loss=0.3507, pruned_loss=0.1028, over 5690605.03 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3475, pruned_loss=0.1083, over 5650449.23 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3504, pruned_loss=0.102, over 5692143.67 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:23:54,242 INFO [optim.py:369] (1/2) 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 04:23:55,005 INFO [train.py:968] (1/2) Epoch 29, batch 36800, giga_loss[loss=0.2485, simple_loss=0.3354, pruned_loss=0.08076, over 29031.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3502, pruned_loss=0.102, over 5689141.62 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3479, pruned_loss=0.1084, over 5652920.82 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3497, pruned_loss=0.1011, over 5688429.92 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:24:25,688 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 36850, giga_loss[loss=0.2351, simple_loss=0.3177, pruned_loss=0.07626, over 28791.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3474, pruned_loss=0.09943, over 5700840.42 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.348, pruned_loss=0.1083, over 5656299.29 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.347, pruned_loss=0.09868, over 5697904.56 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:25:25,433 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 36900, giga_loss[loss=0.2147, simple_loss=0.2958, pruned_loss=0.0668, over 28837.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3425, pruned_loss=0.09659, over 5699724.67 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.348, pruned_loss=0.1081, over 5663392.99 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.342, pruned_loss=0.09591, over 5691872.66 frames. ], batch size: 243, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:25:43,679 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:25:47,927 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1312746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:26:11,041 INFO [train.py:968] (1/2) Epoch 29, batch 36950, giga_loss[loss=0.2597, simple_loss=0.3245, pruned_loss=0.0974, over 28699.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3367, pruned_loss=0.09342, over 5688221.87 frames. ], libri_tot_loss[loss=0.2825, simple_loss=0.3485, pruned_loss=0.1082, over 5659192.29 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3357, pruned_loss=0.09246, over 5687463.96 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:27:04,022 INFO [optim.py:369] (1/2) 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 04:27:04,035 INFO [train.py:968] (1/2) Epoch 29, batch 37000, libri_loss[loss=0.2713, simple_loss=0.3278, pruned_loss=0.1074, over 29681.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3321, pruned_loss=0.09154, over 5678499.77 frames. ], libri_tot_loss[loss=0.2826, simple_loss=0.3487, pruned_loss=0.1083, over 5663776.21 frames. ], giga_tot_loss[loss=0.2556, simple_loss=0.3307, pruned_loss=0.0903, over 5674048.51 frames. ], batch size: 69, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:27:07,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1811, 1.2825, 1.0583, 1.1450], device='cuda:1'), covar=tensor([0.1997, 0.2055, 0.1781, 0.2015], device='cuda:1'), in_proj_covar=tensor([0.2061, 0.2019, 0.1917, 0.2072], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 04:27:46,951 INFO [train.py:968] (1/2) Epoch 29, batch 37050, giga_loss[loss=0.2695, simple_loss=0.3456, pruned_loss=0.09673, over 28277.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3327, pruned_loss=0.09113, over 5677889.47 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3489, pruned_loss=0.1083, over 5666665.85 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3311, pruned_loss=0.0898, over 5672027.13 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:28:02,216 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1312889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:28:04,016 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1312892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:28:06,758 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7124, 1.9664, 1.7177, 1.6453], device='cuda:1'), covar=tensor([0.2357, 0.2506, 0.2697, 0.2687], device='cuda:1'), in_proj_covar=tensor([0.0508, 0.0757, 0.0735, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 04:28:10,398 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 04:28:26,354 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2711, 1.3131, 3.7648, 3.2752], device='cuda:1'), covar=tensor([0.1751, 0.2876, 0.0437, 0.1074], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0677, 0.1010, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 04:28:28,636 INFO [optim.py:369] (1/2) 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 04:28:28,649 INFO [train.py:968] (1/2) Epoch 29, batch 37100, giga_loss[loss=0.2368, simple_loss=0.3217, pruned_loss=0.07589, over 29041.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3336, pruned_loss=0.09098, over 5684038.38 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3494, pruned_loss=0.1085, over 5661011.59 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3316, pruned_loss=0.08942, over 5684211.55 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:28:28,844 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1312921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:29:01,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9363, 3.1167, 1.9064, 0.9480], device='cuda:1'), covar=tensor([0.8708, 0.2722, 0.4736, 0.8551], device='cuda:1'), in_proj_covar=tensor([0.1840, 0.1728, 0.1656, 0.1500], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 04:29:11,603 INFO [train.py:968] (1/2) Epoch 29, batch 37150, giga_loss[loss=0.3109, simple_loss=0.3699, pruned_loss=0.126, over 28373.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3337, pruned_loss=0.0917, over 5680345.01 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3496, pruned_loss=0.1087, over 5652814.53 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.332, pruned_loss=0.09029, over 5687331.85 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:29:17,917 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1312979.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:29:52,668 INFO [optim.py:369] (1/2) 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 04:29:52,680 INFO [train.py:968] (1/2) Epoch 29, batch 37200, giga_loss[loss=0.2814, simple_loss=0.3492, pruned_loss=0.1068, over 27686.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3325, pruned_loss=0.09145, over 5675273.38 frames. ], libri_tot_loss[loss=0.2838, simple_loss=0.3499, pruned_loss=0.1088, over 5645874.60 frames. ], giga_tot_loss[loss=0.2552, simple_loss=0.3305, pruned_loss=0.08992, over 5687614.52 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:29:56,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:30:29,747 INFO [train.py:968] (1/2) Epoch 29, batch 37250, giga_loss[loss=0.2305, simple_loss=0.3149, pruned_loss=0.07301, over 28822.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3316, pruned_loss=0.091, over 5694723.95 frames. ], libri_tot_loss[loss=0.2843, simple_loss=0.3507, pruned_loss=0.1089, over 5652451.81 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3288, pruned_loss=0.08917, over 5699509.20 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:31:01,710 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:31:06,251 INFO [train.py:968] (1/2) Epoch 29, batch 37300, giga_loss[loss=0.1995, simple_loss=0.2788, pruned_loss=0.06009, over 28374.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.33, pruned_loss=0.09038, over 5697756.03 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3519, pruned_loss=0.1095, over 5650548.22 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.326, pruned_loss=0.08783, over 5704983.80 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:31:06,855 INFO [optim.py:369] (1/2) 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 04:31:08,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8272, 2.2406, 1.4542, 1.6445], device='cuda:1'), covar=tensor([0.1166, 0.0657, 0.1097, 0.1275], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0449, 0.0523, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 04:31:45,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:31:45,935 INFO [train.py:968] (1/2) Epoch 29, batch 37350, giga_loss[loss=0.2467, simple_loss=0.3142, pruned_loss=0.08954, over 28818.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.328, pruned_loss=0.08972, over 5695893.85 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3521, pruned_loss=0.1095, over 5650337.85 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3243, pruned_loss=0.08742, over 5702546.71 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:31:47,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1313173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:32:09,499 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1313202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:32:19,914 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9084, 1.1770, 1.3334, 1.0233], device='cuda:1'), covar=tensor([0.2238, 0.1515, 0.2465, 0.1934], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0758, 0.0735, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 04:32:24,674 INFO [train.py:968] (1/2) Epoch 29, batch 37400, giga_loss[loss=0.2175, simple_loss=0.3018, pruned_loss=0.06657, over 28853.00 frames. ], tot_loss[loss=0.252, simple_loss=0.3263, pruned_loss=0.08886, over 5692889.42 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3529, pruned_loss=0.1098, over 5647542.96 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.322, pruned_loss=0.08625, over 5702587.67 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:32:25,274 INFO [optim.py:369] (1/2) 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 04:32:51,847 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:32:53,886 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1313260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:33:01,809 INFO [train.py:968] (1/2) Epoch 29, batch 37450, giga_loss[loss=0.2265, simple_loss=0.3033, pruned_loss=0.0749, over 28629.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3239, pruned_loss=0.08725, over 5707836.93 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3535, pruned_loss=0.1099, over 5653449.83 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3195, pruned_loss=0.08463, over 5711520.01 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:33:16,122 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0069, 3.2173, 2.1717, 1.1640], device='cuda:1'), covar=tensor([0.8830, 0.2548, 0.4140, 0.7690], device='cuda:1'), in_proj_covar=tensor([0.1840, 0.1724, 0.1654, 0.1500], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 04:33:16,696 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1313289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:33:17,392 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:33:23,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3790, 1.5220, 1.4408, 1.2735], device='cuda:1'), covar=tensor([0.3452, 0.3303, 0.2399, 0.3302], device='cuda:1'), in_proj_covar=tensor([0.2059, 0.2019, 0.1918, 0.2079], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 04:33:42,286 INFO [train.py:968] (1/2) Epoch 29, batch 37500, giga_loss[loss=0.2582, simple_loss=0.3239, pruned_loss=0.09621, over 28832.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3209, pruned_loss=0.08563, over 5706380.54 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3538, pruned_loss=0.1101, over 5644593.58 frames. ], giga_tot_loss[loss=0.2419, simple_loss=0.3171, pruned_loss=0.08333, over 5717365.72 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:33:42,863 INFO [optim.py:369] (1/2) 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 04:34:00,669 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4636, 1.6836, 1.6817, 1.4814], device='cuda:1'), covar=tensor([0.2407, 0.2419, 0.2755, 0.2564], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0759, 0.0736, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 04:34:08,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313354.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:34:23,027 INFO [train.py:968] (1/2) Epoch 29, batch 37550, giga_loss[loss=0.2422, simple_loss=0.3181, pruned_loss=0.08314, over 28881.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.319, pruned_loss=0.08433, over 5715912.83 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3538, pruned_loss=0.11, over 5648110.63 frames. ], giga_tot_loss[loss=0.24, simple_loss=0.3154, pruned_loss=0.08231, over 5722501.02 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:34:47,612 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7523, 4.6215, 4.3985, 2.0507], device='cuda:1'), covar=tensor([0.0549, 0.0655, 0.0646, 0.2079], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1196, 0.1005, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:35:03,678 INFO [train.py:968] (1/2) Epoch 29, batch 37600, giga_loss[loss=0.2424, simple_loss=0.3185, pruned_loss=0.08314, over 28986.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3201, pruned_loss=0.08527, over 5716130.24 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3546, pruned_loss=0.1102, over 5653037.92 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3159, pruned_loss=0.08289, over 5718518.34 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:35:04,382 INFO [optim.py:369] (1/2) 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 04:35:43,286 INFO [train.py:968] (1/2) Epoch 29, batch 37650, giga_loss[loss=0.2574, simple_loss=0.3331, pruned_loss=0.09086, over 28757.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3252, pruned_loss=0.08839, over 5698421.84 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3547, pruned_loss=0.1102, over 5639615.08 frames. ], giga_tot_loss[loss=0.2464, simple_loss=0.3209, pruned_loss=0.08595, over 5712893.62 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:36:05,298 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313497.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:36:08,449 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1313500.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:36:20,357 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 04:36:27,766 INFO [train.py:968] (1/2) Epoch 29, batch 37700, giga_loss[loss=0.3007, simple_loss=0.3699, pruned_loss=0.1158, over 28588.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3314, pruned_loss=0.09229, over 5692253.65 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3548, pruned_loss=0.1101, over 5636424.03 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3274, pruned_loss=0.08994, over 5707496.27 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:36:30,013 INFO [optim.py:369] (1/2) 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 04:36:31,623 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1912, 1.2287, 1.0929, 0.8861], device='cuda:1'), covar=tensor([0.0830, 0.0391, 0.0816, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0450, 0.0525, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 04:36:33,082 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:36:34,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1313529.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:36:53,982 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5232, 1.5958, 1.5211, 1.4003], device='cuda:1'), covar=tensor([0.2505, 0.2648, 0.2400, 0.2594], device='cuda:1'), in_proj_covar=tensor([0.2058, 0.2018, 0.1920, 0.2077], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 04:37:12,637 INFO [train.py:968] (1/2) Epoch 29, batch 37750, giga_loss[loss=0.3228, simple_loss=0.3732, pruned_loss=0.1362, over 23504.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.34, pruned_loss=0.09828, over 5676187.84 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3551, pruned_loss=0.1101, over 5634355.67 frames. ], giga_tot_loss[loss=0.2638, simple_loss=0.3359, pruned_loss=0.09588, over 5692764.55 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:37:29,844 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:38:00,753 INFO [train.py:968] (1/2) Epoch 29, batch 37800, libri_loss[loss=0.2891, simple_loss=0.337, pruned_loss=0.1205, over 29655.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.344, pruned_loss=0.1003, over 5669830.59 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3553, pruned_loss=0.1104, over 5642259.68 frames. ], giga_tot_loss[loss=0.2678, simple_loss=0.3401, pruned_loss=0.09773, over 5677197.93 frames. ], batch size: 69, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:38:02,768 INFO [optim.py:369] (1/2) 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 04:38:17,113 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5258, 2.2647, 1.7135, 0.7730], device='cuda:1'), covar=tensor([0.7693, 0.3912, 0.4794, 0.7587], device='cuda:1'), in_proj_covar=tensor([0.1845, 0.1731, 0.1659, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 04:38:37,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:38:42,105 INFO [train.py:968] (1/2) Epoch 29, batch 37850, giga_loss[loss=0.2932, simple_loss=0.3689, pruned_loss=0.1087, over 28899.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3476, pruned_loss=0.1012, over 5676485.57 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3548, pruned_loss=0.1101, over 5645225.45 frames. ], giga_tot_loss[loss=0.2715, simple_loss=0.3445, pruned_loss=0.0992, over 5680708.98 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:39:00,755 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3715, 1.7992, 1.6392, 1.6047], device='cuda:1'), covar=tensor([0.2523, 0.2022, 0.2528, 0.2207], device='cuda:1'), in_proj_covar=tensor([0.0510, 0.0759, 0.0737, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 04:39:28,819 INFO [train.py:968] (1/2) Epoch 29, batch 37900, giga_loss[loss=0.3184, simple_loss=0.3918, pruned_loss=0.1225, over 28644.00 frames. ], tot_loss[loss=0.2806, simple_loss=0.353, pruned_loss=0.1041, over 5668753.92 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3551, pruned_loss=0.1102, over 5647497.05 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3502, pruned_loss=0.1023, over 5670303.60 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:39:30,546 INFO [optim.py:369] (1/2) 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 04:39:31,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5648, 1.7881, 1.3000, 1.3570], device='cuda:1'), covar=tensor([0.1051, 0.0537, 0.1021, 0.1112], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0449, 0.0524, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 04:39:45,154 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:40:02,255 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1313763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:40:07,469 INFO [train.py:968] (1/2) Epoch 29, batch 37950, libri_loss[loss=0.2921, simple_loss=0.3652, pruned_loss=0.1095, over 25832.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3561, pruned_loss=0.1058, over 5669064.74 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3554, pruned_loss=0.1104, over 5649168.05 frames. ], giga_tot_loss[loss=0.2808, simple_loss=0.3535, pruned_loss=0.104, over 5669941.40 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:40:37,495 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1313808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:40:39,391 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 38000, giga_loss[loss=0.2315, simple_loss=0.3199, pruned_loss=0.07154, over 28805.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3507, pruned_loss=0.1013, over 5679376.16 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3558, pruned_loss=0.1106, over 5653868.12 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3482, pruned_loss=0.09947, over 5676081.92 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:40:48,855 INFO [optim.py:369] (1/2) 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 04:41:01,763 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 38050, giga_loss[loss=0.2496, simple_loss=0.3319, pruned_loss=0.08367, over 28259.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3489, pruned_loss=0.09921, over 5678216.42 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3563, pruned_loss=0.111, over 5646873.02 frames. ], giga_tot_loss[loss=0.2706, simple_loss=0.3465, pruned_loss=0.09733, over 5681642.02 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:41:55,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1313902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:42:11,258 INFO [train.py:968] (1/2) Epoch 29, batch 38100, libri_loss[loss=0.2845, simple_loss=0.3421, pruned_loss=0.1135, over 28227.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3481, pruned_loss=0.09852, over 5676664.46 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3562, pruned_loss=0.111, over 5647194.63 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3462, pruned_loss=0.09695, over 5679368.75 frames. ], batch size: 62, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:42:14,716 INFO [optim.py:369] (1/2) 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 04:42:22,939 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4017, 4.2605, 3.9951, 2.1436], device='cuda:1'), covar=tensor([0.0545, 0.0654, 0.0675, 0.2223], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1202, 0.1007, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:42:50,934 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 38150, giga_loss[loss=0.2575, simple_loss=0.3331, pruned_loss=0.09099, over 28327.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3497, pruned_loss=0.0993, over 5681730.61 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3563, pruned_loss=0.111, over 5650742.62 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.348, pruned_loss=0.09791, over 5681134.03 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:43:20,709 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2130, 4.0572, 3.8086, 1.7868], device='cuda:1'), covar=tensor([0.0652, 0.0752, 0.0748, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.1205, 0.1011, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:43:36,990 INFO [train.py:968] (1/2) Epoch 29, batch 38200, giga_loss[loss=0.2732, simple_loss=0.3535, pruned_loss=0.0965, over 29052.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3511, pruned_loss=0.1003, over 5688404.02 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.356, pruned_loss=0.1107, over 5655129.53 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09928, over 5684555.43 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:43:39,343 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 04:44:00,879 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:44:02,900 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 38250, giga_loss[loss=0.2954, simple_loss=0.3659, pruned_loss=0.1124, over 29003.00 frames. ], tot_loss[loss=0.2785, simple_loss=0.3529, pruned_loss=0.1021, over 5692934.43 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3561, pruned_loss=0.1107, over 5656416.88 frames. ], giga_tot_loss[loss=0.2771, simple_loss=0.3519, pruned_loss=0.1011, over 5688895.68 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:44:29,526 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314077.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:44:52,858 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314108.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 04:44:52,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:44:55,847 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:03,865 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:04,939 INFO [train.py:968] (1/2) Epoch 29, batch 38300, giga_loss[loss=0.2454, simple_loss=0.3235, pruned_loss=0.08366, over 28306.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3542, pruned_loss=0.1036, over 5692424.39 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3561, pruned_loss=0.1106, over 5663131.70 frames. ], giga_tot_loss[loss=0.2794, simple_loss=0.3533, pruned_loss=0.1027, over 5684230.03 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:45:08,458 INFO [optim.py:369] (1/2) 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 04:45:19,622 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:20,745 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:45:45,580 INFO [train.py:968] (1/2) Epoch 29, batch 38350, libri_loss[loss=0.3925, simple_loss=0.4432, pruned_loss=0.1709, over 29079.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3547, pruned_loss=0.1041, over 5702054.76 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3561, pruned_loss=0.1107, over 5666565.45 frames. ], giga_tot_loss[loss=0.2802, simple_loss=0.3539, pruned_loss=0.1032, over 5692889.48 frames. ], batch size: 101, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:46:25,715 INFO [train.py:968] (1/2) Epoch 29, batch 38400, giga_loss[loss=0.2585, simple_loss=0.3467, pruned_loss=0.08515, over 28959.00 frames. ], tot_loss[loss=0.2805, simple_loss=0.3544, pruned_loss=0.1033, over 5705249.84 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3561, pruned_loss=0.1107, over 5669222.44 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3537, pruned_loss=0.1025, over 5696250.09 frames. ], batch size: 136, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:46:28,144 INFO [optim.py:369] (1/2) 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 04:46:36,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9455, 3.7866, 3.5583, 1.7849], device='cuda:1'), covar=tensor([0.0768, 0.0914, 0.0857, 0.2199], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1204, 0.1011, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:46:56,854 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:46:58,703 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 38450, giga_loss[loss=0.2605, simple_loss=0.3427, pruned_loss=0.08917, over 28346.00 frames. ], tot_loss[loss=0.2792, simple_loss=0.3541, pruned_loss=0.1021, over 5711328.72 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3568, pruned_loss=0.1111, over 5674795.96 frames. ], giga_tot_loss[loss=0.2773, simple_loss=0.3529, pruned_loss=0.1008, over 5700166.05 frames. ], batch size: 65, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:47:11,951 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:14,788 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:22,740 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:37,652 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:47:43,099 INFO [train.py:968] (1/2) Epoch 29, batch 38500, giga_loss[loss=0.2399, simple_loss=0.327, pruned_loss=0.07638, over 29004.00 frames. ], tot_loss[loss=0.2779, simple_loss=0.3538, pruned_loss=0.101, over 5712482.94 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3564, pruned_loss=0.1108, over 5679337.39 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3531, pruned_loss=0.1001, over 5700130.31 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:47:45,689 INFO [optim.py:369] (1/2) 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 04:48:23,442 INFO [train.py:968] (1/2) Epoch 29, batch 38550, libri_loss[loss=0.3388, simple_loss=0.3997, pruned_loss=0.1389, over 29256.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3513, pruned_loss=0.09959, over 5713231.89 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3563, pruned_loss=0.1106, over 5683549.12 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3507, pruned_loss=0.09877, over 5700356.82 frames. ], batch size: 97, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:48:47,461 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 38600, libri_loss[loss=0.3465, simple_loss=0.4055, pruned_loss=0.1437, over 29516.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.35, pruned_loss=0.09932, over 5705336.63 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.357, pruned_loss=0.111, over 5672327.15 frames. ], giga_tot_loss[loss=0.2724, simple_loss=0.3488, pruned_loss=0.098, over 5705628.41 frames. ], batch size: 82, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:49:05,696 INFO [optim.py:369] (1/2) 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 04:49:25,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5277, 1.2819, 3.9238, 3.3480], device='cuda:1'), covar=tensor([0.1578, 0.2924, 0.0472, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0672, 0.1006, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 04:49:33,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 04:49:41,146 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.03 vs. limit=5.0
+2023-03-15 04:49:44,680 INFO [train.py:968] (1/2) Epoch 29, batch 38650, giga_loss[loss=0.2777, simple_loss=0.3456, pruned_loss=0.1049, over 28931.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.3478, pruned_loss=0.09825, over 5713775.66 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3568, pruned_loss=0.1109, over 5673351.52 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.347, pruned_loss=0.09726, over 5713321.00 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:49:49,850 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3697, 1.6223, 1.7273, 1.4909], device='cuda:1'), covar=tensor([0.2442, 0.2211, 0.2525, 0.2233], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0760, 0.0736, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 04:49:54,508 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314483.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:50:18,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2671, 0.7447, 0.7965, 1.3683], device='cuda:1'), covar=tensor([0.0811, 0.0406, 0.0391, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 04:50:22,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3566, 3.4771, 1.5759, 1.5193], device='cuda:1'), covar=tensor([0.1093, 0.0321, 0.0920, 0.1440], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0569, 0.0411, 0.0448], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 04:50:26,000 INFO [train.py:968] (1/2) Epoch 29, batch 38700, giga_loss[loss=0.2762, simple_loss=0.3464, pruned_loss=0.1031, over 28868.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3477, pruned_loss=0.09879, over 5712212.01 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3568, pruned_loss=0.1109, over 5677727.34 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3468, pruned_loss=0.09777, over 5708567.44 frames. ], batch size: 99, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:50:28,507 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:50:28,880 INFO [optim.py:369] (1/2) 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 04:50:44,657 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:50:46,570 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:51:07,113 INFO [train.py:968] (1/2) Epoch 29, batch 38750, giga_loss[loss=0.3024, simple_loss=0.3679, pruned_loss=0.1185, over 27645.00 frames. ], tot_loss[loss=0.2727, simple_loss=0.3477, pruned_loss=0.0988, over 5714142.49 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3567, pruned_loss=0.1107, over 5677951.34 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3469, pruned_loss=0.09791, over 5711863.82 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:51:09,462 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:51:09,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5455, 1.5101, 1.7547, 1.4030], device='cuda:1'), covar=tensor([0.1585, 0.2361, 0.1343, 0.1750], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0722, 0.0995, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 04:51:16,441 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-15 04:51:31,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6840, 1.7425, 1.7282, 1.5514], device='cuda:1'), covar=tensor([0.2951, 0.3094, 0.2647, 0.2777], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2030, 0.1924, 0.2081], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 04:51:42,941 INFO [train.py:968] (1/2) Epoch 29, batch 38800, giga_loss[loss=0.271, simple_loss=0.3516, pruned_loss=0.09518, over 28765.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.348, pruned_loss=0.09857, over 5722304.37 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3566, pruned_loss=0.1105, over 5686372.23 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3472, pruned_loss=0.09755, over 5714218.26 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 04:51:47,166 INFO [optim.py:369] (1/2) 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 04:51:48,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1314626.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:51:50,082 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1314629.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:51:58,058 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:52:12,069 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1314658.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 04:52:22,165 INFO [train.py:968] (1/2) Epoch 29, batch 38850, giga_loss[loss=0.2381, simple_loss=0.3229, pruned_loss=0.07658, over 28803.00 frames. ], tot_loss[loss=0.2709, simple_loss=0.3471, pruned_loss=0.09732, over 5717040.01 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3564, pruned_loss=0.1104, over 5689903.18 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3465, pruned_loss=0.0965, over 5707807.61 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:52:28,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0395, 2.1696, 2.2043, 1.8346], device='cuda:1'), covar=tensor([0.2838, 0.2805, 0.2735, 0.2803], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2029, 0.1924, 0.2079], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 04:52:47,973 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 04:52:50,311 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:53:02,566 INFO [train.py:968] (1/2) Epoch 29, batch 38900, giga_loss[loss=0.3034, simple_loss=0.3671, pruned_loss=0.1198, over 28899.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.09651, over 5720177.78 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 5692920.43 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.345, pruned_loss=0.09561, over 5710576.41 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:53:07,652 INFO [optim.py:369] (1/2) 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 04:53:07,877 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8035, 3.6430, 3.5106, 1.8885], device='cuda:1'), covar=tensor([0.0790, 0.0935, 0.0951, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.1203, 0.1007, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 04:53:17,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:53:19,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3277, 1.2985, 1.3115, 1.5282], device='cuda:1'), covar=tensor([0.0814, 0.0374, 0.0354, 0.0928], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0122, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 04:53:43,408 INFO [train.py:968] (1/2) Epoch 29, batch 38950, giga_loss[loss=0.2494, simple_loss=0.3346, pruned_loss=0.08212, over 28831.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3444, pruned_loss=0.09598, over 5716597.58 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3565, pruned_loss=0.1104, over 5692802.31 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3434, pruned_loss=0.09488, over 5709693.79 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:53:49,093 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:54:24,206 INFO [train.py:968] (1/2) Epoch 29, batch 39000, libri_loss[loss=0.2927, simple_loss=0.3541, pruned_loss=0.1157, over 29549.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.342, pruned_loss=0.09533, over 5712143.97 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3565, pruned_loss=0.1103, over 5697283.71 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3408, pruned_loss=0.09393, over 5702924.39 frames. ], batch size: 78, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:54:24,207 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 04:54:29,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4043, 1.7210, 1.2406, 1.3527], device='cuda:1'), covar=tensor([0.1134, 0.0485, 0.1044, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0449, 0.0524, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 04:54:32,807 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1300, 1.2435, 3.5087, 3.1579], device='cuda:1'), covar=tensor([0.1716, 0.2806, 0.0489, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0672, 0.1005, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 04:54:34,599 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 04:54:38,666 INFO [optim.py:369] (1/2) 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 04:55:12,878 INFO [train.py:968] (1/2) Epoch 29, batch 39050, giga_loss[loss=0.2543, simple_loss=0.3258, pruned_loss=0.09138, over 28659.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3396, pruned_loss=0.09416, over 5713870.16 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3568, pruned_loss=0.1105, over 5698081.86 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3379, pruned_loss=0.09244, over 5706514.34 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:55:33,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1314899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:55:54,622 INFO [train.py:968] (1/2) Epoch 29, batch 39100, giga_loss[loss=0.284, simple_loss=0.3634, pruned_loss=0.1023, over 28600.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3405, pruned_loss=0.09487, over 5712685.41 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3573, pruned_loss=0.1108, over 5700337.46 frames. ], giga_tot_loss[loss=0.2623, simple_loss=0.3385, pruned_loss=0.09307, over 5705083.37 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:55:59,876 INFO [optim.py:369] (1/2) 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 04:56:08,443 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3074, 1.6492, 1.3291, 0.9998], device='cuda:1'), covar=tensor([0.3069, 0.2885, 0.3414, 0.2599], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1165, 0.1429, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 04:56:34,039 INFO [train.py:968] (1/2) Epoch 29, batch 39150, giga_loss[loss=0.3483, simple_loss=0.3979, pruned_loss=0.1493, over 23600.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3394, pruned_loss=0.09478, over 5704284.12 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1111, over 5701679.20 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.337, pruned_loss=0.09275, over 5697052.00 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 04:56:40,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5432, 3.8304, 1.6129, 1.6668], device='cuda:1'), covar=tensor([0.0902, 0.0475, 0.0843, 0.1283], device='cuda:1'), in_proj_covar=tensor([0.0426, 0.0567, 0.0410, 0.0446], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 04:56:48,956 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1314992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:04,921 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6331, 1.8862, 1.5951, 1.7960], device='cuda:1'), covar=tensor([0.2456, 0.2511, 0.2636, 0.2473], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1165, 0.1429, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 04:57:05,372 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:11,172 INFO [train.py:968] (1/2) Epoch 29, batch 39200, libri_loss[loss=0.2946, simple_loss=0.3609, pruned_loss=0.1141, over 29564.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3375, pruned_loss=0.09409, over 5703205.07 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3582, pruned_loss=0.1113, over 5694928.73 frames. ], giga_tot_loss[loss=0.2591, simple_loss=0.3346, pruned_loss=0.09176, over 5703650.75 frames. ], batch size: 79, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:57:14,772 INFO [optim.py:369] (1/2) 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 04:57:22,610 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:26,658 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:28,461 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:47,524 INFO [train.py:968] (1/2) Epoch 29, batch 39250, giga_loss[loss=0.2455, simple_loss=0.3323, pruned_loss=0.07931, over 28953.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.3351, pruned_loss=0.09302, over 5704696.06 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.358, pruned_loss=0.1112, over 5689482.29 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3326, pruned_loss=0.09096, over 5710052.96 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:57:50,055 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:57:54,751 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:58:20,216 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 39300, giga_loss[loss=0.2529, simple_loss=0.3397, pruned_loss=0.08303, over 29041.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3343, pruned_loss=0.09288, over 5702123.37 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3581, pruned_loss=0.1111, over 5696392.27 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3314, pruned_loss=0.09062, over 5700613.33 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:58:30,663 INFO [optim.py:369] (1/2) 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 04:58:52,839 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:58:55,447 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315157.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:58:57,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:59:02,792 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 39350, giga_loss[loss=0.2423, simple_loss=0.3132, pruned_loss=0.08571, over 28576.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.3339, pruned_loss=0.09228, over 5704110.30 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3585, pruned_loss=0.1112, over 5694048.78 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3302, pruned_loss=0.08967, over 5705457.24 frames. ], batch size: 71, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:59:18,511 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:59:42,160 INFO [train.py:968] (1/2) Epoch 29, batch 39400, giga_loss[loss=0.2378, simple_loss=0.3271, pruned_loss=0.07429, over 28844.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3369, pruned_loss=0.09368, over 5699057.86 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3589, pruned_loss=0.1116, over 5693390.66 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3326, pruned_loss=0.09044, over 5700740.51 frames. ], batch size: 227, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 04:59:44,293 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 04:59:47,058 INFO [optim.py:369] (1/2) 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 04:59:47,375 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0913, 2.4945, 1.7374, 2.1130], device='cuda:1'), covar=tensor([0.0949, 0.0539, 0.0904, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0449, 0.0523, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 04:59:47,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:09,482 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:09,526 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:12,158 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:23,551 INFO [train.py:968] (1/2) Epoch 29, batch 39450, giga_loss[loss=0.2537, simple_loss=0.3392, pruned_loss=0.08414, over 28867.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3396, pruned_loss=0.09513, over 5681774.52 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3586, pruned_loss=0.1115, over 5682555.17 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3359, pruned_loss=0.09217, over 5693505.77 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:00:39,193 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:45,357 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:00:48,236 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:05,060 INFO [train.py:968] (1/2) Epoch 29, batch 39500, libri_loss[loss=0.3341, simple_loss=0.3875, pruned_loss=0.1403, over 29185.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.341, pruned_loss=0.09526, over 5686381.48 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3583, pruned_loss=0.1114, over 5687230.19 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3378, pruned_loss=0.09249, over 5691555.29 frames. ], batch size: 101, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:01:10,747 INFO [optim.py:369] (1/2) 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 05:01:10,955 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:28,641 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0855, 3.9293, 3.7399, 1.7188], device='cuda:1'), covar=tensor([0.0638, 0.0768, 0.0703, 0.2249], device='cuda:1'), in_proj_covar=tensor([0.1301, 0.1206, 0.1011, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 05:01:40,473 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:41,613 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:01:43,007 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4816, 1.7506, 1.2350, 1.2599], device='cuda:1'), covar=tensor([0.1080, 0.0609, 0.1105, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0448, 0.0523, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 05:01:44,020 INFO [train.py:968] (1/2) Epoch 29, batch 39550, giga_loss[loss=0.2505, simple_loss=0.3258, pruned_loss=0.08756, over 28011.00 frames. ], tot_loss[loss=0.2656, simple_loss=0.3416, pruned_loss=0.09484, over 5686324.88 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3579, pruned_loss=0.1111, over 5685656.51 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3387, pruned_loss=0.09229, over 5691380.96 frames. ], batch size: 77, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:01:58,722 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2936, 1.3557, 1.2339, 1.5075], device='cuda:1'), covar=tensor([0.0761, 0.0335, 0.0349, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 05:02:20,061 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 39600, giga_loss[loss=0.2648, simple_loss=0.3464, pruned_loss=0.09162, over 28026.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.34, pruned_loss=0.09323, over 5694095.48 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3578, pruned_loss=0.111, over 5687283.21 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3375, pruned_loss=0.09106, over 5696508.46 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:02:32,372 INFO [optim.py:369] (1/2) 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 05:02:37,124 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3817, 3.2285, 3.1314, 1.5291], device='cuda:1'), covar=tensor([0.1021, 0.1127, 0.1085, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.1299, 0.1206, 0.1011, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 05:03:03,713 INFO [train.py:968] (1/2) Epoch 29, batch 39650, giga_loss[loss=0.2756, simple_loss=0.342, pruned_loss=0.1046, over 28907.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3398, pruned_loss=0.09361, over 5695864.06 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3583, pruned_loss=0.1114, over 5688487.28 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.3371, pruned_loss=0.09123, over 5696642.19 frames. ], batch size: 186, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:03:37,107 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-15 05:03:39,390 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315510.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:03:41,525 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 39700, giga_loss[loss=0.2504, simple_loss=0.3255, pruned_loss=0.08764, over 28526.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3398, pruned_loss=0.09372, over 5708920.83 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1117, over 5689980.01 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3371, pruned_loss=0.09144, over 5708369.58 frames. ], batch size: 85, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:03:52,701 INFO [optim.py:369] (1/2) 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 05:04:03,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3706, 1.5088, 1.4768, 1.3491], device='cuda:1'), covar=tensor([0.3363, 0.2744, 0.2743, 0.2974], device='cuda:1'), in_proj_covar=tensor([0.2060, 0.2028, 0.1922, 0.2076], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 05:04:07,186 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:07,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:16,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5913, 4.4412, 4.2050, 1.9861], device='cuda:1'), covar=tensor([0.0589, 0.0756, 0.0685, 0.1989], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.1203, 0.1009, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 05:04:17,880 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:19,738 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:31,830 INFO [train.py:968] (1/2) Epoch 29, batch 39750, giga_loss[loss=0.3079, simple_loss=0.3711, pruned_loss=0.1224, over 29075.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3427, pruned_loss=0.09537, over 5712555.54 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1115, over 5692257.06 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3406, pruned_loss=0.09362, over 5710309.06 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:04:44,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:04:58,111 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:05:11,228 INFO [train.py:968] (1/2) Epoch 29, batch 39800, giga_loss[loss=0.2564, simple_loss=0.3376, pruned_loss=0.08765, over 28773.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3473, pruned_loss=0.09809, over 5701873.91 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3587, pruned_loss=0.1116, over 5688487.23 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3452, pruned_loss=0.09618, over 5703107.91 frames. ], batch size: 119, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:05:15,727 INFO [optim.py:369] (1/2) 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 05:05:49,947 INFO [train.py:968] (1/2) Epoch 29, batch 39850, giga_loss[loss=0.263, simple_loss=0.3477, pruned_loss=0.08915, over 28909.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3483, pruned_loss=0.09821, over 5712861.59 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3588, pruned_loss=0.1116, over 5694941.95 frames. ], giga_tot_loss[loss=0.2695, simple_loss=0.3462, pruned_loss=0.09635, over 5708384.82 frames. ], batch size: 174, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:05:59,240 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:06:01,047 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:06:03,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:06:26,349 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 39900, giga_loss[loss=0.2823, simple_loss=0.3471, pruned_loss=0.1087, over 23657.00 frames. ], tot_loss[loss=0.2749, simple_loss=0.3504, pruned_loss=0.09969, over 5701430.31 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3593, pruned_loss=0.112, over 5689252.77 frames. ], giga_tot_loss[loss=0.2716, simple_loss=0.348, pruned_loss=0.09756, over 5703247.30 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:06:35,267 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:07:09,738 INFO [train.py:968] (1/2) Epoch 29, batch 39950, giga_loss[loss=0.353, simple_loss=0.4086, pruned_loss=0.1487, over 27580.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3521, pruned_loss=0.1006, over 5696453.63 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3594, pruned_loss=0.1122, over 5684078.82 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3499, pruned_loss=0.09842, over 5702817.07 frames. ], batch size: 472, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:07:25,883 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:07:31,823 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:07:44,386 INFO [train.py:968] (1/2) Epoch 29, batch 40000, giga_loss[loss=0.2653, simple_loss=0.3452, pruned_loss=0.09268, over 28922.00 frames. ], tot_loss[loss=0.2784, simple_loss=0.353, pruned_loss=0.1019, over 5692693.07 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.1129, over 5676619.25 frames. ], giga_tot_loss[loss=0.2742, simple_loss=0.3503, pruned_loss=0.09906, over 5705374.90 frames. ], batch size: 213, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:07:51,020 INFO [optim.py:369] (1/2) 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 05:08:08,492 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1315851.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:08:23,764 INFO [train.py:968] (1/2) Epoch 29, batch 40050, giga_loss[loss=0.2501, simple_loss=0.3265, pruned_loss=0.08682, over 28909.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3514, pruned_loss=0.1011, over 5690871.41 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3601, pruned_loss=0.1128, over 5666668.61 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3492, pruned_loss=0.09886, over 5709301.41 frames. ], batch size: 112, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:08:32,597 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1315883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:08:36,221 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1315886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:08:59,041 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1315915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:09:03,589 INFO [train.py:968] (1/2) Epoch 29, batch 40100, giga_loss[loss=0.234, simple_loss=0.3171, pruned_loss=0.07545, over 28594.00 frames. ], tot_loss[loss=0.2722, simple_loss=0.347, pruned_loss=0.09871, over 5701634.97 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1125, over 5675130.87 frames. ], giga_tot_loss[loss=0.2693, simple_loss=0.3451, pruned_loss=0.09672, over 5709676.80 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:09:09,545 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 40150, giga_loss[loss=0.2607, simple_loss=0.3406, pruned_loss=0.09037, over 28858.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.343, pruned_loss=0.09612, over 5696412.86 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3601, pruned_loss=0.1126, over 5667134.10 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3413, pruned_loss=0.09438, over 5711074.26 frames. ], batch size: 199, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:09:49,267 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1315978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:10:19,468 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-15 05:10:24,480 INFO [train.py:968] (1/2) Epoch 29, batch 40200, giga_loss[loss=0.2512, simple_loss=0.34, pruned_loss=0.08121, over 28969.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3436, pruned_loss=0.09547, over 5703932.18 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1126, over 5669641.38 frames. ], giga_tot_loss[loss=0.2646, simple_loss=0.3419, pruned_loss=0.09362, over 5714220.94 frames. ], batch size: 164, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:10:30,971 INFO [optim.py:369] (1/2) 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 05:10:45,672 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:10:54,930 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:11:00,009 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2092, 1.4713, 1.4597, 1.0922], device='cuda:1'), covar=tensor([0.1931, 0.2636, 0.1655, 0.1838], device='cuda:1'), in_proj_covar=tensor([0.0941, 0.0717, 0.0989, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 05:11:05,796 INFO [train.py:968] (1/2) Epoch 29, batch 40250, giga_loss[loss=0.279, simple_loss=0.3694, pruned_loss=0.09427, over 28345.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3445, pruned_loss=0.09452, over 5699256.32 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1126, over 5672025.10 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.343, pruned_loss=0.09293, over 5705572.19 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:11:31,800 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:11:46,026 INFO [train.py:968] (1/2) Epoch 29, batch 40300, giga_loss[loss=0.2924, simple_loss=0.3677, pruned_loss=0.1085, over 28702.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3441, pruned_loss=0.09492, over 5704332.89 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3596, pruned_loss=0.1124, over 5674461.25 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.3429, pruned_loss=0.09346, over 5707602.85 frames. ], batch size: 307, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:11:46,342 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:11:48,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:11:53,608 INFO [optim.py:369] (1/2) 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 05:12:11,957 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:14,149 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-15 05:12:22,982 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:24,266 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3604, 1.4282, 1.5413, 1.1881], device='cuda:1'), covar=tensor([0.1834, 0.2512, 0.1575, 0.1755], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0716, 0.0989, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 05:12:24,538 INFO [train.py:968] (1/2) Epoch 29, batch 40350, giga_loss[loss=0.2731, simple_loss=0.3497, pruned_loss=0.09826, over 28555.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3434, pruned_loss=0.09566, over 5706481.92 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3593, pruned_loss=0.1122, over 5679501.80 frames. ], giga_tot_loss[loss=0.2653, simple_loss=0.3423, pruned_loss=0.09417, over 5704989.36 frames. ], batch size: 336, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:12:29,014 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:48,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:50,308 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6248, 1.6251, 1.8310, 1.4243], device='cuda:1'), covar=tensor([0.1788, 0.2328, 0.1479, 0.1695], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0715, 0.0988, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 05:12:50,950 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:12:57,452 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4393, 1.4990, 3.1974, 3.1141], device='cuda:1'), covar=tensor([0.1242, 0.2448, 0.0455, 0.1108], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0674, 0.1011, 0.0986], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 05:13:05,511 INFO [train.py:968] (1/2) Epoch 29, batch 40400, giga_loss[loss=0.313, simple_loss=0.3872, pruned_loss=0.1194, over 28293.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3417, pruned_loss=0.09573, over 5705376.88 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3595, pruned_loss=0.1123, over 5682179.08 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3404, pruned_loss=0.09428, over 5702222.64 frames. ], batch size: 368, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:13:10,079 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316226.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:13:13,181 INFO [optim.py:369] (1/2) 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 05:13:16,002 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:13:30,879 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:13:44,943 INFO [train.py:968] (1/2) Epoch 29, batch 40450, giga_loss[loss=0.237, simple_loss=0.3142, pruned_loss=0.07995, over 29041.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3403, pruned_loss=0.0958, over 5707482.71 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.36, pruned_loss=0.1126, over 5678512.49 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3383, pruned_loss=0.09389, over 5709587.83 frames. ], batch size: 155, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:14:18,504 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:20,357 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:24,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:24,647 INFO [train.py:968] (1/2) Epoch 29, batch 40500, giga_loss[loss=0.2691, simple_loss=0.3309, pruned_loss=0.1036, over 28666.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3396, pruned_loss=0.09597, over 5705007.63 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.36, pruned_loss=0.1127, over 5671172.01 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3377, pruned_loss=0.09411, over 5714711.53 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:14:27,036 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316323.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:33,330 INFO [optim.py:369] (1/2) 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 05:14:44,282 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:14:50,850 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:15:02,682 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316369.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:15:05,256 INFO [train.py:968] (1/2) Epoch 29, batch 40550, giga_loss[loss=0.284, simple_loss=0.3415, pruned_loss=0.1132, over 23982.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3371, pruned_loss=0.09448, over 5710559.27 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1126, over 5675792.22 frames. ], giga_tot_loss[loss=0.2604, simple_loss=0.3352, pruned_loss=0.09277, over 5714770.32 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:15:06,183 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316372.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:15:13,168 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4974, 3.5886, 1.6190, 1.6176], device='cuda:1'), covar=tensor([0.0973, 0.0366, 0.0935, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0572, 0.0413, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 05:15:28,892 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1316401.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:15:46,033 INFO [train.py:968] (1/2) Epoch 29, batch 40600, giga_loss[loss=0.2291, simple_loss=0.31, pruned_loss=0.07406, over 28698.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3328, pruned_loss=0.09245, over 5712610.58 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.36, pruned_loss=0.1126, over 5676331.97 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3308, pruned_loss=0.09072, over 5716465.93 frames. ], batch size: 284, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:15:46,937 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:15:50,848 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4747, 4.2916, 1.6617, 1.8405], device='cuda:1'), covar=tensor([0.1027, 0.0312, 0.0959, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0427, 0.0571, 0.0413, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 05:15:51,740 INFO [optim.py:369] (1/2) 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 05:16:23,232 INFO [train.py:968] (1/2) Epoch 29, batch 40650, giga_loss[loss=0.2358, simple_loss=0.3162, pruned_loss=0.07764, over 28692.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3288, pruned_loss=0.09035, over 5717893.85 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3598, pruned_loss=0.1125, over 5679842.16 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.327, pruned_loss=0.08879, over 5718251.54 frames. ], batch size: 242, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:16:30,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:16:41,303 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3667, 1.5062, 1.3887, 1.5219], device='cuda:1'), covar=tensor([0.0767, 0.0342, 0.0345, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 05:16:44,622 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316499.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:17:01,350 INFO [train.py:968] (1/2) Epoch 29, batch 40700, giga_loss[loss=0.2967, simple_loss=0.3546, pruned_loss=0.1194, over 24031.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.33, pruned_loss=0.09093, over 5713177.09 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.1121, over 5686486.48 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3282, pruned_loss=0.08941, over 5708479.87 frames. ], batch size: 705, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:17:08,449 INFO [optim.py:369] (1/2) 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 05:17:29,702 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8019, 1.9692, 1.4798, 1.6230], device='cuda:1'), covar=tensor([0.0852, 0.0485, 0.0894, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0413, 0.0449, 0.0524, 0.0462], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 05:17:34,202 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2572, 1.6285, 1.4248, 1.4497], device='cuda:1'), covar=tensor([0.0750, 0.0370, 0.0351, 0.0898], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 05:17:36,274 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:17:38,177 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:17:38,878 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:17:40,517 INFO [train.py:968] (1/2) Epoch 29, batch 40750, giga_loss[loss=0.259, simple_loss=0.341, pruned_loss=0.08852, over 28904.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3334, pruned_loss=0.0922, over 5720805.24 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5691094.94 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.3312, pruned_loss=0.09058, over 5713938.67 frames. ], batch size: 145, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:18:03,781 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 40800, libri_loss[loss=0.2647, simple_loss=0.3316, pruned_loss=0.09888, over 29638.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3369, pruned_loss=0.09376, over 5713818.26 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3591, pruned_loss=0.112, over 5692233.21 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3348, pruned_loss=0.09207, over 5707614.71 frames. ], batch size: 69, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:18:22,627 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1316621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:24,489 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:25,733 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:18:28,246 INFO [optim.py:369] (1/2) 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 05:18:46,005 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 40850, giga_loss[loss=0.2587, simple_loss=0.3371, pruned_loss=0.09013, over 28924.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3409, pruned_loss=0.09561, over 5713045.55 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.36, pruned_loss=0.1127, over 5687449.41 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3377, pruned_loss=0.09297, over 5713297.99 frames. ], batch size: 66, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:19:21,998 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1747, 2.2097, 1.9373, 2.0048], device='cuda:1'), covar=tensor([0.2154, 0.2722, 0.2515, 0.2482], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0769, 0.0741, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 05:19:30,330 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1316707.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:19:42,175 INFO [train.py:968] (1/2) Epoch 29, batch 40900, giga_loss[loss=0.2727, simple_loss=0.3439, pruned_loss=0.1007, over 28660.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3427, pruned_loss=0.09639, over 5716477.37 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3599, pruned_loss=0.1127, over 5691305.05 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.34, pruned_loss=0.09408, over 5713813.24 frames. ], batch size: 92, lr: 1.09e-03, grad_scale: 8.0
+2023-03-15 05:19:48,017 INFO [optim.py:369] (1/2) 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 05:20:18,990 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 40950, giga_loss[loss=0.2459, simple_loss=0.3235, pruned_loss=0.08417, over 28903.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3448, pruned_loss=0.09814, over 5707620.25 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3595, pruned_loss=0.1125, over 5690606.82 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.3422, pruned_loss=0.09573, over 5707270.13 frames. ], batch size: 106, lr: 1.09e-03, grad_scale: 4.0
+2023-03-15 05:20:20,809 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1316772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:20:25,495 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5737, 1.6114, 1.2319, 1.2335], device='cuda:1'), covar=tensor([0.0838, 0.0464, 0.0923, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0451, 0.0525, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 05:20:31,093 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4138, 1.7206, 1.5803, 1.5608], device='cuda:1'), covar=tensor([0.2205, 0.1997, 0.2593, 0.2092], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0769, 0.0741, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 05:20:52,053 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 41000, giga_loss[loss=0.3593, simple_loss=0.4217, pruned_loss=0.1484, over 27995.00 frames. ], tot_loss[loss=0.2795, simple_loss=0.3513, pruned_loss=0.1039, over 5687818.07 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3593, pruned_loss=0.1124, over 5691753.37 frames. ], giga_tot_loss[loss=0.2766, simple_loss=0.3493, pruned_loss=0.102, over 5686679.62 frames. ], batch size: 412, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:21:22,087 INFO [optim.py:369] (1/2) 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 05:21:24,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8045, 2.1058, 1.7420, 1.9124], device='cuda:1'), covar=tensor([0.2469, 0.2645, 0.2943, 0.2546], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1166, 0.1431, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 05:21:56,187 INFO [train.py:968] (1/2) Epoch 29, batch 41050, giga_loss[loss=0.3008, simple_loss=0.382, pruned_loss=0.1098, over 28695.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.357, pruned_loss=0.1082, over 5687293.91 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3595, pruned_loss=0.1126, over 5698418.56 frames. ], giga_tot_loss[loss=0.2838, simple_loss=0.355, pruned_loss=0.1063, over 5680271.79 frames. ], batch size: 262, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:22:00,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316874.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:22:19,085 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6384, 1.7019, 1.6326, 1.4711], device='cuda:1'), covar=tensor([0.2638, 0.2665, 0.2482, 0.2617], device='cuda:1'), in_proj_covar=tensor([0.2069, 0.2040, 0.1940, 0.2081], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 05:22:41,847 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 41100, giga_loss[loss=0.3082, simple_loss=0.38, pruned_loss=0.1182, over 28995.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3629, pruned_loss=0.1122, over 5685603.74 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3593, pruned_loss=0.1123, over 5698978.00 frames. ], giga_tot_loss[loss=0.2918, simple_loss=0.3617, pruned_loss=0.1109, over 5679326.75 frames. ], batch size: 128, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:22:51,947 INFO [optim.py:369] (1/2) 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 05:23:03,048 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1316942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:23:28,044 INFO [train.py:968] (1/2) Epoch 29, batch 41150, libri_loss[loss=0.2879, simple_loss=0.3614, pruned_loss=0.1072, over 29518.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3687, pruned_loss=0.1174, over 5675344.84 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.359, pruned_loss=0.1121, over 5701040.00 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.368, pruned_loss=0.1166, over 5668387.16 frames. ], batch size: 84, lr: 1.09e-03, grad_scale: 2.0
+2023-03-15 05:23:59,164 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:24:07,016 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317017.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:24:10,485 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317020.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:24:10,810 INFO [train.py:968] (1/2) Epoch 29, batch 41200, giga_loss[loss=0.3154, simple_loss=0.3847, pruned_loss=0.123, over 28976.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3733, pruned_loss=0.1212, over 5682796.10 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.359, pruned_loss=0.1123, over 5707499.18 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3731, pruned_loss=0.1207, over 5670482.33 frames. ], batch size: 136, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:24:20,683 INFO [optim.py:369] (1/2) 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 05:24:40,739 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:24:43,926 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:03,689 INFO [train.py:968] (1/2) Epoch 29, batch 41250, giga_loss[loss=0.2881, simple_loss=0.3652, pruned_loss=0.1055, over 28973.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.377, pruned_loss=0.1247, over 5668245.91 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3589, pruned_loss=0.1121, over 5710671.78 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3773, pruned_loss=0.1247, over 5655079.31 frames. ], batch size: 227, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:25:13,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317082.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:15,822 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:18,201 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:22,779 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6762, 1.9978, 1.9469, 1.5446], device='cuda:1'), covar=tensor([0.3068, 0.2702, 0.2976, 0.3077], device='cuda:1'), in_proj_covar=tensor([0.2063, 0.2035, 0.1933, 0.2078], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 05:25:47,685 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:25:51,828 INFO [train.py:968] (1/2) Epoch 29, batch 41300, giga_loss[loss=0.2947, simple_loss=0.3609, pruned_loss=0.1143, over 28944.00 frames. ], tot_loss[loss=0.3181, simple_loss=0.3798, pruned_loss=0.1282, over 5659843.86 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3589, pruned_loss=0.1123, over 5713688.98 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3807, pruned_loss=0.1284, over 5645235.86 frames. ], batch size: 213, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:25:57,808 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5981, 1.8204, 1.6444, 1.6289], device='cuda:1'), covar=tensor([0.2022, 0.2092, 0.2444, 0.2065], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0767, 0.0739, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 05:26:06,646 INFO [optim.py:369] (1/2) 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 05:26:43,458 INFO [train.py:968] (1/2) Epoch 29, batch 41350, giga_loss[loss=0.339, simple_loss=0.395, pruned_loss=0.1414, over 28737.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.3824, pruned_loss=0.1312, over 5642033.49 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3587, pruned_loss=0.1121, over 5716874.62 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3838, pruned_loss=0.1319, over 5626187.13 frames. ], batch size: 284, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:26:47,932 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:27:31,407 INFO [train.py:968] (1/2) Epoch 29, batch 41400, libri_loss[loss=0.3115, simple_loss=0.3655, pruned_loss=0.1288, over 19686.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.3871, pruned_loss=0.1363, over 5628851.85 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3584, pruned_loss=0.112, over 5713521.66 frames. ], giga_tot_loss[loss=0.3322, simple_loss=0.3892, pruned_loss=0.1376, over 5617717.47 frames. ], batch size: 187, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:27:35,124 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:27:37,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 05:28:05,461 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:28:16,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6261, 1.6202, 1.7802, 1.4216], device='cuda:1'), covar=tensor([0.1684, 0.2479, 0.1390, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.0934, 0.0715, 0.0984, 0.0884], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 05:28:20,971 INFO [train.py:968] (1/2) Epoch 29, batch 41450, giga_loss[loss=0.3183, simple_loss=0.3817, pruned_loss=0.1274, over 28860.00 frames. ], tot_loss[loss=0.3313, simple_loss=0.3886, pruned_loss=0.137, over 5632004.27 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3583, pruned_loss=0.112, over 5703664.29 frames. ], giga_tot_loss[loss=0.3345, simple_loss=0.3914, pruned_loss=0.1389, over 5629829.49 frames. ], batch size: 199, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:28:23,432 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5349, 1.8684, 1.4736, 1.7053], device='cuda:1'), covar=tensor([0.2448, 0.2520, 0.2880, 0.2213], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1167, 0.1433, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 05:28:46,471 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:29:12,712 INFO [train.py:968] (1/2) Epoch 29, batch 41500, giga_loss[loss=0.2888, simple_loss=0.3582, pruned_loss=0.1097, over 28850.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3873, pruned_loss=0.1366, over 5635880.17 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3581, pruned_loss=0.1118, over 5707613.59 frames. ], giga_tot_loss[loss=0.3338, simple_loss=0.3902, pruned_loss=0.1387, over 5629377.87 frames. ], batch size: 174, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:29:22,305 INFO [optim.py:369] (1/2) 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 05:29:57,447 INFO [train.py:968] (1/2) Epoch 29, batch 41550, giga_loss[loss=0.3435, simple_loss=0.4007, pruned_loss=0.1432, over 28710.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3868, pruned_loss=0.1369, over 5624752.55 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3589, pruned_loss=0.1124, over 5703449.35 frames. ], giga_tot_loss[loss=0.3333, simple_loss=0.3893, pruned_loss=0.1387, over 5621028.25 frames. ], batch size: 262, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:30:09,626 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:30:16,023 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4023, 1.5247, 1.4845, 1.3415], device='cuda:1'), covar=tensor([0.2352, 0.2349, 0.2121, 0.2357], device='cuda:1'), in_proj_covar=tensor([0.2070, 0.2043, 0.1938, 0.2087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 05:30:45,173 INFO [train.py:968] (1/2) Epoch 29, batch 41600, giga_loss[loss=0.3167, simple_loss=0.3808, pruned_loss=0.1263, over 28426.00 frames. ], tot_loss[loss=0.329, simple_loss=0.3863, pruned_loss=0.1359, over 5634613.21 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3591, pruned_loss=0.1124, over 5708294.18 frames. ], giga_tot_loss[loss=0.3327, simple_loss=0.3891, pruned_loss=0.1382, over 5624726.19 frames. ], batch size: 65, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:30:52,550 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1317427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:30:57,093 INFO [optim.py:369] (1/2) 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 05:31:01,545 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:31:04,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:31:33,416 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:31:35,367 INFO [train.py:968] (1/2) Epoch 29, batch 41650, giga_loss[loss=0.3893, simple_loss=0.4208, pruned_loss=0.1789, over 26566.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3863, pruned_loss=0.1358, over 5620232.09 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3585, pruned_loss=0.1122, over 5712354.74 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3898, pruned_loss=0.1385, over 5606385.17 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:32:20,109 INFO [train.py:968] (1/2) Epoch 29, batch 41700, libri_loss[loss=0.2248, simple_loss=0.2983, pruned_loss=0.07568, over 29349.00 frames. ], tot_loss[loss=0.3269, simple_loss=0.3849, pruned_loss=0.1344, over 5601806.66 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1118, over 5710751.52 frames. ], giga_tot_loss[loss=0.333, simple_loss=0.3897, pruned_loss=0.1381, over 5587721.50 frames. ], batch size: 71, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:32:21,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8880, 5.6975, 5.4387, 2.9481], device='cuda:1'), covar=tensor([0.0557, 0.0716, 0.0863, 0.1587], device='cuda:1'), in_proj_covar=tensor([0.1320, 0.1221, 0.1024, 0.0757], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 05:32:22,981 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:32:25,516 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:32:28,873 INFO [optim.py:369] (1/2) 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 05:32:50,249 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:33:05,631 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2408, 3.0289, 1.4525, 1.4645], device='cuda:1'), covar=tensor([0.1063, 0.0375, 0.0931, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0576, 0.0415, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 05:33:08,234 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:33:08,631 INFO [train.py:968] (1/2) Epoch 29, batch 41750, libri_loss[loss=0.3018, simple_loss=0.3671, pruned_loss=0.1182, over 29234.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3821, pruned_loss=0.1317, over 5618552.31 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.358, pruned_loss=0.1119, over 5715304.46 frames. ], giga_tot_loss[loss=0.3282, simple_loss=0.3865, pruned_loss=0.135, over 5600993.12 frames. ], batch size: 97, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:33:11,194 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1317573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:33:37,225 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 41800, giga_loss[loss=0.2731, simple_loss=0.3509, pruned_loss=0.09762, over 28685.00 frames. ], tot_loss[loss=0.3178, simple_loss=0.3796, pruned_loss=0.128, over 5634709.55 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3571, pruned_loss=0.1114, over 5717912.34 frames. ], giga_tot_loss[loss=0.3238, simple_loss=0.3844, pruned_loss=0.1316, over 5615836.84 frames. ], batch size: 242, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:34:06,369 INFO [optim.py:369] (1/2) 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 05:34:47,095 INFO [train.py:968] (1/2) Epoch 29, batch 41850, giga_loss[loss=0.2862, simple_loss=0.3593, pruned_loss=0.1065, over 29002.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3778, pruned_loss=0.1264, over 5633608.77 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1115, over 5716952.78 frames. ], giga_tot_loss[loss=0.3199, simple_loss=0.3815, pruned_loss=0.1291, over 5619367.79 frames. ], batch size: 155, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:34:53,484 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 05:34:57,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6000, 1.8571, 1.2577, 1.4184], device='cuda:1'), covar=tensor([0.1117, 0.0623, 0.1156, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0452, 0.0525, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 05:35:08,619 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1317694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:35:10,591 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 41900, giga_loss[loss=0.3101, simple_loss=0.3804, pruned_loss=0.1199, over 28922.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3744, pruned_loss=0.1234, over 5616248.31 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3566, pruned_loss=0.1111, over 5710494.03 frames. ], giga_tot_loss[loss=0.3156, simple_loss=0.3784, pruned_loss=0.1264, over 5608175.38 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:35:39,407 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1317726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:35:45,003 INFO [optim.py:369] (1/2) 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 05:36:21,755 INFO [train.py:968] (1/2) Epoch 29, batch 41950, giga_loss[loss=0.3237, simple_loss=0.3893, pruned_loss=0.1291, over 28947.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.372, pruned_loss=0.1215, over 5643270.47 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3559, pruned_loss=0.1107, over 5714362.30 frames. ], giga_tot_loss[loss=0.3125, simple_loss=0.3762, pruned_loss=0.1245, over 5631900.13 frames. ], batch size: 145, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:37:07,313 INFO [train.py:968] (1/2) Epoch 29, batch 42000, giga_loss[loss=0.2697, simple_loss=0.3416, pruned_loss=0.09894, over 28375.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3723, pruned_loss=0.122, over 5649482.51 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3558, pruned_loss=0.1105, over 5718390.61 frames. ], giga_tot_loss[loss=0.3126, simple_loss=0.376, pruned_loss=0.1246, over 5635823.98 frames. ], batch size: 65, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:37:07,313 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 05:37:16,895 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 05:37:29,444 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 42050, giga_loss[loss=0.375, simple_loss=0.413, pruned_loss=0.1686, over 28028.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3711, pruned_loss=0.1209, over 5639270.04 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3558, pruned_loss=0.1105, over 5710564.74 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3744, pruned_loss=0.1233, over 5634467.16 frames. ], batch size: 412, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:38:54,770 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317920.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:38:55,181 INFO [train.py:968] (1/2) Epoch 29, batch 42100, giga_loss[loss=0.2912, simple_loss=0.3676, pruned_loss=0.1074, over 28667.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3697, pruned_loss=0.1189, over 5641180.84 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3561, pruned_loss=0.1105, over 5712949.98 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3727, pruned_loss=0.1211, over 5632208.59 frames. ], batch size: 92, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:39:10,002 INFO [optim.py:369] (1/2) 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 05:39:19,579 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1448, 1.3301, 1.1711, 0.9950], device='cuda:1'), covar=tensor([0.1006, 0.0459, 0.0994, 0.0997], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0454, 0.0527, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 05:39:44,731 INFO [train.py:968] (1/2) Epoch 29, batch 42150, giga_loss[loss=0.2895, simple_loss=0.3695, pruned_loss=0.1048, over 28681.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3704, pruned_loss=0.1169, over 5642574.34 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3554, pruned_loss=0.1103, over 5708150.49 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3738, pruned_loss=0.1191, over 5637450.11 frames. ], batch size: 99, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:40:02,717 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1317987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:40:04,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3078, 1.4124, 3.8735, 3.3782], device='cuda:1'), covar=tensor([0.1690, 0.2669, 0.0515, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0686, 0.1029, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 05:40:33,770 INFO [train.py:968] (1/2) Epoch 29, batch 42200, giga_loss[loss=0.3134, simple_loss=0.3757, pruned_loss=0.1256, over 28893.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3718, pruned_loss=0.117, over 5655821.00 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3558, pruned_loss=0.1106, over 5709354.53 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3743, pruned_loss=0.1185, over 5650302.45 frames. ], batch size: 186, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:40:44,008 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1318032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:40:45,053 INFO [optim.py:369] (1/2) 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 05:41:12,029 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8803, 3.7373, 3.5713, 2.2163], device='cuda:1'), covar=tensor([0.0670, 0.0792, 0.0793, 0.1628], device='cuda:1'), in_proj_covar=tensor([0.1330, 0.1233, 0.1031, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 05:41:17,443 INFO [train.py:968] (1/2) Epoch 29, batch 42250, giga_loss[loss=0.2931, simple_loss=0.3766, pruned_loss=0.1048, over 28838.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3719, pruned_loss=0.1175, over 5654445.95 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3565, pruned_loss=0.1111, over 5707004.48 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3737, pruned_loss=0.1184, over 5650491.11 frames. ], batch size: 199, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:42:01,208 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2942, 1.7008, 1.4026, 1.4755], device='cuda:1'), covar=tensor([0.2235, 0.2075, 0.2324, 0.2122], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0768, 0.0740, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 05:42:03,030 INFO [train.py:968] (1/2) Epoch 29, batch 42300, giga_loss[loss=0.2867, simple_loss=0.3582, pruned_loss=0.1076, over 28935.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3719, pruned_loss=0.1183, over 5648295.06 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3568, pruned_loss=0.1115, over 5694878.46 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3732, pruned_loss=0.1188, over 5655327.67 frames. ], batch size: 174, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:42:14,005 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 42350, libri_loss[loss=0.2611, simple_loss=0.3409, pruned_loss=0.09061, over 29547.00 frames. ], tot_loss[loss=0.3023, simple_loss=0.3691, pruned_loss=0.1177, over 5661669.27 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3566, pruned_loss=0.1114, over 5701001.00 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3707, pruned_loss=0.1184, over 5660506.38 frames. ], batch size: 83, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:43:32,871 INFO [train.py:968] (1/2) Epoch 29, batch 42400, giga_loss[loss=0.2939, simple_loss=0.3632, pruned_loss=0.1123, over 28859.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3685, pruned_loss=0.1183, over 5660221.35 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3566, pruned_loss=0.1113, over 5704150.84 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.37, pruned_loss=0.119, over 5655844.43 frames. ], batch size: 99, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 05:43:46,353 INFO [optim.py:369] (1/2) 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 05:43:47,577 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8658, 2.9815, 1.9058, 0.9776], device='cuda:1'), covar=tensor([0.9960, 0.3863, 0.4663, 0.8883], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1754, 0.1673, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 05:44:20,127 INFO [train.py:968] (1/2) Epoch 29, batch 42450, giga_loss[loss=0.2498, simple_loss=0.3446, pruned_loss=0.07751, over 29119.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3679, pruned_loss=0.1163, over 5669744.48 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3566, pruned_loss=0.1112, over 5707359.51 frames. ], giga_tot_loss[loss=0.3019, simple_loss=0.3694, pruned_loss=0.1172, over 5661962.25 frames. ], batch size: 155, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 05:44:39,823 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1318295.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:45:04,240 INFO [train.py:968] (1/2) Epoch 29, batch 42500, giga_loss[loss=0.3122, simple_loss=0.3773, pruned_loss=0.1236, over 28925.00 frames. ], tot_loss[loss=0.3007, simple_loss=0.3686, pruned_loss=0.1164, over 5680316.34 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3562, pruned_loss=0.1111, over 5711184.13 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3704, pruned_loss=0.1173, over 5670182.56 frames. ], batch size: 227, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:45:15,636 INFO [optim.py:369] (1/2) 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 05:45:45,176 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1318362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:45:53,513 INFO [train.py:968] (1/2) Epoch 29, batch 42550, giga_loss[loss=0.2566, simple_loss=0.3388, pruned_loss=0.08717, over 29080.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3688, pruned_loss=0.1165, over 5673990.29 frames. ], libri_tot_loss[loss=0.2895, simple_loss=0.3565, pruned_loss=0.1112, over 5713137.45 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3701, pruned_loss=0.1171, over 5664245.90 frames. ], batch size: 128, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:46:26,846 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 42600, giga_loss[loss=0.2735, simple_loss=0.3454, pruned_loss=0.1008, over 29011.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3667, pruned_loss=0.1153, over 5684798.33 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3563, pruned_loss=0.1111, over 5717210.00 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3681, pruned_loss=0.116, over 5673017.11 frames. ], batch size: 136, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:46:51,856 INFO [optim.py:369] (1/2) 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 05:46:55,479 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1318438.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 05:46:57,951 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1318441.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 05:47:21,664 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 42650, giga_loss[loss=0.2941, simple_loss=0.3543, pruned_loss=0.117, over 28964.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3661, pruned_loss=0.1156, over 5676335.22 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3566, pruned_loss=0.1111, over 5719899.98 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3673, pruned_loss=0.1162, over 5663218.85 frames. ], batch size: 106, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:47:53,915 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1318505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:47:55,990 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5245, 4.5076, 1.7128, 1.9009], device='cuda:1'), covar=tensor([0.1039, 0.0290, 0.0902, 0.1266], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0581, 0.0417, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 05:47:56,020 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1318508.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:07,362 INFO [train.py:968] (1/2) Epoch 29, batch 42700, giga_loss[loss=0.2868, simple_loss=0.3563, pruned_loss=0.1087, over 28908.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3644, pruned_loss=0.1153, over 5664610.88 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3568, pruned_loss=0.1112, over 5704478.24 frames. ], giga_tot_loss[loss=0.2985, simple_loss=0.3654, pruned_loss=0.1158, over 5666728.53 frames. ], batch size: 136, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:48:21,934 INFO [optim.py:369] (1/2) 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 05:48:22,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1318537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:36,437 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1318550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:38,401 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1318553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:48:55,097 INFO [train.py:968] (1/2) Epoch 29, batch 42750, giga_loss[loss=0.2666, simple_loss=0.3425, pruned_loss=0.09535, over 29061.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3638, pruned_loss=0.1152, over 5667294.58 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3569, pruned_loss=0.1114, over 5698950.41 frames. ], giga_tot_loss[loss=0.2979, simple_loss=0.3646, pruned_loss=0.1156, over 5673554.26 frames. ], batch size: 136, lr: 1.08e-03, grad_scale: 2.0
+2023-03-15 05:49:06,549 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1318582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:49:30,224 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.52 vs. limit=5.0
+2023-03-15 05:49:40,394 INFO [train.py:968] (1/2) Epoch 29, batch 42800, giga_loss[loss=0.3166, simple_loss=0.3729, pruned_loss=0.1301, over 28905.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3621, pruned_loss=0.1145, over 5667881.31 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3564, pruned_loss=0.1109, over 5701531.79 frames. ], giga_tot_loss[loss=0.2971, simple_loss=0.3634, pruned_loss=0.1154, over 5669706.68 frames. ], batch size: 199, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:49:52,724 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1318647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:50:12,047 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1318655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:50:24,260 INFO [train.py:968] (1/2) Epoch 29, batch 42850, giga_loss[loss=0.2918, simple_loss=0.3582, pruned_loss=0.1127, over 28934.00 frames. ], tot_loss[loss=0.2969, simple_loss=0.3626, pruned_loss=0.1156, over 5665105.50 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3565, pruned_loss=0.1108, over 5709779.59 frames. ], giga_tot_loss[loss=0.2983, simple_loss=0.3637, pruned_loss=0.1165, over 5657391.41 frames. ], batch size: 164, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:51:10,264 INFO [train.py:968] (1/2) Epoch 29, batch 42900, giga_loss[loss=0.2658, simple_loss=0.3424, pruned_loss=0.09459, over 28689.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3622, pruned_loss=0.115, over 5662065.94 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3563, pruned_loss=0.1108, over 5705693.10 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3634, pruned_loss=0.1158, over 5658614.69 frames. ], batch size: 119, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:51:25,838 INFO [optim.py:369] (1/2) 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 05:51:59,167 INFO [train.py:968] (1/2) Epoch 29, batch 42950, libri_loss[loss=0.2728, simple_loss=0.3438, pruned_loss=0.1009, over 29544.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3627, pruned_loss=0.1145, over 5651232.96 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3563, pruned_loss=0.1108, over 5689214.74 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3638, pruned_loss=0.1153, over 5663099.43 frames. ], batch size: 80, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:52:39,815 INFO [train.py:968] (1/2) Epoch 29, batch 43000, giga_loss[loss=0.2612, simple_loss=0.3428, pruned_loss=0.08978, over 28944.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3632, pruned_loss=0.1143, over 5663182.99 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3561, pruned_loss=0.1106, over 5691871.30 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3644, pruned_loss=0.1151, over 5669517.79 frames. ], batch size: 227, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:52:54,487 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 43050, giga_loss[loss=0.3034, simple_loss=0.3752, pruned_loss=0.1158, over 28619.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3629, pruned_loss=0.114, over 5665288.92 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3558, pruned_loss=0.1106, over 5688592.38 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3643, pruned_loss=0.1148, over 5673156.47 frames. ], batch size: 336, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:54:17,094 INFO [train.py:968] (1/2) Epoch 29, batch 43100, libri_loss[loss=0.2055, simple_loss=0.2864, pruned_loss=0.06226, over 29410.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3659, pruned_loss=0.1166, over 5676492.36 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3555, pruned_loss=0.1103, over 5691559.87 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3675, pruned_loss=0.1176, over 5679520.19 frames. ], batch size: 67, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:54:30,769 INFO [optim.py:369] (1/2) 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 05:54:41,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4384, 1.9910, 1.6501, 1.6102], device='cuda:1'), covar=tensor([0.0635, 0.0256, 0.0276, 0.0689], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 05:55:02,497 INFO [train.py:968] (1/2) Epoch 29, batch 43150, giga_loss[loss=0.2551, simple_loss=0.3294, pruned_loss=0.09038, over 28580.00 frames. ], tot_loss[loss=0.301, simple_loss=0.3662, pruned_loss=0.1179, over 5677387.40 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3556, pruned_loss=0.1105, over 5691686.41 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.3676, pruned_loss=0.1187, over 5679214.08 frames. ], batch size: 85, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:55:41,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6826, 4.4942, 1.8667, 1.9797], device='cuda:1'), covar=tensor([0.0970, 0.0403, 0.0846, 0.1176], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0580, 0.0416, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 05:55:46,670 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.65 vs. limit=5.0
+2023-03-15 05:55:54,414 INFO [train.py:968] (1/2) Epoch 29, batch 43200, giga_loss[loss=0.3164, simple_loss=0.3848, pruned_loss=0.124, over 28624.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.367, pruned_loss=0.1196, over 5664735.32 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3557, pruned_loss=0.1105, over 5685735.44 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3683, pruned_loss=0.1204, over 5671579.45 frames. ], batch size: 307, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 05:55:55,370 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1319022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:56:02,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1319030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:56:08,309 INFO [optim.py:369] (1/2) 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 05:56:24,754 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-15 05:56:39,969 INFO [train.py:968] (1/2) Epoch 29, batch 43250, giga_loss[loss=0.3282, simple_loss=0.3803, pruned_loss=0.138, over 28594.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3695, pruned_loss=0.1223, over 5648355.11 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3553, pruned_loss=0.1103, over 5683369.36 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3713, pruned_loss=0.1234, over 5655549.94 frames. ], batch size: 307, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:57:19,497 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 05:57:21,308 INFO [train.py:968] (1/2) Epoch 29, batch 43300, giga_loss[loss=0.2918, simple_loss=0.3596, pruned_loss=0.112, over 28879.00 frames. ], tot_loss[loss=0.3055, simple_loss=0.3684, pruned_loss=0.1214, over 5637935.05 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3563, pruned_loss=0.111, over 5668824.13 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3693, pruned_loss=0.122, over 5655581.58 frames. ], batch size: 199, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:57:34,369 INFO [optim.py:369] (1/2) 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 05:57:58,873 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1319165.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:01,650 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1319168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:03,241 INFO [train.py:968] (1/2) Epoch 29, batch 43350, giga_loss[loss=0.2765, simple_loss=0.3485, pruned_loss=0.1023, over 28731.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3677, pruned_loss=0.1207, over 5652528.70 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.356, pruned_loss=0.1108, over 5675812.03 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.369, pruned_loss=0.1216, over 5659645.05 frames. ], batch size: 119, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:58:06,793 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1319173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:08,515 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1319176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:25,277 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1319197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 05:58:34,789 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 43400, giga_loss[loss=0.3131, simple_loss=0.375, pruned_loss=0.1256, over 28920.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3683, pruned_loss=0.1194, over 5662376.76 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3564, pruned_loss=0.111, over 5680232.72 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.3693, pruned_loss=0.1201, over 5663627.14 frames. ], batch size: 227, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 05:59:00,609 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 43450, giga_loss[loss=0.2868, simple_loss=0.3563, pruned_loss=0.1087, over 28855.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3671, pruned_loss=0.1181, over 5662052.93 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3568, pruned_loss=0.1112, over 5687995.65 frames. ], giga_tot_loss[loss=0.3029, simple_loss=0.368, pruned_loss=0.1189, over 5654810.79 frames. ], batch size: 213, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:00:12,292 INFO [train.py:968] (1/2) Epoch 29, batch 43500, giga_loss[loss=0.2576, simple_loss=0.3338, pruned_loss=0.09067, over 28917.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3658, pruned_loss=0.1177, over 5680182.71 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3575, pruned_loss=0.1117, over 5695471.97 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3662, pruned_loss=0.1181, over 5667085.90 frames. ], batch size: 199, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:00:25,762 INFO [optim.py:369] (1/2) 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 06:00:53,650 INFO [train.py:968] (1/2) Epoch 29, batch 43550, giga_loss[loss=0.2622, simple_loss=0.3369, pruned_loss=0.09375, over 28588.00 frames. ], tot_loss[loss=0.2996, simple_loss=0.3642, pruned_loss=0.1176, over 5663845.19 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3573, pruned_loss=0.1117, over 5685043.74 frames. ], giga_tot_loss[loss=0.3007, simple_loss=0.365, pruned_loss=0.1181, over 5661360.98 frames. ], batch size: 307, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:00:58,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2248, 0.8897, 1.0083, 1.4245], device='cuda:1'), covar=tensor([0.0769, 0.0418, 0.0360, 0.0851], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 06:01:37,772 INFO [train.py:968] (1/2) Epoch 29, batch 43600, giga_loss[loss=0.2991, simple_loss=0.3691, pruned_loss=0.1146, over 28681.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3666, pruned_loss=0.1192, over 5672281.71 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.358, pruned_loss=0.1122, over 5687904.82 frames. ], giga_tot_loss[loss=0.3028, simple_loss=0.3668, pruned_loss=0.1194, over 5667433.23 frames. ], batch size: 242, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 06:01:54,708 INFO [optim.py:369] (1/2) 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 06:01:56,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.91 vs. limit=2.0
+2023-03-15 06:02:25,619 INFO [train.py:968] (1/2) Epoch 29, batch 43650, giga_loss[loss=0.3233, simple_loss=0.3919, pruned_loss=0.1274, over 28890.00 frames. ], tot_loss[loss=0.3082, simple_loss=0.372, pruned_loss=0.1222, over 5665346.78 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3577, pruned_loss=0.1121, over 5691327.44 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3726, pruned_loss=0.1226, over 5657988.13 frames. ], batch size: 174, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:03:16,323 INFO [train.py:968] (1/2) Epoch 29, batch 43700, libri_loss[loss=0.2508, simple_loss=0.3242, pruned_loss=0.08872, over 29555.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3735, pruned_loss=0.1202, over 5672737.28 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3575, pruned_loss=0.112, over 5692425.81 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3743, pruned_loss=0.1207, over 5665783.87 frames. ], batch size: 79, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:03:34,425 INFO [optim.py:369] (1/2) 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 06:04:06,104 INFO [train.py:968] (1/2) Epoch 29, batch 43750, giga_loss[loss=0.3581, simple_loss=0.417, pruned_loss=0.1496, over 28237.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3751, pruned_loss=0.1213, over 5670478.34 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.357, pruned_loss=0.1117, over 5695510.11 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3764, pruned_loss=0.122, over 5662018.55 frames. ], batch size: 368, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:04:41,459 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1319610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:04:51,301 INFO [train.py:968] (1/2) Epoch 29, batch 43800, giga_loss[loss=0.3066, simple_loss=0.3651, pruned_loss=0.124, over 28814.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3768, pruned_loss=0.123, over 5665612.96 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3571, pruned_loss=0.1116, over 5698242.80 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3782, pruned_loss=0.1239, over 5655781.80 frames. ], batch size: 99, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:05:08,619 INFO [optim.py:369] (1/2) 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 06:05:37,647 INFO [train.py:968] (1/2) Epoch 29, batch 43850, giga_loss[loss=0.4568, simple_loss=0.465, pruned_loss=0.2243, over 26616.00 frames. ], tot_loss[loss=0.3122, simple_loss=0.3769, pruned_loss=0.1238, over 5661906.22 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3567, pruned_loss=0.1113, over 5694071.17 frames. ], giga_tot_loss[loss=0.3146, simple_loss=0.3789, pruned_loss=0.1251, over 5656547.33 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:05:57,480 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3786, 1.6003, 1.3786, 1.5619], device='cuda:1'), covar=tensor([0.0753, 0.0353, 0.0336, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 06:06:20,306 INFO [train.py:968] (1/2) Epoch 29, batch 43900, giga_loss[loss=0.2938, simple_loss=0.3678, pruned_loss=0.1099, over 29003.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3762, pruned_loss=0.1242, over 5650334.20 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3574, pruned_loss=0.1117, over 5686697.76 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3777, pruned_loss=0.1252, over 5652300.00 frames. ], batch size: 164, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:06:22,160 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1319722.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:06:37,712 INFO [optim.py:369] (1/2) 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 06:07:07,762 INFO [train.py:968] (1/2) Epoch 29, batch 43950, giga_loss[loss=0.2851, simple_loss=0.3637, pruned_loss=0.1032, over 28321.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3733, pruned_loss=0.123, over 5654221.84 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3571, pruned_loss=0.1115, over 5689033.43 frames. ], giga_tot_loss[loss=0.3115, simple_loss=0.3749, pruned_loss=0.1241, over 5653306.50 frames. ], batch size: 368, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:07:36,655 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 06:07:52,609 INFO [train.py:968] (1/2) Epoch 29, batch 44000, giga_loss[loss=0.2945, simple_loss=0.3583, pruned_loss=0.1153, over 28822.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.371, pruned_loss=0.1219, over 5667415.39 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3577, pruned_loss=0.112, over 5690594.75 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3721, pruned_loss=0.1226, over 5664514.17 frames. ], batch size: 112, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 06:08:00,125 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.94 vs. limit=5.0
+2023-03-15 06:08:08,492 INFO [optim.py:369] (1/2) 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 06:08:41,152 INFO [train.py:968] (1/2) Epoch 29, batch 44050, giga_loss[loss=0.2944, simple_loss=0.357, pruned_loss=0.1159, over 28613.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3708, pruned_loss=0.1222, over 5674081.18 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3575, pruned_loss=0.112, over 5694809.11 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.372, pruned_loss=0.1229, over 5667530.28 frames. ], batch size: 71, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:08:41,977 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1319871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:09:03,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0929, 1.3938, 2.7180, 2.6909], device='cuda:1'), covar=tensor([0.1322, 0.2187, 0.0577, 0.1250], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0684, 0.1029, 0.1003], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 06:09:32,898 INFO [train.py:968] (1/2) Epoch 29, batch 44100, libri_loss[loss=0.3378, simple_loss=0.3922, pruned_loss=0.1417, over 19898.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3722, pruned_loss=0.1236, over 5661311.23 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3577, pruned_loss=0.1121, over 5686233.06 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3731, pruned_loss=0.1241, over 5664637.77 frames. ], batch size: 188, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:09:46,180 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 29, batch 44150, giga_loss[loss=0.2809, simple_loss=0.3488, pruned_loss=0.1065, over 28962.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3702, pruned_loss=0.1226, over 5667678.40 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3581, pruned_loss=0.1123, over 5688630.69 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.371, pruned_loss=0.1231, over 5668008.28 frames. ], batch size: 106, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:10:29,162 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1319985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:11:01,904 INFO [train.py:968] (1/2) Epoch 29, batch 44200, giga_loss[loss=0.2826, simple_loss=0.3558, pruned_loss=0.1047, over 28918.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3683, pruned_loss=0.1211, over 5663299.60 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3579, pruned_loss=0.1121, over 5681129.62 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3693, pruned_loss=0.1219, over 5669642.92 frames. ], batch size: 227, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:11:05,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.18 vs. limit=5.0
+2023-03-15 06:11:14,938 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1990, 1.3528, 1.2244, 1.0132], device='cuda:1'), covar=tensor([0.1031, 0.0448, 0.1018, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0456, 0.0528, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:11:17,235 INFO [optim.py:369] (1/2) 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 06:11:27,436 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-15 06:11:41,966 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:11:50,804 INFO [train.py:968] (1/2) Epoch 29, batch 44250, libri_loss[loss=0.3045, simple_loss=0.3708, pruned_loss=0.1191, over 28561.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3695, pruned_loss=0.1213, over 5663265.43 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3575, pruned_loss=0.1119, over 5684714.44 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3708, pruned_loss=0.1223, over 5664837.35 frames. ], batch size: 106, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:12:14,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:19,968 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:31,839 INFO [train.py:968] (1/2) Epoch 29, batch 44300, libri_loss[loss=0.3022, simple_loss=0.3777, pruned_loss=0.1133, over 29183.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3709, pruned_loss=0.1217, over 5653632.02 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3578, pruned_loss=0.1121, over 5671125.90 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3721, pruned_loss=0.1227, over 5665806.38 frames. ], batch size: 94, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:12:32,083 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:38,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:42,035 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:12:42,063 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2524, 1.6961, 1.2367, 0.5694], device='cuda:1'), covar=tensor([0.4168, 0.2267, 0.2844, 0.6094], device='cuda:1'), in_proj_covar=tensor([0.1869, 0.1760, 0.1671, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 06:12:47,972 INFO [optim.py:369] (1/2) 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 06:13:10,490 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:13:19,421 INFO [train.py:968] (1/2) Epoch 29, batch 44350, libri_loss[loss=0.255, simple_loss=0.3197, pruned_loss=0.09519, over 28160.00 frames. ], tot_loss[loss=0.3069, simple_loss=0.3702, pruned_loss=0.1218, over 5649866.40 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3578, pruned_loss=0.1121, over 5664714.42 frames. ], giga_tot_loss[loss=0.3085, simple_loss=0.3715, pruned_loss=0.1228, over 5664619.94 frames. ], batch size: 62, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:13:41,357 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:13:41,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320196.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:14:00,984 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:05,765 INFO [train.py:968] (1/2) Epoch 29, batch 44400, giga_loss[loss=0.3561, simple_loss=0.3958, pruned_loss=0.1582, over 26746.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3719, pruned_loss=0.1212, over 5641313.61 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3582, pruned_loss=0.1125, over 5657354.66 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3727, pruned_loss=0.1217, over 5658742.37 frames. ], batch size: 555, lr: 1.08e-03, grad_scale: 8.0
+2023-03-15 06:14:22,249 INFO [optim.py:369] (1/2) 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 06:14:22,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:24,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320243.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:28,294 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 44450, giga_loss[loss=0.3241, simple_loss=0.3922, pruned_loss=0.1279, over 28880.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3729, pruned_loss=0.1188, over 5647058.97 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3587, pruned_loss=0.113, over 5640250.34 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3731, pruned_loss=0.1188, over 5677567.29 frames. ], batch size: 66, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:14:49,921 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:14:49,935 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 29, batch 44500, giga_loss[loss=0.2694, simple_loss=0.3547, pruned_loss=0.09208, over 28555.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3758, pruned_loss=0.1201, over 5607068.12 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3594, pruned_loss=0.1136, over 5589727.78 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3757, pruned_loss=0.1197, over 5678416.89 frames. ], batch size: 85, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:15:42,826 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6333, 1.8235, 1.3042, 1.4033], device='cuda:1'), covar=tensor([0.1039, 0.0613, 0.1055, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0455, 0.0528, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:15:49,822 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6325, 1.7822, 1.3250, 1.3593], device='cuda:1'), covar=tensor([0.1062, 0.0639, 0.1032, 0.1213], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0455, 0.0528, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:15:53,666 INFO [optim.py:369] (1/2) 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 06:16:24,936 INFO [train.py:968] (1/2) Epoch 29, batch 44550, giga_loss[loss=0.2805, simple_loss=0.3582, pruned_loss=0.1014, over 28409.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3789, pruned_loss=0.123, over 5605396.36 frames. ], libri_tot_loss[loss=0.2935, simple_loss=0.3596, pruned_loss=0.1137, over 5572472.79 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3787, pruned_loss=0.1226, over 5675680.36 frames. ], batch size: 71, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:16:26,214 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2688, 1.3538, 3.5859, 3.1763], device='cuda:1'), covar=tensor([0.1620, 0.2609, 0.0518, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0684, 0.1028, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 06:16:27,908 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5725, 1.6453, 1.2754, 1.2696], device='cuda:1'), covar=tensor([0.0887, 0.0490, 0.0911, 0.1226], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0456, 0.0528, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:16:43,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:16:45,777 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:17:13,489 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5895, 1.7989, 1.3262, 1.3471], device='cuda:1'), covar=tensor([0.1122, 0.0656, 0.1093, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0455, 0.0528, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:17:15,782 INFO [train.py:968] (1/2) Epoch 29, batch 44600, giga_loss[loss=0.3032, simple_loss=0.3736, pruned_loss=0.1164, over 28919.00 frames. ], tot_loss[loss=0.3166, simple_loss=0.3809, pruned_loss=0.1261, over 5575685.35 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.3599, pruned_loss=0.114, over 5547685.41 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.3807, pruned_loss=0.1256, over 5653658.76 frames. ], batch size: 227, lr: 1.08e-03, grad_scale: 4.0
+2023-03-15 06:17:15,955 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:17:17,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5071, 1.5679, 1.6993, 1.3218], device='cuda:1'), covar=tensor([0.1611, 0.2538, 0.1330, 0.1636], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0721, 0.0989, 0.0887], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 06:17:31,540 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:17:31,892 INFO [optim.py:369] (1/2) 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 06:17:58,622 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-15 06:18:39,466 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:18:56,037 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 50, giga_loss[loss=0.2774, simple_loss=0.3592, pruned_loss=0.09782, over 28453.00 frames. ], tot_loss[loss=0.286, simple_loss=0.3656, pruned_loss=0.1032, over 1263873.07 frames. ], libri_tot_loss[loss=0.2652, simple_loss=0.3439, pruned_loss=0.09324, over 141836.24 frames. ], giga_tot_loss[loss=0.2883, simple_loss=0.3679, pruned_loss=0.1043, over 1150694.68 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:19:35,065 INFO [optim.py:369] (1/2) 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 06:19:56,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6609, 3.6500, 1.6890, 1.7625], device='cuda:1'), covar=tensor([0.0982, 0.0271, 0.0909, 0.1323], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0577, 0.0414, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 06:20:00,647 INFO [train.py:968] (1/2) Epoch 30, batch 100, giga_loss[loss=0.2836, simple_loss=0.3615, pruned_loss=0.1028, over 28990.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3596, pruned_loss=0.1011, over 2230877.12 frames. ], libri_tot_loss[loss=0.2696, simple_loss=0.3499, pruned_loss=0.09463, over 273433.17 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3607, pruned_loss=0.1019, over 2057797.43 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:20:03,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:03,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320571.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:20:12,201 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:15,209 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:19,163 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:24,635 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.50 vs. limit=2.0
+2023-03-15 06:20:30,026 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1320603.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:39,023 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:43,264 INFO [train.py:968] (1/2) Epoch 30, batch 150, giga_loss[loss=0.2404, simple_loss=0.3196, pruned_loss=0.08063, over 28640.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3428, pruned_loss=0.0929, over 3000616.25 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3384, pruned_loss=0.08844, over 411566.24 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.344, pruned_loss=0.09375, over 2790548.87 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:20:46,567 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:48,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:58,959 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:20:59,398 INFO [optim.py:369] (1/2) 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 06:21:01,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:05,002 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:11,409 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:21,388 INFO [train.py:968] (1/2) Epoch 30, batch 200, giga_loss[loss=0.2203, simple_loss=0.2973, pruned_loss=0.07169, over 29033.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.3313, pruned_loss=0.08782, over 3605144.80 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3355, pruned_loss=0.08642, over 626711.78 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3318, pruned_loss=0.08855, over 3342486.08 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:21:23,540 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:56,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320713.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:21:57,049 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320714.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 06:21:59,819 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:00,525 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320717.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 06:22:00,876 INFO [train.py:968] (1/2) Epoch 30, batch 250, libri_loss[loss=0.3013, simple_loss=0.3821, pruned_loss=0.1102, over 29512.00 frames. ], tot_loss[loss=0.2448, simple_loss=0.3224, pruned_loss=0.08362, over 4068679.33 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3396, pruned_loss=0.08833, over 757238.98 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.321, pruned_loss=0.08345, over 3812955.24 frames. ], batch size: 81, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:22:15,062 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:16,920 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:22:23,533 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:24,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320746.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 06:22:41,208 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:22:42,049 INFO [train.py:968] (1/2) Epoch 30, batch 300, giga_loss[loss=0.2588, simple_loss=0.3162, pruned_loss=0.1007, over 28892.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.3135, pruned_loss=0.07991, over 4426760.04 frames. ], libri_tot_loss[loss=0.26, simple_loss=0.3415, pruned_loss=0.08923, over 808135.25 frames. ], giga_tot_loss[loss=0.2352, simple_loss=0.3116, pruned_loss=0.07939, over 4211503.81 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:23:01,266 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1320790.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:23:04,146 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1320793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:23:26,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5190, 2.1424, 1.6728, 0.7679], device='cuda:1'), covar=tensor([0.7866, 0.4076, 0.4922, 0.8452], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1749, 0.1660, 0.1506], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 06:23:27,718 INFO [train.py:968] (1/2) Epoch 30, batch 350, giga_loss[loss=0.2141, simple_loss=0.2934, pruned_loss=0.06738, over 28773.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.3067, pruned_loss=0.0769, over 4704919.04 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3413, pruned_loss=0.08922, over 884310.14 frames. ], giga_tot_loss[loss=0.2285, simple_loss=0.3045, pruned_loss=0.07623, over 4517143.78 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:23:30,698 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1320822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:23:41,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4426, 2.1154, 1.6871, 0.8061], device='cuda:1'), covar=tensor([0.6592, 0.3591, 0.4787, 0.7126], device='cuda:1'), in_proj_covar=tensor([0.1857, 0.1750, 0.1661, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 06:23:43,978 INFO [optim.py:369] (1/2) 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 06:23:52,922 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6162, 1.8779, 1.5266, 1.5111], device='cuda:1'), covar=tensor([0.2885, 0.2948, 0.3320, 0.2810], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1174, 0.1440, 0.1025], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0010], device='cuda:1')
+2023-03-15 06:24:06,262 INFO [train.py:968] (1/2) Epoch 30, batch 400, giga_loss[loss=0.1948, simple_loss=0.2759, pruned_loss=0.05687, over 28854.00 frames. ], tot_loss[loss=0.2255, simple_loss=0.3018, pruned_loss=0.07458, over 4926933.34 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3393, pruned_loss=0.08857, over 933322.23 frames. ], giga_tot_loss[loss=0.2239, simple_loss=0.2999, pruned_loss=0.07397, over 4770592.73 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:24:47,085 INFO [train.py:968] (1/2) Epoch 30, batch 450, giga_loss[loss=0.2478, simple_loss=0.3104, pruned_loss=0.09263, over 26623.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.3003, pruned_loss=0.07445, over 5087158.99 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3409, pruned_loss=0.08918, over 982199.41 frames. ], giga_tot_loss[loss=0.2228, simple_loss=0.2982, pruned_loss=0.07372, over 4956943.29 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:25:02,843 INFO [optim.py:369] (1/2) 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 06:25:23,821 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3361, 2.4383, 2.3574, 2.2966], device='cuda:1'), covar=tensor([0.2419, 0.2645, 0.2300, 0.2386], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0768, 0.0738, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 06:25:26,212 INFO [train.py:968] (1/2) Epoch 30, batch 500, libri_loss[loss=0.2623, simple_loss=0.3465, pruned_loss=0.08902, over 29576.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2979, pruned_loss=0.0731, over 5225578.42 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.339, pruned_loss=0.08837, over 1127889.52 frames. ], giga_tot_loss[loss=0.2198, simple_loss=0.2952, pruned_loss=0.07218, over 5101257.49 frames. ], batch size: 76, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:25:35,647 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1320978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:26:03,318 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.65 vs. limit=5.0
+2023-03-15 06:26:07,292 INFO [train.py:968] (1/2) Epoch 30, batch 550, giga_loss[loss=0.2238, simple_loss=0.3006, pruned_loss=0.07354, over 28285.00 frames. ], tot_loss[loss=0.2208, simple_loss=0.2966, pruned_loss=0.07251, over 5333227.23 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3386, pruned_loss=0.08782, over 1269565.47 frames. ], giga_tot_loss[loss=0.2181, simple_loss=0.2933, pruned_loss=0.07143, over 5216659.89 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:26:16,217 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9689, 2.2087, 2.2770, 1.8783], device='cuda:1'), covar=tensor([0.4014, 0.2802, 0.2662, 0.3414], device='cuda:1'), in_proj_covar=tensor([0.2075, 0.2040, 0.1943, 0.2094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 06:26:26,965 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 600, libri_loss[loss=0.2105, simple_loss=0.2932, pruned_loss=0.06396, over 29671.00 frames. ], tot_loss[loss=0.2199, simple_loss=0.296, pruned_loss=0.07192, over 5397271.22 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3373, pruned_loss=0.08731, over 1510520.91 frames. ], giga_tot_loss[loss=0.2159, simple_loss=0.2913, pruned_loss=0.07028, over 5299079.27 frames. ], batch size: 73, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:26:56,060 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0738, 2.4119, 2.1405, 2.1468], device='cuda:1'), covar=tensor([0.2331, 0.2319, 0.2363, 0.2298], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0765, 0.0736, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 06:27:27,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4368, 1.6682, 1.7020, 1.2869], device='cuda:1'), covar=tensor([0.2026, 0.2827, 0.1610, 0.1881], device='cuda:1'), in_proj_covar=tensor([0.0949, 0.0726, 0.1000, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 06:27:33,586 INFO [train.py:968] (1/2) Epoch 30, batch 650, giga_loss[loss=0.1962, simple_loss=0.2726, pruned_loss=0.05995, over 29022.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2939, pruned_loss=0.07113, over 5456638.60 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3381, pruned_loss=0.08779, over 1585098.39 frames. ], giga_tot_loss[loss=0.2141, simple_loss=0.2892, pruned_loss=0.06945, over 5381557.92 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:27:36,118 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1321121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:27:39,853 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 06:28:03,980 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1321153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:28:19,103 INFO [train.py:968] (1/2) Epoch 30, batch 700, giga_loss[loss=0.1985, simple_loss=0.2726, pruned_loss=0.06221, over 28936.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2921, pruned_loss=0.07009, over 5494252.02 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3387, pruned_loss=0.08811, over 1692290.38 frames. ], giga_tot_loss[loss=0.2116, simple_loss=0.2869, pruned_loss=0.06817, over 5442481.57 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:28:58,700 INFO [train.py:968] (1/2) Epoch 30, batch 750, giga_loss[loss=0.195, simple_loss=0.2725, pruned_loss=0.05872, over 28984.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2908, pruned_loss=0.06919, over 5532363.07 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3397, pruned_loss=0.08811, over 1832175.11 frames. ], giga_tot_loss[loss=0.2094, simple_loss=0.2848, pruned_loss=0.06704, over 5483353.82 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:29:18,549 INFO [optim.py:369] (1/2) 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 06:29:42,778 INFO [train.py:968] (1/2) Epoch 30, batch 800, giga_loss[loss=0.237, simple_loss=0.3154, pruned_loss=0.07925, over 29043.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2892, pruned_loss=0.06884, over 5570092.58 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3388, pruned_loss=0.08735, over 1911837.72 frames. ], giga_tot_loss[loss=0.2089, simple_loss=0.2838, pruned_loss=0.06701, over 5525723.21 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:30:30,555 INFO [train.py:968] (1/2) Epoch 30, batch 850, libri_loss[loss=0.2372, simple_loss=0.3183, pruned_loss=0.07808, over 29574.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2971, pruned_loss=0.073, over 5598670.79 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3377, pruned_loss=0.08661, over 2049999.98 frames. ], giga_tot_loss[loss=0.2171, simple_loss=0.2916, pruned_loss=0.07127, over 5552203.72 frames. ], batch size: 74, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:30:52,861 INFO [optim.py:369] (1/2) 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 06:31:18,806 INFO [train.py:968] (1/2) Epoch 30, batch 900, giga_loss[loss=0.2737, simple_loss=0.3523, pruned_loss=0.09754, over 28902.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.3094, pruned_loss=0.07915, over 5619863.22 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3365, pruned_loss=0.08618, over 2087104.09 frames. ], giga_tot_loss[loss=0.2305, simple_loss=0.3053, pruned_loss=0.07789, over 5581966.41 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:31:50,248 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6649, 4.4769, 4.2159, 1.8931], device='cuda:1'), covar=tensor([0.0565, 0.0763, 0.0793, 0.2196], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.1215, 0.1019, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 06:32:00,003 INFO [train.py:968] (1/2) Epoch 30, batch 950, giga_loss[loss=0.2769, simple_loss=0.3558, pruned_loss=0.09898, over 28601.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3208, pruned_loss=0.08472, over 5635811.70 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3365, pruned_loss=0.08577, over 2218891.46 frames. ], giga_tot_loss[loss=0.2422, simple_loss=0.3169, pruned_loss=0.08375, over 5596394.13 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:32:17,126 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 1000, giga_loss[loss=0.2903, simple_loss=0.3603, pruned_loss=0.1102, over 26602.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3285, pruned_loss=0.08785, over 5653830.24 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3362, pruned_loss=0.08552, over 2327501.55 frames. ], giga_tot_loss[loss=0.2498, simple_loss=0.3251, pruned_loss=0.0872, over 5615899.96 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:32:38,736 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1321468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:33:18,526 INFO [train.py:968] (1/2) Epoch 30, batch 1050, giga_loss[loss=0.2926, simple_loss=0.3796, pruned_loss=0.1028, over 28745.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3319, pruned_loss=0.08808, over 5665200.82 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3362, pruned_loss=0.08544, over 2395319.94 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3292, pruned_loss=0.08765, over 5633480.87 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:33:42,609 INFO [optim.py:369] (1/2) 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 06:33:50,977 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1321552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:34:03,257 INFO [train.py:968] (1/2) Epoch 30, batch 1100, giga_loss[loss=0.2323, simple_loss=0.3154, pruned_loss=0.07456, over 28849.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3338, pruned_loss=0.08832, over 5663573.99 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3364, pruned_loss=0.08537, over 2447716.99 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3316, pruned_loss=0.08805, over 5635091.64 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:34:43,394 INFO [train.py:968] (1/2) Epoch 30, batch 1150, giga_loss[loss=0.2352, simple_loss=0.3218, pruned_loss=0.07428, over 28886.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.336, pruned_loss=0.08964, over 5664691.13 frames. ], libri_tot_loss[loss=0.2532, simple_loss=0.3362, pruned_loss=0.08506, over 2563248.25 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3342, pruned_loss=0.08969, over 5637831.23 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:35:03,296 INFO [optim.py:369] (1/2) 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 06:35:28,409 INFO [train.py:968] (1/2) Epoch 30, batch 1200, giga_loss[loss=0.28, simple_loss=0.3632, pruned_loss=0.0984, over 28961.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.339, pruned_loss=0.09197, over 5673881.69 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3359, pruned_loss=0.08479, over 2596610.11 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3378, pruned_loss=0.09218, over 5650585.52 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:35:48,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3095, 1.3879, 3.3264, 3.0166], device='cuda:1'), covar=tensor([0.1490, 0.2700, 0.0486, 0.2013], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0676, 0.1016, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 06:36:02,047 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3835, 1.7363, 1.6566, 1.6491], device='cuda:1'), covar=tensor([0.2579, 0.2069, 0.2709, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0767, 0.0738, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 06:36:08,252 INFO [train.py:968] (1/2) Epoch 30, batch 1250, giga_loss[loss=0.2557, simple_loss=0.3319, pruned_loss=0.08977, over 28711.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3422, pruned_loss=0.09383, over 5683496.62 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.337, pruned_loss=0.08529, over 2694131.79 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3409, pruned_loss=0.09404, over 5659059.49 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:36:27,736 INFO [optim.py:369] (1/2) 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 06:36:42,117 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2879, 3.2026, 1.4304, 1.4694], device='cuda:1'), covar=tensor([0.1108, 0.0295, 0.0986, 0.1463], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0574, 0.0415, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 06:36:49,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.68 vs. limit=5.0
+2023-03-15 06:36:52,787 INFO [train.py:968] (1/2) Epoch 30, batch 1300, giga_loss[loss=0.3415, simple_loss=0.3943, pruned_loss=0.1444, over 26573.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.347, pruned_loss=0.09597, over 5687900.42 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3375, pruned_loss=0.08552, over 2725655.64 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3458, pruned_loss=0.09614, over 5667190.73 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:37:25,732 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-15 06:37:29,522 INFO [train.py:968] (1/2) Epoch 30, batch 1350, giga_loss[loss=0.2619, simple_loss=0.3472, pruned_loss=0.08832, over 28204.00 frames. ], tot_loss[loss=0.2701, simple_loss=0.348, pruned_loss=0.09607, over 5684905.78 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3375, pruned_loss=0.08569, over 2825283.78 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3474, pruned_loss=0.09646, over 5670247.05 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:37:36,251 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 06:37:49,155 INFO [optim.py:369] (1/2) 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 06:37:51,657 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 1400, giga_loss[loss=0.2634, simple_loss=0.3439, pruned_loss=0.09147, over 28580.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3496, pruned_loss=0.09622, over 5692492.10 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3382, pruned_loss=0.08616, over 2930091.68 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3491, pruned_loss=0.09666, over 5675109.80 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:38:50,831 INFO [train.py:968] (1/2) Epoch 30, batch 1450, libri_loss[loss=0.2759, simple_loss=0.3631, pruned_loss=0.09437, over 28620.00 frames. ], tot_loss[loss=0.2695, simple_loss=0.3493, pruned_loss=0.09482, over 5698321.57 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.338, pruned_loss=0.08599, over 3001292.40 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3492, pruned_loss=0.09544, over 5682121.35 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:38:57,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1321927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:39:05,772 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1321941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:39:28,455 INFO [train.py:968] (1/2) Epoch 30, batch 1500, giga_loss[loss=0.223, simple_loss=0.3148, pruned_loss=0.06558, over 28558.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3476, pruned_loss=0.09294, over 5701967.76 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3384, pruned_loss=0.08616, over 3029717.65 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3475, pruned_loss=0.09342, over 5687975.66 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:39:42,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1321986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:39:45,321 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1321989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:40:05,802 INFO [train.py:968] (1/2) Epoch 30, batch 1550, giga_loss[loss=0.273, simple_loss=0.3546, pruned_loss=0.09572, over 28585.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3445, pruned_loss=0.09019, over 5717045.70 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3372, pruned_loss=0.08539, over 3181606.23 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3454, pruned_loss=0.09125, over 5697885.57 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:40:06,180 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1322018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:40:16,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2386, 0.8398, 0.9028, 1.3966], device='cuda:1'), covar=tensor([0.0795, 0.0408, 0.0374, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 06:40:25,781 INFO [optim.py:369] (1/2) 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 06:40:50,187 INFO [train.py:968] (1/2) Epoch 30, batch 1600, giga_loss[loss=0.2479, simple_loss=0.3308, pruned_loss=0.08252, over 28472.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3473, pruned_loss=0.0936, over 5700343.85 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3374, pruned_loss=0.0854, over 3221540.61 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.348, pruned_loss=0.09451, over 5683438.16 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:40:51,614 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1322070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:40:53,688 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1322073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:41:19,564 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1322102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:41:31,121 INFO [train.py:968] (1/2) Epoch 30, batch 1650, giga_loss[loss=0.3146, simple_loss=0.3731, pruned_loss=0.128, over 28633.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3501, pruned_loss=0.09772, over 5706610.28 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3372, pruned_loss=0.08533, over 3248385.42 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3508, pruned_loss=0.09856, over 5691584.00 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:41:53,416 INFO [optim.py:369] (1/2) 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 06:42:08,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8524, 2.0164, 2.0674, 1.7050], device='cuda:1'), covar=tensor([0.2953, 0.2862, 0.2819, 0.3095], device='cuda:1'), in_proj_covar=tensor([0.2065, 0.2036, 0.1935, 0.2087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 06:42:16,708 INFO [train.py:968] (1/2) Epoch 30, batch 1700, giga_loss[loss=0.2681, simple_loss=0.3407, pruned_loss=0.0978, over 28930.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3508, pruned_loss=0.09975, over 5713247.91 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3373, pruned_loss=0.08537, over 3274661.02 frames. ], giga_tot_loss[loss=0.2762, simple_loss=0.3515, pruned_loss=0.1005, over 5699955.80 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:42:58,026 INFO [train.py:968] (1/2) Epoch 30, batch 1750, giga_loss[loss=0.2783, simple_loss=0.3506, pruned_loss=0.1031, over 28869.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3493, pruned_loss=0.09963, over 5701000.58 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.338, pruned_loss=0.08593, over 3393512.27 frames. ], giga_tot_loss[loss=0.2757, simple_loss=0.3501, pruned_loss=0.1006, over 5689228.82 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:43:09,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5431, 1.5935, 1.7039, 1.3708], device='cuda:1'), covar=tensor([0.1346, 0.2074, 0.1125, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.0948, 0.0725, 0.0998, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 06:43:16,154 INFO [optim.py:369] (1/2) 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 06:43:40,043 INFO [train.py:968] (1/2) Epoch 30, batch 1800, giga_loss[loss=0.2944, simple_loss=0.3676, pruned_loss=0.1106, over 28311.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3474, pruned_loss=0.09925, over 5695716.96 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3379, pruned_loss=0.08586, over 3431052.28 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3482, pruned_loss=0.1003, over 5683525.69 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:44:18,810 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 1850, giga_loss[loss=0.229, simple_loss=0.3187, pruned_loss=0.06969, over 28878.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3462, pruned_loss=0.09807, over 5692686.28 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3383, pruned_loss=0.086, over 3468148.51 frames. ], giga_tot_loss[loss=0.2723, simple_loss=0.3468, pruned_loss=0.09897, over 5680157.64 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:44:38,757 INFO [optim.py:369] (1/2) 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 06:45:00,826 INFO [train.py:968] (1/2) Epoch 30, batch 1900, giga_loss[loss=0.244, simple_loss=0.3246, pruned_loss=0.08172, over 28910.00 frames. ], tot_loss[loss=0.2676, simple_loss=0.3437, pruned_loss=0.09579, over 5700319.04 frames. ], libri_tot_loss[loss=0.2554, simple_loss=0.3385, pruned_loss=0.08609, over 3540207.94 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3442, pruned_loss=0.09676, over 5684489.97 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:45:44,514 INFO [train.py:968] (1/2) Epoch 30, batch 1950, giga_loss[loss=0.2341, simple_loss=0.3154, pruned_loss=0.07643, over 28870.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3405, pruned_loss=0.09393, over 5696274.28 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.339, pruned_loss=0.08641, over 3585997.55 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3407, pruned_loss=0.09467, over 5680830.67 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:46:03,593 INFO [optim.py:369] (1/2) 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 06:46:20,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1322459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:46:23,039 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 2000, giga_loss[loss=0.2183, simple_loss=0.3008, pruned_loss=0.06792, over 28662.00 frames. ], tot_loss[loss=0.2581, simple_loss=0.3345, pruned_loss=0.0908, over 5682375.90 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.339, pruned_loss=0.08623, over 3632343.14 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3346, pruned_loss=0.09161, over 5666372.96 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:46:31,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8123, 2.0929, 1.5139, 1.6977], device='cuda:1'), covar=tensor([0.1037, 0.0538, 0.0993, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0454, 0.0529, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:46:46,678 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1322491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:47:08,603 INFO [train.py:968] (1/2) Epoch 30, batch 2050, giga_loss[loss=0.2338, simple_loss=0.2917, pruned_loss=0.08791, over 23363.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3297, pruned_loss=0.0883, over 5681731.35 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3386, pruned_loss=0.08596, over 3696127.35 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3299, pruned_loss=0.08918, over 5665956.09 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:47:14,356 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.1596, 5.0264, 4.7512, 2.2086], device='cuda:1'), covar=tensor([0.0412, 0.0527, 0.0604, 0.1972], device='cuda:1'), in_proj_covar=tensor([0.1301, 0.1207, 0.1011, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 06:47:32,600 INFO [optim.py:369] (1/2) 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 06:47:55,263 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1322567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:47:55,612 INFO [train.py:968] (1/2) Epoch 30, batch 2100, giga_loss[loss=0.2658, simple_loss=0.3261, pruned_loss=0.1027, over 23568.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.3286, pruned_loss=0.08777, over 5666178.99 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3392, pruned_loss=0.08636, over 3739908.37 frames. ], giga_tot_loss[loss=0.2523, simple_loss=0.328, pruned_loss=0.08825, over 5649253.39 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:48:04,253 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2858, 1.8184, 1.6151, 1.4475], device='cuda:1'), covar=tensor([0.0848, 0.0309, 0.0307, 0.1115], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 06:48:16,693 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1322596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:48:35,200 INFO [train.py:968] (1/2) Epoch 30, batch 2150, giga_loss[loss=0.2472, simple_loss=0.3284, pruned_loss=0.08299, over 28875.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3312, pruned_loss=0.08868, over 5665524.51 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3397, pruned_loss=0.08667, over 3763668.59 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3303, pruned_loss=0.08888, over 5665434.92 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:48:37,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2572, 2.3768, 2.3413, 2.0667], device='cuda:1'), covar=tensor([0.3100, 0.2711, 0.2503, 0.3027], device='cuda:1'), in_proj_covar=tensor([0.2078, 0.2042, 0.1944, 0.2099], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 06:48:51,696 INFO [optim.py:369] (1/2) 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 06:49:11,639 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.34 vs. limit=5.0
+2023-03-15 06:49:13,392 INFO [train.py:968] (1/2) Epoch 30, batch 2200, giga_loss[loss=0.2577, simple_loss=0.3315, pruned_loss=0.09198, over 28039.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3303, pruned_loss=0.08808, over 5678305.74 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3404, pruned_loss=0.08684, over 3805960.56 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3291, pruned_loss=0.08816, over 5674349.40 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:49:52,779 INFO [train.py:968] (1/2) Epoch 30, batch 2250, giga_loss[loss=0.25, simple_loss=0.3308, pruned_loss=0.08458, over 27966.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.3285, pruned_loss=0.08702, over 5687618.54 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3405, pruned_loss=0.08664, over 3875783.24 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.327, pruned_loss=0.08723, over 5679915.49 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:50:06,169 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-15 06:50:11,945 INFO [optim.py:369] (1/2) 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 06:50:34,550 INFO [train.py:968] (1/2) Epoch 30, batch 2300, giga_loss[loss=0.2158, simple_loss=0.2994, pruned_loss=0.06614, over 28994.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.3259, pruned_loss=0.08576, over 5696275.03 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3404, pruned_loss=0.08651, over 3904569.38 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3246, pruned_loss=0.08599, over 5688740.58 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:50:36,682 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1322771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:50:39,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7285, 2.5480, 1.5676, 1.0011], device='cuda:1'), covar=tensor([0.8717, 0.4169, 0.4961, 0.7751], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1735, 0.1658, 0.1503], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 06:51:04,463 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7071, 5.0317, 1.9929, 2.1094], device='cuda:1'), covar=tensor([0.1026, 0.0262, 0.0844, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0573, 0.0416, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 06:51:11,205 INFO [train.py:968] (1/2) Epoch 30, batch 2350, giga_loss[loss=0.2408, simple_loss=0.3195, pruned_loss=0.08102, over 28869.00 frames. ], tot_loss[loss=0.2453, simple_loss=0.3226, pruned_loss=0.08402, over 5702854.05 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3401, pruned_loss=0.08619, over 3944170.00 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3215, pruned_loss=0.08437, over 5693378.14 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:51:12,369 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-15 06:51:31,887 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 2400, giga_loss[loss=0.2056, simple_loss=0.2921, pruned_loss=0.05957, over 29015.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3201, pruned_loss=0.08301, over 5703649.99 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3401, pruned_loss=0.08608, over 3973521.04 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3189, pruned_loss=0.08331, over 5693492.58 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:52:29,133 INFO [train.py:968] (1/2) Epoch 30, batch 2450, giga_loss[loss=0.2955, simple_loss=0.3566, pruned_loss=0.1172, over 26828.00 frames. ], tot_loss[loss=0.2423, simple_loss=0.3193, pruned_loss=0.08263, over 5712675.11 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3402, pruned_loss=0.08591, over 4030739.18 frames. ], giga_tot_loss[loss=0.2418, simple_loss=0.3178, pruned_loss=0.08289, over 5699356.69 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:52:46,975 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1322942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:52:47,385 INFO [optim.py:369] (1/2) 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 06:53:06,781 INFO [train.py:968] (1/2) Epoch 30, batch 2500, giga_loss[loss=0.2297, simple_loss=0.3035, pruned_loss=0.07791, over 28852.00 frames. ], tot_loss[loss=0.2412, simple_loss=0.318, pruned_loss=0.08217, over 5710432.58 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3406, pruned_loss=0.08592, over 4065890.52 frames. ], giga_tot_loss[loss=0.2403, simple_loss=0.316, pruned_loss=0.08229, over 5706482.20 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:53:09,512 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1322971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:53:46,485 INFO [train.py:968] (1/2) Epoch 30, batch 2550, giga_loss[loss=0.2556, simple_loss=0.3359, pruned_loss=0.0877, over 28828.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3166, pruned_loss=0.08144, over 5719077.78 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3404, pruned_loss=0.08566, over 4100796.97 frames. ], giga_tot_loss[loss=0.239, simple_loss=0.3147, pruned_loss=0.08164, over 5714172.28 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:53:51,488 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 06:53:59,685 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4369, 1.5918, 1.2983, 1.0713], device='cuda:1'), covar=tensor([0.1161, 0.0594, 0.1081, 0.1252], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0455, 0.0531, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:54:05,251 INFO [optim.py:369] (1/2) 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 06:54:10,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3781, 1.1996, 3.9458, 3.4206], device='cuda:1'), covar=tensor([0.1693, 0.3068, 0.0429, 0.1564], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0675, 0.1011, 0.0989], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 06:54:25,143 INFO [train.py:968] (1/2) Epoch 30, batch 2600, giga_loss[loss=0.2303, simple_loss=0.3108, pruned_loss=0.07489, over 28749.00 frames. ], tot_loss[loss=0.2377, simple_loss=0.3145, pruned_loss=0.08047, over 5718901.03 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3405, pruned_loss=0.08567, over 4109950.99 frames. ], giga_tot_loss[loss=0.2371, simple_loss=0.3129, pruned_loss=0.08061, over 5714131.44 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:54:25,302 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1323068.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 06:54:29,968 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7014, 1.9721, 1.9093, 1.6633], device='cuda:1'), covar=tensor([0.2937, 0.2316, 0.1787, 0.2311], device='cuda:1'), in_proj_covar=tensor([0.2073, 0.2033, 0.1940, 0.2095], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 06:54:36,786 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:54:39,572 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:54:44,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7233, 1.8038, 1.7531, 1.5793], device='cuda:1'), covar=tensor([0.3199, 0.2914, 0.2288, 0.2831], device='cuda:1'), in_proj_covar=tensor([0.2071, 0.2031, 0.1938, 0.2092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 06:54:58,515 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.75 vs. limit=5.0
+2023-03-15 06:54:58,905 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323114.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:00,878 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:00,905 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 2650, giga_loss[loss=0.2584, simple_loss=0.3279, pruned_loss=0.09448, over 29088.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.3155, pruned_loss=0.0811, over 5724753.24 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3409, pruned_loss=0.08569, over 4163039.54 frames. ], giga_tot_loss[loss=0.2377, simple_loss=0.3133, pruned_loss=0.08107, over 5716101.77 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:55:07,754 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5795, 3.4997, 1.6954, 1.7815], device='cuda:1'), covar=tensor([0.1013, 0.0308, 0.0888, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0572, 0.0415, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 06:55:22,865 INFO [optim.py:369] (1/2) 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 06:55:26,464 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:26,521 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1323146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:55:44,598 INFO [train.py:968] (1/2) Epoch 30, batch 2700, giga_loss[loss=0.2472, simple_loss=0.3285, pruned_loss=0.0829, over 28538.00 frames. ], tot_loss[loss=0.2437, simple_loss=0.3197, pruned_loss=0.08391, over 5721421.42 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3409, pruned_loss=0.08552, over 4187627.39 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3176, pruned_loss=0.08396, over 5713492.01 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:56:26,074 INFO [train.py:968] (1/2) Epoch 30, batch 2750, giga_loss[loss=0.259, simple_loss=0.3447, pruned_loss=0.08665, over 28859.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3247, pruned_loss=0.08674, over 5707589.86 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.341, pruned_loss=0.08548, over 4224566.79 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3225, pruned_loss=0.08679, over 5708843.82 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:56:48,292 INFO [optim.py:369] (1/2) 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 06:57:00,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1621, 1.2843, 1.0800, 0.8711], device='cuda:1'), covar=tensor([0.1000, 0.0447, 0.1019, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0453, 0.0528, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 06:57:09,009 INFO [train.py:968] (1/2) Epoch 30, batch 2800, giga_loss[loss=0.2486, simple_loss=0.3253, pruned_loss=0.08592, over 28814.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.333, pruned_loss=0.09228, over 5703611.76 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3407, pruned_loss=0.08535, over 4249662.85 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3312, pruned_loss=0.09245, over 5701979.05 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 06:57:29,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:57:31,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:57:36,448 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1323300.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:57:51,153 INFO [train.py:968] (1/2) Epoch 30, batch 2850, giga_loss[loss=0.2416, simple_loss=0.3229, pruned_loss=0.08015, over 28976.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3376, pruned_loss=0.09445, over 5703418.27 frames. ], libri_tot_loss[loss=0.2558, simple_loss=0.3407, pruned_loss=0.08545, over 4297390.62 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.336, pruned_loss=0.09475, over 5697869.11 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:57:54,932 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 06:58:13,798 INFO [optim.py:369] (1/2) 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 06:58:34,816 INFO [train.py:968] (1/2) Epoch 30, batch 2900, giga_loss[loss=0.2858, simple_loss=0.3487, pruned_loss=0.1114, over 23633.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3413, pruned_loss=0.095, over 5713501.95 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3405, pruned_loss=0.08527, over 4383060.75 frames. ], giga_tot_loss[loss=0.2659, simple_loss=0.3402, pruned_loss=0.09584, over 5699729.52 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:59:14,111 INFO [train.py:968] (1/2) Epoch 30, batch 2950, giga_loss[loss=0.2605, simple_loss=0.3412, pruned_loss=0.08986, over 28691.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3455, pruned_loss=0.09689, over 5705599.41 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3401, pruned_loss=0.08514, over 4423254.94 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3449, pruned_loss=0.09801, over 5700385.60 frames. ], batch size: 66, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 06:59:40,824 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:00:01,074 INFO [train.py:968] (1/2) Epoch 30, batch 3000, giga_loss[loss=0.2795, simple_loss=0.3585, pruned_loss=0.1003, over 28853.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3515, pruned_loss=0.1012, over 5681220.53 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3398, pruned_loss=0.08497, over 4451401.34 frames. ], giga_tot_loss[loss=0.2782, simple_loss=0.3515, pruned_loss=0.1024, over 5674221.40 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:00:01,075 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 07:00:09,251 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2820, 1.4810, 1.4616, 1.2895], device='cuda:1'), covar=tensor([0.2202, 0.2357, 0.1636, 0.1887], device='cuda:1'), in_proj_covar=tensor([0.2077, 0.2037, 0.1948, 0.2098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:00:09,997 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 07:00:50,251 INFO [train.py:968] (1/2) Epoch 30, batch 3050, giga_loss[loss=0.2589, simple_loss=0.3466, pruned_loss=0.08558, over 28691.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3492, pruned_loss=0.09867, over 5689174.44 frames. ], libri_tot_loss[loss=0.2548, simple_loss=0.3397, pruned_loss=0.08495, over 4492539.90 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3496, pruned_loss=0.1002, over 5683886.32 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:01:09,730 INFO [optim.py:369] (1/2) 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 07:01:29,735 INFO [train.py:968] (1/2) Epoch 30, batch 3100, giga_loss[loss=0.2653, simple_loss=0.345, pruned_loss=0.09277, over 28877.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3465, pruned_loss=0.09634, over 5694528.35 frames. ], libri_tot_loss[loss=0.2549, simple_loss=0.3398, pruned_loss=0.08503, over 4516917.02 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3469, pruned_loss=0.09773, over 5689835.81 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:01:33,673 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-15 07:01:45,828 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323586.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:01:47,805 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323589.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:01:53,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3951, 1.6352, 1.2876, 1.1957], device='cuda:1'), covar=tensor([0.1155, 0.0538, 0.1018, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0454, 0.0528, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 07:01:54,399 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2296, 1.5048, 1.3451, 1.1339], device='cuda:1'), covar=tensor([0.3219, 0.2867, 0.2057, 0.2775], device='cuda:1'), in_proj_covar=tensor([0.2077, 0.2038, 0.1948, 0.2098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:02:11,290 INFO [train.py:968] (1/2) Epoch 30, batch 3150, libri_loss[loss=0.2606, simple_loss=0.3437, pruned_loss=0.08879, over 29657.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.344, pruned_loss=0.09414, over 5706738.92 frames. ], libri_tot_loss[loss=0.2547, simple_loss=0.3394, pruned_loss=0.08504, over 4579885.96 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3448, pruned_loss=0.09572, over 5698245.38 frames. ], batch size: 91, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:02:11,505 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323618.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:02:32,922 INFO [optim.py:369] (1/2) 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 07:02:54,260 INFO [train.py:968] (1/2) Epoch 30, batch 3200, giga_loss[loss=0.2489, simple_loss=0.3303, pruned_loss=0.08374, over 28613.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3447, pruned_loss=0.0942, over 5708778.89 frames. ], libri_tot_loss[loss=0.2546, simple_loss=0.3392, pruned_loss=0.08503, over 4593360.24 frames. ], giga_tot_loss[loss=0.2683, simple_loss=0.3455, pruned_loss=0.09554, over 5700255.48 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:03:00,078 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1323675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:03:34,346 INFO [train.py:968] (1/2) Epoch 30, batch 3250, giga_loss[loss=0.2771, simple_loss=0.3544, pruned_loss=0.09991, over 28435.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3457, pruned_loss=0.09451, over 5717468.46 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3386, pruned_loss=0.0847, over 4651432.72 frames. ], giga_tot_loss[loss=0.2698, simple_loss=0.3471, pruned_loss=0.09623, over 5702526.55 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:03:40,555 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5038, 1.6106, 1.6991, 1.3143], device='cuda:1'), covar=tensor([0.1759, 0.2470, 0.1429, 0.1643], device='cuda:1'), in_proj_covar=tensor([0.0948, 0.0724, 0.0998, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 07:03:56,907 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 3300, giga_loss[loss=0.249, simple_loss=0.3296, pruned_loss=0.08418, over 28682.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3464, pruned_loss=0.09517, over 5714773.40 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3386, pruned_loss=0.08469, over 4670226.91 frames. ], giga_tot_loss[loss=0.2705, simple_loss=0.3477, pruned_loss=0.09669, over 5700302.98 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:04:52,709 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 07:04:57,901 INFO [train.py:968] (1/2) Epoch 30, batch 3350, giga_loss[loss=0.26, simple_loss=0.339, pruned_loss=0.09054, over 28916.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.348, pruned_loss=0.09675, over 5713678.33 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3384, pruned_loss=0.0845, over 4699246.53 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3493, pruned_loss=0.09838, over 5699297.57 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:04:58,111 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1323818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:05:00,187 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1323821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:05:18,935 INFO [optim.py:369] (1/2) 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 07:05:24,853 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1323850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:05:39,263 INFO [train.py:968] (1/2) Epoch 30, batch 3400, giga_loss[loss=0.261, simple_loss=0.3461, pruned_loss=0.08796, over 28798.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3491, pruned_loss=0.09798, over 5721266.38 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3386, pruned_loss=0.08462, over 4728644.50 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3503, pruned_loss=0.09949, over 5705941.54 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:05:55,324 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7141, 1.7781, 1.9055, 1.4942], device='cuda:1'), covar=tensor([0.1913, 0.2619, 0.1537, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.0722, 0.0995, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 07:06:20,228 INFO [train.py:968] (1/2) Epoch 30, batch 3450, giga_loss[loss=0.2767, simple_loss=0.3531, pruned_loss=0.1002, over 28705.00 frames. ], tot_loss[loss=0.272, simple_loss=0.3487, pruned_loss=0.0976, over 5722535.18 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.3385, pruned_loss=0.08445, over 4740661.38 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3498, pruned_loss=0.0991, over 5712958.74 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:06:28,205 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5777, 1.7958, 1.4743, 1.6821], device='cuda:1'), covar=tensor([0.2725, 0.2914, 0.3222, 0.2611], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1167, 0.1431, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 07:06:43,940 INFO [optim.py:369] (1/2) 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 07:07:02,233 INFO [train.py:968] (1/2) Epoch 30, batch 3500, giga_loss[loss=0.2766, simple_loss=0.355, pruned_loss=0.09914, over 28891.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.35, pruned_loss=0.09819, over 5717738.31 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3387, pruned_loss=0.08443, over 4765181.77 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3509, pruned_loss=0.09969, over 5710249.09 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:07:39,491 INFO [train.py:968] (1/2) Epoch 30, batch 3550, giga_loss[loss=0.2716, simple_loss=0.3522, pruned_loss=0.09547, over 29061.00 frames. ], tot_loss[loss=0.2714, simple_loss=0.3492, pruned_loss=0.09679, over 5717483.82 frames. ], libri_tot_loss[loss=0.2544, simple_loss=0.3392, pruned_loss=0.0848, over 4803217.12 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.35, pruned_loss=0.09822, over 5710488.99 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:08:02,095 INFO [optim.py:369] (1/2) 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 07:08:07,581 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3318, 1.6873, 1.6456, 1.5068], device='cuda:1'), covar=tensor([0.2533, 0.2121, 0.2692, 0.2169], device='cuda:1'), in_proj_covar=tensor([0.0519, 0.0765, 0.0739, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 07:08:22,495 INFO [train.py:968] (1/2) Epoch 30, batch 3600, giga_loss[loss=0.2361, simple_loss=0.3209, pruned_loss=0.07558, over 28550.00 frames. ], tot_loss[loss=0.2716, simple_loss=0.3498, pruned_loss=0.09665, over 5722754.97 frames. ], libri_tot_loss[loss=0.2541, simple_loss=0.3388, pruned_loss=0.08466, over 4819816.58 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3509, pruned_loss=0.09804, over 5714726.55 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:09:00,746 INFO [train.py:968] (1/2) Epoch 30, batch 3650, libri_loss[loss=0.236, simple_loss=0.3316, pruned_loss=0.07025, over 29524.00 frames. ], tot_loss[loss=0.271, simple_loss=0.349, pruned_loss=0.09644, over 5723394.43 frames. ], libri_tot_loss[loss=0.2542, simple_loss=0.339, pruned_loss=0.08468, over 4834237.11 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3499, pruned_loss=0.09771, over 5716214.69 frames. ], batch size: 82, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:09:22,638 INFO [optim.py:369] (1/2) 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 07:09:38,565 INFO [train.py:968] (1/2) Epoch 30, batch 3700, libri_loss[loss=0.2424, simple_loss=0.3339, pruned_loss=0.07542, over 29242.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3472, pruned_loss=0.09606, over 5725190.10 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3384, pruned_loss=0.08432, over 4870674.46 frames. ], giga_tot_loss[loss=0.2721, simple_loss=0.3486, pruned_loss=0.09775, over 5714187.52 frames. ], batch size: 97, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:10:00,264 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0654, 1.6212, 1.3189, 1.4086], device='cuda:1'), covar=tensor([0.2673, 0.1970, 0.2664, 0.2074], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0761, 0.0735, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 07:10:14,633 INFO [train.py:968] (1/2) Epoch 30, batch 3750, giga_loss[loss=0.2695, simple_loss=0.3472, pruned_loss=0.09585, over 28993.00 frames. ], tot_loss[loss=0.2673, simple_loss=0.3448, pruned_loss=0.09491, over 5715201.15 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3385, pruned_loss=0.08463, over 4900825.51 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3463, pruned_loss=0.09654, over 5715816.09 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:10:36,724 INFO [optim.py:369] (1/2) 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 07:10:48,163 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.43 vs. limit=5.0
+2023-03-15 07:10:56,636 INFO [train.py:968] (1/2) Epoch 30, batch 3800, giga_loss[loss=0.2582, simple_loss=0.3331, pruned_loss=0.09167, over 28907.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3445, pruned_loss=0.0947, over 5717147.42 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3384, pruned_loss=0.08466, over 4910880.53 frames. ], giga_tot_loss[loss=0.2691, simple_loss=0.3459, pruned_loss=0.09616, over 5723443.44 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:11:21,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4701, 1.7187, 1.7512, 1.5096], device='cuda:1'), covar=tensor([0.2393, 0.2154, 0.2307, 0.2125], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0760, 0.0734, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 07:11:36,739 INFO [train.py:968] (1/2) Epoch 30, batch 3850, giga_loss[loss=0.2705, simple_loss=0.3508, pruned_loss=0.09513, over 28972.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3459, pruned_loss=0.09587, over 5721841.51 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3381, pruned_loss=0.08448, over 4930783.16 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3473, pruned_loss=0.09737, over 5723332.77 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:11:43,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 07:11:48,223 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9438, 1.1639, 3.2282, 2.9613], device='cuda:1'), covar=tensor([0.2162, 0.3139, 0.0887, 0.1409], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0673, 0.1008, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 07:11:50,599 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3695, 1.5141, 1.4286, 1.3221], device='cuda:1'), covar=tensor([0.2995, 0.2795, 0.2176, 0.2546], device='cuda:1'), in_proj_covar=tensor([0.2069, 0.2033, 0.1947, 0.2094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:11:57,730 INFO [optim.py:369] (1/2) 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 07:12:12,736 INFO [train.py:968] (1/2) Epoch 30, batch 3900, giga_loss[loss=0.2808, simple_loss=0.362, pruned_loss=0.09978, over 28733.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3462, pruned_loss=0.09585, over 5718342.47 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.338, pruned_loss=0.08447, over 4951182.22 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3476, pruned_loss=0.09731, over 5718912.68 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:12:45,831 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2947, 1.4652, 1.3227, 1.2248], device='cuda:1'), covar=tensor([0.3079, 0.2883, 0.2365, 0.2897], device='cuda:1'), in_proj_covar=tensor([0.2066, 0.2032, 0.1942, 0.2091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:12:55,352 INFO [train.py:968] (1/2) Epoch 30, batch 3950, giga_loss[loss=0.2742, simple_loss=0.3573, pruned_loss=0.09558, over 28314.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3451, pruned_loss=0.09461, over 5718357.59 frames. ], libri_tot_loss[loss=0.2534, simple_loss=0.3379, pruned_loss=0.08445, over 4970198.25 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3465, pruned_loss=0.09597, over 5715639.73 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:13:17,734 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 4000, giga_loss[loss=0.2596, simple_loss=0.3367, pruned_loss=0.09126, over 28823.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3446, pruned_loss=0.09467, over 5723645.28 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3376, pruned_loss=0.08432, over 4988237.18 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3459, pruned_loss=0.09601, over 5718812.76 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:14:13,748 INFO [train.py:968] (1/2) Epoch 30, batch 4050, giga_loss[loss=0.2475, simple_loss=0.3291, pruned_loss=0.08299, over 28702.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3429, pruned_loss=0.09424, over 5715764.48 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3377, pruned_loss=0.0844, over 4997383.99 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3439, pruned_loss=0.09533, over 5710273.69 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:14:35,195 INFO [optim.py:369] (1/2) 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 07:14:43,546 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1671, 1.7058, 1.2914, 0.4100], device='cuda:1'), covar=tensor([0.6050, 0.3340, 0.5123, 0.7709], device='cuda:1'), in_proj_covar=tensor([0.1839, 0.1720, 0.1645, 0.1497], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 07:14:50,634 INFO [train.py:968] (1/2) Epoch 30, batch 4100, libri_loss[loss=0.2939, simple_loss=0.3813, pruned_loss=0.1032, over 25691.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3401, pruned_loss=0.0928, over 5710662.64 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3383, pruned_loss=0.08479, over 5020427.22 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3406, pruned_loss=0.09361, over 5704898.04 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:15:17,841 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3296, 1.4008, 3.9358, 3.2539], device='cuda:1'), covar=tensor([0.1644, 0.2752, 0.0405, 0.1095], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0672, 0.1004, 0.0984], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 07:15:26,750 INFO [train.py:968] (1/2) Epoch 30, batch 4150, giga_loss[loss=0.2343, simple_loss=0.3155, pruned_loss=0.07657, over 28703.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3394, pruned_loss=0.09284, over 5702206.91 frames. ], libri_tot_loss[loss=0.2545, simple_loss=0.3388, pruned_loss=0.08507, over 5032614.00 frames. ], giga_tot_loss[loss=0.2632, simple_loss=0.3395, pruned_loss=0.09343, over 5701270.61 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:15:48,294 INFO [optim.py:369] (1/2) 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 07:16:03,273 INFO [train.py:968] (1/2) Epoch 30, batch 4200, libri_loss[loss=0.2733, simple_loss=0.3591, pruned_loss=0.09373, over 27854.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3392, pruned_loss=0.09322, over 5705469.35 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3397, pruned_loss=0.08562, over 5056583.66 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3385, pruned_loss=0.09349, over 5701157.88 frames. ], batch size: 116, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:16:06,022 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7822, 1.8520, 1.9616, 1.5676], device='cuda:1'), covar=tensor([0.2029, 0.2560, 0.1621, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0725, 0.0999, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 07:16:13,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4206, 1.5659, 1.5030, 1.3385], device='cuda:1'), covar=tensor([0.3279, 0.2950, 0.2489, 0.2970], device='cuda:1'), in_proj_covar=tensor([0.2072, 0.2039, 0.1949, 0.2098], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:16:41,339 INFO [train.py:968] (1/2) Epoch 30, batch 4250, giga_loss[loss=0.3134, simple_loss=0.3642, pruned_loss=0.1313, over 23894.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3376, pruned_loss=0.09271, over 5708960.95 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3394, pruned_loss=0.08543, over 5085732.70 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3372, pruned_loss=0.0933, over 5699642.98 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:16:53,663 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1324730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:17:05,715 INFO [optim.py:369] (1/2) 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 07:17:13,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3829, 2.3746, 2.3827, 2.1904], device='cuda:1'), covar=tensor([0.3521, 0.2843, 0.2768, 0.2942], device='cuda:1'), in_proj_covar=tensor([0.2076, 0.2044, 0.1953, 0.2101], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:17:21,152 INFO [train.py:968] (1/2) Epoch 30, batch 4300, giga_loss[loss=0.2654, simple_loss=0.3384, pruned_loss=0.09616, over 28885.00 frames. ], tot_loss[loss=0.26, simple_loss=0.336, pruned_loss=0.09196, over 5716910.86 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3402, pruned_loss=0.08585, over 5108926.41 frames. ], giga_tot_loss[loss=0.2598, simple_loss=0.335, pruned_loss=0.09229, over 5705097.13 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:17:58,459 INFO [train.py:968] (1/2) Epoch 30, batch 4350, giga_loss[loss=0.2455, simple_loss=0.3245, pruned_loss=0.08325, over 28905.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3334, pruned_loss=0.09134, over 5708235.83 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.34, pruned_loss=0.0857, over 5111170.34 frames. ], giga_tot_loss[loss=0.2581, simple_loss=0.3327, pruned_loss=0.09178, over 5704566.49 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:18:20,711 INFO [optim.py:369] (1/2) 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 07:18:36,749 INFO [train.py:968] (1/2) Epoch 30, batch 4400, giga_loss[loss=0.2584, simple_loss=0.3315, pruned_loss=0.09272, over 28568.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3307, pruned_loss=0.08982, over 5709842.08 frames. ], libri_tot_loss[loss=0.2561, simple_loss=0.3403, pruned_loss=0.08595, over 5122002.70 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3297, pruned_loss=0.09004, over 5708011.45 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:19:13,864 INFO [train.py:968] (1/2) Epoch 30, batch 4450, giga_loss[loss=0.2396, simple_loss=0.3232, pruned_loss=0.07798, over 29030.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.33, pruned_loss=0.08905, over 5700050.81 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3405, pruned_loss=0.08624, over 5121121.81 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3288, pruned_loss=0.08906, over 5711718.15 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:19:37,735 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 4500, giga_loss[loss=0.2481, simple_loss=0.3362, pruned_loss=0.08002, over 28743.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3336, pruned_loss=0.09067, over 5691425.83 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3409, pruned_loss=0.08666, over 5138628.63 frames. ], giga_tot_loss[loss=0.2564, simple_loss=0.3321, pruned_loss=0.09041, over 5702601.85 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:19:56,442 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1324971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:20:34,070 INFO [train.py:968] (1/2) Epoch 30, batch 4550, giga_loss[loss=0.2572, simple_loss=0.3438, pruned_loss=0.08532, over 28684.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.336, pruned_loss=0.09145, over 5693749.86 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3407, pruned_loss=0.08662, over 5149802.88 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3349, pruned_loss=0.09135, over 5699892.55 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:20:59,971 INFO [optim.py:369] (1/2) 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 07:21:04,844 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7607, 1.9194, 1.6078, 1.9756], device='cuda:1'), covar=tensor([0.2963, 0.3131, 0.3447, 0.2738], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1166, 0.1430, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 07:21:16,847 INFO [train.py:968] (1/2) Epoch 30, batch 4600, giga_loss[loss=0.2406, simple_loss=0.3289, pruned_loss=0.07611, over 28657.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3383, pruned_loss=0.09211, over 5688379.68 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3406, pruned_loss=0.08662, over 5157256.36 frames. ], giga_tot_loss[loss=0.2608, simple_loss=0.3375, pruned_loss=0.09208, over 5691268.51 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:21:24,579 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-15 07:21:48,160 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1325105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:21:50,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3022, 1.6373, 1.4354, 1.4798], device='cuda:1'), covar=tensor([0.0731, 0.0371, 0.0345, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 07:21:57,999 INFO [train.py:968] (1/2) Epoch 30, batch 4650, giga_loss[loss=0.2561, simple_loss=0.3263, pruned_loss=0.09294, over 28778.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3376, pruned_loss=0.09088, over 5689749.10 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3405, pruned_loss=0.08662, over 5173049.93 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.337, pruned_loss=0.09096, over 5689581.24 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:22:23,241 INFO [optim.py:369] (1/2) 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 07:22:36,594 INFO [train.py:968] (1/2) Epoch 30, batch 4700, giga_loss[loss=0.256, simple_loss=0.3347, pruned_loss=0.08864, over 28883.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3369, pruned_loss=0.09021, over 5702977.42 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3406, pruned_loss=0.08653, over 5193936.91 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3362, pruned_loss=0.09046, over 5697947.40 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:23:17,408 INFO [train.py:968] (1/2) Epoch 30, batch 4750, giga_loss[loss=0.2657, simple_loss=0.3434, pruned_loss=0.09404, over 28908.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.338, pruned_loss=0.09131, over 5702746.08 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3408, pruned_loss=0.08675, over 5204595.20 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3371, pruned_loss=0.09138, over 5695685.67 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:23:40,369 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:23:42,301 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1325251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:23:54,148 INFO [train.py:968] (1/2) Epoch 30, batch 4800, giga_loss[loss=0.2802, simple_loss=0.3552, pruned_loss=0.1027, over 28767.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3389, pruned_loss=0.09205, over 5706778.53 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3409, pruned_loss=0.08681, over 5228862.89 frames. ], giga_tot_loss[loss=0.2613, simple_loss=0.3382, pruned_loss=0.09222, over 5694695.81 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:24:04,301 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1325280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:24:34,514 INFO [train.py:968] (1/2) Epoch 30, batch 4850, giga_loss[loss=0.2284, simple_loss=0.3087, pruned_loss=0.07411, over 28710.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3419, pruned_loss=0.09326, over 5705693.83 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3416, pruned_loss=0.08703, over 5244376.59 frames. ], giga_tot_loss[loss=0.2637, simple_loss=0.3407, pruned_loss=0.09333, over 5692795.54 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:24:38,130 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1562, 1.3773, 1.4936, 1.3400], device='cuda:1'), covar=tensor([0.1744, 0.1269, 0.1835, 0.1378], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0763, 0.0737, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 07:24:57,506 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1325346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:24:59,261 INFO [optim.py:369] (1/2) 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 07:25:12,956 INFO [train.py:968] (1/2) Epoch 30, batch 4900, giga_loss[loss=0.2401, simple_loss=0.3217, pruned_loss=0.07921, over 28561.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3438, pruned_loss=0.09389, over 5714386.51 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3414, pruned_loss=0.08705, over 5257369.86 frames. ], giga_tot_loss[loss=0.2656, simple_loss=0.343, pruned_loss=0.09406, over 5700668.94 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:25:57,427 INFO [train.py:968] (1/2) Epoch 30, batch 4950, giga_loss[loss=0.2821, simple_loss=0.3616, pruned_loss=0.1013, over 27994.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3456, pruned_loss=0.09491, over 5713346.52 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3414, pruned_loss=0.08705, over 5257369.86 frames. ], giga_tot_loss[loss=0.2676, simple_loss=0.345, pruned_loss=0.09505, over 5702669.96 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:26:11,887 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8999, 2.1338, 2.0086, 1.5992], device='cuda:1'), covar=tensor([0.3332, 0.2498, 0.2902, 0.3250], device='cuda:1'), in_proj_covar=tensor([0.2090, 0.2055, 0.1960, 0.2105], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:26:14,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3159, 1.5400, 3.6307, 3.2060], device='cuda:1'), covar=tensor([0.1589, 0.2527, 0.0490, 0.0973], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0676, 0.1013, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 07:26:20,724 INFO [optim.py:369] (1/2) 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 07:26:34,073 INFO [train.py:968] (1/2) Epoch 30, batch 5000, giga_loss[loss=0.288, simple_loss=0.3602, pruned_loss=0.1079, over 29046.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3465, pruned_loss=0.09531, over 5710354.64 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3419, pruned_loss=0.08733, over 5256007.35 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3457, pruned_loss=0.09528, over 5709339.53 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:26:53,470 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1325489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:26:56,240 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 5050, giga_loss[loss=0.2407, simple_loss=0.3208, pruned_loss=0.08035, over 28562.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3468, pruned_loss=0.0958, over 5717042.35 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3418, pruned_loss=0.08726, over 5262426.05 frames. ], giga_tot_loss[loss=0.269, simple_loss=0.3463, pruned_loss=0.09591, over 5714633.96 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:27:19,803 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1325521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:27:41,028 INFO [optim.py:369] (1/2) 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 07:27:57,227 INFO [train.py:968] (1/2) Epoch 30, batch 5100, giga_loss[loss=0.2454, simple_loss=0.3306, pruned_loss=0.08009, over 28958.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3454, pruned_loss=0.09521, over 5714504.15 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3418, pruned_loss=0.08716, over 5271533.75 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.345, pruned_loss=0.09554, over 5712426.24 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:27:59,574 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4792, 3.4460, 1.6108, 1.5170], device='cuda:1'), covar=tensor([0.0973, 0.0356, 0.0916, 0.1377], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0573, 0.0415, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 07:28:36,937 INFO [train.py:968] (1/2) Epoch 30, batch 5150, giga_loss[loss=0.2309, simple_loss=0.3051, pruned_loss=0.07834, over 28825.00 frames. ], tot_loss[loss=0.2641, simple_loss=0.342, pruned_loss=0.09315, over 5719603.10 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3418, pruned_loss=0.08715, over 5284210.24 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3417, pruned_loss=0.09358, over 5716779.52 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:29:01,932 INFO [optim.py:369] (1/2) 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 07:29:13,532 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1325665.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:29:15,494 INFO [train.py:968] (1/2) Epoch 30, batch 5200, giga_loss[loss=0.2899, simple_loss=0.3592, pruned_loss=0.1103, over 27931.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3382, pruned_loss=0.09127, over 5715019.13 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3418, pruned_loss=0.08726, over 5288922.98 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.338, pruned_loss=0.09162, over 5718693.39 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:29:21,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8912, 1.1859, 1.3670, 0.9788], device='cuda:1'), covar=tensor([0.2244, 0.1651, 0.2590, 0.2095], device='cuda:1'), in_proj_covar=tensor([0.0518, 0.0765, 0.0739, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 07:29:53,496 INFO [train.py:968] (1/2) Epoch 30, batch 5250, giga_loss[loss=0.2254, simple_loss=0.3037, pruned_loss=0.07357, over 28642.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3377, pruned_loss=0.09078, over 5716280.33 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3418, pruned_loss=0.08727, over 5306213.03 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3374, pruned_loss=0.09119, over 5716567.76 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:30:17,444 INFO [optim.py:369] (1/2) 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 07:30:34,132 INFO [train.py:968] (1/2) Epoch 30, batch 5300, giga_loss[loss=0.2815, simple_loss=0.3685, pruned_loss=0.09721, over 28587.00 frames. ], tot_loss[loss=0.2604, simple_loss=0.3397, pruned_loss=0.09054, over 5709950.29 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3418, pruned_loss=0.08733, over 5320813.76 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3393, pruned_loss=0.09088, over 5706023.80 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:30:55,255 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5953, 1.7951, 1.2749, 1.3499], device='cuda:1'), covar=tensor([0.1138, 0.0741, 0.1182, 0.1349], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0449, 0.0524, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 07:31:13,726 INFO [train.py:968] (1/2) Epoch 30, batch 5350, giga_loss[loss=0.3039, simple_loss=0.374, pruned_loss=0.1169, over 28797.00 frames. ], tot_loss[loss=0.2612, simple_loss=0.3403, pruned_loss=0.09103, over 5704734.34 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.342, pruned_loss=0.08742, over 5328880.89 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3399, pruned_loss=0.09129, over 5701423.43 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:31:36,818 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2835, 1.7707, 1.8096, 1.5273], device='cuda:1'), covar=tensor([0.2222, 0.1733, 0.2170, 0.1978], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0764, 0.0738, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 07:31:39,382 INFO [optim.py:369] (1/2) 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 07:31:39,842 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 07:31:44,812 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1325856.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:31:52,935 INFO [train.py:968] (1/2) Epoch 30, batch 5400, giga_loss[loss=0.2842, simple_loss=0.3598, pruned_loss=0.1043, over 28861.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.3398, pruned_loss=0.09181, over 5711054.83 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3416, pruned_loss=0.08745, over 5348782.41 frames. ], giga_tot_loss[loss=0.262, simple_loss=0.3397, pruned_loss=0.09214, over 5701945.01 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:32:32,756 INFO [train.py:968] (1/2) Epoch 30, batch 5450, giga_loss[loss=0.2476, simple_loss=0.3178, pruned_loss=0.08869, over 28445.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3392, pruned_loss=0.0928, over 5706907.27 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3413, pruned_loss=0.08726, over 5361681.53 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3394, pruned_loss=0.09335, over 5697786.82 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:32:44,412 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1325932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:32:56,359 INFO [optim.py:369] (1/2) 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 07:32:59,212 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5870, 1.8613, 1.4688, 1.8901], device='cuda:1'), covar=tensor([0.2711, 0.2885, 0.3290, 0.2429], device='cuda:1'), in_proj_covar=tensor([0.1613, 0.1162, 0.1427, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 07:33:11,028 INFO [train.py:968] (1/2) Epoch 30, batch 5500, giga_loss[loss=0.2455, simple_loss=0.3185, pruned_loss=0.08623, over 28943.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3378, pruned_loss=0.09308, over 5705704.42 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.341, pruned_loss=0.08711, over 5363635.80 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3381, pruned_loss=0.09374, over 5703579.09 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:33:31,552 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 07:33:49,024 INFO [train.py:968] (1/2) Epoch 30, batch 5550, giga_loss[loss=0.2549, simple_loss=0.3296, pruned_loss=0.09006, over 28710.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.3375, pruned_loss=0.0937, over 5697301.04 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3413, pruned_loss=0.08733, over 5368713.71 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3374, pruned_loss=0.0942, over 5700680.37 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:34:09,114 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:34:15,569 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 5600, giga_loss[loss=0.2252, simple_loss=0.2989, pruned_loss=0.07576, over 28449.00 frames. ], tot_loss[loss=0.2614, simple_loss=0.3362, pruned_loss=0.09329, over 5707404.14 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3415, pruned_loss=0.08742, over 5381325.54 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3359, pruned_loss=0.09374, over 5706096.84 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:35:09,695 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:35:11,555 INFO [train.py:968] (1/2) Epoch 30, batch 5650, giga_loss[loss=0.2122, simple_loss=0.2808, pruned_loss=0.07182, over 28534.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.3328, pruned_loss=0.09133, over 5718246.34 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3407, pruned_loss=0.08714, over 5405319.12 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3329, pruned_loss=0.0922, over 5709722.72 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:35:29,929 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:35:34,871 INFO [optim.py:369] (1/2) 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 07:35:38,527 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7952, 1.9468, 1.9689, 1.6896], device='cuda:1'), covar=tensor([0.2665, 0.2337, 0.1978, 0.2397], device='cuda:1'), in_proj_covar=tensor([0.2081, 0.2047, 0.1958, 0.2094], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:35:45,336 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1857, 1.0488, 3.9222, 3.1671], device='cuda:1'), covar=tensor([0.1771, 0.2982, 0.0553, 0.1160], device='cuda:1'), in_proj_covar=tensor([0.0807, 0.0673, 0.1007, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 07:35:48,745 INFO [train.py:968] (1/2) Epoch 30, batch 5700, giga_loss[loss=0.2616, simple_loss=0.3289, pruned_loss=0.09718, over 28786.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3281, pruned_loss=0.08868, over 5720476.80 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3405, pruned_loss=0.08708, over 5415884.38 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.328, pruned_loss=0.08951, over 5715911.61 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:36:01,216 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:36:03,971 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:36:25,935 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:36:27,545 INFO [train.py:968] (1/2) Epoch 30, batch 5750, giga_loss[loss=0.2973, simple_loss=0.3645, pruned_loss=0.1151, over 28731.00 frames. ], tot_loss[loss=0.2502, simple_loss=0.3256, pruned_loss=0.08742, over 5723357.89 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3402, pruned_loss=0.087, over 5427323.64 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3253, pruned_loss=0.08817, over 5716026.30 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:36:38,826 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326231.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:36:52,152 INFO [optim.py:369] (1/2) 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 07:37:05,418 INFO [train.py:968] (1/2) Epoch 30, batch 5800, giga_loss[loss=0.2708, simple_loss=0.3601, pruned_loss=0.09069, over 28605.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3273, pruned_loss=0.08805, over 5724027.49 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3405, pruned_loss=0.08716, over 5435096.46 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3266, pruned_loss=0.08853, over 5716386.62 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:37:36,707 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:37:44,594 INFO [train.py:968] (1/2) Epoch 30, batch 5850, giga_loss[loss=0.2529, simple_loss=0.331, pruned_loss=0.08744, over 28979.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3308, pruned_loss=0.08959, over 5725090.50 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3404, pruned_loss=0.08715, over 5444095.31 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3301, pruned_loss=0.09, over 5716080.26 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:38:09,901 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 5900, giga_loss[loss=0.307, simple_loss=0.3776, pruned_loss=0.1182, over 28789.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3346, pruned_loss=0.0915, over 5720786.00 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3401, pruned_loss=0.08712, over 5447004.34 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3342, pruned_loss=0.09191, over 5714619.79 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:38:29,403 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326374.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:38:31,732 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326377.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:38:54,451 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326406.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:39:03,612 INFO [train.py:968] (1/2) Epoch 30, batch 5950, giga_loss[loss=0.2582, simple_loss=0.3379, pruned_loss=0.08927, over 28876.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.338, pruned_loss=0.09272, over 5712700.24 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3404, pruned_loss=0.08728, over 5450450.89 frames. ], giga_tot_loss[loss=0.2618, simple_loss=0.3374, pruned_loss=0.09305, over 5711197.29 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:39:03,825 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2687, 1.1298, 3.6604, 3.1629], device='cuda:1'), covar=tensor([0.1838, 0.3031, 0.0800, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0672, 0.1008, 0.0987], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 07:39:27,620 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5882, 2.3024, 1.6949, 0.7988], device='cuda:1'), covar=tensor([0.8198, 0.3626, 0.4939, 0.8696], device='cuda:1'), in_proj_covar=tensor([0.1852, 0.1731, 0.1662, 0.1512], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 07:39:31,888 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:39:32,272 INFO [optim.py:369] (1/2) 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 07:39:35,579 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:39:47,081 INFO [train.py:968] (1/2) Epoch 30, batch 6000, giga_loss[loss=0.325, simple_loss=0.3791, pruned_loss=0.1354, over 26622.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3399, pruned_loss=0.09405, over 5712612.47 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3402, pruned_loss=0.08734, over 5457277.24 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3396, pruned_loss=0.09434, over 5708691.17 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:39:47,081 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 07:39:55,444 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2541, 1.8050, 1.3765, 0.4481], device='cuda:1'), covar=tensor([0.4991, 0.3934, 0.5572, 0.7121], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1731, 0.1664, 0.1512], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 07:39:56,253 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 07:40:07,458 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:40:08,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:40:15,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:40:39,789 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 6050, giga_loss[loss=0.3216, simple_loss=0.3857, pruned_loss=0.1288, over 28988.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.3447, pruned_loss=0.09755, over 5713157.95 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3402, pruned_loss=0.08721, over 5463853.27 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3445, pruned_loss=0.09803, over 5707507.13 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:40:57,468 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:41:07,441 INFO [optim.py:369] (1/2) 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 07:41:08,523 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 6100, giga_loss[loss=0.3119, simple_loss=0.3571, pruned_loss=0.1333, over 23855.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.35, pruned_loss=0.1019, over 5704357.84 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3405, pruned_loss=0.08753, over 5476954.82 frames. ], giga_tot_loss[loss=0.2772, simple_loss=0.3498, pruned_loss=0.1023, over 5694712.70 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:41:52,591 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3080, 1.2732, 3.6519, 3.2598], device='cuda:1'), covar=tensor([0.1829, 0.2832, 0.0838, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0672, 0.1008, 0.0988], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 07:41:52,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4802, 1.6291, 1.5577, 1.4341], device='cuda:1'), covar=tensor([0.1764, 0.1915, 0.2221, 0.1951], device='cuda:1'), in_proj_covar=tensor([0.0517, 0.0765, 0.0739, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 07:42:07,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6144, 1.9094, 1.5246, 1.8273], device='cuda:1'), covar=tensor([0.2587, 0.2685, 0.3014, 0.2488], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1165, 0.1431, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 07:42:08,732 INFO [train.py:968] (1/2) Epoch 30, batch 6150, giga_loss[loss=0.2809, simple_loss=0.3525, pruned_loss=0.1047, over 28694.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.356, pruned_loss=0.1062, over 5685411.09 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3403, pruned_loss=0.0874, over 5485538.31 frames. ], giga_tot_loss[loss=0.2853, simple_loss=0.3563, pruned_loss=0.1071, over 5675873.48 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:42:13,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3665, 1.6406, 1.4293, 1.5260], device='cuda:1'), covar=tensor([0.0797, 0.0338, 0.0341, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 07:42:22,287 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:25,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326636.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:41,969 INFO [optim.py:369] (1/2) 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 07:42:48,267 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:51,536 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1326663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:42:54,985 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 6200, giga_loss[loss=0.3367, simple_loss=0.3997, pruned_loss=0.1368, over 29039.00 frames. ], tot_loss[loss=0.2947, simple_loss=0.364, pruned_loss=0.1126, over 5673851.66 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08761, over 5482674.47 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3644, pruned_loss=0.1135, over 5671948.18 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:43:19,589 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1326692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:43:26,774 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1326700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:43:29,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9356, 2.8358, 1.7356, 1.2171], device='cuda:1'), covar=tensor([0.7598, 0.3367, 0.4360, 0.6634], device='cuda:1'), in_proj_covar=tensor([0.1848, 0.1732, 0.1661, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 07:43:42,405 INFO [train.py:968] (1/2) Epoch 30, batch 6250, giga_loss[loss=0.3452, simple_loss=0.4103, pruned_loss=0.1401, over 28953.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3714, pruned_loss=0.1188, over 5676971.42 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3412, pruned_loss=0.08789, over 5487977.37 frames. ], giga_tot_loss[loss=0.3055, simple_loss=0.3716, pruned_loss=0.1197, over 5674309.25 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:44:09,439 INFO [optim.py:369] (1/2) 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 07:44:24,871 INFO [train.py:968] (1/2) Epoch 30, batch 6300, giga_loss[loss=0.2809, simple_loss=0.3527, pruned_loss=0.1046, over 28829.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3755, pruned_loss=0.1218, over 5661878.47 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3418, pruned_loss=0.0882, over 5489480.17 frames. ], giga_tot_loss[loss=0.3116, simple_loss=0.3762, pruned_loss=0.1235, over 5664614.96 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:44:52,405 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2172, 1.4381, 1.4208, 1.1511], device='cuda:1'), covar=tensor([0.2896, 0.2920, 0.1931, 0.2690], device='cuda:1'), in_proj_covar=tensor([0.2077, 0.2047, 0.1951, 0.2088], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:44:58,952 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2919, 0.8218, 0.9252, 1.4193], device='cuda:1'), covar=tensor([0.0782, 0.0382, 0.0367, 0.0862], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 07:45:02,287 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.6003, 3.4714, 3.3100, 1.7232], device='cuda:1'), covar=tensor([0.0813, 0.0830, 0.0794, 0.2294], device='cuda:1'), in_proj_covar=tensor([0.1316, 0.1213, 0.1018, 0.0759], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 07:45:14,313 INFO [train.py:968] (1/2) Epoch 30, batch 6350, giga_loss[loss=0.2765, simple_loss=0.3528, pruned_loss=0.1, over 28472.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3795, pruned_loss=0.1264, over 5648931.29 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.342, pruned_loss=0.08834, over 5497575.18 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3806, pruned_loss=0.1282, over 5646902.88 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:45:48,090 INFO [optim.py:369] (1/2) 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 07:45:51,102 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:46:04,232 INFO [train.py:968] (1/2) Epoch 30, batch 6400, giga_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.122, over 28894.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.382, pruned_loss=0.1296, over 5624865.89 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3419, pruned_loss=0.08834, over 5495025.20 frames. ], giga_tot_loss[loss=0.3234, simple_loss=0.3835, pruned_loss=0.1317, over 5628719.95 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:46:36,906 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-15 07:46:50,419 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:46:52,849 INFO [train.py:968] (1/2) Epoch 30, batch 6450, giga_loss[loss=0.3668, simple_loss=0.414, pruned_loss=0.1598, over 28545.00 frames. ], tot_loss[loss=0.3251, simple_loss=0.3846, pruned_loss=0.1328, over 5619134.28 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3415, pruned_loss=0.08799, over 5507362.97 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3874, pruned_loss=0.136, over 5614327.06 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:47:02,672 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1326927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:47:06,734 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 07:47:26,203 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 6500, giga_loss[loss=0.3041, simple_loss=0.3777, pruned_loss=0.1152, over 28493.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3857, pruned_loss=0.1336, over 5615516.59 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3404, pruned_loss=0.08737, over 5514622.65 frames. ], giga_tot_loss[loss=0.3329, simple_loss=0.39, pruned_loss=0.1379, over 5608201.14 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:48:13,603 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1326999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:48:16,325 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:48:30,327 INFO [train.py:968] (1/2) Epoch 30, batch 6550, giga_loss[loss=0.4574, simple_loss=0.4697, pruned_loss=0.2225, over 26620.00 frames. ], tot_loss[loss=0.3278, simple_loss=0.3866, pruned_loss=0.1345, over 5626394.41 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3409, pruned_loss=0.08759, over 5519979.09 frames. ], giga_tot_loss[loss=0.3334, simple_loss=0.3902, pruned_loss=0.1383, over 5617358.91 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:48:44,263 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:48:53,286 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:49:05,970 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:19,760 INFO [train.py:968] (1/2) Epoch 30, batch 6600, giga_loss[loss=0.3576, simple_loss=0.4085, pruned_loss=0.1534, over 27906.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3838, pruned_loss=0.1329, over 5631610.53 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3408, pruned_loss=0.08752, over 5522862.65 frames. ], giga_tot_loss[loss=0.3316, simple_loss=0.3882, pruned_loss=0.1375, over 5624785.80 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:49:21,554 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:24,409 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327073.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:26,690 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:39,925 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:49:52,386 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:50:07,073 INFO [train.py:968] (1/2) Epoch 30, batch 6650, libri_loss[loss=0.2607, simple_loss=0.3521, pruned_loss=0.0847, over 29530.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3832, pruned_loss=0.1325, over 5631452.61 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3408, pruned_loss=0.08741, over 5534795.15 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3882, pruned_loss=0.1378, over 5618384.81 frames. ], batch size: 81, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:50:36,145 INFO [optim.py:369] (1/2) 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 07:50:53,901 INFO [train.py:968] (1/2) Epoch 30, batch 6700, giga_loss[loss=0.3333, simple_loss=0.3954, pruned_loss=0.1356, over 28616.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3838, pruned_loss=0.1318, over 5633283.97 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3406, pruned_loss=0.08724, over 5533909.84 frames. ], giga_tot_loss[loss=0.3317, simple_loss=0.389, pruned_loss=0.1372, over 5625906.24 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:51:42,421 INFO [train.py:968] (1/2) Epoch 30, batch 6750, giga_loss[loss=0.3202, simple_loss=0.3608, pruned_loss=0.1398, over 23333.00 frames. ], tot_loss[loss=0.3257, simple_loss=0.3856, pruned_loss=0.1329, over 5622544.41 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3404, pruned_loss=0.08719, over 5535892.29 frames. ], giga_tot_loss[loss=0.3324, simple_loss=0.39, pruned_loss=0.1374, over 5615265.29 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:51:42,763 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:51:47,239 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:52:06,416 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-15 07:52:08,390 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6610, 1.9279, 1.6302, 1.6805], device='cuda:1'), covar=tensor([0.2207, 0.2077, 0.2162, 0.1939], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1167, 0.1433, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 07:52:13,086 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:52:14,100 INFO [optim.py:369] (1/2) 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 07:52:29,591 INFO [train.py:968] (1/2) Epoch 30, batch 6800, giga_loss[loss=0.2717, simple_loss=0.3487, pruned_loss=0.09734, over 29033.00 frames. ], tot_loss[loss=0.3226, simple_loss=0.3832, pruned_loss=0.131, over 5620628.77 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3403, pruned_loss=0.08717, over 5541653.60 frames. ], giga_tot_loss[loss=0.3293, simple_loss=0.3877, pruned_loss=0.1354, over 5611066.73 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 07:53:18,134 INFO [train.py:968] (1/2) Epoch 30, batch 6850, giga_loss[loss=0.2722, simple_loss=0.357, pruned_loss=0.09367, over 28739.00 frames. ], tot_loss[loss=0.3179, simple_loss=0.3806, pruned_loss=0.1276, over 5624129.10 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3406, pruned_loss=0.0874, over 5549073.29 frames. ], giga_tot_loss[loss=0.3244, simple_loss=0.3848, pruned_loss=0.1319, over 5611678.18 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:53:29,416 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:54:05,787 INFO [train.py:968] (1/2) Epoch 30, batch 6900, giga_loss[loss=0.2822, simple_loss=0.3575, pruned_loss=0.1034, over 28478.00 frames. ], tot_loss[loss=0.3118, simple_loss=0.3764, pruned_loss=0.1236, over 5638534.90 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3404, pruned_loss=0.08755, over 5555163.23 frames. ], giga_tot_loss[loss=0.3185, simple_loss=0.3811, pruned_loss=0.128, over 5625136.71 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:54:06,703 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4395, 1.5316, 1.4423, 1.5900], device='cuda:1'), covar=tensor([0.0788, 0.0343, 0.0338, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 07:54:40,721 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8231, 1.9848, 1.9728, 1.7197], device='cuda:1'), covar=tensor([0.2681, 0.2481, 0.2067, 0.2329], device='cuda:1'), in_proj_covar=tensor([0.2087, 0.2056, 0.1961, 0.2096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 07:54:48,344 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:54:48,663 INFO [train.py:968] (1/2) Epoch 30, batch 6950, giga_loss[loss=0.296, simple_loss=0.3671, pruned_loss=0.1125, over 28569.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3727, pruned_loss=0.1203, over 5650051.75 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.34, pruned_loss=0.08736, over 5569130.41 frames. ], giga_tot_loss[loss=0.315, simple_loss=0.3785, pruned_loss=0.1257, over 5630078.21 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:55:25,286 INFO [optim.py:369] (1/2) 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 07:55:35,395 INFO [train.py:968] (1/2) Epoch 30, batch 7000, libri_loss[loss=0.2643, simple_loss=0.3486, pruned_loss=0.08999, over 29770.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3699, pruned_loss=0.1181, over 5657280.97 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3401, pruned_loss=0.08742, over 5577585.54 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3752, pruned_loss=0.1231, over 5635170.39 frames. ], batch size: 87, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:56:09,676 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3197, 1.4693, 1.2948, 1.5192], device='cuda:1'), covar=tensor([0.0794, 0.0351, 0.0345, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 07:56:20,630 INFO [train.py:968] (1/2) Epoch 30, batch 7050, giga_loss[loss=0.288, simple_loss=0.3609, pruned_loss=0.1076, over 28268.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3672, pruned_loss=0.1162, over 5661597.44 frames. ], libri_tot_loss[loss=0.2569, simple_loss=0.3396, pruned_loss=0.08714, over 5586095.52 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3728, pruned_loss=0.1213, over 5638318.91 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:56:31,085 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 07:56:59,569 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9513, 1.3354, 1.0857, 0.2293], device='cuda:1'), covar=tensor([0.4964, 0.3501, 0.4909, 0.7435], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1744, 0.1669, 0.1517], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 07:56:59,572 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327560.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:57:03,782 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327563.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 07:57:07,599 INFO [train.py:968] (1/2) Epoch 30, batch 7100, giga_loss[loss=0.2945, simple_loss=0.366, pruned_loss=0.1115, over 28227.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.368, pruned_loss=0.1166, over 5672580.03 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3398, pruned_loss=0.08729, over 5592005.43 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3727, pruned_loss=0.1211, over 5650295.39 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:57:30,625 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327592.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 07:57:52,700 INFO [train.py:968] (1/2) Epoch 30, batch 7150, giga_loss[loss=0.2747, simple_loss=0.3505, pruned_loss=0.09944, over 29003.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3662, pruned_loss=0.1145, over 5679297.32 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.34, pruned_loss=0.0874, over 5599301.32 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3704, pruned_loss=0.1186, over 5656787.41 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 07:58:03,988 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7653, 1.0187, 2.8045, 2.7177], device='cuda:1'), covar=tensor([0.1893, 0.2878, 0.0722, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0682, 0.1021, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 07:58:18,481 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2847, 1.4262, 1.3564, 1.4786], device='cuda:1'), covar=tensor([0.0829, 0.0375, 0.0356, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 07:58:18,664 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 07:58:27,433 INFO [optim.py:369] (1/2) 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 07:58:41,913 INFO [train.py:968] (1/2) Epoch 30, batch 7200, giga_loss[loss=0.2917, simple_loss=0.3675, pruned_loss=0.108, over 28523.00 frames. ], tot_loss[loss=0.2965, simple_loss=0.3667, pruned_loss=0.1131, over 5670283.59 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.0876, over 5601796.70 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3703, pruned_loss=0.1168, over 5652061.73 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:59:03,263 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2587, 3.6268, 1.4694, 1.5003], device='cuda:1'), covar=tensor([0.1081, 0.0279, 0.0974, 0.1435], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0578, 0.0416, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 07:59:12,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 07:59:12,675 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4530, 4.2891, 4.0923, 2.3751], device='cuda:1'), covar=tensor([0.0698, 0.0829, 0.0889, 0.1625], device='cuda:1'), in_proj_covar=tensor([0.1330, 0.1230, 0.1028, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 07:59:19,853 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 7250, giga_loss[loss=0.3854, simple_loss=0.4261, pruned_loss=0.1724, over 26708.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3692, pruned_loss=0.1133, over 5679028.45 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3408, pruned_loss=0.08778, over 5611461.12 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3727, pruned_loss=0.117, over 5658285.16 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 07:59:44,420 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5219, 1.5777, 1.7112, 1.3127], device='cuda:1'), covar=tensor([0.1699, 0.2492, 0.1447, 0.1719], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0719, 0.0986, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 07:59:56,952 INFO [optim.py:369] (1/2) 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 08:00:12,988 INFO [train.py:968] (1/2) Epoch 30, batch 7300, giga_loss[loss=0.3261, simple_loss=0.3845, pruned_loss=0.1339, over 28254.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3688, pruned_loss=0.1134, over 5678903.47 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3408, pruned_loss=0.08771, over 5618740.33 frames. ], giga_tot_loss[loss=0.3037, simple_loss=0.3727, pruned_loss=0.1174, over 5657582.38 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:00:57,643 INFO [train.py:968] (1/2) Epoch 30, batch 7350, giga_loss[loss=0.2787, simple_loss=0.3514, pruned_loss=0.103, over 28921.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3681, pruned_loss=0.1136, over 5684603.66 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.0877, over 5625835.50 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3721, pruned_loss=0.1174, over 5662758.59 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:01:22,262 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1327846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:01:24,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1327849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:01:29,672 INFO [optim.py:369] (1/2) 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 08:01:45,809 INFO [train.py:968] (1/2) Epoch 30, batch 7400, libri_loss[loss=0.2286, simple_loss=0.3155, pruned_loss=0.07085, over 29609.00 frames. ], tot_loss[loss=0.298, simple_loss=0.3674, pruned_loss=0.1143, over 5667685.82 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3406, pruned_loss=0.08773, over 5621257.02 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.371, pruned_loss=0.1178, over 5654757.56 frames. ], batch size: 74, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:01:53,407 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5428, 1.9823, 1.5511, 1.5536], device='cuda:1'), covar=tensor([0.2587, 0.2625, 0.3088, 0.2450], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1166, 0.1433, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 08:01:54,687 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1327878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:02:00,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4050, 3.5050, 1.5325, 1.5772], device='cuda:1'), covar=tensor([0.1027, 0.0346, 0.0920, 0.1383], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0416, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 08:02:19,549 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1327906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:02:30,974 INFO [train.py:968] (1/2) Epoch 30, batch 7450, giga_loss[loss=0.2668, simple_loss=0.3503, pruned_loss=0.09163, over 28974.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3648, pruned_loss=0.1133, over 5675724.89 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08782, over 5623252.41 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3681, pruned_loss=0.1165, over 5664435.12 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:03:02,062 INFO [optim.py:369] (1/2) 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 08:03:09,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4134, 1.3311, 1.5620, 1.1320], device='cuda:1'), covar=tensor([0.2025, 0.3319, 0.1576, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0720, 0.0987, 0.0887], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 08:03:14,510 INFO [train.py:968] (1/2) Epoch 30, batch 7500, giga_loss[loss=0.2611, simple_loss=0.3541, pruned_loss=0.08404, over 29035.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3631, pruned_loss=0.1118, over 5674302.69 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.34, pruned_loss=0.08752, over 5623030.61 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3671, pruned_loss=0.1155, over 5667646.88 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:03:59,783 INFO [train.py:968] (1/2) Epoch 30, batch 7550, giga_loss[loss=0.2892, simple_loss=0.3621, pruned_loss=0.1081, over 28549.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3633, pruned_loss=0.1105, over 5689324.44 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3406, pruned_loss=0.08766, over 5629207.91 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3666, pruned_loss=0.114, over 5680062.06 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:04:22,418 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328042.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:04:27,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1328049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:04:30,891 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:04:31,822 INFO [optim.py:369] (1/2) 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 08:04:42,111 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4894, 1.1587, 4.1271, 3.4413], device='cuda:1'), covar=tensor([0.1618, 0.2957, 0.0510, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0682, 0.1022, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 08:04:47,572 INFO [train.py:968] (1/2) Epoch 30, batch 7600, giga_loss[loss=0.2728, simple_loss=0.3454, pruned_loss=0.1001, over 28730.00 frames. ], tot_loss[loss=0.2935, simple_loss=0.3645, pruned_loss=0.1113, over 5692938.74 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08773, over 5630044.69 frames. ], giga_tot_loss[loss=0.2978, simple_loss=0.3673, pruned_loss=0.1142, over 5685568.96 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:04:59,329 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1328081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:05:04,184 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1328087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:05:21,781 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 08:05:28,670 INFO [train.py:968] (1/2) Epoch 30, batch 7650, giga_loss[loss=0.3065, simple_loss=0.3676, pruned_loss=0.1227, over 27531.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3647, pruned_loss=0.112, over 5693934.70 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08777, over 5631065.42 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.367, pruned_loss=0.1145, over 5687780.06 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:06:04,867 INFO [optim.py:369] (1/2) 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 08:06:14,345 INFO [train.py:968] (1/2) Epoch 30, batch 7700, giga_loss[loss=0.2437, simple_loss=0.3257, pruned_loss=0.08084, over 28991.00 frames. ], tot_loss[loss=0.2924, simple_loss=0.3624, pruned_loss=0.1112, over 5687868.90 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08774, over 5625937.46 frames. ], giga_tot_loss[loss=0.2962, simple_loss=0.3649, pruned_loss=0.1138, over 5688519.04 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:06:58,894 INFO [train.py:968] (1/2) Epoch 30, batch 7750, giga_loss[loss=0.2891, simple_loss=0.3592, pruned_loss=0.1094, over 28938.00 frames. ], tot_loss[loss=0.2933, simple_loss=0.3627, pruned_loss=0.112, over 5677742.60 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3406, pruned_loss=0.08773, over 5626543.52 frames. ], giga_tot_loss[loss=0.298, simple_loss=0.3657, pruned_loss=0.1151, over 5679295.95 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:07:09,719 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1328230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:07:11,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328233.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:07:25,254 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328248.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:07:32,310 INFO [optim.py:369] (1/2) 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 08:07:36,364 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 7800, giga_loss[loss=0.3165, simple_loss=0.3803, pruned_loss=0.1264, over 28965.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3626, pruned_loss=0.1127, over 5690009.89 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3404, pruned_loss=0.08749, over 5633578.96 frames. ], giga_tot_loss[loss=0.2992, simple_loss=0.3659, pruned_loss=0.1162, over 5686215.62 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:08:22,980 INFO [train.py:968] (1/2) Epoch 30, batch 7850, giga_loss[loss=0.2975, simple_loss=0.3702, pruned_loss=0.1124, over 28599.00 frames. ], tot_loss[loss=0.2914, simple_loss=0.36, pruned_loss=0.1113, over 5695145.31 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.34, pruned_loss=0.08721, over 5638189.23 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.3634, pruned_loss=0.1148, over 5689159.20 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:08:56,272 INFO [optim.py:369] (1/2) 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 08:08:56,588 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2983, 1.9486, 1.4825, 0.6551], device='cuda:1'), covar=tensor([0.6411, 0.3470, 0.4767, 0.7236], device='cuda:1'), in_proj_covar=tensor([0.1867, 0.1753, 0.1676, 0.1521], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 08:09:07,265 INFO [train.py:968] (1/2) Epoch 30, batch 7900, giga_loss[loss=0.3051, simple_loss=0.3672, pruned_loss=0.1215, over 28809.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3587, pruned_loss=0.1113, over 5700798.72 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3399, pruned_loss=0.08713, over 5643666.28 frames. ], giga_tot_loss[loss=0.2954, simple_loss=0.3618, pruned_loss=0.1145, over 5692245.64 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:09:49,352 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1328417.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:09:49,726 INFO [train.py:968] (1/2) Epoch 30, batch 7950, giga_loss[loss=0.3278, simple_loss=0.3898, pruned_loss=0.1329, over 29049.00 frames. ], tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5696087.89 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3398, pruned_loss=0.08707, over 5646426.39 frames. ], giga_tot_loss[loss=0.2957, simple_loss=0.3619, pruned_loss=0.1148, over 5687481.92 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:10:24,887 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 8000, giga_loss[loss=0.2842, simple_loss=0.3538, pruned_loss=0.1073, over 28805.00 frames. ], tot_loss[loss=0.2926, simple_loss=0.3606, pruned_loss=0.1123, over 5693892.39 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.34, pruned_loss=0.08706, over 5651491.30 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.363, pruned_loss=0.1151, over 5683117.78 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:10:52,204 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:11:03,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=4.44 vs. limit=5.0
+2023-03-15 08:11:13,499 INFO [train.py:968] (1/2) Epoch 30, batch 8050, giga_loss[loss=0.3267, simple_loss=0.383, pruned_loss=0.1352, over 28744.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3591, pruned_loss=0.1106, over 5685534.27 frames. ], libri_tot_loss[loss=0.2564, simple_loss=0.3393, pruned_loss=0.08672, over 5657597.94 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.362, pruned_loss=0.1137, over 5672366.32 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:11:45,569 INFO [optim.py:369] (1/2) 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 08:11:50,175 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1328560.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:11:52,336 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328563.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:11:52,604 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.79 vs. limit=2.0
+2023-03-15 08:11:55,533 INFO [train.py:968] (1/2) Epoch 30, batch 8100, giga_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 28278.00 frames. ], tot_loss[loss=0.2917, simple_loss=0.3606, pruned_loss=0.1114, over 5678192.36 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3391, pruned_loss=0.08662, over 5659894.74 frames. ], giga_tot_loss[loss=0.2958, simple_loss=0.3633, pruned_loss=0.1142, over 5665927.40 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:12:16,170 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1328592.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:12:35,719 INFO [train.py:968] (1/2) Epoch 30, batch 8150, giga_loss[loss=0.3537, simple_loss=0.4123, pruned_loss=0.1475, over 28452.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.3621, pruned_loss=0.1122, over 5687897.60 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3395, pruned_loss=0.08688, over 5663667.21 frames. ], giga_tot_loss[loss=0.2972, simple_loss=0.3646, pruned_loss=0.1149, over 5675620.17 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:12:40,545 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1328623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:12:42,248 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.05 vs. limit=5.0
+2023-03-15 08:13:10,802 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 8200, libri_loss[loss=0.2084, simple_loss=0.2935, pruned_loss=0.0617, over 28647.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.365, pruned_loss=0.1152, over 5680917.86 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.34, pruned_loss=0.08716, over 5669570.02 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3674, pruned_loss=0.1179, over 5666309.96 frames. ], batch size: 63, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:13:31,523 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-15 08:13:49,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.01 vs. limit=5.0
+2023-03-15 08:14:08,272 INFO [train.py:968] (1/2) Epoch 30, batch 8250, giga_loss[loss=0.3181, simple_loss=0.3575, pruned_loss=0.1394, over 23660.00 frames. ], tot_loss[loss=0.2991, simple_loss=0.3653, pruned_loss=0.1165, over 5683457.30 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3396, pruned_loss=0.08685, over 5672184.02 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3679, pruned_loss=0.1193, over 5669849.35 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:14:21,110 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4773, 1.5970, 1.6105, 1.4187], device='cuda:1'), covar=tensor([0.2893, 0.2533, 0.2150, 0.2597], device='cuda:1'), in_proj_covar=tensor([0.2096, 0.2069, 0.1974, 0.2111], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 08:14:41,679 INFO [optim.py:369] (1/2) 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 08:14:50,231 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 8300, giga_loss[loss=0.4141, simple_loss=0.4418, pruned_loss=0.1932, over 24072.00 frames. ], tot_loss[loss=0.3063, simple_loss=0.37, pruned_loss=0.1213, over 5673015.37 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3398, pruned_loss=0.0871, over 5677532.00 frames. ], giga_tot_loss[loss=0.3103, simple_loss=0.3725, pruned_loss=0.124, over 5657718.39 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:14:53,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1328769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:15:20,512 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 8350, giga_loss[loss=0.3006, simple_loss=0.3609, pruned_loss=0.1202, over 28916.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3703, pruned_loss=0.1222, over 5673121.09 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3398, pruned_loss=0.0872, over 5680284.18 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3726, pruned_loss=0.1247, over 5658677.07 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:15:39,129 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-15 08:15:47,317 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1328830.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:16:09,220 INFO [optim.py:369] (1/2) 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 08:16:15,846 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 8400, giga_loss[loss=0.3282, simple_loss=0.372, pruned_loss=0.1422, over 28655.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3683, pruned_loss=0.1207, over 5678938.54 frames. ], libri_tot_loss[loss=0.2565, simple_loss=0.3393, pruned_loss=0.08684, over 5688982.28 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3717, pruned_loss=0.1243, over 5659435.70 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:16:58,541 INFO [train.py:968] (1/2) Epoch 30, batch 8450, giga_loss[loss=0.2944, simple_loss=0.3701, pruned_loss=0.1093, over 28499.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3693, pruned_loss=0.1198, over 5682482.92 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3396, pruned_loss=0.08691, over 5689715.30 frames. ], giga_tot_loss[loss=0.3094, simple_loss=0.3723, pruned_loss=0.1232, over 5666104.88 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:17:30,702 INFO [optim.py:369] (1/2) 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 08:17:39,685 INFO [train.py:968] (1/2) Epoch 30, batch 8500, giga_loss[loss=0.2951, simple_loss=0.3674, pruned_loss=0.1114, over 28980.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.367, pruned_loss=0.1174, over 5679674.52 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3394, pruned_loss=0.08689, over 5695698.26 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3702, pruned_loss=0.121, over 5661000.53 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:17:44,777 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 08:18:13,575 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:18:16,108 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 8550, giga_loss[loss=0.3155, simple_loss=0.378, pruned_loss=0.1265, over 28765.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3644, pruned_loss=0.116, over 5686002.51 frames. ], libri_tot_loss[loss=0.2566, simple_loss=0.3394, pruned_loss=0.08688, over 5699129.67 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3675, pruned_loss=0.1195, over 5667667.63 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:18:24,642 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:18:38,767 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:18:45,461 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.77 vs. limit=5.0
+2023-03-15 08:18:54,730 INFO [optim.py:369] (1/2) 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 08:18:59,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5288, 2.2328, 1.6297, 0.7246], device='cuda:1'), covar=tensor([0.6391, 0.3285, 0.4234, 0.6876], device='cuda:1'), in_proj_covar=tensor([0.1866, 0.1758, 0.1678, 0.1525], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 08:19:02,645 INFO [train.py:968] (1/2) Epoch 30, batch 8600, giga_loss[loss=0.3224, simple_loss=0.3876, pruned_loss=0.1286, over 28996.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3629, pruned_loss=0.1152, over 5687025.38 frames. ], libri_tot_loss[loss=0.2572, simple_loss=0.3401, pruned_loss=0.08716, over 5702991.42 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3653, pruned_loss=0.1184, over 5668484.06 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:19:50,492 INFO [train.py:968] (1/2) Epoch 30, batch 8650, giga_loss[loss=0.3021, simple_loss=0.3793, pruned_loss=0.1125, over 29069.00 frames. ], tot_loss[loss=0.2982, simple_loss=0.3636, pruned_loss=0.1164, over 5664345.94 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3401, pruned_loss=0.08729, over 5706458.90 frames. ], giga_tot_loss[loss=0.3025, simple_loss=0.366, pruned_loss=0.1195, over 5646324.81 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:20:26,057 INFO [optim.py:369] (1/2) 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 08:20:35,750 INFO [train.py:968] (1/2) Epoch 30, batch 8700, giga_loss[loss=0.3289, simple_loss=0.4043, pruned_loss=0.1267, over 28866.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3666, pruned_loss=0.1172, over 5664417.56 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3402, pruned_loss=0.08745, over 5700996.63 frames. ], giga_tot_loss[loss=0.3044, simple_loss=0.3689, pruned_loss=0.12, over 5655056.56 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:20:52,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5944, 1.5455, 1.7724, 1.3861], device='cuda:1'), covar=tensor([0.1736, 0.2406, 0.1438, 0.1773], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0722, 0.0991, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 08:20:55,958 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5204, 3.7044, 1.6912, 1.7302], device='cuda:1'), covar=tensor([0.1002, 0.0455, 0.0938, 0.1281], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0416, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 08:21:06,406 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1329205.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:21:16,468 INFO [train.py:968] (1/2) Epoch 30, batch 8750, giga_loss[loss=0.3659, simple_loss=0.4083, pruned_loss=0.1618, over 26623.00 frames. ], tot_loss[loss=0.3001, simple_loss=0.3683, pruned_loss=0.1159, over 5676526.88 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08767, over 5710369.61 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3711, pruned_loss=0.1193, over 5658976.84 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:21:51,779 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 8800, libri_loss[loss=0.2786, simple_loss=0.3597, pruned_loss=0.0987, over 29377.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3705, pruned_loss=0.1164, over 5677578.37 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3407, pruned_loss=0.08774, over 5712160.07 frames. ], giga_tot_loss[loss=0.3057, simple_loss=0.3728, pruned_loss=0.1193, over 5661862.00 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:22:19,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:22:40,731 INFO [train.py:968] (1/2) Epoch 30, batch 8850, giga_loss[loss=0.3001, simple_loss=0.3687, pruned_loss=0.1158, over 29066.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3728, pruned_loss=0.1186, over 5675832.25 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3404, pruned_loss=0.08761, over 5714211.23 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3752, pruned_loss=0.1213, over 5661261.99 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:23:06,448 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329348.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:23:09,051 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7077, 1.6396, 1.9226, 1.5120], device='cuda:1'), covar=tensor([0.1439, 0.2063, 0.1185, 0.1558], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0722, 0.0991, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 08:23:09,558 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1329351.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:23:14,398 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 8900, libri_loss[loss=0.2127, simple_loss=0.2974, pruned_loss=0.06398, over 29347.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3741, pruned_loss=0.1206, over 5655344.02 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08781, over 5708165.72 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3764, pruned_loss=0.1231, over 5648051.08 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:23:26,541 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-15 08:23:28,453 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 08:23:31,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2265, 2.5635, 2.0696, 2.2069], device='cuda:1'), covar=tensor([0.2689, 0.2733, 0.3236, 0.2663], device='cuda:1'), in_proj_covar=tensor([0.1621, 0.1168, 0.1435, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 08:23:33,492 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329380.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:23:45,691 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1329397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:24:02,101 INFO [train.py:968] (1/2) Epoch 30, batch 8950, giga_loss[loss=0.2947, simple_loss=0.3592, pruned_loss=0.1151, over 28450.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3709, pruned_loss=0.1187, over 5661680.05 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3405, pruned_loss=0.08773, over 5712412.61 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3742, pruned_loss=0.1221, over 5650081.97 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:24:40,632 INFO [optim.py:369] (1/2) 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 08:24:50,009 INFO [train.py:968] (1/2) Epoch 30, batch 9000, giga_loss[loss=0.2951, simple_loss=0.3696, pruned_loss=0.1103, over 28882.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3695, pruned_loss=0.1189, over 5649430.92 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3401, pruned_loss=0.08756, over 5714919.81 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3729, pruned_loss=0.1223, over 5636962.73 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:24:50,010 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 08:24:58,707 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 08:25:17,550 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.60 vs. limit=5.0
+2023-03-15 08:25:42,097 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.23 vs. limit=5.0
+2023-03-15 08:25:42,871 INFO [train.py:968] (1/2) Epoch 30, batch 9050, giga_loss[loss=0.2927, simple_loss=0.3586, pruned_loss=0.1134, over 28827.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3685, pruned_loss=0.1191, over 5659787.97 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3402, pruned_loss=0.08755, over 5715831.21 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3712, pruned_loss=0.1219, over 5649011.96 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:25:47,197 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2306, 1.2805, 3.2043, 2.9040], device='cuda:1'), covar=tensor([0.1493, 0.2614, 0.0508, 0.2112], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0686, 0.1026, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 08:25:55,737 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.58 vs. limit=5.0
+2023-03-15 08:26:01,642 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329540.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:26:04,892 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1329543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:26:15,664 INFO [optim.py:369] (1/2) 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 08:26:24,106 INFO [train.py:968] (1/2) Epoch 30, batch 9100, giga_loss[loss=0.2998, simple_loss=0.3633, pruned_loss=0.1182, over 28850.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3683, pruned_loss=0.1193, over 5671345.89 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3406, pruned_loss=0.08786, over 5721773.49 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3712, pruned_loss=0.1224, over 5655134.53 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:26:27,576 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:26:33,164 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6155, 2.0789, 1.4587, 1.5433], device='cuda:1'), covar=tensor([0.1194, 0.0604, 0.1078, 0.1224], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0456, 0.0530, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 08:27:10,864 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2711, 1.4362, 1.3994, 1.4215], device='cuda:1'), covar=tensor([0.0766, 0.0389, 0.0334, 0.0874], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 08:27:11,189 INFO [train.py:968] (1/2) Epoch 30, batch 9150, giga_loss[loss=0.3245, simple_loss=0.3789, pruned_loss=0.135, over 28595.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3693, pruned_loss=0.1208, over 5653247.01 frames. ], libri_tot_loss[loss=0.2581, simple_loss=0.3406, pruned_loss=0.08783, over 5723304.15 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3721, pruned_loss=0.124, over 5637721.11 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:27:44,282 INFO [optim.py:369] (1/2) 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 08:27:48,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1329664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:27:50,953 INFO [train.py:968] (1/2) Epoch 30, batch 9200, libri_loss[loss=0.2276, simple_loss=0.3113, pruned_loss=0.0719, over 29541.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3674, pruned_loss=0.1197, over 5658134.43 frames. ], libri_tot_loss[loss=0.2583, simple_loss=0.3408, pruned_loss=0.08792, over 5718232.97 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3705, pruned_loss=0.1234, over 5648384.89 frames. ], batch size: 79, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:28:33,020 INFO [train.py:968] (1/2) Epoch 30, batch 9250, giga_loss[loss=0.3021, simple_loss=0.3719, pruned_loss=0.1162, over 28543.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3644, pruned_loss=0.1181, over 5660498.04 frames. ], libri_tot_loss[loss=0.2585, simple_loss=0.3407, pruned_loss=0.08817, over 5724206.79 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3677, pruned_loss=0.1217, over 5645901.60 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:29:01,568 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:29:12,246 INFO [train.py:968] (1/2) Epoch 30, batch 9300, giga_loss[loss=0.3588, simple_loss=0.3993, pruned_loss=0.1591, over 26591.00 frames. ], tot_loss[loss=0.2998, simple_loss=0.3648, pruned_loss=0.1174, over 5663035.44 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3411, pruned_loss=0.08827, over 5726998.81 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.3674, pruned_loss=0.1205, over 5648089.18 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:29:37,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4687, 2.0819, 1.5381, 0.7673], device='cuda:1'), covar=tensor([0.6976, 0.3614, 0.4800, 0.7509], device='cuda:1'), in_proj_covar=tensor([0.1866, 0.1754, 0.1675, 0.1521], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 08:29:47,469 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1329807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:29:49,435 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 9350, giga_loss[loss=0.2844, simple_loss=0.3579, pruned_loss=0.1055, over 28895.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3673, pruned_loss=0.118, over 5656947.69 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3416, pruned_loss=0.08843, over 5718292.39 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3696, pruned_loss=0.1212, over 5650299.00 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:29:58,707 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:30:09,418 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3295, 1.1726, 1.0948, 1.4801], device='cuda:1'), covar=tensor([0.0797, 0.0373, 0.0372, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 08:30:13,185 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1329839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:30:21,033 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1329848.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:30:30,381 INFO [optim.py:369] (1/2) 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 08:30:36,306 INFO [train.py:968] (1/2) Epoch 30, batch 9400, giga_loss[loss=0.3149, simple_loss=0.3729, pruned_loss=0.1285, over 28294.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.368, pruned_loss=0.1187, over 5656612.52 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08845, over 5726041.98 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3707, pruned_loss=0.1223, over 5642153.26 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:30:50,861 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.70 vs. limit=2.0
+2023-03-15 08:31:16,822 INFO [train.py:968] (1/2) Epoch 30, batch 9450, giga_loss[loss=0.3028, simple_loss=0.3915, pruned_loss=0.107, over 28786.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3689, pruned_loss=0.1184, over 5662705.56 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3419, pruned_loss=0.08846, over 5729004.38 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3715, pruned_loss=0.122, over 5646938.39 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:31:51,155 INFO [optim.py:369] (1/2) 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 08:31:58,735 INFO [train.py:968] (1/2) Epoch 30, batch 9500, giga_loss[loss=0.279, simple_loss=0.3691, pruned_loss=0.09443, over 28932.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.37, pruned_loss=0.1168, over 5670958.18 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3422, pruned_loss=0.08855, over 5731532.85 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3724, pruned_loss=0.1201, over 5654971.02 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:32:37,769 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:32:38,209 INFO [train.py:968] (1/2) Epoch 30, batch 9550, giga_loss[loss=0.2902, simple_loss=0.3676, pruned_loss=0.1064, over 28972.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3727, pruned_loss=0.1168, over 5681569.53 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3421, pruned_loss=0.08849, over 5731945.19 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3751, pruned_loss=0.1199, over 5667851.35 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:32:54,169 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:33:16,644 INFO [optim.py:369] (1/2) 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 08:33:22,367 INFO [train.py:968] (1/2) Epoch 30, batch 9600, giga_loss[loss=0.2789, simple_loss=0.3569, pruned_loss=0.1004, over 28876.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3767, pruned_loss=0.1202, over 5673035.58 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08848, over 5733000.61 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.379, pruned_loss=0.1229, over 5660745.43 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:33:47,174 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 08:33:57,846 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.04 vs. limit=5.0
+2023-03-15 08:34:05,022 INFO [train.py:968] (1/2) Epoch 30, batch 9650, giga_loss[loss=0.3244, simple_loss=0.3982, pruned_loss=0.1253, over 28860.00 frames. ], tot_loss[loss=0.3123, simple_loss=0.3788, pruned_loss=0.1229, over 5675666.79 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3421, pruned_loss=0.08858, over 5731743.42 frames. ], giga_tot_loss[loss=0.3159, simple_loss=0.381, pruned_loss=0.1254, over 5666474.77 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:34:15,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:34:42,475 INFO [optim.py:369] (1/2) 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 08:34:44,472 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:34:49,202 INFO [train.py:968] (1/2) Epoch 30, batch 9700, giga_loss[loss=0.2932, simple_loss=0.3545, pruned_loss=0.116, over 28697.00 frames. ], tot_loss[loss=0.3124, simple_loss=0.378, pruned_loss=0.1233, over 5662878.04 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.08845, over 5727421.67 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3811, pruned_loss=0.1265, over 5657892.19 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:35:02,312 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330182.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:35:15,486 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 9750, giga_loss[loss=0.3192, simple_loss=0.391, pruned_loss=0.1237, over 29043.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3769, pruned_loss=0.1226, over 5664181.84 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08841, over 5730224.92 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3804, pruned_loss=0.126, over 5656026.57 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:35:37,713 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330223.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 08:35:52,695 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 08:36:01,532 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:36:14,447 INFO [train.py:968] (1/2) Epoch 30, batch 9800, giga_loss[loss=0.2896, simple_loss=0.3717, pruned_loss=0.1037, over 29013.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3751, pruned_loss=0.1198, over 5674056.04 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08842, over 5734372.57 frames. ], giga_tot_loss[loss=0.313, simple_loss=0.3789, pruned_loss=0.1235, over 5662200.85 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:36:15,524 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4786, 1.8038, 1.6762, 1.6164], device='cuda:1'), covar=tensor([0.2145, 0.2069, 0.2386, 0.2134], device='cuda:1'), in_proj_covar=tensor([0.0520, 0.0768, 0.0740, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 08:36:16,904 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:36:20,076 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:36:44,601 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:36:58,019 INFO [train.py:968] (1/2) Epoch 30, batch 9850, giga_loss[loss=0.3422, simple_loss=0.3798, pruned_loss=0.1523, over 23387.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3749, pruned_loss=0.1188, over 5674025.94 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3417, pruned_loss=0.08839, over 5737859.81 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.3788, pruned_loss=0.1225, over 5659944.76 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:37:15,100 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:37:18,706 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 08:37:41,914 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330366.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:37:42,953 INFO [train.py:968] (1/2) Epoch 30, batch 9900, giga_loss[loss=0.299, simple_loss=0.37, pruned_loss=0.114, over 28833.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3745, pruned_loss=0.1179, over 5670249.35 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.0885, over 5729511.81 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3785, pruned_loss=0.1217, over 5664616.42 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:37:43,899 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330369.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:37:46,331 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:38:02,627 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:38:08,299 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330398.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 08:38:17,788 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 9950, giga_loss[loss=0.291, simple_loss=0.3666, pruned_loss=0.1077, over 28689.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3751, pruned_loss=0.1192, over 5672643.11 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3414, pruned_loss=0.0883, over 5736984.16 frames. ], giga_tot_loss[loss=0.314, simple_loss=0.3802, pruned_loss=0.1239, over 5658738.14 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:39:06,345 INFO [optim.py:369] (1/2) 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 08:39:13,088 INFO [train.py:968] (1/2) Epoch 30, batch 10000, giga_loss[loss=0.3077, simple_loss=0.3726, pruned_loss=0.1214, over 28556.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3745, pruned_loss=0.1195, over 5668207.87 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3415, pruned_loss=0.08836, over 5738519.73 frames. ], giga_tot_loss[loss=0.3133, simple_loss=0.3791, pruned_loss=0.1237, over 5654570.67 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:39:58,421 INFO [train.py:968] (1/2) Epoch 30, batch 10050, giga_loss[loss=0.2683, simple_loss=0.343, pruned_loss=0.09682, over 28862.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3736, pruned_loss=0.1204, over 5654409.51 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.3422, pruned_loss=0.08874, over 5735840.25 frames. ], giga_tot_loss[loss=0.3136, simple_loss=0.3779, pruned_loss=0.1246, over 5642848.86 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:40:17,311 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:18,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:19,522 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330538.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:29,190 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:34,766 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:37,311 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:37,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330557.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:40,024 INFO [optim.py:369] (1/2) 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 08:40:46,780 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:40:47,169 INFO [train.py:968] (1/2) Epoch 30, batch 10100, giga_loss[loss=0.2852, simple_loss=0.3619, pruned_loss=0.1042, over 28957.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3717, pruned_loss=0.1194, over 5669656.34 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3426, pruned_loss=0.08892, over 5739596.42 frames. ], giga_tot_loss[loss=0.3108, simple_loss=0.3752, pruned_loss=0.1232, over 5655824.70 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:41:02,630 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:41:30,717 INFO [train.py:968] (1/2) Epoch 30, batch 10150, giga_loss[loss=0.2986, simple_loss=0.3703, pruned_loss=0.1135, over 29093.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3687, pruned_loss=0.1182, over 5663289.02 frames. ], libri_tot_loss[loss=0.2608, simple_loss=0.3429, pruned_loss=0.08934, over 5741900.33 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3727, pruned_loss=0.1223, over 5646133.44 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:41:44,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:12,006 INFO [optim.py:369] (1/2) 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 08:42:18,070 INFO [train.py:968] (1/2) Epoch 30, batch 10200, giga_loss[loss=0.2913, simple_loss=0.3593, pruned_loss=0.1116, over 28775.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3674, pruned_loss=0.1182, over 5667937.13 frames. ], libri_tot_loss[loss=0.2602, simple_loss=0.3423, pruned_loss=0.08907, over 5747306.46 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3718, pruned_loss=0.1225, over 5647194.39 frames. ], batch size: 66, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:42:31,788 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:34,532 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:46,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:49,852 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:42:56,949 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:43:02,229 INFO [train.py:968] (1/2) Epoch 30, batch 10250, libri_loss[loss=0.2523, simple_loss=0.3392, pruned_loss=0.08274, over 29459.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.3661, pruned_loss=0.1171, over 5672422.80 frames. ], libri_tot_loss[loss=0.2599, simple_loss=0.342, pruned_loss=0.08883, over 5753990.44 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.371, pruned_loss=0.1221, over 5646195.10 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:43:06,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1543, 3.3859, 2.2087, 1.1767], device='cuda:1'), covar=tensor([0.9738, 0.4049, 0.4527, 0.8700], device='cuda:1'), in_proj_covar=tensor([0.1860, 0.1748, 0.1666, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 08:43:14,705 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:43:39,880 INFO [optim.py:369] (1/2) 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 08:43:45,169 INFO [train.py:968] (1/2) Epoch 30, batch 10300, giga_loss[loss=0.2647, simple_loss=0.3451, pruned_loss=0.09219, over 29052.00 frames. ], tot_loss[loss=0.2946, simple_loss=0.3628, pruned_loss=0.1133, over 5689500.43 frames. ], libri_tot_loss[loss=0.2597, simple_loss=0.3418, pruned_loss=0.08881, over 5757783.34 frames. ], giga_tot_loss[loss=0.3015, simple_loss=0.3674, pruned_loss=0.1178, over 5663583.75 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:43:49,092 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1330772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:43:50,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1330775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:44:15,039 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1330804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:44:30,389 INFO [train.py:968] (1/2) Epoch 30, batch 10350, giga_loss[loss=0.2465, simple_loss=0.3258, pruned_loss=0.08357, over 29010.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3587, pruned_loss=0.1099, over 5679528.75 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3414, pruned_loss=0.08858, over 5762513.26 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3635, pruned_loss=0.1147, over 5651628.93 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 08:44:52,244 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-15 08:44:59,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4304, 5.0544, 1.6966, 1.8360], device='cuda:1'), covar=tensor([0.1280, 0.0374, 0.1009, 0.1562], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0580, 0.0416, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 08:45:00,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6362, 1.9110, 1.3929, 1.4536], device='cuda:1'), covar=tensor([0.1099, 0.0616, 0.1025, 0.1179], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0453, 0.0526, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 08:45:11,607 INFO [optim.py:369] (1/2) 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 08:45:16,103 INFO [train.py:968] (1/2) Epoch 30, batch 10400, giga_loss[loss=0.29, simple_loss=0.3596, pruned_loss=0.1102, over 28703.00 frames. ], tot_loss[loss=0.2896, simple_loss=0.3594, pruned_loss=0.1099, over 5674830.24 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3416, pruned_loss=0.08861, over 5753210.41 frames. ], giga_tot_loss[loss=0.296, simple_loss=0.3635, pruned_loss=0.1142, over 5659052.79 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:45:17,174 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3505, 2.0894, 1.7035, 1.6132], device='cuda:1'), covar=tensor([0.0819, 0.0284, 0.0326, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0230], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 08:45:41,646 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 10450, giga_loss[loss=0.273, simple_loss=0.3371, pruned_loss=0.1045, over 28947.00 frames. ], tot_loss[loss=0.2888, simple_loss=0.358, pruned_loss=0.1098, over 5676368.90 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.08848, over 5757139.70 frames. ], giga_tot_loss[loss=0.2955, simple_loss=0.3622, pruned_loss=0.1144, over 5656631.50 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:46:16,475 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1330932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:46:32,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8004, 1.8029, 1.9962, 1.5412], device='cuda:1'), covar=tensor([0.1996, 0.2549, 0.1561, 0.1842], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0722, 0.0991, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 08:46:43,193 INFO [optim.py:369] (1/2) 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 08:46:48,854 INFO [train.py:968] (1/2) Epoch 30, batch 10500, giga_loss[loss=0.2951, simple_loss=0.3561, pruned_loss=0.117, over 28716.00 frames. ], tot_loss[loss=0.2878, simple_loss=0.3562, pruned_loss=0.1097, over 5674693.94 frames. ], libri_tot_loss[loss=0.2591, simple_loss=0.3414, pruned_loss=0.08843, over 5759495.70 frames. ], giga_tot_loss[loss=0.2935, simple_loss=0.3598, pruned_loss=0.1137, over 5655950.40 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:47:01,986 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6777, 1.7549, 1.8971, 1.4438], device='cuda:1'), covar=tensor([0.1568, 0.2385, 0.1297, 0.1645], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0722, 0.0991, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 08:47:15,357 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1330998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:47:29,073 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5654, 1.7534, 1.7361, 1.5113], device='cuda:1'), covar=tensor([0.2579, 0.2317, 0.2526, 0.2545], device='cuda:1'), in_proj_covar=tensor([0.2104, 0.2069, 0.1984, 0.2120], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 08:47:34,063 INFO [train.py:968] (1/2) Epoch 30, batch 10550, libri_loss[loss=0.2383, simple_loss=0.3175, pruned_loss=0.07954, over 29347.00 frames. ], tot_loss[loss=0.2903, simple_loss=0.3586, pruned_loss=0.111, over 5669755.33 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.341, pruned_loss=0.08822, over 5752550.07 frames. ], giga_tot_loss[loss=0.2959, simple_loss=0.3622, pruned_loss=0.1149, over 5658336.10 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:48:10,761 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4705, 1.6675, 1.1713, 1.2378], device='cuda:1'), covar=tensor([0.1077, 0.0560, 0.1104, 0.1209], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0453, 0.0527, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 08:48:12,700 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 10600, giga_loss[loss=0.2527, simple_loss=0.3307, pruned_loss=0.08731, over 28643.00 frames. ], tot_loss[loss=0.2929, simple_loss=0.3612, pruned_loss=0.1123, over 5666002.86 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3411, pruned_loss=0.08824, over 5754827.26 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.3643, pruned_loss=0.1159, over 5652946.67 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:48:25,652 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:48:27,433 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:48:56,433 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1331107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:49:07,111 INFO [train.py:968] (1/2) Epoch 30, batch 10650, giga_loss[loss=0.3301, simple_loss=0.3845, pruned_loss=0.1378, over 28480.00 frames. ], tot_loss[loss=0.2941, simple_loss=0.3622, pruned_loss=0.113, over 5661664.77 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3415, pruned_loss=0.0884, over 5760046.60 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.365, pruned_loss=0.1165, over 5643838.87 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:49:47,511 INFO [optim.py:369] (1/2) 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 08:49:53,484 INFO [train.py:968] (1/2) Epoch 30, batch 10700, giga_loss[loss=0.3007, simple_loss=0.368, pruned_loss=0.1167, over 28896.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3623, pruned_loss=0.1136, over 5663420.08 frames. ], libri_tot_loss[loss=0.2594, simple_loss=0.3418, pruned_loss=0.08852, over 5763367.36 frames. ], giga_tot_loss[loss=0.2991, simple_loss=0.3648, pruned_loss=0.1167, over 5644674.67 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:50:16,313 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9688, 3.7743, 3.6257, 1.9004], device='cuda:1'), covar=tensor([0.0768, 0.0932, 0.0889, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.1340, 0.1235, 0.1038, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 08:50:36,618 INFO [train.py:968] (1/2) Epoch 30, batch 10750, giga_loss[loss=0.3183, simple_loss=0.3854, pruned_loss=0.1255, over 27989.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3639, pruned_loss=0.1146, over 5673650.00 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08853, over 5767794.65 frames. ], giga_tot_loss[loss=0.3011, simple_loss=0.3664, pruned_loss=0.118, over 5651557.85 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:51:20,671 INFO [optim.py:369] (1/2) 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 08:51:26,027 INFO [train.py:968] (1/2) Epoch 30, batch 10800, libri_loss[loss=0.2581, simple_loss=0.349, pruned_loss=0.08362, over 29119.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3659, pruned_loss=0.1157, over 5669012.87 frames. ], libri_tot_loss[loss=0.2592, simple_loss=0.3418, pruned_loss=0.08835, over 5769149.59 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.369, pruned_loss=0.1196, over 5646168.03 frames. ], batch size: 101, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:51:28,368 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:51:28,454 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2797, 1.5584, 1.5031, 1.2200], device='cuda:1'), covar=tensor([0.3161, 0.2675, 0.1892, 0.2580], device='cuda:1'), in_proj_covar=tensor([0.2108, 0.2075, 0.1990, 0.2128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 08:52:09,807 INFO [train.py:968] (1/2) Epoch 30, batch 10850, giga_loss[loss=0.3105, simple_loss=0.377, pruned_loss=0.122, over 28666.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3671, pruned_loss=0.1163, over 5678068.68 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.3413, pruned_loss=0.08796, over 5773186.95 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3709, pruned_loss=0.1207, over 5652972.97 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:52:13,712 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1922, 0.9571, 0.9646, 1.3547], device='cuda:1'), covar=tensor([0.0756, 0.0374, 0.0352, 0.0865], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 08:52:21,487 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-15 08:52:51,139 INFO [optim.py:369] (1/2) 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 08:52:56,548 INFO [train.py:968] (1/2) Epoch 30, batch 10900, giga_loss[loss=0.2964, simple_loss=0.3708, pruned_loss=0.111, over 28864.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.369, pruned_loss=0.1179, over 5686563.01 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3411, pruned_loss=0.08783, over 5774123.34 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3725, pruned_loss=0.1219, over 5664683.30 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:53:01,994 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:53:38,925 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5131, 1.8235, 1.4507, 1.7289], device='cuda:1'), covar=tensor([0.2807, 0.2881, 0.3283, 0.2461], device='cuda:1'), in_proj_covar=tensor([0.1622, 0.1170, 0.1437, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 08:53:42,685 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:53:44,954 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:53:45,323 INFO [train.py:968] (1/2) Epoch 30, batch 10950, giga_loss[loss=0.2918, simple_loss=0.3638, pruned_loss=0.1099, over 28938.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3698, pruned_loss=0.1187, over 5684880.98 frames. ], libri_tot_loss[loss=0.2582, simple_loss=0.3409, pruned_loss=0.08772, over 5776150.57 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3731, pruned_loss=0.1225, over 5664374.89 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:54:14,615 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1331446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:54:25,726 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.46 vs. limit=2.0
+2023-03-15 08:54:29,033 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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] (1/2) Epoch 30, batch 11000, giga_loss[loss=0.2845, simple_loss=0.3614, pruned_loss=0.1038, over 28909.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3702, pruned_loss=0.1179, over 5676502.76 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08758, over 5778456.87 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3737, pruned_loss=0.1216, over 5656446.41 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:54:42,608 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:55:17,256 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 11050, giga_loss[loss=0.2874, simple_loss=0.357, pruned_loss=0.1089, over 29011.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3689, pruned_loss=0.1174, over 5676076.36 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08766, over 5783403.16 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3732, pruned_loss=0.1217, over 5650779.71 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:55:20,423 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:55:48,162 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1331548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:56:05,700 INFO [optim.py:369] (1/2) 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 08:56:07,975 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.14 vs. limit=5.0
+2023-03-15 08:56:12,146 INFO [train.py:968] (1/2) Epoch 30, batch 11100, giga_loss[loss=0.2648, simple_loss=0.34, pruned_loss=0.09479, over 28799.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3698, pruned_loss=0.1197, over 5653987.07 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3402, pruned_loss=0.0875, over 5783797.82 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3738, pruned_loss=0.1235, over 5632830.46 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:57:05,417 INFO [train.py:968] (1/2) Epoch 30, batch 11150, giga_loss[loss=0.2415, simple_loss=0.3162, pruned_loss=0.08335, over 28707.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.369, pruned_loss=0.1193, over 5641074.29 frames. ], libri_tot_loss[loss=0.2586, simple_loss=0.341, pruned_loss=0.08806, over 5764511.47 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3725, pruned_loss=0.123, over 5636229.90 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:57:48,723 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 11200, giga_loss[loss=0.2584, simple_loss=0.3322, pruned_loss=0.09227, over 28879.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.3669, pruned_loss=0.1184, over 5646808.83 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3412, pruned_loss=0.08816, over 5765231.91 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3697, pruned_loss=0.1213, over 5641923.93 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:58:07,140 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:58:38,152 INFO [train.py:968] (1/2) Epoch 30, batch 11250, giga_loss[loss=0.2943, simple_loss=0.3697, pruned_loss=0.1095, over 28784.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3674, pruned_loss=0.1191, over 5651941.40 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3411, pruned_loss=0.08808, over 5766376.90 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3698, pruned_loss=0.1218, over 5646102.17 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 08:59:21,692 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 11300, giga_loss[loss=0.3261, simple_loss=0.3913, pruned_loss=0.1305, over 28709.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3675, pruned_loss=0.1198, over 5653049.70 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3413, pruned_loss=0.08821, over 5768709.25 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3697, pruned_loss=0.1224, over 5644232.74 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 08:59:25,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3705, 3.2879, 1.5672, 1.5783], device='cuda:1'), covar=tensor([0.1021, 0.0337, 0.0880, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0415, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 08:59:53,255 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 08:59:56,108 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.57 vs. limit=2.0
+2023-03-15 09:00:13,041 INFO [train.py:968] (1/2) Epoch 30, batch 11350, giga_loss[loss=0.2912, simple_loss=0.3566, pruned_loss=0.1129, over 28565.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3671, pruned_loss=0.1195, over 5653300.44 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3413, pruned_loss=0.08814, over 5766560.53 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3692, pruned_loss=0.122, over 5646681.56 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:00:27,404 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:00:32,294 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:00:44,418 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1331850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:00:56,655 INFO [optim.py:369] (1/2) 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 09:00:59,173 INFO [train.py:968] (1/2) Epoch 30, batch 11400, giga_loss[loss=0.2788, simple_loss=0.3541, pruned_loss=0.1018, over 28833.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3702, pruned_loss=0.1221, over 5658775.62 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3413, pruned_loss=0.0881, over 5768743.17 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3722, pruned_loss=0.1245, over 5650208.23 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:01:49,646 INFO [train.py:968] (1/2) Epoch 30, batch 11450, giga_loss[loss=0.268, simple_loss=0.3328, pruned_loss=0.1016, over 28791.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.37, pruned_loss=0.1222, over 5637904.07 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3414, pruned_loss=0.08817, over 5757656.40 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3719, pruned_loss=0.1244, over 5639934.19 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:02:20,569 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 09:02:36,477 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 11500, giga_loss[loss=0.3074, simple_loss=0.3712, pruned_loss=0.1218, over 28869.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3692, pruned_loss=0.1216, over 5647428.99 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.08824, over 5761174.15 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3714, pruned_loss=0.1243, over 5642729.38 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:02:43,861 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1331973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:02:49,294 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:02:51,890 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:02,868 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1331993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:04,667 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1331996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:18,000 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 11550, libri_loss[loss=0.3761, simple_loss=0.426, pruned_loss=0.1631, over 19812.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3688, pruned_loss=0.1208, over 5648793.37 frames. ], libri_tot_loss[loss=0.259, simple_loss=0.3415, pruned_loss=0.08822, over 5755790.90 frames. ], giga_tot_loss[loss=0.3096, simple_loss=0.3713, pruned_loss=0.1239, over 5647400.85 frames. ], batch size: 187, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:03:28,603 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:30,662 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1332026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:03:57,798 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:04:05,557 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 11600, giga_loss[loss=0.3427, simple_loss=0.3983, pruned_loss=0.1435, over 28983.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3706, pruned_loss=0.1217, over 5635694.18 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08853, over 5741711.46 frames. ], giga_tot_loss[loss=0.3117, simple_loss=0.3732, pruned_loss=0.1251, over 5642832.83 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:04:21,319 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.07 vs. limit=5.0
+2023-03-15 09:04:45,098 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 11650, giga_loss[loss=0.2814, simple_loss=0.3583, pruned_loss=0.1023, over 28886.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3706, pruned_loss=0.1207, over 5657122.44 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08863, over 5743590.06 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3728, pruned_loss=0.1236, over 5660200.12 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:05:35,871 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4836, 1.5662, 1.2048, 1.1428], device='cuda:1'), covar=tensor([0.0998, 0.0556, 0.1014, 0.1278], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0527, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 09:05:43,754 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 11700, giga_loss[loss=0.2873, simple_loss=0.3544, pruned_loss=0.1101, over 28685.00 frames. ], tot_loss[loss=0.3083, simple_loss=0.3719, pruned_loss=0.1223, over 5641406.23 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.3419, pruned_loss=0.08855, over 5737432.45 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3746, pruned_loss=0.1256, over 5645907.72 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:05:47,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:15,429 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:17,359 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:21,228 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:32,254 INFO [train.py:968] (1/2) Epoch 30, batch 11750, giga_loss[loss=0.2943, simple_loss=0.3605, pruned_loss=0.114, over 28939.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3736, pruned_loss=0.1234, over 5645846.31 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.342, pruned_loss=0.08859, over 5737867.51 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3766, pruned_loss=0.1271, over 5646230.85 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:06:33,803 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1332220.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:41,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:06:53,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3430, 1.5892, 1.2998, 1.1494], device='cuda:1'), covar=tensor([0.1090, 0.0548, 0.1015, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0528, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 09:07:10,897 INFO [optim.py:369] (1/2) 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 09:07:12,924 INFO [train.py:968] (1/2) Epoch 30, batch 11800, giga_loss[loss=0.3113, simple_loss=0.3759, pruned_loss=0.1233, over 28971.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3736, pruned_loss=0.1238, over 5650260.24 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3426, pruned_loss=0.08878, over 5743611.10 frames. ], giga_tot_loss[loss=0.3168, simple_loss=0.3771, pruned_loss=0.1283, over 5641431.52 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:07:25,937 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-15 09:07:54,153 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:07:56,906 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:07:58,623 INFO [train.py:968] (1/2) Epoch 30, batch 11850, giga_loss[loss=0.2662, simple_loss=0.358, pruned_loss=0.0872, over 29159.00 frames. ], tot_loss[loss=0.3091, simple_loss=0.3734, pruned_loss=0.1223, over 5658975.89 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3423, pruned_loss=0.08865, over 5747784.12 frames. ], giga_tot_loss[loss=0.3154, simple_loss=0.3772, pruned_loss=0.1268, over 5646242.29 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:08:01,700 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4099, 1.2549, 3.6565, 3.2734], device='cuda:1'), covar=tensor([0.1587, 0.2888, 0.0529, 0.0980], device='cuda:1'), in_proj_covar=tensor([0.0821, 0.0683, 0.1024, 0.1002], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 09:08:23,445 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:08:27,019 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:08:28,210 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:08:31,105 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:08:47,757 INFO [train.py:968] (1/2) Epoch 30, batch 11900, libri_loss[loss=0.2731, simple_loss=0.3616, pruned_loss=0.09229, over 29416.00 frames. ], tot_loss[loss=0.3067, simple_loss=0.3723, pruned_loss=0.1206, over 5660966.22 frames. ], libri_tot_loss[loss=0.2595, simple_loss=0.342, pruned_loss=0.08848, over 5751160.26 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3763, pruned_loss=0.125, over 5645669.57 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:08:59,778 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:09:14,791 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:09:31,060 INFO [train.py:968] (1/2) Epoch 30, batch 11950, giga_loss[loss=0.2548, simple_loss=0.3332, pruned_loss=0.08822, over 28295.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3716, pruned_loss=0.1198, over 5660874.00 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3422, pruned_loss=0.08848, over 5751528.20 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3755, pruned_loss=0.1245, over 5645343.15 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:10:12,387 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 12000, giga_loss[loss=0.2841, simple_loss=0.3586, pruned_loss=0.1048, over 28749.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3696, pruned_loss=0.1189, over 5663701.26 frames. ], libri_tot_loss[loss=0.2598, simple_loss=0.3423, pruned_loss=0.08864, over 5755903.62 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3735, pruned_loss=0.1234, over 5644469.12 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:10:15,061 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 09:10:24,610 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 09:10:37,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:10:48,687 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:10:49,433 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2088, 1.8134, 1.3727, 0.4644], device='cuda:1'), covar=tensor([0.5160, 0.3288, 0.4677, 0.6868], device='cuda:1'), in_proj_covar=tensor([0.1876, 0.1762, 0.1678, 0.1528], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 09:10:50,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:12,888 INFO [train.py:968] (1/2) Epoch 30, batch 12050, giga_loss[loss=0.2769, simple_loss=0.3559, pruned_loss=0.099, over 28920.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.37, pruned_loss=0.1191, over 5667772.88 frames. ], libri_tot_loss[loss=0.2596, simple_loss=0.3422, pruned_loss=0.08851, over 5757050.50 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3736, pruned_loss=0.1231, over 5650352.50 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:11:17,373 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:37,640 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:40,055 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:11:55,766 INFO [optim.py:369] (1/2) 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 09:11:57,319 INFO [train.py:968] (1/2) Epoch 30, batch 12100, giga_loss[loss=0.3005, simple_loss=0.3645, pruned_loss=0.1182, over 29113.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3697, pruned_loss=0.1188, over 5666600.13 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3418, pruned_loss=0.08837, over 5760627.35 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.3738, pruned_loss=0.1233, over 5646005.08 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:12:06,213 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:12:21,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1332595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:12:43,118 INFO [train.py:968] (1/2) Epoch 30, batch 12150, giga_loss[loss=0.3003, simple_loss=0.357, pruned_loss=0.1218, over 28628.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3685, pruned_loss=0.1183, over 5678810.39 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3416, pruned_loss=0.0881, over 5760811.97 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3731, pruned_loss=0.1233, over 5658459.13 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:12:49,659 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:12:53,029 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332627.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:13:20,856 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 09:13:31,373 INFO [train.py:968] (1/2) Epoch 30, batch 12200, giga_loss[loss=0.2714, simple_loss=0.3503, pruned_loss=0.09626, over 28992.00 frames. ], tot_loss[loss=0.3033, simple_loss=0.3685, pruned_loss=0.1191, over 5676836.99 frames. ], libri_tot_loss[loss=0.2587, simple_loss=0.3415, pruned_loss=0.088, over 5762967.44 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3726, pruned_loss=0.1235, over 5657833.65 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:14:21,852 INFO [train.py:968] (1/2) Epoch 30, batch 12250, giga_loss[loss=0.3271, simple_loss=0.3958, pruned_loss=0.1292, over 28709.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3696, pruned_loss=0.1198, over 5680944.19 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3415, pruned_loss=0.088, over 5766125.12 frames. ], giga_tot_loss[loss=0.3112, simple_loss=0.3737, pruned_loss=0.1243, over 5660167.31 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:14:40,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1332738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:14:43,135 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1332741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:15:03,722 INFO [optim.py:369] (1/2) 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 09:15:06,112 INFO [train.py:968] (1/2) Epoch 30, batch 12300, giga_loss[loss=0.2522, simple_loss=0.3288, pruned_loss=0.08777, over 28459.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3706, pruned_loss=0.1202, over 5680125.54 frames. ], libri_tot_loss[loss=0.2589, simple_loss=0.3417, pruned_loss=0.08805, over 5771015.75 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3749, pruned_loss=0.125, over 5655725.95 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:15:07,842 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1332770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:15:49,967 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6103, 1.9431, 1.5652, 1.4552], device='cuda:1'), covar=tensor([0.3023, 0.3032, 0.3537, 0.2768], device='cuda:1'), in_proj_covar=tensor([0.1626, 0.1173, 0.1439, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 09:15:51,607 INFO [train.py:968] (1/2) Epoch 30, batch 12350, giga_loss[loss=0.2662, simple_loss=0.3489, pruned_loss=0.09171, over 29070.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3708, pruned_loss=0.1202, over 5689107.82 frames. ], libri_tot_loss[loss=0.2588, simple_loss=0.3415, pruned_loss=0.08806, over 5774363.27 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.375, pruned_loss=0.1249, over 5664570.08 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:16:39,474 INFO [optim.py:369] (1/2) 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 09:16:40,783 INFO [train.py:968] (1/2) Epoch 30, batch 12400, giga_loss[loss=0.2967, simple_loss=0.3672, pruned_loss=0.1131, over 28877.00 frames. ], tot_loss[loss=0.3047, simple_loss=0.37, pruned_loss=0.1197, over 5674035.57 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3412, pruned_loss=0.08784, over 5775873.35 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3742, pruned_loss=0.1242, over 5651458.27 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:17:15,299 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.21 vs. limit=2.0
+2023-03-15 09:17:23,280 INFO [train.py:968] (1/2) Epoch 30, batch 12450, giga_loss[loss=0.2833, simple_loss=0.3562, pruned_loss=0.1052, over 28620.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3688, pruned_loss=0.1178, over 5686234.02 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3407, pruned_loss=0.08752, over 5778176.38 frames. ], giga_tot_loss[loss=0.3092, simple_loss=0.3733, pruned_loss=0.1225, over 5663641.36 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:18:08,734 INFO [optim.py:369] (1/2) 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 09:18:09,354 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.89 vs. limit=2.0
+2023-03-15 09:18:09,452 INFO [train.py:968] (1/2) Epoch 30, batch 12500, libri_loss[loss=0.2201, simple_loss=0.2985, pruned_loss=0.07086, over 29668.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3676, pruned_loss=0.1166, over 5696440.56 frames. ], libri_tot_loss[loss=0.2578, simple_loss=0.3407, pruned_loss=0.0875, over 5779341.46 frames. ], giga_tot_loss[loss=0.3078, simple_loss=0.3724, pruned_loss=0.1216, over 5673556.07 frames. ], batch size: 69, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:18:55,419 INFO [train.py:968] (1/2) Epoch 30, batch 12550, giga_loss[loss=0.2771, simple_loss=0.3506, pruned_loss=0.1018, over 28767.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3676, pruned_loss=0.1177, over 5679081.93 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3402, pruned_loss=0.08739, over 5772846.95 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3722, pruned_loss=0.1223, over 5664274.11 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:19:41,471 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 12600, giga_loss[loss=0.2809, simple_loss=0.3457, pruned_loss=0.1081, over 28516.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3648, pruned_loss=0.1163, over 5661214.39 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.3399, pruned_loss=0.0872, over 5755589.73 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3697, pruned_loss=0.121, over 5661476.31 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:20:29,574 INFO [train.py:968] (1/2) Epoch 30, batch 12650, libri_loss[loss=0.2791, simple_loss=0.3618, pruned_loss=0.09825, over 29534.00 frames. ], tot_loss[loss=0.2957, simple_loss=0.3618, pruned_loss=0.1148, over 5673182.38 frames. ], libri_tot_loss[loss=0.2575, simple_loss=0.3403, pruned_loss=0.08733, over 5758797.75 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3658, pruned_loss=0.1191, over 5668340.27 frames. ], batch size: 82, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:21:13,416 INFO [optim.py:369] (1/2) 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 09:21:13,428 INFO [train.py:968] (1/2) Epoch 30, batch 12700, giga_loss[loss=0.2973, simple_loss=0.3636, pruned_loss=0.1155, over 28902.00 frames. ], tot_loss[loss=0.2948, simple_loss=0.3603, pruned_loss=0.1146, over 5681733.81 frames. ], libri_tot_loss[loss=0.2574, simple_loss=0.3403, pruned_loss=0.08729, over 5761624.72 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3639, pruned_loss=0.1186, over 5673905.99 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:21:38,524 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-15 09:21:56,974 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-15 09:21:59,737 INFO [train.py:968] (1/2) Epoch 30, batch 12750, giga_loss[loss=0.2739, simple_loss=0.3514, pruned_loss=0.09818, over 28906.00 frames. ], tot_loss[loss=0.2944, simple_loss=0.3598, pruned_loss=0.1145, over 5682826.04 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3408, pruned_loss=0.08755, over 5755737.27 frames. ], giga_tot_loss[loss=0.2993, simple_loss=0.3626, pruned_loss=0.118, over 5680642.99 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:22:52,112 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 12800, giga_loss[loss=0.2805, simple_loss=0.3635, pruned_loss=0.09875, over 28580.00 frames. ], tot_loss[loss=0.2938, simple_loss=0.3601, pruned_loss=0.1137, over 5683257.41 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3406, pruned_loss=0.08741, over 5758873.33 frames. ], giga_tot_loss[loss=0.2986, simple_loss=0.3629, pruned_loss=0.1171, over 5677532.69 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:23:38,907 INFO [train.py:968] (1/2) Epoch 30, batch 12850, giga_loss[loss=0.2793, simple_loss=0.3568, pruned_loss=0.1009, over 27521.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3589, pruned_loss=0.1107, over 5672955.22 frames. ], libri_tot_loss[loss=0.2577, simple_loss=0.3405, pruned_loss=0.08747, over 5748124.01 frames. ], giga_tot_loss[loss=0.2947, simple_loss=0.3616, pruned_loss=0.1139, over 5675575.91 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:23:59,154 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-15 09:24:26,478 INFO [optim.py:369] (1/2) 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 09:24:26,490 INFO [train.py:968] (1/2) Epoch 30, batch 12900, libri_loss[loss=0.2815, simple_loss=0.3506, pruned_loss=0.1062, over 19694.00 frames. ], tot_loss[loss=0.2864, simple_loss=0.3567, pruned_loss=0.108, over 5657571.06 frames. ], libri_tot_loss[loss=0.2579, simple_loss=0.3405, pruned_loss=0.08765, over 5741895.85 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3595, pruned_loss=0.1111, over 5663569.94 frames. ], batch size: 188, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:25:10,821 INFO [train.py:968] (1/2) Epoch 30, batch 12950, giga_loss[loss=0.2884, simple_loss=0.3651, pruned_loss=0.1058, over 28030.00 frames. ], tot_loss[loss=0.2822, simple_loss=0.3538, pruned_loss=0.1053, over 5665524.36 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3405, pruned_loss=0.08816, over 5749940.72 frames. ], giga_tot_loss[loss=0.2868, simple_loss=0.3569, pruned_loss=0.1083, over 5658857.56 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:25:58,632 INFO [optim.py:369] (1/2) 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 09:25:58,645 INFO [train.py:968] (1/2) Epoch 30, batch 13000, giga_loss[loss=0.2878, simple_loss=0.3471, pruned_loss=0.1142, over 26578.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3497, pruned_loss=0.1018, over 5666131.34 frames. ], libri_tot_loss[loss=0.2576, simple_loss=0.3395, pruned_loss=0.08786, over 5751536.36 frames. ], giga_tot_loss[loss=0.2819, simple_loss=0.3535, pruned_loss=0.1051, over 5655814.62 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:26:26,957 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3841, 3.2742, 1.4778, 1.5655], device='cuda:1'), covar=tensor([0.1045, 0.0352, 0.1000, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0579, 0.0417, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 09:26:45,515 INFO [train.py:968] (1/2) Epoch 30, batch 13050, giga_loss[loss=0.2899, simple_loss=0.3751, pruned_loss=0.1024, over 28903.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3487, pruned_loss=0.0996, over 5667219.46 frames. ], libri_tot_loss[loss=0.2573, simple_loss=0.3391, pruned_loss=0.08779, over 5753751.06 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3523, pruned_loss=0.1026, over 5655126.43 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:26:52,104 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2540, 1.5420, 1.6561, 1.3836], device='cuda:1'), covar=tensor([0.3195, 0.2444, 0.1802, 0.2427], device='cuda:1'), in_proj_covar=tensor([0.2084, 0.2061, 0.1966, 0.2108], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 09:26:56,493 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1236, 1.3267, 1.0627, 0.9787], device='cuda:1'), covar=tensor([0.1154, 0.0488, 0.1119, 0.1090], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0452, 0.0525, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 09:27:09,089 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3092, 1.5217, 1.5116, 1.3335], device='cuda:1'), covar=tensor([0.2868, 0.2369, 0.1764, 0.2287], device='cuda:1'), in_proj_covar=tensor([0.2081, 0.2058, 0.1963, 0.2105], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 09:27:33,118 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 13100, giga_loss[loss=0.2666, simple_loss=0.3444, pruned_loss=0.0944, over 28628.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3482, pruned_loss=0.09794, over 5669316.32 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3385, pruned_loss=0.08745, over 5757513.60 frames. ], giga_tot_loss[loss=0.2769, simple_loss=0.352, pruned_loss=0.1009, over 5654482.46 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:27:42,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6987, 1.9439, 1.5546, 1.9277], device='cuda:1'), covar=tensor([0.3076, 0.2981, 0.3455, 0.2535], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1171, 0.1442, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 09:28:24,297 INFO [train.py:968] (1/2) Epoch 30, batch 13150, giga_loss[loss=0.2662, simple_loss=0.3434, pruned_loss=0.09454, over 28951.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.349, pruned_loss=0.09839, over 5664976.97 frames. ], libri_tot_loss[loss=0.2567, simple_loss=0.3384, pruned_loss=0.08753, over 5756394.53 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3522, pruned_loss=0.1009, over 5652787.90 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:28:38,199 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1333635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:28:44,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3356, 1.7855, 1.6669, 1.5602], device='cuda:1'), covar=tensor([0.2231, 0.2057, 0.2085, 0.2074], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0759, 0.0731, 0.0698], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 09:29:08,177 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-15 09:29:08,941 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 13200, giga_loss[loss=0.265, simple_loss=0.3443, pruned_loss=0.09287, over 28892.00 frames. ], tot_loss[loss=0.2694, simple_loss=0.3459, pruned_loss=0.09643, over 5668274.27 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3375, pruned_loss=0.08719, over 5760293.36 frames. ], giga_tot_loss[loss=0.274, simple_loss=0.3497, pruned_loss=0.09917, over 5651288.23 frames. ], batch size: 285, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:29:56,714 INFO [train.py:968] (1/2) Epoch 30, batch 13250, giga_loss[loss=0.2536, simple_loss=0.3413, pruned_loss=0.08301, over 28858.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.343, pruned_loss=0.09419, over 5677484.02 frames. ], libri_tot_loss[loss=0.256, simple_loss=0.3375, pruned_loss=0.0873, over 5762747.90 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3463, pruned_loss=0.09649, over 5659395.29 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:30:27,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4980, 1.6430, 1.6599, 1.2820], device='cuda:1'), covar=tensor([0.1951, 0.2900, 0.1654, 0.2014], device='cuda:1'), in_proj_covar=tensor([0.0936, 0.0716, 0.0985, 0.0886], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 09:30:39,687 INFO [train.py:968] (1/2) Epoch 30, batch 13300, giga_loss[loss=0.2764, simple_loss=0.3513, pruned_loss=0.1007, over 28714.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3432, pruned_loss=0.09422, over 5683033.28 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3376, pruned_loss=0.08746, over 5765669.86 frames. ], giga_tot_loss[loss=0.2692, simple_loss=0.3459, pruned_loss=0.09626, over 5662415.74 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:30:41,066 INFO [optim.py:369] (1/2) 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 09:30:43,345 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2411, 1.5721, 1.4476, 1.1896], device='cuda:1'), covar=tensor([0.2845, 0.2395, 0.1695, 0.2386], device='cuda:1'), in_proj_covar=tensor([0.2071, 0.2047, 0.1952, 0.2092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 09:31:12,428 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8338, 3.6824, 3.4663, 1.6747], device='cuda:1'), covar=tensor([0.0751, 0.0877, 0.0845, 0.2237], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1225, 0.1028, 0.0761], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 09:31:26,060 INFO [train.py:968] (1/2) Epoch 30, batch 13350, giga_loss[loss=0.244, simple_loss=0.329, pruned_loss=0.07947, over 28973.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3429, pruned_loss=0.09408, over 5676920.72 frames. ], libri_tot_loss[loss=0.2562, simple_loss=0.3373, pruned_loss=0.08756, over 5760546.32 frames. ], giga_tot_loss[loss=0.2686, simple_loss=0.3456, pruned_loss=0.09579, over 5662508.70 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:32:13,982 INFO [train.py:968] (1/2) Epoch 30, batch 13400, libri_loss[loss=0.2518, simple_loss=0.3355, pruned_loss=0.084, over 29487.00 frames. ], tot_loss[loss=0.2638, simple_loss=0.3415, pruned_loss=0.09303, over 5673563.17 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08732, over 5763276.59 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3442, pruned_loss=0.09472, over 5658114.93 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:32:15,103 INFO [optim.py:369] (1/2) 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 09:33:03,183 INFO [train.py:968] (1/2) Epoch 30, batch 13450, giga_loss[loss=0.256, simple_loss=0.3373, pruned_loss=0.08739, over 28666.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.339, pruned_loss=0.09102, over 5678048.95 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08735, over 5765742.25 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3413, pruned_loss=0.0924, over 5661848.39 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:33:23,782 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9734, 2.4044, 2.2872, 1.9744], device='cuda:1'), covar=tensor([0.2386, 0.2460, 0.2023, 0.2377], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0757, 0.0729, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 09:33:35,650 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1333949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:33:44,411 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1333959.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 09:33:53,013 INFO [train.py:968] (1/2) Epoch 30, batch 13500, giga_loss[loss=0.2439, simple_loss=0.3198, pruned_loss=0.08403, over 28003.00 frames. ], tot_loss[loss=0.2575, simple_loss=0.3358, pruned_loss=0.08962, over 5667154.85 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3368, pruned_loss=0.08745, over 5765700.26 frames. ], giga_tot_loss[loss=0.2594, simple_loss=0.3375, pruned_loss=0.0907, over 5652232.54 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:33:53,625 INFO [optim.py:369] (1/2) 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 09:33:55,000 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3611, 1.8702, 1.6830, 1.6108], device='cuda:1'), covar=tensor([0.2310, 0.2109, 0.1934, 0.2042], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0757, 0.0729, 0.0697], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 09:34:34,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6129, 1.7270, 1.8151, 1.3978], device='cuda:1'), covar=tensor([0.2011, 0.2968, 0.1692, 0.2033], device='cuda:1'), in_proj_covar=tensor([0.0935, 0.0714, 0.0984, 0.0885], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 09:34:37,145 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334010.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:34:38,694 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6388, 1.8388, 1.9152, 1.6488], device='cuda:1'), covar=tensor([0.2782, 0.2358, 0.1708, 0.2193], device='cuda:1'), in_proj_covar=tensor([0.2063, 0.2038, 0.1943, 0.2082], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 09:34:41,934 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334015.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:34:45,295 INFO [train.py:968] (1/2) Epoch 30, batch 13550, giga_loss[loss=0.271, simple_loss=0.3432, pruned_loss=0.09941, over 28971.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3342, pruned_loss=0.08946, over 5656684.98 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08738, over 5768223.64 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3356, pruned_loss=0.09044, over 5640196.64 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:35:37,706 INFO [train.py:968] (1/2) Epoch 30, batch 13600, giga_loss[loss=0.258, simple_loss=0.3388, pruned_loss=0.0886, over 28863.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3347, pruned_loss=0.09005, over 5645469.66 frames. ], libri_tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08735, over 5760211.99 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3358, pruned_loss=0.09089, over 5638778.25 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:35:38,359 INFO [optim.py:369] (1/2) 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 09:36:31,903 INFO [train.py:968] (1/2) Epoch 30, batch 13650, giga_loss[loss=0.2118, simple_loss=0.2866, pruned_loss=0.06848, over 24246.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3362, pruned_loss=0.09029, over 5637542.08 frames. ], libri_tot_loss[loss=0.2553, simple_loss=0.3363, pruned_loss=0.08714, over 5760466.04 frames. ], giga_tot_loss[loss=0.26, simple_loss=0.3375, pruned_loss=0.09121, over 5629757.16 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:37:11,025 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:37:13,979 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 13700, giga_loss[loss=0.2674, simple_loss=0.3489, pruned_loss=0.09298, over 28932.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3368, pruned_loss=0.0894, over 5644453.40 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.3361, pruned_loss=0.08711, over 5762892.39 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.338, pruned_loss=0.09026, over 5632194.69 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:37:29,942 INFO [optim.py:369] (1/2) 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 09:37:49,017 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:38:27,173 INFO [train.py:968] (1/2) Epoch 30, batch 13750, libri_loss[loss=0.2369, simple_loss=0.3216, pruned_loss=0.07607, over 29521.00 frames. ], tot_loss[loss=0.2594, simple_loss=0.338, pruned_loss=0.09039, over 5643480.90 frames. ], libri_tot_loss[loss=0.255, simple_loss=0.336, pruned_loss=0.08704, over 5762783.57 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.339, pruned_loss=0.09117, over 5632735.31 frames. ], batch size: 82, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:39:24,239 INFO [train.py:968] (1/2) Epoch 30, batch 13800, libri_loss[loss=0.2427, simple_loss=0.3209, pruned_loss=0.08227, over 29544.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3349, pruned_loss=0.08841, over 5654712.13 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3351, pruned_loss=0.08675, over 5763342.14 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3366, pruned_loss=0.08935, over 5642927.64 frames. ], batch size: 81, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:39:26,741 INFO [optim.py:369] (1/2) 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 09:39:51,965 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2839, 1.2067, 3.5255, 3.2003], device='cuda:1'), covar=tensor([0.1628, 0.3005, 0.0529, 0.1125], device='cuda:1'), in_proj_covar=tensor([0.0813, 0.0676, 0.1012, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 09:39:52,103 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-15 09:40:18,982 INFO [train.py:968] (1/2) Epoch 30, batch 13850, giga_loss[loss=0.286, simple_loss=0.3701, pruned_loss=0.1009, over 28502.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3342, pruned_loss=0.08719, over 5641031.82 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3346, pruned_loss=0.08658, over 5756909.69 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.336, pruned_loss=0.08812, over 5634129.29 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:40:28,700 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:40:39,519 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334334.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 09:40:54,704 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3012, 1.5829, 1.3787, 1.5250], device='cuda:1'), covar=tensor([0.0794, 0.0360, 0.0381, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 09:41:17,226 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.83 vs. limit=2.0
+2023-03-15 09:41:22,169 INFO [train.py:968] (1/2) Epoch 30, batch 13900, giga_loss[loss=0.2541, simple_loss=0.3358, pruned_loss=0.08622, over 28407.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3332, pruned_loss=0.08665, over 5647471.24 frames. ], libri_tot_loss[loss=0.2538, simple_loss=0.3345, pruned_loss=0.08658, over 5757231.85 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3347, pruned_loss=0.08737, over 5640340.79 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:41:23,764 INFO [optim.py:369] (1/2) 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 09:41:43,463 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:41:48,932 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:42:17,387 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5157, 1.4749, 1.7016, 1.3421], device='cuda:1'), covar=tensor([0.1844, 0.2819, 0.1586, 0.1966], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0716, 0.0989, 0.0890], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 09:42:20,057 INFO [train.py:968] (1/2) Epoch 30, batch 13950, giga_loss[loss=0.3502, simple_loss=0.4038, pruned_loss=0.1483, over 28375.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3306, pruned_loss=0.08643, over 5650877.43 frames. ], libri_tot_loss[loss=0.253, simple_loss=0.3337, pruned_loss=0.08616, over 5759512.79 frames. ], giga_tot_loss[loss=0.2536, simple_loss=0.3325, pruned_loss=0.08739, over 5640720.13 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:43:16,759 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:43:17,158 INFO [train.py:968] (1/2) Epoch 30, batch 14000, giga_loss[loss=0.2291, simple_loss=0.3115, pruned_loss=0.07335, over 28097.00 frames. ], tot_loss[loss=0.2515, simple_loss=0.3302, pruned_loss=0.08635, over 5657160.15 frames. ], libri_tot_loss[loss=0.2527, simple_loss=0.3334, pruned_loss=0.08598, over 5761762.91 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3319, pruned_loss=0.08729, over 5645186.92 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:43:18,873 INFO [optim.py:369] (1/2) 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 09:43:21,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:43:29,443 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334477.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 09:43:32,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334480.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 09:43:52,272 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7435, 2.0597, 2.0834, 1.8351], device='cuda:1'), covar=tensor([0.1763, 0.1529, 0.1616, 0.1562], device='cuda:1'), in_proj_covar=tensor([0.0509, 0.0754, 0.0726, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 09:43:52,993 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:43:57,840 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:44:03,314 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334509.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 09:44:15,395 INFO [train.py:968] (1/2) Epoch 30, batch 14050, giga_loss[loss=0.2656, simple_loss=0.3462, pruned_loss=0.09251, over 28861.00 frames. ], tot_loss[loss=0.2522, simple_loss=0.3314, pruned_loss=0.08647, over 5661677.78 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3332, pruned_loss=0.08592, over 5755199.72 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3329, pruned_loss=0.08727, over 5656923.29 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:44:27,770 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5241, 4.3561, 4.1575, 1.7570], device='cuda:1'), covar=tensor([0.0558, 0.0688, 0.0766, 0.2085], device='cuda:1'), in_proj_covar=tensor([0.1313, 0.1210, 0.1015, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 09:44:33,616 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:44:37,298 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334536.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:45:13,356 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:45:16,009 INFO [train.py:968] (1/2) Epoch 30, batch 14100, giga_loss[loss=0.2528, simple_loss=0.3325, pruned_loss=0.08654, over 27537.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3348, pruned_loss=0.08782, over 5671752.79 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3329, pruned_loss=0.08582, over 5757533.97 frames. ], giga_tot_loss[loss=0.2567, simple_loss=0.3362, pruned_loss=0.08856, over 5664702.94 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:45:19,765 INFO [optim.py:369] (1/2) 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 09:46:20,834 INFO [train.py:968] (1/2) Epoch 30, batch 14150, giga_loss[loss=0.2253, simple_loss=0.3064, pruned_loss=0.07209, over 28093.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3319, pruned_loss=0.08584, over 5674063.48 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3328, pruned_loss=0.08579, over 5760090.53 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3332, pruned_loss=0.08647, over 5664568.85 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:47:25,445 INFO [train.py:968] (1/2) Epoch 30, batch 14200, giga_loss[loss=0.2618, simple_loss=0.3449, pruned_loss=0.08933, over 28855.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3324, pruned_loss=0.08651, over 5681010.40 frames. ], libri_tot_loss[loss=0.2522, simple_loss=0.3327, pruned_loss=0.08586, over 5762346.35 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3335, pruned_loss=0.08694, over 5670597.54 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:47:29,809 INFO [optim.py:369] (1/2) 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 09:48:28,591 INFO [train.py:968] (1/2) Epoch 30, batch 14250, giga_loss[loss=0.2602, simple_loss=0.3349, pruned_loss=0.09277, over 27696.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3348, pruned_loss=0.08763, over 5667277.00 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3326, pruned_loss=0.08593, over 5766076.70 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3358, pruned_loss=0.08796, over 5652099.58 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:48:29,856 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1334719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:49:16,457 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:49:22,446 INFO [train.py:968] (1/2) Epoch 30, batch 14300, giga_loss[loss=0.2765, simple_loss=0.3644, pruned_loss=0.0943, over 28881.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3372, pruned_loss=0.08675, over 5663509.19 frames. ], libri_tot_loss[loss=0.2515, simple_loss=0.3318, pruned_loss=0.08563, over 5762417.33 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3388, pruned_loss=0.08735, over 5649652.02 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:49:28,231 INFO [optim.py:369] (1/2) 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 09:49:37,809 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2012, 2.4038, 1.2403, 1.2720], device='cuda:1'), covar=tensor([0.0984, 0.0385, 0.0981, 0.1476], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0575, 0.0416, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 09:50:23,132 INFO [train.py:968] (1/2) Epoch 30, batch 14350, libri_loss[loss=0.2743, simple_loss=0.3537, pruned_loss=0.09744, over 29548.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3371, pruned_loss=0.08553, over 5640789.42 frames. ], libri_tot_loss[loss=0.2514, simple_loss=0.3318, pruned_loss=0.08556, over 5755713.53 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3386, pruned_loss=0.08608, over 5633533.54 frames. ], batch size: 83, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 09:51:16,245 INFO [train.py:968] (1/2) Epoch 30, batch 14400, giga_loss[loss=0.2691, simple_loss=0.3623, pruned_loss=0.088, over 28718.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3372, pruned_loss=0.08421, over 5659647.34 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3313, pruned_loss=0.08529, over 5759815.03 frames. ], giga_tot_loss[loss=0.2545, simple_loss=0.3391, pruned_loss=0.0849, over 5646762.90 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:51:19,889 INFO [optim.py:369] (1/2) 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 09:51:28,255 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1334879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:51:57,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1334904.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:52:00,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1334907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:52:14,401 INFO [train.py:968] (1/2) Epoch 30, batch 14450, giga_loss[loss=0.2252, simple_loss=0.3118, pruned_loss=0.06927, over 29032.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.3387, pruned_loss=0.08583, over 5670100.88 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3309, pruned_loss=0.0851, over 5764397.72 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3408, pruned_loss=0.08655, over 5653288.24 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:52:36,298 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1334936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:52:46,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 09:53:00,772 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2856, 3.6036, 2.1896, 1.0668], device='cuda:1'), covar=tensor([0.8060, 0.3231, 0.5037, 0.8627], device='cuda:1'), in_proj_covar=tensor([0.1863, 0.1744, 0.1670, 0.1517], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 09:53:12,871 INFO [train.py:968] (1/2) Epoch 30, batch 14500, giga_loss[loss=0.2228, simple_loss=0.3097, pruned_loss=0.068, over 28909.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3372, pruned_loss=0.08614, over 5672294.28 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3303, pruned_loss=0.08479, over 5764647.67 frames. ], giga_tot_loss[loss=0.2568, simple_loss=0.3396, pruned_loss=0.08705, over 5656258.22 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:53:17,344 INFO [optim.py:369] (1/2) 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 09:53:39,942 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9690, 1.2918, 1.0560, 0.1867], device='cuda:1'), covar=tensor([0.4047, 0.3058, 0.4300, 0.7123], device='cuda:1'), in_proj_covar=tensor([0.1864, 0.1745, 0.1670, 0.1517], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 09:54:05,084 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2370, 1.3536, 3.6965, 3.1529], device='cuda:1'), covar=tensor([0.1684, 0.2717, 0.0509, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0676, 0.1009, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 09:54:26,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2253, 1.5786, 1.5860, 1.4095], device='cuda:1'), covar=tensor([0.1942, 0.1368, 0.1637, 0.1525], device='cuda:1'), in_proj_covar=tensor([0.0507, 0.0752, 0.0726, 0.0694], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 09:54:27,532 INFO [train.py:968] (1/2) Epoch 30, batch 14550, giga_loss[loss=0.2435, simple_loss=0.3284, pruned_loss=0.07935, over 28662.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3375, pruned_loss=0.08692, over 5668501.14 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3301, pruned_loss=0.08465, over 5765852.96 frames. ], giga_tot_loss[loss=0.2576, simple_loss=0.3396, pruned_loss=0.08779, over 5653925.89 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:54:32,884 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1335022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:54:35,986 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1335025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:55:26,057 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 14600, giga_loss[loss=0.2264, simple_loss=0.3113, pruned_loss=0.07077, over 28814.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3335, pruned_loss=0.08443, over 5676800.89 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.33, pruned_loss=0.08458, over 5764407.17 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3354, pruned_loss=0.08518, over 5665426.24 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:55:48,478 INFO [optim.py:369] (1/2) 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 09:56:18,348 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1335094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:56:41,739 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7860, 3.6285, 3.4569, 1.7818], device='cuda:1'), covar=tensor([0.0734, 0.0843, 0.0885, 0.2545], device='cuda:1'), in_proj_covar=tensor([0.1308, 0.1205, 0.1011, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 09:56:45,241 INFO [train.py:968] (1/2) Epoch 30, batch 14650, libri_loss[loss=0.2027, simple_loss=0.281, pruned_loss=0.06217, over 29666.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.3312, pruned_loss=0.08346, over 5664812.62 frames. ], libri_tot_loss[loss=0.2495, simple_loss=0.3298, pruned_loss=0.08455, over 5755971.92 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.333, pruned_loss=0.0841, over 5659944.83 frames. ], batch size: 73, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:57:47,873 INFO [train.py:968] (1/2) Epoch 30, batch 14700, giga_loss[loss=0.2444, simple_loss=0.3255, pruned_loss=0.08166, over 29072.00 frames. ], tot_loss[loss=0.247, simple_loss=0.329, pruned_loss=0.0825, over 5668170.75 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3301, pruned_loss=0.08464, over 5757518.68 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3302, pruned_loss=0.08289, over 5661645.66 frames. ], batch size: 146, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:57:52,192 INFO [optim.py:369] (1/2) 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 09:58:44,524 INFO [train.py:968] (1/2) Epoch 30, batch 14750, giga_loss[loss=0.2699, simple_loss=0.3596, pruned_loss=0.09012, over 28988.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3326, pruned_loss=0.08472, over 5677183.28 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3297, pruned_loss=0.08444, over 5756725.81 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.334, pruned_loss=0.08519, over 5669853.35 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 09:59:09,309 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1335237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:59:12,960 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1335240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:59:44,694 INFO [train.py:968] (1/2) Epoch 30, batch 14800, giga_loss[loss=0.2633, simple_loss=0.3493, pruned_loss=0.08871, over 28352.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3348, pruned_loss=0.08593, over 5673538.51 frames. ], libri_tot_loss[loss=0.2487, simple_loss=0.3291, pruned_loss=0.08413, over 5753675.58 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3365, pruned_loss=0.0866, over 5668729.79 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 09:59:46,506 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1335269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 09:59:50,117 INFO [optim.py:369] (1/2) 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 10:00:41,515 INFO [train.py:968] (1/2) Epoch 30, batch 14850, giga_loss[loss=0.2733, simple_loss=0.3446, pruned_loss=0.1011, over 28974.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3323, pruned_loss=0.08566, over 5681283.81 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.329, pruned_loss=0.08412, over 5757256.22 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3339, pruned_loss=0.08627, over 5671739.06 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:01:35,240 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3189, 1.1635, 1.0795, 1.5840], device='cuda:1'), covar=tensor([0.0775, 0.0385, 0.0387, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 10:01:40,193 INFO [train.py:968] (1/2) Epoch 30, batch 14900, giga_loss[loss=0.2586, simple_loss=0.3408, pruned_loss=0.08819, over 28683.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.3337, pruned_loss=0.08745, over 5677479.04 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3287, pruned_loss=0.08406, over 5760121.35 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3354, pruned_loss=0.08805, over 5665450.63 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:01:45,789 INFO [optim.py:369] (1/2) 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 10:01:59,779 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 14950, giga_loss[loss=0.2698, simple_loss=0.3509, pruned_loss=0.09435, over 28166.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3333, pruned_loss=0.08716, over 5676873.91 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3282, pruned_loss=0.08385, over 5763254.85 frames. ], giga_tot_loss[loss=0.2555, simple_loss=0.3352, pruned_loss=0.08794, over 5662187.64 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:02:48,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.0193, 2.8984, 2.7094, 1.4746], device='cuda:1'), covar=tensor([0.1094, 0.1126, 0.1045, 0.2082], device='cuda:1'), in_proj_covar=tensor([0.1304, 0.1203, 0.1010, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 10:03:41,691 INFO [train.py:968] (1/2) Epoch 30, batch 15000, giga_loss[loss=0.2566, simple_loss=0.344, pruned_loss=0.08459, over 28783.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3352, pruned_loss=0.08705, over 5681662.89 frames. ], libri_tot_loss[loss=0.248, simple_loss=0.3281, pruned_loss=0.08394, over 5768407.79 frames. ], giga_tot_loss[loss=0.2562, simple_loss=0.337, pruned_loss=0.08771, over 5662137.35 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:03:41,691 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 10:03:51,039 INFO [train.py:1012] (1/2) Epoch 30, validation: loss=0.1929, simple_loss=0.2945, pruned_loss=0.04562, over 944034.00 frames. 
+2023-03-15 10:03:51,039 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 10:03:57,278 INFO [optim.py:369] (1/2) 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 10:04:17,498 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3121, 3.1137, 1.4193, 1.5623], device='cuda:1'), covar=tensor([0.1011, 0.0310, 0.0931, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0574, 0.0416, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 10:04:40,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1335504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:05:01,656 INFO [train.py:968] (1/2) Epoch 30, batch 15050, giga_loss[loss=0.2919, simple_loss=0.361, pruned_loss=0.1114, over 28914.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3352, pruned_loss=0.08702, over 5671776.37 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3278, pruned_loss=0.08385, over 5762459.89 frames. ], giga_tot_loss[loss=0.2563, simple_loss=0.3372, pruned_loss=0.0877, over 5659573.32 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 10:05:51,099 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4916, 1.6176, 1.3115, 1.2188], device='cuda:1'), covar=tensor([0.1029, 0.0511, 0.1013, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0414, 0.0446, 0.0522, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 10:06:15,180 INFO [train.py:968] (1/2) Epoch 30, batch 15100, giga_loss[loss=0.2328, simple_loss=0.3099, pruned_loss=0.07781, over 28984.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3312, pruned_loss=0.08499, over 5688573.92 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3278, pruned_loss=0.08386, over 5764414.76 frames. ], giga_tot_loss[loss=0.252, simple_loss=0.3328, pruned_loss=0.08554, over 5676095.46 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 10:06:23,945 INFO [optim.py:369] (1/2) 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 10:07:22,031 INFO [train.py:968] (1/2) Epoch 30, batch 15150, giga_loss[loss=0.2185, simple_loss=0.2961, pruned_loss=0.0704, over 29000.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3265, pruned_loss=0.08355, over 5690585.86 frames. ], libri_tot_loss[loss=0.2475, simple_loss=0.3276, pruned_loss=0.08375, over 5766666.26 frames. ], giga_tot_loss[loss=0.2481, simple_loss=0.328, pruned_loss=0.08411, over 5677303.35 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 10:08:21,148 INFO [train.py:968] (1/2) Epoch 30, batch 15200, giga_loss[loss=0.2082, simple_loss=0.2853, pruned_loss=0.06559, over 24080.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.3261, pruned_loss=0.08383, over 5679484.36 frames. ], libri_tot_loss[loss=0.2477, simple_loss=0.3277, pruned_loss=0.08387, over 5764425.02 frames. ], giga_tot_loss[loss=0.2477, simple_loss=0.3271, pruned_loss=0.08416, over 5670500.40 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:08:28,898 INFO [optim.py:369] (1/2) 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 10:09:13,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4143, 1.6746, 1.3585, 1.6606], device='cuda:1'), covar=tensor([0.0802, 0.0322, 0.0373, 0.0930], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 10:09:15,249 INFO [train.py:968] (1/2) Epoch 30, batch 15250, giga_loss[loss=0.256, simple_loss=0.3367, pruned_loss=0.08762, over 28915.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3268, pruned_loss=0.0848, over 5675041.44 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3273, pruned_loss=0.08375, over 5760026.41 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.328, pruned_loss=0.08518, over 5669091.54 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:09:57,067 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 15300, giga_loss[loss=0.2065, simple_loss=0.2914, pruned_loss=0.06076, over 28053.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3246, pruned_loss=0.08323, over 5661107.59 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3268, pruned_loss=0.08352, over 5762063.30 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3259, pruned_loss=0.08376, over 5652249.78 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:10:15,770 INFO [optim.py:369] (1/2) 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 10:11:07,021 INFO [train.py:968] (1/2) Epoch 30, batch 15350, giga_loss[loss=0.2363, simple_loss=0.3195, pruned_loss=0.07656, over 28048.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.3237, pruned_loss=0.08166, over 5664233.18 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.327, pruned_loss=0.08359, over 5755226.30 frames. ], giga_tot_loss[loss=0.2443, simple_loss=0.3246, pruned_loss=0.082, over 5662247.48 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:12:00,486 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5133, 1.5937, 1.6871, 1.3266], device='cuda:1'), covar=tensor([0.1693, 0.2746, 0.1515, 0.1970], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0716, 0.0990, 0.0891], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 10:12:05,806 INFO [train.py:968] (1/2) Epoch 30, batch 15400, giga_loss[loss=0.2604, simple_loss=0.3435, pruned_loss=0.08864, over 29010.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3226, pruned_loss=0.08113, over 5661776.33 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3267, pruned_loss=0.08337, over 5759428.88 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3235, pruned_loss=0.08154, over 5653032.65 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:12:15,057 INFO [optim.py:369] (1/2) 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 10:12:24,167 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1335879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:12:56,564 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1335902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:12:58,813 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1335905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:13:14,480 INFO [train.py:968] (1/2) Epoch 30, batch 15450, giga_loss[loss=0.2079, simple_loss=0.2987, pruned_loss=0.05852, over 28980.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.3227, pruned_loss=0.08103, over 5674634.68 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3269, pruned_loss=0.08347, over 5759558.05 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3232, pruned_loss=0.08122, over 5666271.00 frames. ], batch size: 120, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:13:37,463 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1335934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:14:19,415 INFO [train.py:968] (1/2) Epoch 30, batch 15500, giga_loss[loss=0.2068, simple_loss=0.2888, pruned_loss=0.06239, over 28553.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.3233, pruned_loss=0.08088, over 5685194.86 frames. ], libri_tot_loss[loss=0.247, simple_loss=0.3269, pruned_loss=0.08351, over 5761031.67 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3236, pruned_loss=0.08096, over 5676561.16 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:14:25,960 INFO [optim.py:369] (1/2) 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 10:14:46,364 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.17 vs. limit=5.0
+2023-03-15 10:15:20,752 INFO [train.py:968] (1/2) Epoch 30, batch 15550, libri_loss[loss=0.2307, simple_loss=0.2959, pruned_loss=0.08268, over 29355.00 frames. ], tot_loss[loss=0.2458, simple_loss=0.3255, pruned_loss=0.08307, over 5687238.20 frames. ], libri_tot_loss[loss=0.2469, simple_loss=0.3267, pruned_loss=0.08351, over 5761508.79 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3259, pruned_loss=0.08312, over 5678297.01 frames. ], batch size: 67, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:15:27,775 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:15:32,185 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:15:46,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4814, 2.1069, 1.5562, 0.7121], device='cuda:1'), covar=tensor([0.7383, 0.3454, 0.4813, 0.7440], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1749, 0.1675, 0.1526], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 10:16:05,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:16:20,010 INFO [train.py:968] (1/2) Epoch 30, batch 15600, libri_loss[loss=0.2238, simple_loss=0.2957, pruned_loss=0.07594, over 29326.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.3234, pruned_loss=0.08209, over 5691067.68 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.326, pruned_loss=0.08321, over 5766256.43 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3243, pruned_loss=0.08238, over 5676968.72 frames. ], batch size: 67, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:16:28,395 INFO [optim.py:369] (1/2) 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 10:16:30,148 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:16:39,161 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.30 vs. limit=2.0
+2023-03-15 10:17:14,884 INFO [train.py:968] (1/2) Epoch 30, batch 15650, giga_loss[loss=0.2326, simple_loss=0.3063, pruned_loss=0.07942, over 24221.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.3238, pruned_loss=0.08118, over 5675120.36 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3261, pruned_loss=0.08319, over 5767881.90 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3244, pruned_loss=0.08137, over 5660687.47 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:18:19,769 INFO [train.py:968] (1/2) Epoch 30, batch 15700, giga_loss[loss=0.2786, simple_loss=0.345, pruned_loss=0.1061, over 26864.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3279, pruned_loss=0.08253, over 5668874.55 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3261, pruned_loss=0.08319, over 5767881.90 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3284, pruned_loss=0.08268, over 5657641.25 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:18:24,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3134, 1.9264, 1.3843, 0.6322], device='cuda:1'), covar=tensor([0.6203, 0.3434, 0.5057, 0.7128], device='cuda:1'), in_proj_covar=tensor([0.1863, 0.1744, 0.1670, 0.1521], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 10:18:26,078 INFO [optim.py:369] (1/2) 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 10:18:36,262 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:19:14,236 INFO [train.py:968] (1/2) Epoch 30, batch 15750, giga_loss[loss=0.2208, simple_loss=0.3188, pruned_loss=0.06141, over 28537.00 frames. ], tot_loss[loss=0.2471, simple_loss=0.3289, pruned_loss=0.08263, over 5671948.16 frames. ], libri_tot_loss[loss=0.2465, simple_loss=0.3263, pruned_loss=0.08334, over 5772079.03 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3291, pruned_loss=0.08259, over 5655753.41 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:20:09,441 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:20:12,323 INFO [train.py:968] (1/2) Epoch 30, batch 15800, giga_loss[loss=0.2538, simple_loss=0.3334, pruned_loss=0.08706, over 28014.00 frames. ], tot_loss[loss=0.2478, simple_loss=0.3294, pruned_loss=0.08314, over 5662111.66 frames. ], libri_tot_loss[loss=0.2466, simple_loss=0.3264, pruned_loss=0.08338, over 5773651.69 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3295, pruned_loss=0.08307, over 5647096.63 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:20:20,149 INFO [optim.py:369] (1/2) 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 10:20:24,939 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.31 vs. limit=2.0
+2023-03-15 10:20:30,853 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:21:07,863 INFO [train.py:968] (1/2) Epoch 30, batch 15850, giga_loss[loss=0.2092, simple_loss=0.298, pruned_loss=0.06019, over 28487.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3273, pruned_loss=0.08226, over 5664850.73 frames. ], libri_tot_loss[loss=0.2464, simple_loss=0.3261, pruned_loss=0.0833, over 5773218.16 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3277, pruned_loss=0.08227, over 5650335.56 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:22:05,735 INFO [train.py:968] (1/2) Epoch 30, batch 15900, giga_loss[loss=0.2191, simple_loss=0.3095, pruned_loss=0.06434, over 28839.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.3248, pruned_loss=0.08085, over 5660413.88 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3255, pruned_loss=0.08304, over 5768369.32 frames. ], giga_tot_loss[loss=0.2439, simple_loss=0.3257, pruned_loss=0.08104, over 5649501.61 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:22:14,402 INFO [optim.py:369] (1/2) 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 10:22:30,624 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1336390.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:23:00,479 INFO [train.py:968] (1/2) Epoch 30, batch 15950, giga_loss[loss=0.2377, simple_loss=0.3227, pruned_loss=0.07638, over 28454.00 frames. ], tot_loss[loss=0.242, simple_loss=0.3232, pruned_loss=0.08044, over 5672869.77 frames. ], libri_tot_loss[loss=0.2454, simple_loss=0.3252, pruned_loss=0.08276, over 5772363.26 frames. ], giga_tot_loss[loss=0.2429, simple_loss=0.3242, pruned_loss=0.08077, over 5657599.26 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:23:39,097 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:23:59,556 INFO [train.py:968] (1/2) Epoch 30, batch 16000, giga_loss[loss=0.2957, simple_loss=0.3605, pruned_loss=0.1154, over 28825.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3233, pruned_loss=0.0802, over 5678001.17 frames. ], libri_tot_loss[loss=0.2451, simple_loss=0.3249, pruned_loss=0.08269, over 5773934.71 frames. ], giga_tot_loss[loss=0.2426, simple_loss=0.3243, pruned_loss=0.08047, over 5661399.98 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:24:08,438 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 16050, giga_loss[loss=0.2256, simple_loss=0.31, pruned_loss=0.0706, over 28896.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.3258, pruned_loss=0.08136, over 5673894.78 frames. ], libri_tot_loss[loss=0.245, simple_loss=0.3248, pruned_loss=0.08264, over 5774611.04 frames. ], giga_tot_loss[loss=0.245, simple_loss=0.3267, pruned_loss=0.08161, over 5659940.01 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:25:51,598 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 16100, giga_loss[loss=0.2159, simple_loss=0.303, pruned_loss=0.06434, over 28822.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.327, pruned_loss=0.0831, over 5668153.08 frames. ], libri_tot_loss[loss=0.2448, simple_loss=0.3246, pruned_loss=0.08249, over 5777408.66 frames. ], giga_tot_loss[loss=0.2474, simple_loss=0.328, pruned_loss=0.08343, over 5651673.88 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:26:13,792 INFO [optim.py:369] (1/2) 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 10:26:32,077 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:26:35,981 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:26:57,193 INFO [train.py:968] (1/2) Epoch 30, batch 16150, giga_loss[loss=0.2585, simple_loss=0.3456, pruned_loss=0.08574, over 28927.00 frames. ], tot_loss[loss=0.2505, simple_loss=0.3308, pruned_loss=0.08509, over 5669646.82 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3245, pruned_loss=0.08245, over 5777370.68 frames. ], giga_tot_loss[loss=0.2513, simple_loss=0.3317, pruned_loss=0.08545, over 5652841.36 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:27:07,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:27:21,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:27:35,050 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5696, 1.6059, 1.3207, 1.1870], device='cuda:1'), covar=tensor([0.0824, 0.0368, 0.0799, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0446, 0.0522, 0.0459], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 10:27:43,358 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:27:50,475 INFO [train.py:968] (1/2) Epoch 30, batch 16200, giga_loss[loss=0.2429, simple_loss=0.3329, pruned_loss=0.07645, over 28642.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3329, pruned_loss=0.08614, over 5665740.16 frames. ], libri_tot_loss[loss=0.2445, simple_loss=0.3242, pruned_loss=0.08238, over 5781391.00 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3343, pruned_loss=0.0866, over 5645387.66 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:28:00,217 INFO [optim.py:369] (1/2) 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 10:28:26,018 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6020, 1.4711, 1.7056, 1.3182], device='cuda:1'), covar=tensor([0.1894, 0.2646, 0.1468, 0.1681], device='cuda:1'), in_proj_covar=tensor([0.0937, 0.0713, 0.0987, 0.0887], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 10:28:29,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:28:33,305 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 16250, giga_loss[loss=0.2442, simple_loss=0.3292, pruned_loss=0.07961, over 28966.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.3333, pruned_loss=0.08586, over 5662463.88 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3241, pruned_loss=0.08226, over 5782839.98 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3346, pruned_loss=0.0864, over 5642941.74 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:29:00,804 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.57 vs. limit=5.0
+2023-03-15 10:29:09,768 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:29:15,784 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.69 vs. limit=2.0
+2023-03-15 10:29:51,685 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1336765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:29:53,958 INFO [train.py:968] (1/2) Epoch 30, batch 16300, giga_loss[loss=0.2435, simple_loss=0.3271, pruned_loss=0.0799, over 28399.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3318, pruned_loss=0.08525, over 5667910.81 frames. ], libri_tot_loss[loss=0.2441, simple_loss=0.3238, pruned_loss=0.08219, over 5786446.62 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.3333, pruned_loss=0.08585, over 5645935.81 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:30:05,949 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1624, 1.2095, 3.3986, 3.0081], device='cuda:1'), covar=tensor([0.1694, 0.2792, 0.0507, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0810, 0.0677, 0.1006, 0.0979], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 10:30:06,353 INFO [optim.py:369] (1/2) 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 10:30:15,645 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336783.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:20,426 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:41,509 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:44,211 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:58,442 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:30:59,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3161, 3.1593, 3.0121, 1.4361], device='cuda:1'), covar=tensor([0.0980, 0.1100, 0.1024, 0.2292], device='cuda:1'), in_proj_covar=tensor([0.1302, 0.1199, 0.1006, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 10:31:00,808 INFO [train.py:968] (1/2) Epoch 30, batch 16350, giga_loss[loss=0.2569, simple_loss=0.3425, pruned_loss=0.08569, over 28675.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.33, pruned_loss=0.08451, over 5671952.58 frames. ], libri_tot_loss[loss=0.2441, simple_loss=0.3238, pruned_loss=0.08221, over 5787188.39 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3312, pruned_loss=0.08496, over 5653616.79 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:31:21,857 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 16400, giga_loss[loss=0.2616, simple_loss=0.3399, pruned_loss=0.09166, over 28882.00 frames. ], tot_loss[loss=0.2501, simple_loss=0.3304, pruned_loss=0.08488, over 5663684.91 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.3242, pruned_loss=0.0825, over 5769862.88 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3312, pruned_loss=0.08506, over 5661309.00 frames. ], batch size: 120, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:32:10,624 INFO [optim.py:369] (1/2) 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 10:32:48,917 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1336908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:32:53,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1336911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:32:59,941 INFO [train.py:968] (1/2) Epoch 30, batch 16450, libri_loss[loss=0.2823, simple_loss=0.3501, pruned_loss=0.1072, over 29531.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3295, pruned_loss=0.08566, over 5646458.63 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3243, pruned_loss=0.08258, over 5762921.16 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3302, pruned_loss=0.08578, over 5648356.35 frames. ], batch size: 89, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:33:23,480 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1336940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:33:55,897 INFO [train.py:968] (1/2) Epoch 30, batch 16500, giga_loss[loss=0.2441, simple_loss=0.3302, pruned_loss=0.079, over 28987.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3278, pruned_loss=0.08504, over 5651066.29 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3242, pruned_loss=0.08259, over 5765846.34 frames. ], giga_tot_loss[loss=0.2495, simple_loss=0.3285, pruned_loss=0.08519, over 5647474.94 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:34:01,403 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 10:34:32,031 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3706, 1.3402, 1.3228, 1.6007], device='cuda:1'), covar=tensor([0.0805, 0.0356, 0.0367, 0.0918], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 10:34:53,733 INFO [train.py:968] (1/2) Epoch 30, batch 16550, giga_loss[loss=0.2474, simple_loss=0.3318, pruned_loss=0.08147, over 28694.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3272, pruned_loss=0.08361, over 5661890.43 frames. ], libri_tot_loss[loss=0.2444, simple_loss=0.3239, pruned_loss=0.08245, over 5766916.70 frames. ], giga_tot_loss[loss=0.248, simple_loss=0.3281, pruned_loss=0.08391, over 5655014.14 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:35:10,272 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:35:48,869 INFO [train.py:968] (1/2) Epoch 30, batch 16600, libri_loss[loss=0.214, simple_loss=0.2965, pruned_loss=0.06572, over 29556.00 frames. ], tot_loss[loss=0.2447, simple_loss=0.3262, pruned_loss=0.08156, over 5673192.29 frames. ], libri_tot_loss[loss=0.2444, simple_loss=0.3238, pruned_loss=0.08243, over 5769186.81 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3271, pruned_loss=0.08183, over 5663303.43 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:35:59,836 INFO [optim.py:369] (1/2) 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 10:36:41,880 INFO [train.py:968] (1/2) Epoch 30, batch 16650, giga_loss[loss=0.2572, simple_loss=0.3469, pruned_loss=0.08376, over 28479.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3284, pruned_loss=0.08118, over 5676820.19 frames. ], libri_tot_loss[loss=0.2443, simple_loss=0.3237, pruned_loss=0.08242, over 5763931.23 frames. ], giga_tot_loss[loss=0.246, simple_loss=0.3294, pruned_loss=0.08136, over 5671344.12 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:37:12,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3668, 3.1141, 1.4670, 1.4897], device='cuda:1'), covar=tensor([0.0969, 0.0326, 0.0956, 0.1365], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0570, 0.0415, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 10:37:32,543 INFO [train.py:968] (1/2) Epoch 30, batch 16700, giga_loss[loss=0.2736, simple_loss=0.3541, pruned_loss=0.09653, over 28703.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3294, pruned_loss=0.0812, over 5682785.92 frames. ], libri_tot_loss[loss=0.2442, simple_loss=0.3236, pruned_loss=0.08237, over 5768560.40 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3304, pruned_loss=0.08134, over 5671854.13 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:37:42,753 INFO [optim.py:369] (1/2) 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 10:38:31,329 INFO [train.py:968] (1/2) Epoch 30, batch 16750, libri_loss[loss=0.199, simple_loss=0.2744, pruned_loss=0.06182, over 29638.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.328, pruned_loss=0.08059, over 5675876.57 frames. ], libri_tot_loss[loss=0.2434, simple_loss=0.3227, pruned_loss=0.08201, over 5771168.57 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3299, pruned_loss=0.081, over 5661573.54 frames. ], batch size: 69, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:39:37,153 INFO [train.py:968] (1/2) Epoch 30, batch 16800, giga_loss[loss=0.2402, simple_loss=0.3221, pruned_loss=0.07909, over 27607.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.3286, pruned_loss=0.08117, over 5668836.97 frames. ], libri_tot_loss[loss=0.2434, simple_loss=0.3228, pruned_loss=0.08199, over 5772900.59 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3301, pruned_loss=0.08151, over 5654736.97 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:39:49,892 INFO [optim.py:369] (1/2) 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 10:40:05,633 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-15 10:40:09,605 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337290.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:40:47,329 INFO [train.py:968] (1/2) Epoch 30, batch 16850, giga_loss[loss=0.2615, simple_loss=0.3424, pruned_loss=0.09036, over 28914.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.3282, pruned_loss=0.08052, over 5665599.65 frames. ], libri_tot_loss[loss=0.2434, simple_loss=0.3228, pruned_loss=0.08202, over 5772404.95 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3295, pruned_loss=0.08074, over 5653841.04 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:41:19,030 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3549, 1.5922, 1.3034, 1.5666], device='cuda:1'), covar=tensor([0.0799, 0.0314, 0.0354, 0.0934], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 10:41:29,715 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337347.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:41:50,807 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.14 vs. limit=5.0
+2023-03-15 10:41:52,907 INFO [train.py:968] (1/2) Epoch 30, batch 16900, giga_loss[loss=0.2952, simple_loss=0.3676, pruned_loss=0.1114, over 28465.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.3296, pruned_loss=0.08086, over 5665533.01 frames. ], libri_tot_loss[loss=0.2437, simple_loss=0.3231, pruned_loss=0.08217, over 5771597.70 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3305, pruned_loss=0.08088, over 5653982.39 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:42:06,754 INFO [optim.py:369] (1/2) 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 10:42:46,574 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:42:46,682 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.56 vs. limit=5.0
+2023-03-15 10:42:47,123 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:42:57,950 INFO [train.py:968] (1/2) Epoch 30, batch 16950, giga_loss[loss=0.2327, simple_loss=0.3229, pruned_loss=0.07126, over 28154.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3324, pruned_loss=0.08222, over 5653535.38 frames. ], libri_tot_loss[loss=0.2434, simple_loss=0.3227, pruned_loss=0.08201, over 5755406.96 frames. ], giga_tot_loss[loss=0.2492, simple_loss=0.3336, pruned_loss=0.08236, over 5655501.52 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:43:01,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7159, 1.8874, 1.5946, 2.0371], device='cuda:1'), covar=tensor([0.2940, 0.3060, 0.3539, 0.2652], device='cuda:1'), in_proj_covar=tensor([0.1622, 0.1164, 0.1437, 0.1017], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 10:43:07,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1835, 1.7586, 1.7115, 1.4964], device='cuda:1'), covar=tensor([0.2494, 0.1942, 0.2274, 0.2255], device='cuda:1'), in_proj_covar=tensor([0.0497, 0.0742, 0.0716, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 10:43:48,011 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9918, 2.1974, 2.1350, 1.8260], device='cuda:1'), covar=tensor([0.2188, 0.2770, 0.2435, 0.2852], device='cuda:1'), in_proj_covar=tensor([0.0496, 0.0742, 0.0716, 0.0686], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 10:44:02,582 INFO [train.py:968] (1/2) Epoch 30, batch 17000, giga_loss[loss=0.2793, simple_loss=0.3542, pruned_loss=0.1022, over 28972.00 frames. ], tot_loss[loss=0.2495, simple_loss=0.3335, pruned_loss=0.08273, over 5653220.04 frames. ], libri_tot_loss[loss=0.2432, simple_loss=0.3225, pruned_loss=0.08197, over 5745393.93 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3348, pruned_loss=0.08288, over 5661375.36 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:44:17,775 INFO [optim.py:369] (1/2) 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 10:44:33,559 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:44:36,752 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 17050, giga_loss[loss=0.2642, simple_loss=0.3437, pruned_loss=0.09241, over 28815.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.3333, pruned_loss=0.08358, over 5659721.00 frames. ], libri_tot_loss[loss=0.243, simple_loss=0.3222, pruned_loss=0.08185, over 5747028.66 frames. ], giga_tot_loss[loss=0.2512, simple_loss=0.3347, pruned_loss=0.08381, over 5664081.09 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:45:16,533 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4413, 1.6716, 1.1604, 1.3251], device='cuda:1'), covar=tensor([0.1035, 0.0580, 0.1121, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0415, 0.0445, 0.0521, 0.0458], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 10:45:17,141 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:45:46,868 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:45:54,935 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:46:00,799 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1337552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:46:20,440 INFO [train.py:968] (1/2) Epoch 30, batch 17100, giga_loss[loss=0.2285, simple_loss=0.3164, pruned_loss=0.07031, over 29117.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3325, pruned_loss=0.08338, over 5668308.35 frames. ], libri_tot_loss[loss=0.2432, simple_loss=0.3225, pruned_loss=0.08196, over 5749825.10 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3336, pruned_loss=0.08349, over 5667990.13 frames. ], batch size: 200, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:46:37,455 INFO [optim.py:369] (1/2) 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 10:46:41,662 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:46:41,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3753, 1.5070, 1.5785, 1.2424], device='cuda:1'), covar=tensor([0.1896, 0.2821, 0.1650, 0.1939], device='cuda:1'), in_proj_covar=tensor([0.0938, 0.0713, 0.0987, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 10:47:13,718 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-15 10:47:30,989 INFO [train.py:968] (1/2) Epoch 30, batch 17150, libri_loss[loss=0.2393, simple_loss=0.3262, pruned_loss=0.07622, over 29512.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3315, pruned_loss=0.08225, over 5666870.16 frames. ], libri_tot_loss[loss=0.2435, simple_loss=0.3227, pruned_loss=0.0821, over 5751336.09 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3322, pruned_loss=0.08223, over 5664450.10 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:48:01,258 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-15 10:48:24,778 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337665.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:48:27,481 INFO [train.py:968] (1/2) Epoch 30, batch 17200, giga_loss[loss=0.233, simple_loss=0.322, pruned_loss=0.07197, over 28680.00 frames. ], tot_loss[loss=0.2481, simple_loss=0.3314, pruned_loss=0.08241, over 5672642.06 frames. ], libri_tot_loss[loss=0.2432, simple_loss=0.3225, pruned_loss=0.08196, over 5755168.05 frames. ], giga_tot_loss[loss=0.2487, simple_loss=0.3323, pruned_loss=0.08253, over 5665545.27 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:48:38,811 INFO [optim.py:369] (1/2) 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 10:49:18,963 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.77 vs. limit=2.0
+2023-03-15 10:49:26,527 INFO [train.py:968] (1/2) Epoch 30, batch 17250, giga_loss[loss=0.2271, simple_loss=0.3206, pruned_loss=0.06681, over 29013.00 frames. ], tot_loss[loss=0.2521, simple_loss=0.335, pruned_loss=0.08456, over 5669903.56 frames. ], libri_tot_loss[loss=0.2431, simple_loss=0.3224, pruned_loss=0.08189, over 5755998.21 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3359, pruned_loss=0.08471, over 5663353.71 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:49:55,572 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.75 vs. limit=2.0
+2023-03-15 10:50:18,247 INFO [train.py:968] (1/2) Epoch 30, batch 17300, libri_loss[loss=0.2046, simple_loss=0.2872, pruned_loss=0.06094, over 29587.00 frames. ], tot_loss[loss=0.2518, simple_loss=0.3342, pruned_loss=0.08464, over 5679169.77 frames. ], libri_tot_loss[loss=0.243, simple_loss=0.3224, pruned_loss=0.08185, over 5759770.22 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3354, pruned_loss=0.08491, over 5667318.00 frames. ], batch size: 75, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:50:28,588 INFO [optim.py:369] (1/2) 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 10:50:32,395 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:51:04,113 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337808.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:51:06,575 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1337811.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:51:14,376 INFO [train.py:968] (1/2) Epoch 30, batch 17350, giga_loss[loss=0.2676, simple_loss=0.3414, pruned_loss=0.09684, over 28792.00 frames. ], tot_loss[loss=0.2488, simple_loss=0.3304, pruned_loss=0.08363, over 5672862.39 frames. ], libri_tot_loss[loss=0.2429, simple_loss=0.3223, pruned_loss=0.08175, over 5761811.51 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3315, pruned_loss=0.08396, over 5660909.09 frames. ], batch size: 263, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:51:41,938 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337840.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 10:51:45,406 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1337843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:52:11,664 INFO [train.py:968] (1/2) Epoch 30, batch 17400, giga_loss[loss=0.2522, simple_loss=0.3369, pruned_loss=0.08376, over 29010.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3299, pruned_loss=0.08401, over 5664888.67 frames. ], libri_tot_loss[loss=0.2427, simple_loss=0.3221, pruned_loss=0.08166, over 5763882.03 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3311, pruned_loss=0.0844, over 5651888.15 frames. ], batch size: 285, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:52:23,831 INFO [optim.py:369] (1/2) 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 10:52:29,658 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5673, 1.8763, 1.5283, 1.4543], device='cuda:1'), covar=tensor([0.2828, 0.2809, 0.3283, 0.2558], device='cuda:1'), in_proj_covar=tensor([0.1618, 0.1162, 0.1435, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 10:53:05,715 INFO [train.py:968] (1/2) Epoch 30, batch 17450, giga_loss[loss=0.3455, simple_loss=0.4113, pruned_loss=0.1398, over 28552.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3346, pruned_loss=0.08727, over 5664487.81 frames. ], libri_tot_loss[loss=0.2426, simple_loss=0.322, pruned_loss=0.08158, over 5767128.68 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.336, pruned_loss=0.08776, over 5648668.75 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:53:06,853 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1337919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:15,802 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1337925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:20,259 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1337928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:44,117 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1337957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:53:52,809 INFO [train.py:968] (1/2) Epoch 30, batch 17500, giga_loss[loss=0.3022, simple_loss=0.3817, pruned_loss=0.1114, over 28830.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3435, pruned_loss=0.09184, over 5673409.76 frames. ], libri_tot_loss[loss=0.2427, simple_loss=0.3221, pruned_loss=0.08164, over 5767534.24 frames. ], giga_tot_loss[loss=0.2648, simple_loss=0.3448, pruned_loss=0.09234, over 5658263.67 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:54:03,234 INFO [optim.py:369] (1/2) 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 10:54:36,735 INFO [train.py:968] (1/2) Epoch 30, batch 17550, libri_loss[loss=0.231, simple_loss=0.3121, pruned_loss=0.0749, over 29571.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3482, pruned_loss=0.09461, over 5680040.22 frames. ], libri_tot_loss[loss=0.2427, simple_loss=0.3221, pruned_loss=0.0817, over 5768489.52 frames. ], giga_tot_loss[loss=0.2699, simple_loss=0.3496, pruned_loss=0.09513, over 5666009.78 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:55:20,108 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:55:22,157 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1338065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:55:25,167 INFO [train.py:968] (1/2) Epoch 30, batch 17600, giga_loss[loss=0.2769, simple_loss=0.3449, pruned_loss=0.1044, over 27742.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3439, pruned_loss=0.09304, over 5672428.12 frames. ], libri_tot_loss[loss=0.2424, simple_loss=0.3217, pruned_loss=0.08153, over 5761954.31 frames. ], giga_tot_loss[loss=0.2669, simple_loss=0.346, pruned_loss=0.09389, over 5665508.06 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:55:33,823 INFO [optim.py:369] (1/2) 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 10:55:45,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1338094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:56:03,025 INFO [train.py:968] (1/2) Epoch 30, batch 17650, giga_loss[loss=0.2479, simple_loss=0.3236, pruned_loss=0.08614, over 29018.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3381, pruned_loss=0.09044, over 5678974.28 frames. ], libri_tot_loss[loss=0.2425, simple_loss=0.3221, pruned_loss=0.08148, over 5759726.73 frames. ], giga_tot_loss[loss=0.2616, simple_loss=0.3401, pruned_loss=0.09154, over 5671922.99 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:56:51,899 INFO [train.py:968] (1/2) Epoch 30, batch 17700, giga_loss[loss=0.2224, simple_loss=0.2999, pruned_loss=0.07249, over 28884.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3309, pruned_loss=0.08716, over 5686144.99 frames. ], libri_tot_loss[loss=0.2429, simple_loss=0.3225, pruned_loss=0.08161, over 5761181.79 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3322, pruned_loss=0.08796, over 5678463.12 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 10:56:59,074 INFO [optim.py:369] (1/2) 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 10:57:32,005 INFO [train.py:968] (1/2) Epoch 30, batch 17750, giga_loss[loss=0.2133, simple_loss=0.2876, pruned_loss=0.06944, over 28471.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.3233, pruned_loss=0.08375, over 5691230.24 frames. ], libri_tot_loss[loss=0.2427, simple_loss=0.3226, pruned_loss=0.08139, over 5763354.38 frames. ], giga_tot_loss[loss=0.2469, simple_loss=0.3244, pruned_loss=0.08475, over 5680210.18 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:57:32,192 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:58:01,265 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:58:11,989 INFO [train.py:968] (1/2) Epoch 30, batch 17800, giga_loss[loss=0.195, simple_loss=0.2746, pruned_loss=0.05768, over 29041.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.3174, pruned_loss=0.08111, over 5700070.52 frames. ], libri_tot_loss[loss=0.2431, simple_loss=0.3232, pruned_loss=0.08149, over 5764151.43 frames. ], giga_tot_loss[loss=0.2407, simple_loss=0.3176, pruned_loss=0.08186, over 5687391.43 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:58:21,537 INFO [optim.py:369] (1/2) 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 10:58:35,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3656, 3.2136, 3.0592, 1.5078], device='cuda:1'), covar=tensor([0.1097, 0.1195, 0.1174, 0.2249], device='cuda:1'), in_proj_covar=tensor([0.1297, 0.1195, 0.1003, 0.0743], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 10:58:49,259 INFO [train.py:968] (1/2) Epoch 30, batch 17850, giga_loss[loss=0.2264, simple_loss=0.2957, pruned_loss=0.07854, over 28751.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.3135, pruned_loss=0.07972, over 5692325.31 frames. ], libri_tot_loss[loss=0.2434, simple_loss=0.3236, pruned_loss=0.0816, over 5759358.17 frames. ], giga_tot_loss[loss=0.2367, simple_loss=0.313, pruned_loss=0.08019, over 5684684.46 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:59:20,425 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:59:23,814 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:59:26,863 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1338364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 10:59:29,186 INFO [train.py:968] (1/2) Epoch 30, batch 17900, giga_loss[loss=0.2268, simple_loss=0.2945, pruned_loss=0.07954, over 28520.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.3101, pruned_loss=0.07848, over 5698398.37 frames. ], libri_tot_loss[loss=0.2439, simple_loss=0.324, pruned_loss=0.08186, over 5760505.83 frames. ], giga_tot_loss[loss=0.2333, simple_loss=0.3093, pruned_loss=0.0786, over 5690795.39 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 10:59:38,852 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1338393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:00:12,475 INFO [train.py:968] (1/2) Epoch 30, batch 17950, giga_loss[loss=0.2012, simple_loss=0.2852, pruned_loss=0.05865, over 28803.00 frames. ], tot_loss[loss=0.2306, simple_loss=0.3069, pruned_loss=0.0771, over 5695655.24 frames. ], libri_tot_loss[loss=0.244, simple_loss=0.3241, pruned_loss=0.08191, over 5761091.25 frames. ], giga_tot_loss[loss=0.2301, simple_loss=0.306, pruned_loss=0.07709, over 5688380.96 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:00:54,047 INFO [train.py:968] (1/2) Epoch 30, batch 18000, giga_loss[loss=0.207, simple_loss=0.2875, pruned_loss=0.06324, over 28947.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.3037, pruned_loss=0.07574, over 5693310.26 frames. ], libri_tot_loss[loss=0.2438, simple_loss=0.324, pruned_loss=0.0818, over 5763354.41 frames. ], giga_tot_loss[loss=0.2271, simple_loss=0.3028, pruned_loss=0.07575, over 5684424.56 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:00:54,047 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 11:01:03,192 INFO [train.py:1012] (1/2) Epoch 30, validation: loss=0.2004, simple_loss=0.3073, pruned_loss=0.04671, over 944034.00 frames. 
+2023-03-15 11:01:03,193 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 11:01:14,968 INFO [optim.py:369] (1/2) 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 11:01:46,133 INFO [train.py:968] (1/2) Epoch 30, batch 18050, giga_loss[loss=0.2048, simple_loss=0.2861, pruned_loss=0.06179, over 28996.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.302, pruned_loss=0.07486, over 5702097.26 frames. ], libri_tot_loss[loss=0.2442, simple_loss=0.3246, pruned_loss=0.08191, over 5764633.08 frames. ], giga_tot_loss[loss=0.2246, simple_loss=0.3, pruned_loss=0.07458, over 5692363.01 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:01:50,463 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.51 vs. limit=5.0
+2023-03-15 11:02:00,857 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3175, 1.1840, 3.8776, 3.2444], device='cuda:1'), covar=tensor([0.1737, 0.3056, 0.0459, 0.1039], device='cuda:1'), in_proj_covar=tensor([0.0811, 0.0677, 0.1010, 0.0982], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 11:02:10,672 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.53 vs. limit=5.0
+2023-03-15 11:02:27,402 INFO [train.py:968] (1/2) Epoch 30, batch 18100, giga_loss[loss=0.209, simple_loss=0.2824, pruned_loss=0.06784, over 28951.00 frames. ], tot_loss[loss=0.224, simple_loss=0.2996, pruned_loss=0.07416, over 5693975.73 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.325, pruned_loss=0.08205, over 5767103.52 frames. ], giga_tot_loss[loss=0.2223, simple_loss=0.2973, pruned_loss=0.07367, over 5682881.51 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:02:27,588 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338568.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:02:38,109 INFO [optim.py:369] (1/2) 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 11:03:08,959 INFO [train.py:968] (1/2) Epoch 30, batch 18150, giga_loss[loss=0.1928, simple_loss=0.2786, pruned_loss=0.05355, over 29048.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.297, pruned_loss=0.07282, over 5698419.09 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3253, pruned_loss=0.08208, over 5768783.76 frames. ], giga_tot_loss[loss=0.2194, simple_loss=0.2943, pruned_loss=0.07224, over 5686633.00 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:03:19,391 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:03:54,795 INFO [train.py:968] (1/2) Epoch 30, batch 18200, libri_loss[loss=0.3034, simple_loss=0.3692, pruned_loss=0.1188, over 19824.00 frames. ], tot_loss[loss=0.219, simple_loss=0.2946, pruned_loss=0.0717, over 5697301.40 frames. ], libri_tot_loss[loss=0.2446, simple_loss=0.3252, pruned_loss=0.08196, over 5762957.62 frames. ], giga_tot_loss[loss=0.2171, simple_loss=0.2919, pruned_loss=0.07111, over 5692591.04 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:04:05,503 INFO [optim.py:369] (1/2) 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 11:04:29,655 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6611, 2.0653, 1.8816, 1.7672], device='cuda:1'), covar=tensor([0.2425, 0.2303, 0.2327, 0.2422], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0749, 0.0722, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:04:34,834 INFO [train.py:968] (1/2) Epoch 30, batch 18250, giga_loss[loss=0.2051, simple_loss=0.2811, pruned_loss=0.06451, over 28892.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.2924, pruned_loss=0.07131, over 5696080.89 frames. ], libri_tot_loss[loss=0.2447, simple_loss=0.3253, pruned_loss=0.08205, over 5763319.63 frames. ], giga_tot_loss[loss=0.2156, simple_loss=0.29, pruned_loss=0.07066, over 5691552.76 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:04:41,118 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:04:41,235 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2657, 1.5578, 1.5395, 1.1621], device='cuda:1'), covar=tensor([0.1742, 0.2469, 0.1408, 0.1680], device='cuda:1'), in_proj_covar=tensor([0.0949, 0.0720, 0.1000, 0.0899], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 11:04:47,603 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:04:47,634 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:25,092 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1707, 1.4989, 1.5753, 1.3164], device='cuda:1'), covar=tensor([0.2455, 0.1758, 0.2498, 0.2061], device='cuda:1'), in_proj_covar=tensor([0.0502, 0.0749, 0.0722, 0.0691], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:05:25,460 INFO [train.py:968] (1/2) Epoch 30, batch 18300, giga_loss[loss=0.2608, simple_loss=0.341, pruned_loss=0.09034, over 28727.00 frames. ], tot_loss[loss=0.227, simple_loss=0.3018, pruned_loss=0.07612, over 5693774.17 frames. ], libri_tot_loss[loss=0.2451, simple_loss=0.3256, pruned_loss=0.08225, over 5764485.78 frames. ], giga_tot_loss[loss=0.225, simple_loss=0.2993, pruned_loss=0.07535, over 5688539.71 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:05:27,692 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 11:05:29,333 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:32,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1338775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:39,378 INFO [optim.py:369] (1/2) 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 11:05:41,037 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1338784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:05:57,728 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 18350, giga_loss[loss=0.3064, simple_loss=0.379, pruned_loss=0.1169, over 28911.00 frames. ], tot_loss[loss=0.2416, simple_loss=0.3162, pruned_loss=0.08351, over 5692632.12 frames. ], libri_tot_loss[loss=0.2454, simple_loss=0.326, pruned_loss=0.08243, over 5766017.67 frames. ], giga_tot_loss[loss=0.2396, simple_loss=0.3137, pruned_loss=0.08271, over 5686162.72 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 11:06:31,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 11:06:48,169 INFO [train.py:968] (1/2) Epoch 30, batch 18400, giga_loss[loss=0.303, simple_loss=0.3845, pruned_loss=0.1108, over 29029.00 frames. ], tot_loss[loss=0.251, simple_loss=0.3264, pruned_loss=0.08782, over 5704731.63 frames. ], libri_tot_loss[loss=0.2457, simple_loss=0.3264, pruned_loss=0.08254, over 5768076.91 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3238, pruned_loss=0.08717, over 5695868.22 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:06:54,152 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1338875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:06:56,799 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:07:20,603 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 18450, giga_loss[loss=0.2431, simple_loss=0.3255, pruned_loss=0.08031, over 28651.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.3333, pruned_loss=0.09055, over 5697682.02 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3264, pruned_loss=0.08256, over 5769699.27 frames. ], giga_tot_loss[loss=0.2557, simple_loss=0.3313, pruned_loss=0.09009, over 5688470.68 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:07:51,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1338943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:08:09,569 INFO [train.py:968] (1/2) Epoch 30, batch 18500, giga_loss[loss=0.2537, simple_loss=0.3446, pruned_loss=0.08137, over 29066.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3368, pruned_loss=0.09125, over 5699928.56 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3265, pruned_loss=0.08263, over 5772609.23 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3353, pruned_loss=0.09098, over 5688799.74 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:08:22,025 INFO [optim.py:369] (1/2) 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 11:08:54,766 INFO [train.py:968] (1/2) Epoch 30, batch 18550, giga_loss[loss=0.2826, simple_loss=0.3562, pruned_loss=0.1045, over 28955.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3373, pruned_loss=0.09081, over 5687309.71 frames. ], libri_tot_loss[loss=0.2459, simple_loss=0.3265, pruned_loss=0.08263, over 5766443.98 frames. ], giga_tot_loss[loss=0.2589, simple_loss=0.3363, pruned_loss=0.09077, over 5683040.02 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:09:15,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7633, 3.6087, 3.4134, 1.9319], device='cuda:1'), covar=tensor([0.0680, 0.0815, 0.0763, 0.2380], device='cuda:1'), in_proj_covar=tensor([0.1292, 0.1194, 0.1002, 0.0744], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 11:09:38,638 INFO [train.py:968] (1/2) Epoch 30, batch 18600, giga_loss[loss=0.2613, simple_loss=0.3444, pruned_loss=0.08907, over 28925.00 frames. ], tot_loss[loss=0.2627, simple_loss=0.34, pruned_loss=0.09273, over 5689407.37 frames. ], libri_tot_loss[loss=0.2458, simple_loss=0.3266, pruned_loss=0.08257, over 5768092.74 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3394, pruned_loss=0.0929, over 5683354.26 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:09:49,138 INFO [optim.py:369] (1/2) 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 11:09:53,829 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:09:56,816 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:03,546 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:10,848 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:21,821 INFO [train.py:968] (1/2) Epoch 30, batch 18650, giga_loss[loss=0.3166, simple_loss=0.3845, pruned_loss=0.1243, over 28912.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3438, pruned_loss=0.0955, over 5698027.71 frames. ], libri_tot_loss[loss=0.2462, simple_loss=0.3268, pruned_loss=0.08275, over 5770547.09 frames. ], giga_tot_loss[loss=0.2673, simple_loss=0.3433, pruned_loss=0.09565, over 5690043.32 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:10:22,039 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:10:53,757 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9419, 1.3027, 1.1196, 0.2580], device='cuda:1'), covar=tensor([0.4778, 0.3829, 0.4587, 0.7110], device='cuda:1'), in_proj_covar=tensor([0.1861, 0.1745, 0.1672, 0.1520], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:10:55,535 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 18700, giga_loss[loss=0.2709, simple_loss=0.356, pruned_loss=0.09295, over 28444.00 frames. ], tot_loss[loss=0.2707, simple_loss=0.347, pruned_loss=0.09716, over 5703550.76 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3276, pruned_loss=0.08303, over 5771595.48 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3465, pruned_loss=0.09754, over 5693439.85 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:11:11,974 INFO [optim.py:369] (1/2) 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 11:11:25,819 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 11:11:40,700 INFO [train.py:968] (1/2) Epoch 30, batch 18750, giga_loss[loss=0.2568, simple_loss=0.3363, pruned_loss=0.08869, over 28937.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3496, pruned_loss=0.09751, over 5701174.94 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.328, pruned_loss=0.08308, over 5765903.13 frames. ], giga_tot_loss[loss=0.2729, simple_loss=0.3495, pruned_loss=0.09813, over 5696886.50 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:12:01,054 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:03,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:10,513 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:12,468 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339253.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:25,026 INFO [train.py:968] (1/2) Epoch 30, batch 18800, giga_loss[loss=0.255, simple_loss=0.3235, pruned_loss=0.09325, over 23410.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3509, pruned_loss=0.09748, over 5701186.95 frames. ], libri_tot_loss[loss=0.2472, simple_loss=0.3282, pruned_loss=0.08315, over 5766957.61 frames. ], giga_tot_loss[loss=0.2735, simple_loss=0.3508, pruned_loss=0.09804, over 5696227.89 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:12:29,939 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:12:35,608 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:51,841 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1339302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:12:54,845 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:13:07,390 INFO [train.py:968] (1/2) Epoch 30, batch 18850, giga_loss[loss=0.2594, simple_loss=0.3471, pruned_loss=0.0859, over 28652.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3515, pruned_loss=0.09713, over 5698163.41 frames. ], libri_tot_loss[loss=0.2471, simple_loss=0.328, pruned_loss=0.0831, over 5768280.20 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3518, pruned_loss=0.09775, over 5692448.73 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:13:09,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3733, 1.6279, 1.3857, 1.5363], device='cuda:1'), covar=tensor([0.0839, 0.0371, 0.0363, 0.0932], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 11:13:19,169 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3753, 1.4564, 1.3449, 1.5610], device='cuda:1'), covar=tensor([0.0847, 0.0356, 0.0355, 0.0933], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 11:13:19,837 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:13:39,942 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:13:44,703 INFO [train.py:968] (1/2) Epoch 30, batch 18900, giga_loss[loss=0.267, simple_loss=0.3477, pruned_loss=0.09321, over 28978.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3493, pruned_loss=0.0946, over 5705620.90 frames. ], libri_tot_loss[loss=0.2467, simple_loss=0.3276, pruned_loss=0.08287, over 5770905.92 frames. ], giga_tot_loss[loss=0.2708, simple_loss=0.3505, pruned_loss=0.0956, over 5697017.40 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:13:56,403 INFO [optim.py:369] (1/2) 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 11:14:24,934 INFO [train.py:968] (1/2) Epoch 30, batch 18950, giga_loss[loss=0.2528, simple_loss=0.3387, pruned_loss=0.08339, over 28824.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3477, pruned_loss=0.0928, over 5708695.79 frames. ], libri_tot_loss[loss=0.2468, simple_loss=0.3278, pruned_loss=0.08289, over 5772177.72 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3487, pruned_loss=0.09368, over 5700253.68 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:15:05,400 INFO [train.py:968] (1/2) Epoch 30, batch 19000, libri_loss[loss=0.2608, simple_loss=0.3455, pruned_loss=0.08806, over 29500.00 frames. ], tot_loss[loss=0.2688, simple_loss=0.3494, pruned_loss=0.09414, over 5705033.01 frames. ], libri_tot_loss[loss=0.2474, simple_loss=0.3284, pruned_loss=0.0832, over 5774384.70 frames. ], giga_tot_loss[loss=0.2696, simple_loss=0.3499, pruned_loss=0.09469, over 5695386.12 frames. ], batch size: 81, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:15:17,879 INFO [optim.py:369] (1/2) 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 11:15:29,650 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:15:40,614 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9571, 1.1285, 1.0538, 0.8861], device='cuda:1'), covar=tensor([0.2389, 0.3112, 0.2012, 0.2487], device='cuda:1'), in_proj_covar=tensor([0.2066, 0.2033, 0.1928, 0.2085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 11:15:44,840 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6027, 1.8474, 1.6096, 1.4075], device='cuda:1'), covar=tensor([0.2058, 0.1884, 0.1970, 0.1935], device='cuda:1'), in_proj_covar=tensor([0.1619, 0.1165, 0.1431, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 11:15:48,745 INFO [train.py:968] (1/2) Epoch 30, batch 19050, giga_loss[loss=0.2906, simple_loss=0.3605, pruned_loss=0.1104, over 28678.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.352, pruned_loss=0.09836, over 5686596.88 frames. ], libri_tot_loss[loss=0.2478, simple_loss=0.3287, pruned_loss=0.08339, over 5766620.23 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3524, pruned_loss=0.09878, over 5684668.33 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:16:14,005 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5956, 1.8625, 1.2829, 1.4583], device='cuda:1'), covar=tensor([0.1121, 0.0660, 0.1031, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0449, 0.0526, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 11:16:30,934 INFO [train.py:968] (1/2) Epoch 30, batch 19100, giga_loss[loss=0.2556, simple_loss=0.3367, pruned_loss=0.08725, over 28107.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3537, pruned_loss=0.1012, over 5679574.01 frames. ], libri_tot_loss[loss=0.2484, simple_loss=0.3294, pruned_loss=0.08367, over 5757556.38 frames. ], giga_tot_loss[loss=0.2789, simple_loss=0.3541, pruned_loss=0.1018, over 5683345.71 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:16:42,605 INFO [optim.py:369] (1/2) 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 11:17:11,785 INFO [train.py:968] (1/2) Epoch 30, batch 19150, giga_loss[loss=0.2571, simple_loss=0.3399, pruned_loss=0.08715, over 28438.00 frames. ], tot_loss[loss=0.2793, simple_loss=0.3538, pruned_loss=0.1023, over 5688394.84 frames. ], libri_tot_loss[loss=0.2486, simple_loss=0.3297, pruned_loss=0.08373, over 5760010.93 frames. ], giga_tot_loss[loss=0.2801, simple_loss=0.3542, pruned_loss=0.103, over 5688329.01 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:17:30,160 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1979, 1.2639, 1.1128, 1.1692], device='cuda:1'), covar=tensor([0.2054, 0.2233, 0.1668, 0.1886], device='cuda:1'), in_proj_covar=tensor([0.2068, 0.2037, 0.1932, 0.2087], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 11:17:31,587 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:17:51,953 INFO [train.py:968] (1/2) Epoch 30, batch 19200, giga_loss[loss=0.2368, simple_loss=0.3201, pruned_loss=0.0768, over 28895.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3517, pruned_loss=0.1015, over 5698946.75 frames. ], libri_tot_loss[loss=0.2491, simple_loss=0.3302, pruned_loss=0.08394, over 5764322.39 frames. ], giga_tot_loss[loss=0.2787, simple_loss=0.3523, pruned_loss=0.1025, over 5692847.62 frames. ], batch size: 112, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:18:00,515 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3892, 1.4814, 3.9902, 3.4568], device='cuda:1'), covar=tensor([0.1701, 0.2755, 0.0440, 0.0920], device='cuda:1'), in_proj_covar=tensor([0.0809, 0.0675, 0.1008, 0.0981], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 11:18:03,587 INFO [optim.py:369] (1/2) 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 11:18:13,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0101, 2.1227, 2.0714, 1.9177], device='cuda:1'), covar=tensor([0.2927, 0.2681, 0.2969, 0.2784], device='cuda:1'), in_proj_covar=tensor([0.2073, 0.2041, 0.1936, 0.2092], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 11:18:37,262 INFO [train.py:968] (1/2) Epoch 30, batch 19250, giga_loss[loss=0.279, simple_loss=0.3466, pruned_loss=0.1057, over 23800.00 frames. ], tot_loss[loss=0.2764, simple_loss=0.3507, pruned_loss=0.1011, over 5677335.21 frames. ], libri_tot_loss[loss=0.2494, simple_loss=0.3306, pruned_loss=0.08415, over 5754335.11 frames. ], giga_tot_loss[loss=0.2774, simple_loss=0.3511, pruned_loss=0.1018, over 5680313.49 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:18:52,405 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:19:16,700 INFO [train.py:968] (1/2) Epoch 30, batch 19300, giga_loss[loss=0.2472, simple_loss=0.3241, pruned_loss=0.08518, over 28515.00 frames. ], tot_loss[loss=0.2726, simple_loss=0.3483, pruned_loss=0.09848, over 5688899.03 frames. ], libri_tot_loss[loss=0.2497, simple_loss=0.3308, pruned_loss=0.08424, over 5758144.92 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3489, pruned_loss=0.09938, over 5686194.56 frames. ], batch size: 65, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:19:28,010 INFO [optim.py:369] (1/2) 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 11:19:44,819 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:19:59,981 INFO [train.py:968] (1/2) Epoch 30, batch 19350, giga_loss[loss=0.2479, simple_loss=0.3293, pruned_loss=0.08321, over 28982.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3444, pruned_loss=0.09572, over 5690257.83 frames. ], libri_tot_loss[loss=0.2493, simple_loss=0.3305, pruned_loss=0.084, over 5761718.97 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3455, pruned_loss=0.09697, over 5683105.13 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:20:45,056 INFO [train.py:968] (1/2) Epoch 30, batch 19400, giga_loss[loss=0.2039, simple_loss=0.2901, pruned_loss=0.0589, over 28966.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.3387, pruned_loss=0.09274, over 5688503.45 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3303, pruned_loss=0.0838, over 5764637.18 frames. ], giga_tot_loss[loss=0.2641, simple_loss=0.3401, pruned_loss=0.0941, over 5679076.45 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:20:49,153 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1339871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:20:58,723 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:21:00,704 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1339883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:21:04,394 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4855, 2.0926, 1.5714, 0.7334], device='cuda:1'), covar=tensor([0.7407, 0.3456, 0.4685, 0.7635], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1739, 0.1664, 0.1515], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:21:25,515 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1339910.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:21:27,823 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1339912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:21:31,700 INFO [train.py:968] (1/2) Epoch 30, batch 19450, libri_loss[loss=0.2815, simple_loss=0.3708, pruned_loss=0.09612, over 29469.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3337, pruned_loss=0.08996, over 5692708.48 frames. ], libri_tot_loss[loss=0.2489, simple_loss=0.3304, pruned_loss=0.08377, over 5766961.92 frames. ], giga_tot_loss[loss=0.2585, simple_loss=0.3347, pruned_loss=0.09119, over 5681675.25 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:22:18,837 INFO [train.py:968] (1/2) Epoch 30, batch 19500, giga_loss[loss=0.2686, simple_loss=0.3558, pruned_loss=0.09073, over 28846.00 frames. ], tot_loss[loss=0.2534, simple_loss=0.3301, pruned_loss=0.08831, over 5688473.73 frames. ], libri_tot_loss[loss=0.2496, simple_loss=0.3309, pruned_loss=0.08411, over 5760818.50 frames. ], giga_tot_loss[loss=0.2543, simple_loss=0.3304, pruned_loss=0.0891, over 5683029.90 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:22:32,986 INFO [optim.py:369] (1/2) 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 11:22:45,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8144, 1.2544, 1.3484, 0.9885], device='cuda:1'), covar=tensor([0.2117, 0.1311, 0.2293, 0.1812], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0759, 0.0732, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:22:59,492 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:02,557 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340017.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:02,987 INFO [train.py:968] (1/2) Epoch 30, batch 19550, giga_loss[loss=0.2418, simple_loss=0.3231, pruned_loss=0.08029, over 28867.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3313, pruned_loss=0.08848, over 5680722.64 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3313, pruned_loss=0.08427, over 5751448.08 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3311, pruned_loss=0.08905, over 5683464.44 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:23:03,243 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:27,513 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:43,534 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:48,465 INFO [train.py:968] (1/2) Epoch 30, batch 19600, giga_loss[loss=0.2386, simple_loss=0.3234, pruned_loss=0.07686, over 28659.00 frames. ], tot_loss[loss=0.2536, simple_loss=0.3313, pruned_loss=0.08791, over 5683616.98 frames. ], libri_tot_loss[loss=0.2499, simple_loss=0.3313, pruned_loss=0.08422, over 5743237.59 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3312, pruned_loss=0.08842, over 5692803.57 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:23:54,382 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:23:56,943 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3784, 2.9998, 1.4794, 1.4557], device='cuda:1'), covar=tensor([0.1009, 0.0333, 0.0923, 0.1376], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0569, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 11:24:00,014 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3902, 4.3085, 1.6219, 1.5849], device='cuda:1'), covar=tensor([0.1126, 0.0295, 0.0958, 0.1471], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0569, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 11:24:01,104 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 19650, giga_loss[loss=0.2455, simple_loss=0.3183, pruned_loss=0.0864, over 28801.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.3307, pruned_loss=0.08796, over 5694071.30 frames. ], libri_tot_loss[loss=0.25, simple_loss=0.3316, pruned_loss=0.08416, over 5746631.04 frames. ], giga_tot_loss[loss=0.2537, simple_loss=0.3304, pruned_loss=0.0885, over 5697480.55 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:24:40,228 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 11:25:03,392 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:06,633 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:09,813 INFO [train.py:968] (1/2) Epoch 30, batch 19700, giga_loss[loss=0.2439, simple_loss=0.3164, pruned_loss=0.08573, over 29050.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3282, pruned_loss=0.08656, over 5708596.27 frames. ], libri_tot_loss[loss=0.2502, simple_loss=0.332, pruned_loss=0.0842, over 5749057.43 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3275, pruned_loss=0.08699, over 5708467.88 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:25:14,152 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:22,775 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:30,777 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4217, 1.4294, 4.0880, 3.4467], device='cuda:1'), covar=tensor([0.1530, 0.2615, 0.0436, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0808, 0.0673, 0.1006, 0.0980], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 11:25:38,747 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340204.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 11:25:42,692 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1340210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:25:48,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2851, 3.0457, 1.3704, 1.4405], device='cuda:1'), covar=tensor([0.1120, 0.0389, 0.0983, 0.1446], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0569, 0.0414, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 11:25:49,461 INFO [train.py:968] (1/2) Epoch 30, batch 19750, giga_loss[loss=0.2782, simple_loss=0.3607, pruned_loss=0.09792, over 28297.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3277, pruned_loss=0.08674, over 5713589.56 frames. ], libri_tot_loss[loss=0.2509, simple_loss=0.3328, pruned_loss=0.08451, over 5752672.60 frames. ], giga_tot_loss[loss=0.25, simple_loss=0.3262, pruned_loss=0.08687, over 5709057.60 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:25:55,328 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9063, 1.3364, 1.3195, 1.1772], device='cuda:1'), covar=tensor([0.2486, 0.1456, 0.2746, 0.1938], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0761, 0.0734, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:26:02,186 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.13 vs. limit=2.0
+2023-03-15 11:26:33,652 INFO [train.py:968] (1/2) Epoch 30, batch 19800, giga_loss[loss=0.2426, simple_loss=0.3206, pruned_loss=0.0823, over 28598.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3261, pruned_loss=0.08668, over 5716544.49 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.3327, pruned_loss=0.0844, over 5754770.55 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3249, pruned_loss=0.08691, over 5710573.85 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:26:49,218 INFO [optim.py:369] (1/2) 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 11:26:49,462 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:26:57,128 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0644, 3.2021, 2.1843, 1.2874], device='cuda:1'), covar=tensor([0.9322, 0.3112, 0.4445, 0.7918], device='cuda:1'), in_proj_covar=tensor([0.1855, 0.1735, 0.1663, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:27:16,101 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 19850, giga_loss[loss=0.2176, simple_loss=0.2915, pruned_loss=0.07184, over 28970.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3234, pruned_loss=0.08554, over 5720710.85 frames. ], libri_tot_loss[loss=0.2506, simple_loss=0.3328, pruned_loss=0.08425, over 5757919.23 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3223, pruned_loss=0.08588, over 5712572.83 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:27:17,888 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:27:41,743 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:27:42,333 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4436, 1.8152, 1.4459, 1.6462], device='cuda:1'), covar=tensor([0.0786, 0.0317, 0.0355, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 11:27:56,904 INFO [train.py:968] (1/2) Epoch 30, batch 19900, giga_loss[loss=0.2364, simple_loss=0.2995, pruned_loss=0.08668, over 28565.00 frames. ], tot_loss[loss=0.2452, simple_loss=0.3212, pruned_loss=0.08462, over 5727122.38 frames. ], libri_tot_loss[loss=0.2508, simple_loss=0.333, pruned_loss=0.08425, over 5761103.67 frames. ], giga_tot_loss[loss=0.2449, simple_loss=0.32, pruned_loss=0.08491, over 5717295.54 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:28:12,793 INFO [optim.py:369] (1/2) 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 11:28:30,381 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3296, 1.5894, 1.5305, 1.4094], device='cuda:1'), covar=tensor([0.2135, 0.2209, 0.2408, 0.2384], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0760, 0.0734, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:28:41,542 INFO [train.py:968] (1/2) Epoch 30, batch 19950, giga_loss[loss=0.237, simple_loss=0.3078, pruned_loss=0.08312, over 28685.00 frames. ], tot_loss[loss=0.244, simple_loss=0.32, pruned_loss=0.08399, over 5724987.28 frames. ], libri_tot_loss[loss=0.2511, simple_loss=0.3336, pruned_loss=0.08427, over 5763584.03 frames. ], giga_tot_loss[loss=0.2433, simple_loss=0.3181, pruned_loss=0.08421, over 5714027.48 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:28:49,199 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:28:49,361 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.09 vs. limit=5.0
+2023-03-15 11:28:51,052 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4972, 3.6002, 1.5743, 1.6178], device='cuda:1'), covar=tensor([0.1033, 0.0316, 0.0910, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0568, 0.0413, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 11:28:51,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:28:58,383 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:29:10,077 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:29:17,297 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:29:23,531 INFO [train.py:968] (1/2) Epoch 30, batch 20000, giga_loss[loss=0.2139, simple_loss=0.3015, pruned_loss=0.06312, over 29023.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.3182, pruned_loss=0.0827, over 5732994.56 frames. ], libri_tot_loss[loss=0.2513, simple_loss=0.334, pruned_loss=0.08432, over 5764727.14 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.316, pruned_loss=0.08282, over 5722035.82 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:29:35,446 INFO [optim.py:369] (1/2) 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 11:30:02,070 INFO [train.py:968] (1/2) Epoch 30, batch 20050, giga_loss[loss=0.2155, simple_loss=0.2955, pruned_loss=0.06773, over 28906.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.3177, pruned_loss=0.08206, over 5736435.82 frames. ], libri_tot_loss[loss=0.2519, simple_loss=0.3348, pruned_loss=0.08449, over 5769313.06 frames. ], giga_tot_loss[loss=0.2393, simple_loss=0.3147, pruned_loss=0.08196, over 5722756.49 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:30:08,806 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.60 vs. limit=5.0
+2023-03-15 11:30:20,346 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-15 11:30:43,725 INFO [train.py:968] (1/2) Epoch 30, batch 20100, giga_loss[loss=0.2691, simple_loss=0.3482, pruned_loss=0.095, over 28746.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.3176, pruned_loss=0.08192, over 5744548.39 frames. ], libri_tot_loss[loss=0.2523, simple_loss=0.3354, pruned_loss=0.08462, over 5771747.56 frames. ], giga_tot_loss[loss=0.2387, simple_loss=0.3142, pruned_loss=0.08163, over 5730372.92 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:30:53,386 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1340579.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 11:30:55,079 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:30:58,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:30:58,798 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 11:31:04,667 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:06,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:23,254 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:28,923 INFO [train.py:968] (1/2) Epoch 30, batch 20150, giga_loss[loss=0.2978, simple_loss=0.3705, pruned_loss=0.1125, over 28852.00 frames. ], tot_loss[loss=0.2477, simple_loss=0.3237, pruned_loss=0.08584, over 5730974.22 frames. ], libri_tot_loss[loss=0.2531, simple_loss=0.3363, pruned_loss=0.08499, over 5773584.42 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3197, pruned_loss=0.08525, over 5717330.59 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:31:36,360 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:31:41,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6278, 1.7476, 1.6474, 1.5307], device='cuda:1'), covar=tensor([0.3078, 0.2969, 0.2617, 0.2969], device='cuda:1'), in_proj_covar=tensor([0.2071, 0.2031, 0.1939, 0.2090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 11:32:05,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-15 11:32:19,287 INFO [train.py:968] (1/2) Epoch 30, batch 20200, giga_loss[loss=0.2649, simple_loss=0.3472, pruned_loss=0.09128, over 28957.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3297, pruned_loss=0.08967, over 5727431.26 frames. ], libri_tot_loss[loss=0.2533, simple_loss=0.3364, pruned_loss=0.08505, over 5774334.69 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3263, pruned_loss=0.08922, over 5715135.26 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:32:22,001 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5149, 4.3680, 4.1383, 2.1651], device='cuda:1'), covar=tensor([0.0563, 0.0705, 0.0647, 0.1917], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1194, 0.1002, 0.0749], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0016, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 11:32:38,660 INFO [optim.py:369] (1/2) 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 11:32:44,585 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.76 vs. limit=2.0
+2023-03-15 11:33:08,153 INFO [train.py:968] (1/2) Epoch 30, batch 20250, giga_loss[loss=0.2664, simple_loss=0.3478, pruned_loss=0.0925, over 28845.00 frames. ], tot_loss[loss=0.2632, simple_loss=0.337, pruned_loss=0.0947, over 5710656.21 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3358, pruned_loss=0.08459, over 5776501.61 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3348, pruned_loss=0.09514, over 5695975.88 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:33:11,548 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5432, 1.9908, 1.5887, 1.6925], device='cuda:1'), covar=tensor([0.0757, 0.0289, 0.0330, 0.0883], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 11:33:12,301 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340722.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 11:33:15,958 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1340725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 11:33:17,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1340728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:33:20,653 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1340754.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 11:33:47,069 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 20300, giga_loss[loss=0.2658, simple_loss=0.3499, pruned_loss=0.09086, over 28917.00 frames. ], tot_loss[loss=0.2672, simple_loss=0.3416, pruned_loss=0.09644, over 5710610.97 frames. ], libri_tot_loss[loss=0.2529, simple_loss=0.3363, pruned_loss=0.08478, over 5780917.54 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3395, pruned_loss=0.09702, over 5692102.87 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:34:10,502 INFO [optim.py:369] (1/2) 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 11:34:37,291 INFO [train.py:968] (1/2) Epoch 30, batch 20350, giga_loss[loss=0.3061, simple_loss=0.3818, pruned_loss=0.1152, over 28609.00 frames. ], tot_loss[loss=0.2699, simple_loss=0.345, pruned_loss=0.09742, over 5692619.44 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.3361, pruned_loss=0.08449, over 5781327.02 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3437, pruned_loss=0.0985, over 5675272.14 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:35:04,259 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 20400, libri_loss[loss=0.1923, simple_loss=0.2811, pruned_loss=0.05173, over 29646.00 frames. ], tot_loss[loss=0.2742, simple_loss=0.349, pruned_loss=0.0997, over 5691327.44 frames. ], libri_tot_loss[loss=0.2525, simple_loss=0.336, pruned_loss=0.08448, over 5781864.05 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3485, pruned_loss=0.1012, over 5673020.99 frames. ], batch size: 73, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:35:39,972 INFO [optim.py:369] (1/2) 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 11:35:47,296 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6144, 2.4079, 1.7604, 0.8567], device='cuda:1'), covar=tensor([0.7311, 0.3605, 0.4660, 0.7797], device='cuda:1'), in_proj_covar=tensor([0.1860, 0.1742, 0.1670, 0.1519], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:36:07,991 INFO [train.py:968] (1/2) Epoch 30, batch 20450, giga_loss[loss=0.359, simple_loss=0.3957, pruned_loss=0.1611, over 26788.00 frames. ], tot_loss[loss=0.2807, simple_loss=0.3547, pruned_loss=0.1033, over 5689011.46 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3362, pruned_loss=0.08455, over 5783618.58 frames. ], giga_tot_loss[loss=0.2822, simple_loss=0.3547, pruned_loss=0.1049, over 5670234.07 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:36:31,352 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6788, 1.9025, 1.3336, 1.3922], device='cuda:1'), covar=tensor([0.1045, 0.0456, 0.0894, 0.1058], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0452, 0.0529, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 11:36:54,679 INFO [train.py:968] (1/2) Epoch 30, batch 20500, giga_loss[loss=0.221, simple_loss=0.3075, pruned_loss=0.0673, over 28618.00 frames. ], tot_loss[loss=0.2753, simple_loss=0.3505, pruned_loss=0.1, over 5691573.74 frames. ], libri_tot_loss[loss=0.2526, simple_loss=0.3362, pruned_loss=0.08448, over 5785908.19 frames. ], giga_tot_loss[loss=0.277, simple_loss=0.3507, pruned_loss=0.1016, over 5672985.32 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:37:10,877 INFO [optim.py:369] (1/2) 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 11:37:11,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0182, 3.2868, 2.2246, 1.0708], device='cuda:1'), covar=tensor([1.0304, 0.3285, 0.4160, 0.8697], device='cuda:1'), in_proj_covar=tensor([0.1863, 0.1743, 0.1675, 0.1521], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:37:33,168 INFO [train.py:968] (1/2) Epoch 30, batch 20550, giga_loss[loss=0.2633, simple_loss=0.3457, pruned_loss=0.09044, over 28917.00 frames. ], tot_loss[loss=0.2721, simple_loss=0.3484, pruned_loss=0.09788, over 5705219.55 frames. ], libri_tot_loss[loss=0.2535, simple_loss=0.3368, pruned_loss=0.08508, over 5788984.53 frames. ], giga_tot_loss[loss=0.2737, simple_loss=0.3486, pruned_loss=0.09941, over 5682561.31 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:38:20,341 INFO [train.py:968] (1/2) Epoch 30, batch 20600, giga_loss[loss=0.2741, simple_loss=0.345, pruned_loss=0.1016, over 28373.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3475, pruned_loss=0.09747, over 5693636.92 frames. ], libri_tot_loss[loss=0.2537, simple_loss=0.337, pruned_loss=0.0852, over 5776973.70 frames. ], giga_tot_loss[loss=0.2727, simple_loss=0.3478, pruned_loss=0.09878, over 5684104.81 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:38:38,022 INFO [optim.py:369] (1/2) 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 11:39:06,133 INFO [train.py:968] (1/2) Epoch 30, batch 20650, giga_loss[loss=0.2842, simple_loss=0.3605, pruned_loss=0.1039, over 28550.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3475, pruned_loss=0.09685, over 5690714.03 frames. ], libri_tot_loss[loss=0.2539, simple_loss=0.3372, pruned_loss=0.08535, over 5775530.56 frames. ], giga_tot_loss[loss=0.2717, simple_loss=0.3477, pruned_loss=0.09792, over 5683481.42 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:39:50,187 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6314, 1.8295, 1.5184, 1.8735], device='cuda:1'), covar=tensor([0.2487, 0.2745, 0.2986, 0.2514], device='cuda:1'), in_proj_covar=tensor([0.1619, 0.1168, 0.1434, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 11:39:51,742 INFO [train.py:968] (1/2) Epoch 30, batch 20700, libri_loss[loss=0.2439, simple_loss=0.3376, pruned_loss=0.07513, over 29695.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3481, pruned_loss=0.0973, over 5695076.30 frames. ], libri_tot_loss[loss=0.2536, simple_loss=0.3369, pruned_loss=0.0852, over 5775979.69 frames. ], giga_tot_loss[loss=0.2728, simple_loss=0.3487, pruned_loss=0.09846, over 5687683.42 frames. ], batch size: 88, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:40:10,556 INFO [optim.py:369] (1/2) 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 11:40:13,800 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3492, 1.7069, 1.5021, 1.6245], device='cuda:1'), covar=tensor([0.0820, 0.0330, 0.0330, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 11:40:34,715 INFO [train.py:968] (1/2) Epoch 30, batch 20750, giga_loss[loss=0.265, simple_loss=0.3441, pruned_loss=0.09296, over 28822.00 frames. ], tot_loss[loss=0.2732, simple_loss=0.3496, pruned_loss=0.0984, over 5707196.64 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3374, pruned_loss=0.08556, over 5777701.04 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3501, pruned_loss=0.09954, over 5696599.87 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:40:38,665 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1341223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:41:22,460 INFO [train.py:968] (1/2) Epoch 30, batch 20800, giga_loss[loss=0.2537, simple_loss=0.3355, pruned_loss=0.08596, over 28720.00 frames. ], tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.09978, over 5689593.55 frames. ], libri_tot_loss[loss=0.254, simple_loss=0.3371, pruned_loss=0.08544, over 5779639.40 frames. ], giga_tot_loss[loss=0.2767, simple_loss=0.3514, pruned_loss=0.101, over 5678489.70 frames. ], batch size: 66, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:41:37,615 INFO [optim.py:369] (1/2) 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 11:41:41,812 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5475, 3.4032, 3.1600, 1.7746], device='cuda:1'), covar=tensor([0.0788, 0.0843, 0.0784, 0.2388], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1200, 0.1006, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 11:41:57,243 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7384, 1.8945, 1.6344, 1.6849], device='cuda:1'), covar=tensor([0.2783, 0.2785, 0.3177, 0.2449], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1166, 0.1432, 0.1014], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 11:42:07,520 INFO [train.py:968] (1/2) Epoch 30, batch 20850, giga_loss[loss=0.3016, simple_loss=0.3647, pruned_loss=0.1192, over 29005.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.3524, pruned_loss=0.1016, over 5696466.65 frames. ], libri_tot_loss[loss=0.2543, simple_loss=0.3373, pruned_loss=0.08562, over 5783926.35 frames. ], giga_tot_loss[loss=0.2795, simple_loss=0.3533, pruned_loss=0.1028, over 5681667.18 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:42:47,682 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1341366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:42:48,600 INFO [train.py:968] (1/2) Epoch 30, batch 20900, giga_loss[loss=0.2618, simple_loss=0.3385, pruned_loss=0.09259, over 28624.00 frames. ], tot_loss[loss=0.2774, simple_loss=0.3523, pruned_loss=0.1012, over 5705414.66 frames. ], libri_tot_loss[loss=0.2552, simple_loss=0.338, pruned_loss=0.08617, over 5785521.66 frames. ], giga_tot_loss[loss=0.2786, simple_loss=0.3528, pruned_loss=0.1022, over 5689395.18 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:42:49,663 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1341369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:43:03,988 INFO [optim.py:369] (1/2) 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 11:43:12,039 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 20950, giga_loss[loss=0.2588, simple_loss=0.3372, pruned_loss=0.09022, over 27634.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3512, pruned_loss=0.09984, over 5696455.11 frames. ], libri_tot_loss[loss=0.2555, simple_loss=0.3383, pruned_loss=0.08633, over 5777034.58 frames. ], giga_tot_loss[loss=0.2764, simple_loss=0.3515, pruned_loss=0.1006, over 5689916.96 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:43:48,937 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8787, 1.9725, 1.7091, 2.1319], device='cuda:1'), covar=tensor([0.3137, 0.2983, 0.3493, 0.2318], device='cuda:1'), in_proj_covar=tensor([0.1616, 0.1165, 0.1430, 0.1012], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 11:44:03,116 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1341457.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:44:12,842 INFO [train.py:968] (1/2) Epoch 30, batch 21000, giga_loss[loss=0.2985, simple_loss=0.3759, pruned_loss=0.1105, over 28691.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3506, pruned_loss=0.09807, over 5702284.03 frames. ], libri_tot_loss[loss=0.2551, simple_loss=0.3379, pruned_loss=0.08618, over 5779608.47 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3514, pruned_loss=0.09907, over 5693393.92 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:44:12,842 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 11:44:23,034 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 11:44:38,090 INFO [optim.py:369] (1/2) 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 11:45:05,061 INFO [train.py:968] (1/2) Epoch 30, batch 21050, giga_loss[loss=0.2383, simple_loss=0.3267, pruned_loss=0.07493, over 29088.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3502, pruned_loss=0.09786, over 5676999.77 frames. ], libri_tot_loss[loss=0.2559, simple_loss=0.3383, pruned_loss=0.08671, over 5753794.77 frames. ], giga_tot_loss[loss=0.2739, simple_loss=0.3509, pruned_loss=0.09851, over 5692742.94 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:45:43,727 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0297, 1.3155, 1.1299, 0.3210], device='cuda:1'), covar=tensor([0.4701, 0.3638, 0.4954, 0.7338], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1730, 0.1660, 0.1510], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:45:45,502 INFO [train.py:968] (1/2) Epoch 30, batch 21100, libri_loss[loss=0.2807, simple_loss=0.3634, pruned_loss=0.09904, over 29770.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3488, pruned_loss=0.09745, over 5684501.40 frames. ], libri_tot_loss[loss=0.2563, simple_loss=0.3387, pruned_loss=0.08696, over 5747092.67 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3492, pruned_loss=0.09792, over 5701687.23 frames. ], batch size: 87, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:45:47,330 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9917, 3.8506, 3.6998, 1.7344], device='cuda:1'), covar=tensor([0.0825, 0.0910, 0.0969, 0.2031], device='cuda:1'), in_proj_covar=tensor([0.1301, 0.1202, 0.1007, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 11:46:00,281 INFO [optim.py:369] (1/2) 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 11:46:25,829 INFO [train.py:968] (1/2) Epoch 30, batch 21150, giga_loss[loss=0.2392, simple_loss=0.3165, pruned_loss=0.08089, over 28602.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3467, pruned_loss=0.09643, over 5693283.55 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.339, pruned_loss=0.08726, over 5748265.54 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.3468, pruned_loss=0.09671, over 5704895.04 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:47:05,308 INFO [train.py:968] (1/2) Epoch 30, batch 21200, giga_loss[loss=0.2146, simple_loss=0.2982, pruned_loss=0.06551, over 28789.00 frames. ], tot_loss[loss=0.2679, simple_loss=0.3445, pruned_loss=0.09565, over 5697679.39 frames. ], libri_tot_loss[loss=0.2568, simple_loss=0.3388, pruned_loss=0.08739, over 5749222.38 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.345, pruned_loss=0.096, over 5704495.21 frames. ], batch size: 60, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:47:24,616 INFO [optim.py:369] (1/2) 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 11:47:49,107 INFO [train.py:968] (1/2) Epoch 30, batch 21250, giga_loss[loss=0.2888, simple_loss=0.361, pruned_loss=0.1083, over 28915.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3456, pruned_loss=0.09684, over 5699913.80 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.339, pruned_loss=0.08754, over 5750987.25 frames. ], giga_tot_loss[loss=0.2701, simple_loss=0.346, pruned_loss=0.0971, over 5703056.82 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:47:58,172 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 11:48:32,114 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 11:48:32,916 INFO [train.py:968] (1/2) Epoch 30, batch 21300, giga_loss[loss=0.2508, simple_loss=0.3333, pruned_loss=0.08415, over 28726.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3459, pruned_loss=0.0968, over 5706683.44 frames. ], libri_tot_loss[loss=0.257, simple_loss=0.3389, pruned_loss=0.08754, over 5753457.53 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3464, pruned_loss=0.09712, over 5706508.94 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:48:50,306 INFO [optim.py:369] (1/2) 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 11:48:55,498 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1341795.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:49:08,711 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3821, 1.4588, 1.2609, 1.5199], device='cuda:1'), covar=tensor([0.0824, 0.0353, 0.0368, 0.0959], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 11:49:15,789 INFO [train.py:968] (1/2) Epoch 30, batch 21350, giga_loss[loss=0.2661, simple_loss=0.3468, pruned_loss=0.09274, over 28560.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3439, pruned_loss=0.09463, over 5704517.18 frames. ], libri_tot_loss[loss=0.2571, simple_loss=0.339, pruned_loss=0.08762, over 5754005.06 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3441, pruned_loss=0.09491, over 5703038.73 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:49:26,670 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1341832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:49:46,485 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-15 11:49:56,103 INFO [train.py:968] (1/2) Epoch 30, batch 21400, giga_loss[loss=0.2915, simple_loss=0.3609, pruned_loss=0.111, over 28319.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3436, pruned_loss=0.0941, over 5719164.87 frames. ], libri_tot_loss[loss=0.258, simple_loss=0.3395, pruned_loss=0.08824, over 5759925.00 frames. ], giga_tot_loss[loss=0.2657, simple_loss=0.3435, pruned_loss=0.09395, over 5711045.48 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:49:59,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3580, 2.7015, 2.5274, 2.4970], device='cuda:1'), covar=tensor([0.2482, 0.2245, 0.2154, 0.2165], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0762, 0.0736, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:50:12,927 INFO [optim.py:369] (1/2) 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 11:50:15,012 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4724, 1.7852, 1.6428, 1.6103], device='cuda:1'), covar=tensor([0.2299, 0.2139, 0.2586, 0.2158], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0762, 0.0736, 0.0701], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:50:36,565 INFO [train.py:968] (1/2) Epoch 30, batch 21450, giga_loss[loss=0.2393, simple_loss=0.3262, pruned_loss=0.07616, over 29028.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3433, pruned_loss=0.0942, over 5728541.38 frames. ], libri_tot_loss[loss=0.2584, simple_loss=0.3397, pruned_loss=0.08861, over 5763180.10 frames. ], giga_tot_loss[loss=0.2655, simple_loss=0.3431, pruned_loss=0.0939, over 5717857.37 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:50:41,188 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5010, 1.6253, 1.6643, 1.4428], device='cuda:1'), covar=tensor([0.3338, 0.3263, 0.2580, 0.3181], device='cuda:1'), in_proj_covar=tensor([0.2087, 0.2052, 0.1959, 0.2104], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 11:51:09,344 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3230, 1.8625, 1.4132, 0.5969], device='cuda:1'), covar=tensor([0.6041, 0.3157, 0.4406, 0.7526], device='cuda:1'), in_proj_covar=tensor([0.1851, 0.1727, 0.1658, 0.1507], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:51:13,310 INFO [train.py:968] (1/2) Epoch 30, batch 21500, giga_loss[loss=0.2224, simple_loss=0.3085, pruned_loss=0.06816, over 28880.00 frames. ], tot_loss[loss=0.264, simple_loss=0.3412, pruned_loss=0.0934, over 5725991.43 frames. ], libri_tot_loss[loss=0.2593, simple_loss=0.3403, pruned_loss=0.08916, over 5759286.41 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3406, pruned_loss=0.09281, over 5720005.80 frames. ], batch size: 186, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:51:14,719 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1442, 1.4289, 1.4624, 1.2347], device='cuda:1'), covar=tensor([0.2256, 0.1835, 0.2515, 0.2120], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0759, 0.0733, 0.0699], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 11:51:19,039 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1341975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:20,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1341977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:21,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1341978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:30,784 INFO [optim.py:369] (1/2) 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 11:51:44,526 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:51:51,836 INFO [train.py:968] (1/2) Epoch 30, batch 21550, libri_loss[loss=0.2739, simple_loss=0.3517, pruned_loss=0.09801, over 29156.00 frames. ], tot_loss[loss=0.262, simple_loss=0.339, pruned_loss=0.09253, over 5725360.10 frames. ], libri_tot_loss[loss=0.2601, simple_loss=0.3407, pruned_loss=0.0897, over 5764368.32 frames. ], giga_tot_loss[loss=0.2607, simple_loss=0.338, pruned_loss=0.09165, over 5714544.84 frames. ], batch size: 101, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:52:34,156 INFO [train.py:968] (1/2) Epoch 30, batch 21600, giga_loss[loss=0.2987, simple_loss=0.3738, pruned_loss=0.1118, over 28238.00 frames. ], tot_loss[loss=0.2626, simple_loss=0.3392, pruned_loss=0.09298, over 5725617.44 frames. ], libri_tot_loss[loss=0.2604, simple_loss=0.341, pruned_loss=0.08991, over 5762431.07 frames. ], giga_tot_loss[loss=0.2612, simple_loss=0.3382, pruned_loss=0.09211, over 5718271.00 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:52:51,901 INFO [optim.py:369] (1/2) 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 11:53:14,593 INFO [train.py:968] (1/2) Epoch 30, batch 21650, giga_loss[loss=0.2406, simple_loss=0.3191, pruned_loss=0.08106, over 29081.00 frames. ], tot_loss[loss=0.2616, simple_loss=0.3378, pruned_loss=0.09274, over 5724227.20 frames. ], libri_tot_loss[loss=0.2605, simple_loss=0.3408, pruned_loss=0.09007, over 5764790.95 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3371, pruned_loss=0.09197, over 5715293.55 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:53:56,831 INFO [train.py:968] (1/2) Epoch 30, batch 21700, giga_loss[loss=0.286, simple_loss=0.3507, pruned_loss=0.1107, over 28311.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3366, pruned_loss=0.09262, over 5717097.19 frames. ], libri_tot_loss[loss=0.2609, simple_loss=0.3411, pruned_loss=0.09036, over 5758563.46 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3357, pruned_loss=0.09178, over 5715337.29 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:53:57,734 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1342169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:53:58,498 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1342170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:54:14,392 INFO [optim.py:369] (1/2) 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 11:54:25,899 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6257, 1.6133, 1.8158, 1.3951], device='cuda:1'), covar=tensor([0.1914, 0.2734, 0.1596, 0.1790], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0721, 0.0995, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 11:54:37,053 INFO [train.py:968] (1/2) Epoch 30, batch 21750, giga_loss[loss=0.2289, simple_loss=0.2996, pruned_loss=0.07911, over 28952.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3349, pruned_loss=0.09189, over 5716713.53 frames. ], libri_tot_loss[loss=0.2623, simple_loss=0.3422, pruned_loss=0.09122, over 5758893.92 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3329, pruned_loss=0.09046, over 5714153.77 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:55:21,893 INFO [train.py:968] (1/2) Epoch 30, batch 21800, giga_loss[loss=0.2391, simple_loss=0.3181, pruned_loss=0.08011, over 28845.00 frames. ], tot_loss[loss=0.257, simple_loss=0.3321, pruned_loss=0.09097, over 5713933.77 frames. ], libri_tot_loss[loss=0.2624, simple_loss=0.3422, pruned_loss=0.09125, over 5759618.45 frames. ], giga_tot_loss[loss=0.2551, simple_loss=0.3305, pruned_loss=0.08982, over 5710994.38 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:55:39,192 INFO [optim.py:369] (1/2) 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 11:55:57,209 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1342313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:55:57,303 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:56:00,119 INFO [train.py:968] (1/2) Epoch 30, batch 21850, libri_loss[loss=0.2786, simple_loss=0.3573, pruned_loss=0.0999, over 26032.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3307, pruned_loss=0.0905, over 5710604.46 frames. ], libri_tot_loss[loss=0.2633, simple_loss=0.3428, pruned_loss=0.09191, over 5760217.01 frames. ], giga_tot_loss[loss=0.2531, simple_loss=0.3285, pruned_loss=0.08891, over 5706310.93 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:56:10,996 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3831, 1.2563, 4.0550, 3.5139], device='cuda:1'), covar=tensor([0.1657, 0.2978, 0.0395, 0.1246], device='cuda:1'), in_proj_covar=tensor([0.0812, 0.0677, 0.1009, 0.0985], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 11:56:23,187 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:56:29,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1342352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:56:42,100 INFO [train.py:968] (1/2) Epoch 30, batch 21900, giga_loss[loss=0.2988, simple_loss=0.3806, pruned_loss=0.1085, over 27690.00 frames. ], tot_loss[loss=0.2556, simple_loss=0.3308, pruned_loss=0.09022, over 5707568.55 frames. ], libri_tot_loss[loss=0.2636, simple_loss=0.3429, pruned_loss=0.09216, over 5761637.40 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3285, pruned_loss=0.08866, over 5701348.71 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:56:59,634 INFO [optim.py:369] (1/2) 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 11:57:24,643 INFO [train.py:968] (1/2) Epoch 30, batch 21950, giga_loss[loss=0.2595, simple_loss=0.3295, pruned_loss=0.09469, over 28567.00 frames. ], tot_loss[loss=0.2593, simple_loss=0.3343, pruned_loss=0.09213, over 5712822.09 frames. ], libri_tot_loss[loss=0.264, simple_loss=0.3429, pruned_loss=0.09249, over 5766388.47 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3321, pruned_loss=0.09053, over 5702218.85 frames. ], batch size: 78, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 11:57:25,522 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4655, 2.1142, 1.4585, 0.7404], device='cuda:1'), covar=tensor([0.7992, 0.3612, 0.5469, 0.8138], device='cuda:1'), in_proj_covar=tensor([0.1859, 0.1732, 0.1667, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 11:57:34,654 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.3119, 4.1700, 3.9551, 1.7651], device='cuda:1'), covar=tensor([0.0575, 0.0707, 0.0725, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.1305, 0.1205, 0.1011, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 11:58:10,043 INFO [train.py:968] (1/2) Epoch 30, batch 22000, libri_loss[loss=0.2391, simple_loss=0.3154, pruned_loss=0.08143, over 29573.00 frames. ], tot_loss[loss=0.2615, simple_loss=0.3369, pruned_loss=0.09306, over 5712639.72 frames. ], libri_tot_loss[loss=0.2639, simple_loss=0.3427, pruned_loss=0.09254, over 5760721.14 frames. ], giga_tot_loss[loss=0.2593, simple_loss=0.3351, pruned_loss=0.09175, over 5706946.36 frames. ], batch size: 75, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:58:26,708 INFO [optim.py:369] (1/2) 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 11:58:32,246 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1342495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:58:34,879 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:58:51,133 INFO [train.py:968] (1/2) Epoch 30, batch 22050, libri_loss[loss=0.2864, simple_loss=0.3535, pruned_loss=0.1097, over 29742.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3389, pruned_loss=0.09343, over 5714054.90 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3429, pruned_loss=0.09289, over 5765639.40 frames. ], giga_tot_loss[loss=0.2606, simple_loss=0.3371, pruned_loss=0.09204, over 5703185.60 frames. ], batch size: 87, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:58:59,204 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:59:13,740 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1342544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 11:59:34,957 INFO [train.py:968] (1/2) Epoch 30, batch 22100, giga_loss[loss=0.2882, simple_loss=0.3454, pruned_loss=0.1155, over 24301.00 frames. ], tot_loss[loss=0.2624, simple_loss=0.339, pruned_loss=0.09289, over 5701875.27 frames. ], libri_tot_loss[loss=0.2643, simple_loss=0.3427, pruned_loss=0.09302, over 5766914.49 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3377, pruned_loss=0.09168, over 5691092.99 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 11:59:56,027 INFO [optim.py:369] (1/2) 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 12:00:19,911 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4845, 1.7630, 1.4600, 1.4285], device='cuda:1'), covar=tensor([0.2802, 0.2927, 0.3391, 0.2562], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1166, 0.1432, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 12:00:20,227 INFO [train.py:968] (1/2) Epoch 30, batch 22150, giga_loss[loss=0.2536, simple_loss=0.3266, pruned_loss=0.0903, over 28921.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3394, pruned_loss=0.09331, over 5702809.17 frames. ], libri_tot_loss[loss=0.2657, simple_loss=0.3436, pruned_loss=0.09393, over 5766249.34 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3375, pruned_loss=0.09152, over 5693376.28 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:01:03,172 INFO [train.py:968] (1/2) Epoch 30, batch 22200, giga_loss[loss=0.2604, simple_loss=0.344, pruned_loss=0.08837, over 28760.00 frames. ], tot_loss[loss=0.2628, simple_loss=0.3393, pruned_loss=0.09312, over 5708185.52 frames. ], libri_tot_loss[loss=0.2664, simple_loss=0.3441, pruned_loss=0.09432, over 5768183.84 frames. ], giga_tot_loss[loss=0.2599, simple_loss=0.3372, pruned_loss=0.09129, over 5697517.27 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:01:08,967 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-15 12:01:20,112 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1342687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:01:20,824 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:01:22,499 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:01:43,684 INFO [train.py:968] (1/2) Epoch 30, batch 22250, giga_loss[loss=0.2769, simple_loss=0.3484, pruned_loss=0.1027, over 28735.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3396, pruned_loss=0.09368, over 5713209.84 frames. ], libri_tot_loss[loss=0.2669, simple_loss=0.3444, pruned_loss=0.09472, over 5769814.46 frames. ], giga_tot_loss[loss=0.2605, simple_loss=0.3374, pruned_loss=0.09178, over 5699970.92 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:01:44,487 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:02:28,632 INFO [train.py:968] (1/2) Epoch 30, batch 22300, giga_loss[loss=0.2659, simple_loss=0.3464, pruned_loss=0.09265, over 28607.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3416, pruned_loss=0.09457, over 5708737.99 frames. ], libri_tot_loss[loss=0.2671, simple_loss=0.3445, pruned_loss=0.09483, over 5770342.39 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3397, pruned_loss=0.09299, over 5697740.44 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:02:47,141 INFO [optim.py:369] (1/2) 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 12:03:12,302 INFO [train.py:968] (1/2) Epoch 30, batch 22350, giga_loss[loss=0.2859, simple_loss=0.3534, pruned_loss=0.1092, over 28920.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3452, pruned_loss=0.09627, over 5714067.56 frames. ], libri_tot_loss[loss=0.2678, simple_loss=0.3451, pruned_loss=0.09524, over 5772331.00 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3432, pruned_loss=0.09466, over 5702729.42 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:03:24,178 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1342831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:03:26,611 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1342834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:03:51,024 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1342863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:03:54,012 INFO [train.py:968] (1/2) Epoch 30, batch 22400, giga_loss[loss=0.3005, simple_loss=0.3799, pruned_loss=0.1106, over 28966.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3477, pruned_loss=0.09783, over 5705362.86 frames. ], libri_tot_loss[loss=0.2688, simple_loss=0.3458, pruned_loss=0.09587, over 5764902.10 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3455, pruned_loss=0.09603, over 5702042.46 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:04:13,154 INFO [optim.py:369] (1/2) 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 12:04:18,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4642, 1.3845, 4.3516, 3.3814], device='cuda:1'), covar=tensor([0.1642, 0.2952, 0.0437, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0680, 0.1015, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 12:04:35,967 INFO [train.py:968] (1/2) Epoch 30, batch 22450, giga_loss[loss=0.3122, simple_loss=0.384, pruned_loss=0.1202, over 28865.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3495, pruned_loss=0.09876, over 5713275.45 frames. ], libri_tot_loss[loss=0.2693, simple_loss=0.3462, pruned_loss=0.09622, over 5765694.22 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3474, pruned_loss=0.09705, over 5709136.72 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:04:46,580 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3585, 1.1949, 4.1494, 3.3811], device='cuda:1'), covar=tensor([0.1774, 0.2985, 0.0434, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0815, 0.0679, 0.1015, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 12:04:59,139 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8440, 1.1277, 2.8575, 2.6820], device='cuda:1'), covar=tensor([0.1690, 0.2568, 0.0587, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0680, 0.1016, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 12:05:07,079 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6862, 1.7535, 1.8229, 1.6275], device='cuda:1'), covar=tensor([0.3215, 0.2843, 0.2238, 0.2981], device='cuda:1'), in_proj_covar=tensor([0.2094, 0.2061, 0.1969, 0.2109], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 12:05:20,246 INFO [train.py:968] (1/2) Epoch 30, batch 22500, giga_loss[loss=0.2597, simple_loss=0.3345, pruned_loss=0.09243, over 28287.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3521, pruned_loss=0.1006, over 5710585.37 frames. ], libri_tot_loss[loss=0.2704, simple_loss=0.347, pruned_loss=0.09687, over 5765045.99 frames. ], giga_tot_loss[loss=0.2736, simple_loss=0.3498, pruned_loss=0.09876, over 5706547.73 frames. ], batch size: 77, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:05:39,239 INFO [optim.py:369] (1/2) 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 12:05:50,272 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.71 vs. limit=2.0
+2023-03-15 12:06:03,153 INFO [train.py:968] (1/2) Epoch 30, batch 22550, giga_loss[loss=0.2369, simple_loss=0.3114, pruned_loss=0.08115, over 28393.00 frames. ], tot_loss[loss=0.2763, simple_loss=0.3513, pruned_loss=0.1006, over 5708708.43 frames. ], libri_tot_loss[loss=0.2711, simple_loss=0.3476, pruned_loss=0.0973, over 5766551.14 frames. ], giga_tot_loss[loss=0.2734, simple_loss=0.3491, pruned_loss=0.09883, over 5703482.14 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:06:15,505 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9376, 1.2894, 1.0560, 0.2673], device='cuda:1'), covar=tensor([0.5291, 0.3681, 0.5668, 0.8357], device='cuda:1'), in_proj_covar=tensor([0.1852, 0.1728, 0.1663, 0.1508], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 12:06:38,679 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.0038, 3.8473, 3.6665, 1.8159], device='cuda:1'), covar=tensor([0.0675, 0.0790, 0.0750, 0.2433], device='cuda:1'), in_proj_covar=tensor([0.1298, 0.1199, 0.1006, 0.0750], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 12:06:49,473 INFO [train.py:968] (1/2) Epoch 30, batch 22600, giga_loss[loss=0.2651, simple_loss=0.3396, pruned_loss=0.09529, over 28911.00 frames. ], tot_loss[loss=0.2736, simple_loss=0.3488, pruned_loss=0.09921, over 5708684.15 frames. ], libri_tot_loss[loss=0.2718, simple_loss=0.3482, pruned_loss=0.09773, over 5762853.25 frames. ], giga_tot_loss[loss=0.2707, simple_loss=0.3465, pruned_loss=0.09743, over 5707265.53 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:07:10,000 INFO [optim.py:369] (1/2) 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 12:07:32,330 INFO [train.py:968] (1/2) Epoch 30, batch 22650, giga_loss[loss=0.2394, simple_loss=0.3198, pruned_loss=0.07949, over 28311.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3439, pruned_loss=0.09642, over 5712240.66 frames. ], libri_tot_loss[loss=0.2719, simple_loss=0.3483, pruned_loss=0.0978, over 5764776.93 frames. ], giga_tot_loss[loss=0.266, simple_loss=0.342, pruned_loss=0.09494, over 5708556.02 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:07:46,997 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 22700, giga_loss[loss=0.2097, simple_loss=0.2955, pruned_loss=0.06194, over 28412.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3431, pruned_loss=0.09588, over 5700036.60 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3487, pruned_loss=0.09811, over 5756607.02 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3412, pruned_loss=0.09443, over 5703620.20 frames. ], batch size: 71, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:08:34,168 INFO [optim.py:369] (1/2) 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 12:08:49,736 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5505, 1.5927, 1.7201, 1.3288], device='cuda:1'), covar=tensor([0.1996, 0.2611, 0.1674, 0.1944], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0718, 0.0993, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 12:09:01,280 INFO [train.py:968] (1/2) Epoch 30, batch 22750, giga_loss[loss=0.2872, simple_loss=0.3619, pruned_loss=0.1062, over 28956.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.344, pruned_loss=0.09469, over 5691145.72 frames. ], libri_tot_loss[loss=0.2725, simple_loss=0.3486, pruned_loss=0.09814, over 5747880.83 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3425, pruned_loss=0.0935, over 5700451.83 frames. ], batch size: 213, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:09:06,179 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5252, 1.7193, 1.1939, 1.3114], device='cuda:1'), covar=tensor([0.1023, 0.0583, 0.1100, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0450, 0.0525, 0.0463], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 12:09:20,263 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.96 vs. limit=2.0
+2023-03-15 12:09:43,168 INFO [train.py:968] (1/2) Epoch 30, batch 22800, libri_loss[loss=0.2816, simple_loss=0.3577, pruned_loss=0.1027, over 29769.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3442, pruned_loss=0.09443, over 5690698.83 frames. ], libri_tot_loss[loss=0.2722, simple_loss=0.3483, pruned_loss=0.098, over 5749018.45 frames. ], giga_tot_loss[loss=0.2652, simple_loss=0.3432, pruned_loss=0.09356, over 5696220.10 frames. ], batch size: 87, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:09:47,048 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6628, 4.5174, 4.2890, 2.3244], device='cuda:1'), covar=tensor([0.0587, 0.0753, 0.0739, 0.1782], device='cuda:1'), in_proj_covar=tensor([0.1300, 0.1201, 0.1008, 0.0751], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 12:09:59,375 INFO [optim.py:369] (1/2) 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 12:10:02,851 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343296.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:10:14,920 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-15 12:10:22,206 INFO [train.py:968] (1/2) Epoch 30, batch 22850, libri_loss[loss=0.2489, simple_loss=0.3141, pruned_loss=0.09187, over 29377.00 frames. ], tot_loss[loss=0.2664, simple_loss=0.3428, pruned_loss=0.09503, over 5686399.54 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3488, pruned_loss=0.09857, over 5742269.19 frames. ], giga_tot_loss[loss=0.2644, simple_loss=0.3414, pruned_loss=0.09368, over 5694223.04 frames. ], batch size: 67, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:10:51,412 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:11:05,537 INFO [train.py:968] (1/2) Epoch 30, batch 22900, giga_loss[loss=0.2414, simple_loss=0.3097, pruned_loss=0.08659, over 28703.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3413, pruned_loss=0.09542, over 5693732.82 frames. ], libri_tot_loss[loss=0.273, simple_loss=0.3487, pruned_loss=0.09862, over 5742811.38 frames. ], giga_tot_loss[loss=0.2643, simple_loss=0.3401, pruned_loss=0.09423, over 5698090.70 frames. ], batch size: 99, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:11:25,624 INFO [optim.py:369] (1/2) 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 12:11:47,731 INFO [train.py:968] (1/2) Epoch 30, batch 22950, giga_loss[loss=0.2552, simple_loss=0.3346, pruned_loss=0.08784, over 29034.00 frames. ], tot_loss[loss=0.2654, simple_loss=0.3398, pruned_loss=0.09546, over 5700529.93 frames. ], libri_tot_loss[loss=0.2739, simple_loss=0.3493, pruned_loss=0.09918, over 5737720.91 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.3381, pruned_loss=0.09395, over 5706687.55 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:11:47,993 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.6342, 4.4910, 4.2450, 2.1377], device='cuda:1'), covar=tensor([0.0605, 0.0773, 0.0726, 0.1955], device='cuda:1'), in_proj_covar=tensor([0.1305, 0.1205, 0.1011, 0.0754], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 12:12:06,298 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5269, 2.1556, 1.7045, 0.8210], device='cuda:1'), covar=tensor([0.8782, 0.3869, 0.4608, 0.9088], device='cuda:1'), in_proj_covar=tensor([0.1853, 0.1728, 0.1667, 0.1513], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 12:12:30,775 INFO [train.py:968] (1/2) Epoch 30, batch 23000, giga_loss[loss=0.2536, simple_loss=0.332, pruned_loss=0.08767, over 28911.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3392, pruned_loss=0.09636, over 5699212.09 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3493, pruned_loss=0.09918, over 5738529.98 frames. ], giga_tot_loss[loss=0.264, simple_loss=0.3378, pruned_loss=0.09512, over 5702926.29 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:12:50,508 INFO [optim.py:369] (1/2) 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 12:12:59,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5536, 1.6892, 1.4490, 1.2100], device='cuda:1'), covar=tensor([0.3201, 0.2988, 0.3704, 0.2710], device='cuda:1'), in_proj_covar=tensor([0.1617, 0.1165, 0.1429, 0.1013], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 12:13:05,308 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343509.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:13:13,538 INFO [train.py:968] (1/2) Epoch 30, batch 23050, giga_loss[loss=0.2628, simple_loss=0.347, pruned_loss=0.0893, over 28538.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3381, pruned_loss=0.09552, over 5708587.28 frames. ], libri_tot_loss[loss=0.2738, simple_loss=0.3493, pruned_loss=0.09918, over 5738529.98 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.337, pruned_loss=0.09456, over 5711478.09 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:13:24,875 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3210, 1.2392, 1.1901, 1.4905], device='cuda:1'), covar=tensor([0.0771, 0.0372, 0.0374, 0.0917], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 12:13:27,335 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4604, 1.7210, 1.6681, 1.5506], device='cuda:1'), covar=tensor([0.2371, 0.2444, 0.2664, 0.2490], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0764, 0.0736, 0.0703], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 12:13:53,335 INFO [train.py:968] (1/2) Epoch 30, batch 23100, libri_loss[loss=0.3329, simple_loss=0.4102, pruned_loss=0.1278, over 29758.00 frames. ], tot_loss[loss=0.2617, simple_loss=0.335, pruned_loss=0.09419, over 5706647.46 frames. ], libri_tot_loss[loss=0.2747, simple_loss=0.3501, pruned_loss=0.09966, over 5738199.09 frames. ], giga_tot_loss[loss=0.2595, simple_loss=0.3332, pruned_loss=0.09291, over 5708417.73 frames. ], batch size: 87, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:14:00,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2245, 1.9641, 1.5559, 0.6046], device='cuda:1'), covar=tensor([0.6194, 0.2984, 0.4532, 0.6286], device='cuda:1'), in_proj_covar=tensor([0.1852, 0.1728, 0.1666, 0.1512], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 12:14:13,452 INFO [optim.py:369] (1/2) 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 12:14:18,192 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:14:21,874 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0320, 1.9697, 2.1756, 1.7170], device='cuda:1'), covar=tensor([0.1661, 0.2650, 0.1457, 0.1734], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0717, 0.0992, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 12:14:33,736 INFO [train.py:968] (1/2) Epoch 30, batch 23150, libri_loss[loss=0.2862, simple_loss=0.3657, pruned_loss=0.1033, over 29680.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.3328, pruned_loss=0.09354, over 5709928.97 frames. ], libri_tot_loss[loss=0.2759, simple_loss=0.351, pruned_loss=0.1004, over 5741809.97 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3298, pruned_loss=0.09162, over 5706758.54 frames. ], batch size: 91, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:14:59,061 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1343652.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:15:00,964 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1343655.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:15:10,826 INFO [train.py:968] (1/2) Epoch 30, batch 23200, giga_loss[loss=0.2383, simple_loss=0.3145, pruned_loss=0.08106, over 28888.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3306, pruned_loss=0.09152, over 5712900.95 frames. ], libri_tot_loss[loss=0.2761, simple_loss=0.3511, pruned_loss=0.1005, over 5744540.99 frames. ], giga_tot_loss[loss=0.2533, simple_loss=0.3274, pruned_loss=0.08953, over 5706927.89 frames. ], batch size: 145, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:15:13,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:15:23,798 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1343684.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:15:27,318 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3432, 1.5725, 1.3478, 1.5665], device='cuda:1'), covar=tensor([0.0734, 0.0331, 0.0338, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 12:15:30,435 INFO [optim.py:369] (1/2) 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 12:15:53,198 INFO [train.py:968] (1/2) Epoch 30, batch 23250, giga_loss[loss=0.2598, simple_loss=0.3394, pruned_loss=0.09012, over 28811.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3335, pruned_loss=0.09304, over 5718419.90 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3517, pruned_loss=0.1011, over 5750416.94 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3296, pruned_loss=0.09054, over 5707065.17 frames. ], batch size: 243, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:15:59,812 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:16:36,474 INFO [train.py:968] (1/2) Epoch 30, batch 23300, giga_loss[loss=0.2728, simple_loss=0.3608, pruned_loss=0.09238, over 28712.00 frames. ], tot_loss[loss=0.2622, simple_loss=0.3364, pruned_loss=0.09403, over 5714197.77 frames. ], libri_tot_loss[loss=0.2769, simple_loss=0.3516, pruned_loss=0.1011, over 5746011.18 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3332, pruned_loss=0.09199, over 5709046.80 frames. ], batch size: 262, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:16:56,245 INFO [optim.py:369] (1/2) 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 12:17:13,986 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1343814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:17:16,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1343817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:17:16,596 INFO [train.py:968] (1/2) Epoch 30, batch 23350, giga_loss[loss=0.2605, simple_loss=0.3459, pruned_loss=0.08755, over 28567.00 frames. ], tot_loss[loss=0.2655, simple_loss=0.3399, pruned_loss=0.09558, over 5715491.41 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.3516, pruned_loss=0.1012, over 5750269.44 frames. ], giga_tot_loss[loss=0.2621, simple_loss=0.337, pruned_loss=0.09366, over 5706561.82 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:17:18,494 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4962, 2.0197, 1.6277, 1.7355], device='cuda:1'), covar=tensor([0.0756, 0.0267, 0.0331, 0.0882], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 12:17:43,422 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1343846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:02,871 INFO [train.py:968] (1/2) Epoch 30, batch 23400, giga_loss[loss=0.2922, simple_loss=0.3525, pruned_loss=0.1159, over 23749.00 frames. ], tot_loss[loss=0.2684, simple_loss=0.3431, pruned_loss=0.09689, over 5704071.47 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3517, pruned_loss=0.1015, over 5748227.71 frames. ], giga_tot_loss[loss=0.2654, simple_loss=0.3406, pruned_loss=0.0951, over 5698595.06 frames. ], batch size: 705, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:18:04,631 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1343870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:07,686 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1343873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:10,312 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3012, 2.6890, 1.4353, 1.4414], device='cuda:1'), covar=tensor([0.0950, 0.0391, 0.0944, 0.1350], device='cuda:1'), in_proj_covar=tensor([0.0428, 0.0573, 0.0415, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 12:18:26,336 INFO [optim.py:369] (1/2) 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 12:18:28,814 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8501, 1.6201, 1.8800, 1.4688], device='cuda:1'), covar=tensor([0.2329, 0.3302, 0.1767, 0.1895], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0715, 0.0989, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 12:18:35,865 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1343902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:18:38,929 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3332, 1.6987, 1.4690, 1.6204], device='cuda:1'), covar=tensor([0.0789, 0.0311, 0.0336, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0122, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0116], device='cuda:1')
+2023-03-15 12:18:48,798 INFO [train.py:968] (1/2) Epoch 30, batch 23450, giga_loss[loss=0.2523, simple_loss=0.3377, pruned_loss=0.0834, over 28810.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3442, pruned_loss=0.0969, over 5700894.82 frames. ], libri_tot_loss[loss=0.2774, simple_loss=0.3518, pruned_loss=0.1015, over 5749042.57 frames. ], giga_tot_loss[loss=0.2665, simple_loss=0.3421, pruned_loss=0.09542, over 5695584.67 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:18:49,038 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1343918.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:19:08,442 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.25 vs. limit=5.0
+2023-03-15 12:19:38,921 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-15 12:19:39,876 INFO [train.py:968] (1/2) Epoch 30, batch 23500, giga_loss[loss=0.4566, simple_loss=0.468, pruned_loss=0.2226, over 26773.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.351, pruned_loss=0.1031, over 5695114.31 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3518, pruned_loss=0.1016, over 5750309.20 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3493, pruned_loss=0.1018, over 5689011.14 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:19:44,624 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1343973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:20:03,657 INFO [optim.py:369] (1/2) 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 12:20:10,120 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 12:20:33,013 INFO [train.py:968] (1/2) Epoch 30, batch 23550, libri_loss[loss=0.3572, simple_loss=0.4029, pruned_loss=0.1558, over 25488.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3555, pruned_loss=0.1067, over 5691552.92 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3522, pruned_loss=0.102, over 5749823.63 frames. ], giga_tot_loss[loss=0.2823, simple_loss=0.3539, pruned_loss=0.1054, over 5686201.77 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:20:52,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5118, 1.6084, 1.5655, 1.4532], device='cuda:1'), covar=tensor([0.2353, 0.2353, 0.1996, 0.2353], device='cuda:1'), in_proj_covar=tensor([0.2093, 0.2057, 0.1967, 0.2103], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 12:21:23,659 INFO [train.py:968] (1/2) Epoch 30, batch 23600, giga_loss[loss=0.3126, simple_loss=0.3857, pruned_loss=0.1198, over 28922.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3631, pruned_loss=0.1121, over 5684332.17 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3522, pruned_loss=0.1022, over 5748666.73 frames. ], giga_tot_loss[loss=0.292, simple_loss=0.3619, pruned_loss=0.1111, over 5679281.56 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:21:39,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0750, 2.2227, 1.6901, 1.5491], device='cuda:1'), covar=tensor([0.0905, 0.0251, 0.0302, 0.1124], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0123, 0.0121, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0106, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 12:21:49,856 INFO [optim.py:369] (1/2) 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 12:22:13,297 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:22:14,323 INFO [train.py:968] (1/2) Epoch 30, batch 23650, giga_loss[loss=0.4791, simple_loss=0.4809, pruned_loss=0.2387, over 26529.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3689, pruned_loss=0.1171, over 5673388.61 frames. ], libri_tot_loss[loss=0.2784, simple_loss=0.3523, pruned_loss=0.1022, over 5740977.33 frames. ], giga_tot_loss[loss=0.3005, simple_loss=0.3681, pruned_loss=0.1164, over 5674458.19 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:22:15,882 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1344119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:22:45,268 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1344148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:23:05,622 INFO [train.py:968] (1/2) Epoch 30, batch 23700, giga_loss[loss=0.4075, simple_loss=0.4391, pruned_loss=0.188, over 27543.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3754, pruned_loss=0.1228, over 5663532.32 frames. ], libri_tot_loss[loss=0.2787, simple_loss=0.3525, pruned_loss=0.1024, over 5745106.66 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3749, pruned_loss=0.1224, over 5659132.07 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:23:32,525 INFO [optim.py:369] (1/2) 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 12:23:56,350 INFO [train.py:968] (1/2) Epoch 30, batch 23750, giga_loss[loss=0.2917, simple_loss=0.3629, pruned_loss=0.1102, over 28616.00 frames. ], tot_loss[loss=0.314, simple_loss=0.3783, pruned_loss=0.1249, over 5669109.71 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3527, pruned_loss=0.1026, over 5745711.28 frames. ], giga_tot_loss[loss=0.3138, simple_loss=0.3781, pruned_loss=0.1248, over 5664087.68 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:24:48,394 INFO [train.py:968] (1/2) Epoch 30, batch 23800, giga_loss[loss=0.3321, simple_loss=0.3903, pruned_loss=0.1369, over 28589.00 frames. ], tot_loss[loss=0.3186, simple_loss=0.381, pruned_loss=0.1281, over 5656373.16 frames. ], libri_tot_loss[loss=0.279, simple_loss=0.3528, pruned_loss=0.1026, over 5737315.62 frames. ], giga_tot_loss[loss=0.3188, simple_loss=0.3811, pruned_loss=0.1282, over 5658442.76 frames. ], batch size: 336, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:25:15,693 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:25:16,899 INFO [optim.py:369] (1/2) 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 12:25:21,452 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 12:25:41,127 INFO [train.py:968] (1/2) Epoch 30, batch 23850, giga_loss[loss=0.2665, simple_loss=0.3395, pruned_loss=0.09678, over 28796.00 frames. ], tot_loss[loss=0.3221, simple_loss=0.3828, pruned_loss=0.1307, over 5639622.52 frames. ], libri_tot_loss[loss=0.2794, simple_loss=0.3531, pruned_loss=0.1029, over 5737801.63 frames. ], giga_tot_loss[loss=0.3228, simple_loss=0.3832, pruned_loss=0.1312, over 5638809.01 frames. ], batch size: 119, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:26:31,452 INFO [train.py:968] (1/2) Epoch 30, batch 23900, giga_loss[loss=0.3444, simple_loss=0.4089, pruned_loss=0.14, over 28695.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3842, pruned_loss=0.1323, over 5639874.34 frames. ], libri_tot_loss[loss=0.2795, simple_loss=0.353, pruned_loss=0.103, over 5729706.72 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3859, pruned_loss=0.1338, over 5642816.66 frames. ], batch size: 242, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:27:04,030 INFO [optim.py:369] (1/2) 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 12:27:14,183 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:27:29,986 INFO [train.py:968] (1/2) Epoch 30, batch 23950, giga_loss[loss=0.2922, simple_loss=0.3598, pruned_loss=0.1124, over 28894.00 frames. ], tot_loss[loss=0.3307, simple_loss=0.3882, pruned_loss=0.1367, over 5608885.43 frames. ], libri_tot_loss[loss=0.2802, simple_loss=0.3534, pruned_loss=0.1035, over 5723129.81 frames. ], giga_tot_loss[loss=0.3332, simple_loss=0.3901, pruned_loss=0.1382, over 5615436.87 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:27:50,290 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:27:54,209 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 24000, giga_loss[loss=0.2728, simple_loss=0.3337, pruned_loss=0.106, over 28650.00 frames. ], tot_loss[loss=0.3293, simple_loss=0.3861, pruned_loss=0.1363, over 5598245.00 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3536, pruned_loss=0.1038, over 5710499.72 frames. ], giga_tot_loss[loss=0.332, simple_loss=0.3882, pruned_loss=0.1379, over 5611007.18 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:28:24,535 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 12:28:33,773 INFO [train.py:1012] (1/2) Epoch 30, validation: loss=0.2015, simple_loss=0.31, pruned_loss=0.04652, over 944034.00 frames. 
+2023-03-15 12:28:33,774 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 12:28:34,113 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:28:34,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1344468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:28:55,112 INFO [optim.py:369] (1/2) 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 12:29:17,903 INFO [train.py:968] (1/2) Epoch 30, batch 24050, giga_loss[loss=0.2727, simple_loss=0.3365, pruned_loss=0.1045, over 28654.00 frames. ], tot_loss[loss=0.3262, simple_loss=0.3832, pruned_loss=0.1346, over 5621649.20 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3534, pruned_loss=0.1039, over 5716369.32 frames. ], giga_tot_loss[loss=0.3297, simple_loss=0.3859, pruned_loss=0.1367, over 5623758.79 frames. ], batch size: 85, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:29:35,577 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-15 12:29:37,357 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5342, 2.2428, 1.6356, 0.7327], device='cuda:1'), covar=tensor([0.6521, 0.3462, 0.4478, 0.7307], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1752, 0.1682, 0.1522], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 12:30:04,606 INFO [train.py:968] (1/2) Epoch 30, batch 24100, giga_loss[loss=0.2939, simple_loss=0.3683, pruned_loss=0.1098, over 28558.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3823, pruned_loss=0.1332, over 5621718.15 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3532, pruned_loss=0.104, over 5718149.19 frames. ], giga_tot_loss[loss=0.3287, simple_loss=0.3857, pruned_loss=0.1359, over 5618738.37 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:30:17,201 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:30:44,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 24150, giga_loss[loss=0.2812, simple_loss=0.3516, pruned_loss=0.1054, over 28805.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3823, pruned_loss=0.1325, over 5616611.92 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3531, pruned_loss=0.104, over 5723797.60 frames. ], giga_tot_loss[loss=0.3283, simple_loss=0.3859, pruned_loss=0.1354, over 5606441.08 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:31:26,561 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9812, 1.3226, 1.1205, 0.2500], device='cuda:1'), covar=tensor([0.4921, 0.3617, 0.4995, 0.7120], device='cuda:1'), in_proj_covar=tensor([0.1871, 0.1751, 0.1683, 0.1523], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 12:31:47,750 INFO [train.py:968] (1/2) Epoch 30, batch 24200, giga_loss[loss=0.3254, simple_loss=0.3886, pruned_loss=0.131, over 28616.00 frames. ], tot_loss[loss=0.3247, simple_loss=0.3838, pruned_loss=0.1328, over 5624848.06 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3532, pruned_loss=0.104, over 5726361.45 frames. ], giga_tot_loss[loss=0.329, simple_loss=0.387, pruned_loss=0.1355, over 5613136.66 frames. ], batch size: 307, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:32:18,128 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 24250, giga_loss[loss=0.3062, simple_loss=0.3891, pruned_loss=0.1117, over 28827.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3827, pruned_loss=0.1313, over 5618320.73 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3534, pruned_loss=0.1042, over 5716805.20 frames. ], giga_tot_loss[loss=0.3268, simple_loss=0.3857, pruned_loss=0.1339, over 5615227.28 frames. ], batch size: 199, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:33:22,652 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-15 12:33:32,093 INFO [train.py:968] (1/2) Epoch 30, batch 24300, giga_loss[loss=0.2637, simple_loss=0.351, pruned_loss=0.08823, over 28940.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.38, pruned_loss=0.1283, over 5628678.90 frames. ], libri_tot_loss[loss=0.2813, simple_loss=0.3536, pruned_loss=0.1045, over 5722253.26 frames. ], giga_tot_loss[loss=0.3222, simple_loss=0.3829, pruned_loss=0.1307, over 5618811.22 frames. ], batch size: 227, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:33:42,385 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:33:50,575 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 12:33:57,703 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([4.5812, 4.4359, 4.2254, 1.9518], device='cuda:1'), covar=tensor([0.0529, 0.0632, 0.0720, 0.2135], device='cuda:1'), in_proj_covar=tensor([0.1329, 0.1229, 0.1031, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 12:34:08,640 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5567, 1.8214, 1.5094, 1.7953], device='cuda:1'), covar=tensor([0.0776, 0.0307, 0.0338, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 12:34:19,681 INFO [train.py:968] (1/2) Epoch 30, batch 24350, giga_loss[loss=0.2501, simple_loss=0.3357, pruned_loss=0.08229, over 28954.00 frames. ], tot_loss[loss=0.3141, simple_loss=0.377, pruned_loss=0.1256, over 5631987.52 frames. ], libri_tot_loss[loss=0.2821, simple_loss=0.3539, pruned_loss=0.1051, over 5723245.97 frames. ], giga_tot_loss[loss=0.3179, simple_loss=0.38, pruned_loss=0.1279, over 5619886.51 frames. ], batch size: 164, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:34:41,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:35:06,074 INFO [train.py:968] (1/2) Epoch 30, batch 24400, giga_loss[loss=0.2808, simple_loss=0.3562, pruned_loss=0.1027, over 29048.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3741, pruned_loss=0.1235, over 5640887.36 frames. ], libri_tot_loss[loss=0.2818, simple_loss=0.3535, pruned_loss=0.1051, over 5728466.63 frames. ], giga_tot_loss[loss=0.3149, simple_loss=0.3777, pruned_loss=0.126, over 5623643.49 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 8.0
+2023-03-15 12:35:07,898 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5527, 1.7959, 1.5001, 1.4991], device='cuda:1'), covar=tensor([0.2569, 0.2602, 0.2787, 0.2632], device='cuda:1'), in_proj_covar=tensor([0.1620, 0.1166, 0.1431, 0.1015], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 12:35:22,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1344884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:35:35,139 INFO [optim.py:369] (1/2) 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] (1/2) attn_weights_entropy = tensor([1.4916, 1.6404, 1.6363, 1.4669], device='cuda:1'), covar=tensor([0.1861, 0.1935, 0.2111, 0.1942], device='cuda:1'), in_proj_covar=tensor([0.0514, 0.0762, 0.0731, 0.0700], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 12:35:54,915 INFO [train.py:968] (1/2) Epoch 30, batch 24450, giga_loss[loss=0.275, simple_loss=0.3498, pruned_loss=0.1001, over 28969.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.3718, pruned_loss=0.1219, over 5645141.01 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.3537, pruned_loss=0.1055, over 5729663.05 frames. ], giga_tot_loss[loss=0.3111, simple_loss=0.3747, pruned_loss=0.1238, over 5628688.10 frames. ], batch size: 136, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:35:59,035 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:02,318 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1344925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:28,602 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1344954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:30,589 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:36:32,062 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3448, 1.4455, 1.4478, 1.3337], device='cuda:1'), covar=tensor([0.2799, 0.2582, 0.2276, 0.2592], device='cuda:1'), in_proj_covar=tensor([0.2102, 0.2062, 0.1974, 0.2117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 12:36:42,514 INFO [train.py:968] (1/2) Epoch 30, batch 24500, giga_loss[loss=0.2741, simple_loss=0.3484, pruned_loss=0.09993, over 29018.00 frames. ], tot_loss[loss=0.3093, simple_loss=0.373, pruned_loss=0.1229, over 5630450.95 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3543, pruned_loss=0.1059, over 5720776.38 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3753, pruned_loss=0.1245, over 5623023.72 frames. ], batch size: 128, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:36:45,928 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1208, 1.3115, 1.0943, 0.8901], device='cuda:1'), covar=tensor([0.1227, 0.0585, 0.1193, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0526, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 12:36:58,134 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1344981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:37:05,991 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1344986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:37:08,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1344989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:37:17,279 INFO [optim.py:369] (1/2) 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 12:37:21,754 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 24550, giga_loss[loss=0.3311, simple_loss=0.3874, pruned_loss=0.1374, over 29031.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3728, pruned_loss=0.1223, over 5628847.92 frames. ], libri_tot_loss[loss=0.2834, simple_loss=0.3546, pruned_loss=0.1061, over 5712239.98 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3745, pruned_loss=0.1235, over 5630776.21 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:37:41,583 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:38:21,755 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.66 vs. limit=5.0
+2023-03-15 12:38:31,276 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-15 12:38:34,675 INFO [train.py:968] (1/2) Epoch 30, batch 24600, libri_loss[loss=0.333, simple_loss=0.3956, pruned_loss=0.1352, over 29077.00 frames. ], tot_loss[loss=0.3049, simple_loss=0.3705, pruned_loss=0.1197, over 5645233.01 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3545, pruned_loss=0.1062, over 5714454.41 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3722, pruned_loss=0.1208, over 5643435.98 frames. ], batch size: 101, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:39:05,980 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1345100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:11,567 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:26,295 INFO [train.py:968] (1/2) Epoch 30, batch 24650, libri_loss[loss=0.3756, simple_loss=0.4239, pruned_loss=0.1637, over 29653.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3727, pruned_loss=0.1186, over 5659456.21 frames. ], libri_tot_loss[loss=0.2844, simple_loss=0.3553, pruned_loss=0.1068, over 5718341.79 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3737, pruned_loss=0.1192, over 5653295.55 frames. ], batch size: 91, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:39:34,014 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1345124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:36,400 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:39:43,383 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:40:10,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:40:26,342 INFO [train.py:968] (1/2) Epoch 30, batch 24700, giga_loss[loss=0.3446, simple_loss=0.3944, pruned_loss=0.1474, over 27584.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3747, pruned_loss=0.1199, over 5646556.71 frames. ], libri_tot_loss[loss=0.2846, simple_loss=0.3554, pruned_loss=0.1069, over 5716008.62 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3755, pruned_loss=0.1203, over 5643202.90 frames. ], batch size: 472, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:40:52,017 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3002, 1.8064, 1.2894, 0.6483], device='cuda:1'), covar=tensor([0.5590, 0.2984, 0.3643, 0.7030], device='cuda:1'), in_proj_covar=tensor([0.1867, 0.1744, 0.1674, 0.1517], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 12:40:54,138 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 24750, giga_loss[loss=0.3223, simple_loss=0.3904, pruned_loss=0.1271, over 28765.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3745, pruned_loss=0.1199, over 5658499.42 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3554, pruned_loss=0.107, over 5711519.40 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3756, pruned_loss=0.1205, over 5658530.77 frames. ], batch size: 284, lr: 1.06e-03, grad_scale: 2.0
+2023-03-15 12:41:57,450 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 24800, giga_loss[loss=0.3213, simple_loss=0.3821, pruned_loss=0.1303, over 29041.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3728, pruned_loss=0.1188, over 5674499.95 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3561, pruned_loss=0.1077, over 5712476.59 frames. ], giga_tot_loss[loss=0.3056, simple_loss=0.3734, pruned_loss=0.1189, over 5672410.53 frames. ], batch size: 155, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:42:35,715 INFO [optim.py:369] (1/2) 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 12:42:54,650 INFO [train.py:968] (1/2) Epoch 30, batch 24850, giga_loss[loss=0.2519, simple_loss=0.3236, pruned_loss=0.09008, over 28976.00 frames. ], tot_loss[loss=0.3057, simple_loss=0.3718, pruned_loss=0.1198, over 5674720.76 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.356, pruned_loss=0.1078, over 5715468.02 frames. ], giga_tot_loss[loss=0.3064, simple_loss=0.3728, pruned_loss=0.12, over 5669701.80 frames. ], batch size: 106, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:43:40,095 INFO [train.py:968] (1/2) Epoch 30, batch 24900, giga_loss[loss=0.2795, simple_loss=0.3655, pruned_loss=0.09669, over 28883.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3711, pruned_loss=0.1203, over 5657310.25 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3564, pruned_loss=0.1084, over 5700794.01 frames. ], giga_tot_loss[loss=0.3061, simple_loss=0.3718, pruned_loss=0.1202, over 5665541.02 frames. ], batch size: 174, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:43:43,483 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1345373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:44:03,740 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6129, 1.6365, 1.7832, 1.3909], device='cuda:1'), covar=tensor([0.1914, 0.2691, 0.1573, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0720, 0.0989, 0.0889], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 12:44:04,776 INFO [optim.py:369] (1/2) 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 12:44:09,485 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1345402.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:44:12,203 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:44:22,765 INFO [train.py:968] (1/2) Epoch 30, batch 24950, giga_loss[loss=0.3744, simple_loss=0.4083, pruned_loss=0.1703, over 26634.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3702, pruned_loss=0.1185, over 5668355.62 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3568, pruned_loss=0.1087, over 5703344.96 frames. ], giga_tot_loss[loss=0.3038, simple_loss=0.3708, pruned_loss=0.1184, over 5671916.75 frames. ], batch size: 555, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:44:41,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:45:16,335 INFO [train.py:968] (1/2) Epoch 30, batch 25000, giga_loss[loss=0.3213, simple_loss=0.3839, pruned_loss=0.1294, over 28680.00 frames. ], tot_loss[loss=0.3014, simple_loss=0.3694, pruned_loss=0.1167, over 5678438.04 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3564, pruned_loss=0.1085, over 5706258.65 frames. ], giga_tot_loss[loss=0.3022, simple_loss=0.3704, pruned_loss=0.117, over 5678130.86 frames. ], batch size: 92, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:45:41,922 INFO [optim.py:369] (1/2) 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 12:46:00,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 25050, giga_loss[loss=0.3077, simple_loss=0.3697, pruned_loss=0.1229, over 27963.00 frames. ], tot_loss[loss=0.3013, simple_loss=0.3689, pruned_loss=0.1168, over 5679462.04 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3561, pruned_loss=0.1084, over 5713110.67 frames. ], giga_tot_loss[loss=0.3026, simple_loss=0.3704, pruned_loss=0.1174, over 5672063.29 frames. ], batch size: 412, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:46:04,006 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1345519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:04,746 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:06,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:36,131 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1345548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:46:54,919 INFO [train.py:968] (1/2) Epoch 30, batch 25100, giga_loss[loss=0.3011, simple_loss=0.3665, pruned_loss=0.1178, over 28387.00 frames. ], tot_loss[loss=0.2993, simple_loss=0.3672, pruned_loss=0.1157, over 5685973.15 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3566, pruned_loss=0.1087, over 5713976.69 frames. ], giga_tot_loss[loss=0.3001, simple_loss=0.3681, pruned_loss=0.1161, over 5678986.38 frames. ], batch size: 65, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:46:57,395 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-15 12:47:07,970 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6023, 1.7555, 1.2401, 1.3247], device='cuda:1'), covar=tensor([0.1033, 0.0620, 0.1066, 0.1247], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0454, 0.0526, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 12:47:27,851 INFO [optim.py:369] (1/2) 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 12:47:46,942 INFO [train.py:968] (1/2) Epoch 30, batch 25150, giga_loss[loss=0.3241, simple_loss=0.3895, pruned_loss=0.1293, over 28264.00 frames. ], tot_loss[loss=0.2995, simple_loss=0.3664, pruned_loss=0.1164, over 5672201.22 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3567, pruned_loss=0.1089, over 5716902.92 frames. ], giga_tot_loss[loss=0.3002, simple_loss=0.3672, pruned_loss=0.1166, over 5663228.66 frames. ], batch size: 368, lr: 1.06e-03, grad_scale: 4.0
+2023-03-15 12:48:36,837 INFO [train.py:968] (1/2) Epoch 30, batch 25200, giga_loss[loss=0.3477, simple_loss=0.3991, pruned_loss=0.1481, over 28779.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3667, pruned_loss=0.1175, over 5669118.26 frames. ], libri_tot_loss[loss=0.2873, simple_loss=0.3567, pruned_loss=0.109, over 5717091.75 frames. ], giga_tot_loss[loss=0.3016, simple_loss=0.3677, pruned_loss=0.1177, over 5660890.68 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 12:49:08,099 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 25250, giga_loss[loss=0.2669, simple_loss=0.3421, pruned_loss=0.09585, over 29075.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3669, pruned_loss=0.1181, over 5674197.13 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3569, pruned_loss=0.109, over 5719289.45 frames. ], giga_tot_loss[loss=0.3021, simple_loss=0.3675, pruned_loss=0.1184, over 5665008.54 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:49:50,348 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345740.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:50:03,846 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3883, 1.7374, 1.5007, 1.6399], device='cuda:1'), covar=tensor([0.0788, 0.0319, 0.0321, 0.0852], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 12:50:11,297 INFO [train.py:968] (1/2) Epoch 30, batch 25300, giga_loss[loss=0.3168, simple_loss=0.3807, pruned_loss=0.1265, over 28528.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3651, pruned_loss=0.1171, over 5673919.26 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3569, pruned_loss=0.109, over 5715248.10 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3659, pruned_loss=0.1177, over 5667870.45 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:50:44,649 INFO [optim.py:369] (1/2) 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 12:51:03,915 INFO [train.py:968] (1/2) Epoch 30, batch 25350, giga_loss[loss=0.3368, simple_loss=0.3859, pruned_loss=0.1438, over 27642.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3648, pruned_loss=0.1176, over 5656876.05 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3573, pruned_loss=0.1094, over 5707545.46 frames. ], giga_tot_loss[loss=0.3004, simple_loss=0.3652, pruned_loss=0.1178, over 5658481.31 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:51:22,644 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:51:51,466 INFO [train.py:968] (1/2) Epoch 30, batch 25400, libri_loss[loss=0.3099, simple_loss=0.3802, pruned_loss=0.1198, over 28711.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.3665, pruned_loss=0.1183, over 5661022.02 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3578, pruned_loss=0.1097, over 5709887.64 frames. ], giga_tot_loss[loss=0.3018, simple_loss=0.3666, pruned_loss=0.1184, over 5659070.39 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:52:16,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1345895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:52:16,743 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1345896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:52:20,067 INFO [optim.py:369] (1/2) 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 12:52:34,974 INFO [train.py:968] (1/2) Epoch 30, batch 25450, giga_loss[loss=0.2719, simple_loss=0.3471, pruned_loss=0.09833, over 28853.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3665, pruned_loss=0.1171, over 5666152.76 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3582, pruned_loss=0.1099, over 5711278.50 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3663, pruned_loss=0.1171, over 5662706.05 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:52:51,664 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6570, 2.0246, 1.7960, 1.6564], device='cuda:1'), covar=tensor([0.2372, 0.2545, 0.2444, 0.2635], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0765, 0.0734, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 12:53:24,245 INFO [train.py:968] (1/2) Epoch 30, batch 25500, giga_loss[loss=0.2868, simple_loss=0.3535, pruned_loss=0.1101, over 28920.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.116, over 5663454.98 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3582, pruned_loss=0.1101, over 5714872.70 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.1159, over 5656870.08 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:53:35,808 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345978.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:53:44,764 INFO [scaling.py:679] (1/2) 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] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1345994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:53:53,844 INFO [optim.py:369] (1/2) 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 12:53:56,724 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:09,556 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.40 vs. limit=2.0
+2023-03-15 12:54:11,543 INFO [train.py:968] (1/2) Epoch 30, batch 25550, giga_loss[loss=0.3468, simple_loss=0.3969, pruned_loss=0.1484, over 28854.00 frames. ], tot_loss[loss=0.2987, simple_loss=0.3654, pruned_loss=0.116, over 5669140.81 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3581, pruned_loss=0.1099, over 5717141.95 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3656, pruned_loss=0.1162, over 5661290.07 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 12:54:29,282 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 12:54:32,753 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:33,911 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:34,031 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.85 vs. limit=5.0
+2023-03-15 12:54:35,511 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:36,231 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:54:59,335 INFO [train.py:968] (1/2) Epoch 30, batch 25600, giga_loss[loss=0.427, simple_loss=0.4519, pruned_loss=0.2011, over 26563.00 frames. ], tot_loss[loss=0.3032, simple_loss=0.3683, pruned_loss=0.1191, over 5657206.62 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.358, pruned_loss=0.1099, over 5719079.00 frames. ], giga_tot_loss[loss=0.3039, simple_loss=0.3688, pruned_loss=0.1195, over 5647450.12 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:55:02,553 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:55:02,740 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-15 12:55:03,239 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 12:55:47,664 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346115.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:55:49,768 INFO [train.py:968] (1/2) Epoch 30, batch 25650, libri_loss[loss=0.3135, simple_loss=0.3754, pruned_loss=0.1258, over 26107.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3696, pruned_loss=0.121, over 5655193.29 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3583, pruned_loss=0.1101, over 5715909.84 frames. ], giga_tot_loss[loss=0.3063, simple_loss=0.3699, pruned_loss=0.1213, over 5649502.46 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:56:39,628 INFO [train.py:968] (1/2) Epoch 30, batch 25700, giga_loss[loss=0.2652, simple_loss=0.3341, pruned_loss=0.09809, over 28439.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3698, pruned_loss=0.1219, over 5667204.53 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3583, pruned_loss=0.1102, over 5718138.74 frames. ], giga_tot_loss[loss=0.3076, simple_loss=0.3704, pruned_loss=0.1224, over 5659529.76 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:56:55,821 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-15 12:57:11,576 INFO [optim.py:369] (1/2) 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 12:57:23,599 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 25750, libri_loss[loss=0.3118, simple_loss=0.379, pruned_loss=0.1223, over 29488.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3708, pruned_loss=0.1231, over 5658283.01 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3579, pruned_loss=0.11, over 5724494.28 frames. ], giga_tot_loss[loss=0.3102, simple_loss=0.372, pruned_loss=0.1242, over 5644142.21 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:57:50,969 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346247.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:58:01,796 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346258.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:58:04,307 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346261.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 12:58:11,529 INFO [train.py:968] (1/2) Epoch 30, batch 25800, libri_loss[loss=0.2727, simple_loss=0.3404, pruned_loss=0.1025, over 29529.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.3688, pruned_loss=0.1214, over 5673952.16 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3579, pruned_loss=0.1102, over 5729974.99 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3702, pruned_loss=0.1225, over 5656029.65 frames. ], batch size: 79, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:58:38,217 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346290.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 12:58:45,529 INFO [optim.py:369] (1/2) 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 12:59:02,122 INFO [train.py:968] (1/2) Epoch 30, batch 25850, giga_loss[loss=0.3176, simple_loss=0.3846, pruned_loss=0.1253, over 28904.00 frames. ], tot_loss[loss=0.3053, simple_loss=0.3687, pruned_loss=0.1209, over 5670442.41 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3579, pruned_loss=0.1103, over 5731429.62 frames. ], giga_tot_loss[loss=0.3066, simple_loss=0.3698, pruned_loss=0.1217, over 5654520.55 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:59:29,373 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1630, 1.3106, 1.1870, 1.1478], device='cuda:1'), covar=tensor([0.2529, 0.2197, 0.1875, 0.2110], device='cuda:1'), in_proj_covar=tensor([0.2113, 0.2078, 0.1984, 0.2127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 12:59:34,428 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:36,843 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:39,796 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:47,679 INFO [train.py:968] (1/2) Epoch 30, batch 25900, giga_loss[loss=0.272, simple_loss=0.3462, pruned_loss=0.09888, over 28815.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3673, pruned_loss=0.1185, over 5661425.43 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3582, pruned_loss=0.1105, over 5722137.78 frames. ], giga_tot_loss[loss=0.3033, simple_loss=0.3682, pruned_loss=0.1192, over 5656398.34 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 12:59:48,656 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 12:59:56,831 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:00:07,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:00:14,821 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 25950, libri_loss[loss=0.2238, simple_loss=0.2936, pruned_loss=0.07702, over 29399.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3659, pruned_loss=0.1175, over 5662963.86 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.359, pruned_loss=0.1112, over 5726830.83 frames. ], giga_tot_loss[loss=0.3008, simple_loss=0.3662, pruned_loss=0.1177, over 5652682.97 frames. ], batch size: 67, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:00:42,485 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5693, 1.5498, 1.7411, 1.3728], device='cuda:1'), covar=tensor([0.1666, 0.2545, 0.1411, 0.1712], device='cuda:1'), in_proj_covar=tensor([0.0939, 0.0720, 0.0987, 0.0888], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 13:01:19,983 INFO [train.py:968] (1/2) Epoch 30, batch 26000, giga_loss[loss=0.2679, simple_loss=0.3343, pruned_loss=0.1008, over 28819.00 frames. ], tot_loss[loss=0.2983, simple_loss=0.3634, pruned_loss=0.1166, over 5667837.69 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3589, pruned_loss=0.1111, over 5725759.03 frames. ], giga_tot_loss[loss=0.2988, simple_loss=0.3639, pruned_loss=0.1169, over 5659761.54 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 13:01:53,078 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346496.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:01:56,835 INFO [optim.py:369] (1/2) 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 13:01:57,171 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:08,815 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:12,790 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:14,702 INFO [train.py:968] (1/2) Epoch 30, batch 26050, giga_loss[loss=0.2576, simple_loss=0.3336, pruned_loss=0.09078, over 28959.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3614, pruned_loss=0.1149, over 5678234.07 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3588, pruned_loss=0.1109, over 5727651.38 frames. ], giga_tot_loss[loss=0.2963, simple_loss=0.3619, pruned_loss=0.1153, over 5669819.07 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 13:02:19,059 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:22,757 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:26,438 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:47,028 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:02:53,120 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1154, 1.2875, 1.0893, 0.9268], device='cuda:1'), covar=tensor([0.1128, 0.0520, 0.1131, 0.1131], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0526, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 13:02:59,096 INFO [train.py:968] (1/2) Epoch 30, batch 26100, giga_loss[loss=0.3293, simple_loss=0.4023, pruned_loss=0.1281, over 28738.00 frames. ], tot_loss[loss=0.3, simple_loss=0.3654, pruned_loss=0.1173, over 5679379.16 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3598, pruned_loss=0.1117, over 5722704.18 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.365, pruned_loss=0.1171, over 5675054.37 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:03:03,037 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5246, 1.7612, 1.2432, 1.3222], device='cuda:1'), covar=tensor([0.1115, 0.0638, 0.1086, 0.1289], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0454, 0.0526, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 13:03:18,782 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.18 vs. limit=2.0
+2023-03-15 13:03:20,305 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4042, 4.2659, 4.0235, 2.0533], device='cuda:1'), covar=tensor([0.0672, 0.0836, 0.0901, 0.1820], device='cuda:1'), in_proj_covar=tensor([0.1336, 0.1234, 0.1035, 0.0766], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 13:03:29,124 INFO [optim.py:369] (1/2) 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 13:03:46,854 INFO [train.py:968] (1/2) Epoch 30, batch 26150, giga_loss[loss=0.2931, simple_loss=0.3735, pruned_loss=0.1064, over 28624.00 frames. ], tot_loss[loss=0.3005, simple_loss=0.3681, pruned_loss=0.1165, over 5679367.80 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3599, pruned_loss=0.1118, over 5723575.42 frames. ], giga_tot_loss[loss=0.3003, simple_loss=0.3678, pruned_loss=0.1164, over 5674248.13 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:03:52,286 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1346622.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 13:04:38,083 INFO [train.py:968] (1/2) Epoch 30, batch 26200, giga_loss[loss=0.2892, simple_loss=0.3608, pruned_loss=0.1088, over 28961.00 frames. ], tot_loss[loss=0.3011, simple_loss=0.3698, pruned_loss=0.1162, over 5681162.84 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3599, pruned_loss=0.1118, over 5723240.63 frames. ], giga_tot_loss[loss=0.301, simple_loss=0.3697, pruned_loss=0.1161, over 5677016.87 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:05:11,149 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 26250, giga_loss[loss=0.2928, simple_loss=0.3652, pruned_loss=0.1102, over 28991.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3721, pruned_loss=0.1181, over 5683717.91 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3602, pruned_loss=0.1118, over 5724164.38 frames. ], giga_tot_loss[loss=0.3042, simple_loss=0.372, pruned_loss=0.1182, over 5678485.64 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:06:01,129 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:06:11,103 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1346765.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 13:06:12,806 INFO [train.py:968] (1/2) Epoch 30, batch 26300, libri_loss[loss=0.2616, simple_loss=0.3428, pruned_loss=0.09015, over 29183.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3733, pruned_loss=0.1197, over 5688274.93 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3598, pruned_loss=0.1117, over 5727166.18 frames. ], giga_tot_loss[loss=0.3072, simple_loss=0.374, pruned_loss=0.1202, over 5680063.50 frames. ], batch size: 97, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:06:13,045 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1346768.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 13:06:21,979 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.42 vs. limit=5.0
+2023-03-15 13:06:45,175 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1346797.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 13:06:47,564 INFO [optim.py:369] (1/2) 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 13:07:00,303 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1346814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:07:03,237 INFO [train.py:968] (1/2) Epoch 30, batch 26350, libri_loss[loss=0.309, simple_loss=0.3838, pruned_loss=0.1171, over 29501.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3735, pruned_loss=0.121, over 5683001.81 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3599, pruned_loss=0.1116, over 5732356.28 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3745, pruned_loss=0.1217, over 5670306.45 frames. ], batch size: 81, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:07:51,374 INFO [train.py:968] (1/2) Epoch 30, batch 26400, giga_loss[loss=0.2943, simple_loss=0.3562, pruned_loss=0.1162, over 28891.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.372, pruned_loss=0.1204, over 5688772.87 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3594, pruned_loss=0.1113, over 5729437.84 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3736, pruned_loss=0.1216, over 5680159.49 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:08:21,033 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 26450, giga_loss[loss=0.2942, simple_loss=0.3583, pruned_loss=0.1151, over 28934.00 frames. ], tot_loss[loss=0.3051, simple_loss=0.3701, pruned_loss=0.12, over 5689765.76 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3599, pruned_loss=0.1117, over 5731875.69 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3711, pruned_loss=0.1207, over 5680263.81 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:08:39,455 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2862, 3.0351, 1.4952, 1.4091], device='cuda:1'), covar=tensor([0.0988, 0.0377, 0.0875, 0.1357], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0582, 0.0418, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 13:08:55,944 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2233, 0.8771, 0.9689, 1.3833], device='cuda:1'), covar=tensor([0.0709, 0.0485, 0.0363, 0.0767], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0124, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 13:09:02,095 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1616, 1.1979, 3.2779, 2.9898], device='cuda:1'), covar=tensor([0.1633, 0.2803, 0.0544, 0.1465], device='cuda:1'), in_proj_covar=tensor([0.0821, 0.0684, 0.1025, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 13:09:05,400 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.83 vs. limit=5.0
+2023-03-15 13:09:29,069 INFO [train.py:968] (1/2) Epoch 30, batch 26500, giga_loss[loss=0.3029, simple_loss=0.3722, pruned_loss=0.1168, over 29034.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3684, pruned_loss=0.1195, over 5687975.37 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.36, pruned_loss=0.1119, over 5725590.26 frames. ], giga_tot_loss[loss=0.3046, simple_loss=0.3693, pruned_loss=0.12, over 5684722.17 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:09:33,105 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5874, 1.7865, 1.2299, 1.3489], device='cuda:1'), covar=tensor([0.1089, 0.0618, 0.1120, 0.1228], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0456, 0.0528, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 13:09:44,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4510, 1.6911, 1.5809, 1.5215], device='cuda:1'), covar=tensor([0.1896, 0.2061, 0.2216, 0.2045], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0764, 0.0734, 0.0704], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 13:10:02,038 INFO [optim.py:369] (1/2) 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 13:10:17,816 INFO [train.py:968] (1/2) Epoch 30, batch 26550, libri_loss[loss=0.2437, simple_loss=0.3196, pruned_loss=0.08387, over 29554.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.3678, pruned_loss=0.1195, over 5673937.84 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3596, pruned_loss=0.1117, over 5720282.89 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3692, pruned_loss=0.1203, over 5674103.90 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:10:22,222 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1347024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:10:24,242 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6595, 2.4169, 1.8254, 0.7927], device='cuda:1'), covar=tensor([0.7779, 0.3794, 0.4573, 0.8121], device='cuda:1'), in_proj_covar=tensor([0.1874, 0.1756, 0.1676, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 13:10:57,415 INFO [train.py:968] (1/2) Epoch 30, batch 26600, giga_loss[loss=0.2552, simple_loss=0.3299, pruned_loss=0.0903, over 29029.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3666, pruned_loss=0.1189, over 5671779.51 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3592, pruned_loss=0.1117, over 5709437.24 frames. ], giga_tot_loss[loss=0.3043, simple_loss=0.3685, pruned_loss=0.12, over 5678040.57 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:11:27,836 INFO [optim.py:369] (1/2) 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 13:11:28,919 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1347103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:11:39,105 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 13:11:44,337 INFO [train.py:968] (1/2) Epoch 30, batch 26650, giga_loss[loss=0.2818, simple_loss=0.3452, pruned_loss=0.1092, over 28237.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3664, pruned_loss=0.1197, over 5663850.83 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3591, pruned_loss=0.1116, over 5710793.27 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3681, pruned_loss=0.1208, over 5666496.58 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:11:54,895 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:12:34,321 INFO [train.py:968] (1/2) Epoch 30, batch 26700, giga_loss[loss=0.3256, simple_loss=0.3762, pruned_loss=0.1375, over 28684.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3656, pruned_loss=0.1196, over 5654630.76 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3592, pruned_loss=0.1118, over 5713058.14 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.367, pruned_loss=0.1205, over 5653673.71 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:12:52,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:13:04,844 INFO [optim.py:369] (1/2) 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 13:13:19,848 INFO [train.py:968] (1/2) Epoch 30, batch 26750, giga_loss[loss=0.2315, simple_loss=0.3141, pruned_loss=0.07447, over 28510.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3657, pruned_loss=0.1186, over 5646875.81 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3586, pruned_loss=0.1116, over 5695525.69 frames. ], giga_tot_loss[loss=0.3035, simple_loss=0.3677, pruned_loss=0.1197, over 5659411.96 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:13:30,941 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3314, 1.6602, 1.2542, 0.8542], device='cuda:1'), covar=tensor([0.4036, 0.2416, 0.2613, 0.5384], device='cuda:1'), in_proj_covar=tensor([0.1881, 0.1763, 0.1681, 0.1522], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 13:14:10,836 INFO [train.py:968] (1/2) Epoch 30, batch 26800, giga_loss[loss=0.3074, simple_loss=0.3774, pruned_loss=0.1187, over 28768.00 frames. ], tot_loss[loss=0.3056, simple_loss=0.3694, pruned_loss=0.121, over 5635906.75 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1118, over 5686199.06 frames. ], giga_tot_loss[loss=0.307, simple_loss=0.3706, pruned_loss=0.1216, over 5654322.84 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:14:17,174 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:14:22,514 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:14:47,180 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:14:54,195 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.7195, 4.5631, 4.3038, 1.8859], device='cuda:1'), covar=tensor([0.0647, 0.0836, 0.0866, 0.2209], device='cuda:1'), in_proj_covar=tensor([0.1343, 0.1240, 0.1039, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 13:15:00,824 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4113, 2.3340, 2.2491, 2.1424], device='cuda:1'), covar=tensor([0.2231, 0.2805, 0.2497, 0.2576], device='cuda:1'), in_proj_covar=tensor([0.0516, 0.0764, 0.0736, 0.0706], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 13:15:02,557 INFO [train.py:968] (1/2) Epoch 30, batch 26850, giga_loss[loss=0.2726, simple_loss=0.3564, pruned_loss=0.09439, over 28914.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3686, pruned_loss=0.1206, over 5647170.51 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1118, over 5688517.92 frames. ], giga_tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5659139.94 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:15:17,107 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347332.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:15:19,726 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:15:49,189 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:15:54,141 INFO [train.py:968] (1/2) Epoch 30, batch 26900, giga_loss[loss=0.2975, simple_loss=0.3758, pruned_loss=0.1095, over 28926.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3703, pruned_loss=0.1189, over 5654456.50 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1118, over 5688517.92 frames. ], giga_tot_loss[loss=0.3049, simple_loss=0.3712, pruned_loss=0.1194, over 5663772.46 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:16:27,123 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:16:29,987 INFO [optim.py:369] (1/2) 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 13:16:35,997 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 13:16:42,511 INFO [train.py:968] (1/2) Epoch 30, batch 26950, giga_loss[loss=0.368, simple_loss=0.4286, pruned_loss=0.1537, over 28872.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3724, pruned_loss=0.1179, over 5671720.40 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5692352.66 frames. ], giga_tot_loss[loss=0.3047, simple_loss=0.373, pruned_loss=0.1182, over 5675117.87 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:16:58,362 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-15 13:17:01,799 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2587, 1.2236, 3.8116, 3.4283], device='cuda:1'), covar=tensor([0.2096, 0.3200, 0.0964, 0.1136], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0685, 0.1026, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 13:17:24,377 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.78 vs. limit=5.0
+2023-03-15 13:17:32,208 INFO [train.py:968] (1/2) Epoch 30, batch 27000, giga_loss[loss=0.3666, simple_loss=0.3966, pruned_loss=0.1683, over 23443.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3752, pruned_loss=0.1198, over 5677528.39 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3589, pruned_loss=0.1118, over 5694826.79 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3763, pruned_loss=0.1203, over 5677695.68 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:17:32,208 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 13:17:41,088 INFO [train.py:1012] (1/2) Epoch 30, validation: loss=0.202, simple_loss=0.3097, pruned_loss=0.04714, over 944034.00 frames. 
+2023-03-15 13:17:41,089 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 13:17:49,216 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1347478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:18:14,075 INFO [optim.py:369] (1/2) 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 13:18:29,890 INFO [train.py:968] (1/2) Epoch 30, batch 27050, giga_loss[loss=0.2841, simple_loss=0.3597, pruned_loss=0.1043, over 28997.00 frames. ], tot_loss[loss=0.3111, simple_loss=0.3775, pruned_loss=0.1224, over 5676975.69 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3589, pruned_loss=0.1118, over 5699916.24 frames. ], giga_tot_loss[loss=0.3124, simple_loss=0.3787, pruned_loss=0.123, over 5671828.03 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:18:54,391 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:18:56,903 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:19:22,726 INFO [train.py:968] (1/2) Epoch 30, batch 27100, giga_loss[loss=0.2844, simple_loss=0.3607, pruned_loss=0.1041, over 28851.00 frames. ], tot_loss[loss=0.315, simple_loss=0.3793, pruned_loss=0.1253, over 5645236.02 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.3591, pruned_loss=0.1119, over 5691359.24 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3805, pruned_loss=0.1259, over 5647616.91 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:19:28,127 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:19:56,504 INFO [optim.py:369] (1/2) 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 13:19:59,241 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2889, 1.4443, 1.4418, 1.2400], device='cuda:1'), covar=tensor([0.3003, 0.3086, 0.1939, 0.2702], device='cuda:1'), in_proj_covar=tensor([0.2093, 0.2062, 0.1973, 0.2115], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 13:20:13,458 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1627, 1.2456, 3.2647, 2.9676], device='cuda:1'), covar=tensor([0.1567, 0.2637, 0.0537, 0.1598], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0684, 0.1026, 0.1001], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 13:20:13,774 INFO [train.py:968] (1/2) Epoch 30, batch 27150, giga_loss[loss=0.3343, simple_loss=0.3926, pruned_loss=0.138, over 27997.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3776, pruned_loss=0.1242, over 5658770.06 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5693749.20 frames. ], giga_tot_loss[loss=0.3141, simple_loss=0.3786, pruned_loss=0.1248, over 5658139.14 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:20:16,657 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1347621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:20:19,326 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1347624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:20:47,292 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1347653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:20:57,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.29 vs. limit=2.0
+2023-03-15 13:21:00,963 INFO [train.py:968] (1/2) Epoch 30, batch 27200, giga_loss[loss=0.3222, simple_loss=0.3815, pruned_loss=0.1315, over 28371.00 frames. ], tot_loss[loss=0.3125, simple_loss=0.3774, pruned_loss=0.1239, over 5642401.88 frames. ], libri_tot_loss[loss=0.292, simple_loss=0.3596, pruned_loss=0.1122, over 5690898.68 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3782, pruned_loss=0.1244, over 5643602.50 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:21:32,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.50 vs. limit=5.0
+2023-03-15 13:21:33,299 INFO [optim.py:369] (1/2) 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 13:21:43,760 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3625, 1.4859, 1.4789, 1.3275], device='cuda:1'), covar=tensor([0.3333, 0.3191, 0.2322, 0.2794], device='cuda:1'), in_proj_covar=tensor([0.2095, 0.2063, 0.1975, 0.2116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 13:21:46,207 INFO [train.py:968] (1/2) Epoch 30, batch 27250, giga_loss[loss=0.3581, simple_loss=0.3871, pruned_loss=0.1646, over 23544.00 frames. ], tot_loss[loss=0.3092, simple_loss=0.376, pruned_loss=0.1212, over 5646994.17 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5685566.46 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3773, pruned_loss=0.122, over 5652775.03 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:21:52,673 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1443, 1.2235, 1.0507, 0.8653], device='cuda:1'), covar=tensor([0.0860, 0.0386, 0.0821, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0458, 0.0530, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 13:22:11,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.7991, 3.6283, 3.4546, 1.6593], device='cuda:1'), covar=tensor([0.0824, 0.0936, 0.0948, 0.2230], device='cuda:1'), in_proj_covar=tensor([0.1347, 0.1241, 0.1041, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 13:22:25,916 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1347754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:22:40,220 INFO [train.py:968] (1/2) Epoch 30, batch 27300, giga_loss[loss=0.3191, simple_loss=0.3849, pruned_loss=0.1267, over 28983.00 frames. ], tot_loss[loss=0.309, simple_loss=0.3764, pruned_loss=0.1208, over 5650330.95 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3593, pruned_loss=0.1122, over 5685365.48 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3775, pruned_loss=0.1214, over 5654694.14 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:23:21,048 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 27350, giga_loss[loss=0.2869, simple_loss=0.3589, pruned_loss=0.1075, over 28923.00 frames. ], tot_loss[loss=0.3094, simple_loss=0.3766, pruned_loss=0.1211, over 5655560.01 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3594, pruned_loss=0.1122, over 5686486.27 frames. ], giga_tot_loss[loss=0.3104, simple_loss=0.3776, pruned_loss=0.1216, over 5657485.33 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:23:43,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3272, 1.4256, 3.8021, 3.3114], device='cuda:1'), covar=tensor([0.1667, 0.2682, 0.0488, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0687, 0.1028, 0.1005], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 13:24:21,476 INFO [train.py:968] (1/2) Epoch 30, batch 27400, giga_loss[loss=0.3365, simple_loss=0.3928, pruned_loss=0.1401, over 28572.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.375, pruned_loss=0.1204, over 5668207.48 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.1121, over 5690776.59 frames. ], giga_tot_loss[loss=0.3091, simple_loss=0.3762, pruned_loss=0.1211, over 5665526.13 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:24:58,143 INFO [optim.py:369] (1/2) 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 13:25:15,648 INFO [train.py:968] (1/2) Epoch 30, batch 27450, giga_loss[loss=0.3119, simple_loss=0.372, pruned_loss=0.1259, over 28260.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.373, pruned_loss=0.121, over 5652208.63 frames. ], libri_tot_loss[loss=0.2918, simple_loss=0.3594, pruned_loss=0.1121, over 5691237.23 frames. ], giga_tot_loss[loss=0.309, simple_loss=0.3744, pruned_loss=0.1218, over 5649025.30 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:26:02,450 INFO [train.py:968] (1/2) Epoch 30, batch 27500, giga_loss[loss=0.3002, simple_loss=0.3642, pruned_loss=0.1181, over 28964.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.372, pruned_loss=0.1214, over 5651928.91 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3587, pruned_loss=0.1115, over 5698457.08 frames. ], giga_tot_loss[loss=0.31, simple_loss=0.3742, pruned_loss=0.1229, over 5640878.76 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:26:17,116 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2285, 2.4846, 1.3029, 1.3467], device='cuda:1'), covar=tensor([0.0992, 0.0402, 0.0885, 0.1388], device='cuda:1'), in_proj_covar=tensor([0.0435, 0.0585, 0.0420, 0.0456], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 13:26:35,501 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 13:26:35,581 INFO [optim.py:369] (1/2) 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 13:26:49,652 INFO [train.py:968] (1/2) Epoch 30, batch 27550, libri_loss[loss=0.3029, simple_loss=0.3722, pruned_loss=0.1168, over 29477.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.37, pruned_loss=0.1204, over 5644384.86 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.359, pruned_loss=0.1117, over 5683614.83 frames. ], giga_tot_loss[loss=0.3079, simple_loss=0.372, pruned_loss=0.1219, over 5646965.12 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:27:34,285 INFO [train.py:968] (1/2) Epoch 30, batch 27600, giga_loss[loss=0.3044, simple_loss=0.3664, pruned_loss=0.1212, over 28624.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3696, pruned_loss=0.1212, over 5650125.88 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.112, over 5691573.19 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3713, pruned_loss=0.1224, over 5643931.55 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:28:06,815 INFO [optim.py:369] (1/2) 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 13:28:14,249 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:28:19,221 INFO [train.py:968] (1/2) Epoch 30, batch 27650, giga_loss[loss=0.2796, simple_loss=0.3513, pruned_loss=0.1039, over 28547.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1196, over 5652160.08 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3595, pruned_loss=0.1122, over 5693760.95 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3692, pruned_loss=0.1206, over 5644539.05 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:28:29,066 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:28:32,936 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7865, 1.9974, 1.7777, 1.7060], device='cuda:1'), covar=tensor([0.2300, 0.2740, 0.2531, 0.2564], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0762, 0.0736, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 13:29:02,337 INFO [train.py:968] (1/2) Epoch 30, batch 27700, giga_loss[loss=0.266, simple_loss=0.3414, pruned_loss=0.09525, over 28986.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3637, pruned_loss=0.1151, over 5664795.07 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.112, over 5699848.70 frames. ], giga_tot_loss[loss=0.2987, simple_loss=0.3651, pruned_loss=0.1162, over 5652029.44 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:29:06,056 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2914, 1.5317, 1.2849, 1.5495], device='cuda:1'), covar=tensor([0.0809, 0.0357, 0.0363, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0124, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0077, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 13:29:14,157 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2015, 1.5438, 1.1715, 0.7255], device='cuda:1'), covar=tensor([0.4775, 0.2565, 0.3206, 0.6499], device='cuda:1'), in_proj_covar=tensor([0.1885, 0.1764, 0.1684, 0.1525], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 13:29:21,371 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7420, 1.8470, 1.6640, 1.7033], device='cuda:1'), covar=tensor([0.2322, 0.2402, 0.2394, 0.2237], device='cuda:1'), in_proj_covar=tensor([0.1627, 0.1172, 0.1440, 0.1023], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 13:29:34,776 INFO [optim.py:369] (1/2) 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 13:29:51,194 INFO [train.py:968] (1/2) Epoch 30, batch 27750, giga_loss[loss=0.246, simple_loss=0.3214, pruned_loss=0.08535, over 28935.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3614, pruned_loss=0.1127, over 5670574.92 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3594, pruned_loss=0.1122, over 5702259.24 frames. ], giga_tot_loss[loss=0.2946, simple_loss=0.3625, pruned_loss=0.1134, over 5657765.26 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:30:39,816 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9603, 1.1409, 1.1186, 0.8832], device='cuda:1'), covar=tensor([0.2661, 0.3075, 0.2031, 0.2502], device='cuda:1'), in_proj_covar=tensor([0.2095, 0.2069, 0.1975, 0.2119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 13:30:40,119 INFO [train.py:968] (1/2) Epoch 30, batch 27800, giga_loss[loss=0.2882, simple_loss=0.3644, pruned_loss=0.106, over 28933.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.361, pruned_loss=0.1126, over 5667828.33 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.112, over 5708244.36 frames. ], giga_tot_loss[loss=0.2943, simple_loss=0.3621, pruned_loss=0.1133, over 5650966.29 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:30:45,562 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1348272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:30:47,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:31:00,880 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7477, 1.9518, 1.4042, 1.4661], device='cuda:1'), covar=tensor([0.1104, 0.0671, 0.1084, 0.1171], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0458, 0.0529, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 13:31:18,959 INFO [optim.py:369] (1/2) 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 13:31:19,228 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 27850, giga_loss[loss=0.2689, simple_loss=0.3389, pruned_loss=0.09943, over 28851.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3571, pruned_loss=0.1109, over 5670898.78 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3589, pruned_loss=0.1118, over 5710273.87 frames. ], giga_tot_loss[loss=0.2908, simple_loss=0.3582, pruned_loss=0.1117, over 5655553.25 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:32:30,781 INFO [train.py:968] (1/2) Epoch 30, batch 27900, giga_loss[loss=0.3482, simple_loss=0.394, pruned_loss=0.1512, over 27994.00 frames. ], tot_loss[loss=0.2911, simple_loss=0.3578, pruned_loss=0.1122, over 5670444.14 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3587, pruned_loss=0.1115, over 5715862.64 frames. ], giga_tot_loss[loss=0.2925, simple_loss=0.3588, pruned_loss=0.113, over 5652067.04 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:32:51,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4264, 1.3950, 3.8399, 3.2040], device='cuda:1'), covar=tensor([0.1637, 0.2824, 0.0482, 0.1238], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0687, 0.1031, 0.1009], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 13:33:04,429 INFO [optim.py:369] (1/2) 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 13:33:17,567 INFO [train.py:968] (1/2) Epoch 30, batch 27950, giga_loss[loss=0.2932, simple_loss=0.3716, pruned_loss=0.1074, over 28920.00 frames. ], tot_loss[loss=0.2922, simple_loss=0.3597, pruned_loss=0.1123, over 5674322.62 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3586, pruned_loss=0.1115, over 5711310.32 frames. ], giga_tot_loss[loss=0.2934, simple_loss=0.3606, pruned_loss=0.1131, over 5662132.68 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:33:32,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9467, 1.1596, 1.0566, 0.8957], device='cuda:1'), covar=tensor([0.2543, 0.3143, 0.2043, 0.2590], device='cuda:1'), in_proj_covar=tensor([0.2101, 0.2072, 0.1979, 0.2124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 13:34:01,145 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 13:34:06,387 INFO [train.py:968] (1/2) Epoch 30, batch 28000, giga_loss[loss=0.2687, simple_loss=0.3443, pruned_loss=0.09659, over 28965.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3623, pruned_loss=0.1142, over 5661379.84 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3584, pruned_loss=0.1113, over 5710674.66 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3634, pruned_loss=0.115, over 5651247.55 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:34:14,474 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:34:16,865 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0141, 1.3032, 1.1013, 0.2456], device='cuda:1'), covar=tensor([0.4840, 0.3542, 0.4722, 0.7591], device='cuda:1'), in_proj_covar=tensor([0.1880, 0.1760, 0.1679, 0.1521], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 13:34:27,038 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:34:43,502 INFO [optim.py:369] (1/2) 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 13:34:45,029 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.86 vs. limit=5.0
+2023-03-15 13:34:54,742 INFO [train.py:968] (1/2) Epoch 30, batch 28050, giga_loss[loss=0.3201, simple_loss=0.3748, pruned_loss=0.1327, over 27814.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3648, pruned_loss=0.1162, over 5649417.82 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3586, pruned_loss=0.1114, over 5703376.09 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3654, pruned_loss=0.1167, over 5647948.34 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:35:43,386 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9521, 1.1369, 1.0857, 0.8541], device='cuda:1'), covar=tensor([0.2502, 0.3084, 0.1879, 0.2588], device='cuda:1'), in_proj_covar=tensor([0.2098, 0.2070, 0.1974, 0.2121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 13:35:43,704 INFO [train.py:968] (1/2) Epoch 30, batch 28100, giga_loss[loss=0.3121, simple_loss=0.3744, pruned_loss=0.1249, over 29097.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3653, pruned_loss=0.1165, over 5651900.62 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3589, pruned_loss=0.1114, over 5707463.19 frames. ], giga_tot_loss[loss=0.2999, simple_loss=0.3656, pruned_loss=0.1171, over 5646084.67 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:36:12,945 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348601.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:36:15,707 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 28150, giga_loss[loss=0.3148, simple_loss=0.3802, pruned_loss=0.1247, over 28675.00 frames. ], tot_loss[loss=0.3012, simple_loss=0.367, pruned_loss=0.1177, over 5664861.57 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3589, pruned_loss=0.1113, over 5710853.08 frames. ], giga_tot_loss[loss=0.302, simple_loss=0.3675, pruned_loss=0.1183, over 5656500.34 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:36:42,099 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1348630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:36:45,389 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:37:11,859 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1348662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:37:15,574 INFO [train.py:968] (1/2) Epoch 30, batch 28200, libri_loss[loss=0.2926, simple_loss=0.3688, pruned_loss=0.1081, over 29538.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3696, pruned_loss=0.1193, over 5671210.98 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.359, pruned_loss=0.1114, over 5716845.93 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3702, pruned_loss=0.12, over 5657647.33 frames. ], batch size: 89, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:37:15,974 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6359, 1.4366, 4.8435, 3.7054], device='cuda:1'), covar=tensor([0.1647, 0.2738, 0.0404, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0685, 0.1026, 0.1006], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 13:37:50,516 INFO [optim.py:369] (1/2) 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 13:38:04,710 INFO [train.py:968] (1/2) Epoch 30, batch 28250, giga_loss[loss=0.2842, simple_loss=0.3588, pruned_loss=0.1048, over 28677.00 frames. ], tot_loss[loss=0.3054, simple_loss=0.3704, pruned_loss=0.1202, over 5656938.22 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3588, pruned_loss=0.1114, over 5708658.46 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3712, pruned_loss=0.1208, over 5652203.42 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:38:55,396 INFO [train.py:968] (1/2) Epoch 30, batch 28300, giga_loss[loss=0.3903, simple_loss=0.4185, pruned_loss=0.181, over 26528.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3717, pruned_loss=0.122, over 5652150.01 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3589, pruned_loss=0.1114, over 5707512.65 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3724, pruned_loss=0.1226, over 5648962.83 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:39:34,517 INFO [optim.py:369] (1/2) 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 13:39:48,624 INFO [train.py:968] (1/2) Epoch 30, batch 28350, giga_loss[loss=0.2956, simple_loss=0.3703, pruned_loss=0.1104, over 28569.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.373, pruned_loss=0.1212, over 5654629.49 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.3592, pruned_loss=0.1117, over 5708654.25 frames. ], giga_tot_loss[loss=0.3084, simple_loss=0.3735, pruned_loss=0.1216, over 5650116.49 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:40:23,843 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:40:37,920 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7914, 2.0211, 1.9505, 1.6129], device='cuda:1'), covar=tensor([0.3145, 0.2713, 0.3097, 0.3002], device='cuda:1'), in_proj_covar=tensor([0.2102, 0.2073, 0.1978, 0.2121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 13:40:41,097 INFO [train.py:968] (1/2) Epoch 30, batch 28400, libri_loss[loss=0.268, simple_loss=0.3482, pruned_loss=0.09389, over 28593.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3727, pruned_loss=0.1201, over 5665268.73 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3591, pruned_loss=0.1117, over 5711472.45 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3735, pruned_loss=0.1206, over 5658241.77 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:40:50,541 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:40:59,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 13:41:17,203 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4357, 1.3867, 4.1344, 3.3885], device='cuda:1'), covar=tensor([0.1675, 0.2738, 0.0464, 0.0902], device='cuda:1'), in_proj_covar=tensor([0.0824, 0.0686, 0.1027, 0.1007], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 13:41:20,206 INFO [optim.py:369] (1/2) 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 13:41:29,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2925, 4.1236, 2.3630, 2.3426], device='cuda:1'), covar=tensor([0.0710, 0.0394, 0.0680, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0582, 0.0418, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 13:41:32,264 INFO [train.py:968] (1/2) Epoch 30, batch 28450, giga_loss[loss=0.2829, simple_loss=0.3536, pruned_loss=0.1061, over 28784.00 frames. ], tot_loss[loss=0.3061, simple_loss=0.3715, pruned_loss=0.1204, over 5661411.18 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3595, pruned_loss=0.1122, over 5704268.52 frames. ], giga_tot_loss[loss=0.3065, simple_loss=0.3719, pruned_loss=0.1205, over 5662108.20 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:42:25,445 INFO [train.py:968] (1/2) Epoch 30, batch 28500, giga_loss[loss=0.2653, simple_loss=0.3432, pruned_loss=0.09372, over 28624.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3702, pruned_loss=0.1201, over 5658481.31 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3593, pruned_loss=0.1121, over 5698588.79 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.371, pruned_loss=0.1204, over 5664036.79 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:42:33,460 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1348976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:40,264 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1348982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:52,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1309, 1.4374, 1.4072, 1.4958], device='cuda:1'), covar=tensor([0.0818, 0.0335, 0.0300, 0.1025], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 13:42:53,779 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1348993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:57,083 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1348996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:42:59,288 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4268, 1.7140, 1.4179, 1.0089], device='cuda:1'), covar=tensor([0.2681, 0.2803, 0.3151, 0.2584], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1173, 0.1441, 0.1023], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 13:43:06,894 INFO [optim.py:369] (1/2) 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 13:43:16,360 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.16 vs. limit=2.0
+2023-03-15 13:43:22,202 INFO [train.py:968] (1/2) Epoch 30, batch 28550, giga_loss[loss=0.2826, simple_loss=0.3572, pruned_loss=0.104, over 28943.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3679, pruned_loss=0.1187, over 5638331.68 frames. ], libri_tot_loss[loss=0.2924, simple_loss=0.3597, pruned_loss=0.1125, over 5669738.64 frames. ], giga_tot_loss[loss=0.3031, simple_loss=0.3684, pruned_loss=0.1189, over 5666733.50 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:43:27,665 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:44:07,144 INFO [train.py:968] (1/2) Epoch 30, batch 28600, giga_loss[loss=0.3109, simple_loss=0.3754, pruned_loss=0.1232, over 28895.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3675, pruned_loss=0.1186, over 5652021.13 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3597, pruned_loss=0.1124, over 5671139.11 frames. ], giga_tot_loss[loss=0.303, simple_loss=0.3681, pruned_loss=0.1189, over 5672849.47 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:44:20,523 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4048, 1.6506, 1.4546, 1.4849], device='cuda:1'), covar=tensor([0.0692, 0.0421, 0.0337, 0.0759], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0124, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 13:44:26,605 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.22 vs. limit=2.0
+2023-03-15 13:44:41,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8342, 4.6969, 4.4809, 2.2684], device='cuda:1'), covar=tensor([0.0499, 0.0609, 0.0708, 0.1831], device='cuda:1'), in_proj_covar=tensor([0.1342, 0.1238, 0.1038, 0.0763], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 13:44:49,048 INFO [optim.py:369] (1/2) 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 13:44:54,786 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6667, 1.9044, 1.3813, 1.5200], device='cuda:1'), covar=tensor([0.0975, 0.0554, 0.0994, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0459, 0.0529, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 13:44:58,756 INFO [train.py:968] (1/2) Epoch 30, batch 28650, giga_loss[loss=0.2831, simple_loss=0.3603, pruned_loss=0.103, over 28838.00 frames. ], tot_loss[loss=0.3036, simple_loss=0.3679, pruned_loss=0.1196, over 5631755.23 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.36, pruned_loss=0.1125, over 5665188.86 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3682, pruned_loss=0.1199, over 5652891.72 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:44:59,837 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:45:01,781 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:45:32,260 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:45:40,038 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 13:45:48,526 INFO [train.py:968] (1/2) Epoch 30, batch 28700, giga_loss[loss=0.2354, simple_loss=0.3157, pruned_loss=0.07751, over 28595.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3679, pruned_loss=0.1199, over 5640254.46 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3602, pruned_loss=0.1128, over 5663876.67 frames. ], giga_tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.12, over 5658009.01 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:46:23,859 INFO [optim.py:369] (1/2) 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 13:46:34,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2606, 1.4136, 1.4262, 1.2064], device='cuda:1'), covar=tensor([0.2798, 0.2537, 0.1917, 0.2431], device='cuda:1'), in_proj_covar=tensor([0.2096, 0.2069, 0.1974, 0.2117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 13:46:35,347 INFO [train.py:968] (1/2) Epoch 30, batch 28750, giga_loss[loss=0.2897, simple_loss=0.3503, pruned_loss=0.1145, over 28841.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3713, pruned_loss=0.1232, over 5622775.12 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5641755.03 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3712, pruned_loss=0.1231, over 5655189.05 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 13:46:39,154 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3411, 3.1827, 3.0499, 1.4331], device='cuda:1'), covar=tensor([0.0989, 0.1103, 0.0995, 0.2232], device='cuda:1'), in_proj_covar=tensor([0.1344, 0.1242, 0.1042, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 13:46:47,637 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9268, 3.7625, 3.5693, 1.6090], device='cuda:1'), covar=tensor([0.0767, 0.0895, 0.0852, 0.2211], device='cuda:1'), in_proj_covar=tensor([0.1344, 0.1242, 0.1042, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 13:47:08,051 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 28800, giga_loss[loss=0.2721, simple_loss=0.3447, pruned_loss=0.09978, over 28195.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3728, pruned_loss=0.1246, over 5624043.00 frames. ], libri_tot_loss[loss=0.2936, simple_loss=0.3608, pruned_loss=0.1132, over 5645924.01 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3728, pruned_loss=0.1246, over 5645636.27 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:48:05,682 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8766, 5.2240, 2.0787, 2.2434], device='cuda:1'), covar=tensor([0.0986, 0.0276, 0.0851, 0.1264], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0582, 0.0418, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 13:48:06,939 INFO [optim.py:369] (1/2) 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 13:48:17,879 INFO [train.py:968] (1/2) Epoch 30, batch 28850, giga_loss[loss=0.4394, simple_loss=0.4562, pruned_loss=0.2113, over 28279.00 frames. ], tot_loss[loss=0.3119, simple_loss=0.3729, pruned_loss=0.1254, over 5623880.62 frames. ], libri_tot_loss[loss=0.2937, simple_loss=0.3609, pruned_loss=0.1133, over 5648251.43 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.373, pruned_loss=0.1255, over 5638752.04 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:48:28,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5948, 1.8207, 1.5543, 1.6185], device='cuda:1'), covar=tensor([0.2127, 0.2198, 0.2281, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1174, 0.1441, 0.1023], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 13:48:56,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1349357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:06,101 INFO [train.py:968] (1/2) Epoch 30, batch 28900, giga_loss[loss=0.2891, simple_loss=0.357, pruned_loss=0.1106, over 28257.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3717, pruned_loss=0.1245, over 5636606.50 frames. ], libri_tot_loss[loss=0.2939, simple_loss=0.361, pruned_loss=0.1134, over 5650975.90 frames. ], giga_tot_loss[loss=0.3105, simple_loss=0.3718, pruned_loss=0.1246, over 5645956.56 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:49:07,780 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1349370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:33,173 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349395.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:35,617 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:49:43,221 INFO [optim.py:369] (1/2) 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 13:49:52,623 INFO [train.py:968] (1/2) Epoch 30, batch 28950, giga_loss[loss=0.2695, simple_loss=0.3426, pruned_loss=0.09822, over 28953.00 frames. ], tot_loss[loss=0.3088, simple_loss=0.3706, pruned_loss=0.1235, over 5636809.42 frames. ], libri_tot_loss[loss=0.2933, simple_loss=0.3605, pruned_loss=0.1131, over 5657157.44 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.3714, pruned_loss=0.1241, over 5638318.44 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:50:01,607 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:50:36,041 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-15 13:50:39,842 INFO [train.py:968] (1/2) Epoch 30, batch 29000, giga_loss[loss=0.2827, simple_loss=0.3565, pruned_loss=0.1044, over 28760.00 frames. ], tot_loss[loss=0.307, simple_loss=0.3698, pruned_loss=0.1221, over 5645747.47 frames. ], libri_tot_loss[loss=0.293, simple_loss=0.3601, pruned_loss=0.1129, over 5663940.15 frames. ], giga_tot_loss[loss=0.3086, simple_loss=0.3711, pruned_loss=0.123, over 5640097.14 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:51:03,643 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2400, 2.9026, 1.4355, 1.3835], device='cuda:1'), covar=tensor([0.1075, 0.0387, 0.0922, 0.1417], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0582, 0.0418, 0.0455], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 13:51:11,398 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:51:14,219 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:51:17,207 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1349505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:51:18,130 INFO [optim.py:369] (1/2) 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 13:51:26,958 INFO [train.py:968] (1/2) Epoch 30, batch 29050, giga_loss[loss=0.3345, simple_loss=0.3819, pruned_loss=0.1435, over 27467.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3699, pruned_loss=0.1214, over 5658665.77 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.36, pruned_loss=0.1129, over 5669860.13 frames. ], giga_tot_loss[loss=0.308, simple_loss=0.3712, pruned_loss=0.1224, over 5648457.40 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:51:41,433 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:52:15,498 INFO [train.py:968] (1/2) Epoch 30, batch 29100, giga_loss[loss=0.2981, simple_loss=0.3687, pruned_loss=0.1138, over 28939.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3704, pruned_loss=0.1221, over 5668626.38 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3594, pruned_loss=0.1125, over 5673028.35 frames. ], giga_tot_loss[loss=0.3095, simple_loss=0.3722, pruned_loss=0.1234, over 5657434.53 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:52:45,273 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.11 vs. limit=2.0
+2023-03-15 13:52:50,970 INFO [optim.py:369] (1/2) 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 13:53:00,927 INFO [train.py:968] (1/2) Epoch 30, batch 29150, giga_loss[loss=0.3662, simple_loss=0.4205, pruned_loss=0.156, over 28258.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3716, pruned_loss=0.1229, over 5674523.54 frames. ], libri_tot_loss[loss=0.2919, simple_loss=0.3592, pruned_loss=0.1123, over 5670619.88 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3734, pruned_loss=0.1243, over 5667361.37 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:53:47,318 INFO [train.py:968] (1/2) Epoch 30, batch 29200, giga_loss[loss=0.3101, simple_loss=0.3903, pruned_loss=0.115, over 28728.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3734, pruned_loss=0.1239, over 5671577.55 frames. ], libri_tot_loss[loss=0.2915, simple_loss=0.359, pruned_loss=0.112, over 5677364.31 frames. ], giga_tot_loss[loss=0.3135, simple_loss=0.3756, pruned_loss=0.1257, over 5659488.72 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 13:54:27,646 INFO [optim.py:369] (1/2) 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 13:54:38,221 INFO [train.py:968] (1/2) Epoch 30, batch 29250, giga_loss[loss=0.2825, simple_loss=0.3623, pruned_loss=0.1014, over 28698.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3739, pruned_loss=0.1229, over 5671958.06 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3587, pruned_loss=0.1117, over 5681940.64 frames. ], giga_tot_loss[loss=0.3129, simple_loss=0.3763, pruned_loss=0.1248, over 5658141.05 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:55:07,349 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1349745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:55:26,917 INFO [train.py:968] (1/2) Epoch 30, batch 29300, giga_loss[loss=0.2515, simple_loss=0.3317, pruned_loss=0.08571, over 28378.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3734, pruned_loss=0.1218, over 5673452.96 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3591, pruned_loss=0.1121, over 5685009.85 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3751, pruned_loss=0.1232, over 5659651.31 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:56:05,769 INFO [optim.py:369] (1/2) 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 13:56:09,253 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1349812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:56:14,940 INFO [train.py:968] (1/2) Epoch 30, batch 29350, giga_loss[loss=0.3154, simple_loss=0.378, pruned_loss=0.1264, over 28608.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.371, pruned_loss=0.1207, over 5671769.17 frames. ], libri_tot_loss[loss=0.2916, simple_loss=0.3592, pruned_loss=0.112, over 5688997.56 frames. ], giga_tot_loss[loss=0.3082, simple_loss=0.3725, pruned_loss=0.122, over 5656967.84 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:56:52,513 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-15 13:57:00,304 INFO [train.py:968] (1/2) Epoch 30, batch 29400, giga_loss[loss=0.3184, simple_loss=0.3801, pruned_loss=0.1283, over 28889.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3714, pruned_loss=0.1208, over 5668623.63 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1117, over 5692386.79 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3732, pruned_loss=0.1223, over 5653365.88 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:57:15,801 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1349880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:57:24,114 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1349888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:57:27,641 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1349891.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:57:41,567 INFO [optim.py:369] (1/2) 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 13:57:54,963 INFO [train.py:968] (1/2) Epoch 30, batch 29450, giga_loss[loss=0.2894, simple_loss=0.3596, pruned_loss=0.1095, over 28545.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.372, pruned_loss=0.1211, over 5667712.23 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3584, pruned_loss=0.1115, over 5697232.86 frames. ], giga_tot_loss[loss=0.3098, simple_loss=0.374, pruned_loss=0.1227, over 5650739.63 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:57:57,244 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1349920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:58:31,721 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 13:58:42,412 INFO [train.py:968] (1/2) Epoch 30, batch 29500, giga_loss[loss=0.372, simple_loss=0.4099, pruned_loss=0.1671, over 27601.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.3724, pruned_loss=0.1222, over 5673539.13 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1117, over 5701991.17 frames. ], giga_tot_loss[loss=0.3107, simple_loss=0.3741, pruned_loss=0.1236, over 5654850.91 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:59:21,431 INFO [optim.py:369] (1/2) 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 13:59:30,499 INFO [train.py:968] (1/2) Epoch 30, batch 29550, giga_loss[loss=0.3354, simple_loss=0.368, pruned_loss=0.1514, over 23394.00 frames. ], tot_loss[loss=0.3102, simple_loss=0.3732, pruned_loss=0.1235, over 5670391.85 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3588, pruned_loss=0.1117, over 5704286.00 frames. ], giga_tot_loss[loss=0.3121, simple_loss=0.3746, pruned_loss=0.1247, over 5653435.62 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 13:59:36,531 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1350023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 13:59:38,528 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1350026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:00:04,498 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 29600, giga_loss[loss=0.3117, simple_loss=0.3751, pruned_loss=0.1241, over 28712.00 frames. ], tot_loss[loss=0.311, simple_loss=0.3739, pruned_loss=0.1241, over 5669831.72 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.359, pruned_loss=0.1117, over 5705950.62 frames. ], giga_tot_loss[loss=0.3134, simple_loss=0.3756, pruned_loss=0.1257, over 5652423.57 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:00:20,445 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5842, 2.0457, 1.8265, 1.8378], device='cuda:1'), covar=tensor([0.0764, 0.0275, 0.0301, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0124, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 14:00:48,698 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6218, 1.9755, 2.0878, 1.5887], device='cuda:1'), covar=tensor([0.2027, 0.1792, 0.1929, 0.2141], device='cuda:1'), in_proj_covar=tensor([0.0521, 0.0771, 0.0744, 0.0712], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 14:00:52,822 INFO [optim.py:369] (1/2) 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 14:01:04,027 INFO [train.py:968] (1/2) Epoch 30, batch 29650, giga_loss[loss=0.3224, simple_loss=0.3843, pruned_loss=0.1302, over 28956.00 frames. ], tot_loss[loss=0.3097, simple_loss=0.3729, pruned_loss=0.1233, over 5671431.31 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3585, pruned_loss=0.1114, over 5711425.65 frames. ], giga_tot_loss[loss=0.3128, simple_loss=0.3752, pruned_loss=0.1252, over 5651248.03 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:01:22,397 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9043, 3.7604, 3.5701, 1.8785], device='cuda:1'), covar=tensor([0.0710, 0.0786, 0.0767, 0.2111], device='cuda:1'), in_proj_covar=tensor([0.1350, 0.1246, 0.1045, 0.0767], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 14:01:54,731 INFO [train.py:968] (1/2) Epoch 30, batch 29700, giga_loss[loss=0.4103, simple_loss=0.4537, pruned_loss=0.1835, over 28591.00 frames. ], tot_loss[loss=0.3096, simple_loss=0.3729, pruned_loss=0.1232, over 5669194.96 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3587, pruned_loss=0.1115, over 5709828.58 frames. ], giga_tot_loss[loss=0.312, simple_loss=0.3747, pruned_loss=0.1247, over 5654440.63 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:02:11,462 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:02:40,876 INFO [train.py:968] (1/2) Epoch 30, batch 29750, libri_loss[loss=0.3217, simple_loss=0.3888, pruned_loss=0.1273, over 29699.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3723, pruned_loss=0.1219, over 5669066.44 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.359, pruned_loss=0.1119, over 5695464.51 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3738, pruned_loss=0.123, over 5668848.11 frames. ], batch size: 91, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:03:26,328 INFO [train.py:968] (1/2) Epoch 30, batch 29800, giga_loss[loss=0.2794, simple_loss=0.3539, pruned_loss=0.1025, over 28963.00 frames. ], tot_loss[loss=0.306, simple_loss=0.3709, pruned_loss=0.1206, over 5658582.46 frames. ], libri_tot_loss[loss=0.2913, simple_loss=0.359, pruned_loss=0.1118, over 5694927.55 frames. ], giga_tot_loss[loss=0.3083, simple_loss=0.3726, pruned_loss=0.1219, over 5657587.36 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:03:41,181 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7065, 1.9502, 1.3405, 1.4838], device='cuda:1'), covar=tensor([0.1092, 0.0621, 0.1073, 0.1174], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0458, 0.0529, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 14:04:02,266 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 29850, giga_loss[loss=0.3262, simple_loss=0.374, pruned_loss=0.1392, over 23467.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3706, pruned_loss=0.1199, over 5655868.15 frames. ], libri_tot_loss[loss=0.2922, simple_loss=0.3596, pruned_loss=0.1123, over 5692864.96 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3719, pruned_loss=0.1209, over 5656164.34 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:04:24,149 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1350330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:04:26,287 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1350333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:04:53,980 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1350362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:04:57,929 INFO [train.py:968] (1/2) Epoch 30, batch 29900, giga_loss[loss=0.2768, simple_loss=0.3463, pruned_loss=0.1037, over 28895.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3692, pruned_loss=0.1196, over 5657467.38 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3599, pruned_loss=0.1125, over 5691505.43 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.37, pruned_loss=0.1203, over 5658583.67 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:05:37,620 INFO [optim.py:369] (1/2) 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 14:05:44,391 INFO [train.py:968] (1/2) Epoch 30, batch 29950, giga_loss[loss=0.2858, simple_loss=0.3538, pruned_loss=0.1089, over 28899.00 frames. ], tot_loss[loss=0.3026, simple_loss=0.3675, pruned_loss=0.1188, over 5657843.12 frames. ], libri_tot_loss[loss=0.2927, simple_loss=0.3601, pruned_loss=0.1127, over 5693150.72 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3682, pruned_loss=0.1194, over 5657006.29 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:06:36,090 INFO [train.py:968] (1/2) Epoch 30, batch 30000, giga_loss[loss=0.2565, simple_loss=0.3321, pruned_loss=0.09044, over 28920.00 frames. ], tot_loss[loss=0.2986, simple_loss=0.3635, pruned_loss=0.1169, over 5666019.75 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3591, pruned_loss=0.1122, over 5697259.45 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.3652, pruned_loss=0.118, over 5660248.10 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:06:36,091 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 14:06:44,918 INFO [train.py:1012] (1/2) Epoch 30, validation: loss=0.2021, simple_loss=0.3107, pruned_loss=0.04675, over 944034.00 frames. 
+2023-03-15 14:06:44,919 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 14:07:25,680 INFO [optim.py:369] (1/2) 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 14:07:31,146 INFO [train.py:968] (1/2) Epoch 30, batch 30050, giga_loss[loss=0.3266, simple_loss=0.3828, pruned_loss=0.1352, over 28789.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3622, pruned_loss=0.1167, over 5681713.33 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3597, pruned_loss=0.1125, over 5699895.09 frames. ], giga_tot_loss[loss=0.2989, simple_loss=0.3631, pruned_loss=0.1173, over 5674434.33 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:07:36,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6939, 1.4651, 5.0914, 3.7486], device='cuda:1'), covar=tensor([0.1728, 0.2849, 0.0433, 0.0872], device='cuda:1'), in_proj_covar=tensor([0.0829, 0.0687, 0.1031, 0.1010], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 14:07:41,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2059, 4.0647, 3.8683, 1.8742], device='cuda:1'), covar=tensor([0.0636, 0.0760, 0.0774, 0.2113], device='cuda:1'), in_proj_covar=tensor([0.1352, 0.1248, 0.1047, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0013], device='cuda:1')
+2023-03-15 14:07:49,320 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1350537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:08:21,023 INFO [train.py:968] (1/2) Epoch 30, batch 30100, giga_loss[loss=0.2979, simple_loss=0.3664, pruned_loss=0.1147, over 29012.00 frames. ], tot_loss[loss=0.2975, simple_loss=0.3617, pruned_loss=0.1167, over 5691708.68 frames. ], libri_tot_loss[loss=0.2929, simple_loss=0.3602, pruned_loss=0.1128, over 5701203.59 frames. ], giga_tot_loss[loss=0.2981, simple_loss=0.362, pruned_loss=0.1171, over 5684161.22 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:08:30,193 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-15 14:09:03,837 INFO [optim.py:369] (1/2) 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 14:09:10,459 INFO [train.py:968] (1/2) Epoch 30, batch 30150, libri_loss[loss=0.2687, simple_loss=0.3405, pruned_loss=0.09842, over 29525.00 frames. ], tot_loss[loss=0.2961, simple_loss=0.3608, pruned_loss=0.1157, over 5691767.39 frames. ], libri_tot_loss[loss=0.2928, simple_loss=0.3601, pruned_loss=0.1128, over 5706524.75 frames. ], giga_tot_loss[loss=0.2967, simple_loss=0.3611, pruned_loss=0.1162, over 5680564.66 frames. ], batch size: 80, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:09:58,608 INFO [train.py:968] (1/2) Epoch 30, batch 30200, giga_loss[loss=0.2724, simple_loss=0.3563, pruned_loss=0.09426, over 28946.00 frames. ], tot_loss[loss=0.2932, simple_loss=0.36, pruned_loss=0.1132, over 5688173.21 frames. ], libri_tot_loss[loss=0.2926, simple_loss=0.3601, pruned_loss=0.1126, over 5711409.24 frames. ], giga_tot_loss[loss=0.2939, simple_loss=0.3604, pruned_loss=0.1137, over 5674636.69 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:10:50,344 INFO [optim.py:369] (1/2) 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 14:10:57,134 INFO [train.py:968] (1/2) Epoch 30, batch 30250, giga_loss[loss=0.2726, simple_loss=0.3493, pruned_loss=0.09796, over 28905.00 frames. ], tot_loss[loss=0.2885, simple_loss=0.3577, pruned_loss=0.1096, over 5676922.49 frames. ], libri_tot_loss[loss=0.2923, simple_loss=0.3598, pruned_loss=0.1124, over 5713123.93 frames. ], giga_tot_loss[loss=0.2892, simple_loss=0.3581, pruned_loss=0.1102, over 5664290.61 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:11:06,586 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5703, 1.8395, 1.8151, 1.5682], device='cuda:1'), covar=tensor([0.2679, 0.2239, 0.1867, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.2087, 0.2059, 0.1965, 0.2108], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 14:11:47,636 INFO [train.py:968] (1/2) Epoch 30, batch 30300, giga_loss[loss=0.2513, simple_loss=0.339, pruned_loss=0.08178, over 29053.00 frames. ], tot_loss[loss=0.284, simple_loss=0.3546, pruned_loss=0.1067, over 5668697.45 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3597, pruned_loss=0.1126, over 5713861.37 frames. ], giga_tot_loss[loss=0.2844, simple_loss=0.355, pruned_loss=0.1069, over 5657142.39 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:12:33,170 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 30350, giga_loss[loss=0.2246, simple_loss=0.315, pruned_loss=0.0671, over 29104.00 frames. ], tot_loss[loss=0.2787, simple_loss=0.3507, pruned_loss=0.1034, over 5664853.07 frames. ], libri_tot_loss[loss=0.2925, simple_loss=0.3595, pruned_loss=0.1127, over 5716894.37 frames. ], giga_tot_loss[loss=0.2788, simple_loss=0.3511, pruned_loss=0.1032, over 5652085.28 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:12:43,706 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4363, 1.6940, 1.3477, 1.5562], device='cuda:1'), covar=tensor([0.0781, 0.0382, 0.0384, 0.0886], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 14:12:54,141 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 14:12:57,659 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1350833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:13:09,610 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7686, 1.2526, 1.3819, 1.0228], device='cuda:1'), covar=tensor([0.2583, 0.1504, 0.2302, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0762, 0.0736, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 14:13:31,477 INFO [train.py:968] (1/2) Epoch 30, batch 30400, giga_loss[loss=0.2461, simple_loss=0.331, pruned_loss=0.08067, over 27941.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3497, pruned_loss=0.1011, over 5662769.50 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.3592, pruned_loss=0.1126, over 5718698.77 frames. ], giga_tot_loss[loss=0.276, simple_loss=0.3501, pruned_loss=0.1009, over 5649961.01 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:13:33,553 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4088, 1.6035, 1.2509, 1.5226], device='cuda:1'), covar=tensor([0.0799, 0.0346, 0.0383, 0.0925], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 14:14:17,512 INFO [optim.py:369] (1/2) 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 14:14:18,459 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1350912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:14:24,256 INFO [train.py:968] (1/2) Epoch 30, batch 30450, libri_loss[loss=0.2637, simple_loss=0.3393, pruned_loss=0.09404, over 29532.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3491, pruned_loss=0.1, over 5636933.69 frames. ], libri_tot_loss[loss=0.2921, simple_loss=0.359, pruned_loss=0.1126, over 5702375.24 frames. ], giga_tot_loss[loss=0.2741, simple_loss=0.3493, pruned_loss=0.09947, over 5639428.74 frames. ], batch size: 84, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:14:25,300 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3781, 1.2636, 3.9583, 3.4409], device='cuda:1'), covar=tensor([0.1718, 0.3182, 0.0474, 0.1519], device='cuda:1'), in_proj_covar=tensor([0.0825, 0.0685, 0.1027, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 14:15:12,216 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1350964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:15:15,274 INFO [train.py:968] (1/2) Epoch 30, batch 30500, giga_loss[loss=0.2474, simple_loss=0.3147, pruned_loss=0.09002, over 24248.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3494, pruned_loss=0.1004, over 5638378.47 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3584, pruned_loss=0.1124, over 5705774.26 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3498, pruned_loss=0.09988, over 5635951.57 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:15:16,986 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 14:15:49,432 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:15:56,810 INFO [optim.py:369] (1/2) 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 14:16:03,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-15 14:16:05,575 INFO [train.py:968] (1/2) Epoch 30, batch 30550, giga_loss[loss=0.2328, simple_loss=0.3159, pruned_loss=0.07484, over 28701.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3471, pruned_loss=0.09938, over 5617429.59 frames. ], libri_tot_loss[loss=0.2917, simple_loss=0.3581, pruned_loss=0.1126, over 5689201.59 frames. ], giga_tot_loss[loss=0.2719, simple_loss=0.3474, pruned_loss=0.09825, over 5628427.00 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:16:41,518 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:16:43,281 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:16:53,639 INFO [train.py:968] (1/2) Epoch 30, batch 30600, giga_loss[loss=0.2355, simple_loss=0.3218, pruned_loss=0.07466, over 28831.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.344, pruned_loss=0.09706, over 5633795.32 frames. ], libri_tot_loss[loss=0.291, simple_loss=0.3574, pruned_loss=0.1123, over 5693043.68 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3446, pruned_loss=0.09607, over 5637557.76 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:17:16,431 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351087.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:17:34,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4193, 4.2558, 4.0332, 2.3132], device='cuda:1'), covar=tensor([0.0579, 0.0720, 0.0893, 0.2056], device='cuda:1'), in_proj_covar=tensor([0.1327, 0.1228, 0.1026, 0.0756], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 14:17:38,093 INFO [optim.py:369] (1/2) 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 14:17:43,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.33 vs. limit=2.0
+2023-03-15 14:17:45,481 INFO [train.py:968] (1/2) Epoch 30, batch 30650, giga_loss[loss=0.2407, simple_loss=0.306, pruned_loss=0.08767, over 24063.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3431, pruned_loss=0.0966, over 5633855.01 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3571, pruned_loss=0.1124, over 5697322.60 frames. ], giga_tot_loss[loss=0.2672, simple_loss=0.3435, pruned_loss=0.09548, over 5631875.11 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:18:26,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:18:34,787 INFO [train.py:968] (1/2) Epoch 30, batch 30700, giga_loss[loss=0.2424, simple_loss=0.3331, pruned_loss=0.07581, over 28790.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3431, pruned_loss=0.09593, over 5642725.54 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3568, pruned_loss=0.1123, over 5696107.26 frames. ], giga_tot_loss[loss=0.2666, simple_loss=0.3435, pruned_loss=0.09483, over 5641202.38 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:18:53,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4605, 1.5252, 1.2384, 1.1645], device='cuda:1'), covar=tensor([0.0853, 0.0310, 0.0738, 0.0948], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0452, 0.0522, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 14:19:18,493 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351208.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:19:21,314 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 30750, giga_loss[loss=0.2508, simple_loss=0.328, pruned_loss=0.08683, over 27632.00 frames. ], tot_loss[loss=0.2634, simple_loss=0.3402, pruned_loss=0.0933, over 5645177.03 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3564, pruned_loss=0.1121, over 5698222.83 frames. ], giga_tot_loss[loss=0.2628, simple_loss=0.3407, pruned_loss=0.09243, over 5641625.61 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:20:26,985 INFO [train.py:968] (1/2) Epoch 30, batch 30800, giga_loss[loss=0.2383, simple_loss=0.3236, pruned_loss=0.07652, over 28986.00 frames. ], tot_loss[loss=0.2596, simple_loss=0.3373, pruned_loss=0.09094, over 5641295.61 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3562, pruned_loss=0.112, over 5700338.37 frames. ], giga_tot_loss[loss=0.259, simple_loss=0.3377, pruned_loss=0.09019, over 5636195.02 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:21:05,976 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 14:21:10,720 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 30850, giga_loss[loss=0.2478, simple_loss=0.3289, pruned_loss=0.08332, over 28661.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3335, pruned_loss=0.08919, over 5638747.32 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3558, pruned_loss=0.1119, over 5695507.02 frames. ], giga_tot_loss[loss=0.255, simple_loss=0.3337, pruned_loss=0.08818, over 5637883.84 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:21:40,065 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:21:53,857 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:21:57,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:22:11,757 INFO [train.py:968] (1/2) Epoch 30, batch 30900, giga_loss[loss=0.216, simple_loss=0.2834, pruned_loss=0.07435, over 23815.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3325, pruned_loss=0.08911, over 5640618.63 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3557, pruned_loss=0.1118, over 5696563.48 frames. ], giga_tot_loss[loss=0.2546, simple_loss=0.3327, pruned_loss=0.08827, over 5638667.38 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:22:21,942 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:22:27,997 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351383.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:23:01,860 INFO [optim.py:369] (1/2) 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] (1/2) 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] (1/2) Epoch 30, batch 30950, giga_loss[loss=0.3112, simple_loss=0.3626, pruned_loss=0.1299, over 23948.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3333, pruned_loss=0.08991, over 5621825.69 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3558, pruned_loss=0.112, over 5697490.85 frames. ], giga_tot_loss[loss=0.2554, simple_loss=0.3331, pruned_loss=0.08887, over 5618803.68 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:24:05,907 INFO [train.py:968] (1/2) Epoch 30, batch 31000, giga_loss[loss=0.2625, simple_loss=0.3426, pruned_loss=0.09126, over 29068.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.337, pruned_loss=0.09177, over 5634235.64 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3558, pruned_loss=0.1124, over 5702142.73 frames. ], giga_tot_loss[loss=0.2578, simple_loss=0.3361, pruned_loss=0.08977, over 5625349.44 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:24:21,551 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:24:26,764 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:24:57,362 INFO [optim.py:369] (1/2) 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 14:24:59,263 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:03,087 INFO [train.py:968] (1/2) Epoch 30, batch 31050, giga_loss[loss=0.2383, simple_loss=0.3268, pruned_loss=0.07489, over 28629.00 frames. ], tot_loss[loss=0.2609, simple_loss=0.3384, pruned_loss=0.0917, over 5640659.71 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3552, pruned_loss=0.1122, over 5695320.15 frames. ], giga_tot_loss[loss=0.2587, simple_loss=0.3377, pruned_loss=0.08979, over 5638556.46 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:25:06,707 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:08,806 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:23,320 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:25:31,658 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 14:25:48,274 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:26:05,100 INFO [train.py:968] (1/2) Epoch 30, batch 31100, giga_loss[loss=0.2485, simple_loss=0.3329, pruned_loss=0.08202, over 28905.00 frames. ], tot_loss[loss=0.2623, simple_loss=0.3395, pruned_loss=0.09253, over 5655583.65 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.355, pruned_loss=0.1123, over 5692152.48 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.3386, pruned_loss=0.09027, over 5655792.86 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:27:04,089 INFO [optim.py:369] (1/2) 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 14:27:09,556 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4527, 1.6647, 1.4619, 1.6180], device='cuda:1'), covar=tensor([0.0688, 0.0301, 0.0321, 0.0743], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 14:27:11,547 INFO [train.py:968] (1/2) Epoch 30, batch 31150, giga_loss[loss=0.249, simple_loss=0.3319, pruned_loss=0.08309, over 28947.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.3368, pruned_loss=0.09105, over 5650703.13 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.3549, pruned_loss=0.1124, over 5696683.40 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3358, pruned_loss=0.08868, over 5645969.60 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:27:13,232 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:28:16,297 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.66 vs. limit=2.0
+2023-03-15 14:28:19,733 INFO [train.py:968] (1/2) Epoch 30, batch 31200, giga_loss[loss=0.2394, simple_loss=0.3274, pruned_loss=0.07572, over 28945.00 frames. ], tot_loss[loss=0.2574, simple_loss=0.3362, pruned_loss=0.08926, over 5656147.36 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3548, pruned_loss=0.1123, over 5697888.53 frames. ], giga_tot_loss[loss=0.2548, simple_loss=0.3353, pruned_loss=0.08712, over 5650894.93 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:28:29,432 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:28:32,050 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:29:07,219 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1351709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:29:12,592 INFO [optim.py:369] (1/2) 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 14:29:18,855 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3717, 1.6161, 1.2246, 1.1596], device='cuda:1'), covar=tensor([0.1085, 0.0451, 0.0947, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0416, 0.0451, 0.0523, 0.0461], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 14:29:19,966 INFO [train.py:968] (1/2) Epoch 30, batch 31250, giga_loss[loss=0.2445, simple_loss=0.3234, pruned_loss=0.08284, over 28900.00 frames. ], tot_loss[loss=0.2552, simple_loss=0.334, pruned_loss=0.0882, over 5663649.36 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3547, pruned_loss=0.1124, over 5697303.19 frames. ], giga_tot_loss[loss=0.2525, simple_loss=0.333, pruned_loss=0.08596, over 5659124.16 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:30:27,705 INFO [train.py:968] (1/2) Epoch 30, batch 31300, giga_loss[loss=0.2264, simple_loss=0.3055, pruned_loss=0.07366, over 28590.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3315, pruned_loss=0.08749, over 5658553.52 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3546, pruned_loss=0.1123, over 5698294.68 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3306, pruned_loss=0.08567, over 5654009.46 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:30:53,968 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:31:22,431 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 31350, giga_loss[loss=0.286, simple_loss=0.3649, pruned_loss=0.1035, over 28530.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3308, pruned_loss=0.08741, over 5662345.12 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3548, pruned_loss=0.1125, over 5697899.42 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3297, pruned_loss=0.08547, over 5658559.02 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:31:29,608 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5930, 1.8698, 1.5363, 1.7766], device='cuda:1'), covar=tensor([0.2878, 0.2837, 0.3088, 0.2781], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1168, 0.1442, 0.1021], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 14:31:31,316 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 14:31:49,537 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 14:32:19,462 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6860, 1.9331, 1.6444, 1.5772], device='cuda:1'), covar=tensor([0.2966, 0.2596, 0.2732, 0.2650], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1168, 0.1441, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 14:32:21,943 INFO [train.py:968] (1/2) Epoch 30, batch 31400, giga_loss[loss=0.2704, simple_loss=0.3546, pruned_loss=0.09307, over 28722.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.3317, pruned_loss=0.08732, over 5658588.91 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3542, pruned_loss=0.1123, over 5696547.10 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3305, pruned_loss=0.08507, over 5655666.00 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:32:44,162 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:33:18,548 INFO [optim.py:369] (1/2) 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 14:33:21,810 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 14:33:23,375 INFO [train.py:968] (1/2) Epoch 30, batch 31450, giga_loss[loss=0.2654, simple_loss=0.3459, pruned_loss=0.09248, over 28130.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3338, pruned_loss=0.08822, over 5668198.01 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3529, pruned_loss=0.1117, over 5703347.75 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3332, pruned_loss=0.086, over 5658209.27 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:33:43,815 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1351932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:33:48,673 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1351935.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:34:05,296 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1351949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:34:20,700 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 31500, giga_loss[loss=0.2793, simple_loss=0.3532, pruned_loss=0.1027, over 28726.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08703, over 5677567.40 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3522, pruned_loss=0.1113, over 5709040.65 frames. ], giga_tot_loss[loss=0.2506, simple_loss=0.3314, pruned_loss=0.08492, over 5663500.98 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:35:05,199 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1351994.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:35:28,470 INFO [optim.py:369] (1/2) 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 14:35:35,671 INFO [train.py:968] (1/2) Epoch 30, batch 31550, giga_loss[loss=0.2782, simple_loss=0.3532, pruned_loss=0.1015, over 28088.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3324, pruned_loss=0.0875, over 5679061.10 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3523, pruned_loss=0.1114, over 5709750.02 frames. ], giga_tot_loss[loss=0.2509, simple_loss=0.3315, pruned_loss=0.08519, over 5666425.49 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:36:42,605 INFO [train.py:968] (1/2) Epoch 30, batch 31600, giga_loss[loss=0.2607, simple_loss=0.3561, pruned_loss=0.08268, over 28994.00 frames. ], tot_loss[loss=0.255, simple_loss=0.3348, pruned_loss=0.08762, over 5675305.82 frames. ], libri_tot_loss[loss=0.2872, simple_loss=0.3519, pruned_loss=0.1112, over 5711843.57 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3343, pruned_loss=0.08573, over 5663388.24 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:37:43,595 INFO [optim.py:369] (1/2) 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 14:37:47,098 INFO [train.py:968] (1/2) Epoch 30, batch 31650, giga_loss[loss=0.3081, simple_loss=0.3957, pruned_loss=0.1102, over 28867.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.339, pruned_loss=0.08775, over 5656504.96 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3516, pruned_loss=0.1112, over 5697837.53 frames. ], giga_tot_loss[loss=0.2549, simple_loss=0.3384, pruned_loss=0.08566, over 5657688.45 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:38:16,066 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:38:19,957 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1352140.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:38:54,381 INFO [train.py:968] (1/2) Epoch 30, batch 31700, giga_loss[loss=0.2589, simple_loss=0.3373, pruned_loss=0.09023, over 26623.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3389, pruned_loss=0.08644, over 5648588.90 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3515, pruned_loss=0.1111, over 5698852.35 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3385, pruned_loss=0.08474, over 5648377.81 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:38:56,100 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:39:22,008 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2985, 1.5458, 1.6254, 1.3553], device='cuda:1'), covar=tensor([0.3239, 0.2751, 0.1732, 0.2551], device='cuda:1'), in_proj_covar=tensor([0.2054, 0.2022, 0.1918, 0.2065], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 14:39:49,085 INFO [optim.py:369] (1/2) 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 14:39:53,859 INFO [train.py:968] (1/2) Epoch 30, batch 31750, libri_loss[loss=0.2675, simple_loss=0.3279, pruned_loss=0.1035, over 29539.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3382, pruned_loss=0.08563, over 5647997.01 frames. ], libri_tot_loss[loss=0.2864, simple_loss=0.3512, pruned_loss=0.1109, over 5695879.71 frames. ], giga_tot_loss[loss=0.2527, simple_loss=0.3378, pruned_loss=0.08382, over 5649296.75 frames. ], batch size: 74, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:40:46,692 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1352262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:40:49,987 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.3784, 1.5571, 5.6285, 4.3071], device='cuda:1'), covar=tensor([0.1417, 0.2699, 0.0697, 0.0869], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0683, 0.1017, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 14:40:52,845 INFO [train.py:968] (1/2) Epoch 30, batch 31800, giga_loss[loss=0.2329, simple_loss=0.3218, pruned_loss=0.07196, over 28851.00 frames. ], tot_loss[loss=0.2559, simple_loss=0.3383, pruned_loss=0.08676, over 5645345.09 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3503, pruned_loss=0.1105, over 5691058.97 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3384, pruned_loss=0.08459, over 5648252.11 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:41:50,490 INFO [optim.py:369] (1/2) 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 14:41:56,399 INFO [train.py:968] (1/2) Epoch 30, batch 31850, giga_loss[loss=0.2553, simple_loss=0.3324, pruned_loss=0.0891, over 28158.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3372, pruned_loss=0.08719, over 5660267.29 frames. ], libri_tot_loss[loss=0.2855, simple_loss=0.3501, pruned_loss=0.1104, over 5698402.98 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3369, pruned_loss=0.08475, over 5654395.39 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:42:08,916 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1352324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:43:14,640 INFO [train.py:968] (1/2) Epoch 30, batch 31900, giga_loss[loss=0.2565, simple_loss=0.3395, pruned_loss=0.08674, over 28906.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3367, pruned_loss=0.08735, over 5671468.49 frames. ], libri_tot_loss[loss=0.2849, simple_loss=0.3495, pruned_loss=0.1101, over 5701873.92 frames. ], giga_tot_loss[loss=0.2532, simple_loss=0.3365, pruned_loss=0.08491, over 5662493.32 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:44:12,329 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352405.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:44:17,628 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 14:44:29,815 INFO [train.py:968] (1/2) Epoch 30, batch 31950, giga_loss[loss=0.2109, simple_loss=0.3022, pruned_loss=0.05981, over 28368.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3325, pruned_loss=0.08487, over 5680313.86 frames. ], libri_tot_loss[loss=0.284, simple_loss=0.3488, pruned_loss=0.1096, over 5707440.74 frames. ], giga_tot_loss[loss=0.2491, simple_loss=0.3327, pruned_loss=0.08272, over 5667385.60 frames. ], batch size: 369, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:44:59,169 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:01,499 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:27,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6174, 1.9640, 1.5375, 1.6906], device='cuda:1'), covar=tensor([0.2743, 0.2750, 0.3193, 0.2548], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1169, 0.1443, 0.1022], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 14:45:33,842 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:34,178 INFO [train.py:968] (1/2) Epoch 30, batch 32000, giga_loss[loss=0.218, simple_loss=0.3098, pruned_loss=0.06309, over 29036.00 frames. ], tot_loss[loss=0.249, simple_loss=0.3303, pruned_loss=0.08386, over 5678180.94 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3485, pruned_loss=0.1094, over 5712955.59 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3301, pruned_loss=0.08142, over 5661476.21 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 14:45:38,856 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1352470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:45:44,222 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3731, 1.6451, 1.4372, 1.5729], device='cuda:1'), covar=tensor([0.0775, 0.0362, 0.0359, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 14:45:46,871 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.57 vs. limit=5.0
+2023-03-15 14:46:12,414 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352499.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:46:31,485 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 32050, giga_loss[loss=0.2686, simple_loss=0.346, pruned_loss=0.09566, over 28958.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.3275, pruned_loss=0.08307, over 5682353.86 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3478, pruned_loss=0.1091, over 5718295.96 frames. ], giga_tot_loss[loss=0.2437, simple_loss=0.3271, pruned_loss=0.08014, over 5662044.14 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:47:37,382 INFO [train.py:968] (1/2) Epoch 30, batch 32100, giga_loss[loss=0.2769, simple_loss=0.3674, pruned_loss=0.09327, over 29116.00 frames. ], tot_loss[loss=0.25, simple_loss=0.3305, pruned_loss=0.08477, over 5678474.76 frames. ], libri_tot_loss[loss=0.2831, simple_loss=0.3478, pruned_loss=0.1092, over 5713898.76 frames. ], giga_tot_loss[loss=0.2467, simple_loss=0.3297, pruned_loss=0.08181, over 5665028.63 frames. ], batch size: 285, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:48:25,052 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.42 vs. limit=2.0
+2023-03-15 14:48:36,042 INFO [optim.py:369] (1/2) 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 14:48:38,519 INFO [train.py:968] (1/2) Epoch 30, batch 32150, giga_loss[loss=0.2416, simple_loss=0.3267, pruned_loss=0.07828, over 28898.00 frames. ], tot_loss[loss=0.2541, simple_loss=0.3341, pruned_loss=0.08705, over 5680907.94 frames. ], libri_tot_loss[loss=0.2823, simple_loss=0.347, pruned_loss=0.1088, over 5717858.67 frames. ], giga_tot_loss[loss=0.2514, simple_loss=0.3339, pruned_loss=0.08447, over 5665819.62 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:49:39,827 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 14:49:41,008 INFO [train.py:968] (1/2) Epoch 30, batch 32200, giga_loss[loss=0.2247, simple_loss=0.2907, pruned_loss=0.07929, over 24617.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.3318, pruned_loss=0.08699, over 5678089.09 frames. ], libri_tot_loss[loss=0.2816, simple_loss=0.3462, pruned_loss=0.1085, over 5725025.62 frames. ], giga_tot_loss[loss=0.2503, simple_loss=0.3318, pruned_loss=0.0844, over 5657648.73 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:49:46,061 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352672.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:50:41,283 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 32250, giga_loss[loss=0.2367, simple_loss=0.3209, pruned_loss=0.07627, over 29002.00 frames. ], tot_loss[loss=0.2544, simple_loss=0.3325, pruned_loss=0.08819, over 5680630.76 frames. ], libri_tot_loss[loss=0.2812, simple_loss=0.3459, pruned_loss=0.1082, over 5724650.88 frames. ], giga_tot_loss[loss=0.2518, simple_loss=0.3324, pruned_loss=0.08565, over 5662903.46 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:51:24,370 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352751.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:51:44,067 INFO [train.py:968] (1/2) Epoch 30, batch 32300, giga_loss[loss=0.2395, simple_loss=0.3295, pruned_loss=0.07478, over 28743.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3327, pruned_loss=0.08828, over 5674010.38 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3452, pruned_loss=0.108, over 5720156.45 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3328, pruned_loss=0.08577, over 5661816.79 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:52:53,291 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1352815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:52:53,657 INFO [optim.py:369] (1/2) 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 14:52:56,699 INFO [train.py:968] (1/2) Epoch 30, batch 32350, giga_loss[loss=0.2413, simple_loss=0.3417, pruned_loss=0.07043, over 28897.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.335, pruned_loss=0.08854, over 5663939.70 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3449, pruned_loss=0.1078, over 5712849.47 frames. ], giga_tot_loss[loss=0.254, simple_loss=0.3353, pruned_loss=0.08641, over 5660645.64 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 14:53:30,079 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1352838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:54:19,526 INFO [train.py:968] (1/2) Epoch 30, batch 32400, giga_loss[loss=0.2426, simple_loss=0.3224, pruned_loss=0.0814, over 28784.00 frames. ], tot_loss[loss=0.254, simple_loss=0.3342, pruned_loss=0.08691, over 5659194.31 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3447, pruned_loss=0.1078, over 5705742.38 frames. ], giga_tot_loss[loss=0.2522, simple_loss=0.3344, pruned_loss=0.08505, over 5661526.04 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:54:32,269 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3111, 1.4809, 1.4500, 1.2846], device='cuda:1'), covar=tensor([0.2768, 0.2513, 0.1954, 0.2535], device='cuda:1'), in_proj_covar=tensor([0.2050, 0.2013, 0.1910, 0.2063], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 14:55:23,046 INFO [optim.py:369] (1/2) 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 14:55:24,485 INFO [train.py:968] (1/2) Epoch 30, batch 32450, libri_loss[loss=0.3513, simple_loss=0.3904, pruned_loss=0.1561, over 18921.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3323, pruned_loss=0.08697, over 5660946.24 frames. ], libri_tot_loss[loss=0.2805, simple_loss=0.3449, pruned_loss=0.108, over 5701221.37 frames. ], giga_tot_loss[loss=0.2504, simple_loss=0.3318, pruned_loss=0.08446, over 5666456.76 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:56:14,840 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1352958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:56:19,024 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1352961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:56:19,243 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-15 14:56:29,621 INFO [train.py:968] (1/2) Epoch 30, batch 32500, libri_loss[loss=0.2426, simple_loss=0.3194, pruned_loss=0.08293, over 29502.00 frames. ], tot_loss[loss=0.2492, simple_loss=0.3276, pruned_loss=0.08539, over 5651423.51 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.345, pruned_loss=0.1081, over 5689301.44 frames. ], giga_tot_loss[loss=0.2458, simple_loss=0.3266, pruned_loss=0.08252, over 5665135.89 frames. ], batch size: 81, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:56:56,551 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1352990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:57:31,598 INFO [optim.py:369] (1/2) 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 14:57:32,749 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 14:57:32,959 INFO [train.py:968] (1/2) Epoch 30, batch 32550, libri_loss[loss=0.2619, simple_loss=0.3381, pruned_loss=0.09282, over 29403.00 frames. ], tot_loss[loss=0.247, simple_loss=0.325, pruned_loss=0.08453, over 5656040.74 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3443, pruned_loss=0.1076, over 5697037.95 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.324, pruned_loss=0.08163, over 5658427.96 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:58:07,510 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:58:33,774 INFO [train.py:968] (1/2) Epoch 30, batch 32600, giga_loss[loss=0.2318, simple_loss=0.3172, pruned_loss=0.07321, over 28882.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3263, pruned_loss=0.08535, over 5648845.54 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3443, pruned_loss=0.1076, over 5695152.43 frames. ], giga_tot_loss[loss=0.2455, simple_loss=0.3254, pruned_loss=0.08286, over 5651737.74 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:59:23,891 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.6844, 1.1948, 1.2234, 0.9975], device='cuda:1'), covar=tensor([0.2350, 0.1381, 0.2127, 0.1700], device='cuda:1'), in_proj_covar=tensor([0.0504, 0.0750, 0.0725, 0.0696], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 14:59:32,270 INFO [optim.py:369] (1/2) 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 14:59:35,904 INFO [train.py:968] (1/2) Epoch 30, batch 32650, giga_loss[loss=0.2283, simple_loss=0.2996, pruned_loss=0.07849, over 26832.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.3265, pruned_loss=0.08523, over 5645141.14 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3444, pruned_loss=0.1077, over 5687642.09 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3253, pruned_loss=0.08268, over 5653637.09 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 14:59:47,126 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 14:59:49,814 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:00:26,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4353, 1.8304, 1.4907, 1.6133], device='cuda:1'), covar=tensor([0.0729, 0.0410, 0.0365, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0121, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 15:00:38,551 INFO [train.py:968] (1/2) Epoch 30, batch 32700, giga_loss[loss=0.2454, simple_loss=0.3261, pruned_loss=0.08231, over 28956.00 frames. ], tot_loss[loss=0.2476, simple_loss=0.3259, pruned_loss=0.08462, over 5639514.33 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3444, pruned_loss=0.1079, over 5683577.88 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3244, pruned_loss=0.08178, over 5648357.90 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:01:10,165 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:01:14,190 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:01:42,279 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:01:45,503 INFO [optim.py:369] (1/2) 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 15:01:47,878 INFO [train.py:968] (1/2) Epoch 30, batch 32750, giga_loss[loss=0.2702, simple_loss=0.3408, pruned_loss=0.09981, over 27585.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3247, pruned_loss=0.08354, over 5648402.62 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3446, pruned_loss=0.1081, over 5684707.99 frames. ], giga_tot_loss[loss=0.2427, simple_loss=0.3233, pruned_loss=0.08106, over 5654135.19 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:01:55,129 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353222.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:02:57,426 INFO [train.py:968] (1/2) Epoch 30, batch 32800, giga_loss[loss=0.1994, simple_loss=0.2877, pruned_loss=0.05551, over 28866.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.3234, pruned_loss=0.0828, over 5650969.18 frames. ], libri_tot_loss[loss=0.2807, simple_loss=0.3448, pruned_loss=0.1083, over 5682096.54 frames. ], giga_tot_loss[loss=0.2408, simple_loss=0.3216, pruned_loss=0.08, over 5657212.19 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:03:00,910 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:03:04,455 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:03:44,349 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:03:44,568 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.74 vs. limit=2.0
+2023-03-15 15:04:02,415 INFO [optim.py:369] (1/2) 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 15:04:03,058 INFO [train.py:968] (1/2) Epoch 30, batch 32850, giga_loss[loss=0.2593, simple_loss=0.3412, pruned_loss=0.08871, over 28346.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.3241, pruned_loss=0.08303, over 5651899.12 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3444, pruned_loss=0.1081, over 5690191.74 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3223, pruned_loss=0.0802, over 5648191.91 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:04:12,466 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.28 vs. limit=2.0
+2023-03-15 15:04:53,576 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:04:54,425 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:04:57,785 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 32900, giga_loss[loss=0.2581, simple_loss=0.3252, pruned_loss=0.09546, over 26762.00 frames. ], tot_loss[loss=0.2459, simple_loss=0.3245, pruned_loss=0.08367, over 5655509.06 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3444, pruned_loss=0.1081, over 5691530.68 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3227, pruned_loss=0.08094, over 5650617.07 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:05:34,593 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 15:06:14,037 INFO [train.py:968] (1/2) Epoch 30, batch 32950, giga_loss[loss=0.2343, simple_loss=0.3156, pruned_loss=0.07646, over 28920.00 frames. ], tot_loss[loss=0.2465, simple_loss=0.3247, pruned_loss=0.08418, over 5651655.89 frames. ], libri_tot_loss[loss=0.2809, simple_loss=0.3448, pruned_loss=0.1085, over 5680654.98 frames. ], giga_tot_loss[loss=0.2428, simple_loss=0.3227, pruned_loss=0.08142, over 5656194.78 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:06:29,691 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4574, 1.8069, 1.4078, 1.6218], device='cuda:1'), covar=tensor([0.3028, 0.3013, 0.3538, 0.2467], device='cuda:1'), in_proj_covar=tensor([0.1636, 0.1170, 0.1447, 0.1026], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0010], device='cuda:1')
+2023-03-15 15:07:19,608 INFO [train.py:968] (1/2) Epoch 30, batch 33000, giga_loss[loss=0.2417, simple_loss=0.3327, pruned_loss=0.0754, over 28539.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.325, pruned_loss=0.08306, over 5654579.03 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3445, pruned_loss=0.1084, over 5683876.54 frames. ], giga_tot_loss[loss=0.2423, simple_loss=0.3234, pruned_loss=0.08063, over 5654893.17 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:07:19,609 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 15:07:28,210 INFO [train.py:1012] (1/2) Epoch 30, validation: loss=0.1937, simple_loss=0.2953, pruned_loss=0.04605, over 944034.00 frames. 
+2023-03-15 15:07:28,211 INFO [train.py:1013] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 15:08:09,529 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353503.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:08:28,313 INFO [optim.py:369] (1/2) 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 15:08:30,574 INFO [train.py:968] (1/2) Epoch 30, batch 33050, giga_loss[loss=0.219, simple_loss=0.3212, pruned_loss=0.05841, over 28879.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.3273, pruned_loss=0.08298, over 5658313.02 frames. ], libri_tot_loss[loss=0.2803, simple_loss=0.3442, pruned_loss=0.1082, over 5685917.66 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3261, pruned_loss=0.08099, over 5656574.43 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:08:43,361 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4004, 1.5386, 1.5451, 1.4142], device='cuda:1'), covar=tensor([0.2422, 0.2332, 0.1753, 0.2136], device='cuda:1'), in_proj_covar=tensor([0.2045, 0.2004, 0.1904, 0.2056], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 15:08:46,512 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.72 vs. limit=2.0
+2023-03-15 15:09:30,711 INFO [train.py:968] (1/2) Epoch 30, batch 33100, giga_loss[loss=0.2571, simple_loss=0.3382, pruned_loss=0.08798, over 28086.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.329, pruned_loss=0.08404, over 5650185.05 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3438, pruned_loss=0.1077, over 5687844.57 frames. ], giga_tot_loss[loss=0.2462, simple_loss=0.3281, pruned_loss=0.08218, over 5646188.16 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:10:36,548 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 33150, giga_loss[loss=0.2459, simple_loss=0.3308, pruned_loss=0.08045, over 28872.00 frames. ], tot_loss[loss=0.25, simple_loss=0.33, pruned_loss=0.08501, over 5637317.35 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3437, pruned_loss=0.1078, over 5673979.10 frames. ], giga_tot_loss[loss=0.2473, simple_loss=0.3289, pruned_loss=0.08289, over 5645862.21 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:10:38,026 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.49 vs. limit=5.0
+2023-03-15 15:11:07,218 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-15 15:11:13,352 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:11:15,571 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 33200, giga_loss[loss=0.2071, simple_loss=0.2972, pruned_loss=0.0585, over 28821.00 frames. ], tot_loss[loss=0.2484, simple_loss=0.3283, pruned_loss=0.08418, over 5651342.52 frames. ], libri_tot_loss[loss=0.2783, simple_loss=0.3425, pruned_loss=0.1071, over 5678869.25 frames. ], giga_tot_loss[loss=0.2466, simple_loss=0.3283, pruned_loss=0.08248, over 5652807.82 frames. ], batch size: 243, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:11:46,694 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:12:35,819 INFO [optim.py:369] (1/2) 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 15:12:36,833 INFO [train.py:968] (1/2) Epoch 30, batch 33250, giga_loss[loss=0.2667, simple_loss=0.3435, pruned_loss=0.09498, over 28983.00 frames. ], tot_loss[loss=0.2461, simple_loss=0.3263, pruned_loss=0.08292, over 5648953.93 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3422, pruned_loss=0.107, over 5672769.30 frames. ], giga_tot_loss[loss=0.2444, simple_loss=0.3262, pruned_loss=0.08128, over 5654810.97 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:12:51,362 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1353730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:13:37,423 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:13:39,051 INFO [train.py:968] (1/2) Epoch 30, batch 33300, giga_loss[loss=0.2183, simple_loss=0.2982, pruned_loss=0.06919, over 28870.00 frames. ], tot_loss[loss=0.245, simple_loss=0.3243, pruned_loss=0.08282, over 5645580.38 frames. ], libri_tot_loss[loss=0.2779, simple_loss=0.3418, pruned_loss=0.107, over 5666274.09 frames. ], giga_tot_loss[loss=0.2432, simple_loss=0.3243, pruned_loss=0.08105, over 5656315.75 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:13:48,114 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5115, 1.8160, 1.4855, 1.2558], device='cuda:1'), covar=tensor([0.2877, 0.2934, 0.3477, 0.2686], device='cuda:1'), in_proj_covar=tensor([0.1633, 0.1168, 0.1444, 0.1022], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 15:14:32,668 INFO [optim.py:369] (1/2) 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 15:14:33,602 INFO [train.py:968] (1/2) Epoch 30, batch 33350, giga_loss[loss=0.2596, simple_loss=0.3436, pruned_loss=0.08782, over 28367.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3255, pruned_loss=0.08355, over 5662862.71 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3415, pruned_loss=0.1068, over 5674829.43 frames. ], giga_tot_loss[loss=0.2438, simple_loss=0.325, pruned_loss=0.08127, over 5663343.63 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:15:42,878 INFO [train.py:968] (1/2) Epoch 30, batch 33400, libri_loss[loss=0.2878, simple_loss=0.3568, pruned_loss=0.1094, over 29655.00 frames. ], tot_loss[loss=0.2496, simple_loss=0.3292, pruned_loss=0.08503, over 5661838.83 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3417, pruned_loss=0.1067, over 5677266.32 frames. ], giga_tot_loss[loss=0.2472, simple_loss=0.3284, pruned_loss=0.08303, over 5659659.87 frames. ], batch size: 88, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:15:50,282 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1353873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:15:52,861 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1353876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:15:55,511 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8801, 3.7410, 3.5448, 1.9918], device='cuda:1'), covar=tensor([0.0744, 0.0863, 0.0928, 0.1973], device='cuda:1'), in_proj_covar=tensor([0.1310, 0.1207, 0.1013, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 15:16:30,035 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1353905.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:16:43,010 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 33450, giga_loss[loss=0.2193, simple_loss=0.2905, pruned_loss=0.07408, over 24254.00 frames. ], tot_loss[loss=0.2506, simple_loss=0.3296, pruned_loss=0.08577, over 5655877.64 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3418, pruned_loss=0.1069, over 5678466.88 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3286, pruned_loss=0.08349, over 5652532.74 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:16:59,136 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 15:17:26,257 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.27 vs. limit=2.0
+2023-03-15 15:17:29,348 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-15 15:17:48,019 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.84 vs. limit=5.0
+2023-03-15 15:17:54,926 INFO [train.py:968] (1/2) Epoch 30, batch 33500, giga_loss[loss=0.2274, simple_loss=0.3051, pruned_loss=0.07489, over 28451.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3318, pruned_loss=0.08673, over 5673718.18 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3417, pruned_loss=0.1068, over 5682368.89 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3308, pruned_loss=0.08452, over 5667225.85 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:18:18,247 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1353987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:18:52,342 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 33550, giga_loss[loss=0.196, simple_loss=0.2739, pruned_loss=0.05901, over 24252.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3344, pruned_loss=0.08746, over 5652010.02 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3416, pruned_loss=0.1069, over 5666811.40 frames. ], giga_tot_loss[loss=0.2521, simple_loss=0.3335, pruned_loss=0.08529, over 5660688.81 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:20:00,326 INFO [train.py:968] (1/2) Epoch 30, batch 33600, giga_loss[loss=0.2656, simple_loss=0.3445, pruned_loss=0.09339, over 28020.00 frames. ], tot_loss[loss=0.2542, simple_loss=0.3345, pruned_loss=0.08693, over 5649463.56 frames. ], libri_tot_loss[loss=0.2782, simple_loss=0.342, pruned_loss=0.1072, over 5665056.67 frames. ], giga_tot_loss[loss=0.2515, simple_loss=0.3334, pruned_loss=0.08475, over 5657713.84 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:20:12,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.4283, 4.2593, 4.0459, 2.0216], device='cuda:1'), covar=tensor([0.0577, 0.0721, 0.0835, 0.2335], device='cuda:1'), in_proj_covar=tensor([0.1308, 0.1203, 0.1010, 0.0742], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 15:20:19,845 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.62 vs. limit=5.0
+2023-03-15 15:20:22,171 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5300, 1.6094, 1.7100, 1.3305], device='cuda:1'), covar=tensor([0.1867, 0.2826, 0.1606, 0.1917], device='cuda:1'), in_proj_covar=tensor([0.0940, 0.0714, 0.0992, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 15:21:08,475 INFO [train.py:968] (1/2) Epoch 30, batch 33650, giga_loss[loss=0.2499, simple_loss=0.3303, pruned_loss=0.08471, over 28990.00 frames. ], tot_loss[loss=0.2532, simple_loss=0.333, pruned_loss=0.08668, over 5658191.35 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3414, pruned_loss=0.1069, over 5672252.45 frames. ], giga_tot_loss[loss=0.2508, simple_loss=0.3324, pruned_loss=0.08464, over 5657694.93 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:21:09,154 INFO [optim.py:369] (1/2) 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 15:21:43,297 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 33700, giga_loss[loss=0.2422, simple_loss=0.3213, pruned_loss=0.08158, over 28096.00 frames. ], tot_loss[loss=0.2527, simple_loss=0.3318, pruned_loss=0.08682, over 5657301.86 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3413, pruned_loss=0.1067, over 5672521.99 frames. ], giga_tot_loss[loss=0.2507, simple_loss=0.3313, pruned_loss=0.08502, over 5656495.51 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:22:33,478 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4734, 3.7107, 1.6724, 1.6611], device='cuda:1'), covar=tensor([0.1065, 0.0393, 0.0973, 0.1430], device='cuda:1'), in_proj_covar=tensor([0.0432, 0.0574, 0.0416, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 15:23:20,721 INFO [train.py:968] (1/2) Epoch 30, batch 33750, giga_loss[loss=0.2305, simple_loss=0.314, pruned_loss=0.07351, over 28944.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3309, pruned_loss=0.08618, over 5648736.82 frames. ], libri_tot_loss[loss=0.277, simple_loss=0.341, pruned_loss=0.1065, over 5669416.37 frames. ], giga_tot_loss[loss=0.2497, simple_loss=0.3306, pruned_loss=0.08437, over 5649534.58 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:23:21,303 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 33800, giga_loss[loss=0.2116, simple_loss=0.3003, pruned_loss=0.06144, over 28889.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3305, pruned_loss=0.08705, over 5651353.73 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3405, pruned_loss=0.1064, over 5667438.90 frames. ], giga_tot_loss[loss=0.2499, simple_loss=0.3302, pruned_loss=0.08474, over 5652703.54 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:24:45,077 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1354284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:24:50,238 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1354287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:25:28,388 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1354316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:25:29,664 INFO [train.py:968] (1/2) Epoch 30, batch 33850, giga_loss[loss=0.2432, simple_loss=0.3194, pruned_loss=0.08349, over 28952.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3286, pruned_loss=0.08692, over 5645106.45 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3401, pruned_loss=0.1062, over 5670707.90 frames. ], giga_tot_loss[loss=0.2494, simple_loss=0.3286, pruned_loss=0.08508, over 5642948.79 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:25:30,631 INFO [optim.py:369] (1/2) 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 15:26:06,749 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 15:26:20,812 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 15:26:23,523 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1354362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:26:29,697 INFO [train.py:968] (1/2) Epoch 30, batch 33900, giga_loss[loss=0.2353, simple_loss=0.32, pruned_loss=0.07528, over 28938.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.3276, pruned_loss=0.08511, over 5653216.39 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.34, pruned_loss=0.106, over 5676987.42 frames. ], giga_tot_loss[loss=0.247, simple_loss=0.3274, pruned_loss=0.0833, over 5644968.44 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:26:31,646 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-15 15:27:31,438 INFO [train.py:968] (1/2) Epoch 30, batch 33950, giga_loss[loss=0.2592, simple_loss=0.3414, pruned_loss=0.08857, over 27985.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.3271, pruned_loss=0.08373, over 5669151.52 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3396, pruned_loss=0.1058, over 5682916.93 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3269, pruned_loss=0.08187, over 5656915.31 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:27:33,126 INFO [optim.py:369] (1/2) 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 15:27:36,534 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3684, 1.5462, 1.3924, 1.5653], device='cuda:1'), covar=tensor([0.0741, 0.0430, 0.0359, 0.0814], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0123, 0.0122, 0.0234], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0118], device='cuda:1')
+2023-03-15 15:28:00,147 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 15:28:03,919 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7103, 0.8752, 0.8089, 0.6880], device='cuda:1'), covar=tensor([0.1548, 0.1788, 0.1432, 0.1640], device='cuda:1'), in_proj_covar=tensor([0.2062, 0.2021, 0.1920, 0.2077], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 15:28:04,801 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 15:28:32,876 INFO [train.py:968] (1/2) Epoch 30, batch 34000, giga_loss[loss=0.2131, simple_loss=0.3034, pruned_loss=0.06143, over 28478.00 frames. ], tot_loss[loss=0.2472, simple_loss=0.3293, pruned_loss=0.08253, over 5669254.49 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3396, pruned_loss=0.1058, over 5675380.98 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3291, pruned_loss=0.08083, over 5666258.27 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:29:18,923 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1354505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:29:23,239 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 34050, giga_loss[loss=0.2278, simple_loss=0.3172, pruned_loss=0.0692, over 28121.00 frames. ], tot_loss[loss=0.2474, simple_loss=0.3299, pruned_loss=0.08246, over 5656571.58 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3396, pruned_loss=0.106, over 5667785.07 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.3295, pruned_loss=0.08051, over 5660206.54 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:29:35,187 INFO [optim.py:369] (1/2) 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 15:29:56,061 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1354537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:30:09,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4693, 1.7832, 1.4228, 1.4599], device='cuda:1'), covar=tensor([0.2833, 0.2775, 0.3183, 0.2502], device='cuda:1'), in_proj_covar=tensor([0.1630, 0.1165, 0.1440, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 15:30:17,101 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2110, 1.3665, 3.6662, 3.1695], device='cuda:1'), covar=tensor([0.1787, 0.2767, 0.0511, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0683, 0.1014, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 15:30:42,071 INFO [train.py:968] (1/2) Epoch 30, batch 34100, giga_loss[loss=0.2564, simple_loss=0.3385, pruned_loss=0.08718, over 28384.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.3305, pruned_loss=0.0826, over 5659168.71 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3399, pruned_loss=0.1062, over 5668367.69 frames. ], giga_tot_loss[loss=0.2454, simple_loss=0.3298, pruned_loss=0.08046, over 5661339.50 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:31:52,646 INFO [train.py:968] (1/2) Epoch 30, batch 34150, giga_loss[loss=0.2728, simple_loss=0.3576, pruned_loss=0.09395, over 28537.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3308, pruned_loss=0.08258, over 5669382.80 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.34, pruned_loss=0.1062, over 5674177.49 frames. ], giga_tot_loss[loss=0.2453, simple_loss=0.33, pruned_loss=0.08037, over 5665888.79 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:31:54,003 INFO [optim.py:369] (1/2) 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 15:32:59,736 INFO [train.py:968] (1/2) Epoch 30, batch 34200, giga_loss[loss=0.2468, simple_loss=0.3331, pruned_loss=0.0802, over 29014.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3312, pruned_loss=0.08268, over 5673090.26 frames. ], libri_tot_loss[loss=0.2756, simple_loss=0.3396, pruned_loss=0.1058, over 5678959.99 frames. ], giga_tot_loss[loss=0.2461, simple_loss=0.3307, pruned_loss=0.08072, over 5666070.56 frames. ], batch size: 285, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:33:21,176 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5309, 1.7381, 1.4383, 1.4232], device='cuda:1'), covar=tensor([0.2886, 0.2842, 0.3405, 0.2453], device='cuda:1'), in_proj_covar=tensor([0.1629, 0.1165, 0.1441, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 15:33:31,358 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.5803, 1.3505, 4.4506, 3.5626], device='cuda:1'), covar=tensor([0.1842, 0.3057, 0.0665, 0.1253], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0684, 0.1016, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 15:33:36,320 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.20 vs. limit=5.0
+2023-03-15 15:34:09,885 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.48 vs. limit=5.0
+2023-03-15 15:34:16,540 INFO [train.py:968] (1/2) Epoch 30, batch 34250, giga_loss[loss=0.2399, simple_loss=0.3267, pruned_loss=0.07651, over 28953.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3312, pruned_loss=0.08242, over 5669148.06 frames. ], libri_tot_loss[loss=0.2755, simple_loss=0.3395, pruned_loss=0.1057, over 5682300.77 frames. ], giga_tot_loss[loss=0.2459, simple_loss=0.3307, pruned_loss=0.08054, over 5660422.97 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:34:19,685 INFO [optim.py:369] (1/2) 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 15:35:15,932 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4388, 1.7216, 1.4080, 1.2290], device='cuda:1'), covar=tensor([0.2371, 0.2379, 0.2725, 0.2389], device='cuda:1'), in_proj_covar=tensor([0.1628, 0.1164, 0.1440, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0012, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 15:35:21,564 INFO [train.py:968] (1/2) Epoch 30, batch 34300, giga_loss[loss=0.2916, simple_loss=0.3745, pruned_loss=0.1044, over 28711.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.3345, pruned_loss=0.08444, over 5662328.05 frames. ], libri_tot_loss[loss=0.2758, simple_loss=0.3398, pruned_loss=0.1059, over 5680198.63 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3336, pruned_loss=0.08204, over 5657038.92 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:36:27,951 INFO [train.py:968] (1/2) Epoch 30, batch 34350, giga_loss[loss=0.2346, simple_loss=0.3233, pruned_loss=0.07295, over 28952.00 frames. ], tot_loss[loss=0.2523, simple_loss=0.3358, pruned_loss=0.08436, over 5669298.48 frames. ], libri_tot_loss[loss=0.2757, simple_loss=0.3397, pruned_loss=0.1058, over 5675367.15 frames. ], giga_tot_loss[loss=0.2496, simple_loss=0.3351, pruned_loss=0.08207, over 5669365.46 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:36:31,533 INFO [optim.py:369] (1/2) 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 15:36:52,795 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7078, 1.9175, 1.9010, 1.6622], device='cuda:1'), covar=tensor([0.2346, 0.1941, 0.1667, 0.2008], device='cuda:1'), in_proj_covar=tensor([0.2052, 0.2009, 0.1908, 0.2063], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 15:37:05,259 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6022, 2.2201, 1.5596, 0.7732], device='cuda:1'), covar=tensor([0.6896, 0.3308, 0.4868, 0.8118], device='cuda:1'), in_proj_covar=tensor([0.1870, 0.1755, 0.1676, 0.1525], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 15:37:18,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.8737, 4.7039, 4.5171, 2.2095], device='cuda:1'), covar=tensor([0.0500, 0.0637, 0.0758, 0.2011], device='cuda:1'), in_proj_covar=tensor([0.1310, 0.1206, 0.1010, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 15:37:27,357 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1354860.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 15:37:36,886 INFO [train.py:968] (1/2) Epoch 30, batch 34400, giga_loss[loss=0.2148, simple_loss=0.3038, pruned_loss=0.06291, over 28933.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.3354, pruned_loss=0.08495, over 5672873.92 frames. ], libri_tot_loss[loss=0.2762, simple_loss=0.3403, pruned_loss=0.1061, over 5669311.85 frames. ], giga_tot_loss[loss=0.2493, simple_loss=0.3342, pruned_loss=0.08224, over 5678234.82 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:38:16,729 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1354896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:38:36,309 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4958, 1.7590, 1.7231, 1.4650], device='cuda:1'), covar=tensor([0.3580, 0.2465, 0.2089, 0.2868], device='cuda:1'), in_proj_covar=tensor([0.2047, 0.2003, 0.1904, 0.2060], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 15:38:52,188 INFO [train.py:968] (1/2) Epoch 30, batch 34450, giga_loss[loss=0.2188, simple_loss=0.3036, pruned_loss=0.06697, over 29083.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3334, pruned_loss=0.08438, over 5675260.69 frames. ], libri_tot_loss[loss=0.276, simple_loss=0.3401, pruned_loss=0.1059, over 5672676.23 frames. ], giga_tot_loss[loss=0.2483, simple_loss=0.3325, pruned_loss=0.08205, over 5676533.81 frames. ], batch size: 120, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:38:57,534 INFO [optim.py:369] (1/2) 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 15:39:18,593 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1354932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:40:05,114 INFO [train.py:968] (1/2) Epoch 30, batch 34500, giga_loss[loss=0.2167, simple_loss=0.3074, pruned_loss=0.06301, over 29043.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.3317, pruned_loss=0.08244, over 5666134.44 frames. ], libri_tot_loss[loss=0.2766, simple_loss=0.3405, pruned_loss=0.1063, over 5655958.11 frames. ], giga_tot_loss[loss=0.2451, simple_loss=0.3304, pruned_loss=0.07986, over 5684049.18 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:40:48,097 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2415, 2.4955, 1.2820, 1.3475], device='cuda:1'), covar=tensor([0.1023, 0.0389, 0.0993, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0575, 0.0417, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 15:41:06,499 INFO [train.py:968] (1/2) Epoch 30, batch 34550, libri_loss[loss=0.2576, simple_loss=0.3232, pruned_loss=0.09601, over 29543.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.331, pruned_loss=0.08283, over 5681584.69 frames. ], libri_tot_loss[loss=0.2768, simple_loss=0.3407, pruned_loss=0.1065, over 5665376.45 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3293, pruned_loss=0.07935, over 5688272.52 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:41:10,509 INFO [optim.py:369] (1/2) 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 15:42:04,033 INFO [train.py:968] (1/2) Epoch 30, batch 34600, libri_loss[loss=0.3026, simple_loss=0.3619, pruned_loss=0.1217, over 29388.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3318, pruned_loss=0.08386, over 5679563.11 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3412, pruned_loss=0.107, over 5672388.49 frames. ], giga_tot_loss[loss=0.2442, simple_loss=0.3295, pruned_loss=0.07946, over 5678582.96 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:43:04,939 INFO [train.py:968] (1/2) Epoch 30, batch 34650, giga_loss[loss=0.2381, simple_loss=0.3199, pruned_loss=0.07811, over 28892.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3345, pruned_loss=0.08559, over 5672527.78 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3412, pruned_loss=0.1071, over 5675261.78 frames. ], giga_tot_loss[loss=0.2479, simple_loss=0.3325, pruned_loss=0.08163, over 5669364.06 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:43:10,055 INFO [optim.py:369] (1/2) 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 15:43:32,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5436, 4.4038, 4.1407, 1.9602], device='cuda:1'), covar=tensor([0.0517, 0.0652, 0.0697, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.1312, 0.1206, 0.1011, 0.0745], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 15:44:04,372 INFO [train.py:968] (1/2) Epoch 30, batch 34700, giga_loss[loss=0.1993, simple_loss=0.2882, pruned_loss=0.05516, over 28755.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3321, pruned_loss=0.08513, over 5667134.18 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3413, pruned_loss=0.1072, over 5669330.98 frames. ], giga_tot_loss[loss=0.2465, simple_loss=0.3303, pruned_loss=0.08133, over 5669108.25 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:45:02,921 INFO [train.py:968] (1/2) Epoch 30, batch 34750, giga_loss[loss=0.2767, simple_loss=0.341, pruned_loss=0.1062, over 26768.00 frames. ], tot_loss[loss=0.2498, simple_loss=0.3303, pruned_loss=0.08464, over 5662038.07 frames. ], libri_tot_loss[loss=0.2776, simple_loss=0.3411, pruned_loss=0.107, over 5664562.05 frames. ], giga_tot_loss[loss=0.2457, simple_loss=0.3288, pruned_loss=0.08133, over 5668075.77 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 15:45:08,369 INFO [optim.py:369] (1/2) 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 15:45:25,224 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 34800, giga_loss[loss=0.2881, simple_loss=0.3717, pruned_loss=0.1022, over 28697.00 frames. ], tot_loss[loss=0.2511, simple_loss=0.3313, pruned_loss=0.08547, over 5661308.27 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.3409, pruned_loss=0.107, over 5667129.23 frames. ], giga_tot_loss[loss=0.2478, simple_loss=0.3302, pruned_loss=0.08267, over 5663614.27 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:46:09,086 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1355271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:46:43,482 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 34850, giga_loss[loss=0.2878, simple_loss=0.3624, pruned_loss=0.1066, over 27944.00 frames. ], tot_loss[loss=0.2598, simple_loss=0.3392, pruned_loss=0.09018, over 5659382.85 frames. ], libri_tot_loss[loss=0.2771, simple_loss=0.3406, pruned_loss=0.1068, over 5663504.09 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3384, pruned_loss=0.08771, over 5664288.79 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:46:55,383 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 34900, giga_loss[loss=0.2872, simple_loss=0.3494, pruned_loss=0.1125, over 23746.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3466, pruned_loss=0.09419, over 5652775.49 frames. ], libri_tot_loss[loss=0.2773, simple_loss=0.3408, pruned_loss=0.107, over 5650368.21 frames. ], giga_tot_loss[loss=0.2647, simple_loss=0.3459, pruned_loss=0.0917, over 5668708.62 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:47:42,787 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:47:51,435 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355378.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 15:47:54,614 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355381.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 15:47:59,983 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8703, 1.0767, 2.9653, 2.8246], device='cuda:1'), covar=tensor([0.1683, 0.2566, 0.0618, 0.1015], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0686, 0.1020, 0.0995], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 15:48:20,294 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355410.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 15:48:22,747 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:24,718 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:25,050 INFO [train.py:968] (1/2) Epoch 30, batch 34950, giga_loss[loss=0.2851, simple_loss=0.3591, pruned_loss=0.1056, over 29086.00 frames. ], tot_loss[loss=0.2682, simple_loss=0.3468, pruned_loss=0.09481, over 5668515.46 frames. ], libri_tot_loss[loss=0.2767, simple_loss=0.3403, pruned_loss=0.1066, over 5656655.90 frames. ], giga_tot_loss[loss=0.2663, simple_loss=0.3469, pruned_loss=0.09284, over 5675960.36 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:48:29,515 INFO [optim.py:369] (1/2) 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 15:48:49,869 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:54,073 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:57,266 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:48:59,234 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.16 vs. limit=5.0
+2023-03-15 15:49:08,544 INFO [train.py:968] (1/2) Epoch 30, batch 35000, giga_loss[loss=0.2461, simple_loss=0.3143, pruned_loss=0.08898, over 28865.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3437, pruned_loss=0.09435, over 5669803.33 frames. ], libri_tot_loss[loss=0.2777, simple_loss=0.3411, pruned_loss=0.1072, over 5657839.71 frames. ], giga_tot_loss[loss=0.2635, simple_loss=0.3432, pruned_loss=0.09188, over 5674842.27 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:49:21,858 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355482.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:49:54,888 INFO [train.py:968] (1/2) Epoch 30, batch 35050, giga_loss[loss=0.2693, simple_loss=0.3398, pruned_loss=0.09942, over 27940.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.3361, pruned_loss=0.09083, over 5668066.89 frames. ], libri_tot_loss[loss=0.2778, simple_loss=0.3413, pruned_loss=0.1071, over 5657008.90 frames. ], giga_tot_loss[loss=0.2565, simple_loss=0.3355, pruned_loss=0.08871, over 5672963.44 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:50:00,515 INFO [optim.py:369] (1/2) 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 15:50:38,999 INFO [train.py:968] (1/2) Epoch 30, batch 35100, giga_loss[loss=0.1925, simple_loss=0.2665, pruned_loss=0.05921, over 28427.00 frames. ], tot_loss[loss=0.2508, simple_loss=0.3276, pruned_loss=0.08697, over 5678873.31 frames. ], libri_tot_loss[loss=0.2775, simple_loss=0.341, pruned_loss=0.107, over 5660870.97 frames. ], giga_tot_loss[loss=0.2488, simple_loss=0.3272, pruned_loss=0.08515, over 5679524.95 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:51:10,674 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5223, 1.7259, 1.7445, 1.3428], device='cuda:1'), covar=tensor([0.1881, 0.2721, 0.1563, 0.1811], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0713, 0.0994, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 15:51:19,418 INFO [train.py:968] (1/2) Epoch 30, batch 35150, giga_loss[loss=0.2167, simple_loss=0.2897, pruned_loss=0.07185, over 28646.00 frames. ], tot_loss[loss=0.247, simple_loss=0.3227, pruned_loss=0.08565, over 5678333.55 frames. ], libri_tot_loss[loss=0.2781, simple_loss=0.3416, pruned_loss=0.1073, over 5659111.64 frames. ], giga_tot_loss[loss=0.244, simple_loss=0.3214, pruned_loss=0.08331, over 5681544.85 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:51:22,242 INFO [optim.py:369] (1/2) 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 15:51:35,705 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5848, 2.0295, 1.8134, 1.8375], device='cuda:1'), covar=tensor([0.2476, 0.2254, 0.2619, 0.2258], device='cuda:1'), in_proj_covar=tensor([0.0500, 0.0750, 0.0723, 0.0693], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 15:51:47,326 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:52:02,520 INFO [train.py:968] (1/2) Epoch 30, batch 35200, giga_loss[loss=0.2033, simple_loss=0.2816, pruned_loss=0.06253, over 28872.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.3178, pruned_loss=0.08373, over 5672327.14 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.3427, pruned_loss=0.1079, over 5655586.64 frames. ], giga_tot_loss[loss=0.2382, simple_loss=0.3152, pruned_loss=0.08065, over 5678883.76 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:52:45,577 INFO [train.py:968] (1/2) Epoch 30, batch 35250, giga_loss[loss=0.2239, simple_loss=0.3039, pruned_loss=0.07197, over 28831.00 frames. ], tot_loss[loss=0.2397, simple_loss=0.3147, pruned_loss=0.0824, over 5687988.88 frames. ], libri_tot_loss[loss=0.2788, simple_loss=0.3425, pruned_loss=0.1076, over 5660877.05 frames. ], giga_tot_loss[loss=0.2354, simple_loss=0.3119, pruned_loss=0.0794, over 5689399.53 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:52:48,194 INFO [optim.py:369] (1/2) 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 15:53:10,750 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1355744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:53:16,249 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:53:31,992 INFO [train.py:968] (1/2) Epoch 30, batch 35300, giga_loss[loss=0.2185, simple_loss=0.2922, pruned_loss=0.07238, over 28457.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.3108, pruned_loss=0.08071, over 5688444.96 frames. ], libri_tot_loss[loss=0.2789, simple_loss=0.3425, pruned_loss=0.1076, over 5663854.83 frames. ], giga_tot_loss[loss=0.2318, simple_loss=0.3081, pruned_loss=0.07782, over 5687461.81 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:54:13,373 INFO [train.py:968] (1/2) Epoch 30, batch 35350, libri_loss[loss=0.2168, simple_loss=0.296, pruned_loss=0.06876, over 29582.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.3088, pruned_loss=0.07979, over 5678603.98 frames. ], libri_tot_loss[loss=0.2792, simple_loss=0.343, pruned_loss=0.1077, over 5658915.87 frames. ], giga_tot_loss[loss=0.2291, simple_loss=0.3052, pruned_loss=0.07656, over 5683463.54 frames. ], batch size: 75, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:54:14,989 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6335, 2.2875, 1.7361, 0.9227], device='cuda:1'), covar=tensor([0.7592, 0.3662, 0.4837, 0.7939], device='cuda:1'), in_proj_covar=tensor([0.1865, 0.1753, 0.1669, 0.1518], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 15:54:17,085 INFO [optim.py:369] (1/2) 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 15:54:23,867 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:54:57,975 INFO [train.py:968] (1/2) Epoch 30, batch 35400, giga_loss[loss=0.2114, simple_loss=0.2846, pruned_loss=0.06914, over 28956.00 frames. ], tot_loss[loss=0.2314, simple_loss=0.3058, pruned_loss=0.07849, over 5665641.01 frames. ], libri_tot_loss[loss=0.2793, simple_loss=0.3432, pruned_loss=0.1077, over 5654122.83 frames. ], giga_tot_loss[loss=0.226, simple_loss=0.3017, pruned_loss=0.07518, over 5675076.31 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:55:15,200 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1355887.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:55:17,420 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1355890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:55:26,090 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5846, 1.9250, 1.5643, 1.5033], device='cuda:1'), covar=tensor([0.2703, 0.2784, 0.3211, 0.2570], device='cuda:1'), in_proj_covar=tensor([0.1635, 0.1169, 0.1444, 0.1022], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 15:55:40,452 INFO [train.py:968] (1/2) Epoch 30, batch 35450, giga_loss[loss=0.1971, simple_loss=0.2759, pruned_loss=0.05921, over 28787.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.3037, pruned_loss=0.07745, over 5677411.01 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3437, pruned_loss=0.1077, over 5660923.64 frames. ], giga_tot_loss[loss=0.2234, simple_loss=0.2989, pruned_loss=0.07394, over 5679452.08 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:55:41,468 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1355919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:55:43,063 INFO [optim.py:369] (1/2) 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 15:56:21,212 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1355962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:56:25,797 INFO [train.py:968] (1/2) Epoch 30, batch 35500, giga_loss[loss=0.2068, simple_loss=0.2844, pruned_loss=0.06458, over 28622.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.3005, pruned_loss=0.07584, over 5676270.48 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3439, pruned_loss=0.1078, over 5653494.45 frames. ], giga_tot_loss[loss=0.2207, simple_loss=0.2961, pruned_loss=0.07264, over 5685406.42 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:57:04,686 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4520, 3.0205, 1.5307, 1.5796], device='cuda:1'), covar=tensor([0.0998, 0.0362, 0.0932, 0.1304], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0576, 0.0418, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 15:57:11,979 INFO [train.py:968] (1/2) Epoch 30, batch 35550, libri_loss[loss=0.3387, simple_loss=0.4075, pruned_loss=0.135, over 29479.00 frames. ], tot_loss[loss=0.2245, simple_loss=0.2987, pruned_loss=0.07518, over 5677745.17 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3444, pruned_loss=0.1079, over 5658080.64 frames. ], giga_tot_loss[loss=0.2188, simple_loss=0.2937, pruned_loss=0.07189, over 5681078.83 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 15:57:12,339 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7638, 2.0691, 1.6772, 1.7916], device='cuda:1'), covar=tensor([0.2816, 0.2841, 0.3421, 0.2693], device='cuda:1'), in_proj_covar=tensor([0.1637, 0.1171, 0.1446, 0.1023], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0010], device='cuda:1')
+2023-03-15 15:57:15,302 INFO [optim.py:369] (1/2) 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 15:57:17,500 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:57:24,204 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4548, 1.3505, 4.0596, 3.2106], device='cuda:1'), covar=tensor([0.1686, 0.2907, 0.0448, 0.0992], device='cuda:1'), in_proj_covar=tensor([0.0820, 0.0683, 0.1019, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 15:57:58,123 INFO [train.py:968] (1/2) Epoch 30, batch 35600, giga_loss[loss=0.2249, simple_loss=0.2941, pruned_loss=0.07789, over 28668.00 frames. ], tot_loss[loss=0.2218, simple_loss=0.2957, pruned_loss=0.07401, over 5676815.24 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3441, pruned_loss=0.1076, over 5662505.57 frames. ], giga_tot_loss[loss=0.2163, simple_loss=0.2909, pruned_loss=0.07085, over 5675895.08 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:58:45,860 INFO [train.py:968] (1/2) Epoch 30, batch 35650, giga_loss[loss=0.2433, simple_loss=0.3229, pruned_loss=0.08186, over 29055.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.2968, pruned_loss=0.07505, over 5676449.25 frames. ], libri_tot_loss[loss=0.2796, simple_loss=0.3441, pruned_loss=0.1076, over 5667379.81 frames. ], giga_tot_loss[loss=0.218, simple_loss=0.2921, pruned_loss=0.07197, over 5671732.64 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:58:51,213 INFO [optim.py:369] (1/2) 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 15:58:53,274 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:59:34,694 INFO [train.py:968] (1/2) Epoch 30, batch 35700, giga_loss[loss=0.2594, simple_loss=0.3401, pruned_loss=0.08936, over 28340.00 frames. ], tot_loss[loss=0.2344, simple_loss=0.3082, pruned_loss=0.08031, over 5687696.18 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3445, pruned_loss=0.1078, over 5669739.14 frames. ], giga_tot_loss[loss=0.2294, simple_loss=0.3039, pruned_loss=0.07745, over 5681980.30 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 15:59:35,871 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 15:59:39,069 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:00:08,669 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1356200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:00:13,361 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:00:24,038 INFO [train.py:968] (1/2) Epoch 30, batch 35750, giga_loss[loss=0.2715, simple_loss=0.355, pruned_loss=0.09395, over 28407.00 frames. ], tot_loss[loss=0.248, simple_loss=0.3216, pruned_loss=0.08721, over 5690144.36 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3443, pruned_loss=0.1075, over 5674641.35 frames. ], giga_tot_loss[loss=0.2436, simple_loss=0.3178, pruned_loss=0.08474, over 5681310.28 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:00:30,787 INFO [optim.py:369] (1/2) 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 16:00:31,179 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 16:01:12,521 INFO [train.py:968] (1/2) Epoch 30, batch 35800, giga_loss[loss=0.3037, simple_loss=0.3613, pruned_loss=0.123, over 23493.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3314, pruned_loss=0.09211, over 5684260.51 frames. ], libri_tot_loss[loss=0.2798, simple_loss=0.3444, pruned_loss=0.1076, over 5676930.86 frames. ], giga_tot_loss[loss=0.2541, simple_loss=0.3282, pruned_loss=0.08993, over 5675269.76 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:01:12,790 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:01:14,260 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.61 vs. limit=5.0
+2023-03-15 16:01:14,855 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:01:28,798 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1712, 1.2971, 3.3103, 2.7828], device='cuda:1'), covar=tensor([0.1671, 0.2812, 0.0535, 0.1249], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0681, 0.1016, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 16:01:39,810 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 35850, libri_loss[loss=0.2271, simple_loss=0.2964, pruned_loss=0.0789, over 29655.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.337, pruned_loss=0.094, over 5693321.07 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3447, pruned_loss=0.1078, over 5684292.11 frames. ], giga_tot_loss[loss=0.2586, simple_loss=0.3339, pruned_loss=0.09162, over 5679693.58 frames. ], batch size: 69, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:01:59,718 INFO [optim.py:369] (1/2) 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 16:02:02,287 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.53 vs. limit=2.0
+2023-03-15 16:02:09,705 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:02:19,075 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:02:20,597 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.5390, 3.3742, 3.1740, 1.7301], device='cuda:1'), covar=tensor([0.0767, 0.0894, 0.0774, 0.2424], device='cuda:1'), in_proj_covar=tensor([0.1305, 0.1207, 0.1009, 0.0747], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 16:02:21,313 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:02:34,657 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2870, 1.8462, 1.4292, 0.5693], device='cuda:1'), covar=tensor([0.7277, 0.3767, 0.5001, 0.7811], device='cuda:1'), in_proj_covar=tensor([0.1869, 0.1759, 0.1675, 0.1519], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:02:36,434 INFO [train.py:968] (1/2) Epoch 30, batch 35900, giga_loss[loss=0.2905, simple_loss=0.3692, pruned_loss=0.1059, over 28624.00 frames. ], tot_loss[loss=0.2631, simple_loss=0.3387, pruned_loss=0.09377, over 5689374.78 frames. ], libri_tot_loss[loss=0.2797, simple_loss=0.3446, pruned_loss=0.1074, over 5692488.05 frames. ], giga_tot_loss[loss=0.2596, simple_loss=0.336, pruned_loss=0.09164, over 5671416.24 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:02:49,604 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1356381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:03:21,459 INFO [train.py:968] (1/2) Epoch 30, batch 35950, giga_loss[loss=0.2422, simple_loss=0.327, pruned_loss=0.07866, over 29025.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3404, pruned_loss=0.09432, over 5688896.05 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.3449, pruned_loss=0.1074, over 5700879.74 frames. ], giga_tot_loss[loss=0.2609, simple_loss=0.3376, pruned_loss=0.09207, over 5666585.50 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:03:25,407 INFO [optim.py:369] (1/2) 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 16:03:49,963 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1356452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:03:54,609 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1694, 1.2006, 3.5341, 3.0037], device='cuda:1'), covar=tensor([0.1735, 0.3030, 0.0458, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0680, 0.1017, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 16:04:07,339 INFO [train.py:968] (1/2) Epoch 30, batch 36000, giga_loss[loss=0.2877, simple_loss=0.3613, pruned_loss=0.1071, over 28815.00 frames. ], tot_loss[loss=0.2665, simple_loss=0.3423, pruned_loss=0.09539, over 5696590.36 frames. ], libri_tot_loss[loss=0.2799, simple_loss=0.345, pruned_loss=0.1074, over 5702728.20 frames. ], giga_tot_loss[loss=0.2634, simple_loss=0.34, pruned_loss=0.0934, over 5677199.18 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:04:07,339 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 16:04:12,731 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1674, 1.3342, 3.3491, 3.0481], device='cuda:1'), covar=tensor([0.1806, 0.3019, 0.0538, 0.1119], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0680, 0.1016, 0.0991], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 16:04:16,004 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 16:04:24,690 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:04:27,553 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:04:39,112 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5161, 2.1721, 1.7617, 0.7394], device='cuda:1'), covar=tensor([0.7831, 0.4037, 0.5018, 0.7997], device='cuda:1'), in_proj_covar=tensor([0.1864, 0.1754, 0.1672, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:04:54,218 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1356512.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:04:57,844 INFO [train.py:968] (1/2) Epoch 30, batch 36050, giga_loss[loss=0.2718, simple_loss=0.3466, pruned_loss=0.09851, over 28960.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.3458, pruned_loss=0.09795, over 5693257.94 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3451, pruned_loss=0.1075, over 5704250.38 frames. ], giga_tot_loss[loss=0.268, simple_loss=0.3438, pruned_loss=0.09609, over 5676199.47 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:05:03,678 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 36100, giga_loss[loss=0.2717, simple_loss=0.3593, pruned_loss=0.09203, over 28333.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3494, pruned_loss=0.09987, over 5686086.05 frames. ], libri_tot_loss[loss=0.28, simple_loss=0.3451, pruned_loss=0.1074, over 5693439.24 frames. ], giga_tot_loss[loss=0.2722, simple_loss=0.3479, pruned_loss=0.09822, over 5682276.46 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:05:52,697 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4286, 2.0586, 1.5078, 0.6473], device='cuda:1'), covar=tensor([0.8022, 0.3889, 0.5026, 0.8085], device='cuda:1'), in_proj_covar=tensor([0.1870, 0.1759, 0.1676, 0.1519], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:06:03,842 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3910, 1.4718, 1.5053, 1.2998], device='cuda:1'), covar=tensor([0.2971, 0.3534, 0.2479, 0.3048], device='cuda:1'), in_proj_covar=tensor([0.2063, 0.2019, 0.1925, 0.2082], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 16:06:21,183 INFO [train.py:968] (1/2) Epoch 30, batch 36150, giga_loss[loss=0.2601, simple_loss=0.3429, pruned_loss=0.08862, over 27982.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3525, pruned_loss=0.1003, over 5693027.28 frames. ], libri_tot_loss[loss=0.2808, simple_loss=0.3458, pruned_loss=0.1079, over 5686746.40 frames. ], giga_tot_loss[loss=0.2738, simple_loss=0.3507, pruned_loss=0.09849, over 5696681.01 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:06:30,374 INFO [optim.py:369] (1/2) 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 16:07:02,969 INFO [train.py:968] (1/2) Epoch 30, batch 36200, giga_loss[loss=0.2615, simple_loss=0.3469, pruned_loss=0.08806, over 28923.00 frames. ], tot_loss[loss=0.2768, simple_loss=0.3532, pruned_loss=0.1002, over 5694977.82 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3455, pruned_loss=0.1073, over 5694397.67 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3522, pruned_loss=0.09894, over 5690866.99 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:07:44,371 INFO [train.py:968] (1/2) Epoch 30, batch 36250, libri_loss[loss=0.2939, simple_loss=0.3684, pruned_loss=0.1097, over 27775.00 frames. ], tot_loss[loss=0.2766, simple_loss=0.3538, pruned_loss=0.09968, over 5699305.93 frames. ], libri_tot_loss[loss=0.2801, simple_loss=0.3458, pruned_loss=0.1072, over 5698531.70 frames. ], giga_tot_loss[loss=0.2751, simple_loss=0.3531, pruned_loss=0.09851, over 5692383.22 frames. ], batch size: 116, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:07:52,552 INFO [optim.py:369] (1/2) 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 16:08:25,581 INFO [train.py:968] (1/2) Epoch 30, batch 36300, giga_loss[loss=0.2492, simple_loss=0.3405, pruned_loss=0.07895, over 28755.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3539, pruned_loss=0.0989, over 5702730.28 frames. ], libri_tot_loss[loss=0.2804, simple_loss=0.3461, pruned_loss=0.1074, over 5701694.21 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3532, pruned_loss=0.09776, over 5694559.05 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:08:41,484 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1356786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:09:08,088 INFO [train.py:968] (1/2) Epoch 30, batch 36350, giga_loss[loss=0.261, simple_loss=0.3428, pruned_loss=0.08954, over 28672.00 frames. ], tot_loss[loss=0.2729, simple_loss=0.3519, pruned_loss=0.09689, over 5701164.09 frames. ], libri_tot_loss[loss=0.2806, simple_loss=0.3462, pruned_loss=0.1075, over 5701450.84 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.3514, pruned_loss=0.09574, over 5694760.26 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 1.0
+2023-03-15 16:09:14,393 INFO [optim.py:369] (1/2) 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 16:09:14,708 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1356827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:09:22,521 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.90 vs. limit=5.0
+2023-03-15 16:09:49,463 INFO [train.py:968] (1/2) Epoch 30, batch 36400, libri_loss[loss=0.2846, simple_loss=0.3515, pruned_loss=0.1089, over 28558.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3524, pruned_loss=0.09773, over 5696817.10 frames. ], libri_tot_loss[loss=0.2811, simple_loss=0.3469, pruned_loss=0.1077, over 5708010.09 frames. ], giga_tot_loss[loss=0.272, simple_loss=0.3516, pruned_loss=0.09621, over 5685945.72 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:10:36,614 INFO [train.py:968] (1/2) Epoch 30, batch 36450, giga_loss[loss=0.2658, simple_loss=0.3536, pruned_loss=0.08899, over 28144.00 frames. ], tot_loss[loss=0.2786, simple_loss=0.3549, pruned_loss=0.1012, over 5682754.29 frames. ], libri_tot_loss[loss=0.2817, simple_loss=0.3474, pruned_loss=0.108, over 5699673.06 frames. ], giga_tot_loss[loss=0.2765, simple_loss=0.3539, pruned_loss=0.09951, over 5680990.64 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:10:44,675 INFO [optim.py:369] (1/2) 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 16:11:20,353 INFO [train.py:968] (1/2) Epoch 30, batch 36500, libri_loss[loss=0.3457, simple_loss=0.4085, pruned_loss=0.1414, over 25838.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3569, pruned_loss=0.1045, over 5688963.06 frames. ], libri_tot_loss[loss=0.2819, simple_loss=0.3476, pruned_loss=0.1081, over 5702204.35 frames. ], giga_tot_loss[loss=0.281, simple_loss=0.3561, pruned_loss=0.1029, over 5685283.94 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:11:22,775 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1356970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:11:24,632 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1356973.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:11:29,602 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8647, 2.2018, 1.7259, 2.1372], device='cuda:1'), covar=tensor([0.2549, 0.2600, 0.2980, 0.2444], device='cuda:1'), in_proj_covar=tensor([0.1635, 0.1171, 0.1442, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 16:11:52,628 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:12:08,286 INFO [train.py:968] (1/2) Epoch 30, batch 36550, libri_loss[loss=0.4104, simple_loss=0.4297, pruned_loss=0.1956, over 19837.00 frames. ], tot_loss[loss=0.2846, simple_loss=0.357, pruned_loss=0.1061, over 5676337.96 frames. ], libri_tot_loss[loss=0.2824, simple_loss=0.3479, pruned_loss=0.1084, over 5693530.33 frames. ], giga_tot_loss[loss=0.2826, simple_loss=0.3561, pruned_loss=0.1046, over 5682047.70 frames. ], batch size: 188, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:12:15,151 INFO [optim.py:369] (1/2) 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 16:12:32,215 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3671, 1.6056, 1.2414, 1.0597], device='cuda:1'), covar=tensor([0.1140, 0.0538, 0.1072, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0451, 0.0528, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 16:12:51,193 INFO [train.py:968] (1/2) Epoch 30, batch 36600, giga_loss[loss=0.2473, simple_loss=0.327, pruned_loss=0.08378, over 28881.00 frames. ], tot_loss[loss=0.2829, simple_loss=0.3549, pruned_loss=0.1054, over 5689701.97 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3483, pruned_loss=0.1086, over 5696550.90 frames. ], giga_tot_loss[loss=0.2809, simple_loss=0.354, pruned_loss=0.1039, over 5691238.18 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:13:37,668 INFO [train.py:968] (1/2) Epoch 30, batch 36650, giga_loss[loss=0.2824, simple_loss=0.3531, pruned_loss=0.1059, over 28942.00 frames. ], tot_loss[loss=0.281, simple_loss=0.353, pruned_loss=0.1045, over 5690854.60 frames. ], libri_tot_loss[loss=0.283, simple_loss=0.3486, pruned_loss=0.1087, over 5694589.58 frames. ], giga_tot_loss[loss=0.2793, simple_loss=0.3522, pruned_loss=0.1032, over 5693658.30 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:13:43,547 INFO [optim.py:369] (1/2) 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 16:14:13,226 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5847, 1.8732, 1.5464, 1.5217], device='cuda:1'), covar=tensor([0.2665, 0.2728, 0.3129, 0.2465], device='cuda:1'), in_proj_covar=tensor([0.1632, 0.1168, 0.1440, 0.1019], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 16:14:13,725 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1357160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:14:14,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357161.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:14:20,916 INFO [train.py:968] (1/2) Epoch 30, batch 36700, libri_loss[loss=0.246, simple_loss=0.3194, pruned_loss=0.08628, over 29365.00 frames. ], tot_loss[loss=0.2789, simple_loss=0.3515, pruned_loss=0.1031, over 5696974.53 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3485, pruned_loss=0.1086, over 5701957.09 frames. ], giga_tot_loss[loss=0.2775, simple_loss=0.351, pruned_loss=0.102, over 5692095.78 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:15:01,321 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4109, 2.1457, 1.6641, 0.6972], device='cuda:1'), covar=tensor([0.7481, 0.3628, 0.4967, 0.7721], device='cuda:1'), in_proj_covar=tensor([0.1869, 0.1757, 0.1673, 0.1517], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:15:03,824 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.68 vs. limit=5.0
+2023-03-15 16:15:04,604 INFO [train.py:968] (1/2) Epoch 30, batch 36750, giga_loss[loss=0.2802, simple_loss=0.3565, pruned_loss=0.1019, over 28549.00 frames. ], tot_loss[loss=0.2758, simple_loss=0.3496, pruned_loss=0.101, over 5705300.19 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3487, pruned_loss=0.1084, over 5706572.13 frames. ], giga_tot_loss[loss=0.2746, simple_loss=0.3491, pruned_loss=0.1001, over 5697321.86 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:15:14,487 INFO [optim.py:369] (1/2) 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 16:15:52,406 INFO [train.py:968] (1/2) Epoch 30, batch 36800, giga_loss[loss=0.2311, simple_loss=0.3145, pruned_loss=0.07388, over 28848.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3447, pruned_loss=0.098, over 5700492.06 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3491, pruned_loss=0.1086, over 5709468.85 frames. ], giga_tot_loss[loss=0.2688, simple_loss=0.3438, pruned_loss=0.09686, over 5691317.71 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:16:27,467 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:16:30,138 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1357307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:16:41,827 INFO [train.py:968] (1/2) Epoch 30, batch 36850, giga_loss[loss=0.2108, simple_loss=0.2988, pruned_loss=0.0614, over 28854.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3388, pruned_loss=0.09481, over 5704945.47 frames. ], libri_tot_loss[loss=0.2833, simple_loss=0.3492, pruned_loss=0.1087, over 5715558.35 frames. ], giga_tot_loss[loss=0.2624, simple_loss=0.3378, pruned_loss=0.0935, over 5692016.60 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:16:52,871 INFO [optim.py:369] (1/2) 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 16:16:54,495 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1357328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:17:04,673 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:17:19,696 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.09 vs. limit=2.0
+2023-03-15 16:17:38,840 INFO [train.py:968] (1/2) Epoch 30, batch 36900, giga_loss[loss=0.2322, simple_loss=0.29, pruned_loss=0.08722, over 23339.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3326, pruned_loss=0.09162, over 5686147.22 frames. ], libri_tot_loss[loss=0.2827, simple_loss=0.3489, pruned_loss=0.1083, over 5716927.48 frames. ], giga_tot_loss[loss=0.2566, simple_loss=0.3318, pruned_loss=0.09068, over 5674195.43 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:18:23,676 INFO [train.py:968] (1/2) Epoch 30, batch 36950, giga_loss[loss=0.2458, simple_loss=0.3249, pruned_loss=0.08342, over 29133.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.3341, pruned_loss=0.09203, over 5686267.36 frames. ], libri_tot_loss[loss=0.2829, simple_loss=0.3492, pruned_loss=0.1083, over 5721026.24 frames. ], giga_tot_loss[loss=0.2573, simple_loss=0.3328, pruned_loss=0.09088, over 5672397.33 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:18:31,580 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:19:00,392 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6384, 1.6733, 1.8431, 1.4063], device='cuda:1'), covar=tensor([0.1931, 0.2654, 0.1611, 0.1829], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0717, 0.0996, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:19:05,379 INFO [train.py:968] (1/2) Epoch 30, batch 37000, giga_loss[loss=0.237, simple_loss=0.3243, pruned_loss=0.07492, over 28737.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.334, pruned_loss=0.09143, over 5688311.25 frames. ], libri_tot_loss[loss=0.2828, simple_loss=0.3491, pruned_loss=0.1082, over 5714682.32 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3329, pruned_loss=0.0904, over 5683588.49 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:19:47,247 INFO [train.py:968] (1/2) Epoch 30, batch 37050, giga_loss[loss=0.2344, simple_loss=0.3167, pruned_loss=0.07608, over 28816.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3341, pruned_loss=0.09165, over 5694600.99 frames. ], libri_tot_loss[loss=0.2832, simple_loss=0.3496, pruned_loss=0.1083, over 5716940.69 frames. ], giga_tot_loss[loss=0.2561, simple_loss=0.3321, pruned_loss=0.09005, over 5687501.91 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:19:54,723 INFO [optim.py:369] (1/2) 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 16:20:02,356 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:20:04,502 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5569, 1.6509, 1.6040, 1.4287], device='cuda:1'), covar=tensor([0.3611, 0.3311, 0.2353, 0.3370], device='cuda:1'), in_proj_covar=tensor([0.2063, 0.2016, 0.1925, 0.2080], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 16:20:25,673 INFO [train.py:968] (1/2) Epoch 30, batch 37100, giga_loss[loss=0.2686, simple_loss=0.3276, pruned_loss=0.1048, over 28863.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.3334, pruned_loss=0.0917, over 5700793.82 frames. ], libri_tot_loss[loss=0.2835, simple_loss=0.3502, pruned_loss=0.1084, over 5723803.55 frames. ], giga_tot_loss[loss=0.2553, simple_loss=0.3309, pruned_loss=0.08983, over 5688101.08 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:20:28,560 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.1885, 3.3820, 2.1675, 1.4179], device='cuda:1'), covar=tensor([0.8581, 0.3013, 0.4912, 0.7367], device='cuda:1'), in_proj_covar=tensor([0.1865, 0.1750, 0.1671, 0.1513], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:21:06,988 INFO [train.py:968] (1/2) Epoch 30, batch 37150, libri_loss[loss=0.2807, simple_loss=0.3485, pruned_loss=0.1065, over 29363.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3315, pruned_loss=0.09065, over 5708535.13 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3506, pruned_loss=0.1084, over 5726037.48 frames. ], giga_tot_loss[loss=0.2526, simple_loss=0.3283, pruned_loss=0.08847, over 5695643.80 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:21:12,812 INFO [optim.py:369] (1/2) 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 16:21:24,461 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6226, 1.7482, 1.8259, 1.3906], device='cuda:1'), covar=tensor([0.1926, 0.2663, 0.1585, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0718, 0.0997, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:21:46,819 INFO [train.py:968] (1/2) Epoch 30, batch 37200, giga_loss[loss=0.23, simple_loss=0.3037, pruned_loss=0.07821, over 28970.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.3288, pruned_loss=0.08941, over 5715599.49 frames. ], libri_tot_loss[loss=0.2837, simple_loss=0.3507, pruned_loss=0.1083, over 5728712.73 frames. ], giga_tot_loss[loss=0.2505, simple_loss=0.3259, pruned_loss=0.08753, over 5702880.11 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:21:54,820 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:21:57,725 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1357681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:22:13,561 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:22:18,647 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:22:25,019 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9646, 1.2024, 0.9715, 0.2559], device='cuda:1'), covar=tensor([0.3580, 0.2442, 0.3462, 0.6955], device='cuda:1'), in_proj_covar=tensor([0.1860, 0.1744, 0.1665, 0.1508], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:22:25,919 INFO [train.py:968] (1/2) Epoch 30, batch 37250, giga_loss[loss=0.239, simple_loss=0.3154, pruned_loss=0.08129, over 28985.00 frames. ], tot_loss[loss=0.2531, simple_loss=0.3278, pruned_loss=0.08918, over 5720882.72 frames. ], libri_tot_loss[loss=0.2847, simple_loss=0.3518, pruned_loss=0.1088, over 5731921.55 frames. ], giga_tot_loss[loss=0.2485, simple_loss=0.3238, pruned_loss=0.0866, over 5707508.04 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:22:34,385 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2592, 1.4751, 1.5212, 1.1389], device='cuda:1'), covar=tensor([0.1842, 0.2840, 0.1515, 0.1814], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0717, 0.0996, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:22:35,363 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 37300, giga_loss[loss=0.2071, simple_loss=0.2893, pruned_loss=0.06241, over 28776.00 frames. ], tot_loss[loss=0.2516, simple_loss=0.3262, pruned_loss=0.08847, over 5719395.15 frames. ], libri_tot_loss[loss=0.2851, simple_loss=0.3522, pruned_loss=0.109, over 5734595.46 frames. ], giga_tot_loss[loss=0.2468, simple_loss=0.322, pruned_loss=0.08577, over 5706065.42 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:23:12,012 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.73 vs. limit=2.0
+2023-03-15 16:23:25,504 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5002, 4.3247, 4.1719, 2.0559], device='cuda:1'), covar=tensor([0.0711, 0.0881, 0.0987, 0.1912], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.1207, 0.1011, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 16:23:34,008 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1357801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:23:41,585 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1357809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:23:46,342 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2704, 4.0979, 3.9378, 1.8101], device='cuda:1'), covar=tensor([0.0705, 0.0865, 0.0964, 0.2062], device='cuda:1'), in_proj_covar=tensor([0.1308, 0.1205, 0.1010, 0.0752], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 16:23:47,434 INFO [train.py:968] (1/2) Epoch 30, batch 37350, giga_loss[loss=0.2297, simple_loss=0.296, pruned_loss=0.08173, over 28771.00 frames. ], tot_loss[loss=0.2497, simple_loss=0.3243, pruned_loss=0.08758, over 5722384.93 frames. ], libri_tot_loss[loss=0.2857, simple_loss=0.3529, pruned_loss=0.1092, over 5737038.08 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3197, pruned_loss=0.08471, over 5709140.34 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:23:55,502 INFO [optim.py:369] (1/2) 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 16:23:55,834 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7720, 2.5350, 1.6905, 1.0462], device='cuda:1'), covar=tensor([0.9751, 0.4595, 0.4827, 0.8325], device='cuda:1'), in_proj_covar=tensor([0.1863, 0.1748, 0.1668, 0.1511], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:24:10,186 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:24:13,551 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1357849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:24:16,083 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6493, 1.4576, 4.7557, 3.4913], device='cuda:1'), covar=tensor([0.1760, 0.2917, 0.0392, 0.0951], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0679, 0.1014, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 16:24:27,554 INFO [train.py:968] (1/2) Epoch 30, batch 37400, giga_loss[loss=0.2658, simple_loss=0.3354, pruned_loss=0.09813, over 28583.00 frames. ], tot_loss[loss=0.2463, simple_loss=0.3213, pruned_loss=0.08565, over 5719543.48 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3534, pruned_loss=0.1094, over 5729532.49 frames. ], giga_tot_loss[loss=0.2414, simple_loss=0.3168, pruned_loss=0.08295, over 5715587.39 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:24:35,747 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:25:10,225 INFO [train.py:968] (1/2) Epoch 30, batch 37450, giga_loss[loss=0.2419, simple_loss=0.3166, pruned_loss=0.08358, over 28178.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.3192, pruned_loss=0.08445, over 5725942.42 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3535, pruned_loss=0.1093, over 5731407.90 frames. ], giga_tot_loss[loss=0.2399, simple_loss=0.3154, pruned_loss=0.08219, over 5721170.17 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:25:17,859 INFO [optim.py:369] (1/2) 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 16:25:24,178 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.76 vs. limit=2.0
+2023-03-15 16:25:34,284 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1357944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:25:37,073 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 37500, giga_loss[loss=0.2188, simple_loss=0.2932, pruned_loss=0.07222, over 28536.00 frames. ], tot_loss[loss=0.2432, simple_loss=0.3184, pruned_loss=0.08402, over 5726003.84 frames. ], libri_tot_loss[loss=0.2858, simple_loss=0.3536, pruned_loss=0.109, over 5733750.62 frames. ], giga_tot_loss[loss=0.2395, simple_loss=0.3148, pruned_loss=0.08215, over 5720206.53 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:26:02,167 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1357976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:26:38,505 INFO [train.py:968] (1/2) Epoch 30, batch 37550, giga_loss[loss=0.2453, simple_loss=0.3124, pruned_loss=0.08911, over 28530.00 frames. ], tot_loss[loss=0.2482, simple_loss=0.3228, pruned_loss=0.08682, over 5708644.77 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3538, pruned_loss=0.1091, over 5723582.82 frames. ], giga_tot_loss[loss=0.2445, simple_loss=0.3192, pruned_loss=0.08492, over 5712797.96 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:26:42,545 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6749, 4.6097, 1.7808, 2.0145], device='cuda:1'), covar=tensor([0.1038, 0.0274, 0.0900, 0.1248], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0574, 0.0418, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 16:26:47,173 INFO [optim.py:369] (1/2) 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 16:27:23,400 INFO [train.py:968] (1/2) Epoch 30, batch 37600, giga_loss[loss=0.3334, simple_loss=0.3945, pruned_loss=0.1361, over 28574.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.3291, pruned_loss=0.09077, over 5694563.39 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3546, pruned_loss=0.1096, over 5715965.28 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.325, pruned_loss=0.08846, over 5705170.34 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:27:31,537 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.2772, 3.1329, 2.9691, 1.5224], device='cuda:1'), covar=tensor([0.0934, 0.1015, 0.0869, 0.2171], device='cuda:1'), in_proj_covar=tensor([0.1309, 0.1209, 0.1013, 0.0753], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 16:28:11,360 INFO [train.py:968] (1/2) Epoch 30, batch 37650, giga_loss[loss=0.2824, simple_loss=0.3543, pruned_loss=0.1053, over 28909.00 frames. ], tot_loss[loss=0.2659, simple_loss=0.3381, pruned_loss=0.09686, over 5693970.36 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3547, pruned_loss=0.1094, over 5719177.63 frames. ], giga_tot_loss[loss=0.2617, simple_loss=0.3341, pruned_loss=0.09463, over 5698708.91 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:28:22,347 INFO [optim.py:369] (1/2) 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 16:29:04,245 INFO [train.py:968] (1/2) Epoch 30, batch 37700, giga_loss[loss=0.2848, simple_loss=0.3627, pruned_loss=0.1035, over 28844.00 frames. ], tot_loss[loss=0.2708, simple_loss=0.343, pruned_loss=0.09928, over 5672543.15 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3549, pruned_loss=0.1095, over 5712494.27 frames. ], giga_tot_loss[loss=0.2668, simple_loss=0.3392, pruned_loss=0.09716, over 5682177.01 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:29:18,859 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:29:21,595 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358184.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:29:34,744 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5429, 1.7542, 1.8556, 1.4948], device='cuda:1'), covar=tensor([0.3393, 0.3019, 0.2947, 0.3302], device='cuda:1'), in_proj_covar=tensor([0.2069, 0.2023, 0.1933, 0.2085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 16:29:51,986 INFO [train.py:968] (1/2) Epoch 30, batch 37750, giga_loss[loss=0.2684, simple_loss=0.3522, pruned_loss=0.09232, over 28818.00 frames. ], tot_loss[loss=0.2738, simple_loss=0.3468, pruned_loss=0.1004, over 5674674.00 frames. ], libri_tot_loss[loss=0.2868, simple_loss=0.3548, pruned_loss=0.1094, over 5713467.96 frames. ], giga_tot_loss[loss=0.2704, simple_loss=0.3436, pruned_loss=0.09858, over 5680470.52 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:30:04,016 INFO [optim.py:369] (1/2) 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 16:30:04,887 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:30:06,984 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5555, 3.4052, 1.5229, 1.6631], device='cuda:1'), covar=tensor([0.1063, 0.0285, 0.0969, 0.1424], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0576, 0.0417, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 16:30:25,173 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2295, 1.5473, 1.5124, 1.1148], device='cuda:1'), covar=tensor([0.1750, 0.2578, 0.1424, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0718, 0.0998, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:30:42,142 INFO [train.py:968] (1/2) Epoch 30, batch 37800, giga_loss[loss=0.2952, simple_loss=0.3732, pruned_loss=0.1086, over 28576.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3511, pruned_loss=0.1024, over 5668933.01 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3546, pruned_loss=0.1091, over 5716836.46 frames. ], giga_tot_loss[loss=0.2754, simple_loss=0.3487, pruned_loss=0.1011, over 5669595.22 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:30:55,834 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.12 vs. limit=2.0
+2023-03-15 16:30:57,843 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4009, 1.4395, 1.2067, 1.5943], device='cuda:1'), covar=tensor([0.0794, 0.0369, 0.0365, 0.0885], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 16:31:26,652 INFO [train.py:968] (1/2) Epoch 30, batch 37850, giga_loss[loss=0.2368, simple_loss=0.3268, pruned_loss=0.07346, over 28700.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3544, pruned_loss=0.1041, over 5671199.29 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3548, pruned_loss=0.1092, over 5718552.86 frames. ], giga_tot_loss[loss=0.2792, simple_loss=0.3524, pruned_loss=0.1029, over 5669941.40 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:31:35,424 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 16:31:37,635 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1358330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:32:03,044 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1358359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:32:09,455 INFO [train.py:968] (1/2) Epoch 30, batch 37900, giga_loss[loss=0.2576, simple_loss=0.3484, pruned_loss=0.08335, over 28642.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3494, pruned_loss=0.09998, over 5672855.13 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3549, pruned_loss=0.1094, over 5712839.39 frames. ], giga_tot_loss[loss=0.2725, simple_loss=0.3476, pruned_loss=0.09868, over 5675655.20 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:32:47,854 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.8826, 1.2055, 1.3526, 1.0314], device='cuda:1'), covar=tensor([0.2561, 0.1637, 0.2695, 0.2097], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0760, 0.0733, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 16:32:51,755 INFO [train.py:968] (1/2) Epoch 30, batch 37950, giga_loss[loss=0.2559, simple_loss=0.3365, pruned_loss=0.08766, over 28852.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.3471, pruned_loss=0.0978, over 5677735.47 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3544, pruned_loss=0.1091, over 5708711.51 frames. ], giga_tot_loss[loss=0.2697, simple_loss=0.3459, pruned_loss=0.09676, over 5682724.88 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:33:03,504 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 38000, libri_loss[loss=0.3383, simple_loss=0.394, pruned_loss=0.1413, over 29156.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.346, pruned_loss=0.09676, over 5678408.89 frames. ], libri_tot_loss[loss=0.2867, simple_loss=0.3547, pruned_loss=0.1094, over 5712130.85 frames. ], giga_tot_loss[loss=0.2677, simple_loss=0.3446, pruned_loss=0.09545, over 5678650.75 frames. ], batch size: 101, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:33:40,069 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.10 vs. limit=2.0
+2023-03-15 16:34:02,628 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:34:18,558 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2735, 2.5863, 1.4542, 1.4642], device='cuda:1'), covar=tensor([0.0970, 0.0383, 0.0806, 0.1302], device='cuda:1'), in_proj_covar=tensor([0.0433, 0.0575, 0.0417, 0.0451], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 16:34:19,566 INFO [train.py:968] (1/2) Epoch 30, batch 38050, libri_loss[loss=0.3094, simple_loss=0.3724, pruned_loss=0.1231, over 29522.00 frames. ], tot_loss[loss=0.2717, simple_loss=0.3476, pruned_loss=0.09788, over 5690316.84 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3548, pruned_loss=0.1095, over 5718416.76 frames. ], giga_tot_loss[loss=0.2694, simple_loss=0.3461, pruned_loss=0.09631, over 5683815.78 frames. ], batch size: 80, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:34:29,360 INFO [optim.py:369] (1/2) 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 16:34:43,819 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3905, 5.2433, 4.9122, 2.6244], device='cuda:1'), covar=tensor([0.0456, 0.0562, 0.0623, 0.1723], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.1217, 0.1018, 0.0758], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 16:34:50,193 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:35:01,411 INFO [train.py:968] (1/2) Epoch 30, batch 38100, giga_loss[loss=0.3202, simple_loss=0.3833, pruned_loss=0.1285, over 28668.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3504, pruned_loss=0.09993, over 5691181.58 frames. ], libri_tot_loss[loss=0.2865, simple_loss=0.3545, pruned_loss=0.1092, over 5720957.47 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3493, pruned_loss=0.09864, over 5683152.67 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:35:08,422 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:35:23,198 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6756, 1.8607, 1.3251, 1.3509], device='cuda:1'), covar=tensor([0.1171, 0.0685, 0.1177, 0.1296], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0451, 0.0527, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 16:35:36,612 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:35:47,967 INFO [train.py:968] (1/2) Epoch 30, batch 38150, libri_loss[loss=0.2873, simple_loss=0.3631, pruned_loss=0.1058, over 29251.00 frames. ], tot_loss[loss=0.2765, simple_loss=0.3516, pruned_loss=0.1007, over 5686938.30 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3551, pruned_loss=0.1094, over 5712803.15 frames. ], giga_tot_loss[loss=0.2743, simple_loss=0.3501, pruned_loss=0.09924, over 5686506.82 frames. ], batch size: 94, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:35:54,583 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-15 16:36:00,058 INFO [optim.py:369] (1/2) 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 16:36:25,000 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:36:33,716 INFO [train.py:968] (1/2) Epoch 30, batch 38200, giga_loss[loss=0.2781, simple_loss=0.3522, pruned_loss=0.102, over 28928.00 frames. ], tot_loss[loss=0.2767, simple_loss=0.3516, pruned_loss=0.1009, over 5692428.84 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.355, pruned_loss=0.1093, over 5714752.25 frames. ], giga_tot_loss[loss=0.275, simple_loss=0.3505, pruned_loss=0.09972, over 5690177.38 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:37:00,373 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1358698.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:02,703 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1358701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:16,000 INFO [train.py:968] (1/2) Epoch 30, batch 38250, giga_loss[loss=0.3081, simple_loss=0.3742, pruned_loss=0.121, over 28817.00 frames. ], tot_loss[loss=0.2777, simple_loss=0.352, pruned_loss=0.1017, over 5695030.30 frames. ], libri_tot_loss[loss=0.2866, simple_loss=0.3549, pruned_loss=0.1092, over 5719999.79 frames. ], giga_tot_loss[loss=0.2763, simple_loss=0.3511, pruned_loss=0.1008, over 5687947.69 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:37:25,963 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1358730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:26,325 INFO [optim.py:369] (1/2) 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 16:37:26,793 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-15 16:37:29,659 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4020, 1.6625, 1.2550, 1.1775], device='cuda:1'), covar=tensor([0.1117, 0.0546, 0.1081, 0.1143], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0451, 0.0528, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 16:37:39,557 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1358747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:42,392 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1358750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:53,358 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1358761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:37:59,051 INFO [train.py:968] (1/2) Epoch 30, batch 38300, giga_loss[loss=0.2517, simple_loss=0.3293, pruned_loss=0.08709, over 28125.00 frames. ], tot_loss[loss=0.278, simple_loss=0.3523, pruned_loss=0.1018, over 5704140.72 frames. ], libri_tot_loss[loss=0.2862, simple_loss=0.3548, pruned_loss=0.1089, over 5724078.09 frames. ], giga_tot_loss[loss=0.2768, simple_loss=0.3516, pruned_loss=0.101, over 5693828.47 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:38:10,009 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1358779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:38:41,741 INFO [train.py:968] (1/2) Epoch 30, batch 38350, giga_loss[loss=0.2744, simple_loss=0.3491, pruned_loss=0.09989, over 28333.00 frames. ], tot_loss[loss=0.2752, simple_loss=0.3508, pruned_loss=0.09981, over 5708080.14 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3546, pruned_loss=0.1087, over 5726641.51 frames. ], giga_tot_loss[loss=0.2744, simple_loss=0.3504, pruned_loss=0.09919, over 5697338.84 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:38:52,193 INFO [optim.py:369] (1/2) 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 16:39:25,896 INFO [train.py:968] (1/2) Epoch 30, batch 38400, giga_loss[loss=0.2655, simple_loss=0.3574, pruned_loss=0.08676, over 28856.00 frames. ], tot_loss[loss=0.2756, simple_loss=0.3519, pruned_loss=0.09962, over 5700804.91 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3545, pruned_loss=0.1087, over 5719123.16 frames. ], giga_tot_loss[loss=0.2748, simple_loss=0.3516, pruned_loss=0.099, over 5699156.82 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:39:29,555 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:39:42,767 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6107, 1.6666, 1.7961, 1.4015], device='cuda:1'), covar=tensor([0.1948, 0.2845, 0.1594, 0.1923], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0721, 0.0999, 0.0897], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:40:05,859 INFO [train.py:968] (1/2) Epoch 30, batch 38450, giga_loss[loss=0.2934, simple_loss=0.3588, pruned_loss=0.114, over 27582.00 frames. ], tot_loss[loss=0.2737, simple_loss=0.3504, pruned_loss=0.09854, over 5711947.86 frames. ], libri_tot_loss[loss=0.2854, simple_loss=0.354, pruned_loss=0.1084, over 5726228.11 frames. ], giga_tot_loss[loss=0.2732, simple_loss=0.3505, pruned_loss=0.09797, over 5703513.77 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:40:17,462 INFO [optim.py:369] (1/2) 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 16:40:31,497 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1358949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:40:46,336 INFO [train.py:968] (1/2) Epoch 30, batch 38500, giga_loss[loss=0.269, simple_loss=0.346, pruned_loss=0.09598, over 28703.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3488, pruned_loss=0.09809, over 5703338.16 frames. ], libri_tot_loss[loss=0.286, simple_loss=0.3546, pruned_loss=0.1087, over 5717570.17 frames. ], giga_tot_loss[loss=0.2712, simple_loss=0.3483, pruned_loss=0.09706, over 5703314.48 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:41:26,906 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:41:28,002 INFO [train.py:968] (1/2) Epoch 30, batch 38550, giga_loss[loss=0.262, simple_loss=0.3369, pruned_loss=0.09358, over 28539.00 frames. ], tot_loss[loss=0.2724, simple_loss=0.3484, pruned_loss=0.09819, over 5703791.54 frames. ], libri_tot_loss[loss=0.2869, simple_loss=0.3552, pruned_loss=0.1093, over 5710431.45 frames. ], giga_tot_loss[loss=0.2702, simple_loss=0.3471, pruned_loss=0.09659, over 5709003.39 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:41:30,825 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:41:39,996 INFO [optim.py:369] (1/2) 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 16:41:41,730 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:41:43,078 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2965, 2.8875, 1.4034, 1.4829], device='cuda:1'), covar=tensor([0.1069, 0.0328, 0.0899, 0.1347], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0573, 0.0415, 0.0450], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 16:41:52,401 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:41:56,423 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4337, 2.5230, 2.4146, 2.2361], device='cuda:1'), covar=tensor([0.2282, 0.2642, 0.2317, 0.2321], device='cuda:1'), in_proj_covar=tensor([0.0511, 0.0759, 0.0731, 0.0702], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0009], device='cuda:1')
+2023-03-15 16:42:08,723 INFO [train.py:968] (1/2) Epoch 30, batch 38600, giga_loss[loss=0.2722, simple_loss=0.3452, pruned_loss=0.09959, over 27895.00 frames. ], tot_loss[loss=0.2712, simple_loss=0.3469, pruned_loss=0.09777, over 5706211.83 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3553, pruned_loss=0.1094, over 5711090.20 frames. ], giga_tot_loss[loss=0.2687, simple_loss=0.3455, pruned_loss=0.09596, over 5709980.81 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:42:09,779 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=3.16 vs. limit=5.0
+2023-03-15 16:42:29,090 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:42:31,053 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:42:40,545 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.20 vs. limit=2.0
+2023-03-15 16:42:50,381 INFO [train.py:968] (1/2) Epoch 30, batch 38650, giga_loss[loss=0.275, simple_loss=0.3547, pruned_loss=0.09763, over 28944.00 frames. ], tot_loss[loss=0.2735, simple_loss=0.3485, pruned_loss=0.09922, over 5709488.29 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3558, pruned_loss=0.1096, over 5713353.09 frames. ], giga_tot_loss[loss=0.2709, simple_loss=0.347, pruned_loss=0.09739, over 5710459.99 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:42:55,427 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:42:59,825 INFO [optim.py:369] (1/2) 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 16:43:04,596 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359136.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:43:24,145 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359160.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:43:29,237 INFO [train.py:968] (1/2) Epoch 30, batch 38700, giga_loss[loss=0.2739, simple_loss=0.3564, pruned_loss=0.09568, over 28929.00 frames. ], tot_loss[loss=0.274, simple_loss=0.3489, pruned_loss=0.09955, over 5695411.72 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.356, pruned_loss=0.1099, over 5699326.05 frames. ], giga_tot_loss[loss=0.271, simple_loss=0.3471, pruned_loss=0.09746, over 5708897.73 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:43:37,316 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:43:39,262 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:43:51,231 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1920, 3.3372, 1.3206, 1.4159], device='cuda:1'), covar=tensor([0.1316, 0.0361, 0.1116, 0.1714], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0571, 0.0415, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 16:44:02,816 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 38750, giga_loss[loss=0.3268, simple_loss=0.395, pruned_loss=0.1293, over 27863.00 frames. ], tot_loss[loss=0.273, simple_loss=0.3487, pruned_loss=0.09868, over 5692374.29 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3561, pruned_loss=0.11, over 5693072.40 frames. ], giga_tot_loss[loss=0.2703, simple_loss=0.3471, pruned_loss=0.09676, over 5708292.04 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:44:21,456 INFO [optim.py:369] (1/2) 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 16:44:25,587 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.20 vs. limit=5.0
+2023-03-15 16:44:31,496 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.59 vs. limit=5.0
+2023-03-15 16:44:50,972 INFO [train.py:968] (1/2) Epoch 30, batch 38800, giga_loss[loss=0.2706, simple_loss=0.3416, pruned_loss=0.09984, over 28882.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3471, pruned_loss=0.09709, over 5702309.43 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3559, pruned_loss=0.1098, over 5698605.09 frames. ], giga_tot_loss[loss=0.2684, simple_loss=0.3459, pruned_loss=0.09548, over 5710185.19 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:44:51,832 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1481, 1.4419, 1.3835, 1.0216], device='cuda:1'), covar=tensor([0.1423, 0.2447, 0.1292, 0.1561], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0721, 0.0998, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:44:59,326 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:02,011 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:04,526 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4789, 3.4324, 1.5200, 1.7380], device='cuda:1'), covar=tensor([0.1049, 0.0276, 0.0978, 0.1332], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0570, 0.0415, 0.0449], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0034, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 16:45:26,077 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:32,517 INFO [train.py:968] (1/2) Epoch 30, batch 38850, giga_loss[loss=0.2782, simple_loss=0.3541, pruned_loss=0.1012, over 28573.00 frames. ], tot_loss[loss=0.2705, simple_loss=0.3466, pruned_loss=0.09722, over 5702356.13 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3561, pruned_loss=0.11, over 5698525.91 frames. ], giga_tot_loss[loss=0.2681, simple_loss=0.3452, pruned_loss=0.09546, over 5708582.87 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:45:36,782 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:45:41,486 INFO [optim.py:369] (1/2) 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 16:46:10,902 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359365.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 16:46:12,949 INFO [train.py:968] (1/2) Epoch 30, batch 38900, giga_loss[loss=0.2748, simple_loss=0.3582, pruned_loss=0.09576, over 28781.00 frames. ], tot_loss[loss=0.2668, simple_loss=0.3431, pruned_loss=0.09526, over 5702286.90 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.356, pruned_loss=0.1099, over 5700647.28 frames. ], giga_tot_loss[loss=0.2649, simple_loss=0.342, pruned_loss=0.09384, over 5705431.36 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:46:37,479 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7033, 1.6687, 1.8692, 1.4542], device='cuda:1'), covar=tensor([0.2015, 0.2696, 0.1611, 0.1933], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0722, 0.0999, 0.0896], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:46:53,795 INFO [train.py:968] (1/2) Epoch 30, batch 38950, giga_loss[loss=0.2599, simple_loss=0.334, pruned_loss=0.0929, over 28706.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3393, pruned_loss=0.09332, over 5699092.82 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.356, pruned_loss=0.1099, over 5694894.48 frames. ], giga_tot_loss[loss=0.2611, simple_loss=0.3382, pruned_loss=0.09196, over 5706420.84 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:47:03,196 INFO [optim.py:369] (1/2) 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 16:47:27,915 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3773, 1.4898, 1.4171, 1.3538], device='cuda:1'), covar=tensor([0.2412, 0.2637, 0.2168, 0.2321], device='cuda:1'), in_proj_covar=tensor([0.2082, 0.2042, 0.1947, 0.2100], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 16:47:33,164 INFO [train.py:968] (1/2) Epoch 30, batch 39000, giga_loss[loss=0.2718, simple_loss=0.3524, pruned_loss=0.0956, over 28662.00 frames. ], tot_loss[loss=0.2629, simple_loss=0.3388, pruned_loss=0.09349, over 5701810.27 frames. ], libri_tot_loss[loss=0.2879, simple_loss=0.3559, pruned_loss=0.1099, over 5699144.40 frames. ], giga_tot_loss[loss=0.261, simple_loss=0.3377, pruned_loss=0.09211, over 5703990.01 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:47:33,164 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 16:47:42,177 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 16:48:27,636 INFO [train.py:968] (1/2) Epoch 30, batch 39050, giga_loss[loss=0.2323, simple_loss=0.3123, pruned_loss=0.07618, over 28975.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3388, pruned_loss=0.09394, over 5699071.50 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.3557, pruned_loss=0.1098, over 5703325.31 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3378, pruned_loss=0.09264, over 5696916.73 frames. ], batch size: 145, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:48:37,291 INFO [optim.py:369] (1/2) 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 16:48:41,630 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:49:07,015 INFO [train.py:968] (1/2) Epoch 30, batch 39100, giga_loss[loss=0.2621, simple_loss=0.3343, pruned_loss=0.09489, over 29070.00 frames. ], tot_loss[loss=0.2613, simple_loss=0.3364, pruned_loss=0.09312, over 5705186.59 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3559, pruned_loss=0.1101, over 5705948.42 frames. ], giga_tot_loss[loss=0.2592, simple_loss=0.3352, pruned_loss=0.09165, over 5701206.41 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:49:47,683 INFO [train.py:968] (1/2) Epoch 30, batch 39150, giga_loss[loss=0.2987, simple_loss=0.3558, pruned_loss=0.1208, over 28587.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3347, pruned_loss=0.09275, over 5699049.11 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3564, pruned_loss=0.1105, over 5695954.34 frames. ], giga_tot_loss[loss=0.2569, simple_loss=0.3326, pruned_loss=0.09064, over 5705783.99 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:49:58,728 INFO [optim.py:369] (1/2) 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 16:50:25,694 INFO [train.py:968] (1/2) Epoch 30, batch 39200, giga_loss[loss=0.2415, simple_loss=0.3205, pruned_loss=0.08127, over 28621.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3325, pruned_loss=0.09162, over 5700536.19 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3564, pruned_loss=0.1104, over 5699277.13 frames. ], giga_tot_loss[loss=0.2547, simple_loss=0.3303, pruned_loss=0.08953, over 5703142.11 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 16:50:31,103 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3907, 1.6125, 1.3754, 1.6300], device='cuda:1'), covar=tensor([0.0752, 0.0323, 0.0350, 0.0881], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0121, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0067, 0.0117], device='cuda:1')
+2023-03-15 16:50:34,883 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:50:37,099 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:50:52,150 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:50:56,140 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-15 16:51:00,281 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:51:06,710 INFO [train.py:968] (1/2) Epoch 30, batch 39250, giga_loss[loss=0.2579, simple_loss=0.3332, pruned_loss=0.09134, over 28738.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3305, pruned_loss=0.09015, over 5711244.83 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.356, pruned_loss=0.11, over 5706483.39 frames. ], giga_tot_loss[loss=0.2524, simple_loss=0.3283, pruned_loss=0.08821, over 5706971.94 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:51:15,976 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5828, 2.2310, 1.5691, 0.8498], device='cuda:1'), covar=tensor([0.7569, 0.3368, 0.5025, 0.7795], device='cuda:1'), in_proj_covar=tensor([0.1863, 0.1746, 0.1669, 0.1513], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:51:22,407 INFO [optim.py:369] (1/2) 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 16:51:26,679 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1359740.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 16:51:49,780 INFO [train.py:968] (1/2) Epoch 30, batch 39300, libri_loss[loss=0.2853, simple_loss=0.3594, pruned_loss=0.1056, over 29510.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3327, pruned_loss=0.09123, over 5706396.56 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3564, pruned_loss=0.1104, over 5707032.63 frames. ], giga_tot_loss[loss=0.2535, simple_loss=0.3297, pruned_loss=0.08866, over 5702717.43 frames. ], batch size: 82, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:51:53,501 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.9630, 1.3091, 1.1248, 0.2277], device='cuda:1'), covar=tensor([0.5070, 0.3852, 0.5214, 0.7907], device='cuda:1'), in_proj_covar=tensor([0.1866, 0.1748, 0.1672, 0.1516], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 16:52:35,878 INFO [train.py:968] (1/2) Epoch 30, batch 39350, giga_loss[loss=0.2478, simple_loss=0.3309, pruned_loss=0.08236, over 29026.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.3363, pruned_loss=0.09294, over 5704268.80 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3564, pruned_loss=0.1104, over 5710118.55 frames. ], giga_tot_loss[loss=0.2575, simple_loss=0.3336, pruned_loss=0.09066, over 5698413.13 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:52:42,376 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3503, 1.4887, 1.4430, 1.3445], device='cuda:1'), covar=tensor([0.2364, 0.2128, 0.2303, 0.2181], device='cuda:1'), in_proj_covar=tensor([0.2090, 0.2051, 0.1954, 0.2108], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 16:52:51,384 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:00,275 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:11,076 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:20,840 INFO [train.py:968] (1/2) Epoch 30, batch 39400, giga_loss[loss=0.2803, simple_loss=0.3608, pruned_loss=0.09993, over 28298.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3391, pruned_loss=0.09401, over 5695455.60 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3562, pruned_loss=0.1102, over 5704952.84 frames. ], giga_tot_loss[loss=0.2601, simple_loss=0.3365, pruned_loss=0.0918, over 5694159.33 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:53:26,584 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1359874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:53:35,346 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1359883.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 16:53:39,060 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1359886.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 16:54:04,859 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 39450, giga_loss[loss=0.2569, simple_loss=0.3428, pruned_loss=0.08552, over 28697.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.3411, pruned_loss=0.09476, over 5693550.41 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3562, pruned_loss=0.1102, over 5707533.86 frames. ], giga_tot_loss[loss=0.2619, simple_loss=0.3387, pruned_loss=0.09255, over 5689762.56 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:54:15,610 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1359931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:54:17,601 INFO [optim.py:369] (1/2) 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 16:54:37,467 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 39500, giga_loss[loss=0.2495, simple_loss=0.3272, pruned_loss=0.08594, over 29036.00 frames. ], tot_loss[loss=0.2647, simple_loss=0.3411, pruned_loss=0.09419, over 5701128.48 frames. ], libri_tot_loss[loss=0.2882, simple_loss=0.3562, pruned_loss=0.1101, over 5711234.15 frames. ], giga_tot_loss[loss=0.2614, simple_loss=0.3387, pruned_loss=0.09206, over 5694687.37 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:54:55,167 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5129, 1.6201, 1.1713, 1.2423], device='cuda:1'), covar=tensor([0.0960, 0.0610, 0.1041, 0.1385], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0451, 0.0525, 0.0465], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 16:55:26,508 INFO [train.py:968] (1/2) Epoch 30, batch 39550, giga_loss[loss=0.2278, simple_loss=0.3036, pruned_loss=0.07596, over 28697.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.34, pruned_loss=0.09352, over 5696567.91 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3561, pruned_loss=0.11, over 5703728.62 frames. ], giga_tot_loss[loss=0.2603, simple_loss=0.3377, pruned_loss=0.09148, over 5697773.23 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:55:39,868 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 39600, giga_loss[loss=0.2952, simple_loss=0.3725, pruned_loss=0.1089, over 29075.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3408, pruned_loss=0.09404, over 5703753.74 frames. ], libri_tot_loss[loss=0.2881, simple_loss=0.3563, pruned_loss=0.11, over 5705092.37 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3387, pruned_loss=0.09219, over 5703283.58 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:56:51,090 INFO [train.py:968] (1/2) Epoch 30, batch 39650, giga_loss[loss=0.2603, simple_loss=0.3396, pruned_loss=0.09044, over 28905.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3425, pruned_loss=0.09541, over 5717180.96 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.357, pruned_loss=0.1106, over 5710874.24 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3395, pruned_loss=0.09288, over 5711625.13 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 16:57:00,678 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.0832, 1.3147, 1.3236, 1.0078], device='cuda:1'), covar=tensor([0.1414, 0.2209, 0.1226, 0.1609], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0719, 0.0996, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 16:57:05,807 INFO [optim.py:369] (1/2) 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 16:57:35,390 INFO [train.py:968] (1/2) Epoch 30, batch 39700, giga_loss[loss=0.3311, simple_loss=0.3976, pruned_loss=0.1323, over 28574.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.346, pruned_loss=0.09731, over 5713353.95 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3569, pruned_loss=0.1106, over 5712905.97 frames. ], giga_tot_loss[loss=0.267, simple_loss=0.3437, pruned_loss=0.09513, over 5707236.31 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:57:36,582 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360169.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 16:57:50,943 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 16:58:13,873 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3740, 1.4368, 1.5107, 1.3540], device='cuda:1'), covar=tensor([0.3212, 0.2940, 0.2648, 0.3019], device='cuda:1'), in_proj_covar=tensor([0.2088, 0.2054, 0.1955, 0.2106], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 16:58:17,732 INFO [train.py:968] (1/2) Epoch 30, batch 39750, giga_loss[loss=0.281, simple_loss=0.3454, pruned_loss=0.1083, over 28523.00 frames. ], tot_loss[loss=0.2744, simple_loss=0.3497, pruned_loss=0.09955, over 5704057.53 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3572, pruned_loss=0.1107, over 5702206.14 frames. ], giga_tot_loss[loss=0.2713, simple_loss=0.3474, pruned_loss=0.09754, over 5708665.21 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:58:18,572 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:58:28,884 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:58:32,141 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 39800, giga_loss[loss=0.2488, simple_loss=0.3299, pruned_loss=0.08386, over 28414.00 frames. ], tot_loss[loss=0.2734, simple_loss=0.3493, pruned_loss=0.09871, over 5705103.42 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.357, pruned_loss=0.1104, over 5704114.63 frames. ], giga_tot_loss[loss=0.2711, simple_loss=0.3476, pruned_loss=0.09726, over 5706878.89 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:59:00,892 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3795, 1.2234, 4.0769, 3.3870], device='cuda:1'), covar=tensor([0.1590, 0.2707, 0.0477, 0.1003], device='cuda:1'), in_proj_covar=tensor([0.0818, 0.0679, 0.1014, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 16:59:32,449 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:59:35,377 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 16:59:38,432 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-15 16:59:44,537 INFO [train.py:968] (1/2) Epoch 30, batch 39850, giga_loss[loss=0.2802, simple_loss=0.3611, pruned_loss=0.09965, over 28607.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3507, pruned_loss=0.09927, over 5705406.67 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3567, pruned_loss=0.1103, over 5703724.06 frames. ], giga_tot_loss[loss=0.273, simple_loss=0.3496, pruned_loss=0.09821, over 5707104.86 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 16:59:58,820 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:00:03,776 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 17:00:24,074 INFO [train.py:968] (1/2) Epoch 30, batch 39900, giga_loss[loss=0.2897, simple_loss=0.3723, pruned_loss=0.1035, over 28666.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3506, pruned_loss=0.09954, over 5707571.05 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3566, pruned_loss=0.1102, over 5710243.73 frames. ], giga_tot_loss[loss=0.2733, simple_loss=0.3496, pruned_loss=0.09845, over 5702832.05 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:00:28,252 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:00:31,164 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360377.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:00:48,914 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-15 17:00:55,001 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:01:03,653 INFO [train.py:968] (1/2) Epoch 30, batch 39950, giga_loss[loss=0.2572, simple_loss=0.3294, pruned_loss=0.09251, over 28688.00 frames. ], tot_loss[loss=0.2733, simple_loss=0.3492, pruned_loss=0.09874, over 5710935.02 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3569, pruned_loss=0.1104, over 5705844.58 frames. ], giga_tot_loss[loss=0.2714, simple_loss=0.348, pruned_loss=0.09741, over 5710742.43 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:01:04,496 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3565, 2.9386, 1.4447, 1.4222], device='cuda:1'), covar=tensor([0.0924, 0.0339, 0.0926, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0576, 0.0418, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 17:01:16,463 INFO [optim.py:369] (1/2) 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 17:01:30,347 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:01:33,565 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:01:46,617 INFO [train.py:968] (1/2) Epoch 30, batch 40000, giga_loss[loss=0.2281, simple_loss=0.3024, pruned_loss=0.07687, over 28744.00 frames. ], tot_loss[loss=0.2702, simple_loss=0.3459, pruned_loss=0.09725, over 5715630.23 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.357, pruned_loss=0.1104, over 5707502.88 frames. ], giga_tot_loss[loss=0.2685, simple_loss=0.3448, pruned_loss=0.09606, over 5714126.37 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:01:54,115 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360475.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:01:56,842 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:02:01,042 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 17:02:17,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5950, 1.6430, 1.7531, 1.3889], device='cuda:1'), covar=tensor([0.1867, 0.2559, 0.1587, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.0946, 0.0720, 0.0997, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 17:02:18,790 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 40050, libri_loss[loss=0.3093, simple_loss=0.3839, pruned_loss=0.1174, over 28751.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3432, pruned_loss=0.09588, over 5713865.48 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3576, pruned_loss=0.1108, over 5711833.76 frames. ], giga_tot_loss[loss=0.265, simple_loss=0.3415, pruned_loss=0.09428, over 5708976.91 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:02:41,387 INFO [optim.py:369] (1/2) 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 17:02:48,628 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360544.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 17:03:01,774 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.6332, 1.1956, 4.4355, 3.4648], device='cuda:1'), covar=tensor([0.1606, 0.3160, 0.0428, 0.1070], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0678, 0.1014, 0.0992], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:03:07,319 INFO [train.py:968] (1/2) Epoch 30, batch 40100, giga_loss[loss=0.2571, simple_loss=0.3444, pruned_loss=0.08494, over 28938.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3417, pruned_loss=0.09421, over 5722913.27 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1104, over 5716936.43 frames. ], giga_tot_loss[loss=0.2631, simple_loss=0.3404, pruned_loss=0.09291, over 5714510.09 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:03:10,493 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 17:03:27,100 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4726, 3.1427, 1.4397, 1.5512], device='cuda:1'), covar=tensor([0.0967, 0.0325, 0.1010, 0.1345], device='cuda:1'), in_proj_covar=tensor([0.0430, 0.0575, 0.0418, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 17:03:28,721 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:03:31,733 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7999, 2.0688, 2.1551, 1.6910], device='cuda:1'), covar=tensor([0.3579, 0.2811, 0.2782, 0.3217], device='cuda:1'), in_proj_covar=tensor([0.2088, 0.2053, 0.1957, 0.2105], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 17:03:34,170 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4860, 1.4365, 4.2439, 3.4892], device='cuda:1'), covar=tensor([0.1541, 0.2749, 0.0405, 0.1130], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0677, 0.1013, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:03:48,018 INFO [train.py:968] (1/2) Epoch 30, batch 40150, giga_loss[loss=0.2195, simple_loss=0.3073, pruned_loss=0.06584, over 28712.00 frames. ], tot_loss[loss=0.266, simple_loss=0.344, pruned_loss=0.09405, over 5718569.46 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1105, over 5720160.19 frames. ], giga_tot_loss[loss=0.2639, simple_loss=0.3427, pruned_loss=0.09255, over 5708917.15 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:03:53,026 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0644, 2.2221, 2.2628, 1.7899], device='cuda:1'), covar=tensor([0.1907, 0.2255, 0.1557, 0.1751], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.0719, 0.0995, 0.0894], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 17:04:02,641 INFO [optim.py:369] (1/2) 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 17:04:21,424 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9884, 1.1443, 2.8224, 2.6629], device='cuda:1'), covar=tensor([0.1612, 0.2658, 0.0621, 0.1295], device='cuda:1'), in_proj_covar=tensor([0.0816, 0.0677, 0.1012, 0.0990], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:04:29,836 INFO [train.py:968] (1/2) Epoch 30, batch 40200, giga_loss[loss=0.2405, simple_loss=0.3268, pruned_loss=0.07707, over 28875.00 frames. ], tot_loss[loss=0.2651, simple_loss=0.3432, pruned_loss=0.09353, over 5717170.67 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3571, pruned_loss=0.1104, over 5723906.40 frames. ], giga_tot_loss[loss=0.2629, simple_loss=0.3418, pruned_loss=0.09198, over 5705996.31 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:04:38,414 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1360678.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:04:45,226 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360687.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:04:47,315 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360690.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 17:04:52,447 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3359, 2.6170, 1.9187, 2.2652], device='cuda:1'), covar=tensor([0.0855, 0.0519, 0.0862, 0.0979], device='cuda:1'), in_proj_covar=tensor([0.0417, 0.0451, 0.0525, 0.0464], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 17:05:08,398 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:05:09,527 INFO [train.py:968] (1/2) Epoch 30, batch 40250, giga_loss[loss=0.2613, simple_loss=0.3377, pruned_loss=0.09245, over 28791.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3421, pruned_loss=0.09379, over 5722990.88 frames. ], libri_tot_loss[loss=0.2884, simple_loss=0.3567, pruned_loss=0.1101, over 5728242.53 frames. ], giga_tot_loss[loss=0.263, simple_loss=0.341, pruned_loss=0.09248, over 5709994.33 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:05:10,765 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360719.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 17:05:23,307 INFO [optim.py:369] (1/2) 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 17:05:24,872 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:05:27,711 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 40300, giga_loss[loss=0.2292, simple_loss=0.298, pruned_loss=0.08019, over 28181.00 frames. ], tot_loss[loss=0.2663, simple_loss=0.3421, pruned_loss=0.09531, over 5719089.59 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3568, pruned_loss=0.1101, over 5731258.29 frames. ], giga_tot_loss[loss=0.2642, simple_loss=0.3408, pruned_loss=0.09383, over 5705781.30 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:05:50,509 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:06:32,238 INFO [train.py:968] (1/2) Epoch 30, batch 40350, giga_loss[loss=0.2571, simple_loss=0.3344, pruned_loss=0.08986, over 28938.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3412, pruned_loss=0.096, over 5708245.65 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3568, pruned_loss=0.1101, over 5721265.56 frames. ], giga_tot_loss[loss=0.2645, simple_loss=0.3399, pruned_loss=0.09457, over 5705894.82 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:06:37,296 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1360821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:06:40,149 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1360824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:06:47,037 INFO [optim.py:369] (1/2) 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 17:07:02,522 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1360853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:07:13,267 INFO [train.py:968] (1/2) Epoch 30, batch 40400, giga_loss[loss=0.2265, simple_loss=0.2974, pruned_loss=0.07785, over 28522.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3393, pruned_loss=0.09523, over 5716750.69 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3568, pruned_loss=0.1099, over 5723958.77 frames. ], giga_tot_loss[loss=0.2626, simple_loss=0.3377, pruned_loss=0.09377, over 5712236.46 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:07:47,213 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4177, 3.8990, 1.5207, 1.5740], device='cuda:1'), covar=tensor([0.0995, 0.0365, 0.0983, 0.1312], device='cuda:1'), in_proj_covar=tensor([0.0429, 0.0576, 0.0417, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0028, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 17:07:55,354 INFO [train.py:968] (1/2) Epoch 30, batch 40450, giga_loss[loss=0.2509, simple_loss=0.3188, pruned_loss=0.09154, over 28641.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.3384, pruned_loss=0.09509, over 5724932.86 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3571, pruned_loss=0.1101, over 5729725.63 frames. ], giga_tot_loss[loss=0.2615, simple_loss=0.3364, pruned_loss=0.09335, over 5715979.98 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:08:09,327 INFO [optim.py:369] (1/2) 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 17:08:31,748 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1360967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:08:32,167 INFO [train.py:968] (1/2) Epoch 30, batch 40500, giga_loss[loss=0.2259, simple_loss=0.3, pruned_loss=0.07587, over 28815.00 frames. ], tot_loss[loss=0.262, simple_loss=0.3357, pruned_loss=0.09415, over 5728810.94 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3569, pruned_loss=0.11, over 5735570.69 frames. ], giga_tot_loss[loss=0.2588, simple_loss=0.3333, pruned_loss=0.09213, over 5715956.07 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:09:00,171 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:09:11,588 INFO [train.py:968] (1/2) Epoch 30, batch 40550, giga_loss[loss=0.2122, simple_loss=0.2969, pruned_loss=0.06381, over 28619.00 frames. ], tot_loss[loss=0.2571, simple_loss=0.3308, pruned_loss=0.0917, over 5724496.97 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3568, pruned_loss=0.1101, over 5728965.27 frames. ], giga_tot_loss[loss=0.2538, simple_loss=0.3284, pruned_loss=0.08962, over 5719364.01 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:09:27,939 INFO [optim.py:369] (1/2) 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 17:09:28,935 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3710, 1.4070, 1.2566, 1.5263], device='cuda:1'), covar=tensor([0.0726, 0.0393, 0.0370, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0076, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 17:09:51,597 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.99 vs. limit=5.0
+2023-03-15 17:09:54,161 INFO [train.py:968] (1/2) Epoch 30, batch 40600, giga_loss[loss=0.2329, simple_loss=0.3179, pruned_loss=0.07399, over 28911.00 frames. ], tot_loss[loss=0.2546, simple_loss=0.3285, pruned_loss=0.09034, over 5716749.60 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3569, pruned_loss=0.1101, over 5724764.20 frames. ], giga_tot_loss[loss=0.251, simple_loss=0.3258, pruned_loss=0.08809, over 5716222.06 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:10:03,847 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.3802, 5.2263, 4.9657, 2.5903], device='cuda:1'), covar=tensor([0.0469, 0.0585, 0.0583, 0.1661], device='cuda:1'), in_proj_covar=tensor([0.1317, 0.1214, 0.1020, 0.0755], device='cuda:1'), out_proj_covar=tensor([0.0019, 0.0017, 0.0013, 0.0012], device='cuda:1')
+2023-03-15 17:10:11,000 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:10:34,359 INFO [train.py:968] (1/2) Epoch 30, batch 40650, giga_loss[loss=0.2748, simple_loss=0.3554, pruned_loss=0.09713, over 29047.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.3312, pruned_loss=0.09125, over 5708713.05 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3572, pruned_loss=0.1104, over 5724341.04 frames. ], giga_tot_loss[loss=0.2529, simple_loss=0.3281, pruned_loss=0.08882, over 5708577.97 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:10:49,141 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 40700, giga_loss[loss=0.2684, simple_loss=0.3498, pruned_loss=0.09345, over 28870.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.3345, pruned_loss=0.09244, over 5712836.71 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.3574, pruned_loss=0.1105, over 5727244.21 frames. ], giga_tot_loss[loss=0.2558, simple_loss=0.3314, pruned_loss=0.09009, over 5710068.21 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:11:55,084 INFO [train.py:968] (1/2) Epoch 30, batch 40750, giga_loss[loss=0.3002, simple_loss=0.3638, pruned_loss=0.1184, over 28892.00 frames. ], tot_loss[loss=0.2625, simple_loss=0.3375, pruned_loss=0.09373, over 5707528.82 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3574, pruned_loss=0.1104, over 5722996.59 frames. ], giga_tot_loss[loss=0.2582, simple_loss=0.3342, pruned_loss=0.09113, over 5709401.76 frames. ], batch size: 66, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:11:55,913 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361219.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:06,970 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:08,149 INFO [optim.py:369] (1/2) 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 17:12:09,041 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:33,956 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:12:34,931 INFO [train.py:968] (1/2) Epoch 30, batch 40800, giga_loss[loss=0.2748, simple_loss=0.3477, pruned_loss=0.101, over 28954.00 frames. ], tot_loss[loss=0.2643, simple_loss=0.34, pruned_loss=0.09431, over 5702079.36 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3576, pruned_loss=0.1106, over 5709477.81 frames. ], giga_tot_loss[loss=0.2597, simple_loss=0.3364, pruned_loss=0.09152, over 5715775.64 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:12:40,155 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.87 vs. limit=2.0
+2023-03-15 17:12:49,104 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 17:12:52,260 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4611, 2.0701, 1.4580, 0.7858], device='cuda:1'), covar=tensor([0.7826, 0.3595, 0.4819, 0.8164], device='cuda:1'), in_proj_covar=tensor([0.1865, 0.1751, 0.1667, 0.1514], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 17:13:04,438 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3435, 1.5896, 1.2304, 1.1875], device='cuda:1'), covar=tensor([0.1145, 0.0643, 0.1145, 0.1259], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0453, 0.0525, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 17:13:15,632 INFO [train.py:968] (1/2) Epoch 30, batch 40850, giga_loss[loss=0.2754, simple_loss=0.3457, pruned_loss=0.1026, over 28797.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3436, pruned_loss=0.09654, over 5699662.01 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.358, pruned_loss=0.1108, over 5705114.37 frames. ], giga_tot_loss[loss=0.2636, simple_loss=0.3399, pruned_loss=0.09368, over 5713762.24 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:13:23,230 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.14 vs. limit=2.0
+2023-03-15 17:13:30,513 INFO [optim.py:369] (1/2) 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 17:13:32,863 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.80 vs. limit=2.0
+2023-03-15 17:13:33,176 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:13:36,222 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:14:02,062 INFO [train.py:968] (1/2) Epoch 30, batch 40900, giga_loss[loss=0.2475, simple_loss=0.3213, pruned_loss=0.08687, over 28690.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3447, pruned_loss=0.09767, over 5696117.57 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3578, pruned_loss=0.1107, over 5707723.82 frames. ], giga_tot_loss[loss=0.2662, simple_loss=0.3418, pruned_loss=0.09527, over 5705031.90 frames. ], batch size: 92, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:14:12,754 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 40950, giga_loss[loss=0.3107, simple_loss=0.3564, pruned_loss=0.1325, over 23527.00 frames. ], tot_loss[loss=0.2804, simple_loss=0.3521, pruned_loss=0.1044, over 5665272.64 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.358, pruned_loss=0.111, over 5693969.24 frames. ], giga_tot_loss[loss=0.2761, simple_loss=0.349, pruned_loss=0.1016, over 5683545.74 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:15:07,203 INFO [optim.py:369] (1/2) 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 17:15:37,648 INFO [train.py:968] (1/2) Epoch 30, batch 41000, giga_loss[loss=0.2989, simple_loss=0.3684, pruned_loss=0.1147, over 28849.00 frames. ], tot_loss[loss=0.2892, simple_loss=0.3593, pruned_loss=0.1095, over 5673741.18 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.358, pruned_loss=0.1109, over 5697100.93 frames. ], giga_tot_loss[loss=0.2858, simple_loss=0.3569, pruned_loss=0.1074, over 5685032.90 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:15:47,983 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:15:54,070 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:15:56,570 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:16:09,364 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4230, 1.5413, 3.1858, 3.1328], device='cuda:1'), covar=tensor([0.1261, 0.2332, 0.0484, 0.1079], device='cuda:1'), in_proj_covar=tensor([0.0817, 0.0678, 0.1015, 0.0993], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:16:12,257 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2110, 2.5197, 1.3225, 1.3293], device='cuda:1'), covar=tensor([0.1045, 0.0414, 0.0917, 0.1407], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0578, 0.0418, 0.0452], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0031], device='cuda:1')
+2023-03-15 17:16:17,067 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9839, 2.3763, 1.5604, 1.7490], device='cuda:1'), covar=tensor([0.1038, 0.0614, 0.1004, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0454, 0.0527, 0.0467], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 17:16:23,747 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 41050, giga_loss[loss=0.3084, simple_loss=0.3707, pruned_loss=0.1231, over 28626.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3645, pruned_loss=0.1138, over 5667824.76 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3575, pruned_loss=0.1106, over 5703856.37 frames. ], giga_tot_loss[loss=0.294, simple_loss=0.3632, pruned_loss=0.1124, over 5669975.79 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:16:26,845 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:16:28,828 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:16:39,943 INFO [optim.py:369] (1/2) 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 17:16:53,778 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:17:04,349 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.41 vs. limit=2.0
+2023-03-15 17:17:08,342 INFO [train.py:968] (1/2) Epoch 30, batch 41100, giga_loss[loss=0.36, simple_loss=0.4085, pruned_loss=0.1558, over 28534.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3707, pruned_loss=0.119, over 5652845.85 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3579, pruned_loss=0.1109, over 5685153.55 frames. ], giga_tot_loss[loss=0.3024, simple_loss=0.3694, pruned_loss=0.1177, over 5670270.08 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:17:34,070 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1361594.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:17:56,572 INFO [train.py:968] (1/2) Epoch 30, batch 41150, giga_loss[loss=0.3524, simple_loss=0.4094, pruned_loss=0.1477, over 28783.00 frames. ], tot_loss[loss=0.3121, simple_loss=0.3765, pruned_loss=0.1238, over 5658132.80 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3577, pruned_loss=0.1108, over 5688579.29 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3759, pruned_loss=0.123, over 5668317.23 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:18:15,664 INFO [optim.py:369] (1/2) 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 17:18:23,710 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:18:29,201 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:18:49,722 INFO [train.py:968] (1/2) Epoch 30, batch 41200, giga_loss[loss=0.3383, simple_loss=0.3969, pruned_loss=0.1399, over 28705.00 frames. ], tot_loss[loss=0.3158, simple_loss=0.3786, pruned_loss=0.1265, over 5655611.58 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.3579, pruned_loss=0.111, over 5695387.76 frames. ], giga_tot_loss[loss=0.3155, simple_loss=0.3786, pruned_loss=0.1262, over 5656843.43 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:19:07,384 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4131, 3.7095, 1.6312, 1.6378], device='cuda:1'), covar=tensor([0.1048, 0.0396, 0.0908, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0431, 0.0580, 0.0418, 0.0453], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 17:19:19,590 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8269, 2.7898, 1.7114, 1.0076], device='cuda:1'), covar=tensor([0.8899, 0.3928, 0.4618, 0.8064], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1757, 0.1668, 0.1515], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 17:19:37,349 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3243, 1.7273, 1.2575, 0.8147], device='cuda:1'), covar=tensor([0.4645, 0.2843, 0.2717, 0.5678], device='cuda:1'), in_proj_covar=tensor([0.1868, 0.1756, 0.1668, 0.1515], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 17:19:41,511 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 41250, giga_loss[loss=0.2452, simple_loss=0.3284, pruned_loss=0.081, over 28629.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3814, pruned_loss=0.1299, over 5638882.48 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.358, pruned_loss=0.111, over 5697273.38 frames. ], giga_tot_loss[loss=0.3208, simple_loss=0.3817, pruned_loss=0.1299, over 5637425.30 frames. ], batch size: 71, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:20:08,163 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:20:08,482 INFO [optim.py:369] (1/2) 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 17:20:10,792 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:20:39,630 INFO [train.py:968] (1/2) Epoch 30, batch 41300, giga_loss[loss=0.4145, simple_loss=0.448, pruned_loss=0.1906, over 27491.00 frames. ], tot_loss[loss=0.3241, simple_loss=0.3834, pruned_loss=0.1324, over 5628884.49 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3577, pruned_loss=0.1108, over 5700308.98 frames. ], giga_tot_loss[loss=0.3249, simple_loss=0.3842, pruned_loss=0.1328, over 5624002.80 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:20:42,099 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:20:42,930 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.15 vs. limit=2.0
+2023-03-15 17:21:00,821 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1361788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:21:30,081 INFO [train.py:968] (1/2) Epoch 30, batch 41350, libri_loss[loss=0.364, simple_loss=0.4115, pruned_loss=0.1582, over 29754.00 frames. ], tot_loss[loss=0.3289, simple_loss=0.3866, pruned_loss=0.1356, over 5621754.28 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3578, pruned_loss=0.111, over 5695077.81 frames. ], giga_tot_loss[loss=0.331, simple_loss=0.3884, pruned_loss=0.1368, over 5620015.40 frames. ], batch size: 87, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:21:53,055 INFO [optim.py:369] (1/2) 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 17:22:07,849 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361856.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:22:14,665 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 41400, giga_loss[loss=0.2903, simple_loss=0.3551, pruned_loss=0.1128, over 28991.00 frames. ], tot_loss[loss=0.3308, simple_loss=0.3881, pruned_loss=0.1368, over 5633294.35 frames. ], libri_tot_loss[loss=0.2897, simple_loss=0.3575, pruned_loss=0.1109, over 5697808.06 frames. ], giga_tot_loss[loss=0.3331, simple_loss=0.3901, pruned_loss=0.1381, over 5628636.85 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:22:50,653 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1361888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:23:00,434 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.9271, 2.9670, 1.9363, 1.1090], device='cuda:1'), covar=tensor([0.9318, 0.3647, 0.4231, 0.8235], device='cuda:1'), in_proj_covar=tensor([0.1877, 0.1764, 0.1672, 0.1521], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 17:23:19,678 INFO [train.py:968] (1/2) Epoch 30, batch 41450, giga_loss[loss=0.2769, simple_loss=0.3399, pruned_loss=0.107, over 28659.00 frames. ], tot_loss[loss=0.33, simple_loss=0.3868, pruned_loss=0.1366, over 5620116.84 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.3579, pruned_loss=0.1112, over 5687056.37 frames. ], giga_tot_loss[loss=0.3319, simple_loss=0.3884, pruned_loss=0.1376, over 5625105.57 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:23:41,431 INFO [optim.py:369] (1/2) 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 17:24:07,265 INFO [train.py:968] (1/2) Epoch 30, batch 41500, giga_loss[loss=0.3019, simple_loss=0.3698, pruned_loss=0.117, over 28705.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3864, pruned_loss=0.1363, over 5628778.15 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3581, pruned_loss=0.1114, over 5693618.44 frames. ], giga_tot_loss[loss=0.3321, simple_loss=0.3884, pruned_loss=0.1379, over 5624802.93 frames. ], batch size: 284, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:24:33,577 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1361996.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:24:36,728 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1361999.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:24:40,353 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0389, 2.2792, 2.2520, 1.7651], device='cuda:1'), covar=tensor([0.3090, 0.2818, 0.2769, 0.3190], device='cuda:1'), in_proj_covar=tensor([0.2095, 0.2059, 0.1964, 0.2110], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 17:25:00,476 INFO [train.py:968] (1/2) Epoch 30, batch 41550, giga_loss[loss=0.3612, simple_loss=0.4172, pruned_loss=0.1526, over 28795.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3861, pruned_loss=0.1351, over 5622933.11 frames. ], libri_tot_loss[loss=0.2902, simple_loss=0.358, pruned_loss=0.1112, over 5696378.63 frames. ], giga_tot_loss[loss=0.3315, simple_loss=0.3886, pruned_loss=0.1372, over 5615745.52 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:25:02,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:25:06,404 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:25:10,246 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:25:24,156 INFO [optim.py:369] (1/2) 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 17:25:25,236 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.8670, 3.7259, 3.5600, 1.9015], device='cuda:1'), covar=tensor([0.0677, 0.0794, 0.0768, 0.2311], device='cuda:1'), in_proj_covar=tensor([0.1338, 0.1234, 0.1036, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 17:25:31,840 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.85 vs. limit=2.0
+2023-03-15 17:25:53,527 INFO [train.py:968] (1/2) Epoch 30, batch 41600, giga_loss[loss=0.3003, simple_loss=0.3768, pruned_loss=0.1119, over 28938.00 frames. ], tot_loss[loss=0.3341, simple_loss=0.3907, pruned_loss=0.1388, over 5602587.53 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3586, pruned_loss=0.1118, over 5678099.70 frames. ], giga_tot_loss[loss=0.3362, simple_loss=0.3924, pruned_loss=0.14, over 5612842.52 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:25:59,518 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.2704, 4.1030, 3.9152, 2.0454], device='cuda:1'), covar=tensor([0.0847, 0.1020, 0.1091, 0.2038], device='cuda:1'), in_proj_covar=tensor([0.1336, 0.1232, 0.1034, 0.0764], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 17:26:49,119 INFO [train.py:968] (1/2) Epoch 30, batch 41650, giga_loss[loss=0.3225, simple_loss=0.3851, pruned_loss=0.1299, over 28369.00 frames. ], tot_loss[loss=0.3324, simple_loss=0.3892, pruned_loss=0.1378, over 5581247.86 frames. ], libri_tot_loss[loss=0.2911, simple_loss=0.3585, pruned_loss=0.1118, over 5669477.62 frames. ], giga_tot_loss[loss=0.3348, simple_loss=0.3911, pruned_loss=0.1392, over 5595698.71 frames. ], batch size: 369, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:26:56,340 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.24 vs. limit=2.0
+2023-03-15 17:27:11,053 INFO [optim.py:369] (1/2) 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 17:27:36,752 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:27:36,785 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:27:39,297 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:27:40,186 INFO [train.py:968] (1/2) Epoch 30, batch 41700, giga_loss[loss=0.3288, simple_loss=0.3925, pruned_loss=0.1325, over 28913.00 frames. ], tot_loss[loss=0.3266, simple_loss=0.3858, pruned_loss=0.1337, over 5593009.00 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3587, pruned_loss=0.112, over 5665363.73 frames. ], giga_tot_loss[loss=0.3292, simple_loss=0.3879, pruned_loss=0.1353, over 5606178.19 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:27:42,520 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:28:07,713 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:28:12,198 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:28:29,245 INFO [train.py:968] (1/2) Epoch 30, batch 41750, giga_loss[loss=0.3225, simple_loss=0.3813, pruned_loss=0.1318, over 27592.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.3835, pruned_loss=0.1306, over 5615963.98 frames. ], libri_tot_loss[loss=0.2912, simple_loss=0.3585, pruned_loss=0.1119, over 5671753.18 frames. ], giga_tot_loss[loss=0.3255, simple_loss=0.3861, pruned_loss=0.1325, over 5619297.17 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:28:52,786 INFO [optim.py:369] (1/2) 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 17:29:05,238 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-15 17:29:09,630 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8730, 1.1314, 2.8666, 2.7011], device='cuda:1'), covar=tensor([0.1695, 0.2646, 0.0613, 0.1207], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0681, 0.1020, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:29:15,766 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=2.63 vs. limit=5.0
+2023-03-15 17:29:20,617 INFO [train.py:968] (1/2) Epoch 30, batch 41800, libri_loss[loss=0.2773, simple_loss=0.3477, pruned_loss=0.1035, over 29523.00 frames. ], tot_loss[loss=0.3177, simple_loss=0.3797, pruned_loss=0.1278, over 5615881.94 frames. ], libri_tot_loss[loss=0.2914, simple_loss=0.3586, pruned_loss=0.1121, over 5672470.69 frames. ], giga_tot_loss[loss=0.3205, simple_loss=0.382, pruned_loss=0.1295, over 5616782.87 frames. ], batch size: 83, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:29:58,931 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:30:01,959 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 41850, giga_loss[loss=0.2718, simple_loss=0.3459, pruned_loss=0.09882, over 28858.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3763, pruned_loss=0.125, over 5622952.54 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.358, pruned_loss=0.1119, over 5675233.45 frames. ], giga_tot_loss[loss=0.3166, simple_loss=0.3793, pruned_loss=0.127, over 5619631.91 frames. ], batch size: 174, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:30:34,082 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:30:34,422 INFO [optim.py:369] (1/2) 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 17:31:01,445 INFO [train.py:968] (1/2) Epoch 30, batch 41900, giga_loss[loss=0.2664, simple_loss=0.345, pruned_loss=0.09388, over 29076.00 frames. ], tot_loss[loss=0.3101, simple_loss=0.374, pruned_loss=0.1231, over 5637546.66 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3579, pruned_loss=0.1118, over 5677645.26 frames. ], giga_tot_loss[loss=0.3132, simple_loss=0.3767, pruned_loss=0.1249, over 5632038.84 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:31:04,735 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7596, 1.8336, 1.2799, 1.4805], device='cuda:1'), covar=tensor([0.0999, 0.0654, 0.1116, 0.1200], device='cuda:1'), in_proj_covar=tensor([0.0421, 0.0456, 0.0529, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 17:31:07,920 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:31:51,748 INFO [train.py:968] (1/2) Epoch 30, batch 41950, giga_loss[loss=0.2739, simple_loss=0.3574, pruned_loss=0.09523, over 29024.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3728, pruned_loss=0.1223, over 5648322.37 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3573, pruned_loss=0.1115, over 5681280.65 frames. ], giga_tot_loss[loss=0.3123, simple_loss=0.3759, pruned_loss=0.1243, over 5639496.69 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:32:16,274 INFO [optim.py:369] (1/2) 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 17:32:20,466 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:32:28,372 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8038, 2.9115, 1.7943, 1.0001], device='cuda:1'), covar=tensor([1.0368, 0.3873, 0.5111, 0.8789], device='cuda:1'), in_proj_covar=tensor([0.1883, 0.1767, 0.1678, 0.1528], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 17:32:44,939 INFO [train.py:968] (1/2) Epoch 30, batch 42000, giga_loss[loss=0.3113, simple_loss=0.3683, pruned_loss=0.1272, over 27544.00 frames. ], tot_loss[loss=0.3077, simple_loss=0.372, pruned_loss=0.1217, over 5636875.07 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3575, pruned_loss=0.1116, over 5682255.28 frames. ], giga_tot_loss[loss=0.3109, simple_loss=0.3747, pruned_loss=0.1236, over 5627424.54 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:32:44,940 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 17:32:54,621 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 17:33:25,094 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.73 vs. limit=2.0
+2023-03-15 17:33:49,771 INFO [train.py:968] (1/2) Epoch 30, batch 42050, giga_loss[loss=0.269, simple_loss=0.3624, pruned_loss=0.0878, over 28679.00 frames. ], tot_loss[loss=0.3021, simple_loss=0.3694, pruned_loss=0.1174, over 5633504.52 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3574, pruned_loss=0.1116, over 5673047.21 frames. ], giga_tot_loss[loss=0.305, simple_loss=0.3719, pruned_loss=0.119, over 5632986.58 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:34:10,469 INFO [optim.py:369] (1/2) 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 17:34:28,351 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.4179, 1.2544, 4.3000, 3.3426], device='cuda:1'), covar=tensor([0.1696, 0.2930, 0.0445, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0682, 0.1023, 0.0998], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:34:42,227 INFO [train.py:968] (1/2) Epoch 30, batch 42100, giga_loss[loss=0.3199, simple_loss=0.3859, pruned_loss=0.1269, over 29044.00 frames. ], tot_loss[loss=0.3018, simple_loss=0.371, pruned_loss=0.1163, over 5653790.55 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3568, pruned_loss=0.1112, over 5676540.29 frames. ], giga_tot_loss[loss=0.3048, simple_loss=0.3737, pruned_loss=0.1179, over 5649924.57 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:34:49,053 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6375, 1.8580, 2.0104, 1.5975], device='cuda:1'), covar=tensor([0.3036, 0.2626, 0.2433, 0.2828], device='cuda:1'), in_proj_covar=tensor([0.2105, 0.2064, 0.1969, 0.2121], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 17:35:03,868 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([4.5284, 4.3705, 4.1690, 2.1853], device='cuda:1'), covar=tensor([0.0579, 0.0708, 0.0776, 0.2128], device='cuda:1'), in_proj_covar=tensor([0.1340, 0.1235, 0.1035, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 17:35:33,051 INFO [train.py:968] (1/2) Epoch 30, batch 42150, giga_loss[loss=0.2834, simple_loss=0.3616, pruned_loss=0.1026, over 29061.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3726, pruned_loss=0.1175, over 5660580.22 frames. ], libri_tot_loss[loss=0.2898, simple_loss=0.357, pruned_loss=0.1113, over 5680258.66 frames. ], giga_tot_loss[loss=0.3062, simple_loss=0.3748, pruned_loss=0.1188, over 5653905.24 frames. ], batch size: 155, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:35:54,001 INFO [optim.py:369] (1/2) 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 17:35:56,177 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3073, 1.3708, 3.3729, 3.1304], device='cuda:1'), covar=tensor([0.1523, 0.2607, 0.0523, 0.1062], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0682, 0.1022, 0.0997], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:36:19,360 INFO [train.py:968] (1/2) Epoch 30, batch 42200, giga_loss[loss=0.3148, simple_loss=0.3658, pruned_loss=0.132, over 23465.00 frames. ], tot_loss[loss=0.3027, simple_loss=0.3712, pruned_loss=0.1171, over 5660410.42 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3567, pruned_loss=0.1111, over 5684105.24 frames. ], giga_tot_loss[loss=0.3053, simple_loss=0.3736, pruned_loss=0.1185, over 5651275.50 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:36:26,219 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:36:58,805 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.19 vs. limit=2.0
+2023-03-15 17:37:01,906 INFO [train.py:968] (1/2) Epoch 30, batch 42250, giga_loss[loss=0.2927, simple_loss=0.3559, pruned_loss=0.1148, over 28986.00 frames. ], tot_loss[loss=0.3025, simple_loss=0.3702, pruned_loss=0.1174, over 5663574.92 frames. ], libri_tot_loss[loss=0.2901, simple_loss=0.3571, pruned_loss=0.1116, over 5677759.11 frames. ], giga_tot_loss[loss=0.3045, simple_loss=0.3723, pruned_loss=0.1183, over 5660981.21 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:37:19,016 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6116, 1.6037, 1.7994, 1.3908], device='cuda:1'), covar=tensor([0.1656, 0.2481, 0.1382, 0.1700], device='cuda:1'), in_proj_covar=tensor([0.0942, 0.0720, 0.0992, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 17:37:25,727 INFO [optim.py:369] (1/2) 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 17:37:33,672 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 42300, giga_loss[loss=0.2602, simple_loss=0.3404, pruned_loss=0.08996, over 28558.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.368, pruned_loss=0.1174, over 5662622.14 frames. ], libri_tot_loss[loss=0.2899, simple_loss=0.3568, pruned_loss=0.1114, over 5682749.07 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3703, pruned_loss=0.1184, over 5655582.37 frames. ], batch size: 78, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:38:27,185 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1362808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:38:38,938 INFO [train.py:968] (1/2) Epoch 30, batch 42350, giga_loss[loss=0.3057, simple_loss=0.3808, pruned_loss=0.1154, over 28546.00 frames. ], tot_loss[loss=0.3016, simple_loss=0.368, pruned_loss=0.1176, over 5639060.60 frames. ], libri_tot_loss[loss=0.2904, simple_loss=0.3573, pruned_loss=0.1118, over 5658739.97 frames. ], giga_tot_loss[loss=0.3032, simple_loss=0.3698, pruned_loss=0.1183, over 5655423.77 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 17:38:40,649 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1362820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:38:49,536 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1712, 1.6069, 1.5982, 1.3473], device='cuda:1'), covar=tensor([0.2022, 0.1518, 0.2199, 0.1758], device='cuda:1'), in_proj_covar=tensor([0.0515, 0.0767, 0.0736, 0.0709], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 17:38:53,244 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8659, 1.1560, 2.8307, 2.7559], device='cuda:1'), covar=tensor([0.1707, 0.2669, 0.0624, 0.1380], device='cuda:1'), in_proj_covar=tensor([0.0819, 0.0681, 0.1020, 0.0996], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:39:03,263 INFO [optim.py:369] (1/2) 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 17:39:27,503 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([0.7665, 0.9777, 0.8313, 0.2939], device='cuda:1'), covar=tensor([0.3099, 0.2978, 0.3137, 0.4965], device='cuda:1'), in_proj_covar=tensor([0.1880, 0.1769, 0.1678, 0.1527], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 17:39:27,906 INFO [train.py:968] (1/2) Epoch 30, batch 42400, libri_loss[loss=0.3835, simple_loss=0.4252, pruned_loss=0.1709, over 20076.00 frames. ], tot_loss[loss=0.2999, simple_loss=0.3677, pruned_loss=0.1161, over 5649719.55 frames. ], libri_tot_loss[loss=0.2907, simple_loss=0.3575, pruned_loss=0.1119, over 5658237.70 frames. ], giga_tot_loss[loss=0.3013, simple_loss=0.3692, pruned_loss=0.1167, over 5663824.61 frames. ], batch size: 187, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:39:39,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.59 vs. limit=5.0
+2023-03-15 17:39:51,137 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:39:54,397 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362895.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:40:14,878 INFO [train.py:968] (1/2) Epoch 30, batch 42450, giga_loss[loss=0.2841, simple_loss=0.3557, pruned_loss=0.1062, over 28625.00 frames. ], tot_loss[loss=0.2994, simple_loss=0.3676, pruned_loss=0.1156, over 5662087.62 frames. ], libri_tot_loss[loss=0.2909, simple_loss=0.3576, pruned_loss=0.112, over 5659406.01 frames. ], giga_tot_loss[loss=0.3006, simple_loss=0.369, pruned_loss=0.1161, over 5672256.99 frames. ], batch size: 85, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:40:21,383 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:40:40,932 INFO [optim.py:369] (1/2) 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 17:41:02,695 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5687, 1.6202, 1.3204, 1.1996], device='cuda:1'), covar=tensor([0.0860, 0.0423, 0.0840, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0454, 0.0527, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 17:41:03,279 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1362963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:41:04,034 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3956, 1.6200, 1.6545, 1.4740], device='cuda:1'), covar=tensor([0.2160, 0.2118, 0.2580, 0.2280], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0764, 0.0734, 0.0707], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 17:41:05,448 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1362966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:41:06,404 INFO [train.py:968] (1/2) Epoch 30, batch 42500, giga_loss[loss=0.3153, simple_loss=0.3768, pruned_loss=0.1269, over 28334.00 frames. ], tot_loss[loss=0.3002, simple_loss=0.368, pruned_loss=0.1162, over 5654146.91 frames. ], libri_tot_loss[loss=0.2908, simple_loss=0.3576, pruned_loss=0.112, over 5659534.45 frames. ], giga_tot_loss[loss=0.3014, simple_loss=0.3693, pruned_loss=0.1167, over 5662090.52 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:41:34,766 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1362995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:41:55,582 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.26 vs. limit=2.0
+2023-03-15 17:41:55,711 INFO [train.py:968] (1/2) Epoch 30, batch 42550, giga_loss[loss=0.2916, simple_loss=0.3574, pruned_loss=0.1129, over 28959.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3658, pruned_loss=0.1152, over 5657814.12 frames. ], libri_tot_loss[loss=0.2903, simple_loss=0.3572, pruned_loss=0.1117, over 5653943.05 frames. ], giga_tot_loss[loss=0.2997, simple_loss=0.3674, pruned_loss=0.116, over 5669401.31 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:42:03,517 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-15 17:42:16,764 INFO [optim.py:369] (1/2) 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 17:42:24,025 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7225, 1.6163, 1.9178, 1.5191], device='cuda:1'), covar=tensor([0.1534, 0.2118, 0.1283, 0.1565], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0722, 0.0994, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 17:42:29,090 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:42:42,978 INFO [train.py:968] (1/2) Epoch 30, batch 42600, giga_loss[loss=0.2801, simple_loss=0.3429, pruned_loss=0.1087, over 28773.00 frames. ], tot_loss[loss=0.2978, simple_loss=0.3652, pruned_loss=0.1152, over 5660949.76 frames. ], libri_tot_loss[loss=0.2905, simple_loss=0.3575, pruned_loss=0.1118, over 5659357.32 frames. ], giga_tot_loss[loss=0.299, simple_loss=0.3664, pruned_loss=0.1158, over 5665532.69 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:43:32,551 INFO [train.py:968] (1/2) Epoch 30, batch 42650, giga_loss[loss=0.3401, simple_loss=0.3931, pruned_loss=0.1436, over 27600.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3633, pruned_loss=0.1147, over 5671598.55 frames. ], libri_tot_loss[loss=0.2906, simple_loss=0.3575, pruned_loss=0.1118, over 5665837.72 frames. ], giga_tot_loss[loss=0.2975, simple_loss=0.3646, pruned_loss=0.1152, over 5669485.42 frames. ], batch size: 472, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:43:58,291 INFO [optim.py:369] (1/2) 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 17:44:04,896 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 17:44:10,435 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8787, 1.9739, 1.7523, 1.7840], device='cuda:1'), covar=tensor([0.2830, 0.3025, 0.3341, 0.2862], device='cuda:1'), in_proj_covar=tensor([0.1633, 0.1172, 0.1441, 0.1020], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 17:44:19,952 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.17 vs. limit=2.0
+2023-03-15 17:44:27,290 INFO [train.py:968] (1/2) Epoch 30, batch 42700, giga_loss[loss=0.3314, simple_loss=0.375, pruned_loss=0.1439, over 26874.00 frames. ], tot_loss[loss=0.2959, simple_loss=0.3623, pruned_loss=0.1147, over 5676018.99 frames. ], libri_tot_loss[loss=0.29, simple_loss=0.357, pruned_loss=0.1115, over 5667215.07 frames. ], giga_tot_loss[loss=0.2973, simple_loss=0.3638, pruned_loss=0.1154, over 5673200.66 frames. ], batch size: 555, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:44:41,564 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:44:52,303 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.78 vs. limit=2.0
+2023-03-15 17:44:56,117 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:45:14,493 INFO [train.py:968] (1/2) Epoch 30, batch 42750, giga_loss[loss=0.3329, simple_loss=0.3956, pruned_loss=0.1351, over 28874.00 frames. ], tot_loss[loss=0.2943, simple_loss=0.3606, pruned_loss=0.114, over 5677587.08 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3564, pruned_loss=0.1111, over 5671156.35 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.3625, pruned_loss=0.1151, over 5671973.10 frames. ], batch size: 285, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:45:17,540 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5532, 1.5951, 1.7189, 1.3392], device='cuda:1'), covar=tensor([0.1727, 0.2598, 0.1479, 0.1741], device='cuda:1'), in_proj_covar=tensor([0.0943, 0.0722, 0.0995, 0.0893], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 17:45:24,041 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 17:45:38,855 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 42800, giga_loss[loss=0.2536, simple_loss=0.3344, pruned_loss=0.08641, over 28806.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3611, pruned_loss=0.115, over 5654190.55 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3562, pruned_loss=0.1109, over 5664918.28 frames. ], giga_tot_loss[loss=0.2976, simple_loss=0.3629, pruned_loss=0.1162, over 5654882.55 frames. ], batch size: 66, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:46:29,106 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.1526, 1.2792, 1.1447, 0.9024], device='cuda:1'), covar=tensor([0.1182, 0.0569, 0.1136, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0419, 0.0454, 0.0528, 0.0468], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 17:46:36,040 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.0094, 2.3869, 2.2271, 1.7172], device='cuda:1'), covar=tensor([0.3725, 0.2575, 0.2965, 0.3655], device='cuda:1'), in_proj_covar=tensor([0.2103, 0.2060, 0.1968, 0.2117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 17:46:47,817 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1363311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:46:47,861 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([6.0454, 5.8447, 5.5584, 3.1866], device='cuda:1'), covar=tensor([0.0548, 0.0702, 0.0743, 0.1472], device='cuda:1'), in_proj_covar=tensor([0.1341, 0.1238, 0.1035, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 17:46:52,487 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.2943, 5.1245, 4.8782, 2.6350], device='cuda:1'), covar=tensor([0.0528, 0.0659, 0.0716, 0.1690], device='cuda:1'), in_proj_covar=tensor([0.1341, 0.1238, 0.1035, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 17:46:53,571 INFO [train.py:968] (1/2) Epoch 30, batch 42850, libri_loss[loss=0.3107, simple_loss=0.3788, pruned_loss=0.1213, over 29121.00 frames. ], tot_loss[loss=0.2951, simple_loss=0.3614, pruned_loss=0.1144, over 5659261.15 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3562, pruned_loss=0.1108, over 5665715.34 frames. ], giga_tot_loss[loss=0.2969, simple_loss=0.3629, pruned_loss=0.1154, over 5658916.45 frames. ], batch size: 101, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:47:01,170 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:47:05,039 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:47:17,608 INFO [optim.py:369] (1/2) 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 17:47:31,133 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.3771, 1.6673, 1.4579, 1.6124], device='cuda:1'), covar=tensor([0.0768, 0.0348, 0.0333, 0.0854], device='cuda:1'), in_proj_covar=tensor([0.0195, 0.0123, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 17:47:32,430 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:47:41,009 INFO [train.py:968] (1/2) Epoch 30, batch 42900, libri_loss[loss=0.3382, simple_loss=0.3873, pruned_loss=0.1446, over 19028.00 frames. ], tot_loss[loss=0.2952, simple_loss=0.3622, pruned_loss=0.1141, over 5648617.86 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3563, pruned_loss=0.1109, over 5648478.49 frames. ], giga_tot_loss[loss=0.2966, simple_loss=0.3633, pruned_loss=0.1149, over 5665711.53 frames. ], batch size: 187, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:48:18,290 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1363406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:48:29,876 INFO [train.py:968] (1/2) Epoch 30, batch 42950, giga_loss[loss=0.2904, simple_loss=0.3652, pruned_loss=0.1078, over 28978.00 frames. ], tot_loss[loss=0.2953, simple_loss=0.3629, pruned_loss=0.1138, over 5658727.88 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3561, pruned_loss=0.1108, over 5652179.27 frames. ], giga_tot_loss[loss=0.2965, simple_loss=0.364, pruned_loss=0.1145, over 5669228.55 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:48:55,336 INFO [optim.py:369] (1/2) 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] (1/2) Epoch 30, batch 43000, giga_loss[loss=0.3043, simple_loss=0.3705, pruned_loss=0.1191, over 28627.00 frames. ], tot_loss[loss=0.2954, simple_loss=0.3631, pruned_loss=0.1139, over 5667061.39 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3563, pruned_loss=0.1108, over 5653572.60 frames. ], giga_tot_loss[loss=0.2964, simple_loss=0.364, pruned_loss=0.1144, over 5674132.82 frames. ], batch size: 307, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:49:33,203 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.23 vs. limit=2.0
+2023-03-15 17:50:12,187 INFO [train.py:968] (1/2) Epoch 30, batch 43050, giga_loss[loss=0.2818, simple_loss=0.3481, pruned_loss=0.1077, over 29106.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3667, pruned_loss=0.1169, over 5671374.82 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3565, pruned_loss=0.111, over 5652242.07 frames. ], giga_tot_loss[loss=0.3012, simple_loss=0.3675, pruned_loss=0.1174, over 5678243.66 frames. ], batch size: 128, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:50:16,860 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.4685, 1.7380, 1.4741, 1.6208], device='cuda:1'), covar=tensor([0.0769, 0.0325, 0.0326, 0.0860], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0124, 0.0122, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 17:50:24,646 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2334, 1.5047, 1.5032, 1.3486], device='cuda:1'), covar=tensor([0.1738, 0.1420, 0.2063, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.0512, 0.0763, 0.0731, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 17:50:37,546 INFO [optim.py:369] (1/2) 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 17:51:10,949 INFO [train.py:968] (1/2) Epoch 30, batch 43100, libri_loss[loss=0.2541, simple_loss=0.3311, pruned_loss=0.08851, over 29552.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3679, pruned_loss=0.1191, over 5678549.73 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3564, pruned_loss=0.1108, over 5654553.52 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3688, pruned_loss=0.1197, over 5682053.39 frames. ], batch size: 79, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:51:24,254 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7975, 1.1433, 2.8782, 2.8442], device='cuda:1'), covar=tensor([0.1795, 0.2576, 0.0640, 0.0916], device='cuda:1'), in_proj_covar=tensor([0.0822, 0.0683, 0.1023, 0.0999], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:51:54,061 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5864, 1.8261, 1.2763, 1.3648], device='cuda:1'), covar=tensor([0.1043, 0.0601, 0.1082, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0418, 0.0452, 0.0526, 0.0466], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0009, 0.0011, 0.0010], device='cuda:1')
+2023-03-15 17:52:01,601 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 43150, giga_loss[loss=0.289, simple_loss=0.3587, pruned_loss=0.1097, over 28722.00 frames. ], tot_loss[loss=0.3062, simple_loss=0.3695, pruned_loss=0.1215, over 5672266.65 frames. ], libri_tot_loss[loss=0.2889, simple_loss=0.3564, pruned_loss=0.1107, over 5653974.18 frames. ], giga_tot_loss[loss=0.3074, simple_loss=0.3704, pruned_loss=0.1222, over 5675908.24 frames. ], batch size: 99, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:52:23,725 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.8366, 5.6823, 5.3587, 3.0083], device='cuda:1'), covar=tensor([0.0503, 0.0701, 0.0697, 0.1555], device='cuda:1'), in_proj_covar=tensor([0.1341, 0.1238, 0.1036, 0.0765], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 17:52:29,213 INFO [optim.py:369] (1/2) 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 17:52:54,705 INFO [train.py:968] (1/2) Epoch 30, batch 43200, giga_loss[loss=0.3182, simple_loss=0.3849, pruned_loss=0.1257, over 28757.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3719, pruned_loss=0.1236, over 5648178.37 frames. ], libri_tot_loss[loss=0.2896, simple_loss=0.3569, pruned_loss=0.1111, over 5645700.81 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3723, pruned_loss=0.1239, over 5658883.93 frames. ], batch size: 242, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:53:10,922 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:53:28,678 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1363706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:53:39,167 INFO [train.py:968] (1/2) Epoch 30, batch 43250, giga_loss[loss=0.3439, simple_loss=0.3854, pruned_loss=0.1512, over 23879.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1218, over 5652735.28 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3569, pruned_loss=0.111, over 5649234.26 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3702, pruned_loss=0.1225, over 5658055.84 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:53:44,308 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-15 17:53:59,464 INFO [optim.py:369] (1/2) 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 17:54:18,632 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:54:20,675 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:54:25,126 INFO [train.py:968] (1/2) Epoch 30, batch 43300, giga_loss[loss=0.2858, simple_loss=0.3628, pruned_loss=0.1044, over 28879.00 frames. ], tot_loss[loss=0.3045, simple_loss=0.3692, pruned_loss=0.1199, over 5664134.83 frames. ], libri_tot_loss[loss=0.2893, simple_loss=0.3569, pruned_loss=0.1108, over 5653912.56 frames. ], giga_tot_loss[loss=0.3058, simple_loss=0.3699, pruned_loss=0.1208, over 5664647.58 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:54:38,258 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1363781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:54:50,161 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363792.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:55:16,779 INFO [train.py:968] (1/2) Epoch 30, batch 43350, giga_loss[loss=0.3113, simple_loss=0.3486, pruned_loss=0.137, over 23932.00 frames. ], tot_loss[loss=0.3004, simple_loss=0.3667, pruned_loss=0.1171, over 5660183.70 frames. ], libri_tot_loss[loss=0.2892, simple_loss=0.357, pruned_loss=0.1107, over 5658867.25 frames. ], giga_tot_loss[loss=0.3017, simple_loss=0.3674, pruned_loss=0.118, over 5656266.17 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:55:18,427 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=48, metric=1.25 vs. limit=2.0
+2023-03-15 17:55:26,456 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:55:29,437 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:55:38,317 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.8634, 2.2726, 1.7521, 1.9293], device='cuda:1'), covar=tensor([0.2528, 0.2575, 0.2982, 0.2601], device='cuda:1'), in_proj_covar=tensor([0.1631, 0.1170, 0.1438, 0.1018], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 17:55:38,673 INFO [optim.py:369] (1/2) 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 17:55:59,081 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1363861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:56:04,489 INFO [train.py:968] (1/2) Epoch 30, batch 43400, giga_loss[loss=0.2846, simple_loss=0.3564, pruned_loss=0.1064, over 28864.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3647, pruned_loss=0.1162, over 5667611.64 frames. ], libri_tot_loss[loss=0.2894, simple_loss=0.3571, pruned_loss=0.1108, over 5658809.40 frames. ], giga_tot_loss[loss=0.2996, simple_loss=0.3653, pruned_loss=0.1169, over 5664430.46 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:56:51,454 INFO [train.py:968] (1/2) Epoch 30, batch 43450, giga_loss[loss=0.3613, simple_loss=0.3851, pruned_loss=0.1687, over 23585.00 frames. ], tot_loss[loss=0.2985, simple_loss=0.3638, pruned_loss=0.1166, over 5663250.77 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.357, pruned_loss=0.1106, over 5661611.41 frames. ], giga_tot_loss[loss=0.2998, simple_loss=0.3646, pruned_loss=0.1175, over 5658587.71 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:56:57,623 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1363924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:56:59,585 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1363927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 17:57:15,369 INFO [optim.py:369] (1/2) 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 17:57:27,757 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 43500, giga_loss[loss=0.3094, simple_loss=0.3761, pruned_loss=0.1214, over 27926.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.363, pruned_loss=0.1162, over 5672761.49 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3565, pruned_loss=0.1104, over 5667685.82 frames. ], giga_tot_loss[loss=0.2994, simple_loss=0.3643, pruned_loss=0.1173, over 5663907.94 frames. ], batch size: 412, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:58:26,347 INFO [train.py:968] (1/2) Epoch 30, batch 43550, giga_loss[loss=0.2941, simple_loss=0.37, pruned_loss=0.1091, over 28687.00 frames. ], tot_loss[loss=0.3022, simple_loss=0.3673, pruned_loss=0.1185, over 5672983.84 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3568, pruned_loss=0.1106, over 5670026.70 frames. ], giga_tot_loss[loss=0.3036, simple_loss=0.3683, pruned_loss=0.1194, over 5663754.27 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 17:58:50,399 INFO [optim.py:369] (1/2) 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 17:59:05,146 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.7886, 1.3252, 4.7934, 3.6637], device='cuda:1'), covar=tensor([0.1561, 0.2947, 0.0473, 0.1032], device='cuda:1'), in_proj_covar=tensor([0.0823, 0.0683, 0.1024, 0.1000], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0011, 0.0012], device='cuda:1')
+2023-03-15 17:59:14,333 INFO [train.py:968] (1/2) Epoch 30, batch 43600, giga_loss[loss=0.3093, simple_loss=0.3872, pruned_loss=0.1157, over 28380.00 frames. ], tot_loss[loss=0.3017, simple_loss=0.3693, pruned_loss=0.1171, over 5669110.40 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3565, pruned_loss=0.1104, over 5665169.81 frames. ], giga_tot_loss[loss=0.3034, simple_loss=0.3706, pruned_loss=0.1181, over 5666449.16 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 17:59:31,249 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1364081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:00:07,191 INFO [train.py:968] (1/2) Epoch 30, batch 43650, giga_loss[loss=0.3516, simple_loss=0.4184, pruned_loss=0.1424, over 28318.00 frames. ], tot_loss[loss=0.302, simple_loss=0.3699, pruned_loss=0.117, over 5667254.91 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3562, pruned_loss=0.1102, over 5666408.32 frames. ], giga_tot_loss[loss=0.3041, simple_loss=0.3716, pruned_loss=0.1183, over 5663725.77 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:00:33,058 INFO [optim.py:369] (1/2) 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 18:00:57,623 INFO [train.py:968] (1/2) Epoch 30, batch 43700, giga_loss[loss=0.3316, simple_loss=0.3855, pruned_loss=0.1389, over 28276.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3718, pruned_loss=0.1187, over 5670678.15 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3556, pruned_loss=0.1097, over 5671738.35 frames. ], giga_tot_loss[loss=0.3075, simple_loss=0.3742, pruned_loss=0.1204, over 5663068.15 frames. ], batch size: 368, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:01:29,178 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364200.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:01:47,608 INFO [train.py:968] (1/2) Epoch 30, batch 43750, giga_loss[loss=0.2819, simple_loss=0.3592, pruned_loss=0.1023, over 28912.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3728, pruned_loss=0.1202, over 5664827.16 frames. ], libri_tot_loss[loss=0.2878, simple_loss=0.3558, pruned_loss=0.1099, over 5675277.43 frames. ], giga_tot_loss[loss=0.3088, simple_loss=0.3747, pruned_loss=0.1214, over 5655562.95 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:01:52,889 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1364224.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:01:54,701 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364227.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:02:08,050 INFO [optim.py:369] (1/2) 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 18:02:19,191 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 43800, giga_loss[loss=0.256, simple_loss=0.3345, pruned_loss=0.08878, over 28246.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3726, pruned_loss=0.1209, over 5674988.85 frames. ], libri_tot_loss[loss=0.2883, simple_loss=0.3562, pruned_loss=0.1102, over 5681705.24 frames. ], giga_tot_loss[loss=0.3093, simple_loss=0.3745, pruned_loss=0.122, over 5661348.44 frames. ], batch size: 77, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:02:45,149 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:02:54,944 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-15 18:03:09,817 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.3029, 1.2727, 3.8799, 3.2418], device='cuda:1'), covar=tensor([0.1714, 0.2862, 0.0482, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0826, 0.0685, 0.1026, 0.1004], device='cuda:1'), out_proj_covar=tensor([0.0011, 0.0010, 0.0012, 0.0012], device='cuda:1')
+2023-03-15 18:03:17,915 INFO [train.py:968] (1/2) Epoch 30, batch 43850, libri_loss[loss=0.3131, simple_loss=0.3787, pruned_loss=0.1238, over 29676.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3719, pruned_loss=0.1212, over 5663138.04 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3569, pruned_loss=0.1106, over 5673463.97 frames. ], giga_tot_loss[loss=0.3087, simple_loss=0.3732, pruned_loss=0.1221, over 5659357.27 frames. ], batch size: 91, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:03:40,784 INFO [optim.py:369] (1/2) 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 18:03:45,926 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 43900, giga_loss[loss=0.2484, simple_loss=0.3191, pruned_loss=0.08891, over 28950.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3687, pruned_loss=0.1194, over 5667290.72 frames. ], libri_tot_loss[loss=0.2888, simple_loss=0.3566, pruned_loss=0.1105, over 5673310.09 frames. ], giga_tot_loss[loss=0.3059, simple_loss=0.3705, pruned_loss=0.1206, over 5663584.55 frames. ], batch size: 86, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:04:23,227 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6364, 1.6086, 1.8456, 1.4357], device='cuda:1'), covar=tensor([0.1740, 0.2598, 0.1410, 0.1731], device='cuda:1'), in_proj_covar=tensor([0.0944, 0.0722, 0.0994, 0.0892], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 18:04:52,043 INFO [train.py:968] (1/2) Epoch 30, batch 43950, giga_loss[loss=0.2824, simple_loss=0.3525, pruned_loss=0.1062, over 28741.00 frames. ], tot_loss[loss=0.3035, simple_loss=0.3679, pruned_loss=0.1195, over 5669977.75 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.3569, pruned_loss=0.1106, over 5678510.35 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3695, pruned_loss=0.1207, over 5662389.24 frames. ], batch size: 60, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:05:12,301 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-15 18:05:22,393 INFO [optim.py:369] (1/2) 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 18:05:46,234 INFO [train.py:968] (1/2) Epoch 30, batch 44000, giga_loss[loss=0.2972, simple_loss=0.3702, pruned_loss=0.1121, over 28840.00 frames. ], tot_loss[loss=0.3052, simple_loss=0.3691, pruned_loss=0.1206, over 5679510.00 frames. ], libri_tot_loss[loss=0.2887, simple_loss=0.3566, pruned_loss=0.1104, over 5679754.38 frames. ], giga_tot_loss[loss=0.3073, simple_loss=0.3709, pruned_loss=0.1219, over 5672141.91 frames. ], batch size: 199, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:06:36,121 INFO [train.py:968] (1/2) Epoch 30, batch 44050, giga_loss[loss=0.2995, simple_loss=0.3663, pruned_loss=0.1164, over 28925.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3696, pruned_loss=0.1217, over 5663690.06 frames. ], libri_tot_loss[loss=0.2885, simple_loss=0.3565, pruned_loss=0.1103, over 5675566.94 frames. ], giga_tot_loss[loss=0.3089, simple_loss=0.3715, pruned_loss=0.1231, over 5661821.99 frames. ], batch size: 136, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:06:57,220 INFO [optim.py:369] (1/2) 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 18:07:13,578 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([5.9321, 5.7433, 5.4529, 3.0821], device='cuda:1'), covar=tensor([0.0555, 0.0714, 0.0879, 0.1548], device='cuda:1'), in_proj_covar=tensor([0.1344, 0.1241, 0.1040, 0.0768], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 18:07:20,042 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.7060, 1.9806, 1.3163, 1.6404], device='cuda:1'), covar=tensor([0.1177, 0.0778, 0.1158, 0.1293], device='cuda:1'), in_proj_covar=tensor([0.0420, 0.0456, 0.0527, 0.0469], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 18:07:20,358 INFO [train.py:968] (1/2) Epoch 30, batch 44100, giga_loss[loss=0.2568, simple_loss=0.3397, pruned_loss=0.08697, over 28848.00 frames. ], tot_loss[loss=0.3048, simple_loss=0.3682, pruned_loss=0.1208, over 5659995.47 frames. ], libri_tot_loss[loss=0.289, simple_loss=0.357, pruned_loss=0.1104, over 5669557.68 frames. ], giga_tot_loss[loss=0.3069, simple_loss=0.3696, pruned_loss=0.1221, over 5664351.60 frames. ], batch size: 227, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:07:28,198 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1364575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:08:07,408 INFO [train.py:968] (1/2) Epoch 30, batch 44150, giga_loss[loss=0.2351, simple_loss=0.3122, pruned_loss=0.07901, over 28260.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3668, pruned_loss=0.1197, over 5665223.60 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3568, pruned_loss=0.1102, over 5672909.53 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3683, pruned_loss=0.1212, over 5665630.83 frames. ], batch size: 65, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:08:33,542 INFO [optim.py:369] (1/2) 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 18:08:53,989 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 44200, giga_loss[loss=0.2881, simple_loss=0.362, pruned_loss=0.1071, over 28902.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3675, pruned_loss=0.1192, over 5667400.79 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3563, pruned_loss=0.1099, over 5677684.92 frames. ], giga_tot_loss[loss=0.3054, simple_loss=0.3692, pruned_loss=0.1208, over 5663358.05 frames. ], batch size: 112, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:09:24,009 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:09:24,094 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5197, 1.8224, 1.2327, 1.4147], device='cuda:1'), covar=tensor([0.1135, 0.0592, 0.1123, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0422, 0.0457, 0.0529, 0.0470], device='cuda:1'), out_proj_covar=tensor([0.0009, 0.0010, 0.0011, 0.0011], device='cuda:1')
+2023-03-15 18:09:38,274 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364706.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:09:49,581 INFO [train.py:968] (1/2) Epoch 30, batch 44250, libri_loss[loss=0.3003, simple_loss=0.3697, pruned_loss=0.1155, over 28807.00 frames. ], tot_loss[loss=0.3041, simple_loss=0.3687, pruned_loss=0.1197, over 5677007.64 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3559, pruned_loss=0.1095, over 5684063.06 frames. ], giga_tot_loss[loss=0.3071, simple_loss=0.3708, pruned_loss=0.1217, over 5667693.45 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:09:49,946 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1364718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:09:53,413 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:09:59,590 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1364726.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:10:09,519 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6264, 1.8539, 1.6702, 1.4515], device='cuda:1'), covar=tensor([0.3179, 0.2524, 0.2618, 0.2950], device='cuda:1'), in_proj_covar=tensor([0.2101, 0.2057, 0.1971, 0.2117], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1')
+2023-03-15 18:10:18,543 INFO [optim.py:369] (1/2) 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] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1364750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:10:24,810 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:10:39,861 INFO [train.py:968] (1/2) Epoch 30, batch 44300, giga_loss[loss=0.332, simple_loss=0.3684, pruned_loss=0.1478, over 23811.00 frames. ], tot_loss[loss=0.304, simple_loss=0.3681, pruned_loss=0.12, over 5665868.10 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3559, pruned_loss=0.1095, over 5683930.03 frames. ], giga_tot_loss[loss=0.3068, simple_loss=0.3702, pruned_loss=0.1217, over 5658141.98 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:11:12,910 INFO [zipformer.py:1188] (1/2) 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] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:11:27,050 INFO [train.py:968] (1/2) Epoch 30, batch 44350, giga_loss[loss=0.2629, simple_loss=0.3511, pruned_loss=0.08735, over 28748.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.37, pruned_loss=0.1188, over 5673956.03 frames. ], libri_tot_loss[loss=0.2871, simple_loss=0.3557, pruned_loss=0.1093, over 5689496.32 frames. ], giga_tot_loss[loss=0.3067, simple_loss=0.3721, pruned_loss=0.1206, over 5662602.75 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:11:29,633 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2371, 1.4129, 1.4434, 1.0669], device='cuda:1'), covar=tensor([0.2003, 0.2927, 0.1752, 0.2046], device='cuda:1'), in_proj_covar=tensor([0.0945, 0.0725, 0.0996, 0.0895], device='cuda:1'), out_proj_covar=tensor([0.0016, 0.0016, 0.0017, 0.0015], device='cuda:1')
+2023-03-15 18:11:34,240 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1364826.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:11:41,012 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1364835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:11:48,221 INFO [optim.py:369] (1/2) 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 18:12:08,124 INFO [train.py:968] (1/2) Epoch 30, batch 44400, giga_loss[loss=0.3167, simple_loss=0.3953, pruned_loss=0.119, over 28976.00 frames. ], tot_loss[loss=0.3024, simple_loss=0.3707, pruned_loss=0.117, over 5672852.35 frames. ], libri_tot_loss[loss=0.2877, simple_loss=0.3562, pruned_loss=0.1096, over 5673397.23 frames. ], giga_tot_loss[loss=0.3051, simple_loss=0.3728, pruned_loss=0.1187, over 5677072.65 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:12:09,189 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1364869.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:12:11,527 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1364872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:12:38,334 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1364901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:12:55,312 INFO [train.py:968] (1/2) Epoch 30, batch 44450, giga_loss[loss=0.3209, simple_loss=0.3883, pruned_loss=0.1267, over 28965.00 frames. ], tot_loss[loss=0.305, simple_loss=0.3732, pruned_loss=0.1184, over 5673334.85 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3562, pruned_loss=0.1099, over 5672864.42 frames. ], giga_tot_loss[loss=0.3077, simple_loss=0.3756, pruned_loss=0.1199, over 5677676.77 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:13:09,280 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2591, 2.4323, 2.2512, 2.1099], device='cuda:1'), covar=tensor([0.2259, 0.2395, 0.2294, 0.2623], device='cuda:1'), in_proj_covar=tensor([0.0513, 0.0764, 0.0733, 0.0705], device='cuda:1'), out_proj_covar=tensor([0.0007, 0.0010, 0.0010, 0.0010], device='cuda:1')
+2023-03-15 18:13:10,232 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.5865, 4.8238, 1.8321, 1.8817], device='cuda:1'), covar=tensor([0.1063, 0.0466, 0.0939, 0.1346], device='cuda:1'), in_proj_covar=tensor([0.0434, 0.0584, 0.0419, 0.0454], device='cuda:1'), out_proj_covar=tensor([0.0029, 0.0035, 0.0027, 0.0032], device='cuda:1')
+2023-03-15 18:13:22,038 INFO [optim.py:369] (1/2) 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 18:13:41,298 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-15 18:13:45,299 INFO [train.py:968] (1/2) Epoch 30, batch 44500, giga_loss[loss=0.3427, simple_loss=0.3964, pruned_loss=0.1445, over 28830.00 frames. ], tot_loss[loss=0.3098, simple_loss=0.3764, pruned_loss=0.1216, over 5672080.11 frames. ], libri_tot_loss[loss=0.288, simple_loss=0.3562, pruned_loss=0.1099, over 5674177.04 frames. ], giga_tot_loss[loss=0.3122, simple_loss=0.3786, pruned_loss=0.1229, over 5674667.57 frames. ], batch size: 119, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:14:35,787 INFO [train.py:968] (1/2) Epoch 30, batch 44550, giga_loss[loss=0.3686, simple_loss=0.3993, pruned_loss=0.169, over 23402.00 frames. ], tot_loss[loss=0.3132, simple_loss=0.3781, pruned_loss=0.1242, over 5663567.98 frames. ], libri_tot_loss[loss=0.2874, simple_loss=0.3559, pruned_loss=0.1095, over 5679350.41 frames. ], giga_tot_loss[loss=0.3162, simple_loss=0.3807, pruned_loss=0.1258, over 5661172.65 frames. ], batch size: 705, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:14:36,653 INFO [zipformer.py:1188] (1/2) warmup_begin=666.7, warmup_end=1333.3, batch_count=1365019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:15:04,471 INFO [optim.py:369] (1/2) 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 18:15:21,770 INFO [zipformer.py:1188] (1/2) 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] (1/2) Epoch 30, batch 44600, giga_loss[loss=0.2852, simple_loss=0.3547, pruned_loss=0.1078, over 28971.00 frames. ], tot_loss[loss=0.314, simple_loss=0.378, pruned_loss=0.125, over 5651192.03 frames. ], libri_tot_loss[loss=0.2875, simple_loss=0.3559, pruned_loss=0.1095, over 5675455.16 frames. ], giga_tot_loss[loss=0.317, simple_loss=0.3805, pruned_loss=0.1267, over 5651227.98 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:15:34,431 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365081.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:16:10,339 INFO [train.py:968] (1/2) Epoch 30, batch 44650, giga_loss[loss=0.2827, simple_loss=0.3505, pruned_loss=0.1075, over 28908.00 frames. ], tot_loss[loss=0.3112, simple_loss=0.376, pruned_loss=0.1232, over 5657982.94 frames. ], libri_tot_loss[loss=0.287, simple_loss=0.3555, pruned_loss=0.1093, over 5678737.60 frames. ], giga_tot_loss[loss=0.3143, simple_loss=0.3787, pruned_loss=0.125, over 5655031.27 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:16:20,435 INFO [zipformer.py:1188] (1/2) 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] (1/2) 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 18:16:56,131 INFO [train.py:968] (1/2) Epoch 30, batch 44700, libri_loss[loss=0.2832, simple_loss=0.3569, pruned_loss=0.1047, over 28664.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.3746, pruned_loss=0.1208, over 5664510.84 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3547, pruned_loss=0.1088, over 5683699.46 frames. ], giga_tot_loss[loss=0.3119, simple_loss=0.378, pruned_loss=0.1229, over 5657233.19 frames. ], batch size: 106, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:17:29,513 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365201.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:17:36,039 INFO [scaling.py:679] (1/2) Whitening: num_groups=1, num_channels=128, metric=1.98 vs. limit=5.0
+2023-03-15 18:17:36,450 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:17:38,597 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:17:44,768 INFO [train.py:968] (1/2) Epoch 30, batch 44750, giga_loss[loss=0.3107, simple_loss=0.385, pruned_loss=0.1182, over 28536.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3756, pruned_loss=0.12, over 5666744.97 frames. ], libri_tot_loss[loss=0.2859, simple_loss=0.3545, pruned_loss=0.1086, over 5685398.82 frames. ], giga_tot_loss[loss=0.3113, simple_loss=0.3786, pruned_loss=0.122, over 5659492.27 frames. ], batch size: 336, lr: 1.05e-03, grad_scale: 4.0
+2023-03-15 18:17:49,058 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:17:51,293 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365227.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:18:00,516 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.6812, 1.8637, 1.5275, 1.7779], device='cuda:1'), covar=tensor([0.2695, 0.2833, 0.3289, 0.2358], device='cuda:1'), in_proj_covar=tensor([0.1634, 0.1171, 0.1443, 0.1022], device='cuda:1'), out_proj_covar=tensor([0.0015, 0.0013, 0.0013, 0.0009], device='cuda:1')
+2023-03-15 18:18:02,604 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:18:09,858 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([3.9252, 3.7325, 3.5765, 1.8983], device='cuda:1'), covar=tensor([0.0797, 0.0949, 0.0885, 0.2012], device='cuda:1'), in_proj_covar=tensor([0.1350, 0.1246, 0.1044, 0.0769], device='cuda:1'), out_proj_covar=tensor([0.0020, 0.0017, 0.0014, 0.0012], device='cuda:1')
+2023-03-15 18:18:10,912 INFO [optim.py:369] (1/2) 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 18:18:21,701 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365256.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:18:32,867 INFO [train.py:968] (1/2) Epoch 30, batch 44800, giga_loss[loss=0.3049, simple_loss=0.3695, pruned_loss=0.1202, over 28959.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3755, pruned_loss=0.1199, over 5682410.57 frames. ], libri_tot_loss[loss=0.2861, simple_loss=0.3549, pruned_loss=0.1087, over 5693446.53 frames. ], giga_tot_loss[loss=0.311, simple_loss=0.3784, pruned_loss=0.1218, over 5668956.79 frames. ], batch size: 213, lr: 1.05e-03, grad_scale: 8.0
+2023-03-15 18:18:37,267 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:18:41,592 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:19:11,873 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365303.0, num_to_drop=0, layers_to_drop=set()
+2023-03-15 18:19:21,560 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-15 18:19:25,097 INFO [train.py:968] (1/2) Epoch 30, batch 44850, giga_loss[loss=0.3001, simple_loss=0.3682, pruned_loss=0.1161, over 28711.00 frames. ], tot_loss[loss=0.308, simple_loss=0.3754, pruned_loss=0.1203, over 5663515.82 frames. ], libri_tot_loss[loss=0.2863, simple_loss=0.3551, pruned_loss=0.1088, over 5675374.64 frames. ], giga_tot_loss[loss=0.3106, simple_loss=0.3776, pruned_loss=0.1218, over 5669790.56 frames. ], batch size: 262, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:19:34,196 INFO [scaling.py:679] (1/2) Whitening: num_groups=4, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-15 18:19:48,510 INFO [zipformer.py:1188] (1/2) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1365344.0, num_to_drop=1, layers_to_drop={0}
+2023-03-15 18:19:51,465 INFO [zipformer.py:1188] (1/2) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1365347.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:19:51,739 INFO [optim.py:369] (1/2) 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 18:20:11,066 INFO [train.py:968] (1/2) Epoch 30, batch 44900, giga_loss[loss=0.2818, simple_loss=0.3524, pruned_loss=0.1056, over 28927.00 frames. ], tot_loss[loss=0.3085, simple_loss=0.3749, pruned_loss=0.121, over 5604532.67 frames. ], libri_tot_loss[loss=0.2876, simple_loss=0.356, pruned_loss=0.1096, over 5613980.37 frames. ], giga_tot_loss[loss=0.3099, simple_loss=0.3763, pruned_loss=0.1218, over 5663411.21 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:20:18,077 INFO [zipformer.py:1188] (1/2) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1365376.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:20:37,985 INFO [zipformer.py:1188] (1/2) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1365394.0, num_to_drop=1, layers_to_drop={1}
+2023-03-15 18:21:02,633 INFO [train.py:968] (1/2) Epoch 30, batch 44950, giga_loss[loss=0.2672, simple_loss=0.3445, pruned_loss=0.0949, over 28998.00 frames. ], tot_loss[loss=0.3095, simple_loss=0.3743, pruned_loss=0.1224, over 5575871.84 frames. ], libri_tot_loss[loss=0.2886, simple_loss=0.3567, pruned_loss=0.1102, over 5580736.42 frames. ], giga_tot_loss[loss=0.3101, simple_loss=0.3751, pruned_loss=0.1226, over 5651433.56 frames. ], batch size: 164, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:21:33,467 INFO [optim.py:369] (1/2) 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 18:21:49,152 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([2.2002, 2.5740, 2.2025, 2.3065], device='cuda:1'), covar=tensor([0.0519, 0.0230, 0.0241, 0.0608], device='cuda:1'), in_proj_covar=tensor([0.0196, 0.0124, 0.0122, 0.0232], device='cuda:1'), out_proj_covar=tensor([0.0107, 0.0077, 0.0068, 0.0117], device='cuda:1')
+2023-03-15 18:21:51,564 INFO [train.py:968] (1/2) Epoch 30, batch 45000, giga_loss[loss=0.2962, simple_loss=0.3633, pruned_loss=0.1145, over 28869.00 frames. ], tot_loss[loss=0.3079, simple_loss=0.3722, pruned_loss=0.1218, over 5558429.46 frames. ], libri_tot_loss[loss=0.2891, simple_loss=0.3571, pruned_loss=0.1106, over 5547771.21 frames. ], giga_tot_loss[loss=0.3081, simple_loss=0.3727, pruned_loss=0.1217, over 5650287.12 frames. ], batch size: 186, lr: 1.05e-03, grad_scale: 2.0
+2023-03-15 18:21:51,565 INFO [train.py:1003] (1/2) Computing validation loss
+2023-03-15 18:21:57,246 INFO [zipformer.py:2441] (1/2) attn_weights_entropy = tensor([1.2397, 1.7700, 1.3274, 0.4571], device='cuda:1'), covar=tensor([0.6156, 0.4525, 0.5700, 0.7816], device='cuda:1'), in_proj_covar=tensor([0.1889, 0.1776, 0.1685, 0.1537], device='cuda:1'), out_proj_covar=tensor([0.0005, 0.0005, 0.0004, 0.0004], device='cuda:1')
+2023-03-15 18:22:00,673 INFO [train.py:1012] (1/2) 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] (1/2) Maximum memory allocated so far is 19850MB
+2023-03-15 18:22:21,415 INFO [train.py:866] (1/2) libri reaches end of dataloader
+2023-03-15 18:22:22,695 INFO [train.py:1284] (1/2) Done!